]> Git Repo - J-linux.git/blob - drivers/net/ethernet/intel/ixgbe/ixgbe_main.c
Merge tag 'vfs-6.13-rc7.fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs
[J-linux.git] / drivers / net / ethernet / intel / ixgbe / ixgbe_main.c
1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright(c) 1999 - 2018 Intel Corporation. */
3
4 #include <linux/types.h>
5 #include <linux/module.h>
6 #include <linux/pci.h>
7 #include <linux/netdevice.h>
8 #include <linux/vmalloc.h>
9 #include <linux/string.h>
10 #include <linux/in.h>
11 #include <linux/interrupt.h>
12 #include <linux/ip.h>
13 #include <linux/tcp.h>
14 #include <linux/sctp.h>
15 #include <linux/pkt_sched.h>
16 #include <linux/ipv6.h>
17 #include <linux/slab.h>
18 #include <net/checksum.h>
19 #include <net/ip6_checksum.h>
20 #include <linux/etherdevice.h>
21 #include <linux/ethtool.h>
22 #include <linux/if.h>
23 #include <linux/if_vlan.h>
24 #include <linux/if_macvlan.h>
25 #include <linux/if_bridge.h>
26 #include <linux/prefetch.h>
27 #include <linux/bpf.h>
28 #include <linux/bpf_trace.h>
29 #include <linux/atomic.h>
30 #include <linux/numa.h>
31 #include <generated/utsrelease.h>
32 #include <scsi/fc/fc_fcoe.h>
33 #include <net/udp_tunnel.h>
34 #include <net/pkt_cls.h>
35 #include <net/tc_act/tc_gact.h>
36 #include <net/tc_act/tc_mirred.h>
37 #include <net/vxlan.h>
38 #include <net/mpls.h>
39 #include <net/netdev_queues.h>
40 #include <net/xdp_sock_drv.h>
41 #include <net/xfrm.h>
42
43 #include "ixgbe.h"
44 #include "ixgbe_common.h"
45 #include "ixgbe_dcb_82599.h"
46 #include "ixgbe_mbx.h"
47 #include "ixgbe_phy.h"
48 #include "ixgbe_sriov.h"
49 #include "ixgbe_model.h"
50 #include "ixgbe_txrx_common.h"
51
52 char ixgbe_driver_name[] = "ixgbe";
53 static const char ixgbe_driver_string[] =
54                               "Intel(R) 10 Gigabit PCI Express Network Driver";
55 #ifdef IXGBE_FCOE
56 char ixgbe_default_device_descr[] =
57                               "Intel(R) 10 Gigabit Network Connection";
58 #else
59 static char ixgbe_default_device_descr[] =
60                               "Intel(R) 10 Gigabit Network Connection";
61 #endif
62 static const char ixgbe_copyright[] =
63                                 "Copyright (c) 1999-2016 Intel Corporation.";
64
65 static const char ixgbe_overheat_msg[] = "Network adapter has been stopped because it has over heated. Restart the computer. If the problem persists, power off the system and replace the adapter";
66
67 static const struct ixgbe_info *ixgbe_info_tbl[] = {
68         [board_82598]           = &ixgbe_82598_info,
69         [board_82599]           = &ixgbe_82599_info,
70         [board_X540]            = &ixgbe_X540_info,
71         [board_X550]            = &ixgbe_X550_info,
72         [board_X550EM_x]        = &ixgbe_X550EM_x_info,
73         [board_x550em_x_fw]     = &ixgbe_x550em_x_fw_info,
74         [board_x550em_a]        = &ixgbe_x550em_a_info,
75         [board_x550em_a_fw]     = &ixgbe_x550em_a_fw_info,
76 };
77
78 /* ixgbe_pci_tbl - PCI Device ID Table
79  *
80  * Wildcard entries (PCI_ANY_ID) should come last
81  * Last entry must be all 0s
82  *
83  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
84  *   Class, Class Mask, private data (not used) }
85  */
86 static const struct pci_device_id ixgbe_pci_tbl[] = {
87         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598), board_82598 },
88         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT), board_82598 },
89         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT), board_82598 },
90         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT), board_82598 },
91         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT2), board_82598 },
92         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4), board_82598 },
93         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT), board_82598 },
94         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT), board_82598 },
95         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM), board_82598 },
96         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR), board_82598 },
97         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM), board_82598 },
98         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX), board_82598 },
99         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4), board_82599 },
100         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM), board_82599 },
101         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KR), board_82599 },
102         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP), board_82599 },
103         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_EM), board_82599 },
104         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4_MEZZ), board_82599 },
105         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_CX4), board_82599 },
106         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_BACKPLANE_FCOE), board_82599 },
107         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_FCOE), board_82599 },
108         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_T3_LOM), board_82599 },
109         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_COMBO_BACKPLANE), board_82599 },
110         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T), board_X540 },
111         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF2), board_82599 },
112         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_LS), board_82599 },
113         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_QSFP_SF_QP), board_82599 },
114         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599EN_SFP), board_82599 },
115         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF_QP), board_82599 },
116         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T1), board_X540 },
117         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550T), board_X550},
118         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550T1), board_X550},
119         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_KX4), board_X550EM_x},
120         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_XFI), board_X550EM_x},
121         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_KR), board_X550EM_x},
122         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_10G_T), board_X550EM_x},
123         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_SFP), board_X550EM_x},
124         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_1G_T), board_x550em_x_fw},
125         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_KR), board_x550em_a },
126         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_KR_L), board_x550em_a },
127         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_SFP_N), board_x550em_a },
128         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_SGMII), board_x550em_a },
129         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_SGMII_L), board_x550em_a },
130         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_10G_T), board_x550em_a},
131         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_SFP), board_x550em_a },
132         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_1G_T), board_x550em_a_fw },
133         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_1G_T_L), board_x550em_a_fw },
134         /* required last entry */
135         {0, }
136 };
137 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
138
139 #ifdef CONFIG_IXGBE_DCA
140 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
141                             void *p);
142 static struct notifier_block dca_notifier = {
143         .notifier_call = ixgbe_notify_dca,
144         .next          = NULL,
145         .priority      = 0
146 };
147 #endif
148
149 #ifdef CONFIG_PCI_IOV
150 static unsigned int max_vfs;
151 module_param(max_vfs, uint, 0);
152 MODULE_PARM_DESC(max_vfs,
153                  "Maximum number of virtual functions to allocate per physical function - default is zero and maximum value is 63. (Deprecated)");
154 #endif /* CONFIG_PCI_IOV */
155
156 static bool allow_unsupported_sfp;
157 module_param(allow_unsupported_sfp, bool, 0444);
158 MODULE_PARM_DESC(allow_unsupported_sfp,
159                  "Allow unsupported and untested SFP+ modules on 82599-based adapters");
160
161 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
162 static int debug = -1;
163 module_param(debug, int, 0);
164 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
165
166 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
167 MODULE_LICENSE("GPL v2");
168
169 DEFINE_STATIC_KEY_FALSE(ixgbe_xdp_locking_key);
170 EXPORT_SYMBOL(ixgbe_xdp_locking_key);
171
172 static struct workqueue_struct *ixgbe_wq;
173
174 static bool ixgbe_check_cfg_remove(struct ixgbe_hw *hw, struct pci_dev *pdev);
175 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *);
176
177 static const struct net_device_ops ixgbe_netdev_ops;
178
179 static bool netif_is_ixgbe(struct net_device *dev)
180 {
181         return dev && (dev->netdev_ops == &ixgbe_netdev_ops);
182 }
183
184 static int ixgbe_read_pci_cfg_word_parent(struct ixgbe_adapter *adapter,
185                                           u32 reg, u16 *value)
186 {
187         struct pci_dev *parent_dev;
188         struct pci_bus *parent_bus;
189
190         parent_bus = adapter->pdev->bus->parent;
191         if (!parent_bus)
192                 return -1;
193
194         parent_dev = parent_bus->self;
195         if (!parent_dev)
196                 return -1;
197
198         if (!pci_is_pcie(parent_dev))
199                 return -1;
200
201         pcie_capability_read_word(parent_dev, reg, value);
202         if (*value == IXGBE_FAILED_READ_CFG_WORD &&
203             ixgbe_check_cfg_remove(&adapter->hw, parent_dev))
204                 return -1;
205         return 0;
206 }
207
208 static int ixgbe_get_parent_bus_info(struct ixgbe_adapter *adapter)
209 {
210         struct ixgbe_hw *hw = &adapter->hw;
211         u16 link_status = 0;
212         int err;
213
214         hw->bus.type = ixgbe_bus_type_pci_express;
215
216         /* Get the negotiated link width and speed from PCI config space of the
217          * parent, as this device is behind a switch
218          */
219         err = ixgbe_read_pci_cfg_word_parent(adapter, 18, &link_status);
220
221         /* assume caller will handle error case */
222         if (err)
223                 return err;
224
225         hw->bus.width = ixgbe_convert_bus_width(link_status);
226         hw->bus.speed = ixgbe_convert_bus_speed(link_status);
227
228         return 0;
229 }
230
231 /**
232  * ixgbe_pcie_from_parent - Determine whether PCIe info should come from parent
233  * @hw: hw specific details
234  *
235  * This function is used by probe to determine whether a device's PCI-Express
236  * bandwidth details should be gathered from the parent bus instead of from the
237  * device. Used to ensure that various locations all have the correct device ID
238  * checks.
239  */
240 static inline bool ixgbe_pcie_from_parent(struct ixgbe_hw *hw)
241 {
242         switch (hw->device_id) {
243         case IXGBE_DEV_ID_82599_SFP_SF_QP:
244         case IXGBE_DEV_ID_82599_QSFP_SF_QP:
245                 return true;
246         default:
247                 return false;
248         }
249 }
250
251 static void ixgbe_check_minimum_link(struct ixgbe_adapter *adapter,
252                                      int expected_gts)
253 {
254         struct ixgbe_hw *hw = &adapter->hw;
255         struct pci_dev *pdev;
256
257         /* Some devices are not connected over PCIe and thus do not negotiate
258          * speed. These devices do not have valid bus info, and thus any report
259          * we generate may not be correct.
260          */
261         if (hw->bus.type == ixgbe_bus_type_internal)
262                 return;
263
264         /* determine whether to use the parent device */
265         if (ixgbe_pcie_from_parent(&adapter->hw))
266                 pdev = adapter->pdev->bus->parent->self;
267         else
268                 pdev = adapter->pdev;
269
270         pcie_print_link_status(pdev);
271 }
272
273 static void ixgbe_service_event_schedule(struct ixgbe_adapter *adapter)
274 {
275         if (!test_bit(__IXGBE_DOWN, &adapter->state) &&
276             !test_bit(__IXGBE_REMOVING, &adapter->state) &&
277             !test_and_set_bit(__IXGBE_SERVICE_SCHED, &adapter->state))
278                 queue_work(ixgbe_wq, &adapter->service_task);
279 }
280
281 static void ixgbe_remove_adapter(struct ixgbe_hw *hw)
282 {
283         struct ixgbe_adapter *adapter = hw->back;
284
285         if (!hw->hw_addr)
286                 return;
287         hw->hw_addr = NULL;
288         e_dev_err("Adapter removed\n");
289         if (test_bit(__IXGBE_SERVICE_INITED, &adapter->state))
290                 ixgbe_service_event_schedule(adapter);
291 }
292
293 static u32 ixgbe_check_remove(struct ixgbe_hw *hw, u32 reg)
294 {
295         u8 __iomem *reg_addr;
296         u32 value;
297         int i;
298
299         reg_addr = READ_ONCE(hw->hw_addr);
300         if (ixgbe_removed(reg_addr))
301                 return IXGBE_FAILED_READ_REG;
302
303         /* Register read of 0xFFFFFFF can indicate the adapter has been removed,
304          * so perform several status register reads to determine if the adapter
305          * has been removed.
306          */
307         for (i = 0; i < IXGBE_FAILED_READ_RETRIES; i++) {
308                 value = readl(reg_addr + IXGBE_STATUS);
309                 if (value != IXGBE_FAILED_READ_REG)
310                         break;
311                 mdelay(3);
312         }
313
314         if (value == IXGBE_FAILED_READ_REG)
315                 ixgbe_remove_adapter(hw);
316         else
317                 value = readl(reg_addr + reg);
318         return value;
319 }
320
321 /**
322  * ixgbe_read_reg - Read from device register
323  * @hw: hw specific details
324  * @reg: offset of register to read
325  *
326  * Returns : value read or IXGBE_FAILED_READ_REG if removed
327  *
328  * This function is used to read device registers. It checks for device
329  * removal by confirming any read that returns all ones by checking the
330  * status register value for all ones. This function avoids reading from
331  * the hardware if a removal was previously detected in which case it
332  * returns IXGBE_FAILED_READ_REG (all ones).
333  */
334 u32 ixgbe_read_reg(struct ixgbe_hw *hw, u32 reg)
335 {
336         u8 __iomem *reg_addr = READ_ONCE(hw->hw_addr);
337         u32 value;
338
339         if (ixgbe_removed(reg_addr))
340                 return IXGBE_FAILED_READ_REG;
341         if (unlikely(hw->phy.nw_mng_if_sel &
342                      IXGBE_NW_MNG_IF_SEL_SGMII_ENABLE)) {
343                 struct ixgbe_adapter *adapter;
344                 int i;
345
346                 for (i = 0; i < 200; ++i) {
347                         value = readl(reg_addr + IXGBE_MAC_SGMII_BUSY);
348                         if (likely(!value))
349                                 goto writes_completed;
350                         if (value == IXGBE_FAILED_READ_REG) {
351                                 ixgbe_remove_adapter(hw);
352                                 return IXGBE_FAILED_READ_REG;
353                         }
354                         udelay(5);
355                 }
356
357                 adapter = hw->back;
358                 e_warn(hw, "register writes incomplete %08x\n", value);
359         }
360
361 writes_completed:
362         value = readl(reg_addr + reg);
363         if (unlikely(value == IXGBE_FAILED_READ_REG))
364                 value = ixgbe_check_remove(hw, reg);
365         return value;
366 }
367
368 static bool ixgbe_check_cfg_remove(struct ixgbe_hw *hw, struct pci_dev *pdev)
369 {
370         u16 value;
371
372         pci_read_config_word(pdev, PCI_VENDOR_ID, &value);
373         if (value == IXGBE_FAILED_READ_CFG_WORD) {
374                 ixgbe_remove_adapter(hw);
375                 return true;
376         }
377         return false;
378 }
379
380 u16 ixgbe_read_pci_cfg_word(struct ixgbe_hw *hw, u32 reg)
381 {
382         struct ixgbe_adapter *adapter = hw->back;
383         u16 value;
384
385         if (ixgbe_removed(hw->hw_addr))
386                 return IXGBE_FAILED_READ_CFG_WORD;
387         pci_read_config_word(adapter->pdev, reg, &value);
388         if (value == IXGBE_FAILED_READ_CFG_WORD &&
389             ixgbe_check_cfg_remove(hw, adapter->pdev))
390                 return IXGBE_FAILED_READ_CFG_WORD;
391         return value;
392 }
393
394 #ifdef CONFIG_PCI_IOV
395 static u32 ixgbe_read_pci_cfg_dword(struct ixgbe_hw *hw, u32 reg)
396 {
397         struct ixgbe_adapter *adapter = hw->back;
398         u32 value;
399
400         if (ixgbe_removed(hw->hw_addr))
401                 return IXGBE_FAILED_READ_CFG_DWORD;
402         pci_read_config_dword(adapter->pdev, reg, &value);
403         if (value == IXGBE_FAILED_READ_CFG_DWORD &&
404             ixgbe_check_cfg_remove(hw, adapter->pdev))
405                 return IXGBE_FAILED_READ_CFG_DWORD;
406         return value;
407 }
408 #endif /* CONFIG_PCI_IOV */
409
410 void ixgbe_write_pci_cfg_word(struct ixgbe_hw *hw, u32 reg, u16 value)
411 {
412         struct ixgbe_adapter *adapter = hw->back;
413
414         if (ixgbe_removed(hw->hw_addr))
415                 return;
416         pci_write_config_word(adapter->pdev, reg, value);
417 }
418
419 static void ixgbe_service_event_complete(struct ixgbe_adapter *adapter)
420 {
421         BUG_ON(!test_bit(__IXGBE_SERVICE_SCHED, &adapter->state));
422
423         /* flush memory to make sure state is correct before next watchdog */
424         smp_mb__before_atomic();
425         clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
426 }
427
428 struct ixgbe_reg_info {
429         u32 ofs;
430         char *name;
431 };
432
433 static const struct ixgbe_reg_info ixgbe_reg_info_tbl[] = {
434
435         /* General Registers */
436         {IXGBE_CTRL, "CTRL"},
437         {IXGBE_STATUS, "STATUS"},
438         {IXGBE_CTRL_EXT, "CTRL_EXT"},
439
440         /* Interrupt Registers */
441         {IXGBE_EICR, "EICR"},
442
443         /* RX Registers */
444         {IXGBE_SRRCTL(0), "SRRCTL"},
445         {IXGBE_DCA_RXCTRL(0), "DRXCTL"},
446         {IXGBE_RDLEN(0), "RDLEN"},
447         {IXGBE_RDH(0), "RDH"},
448         {IXGBE_RDT(0), "RDT"},
449         {IXGBE_RXDCTL(0), "RXDCTL"},
450         {IXGBE_RDBAL(0), "RDBAL"},
451         {IXGBE_RDBAH(0), "RDBAH"},
452
453         /* TX Registers */
454         {IXGBE_TDBAL(0), "TDBAL"},
455         {IXGBE_TDBAH(0), "TDBAH"},
456         {IXGBE_TDLEN(0), "TDLEN"},
457         {IXGBE_TDH(0), "TDH"},
458         {IXGBE_TDT(0), "TDT"},
459         {IXGBE_TXDCTL(0), "TXDCTL"},
460
461         /* List Terminator */
462         { .name = NULL }
463 };
464
465
466 /*
467  * ixgbe_regdump - register printout routine
468  */
469 static void ixgbe_regdump(struct ixgbe_hw *hw, struct ixgbe_reg_info *reginfo)
470 {
471         int i;
472         char rname[16];
473         u32 regs[64];
474
475         switch (reginfo->ofs) {
476         case IXGBE_SRRCTL(0):
477                 for (i = 0; i < 64; i++)
478                         regs[i] = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i));
479                 break;
480         case IXGBE_DCA_RXCTRL(0):
481                 for (i = 0; i < 64; i++)
482                         regs[i] = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(i));
483                 break;
484         case IXGBE_RDLEN(0):
485                 for (i = 0; i < 64; i++)
486                         regs[i] = IXGBE_READ_REG(hw, IXGBE_RDLEN(i));
487                 break;
488         case IXGBE_RDH(0):
489                 for (i = 0; i < 64; i++)
490                         regs[i] = IXGBE_READ_REG(hw, IXGBE_RDH(i));
491                 break;
492         case IXGBE_RDT(0):
493                 for (i = 0; i < 64; i++)
494                         regs[i] = IXGBE_READ_REG(hw, IXGBE_RDT(i));
495                 break;
496         case IXGBE_RXDCTL(0):
497                 for (i = 0; i < 64; i++)
498                         regs[i] = IXGBE_READ_REG(hw, IXGBE_RXDCTL(i));
499                 break;
500         case IXGBE_RDBAL(0):
501                 for (i = 0; i < 64; i++)
502                         regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAL(i));
503                 break;
504         case IXGBE_RDBAH(0):
505                 for (i = 0; i < 64; i++)
506                         regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAH(i));
507                 break;
508         case IXGBE_TDBAL(0):
509                 for (i = 0; i < 64; i++)
510                         regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAL(i));
511                 break;
512         case IXGBE_TDBAH(0):
513                 for (i = 0; i < 64; i++)
514                         regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAH(i));
515                 break;
516         case IXGBE_TDLEN(0):
517                 for (i = 0; i < 64; i++)
518                         regs[i] = IXGBE_READ_REG(hw, IXGBE_TDLEN(i));
519                 break;
520         case IXGBE_TDH(0):
521                 for (i = 0; i < 64; i++)
522                         regs[i] = IXGBE_READ_REG(hw, IXGBE_TDH(i));
523                 break;
524         case IXGBE_TDT(0):
525                 for (i = 0; i < 64; i++)
526                         regs[i] = IXGBE_READ_REG(hw, IXGBE_TDT(i));
527                 break;
528         case IXGBE_TXDCTL(0):
529                 for (i = 0; i < 64; i++)
530                         regs[i] = IXGBE_READ_REG(hw, IXGBE_TXDCTL(i));
531                 break;
532         default:
533                 pr_info("%-15s %08x\n",
534                         reginfo->name, IXGBE_READ_REG(hw, reginfo->ofs));
535                 return;
536         }
537
538         i = 0;
539         while (i < 64) {
540                 int j;
541                 char buf[9 * 8 + 1];
542                 char *p = buf;
543
544                 snprintf(rname, 16, "%s[%d-%d]", reginfo->name, i, i + 7);
545                 for (j = 0; j < 8; j++)
546                         p += sprintf(p, " %08x", regs[i++]);
547                 pr_err("%-15s%s\n", rname, buf);
548         }
549
550 }
551
552 static void ixgbe_print_buffer(struct ixgbe_ring *ring, int n)
553 {
554         struct ixgbe_tx_buffer *tx_buffer;
555
556         tx_buffer = &ring->tx_buffer_info[ring->next_to_clean];
557         pr_info(" %5d %5X %5X %016llX %08X %p %016llX\n",
558                 n, ring->next_to_use, ring->next_to_clean,
559                 (u64)dma_unmap_addr(tx_buffer, dma),
560                 dma_unmap_len(tx_buffer, len),
561                 tx_buffer->next_to_watch,
562                 (u64)tx_buffer->time_stamp);
563 }
564
565 /*
566  * ixgbe_dump - Print registers, tx-rings and rx-rings
567  */
568 static void ixgbe_dump(struct ixgbe_adapter *adapter)
569 {
570         struct net_device *netdev = adapter->netdev;
571         struct ixgbe_hw *hw = &adapter->hw;
572         struct ixgbe_reg_info *reginfo;
573         int n = 0;
574         struct ixgbe_ring *ring;
575         struct ixgbe_tx_buffer *tx_buffer;
576         union ixgbe_adv_tx_desc *tx_desc;
577         struct my_u0 { u64 a; u64 b; } *u0;
578         struct ixgbe_ring *rx_ring;
579         union ixgbe_adv_rx_desc *rx_desc;
580         struct ixgbe_rx_buffer *rx_buffer_info;
581         int i = 0;
582
583         if (!netif_msg_hw(adapter))
584                 return;
585
586         /* Print netdevice Info */
587         if (netdev) {
588                 dev_info(&adapter->pdev->dev, "Net device Info\n");
589                 pr_info("Device Name     state            "
590                         "trans_start\n");
591                 pr_info("%-15s %016lX %016lX\n",
592                         netdev->name,
593                         netdev->state,
594                         dev_trans_start(netdev));
595         }
596
597         /* Print Registers */
598         dev_info(&adapter->pdev->dev, "Register Dump\n");
599         pr_info(" Register Name   Value\n");
600         for (reginfo = (struct ixgbe_reg_info *)ixgbe_reg_info_tbl;
601              reginfo->name; reginfo++) {
602                 ixgbe_regdump(hw, reginfo);
603         }
604
605         /* Print TX Ring Summary */
606         if (!netdev || !netif_running(netdev))
607                 return;
608
609         dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
610         pr_info(" %s     %s              %s        %s\n",
611                 "Queue [NTU] [NTC] [bi(ntc)->dma  ]",
612                 "leng", "ntw", "timestamp");
613         for (n = 0; n < adapter->num_tx_queues; n++) {
614                 ring = adapter->tx_ring[n];
615                 ixgbe_print_buffer(ring, n);
616         }
617
618         for (n = 0; n < adapter->num_xdp_queues; n++) {
619                 ring = adapter->xdp_ring[n];
620                 ixgbe_print_buffer(ring, n);
621         }
622
623         /* Print TX Rings */
624         if (!netif_msg_tx_done(adapter))
625                 goto rx_ring_summary;
626
627         dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
628
629         /* Transmit Descriptor Formats
630          *
631          * 82598 Advanced Transmit Descriptor
632          *   +--------------------------------------------------------------+
633          * 0 |         Buffer Address [63:0]                                |
634          *   +--------------------------------------------------------------+
635          * 8 |  PAYLEN  | POPTS  | IDX | STA | DCMD  |DTYP |  RSV |  DTALEN |
636          *   +--------------------------------------------------------------+
637          *   63       46 45    40 39 36 35 32 31   24 23 20 19              0
638          *
639          * 82598 Advanced Transmit Descriptor (Write-Back Format)
640          *   +--------------------------------------------------------------+
641          * 0 |                          RSV [63:0]                          |
642          *   +--------------------------------------------------------------+
643          * 8 |            RSV           |  STA  |          NXTSEQ           |
644          *   +--------------------------------------------------------------+
645          *   63                       36 35   32 31                         0
646          *
647          * 82599+ Advanced Transmit Descriptor
648          *   +--------------------------------------------------------------+
649          * 0 |         Buffer Address [63:0]                                |
650          *   +--------------------------------------------------------------+
651          * 8 |PAYLEN  |POPTS|CC|IDX  |STA  |DCMD  |DTYP |MAC  |RSV  |DTALEN |
652          *   +--------------------------------------------------------------+
653          *   63     46 45 40 39 38 36 35 32 31  24 23 20 19 18 17 16 15     0
654          *
655          * 82599+ Advanced Transmit Descriptor (Write-Back Format)
656          *   +--------------------------------------------------------------+
657          * 0 |                          RSV [63:0]                          |
658          *   +--------------------------------------------------------------+
659          * 8 |            RSV           |  STA  |           RSV             |
660          *   +--------------------------------------------------------------+
661          *   63                       36 35   32 31                         0
662          */
663
664         for (n = 0; n < adapter->num_tx_queues; n++) {
665                 ring = adapter->tx_ring[n];
666                 pr_info("------------------------------------\n");
667                 pr_info("TX QUEUE INDEX = %d\n", ring->queue_index);
668                 pr_info("------------------------------------\n");
669                 pr_info("%s%s    %s              %s        %s          %s\n",
670                         "T [desc]     [address 63:0  ] ",
671                         "[PlPOIdStDDt Ln] [bi->dma       ] ",
672                         "leng", "ntw", "timestamp", "bi->skb");
673
674                 for (i = 0; ring->desc && (i < ring->count); i++) {
675                         tx_desc = IXGBE_TX_DESC(ring, i);
676                         tx_buffer = &ring->tx_buffer_info[i];
677                         u0 = (struct my_u0 *)tx_desc;
678                         if (dma_unmap_len(tx_buffer, len) > 0) {
679                                 const char *ring_desc;
680
681                                 if (i == ring->next_to_use &&
682                                     i == ring->next_to_clean)
683                                         ring_desc = " NTC/U";
684                                 else if (i == ring->next_to_use)
685                                         ring_desc = " NTU";
686                                 else if (i == ring->next_to_clean)
687                                         ring_desc = " NTC";
688                                 else
689                                         ring_desc = "";
690                                 pr_info("T [0x%03X]    %016llX %016llX %016llX %08X %p %016llX %p%s",
691                                         i,
692                                         le64_to_cpu((__force __le64)u0->a),
693                                         le64_to_cpu((__force __le64)u0->b),
694                                         (u64)dma_unmap_addr(tx_buffer, dma),
695                                         dma_unmap_len(tx_buffer, len),
696                                         tx_buffer->next_to_watch,
697                                         (u64)tx_buffer->time_stamp,
698                                         tx_buffer->skb,
699                                         ring_desc);
700
701                                 if (netif_msg_pktdata(adapter) &&
702                                     tx_buffer->skb)
703                                         print_hex_dump(KERN_INFO, "",
704                                                 DUMP_PREFIX_ADDRESS, 16, 1,
705                                                 tx_buffer->skb->data,
706                                                 dma_unmap_len(tx_buffer, len),
707                                                 true);
708                         }
709                 }
710         }
711
712         /* Print RX Rings Summary */
713 rx_ring_summary:
714         dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
715         pr_info("Queue [NTU] [NTC]\n");
716         for (n = 0; n < adapter->num_rx_queues; n++) {
717                 rx_ring = adapter->rx_ring[n];
718                 pr_info("%5d %5X %5X\n",
719                         n, rx_ring->next_to_use, rx_ring->next_to_clean);
720         }
721
722         /* Print RX Rings */
723         if (!netif_msg_rx_status(adapter))
724                 return;
725
726         dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
727
728         /* Receive Descriptor Formats
729          *
730          * 82598 Advanced Receive Descriptor (Read) Format
731          *    63                                           1        0
732          *    +-----------------------------------------------------+
733          *  0 |       Packet Buffer Address [63:1]           |A0/NSE|
734          *    +----------------------------------------------+------+
735          *  8 |       Header Buffer Address [63:1]           |  DD  |
736          *    +-----------------------------------------------------+
737          *
738          *
739          * 82598 Advanced Receive Descriptor (Write-Back) Format
740          *
741          *   63       48 47    32 31  30      21 20 16 15   4 3     0
742          *   +------------------------------------------------------+
743          * 0 |       RSS Hash /  |SPH| HDR_LEN  | RSV |Packet|  RSS |
744          *   | Packet   | IP     |   |          |     | Type | Type |
745          *   | Checksum | Ident  |   |          |     |      |      |
746          *   +------------------------------------------------------+
747          * 8 | VLAN Tag | Length | Extended Error | Extended Status |
748          *   +------------------------------------------------------+
749          *   63       48 47    32 31            20 19               0
750          *
751          * 82599+ Advanced Receive Descriptor (Read) Format
752          *    63                                           1        0
753          *    +-----------------------------------------------------+
754          *  0 |       Packet Buffer Address [63:1]           |A0/NSE|
755          *    +----------------------------------------------+------+
756          *  8 |       Header Buffer Address [63:1]           |  DD  |
757          *    +-----------------------------------------------------+
758          *
759          *
760          * 82599+ Advanced Receive Descriptor (Write-Back) Format
761          *
762          *   63       48 47    32 31  30      21 20 17 16   4 3     0
763          *   +------------------------------------------------------+
764          * 0 |RSS / Frag Checksum|SPH| HDR_LEN  |RSC- |Packet|  RSS |
765          *   |/ RTT / PCoE_PARAM |   |          | CNT | Type | Type |
766          *   |/ Flow Dir Flt ID  |   |          |     |      |      |
767          *   +------------------------------------------------------+
768          * 8 | VLAN Tag | Length |Extended Error| Xtnd Status/NEXTP |
769          *   +------------------------------------------------------+
770          *   63       48 47    32 31          20 19                 0
771          */
772
773         for (n = 0; n < adapter->num_rx_queues; n++) {
774                 rx_ring = adapter->rx_ring[n];
775                 pr_info("------------------------------------\n");
776                 pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
777                 pr_info("------------------------------------\n");
778                 pr_info("%s%s%s\n",
779                         "R  [desc]      [ PktBuf     A0] ",
780                         "[  HeadBuf   DD] [bi->dma       ] [bi->skb       ] ",
781                         "<-- Adv Rx Read format");
782                 pr_info("%s%s%s\n",
783                         "RWB[desc]      [PcsmIpSHl PtRs] ",
784                         "[vl er S cks ln] ---------------- [bi->skb       ] ",
785                         "<-- Adv Rx Write-Back format");
786
787                 for (i = 0; i < rx_ring->count; i++) {
788                         const char *ring_desc;
789
790                         if (i == rx_ring->next_to_use)
791                                 ring_desc = " NTU";
792                         else if (i == rx_ring->next_to_clean)
793                                 ring_desc = " NTC";
794                         else
795                                 ring_desc = "";
796
797                         rx_buffer_info = &rx_ring->rx_buffer_info[i];
798                         rx_desc = IXGBE_RX_DESC(rx_ring, i);
799                         u0 = (struct my_u0 *)rx_desc;
800                         if (rx_desc->wb.upper.length) {
801                                 /* Descriptor Done */
802                                 pr_info("RWB[0x%03X]     %016llX %016llX ---------------- %p%s\n",
803                                         i,
804                                         le64_to_cpu((__force __le64)u0->a),
805                                         le64_to_cpu((__force __le64)u0->b),
806                                         rx_buffer_info->skb,
807                                         ring_desc);
808                         } else {
809                                 pr_info("R  [0x%03X]     %016llX %016llX %016llX %p%s\n",
810                                         i,
811                                         le64_to_cpu((__force __le64)u0->a),
812                                         le64_to_cpu((__force __le64)u0->b),
813                                         (u64)rx_buffer_info->dma,
814                                         rx_buffer_info->skb,
815                                         ring_desc);
816
817                                 if (netif_msg_pktdata(adapter) &&
818                                     rx_buffer_info->dma) {
819                                         print_hex_dump(KERN_INFO, "",
820                                            DUMP_PREFIX_ADDRESS, 16, 1,
821                                            page_address(rx_buffer_info->page) +
822                                                     rx_buffer_info->page_offset,
823                                            ixgbe_rx_bufsz(rx_ring), true);
824                                 }
825                         }
826                 }
827         }
828 }
829
830 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
831 {
832         u32 ctrl_ext;
833
834         /* Let firmware take over control of h/w */
835         ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
836         IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
837                         ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
838 }
839
840 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
841 {
842         u32 ctrl_ext;
843
844         /* Let firmware know the driver has taken over */
845         ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
846         IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
847                         ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
848 }
849
850 /**
851  * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
852  * @adapter: pointer to adapter struct
853  * @direction: 0 for Rx, 1 for Tx, -1 for other causes
854  * @queue: queue to map the corresponding interrupt to
855  * @msix_vector: the vector to map to the corresponding queue
856  *
857  */
858 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
859                            u8 queue, u8 msix_vector)
860 {
861         u32 ivar, index;
862         struct ixgbe_hw *hw = &adapter->hw;
863         switch (hw->mac.type) {
864         case ixgbe_mac_82598EB:
865                 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
866                 if (direction == -1)
867                         direction = 0;
868                 index = (((direction * 64) + queue) >> 2) & 0x1F;
869                 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
870                 ivar &= ~(0xFF << (8 * (queue & 0x3)));
871                 ivar |= (msix_vector << (8 * (queue & 0x3)));
872                 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
873                 break;
874         case ixgbe_mac_82599EB:
875         case ixgbe_mac_X540:
876         case ixgbe_mac_X550:
877         case ixgbe_mac_X550EM_x:
878         case ixgbe_mac_x550em_a:
879                 if (direction == -1) {
880                         /* other causes */
881                         msix_vector |= IXGBE_IVAR_ALLOC_VAL;
882                         index = ((queue & 1) * 8);
883                         ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
884                         ivar &= ~(0xFF << index);
885                         ivar |= (msix_vector << index);
886                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
887                         break;
888                 } else {
889                         /* tx or rx causes */
890                         msix_vector |= IXGBE_IVAR_ALLOC_VAL;
891                         index = ((16 * (queue & 1)) + (8 * direction));
892                         ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
893                         ivar &= ~(0xFF << index);
894                         ivar |= (msix_vector << index);
895                         IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
896                         break;
897                 }
898         default:
899                 break;
900         }
901 }
902
903 void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
904                             u64 qmask)
905 {
906         u32 mask;
907
908         switch (adapter->hw.mac.type) {
909         case ixgbe_mac_82598EB:
910                 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
911                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
912                 break;
913         case ixgbe_mac_82599EB:
914         case ixgbe_mac_X540:
915         case ixgbe_mac_X550:
916         case ixgbe_mac_X550EM_x:
917         case ixgbe_mac_x550em_a:
918                 mask = (qmask & 0xFFFFFFFF);
919                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
920                 mask = (qmask >> 32);
921                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
922                 break;
923         default:
924                 break;
925         }
926 }
927
928 static void ixgbe_update_xoff_rx_lfc(struct ixgbe_adapter *adapter)
929 {
930         struct ixgbe_hw *hw = &adapter->hw;
931         struct ixgbe_hw_stats *hwstats = &adapter->stats;
932         int i;
933         u32 data;
934
935         if ((hw->fc.current_mode != ixgbe_fc_full) &&
936             (hw->fc.current_mode != ixgbe_fc_rx_pause))
937                 return;
938
939         switch (hw->mac.type) {
940         case ixgbe_mac_82598EB:
941                 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
942                 break;
943         default:
944                 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
945         }
946         hwstats->lxoffrxc += data;
947
948         /* refill credits (no tx hang) if we received xoff */
949         if (!data)
950                 return;
951
952         for (i = 0; i < adapter->num_tx_queues; i++)
953                 clear_bit(__IXGBE_HANG_CHECK_ARMED,
954                           &adapter->tx_ring[i]->state);
955
956         for (i = 0; i < adapter->num_xdp_queues; i++)
957                 clear_bit(__IXGBE_HANG_CHECK_ARMED,
958                           &adapter->xdp_ring[i]->state);
959 }
960
961 static void ixgbe_update_xoff_received(struct ixgbe_adapter *adapter)
962 {
963         struct ixgbe_hw *hw = &adapter->hw;
964         struct ixgbe_hw_stats *hwstats = &adapter->stats;
965         u32 xoff[8] = {0};
966         u8 tc;
967         int i;
968         bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
969
970         if (adapter->ixgbe_ieee_pfc)
971                 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
972
973         if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED) || !pfc_en) {
974                 ixgbe_update_xoff_rx_lfc(adapter);
975                 return;
976         }
977
978         /* update stats for each tc, only valid with PFC enabled */
979         for (i = 0; i < MAX_TX_PACKET_BUFFERS; i++) {
980                 u32 pxoffrxc;
981
982                 switch (hw->mac.type) {
983                 case ixgbe_mac_82598EB:
984                         pxoffrxc = IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i));
985                         break;
986                 default:
987                         pxoffrxc = IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i));
988                 }
989                 hwstats->pxoffrxc[i] += pxoffrxc;
990                 /* Get the TC for given UP */
991                 tc = netdev_get_prio_tc_map(adapter->netdev, i);
992                 xoff[tc] += pxoffrxc;
993         }
994
995         /* disarm tx queues that have received xoff frames */
996         for (i = 0; i < adapter->num_tx_queues; i++) {
997                 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
998
999                 tc = tx_ring->dcb_tc;
1000                 if (xoff[tc])
1001                         clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
1002         }
1003
1004         for (i = 0; i < adapter->num_xdp_queues; i++) {
1005                 struct ixgbe_ring *xdp_ring = adapter->xdp_ring[i];
1006
1007                 tc = xdp_ring->dcb_tc;
1008                 if (xoff[tc])
1009                         clear_bit(__IXGBE_HANG_CHECK_ARMED, &xdp_ring->state);
1010         }
1011 }
1012
1013 static u64 ixgbe_get_tx_completed(struct ixgbe_ring *ring)
1014 {
1015         return ring->stats.packets;
1016 }
1017
1018 static u64 ixgbe_get_tx_pending(struct ixgbe_ring *ring)
1019 {
1020         unsigned int head, tail;
1021
1022         head = ring->next_to_clean;
1023         tail = ring->next_to_use;
1024
1025         return ((head <= tail) ? tail : tail + ring->count) - head;
1026 }
1027
1028 static inline bool ixgbe_check_tx_hang(struct ixgbe_ring *tx_ring)
1029 {
1030         u32 tx_done = ixgbe_get_tx_completed(tx_ring);
1031         u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
1032         u32 tx_pending = ixgbe_get_tx_pending(tx_ring);
1033
1034         clear_check_for_tx_hang(tx_ring);
1035
1036         /*
1037          * Check for a hung queue, but be thorough. This verifies
1038          * that a transmit has been completed since the previous
1039          * check AND there is at least one packet pending. The
1040          * ARMED bit is set to indicate a potential hang. The
1041          * bit is cleared if a pause frame is received to remove
1042          * false hang detection due to PFC or 802.3x frames. By
1043          * requiring this to fail twice we avoid races with
1044          * pfc clearing the ARMED bit and conditions where we
1045          * run the check_tx_hang logic with a transmit completion
1046          * pending but without time to complete it yet.
1047          */
1048         if (tx_done_old == tx_done && tx_pending)
1049                 /* make sure it is true for two checks in a row */
1050                 return test_and_set_bit(__IXGBE_HANG_CHECK_ARMED,
1051                                         &tx_ring->state);
1052         /* update completed stats and continue */
1053         tx_ring->tx_stats.tx_done_old = tx_done;
1054         /* reset the countdown */
1055         clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
1056
1057         return false;
1058 }
1059
1060 /**
1061  * ixgbe_tx_timeout_reset - initiate reset due to Tx timeout
1062  * @adapter: driver private struct
1063  **/
1064 static void ixgbe_tx_timeout_reset(struct ixgbe_adapter *adapter)
1065 {
1066
1067         /* Do the reset outside of interrupt context */
1068         if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1069                 set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
1070                 e_warn(drv, "initiating reset due to tx timeout\n");
1071                 ixgbe_service_event_schedule(adapter);
1072         }
1073 }
1074
1075 /**
1076  * ixgbe_tx_maxrate - callback to set the maximum per-queue bitrate
1077  * @netdev: network interface device structure
1078  * @queue_index: Tx queue to set
1079  * @maxrate: desired maximum transmit bitrate
1080  **/
1081 static int ixgbe_tx_maxrate(struct net_device *netdev,
1082                             int queue_index, u32 maxrate)
1083 {
1084         struct ixgbe_adapter *adapter = netdev_priv(netdev);
1085         struct ixgbe_hw *hw = &adapter->hw;
1086         u32 bcnrc_val = ixgbe_link_mbps(adapter);
1087
1088         if (!maxrate)
1089                 return 0;
1090
1091         /* Calculate the rate factor values to set */
1092         bcnrc_val <<= IXGBE_RTTBCNRC_RF_INT_SHIFT;
1093         bcnrc_val /= maxrate;
1094
1095         /* clear everything but the rate factor */
1096         bcnrc_val &= IXGBE_RTTBCNRC_RF_INT_MASK |
1097         IXGBE_RTTBCNRC_RF_DEC_MASK;
1098
1099         /* enable the rate scheduler */
1100         bcnrc_val |= IXGBE_RTTBCNRC_RS_ENA;
1101
1102         IXGBE_WRITE_REG(hw, IXGBE_RTTDQSEL, queue_index);
1103         IXGBE_WRITE_REG(hw, IXGBE_RTTBCNRC, bcnrc_val);
1104
1105         return 0;
1106 }
1107
1108 /**
1109  * ixgbe_update_tx_ring_stats - Update Tx ring specific counters
1110  * @tx_ring: ring to update
1111  * @q_vector: queue vector ring belongs to
1112  * @pkts: number of processed packets
1113  * @bytes: number of processed bytes
1114  */
1115 void ixgbe_update_tx_ring_stats(struct ixgbe_ring *tx_ring,
1116                                 struct ixgbe_q_vector *q_vector, u64 pkts,
1117                                 u64 bytes)
1118 {
1119         u64_stats_update_begin(&tx_ring->syncp);
1120         tx_ring->stats.bytes += bytes;
1121         tx_ring->stats.packets += pkts;
1122         u64_stats_update_end(&tx_ring->syncp);
1123         q_vector->tx.total_bytes += bytes;
1124         q_vector->tx.total_packets += pkts;
1125 }
1126
1127 /**
1128  * ixgbe_update_rx_ring_stats - Update Rx ring specific counters
1129  * @rx_ring: ring to update
1130  * @q_vector: queue vector ring belongs to
1131  * @pkts: number of processed packets
1132  * @bytes: number of processed bytes
1133  */
1134 void ixgbe_update_rx_ring_stats(struct ixgbe_ring *rx_ring,
1135                                 struct ixgbe_q_vector *q_vector, u64 pkts,
1136                                 u64 bytes)
1137 {
1138         u64_stats_update_begin(&rx_ring->syncp);
1139         rx_ring->stats.bytes += bytes;
1140         rx_ring->stats.packets += pkts;
1141         u64_stats_update_end(&rx_ring->syncp);
1142         q_vector->rx.total_bytes += bytes;
1143         q_vector->rx.total_packets += pkts;
1144 }
1145
1146 /**
1147  * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
1148  * @q_vector: structure containing interrupt and ring information
1149  * @tx_ring: tx ring to clean
1150  * @napi_budget: Used to determine if we are in netpoll
1151  **/
1152 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
1153                                struct ixgbe_ring *tx_ring, int napi_budget)
1154 {
1155         struct ixgbe_adapter *adapter = q_vector->adapter;
1156         struct ixgbe_tx_buffer *tx_buffer;
1157         union ixgbe_adv_tx_desc *tx_desc;
1158         unsigned int total_bytes = 0, total_packets = 0, total_ipsec = 0;
1159         unsigned int budget = q_vector->tx.work_limit;
1160         unsigned int i = tx_ring->next_to_clean;
1161         struct netdev_queue *txq;
1162
1163         if (test_bit(__IXGBE_DOWN, &adapter->state))
1164                 return true;
1165
1166         tx_buffer = &tx_ring->tx_buffer_info[i];
1167         tx_desc = IXGBE_TX_DESC(tx_ring, i);
1168         i -= tx_ring->count;
1169
1170         do {
1171                 union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
1172
1173                 /* if next_to_watch is not set then there is no work pending */
1174                 if (!eop_desc)
1175                         break;
1176
1177                 /* prevent any other reads prior to eop_desc */
1178                 smp_rmb();
1179
1180                 /* if DD is not set pending work has not been completed */
1181                 if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
1182                         break;
1183
1184                 /* clear next_to_watch to prevent false hangs */
1185                 tx_buffer->next_to_watch = NULL;
1186
1187                 /* update the statistics for this packet */
1188                 total_bytes += tx_buffer->bytecount;
1189                 total_packets += tx_buffer->gso_segs;
1190                 if (tx_buffer->tx_flags & IXGBE_TX_FLAGS_IPSEC)
1191                         total_ipsec++;
1192
1193                 /* free the skb */
1194                 if (ring_is_xdp(tx_ring))
1195                         xdp_return_frame(tx_buffer->xdpf);
1196                 else
1197                         napi_consume_skb(tx_buffer->skb, napi_budget);
1198
1199                 /* unmap skb header data */
1200                 dma_unmap_single(tx_ring->dev,
1201                                  dma_unmap_addr(tx_buffer, dma),
1202                                  dma_unmap_len(tx_buffer, len),
1203                                  DMA_TO_DEVICE);
1204
1205                 /* clear tx_buffer data */
1206                 dma_unmap_len_set(tx_buffer, len, 0);
1207
1208                 /* unmap remaining buffers */
1209                 while (tx_desc != eop_desc) {
1210                         tx_buffer++;
1211                         tx_desc++;
1212                         i++;
1213                         if (unlikely(!i)) {
1214                                 i -= tx_ring->count;
1215                                 tx_buffer = tx_ring->tx_buffer_info;
1216                                 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
1217                         }
1218
1219                         /* unmap any remaining paged data */
1220                         if (dma_unmap_len(tx_buffer, len)) {
1221                                 dma_unmap_page(tx_ring->dev,
1222                                                dma_unmap_addr(tx_buffer, dma),
1223                                                dma_unmap_len(tx_buffer, len),
1224                                                DMA_TO_DEVICE);
1225                                 dma_unmap_len_set(tx_buffer, len, 0);
1226                         }
1227                 }
1228
1229                 /* move us one more past the eop_desc for start of next pkt */
1230                 tx_buffer++;
1231                 tx_desc++;
1232                 i++;
1233                 if (unlikely(!i)) {
1234                         i -= tx_ring->count;
1235                         tx_buffer = tx_ring->tx_buffer_info;
1236                         tx_desc = IXGBE_TX_DESC(tx_ring, 0);
1237                 }
1238
1239                 /* issue prefetch for next Tx descriptor */
1240                 prefetch(tx_desc);
1241
1242                 /* update budget accounting */
1243                 budget--;
1244         } while (likely(budget));
1245
1246         i += tx_ring->count;
1247         tx_ring->next_to_clean = i;
1248         ixgbe_update_tx_ring_stats(tx_ring, q_vector, total_packets,
1249                                    total_bytes);
1250         adapter->tx_ipsec += total_ipsec;
1251
1252         if (check_for_tx_hang(tx_ring) && ixgbe_check_tx_hang(tx_ring)) {
1253                 /* schedule immediate reset if we believe we hung */
1254                 struct ixgbe_hw *hw = &adapter->hw;
1255                 e_err(drv, "Detected Tx Unit Hang %s\n"
1256                         "  Tx Queue             <%d>\n"
1257                         "  TDH, TDT             <%x>, <%x>\n"
1258                         "  next_to_use          <%x>\n"
1259                         "  next_to_clean        <%x>\n"
1260                         "tx_buffer_info[next_to_clean]\n"
1261                         "  time_stamp           <%lx>\n"
1262                         "  jiffies              <%lx>\n",
1263                         ring_is_xdp(tx_ring) ? "(XDP)" : "",
1264                         tx_ring->queue_index,
1265                         IXGBE_READ_REG(hw, IXGBE_TDH(tx_ring->reg_idx)),
1266                         IXGBE_READ_REG(hw, IXGBE_TDT(tx_ring->reg_idx)),
1267                         tx_ring->next_to_use, i,
1268                         tx_ring->tx_buffer_info[i].time_stamp, jiffies);
1269
1270                 if (!ring_is_xdp(tx_ring))
1271                         netif_stop_subqueue(tx_ring->netdev,
1272                                             tx_ring->queue_index);
1273
1274                 e_info(probe,
1275                        "tx hang %d detected on queue %d, resetting adapter\n",
1276                         adapter->tx_timeout_count + 1, tx_ring->queue_index);
1277
1278                 /* schedule immediate reset if we believe we hung */
1279                 ixgbe_tx_timeout_reset(adapter);
1280
1281                 /* the adapter is about to reset, no point in enabling stuff */
1282                 return true;
1283         }
1284
1285         if (ring_is_xdp(tx_ring))
1286                 return !!budget;
1287
1288 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
1289         txq = netdev_get_tx_queue(tx_ring->netdev, tx_ring->queue_index);
1290         if (!__netif_txq_completed_wake(txq, total_packets, total_bytes,
1291                                         ixgbe_desc_unused(tx_ring),
1292                                         TX_WAKE_THRESHOLD,
1293                                         !netif_carrier_ok(tx_ring->netdev) ||
1294                                         test_bit(__IXGBE_DOWN, &adapter->state)))
1295                 ++tx_ring->tx_stats.restart_queue;
1296
1297         return !!budget;
1298 }
1299
1300 #ifdef CONFIG_IXGBE_DCA
1301 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
1302                                 struct ixgbe_ring *tx_ring,
1303                                 int cpu)
1304 {
1305         struct ixgbe_hw *hw = &adapter->hw;
1306         u32 txctrl = 0;
1307         u16 reg_offset;
1308
1309         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1310                 txctrl = dca3_get_tag(tx_ring->dev, cpu);
1311
1312         switch (hw->mac.type) {
1313         case ixgbe_mac_82598EB:
1314                 reg_offset = IXGBE_DCA_TXCTRL(tx_ring->reg_idx);
1315                 break;
1316         case ixgbe_mac_82599EB:
1317         case ixgbe_mac_X540:
1318                 reg_offset = IXGBE_DCA_TXCTRL_82599(tx_ring->reg_idx);
1319                 txctrl <<= IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599;
1320                 break;
1321         default:
1322                 /* for unknown hardware do not write register */
1323                 return;
1324         }
1325
1326         /*
1327          * We can enable relaxed ordering for reads, but not writes when
1328          * DCA is enabled.  This is due to a known issue in some chipsets
1329          * which will cause the DCA tag to be cleared.
1330          */
1331         txctrl |= IXGBE_DCA_TXCTRL_DESC_RRO_EN |
1332                   IXGBE_DCA_TXCTRL_DATA_RRO_EN |
1333                   IXGBE_DCA_TXCTRL_DESC_DCA_EN;
1334
1335         IXGBE_WRITE_REG(hw, reg_offset, txctrl);
1336 }
1337
1338 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
1339                                 struct ixgbe_ring *rx_ring,
1340                                 int cpu)
1341 {
1342         struct ixgbe_hw *hw = &adapter->hw;
1343         u32 rxctrl = 0;
1344         u8 reg_idx = rx_ring->reg_idx;
1345
1346         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1347                 rxctrl = dca3_get_tag(rx_ring->dev, cpu);
1348
1349         switch (hw->mac.type) {
1350         case ixgbe_mac_82599EB:
1351         case ixgbe_mac_X540:
1352                 rxctrl <<= IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599;
1353                 break;
1354         default:
1355                 break;
1356         }
1357
1358         /*
1359          * We can enable relaxed ordering for reads, but not writes when
1360          * DCA is enabled.  This is due to a known issue in some chipsets
1361          * which will cause the DCA tag to be cleared.
1362          */
1363         rxctrl |= IXGBE_DCA_RXCTRL_DESC_RRO_EN |
1364                   IXGBE_DCA_RXCTRL_DATA_DCA_EN |
1365                   IXGBE_DCA_RXCTRL_DESC_DCA_EN;
1366
1367         IXGBE_WRITE_REG(hw, IXGBE_DCA_RXCTRL(reg_idx), rxctrl);
1368 }
1369
1370 static void ixgbe_update_dca(struct ixgbe_q_vector *q_vector)
1371 {
1372         struct ixgbe_adapter *adapter = q_vector->adapter;
1373         struct ixgbe_ring *ring;
1374         int cpu = get_cpu();
1375
1376         if (q_vector->cpu == cpu)
1377                 goto out_no_update;
1378
1379         ixgbe_for_each_ring(ring, q_vector->tx)
1380                 ixgbe_update_tx_dca(adapter, ring, cpu);
1381
1382         ixgbe_for_each_ring(ring, q_vector->rx)
1383                 ixgbe_update_rx_dca(adapter, ring, cpu);
1384
1385         q_vector->cpu = cpu;
1386 out_no_update:
1387         put_cpu();
1388 }
1389
1390 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
1391 {
1392         int i;
1393
1394         /* always use CB2 mode, difference is masked in the CB driver */
1395         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1396                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1397                                 IXGBE_DCA_CTRL_DCA_MODE_CB2);
1398         else
1399                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1400                                 IXGBE_DCA_CTRL_DCA_DISABLE);
1401
1402         for (i = 0; i < adapter->num_q_vectors; i++) {
1403                 adapter->q_vector[i]->cpu = -1;
1404                 ixgbe_update_dca(adapter->q_vector[i]);
1405         }
1406 }
1407
1408 static int __ixgbe_notify_dca(struct device *dev, void *data)
1409 {
1410         struct ixgbe_adapter *adapter = dev_get_drvdata(dev);
1411         unsigned long event = *(unsigned long *)data;
1412
1413         if (!(adapter->flags & IXGBE_FLAG_DCA_CAPABLE))
1414                 return 0;
1415
1416         switch (event) {
1417         case DCA_PROVIDER_ADD:
1418                 /* if we're already enabled, don't do it again */
1419                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1420                         break;
1421                 if (dca_add_requester(dev) == 0) {
1422                         adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
1423                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1424                                         IXGBE_DCA_CTRL_DCA_MODE_CB2);
1425                         break;
1426                 }
1427                 fallthrough; /* DCA is disabled. */
1428         case DCA_PROVIDER_REMOVE:
1429                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
1430                         dca_remove_requester(dev);
1431                         adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
1432                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1433                                         IXGBE_DCA_CTRL_DCA_DISABLE);
1434                 }
1435                 break;
1436         }
1437
1438         return 0;
1439 }
1440
1441 #endif /* CONFIG_IXGBE_DCA */
1442
1443 #define IXGBE_RSS_L4_TYPES_MASK \
1444         ((1ul << IXGBE_RXDADV_RSSTYPE_IPV4_TCP) | \
1445          (1ul << IXGBE_RXDADV_RSSTYPE_IPV4_UDP) | \
1446          (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_TCP) | \
1447          (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_UDP))
1448
1449 static inline void ixgbe_rx_hash(struct ixgbe_ring *ring,
1450                                  union ixgbe_adv_rx_desc *rx_desc,
1451                                  struct sk_buff *skb)
1452 {
1453         u16 rss_type;
1454
1455         if (!(ring->netdev->features & NETIF_F_RXHASH))
1456                 return;
1457
1458         rss_type = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.pkt_info) &
1459                    IXGBE_RXDADV_RSSTYPE_MASK;
1460
1461         if (!rss_type)
1462                 return;
1463
1464         skb_set_hash(skb, le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
1465                      (IXGBE_RSS_L4_TYPES_MASK & (1ul << rss_type)) ?
1466                      PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3);
1467 }
1468
1469 #ifdef IXGBE_FCOE
1470 /**
1471  * ixgbe_rx_is_fcoe - check the rx desc for incoming pkt type
1472  * @ring: structure containing ring specific data
1473  * @rx_desc: advanced rx descriptor
1474  *
1475  * Returns : true if it is FCoE pkt
1476  */
1477 static inline bool ixgbe_rx_is_fcoe(struct ixgbe_ring *ring,
1478                                     union ixgbe_adv_rx_desc *rx_desc)
1479 {
1480         __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1481
1482         return test_bit(__IXGBE_RX_FCOE, &ring->state) &&
1483                ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_ETQF_MASK)) ==
1484                 (cpu_to_le16(IXGBE_ETQF_FILTER_FCOE <<
1485                              IXGBE_RXDADV_PKTTYPE_ETQF_SHIFT)));
1486 }
1487
1488 #endif /* IXGBE_FCOE */
1489 /**
1490  * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
1491  * @ring: structure containing ring specific data
1492  * @rx_desc: current Rx descriptor being processed
1493  * @skb: skb currently being received and modified
1494  **/
1495 static inline void ixgbe_rx_checksum(struct ixgbe_ring *ring,
1496                                      union ixgbe_adv_rx_desc *rx_desc,
1497                                      struct sk_buff *skb)
1498 {
1499         __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1500         bool encap_pkt = false;
1501
1502         skb_checksum_none_assert(skb);
1503
1504         /* Rx csum disabled */
1505         if (!(ring->netdev->features & NETIF_F_RXCSUM))
1506                 return;
1507
1508         /* check for VXLAN and Geneve packets */
1509         if (pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_VXLAN)) {
1510                 encap_pkt = true;
1511                 skb->encapsulation = 1;
1512         }
1513
1514         /* if IP and error */
1515         if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
1516             ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
1517                 ring->rx_stats.csum_err++;
1518                 return;
1519         }
1520
1521         if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
1522                 return;
1523
1524         if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
1525                 /*
1526                  * 82599 errata, UDP frames with a 0 checksum can be marked as
1527                  * checksum errors.
1528                  */
1529                 if ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_UDP)) &&
1530                     test_bit(__IXGBE_RX_CSUM_UDP_ZERO_ERR, &ring->state))
1531                         return;
1532
1533                 ring->rx_stats.csum_err++;
1534                 return;
1535         }
1536
1537         /* It must be a TCP or UDP packet with a valid checksum */
1538         skb->ip_summed = CHECKSUM_UNNECESSARY;
1539         if (encap_pkt) {
1540                 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_OUTERIPCS))
1541                         return;
1542
1543                 if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_OUTERIPER)) {
1544                         skb->ip_summed = CHECKSUM_NONE;
1545                         return;
1546                 }
1547                 /* If we checked the outer header let the stack know */
1548                 skb->csum_level = 1;
1549         }
1550 }
1551
1552 static unsigned int ixgbe_rx_offset(struct ixgbe_ring *rx_ring)
1553 {
1554         return ring_uses_build_skb(rx_ring) ? IXGBE_SKB_PAD : 0;
1555 }
1556
1557 static bool ixgbe_alloc_mapped_page(struct ixgbe_ring *rx_ring,
1558                                     struct ixgbe_rx_buffer *bi)
1559 {
1560         struct page *page = bi->page;
1561         dma_addr_t dma;
1562
1563         /* since we are recycling buffers we should seldom need to alloc */
1564         if (likely(page))
1565                 return true;
1566
1567         /* alloc new page for storage */
1568         page = dev_alloc_pages(ixgbe_rx_pg_order(rx_ring));
1569         if (unlikely(!page)) {
1570                 rx_ring->rx_stats.alloc_rx_page_failed++;
1571                 return false;
1572         }
1573
1574         /* map page for use */
1575         dma = dma_map_page_attrs(rx_ring->dev, page, 0,
1576                                  ixgbe_rx_pg_size(rx_ring),
1577                                  DMA_FROM_DEVICE,
1578                                  IXGBE_RX_DMA_ATTR);
1579
1580         /*
1581          * if mapping failed free memory back to system since
1582          * there isn't much point in holding memory we can't use
1583          */
1584         if (dma_mapping_error(rx_ring->dev, dma)) {
1585                 __free_pages(page, ixgbe_rx_pg_order(rx_ring));
1586
1587                 rx_ring->rx_stats.alloc_rx_page_failed++;
1588                 return false;
1589         }
1590
1591         bi->dma = dma;
1592         bi->page = page;
1593         bi->page_offset = rx_ring->rx_offset;
1594         page_ref_add(page, USHRT_MAX - 1);
1595         bi->pagecnt_bias = USHRT_MAX;
1596         rx_ring->rx_stats.alloc_rx_page++;
1597
1598         return true;
1599 }
1600
1601 /**
1602  * ixgbe_alloc_rx_buffers - Replace used receive buffers
1603  * @rx_ring: ring to place buffers on
1604  * @cleaned_count: number of buffers to replace
1605  **/
1606 void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
1607 {
1608         union ixgbe_adv_rx_desc *rx_desc;
1609         struct ixgbe_rx_buffer *bi;
1610         u16 i = rx_ring->next_to_use;
1611         u16 bufsz;
1612
1613         /* nothing to do */
1614         if (!cleaned_count)
1615                 return;
1616
1617         rx_desc = IXGBE_RX_DESC(rx_ring, i);
1618         bi = &rx_ring->rx_buffer_info[i];
1619         i -= rx_ring->count;
1620
1621         bufsz = ixgbe_rx_bufsz(rx_ring);
1622
1623         do {
1624                 if (!ixgbe_alloc_mapped_page(rx_ring, bi))
1625                         break;
1626
1627                 /* sync the buffer for use by the device */
1628                 dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
1629                                                  bi->page_offset, bufsz,
1630                                                  DMA_FROM_DEVICE);
1631
1632                 /*
1633                  * Refresh the desc even if buffer_addrs didn't change
1634                  * because each write-back erases this info.
1635                  */
1636                 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
1637
1638                 rx_desc++;
1639                 bi++;
1640                 i++;
1641                 if (unlikely(!i)) {
1642                         rx_desc = IXGBE_RX_DESC(rx_ring, 0);
1643                         bi = rx_ring->rx_buffer_info;
1644                         i -= rx_ring->count;
1645                 }
1646
1647                 /* clear the length for the next_to_use descriptor */
1648                 rx_desc->wb.upper.length = 0;
1649
1650                 cleaned_count--;
1651         } while (cleaned_count);
1652
1653         i += rx_ring->count;
1654
1655         if (rx_ring->next_to_use != i) {
1656                 rx_ring->next_to_use = i;
1657
1658                 /* update next to alloc since we have filled the ring */
1659                 rx_ring->next_to_alloc = i;
1660
1661                 /* Force memory writes to complete before letting h/w
1662                  * know there are new descriptors to fetch.  (Only
1663                  * applicable for weak-ordered memory model archs,
1664                  * such as IA-64).
1665                  */
1666                 wmb();
1667                 writel(i, rx_ring->tail);
1668         }
1669 }
1670
1671 static void ixgbe_set_rsc_gso_size(struct ixgbe_ring *ring,
1672                                    struct sk_buff *skb)
1673 {
1674         u16 hdr_len = skb_headlen(skb);
1675
1676         /* set gso_size to avoid messing up TCP MSS */
1677         skb_shinfo(skb)->gso_size = DIV_ROUND_UP((skb->len - hdr_len),
1678                                                  IXGBE_CB(skb)->append_cnt);
1679         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1680 }
1681
1682 static void ixgbe_update_rsc_stats(struct ixgbe_ring *rx_ring,
1683                                    struct sk_buff *skb)
1684 {
1685         /* if append_cnt is 0 then frame is not RSC */
1686         if (!IXGBE_CB(skb)->append_cnt)
1687                 return;
1688
1689         rx_ring->rx_stats.rsc_count += IXGBE_CB(skb)->append_cnt;
1690         rx_ring->rx_stats.rsc_flush++;
1691
1692         ixgbe_set_rsc_gso_size(rx_ring, skb);
1693
1694         /* gso_size is computed using append_cnt so always clear it last */
1695         IXGBE_CB(skb)->append_cnt = 0;
1696 }
1697
1698 /**
1699  * ixgbe_process_skb_fields - Populate skb header fields from Rx descriptor
1700  * @rx_ring: rx descriptor ring packet is being transacted on
1701  * @rx_desc: pointer to the EOP Rx descriptor
1702  * @skb: pointer to current skb being populated
1703  *
1704  * This function checks the ring, descriptor, and packet information in
1705  * order to populate the hash, checksum, VLAN, timestamp, protocol, and
1706  * other fields within the skb.
1707  **/
1708 void ixgbe_process_skb_fields(struct ixgbe_ring *rx_ring,
1709                               union ixgbe_adv_rx_desc *rx_desc,
1710                               struct sk_buff *skb)
1711 {
1712         struct net_device *dev = rx_ring->netdev;
1713         u32 flags = rx_ring->q_vector->adapter->flags;
1714
1715         ixgbe_update_rsc_stats(rx_ring, skb);
1716
1717         ixgbe_rx_hash(rx_ring, rx_desc, skb);
1718
1719         ixgbe_rx_checksum(rx_ring, rx_desc, skb);
1720
1721         if (unlikely(flags & IXGBE_FLAG_RX_HWTSTAMP_ENABLED))
1722                 ixgbe_ptp_rx_hwtstamp(rx_ring, rx_desc, skb);
1723
1724         if ((dev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
1725             ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
1726                 u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
1727                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
1728         }
1729
1730         if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_STAT_SECP))
1731                 ixgbe_ipsec_rx(rx_ring, rx_desc, skb);
1732
1733         /* record Rx queue, or update MACVLAN statistics */
1734         if (netif_is_ixgbe(dev))
1735                 skb_record_rx_queue(skb, rx_ring->queue_index);
1736         else
1737                 macvlan_count_rx(netdev_priv(dev), skb->len + ETH_HLEN, true,
1738                                  false);
1739
1740         skb->protocol = eth_type_trans(skb, dev);
1741 }
1742
1743 void ixgbe_rx_skb(struct ixgbe_q_vector *q_vector,
1744                   struct sk_buff *skb)
1745 {
1746         napi_gro_receive(&q_vector->napi, skb);
1747 }
1748
1749 /**
1750  * ixgbe_is_non_eop - process handling of non-EOP buffers
1751  * @rx_ring: Rx ring being processed
1752  * @rx_desc: Rx descriptor for current buffer
1753  * @skb: Current socket buffer containing buffer in progress
1754  *
1755  * This function updates next to clean.  If the buffer is an EOP buffer
1756  * this function exits returning false, otherwise it will place the
1757  * sk_buff in the next buffer to be chained and return true indicating
1758  * that this is in fact a non-EOP buffer.
1759  **/
1760 static bool ixgbe_is_non_eop(struct ixgbe_ring *rx_ring,
1761                              union ixgbe_adv_rx_desc *rx_desc,
1762                              struct sk_buff *skb)
1763 {
1764         u32 ntc = rx_ring->next_to_clean + 1;
1765
1766         /* fetch, update, and store next to clean */
1767         ntc = (ntc < rx_ring->count) ? ntc : 0;
1768         rx_ring->next_to_clean = ntc;
1769
1770         prefetch(IXGBE_RX_DESC(rx_ring, ntc));
1771
1772         /* update RSC append count if present */
1773         if (ring_is_rsc_enabled(rx_ring)) {
1774                 __le32 rsc_enabled = rx_desc->wb.lower.lo_dword.data &
1775                                      cpu_to_le32(IXGBE_RXDADV_RSCCNT_MASK);
1776
1777                 if (unlikely(rsc_enabled)) {
1778                         u32 rsc_cnt = le32_to_cpu(rsc_enabled);
1779
1780                         rsc_cnt >>= IXGBE_RXDADV_RSCCNT_SHIFT;
1781                         IXGBE_CB(skb)->append_cnt += rsc_cnt - 1;
1782
1783                         /* update ntc based on RSC value */
1784                         ntc = le32_to_cpu(rx_desc->wb.upper.status_error);
1785                         ntc &= IXGBE_RXDADV_NEXTP_MASK;
1786                         ntc >>= IXGBE_RXDADV_NEXTP_SHIFT;
1787                 }
1788         }
1789
1790         /* if we are the last buffer then there is nothing else to do */
1791         if (likely(ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
1792                 return false;
1793
1794         /* place skb in next buffer to be received */
1795         rx_ring->rx_buffer_info[ntc].skb = skb;
1796         rx_ring->rx_stats.non_eop_descs++;
1797
1798         return true;
1799 }
1800
1801 /**
1802  * ixgbe_pull_tail - ixgbe specific version of skb_pull_tail
1803  * @rx_ring: rx descriptor ring packet is being transacted on
1804  * @skb: pointer to current skb being adjusted
1805  *
1806  * This function is an ixgbe specific version of __pskb_pull_tail.  The
1807  * main difference between this version and the original function is that
1808  * this function can make several assumptions about the state of things
1809  * that allow for significant optimizations versus the standard function.
1810  * As a result we can do things like drop a frag and maintain an accurate
1811  * truesize for the skb.
1812  */
1813 static void ixgbe_pull_tail(struct ixgbe_ring *rx_ring,
1814                             struct sk_buff *skb)
1815 {
1816         skb_frag_t *frag = &skb_shinfo(skb)->frags[0];
1817         unsigned char *va;
1818         unsigned int pull_len;
1819
1820         /*
1821          * it is valid to use page_address instead of kmap since we are
1822          * working with pages allocated out of the lomem pool per
1823          * alloc_page(GFP_ATOMIC)
1824          */
1825         va = skb_frag_address(frag);
1826
1827         /*
1828          * we need the header to contain the greater of either ETH_HLEN or
1829          * 60 bytes if the skb->len is less than 60 for skb_pad.
1830          */
1831         pull_len = eth_get_headlen(skb->dev, va, IXGBE_RX_HDR_SIZE);
1832
1833         /* align pull length to size of long to optimize memcpy performance */
1834         skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
1835
1836         /* update all of the pointers */
1837         skb_frag_size_sub(frag, pull_len);
1838         skb_frag_off_add(frag, pull_len);
1839         skb->data_len -= pull_len;
1840         skb->tail += pull_len;
1841 }
1842
1843 /**
1844  * ixgbe_dma_sync_frag - perform DMA sync for first frag of SKB
1845  * @rx_ring: rx descriptor ring packet is being transacted on
1846  * @skb: pointer to current skb being updated
1847  *
1848  * This function provides a basic DMA sync up for the first fragment of an
1849  * skb.  The reason for doing this is that the first fragment cannot be
1850  * unmapped until we have reached the end of packet descriptor for a buffer
1851  * chain.
1852  */
1853 static void ixgbe_dma_sync_frag(struct ixgbe_ring *rx_ring,
1854                                 struct sk_buff *skb)
1855 {
1856         if (ring_uses_build_skb(rx_ring)) {
1857                 unsigned long mask = (unsigned long)ixgbe_rx_pg_size(rx_ring) - 1;
1858                 unsigned long offset = (unsigned long)(skb->data) & mask;
1859
1860                 dma_sync_single_range_for_cpu(rx_ring->dev,
1861                                               IXGBE_CB(skb)->dma,
1862                                               offset,
1863                                               skb_headlen(skb),
1864                                               DMA_FROM_DEVICE);
1865         } else {
1866                 skb_frag_t *frag = &skb_shinfo(skb)->frags[0];
1867
1868                 dma_sync_single_range_for_cpu(rx_ring->dev,
1869                                               IXGBE_CB(skb)->dma,
1870                                               skb_frag_off(frag),
1871                                               skb_frag_size(frag),
1872                                               DMA_FROM_DEVICE);
1873         }
1874
1875         /* If the page was released, just unmap it. */
1876         if (unlikely(IXGBE_CB(skb)->page_released)) {
1877                 dma_unmap_page_attrs(rx_ring->dev, IXGBE_CB(skb)->dma,
1878                                      ixgbe_rx_pg_size(rx_ring),
1879                                      DMA_FROM_DEVICE,
1880                                      IXGBE_RX_DMA_ATTR);
1881         }
1882 }
1883
1884 /**
1885  * ixgbe_cleanup_headers - Correct corrupted or empty headers
1886  * @rx_ring: rx descriptor ring packet is being transacted on
1887  * @rx_desc: pointer to the EOP Rx descriptor
1888  * @skb: pointer to current skb being fixed
1889  *
1890  * Check if the skb is valid in the XDP case it will be an error pointer.
1891  * Return true in this case to abort processing and advance to next
1892  * descriptor.
1893  *
1894  * Check for corrupted packet headers caused by senders on the local L2
1895  * embedded NIC switch not setting up their Tx Descriptors right.  These
1896  * should be very rare.
1897  *
1898  * Also address the case where we are pulling data in on pages only
1899  * and as such no data is present in the skb header.
1900  *
1901  * In addition if skb is not at least 60 bytes we need to pad it so that
1902  * it is large enough to qualify as a valid Ethernet frame.
1903  *
1904  * Returns true if an error was encountered and skb was freed.
1905  **/
1906 bool ixgbe_cleanup_headers(struct ixgbe_ring *rx_ring,
1907                            union ixgbe_adv_rx_desc *rx_desc,
1908                            struct sk_buff *skb)
1909 {
1910         struct net_device *netdev = rx_ring->netdev;
1911
1912         /* XDP packets use error pointer so abort at this point */
1913         if (IS_ERR(skb))
1914                 return true;
1915
1916         /* Verify netdev is present, and that packet does not have any
1917          * errors that would be unacceptable to the netdev.
1918          */
1919         if (!netdev ||
1920             (unlikely(ixgbe_test_staterr(rx_desc,
1921                                          IXGBE_RXDADV_ERR_FRAME_ERR_MASK) &&
1922              !(netdev->features & NETIF_F_RXALL)))) {
1923                 dev_kfree_skb_any(skb);
1924                 return true;
1925         }
1926
1927         /* place header in linear portion of buffer */
1928         if (!skb_headlen(skb))
1929                 ixgbe_pull_tail(rx_ring, skb);
1930
1931 #ifdef IXGBE_FCOE
1932         /* do not attempt to pad FCoE Frames as this will disrupt DDP */
1933         if (ixgbe_rx_is_fcoe(rx_ring, rx_desc))
1934                 return false;
1935
1936 #endif
1937         /* if eth_skb_pad returns an error the skb was freed */
1938         if (eth_skb_pad(skb))
1939                 return true;
1940
1941         return false;
1942 }
1943
1944 /**
1945  * ixgbe_reuse_rx_page - page flip buffer and store it back on the ring
1946  * @rx_ring: rx descriptor ring to store buffers on
1947  * @old_buff: donor buffer to have page reused
1948  *
1949  * Synchronizes page for reuse by the adapter
1950  **/
1951 static void ixgbe_reuse_rx_page(struct ixgbe_ring *rx_ring,
1952                                 struct ixgbe_rx_buffer *old_buff)
1953 {
1954         struct ixgbe_rx_buffer *new_buff;
1955         u16 nta = rx_ring->next_to_alloc;
1956
1957         new_buff = &rx_ring->rx_buffer_info[nta];
1958
1959         /* update, and store next to alloc */
1960         nta++;
1961         rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
1962
1963         /* Transfer page from old buffer to new buffer.
1964          * Move each member individually to avoid possible store
1965          * forwarding stalls and unnecessary copy of skb.
1966          */
1967         new_buff->dma           = old_buff->dma;
1968         new_buff->page          = old_buff->page;
1969         new_buff->page_offset   = old_buff->page_offset;
1970         new_buff->pagecnt_bias  = old_buff->pagecnt_bias;
1971 }
1972
1973 static bool ixgbe_can_reuse_rx_page(struct ixgbe_rx_buffer *rx_buffer,
1974                                     int rx_buffer_pgcnt)
1975 {
1976         unsigned int pagecnt_bias = rx_buffer->pagecnt_bias;
1977         struct page *page = rx_buffer->page;
1978
1979         /* avoid re-using remote and pfmemalloc pages */
1980         if (!dev_page_is_reusable(page))
1981                 return false;
1982
1983 #if (PAGE_SIZE < 8192)
1984         /* if we are only owner of page we can reuse it */
1985         if (unlikely((rx_buffer_pgcnt - pagecnt_bias) > 1))
1986                 return false;
1987 #else
1988         /* The last offset is a bit aggressive in that we assume the
1989          * worst case of FCoE being enabled and using a 3K buffer.
1990          * However this should have minimal impact as the 1K extra is
1991          * still less than one buffer in size.
1992          */
1993 #define IXGBE_LAST_OFFSET \
1994         (SKB_WITH_OVERHEAD(PAGE_SIZE) - IXGBE_RXBUFFER_3K)
1995         if (rx_buffer->page_offset > IXGBE_LAST_OFFSET)
1996                 return false;
1997 #endif
1998
1999         /* If we have drained the page fragment pool we need to update
2000          * the pagecnt_bias and page count so that we fully restock the
2001          * number of references the driver holds.
2002          */
2003         if (unlikely(pagecnt_bias == 1)) {
2004                 page_ref_add(page, USHRT_MAX - 1);
2005                 rx_buffer->pagecnt_bias = USHRT_MAX;
2006         }
2007
2008         return true;
2009 }
2010
2011 /**
2012  * ixgbe_add_rx_frag - Add contents of Rx buffer to sk_buff
2013  * @rx_ring: rx descriptor ring to transact packets on
2014  * @rx_buffer: buffer containing page to add
2015  * @skb: sk_buff to place the data into
2016  * @size: size of data in rx_buffer
2017  *
2018  * This function will add the data contained in rx_buffer->page to the skb.
2019  * This is done either through a direct copy if the data in the buffer is
2020  * less than the skb header size, otherwise it will just attach the page as
2021  * a frag to the skb.
2022  *
2023  * The function will then update the page offset if necessary and return
2024  * true if the buffer can be reused by the adapter.
2025  **/
2026 static void ixgbe_add_rx_frag(struct ixgbe_ring *rx_ring,
2027                               struct ixgbe_rx_buffer *rx_buffer,
2028                               struct sk_buff *skb,
2029                               unsigned int size)
2030 {
2031 #if (PAGE_SIZE < 8192)
2032         unsigned int truesize = ixgbe_rx_pg_size(rx_ring) / 2;
2033 #else
2034         unsigned int truesize = rx_ring->rx_offset ?
2035                                 SKB_DATA_ALIGN(rx_ring->rx_offset + size) :
2036                                 SKB_DATA_ALIGN(size);
2037 #endif
2038         skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
2039                         rx_buffer->page_offset, size, truesize);
2040 #if (PAGE_SIZE < 8192)
2041         rx_buffer->page_offset ^= truesize;
2042 #else
2043         rx_buffer->page_offset += truesize;
2044 #endif
2045 }
2046
2047 static struct ixgbe_rx_buffer *ixgbe_get_rx_buffer(struct ixgbe_ring *rx_ring,
2048                                                    union ixgbe_adv_rx_desc *rx_desc,
2049                                                    struct sk_buff **skb,
2050                                                    const unsigned int size,
2051                                                    int *rx_buffer_pgcnt)
2052 {
2053         struct ixgbe_rx_buffer *rx_buffer;
2054
2055         rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
2056         *rx_buffer_pgcnt =
2057 #if (PAGE_SIZE < 8192)
2058                 page_count(rx_buffer->page);
2059 #else
2060                 0;
2061 #endif
2062         prefetchw(rx_buffer->page);
2063         *skb = rx_buffer->skb;
2064
2065         /* Delay unmapping of the first packet. It carries the header
2066          * information, HW may still access the header after the writeback.
2067          * Only unmap it when EOP is reached
2068          */
2069         if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)) {
2070                 if (!*skb)
2071                         goto skip_sync;
2072         } else {
2073                 if (*skb)
2074                         ixgbe_dma_sync_frag(rx_ring, *skb);
2075         }
2076
2077         /* we are reusing so sync this buffer for CPU use */
2078         dma_sync_single_range_for_cpu(rx_ring->dev,
2079                                       rx_buffer->dma,
2080                                       rx_buffer->page_offset,
2081                                       size,
2082                                       DMA_FROM_DEVICE);
2083 skip_sync:
2084         rx_buffer->pagecnt_bias--;
2085
2086         return rx_buffer;
2087 }
2088
2089 static void ixgbe_put_rx_buffer(struct ixgbe_ring *rx_ring,
2090                                 struct ixgbe_rx_buffer *rx_buffer,
2091                                 struct sk_buff *skb,
2092                                 int rx_buffer_pgcnt)
2093 {
2094         if (ixgbe_can_reuse_rx_page(rx_buffer, rx_buffer_pgcnt)) {
2095                 /* hand second half of page back to the ring */
2096                 ixgbe_reuse_rx_page(rx_ring, rx_buffer);
2097         } else {
2098                 if (!IS_ERR(skb) && IXGBE_CB(skb)->dma == rx_buffer->dma) {
2099                         /* the page has been released from the ring */
2100                         IXGBE_CB(skb)->page_released = true;
2101                 } else {
2102                         /* we are not reusing the buffer so unmap it */
2103                         dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
2104                                              ixgbe_rx_pg_size(rx_ring),
2105                                              DMA_FROM_DEVICE,
2106                                              IXGBE_RX_DMA_ATTR);
2107                 }
2108                 __page_frag_cache_drain(rx_buffer->page,
2109                                         rx_buffer->pagecnt_bias);
2110         }
2111
2112         /* clear contents of rx_buffer */
2113         rx_buffer->page = NULL;
2114         rx_buffer->skb = NULL;
2115 }
2116
2117 static struct sk_buff *ixgbe_construct_skb(struct ixgbe_ring *rx_ring,
2118                                            struct ixgbe_rx_buffer *rx_buffer,
2119                                            struct xdp_buff *xdp,
2120                                            union ixgbe_adv_rx_desc *rx_desc)
2121 {
2122         unsigned int size = xdp->data_end - xdp->data;
2123 #if (PAGE_SIZE < 8192)
2124         unsigned int truesize = ixgbe_rx_pg_size(rx_ring) / 2;
2125 #else
2126         unsigned int truesize = SKB_DATA_ALIGN(xdp->data_end -
2127                                                xdp->data_hard_start);
2128 #endif
2129         struct sk_buff *skb;
2130
2131         /* prefetch first cache line of first page */
2132         net_prefetch(xdp->data);
2133
2134         /* Note, we get here by enabling legacy-rx via:
2135          *
2136          *    ethtool --set-priv-flags <dev> legacy-rx on
2137          *
2138          * In this mode, we currently get 0 extra XDP headroom as
2139          * opposed to having legacy-rx off, where we process XDP
2140          * packets going to stack via ixgbe_build_skb(). The latter
2141          * provides us currently with 192 bytes of headroom.
2142          *
2143          * For ixgbe_construct_skb() mode it means that the
2144          * xdp->data_meta will always point to xdp->data, since
2145          * the helper cannot expand the head. Should this ever
2146          * change in future for legacy-rx mode on, then lets also
2147          * add xdp->data_meta handling here.
2148          */
2149
2150         /* allocate a skb to store the frags */
2151         skb = napi_alloc_skb(&rx_ring->q_vector->napi, IXGBE_RX_HDR_SIZE);
2152         if (unlikely(!skb))
2153                 return NULL;
2154
2155         if (size > IXGBE_RX_HDR_SIZE) {
2156                 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP))
2157                         IXGBE_CB(skb)->dma = rx_buffer->dma;
2158
2159                 skb_add_rx_frag(skb, 0, rx_buffer->page,
2160                                 xdp->data - page_address(rx_buffer->page),
2161                                 size, truesize);
2162 #if (PAGE_SIZE < 8192)
2163                 rx_buffer->page_offset ^= truesize;
2164 #else
2165                 rx_buffer->page_offset += truesize;
2166 #endif
2167         } else {
2168                 memcpy(__skb_put(skb, size),
2169                        xdp->data, ALIGN(size, sizeof(long)));
2170                 rx_buffer->pagecnt_bias++;
2171         }
2172
2173         return skb;
2174 }
2175
2176 static struct sk_buff *ixgbe_build_skb(struct ixgbe_ring *rx_ring,
2177                                        struct ixgbe_rx_buffer *rx_buffer,
2178                                        struct xdp_buff *xdp,
2179                                        union ixgbe_adv_rx_desc *rx_desc)
2180 {
2181         unsigned int metasize = xdp->data - xdp->data_meta;
2182 #if (PAGE_SIZE < 8192)
2183         unsigned int truesize = ixgbe_rx_pg_size(rx_ring) / 2;
2184 #else
2185         unsigned int truesize = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) +
2186                                 SKB_DATA_ALIGN(xdp->data_end -
2187                                                xdp->data_hard_start);
2188 #endif
2189         struct sk_buff *skb;
2190
2191         /* Prefetch first cache line of first page. If xdp->data_meta
2192          * is unused, this points extactly as xdp->data, otherwise we
2193          * likely have a consumer accessing first few bytes of meta
2194          * data, and then actual data.
2195          */
2196         net_prefetch(xdp->data_meta);
2197
2198         /* build an skb to around the page buffer */
2199         skb = napi_build_skb(xdp->data_hard_start, truesize);
2200         if (unlikely(!skb))
2201                 return NULL;
2202
2203         /* update pointers within the skb to store the data */
2204         skb_reserve(skb, xdp->data - xdp->data_hard_start);
2205         __skb_put(skb, xdp->data_end - xdp->data);
2206         if (metasize)
2207                 skb_metadata_set(skb, metasize);
2208
2209         /* record DMA address if this is the start of a chain of buffers */
2210         if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP))
2211                 IXGBE_CB(skb)->dma = rx_buffer->dma;
2212
2213         /* update buffer offset */
2214 #if (PAGE_SIZE < 8192)
2215         rx_buffer->page_offset ^= truesize;
2216 #else
2217         rx_buffer->page_offset += truesize;
2218 #endif
2219
2220         return skb;
2221 }
2222
2223 static struct sk_buff *ixgbe_run_xdp(struct ixgbe_adapter *adapter,
2224                                      struct ixgbe_ring *rx_ring,
2225                                      struct xdp_buff *xdp)
2226 {
2227         int err, result = IXGBE_XDP_PASS;
2228         struct bpf_prog *xdp_prog;
2229         struct ixgbe_ring *ring;
2230         struct xdp_frame *xdpf;
2231         u32 act;
2232
2233         xdp_prog = READ_ONCE(rx_ring->xdp_prog);
2234
2235         if (!xdp_prog)
2236                 goto xdp_out;
2237
2238         prefetchw(xdp->data_hard_start); /* xdp_frame write */
2239
2240         act = bpf_prog_run_xdp(xdp_prog, xdp);
2241         switch (act) {
2242         case XDP_PASS:
2243                 break;
2244         case XDP_TX:
2245                 xdpf = xdp_convert_buff_to_frame(xdp);
2246                 if (unlikely(!xdpf))
2247                         goto out_failure;
2248                 ring = ixgbe_determine_xdp_ring(adapter);
2249                 if (static_branch_unlikely(&ixgbe_xdp_locking_key))
2250                         spin_lock(&ring->tx_lock);
2251                 result = ixgbe_xmit_xdp_ring(ring, xdpf);
2252                 if (static_branch_unlikely(&ixgbe_xdp_locking_key))
2253                         spin_unlock(&ring->tx_lock);
2254                 if (result == IXGBE_XDP_CONSUMED)
2255                         goto out_failure;
2256                 break;
2257         case XDP_REDIRECT:
2258                 err = xdp_do_redirect(adapter->netdev, xdp, xdp_prog);
2259                 if (err)
2260                         goto out_failure;
2261                 result = IXGBE_XDP_REDIR;
2262                 break;
2263         default:
2264                 bpf_warn_invalid_xdp_action(rx_ring->netdev, xdp_prog, act);
2265                 fallthrough;
2266         case XDP_ABORTED:
2267 out_failure:
2268                 trace_xdp_exception(rx_ring->netdev, xdp_prog, act);
2269                 fallthrough; /* handle aborts by dropping packet */
2270         case XDP_DROP:
2271                 result = IXGBE_XDP_CONSUMED;
2272                 break;
2273         }
2274 xdp_out:
2275         return ERR_PTR(-result);
2276 }
2277
2278 static unsigned int ixgbe_rx_frame_truesize(struct ixgbe_ring *rx_ring,
2279                                             unsigned int size)
2280 {
2281         unsigned int truesize;
2282
2283 #if (PAGE_SIZE < 8192)
2284         truesize = ixgbe_rx_pg_size(rx_ring) / 2; /* Must be power-of-2 */
2285 #else
2286         truesize = rx_ring->rx_offset ?
2287                 SKB_DATA_ALIGN(rx_ring->rx_offset + size) +
2288                 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) :
2289                 SKB_DATA_ALIGN(size);
2290 #endif
2291         return truesize;
2292 }
2293
2294 static void ixgbe_rx_buffer_flip(struct ixgbe_ring *rx_ring,
2295                                  struct ixgbe_rx_buffer *rx_buffer,
2296                                  unsigned int size)
2297 {
2298         unsigned int truesize = ixgbe_rx_frame_truesize(rx_ring, size);
2299 #if (PAGE_SIZE < 8192)
2300         rx_buffer->page_offset ^= truesize;
2301 #else
2302         rx_buffer->page_offset += truesize;
2303 #endif
2304 }
2305
2306 /**
2307  * ixgbe_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
2308  * @q_vector: structure containing interrupt and ring information
2309  * @rx_ring: rx descriptor ring to transact packets on
2310  * @budget: Total limit on number of packets to process
2311  *
2312  * This function provides a "bounce buffer" approach to Rx interrupt
2313  * processing.  The advantage to this is that on systems that have
2314  * expensive overhead for IOMMU access this provides a means of avoiding
2315  * it by maintaining the mapping of the page to the syste.
2316  *
2317  * Returns amount of work completed
2318  **/
2319 static int ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
2320                                struct ixgbe_ring *rx_ring,
2321                                const int budget)
2322 {
2323         unsigned int total_rx_bytes = 0, total_rx_packets = 0, frame_sz = 0;
2324         struct ixgbe_adapter *adapter = q_vector->adapter;
2325 #ifdef IXGBE_FCOE
2326         int ddp_bytes;
2327         unsigned int mss = 0;
2328 #endif /* IXGBE_FCOE */
2329         u16 cleaned_count = ixgbe_desc_unused(rx_ring);
2330         unsigned int offset = rx_ring->rx_offset;
2331         unsigned int xdp_xmit = 0;
2332         struct xdp_buff xdp;
2333
2334         /* Frame size depend on rx_ring setup when PAGE_SIZE=4K */
2335 #if (PAGE_SIZE < 8192)
2336         frame_sz = ixgbe_rx_frame_truesize(rx_ring, 0);
2337 #endif
2338         xdp_init_buff(&xdp, frame_sz, &rx_ring->xdp_rxq);
2339
2340         while (likely(total_rx_packets < budget)) {
2341                 union ixgbe_adv_rx_desc *rx_desc;
2342                 struct ixgbe_rx_buffer *rx_buffer;
2343                 struct sk_buff *skb;
2344                 int rx_buffer_pgcnt;
2345                 unsigned int size;
2346
2347                 /* return some buffers to hardware, one at a time is too slow */
2348                 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
2349                         ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
2350                         cleaned_count = 0;
2351                 }
2352
2353                 rx_desc = IXGBE_RX_DESC(rx_ring, rx_ring->next_to_clean);
2354                 size = le16_to_cpu(rx_desc->wb.upper.length);
2355                 if (!size)
2356                         break;
2357
2358                 /* This memory barrier is needed to keep us from reading
2359                  * any other fields out of the rx_desc until we know the
2360                  * descriptor has been written back
2361                  */
2362                 dma_rmb();
2363
2364                 rx_buffer = ixgbe_get_rx_buffer(rx_ring, rx_desc, &skb, size, &rx_buffer_pgcnt);
2365
2366                 /* retrieve a buffer from the ring */
2367                 if (!skb) {
2368                         unsigned char *hard_start;
2369
2370                         hard_start = page_address(rx_buffer->page) +
2371                                      rx_buffer->page_offset - offset;
2372                         xdp_prepare_buff(&xdp, hard_start, offset, size, true);
2373                         xdp_buff_clear_frags_flag(&xdp);
2374 #if (PAGE_SIZE > 4096)
2375                         /* At larger PAGE_SIZE, frame_sz depend on len size */
2376                         xdp.frame_sz = ixgbe_rx_frame_truesize(rx_ring, size);
2377 #endif
2378                         skb = ixgbe_run_xdp(adapter, rx_ring, &xdp);
2379                 }
2380
2381                 if (IS_ERR(skb)) {
2382                         unsigned int xdp_res = -PTR_ERR(skb);
2383
2384                         if (xdp_res & (IXGBE_XDP_TX | IXGBE_XDP_REDIR)) {
2385                                 xdp_xmit |= xdp_res;
2386                                 ixgbe_rx_buffer_flip(rx_ring, rx_buffer, size);
2387                         } else {
2388                                 rx_buffer->pagecnt_bias++;
2389                         }
2390                         total_rx_packets++;
2391                         total_rx_bytes += size;
2392                 } else if (skb) {
2393                         ixgbe_add_rx_frag(rx_ring, rx_buffer, skb, size);
2394                 } else if (ring_uses_build_skb(rx_ring)) {
2395                         skb = ixgbe_build_skb(rx_ring, rx_buffer,
2396                                               &xdp, rx_desc);
2397                 } else {
2398                         skb = ixgbe_construct_skb(rx_ring, rx_buffer,
2399                                                   &xdp, rx_desc);
2400                 }
2401
2402                 /* exit if we failed to retrieve a buffer */
2403                 if (!skb) {
2404                         rx_ring->rx_stats.alloc_rx_buff_failed++;
2405                         rx_buffer->pagecnt_bias++;
2406                         break;
2407                 }
2408
2409                 ixgbe_put_rx_buffer(rx_ring, rx_buffer, skb, rx_buffer_pgcnt);
2410                 cleaned_count++;
2411
2412                 /* place incomplete frames back on ring for completion */
2413                 if (ixgbe_is_non_eop(rx_ring, rx_desc, skb))
2414                         continue;
2415
2416                 /* verify the packet layout is correct */
2417                 if (ixgbe_cleanup_headers(rx_ring, rx_desc, skb))
2418                         continue;
2419
2420                 /* probably a little skewed due to removing CRC */
2421                 total_rx_bytes += skb->len;
2422
2423                 /* populate checksum, timestamp, VLAN, and protocol */
2424                 ixgbe_process_skb_fields(rx_ring, rx_desc, skb);
2425
2426 #ifdef IXGBE_FCOE
2427                 /* if ddp, not passing to ULD unless for FCP_RSP or error */
2428                 if (ixgbe_rx_is_fcoe(rx_ring, rx_desc)) {
2429                         ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
2430                         /* include DDPed FCoE data */
2431                         if (ddp_bytes > 0) {
2432                                 if (!mss) {
2433                                         mss = rx_ring->netdev->mtu -
2434                                                 sizeof(struct fcoe_hdr) -
2435                                                 sizeof(struct fc_frame_header) -
2436                                                 sizeof(struct fcoe_crc_eof);
2437                                         if (mss > 512)
2438                                                 mss &= ~511;
2439                                 }
2440                                 total_rx_bytes += ddp_bytes;
2441                                 total_rx_packets += DIV_ROUND_UP(ddp_bytes,
2442                                                                  mss);
2443                         }
2444                         if (!ddp_bytes) {
2445                                 dev_kfree_skb_any(skb);
2446                                 continue;
2447                         }
2448                 }
2449
2450 #endif /* IXGBE_FCOE */
2451                 ixgbe_rx_skb(q_vector, skb);
2452
2453                 /* update budget accounting */
2454                 total_rx_packets++;
2455         }
2456
2457         if (xdp_xmit & IXGBE_XDP_REDIR)
2458                 xdp_do_flush();
2459
2460         if (xdp_xmit & IXGBE_XDP_TX) {
2461                 struct ixgbe_ring *ring = ixgbe_determine_xdp_ring(adapter);
2462
2463                 ixgbe_xdp_ring_update_tail_locked(ring);
2464         }
2465
2466         ixgbe_update_rx_ring_stats(rx_ring, q_vector, total_rx_packets,
2467                                    total_rx_bytes);
2468
2469         return total_rx_packets;
2470 }
2471
2472 /**
2473  * ixgbe_configure_msix - Configure MSI-X hardware
2474  * @adapter: board private structure
2475  *
2476  * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
2477  * interrupts.
2478  **/
2479 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
2480 {
2481         struct ixgbe_q_vector *q_vector;
2482         int v_idx;
2483         u32 mask;
2484
2485         /* Populate MSIX to EITR Select */
2486         if (adapter->num_vfs > 32) {
2487                 u32 eitrsel = BIT(adapter->num_vfs - 32) - 1;
2488                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
2489         }
2490
2491         /*
2492          * Populate the IVAR table and set the ITR values to the
2493          * corresponding register.
2494          */
2495         for (v_idx = 0; v_idx < adapter->num_q_vectors; v_idx++) {
2496                 struct ixgbe_ring *ring;
2497                 q_vector = adapter->q_vector[v_idx];
2498
2499                 ixgbe_for_each_ring(ring, q_vector->rx)
2500                         ixgbe_set_ivar(adapter, 0, ring->reg_idx, v_idx);
2501
2502                 ixgbe_for_each_ring(ring, q_vector->tx)
2503                         ixgbe_set_ivar(adapter, 1, ring->reg_idx, v_idx);
2504
2505                 ixgbe_write_eitr(q_vector);
2506         }
2507
2508         switch (adapter->hw.mac.type) {
2509         case ixgbe_mac_82598EB:
2510                 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
2511                                v_idx);
2512                 break;
2513         case ixgbe_mac_82599EB:
2514         case ixgbe_mac_X540:
2515         case ixgbe_mac_X550:
2516         case ixgbe_mac_X550EM_x:
2517         case ixgbe_mac_x550em_a:
2518                 ixgbe_set_ivar(adapter, -1, 1, v_idx);
2519                 break;
2520         default:
2521                 break;
2522         }
2523         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
2524
2525         /* set up to autoclear timer, and the vectors */
2526         mask = IXGBE_EIMS_ENABLE_MASK;
2527         mask &= ~(IXGBE_EIMS_OTHER |
2528                   IXGBE_EIMS_MAILBOX |
2529                   IXGBE_EIMS_LSC);
2530
2531         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
2532 }
2533
2534 /**
2535  * ixgbe_update_itr - update the dynamic ITR value based on statistics
2536  * @q_vector: structure containing interrupt and ring information
2537  * @ring_container: structure containing ring performance data
2538  *
2539  *      Stores a new ITR value based on packets and byte
2540  *      counts during the last interrupt.  The advantage of per interrupt
2541  *      computation is faster updates and more accurate ITR for the current
2542  *      traffic pattern.  Constants in this function were computed
2543  *      based on theoretical maximum wire speed and thresholds were set based
2544  *      on testing data as well as attempting to minimize response time
2545  *      while increasing bulk throughput.
2546  **/
2547 static void ixgbe_update_itr(struct ixgbe_q_vector *q_vector,
2548                              struct ixgbe_ring_container *ring_container)
2549 {
2550         unsigned int itr = IXGBE_ITR_ADAPTIVE_MIN_USECS |
2551                            IXGBE_ITR_ADAPTIVE_LATENCY;
2552         unsigned int avg_wire_size, packets, bytes;
2553         unsigned long next_update = jiffies;
2554
2555         /* If we don't have any rings just leave ourselves set for maximum
2556          * possible latency so we take ourselves out of the equation.
2557          */
2558         if (!ring_container->ring)
2559                 return;
2560
2561         /* If we didn't update within up to 1 - 2 jiffies we can assume
2562          * that either packets are coming in so slow there hasn't been
2563          * any work, or that there is so much work that NAPI is dealing
2564          * with interrupt moderation and we don't need to do anything.
2565          */
2566         if (time_after(next_update, ring_container->next_update))
2567                 goto clear_counts;
2568
2569         packets = ring_container->total_packets;
2570
2571         /* We have no packets to actually measure against. This means
2572          * either one of the other queues on this vector is active or
2573          * we are a Tx queue doing TSO with too high of an interrupt rate.
2574          *
2575          * When this occurs just tick up our delay by the minimum value
2576          * and hope that this extra delay will prevent us from being called
2577          * without any work on our queue.
2578          */
2579         if (!packets) {
2580                 itr = (q_vector->itr >> 2) + IXGBE_ITR_ADAPTIVE_MIN_INC;
2581                 if (itr > IXGBE_ITR_ADAPTIVE_MAX_USECS)
2582                         itr = IXGBE_ITR_ADAPTIVE_MAX_USECS;
2583                 itr += ring_container->itr & IXGBE_ITR_ADAPTIVE_LATENCY;
2584                 goto clear_counts;
2585         }
2586
2587         bytes = ring_container->total_bytes;
2588
2589         /* If packets are less than 4 or bytes are less than 9000 assume
2590          * insufficient data to use bulk rate limiting approach. We are
2591          * likely latency driven.
2592          */
2593         if (packets < 4 && bytes < 9000) {
2594                 itr = IXGBE_ITR_ADAPTIVE_LATENCY;
2595                 goto adjust_by_size;
2596         }
2597
2598         /* Between 4 and 48 we can assume that our current interrupt delay
2599          * is only slightly too low. As such we should increase it by a small
2600          * fixed amount.
2601          */
2602         if (packets < 48) {
2603                 itr = (q_vector->itr >> 2) + IXGBE_ITR_ADAPTIVE_MIN_INC;
2604                 if (itr > IXGBE_ITR_ADAPTIVE_MAX_USECS)
2605                         itr = IXGBE_ITR_ADAPTIVE_MAX_USECS;
2606                 goto clear_counts;
2607         }
2608
2609         /* Between 48 and 96 is our "goldilocks" zone where we are working
2610          * out "just right". Just report that our current ITR is good for us.
2611          */
2612         if (packets < 96) {
2613                 itr = q_vector->itr >> 2;
2614                 goto clear_counts;
2615         }
2616
2617         /* If packet count is 96 or greater we are likely looking at a slight
2618          * overrun of the delay we want. Try halving our delay to see if that
2619          * will cut the number of packets in half per interrupt.
2620          */
2621         if (packets < 256) {
2622                 itr = q_vector->itr >> 3;
2623                 if (itr < IXGBE_ITR_ADAPTIVE_MIN_USECS)
2624                         itr = IXGBE_ITR_ADAPTIVE_MIN_USECS;
2625                 goto clear_counts;
2626         }
2627
2628         /* The paths below assume we are dealing with a bulk ITR since number
2629          * of packets is 256 or greater. We are just going to have to compute
2630          * a value and try to bring the count under control, though for smaller
2631          * packet sizes there isn't much we can do as NAPI polling will likely
2632          * be kicking in sooner rather than later.
2633          */
2634         itr = IXGBE_ITR_ADAPTIVE_BULK;
2635
2636 adjust_by_size:
2637         /* If packet counts are 256 or greater we can assume we have a gross
2638          * overestimation of what the rate should be. Instead of trying to fine
2639          * tune it just use the formula below to try and dial in an exact value
2640          * give the current packet size of the frame.
2641          */
2642         avg_wire_size = bytes / packets;
2643
2644         /* The following is a crude approximation of:
2645          *  wmem_default / (size + overhead) = desired_pkts_per_int
2646          *  rate / bits_per_byte / (size + ethernet overhead) = pkt_rate
2647          *  (desired_pkt_rate / pkt_rate) * usecs_per_sec = ITR value
2648          *
2649          * Assuming wmem_default is 212992 and overhead is 640 bytes per
2650          * packet, (256 skb, 64 headroom, 320 shared info), we can reduce the
2651          * formula down to
2652          *
2653          *  (170 * (size + 24)) / (size + 640) = ITR
2654          *
2655          * We first do some math on the packet size and then finally bitshift
2656          * by 8 after rounding up. We also have to account for PCIe link speed
2657          * difference as ITR scales based on this.
2658          */
2659         if (avg_wire_size <= 60) {
2660                 /* Start at 50k ints/sec */
2661                 avg_wire_size = 5120;
2662         } else if (avg_wire_size <= 316) {
2663                 /* 50K ints/sec to 16K ints/sec */
2664                 avg_wire_size *= 40;
2665                 avg_wire_size += 2720;
2666         } else if (avg_wire_size <= 1084) {
2667                 /* 16K ints/sec to 9.2K ints/sec */
2668                 avg_wire_size *= 15;
2669                 avg_wire_size += 11452;
2670         } else if (avg_wire_size < 1968) {
2671                 /* 9.2K ints/sec to 8K ints/sec */
2672                 avg_wire_size *= 5;
2673                 avg_wire_size += 22420;
2674         } else {
2675                 /* plateau at a limit of 8K ints/sec */
2676                 avg_wire_size = 32256;
2677         }
2678
2679         /* If we are in low latency mode half our delay which doubles the rate
2680          * to somewhere between 100K to 16K ints/sec
2681          */
2682         if (itr & IXGBE_ITR_ADAPTIVE_LATENCY)
2683                 avg_wire_size >>= 1;
2684
2685         /* Resultant value is 256 times larger than it needs to be. This
2686          * gives us room to adjust the value as needed to either increase
2687          * or decrease the value based on link speeds of 10G, 2.5G, 1G, etc.
2688          *
2689          * Use addition as we have already recorded the new latency flag
2690          * for the ITR value.
2691          */
2692         switch (q_vector->adapter->link_speed) {
2693         case IXGBE_LINK_SPEED_10GB_FULL:
2694         case IXGBE_LINK_SPEED_100_FULL:
2695         default:
2696                 itr += DIV_ROUND_UP(avg_wire_size,
2697                                     IXGBE_ITR_ADAPTIVE_MIN_INC * 256) *
2698                        IXGBE_ITR_ADAPTIVE_MIN_INC;
2699                 break;
2700         case IXGBE_LINK_SPEED_2_5GB_FULL:
2701         case IXGBE_LINK_SPEED_1GB_FULL:
2702         case IXGBE_LINK_SPEED_10_FULL:
2703                 if (avg_wire_size > 8064)
2704                         avg_wire_size = 8064;
2705                 itr += DIV_ROUND_UP(avg_wire_size,
2706                                     IXGBE_ITR_ADAPTIVE_MIN_INC * 64) *
2707                        IXGBE_ITR_ADAPTIVE_MIN_INC;
2708                 break;
2709         }
2710
2711 clear_counts:
2712         /* write back value */
2713         ring_container->itr = itr;
2714
2715         /* next update should occur within next jiffy */
2716         ring_container->next_update = next_update + 1;
2717
2718         ring_container->total_bytes = 0;
2719         ring_container->total_packets = 0;
2720 }
2721
2722 /**
2723  * ixgbe_write_eitr - write EITR register in hardware specific way
2724  * @q_vector: structure containing interrupt and ring information
2725  *
2726  * This function is made to be called by ethtool and by the driver
2727  * when it needs to update EITR registers at runtime.  Hardware
2728  * specific quirks/differences are taken care of here.
2729  */
2730 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
2731 {
2732         struct ixgbe_adapter *adapter = q_vector->adapter;
2733         struct ixgbe_hw *hw = &adapter->hw;
2734         int v_idx = q_vector->v_idx;
2735         u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
2736
2737         switch (adapter->hw.mac.type) {
2738         case ixgbe_mac_82598EB:
2739                 /* must write high and low 16 bits to reset counter */
2740                 itr_reg |= (itr_reg << 16);
2741                 break;
2742         case ixgbe_mac_82599EB:
2743         case ixgbe_mac_X540:
2744         case ixgbe_mac_X550:
2745         case ixgbe_mac_X550EM_x:
2746         case ixgbe_mac_x550em_a:
2747                 /*
2748                  * set the WDIS bit to not clear the timer bits and cause an
2749                  * immediate assertion of the interrupt
2750                  */
2751                 itr_reg |= IXGBE_EITR_CNT_WDIS;
2752                 break;
2753         default:
2754                 break;
2755         }
2756         IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
2757 }
2758
2759 static void ixgbe_set_itr(struct ixgbe_q_vector *q_vector)
2760 {
2761         u32 new_itr;
2762
2763         ixgbe_update_itr(q_vector, &q_vector->tx);
2764         ixgbe_update_itr(q_vector, &q_vector->rx);
2765
2766         /* use the smallest value of new ITR delay calculations */
2767         new_itr = min(q_vector->rx.itr, q_vector->tx.itr);
2768
2769         /* Clear latency flag if set, shift into correct position */
2770         new_itr &= ~IXGBE_ITR_ADAPTIVE_LATENCY;
2771         new_itr <<= 2;
2772
2773         if (new_itr != q_vector->itr) {
2774                 /* save the algorithm value here */
2775                 q_vector->itr = new_itr;
2776
2777                 ixgbe_write_eitr(q_vector);
2778         }
2779 }
2780
2781 /**
2782  * ixgbe_check_overtemp_subtask - check for over temperature
2783  * @adapter: pointer to adapter
2784  **/
2785 static void ixgbe_check_overtemp_subtask(struct ixgbe_adapter *adapter)
2786 {
2787         struct ixgbe_hw *hw = &adapter->hw;
2788         u32 eicr = adapter->interrupt_event;
2789
2790         if (test_bit(__IXGBE_DOWN, &adapter->state))
2791                 return;
2792
2793         if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_EVENT))
2794                 return;
2795
2796         adapter->flags2 &= ~IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2797
2798         switch (hw->device_id) {
2799         case IXGBE_DEV_ID_82599_T3_LOM:
2800                 /*
2801                  * Since the warning interrupt is for both ports
2802                  * we don't have to check if:
2803                  *  - This interrupt wasn't for our port.
2804                  *  - We may have missed the interrupt so always have to
2805                  *    check if we  got a LSC
2806                  */
2807                 if (!(eicr & IXGBE_EICR_GPI_SDP0_8259X) &&
2808                     !(eicr & IXGBE_EICR_LSC))
2809                         return;
2810
2811                 if (!(eicr & IXGBE_EICR_LSC) && hw->mac.ops.check_link) {
2812                         u32 speed;
2813                         bool link_up = false;
2814
2815                         hw->mac.ops.check_link(hw, &speed, &link_up, false);
2816
2817                         if (link_up)
2818                                 return;
2819                 }
2820
2821                 /* Check if this is not due to overtemp */
2822                 if (!hw->phy.ops.check_overtemp(hw))
2823                         return;
2824
2825                 break;
2826         case IXGBE_DEV_ID_X550EM_A_1G_T:
2827         case IXGBE_DEV_ID_X550EM_A_1G_T_L:
2828                 if (!hw->phy.ops.check_overtemp(hw))
2829                         return;
2830                 break;
2831         default:
2832                 if (adapter->hw.mac.type >= ixgbe_mac_X540)
2833                         return;
2834                 if (!(eicr & IXGBE_EICR_GPI_SDP0(hw)))
2835                         return;
2836                 break;
2837         }
2838         e_crit(drv, "%s\n", ixgbe_overheat_msg);
2839
2840         adapter->interrupt_event = 0;
2841 }
2842
2843 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
2844 {
2845         struct ixgbe_hw *hw = &adapter->hw;
2846
2847         if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
2848             (eicr & IXGBE_EICR_GPI_SDP1(hw))) {
2849                 e_crit(probe, "Fan has stopped, replace the adapter\n");
2850                 /* write to clear the interrupt */
2851                 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1(hw));
2852         }
2853 }
2854
2855 static void ixgbe_check_overtemp_event(struct ixgbe_adapter *adapter, u32 eicr)
2856 {
2857         struct ixgbe_hw *hw = &adapter->hw;
2858
2859         if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
2860                 return;
2861
2862         switch (adapter->hw.mac.type) {
2863         case ixgbe_mac_82599EB:
2864                 /*
2865                  * Need to check link state so complete overtemp check
2866                  * on service task
2867                  */
2868                 if (((eicr & IXGBE_EICR_GPI_SDP0(hw)) ||
2869                      (eicr & IXGBE_EICR_LSC)) &&
2870                     (!test_bit(__IXGBE_DOWN, &adapter->state))) {
2871                         adapter->interrupt_event = eicr;
2872                         adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2873                         ixgbe_service_event_schedule(adapter);
2874                         return;
2875                 }
2876                 return;
2877         case ixgbe_mac_x550em_a:
2878                 if (eicr & IXGBE_EICR_GPI_SDP0_X550EM_a) {
2879                         adapter->interrupt_event = eicr;
2880                         adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2881                         ixgbe_service_event_schedule(adapter);
2882                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC,
2883                                         IXGBE_EICR_GPI_SDP0_X550EM_a);
2884                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICR,
2885                                         IXGBE_EICR_GPI_SDP0_X550EM_a);
2886                 }
2887                 return;
2888         case ixgbe_mac_X550:
2889         case ixgbe_mac_X540:
2890                 if (!(eicr & IXGBE_EICR_TS))
2891                         return;
2892                 break;
2893         default:
2894                 return;
2895         }
2896
2897         e_crit(drv, "%s\n", ixgbe_overheat_msg);
2898 }
2899
2900 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
2901 {
2902         switch (hw->mac.type) {
2903         case ixgbe_mac_82598EB:
2904                 if (hw->phy.type == ixgbe_phy_nl)
2905                         return true;
2906                 return false;
2907         case ixgbe_mac_82599EB:
2908         case ixgbe_mac_X550EM_x:
2909         case ixgbe_mac_x550em_a:
2910                 switch (hw->mac.ops.get_media_type(hw)) {
2911                 case ixgbe_media_type_fiber:
2912                 case ixgbe_media_type_fiber_qsfp:
2913                         return true;
2914                 default:
2915                         return false;
2916                 }
2917         default:
2918                 return false;
2919         }
2920 }
2921
2922 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
2923 {
2924         struct ixgbe_hw *hw = &adapter->hw;
2925         u32 eicr_mask = IXGBE_EICR_GPI_SDP2(hw);
2926
2927         if (!ixgbe_is_sfp(hw))
2928                 return;
2929
2930         /* Later MAC's use different SDP */
2931         if (hw->mac.type >= ixgbe_mac_X540)
2932                 eicr_mask = IXGBE_EICR_GPI_SDP0_X540;
2933
2934         if (eicr & eicr_mask) {
2935                 /* Clear the interrupt */
2936                 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr_mask);
2937                 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2938                         adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
2939                         adapter->sfp_poll_time = 0;
2940                         ixgbe_service_event_schedule(adapter);
2941                 }
2942         }
2943
2944         if (adapter->hw.mac.type == ixgbe_mac_82599EB &&
2945             (eicr & IXGBE_EICR_GPI_SDP1(hw))) {
2946                 /* Clear the interrupt */
2947                 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1(hw));
2948                 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2949                         adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
2950                         ixgbe_service_event_schedule(adapter);
2951                 }
2952         }
2953 }
2954
2955 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
2956 {
2957         struct ixgbe_hw *hw = &adapter->hw;
2958
2959         adapter->lsc_int++;
2960         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
2961         adapter->link_check_timeout = jiffies;
2962         if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2963                 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
2964                 IXGBE_WRITE_FLUSH(hw);
2965                 ixgbe_service_event_schedule(adapter);
2966         }
2967 }
2968
2969 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
2970                                            u64 qmask)
2971 {
2972         struct ixgbe_hw *hw = &adapter->hw;
2973         u32 mask;
2974
2975         switch (hw->mac.type) {
2976         case ixgbe_mac_82598EB:
2977                 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2978                 IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
2979                 break;
2980         case ixgbe_mac_82599EB:
2981         case ixgbe_mac_X540:
2982         case ixgbe_mac_X550:
2983         case ixgbe_mac_X550EM_x:
2984         case ixgbe_mac_x550em_a:
2985                 mask = (qmask & 0xFFFFFFFF);
2986                 if (mask)
2987                         IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
2988                 mask = (qmask >> 32);
2989                 if (mask)
2990                         IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
2991                 break;
2992         default:
2993                 break;
2994         }
2995         /* skip the flush */
2996 }
2997
2998 /**
2999  * ixgbe_irq_enable - Enable default interrupt generation settings
3000  * @adapter: board private structure
3001  * @queues: enable irqs for queues
3002  * @flush: flush register write
3003  **/
3004 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
3005                                     bool flush)
3006 {
3007         struct ixgbe_hw *hw = &adapter->hw;
3008         u32 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
3009
3010         /* don't reenable LSC while waiting for link */
3011         if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
3012                 mask &= ~IXGBE_EIMS_LSC;
3013
3014         if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
3015                 switch (adapter->hw.mac.type) {
3016                 case ixgbe_mac_82599EB:
3017                         mask |= IXGBE_EIMS_GPI_SDP0(hw);
3018                         break;
3019                 case ixgbe_mac_X540:
3020                 case ixgbe_mac_X550:
3021                 case ixgbe_mac_X550EM_x:
3022                 case ixgbe_mac_x550em_a:
3023                         mask |= IXGBE_EIMS_TS;
3024                         break;
3025                 default:
3026                         break;
3027                 }
3028         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
3029                 mask |= IXGBE_EIMS_GPI_SDP1(hw);
3030         switch (adapter->hw.mac.type) {
3031         case ixgbe_mac_82599EB:
3032                 mask |= IXGBE_EIMS_GPI_SDP1(hw);
3033                 mask |= IXGBE_EIMS_GPI_SDP2(hw);
3034                 fallthrough;
3035         case ixgbe_mac_X540:
3036         case ixgbe_mac_X550:
3037         case ixgbe_mac_X550EM_x:
3038         case ixgbe_mac_x550em_a:
3039                 if (adapter->hw.device_id == IXGBE_DEV_ID_X550EM_X_SFP ||
3040                     adapter->hw.device_id == IXGBE_DEV_ID_X550EM_A_SFP ||
3041                     adapter->hw.device_id == IXGBE_DEV_ID_X550EM_A_SFP_N)
3042                         mask |= IXGBE_EIMS_GPI_SDP0(&adapter->hw);
3043                 if (adapter->hw.phy.type == ixgbe_phy_x550em_ext_t)
3044                         mask |= IXGBE_EICR_GPI_SDP0_X540;
3045                 mask |= IXGBE_EIMS_ECC;
3046                 mask |= IXGBE_EIMS_MAILBOX;
3047                 break;
3048         default:
3049                 break;
3050         }
3051
3052         if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
3053             !(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
3054                 mask |= IXGBE_EIMS_FLOW_DIR;
3055
3056         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
3057         if (queues)
3058                 ixgbe_irq_enable_queues(adapter, ~0);
3059         if (flush)
3060                 IXGBE_WRITE_FLUSH(&adapter->hw);
3061 }
3062
3063 static irqreturn_t ixgbe_msix_other(int irq, void *data)
3064 {
3065         struct ixgbe_adapter *adapter = data;
3066         struct ixgbe_hw *hw = &adapter->hw;
3067         u32 eicr;
3068
3069         /*
3070          * Workaround for Silicon errata.  Use clear-by-write instead
3071          * of clear-by-read.  Reading with EICS will return the
3072          * interrupt causes without clearing, which later be done
3073          * with the write to EICR.
3074          */
3075         eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
3076
3077         /* The lower 16bits of the EICR register are for the queue interrupts
3078          * which should be masked here in order to not accidentally clear them if
3079          * the bits are high when ixgbe_msix_other is called. There is a race
3080          * condition otherwise which results in possible performance loss
3081          * especially if the ixgbe_msix_other interrupt is triggering
3082          * consistently (as it would when PPS is turned on for the X540 device)
3083          */
3084         eicr &= 0xFFFF0000;
3085
3086         IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
3087
3088         if (eicr & IXGBE_EICR_LSC)
3089                 ixgbe_check_lsc(adapter);
3090
3091         if (eicr & IXGBE_EICR_MAILBOX)
3092                 ixgbe_msg_task(adapter);
3093
3094         switch (hw->mac.type) {
3095         case ixgbe_mac_82599EB:
3096         case ixgbe_mac_X540:
3097         case ixgbe_mac_X550:
3098         case ixgbe_mac_X550EM_x:
3099         case ixgbe_mac_x550em_a:
3100                 if (hw->phy.type == ixgbe_phy_x550em_ext_t &&
3101                     (eicr & IXGBE_EICR_GPI_SDP0_X540)) {
3102                         adapter->flags2 |= IXGBE_FLAG2_PHY_INTERRUPT;
3103                         ixgbe_service_event_schedule(adapter);
3104                         IXGBE_WRITE_REG(hw, IXGBE_EICR,
3105                                         IXGBE_EICR_GPI_SDP0_X540);
3106                 }
3107                 if (eicr & IXGBE_EICR_ECC) {
3108                         e_info(link, "Received ECC Err, initiating reset\n");
3109                         set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
3110                         ixgbe_service_event_schedule(adapter);
3111                         IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC);
3112                 }
3113                 /* Handle Flow Director Full threshold interrupt */
3114                 if (eicr & IXGBE_EICR_FLOW_DIR) {
3115                         int reinit_count = 0;
3116                         int i;
3117                         for (i = 0; i < adapter->num_tx_queues; i++) {
3118                                 struct ixgbe_ring *ring = adapter->tx_ring[i];
3119                                 if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
3120                                                        &ring->state))
3121                                         reinit_count++;
3122                         }
3123                         if (reinit_count) {
3124                                 /* no more flow director interrupts until after init */
3125                                 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_FLOW_DIR);
3126                                 adapter->flags2 |= IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
3127                                 ixgbe_service_event_schedule(adapter);
3128                         }
3129                 }
3130                 ixgbe_check_sfp_event(adapter, eicr);
3131                 ixgbe_check_overtemp_event(adapter, eicr);
3132                 break;
3133         default:
3134                 break;
3135         }
3136
3137         ixgbe_check_fan_failure(adapter, eicr);
3138
3139         if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
3140                 ixgbe_ptp_check_pps_event(adapter);
3141
3142         /* re-enable the original interrupt state, no lsc, no queues */
3143         if (!test_bit(__IXGBE_DOWN, &adapter->state))
3144                 ixgbe_irq_enable(adapter, false, false);
3145
3146         return IRQ_HANDLED;
3147 }
3148
3149 static irqreturn_t ixgbe_msix_clean_rings(int irq, void *data)
3150 {
3151         struct ixgbe_q_vector *q_vector = data;
3152
3153         /* EIAM disabled interrupts (on this vector) for us */
3154
3155         if (q_vector->rx.ring || q_vector->tx.ring)
3156                 napi_schedule_irqoff(&q_vector->napi);
3157
3158         return IRQ_HANDLED;
3159 }
3160
3161 /**
3162  * ixgbe_poll - NAPI Rx polling callback
3163  * @napi: structure for representing this polling device
3164  * @budget: how many packets driver is allowed to clean
3165  *
3166  * This function is used for legacy and MSI, NAPI mode
3167  **/
3168 int ixgbe_poll(struct napi_struct *napi, int budget)
3169 {
3170         struct ixgbe_q_vector *q_vector =
3171                                 container_of(napi, struct ixgbe_q_vector, napi);
3172         struct ixgbe_adapter *adapter = q_vector->adapter;
3173         struct ixgbe_ring *ring;
3174         int per_ring_budget, work_done = 0;
3175         bool clean_complete = true;
3176
3177 #ifdef CONFIG_IXGBE_DCA
3178         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
3179                 ixgbe_update_dca(q_vector);
3180 #endif
3181
3182         ixgbe_for_each_ring(ring, q_vector->tx) {
3183                 bool wd = ring->xsk_pool ?
3184                           ixgbe_clean_xdp_tx_irq(q_vector, ring, budget) :
3185                           ixgbe_clean_tx_irq(q_vector, ring, budget);
3186
3187                 if (!wd)
3188                         clean_complete = false;
3189         }
3190
3191         /* Exit if we are called by netpoll */
3192         if (budget <= 0)
3193                 return budget;
3194
3195         /* attempt to distribute budget to each queue fairly, but don't allow
3196          * the budget to go below 1 because we'll exit polling */
3197         if (q_vector->rx.count > 1)
3198                 per_ring_budget = max(budget/q_vector->rx.count, 1);
3199         else
3200                 per_ring_budget = budget;
3201
3202         ixgbe_for_each_ring(ring, q_vector->rx) {
3203                 int cleaned = ring->xsk_pool ?
3204                               ixgbe_clean_rx_irq_zc(q_vector, ring,
3205                                                     per_ring_budget) :
3206                               ixgbe_clean_rx_irq(q_vector, ring,
3207                                                  per_ring_budget);
3208
3209                 work_done += cleaned;
3210                 if (cleaned >= per_ring_budget)
3211                         clean_complete = false;
3212         }
3213
3214         /* If all work not completed, return budget and keep polling */
3215         if (!clean_complete)
3216                 return budget;
3217
3218         /* all work done, exit the polling mode */
3219         if (likely(napi_complete_done(napi, work_done))) {
3220                 if (adapter->rx_itr_setting & 1)
3221                         ixgbe_set_itr(q_vector);
3222                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3223                         ixgbe_irq_enable_queues(adapter,
3224                                                 BIT_ULL(q_vector->v_idx));
3225         }
3226
3227         return min(work_done, budget - 1);
3228 }
3229
3230 /**
3231  * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
3232  * @adapter: board private structure
3233  *
3234  * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
3235  * interrupts from the kernel.
3236  **/
3237 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
3238 {
3239         struct net_device *netdev = adapter->netdev;
3240         unsigned int ri = 0, ti = 0;
3241         int vector, err;
3242
3243         for (vector = 0; vector < adapter->num_q_vectors; vector++) {
3244                 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
3245                 struct msix_entry *entry = &adapter->msix_entries[vector];
3246
3247                 if (q_vector->tx.ring && q_vector->rx.ring) {
3248                         snprintf(q_vector->name, sizeof(q_vector->name),
3249                                  "%s-TxRx-%u", netdev->name, ri++);
3250                         ti++;
3251                 } else if (q_vector->rx.ring) {
3252                         snprintf(q_vector->name, sizeof(q_vector->name),
3253                                  "%s-rx-%u", netdev->name, ri++);
3254                 } else if (q_vector->tx.ring) {
3255                         snprintf(q_vector->name, sizeof(q_vector->name),
3256                                  "%s-tx-%u", netdev->name, ti++);
3257                 } else {
3258                         /* skip this unused q_vector */
3259                         continue;
3260                 }
3261                 err = request_irq(entry->vector, &ixgbe_msix_clean_rings, 0,
3262                                   q_vector->name, q_vector);
3263                 if (err) {
3264                         e_err(probe, "request_irq failed for MSIX interrupt "
3265                               "Error: %d\n", err);
3266                         goto free_queue_irqs;
3267                 }
3268                 /* If Flow Director is enabled, set interrupt affinity */
3269                 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3270                         /* assign the mask for this irq */
3271                         irq_update_affinity_hint(entry->vector,
3272                                                  &q_vector->affinity_mask);
3273                 }
3274         }
3275
3276         err = request_irq(adapter->msix_entries[vector].vector,
3277                           ixgbe_msix_other, 0, netdev->name, adapter);
3278         if (err) {
3279                 e_err(probe, "request_irq for msix_other failed: %d\n", err);
3280                 goto free_queue_irqs;
3281         }
3282
3283         return 0;
3284
3285 free_queue_irqs:
3286         while (vector) {
3287                 vector--;
3288                 irq_update_affinity_hint(adapter->msix_entries[vector].vector,
3289                                          NULL);
3290                 free_irq(adapter->msix_entries[vector].vector,
3291                          adapter->q_vector[vector]);
3292         }
3293         adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
3294         pci_disable_msix(adapter->pdev);
3295         kfree(adapter->msix_entries);
3296         adapter->msix_entries = NULL;
3297         return err;
3298 }
3299
3300 /**
3301  * ixgbe_intr - legacy mode Interrupt Handler
3302  * @irq: interrupt number
3303  * @data: pointer to a network interface device structure
3304  **/
3305 static irqreturn_t ixgbe_intr(int irq, void *data)
3306 {
3307         struct ixgbe_adapter *adapter = data;
3308         struct ixgbe_hw *hw = &adapter->hw;
3309         struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
3310         u32 eicr;
3311
3312         /*
3313          * Workaround for silicon errata #26 on 82598.  Mask the interrupt
3314          * before the read of EICR.
3315          */
3316         IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
3317
3318         /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
3319          * therefore no explicit interrupt disable is necessary */
3320         eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
3321         if (!eicr) {
3322                 /*
3323                  * shared interrupt alert!
3324                  * make sure interrupts are enabled because the read will
3325                  * have disabled interrupts due to EIAM
3326                  * finish the workaround of silicon errata on 82598.  Unmask
3327                  * the interrupt that we masked before the EICR read.
3328                  */
3329                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3330                         ixgbe_irq_enable(adapter, true, true);
3331                 return IRQ_NONE;        /* Not our interrupt */
3332         }
3333
3334         if (eicr & IXGBE_EICR_LSC)
3335                 ixgbe_check_lsc(adapter);
3336
3337         switch (hw->mac.type) {
3338         case ixgbe_mac_82599EB:
3339                 ixgbe_check_sfp_event(adapter, eicr);
3340                 fallthrough;
3341         case ixgbe_mac_X540:
3342         case ixgbe_mac_X550:
3343         case ixgbe_mac_X550EM_x:
3344         case ixgbe_mac_x550em_a:
3345                 if (eicr & IXGBE_EICR_ECC) {
3346                         e_info(link, "Received ECC Err, initiating reset\n");
3347                         set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
3348                         ixgbe_service_event_schedule(adapter);
3349                         IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC);
3350                 }
3351                 ixgbe_check_overtemp_event(adapter, eicr);
3352                 break;
3353         default:
3354                 break;
3355         }
3356
3357         ixgbe_check_fan_failure(adapter, eicr);
3358         if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
3359                 ixgbe_ptp_check_pps_event(adapter);
3360
3361         /* would disable interrupts here but EIAM disabled it */
3362         napi_schedule_irqoff(&q_vector->napi);
3363
3364         /*
3365          * re-enable link(maybe) and non-queue interrupts, no flush.
3366          * ixgbe_poll will re-enable the queue interrupts
3367          */
3368         if (!test_bit(__IXGBE_DOWN, &adapter->state))
3369                 ixgbe_irq_enable(adapter, false, false);
3370
3371         return IRQ_HANDLED;
3372 }
3373
3374 /**
3375  * ixgbe_request_irq - initialize interrupts
3376  * @adapter: board private structure
3377  *
3378  * Attempts to configure interrupts using the best available
3379  * capabilities of the hardware and kernel.
3380  **/
3381 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
3382 {
3383         struct net_device *netdev = adapter->netdev;
3384         int err;
3385
3386         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3387                 err = ixgbe_request_msix_irqs(adapter);
3388         else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED)
3389                 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
3390                                   netdev->name, adapter);
3391         else
3392                 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
3393                                   netdev->name, adapter);
3394
3395         if (err)
3396                 e_err(probe, "request_irq failed, Error %d\n", err);
3397
3398         return err;
3399 }
3400
3401 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
3402 {
3403         int vector;
3404
3405         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
3406                 free_irq(adapter->pdev->irq, adapter);
3407                 return;
3408         }
3409
3410         if (!adapter->msix_entries)
3411                 return;
3412
3413         for (vector = 0; vector < adapter->num_q_vectors; vector++) {
3414                 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
3415                 struct msix_entry *entry = &adapter->msix_entries[vector];
3416
3417                 /* free only the irqs that were actually requested */
3418                 if (!q_vector->rx.ring && !q_vector->tx.ring)
3419                         continue;
3420
3421                 /* clear the affinity_mask in the IRQ descriptor */
3422                 irq_update_affinity_hint(entry->vector, NULL);
3423
3424                 free_irq(entry->vector, q_vector);
3425         }
3426
3427         free_irq(adapter->msix_entries[vector].vector, adapter);
3428 }
3429
3430 /**
3431  * ixgbe_irq_disable - Mask off interrupt generation on the NIC
3432  * @adapter: board private structure
3433  **/
3434 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
3435 {
3436         switch (adapter->hw.mac.type) {
3437         case ixgbe_mac_82598EB:
3438                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
3439                 break;
3440         case ixgbe_mac_82599EB:
3441         case ixgbe_mac_X540:
3442         case ixgbe_mac_X550:
3443         case ixgbe_mac_X550EM_x:
3444         case ixgbe_mac_x550em_a:
3445                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
3446                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
3447                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
3448                 break;
3449         default:
3450                 break;
3451         }
3452         IXGBE_WRITE_FLUSH(&adapter->hw);
3453         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3454                 int vector;
3455
3456                 for (vector = 0; vector < adapter->num_q_vectors; vector++)
3457                         synchronize_irq(adapter->msix_entries[vector].vector);
3458
3459                 synchronize_irq(adapter->msix_entries[vector++].vector);
3460         } else {
3461                 synchronize_irq(adapter->pdev->irq);
3462         }
3463 }
3464
3465 /**
3466  * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
3467  * @adapter: board private structure
3468  *
3469  **/
3470 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
3471 {
3472         struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
3473
3474         ixgbe_write_eitr(q_vector);
3475
3476         ixgbe_set_ivar(adapter, 0, 0, 0);
3477         ixgbe_set_ivar(adapter, 1, 0, 0);
3478
3479         e_info(hw, "Legacy interrupt IVAR setup done\n");
3480 }
3481
3482 /**
3483  * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
3484  * @adapter: board private structure
3485  * @ring: structure containing ring specific data
3486  *
3487  * Configure the Tx descriptor ring after a reset.
3488  **/
3489 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
3490                              struct ixgbe_ring *ring)
3491 {
3492         struct ixgbe_hw *hw = &adapter->hw;
3493         u64 tdba = ring->dma;
3494         int wait_loop = 10;
3495         u32 txdctl = IXGBE_TXDCTL_ENABLE;
3496         u8 reg_idx = ring->reg_idx;
3497
3498         ring->xsk_pool = NULL;
3499         if (ring_is_xdp(ring))
3500                 ring->xsk_pool = ixgbe_xsk_pool(adapter, ring);
3501
3502         /* disable queue to avoid issues while updating state */
3503         IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), 0);
3504         IXGBE_WRITE_FLUSH(hw);
3505
3506         IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
3507                         (tdba & DMA_BIT_MASK(32)));
3508         IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
3509         IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
3510                         ring->count * sizeof(union ixgbe_adv_tx_desc));
3511         IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
3512         IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
3513         ring->tail = adapter->io_addr + IXGBE_TDT(reg_idx);
3514
3515         /*
3516          * set WTHRESH to encourage burst writeback, it should not be set
3517          * higher than 1 when:
3518          * - ITR is 0 as it could cause false TX hangs
3519          * - ITR is set to > 100k int/sec and BQL is enabled
3520          *
3521          * In order to avoid issues WTHRESH + PTHRESH should always be equal
3522          * to or less than the number of on chip descriptors, which is
3523          * currently 40.
3524          */
3525         if (!ring->q_vector || (ring->q_vector->itr < IXGBE_100K_ITR))
3526                 txdctl |= 1u << 16;     /* WTHRESH = 1 */
3527         else
3528                 txdctl |= 8u << 16;     /* WTHRESH = 8 */
3529
3530         /*
3531          * Setting PTHRESH to 32 both improves performance
3532          * and avoids a TX hang with DFP enabled
3533          */
3534         txdctl |= (1u << 8) |   /* HTHRESH = 1 */
3535                    32;          /* PTHRESH = 32 */
3536
3537         /* reinitialize flowdirector state */
3538         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3539                 ring->atr_sample_rate = adapter->atr_sample_rate;
3540                 ring->atr_count = 0;
3541                 set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
3542         } else {
3543                 ring->atr_sample_rate = 0;
3544         }
3545
3546         /* initialize XPS */
3547         if (!test_and_set_bit(__IXGBE_TX_XPS_INIT_DONE, &ring->state)) {
3548                 struct ixgbe_q_vector *q_vector = ring->q_vector;
3549
3550                 if (q_vector)
3551                         netif_set_xps_queue(ring->netdev,
3552                                             &q_vector->affinity_mask,
3553                                             ring->queue_index);
3554         }
3555
3556         clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
3557
3558         /* reinitialize tx_buffer_info */
3559         memset(ring->tx_buffer_info, 0,
3560                sizeof(struct ixgbe_tx_buffer) * ring->count);
3561
3562         /* enable queue */
3563         IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
3564
3565         /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3566         if (hw->mac.type == ixgbe_mac_82598EB &&
3567             !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3568                 return;
3569
3570         /* poll to verify queue is enabled */
3571         do {
3572                 usleep_range(1000, 2000);
3573                 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
3574         } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
3575         if (!wait_loop)
3576                 hw_dbg(hw, "Could not enable Tx Queue %d\n", reg_idx);
3577 }
3578
3579 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
3580 {
3581         struct ixgbe_hw *hw = &adapter->hw;
3582         u32 rttdcs, mtqc;
3583         u8 tcs = adapter->hw_tcs;
3584
3585         if (hw->mac.type == ixgbe_mac_82598EB)
3586                 return;
3587
3588         /* disable the arbiter while setting MTQC */
3589         rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3590         rttdcs |= IXGBE_RTTDCS_ARBDIS;
3591         IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3592
3593         /* set transmit pool layout */
3594         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3595                 mtqc = IXGBE_MTQC_VT_ENA;
3596                 if (tcs > 4)
3597                         mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3598                 else if (tcs > 1)
3599                         mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3600                 else if (adapter->ring_feature[RING_F_VMDQ].mask ==
3601                          IXGBE_82599_VMDQ_4Q_MASK)
3602                         mtqc |= IXGBE_MTQC_32VF;
3603                 else
3604                         mtqc |= IXGBE_MTQC_64VF;
3605         } else {
3606                 if (tcs > 4) {
3607                         mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3608                 } else if (tcs > 1) {
3609                         mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3610                 } else {
3611                         u8 max_txq = adapter->num_tx_queues +
3612                                 adapter->num_xdp_queues;
3613                         if (max_txq > 63)
3614                                 mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3615                         else
3616                                 mtqc = IXGBE_MTQC_64Q_1PB;
3617                 }
3618         }
3619
3620         IXGBE_WRITE_REG(hw, IXGBE_MTQC, mtqc);
3621
3622         /* Enable Security TX Buffer IFG for multiple pb */
3623         if (tcs) {
3624                 u32 sectx = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
3625                 sectx |= IXGBE_SECTX_DCB;
3626                 IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, sectx);
3627         }
3628
3629         /* re-enable the arbiter */
3630         rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
3631         IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3632 }
3633
3634 /**
3635  * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
3636  * @adapter: board private structure
3637  *
3638  * Configure the Tx unit of the MAC after a reset.
3639  **/
3640 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
3641 {
3642         struct ixgbe_hw *hw = &adapter->hw;
3643         u32 dmatxctl;
3644         u32 i;
3645
3646         ixgbe_setup_mtqc(adapter);
3647
3648         if (hw->mac.type != ixgbe_mac_82598EB) {
3649                 /* DMATXCTL.EN must be before Tx queues are enabled */
3650                 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
3651                 dmatxctl |= IXGBE_DMATXCTL_TE;
3652                 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
3653         }
3654
3655         /* Setup the HW Tx Head and Tail descriptor pointers */
3656         for (i = 0; i < adapter->num_tx_queues; i++)
3657                 ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
3658         for (i = 0; i < adapter->num_xdp_queues; i++)
3659                 ixgbe_configure_tx_ring(adapter, adapter->xdp_ring[i]);
3660 }
3661
3662 static void ixgbe_enable_rx_drop(struct ixgbe_adapter *adapter,
3663                                  struct ixgbe_ring *ring)
3664 {
3665         struct ixgbe_hw *hw = &adapter->hw;
3666         u8 reg_idx = ring->reg_idx;
3667         u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3668
3669         srrctl |= IXGBE_SRRCTL_DROP_EN;
3670
3671         IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3672 }
3673
3674 static void ixgbe_disable_rx_drop(struct ixgbe_adapter *adapter,
3675                                   struct ixgbe_ring *ring)
3676 {
3677         struct ixgbe_hw *hw = &adapter->hw;
3678         u8 reg_idx = ring->reg_idx;
3679         u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3680
3681         srrctl &= ~IXGBE_SRRCTL_DROP_EN;
3682
3683         IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3684 }
3685
3686 #ifdef CONFIG_IXGBE_DCB
3687 void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3688 #else
3689 static void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3690 #endif
3691 {
3692         int i;
3693         bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
3694
3695         if (adapter->ixgbe_ieee_pfc)
3696                 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
3697
3698         /*
3699          * We should set the drop enable bit if:
3700          *  SR-IOV is enabled
3701          *   or
3702          *  Number of Rx queues > 1 and flow control is disabled
3703          *
3704          *  This allows us to avoid head of line blocking for security
3705          *  and performance reasons.
3706          */
3707         if (adapter->num_vfs || (adapter->num_rx_queues > 1 &&
3708             !(adapter->hw.fc.current_mode & ixgbe_fc_tx_pause) && !pfc_en)) {
3709                 for (i = 0; i < adapter->num_rx_queues; i++)
3710                         ixgbe_enable_rx_drop(adapter, adapter->rx_ring[i]);
3711         } else {
3712                 for (i = 0; i < adapter->num_rx_queues; i++)
3713                         ixgbe_disable_rx_drop(adapter, adapter->rx_ring[i]);
3714         }
3715 }
3716
3717 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
3718
3719 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
3720                                    struct ixgbe_ring *rx_ring)
3721 {
3722         struct ixgbe_hw *hw = &adapter->hw;
3723         u32 srrctl;
3724         u8 reg_idx = rx_ring->reg_idx;
3725
3726         if (hw->mac.type == ixgbe_mac_82598EB) {
3727                 u16 mask = adapter->ring_feature[RING_F_RSS].mask;
3728
3729                 /*
3730                  * if VMDq is not active we must program one srrctl register
3731                  * per RSS queue since we have enabled RDRXCTL.MVMEN
3732                  */
3733                 reg_idx &= mask;
3734         }
3735
3736         /* configure header buffer length, needed for RSC */
3737         srrctl = IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
3738
3739         /* configure the packet buffer length */
3740         if (rx_ring->xsk_pool) {
3741                 u32 xsk_buf_len = xsk_pool_get_rx_frame_size(rx_ring->xsk_pool);
3742
3743                 /* If the MAC support setting RXDCTL.RLPML, the
3744                  * SRRCTL[n].BSIZEPKT is set to PAGE_SIZE and
3745                  * RXDCTL.RLPML is set to the actual UMEM buffer
3746                  * size. If not, then we are stuck with a 1k buffer
3747                  * size resolution. In this case frames larger than
3748                  * the UMEM buffer size viewed in a 1k resolution will
3749                  * be dropped.
3750                  */
3751                 if (hw->mac.type != ixgbe_mac_82599EB)
3752                         srrctl |= PAGE_SIZE >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3753                 else
3754                         srrctl |= xsk_buf_len >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3755         } else if (test_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state)) {
3756                 srrctl |= IXGBE_RXBUFFER_3K >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3757         } else {
3758                 srrctl |= IXGBE_RXBUFFER_2K >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3759         }
3760
3761         /* configure descriptor type */
3762         srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
3763
3764         IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3765 }
3766
3767 /**
3768  * ixgbe_rss_indir_tbl_entries - Return RSS indirection table entries
3769  * @adapter: device handle
3770  *
3771  *  - 82598/82599/X540:     128
3772  *  - X550(non-SRIOV mode): 512
3773  *  - X550(SRIOV mode):     64
3774  */
3775 u32 ixgbe_rss_indir_tbl_entries(struct ixgbe_adapter *adapter)
3776 {
3777         if (adapter->hw.mac.type < ixgbe_mac_X550)
3778                 return 128;
3779         else if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3780                 return 64;
3781         else
3782                 return 512;
3783 }
3784
3785 /**
3786  * ixgbe_store_key - Write the RSS key to HW
3787  * @adapter: device handle
3788  *
3789  * Write the RSS key stored in adapter.rss_key to HW.
3790  */
3791 void ixgbe_store_key(struct ixgbe_adapter *adapter)
3792 {
3793         struct ixgbe_hw *hw = &adapter->hw;
3794         int i;
3795
3796         for (i = 0; i < 10; i++)
3797                 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), adapter->rss_key[i]);
3798 }
3799
3800 /**
3801  * ixgbe_init_rss_key - Initialize adapter RSS key
3802  * @adapter: device handle
3803  *
3804  * Allocates and initializes the RSS key if it is not allocated.
3805  **/
3806 static inline int ixgbe_init_rss_key(struct ixgbe_adapter *adapter)
3807 {
3808         u32 *rss_key;
3809
3810         if (!adapter->rss_key) {
3811                 rss_key = kzalloc(IXGBE_RSS_KEY_SIZE, GFP_KERNEL);
3812                 if (unlikely(!rss_key))
3813                         return -ENOMEM;
3814
3815                 netdev_rss_key_fill(rss_key, IXGBE_RSS_KEY_SIZE);
3816                 adapter->rss_key = rss_key;
3817         }
3818
3819         return 0;
3820 }
3821
3822 /**
3823  * ixgbe_store_reta - Write the RETA table to HW
3824  * @adapter: device handle
3825  *
3826  * Write the RSS redirection table stored in adapter.rss_indir_tbl[] to HW.
3827  */
3828 void ixgbe_store_reta(struct ixgbe_adapter *adapter)
3829 {
3830         u32 i, reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3831         struct ixgbe_hw *hw = &adapter->hw;
3832         u32 reta = 0;
3833         u32 indices_multi;
3834         u8 *indir_tbl = adapter->rss_indir_tbl;
3835
3836         /* Fill out the redirection table as follows:
3837          *  - 82598:      8 bit wide entries containing pair of 4 bit RSS
3838          *    indices.
3839          *  - 82599/X540: 8 bit wide entries containing 4 bit RSS index
3840          *  - X550:       8 bit wide entries containing 6 bit RSS index
3841          */
3842         if (adapter->hw.mac.type == ixgbe_mac_82598EB)
3843                 indices_multi = 0x11;
3844         else
3845                 indices_multi = 0x1;
3846
3847         /* Write redirection table to HW */
3848         for (i = 0; i < reta_entries; i++) {
3849                 reta |= indices_multi * indir_tbl[i] << (i & 0x3) * 8;
3850                 if ((i & 3) == 3) {
3851                         if (i < 128)
3852                                 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
3853                         else
3854                                 IXGBE_WRITE_REG(hw, IXGBE_ERETA((i >> 2) - 32),
3855                                                 reta);
3856                         reta = 0;
3857                 }
3858         }
3859 }
3860
3861 /**
3862  * ixgbe_store_vfreta - Write the RETA table to HW (x550 devices in SRIOV mode)
3863  * @adapter: device handle
3864  *
3865  * Write the RSS redirection table stored in adapter.rss_indir_tbl[] to HW.
3866  */
3867 static void ixgbe_store_vfreta(struct ixgbe_adapter *adapter)
3868 {
3869         u32 i, reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3870         struct ixgbe_hw *hw = &adapter->hw;
3871         u32 vfreta = 0;
3872
3873         /* Write redirection table to HW */
3874         for (i = 0; i < reta_entries; i++) {
3875                 u16 pool = adapter->num_rx_pools;
3876
3877                 vfreta |= (u32)adapter->rss_indir_tbl[i] << (i & 0x3) * 8;
3878                 if ((i & 3) != 3)
3879                         continue;
3880
3881                 while (pool--)
3882                         IXGBE_WRITE_REG(hw,
3883                                         IXGBE_PFVFRETA(i >> 2, VMDQ_P(pool)),
3884                                         vfreta);
3885                 vfreta = 0;
3886         }
3887 }
3888
3889 static void ixgbe_setup_reta(struct ixgbe_adapter *adapter)
3890 {
3891         u32 i, j;
3892         u32 reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3893         u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3894
3895         /* Program table for at least 4 queues w/ SR-IOV so that VFs can
3896          * make full use of any rings they may have.  We will use the
3897          * PSRTYPE register to control how many rings we use within the PF.
3898          */
3899         if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) && (rss_i < 4))
3900                 rss_i = 4;
3901
3902         /* Fill out hash function seeds */
3903         ixgbe_store_key(adapter);
3904
3905         /* Fill out redirection table */
3906         memset(adapter->rss_indir_tbl, 0, sizeof(adapter->rss_indir_tbl));
3907
3908         for (i = 0, j = 0; i < reta_entries; i++, j++) {
3909                 if (j == rss_i)
3910                         j = 0;
3911
3912                 adapter->rss_indir_tbl[i] = j;
3913         }
3914
3915         ixgbe_store_reta(adapter);
3916 }
3917
3918 static void ixgbe_setup_vfreta(struct ixgbe_adapter *adapter)
3919 {
3920         struct ixgbe_hw *hw = &adapter->hw;
3921         u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3922         int i, j;
3923
3924         /* Fill out hash function seeds */
3925         for (i = 0; i < 10; i++) {
3926                 u16 pool = adapter->num_rx_pools;
3927
3928                 while (pool--)
3929                         IXGBE_WRITE_REG(hw,
3930                                         IXGBE_PFVFRSSRK(i, VMDQ_P(pool)),
3931                                         *(adapter->rss_key + i));
3932         }
3933
3934         /* Fill out the redirection table */
3935         for (i = 0, j = 0; i < 64; i++, j++) {
3936                 if (j == rss_i)
3937                         j = 0;
3938
3939                 adapter->rss_indir_tbl[i] = j;
3940         }
3941
3942         ixgbe_store_vfreta(adapter);
3943 }
3944
3945 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
3946 {
3947         struct ixgbe_hw *hw = &adapter->hw;
3948         u32 mrqc = 0, rss_field = 0, vfmrqc = 0;
3949         u32 rxcsum;
3950
3951         /* Disable indicating checksum in descriptor, enables RSS hash */
3952         rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
3953         rxcsum |= IXGBE_RXCSUM_PCSD;
3954         IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
3955
3956         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
3957                 if (adapter->ring_feature[RING_F_RSS].mask)
3958                         mrqc = IXGBE_MRQC_RSSEN;
3959         } else {
3960                 u8 tcs = adapter->hw_tcs;
3961
3962                 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3963                         if (tcs > 4)
3964                                 mrqc = IXGBE_MRQC_VMDQRT8TCEN;  /* 8 TCs */
3965                         else if (tcs > 1)
3966                                 mrqc = IXGBE_MRQC_VMDQRT4TCEN;  /* 4 TCs */
3967                         else if (adapter->ring_feature[RING_F_VMDQ].mask ==
3968                                  IXGBE_82599_VMDQ_4Q_MASK)
3969                                 mrqc = IXGBE_MRQC_VMDQRSS32EN;
3970                         else
3971                                 mrqc = IXGBE_MRQC_VMDQRSS64EN;
3972
3973                         /* Enable L3/L4 for Tx Switched packets only for X550,
3974                          * older devices do not support this feature
3975                          */
3976                         if (hw->mac.type >= ixgbe_mac_X550)
3977                                 mrqc |= IXGBE_MRQC_L3L4TXSWEN;
3978                 } else {
3979                         if (tcs > 4)
3980                                 mrqc = IXGBE_MRQC_RTRSS8TCEN;
3981                         else if (tcs > 1)
3982                                 mrqc = IXGBE_MRQC_RTRSS4TCEN;
3983                         else
3984                                 mrqc = IXGBE_MRQC_RSSEN;
3985                 }
3986         }
3987
3988         /* Perform hash on these packet types */
3989         rss_field |= IXGBE_MRQC_RSS_FIELD_IPV4 |
3990                      IXGBE_MRQC_RSS_FIELD_IPV4_TCP |
3991                      IXGBE_MRQC_RSS_FIELD_IPV6 |
3992                      IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
3993
3994         if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV4_UDP)
3995                 rss_field |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP;
3996         if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV6_UDP)
3997                 rss_field |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
3998
3999         if ((hw->mac.type >= ixgbe_mac_X550) &&
4000             (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)) {
4001                 u16 pool = adapter->num_rx_pools;
4002
4003                 /* Enable VF RSS mode */
4004                 mrqc |= IXGBE_MRQC_MULTIPLE_RSS;
4005                 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
4006
4007                 /* Setup RSS through the VF registers */
4008                 ixgbe_setup_vfreta(adapter);
4009                 vfmrqc = IXGBE_MRQC_RSSEN;
4010                 vfmrqc |= rss_field;
4011
4012                 while (pool--)
4013                         IXGBE_WRITE_REG(hw,
4014                                         IXGBE_PFVFMRQC(VMDQ_P(pool)),
4015                                         vfmrqc);
4016         } else {
4017                 ixgbe_setup_reta(adapter);
4018                 mrqc |= rss_field;
4019                 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
4020         }
4021 }
4022
4023 /**
4024  * ixgbe_configure_rscctl - enable RSC for the indicated ring
4025  * @adapter: address of board private structure
4026  * @ring: structure containing ring specific data
4027  **/
4028 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
4029                                    struct ixgbe_ring *ring)
4030 {
4031         struct ixgbe_hw *hw = &adapter->hw;
4032         u32 rscctrl;
4033         u8 reg_idx = ring->reg_idx;
4034
4035         if (!ring_is_rsc_enabled(ring))
4036                 return;
4037
4038         rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
4039         rscctrl |= IXGBE_RSCCTL_RSCEN;
4040         /*
4041          * we must limit the number of descriptors so that the
4042          * total size of max desc * buf_len is not greater
4043          * than 65536
4044          */
4045         rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
4046         IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
4047 }
4048
4049 #define IXGBE_MAX_RX_DESC_POLL 10
4050 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
4051                                        struct ixgbe_ring *ring)
4052 {
4053         struct ixgbe_hw *hw = &adapter->hw;
4054         int wait_loop = IXGBE_MAX_RX_DESC_POLL;
4055         u32 rxdctl;
4056         u8 reg_idx = ring->reg_idx;
4057
4058         if (ixgbe_removed(hw->hw_addr))
4059                 return;
4060         /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
4061         if (hw->mac.type == ixgbe_mac_82598EB &&
4062             !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
4063                 return;
4064
4065         do {
4066                 usleep_range(1000, 2000);
4067                 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
4068         } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
4069
4070         if (!wait_loop) {
4071                 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
4072                       "the polling period\n", reg_idx);
4073         }
4074 }
4075
4076 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
4077                              struct ixgbe_ring *ring)
4078 {
4079         struct ixgbe_hw *hw = &adapter->hw;
4080         union ixgbe_adv_rx_desc *rx_desc;
4081         u64 rdba = ring->dma;
4082         u32 rxdctl;
4083         u8 reg_idx = ring->reg_idx;
4084
4085         xdp_rxq_info_unreg_mem_model(&ring->xdp_rxq);
4086         ring->xsk_pool = ixgbe_xsk_pool(adapter, ring);
4087         if (ring->xsk_pool) {
4088                 WARN_ON(xdp_rxq_info_reg_mem_model(&ring->xdp_rxq,
4089                                                    MEM_TYPE_XSK_BUFF_POOL,
4090                                                    NULL));
4091                 xsk_pool_set_rxq_info(ring->xsk_pool, &ring->xdp_rxq);
4092         } else {
4093                 WARN_ON(xdp_rxq_info_reg_mem_model(&ring->xdp_rxq,
4094                                                    MEM_TYPE_PAGE_SHARED, NULL));
4095         }
4096
4097         /* disable queue to avoid use of these values while updating state */
4098         rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
4099         rxdctl &= ~IXGBE_RXDCTL_ENABLE;
4100
4101         /* write value back with RXDCTL.ENABLE bit cleared */
4102         IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
4103         IXGBE_WRITE_FLUSH(hw);
4104
4105         IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
4106         IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
4107         IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
4108                         ring->count * sizeof(union ixgbe_adv_rx_desc));
4109         /* Force flushing of IXGBE_RDLEN to prevent MDD */
4110         IXGBE_WRITE_FLUSH(hw);
4111
4112         IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
4113         IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
4114         ring->tail = adapter->io_addr + IXGBE_RDT(reg_idx);
4115
4116         ixgbe_configure_srrctl(adapter, ring);
4117         ixgbe_configure_rscctl(adapter, ring);
4118
4119         if (hw->mac.type == ixgbe_mac_82598EB) {
4120                 /*
4121                  * enable cache line friendly hardware writes:
4122                  * PTHRESH=32 descriptors (half the internal cache),
4123                  * this also removes ugly rx_no_buffer_count increment
4124                  * HTHRESH=4 descriptors (to minimize latency on fetch)
4125                  * WTHRESH=8 burst writeback up to two cache lines
4126                  */
4127                 rxdctl &= ~0x3FFFFF;
4128                 rxdctl |=  0x080420;
4129 #if (PAGE_SIZE < 8192)
4130         /* RXDCTL.RLPML does not work on 82599 */
4131         } else if (hw->mac.type != ixgbe_mac_82599EB) {
4132                 rxdctl &= ~(IXGBE_RXDCTL_RLPMLMASK |
4133                             IXGBE_RXDCTL_RLPML_EN);
4134
4135                 /* Limit the maximum frame size so we don't overrun the skb.
4136                  * This can happen in SRIOV mode when the MTU of the VF is
4137                  * higher than the MTU of the PF.
4138                  */
4139                 if (ring_uses_build_skb(ring) &&
4140                     !test_bit(__IXGBE_RX_3K_BUFFER, &ring->state))
4141                         rxdctl |= IXGBE_MAX_2K_FRAME_BUILD_SKB |
4142                                   IXGBE_RXDCTL_RLPML_EN;
4143 #endif
4144         }
4145
4146         ring->rx_offset = ixgbe_rx_offset(ring);
4147
4148         if (ring->xsk_pool && hw->mac.type != ixgbe_mac_82599EB) {
4149                 u32 xsk_buf_len = xsk_pool_get_rx_frame_size(ring->xsk_pool);
4150
4151                 rxdctl &= ~(IXGBE_RXDCTL_RLPMLMASK |
4152                             IXGBE_RXDCTL_RLPML_EN);
4153                 rxdctl |= xsk_buf_len | IXGBE_RXDCTL_RLPML_EN;
4154
4155                 ring->rx_buf_len = xsk_buf_len;
4156         }
4157
4158         /* initialize rx_buffer_info */
4159         memset(ring->rx_buffer_info, 0,
4160                sizeof(struct ixgbe_rx_buffer) * ring->count);
4161
4162         /* initialize Rx descriptor 0 */
4163         rx_desc = IXGBE_RX_DESC(ring, 0);
4164         rx_desc->wb.upper.length = 0;
4165
4166         /* enable receive descriptor ring */
4167         rxdctl |= IXGBE_RXDCTL_ENABLE;
4168         IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
4169
4170         ixgbe_rx_desc_queue_enable(adapter, ring);
4171         if (ring->xsk_pool)
4172                 ixgbe_alloc_rx_buffers_zc(ring, ixgbe_desc_unused(ring));
4173         else
4174                 ixgbe_alloc_rx_buffers(ring, ixgbe_desc_unused(ring));
4175 }
4176
4177 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
4178 {
4179         struct ixgbe_hw *hw = &adapter->hw;
4180         int rss_i = adapter->ring_feature[RING_F_RSS].indices;
4181         u16 pool = adapter->num_rx_pools;
4182
4183         /* PSRTYPE must be initialized in non 82598 adapters */
4184         u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
4185                       IXGBE_PSRTYPE_UDPHDR |
4186                       IXGBE_PSRTYPE_IPV4HDR |
4187                       IXGBE_PSRTYPE_L2HDR |
4188                       IXGBE_PSRTYPE_IPV6HDR;
4189
4190         if (hw->mac.type == ixgbe_mac_82598EB)
4191                 return;
4192
4193         if (rss_i > 3)
4194                 psrtype |= 2u << 29;
4195         else if (rss_i > 1)
4196                 psrtype |= 1u << 29;
4197
4198         while (pool--)
4199                 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(VMDQ_P(pool)), psrtype);
4200 }
4201
4202 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
4203 {
4204         struct ixgbe_hw *hw = &adapter->hw;
4205         u16 pool = adapter->num_rx_pools;
4206         u32 reg_offset, vf_shift, vmolr;
4207         u32 gcr_ext, vmdctl;
4208         int i;
4209
4210         if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
4211                 return;
4212
4213         vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
4214         vmdctl |= IXGBE_VMD_CTL_VMDQ_EN;
4215         vmdctl &= ~IXGBE_VT_CTL_POOL_MASK;
4216         vmdctl |= VMDQ_P(0) << IXGBE_VT_CTL_POOL_SHIFT;
4217         vmdctl |= IXGBE_VT_CTL_REPLEN;
4218         IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl);
4219
4220         /* accept untagged packets until a vlan tag is
4221          * specifically set for the VMDQ queue/pool
4222          */
4223         vmolr = IXGBE_VMOLR_AUPE;
4224         while (pool--)
4225                 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(VMDQ_P(pool)), vmolr);
4226
4227         vf_shift = VMDQ_P(0) % 32;
4228         reg_offset = (VMDQ_P(0) >= 32) ? 1 : 0;
4229
4230         /* Enable only the PF's pool for Tx/Rx */
4231         IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), GENMASK(31, vf_shift));
4232         IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), reg_offset - 1);
4233         IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), GENMASK(31, vf_shift));
4234         IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), reg_offset - 1);
4235         if (adapter->bridge_mode == BRIDGE_MODE_VEB)
4236                 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
4237
4238         /* Map PF MAC address in RAR Entry 0 to first pool following VFs */
4239         hw->mac.ops.set_vmdq(hw, 0, VMDQ_P(0));
4240
4241         /* clear VLAN promisc flag so VFTA will be updated if necessary */
4242         adapter->flags2 &= ~IXGBE_FLAG2_VLAN_PROMISC;
4243
4244         /*
4245          * Set up VF register offsets for selected VT Mode,
4246          * i.e. 32 or 64 VFs for SR-IOV
4247          */
4248         switch (adapter->ring_feature[RING_F_VMDQ].mask) {
4249         case IXGBE_82599_VMDQ_8Q_MASK:
4250                 gcr_ext = IXGBE_GCR_EXT_VT_MODE_16;
4251                 break;
4252         case IXGBE_82599_VMDQ_4Q_MASK:
4253                 gcr_ext = IXGBE_GCR_EXT_VT_MODE_32;
4254                 break;
4255         default:
4256                 gcr_ext = IXGBE_GCR_EXT_VT_MODE_64;
4257                 break;
4258         }
4259
4260         IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
4261
4262         for (i = 0; i < adapter->num_vfs; i++) {
4263                 /* configure spoof checking */
4264                 ixgbe_ndo_set_vf_spoofchk(adapter->netdev, i,
4265                                           adapter->vfinfo[i].spoofchk_enabled);
4266
4267                 /* Enable/Disable RSS query feature  */
4268                 ixgbe_ndo_set_vf_rss_query_en(adapter->netdev, i,
4269                                           adapter->vfinfo[i].rss_query_enabled);
4270         }
4271 }
4272
4273 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
4274 {
4275         struct ixgbe_hw *hw = &adapter->hw;
4276         struct net_device *netdev = adapter->netdev;
4277         int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
4278         struct ixgbe_ring *rx_ring;
4279         int i;
4280         u32 mhadd, hlreg0;
4281
4282 #ifdef IXGBE_FCOE
4283         /* adjust max frame to be able to do baby jumbo for FCoE */
4284         if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
4285             (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
4286                 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
4287
4288 #endif /* IXGBE_FCOE */
4289
4290         /* adjust max frame to be at least the size of a standard frame */
4291         if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
4292                 max_frame = (ETH_FRAME_LEN + ETH_FCS_LEN);
4293
4294         mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
4295         if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
4296                 mhadd &= ~IXGBE_MHADD_MFS_MASK;
4297                 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
4298
4299                 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
4300         }
4301
4302         hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
4303         /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
4304         hlreg0 |= IXGBE_HLREG0_JUMBOEN;
4305         IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
4306
4307         /*
4308          * Setup the HW Rx Head and Tail Descriptor Pointers and
4309          * the Base and Length of the Rx Descriptor Ring
4310          */
4311         for (i = 0; i < adapter->num_rx_queues; i++) {
4312                 rx_ring = adapter->rx_ring[i];
4313
4314                 clear_ring_rsc_enabled(rx_ring);
4315                 clear_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4316                 clear_bit(__IXGBE_RX_BUILD_SKB_ENABLED, &rx_ring->state);
4317
4318                 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
4319                         set_ring_rsc_enabled(rx_ring);
4320
4321                 if (test_bit(__IXGBE_RX_FCOE, &rx_ring->state))
4322                         set_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4323
4324                 if (adapter->flags2 & IXGBE_FLAG2_RX_LEGACY)
4325                         continue;
4326
4327                 set_bit(__IXGBE_RX_BUILD_SKB_ENABLED, &rx_ring->state);
4328
4329 #if (PAGE_SIZE < 8192)
4330                 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
4331                         set_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4332
4333                 if (IXGBE_2K_TOO_SMALL_WITH_PADDING ||
4334                     (max_frame > (ETH_FRAME_LEN + ETH_FCS_LEN)))
4335                         set_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4336 #endif
4337         }
4338 }
4339
4340 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
4341 {
4342         struct ixgbe_hw *hw = &adapter->hw;
4343         u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
4344
4345         switch (hw->mac.type) {
4346         case ixgbe_mac_82598EB:
4347                 /*
4348                  * For VMDq support of different descriptor types or
4349                  * buffer sizes through the use of multiple SRRCTL
4350                  * registers, RDRXCTL.MVMEN must be set to 1
4351                  *
4352                  * also, the manual doesn't mention it clearly but DCA hints
4353                  * will only use queue 0's tags unless this bit is set.  Side
4354                  * effects of setting this bit are only that SRRCTL must be
4355                  * fully programmed [0..15]
4356                  */
4357                 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
4358                 break;
4359         case ixgbe_mac_X550:
4360         case ixgbe_mac_X550EM_x:
4361         case ixgbe_mac_x550em_a:
4362                 if (adapter->num_vfs)
4363                         rdrxctl |= IXGBE_RDRXCTL_PSP;
4364                 fallthrough;
4365         case ixgbe_mac_82599EB:
4366         case ixgbe_mac_X540:
4367                 /* Disable RSC for ACK packets */
4368                 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
4369                    (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
4370                 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
4371                 /* hardware requires some bits to be set by default */
4372                 rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
4373                 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
4374                 break;
4375         default:
4376                 /* We should do nothing since we don't know this hardware */
4377                 return;
4378         }
4379
4380         IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
4381 }
4382
4383 /**
4384  * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
4385  * @adapter: board private structure
4386  *
4387  * Configure the Rx unit of the MAC after a reset.
4388  **/
4389 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
4390 {
4391         struct ixgbe_hw *hw = &adapter->hw;
4392         int i;
4393         u32 rxctrl, rfctl;
4394
4395         /* disable receives while setting up the descriptors */
4396         hw->mac.ops.disable_rx(hw);
4397
4398         ixgbe_setup_psrtype(adapter);
4399         ixgbe_setup_rdrxctl(adapter);
4400
4401         /* RSC Setup */
4402         rfctl = IXGBE_READ_REG(hw, IXGBE_RFCTL);
4403         rfctl &= ~IXGBE_RFCTL_RSC_DIS;
4404         if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
4405                 rfctl |= IXGBE_RFCTL_RSC_DIS;
4406
4407         /* disable NFS filtering */
4408         rfctl |= (IXGBE_RFCTL_NFSW_DIS | IXGBE_RFCTL_NFSR_DIS);
4409         IXGBE_WRITE_REG(hw, IXGBE_RFCTL, rfctl);
4410
4411         /* Program registers for the distribution of queues */
4412         ixgbe_setup_mrqc(adapter);
4413
4414         /* set_rx_buffer_len must be called before ring initialization */
4415         ixgbe_set_rx_buffer_len(adapter);
4416
4417         /*
4418          * Setup the HW Rx Head and Tail Descriptor Pointers and
4419          * the Base and Length of the Rx Descriptor Ring
4420          */
4421         for (i = 0; i < adapter->num_rx_queues; i++)
4422                 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
4423
4424         rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4425         /* disable drop enable for 82598 parts */
4426         if (hw->mac.type == ixgbe_mac_82598EB)
4427                 rxctrl |= IXGBE_RXCTRL_DMBYPS;
4428
4429         /* enable all receives */
4430         rxctrl |= IXGBE_RXCTRL_RXEN;
4431         hw->mac.ops.enable_rx_dma(hw, rxctrl);
4432 }
4433
4434 static int ixgbe_vlan_rx_add_vid(struct net_device *netdev,
4435                                  __be16 proto, u16 vid)
4436 {
4437         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4438         struct ixgbe_hw *hw = &adapter->hw;
4439
4440         /* add VID to filter table */
4441         if (!vid || !(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4442                 hw->mac.ops.set_vfta(&adapter->hw, vid, VMDQ_P(0), true, !!vid);
4443
4444         set_bit(vid, adapter->active_vlans);
4445
4446         return 0;
4447 }
4448
4449 static int ixgbe_find_vlvf_entry(struct ixgbe_hw *hw, u32 vlan)
4450 {
4451         u32 vlvf;
4452         int idx;
4453
4454         /* short cut the special case */
4455         if (vlan == 0)
4456                 return 0;
4457
4458         /* Search for the vlan id in the VLVF entries */
4459         for (idx = IXGBE_VLVF_ENTRIES; --idx;) {
4460                 vlvf = IXGBE_READ_REG(hw, IXGBE_VLVF(idx));
4461                 if ((vlvf & VLAN_VID_MASK) == vlan)
4462                         break;
4463         }
4464
4465         return idx;
4466 }
4467
4468 void ixgbe_update_pf_promisc_vlvf(struct ixgbe_adapter *adapter, u32 vid)
4469 {
4470         struct ixgbe_hw *hw = &adapter->hw;
4471         u32 bits, word;
4472         int idx;
4473
4474         idx = ixgbe_find_vlvf_entry(hw, vid);
4475         if (!idx)
4476                 return;
4477
4478         /* See if any other pools are set for this VLAN filter
4479          * entry other than the PF.
4480          */
4481         word = idx * 2 + (VMDQ_P(0) / 32);
4482         bits = ~BIT(VMDQ_P(0) % 32);
4483         bits &= IXGBE_READ_REG(hw, IXGBE_VLVFB(word));
4484
4485         /* Disable the filter so this falls into the default pool. */
4486         if (!bits && !IXGBE_READ_REG(hw, IXGBE_VLVFB(word ^ 1))) {
4487                 if (!(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4488                         IXGBE_WRITE_REG(hw, IXGBE_VLVFB(word), 0);
4489                 IXGBE_WRITE_REG(hw, IXGBE_VLVF(idx), 0);
4490         }
4491 }
4492
4493 static int ixgbe_vlan_rx_kill_vid(struct net_device *netdev,
4494                                   __be16 proto, u16 vid)
4495 {
4496         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4497         struct ixgbe_hw *hw = &adapter->hw;
4498
4499         /* remove VID from filter table */
4500         if (vid && !(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4501                 hw->mac.ops.set_vfta(hw, vid, VMDQ_P(0), false, true);
4502
4503         clear_bit(vid, adapter->active_vlans);
4504
4505         return 0;
4506 }
4507
4508 /**
4509  * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
4510  * @adapter: driver data
4511  */
4512 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
4513 {
4514         struct ixgbe_hw *hw = &adapter->hw;
4515         u32 vlnctrl;
4516         int i, j;
4517
4518         switch (hw->mac.type) {
4519         case ixgbe_mac_82598EB:
4520                 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4521                 vlnctrl &= ~IXGBE_VLNCTRL_VME;
4522                 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4523                 break;
4524         case ixgbe_mac_82599EB:
4525         case ixgbe_mac_X540:
4526         case ixgbe_mac_X550:
4527         case ixgbe_mac_X550EM_x:
4528         case ixgbe_mac_x550em_a:
4529                 for (i = 0; i < adapter->num_rx_queues; i++) {
4530                         struct ixgbe_ring *ring = adapter->rx_ring[i];
4531
4532                         if (!netif_is_ixgbe(ring->netdev))
4533                                 continue;
4534
4535                         j = ring->reg_idx;
4536                         vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
4537                         vlnctrl &= ~IXGBE_RXDCTL_VME;
4538                         IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
4539                 }
4540                 break;
4541         default:
4542                 break;
4543         }
4544 }
4545
4546 /**
4547  * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
4548  * @adapter: driver data
4549  */
4550 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
4551 {
4552         struct ixgbe_hw *hw = &adapter->hw;
4553         u32 vlnctrl;
4554         int i, j;
4555
4556         switch (hw->mac.type) {
4557         case ixgbe_mac_82598EB:
4558                 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4559                 vlnctrl |= IXGBE_VLNCTRL_VME;
4560                 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4561                 break;
4562         case ixgbe_mac_82599EB:
4563         case ixgbe_mac_X540:
4564         case ixgbe_mac_X550:
4565         case ixgbe_mac_X550EM_x:
4566         case ixgbe_mac_x550em_a:
4567                 for (i = 0; i < adapter->num_rx_queues; i++) {
4568                         struct ixgbe_ring *ring = adapter->rx_ring[i];
4569
4570                         if (!netif_is_ixgbe(ring->netdev))
4571                                 continue;
4572
4573                         j = ring->reg_idx;
4574                         vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
4575                         vlnctrl |= IXGBE_RXDCTL_VME;
4576                         IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
4577                 }
4578                 break;
4579         default:
4580                 break;
4581         }
4582 }
4583
4584 static void ixgbe_vlan_promisc_enable(struct ixgbe_adapter *adapter)
4585 {
4586         struct ixgbe_hw *hw = &adapter->hw;
4587         u32 vlnctrl, i;
4588
4589         vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4590
4591         if (adapter->flags & IXGBE_FLAG_VMDQ_ENABLED) {
4592         /* For VMDq and SR-IOV we must leave VLAN filtering enabled */
4593                 vlnctrl |= IXGBE_VLNCTRL_VFE;
4594                 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4595         } else {
4596                 vlnctrl &= ~IXGBE_VLNCTRL_VFE;
4597                 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4598                 return;
4599         }
4600
4601         /* Nothing to do for 82598 */
4602         if (hw->mac.type == ixgbe_mac_82598EB)
4603                 return;
4604
4605         /* We are already in VLAN promisc, nothing to do */
4606         if (adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC)
4607                 return;
4608
4609         /* Set flag so we don't redo unnecessary work */
4610         adapter->flags2 |= IXGBE_FLAG2_VLAN_PROMISC;
4611
4612         /* Add PF to all active pools */
4613         for (i = IXGBE_VLVF_ENTRIES; --i;) {
4614                 u32 reg_offset = IXGBE_VLVFB(i * 2 + VMDQ_P(0) / 32);
4615                 u32 vlvfb = IXGBE_READ_REG(hw, reg_offset);
4616
4617                 vlvfb |= BIT(VMDQ_P(0) % 32);
4618                 IXGBE_WRITE_REG(hw, reg_offset, vlvfb);
4619         }
4620
4621         /* Set all bits in the VLAN filter table array */
4622         for (i = hw->mac.vft_size; i--;)
4623                 IXGBE_WRITE_REG(hw, IXGBE_VFTA(i), ~0U);
4624 }
4625
4626 #define VFTA_BLOCK_SIZE 8
4627 static void ixgbe_scrub_vfta(struct ixgbe_adapter *adapter, u32 vfta_offset)
4628 {
4629         struct ixgbe_hw *hw = &adapter->hw;
4630         u32 vfta[VFTA_BLOCK_SIZE] = { 0 };
4631         u32 vid_start = vfta_offset * 32;
4632         u32 vid_end = vid_start + (VFTA_BLOCK_SIZE * 32);
4633         u32 i, vid, word, bits;
4634
4635         for (i = IXGBE_VLVF_ENTRIES; --i;) {
4636                 u32 vlvf = IXGBE_READ_REG(hw, IXGBE_VLVF(i));
4637
4638                 /* pull VLAN ID from VLVF */
4639                 vid = vlvf & VLAN_VID_MASK;
4640
4641                 /* only concern outselves with a certain range */
4642                 if (vid < vid_start || vid >= vid_end)
4643                         continue;
4644
4645                 if (vlvf) {
4646                         /* record VLAN ID in VFTA */
4647                         vfta[(vid - vid_start) / 32] |= BIT(vid % 32);
4648
4649                         /* if PF is part of this then continue */
4650                         if (test_bit(vid, adapter->active_vlans))
4651                                 continue;
4652                 }
4653
4654                 /* remove PF from the pool */
4655                 word = i * 2 + VMDQ_P(0) / 32;
4656                 bits = ~BIT(VMDQ_P(0) % 32);
4657                 bits &= IXGBE_READ_REG(hw, IXGBE_VLVFB(word));
4658                 IXGBE_WRITE_REG(hw, IXGBE_VLVFB(word), bits);
4659         }
4660
4661         /* extract values from active_vlans and write back to VFTA */
4662         for (i = VFTA_BLOCK_SIZE; i--;) {
4663                 vid = (vfta_offset + i) * 32;
4664                 word = vid / BITS_PER_LONG;
4665                 bits = vid % BITS_PER_LONG;
4666
4667                 vfta[i] |= adapter->active_vlans[word] >> bits;
4668
4669                 IXGBE_WRITE_REG(hw, IXGBE_VFTA(vfta_offset + i), vfta[i]);
4670         }
4671 }
4672
4673 static void ixgbe_vlan_promisc_disable(struct ixgbe_adapter *adapter)
4674 {
4675         struct ixgbe_hw *hw = &adapter->hw;
4676         u32 vlnctrl, i;
4677
4678         /* Set VLAN filtering to enabled */
4679         vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4680         vlnctrl |= IXGBE_VLNCTRL_VFE;
4681         IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4682
4683         if (!(adapter->flags & IXGBE_FLAG_VMDQ_ENABLED) ||
4684             hw->mac.type == ixgbe_mac_82598EB)
4685                 return;
4686
4687         /* We are not in VLAN promisc, nothing to do */
4688         if (!(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4689                 return;
4690
4691         /* Set flag so we don't redo unnecessary work */
4692         adapter->flags2 &= ~IXGBE_FLAG2_VLAN_PROMISC;
4693
4694         for (i = 0; i < hw->mac.vft_size; i += VFTA_BLOCK_SIZE)
4695                 ixgbe_scrub_vfta(adapter, i);
4696 }
4697
4698 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
4699 {
4700         u16 vid = 1;
4701
4702         ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0);
4703
4704         for_each_set_bit_from(vid, adapter->active_vlans, VLAN_N_VID)
4705                 ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
4706 }
4707
4708 /**
4709  * ixgbe_write_mc_addr_list - write multicast addresses to MTA
4710  * @netdev: network interface device structure
4711  *
4712  * Writes multicast address list to the MTA hash table.
4713  * Returns: -ENOMEM on failure
4714  *                0 on no addresses written
4715  *                X on writing X addresses to MTA
4716  **/
4717 static int ixgbe_write_mc_addr_list(struct net_device *netdev)
4718 {
4719         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4720         struct ixgbe_hw *hw = &adapter->hw;
4721
4722         if (!netif_running(netdev))
4723                 return 0;
4724
4725         if (hw->mac.ops.update_mc_addr_list)
4726                 hw->mac.ops.update_mc_addr_list(hw, netdev);
4727         else
4728                 return -ENOMEM;
4729
4730 #ifdef CONFIG_PCI_IOV
4731         ixgbe_restore_vf_multicasts(adapter);
4732 #endif
4733
4734         return netdev_mc_count(netdev);
4735 }
4736
4737 #ifdef CONFIG_PCI_IOV
4738 void ixgbe_full_sync_mac_table(struct ixgbe_adapter *adapter)
4739 {
4740         struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4741         struct ixgbe_hw *hw = &adapter->hw;
4742         int i;
4743
4744         for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4745                 mac_table->state &= ~IXGBE_MAC_STATE_MODIFIED;
4746
4747                 if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4748                         hw->mac.ops.set_rar(hw, i,
4749                                             mac_table->addr,
4750                                             mac_table->pool,
4751                                             IXGBE_RAH_AV);
4752                 else
4753                         hw->mac.ops.clear_rar(hw, i);
4754         }
4755 }
4756
4757 #endif
4758 static void ixgbe_sync_mac_table(struct ixgbe_adapter *adapter)
4759 {
4760         struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4761         struct ixgbe_hw *hw = &adapter->hw;
4762         int i;
4763
4764         for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4765                 if (!(mac_table->state & IXGBE_MAC_STATE_MODIFIED))
4766                         continue;
4767
4768                 mac_table->state &= ~IXGBE_MAC_STATE_MODIFIED;
4769
4770                 if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4771                         hw->mac.ops.set_rar(hw, i,
4772                                             mac_table->addr,
4773                                             mac_table->pool,
4774                                             IXGBE_RAH_AV);
4775                 else
4776                         hw->mac.ops.clear_rar(hw, i);
4777         }
4778 }
4779
4780 static void ixgbe_flush_sw_mac_table(struct ixgbe_adapter *adapter)
4781 {
4782         struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4783         struct ixgbe_hw *hw = &adapter->hw;
4784         int i;
4785
4786         for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4787                 mac_table->state |= IXGBE_MAC_STATE_MODIFIED;
4788                 mac_table->state &= ~IXGBE_MAC_STATE_IN_USE;
4789         }
4790
4791         ixgbe_sync_mac_table(adapter);
4792 }
4793
4794 static int ixgbe_available_rars(struct ixgbe_adapter *adapter, u16 pool)
4795 {
4796         struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4797         struct ixgbe_hw *hw = &adapter->hw;
4798         int i, count = 0;
4799
4800         for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4801                 /* do not count default RAR as available */
4802                 if (mac_table->state & IXGBE_MAC_STATE_DEFAULT)
4803                         continue;
4804
4805                 /* only count unused and addresses that belong to us */
4806                 if (mac_table->state & IXGBE_MAC_STATE_IN_USE) {
4807                         if (mac_table->pool != pool)
4808                                 continue;
4809                 }
4810
4811                 count++;
4812         }
4813
4814         return count;
4815 }
4816
4817 /* this function destroys the first RAR entry */
4818 static void ixgbe_mac_set_default_filter(struct ixgbe_adapter *adapter)
4819 {
4820         struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4821         struct ixgbe_hw *hw = &adapter->hw;
4822
4823         memcpy(&mac_table->addr, hw->mac.addr, ETH_ALEN);
4824         mac_table->pool = VMDQ_P(0);
4825
4826         mac_table->state = IXGBE_MAC_STATE_DEFAULT | IXGBE_MAC_STATE_IN_USE;
4827
4828         hw->mac.ops.set_rar(hw, 0, mac_table->addr, mac_table->pool,
4829                             IXGBE_RAH_AV);
4830 }
4831
4832 int ixgbe_add_mac_filter(struct ixgbe_adapter *adapter,
4833                          const u8 *addr, u16 pool)
4834 {
4835         struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4836         struct ixgbe_hw *hw = &adapter->hw;
4837         int i;
4838
4839         if (is_zero_ether_addr(addr))
4840                 return -EINVAL;
4841
4842         for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4843                 if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4844                         continue;
4845
4846                 ether_addr_copy(mac_table->addr, addr);
4847                 mac_table->pool = pool;
4848
4849                 mac_table->state |= IXGBE_MAC_STATE_MODIFIED |
4850                                     IXGBE_MAC_STATE_IN_USE;
4851
4852                 ixgbe_sync_mac_table(adapter);
4853
4854                 return i;
4855         }
4856
4857         return -ENOMEM;
4858 }
4859
4860 int ixgbe_del_mac_filter(struct ixgbe_adapter *adapter,
4861                          const u8 *addr, u16 pool)
4862 {
4863         struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4864         struct ixgbe_hw *hw = &adapter->hw;
4865         int i;
4866
4867         if (is_zero_ether_addr(addr))
4868                 return -EINVAL;
4869
4870         /* search table for addr, if found clear IN_USE flag and sync */
4871         for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4872                 /* we can only delete an entry if it is in use */
4873                 if (!(mac_table->state & IXGBE_MAC_STATE_IN_USE))
4874                         continue;
4875                 /* we only care about entries that belong to the given pool */
4876                 if (mac_table->pool != pool)
4877                         continue;
4878                 /* we only care about a specific MAC address */
4879                 if (!ether_addr_equal(addr, mac_table->addr))
4880                         continue;
4881
4882                 mac_table->state |= IXGBE_MAC_STATE_MODIFIED;
4883                 mac_table->state &= ~IXGBE_MAC_STATE_IN_USE;
4884
4885                 ixgbe_sync_mac_table(adapter);
4886
4887                 return 0;
4888         }
4889
4890         return -ENOMEM;
4891 }
4892
4893 static int ixgbe_uc_sync(struct net_device *netdev, const unsigned char *addr)
4894 {
4895         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4896         int ret;
4897
4898         ret = ixgbe_add_mac_filter(adapter, addr, VMDQ_P(0));
4899
4900         return min_t(int, ret, 0);
4901 }
4902
4903 static int ixgbe_uc_unsync(struct net_device *netdev, const unsigned char *addr)
4904 {
4905         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4906
4907         ixgbe_del_mac_filter(adapter, addr, VMDQ_P(0));
4908
4909         return 0;
4910 }
4911
4912 /**
4913  * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
4914  * @netdev: network interface device structure
4915  *
4916  * The set_rx_method entry point is called whenever the unicast/multicast
4917  * address list or the network interface flags are updated.  This routine is
4918  * responsible for configuring the hardware for proper unicast, multicast and
4919  * promiscuous mode.
4920  **/
4921 void ixgbe_set_rx_mode(struct net_device *netdev)
4922 {
4923         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4924         struct ixgbe_hw *hw = &adapter->hw;
4925         u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
4926         netdev_features_t features = netdev->features;
4927         int count;
4928
4929         /* Check for Promiscuous and All Multicast modes */
4930         fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4931
4932         /* set all bits that we expect to always be set */
4933         fctrl &= ~IXGBE_FCTRL_SBP; /* disable store-bad-packets */
4934         fctrl |= IXGBE_FCTRL_BAM;
4935         fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
4936         fctrl |= IXGBE_FCTRL_PMCF;
4937
4938         /* clear the bits we are changing the status of */
4939         fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
4940         if (netdev->flags & IFF_PROMISC) {
4941                 hw->addr_ctrl.user_set_promisc = true;
4942                 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
4943                 vmolr |= IXGBE_VMOLR_MPE;
4944                 features &= ~NETIF_F_HW_VLAN_CTAG_FILTER;
4945         } else {
4946                 if (netdev->flags & IFF_ALLMULTI) {
4947                         fctrl |= IXGBE_FCTRL_MPE;
4948                         vmolr |= IXGBE_VMOLR_MPE;
4949                 }
4950                 hw->addr_ctrl.user_set_promisc = false;
4951         }
4952
4953         /*
4954          * Write addresses to available RAR registers, if there is not
4955          * sufficient space to store all the addresses then enable
4956          * unicast promiscuous mode
4957          */
4958         if (__dev_uc_sync(netdev, ixgbe_uc_sync, ixgbe_uc_unsync)) {
4959                 fctrl |= IXGBE_FCTRL_UPE;
4960                 vmolr |= IXGBE_VMOLR_ROPE;
4961         }
4962
4963         /* Write addresses to the MTA, if the attempt fails
4964          * then we should just turn on promiscuous mode so
4965          * that we can at least receive multicast traffic
4966          */
4967         count = ixgbe_write_mc_addr_list(netdev);
4968         if (count < 0) {
4969                 fctrl |= IXGBE_FCTRL_MPE;
4970                 vmolr |= IXGBE_VMOLR_MPE;
4971         } else if (count) {
4972                 vmolr |= IXGBE_VMOLR_ROMPE;
4973         }
4974
4975         if (hw->mac.type != ixgbe_mac_82598EB) {
4976                 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(VMDQ_P(0))) &
4977                          ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
4978                            IXGBE_VMOLR_ROPE);
4979                 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(VMDQ_P(0)), vmolr);
4980         }
4981
4982         /* This is useful for sniffing bad packets. */
4983         if (features & NETIF_F_RXALL) {
4984                 /* UPE and MPE will be handled by normal PROMISC logic
4985                  * in e1000e_set_rx_mode */
4986                 fctrl |= (IXGBE_FCTRL_SBP | /* Receive bad packets */
4987                           IXGBE_FCTRL_BAM | /* RX All Bcast Pkts */
4988                           IXGBE_FCTRL_PMCF); /* RX All MAC Ctrl Pkts */
4989
4990                 fctrl &= ~(IXGBE_FCTRL_DPF);
4991                 /* NOTE:  VLAN filtering is disabled by setting PROMISC */
4992         }
4993
4994         IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
4995
4996         if (features & NETIF_F_HW_VLAN_CTAG_RX)
4997                 ixgbe_vlan_strip_enable(adapter);
4998         else
4999                 ixgbe_vlan_strip_disable(adapter);
5000
5001         if (features & NETIF_F_HW_VLAN_CTAG_FILTER)
5002                 ixgbe_vlan_promisc_disable(adapter);
5003         else
5004                 ixgbe_vlan_promisc_enable(adapter);
5005 }
5006
5007 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
5008 {
5009         int q_idx;
5010
5011         for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++)
5012                 napi_enable(&adapter->q_vector[q_idx]->napi);
5013 }
5014
5015 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
5016 {
5017         int q_idx;
5018
5019         for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++)
5020                 napi_disable(&adapter->q_vector[q_idx]->napi);
5021 }
5022
5023 static int ixgbe_udp_tunnel_sync(struct net_device *dev, unsigned int table)
5024 {
5025         struct ixgbe_adapter *adapter = netdev_priv(dev);
5026         struct ixgbe_hw *hw = &adapter->hw;
5027         struct udp_tunnel_info ti;
5028
5029         udp_tunnel_nic_get_port(dev, table, 0, &ti);
5030         if (ti.type == UDP_TUNNEL_TYPE_VXLAN)
5031                 adapter->vxlan_port = ti.port;
5032         else
5033                 adapter->geneve_port = ti.port;
5034
5035         IXGBE_WRITE_REG(hw, IXGBE_VXLANCTRL,
5036                         ntohs(adapter->vxlan_port) |
5037                         ntohs(adapter->geneve_port) <<
5038                                 IXGBE_VXLANCTRL_GENEVE_UDPPORT_SHIFT);
5039         return 0;
5040 }
5041
5042 static const struct udp_tunnel_nic_info ixgbe_udp_tunnels_x550 = {
5043         .sync_table     = ixgbe_udp_tunnel_sync,
5044         .flags          = UDP_TUNNEL_NIC_INFO_IPV4_ONLY,
5045         .tables         = {
5046                 { .n_entries = 1, .tunnel_types = UDP_TUNNEL_TYPE_VXLAN,  },
5047         },
5048 };
5049
5050 static const struct udp_tunnel_nic_info ixgbe_udp_tunnels_x550em_a = {
5051         .sync_table     = ixgbe_udp_tunnel_sync,
5052         .flags          = UDP_TUNNEL_NIC_INFO_IPV4_ONLY,
5053         .tables         = {
5054                 { .n_entries = 1, .tunnel_types = UDP_TUNNEL_TYPE_VXLAN,  },
5055                 { .n_entries = 1, .tunnel_types = UDP_TUNNEL_TYPE_GENEVE, },
5056         },
5057 };
5058
5059 #ifdef CONFIG_IXGBE_DCB
5060 /**
5061  * ixgbe_configure_dcb - Configure DCB hardware
5062  * @adapter: ixgbe adapter struct
5063  *
5064  * This is called by the driver on open to configure the DCB hardware.
5065  * This is also called by the gennetlink interface when reconfiguring
5066  * the DCB state.
5067  */
5068 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
5069 {
5070         struct ixgbe_hw *hw = &adapter->hw;
5071         int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
5072
5073         if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
5074                 if (hw->mac.type == ixgbe_mac_82598EB)
5075                         netif_set_tso_max_size(adapter->netdev, 65536);
5076                 return;
5077         }
5078
5079         if (hw->mac.type == ixgbe_mac_82598EB)
5080                 netif_set_tso_max_size(adapter->netdev, 32768);
5081
5082 #ifdef IXGBE_FCOE
5083         if (adapter->netdev->fcoe_mtu)
5084                 max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
5085 #endif
5086
5087         /* reconfigure the hardware */
5088         if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE) {
5089                 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
5090                                                 DCB_TX_CONFIG);
5091                 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
5092                                                 DCB_RX_CONFIG);
5093                 ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
5094         } else if (adapter->ixgbe_ieee_ets && adapter->ixgbe_ieee_pfc) {
5095                 ixgbe_dcb_hw_ets(&adapter->hw,
5096                                  adapter->ixgbe_ieee_ets,
5097                                  max_frame);
5098                 ixgbe_dcb_hw_pfc_config(&adapter->hw,
5099                                         adapter->ixgbe_ieee_pfc->pfc_en,
5100                                         adapter->ixgbe_ieee_ets->prio_tc);
5101         }
5102
5103         /* Enable RSS Hash per TC */
5104         if (hw->mac.type != ixgbe_mac_82598EB) {
5105                 u32 msb = 0;
5106                 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices - 1;
5107
5108                 while (rss_i) {
5109                         msb++;
5110                         rss_i >>= 1;
5111                 }
5112
5113                 /* write msb to all 8 TCs in one write */
5114                 IXGBE_WRITE_REG(hw, IXGBE_RQTC, msb * 0x11111111);
5115         }
5116 }
5117 #endif
5118
5119 /* Additional bittime to account for IXGBE framing */
5120 #define IXGBE_ETH_FRAMING 20
5121
5122 /**
5123  * ixgbe_hpbthresh - calculate high water mark for flow control
5124  *
5125  * @adapter: board private structure to calculate for
5126  * @pb: packet buffer to calculate
5127  */
5128 static int ixgbe_hpbthresh(struct ixgbe_adapter *adapter, int pb)
5129 {
5130         struct ixgbe_hw *hw = &adapter->hw;
5131         struct net_device *dev = adapter->netdev;
5132         int link, tc, kb, marker;
5133         u32 dv_id, rx_pba;
5134
5135         /* Calculate max LAN frame size */
5136         tc = link = dev->mtu + ETH_HLEN + ETH_FCS_LEN + IXGBE_ETH_FRAMING;
5137
5138 #ifdef IXGBE_FCOE
5139         /* FCoE traffic class uses FCOE jumbo frames */
5140         if (dev->fcoe_mtu && tc < IXGBE_FCOE_JUMBO_FRAME_SIZE &&
5141             (pb == ixgbe_fcoe_get_tc(adapter)))
5142                 tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
5143 #endif
5144
5145         /* Calculate delay value for device */
5146         switch (hw->mac.type) {
5147         case ixgbe_mac_X540:
5148         case ixgbe_mac_X550:
5149         case ixgbe_mac_X550EM_x:
5150         case ixgbe_mac_x550em_a:
5151                 dv_id = IXGBE_DV_X540(link, tc);
5152                 break;
5153         default:
5154                 dv_id = IXGBE_DV(link, tc);
5155                 break;
5156         }
5157
5158         /* Loopback switch introduces additional latency */
5159         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
5160                 dv_id += IXGBE_B2BT(tc);
5161
5162         /* Delay value is calculated in bit times convert to KB */
5163         kb = IXGBE_BT2KB(dv_id);
5164         rx_pba = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(pb)) >> 10;
5165
5166         marker = rx_pba - kb;
5167
5168         /* It is possible that the packet buffer is not large enough
5169          * to provide required headroom. In this case throw an error
5170          * to user and a do the best we can.
5171          */
5172         if (marker < 0) {
5173                 e_warn(drv, "Packet Buffer(%i) can not provide enough"
5174                             "headroom to support flow control."
5175                             "Decrease MTU or number of traffic classes\n", pb);
5176                 marker = tc + 1;
5177         }
5178
5179         return marker;
5180 }
5181
5182 /**
5183  * ixgbe_lpbthresh - calculate low water mark for flow control
5184  *
5185  * @adapter: board private structure to calculate for
5186  * @pb: packet buffer to calculate
5187  */
5188 static int ixgbe_lpbthresh(struct ixgbe_adapter *adapter, int pb)
5189 {
5190         struct ixgbe_hw *hw = &adapter->hw;
5191         struct net_device *dev = adapter->netdev;
5192         int tc;
5193         u32 dv_id;
5194
5195         /* Calculate max LAN frame size */
5196         tc = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
5197
5198 #ifdef IXGBE_FCOE
5199         /* FCoE traffic class uses FCOE jumbo frames */
5200         if (dev->fcoe_mtu && tc < IXGBE_FCOE_JUMBO_FRAME_SIZE &&
5201             (pb == netdev_get_prio_tc_map(dev, adapter->fcoe.up)))
5202                 tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
5203 #endif
5204
5205         /* Calculate delay value for device */
5206         switch (hw->mac.type) {
5207         case ixgbe_mac_X540:
5208         case ixgbe_mac_X550:
5209         case ixgbe_mac_X550EM_x:
5210         case ixgbe_mac_x550em_a:
5211                 dv_id = IXGBE_LOW_DV_X540(tc);
5212                 break;
5213         default:
5214                 dv_id = IXGBE_LOW_DV(tc);
5215                 break;
5216         }
5217
5218         /* Delay value is calculated in bit times convert to KB */
5219         return IXGBE_BT2KB(dv_id);
5220 }
5221
5222 /*
5223  * ixgbe_pbthresh_setup - calculate and setup high low water marks
5224  */
5225 static void ixgbe_pbthresh_setup(struct ixgbe_adapter *adapter)
5226 {
5227         struct ixgbe_hw *hw = &adapter->hw;
5228         int num_tc = adapter->hw_tcs;
5229         int i;
5230
5231         if (!num_tc)
5232                 num_tc = 1;
5233
5234         for (i = 0; i < num_tc; i++) {
5235                 hw->fc.high_water[i] = ixgbe_hpbthresh(adapter, i);
5236                 hw->fc.low_water[i] = ixgbe_lpbthresh(adapter, i);
5237
5238                 /* Low water marks must not be larger than high water marks */
5239                 if (hw->fc.low_water[i] > hw->fc.high_water[i])
5240                         hw->fc.low_water[i] = 0;
5241         }
5242
5243         for (; i < MAX_TRAFFIC_CLASS; i++)
5244                 hw->fc.high_water[i] = 0;
5245 }
5246
5247 static void ixgbe_configure_pb(struct ixgbe_adapter *adapter)
5248 {
5249         struct ixgbe_hw *hw = &adapter->hw;
5250         int hdrm;
5251         u8 tc = adapter->hw_tcs;
5252
5253         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
5254             adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
5255                 hdrm = 32 << adapter->fdir_pballoc;
5256         else
5257                 hdrm = 0;
5258
5259         hw->mac.ops.set_rxpba(hw, tc, hdrm, PBA_STRATEGY_EQUAL);
5260         ixgbe_pbthresh_setup(adapter);
5261 }
5262
5263 static void ixgbe_fdir_filter_restore(struct ixgbe_adapter *adapter)
5264 {
5265         struct ixgbe_hw *hw = &adapter->hw;
5266         struct hlist_node *node2;
5267         struct ixgbe_fdir_filter *filter;
5268         u8 queue;
5269
5270         spin_lock(&adapter->fdir_perfect_lock);
5271
5272         if (!hlist_empty(&adapter->fdir_filter_list))
5273                 ixgbe_fdir_set_input_mask_82599(hw, &adapter->fdir_mask);
5274
5275         hlist_for_each_entry_safe(filter, node2,
5276                                   &adapter->fdir_filter_list, fdir_node) {
5277                 if (filter->action == IXGBE_FDIR_DROP_QUEUE) {
5278                         queue = IXGBE_FDIR_DROP_QUEUE;
5279                 } else {
5280                         u32 ring = ethtool_get_flow_spec_ring(filter->action);
5281                         u8 vf = ethtool_get_flow_spec_ring_vf(filter->action);
5282
5283                         if (!vf && (ring >= adapter->num_rx_queues)) {
5284                                 e_err(drv, "FDIR restore failed without VF, ring: %u\n",
5285                                       ring);
5286                                 continue;
5287                         } else if (vf &&
5288                                    ((vf > adapter->num_vfs) ||
5289                                      ring >= adapter->num_rx_queues_per_pool)) {
5290                                 e_err(drv, "FDIR restore failed with VF, vf: %hhu, ring: %u\n",
5291                                       vf, ring);
5292                                 continue;
5293                         }
5294
5295                         /* Map the ring onto the absolute queue index */
5296                         if (!vf)
5297                                 queue = adapter->rx_ring[ring]->reg_idx;
5298                         else
5299                                 queue = ((vf - 1) *
5300                                         adapter->num_rx_queues_per_pool) + ring;
5301                 }
5302
5303                 ixgbe_fdir_write_perfect_filter_82599(hw,
5304                                 &filter->filter, filter->sw_idx, queue);
5305         }
5306
5307         spin_unlock(&adapter->fdir_perfect_lock);
5308 }
5309
5310 /**
5311  * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
5312  * @rx_ring: ring to free buffers from
5313  **/
5314 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
5315 {
5316         u16 i = rx_ring->next_to_clean;
5317         struct ixgbe_rx_buffer *rx_buffer = &rx_ring->rx_buffer_info[i];
5318
5319         if (rx_ring->xsk_pool) {
5320                 ixgbe_xsk_clean_rx_ring(rx_ring);
5321                 goto skip_free;
5322         }
5323
5324         /* Free all the Rx ring sk_buffs */
5325         while (i != rx_ring->next_to_alloc) {
5326                 if (rx_buffer->skb) {
5327                         struct sk_buff *skb = rx_buffer->skb;
5328                         if (IXGBE_CB(skb)->page_released)
5329                                 dma_unmap_page_attrs(rx_ring->dev,
5330                                                      IXGBE_CB(skb)->dma,
5331                                                      ixgbe_rx_pg_size(rx_ring),
5332                                                      DMA_FROM_DEVICE,
5333                                                      IXGBE_RX_DMA_ATTR);
5334                         dev_kfree_skb(skb);
5335                 }
5336
5337                 /* Invalidate cache lines that may have been written to by
5338                  * device so that we avoid corrupting memory.
5339                  */
5340                 dma_sync_single_range_for_cpu(rx_ring->dev,
5341                                               rx_buffer->dma,
5342                                               rx_buffer->page_offset,
5343                                               ixgbe_rx_bufsz(rx_ring),
5344                                               DMA_FROM_DEVICE);
5345
5346                 /* free resources associated with mapping */
5347                 dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
5348                                      ixgbe_rx_pg_size(rx_ring),
5349                                      DMA_FROM_DEVICE,
5350                                      IXGBE_RX_DMA_ATTR);
5351                 __page_frag_cache_drain(rx_buffer->page,
5352                                         rx_buffer->pagecnt_bias);
5353
5354                 i++;
5355                 rx_buffer++;
5356                 if (i == rx_ring->count) {
5357                         i = 0;
5358                         rx_buffer = rx_ring->rx_buffer_info;
5359                 }
5360         }
5361
5362 skip_free:
5363         rx_ring->next_to_alloc = 0;
5364         rx_ring->next_to_clean = 0;
5365         rx_ring->next_to_use = 0;
5366 }
5367
5368 static int ixgbe_fwd_ring_up(struct ixgbe_adapter *adapter,
5369                              struct ixgbe_fwd_adapter *accel)
5370 {
5371         u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
5372         int num_tc = netdev_get_num_tc(adapter->netdev);
5373         struct net_device *vdev = accel->netdev;
5374         int i, baseq, err;
5375
5376         baseq = accel->pool * adapter->num_rx_queues_per_pool;
5377         netdev_dbg(vdev, "pool %i:%i queues %i:%i\n",
5378                    accel->pool, adapter->num_rx_pools,
5379                    baseq, baseq + adapter->num_rx_queues_per_pool);
5380
5381         accel->rx_base_queue = baseq;
5382         accel->tx_base_queue = baseq;
5383
5384         /* record configuration for macvlan interface in vdev */
5385         for (i = 0; i < num_tc; i++)
5386                 netdev_bind_sb_channel_queue(adapter->netdev, vdev,
5387                                              i, rss_i, baseq + (rss_i * i));
5388
5389         for (i = 0; i < adapter->num_rx_queues_per_pool; i++)
5390                 adapter->rx_ring[baseq + i]->netdev = vdev;
5391
5392         /* Guarantee all rings are updated before we update the
5393          * MAC address filter.
5394          */
5395         wmb();
5396
5397         /* ixgbe_add_mac_filter will return an index if it succeeds, so we
5398          * need to only treat it as an error value if it is negative.
5399          */
5400         err = ixgbe_add_mac_filter(adapter, vdev->dev_addr,
5401                                    VMDQ_P(accel->pool));
5402         if (err >= 0)
5403                 return 0;
5404
5405         /* if we cannot add the MAC rule then disable the offload */
5406         macvlan_release_l2fw_offload(vdev);
5407
5408         for (i = 0; i < adapter->num_rx_queues_per_pool; i++)
5409                 adapter->rx_ring[baseq + i]->netdev = NULL;
5410
5411         netdev_err(vdev, "L2FW offload disabled due to L2 filter error\n");
5412
5413         /* unbind the queues and drop the subordinate channel config */
5414         netdev_unbind_sb_channel(adapter->netdev, vdev);
5415         netdev_set_sb_channel(vdev, 0);
5416
5417         clear_bit(accel->pool, adapter->fwd_bitmask);
5418         kfree(accel);
5419
5420         return err;
5421 }
5422
5423 static int ixgbe_macvlan_up(struct net_device *vdev,
5424                             struct netdev_nested_priv *priv)
5425 {
5426         struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)priv->data;
5427         struct ixgbe_fwd_adapter *accel;
5428
5429         if (!netif_is_macvlan(vdev))
5430                 return 0;
5431
5432         accel = macvlan_accel_priv(vdev);
5433         if (!accel)
5434                 return 0;
5435
5436         ixgbe_fwd_ring_up(adapter, accel);
5437
5438         return 0;
5439 }
5440
5441 static void ixgbe_configure_dfwd(struct ixgbe_adapter *adapter)
5442 {
5443         struct netdev_nested_priv priv = {
5444                 .data = (void *)adapter,
5445         };
5446
5447         netdev_walk_all_upper_dev_rcu(adapter->netdev,
5448                                       ixgbe_macvlan_up, &priv);
5449 }
5450
5451 static void ixgbe_configure(struct ixgbe_adapter *adapter)
5452 {
5453         struct ixgbe_hw *hw = &adapter->hw;
5454
5455         ixgbe_configure_pb(adapter);
5456 #ifdef CONFIG_IXGBE_DCB
5457         ixgbe_configure_dcb(adapter);
5458 #endif
5459         /*
5460          * We must restore virtualization before VLANs or else
5461          * the VLVF registers will not be populated
5462          */
5463         ixgbe_configure_virtualization(adapter);
5464
5465         ixgbe_set_rx_mode(adapter->netdev);
5466         ixgbe_restore_vlan(adapter);
5467         ixgbe_ipsec_restore(adapter);
5468
5469         switch (hw->mac.type) {
5470         case ixgbe_mac_82599EB:
5471         case ixgbe_mac_X540:
5472                 hw->mac.ops.disable_rx_buff(hw);
5473                 break;
5474         default:
5475                 break;
5476         }
5477
5478         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
5479                 ixgbe_init_fdir_signature_82599(&adapter->hw,
5480                                                 adapter->fdir_pballoc);
5481         } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
5482                 ixgbe_init_fdir_perfect_82599(&adapter->hw,
5483                                               adapter->fdir_pballoc);
5484                 ixgbe_fdir_filter_restore(adapter);
5485         }
5486
5487         switch (hw->mac.type) {
5488         case ixgbe_mac_82599EB:
5489         case ixgbe_mac_X540:
5490                 hw->mac.ops.enable_rx_buff(hw);
5491                 break;
5492         default:
5493                 break;
5494         }
5495
5496 #ifdef CONFIG_IXGBE_DCA
5497         /* configure DCA */
5498         if (adapter->flags & IXGBE_FLAG_DCA_CAPABLE)
5499                 ixgbe_setup_dca(adapter);
5500 #endif /* CONFIG_IXGBE_DCA */
5501
5502 #ifdef IXGBE_FCOE
5503         /* configure FCoE L2 filters, redirection table, and Rx control */
5504         ixgbe_configure_fcoe(adapter);
5505
5506 #endif /* IXGBE_FCOE */
5507         ixgbe_configure_tx(adapter);
5508         ixgbe_configure_rx(adapter);
5509         ixgbe_configure_dfwd(adapter);
5510 }
5511
5512 /**
5513  * ixgbe_sfp_link_config - set up SFP+ link
5514  * @adapter: pointer to private adapter struct
5515  **/
5516 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
5517 {
5518         /*
5519          * We are assuming the worst case scenario here, and that
5520          * is that an SFP was inserted/removed after the reset
5521          * but before SFP detection was enabled.  As such the best
5522          * solution is to just start searching as soon as we start
5523          */
5524         if (adapter->hw.mac.type == ixgbe_mac_82598EB)
5525                 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
5526
5527         adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
5528         adapter->sfp_poll_time = 0;
5529 }
5530
5531 /**
5532  * ixgbe_non_sfp_link_config - set up non-SFP+ link
5533  * @hw: pointer to private hardware struct
5534  *
5535  * Returns 0 on success, negative on failure
5536  **/
5537 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
5538 {
5539         u32 speed;
5540         bool autoneg, link_up = false;
5541         int ret = -EIO;
5542
5543         if (hw->mac.ops.check_link)
5544                 ret = hw->mac.ops.check_link(hw, &speed, &link_up, false);
5545
5546         if (ret)
5547                 return ret;
5548
5549         speed = hw->phy.autoneg_advertised;
5550         if (!speed && hw->mac.ops.get_link_capabilities) {
5551                 ret = hw->mac.ops.get_link_capabilities(hw, &speed,
5552                                                         &autoneg);
5553                 /* remove NBASE-T speeds from default autonegotiation
5554                  * to accommodate broken network switches in the field
5555                  * which cannot cope with advertised NBASE-T speeds
5556                  */
5557                 speed &= ~(IXGBE_LINK_SPEED_5GB_FULL |
5558                            IXGBE_LINK_SPEED_2_5GB_FULL);
5559         }
5560
5561         if (ret)
5562                 return ret;
5563
5564         if (hw->mac.ops.setup_link)
5565                 ret = hw->mac.ops.setup_link(hw, speed, link_up);
5566
5567         return ret;
5568 }
5569
5570 /**
5571  * ixgbe_clear_vf_stats_counters - Clear out VF stats after reset
5572  * @adapter: board private structure
5573  *
5574  * On a reset we need to clear out the VF stats or accounting gets
5575  * messed up because they're not clear on read.
5576  **/
5577 static void ixgbe_clear_vf_stats_counters(struct ixgbe_adapter *adapter)
5578 {
5579         struct ixgbe_hw *hw = &adapter->hw;
5580         int i;
5581
5582         for (i = 0; i < adapter->num_vfs; i++) {
5583                 adapter->vfinfo[i].last_vfstats.gprc =
5584                         IXGBE_READ_REG(hw, IXGBE_PVFGPRC(i));
5585                 adapter->vfinfo[i].saved_rst_vfstats.gprc +=
5586                         adapter->vfinfo[i].vfstats.gprc;
5587                 adapter->vfinfo[i].vfstats.gprc = 0;
5588                 adapter->vfinfo[i].last_vfstats.gptc =
5589                         IXGBE_READ_REG(hw, IXGBE_PVFGPTC(i));
5590                 adapter->vfinfo[i].saved_rst_vfstats.gptc +=
5591                         adapter->vfinfo[i].vfstats.gptc;
5592                 adapter->vfinfo[i].vfstats.gptc = 0;
5593                 adapter->vfinfo[i].last_vfstats.gorc =
5594                         IXGBE_READ_REG(hw, IXGBE_PVFGORC_LSB(i));
5595                 adapter->vfinfo[i].saved_rst_vfstats.gorc +=
5596                         adapter->vfinfo[i].vfstats.gorc;
5597                 adapter->vfinfo[i].vfstats.gorc = 0;
5598                 adapter->vfinfo[i].last_vfstats.gotc =
5599                         IXGBE_READ_REG(hw, IXGBE_PVFGOTC_LSB(i));
5600                 adapter->vfinfo[i].saved_rst_vfstats.gotc +=
5601                         adapter->vfinfo[i].vfstats.gotc;
5602                 adapter->vfinfo[i].vfstats.gotc = 0;
5603                 adapter->vfinfo[i].last_vfstats.mprc =
5604                         IXGBE_READ_REG(hw, IXGBE_PVFMPRC(i));
5605                 adapter->vfinfo[i].saved_rst_vfstats.mprc +=
5606                         adapter->vfinfo[i].vfstats.mprc;
5607                 adapter->vfinfo[i].vfstats.mprc = 0;
5608         }
5609 }
5610
5611 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
5612 {
5613         struct ixgbe_hw *hw = &adapter->hw;
5614         u32 gpie = 0;
5615
5616         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
5617                 gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
5618                        IXGBE_GPIE_OCD;
5619                 gpie |= IXGBE_GPIE_EIAME;
5620                 /*
5621                  * use EIAM to auto-mask when MSI-X interrupt is asserted
5622                  * this saves a register write for every interrupt
5623                  */
5624                 switch (hw->mac.type) {
5625                 case ixgbe_mac_82598EB:
5626                         IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
5627                         break;
5628                 case ixgbe_mac_82599EB:
5629                 case ixgbe_mac_X540:
5630                 case ixgbe_mac_X550:
5631                 case ixgbe_mac_X550EM_x:
5632                 case ixgbe_mac_x550em_a:
5633                 default:
5634                         IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
5635                         IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
5636                         break;
5637                 }
5638         } else {
5639                 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
5640                  * specifically only auto mask tx and rx interrupts */
5641                 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
5642         }
5643
5644         /* XXX: to interrupt immediately for EICS writes, enable this */
5645         /* gpie |= IXGBE_GPIE_EIMEN; */
5646
5647         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
5648                 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
5649
5650                 switch (adapter->ring_feature[RING_F_VMDQ].mask) {
5651                 case IXGBE_82599_VMDQ_8Q_MASK:
5652                         gpie |= IXGBE_GPIE_VTMODE_16;
5653                         break;
5654                 case IXGBE_82599_VMDQ_4Q_MASK:
5655                         gpie |= IXGBE_GPIE_VTMODE_32;
5656                         break;
5657                 default:
5658                         gpie |= IXGBE_GPIE_VTMODE_64;
5659                         break;
5660                 }
5661         }
5662
5663         /* Enable Thermal over heat sensor interrupt */
5664         if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
5665                 switch (adapter->hw.mac.type) {
5666                 case ixgbe_mac_82599EB:
5667                         gpie |= IXGBE_SDP0_GPIEN_8259X;
5668                         break;
5669                 default:
5670                         break;
5671                 }
5672         }
5673
5674         /* Enable fan failure interrupt */
5675         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
5676                 gpie |= IXGBE_SDP1_GPIEN(hw);
5677
5678         switch (hw->mac.type) {
5679         case ixgbe_mac_82599EB:
5680                 gpie |= IXGBE_SDP1_GPIEN_8259X | IXGBE_SDP2_GPIEN_8259X;
5681                 break;
5682         case ixgbe_mac_X550EM_x:
5683         case ixgbe_mac_x550em_a:
5684                 gpie |= IXGBE_SDP0_GPIEN_X540;
5685                 break;
5686         default:
5687                 break;
5688         }
5689
5690         IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
5691 }
5692
5693 static void ixgbe_up_complete(struct ixgbe_adapter *adapter)
5694 {
5695         struct ixgbe_hw *hw = &adapter->hw;
5696         int err;
5697         u32 ctrl_ext;
5698
5699         ixgbe_get_hw_control(adapter);
5700         ixgbe_setup_gpie(adapter);
5701
5702         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
5703                 ixgbe_configure_msix(adapter);
5704         else
5705                 ixgbe_configure_msi_and_legacy(adapter);
5706
5707         /* enable the optics for 82599 SFP+ fiber */
5708         if (hw->mac.ops.enable_tx_laser)
5709                 hw->mac.ops.enable_tx_laser(hw);
5710
5711         if (hw->phy.ops.set_phy_power)
5712                 hw->phy.ops.set_phy_power(hw, true);
5713
5714         smp_mb__before_atomic();
5715         clear_bit(__IXGBE_DOWN, &adapter->state);
5716         ixgbe_napi_enable_all(adapter);
5717
5718         if (ixgbe_is_sfp(hw)) {
5719                 ixgbe_sfp_link_config(adapter);
5720         } else {
5721                 err = ixgbe_non_sfp_link_config(hw);
5722                 if (err)
5723                         e_err(probe, "link_config FAILED %d\n", err);
5724         }
5725
5726         /* clear any pending interrupts, may auto mask */
5727         IXGBE_READ_REG(hw, IXGBE_EICR);
5728         ixgbe_irq_enable(adapter, true, true);
5729
5730         /*
5731          * If this adapter has a fan, check to see if we had a failure
5732          * before we enabled the interrupt.
5733          */
5734         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
5735                 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
5736                 if (esdp & IXGBE_ESDP_SDP1)
5737                         e_crit(drv, "Fan has stopped, replace the adapter\n");
5738         }
5739
5740         /* bring the link up in the watchdog, this could race with our first
5741          * link up interrupt but shouldn't be a problem */
5742         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5743         adapter->link_check_timeout = jiffies;
5744         mod_timer(&adapter->service_timer, jiffies);
5745
5746         ixgbe_clear_vf_stats_counters(adapter);
5747         /* Set PF Reset Done bit so PF/VF Mail Ops can work */
5748         ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
5749         ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
5750         IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
5751
5752         /* update setting rx tx for all active vfs */
5753         ixgbe_set_all_vfs(adapter);
5754 }
5755
5756 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
5757 {
5758         /* put off any impending NetWatchDogTimeout */
5759         netif_trans_update(adapter->netdev);
5760
5761         while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
5762                 usleep_range(1000, 2000);
5763         if (adapter->hw.phy.type == ixgbe_phy_fw)
5764                 ixgbe_watchdog_link_is_down(adapter);
5765         ixgbe_down(adapter);
5766         /*
5767          * If SR-IOV enabled then wait a bit before bringing the adapter
5768          * back up to give the VFs time to respond to the reset.  The
5769          * two second wait is based upon the watchdog timer cycle in
5770          * the VF driver.
5771          */
5772         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
5773                 msleep(2000);
5774         ixgbe_up(adapter);
5775         clear_bit(__IXGBE_RESETTING, &adapter->state);
5776 }
5777
5778 void ixgbe_up(struct ixgbe_adapter *adapter)
5779 {
5780         /* hardware has been reset, we need to reload some things */
5781         ixgbe_configure(adapter);
5782
5783         ixgbe_up_complete(adapter);
5784 }
5785
5786 static unsigned long ixgbe_get_completion_timeout(struct ixgbe_adapter *adapter)
5787 {
5788         u16 devctl2;
5789
5790         pcie_capability_read_word(adapter->pdev, PCI_EXP_DEVCTL2, &devctl2);
5791
5792         switch (devctl2 & IXGBE_PCIDEVCTRL2_TIMEO_MASK) {
5793         case IXGBE_PCIDEVCTRL2_17_34s:
5794         case IXGBE_PCIDEVCTRL2_4_8s:
5795                 /* For now we cap the upper limit on delay to 2 seconds
5796                  * as we end up going up to 34 seconds of delay in worst
5797                  * case timeout value.
5798                  */
5799         case IXGBE_PCIDEVCTRL2_1_2s:
5800                 return 2000000ul;       /* 2.0 s */
5801         case IXGBE_PCIDEVCTRL2_260_520ms:
5802                 return 520000ul;        /* 520 ms */
5803         case IXGBE_PCIDEVCTRL2_65_130ms:
5804                 return 130000ul;        /* 130 ms */
5805         case IXGBE_PCIDEVCTRL2_16_32ms:
5806                 return 32000ul;         /* 32 ms */
5807         case IXGBE_PCIDEVCTRL2_1_2ms:
5808                 return 2000ul;          /* 2 ms */
5809         case IXGBE_PCIDEVCTRL2_50_100us:
5810                 return 100ul;           /* 100 us */
5811         case IXGBE_PCIDEVCTRL2_16_32ms_def:
5812                 return 32000ul;         /* 32 ms */
5813         default:
5814                 break;
5815         }
5816
5817         /* We shouldn't need to hit this path, but just in case default as
5818          * though completion timeout is not supported and support 32ms.
5819          */
5820         return 32000ul;
5821 }
5822
5823 void ixgbe_disable_rx(struct ixgbe_adapter *adapter)
5824 {
5825         unsigned long wait_delay, delay_interval;
5826         struct ixgbe_hw *hw = &adapter->hw;
5827         int i, wait_loop;
5828         u32 rxdctl;
5829
5830         /* disable receives */
5831         hw->mac.ops.disable_rx(hw);
5832
5833         if (ixgbe_removed(hw->hw_addr))
5834                 return;
5835
5836         /* disable all enabled Rx queues */
5837         for (i = 0; i < adapter->num_rx_queues; i++) {
5838                 struct ixgbe_ring *ring = adapter->rx_ring[i];
5839                 u8 reg_idx = ring->reg_idx;
5840
5841                 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
5842                 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
5843                 rxdctl |= IXGBE_RXDCTL_SWFLSH;
5844
5845                 /* write value back with RXDCTL.ENABLE bit cleared */
5846                 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
5847         }
5848
5849         /* RXDCTL.EN may not change on 82598 if link is down, so skip it */
5850         if (hw->mac.type == ixgbe_mac_82598EB &&
5851             !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
5852                 return;
5853
5854         /* Determine our minimum delay interval. We will increase this value
5855          * with each subsequent test. This way if the device returns quickly
5856          * we should spend as little time as possible waiting, however as
5857          * the time increases we will wait for larger periods of time.
5858          *
5859          * The trick here is that we increase the interval using the
5860          * following pattern: 1x 3x 5x 7x 9x 11x 13x 15x 17x 19x. The result
5861          * of that wait is that it totals up to 100x whatever interval we
5862          * choose. Since our minimum wait is 100us we can just divide the
5863          * total timeout by 100 to get our minimum delay interval.
5864          */
5865         delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
5866
5867         wait_loop = IXGBE_MAX_RX_DESC_POLL;
5868         wait_delay = delay_interval;
5869
5870         while (wait_loop--) {
5871                 usleep_range(wait_delay, wait_delay + 10);
5872                 wait_delay += delay_interval * 2;
5873                 rxdctl = 0;
5874
5875                 /* OR together the reading of all the active RXDCTL registers,
5876                  * and then test the result. We need the disable to complete
5877                  * before we start freeing the memory and invalidating the
5878                  * DMA mappings.
5879                  */
5880                 for (i = 0; i < adapter->num_rx_queues; i++) {
5881                         struct ixgbe_ring *ring = adapter->rx_ring[i];
5882                         u8 reg_idx = ring->reg_idx;
5883
5884                         rxdctl |= IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
5885                 }
5886
5887                 if (!(rxdctl & IXGBE_RXDCTL_ENABLE))
5888                         return;
5889         }
5890
5891         e_err(drv,
5892               "RXDCTL.ENABLE for one or more queues not cleared within the polling period\n");
5893 }
5894
5895 void ixgbe_disable_tx(struct ixgbe_adapter *adapter)
5896 {
5897         unsigned long wait_delay, delay_interval;
5898         struct ixgbe_hw *hw = &adapter->hw;
5899         int i, wait_loop;
5900         u32 txdctl;
5901
5902         if (ixgbe_removed(hw->hw_addr))
5903                 return;
5904
5905         /* disable all enabled Tx queues */
5906         for (i = 0; i < adapter->num_tx_queues; i++) {
5907                 struct ixgbe_ring *ring = adapter->tx_ring[i];
5908                 u8 reg_idx = ring->reg_idx;
5909
5910                 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
5911         }
5912
5913         /* disable all enabled XDP Tx queues */
5914         for (i = 0; i < adapter->num_xdp_queues; i++) {
5915                 struct ixgbe_ring *ring = adapter->xdp_ring[i];
5916                 u8 reg_idx = ring->reg_idx;
5917
5918                 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
5919         }
5920
5921         /* If the link is not up there shouldn't be much in the way of
5922          * pending transactions. Those that are left will be flushed out
5923          * when the reset logic goes through the flush sequence to clean out
5924          * the pending Tx transactions.
5925          */
5926         if (!(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
5927                 goto dma_engine_disable;
5928
5929         /* Determine our minimum delay interval. We will increase this value
5930          * with each subsequent test. This way if the device returns quickly
5931          * we should spend as little time as possible waiting, however as
5932          * the time increases we will wait for larger periods of time.
5933          *
5934          * The trick here is that we increase the interval using the
5935          * following pattern: 1x 3x 5x 7x 9x 11x 13x 15x 17x 19x. The result
5936          * of that wait is that it totals up to 100x whatever interval we
5937          * choose. Since our minimum wait is 100us we can just divide the
5938          * total timeout by 100 to get our minimum delay interval.
5939          */
5940         delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
5941
5942         wait_loop = IXGBE_MAX_RX_DESC_POLL;
5943         wait_delay = delay_interval;
5944
5945         while (wait_loop--) {
5946                 usleep_range(wait_delay, wait_delay + 10);
5947                 wait_delay += delay_interval * 2;
5948                 txdctl = 0;
5949
5950                 /* OR together the reading of all the active TXDCTL registers,
5951                  * and then test the result. We need the disable to complete
5952                  * before we start freeing the memory and invalidating the
5953                  * DMA mappings.
5954                  */
5955                 for (i = 0; i < adapter->num_tx_queues; i++) {
5956                         struct ixgbe_ring *ring = adapter->tx_ring[i];
5957                         u8 reg_idx = ring->reg_idx;
5958
5959                         txdctl |= IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
5960                 }
5961                 for (i = 0; i < adapter->num_xdp_queues; i++) {
5962                         struct ixgbe_ring *ring = adapter->xdp_ring[i];
5963                         u8 reg_idx = ring->reg_idx;
5964
5965                         txdctl |= IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
5966                 }
5967
5968                 if (!(txdctl & IXGBE_TXDCTL_ENABLE))
5969                         goto dma_engine_disable;
5970         }
5971
5972         e_err(drv,
5973               "TXDCTL.ENABLE for one or more queues not cleared within the polling period\n");
5974
5975 dma_engine_disable:
5976         /* Disable the Tx DMA engine on 82599 and later MAC */
5977         switch (hw->mac.type) {
5978         case ixgbe_mac_82599EB:
5979         case ixgbe_mac_X540:
5980         case ixgbe_mac_X550:
5981         case ixgbe_mac_X550EM_x:
5982         case ixgbe_mac_x550em_a:
5983                 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
5984                                 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
5985                                  ~IXGBE_DMATXCTL_TE));
5986                 fallthrough;
5987         default:
5988                 break;
5989         }
5990 }
5991
5992 void ixgbe_reset(struct ixgbe_adapter *adapter)
5993 {
5994         struct ixgbe_hw *hw = &adapter->hw;
5995         struct net_device *netdev = adapter->netdev;
5996         int err;
5997
5998         if (ixgbe_removed(hw->hw_addr))
5999                 return;
6000         /* lock SFP init bit to prevent race conditions with the watchdog */
6001         while (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
6002                 usleep_range(1000, 2000);
6003
6004         /* clear all SFP and link config related flags while holding SFP_INIT */
6005         adapter->flags2 &= ~(IXGBE_FLAG2_SEARCH_FOR_SFP |
6006                              IXGBE_FLAG2_SFP_NEEDS_RESET);
6007         adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
6008
6009         err = hw->mac.ops.init_hw(hw);
6010         switch (err) {
6011         case 0:
6012         case -ENOENT:
6013         case -EOPNOTSUPP:
6014                 break;
6015         case -EALREADY:
6016                 e_dev_err("primary disable timed out\n");
6017                 break;
6018         case -EACCES:
6019                 /* We are running on a pre-production device, log a warning */
6020                 e_dev_warn("This device is a pre-production adapter/LOM. "
6021                            "Please be aware there may be issues associated with "
6022                            "your hardware.  If you are experiencing problems "
6023                            "please contact your Intel or hardware "
6024                            "representative who provided you with this "
6025                            "hardware.\n");
6026                 break;
6027         default:
6028                 e_dev_err("Hardware Error: %d\n", err);
6029         }
6030
6031         clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
6032
6033         /* flush entries out of MAC table */
6034         ixgbe_flush_sw_mac_table(adapter);
6035         __dev_uc_unsync(netdev, NULL);
6036
6037         /* do not flush user set addresses */
6038         ixgbe_mac_set_default_filter(adapter);
6039
6040         /* update SAN MAC vmdq pool selection */
6041         if (hw->mac.san_mac_rar_index)
6042                 hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
6043
6044         if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
6045                 ixgbe_ptp_reset(adapter);
6046
6047         if (hw->phy.ops.set_phy_power) {
6048                 if (!netif_running(adapter->netdev) && !adapter->wol)
6049                         hw->phy.ops.set_phy_power(hw, false);
6050                 else
6051                         hw->phy.ops.set_phy_power(hw, true);
6052         }
6053 }
6054
6055 /**
6056  * ixgbe_clean_tx_ring - Free Tx Buffers
6057  * @tx_ring: ring to be cleaned
6058  **/
6059 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
6060 {
6061         u16 i = tx_ring->next_to_clean;
6062         struct ixgbe_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i];
6063
6064         if (tx_ring->xsk_pool) {
6065                 ixgbe_xsk_clean_tx_ring(tx_ring);
6066                 goto out;
6067         }
6068
6069         while (i != tx_ring->next_to_use) {
6070                 union ixgbe_adv_tx_desc *eop_desc, *tx_desc;
6071
6072                 /* Free all the Tx ring sk_buffs */
6073                 if (ring_is_xdp(tx_ring))
6074                         xdp_return_frame(tx_buffer->xdpf);
6075                 else
6076                         dev_kfree_skb_any(tx_buffer->skb);
6077
6078                 /* unmap skb header data */
6079                 dma_unmap_single(tx_ring->dev,
6080                                  dma_unmap_addr(tx_buffer, dma),
6081                                  dma_unmap_len(tx_buffer, len),
6082                                  DMA_TO_DEVICE);
6083
6084                 /* check for eop_desc to determine the end of the packet */
6085                 eop_desc = tx_buffer->next_to_watch;
6086                 tx_desc = IXGBE_TX_DESC(tx_ring, i);
6087
6088                 /* unmap remaining buffers */
6089                 while (tx_desc != eop_desc) {
6090                         tx_buffer++;
6091                         tx_desc++;
6092                         i++;
6093                         if (unlikely(i == tx_ring->count)) {
6094                                 i = 0;
6095                                 tx_buffer = tx_ring->tx_buffer_info;
6096                                 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
6097                         }
6098
6099                         /* unmap any remaining paged data */
6100                         if (dma_unmap_len(tx_buffer, len))
6101                                 dma_unmap_page(tx_ring->dev,
6102                                                dma_unmap_addr(tx_buffer, dma),
6103                                                dma_unmap_len(tx_buffer, len),
6104                                                DMA_TO_DEVICE);
6105                 }
6106
6107                 /* move us one more past the eop_desc for start of next pkt */
6108                 tx_buffer++;
6109                 i++;
6110                 if (unlikely(i == tx_ring->count)) {
6111                         i = 0;
6112                         tx_buffer = tx_ring->tx_buffer_info;
6113                 }
6114         }
6115
6116         /* reset BQL for queue */
6117         if (!ring_is_xdp(tx_ring))
6118                 netdev_tx_reset_queue(txring_txq(tx_ring));
6119
6120 out:
6121         /* reset next_to_use and next_to_clean */
6122         tx_ring->next_to_use = 0;
6123         tx_ring->next_to_clean = 0;
6124 }
6125
6126 /**
6127  * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
6128  * @adapter: board private structure
6129  **/
6130 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
6131 {
6132         int i;
6133
6134         for (i = 0; i < adapter->num_rx_queues; i++)
6135                 ixgbe_clean_rx_ring(adapter->rx_ring[i]);
6136 }
6137
6138 /**
6139  * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
6140  * @adapter: board private structure
6141  **/
6142 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
6143 {
6144         int i;
6145
6146         for (i = 0; i < adapter->num_tx_queues; i++)
6147                 ixgbe_clean_tx_ring(adapter->tx_ring[i]);
6148         for (i = 0; i < adapter->num_xdp_queues; i++)
6149                 ixgbe_clean_tx_ring(adapter->xdp_ring[i]);
6150 }
6151
6152 static void ixgbe_fdir_filter_exit(struct ixgbe_adapter *adapter)
6153 {
6154         struct hlist_node *node2;
6155         struct ixgbe_fdir_filter *filter;
6156
6157         spin_lock(&adapter->fdir_perfect_lock);
6158
6159         hlist_for_each_entry_safe(filter, node2,
6160                                   &adapter->fdir_filter_list, fdir_node) {
6161                 hlist_del(&filter->fdir_node);
6162                 kfree(filter);
6163         }
6164         adapter->fdir_filter_count = 0;
6165
6166         spin_unlock(&adapter->fdir_perfect_lock);
6167 }
6168
6169 void ixgbe_down(struct ixgbe_adapter *adapter)
6170 {
6171         struct net_device *netdev = adapter->netdev;
6172         struct ixgbe_hw *hw = &adapter->hw;
6173         int i;
6174
6175         /* signal that we are down to the interrupt handler */
6176         if (test_and_set_bit(__IXGBE_DOWN, &adapter->state))
6177                 return; /* do nothing if already down */
6178
6179         /* Shut off incoming Tx traffic */
6180         netif_tx_stop_all_queues(netdev);
6181
6182         /* call carrier off first to avoid false dev_watchdog timeouts */
6183         netif_carrier_off(netdev);
6184         netif_tx_disable(netdev);
6185
6186         /* Disable Rx */
6187         ixgbe_disable_rx(adapter);
6188
6189         /* synchronize_rcu() needed for pending XDP buffers to drain */
6190         if (adapter->xdp_ring[0])
6191                 synchronize_rcu();
6192
6193         ixgbe_irq_disable(adapter);
6194
6195         ixgbe_napi_disable_all(adapter);
6196
6197         clear_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
6198         adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
6199         adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
6200
6201         del_timer_sync(&adapter->service_timer);
6202
6203         if (adapter->num_vfs) {
6204                 /* Clear EITR Select mapping */
6205                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
6206
6207                 /* Mark all the VFs as inactive */
6208                 for (i = 0 ; i < adapter->num_vfs; i++)
6209                         adapter->vfinfo[i].clear_to_send = false;
6210
6211                 /* update setting rx tx for all active vfs */
6212                 ixgbe_set_all_vfs(adapter);
6213         }
6214
6215         /* disable transmits in the hardware now that interrupts are off */
6216         ixgbe_disable_tx(adapter);
6217
6218         if (!pci_channel_offline(adapter->pdev))
6219                 ixgbe_reset(adapter);
6220
6221         /* power down the optics for 82599 SFP+ fiber */
6222         if (hw->mac.ops.disable_tx_laser)
6223                 hw->mac.ops.disable_tx_laser(hw);
6224
6225         ixgbe_clean_all_tx_rings(adapter);
6226         ixgbe_clean_all_rx_rings(adapter);
6227 }
6228
6229 /**
6230  * ixgbe_set_eee_capable - helper function to determine EEE support on X550
6231  * @adapter: board private structure
6232  */
6233 static void ixgbe_set_eee_capable(struct ixgbe_adapter *adapter)
6234 {
6235         struct ixgbe_hw *hw = &adapter->hw;
6236
6237         switch (hw->device_id) {
6238         case IXGBE_DEV_ID_X550EM_A_1G_T:
6239         case IXGBE_DEV_ID_X550EM_A_1G_T_L:
6240                 if (!hw->phy.eee_speeds_supported)
6241                         break;
6242                 adapter->flags2 |= IXGBE_FLAG2_EEE_CAPABLE;
6243                 if (!hw->phy.eee_speeds_advertised)
6244                         break;
6245                 adapter->flags2 |= IXGBE_FLAG2_EEE_ENABLED;
6246                 break;
6247         default:
6248                 adapter->flags2 &= ~IXGBE_FLAG2_EEE_CAPABLE;
6249                 adapter->flags2 &= ~IXGBE_FLAG2_EEE_ENABLED;
6250                 break;
6251         }
6252 }
6253
6254 /**
6255  * ixgbe_tx_timeout - Respond to a Tx Hang
6256  * @netdev: network interface device structure
6257  * @txqueue: queue number that timed out
6258  **/
6259 static void ixgbe_tx_timeout(struct net_device *netdev, unsigned int __always_unused txqueue)
6260 {
6261         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6262
6263         /* Do the reset outside of interrupt context */
6264         ixgbe_tx_timeout_reset(adapter);
6265 }
6266
6267 #ifdef CONFIG_IXGBE_DCB
6268 static void ixgbe_init_dcb(struct ixgbe_adapter *adapter)
6269 {
6270         struct ixgbe_hw *hw = &adapter->hw;
6271         struct tc_configuration *tc;
6272         int j;
6273
6274         switch (hw->mac.type) {
6275         case ixgbe_mac_82598EB:
6276         case ixgbe_mac_82599EB:
6277                 adapter->dcb_cfg.num_tcs.pg_tcs = MAX_TRAFFIC_CLASS;
6278                 adapter->dcb_cfg.num_tcs.pfc_tcs = MAX_TRAFFIC_CLASS;
6279                 break;
6280         case ixgbe_mac_X540:
6281         case ixgbe_mac_X550:
6282                 adapter->dcb_cfg.num_tcs.pg_tcs = X540_TRAFFIC_CLASS;
6283                 adapter->dcb_cfg.num_tcs.pfc_tcs = X540_TRAFFIC_CLASS;
6284                 break;
6285         case ixgbe_mac_X550EM_x:
6286         case ixgbe_mac_x550em_a:
6287         default:
6288                 adapter->dcb_cfg.num_tcs.pg_tcs = DEF_TRAFFIC_CLASS;
6289                 adapter->dcb_cfg.num_tcs.pfc_tcs = DEF_TRAFFIC_CLASS;
6290                 break;
6291         }
6292
6293         /* Configure DCB traffic classes */
6294         for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
6295                 tc = &adapter->dcb_cfg.tc_config[j];
6296                 tc->path[DCB_TX_CONFIG].bwg_id = 0;
6297                 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
6298                 tc->path[DCB_RX_CONFIG].bwg_id = 0;
6299                 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
6300                 tc->dcb_pfc = pfc_disabled;
6301         }
6302
6303         /* Initialize default user to priority mapping, UPx->TC0 */
6304         tc = &adapter->dcb_cfg.tc_config[0];
6305         tc->path[DCB_TX_CONFIG].up_to_tc_bitmap = 0xFF;
6306         tc->path[DCB_RX_CONFIG].up_to_tc_bitmap = 0xFF;
6307
6308         adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
6309         adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
6310         adapter->dcb_cfg.pfc_mode_enable = false;
6311         adapter->dcb_set_bitmap = 0x00;
6312         if (adapter->flags & IXGBE_FLAG_DCB_CAPABLE)
6313                 adapter->dcbx_cap = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE;
6314         memcpy(&adapter->temp_dcb_cfg, &adapter->dcb_cfg,
6315                sizeof(adapter->temp_dcb_cfg));
6316 }
6317 #endif
6318
6319 /**
6320  * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
6321  * @adapter: board private structure to initialize
6322  * @ii: pointer to ixgbe_info for device
6323  *
6324  * ixgbe_sw_init initializes the Adapter private data structure.
6325  * Fields are initialized based on PCI device information and
6326  * OS network device settings (MTU size).
6327  **/
6328 static int ixgbe_sw_init(struct ixgbe_adapter *adapter,
6329                          const struct ixgbe_info *ii)
6330 {
6331         struct ixgbe_hw *hw = &adapter->hw;
6332         struct pci_dev *pdev = adapter->pdev;
6333         unsigned int rss, fdir;
6334         u32 fwsm;
6335         int i;
6336
6337         /* PCI config space info */
6338
6339         hw->vendor_id = pdev->vendor;
6340         hw->device_id = pdev->device;
6341         hw->revision_id = pdev->revision;
6342         hw->subsystem_vendor_id = pdev->subsystem_vendor;
6343         hw->subsystem_device_id = pdev->subsystem_device;
6344
6345         /* get_invariants needs the device IDs */
6346         ii->get_invariants(hw);
6347
6348         /* Set common capability flags and settings */
6349         rss = min_t(int, ixgbe_max_rss_indices(adapter), num_online_cpus());
6350         adapter->ring_feature[RING_F_RSS].limit = rss;
6351         adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
6352         adapter->max_q_vectors = MAX_Q_VECTORS_82599;
6353         adapter->atr_sample_rate = 20;
6354         fdir = min_t(int, IXGBE_MAX_FDIR_INDICES, num_online_cpus());
6355         adapter->ring_feature[RING_F_FDIR].limit = fdir;
6356         adapter->fdir_pballoc = IXGBE_FDIR_PBALLOC_64K;
6357         adapter->ring_feature[RING_F_VMDQ].limit = 1;
6358 #ifdef CONFIG_IXGBE_DCA
6359         adapter->flags |= IXGBE_FLAG_DCA_CAPABLE;
6360 #endif
6361 #ifdef CONFIG_IXGBE_DCB
6362         adapter->flags |= IXGBE_FLAG_DCB_CAPABLE;
6363         adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
6364 #endif
6365 #ifdef IXGBE_FCOE
6366         adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
6367         adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
6368 #ifdef CONFIG_IXGBE_DCB
6369         /* Default traffic class to use for FCoE */
6370         adapter->fcoe.up = IXGBE_FCOE_DEFTC;
6371 #endif /* CONFIG_IXGBE_DCB */
6372 #endif /* IXGBE_FCOE */
6373
6374         /* initialize static ixgbe jump table entries */
6375         adapter->jump_tables[0] = kzalloc(sizeof(*adapter->jump_tables[0]),
6376                                           GFP_KERNEL);
6377         if (!adapter->jump_tables[0])
6378                 return -ENOMEM;
6379         adapter->jump_tables[0]->mat = ixgbe_ipv4_fields;
6380
6381         for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++)
6382                 adapter->jump_tables[i] = NULL;
6383
6384         adapter->mac_table = kcalloc(hw->mac.num_rar_entries,
6385                                      sizeof(struct ixgbe_mac_addr),
6386                                      GFP_KERNEL);
6387         if (!adapter->mac_table)
6388                 return -ENOMEM;
6389
6390         if (ixgbe_init_rss_key(adapter))
6391                 return -ENOMEM;
6392
6393         adapter->af_xdp_zc_qps = bitmap_zalloc(IXGBE_MAX_XDP_QS, GFP_KERNEL);
6394         if (!adapter->af_xdp_zc_qps)
6395                 return -ENOMEM;
6396
6397         /* Set MAC specific capability flags and exceptions */
6398         switch (hw->mac.type) {
6399         case ixgbe_mac_82598EB:
6400                 adapter->flags2 &= ~IXGBE_FLAG2_RSC_CAPABLE;
6401
6402                 if (hw->device_id == IXGBE_DEV_ID_82598AT)
6403                         adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
6404
6405                 adapter->max_q_vectors = MAX_Q_VECTORS_82598;
6406                 adapter->ring_feature[RING_F_FDIR].limit = 0;
6407                 adapter->atr_sample_rate = 0;
6408                 adapter->fdir_pballoc = 0;
6409 #ifdef IXGBE_FCOE
6410                 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
6411                 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
6412 #ifdef CONFIG_IXGBE_DCB
6413                 adapter->fcoe.up = 0;
6414 #endif /* IXGBE_DCB */
6415 #endif /* IXGBE_FCOE */
6416                 break;
6417         case ixgbe_mac_82599EB:
6418                 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
6419                         adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6420                 break;
6421         case ixgbe_mac_X540:
6422                 fwsm = IXGBE_READ_REG(hw, IXGBE_FWSM(hw));
6423                 if (fwsm & IXGBE_FWSM_TS_ENABLED)
6424                         adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6425                 break;
6426         case ixgbe_mac_x550em_a:
6427                 switch (hw->device_id) {
6428                 case IXGBE_DEV_ID_X550EM_A_1G_T:
6429                 case IXGBE_DEV_ID_X550EM_A_1G_T_L:
6430                         adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6431                         break;
6432                 default:
6433                         break;
6434                 }
6435                 fallthrough;
6436         case ixgbe_mac_X550EM_x:
6437 #ifdef CONFIG_IXGBE_DCB
6438                 adapter->flags &= ~IXGBE_FLAG_DCB_CAPABLE;
6439 #endif
6440 #ifdef IXGBE_FCOE
6441                 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
6442 #ifdef CONFIG_IXGBE_DCB
6443                 adapter->fcoe.up = 0;
6444 #endif /* IXGBE_DCB */
6445 #endif /* IXGBE_FCOE */
6446                 fallthrough;
6447         case ixgbe_mac_X550:
6448                 if (hw->mac.type == ixgbe_mac_X550)
6449                         adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6450 #ifdef CONFIG_IXGBE_DCA
6451                 adapter->flags &= ~IXGBE_FLAG_DCA_CAPABLE;
6452 #endif
6453                 break;
6454         default:
6455                 break;
6456         }
6457
6458 #ifdef IXGBE_FCOE
6459         /* FCoE support exists, always init the FCoE lock */
6460         spin_lock_init(&adapter->fcoe.lock);
6461
6462 #endif
6463         /* n-tuple support exists, always init our spinlock */
6464         spin_lock_init(&adapter->fdir_perfect_lock);
6465
6466         /* init spinlock to avoid concurrency of VF resources */
6467         spin_lock_init(&adapter->vfs_lock);
6468
6469 #ifdef CONFIG_IXGBE_DCB
6470         ixgbe_init_dcb(adapter);
6471 #endif
6472         ixgbe_init_ipsec_offload(adapter);
6473
6474         /* default flow control settings */
6475         hw->fc.requested_mode = ixgbe_fc_full;
6476         hw->fc.current_mode = ixgbe_fc_full;    /* init for ethtool output */
6477         ixgbe_pbthresh_setup(adapter);
6478         hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
6479         hw->fc.send_xon = true;
6480         hw->fc.disable_fc_autoneg = ixgbe_device_supports_autoneg_fc(hw);
6481
6482 #ifdef CONFIG_PCI_IOV
6483         if (max_vfs > 0)
6484                 e_dev_warn("Enabling SR-IOV VFs using the max_vfs module parameter is deprecated - please use the pci sysfs interface instead.\n");
6485
6486         /* assign number of SR-IOV VFs */
6487         if (hw->mac.type != ixgbe_mac_82598EB) {
6488                 if (max_vfs > IXGBE_MAX_VFS_DRV_LIMIT) {
6489                         max_vfs = 0;
6490                         e_dev_warn("max_vfs parameter out of range. Not assigning any SR-IOV VFs\n");
6491                 }
6492         }
6493 #endif /* CONFIG_PCI_IOV */
6494
6495         /* enable itr by default in dynamic mode */
6496         adapter->rx_itr_setting = 1;
6497         adapter->tx_itr_setting = 1;
6498
6499         /* set default ring sizes */
6500         adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
6501         adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
6502
6503         /* set default work limits */
6504         adapter->tx_work_limit = IXGBE_DEFAULT_TX_WORK;
6505
6506         /* initialize eeprom parameters */
6507         if (ixgbe_init_eeprom_params_generic(hw)) {
6508                 e_dev_err("EEPROM initialization failed\n");
6509                 return -EIO;
6510         }
6511
6512         /* PF holds first pool slot */
6513         set_bit(0, adapter->fwd_bitmask);
6514         set_bit(__IXGBE_DOWN, &adapter->state);
6515
6516         /* enable locking for XDP_TX if we have more CPUs than queues */
6517         if (nr_cpu_ids > IXGBE_MAX_XDP_QS)
6518                 static_branch_enable(&ixgbe_xdp_locking_key);
6519
6520         return 0;
6521 }
6522
6523 /**
6524  * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
6525  * @tx_ring:    tx descriptor ring (for a specific queue) to setup
6526  *
6527  * Return 0 on success, negative on failure
6528  **/
6529 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
6530 {
6531         struct device *dev = tx_ring->dev;
6532         int orig_node = dev_to_node(dev);
6533         int ring_node = NUMA_NO_NODE;
6534         int size;
6535
6536         size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
6537
6538         if (tx_ring->q_vector)
6539                 ring_node = tx_ring->q_vector->numa_node;
6540
6541         tx_ring->tx_buffer_info = vmalloc_node(size, ring_node);
6542         if (!tx_ring->tx_buffer_info)
6543                 tx_ring->tx_buffer_info = vmalloc(size);
6544         if (!tx_ring->tx_buffer_info)
6545                 goto err;
6546
6547         /* round up to nearest 4K */
6548         tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
6549         tx_ring->size = ALIGN(tx_ring->size, 4096);
6550
6551         set_dev_node(dev, ring_node);
6552         tx_ring->desc = dma_alloc_coherent(dev,
6553                                            tx_ring->size,
6554                                            &tx_ring->dma,
6555                                            GFP_KERNEL);
6556         set_dev_node(dev, orig_node);
6557         if (!tx_ring->desc)
6558                 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
6559                                                    &tx_ring->dma, GFP_KERNEL);
6560         if (!tx_ring->desc)
6561                 goto err;
6562
6563         tx_ring->next_to_use = 0;
6564         tx_ring->next_to_clean = 0;
6565         return 0;
6566
6567 err:
6568         vfree(tx_ring->tx_buffer_info);
6569         tx_ring->tx_buffer_info = NULL;
6570         dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
6571         return -ENOMEM;
6572 }
6573
6574 /**
6575  * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
6576  * @adapter: board private structure
6577  *
6578  * If this function returns with an error, then it's possible one or
6579  * more of the rings is populated (while the rest are not).  It is the
6580  * callers duty to clean those orphaned rings.
6581  *
6582  * Return 0 on success, negative on failure
6583  **/
6584 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
6585 {
6586         int i, j = 0, err = 0;
6587
6588         for (i = 0; i < adapter->num_tx_queues; i++) {
6589                 err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
6590                 if (!err)
6591                         continue;
6592
6593                 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
6594                 goto err_setup_tx;
6595         }
6596         for (j = 0; j < adapter->num_xdp_queues; j++) {
6597                 err = ixgbe_setup_tx_resources(adapter->xdp_ring[j]);
6598                 if (!err)
6599                         continue;
6600
6601                 e_err(probe, "Allocation for Tx Queue %u failed\n", j);
6602                 goto err_setup_tx;
6603         }
6604
6605         return 0;
6606 err_setup_tx:
6607         /* rewind the index freeing the rings as we go */
6608         while (j--)
6609                 ixgbe_free_tx_resources(adapter->xdp_ring[j]);
6610         while (i--)
6611                 ixgbe_free_tx_resources(adapter->tx_ring[i]);
6612         return err;
6613 }
6614
6615 static int ixgbe_rx_napi_id(struct ixgbe_ring *rx_ring)
6616 {
6617         struct ixgbe_q_vector *q_vector = rx_ring->q_vector;
6618
6619         return q_vector ? q_vector->napi.napi_id : 0;
6620 }
6621
6622 /**
6623  * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
6624  * @adapter: pointer to ixgbe_adapter
6625  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
6626  *
6627  * Returns 0 on success, negative on failure
6628  **/
6629 int ixgbe_setup_rx_resources(struct ixgbe_adapter *adapter,
6630                              struct ixgbe_ring *rx_ring)
6631 {
6632         struct device *dev = rx_ring->dev;
6633         int orig_node = dev_to_node(dev);
6634         int ring_node = NUMA_NO_NODE;
6635         int size;
6636
6637         size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
6638
6639         if (rx_ring->q_vector)
6640                 ring_node = rx_ring->q_vector->numa_node;
6641
6642         rx_ring->rx_buffer_info = vmalloc_node(size, ring_node);
6643         if (!rx_ring->rx_buffer_info)
6644                 rx_ring->rx_buffer_info = vmalloc(size);
6645         if (!rx_ring->rx_buffer_info)
6646                 goto err;
6647
6648         /* Round up to nearest 4K */
6649         rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
6650         rx_ring->size = ALIGN(rx_ring->size, 4096);
6651
6652         set_dev_node(dev, ring_node);
6653         rx_ring->desc = dma_alloc_coherent(dev,
6654                                            rx_ring->size,
6655                                            &rx_ring->dma,
6656                                            GFP_KERNEL);
6657         set_dev_node(dev, orig_node);
6658         if (!rx_ring->desc)
6659                 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
6660                                                    &rx_ring->dma, GFP_KERNEL);
6661         if (!rx_ring->desc)
6662                 goto err;
6663
6664         rx_ring->next_to_clean = 0;
6665         rx_ring->next_to_use = 0;
6666
6667         /* XDP RX-queue info */
6668         if (xdp_rxq_info_reg(&rx_ring->xdp_rxq, adapter->netdev,
6669                              rx_ring->queue_index, ixgbe_rx_napi_id(rx_ring)) < 0)
6670                 goto err;
6671
6672         WRITE_ONCE(rx_ring->xdp_prog, adapter->xdp_prog);
6673
6674         return 0;
6675 err:
6676         vfree(rx_ring->rx_buffer_info);
6677         rx_ring->rx_buffer_info = NULL;
6678         dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
6679         return -ENOMEM;
6680 }
6681
6682 /**
6683  * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
6684  * @adapter: board private structure
6685  *
6686  * If this function returns with an error, then it's possible one or
6687  * more of the rings is populated (while the rest are not).  It is the
6688  * callers duty to clean those orphaned rings.
6689  *
6690  * Return 0 on success, negative on failure
6691  **/
6692 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
6693 {
6694         int i, err = 0;
6695
6696         for (i = 0; i < adapter->num_rx_queues; i++) {
6697                 err = ixgbe_setup_rx_resources(adapter, adapter->rx_ring[i]);
6698                 if (!err)
6699                         continue;
6700
6701                 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
6702                 goto err_setup_rx;
6703         }
6704
6705 #ifdef IXGBE_FCOE
6706         err = ixgbe_setup_fcoe_ddp_resources(adapter);
6707         if (!err)
6708 #endif
6709                 return 0;
6710 err_setup_rx:
6711         /* rewind the index freeing the rings as we go */
6712         while (i--)
6713                 ixgbe_free_rx_resources(adapter->rx_ring[i]);
6714         return err;
6715 }
6716
6717 /**
6718  * ixgbe_free_tx_resources - Free Tx Resources per Queue
6719  * @tx_ring: Tx descriptor ring for a specific queue
6720  *
6721  * Free all transmit software resources
6722  **/
6723 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
6724 {
6725         ixgbe_clean_tx_ring(tx_ring);
6726
6727         vfree(tx_ring->tx_buffer_info);
6728         tx_ring->tx_buffer_info = NULL;
6729
6730         /* if not set, then don't free */
6731         if (!tx_ring->desc)
6732                 return;
6733
6734         dma_free_coherent(tx_ring->dev, tx_ring->size,
6735                           tx_ring->desc, tx_ring->dma);
6736
6737         tx_ring->desc = NULL;
6738 }
6739
6740 /**
6741  * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
6742  * @adapter: board private structure
6743  *
6744  * Free all transmit software resources
6745  **/
6746 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
6747 {
6748         int i;
6749
6750         for (i = 0; i < adapter->num_tx_queues; i++)
6751                 if (adapter->tx_ring[i]->desc)
6752                         ixgbe_free_tx_resources(adapter->tx_ring[i]);
6753         for (i = 0; i < adapter->num_xdp_queues; i++)
6754                 if (adapter->xdp_ring[i]->desc)
6755                         ixgbe_free_tx_resources(adapter->xdp_ring[i]);
6756 }
6757
6758 /**
6759  * ixgbe_free_rx_resources - Free Rx Resources
6760  * @rx_ring: ring to clean the resources from
6761  *
6762  * Free all receive software resources
6763  **/
6764 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
6765 {
6766         ixgbe_clean_rx_ring(rx_ring);
6767
6768         rx_ring->xdp_prog = NULL;
6769         xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
6770         vfree(rx_ring->rx_buffer_info);
6771         rx_ring->rx_buffer_info = NULL;
6772
6773         /* if not set, then don't free */
6774         if (!rx_ring->desc)
6775                 return;
6776
6777         dma_free_coherent(rx_ring->dev, rx_ring->size,
6778                           rx_ring->desc, rx_ring->dma);
6779
6780         rx_ring->desc = NULL;
6781 }
6782
6783 /**
6784  * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
6785  * @adapter: board private structure
6786  *
6787  * Free all receive software resources
6788  **/
6789 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
6790 {
6791         int i;
6792
6793 #ifdef IXGBE_FCOE
6794         ixgbe_free_fcoe_ddp_resources(adapter);
6795
6796 #endif
6797         for (i = 0; i < adapter->num_rx_queues; i++)
6798                 if (adapter->rx_ring[i]->desc)
6799                         ixgbe_free_rx_resources(adapter->rx_ring[i]);
6800 }
6801
6802 /**
6803  * ixgbe_max_xdp_frame_size - returns the maximum allowed frame size for XDP
6804  * @adapter: device handle, pointer to adapter
6805  */
6806 static int ixgbe_max_xdp_frame_size(struct ixgbe_adapter *adapter)
6807 {
6808         if (PAGE_SIZE >= 8192 || adapter->flags2 & IXGBE_FLAG2_RX_LEGACY)
6809                 return IXGBE_RXBUFFER_2K;
6810         else
6811                 return IXGBE_RXBUFFER_3K;
6812 }
6813
6814 /**
6815  * ixgbe_change_mtu - Change the Maximum Transfer Unit
6816  * @netdev: network interface device structure
6817  * @new_mtu: new value for maximum frame size
6818  *
6819  * Returns 0 on success, negative on failure
6820  **/
6821 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
6822 {
6823         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6824
6825         if (ixgbe_enabled_xdp_adapter(adapter)) {
6826                 int new_frame_size = new_mtu + IXGBE_PKT_HDR_PAD;
6827
6828                 if (new_frame_size > ixgbe_max_xdp_frame_size(adapter)) {
6829                         e_warn(probe, "Requested MTU size is not supported with XDP\n");
6830                         return -EINVAL;
6831                 }
6832         }
6833
6834         /*
6835          * For 82599EB we cannot allow legacy VFs to enable their receive
6836          * paths when MTU greater than 1500 is configured.  So display a
6837          * warning that legacy VFs will be disabled.
6838          */
6839         if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
6840             (adapter->hw.mac.type == ixgbe_mac_82599EB) &&
6841             (new_mtu > ETH_DATA_LEN))
6842                 e_warn(probe, "Setting MTU > 1500 will disable legacy VFs\n");
6843
6844         netdev_dbg(netdev, "changing MTU from %d to %d\n",
6845                    netdev->mtu, new_mtu);
6846
6847         /* must set new MTU before calling down or up */
6848         WRITE_ONCE(netdev->mtu, new_mtu);
6849
6850         if (netif_running(netdev))
6851                 ixgbe_reinit_locked(adapter);
6852
6853         return 0;
6854 }
6855
6856 /**
6857  * ixgbe_open - Called when a network interface is made active
6858  * @netdev: network interface device structure
6859  *
6860  * Returns 0 on success, negative value on failure
6861  *
6862  * The open entry point is called when a network interface is made
6863  * active by the system (IFF_UP).  At this point all resources needed
6864  * for transmit and receive operations are allocated, the interrupt
6865  * handler is registered with the OS, the watchdog timer is started,
6866  * and the stack is notified that the interface is ready.
6867  **/
6868 int ixgbe_open(struct net_device *netdev)
6869 {
6870         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6871         struct ixgbe_hw *hw = &adapter->hw;
6872         int err, queues;
6873
6874         /* disallow open during test */
6875         if (test_bit(__IXGBE_TESTING, &adapter->state))
6876                 return -EBUSY;
6877
6878         netif_carrier_off(netdev);
6879
6880         /* allocate transmit descriptors */
6881         err = ixgbe_setup_all_tx_resources(adapter);
6882         if (err)
6883                 goto err_setup_tx;
6884
6885         /* allocate receive descriptors */
6886         err = ixgbe_setup_all_rx_resources(adapter);
6887         if (err)
6888                 goto err_setup_rx;
6889
6890         ixgbe_configure(adapter);
6891
6892         err = ixgbe_request_irq(adapter);
6893         if (err)
6894                 goto err_req_irq;
6895
6896         /* Notify the stack of the actual queue counts. */
6897         queues = adapter->num_tx_queues;
6898         err = netif_set_real_num_tx_queues(netdev, queues);
6899         if (err)
6900                 goto err_set_queues;
6901
6902         queues = adapter->num_rx_queues;
6903         err = netif_set_real_num_rx_queues(netdev, queues);
6904         if (err)
6905                 goto err_set_queues;
6906
6907         ixgbe_ptp_init(adapter);
6908
6909         ixgbe_up_complete(adapter);
6910
6911         udp_tunnel_nic_reset_ntf(netdev);
6912
6913         return 0;
6914
6915 err_set_queues:
6916         ixgbe_free_irq(adapter);
6917 err_req_irq:
6918         ixgbe_free_all_rx_resources(adapter);
6919         if (hw->phy.ops.set_phy_power && !adapter->wol)
6920                 hw->phy.ops.set_phy_power(&adapter->hw, false);
6921 err_setup_rx:
6922         ixgbe_free_all_tx_resources(adapter);
6923 err_setup_tx:
6924         ixgbe_reset(adapter);
6925
6926         return err;
6927 }
6928
6929 static void ixgbe_close_suspend(struct ixgbe_adapter *adapter)
6930 {
6931         ixgbe_ptp_suspend(adapter);
6932
6933         if (adapter->hw.phy.ops.enter_lplu) {
6934                 adapter->hw.phy.reset_disable = true;
6935                 ixgbe_down(adapter);
6936                 adapter->hw.phy.ops.enter_lplu(&adapter->hw);
6937                 adapter->hw.phy.reset_disable = false;
6938         } else {
6939                 ixgbe_down(adapter);
6940         }
6941
6942         ixgbe_free_irq(adapter);
6943
6944         ixgbe_free_all_tx_resources(adapter);
6945         ixgbe_free_all_rx_resources(adapter);
6946 }
6947
6948 /**
6949  * ixgbe_close - Disables a network interface
6950  * @netdev: network interface device structure
6951  *
6952  * Returns 0, this is not allowed to fail
6953  *
6954  * The close entry point is called when an interface is de-activated
6955  * by the OS.  The hardware is still under the drivers control, but
6956  * needs to be disabled.  A global MAC reset is issued to stop the
6957  * hardware, and all transmit and receive resources are freed.
6958  **/
6959 int ixgbe_close(struct net_device *netdev)
6960 {
6961         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6962
6963         ixgbe_ptp_stop(adapter);
6964
6965         if (netif_device_present(netdev))
6966                 ixgbe_close_suspend(adapter);
6967
6968         ixgbe_fdir_filter_exit(adapter);
6969
6970         ixgbe_release_hw_control(adapter);
6971
6972         return 0;
6973 }
6974
6975 static int ixgbe_resume(struct device *dev_d)
6976 {
6977         struct pci_dev *pdev = to_pci_dev(dev_d);
6978         struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
6979         struct net_device *netdev = adapter->netdev;
6980         u32 err;
6981
6982         adapter->hw.hw_addr = adapter->io_addr;
6983
6984         err = pci_enable_device_mem(pdev);
6985         if (err) {
6986                 e_dev_err("Cannot enable PCI device from suspend\n");
6987                 return err;
6988         }
6989         smp_mb__before_atomic();
6990         clear_bit(__IXGBE_DISABLED, &adapter->state);
6991         pci_set_master(pdev);
6992
6993         device_wakeup_disable(dev_d);
6994
6995         ixgbe_reset(adapter);
6996
6997         IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
6998
6999         rtnl_lock();
7000         err = ixgbe_init_interrupt_scheme(adapter);
7001         if (!err && netif_running(netdev))
7002                 err = ixgbe_open(netdev);
7003
7004
7005         if (!err)
7006                 netif_device_attach(netdev);
7007         rtnl_unlock();
7008
7009         return err;
7010 }
7011
7012 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
7013 {
7014         struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7015         struct net_device *netdev = adapter->netdev;
7016         struct ixgbe_hw *hw = &adapter->hw;
7017         u32 ctrl;
7018         u32 wufc = adapter->wol;
7019
7020         rtnl_lock();
7021         netif_device_detach(netdev);
7022
7023         if (netif_running(netdev))
7024                 ixgbe_close_suspend(adapter);
7025
7026         ixgbe_clear_interrupt_scheme(adapter);
7027         rtnl_unlock();
7028
7029         if (hw->mac.ops.stop_link_on_d3)
7030                 hw->mac.ops.stop_link_on_d3(hw);
7031
7032         if (wufc) {
7033                 u32 fctrl;
7034
7035                 ixgbe_set_rx_mode(netdev);
7036
7037                 /* enable the optics for 82599 SFP+ fiber as we can WoL */
7038                 if (hw->mac.ops.enable_tx_laser)
7039                         hw->mac.ops.enable_tx_laser(hw);
7040
7041                 /* enable the reception of multicast packets */
7042                 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
7043                 fctrl |= IXGBE_FCTRL_MPE;
7044                 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
7045
7046                 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
7047                 ctrl |= IXGBE_CTRL_GIO_DIS;
7048                 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
7049
7050                 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
7051         } else {
7052                 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
7053                 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
7054         }
7055
7056         switch (hw->mac.type) {
7057         case ixgbe_mac_82598EB:
7058                 pci_wake_from_d3(pdev, false);
7059                 break;
7060         case ixgbe_mac_82599EB:
7061         case ixgbe_mac_X540:
7062         case ixgbe_mac_X550:
7063         case ixgbe_mac_X550EM_x:
7064         case ixgbe_mac_x550em_a:
7065                 pci_wake_from_d3(pdev, !!wufc);
7066                 break;
7067         default:
7068                 break;
7069         }
7070
7071         *enable_wake = !!wufc;
7072         if (hw->phy.ops.set_phy_power && !*enable_wake)
7073                 hw->phy.ops.set_phy_power(hw, false);
7074
7075         ixgbe_release_hw_control(adapter);
7076
7077         if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state))
7078                 pci_disable_device(pdev);
7079
7080         return 0;
7081 }
7082
7083 static int ixgbe_suspend(struct device *dev_d)
7084 {
7085         struct pci_dev *pdev = to_pci_dev(dev_d);
7086         int retval;
7087         bool wake;
7088
7089         retval = __ixgbe_shutdown(pdev, &wake);
7090
7091         device_set_wakeup_enable(dev_d, wake);
7092
7093         return retval;
7094 }
7095
7096 static void ixgbe_shutdown(struct pci_dev *pdev)
7097 {
7098         bool wake;
7099
7100         __ixgbe_shutdown(pdev, &wake);
7101
7102         if (system_state == SYSTEM_POWER_OFF) {
7103                 pci_wake_from_d3(pdev, wake);
7104                 pci_set_power_state(pdev, PCI_D3hot);
7105         }
7106 }
7107
7108 /**
7109  * ixgbe_update_stats - Update the board statistics counters.
7110  * @adapter: board private structure
7111  **/
7112 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
7113 {
7114         struct net_device *netdev = adapter->netdev;
7115         struct ixgbe_hw *hw = &adapter->hw;
7116         struct ixgbe_hw_stats *hwstats = &adapter->stats;
7117         u64 total_mpc = 0;
7118         u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
7119         u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
7120         u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
7121         u64 alloc_rx_page = 0;
7122         u64 bytes = 0, packets = 0, hw_csum_rx_error = 0;
7123
7124         if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7125             test_bit(__IXGBE_RESETTING, &adapter->state))
7126                 return;
7127
7128         if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
7129                 u64 rsc_count = 0;
7130                 u64 rsc_flush = 0;
7131                 for (i = 0; i < adapter->num_rx_queues; i++) {
7132                         rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
7133                         rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
7134                 }
7135                 adapter->rsc_total_count = rsc_count;
7136                 adapter->rsc_total_flush = rsc_flush;
7137         }
7138
7139         for (i = 0; i < adapter->num_rx_queues; i++) {
7140                 struct ixgbe_ring *rx_ring = READ_ONCE(adapter->rx_ring[i]);
7141
7142                 if (!rx_ring)
7143                         continue;
7144                 non_eop_descs += rx_ring->rx_stats.non_eop_descs;
7145                 alloc_rx_page += rx_ring->rx_stats.alloc_rx_page;
7146                 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
7147                 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
7148                 hw_csum_rx_error += rx_ring->rx_stats.csum_err;
7149                 bytes += rx_ring->stats.bytes;
7150                 packets += rx_ring->stats.packets;
7151         }
7152         adapter->non_eop_descs = non_eop_descs;
7153         adapter->alloc_rx_page = alloc_rx_page;
7154         adapter->alloc_rx_page_failed = alloc_rx_page_failed;
7155         adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
7156         adapter->hw_csum_rx_error = hw_csum_rx_error;
7157         netdev->stats.rx_bytes = bytes;
7158         netdev->stats.rx_packets = packets;
7159
7160         bytes = 0;
7161         packets = 0;
7162         /* gather some stats to the adapter struct that are per queue */
7163         for (i = 0; i < adapter->num_tx_queues; i++) {
7164                 struct ixgbe_ring *tx_ring = READ_ONCE(adapter->tx_ring[i]);
7165
7166                 if (!tx_ring)
7167                         continue;
7168                 restart_queue += tx_ring->tx_stats.restart_queue;
7169                 tx_busy += tx_ring->tx_stats.tx_busy;
7170                 bytes += tx_ring->stats.bytes;
7171                 packets += tx_ring->stats.packets;
7172         }
7173         for (i = 0; i < adapter->num_xdp_queues; i++) {
7174                 struct ixgbe_ring *xdp_ring = READ_ONCE(adapter->xdp_ring[i]);
7175
7176                 if (!xdp_ring)
7177                         continue;
7178                 restart_queue += xdp_ring->tx_stats.restart_queue;
7179                 tx_busy += xdp_ring->tx_stats.tx_busy;
7180                 bytes += xdp_ring->stats.bytes;
7181                 packets += xdp_ring->stats.packets;
7182         }
7183         adapter->restart_queue = restart_queue;
7184         adapter->tx_busy = tx_busy;
7185         netdev->stats.tx_bytes = bytes;
7186         netdev->stats.tx_packets = packets;
7187
7188         hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
7189
7190         /* 8 register reads */
7191         for (i = 0; i < 8; i++) {
7192                 /* for packet buffers not used, the register should read 0 */
7193                 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
7194                 missed_rx += mpc;
7195                 hwstats->mpc[i] += mpc;
7196                 total_mpc += hwstats->mpc[i];
7197                 hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
7198                 hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
7199                 switch (hw->mac.type) {
7200                 case ixgbe_mac_82598EB:
7201                         hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
7202                         hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
7203                         hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
7204                         hwstats->pxonrxc[i] +=
7205                                 IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
7206                         break;
7207                 case ixgbe_mac_82599EB:
7208                 case ixgbe_mac_X540:
7209                 case ixgbe_mac_X550:
7210                 case ixgbe_mac_X550EM_x:
7211                 case ixgbe_mac_x550em_a:
7212                         hwstats->pxonrxc[i] +=
7213                                 IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
7214                         break;
7215                 default:
7216                         break;
7217                 }
7218         }
7219
7220         /*16 register reads */
7221         for (i = 0; i < 16; i++) {
7222                 hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
7223                 hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
7224                 if ((hw->mac.type == ixgbe_mac_82599EB) ||
7225                     (hw->mac.type == ixgbe_mac_X540) ||
7226                     (hw->mac.type == ixgbe_mac_X550) ||
7227                     (hw->mac.type == ixgbe_mac_X550EM_x) ||
7228                     (hw->mac.type == ixgbe_mac_x550em_a)) {
7229                         hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC_L(i));
7230                         IXGBE_READ_REG(hw, IXGBE_QBTC_H(i)); /* to clear */
7231                         hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC_L(i));
7232                         IXGBE_READ_REG(hw, IXGBE_QBRC_H(i)); /* to clear */
7233                 }
7234         }
7235
7236         hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
7237         /* work around hardware counting issue */
7238         hwstats->gprc -= missed_rx;
7239
7240         ixgbe_update_xoff_received(adapter);
7241
7242         /* 82598 hardware only has a 32 bit counter in the high register */
7243         switch (hw->mac.type) {
7244         case ixgbe_mac_82598EB:
7245                 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
7246                 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
7247                 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
7248                 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
7249                 break;
7250         case ixgbe_mac_X540:
7251         case ixgbe_mac_X550:
7252         case ixgbe_mac_X550EM_x:
7253         case ixgbe_mac_x550em_a:
7254                 /* OS2BMC stats are X540 and later */
7255                 hwstats->o2bgptc += IXGBE_READ_REG(hw, IXGBE_O2BGPTC);
7256                 hwstats->o2bspc += IXGBE_READ_REG(hw, IXGBE_O2BSPC);
7257                 hwstats->b2ospc += IXGBE_READ_REG(hw, IXGBE_B2OSPC);
7258                 hwstats->b2ogprc += IXGBE_READ_REG(hw, IXGBE_B2OGPRC);
7259                 fallthrough;
7260         case ixgbe_mac_82599EB:
7261                 for (i = 0; i < 16; i++)
7262                         adapter->hw_rx_no_dma_resources +=
7263                                              IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
7264                 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
7265                 IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
7266                 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
7267                 IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
7268                 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
7269                 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
7270                 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
7271                 hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
7272                 hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
7273 #ifdef IXGBE_FCOE
7274                 hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
7275                 hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
7276                 hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
7277                 hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
7278                 hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
7279                 hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
7280                 /* Add up per cpu counters for total ddp aloc fail */
7281                 if (adapter->fcoe.ddp_pool) {
7282                         struct ixgbe_fcoe *fcoe = &adapter->fcoe;
7283                         struct ixgbe_fcoe_ddp_pool *ddp_pool;
7284                         unsigned int cpu;
7285                         u64 noddp = 0, noddp_ext_buff = 0;
7286                         for_each_possible_cpu(cpu) {
7287                                 ddp_pool = per_cpu_ptr(fcoe->ddp_pool, cpu);
7288                                 noddp += ddp_pool->noddp;
7289                                 noddp_ext_buff += ddp_pool->noddp_ext_buff;
7290                         }
7291                         hwstats->fcoe_noddp = noddp;
7292                         hwstats->fcoe_noddp_ext_buff = noddp_ext_buff;
7293                 }
7294 #endif /* IXGBE_FCOE */
7295                 break;
7296         default:
7297                 break;
7298         }
7299         bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
7300         hwstats->bprc += bprc;
7301         hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
7302         if (hw->mac.type == ixgbe_mac_82598EB)
7303                 hwstats->mprc -= bprc;
7304         hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
7305         hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
7306         hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
7307         hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
7308         hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
7309         hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
7310         hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
7311         hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
7312         lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
7313         hwstats->lxontxc += lxon;
7314         lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
7315         hwstats->lxofftxc += lxoff;
7316         hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
7317         hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
7318         /*
7319          * 82598 errata - tx of flow control packets is included in tx counters
7320          */
7321         xon_off_tot = lxon + lxoff;
7322         hwstats->gptc -= xon_off_tot;
7323         hwstats->mptc -= xon_off_tot;
7324         hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
7325         hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
7326         hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
7327         hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
7328         hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
7329         hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
7330         hwstats->ptc64 -= xon_off_tot;
7331         hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
7332         hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
7333         hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
7334         hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
7335         hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
7336         hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
7337
7338         /* Fill out the OS statistics structure */
7339         netdev->stats.multicast = hwstats->mprc;
7340
7341         /* Rx Errors */
7342         netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
7343         netdev->stats.rx_dropped = 0;
7344         netdev->stats.rx_length_errors = hwstats->rlec;
7345         netdev->stats.rx_crc_errors = hwstats->crcerrs;
7346         netdev->stats.rx_missed_errors = total_mpc;
7347
7348         /* VF Stats Collection - skip while resetting because these
7349          * are not clear on read and otherwise you'll sometimes get
7350          * crazy values.
7351          */
7352         if (!test_bit(__IXGBE_RESETTING, &adapter->state)) {
7353                 for (i = 0; i < adapter->num_vfs; i++) {
7354                         UPDATE_VF_COUNTER_32bit(IXGBE_PVFGPRC(i),
7355                                                 adapter->vfinfo[i].last_vfstats.gprc,
7356                                                 adapter->vfinfo[i].vfstats.gprc);
7357                         UPDATE_VF_COUNTER_32bit(IXGBE_PVFGPTC(i),
7358                                                 adapter->vfinfo[i].last_vfstats.gptc,
7359                                                 adapter->vfinfo[i].vfstats.gptc);
7360                         UPDATE_VF_COUNTER_36bit(IXGBE_PVFGORC_LSB(i),
7361                                                 IXGBE_PVFGORC_MSB(i),
7362                                                 adapter->vfinfo[i].last_vfstats.gorc,
7363                                                 adapter->vfinfo[i].vfstats.gorc);
7364                         UPDATE_VF_COUNTER_36bit(IXGBE_PVFGOTC_LSB(i),
7365                                                 IXGBE_PVFGOTC_MSB(i),
7366                                                 adapter->vfinfo[i].last_vfstats.gotc,
7367                                                 adapter->vfinfo[i].vfstats.gotc);
7368                         UPDATE_VF_COUNTER_32bit(IXGBE_PVFMPRC(i),
7369                                                 adapter->vfinfo[i].last_vfstats.mprc,
7370                                                 adapter->vfinfo[i].vfstats.mprc);
7371                 }
7372         }
7373 }
7374
7375 /**
7376  * ixgbe_fdir_reinit_subtask - worker thread to reinit FDIR filter table
7377  * @adapter: pointer to the device adapter structure
7378  **/
7379 static void ixgbe_fdir_reinit_subtask(struct ixgbe_adapter *adapter)
7380 {
7381         struct ixgbe_hw *hw = &adapter->hw;
7382         int i;
7383
7384         if (!(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
7385                 return;
7386
7387         adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
7388
7389         /* if interface is down do nothing */
7390         if (test_bit(__IXGBE_DOWN, &adapter->state))
7391                 return;
7392
7393         /* do nothing if we are not using signature filters */
7394         if (!(adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE))
7395                 return;
7396
7397         adapter->fdir_overflow++;
7398
7399         if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
7400                 for (i = 0; i < adapter->num_tx_queues; i++)
7401                         set_bit(__IXGBE_TX_FDIR_INIT_DONE,
7402                                 &(adapter->tx_ring[i]->state));
7403                 for (i = 0; i < adapter->num_xdp_queues; i++)
7404                         set_bit(__IXGBE_TX_FDIR_INIT_DONE,
7405                                 &adapter->xdp_ring[i]->state);
7406                 /* re-enable flow director interrupts */
7407                 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_FLOW_DIR);
7408         } else {
7409                 e_err(probe, "failed to finish FDIR re-initialization, "
7410                       "ignored adding FDIR ATR filters\n");
7411         }
7412 }
7413
7414 /**
7415  * ixgbe_check_hang_subtask - check for hung queues and dropped interrupts
7416  * @adapter: pointer to the device adapter structure
7417  *
7418  * This function serves two purposes.  First it strobes the interrupt lines
7419  * in order to make certain interrupts are occurring.  Secondly it sets the
7420  * bits needed to check for TX hangs.  As a result we should immediately
7421  * determine if a hang has occurred.
7422  */
7423 static void ixgbe_check_hang_subtask(struct ixgbe_adapter *adapter)
7424 {
7425         struct ixgbe_hw *hw = &adapter->hw;
7426         u64 eics = 0;
7427         int i;
7428
7429         /* If we're down, removing or resetting, just bail */
7430         if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7431             test_bit(__IXGBE_REMOVING, &adapter->state) ||
7432             test_bit(__IXGBE_RESETTING, &adapter->state))
7433                 return;
7434
7435         /* Force detection of hung controller */
7436         if (netif_carrier_ok(adapter->netdev)) {
7437                 for (i = 0; i < adapter->num_tx_queues; i++)
7438                         set_check_for_tx_hang(adapter->tx_ring[i]);
7439                 for (i = 0; i < adapter->num_xdp_queues; i++)
7440                         set_check_for_tx_hang(adapter->xdp_ring[i]);
7441         }
7442
7443         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
7444                 /*
7445                  * for legacy and MSI interrupts don't set any bits
7446                  * that are enabled for EIAM, because this operation
7447                  * would set *both* EIMS and EICS for any bit in EIAM
7448                  */
7449                 IXGBE_WRITE_REG(hw, IXGBE_EICS,
7450                         (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
7451         } else {
7452                 /* get one bit for every active tx/rx interrupt vector */
7453                 for (i = 0; i < adapter->num_q_vectors; i++) {
7454                         struct ixgbe_q_vector *qv = adapter->q_vector[i];
7455                         if (qv->rx.ring || qv->tx.ring)
7456                                 eics |= BIT_ULL(i);
7457                 }
7458         }
7459
7460         /* Cause software interrupt to ensure rings are cleaned */
7461         ixgbe_irq_rearm_queues(adapter, eics);
7462 }
7463
7464 /**
7465  * ixgbe_watchdog_update_link - update the link status
7466  * @adapter: pointer to the device adapter structure
7467  **/
7468 static void ixgbe_watchdog_update_link(struct ixgbe_adapter *adapter)
7469 {
7470         struct ixgbe_hw *hw = &adapter->hw;
7471         u32 link_speed = adapter->link_speed;
7472         bool link_up = adapter->link_up;
7473         bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
7474
7475         if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
7476                 return;
7477
7478         if (hw->mac.ops.check_link) {
7479                 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
7480         } else {
7481                 /* always assume link is up, if no check link function */
7482                 link_speed = IXGBE_LINK_SPEED_10GB_FULL;
7483                 link_up = true;
7484         }
7485
7486         if (adapter->ixgbe_ieee_pfc)
7487                 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
7488
7489         if (link_up && !((adapter->flags & IXGBE_FLAG_DCB_ENABLED) && pfc_en)) {
7490                 hw->mac.ops.fc_enable(hw);
7491                 ixgbe_set_rx_drop_en(adapter);
7492         }
7493
7494         if (link_up ||
7495             time_after(jiffies, (adapter->link_check_timeout +
7496                                  IXGBE_TRY_LINK_TIMEOUT))) {
7497                 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
7498                 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
7499                 IXGBE_WRITE_FLUSH(hw);
7500         }
7501
7502         adapter->link_up = link_up;
7503         adapter->link_speed = link_speed;
7504 }
7505
7506 static void ixgbe_update_default_up(struct ixgbe_adapter *adapter)
7507 {
7508 #ifdef CONFIG_IXGBE_DCB
7509         struct net_device *netdev = adapter->netdev;
7510         struct dcb_app app = {
7511                               .selector = IEEE_8021QAZ_APP_SEL_ETHERTYPE,
7512                               .protocol = 0,
7513                              };
7514         u8 up = 0;
7515
7516         if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_IEEE)
7517                 up = dcb_ieee_getapp_mask(netdev, &app);
7518
7519         adapter->default_up = (up > 1) ? (ffs(up) - 1) : 0;
7520 #endif
7521 }
7522
7523 /**
7524  * ixgbe_watchdog_link_is_up - update netif_carrier status and
7525  *                             print link up message
7526  * @adapter: pointer to the device adapter structure
7527  **/
7528 static void ixgbe_watchdog_link_is_up(struct ixgbe_adapter *adapter)
7529 {
7530         struct net_device *netdev = adapter->netdev;
7531         struct ixgbe_hw *hw = &adapter->hw;
7532         u32 link_speed = adapter->link_speed;
7533         const char *speed_str;
7534         bool flow_rx, flow_tx;
7535
7536         /* only continue if link was previously down */
7537         if (netif_carrier_ok(netdev))
7538                 return;
7539
7540         adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
7541
7542         switch (hw->mac.type) {
7543         case ixgbe_mac_82598EB: {
7544                 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
7545                 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
7546                 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
7547                 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
7548         }
7549                 break;
7550         case ixgbe_mac_X540:
7551         case ixgbe_mac_X550:
7552         case ixgbe_mac_X550EM_x:
7553         case ixgbe_mac_x550em_a:
7554         case ixgbe_mac_82599EB: {
7555                 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
7556                 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
7557                 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
7558                 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
7559         }
7560                 break;
7561         default:
7562                 flow_tx = false;
7563                 flow_rx = false;
7564                 break;
7565         }
7566
7567         adapter->last_rx_ptp_check = jiffies;
7568
7569         if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
7570                 ixgbe_ptp_start_cyclecounter(adapter);
7571
7572         switch (link_speed) {
7573         case IXGBE_LINK_SPEED_10GB_FULL:
7574                 speed_str = "10 Gbps";
7575                 break;
7576         case IXGBE_LINK_SPEED_5GB_FULL:
7577                 speed_str = "5 Gbps";
7578                 break;
7579         case IXGBE_LINK_SPEED_2_5GB_FULL:
7580                 speed_str = "2.5 Gbps";
7581                 break;
7582         case IXGBE_LINK_SPEED_1GB_FULL:
7583                 speed_str = "1 Gbps";
7584                 break;
7585         case IXGBE_LINK_SPEED_100_FULL:
7586                 speed_str = "100 Mbps";
7587                 break;
7588         case IXGBE_LINK_SPEED_10_FULL:
7589                 speed_str = "10 Mbps";
7590                 break;
7591         default:
7592                 speed_str = "unknown speed";
7593                 break;
7594         }
7595         e_info(drv, "NIC Link is Up %s, Flow Control: %s\n", speed_str,
7596                ((flow_rx && flow_tx) ? "RX/TX" :
7597                (flow_rx ? "RX" :
7598                (flow_tx ? "TX" : "None"))));
7599
7600         netif_carrier_on(netdev);
7601         ixgbe_check_vf_rate_limit(adapter);
7602
7603         /* enable transmits */
7604         netif_tx_wake_all_queues(adapter->netdev);
7605
7606         /* update the default user priority for VFs */
7607         ixgbe_update_default_up(adapter);
7608
7609         /* ping all the active vfs to let them know link has changed */
7610         ixgbe_ping_all_vfs(adapter);
7611 }
7612
7613 /**
7614  * ixgbe_watchdog_link_is_down - update netif_carrier status and
7615  *                               print link down message
7616  * @adapter: pointer to the adapter structure
7617  **/
7618 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *adapter)
7619 {
7620         struct net_device *netdev = adapter->netdev;
7621         struct ixgbe_hw *hw = &adapter->hw;
7622
7623         adapter->link_up = false;
7624         adapter->link_speed = 0;
7625
7626         /* only continue if link was up previously */
7627         if (!netif_carrier_ok(netdev))
7628                 return;
7629
7630         /* poll for SFP+ cable when link is down */
7631         if (ixgbe_is_sfp(hw) && hw->mac.type == ixgbe_mac_82598EB)
7632                 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
7633
7634         if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
7635                 ixgbe_ptp_start_cyclecounter(adapter);
7636
7637         e_info(drv, "NIC Link is Down\n");
7638         netif_carrier_off(netdev);
7639
7640         /* ping all the active vfs to let them know link has changed */
7641         ixgbe_ping_all_vfs(adapter);
7642 }
7643
7644 static bool ixgbe_ring_tx_pending(struct ixgbe_adapter *adapter)
7645 {
7646         int i;
7647
7648         for (i = 0; i < adapter->num_tx_queues; i++) {
7649                 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
7650
7651                 if (tx_ring->next_to_use != tx_ring->next_to_clean)
7652                         return true;
7653         }
7654
7655         for (i = 0; i < adapter->num_xdp_queues; i++) {
7656                 struct ixgbe_ring *ring = adapter->xdp_ring[i];
7657
7658                 if (ring->next_to_use != ring->next_to_clean)
7659                         return true;
7660         }
7661
7662         return false;
7663 }
7664
7665 static bool ixgbe_vf_tx_pending(struct ixgbe_adapter *adapter)
7666 {
7667         struct ixgbe_hw *hw = &adapter->hw;
7668         struct ixgbe_ring_feature *vmdq = &adapter->ring_feature[RING_F_VMDQ];
7669         u32 q_per_pool = __ALIGN_MASK(1, ~vmdq->mask);
7670
7671         int i, j;
7672
7673         if (!adapter->num_vfs)
7674                 return false;
7675
7676         /* resetting the PF is only needed for MAC before X550 */
7677         if (hw->mac.type >= ixgbe_mac_X550)
7678                 return false;
7679
7680         for (i = 0; i < adapter->num_vfs; i++) {
7681                 for (j = 0; j < q_per_pool; j++) {
7682                         u32 h, t;
7683
7684                         h = IXGBE_READ_REG(hw, IXGBE_PVFTDHN(q_per_pool, i, j));
7685                         t = IXGBE_READ_REG(hw, IXGBE_PVFTDTN(q_per_pool, i, j));
7686
7687                         if (h != t)
7688                                 return true;
7689                 }
7690         }
7691
7692         return false;
7693 }
7694
7695 /**
7696  * ixgbe_watchdog_flush_tx - flush queues on link down
7697  * @adapter: pointer to the device adapter structure
7698  **/
7699 static void ixgbe_watchdog_flush_tx(struct ixgbe_adapter *adapter)
7700 {
7701         if (!netif_carrier_ok(adapter->netdev)) {
7702                 if (ixgbe_ring_tx_pending(adapter) ||
7703                     ixgbe_vf_tx_pending(adapter)) {
7704                         /* We've lost link, so the controller stops DMA,
7705                          * but we've got queued Tx work that's never going
7706                          * to get done, so reset controller to flush Tx.
7707                          * (Do the reset outside of interrupt context).
7708                          */
7709                         e_warn(drv, "initiating reset to clear Tx work after link loss\n");
7710                         set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
7711                 }
7712         }
7713 }
7714
7715 #ifdef CONFIG_PCI_IOV
7716 static void ixgbe_bad_vf_abort(struct ixgbe_adapter *adapter, u32 vf)
7717 {
7718         struct ixgbe_hw *hw = &adapter->hw;
7719
7720         if (adapter->hw.mac.type == ixgbe_mac_82599EB &&
7721             adapter->flags2 & IXGBE_FLAG2_AUTO_DISABLE_VF) {
7722                 adapter->vfinfo[vf].primary_abort_count++;
7723                 if (adapter->vfinfo[vf].primary_abort_count ==
7724                     IXGBE_PRIMARY_ABORT_LIMIT) {
7725                         ixgbe_set_vf_link_state(adapter, vf,
7726                                                 IFLA_VF_LINK_STATE_DISABLE);
7727                         adapter->vfinfo[vf].primary_abort_count = 0;
7728
7729                         e_info(drv,
7730                                "Malicious Driver Detection event detected on PF %d VF %d MAC: %pM mdd-disable-vf=on",
7731                                hw->bus.func, vf,
7732                                adapter->vfinfo[vf].vf_mac_addresses);
7733                 }
7734         }
7735 }
7736
7737 static void ixgbe_check_for_bad_vf(struct ixgbe_adapter *adapter)
7738 {
7739         struct ixgbe_hw *hw = &adapter->hw;
7740         struct pci_dev *pdev = adapter->pdev;
7741         unsigned int vf;
7742         u32 gpc;
7743
7744         if (!(netif_carrier_ok(adapter->netdev)))
7745                 return;
7746
7747         gpc = IXGBE_READ_REG(hw, IXGBE_TXDGPC);
7748         if (gpc) /* If incrementing then no need for the check below */
7749                 return;
7750         /* Check to see if a bad DMA write target from an errant or
7751          * malicious VF has caused a PCIe error.  If so then we can
7752          * issue a VFLR to the offending VF(s) and then resume without
7753          * requesting a full slot reset.
7754          */
7755
7756         if (!pdev)
7757                 return;
7758
7759         /* check status reg for all VFs owned by this PF */
7760         for (vf = 0; vf < adapter->num_vfs; ++vf) {
7761                 struct pci_dev *vfdev = adapter->vfinfo[vf].vfdev;
7762                 u16 status_reg;
7763
7764                 if (!vfdev)
7765                         continue;
7766                 pci_read_config_word(vfdev, PCI_STATUS, &status_reg);
7767                 if (status_reg != IXGBE_FAILED_READ_CFG_WORD &&
7768                     status_reg & PCI_STATUS_REC_MASTER_ABORT) {
7769                         ixgbe_bad_vf_abort(adapter, vf);
7770                         pcie_flr(vfdev);
7771                 }
7772         }
7773 }
7774
7775 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
7776 {
7777         u32 ssvpc;
7778
7779         /* Do not perform spoof check for 82598 or if not in IOV mode */
7780         if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
7781             adapter->num_vfs == 0)
7782                 return;
7783
7784         ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
7785
7786         /*
7787          * ssvpc register is cleared on read, if zero then no
7788          * spoofed packets in the last interval.
7789          */
7790         if (!ssvpc)
7791                 return;
7792
7793         e_warn(drv, "%u Spoofed packets detected\n", ssvpc);
7794 }
7795 #else
7796 static void ixgbe_spoof_check(struct ixgbe_adapter __always_unused *adapter)
7797 {
7798 }
7799
7800 static void
7801 ixgbe_check_for_bad_vf(struct ixgbe_adapter __always_unused *adapter)
7802 {
7803 }
7804 #endif /* CONFIG_PCI_IOV */
7805
7806
7807 /**
7808  * ixgbe_watchdog_subtask - check and bring link up
7809  * @adapter: pointer to the device adapter structure
7810  **/
7811 static void ixgbe_watchdog_subtask(struct ixgbe_adapter *adapter)
7812 {
7813         /* if interface is down, removing or resetting, do nothing */
7814         if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7815             test_bit(__IXGBE_REMOVING, &adapter->state) ||
7816             test_bit(__IXGBE_RESETTING, &adapter->state))
7817                 return;
7818
7819         ixgbe_watchdog_update_link(adapter);
7820
7821         if (adapter->link_up)
7822                 ixgbe_watchdog_link_is_up(adapter);
7823         else
7824                 ixgbe_watchdog_link_is_down(adapter);
7825
7826         ixgbe_check_for_bad_vf(adapter);
7827         ixgbe_spoof_check(adapter);
7828         ixgbe_update_stats(adapter);
7829
7830         ixgbe_watchdog_flush_tx(adapter);
7831 }
7832
7833 /**
7834  * ixgbe_sfp_detection_subtask - poll for SFP+ cable
7835  * @adapter: the ixgbe adapter structure
7836  **/
7837 static void ixgbe_sfp_detection_subtask(struct ixgbe_adapter *adapter)
7838 {
7839         struct ixgbe_hw *hw = &adapter->hw;
7840         int err;
7841
7842         /* not searching for SFP so there is nothing to do here */
7843         if (!(adapter->flags2 & IXGBE_FLAG2_SEARCH_FOR_SFP) &&
7844             !(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
7845                 return;
7846
7847         if (adapter->sfp_poll_time &&
7848             time_after(adapter->sfp_poll_time, jiffies))
7849                 return; /* If not yet time to poll for SFP */
7850
7851         /* someone else is in init, wait until next service event */
7852         if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
7853                 return;
7854
7855         adapter->sfp_poll_time = jiffies + IXGBE_SFP_POLL_JIFFIES - 1;
7856
7857         err = hw->phy.ops.identify_sfp(hw);
7858         if (err == -EOPNOTSUPP)
7859                 goto sfp_out;
7860
7861         if (err == -ENOENT) {
7862                 /* If no cable is present, then we need to reset
7863                  * the next time we find a good cable. */
7864                 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
7865         }
7866
7867         /* exit on error */
7868         if (err)
7869                 goto sfp_out;
7870
7871         /* exit if reset not needed */
7872         if (!(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
7873                 goto sfp_out;
7874
7875         adapter->flags2 &= ~IXGBE_FLAG2_SFP_NEEDS_RESET;
7876
7877         /*
7878          * A module may be identified correctly, but the EEPROM may not have
7879          * support for that module.  setup_sfp() will fail in that case, so
7880          * we should not allow that module to load.
7881          */
7882         if (hw->mac.type == ixgbe_mac_82598EB)
7883                 err = hw->phy.ops.reset(hw);
7884         else
7885                 err = hw->mac.ops.setup_sfp(hw);
7886
7887         if (err == -EOPNOTSUPP)
7888                 goto sfp_out;
7889
7890         adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
7891         e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
7892
7893 sfp_out:
7894         clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
7895
7896         if (err == -EOPNOTSUPP &&
7897             adapter->netdev->reg_state == NETREG_REGISTERED) {
7898                 e_dev_err("failed to initialize because an unsupported "
7899                           "SFP+ module type was detected.\n");
7900                 e_dev_err("Reload the driver after installing a "
7901                           "supported module.\n");
7902                 unregister_netdev(adapter->netdev);
7903         }
7904 }
7905
7906 /**
7907  * ixgbe_sfp_link_config_subtask - set up link SFP after module install
7908  * @adapter: the ixgbe adapter structure
7909  **/
7910 static void ixgbe_sfp_link_config_subtask(struct ixgbe_adapter *adapter)
7911 {
7912         struct ixgbe_hw *hw = &adapter->hw;
7913         u32 cap_speed;
7914         u32 speed;
7915         bool autoneg = false;
7916
7917         if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_CONFIG))
7918                 return;
7919
7920         /* someone else is in init, wait until next service event */
7921         if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
7922                 return;
7923
7924         adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
7925
7926         hw->mac.ops.get_link_capabilities(hw, &cap_speed, &autoneg);
7927
7928         /* advertise highest capable link speed */
7929         if (!autoneg && (cap_speed & IXGBE_LINK_SPEED_10GB_FULL))
7930                 speed = IXGBE_LINK_SPEED_10GB_FULL;
7931         else
7932                 speed = cap_speed & (IXGBE_LINK_SPEED_10GB_FULL |
7933                                      IXGBE_LINK_SPEED_1GB_FULL);
7934
7935         if (hw->mac.ops.setup_link)
7936                 hw->mac.ops.setup_link(hw, speed, true);
7937
7938         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
7939         adapter->link_check_timeout = jiffies;
7940         clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
7941 }
7942
7943 /**
7944  * ixgbe_service_timer - Timer Call-back
7945  * @t: pointer to timer_list structure
7946  **/
7947 static void ixgbe_service_timer(struct timer_list *t)
7948 {
7949         struct ixgbe_adapter *adapter = from_timer(adapter, t, service_timer);
7950         unsigned long next_event_offset;
7951
7952         /* poll faster when waiting for link */
7953         if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
7954                 next_event_offset = HZ / 10;
7955         else
7956                 next_event_offset = HZ * 2;
7957
7958         /* Reset the timer */
7959         mod_timer(&adapter->service_timer, next_event_offset + jiffies);
7960
7961         ixgbe_service_event_schedule(adapter);
7962 }
7963
7964 static void ixgbe_phy_interrupt_subtask(struct ixgbe_adapter *adapter)
7965 {
7966         struct ixgbe_hw *hw = &adapter->hw;
7967         bool overtemp;
7968
7969         if (!(adapter->flags2 & IXGBE_FLAG2_PHY_INTERRUPT))
7970                 return;
7971
7972         adapter->flags2 &= ~IXGBE_FLAG2_PHY_INTERRUPT;
7973
7974         if (!hw->phy.ops.handle_lasi)
7975                 return;
7976
7977         hw->phy.ops.handle_lasi(&adapter->hw, &overtemp);
7978         if (overtemp)
7979                 e_crit(drv, "%s\n", ixgbe_overheat_msg);
7980 }
7981
7982 static void ixgbe_reset_subtask(struct ixgbe_adapter *adapter)
7983 {
7984         if (!test_and_clear_bit(__IXGBE_RESET_REQUESTED, &adapter->state))
7985                 return;
7986
7987         rtnl_lock();
7988         /* If we're already down, removing or resetting, just bail */
7989         if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7990             test_bit(__IXGBE_REMOVING, &adapter->state) ||
7991             test_bit(__IXGBE_RESETTING, &adapter->state)) {
7992                 rtnl_unlock();
7993                 return;
7994         }
7995
7996         ixgbe_dump(adapter);
7997         netdev_err(adapter->netdev, "Reset adapter\n");
7998         adapter->tx_timeout_count++;
7999
8000         ixgbe_reinit_locked(adapter);
8001         rtnl_unlock();
8002 }
8003
8004 /**
8005  * ixgbe_check_fw_error - Check firmware for errors
8006  * @adapter: the adapter private structure
8007  *
8008  * Check firmware errors in register FWSM
8009  */
8010 static bool ixgbe_check_fw_error(struct ixgbe_adapter *adapter)
8011 {
8012         struct ixgbe_hw *hw = &adapter->hw;
8013         u32 fwsm;
8014
8015         /* read fwsm.ext_err_ind register and log errors */
8016         fwsm = IXGBE_READ_REG(hw, IXGBE_FWSM(hw));
8017
8018         if (fwsm & IXGBE_FWSM_EXT_ERR_IND_MASK ||
8019             !(fwsm & IXGBE_FWSM_FW_VAL_BIT))
8020                 e_dev_warn("Warning firmware error detected FWSM: 0x%08X\n",
8021                            fwsm);
8022
8023         if (hw->mac.ops.fw_recovery_mode && hw->mac.ops.fw_recovery_mode(hw)) {
8024                 e_dev_err("Firmware recovery mode detected. Limiting functionality. Refer to the Intel(R) Ethernet Adapters and Devices User Guide for details on firmware recovery mode.\n");
8025                 return true;
8026         }
8027
8028         return false;
8029 }
8030
8031 /**
8032  * ixgbe_service_task - manages and runs subtasks
8033  * @work: pointer to work_struct containing our data
8034  **/
8035 static void ixgbe_service_task(struct work_struct *work)
8036 {
8037         struct ixgbe_adapter *adapter = container_of(work,
8038                                                      struct ixgbe_adapter,
8039                                                      service_task);
8040         if (ixgbe_removed(adapter->hw.hw_addr)) {
8041                 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
8042                         rtnl_lock();
8043                         ixgbe_down(adapter);
8044                         rtnl_unlock();
8045                 }
8046                 ixgbe_service_event_complete(adapter);
8047                 return;
8048         }
8049         if (ixgbe_check_fw_error(adapter)) {
8050                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
8051                         unregister_netdev(adapter->netdev);
8052                 ixgbe_service_event_complete(adapter);
8053                 return;
8054         }
8055         ixgbe_reset_subtask(adapter);
8056         ixgbe_phy_interrupt_subtask(adapter);
8057         ixgbe_sfp_detection_subtask(adapter);
8058         ixgbe_sfp_link_config_subtask(adapter);
8059         ixgbe_check_overtemp_subtask(adapter);
8060         ixgbe_watchdog_subtask(adapter);
8061         ixgbe_fdir_reinit_subtask(adapter);
8062         ixgbe_check_hang_subtask(adapter);
8063
8064         if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state)) {
8065                 ixgbe_ptp_overflow_check(adapter);
8066                 if (adapter->flags & IXGBE_FLAG_RX_HWTSTAMP_IN_REGISTER)
8067                         ixgbe_ptp_rx_hang(adapter);
8068                 ixgbe_ptp_tx_hang(adapter);
8069         }
8070
8071         ixgbe_service_event_complete(adapter);
8072 }
8073
8074 static int ixgbe_tso(struct ixgbe_ring *tx_ring,
8075                      struct ixgbe_tx_buffer *first,
8076                      u8 *hdr_len,
8077                      struct ixgbe_ipsec_tx_data *itd)
8078 {
8079         u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
8080         struct sk_buff *skb = first->skb;
8081         union {
8082                 struct iphdr *v4;
8083                 struct ipv6hdr *v6;
8084                 unsigned char *hdr;
8085         } ip;
8086         union {
8087                 struct tcphdr *tcp;
8088                 struct udphdr *udp;
8089                 unsigned char *hdr;
8090         } l4;
8091         u32 paylen, l4_offset;
8092         u32 fceof_saidx = 0;
8093         int err;
8094
8095         if (skb->ip_summed != CHECKSUM_PARTIAL)
8096                 return 0;
8097
8098         if (!skb_is_gso(skb))
8099                 return 0;
8100
8101         err = skb_cow_head(skb, 0);
8102         if (err < 0)
8103                 return err;
8104
8105         if (eth_p_mpls(first->protocol))
8106                 ip.hdr = skb_inner_network_header(skb);
8107         else
8108                 ip.hdr = skb_network_header(skb);
8109         l4.hdr = skb_checksum_start(skb);
8110
8111         /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
8112         type_tucmd = (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ?
8113                       IXGBE_ADVTXD_TUCMD_L4T_UDP : IXGBE_ADVTXD_TUCMD_L4T_TCP;
8114
8115         /* initialize outer IP header fields */
8116         if (ip.v4->version == 4) {
8117                 unsigned char *csum_start = skb_checksum_start(skb);
8118                 unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
8119                 int len = csum_start - trans_start;
8120
8121                 /* IP header will have to cancel out any data that
8122                  * is not a part of the outer IP header, so set to
8123                  * a reverse csum if needed, else init check to 0.
8124                  */
8125                 ip.v4->check = (skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL) ?
8126                                            csum_fold(csum_partial(trans_start,
8127                                                                   len, 0)) : 0;
8128                 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
8129
8130                 ip.v4->tot_len = 0;
8131                 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
8132                                    IXGBE_TX_FLAGS_CSUM |
8133                                    IXGBE_TX_FLAGS_IPV4;
8134         } else {
8135                 ip.v6->payload_len = 0;
8136                 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
8137                                    IXGBE_TX_FLAGS_CSUM;
8138         }
8139
8140         /* determine offset of inner transport header */
8141         l4_offset = l4.hdr - skb->data;
8142
8143         /* remove payload length from inner checksum */
8144         paylen = skb->len - l4_offset;
8145
8146         if (type_tucmd & IXGBE_ADVTXD_TUCMD_L4T_TCP) {
8147                 /* compute length of segmentation header */
8148                 *hdr_len = (l4.tcp->doff * 4) + l4_offset;
8149                 csum_replace_by_diff(&l4.tcp->check,
8150                                      (__force __wsum)htonl(paylen));
8151         } else {
8152                 /* compute length of segmentation header */
8153                 *hdr_len = sizeof(*l4.udp) + l4_offset;
8154                 csum_replace_by_diff(&l4.udp->check,
8155                                      (__force __wsum)htonl(paylen));
8156         }
8157
8158         /* update gso size and bytecount with header size */
8159         first->gso_segs = skb_shinfo(skb)->gso_segs;
8160         first->bytecount += (first->gso_segs - 1) * *hdr_len;
8161
8162         /* mss_l4len_id: use 0 as index for TSO */
8163         mss_l4len_idx = (*hdr_len - l4_offset) << IXGBE_ADVTXD_L4LEN_SHIFT;
8164         mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
8165
8166         fceof_saidx |= itd->sa_idx;
8167         type_tucmd |= itd->flags | itd->trailer_len;
8168
8169         /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
8170         vlan_macip_lens = l4.hdr - ip.hdr;
8171         vlan_macip_lens |= (ip.hdr - skb->data) << IXGBE_ADVTXD_MACLEN_SHIFT;
8172         vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
8173
8174         ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, fceof_saidx, type_tucmd,
8175                           mss_l4len_idx);
8176
8177         return 1;
8178 }
8179
8180 static void ixgbe_tx_csum(struct ixgbe_ring *tx_ring,
8181                           struct ixgbe_tx_buffer *first,
8182                           struct ixgbe_ipsec_tx_data *itd)
8183 {
8184         struct sk_buff *skb = first->skb;
8185         u32 vlan_macip_lens = 0;
8186         u32 fceof_saidx = 0;
8187         u32 type_tucmd = 0;
8188
8189         if (skb->ip_summed != CHECKSUM_PARTIAL) {
8190 csum_failed:
8191                 if (!(first->tx_flags & (IXGBE_TX_FLAGS_HW_VLAN |
8192                                          IXGBE_TX_FLAGS_CC)))
8193                         return;
8194                 goto no_csum;
8195         }
8196
8197         switch (skb->csum_offset) {
8198         case offsetof(struct tcphdr, check):
8199                 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
8200                 fallthrough;
8201         case offsetof(struct udphdr, check):
8202                 break;
8203         case offsetof(struct sctphdr, checksum):
8204                 /* validate that this is actually an SCTP request */
8205                 if (skb_csum_is_sctp(skb)) {
8206                         type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_SCTP;
8207                         break;
8208                 }
8209                 fallthrough;
8210         default:
8211                 skb_checksum_help(skb);
8212                 goto csum_failed;
8213         }
8214
8215         /* update TX checksum flag */
8216         first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
8217         vlan_macip_lens = skb_checksum_start_offset(skb) -
8218                           skb_network_offset(skb);
8219 no_csum:
8220         /* vlan_macip_lens: MACLEN, VLAN tag */
8221         vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
8222         vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
8223
8224         fceof_saidx |= itd->sa_idx;
8225         type_tucmd |= itd->flags | itd->trailer_len;
8226
8227         ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, fceof_saidx, type_tucmd, 0);
8228 }
8229
8230 #define IXGBE_SET_FLAG(_input, _flag, _result) \
8231         ((_flag <= _result) ? \
8232          ((u32)(_input & _flag) * (_result / _flag)) : \
8233          ((u32)(_input & _flag) / (_flag / _result)))
8234
8235 static u32 ixgbe_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
8236 {
8237         /* set type for advanced descriptor with frame checksum insertion */
8238         u32 cmd_type = IXGBE_ADVTXD_DTYP_DATA |
8239                        IXGBE_ADVTXD_DCMD_DEXT |
8240                        IXGBE_ADVTXD_DCMD_IFCS;
8241
8242         /* set HW vlan bit if vlan is present */
8243         cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_HW_VLAN,
8244                                    IXGBE_ADVTXD_DCMD_VLE);
8245
8246         /* set segmentation enable bits for TSO/FSO */
8247         cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSO,
8248                                    IXGBE_ADVTXD_DCMD_TSE);
8249
8250         /* set timestamp bit if present */
8251         cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSTAMP,
8252                                    IXGBE_ADVTXD_MAC_TSTAMP);
8253
8254         /* insert frame checksum */
8255         cmd_type ^= IXGBE_SET_FLAG(skb->no_fcs, 1, IXGBE_ADVTXD_DCMD_IFCS);
8256
8257         return cmd_type;
8258 }
8259
8260 static void ixgbe_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
8261                                    u32 tx_flags, unsigned int paylen)
8262 {
8263         u32 olinfo_status = paylen << IXGBE_ADVTXD_PAYLEN_SHIFT;
8264
8265         /* enable L4 checksum for TSO and TX checksum offload */
8266         olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8267                                         IXGBE_TX_FLAGS_CSUM,
8268                                         IXGBE_ADVTXD_POPTS_TXSM);
8269
8270         /* enable IPv4 checksum for TSO */
8271         olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8272                                         IXGBE_TX_FLAGS_IPV4,
8273                                         IXGBE_ADVTXD_POPTS_IXSM);
8274
8275         /* enable IPsec */
8276         olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8277                                         IXGBE_TX_FLAGS_IPSEC,
8278                                         IXGBE_ADVTXD_POPTS_IPSEC);
8279
8280         /*
8281          * Check Context must be set if Tx switch is enabled, which it
8282          * always is for case where virtual functions are running
8283          */
8284         olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8285                                         IXGBE_TX_FLAGS_CC,
8286                                         IXGBE_ADVTXD_CC);
8287
8288         tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
8289 }
8290
8291 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
8292 {
8293         if (!netif_subqueue_try_stop(tx_ring->netdev, tx_ring->queue_index,
8294                                      ixgbe_desc_unused(tx_ring), size))
8295                 return -EBUSY;
8296
8297         ++tx_ring->tx_stats.restart_queue;
8298         return 0;
8299 }
8300
8301 static inline int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
8302 {
8303         if (likely(ixgbe_desc_unused(tx_ring) >= size))
8304                 return 0;
8305
8306         return __ixgbe_maybe_stop_tx(tx_ring, size);
8307 }
8308
8309 static int ixgbe_tx_map(struct ixgbe_ring *tx_ring,
8310                         struct ixgbe_tx_buffer *first,
8311                         const u8 hdr_len)
8312 {
8313         struct sk_buff *skb = first->skb;
8314         struct ixgbe_tx_buffer *tx_buffer;
8315         union ixgbe_adv_tx_desc *tx_desc;
8316         skb_frag_t *frag;
8317         dma_addr_t dma;
8318         unsigned int data_len, size;
8319         u32 tx_flags = first->tx_flags;
8320         u32 cmd_type = ixgbe_tx_cmd_type(skb, tx_flags);
8321         u16 i = tx_ring->next_to_use;
8322
8323         tx_desc = IXGBE_TX_DESC(tx_ring, i);
8324
8325         ixgbe_tx_olinfo_status(tx_desc, tx_flags, skb->len - hdr_len);
8326
8327         size = skb_headlen(skb);
8328         data_len = skb->data_len;
8329
8330 #ifdef IXGBE_FCOE
8331         if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
8332                 if (data_len < sizeof(struct fcoe_crc_eof)) {
8333                         size -= sizeof(struct fcoe_crc_eof) - data_len;
8334                         data_len = 0;
8335                 } else {
8336                         data_len -= sizeof(struct fcoe_crc_eof);
8337                 }
8338         }
8339
8340 #endif
8341         dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
8342
8343         tx_buffer = first;
8344
8345         for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
8346                 if (dma_mapping_error(tx_ring->dev, dma))
8347                         goto dma_error;
8348
8349                 /* record length, and DMA address */
8350                 dma_unmap_len_set(tx_buffer, len, size);
8351                 dma_unmap_addr_set(tx_buffer, dma, dma);
8352
8353                 tx_desc->read.buffer_addr = cpu_to_le64(dma);
8354
8355                 while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
8356                         tx_desc->read.cmd_type_len =
8357                                 cpu_to_le32(cmd_type ^ IXGBE_MAX_DATA_PER_TXD);
8358
8359                         i++;
8360                         tx_desc++;
8361                         if (i == tx_ring->count) {
8362                                 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
8363                                 i = 0;
8364                         }
8365                         tx_desc->read.olinfo_status = 0;
8366
8367                         dma += IXGBE_MAX_DATA_PER_TXD;
8368                         size -= IXGBE_MAX_DATA_PER_TXD;
8369
8370                         tx_desc->read.buffer_addr = cpu_to_le64(dma);
8371                 }
8372
8373                 if (likely(!data_len))
8374                         break;
8375
8376                 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
8377
8378                 i++;
8379                 tx_desc++;
8380                 if (i == tx_ring->count) {
8381                         tx_desc = IXGBE_TX_DESC(tx_ring, 0);
8382                         i = 0;
8383                 }
8384                 tx_desc->read.olinfo_status = 0;
8385
8386 #ifdef IXGBE_FCOE
8387                 size = min_t(unsigned int, data_len, skb_frag_size(frag));
8388 #else
8389                 size = skb_frag_size(frag);
8390 #endif
8391                 data_len -= size;
8392
8393                 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
8394                                        DMA_TO_DEVICE);
8395
8396                 tx_buffer = &tx_ring->tx_buffer_info[i];
8397         }
8398
8399         /* write last descriptor with RS and EOP bits */
8400         cmd_type |= size | IXGBE_TXD_CMD;
8401         tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
8402
8403         netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
8404
8405         /* set the timestamp */
8406         first->time_stamp = jiffies;
8407
8408         skb_tx_timestamp(skb);
8409
8410         /*
8411          * Force memory writes to complete before letting h/w know there
8412          * are new descriptors to fetch.  (Only applicable for weak-ordered
8413          * memory model archs, such as IA-64).
8414          *
8415          * We also need this memory barrier to make certain all of the
8416          * status bits have been updated before next_to_watch is written.
8417          */
8418         wmb();
8419
8420         /* set next_to_watch value indicating a packet is present */
8421         first->next_to_watch = tx_desc;
8422
8423         i++;
8424         if (i == tx_ring->count)
8425                 i = 0;
8426
8427         tx_ring->next_to_use = i;
8428
8429         ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
8430
8431         if (netif_xmit_stopped(txring_txq(tx_ring)) || !netdev_xmit_more()) {
8432                 writel(i, tx_ring->tail);
8433         }
8434
8435         return 0;
8436 dma_error:
8437         dev_err(tx_ring->dev, "TX DMA map failed\n");
8438
8439         /* clear dma mappings for failed tx_buffer_info map */
8440         for (;;) {
8441                 tx_buffer = &tx_ring->tx_buffer_info[i];
8442                 if (dma_unmap_len(tx_buffer, len))
8443                         dma_unmap_page(tx_ring->dev,
8444                                        dma_unmap_addr(tx_buffer, dma),
8445                                        dma_unmap_len(tx_buffer, len),
8446                                        DMA_TO_DEVICE);
8447                 dma_unmap_len_set(tx_buffer, len, 0);
8448                 if (tx_buffer == first)
8449                         break;
8450                 if (i == 0)
8451                         i += tx_ring->count;
8452                 i--;
8453         }
8454
8455         dev_kfree_skb_any(first->skb);
8456         first->skb = NULL;
8457
8458         tx_ring->next_to_use = i;
8459
8460         return -1;
8461 }
8462
8463 static void ixgbe_atr(struct ixgbe_ring *ring,
8464                       struct ixgbe_tx_buffer *first)
8465 {
8466         struct ixgbe_q_vector *q_vector = ring->q_vector;
8467         union ixgbe_atr_hash_dword input = { .dword = 0 };
8468         union ixgbe_atr_hash_dword common = { .dword = 0 };
8469         union {
8470                 unsigned char *network;
8471                 struct iphdr *ipv4;
8472                 struct ipv6hdr *ipv6;
8473         } hdr;
8474         struct tcphdr *th;
8475         unsigned int hlen;
8476         struct sk_buff *skb;
8477         __be16 vlan_id;
8478         int l4_proto;
8479
8480         /* if ring doesn't have a interrupt vector, cannot perform ATR */
8481         if (!q_vector)
8482                 return;
8483
8484         /* do nothing if sampling is disabled */
8485         if (!ring->atr_sample_rate)
8486                 return;
8487
8488         ring->atr_count++;
8489
8490         /* currently only IPv4/IPv6 with TCP is supported */
8491         if ((first->protocol != htons(ETH_P_IP)) &&
8492             (first->protocol != htons(ETH_P_IPV6)))
8493                 return;
8494
8495         /* snag network header to get L4 type and address */
8496         skb = first->skb;
8497         hdr.network = skb_network_header(skb);
8498         if (unlikely(hdr.network <= skb->data))
8499                 return;
8500         if (skb->encapsulation &&
8501             first->protocol == htons(ETH_P_IP) &&
8502             hdr.ipv4->protocol == IPPROTO_UDP) {
8503                 struct ixgbe_adapter *adapter = q_vector->adapter;
8504
8505                 if (unlikely(skb_tail_pointer(skb) < hdr.network +
8506                              vxlan_headroom(0)))
8507                         return;
8508
8509                 /* verify the port is recognized as VXLAN */
8510                 if (adapter->vxlan_port &&
8511                     udp_hdr(skb)->dest == adapter->vxlan_port)
8512                         hdr.network = skb_inner_network_header(skb);
8513
8514                 if (adapter->geneve_port &&
8515                     udp_hdr(skb)->dest == adapter->geneve_port)
8516                         hdr.network = skb_inner_network_header(skb);
8517         }
8518
8519         /* Make sure we have at least [minimum IPv4 header + TCP]
8520          * or [IPv6 header] bytes
8521          */
8522         if (unlikely(skb_tail_pointer(skb) < hdr.network + 40))
8523                 return;
8524
8525         /* Currently only IPv4/IPv6 with TCP is supported */
8526         switch (hdr.ipv4->version) {
8527         case IPVERSION:
8528                 /* access ihl as u8 to avoid unaligned access on ia64 */
8529                 hlen = (hdr.network[0] & 0x0F) << 2;
8530                 l4_proto = hdr.ipv4->protocol;
8531                 break;
8532         case 6:
8533                 hlen = hdr.network - skb->data;
8534                 l4_proto = ipv6_find_hdr(skb, &hlen, IPPROTO_TCP, NULL, NULL);
8535                 hlen -= hdr.network - skb->data;
8536                 break;
8537         default:
8538                 return;
8539         }
8540
8541         if (l4_proto != IPPROTO_TCP)
8542                 return;
8543
8544         if (unlikely(skb_tail_pointer(skb) < hdr.network +
8545                      hlen + sizeof(struct tcphdr)))
8546                 return;
8547
8548         th = (struct tcphdr *)(hdr.network + hlen);
8549
8550         /* skip this packet since the socket is closing */
8551         if (th->fin)
8552                 return;
8553
8554         /* sample on all syn packets or once every atr sample count */
8555         if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
8556                 return;
8557
8558         /* reset sample count */
8559         ring->atr_count = 0;
8560
8561         vlan_id = htons(first->tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
8562
8563         /*
8564          * src and dst are inverted, think how the receiver sees them
8565          *
8566          * The input is broken into two sections, a non-compressed section
8567          * containing vm_pool, vlan_id, and flow_type.  The rest of the data
8568          * is XORed together and stored in the compressed dword.
8569          */
8570         input.formatted.vlan_id = vlan_id;
8571
8572         /*
8573          * since src port and flex bytes occupy the same word XOR them together
8574          * and write the value to source port portion of compressed dword
8575          */
8576         if (first->tx_flags & (IXGBE_TX_FLAGS_SW_VLAN | IXGBE_TX_FLAGS_HW_VLAN))
8577                 common.port.src ^= th->dest ^ htons(ETH_P_8021Q);
8578         else
8579                 common.port.src ^= th->dest ^ first->protocol;
8580         common.port.dst ^= th->source;
8581
8582         switch (hdr.ipv4->version) {
8583         case IPVERSION:
8584                 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
8585                 common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
8586                 break;
8587         case 6:
8588                 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
8589                 common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
8590                              hdr.ipv6->saddr.s6_addr32[1] ^
8591                              hdr.ipv6->saddr.s6_addr32[2] ^
8592                              hdr.ipv6->saddr.s6_addr32[3] ^
8593                              hdr.ipv6->daddr.s6_addr32[0] ^
8594                              hdr.ipv6->daddr.s6_addr32[1] ^
8595                              hdr.ipv6->daddr.s6_addr32[2] ^
8596                              hdr.ipv6->daddr.s6_addr32[3];
8597                 break;
8598         default:
8599                 break;
8600         }
8601
8602         if (hdr.network != skb_network_header(skb))
8603                 input.formatted.flow_type |= IXGBE_ATR_L4TYPE_TUNNEL_MASK;
8604
8605         /* This assumes the Rx queue and Tx queue are bound to the same CPU */
8606         ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
8607                                               input, common, ring->queue_index);
8608 }
8609
8610 #ifdef IXGBE_FCOE
8611 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb,
8612                               struct net_device *sb_dev)
8613 {
8614         struct ixgbe_adapter *adapter;
8615         struct ixgbe_ring_feature *f;
8616         int txq;
8617
8618         if (sb_dev) {
8619                 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
8620                 struct net_device *vdev = sb_dev;
8621
8622                 txq = vdev->tc_to_txq[tc].offset;
8623                 txq += reciprocal_scale(skb_get_hash(skb),
8624                                         vdev->tc_to_txq[tc].count);
8625
8626                 return txq;
8627         }
8628
8629         /*
8630          * only execute the code below if protocol is FCoE
8631          * or FIP and we have FCoE enabled on the adapter
8632          */
8633         switch (vlan_get_protocol(skb)) {
8634         case htons(ETH_P_FCOE):
8635         case htons(ETH_P_FIP):
8636                 adapter = netdev_priv(dev);
8637
8638                 if (!sb_dev && (adapter->flags & IXGBE_FLAG_FCOE_ENABLED))
8639                         break;
8640                 fallthrough;
8641         default:
8642                 return netdev_pick_tx(dev, skb, sb_dev);
8643         }
8644
8645         f = &adapter->ring_feature[RING_F_FCOE];
8646
8647         txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) :
8648                                            smp_processor_id();
8649
8650         while (txq >= f->indices)
8651                 txq -= f->indices;
8652
8653         return txq + f->offset;
8654 }
8655
8656 #endif
8657 int ixgbe_xmit_xdp_ring(struct ixgbe_ring *ring,
8658                         struct xdp_frame *xdpf)
8659 {
8660         struct skb_shared_info *sinfo = xdp_get_shared_info_from_frame(xdpf);
8661         u8 nr_frags = unlikely(xdp_frame_has_frags(xdpf)) ? sinfo->nr_frags : 0;
8662         u16 i = 0, index = ring->next_to_use;
8663         struct ixgbe_tx_buffer *tx_head = &ring->tx_buffer_info[index];
8664         struct ixgbe_tx_buffer *tx_buff = tx_head;
8665         union ixgbe_adv_tx_desc *tx_desc = IXGBE_TX_DESC(ring, index);
8666         u32 cmd_type, len = xdpf->len;
8667         void *data = xdpf->data;
8668
8669         if (unlikely(ixgbe_desc_unused(ring) < 1 + nr_frags))
8670                 return IXGBE_XDP_CONSUMED;
8671
8672         tx_head->bytecount = xdp_get_frame_len(xdpf);
8673         tx_head->gso_segs = 1;
8674         tx_head->xdpf = xdpf;
8675
8676         tx_desc->read.olinfo_status =
8677                 cpu_to_le32(tx_head->bytecount << IXGBE_ADVTXD_PAYLEN_SHIFT);
8678
8679         for (;;) {
8680                 dma_addr_t dma;
8681
8682                 dma = dma_map_single(ring->dev, data, len, DMA_TO_DEVICE);
8683                 if (dma_mapping_error(ring->dev, dma))
8684                         goto unmap;
8685
8686                 dma_unmap_len_set(tx_buff, len, len);
8687                 dma_unmap_addr_set(tx_buff, dma, dma);
8688
8689                 cmd_type = IXGBE_ADVTXD_DTYP_DATA | IXGBE_ADVTXD_DCMD_DEXT |
8690                            IXGBE_ADVTXD_DCMD_IFCS | len;
8691                 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
8692                 tx_desc->read.buffer_addr = cpu_to_le64(dma);
8693                 tx_buff->protocol = 0;
8694
8695                 if (++index == ring->count)
8696                         index = 0;
8697
8698                 if (i == nr_frags)
8699                         break;
8700
8701                 tx_buff = &ring->tx_buffer_info[index];
8702                 tx_desc = IXGBE_TX_DESC(ring, index);
8703                 tx_desc->read.olinfo_status = 0;
8704
8705                 data = skb_frag_address(&sinfo->frags[i]);
8706                 len = skb_frag_size(&sinfo->frags[i]);
8707                 i++;
8708         }
8709         /* put descriptor type bits */
8710         tx_desc->read.cmd_type_len |= cpu_to_le32(IXGBE_TXD_CMD);
8711
8712         /* Avoid any potential race with xdp_xmit and cleanup */
8713         smp_wmb();
8714
8715         tx_head->next_to_watch = tx_desc;
8716         ring->next_to_use = index;
8717
8718         return IXGBE_XDP_TX;
8719
8720 unmap:
8721         for (;;) {
8722                 tx_buff = &ring->tx_buffer_info[index];
8723                 if (dma_unmap_len(tx_buff, len))
8724                         dma_unmap_page(ring->dev, dma_unmap_addr(tx_buff, dma),
8725                                        dma_unmap_len(tx_buff, len),
8726                                        DMA_TO_DEVICE);
8727                 dma_unmap_len_set(tx_buff, len, 0);
8728                 if (tx_buff == tx_head)
8729                         break;
8730
8731                 if (!index)
8732                         index += ring->count;
8733                 index--;
8734         }
8735
8736         return IXGBE_XDP_CONSUMED;
8737 }
8738
8739 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
8740                           struct ixgbe_adapter *adapter,
8741                           struct ixgbe_ring *tx_ring)
8742 {
8743         struct ixgbe_tx_buffer *first;
8744         int tso;
8745         u32 tx_flags = 0;
8746         unsigned short f;
8747         u16 count = TXD_USE_COUNT(skb_headlen(skb));
8748         struct ixgbe_ipsec_tx_data ipsec_tx = { 0 };
8749         __be16 protocol = skb->protocol;
8750         u8 hdr_len = 0;
8751
8752         /*
8753          * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
8754          *       + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
8755          *       + 2 desc gap to keep tail from touching head,
8756          *       + 1 desc for context descriptor,
8757          * otherwise try next time
8758          */
8759         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
8760                 count += TXD_USE_COUNT(skb_frag_size(
8761                                                 &skb_shinfo(skb)->frags[f]));
8762
8763         if (ixgbe_maybe_stop_tx(tx_ring, count + 3)) {
8764                 tx_ring->tx_stats.tx_busy++;
8765                 return NETDEV_TX_BUSY;
8766         }
8767
8768         /* record the location of the first descriptor for this packet */
8769         first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
8770         first->skb = skb;
8771         first->bytecount = skb->len;
8772         first->gso_segs = 1;
8773
8774         /* if we have a HW VLAN tag being added default to the HW one */
8775         if (skb_vlan_tag_present(skb)) {
8776                 tx_flags |= skb_vlan_tag_get(skb) << IXGBE_TX_FLAGS_VLAN_SHIFT;
8777                 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
8778         /* else if it is a SW VLAN check the next protocol and store the tag */
8779         } else if (protocol == htons(ETH_P_8021Q)) {
8780                 struct vlan_hdr *vhdr, _vhdr;
8781                 vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
8782                 if (!vhdr)
8783                         goto out_drop;
8784
8785                 tx_flags |= ntohs(vhdr->h_vlan_TCI) <<
8786                                   IXGBE_TX_FLAGS_VLAN_SHIFT;
8787                 tx_flags |= IXGBE_TX_FLAGS_SW_VLAN;
8788         }
8789         protocol = vlan_get_protocol(skb);
8790
8791         if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) &&
8792             adapter->ptp_clock) {
8793                 if (adapter->tstamp_config.tx_type == HWTSTAMP_TX_ON &&
8794                     !test_and_set_bit_lock(__IXGBE_PTP_TX_IN_PROGRESS,
8795                                            &adapter->state)) {
8796                         skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
8797                         tx_flags |= IXGBE_TX_FLAGS_TSTAMP;
8798
8799                         /* schedule check for Tx timestamp */
8800                         adapter->ptp_tx_skb = skb_get(skb);
8801                         adapter->ptp_tx_start = jiffies;
8802                         schedule_work(&adapter->ptp_tx_work);
8803                 } else {
8804                         adapter->tx_hwtstamp_skipped++;
8805                 }
8806         }
8807
8808 #ifdef CONFIG_PCI_IOV
8809         /*
8810          * Use the l2switch_enable flag - would be false if the DMA
8811          * Tx switch had been disabled.
8812          */
8813         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
8814                 tx_flags |= IXGBE_TX_FLAGS_CC;
8815
8816 #endif
8817         /* DCB maps skb priorities 0-7 onto 3 bit PCP of VLAN tag. */
8818         if ((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
8819             ((tx_flags & (IXGBE_TX_FLAGS_HW_VLAN | IXGBE_TX_FLAGS_SW_VLAN)) ||
8820              (skb->priority != TC_PRIO_CONTROL))) {
8821                 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
8822                 tx_flags |= (skb->priority & 0x7) <<
8823                                         IXGBE_TX_FLAGS_VLAN_PRIO_SHIFT;
8824                 if (tx_flags & IXGBE_TX_FLAGS_SW_VLAN) {
8825                         struct vlan_ethhdr *vhdr;
8826
8827                         if (skb_cow_head(skb, 0))
8828                                 goto out_drop;
8829                         vhdr = skb_vlan_eth_hdr(skb);
8830                         vhdr->h_vlan_TCI = htons(tx_flags >>
8831                                                  IXGBE_TX_FLAGS_VLAN_SHIFT);
8832                 } else {
8833                         tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
8834                 }
8835         }
8836
8837         /* record initial flags and protocol */
8838         first->tx_flags = tx_flags;
8839         first->protocol = protocol;
8840
8841 #ifdef IXGBE_FCOE
8842         /* setup tx offload for FCoE */
8843         if ((protocol == htons(ETH_P_FCOE)) &&
8844             (tx_ring->netdev->features & (NETIF_F_FSO | NETIF_F_FCOE_CRC))) {
8845                 tso = ixgbe_fso(tx_ring, first, &hdr_len);
8846                 if (tso < 0)
8847                         goto out_drop;
8848
8849                 goto xmit_fcoe;
8850         }
8851
8852 #endif /* IXGBE_FCOE */
8853
8854 #ifdef CONFIG_IXGBE_IPSEC
8855         if (xfrm_offload(skb) &&
8856             !ixgbe_ipsec_tx(tx_ring, first, &ipsec_tx))
8857                 goto out_drop;
8858 #endif
8859         tso = ixgbe_tso(tx_ring, first, &hdr_len, &ipsec_tx);
8860         if (tso < 0)
8861                 goto out_drop;
8862         else if (!tso)
8863                 ixgbe_tx_csum(tx_ring, first, &ipsec_tx);
8864
8865         /* add the ATR filter if ATR is on */
8866         if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
8867                 ixgbe_atr(tx_ring, first);
8868
8869 #ifdef IXGBE_FCOE
8870 xmit_fcoe:
8871 #endif /* IXGBE_FCOE */
8872         if (ixgbe_tx_map(tx_ring, first, hdr_len))
8873                 goto cleanup_tx_timestamp;
8874
8875         return NETDEV_TX_OK;
8876
8877 out_drop:
8878         dev_kfree_skb_any(first->skb);
8879         first->skb = NULL;
8880 cleanup_tx_timestamp:
8881         if (unlikely(tx_flags & IXGBE_TX_FLAGS_TSTAMP)) {
8882                 dev_kfree_skb_any(adapter->ptp_tx_skb);
8883                 adapter->ptp_tx_skb = NULL;
8884                 cancel_work_sync(&adapter->ptp_tx_work);
8885                 clear_bit_unlock(__IXGBE_PTP_TX_IN_PROGRESS, &adapter->state);
8886         }
8887
8888         return NETDEV_TX_OK;
8889 }
8890
8891 static netdev_tx_t __ixgbe_xmit_frame(struct sk_buff *skb,
8892                                       struct net_device *netdev,
8893                                       struct ixgbe_ring *ring)
8894 {
8895         struct ixgbe_adapter *adapter = netdev_priv(netdev);
8896         struct ixgbe_ring *tx_ring;
8897
8898         /*
8899          * The minimum packet size for olinfo paylen is 17 so pad the skb
8900          * in order to meet this minimum size requirement.
8901          */
8902         if (skb_put_padto(skb, 17))
8903                 return NETDEV_TX_OK;
8904
8905         tx_ring = ring ? ring : adapter->tx_ring[skb_get_queue_mapping(skb)];
8906         if (unlikely(test_bit(__IXGBE_TX_DISABLED, &tx_ring->state)))
8907                 return NETDEV_TX_BUSY;
8908
8909         return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
8910 }
8911
8912 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
8913                                     struct net_device *netdev)
8914 {
8915         return __ixgbe_xmit_frame(skb, netdev, NULL);
8916 }
8917
8918 /**
8919  * ixgbe_set_mac - Change the Ethernet Address of the NIC
8920  * @netdev: network interface device structure
8921  * @p: pointer to an address structure
8922  *
8923  * Returns 0 on success, negative on failure
8924  **/
8925 static int ixgbe_set_mac(struct net_device *netdev, void *p)
8926 {
8927         struct ixgbe_adapter *adapter = netdev_priv(netdev);
8928         struct ixgbe_hw *hw = &adapter->hw;
8929         struct sockaddr *addr = p;
8930
8931         if (!is_valid_ether_addr(addr->sa_data))
8932                 return -EADDRNOTAVAIL;
8933
8934         eth_hw_addr_set(netdev, addr->sa_data);
8935         memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
8936
8937         ixgbe_mac_set_default_filter(adapter);
8938
8939         return 0;
8940 }
8941
8942 static int
8943 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
8944 {
8945         struct ixgbe_adapter *adapter = netdev_priv(netdev);
8946         struct ixgbe_hw *hw = &adapter->hw;
8947         u16 value;
8948         int rc;
8949
8950         if (adapter->mii_bus) {
8951                 int regnum = addr;
8952
8953                 if (devad != MDIO_DEVAD_NONE)
8954                         return mdiobus_c45_read(adapter->mii_bus, prtad,
8955                                                 devad, regnum);
8956
8957                 return mdiobus_read(adapter->mii_bus, prtad, regnum);
8958         }
8959
8960         if (prtad != hw->phy.mdio.prtad)
8961                 return -EINVAL;
8962         rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
8963         if (!rc)
8964                 rc = value;
8965         return rc;
8966 }
8967
8968 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
8969                             u16 addr, u16 value)
8970 {
8971         struct ixgbe_adapter *adapter = netdev_priv(netdev);
8972         struct ixgbe_hw *hw = &adapter->hw;
8973
8974         if (adapter->mii_bus) {
8975                 int regnum = addr;
8976
8977                 if (devad != MDIO_DEVAD_NONE)
8978                         return mdiobus_c45_write(adapter->mii_bus, prtad, devad,
8979                                                  regnum, value);
8980
8981                 return mdiobus_write(adapter->mii_bus, prtad, regnum, value);
8982         }
8983
8984         if (prtad != hw->phy.mdio.prtad)
8985                 return -EINVAL;
8986         return hw->phy.ops.write_reg(hw, addr, devad, value);
8987 }
8988
8989 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
8990 {
8991         struct ixgbe_adapter *adapter = netdev_priv(netdev);
8992
8993         switch (cmd) {
8994         case SIOCSHWTSTAMP:
8995                 return ixgbe_ptp_set_ts_config(adapter, req);
8996         case SIOCGHWTSTAMP:
8997                 return ixgbe_ptp_get_ts_config(adapter, req);
8998         case SIOCGMIIPHY:
8999                 if (!adapter->hw.phy.ops.read_reg)
9000                         return -EOPNOTSUPP;
9001                 fallthrough;
9002         default:
9003                 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
9004         }
9005 }
9006
9007 /**
9008  * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
9009  * netdev->dev_addrs
9010  * @dev: network interface device structure
9011  *
9012  * Returns non-zero on failure
9013  **/
9014 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
9015 {
9016         int err = 0;
9017         struct ixgbe_adapter *adapter = netdev_priv(dev);
9018         struct ixgbe_hw *hw = &adapter->hw;
9019
9020         if (is_valid_ether_addr(hw->mac.san_addr)) {
9021                 rtnl_lock();
9022                 err = dev_addr_add(dev, hw->mac.san_addr, NETDEV_HW_ADDR_T_SAN);
9023                 rtnl_unlock();
9024
9025                 /* update SAN MAC vmdq pool selection */
9026                 hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
9027         }
9028         return err;
9029 }
9030
9031 /**
9032  * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
9033  * netdev->dev_addrs
9034  * @dev: network interface device structure
9035  *
9036  * Returns non-zero on failure
9037  **/
9038 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
9039 {
9040         int err = 0;
9041         struct ixgbe_adapter *adapter = netdev_priv(dev);
9042         struct ixgbe_mac_info *mac = &adapter->hw.mac;
9043
9044         if (is_valid_ether_addr(mac->san_addr)) {
9045                 rtnl_lock();
9046                 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
9047                 rtnl_unlock();
9048         }
9049         return err;
9050 }
9051
9052 static void ixgbe_get_ring_stats64(struct rtnl_link_stats64 *stats,
9053                                    struct ixgbe_ring *ring)
9054 {
9055         u64 bytes, packets;
9056         unsigned int start;
9057
9058         if (ring) {
9059                 do {
9060                         start = u64_stats_fetch_begin(&ring->syncp);
9061                         packets = ring->stats.packets;
9062                         bytes   = ring->stats.bytes;
9063                 } while (u64_stats_fetch_retry(&ring->syncp, start));
9064                 stats->tx_packets += packets;
9065                 stats->tx_bytes   += bytes;
9066         }
9067 }
9068
9069 static void ixgbe_get_stats64(struct net_device *netdev,
9070                               struct rtnl_link_stats64 *stats)
9071 {
9072         struct ixgbe_adapter *adapter = netdev_priv(netdev);
9073         int i;
9074
9075         rcu_read_lock();
9076         for (i = 0; i < adapter->num_rx_queues; i++) {
9077                 struct ixgbe_ring *ring = READ_ONCE(adapter->rx_ring[i]);
9078                 u64 bytes, packets;
9079                 unsigned int start;
9080
9081                 if (ring) {
9082                         do {
9083                                 start = u64_stats_fetch_begin(&ring->syncp);
9084                                 packets = ring->stats.packets;
9085                                 bytes   = ring->stats.bytes;
9086                         } while (u64_stats_fetch_retry(&ring->syncp, start));
9087                         stats->rx_packets += packets;
9088                         stats->rx_bytes   += bytes;
9089                 }
9090         }
9091
9092         for (i = 0; i < adapter->num_tx_queues; i++) {
9093                 struct ixgbe_ring *ring = READ_ONCE(adapter->tx_ring[i]);
9094
9095                 ixgbe_get_ring_stats64(stats, ring);
9096         }
9097         for (i = 0; i < adapter->num_xdp_queues; i++) {
9098                 struct ixgbe_ring *ring = READ_ONCE(adapter->xdp_ring[i]);
9099
9100                 ixgbe_get_ring_stats64(stats, ring);
9101         }
9102         rcu_read_unlock();
9103
9104         /* following stats updated by ixgbe_watchdog_task() */
9105         stats->multicast        = netdev->stats.multicast;
9106         stats->rx_errors        = netdev->stats.rx_errors;
9107         stats->rx_length_errors = netdev->stats.rx_length_errors;
9108         stats->rx_crc_errors    = netdev->stats.rx_crc_errors;
9109         stats->rx_missed_errors = netdev->stats.rx_missed_errors;
9110 }
9111
9112 static int ixgbe_ndo_get_vf_stats(struct net_device *netdev, int vf,
9113                                   struct ifla_vf_stats *vf_stats)
9114 {
9115         struct ixgbe_adapter *adapter = netdev_priv(netdev);
9116
9117         if (vf < 0 || vf >= adapter->num_vfs)
9118                 return -EINVAL;
9119
9120         vf_stats->rx_packets = adapter->vfinfo[vf].vfstats.gprc;
9121         vf_stats->rx_bytes   = adapter->vfinfo[vf].vfstats.gorc;
9122         vf_stats->tx_packets = adapter->vfinfo[vf].vfstats.gptc;
9123         vf_stats->tx_bytes   = adapter->vfinfo[vf].vfstats.gotc;
9124         vf_stats->multicast  = adapter->vfinfo[vf].vfstats.mprc;
9125
9126         return 0;
9127 }
9128
9129 #ifdef CONFIG_IXGBE_DCB
9130 /**
9131  * ixgbe_validate_rtr - verify 802.1Qp to Rx packet buffer mapping is valid.
9132  * @adapter: pointer to ixgbe_adapter
9133  * @tc: number of traffic classes currently enabled
9134  *
9135  * Configure a valid 802.1Qp to Rx packet buffer mapping ie confirm
9136  * 802.1Q priority maps to a packet buffer that exists.
9137  */
9138 static void ixgbe_validate_rtr(struct ixgbe_adapter *adapter, u8 tc)
9139 {
9140         struct ixgbe_hw *hw = &adapter->hw;
9141         u32 reg, rsave;
9142         int i;
9143
9144         /* 82598 have a static priority to TC mapping that can not
9145          * be changed so no validation is needed.
9146          */
9147         if (hw->mac.type == ixgbe_mac_82598EB)
9148                 return;
9149
9150         reg = IXGBE_READ_REG(hw, IXGBE_RTRUP2TC);
9151         rsave = reg;
9152
9153         for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
9154                 u8 up2tc = reg >> (i * IXGBE_RTRUP2TC_UP_SHIFT);
9155
9156                 /* If up2tc is out of bounds default to zero */
9157                 if (up2tc > tc)
9158                         reg &= ~(0x7 << IXGBE_RTRUP2TC_UP_SHIFT);
9159         }
9160
9161         if (reg != rsave)
9162                 IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, reg);
9163
9164         return;
9165 }
9166
9167 /**
9168  * ixgbe_set_prio_tc_map - Configure netdev prio tc map
9169  * @adapter: Pointer to adapter struct
9170  *
9171  * Populate the netdev user priority to tc map
9172  */
9173 static void ixgbe_set_prio_tc_map(struct ixgbe_adapter *adapter)
9174 {
9175         struct net_device *dev = adapter->netdev;
9176         struct ixgbe_dcb_config *dcb_cfg = &adapter->dcb_cfg;
9177         struct ieee_ets *ets = adapter->ixgbe_ieee_ets;
9178         u8 prio;
9179
9180         for (prio = 0; prio < MAX_USER_PRIORITY; prio++) {
9181                 u8 tc = 0;
9182
9183                 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE)
9184                         tc = ixgbe_dcb_get_tc_from_up(dcb_cfg, 0, prio);
9185                 else if (ets)
9186                         tc = ets->prio_tc[prio];
9187
9188                 netdev_set_prio_tc_map(dev, prio, tc);
9189         }
9190 }
9191
9192 #endif /* CONFIG_IXGBE_DCB */
9193 static int ixgbe_reassign_macvlan_pool(struct net_device *vdev,
9194                                        struct netdev_nested_priv *priv)
9195 {
9196         struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)priv->data;
9197         struct ixgbe_fwd_adapter *accel;
9198         int pool;
9199
9200         /* we only care about macvlans... */
9201         if (!netif_is_macvlan(vdev))
9202                 return 0;
9203
9204         /* that have hardware offload enabled... */
9205         accel = macvlan_accel_priv(vdev);
9206         if (!accel)
9207                 return 0;
9208
9209         /* If we can relocate to a different bit do so */
9210         pool = find_first_zero_bit(adapter->fwd_bitmask, adapter->num_rx_pools);
9211         if (pool < adapter->num_rx_pools) {
9212                 set_bit(pool, adapter->fwd_bitmask);
9213                 accel->pool = pool;
9214                 return 0;
9215         }
9216
9217         /* if we cannot find a free pool then disable the offload */
9218         netdev_err(vdev, "L2FW offload disabled due to lack of queue resources\n");
9219         macvlan_release_l2fw_offload(vdev);
9220
9221         /* unbind the queues and drop the subordinate channel config */
9222         netdev_unbind_sb_channel(adapter->netdev, vdev);
9223         netdev_set_sb_channel(vdev, 0);
9224
9225         kfree(accel);
9226
9227         return 0;
9228 }
9229
9230 static void ixgbe_defrag_macvlan_pools(struct net_device *dev)
9231 {
9232         struct ixgbe_adapter *adapter = netdev_priv(dev);
9233         struct netdev_nested_priv priv = {
9234                 .data = (void *)adapter,
9235         };
9236
9237         /* flush any stale bits out of the fwd bitmask */
9238         bitmap_clear(adapter->fwd_bitmask, 1, 63);
9239
9240         /* walk through upper devices reassigning pools */
9241         netdev_walk_all_upper_dev_rcu(dev, ixgbe_reassign_macvlan_pool,
9242                                       &priv);
9243 }
9244
9245 /**
9246  * ixgbe_setup_tc - configure net_device for multiple traffic classes
9247  *
9248  * @dev: net device to configure
9249  * @tc: number of traffic classes to enable
9250  */
9251 int ixgbe_setup_tc(struct net_device *dev, u8 tc)
9252 {
9253         struct ixgbe_adapter *adapter = netdev_priv(dev);
9254         struct ixgbe_hw *hw = &adapter->hw;
9255
9256         /* Hardware supports up to 8 traffic classes */
9257         if (tc > adapter->dcb_cfg.num_tcs.pg_tcs)
9258                 return -EINVAL;
9259
9260         if (hw->mac.type == ixgbe_mac_82598EB && tc && tc < MAX_TRAFFIC_CLASS)
9261                 return -EINVAL;
9262
9263         /* Hardware has to reinitialize queues and interrupts to
9264          * match packet buffer alignment. Unfortunately, the
9265          * hardware is not flexible enough to do this dynamically.
9266          */
9267         if (netif_running(dev))
9268                 ixgbe_close(dev);
9269         else
9270                 ixgbe_reset(adapter);
9271
9272         ixgbe_clear_interrupt_scheme(adapter);
9273
9274 #ifdef CONFIG_IXGBE_DCB
9275         if (tc) {
9276                 if (adapter->xdp_prog) {
9277                         e_warn(probe, "DCB is not supported with XDP\n");
9278
9279                         ixgbe_init_interrupt_scheme(adapter);
9280                         if (netif_running(dev))
9281                                 ixgbe_open(dev);
9282                         return -EINVAL;
9283                 }
9284
9285                 netdev_set_num_tc(dev, tc);
9286                 ixgbe_set_prio_tc_map(adapter);
9287
9288                 adapter->hw_tcs = tc;
9289                 adapter->flags |= IXGBE_FLAG_DCB_ENABLED;
9290
9291                 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
9292                         adapter->last_lfc_mode = adapter->hw.fc.requested_mode;
9293                         adapter->hw.fc.requested_mode = ixgbe_fc_none;
9294                 }
9295         } else {
9296                 netdev_reset_tc(dev);
9297
9298                 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
9299                         adapter->hw.fc.requested_mode = adapter->last_lfc_mode;
9300
9301                 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
9302                 adapter->hw_tcs = tc;
9303
9304                 adapter->temp_dcb_cfg.pfc_mode_enable = false;
9305                 adapter->dcb_cfg.pfc_mode_enable = false;
9306         }
9307
9308         ixgbe_validate_rtr(adapter, tc);
9309
9310 #endif /* CONFIG_IXGBE_DCB */
9311         ixgbe_init_interrupt_scheme(adapter);
9312
9313         ixgbe_defrag_macvlan_pools(dev);
9314
9315         if (netif_running(dev))
9316                 return ixgbe_open(dev);
9317
9318         return 0;
9319 }
9320
9321 static int ixgbe_delete_clsu32(struct ixgbe_adapter *adapter,
9322                                struct tc_cls_u32_offload *cls)
9323 {
9324         u32 hdl = cls->knode.handle;
9325         u32 uhtid = TC_U32_USERHTID(cls->knode.handle);
9326         u32 loc = cls->knode.handle & 0xfffff;
9327         int err = 0, i, j;
9328         struct ixgbe_jump_table *jump = NULL;
9329
9330         if (loc > IXGBE_MAX_HW_ENTRIES)
9331                 return -EINVAL;
9332
9333         if ((uhtid != 0x800) && (uhtid >= IXGBE_MAX_LINK_HANDLE))
9334                 return -EINVAL;
9335
9336         /* Clear this filter in the link data it is associated with */
9337         if (uhtid != 0x800) {
9338                 jump = adapter->jump_tables[uhtid];
9339                 if (!jump)
9340                         return -EINVAL;
9341                 if (!test_bit(loc - 1, jump->child_loc_map))
9342                         return -EINVAL;
9343                 clear_bit(loc - 1, jump->child_loc_map);
9344         }
9345
9346         /* Check if the filter being deleted is a link */
9347         for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++) {
9348                 jump = adapter->jump_tables[i];
9349                 if (jump && jump->link_hdl == hdl) {
9350                         /* Delete filters in the hardware in the child hash
9351                          * table associated with this link
9352                          */
9353                         for (j = 0; j < IXGBE_MAX_HW_ENTRIES; j++) {
9354                                 if (!test_bit(j, jump->child_loc_map))
9355                                         continue;
9356                                 spin_lock(&adapter->fdir_perfect_lock);
9357                                 err = ixgbe_update_ethtool_fdir_entry(adapter,
9358                                                                       NULL,
9359                                                                       j + 1);
9360                                 spin_unlock(&adapter->fdir_perfect_lock);
9361                                 clear_bit(j, jump->child_loc_map);
9362                         }
9363                         /* Remove resources for this link */
9364                         kfree(jump->input);
9365                         kfree(jump->mask);
9366                         kfree(jump);
9367                         adapter->jump_tables[i] = NULL;
9368                         return err;
9369                 }
9370         }
9371
9372         spin_lock(&adapter->fdir_perfect_lock);
9373         err = ixgbe_update_ethtool_fdir_entry(adapter, NULL, loc);
9374         spin_unlock(&adapter->fdir_perfect_lock);
9375         return err;
9376 }
9377
9378 static int ixgbe_configure_clsu32_add_hnode(struct ixgbe_adapter *adapter,
9379                                             struct tc_cls_u32_offload *cls)
9380 {
9381         u32 uhtid = TC_U32_USERHTID(cls->hnode.handle);
9382
9383         if (uhtid >= IXGBE_MAX_LINK_HANDLE)
9384                 return -EINVAL;
9385
9386         /* This ixgbe devices do not support hash tables at the moment
9387          * so abort when given hash tables.
9388          */
9389         if (cls->hnode.divisor > 0)
9390                 return -EINVAL;
9391
9392         set_bit(uhtid - 1, &adapter->tables);
9393         return 0;
9394 }
9395
9396 static int ixgbe_configure_clsu32_del_hnode(struct ixgbe_adapter *adapter,
9397                                             struct tc_cls_u32_offload *cls)
9398 {
9399         u32 uhtid = TC_U32_USERHTID(cls->hnode.handle);
9400
9401         if (uhtid >= IXGBE_MAX_LINK_HANDLE)
9402                 return -EINVAL;
9403
9404         clear_bit(uhtid - 1, &adapter->tables);
9405         return 0;
9406 }
9407
9408 #ifdef CONFIG_NET_CLS_ACT
9409 struct upper_walk_data {
9410         struct ixgbe_adapter *adapter;
9411         u64 action;
9412         int ifindex;
9413         u8 queue;
9414 };
9415
9416 static int get_macvlan_queue(struct net_device *upper,
9417                              struct netdev_nested_priv *priv)
9418 {
9419         if (netif_is_macvlan(upper)) {
9420                 struct ixgbe_fwd_adapter *vadapter = macvlan_accel_priv(upper);
9421                 struct ixgbe_adapter *adapter;
9422                 struct upper_walk_data *data;
9423                 int ifindex;
9424
9425                 data = (struct upper_walk_data *)priv->data;
9426                 ifindex = data->ifindex;
9427                 adapter = data->adapter;
9428                 if (vadapter && upper->ifindex == ifindex) {
9429                         data->queue = adapter->rx_ring[vadapter->rx_base_queue]->reg_idx;
9430                         data->action = data->queue;
9431                         return 1;
9432                 }
9433         }
9434
9435         return 0;
9436 }
9437
9438 static int handle_redirect_action(struct ixgbe_adapter *adapter, int ifindex,
9439                                   u8 *queue, u64 *action)
9440 {
9441         struct ixgbe_ring_feature *vmdq = &adapter->ring_feature[RING_F_VMDQ];
9442         unsigned int num_vfs = adapter->num_vfs, vf;
9443         struct netdev_nested_priv priv;
9444         struct upper_walk_data data;
9445         struct net_device *upper;
9446
9447         /* redirect to a SRIOV VF */
9448         for (vf = 0; vf < num_vfs; ++vf) {
9449                 upper = pci_get_drvdata(adapter->vfinfo[vf].vfdev);
9450                 if (upper->ifindex == ifindex) {
9451                         *queue = vf * __ALIGN_MASK(1, ~vmdq->mask);
9452                         *action = vf + 1;
9453                         *action <<= ETHTOOL_RX_FLOW_SPEC_RING_VF_OFF;
9454                         return 0;
9455                 }
9456         }
9457
9458         /* redirect to a offloaded macvlan netdev */
9459         data.adapter = adapter;
9460         data.ifindex = ifindex;
9461         data.action = 0;
9462         data.queue = 0;
9463         priv.data = (void *)&data;
9464         if (netdev_walk_all_upper_dev_rcu(adapter->netdev,
9465                                           get_macvlan_queue, &priv)) {
9466                 *action = data.action;
9467                 *queue = data.queue;
9468
9469                 return 0;
9470         }
9471
9472         return -EINVAL;
9473 }
9474
9475 static int parse_tc_actions(struct ixgbe_adapter *adapter,
9476                             struct tcf_exts *exts, u64 *action, u8 *queue)
9477 {
9478         const struct tc_action *a;
9479         int i;
9480
9481         if (!tcf_exts_has_actions(exts))
9482                 return -EINVAL;
9483
9484         tcf_exts_for_each_action(i, a, exts) {
9485                 /* Drop action */
9486                 if (is_tcf_gact_shot(a)) {
9487                         *action = IXGBE_FDIR_DROP_QUEUE;
9488                         *queue = IXGBE_FDIR_DROP_QUEUE;
9489                         return 0;
9490                 }
9491
9492                 /* Redirect to a VF or a offloaded macvlan */
9493                 if (is_tcf_mirred_egress_redirect(a)) {
9494                         struct net_device *dev = tcf_mirred_dev(a);
9495
9496                         if (!dev)
9497                                 return -EINVAL;
9498                         return handle_redirect_action(adapter, dev->ifindex,
9499                                                       queue, action);
9500                 }
9501
9502                 return -EINVAL;
9503         }
9504
9505         return -EINVAL;
9506 }
9507 #else
9508 static int parse_tc_actions(struct ixgbe_adapter *adapter,
9509                             struct tcf_exts *exts, u64 *action, u8 *queue)
9510 {
9511         return -EINVAL;
9512 }
9513 #endif /* CONFIG_NET_CLS_ACT */
9514
9515 static int ixgbe_clsu32_build_input(struct ixgbe_fdir_filter *input,
9516                                     union ixgbe_atr_input *mask,
9517                                     struct tc_cls_u32_offload *cls,
9518                                     struct ixgbe_mat_field *field_ptr,
9519                                     struct ixgbe_nexthdr *nexthdr)
9520 {
9521         int i, j, off;
9522         __be32 val, m;
9523         bool found_entry = false, found_jump_field = false;
9524
9525         for (i = 0; i < cls->knode.sel->nkeys; i++) {
9526                 off = cls->knode.sel->keys[i].off;
9527                 val = cls->knode.sel->keys[i].val;
9528                 m = cls->knode.sel->keys[i].mask;
9529
9530                 for (j = 0; field_ptr[j].val; j++) {
9531                         if (field_ptr[j].off == off) {
9532                                 field_ptr[j].val(input, mask, (__force u32)val,
9533                                                  (__force u32)m);
9534                                 input->filter.formatted.flow_type |=
9535                                         field_ptr[j].type;
9536                                 found_entry = true;
9537                                 break;
9538                         }
9539                 }
9540                 if (nexthdr) {
9541                         if (nexthdr->off == cls->knode.sel->keys[i].off &&
9542                             nexthdr->val ==
9543                             (__force u32)cls->knode.sel->keys[i].val &&
9544                             nexthdr->mask ==
9545                             (__force u32)cls->knode.sel->keys[i].mask)
9546                                 found_jump_field = true;
9547                         else
9548                                 continue;
9549                 }
9550         }
9551
9552         if (nexthdr && !found_jump_field)
9553                 return -EINVAL;
9554
9555         if (!found_entry)
9556                 return 0;
9557
9558         mask->formatted.flow_type = IXGBE_ATR_L4TYPE_IPV6_MASK |
9559                                     IXGBE_ATR_L4TYPE_MASK;
9560
9561         if (input->filter.formatted.flow_type == IXGBE_ATR_FLOW_TYPE_IPV4)
9562                 mask->formatted.flow_type &= IXGBE_ATR_L4TYPE_IPV6_MASK;
9563
9564         return 0;
9565 }
9566
9567 static int ixgbe_configure_clsu32(struct ixgbe_adapter *adapter,
9568                                   struct tc_cls_u32_offload *cls)
9569 {
9570         __be16 protocol = cls->common.protocol;
9571         u32 loc = cls->knode.handle & 0xfffff;
9572         struct ixgbe_hw *hw = &adapter->hw;
9573         struct ixgbe_mat_field *field_ptr;
9574         struct ixgbe_fdir_filter *input = NULL;
9575         union ixgbe_atr_input *mask = NULL;
9576         struct ixgbe_jump_table *jump = NULL;
9577         int i, err = -EINVAL;
9578         u8 queue;
9579         u32 uhtid, link_uhtid;
9580
9581         uhtid = TC_U32_USERHTID(cls->knode.handle);
9582         link_uhtid = TC_U32_USERHTID(cls->knode.link_handle);
9583
9584         /* At the moment cls_u32 jumps to network layer and skips past
9585          * L2 headers. The canonical method to match L2 frames is to use
9586          * negative values. However this is error prone at best but really
9587          * just broken because there is no way to "know" what sort of hdr
9588          * is in front of the network layer. Fix cls_u32 to support L2
9589          * headers when needed.
9590          */
9591         if (protocol != htons(ETH_P_IP))
9592                 return err;
9593
9594         if (loc >= ((1024 << adapter->fdir_pballoc) - 2)) {
9595                 e_err(drv, "Location out of range\n");
9596                 return err;
9597         }
9598
9599         /* cls u32 is a graph starting at root node 0x800. The driver tracks
9600          * links and also the fields used to advance the parser across each
9601          * link (e.g. nexthdr/eat parameters from 'tc'). This way we can map
9602          * the u32 graph onto the hardware parse graph denoted in ixgbe_model.h
9603          * To add support for new nodes update ixgbe_model.h parse structures
9604          * this function _should_ be generic try not to hardcode values here.
9605          */
9606         if (uhtid == 0x800) {
9607                 field_ptr = (adapter->jump_tables[0])->mat;
9608         } else {
9609                 if (uhtid >= IXGBE_MAX_LINK_HANDLE)
9610                         return err;
9611                 if (!adapter->jump_tables[uhtid])
9612                         return err;
9613                 field_ptr = (adapter->jump_tables[uhtid])->mat;
9614         }
9615
9616         if (!field_ptr)
9617                 return err;
9618
9619         /* At this point we know the field_ptr is valid and need to either
9620          * build cls_u32 link or attach filter. Because adding a link to
9621          * a handle that does not exist is invalid and the same for adding
9622          * rules to handles that don't exist.
9623          */
9624
9625         if (link_uhtid) {
9626                 struct ixgbe_nexthdr *nexthdr = ixgbe_ipv4_jumps;
9627
9628                 if (link_uhtid >= IXGBE_MAX_LINK_HANDLE)
9629                         return err;
9630
9631                 if (!test_bit(link_uhtid - 1, &adapter->tables))
9632                         return err;
9633
9634                 /* Multiple filters as links to the same hash table are not
9635                  * supported. To add a new filter with the same next header
9636                  * but different match/jump conditions, create a new hash table
9637                  * and link to it.
9638                  */
9639                 if (adapter->jump_tables[link_uhtid] &&
9640                     (adapter->jump_tables[link_uhtid])->link_hdl) {
9641                         e_err(drv, "Link filter exists for link: %x\n",
9642                               link_uhtid);
9643                         return err;
9644                 }
9645
9646                 for (i = 0; nexthdr[i].jump; i++) {
9647                         if (nexthdr[i].o != cls->knode.sel->offoff ||
9648                             nexthdr[i].s != cls->knode.sel->offshift ||
9649                             nexthdr[i].m !=
9650                             (__force u32)cls->knode.sel->offmask)
9651                                 return err;
9652
9653                         jump = kzalloc(sizeof(*jump), GFP_KERNEL);
9654                         if (!jump)
9655                                 return -ENOMEM;
9656                         input = kzalloc(sizeof(*input), GFP_KERNEL);
9657                         if (!input) {
9658                                 err = -ENOMEM;
9659                                 goto free_jump;
9660                         }
9661                         mask = kzalloc(sizeof(*mask), GFP_KERNEL);
9662                         if (!mask) {
9663                                 err = -ENOMEM;
9664                                 goto free_input;
9665                         }
9666                         jump->input = input;
9667                         jump->mask = mask;
9668                         jump->link_hdl = cls->knode.handle;
9669
9670                         err = ixgbe_clsu32_build_input(input, mask, cls,
9671                                                        field_ptr, &nexthdr[i]);
9672                         if (!err) {
9673                                 jump->mat = nexthdr[i].jump;
9674                                 adapter->jump_tables[link_uhtid] = jump;
9675                                 break;
9676                         } else {
9677                                 kfree(mask);
9678                                 kfree(input);
9679                                 kfree(jump);
9680                         }
9681                 }
9682                 return 0;
9683         }
9684
9685         input = kzalloc(sizeof(*input), GFP_KERNEL);
9686         if (!input)
9687                 return -ENOMEM;
9688         mask = kzalloc(sizeof(*mask), GFP_KERNEL);
9689         if (!mask) {
9690                 err = -ENOMEM;
9691                 goto free_input;
9692         }
9693
9694         if ((uhtid != 0x800) && (adapter->jump_tables[uhtid])) {
9695                 if ((adapter->jump_tables[uhtid])->input)
9696                         memcpy(input, (adapter->jump_tables[uhtid])->input,
9697                                sizeof(*input));
9698                 if ((adapter->jump_tables[uhtid])->mask)
9699                         memcpy(mask, (adapter->jump_tables[uhtid])->mask,
9700                                sizeof(*mask));
9701
9702                 /* Lookup in all child hash tables if this location is already
9703                  * filled with a filter
9704                  */
9705                 for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++) {
9706                         struct ixgbe_jump_table *link = adapter->jump_tables[i];
9707
9708                         if (link && (test_bit(loc - 1, link->child_loc_map))) {
9709                                 e_err(drv, "Filter exists in location: %x\n",
9710                                       loc);
9711                                 err = -EINVAL;
9712                                 goto err_out;
9713                         }
9714                 }
9715         }
9716         err = ixgbe_clsu32_build_input(input, mask, cls, field_ptr, NULL);
9717         if (err)
9718                 goto err_out;
9719
9720         err = parse_tc_actions(adapter, cls->knode.exts, &input->action,
9721                                &queue);
9722         if (err < 0)
9723                 goto err_out;
9724
9725         input->sw_idx = loc;
9726
9727         spin_lock(&adapter->fdir_perfect_lock);
9728
9729         if (hlist_empty(&adapter->fdir_filter_list)) {
9730                 memcpy(&adapter->fdir_mask, mask, sizeof(*mask));
9731                 err = ixgbe_fdir_set_input_mask_82599(hw, mask);
9732                 if (err)
9733                         goto err_out_w_lock;
9734         } else if (memcmp(&adapter->fdir_mask, mask, sizeof(*mask))) {
9735                 err = -EINVAL;
9736                 goto err_out_w_lock;
9737         }
9738
9739         ixgbe_atr_compute_perfect_hash_82599(&input->filter, mask);
9740         err = ixgbe_fdir_write_perfect_filter_82599(hw, &input->filter,
9741                                                     input->sw_idx, queue);
9742         if (err)
9743                 goto err_out_w_lock;
9744
9745         ixgbe_update_ethtool_fdir_entry(adapter, input, input->sw_idx);
9746         spin_unlock(&adapter->fdir_perfect_lock);
9747
9748         if ((uhtid != 0x800) && (adapter->jump_tables[uhtid]))
9749                 set_bit(loc - 1, (adapter->jump_tables[uhtid])->child_loc_map);
9750
9751         kfree(mask);
9752         return err;
9753 err_out_w_lock:
9754         spin_unlock(&adapter->fdir_perfect_lock);
9755 err_out:
9756         kfree(mask);
9757 free_input:
9758         kfree(input);
9759 free_jump:
9760         kfree(jump);
9761         return err;
9762 }
9763
9764 static int ixgbe_setup_tc_cls_u32(struct ixgbe_adapter *adapter,
9765                                   struct tc_cls_u32_offload *cls_u32)
9766 {
9767         switch (cls_u32->command) {
9768         case TC_CLSU32_NEW_KNODE:
9769         case TC_CLSU32_REPLACE_KNODE:
9770                 return ixgbe_configure_clsu32(adapter, cls_u32);
9771         case TC_CLSU32_DELETE_KNODE:
9772                 return ixgbe_delete_clsu32(adapter, cls_u32);
9773         case TC_CLSU32_NEW_HNODE:
9774         case TC_CLSU32_REPLACE_HNODE:
9775                 return ixgbe_configure_clsu32_add_hnode(adapter, cls_u32);
9776         case TC_CLSU32_DELETE_HNODE:
9777                 return ixgbe_configure_clsu32_del_hnode(adapter, cls_u32);
9778         default:
9779                 return -EOPNOTSUPP;
9780         }
9781 }
9782
9783 static int ixgbe_setup_tc_block_cb(enum tc_setup_type type, void *type_data,
9784                                    void *cb_priv)
9785 {
9786         struct ixgbe_adapter *adapter = cb_priv;
9787
9788         if (!tc_cls_can_offload_and_chain0(adapter->netdev, type_data))
9789                 return -EOPNOTSUPP;
9790
9791         switch (type) {
9792         case TC_SETUP_CLSU32:
9793                 return ixgbe_setup_tc_cls_u32(adapter, type_data);
9794         default:
9795                 return -EOPNOTSUPP;
9796         }
9797 }
9798
9799 static int ixgbe_setup_tc_mqprio(struct net_device *dev,
9800                                  struct tc_mqprio_qopt *mqprio)
9801 {
9802         mqprio->hw = TC_MQPRIO_HW_OFFLOAD_TCS;
9803         return ixgbe_setup_tc(dev, mqprio->num_tc);
9804 }
9805
9806 static LIST_HEAD(ixgbe_block_cb_list);
9807
9808 static int __ixgbe_setup_tc(struct net_device *dev, enum tc_setup_type type,
9809                             void *type_data)
9810 {
9811         struct ixgbe_adapter *adapter = netdev_priv(dev);
9812
9813         switch (type) {
9814         case TC_SETUP_BLOCK:
9815                 return flow_block_cb_setup_simple(type_data,
9816                                                   &ixgbe_block_cb_list,
9817                                                   ixgbe_setup_tc_block_cb,
9818                                                   adapter, adapter, true);
9819         case TC_SETUP_QDISC_MQPRIO:
9820                 return ixgbe_setup_tc_mqprio(dev, type_data);
9821         default:
9822                 return -EOPNOTSUPP;
9823         }
9824 }
9825
9826 #ifdef CONFIG_PCI_IOV
9827 void ixgbe_sriov_reinit(struct ixgbe_adapter *adapter)
9828 {
9829         struct net_device *netdev = adapter->netdev;
9830
9831         rtnl_lock();
9832         ixgbe_setup_tc(netdev, adapter->hw_tcs);
9833         rtnl_unlock();
9834 }
9835
9836 #endif
9837 void ixgbe_do_reset(struct net_device *netdev)
9838 {
9839         struct ixgbe_adapter *adapter = netdev_priv(netdev);
9840
9841         if (netif_running(netdev))
9842                 ixgbe_reinit_locked(adapter);
9843         else
9844                 ixgbe_reset(adapter);
9845 }
9846
9847 static netdev_features_t ixgbe_fix_features(struct net_device *netdev,
9848                                             netdev_features_t features)
9849 {
9850         struct ixgbe_adapter *adapter = netdev_priv(netdev);
9851
9852         /* If Rx checksum is disabled, then RSC/LRO should also be disabled */
9853         if (!(features & NETIF_F_RXCSUM))
9854                 features &= ~NETIF_F_LRO;
9855
9856         /* Turn off LRO if not RSC capable */
9857         if (!(adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE))
9858                 features &= ~NETIF_F_LRO;
9859
9860         if (adapter->xdp_prog && (features & NETIF_F_LRO)) {
9861                 e_dev_err("LRO is not supported with XDP\n");
9862                 features &= ~NETIF_F_LRO;
9863         }
9864
9865         return features;
9866 }
9867
9868 static void ixgbe_reset_l2fw_offload(struct ixgbe_adapter *adapter)
9869 {
9870         int rss = min_t(int, ixgbe_max_rss_indices(adapter),
9871                         num_online_cpus());
9872
9873         /* go back to full RSS if we're not running SR-IOV */
9874         if (!adapter->ring_feature[RING_F_VMDQ].offset)
9875                 adapter->flags &= ~(IXGBE_FLAG_VMDQ_ENABLED |
9876                                     IXGBE_FLAG_SRIOV_ENABLED);
9877
9878         adapter->ring_feature[RING_F_RSS].limit = rss;
9879         adapter->ring_feature[RING_F_VMDQ].limit = 1;
9880
9881         ixgbe_setup_tc(adapter->netdev, adapter->hw_tcs);
9882 }
9883
9884 static int ixgbe_set_features(struct net_device *netdev,
9885                               netdev_features_t features)
9886 {
9887         struct ixgbe_adapter *adapter = netdev_priv(netdev);
9888         netdev_features_t changed = netdev->features ^ features;
9889         bool need_reset = false;
9890
9891         /* Make sure RSC matches LRO, reset if change */
9892         if (!(features & NETIF_F_LRO)) {
9893                 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
9894                         need_reset = true;
9895                 adapter->flags2 &= ~IXGBE_FLAG2_RSC_ENABLED;
9896         } else if ((adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE) &&
9897                    !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)) {
9898                 if (adapter->rx_itr_setting == 1 ||
9899                     adapter->rx_itr_setting > IXGBE_MIN_RSC_ITR) {
9900                         adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
9901                         need_reset = true;
9902                 } else if ((changed ^ features) & NETIF_F_LRO) {
9903                         e_info(probe, "rx-usecs set too low, "
9904                                "disabling RSC\n");
9905                 }
9906         }
9907
9908         /*
9909          * Check if Flow Director n-tuple support or hw_tc support was
9910          * enabled or disabled.  If the state changed, we need to reset.
9911          */
9912         if ((features & NETIF_F_NTUPLE) || (features & NETIF_F_HW_TC)) {
9913                 /* turn off ATR, enable perfect filters and reset */
9914                 if (!(adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
9915                         need_reset = true;
9916
9917                 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
9918                 adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
9919         } else {
9920                 /* turn off perfect filters, enable ATR and reset */
9921                 if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
9922                         need_reset = true;
9923
9924                 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
9925
9926                 /* We cannot enable ATR if SR-IOV is enabled */
9927                 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED ||
9928                     /* We cannot enable ATR if we have 2 or more tcs */
9929                     (adapter->hw_tcs > 1) ||
9930                     /* We cannot enable ATR if RSS is disabled */
9931                     (adapter->ring_feature[RING_F_RSS].limit <= 1) ||
9932                     /* A sample rate of 0 indicates ATR disabled */
9933                     (!adapter->atr_sample_rate))
9934                         ; /* do nothing not supported */
9935                 else /* otherwise supported and set the flag */
9936                         adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
9937         }
9938
9939         if (changed & NETIF_F_RXALL)
9940                 need_reset = true;
9941
9942         netdev->features = features;
9943
9944         if ((changed & NETIF_F_HW_L2FW_DOFFLOAD) && adapter->num_rx_pools > 1)
9945                 ixgbe_reset_l2fw_offload(adapter);
9946         else if (need_reset)
9947                 ixgbe_do_reset(netdev);
9948         else if (changed & (NETIF_F_HW_VLAN_CTAG_RX |
9949                             NETIF_F_HW_VLAN_CTAG_FILTER))
9950                 ixgbe_set_rx_mode(netdev);
9951
9952         return 1;
9953 }
9954
9955 static int ixgbe_ndo_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
9956                              struct net_device *dev,
9957                              const unsigned char *addr, u16 vid,
9958                              u16 flags, bool *notified,
9959                              struct netlink_ext_ack *extack)
9960 {
9961         /* guarantee we can provide a unique filter for the unicast address */
9962         if (is_unicast_ether_addr(addr) || is_link_local_ether_addr(addr)) {
9963                 struct ixgbe_adapter *adapter = netdev_priv(dev);
9964                 u16 pool = VMDQ_P(0);
9965
9966                 if (netdev_uc_count(dev) >= ixgbe_available_rars(adapter, pool))
9967                         return -ENOMEM;
9968         }
9969
9970         return ndo_dflt_fdb_add(ndm, tb, dev, addr, vid, flags);
9971 }
9972
9973 /**
9974  * ixgbe_configure_bridge_mode - set various bridge modes
9975  * @adapter: the private structure
9976  * @mode: requested bridge mode
9977  *
9978  * Configure some settings require for various bridge modes.
9979  **/
9980 static int ixgbe_configure_bridge_mode(struct ixgbe_adapter *adapter,
9981                                        __u16 mode)
9982 {
9983         struct ixgbe_hw *hw = &adapter->hw;
9984         unsigned int p, num_pools;
9985         u32 vmdctl;
9986
9987         switch (mode) {
9988         case BRIDGE_MODE_VEPA:
9989                 /* disable Tx loopback, rely on switch hairpin mode */
9990                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC, 0);
9991
9992                 /* must enable Rx switching replication to allow multicast
9993                  * packet reception on all VFs, and to enable source address
9994                  * pruning.
9995                  */
9996                 vmdctl = IXGBE_READ_REG(hw, IXGBE_VMD_CTL);
9997                 vmdctl |= IXGBE_VT_CTL_REPLEN;
9998                 IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
9999
10000                 /* enable Rx source address pruning. Note, this requires
10001                  * replication to be enabled or else it does nothing.
10002                  */
10003                 num_pools = adapter->num_vfs + adapter->num_rx_pools;
10004                 for (p = 0; p < num_pools; p++) {
10005                         if (hw->mac.ops.set_source_address_pruning)
10006                                 hw->mac.ops.set_source_address_pruning(hw,
10007                                                                        true,
10008                                                                        p);
10009                 }
10010                 break;
10011         case BRIDGE_MODE_VEB:
10012                 /* enable Tx loopback for internal VF/PF communication */
10013                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC,
10014                                 IXGBE_PFDTXGSWC_VT_LBEN);
10015
10016                 /* disable Rx switching replication unless we have SR-IOV
10017                  * virtual functions
10018                  */
10019                 vmdctl = IXGBE_READ_REG(hw, IXGBE_VMD_CTL);
10020                 if (!adapter->num_vfs)
10021                         vmdctl &= ~IXGBE_VT_CTL_REPLEN;
10022                 IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
10023
10024                 /* disable Rx source address pruning, since we don't expect to
10025                  * be receiving external loopback of our transmitted frames.
10026                  */
10027                 num_pools = adapter->num_vfs + adapter->num_rx_pools;
10028                 for (p = 0; p < num_pools; p++) {
10029                         if (hw->mac.ops.set_source_address_pruning)
10030                                 hw->mac.ops.set_source_address_pruning(hw,
10031                                                                        false,
10032                                                                        p);
10033                 }
10034                 break;
10035         default:
10036                 return -EINVAL;
10037         }
10038
10039         adapter->bridge_mode = mode;
10040
10041         e_info(drv, "enabling bridge mode: %s\n",
10042                mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
10043
10044         return 0;
10045 }
10046
10047 static int ixgbe_ndo_bridge_setlink(struct net_device *dev,
10048                                     struct nlmsghdr *nlh, u16 flags,
10049                                     struct netlink_ext_ack *extack)
10050 {
10051         struct ixgbe_adapter *adapter = netdev_priv(dev);
10052         struct nlattr *attr, *br_spec;
10053         int rem;
10054
10055         if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
10056                 return -EOPNOTSUPP;
10057
10058         br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
10059         if (!br_spec)
10060                 return -EINVAL;
10061
10062         nla_for_each_nested_type(attr, IFLA_BRIDGE_MODE, br_spec, rem) {
10063                 __u16 mode = nla_get_u16(attr);
10064                 int status = ixgbe_configure_bridge_mode(adapter, mode);
10065
10066                 if (status)
10067                         return status;
10068
10069                 break;
10070         }
10071
10072         return 0;
10073 }
10074
10075 static int ixgbe_ndo_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
10076                                     struct net_device *dev,
10077                                     u32 filter_mask, int nlflags)
10078 {
10079         struct ixgbe_adapter *adapter = netdev_priv(dev);
10080
10081         if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
10082                 return 0;
10083
10084         return ndo_dflt_bridge_getlink(skb, pid, seq, dev,
10085                                        adapter->bridge_mode, 0, 0, nlflags,
10086                                        filter_mask, NULL);
10087 }
10088
10089 static void *ixgbe_fwd_add(struct net_device *pdev, struct net_device *vdev)
10090 {
10091         struct ixgbe_adapter *adapter = netdev_priv(pdev);
10092         struct ixgbe_fwd_adapter *accel;
10093         int tcs = adapter->hw_tcs ? : 1;
10094         int pool, err;
10095
10096         if (adapter->xdp_prog) {
10097                 e_warn(probe, "L2FW offload is not supported with XDP\n");
10098                 return ERR_PTR(-EINVAL);
10099         }
10100
10101         /* The hardware supported by ixgbe only filters on the destination MAC
10102          * address. In order to avoid issues we only support offloading modes
10103          * where the hardware can actually provide the functionality.
10104          */
10105         if (!macvlan_supports_dest_filter(vdev))
10106                 return ERR_PTR(-EMEDIUMTYPE);
10107
10108         /* We need to lock down the macvlan to be a single queue device so that
10109          * we can reuse the tc_to_txq field in the macvlan netdev to represent
10110          * the queue mapping to our netdev.
10111          */
10112         if (netif_is_multiqueue(vdev))
10113                 return ERR_PTR(-ERANGE);
10114
10115         pool = find_first_zero_bit(adapter->fwd_bitmask, adapter->num_rx_pools);
10116         if (pool == adapter->num_rx_pools) {
10117                 u16 used_pools = adapter->num_vfs + adapter->num_rx_pools;
10118                 u16 reserved_pools;
10119
10120                 if (((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
10121                      adapter->num_rx_pools >= (MAX_TX_QUEUES / tcs)) ||
10122                     adapter->num_rx_pools > IXGBE_MAX_MACVLANS)
10123                         return ERR_PTR(-EBUSY);
10124
10125                 /* Hardware has a limited number of available pools. Each VF,
10126                  * and the PF require a pool. Check to ensure we don't
10127                  * attempt to use more then the available number of pools.
10128                  */
10129                 if (used_pools >= IXGBE_MAX_VF_FUNCTIONS)
10130                         return ERR_PTR(-EBUSY);
10131
10132                 /* Enable VMDq flag so device will be set in VM mode */
10133                 adapter->flags |= IXGBE_FLAG_VMDQ_ENABLED |
10134                                   IXGBE_FLAG_SRIOV_ENABLED;
10135
10136                 /* Try to reserve as many queues per pool as possible,
10137                  * we start with the configurations that support 4 queues
10138                  * per pools, followed by 2, and then by just 1 per pool.
10139                  */
10140                 if (used_pools < 32 && adapter->num_rx_pools < 16)
10141                         reserved_pools = min_t(u16,
10142                                                32 - used_pools,
10143                                                16 - adapter->num_rx_pools);
10144                 else if (adapter->num_rx_pools < 32)
10145                         reserved_pools = min_t(u16,
10146                                                64 - used_pools,
10147                                                32 - adapter->num_rx_pools);
10148                 else
10149                         reserved_pools = 64 - used_pools;
10150
10151
10152                 if (!reserved_pools)
10153                         return ERR_PTR(-EBUSY);
10154
10155                 adapter->ring_feature[RING_F_VMDQ].limit += reserved_pools;
10156
10157                 /* Force reinit of ring allocation with VMDQ enabled */
10158                 err = ixgbe_setup_tc(pdev, adapter->hw_tcs);
10159                 if (err)
10160                         return ERR_PTR(err);
10161
10162                 if (pool >= adapter->num_rx_pools)
10163                         return ERR_PTR(-ENOMEM);
10164         }
10165
10166         accel = kzalloc(sizeof(*accel), GFP_KERNEL);
10167         if (!accel)
10168                 return ERR_PTR(-ENOMEM);
10169
10170         set_bit(pool, adapter->fwd_bitmask);
10171         netdev_set_sb_channel(vdev, pool);
10172         accel->pool = pool;
10173         accel->netdev = vdev;
10174
10175         if (!netif_running(pdev))
10176                 return accel;
10177
10178         err = ixgbe_fwd_ring_up(adapter, accel);
10179         if (err)
10180                 return ERR_PTR(err);
10181
10182         return accel;
10183 }
10184
10185 static void ixgbe_fwd_del(struct net_device *pdev, void *priv)
10186 {
10187         struct ixgbe_fwd_adapter *accel = priv;
10188         struct ixgbe_adapter *adapter = netdev_priv(pdev);
10189         unsigned int rxbase = accel->rx_base_queue;
10190         unsigned int i;
10191
10192         /* delete unicast filter associated with offloaded interface */
10193         ixgbe_del_mac_filter(adapter, accel->netdev->dev_addr,
10194                              VMDQ_P(accel->pool));
10195
10196         /* Allow remaining Rx packets to get flushed out of the
10197          * Rx FIFO before we drop the netdev for the ring.
10198          */
10199         usleep_range(10000, 20000);
10200
10201         for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
10202                 struct ixgbe_ring *ring = adapter->rx_ring[rxbase + i];
10203                 struct ixgbe_q_vector *qv = ring->q_vector;
10204
10205                 /* Make sure we aren't processing any packets and clear
10206                  * netdev to shut down the ring.
10207                  */
10208                 if (netif_running(adapter->netdev))
10209                         napi_synchronize(&qv->napi);
10210                 ring->netdev = NULL;
10211         }
10212
10213         /* unbind the queues and drop the subordinate channel config */
10214         netdev_unbind_sb_channel(pdev, accel->netdev);
10215         netdev_set_sb_channel(accel->netdev, 0);
10216
10217         clear_bit(accel->pool, adapter->fwd_bitmask);
10218         kfree(accel);
10219 }
10220
10221 #define IXGBE_MAX_MAC_HDR_LEN           127
10222 #define IXGBE_MAX_NETWORK_HDR_LEN       511
10223
10224 static netdev_features_t
10225 ixgbe_features_check(struct sk_buff *skb, struct net_device *dev,
10226                      netdev_features_t features)
10227 {
10228         unsigned int network_hdr_len, mac_hdr_len;
10229
10230         /* Make certain the headers can be described by a context descriptor */
10231         mac_hdr_len = skb_network_offset(skb);
10232         if (unlikely(mac_hdr_len > IXGBE_MAX_MAC_HDR_LEN))
10233                 return features & ~(NETIF_F_HW_CSUM |
10234                                     NETIF_F_SCTP_CRC |
10235                                     NETIF_F_GSO_UDP_L4 |
10236                                     NETIF_F_HW_VLAN_CTAG_TX |
10237                                     NETIF_F_TSO |
10238                                     NETIF_F_TSO6);
10239
10240         network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
10241         if (unlikely(network_hdr_len >  IXGBE_MAX_NETWORK_HDR_LEN))
10242                 return features & ~(NETIF_F_HW_CSUM |
10243                                     NETIF_F_SCTP_CRC |
10244                                     NETIF_F_GSO_UDP_L4 |
10245                                     NETIF_F_TSO |
10246                                     NETIF_F_TSO6);
10247
10248         /* We can only support IPV4 TSO in tunnels if we can mangle the
10249          * inner IP ID field, so strip TSO if MANGLEID is not supported.
10250          * IPsec offoad sets skb->encapsulation but still can handle
10251          * the TSO, so it's the exception.
10252          */
10253         if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID)) {
10254 #ifdef CONFIG_IXGBE_IPSEC
10255                 if (!secpath_exists(skb))
10256 #endif
10257                         features &= ~NETIF_F_TSO;
10258         }
10259
10260         return features;
10261 }
10262
10263 static int ixgbe_xdp_setup(struct net_device *dev, struct bpf_prog *prog)
10264 {
10265         int i, frame_size = dev->mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
10266         struct ixgbe_adapter *adapter = netdev_priv(dev);
10267         struct bpf_prog *old_prog;
10268         bool need_reset;
10269         int num_queues;
10270
10271         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
10272                 return -EINVAL;
10273
10274         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
10275                 return -EINVAL;
10276
10277         /* verify ixgbe ring attributes are sufficient for XDP */
10278         for (i = 0; i < adapter->num_rx_queues; i++) {
10279                 struct ixgbe_ring *ring = adapter->rx_ring[i];
10280
10281                 if (ring_is_rsc_enabled(ring))
10282                         return -EINVAL;
10283
10284                 if (frame_size > ixgbe_rx_bufsz(ring))
10285                         return -EINVAL;
10286         }
10287
10288         /* if the number of cpus is much larger than the maximum of queues,
10289          * we should stop it and then return with ENOMEM like before.
10290          */
10291         if (nr_cpu_ids > IXGBE_MAX_XDP_QS * 2)
10292                 return -ENOMEM;
10293
10294         old_prog = xchg(&adapter->xdp_prog, prog);
10295         need_reset = (!!prog != !!old_prog);
10296
10297         /* If transitioning XDP modes reconfigure rings */
10298         if (need_reset) {
10299                 int err;
10300
10301                 if (!prog)
10302                         /* Wait until ndo_xsk_wakeup completes. */
10303                         synchronize_rcu();
10304                 err = ixgbe_setup_tc(dev, adapter->hw_tcs);
10305
10306                 if (err)
10307                         return -EINVAL;
10308                 if (!prog)
10309                         xdp_features_clear_redirect_target(dev);
10310         } else {
10311                 for (i = 0; i < adapter->num_rx_queues; i++) {
10312                         WRITE_ONCE(adapter->rx_ring[i]->xdp_prog,
10313                                    adapter->xdp_prog);
10314                 }
10315         }
10316
10317         if (old_prog)
10318                 bpf_prog_put(old_prog);
10319
10320         /* Kick start the NAPI context if there is an AF_XDP socket open
10321          * on that queue id. This so that receiving will start.
10322          */
10323         if (need_reset && prog) {
10324                 num_queues = min_t(int, adapter->num_rx_queues,
10325                                    adapter->num_xdp_queues);
10326                 for (i = 0; i < num_queues; i++)
10327                         if (adapter->xdp_ring[i]->xsk_pool)
10328                                 (void)ixgbe_xsk_wakeup(adapter->netdev, i,
10329                                                        XDP_WAKEUP_RX);
10330                 xdp_features_set_redirect_target(dev, true);
10331         }
10332
10333         return 0;
10334 }
10335
10336 static int ixgbe_xdp(struct net_device *dev, struct netdev_bpf *xdp)
10337 {
10338         struct ixgbe_adapter *adapter = netdev_priv(dev);
10339
10340         switch (xdp->command) {
10341         case XDP_SETUP_PROG:
10342                 return ixgbe_xdp_setup(dev, xdp->prog);
10343         case XDP_SETUP_XSK_POOL:
10344                 return ixgbe_xsk_pool_setup(adapter, xdp->xsk.pool,
10345                                             xdp->xsk.queue_id);
10346
10347         default:
10348                 return -EINVAL;
10349         }
10350 }
10351
10352 void ixgbe_xdp_ring_update_tail(struct ixgbe_ring *ring)
10353 {
10354         /* Force memory writes to complete before letting h/w know there
10355          * are new descriptors to fetch.
10356          */
10357         wmb();
10358         writel(ring->next_to_use, ring->tail);
10359 }
10360
10361 void ixgbe_xdp_ring_update_tail_locked(struct ixgbe_ring *ring)
10362 {
10363         if (static_branch_unlikely(&ixgbe_xdp_locking_key))
10364                 spin_lock(&ring->tx_lock);
10365         ixgbe_xdp_ring_update_tail(ring);
10366         if (static_branch_unlikely(&ixgbe_xdp_locking_key))
10367                 spin_unlock(&ring->tx_lock);
10368 }
10369
10370 static int ixgbe_xdp_xmit(struct net_device *dev, int n,
10371                           struct xdp_frame **frames, u32 flags)
10372 {
10373         struct ixgbe_adapter *adapter = netdev_priv(dev);
10374         struct ixgbe_ring *ring;
10375         int nxmit = 0;
10376         int i;
10377
10378         if (unlikely(test_bit(__IXGBE_DOWN, &adapter->state)))
10379                 return -ENETDOWN;
10380
10381         if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
10382                 return -EINVAL;
10383
10384         /* During program transitions its possible adapter->xdp_prog is assigned
10385          * but ring has not been configured yet. In this case simply abort xmit.
10386          */
10387         ring = adapter->xdp_prog ? ixgbe_determine_xdp_ring(adapter) : NULL;
10388         if (unlikely(!ring))
10389                 return -ENXIO;
10390
10391         if (unlikely(test_bit(__IXGBE_TX_DISABLED, &ring->state)))
10392                 return -ENXIO;
10393
10394         if (static_branch_unlikely(&ixgbe_xdp_locking_key))
10395                 spin_lock(&ring->tx_lock);
10396
10397         for (i = 0; i < n; i++) {
10398                 struct xdp_frame *xdpf = frames[i];
10399                 int err;
10400
10401                 err = ixgbe_xmit_xdp_ring(ring, xdpf);
10402                 if (err != IXGBE_XDP_TX)
10403                         break;
10404                 nxmit++;
10405         }
10406
10407         if (unlikely(flags & XDP_XMIT_FLUSH))
10408                 ixgbe_xdp_ring_update_tail(ring);
10409
10410         if (static_branch_unlikely(&ixgbe_xdp_locking_key))
10411                 spin_unlock(&ring->tx_lock);
10412
10413         return nxmit;
10414 }
10415
10416 static const struct net_device_ops ixgbe_netdev_ops = {
10417         .ndo_open               = ixgbe_open,
10418         .ndo_stop               = ixgbe_close,
10419         .ndo_start_xmit         = ixgbe_xmit_frame,
10420         .ndo_set_rx_mode        = ixgbe_set_rx_mode,
10421         .ndo_validate_addr      = eth_validate_addr,
10422         .ndo_set_mac_address    = ixgbe_set_mac,
10423         .ndo_change_mtu         = ixgbe_change_mtu,
10424         .ndo_tx_timeout         = ixgbe_tx_timeout,
10425         .ndo_set_tx_maxrate     = ixgbe_tx_maxrate,
10426         .ndo_vlan_rx_add_vid    = ixgbe_vlan_rx_add_vid,
10427         .ndo_vlan_rx_kill_vid   = ixgbe_vlan_rx_kill_vid,
10428         .ndo_eth_ioctl          = ixgbe_ioctl,
10429         .ndo_set_vf_mac         = ixgbe_ndo_set_vf_mac,
10430         .ndo_set_vf_vlan        = ixgbe_ndo_set_vf_vlan,
10431         .ndo_set_vf_rate        = ixgbe_ndo_set_vf_bw,
10432         .ndo_set_vf_spoofchk    = ixgbe_ndo_set_vf_spoofchk,
10433         .ndo_set_vf_link_state  = ixgbe_ndo_set_vf_link_state,
10434         .ndo_set_vf_rss_query_en = ixgbe_ndo_set_vf_rss_query_en,
10435         .ndo_set_vf_trust       = ixgbe_ndo_set_vf_trust,
10436         .ndo_get_vf_config      = ixgbe_ndo_get_vf_config,
10437         .ndo_get_vf_stats       = ixgbe_ndo_get_vf_stats,
10438         .ndo_get_stats64        = ixgbe_get_stats64,
10439         .ndo_setup_tc           = __ixgbe_setup_tc,
10440 #ifdef IXGBE_FCOE
10441         .ndo_select_queue       = ixgbe_select_queue,
10442         .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
10443         .ndo_fcoe_ddp_target = ixgbe_fcoe_ddp_target,
10444         .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
10445         .ndo_fcoe_enable = ixgbe_fcoe_enable,
10446         .ndo_fcoe_disable = ixgbe_fcoe_disable,
10447         .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
10448         .ndo_fcoe_get_hbainfo = ixgbe_fcoe_get_hbainfo,
10449 #endif /* IXGBE_FCOE */
10450         .ndo_set_features = ixgbe_set_features,
10451         .ndo_fix_features = ixgbe_fix_features,
10452         .ndo_fdb_add            = ixgbe_ndo_fdb_add,
10453         .ndo_bridge_setlink     = ixgbe_ndo_bridge_setlink,
10454         .ndo_bridge_getlink     = ixgbe_ndo_bridge_getlink,
10455         .ndo_dfwd_add_station   = ixgbe_fwd_add,
10456         .ndo_dfwd_del_station   = ixgbe_fwd_del,
10457         .ndo_features_check     = ixgbe_features_check,
10458         .ndo_bpf                = ixgbe_xdp,
10459         .ndo_xdp_xmit           = ixgbe_xdp_xmit,
10460         .ndo_xsk_wakeup         = ixgbe_xsk_wakeup,
10461 };
10462
10463 static void ixgbe_disable_txr_hw(struct ixgbe_adapter *adapter,
10464                                  struct ixgbe_ring *tx_ring)
10465 {
10466         unsigned long wait_delay, delay_interval;
10467         struct ixgbe_hw *hw = &adapter->hw;
10468         u8 reg_idx = tx_ring->reg_idx;
10469         int wait_loop;
10470         u32 txdctl;
10471
10472         IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
10473
10474         /* delay mechanism from ixgbe_disable_tx */
10475         delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
10476
10477         wait_loop = IXGBE_MAX_RX_DESC_POLL;
10478         wait_delay = delay_interval;
10479
10480         while (wait_loop--) {
10481                 usleep_range(wait_delay, wait_delay + 10);
10482                 wait_delay += delay_interval * 2;
10483                 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
10484
10485                 if (!(txdctl & IXGBE_TXDCTL_ENABLE))
10486                         return;
10487         }
10488
10489         e_err(drv, "TXDCTL.ENABLE not cleared within the polling period\n");
10490 }
10491
10492 static void ixgbe_disable_txr(struct ixgbe_adapter *adapter,
10493                               struct ixgbe_ring *tx_ring)
10494 {
10495         set_bit(__IXGBE_TX_DISABLED, &tx_ring->state);
10496         ixgbe_disable_txr_hw(adapter, tx_ring);
10497 }
10498
10499 static void ixgbe_disable_rxr_hw(struct ixgbe_adapter *adapter,
10500                                  struct ixgbe_ring *rx_ring)
10501 {
10502         unsigned long wait_delay, delay_interval;
10503         struct ixgbe_hw *hw = &adapter->hw;
10504         u8 reg_idx = rx_ring->reg_idx;
10505         int wait_loop;
10506         u32 rxdctl;
10507
10508         rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
10509         rxdctl &= ~IXGBE_RXDCTL_ENABLE;
10510         rxdctl |= IXGBE_RXDCTL_SWFLSH;
10511
10512         /* write value back with RXDCTL.ENABLE bit cleared */
10513         IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
10514
10515         /* RXDCTL.EN may not change on 82598 if link is down, so skip it */
10516         if (hw->mac.type == ixgbe_mac_82598EB &&
10517             !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
10518                 return;
10519
10520         /* delay mechanism from ixgbe_disable_rx */
10521         delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
10522
10523         wait_loop = IXGBE_MAX_RX_DESC_POLL;
10524         wait_delay = delay_interval;
10525
10526         while (wait_loop--) {
10527                 usleep_range(wait_delay, wait_delay + 10);
10528                 wait_delay += delay_interval * 2;
10529                 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
10530
10531                 if (!(rxdctl & IXGBE_RXDCTL_ENABLE))
10532                         return;
10533         }
10534
10535         e_err(drv, "RXDCTL.ENABLE not cleared within the polling period\n");
10536 }
10537
10538 static void ixgbe_reset_txr_stats(struct ixgbe_ring *tx_ring)
10539 {
10540         memset(&tx_ring->stats, 0, sizeof(tx_ring->stats));
10541         memset(&tx_ring->tx_stats, 0, sizeof(tx_ring->tx_stats));
10542 }
10543
10544 static void ixgbe_reset_rxr_stats(struct ixgbe_ring *rx_ring)
10545 {
10546         memset(&rx_ring->stats, 0, sizeof(rx_ring->stats));
10547         memset(&rx_ring->rx_stats, 0, sizeof(rx_ring->rx_stats));
10548 }
10549
10550 /**
10551  * ixgbe_irq_disable_single - Disable single IRQ vector
10552  * @adapter: adapter structure
10553  * @ring: ring index
10554  **/
10555 static void ixgbe_irq_disable_single(struct ixgbe_adapter *adapter, u32 ring)
10556 {
10557         struct ixgbe_hw *hw = &adapter->hw;
10558         u64 qmask = BIT_ULL(ring);
10559         u32 mask;
10560
10561         switch (adapter->hw.mac.type) {
10562         case ixgbe_mac_82598EB:
10563                 mask = qmask & IXGBE_EIMC_RTX_QUEUE;
10564                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, mask);
10565                 break;
10566         case ixgbe_mac_82599EB:
10567         case ixgbe_mac_X540:
10568         case ixgbe_mac_X550:
10569         case ixgbe_mac_X550EM_x:
10570         case ixgbe_mac_x550em_a:
10571                 mask = (qmask & 0xFFFFFFFF);
10572                 if (mask)
10573                         IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
10574                 mask = (qmask >> 32);
10575                 if (mask)
10576                         IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
10577                 break;
10578         default:
10579                 break;
10580         }
10581         IXGBE_WRITE_FLUSH(&adapter->hw);
10582         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
10583                 synchronize_irq(adapter->msix_entries[ring].vector);
10584         else
10585                 synchronize_irq(adapter->pdev->irq);
10586 }
10587
10588 /**
10589  * ixgbe_txrx_ring_disable - Disable Rx/Tx/XDP Tx rings
10590  * @adapter: adapter structure
10591  * @ring: ring index
10592  *
10593  * This function disables a certain Rx/Tx/XDP Tx ring. The function
10594  * assumes that the netdev is running.
10595  **/
10596 void ixgbe_txrx_ring_disable(struct ixgbe_adapter *adapter, int ring)
10597 {
10598         struct ixgbe_ring *rx_ring, *tx_ring, *xdp_ring;
10599
10600         rx_ring = adapter->rx_ring[ring];
10601         tx_ring = adapter->tx_ring[ring];
10602         xdp_ring = adapter->xdp_ring[ring];
10603
10604         ixgbe_irq_disable_single(adapter, ring);
10605
10606         /* Rx/Tx/XDP Tx share the same napi context. */
10607         napi_disable(&rx_ring->q_vector->napi);
10608
10609         ixgbe_disable_txr(adapter, tx_ring);
10610         if (xdp_ring)
10611                 ixgbe_disable_txr(adapter, xdp_ring);
10612         ixgbe_disable_rxr_hw(adapter, rx_ring);
10613
10614         if (xdp_ring)
10615                 synchronize_rcu();
10616
10617         ixgbe_clean_tx_ring(tx_ring);
10618         if (xdp_ring)
10619                 ixgbe_clean_tx_ring(xdp_ring);
10620         ixgbe_clean_rx_ring(rx_ring);
10621
10622         ixgbe_reset_txr_stats(tx_ring);
10623         if (xdp_ring)
10624                 ixgbe_reset_txr_stats(xdp_ring);
10625         ixgbe_reset_rxr_stats(rx_ring);
10626 }
10627
10628 /**
10629  * ixgbe_txrx_ring_enable - Enable Rx/Tx/XDP Tx rings
10630  * @adapter: adapter structure
10631  * @ring: ring index
10632  *
10633  * This function enables a certain Rx/Tx/XDP Tx ring. The function
10634  * assumes that the netdev is running.
10635  **/
10636 void ixgbe_txrx_ring_enable(struct ixgbe_adapter *adapter, int ring)
10637 {
10638         struct ixgbe_ring *rx_ring, *tx_ring, *xdp_ring;
10639
10640         rx_ring = adapter->rx_ring[ring];
10641         tx_ring = adapter->tx_ring[ring];
10642         xdp_ring = adapter->xdp_ring[ring];
10643
10644         ixgbe_configure_tx_ring(adapter, tx_ring);
10645         if (xdp_ring)
10646                 ixgbe_configure_tx_ring(adapter, xdp_ring);
10647         ixgbe_configure_rx_ring(adapter, rx_ring);
10648
10649         clear_bit(__IXGBE_TX_DISABLED, &tx_ring->state);
10650         if (xdp_ring)
10651                 clear_bit(__IXGBE_TX_DISABLED, &xdp_ring->state);
10652
10653         /* Rx/Tx/XDP Tx share the same napi context. */
10654         napi_enable(&rx_ring->q_vector->napi);
10655         ixgbe_irq_enable_queues(adapter, BIT_ULL(ring));
10656         IXGBE_WRITE_FLUSH(&adapter->hw);
10657 }
10658
10659 /**
10660  * ixgbe_enumerate_functions - Get the number of ports this device has
10661  * @adapter: adapter structure
10662  *
10663  * This function enumerates the phsyical functions co-located on a single slot,
10664  * in order to determine how many ports a device has. This is most useful in
10665  * determining the required GT/s of PCIe bandwidth necessary for optimal
10666  * performance.
10667  **/
10668 static inline int ixgbe_enumerate_functions(struct ixgbe_adapter *adapter)
10669 {
10670         struct pci_dev *entry, *pdev = adapter->pdev;
10671         int physfns = 0;
10672
10673         /* Some cards can not use the generic count PCIe functions method,
10674          * because they are behind a parent switch, so we hardcode these with
10675          * the correct number of functions.
10676          */
10677         if (ixgbe_pcie_from_parent(&adapter->hw))
10678                 physfns = 4;
10679
10680         list_for_each_entry(entry, &adapter->pdev->bus->devices, bus_list) {
10681                 /* don't count virtual functions */
10682                 if (entry->is_virtfn)
10683                         continue;
10684
10685                 /* When the devices on the bus don't all match our device ID,
10686                  * we can't reliably determine the correct number of
10687                  * functions. This can occur if a function has been direct
10688                  * attached to a virtual machine using VT-d, for example. In
10689                  * this case, simply return -1 to indicate this.
10690                  */
10691                 if ((entry->vendor != pdev->vendor) ||
10692                     (entry->device != pdev->device))
10693                         return -1;
10694
10695                 physfns++;
10696         }
10697
10698         return physfns;
10699 }
10700
10701 /**
10702  * ixgbe_wol_supported - Check whether device supports WoL
10703  * @adapter: the adapter private structure
10704  * @device_id: the device ID
10705  * @subdevice_id: the subsystem device ID
10706  *
10707  * This function is used by probe and ethtool to determine
10708  * which devices have WoL support
10709  *
10710  **/
10711 bool ixgbe_wol_supported(struct ixgbe_adapter *adapter, u16 device_id,
10712                          u16 subdevice_id)
10713 {
10714         struct ixgbe_hw *hw = &adapter->hw;
10715         u16 wol_cap = adapter->eeprom_cap & IXGBE_DEVICE_CAPS_WOL_MASK;
10716
10717         /* WOL not supported on 82598 */
10718         if (hw->mac.type == ixgbe_mac_82598EB)
10719                 return false;
10720
10721         /* check eeprom to see if WOL is enabled for X540 and newer */
10722         if (hw->mac.type >= ixgbe_mac_X540) {
10723                 if ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0_1) ||
10724                     ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0) &&
10725                      (hw->bus.func == 0)))
10726                         return true;
10727         }
10728
10729         /* WOL is determined based on device IDs for 82599 MACs */
10730         switch (device_id) {
10731         case IXGBE_DEV_ID_82599_SFP:
10732                 /* Only these subdevices could supports WOL */
10733                 switch (subdevice_id) {
10734                 case IXGBE_SUBDEV_ID_82599_560FLR:
10735                 case IXGBE_SUBDEV_ID_82599_LOM_SNAP6:
10736                 case IXGBE_SUBDEV_ID_82599_SFP_WOL0:
10737                 case IXGBE_SUBDEV_ID_82599_SFP_2OCP:
10738                         /* only support first port */
10739                         if (hw->bus.func != 0)
10740                                 break;
10741                         fallthrough;
10742                 case IXGBE_SUBDEV_ID_82599_SP_560FLR:
10743                 case IXGBE_SUBDEV_ID_82599_SFP:
10744                 case IXGBE_SUBDEV_ID_82599_RNDC:
10745                 case IXGBE_SUBDEV_ID_82599_ECNA_DP:
10746                 case IXGBE_SUBDEV_ID_82599_SFP_1OCP:
10747                 case IXGBE_SUBDEV_ID_82599_SFP_LOM_OEM1:
10748                 case IXGBE_SUBDEV_ID_82599_SFP_LOM_OEM2:
10749                         return true;
10750                 }
10751                 break;
10752         case IXGBE_DEV_ID_82599EN_SFP:
10753                 /* Only these subdevices support WOL */
10754                 switch (subdevice_id) {
10755                 case IXGBE_SUBDEV_ID_82599EN_SFP_OCP1:
10756                         return true;
10757                 }
10758                 break;
10759         case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
10760                 /* All except this subdevice support WOL */
10761                 if (subdevice_id != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
10762                         return true;
10763                 break;
10764         case IXGBE_DEV_ID_82599_KX4:
10765                 return  true;
10766         default:
10767                 break;
10768         }
10769
10770         return false;
10771 }
10772
10773 /**
10774  * ixgbe_set_fw_version - Set FW version
10775  * @adapter: the adapter private structure
10776  *
10777  * This function is used by probe and ethtool to determine the FW version to
10778  * format to display. The FW version is taken from the EEPROM/NVM.
10779  */
10780 static void ixgbe_set_fw_version(struct ixgbe_adapter *adapter)
10781 {
10782         struct ixgbe_hw *hw = &adapter->hw;
10783         struct ixgbe_nvm_version nvm_ver;
10784
10785         ixgbe_get_oem_prod_version(hw, &nvm_ver);
10786         if (nvm_ver.oem_valid) {
10787                 snprintf(adapter->eeprom_id, sizeof(adapter->eeprom_id),
10788                          "%x.%x.%x", nvm_ver.oem_major, nvm_ver.oem_minor,
10789                          nvm_ver.oem_release);
10790                 return;
10791         }
10792
10793         ixgbe_get_etk_id(hw, &nvm_ver);
10794         ixgbe_get_orom_version(hw, &nvm_ver);
10795
10796         if (nvm_ver.or_valid) {
10797                 snprintf(adapter->eeprom_id, sizeof(adapter->eeprom_id),
10798                          "0x%08x, %d.%d.%d", nvm_ver.etk_id, nvm_ver.or_major,
10799                          nvm_ver.or_build, nvm_ver.or_patch);
10800                 return;
10801         }
10802
10803         /* Set ETrack ID format */
10804         snprintf(adapter->eeprom_id, sizeof(adapter->eeprom_id),
10805                  "0x%08x", nvm_ver.etk_id);
10806 }
10807
10808 /**
10809  * ixgbe_probe - Device Initialization Routine
10810  * @pdev: PCI device information struct
10811  * @ent: entry in ixgbe_pci_tbl
10812  *
10813  * Returns 0 on success, negative on failure
10814  *
10815  * ixgbe_probe initializes an adapter identified by a pci_dev structure.
10816  * The OS initialization, configuring of the adapter private structure,
10817  * and a hardware reset occur.
10818  **/
10819 static int ixgbe_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
10820 {
10821         struct net_device *netdev;
10822         struct ixgbe_adapter *adapter = NULL;
10823         struct ixgbe_hw *hw;
10824         const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
10825         unsigned int indices = MAX_TX_QUEUES;
10826         u8 part_str[IXGBE_PBANUM_LENGTH];
10827         int i, err, expected_gts;
10828         bool disable_dev = false;
10829 #ifdef IXGBE_FCOE
10830         u16 device_caps;
10831 #endif
10832         u32 eec;
10833
10834         /* Catch broken hardware that put the wrong VF device ID in
10835          * the PCIe SR-IOV capability.
10836          */
10837         if (pdev->is_virtfn) {
10838                 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
10839                      pci_name(pdev), pdev->vendor, pdev->device);
10840                 return -EINVAL;
10841         }
10842
10843         err = pci_enable_device_mem(pdev);
10844         if (err)
10845                 return err;
10846
10847         err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
10848         if (err) {
10849                 dev_err(&pdev->dev,
10850                         "No usable DMA configuration, aborting\n");
10851                 goto err_dma;
10852         }
10853
10854         err = pci_request_mem_regions(pdev, ixgbe_driver_name);
10855         if (err) {
10856                 dev_err(&pdev->dev,
10857                         "pci_request_selected_regions failed 0x%x\n", err);
10858                 goto err_pci_reg;
10859         }
10860
10861         pci_set_master(pdev);
10862         pci_save_state(pdev);
10863
10864         if (ii->mac == ixgbe_mac_82598EB) {
10865 #ifdef CONFIG_IXGBE_DCB
10866                 /* 8 TC w/ 4 queues per TC */
10867                 indices = 4 * MAX_TRAFFIC_CLASS;
10868 #else
10869                 indices = IXGBE_MAX_RSS_INDICES;
10870 #endif
10871         }
10872
10873         netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
10874         if (!netdev) {
10875                 err = -ENOMEM;
10876                 goto err_alloc_etherdev;
10877         }
10878
10879         SET_NETDEV_DEV(netdev, &pdev->dev);
10880
10881         adapter = netdev_priv(netdev);
10882
10883         adapter->netdev = netdev;
10884         adapter->pdev = pdev;
10885         hw = &adapter->hw;
10886         hw->back = adapter;
10887         adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
10888
10889         hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
10890                               pci_resource_len(pdev, 0));
10891         adapter->io_addr = hw->hw_addr;
10892         if (!hw->hw_addr) {
10893                 err = -EIO;
10894                 goto err_ioremap;
10895         }
10896
10897         netdev->netdev_ops = &ixgbe_netdev_ops;
10898         ixgbe_set_ethtool_ops(netdev);
10899         netdev->watchdog_timeo = 5 * HZ;
10900         strscpy(netdev->name, pci_name(pdev), sizeof(netdev->name));
10901
10902         /* Setup hw api */
10903         hw->mac.ops   = *ii->mac_ops;
10904         hw->mac.type  = ii->mac;
10905         hw->mvals     = ii->mvals;
10906         if (ii->link_ops)
10907                 hw->link.ops  = *ii->link_ops;
10908
10909         /* EEPROM */
10910         hw->eeprom.ops = *ii->eeprom_ops;
10911         eec = IXGBE_READ_REG(hw, IXGBE_EEC(hw));
10912         if (ixgbe_removed(hw->hw_addr)) {
10913                 err = -EIO;
10914                 goto err_ioremap;
10915         }
10916         /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
10917         if (!(eec & BIT(8)))
10918                 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
10919
10920         /* PHY */
10921         hw->phy.ops = *ii->phy_ops;
10922         hw->phy.sfp_type = ixgbe_sfp_type_unknown;
10923         /* ixgbe_identify_phy_generic will set prtad and mmds properly */
10924         hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
10925         hw->phy.mdio.mmds = 0;
10926         hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
10927         hw->phy.mdio.dev = netdev;
10928         hw->phy.mdio.mdio_read = ixgbe_mdio_read;
10929         hw->phy.mdio.mdio_write = ixgbe_mdio_write;
10930
10931         /* setup the private structure */
10932         err = ixgbe_sw_init(adapter, ii);
10933         if (err)
10934                 goto err_sw_init;
10935
10936         if (adapter->hw.mac.type == ixgbe_mac_82599EB)
10937                 adapter->flags2 |= IXGBE_FLAG2_AUTO_DISABLE_VF;
10938
10939         switch (adapter->hw.mac.type) {
10940         case ixgbe_mac_X550:
10941         case ixgbe_mac_X550EM_x:
10942                 netdev->udp_tunnel_nic_info = &ixgbe_udp_tunnels_x550;
10943                 break;
10944         case ixgbe_mac_x550em_a:
10945                 netdev->udp_tunnel_nic_info = &ixgbe_udp_tunnels_x550em_a;
10946                 break;
10947         default:
10948                 break;
10949         }
10950
10951         /* Make sure the SWFW semaphore is in a valid state */
10952         if (hw->mac.ops.init_swfw_sync)
10953                 hw->mac.ops.init_swfw_sync(hw);
10954
10955         /* Make it possible the adapter to be woken up via WOL */
10956         switch (adapter->hw.mac.type) {
10957         case ixgbe_mac_82599EB:
10958         case ixgbe_mac_X540:
10959         case ixgbe_mac_X550:
10960         case ixgbe_mac_X550EM_x:
10961         case ixgbe_mac_x550em_a:
10962                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
10963                 break;
10964         default:
10965                 break;
10966         }
10967
10968         /*
10969          * If there is a fan on this device and it has failed log the
10970          * failure.
10971          */
10972         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
10973                 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
10974                 if (esdp & IXGBE_ESDP_SDP1)
10975                         e_crit(probe, "Fan has stopped, replace the adapter\n");
10976         }
10977
10978         if (allow_unsupported_sfp)
10979                 hw->allow_unsupported_sfp = allow_unsupported_sfp;
10980
10981         /* reset_hw fills in the perm_addr as well */
10982         hw->phy.reset_if_overtemp = true;
10983         err = hw->mac.ops.reset_hw(hw);
10984         hw->phy.reset_if_overtemp = false;
10985         ixgbe_set_eee_capable(adapter);
10986         if (err == -ENOENT) {
10987                 err = 0;
10988         } else if (err == -EOPNOTSUPP) {
10989                 e_dev_err("failed to load because an unsupported SFP+ or QSFP module type was detected.\n");
10990                 e_dev_err("Reload the driver after installing a supported module.\n");
10991                 goto err_sw_init;
10992         } else if (err) {
10993                 e_dev_err("HW Init failed: %d\n", err);
10994                 goto err_sw_init;
10995         }
10996
10997 #ifdef CONFIG_PCI_IOV
10998         /* SR-IOV not supported on the 82598 */
10999         if (adapter->hw.mac.type == ixgbe_mac_82598EB)
11000                 goto skip_sriov;
11001         /* Mailbox */
11002         ixgbe_init_mbx_params_pf(hw);
11003         hw->mbx.ops = ii->mbx_ops;
11004         pci_sriov_set_totalvfs(pdev, IXGBE_MAX_VFS_DRV_LIMIT);
11005         ixgbe_enable_sriov(adapter, max_vfs);
11006 skip_sriov:
11007
11008 #endif
11009         netdev->features = NETIF_F_SG |
11010                            NETIF_F_TSO |
11011                            NETIF_F_TSO6 |
11012                            NETIF_F_RXHASH |
11013                            NETIF_F_RXCSUM |
11014                            NETIF_F_HW_CSUM;
11015
11016 #define IXGBE_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
11017                                     NETIF_F_GSO_GRE_CSUM | \
11018                                     NETIF_F_GSO_IPXIP4 | \
11019                                     NETIF_F_GSO_IPXIP6 | \
11020                                     NETIF_F_GSO_UDP_TUNNEL | \
11021                                     NETIF_F_GSO_UDP_TUNNEL_CSUM)
11022
11023         netdev->gso_partial_features = IXGBE_GSO_PARTIAL_FEATURES;
11024         netdev->features |= NETIF_F_GSO_PARTIAL |
11025                             IXGBE_GSO_PARTIAL_FEATURES;
11026
11027         if (hw->mac.type >= ixgbe_mac_82599EB)
11028                 netdev->features |= NETIF_F_SCTP_CRC | NETIF_F_GSO_UDP_L4;
11029
11030 #ifdef CONFIG_IXGBE_IPSEC
11031 #define IXGBE_ESP_FEATURES      (NETIF_F_HW_ESP | \
11032                                  NETIF_F_HW_ESP_TX_CSUM | \
11033                                  NETIF_F_GSO_ESP)
11034
11035         if (adapter->ipsec)
11036                 netdev->features |= IXGBE_ESP_FEATURES;
11037 #endif
11038         /* copy netdev features into list of user selectable features */
11039         netdev->hw_features |= netdev->features |
11040                                NETIF_F_HW_VLAN_CTAG_FILTER |
11041                                NETIF_F_HW_VLAN_CTAG_RX |
11042                                NETIF_F_HW_VLAN_CTAG_TX |
11043                                NETIF_F_RXALL |
11044                                NETIF_F_HW_L2FW_DOFFLOAD;
11045
11046         if (hw->mac.type >= ixgbe_mac_82599EB)
11047                 netdev->hw_features |= NETIF_F_NTUPLE |
11048                                        NETIF_F_HW_TC;
11049
11050         netdev->features |= NETIF_F_HIGHDMA;
11051
11052         netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
11053         netdev->hw_enc_features |= netdev->vlan_features;
11054         netdev->mpls_features |= NETIF_F_SG |
11055                                  NETIF_F_TSO |
11056                                  NETIF_F_TSO6 |
11057                                  NETIF_F_HW_CSUM;
11058         netdev->mpls_features |= IXGBE_GSO_PARTIAL_FEATURES;
11059
11060         /* set this bit last since it cannot be part of vlan_features */
11061         netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
11062                             NETIF_F_HW_VLAN_CTAG_RX |
11063                             NETIF_F_HW_VLAN_CTAG_TX;
11064
11065         netdev->priv_flags |= IFF_UNICAST_FLT;
11066         netdev->priv_flags |= IFF_SUPP_NOFCS;
11067
11068         netdev->xdp_features = NETDEV_XDP_ACT_BASIC | NETDEV_XDP_ACT_REDIRECT |
11069                                NETDEV_XDP_ACT_XSK_ZEROCOPY;
11070
11071         /* MTU range: 68 - 9710 */
11072         netdev->min_mtu = ETH_MIN_MTU;
11073         netdev->max_mtu = IXGBE_MAX_JUMBO_FRAME_SIZE - (ETH_HLEN + ETH_FCS_LEN);
11074
11075 #ifdef CONFIG_IXGBE_DCB
11076         if (adapter->flags & IXGBE_FLAG_DCB_CAPABLE)
11077                 netdev->dcbnl_ops = &ixgbe_dcbnl_ops;
11078 #endif
11079
11080 #ifdef IXGBE_FCOE
11081         if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
11082                 unsigned int fcoe_l;
11083
11084                 if (hw->mac.ops.get_device_caps) {
11085                         hw->mac.ops.get_device_caps(hw, &device_caps);
11086                         if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
11087                                 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
11088                 }
11089
11090
11091                 fcoe_l = min_t(int, IXGBE_FCRETA_SIZE, num_online_cpus());
11092                 adapter->ring_feature[RING_F_FCOE].limit = fcoe_l;
11093
11094                 netdev->features |= NETIF_F_FSO |
11095                                     NETIF_F_FCOE_CRC;
11096
11097                 netdev->vlan_features |= NETIF_F_FSO |
11098                                          NETIF_F_FCOE_CRC;
11099         }
11100 #endif /* IXGBE_FCOE */
11101         if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
11102                 netdev->hw_features |= NETIF_F_LRO;
11103         if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
11104                 netdev->features |= NETIF_F_LRO;
11105
11106         if (ixgbe_check_fw_error(adapter)) {
11107                 err = -EIO;
11108                 goto err_sw_init;
11109         }
11110
11111         /* make sure the EEPROM is good */
11112         if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
11113                 e_dev_err("The EEPROM Checksum Is Not Valid\n");
11114                 err = -EIO;
11115                 goto err_sw_init;
11116         }
11117
11118         eth_platform_get_mac_address(&adapter->pdev->dev,
11119                                      adapter->hw.mac.perm_addr);
11120
11121         eth_hw_addr_set(netdev, hw->mac.perm_addr);
11122
11123         if (!is_valid_ether_addr(netdev->dev_addr)) {
11124                 e_dev_err("invalid MAC address\n");
11125                 err = -EIO;
11126                 goto err_sw_init;
11127         }
11128
11129         /* Set hw->mac.addr to permanent MAC address */
11130         ether_addr_copy(hw->mac.addr, hw->mac.perm_addr);
11131         ixgbe_mac_set_default_filter(adapter);
11132
11133         timer_setup(&adapter->service_timer, ixgbe_service_timer, 0);
11134
11135         if (ixgbe_removed(hw->hw_addr)) {
11136                 err = -EIO;
11137                 goto err_sw_init;
11138         }
11139         INIT_WORK(&adapter->service_task, ixgbe_service_task);
11140         set_bit(__IXGBE_SERVICE_INITED, &adapter->state);
11141         clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
11142
11143         err = ixgbe_init_interrupt_scheme(adapter);
11144         if (err)
11145                 goto err_sw_init;
11146
11147         for (i = 0; i < adapter->num_rx_queues; i++)
11148                 u64_stats_init(&adapter->rx_ring[i]->syncp);
11149         for (i = 0; i < adapter->num_tx_queues; i++)
11150                 u64_stats_init(&adapter->tx_ring[i]->syncp);
11151         for (i = 0; i < adapter->num_xdp_queues; i++)
11152                 u64_stats_init(&adapter->xdp_ring[i]->syncp);
11153
11154         /* WOL not supported for all devices */
11155         adapter->wol = 0;
11156         hw->eeprom.ops.read(hw, 0x2c, &adapter->eeprom_cap);
11157         hw->wol_enabled = ixgbe_wol_supported(adapter, pdev->device,
11158                                                 pdev->subsystem_device);
11159         if (hw->wol_enabled)
11160                 adapter->wol = IXGBE_WUFC_MAG;
11161
11162         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
11163
11164         /* save off EEPROM version number */
11165         ixgbe_set_fw_version(adapter);
11166
11167         /* pick up the PCI bus settings for reporting later */
11168         if (ixgbe_pcie_from_parent(hw))
11169                 ixgbe_get_parent_bus_info(adapter);
11170         else
11171                  hw->mac.ops.get_bus_info(hw);
11172
11173         /* calculate the expected PCIe bandwidth required for optimal
11174          * performance. Note that some older parts will never have enough
11175          * bandwidth due to being older generation PCIe parts. We clamp these
11176          * parts to ensure no warning is displayed if it can't be fixed.
11177          */
11178         switch (hw->mac.type) {
11179         case ixgbe_mac_82598EB:
11180                 expected_gts = min(ixgbe_enumerate_functions(adapter) * 10, 16);
11181                 break;
11182         default:
11183                 expected_gts = ixgbe_enumerate_functions(adapter) * 10;
11184                 break;
11185         }
11186
11187         /* don't check link if we failed to enumerate functions */
11188         if (expected_gts > 0)
11189                 ixgbe_check_minimum_link(adapter, expected_gts);
11190
11191         err = ixgbe_read_pba_string_generic(hw, part_str, sizeof(part_str));
11192         if (err)
11193                 strscpy(part_str, "Unknown", sizeof(part_str));
11194         if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
11195                 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
11196                            hw->mac.type, hw->phy.type, hw->phy.sfp_type,
11197                            part_str);
11198         else
11199                 e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
11200                            hw->mac.type, hw->phy.type, part_str);
11201
11202         e_dev_info("%pM\n", netdev->dev_addr);
11203
11204         /* reset the hardware with the new settings */
11205         err = hw->mac.ops.start_hw(hw);
11206         if (err == -EACCES) {
11207                 /* We are running on a pre-production device, log a warning */
11208                 e_dev_warn("This device is a pre-production adapter/LOM. "
11209                            "Please be aware there may be issues associated "
11210                            "with your hardware.  If you are experiencing "
11211                            "problems please contact your Intel or hardware "
11212                            "representative who provided you with this "
11213                            "hardware.\n");
11214         }
11215         strcpy(netdev->name, "eth%d");
11216         pci_set_drvdata(pdev, adapter);
11217         err = register_netdev(netdev);
11218         if (err)
11219                 goto err_register;
11220
11221
11222         /* power down the optics for 82599 SFP+ fiber */
11223         if (hw->mac.ops.disable_tx_laser)
11224                 hw->mac.ops.disable_tx_laser(hw);
11225
11226         /* carrier off reporting is important to ethtool even BEFORE open */
11227         netif_carrier_off(netdev);
11228
11229 #ifdef CONFIG_IXGBE_DCA
11230         if (dca_add_requester(&pdev->dev) == 0) {
11231                 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
11232                 ixgbe_setup_dca(adapter);
11233         }
11234 #endif
11235         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
11236                 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
11237                 for (i = 0; i < adapter->num_vfs; i++)
11238                         ixgbe_vf_configuration(pdev, (i | 0x10000000));
11239         }
11240
11241         /* firmware requires driver version to be 0xFFFFFFFF
11242          * since os does not support feature
11243          */
11244         if (hw->mac.ops.set_fw_drv_ver)
11245                 hw->mac.ops.set_fw_drv_ver(hw, 0xFF, 0xFF, 0xFF, 0xFF,
11246                                            sizeof(UTS_RELEASE) - 1,
11247                                            UTS_RELEASE);
11248
11249         /* add san mac addr to netdev */
11250         ixgbe_add_sanmac_netdev(netdev);
11251
11252         e_dev_info("%s\n", ixgbe_default_device_descr);
11253
11254 #ifdef CONFIG_IXGBE_HWMON
11255         if (ixgbe_sysfs_init(adapter))
11256                 e_err(probe, "failed to allocate sysfs resources\n");
11257 #endif /* CONFIG_IXGBE_HWMON */
11258
11259         ixgbe_dbg_adapter_init(adapter);
11260
11261         /* setup link for SFP devices with MNG FW, else wait for IXGBE_UP */
11262         if (ixgbe_mng_enabled(hw) && ixgbe_is_sfp(hw) && hw->mac.ops.setup_link)
11263                 hw->mac.ops.setup_link(hw,
11264                         IXGBE_LINK_SPEED_10GB_FULL | IXGBE_LINK_SPEED_1GB_FULL,
11265                         true);
11266
11267         err = ixgbe_mii_bus_init(hw);
11268         if (err)
11269                 goto err_netdev;
11270
11271         return 0;
11272
11273 err_netdev:
11274         unregister_netdev(netdev);
11275 err_register:
11276         ixgbe_release_hw_control(adapter);
11277         ixgbe_clear_interrupt_scheme(adapter);
11278 err_sw_init:
11279         ixgbe_disable_sriov(adapter);
11280         adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
11281         iounmap(adapter->io_addr);
11282         kfree(adapter->jump_tables[0]);
11283         kfree(adapter->mac_table);
11284         kfree(adapter->rss_key);
11285         bitmap_free(adapter->af_xdp_zc_qps);
11286 err_ioremap:
11287         disable_dev = !test_and_set_bit(__IXGBE_DISABLED, &adapter->state);
11288         free_netdev(netdev);
11289 err_alloc_etherdev:
11290         pci_release_mem_regions(pdev);
11291 err_pci_reg:
11292 err_dma:
11293         if (!adapter || disable_dev)
11294                 pci_disable_device(pdev);
11295         return err;
11296 }
11297
11298 /**
11299  * ixgbe_remove - Device Removal Routine
11300  * @pdev: PCI device information struct
11301  *
11302  * ixgbe_remove is called by the PCI subsystem to alert the driver
11303  * that it should release a PCI device.  The could be caused by a
11304  * Hot-Plug event, or because the driver is going to be removed from
11305  * memory.
11306  **/
11307 static void ixgbe_remove(struct pci_dev *pdev)
11308 {
11309         struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11310         struct net_device *netdev;
11311         bool disable_dev;
11312         int i;
11313
11314         /* if !adapter then we already cleaned up in probe */
11315         if (!adapter)
11316                 return;
11317
11318         netdev  = adapter->netdev;
11319         ixgbe_dbg_adapter_exit(adapter);
11320
11321         set_bit(__IXGBE_REMOVING, &adapter->state);
11322         cancel_work_sync(&adapter->service_task);
11323
11324         if (adapter->mii_bus)
11325                 mdiobus_unregister(adapter->mii_bus);
11326
11327 #ifdef CONFIG_IXGBE_DCA
11328         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
11329                 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
11330                 dca_remove_requester(&pdev->dev);
11331                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
11332                                 IXGBE_DCA_CTRL_DCA_DISABLE);
11333         }
11334
11335 #endif
11336 #ifdef CONFIG_IXGBE_HWMON
11337         ixgbe_sysfs_exit(adapter);
11338 #endif /* CONFIG_IXGBE_HWMON */
11339
11340         /* remove the added san mac */
11341         ixgbe_del_sanmac_netdev(netdev);
11342
11343 #ifdef CONFIG_PCI_IOV
11344         ixgbe_disable_sriov(adapter);
11345 #endif
11346         if (netdev->reg_state == NETREG_REGISTERED)
11347                 unregister_netdev(netdev);
11348
11349         ixgbe_stop_ipsec_offload(adapter);
11350         ixgbe_clear_interrupt_scheme(adapter);
11351
11352         ixgbe_release_hw_control(adapter);
11353
11354 #ifdef CONFIG_DCB
11355         kfree(adapter->ixgbe_ieee_pfc);
11356         kfree(adapter->ixgbe_ieee_ets);
11357
11358 #endif
11359         iounmap(adapter->io_addr);
11360         pci_release_mem_regions(pdev);
11361
11362         e_dev_info("complete\n");
11363
11364         for (i = 0; i < IXGBE_MAX_LINK_HANDLE; i++) {
11365                 if (adapter->jump_tables[i]) {
11366                         kfree(adapter->jump_tables[i]->input);
11367                         kfree(adapter->jump_tables[i]->mask);
11368                 }
11369                 kfree(adapter->jump_tables[i]);
11370         }
11371
11372         kfree(adapter->mac_table);
11373         kfree(adapter->rss_key);
11374         bitmap_free(adapter->af_xdp_zc_qps);
11375         disable_dev = !test_and_set_bit(__IXGBE_DISABLED, &adapter->state);
11376         free_netdev(netdev);
11377
11378         if (disable_dev)
11379                 pci_disable_device(pdev);
11380 }
11381
11382 /**
11383  * ixgbe_io_error_detected - called when PCI error is detected
11384  * @pdev: Pointer to PCI device
11385  * @state: The current pci connection state
11386  *
11387  * This function is called after a PCI bus error affecting
11388  * this device has been detected.
11389  */
11390 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
11391                                                 pci_channel_state_t state)
11392 {
11393         struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11394         struct net_device *netdev = adapter->netdev;
11395
11396 #ifdef CONFIG_PCI_IOV
11397         struct ixgbe_hw *hw = &adapter->hw;
11398         struct pci_dev *bdev, *vfdev;
11399         u32 dw0, dw1, dw2, dw3;
11400         int vf, pos;
11401         u16 req_id, pf_func;
11402
11403         if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
11404             adapter->num_vfs == 0)
11405                 goto skip_bad_vf_detection;
11406
11407         bdev = pdev->bus->self;
11408         while (bdev && (pci_pcie_type(bdev) != PCI_EXP_TYPE_ROOT_PORT))
11409                 bdev = bdev->bus->self;
11410
11411         if (!bdev)
11412                 goto skip_bad_vf_detection;
11413
11414         pos = pci_find_ext_capability(bdev, PCI_EXT_CAP_ID_ERR);
11415         if (!pos)
11416                 goto skip_bad_vf_detection;
11417
11418         dw0 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG);
11419         dw1 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 4);
11420         dw2 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 8);
11421         dw3 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 12);
11422         if (ixgbe_removed(hw->hw_addr))
11423                 goto skip_bad_vf_detection;
11424
11425         req_id = dw1 >> 16;
11426         /* On the 82599 if bit 7 of the requestor ID is set then it's a VF */
11427         if (!(req_id & 0x0080))
11428                 goto skip_bad_vf_detection;
11429
11430         pf_func = req_id & 0x01;
11431         if ((pf_func & 1) == (pdev->devfn & 1)) {
11432                 unsigned int device_id;
11433
11434                 vf = FIELD_GET(0x7F, req_id);
11435                 e_dev_err("VF %d has caused a PCIe error\n", vf);
11436                 e_dev_err("TLP: dw0: %8.8x\tdw1: %8.8x\tdw2: "
11437                                 "%8.8x\tdw3: %8.8x\n",
11438                 dw0, dw1, dw2, dw3);
11439                 switch (adapter->hw.mac.type) {
11440                 case ixgbe_mac_82599EB:
11441                         device_id = IXGBE_82599_VF_DEVICE_ID;
11442                         break;
11443                 case ixgbe_mac_X540:
11444                         device_id = IXGBE_X540_VF_DEVICE_ID;
11445                         break;
11446                 case ixgbe_mac_X550:
11447                         device_id = IXGBE_DEV_ID_X550_VF;
11448                         break;
11449                 case ixgbe_mac_X550EM_x:
11450                         device_id = IXGBE_DEV_ID_X550EM_X_VF;
11451                         break;
11452                 case ixgbe_mac_x550em_a:
11453                         device_id = IXGBE_DEV_ID_X550EM_A_VF;
11454                         break;
11455                 default:
11456                         device_id = 0;
11457                         break;
11458                 }
11459
11460                 /* Find the pci device of the offending VF */
11461                 vfdev = pci_get_device(PCI_VENDOR_ID_INTEL, device_id, NULL);
11462                 while (vfdev) {
11463                         if (vfdev->devfn == (req_id & 0xFF))
11464                                 break;
11465                         vfdev = pci_get_device(PCI_VENDOR_ID_INTEL,
11466                                                device_id, vfdev);
11467                 }
11468                 /*
11469                  * There's a slim chance the VF could have been hot plugged,
11470                  * so if it is no longer present we don't need to issue the
11471                  * VFLR.  Just clean up the AER in that case.
11472                  */
11473                 if (vfdev) {
11474                         pcie_flr(vfdev);
11475                         /* Free device reference count */
11476                         pci_dev_put(vfdev);
11477                 }
11478         }
11479
11480         /*
11481          * Even though the error may have occurred on the other port
11482          * we still need to increment the vf error reference count for
11483          * both ports because the I/O resume function will be called
11484          * for both of them.
11485          */
11486         adapter->vferr_refcount++;
11487
11488         return PCI_ERS_RESULT_RECOVERED;
11489
11490 skip_bad_vf_detection:
11491 #endif /* CONFIG_PCI_IOV */
11492         if (!test_bit(__IXGBE_SERVICE_INITED, &adapter->state))
11493                 return PCI_ERS_RESULT_DISCONNECT;
11494
11495         if (!netif_device_present(netdev))
11496                 return PCI_ERS_RESULT_DISCONNECT;
11497
11498         rtnl_lock();
11499         netif_device_detach(netdev);
11500
11501         if (netif_running(netdev))
11502                 ixgbe_close_suspend(adapter);
11503
11504         if (state == pci_channel_io_perm_failure) {
11505                 rtnl_unlock();
11506                 return PCI_ERS_RESULT_DISCONNECT;
11507         }
11508
11509         if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state))
11510                 pci_disable_device(pdev);
11511         rtnl_unlock();
11512
11513         /* Request a slot reset. */
11514         return PCI_ERS_RESULT_NEED_RESET;
11515 }
11516
11517 /**
11518  * ixgbe_io_slot_reset - called after the pci bus has been reset.
11519  * @pdev: Pointer to PCI device
11520  *
11521  * Restart the card from scratch, as if from a cold-boot.
11522  */
11523 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
11524 {
11525         struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11526         pci_ers_result_t result;
11527
11528         if (pci_enable_device_mem(pdev)) {
11529                 e_err(probe, "Cannot re-enable PCI device after reset.\n");
11530                 result = PCI_ERS_RESULT_DISCONNECT;
11531         } else {
11532                 smp_mb__before_atomic();
11533                 clear_bit(__IXGBE_DISABLED, &adapter->state);
11534                 adapter->hw.hw_addr = adapter->io_addr;
11535                 pci_set_master(pdev);
11536                 pci_restore_state(pdev);
11537                 pci_save_state(pdev);
11538
11539                 pci_wake_from_d3(pdev, false);
11540
11541                 ixgbe_reset(adapter);
11542                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
11543                 result = PCI_ERS_RESULT_RECOVERED;
11544         }
11545
11546         return result;
11547 }
11548
11549 /**
11550  * ixgbe_io_resume - called when traffic can start flowing again.
11551  * @pdev: Pointer to PCI device
11552  *
11553  * This callback is called when the error recovery driver tells us that
11554  * its OK to resume normal operation.
11555  */
11556 static void ixgbe_io_resume(struct pci_dev *pdev)
11557 {
11558         struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11559         struct net_device *netdev = adapter->netdev;
11560
11561 #ifdef CONFIG_PCI_IOV
11562         if (adapter->vferr_refcount) {
11563                 e_info(drv, "Resuming after VF err\n");
11564                 adapter->vferr_refcount--;
11565                 return;
11566         }
11567
11568 #endif
11569         rtnl_lock();
11570         if (netif_running(netdev))
11571                 ixgbe_open(netdev);
11572
11573         netif_device_attach(netdev);
11574         rtnl_unlock();
11575 }
11576
11577 static const struct pci_error_handlers ixgbe_err_handler = {
11578         .error_detected = ixgbe_io_error_detected,
11579         .slot_reset = ixgbe_io_slot_reset,
11580         .resume = ixgbe_io_resume,
11581 };
11582
11583 static DEFINE_SIMPLE_DEV_PM_OPS(ixgbe_pm_ops, ixgbe_suspend, ixgbe_resume);
11584
11585 static struct pci_driver ixgbe_driver = {
11586         .name      = ixgbe_driver_name,
11587         .id_table  = ixgbe_pci_tbl,
11588         .probe     = ixgbe_probe,
11589         .remove    = ixgbe_remove,
11590         .driver.pm = pm_sleep_ptr(&ixgbe_pm_ops),
11591         .shutdown  = ixgbe_shutdown,
11592         .sriov_configure = ixgbe_pci_sriov_configure,
11593         .err_handler = &ixgbe_err_handler
11594 };
11595
11596 /**
11597  * ixgbe_init_module - Driver Registration Routine
11598  *
11599  * ixgbe_init_module is the first routine called when the driver is
11600  * loaded. All it does is register with the PCI subsystem.
11601  **/
11602 static int __init ixgbe_init_module(void)
11603 {
11604         int ret;
11605         pr_info("%s\n", ixgbe_driver_string);
11606         pr_info("%s\n", ixgbe_copyright);
11607
11608         ixgbe_wq = create_singlethread_workqueue(ixgbe_driver_name);
11609         if (!ixgbe_wq) {
11610                 pr_err("%s: Failed to create workqueue\n", ixgbe_driver_name);
11611                 return -ENOMEM;
11612         }
11613
11614         ixgbe_dbg_init();
11615
11616         ret = pci_register_driver(&ixgbe_driver);
11617         if (ret) {
11618                 destroy_workqueue(ixgbe_wq);
11619                 ixgbe_dbg_exit();
11620                 return ret;
11621         }
11622
11623 #ifdef CONFIG_IXGBE_DCA
11624         dca_register_notify(&dca_notifier);
11625 #endif
11626
11627         return 0;
11628 }
11629
11630 module_init(ixgbe_init_module);
11631
11632 /**
11633  * ixgbe_exit_module - Driver Exit Cleanup Routine
11634  *
11635  * ixgbe_exit_module is called just before the driver is removed
11636  * from memory.
11637  **/
11638 static void __exit ixgbe_exit_module(void)
11639 {
11640 #ifdef CONFIG_IXGBE_DCA
11641         dca_unregister_notify(&dca_notifier);
11642 #endif
11643         pci_unregister_driver(&ixgbe_driver);
11644
11645         ixgbe_dbg_exit();
11646         if (ixgbe_wq) {
11647                 destroy_workqueue(ixgbe_wq);
11648                 ixgbe_wq = NULL;
11649         }
11650 }
11651
11652 #ifdef CONFIG_IXGBE_DCA
11653 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
11654                             void *p)
11655 {
11656         int ret_val;
11657
11658         ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
11659                                          __ixgbe_notify_dca);
11660
11661         return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
11662 }
11663
11664 #endif /* CONFIG_IXGBE_DCA */
11665
11666 module_exit(ixgbe_exit_module);
11667
11668 /* ixgbe_main.c */
This page took 0.725659 seconds and 4 git commands to generate.