]>
Commit | Line | Data |
---|---|---|
4863dea3 SG |
1 | /* |
2 | * Copyright (C) 2015 Cavium, Inc. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or modify it | |
5 | * under the terms of version 2 of the GNU General Public License | |
6 | * as published by the Free Software Foundation. | |
7 | */ | |
8 | ||
9 | #include <linux/module.h> | |
10 | #include <linux/interrupt.h> | |
11 | #include <linux/pci.h> | |
12 | #include <linux/netdevice.h> | |
aa2e259b | 13 | #include <linux/if_vlan.h> |
4863dea3 SG |
14 | #include <linux/etherdevice.h> |
15 | #include <linux/ethtool.h> | |
16 | #include <linux/log2.h> | |
17 | #include <linux/prefetch.h> | |
18 | #include <linux/irq.h> | |
83abb7d7 | 19 | #include <linux/iommu.h> |
05c773f5 | 20 | #include <linux/bpf.h> |
c56d91ce | 21 | #include <linux/bpf_trace.h> |
05c773f5 | 22 | #include <linux/filter.h> |
4a875509 | 23 | #include <linux/net_tstamp.h> |
37c3347e | 24 | #include <linux/workqueue.h> |
4863dea3 SG |
25 | |
26 | #include "nic_reg.h" | |
27 | #include "nic.h" | |
28 | #include "nicvf_queues.h" | |
29 | #include "thunder_bgx.h" | |
4a875509 | 30 | #include "../common/cavium_ptp.h" |
4863dea3 | 31 | |
6b9e6547 | 32 | #define DRV_NAME "nicvf" |
4863dea3 SG |
33 | #define DRV_VERSION "1.0" |
34 | ||
35 | /* Supported devices */ | |
36 | static const struct pci_device_id nicvf_id_table[] = { | |
37 | { PCI_DEVICE_SUB(PCI_VENDOR_ID_CAVIUM, | |
38 | PCI_DEVICE_ID_THUNDER_NIC_VF, | |
f7ff0ae8 SG |
39 | PCI_VENDOR_ID_CAVIUM, |
40 | PCI_SUBSYS_DEVID_88XX_NIC_VF) }, | |
4863dea3 SG |
41 | { PCI_DEVICE_SUB(PCI_VENDOR_ID_CAVIUM, |
42 | PCI_DEVICE_ID_THUNDER_PASS1_NIC_VF, | |
f7ff0ae8 SG |
43 | PCI_VENDOR_ID_CAVIUM, |
44 | PCI_SUBSYS_DEVID_88XX_PASS1_NIC_VF) }, | |
45 | { PCI_DEVICE_SUB(PCI_VENDOR_ID_CAVIUM, | |
46 | PCI_DEVICE_ID_THUNDER_NIC_VF, | |
47 | PCI_VENDOR_ID_CAVIUM, | |
48 | PCI_SUBSYS_DEVID_81XX_NIC_VF) }, | |
49 | { PCI_DEVICE_SUB(PCI_VENDOR_ID_CAVIUM, | |
50 | PCI_DEVICE_ID_THUNDER_NIC_VF, | |
51 | PCI_VENDOR_ID_CAVIUM, | |
52 | PCI_SUBSYS_DEVID_83XX_NIC_VF) }, | |
4863dea3 SG |
53 | { 0, } /* end of table */ |
54 | }; | |
55 | ||
56 | MODULE_AUTHOR("Sunil Goutham"); | |
57 | MODULE_DESCRIPTION("Cavium Thunder NIC Virtual Function Driver"); | |
58 | MODULE_LICENSE("GPL v2"); | |
59 | MODULE_VERSION(DRV_VERSION); | |
60 | MODULE_DEVICE_TABLE(pci, nicvf_id_table); | |
61 | ||
62 | static int debug = 0x00; | |
63 | module_param(debug, int, 0644); | |
64 | MODULE_PARM_DESC(debug, "Debug message level bitmap"); | |
65 | ||
66 | static int cpi_alg = CPI_ALG_NONE; | |
d3757ba4 | 67 | module_param(cpi_alg, int, 0444); |
4863dea3 SG |
68 | MODULE_PARM_DESC(cpi_alg, |
69 | "PFC algorithm (0=none, 1=VLAN, 2=VLAN16, 3=IP Diffserv)"); | |
70 | ||
37c3347e VL |
71 | /* workqueue for handling kernel ndo_set_rx_mode() calls */ |
72 | static struct workqueue_struct *nicvf_rx_mode_wq; | |
73 | ||
92dc8769 SG |
74 | static inline u8 nicvf_netdev_qidx(struct nicvf *nic, u8 qidx) |
75 | { | |
76 | if (nic->sqs_mode) | |
77 | return qidx + ((nic->sqs_id + 1) * MAX_CMP_QUEUES_PER_QS); | |
78 | else | |
79 | return qidx; | |
80 | } | |
81 | ||
4863dea3 SG |
82 | /* The Cavium ThunderX network controller can *only* be found in SoCs |
83 | * containing the ThunderX ARM64 CPU implementation. All accesses to the device | |
84 | * registers on this platform are implicitly strongly ordered with respect | |
85 | * to memory accesses. So writeq_relaxed() and readq_relaxed() are safe to use | |
86 | * with no memory barriers in this driver. The readq()/writeq() functions add | |
87 | * explicit ordering operation which in this case are redundant, and only | |
88 | * add overhead. | |
89 | */ | |
90 | ||
91 | /* Register read/write APIs */ | |
92 | void nicvf_reg_write(struct nicvf *nic, u64 offset, u64 val) | |
93 | { | |
94 | writeq_relaxed(val, nic->reg_base + offset); | |
95 | } | |
96 | ||
97 | u64 nicvf_reg_read(struct nicvf *nic, u64 offset) | |
98 | { | |
99 | return readq_relaxed(nic->reg_base + offset); | |
100 | } | |
101 | ||
102 | void nicvf_queue_reg_write(struct nicvf *nic, u64 offset, | |
103 | u64 qidx, u64 val) | |
104 | { | |
105 | void __iomem *addr = nic->reg_base + offset; | |
106 | ||
107 | writeq_relaxed(val, addr + (qidx << NIC_Q_NUM_SHIFT)); | |
108 | } | |
109 | ||
110 | u64 nicvf_queue_reg_read(struct nicvf *nic, u64 offset, u64 qidx) | |
111 | { | |
112 | void __iomem *addr = nic->reg_base + offset; | |
113 | ||
114 | return readq_relaxed(addr + (qidx << NIC_Q_NUM_SHIFT)); | |
115 | } | |
116 | ||
117 | /* VF -> PF mailbox communication */ | |
2cd2a196 AM |
118 | static void nicvf_write_to_mbx(struct nicvf *nic, union nic_mbx *mbx) |
119 | { | |
120 | u64 *msg = (u64 *)mbx; | |
121 | ||
122 | nicvf_reg_write(nic, NIC_VF_PF_MAILBOX_0_1 + 0, msg[0]); | |
123 | nicvf_reg_write(nic, NIC_VF_PF_MAILBOX_0_1 + 8, msg[1]); | |
124 | } | |
125 | ||
4863dea3 SG |
126 | int nicvf_send_msg_to_pf(struct nicvf *nic, union nic_mbx *mbx) |
127 | { | |
128 | int timeout = NIC_MBOX_MSG_TIMEOUT; | |
129 | int sleep = 10; | |
4863dea3 SG |
130 | |
131 | nic->pf_acked = false; | |
132 | nic->pf_nacked = false; | |
133 | ||
2cd2a196 | 134 | nicvf_write_to_mbx(nic, mbx); |
4863dea3 SG |
135 | |
136 | /* Wait for previous message to be acked, timeout 2sec */ | |
137 | while (!nic->pf_acked) { | |
ecae29cb RB |
138 | if (nic->pf_nacked) { |
139 | netdev_err(nic->netdev, | |
140 | "PF NACK to mbox msg 0x%02x from VF%d\n", | |
141 | (mbx->msg.msg & 0xFF), nic->vf_id); | |
4863dea3 | 142 | return -EINVAL; |
ecae29cb | 143 | } |
4863dea3 SG |
144 | msleep(sleep); |
145 | if (nic->pf_acked) | |
146 | break; | |
147 | timeout -= sleep; | |
148 | if (!timeout) { | |
149 | netdev_err(nic->netdev, | |
ecae29cb | 150 | "PF didn't ACK to mbox msg 0x%02x from VF%d\n", |
4863dea3 SG |
151 | (mbx->msg.msg & 0xFF), nic->vf_id); |
152 | return -EBUSY; | |
153 | } | |
154 | } | |
155 | return 0; | |
156 | } | |
157 | ||
158 | /* Checks if VF is able to comminicate with PF | |
159 | * and also gets the VNIC number this VF is associated to. | |
160 | */ | |
161 | static int nicvf_check_pf_ready(struct nicvf *nic) | |
162 | { | |
2cd2a196 AM |
163 | union nic_mbx mbx = {}; |
164 | ||
165 | mbx.msg.msg = NIC_MBOX_MSG_READY; | |
6051cba7 SG |
166 | if (nicvf_send_msg_to_pf(nic, &mbx)) { |
167 | netdev_err(nic->netdev, | |
168 | "PF didn't respond to READY msg\n"); | |
169 | return 0; | |
4863dea3 | 170 | } |
6051cba7 | 171 | |
4863dea3 SG |
172 | return 1; |
173 | } | |
174 | ||
fd7ec062 AM |
175 | static void nicvf_read_bgx_stats(struct nicvf *nic, struct bgx_stats_msg *bgx) |
176 | { | |
177 | if (bgx->rx) | |
178 | nic->bgx_stats.rx_stats[bgx->idx] = bgx->stats; | |
179 | else | |
180 | nic->bgx_stats.tx_stats[bgx->idx] = bgx->stats; | |
181 | } | |
182 | ||
4863dea3 SG |
183 | static void nicvf_handle_mbx_intr(struct nicvf *nic) |
184 | { | |
185 | union nic_mbx mbx = {}; | |
186 | u64 *mbx_data; | |
187 | u64 mbx_addr; | |
188 | int i; | |
189 | ||
190 | mbx_addr = NIC_VF_PF_MAILBOX_0_1; | |
191 | mbx_data = (u64 *)&mbx; | |
192 | ||
193 | for (i = 0; i < NIC_PF_VF_MAILBOX_SIZE; i++) { | |
194 | *mbx_data = nicvf_reg_read(nic, mbx_addr); | |
195 | mbx_data++; | |
196 | mbx_addr += sizeof(u64); | |
197 | } | |
198 | ||
199 | netdev_dbg(nic->netdev, "Mbox message: msg: 0x%x\n", mbx.msg.msg); | |
200 | switch (mbx.msg.msg) { | |
201 | case NIC_MBOX_MSG_READY: | |
6051cba7 | 202 | nic->pf_acked = true; |
4863dea3 SG |
203 | nic->vf_id = mbx.nic_cfg.vf_id & 0x7F; |
204 | nic->tns_mode = mbx.nic_cfg.tns_mode & 0x7F; | |
205 | nic->node = mbx.nic_cfg.node_id; | |
bd049a90 PF |
206 | if (!nic->set_mac_pending) |
207 | ether_addr_copy(nic->netdev->dev_addr, | |
208 | mbx.nic_cfg.mac_addr); | |
92dc8769 | 209 | nic->sqs_mode = mbx.nic_cfg.sqs_mode; |
d77a2384 | 210 | nic->loopback_supported = mbx.nic_cfg.loopback_supported; |
4863dea3 SG |
211 | nic->link_up = false; |
212 | nic->duplex = 0; | |
213 | nic->speed = 0; | |
214 | break; | |
215 | case NIC_MBOX_MSG_ACK: | |
216 | nic->pf_acked = true; | |
217 | break; | |
218 | case NIC_MBOX_MSG_NACK: | |
219 | nic->pf_nacked = true; | |
220 | break; | |
221 | case NIC_MBOX_MSG_RSS_SIZE: | |
222 | nic->rss_info.rss_size = mbx.rss_size.ind_tbl_size; | |
223 | nic->pf_acked = true; | |
224 | break; | |
225 | case NIC_MBOX_MSG_BGX_STATS: | |
226 | nicvf_read_bgx_stats(nic, &mbx.bgx_stats); | |
227 | nic->pf_acked = true; | |
4863dea3 SG |
228 | break; |
229 | case NIC_MBOX_MSG_BGX_LINK_CHANGE: | |
230 | nic->pf_acked = true; | |
231 | nic->link_up = mbx.link_status.link_up; | |
232 | nic->duplex = mbx.link_status.duplex; | |
233 | nic->speed = mbx.link_status.speed; | |
1cc70259 | 234 | nic->mac_type = mbx.link_status.mac_type; |
4863dea3 | 235 | if (nic->link_up) { |
bf24e136 JP |
236 | netdev_info(nic->netdev, "Link is Up %d Mbps %s duplex\n", |
237 | nic->speed, | |
4863dea3 | 238 | nic->duplex == DUPLEX_FULL ? |
bf24e136 | 239 | "Full" : "Half"); |
4863dea3 | 240 | netif_carrier_on(nic->netdev); |
b49087dd | 241 | netif_tx_start_all_queues(nic->netdev); |
4863dea3 | 242 | } else { |
bf24e136 | 243 | netdev_info(nic->netdev, "Link is Down\n"); |
4863dea3 SG |
244 | netif_carrier_off(nic->netdev); |
245 | netif_tx_stop_all_queues(nic->netdev); | |
246 | } | |
247 | break; | |
92dc8769 SG |
248 | case NIC_MBOX_MSG_ALLOC_SQS: |
249 | nic->sqs_count = mbx.sqs_alloc.qs_count; | |
250 | nic->pf_acked = true; | |
251 | break; | |
252 | case NIC_MBOX_MSG_SNICVF_PTR: | |
253 | /* Primary VF: make note of secondary VF's pointer | |
254 | * to be used while packet transmission. | |
255 | */ | |
256 | nic->snicvf[mbx.nicvf.sqs_id] = | |
257 | (struct nicvf *)mbx.nicvf.nicvf; | |
258 | nic->pf_acked = true; | |
259 | break; | |
260 | case NIC_MBOX_MSG_PNICVF_PTR: | |
261 | /* Secondary VF/Qset: make note of primary VF's pointer | |
262 | * to be used while packet reception, to handover packet | |
263 | * to primary VF's netdev. | |
264 | */ | |
265 | nic->pnicvf = (struct nicvf *)mbx.nicvf.nicvf; | |
266 | nic->pf_acked = true; | |
267 | break; | |
430da208 SG |
268 | case NIC_MBOX_MSG_PFC: |
269 | nic->pfc.autoneg = mbx.pfc.autoneg; | |
270 | nic->pfc.fc_rx = mbx.pfc.fc_rx; | |
271 | nic->pfc.fc_tx = mbx.pfc.fc_tx; | |
272 | nic->pf_acked = true; | |
273 | break; | |
4863dea3 SG |
274 | default: |
275 | netdev_err(nic->netdev, | |
276 | "Invalid message from PF, msg 0x%x\n", mbx.msg.msg); | |
277 | break; | |
278 | } | |
279 | nicvf_clear_intr(nic, NICVF_INTR_MBOX, 0); | |
280 | } | |
281 | ||
282 | static int nicvf_hw_set_mac_addr(struct nicvf *nic, struct net_device *netdev) | |
283 | { | |
284 | union nic_mbx mbx = {}; | |
4863dea3 SG |
285 | |
286 | mbx.mac.msg = NIC_MBOX_MSG_SET_MAC; | |
287 | mbx.mac.vf_id = nic->vf_id; | |
e610cb32 | 288 | ether_addr_copy(mbx.mac.mac_addr, netdev->dev_addr); |
4863dea3 SG |
289 | |
290 | return nicvf_send_msg_to_pf(nic, &mbx); | |
291 | } | |
292 | ||
fd7ec062 | 293 | static void nicvf_config_cpi(struct nicvf *nic) |
4863dea3 SG |
294 | { |
295 | union nic_mbx mbx = {}; | |
296 | ||
297 | mbx.cpi_cfg.msg = NIC_MBOX_MSG_CPI_CFG; | |
298 | mbx.cpi_cfg.vf_id = nic->vf_id; | |
299 | mbx.cpi_cfg.cpi_alg = nic->cpi_alg; | |
300 | mbx.cpi_cfg.rq_cnt = nic->qs->rq_cnt; | |
301 | ||
302 | nicvf_send_msg_to_pf(nic, &mbx); | |
303 | } | |
304 | ||
fd7ec062 | 305 | static void nicvf_get_rss_size(struct nicvf *nic) |
4863dea3 SG |
306 | { |
307 | union nic_mbx mbx = {}; | |
308 | ||
309 | mbx.rss_size.msg = NIC_MBOX_MSG_RSS_SIZE; | |
310 | mbx.rss_size.vf_id = nic->vf_id; | |
311 | nicvf_send_msg_to_pf(nic, &mbx); | |
312 | } | |
313 | ||
314 | void nicvf_config_rss(struct nicvf *nic) | |
315 | { | |
316 | union nic_mbx mbx = {}; | |
317 | struct nicvf_rss_info *rss = &nic->rss_info; | |
318 | int ind_tbl_len = rss->rss_size; | |
319 | int i, nextq = 0; | |
320 | ||
321 | mbx.rss_cfg.vf_id = nic->vf_id; | |
322 | mbx.rss_cfg.hash_bits = rss->hash_bits; | |
323 | while (ind_tbl_len) { | |
324 | mbx.rss_cfg.tbl_offset = nextq; | |
325 | mbx.rss_cfg.tbl_len = min(ind_tbl_len, | |
326 | RSS_IND_TBL_LEN_PER_MBX_MSG); | |
327 | mbx.rss_cfg.msg = mbx.rss_cfg.tbl_offset ? | |
328 | NIC_MBOX_MSG_RSS_CFG_CONT : NIC_MBOX_MSG_RSS_CFG; | |
329 | ||
330 | for (i = 0; i < mbx.rss_cfg.tbl_len; i++) | |
331 | mbx.rss_cfg.ind_tbl[i] = rss->ind_tbl[nextq++]; | |
332 | ||
333 | nicvf_send_msg_to_pf(nic, &mbx); | |
334 | ||
335 | ind_tbl_len -= mbx.rss_cfg.tbl_len; | |
336 | } | |
337 | } | |
338 | ||
339 | void nicvf_set_rss_key(struct nicvf *nic) | |
340 | { | |
341 | struct nicvf_rss_info *rss = &nic->rss_info; | |
342 | u64 key_addr = NIC_VNIC_RSS_KEY_0_4; | |
343 | int idx; | |
344 | ||
345 | for (idx = 0; idx < RSS_HASH_KEY_SIZE; idx++) { | |
346 | nicvf_reg_write(nic, key_addr, rss->key[idx]); | |
347 | key_addr += sizeof(u64); | |
348 | } | |
349 | } | |
350 | ||
351 | static int nicvf_rss_init(struct nicvf *nic) | |
352 | { | |
353 | struct nicvf_rss_info *rss = &nic->rss_info; | |
354 | int idx; | |
355 | ||
356 | nicvf_get_rss_size(nic); | |
357 | ||
38bb5d4f | 358 | if (cpi_alg != CPI_ALG_NONE) { |
4863dea3 SG |
359 | rss->enable = false; |
360 | rss->hash_bits = 0; | |
361 | return 0; | |
362 | } | |
363 | ||
364 | rss->enable = true; | |
365 | ||
0052c92f | 366 | netdev_rss_key_fill(rss->key, RSS_HASH_KEY_SIZE * sizeof(u64)); |
4863dea3 SG |
367 | nicvf_set_rss_key(nic); |
368 | ||
369 | rss->cfg = RSS_IP_HASH_ENA | RSS_TCP_HASH_ENA | RSS_UDP_HASH_ENA; | |
370 | nicvf_reg_write(nic, NIC_VNIC_RSS_CFG, rss->cfg); | |
371 | ||
372 | rss->hash_bits = ilog2(rounddown_pow_of_two(rss->rss_size)); | |
373 | ||
374 | for (idx = 0; idx < rss->rss_size; idx++) | |
375 | rss->ind_tbl[idx] = ethtool_rxfh_indir_default(idx, | |
92dc8769 | 376 | nic->rx_queues); |
4863dea3 SG |
377 | nicvf_config_rss(nic); |
378 | return 1; | |
379 | } | |
380 | ||
92dc8769 SG |
381 | /* Request PF to allocate additional Qsets */ |
382 | static void nicvf_request_sqs(struct nicvf *nic) | |
383 | { | |
384 | union nic_mbx mbx = {}; | |
385 | int sqs; | |
386 | int sqs_count = nic->sqs_count; | |
387 | int rx_queues = 0, tx_queues = 0; | |
388 | ||
389 | /* Only primary VF should request */ | |
390 | if (nic->sqs_mode || !nic->sqs_count) | |
391 | return; | |
392 | ||
393 | mbx.sqs_alloc.msg = NIC_MBOX_MSG_ALLOC_SQS; | |
394 | mbx.sqs_alloc.vf_id = nic->vf_id; | |
395 | mbx.sqs_alloc.qs_count = nic->sqs_count; | |
396 | if (nicvf_send_msg_to_pf(nic, &mbx)) { | |
397 | /* No response from PF */ | |
398 | nic->sqs_count = 0; | |
399 | return; | |
400 | } | |
401 | ||
402 | /* Return if no Secondary Qsets available */ | |
403 | if (!nic->sqs_count) | |
404 | return; | |
405 | ||
406 | if (nic->rx_queues > MAX_RCV_QUEUES_PER_QS) | |
407 | rx_queues = nic->rx_queues - MAX_RCV_QUEUES_PER_QS; | |
05c773f5 SG |
408 | |
409 | tx_queues = nic->tx_queues + nic->xdp_tx_queues; | |
410 | if (tx_queues > MAX_SND_QUEUES_PER_QS) | |
411 | tx_queues = tx_queues - MAX_SND_QUEUES_PER_QS; | |
92dc8769 SG |
412 | |
413 | /* Set no of Rx/Tx queues in each of the SQsets */ | |
414 | for (sqs = 0; sqs < nic->sqs_count; sqs++) { | |
415 | mbx.nicvf.msg = NIC_MBOX_MSG_SNICVF_PTR; | |
416 | mbx.nicvf.vf_id = nic->vf_id; | |
417 | mbx.nicvf.sqs_id = sqs; | |
418 | nicvf_send_msg_to_pf(nic, &mbx); | |
419 | ||
420 | nic->snicvf[sqs]->sqs_id = sqs; | |
421 | if (rx_queues > MAX_RCV_QUEUES_PER_QS) { | |
422 | nic->snicvf[sqs]->qs->rq_cnt = MAX_RCV_QUEUES_PER_QS; | |
423 | rx_queues -= MAX_RCV_QUEUES_PER_QS; | |
424 | } else { | |
425 | nic->snicvf[sqs]->qs->rq_cnt = rx_queues; | |
426 | rx_queues = 0; | |
427 | } | |
428 | ||
429 | if (tx_queues > MAX_SND_QUEUES_PER_QS) { | |
430 | nic->snicvf[sqs]->qs->sq_cnt = MAX_SND_QUEUES_PER_QS; | |
431 | tx_queues -= MAX_SND_QUEUES_PER_QS; | |
432 | } else { | |
433 | nic->snicvf[sqs]->qs->sq_cnt = tx_queues; | |
434 | tx_queues = 0; | |
435 | } | |
436 | ||
437 | nic->snicvf[sqs]->qs->cq_cnt = | |
438 | max(nic->snicvf[sqs]->qs->rq_cnt, nic->snicvf[sqs]->qs->sq_cnt); | |
439 | ||
440 | /* Initialize secondary Qset's queues and its interrupts */ | |
441 | nicvf_open(nic->snicvf[sqs]->netdev); | |
442 | } | |
443 | ||
444 | /* Update stack with actual Rx/Tx queue count allocated */ | |
445 | if (sqs_count != nic->sqs_count) | |
446 | nicvf_set_real_num_queues(nic->netdev, | |
447 | nic->tx_queues, nic->rx_queues); | |
448 | } | |
449 | ||
450 | /* Send this Qset's nicvf pointer to PF. | |
451 | * PF inturn sends primary VF's nicvf struct to secondary Qsets/VFs | |
452 | * so that packets received by these Qsets can use primary VF's netdev | |
453 | */ | |
454 | static void nicvf_send_vf_struct(struct nicvf *nic) | |
455 | { | |
456 | union nic_mbx mbx = {}; | |
457 | ||
458 | mbx.nicvf.msg = NIC_MBOX_MSG_NICVF_PTR; | |
459 | mbx.nicvf.sqs_mode = nic->sqs_mode; | |
460 | mbx.nicvf.nicvf = (u64)nic; | |
461 | nicvf_send_msg_to_pf(nic, &mbx); | |
462 | } | |
463 | ||
464 | static void nicvf_get_primary_vf_struct(struct nicvf *nic) | |
465 | { | |
466 | union nic_mbx mbx = {}; | |
467 | ||
468 | mbx.nicvf.msg = NIC_MBOX_MSG_PNICVF_PTR; | |
469 | nicvf_send_msg_to_pf(nic, &mbx); | |
470 | } | |
471 | ||
4863dea3 SG |
472 | int nicvf_set_real_num_queues(struct net_device *netdev, |
473 | int tx_queues, int rx_queues) | |
474 | { | |
475 | int err = 0; | |
476 | ||
477 | err = netif_set_real_num_tx_queues(netdev, tx_queues); | |
478 | if (err) { | |
479 | netdev_err(netdev, | |
480 | "Failed to set no of Tx queues: %d\n", tx_queues); | |
481 | return err; | |
482 | } | |
483 | ||
484 | err = netif_set_real_num_rx_queues(netdev, rx_queues); | |
485 | if (err) | |
486 | netdev_err(netdev, | |
487 | "Failed to set no of Rx queues: %d\n", rx_queues); | |
488 | return err; | |
489 | } | |
490 | ||
491 | static int nicvf_init_resources(struct nicvf *nic) | |
492 | { | |
493 | int err; | |
4863dea3 SG |
494 | |
495 | /* Enable Qset */ | |
496 | nicvf_qset_config(nic, true); | |
497 | ||
498 | /* Initialize queues and HW for data transfer */ | |
499 | err = nicvf_config_data_transfer(nic, true); | |
500 | if (err) { | |
501 | netdev_err(nic->netdev, | |
502 | "Failed to alloc/config VF's QSet resources\n"); | |
503 | return err; | |
504 | } | |
505 | ||
4863dea3 SG |
506 | return 0; |
507 | } | |
508 | ||
16f2bccd | 509 | static inline bool nicvf_xdp_rx(struct nicvf *nic, struct bpf_prog *prog, |
e3d06ff9 | 510 | struct cqe_rx_t *cqe_rx, struct snd_queue *sq, |
27e95e36 | 511 | struct rcv_queue *rq, struct sk_buff **skb) |
05c773f5 SG |
512 | { |
513 | struct xdp_buff xdp; | |
c56d91ce | 514 | struct page *page; |
05c773f5 | 515 | u32 action; |
e6dbe939 | 516 | u16 len, offset = 0; |
05c773f5 | 517 | u64 dma_addr, cpu_addr; |
e3d06ff9 | 518 | void *orig_data; |
05c773f5 SG |
519 | |
520 | /* Retrieve packet buffer's DMA address and length */ | |
521 | len = *((u16 *)((void *)cqe_rx + (3 * sizeof(u64)))); | |
522 | dma_addr = *((u64 *)((void *)cqe_rx + (7 * sizeof(u64)))); | |
523 | ||
524 | cpu_addr = nicvf_iova_to_phys(nic, dma_addr); | |
525 | if (!cpu_addr) | |
526 | return false; | |
e3d06ff9 SG |
527 | cpu_addr = (u64)phys_to_virt(cpu_addr); |
528 | page = virt_to_page((void *)cpu_addr); | |
05c773f5 | 529 | |
e6dbe939 | 530 | xdp.data_hard_start = page_address(page); |
e3d06ff9 | 531 | xdp.data = (void *)cpu_addr; |
de8f3a83 | 532 | xdp_set_data_meta_invalid(&xdp); |
05c773f5 | 533 | xdp.data_end = xdp.data + len; |
27e95e36 | 534 | xdp.rxq = &rq->xdp_rxq; |
e3d06ff9 | 535 | orig_data = xdp.data; |
05c773f5 SG |
536 | |
537 | rcu_read_lock(); | |
538 | action = bpf_prog_run_xdp(prog, &xdp); | |
539 | rcu_read_unlock(); | |
540 | ||
a48ce00f | 541 | len = xdp.data_end - xdp.data; |
e3d06ff9 SG |
542 | /* Check if XDP program has changed headers */ |
543 | if (orig_data != xdp.data) { | |
e3d06ff9 SG |
544 | offset = orig_data - xdp.data; |
545 | dma_addr -= offset; | |
546 | } | |
547 | ||
05c773f5 SG |
548 | switch (action) { |
549 | case XDP_PASS: | |
e6dbe939 JDB |
550 | /* Check if it's a recycled page, if not |
551 | * unmap the DMA mapping. | |
552 | * | |
553 | * Recycled page holds an extra reference. | |
554 | */ | |
555 | if (page_ref_count(page) == 1) { | |
556 | dma_addr &= PAGE_MASK; | |
557 | dma_unmap_page_attrs(&nic->pdev->dev, dma_addr, | |
558 | RCV_FRAG_LEN + XDP_PACKET_HEADROOM, | |
559 | DMA_FROM_DEVICE, | |
560 | DMA_ATTR_SKIP_CPU_SYNC); | |
561 | } | |
e3d06ff9 SG |
562 | |
563 | /* Build SKB and pass on packet to network stack */ | |
564 | *skb = build_skb(xdp.data, | |
565 | RCV_FRAG_LEN - cqe_rx->align_pad + offset); | |
566 | if (!*skb) | |
567 | put_page(page); | |
568 | else | |
569 | skb_put(*skb, len); | |
05c773f5 | 570 | return false; |
16f2bccd SG |
571 | case XDP_TX: |
572 | nicvf_xdp_sq_append_pkt(nic, sq, (u64)xdp.data, dma_addr, len); | |
573 | return true; | |
05c773f5 SG |
574 | default: |
575 | bpf_warn_invalid_xdp_action(action); | |
1a2ace56 | 576 | /* fall through */ |
c56d91ce SG |
577 | case XDP_ABORTED: |
578 | trace_xdp_exception(nic->netdev, prog, action); | |
1a2ace56 | 579 | /* fall through */ |
c56d91ce | 580 | case XDP_DROP: |
e6dbe939 JDB |
581 | /* Check if it's a recycled page, if not |
582 | * unmap the DMA mapping. | |
583 | * | |
584 | * Recycled page holds an extra reference. | |
585 | */ | |
586 | if (page_ref_count(page) == 1) { | |
587 | dma_addr &= PAGE_MASK; | |
588 | dma_unmap_page_attrs(&nic->pdev->dev, dma_addr, | |
589 | RCV_FRAG_LEN + XDP_PACKET_HEADROOM, | |
590 | DMA_FROM_DEVICE, | |
591 | DMA_ATTR_SKIP_CPU_SYNC); | |
592 | } | |
c56d91ce SG |
593 | put_page(page); |
594 | return true; | |
05c773f5 SG |
595 | } |
596 | return false; | |
597 | } | |
598 | ||
4a875509 SG |
599 | static void nicvf_snd_ptp_handler(struct net_device *netdev, |
600 | struct cqe_send_t *cqe_tx) | |
601 | { | |
602 | struct nicvf *nic = netdev_priv(netdev); | |
603 | struct skb_shared_hwtstamps ts; | |
604 | u64 ns; | |
605 | ||
606 | nic = nic->pnicvf; | |
607 | ||
608 | /* Sync for 'ptp_skb' */ | |
609 | smp_rmb(); | |
610 | ||
611 | /* New timestamp request can be queued now */ | |
612 | atomic_set(&nic->tx_ptp_skbs, 0); | |
613 | ||
614 | /* Check for timestamp requested skb */ | |
615 | if (!nic->ptp_skb) | |
616 | return; | |
617 | ||
618 | /* Check if timestamping is timedout, which is set to 10us */ | |
619 | if (cqe_tx->send_status == CQ_TX_ERROP_TSTMP_TIMEOUT || | |
620 | cqe_tx->send_status == CQ_TX_ERROP_TSTMP_CONFLICT) | |
621 | goto no_tstamp; | |
622 | ||
623 | /* Get the timestamp */ | |
624 | memset(&ts, 0, sizeof(ts)); | |
625 | ns = cavium_ptp_tstamp2time(nic->ptp_clock, cqe_tx->ptp_timestamp); | |
626 | ts.hwtstamp = ns_to_ktime(ns); | |
627 | skb_tstamp_tx(nic->ptp_skb, &ts); | |
628 | ||
629 | no_tstamp: | |
630 | /* Free the original skb */ | |
631 | dev_kfree_skb_any(nic->ptp_skb); | |
632 | nic->ptp_skb = NULL; | |
633 | /* Sync 'ptp_skb' */ | |
634 | smp_wmb(); | |
635 | } | |
636 | ||
4863dea3 | 637 | static void nicvf_snd_pkt_handler(struct net_device *netdev, |
c43548d2 | 638 | struct cqe_send_t *cqe_tx, |
0dada88b | 639 | int budget, int *subdesc_cnt, |
2c204c2b | 640 | unsigned int *tx_pkts, unsigned int *tx_bytes) |
4863dea3 SG |
641 | { |
642 | struct sk_buff *skb = NULL; | |
16f2bccd | 643 | struct page *page; |
4863dea3 SG |
644 | struct nicvf *nic = netdev_priv(netdev); |
645 | struct snd_queue *sq; | |
646 | struct sq_hdr_subdesc *hdr; | |
7ceb8a13 | 647 | struct sq_hdr_subdesc *tso_sqe; |
4863dea3 SG |
648 | |
649 | sq = &nic->qs->sq[cqe_tx->sq_idx]; | |
650 | ||
651 | hdr = (struct sq_hdr_subdesc *)GET_SQ_DESC(sq, cqe_tx->sqe_ptr); | |
652 | if (hdr->subdesc_type != SQ_DESC_TYPE_HEADER) | |
653 | return; | |
654 | ||
0dada88b SG |
655 | /* Check for errors */ |
656 | if (cqe_tx->send_status) | |
657 | nicvf_check_cqe_tx_errs(nic->pnicvf, cqe_tx); | |
4863dea3 | 658 | |
16f2bccd SG |
659 | /* Is this a XDP designated Tx queue */ |
660 | if (sq->is_xdp) { | |
661 | page = (struct page *)sq->xdp_page[cqe_tx->sqe_ptr]; | |
662 | /* Check if it's recycled page or else unmap DMA mapping */ | |
663 | if (page && (page_ref_count(page) == 1)) | |
664 | nicvf_unmap_sndq_buffers(nic, sq, cqe_tx->sqe_ptr, | |
665 | hdr->subdesc_cnt); | |
666 | ||
667 | /* Release page reference for recycling */ | |
668 | if (page) | |
669 | put_page(page); | |
670 | sq->xdp_page[cqe_tx->sqe_ptr] = (u64)NULL; | |
671 | *subdesc_cnt += hdr->subdesc_cnt + 1; | |
672 | return; | |
673 | } | |
674 | ||
4863dea3 | 675 | skb = (struct sk_buff *)sq->skbuff[cqe_tx->sqe_ptr]; |
4863dea3 | 676 | if (skb) { |
7ceb8a13 SG |
677 | /* Check for dummy descriptor used for HW TSO offload on 88xx */ |
678 | if (hdr->dont_send) { | |
679 | /* Get actual TSO descriptors and free them */ | |
680 | tso_sqe = | |
681 | (struct sq_hdr_subdesc *)GET_SQ_DESC(sq, hdr->rsvd2); | |
83abb7d7 SG |
682 | nicvf_unmap_sndq_buffers(nic, sq, hdr->rsvd2, |
683 | tso_sqe->subdesc_cnt); | |
0dada88b | 684 | *subdesc_cnt += tso_sqe->subdesc_cnt + 1; |
83abb7d7 SG |
685 | } else { |
686 | nicvf_unmap_sndq_buffers(nic, sq, cqe_tx->sqe_ptr, | |
687 | hdr->subdesc_cnt); | |
7ceb8a13 | 688 | } |
0dada88b | 689 | *subdesc_cnt += hdr->subdesc_cnt + 1; |
4863dea3 | 690 | prefetch(skb); |
2c204c2b SG |
691 | (*tx_pkts)++; |
692 | *tx_bytes += skb->len; | |
4a875509 SG |
693 | /* If timestamp is requested for this skb, don't free it */ |
694 | if (skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS && | |
695 | !nic->pnicvf->ptp_skb) | |
696 | nic->pnicvf->ptp_skb = skb; | |
697 | else | |
698 | napi_consume_skb(skb, budget); | |
143ceb0b | 699 | sq->skbuff[cqe_tx->sqe_ptr] = (u64)NULL; |
40fb5f8a | 700 | } else { |
7ceb8a13 SG |
701 | /* In case of SW TSO on 88xx, only last segment will have |
702 | * a SKB attached, so just free SQEs here. | |
40fb5f8a SG |
703 | */ |
704 | if (!nic->hw_tso) | |
0dada88b | 705 | *subdesc_cnt += hdr->subdesc_cnt + 1; |
4863dea3 SG |
706 | } |
707 | } | |
708 | ||
38bb5d4f SG |
709 | static inline void nicvf_set_rxhash(struct net_device *netdev, |
710 | struct cqe_rx_t *cqe_rx, | |
711 | struct sk_buff *skb) | |
712 | { | |
713 | u8 hash_type; | |
714 | u32 hash; | |
715 | ||
716 | if (!(netdev->features & NETIF_F_RXHASH)) | |
717 | return; | |
718 | ||
719 | switch (cqe_rx->rss_alg) { | |
720 | case RSS_ALG_TCP_IP: | |
721 | case RSS_ALG_UDP_IP: | |
722 | hash_type = PKT_HASH_TYPE_L4; | |
723 | hash = cqe_rx->rss_tag; | |
724 | break; | |
725 | case RSS_ALG_IP: | |
726 | hash_type = PKT_HASH_TYPE_L3; | |
727 | hash = cqe_rx->rss_tag; | |
728 | break; | |
729 | default: | |
730 | hash_type = PKT_HASH_TYPE_NONE; | |
731 | hash = 0; | |
732 | } | |
733 | ||
734 | skb_set_hash(skb, hash, hash_type); | |
735 | } | |
736 | ||
4a875509 SG |
737 | static inline void nicvf_set_rxtstamp(struct nicvf *nic, struct sk_buff *skb) |
738 | { | |
739 | u64 ns; | |
740 | ||
741 | if (!nic->ptp_clock || !nic->hw_rx_tstamp) | |
742 | return; | |
743 | ||
744 | /* The first 8 bytes is the timestamp */ | |
745 | ns = cavium_ptp_tstamp2time(nic->ptp_clock, | |
746 | be64_to_cpu(*(__be64 *)skb->data)); | |
747 | skb_hwtstamps(skb)->hwtstamp = ns_to_ktime(ns); | |
748 | ||
749 | __skb_pull(skb, 8); | |
750 | } | |
751 | ||
4863dea3 SG |
752 | static void nicvf_rcv_pkt_handler(struct net_device *netdev, |
753 | struct napi_struct *napi, | |
27e95e36 JDB |
754 | struct cqe_rx_t *cqe_rx, |
755 | struct snd_queue *sq, struct rcv_queue *rq) | |
4863dea3 | 756 | { |
e3d06ff9 | 757 | struct sk_buff *skb = NULL; |
4863dea3 | 758 | struct nicvf *nic = netdev_priv(netdev); |
83abb7d7 | 759 | struct nicvf *snic = nic; |
4863dea3 | 760 | int err = 0; |
92dc8769 SG |
761 | int rq_idx; |
762 | ||
763 | rq_idx = nicvf_netdev_qidx(nic, cqe_rx->rq_idx); | |
764 | ||
765 | if (nic->sqs_mode) { | |
766 | /* Use primary VF's 'nicvf' struct */ | |
767 | nic = nic->pnicvf; | |
768 | netdev = nic->netdev; | |
769 | } | |
4863dea3 SG |
770 | |
771 | /* Check for errors */ | |
0dada88b SG |
772 | if (cqe_rx->err_level || cqe_rx->err_opcode) { |
773 | err = nicvf_check_cqe_rx_errs(nic, cqe_rx); | |
774 | if (err && !cqe_rx->rb_cnt) | |
775 | return; | |
776 | } | |
4863dea3 | 777 | |
05c773f5 | 778 | /* For XDP, ignore pkts spanning multiple pages */ |
e3d06ff9 SG |
779 | if (nic->xdp_prog && (cqe_rx->rb_cnt == 1)) { |
780 | /* Packet consumed by XDP */ | |
27e95e36 | 781 | if (nicvf_xdp_rx(snic, nic->xdp_prog, cqe_rx, sq, rq, &skb)) |
05c773f5 | 782 | return; |
e3d06ff9 SG |
783 | } else { |
784 | skb = nicvf_get_rcv_skb(snic, cqe_rx, | |
785 | nic->xdp_prog ? true : false); | |
786 | } | |
05c773f5 | 787 | |
e3d06ff9 | 788 | if (!skb) |
4863dea3 | 789 | return; |
4863dea3 SG |
790 | |
791 | if (netif_msg_pktdata(nic)) { | |
bf24e136 | 792 | netdev_info(nic->netdev, "skb 0x%p, len=%d\n", skb, skb->len); |
4863dea3 SG |
793 | print_hex_dump(KERN_INFO, "", DUMP_PREFIX_OFFSET, 16, 1, |
794 | skb->data, skb->len, true); | |
795 | } | |
796 | ||
a2dc5ded SG |
797 | /* If error packet, drop it here */ |
798 | if (err) { | |
799 | dev_kfree_skb_any(skb); | |
800 | return; | |
801 | } | |
802 | ||
4a875509 | 803 | nicvf_set_rxtstamp(nic, skb); |
38bb5d4f SG |
804 | nicvf_set_rxhash(netdev, cqe_rx, skb); |
805 | ||
92dc8769 | 806 | skb_record_rx_queue(skb, rq_idx); |
4863dea3 SG |
807 | if (netdev->hw_features & NETIF_F_RXCSUM) { |
808 | /* HW by default verifies TCP/UDP/SCTP checksums */ | |
809 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
810 | } else { | |
811 | skb_checksum_none_assert(skb); | |
812 | } | |
813 | ||
814 | skb->protocol = eth_type_trans(skb, netdev); | |
815 | ||
aa2e259b SG |
816 | /* Check for stripped VLAN */ |
817 | if (cqe_rx->vlan_found && cqe_rx->vlan_stripped) | |
818 | __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), | |
819 | ntohs((__force __be16)cqe_rx->vlan_tci)); | |
820 | ||
4863dea3 SG |
821 | if (napi && (netdev->features & NETIF_F_GRO)) |
822 | napi_gro_receive(napi, skb); | |
823 | else | |
824 | netif_receive_skb(skb); | |
825 | } | |
826 | ||
827 | static int nicvf_cq_intr_handler(struct net_device *netdev, u8 cq_idx, | |
828 | struct napi_struct *napi, int budget) | |
829 | { | |
74840b83 | 830 | int processed_cqe, work_done = 0, tx_done = 0; |
4863dea3 | 831 | int cqe_count, cqe_head; |
0dada88b | 832 | int subdesc_cnt = 0; |
4863dea3 SG |
833 | struct nicvf *nic = netdev_priv(netdev); |
834 | struct queue_set *qs = nic->qs; | |
835 | struct cmp_queue *cq = &qs->cq[cq_idx]; | |
836 | struct cqe_rx_t *cq_desc; | |
74840b83 | 837 | struct netdev_queue *txq; |
16f2bccd | 838 | struct snd_queue *sq = &qs->sq[cq_idx]; |
27e95e36 | 839 | struct rcv_queue *rq = &qs->rq[cq_idx]; |
16f2bccd | 840 | unsigned int tx_pkts = 0, tx_bytes = 0, txq_idx; |
4863dea3 SG |
841 | |
842 | spin_lock_bh(&cq->lock); | |
843 | loop: | |
844 | processed_cqe = 0; | |
845 | /* Get no of valid CQ entries to process */ | |
846 | cqe_count = nicvf_queue_reg_read(nic, NIC_QSET_CQ_0_7_STATUS, cq_idx); | |
847 | cqe_count &= CQ_CQE_COUNT; | |
848 | if (!cqe_count) | |
849 | goto done; | |
850 | ||
851 | /* Get head of the valid CQ entries */ | |
852 | cqe_head = nicvf_queue_reg_read(nic, NIC_QSET_CQ_0_7_HEAD, cq_idx) >> 9; | |
853 | cqe_head &= 0xFFFF; | |
854 | ||
4863dea3 SG |
855 | while (processed_cqe < cqe_count) { |
856 | /* Get the CQ descriptor */ | |
857 | cq_desc = (struct cqe_rx_t *)GET_CQ_DESC(cq, cqe_head); | |
858 | cqe_head++; | |
859 | cqe_head &= (cq->dmem.q_len - 1); | |
860 | /* Initiate prefetch for next descriptor */ | |
861 | prefetch((struct cqe_rx_t *)GET_CQ_DESC(cq, cqe_head)); | |
862 | ||
863 | if ((work_done >= budget) && napi && | |
864 | (cq_desc->cqe_type != CQE_TYPE_SEND)) { | |
865 | break; | |
866 | } | |
867 | ||
4863dea3 SG |
868 | switch (cq_desc->cqe_type) { |
869 | case CQE_TYPE_RX: | |
27e95e36 | 870 | nicvf_rcv_pkt_handler(netdev, napi, cq_desc, sq, rq); |
4863dea3 SG |
871 | work_done++; |
872 | break; | |
873 | case CQE_TYPE_SEND: | |
0dada88b SG |
874 | nicvf_snd_pkt_handler(netdev, (void *)cq_desc, |
875 | budget, &subdesc_cnt, | |
876 | &tx_pkts, &tx_bytes); | |
74840b83 | 877 | tx_done++; |
4863dea3 | 878 | break; |
4a875509 SG |
879 | case CQE_TYPE_SEND_PTP: |
880 | nicvf_snd_ptp_handler(netdev, (void *)cq_desc); | |
881 | break; | |
4863dea3 SG |
882 | case CQE_TYPE_INVALID: |
883 | case CQE_TYPE_RX_SPLIT: | |
884 | case CQE_TYPE_RX_TCP: | |
4863dea3 SG |
885 | /* Ignore for now */ |
886 | break; | |
887 | } | |
888 | processed_cqe++; | |
889 | } | |
4863dea3 SG |
890 | |
891 | /* Ring doorbell to inform H/W to reuse processed CQEs */ | |
892 | nicvf_queue_reg_write(nic, NIC_QSET_CQ_0_7_DOOR, | |
893 | cq_idx, processed_cqe); | |
894 | ||
895 | if ((work_done < budget) && napi) | |
896 | goto loop; | |
897 | ||
898 | done: | |
0dada88b SG |
899 | /* Update SQ's descriptor free count */ |
900 | if (subdesc_cnt) | |
901 | nicvf_put_sq_desc(sq, subdesc_cnt); | |
902 | ||
16f2bccd SG |
903 | txq_idx = nicvf_netdev_qidx(nic, cq_idx); |
904 | /* Handle XDP TX queues */ | |
905 | if (nic->pnicvf->xdp_prog) { | |
906 | if (txq_idx < nic->pnicvf->xdp_tx_queues) { | |
907 | nicvf_xdp_sq_doorbell(nic, sq, cq_idx); | |
908 | goto out; | |
909 | } | |
910 | nic = nic->pnicvf; | |
911 | txq_idx -= nic->pnicvf->xdp_tx_queues; | |
912 | } | |
913 | ||
0dada88b | 914 | /* Wakeup TXQ if its stopped earlier due to SQ full */ |
bd3ad7d3 SG |
915 | if (tx_done || |
916 | (atomic_read(&sq->free_cnt) >= MIN_SQ_DESC_PER_PKT_XMIT)) { | |
92dc8769 | 917 | netdev = nic->pnicvf->netdev; |
16f2bccd | 918 | txq = netdev_get_tx_queue(netdev, txq_idx); |
2c204c2b SG |
919 | if (tx_pkts) |
920 | netdev_tx_completed_queue(txq, tx_pkts, tx_bytes); | |
921 | ||
bd3ad7d3 SG |
922 | /* To read updated queue and carrier status */ |
923 | smp_mb(); | |
92dc8769 | 924 | if (netif_tx_queue_stopped(txq) && netif_carrier_ok(netdev)) { |
bd3ad7d3 SG |
925 | netif_tx_wake_queue(txq); |
926 | nic = nic->pnicvf; | |
964cb69b | 927 | this_cpu_inc(nic->drv_stats->txq_wake); |
bf24e136 JP |
928 | netif_warn(nic, tx_err, netdev, |
929 | "Transmit queue wakeup SQ%d\n", txq_idx); | |
74840b83 SG |
930 | } |
931 | } | |
932 | ||
16f2bccd | 933 | out: |
4863dea3 SG |
934 | spin_unlock_bh(&cq->lock); |
935 | return work_done; | |
936 | } | |
937 | ||
938 | static int nicvf_poll(struct napi_struct *napi, int budget) | |
939 | { | |
940 | u64 cq_head; | |
941 | int work_done = 0; | |
942 | struct net_device *netdev = napi->dev; | |
943 | struct nicvf *nic = netdev_priv(netdev); | |
944 | struct nicvf_cq_poll *cq; | |
4863dea3 SG |
945 | |
946 | cq = container_of(napi, struct nicvf_cq_poll, napi); | |
947 | work_done = nicvf_cq_intr_handler(netdev, cq->cq_idx, napi, budget); | |
948 | ||
4863dea3 SG |
949 | if (work_done < budget) { |
950 | /* Slow packet rate, exit polling */ | |
6ad20165 | 951 | napi_complete_done(napi, work_done); |
4863dea3 SG |
952 | /* Re-enable interrupts */ |
953 | cq_head = nicvf_queue_reg_read(nic, NIC_QSET_CQ_0_7_HEAD, | |
954 | cq->cq_idx); | |
955 | nicvf_clear_intr(nic, NICVF_INTR_CQ, cq->cq_idx); | |
956 | nicvf_queue_reg_write(nic, NIC_QSET_CQ_0_7_HEAD, | |
957 | cq->cq_idx, cq_head); | |
958 | nicvf_enable_intr(nic, NICVF_INTR_CQ, cq->cq_idx); | |
959 | } | |
960 | return work_done; | |
961 | } | |
962 | ||
963 | /* Qset error interrupt handler | |
964 | * | |
965 | * As of now only CQ errors are handled | |
966 | */ | |
fd7ec062 | 967 | static void nicvf_handle_qs_err(unsigned long data) |
4863dea3 SG |
968 | { |
969 | struct nicvf *nic = (struct nicvf *)data; | |
970 | struct queue_set *qs = nic->qs; | |
971 | int qidx; | |
972 | u64 status; | |
973 | ||
974 | netif_tx_disable(nic->netdev); | |
975 | ||
976 | /* Check if it is CQ err */ | |
977 | for (qidx = 0; qidx < qs->cq_cnt; qidx++) { | |
978 | status = nicvf_queue_reg_read(nic, NIC_QSET_CQ_0_7_STATUS, | |
979 | qidx); | |
980 | if (!(status & CQ_ERR_MASK)) | |
981 | continue; | |
982 | /* Process already queued CQEs and reconfig CQ */ | |
983 | nicvf_disable_intr(nic, NICVF_INTR_CQ, qidx); | |
984 | nicvf_sq_disable(nic, qidx); | |
985 | nicvf_cq_intr_handler(nic->netdev, qidx, NULL, 0); | |
986 | nicvf_cmp_queue_config(nic, qs, qidx, true); | |
987 | nicvf_sq_free_used_descs(nic->netdev, &qs->sq[qidx], qidx); | |
988 | nicvf_sq_enable(nic, &qs->sq[qidx], qidx); | |
989 | ||
990 | nicvf_enable_intr(nic, NICVF_INTR_CQ, qidx); | |
991 | } | |
992 | ||
993 | netif_tx_start_all_queues(nic->netdev); | |
994 | /* Re-enable Qset error interrupt */ | |
995 | nicvf_enable_intr(nic, NICVF_INTR_QS_ERR, 0); | |
996 | } | |
997 | ||
39ad6eea SG |
998 | static void nicvf_dump_intr_status(struct nicvf *nic) |
999 | { | |
bf24e136 JP |
1000 | netif_info(nic, intr, nic->netdev, "interrupt status 0x%llx\n", |
1001 | nicvf_reg_read(nic, NIC_VF_INT)); | |
39ad6eea SG |
1002 | } |
1003 | ||
4863dea3 SG |
1004 | static irqreturn_t nicvf_misc_intr_handler(int irq, void *nicvf_irq) |
1005 | { | |
1006 | struct nicvf *nic = (struct nicvf *)nicvf_irq; | |
1007 | u64 intr; | |
1008 | ||
39ad6eea SG |
1009 | nicvf_dump_intr_status(nic); |
1010 | ||
4863dea3 SG |
1011 | intr = nicvf_reg_read(nic, NIC_VF_INT); |
1012 | /* Check for spurious interrupt */ | |
1013 | if (!(intr & NICVF_INTR_MBOX_MASK)) | |
1014 | return IRQ_HANDLED; | |
1015 | ||
1016 | nicvf_handle_mbx_intr(nic); | |
1017 | ||
1018 | return IRQ_HANDLED; | |
1019 | } | |
1020 | ||
39ad6eea SG |
1021 | static irqreturn_t nicvf_intr_handler(int irq, void *cq_irq) |
1022 | { | |
1023 | struct nicvf_cq_poll *cq_poll = (struct nicvf_cq_poll *)cq_irq; | |
1024 | struct nicvf *nic = cq_poll->nicvf; | |
1025 | int qidx = cq_poll->cq_idx; | |
1026 | ||
1027 | nicvf_dump_intr_status(nic); | |
1028 | ||
1029 | /* Disable interrupts */ | |
1030 | nicvf_disable_intr(nic, NICVF_INTR_CQ, qidx); | |
1031 | ||
1032 | /* Schedule NAPI */ | |
ef0a4d86 | 1033 | napi_schedule_irqoff(&cq_poll->napi); |
39ad6eea SG |
1034 | |
1035 | /* Clear interrupt */ | |
1036 | nicvf_clear_intr(nic, NICVF_INTR_CQ, qidx); | |
1037 | ||
1038 | return IRQ_HANDLED; | |
1039 | } | |
1040 | ||
1041 | static irqreturn_t nicvf_rbdr_intr_handler(int irq, void *nicvf_irq) | |
4863dea3 | 1042 | { |
4863dea3 | 1043 | struct nicvf *nic = (struct nicvf *)nicvf_irq; |
39ad6eea | 1044 | u8 qidx; |
4863dea3 | 1045 | |
4863dea3 | 1046 | |
39ad6eea SG |
1047 | nicvf_dump_intr_status(nic); |
1048 | ||
1049 | /* Disable RBDR interrupt and schedule softirq */ | |
1050 | for (qidx = 0; qidx < nic->qs->rbdr_cnt; qidx++) { | |
1051 | if (!nicvf_is_intr_enabled(nic, NICVF_INTR_RBDR, qidx)) | |
4863dea3 | 1052 | continue; |
39ad6eea SG |
1053 | nicvf_disable_intr(nic, NICVF_INTR_RBDR, qidx); |
1054 | tasklet_hi_schedule(&nic->rbdr_task); | |
1055 | /* Clear interrupt */ | |
1056 | nicvf_clear_intr(nic, NICVF_INTR_RBDR, qidx); | |
1057 | } | |
4863dea3 | 1058 | |
39ad6eea SG |
1059 | return IRQ_HANDLED; |
1060 | } | |
4863dea3 | 1061 | |
39ad6eea SG |
1062 | static irqreturn_t nicvf_qs_err_intr_handler(int irq, void *nicvf_irq) |
1063 | { | |
1064 | struct nicvf *nic = (struct nicvf *)nicvf_irq; | |
4863dea3 | 1065 | |
39ad6eea SG |
1066 | nicvf_dump_intr_status(nic); |
1067 | ||
1068 | /* Disable Qset err interrupt and schedule softirq */ | |
1069 | nicvf_disable_intr(nic, NICVF_INTR_QS_ERR, 0); | |
1070 | tasklet_hi_schedule(&nic->qs_err_task); | |
1071 | nicvf_clear_intr(nic, NICVF_INTR_QS_ERR, 0); | |
4863dea3 | 1072 | |
4863dea3 SG |
1073 | return IRQ_HANDLED; |
1074 | } | |
1075 | ||
fb4b7d98 SG |
1076 | static void nicvf_set_irq_affinity(struct nicvf *nic) |
1077 | { | |
1078 | int vec, cpu; | |
fb4b7d98 SG |
1079 | |
1080 | for (vec = 0; vec < nic->num_vec; vec++) { | |
1081 | if (!nic->irq_allocated[vec]) | |
1082 | continue; | |
1083 | ||
1084 | if (!zalloc_cpumask_var(&nic->affinity_mask[vec], GFP_KERNEL)) | |
1085 | return; | |
1086 | /* CQ interrupts */ | |
1087 | if (vec < NICVF_INTR_ID_SQ) | |
1088 | /* Leave CPU0 for RBDR and other interrupts */ | |
1089 | cpu = nicvf_netdev_qidx(nic, vec) + 1; | |
1090 | else | |
1091 | cpu = 0; | |
1092 | ||
1093 | cpumask_set_cpu(cpumask_local_spread(cpu, nic->node), | |
1094 | nic->affinity_mask[vec]); | |
ce211b17 TS |
1095 | irq_set_affinity_hint(pci_irq_vector(nic->pdev, vec), |
1096 | nic->affinity_mask[vec]); | |
fb4b7d98 SG |
1097 | } |
1098 | } | |
1099 | ||
4863dea3 SG |
1100 | static int nicvf_register_interrupts(struct nicvf *nic) |
1101 | { | |
39ad6eea | 1102 | int irq, ret = 0; |
4863dea3 SG |
1103 | |
1104 | for_each_cq_irq(irq) | |
e4126213 SG |
1105 | sprintf(nic->irq_name[irq], "%s-rxtx-%d", |
1106 | nic->pnicvf->netdev->name, | |
1107 | nicvf_netdev_qidx(nic, irq)); | |
4863dea3 SG |
1108 | |
1109 | for_each_sq_irq(irq) | |
e4126213 SG |
1110 | sprintf(nic->irq_name[irq], "%s-sq-%d", |
1111 | nic->pnicvf->netdev->name, | |
1112 | nicvf_netdev_qidx(nic, irq - NICVF_INTR_ID_SQ)); | |
4863dea3 SG |
1113 | |
1114 | for_each_rbdr_irq(irq) | |
e4126213 SG |
1115 | sprintf(nic->irq_name[irq], "%s-rbdr-%d", |
1116 | nic->pnicvf->netdev->name, | |
1117 | nic->sqs_mode ? (nic->sqs_id + 1) : 0); | |
4863dea3 | 1118 | |
39ad6eea SG |
1119 | /* Register CQ interrupts */ |
1120 | for (irq = 0; irq < nic->qs->cq_cnt; irq++) { | |
ce211b17 TS |
1121 | ret = request_irq(pci_irq_vector(nic->pdev, irq), |
1122 | nicvf_intr_handler, | |
39ad6eea | 1123 | 0, nic->irq_name[irq], nic->napi[irq]); |
4863dea3 | 1124 | if (ret) |
39ad6eea | 1125 | goto err; |
4863dea3 SG |
1126 | nic->irq_allocated[irq] = true; |
1127 | } | |
1128 | ||
39ad6eea SG |
1129 | /* Register RBDR interrupt */ |
1130 | for (irq = NICVF_INTR_ID_RBDR; | |
1131 | irq < (NICVF_INTR_ID_RBDR + nic->qs->rbdr_cnt); irq++) { | |
ce211b17 TS |
1132 | ret = request_irq(pci_irq_vector(nic->pdev, irq), |
1133 | nicvf_rbdr_intr_handler, | |
4863dea3 SG |
1134 | 0, nic->irq_name[irq], nic); |
1135 | if (ret) | |
39ad6eea | 1136 | goto err; |
4863dea3 SG |
1137 | nic->irq_allocated[irq] = true; |
1138 | } | |
1139 | ||
39ad6eea | 1140 | /* Register QS error interrupt */ |
e4126213 SG |
1141 | sprintf(nic->irq_name[NICVF_INTR_ID_QS_ERR], "%s-qset-err-%d", |
1142 | nic->pnicvf->netdev->name, | |
1143 | nic->sqs_mode ? (nic->sqs_id + 1) : 0); | |
39ad6eea | 1144 | irq = NICVF_INTR_ID_QS_ERR; |
ce211b17 | 1145 | ret = request_irq(pci_irq_vector(nic->pdev, irq), |
39ad6eea SG |
1146 | nicvf_qs_err_intr_handler, |
1147 | 0, nic->irq_name[irq], nic); | |
fb4b7d98 SG |
1148 | if (ret) |
1149 | goto err; | |
1150 | ||
1151 | nic->irq_allocated[irq] = true; | |
1152 | ||
1153 | /* Set IRQ affinities */ | |
1154 | nicvf_set_irq_affinity(nic); | |
4863dea3 | 1155 | |
39ad6eea SG |
1156 | err: |
1157 | if (ret) | |
1158 | netdev_err(nic->netdev, "request_irq failed, vector %d\n", irq); | |
4863dea3 | 1159 | |
39ad6eea | 1160 | return ret; |
4863dea3 SG |
1161 | } |
1162 | ||
1163 | static void nicvf_unregister_interrupts(struct nicvf *nic) | |
1164 | { | |
ce211b17 | 1165 | struct pci_dev *pdev = nic->pdev; |
4863dea3 SG |
1166 | int irq; |
1167 | ||
1168 | /* Free registered interrupts */ | |
1169 | for (irq = 0; irq < nic->num_vec; irq++) { | |
39ad6eea SG |
1170 | if (!nic->irq_allocated[irq]) |
1171 | continue; | |
1172 | ||
ce211b17 | 1173 | irq_set_affinity_hint(pci_irq_vector(pdev, irq), NULL); |
fb4b7d98 SG |
1174 | free_cpumask_var(nic->affinity_mask[irq]); |
1175 | ||
39ad6eea | 1176 | if (irq < NICVF_INTR_ID_SQ) |
ce211b17 | 1177 | free_irq(pci_irq_vector(pdev, irq), nic->napi[irq]); |
39ad6eea | 1178 | else |
ce211b17 | 1179 | free_irq(pci_irq_vector(pdev, irq), nic); |
39ad6eea | 1180 | |
4863dea3 SG |
1181 | nic->irq_allocated[irq] = false; |
1182 | } | |
1183 | ||
1184 | /* Disable MSI-X */ | |
ce211b17 TS |
1185 | pci_free_irq_vectors(pdev); |
1186 | nic->num_vec = 0; | |
4863dea3 SG |
1187 | } |
1188 | ||
1189 | /* Initialize MSIX vectors and register MISC interrupt. | |
1190 | * Send READY message to PF to check if its alive | |
1191 | */ | |
1192 | static int nicvf_register_misc_interrupt(struct nicvf *nic) | |
1193 | { | |
1194 | int ret = 0; | |
1195 | int irq = NICVF_INTR_ID_MISC; | |
1196 | ||
1197 | /* Return if mailbox interrupt is already registered */ | |
ce211b17 | 1198 | if (nic->pdev->msix_enabled) |
4863dea3 SG |
1199 | return 0; |
1200 | ||
1201 | /* Enable MSI-X */ | |
ce211b17 TS |
1202 | nic->num_vec = pci_msix_vec_count(nic->pdev); |
1203 | ret = pci_alloc_irq_vectors(nic->pdev, nic->num_vec, nic->num_vec, | |
1204 | PCI_IRQ_MSIX); | |
1205 | if (ret < 0) { | |
1206 | netdev_err(nic->netdev, | |
1207 | "Req for #%d msix vectors failed\n", nic->num_vec); | |
4863dea3 | 1208 | return 1; |
ce211b17 | 1209 | } |
4863dea3 SG |
1210 | |
1211 | sprintf(nic->irq_name[irq], "%s Mbox", "NICVF"); | |
1212 | /* Register Misc interrupt */ | |
ce211b17 | 1213 | ret = request_irq(pci_irq_vector(nic->pdev, irq), |
4863dea3 SG |
1214 | nicvf_misc_intr_handler, 0, nic->irq_name[irq], nic); |
1215 | ||
1216 | if (ret) | |
1217 | return ret; | |
1218 | nic->irq_allocated[irq] = true; | |
1219 | ||
1220 | /* Enable mailbox interrupt */ | |
1221 | nicvf_enable_intr(nic, NICVF_INTR_MBOX, 0); | |
1222 | ||
1223 | /* Check if VF is able to communicate with PF */ | |
1224 | if (!nicvf_check_pf_ready(nic)) { | |
1225 | nicvf_disable_intr(nic, NICVF_INTR_MBOX, 0); | |
1226 | nicvf_unregister_interrupts(nic); | |
1227 | return 1; | |
1228 | } | |
1229 | ||
1230 | return 0; | |
1231 | } | |
1232 | ||
1233 | static netdev_tx_t nicvf_xmit(struct sk_buff *skb, struct net_device *netdev) | |
1234 | { | |
1235 | struct nicvf *nic = netdev_priv(netdev); | |
1236 | int qid = skb_get_queue_mapping(skb); | |
1237 | struct netdev_queue *txq = netdev_get_tx_queue(netdev, qid); | |
bd3ad7d3 SG |
1238 | struct nicvf *snic; |
1239 | struct snd_queue *sq; | |
1240 | int tmp; | |
4863dea3 SG |
1241 | |
1242 | /* Check for minimum packet length */ | |
1243 | if (skb->len <= ETH_HLEN) { | |
1244 | dev_kfree_skb(skb); | |
1245 | return NETDEV_TX_OK; | |
1246 | } | |
1247 | ||
16f2bccd SG |
1248 | /* In XDP case, initial HW tx queues are used for XDP, |
1249 | * but stack's queue mapping starts at '0', so skip the | |
1250 | * Tx queues attached to Rx queues for XDP. | |
1251 | */ | |
1252 | if (nic->xdp_prog) | |
1253 | qid += nic->xdp_tx_queues; | |
1254 | ||
bd3ad7d3 SG |
1255 | snic = nic; |
1256 | /* Get secondary Qset's SQ structure */ | |
1257 | if (qid >= MAX_SND_QUEUES_PER_QS) { | |
1258 | tmp = qid / MAX_SND_QUEUES_PER_QS; | |
1259 | snic = (struct nicvf *)nic->snicvf[tmp - 1]; | |
1260 | if (!snic) { | |
1261 | netdev_warn(nic->netdev, | |
1262 | "Secondary Qset#%d's ptr not initialized\n", | |
1263 | tmp - 1); | |
1264 | dev_kfree_skb(skb); | |
1265 | return NETDEV_TX_OK; | |
1266 | } | |
1267 | qid = qid % MAX_SND_QUEUES_PER_QS; | |
1268 | } | |
1269 | ||
1270 | sq = &snic->qs->sq[qid]; | |
1271 | if (!netif_tx_queue_stopped(txq) && | |
1272 | !nicvf_sq_append_skb(snic, sq, skb, qid)) { | |
4863dea3 | 1273 | netif_tx_stop_queue(txq); |
bd3ad7d3 SG |
1274 | |
1275 | /* Barrier, so that stop_queue visible to other cpus */ | |
1276 | smp_mb(); | |
1277 | ||
1278 | /* Check again, incase another cpu freed descriptors */ | |
1279 | if (atomic_read(&sq->free_cnt) > MIN_SQ_DESC_PER_PKT_XMIT) { | |
1280 | netif_tx_wake_queue(txq); | |
1281 | } else { | |
1282 | this_cpu_inc(nic->drv_stats->txq_stop); | |
bf24e136 JP |
1283 | netif_warn(nic, tx_err, netdev, |
1284 | "Transmit ring full, stopping SQ%d\n", qid); | |
bd3ad7d3 | 1285 | } |
4863dea3 SG |
1286 | return NETDEV_TX_BUSY; |
1287 | } | |
1288 | ||
1289 | return NETDEV_TX_OK; | |
1290 | } | |
1291 | ||
39ad6eea SG |
1292 | static inline void nicvf_free_cq_poll(struct nicvf *nic) |
1293 | { | |
1294 | struct nicvf_cq_poll *cq_poll; | |
1295 | int qidx; | |
1296 | ||
1297 | for (qidx = 0; qidx < nic->qs->cq_cnt; qidx++) { | |
1298 | cq_poll = nic->napi[qidx]; | |
1299 | if (!cq_poll) | |
1300 | continue; | |
1301 | nic->napi[qidx] = NULL; | |
1302 | kfree(cq_poll); | |
1303 | } | |
1304 | } | |
1305 | ||
4863dea3 SG |
1306 | int nicvf_stop(struct net_device *netdev) |
1307 | { | |
1308 | int irq, qidx; | |
1309 | struct nicvf *nic = netdev_priv(netdev); | |
1310 | struct queue_set *qs = nic->qs; | |
1311 | struct nicvf_cq_poll *cq_poll = NULL; | |
1312 | union nic_mbx mbx = {}; | |
1313 | ||
1314 | mbx.msg.msg = NIC_MBOX_MSG_SHUTDOWN; | |
1315 | nicvf_send_msg_to_pf(nic, &mbx); | |
1316 | ||
1317 | netif_carrier_off(netdev); | |
92dc8769 | 1318 | netif_tx_stop_all_queues(nic->netdev); |
0b72a9a1 | 1319 | nic->link_up = false; |
92dc8769 SG |
1320 | |
1321 | /* Teardown secondary qsets first */ | |
1322 | if (!nic->sqs_mode) { | |
1323 | for (qidx = 0; qidx < nic->sqs_count; qidx++) { | |
1324 | if (!nic->snicvf[qidx]) | |
1325 | continue; | |
1326 | nicvf_stop(nic->snicvf[qidx]->netdev); | |
1327 | nic->snicvf[qidx] = NULL; | |
1328 | } | |
1329 | } | |
4863dea3 SG |
1330 | |
1331 | /* Disable RBDR & QS error interrupts */ | |
1332 | for (qidx = 0; qidx < qs->rbdr_cnt; qidx++) { | |
1333 | nicvf_disable_intr(nic, NICVF_INTR_RBDR, qidx); | |
1334 | nicvf_clear_intr(nic, NICVF_INTR_RBDR, qidx); | |
1335 | } | |
1336 | nicvf_disable_intr(nic, NICVF_INTR_QS_ERR, 0); | |
1337 | nicvf_clear_intr(nic, NICVF_INTR_QS_ERR, 0); | |
1338 | ||
1339 | /* Wait for pending IRQ handlers to finish */ | |
1340 | for (irq = 0; irq < nic->num_vec; irq++) | |
ce211b17 | 1341 | synchronize_irq(pci_irq_vector(nic->pdev, irq)); |
4863dea3 SG |
1342 | |
1343 | tasklet_kill(&nic->rbdr_task); | |
1344 | tasklet_kill(&nic->qs_err_task); | |
1345 | if (nic->rb_work_scheduled) | |
1346 | cancel_delayed_work_sync(&nic->rbdr_work); | |
1347 | ||
1348 | for (qidx = 0; qidx < nic->qs->cq_cnt; qidx++) { | |
1349 | cq_poll = nic->napi[qidx]; | |
1350 | if (!cq_poll) | |
1351 | continue; | |
4863dea3 SG |
1352 | napi_synchronize(&cq_poll->napi); |
1353 | /* CQ intr is enabled while napi_complete, | |
1354 | * so disable it now | |
1355 | */ | |
1356 | nicvf_disable_intr(nic, NICVF_INTR_CQ, qidx); | |
1357 | nicvf_clear_intr(nic, NICVF_INTR_CQ, qidx); | |
1358 | napi_disable(&cq_poll->napi); | |
1359 | netif_napi_del(&cq_poll->napi); | |
4863dea3 SG |
1360 | } |
1361 | ||
b49087dd SG |
1362 | netif_tx_disable(netdev); |
1363 | ||
2c204c2b SG |
1364 | for (qidx = 0; qidx < netdev->num_tx_queues; qidx++) |
1365 | netdev_tx_reset_queue(netdev_get_tx_queue(netdev, qidx)); | |
1366 | ||
4863dea3 SG |
1367 | /* Free resources */ |
1368 | nicvf_config_data_transfer(nic, false); | |
1369 | ||
1370 | /* Disable HW Qset */ | |
1371 | nicvf_qset_config(nic, false); | |
1372 | ||
1373 | /* disable mailbox interrupt */ | |
1374 | nicvf_disable_intr(nic, NICVF_INTR_MBOX, 0); | |
1375 | ||
1376 | nicvf_unregister_interrupts(nic); | |
1377 | ||
39ad6eea SG |
1378 | nicvf_free_cq_poll(nic); |
1379 | ||
4a875509 SG |
1380 | /* Free any pending SKB saved to receive timestamp */ |
1381 | if (nic->ptp_skb) { | |
1382 | dev_kfree_skb_any(nic->ptp_skb); | |
1383 | nic->ptp_skb = NULL; | |
1384 | } | |
1385 | ||
92dc8769 SG |
1386 | /* Clear multiqset info */ |
1387 | nic->pnicvf = nic; | |
92dc8769 | 1388 | |
4863dea3 SG |
1389 | return 0; |
1390 | } | |
1391 | ||
4a875509 SG |
1392 | static int nicvf_config_hw_rx_tstamp(struct nicvf *nic, bool enable) |
1393 | { | |
1394 | union nic_mbx mbx = {}; | |
1395 | ||
1396 | mbx.ptp.msg = NIC_MBOX_MSG_PTP_CFG; | |
1397 | mbx.ptp.enable = enable; | |
1398 | ||
1399 | return nicvf_send_msg_to_pf(nic, &mbx); | |
1400 | } | |
1401 | ||
712c3185 SG |
1402 | static int nicvf_update_hw_max_frs(struct nicvf *nic, int mtu) |
1403 | { | |
1404 | union nic_mbx mbx = {}; | |
1405 | ||
1406 | mbx.frs.msg = NIC_MBOX_MSG_SET_MAX_FRS; | |
1407 | mbx.frs.max_frs = mtu; | |
1408 | mbx.frs.vf_id = nic->vf_id; | |
1409 | ||
1410 | return nicvf_send_msg_to_pf(nic, &mbx); | |
1411 | } | |
1412 | ||
4863dea3 SG |
1413 | int nicvf_open(struct net_device *netdev) |
1414 | { | |
964cb69b | 1415 | int cpu, err, qidx; |
4863dea3 SG |
1416 | struct nicvf *nic = netdev_priv(netdev); |
1417 | struct queue_set *qs = nic->qs; | |
1418 | struct nicvf_cq_poll *cq_poll = NULL; | |
c94acf80 | 1419 | union nic_mbx mbx = {}; |
4863dea3 SG |
1420 | |
1421 | netif_carrier_off(netdev); | |
1422 | ||
1423 | err = nicvf_register_misc_interrupt(nic); | |
1424 | if (err) | |
1425 | return err; | |
1426 | ||
1427 | /* Register NAPI handler for processing CQEs */ | |
1428 | for (qidx = 0; qidx < qs->cq_cnt; qidx++) { | |
1429 | cq_poll = kzalloc(sizeof(*cq_poll), GFP_KERNEL); | |
1430 | if (!cq_poll) { | |
1431 | err = -ENOMEM; | |
1432 | goto napi_del; | |
1433 | } | |
1434 | cq_poll->cq_idx = qidx; | |
39ad6eea | 1435 | cq_poll->nicvf = nic; |
4863dea3 SG |
1436 | netif_napi_add(netdev, &cq_poll->napi, nicvf_poll, |
1437 | NAPI_POLL_WEIGHT); | |
1438 | napi_enable(&cq_poll->napi); | |
1439 | nic->napi[qidx] = cq_poll; | |
1440 | } | |
1441 | ||
1442 | /* Check if we got MAC address from PF or else generate a radom MAC */ | |
a3a8ce4c | 1443 | if (!nic->sqs_mode && is_zero_ether_addr(netdev->dev_addr)) { |
4863dea3 SG |
1444 | eth_hw_addr_random(netdev); |
1445 | nicvf_hw_set_mac_addr(nic, netdev); | |
1446 | } | |
1447 | ||
bd049a90 PF |
1448 | if (nic->set_mac_pending) { |
1449 | nic->set_mac_pending = false; | |
1450 | nicvf_hw_set_mac_addr(nic, netdev); | |
1451 | } | |
1452 | ||
4863dea3 SG |
1453 | /* Init tasklet for handling Qset err interrupt */ |
1454 | tasklet_init(&nic->qs_err_task, nicvf_handle_qs_err, | |
1455 | (unsigned long)nic); | |
1456 | ||
1457 | /* Init RBDR tasklet which will refill RBDR */ | |
1458 | tasklet_init(&nic->rbdr_task, nicvf_rbdr_task, | |
1459 | (unsigned long)nic); | |
1460 | INIT_DELAYED_WORK(&nic->rbdr_work, nicvf_rbdr_work); | |
1461 | ||
1462 | /* Configure CPI alorithm */ | |
1463 | nic->cpi_alg = cpi_alg; | |
92dc8769 SG |
1464 | if (!nic->sqs_mode) |
1465 | nicvf_config_cpi(nic); | |
1466 | ||
1467 | nicvf_request_sqs(nic); | |
1468 | if (nic->sqs_mode) | |
1469 | nicvf_get_primary_vf_struct(nic); | |
4863dea3 | 1470 | |
4a875509 SG |
1471 | /* Configure PTP timestamp */ |
1472 | if (nic->ptp_clock) | |
1473 | nicvf_config_hw_rx_tstamp(nic, nic->hw_rx_tstamp); | |
1474 | atomic_set(&nic->tx_ptp_skbs, 0); | |
1475 | nic->ptp_skb = NULL; | |
1476 | ||
712c3185 SG |
1477 | /* Configure receive side scaling and MTU */ |
1478 | if (!nic->sqs_mode) { | |
92dc8769 | 1479 | nicvf_rss_init(nic); |
60dce04b WY |
1480 | err = nicvf_update_hw_max_frs(nic, netdev->mtu); |
1481 | if (err) | |
712c3185 | 1482 | goto cleanup; |
964cb69b SG |
1483 | |
1484 | /* Clear percpu stats */ | |
1485 | for_each_possible_cpu(cpu) | |
1486 | memset(per_cpu_ptr(nic->drv_stats, cpu), 0, | |
1487 | sizeof(struct nicvf_drv_stats)); | |
712c3185 | 1488 | } |
4863dea3 SG |
1489 | |
1490 | err = nicvf_register_interrupts(nic); | |
1491 | if (err) | |
1492 | goto cleanup; | |
1493 | ||
1494 | /* Initialize the queues */ | |
1495 | err = nicvf_init_resources(nic); | |
1496 | if (err) | |
1497 | goto cleanup; | |
1498 | ||
1499 | /* Make sure queue initialization is written */ | |
1500 | wmb(); | |
1501 | ||
1502 | nicvf_reg_write(nic, NIC_VF_INT, -1); | |
1503 | /* Enable Qset err interrupt */ | |
1504 | nicvf_enable_intr(nic, NICVF_INTR_QS_ERR, 0); | |
1505 | ||
1506 | /* Enable completion queue interrupt */ | |
1507 | for (qidx = 0; qidx < qs->cq_cnt; qidx++) | |
1508 | nicvf_enable_intr(nic, NICVF_INTR_CQ, qidx); | |
1509 | ||
1510 | /* Enable RBDR threshold interrupt */ | |
1511 | for (qidx = 0; qidx < qs->rbdr_cnt; qidx++) | |
1512 | nicvf_enable_intr(nic, NICVF_INTR_RBDR, qidx); | |
1513 | ||
c94acf80 SG |
1514 | /* Send VF config done msg to PF */ |
1515 | mbx.msg.msg = NIC_MBOX_MSG_CFG_DONE; | |
1516 | nicvf_write_to_mbx(nic, &mbx); | |
74840b83 | 1517 | |
4863dea3 SG |
1518 | return 0; |
1519 | cleanup: | |
1520 | nicvf_disable_intr(nic, NICVF_INTR_MBOX, 0); | |
1521 | nicvf_unregister_interrupts(nic); | |
39ad6eea SG |
1522 | tasklet_kill(&nic->qs_err_task); |
1523 | tasklet_kill(&nic->rbdr_task); | |
4863dea3 SG |
1524 | napi_del: |
1525 | for (qidx = 0; qidx < qs->cq_cnt; qidx++) { | |
1526 | cq_poll = nic->napi[qidx]; | |
1527 | if (!cq_poll) | |
1528 | continue; | |
1529 | napi_disable(&cq_poll->napi); | |
1530 | netif_napi_del(&cq_poll->napi); | |
4863dea3 | 1531 | } |
39ad6eea | 1532 | nicvf_free_cq_poll(nic); |
4863dea3 SG |
1533 | return err; |
1534 | } | |
1535 | ||
4863dea3 SG |
1536 | static int nicvf_change_mtu(struct net_device *netdev, int new_mtu) |
1537 | { | |
1538 | struct nicvf *nic = netdev_priv(netdev); | |
f9aa9dc7 | 1539 | int orig_mtu = netdev->mtu; |
4863dea3 | 1540 | |
4863dea3 | 1541 | netdev->mtu = new_mtu; |
712c3185 SG |
1542 | |
1543 | if (!netif_running(netdev)) | |
1544 | return 0; | |
1545 | ||
f9aa9dc7 DM |
1546 | if (nicvf_update_hw_max_frs(nic, new_mtu)) { |
1547 | netdev->mtu = orig_mtu; | |
4863dea3 | 1548 | return -EINVAL; |
f9aa9dc7 | 1549 | } |
4863dea3 SG |
1550 | |
1551 | return 0; | |
1552 | } | |
1553 | ||
1554 | static int nicvf_set_mac_address(struct net_device *netdev, void *p) | |
1555 | { | |
1556 | struct sockaddr *addr = p; | |
1557 | struct nicvf *nic = netdev_priv(netdev); | |
1558 | ||
1559 | if (!is_valid_ether_addr(addr->sa_data)) | |
1560 | return -EADDRNOTAVAIL; | |
1561 | ||
1562 | memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len); | |
1563 | ||
ce211b17 | 1564 | if (nic->pdev->msix_enabled) { |
4863dea3 SG |
1565 | if (nicvf_hw_set_mac_addr(nic, netdev)) |
1566 | return -EBUSY; | |
bd049a90 PF |
1567 | } else { |
1568 | nic->set_mac_pending = true; | |
1569 | } | |
4863dea3 SG |
1570 | |
1571 | return 0; | |
1572 | } | |
1573 | ||
4863dea3 SG |
1574 | void nicvf_update_lmac_stats(struct nicvf *nic) |
1575 | { | |
1576 | int stat = 0; | |
1577 | union nic_mbx mbx = {}; | |
4863dea3 SG |
1578 | |
1579 | if (!netif_running(nic->netdev)) | |
1580 | return; | |
1581 | ||
1582 | mbx.bgx_stats.msg = NIC_MBOX_MSG_BGX_STATS; | |
1583 | mbx.bgx_stats.vf_id = nic->vf_id; | |
1584 | /* Rx stats */ | |
1585 | mbx.bgx_stats.rx = 1; | |
1586 | while (stat < BGX_RX_STATS_COUNT) { | |
4863dea3 | 1587 | mbx.bgx_stats.idx = stat; |
6051cba7 SG |
1588 | if (nicvf_send_msg_to_pf(nic, &mbx)) |
1589 | return; | |
4863dea3 SG |
1590 | stat++; |
1591 | } | |
1592 | ||
1593 | stat = 0; | |
1594 | ||
1595 | /* Tx stats */ | |
1596 | mbx.bgx_stats.rx = 0; | |
1597 | while (stat < BGX_TX_STATS_COUNT) { | |
4863dea3 | 1598 | mbx.bgx_stats.idx = stat; |
6051cba7 SG |
1599 | if (nicvf_send_msg_to_pf(nic, &mbx)) |
1600 | return; | |
4863dea3 SG |
1601 | stat++; |
1602 | } | |
1603 | } | |
1604 | ||
1605 | void nicvf_update_stats(struct nicvf *nic) | |
1606 | { | |
964cb69b SG |
1607 | int qidx, cpu; |
1608 | u64 tmp_stats = 0; | |
a2dc5ded | 1609 | struct nicvf_hw_stats *stats = &nic->hw_stats; |
964cb69b | 1610 | struct nicvf_drv_stats *drv_stats; |
4863dea3 SG |
1611 | struct queue_set *qs = nic->qs; |
1612 | ||
1613 | #define GET_RX_STATS(reg) \ | |
1614 | nicvf_reg_read(nic, NIC_VNIC_RX_STAT_0_13 | (reg << 3)) | |
1615 | #define GET_TX_STATS(reg) \ | |
1616 | nicvf_reg_read(nic, NIC_VNIC_TX_STAT_0_4 | (reg << 3)) | |
1617 | ||
a2dc5ded SG |
1618 | stats->rx_bytes = GET_RX_STATS(RX_OCTS); |
1619 | stats->rx_ucast_frames = GET_RX_STATS(RX_UCAST); | |
1620 | stats->rx_bcast_frames = GET_RX_STATS(RX_BCAST); | |
1621 | stats->rx_mcast_frames = GET_RX_STATS(RX_MCAST); | |
4863dea3 SG |
1622 | stats->rx_fcs_errors = GET_RX_STATS(RX_FCS); |
1623 | stats->rx_l2_errors = GET_RX_STATS(RX_L2ERR); | |
1624 | stats->rx_drop_red = GET_RX_STATS(RX_RED); | |
a2dc5ded | 1625 | stats->rx_drop_red_bytes = GET_RX_STATS(RX_RED_OCTS); |
4863dea3 | 1626 | stats->rx_drop_overrun = GET_RX_STATS(RX_ORUN); |
a2dc5ded | 1627 | stats->rx_drop_overrun_bytes = GET_RX_STATS(RX_ORUN_OCTS); |
4863dea3 SG |
1628 | stats->rx_drop_bcast = GET_RX_STATS(RX_DRP_BCAST); |
1629 | stats->rx_drop_mcast = GET_RX_STATS(RX_DRP_MCAST); | |
1630 | stats->rx_drop_l3_bcast = GET_RX_STATS(RX_DRP_L3BCAST); | |
1631 | stats->rx_drop_l3_mcast = GET_RX_STATS(RX_DRP_L3MCAST); | |
1632 | ||
964cb69b SG |
1633 | stats->tx_bytes = GET_TX_STATS(TX_OCTS); |
1634 | stats->tx_ucast_frames = GET_TX_STATS(TX_UCAST); | |
1635 | stats->tx_bcast_frames = GET_TX_STATS(TX_BCAST); | |
1636 | stats->tx_mcast_frames = GET_TX_STATS(TX_MCAST); | |
4863dea3 SG |
1637 | stats->tx_drops = GET_TX_STATS(TX_DROP); |
1638 | ||
964cb69b SG |
1639 | /* On T88 pass 2.0, the dummy SQE added for TSO notification |
1640 | * via CQE has 'dont_send' set. Hence HW drops the pkt pointed | |
1641 | * pointed by dummy SQE and results in tx_drops counter being | |
1642 | * incremented. Subtracting it from tx_tso counter will give | |
1643 | * exact tx_drops counter. | |
1644 | */ | |
1645 | if (nic->t88 && nic->hw_tso) { | |
1646 | for_each_possible_cpu(cpu) { | |
1647 | drv_stats = per_cpu_ptr(nic->drv_stats, cpu); | |
1648 | tmp_stats += drv_stats->tx_tso; | |
1649 | } | |
1650 | stats->tx_drops = tmp_stats - stats->tx_drops; | |
1651 | } | |
1652 | stats->tx_frames = stats->tx_ucast_frames + | |
1653 | stats->tx_bcast_frames + | |
1654 | stats->tx_mcast_frames; | |
1655 | stats->rx_frames = stats->rx_ucast_frames + | |
1656 | stats->rx_bcast_frames + | |
1657 | stats->rx_mcast_frames; | |
1658 | stats->rx_drops = stats->rx_drop_red + | |
1659 | stats->rx_drop_overrun; | |
4863dea3 SG |
1660 | |
1661 | /* Update RQ and SQ stats */ | |
1662 | for (qidx = 0; qidx < qs->rq_cnt; qidx++) | |
1663 | nicvf_update_rq_stats(nic, qidx); | |
1664 | for (qidx = 0; qidx < qs->sq_cnt; qidx++) | |
1665 | nicvf_update_sq_stats(nic, qidx); | |
1666 | } | |
1667 | ||
bc1f4470 | 1668 | static void nicvf_get_stats64(struct net_device *netdev, |
1669 | struct rtnl_link_stats64 *stats) | |
4863dea3 SG |
1670 | { |
1671 | struct nicvf *nic = netdev_priv(netdev); | |
a2dc5ded | 1672 | struct nicvf_hw_stats *hw_stats = &nic->hw_stats; |
4863dea3 SG |
1673 | |
1674 | nicvf_update_stats(nic); | |
1675 | ||
a2dc5ded | 1676 | stats->rx_bytes = hw_stats->rx_bytes; |
964cb69b SG |
1677 | stats->rx_packets = hw_stats->rx_frames; |
1678 | stats->rx_dropped = hw_stats->rx_drops; | |
a2dc5ded | 1679 | stats->multicast = hw_stats->rx_mcast_frames; |
4863dea3 | 1680 | |
964cb69b SG |
1681 | stats->tx_bytes = hw_stats->tx_bytes; |
1682 | stats->tx_packets = hw_stats->tx_frames; | |
1683 | stats->tx_dropped = hw_stats->tx_drops; | |
4863dea3 | 1684 | |
4863dea3 SG |
1685 | } |
1686 | ||
1687 | static void nicvf_tx_timeout(struct net_device *dev) | |
1688 | { | |
1689 | struct nicvf *nic = netdev_priv(dev); | |
1690 | ||
bf24e136 | 1691 | netif_warn(nic, tx_err, dev, "Transmit timed out, resetting\n"); |
4863dea3 | 1692 | |
964cb69b | 1693 | this_cpu_inc(nic->drv_stats->tx_timeout); |
4863dea3 SG |
1694 | schedule_work(&nic->reset_task); |
1695 | } | |
1696 | ||
1697 | static void nicvf_reset_task(struct work_struct *work) | |
1698 | { | |
1699 | struct nicvf *nic; | |
1700 | ||
1701 | nic = container_of(work, struct nicvf, reset_task); | |
1702 | ||
1703 | if (!netif_running(nic->netdev)) | |
1704 | return; | |
1705 | ||
1706 | nicvf_stop(nic->netdev); | |
1707 | nicvf_open(nic->netdev); | |
860e9538 | 1708 | netif_trans_update(nic->netdev); |
4863dea3 SG |
1709 | } |
1710 | ||
d77a2384 SG |
1711 | static int nicvf_config_loopback(struct nicvf *nic, |
1712 | netdev_features_t features) | |
1713 | { | |
1714 | union nic_mbx mbx = {}; | |
1715 | ||
1716 | mbx.lbk.msg = NIC_MBOX_MSG_LOOPBACK; | |
1717 | mbx.lbk.vf_id = nic->vf_id; | |
1718 | mbx.lbk.enable = (features & NETIF_F_LOOPBACK) != 0; | |
1719 | ||
1720 | return nicvf_send_msg_to_pf(nic, &mbx); | |
1721 | } | |
1722 | ||
1723 | static netdev_features_t nicvf_fix_features(struct net_device *netdev, | |
1724 | netdev_features_t features) | |
1725 | { | |
1726 | struct nicvf *nic = netdev_priv(netdev); | |
1727 | ||
1728 | if ((features & NETIF_F_LOOPBACK) && | |
1729 | netif_running(netdev) && !nic->loopback_supported) | |
1730 | features &= ~NETIF_F_LOOPBACK; | |
1731 | ||
1732 | return features; | |
1733 | } | |
1734 | ||
aa2e259b SG |
1735 | static int nicvf_set_features(struct net_device *netdev, |
1736 | netdev_features_t features) | |
1737 | { | |
1738 | struct nicvf *nic = netdev_priv(netdev); | |
1739 | netdev_features_t changed = features ^ netdev->features; | |
1740 | ||
1741 | if (changed & NETIF_F_HW_VLAN_CTAG_RX) | |
1742 | nicvf_config_vlan_stripping(nic, features); | |
1743 | ||
d77a2384 SG |
1744 | if ((changed & NETIF_F_LOOPBACK) && netif_running(netdev)) |
1745 | return nicvf_config_loopback(nic, features); | |
1746 | ||
aa2e259b SG |
1747 | return 0; |
1748 | } | |
1749 | ||
05c773f5 SG |
1750 | static void nicvf_set_xdp_queues(struct nicvf *nic, bool bpf_attached) |
1751 | { | |
1752 | u8 cq_count, txq_count; | |
1753 | ||
1754 | /* Set XDP Tx queue count same as Rx queue count */ | |
1755 | if (!bpf_attached) | |
1756 | nic->xdp_tx_queues = 0; | |
1757 | else | |
1758 | nic->xdp_tx_queues = nic->rx_queues; | |
1759 | ||
1760 | /* If queue count > MAX_CMP_QUEUES_PER_QS, then additional qsets | |
1761 | * needs to be allocated, check how many. | |
1762 | */ | |
1763 | txq_count = nic->xdp_tx_queues + nic->tx_queues; | |
1764 | cq_count = max(nic->rx_queues, txq_count); | |
1765 | if (cq_count > MAX_CMP_QUEUES_PER_QS) { | |
1766 | nic->sqs_count = roundup(cq_count, MAX_CMP_QUEUES_PER_QS); | |
1767 | nic->sqs_count = (nic->sqs_count / MAX_CMP_QUEUES_PER_QS) - 1; | |
1768 | } else { | |
1769 | nic->sqs_count = 0; | |
1770 | } | |
1771 | ||
1772 | /* Set primary Qset's resources */ | |
1773 | nic->qs->rq_cnt = min_t(u8, nic->rx_queues, MAX_RCV_QUEUES_PER_QS); | |
1774 | nic->qs->sq_cnt = min_t(u8, txq_count, MAX_SND_QUEUES_PER_QS); | |
1775 | nic->qs->cq_cnt = max_t(u8, nic->qs->rq_cnt, nic->qs->sq_cnt); | |
1776 | ||
1777 | /* Update stack */ | |
1778 | nicvf_set_real_num_queues(nic->netdev, nic->tx_queues, nic->rx_queues); | |
1779 | } | |
1780 | ||
1781 | static int nicvf_xdp_setup(struct nicvf *nic, struct bpf_prog *prog) | |
1782 | { | |
1783 | struct net_device *dev = nic->netdev; | |
1784 | bool if_up = netif_running(nic->netdev); | |
1785 | struct bpf_prog *old_prog; | |
1786 | bool bpf_attached = false; | |
6d0f60b0 | 1787 | int ret = 0; |
05c773f5 SG |
1788 | |
1789 | /* For now just support only the usual MTU sized frames */ | |
1790 | if (prog && (dev->mtu > 1500)) { | |
1791 | netdev_warn(dev, "Jumbo frames not yet supported with XDP, current MTU %d.\n", | |
1792 | dev->mtu); | |
1793 | return -EOPNOTSUPP; | |
1794 | } | |
1795 | ||
05c773f5 SG |
1796 | /* ALL SQs attached to CQs i.e same as RQs, are treated as |
1797 | * XDP Tx queues and more Tx queues are allocated for | |
1798 | * network stack to send pkts out. | |
1799 | * | |
1800 | * No of Tx queues are either same as Rx queues or whatever | |
1801 | * is left in max no of queues possible. | |
1802 | */ | |
1803 | if ((nic->rx_queues + nic->tx_queues) > nic->max_queues) { | |
1804 | netdev_warn(dev, | |
1805 | "Failed to attach BPF prog, RXQs + TXQs > Max %d\n", | |
1806 | nic->max_queues); | |
1807 | return -ENOMEM; | |
1808 | } | |
1809 | ||
1810 | if (if_up) | |
1811 | nicvf_stop(nic->netdev); | |
1812 | ||
1813 | old_prog = xchg(&nic->xdp_prog, prog); | |
1814 | /* Detach old prog, if any */ | |
1815 | if (old_prog) | |
1816 | bpf_prog_put(old_prog); | |
1817 | ||
1818 | if (nic->xdp_prog) { | |
1819 | /* Attach BPF program */ | |
1820 | nic->xdp_prog = bpf_prog_add(nic->xdp_prog, nic->rx_queues - 1); | |
6d0f60b0 | 1821 | if (!IS_ERR(nic->xdp_prog)) { |
05c773f5 | 1822 | bpf_attached = true; |
6d0f60b0 LB |
1823 | } else { |
1824 | ret = PTR_ERR(nic->xdp_prog); | |
1825 | nic->xdp_prog = NULL; | |
1826 | } | |
05c773f5 SG |
1827 | } |
1828 | ||
1829 | /* Calculate Tx queues needed for XDP and network stack */ | |
1830 | nicvf_set_xdp_queues(nic, bpf_attached); | |
1831 | ||
1832 | if (if_up) { | |
1833 | /* Reinitialize interface, clean slate */ | |
1834 | nicvf_open(nic->netdev); | |
1835 | netif_trans_update(nic->netdev); | |
1836 | } | |
1837 | ||
6d0f60b0 | 1838 | return ret; |
05c773f5 SG |
1839 | } |
1840 | ||
f4e63525 | 1841 | static int nicvf_xdp(struct net_device *netdev, struct netdev_bpf *xdp) |
05c773f5 SG |
1842 | { |
1843 | struct nicvf *nic = netdev_priv(netdev); | |
1844 | ||
1845 | /* To avoid checks while retrieving buffer address from CQE_RX, | |
1846 | * do not support XDP for T88 pass1.x silicons which are anyway | |
1847 | * not in use widely. | |
1848 | */ | |
1849 | if (pass1_silicon(nic->pdev)) | |
1850 | return -EOPNOTSUPP; | |
1851 | ||
1852 | switch (xdp->command) { | |
1853 | case XDP_SETUP_PROG: | |
1854 | return nicvf_xdp_setup(nic, xdp->prog); | |
1855 | case XDP_QUERY_PROG: | |
1efde2b6 | 1856 | xdp->prog_id = nic->xdp_prog ? nic->xdp_prog->aux->id : 0; |
05c773f5 SG |
1857 | return 0; |
1858 | default: | |
1859 | return -EINVAL; | |
1860 | } | |
1861 | } | |
1862 | ||
4a875509 SG |
1863 | static int nicvf_config_hwtstamp(struct net_device *netdev, struct ifreq *ifr) |
1864 | { | |
1865 | struct hwtstamp_config config; | |
1866 | struct nicvf *nic = netdev_priv(netdev); | |
1867 | ||
1868 | if (!nic->ptp_clock) | |
1869 | return -ENODEV; | |
1870 | ||
1871 | if (copy_from_user(&config, ifr->ifr_data, sizeof(config))) | |
1872 | return -EFAULT; | |
1873 | ||
1874 | /* reserved for future extensions */ | |
1875 | if (config.flags) | |
1876 | return -EINVAL; | |
1877 | ||
1878 | switch (config.tx_type) { | |
1879 | case HWTSTAMP_TX_OFF: | |
1880 | case HWTSTAMP_TX_ON: | |
1881 | break; | |
1882 | default: | |
1883 | return -ERANGE; | |
1884 | } | |
1885 | ||
1886 | switch (config.rx_filter) { | |
1887 | case HWTSTAMP_FILTER_NONE: | |
1888 | nic->hw_rx_tstamp = false; | |
1889 | break; | |
1890 | case HWTSTAMP_FILTER_ALL: | |
1891 | case HWTSTAMP_FILTER_SOME: | |
1892 | case HWTSTAMP_FILTER_PTP_V1_L4_EVENT: | |
1893 | case HWTSTAMP_FILTER_PTP_V1_L4_SYNC: | |
1894 | case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ: | |
1895 | case HWTSTAMP_FILTER_PTP_V2_L4_EVENT: | |
1896 | case HWTSTAMP_FILTER_PTP_V2_L4_SYNC: | |
1897 | case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ: | |
1898 | case HWTSTAMP_FILTER_PTP_V2_L2_EVENT: | |
1899 | case HWTSTAMP_FILTER_PTP_V2_L2_SYNC: | |
1900 | case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ: | |
1901 | case HWTSTAMP_FILTER_PTP_V2_EVENT: | |
1902 | case HWTSTAMP_FILTER_PTP_V2_SYNC: | |
1903 | case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ: | |
1904 | nic->hw_rx_tstamp = true; | |
1905 | config.rx_filter = HWTSTAMP_FILTER_ALL; | |
1906 | break; | |
1907 | default: | |
1908 | return -ERANGE; | |
1909 | } | |
1910 | ||
1911 | if (netif_running(netdev)) | |
1912 | nicvf_config_hw_rx_tstamp(nic, nic->hw_rx_tstamp); | |
1913 | ||
1914 | if (copy_to_user(ifr->ifr_data, &config, sizeof(config))) | |
1915 | return -EFAULT; | |
1916 | ||
1917 | return 0; | |
1918 | } | |
1919 | ||
1920 | static int nicvf_ioctl(struct net_device *netdev, struct ifreq *req, int cmd) | |
1921 | { | |
1922 | switch (cmd) { | |
1923 | case SIOCSHWTSTAMP: | |
1924 | return nicvf_config_hwtstamp(netdev, req); | |
1925 | default: | |
1926 | return -EOPNOTSUPP; | |
1927 | } | |
1928 | } | |
1929 | ||
469998c8 VL |
1930 | static void __nicvf_set_rx_mode_task(u8 mode, struct xcast_addr_list *mc_addrs, |
1931 | struct nicvf *nic) | |
37c3347e | 1932 | { |
37c3347e | 1933 | union nic_mbx mbx = {}; |
9b5c4dfb | 1934 | int idx; |
37c3347e | 1935 | |
37c3347e VL |
1936 | /* From the inside of VM code flow we have only 128 bits memory |
1937 | * available to send message to host's PF, so send all mc addrs | |
1938 | * one by one, starting from flush command in case if kernel | |
1939 | * requests to configure specific MAC filtering | |
1940 | */ | |
1941 | ||
1942 | /* flush DMAC filters and reset RX mode */ | |
1943 | mbx.xcast.msg = NIC_MBOX_MSG_RESET_XCAST; | |
1944 | nicvf_send_msg_to_pf(nic, &mbx); | |
1945 | ||
469998c8 | 1946 | if (mode & BGX_XCAST_MCAST_FILTER) { |
37c3347e VL |
1947 | /* once enabling filtering, we need to signal to PF to add |
1948 | * its' own LMAC to the filter to accept packets for it. | |
1949 | */ | |
1950 | mbx.xcast.msg = NIC_MBOX_MSG_ADD_MCAST; | |
1951 | mbx.xcast.data.mac = 0; | |
1952 | nicvf_send_msg_to_pf(nic, &mbx); | |
1953 | } | |
1954 | ||
1955 | /* check if we have any specific MACs to be added to PF DMAC filter */ | |
469998c8 | 1956 | if (mc_addrs) { |
37c3347e | 1957 | /* now go through kernel list of MACs and add them one by one */ |
469998c8 | 1958 | for (idx = 0; idx < mc_addrs->count; idx++) { |
37c3347e | 1959 | mbx.xcast.msg = NIC_MBOX_MSG_ADD_MCAST; |
469998c8 | 1960 | mbx.xcast.data.mac = mc_addrs->mc[idx]; |
37c3347e | 1961 | nicvf_send_msg_to_pf(nic, &mbx); |
37c3347e | 1962 | } |
469998c8 | 1963 | kfree(mc_addrs); |
37c3347e VL |
1964 | } |
1965 | ||
1966 | /* and finally set rx mode for PF accordingly */ | |
1967 | mbx.xcast.msg = NIC_MBOX_MSG_SET_XCAST; | |
469998c8 | 1968 | mbx.xcast.data.mode = mode; |
37c3347e VL |
1969 | |
1970 | nicvf_send_msg_to_pf(nic, &mbx); | |
1971 | } | |
1972 | ||
469998c8 VL |
1973 | static void nicvf_set_rx_mode_task(struct work_struct *work_arg) |
1974 | { | |
1975 | struct nicvf_work *vf_work = container_of(work_arg, struct nicvf_work, | |
1976 | work.work); | |
1977 | struct nicvf *nic = container_of(vf_work, struct nicvf, rx_mode_work); | |
1978 | u8 mode; | |
1979 | struct xcast_addr_list *mc; | |
1980 | ||
1981 | if (!vf_work) | |
1982 | return; | |
1983 | ||
1984 | /* Save message data locally to prevent them from | |
1985 | * being overwritten by next ndo_set_rx_mode call(). | |
1986 | */ | |
1987 | spin_lock(&nic->rx_mode_wq_lock); | |
1988 | mode = vf_work->mode; | |
1989 | mc = vf_work->mc; | |
1990 | vf_work->mc = NULL; | |
1991 | spin_unlock(&nic->rx_mode_wq_lock); | |
1992 | ||
1993 | __nicvf_set_rx_mode_task(mode, mc, nic); | |
1994 | } | |
1995 | ||
37c3347e VL |
1996 | static void nicvf_set_rx_mode(struct net_device *netdev) |
1997 | { | |
1998 | struct nicvf *nic = netdev_priv(netdev); | |
1999 | struct netdev_hw_addr *ha; | |
2000 | struct xcast_addr_list *mc_list = NULL; | |
2001 | u8 mode = 0; | |
2002 | ||
2003 | if (netdev->flags & IFF_PROMISC) { | |
2004 | mode = BGX_XCAST_BCAST_ACCEPT | BGX_XCAST_MCAST_ACCEPT; | |
2005 | } else { | |
2006 | if (netdev->flags & IFF_BROADCAST) | |
2007 | mode |= BGX_XCAST_BCAST_ACCEPT; | |
2008 | ||
2009 | if (netdev->flags & IFF_ALLMULTI) { | |
2010 | mode |= BGX_XCAST_MCAST_ACCEPT; | |
2011 | } else if (netdev->flags & IFF_MULTICAST) { | |
2012 | mode |= BGX_XCAST_MCAST_FILTER; | |
2013 | /* here we need to copy mc addrs */ | |
2014 | if (netdev_mc_count(netdev)) { | |
9b5c4dfb VL |
2015 | mc_list = kmalloc(offsetof(typeof(*mc_list), |
2016 | mc[netdev_mc_count(netdev)]), | |
2017 | GFP_ATOMIC); | |
2018 | if (unlikely(!mc_list)) | |
2019 | return; | |
2020 | mc_list->count = 0; | |
37c3347e | 2021 | netdev_hw_addr_list_for_each(ha, &netdev->mc) { |
9b5c4dfb | 2022 | mc_list->mc[mc_list->count] = |
37c3347e | 2023 | ether_addr_to_u64(ha->addr); |
37c3347e VL |
2024 | mc_list->count++; |
2025 | } | |
2026 | } | |
2027 | } | |
2028 | } | |
469998c8 VL |
2029 | spin_lock(&nic->rx_mode_wq_lock); |
2030 | kfree(nic->rx_mode_work.mc); | |
37c3347e VL |
2031 | nic->rx_mode_work.mc = mc_list; |
2032 | nic->rx_mode_work.mode = mode; | |
469998c8 VL |
2033 | queue_delayed_work(nicvf_rx_mode_wq, &nic->rx_mode_work.work, 0); |
2034 | spin_unlock(&nic->rx_mode_wq_lock); | |
37c3347e VL |
2035 | } |
2036 | ||
4863dea3 SG |
2037 | static const struct net_device_ops nicvf_netdev_ops = { |
2038 | .ndo_open = nicvf_open, | |
2039 | .ndo_stop = nicvf_stop, | |
2040 | .ndo_start_xmit = nicvf_xmit, | |
2041 | .ndo_change_mtu = nicvf_change_mtu, | |
2042 | .ndo_set_mac_address = nicvf_set_mac_address, | |
2043 | .ndo_get_stats64 = nicvf_get_stats64, | |
2044 | .ndo_tx_timeout = nicvf_tx_timeout, | |
d77a2384 | 2045 | .ndo_fix_features = nicvf_fix_features, |
aa2e259b | 2046 | .ndo_set_features = nicvf_set_features, |
f4e63525 | 2047 | .ndo_bpf = nicvf_xdp, |
4a875509 | 2048 | .ndo_do_ioctl = nicvf_ioctl, |
37c3347e | 2049 | .ndo_set_rx_mode = nicvf_set_rx_mode, |
4863dea3 SG |
2050 | }; |
2051 | ||
2052 | static int nicvf_probe(struct pci_dev *pdev, const struct pci_device_id *ent) | |
2053 | { | |
2054 | struct device *dev = &pdev->dev; | |
2055 | struct net_device *netdev; | |
2056 | struct nicvf *nic; | |
92dc8769 | 2057 | int err, qcount; |
7ceb8a13 | 2058 | u16 sdevid; |
4a875509 SG |
2059 | struct cavium_ptp *ptp_clock; |
2060 | ||
2061 | ptp_clock = cavium_ptp_get(); | |
2062 | if (IS_ERR(ptp_clock)) { | |
2063 | if (PTR_ERR(ptp_clock) == -ENODEV) | |
2064 | /* In virtualized environment we proceed without ptp */ | |
2065 | ptp_clock = NULL; | |
2066 | else | |
2067 | return PTR_ERR(ptp_clock); | |
2068 | } | |
4863dea3 SG |
2069 | |
2070 | err = pci_enable_device(pdev); | |
2071 | if (err) { | |
2072 | dev_err(dev, "Failed to enable PCI device\n"); | |
2073 | return err; | |
2074 | } | |
2075 | ||
2076 | err = pci_request_regions(pdev, DRV_NAME); | |
2077 | if (err) { | |
2078 | dev_err(dev, "PCI request regions failed 0x%x\n", err); | |
2079 | goto err_disable_device; | |
2080 | } | |
2081 | ||
2082 | err = pci_set_dma_mask(pdev, DMA_BIT_MASK(48)); | |
2083 | if (err) { | |
2084 | dev_err(dev, "Unable to get usable DMA configuration\n"); | |
2085 | goto err_release_regions; | |
2086 | } | |
2087 | ||
2088 | err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(48)); | |
2089 | if (err) { | |
2090 | dev_err(dev, "unable to get 48-bit DMA for consistent allocations\n"); | |
2091 | goto err_release_regions; | |
2092 | } | |
2093 | ||
3a397ebe | 2094 | qcount = netif_get_num_default_rss_queues(); |
92dc8769 SG |
2095 | |
2096 | /* Restrict multiqset support only for host bound VFs */ | |
2097 | if (pdev->is_virtfn) { | |
2098 | /* Set max number of queues per VF */ | |
3a397ebe SG |
2099 | qcount = min_t(int, num_online_cpus(), |
2100 | (MAX_SQS_PER_VF + 1) * MAX_CMP_QUEUES_PER_QS); | |
92dc8769 SG |
2101 | } |
2102 | ||
2103 | netdev = alloc_etherdev_mqs(sizeof(struct nicvf), qcount, qcount); | |
4863dea3 SG |
2104 | if (!netdev) { |
2105 | err = -ENOMEM; | |
2106 | goto err_release_regions; | |
2107 | } | |
2108 | ||
2109 | pci_set_drvdata(pdev, netdev); | |
2110 | ||
2111 | SET_NETDEV_DEV(netdev, &pdev->dev); | |
2112 | ||
2113 | nic = netdev_priv(netdev); | |
2114 | nic->netdev = netdev; | |
2115 | nic->pdev = pdev; | |
92dc8769 SG |
2116 | nic->pnicvf = nic; |
2117 | nic->max_queues = qcount; | |
87de0838 SG |
2118 | /* If no of CPUs are too low, there won't be any queues left |
2119 | * for XDP_TX, hence double it. | |
2120 | */ | |
2121 | if (!nic->t88) | |
2122 | nic->max_queues *= 2; | |
4a875509 | 2123 | nic->ptp_clock = ptp_clock; |
4863dea3 SG |
2124 | |
2125 | /* MAP VF's configuration registers */ | |
2126 | nic->reg_base = pcim_iomap(pdev, PCI_CFG_REG_BAR_NUM, 0); | |
2127 | if (!nic->reg_base) { | |
2128 | dev_err(dev, "Cannot map config register space, aborting\n"); | |
2129 | err = -ENOMEM; | |
2130 | goto err_free_netdev; | |
2131 | } | |
2132 | ||
964cb69b SG |
2133 | nic->drv_stats = netdev_alloc_pcpu_stats(struct nicvf_drv_stats); |
2134 | if (!nic->drv_stats) { | |
2135 | err = -ENOMEM; | |
2136 | goto err_free_netdev; | |
2137 | } | |
2138 | ||
4863dea3 SG |
2139 | err = nicvf_set_qset_resources(nic); |
2140 | if (err) | |
2141 | goto err_free_netdev; | |
2142 | ||
4863dea3 SG |
2143 | /* Check if PF is alive and get MAC address for this VF */ |
2144 | err = nicvf_register_misc_interrupt(nic); | |
2145 | if (err) | |
2146 | goto err_free_netdev; | |
2147 | ||
92dc8769 SG |
2148 | nicvf_send_vf_struct(nic); |
2149 | ||
8d210d54 SG |
2150 | if (!pass1_silicon(nic->pdev)) |
2151 | nic->hw_tso = true; | |
2152 | ||
83abb7d7 SG |
2153 | /* Get iommu domain for iova to physical addr conversion */ |
2154 | nic->iommu_domain = iommu_get_domain_for_dev(dev); | |
2155 | ||
7ceb8a13 SG |
2156 | pci_read_config_word(nic->pdev, PCI_SUBSYSTEM_ID, &sdevid); |
2157 | if (sdevid == 0xA134) | |
2158 | nic->t88 = true; | |
2159 | ||
92dc8769 SG |
2160 | /* Check if this VF is in QS only mode */ |
2161 | if (nic->sqs_mode) | |
2162 | return 0; | |
2163 | ||
2164 | err = nicvf_set_real_num_queues(netdev, nic->tx_queues, nic->rx_queues); | |
2165 | if (err) | |
2166 | goto err_unregister_interrupts; | |
2167 | ||
3a9024f5 TS |
2168 | netdev->hw_features = (NETIF_F_RXCSUM | NETIF_F_SG | |
2169 | NETIF_F_TSO | NETIF_F_GRO | NETIF_F_TSO6 | | |
2170 | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | | |
92dc8769 SG |
2171 | NETIF_F_HW_VLAN_CTAG_RX); |
2172 | ||
2173 | netdev->hw_features |= NETIF_F_RXHASH; | |
aa2e259b SG |
2174 | |
2175 | netdev->features |= netdev->hw_features; | |
d77a2384 | 2176 | netdev->hw_features |= NETIF_F_LOOPBACK; |
38bb5d4f | 2177 | |
3a9024f5 TS |
2178 | netdev->vlan_features = NETIF_F_SG | NETIF_F_IP_CSUM | |
2179 | NETIF_F_IPV6_CSUM | NETIF_F_TSO | NETIF_F_TSO6; | |
4863dea3 SG |
2180 | |
2181 | netdev->netdev_ops = &nicvf_netdev_ops; | |
3d7a8aaa | 2182 | netdev->watchdog_timeo = NICVF_TX_TIMEOUT; |
4863dea3 | 2183 | |
109cc165 JW |
2184 | /* MTU range: 64 - 9200 */ |
2185 | netdev->min_mtu = NIC_HW_MIN_FRS; | |
2186 | netdev->max_mtu = NIC_HW_MAX_FRS; | |
2187 | ||
4863dea3 SG |
2188 | INIT_WORK(&nic->reset_task, nicvf_reset_task); |
2189 | ||
37c3347e | 2190 | INIT_DELAYED_WORK(&nic->rx_mode_work.work, nicvf_set_rx_mode_task); |
469998c8 | 2191 | spin_lock_init(&nic->rx_mode_wq_lock); |
37c3347e | 2192 | |
4863dea3 SG |
2193 | err = register_netdev(netdev); |
2194 | if (err) { | |
2195 | dev_err(dev, "Failed to register netdevice\n"); | |
2196 | goto err_unregister_interrupts; | |
2197 | } | |
2198 | ||
2199 | nic->msg_enable = debug; | |
2200 | ||
2201 | nicvf_set_ethtool_ops(netdev); | |
2202 | ||
2203 | return 0; | |
2204 | ||
2205 | err_unregister_interrupts: | |
2206 | nicvf_unregister_interrupts(nic); | |
2207 | err_free_netdev: | |
2208 | pci_set_drvdata(pdev, NULL); | |
964cb69b SG |
2209 | if (nic->drv_stats) |
2210 | free_percpu(nic->drv_stats); | |
4863dea3 SG |
2211 | free_netdev(netdev); |
2212 | err_release_regions: | |
2213 | pci_release_regions(pdev); | |
2214 | err_disable_device: | |
2215 | pci_disable_device(pdev); | |
2216 | return err; | |
2217 | } | |
2218 | ||
2219 | static void nicvf_remove(struct pci_dev *pdev) | |
2220 | { | |
2221 | struct net_device *netdev = pci_get_drvdata(pdev); | |
7750130d PF |
2222 | struct nicvf *nic; |
2223 | struct net_device *pnetdev; | |
2224 | ||
2225 | if (!netdev) | |
2226 | return; | |
2227 | ||
2228 | nic = netdev_priv(netdev); | |
2229 | pnetdev = nic->pnicvf->netdev; | |
4863dea3 | 2230 | |
37c3347e VL |
2231 | cancel_delayed_work_sync(&nic->rx_mode_work.work); |
2232 | ||
92dc8769 SG |
2233 | /* Check if this Qset is assigned to different VF. |
2234 | * If yes, clean primary and all secondary Qsets. | |
2235 | */ | |
2236 | if (pnetdev && (pnetdev->reg_state == NETREG_REGISTERED)) | |
2237 | unregister_netdev(pnetdev); | |
4863dea3 SG |
2238 | nicvf_unregister_interrupts(nic); |
2239 | pci_set_drvdata(pdev, NULL); | |
964cb69b SG |
2240 | if (nic->drv_stats) |
2241 | free_percpu(nic->drv_stats); | |
4a875509 | 2242 | cavium_ptp_put(nic->ptp_clock); |
4863dea3 SG |
2243 | free_netdev(netdev); |
2244 | pci_release_regions(pdev); | |
2245 | pci_disable_device(pdev); | |
2246 | } | |
2247 | ||
4adf4351 SG |
2248 | static void nicvf_shutdown(struct pci_dev *pdev) |
2249 | { | |
2250 | nicvf_remove(pdev); | |
2251 | } | |
2252 | ||
4863dea3 SG |
2253 | static struct pci_driver nicvf_driver = { |
2254 | .name = DRV_NAME, | |
2255 | .id_table = nicvf_id_table, | |
2256 | .probe = nicvf_probe, | |
2257 | .remove = nicvf_remove, | |
4adf4351 | 2258 | .shutdown = nicvf_shutdown, |
4863dea3 SG |
2259 | }; |
2260 | ||
2261 | static int __init nicvf_init_module(void) | |
2262 | { | |
2263 | pr_info("%s, ver %s\n", DRV_NAME, DRV_VERSION); | |
37c3347e VL |
2264 | nicvf_rx_mode_wq = alloc_ordered_workqueue("nicvf_generic", |
2265 | WQ_MEM_RECLAIM); | |
4863dea3 SG |
2266 | return pci_register_driver(&nicvf_driver); |
2267 | } | |
2268 | ||
2269 | static void __exit nicvf_cleanup_module(void) | |
2270 | { | |
37c3347e VL |
2271 | if (nicvf_rx_mode_wq) { |
2272 | destroy_workqueue(nicvf_rx_mode_wq); | |
2273 | nicvf_rx_mode_wq = NULL; | |
2274 | } | |
4863dea3 SG |
2275 | pci_unregister_driver(&nicvf_driver); |
2276 | } | |
2277 | ||
2278 | module_init(nicvf_init_module); | |
2279 | module_exit(nicvf_cleanup_module); |