drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c

   1 /*
   2  * This file is part of the Chelsio T4 Ethernet driver for Linux.
   3  *
   4  * Copyright (c) 2003-2016 Chelsio Communications, Inc. All rights reserved.
   5  *
   6  * This software is available to you under a choice of one of two
   7  * licenses.  You may choose to be licensed under the terms of the GNU
   8  * General Public License (GPL) Version 2, available from the file
   9  * COPYING in the main directory of this source tree, or the
  10  * OpenIB.org BSD license below:
  11  *
  12  *     Redistribution and use in source and binary forms, with or
  13  *     without modification, are permitted provided that the following
  14  *     conditions are met:
  15  *
  16  *      - Redistributions of source code must retain the above
  17  *        copyright notice, this list of conditions and the following
  18  *        disclaimer.
  19  *
  20  *      - Redistributions in binary form must reproduce the above
  21  *        copyright notice, this list of conditions and the following
  22  *        disclaimer in the documentation and/or other materials
  23  *        provided with the distribution.
  24  *
  25  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  26  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  27  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  28  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  29  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  30  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  31  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  32  * SOFTWARE.
  33  */
  34
  35 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  36
  37 #include <linux/bitmap.h>
  38 #include <linux/crc32.h>
  39 #include <linux/ctype.h>
  40 #include <linux/debugfs.h>
  41 #include <linux/err.h>
  42 #include <linux/etherdevice.h>
  43 #include <linux/firmware.h>
  44 #include <linux/if.h>
  45 #include <linux/if_vlan.h>
  46 #include <linux/init.h>
  47 #include <linux/log2.h>
  48 #include <linux/mdio.h>
  49 #include <linux/module.h>
  50 #include <linux/moduleparam.h>
  51 #include <linux/mutex.h>
  52 #include <linux/netdevice.h>
  53 #include <linux/pci.h>
  54 #include <linux/aer.h>
  55 #include <linux/rtnetlink.h>
  56 #include <linux/sched.h>
  57 #include <linux/seq_file.h>
  58 #include <linux/sockios.h>
  59 #include <linux/vmalloc.h>
  60 #include <linux/workqueue.h>
  61 #include <net/neighbour.h>
  62 #include <net/netevent.h>
  63 #include <net/addrconf.h>
  64 #include <net/bonding.h>
  65 #include <linux/uaccess.h>
  66 #include <linux/crash_dump.h>
  67 #include <net/udp_tunnel.h>
  68
  69 #include "cxgb4.h"
  70 #include "cxgb4_filter.h"
  71 #include "t4_regs.h"
  72 #include "t4_values.h"
  73 #include "t4_msg.h"
  74 #include "t4fw_api.h"
  75 #include "t4fw_version.h"
  76 #include "cxgb4_dcb.h"
  77 #include "srq.h"
  78 #include "cxgb4_debugfs.h"
  79 #include "clip_tbl.h"
  80 #include "l2t.h"
  81 #include "smt.h"
  82 #include "sched.h"
  83 #include "cxgb4_tc_u32.h"
  84 #include "cxgb4_tc_flower.h"
  85 #include "cxgb4_ptp.h"
  86 #include "cxgb4_cudbg.h"
  87
  88 char cxgb4_driver_name[] = KBUILD_MODNAME;
  89
  90 #ifdef DRV_VERSION
  91 #undef DRV_VERSION
  92 #endif
  93 #define DRV_VERSION "2.0.0-ko"
  94 const char cxgb4_driver_version[] = DRV_VERSION;
  95 #define DRV_DESC "Chelsio T4/T5/T6 Network Driver"
  96
  97 #define DFLT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_LINK | \
  98                          NETIF_MSG_TIMER | NETIF_MSG_IFDOWN | NETIF_MSG_IFUP |\
  99                          NETIF_MSG_RX_ERR | NETIF_MSG_TX_ERR)
 100
 101 /* Macros needed to support the PCI Device ID Table ...
 102  */
 103 #define CH_PCI_DEVICE_ID_TABLE_DEFINE_BEGIN \
 104         static const struct pci_device_id cxgb4_pci_tbl[] = {
 105 #define CXGB4_UNIFIED_PF 0x4
 106
 107 #define CH_PCI_DEVICE_ID_FUNCTION CXGB4_UNIFIED_PF
 108
 109 /* Include PCI Device IDs for both PF4 and PF0-3 so our PCI probe() routine is
 110  * called for both.
 111  */
 112 #define CH_PCI_DEVICE_ID_FUNCTION2 0x0
 113
 114 #define CH_PCI_ID_TABLE_ENTRY(devid) \
 115                 {PCI_VDEVICE(CHELSIO, (devid)), CXGB4_UNIFIED_PF}
 116
 117 #define CH_PCI_DEVICE_ID_TABLE_DEFINE_END \
 118                 { 0, } \
 119         }
 120
 121 #include "t4_pci_id_tbl.h"
 122
 123 #define FW4_FNAME "cxgb4/t4fw.bin"
 124 #define FW5_FNAME "cxgb4/t5fw.bin"
 125 #define FW6_FNAME "cxgb4/t6fw.bin"
 126 #define FW4_CFNAME "cxgb4/t4-config.txt"
 127 #define FW5_CFNAME "cxgb4/t5-config.txt"
 128 #define FW6_CFNAME "cxgb4/t6-config.txt"
 129 #define PHY_AQ1202_FIRMWARE "cxgb4/aq1202_fw.cld"
 130 #define PHY_BCM84834_FIRMWARE "cxgb4/bcm8483.bin"
 131 #define PHY_AQ1202_DEVICEID 0x4409
 132 #define PHY_BCM84834_DEVICEID 0x4486
 133
 134 MODULE_DESCRIPTION(DRV_DESC);
 135 MODULE_AUTHOR("Chelsio Communications");
 136 MODULE_LICENSE("Dual BSD/GPL");
 137 MODULE_VERSION(DRV_VERSION);
 138 MODULE_DEVICE_TABLE(pci, cxgb4_pci_tbl);
 139 MODULE_FIRMWARE(FW4_FNAME);
 140 MODULE_FIRMWARE(FW5_FNAME);
 141 MODULE_FIRMWARE(FW6_FNAME);
 142
 143 /*
 144  * The driver uses the best interrupt scheme available on a platform in the
 145  * order MSI-X, MSI, legacy INTx interrupts.  This parameter determines which
 146  * of these schemes the driver may consider as follows:
 147  *
 148  * msi = 2: choose from among all three options
 149  * msi = 1: only consider MSI and INTx interrupts
 150  * msi = 0: force INTx interrupts
 151  */
 152 static int msi = 2;
 153
 154 module_param(msi, int, 0644);
 155 MODULE_PARM_DESC(msi, "whether to use INTx (0), MSI (1) or MSI-X (2)");
 156
 157 /*
 158  * Normally we tell the chip to deliver Ingress Packets into our DMA buffers
 159  * offset by 2 bytes in order to have the IP headers line up on 4-byte
 160  * boundaries.  This is a requirement for many architectures which will throw
 161  * a machine check fault if an attempt is made to access one of the 4-byte IP
 162  * header fields on a non-4-byte boundary.  And it's a major performance issue
 163  * even on some architectures which allow it like some implementations of the
 164  * x86 ISA.  However, some architectures don't mind this and for some very
 165  * edge-case performance sensitive applications (like forwarding large volumes
 166  * of small packets), setting this DMA offset to 0 will decrease the number of
 167  * PCI-E Bus transfers enough to measurably affect performance.
 168  */
 169 static int rx_dma_offset = 2;
 170
 171 /* TX Queue select used to determine what algorithm to use for selecting TX
 172  * queue. Select between the kernel provided function (select_queue=0) or user
 173  * cxgb_select_queue function (select_queue=1)
 174  *
 175  * Default: select_queue=0
 176  */
 177 static int select_queue;
 178 module_param(select_queue, int, 0644);
 179 MODULE_PARM_DESC(select_queue,
 180                  "Select between kernel provided method of selecting or driver method of selecting TX queue. Default is kernel method.");
 181
 182 static struct dentry *cxgb4_debugfs_root;
 183
 184 LIST_HEAD(adapter_list);
 185 DEFINE_MUTEX(uld_mutex);
 186
 187 static void link_report(struct net_device *dev)
 188 {
 189         if (!netif_carrier_ok(dev))
 190                 netdev_info(dev, "link down\n");
 191         else {
 192                 static const char *fc[] = { "no", "Rx", "Tx", "Tx/Rx" };
 193
 194                 const char *s;
 195                 const struct port_info *p = netdev_priv(dev);
 196
 197                 switch (p->link_cfg.speed) {
 198                 case 100:
 199                         s = "100Mbps";
 200                         break;
 201                 case 1000:
 202                         s = "1Gbps";
 203                         break;
 204                 case 10000:
 205                         s = "10Gbps";
 206                         break;
 207                 case 25000:
 208                         s = "25Gbps";
 209                         break;
 210                 case 40000:
 211                         s = "40Gbps";
 212                         break;
 213                 case 50000:
 214                         s = "50Gbps";
 215                         break;
 216                 case 100000:
 217                         s = "100Gbps";
 218                         break;
 219                 default:
 220                         pr_info("%s: unsupported speed: %d\n",
 221                                 dev->name, p->link_cfg.speed);
 222                         return;
 223                 }
 224
 225                 netdev_info(dev, "link up, %s, full-duplex, %s PAUSE\n", s,
 226                             fc[p->link_cfg.fc]);
 227         }
 228 }
 229
 230 #ifdef CONFIG_CHELSIO_T4_DCB
 231 /* Set up/tear down Data Center Bridging Priority mapping for a net device. */
 232 static void dcb_tx_queue_prio_enable(struct net_device *dev, int enable)
 233 {
 234         struct port_info *pi = netdev_priv(dev);
 235         struct adapter *adap = pi->adapter;
 236         struct sge_eth_txq *txq = &adap->sge.ethtxq[pi->first_qset];
 237         int i;
 238
 239         /* We use a simple mapping of Port TX Queue Index to DCB
 240          * Priority when we're enabling DCB.
 241          */
 242         for (i = 0; i < pi->nqsets; i++, txq++) {
 243                 u32 name, value;
 244                 int err;
 245
 246                 name = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DMAQ) |
 247                         FW_PARAMS_PARAM_X_V(
 248                                 FW_PARAMS_PARAM_DMAQ_EQ_DCBPRIO_ETH) |
 249                         FW_PARAMS_PARAM_YZ_V(txq->q.cntxt_id));
 250                 value = enable ? i : 0xffffffff;
 251
 252                 /* Since we can be called while atomic (from "interrupt
 253                  * level") we need to issue the Set Parameters Commannd
 254                  * without sleeping (timeout < 0).
 255                  */
 256                 err = t4_set_params_timeout(adap, adap->mbox, adap->pf, 0, 1,
 257                                             &name, &value,
 258                                             -FW_CMD_MAX_TIMEOUT);
 259
 260                 if (err)
 261                         dev_err(adap->pdev_dev,
 262                                 "Can't %s DCB Priority on port %d, TX Queue %d: err=%d\n",
 263                                 enable ? "set" : "unset", pi->port_id, i, -err);
 264                 else
 265                         txq->dcb_prio = enable ? value : 0;
 266         }
 267 }
 268
 269 int cxgb4_dcb_enabled(const struct net_device *dev)
 270 {
 271         struct port_info *pi = netdev_priv(dev);
 272
 273         if (!pi->dcb.enabled)
 274                 return 0;
 275
 276         return ((pi->dcb.state == CXGB4_DCB_STATE_FW_ALLSYNCED) ||
 277                 (pi->dcb.state == CXGB4_DCB_STATE_HOST));
 278 }
 279 #endif /* CONFIG_CHELSIO_T4_DCB */
 280
 281 void t4_os_link_changed(struct adapter *adapter, int port_id, int link_stat)
 282 {
 283         struct net_device *dev = adapter->port[port_id];
 284
 285         /* Skip changes from disabled ports. */
 286         if (netif_running(dev) && link_stat != netif_carrier_ok(dev)) {
 287                 if (link_stat)
 288                         netif_carrier_on(dev);
 289                 else {
 290 #ifdef CONFIG_CHELSIO_T4_DCB
 291                         if (cxgb4_dcb_enabled(dev)) {
 292                                 cxgb4_dcb_reset(dev);
 293                                 dcb_tx_queue_prio_enable(dev, false);
 294                         }
 295 #endif /* CONFIG_CHELSIO_T4_DCB */
 296                         netif_carrier_off(dev);
 297                 }
 298
 299                 link_report(dev);
 300         }
 301 }
 302
 303 void t4_os_portmod_changed(struct adapter *adap, int port_id)
 304 {
 305         static const char *mod_str[] = {
 306                 NULL, "LR", "SR", "ER", "passive DA", "active DA", "LRM"
 307         };
 308
 309         struct net_device *dev = adap->port[port_id];
 310         struct port_info *pi = netdev_priv(dev);
 311
 312         if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
 313                 netdev_info(dev, "port module unplugged\n");
 314         else if (pi->mod_type < ARRAY_SIZE(mod_str))
 315                 netdev_info(dev, "%s module inserted\n", mod_str[pi->mod_type]);
 316         else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
 317                 netdev_info(dev, "%s: unsupported port module inserted\n",
 318                             dev->name);
 319         else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
 320                 netdev_info(dev, "%s: unknown port module inserted\n",
 321                             dev->name);
 322         else if (pi->mod_type == FW_PORT_MOD_TYPE_ERROR)
 323                 netdev_info(dev, "%s: transceiver module error\n", dev->name);
 324         else
 325                 netdev_info(dev, "%s: unknown module type %d inserted\n",
 326                             dev->name, pi->mod_type);
 327
 328         /* If the interface is running, then we'll need any "sticky" Link
 329          * Parameters redone with a new Transceiver Module.
 330          */
 331         pi->link_cfg.redo_l1cfg = netif_running(dev);
 332 }
 333
 334 int dbfifo_int_thresh = 10; /* 10 == 640 entry threshold */
 335 module_param(dbfifo_int_thresh, int, 0644);
 336 MODULE_PARM_DESC(dbfifo_int_thresh, "doorbell fifo interrupt threshold");
 337
 338 /*
 339  * usecs to sleep while draining the dbfifo
 340  */
 341 static int dbfifo_drain_delay = 1000;
 342 module_param(dbfifo_drain_delay, int, 0644);
 343 MODULE_PARM_DESC(dbfifo_drain_delay,
 344                  "usecs to sleep while draining the dbfifo");
 345
 346 static inline int cxgb4_set_addr_hash(struct port_info *pi)
 347 {
 348         struct adapter *adap = pi->adapter;
 349         u64 vec = 0;
 350         bool ucast = false;
 351         struct hash_mac_addr *entry;
 352
 353         /* Calculate the hash vector for the updated list and program it */
 354         list_for_each_entry(entry, &adap->mac_hlist, list) {
 355                 ucast |= is_unicast_ether_addr(entry->addr);
 356                 vec |= (1ULL << hash_mac_addr(entry->addr));
 357         }
 358         return t4_set_addr_hash(adap, adap->mbox, pi->viid, ucast,
 359                                 vec, false);
 360 }
 361
 362 static int cxgb4_mac_sync(struct net_device *netdev, const u8 *mac_addr)
 363 {
 364         struct port_info *pi = netdev_priv(netdev);
 365         struct adapter *adap = pi->adapter;
 366         int ret;
 367         u64 mhash = 0;
 368         u64 uhash = 0;
 369         bool free = false;
 370         bool ucast = is_unicast_ether_addr(mac_addr);
 371         const u8 *maclist[1] = {mac_addr};
 372         struct hash_mac_addr *new_entry;
 373
 374         ret = t4_alloc_mac_filt(adap, adap->mbox, pi->viid, free, 1, maclist,
 375                                 NULL, ucast ? &uhash : &mhash, false);
 376         if (ret < 0)
 377                 goto out;
 378         /* if hash != 0, then add the addr to hash addr list
 379          * so on the end we will calculate the hash for the
 380          * list and program it
 381          */
 382         if (uhash || mhash) {
 383                 new_entry = kzalloc(sizeof(*new_entry), GFP_ATOMIC);
 384                 if (!new_entry)
 385                         return -ENOMEM;
 386                 ether_addr_copy(new_entry->addr, mac_addr);
 387                 list_add_tail(&new_entry->list, &adap->mac_hlist);
 388                 ret = cxgb4_set_addr_hash(pi);
 389         }
 390 out:
 391         return ret < 0 ? ret : 0;
 392 }
 393
 394 static int cxgb4_mac_unsync(struct net_device *netdev, const u8 *mac_addr)
 395 {
 396         struct port_info *pi = netdev_priv(netdev);
 397         struct adapter *adap = pi->adapter;
 398         int ret;
 399         const u8 *maclist[1] = {mac_addr};
 400         struct hash_mac_addr *entry, *tmp;
 401
 402         /* If the MAC address to be removed is in the hash addr
 403          * list, delete it from the list and update hash vector
 404          */
 405         list_for_each_entry_safe(entry, tmp, &adap->mac_hlist, list) {
 406                 if (ether_addr_equal(entry->addr, mac_addr)) {
 407                         list_del(&entry->list);
 408                         kfree(entry);
 409                         return cxgb4_set_addr_hash(pi);
 410                 }
 411         }
 412
 413         ret = t4_free_mac_filt(adap, adap->mbox, pi->viid, 1, maclist, false);
 414         return ret < 0 ? -EINVAL : 0;
 415 }
 416
 417 /*
 418  * Set Rx properties of a port, such as promiscruity, address filters, and MTU.
 419  * If @mtu is -1 it is left unchanged.
 420  */
 421 static int set_rxmode(struct net_device *dev, int mtu, bool sleep_ok)
 422 {
 423         struct port_info *pi = netdev_priv(dev);
 424         struct adapter *adapter = pi->adapter;
 425
 426         __dev_uc_sync(dev, cxgb4_mac_sync, cxgb4_mac_unsync);
 427         __dev_mc_sync(dev, cxgb4_mac_sync, cxgb4_mac_unsync);
 428
 429         return t4_set_rxmode(adapter, adapter->mbox, pi->viid, mtu,
 430                              (dev->flags & IFF_PROMISC) ? 1 : 0,
 431                              (dev->flags & IFF_ALLMULTI) ? 1 : 0, 1, -1,
 432                              sleep_ok);
 433 }
 434
 435 /**
 436  *      link_start - enable a port
 437  *      @dev: the port to enable
 438  *
 439  *      Performs the MAC and PHY actions needed to enable a port.
 440  */
 441 static int link_start(struct net_device *dev)
 442 {
 443         int ret;
 444         struct port_info *pi = netdev_priv(dev);
 445         unsigned int mb = pi->adapter->pf;
 446
 447         /*
 448          * We do not set address filters and promiscuity here, the stack does
 449          * that step explicitly.
 450          */
 451         ret = t4_set_rxmode(pi->adapter, mb, pi->viid, dev->mtu, -1, -1, -1,
 452                             !!(dev->features & NETIF_F_HW_VLAN_CTAG_RX), true);
 453         if (ret == 0) {
 454                 ret = t4_change_mac(pi->adapter, mb, pi->viid,
 455                                     pi->xact_addr_filt, dev->dev_addr, true,
 456                                     true);
 457                 if (ret >= 0) {
 458                         pi->xact_addr_filt = ret;
 459                         ret = 0;
 460                 }
 461         }
 462         if (ret == 0)
 463                 ret = t4_link_l1cfg(pi->adapter, mb, pi->tx_chan,
 464                                     &pi->link_cfg);
 465         if (ret == 0) {
 466                 local_bh_disable();
 467                 ret = t4_enable_pi_params(pi->adapter, mb, pi, true,
 468                                           true, CXGB4_DCB_ENABLED);
 469                 local_bh_enable();
 470         }
 471
 472         return ret;
 473 }
 474
 475 #ifdef CONFIG_CHELSIO_T4_DCB
 476 /* Handle a Data Center Bridging update message from the firmware. */
 477 static void dcb_rpl(struct adapter *adap, const struct fw_port_cmd *pcmd)
 478 {
 479         int port = FW_PORT_CMD_PORTID_G(ntohl(pcmd->op_to_portid));
 480         struct net_device *dev = adap->port[adap->chan_map[port]];
 481         int old_dcb_enabled = cxgb4_dcb_enabled(dev);
 482         int new_dcb_enabled;
 483
 484         cxgb4_dcb_handle_fw_update(adap, pcmd);
 485         new_dcb_enabled = cxgb4_dcb_enabled(dev);
 486
 487         /* If the DCB has become enabled or disabled on the port then we're
 488          * going to need to set up/tear down DCB Priority parameters for the
 489          * TX Queues associated with the port.
 490          */
 491         if (new_dcb_enabled != old_dcb_enabled)
 492                 dcb_tx_queue_prio_enable(dev, new_dcb_enabled);
 493 }
 494 #endif /* CONFIG_CHELSIO_T4_DCB */
 495
 496 /* Response queue handler for the FW event queue.
 497  */
 498 static int fwevtq_handler(struct sge_rspq *q, const __be64 *rsp,
 499                           const struct pkt_gl *gl)
 500 {
 501         u8 opcode = ((const struct rss_header *)rsp)->opcode;
 502
 503         rsp++;                                          /* skip RSS header */
 504
 505         /* FW can send EGR_UPDATEs encapsulated in a CPL_FW4_MSG.
 506          */
 507         if (unlikely(opcode == CPL_FW4_MSG &&
 508            ((const struct cpl_fw4_msg *)rsp)->type == FW_TYPE_RSSCPL)) {
 509                 rsp++;
 510                 opcode = ((const struct rss_header *)rsp)->opcode;
 511                 rsp++;
 512                 if (opcode != CPL_SGE_EGR_UPDATE) {
 513                         dev_err(q->adap->pdev_dev, "unexpected FW4/CPL %#x on FW event queue\n"
 514                                 , opcode);
 515                         goto out;
 516                 }
 517         }
 518
 519         if (likely(opcode == CPL_SGE_EGR_UPDATE)) {
 520                 const struct cpl_sge_egr_update *p = (void *)rsp;
 521                 unsigned int qid = EGR_QID_G(ntohl(p->opcode_qid));
 522                 struct sge_txq *txq;
 523
 524                 txq = q->adap->sge.egr_map[qid - q->adap->sge.egr_start];
 525                 txq->restarts++;
 526                 if (txq->q_type == CXGB4_TXQ_ETH) {
 527                         struct sge_eth_txq *eq;
 528
 529                         eq = container_of(txq, struct sge_eth_txq, q);
 530                         netif_tx_wake_queue(eq->txq);
 531                 } else {
 532                         struct sge_uld_txq *oq;
 533
 534                         oq = container_of(txq, struct sge_uld_txq, q);
 535                         tasklet_schedule(&oq->qresume_tsk);
 536                 }
 537         } else if (opcode == CPL_FW6_MSG || opcode == CPL_FW4_MSG) {
 538                 const struct cpl_fw6_msg *p = (void *)rsp;
 539
 540 #ifdef CONFIG_CHELSIO_T4_DCB
 541                 const struct fw_port_cmd *pcmd = (const void *)p->data;
 542                 unsigned int cmd = FW_CMD_OP_G(ntohl(pcmd->op_to_portid));
 543                 unsigned int action =
 544                         FW_PORT_CMD_ACTION_G(ntohl(pcmd->action_to_len16));
 545
 546                 if (cmd == FW_PORT_CMD &&
 547                     (action == FW_PORT_ACTION_GET_PORT_INFO ||
 548                      action == FW_PORT_ACTION_GET_PORT_INFO32)) {
 549                         int port = FW_PORT_CMD_PORTID_G(
 550                                         be32_to_cpu(pcmd->op_to_portid));
 551                         struct net_device *dev;
 552                         int dcbxdis, state_input;
 553
 554                         dev = q->adap->port[q->adap->chan_map[port]];
 555                         dcbxdis = (action == FW_PORT_ACTION_GET_PORT_INFO
 556                           ? !!(pcmd->u.info.dcbxdis_pkd & FW_PORT_CMD_DCBXDIS_F)
 557                           : !!(be32_to_cpu(pcmd->u.info32.lstatus32_to_cbllen32)
 558                                & FW_PORT_CMD_DCBXDIS32_F));
 559                         state_input = (dcbxdis
 560                                        ? CXGB4_DCB_INPUT_FW_DISABLED
 561                                        : CXGB4_DCB_INPUT_FW_ENABLED);
 562
 563                         cxgb4_dcb_state_fsm(dev, state_input);
 564                 }
 565
 566                 if (cmd == FW_PORT_CMD &&
 567                     action == FW_PORT_ACTION_L2_DCB_CFG)
 568                         dcb_rpl(q->adap, pcmd);
 569                 else
 570 #endif
 571                         if (p->type == 0)
 572                                 t4_handle_fw_rpl(q->adap, p->data);
 573         } else if (opcode == CPL_L2T_WRITE_RPL) {
 574                 const struct cpl_l2t_write_rpl *p = (void *)rsp;
 575
 576                 do_l2t_write_rpl(q->adap, p);
 577         } else if (opcode == CPL_SMT_WRITE_RPL) {
 578                 const struct cpl_smt_write_rpl *p = (void *)rsp;
 579
 580                 do_smt_write_rpl(q->adap, p);
 581         } else if (opcode == CPL_SET_TCB_RPL) {
 582                 const struct cpl_set_tcb_rpl *p = (void *)rsp;
 583
 584                 filter_rpl(q->adap, p);
 585         } else if (opcode == CPL_ACT_OPEN_RPL) {
 586                 const struct cpl_act_open_rpl *p = (void *)rsp;
 587
 588                 hash_filter_rpl(q->adap, p);
 589         } else if (opcode == CPL_ABORT_RPL_RSS) {
 590                 const struct cpl_abort_rpl_rss *p = (void *)rsp;
 591
 592                 hash_del_filter_rpl(q->adap, p);
 593         } else if (opcode == CPL_SRQ_TABLE_RPL) {
 594                 const struct cpl_srq_table_rpl *p = (void *)rsp;
 595
 596                 do_srq_table_rpl(q->adap, p);
 597         } else
 598                 dev_err(q->adap->pdev_dev,
 599                         "unexpected CPL %#x on FW event queue\n", opcode);
 600 out:
 601         return 0;
 602 }
 603
 604 static void disable_msi(struct adapter *adapter)
 605 {
 606         if (adapter->flags & USING_MSIX) {
 607                 pci_disable_msix(adapter->pdev);
 608                 adapter->flags &= ~USING_MSIX;
 609         } else if (adapter->flags & USING_MSI) {
 610                 pci_disable_msi(adapter->pdev);
 611                 adapter->flags &= ~USING_MSI;
 612         }
 613 }
 614
 615 /*
 616  * Interrupt handler for non-data events used with MSI-X.
 617  */
 618 static irqreturn_t t4_nondata_intr(int irq, void *cookie)
 619 {
 620         struct adapter *adap = cookie;
 621         u32 v = t4_read_reg(adap, MYPF_REG(PL_PF_INT_CAUSE_A));
 622
 623         if (v & PFSW_F) {
 624                 adap->swintr = 1;
 625                 t4_write_reg(adap, MYPF_REG(PL_PF_INT_CAUSE_A), v);
 626         }
 627         if (adap->flags & MASTER_PF)
 628                 t4_slow_intr_handler(adap);
 629         return IRQ_HANDLED;
 630 }
 631
 632 /*
 633  * Name the MSI-X interrupts.
 634  */
 635 static void name_msix_vecs(struct adapter *adap)
 636 {
 637         int i, j, msi_idx = 2, n = sizeof(adap->msix_info[0].desc);
 638
 639         /* non-data interrupts */
 640         snprintf(adap->msix_info[0].desc, n, "%s", adap->port[0]->name);
 641
 642         /* FW events */
 643         snprintf(adap->msix_info[1].desc, n, "%s-FWeventq",
 644                  adap->port[0]->name);
 645
 646         /* Ethernet queues */
 647         for_each_port(adap, j) {
 648                 struct net_device *d = adap->port[j];
 649                 const struct port_info *pi = netdev_priv(d);
 650
 651                 for (i = 0; i < pi->nqsets; i++, msi_idx++)
 652                         snprintf(adap->msix_info[msi_idx].desc, n, "%s-Rx%d",
 653                                  d->name, i);
 654         }
 655 }
 656
 657 static int request_msix_queue_irqs(struct adapter *adap)
 658 {
 659         struct sge *s = &adap->sge;
 660         int err, ethqidx;
 661         int msi_index = 2;
 662
 663         err = request_irq(adap->msix_info[1].vec, t4_sge_intr_msix, 0,
 664                           adap->msix_info[1].desc, &s->fw_evtq);
 665         if (err)
 666                 return err;
 667
 668         for_each_ethrxq(s, ethqidx) {
 669                 err = request_irq(adap->msix_info[msi_index].vec,
 670                                   t4_sge_intr_msix, 0,
 671                                   adap->msix_info[msi_index].desc,
 672                                   &s->ethrxq[ethqidx].rspq);
 673                 if (err)
 674                         goto unwind;
 675                 msi_index++;
 676         }
 677         return 0;
 678
 679 unwind:
 680         while (--ethqidx >= 0)
 681                 free_irq(adap->msix_info[--msi_index].vec,
 682                          &s->ethrxq[ethqidx].rspq);
 683         free_irq(adap->msix_info[1].vec, &s->fw_evtq);
 684         return err;
 685 }
 686
 687 static void free_msix_queue_irqs(struct adapter *adap)
 688 {
 689         int i, msi_index = 2;
 690         struct sge *s = &adap->sge;
 691
 692         free_irq(adap->msix_info[1].vec, &s->fw_evtq);
 693         for_each_ethrxq(s, i)
 694                 free_irq(adap->msix_info[msi_index++].vec, &s->ethrxq[i].rspq);
 695 }
 696
 697 /**
 698  *      cxgb4_write_rss - write the RSS table for a given port
 699  *      @pi: the port
 700  *      @queues: array of queue indices for RSS
 701  *
 702  *      Sets up the portion of the HW RSS table for the port's VI to distribute
 703  *      packets to the Rx queues in @queues.
 704  *      Should never be called before setting up sge eth rx queues
 705  */
 706 int cxgb4_write_rss(const struct port_info *pi, const u16 *queues)
 707 {
 708         u16 *rss;
 709         int i, err;
 710         struct adapter *adapter = pi->adapter;
 711         const struct sge_eth_rxq *rxq;
 712
 713         rxq = &adapter->sge.ethrxq[pi->first_qset];
 714         rss = kmalloc_array(pi->rss_size, sizeof(u16), GFP_KERNEL);
 715         if (!rss)
 716                 return -ENOMEM;
 717
 718         /* map the queue indices to queue ids */
 719         for (i = 0; i < pi->rss_size; i++, queues++)
 720                 rss[i] = rxq[*queues].rspq.abs_id;
 721
 722         err = t4_config_rss_range(adapter, adapter->pf, pi->viid, 0,
 723                                   pi->rss_size, rss, pi->rss_size);
 724         /* If Tunnel All Lookup isn't specified in the global RSS
 725          * Configuration, then we need to specify a default Ingress
 726          * Queue for any ingress packets which aren't hashed.  We'll
 727          * use our first ingress queue ...
 728          */
 729         if (!err)
 730                 err = t4_config_vi_rss(adapter, adapter->mbox, pi->viid,
 731                                        FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN_F |
 732                                        FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN_F |
 733                                        FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN_F |
 734                                        FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN_F |
 735                                        FW_RSS_VI_CONFIG_CMD_UDPEN_F,
 736                                        rss[0]);
 737         kfree(rss);
 738         return err;
 739 }
 740
 741 /**
 742  *      setup_rss - configure RSS
 743  *      @adap: the adapter
 744  *
 745  *      Sets up RSS for each port.
 746  */
 747 static int setup_rss(struct adapter *adap)
 748 {
 749         int i, j, err;
 750
 751         for_each_port(adap, i) {
 752                 const struct port_info *pi = adap2pinfo(adap, i);
 753
 754                 /* Fill default values with equal distribution */
 755                 for (j = 0; j < pi->rss_size; j++)
 756                         pi->rss[j] = j % pi->nqsets;
 757
 758                 err = cxgb4_write_rss(pi, pi->rss);
 759                 if (err)
 760                         return err;
 761         }
 762         return 0;
 763 }
 764
 765 /*
 766  * Return the channel of the ingress queue with the given qid.
 767  */
 768 static unsigned int rxq_to_chan(const struct sge *p, unsigned int qid)
 769 {
 770         qid -= p->ingr_start;
 771         return netdev2pinfo(p->ingr_map[qid]->netdev)->tx_chan;
 772 }
 773
 774 /*
 775  * Wait until all NAPI handlers are descheduled.
 776  */
 777 static void quiesce_rx(struct adapter *adap)
 778 {
 779         int i;
 780
 781         for (i = 0; i < adap->sge.ingr_sz; i++) {
 782                 struct sge_rspq *q = adap->sge.ingr_map[i];
 783
 784                 if (q && q->handler)
 785                         napi_disable(&q->napi);
 786         }
 787 }
 788
 789 /* Disable interrupt and napi handler */
 790 static void disable_interrupts(struct adapter *adap)
 791 {
 792         if (adap->flags & FULL_INIT_DONE) {
 793                 t4_intr_disable(adap);
 794                 if (adap->flags & USING_MSIX) {
 795                         free_msix_queue_irqs(adap);
 796                         free_irq(adap->msix_info[0].vec, adap);
 797                 } else {
 798                         free_irq(adap->pdev->irq, adap);
 799                 }
 800                 quiesce_rx(adap);
 801         }
 802 }
 803
 804 /*
 805  * Enable NAPI scheduling and interrupt generation for all Rx queues.
 806  */
 807 static void enable_rx(struct adapter *adap)
 808 {
 809         int i;
 810
 811         for (i = 0; i < adap->sge.ingr_sz; i++) {
 812                 struct sge_rspq *q = adap->sge.ingr_map[i];
 813
 814                 if (!q)
 815                         continue;
 816                 if (q->handler)
 817                         napi_enable(&q->napi);
 818
 819                 /* 0-increment GTS to start the timer and enable interrupts */
 820                 t4_write_reg(adap, MYPF_REG(SGE_PF_GTS_A),
 821                              SEINTARM_V(q->intr_params) |
 822                              INGRESSQID_V(q->cntxt_id));
 823         }
 824 }
 825
 826
 827 static int setup_fw_sge_queues(struct adapter *adap)
 828 {
 829         struct sge *s = &adap->sge;
 830         int err = 0;
 831
 832         bitmap_zero(s->starving_fl, s->egr_sz);
 833         bitmap_zero(s->txq_maperr, s->egr_sz);
 834
 835         if (adap->flags & USING_MSIX)
 836                 adap->msi_idx = 1;         /* vector 0 is for non-queue interrupts */
 837         else {
 838                 err = t4_sge_alloc_rxq(adap, &s->intrq, false, adap->port[0], 0,
 839                                        NULL, NULL, NULL, -1);
 840                 if (err)
 841                         return err;
 842                 adap->msi_idx = -((int)s->intrq.abs_id + 1);
 843         }
 844
 845         err = t4_sge_alloc_rxq(adap, &s->fw_evtq, true, adap->port[0],
 846                                adap->msi_idx, NULL, fwevtq_handler, NULL, -1);
 847         return err;
 848 }
 849
 850 /**
 851  *      setup_sge_queues - configure SGE Tx/Rx/response queues
 852  *      @adap: the adapter
 853  *
 854  *      Determines how many sets of SGE queues to use and initializes them.
 855  *      We support multiple queue sets per port if we have MSI-X, otherwise
 856  *      just one queue set per port.
 857  */
 858 static int setup_sge_queues(struct adapter *adap)
 859 {
 860         int err, i, j;
 861         struct sge *s = &adap->sge;
 862         struct sge_uld_rxq_info *rxq_info = NULL;
 863         unsigned int cmplqid = 0;
 864
 865         if (is_uld(adap))
 866                 rxq_info = s->uld_rxq_info[CXGB4_ULD_RDMA];
 867
 868         for_each_port(adap, i) {
 869                 struct net_device *dev = adap->port[i];
 870                 struct port_info *pi = netdev_priv(dev);
 871                 struct sge_eth_rxq *q = &s->ethrxq[pi->first_qset];
 872                 struct sge_eth_txq *t = &s->ethtxq[pi->first_qset];
 873
 874                 for (j = 0; j < pi->nqsets; j++, q++) {
 875                         if (adap->msi_idx > 0)
 876                                 adap->msi_idx++;
 877                         err = t4_sge_alloc_rxq(adap, &q->rspq, false, dev,
 878                                                adap->msi_idx, &q->fl,
 879                                                t4_ethrx_handler,
 880                                                NULL,
 881                                                t4_get_tp_ch_map(adap,
 882                                                                 pi->tx_chan));
 883                         if (err)
 884                                 goto freeout;
 885                         q->rspq.idx = j;
 886                         memset(&q->stats, 0, sizeof(q->stats));
 887                 }
 888                 for (j = 0; j < pi->nqsets; j++, t++) {
 889                         err = t4_sge_alloc_eth_txq(adap, t, dev,
 890                                         netdev_get_tx_queue(dev, j),
 891                                         s->fw_evtq.cntxt_id);
 892                         if (err)
 893                                 goto freeout;
 894                 }
 895         }
 896
 897         for_each_port(adap, i) {
 898                 /* Note that cmplqid below is 0 if we don't
 899                  * have RDMA queues, and that's the right value.
 900                  */
 901                 if (rxq_info)
 902                         cmplqid = rxq_info->uldrxq[i].rspq.cntxt_id;
 903
 904                 err = t4_sge_alloc_ctrl_txq(adap, &s->ctrlq[i], adap->port[i],
 905                                             s->fw_evtq.cntxt_id, cmplqid);
 906                 if (err)
 907                         goto freeout;
 908         }
 909
 910         if (!is_t4(adap->params.chip)) {
 911                 err = t4_sge_alloc_eth_txq(adap, &s->ptptxq, adap->port[0],
 912                                            netdev_get_tx_queue(adap->port[0], 0)
 913                                            , s->fw_evtq.cntxt_id);
 914                 if (err)
 915                         goto freeout;
 916         }
 917
 918         t4_write_reg(adap, is_t4(adap->params.chip) ?
 919                                 MPS_TRC_RSS_CONTROL_A :
 920                                 MPS_T5_TRC_RSS_CONTROL_A,
 921                      RSSCONTROL_V(netdev2pinfo(adap->port[0])->tx_chan) |
 922                      QUEUENUMBER_V(s->ethrxq[0].rspq.abs_id));
 923         return 0;
 924 freeout:
 925         dev_err(adap->pdev_dev, "Can't allocate queues, err=%d\n", -err);
 926         t4_free_sge_resources(adap);
 927         return err;
 928 }
 929
 930 static u16 cxgb_select_queue(struct net_device *dev, struct sk_buff *skb,
 931                              struct net_device *sb_dev,
 932                              select_queue_fallback_t fallback)
 933 {
 934         int txq;
 935
 936 #ifdef CONFIG_CHELSIO_T4_DCB
 937         /* If a Data Center Bridging has been successfully negotiated on this
 938          * link then we'll use the skb's priority to map it to a TX Queue.
 939          * The skb's priority is determined via the VLAN Tag Priority Code
 940          * Point field.
 941          */
 942         if (cxgb4_dcb_enabled(dev) && !is_kdump_kernel()) {
 943                 u16 vlan_tci;
 944                 int err;
 945
 946                 err = vlan_get_tag(skb, &vlan_tci);
 947                 if (unlikely(err)) {
 948                         if (net_ratelimit())
 949                                 netdev_warn(dev,
 950                                             "TX Packet without VLAN Tag on DCB Link\n");
 951                         txq = 0;
 952                 } else {
 953                         txq = (vlan_tci & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
 954 #ifdef CONFIG_CHELSIO_T4_FCOE
 955                         if (skb->protocol == htons(ETH_P_FCOE))
 956                                 txq = skb->priority & 0x7;
 957 #endif /* CONFIG_CHELSIO_T4_FCOE */
 958                 }
 959                 return txq;
 960         }
 961 #endif /* CONFIG_CHELSIO_T4_DCB */
 962
 963         if (select_queue) {
 964                 txq = (skb_rx_queue_recorded(skb)
 965                         ? skb_get_rx_queue(skb)
 966                         : smp_processor_id());
 967
 968                 while (unlikely(txq >= dev->real_num_tx_queues))
 969                         txq -= dev->real_num_tx_queues;
 970
 971                 return txq;
 972         }
 973
 974         return fallback(dev, skb, NULL) % dev->real_num_tx_queues;
 975 }
 976
 977 static int closest_timer(const struct sge *s, int time)
 978 {
 979         int i, delta, match = 0, min_delta = INT_MAX;
 980
 981         for (i = 0; i < ARRAY_SIZE(s->timer_val); i++) {
 982                 delta = time - s->timer_val[i];
 983                 if (delta < 0)
 984                         delta = -delta;
 985                 if (delta < min_delta) {
 986                         min_delta = delta;
 987                         match = i;
 988                 }
 989         }
 990         return match;
 991 }
 992
 993 static int closest_thres(const struct sge *s, int thres)
 994 {
 995         int i, delta, match = 0, min_delta = INT_MAX;
 996
 997         for (i = 0; i < ARRAY_SIZE(s->counter_val); i++) {
 998                 delta = thres - s->counter_val[i];
 999                 if (delta < 0)
1000                         delta = -delta;
1001                 if (delta < min_delta) {
1002                         min_delta = delta;
1003                         match = i;
1004                 }
1005         }
1006         return match;
1007 }
1008
1009 /**
1010  *      cxgb4_set_rspq_intr_params - set a queue's interrupt holdoff parameters
1011  *      @q: the Rx queue
1012  *      @us: the hold-off time in us, or 0 to disable timer
1013  *      @cnt: the hold-off packet count, or 0 to disable counter
1014  *
1015  *      Sets an Rx queue's interrupt hold-off time and packet count.  At least
1016  *      one of the two needs to be enabled for the queue to generate interrupts.
1017  */
1018 int cxgb4_set_rspq_intr_params(struct sge_rspq *q,
1019                                unsigned int us, unsigned int cnt)
1020 {
1021         struct adapter *adap = q->adap;
1022
1023         if ((us | cnt) == 0)
1024                 cnt = 1;
1025
1026         if (cnt) {
1027                 int err;
1028                 u32 v, new_idx;
1029
1030                 new_idx = closest_thres(&adap->sge, cnt);
1031                 if (q->desc && q->pktcnt_idx != new_idx) {
1032                         /* the queue has already been created, update it */
1033                         v = FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DMAQ) |
1034                             FW_PARAMS_PARAM_X_V(
1035                                         FW_PARAMS_PARAM_DMAQ_IQ_INTCNTTHRESH) |
1036                             FW_PARAMS_PARAM_YZ_V(q->cntxt_id);
1037                         err = t4_set_params(adap, adap->mbox, adap->pf, 0, 1,
1038                                             &v, &new_idx);
1039                         if (err)
1040                                 return err;
1041                 }
1042                 q->pktcnt_idx = new_idx;
1043         }
1044
1045         us = us == 0 ? 6 : closest_timer(&adap->sge, us);
1046         q->intr_params = QINTR_TIMER_IDX_V(us) | QINTR_CNT_EN_V(cnt > 0);
1047         return 0;
1048 }
1049
1050 static int cxgb_set_features(struct net_device *dev, netdev_features_t features)
1051 {
1052         const struct port_info *pi = netdev_priv(dev);
1053         netdev_features_t changed = dev->features ^ features;
1054         int err;
1055
1056         if (!(changed & NETIF_F_HW_VLAN_CTAG_RX))
1057                 return 0;
1058
1059         err = t4_set_rxmode(pi->adapter, pi->adapter->pf, pi->viid, -1,
1060                             -1, -1, -1,
1061                             !!(features & NETIF_F_HW_VLAN_CTAG_RX), true);
1062         if (unlikely(err))
1063                 dev->features = features ^ NETIF_F_HW_VLAN_CTAG_RX;
1064         return err;
1065 }
1066
1067 static int setup_debugfs(struct adapter *adap)
1068 {
1069         if (IS_ERR_OR_NULL(adap->debugfs_root))
1070                 return -1;
1071
1072 #ifdef CONFIG_DEBUG_FS
1073         t4_setup_debugfs(adap);
1074 #endif
1075         return 0;
1076 }
1077
1078 /*
1079  * upper-layer driver support
1080  */
1081
1082 /*
1083  * Allocate an active-open TID and set it to the supplied value.
1084  */
1085 int cxgb4_alloc_atid(struct tid_info *t, void *data)
1086 {
1087         int atid = -1;
1088
1089         spin_lock_bh(&t->atid_lock);
1090         if (t->afree) {
1091                 union aopen_entry *p = t->afree;
1092
1093                 atid = (p - t->atid_tab) + t->atid_base;
1094                 t->afree = p->next;
1095                 p->data = data;
1096                 t->atids_in_use++;
1097         }
1098         spin_unlock_bh(&t->atid_lock);
1099         return atid;
1100 }
1101 EXPORT_SYMBOL(cxgb4_alloc_atid);
1102
1103 /*
1104  * Release an active-open TID.
1105  */
1106 void cxgb4_free_atid(struct tid_info *t, unsigned int atid)
1107 {
1108         union aopen_entry *p = &t->atid_tab[atid - t->atid_base];
1109
1110         spin_lock_bh(&t->atid_lock);
1111         p->next = t->afree;
1112         t->afree = p;
1113         t->atids_in_use--;
1114         spin_unlock_bh(&t->atid_lock);
1115 }
1116 EXPORT_SYMBOL(cxgb4_free_atid);
1117
1118 /*
1119  * Allocate a server TID and set it to the supplied value.
1120  */
1121 int cxgb4_alloc_stid(struct tid_info *t, int family, void *data)
1122 {
1123         int stid;
1124
1125         spin_lock_bh(&t->stid_lock);
1126         if (family == PF_INET) {
1127                 stid = find_first_zero_bit(t->stid_bmap, t->nstids);
1128                 if (stid < t->nstids)
1129                         __set_bit(stid, t->stid_bmap);
1130                 else
1131                         stid = -1;
1132         } else {
1133                 stid = bitmap_find_free_region(t->stid_bmap, t->nstids, 1);
1134                 if (stid < 0)
1135                         stid = -1;
1136         }
1137         if (stid >= 0) {
1138                 t->stid_tab[stid].data = data;
1139                 stid += t->stid_base;
1140                 /* IPv6 requires max of 520 bits or 16 cells in TCAM
1141                  * This is equivalent to 4 TIDs. With CLIP enabled it
1142                  * needs 2 TIDs.
1143                  */
1144                 if (family == PF_INET6) {
1145                         t->stids_in_use += 2;
1146                         t->v6_stids_in_use += 2;
1147                 } else {
1148                         t->stids_in_use++;
1149                 }
1150         }
1151         spin_unlock_bh(&t->stid_lock);
1152         return stid;
1153 }
1154 EXPORT_SYMBOL(cxgb4_alloc_stid);
1155
1156 /* Allocate a server filter TID and set it to the supplied value.
1157  */
1158 int cxgb4_alloc_sftid(struct tid_info *t, int family, void *data)
1159 {
1160         int stid;
1161
1162         spin_lock_bh(&t->stid_lock);
1163         if (family == PF_INET) {
1164                 stid = find_next_zero_bit(t->stid_bmap,
1165                                 t->nstids + t->nsftids, t->nstids);
1166                 if (stid < (t->nstids + t->nsftids))
1167                         __set_bit(stid, t->stid_bmap);
1168                 else
1169                         stid = -1;
1170         } else {
1171                 stid = -1;
1172         }
1173         if (stid >= 0) {
1174                 t->stid_tab[stid].data = data;
1175                 stid -= t->nstids;
1176                 stid += t->sftid_base;
1177                 t->sftids_in_use++;
1178         }
1179         spin_unlock_bh(&t->stid_lock);
1180         return stid;
1181 }
1182 EXPORT_SYMBOL(cxgb4_alloc_sftid);
1183
1184 /* Release a server TID.
1185  */
1186 void cxgb4_free_stid(struct tid_info *t, unsigned int stid, int family)
1187 {
1188         /* Is it a server filter TID? */
1189         if (t->nsftids && (stid >= t->sftid_base)) {
1190                 stid -= t->sftid_base;
1191                 stid += t->nstids;
1192         } else {
1193                 stid -= t->stid_base;
1194         }
1195
1196         spin_lock_bh(&t->stid_lock);
1197         if (family == PF_INET)
1198                 __clear_bit(stid, t->stid_bmap);
1199         else
1200                 bitmap_release_region(t->stid_bmap, stid, 1);
1201         t->stid_tab[stid].data = NULL;
1202         if (stid < t->nstids) {
1203                 if (family == PF_INET6) {
1204                         t->stids_in_use -= 2;
1205                         t->v6_stids_in_use -= 2;
1206                 } else {
1207                         t->stids_in_use--;
1208                 }
1209         } else {
1210                 t->sftids_in_use--;
1211         }
1212
1213         spin_unlock_bh(&t->stid_lock);
1214 }
1215 EXPORT_SYMBOL(cxgb4_free_stid);
1216
1217 /*
1218  * Populate a TID_RELEASE WR.  Caller must properly size the skb.
1219  */
1220 static void mk_tid_release(struct sk_buff *skb, unsigned int chan,
1221                            unsigned int tid)
1222 {
1223         struct cpl_tid_release *req;
1224
1225         set_wr_txq(skb, CPL_PRIORITY_SETUP, chan);
1226         req = __skb_put(skb, sizeof(*req));
1227         INIT_TP_WR(req, tid);
1228         OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_TID_RELEASE, tid));
1229 }
1230
1231 /*
1232  * Queue a TID release request and if necessary schedule a work queue to
1233  * process it.
1234  */
1235 static void cxgb4_queue_tid_release(struct tid_info *t, unsigned int chan,
1236                                     unsigned int tid)
1237 {
1238         void **p = &t->tid_tab[tid];
1239         struct adapter *adap = container_of(t, struct adapter, tids);
1240
1241         spin_lock_bh(&adap->tid_release_lock);
1242         *p = adap->tid_release_head;
1243         /* Low 2 bits encode the Tx channel number */
1244         adap->tid_release_head = (void **)((uintptr_t)p | chan);
1245         if (!adap->tid_release_task_busy) {
1246                 adap->tid_release_task_busy = true;
1247                 queue_work(adap->workq, &adap->tid_release_task);
1248         }
1249         spin_unlock_bh(&adap->tid_release_lock);
1250 }
1251
1252 /*
1253  * Process the list of pending TID release requests.
1254  */
1255 static void process_tid_release_list(struct work_struct *work)
1256 {
1257         struct sk_buff *skb;
1258         struct adapter *adap;
1259
1260         adap = container_of(work, struct adapter, tid_release_task);
1261
1262         spin_lock_bh(&adap->tid_release_lock);
1263         while (adap->tid_release_head) {
1264                 void **p = adap->tid_release_head;
1265                 unsigned int chan = (uintptr_t)p & 3;
1266                 p = (void *)p - chan;
1267
1268                 adap->tid_release_head = *p;
1269                 *p = NULL;
1270                 spin_unlock_bh(&adap->tid_release_lock);
1271
1272                 while (!(skb = alloc_skb(sizeof(struct cpl_tid_release),
1273                                          GFP_KERNEL)))
1274                         schedule_timeout_uninterruptible(1);
1275
1276                 mk_tid_release(skb, chan, p - adap->tids.tid_tab);
1277                 t4_ofld_send(adap, skb);
1278                 spin_lock_bh(&adap->tid_release_lock);
1279         }
1280         adap->tid_release_task_busy = false;
1281         spin_unlock_bh(&adap->tid_release_lock);
1282 }
1283
1284 /*
1285  * Release a TID and inform HW.  If we are unable to allocate the release
1286  * message we defer to a work queue.
1287  */
1288 void cxgb4_remove_tid(struct tid_info *t, unsigned int chan, unsigned int tid,
1289                       unsigned short family)
1290 {
1291         struct sk_buff *skb;
1292         struct adapter *adap = container_of(t, struct adapter, tids);
1293
1294         WARN_ON(tid >= t->ntids);
1295
1296         if (t->tid_tab[tid]) {
1297                 t->tid_tab[tid] = NULL;
1298                 atomic_dec(&t->conns_in_use);
1299                 if (t->hash_base && (tid >= t->hash_base)) {
1300                         if (family == AF_INET6)
1301                                 atomic_sub(2, &t->hash_tids_in_use);
1302                         else
1303                                 atomic_dec(&t->hash_tids_in_use);
1304                 } else {
1305                         if (family == AF_INET6)
1306                                 atomic_sub(2, &t->tids_in_use);
1307                         else
1308                                 atomic_dec(&t->tids_in_use);
1309                 }
1310         }
1311
1312         skb = alloc_skb(sizeof(struct cpl_tid_release), GFP_ATOMIC);
1313         if (likely(skb)) {
1314                 mk_tid_release(skb, chan, tid);
1315                 t4_ofld_send(adap, skb);
1316         } else
1317                 cxgb4_queue_tid_release(t, chan, tid);
1318 }
1319 EXPORT_SYMBOL(cxgb4_remove_tid);
1320
1321 /*
1322  * Allocate and initialize the TID tables.  Returns 0 on success.
1323  */
1324 static int tid_init(struct tid_info *t)
1325 {
1326         struct adapter *adap = container_of(t, struct adapter, tids);
1327         unsigned int max_ftids = t->nftids + t->nsftids;
1328         unsigned int natids = t->natids;
1329         unsigned int stid_bmap_size;
1330         unsigned int ftid_bmap_size;
1331         size_t size;
1332
1333         stid_bmap_size = BITS_TO_LONGS(t->nstids + t->nsftids);
1334         ftid_bmap_size = BITS_TO_LONGS(t->nftids);
1335         size = t->ntids * sizeof(*t->tid_tab) +
1336                natids * sizeof(*t->atid_tab) +
1337                t->nstids * sizeof(*t->stid_tab) +
1338                t->nsftids * sizeof(*t->stid_tab) +
1339                stid_bmap_size * sizeof(long) +
1340                max_ftids * sizeof(*t->ftid_tab) +
1341                ftid_bmap_size * sizeof(long);
1342
1343         t->tid_tab = kvzalloc(size, GFP_KERNEL);
1344         if (!t->tid_tab)
1345                 return -ENOMEM;
1346
1347         t->atid_tab = (union aopen_entry *)&t->tid_tab[t->ntids];
1348         t->stid_tab = (struct serv_entry *)&t->atid_tab[natids];
1349         t->stid_bmap = (unsigned long *)&t->stid_tab[t->nstids + t->nsftids];
1350         t->ftid_tab = (struct filter_entry *)&t->stid_bmap[stid_bmap_size];
1351         t->ftid_bmap = (unsigned long *)&t->ftid_tab[max_ftids];
1352         spin_lock_init(&t->stid_lock);
1353         spin_lock_init(&t->atid_lock);
1354         spin_lock_init(&t->ftid_lock);
1355
1356         t->stids_in_use = 0;
1357         t->v6_stids_in_use = 0;
1358         t->sftids_in_use = 0;
1359         t->afree = NULL;
1360         t->atids_in_use = 0;
1361         atomic_set(&t->tids_in_use, 0);
1362         atomic_set(&t->conns_in_use, 0);
1363         atomic_set(&t->hash_tids_in_use, 0);
1364
1365         /* Setup the free list for atid_tab and clear the stid bitmap. */
1366         if (natids) {
1367                 while (--natids)
1368                         t->atid_tab[natids - 1].next = &t->atid_tab[natids];
1369                 t->afree = t->atid_tab;
1370         }
1371
1372         if (is_offload(adap)) {
1373                 bitmap_zero(t->stid_bmap, t->nstids + t->nsftids);
1374                 /* Reserve stid 0 for T4/T5 adapters */
1375                 if (!t->stid_base &&
1376                     CHELSIO_CHIP_VERSION(adap->params.chip) <= CHELSIO_T5)
1377                         __set_bit(0, t->stid_bmap);
1378         }
1379
1380         bitmap_zero(t->ftid_bmap, t->nftids);
1381         return 0;
1382 }
1383
1384 /**
1385  *      cxgb4_create_server - create an IP server
1386  *      @dev: the device
1387  *      @stid: the server TID
1388  *      @sip: local IP address to bind server to
1389  *      @sport: the server's TCP port
1390  *      @queue: queue to direct messages from this server to
1391  *
1392  *      Create an IP server for the given port and address.
1393  *      Returns <0 on error and one of the %NET_XMIT_* values on success.
1394  */
1395 int cxgb4_create_server(const struct net_device *dev, unsigned int stid,
1396                         __be32 sip, __be16 sport, __be16 vlan,
1397                         unsigned int queue)
1398 {
1399         unsigned int chan;
1400         struct sk_buff *skb;
1401         struct adapter *adap;
1402         struct cpl_pass_open_req *req;
1403         int ret;
1404
1405         skb = alloc_skb(sizeof(*req), GFP_KERNEL);
1406         if (!skb)
1407                 return -ENOMEM;
1408
1409         adap = netdev2adap(dev);
1410         req = __skb_put(skb, sizeof(*req));
1411         INIT_TP_WR(req, 0);
1412         OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ, stid));
1413         req->local_port = sport;
1414         req->peer_port = htons(0);
1415         req->local_ip = sip;
1416         req->peer_ip = htonl(0);
1417         chan = rxq_to_chan(&adap->sge, queue);
1418         req->opt0 = cpu_to_be64(TX_CHAN_V(chan));
1419         req->opt1 = cpu_to_be64(CONN_POLICY_V(CPL_CONN_POLICY_ASK) |
1420                                 SYN_RSS_ENABLE_F | SYN_RSS_QUEUE_V(queue));
1421         ret = t4_mgmt_tx(adap, skb);
1422         return net_xmit_eval(ret);
1423 }
1424 EXPORT_SYMBOL(cxgb4_create_server);
1425
1426 /*      cxgb4_create_server6 - create an IPv6 server
1427  *      @dev: the device
1428  *      @stid: the server TID
1429  *      @sip: local IPv6 address to bind server to
1430  *      @sport: the server's TCP port
1431  *      @queue: queue to direct messages from this server to
1432  *
1433  *      Create an IPv6 server for the given port and address.
1434  *      Returns <0 on error and one of the %NET_XMIT_* values on success.
1435  */
1436 int cxgb4_create_server6(const struct net_device *dev, unsigned int stid,
1437                          const struct in6_addr *sip, __be16 sport,
1438                          unsigned int queue)
1439 {
1440         unsigned int chan;
1441         struct sk_buff *skb;
1442         struct adapter *adap;
1443         struct cpl_pass_open_req6 *req;
1444         int ret;
1445
1446         skb = alloc_skb(sizeof(*req), GFP_KERNEL);
1447         if (!skb)
1448                 return -ENOMEM;
1449
1450         adap = netdev2adap(dev);
1451         req = __skb_put(skb, sizeof(*req));
1452         INIT_TP_WR(req, 0);
1453         OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ6, stid));
1454         req->local_port = sport;
1455         req->peer_port = htons(0);
1456         req->local_ip_hi = *(__be64 *)(sip->s6_addr);
1457         req->local_ip_lo = *(__be64 *)(sip->s6_addr + 8);
1458         req->peer_ip_hi = cpu_to_be64(0);
1459         req->peer_ip_lo = cpu_to_be64(0);
1460         chan = rxq_to_chan(&adap->sge, queue);
1461         req->opt0 = cpu_to_be64(TX_CHAN_V(chan));
1462         req->opt1 = cpu_to_be64(CONN_POLICY_V(CPL_CONN_POLICY_ASK) |
1463                                 SYN_RSS_ENABLE_F | SYN_RSS_QUEUE_V(queue));
1464         ret = t4_mgmt_tx(adap, skb);
1465         return net_xmit_eval(ret);
1466 }
1467 EXPORT_SYMBOL(cxgb4_create_server6);
1468
1469 int cxgb4_remove_server(const struct net_device *dev, unsigned int stid,
1470                         unsigned int queue, bool ipv6)
1471 {
1472         struct sk_buff *skb;
1473         struct adapter *adap;
1474         struct cpl_close_listsvr_req *req;
1475         int ret;
1476
1477         adap = netdev2adap(dev);
1478
1479         skb = alloc_skb(sizeof(*req), GFP_KERNEL);
1480         if (!skb)
1481                 return -ENOMEM;
1482
1483         req = __skb_put(skb, sizeof(*req));
1484         INIT_TP_WR(req, 0);
1485         OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_LISTSRV_REQ, stid));
1486         req->reply_ctrl = htons(NO_REPLY_V(0) | (ipv6 ? LISTSVR_IPV6_V(1) :
1487                                 LISTSVR_IPV6_V(0)) | QUEUENO_V(queue));
1488         ret = t4_mgmt_tx(adap, skb);
1489         return net_xmit_eval(ret);
1490 }
1491 EXPORT_SYMBOL(cxgb4_remove_server);
1492
1493 /**
1494  *      cxgb4_best_mtu - find the entry in the MTU table closest to an MTU
1495  *      @mtus: the HW MTU table
1496  *      @mtu: the target MTU
1497  *      @idx: index of selected entry in the MTU table
1498  *
1499  *      Returns the index and the value in the HW MTU table that is closest to
1500  *      but does not exceed @mtu, unless @mtu is smaller than any value in the
1501  *      table, in which case that smallest available value is selected.
1502  */
1503 unsigned int cxgb4_best_mtu(const unsigned short *mtus, unsigned short mtu,
1504                             unsigned int *idx)
1505 {
1506         unsigned int i = 0;
1507
1508         while (i < NMTUS - 1 && mtus[i + 1] <= mtu)
1509                 ++i;
1510         if (idx)
1511                 *idx = i;
1512         return mtus[i];
1513 }
1514 EXPORT_SYMBOL(cxgb4_best_mtu);
1515
1516 /**
1517  *     cxgb4_best_aligned_mtu - find best MTU, [hopefully] data size aligned
1518  *     @mtus: the HW MTU table
1519  *     @header_size: Header Size
1520  *     @data_size_max: maximum Data Segment Size
1521  *     @data_size_align: desired Data Segment Size Alignment (2^N)
1522  *     @mtu_idxp: HW MTU Table Index return value pointer (possibly NULL)
1523  *
1524  *     Similar to cxgb4_best_mtu() but instead of searching the Hardware
1525  *     MTU Table based solely on a Maximum MTU parameter, we break that
1526  *     parameter up into a Header Size and Maximum Data Segment Size, and
1527  *     provide a desired Data Segment Size Alignment.  If we find an MTU in
1528  *     the Hardware MTU Table which will result in a Data Segment Size with
1529  *     the requested alignment _and_ that MTU isn't "too far" from the
1530  *     closest MTU, then we'll return that rather than the closest MTU.
1531  */
1532 unsigned int cxgb4_best_aligned_mtu(const unsigned short *mtus,
1533                                     unsigned short header_size,
1534                                     unsigned short data_size_max,
1535                                     unsigned short data_size_align,
1536                                     unsigned int *mtu_idxp)
1537 {
1538         unsigned short max_mtu = header_size + data_size_max;
1539         unsigned short data_size_align_mask = data_size_align - 1;
1540         int mtu_idx, aligned_mtu_idx;
1541
1542         /* Scan the MTU Table till we find an MTU which is larger than our
1543          * Maximum MTU or we reach the end of the table.  Along the way,
1544          * record the last MTU found, if any, which will result in a Data
1545          * Segment Length matching the requested alignment.
1546          */
1547         for (mtu_idx = 0, aligned_mtu_idx = -1; mtu_idx < NMTUS; mtu_idx++) {
1548                 unsigned short data_size = mtus[mtu_idx] - header_size;
1549
1550                 /* If this MTU minus the Header Size would result in a
1551                  * Data Segment Size of the desired alignment, remember it.
1552                  */
1553                 if ((data_size & data_size_align_mask) == 0)
1554                         aligned_mtu_idx = mtu_idx;
1555
1556                 /* If we're not at the end of the Hardware MTU Table and the
1557                  * next element is larger than our Maximum MTU, drop out of
1558                  * the loop.
1559                  */
1560                 if (mtu_idx+1 < NMTUS && mtus[mtu_idx+1] > max_mtu)
1561                         break;
1562         }
1563
1564         /* If we fell out of the loop because we ran to the end of the table,
1565          * then we just have to use the last [largest] entry.
1566          */
1567         if (mtu_idx == NMTUS)
1568                 mtu_idx--;
1569
1570         /* If we found an MTU which resulted in the requested Data Segment
1571          * Length alignment and that's "not far" from the largest MTU which is
1572          * less than or equal to the maximum MTU, then use that.
1573          */
1574         if (aligned_mtu_idx >= 0 &&
1575             mtu_idx - aligned_mtu_idx <= 1)
1576                 mtu_idx = aligned_mtu_idx;
1577
1578         /* If the caller has passed in an MTU Index pointer, pass the
1579          * MTU Index back.  Return the MTU value.
1580          */
1581         if (mtu_idxp)
1582                 *mtu_idxp = mtu_idx;
1583         return mtus[mtu_idx];
1584 }
1585 EXPORT_SYMBOL(cxgb4_best_aligned_mtu);
1586
1587 /**
1588  *      cxgb4_tp_smt_idx - Get the Source Mac Table index for this VI
1589  *      @chip: chip type
1590  *      @viid: VI id of the given port
1591  *
1592  *      Return the SMT index for this VI.
1593  */
1594 unsigned int cxgb4_tp_smt_idx(enum chip_type chip, unsigned int viid)
1595 {
1596         /* In T4/T5, SMT contains 256 SMAC entries organized in
1597          * 128 rows of 2 entries each.
1598          * In T6, SMT contains 256 SMAC entries in 256 rows.
1599          * TODO: The below code needs to be updated when we add support
1600          * for 256 VFs.
1601          */
1602         if (CHELSIO_CHIP_VERSION(chip) <= CHELSIO_T5)
1603                 return ((viid & 0x7f) << 1);
1604         else
1605                 return (viid & 0x7f);
1606 }
1607 EXPORT_SYMBOL(cxgb4_tp_smt_idx);
1608
1609 /**
1610  *      cxgb4_port_chan - get the HW channel of a port
1611  *      @dev: the net device for the port
1612  *
1613  *      Return the HW Tx channel of the given port.
1614  */
1615 unsigned int cxgb4_port_chan(const struct net_device *dev)
1616 {
1617         return netdev2pinfo(dev)->tx_chan;
1618 }
1619 EXPORT_SYMBOL(cxgb4_port_chan);
1620
1621 unsigned int cxgb4_dbfifo_count(const struct net_device *dev, int lpfifo)
1622 {
1623         struct adapter *adap = netdev2adap(dev);
1624         u32 v1, v2, lp_count, hp_count;
1625
1626         v1 = t4_read_reg(adap, SGE_DBFIFO_STATUS_A);
1627         v2 = t4_read_reg(adap, SGE_DBFIFO_STATUS2_A);
1628         if (is_t4(adap->params.chip)) {
1629                 lp_count = LP_COUNT_G(v1);
1630                 hp_count = HP_COUNT_G(v1);
1631         } else {
1632                 lp_count = LP_COUNT_T5_G(v1);
1633                 hp_count = HP_COUNT_T5_G(v2);
1634         }
1635         return lpfifo ? lp_count : hp_count;
1636 }
1637 EXPORT_SYMBOL(cxgb4_dbfifo_count);
1638
1639 /**
1640  *      cxgb4_port_viid - get the VI id of a port
1641  *      @dev: the net device for the port
1642  *
1643  *      Return the VI id of the given port.
1644  */
1645 unsigned int cxgb4_port_viid(const struct net_device *dev)
1646 {
1647         return netdev2pinfo(dev)->viid;
1648 }
1649 EXPORT_SYMBOL(cxgb4_port_viid);
1650
1651 /**
1652  *      cxgb4_port_idx - get the index of a port
1653  *      @dev: the net device for the port
1654  *
1655  *      Return the index of the given port.
1656  */
1657 unsigned int cxgb4_port_idx(const struct net_device *dev)
1658 {
1659         return netdev2pinfo(dev)->port_id;
1660 }
1661 EXPORT_SYMBOL(cxgb4_port_idx);
1662
1663 void cxgb4_get_tcp_stats(struct pci_dev *pdev, struct tp_tcp_stats *v4,
1664                          struct tp_tcp_stats *v6)
1665 {
1666         struct adapter *adap = pci_get_drvdata(pdev);
1667
1668         spin_lock(&adap->stats_lock);
1669         t4_tp_get_tcp_stats(adap, v4, v6, false);
1670         spin_unlock(&adap->stats_lock);
1671 }
1672 EXPORT_SYMBOL(cxgb4_get_tcp_stats);
1673
1674 void cxgb4_iscsi_init(struct net_device *dev, unsigned int tag_mask,
1675                       const unsigned int *pgsz_order)
1676 {
1677         struct adapter *adap = netdev2adap(dev);
1678
1679         t4_write_reg(adap, ULP_RX_ISCSI_TAGMASK_A, tag_mask);
1680         t4_write_reg(adap, ULP_RX_ISCSI_PSZ_A, HPZ0_V(pgsz_order[0]) |
1681                      HPZ1_V(pgsz_order[1]) | HPZ2_V(pgsz_order[2]) |
1682                      HPZ3_V(pgsz_order[3]));
1683 }
1684 EXPORT_SYMBOL(cxgb4_iscsi_init);
1685
1686 int cxgb4_flush_eq_cache(struct net_device *dev)
1687 {
1688         struct adapter *adap = netdev2adap(dev);
1689
1690         return t4_sge_ctxt_flush(adap, adap->mbox, CTXT_EGRESS);
1691 }
1692 EXPORT_SYMBOL(cxgb4_flush_eq_cache);
1693
1694 static int read_eq_indices(struct adapter *adap, u16 qid, u16 *pidx, u16 *cidx)
1695 {
1696         u32 addr = t4_read_reg(adap, SGE_DBQ_CTXT_BADDR_A) + 24 * qid + 8;
1697         __be64 indices;
1698         int ret;
1699
1700         spin_lock(&adap->win0_lock);
1701         ret = t4_memory_rw(adap, 0, MEM_EDC0, addr,
1702                            sizeof(indices), (__be32 *)&indices,
1703                            T4_MEMORY_READ);
1704         spin_unlock(&adap->win0_lock);
1705         if (!ret) {
1706                 *cidx = (be64_to_cpu(indices) >> 25) & 0xffff;
1707                 *pidx = (be64_to_cpu(indices) >> 9) & 0xffff;
1708         }
1709         return ret;
1710 }
1711
1712 int cxgb4_sync_txq_pidx(struct net_device *dev, u16 qid, u16 pidx,
1713                         u16 size)
1714 {
1715         struct adapter *adap = netdev2adap(dev);
1716         u16 hw_pidx, hw_cidx;
1717         int ret;
1718
1719         ret = read_eq_indices(adap, qid, &hw_pidx, &hw_cidx);
1720         if (ret)
1721                 goto out;
1722
1723         if (pidx != hw_pidx) {
1724                 u16 delta;
1725                 u32 val;
1726
1727                 if (pidx >= hw_pidx)
1728                         delta = pidx - hw_pidx;
1729                 else
1730                         delta = size - hw_pidx + pidx;
1731
1732                 if (is_t4(adap->params.chip))
1733                         val = PIDX_V(delta);
1734                 else
1735                         val = PIDX_T5_V(delta);
1736                 wmb();
1737                 t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL_A),
1738                              QID_V(qid) | val);
1739         }
1740 out:
1741         return ret;
1742 }
1743 EXPORT_SYMBOL(cxgb4_sync_txq_pidx);
1744
1745 int cxgb4_read_tpte(struct net_device *dev, u32 stag, __be32 *tpte)
1746 {
1747         u32 edc0_size, edc1_size, mc0_size, mc1_size, size;
1748         u32 edc0_end, edc1_end, mc0_end, mc1_end;
1749         u32 offset, memtype, memaddr;
1750         struct adapter *adap;
1751         u32 hma_size = 0;
1752         int ret;
1753
1754         adap = netdev2adap(dev);
1755
1756         offset = ((stag >> 8) * 32) + adap->vres.stag.start;
1757
1758         /* Figure out where the offset lands in the Memory Type/Address scheme.
1759          * This code assumes that the memory is laid out starting at offset 0
1760          * with no breaks as: EDC0, EDC1, MC0, MC1. All cards have both EDC0
1761          * and EDC1.  Some cards will have neither MC0 nor MC1, most cards have
1762          * MC0, and some have both MC0 and MC1.
1763          */
1764         size = t4_read_reg(adap, MA_EDRAM0_BAR_A);
1765         edc0_size = EDRAM0_SIZE_G(size) << 20;
1766         size = t4_read_reg(adap, MA_EDRAM1_BAR_A);
1767         edc1_size = EDRAM1_SIZE_G(size) << 20;
1768         size = t4_read_reg(adap, MA_EXT_MEMORY0_BAR_A);
1769         mc0_size = EXT_MEM0_SIZE_G(size) << 20;
1770
1771         if (t4_read_reg(adap, MA_TARGET_MEM_ENABLE_A) & HMA_MUX_F) {
1772                 size = t4_read_reg(adap, MA_EXT_MEMORY1_BAR_A);
1773                 hma_size = EXT_MEM1_SIZE_G(size) << 20;
1774         }
1775         edc0_end = edc0_size;
1776         edc1_end = edc0_end + edc1_size;
1777         mc0_end = edc1_end + mc0_size;
1778
1779         if (offset < edc0_end) {
1780                 memtype = MEM_EDC0;
1781                 memaddr = offset;
1782         } else if (offset < edc1_end) {
1783                 memtype = MEM_EDC1;
1784                 memaddr = offset - edc0_end;
1785         } else {
1786                 if (hma_size && (offset < (edc1_end + hma_size))) {
1787                         memtype = MEM_HMA;
1788                         memaddr = offset - edc1_end;
1789                 } else if (offset < mc0_end) {
1790                         memtype = MEM_MC0;
1791                         memaddr = offset - edc1_end;
1792                 } else if (is_t5(adap->params.chip)) {
1793                         size = t4_read_reg(adap, MA_EXT_MEMORY1_BAR_A);
1794                         mc1_size = EXT_MEM1_SIZE_G(size) << 20;
1795                         mc1_end = mc0_end + mc1_size;
1796                         if (offset < mc1_end) {
1797                                 memtype = MEM_MC1;
1798                                 memaddr = offset - mc0_end;
1799                         } else {
1800                                 /* offset beyond the end of any memory */
1801                                 goto err;
1802                         }
1803                 } else {
1804                         /* T4/T6 only has a single memory channel */
1805                         goto err;
1806                 }
1807         }
1808
1809         spin_lock(&adap->win0_lock);
1810         ret = t4_memory_rw(adap, 0, memtype, memaddr, 32, tpte, T4_MEMORY_READ);
1811         spin_unlock(&adap->win0_lock);
1812         return ret;
1813
1814 err:
1815         dev_err(adap->pdev_dev, "stag %#x, offset %#x out of range\n",
1816                 stag, offset);
1817         return -EINVAL;
1818 }
1819 EXPORT_SYMBOL(cxgb4_read_tpte);
1820
1821 u64 cxgb4_read_sge_timestamp(struct net_device *dev)
1822 {
1823         u32 hi, lo;
1824         struct adapter *adap;
1825
1826         adap = netdev2adap(dev);
1827         lo = t4_read_reg(adap, SGE_TIMESTAMP_LO_A);
1828         hi = TSVAL_G(t4_read_reg(adap, SGE_TIMESTAMP_HI_A));
1829
1830         return ((u64)hi << 32) | (u64)lo;
1831 }
1832 EXPORT_SYMBOL(cxgb4_read_sge_timestamp);
1833
1834 int cxgb4_bar2_sge_qregs(struct net_device *dev,
1835                          unsigned int qid,
1836                          enum cxgb4_bar2_qtype qtype,
1837                          int user,
1838                          u64 *pbar2_qoffset,
1839                          unsigned int *pbar2_qid)
1840 {
1841         return t4_bar2_sge_qregs(netdev2adap(dev),
1842                                  qid,
1843                                  (qtype == CXGB4_BAR2_QTYPE_EGRESS
1844                                   ? T4_BAR2_QTYPE_EGRESS
1845                                   : T4_BAR2_QTYPE_INGRESS),
1846                                  user,
1847                                  pbar2_qoffset,
1848                                  pbar2_qid);
1849 }
1850 EXPORT_SYMBOL(cxgb4_bar2_sge_qregs);
1851
1852 static struct pci_driver cxgb4_driver;
1853
1854 static void check_neigh_update(struct neighbour *neigh)
1855 {
1856         const struct device *parent;
1857         const struct net_device *netdev = neigh->dev;
1858
1859         if (is_vlan_dev(netdev))
1860                 netdev = vlan_dev_real_dev(netdev);
1861         parent = netdev->dev.parent;
1862         if (parent && parent->driver == &cxgb4_driver.driver)
1863                 t4_l2t_update(dev_get_drvdata(parent), neigh);
1864 }
1865
1866 static int netevent_cb(struct notifier_block *nb, unsigned long event,
1867                        void *data)
1868 {
1869         switch (event) {
1870         case NETEVENT_NEIGH_UPDATE:
1871                 check_neigh_update(data);
1872                 break;
1873         case NETEVENT_REDIRECT:
1874         default:
1875                 break;
1876         }
1877         return 0;
1878 }
1879
1880 static bool netevent_registered;
1881 static struct notifier_block cxgb4_netevent_nb = {
1882         .notifier_call = netevent_cb
1883 };
1884
1885 static void drain_db_fifo(struct adapter *adap, int usecs)
1886 {
1887         u32 v1, v2, lp_count, hp_count;
1888
1889         do {
1890                 v1 = t4_read_reg(adap, SGE_DBFIFO_STATUS_A);
1891                 v2 = t4_read_reg(adap, SGE_DBFIFO_STATUS2_A);
1892                 if (is_t4(adap->params.chip)) {
1893                         lp_count = LP_COUNT_G(v1);
1894                         hp_count = HP_COUNT_G(v1);
1895                 } else {
1896                         lp_count = LP_COUNT_T5_G(v1);
1897                         hp_count = HP_COUNT_T5_G(v2);
1898                 }
1899
1900                 if (lp_count == 0 && hp_count == 0)
1901                         break;
1902                 set_current_state(TASK_UNINTERRUPTIBLE);
1903                 schedule_timeout(usecs_to_jiffies(usecs));
1904         } while (1);
1905 }
1906
1907 static void disable_txq_db(struct sge_txq *q)
1908 {
1909         unsigned long flags;
1910
1911         spin_lock_irqsave(&q->db_lock, flags);
1912         q->db_disabled = 1;
1913         spin_unlock_irqrestore(&q->db_lock, flags);
1914 }
1915
1916 static void enable_txq_db(struct adapter *adap, struct sge_txq *q)
1917 {
1918         spin_lock_irq(&q->db_lock);
1919         if (q->db_pidx_inc) {
1920                 /* Make sure that all writes to the TX descriptors
1921                  * are committed before we tell HW about them.
1922                  */
1923                 wmb();
1924                 t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL_A),
1925                              QID_V(q->cntxt_id) | PIDX_V(q->db_pidx_inc));
1926                 q->db_pidx_inc = 0;
1927         }
1928         q->db_disabled = 0;
1929         spin_unlock_irq(&q->db_lock);
1930 }
1931
1932 static void disable_dbs(struct adapter *adap)
1933 {
1934         int i;
1935
1936         for_each_ethrxq(&adap->sge, i)
1937                 disable_txq_db(&adap->sge.ethtxq[i].q);
1938         if (is_offload(adap)) {
1939                 struct sge_uld_txq_info *txq_info =
1940                         adap->sge.uld_txq_info[CXGB4_TX_OFLD];
1941
1942                 if (txq_info) {
1943                         for_each_ofldtxq(&adap->sge, i) {
1944                                 struct sge_uld_txq *txq = &txq_info->uldtxq[i];
1945
1946                                 disable_txq_db(&txq->q);
1947                         }
1948                 }
1949         }
1950         for_each_port(adap, i)
1951                 disable_txq_db(&adap->sge.ctrlq[i].q);
1952 }
1953
1954 static void enable_dbs(struct adapter *adap)
1955 {
1956         int i;
1957
1958         for_each_ethrxq(&adap->sge, i)
1959                 enable_txq_db(adap, &adap->sge.ethtxq[i].q);
1960         if (is_offload(adap)) {
1961                 struct sge_uld_txq_info *txq_info =
1962                         adap->sge.uld_txq_info[CXGB4_TX_OFLD];
1963
1964                 if (txq_info) {
1965                         for_each_ofldtxq(&adap->sge, i) {
1966                                 struct sge_uld_txq *txq = &txq_info->uldtxq[i];
1967
1968                                 enable_txq_db(adap, &txq->q);
1969                         }
1970                 }
1971         }
1972         for_each_port(adap, i)
1973                 enable_txq_db(adap, &adap->sge.ctrlq[i].q);
1974 }
1975
1976 static void notify_rdma_uld(struct adapter *adap, enum cxgb4_control cmd)
1977 {
1978         enum cxgb4_uld type = CXGB4_ULD_RDMA;
1979
1980         if (adap->uld && adap->uld[type].handle)
1981                 adap->uld[type].control(adap->uld[type].handle, cmd);
1982 }
1983
1984 static void process_db_full(struct work_struct *work)
1985 {
1986         struct adapter *adap;
1987
1988         adap = container_of(work, struct adapter, db_full_task);
1989
1990         drain_db_fifo(adap, dbfifo_drain_delay);
1991         enable_dbs(adap);
1992         notify_rdma_uld(adap, CXGB4_CONTROL_DB_EMPTY);
1993         if (CHELSIO_CHIP_VERSION(adap->params.chip) <= CHELSIO_T5)
1994                 t4_set_reg_field(adap, SGE_INT_ENABLE3_A,
1995                                  DBFIFO_HP_INT_F | DBFIFO_LP_INT_F,
1996                                  DBFIFO_HP_INT_F | DBFIFO_LP_INT_F);
1997         else
1998                 t4_set_reg_field(adap, SGE_INT_ENABLE3_A,
1999                                  DBFIFO_LP_INT_F, DBFIFO_LP_INT_F);
2000 }
2001
2002 static void sync_txq_pidx(struct adapter *adap, struct sge_txq *q)
2003 {
2004         u16 hw_pidx, hw_cidx;
2005         int ret;
2006
2007         spin_lock_irq(&q->db_lock);
2008         ret = read_eq_indices(adap, (u16)q->cntxt_id, &hw_pidx, &hw_cidx);
2009         if (ret)
2010                 goto out;
2011         if (q->db_pidx != hw_pidx) {
2012                 u16 delta;
2013                 u32 val;
2014
2015                 if (q->db_pidx >= hw_pidx)
2016                         delta = q->db_pidx - hw_pidx;
2017                 else
2018                         delta = q->size - hw_pidx + q->db_pidx;
2019
2020                 if (is_t4(adap->params.chip))
2021                         val = PIDX_V(delta);
2022                 else
2023                         val = PIDX_T5_V(delta);
2024                 wmb();
2025                 t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL_A),
2026                              QID_V(q->cntxt_id) | val);
2027         }
2028 out:
2029         q->db_disabled = 0;
2030         q->db_pidx_inc = 0;
2031         spin_unlock_irq(&q->db_lock);
2032         if (ret)
2033                 CH_WARN(adap, "DB drop recovery failed.\n");
2034 }
2035
2036 static void recover_all_queues(struct adapter *adap)
2037 {
2038         int i;
2039
2040         for_each_ethrxq(&adap->sge, i)
2041                 sync_txq_pidx(adap, &adap->sge.ethtxq[i].q);
2042         if (is_offload(adap)) {
2043                 struct sge_uld_txq_info *txq_info =
2044                         adap->sge.uld_txq_info[CXGB4_TX_OFLD];
2045                 if (txq_info) {
2046                         for_each_ofldtxq(&adap->sge, i) {
2047                                 struct sge_uld_txq *txq = &txq_info->uldtxq[i];
2048
2049                                 sync_txq_pidx(adap, &txq->q);
2050                         }
2051                 }
2052         }
2053         for_each_port(adap, i)
2054                 sync_txq_pidx(adap, &adap->sge.ctrlq[i].q);
2055 }
2056
2057 static void process_db_drop(struct work_struct *work)
2058 {
2059         struct adapter *adap;
2060
2061         adap = container_of(work, struct adapter, db_drop_task);
2062
2063         if (is_t4(adap->params.chip)) {
2064                 drain_db_fifo(adap, dbfifo_drain_delay);
2065                 notify_rdma_uld(adap, CXGB4_CONTROL_DB_DROP);
2066                 drain_db_fifo(adap, dbfifo_drain_delay);
2067                 recover_all_queues(adap);
2068                 drain_db_fifo(adap, dbfifo_drain_delay);
2069                 enable_dbs(adap);
2070                 notify_rdma_uld(adap, CXGB4_CONTROL_DB_EMPTY);
2071         } else if (is_t5(adap->params.chip)) {
2072                 u32 dropped_db = t4_read_reg(adap, 0x010ac);
2073                 u16 qid = (dropped_db >> 15) & 0x1ffff;
2074                 u16 pidx_inc = dropped_db & 0x1fff;
2075                 u64 bar2_qoffset;
2076                 unsigned int bar2_qid;
2077                 int ret;
2078
2079                 ret = t4_bar2_sge_qregs(adap, qid, T4_BAR2_QTYPE_EGRESS,
2080                                         0, &bar2_qoffset, &bar2_qid);
2081                 if (ret)
2082                         dev_err(adap->pdev_dev, "doorbell drop recovery: "
2083                                 "qid=%d, pidx_inc=%d\n", qid, pidx_inc);
2084                 else
2085                         writel(PIDX_T5_V(pidx_inc) | QID_V(bar2_qid),
2086                                adap->bar2 + bar2_qoffset + SGE_UDB_KDOORBELL);
2087
2088                 /* Re-enable BAR2 WC */
2089                 t4_set_reg_field(adap, 0x10b0, 1<<15, 1<<15);
2090         }
2091
2092         if (CHELSIO_CHIP_VERSION(adap->params.chip) <= CHELSIO_T5)
2093                 t4_set_reg_field(adap, SGE_DOORBELL_CONTROL_A, DROPPED_DB_F, 0);
2094 }
2095
2096 void t4_db_full(struct adapter *adap)
2097 {
2098         if (is_t4(adap->params.chip)) {
2099                 disable_dbs(adap);
2100                 notify_rdma_uld(adap, CXGB4_CONTROL_DB_FULL);
2101                 t4_set_reg_field(adap, SGE_INT_ENABLE3_A,
2102                                  DBFIFO_HP_INT_F | DBFIFO_LP_INT_F, 0);
2103                 queue_work(adap->workq, &adap->db_full_task);
2104         }
2105 }
2106
2107 void t4_db_dropped(struct adapter *adap)
2108 {
2109         if (is_t4(adap->params.chip)) {
2110                 disable_dbs(adap);
2111                 notify_rdma_uld(adap, CXGB4_CONTROL_DB_FULL);
2112         }
2113         queue_work(adap->workq, &adap->db_drop_task);
2114 }
2115
2116 void t4_register_netevent_notifier(void)
2117 {
2118         if (!netevent_registered) {
2119                 register_netevent_notifier(&cxgb4_netevent_nb);
2120                 netevent_registered = true;
2121         }
2122 }
2123
2124 static void detach_ulds(struct adapter *adap)
2125 {
2126         unsigned int i;
2127
2128         mutex_lock(&uld_mutex);
2129         list_del(&adap->list_node);
2130
2131         for (i = 0; i < CXGB4_ULD_MAX; i++)
2132                 if (adap->uld && adap->uld[i].handle)
2133                         adap->uld[i].state_change(adap->uld[i].handle,
2134                                              CXGB4_STATE_DETACH);
2135
2136         if (netevent_registered && list_empty(&adapter_list)) {
2137                 unregister_netevent_notifier(&cxgb4_netevent_nb);
2138                 netevent_registered = false;
2139         }
2140         mutex_unlock(&uld_mutex);
2141 }
2142
2143 static void notify_ulds(struct adapter *adap, enum cxgb4_state new_state)
2144 {
2145         unsigned int i;
2146
2147         mutex_lock(&uld_mutex);
2148         for (i = 0; i < CXGB4_ULD_MAX; i++)
2149                 if (adap->uld && adap->uld[i].handle)
2150                         adap->uld[i].state_change(adap->uld[i].handle,
2151                                                   new_state);
2152         mutex_unlock(&uld_mutex);
2153 }
2154
2155 #if IS_ENABLED(CONFIG_IPV6)
2156 static int cxgb4_inet6addr_handler(struct notifier_block *this,
2157                                    unsigned long event, void *data)
2158 {
2159         struct inet6_ifaddr *ifa = data;
2160         struct net_device *event_dev = ifa->idev->dev;
2161         const struct device *parent = NULL;
2162 #if IS_ENABLED(CONFIG_BONDING)
2163         struct adapter *adap;
2164 #endif
2165         if (is_vlan_dev(event_dev))
2166                 event_dev = vlan_dev_real_dev(event_dev);
2167 #if IS_ENABLED(CONFIG_BONDING)
2168         if (event_dev->flags & IFF_MASTER) {
2169                 list_for_each_entry(adap, &adapter_list, list_node) {
2170                         switch (event) {
2171                         case NETDEV_UP:
2172                                 cxgb4_clip_get(adap->port[0],
2173                                                (const u32 *)ifa, 1);
2174                                 break;
2175                         case NETDEV_DOWN:
2176                                 cxgb4_clip_release(adap->port[0],
2177                                                    (const u32 *)ifa, 1);
2178                                 break;
2179                         default:
2180                                 break;
2181                         }
2182                 }
2183                 return NOTIFY_OK;
2184         }
2185 #endif
2186
2187         if (event_dev)
2188                 parent = event_dev->dev.parent;
2189
2190         if (parent && parent->driver == &cxgb4_driver.driver) {
2191                 switch (event) {
2192                 case NETDEV_UP:
2193                         cxgb4_clip_get(event_dev, (const u32 *)ifa, 1);
2194                         break;
2195                 case NETDEV_DOWN:
2196                         cxgb4_clip_release(event_dev, (const u32 *)ifa, 1);
2197                         break;
2198                 default:
2199                         break;
2200                 }
2201         }
2202         return NOTIFY_OK;
2203 }
2204
2205 static bool inet6addr_registered;
2206 static struct notifier_block cxgb4_inet6addr_notifier = {
2207         .notifier_call = cxgb4_inet6addr_handler
2208 };
2209
2210 static void update_clip(const struct adapter *adap)
2211 {
2212         int i;
2213         struct net_device *dev;
2214         int ret;
2215
2216         rcu_read_lock();
2217
2218         for (i = 0; i < MAX_NPORTS; i++) {
2219                 dev = adap->port[i];
2220                 ret = 0;
2221
2222                 if (dev)
2223                         ret = cxgb4_update_root_dev_clip(dev);
2224
2225                 if (ret < 0)
2226                         break;
2227         }
2228         rcu_read_unlock();
2229 }
2230 #endif /* IS_ENABLED(CONFIG_IPV6) */
2231
2232 /**
2233  *      cxgb_up - enable the adapter
2234  *      @adap: adapter being enabled
2235  *
2236  *      Called when the first port is enabled, this function performs the
2237  *      actions necessary to make an adapter operational, such as completing
2238  *      the initialization of HW modules, and enabling interrupts.
2239  *
2240  *      Must be called with the rtnl lock held.
2241  */
2242 static int cxgb_up(struct adapter *adap)
2243 {
2244         int err;
2245
2246         mutex_lock(&uld_mutex);
2247         err = setup_sge_queues(adap);
2248         if (err)
2249                 goto rel_lock;
2250         err = setup_rss(adap);
2251         if (err)
2252                 goto freeq;
2253
2254         if (adap->flags & USING_MSIX) {
2255                 name_msix_vecs(adap);
2256                 err = request_irq(adap->msix_info[0].vec, t4_nondata_intr, 0,
2257                                   adap->msix_info[0].desc, adap);
2258                 if (err)
2259                         goto irq_err;
2260                 err = request_msix_queue_irqs(adap);
2261                 if (err) {
2262                         free_irq(adap->msix_info[0].vec, adap);
2263                         goto irq_err;
2264                 }
2265         } else {
2266                 err = request_irq(adap->pdev->irq, t4_intr_handler(adap),
2267                                   (adap->flags & USING_MSI) ? 0 : IRQF_SHARED,
2268                                   adap->port[0]->name, adap);
2269                 if (err)
2270                         goto irq_err;
2271         }
2272
2273         enable_rx(adap);
2274         t4_sge_start(adap);
2275         t4_intr_enable(adap);
2276         adap->flags |= FULL_INIT_DONE;
2277         mutex_unlock(&uld_mutex);
2278
2279         notify_ulds(adap, CXGB4_STATE_UP);
2280 #if IS_ENABLED(CONFIG_IPV6)
2281         update_clip(adap);
2282 #endif
2283         /* Initialize hash mac addr list*/
2284         INIT_LIST_HEAD(&adap->mac_hlist);
2285         return err;
2286
2287  irq_err:
2288         dev_err(adap->pdev_dev, "request_irq failed, err %d\n", err);
2289  freeq:
2290         t4_free_sge_resources(adap);
2291  rel_lock:
2292         mutex_unlock(&uld_mutex);
2293         return err;
2294 }
2295
2296 static void cxgb_down(struct adapter *adapter)
2297 {
2298         cancel_work_sync(&adapter->tid_release_task);
2299         cancel_work_sync(&adapter->db_full_task);
2300         cancel_work_sync(&adapter->db_drop_task);
2301         adapter->tid_release_task_busy = false;
2302         adapter->tid_release_head = NULL;
2303
2304         t4_sge_stop(adapter);
2305         t4_free_sge_resources(adapter);
2306         adapter->flags &= ~FULL_INIT_DONE;
2307 }
2308
2309 /*
2310  * net_device operations
2311  */
2312 static int cxgb_open(struct net_device *dev)
2313 {
2314         int err;
2315         struct port_info *pi = netdev_priv(dev);
2316         struct adapter *adapter = pi->adapter;
2317
2318         netif_carrier_off(dev);
2319
2320         if (!(adapter->flags & FULL_INIT_DONE)) {
2321                 err = cxgb_up(adapter);
2322                 if (err < 0)
2323                         return err;
2324         }
2325
2326         /* It's possible that the basic port information could have
2327          * changed since we first read it.
2328          */
2329         err = t4_update_port_info(pi);
2330         if (err < 0)
2331                 return err;
2332
2333         err = link_start(dev);
2334         if (!err)
2335                 netif_tx_start_all_queues(dev);
2336         return err;
2337 }
2338
2339 static int cxgb_close(struct net_device *dev)
2340 {
2341         struct port_info *pi = netdev_priv(dev);
2342         struct adapter *adapter = pi->adapter;
2343         int ret;
2344
2345         netif_tx_stop_all_queues(dev);
2346         netif_carrier_off(dev);
2347         ret = t4_enable_pi_params(adapter, adapter->pf, pi,
2348                                   false, false, false);
2349 #ifdef CONFIG_CHELSIO_T4_DCB
2350         cxgb4_dcb_reset(dev);
2351         dcb_tx_queue_prio_enable(dev, false);
2352 #endif
2353         return ret;
2354 }
2355
2356 int cxgb4_create_server_filter(const struct net_device *dev, unsigned int stid,
2357                 __be32 sip, __be16 sport, __be16 vlan,
2358                 unsigned int queue, unsigned char port, unsigned char mask)
2359 {
2360         int ret;
2361         struct filter_entry *f;
2362         struct adapter *adap;
2363         int i;
2364         u8 *val;
2365
2366         adap = netdev2adap(dev);
2367
2368         /* Adjust stid to correct filter index */
2369         stid -= adap->tids.sftid_base;
2370         stid += adap->tids.nftids;
2371
2372         /* Check to make sure the filter requested is writable ...
2373          */
2374         f = &adap->tids.ftid_tab[stid];
2375         ret = writable_filter(f);
2376         if (ret)
2377                 return ret;
2378
2379         /* Clear out any old resources being used by the filter before
2380          * we start constructing the new filter.
2381          */
2382         if (f->valid)
2383                 clear_filter(adap, f);
2384
2385         /* Clear out filter specifications */
2386         memset(&f->fs, 0, sizeof(struct ch_filter_specification));
2387         f->fs.val.lport = cpu_to_be16(sport);
2388         f->fs.mask.lport  = ~0;
2389         val = (u8 *)&sip;
2390         if ((val[0] | val[1] | val[2] | val[3]) != 0) {
2391                 for (i = 0; i < 4; i++) {
2392                         f->fs.val.lip[i] = val[i];
2393                         f->fs.mask.lip[i] = ~0;
2394                 }
2395                 if (adap->params.tp.vlan_pri_map & PORT_F) {
2396                         f->fs.val.iport = port;
2397                         f->fs.mask.iport = mask;
2398                 }
2399         }
2400
2401         if (adap->params.tp.vlan_pri_map & PROTOCOL_F) {
2402                 f->fs.val.proto = IPPROTO_TCP;
2403                 f->fs.mask.proto = ~0;
2404         }
2405
2406         f->fs.dirsteer = 1;
2407         f->fs.iq = queue;
2408         /* Mark filter as locked */
2409         f->locked = 1;
2410         f->fs.rpttid = 1;
2411
2412         /* Save the actual tid. We need this to get the corresponding
2413          * filter entry structure in filter_rpl.
2414          */
2415         f->tid = stid + adap->tids.ftid_base;
2416         ret = set_filter_wr(adap, stid);
2417         if (ret) {
2418                 clear_filter(adap, f);
2419                 return ret;
2420         }
2421
2422         return 0;
2423 }
2424 EXPORT_SYMBOL(cxgb4_create_server_filter);
2425
2426 int cxgb4_remove_server_filter(const struct net_device *dev, unsigned int stid,
2427                 unsigned int queue, bool ipv6)
2428 {
2429         struct filter_entry *f;
2430         struct adapter *adap;
2431
2432         adap = netdev2adap(dev);
2433
2434         /* Adjust stid to correct filter index */
2435         stid -= adap->tids.sftid_base;
2436         stid += adap->tids.nftids;
2437
2438         f = &adap->tids.ftid_tab[stid];
2439         /* Unlock the filter */
2440         f->locked = 0;
2441
2442         return delete_filter(adap, stid);
2443 }
2444 EXPORT_SYMBOL(cxgb4_remove_server_filter);
2445
2446 static void cxgb_get_stats(struct net_device *dev,
2447                            struct rtnl_link_stats64 *ns)
2448 {
2449         struct port_stats stats;
2450         struct port_info *p = netdev_priv(dev);
2451         struct adapter *adapter = p->adapter;
2452
2453         /* Block retrieving statistics during EEH error
2454          * recovery. Otherwise, the recovery might fail
2455          * and the PCI device will be removed permanently
2456          */
2457         spin_lock(&adapter->stats_lock);
2458         if (!netif_device_present(dev)) {
2459                 spin_unlock(&adapter->stats_lock);
2460                 return;
2461         }
2462         t4_get_port_stats_offset(adapter, p->tx_chan, &stats,
2463                                  &p->stats_base);
2464         spin_unlock(&adapter->stats_lock);
2465
2466         ns->tx_bytes   = stats.tx_octets;
2467         ns->tx_packets = stats.tx_frames;
2468         ns->rx_bytes   = stats.rx_octets;
2469         ns->rx_packets = stats.rx_frames;
2470         ns->multicast  = stats.rx_mcast_frames;
2471
2472         /* detailed rx_errors */
2473         ns->rx_length_errors = stats.rx_jabber + stats.rx_too_long +
2474                                stats.rx_runt;
2475         ns->rx_over_errors   = 0;
2476         ns->rx_crc_errors    = stats.rx_fcs_err;
2477         ns->rx_frame_errors  = stats.rx_symbol_err;
2478         ns->rx_dropped       = stats.rx_ovflow0 + stats.rx_ovflow1 +
2479                                stats.rx_ovflow2 + stats.rx_ovflow3 +
2480                                stats.rx_trunc0 + stats.rx_trunc1 +
2481                                stats.rx_trunc2 + stats.rx_trunc3;
2482         ns->rx_missed_errors = 0;
2483
2484         /* detailed tx_errors */
2485         ns->tx_aborted_errors   = 0;
2486         ns->tx_carrier_errors   = 0;
2487         ns->tx_fifo_errors      = 0;
2488         ns->tx_heartbeat_errors = 0;
2489         ns->tx_window_errors    = 0;
2490
2491         ns->tx_errors = stats.tx_error_frames;
2492         ns->rx_errors = stats.rx_symbol_err + stats.rx_fcs_err +
2493                 ns->rx_length_errors + stats.rx_len_err + ns->rx_fifo_errors;
2494 }
2495
2496 static int cxgb_ioctl(struct net_device *dev, struct ifreq *req, int cmd)
2497 {
2498         unsigned int mbox;
2499         int ret = 0, prtad, devad;
2500         struct port_info *pi = netdev_priv(dev);
2501         struct adapter *adapter = pi->adapter;
2502         struct mii_ioctl_data *data = (struct mii_ioctl_data *)&req->ifr_data;
2503
2504         switch (cmd) {
2505         case SIOCGMIIPHY:
2506                 if (pi->mdio_addr < 0)
2507                         return -EOPNOTSUPP;
2508                 data->phy_id = pi->mdio_addr;
2509                 break;
2510         case SIOCGMIIREG:
2511         case SIOCSMIIREG:
2512                 if (mdio_phy_id_is_c45(data->phy_id)) {
2513                         prtad = mdio_phy_id_prtad(data->phy_id);
2514                         devad = mdio_phy_id_devad(data->phy_id);
2515                 } else if (data->phy_id < 32) {
2516                         prtad = data->phy_id;
2517                         devad = 0;
2518                         data->reg_num &= 0x1f;
2519                 } else
2520                         return -EINVAL;
2521
2522                 mbox = pi->adapter->pf;
2523                 if (cmd == SIOCGMIIREG)
2524                         ret = t4_mdio_rd(pi->adapter, mbox, prtad, devad,
2525                                          data->reg_num, &data->val_out);
2526                 else
2527                         ret = t4_mdio_wr(pi->adapter, mbox, prtad, devad,
2528                                          data->reg_num, data->val_in);
2529                 break;
2530         case SIOCGHWTSTAMP:
2531                 return copy_to_user(req->ifr_data, &pi->tstamp_config,
2532                                     sizeof(pi->tstamp_config)) ?
2533                         -EFAULT : 0;
2534         case SIOCSHWTSTAMP:
2535                 if (copy_from_user(&pi->tstamp_config, req->ifr_data,
2536                                    sizeof(pi->tstamp_config)))
2537                         return -EFAULT;
2538
2539                 if (!is_t4(adapter->params.chip)) {
2540                         switch (pi->tstamp_config.tx_type) {
2541                         case HWTSTAMP_TX_OFF:
2542                         case HWTSTAMP_TX_ON:
2543                                 break;
2544                         default:
2545                                 return -ERANGE;
2546                         }
2547
2548                         switch (pi->tstamp_config.rx_filter) {
2549                         case HWTSTAMP_FILTER_NONE:
2550                                 pi->rxtstamp = false;
2551                                 break;
2552                         case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
2553                         case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
2554                                 cxgb4_ptprx_timestamping(pi, pi->port_id,
2555                                                          PTP_TS_L4);
2556                                 break;
2557                         case HWTSTAMP_FILTER_PTP_V2_EVENT:
2558                                 cxgb4_ptprx_timestamping(pi, pi->port_id,
2559                                                          PTP_TS_L2_L4);
2560                                 break;
2561                         case HWTSTAMP_FILTER_ALL:
2562                         case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
2563                         case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
2564                         case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
2565                         case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
2566                                 pi->rxtstamp = true;
2567                                 break;
2568                         default:
2569                                 pi->tstamp_config.rx_filter =
2570                                         HWTSTAMP_FILTER_NONE;
2571                                 return -ERANGE;
2572                         }
2573
2574                         if ((pi->tstamp_config.tx_type == HWTSTAMP_TX_OFF) &&
2575                             (pi->tstamp_config.rx_filter ==
2576                                 HWTSTAMP_FILTER_NONE)) {
2577                                 if (cxgb4_ptp_txtype(adapter, pi->port_id) >= 0)
2578                                         pi->ptp_enable = false;
2579                         }
2580
2581                         if (pi->tstamp_config.rx_filter !=
2582                                 HWTSTAMP_FILTER_NONE) {
2583                                 if (cxgb4_ptp_redirect_rx_packet(adapter,
2584                                                                  pi) >= 0)
2585                                         pi->ptp_enable = true;
2586                         }
2587                 } else {
2588                         /* For T4 Adapters */
2589                         switch (pi->tstamp_config.rx_filter) {
2590                         case HWTSTAMP_FILTER_NONE:
2591                         pi->rxtstamp = false;
2592                         break;
2593                         case HWTSTAMP_FILTER_ALL:
2594                         pi->rxtstamp = true;
2595                         break;
2596                         default:
2597                         pi->tstamp_config.rx_filter =
2598                         HWTSTAMP_FILTER_NONE;
2599                         return -ERANGE;
2600                         }
2601                 }
2602                 return copy_to_user(req->ifr_data, &pi->tstamp_config,
2603                                     sizeof(pi->tstamp_config)) ?
2604                         -EFAULT : 0;
2605         default:
2606                 return -EOPNOTSUPP;
2607         }
2608         return ret;
2609 }
2610
2611 static void cxgb_set_rxmode(struct net_device *dev)
2612 {
2613         /* unfortunately we can't return errors to the stack */
2614         set_rxmode(dev, -1, false);
2615 }
2616
2617 static int cxgb_change_mtu(struct net_device *dev, int new_mtu)
2618 {
2619         int ret;
2620         struct port_info *pi = netdev_priv(dev);
2621
2622         ret = t4_set_rxmode(pi->adapter, pi->adapter->pf, pi->viid, new_mtu, -1,
2623                             -1, -1, -1, true);
2624         if (!ret)
2625                 dev->mtu = new_mtu;
2626         return ret;
2627 }
2628
2629 #ifdef CONFIG_PCI_IOV
2630 static int cxgb4_mgmt_open(struct net_device *dev)
2631 {
2632         /* Turn carrier off since we don't have to transmit anything on this
2633          * interface.
2634          */
2635         netif_carrier_off(dev);
2636         return 0;
2637 }
2638
2639 /* Fill MAC address that will be assigned by the FW */
2640 static void cxgb4_mgmt_fill_vf_station_mac_addr(struct adapter *adap)
2641 {
2642         u8 hw_addr[ETH_ALEN], macaddr[ETH_ALEN];
2643         unsigned int i, vf, nvfs;
2644         u16 a, b;
2645         int err;
2646         u8 *na;
2647
2648         adap->params.pci.vpd_cap_addr = pci_find_capability(adap->pdev,
2649                                                             PCI_CAP_ID_VPD);
2650         err = t4_get_raw_vpd_params(adap, &adap->params.vpd);
2651         if (err)
2652                 return;
2653
2654         na = adap->params.vpd.na;
2655         for (i = 0; i < ETH_ALEN; i++)
2656                 hw_addr[i] = (hex2val(na[2 * i + 0]) * 16 +
2657                               hex2val(na[2 * i + 1]));
2658
2659         a = (hw_addr[0] << 8) | hw_addr[1];
2660         b = (hw_addr[1] << 8) | hw_addr[2];
2661         a ^= b;
2662         a |= 0x0200;    /* locally assigned Ethernet MAC address */
2663         a &= ~0x0100;   /* not a multicast Ethernet MAC address */
2664         macaddr[0] = a >> 8;
2665         macaddr[1] = a & 0xff;
2666
2667         for (i = 2; i < 5; i++)
2668                 macaddr[i] = hw_addr[i + 1];
2669
2670         for (vf = 0, nvfs = pci_sriov_get_totalvfs(adap->pdev);
2671                 vf < nvfs; vf++) {
2672                 macaddr[5] = adap->pf * 16 + vf;
2673                 ether_addr_copy(adap->vfinfo[vf].vf_mac_addr, macaddr);
2674         }
2675 }
2676
2677 static int cxgb4_mgmt_set_vf_mac(struct net_device *dev, int vf, u8 *mac)
2678 {
2679         struct port_info *pi = netdev_priv(dev);
2680         struct adapter *adap = pi->adapter;
2681         int ret;
2682
2683         /* verify MAC addr is valid */
2684         if (!is_valid_ether_addr(mac)) {
2685                 dev_err(pi->adapter->pdev_dev,
2686                         "Invalid Ethernet address %pM for VF %d\n",
2687                         mac, vf);
2688                 return -EINVAL;
2689         }
2690
2691         dev_info(pi->adapter->pdev_dev,
2692                  "Setting MAC %pM on VF %d\n", mac, vf);
2693         ret = t4_set_vf_mac_acl(adap, vf + 1, 1, mac);
2694         if (!ret)
2695                 ether_addr_copy(adap->vfinfo[vf].vf_mac_addr, mac);
2696         return ret;
2697 }
2698
2699 static int cxgb4_mgmt_get_vf_config(struct net_device *dev,
2700                                     int vf, struct ifla_vf_info *ivi)
2701 {
2702         struct port_info *pi = netdev_priv(dev);
2703         struct adapter *adap = pi->adapter;
2704         struct vf_info *vfinfo;
2705
2706         if (vf >= adap->num_vfs)
2707                 return -EINVAL;
2708         vfinfo = &adap->vfinfo[vf];
2709
2710         ivi->vf = vf;
2711         ivi->max_tx_rate = vfinfo->tx_rate;
2712         ivi->min_tx_rate = 0;
2713         ether_addr_copy(ivi->mac, vfinfo->vf_mac_addr);
2714         ivi->vlan = vfinfo->vlan;
2715         return 0;
2716 }
2717
2718 static int cxgb4_mgmt_get_phys_port_id(struct net_device *dev,
2719                                        struct netdev_phys_item_id *ppid)
2720 {
2721         struct port_info *pi = netdev_priv(dev);
2722         unsigned int phy_port_id;
2723
2724         phy_port_id = pi->adapter->adap_idx * 10 + pi->port_id;
2725         ppid->id_len = sizeof(phy_port_id);
2726         memcpy(ppid->id, &phy_port_id, ppid->id_len);
2727         return 0;
2728 }
2729
2730 static int cxgb4_mgmt_set_vf_rate(struct net_device *dev, int vf,
2731                                   int min_tx_rate, int max_tx_rate)
2732 {
2733         struct port_info *pi = netdev_priv(dev);
2734         struct adapter *adap = pi->adapter;
2735         unsigned int link_ok, speed, mtu;
2736         u32 fw_pfvf, fw_class;
2737         int class_id = vf;
2738         int ret;
2739         u16 pktsize;
2740
2741         if (vf >= adap->num_vfs)
2742                 return -EINVAL;
2743
2744         if (min_tx_rate) {
2745                 dev_err(adap->pdev_dev,
2746                         "Min tx rate (%d) (> 0) for VF %d is Invalid.\n",
2747                         min_tx_rate, vf);
2748                 return -EINVAL;
2749         }
2750
2751         if (max_tx_rate == 0) {
2752                 /* unbind VF to to any Traffic Class */
2753                 fw_pfvf =
2754                     (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_PFVF) |
2755                      FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_PFVF_SCHEDCLASS_ETH));
2756                 fw_class = 0xffffffff;
2757                 ret = t4_set_params(adap, adap->mbox, adap->pf, vf + 1, 1,
2758                                     &fw_pfvf, &fw_class);
2759                 if (ret) {
2760                         dev_err(adap->pdev_dev,
2761                                 "Err %d in unbinding PF %d VF %d from TX Rate Limiting\n",
2762                                 ret, adap->pf, vf);
2763                         return -EINVAL;
2764                 }
2765                 dev_info(adap->pdev_dev,
2766                          "PF %d VF %d is unbound from TX Rate Limiting\n",
2767                          adap->pf, vf);
2768                 adap->vfinfo[vf].tx_rate = 0;
2769                 return 0;
2770         }
2771
2772         ret = t4_get_link_params(pi, &link_ok, &speed, &mtu);
2773         if (ret != FW_SUCCESS) {
2774                 dev_err(adap->pdev_dev,
2775                         "Failed to get link information for VF %d\n", vf);
2776                 return -EINVAL;
2777         }
2778
2779         if (!link_ok) {
2780                 dev_err(adap->pdev_dev, "Link down for VF %d\n", vf);
2781                 return -EINVAL;
2782         }
2783
2784         if (max_tx_rate > speed) {
2785                 dev_err(adap->pdev_dev,
2786                         "Max tx rate %d for VF %d can't be > link-speed %u",
2787                         max_tx_rate, vf, speed);
2788                 return -EINVAL;
2789         }
2790
2791         pktsize = mtu;
2792         /* subtract ethhdr size and 4 bytes crc since, f/w appends it */
2793         pktsize = pktsize - sizeof(struct ethhdr) - 4;
2794         /* subtract ipv4 hdr size, tcp hdr size to get typical IPv4 MSS size */
2795         pktsize = pktsize - sizeof(struct iphdr) - sizeof(struct tcphdr);
2796         /* configure Traffic Class for rate-limiting */
2797         ret = t4_sched_params(adap, SCHED_CLASS_TYPE_PACKET,
2798                               SCHED_CLASS_LEVEL_CL_RL,
2799                               SCHED_CLASS_MODE_CLASS,
2800                               SCHED_CLASS_RATEUNIT_BITS,
2801                               SCHED_CLASS_RATEMODE_ABS,
2802                               pi->tx_chan, class_id, 0,
2803                               max_tx_rate * 1000, 0, pktsize);
2804         if (ret) {
2805                 dev_err(adap->pdev_dev, "Err %d for Traffic Class config\n",
2806                         ret);
2807                 return -EINVAL;
2808         }
2809         dev_info(adap->pdev_dev,
2810                  "Class %d with MSS %u configured with rate %u\n",
2811                  class_id, pktsize, max_tx_rate);
2812
2813         /* bind VF to configured Traffic Class */
2814         fw_pfvf = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_PFVF) |
2815                    FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_PFVF_SCHEDCLASS_ETH));
2816         fw_class = class_id;
2817         ret = t4_set_params(adap, adap->mbox, adap->pf, vf + 1, 1, &fw_pfvf,
2818                             &fw_class);
2819         if (ret) {
2820                 dev_err(adap->pdev_dev,
2821                         "Err %d in binding PF %d VF %d to Traffic Class %d\n",
2822                         ret, adap->pf, vf, class_id);
2823                 return -EINVAL;
2824         }
2825         dev_info(adap->pdev_dev, "PF %d VF %d is bound to Class %d\n",
2826                  adap->pf, vf, class_id);
2827         adap->vfinfo[vf].tx_rate = max_tx_rate;
2828         return 0;
2829 }
2830
2831 static int cxgb4_mgmt_set_vf_vlan(struct net_device *dev, int vf,
2832                                   u16 vlan, u8 qos, __be16 vlan_proto)
2833 {
2834         struct port_info *pi = netdev_priv(dev);
2835         struct adapter *adap = pi->adapter;
2836         int ret;
2837
2838         if (vf >= adap->num_vfs || vlan > 4095 || qos > 7)
2839                 return -EINVAL;
2840
2841         if (vlan_proto != htons(ETH_P_8021Q) || qos != 0)
2842                 return -EPROTONOSUPPORT;
2843
2844         ret = t4_set_vlan_acl(adap, adap->mbox, vf + 1, vlan);
2845         if (!ret) {
2846                 adap->vfinfo[vf].vlan = vlan;
2847                 return 0;
2848         }
2849
2850         dev_err(adap->pdev_dev, "Err %d %s VLAN ACL for PF/VF %d/%d\n",
2851                 ret, (vlan ? "setting" : "clearing"), adap->pf, vf);
2852         return ret;
2853 }
2854 #endif /* CONFIG_PCI_IOV */
2855
2856 static int cxgb_set_mac_addr(struct net_device *dev, void *p)
2857 {
2858         int ret;
2859         struct sockaddr *addr = p;
2860         struct port_info *pi = netdev_priv(dev);
2861
2862         if (!is_valid_ether_addr(addr->sa_data))
2863                 return -EADDRNOTAVAIL;
2864
2865         ret = t4_change_mac(pi->adapter, pi->adapter->pf, pi->viid,
2866                             pi->xact_addr_filt, addr->sa_data, true, true);
2867         if (ret < 0)
2868                 return ret;
2869
2870         memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
2871         pi->xact_addr_filt = ret;
2872         return 0;
2873 }
2874
2875 #ifdef CONFIG_NET_POLL_CONTROLLER
2876 static void cxgb_netpoll(struct net_device *dev)
2877 {
2878         struct port_info *pi = netdev_priv(dev);
2879         struct adapter *adap = pi->adapter;
2880
2881         if (adap->flags & USING_MSIX) {
2882                 int i;
2883                 struct sge_eth_rxq *rx = &adap->sge.ethrxq[pi->first_qset];
2884
2885                 for (i = pi->nqsets; i; i--, rx++)
2886                         t4_sge_intr_msix(0, &rx->rspq);
2887         } else
2888                 t4_intr_handler(adap)(0, adap);
2889 }
2890 #endif
2891
2892 static int cxgb_set_tx_maxrate(struct net_device *dev, int index, u32 rate)
2893 {
2894         struct port_info *pi = netdev_priv(dev);
2895         struct adapter *adap = pi->adapter;
2896         struct sched_class *e;
2897         struct ch_sched_params p;
2898         struct ch_sched_queue qe;
2899         u32 req_rate;
2900         int err = 0;
2901
2902         if (!can_sched(dev))
2903                 return -ENOTSUPP;
2904
2905         if (index < 0 || index > pi->nqsets - 1)
2906                 return -EINVAL;
2907
2908         if (!(adap->flags & FULL_INIT_DONE)) {
2909                 dev_err(adap->pdev_dev,
2910                         "Failed to rate limit on queue %d. Link Down?\n",
2911                         index);
2912                 return -EINVAL;
2913         }
2914
2915         /* Convert from Mbps to Kbps */
2916         req_rate = rate * 1000;
2917
2918         /* Max rate is 100 Gbps */
2919         if (req_rate > SCHED_MAX_RATE_KBPS) {
2920                 dev_err(adap->pdev_dev,
2921                         "Invalid rate %u Mbps, Max rate is %u Mbps\n",
2922                         rate, SCHED_MAX_RATE_KBPS / 1000);
2923                 return -ERANGE;
2924         }
2925
2926         /* First unbind the queue from any existing class */
2927         memset(&qe, 0, sizeof(qe));
2928         qe.queue = index;
2929         qe.class = SCHED_CLS_NONE;
2930
2931         err = cxgb4_sched_class_unbind(dev, (void *)(&qe), SCHED_QUEUE);
2932         if (err) {
2933                 dev_err(adap->pdev_dev,
2934                         "Unbinding Queue %d on port %d fail. Err: %d\n",
2935                         index, pi->port_id, err);
2936                 return err;
2937         }
2938
2939         /* Queue already unbound */
2940         if (!req_rate)
2941                 return 0;
2942
2943         /* Fetch any available unused or matching scheduling class */
2944         memset(&p, 0, sizeof(p));
2945         p.type = SCHED_CLASS_TYPE_PACKET;
2946         p.u.params.level    = SCHED_CLASS_LEVEL_CL_RL;
2947         p.u.params.mode     = SCHED_CLASS_MODE_CLASS;
2948         p.u.params.rateunit = SCHED_CLASS_RATEUNIT_BITS;
2949         p.u.params.ratemode = SCHED_CLASS_RATEMODE_ABS;
2950         p.u.params.channel  = pi->tx_chan;
2951         p.u.params.class    = SCHED_CLS_NONE;
2952         p.u.params.minrate  = 0;
2953         p.u.params.maxrate  = req_rate;
2954         p.u.params.weight   = 0;
2955         p.u.params.pktsize  = dev->mtu;
2956
2957         e = cxgb4_sched_class_alloc(dev, &p);
2958         if (!e)
2959                 return -ENOMEM;
2960
2961         /* Bind the queue to a scheduling class */
2962         memset(&qe, 0, sizeof(qe));
2963         qe.queue = index;
2964         qe.class = e->idx;
2965
2966         err = cxgb4_sched_class_bind(dev, (void *)(&qe), SCHED_QUEUE);
2967         if (err)
2968                 dev_err(adap->pdev_dev,
2969                         "Queue rate limiting failed. Err: %d\n", err);
2970         return err;
2971 }
2972
2973 static int cxgb_setup_tc_flower(struct net_device *dev,
2974                                 struct tc_cls_flower_offload *cls_flower)
2975 {
2976         switch (cls_flower->command) {
2977         case TC_CLSFLOWER_REPLACE:
2978                 return cxgb4_tc_flower_replace(dev, cls_flower);
2979         case TC_CLSFLOWER_DESTROY:
2980                 return cxgb4_tc_flower_destroy(dev, cls_flower);
2981         case TC_CLSFLOWER_STATS:
2982                 return cxgb4_tc_flower_stats(dev, cls_flower);
2983         default:
2984                 return -EOPNOTSUPP;
2985         }
2986 }
2987
2988 static int cxgb_setup_tc_cls_u32(struct net_device *dev,
2989                                  struct tc_cls_u32_offload *cls_u32)
2990 {
2991         switch (cls_u32->command) {
2992         case TC_CLSU32_NEW_KNODE:
2993         case TC_CLSU32_REPLACE_KNODE:
2994                 return cxgb4_config_knode(dev, cls_u32);
2995         case TC_CLSU32_DELETE_KNODE:
2996                 return cxgb4_delete_knode(dev, cls_u32);
2997         default:
2998                 return -EOPNOTSUPP;
2999         }
3000 }
3001
3002 static int cxgb_setup_tc_block_cb(enum tc_setup_type type, void *type_data,
3003                                   void *cb_priv)
3004 {
3005         struct net_device *dev = cb_priv;
3006         struct port_info *pi = netdev2pinfo(dev);
3007         struct adapter *adap = netdev2adap(dev);
3008
3009         if (!(adap->flags & FULL_INIT_DONE)) {
3010                 dev_err(adap->pdev_dev,
3011                         "Failed to setup tc on port %d. Link Down?\n",
3012                         pi->port_id);
3013                 return -EINVAL;
3014         }
3015
3016         if (!tc_cls_can_offload_and_chain0(dev, type_data))
3017                 return -EOPNOTSUPP;
3018
3019         switch (type) {
3020         case TC_SETUP_CLSU32:
3021                 return cxgb_setup_tc_cls_u32(dev, type_data);
3022         case TC_SETUP_CLSFLOWER:
3023                 return cxgb_setup_tc_flower(dev, type_data);
3024         default:
3025                 return -EOPNOTSUPP;
3026         }
3027 }
3028
3029 static int cxgb_setup_tc_block(struct net_device *dev,
3030                                struct tc_block_offload *f)
3031 {
3032         struct port_info *pi = netdev2pinfo(dev);
3033
3034         if (f->binder_type != TCF_BLOCK_BINDER_TYPE_CLSACT_INGRESS)
3035                 return -EOPNOTSUPP;
3036
3037         switch (f->command) {
3038         case TC_BLOCK_BIND:
3039                 return tcf_block_cb_register(f->block, cxgb_setup_tc_block_cb,
3040                                              pi, dev, f->extack);
3041         case TC_BLOCK_UNBIND:
3042                 tcf_block_cb_unregister(f->block, cxgb_setup_tc_block_cb, pi);
3043                 return 0;
3044         default:
3045                 return -EOPNOTSUPP;
3046         }
3047 }
3048
3049 static int cxgb_setup_tc(struct net_device *dev, enum tc_setup_type type,
3050                          void *type_data)
3051 {
3052         switch (type) {
3053         case TC_SETUP_BLOCK:
3054                 return cxgb_setup_tc_block(dev, type_data);
3055         default:
3056                 return -EOPNOTSUPP;
3057         }
3058 }
3059
3060 static void cxgb_del_udp_tunnel(struct net_device *netdev,
3061                                 struct udp_tunnel_info *ti)
3062 {
3063         struct port_info *pi = netdev_priv(netdev);
3064         struct adapter *adapter = pi->adapter;
3065         unsigned int chip_ver = CHELSIO_CHIP_VERSION(adapter->params.chip);
3066         u8 match_all_mac[] = { 0, 0, 0, 0, 0, 0 };
3067         int ret = 0, i;
3068
3069         if (chip_ver < CHELSIO_T6)
3070                 return;
3071
3072         switch (ti->type) {
3073         case UDP_TUNNEL_TYPE_VXLAN:
3074                 if (!adapter->vxlan_port_cnt ||
3075                     adapter->vxlan_port != ti->port)
3076                         return; /* Invalid VxLAN destination port */
3077
3078                 adapter->vxlan_port_cnt--;
3079                 if (adapter->vxlan_port_cnt)
3080                         return;
3081
3082                 adapter->vxlan_port = 0;
3083                 t4_write_reg(adapter, MPS_RX_VXLAN_TYPE_A, 0);
3084                 break;
3085         case UDP_TUNNEL_TYPE_GENEVE:
3086                 if (!adapter->geneve_port_cnt ||
3087                     adapter->geneve_port != ti->port)
3088                         return; /* Invalid GENEVE destination port */
3089
3090                 adapter->geneve_port_cnt--;
3091                 if (adapter->geneve_port_cnt)
3092                         return;
3093
3094                 adapter->geneve_port = 0;
3095                 t4_write_reg(adapter, MPS_RX_GENEVE_TYPE_A, 0);
3096                 break;
3097         default:
3098                 return;
3099         }
3100
3101         /* Matchall mac entries can be deleted only after all tunnel ports
3102          * are brought down or removed.
3103          */
3104         if (!adapter->rawf_cnt)
3105                 return;
3106         for_each_port(adapter, i) {
3107                 pi = adap2pinfo(adapter, i);
3108                 ret = t4_free_raw_mac_filt(adapter, pi->viid,
3109                                            match_all_mac, match_all_mac,
3110                                            adapter->rawf_start +
3111                                             pi->port_id,
3112                                            1, pi->port_id, false);
3113                 if (ret < 0) {
3114                         netdev_info(netdev, "Failed to free mac filter entry, for port %d\n",
3115                                     i);
3116                         return;
3117                 }
3118                 atomic_dec(&adapter->mps_encap[adapter->rawf_start +
3119                            pi->port_id].refcnt);
3120         }
3121 }
3122
3123 static void cxgb_add_udp_tunnel(struct net_device *netdev,
3124                                 struct udp_tunnel_info *ti)
3125 {
3126         struct port_info *pi = netdev_priv(netdev);
3127         struct adapter *adapter = pi->adapter;
3128         unsigned int chip_ver = CHELSIO_CHIP_VERSION(adapter->params.chip);
3129         u8 match_all_mac[] = { 0, 0, 0, 0, 0, 0 };
3130         int i, ret;
3131
3132         if (chip_ver < CHELSIO_T6 || !adapter->rawf_cnt)
3133                 return;
3134
3135         switch (ti->type) {
3136         case UDP_TUNNEL_TYPE_VXLAN:
3137                 /* Callback for adding vxlan port can be called with the same
3138                  * port for both IPv4 and IPv6. We should not disable the
3139                  * offloading when the same port for both protocols is added
3140                  * and later one of them is removed.
3141                  */
3142                 if (adapter->vxlan_port_cnt &&
3143                     adapter->vxlan_port == ti->port) {
3144                         adapter->vxlan_port_cnt++;
3145                         return;
3146                 }
3147
3148                 /* We will support only one VxLAN port */
3149                 if (adapter->vxlan_port_cnt) {
3150                         netdev_info(netdev, "UDP port %d already offloaded, not adding port %d\n",
3151                                     be16_to_cpu(adapter->vxlan_port),
3152                                     be16_to_cpu(ti->port));
3153                         return;
3154                 }
3155
3156                 adapter->vxlan_port = ti->port;
3157                 adapter->vxlan_port_cnt = 1;
3158
3159                 t4_write_reg(adapter, MPS_RX_VXLAN_TYPE_A,
3160                              VXLAN_V(be16_to_cpu(ti->port)) | VXLAN_EN_F);
3161                 break;
3162         case UDP_TUNNEL_TYPE_GENEVE:
3163                 if (adapter->geneve_port_cnt &&
3164                     adapter->geneve_port == ti->port) {
3165                         adapter->geneve_port_cnt++;
3166                         return;
3167                 }
3168
3169                 /* We will support only one GENEVE port */
3170                 if (adapter->geneve_port_cnt) {
3171                         netdev_info(netdev, "UDP port %d already offloaded, not adding port %d\n",
3172                                     be16_to_cpu(adapter->geneve_port),
3173                                     be16_to_cpu(ti->port));
3174                         return;
3175                 }
3176
3177                 adapter->geneve_port = ti->port;
3178                 adapter->geneve_port_cnt = 1;
3179
3180                 t4_write_reg(adapter, MPS_RX_GENEVE_TYPE_A,
3181                              GENEVE_V(be16_to_cpu(ti->port)) | GENEVE_EN_F);
3182                 break;
3183         default:
3184                 return;
3185         }
3186
3187         /* Create a 'match all' mac filter entry for inner mac,
3188          * if raw mac interface is supported. Once the linux kernel provides
3189          * driver entry points for adding/deleting the inner mac addresses,
3190          * we will remove this 'match all' entry and fallback to adding
3191          * exact match filters.
3192          */
3193         for_each_port(adapter, i) {
3194                 pi = adap2pinfo(adapter, i);
3195
3196                 ret = t4_alloc_raw_mac_filt(adapter, pi->viid,
3197                                             match_all_mac,
3198                                             match_all_mac,
3199                                             adapter->rawf_start +
3200                                             pi->port_id,
3201                                             1, pi->port_id, false);
3202                 if (ret < 0) {
3203                         netdev_info(netdev, "Failed to allocate a mac filter entry, not adding port %d\n",
3204                                     be16_to_cpu(ti->port));
3205                         cxgb_del_udp_tunnel(netdev, ti);
3206                         return;
3207                 }
3208                 atomic_inc(&adapter->mps_encap[ret].refcnt);
3209         }
3210 }
3211
3212 static netdev_features_t cxgb_features_check(struct sk_buff *skb,
3213                                              struct net_device *dev,
3214                                              netdev_features_t features)
3215 {
3216         struct port_info *pi = netdev_priv(dev);
3217         struct adapter *adapter = pi->adapter;
3218
3219         if (CHELSIO_CHIP_VERSION(adapter->params.chip) < CHELSIO_T6)
3220                 return features;
3221
3222         /* Check if hw supports offload for this packet */
3223         if (!skb->encapsulation || cxgb_encap_offload_supported(skb))
3224                 return features;
3225
3226         /* Offload is not supported for this encapsulated packet */
3227         return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3228 }
3229
3230 static netdev_features_t cxgb_fix_features(struct net_device *dev,
3231                                            netdev_features_t features)
3232 {
3233         /* Disable GRO, if RX_CSUM is disabled */
3234         if (!(features & NETIF_F_RXCSUM))
3235                 features &= ~NETIF_F_GRO;
3236
3237         return features;
3238 }
3239
3240 static const struct net_device_ops cxgb4_netdev_ops = {
3241         .ndo_open             = cxgb_open,
3242         .ndo_stop             = cxgb_close,
3243         .ndo_start_xmit       = t4_start_xmit,
3244         .ndo_select_queue     = cxgb_select_queue,
3245         .ndo_get_stats64      = cxgb_get_stats,
3246         .ndo_set_rx_mode      = cxgb_set_rxmode,
3247         .ndo_set_mac_address  = cxgb_set_mac_addr,
3248         .ndo_set_features     = cxgb_set_features,
3249         .ndo_validate_addr    = eth_validate_addr,
3250         .ndo_do_ioctl         = cxgb_ioctl,
3251         .ndo_change_mtu       = cxgb_change_mtu,
3252 #ifdef CONFIG_NET_POLL_CONTROLLER
3253         .ndo_poll_controller  = cxgb_netpoll,
3254 #endif
3255 #ifdef CONFIG_CHELSIO_T4_FCOE
3256         .ndo_fcoe_enable      = cxgb_fcoe_enable,
3257         .ndo_fcoe_disable     = cxgb_fcoe_disable,
3258 #endif /* CONFIG_CHELSIO_T4_FCOE */
3259         .ndo_set_tx_maxrate   = cxgb_set_tx_maxrate,
3260         .ndo_setup_tc         = cxgb_setup_tc,
3261         .ndo_udp_tunnel_add   = cxgb_add_udp_tunnel,
3262         .ndo_udp_tunnel_del   = cxgb_del_udp_tunnel,
3263         .ndo_features_check   = cxgb_features_check,
3264         .ndo_fix_features     = cxgb_fix_features,
3265 };
3266
3267 #ifdef CONFIG_PCI_IOV
3268 static const struct net_device_ops cxgb4_mgmt_netdev_ops = {
3269         .ndo_open             = cxgb4_mgmt_open,
3270         .ndo_set_vf_mac       = cxgb4_mgmt_set_vf_mac,
3271         .ndo_get_vf_config    = cxgb4_mgmt_get_vf_config,
3272         .ndo_set_vf_rate      = cxgb4_mgmt_set_vf_rate,
3273         .ndo_get_phys_port_id = cxgb4_mgmt_get_phys_port_id,
3274         .ndo_set_vf_vlan      = cxgb4_mgmt_set_vf_vlan,
3275 };
3276 #endif
3277
3278 static void cxgb4_mgmt_get_drvinfo(struct net_device *dev,
3279                                    struct ethtool_drvinfo *info)
3280 {
3281         struct adapter *adapter = netdev2adap(dev);
3282
3283         strlcpy(info->driver, cxgb4_driver_name, sizeof(info->driver));
3284         strlcpy(info->version, cxgb4_driver_version,
3285                 sizeof(info->version));
3286         strlcpy(info->bus_info, pci_name(adapter->pdev),
3287                 sizeof(info->bus_info));
3288 }
3289
3290 static const struct ethtool_ops cxgb4_mgmt_ethtool_ops = {
3291         .get_drvinfo       = cxgb4_mgmt_get_drvinfo,
3292 };
3293
3294 static void notify_fatal_err(struct work_struct *work)
3295 {
3296         struct adapter *adap;
3297
3298         adap = container_of(work, struct adapter, fatal_err_notify_task);
3299         notify_ulds(adap, CXGB4_STATE_FATAL_ERROR);
3300 }
3301
3302 void t4_fatal_err(struct adapter *adap)
3303 {
3304         int port;
3305
3306         if (pci_channel_offline(adap->pdev))
3307                 return;
3308
3309         /* Disable the SGE since ULDs are going to free resources that
3310          * could be exposed to the adapter.  RDMA MWs for example...
3311          */
3312         t4_shutdown_adapter(adap);
3313         for_each_port(adap, port) {
3314                 struct net_device *dev = adap->port[port];
3315
3316                 /* If we get here in very early initialization the network
3317                  * devices may not have been set up yet.
3318                  */
3319                 if (!dev)
3320                         continue;
3321
3322                 netif_tx_stop_all_queues(dev);
3323                 netif_carrier_off(dev);
3324         }
3325         dev_alert(adap->pdev_dev, "encountered fatal error, adapter stopped\n");
3326         queue_work(adap->workq, &adap->fatal_err_notify_task);
3327 }
3328
3329 static void setup_memwin(struct adapter *adap)
3330 {
3331         u32 nic_win_base = t4_get_util_window(adap);
3332
3333         t4_setup_memwin(adap, nic_win_base, MEMWIN_NIC);
3334 }
3335
3336 static void setup_memwin_rdma(struct adapter *adap)
3337 {
3338         if (adap->vres.ocq.size) {
3339                 u32 start;
3340                 unsigned int sz_kb;
3341
3342                 start = t4_read_pcie_cfg4(adap, PCI_BASE_ADDRESS_2);
3343                 start &= PCI_BASE_ADDRESS_MEM_MASK;
3344                 start += OCQ_WIN_OFFSET(adap->pdev, &adap->vres);
3345                 sz_kb = roundup_pow_of_two(adap->vres.ocq.size) >> 10;
3346                 t4_write_reg(adap,
3347                              PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN_A, 3),
3348                              start | BIR_V(1) | WINDOW_V(ilog2(sz_kb)));
3349                 t4_write_reg(adap,
3350                              PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_OFFSET_A, 3),
3351                              adap->vres.ocq.start);
3352                 t4_read_reg(adap,
3353                             PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_OFFSET_A, 3));
3354         }
3355 }
3356
3357 /* HMA Definitions */
3358
3359 /* The maximum number of address that can be send in a single FW cmd */
3360 #define HMA_MAX_ADDR_IN_CMD     5
3361
3362 #define HMA_PAGE_SIZE           PAGE_SIZE
3363
3364 #define HMA_MAX_NO_FW_ADDRESS   (16 << 10)  /* FW supports 16K addresses */
3365
3366 #define HMA_PAGE_ORDER                                  \
3367         ((HMA_PAGE_SIZE < HMA_MAX_NO_FW_ADDRESS) ?      \
3368         ilog2(HMA_MAX_NO_FW_ADDRESS / HMA_PAGE_SIZE) : 0)
3369
3370 /* The minimum and maximum possible HMA sizes that can be specified in the FW
3371  * configuration(in units of MB).
3372  */
3373 #define HMA_MIN_TOTAL_SIZE      1
3374 #define HMA_MAX_TOTAL_SIZE                              \
3375         (((HMA_PAGE_SIZE << HMA_PAGE_ORDER) *           \
3376           HMA_MAX_NO_FW_ADDRESS) >> 20)
3377
3378 static void adap_free_hma_mem(struct adapter *adapter)
3379 {
3380         struct scatterlist *iter;
3381         struct page *page;
3382         int i;
3383
3384         if (!adapter->hma.sgt)
3385                 return;
3386
3387         if (adapter->hma.flags & HMA_DMA_MAPPED_FLAG) {
3388                 dma_unmap_sg(adapter->pdev_dev, adapter->hma.sgt->sgl,
3389                              adapter->hma.sgt->nents, PCI_DMA_BIDIRECTIONAL);
3390                 adapter->hma.flags &= ~HMA_DMA_MAPPED_FLAG;
3391         }
3392
3393         for_each_sg(adapter->hma.sgt->sgl, iter,
3394                     adapter->hma.sgt->orig_nents, i) {
3395                 page = sg_page(iter);
3396                 if (page)
3397                         __free_pages(page, HMA_PAGE_ORDER);
3398         }
3399
3400         kfree(adapter->hma.phy_addr);
3401         sg_free_table(adapter->hma.sgt);
3402         kfree(adapter->hma.sgt);
3403         adapter->hma.sgt = NULL;
3404 }
3405
3406 static int adap_config_hma(struct adapter *adapter)
3407 {
3408         struct scatterlist *sgl, *iter;
3409         struct sg_table *sgt;
3410         struct page *newpage;
3411         unsigned int i, j, k;
3412         u32 param, hma_size;
3413         unsigned int ncmds;
3414         size_t page_size;
3415         u32 page_order;
3416         int node, ret;
3417
3418         /* HMA is supported only for T6+ cards.
3419          * Avoid initializing HMA in kdump kernels.
3420          */
3421         if (is_kdump_kernel() ||
3422             CHELSIO_CHIP_VERSION(adapter->params.chip) < CHELSIO_T6)
3423                 return 0;
3424
3425         /* Get the HMA region size required by fw */
3426         param = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) |
3427                  FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_HMA_SIZE));
3428         ret = t4_query_params(adapter, adapter->mbox, adapter->pf, 0,
3429                               1, &param, &hma_size);
3430         /* An error means card has its own memory or HMA is not supported by
3431          * the firmware. Return without any errors.
3432          */
3433         if (ret || !hma_size)
3434                 return 0;
3435
3436         if (hma_size < HMA_MIN_TOTAL_SIZE ||
3437             hma_size > HMA_MAX_TOTAL_SIZE) {
3438                 dev_err(adapter->pdev_dev,
3439                         "HMA size %uMB beyond bounds(%u-%lu)MB\n",
3440                         hma_size, HMA_MIN_TOTAL_SIZE, HMA_MAX_TOTAL_SIZE);
3441                 return -EINVAL;
3442         }
3443
3444         page_size = HMA_PAGE_SIZE;
3445         page_order = HMA_PAGE_ORDER;
3446         adapter->hma.sgt = kzalloc(sizeof(*adapter->hma.sgt), GFP_KERNEL);
3447         if (unlikely(!adapter->hma.sgt)) {
3448                 dev_err(adapter->pdev_dev, "HMA SG table allocation failed\n");
3449                 return -ENOMEM;
3450         }
3451         sgt = adapter->hma.sgt;
3452         /* FW returned value will be in MB's
3453          */
3454         sgt->orig_nents = (hma_size << 20) / (page_size << page_order);
3455         if (sg_alloc_table(sgt, sgt->orig_nents, GFP_KERNEL)) {
3456                 dev_err(adapter->pdev_dev, "HMA SGL allocation failed\n");
3457                 kfree(adapter->hma.sgt);
3458                 adapter->hma.sgt = NULL;
3459                 return -ENOMEM;
3460         }
3461
3462         sgl = adapter->hma.sgt->sgl;
3463         node = dev_to_node(adapter->pdev_dev);
3464         for_each_sg(sgl, iter, sgt->orig_nents, i) {
3465                 newpage = alloc_pages_node(node, __GFP_NOWARN | GFP_KERNEL |
3466                                            __GFP_ZERO, page_order);
3467                 if (!newpage) {
3468                         dev_err(adapter->pdev_dev,
3469                                 "Not enough memory for HMA page allocation\n");
3470                         ret = -ENOMEM;
3471                         goto free_hma;
3472                 }
3473                 sg_set_page(iter, newpage, page_size << page_order, 0);
3474         }
3475
3476         sgt->nents = dma_map_sg(adapter->pdev_dev, sgl, sgt->orig_nents,
3477                                 DMA_BIDIRECTIONAL);
3478         if (!sgt->nents) {
3479                 dev_err(adapter->pdev_dev,
3480                         "Not enough memory for HMA DMA mapping");
3481                 ret = -ENOMEM;
3482                 goto free_hma;
3483         }
3484         adapter->hma.flags |= HMA_DMA_MAPPED_FLAG;
3485
3486         adapter->hma.phy_addr = kcalloc(sgt->nents, sizeof(dma_addr_t),
3487                                         GFP_KERNEL);
3488         if (unlikely(!adapter->hma.phy_addr))
3489                 goto free_hma;
3490
3491         for_each_sg(sgl, iter, sgt->nents, i) {
3492                 newpage = sg_page(iter);
3493                 adapter->hma.phy_addr[i] = sg_dma_address(iter);
3494         }
3495
3496         ncmds = DIV_ROUND_UP(sgt->nents, HMA_MAX_ADDR_IN_CMD);
3497         /* Pass on the addresses to firmware */
3498         for (i = 0, k = 0; i < ncmds; i++, k += HMA_MAX_ADDR_IN_CMD) {
3499                 struct fw_hma_cmd hma_cmd;
3500                 u8 naddr = HMA_MAX_ADDR_IN_CMD;
3501                 u8 soc = 0, eoc = 0;
3502                 u8 hma_mode = 1; /* Presently we support only Page table mode */
3503
3504                 soc = (i == 0) ? 1 : 0;
3505                 eoc = (i == ncmds - 1) ? 1 : 0;
3506
3507                 /* For last cmd, set naddr corresponding to remaining
3508                  * addresses
3509                  */
3510                 if (i == ncmds - 1) {
3511                         naddr = sgt->nents % HMA_MAX_ADDR_IN_CMD;
3512                         naddr = naddr ? naddr : HMA_MAX_ADDR_IN_CMD;
3513                 }
3514                 memset(&hma_cmd, 0, sizeof(hma_cmd));
3515                 hma_cmd.op_pkd = htonl(FW_CMD_OP_V(FW_HMA_CMD) |
3516                                        FW_CMD_REQUEST_F | FW_CMD_WRITE_F);
3517                 hma_cmd.retval_len16 = htonl(FW_LEN16(hma_cmd));
3518
3519                 hma_cmd.mode_to_pcie_params =
3520                         htonl(FW_HMA_CMD_MODE_V(hma_mode) |
3521                               FW_HMA_CMD_SOC_V(soc) | FW_HMA_CMD_EOC_V(eoc));
3522
3523                 /* HMA cmd size specified in MB's */
3524                 hma_cmd.naddr_size =
3525                         htonl(FW_HMA_CMD_SIZE_V(hma_size) |
3526                               FW_HMA_CMD_NADDR_V(naddr));
3527
3528                 /* Total Page size specified in units of 4K */
3529                 hma_cmd.addr_size_pkd =
3530                         htonl(FW_HMA_CMD_ADDR_SIZE_V
3531                                 ((page_size << page_order) >> 12));
3532
3533                 /* Fill the 5 addresses */
3534                 for (j = 0; j < naddr; j++) {
3535                         hma_cmd.phy_address[j] =
3536                                 cpu_to_be64(adapter->hma.phy_addr[j + k]);
3537                 }
3538                 ret = t4_wr_mbox(adapter, adapter->mbox, &hma_cmd,
3539                                  sizeof(hma_cmd), &hma_cmd);
3540                 if (ret) {
3541                         dev_err(adapter->pdev_dev,
3542                                 "HMA FW command failed with err %d\n", ret);
3543                         goto free_hma;
3544                 }
3545         }
3546
3547         if (!ret)
3548                 dev_info(adapter->pdev_dev,
3549                          "Reserved %uMB host memory for HMA\n", hma_size);
3550         return ret;
3551
3552 free_hma:
3553         adap_free_hma_mem(adapter);
3554         return ret;
3555 }
3556
3557 static int adap_init1(struct adapter *adap, struct fw_caps_config_cmd *c)
3558 {
3559         u32 v;
3560         int ret;
3561
3562         /* Now that we've successfully configured and initialized the adapter
3563          * can ask the Firmware what resources it has provisioned for us.
3564          */
3565         ret = t4_get_pfres(adap);
3566         if (ret) {
3567                 dev_err(adap->pdev_dev,
3568                         "Unable to retrieve resource provisioning information\n");
3569                 return ret;
3570         }
3571
3572         /* get device capabilities */
3573         memset(c, 0, sizeof(*c));
3574         c->op_to_write = htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
3575                                FW_CMD_REQUEST_F | FW_CMD_READ_F);
3576         c->cfvalid_to_len16 = htonl(FW_LEN16(*c));
3577         ret = t4_wr_mbox(adap, adap->mbox, c, sizeof(*c), c);
3578         if (ret < 0)
3579                 return ret;
3580
3581         c->op_to_write = htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
3582                                FW_CMD_REQUEST_F | FW_CMD_WRITE_F);
3583         ret = t4_wr_mbox(adap, adap->mbox, c, sizeof(*c), NULL);
3584         if (ret < 0)
3585                 return ret;
3586
3587         ret = t4_config_glbl_rss(adap, adap->pf,
3588                                  FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL,
3589                                  FW_RSS_GLB_CONFIG_CMD_TNLMAPEN_F |
3590                                  FW_RSS_GLB_CONFIG_CMD_TNLALLLKP_F);
3591         if (ret < 0)
3592                 return ret;
3593
3594         ret = t4_cfg_pfvf(adap, adap->mbox, adap->pf, 0, adap->sge.egr_sz, 64,
3595                           MAX_INGQ, 0, 0, 4, 0xf, 0xf, 16, FW_CMD_CAP_PF,
3596                           FW_CMD_CAP_PF);
3597         if (ret < 0)
3598                 return ret;
3599
3600         t4_sge_init(adap);
3601
3602         /* tweak some settings */
3603         t4_write_reg(adap, TP_SHIFT_CNT_A, 0x64f8849);
3604         t4_write_reg(adap, ULP_RX_TDDP_PSZ_A, HPZ0_V(PAGE_SHIFT - 12));
3605         t4_write_reg(adap, TP_PIO_ADDR_A, TP_INGRESS_CONFIG_A);
3606         v = t4_read_reg(adap, TP_PIO_DATA_A);
3607         t4_write_reg(adap, TP_PIO_DATA_A, v & ~CSUM_HAS_PSEUDO_HDR_F);
3608
3609         /* first 4 Tx modulation queues point to consecutive Tx channels */
3610         adap->params.tp.tx_modq_map = 0xE4;
3611         t4_write_reg(adap, TP_TX_MOD_QUEUE_REQ_MAP_A,
3612                      TX_MOD_QUEUE_REQ_MAP_V(adap->params.tp.tx_modq_map));
3613
3614         /* associate each Tx modulation queue with consecutive Tx channels */
3615         v = 0x84218421;
3616         t4_write_indirect(adap, TP_PIO_ADDR_A, TP_PIO_DATA_A,
3617                           &v, 1, TP_TX_SCHED_HDR_A);
3618         t4_write_indirect(adap, TP_PIO_ADDR_A, TP_PIO_DATA_A,
3619                           &v, 1, TP_TX_SCHED_FIFO_A);
3620         t4_write_indirect(adap, TP_PIO_ADDR_A, TP_PIO_DATA_A,
3621                           &v, 1, TP_TX_SCHED_PCMD_A);
3622
3623 #define T4_TX_MODQ_10G_WEIGHT_DEFAULT 16 /* in KB units */
3624         if (is_offload(adap)) {
3625                 t4_write_reg(adap, TP_TX_MOD_QUEUE_WEIGHT0_A,
3626                              TX_MODQ_WEIGHT0_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
3627                              TX_MODQ_WEIGHT1_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
3628                              TX_MODQ_WEIGHT2_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
3629                              TX_MODQ_WEIGHT3_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT));
3630                 t4_write_reg(adap, TP_TX_MOD_CHANNEL_WEIGHT_A,
3631                              TX_MODQ_WEIGHT0_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
3632                              TX_MODQ_WEIGHT1_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
3633                              TX_MODQ_WEIGHT2_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
3634                              TX_MODQ_WEIGHT3_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT));
3635         }
3636
3637         /* get basic stuff going */
3638         return t4_early_init(adap, adap->pf);
3639 }
3640
3641 /*
3642  * Max # of ATIDs.  The absolute HW max is 16K but we keep it lower.
3643  */
3644 #define MAX_ATIDS 8192U
3645
3646 /*
3647  * Phase 0 of initialization: contact FW, obtain config, perform basic init.
3648  *
3649  * If the firmware we're dealing with has Configuration File support, then
3650  * we use that to perform all configuration
3651  */
3652
3653 /*
3654  * Tweak configuration based on module parameters, etc.  Most of these have
3655  * defaults assigned to them by Firmware Configuration Files (if we're using
3656  * them) but need to be explicitly set if we're using hard-coded
3657  * initialization.  But even in the case of using Firmware Configuration
3658  * Files, we'd like to expose the ability to change these via module
3659  * parameters so these are essentially common tweaks/settings for
3660  * Configuration Files and hard-coded initialization ...
3661  */
3662 static int adap_init0_tweaks(struct adapter *adapter)
3663 {
3664         /*
3665          * Fix up various Host-Dependent Parameters like Page Size, Cache
3666          * Line Size, etc.  The firmware default is for a 4KB Page Size and
3667          * 64B Cache Line Size ...
3668          */
3669         t4_fixup_host_params(adapter, PAGE_SIZE, L1_CACHE_BYTES);
3670
3671         /*
3672          * Process module parameters which affect early initialization.
3673          */
3674         if (rx_dma_offset != 2 && rx_dma_offset != 0) {
3675                 dev_err(&adapter->pdev->dev,
3676                         "Ignoring illegal rx_dma_offset=%d, using 2\n",
3677                         rx_dma_offset);
3678                 rx_dma_offset = 2;
3679         }
3680         t4_set_reg_field(adapter, SGE_CONTROL_A,
3681                          PKTSHIFT_V(PKTSHIFT_M),
3682                          PKTSHIFT_V(rx_dma_offset));
3683
3684         /*
3685          * Don't include the "IP Pseudo Header" in CPL_RX_PKT checksums: Linux
3686          * adds the pseudo header itself.
3687          */
3688         t4_tp_wr_bits_indirect(adapter, TP_INGRESS_CONFIG_A,
3689                                CSUM_HAS_PSEUDO_HDR_F, 0);
3690
3691         return 0;
3692 }
3693
3694 /* 10Gb/s-BT PHY Support. chip-external 10Gb/s-BT PHYs are complex chips
3695  * unto themselves and they contain their own firmware to perform their
3696  * tasks ...
3697  */
3698 static int phy_aq1202_version(const u8 *phy_fw_data,
3699                               size_t phy_fw_size)
3700 {
3701         int offset;
3702
3703         /* At offset 0x8 you're looking for the primary image's
3704          * starting offset which is 3 Bytes wide
3705          *
3706          * At offset 0xa of the primary image, you look for the offset
3707          * of the DRAM segment which is 3 Bytes wide.
3708          *
3709          * The FW version is at offset 0x27e of the DRAM and is 2 Bytes
3710          * wide
3711          */
3712         #define be16(__p) (((__p)[0] << 8) | (__p)[1])
3713         #define le16(__p) ((__p)[0] | ((__p)[1] << 8))
3714         #define le24(__p) (le16(__p) | ((__p)[2] << 16))
3715
3716         offset = le24(phy_fw_data + 0x8) << 12;
3717         offset = le24(phy_fw_data + offset + 0xa);
3718         return be16(phy_fw_data + offset + 0x27e);
3719
3720         #undef be16
3721         #undef le16
3722         #undef le24
3723 }
3724
3725 static struct info_10gbt_phy_fw {
3726         unsigned int phy_fw_id;         /* PCI Device ID */
3727         char *phy_fw_file;              /* /lib/firmware/ PHY Firmware file */
3728         int (*phy_fw_version)(const u8 *phy_fw_data, size_t phy_fw_size);
3729         int phy_flash;                  /* Has FLASH for PHY Firmware */
3730 } phy_info_array[] = {
3731         {
3732                 PHY_AQ1202_DEVICEID,
3733                 PHY_AQ1202_FIRMWARE,
3734                 phy_aq1202_version,
3735                 1,
3736         },
3737         {
3738                 PHY_BCM84834_DEVICEID,
3739                 PHY_BCM84834_FIRMWARE,
3740                 NULL,
3741                 0,
3742         },
3743         { 0, NULL, NULL },
3744 };
3745
3746 static struct info_10gbt_phy_fw *find_phy_info(int devid)
3747 {
3748         int i;
3749
3750         for (i = 0; i < ARRAY_SIZE(phy_info_array); i++) {
3751                 if (phy_info_array[i].phy_fw_id == devid)
3752                         return &phy_info_array[i];
3753         }
3754         return NULL;
3755 }
3756
3757 /* Handle updating of chip-external 10Gb/s-BT PHY firmware.  This needs to
3758  * happen after the FW_RESET_CMD but before the FW_INITIALIZE_CMD.  On error
3759  * we return a negative error number.  If we transfer new firmware we return 1
3760  * (from t4_load_phy_fw()).  If we don't do anything we return 0.
3761  */
3762 static int adap_init0_phy(struct adapter *adap)
3763 {
3764         const struct firmware *phyf;
3765         int ret;
3766         struct info_10gbt_phy_fw *phy_info;
3767
3768         /* Use the device ID to determine which PHY file to flash.
3769          */
3770         phy_info = find_phy_info(adap->pdev->device);
3771         if (!phy_info) {
3772                 dev_warn(adap->pdev_dev,
3773                          "No PHY Firmware file found for this PHY\n");
3774                 return -EOPNOTSUPP;
3775         }
3776
3777         /* If we have a T4 PHY firmware file under /lib/firmware/cxgb4/, then
3778          * use that. The adapter firmware provides us with a memory buffer
3779          * where we can load a PHY firmware file from the host if we want to
3780          * override the PHY firmware File in flash.
3781          */
3782         ret = request_firmware_direct(&phyf, phy_info->phy_fw_file,
3783                                       adap->pdev_dev);
3784         if (ret < 0) {
3785                 /* For adapters without FLASH attached to PHY for their
3786                  * firmware, it's obviously a fatal error if we can't get the
3787                  * firmware to the adapter.  For adapters with PHY firmware
3788                  * FLASH storage, it's worth a warning if we can't find the
3789                  * PHY Firmware but we'll neuter the error ...
3790                  */
3791                 dev_err(adap->pdev_dev, "unable to find PHY Firmware image "
3792                         "/lib/firmware/%s, error %d\n",
3793                         phy_info->phy_fw_file, -ret);
3794                 if (phy_info->phy_flash) {
3795                         int cur_phy_fw_ver = 0;
3796
3797                         t4_phy_fw_ver(adap, &cur_phy_fw_ver);
3798                         dev_warn(adap->pdev_dev, "continuing with, on-adapter "
3799                                  "FLASH copy, version %#x\n", cur_phy_fw_ver);
3800                         ret = 0;
3801                 }
3802
3803                 return ret;
3804         }
3805
3806         /* Load PHY Firmware onto adapter.
3807          */
3808         ret = t4_load_phy_fw(adap, MEMWIN_NIC, &adap->win0_lock,
3809                              phy_info->phy_fw_version,
3810                              (u8 *)phyf->data, phyf->size);
3811         if (ret < 0)
3812                 dev_err(adap->pdev_dev, "PHY Firmware transfer error %d\n",
3813                         -ret);
3814         else if (ret > 0) {
3815                 int new_phy_fw_ver = 0;
3816
3817                 if (phy_info->phy_fw_version)
3818                         new_phy_fw_ver = phy_info->phy_fw_version(phyf->data,
3819                                                                   phyf->size);
3820                 dev_info(adap->pdev_dev, "Successfully transferred PHY "
3821                          "Firmware /lib/firmware/%s, version %#x\n",
3822                          phy_info->phy_fw_file, new_phy_fw_ver);
3823         }
3824
3825         release_firmware(phyf);
3826
3827         return ret;
3828 }
3829
3830 /*
3831  * Attempt to initialize the adapter via a Firmware Configuration File.
3832  */
3833 static int adap_init0_config(struct adapter *adapter, int reset)
3834 {
3835         struct fw_caps_config_cmd caps_cmd;
3836         const struct firmware *cf;
3837         unsigned long mtype = 0, maddr = 0;
3838         u32 finiver, finicsum, cfcsum;
3839         int ret;
3840         int config_issued = 0;
3841         char *fw_config_file, fw_config_file_path[256];
3842         char *config_name = NULL;
3843
3844         /*
3845          * Reset device if necessary.
3846          */
3847         if (reset) {
3848                 ret = t4_fw_reset(adapter, adapter->mbox,
3849                                   PIORSTMODE_F | PIORST_F);
3850                 if (ret < 0)
3851                         goto bye;
3852         }
3853
3854         /* If this is a 10Gb/s-BT adapter make sure the chip-external
3855          * 10Gb/s-BT PHYs have up-to-date firmware.  Note that this step needs
3856          * to be performed after any global adapter RESET above since some
3857          * PHYs only have local RAM copies of the PHY firmware.
3858          */
3859         if (is_10gbt_device(adapter->pdev->device)) {
3860                 ret = adap_init0_phy(adapter);
3861                 if (ret < 0)
3862                         goto bye;
3863         }
3864         /*
3865          * If we have a T4 configuration file under /lib/firmware/cxgb4/,
3866          * then use that.  Otherwise, use the configuration file stored
3867          * in the adapter flash ...
3868          */
3869         switch (CHELSIO_CHIP_VERSION(adapter->params.chip)) {
3870         case CHELSIO_T4:
3871                 fw_config_file = FW4_CFNAME;
3872                 break;
3873         case CHELSIO_T5:
3874                 fw_config_file = FW5_CFNAME;
3875                 break;
3876         case CHELSIO_T6:
3877                 fw_config_file = FW6_CFNAME;
3878                 break;
3879         default:
3880                 dev_err(adapter->pdev_dev, "Device %d is not supported\n",
3881                        adapter->pdev->device);
3882                 ret = -EINVAL;
3883                 goto bye;
3884         }
3885
3886         ret = request_firmware(&cf, fw_config_file, adapter->pdev_dev);
3887         if (ret < 0) {
3888                 config_name = "On FLASH";
3889                 mtype = FW_MEMTYPE_CF_FLASH;
3890                 maddr = t4_flash_cfg_addr(adapter);
3891         } else {
3892                 u32 params[7], val[7];
3893
3894                 sprintf(fw_config_file_path,
3895                         "/lib/firmware/%s", fw_config_file);
3896                 config_name = fw_config_file_path;
3897
3898                 if (cf->size >= FLASH_CFG_MAX_SIZE)
3899                         ret = -ENOMEM;
3900                 else {
3901                         params[0] = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) |
3902                              FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_CF));
3903                         ret = t4_query_params(adapter, adapter->mbox,
3904                                               adapter->pf, 0, 1, params, val);
3905                         if (ret == 0) {
3906                                 /*
3907                                  * For t4_memory_rw() below addresses and
3908                                  * sizes have to be in terms of multiples of 4
3909                                  * bytes.  So, if the Configuration File isn't
3910                                  * a multiple of 4 bytes in length we'll have
3911                                  * to write that out separately since we can't
3912                                  * guarantee that the bytes following the
3913                                  * residual byte in the buffer returned by
3914                                  * request_firmware() are zeroed out ...
3915                                  */
3916                                 size_t resid = cf->size & 0x3;
3917                                 size_t size = cf->size & ~0x3;
3918                                 __be32 *data = (__be32 *)cf->data;
3919
3920                                 mtype = FW_PARAMS_PARAM_Y_G(val[0]);
3921                                 maddr = FW_PARAMS_PARAM_Z_G(val[0]) << 16;
3922
3923                                 spin_lock(&adapter->win0_lock);
3924                                 ret = t4_memory_rw(adapter, 0, mtype, maddr,
3925                                                    size, data, T4_MEMORY_WRITE);
3926                                 if (ret == 0 && resid != 0) {
3927                                         union {
3928                                                 __be32 word;
3929                                                 char buf[4];
3930                                         } last;
3931                                         int i;
3932
3933                                         last.word = data[size >> 2];
3934                                         for (i = resid; i < 4; i++)
3935                                                 last.buf[i] = 0;
3936                                         ret = t4_memory_rw(adapter, 0, mtype,
3937                                                            maddr + size,
3938                                                            4, &last.word,
3939                                                            T4_MEMORY_WRITE);
3940                                 }
3941                                 spin_unlock(&adapter->win0_lock);
3942                         }
3943                 }
3944
3945                 release_firmware(cf);
3946                 if (ret)
3947                         goto bye;
3948         }
3949
3950         /*
3951          * Issue a Capability Configuration command to the firmware to get it
3952          * to parse the Configuration File.  We don't use t4_fw_config_file()
3953          * because we want the ability to modify various features after we've
3954          * processed the configuration file ...
3955          */
3956         memset(&caps_cmd, 0, sizeof(caps_cmd));
3957         caps_cmd.op_to_write =
3958                 htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
3959                       FW_CMD_REQUEST_F |
3960                       FW_CMD_READ_F);
3961         caps_cmd.cfvalid_to_len16 =
3962                 htonl(FW_CAPS_CONFIG_CMD_CFVALID_F |
3963                       FW_CAPS_CONFIG_CMD_MEMTYPE_CF_V(mtype) |
3964                       FW_CAPS_CONFIG_CMD_MEMADDR64K_CF_V(maddr >> 16) |
3965                       FW_LEN16(caps_cmd));
3966         ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
3967                          &caps_cmd);
3968
3969         /* If the CAPS_CONFIG failed with an ENOENT (for a Firmware
3970          * Configuration File in FLASH), our last gasp effort is to use the
3971          * Firmware Configuration File which is embedded in the firmware.  A
3972          * very few early versions of the firmware didn't have one embedded
3973          * but we can ignore those.
3974          */
3975         if (ret == -ENOENT) {
3976                 memset(&caps_cmd, 0, sizeof(caps_cmd));
3977                 caps_cmd.op_to_write =
3978                         htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
3979                                         FW_CMD_REQUEST_F |
3980                                         FW_CMD_READ_F);
3981                 caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
3982                 ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd,
3983                                 sizeof(caps_cmd), &caps_cmd);
3984                 config_name = "Firmware Default";
3985         }
3986
3987         config_issued = 1;
3988         if (ret < 0)
3989                 goto bye;
3990
3991         finiver = ntohl(caps_cmd.finiver);
3992         finicsum = ntohl(caps_cmd.finicsum);
3993         cfcsum = ntohl(caps_cmd.cfcsum);
3994         if (finicsum != cfcsum)
3995                 dev_warn(adapter->pdev_dev, "Configuration File checksum "\
3996                          "mismatch: [fini] csum=%#x, computed csum=%#x\n",
3997                          finicsum, cfcsum);
3998
3999         /*
4000          * And now tell the firmware to use the configuration we just loaded.
4001          */
4002         caps_cmd.op_to_write =
4003                 htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
4004                       FW_CMD_REQUEST_F |
4005                       FW_CMD_WRITE_F);
4006         caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
4007         ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
4008                          NULL);
4009         if (ret < 0)
4010                 goto bye;
4011
4012         /*
4013          * Tweak configuration based on system architecture, module
4014          * parameters, etc.
4015          */
4016         ret = adap_init0_tweaks(adapter);
4017         if (ret < 0)
4018                 goto bye;
4019
4020         /* We will proceed even if HMA init fails. */
4021         ret = adap_config_hma(adapter);
4022         if (ret)
4023                 dev_err(adapter->pdev_dev,
4024                         "HMA configuration failed with error %d\n", ret);
4025
4026         /*
4027          * And finally tell the firmware to initialize itself using the
4028          * parameters from the Configuration File.
4029          */
4030         ret = t4_fw_initialize(adapter, adapter->mbox);
4031         if (ret < 0)
4032                 goto bye;
4033
4034         /* Emit Firmware Configuration File information and return
4035          * successfully.
4036          */
4037         dev_info(adapter->pdev_dev, "Successfully configured using Firmware "\
4038                  "Configuration File \"%s\", version %#x, computed checksum %#x\n",
4039                  config_name, finiver, cfcsum);
4040         return 0;
4041
4042         /*
4043          * Something bad happened.  Return the error ...  (If the "error"
4044          * is that there's no Configuration File on the adapter we don't
4045          * want to issue a warning since this is fairly common.)
4046          */
4047 bye:
4048         if (config_issued && ret != -ENOENT)
4049                 dev_warn(adapter->pdev_dev, "\"%s\" configuration file error %d\n",
4050                          config_name, -ret);
4051         return ret;
4052 }
4053
4054 static struct fw_info fw_info_array[] = {
4055         {
4056                 .chip = CHELSIO_T4,
4057                 .fs_name = FW4_CFNAME,
4058                 .fw_mod_name = FW4_FNAME,
4059                 .fw_hdr = {
4060                         .chip = FW_HDR_CHIP_T4,
4061                         .fw_ver = __cpu_to_be32(FW_VERSION(T4)),
4062                         .intfver_nic = FW_INTFVER(T4, NIC),
4063                         .intfver_vnic = FW_INTFVER(T4, VNIC),
4064                         .intfver_ri = FW_INTFVER(T4, RI),
4065                         .intfver_iscsi = FW_INTFVER(T4, ISCSI),
4066                         .intfver_fcoe = FW_INTFVER(T4, FCOE),
4067                 },
4068         }, {
4069                 .chip = CHELSIO_T5,
4070                 .fs_name = FW5_CFNAME,
4071                 .fw_mod_name = FW5_FNAME,
4072                 .fw_hdr = {
4073                         .chip = FW_HDR_CHIP_T5,
4074                         .fw_ver = __cpu_to_be32(FW_VERSION(T5)),
4075                         .intfver_nic = FW_INTFVER(T5, NIC),
4076                         .intfver_vnic = FW_INTFVER(T5, VNIC),
4077                         .intfver_ri = FW_INTFVER(T5, RI),
4078                         .intfver_iscsi = FW_INTFVER(T5, ISCSI),
4079                         .intfver_fcoe = FW_INTFVER(T5, FCOE),
4080                 },
4081         }, {
4082                 .chip = CHELSIO_T6,
4083                 .fs_name = FW6_CFNAME,
4084                 .fw_mod_name = FW6_FNAME,
4085                 .fw_hdr = {
4086                         .chip = FW_HDR_CHIP_T6,
4087                         .fw_ver = __cpu_to_be32(FW_VERSION(T6)),
4088                         .intfver_nic = FW_INTFVER(T6, NIC),
4089                         .intfver_vnic = FW_INTFVER(T6, VNIC),
4090                         .intfver_ofld = FW_INTFVER(T6, OFLD),
4091                         .intfver_ri = FW_INTFVER(T6, RI),
4092                         .intfver_iscsipdu = FW_INTFVER(T6, ISCSIPDU),
4093                         .intfver_iscsi = FW_INTFVER(T6, ISCSI),
4094                         .intfver_fcoepdu = FW_INTFVER(T6, FCOEPDU),
4095                         .intfver_fcoe = FW_INTFVER(T6, FCOE),
4096                 },
4097         }
4098
4099 };
4100
4101 static struct fw_info *find_fw_info(int chip)
4102 {
4103         int i;
4104
4105         for (i = 0; i < ARRAY_SIZE(fw_info_array); i++) {
4106                 if (fw_info_array[i].chip == chip)
4107                         return &fw_info_array[i];
4108         }
4109         return NULL;
4110 }
4111
4112 /*
4113  * Phase 0 of initialization: contact FW, obtain config, perform basic init.
4114  */
4115 static int adap_init0(struct adapter *adap)
4116 {
4117         int ret;
4118         u32 v, port_vec;
4119         enum dev_state state;
4120         u32 params[7], val[7];
4121         struct fw_caps_config_cmd caps_cmd;
4122         int reset = 1;
4123
4124         /* Grab Firmware Device Log parameters as early as possible so we have
4125          * access to it for debugging, etc.
4126          */
4127         ret = t4_init_devlog_params(adap);
4128         if (ret < 0)
4129                 return ret;
4130
4131         /* Contact FW, advertising Master capability */
4132         ret = t4_fw_hello(adap, adap->mbox, adap->mbox,
4133                           is_kdump_kernel() ? MASTER_MUST : MASTER_MAY, &state);
4134         if (ret < 0) {
4135                 dev_err(adap->pdev_dev, "could not connect to FW, error %d\n",
4136                         ret);
4137                 return ret;
4138         }
4139         if (ret == adap->mbox)
4140                 adap->flags |= MASTER_PF;
4141
4142         /*
4143          * If we're the Master PF Driver and the device is uninitialized,
4144          * then let's consider upgrading the firmware ...  (We always want
4145          * to check the firmware version number in order to A. get it for
4146          * later reporting and B. to warn if the currently loaded firmware
4147          * is excessively mismatched relative to the driver.)
4148          */
4149
4150         t4_get_version_info(adap);
4151         ret = t4_check_fw_version(adap);
4152         /* If firmware is too old (not supported by driver) force an update. */
4153         if (ret)
4154                 state = DEV_STATE_UNINIT;
4155         if ((adap->flags & MASTER_PF) && state != DEV_STATE_INIT) {
4156                 struct fw_info *fw_info;
4157                 struct fw_hdr *card_fw;
4158                 const struct firmware *fw;
4159                 const u8 *fw_data = NULL;
4160                 unsigned int fw_size = 0;
4161
4162                 /* This is the firmware whose headers the driver was compiled
4163                  * against
4164                  */
4165                 fw_info = find_fw_info(CHELSIO_CHIP_VERSION(adap->params.chip));
4166                 if (fw_info == NULL) {
4167                         dev_err(adap->pdev_dev,
4168                                 "unable to get firmware info for chip %d.\n",
4169                                 CHELSIO_CHIP_VERSION(adap->params.chip));
4170                         return -EINVAL;
4171                 }
4172
4173                 /* allocate memory to read the header of the firmware on the
4174                  * card
4175                  */
4176                 card_fw = kvzalloc(sizeof(*card_fw), GFP_KERNEL);
4177                 if (!card_fw) {
4178                         ret = -ENOMEM;
4179                         goto bye;
4180                 }
4181
4182                 /* Get FW from from /lib/firmware/ */
4183                 ret = request_firmware(&fw, fw_info->fw_mod_name,
4184                                        adap->pdev_dev);
4185                 if (ret < 0) {
4186                         dev_err(adap->pdev_dev,
4187                                 "unable to load firmware image %s, error %d\n",
4188                                 fw_info->fw_mod_name, ret);
4189                 } else {
4190                         fw_data = fw->data;
4191                         fw_size = fw->size;
4192                 }
4193
4194                 /* upgrade FW logic */
4195                 ret = t4_prep_fw(adap, fw_info, fw_data, fw_size, card_fw,
4196                                  state, &reset);
4197
4198                 /* Cleaning up */
4199                 release_firmware(fw);
4200                 kvfree(card_fw);
4201
4202                 if (ret < 0)
4203                         goto bye;
4204         }
4205
4206         /* If the firmware is initialized already, emit a simply note to that
4207          * effect. Otherwise, it's time to try initializing the adapter.
4208          */
4209         if (state == DEV_STATE_INIT) {
4210                 ret = adap_config_hma(adap);
4211                 if (ret)
4212                         dev_err(adap->pdev_dev,
4213                                 "HMA configuration failed with error %d\n",
4214                                 ret);
4215                 dev_info(adap->pdev_dev, "Coming up as %s: "\
4216                          "Adapter already initialized\n",
4217                          adap->flags & MASTER_PF ? "MASTER" : "SLAVE");
4218         } else {
4219                 dev_info(adap->pdev_dev, "Coming up as MASTER: "\
4220                          "Initializing adapter\n");
4221
4222                 /* Find out whether we're dealing with a version of the
4223                  * firmware which has configuration file support.
4224                  */
4225                 params[0] = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) |
4226                              FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_CF));
4227                 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1,
4228                                       params, val);
4229
4230                 /* If the firmware doesn't support Configuration Files,
4231                  * return an error.
4232                  */
4233                 if (ret < 0) {
4234                         dev_err(adap->pdev_dev, "firmware doesn't support "
4235                                 "Firmware Configuration Files\n");
4236                         goto bye;
4237                 }
4238
4239                 /* The firmware provides us with a memory buffer where we can
4240                  * load a Configuration File from the host if we want to
4241                  * override the Configuration File in flash.
4242                  */
4243                 ret = adap_init0_config(adap, reset);
4244                 if (ret == -ENOENT) {
4245                         dev_err(adap->pdev_dev, "no Configuration File "
4246                                 "present on adapter.\n");
4247                         goto bye;
4248                 }
4249                 if (ret < 0) {
4250                         dev_err(adap->pdev_dev, "could not initialize "
4251                                 "adapter, error %d\n", -ret);
4252                         goto bye;
4253                 }
4254         }
4255
4256         /* Now that we've successfully configured and initialized the adapter
4257          * (or found it already initialized), we can ask the Firmware what
4258          * resources it has provisioned for us.
4259          */
4260         ret = t4_get_pfres(adap);
4261         if (ret) {
4262                 dev_err(adap->pdev_dev,
4263                         "Unable to retrieve resource provisioning information\n");
4264                 goto bye;
4265         }
4266
4267         /* Grab VPD parameters.  This should be done after we establish a
4268          * connection to the firmware since some of the VPD parameters
4269          * (notably the Core Clock frequency) are retrieved via requests to
4270          * the firmware.  On the other hand, we need these fairly early on
4271          * so we do this right after getting ahold of the firmware.
4272          *
4273          * We need to do this after initializing the adapter because someone
4274          * could have FLASHed a new VPD which won't be read by the firmware
4275          * until we do the RESET ...
4276          */
4277         ret = t4_get_vpd_params(adap, &adap->params.vpd);
4278         if (ret < 0)
4279                 goto bye;
4280
4281         /* Find out what ports are available to us.  Note that we need to do
4282          * this before calling adap_init0_no_config() since it needs nports
4283          * and portvec ...
4284          */
4285         v =
4286             FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) |
4287             FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_PORTVEC);
4288         ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1, &v, &port_vec);
4289         if (ret < 0)
4290                 goto bye;
4291
4292         adap->params.nports = hweight32(port_vec);
4293         adap->params.portvec = port_vec;
4294
4295         /* Give the SGE code a chance to pull in anything that it needs ...
4296          * Note that this must be called after we retrieve our VPD parameters
4297          * in order to know how to convert core ticks to seconds, etc.
4298          */
4299         ret = t4_sge_init(adap);
4300         if (ret < 0)
4301                 goto bye;
4302
4303         if (is_bypass_device(adap->pdev->device))
4304                 adap->params.bypass = 1;
4305
4306         /*
4307          * Grab some of our basic fundamental operating parameters.
4308          */
4309 #define FW_PARAM_DEV(param) \
4310         (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) | \
4311         FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_##param))
4312
4313 #define FW_PARAM_PFVF(param) \
4314         FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_PFVF) | \
4315         FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_PFVF_##param)|  \
4316         FW_PARAMS_PARAM_Y_V(0) | \
4317         FW_PARAMS_PARAM_Z_V(0)
4318
4319         params[0] = FW_PARAM_PFVF(EQ_START);
4320         params[1] = FW_PARAM_PFVF(L2T_START);
4321         params[2] = FW_PARAM_PFVF(L2T_END);
4322         params[3] = FW_PARAM_PFVF(FILTER_START);
4323         params[4] = FW_PARAM_PFVF(FILTER_END);
4324         params[5] = FW_PARAM_PFVF(IQFLINT_START);
4325         ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 6, params, val);
4326         if (ret < 0)
4327                 goto bye;
4328         adap->sge.egr_start = val[0];
4329         adap->l2t_start = val[1];
4330         adap->l2t_end = val[2];
4331         adap->tids.ftid_base = val[3];
4332         adap->tids.nftids = val[4] - val[3] + 1;
4333         adap->sge.ingr_start = val[5];
4334
4335         if (CHELSIO_CHIP_VERSION(adap->params.chip) > CHELSIO_T5) {
4336                 /* Read the raw mps entries. In T6, the last 2 tcam entries
4337                  * are reserved for raw mac addresses (rawf = 2, one per port).
4338                  */
4339                 params[0] = FW_PARAM_PFVF(RAWF_START);
4340                 params[1] = FW_PARAM_PFVF(RAWF_END);
4341                 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2,
4342                                       params, val);
4343                 if (ret == 0) {
4344                         adap->rawf_start = val[0];
4345                         adap->rawf_cnt = val[1] - val[0] + 1;
4346                 }
4347         }
4348
4349         /* qids (ingress/egress) returned from firmware can be anywhere
4350          * in the range from EQ(IQFLINT)_START to EQ(IQFLINT)_END.
4351          * Hence driver needs to allocate memory for this range to
4352          * store the queue info. Get the highest IQFLINT/EQ index returned
4353          * in FW_EQ_*_CMD.alloc command.
4354          */
4355         params[0] = FW_PARAM_PFVF(EQ_END);
4356         params[1] = FW_PARAM_PFVF(IQFLINT_END);
4357         ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2, params, val);
4358         if (ret < 0)
4359                 goto bye;
4360         adap->sge.egr_sz = val[0] - adap->sge.egr_start + 1;
4361         adap->sge.ingr_sz = val[1] - adap->sge.ingr_start + 1;
4362
4363         adap->sge.egr_map = kcalloc(adap->sge.egr_sz,
4364                                     sizeof(*adap->sge.egr_map), GFP_KERNEL);
4365         if (!adap->sge.egr_map) {
4366                 ret = -ENOMEM;
4367                 goto bye;
4368         }
4369
4370         adap->sge.ingr_map = kcalloc(adap->sge.ingr_sz,
4371                                      sizeof(*adap->sge.ingr_map), GFP_KERNEL);
4372         if (!adap->sge.ingr_map) {
4373                 ret = -ENOMEM;
4374                 goto bye;
4375         }
4376
4377         /* Allocate the memory for the vaious egress queue bitmaps
4378          * ie starving_fl, txq_maperr and blocked_fl.
4379          */
4380         adap->sge.starving_fl = kcalloc(BITS_TO_LONGS(adap->sge.egr_sz),
4381                                         sizeof(long), GFP_KERNEL);
4382         if (!adap->sge.starving_fl) {
4383                 ret = -ENOMEM;
4384                 goto bye;
4385         }
4386
4387         adap->sge.txq_maperr = kcalloc(BITS_TO_LONGS(adap->sge.egr_sz),
4388                                        sizeof(long), GFP_KERNEL);
4389         if (!adap->sge.txq_maperr) {
4390                 ret = -ENOMEM;
4391                 goto bye;
4392         }
4393
4394 #ifdef CONFIG_DEBUG_FS
4395         adap->sge.blocked_fl = kcalloc(BITS_TO_LONGS(adap->sge.egr_sz),
4396                                        sizeof(long), GFP_KERNEL);
4397         if (!adap->sge.blocked_fl) {
4398                 ret = -ENOMEM;
4399                 goto bye;
4400         }
4401 #endif
4402
4403         params[0] = FW_PARAM_PFVF(CLIP_START);
4404         params[1] = FW_PARAM_PFVF(CLIP_END);
4405         ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2, params, val);
4406         if (ret < 0)
4407                 goto bye;
4408         adap->clipt_start = val[0];
4409         adap->clipt_end = val[1];
4410
4411         /* We don't yet have a PARAMs calls to retrieve the number of Traffic
4412          * Classes supported by the hardware/firmware so we hard code it here
4413          * for now.
4414          */
4415         adap->params.nsched_cls = is_t4(adap->params.chip) ? 15 : 16;
4416
4417         /* query params related to active filter region */
4418         params[0] = FW_PARAM_PFVF(ACTIVE_FILTER_START);
4419         params[1] = FW_PARAM_PFVF(ACTIVE_FILTER_END);
4420         ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2, params, val);
4421         /* If Active filter size is set we enable establishing
4422          * offload connection through firmware work request
4423          */
4424         if ((val[0] != val[1]) && (ret >= 0)) {
4425                 adap->flags |= FW_OFLD_CONN;
4426                 adap->tids.aftid_base = val[0];
4427                 adap->tids.aftid_end = val[1];
4428         }
4429
4430         /* If we're running on newer firmware, let it know that we're
4431          * prepared to deal with encapsulated CPL messages.  Older
4432          * firmware won't understand this and we'll just get
4433          * unencapsulated messages ...
4434          */
4435         params[0] = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
4436         val[0] = 1;
4437         (void)t4_set_params(adap, adap->mbox, adap->pf, 0, 1, params, val);
4438
4439         /*
4440          * Find out whether we're allowed to use the T5+ ULPTX MEMWRITE DSGL
4441          * capability.  Earlier versions of the firmware didn't have the
4442          * ULPTX_MEMWRITE_DSGL so we'll interpret a query failure as no
4443          * permission to use ULPTX MEMWRITE DSGL.
4444          */
4445         if (is_t4(adap->params.chip)) {
4446                 adap->params.ulptx_memwrite_dsgl = false;
4447         } else {
4448                 params[0] = FW_PARAM_DEV(ULPTX_MEMWRITE_DSGL);
4449                 ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
4450                                       1, params, val);
4451                 adap->params.ulptx_memwrite_dsgl = (ret == 0 && val[0] != 0);
4452         }
4453
4454         /* See if FW supports FW_RI_FR_NSMR_TPTE_WR work request */
4455         params[0] = FW_PARAM_DEV(RI_FR_NSMR_TPTE_WR);
4456         ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
4457                               1, params, val);
4458         adap->params.fr_nsmr_tpte_wr_support = (ret == 0 && val[0] != 0);
4459
4460         /* See if FW supports FW_FILTER2 work request */
4461         if (is_t4(adap->params.chip)) {
4462                 adap->params.filter2_wr_support = 0;
4463         } else {
4464                 params[0] = FW_PARAM_DEV(FILTER2_WR);
4465                 ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
4466                                       1, params, val);
4467                 adap->params.filter2_wr_support = (ret == 0 && val[0] != 0);
4468         }
4469
4470         /*
4471          * Get device capabilities so we can determine what resources we need
4472          * to manage.
4473          */
4474         memset(&caps_cmd, 0, sizeof(caps_cmd));
4475         caps_cmd.op_to_write = htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
4476                                      FW_CMD_REQUEST_F | FW_CMD_READ_F);
4477         caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
4478         ret = t4_wr_mbox(adap, adap->mbox, &caps_cmd, sizeof(caps_cmd),
4479                          &caps_cmd);
4480         if (ret < 0)
4481                 goto bye;
4482
4483         if (caps_cmd.ofldcaps ||
4484             (caps_cmd.niccaps & htons(FW_CAPS_CONFIG_NIC_HASHFILTER))) {
4485                 /* query offload-related parameters */
4486                 params[0] = FW_PARAM_DEV(NTID);
4487                 params[1] = FW_PARAM_PFVF(SERVER_START);
4488                 params[2] = FW_PARAM_PFVF(SERVER_END);
4489                 params[3] = FW_PARAM_PFVF(TDDP_START);
4490                 params[4] = FW_PARAM_PFVF(TDDP_END);
4491                 params[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
4492                 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 6,
4493                                       params, val);
4494                 if (ret < 0)
4495                         goto bye;
4496                 adap->tids.ntids = val[0];
4497                 adap->tids.natids = min(adap->tids.ntids / 2, MAX_ATIDS);
4498                 adap->tids.stid_base = val[1];
4499                 adap->tids.nstids = val[2] - val[1] + 1;
4500                 /*
4501                  * Setup server filter region. Divide the available filter
4502                  * region into two parts. Regular filters get 1/3rd and server
4503                  * filters get 2/3rd part. This is only enabled if workarond
4504                  * path is enabled.
4505                  * 1. For regular filters.
4506                  * 2. Server filter: This are special filters which are used
4507                  * to redirect SYN packets to offload queue.
4508                  */
4509                 if (adap->flags & FW_OFLD_CONN && !is_bypass(adap)) {
4510                         adap->tids.sftid_base = adap->tids.ftid_base +
4511                                         DIV_ROUND_UP(adap->tids.nftids, 3);
4512                         adap->tids.nsftids = adap->tids.nftids -
4513                                          DIV_ROUND_UP(adap->tids.nftids, 3);
4514                         adap->tids.nftids = adap->tids.sftid_base -
4515                                                 adap->tids.ftid_base;
4516                 }
4517                 adap->vres.ddp.start = val[3];
4518                 adap->vres.ddp.size = val[4] - val[3] + 1;
4519                 adap->params.ofldq_wr_cred = val[5];
4520
4521                 if (caps_cmd.niccaps & htons(FW_CAPS_CONFIG_NIC_HASHFILTER)) {
4522                         ret = init_hash_filter(adap);
4523                         if (ret < 0)
4524                                 goto bye;
4525                 } else {
4526                         adap->params.offload = 1;
4527                         adap->num_ofld_uld += 1;
4528                 }
4529         }
4530         if (caps_cmd.rdmacaps) {
4531                 params[0] = FW_PARAM_PFVF(STAG_START);
4532                 params[1] = FW_PARAM_PFVF(STAG_END);
4533                 params[2] = FW_PARAM_PFVF(RQ_START);
4534                 params[3] = FW_PARAM_PFVF(RQ_END);
4535                 params[4] = FW_PARAM_PFVF(PBL_START);
4536                 params[5] = FW_PARAM_PFVF(PBL_END);
4537                 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 6,
4538                                       params, val);
4539                 if (ret < 0)
4540                         goto bye;
4541                 adap->vres.stag.start = val[0];
4542                 adap->vres.stag.size = val[1] - val[0] + 1;
4543                 adap->vres.rq.start = val[2];
4544                 adap->vres.rq.size = val[3] - val[2] + 1;
4545                 adap->vres.pbl.start = val[4];
4546                 adap->vres.pbl.size = val[5] - val[4] + 1;
4547
4548                 params[0] = FW_PARAM_PFVF(SRQ_START);
4549                 params[1] = FW_PARAM_PFVF(SRQ_END);
4550                 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2,
4551                                       params, val);
4552                 if (!ret) {
4553                         adap->vres.srq.start = val[0];
4554                         adap->vres.srq.size = val[1] - val[0] + 1;
4555                 }
4556                 if (adap->vres.srq.size) {
4557                         adap->srq = t4_init_srq(adap->vres.srq.size);
4558                         if (!adap->srq)
4559                                 dev_warn(&adap->pdev->dev, "could not allocate SRQ, continuing\n");
4560                 }
4561
4562                 params[0] = FW_PARAM_PFVF(SQRQ_START);
4563                 params[1] = FW_PARAM_PFVF(SQRQ_END);
4564                 params[2] = FW_PARAM_PFVF(CQ_START);
4565                 params[3] = FW_PARAM_PFVF(CQ_END);
4566                 params[4] = FW_PARAM_PFVF(OCQ_START);
4567                 params[5] = FW_PARAM_PFVF(OCQ_END);
4568                 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 6, params,
4569                                       val);
4570                 if (ret < 0)
4571                         goto bye;
4572                 adap->vres.qp.start = val[0];
4573                 adap->vres.qp.size = val[1] - val[0] + 1;
4574                 adap->vres.cq.start = val[2];
4575                 adap->vres.cq.size = val[3] - val[2] + 1;
4576                 adap->vres.ocq.start = val[4];
4577                 adap->vres.ocq.size = val[5] - val[4] + 1;
4578
4579                 params[0] = FW_PARAM_DEV(MAXORDIRD_QP);
4580                 params[1] = FW_PARAM_DEV(MAXIRD_ADAPTER);
4581                 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2, params,
4582                                       val);
4583                 if (ret < 0) {
4584                         adap->params.max_ordird_qp = 8;
4585                         adap->params.max_ird_adapter = 32 * adap->tids.ntids;
4586                         ret = 0;
4587                 } else {
4588                         adap->params.max_ordird_qp = val[0];
4589                         adap->params.max_ird_adapter = val[1];
4590                 }
4591                 dev_info(adap->pdev_dev,
4592                          "max_ordird_qp %d max_ird_adapter %d\n",
4593                          adap->params.max_ordird_qp,
4594                          adap->params.max_ird_adapter);
4595
4596                 /* Enable write_with_immediate if FW supports it */
4597                 params[0] = FW_PARAM_DEV(RDMA_WRITE_WITH_IMM);
4598                 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1, params,
4599                                       val);
4600                 adap->params.write_w_imm_support = (ret == 0 && val[0] != 0);
4601
4602                 /* Enable write_cmpl if FW supports it */
4603                 params[0] = FW_PARAM_DEV(RI_WRITE_CMPL_WR);
4604                 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1, params,
4605                                       val);
4606                 adap->params.write_cmpl_support = (ret == 0 && val[0] != 0);
4607                 adap->num_ofld_uld += 2;
4608         }
4609         if (caps_cmd.iscsicaps) {
4610                 params[0] = FW_PARAM_PFVF(ISCSI_START);
4611                 params[1] = FW_PARAM_PFVF(ISCSI_END);
4612                 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2,
4613                                       params, val);
4614                 if (ret < 0)
4615                         goto bye;
4616                 adap->vres.iscsi.start = val[0];
4617                 adap->vres.iscsi.size = val[1] - val[0] + 1;
4618                 /* LIO target and cxgb4i initiaitor */
4619                 adap->num_ofld_uld += 2;
4620         }
4621         if (caps_cmd.cryptocaps) {
4622                 if (ntohs(caps_cmd.cryptocaps) &
4623                     FW_CAPS_CONFIG_CRYPTO_LOOKASIDE) {
4624                         params[0] = FW_PARAM_PFVF(NCRYPTO_LOOKASIDE);
4625                         ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
4626                                               2, params, val);
4627                         if (ret < 0) {
4628                                 if (ret != -EINVAL)
4629                                         goto bye;
4630                         } else {
4631                                 adap->vres.ncrypto_fc = val[0];
4632                         }
4633                         adap->num_ofld_uld += 1;
4634                 }
4635                 if (ntohs(caps_cmd.cryptocaps) &
4636                     FW_CAPS_CONFIG_TLS_INLINE) {
4637                         params[0] = FW_PARAM_PFVF(TLS_START);
4638                         params[1] = FW_PARAM_PFVF(TLS_END);
4639                         ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
4640                                               2, params, val);
4641                         if (ret < 0)
4642                                 goto bye;
4643                         adap->vres.key.start = val[0];
4644                         adap->vres.key.size = val[1] - val[0] + 1;
4645                         adap->num_uld += 1;
4646                 }
4647                 adap->params.crypto = ntohs(caps_cmd.cryptocaps);
4648         }
4649 #undef FW_PARAM_PFVF
4650 #undef FW_PARAM_DEV
4651
4652         /* The MTU/MSS Table is initialized by now, so load their values.  If
4653          * we're initializing the adapter, then we'll make any modifications
4654          * we want to the MTU/MSS Table and also initialize the congestion
4655          * parameters.
4656          */
4657         t4_read_mtu_tbl(adap, adap->params.mtus, NULL);
4658         if (state != DEV_STATE_INIT) {
4659                 int i;
4660
4661                 /* The default MTU Table contains values 1492 and 1500.
4662                  * However, for TCP, it's better to have two values which are
4663                  * a multiple of 8 +/- 4 bytes apart near this popular MTU.
4664                  * This allows us to have a TCP Data Payload which is a
4665                  * multiple of 8 regardless of what combination of TCP Options
4666                  * are in use (always a multiple of 4 bytes) which is
4667                  * important for performance reasons.  For instance, if no
4668                  * options are in use, then we have a 20-byte IP header and a
4669                  * 20-byte TCP header.  In this case, a 1500-byte MSS would
4670                  * result in a TCP Data Payload of 1500 - 40 == 1460 bytes
4671                  * which is not a multiple of 8.  So using an MSS of 1488 in
4672                  * this case results in a TCP Data Payload of 1448 bytes which
4673                  * is a multiple of 8.  On the other hand, if 12-byte TCP Time
4674                  * Stamps have been negotiated, then an MTU of 1500 bytes
4675                  * results in a TCP Data Payload of 1448 bytes which, as
4676                  * above, is a multiple of 8 bytes ...
4677                  */
4678                 for (i = 0; i < NMTUS; i++)
4679                         if (adap->params.mtus[i] == 1492) {
4680                                 adap->params.mtus[i] = 1488;
4681                                 break;
4682                         }
4683
4684                 t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
4685                              adap->params.b_wnd);
4686         }
4687         t4_init_sge_params(adap);
4688         adap->flags |= FW_OK;
4689         t4_init_tp_params(adap, true);
4690         return 0;
4691
4692         /*
4693          * Something bad happened.  If a command timed out or failed with EIO
4694          * FW does not operate within its spec or something catastrophic
4695          * happened to HW/FW, stop issuing commands.
4696          */
4697 bye:
4698         adap_free_hma_mem(adap);
4699         kfree(adap->sge.egr_map);
4700         kfree(adap->sge.ingr_map);
4701         kfree(adap->sge.starving_fl);
4702         kfree(adap->sge.txq_maperr);
4703 #ifdef CONFIG_DEBUG_FS
4704         kfree(adap->sge.blocked_fl);
4705 #endif
4706         if (ret != -ETIMEDOUT && ret != -EIO)
4707                 t4_fw_bye(adap, adap->mbox);
4708         return ret;
4709 }
4710
4711 /* EEH callbacks */
4712
4713 static pci_ers_result_t eeh_err_detected(struct pci_dev *pdev,
4714                                          pci_channel_state_t state)
4715 {
4716         int i;
4717         struct adapter *adap = pci_get_drvdata(pdev);
4718
4719         if (!adap)
4720                 goto out;
4721
4722         rtnl_lock();
4723         adap->flags &= ~FW_OK;
4724         notify_ulds(adap, CXGB4_STATE_START_RECOVERY);
4725         spin_lock(&adap->stats_lock);
4726         for_each_port(adap, i) {
4727                 struct net_device *dev = adap->port[i];
4728                 if (dev) {
4729                         netif_device_detach(dev);
4730                         netif_carrier_off(dev);
4731                 }
4732         }
4733         spin_unlock(&adap->stats_lock);
4734         disable_interrupts(adap);
4735         if (adap->flags & FULL_INIT_DONE)
4736                 cxgb_down(adap);
4737         rtnl_unlock();
4738         if ((adap->flags & DEV_ENABLED)) {
4739                 pci_disable_device(pdev);
4740                 adap->flags &= ~DEV_ENABLED;
4741         }
4742 out:    return state == pci_channel_io_perm_failure ?
4743                 PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET;
4744 }
4745
4746 static pci_ers_result_t eeh_slot_reset(struct pci_dev *pdev)
4747 {
4748         int i, ret;
4749         struct fw_caps_config_cmd c;
4750         struct adapter *adap = pci_get_drvdata(pdev);
4751
4752         if (!adap) {
4753                 pci_restore_state(pdev);
4754                 pci_save_state(pdev);
4755                 return PCI_ERS_RESULT_RECOVERED;
4756         }
4757
4758         if (!(adap->flags & DEV_ENABLED)) {
4759                 if (pci_enable_device(pdev)) {
4760                         dev_err(&pdev->dev, "Cannot reenable PCI "
4761                                             "device after reset\n");
4762                         return PCI_ERS_RESULT_DISCONNECT;
4763                 }
4764                 adap->flags |= DEV_ENABLED;
4765         }
4766
4767         pci_set_master(pdev);
4768         pci_restore_state(pdev);
4769         pci_save_state(pdev);
4770
4771         if (t4_wait_dev_ready(adap->regs) < 0)
4772                 return PCI_ERS_RESULT_DISCONNECT;
4773         if (t4_fw_hello(adap, adap->mbox, adap->pf, MASTER_MUST, NULL) < 0)
4774                 return PCI_ERS_RESULT_DISCONNECT;
4775         adap->flags |= FW_OK;
4776         if (adap_init1(adap, &c))
4777                 return PCI_ERS_RESULT_DISCONNECT;
4778
4779         for_each_port(adap, i) {
4780                 struct port_info *p = adap2pinfo(adap, i);
4781
4782                 ret = t4_alloc_vi(adap, adap->mbox, p->tx_chan, adap->pf, 0, 1,
4783                                   NULL, NULL);
4784                 if (ret < 0)
4785                         return PCI_ERS_RESULT_DISCONNECT;
4786                 p->viid = ret;
4787                 p->xact_addr_filt = -1;
4788         }
4789
4790         t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
4791                      adap->params.b_wnd);
4792         setup_memwin(adap);
4793         if (cxgb_up(adap))
4794                 return PCI_ERS_RESULT_DISCONNECT;
4795         return PCI_ERS_RESULT_RECOVERED;
4796 }
4797
4798 static void eeh_resume(struct pci_dev *pdev)
4799 {
4800         int i;
4801         struct adapter *adap = pci_get_drvdata(pdev);
4802
4803         if (!adap)
4804                 return;
4805
4806         rtnl_lock();
4807         for_each_port(adap, i) {
4808                 struct net_device *dev = adap->port[i];
4809                 if (dev) {
4810                         if (netif_running(dev)) {
4811                                 link_start(dev);
4812                                 cxgb_set_rxmode(dev);
4813                         }
4814                         netif_device_attach(dev);
4815                 }
4816         }
4817         rtnl_unlock();
4818 }
4819
4820 static const struct pci_error_handlers cxgb4_eeh = {
4821         .error_detected = eeh_err_detected,
4822         .slot_reset     = eeh_slot_reset,
4823         .resume         = eeh_resume,
4824 };
4825
4826 /* Return true if the Link Configuration supports "High Speeds" (those greater
4827  * than 1Gb/s).
4828  */
4829 static inline bool is_x_10g_port(const struct link_config *lc)
4830 {
4831         unsigned int speeds, high_speeds;
4832
4833         speeds = FW_PORT_CAP32_SPEED_V(FW_PORT_CAP32_SPEED_G(lc->pcaps));
4834         high_speeds = speeds &
4835                         ~(FW_PORT_CAP32_SPEED_100M | FW_PORT_CAP32_SPEED_1G);
4836
4837         return high_speeds != 0;
4838 }
4839
4840 /*
4841  * Perform default configuration of DMA queues depending on the number and type
4842  * of ports we found and the number of available CPUs.  Most settings can be
4843  * modified by the admin prior to actual use.
4844  */
4845 static int cfg_queues(struct adapter *adap)
4846 {
4847         struct sge *s = &adap->sge;
4848         int i, n10g = 0, qidx = 0;
4849         int niqflint, neq, avail_eth_qsets;
4850         int max_eth_qsets = 32;
4851 #ifndef CONFIG_CHELSIO_T4_DCB
4852         int q10g = 0;
4853 #endif
4854
4855         /* Reduce memory usage in kdump environment, disable all offload.
4856          */
4857         if (is_kdump_kernel() || (is_uld(adap) && t4_uld_mem_alloc(adap))) {
4858                 adap->params.offload = 0;
4859                 adap->params.crypto = 0;
4860         }
4861
4862         /* Calculate the number of Ethernet Queue Sets available based on
4863          * resources provisioned for us.  We always have an Asynchronous
4864          * Firmware Event Ingress Queue.  If we're operating in MSI or Legacy
4865          * IRQ Pin Interrupt mode, then we'll also have a Forwarded Interrupt
4866          * Ingress Queue.  Meanwhile, we need two Egress Queues for each
4867          * Queue Set: one for the Free List and one for the Ethernet TX Queue.
4868          *
4869          * Note that we should also take into account all of the various
4870          * Offload Queues.  But, in any situation where we're operating in
4871          * a Resource Constrained Provisioning environment, doing any Offload
4872          * at all is problematic ...
4873          */
4874         niqflint = adap->params.pfres.niqflint - 1;
4875         if (!(adap->flags & USING_MSIX))
4876                 niqflint--;
4877         neq = adap->params.pfres.neq / 2;
4878         avail_eth_qsets = min(niqflint, neq);
4879
4880         if (avail_eth_qsets > max_eth_qsets)
4881                 avail_eth_qsets = max_eth_qsets;
4882
4883         if (avail_eth_qsets < adap->params.nports) {
4884                 dev_err(adap->pdev_dev, "avail_eth_qsets=%d < nports=%d\n",
4885                         avail_eth_qsets, adap->params.nports);
4886                 return -ENOMEM;
4887         }
4888
4889         /* Count the number of 10Gb/s or better ports */
4890         for_each_port(adap, i)
4891                 n10g += is_x_10g_port(&adap2pinfo(adap, i)->link_cfg);
4892
4893 #ifdef CONFIG_CHELSIO_T4_DCB
4894         /* For Data Center Bridging support we need to be able to support up
4895          * to 8 Traffic Priorities; each of which will be assigned to its
4896          * own TX Queue in order to prevent Head-Of-Line Blocking.
4897          */
4898         if (adap->params.nports * 8 > avail_eth_qsets) {
4899                 dev_err(adap->pdev_dev, "DCB avail_eth_qsets=%d < %d!\n",
4900                         avail_eth_qsets, adap->params.nports * 8);
4901                 return -ENOMEM;
4902         }
4903
4904         for_each_port(adap, i) {
4905                 struct port_info *pi = adap2pinfo(adap, i);
4906
4907                 pi->first_qset = qidx;
4908                 pi->nqsets = is_kdump_kernel() ? 1 : 8;
4909                 qidx += pi->nqsets;
4910         }
4911 #else /* !CONFIG_CHELSIO_T4_DCB */
4912         /*
4913          * We default to 1 queue per non-10G port and up to # of cores queues
4914          * per 10G port.
4915          */
4916         if (n10g)
4917                 q10g = (avail_eth_qsets - (adap->params.nports - n10g)) / n10g;
4918         if (q10g > netif_get_num_default_rss_queues())
4919                 q10g = netif_get_num_default_rss_queues();
4920
4921         if (is_kdump_kernel())
4922                 q10g = 1;
4923
4924         for_each_port(adap, i) {
4925                 struct port_info *pi = adap2pinfo(adap, i);
4926
4927                 pi->first_qset = qidx;
4928                 pi->nqsets = is_x_10g_port(&pi->link_cfg) ? q10g : 1;
4929                 qidx += pi->nqsets;
4930         }
4931 #endif /* !CONFIG_CHELSIO_T4_DCB */
4932
4933         s->ethqsets = qidx;
4934         s->max_ethqsets = qidx;   /* MSI-X may lower it later */
4935
4936         if (is_uld(adap)) {
4937                 /*
4938                  * For offload we use 1 queue/channel if all ports are up to 1G,
4939                  * otherwise we divide all available queues amongst the channels
4940                  * capped by the number of available cores.
4941                  */
4942                 if (n10g) {
4943                         i = min_t(int, MAX_OFLD_QSETS, num_online_cpus());
4944                         s->ofldqsets = roundup(i, adap->params.nports);
4945                 } else {
4946                         s->ofldqsets = adap->params.nports;
4947                 }
4948         }
4949
4950         for (i = 0; i < ARRAY_SIZE(s->ethrxq); i++) {
4951                 struct sge_eth_rxq *r = &s->ethrxq[i];
4952
4953                 init_rspq(adap, &r->rspq, 5, 10, 1024, 64);
4954                 r->fl.size = 72;
4955         }
4956
4957         for (i = 0; i < ARRAY_SIZE(s->ethtxq); i++)
4958                 s->ethtxq[i].q.size = 1024;
4959
4960         for (i = 0; i < ARRAY_SIZE(s->ctrlq); i++)
4961                 s->ctrlq[i].q.size = 512;
4962
4963         if (!is_t4(adap->params.chip))
4964                 s->ptptxq.q.size = 8;
4965
4966         init_rspq(adap, &s->fw_evtq, 0, 1, 1024, 64);
4967         init_rspq(adap, &s->intrq, 0, 1, 512, 64);
4968
4969         return 0;
4970 }
4971
4972 /*
4973  * Reduce the number of Ethernet queues across all ports to at most n.
4974  * n provides at least one queue per port.
4975  */
4976 static void reduce_ethqs(struct adapter *adap, int n)
4977 {
4978         int i;
4979         struct port_info *pi;
4980
4981         while (n < adap->sge.ethqsets)
4982                 for_each_port(adap, i) {
4983                         pi = adap2pinfo(adap, i);
4984                         if (pi->nqsets > 1) {
4985                                 pi->nqsets--;
4986                                 adap->sge.ethqsets--;
4987                                 if (adap->sge.ethqsets <= n)
4988                                         break;
4989                         }
4990                 }
4991
4992         n = 0;
4993         for_each_port(adap, i) {
4994                 pi = adap2pinfo(adap, i);
4995                 pi->first_qset = n;
4996                 n += pi->nqsets;
4997         }
4998 }
4999
5000 static int get_msix_info(struct adapter *adap)
5001 {
5002         struct uld_msix_info *msix_info;
5003         unsigned int max_ingq = 0;
5004
5005         if (is_offload(adap))
5006                 max_ingq += MAX_OFLD_QSETS * adap->num_ofld_uld;
5007         if (is_pci_uld(adap))
5008                 max_ingq += MAX_OFLD_QSETS * adap->num_uld;
5009
5010         if (!max_ingq)
5011                 goto out;
5012
5013         msix_info = kcalloc(max_ingq, sizeof(*msix_info), GFP_KERNEL);
5014         if (!msix_info)
5015                 return -ENOMEM;
5016
5017         adap->msix_bmap_ulds.msix_bmap = kcalloc(BITS_TO_LONGS(max_ingq),
5018                                                  sizeof(long), GFP_KERNEL);
5019         if (!adap->msix_bmap_ulds.msix_bmap) {
5020                 kfree(msix_info);
5021                 return -ENOMEM;
5022         }
5023         spin_lock_init(&adap->msix_bmap_ulds.lock);
5024         adap->msix_info_ulds = msix_info;
5025 out:
5026         return 0;
5027 }
5028
5029 static void free_msix_info(struct adapter *adap)
5030 {
5031         if (!(adap->num_uld && adap->num_ofld_uld))
5032                 return;
5033
5034         kfree(adap->msix_info_ulds);
5035         kfree(adap->msix_bmap_ulds.msix_bmap);
5036 }
5037
5038 /* 2 MSI-X vectors needed for the FW queue and non-data interrupts */
5039 #define EXTRA_VECS 2
5040
5041 static int enable_msix(struct adapter *adap)
5042 {
5043         int ofld_need = 0, uld_need = 0;
5044         int i, j, want, need, allocated;
5045         struct sge *s = &adap->sge;
5046         unsigned int nchan = adap->params.nports;
5047         struct msix_entry *entries;
5048         int max_ingq = MAX_INGQ;
5049
5050         if (is_pci_uld(adap))
5051                 max_ingq += (MAX_OFLD_QSETS * adap->num_uld);
5052         if (is_offload(adap))
5053                 max_ingq += (MAX_OFLD_QSETS * adap->num_ofld_uld);
5054         entries = kmalloc_array(max_ingq + 1, sizeof(*entries),
5055                                 GFP_KERNEL);
5056         if (!entries)
5057                 return -ENOMEM;
5058
5059         /* map for msix */
5060         if (get_msix_info(adap)) {
5061                 adap->params.offload = 0;
5062                 adap->params.crypto = 0;
5063         }
5064
5065         for (i = 0; i < max_ingq + 1; ++i)
5066                 entries[i].entry = i;
5067
5068         want = s->max_ethqsets + EXTRA_VECS;
5069         if (is_offload(adap)) {
5070                 want += adap->num_ofld_uld * s->ofldqsets;
5071                 ofld_need = adap->num_ofld_uld * nchan;
5072         }
5073         if (is_pci_uld(adap)) {
5074                 want += adap->num_uld * s->ofldqsets;
5075                 uld_need = adap->num_uld * nchan;
5076         }
5077 #ifdef CONFIG_CHELSIO_T4_DCB
5078         /* For Data Center Bridging we need 8 Ethernet TX Priority Queues for
5079          * each port.
5080          */
5081         need = 8 * adap->params.nports + EXTRA_VECS + ofld_need + uld_need;
5082 #else
5083         need = adap->params.nports + EXTRA_VECS + ofld_need + uld_need;
5084 #endif
5085         allocated = pci_enable_msix_range(adap->pdev, entries, need, want);
5086         if (allocated < 0) {
5087                 dev_info(adap->pdev_dev, "not enough MSI-X vectors left,"
5088                          " not using MSI-X\n");
5089                 kfree(entries);
5090                 return allocated;
5091         }
5092
5093         /* Distribute available vectors to the various queue groups.
5094          * Every group gets its minimum requirement and NIC gets top
5095          * priority for leftovers.
5096          */
5097         i = allocated - EXTRA_VECS - ofld_need - uld_need;
5098         if (i < s->max_ethqsets) {
5099                 s->max_ethqsets = i;
5100                 if (i < s->ethqsets)
5101                         reduce_ethqs(adap, i);
5102         }
5103         if (is_uld(adap)) {
5104                 if (allocated < want)
5105                         s->nqs_per_uld = nchan;
5106                 else
5107                         s->nqs_per_uld = s->ofldqsets;
5108         }
5109
5110         for (i = 0; i < (s->max_ethqsets + EXTRA_VECS); ++i)
5111                 adap->msix_info[i].vec = entries[i].vector;
5112         if (is_uld(adap)) {
5113                 for (j = 0 ; i < allocated; ++i, j++) {
5114                         adap->msix_info_ulds[j].vec = entries[i].vector;
5115                         adap->msix_info_ulds[j].idx = i;
5116                 }
5117                 adap->msix_bmap_ulds.mapsize = j;
5118         }
5119         dev_info(adap->pdev_dev, "%d MSI-X vectors allocated, "
5120                  "nic %d per uld %d\n",
5121                  allocated, s->max_ethqsets, s->nqs_per_uld);
5122
5123         kfree(entries);
5124         return 0;
5125 }
5126
5127 #undef EXTRA_VECS
5128
5129 static int init_rss(struct adapter *adap)
5130 {
5131         unsigned int i;
5132         int err;
5133
5134         err = t4_init_rss_mode(adap, adap->mbox);
5135         if (err)
5136                 return err;
5137
5138         for_each_port(adap, i) {
5139                 struct port_info *pi = adap2pinfo(adap, i);
5140
5141                 pi->rss = kcalloc(pi->rss_size, sizeof(u16), GFP_KERNEL);
5142                 if (!pi->rss)
5143                         return -ENOMEM;
5144         }
5145         return 0;
5146 }
5147
5148 /* Dump basic information about the adapter */
5149 static void print_adapter_info(struct adapter *adapter)
5150 {
5151         /* Hardware/Firmware/etc. Version/Revision IDs */
5152         t4_dump_version_info(adapter);
5153
5154         /* Software/Hardware configuration */
5155         dev_info(adapter->pdev_dev, "Configuration: %sNIC %s, %s capable\n",
5156                  is_offload(adapter) ? "R" : "",
5157                  ((adapter->flags & USING_MSIX) ? "MSI-X" :
5158                   (adapter->flags & USING_MSI) ? "MSI" : ""),
5159                  is_offload(adapter) ? "Offload" : "non-Offload");
5160 }
5161
5162 static void print_port_info(const struct net_device *dev)
5163 {
5164         char buf[80];
5165         char *bufp = buf;
5166         const struct port_info *pi = netdev_priv(dev);
5167         const struct adapter *adap = pi->adapter;
5168
5169         if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_100M)
5170                 bufp += sprintf(bufp, "100M/");
5171         if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_1G)
5172                 bufp += sprintf(bufp, "1G/");
5173         if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_10G)
5174                 bufp += sprintf(bufp, "10G/");
5175         if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_25G)
5176                 bufp += sprintf(bufp, "25G/");
5177         if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_40G)
5178                 bufp += sprintf(bufp, "40G/");
5179         if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_50G)
5180                 bufp += sprintf(bufp, "50G/");
5181         if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_100G)
5182                 bufp += sprintf(bufp, "100G/");
5183         if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_200G)
5184                 bufp += sprintf(bufp, "200G/");
5185         if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_400G)
5186                 bufp += sprintf(bufp, "400G/");
5187         if (bufp != buf)
5188                 --bufp;
5189         sprintf(bufp, "BASE-%s", t4_get_port_type_description(pi->port_type));
5190
5191         netdev_info(dev, "%s: Chelsio %s (%s) %s\n",
5192                     dev->name, adap->params.vpd.id, adap->name, buf);
5193 }
5194
5195 /*
5196  * Free the following resources:
5197  * - memory used for tables
5198  * - MSI/MSI-X
5199  * - net devices
5200  * - resources FW is holding for us
5201  */
5202 static void free_some_resources(struct adapter *adapter)
5203 {
5204         unsigned int i;
5205
5206         kvfree(adapter->mps_encap);
5207         kvfree(adapter->smt);
5208         kvfree(adapter->l2t);
5209         kvfree(adapter->srq);
5210         t4_cleanup_sched(adapter);
5211         kvfree(adapter->tids.tid_tab);
5212         cxgb4_cleanup_tc_flower(adapter);
5213         cxgb4_cleanup_tc_u32(adapter);
5214         kfree(adapter->sge.egr_map);
5215         kfree(adapter->sge.ingr_map);
5216         kfree(adapter->sge.starving_fl);
5217         kfree(adapter->sge.txq_maperr);
5218 #ifdef CONFIG_DEBUG_FS
5219         kfree(adapter->sge.blocked_fl);
5220 #endif
5221         disable_msi(adapter);
5222
5223         for_each_port(adapter, i)
5224                 if (adapter->port[i]) {
5225                         struct port_info *pi = adap2pinfo(adapter, i);
5226
5227                         if (pi->viid != 0)
5228                                 t4_free_vi(adapter, adapter->mbox, adapter->pf,
5229                                            0, pi->viid);
5230                         kfree(adap2pinfo(adapter, i)->rss);
5231                         free_netdev(adapter->port[i]);
5232                 }
5233         if (adapter->flags & FW_OK)
5234                 t4_fw_bye(adapter, adapter->pf);
5235 }
5236
5237 #define TSO_FLAGS (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN)
5238 #define VLAN_FEAT (NETIF_F_SG | NETIF_F_IP_CSUM | TSO_FLAGS | \
5239                    NETIF_F_IPV6_CSUM | NETIF_F_HIGHDMA)
5240 #define SEGMENT_SIZE 128
5241
5242 static int t4_get_chip_type(struct adapter *adap, int ver)
5243 {
5244         u32 pl_rev = REV_G(t4_read_reg(adap, PL_REV_A));
5245
5246         switch (ver) {
5247         case CHELSIO_T4:
5248                 return CHELSIO_CHIP_CODE(CHELSIO_T4, pl_rev);
5249         case CHELSIO_T5:
5250                 return CHELSIO_CHIP_CODE(CHELSIO_T5, pl_rev);
5251         case CHELSIO_T6:
5252                 return CHELSIO_CHIP_CODE(CHELSIO_T6, pl_rev);
5253         default:
5254                 break;
5255         }
5256         return -EINVAL;
5257 }
5258
5259 #ifdef CONFIG_PCI_IOV
5260 static void cxgb4_mgmt_setup(struct net_device *dev)
5261 {
5262         dev->type = ARPHRD_NONE;
5263         dev->mtu = 0;
5264         dev->hard_header_len = 0;
5265         dev->addr_len = 0;
5266         dev->tx_queue_len = 0;
5267         dev->flags |= IFF_NOARP;
5268         dev->priv_flags |= IFF_NO_QUEUE;
5269
5270         /* Initialize the device structure. */
5271         dev->netdev_ops = &cxgb4_mgmt_netdev_ops;
5272         dev->ethtool_ops = &cxgb4_mgmt_ethtool_ops;
5273 }
5274
5275 static int cxgb4_iov_configure(struct pci_dev *pdev, int num_vfs)
5276 {
5277         struct adapter *adap = pci_get_drvdata(pdev);
5278         int err = 0;
5279         int current_vfs = pci_num_vf(pdev);
5280         u32 pcie_fw;
5281
5282         pcie_fw = readl(adap->regs + PCIE_FW_A);
5283         /* Check if fw is initialized */
5284         if (!(pcie_fw & PCIE_FW_INIT_F)) {
5285                 dev_warn(&pdev->dev, "Device not initialized\n");
5286                 return -EOPNOTSUPP;
5287         }
5288
5289         /* If any of the VF's is already assigned to Guest OS, then
5290          * SRIOV for the same cannot be modified
5291          */
5292         if (current_vfs && pci_vfs_assigned(pdev)) {
5293                 dev_err(&pdev->dev,
5294                         "Cannot modify SR-IOV while VFs are assigned\n");
5295                 return current_vfs;
5296         }
5297         /* Note that the upper-level code ensures that we're never called with
5298          * a non-zero "num_vfs" when we already have VFs instantiated.  But
5299          * it never hurts to code defensively.
5300          */
5301         if (num_vfs != 0 && current_vfs != 0)
5302                 return -EBUSY;
5303
5304         /* Nothing to do for no change. */
5305         if (num_vfs == current_vfs)
5306                 return num_vfs;
5307
5308         /* Disable SRIOV when zero is passed. */
5309         if (!num_vfs) {
5310                 pci_disable_sriov(pdev);
5311                 /* free VF Management Interface */
5312                 unregister_netdev(adap->port[0]);
5313                 free_netdev(adap->port[0]);
5314                 adap->port[0] = NULL;
5315
5316                 /* free VF resources */
5317                 adap->num_vfs = 0;
5318                 kfree(adap->vfinfo);
5319                 adap->vfinfo = NULL;
5320                 return 0;
5321         }
5322
5323         if (!current_vfs) {
5324                 struct fw_pfvf_cmd port_cmd, port_rpl;
5325                 struct net_device *netdev;
5326                 unsigned int pmask, port;
5327                 struct pci_dev *pbridge;
5328                 struct port_info *pi;
5329                 char name[IFNAMSIZ];
5330                 u32 devcap2;
5331                 u16 flags;
5332                 int pos;
5333
5334                 /* If we want to instantiate Virtual Functions, then our
5335                  * parent bridge's PCI-E needs to support Alternative Routing
5336                  * ID (ARI) because our VFs will show up at function offset 8
5337                  * and above.
5338                  */
5339                 pbridge = pdev->bus->self;
5340                 pos = pci_find_capability(pbridge, PCI_CAP_ID_EXP);
5341                 pci_read_config_word(pbridge, pos + PCI_EXP_FLAGS, &flags);
5342                 pci_read_config_dword(pbridge, pos + PCI_EXP_DEVCAP2, &devcap2);
5343
5344                 if ((flags & PCI_EXP_FLAGS_VERS) < 2 ||
5345                     !(devcap2 & PCI_EXP_DEVCAP2_ARI)) {
5346                         /* Our parent bridge does not support ARI so issue a
5347                          * warning and skip instantiating the VFs.  They
5348                          * won't be reachable.
5349                          */
5350                         dev_warn(&pdev->dev, "Parent bridge %02x:%02x.%x doesn't support ARI; can't instantiate Virtual Functions\n",
5351                                  pbridge->bus->number, PCI_SLOT(pbridge->devfn),
5352                                  PCI_FUNC(pbridge->devfn));
5353                         return -ENOTSUPP;
5354                 }
5355                 memset(&port_cmd, 0, sizeof(port_cmd));
5356                 port_cmd.op_to_vfn = cpu_to_be32(FW_CMD_OP_V(FW_PFVF_CMD) |
5357                                                  FW_CMD_REQUEST_F |
5358                                                  FW_CMD_READ_F |
5359                                                  FW_PFVF_CMD_PFN_V(adap->pf) |
5360                                                  FW_PFVF_CMD_VFN_V(0));
5361                 port_cmd.retval_len16 = cpu_to_be32(FW_LEN16(port_cmd));
5362                 err = t4_wr_mbox(adap, adap->mbox, &port_cmd, sizeof(port_cmd),
5363                                  &port_rpl);
5364                 if (err)
5365                         return err;
5366                 pmask = FW_PFVF_CMD_PMASK_G(be32_to_cpu(port_rpl.type_to_neq));
5367                 port = ffs(pmask) - 1;
5368                 /* Allocate VF Management Interface. */
5369                 snprintf(name, IFNAMSIZ, "mgmtpf%d,%d", adap->adap_idx,
5370                          adap->pf);
5371                 netdev = alloc_netdev(sizeof(struct port_info),
5372                                       name, NET_NAME_UNKNOWN, cxgb4_mgmt_setup);
5373                 if (!netdev)
5374                         return -ENOMEM;
5375
5376                 pi = netdev_priv(netdev);
5377                 pi->adapter = adap;
5378                 pi->lport = port;
5379                 pi->tx_chan = port;
5380                 SET_NETDEV_DEV(netdev, &pdev->dev);
5381
5382                 adap->port[0] = netdev;
5383                 pi->port_id = 0;
5384
5385                 err = register_netdev(adap->port[0]);
5386                 if (err) {
5387                         pr_info("Unable to register VF mgmt netdev %s\n", name);
5388                         free_netdev(adap->port[0]);
5389                         adap->port[0] = NULL;
5390                         return err;
5391                 }
5392                 /* Allocate and set up VF Information. */
5393                 adap->vfinfo = kcalloc(pci_sriov_get_totalvfs(pdev),
5394                                        sizeof(struct vf_info), GFP_KERNEL);
5395                 if (!adap->vfinfo) {
5396                         unregister_netdev(adap->port[0]);
5397                         free_netdev(adap->port[0]);
5398                         adap->port[0] = NULL;
5399                         return -ENOMEM;
5400                 }
5401                 cxgb4_mgmt_fill_vf_station_mac_addr(adap);
5402         }
5403         /* Instantiate the requested number of VFs. */
5404         err = pci_enable_sriov(pdev, num_vfs);
5405         if (err) {
5406                 pr_info("Unable to instantiate %d VFs\n", num_vfs);
5407                 if (!current_vfs) {
5408                         unregister_netdev(adap->port[0]);
5409                         free_netdev(adap->port[0]);
5410                         adap->port[0] = NULL;
5411                         kfree(adap->vfinfo);
5412                         adap->vfinfo = NULL;
5413                 }
5414                 return err;
5415         }
5416
5417         adap->num_vfs = num_vfs;
5418         return num_vfs;
5419 }
5420 #endif /* CONFIG_PCI_IOV */
5421
5422 static int init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
5423 {
5424         struct net_device *netdev;
5425         struct adapter *adapter;
5426         static int adap_idx = 1;
5427         int s_qpp, qpp, num_seg;
5428         struct port_info *pi;
5429         bool highdma = false;
5430         enum chip_type chip;
5431         void __iomem *regs;
5432         int func, chip_ver;
5433         u16 device_id;
5434         int i, err;
5435         u32 whoami;
5436
5437         printk_once(KERN_INFO "%s - version %s\n", DRV_DESC, DRV_VERSION);
5438
5439         err = pci_request_regions(pdev, KBUILD_MODNAME);
5440         if (err) {
5441                 /* Just info, some other driver may have claimed the device. */
5442                 dev_info(&pdev->dev, "cannot obtain PCI resources\n");
5443                 return err;
5444         }
5445
5446         err = pci_enable_device(pdev);
5447         if (err) {
5448                 dev_err(&pdev->dev, "cannot enable PCI device\n");
5449                 goto out_release_regions;
5450         }
5451
5452         regs = pci_ioremap_bar(pdev, 0);
5453         if (!regs) {
5454                 dev_err(&pdev->dev, "cannot map device registers\n");
5455                 err = -ENOMEM;
5456                 goto out_disable_device;
5457         }
5458
5459         adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
5460         if (!adapter) {
5461                 err = -ENOMEM;
5462                 goto out_unmap_bar0;
5463         }
5464
5465         adapter->regs = regs;
5466         err = t4_wait_dev_ready(regs);
5467         if (err < 0)
5468                 goto out_free_adapter;
5469
5470         /* We control everything through one PF */
5471         whoami = t4_read_reg(adapter, PL_WHOAMI_A);
5472         pci_read_config_word(pdev, PCI_DEVICE_ID, &device_id);
5473         chip = t4_get_chip_type(adapter, CHELSIO_PCI_ID_VER(device_id));
5474         if (chip < 0) {
5475                 dev_err(&pdev->dev, "Device %d is not supported\n", device_id);
5476                 err = chip;
5477                 goto out_free_adapter;
5478         }
5479         chip_ver = CHELSIO_CHIP_VERSION(chip);
5480         func = chip_ver <= CHELSIO_T5 ?
5481                SOURCEPF_G(whoami) : T6_SOURCEPF_G(whoami);
5482
5483         adapter->pdev = pdev;
5484         adapter->pdev_dev = &pdev->dev;
5485         adapter->name = pci_name(pdev);
5486         adapter->mbox = func;
5487         adapter->pf = func;
5488         adapter->params.chip = chip;
5489         adapter->adap_idx = adap_idx;
5490         adapter->msg_enable = DFLT_MSG_ENABLE;
5491         adapter->mbox_log = kzalloc(sizeof(*adapter->mbox_log) +
5492                                     (sizeof(struct mbox_cmd) *
5493                                      T4_OS_LOG_MBOX_CMDS),
5494                                     GFP_KERNEL);
5495         if (!adapter->mbox_log) {
5496                 err = -ENOMEM;
5497                 goto out_free_adapter;
5498         }
5499         spin_lock_init(&adapter->mbox_lock);
5500         INIT_LIST_HEAD(&adapter->mlist.list);
5501         adapter->mbox_log->size = T4_OS_LOG_MBOX_CMDS;
5502         pci_set_drvdata(pdev, adapter);
5503
5504         if (func != ent->driver_data) {
5505                 pci_disable_device(pdev);
5506                 pci_save_state(pdev);        /* to restore SR-IOV later */
5507                 return 0;
5508         }
5509
5510         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64))) {
5511                 highdma = true;
5512                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
5513                 if (err) {
5514                         dev_err(&pdev->dev, "unable to obtain 64-bit DMA for "
5515                                 "coherent allocations\n");
5516                         goto out_free_adapter;
5517                 }
5518         } else {
5519                 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
5520                 if (err) {
5521                         dev_err(&pdev->dev, "no usable DMA configuration\n");
5522                         goto out_free_adapter;
5523                 }
5524         }
5525
5526         pci_enable_pcie_error_reporting(pdev);
5527         pci_set_master(pdev);
5528         pci_save_state(pdev);
5529         adap_idx++;
5530         adapter->workq = create_singlethread_workqueue("cxgb4");
5531         if (!adapter->workq) {
5532                 err = -ENOMEM;
5533                 goto out_free_adapter;
5534         }
5535
5536         /* PCI device has been enabled */
5537         adapter->flags |= DEV_ENABLED;
5538         memset(adapter->chan_map, 0xff, sizeof(adapter->chan_map));
5539
5540         /* If possible, we use PCIe Relaxed Ordering Attribute to deliver
5541          * Ingress Packet Data to Free List Buffers in order to allow for
5542          * chipset performance optimizations between the Root Complex and
5543          * Memory Controllers.  (Messages to the associated Ingress Queue
5544          * notifying new Packet Placement in the Free Lists Buffers will be
5545          * send without the Relaxed Ordering Attribute thus guaranteeing that
5546          * all preceding PCIe Transaction Layer Packets will be processed
5547          * first.)  But some Root Complexes have various issues with Upstream
5548          * Transaction Layer Packets with the Relaxed Ordering Attribute set.
5549          * The PCIe devices which under the Root Complexes will be cleared the
5550          * Relaxed Ordering bit in the configuration space, So we check our
5551          * PCIe configuration space to see if it's flagged with advice against
5552          * using Relaxed Ordering.
5553          */
5554         if (!pcie_relaxed_ordering_enabled(pdev))
5555                 adapter->flags |= ROOT_NO_RELAXED_ORDERING;
5556
5557         spin_lock_init(&adapter->stats_lock);
5558         spin_lock_init(&adapter->tid_release_lock);
5559         spin_lock_init(&adapter->win0_lock);
5560
5561         INIT_WORK(&adapter->tid_release_task, process_tid_release_list);
5562         INIT_WORK(&adapter->db_full_task, process_db_full);
5563         INIT_WORK(&adapter->db_drop_task, process_db_drop);
5564         INIT_WORK(&adapter->fatal_err_notify_task, notify_fatal_err);
5565
5566         err = t4_prep_adapter(adapter);
5567         if (err)
5568                 goto out_free_adapter;
5569
5570         if (is_kdump_kernel()) {
5571                 /* Collect hardware state and append to /proc/vmcore */
5572                 err = cxgb4_cudbg_vmcore_add_dump(adapter);
5573                 if (err) {
5574                         dev_warn(adapter->pdev_dev,
5575                                  "Fail collecting vmcore device dump, err: %d. Continuing\n",
5576                                  err);
5577                         err = 0;
5578                 }
5579         }
5580
5581         if (!is_t4(adapter->params.chip)) {
5582                 s_qpp = (QUEUESPERPAGEPF0_S +
5583                         (QUEUESPERPAGEPF1_S - QUEUESPERPAGEPF0_S) *
5584                         adapter->pf);
5585                 qpp = 1 << QUEUESPERPAGEPF0_G(t4_read_reg(adapter,
5586                       SGE_EGRESS_QUEUES_PER_PAGE_PF_A) >> s_qpp);
5587                 num_seg = PAGE_SIZE / SEGMENT_SIZE;
5588
5589                 /* Each segment size is 128B. Write coalescing is enabled only
5590                  * when SGE_EGRESS_QUEUES_PER_PAGE_PF reg value for the
5591                  * queue is less no of segments that can be accommodated in
5592                  * a page size.
5593                  */
5594                 if (qpp > num_seg) {
5595                         dev_err(&pdev->dev,
5596                                 "Incorrect number of egress queues per page\n");
5597                         err = -EINVAL;
5598                         goto out_free_adapter;
5599                 }
5600                 adapter->bar2 = ioremap_wc(pci_resource_start(pdev, 2),
5601                 pci_resource_len(pdev, 2));
5602                 if (!adapter->bar2) {
5603                         dev_err(&pdev->dev, "cannot map device bar2 region\n");
5604                         err = -ENOMEM;
5605                         goto out_free_adapter;
5606                 }
5607         }
5608
5609         setup_memwin(adapter);
5610         err = adap_init0(adapter);
5611 #ifdef CONFIG_DEBUG_FS
5612         bitmap_zero(adapter->sge.blocked_fl, adapter->sge.egr_sz);
5613 #endif
5614         setup_memwin_rdma(adapter);
5615         if (err)
5616                 goto out_unmap_bar;
5617
5618         /* configure SGE_STAT_CFG_A to read WC stats */
5619         if (!is_t4(adapter->params.chip))
5620                 t4_write_reg(adapter, SGE_STAT_CFG_A, STATSOURCE_T5_V(7) |
5621                              (is_t5(adapter->params.chip) ? STATMODE_V(0) :
5622                               T6_STATMODE_V(0)));
5623
5624         for_each_port(adapter, i) {
5625                 netdev = alloc_etherdev_mq(sizeof(struct port_info),
5626                                            MAX_ETH_QSETS);
5627                 if (!netdev) {
5628                         err = -ENOMEM;
5629                         goto out_free_dev;
5630                 }
5631
5632                 SET_NETDEV_DEV(netdev, &pdev->dev);
5633
5634                 adapter->port[i] = netdev;
5635                 pi = netdev_priv(netdev);
5636                 pi->adapter = adapter;
5637                 pi->xact_addr_filt = -1;
5638                 pi->port_id = i;
5639                 netdev->irq = pdev->irq;
5640
5641                 netdev->hw_features = NETIF_F_SG | TSO_FLAGS |
5642                         NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
5643                         NETIF_F_RXCSUM | NETIF_F_RXHASH |
5644                         NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
5645                         NETIF_F_HW_TC;
5646
5647                 if (chip_ver > CHELSIO_T5) {
5648                         netdev->hw_enc_features |= NETIF_F_IP_CSUM |
5649                                                    NETIF_F_IPV6_CSUM |
5650                                                    NETIF_F_RXCSUM |
5651                                                    NETIF_F_GSO_UDP_TUNNEL |
5652                                                    NETIF_F_TSO | NETIF_F_TSO6;
5653
5654                         netdev->hw_features |= NETIF_F_GSO_UDP_TUNNEL;
5655                 }
5656
5657                 if (highdma)
5658                         netdev->hw_features |= NETIF_F_HIGHDMA;
5659                 netdev->features |= netdev->hw_features;
5660                 netdev->vlan_features = netdev->features & VLAN_FEAT;
5661
5662                 netdev->priv_flags |= IFF_UNICAST_FLT;
5663
5664                 /* MTU range: 81 - 9600 */
5665                 netdev->min_mtu = 81;              /* accommodate SACK */
5666                 netdev->max_mtu = MAX_MTU;
5667
5668                 netdev->netdev_ops = &cxgb4_netdev_ops;
5669 #ifdef CONFIG_CHELSIO_T4_DCB
5670                 netdev->dcbnl_ops = &cxgb4_dcb_ops;
5671                 cxgb4_dcb_state_init(netdev);
5672                 cxgb4_dcb_version_init(netdev);
5673 #endif
5674                 cxgb4_set_ethtool_ops(netdev);
5675         }
5676
5677         cxgb4_init_ethtool_dump(adapter);
5678
5679         pci_set_drvdata(pdev, adapter);
5680
5681         if (adapter->flags & FW_OK) {
5682                 err = t4_port_init(adapter, func, func, 0);
5683                 if (err)
5684                         goto out_free_dev;
5685         } else if (adapter->params.nports == 1) {
5686                 /* If we don't have a connection to the firmware -- possibly
5687                  * because of an error -- grab the raw VPD parameters so we
5688                  * can set the proper MAC Address on the debug network
5689                  * interface that we've created.
5690                  */
5691                 u8 hw_addr[ETH_ALEN];
5692                 u8 *na = adapter->params.vpd.na;
5693
5694                 err = t4_get_raw_vpd_params(adapter, &adapter->params.vpd);
5695                 if (!err) {
5696                         for (i = 0; i < ETH_ALEN; i++)
5697                                 hw_addr[i] = (hex2val(na[2 * i + 0]) * 16 +
5698                                               hex2val(na[2 * i + 1]));
5699                         t4_set_hw_addr(adapter, 0, hw_addr);
5700                 }
5701         }
5702
5703         if (!(adapter->flags & FW_OK))
5704                 goto fw_attach_fail;
5705
5706         /* Configure queues and allocate tables now, they can be needed as
5707          * soon as the first register_netdev completes.
5708          */
5709         err = cfg_queues(adapter);
5710         if (err)
5711                 goto out_free_dev;
5712
5713         adapter->smt = t4_init_smt();
5714         if (!adapter->smt) {
5715                 /* We tolerate a lack of SMT, giving up some functionality */
5716                 dev_warn(&pdev->dev, "could not allocate SMT, continuing\n");
5717         }
5718
5719         adapter->l2t = t4_init_l2t(adapter->l2t_start, adapter->l2t_end);
5720         if (!adapter->l2t) {
5721                 /* We tolerate a lack of L2T, giving up some functionality */
5722                 dev_warn(&pdev->dev, "could not allocate L2T, continuing\n");
5723                 adapter->params.offload = 0;
5724         }
5725
5726         adapter->mps_encap = kvcalloc(adapter->params.arch.mps_tcam_size,
5727                                       sizeof(struct mps_encap_entry),
5728                                       GFP_KERNEL);
5729         if (!adapter->mps_encap)
5730                 dev_warn(&pdev->dev, "could not allocate MPS Encap entries, continuing\n");
5731
5732 #if IS_ENABLED(CONFIG_IPV6)
5733         if (chip_ver <= CHELSIO_T5 &&
5734             (!(t4_read_reg(adapter, LE_DB_CONFIG_A) & ASLIPCOMPEN_F))) {
5735                 /* CLIP functionality is not present in hardware,
5736                  * hence disable all offload features
5737                  */
5738                 dev_warn(&pdev->dev,
5739                          "CLIP not enabled in hardware, continuing\n");
5740                 adapter->params.offload = 0;
5741         } else {
5742                 adapter->clipt = t4_init_clip_tbl(adapter->clipt_start,
5743                                                   adapter->clipt_end);
5744                 if (!adapter->clipt) {
5745                         /* We tolerate a lack of clip_table, giving up
5746                          * some functionality
5747                          */
5748                         dev_warn(&pdev->dev,
5749                                  "could not allocate Clip table, continuing\n");
5750                         adapter->params.offload = 0;
5751                 }
5752         }
5753 #endif
5754
5755         for_each_port(adapter, i) {
5756                 pi = adap2pinfo(adapter, i);
5757                 pi->sched_tbl = t4_init_sched(adapter->params.nsched_cls);
5758                 if (!pi->sched_tbl)
5759                         dev_warn(&pdev->dev,
5760                                  "could not activate scheduling on port %d\n",
5761                                  i);
5762         }
5763
5764         if (tid_init(&adapter->tids) < 0) {
5765                 dev_warn(&pdev->dev, "could not allocate TID table, "
5766                          "continuing\n");
5767                 adapter->params.offload = 0;
5768         } else {
5769                 adapter->tc_u32 = cxgb4_init_tc_u32(adapter);
5770                 if (!adapter->tc_u32)
5771                         dev_warn(&pdev->dev,
5772                                  "could not offload tc u32, continuing\n");
5773
5774                 if (cxgb4_init_tc_flower(adapter))
5775                         dev_warn(&pdev->dev,
5776                                  "could not offload tc flower, continuing\n");
5777         }
5778
5779         if (is_offload(adapter) || is_hashfilter(adapter)) {
5780                 if (t4_read_reg(adapter, LE_DB_CONFIG_A) & HASHEN_F) {
5781                         u32 hash_base, hash_reg;
5782
5783                         if (chip_ver <= CHELSIO_T5) {
5784                                 hash_reg = LE_DB_TID_HASHBASE_A;
5785                                 hash_base = t4_read_reg(adapter, hash_reg);
5786                                 adapter->tids.hash_base = hash_base / 4;
5787                         } else {
5788                                 hash_reg = T6_LE_DB_HASH_TID_BASE_A;
5789                                 hash_base = t4_read_reg(adapter, hash_reg);
5790                                 adapter->tids.hash_base = hash_base;
5791                         }
5792                 }
5793         }
5794
5795         /* See what interrupts we'll be using */
5796         if (msi > 1 && enable_msix(adapter) == 0)
5797                 adapter->flags |= USING_MSIX;
5798         else if (msi > 0 && pci_enable_msi(pdev) == 0) {
5799                 adapter->flags |= USING_MSI;
5800                 if (msi > 1)
5801                         free_msix_info(adapter);
5802         }
5803
5804         /* check for PCI Express bandwidth capabiltites */
5805         pcie_print_link_status(pdev);
5806
5807         err = init_rss(adapter);
5808         if (err)
5809                 goto out_free_dev;
5810
5811         err = setup_fw_sge_queues(adapter);
5812         if (err) {
5813                 dev_err(adapter->pdev_dev,
5814                         "FW sge queue allocation failed, err %d", err);
5815                 goto out_free_dev;
5816         }
5817
5818 fw_attach_fail:
5819         /*
5820          * The card is now ready to go.  If any errors occur during device
5821          * registration we do not fail the whole card but rather proceed only
5822          * with the ports we manage to register successfully.  However we must
5823          * register at least one net device.
5824          */
5825         for_each_port(adapter, i) {
5826                 pi = adap2pinfo(adapter, i);
5827                 adapter->port[i]->dev_port = pi->lport;
5828                 netif_set_real_num_tx_queues(adapter->port[i], pi->nqsets);
5829                 netif_set_real_num_rx_queues(adapter->port[i], pi->nqsets);
5830
5831                 netif_carrier_off(adapter->port[i]);
5832
5833                 err = register_netdev(adapter->port[i]);
5834                 if (err)
5835                         break;
5836                 adapter->chan_map[pi->tx_chan] = i;
5837                 print_port_info(adapter->port[i]);
5838         }
5839         if (i == 0) {
5840                 dev_err(&pdev->dev, "could not register any net devices\n");
5841                 goto out_free_dev;
5842         }
5843         if (err) {
5844                 dev_warn(&pdev->dev, "only %d net devices registered\n", i);
5845                 err = 0;
5846         }
5847
5848         if (cxgb4_debugfs_root) {
5849                 adapter->debugfs_root = debugfs_create_dir(pci_name(pdev),
5850                                                            cxgb4_debugfs_root);
5851                 setup_debugfs(adapter);
5852         }
5853
5854         /* PCIe EEH recovery on powerpc platforms needs fundamental reset */
5855         pdev->needs_freset = 1;
5856
5857         if (is_uld(adapter)) {
5858                 mutex_lock(&uld_mutex);
5859                 list_add_tail(&adapter->list_node, &adapter_list);
5860                 mutex_unlock(&uld_mutex);
5861         }
5862
5863         if (!is_t4(adapter->params.chip))
5864                 cxgb4_ptp_init(adapter);
5865
5866         if (IS_REACHABLE(CONFIG_THERMAL) &&
5867             !is_t4(adapter->params.chip) && (adapter->flags & FW_OK))
5868                 cxgb4_thermal_init(adapter);
5869
5870         print_adapter_info(adapter);
5871         return 0;
5872
5873  out_free_dev:
5874         t4_free_sge_resources(adapter);
5875         free_some_resources(adapter);
5876         if (adapter->flags & USING_MSIX)
5877                 free_msix_info(adapter);
5878         if (adapter->num_uld || adapter->num_ofld_uld)
5879                 t4_uld_mem_free(adapter);
5880  out_unmap_bar:
5881         if (!is_t4(adapter->params.chip))
5882                 iounmap(adapter->bar2);
5883  out_free_adapter:
5884         if (adapter->workq)
5885                 destroy_workqueue(adapter->workq);
5886
5887         kfree(adapter->mbox_log);
5888         kfree(adapter);
5889  out_unmap_bar0:
5890         iounmap(regs);
5891  out_disable_device:
5892         pci_disable_pcie_error_reporting(pdev);
5893         pci_disable_device(pdev);
5894  out_release_regions:
5895         pci_release_regions(pdev);
5896         return err;
5897 }
5898
5899 static void remove_one(struct pci_dev *pdev)
5900 {
5901         struct adapter *adapter = pci_get_drvdata(pdev);
5902
5903         if (!adapter) {
5904                 pci_release_regions(pdev);
5905                 return;
5906         }
5907
5908         adapter->flags |= SHUTTING_DOWN;
5909
5910         if (adapter->pf == 4) {
5911                 int i;
5912
5913                 /* Tear down per-adapter Work Queue first since it can contain
5914                  * references to our adapter data structure.
5915                  */
5916                 destroy_workqueue(adapter->workq);
5917
5918                 if (is_uld(adapter)) {
5919                         detach_ulds(adapter);
5920                         t4_uld_clean_up(adapter);
5921                 }
5922
5923                 adap_free_hma_mem(adapter);
5924
5925                 disable_interrupts(adapter);
5926
5927                 for_each_port(adapter, i)
5928                         if (adapter->port[i]->reg_state == NETREG_REGISTERED)
5929                                 unregister_netdev(adapter->port[i]);
5930
5931                 debugfs_remove_recursive(adapter->debugfs_root);
5932
5933                 if (!is_t4(adapter->params.chip))
5934                         cxgb4_ptp_stop(adapter);
5935                 if (IS_REACHABLE(CONFIG_THERMAL))
5936                         cxgb4_thermal_remove(adapter);
5937
5938                 /* If we allocated filters, free up state associated with any
5939                  * valid filters ...
5940                  */
5941                 clear_all_filters(adapter);
5942
5943                 if (adapter->flags & FULL_INIT_DONE)
5944                         cxgb_down(adapter);
5945
5946                 if (adapter->flags & USING_MSIX)
5947                         free_msix_info(adapter);
5948                 if (adapter->num_uld || adapter->num_ofld_uld)
5949                         t4_uld_mem_free(adapter);
5950                 free_some_resources(adapter);
5951 #if IS_ENABLED(CONFIG_IPV6)
5952                 t4_cleanup_clip_tbl(adapter);
5953 #endif
5954                 if (!is_t4(adapter->params.chip))
5955                         iounmap(adapter->bar2);
5956         }
5957 #ifdef CONFIG_PCI_IOV
5958         else {
5959                 cxgb4_iov_configure(adapter->pdev, 0);
5960         }
5961 #endif
5962         iounmap(adapter->regs);
5963         pci_disable_pcie_error_reporting(pdev);
5964         if ((adapter->flags & DEV_ENABLED)) {
5965                 pci_disable_device(pdev);
5966                 adapter->flags &= ~DEV_ENABLED;
5967         }
5968         pci_release_regions(pdev);
5969         kfree(adapter->mbox_log);
5970         synchronize_rcu();
5971         kfree(adapter);
5972 }
5973
5974 /* "Shutdown" quiesces the device, stopping Ingress Packet and Interrupt
5975  * delivery.  This is essentially a stripped down version of the PCI remove()
5976  * function where we do the minimal amount of work necessary to shutdown any
5977  * further activity.
5978  */
5979 static void shutdown_one(struct pci_dev *pdev)
5980 {
5981         struct adapter *adapter = pci_get_drvdata(pdev);
5982
5983         /* As with remove_one() above (see extended comment), we only want do
5984          * do cleanup on PCI Devices which went all the way through init_one()
5985          * ...
5986          */
5987         if (!adapter) {
5988                 pci_release_regions(pdev);
5989                 return;
5990         }
5991
5992         adapter->flags |= SHUTTING_DOWN;
5993
5994         if (adapter->pf == 4) {
5995                 int i;
5996
5997                 for_each_port(adapter, i)
5998                         if (adapter->port[i]->reg_state == NETREG_REGISTERED)
5999                                 cxgb_close(adapter->port[i]);
6000
6001                 if (is_uld(adapter)) {
6002                         detach_ulds(adapter);
6003                         t4_uld_clean_up(adapter);
6004                 }
6005
6006                 disable_interrupts(adapter);
6007                 disable_msi(adapter);
6008
6009                 t4_sge_stop(adapter);
6010                 if (adapter->flags & FW_OK)
6011                         t4_fw_bye(adapter, adapter->mbox);
6012         }
6013 }
6014
6015 static struct pci_driver cxgb4_driver = {
6016         .name     = KBUILD_MODNAME,
6017         .id_table = cxgb4_pci_tbl,
6018         .probe    = init_one,
6019         .remove   = remove_one,
6020         .shutdown = shutdown_one,
6021 #ifdef CONFIG_PCI_IOV
6022         .sriov_configure = cxgb4_iov_configure,
6023 #endif
6024         .err_handler = &cxgb4_eeh,
6025 };
6026
6027 static int __init cxgb4_init_module(void)
6028 {
6029         int ret;
6030
6031         /* Debugfs support is optional, just warn if this fails */
6032         cxgb4_debugfs_root = debugfs_create_dir(KBUILD_MODNAME, NULL);
6033         if (!cxgb4_debugfs_root)
6034                 pr_warn("could not create debugfs entry, continuing\n");
6035
6036         ret = pci_register_driver(&cxgb4_driver);
6037         if (ret < 0)
6038                 debugfs_remove(cxgb4_debugfs_root);
6039
6040 #if IS_ENABLED(CONFIG_IPV6)
6041         if (!inet6addr_registered) {
6042                 register_inet6addr_notifier(&cxgb4_inet6addr_notifier);
6043                 inet6addr_registered = true;
6044         }
6045 #endif
6046
6047         return ret;
6048 }
6049
6050 static void __exit cxgb4_cleanup_module(void)
6051 {
6052 #if IS_ENABLED(CONFIG_IPV6)
6053         if (inet6addr_registered) {
6054                 unregister_inet6addr_notifier(&cxgb4_inet6addr_notifier);
6055                 inet6addr_registered = false;
6056         }
6057 #endif
6058         pci_unregister_driver(&cxgb4_driver);
6059         debugfs_remove(cxgb4_debugfs_root);  /* NULL ok */
6060 }
6061
6062 module_init(cxgb4_init_module);
6063 module_exit(cxgb4_cleanup_module);