drivers/net/ethernet/cisco/enic/enic_main.c

   1 /*
   2  * Copyright 2008-2010 Cisco Systems, Inc.  All rights reserved.
   3  * Copyright 2007 Nuova Systems, Inc.  All rights reserved.
   4  *
   5  * This program is free software; you may redistribute it and/or modify
   6  * it under the terms of the GNU General Public License as published by
   7  * the Free Software Foundation; version 2 of the License.
   8  *
   9  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  10  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  11  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  12  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  13  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  14  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  15  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  16  * SOFTWARE.
  17  *
  18  */
  19
  20 #include <linux/module.h>
  21 #include <linux/kernel.h>
  22 #include <linux/string.h>
  23 #include <linux/errno.h>
  24 #include <linux/types.h>
  25 #include <linux/init.h>
  26 #include <linux/interrupt.h>
  27 #include <linux/workqueue.h>
  28 #include <linux/pci.h>
  29 #include <linux/netdevice.h>
  30 #include <linux/etherdevice.h>
  31 #include <linux/if.h>
  32 #include <linux/if_ether.h>
  33 #include <linux/if_vlan.h>
  34 #include <linux/in.h>
  35 #include <linux/ip.h>
  36 #include <linux/ipv6.h>
  37 #include <linux/tcp.h>
  38 #include <linux/rtnetlink.h>
  39 #include <linux/prefetch.h>
  40 #include <net/ip6_checksum.h>
  41 #include <linux/ktime.h>
  42 #include <linux/numa.h>
  43 #ifdef CONFIG_RFS_ACCEL
  44 #include <linux/cpu_rmap.h>
  45 #endif
  46 #include <linux/crash_dump.h>
  47 #include <net/busy_poll.h>
  48 #include <net/vxlan.h>
  49 #include <net/netdev_queues.h>
  50
  51 #include "cq_enet_desc.h"
  52 #include "vnic_dev.h"
  53 #include "vnic_intr.h"
  54 #include "vnic_stats.h"
  55 #include "vnic_vic.h"
  56 #include "enic_res.h"
  57 #include "enic.h"
  58 #include "enic_dev.h"
  59 #include "enic_pp.h"
  60 #include "enic_clsf.h"
  61
  62 #define ENIC_NOTIFY_TIMER_PERIOD        (2 * HZ)
  63 #define WQ_ENET_MAX_DESC_LEN            (1 << WQ_ENET_LEN_BITS)
  64 #define MAX_TSO                         (1 << 16)
  65 #define ENIC_DESC_MAX_SPLITS            (MAX_TSO / WQ_ENET_MAX_DESC_LEN + 1)
  66
  67 #define PCI_DEVICE_ID_CISCO_VIC_ENET         0x0043  /* ethernet vnic */
  68 #define PCI_DEVICE_ID_CISCO_VIC_ENET_DYN     0x0044  /* enet dynamic vnic */
  69 #define PCI_DEVICE_ID_CISCO_VIC_ENET_VF      0x0071  /* enet SRIOV VF */
  70
  71 #define RX_COPYBREAK_DEFAULT            256
  72
  73 /* Supported devices */
  74 static const struct pci_device_id enic_id_table[] = {
  75         { PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET) },
  76         { PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET_DYN) },
  77         { PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET_VF) },
  78         { 0, }  /* end of table */
  79 };
  80
  81 MODULE_DESCRIPTION(DRV_DESCRIPTION);
  82 MODULE_AUTHOR("Scott Feldman <[email protected]>");
  83 MODULE_LICENSE("GPL");
  84 MODULE_DEVICE_TABLE(pci, enic_id_table);
  85
  86 #define ENIC_LARGE_PKT_THRESHOLD                1000
  87 #define ENIC_MAX_COALESCE_TIMERS                10
  88 /*  Interrupt moderation table, which will be used to decide the
  89  *  coalescing timer values
  90  *  {rx_rate in Mbps, mapping percentage of the range}
  91  */
  92 static struct enic_intr_mod_table mod_table[ENIC_MAX_COALESCE_TIMERS + 1] = {
  93         {4000,  0},
  94         {4400, 10},
  95         {5060, 20},
  96         {5230, 30},
  97         {5540, 40},
  98         {5820, 50},
  99         {6120, 60},
 100         {6435, 70},
 101         {6745, 80},
 102         {7000, 90},
 103         {0xFFFFFFFF, 100}
 104 };
 105
 106 /* This table helps the driver to pick different ranges for rx coalescing
 107  * timer depending on the link speed.
 108  */
 109 static struct enic_intr_mod_range mod_range[ENIC_MAX_LINK_SPEEDS] = {
 110         {0,  0}, /* 0  - 4  Gbps */
 111         {0,  3}, /* 4  - 10 Gbps */
 112         {3,  6}, /* 10+ Gbps */
 113 };
 114
 115 static void enic_init_affinity_hint(struct enic *enic)
 116 {
 117         int numa_node = dev_to_node(&enic->pdev->dev);
 118         int i;
 119
 120         for (i = 0; i < enic->intr_count; i++) {
 121                 if (enic_is_err_intr(enic, i) || enic_is_notify_intr(enic, i) ||
 122                     (cpumask_available(enic->msix[i].affinity_mask) &&
 123                      !cpumask_empty(enic->msix[i].affinity_mask)))
 124                         continue;
 125                 if (zalloc_cpumask_var(&enic->msix[i].affinity_mask,
 126                                        GFP_KERNEL))
 127                         cpumask_set_cpu(cpumask_local_spread(i, numa_node),
 128                                         enic->msix[i].affinity_mask);
 129         }
 130 }
 131
 132 static void enic_free_affinity_hint(struct enic *enic)
 133 {
 134         int i;
 135
 136         for (i = 0; i < enic->intr_count; i++) {
 137                 if (enic_is_err_intr(enic, i) || enic_is_notify_intr(enic, i))
 138                         continue;
 139                 free_cpumask_var(enic->msix[i].affinity_mask);
 140         }
 141 }
 142
 143 static void enic_set_affinity_hint(struct enic *enic)
 144 {
 145         int i;
 146         int err;
 147
 148         for (i = 0; i < enic->intr_count; i++) {
 149                 if (enic_is_err_intr(enic, i)           ||
 150                     enic_is_notify_intr(enic, i)        ||
 151                     !cpumask_available(enic->msix[i].affinity_mask) ||
 152                     cpumask_empty(enic->msix[i].affinity_mask))
 153                         continue;
 154                 err = irq_update_affinity_hint(enic->msix_entry[i].vector,
 155                                                enic->msix[i].affinity_mask);
 156                 if (err)
 157                         netdev_warn(enic->netdev, "irq_update_affinity_hint failed, err %d\n",
 158                                     err);
 159         }
 160
 161         for (i = 0; i < enic->wq_count; i++) {
 162                 int wq_intr = enic_msix_wq_intr(enic, i);
 163
 164                 if (cpumask_available(enic->msix[wq_intr].affinity_mask) &&
 165                     !cpumask_empty(enic->msix[wq_intr].affinity_mask))
 166                         netif_set_xps_queue(enic->netdev,
 167                                             enic->msix[wq_intr].affinity_mask,
 168                                             i);
 169         }
 170 }
 171
 172 static void enic_unset_affinity_hint(struct enic *enic)
 173 {
 174         int i;
 175
 176         for (i = 0; i < enic->intr_count; i++)
 177                 irq_update_affinity_hint(enic->msix_entry[i].vector, NULL);
 178 }
 179
 180 static int enic_udp_tunnel_set_port(struct net_device *netdev,
 181                                     unsigned int table, unsigned int entry,
 182                                     struct udp_tunnel_info *ti)
 183 {
 184         struct enic *enic = netdev_priv(netdev);
 185         int err;
 186
 187         spin_lock_bh(&enic->devcmd_lock);
 188
 189         err = vnic_dev_overlay_offload_cfg(enic->vdev,
 190                                            OVERLAY_CFG_VXLAN_PORT_UPDATE,
 191                                            ntohs(ti->port));
 192         if (err)
 193                 goto error;
 194
 195         err = vnic_dev_overlay_offload_ctrl(enic->vdev, OVERLAY_FEATURE_VXLAN,
 196                                             enic->vxlan.patch_level);
 197         if (err)
 198                 goto error;
 199
 200         enic->vxlan.vxlan_udp_port_number = ntohs(ti->port);
 201 error:
 202         spin_unlock_bh(&enic->devcmd_lock);
 203
 204         return err;
 205 }
 206
 207 static int enic_udp_tunnel_unset_port(struct net_device *netdev,
 208                                       unsigned int table, unsigned int entry,
 209                                       struct udp_tunnel_info *ti)
 210 {
 211         struct enic *enic = netdev_priv(netdev);
 212         int err;
 213
 214         spin_lock_bh(&enic->devcmd_lock);
 215
 216         err = vnic_dev_overlay_offload_ctrl(enic->vdev, OVERLAY_FEATURE_VXLAN,
 217                                             OVERLAY_OFFLOAD_DISABLE);
 218         if (err)
 219                 goto unlock;
 220
 221         enic->vxlan.vxlan_udp_port_number = 0;
 222
 223 unlock:
 224         spin_unlock_bh(&enic->devcmd_lock);
 225
 226         return err;
 227 }
 228
 229 static const struct udp_tunnel_nic_info enic_udp_tunnels = {
 230         .set_port       = enic_udp_tunnel_set_port,
 231         .unset_port     = enic_udp_tunnel_unset_port,
 232         .tables         = {
 233                 { .n_entries = 1, .tunnel_types = UDP_TUNNEL_TYPE_VXLAN, },
 234         },
 235 }, enic_udp_tunnels_v4 = {
 236         .set_port       = enic_udp_tunnel_set_port,
 237         .unset_port     = enic_udp_tunnel_unset_port,
 238         .flags          = UDP_TUNNEL_NIC_INFO_IPV4_ONLY,
 239         .tables         = {
 240                 { .n_entries = 1, .tunnel_types = UDP_TUNNEL_TYPE_VXLAN, },
 241         },
 242 };
 243
 244 static netdev_features_t enic_features_check(struct sk_buff *skb,
 245                                              struct net_device *dev,
 246                                              netdev_features_t features)
 247 {
 248         const struct ethhdr *eth = (struct ethhdr *)skb_inner_mac_header(skb);
 249         struct enic *enic = netdev_priv(dev);
 250         struct udphdr *udph;
 251         u16 port = 0;
 252         u8 proto;
 253
 254         if (!skb->encapsulation)
 255                 return features;
 256
 257         features = vxlan_features_check(skb, features);
 258
 259         switch (vlan_get_protocol(skb)) {
 260         case htons(ETH_P_IPV6):
 261                 if (!(enic->vxlan.flags & ENIC_VXLAN_OUTER_IPV6))
 262                         goto out;
 263                 proto = ipv6_hdr(skb)->nexthdr;
 264                 break;
 265         case htons(ETH_P_IP):
 266                 proto = ip_hdr(skb)->protocol;
 267                 break;
 268         default:
 269                 goto out;
 270         }
 271
 272         switch (eth->h_proto) {
 273         case ntohs(ETH_P_IPV6):
 274                 if (!(enic->vxlan.flags & ENIC_VXLAN_INNER_IPV6))
 275                         goto out;
 276                 fallthrough;
 277         case ntohs(ETH_P_IP):
 278                 break;
 279         default:
 280                 goto out;
 281         }
 282
 283
 284         if (proto == IPPROTO_UDP) {
 285                 udph = udp_hdr(skb);
 286                 port = be16_to_cpu(udph->dest);
 287         }
 288
 289         /* HW supports offload of only one UDP port. Remove CSUM and GSO MASK
 290          * for other UDP port tunnels
 291          */
 292         if (port  != enic->vxlan.vxlan_udp_port_number)
 293                 goto out;
 294
 295         return features;
 296
 297 out:
 298         return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
 299 }
 300
 301 int enic_is_dynamic(struct enic *enic)
 302 {
 303         return enic->pdev->device == PCI_DEVICE_ID_CISCO_VIC_ENET_DYN;
 304 }
 305
 306 int enic_sriov_enabled(struct enic *enic)
 307 {
 308         return (enic->priv_flags & ENIC_SRIOV_ENABLED) ? 1 : 0;
 309 }
 310
 311 static int enic_is_sriov_vf(struct enic *enic)
 312 {
 313         return enic->pdev->device == PCI_DEVICE_ID_CISCO_VIC_ENET_VF;
 314 }
 315
 316 int enic_is_valid_vf(struct enic *enic, int vf)
 317 {
 318 #ifdef CONFIG_PCI_IOV
 319         return vf >= 0 && vf < enic->num_vfs;
 320 #else
 321         return 0;
 322 #endif
 323 }
 324
 325 static void enic_free_wq_buf(struct vnic_wq *wq, struct vnic_wq_buf *buf)
 326 {
 327         struct enic *enic = vnic_dev_priv(wq->vdev);
 328
 329         if (buf->sop)
 330                 dma_unmap_single(&enic->pdev->dev, buf->dma_addr, buf->len,
 331                                  DMA_TO_DEVICE);
 332         else
 333                 dma_unmap_page(&enic->pdev->dev, buf->dma_addr, buf->len,
 334                                DMA_TO_DEVICE);
 335
 336         if (buf->os_buf)
 337                 dev_kfree_skb_any(buf->os_buf);
 338 }
 339
 340 static void enic_wq_free_buf(struct vnic_wq *wq,
 341         struct cq_desc *cq_desc, struct vnic_wq_buf *buf, void *opaque)
 342 {
 343         struct enic *enic = vnic_dev_priv(wq->vdev);
 344
 345         enic->wq[wq->index].stats.cq_work++;
 346         enic->wq[wq->index].stats.cq_bytes += buf->len;
 347         enic_free_wq_buf(wq, buf);
 348 }
 349
 350 static int enic_wq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
 351         u8 type, u16 q_number, u16 completed_index, void *opaque)
 352 {
 353         struct enic *enic = vnic_dev_priv(vdev);
 354
 355         spin_lock(&enic->wq[q_number].lock);
 356
 357         vnic_wq_service(&enic->wq[q_number].vwq, cq_desc,
 358                 completed_index, enic_wq_free_buf,
 359                 opaque);
 360
 361         if (netif_tx_queue_stopped(netdev_get_tx_queue(enic->netdev, q_number)) &&
 362             vnic_wq_desc_avail(&enic->wq[q_number].vwq) >=
 363             (MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS)) {
 364                 netif_wake_subqueue(enic->netdev, q_number);
 365                 enic->wq[q_number].stats.wake++;
 366         }
 367
 368         spin_unlock(&enic->wq[q_number].lock);
 369
 370         return 0;
 371 }
 372
 373 static bool enic_log_q_error(struct enic *enic)
 374 {
 375         unsigned int i;
 376         u32 error_status;
 377         bool err = false;
 378
 379         for (i = 0; i < enic->wq_count; i++) {
 380                 error_status = vnic_wq_error_status(&enic->wq[i].vwq);
 381                 err |= error_status;
 382                 if (error_status)
 383                         netdev_err(enic->netdev, "WQ[%d] error_status %d\n",
 384                                 i, error_status);
 385         }
 386
 387         for (i = 0; i < enic->rq_count; i++) {
 388                 error_status = vnic_rq_error_status(&enic->rq[i].vrq);
 389                 err |= error_status;
 390                 if (error_status)
 391                         netdev_err(enic->netdev, "RQ[%d] error_status %d\n",
 392                                 i, error_status);
 393         }
 394
 395         return err;
 396 }
 397
 398 static void enic_msglvl_check(struct enic *enic)
 399 {
 400         u32 msg_enable = vnic_dev_msg_lvl(enic->vdev);
 401
 402         if (msg_enable != enic->msg_enable) {
 403                 netdev_info(enic->netdev, "msg lvl changed from 0x%x to 0x%x\n",
 404                         enic->msg_enable, msg_enable);
 405                 enic->msg_enable = msg_enable;
 406         }
 407 }
 408
 409 static void enic_mtu_check(struct enic *enic)
 410 {
 411         u32 mtu = vnic_dev_mtu(enic->vdev);
 412         struct net_device *netdev = enic->netdev;
 413
 414         if (mtu && mtu != enic->port_mtu) {
 415                 enic->port_mtu = mtu;
 416                 if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic)) {
 417                         mtu = max_t(int, ENIC_MIN_MTU,
 418                                 min_t(int, ENIC_MAX_MTU, mtu));
 419                         if (mtu != netdev->mtu)
 420                                 schedule_work(&enic->change_mtu_work);
 421                 } else {
 422                         if (mtu < netdev->mtu)
 423                                 netdev_warn(netdev,
 424                                         "interface MTU (%d) set higher "
 425                                         "than switch port MTU (%d)\n",
 426                                         netdev->mtu, mtu);
 427                 }
 428         }
 429 }
 430
 431 static void enic_set_rx_coal_setting(struct enic *enic)
 432 {
 433         unsigned int speed;
 434         int index = -1;
 435         struct enic_rx_coal *rx_coal = &enic->rx_coalesce_setting;
 436
 437         /* 1. Read the link speed from fw
 438          * 2. Pick the default range for the speed
 439          * 3. Update it in enic->rx_coalesce_setting
 440          */
 441         speed = vnic_dev_port_speed(enic->vdev);
 442         if (speed > ENIC_LINK_SPEED_10G)
 443                 index = ENIC_LINK_40G_INDEX;
 444         else if (speed > ENIC_LINK_SPEED_4G)
 445                 index = ENIC_LINK_10G_INDEX;
 446         else
 447                 index = ENIC_LINK_4G_INDEX;
 448
 449         rx_coal->small_pkt_range_start = mod_range[index].small_pkt_range_start;
 450         rx_coal->large_pkt_range_start = mod_range[index].large_pkt_range_start;
 451         rx_coal->range_end = ENIC_RX_COALESCE_RANGE_END;
 452
 453         /* Start with the value provided by UCSM */
 454         for (index = 0; index < enic->rq_count; index++)
 455                 enic->cq[index].cur_rx_coal_timeval =
 456                                 enic->config.intr_timer_usec;
 457
 458         rx_coal->use_adaptive_rx_coalesce = 1;
 459 }
 460
 461 static void enic_link_check(struct enic *enic)
 462 {
 463         int link_status = vnic_dev_link_status(enic->vdev);
 464         int carrier_ok = netif_carrier_ok(enic->netdev);
 465
 466         if (link_status && !carrier_ok) {
 467                 netdev_info(enic->netdev, "Link UP\n");
 468                 netif_carrier_on(enic->netdev);
 469                 enic_set_rx_coal_setting(enic);
 470         } else if (!link_status && carrier_ok) {
 471                 netdev_info(enic->netdev, "Link DOWN\n");
 472                 netif_carrier_off(enic->netdev);
 473         }
 474 }
 475
 476 static void enic_notify_check(struct enic *enic)
 477 {
 478         enic_msglvl_check(enic);
 479         enic_mtu_check(enic);
 480         enic_link_check(enic);
 481 }
 482
 483 #define ENIC_TEST_INTR(pba, i) (pba & (1 << i))
 484
 485 static irqreturn_t enic_isr_legacy(int irq, void *data)
 486 {
 487         struct net_device *netdev = data;
 488         struct enic *enic = netdev_priv(netdev);
 489         unsigned int io_intr = ENIC_LEGACY_IO_INTR;
 490         unsigned int err_intr = ENIC_LEGACY_ERR_INTR;
 491         unsigned int notify_intr = ENIC_LEGACY_NOTIFY_INTR;
 492         u32 pba;
 493
 494         vnic_intr_mask(&enic->intr[io_intr]);
 495
 496         pba = vnic_intr_legacy_pba(enic->legacy_pba);
 497         if (!pba) {
 498                 vnic_intr_unmask(&enic->intr[io_intr]);
 499                 return IRQ_NONE;        /* not our interrupt */
 500         }
 501
 502         if (ENIC_TEST_INTR(pba, notify_intr)) {
 503                 enic_notify_check(enic);
 504                 vnic_intr_return_all_credits(&enic->intr[notify_intr]);
 505         }
 506
 507         if (ENIC_TEST_INTR(pba, err_intr)) {
 508                 vnic_intr_return_all_credits(&enic->intr[err_intr]);
 509                 enic_log_q_error(enic);
 510                 /* schedule recovery from WQ/RQ error */
 511                 schedule_work(&enic->reset);
 512                 return IRQ_HANDLED;
 513         }
 514
 515         if (ENIC_TEST_INTR(pba, io_intr))
 516                 napi_schedule_irqoff(&enic->napi[0]);
 517         else
 518                 vnic_intr_unmask(&enic->intr[io_intr]);
 519
 520         return IRQ_HANDLED;
 521 }
 522
 523 static irqreturn_t enic_isr_msi(int irq, void *data)
 524 {
 525         struct enic *enic = data;
 526
 527         /* With MSI, there is no sharing of interrupts, so this is
 528          * our interrupt and there is no need to ack it.  The device
 529          * is not providing per-vector masking, so the OS will not
 530          * write to PCI config space to mask/unmask the interrupt.
 531          * We're using mask_on_assertion for MSI, so the device
 532          * automatically masks the interrupt when the interrupt is
 533          * generated.  Later, when exiting polling, the interrupt
 534          * will be unmasked (see enic_poll).
 535          *
 536          * Also, the device uses the same PCIe Traffic Class (TC)
 537          * for Memory Write data and MSI, so there are no ordering
 538          * issues; the MSI will always arrive at the Root Complex
 539          * _after_ corresponding Memory Writes (i.e. descriptor
 540          * writes).
 541          */
 542
 543         napi_schedule_irqoff(&enic->napi[0]);
 544
 545         return IRQ_HANDLED;
 546 }
 547
 548 static irqreturn_t enic_isr_msix(int irq, void *data)
 549 {
 550         struct napi_struct *napi = data;
 551
 552         napi_schedule_irqoff(napi);
 553
 554         return IRQ_HANDLED;
 555 }
 556
 557 static irqreturn_t enic_isr_msix_err(int irq, void *data)
 558 {
 559         struct enic *enic = data;
 560         unsigned int intr = enic_msix_err_intr(enic);
 561
 562         vnic_intr_return_all_credits(&enic->intr[intr]);
 563
 564         if (enic_log_q_error(enic))
 565                 /* schedule recovery from WQ/RQ error */
 566                 schedule_work(&enic->reset);
 567
 568         return IRQ_HANDLED;
 569 }
 570
 571 static irqreturn_t enic_isr_msix_notify(int irq, void *data)
 572 {
 573         struct enic *enic = data;
 574         unsigned int intr = enic_msix_notify_intr(enic);
 575
 576         enic_notify_check(enic);
 577         vnic_intr_return_all_credits(&enic->intr[intr]);
 578
 579         return IRQ_HANDLED;
 580 }
 581
 582 static int enic_queue_wq_skb_cont(struct enic *enic, struct vnic_wq *wq,
 583                                   struct sk_buff *skb, unsigned int len_left,
 584                                   int loopback)
 585 {
 586         const skb_frag_t *frag;
 587         dma_addr_t dma_addr;
 588
 589         /* Queue additional data fragments */
 590         for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
 591                 len_left -= skb_frag_size(frag);
 592                 dma_addr = skb_frag_dma_map(&enic->pdev->dev, frag, 0,
 593                                             skb_frag_size(frag),
 594                                             DMA_TO_DEVICE);
 595                 if (unlikely(enic_dma_map_check(enic, dma_addr)))
 596                         return -ENOMEM;
 597                 enic_queue_wq_desc_cont(wq, skb, dma_addr, skb_frag_size(frag),
 598                                         (len_left == 0),        /* EOP? */
 599                                         loopback);
 600         }
 601
 602         return 0;
 603 }
 604
 605 static int enic_queue_wq_skb_vlan(struct enic *enic, struct vnic_wq *wq,
 606                                   struct sk_buff *skb, int vlan_tag_insert,
 607                                   unsigned int vlan_tag, int loopback)
 608 {
 609         unsigned int head_len = skb_headlen(skb);
 610         unsigned int len_left = skb->len - head_len;
 611         int eop = (len_left == 0);
 612         dma_addr_t dma_addr;
 613         int err = 0;
 614
 615         dma_addr = dma_map_single(&enic->pdev->dev, skb->data, head_len,
 616                                   DMA_TO_DEVICE);
 617         if (unlikely(enic_dma_map_check(enic, dma_addr)))
 618                 return -ENOMEM;
 619
 620         /* Queue the main skb fragment. The fragments are no larger
 621          * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less
 622          * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor
 623          * per fragment is queued.
 624          */
 625         enic_queue_wq_desc(wq, skb, dma_addr, head_len, vlan_tag_insert,
 626                            vlan_tag, eop, loopback);
 627
 628         if (!eop)
 629                 err = enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback);
 630
 631         /* The enic_queue_wq_desc() above does not do HW checksum */
 632         enic->wq[wq->index].stats.csum_none++;
 633         enic->wq[wq->index].stats.packets++;
 634         enic->wq[wq->index].stats.bytes += skb->len;
 635
 636         return err;
 637 }
 638
 639 static int enic_queue_wq_skb_csum_l4(struct enic *enic, struct vnic_wq *wq,
 640                                      struct sk_buff *skb, int vlan_tag_insert,
 641                                      unsigned int vlan_tag, int loopback)
 642 {
 643         unsigned int head_len = skb_headlen(skb);
 644         unsigned int len_left = skb->len - head_len;
 645         unsigned int hdr_len = skb_checksum_start_offset(skb);
 646         unsigned int csum_offset = hdr_len + skb->csum_offset;
 647         int eop = (len_left == 0);
 648         dma_addr_t dma_addr;
 649         int err = 0;
 650
 651         dma_addr = dma_map_single(&enic->pdev->dev, skb->data, head_len,
 652                                   DMA_TO_DEVICE);
 653         if (unlikely(enic_dma_map_check(enic, dma_addr)))
 654                 return -ENOMEM;
 655
 656         /* Queue the main skb fragment. The fragments are no larger
 657          * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less
 658          * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor
 659          * per fragment is queued.
 660          */
 661         enic_queue_wq_desc_csum_l4(wq, skb, dma_addr, head_len, csum_offset,
 662                                    hdr_len, vlan_tag_insert, vlan_tag, eop,
 663                                    loopback);
 664
 665         if (!eop)
 666                 err = enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback);
 667
 668         enic->wq[wq->index].stats.csum_partial++;
 669         enic->wq[wq->index].stats.packets++;
 670         enic->wq[wq->index].stats.bytes += skb->len;
 671
 672         return err;
 673 }
 674
 675 static void enic_preload_tcp_csum_encap(struct sk_buff *skb)
 676 {
 677         const struct ethhdr *eth = (struct ethhdr *)skb_inner_mac_header(skb);
 678
 679         switch (eth->h_proto) {
 680         case ntohs(ETH_P_IP):
 681                 inner_ip_hdr(skb)->check = 0;
 682                 inner_tcp_hdr(skb)->check =
 683                         ~csum_tcpudp_magic(inner_ip_hdr(skb)->saddr,
 684                                            inner_ip_hdr(skb)->daddr, 0,
 685                                            IPPROTO_TCP, 0);
 686                 break;
 687         case ntohs(ETH_P_IPV6):
 688                 inner_tcp_hdr(skb)->check =
 689                         ~csum_ipv6_magic(&inner_ipv6_hdr(skb)->saddr,
 690                                          &inner_ipv6_hdr(skb)->daddr, 0,
 691                                          IPPROTO_TCP, 0);
 692                 break;
 693         default:
 694                 WARN_ONCE(1, "Non ipv4/ipv6 inner pkt for encap offload");
 695                 break;
 696         }
 697 }
 698
 699 static void enic_preload_tcp_csum(struct sk_buff *skb)
 700 {
 701         /* Preload TCP csum field with IP pseudo hdr calculated
 702          * with IP length set to zero.  HW will later add in length
 703          * to each TCP segment resulting from the TSO.
 704          */
 705
 706         if (skb->protocol == cpu_to_be16(ETH_P_IP)) {
 707                 ip_hdr(skb)->check = 0;
 708                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
 709                         ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 710         } else if (skb->protocol == cpu_to_be16(ETH_P_IPV6)) {
 711                 tcp_v6_gso_csum_prep(skb);
 712         }
 713 }
 714
 715 static int enic_queue_wq_skb_tso(struct enic *enic, struct vnic_wq *wq,
 716                                  struct sk_buff *skb, unsigned int mss,
 717                                  int vlan_tag_insert, unsigned int vlan_tag,
 718                                  int loopback)
 719 {
 720         unsigned int frag_len_left = skb_headlen(skb);
 721         unsigned int len_left = skb->len - frag_len_left;
 722         int eop = (len_left == 0);
 723         unsigned int offset = 0;
 724         unsigned int hdr_len;
 725         dma_addr_t dma_addr;
 726         unsigned int pkts;
 727         unsigned int len;
 728         skb_frag_t *frag;
 729
 730         if (skb->encapsulation) {
 731                 hdr_len = skb_inner_tcp_all_headers(skb);
 732                 enic_preload_tcp_csum_encap(skb);
 733                 enic->wq[wq->index].stats.encap_tso++;
 734         } else {
 735                 hdr_len = skb_tcp_all_headers(skb);
 736                 enic_preload_tcp_csum(skb);
 737                 enic->wq[wq->index].stats.tso++;
 738         }
 739
 740         /* Queue WQ_ENET_MAX_DESC_LEN length descriptors
 741          * for the main skb fragment
 742          */
 743         while (frag_len_left) {
 744                 len = min(frag_len_left, (unsigned int)WQ_ENET_MAX_DESC_LEN);
 745                 dma_addr = dma_map_single(&enic->pdev->dev,
 746                                           skb->data + offset, len,
 747                                           DMA_TO_DEVICE);
 748                 if (unlikely(enic_dma_map_check(enic, dma_addr)))
 749                         return -ENOMEM;
 750                 enic_queue_wq_desc_tso(wq, skb, dma_addr, len, mss, hdr_len,
 751                                        vlan_tag_insert, vlan_tag,
 752                                        eop && (len == frag_len_left), loopback);
 753                 frag_len_left -= len;
 754                 offset += len;
 755         }
 756
 757         if (eop)
 758                 goto tso_out_stats;
 759
 760         /* Queue WQ_ENET_MAX_DESC_LEN length descriptors
 761          * for additional data fragments
 762          */
 763         for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
 764                 len_left -= skb_frag_size(frag);
 765                 frag_len_left = skb_frag_size(frag);
 766                 offset = 0;
 767
 768                 while (frag_len_left) {
 769                         len = min(frag_len_left,
 770                                 (unsigned int)WQ_ENET_MAX_DESC_LEN);
 771                         dma_addr = skb_frag_dma_map(&enic->pdev->dev, frag,
 772                                                     offset, len,
 773                                                     DMA_TO_DEVICE);
 774                         if (unlikely(enic_dma_map_check(enic, dma_addr)))
 775                                 return -ENOMEM;
 776                         enic_queue_wq_desc_cont(wq, skb, dma_addr, len,
 777                                                 (len_left == 0) &&
 778                                                  (len == frag_len_left),/*EOP*/
 779                                                 loopback);
 780                         frag_len_left -= len;
 781                         offset += len;
 782                 }
 783         }
 784
 785 tso_out_stats:
 786         /* calculate how many packets tso sent */
 787         len = skb->len - hdr_len;
 788         pkts = len / mss;
 789         if ((len % mss) > 0)
 790                 pkts++;
 791         enic->wq[wq->index].stats.packets += pkts;
 792         enic->wq[wq->index].stats.bytes += (len + (pkts * hdr_len));
 793
 794         return 0;
 795 }
 796
 797 static inline int enic_queue_wq_skb_encap(struct enic *enic, struct vnic_wq *wq,
 798                                           struct sk_buff *skb,
 799                                           int vlan_tag_insert,
 800                                           unsigned int vlan_tag, int loopback)
 801 {
 802         unsigned int head_len = skb_headlen(skb);
 803         unsigned int len_left = skb->len - head_len;
 804         /* Hardware will overwrite the checksum fields, calculating from
 805          * scratch and ignoring the value placed by software.
 806          * Offload mode = 00
 807          * mss[2], mss[1], mss[0] bits are set
 808          */
 809         unsigned int mss_or_csum = 7;
 810         int eop = (len_left == 0);
 811         dma_addr_t dma_addr;
 812         int err = 0;
 813
 814         dma_addr = dma_map_single(&enic->pdev->dev, skb->data, head_len,
 815                                   DMA_TO_DEVICE);
 816         if (unlikely(enic_dma_map_check(enic, dma_addr)))
 817                 return -ENOMEM;
 818
 819         enic_queue_wq_desc_ex(wq, skb, dma_addr, head_len, mss_or_csum, 0,
 820                               vlan_tag_insert, vlan_tag,
 821                               WQ_ENET_OFFLOAD_MODE_CSUM, eop, 1 /* SOP */, eop,
 822                               loopback);
 823         if (!eop)
 824                 err = enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback);
 825
 826         enic->wq[wq->index].stats.encap_csum++;
 827         enic->wq[wq->index].stats.packets++;
 828         enic->wq[wq->index].stats.bytes += skb->len;
 829
 830         return err;
 831 }
 832
 833 static inline int enic_queue_wq_skb(struct enic *enic,
 834         struct vnic_wq *wq, struct sk_buff *skb)
 835 {
 836         unsigned int mss = skb_shinfo(skb)->gso_size;
 837         unsigned int vlan_tag = 0;
 838         int vlan_tag_insert = 0;
 839         int loopback = 0;
 840         int err;
 841
 842         if (skb_vlan_tag_present(skb)) {
 843                 /* VLAN tag from trunking driver */
 844                 vlan_tag_insert = 1;
 845                 vlan_tag = skb_vlan_tag_get(skb);
 846                 enic->wq[wq->index].stats.add_vlan++;
 847         } else if (enic->loop_enable) {
 848                 vlan_tag = enic->loop_tag;
 849                 loopback = 1;
 850         }
 851
 852         if (mss)
 853                 err = enic_queue_wq_skb_tso(enic, wq, skb, mss,
 854                                             vlan_tag_insert, vlan_tag,
 855                                             loopback);
 856         else if (skb->encapsulation)
 857                 err = enic_queue_wq_skb_encap(enic, wq, skb, vlan_tag_insert,
 858                                               vlan_tag, loopback);
 859         else if (skb->ip_summed == CHECKSUM_PARTIAL)
 860                 err = enic_queue_wq_skb_csum_l4(enic, wq, skb, vlan_tag_insert,
 861                                                 vlan_tag, loopback);
 862         else
 863                 err = enic_queue_wq_skb_vlan(enic, wq, skb, vlan_tag_insert,
 864                                              vlan_tag, loopback);
 865         if (unlikely(err)) {
 866                 struct vnic_wq_buf *buf;
 867
 868                 buf = wq->to_use->prev;
 869                 /* while not EOP of previous pkt && queue not empty.
 870                  * For all non EOP bufs, os_buf is NULL.
 871                  */
 872                 while (!buf->os_buf && (buf->next != wq->to_clean)) {
 873                         enic_free_wq_buf(wq, buf);
 874                         wq->ring.desc_avail++;
 875                         buf = buf->prev;
 876                 }
 877                 wq->to_use = buf->next;
 878                 dev_kfree_skb(skb);
 879         }
 880         return err;
 881 }
 882
 883 /* netif_tx_lock held, process context with BHs disabled, or BH */
 884 static netdev_tx_t enic_hard_start_xmit(struct sk_buff *skb,
 885         struct net_device *netdev)
 886 {
 887         struct enic *enic = netdev_priv(netdev);
 888         struct vnic_wq *wq;
 889         unsigned int txq_map;
 890         struct netdev_queue *txq;
 891
 892         txq_map = skb_get_queue_mapping(skb) % enic->wq_count;
 893         wq = &enic->wq[txq_map].vwq;
 894
 895         if (skb->len <= 0) {
 896                 dev_kfree_skb_any(skb);
 897                 enic->wq[wq->index].stats.null_pkt++;
 898                 return NETDEV_TX_OK;
 899         }
 900
 901         txq = netdev_get_tx_queue(netdev, txq_map);
 902
 903         /* Non-TSO sends must fit within ENIC_NON_TSO_MAX_DESC descs,
 904          * which is very likely.  In the off chance it's going to take
 905          * more than * ENIC_NON_TSO_MAX_DESC, linearize the skb.
 906          */
 907
 908         if (skb_shinfo(skb)->gso_size == 0 &&
 909             skb_shinfo(skb)->nr_frags + 1 > ENIC_NON_TSO_MAX_DESC &&
 910             skb_linearize(skb)) {
 911                 dev_kfree_skb_any(skb);
 912                 enic->wq[wq->index].stats.skb_linear_fail++;
 913                 return NETDEV_TX_OK;
 914         }
 915
 916         spin_lock(&enic->wq[txq_map].lock);
 917
 918         if (vnic_wq_desc_avail(wq) <
 919             skb_shinfo(skb)->nr_frags + ENIC_DESC_MAX_SPLITS) {
 920                 netif_tx_stop_queue(txq);
 921                 /* This is a hard error, log it */
 922                 netdev_err(netdev, "BUG! Tx ring full when queue awake!\n");
 923                 spin_unlock(&enic->wq[txq_map].lock);
 924                 enic->wq[wq->index].stats.desc_full_awake++;
 925                 return NETDEV_TX_BUSY;
 926         }
 927
 928         if (enic_queue_wq_skb(enic, wq, skb))
 929                 goto error;
 930
 931         if (vnic_wq_desc_avail(wq) < MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS) {
 932                 netif_tx_stop_queue(txq);
 933                 enic->wq[wq->index].stats.stopped++;
 934         }
 935         skb_tx_timestamp(skb);
 936         if (!netdev_xmit_more() || netif_xmit_stopped(txq))
 937                 vnic_wq_doorbell(wq);
 938
 939 error:
 940         spin_unlock(&enic->wq[txq_map].lock);
 941
 942         return NETDEV_TX_OK;
 943 }
 944
 945 /* rcu_read_lock potentially held, nominally process context */
 946 static void enic_get_stats(struct net_device *netdev,
 947                            struct rtnl_link_stats64 *net_stats)
 948 {
 949         struct enic *enic = netdev_priv(netdev);
 950         struct vnic_stats *stats;
 951         u64 pkt_truncated = 0;
 952         u64 bad_fcs = 0;
 953         int err;
 954         int i;
 955
 956         err = enic_dev_stats_dump(enic, &stats);
 957         /* return only when dma_alloc_coherent fails in vnic_dev_stats_dump
 958          * For other failures, like devcmd failure, we return previously
 959          * recorded stats.
 960          */
 961         if (err == -ENOMEM)
 962                 return;
 963
 964         net_stats->tx_packets = stats->tx.tx_frames_ok;
 965         net_stats->tx_bytes = stats->tx.tx_bytes_ok;
 966         net_stats->tx_errors = stats->tx.tx_errors;
 967         net_stats->tx_dropped = stats->tx.tx_drops;
 968
 969         net_stats->rx_packets = stats->rx.rx_frames_ok;
 970         net_stats->rx_bytes = stats->rx.rx_bytes_ok;
 971         net_stats->rx_errors = stats->rx.rx_errors;
 972         net_stats->multicast = stats->rx.rx_multicast_frames_ok;
 973
 974         for (i = 0; i < enic->rq_count; i++) {
 975                 struct enic_rq_stats *rqs = &enic->rq[i].stats;
 976
 977                 if (!enic->rq[i].vrq.ctrl)
 978                         break;
 979                 pkt_truncated += rqs->pkt_truncated;
 980                 bad_fcs += rqs->bad_fcs;
 981         }
 982         net_stats->rx_over_errors = pkt_truncated;
 983         net_stats->rx_crc_errors = bad_fcs;
 984         net_stats->rx_dropped = stats->rx.rx_no_bufs + stats->rx.rx_drop;
 985 }
 986
 987 static int enic_mc_sync(struct net_device *netdev, const u8 *mc_addr)
 988 {
 989         struct enic *enic = netdev_priv(netdev);
 990
 991         if (enic->mc_count == ENIC_MULTICAST_PERFECT_FILTERS) {
 992                 unsigned int mc_count = netdev_mc_count(netdev);
 993
 994                 netdev_warn(netdev, "Registering only %d out of %d multicast addresses\n",
 995                             ENIC_MULTICAST_PERFECT_FILTERS, mc_count);
 996
 997                 return -ENOSPC;
 998         }
 999
1000         enic_dev_add_addr(enic, mc_addr);
1001         enic->mc_count++;
1002
1003         return 0;
1004 }
1005
1006 static int enic_mc_unsync(struct net_device *netdev, const u8 *mc_addr)
1007 {
1008         struct enic *enic = netdev_priv(netdev);
1009
1010         enic_dev_del_addr(enic, mc_addr);
1011         enic->mc_count--;
1012
1013         return 0;
1014 }
1015
1016 static int enic_uc_sync(struct net_device *netdev, const u8 *uc_addr)
1017 {
1018         struct enic *enic = netdev_priv(netdev);
1019
1020         if (enic->uc_count == ENIC_UNICAST_PERFECT_FILTERS) {
1021                 unsigned int uc_count = netdev_uc_count(netdev);
1022
1023                 netdev_warn(netdev, "Registering only %d out of %d unicast addresses\n",
1024                             ENIC_UNICAST_PERFECT_FILTERS, uc_count);
1025
1026                 return -ENOSPC;
1027         }
1028
1029         enic_dev_add_addr(enic, uc_addr);
1030         enic->uc_count++;
1031
1032         return 0;
1033 }
1034
1035 static int enic_uc_unsync(struct net_device *netdev, const u8 *uc_addr)
1036 {
1037         struct enic *enic = netdev_priv(netdev);
1038
1039         enic_dev_del_addr(enic, uc_addr);
1040         enic->uc_count--;
1041
1042         return 0;
1043 }
1044
1045 void enic_reset_addr_lists(struct enic *enic)
1046 {
1047         struct net_device *netdev = enic->netdev;
1048
1049         __dev_uc_unsync(netdev, NULL);
1050         __dev_mc_unsync(netdev, NULL);
1051
1052         enic->mc_count = 0;
1053         enic->uc_count = 0;
1054         enic->flags = 0;
1055 }
1056
1057 static int enic_set_mac_addr(struct net_device *netdev, char *addr)
1058 {
1059         struct enic *enic = netdev_priv(netdev);
1060
1061         if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic)) {
1062                 if (!is_valid_ether_addr(addr) && !is_zero_ether_addr(addr))
1063                         return -EADDRNOTAVAIL;
1064         } else {
1065                 if (!is_valid_ether_addr(addr))
1066                         return -EADDRNOTAVAIL;
1067         }
1068
1069         eth_hw_addr_set(netdev, addr);
1070
1071         return 0;
1072 }
1073
1074 static int enic_set_mac_address_dynamic(struct net_device *netdev, void *p)
1075 {
1076         struct enic *enic = netdev_priv(netdev);
1077         struct sockaddr *saddr = p;
1078         char *addr = saddr->sa_data;
1079         int err;
1080
1081         if (netif_running(enic->netdev)) {
1082                 err = enic_dev_del_station_addr(enic);
1083                 if (err)
1084                         return err;
1085         }
1086
1087         err = enic_set_mac_addr(netdev, addr);
1088         if (err)
1089                 return err;
1090
1091         if (netif_running(enic->netdev)) {
1092                 err = enic_dev_add_station_addr(enic);
1093                 if (err)
1094                         return err;
1095         }
1096
1097         return err;
1098 }
1099
1100 static int enic_set_mac_address(struct net_device *netdev, void *p)
1101 {
1102         struct sockaddr *saddr = p;
1103         char *addr = saddr->sa_data;
1104         struct enic *enic = netdev_priv(netdev);
1105         int err;
1106
1107         err = enic_dev_del_station_addr(enic);
1108         if (err)
1109                 return err;
1110
1111         err = enic_set_mac_addr(netdev, addr);
1112         if (err)
1113                 return err;
1114
1115         return enic_dev_add_station_addr(enic);
1116 }
1117
1118 /* netif_tx_lock held, BHs disabled */
1119 static void enic_set_rx_mode(struct net_device *netdev)
1120 {
1121         struct enic *enic = netdev_priv(netdev);
1122         int directed = 1;
1123         int multicast = (netdev->flags & IFF_MULTICAST) ? 1 : 0;
1124         int broadcast = (netdev->flags & IFF_BROADCAST) ? 1 : 0;
1125         int promisc = (netdev->flags & IFF_PROMISC) ||
1126                 netdev_uc_count(netdev) > ENIC_UNICAST_PERFECT_FILTERS;
1127         int allmulti = (netdev->flags & IFF_ALLMULTI) ||
1128                 netdev_mc_count(netdev) > ENIC_MULTICAST_PERFECT_FILTERS;
1129         unsigned int flags = netdev->flags |
1130                 (allmulti ? IFF_ALLMULTI : 0) |
1131                 (promisc ? IFF_PROMISC : 0);
1132
1133         if (enic->flags != flags) {
1134                 enic->flags = flags;
1135                 enic_dev_packet_filter(enic, directed,
1136                         multicast, broadcast, promisc, allmulti);
1137         }
1138
1139         if (!promisc) {
1140                 __dev_uc_sync(netdev, enic_uc_sync, enic_uc_unsync);
1141                 if (!allmulti)
1142                         __dev_mc_sync(netdev, enic_mc_sync, enic_mc_unsync);
1143         }
1144 }
1145
1146 /* netif_tx_lock held, BHs disabled */
1147 static void enic_tx_timeout(struct net_device *netdev, unsigned int txqueue)
1148 {
1149         struct enic *enic = netdev_priv(netdev);
1150         schedule_work(&enic->tx_hang_reset);
1151 }
1152
1153 static int enic_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
1154 {
1155         struct enic *enic = netdev_priv(netdev);
1156         struct enic_port_profile *pp;
1157         int err;
1158
1159         ENIC_PP_BY_INDEX(enic, vf, pp, &err);
1160         if (err)
1161                 return err;
1162
1163         if (is_valid_ether_addr(mac) || is_zero_ether_addr(mac)) {
1164                 if (vf == PORT_SELF_VF) {
1165                         memcpy(pp->vf_mac, mac, ETH_ALEN);
1166                         return 0;
1167                 } else {
1168                         /*
1169                          * For sriov vf's set the mac in hw
1170                          */
1171                         ENIC_DEVCMD_PROXY_BY_INDEX(vf, err, enic,
1172                                 vnic_dev_set_mac_addr, mac);
1173                         return enic_dev_status_to_errno(err);
1174                 }
1175         } else
1176                 return -EINVAL;
1177 }
1178
1179 static int enic_set_vf_port(struct net_device *netdev, int vf,
1180         struct nlattr *port[])
1181 {
1182         static const u8 zero_addr[ETH_ALEN] = {};
1183         struct enic *enic = netdev_priv(netdev);
1184         struct enic_port_profile prev_pp;
1185         struct enic_port_profile *pp;
1186         int err = 0, restore_pp = 1;
1187
1188         ENIC_PP_BY_INDEX(enic, vf, pp, &err);
1189         if (err)
1190                 return err;
1191
1192         if (!port[IFLA_PORT_REQUEST])
1193                 return -EOPNOTSUPP;
1194
1195         memcpy(&prev_pp, pp, sizeof(*enic->pp));
1196         memset(pp, 0, sizeof(*enic->pp));
1197
1198         pp->set |= ENIC_SET_REQUEST;
1199         pp->request = nla_get_u8(port[IFLA_PORT_REQUEST]);
1200
1201         if (port[IFLA_PORT_PROFILE]) {
1202                 if (nla_len(port[IFLA_PORT_PROFILE]) != PORT_PROFILE_MAX) {
1203                         memcpy(pp, &prev_pp, sizeof(*pp));
1204                         return -EINVAL;
1205                 }
1206                 pp->set |= ENIC_SET_NAME;
1207                 memcpy(pp->name, nla_data(port[IFLA_PORT_PROFILE]),
1208                         PORT_PROFILE_MAX);
1209         }
1210
1211         if (port[IFLA_PORT_INSTANCE_UUID]) {
1212                 if (nla_len(port[IFLA_PORT_INSTANCE_UUID]) != PORT_UUID_MAX) {
1213                         memcpy(pp, &prev_pp, sizeof(*pp));
1214                         return -EINVAL;
1215                 }
1216                 pp->set |= ENIC_SET_INSTANCE;
1217                 memcpy(pp->instance_uuid,
1218                         nla_data(port[IFLA_PORT_INSTANCE_UUID]), PORT_UUID_MAX);
1219         }
1220
1221         if (port[IFLA_PORT_HOST_UUID]) {
1222                 if (nla_len(port[IFLA_PORT_HOST_UUID]) != PORT_UUID_MAX) {
1223                         memcpy(pp, &prev_pp, sizeof(*pp));
1224                         return -EINVAL;
1225                 }
1226                 pp->set |= ENIC_SET_HOST;
1227                 memcpy(pp->host_uuid,
1228                         nla_data(port[IFLA_PORT_HOST_UUID]), PORT_UUID_MAX);
1229         }
1230
1231         if (vf == PORT_SELF_VF) {
1232                 /* Special case handling: mac came from IFLA_VF_MAC */
1233                 if (!is_zero_ether_addr(prev_pp.vf_mac))
1234                         memcpy(pp->mac_addr, prev_pp.vf_mac, ETH_ALEN);
1235
1236                 if (is_zero_ether_addr(netdev->dev_addr))
1237                         eth_hw_addr_random(netdev);
1238         } else {
1239                 /* SR-IOV VF: get mac from adapter */
1240                 ENIC_DEVCMD_PROXY_BY_INDEX(vf, err, enic,
1241                         vnic_dev_get_mac_addr, pp->mac_addr);
1242                 if (err) {
1243                         netdev_err(netdev, "Error getting mac for vf %d\n", vf);
1244                         memcpy(pp, &prev_pp, sizeof(*pp));
1245                         return enic_dev_status_to_errno(err);
1246                 }
1247         }
1248
1249         err = enic_process_set_pp_request(enic, vf, &prev_pp, &restore_pp);
1250         if (err) {
1251                 if (restore_pp) {
1252                         /* Things are still the way they were: Implicit
1253                          * DISASSOCIATE failed
1254                          */
1255                         memcpy(pp, &prev_pp, sizeof(*pp));
1256                 } else {
1257                         memset(pp, 0, sizeof(*pp));
1258                         if (vf == PORT_SELF_VF)
1259                                 eth_hw_addr_set(netdev, zero_addr);
1260                 }
1261         } else {
1262                 /* Set flag to indicate that the port assoc/disassoc
1263                  * request has been sent out to fw
1264                  */
1265                 pp->set |= ENIC_PORT_REQUEST_APPLIED;
1266
1267                 /* If DISASSOCIATE, clean up all assigned/saved macaddresses */
1268                 if (pp->request == PORT_REQUEST_DISASSOCIATE) {
1269                         eth_zero_addr(pp->mac_addr);
1270                         if (vf == PORT_SELF_VF)
1271                                 eth_hw_addr_set(netdev, zero_addr);
1272                 }
1273         }
1274
1275         if (vf == PORT_SELF_VF)
1276                 eth_zero_addr(pp->vf_mac);
1277
1278         return err;
1279 }
1280
1281 static int enic_get_vf_port(struct net_device *netdev, int vf,
1282         struct sk_buff *skb)
1283 {
1284         struct enic *enic = netdev_priv(netdev);
1285         u16 response = PORT_PROFILE_RESPONSE_SUCCESS;
1286         struct enic_port_profile *pp;
1287         int err;
1288
1289         ENIC_PP_BY_INDEX(enic, vf, pp, &err);
1290         if (err)
1291                 return err;
1292
1293         if (!(pp->set & ENIC_PORT_REQUEST_APPLIED))
1294                 return -ENODATA;
1295
1296         err = enic_process_get_pp_request(enic, vf, pp->request, &response);
1297         if (err)
1298                 return err;
1299
1300         if (nla_put_u16(skb, IFLA_PORT_REQUEST, pp->request) ||
1301             nla_put_u16(skb, IFLA_PORT_RESPONSE, response) ||
1302             ((pp->set & ENIC_SET_NAME) &&
1303              nla_put(skb, IFLA_PORT_PROFILE, PORT_PROFILE_MAX, pp->name)) ||
1304             ((pp->set & ENIC_SET_INSTANCE) &&
1305              nla_put(skb, IFLA_PORT_INSTANCE_UUID, PORT_UUID_MAX,
1306                      pp->instance_uuid)) ||
1307             ((pp->set & ENIC_SET_HOST) &&
1308              nla_put(skb, IFLA_PORT_HOST_UUID, PORT_UUID_MAX, pp->host_uuid)))
1309                 goto nla_put_failure;
1310         return 0;
1311
1312 nla_put_failure:
1313         return -EMSGSIZE;
1314 }
1315
1316 static void enic_free_rq_buf(struct vnic_rq *rq, struct vnic_rq_buf *buf)
1317 {
1318         struct enic *enic = vnic_dev_priv(rq->vdev);
1319
1320         if (!buf->os_buf)
1321                 return;
1322
1323         dma_unmap_single(&enic->pdev->dev, buf->dma_addr, buf->len,
1324                          DMA_FROM_DEVICE);
1325         dev_kfree_skb_any(buf->os_buf);
1326         buf->os_buf = NULL;
1327 }
1328
1329 static int enic_rq_alloc_buf(struct vnic_rq *rq)
1330 {
1331         struct enic *enic = vnic_dev_priv(rq->vdev);
1332         struct net_device *netdev = enic->netdev;
1333         struct sk_buff *skb;
1334         unsigned int len = netdev->mtu + VLAN_ETH_HLEN;
1335         unsigned int os_buf_index = 0;
1336         dma_addr_t dma_addr;
1337         struct vnic_rq_buf *buf = rq->to_use;
1338
1339         if (buf->os_buf) {
1340                 enic_queue_rq_desc(rq, buf->os_buf, os_buf_index, buf->dma_addr,
1341                                    buf->len);
1342
1343                 return 0;
1344         }
1345         skb = netdev_alloc_skb_ip_align(netdev, len);
1346         if (!skb) {
1347                 enic->rq[rq->index].stats.no_skb++;
1348                 return -ENOMEM;
1349         }
1350
1351         dma_addr = dma_map_single(&enic->pdev->dev, skb->data, len,
1352                                   DMA_FROM_DEVICE);
1353         if (unlikely(enic_dma_map_check(enic, dma_addr))) {
1354                 dev_kfree_skb(skb);
1355                 return -ENOMEM;
1356         }
1357
1358         enic_queue_rq_desc(rq, skb, os_buf_index,
1359                 dma_addr, len);
1360
1361         return 0;
1362 }
1363
1364 static void enic_intr_update_pkt_size(struct vnic_rx_bytes_counter *pkt_size,
1365                                       u32 pkt_len)
1366 {
1367         if (ENIC_LARGE_PKT_THRESHOLD <= pkt_len)
1368                 pkt_size->large_pkt_bytes_cnt += pkt_len;
1369         else
1370                 pkt_size->small_pkt_bytes_cnt += pkt_len;
1371 }
1372
1373 static bool enic_rxcopybreak(struct net_device *netdev, struct sk_buff **skb,
1374                              struct vnic_rq_buf *buf, u16 len)
1375 {
1376         struct enic *enic = netdev_priv(netdev);
1377         struct sk_buff *new_skb;
1378
1379         if (len > enic->rx_copybreak)
1380                 return false;
1381         new_skb = netdev_alloc_skb_ip_align(netdev, len);
1382         if (!new_skb)
1383                 return false;
1384         dma_sync_single_for_cpu(&enic->pdev->dev, buf->dma_addr, len,
1385                                 DMA_FROM_DEVICE);
1386         memcpy(new_skb->data, (*skb)->data, len);
1387         *skb = new_skb;
1388
1389         return true;
1390 }
1391
1392 static void enic_rq_indicate_buf(struct vnic_rq *rq,
1393         struct cq_desc *cq_desc, struct vnic_rq_buf *buf,
1394         int skipped, void *opaque)
1395 {
1396         struct enic *enic = vnic_dev_priv(rq->vdev);
1397         struct net_device *netdev = enic->netdev;
1398         struct sk_buff *skb;
1399         struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
1400         struct enic_rq_stats *rqstats = &enic->rq[rq->index].stats;
1401
1402         u8 type, color, eop, sop, ingress_port, vlan_stripped;
1403         u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof;
1404         u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok;
1405         u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc;
1406         u8 packet_error;
1407         u16 q_number, completed_index, bytes_written, vlan_tci, checksum;
1408         u32 rss_hash;
1409         bool outer_csum_ok = true, encap = false;
1410
1411         rqstats->packets++;
1412         if (skipped) {
1413                 rqstats->desc_skip++;
1414                 return;
1415         }
1416
1417         skb = buf->os_buf;
1418
1419         cq_enet_rq_desc_dec((struct cq_enet_rq_desc *)cq_desc,
1420                 &type, &color, &q_number, &completed_index,
1421                 &ingress_port, &fcoe, &eop, &sop, &rss_type,
1422                 &csum_not_calc, &rss_hash, &bytes_written,
1423                 &packet_error, &vlan_stripped, &vlan_tci, &checksum,
1424                 &fcoe_sof, &fcoe_fc_crc_ok, &fcoe_enc_error,
1425                 &fcoe_eof, &tcp_udp_csum_ok, &udp, &tcp,
1426                 &ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment,
1427                 &fcs_ok);
1428
1429         if (packet_error) {
1430
1431                 if (!fcs_ok) {
1432                         if (bytes_written > 0)
1433                                 rqstats->bad_fcs++;
1434                         else if (bytes_written == 0)
1435                                 rqstats->pkt_truncated++;
1436                 }
1437
1438                 dma_unmap_single(&enic->pdev->dev, buf->dma_addr, buf->len,
1439                                  DMA_FROM_DEVICE);
1440                 dev_kfree_skb_any(skb);
1441                 buf->os_buf = NULL;
1442
1443                 return;
1444         }
1445
1446         if (eop && bytes_written > 0) {
1447
1448                 /* Good receive
1449                  */
1450                 rqstats->bytes += bytes_written;
1451                 if (!enic_rxcopybreak(netdev, &skb, buf, bytes_written)) {
1452                         buf->os_buf = NULL;
1453                         dma_unmap_single(&enic->pdev->dev, buf->dma_addr,
1454                                          buf->len, DMA_FROM_DEVICE);
1455                 }
1456                 prefetch(skb->data - NET_IP_ALIGN);
1457
1458                 skb_put(skb, bytes_written);
1459                 skb->protocol = eth_type_trans(skb, netdev);
1460                 skb_record_rx_queue(skb, q_number);
1461                 if ((netdev->features & NETIF_F_RXHASH) && rss_hash &&
1462                     (type == 3)) {
1463                         switch (rss_type) {
1464                         case CQ_ENET_RQ_DESC_RSS_TYPE_TCP_IPv4:
1465                         case CQ_ENET_RQ_DESC_RSS_TYPE_TCP_IPv6:
1466                         case CQ_ENET_RQ_DESC_RSS_TYPE_TCP_IPv6_EX:
1467                                 skb_set_hash(skb, rss_hash, PKT_HASH_TYPE_L4);
1468                                 rqstats->l4_rss_hash++;
1469                                 break;
1470                         case CQ_ENET_RQ_DESC_RSS_TYPE_IPv4:
1471                         case CQ_ENET_RQ_DESC_RSS_TYPE_IPv6:
1472                         case CQ_ENET_RQ_DESC_RSS_TYPE_IPv6_EX:
1473                                 skb_set_hash(skb, rss_hash, PKT_HASH_TYPE_L3);
1474                                 rqstats->l3_rss_hash++;
1475                                 break;
1476                         }
1477                 }
1478                 if (enic->vxlan.vxlan_udp_port_number) {
1479                         switch (enic->vxlan.patch_level) {
1480                         case 0:
1481                                 if (fcoe) {
1482                                         encap = true;
1483                                         outer_csum_ok = fcoe_fc_crc_ok;
1484                                 }
1485                                 break;
1486                         case 2:
1487                                 if ((type == 7) &&
1488                                     (rss_hash & BIT(0))) {
1489                                         encap = true;
1490                                         outer_csum_ok = (rss_hash & BIT(1)) &&
1491                                                         (rss_hash & BIT(2));
1492                                 }
1493                                 break;
1494                         }
1495                 }
1496
1497                 /* Hardware does not provide whole packet checksum. It only
1498                  * provides pseudo checksum. Since hw validates the packet
1499                  * checksum but not provide us the checksum value. use
1500                  * CHECSUM_UNNECESSARY.
1501                  *
1502                  * In case of encap pkt tcp_udp_csum_ok/tcp_udp_csum_ok is
1503                  * inner csum_ok. outer_csum_ok is set by hw when outer udp
1504                  * csum is correct or is zero.
1505                  */
1506                 if ((netdev->features & NETIF_F_RXCSUM) && !csum_not_calc &&
1507                     tcp_udp_csum_ok && outer_csum_ok &&
1508                     (ipv4_csum_ok || ipv6)) {
1509                         skb->ip_summed = CHECKSUM_UNNECESSARY;
1510                         skb->csum_level = encap;
1511                         if (encap)
1512                                 rqstats->csum_unnecessary_encap++;
1513                         else
1514                                 rqstats->csum_unnecessary++;
1515                 }
1516
1517                 if (vlan_stripped) {
1518                         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tci);
1519                         rqstats->vlan_stripped++;
1520                 }
1521                 skb_mark_napi_id(skb, &enic->napi[rq->index]);
1522                 if (!(netdev->features & NETIF_F_GRO))
1523                         netif_receive_skb(skb);
1524                 else
1525                         napi_gro_receive(&enic->napi[q_number], skb);
1526                 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1527                         enic_intr_update_pkt_size(&cq->pkt_size_counter,
1528                                                   bytes_written);
1529         } else {
1530
1531                 /* Buffer overflow
1532                  */
1533                 rqstats->pkt_truncated++;
1534                 dma_unmap_single(&enic->pdev->dev, buf->dma_addr, buf->len,
1535                                  DMA_FROM_DEVICE);
1536                 dev_kfree_skb_any(skb);
1537                 buf->os_buf = NULL;
1538         }
1539 }
1540
1541 static int enic_rq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
1542         u8 type, u16 q_number, u16 completed_index, void *opaque)
1543 {
1544         struct enic *enic = vnic_dev_priv(vdev);
1545
1546         vnic_rq_service(&enic->rq[q_number].vrq, cq_desc,
1547                 completed_index, VNIC_RQ_RETURN_DESC,
1548                 enic_rq_indicate_buf, opaque);
1549
1550         return 0;
1551 }
1552
1553 static void enic_set_int_moderation(struct enic *enic, struct vnic_rq *rq)
1554 {
1555         unsigned int intr = enic_msix_rq_intr(enic, rq->index);
1556         struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
1557         u32 timer = cq->tobe_rx_coal_timeval;
1558
1559         if (cq->tobe_rx_coal_timeval != cq->cur_rx_coal_timeval) {
1560                 vnic_intr_coalescing_timer_set(&enic->intr[intr], timer);
1561                 cq->cur_rx_coal_timeval = cq->tobe_rx_coal_timeval;
1562         }
1563 }
1564
1565 static void enic_calc_int_moderation(struct enic *enic, struct vnic_rq *rq)
1566 {
1567         struct enic_rx_coal *rx_coal = &enic->rx_coalesce_setting;
1568         struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
1569         struct vnic_rx_bytes_counter *pkt_size_counter = &cq->pkt_size_counter;
1570         int index;
1571         u32 timer;
1572         u32 range_start;
1573         u32 traffic;
1574         u64 delta;
1575         ktime_t now = ktime_get();
1576
1577         delta = ktime_us_delta(now, cq->prev_ts);
1578         if (delta < ENIC_AIC_TS_BREAK)
1579                 return;
1580         cq->prev_ts = now;
1581
1582         traffic = pkt_size_counter->large_pkt_bytes_cnt +
1583                   pkt_size_counter->small_pkt_bytes_cnt;
1584         /* The table takes Mbps
1585          * traffic *= 8    => bits
1586          * traffic *= (10^6 / delta)    => bps
1587          * traffic /= 10^6     => Mbps
1588          *
1589          * Combining, traffic *= (8 / delta)
1590          */
1591
1592         traffic <<= 3;
1593         traffic = delta > UINT_MAX ? 0 : traffic / (u32)delta;
1594
1595         for (index = 0; index < ENIC_MAX_COALESCE_TIMERS; index++)
1596                 if (traffic < mod_table[index].rx_rate)
1597                         break;
1598         range_start = (pkt_size_counter->small_pkt_bytes_cnt >
1599                        pkt_size_counter->large_pkt_bytes_cnt << 1) ?
1600                       rx_coal->small_pkt_range_start :
1601                       rx_coal->large_pkt_range_start;
1602         timer = range_start + ((rx_coal->range_end - range_start) *
1603                                mod_table[index].range_percent / 100);
1604         /* Damping */
1605         cq->tobe_rx_coal_timeval = (timer + cq->tobe_rx_coal_timeval) >> 1;
1606
1607         pkt_size_counter->large_pkt_bytes_cnt = 0;
1608         pkt_size_counter->small_pkt_bytes_cnt = 0;
1609 }
1610
1611 static int enic_poll(struct napi_struct *napi, int budget)
1612 {
1613         struct net_device *netdev = napi->dev;
1614         struct enic *enic = netdev_priv(netdev);
1615         unsigned int cq_rq = enic_cq_rq(enic, 0);
1616         unsigned int cq_wq = enic_cq_wq(enic, 0);
1617         unsigned int intr = ENIC_LEGACY_IO_INTR;
1618         unsigned int rq_work_to_do = budget;
1619         unsigned int wq_work_to_do = ENIC_WQ_NAPI_BUDGET;
1620         unsigned int  work_done, rq_work_done = 0, wq_work_done;
1621         int err;
1622
1623         wq_work_done = vnic_cq_service(&enic->cq[cq_wq], wq_work_to_do,
1624                                        enic_wq_service, NULL);
1625
1626         if (budget > 0)
1627                 rq_work_done = vnic_cq_service(&enic->cq[cq_rq],
1628                         rq_work_to_do, enic_rq_service, NULL);
1629
1630         /* Accumulate intr event credits for this polling
1631          * cycle.  An intr event is the completion of a
1632          * a WQ or RQ packet.
1633          */
1634
1635         work_done = rq_work_done + wq_work_done;
1636
1637         if (work_done > 0)
1638                 vnic_intr_return_credits(&enic->intr[intr],
1639                         work_done,
1640                         0 /* don't unmask intr */,
1641                         0 /* don't reset intr timer */);
1642
1643         err = vnic_rq_fill(&enic->rq[0].vrq, enic_rq_alloc_buf);
1644
1645         /* Buffer allocation failed. Stay in polling
1646          * mode so we can try to fill the ring again.
1647          */
1648
1649         if (err)
1650                 rq_work_done = rq_work_to_do;
1651         if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1652                 /* Call the function which refreshes the intr coalescing timer
1653                  * value based on the traffic.
1654                  */
1655                 enic_calc_int_moderation(enic, &enic->rq[0].vrq);
1656
1657         if ((rq_work_done < budget) && napi_complete_done(napi, rq_work_done)) {
1658
1659                 /* Some work done, but not enough to stay in polling,
1660                  * exit polling
1661                  */
1662
1663                 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1664                         enic_set_int_moderation(enic, &enic->rq[0].vrq);
1665                 vnic_intr_unmask(&enic->intr[intr]);
1666                 enic->rq[0].stats.napi_complete++;
1667         } else {
1668                 enic->rq[0].stats.napi_repoll++;
1669         }
1670
1671         return rq_work_done;
1672 }
1673
1674 #ifdef CONFIG_RFS_ACCEL
1675 static void enic_free_rx_cpu_rmap(struct enic *enic)
1676 {
1677         free_irq_cpu_rmap(enic->netdev->rx_cpu_rmap);
1678         enic->netdev->rx_cpu_rmap = NULL;
1679 }
1680
1681 static void enic_set_rx_cpu_rmap(struct enic *enic)
1682 {
1683         int i, res;
1684
1685         if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX) {
1686                 enic->netdev->rx_cpu_rmap = alloc_irq_cpu_rmap(enic->rq_count);
1687                 if (unlikely(!enic->netdev->rx_cpu_rmap))
1688                         return;
1689                 for (i = 0; i < enic->rq_count; i++) {
1690                         res = irq_cpu_rmap_add(enic->netdev->rx_cpu_rmap,
1691                                                enic->msix_entry[i].vector);
1692                         if (unlikely(res)) {
1693                                 enic_free_rx_cpu_rmap(enic);
1694                                 return;
1695                         }
1696                 }
1697         }
1698 }
1699
1700 #else
1701
1702 static void enic_free_rx_cpu_rmap(struct enic *enic)
1703 {
1704 }
1705
1706 static void enic_set_rx_cpu_rmap(struct enic *enic)
1707 {
1708 }
1709
1710 #endif /* CONFIG_RFS_ACCEL */
1711
1712 static int enic_poll_msix_wq(struct napi_struct *napi, int budget)
1713 {
1714         struct net_device *netdev = napi->dev;
1715         struct enic *enic = netdev_priv(netdev);
1716         unsigned int wq_index = (napi - &enic->napi[0]) - enic->rq_count;
1717         struct vnic_wq *wq = &enic->wq[wq_index].vwq;
1718         unsigned int cq;
1719         unsigned int intr;
1720         unsigned int wq_work_to_do = ENIC_WQ_NAPI_BUDGET;
1721         unsigned int wq_work_done;
1722         unsigned int wq_irq;
1723
1724         wq_irq = wq->index;
1725         cq = enic_cq_wq(enic, wq_irq);
1726         intr = enic_msix_wq_intr(enic, wq_irq);
1727         wq_work_done = vnic_cq_service(&enic->cq[cq], wq_work_to_do,
1728                                        enic_wq_service, NULL);
1729
1730         vnic_intr_return_credits(&enic->intr[intr], wq_work_done,
1731                                  0 /* don't unmask intr */,
1732                                  1 /* reset intr timer */);
1733         if (!wq_work_done) {
1734                 napi_complete(napi);
1735                 vnic_intr_unmask(&enic->intr[intr]);
1736                 return 0;
1737         }
1738
1739         return budget;
1740 }
1741
1742 static int enic_poll_msix_rq(struct napi_struct *napi, int budget)
1743 {
1744         struct net_device *netdev = napi->dev;
1745         struct enic *enic = netdev_priv(netdev);
1746         unsigned int rq = (napi - &enic->napi[0]);
1747         unsigned int cq = enic_cq_rq(enic, rq);
1748         unsigned int intr = enic_msix_rq_intr(enic, rq);
1749         unsigned int work_to_do = budget;
1750         unsigned int work_done = 0;
1751         int err;
1752
1753         /* Service RQ
1754          */
1755
1756         if (budget > 0)
1757                 work_done = vnic_cq_service(&enic->cq[cq],
1758                         work_to_do, enic_rq_service, NULL);
1759
1760         /* Return intr event credits for this polling
1761          * cycle.  An intr event is the completion of a
1762          * RQ packet.
1763          */
1764
1765         if (work_done > 0)
1766                 vnic_intr_return_credits(&enic->intr[intr],
1767                         work_done,
1768                         0 /* don't unmask intr */,
1769                         0 /* don't reset intr timer */);
1770
1771         err = vnic_rq_fill(&enic->rq[rq].vrq, enic_rq_alloc_buf);
1772
1773         /* Buffer allocation failed. Stay in polling mode
1774          * so we can try to fill the ring again.
1775          */
1776
1777         if (err)
1778                 work_done = work_to_do;
1779         if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1780                 /* Call the function which refreshes the intr coalescing timer
1781                  * value based on the traffic.
1782                  */
1783                 enic_calc_int_moderation(enic, &enic->rq[rq].vrq);
1784
1785         if ((work_done < budget) && napi_complete_done(napi, work_done)) {
1786
1787                 /* Some work done, but not enough to stay in polling,
1788                  * exit polling
1789                  */
1790
1791                 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1792                         enic_set_int_moderation(enic, &enic->rq[rq].vrq);
1793                 vnic_intr_unmask(&enic->intr[intr]);
1794                 enic->rq[rq].stats.napi_complete++;
1795         } else {
1796                 enic->rq[rq].stats.napi_repoll++;
1797         }
1798
1799         return work_done;
1800 }
1801
1802 static void enic_notify_timer(struct timer_list *t)
1803 {
1804         struct enic *enic = from_timer(enic, t, notify_timer);
1805
1806         enic_notify_check(enic);
1807
1808         mod_timer(&enic->notify_timer,
1809                 round_jiffies(jiffies + ENIC_NOTIFY_TIMER_PERIOD));
1810 }
1811
1812 static void enic_free_intr(struct enic *enic)
1813 {
1814         struct net_device *netdev = enic->netdev;
1815         unsigned int i;
1816
1817         enic_free_rx_cpu_rmap(enic);
1818         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1819         case VNIC_DEV_INTR_MODE_INTX:
1820                 free_irq(enic->pdev->irq, netdev);
1821                 break;
1822         case VNIC_DEV_INTR_MODE_MSI:
1823                 free_irq(enic->pdev->irq, enic);
1824                 break;
1825         case VNIC_DEV_INTR_MODE_MSIX:
1826                 for (i = 0; i < enic->intr_count; i++)
1827                         if (enic->msix[i].requested)
1828                                 free_irq(enic->msix_entry[i].vector,
1829                                         enic->msix[i].devid);
1830                 break;
1831         default:
1832                 break;
1833         }
1834 }
1835
1836 static int enic_request_intr(struct enic *enic)
1837 {
1838         struct net_device *netdev = enic->netdev;
1839         unsigned int i, intr;
1840         int err = 0;
1841
1842         enic_set_rx_cpu_rmap(enic);
1843         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1844
1845         case VNIC_DEV_INTR_MODE_INTX:
1846
1847                 err = request_irq(enic->pdev->irq, enic_isr_legacy,
1848                         IRQF_SHARED, netdev->name, netdev);
1849                 break;
1850
1851         case VNIC_DEV_INTR_MODE_MSI:
1852
1853                 err = request_irq(enic->pdev->irq, enic_isr_msi,
1854                         0, netdev->name, enic);
1855                 break;
1856
1857         case VNIC_DEV_INTR_MODE_MSIX:
1858
1859                 for (i = 0; i < enic->rq_count; i++) {
1860                         intr = enic_msix_rq_intr(enic, i);
1861                         snprintf(enic->msix[intr].devname,
1862                                 sizeof(enic->msix[intr].devname),
1863                                 "%s-rx-%u", netdev->name, i);
1864                         enic->msix[intr].isr = enic_isr_msix;
1865                         enic->msix[intr].devid = &enic->napi[i];
1866                 }
1867
1868                 for (i = 0; i < enic->wq_count; i++) {
1869                         int wq = enic_cq_wq(enic, i);
1870
1871                         intr = enic_msix_wq_intr(enic, i);
1872                         snprintf(enic->msix[intr].devname,
1873                                 sizeof(enic->msix[intr].devname),
1874                                 "%s-tx-%u", netdev->name, i);
1875                         enic->msix[intr].isr = enic_isr_msix;
1876                         enic->msix[intr].devid = &enic->napi[wq];
1877                 }
1878
1879                 intr = enic_msix_err_intr(enic);
1880                 snprintf(enic->msix[intr].devname,
1881                         sizeof(enic->msix[intr].devname),
1882                         "%s-err", netdev->name);
1883                 enic->msix[intr].isr = enic_isr_msix_err;
1884                 enic->msix[intr].devid = enic;
1885
1886                 intr = enic_msix_notify_intr(enic);
1887                 snprintf(enic->msix[intr].devname,
1888                         sizeof(enic->msix[intr].devname),
1889                         "%s-notify", netdev->name);
1890                 enic->msix[intr].isr = enic_isr_msix_notify;
1891                 enic->msix[intr].devid = enic;
1892
1893                 for (i = 0; i < enic->intr_count; i++)
1894                         enic->msix[i].requested = 0;
1895
1896                 for (i = 0; i < enic->intr_count; i++) {
1897                         err = request_irq(enic->msix_entry[i].vector,
1898                                 enic->msix[i].isr, 0,
1899                                 enic->msix[i].devname,
1900                                 enic->msix[i].devid);
1901                         if (err) {
1902                                 enic_free_intr(enic);
1903                                 break;
1904                         }
1905                         enic->msix[i].requested = 1;
1906                 }
1907
1908                 break;
1909
1910         default:
1911                 break;
1912         }
1913
1914         return err;
1915 }
1916
1917 static void enic_synchronize_irqs(struct enic *enic)
1918 {
1919         unsigned int i;
1920
1921         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1922         case VNIC_DEV_INTR_MODE_INTX:
1923         case VNIC_DEV_INTR_MODE_MSI:
1924                 synchronize_irq(enic->pdev->irq);
1925                 break;
1926         case VNIC_DEV_INTR_MODE_MSIX:
1927                 for (i = 0; i < enic->intr_count; i++)
1928                         synchronize_irq(enic->msix_entry[i].vector);
1929                 break;
1930         default:
1931                 break;
1932         }
1933 }
1934
1935 static int enic_dev_notify_set(struct enic *enic)
1936 {
1937         int err;
1938
1939         spin_lock_bh(&enic->devcmd_lock);
1940         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1941         case VNIC_DEV_INTR_MODE_INTX:
1942                 err = vnic_dev_notify_set(enic->vdev, ENIC_LEGACY_NOTIFY_INTR);
1943                 break;
1944         case VNIC_DEV_INTR_MODE_MSIX:
1945                 err = vnic_dev_notify_set(enic->vdev,
1946                         enic_msix_notify_intr(enic));
1947                 break;
1948         default:
1949                 err = vnic_dev_notify_set(enic->vdev, -1 /* no intr */);
1950                 break;
1951         }
1952         spin_unlock_bh(&enic->devcmd_lock);
1953
1954         return err;
1955 }
1956
1957 static void enic_notify_timer_start(struct enic *enic)
1958 {
1959         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1960         case VNIC_DEV_INTR_MODE_MSI:
1961                 mod_timer(&enic->notify_timer, jiffies);
1962                 break;
1963         default:
1964                 /* Using intr for notification for INTx/MSI-X */
1965                 break;
1966         }
1967 }
1968
1969 /* rtnl lock is held, process context */
1970 static int enic_open(struct net_device *netdev)
1971 {
1972         struct enic *enic = netdev_priv(netdev);
1973         unsigned int i;
1974         int err, ret;
1975
1976         err = enic_request_intr(enic);
1977         if (err) {
1978                 netdev_err(netdev, "Unable to request irq.\n");
1979                 return err;
1980         }
1981         enic_init_affinity_hint(enic);
1982         enic_set_affinity_hint(enic);
1983
1984         err = enic_dev_notify_set(enic);
1985         if (err) {
1986                 netdev_err(netdev,
1987                         "Failed to alloc notify buffer, aborting.\n");
1988                 goto err_out_free_intr;
1989         }
1990
1991         for (i = 0; i < enic->rq_count; i++) {
1992                 /* enable rq before updating rq desc */
1993                 vnic_rq_enable(&enic->rq[i].vrq);
1994                 vnic_rq_fill(&enic->rq[i].vrq, enic_rq_alloc_buf);
1995                 /* Need at least one buffer on ring to get going */
1996                 if (vnic_rq_desc_used(&enic->rq[i].vrq) == 0) {
1997                         netdev_err(netdev, "Unable to alloc receive buffers\n");
1998                         err = -ENOMEM;
1999                         goto err_out_free_rq;
2000                 }
2001         }
2002
2003         for (i = 0; i < enic->wq_count; i++)
2004                 vnic_wq_enable(&enic->wq[i].vwq);
2005
2006         if (!enic_is_dynamic(enic) && !enic_is_sriov_vf(enic))
2007                 enic_dev_add_station_addr(enic);
2008
2009         enic_set_rx_mode(netdev);
2010
2011         netif_tx_wake_all_queues(netdev);
2012
2013         for (i = 0; i < enic->rq_count; i++)
2014                 napi_enable(&enic->napi[i]);
2015
2016         if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX)
2017                 for (i = 0; i < enic->wq_count; i++)
2018                         napi_enable(&enic->napi[enic_cq_wq(enic, i)]);
2019         enic_dev_enable(enic);
2020
2021         for (i = 0; i < enic->intr_count; i++)
2022                 vnic_intr_unmask(&enic->intr[i]);
2023
2024         enic_notify_timer_start(enic);
2025         enic_rfs_timer_start(enic);
2026
2027         return 0;
2028
2029 err_out_free_rq:
2030         for (i = 0; i < enic->rq_count; i++) {
2031                 ret = vnic_rq_disable(&enic->rq[i].vrq);
2032                 if (!ret)
2033                         vnic_rq_clean(&enic->rq[i].vrq, enic_free_rq_buf);
2034         }
2035         enic_dev_notify_unset(enic);
2036 err_out_free_intr:
2037         enic_unset_affinity_hint(enic);
2038         enic_free_intr(enic);
2039
2040         return err;
2041 }
2042
2043 /* rtnl lock is held, process context */
2044 static int enic_stop(struct net_device *netdev)
2045 {
2046         struct enic *enic = netdev_priv(netdev);
2047         unsigned int i;
2048         int err;
2049
2050         for (i = 0; i < enic->intr_count; i++) {
2051                 vnic_intr_mask(&enic->intr[i]);
2052                 (void)vnic_intr_masked(&enic->intr[i]); /* flush write */
2053         }
2054
2055         enic_synchronize_irqs(enic);
2056
2057         del_timer_sync(&enic->notify_timer);
2058         enic_rfs_flw_tbl_free(enic);
2059
2060         enic_dev_disable(enic);
2061
2062         for (i = 0; i < enic->rq_count; i++)
2063                 napi_disable(&enic->napi[i]);
2064
2065         netif_carrier_off(netdev);
2066         if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX)
2067                 for (i = 0; i < enic->wq_count; i++)
2068                         napi_disable(&enic->napi[enic_cq_wq(enic, i)]);
2069         netif_tx_disable(netdev);
2070
2071         if (!enic_is_dynamic(enic) && !enic_is_sriov_vf(enic))
2072                 enic_dev_del_station_addr(enic);
2073
2074         for (i = 0; i < enic->wq_count; i++) {
2075                 err = vnic_wq_disable(&enic->wq[i].vwq);
2076                 if (err)
2077                         return err;
2078         }
2079         for (i = 0; i < enic->rq_count; i++) {
2080                 err = vnic_rq_disable(&enic->rq[i].vrq);
2081                 if (err)
2082                         return err;
2083         }
2084
2085         enic_dev_notify_unset(enic);
2086         enic_unset_affinity_hint(enic);
2087         enic_free_intr(enic);
2088
2089         for (i = 0; i < enic->wq_count; i++)
2090                 vnic_wq_clean(&enic->wq[i].vwq, enic_free_wq_buf);
2091         for (i = 0; i < enic->rq_count; i++)
2092                 vnic_rq_clean(&enic->rq[i].vrq, enic_free_rq_buf);
2093         for (i = 0; i < enic->cq_count; i++)
2094                 vnic_cq_clean(&enic->cq[i]);
2095         for (i = 0; i < enic->intr_count; i++)
2096                 vnic_intr_clean(&enic->intr[i]);
2097
2098         return 0;
2099 }
2100
2101 static int _enic_change_mtu(struct net_device *netdev, int new_mtu)
2102 {
2103         bool running = netif_running(netdev);
2104         int err = 0;
2105
2106         ASSERT_RTNL();
2107         if (running) {
2108                 err = enic_stop(netdev);
2109                 if (err)
2110                         return err;
2111         }
2112
2113         WRITE_ONCE(netdev->mtu, new_mtu);
2114
2115         if (running) {
2116                 err = enic_open(netdev);
2117                 if (err)
2118                         return err;
2119         }
2120
2121         return 0;
2122 }
2123
2124 static int enic_change_mtu(struct net_device *netdev, int new_mtu)
2125 {
2126         struct enic *enic = netdev_priv(netdev);
2127
2128         if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic))
2129                 return -EOPNOTSUPP;
2130
2131         if (netdev->mtu > enic->port_mtu)
2132                 netdev_warn(netdev,
2133                             "interface MTU (%d) set higher than port MTU (%d)\n",
2134                             netdev->mtu, enic->port_mtu);
2135
2136         return _enic_change_mtu(netdev, new_mtu);
2137 }
2138
2139 static void enic_change_mtu_work(struct work_struct *work)
2140 {
2141         struct enic *enic = container_of(work, struct enic, change_mtu_work);
2142         struct net_device *netdev = enic->netdev;
2143         int new_mtu = vnic_dev_mtu(enic->vdev);
2144
2145         rtnl_lock();
2146         (void)_enic_change_mtu(netdev, new_mtu);
2147         rtnl_unlock();
2148
2149         netdev_info(netdev, "interface MTU set as %d\n", netdev->mtu);
2150 }
2151
2152 #ifdef CONFIG_NET_POLL_CONTROLLER
2153 static void enic_poll_controller(struct net_device *netdev)
2154 {
2155         struct enic *enic = netdev_priv(netdev);
2156         struct vnic_dev *vdev = enic->vdev;
2157         unsigned int i, intr;
2158
2159         switch (vnic_dev_get_intr_mode(vdev)) {
2160         case VNIC_DEV_INTR_MODE_MSIX:
2161                 for (i = 0; i < enic->rq_count; i++) {
2162                         intr = enic_msix_rq_intr(enic, i);
2163                         enic_isr_msix(enic->msix_entry[intr].vector,
2164                                       &enic->napi[i]);
2165                 }
2166
2167                 for (i = 0; i < enic->wq_count; i++) {
2168                         intr = enic_msix_wq_intr(enic, i);
2169                         enic_isr_msix(enic->msix_entry[intr].vector,
2170                                       &enic->napi[enic_cq_wq(enic, i)]);
2171                 }
2172
2173                 break;
2174         case VNIC_DEV_INTR_MODE_MSI:
2175                 enic_isr_msi(enic->pdev->irq, enic);
2176                 break;
2177         case VNIC_DEV_INTR_MODE_INTX:
2178                 enic_isr_legacy(enic->pdev->irq, netdev);
2179                 break;
2180         default:
2181                 break;
2182         }
2183 }
2184 #endif
2185
2186 static int enic_dev_wait(struct vnic_dev *vdev,
2187         int (*start)(struct vnic_dev *, int),
2188         int (*finished)(struct vnic_dev *, int *),
2189         int arg)
2190 {
2191         unsigned long time;
2192         int done;
2193         int err;
2194
2195         err = start(vdev, arg);
2196         if (err)
2197                 return err;
2198
2199         /* Wait for func to complete...2 seconds max
2200          */
2201
2202         time = jiffies + (HZ * 2);
2203         do {
2204
2205                 err = finished(vdev, &done);
2206                 if (err)
2207                         return err;
2208
2209                 if (done)
2210                         return 0;
2211
2212                 schedule_timeout_uninterruptible(HZ / 10);
2213
2214         } while (time_after(time, jiffies));
2215
2216         return -ETIMEDOUT;
2217 }
2218
2219 static int enic_dev_open(struct enic *enic)
2220 {
2221         int err;
2222         u32 flags = CMD_OPENF_IG_DESCCACHE;
2223
2224         err = enic_dev_wait(enic->vdev, vnic_dev_open,
2225                 vnic_dev_open_done, flags);
2226         if (err)
2227                 dev_err(enic_get_dev(enic), "vNIC device open failed, err %d\n",
2228                         err);
2229
2230         return err;
2231 }
2232
2233 static int enic_dev_soft_reset(struct enic *enic)
2234 {
2235         int err;
2236
2237         err = enic_dev_wait(enic->vdev, vnic_dev_soft_reset,
2238                             vnic_dev_soft_reset_done, 0);
2239         if (err)
2240                 netdev_err(enic->netdev, "vNIC soft reset failed, err %d\n",
2241                            err);
2242
2243         return err;
2244 }
2245
2246 static int enic_dev_hang_reset(struct enic *enic)
2247 {
2248         int err;
2249
2250         err = enic_dev_wait(enic->vdev, vnic_dev_hang_reset,
2251                 vnic_dev_hang_reset_done, 0);
2252         if (err)
2253                 netdev_err(enic->netdev, "vNIC hang reset failed, err %d\n",
2254                         err);
2255
2256         return err;
2257 }
2258
2259 int __enic_set_rsskey(struct enic *enic)
2260 {
2261         union vnic_rss_key *rss_key_buf_va;
2262         dma_addr_t rss_key_buf_pa;
2263         int i, kidx, bidx, err;
2264
2265         rss_key_buf_va = dma_alloc_coherent(&enic->pdev->dev,
2266                                             sizeof(union vnic_rss_key),
2267                                             &rss_key_buf_pa, GFP_ATOMIC);
2268         if (!rss_key_buf_va)
2269                 return -ENOMEM;
2270
2271         for (i = 0; i < ENIC_RSS_LEN; i++) {
2272                 kidx = i / ENIC_RSS_BYTES_PER_KEY;
2273                 bidx = i % ENIC_RSS_BYTES_PER_KEY;
2274                 rss_key_buf_va->key[kidx].b[bidx] = enic->rss_key[i];
2275         }
2276         spin_lock_bh(&enic->devcmd_lock);
2277         err = enic_set_rss_key(enic,
2278                 rss_key_buf_pa,
2279                 sizeof(union vnic_rss_key));
2280         spin_unlock_bh(&enic->devcmd_lock);
2281
2282         dma_free_coherent(&enic->pdev->dev, sizeof(union vnic_rss_key),
2283                           rss_key_buf_va, rss_key_buf_pa);
2284
2285         return err;
2286 }
2287
2288 static int enic_set_rsskey(struct enic *enic)
2289 {
2290         netdev_rss_key_fill(enic->rss_key, ENIC_RSS_LEN);
2291
2292         return __enic_set_rsskey(enic);
2293 }
2294
2295 static int enic_set_rsscpu(struct enic *enic, u8 rss_hash_bits)
2296 {
2297         dma_addr_t rss_cpu_buf_pa;
2298         union vnic_rss_cpu *rss_cpu_buf_va = NULL;
2299         unsigned int i;
2300         int err;
2301
2302         rss_cpu_buf_va = dma_alloc_coherent(&enic->pdev->dev,
2303                                             sizeof(union vnic_rss_cpu),
2304                                             &rss_cpu_buf_pa, GFP_ATOMIC);
2305         if (!rss_cpu_buf_va)
2306                 return -ENOMEM;
2307
2308         for (i = 0; i < (1 << rss_hash_bits); i++)
2309                 (*rss_cpu_buf_va).cpu[i/4].b[i%4] = i % enic->rq_count;
2310
2311         spin_lock_bh(&enic->devcmd_lock);
2312         err = enic_set_rss_cpu(enic,
2313                 rss_cpu_buf_pa,
2314                 sizeof(union vnic_rss_cpu));
2315         spin_unlock_bh(&enic->devcmd_lock);
2316
2317         dma_free_coherent(&enic->pdev->dev, sizeof(union vnic_rss_cpu),
2318                           rss_cpu_buf_va, rss_cpu_buf_pa);
2319
2320         return err;
2321 }
2322
2323 static int enic_set_niccfg(struct enic *enic, u8 rss_default_cpu,
2324         u8 rss_hash_type, u8 rss_hash_bits, u8 rss_base_cpu, u8 rss_enable)
2325 {
2326         const u8 tso_ipid_split_en = 0;
2327         const u8 ig_vlan_strip_en = 1;
2328         int err;
2329
2330         /* Enable VLAN tag stripping.
2331         */
2332
2333         spin_lock_bh(&enic->devcmd_lock);
2334         err = enic_set_nic_cfg(enic,
2335                 rss_default_cpu, rss_hash_type,
2336                 rss_hash_bits, rss_base_cpu,
2337                 rss_enable, tso_ipid_split_en,
2338                 ig_vlan_strip_en);
2339         spin_unlock_bh(&enic->devcmd_lock);
2340
2341         return err;
2342 }
2343
2344 static int enic_set_rss_nic_cfg(struct enic *enic)
2345 {
2346         struct device *dev = enic_get_dev(enic);
2347         const u8 rss_default_cpu = 0;
2348         const u8 rss_hash_bits = 7;
2349         const u8 rss_base_cpu = 0;
2350         u8 rss_hash_type;
2351         int res;
2352         u8 rss_enable = ENIC_SETTING(enic, RSS) && (enic->rq_count > 1);
2353
2354         spin_lock_bh(&enic->devcmd_lock);
2355         res = vnic_dev_capable_rss_hash_type(enic->vdev, &rss_hash_type);
2356         spin_unlock_bh(&enic->devcmd_lock);
2357         if (res) {
2358                 /* defaults for old adapters
2359                  */
2360                 rss_hash_type = NIC_CFG_RSS_HASH_TYPE_IPV4      |
2361                                 NIC_CFG_RSS_HASH_TYPE_TCP_IPV4  |
2362                                 NIC_CFG_RSS_HASH_TYPE_IPV6      |
2363                                 NIC_CFG_RSS_HASH_TYPE_TCP_IPV6;
2364         }
2365
2366         if (rss_enable) {
2367                 if (!enic_set_rsskey(enic)) {
2368                         if (enic_set_rsscpu(enic, rss_hash_bits)) {
2369                                 rss_enable = 0;
2370                                 dev_warn(dev, "RSS disabled, "
2371                                         "Failed to set RSS cpu indirection table.");
2372                         }
2373                 } else {
2374                         rss_enable = 0;
2375                         dev_warn(dev, "RSS disabled, Failed to set RSS key.\n");
2376                 }
2377         }
2378
2379         return enic_set_niccfg(enic, rss_default_cpu, rss_hash_type,
2380                 rss_hash_bits, rss_base_cpu, rss_enable);
2381 }
2382
2383 static void enic_set_api_busy(struct enic *enic, bool busy)
2384 {
2385         spin_lock(&enic->enic_api_lock);
2386         enic->enic_api_busy = busy;
2387         spin_unlock(&enic->enic_api_lock);
2388 }
2389
2390 static void enic_reset(struct work_struct *work)
2391 {
2392         struct enic *enic = container_of(work, struct enic, reset);
2393
2394         if (!netif_running(enic->netdev))
2395                 return;
2396
2397         rtnl_lock();
2398
2399         /* Stop any activity from infiniband */
2400         enic_set_api_busy(enic, true);
2401
2402         enic_stop(enic->netdev);
2403         enic_dev_soft_reset(enic);
2404         enic_reset_addr_lists(enic);
2405         enic_init_vnic_resources(enic);
2406         enic_set_rss_nic_cfg(enic);
2407         enic_dev_set_ig_vlan_rewrite_mode(enic);
2408         enic_open(enic->netdev);
2409
2410         /* Allow infiniband to fiddle with the device again */
2411         enic_set_api_busy(enic, false);
2412
2413         call_netdevice_notifiers(NETDEV_REBOOT, enic->netdev);
2414
2415         rtnl_unlock();
2416 }
2417
2418 static void enic_tx_hang_reset(struct work_struct *work)
2419 {
2420         struct enic *enic = container_of(work, struct enic, tx_hang_reset);
2421
2422         rtnl_lock();
2423
2424         /* Stop any activity from infiniband */
2425         enic_set_api_busy(enic, true);
2426
2427         enic_dev_hang_notify(enic);
2428         enic_stop(enic->netdev);
2429         enic_dev_hang_reset(enic);
2430         enic_reset_addr_lists(enic);
2431         enic_init_vnic_resources(enic);
2432         enic_set_rss_nic_cfg(enic);
2433         enic_dev_set_ig_vlan_rewrite_mode(enic);
2434         enic_open(enic->netdev);
2435
2436         /* Allow infiniband to fiddle with the device again */
2437         enic_set_api_busy(enic, false);
2438
2439         call_netdevice_notifiers(NETDEV_REBOOT, enic->netdev);
2440
2441         rtnl_unlock();
2442 }
2443
2444 static int enic_set_intr_mode(struct enic *enic)
2445 {
2446         unsigned int i;
2447         int num_intr;
2448
2449         /* Set interrupt mode (INTx, MSI, MSI-X) depending
2450          * on system capabilities.
2451          *
2452          * Try MSI-X first
2453          */
2454
2455         if (enic->config.intr_mode < 1 &&
2456             enic->intr_avail >= ENIC_MSIX_MIN_INTR) {
2457                 for (i = 0; i < enic->intr_avail; i++)
2458                         enic->msix_entry[i].entry = i;
2459
2460                 num_intr = pci_enable_msix_range(enic->pdev, enic->msix_entry,
2461                                                  ENIC_MSIX_MIN_INTR,
2462                                                  enic->intr_avail);
2463                 if (num_intr > 0) {
2464                         vnic_dev_set_intr_mode(enic->vdev,
2465                                                VNIC_DEV_INTR_MODE_MSIX);
2466                         enic->intr_avail = num_intr;
2467                         return 0;
2468                 }
2469         }
2470
2471         /* Next try MSI
2472          *
2473          * We need 1 INTR
2474          */
2475
2476         if (enic->config.intr_mode < 2 &&
2477             enic->intr_avail >= 1 &&
2478             !pci_enable_msi(enic->pdev)) {
2479                 enic->intr_avail = 1;
2480                 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSI);
2481                 return 0;
2482         }
2483
2484         /* Next try INTx
2485          *
2486          * We need 3 INTRs
2487          * (the first INTR is used for WQ/RQ)
2488          * (the second INTR is used for WQ/RQ errors)
2489          * (the last INTR is used for notifications)
2490          */
2491
2492         if (enic->config.intr_mode < 3 &&
2493             enic->intr_avail >= 3) {
2494                 enic->intr_avail = 3;
2495                 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_INTX);
2496                 return 0;
2497         }
2498
2499         vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
2500
2501         return -EINVAL;
2502 }
2503
2504 static void enic_clear_intr_mode(struct enic *enic)
2505 {
2506         switch (vnic_dev_get_intr_mode(enic->vdev)) {
2507         case VNIC_DEV_INTR_MODE_MSIX:
2508                 pci_disable_msix(enic->pdev);
2509                 break;
2510         case VNIC_DEV_INTR_MODE_MSI:
2511                 pci_disable_msi(enic->pdev);
2512                 break;
2513         default:
2514                 break;
2515         }
2516
2517         vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
2518 }
2519
2520 static int enic_adjust_resources(struct enic *enic)
2521 {
2522         unsigned int max_queues;
2523         unsigned int rq_default;
2524         unsigned int rq_avail;
2525         unsigned int wq_avail;
2526
2527         if (enic->rq_avail < 1 || enic->wq_avail < 1 || enic->cq_avail < 2) {
2528                 dev_err(enic_get_dev(enic),
2529                         "Not enough resources available rq: %d wq: %d cq: %d\n",
2530                         enic->rq_avail, enic->wq_avail,
2531                         enic->cq_avail);
2532                 return -ENOSPC;
2533         }
2534
2535         if (is_kdump_kernel()) {
2536                 dev_info(enic_get_dev(enic), "Running from within kdump kernel. Using minimal resources\n");
2537                 enic->rq_avail = 1;
2538                 enic->wq_avail = 1;
2539                 enic->config.rq_desc_count = ENIC_MIN_RQ_DESCS;
2540                 enic->config.wq_desc_count = ENIC_MIN_WQ_DESCS;
2541                 enic->config.mtu = min_t(u16, 1500, enic->config.mtu);
2542         }
2543
2544         /* if RSS isn't set, then we can only use one RQ */
2545         if (!ENIC_SETTING(enic, RSS))
2546                 enic->rq_avail = 1;
2547
2548         switch (vnic_dev_get_intr_mode(enic->vdev)) {
2549         case VNIC_DEV_INTR_MODE_INTX:
2550         case VNIC_DEV_INTR_MODE_MSI:
2551                 enic->rq_count = 1;
2552                 enic->wq_count = 1;
2553                 enic->cq_count = 2;
2554                 enic->intr_count = enic->intr_avail;
2555                 break;
2556         case VNIC_DEV_INTR_MODE_MSIX:
2557                 /* Adjust the number of wqs/rqs/cqs/interrupts that will be
2558                  * used based on which resource is the most constrained
2559                  */
2560                 wq_avail = min(enic->wq_avail, ENIC_WQ_MAX);
2561                 rq_default = netif_get_num_default_rss_queues();
2562                 rq_avail = min3(enic->rq_avail, ENIC_RQ_MAX, rq_default);
2563                 max_queues = min(enic->cq_avail,
2564                                  enic->intr_avail - ENIC_MSIX_RESERVED_INTR);
2565                 if (wq_avail + rq_avail <= max_queues) {
2566                         enic->rq_count = rq_avail;
2567                         enic->wq_count = wq_avail;
2568                 } else {
2569                         /* recalculate wq/rq count */
2570                         if (rq_avail < wq_avail) {
2571                                 enic->rq_count = min(rq_avail, max_queues / 2);
2572                                 enic->wq_count = max_queues - enic->rq_count;
2573                         } else {
2574                                 enic->wq_count = min(wq_avail, max_queues / 2);
2575                                 enic->rq_count = max_queues - enic->wq_count;
2576                         }
2577                 }
2578                 enic->cq_count = enic->rq_count + enic->wq_count;
2579                 enic->intr_count = enic->cq_count + ENIC_MSIX_RESERVED_INTR;
2580
2581                 break;
2582         default:
2583                 dev_err(enic_get_dev(enic), "Unknown interrupt mode\n");
2584                 return -EINVAL;
2585         }
2586
2587         return 0;
2588 }
2589
2590 static void enic_get_queue_stats_rx(struct net_device *dev, int idx,
2591                                     struct netdev_queue_stats_rx *rxs)
2592 {
2593         struct enic *enic = netdev_priv(dev);
2594         struct enic_rq_stats *rqstats = &enic->rq[idx].stats;
2595
2596         rxs->bytes = rqstats->bytes;
2597         rxs->packets = rqstats->packets;
2598         rxs->hw_drops = rqstats->bad_fcs + rqstats->pkt_truncated;
2599         rxs->hw_drop_overruns = rqstats->pkt_truncated;
2600         rxs->csum_unnecessary = rqstats->csum_unnecessary +
2601                                 rqstats->csum_unnecessary_encap;
2602 }
2603
2604 static void enic_get_queue_stats_tx(struct net_device *dev, int idx,
2605                                     struct netdev_queue_stats_tx *txs)
2606 {
2607         struct enic *enic = netdev_priv(dev);
2608         struct enic_wq_stats *wqstats = &enic->wq[idx].stats;
2609
2610         txs->bytes = wqstats->bytes;
2611         txs->packets = wqstats->packets;
2612         txs->csum_none = wqstats->csum_none;
2613         txs->needs_csum = wqstats->csum_partial + wqstats->encap_csum +
2614                           wqstats->tso;
2615         txs->hw_gso_packets = wqstats->tso;
2616         txs->stop = wqstats->stopped;
2617         txs->wake = wqstats->wake;
2618 }
2619
2620 static void enic_get_base_stats(struct net_device *dev,
2621                                 struct netdev_queue_stats_rx *rxs,
2622                                 struct netdev_queue_stats_tx *txs)
2623 {
2624         rxs->bytes = 0;
2625         rxs->packets = 0;
2626         rxs->hw_drops = 0;
2627         rxs->hw_drop_overruns = 0;
2628         rxs->csum_unnecessary = 0;
2629         txs->bytes = 0;
2630         txs->packets = 0;
2631         txs->csum_none = 0;
2632         txs->needs_csum = 0;
2633         txs->hw_gso_packets = 0;
2634         txs->stop = 0;
2635         txs->wake = 0;
2636 }
2637
2638 static const struct net_device_ops enic_netdev_dynamic_ops = {
2639         .ndo_open               = enic_open,
2640         .ndo_stop               = enic_stop,
2641         .ndo_start_xmit         = enic_hard_start_xmit,
2642         .ndo_get_stats64        = enic_get_stats,
2643         .ndo_validate_addr      = eth_validate_addr,
2644         .ndo_set_rx_mode        = enic_set_rx_mode,
2645         .ndo_set_mac_address    = enic_set_mac_address_dynamic,
2646         .ndo_change_mtu         = enic_change_mtu,
2647         .ndo_vlan_rx_add_vid    = enic_vlan_rx_add_vid,
2648         .ndo_vlan_rx_kill_vid   = enic_vlan_rx_kill_vid,
2649         .ndo_tx_timeout         = enic_tx_timeout,
2650         .ndo_set_vf_port        = enic_set_vf_port,
2651         .ndo_get_vf_port        = enic_get_vf_port,
2652         .ndo_set_vf_mac         = enic_set_vf_mac,
2653 #ifdef CONFIG_NET_POLL_CONTROLLER
2654         .ndo_poll_controller    = enic_poll_controller,
2655 #endif
2656 #ifdef CONFIG_RFS_ACCEL
2657         .ndo_rx_flow_steer      = enic_rx_flow_steer,
2658 #endif
2659         .ndo_features_check     = enic_features_check,
2660 };
2661
2662 static const struct net_device_ops enic_netdev_ops = {
2663         .ndo_open               = enic_open,
2664         .ndo_stop               = enic_stop,
2665         .ndo_start_xmit         = enic_hard_start_xmit,
2666         .ndo_get_stats64        = enic_get_stats,
2667         .ndo_validate_addr      = eth_validate_addr,
2668         .ndo_set_mac_address    = enic_set_mac_address,
2669         .ndo_set_rx_mode        = enic_set_rx_mode,
2670         .ndo_change_mtu         = enic_change_mtu,
2671         .ndo_vlan_rx_add_vid    = enic_vlan_rx_add_vid,
2672         .ndo_vlan_rx_kill_vid   = enic_vlan_rx_kill_vid,
2673         .ndo_tx_timeout         = enic_tx_timeout,
2674         .ndo_set_vf_port        = enic_set_vf_port,
2675         .ndo_get_vf_port        = enic_get_vf_port,
2676         .ndo_set_vf_mac         = enic_set_vf_mac,
2677 #ifdef CONFIG_NET_POLL_CONTROLLER
2678         .ndo_poll_controller    = enic_poll_controller,
2679 #endif
2680 #ifdef CONFIG_RFS_ACCEL
2681         .ndo_rx_flow_steer      = enic_rx_flow_steer,
2682 #endif
2683         .ndo_features_check     = enic_features_check,
2684 };
2685
2686 static const struct netdev_stat_ops enic_netdev_stat_ops = {
2687         .get_queue_stats_rx     = enic_get_queue_stats_rx,
2688         .get_queue_stats_tx     = enic_get_queue_stats_tx,
2689         .get_base_stats         = enic_get_base_stats,
2690 };
2691
2692 static void enic_free_enic_resources(struct enic *enic)
2693 {
2694         kfree(enic->wq);
2695         enic->wq = NULL;
2696
2697         kfree(enic->rq);
2698         enic->rq = NULL;
2699
2700         kfree(enic->cq);
2701         enic->cq = NULL;
2702
2703         kfree(enic->napi);
2704         enic->napi = NULL;
2705
2706         kfree(enic->msix_entry);
2707         enic->msix_entry = NULL;
2708
2709         kfree(enic->msix);
2710         enic->msix = NULL;
2711
2712         kfree(enic->intr);
2713         enic->intr = NULL;
2714 }
2715
2716 static int enic_alloc_enic_resources(struct enic *enic)
2717 {
2718         enic->wq = kcalloc(enic->wq_avail, sizeof(struct enic_wq), GFP_KERNEL);
2719         if (!enic->wq)
2720                 goto free_queues;
2721
2722         enic->rq = kcalloc(enic->rq_avail, sizeof(struct enic_rq), GFP_KERNEL);
2723         if (!enic->rq)
2724                 goto free_queues;
2725
2726         enic->cq = kcalloc(enic->cq_avail, sizeof(struct vnic_cq), GFP_KERNEL);
2727         if (!enic->cq)
2728                 goto free_queues;
2729
2730         enic->napi = kcalloc(enic->wq_avail + enic->rq_avail,
2731                              sizeof(struct napi_struct), GFP_KERNEL);
2732         if (!enic->napi)
2733                 goto free_queues;
2734
2735         enic->msix_entry = kcalloc(enic->intr_avail, sizeof(struct msix_entry),
2736                                    GFP_KERNEL);
2737         if (!enic->msix_entry)
2738                 goto free_queues;
2739
2740         enic->msix = kcalloc(enic->intr_avail, sizeof(struct enic_msix_entry),
2741                              GFP_KERNEL);
2742         if (!enic->msix)
2743                 goto free_queues;
2744
2745         enic->intr = kcalloc(enic->intr_avail, sizeof(struct vnic_intr),
2746                              GFP_KERNEL);
2747         if (!enic->intr)
2748                 goto free_queues;
2749
2750         return 0;
2751
2752 free_queues:
2753         enic_free_enic_resources(enic);
2754         return -ENOMEM;
2755 }
2756
2757 static void enic_dev_deinit(struct enic *enic)
2758 {
2759         unsigned int i;
2760
2761         for (i = 0; i < enic->rq_count; i++)
2762                 __netif_napi_del(&enic->napi[i]);
2763
2764         if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX)
2765                 for (i = 0; i < enic->wq_count; i++)
2766                         __netif_napi_del(&enic->napi[enic_cq_wq(enic, i)]);
2767
2768         /* observe RCU grace period after __netif_napi_del() calls */
2769         synchronize_net();
2770
2771         enic_free_vnic_resources(enic);
2772         enic_clear_intr_mode(enic);
2773         enic_free_affinity_hint(enic);
2774         enic_free_enic_resources(enic);
2775 }
2776
2777 static int enic_dev_init(struct enic *enic)
2778 {
2779         struct device *dev = enic_get_dev(enic);
2780         struct net_device *netdev = enic->netdev;
2781         unsigned int i;
2782         int err;
2783
2784         /* Get interrupt coalesce timer info */
2785         err = enic_dev_intr_coal_timer_info(enic);
2786         if (err) {
2787                 dev_warn(dev, "Using default conversion factor for "
2788                         "interrupt coalesce timer\n");
2789                 vnic_dev_intr_coal_timer_info_default(enic->vdev);
2790         }
2791
2792         /* Get vNIC configuration
2793          */
2794
2795         err = enic_get_vnic_config(enic);
2796         if (err) {
2797                 dev_err(dev, "Get vNIC configuration failed, aborting\n");
2798                 return err;
2799         }
2800
2801         /* Get available resource counts
2802          */
2803
2804         enic_get_res_counts(enic);
2805
2806         err = enic_alloc_enic_resources(enic);
2807         if (err) {
2808                 dev_err(dev, "Failed to allocate enic resources\n");
2809                 return err;
2810         }
2811
2812         /* Set interrupt mode based on system capabilities */
2813
2814         err = enic_set_intr_mode(enic);
2815         if (err) {
2816                 dev_err(dev, "Failed to set intr mode based on resource "
2817                         "counts and system capabilities, aborting\n");
2818                 goto err_out_free_vnic_resources;
2819         }
2820
2821         /* Adjust resource counts based on most constrained resources */
2822         err = enic_adjust_resources(enic);
2823         if (err) {
2824                 dev_err(dev, "Failed to adjust resources\n");
2825                 goto err_out_free_vnic_resources;
2826         }
2827
2828         /* Allocate and configure vNIC resources
2829          */
2830
2831         err = enic_alloc_vnic_resources(enic);
2832         if (err) {
2833                 dev_err(dev, "Failed to alloc vNIC resources, aborting\n");
2834                 goto err_out_free_vnic_resources;
2835         }
2836
2837         enic_init_vnic_resources(enic);
2838
2839         err = enic_set_rss_nic_cfg(enic);
2840         if (err) {
2841                 dev_err(dev, "Failed to config nic, aborting\n");
2842                 goto err_out_free_vnic_resources;
2843         }
2844
2845         switch (vnic_dev_get_intr_mode(enic->vdev)) {
2846         default:
2847                 netif_napi_add(netdev, &enic->napi[0], enic_poll);
2848                 break;
2849         case VNIC_DEV_INTR_MODE_MSIX:
2850                 for (i = 0; i < enic->rq_count; i++) {
2851                         netif_napi_add(netdev, &enic->napi[i],
2852                                        enic_poll_msix_rq);
2853                 }
2854                 for (i = 0; i < enic->wq_count; i++)
2855                         netif_napi_add(netdev,
2856                                        &enic->napi[enic_cq_wq(enic, i)],
2857                                        enic_poll_msix_wq);
2858                 break;
2859         }
2860
2861         return 0;
2862
2863 err_out_free_vnic_resources:
2864         enic_free_affinity_hint(enic);
2865         enic_clear_intr_mode(enic);
2866         enic_free_vnic_resources(enic);
2867         enic_free_enic_resources(enic);
2868
2869         return err;
2870 }
2871
2872 static void enic_iounmap(struct enic *enic)
2873 {
2874         unsigned int i;
2875
2876         for (i = 0; i < ARRAY_SIZE(enic->bar); i++)
2877                 if (enic->bar[i].vaddr)
2878                         iounmap(enic->bar[i].vaddr);
2879 }
2880
2881 static int enic_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2882 {
2883         struct device *dev = &pdev->dev;
2884         struct net_device *netdev;
2885         struct enic *enic;
2886         int using_dac = 0;
2887         unsigned int i;
2888         int err;
2889 #ifdef CONFIG_PCI_IOV
2890         int pos = 0;
2891 #endif
2892         int num_pps = 1;
2893
2894         /* Allocate net device structure and initialize.  Private
2895          * instance data is initialized to zero.
2896          */
2897
2898         netdev = alloc_etherdev_mqs(sizeof(struct enic),
2899                                     ENIC_RQ_MAX, ENIC_WQ_MAX);
2900         if (!netdev)
2901                 return -ENOMEM;
2902
2903         pci_set_drvdata(pdev, netdev);
2904
2905         SET_NETDEV_DEV(netdev, &pdev->dev);
2906
2907         enic = netdev_priv(netdev);
2908         enic->netdev = netdev;
2909         enic->pdev = pdev;
2910
2911         /* Setup PCI resources
2912          */
2913
2914         err = pci_enable_device_mem(pdev);
2915         if (err) {
2916                 dev_err(dev, "Cannot enable PCI device, aborting\n");
2917                 goto err_out_free_netdev;
2918         }
2919
2920         err = pci_request_regions(pdev, DRV_NAME);
2921         if (err) {
2922                 dev_err(dev, "Cannot request PCI regions, aborting\n");
2923                 goto err_out_disable_device;
2924         }
2925
2926         pci_set_master(pdev);
2927
2928         /* Query PCI controller on system for DMA addressing
2929          * limitation for the device.  Try 47-bit first, and
2930          * fail to 32-bit.
2931          */
2932
2933         err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(47));
2934         if (err) {
2935                 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
2936                 if (err) {
2937                         dev_err(dev, "No usable DMA configuration, aborting\n");
2938                         goto err_out_release_regions;
2939                 }
2940         } else {
2941                 using_dac = 1;
2942         }
2943
2944         /* Map vNIC resources from BAR0-5
2945          */
2946
2947         for (i = 0; i < ARRAY_SIZE(enic->bar); i++) {
2948                 if (!(pci_resource_flags(pdev, i) & IORESOURCE_MEM))
2949                         continue;
2950                 enic->bar[i].len = pci_resource_len(pdev, i);
2951                 enic->bar[i].vaddr = pci_iomap(pdev, i, enic->bar[i].len);
2952                 if (!enic->bar[i].vaddr) {
2953                         dev_err(dev, "Cannot memory-map BAR %d, aborting\n", i);
2954                         err = -ENODEV;
2955                         goto err_out_iounmap;
2956                 }
2957                 enic->bar[i].bus_addr = pci_resource_start(pdev, i);
2958         }
2959
2960         /* Register vNIC device
2961          */
2962
2963         enic->vdev = vnic_dev_register(NULL, enic, pdev, enic->bar,
2964                 ARRAY_SIZE(enic->bar));
2965         if (!enic->vdev) {
2966                 dev_err(dev, "vNIC registration failed, aborting\n");
2967                 err = -ENODEV;
2968                 goto err_out_iounmap;
2969         }
2970
2971         err = vnic_devcmd_init(enic->vdev);
2972
2973         if (err)
2974                 goto err_out_vnic_unregister;
2975
2976 #ifdef CONFIG_PCI_IOV
2977         /* Get number of subvnics */
2978         pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_SRIOV);
2979         if (pos) {
2980                 pci_read_config_word(pdev, pos + PCI_SRIOV_TOTAL_VF,
2981                         &enic->num_vfs);
2982                 if (enic->num_vfs) {
2983                         err = pci_enable_sriov(pdev, enic->num_vfs);
2984                         if (err) {
2985                                 dev_err(dev, "SRIOV enable failed, aborting."
2986                                         " pci_enable_sriov() returned %d\n",
2987                                         err);
2988                                 goto err_out_vnic_unregister;
2989                         }
2990                         enic->priv_flags |= ENIC_SRIOV_ENABLED;
2991                         num_pps = enic->num_vfs;
2992                 }
2993         }
2994 #endif
2995
2996         /* Allocate structure for port profiles */
2997         enic->pp = kcalloc(num_pps, sizeof(*enic->pp), GFP_KERNEL);
2998         if (!enic->pp) {
2999                 err = -ENOMEM;
3000                 goto err_out_disable_sriov_pp;
3001         }
3002
3003         /* Issue device open to get device in known state
3004          */
3005
3006         err = enic_dev_open(enic);
3007         if (err) {
3008                 dev_err(dev, "vNIC dev open failed, aborting\n");
3009                 goto err_out_disable_sriov;
3010         }
3011
3012         /* Setup devcmd lock
3013          */
3014
3015         spin_lock_init(&enic->devcmd_lock);
3016         spin_lock_init(&enic->enic_api_lock);
3017
3018         /*
3019          * Set ingress vlan rewrite mode before vnic initialization
3020          */
3021
3022         err = enic_dev_set_ig_vlan_rewrite_mode(enic);
3023         if (err) {
3024                 dev_err(dev,
3025                         "Failed to set ingress vlan rewrite mode, aborting.\n");
3026                 goto err_out_dev_close;
3027         }
3028
3029         /* Issue device init to initialize the vnic-to-switch link.
3030          * We'll start with carrier off and wait for link UP
3031          * notification later to turn on carrier.  We don't need
3032          * to wait here for the vnic-to-switch link initialization
3033          * to complete; link UP notification is the indication that
3034          * the process is complete.
3035          */
3036
3037         netif_carrier_off(netdev);
3038
3039         /* Do not call dev_init for a dynamic vnic.
3040          * For a dynamic vnic, init_prov_info will be
3041          * called later by an upper layer.
3042          */
3043
3044         if (!enic_is_dynamic(enic)) {
3045                 err = vnic_dev_init(enic->vdev, 0);
3046                 if (err) {
3047                         dev_err(dev, "vNIC dev init failed, aborting\n");
3048                         goto err_out_dev_close;
3049                 }
3050         }
3051
3052         err = enic_dev_init(enic);
3053         if (err) {
3054                 dev_err(dev, "Device initialization failed, aborting\n");
3055                 goto err_out_dev_close;
3056         }
3057
3058         netif_set_real_num_tx_queues(netdev, enic->wq_count);
3059         netif_set_real_num_rx_queues(netdev, enic->rq_count);
3060
3061         /* Setup notification timer, HW reset task, and wq locks
3062          */
3063
3064         timer_setup(&enic->notify_timer, enic_notify_timer, 0);
3065
3066         enic_rfs_flw_tbl_init(enic);
3067         INIT_WORK(&enic->reset, enic_reset);
3068         INIT_WORK(&enic->tx_hang_reset, enic_tx_hang_reset);
3069         INIT_WORK(&enic->change_mtu_work, enic_change_mtu_work);
3070
3071         for (i = 0; i < enic->wq_count; i++)
3072                 spin_lock_init(&enic->wq[i].lock);
3073
3074         /* Register net device
3075          */
3076
3077         enic->port_mtu = enic->config.mtu;
3078
3079         err = enic_set_mac_addr(netdev, enic->mac_addr);
3080         if (err) {
3081                 dev_err(dev, "Invalid MAC address, aborting\n");
3082                 goto err_out_dev_deinit;
3083         }
3084
3085         enic->tx_coalesce_usecs = enic->config.intr_timer_usec;
3086         /* rx coalesce time already got initialized. This gets used
3087          * if adaptive coal is turned off
3088          */
3089         enic->rx_coalesce_usecs = enic->tx_coalesce_usecs;
3090
3091         if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic))
3092                 netdev->netdev_ops = &enic_netdev_dynamic_ops;
3093         else
3094                 netdev->netdev_ops = &enic_netdev_ops;
3095         netdev->stat_ops = &enic_netdev_stat_ops;
3096
3097         netdev->watchdog_timeo = 2 * HZ;
3098         enic_set_ethtool_ops(netdev);
3099
3100         netdev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
3101         if (ENIC_SETTING(enic, LOOP)) {
3102                 netdev->features &= ~NETIF_F_HW_VLAN_CTAG_TX;
3103                 enic->loop_enable = 1;
3104                 enic->loop_tag = enic->config.loop_tag;
3105                 dev_info(dev, "loopback tag=0x%04x\n", enic->loop_tag);
3106         }
3107         if (ENIC_SETTING(enic, TXCSUM))
3108                 netdev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM;
3109         if (ENIC_SETTING(enic, TSO))
3110                 netdev->hw_features |= NETIF_F_TSO |
3111                         NETIF_F_TSO6 | NETIF_F_TSO_ECN;
3112         if (ENIC_SETTING(enic, RSS))
3113                 netdev->hw_features |= NETIF_F_RXHASH;
3114         if (ENIC_SETTING(enic, RXCSUM))
3115                 netdev->hw_features |= NETIF_F_RXCSUM;
3116         if (ENIC_SETTING(enic, VXLAN)) {
3117                 u64 patch_level;
3118                 u64 a1 = 0;
3119
3120                 netdev->hw_enc_features |= NETIF_F_RXCSUM               |
3121                                            NETIF_F_TSO                  |
3122                                            NETIF_F_TSO6                 |
3123                                            NETIF_F_TSO_ECN              |
3124                                            NETIF_F_GSO_UDP_TUNNEL       |
3125                                            NETIF_F_HW_CSUM              |
3126                                            NETIF_F_GSO_UDP_TUNNEL_CSUM;
3127                 netdev->hw_features |= netdev->hw_enc_features;
3128                 /* get bit mask from hw about supported offload bit level
3129                  * BIT(0) = fw supports patch_level 0
3130                  *          fcoe bit = encap
3131                  *          fcoe_fc_crc_ok = outer csum ok
3132                  * BIT(1) = always set by fw
3133                  * BIT(2) = fw supports patch_level 2
3134                  *          BIT(0) in rss_hash = encap
3135                  *          BIT(1,2) in rss_hash = outer_ip_csum_ok/
3136                  *                                 outer_tcp_csum_ok
3137                  * used in enic_rq_indicate_buf
3138                  */
3139                 err = vnic_dev_get_supported_feature_ver(enic->vdev,
3140                                                          VIC_FEATURE_VXLAN,
3141                                                          &patch_level, &a1);
3142                 if (err)
3143                         patch_level = 0;
3144                 enic->vxlan.flags = (u8)a1;
3145                 /* mask bits that are supported by driver
3146                  */
3147                 patch_level &= BIT_ULL(0) | BIT_ULL(2);
3148                 patch_level = fls(patch_level);
3149                 patch_level = patch_level ? patch_level - 1 : 0;
3150                 enic->vxlan.patch_level = patch_level;
3151
3152                 if (vnic_dev_get_res_count(enic->vdev, RES_TYPE_WQ) == 1 ||
3153                     enic->vxlan.flags & ENIC_VXLAN_MULTI_WQ) {
3154                         netdev->udp_tunnel_nic_info = &enic_udp_tunnels_v4;
3155                         if (enic->vxlan.flags & ENIC_VXLAN_OUTER_IPV6)
3156                                 netdev->udp_tunnel_nic_info = &enic_udp_tunnels;
3157                 }
3158         }
3159
3160         netdev->features |= netdev->hw_features;
3161         netdev->vlan_features |= netdev->features;
3162
3163 #ifdef CONFIG_RFS_ACCEL
3164         netdev->hw_features |= NETIF_F_NTUPLE;
3165 #endif
3166
3167         if (using_dac)
3168                 netdev->features |= NETIF_F_HIGHDMA;
3169
3170         netdev->priv_flags |= IFF_UNICAST_FLT;
3171
3172         /* MTU range: 68 - 9000 */
3173         netdev->min_mtu = ENIC_MIN_MTU;
3174         netdev->max_mtu = ENIC_MAX_MTU;
3175         netdev->mtu     = enic->port_mtu;
3176
3177         err = register_netdev(netdev);
3178         if (err) {
3179                 dev_err(dev, "Cannot register net device, aborting\n");
3180                 goto err_out_dev_deinit;
3181         }
3182         enic->rx_copybreak = RX_COPYBREAK_DEFAULT;
3183
3184         return 0;
3185
3186 err_out_dev_deinit:
3187         enic_dev_deinit(enic);
3188 err_out_dev_close:
3189         vnic_dev_close(enic->vdev);
3190 err_out_disable_sriov:
3191         kfree(enic->pp);
3192 err_out_disable_sriov_pp:
3193 #ifdef CONFIG_PCI_IOV
3194         if (enic_sriov_enabled(enic)) {
3195                 pci_disable_sriov(pdev);
3196                 enic->priv_flags &= ~ENIC_SRIOV_ENABLED;
3197         }
3198 #endif
3199 err_out_vnic_unregister:
3200         vnic_dev_unregister(enic->vdev);
3201 err_out_iounmap:
3202         enic_iounmap(enic);
3203 err_out_release_regions:
3204         pci_release_regions(pdev);
3205 err_out_disable_device:
3206         pci_disable_device(pdev);
3207 err_out_free_netdev:
3208         free_netdev(netdev);
3209
3210         return err;
3211 }
3212
3213 static void enic_remove(struct pci_dev *pdev)
3214 {
3215         struct net_device *netdev = pci_get_drvdata(pdev);
3216
3217         if (netdev) {
3218                 struct enic *enic = netdev_priv(netdev);
3219
3220                 cancel_work_sync(&enic->reset);
3221                 cancel_work_sync(&enic->change_mtu_work);
3222                 unregister_netdev(netdev);
3223                 enic_dev_deinit(enic);
3224                 vnic_dev_close(enic->vdev);
3225 #ifdef CONFIG_PCI_IOV
3226                 if (enic_sriov_enabled(enic)) {
3227                         pci_disable_sriov(pdev);
3228                         enic->priv_flags &= ~ENIC_SRIOV_ENABLED;
3229                 }
3230 #endif
3231                 kfree(enic->pp);
3232                 vnic_dev_unregister(enic->vdev);
3233                 enic_iounmap(enic);
3234                 pci_release_regions(pdev);
3235                 pci_disable_device(pdev);
3236                 free_netdev(netdev);
3237         }
3238 }
3239
3240 static struct pci_driver enic_driver = {
3241         .name = DRV_NAME,
3242         .id_table = enic_id_table,
3243         .probe = enic_probe,
3244         .remove = enic_remove,
3245 };
3246
3247 module_pci_driver(enic_driver);