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