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_set_affinity_hint(enic->msix_entry[i].vector,
 154                                             enic->msix[i].affinity_mask);
 155                 if (err)
 156                         netdev_warn(enic->netdev, "irq_set_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_set_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                 pci_unmap_single(enic->pdev, buf->dma_addr,
 330                         buf->len, PCI_DMA_TODEVICE);
 331         else
 332                 pci_unmap_page(enic->pdev, buf->dma_addr,
 333                         buf->len, PCI_DMA_TODEVICE);
 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 = pci_map_single(enic->pdev, skb->data, head_len,
 578                                   PCI_DMA_TODEVICE);
 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 = pci_map_single(enic->pdev, skb->data, head_len,
 609                                   PCI_DMA_TODEVICE);
 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_transport_header(skb) - skb->data;
 684                 hdr_len += inner_tcp_hdrlen(skb);
 685                 enic_preload_tcp_csum_encap(skb);
 686         } else {
 687                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
 688                 enic_preload_tcp_csum(skb);
 689         }
 690
 691         /* Queue WQ_ENET_MAX_DESC_LEN length descriptors
 692          * for the main skb fragment
 693          */
 694         while (frag_len_left) {
 695                 len = min(frag_len_left, (unsigned int)WQ_ENET_MAX_DESC_LEN);
 696                 dma_addr = pci_map_single(enic->pdev, skb->data + offset, len,
 697                                           PCI_DMA_TODEVICE);
 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 = pci_map_single(enic->pdev, skb->data, head_len,
 756                                   PCI_DMA_TODEVICE);
 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 void 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 }
 817
 818 /* netif_tx_lock held, process context with BHs disabled, or BH */
 819 static netdev_tx_t enic_hard_start_xmit(struct sk_buff *skb,
 820         struct net_device *netdev)
 821 {
 822         struct enic *enic = netdev_priv(netdev);
 823         struct vnic_wq *wq;
 824         unsigned int txq_map;
 825         struct netdev_queue *txq;
 826
 827         if (skb->len <= 0) {
 828                 dev_kfree_skb_any(skb);
 829                 return NETDEV_TX_OK;
 830         }
 831
 832         txq_map = skb_get_queue_mapping(skb) % enic->wq_count;
 833         wq = &enic->wq[txq_map];
 834         txq = netdev_get_tx_queue(netdev, txq_map);
 835
 836         /* Non-TSO sends must fit within ENIC_NON_TSO_MAX_DESC descs,
 837          * which is very likely.  In the off chance it's going to take
 838          * more than * ENIC_NON_TSO_MAX_DESC, linearize the skb.
 839          */
 840
 841         if (skb_shinfo(skb)->gso_size == 0 &&
 842             skb_shinfo(skb)->nr_frags + 1 > ENIC_NON_TSO_MAX_DESC &&
 843             skb_linearize(skb)) {
 844                 dev_kfree_skb_any(skb);
 845                 return NETDEV_TX_OK;
 846         }
 847
 848         spin_lock(&enic->wq_lock[txq_map]);
 849
 850         if (vnic_wq_desc_avail(wq) <
 851             skb_shinfo(skb)->nr_frags + ENIC_DESC_MAX_SPLITS) {
 852                 netif_tx_stop_queue(txq);
 853                 /* This is a hard error, log it */
 854                 netdev_err(netdev, "BUG! Tx ring full when queue awake!\n");
 855                 spin_unlock(&enic->wq_lock[txq_map]);
 856                 return NETDEV_TX_BUSY;
 857         }
 858
 859         enic_queue_wq_skb(enic, wq, skb);
 860
 861         if (vnic_wq_desc_avail(wq) < MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS)
 862                 netif_tx_stop_queue(txq);
 863         skb_tx_timestamp(skb);
 864         if (!netdev_xmit_more() || netif_xmit_stopped(txq))
 865                 vnic_wq_doorbell(wq);
 866
 867         spin_unlock(&enic->wq_lock[txq_map]);
 868
 869         return NETDEV_TX_OK;
 870 }
 871
 872 /* dev_base_lock rwlock held, nominally process context */
 873 static void enic_get_stats(struct net_device *netdev,
 874                            struct rtnl_link_stats64 *net_stats)
 875 {
 876         struct enic *enic = netdev_priv(netdev);
 877         struct vnic_stats *stats;
 878         int err;
 879
 880         err = enic_dev_stats_dump(enic, &stats);
 881         /* return only when pci_zalloc_consistent fails in vnic_dev_stats_dump
 882          * For other failures, like devcmd failure, we return previously
 883          * recorded stats.
 884          */
 885         if (err == -ENOMEM)
 886                 return;
 887
 888         net_stats->tx_packets = stats->tx.tx_frames_ok;
 889         net_stats->tx_bytes = stats->tx.tx_bytes_ok;
 890         net_stats->tx_errors = stats->tx.tx_errors;
 891         net_stats->tx_dropped = stats->tx.tx_drops;
 892
 893         net_stats->rx_packets = stats->rx.rx_frames_ok;
 894         net_stats->rx_bytes = stats->rx.rx_bytes_ok;
 895         net_stats->rx_errors = stats->rx.rx_errors;
 896         net_stats->multicast = stats->rx.rx_multicast_frames_ok;
 897         net_stats->rx_over_errors = enic->rq_truncated_pkts;
 898         net_stats->rx_crc_errors = enic->rq_bad_fcs;
 899         net_stats->rx_dropped = stats->rx.rx_no_bufs + stats->rx.rx_drop;
 900 }
 901
 902 static int enic_mc_sync(struct net_device *netdev, const u8 *mc_addr)
 903 {
 904         struct enic *enic = netdev_priv(netdev);
 905
 906         if (enic->mc_count == ENIC_MULTICAST_PERFECT_FILTERS) {
 907                 unsigned int mc_count = netdev_mc_count(netdev);
 908
 909                 netdev_warn(netdev, "Registering only %d out of %d multicast addresses\n",
 910                             ENIC_MULTICAST_PERFECT_FILTERS, mc_count);
 911
 912                 return -ENOSPC;
 913         }
 914
 915         enic_dev_add_addr(enic, mc_addr);
 916         enic->mc_count++;
 917
 918         return 0;
 919 }
 920
 921 static int enic_mc_unsync(struct net_device *netdev, const u8 *mc_addr)
 922 {
 923         struct enic *enic = netdev_priv(netdev);
 924
 925         enic_dev_del_addr(enic, mc_addr);
 926         enic->mc_count--;
 927
 928         return 0;
 929 }
 930
 931 static int enic_uc_sync(struct net_device *netdev, const u8 *uc_addr)
 932 {
 933         struct enic *enic = netdev_priv(netdev);
 934
 935         if (enic->uc_count == ENIC_UNICAST_PERFECT_FILTERS) {
 936                 unsigned int uc_count = netdev_uc_count(netdev);
 937
 938                 netdev_warn(netdev, "Registering only %d out of %d unicast addresses\n",
 939                             ENIC_UNICAST_PERFECT_FILTERS, uc_count);
 940
 941                 return -ENOSPC;
 942         }
 943
 944         enic_dev_add_addr(enic, uc_addr);
 945         enic->uc_count++;
 946
 947         return 0;
 948 }
 949
 950 static int enic_uc_unsync(struct net_device *netdev, const u8 *uc_addr)
 951 {
 952         struct enic *enic = netdev_priv(netdev);
 953
 954         enic_dev_del_addr(enic, uc_addr);
 955         enic->uc_count--;
 956
 957         return 0;
 958 }
 959
 960 void enic_reset_addr_lists(struct enic *enic)
 961 {
 962         struct net_device *netdev = enic->netdev;
 963
 964         __dev_uc_unsync(netdev, NULL);
 965         __dev_mc_unsync(netdev, NULL);
 966
 967         enic->mc_count = 0;
 968         enic->uc_count = 0;
 969         enic->flags = 0;
 970 }
 971
 972 static int enic_set_mac_addr(struct net_device *netdev, char *addr)
 973 {
 974         struct enic *enic = netdev_priv(netdev);
 975
 976         if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic)) {
 977                 if (!is_valid_ether_addr(addr) && !is_zero_ether_addr(addr))
 978                         return -EADDRNOTAVAIL;
 979         } else {
 980                 if (!is_valid_ether_addr(addr))
 981                         return -EADDRNOTAVAIL;
 982         }
 983
 984         memcpy(netdev->dev_addr, addr, netdev->addr_len);
 985
 986         return 0;
 987 }
 988
 989 static int enic_set_mac_address_dynamic(struct net_device *netdev, void *p)
 990 {
 991         struct enic *enic = netdev_priv(netdev);
 992         struct sockaddr *saddr = p;
 993         char *addr = saddr->sa_data;
 994         int err;
 995
 996         if (netif_running(enic->netdev)) {
 997                 err = enic_dev_del_station_addr(enic);
 998                 if (err)
 999                         return err;
1000         }
1001
1002         err = enic_set_mac_addr(netdev, addr);
1003         if (err)
1004                 return err;
1005
1006         if (netif_running(enic->netdev)) {
1007                 err = enic_dev_add_station_addr(enic);
1008                 if (err)
1009                         return err;
1010         }
1011
1012         return err;
1013 }
1014
1015 static int enic_set_mac_address(struct net_device *netdev, void *p)
1016 {
1017         struct sockaddr *saddr = p;
1018         char *addr = saddr->sa_data;
1019         struct enic *enic = netdev_priv(netdev);
1020         int err;
1021
1022         err = enic_dev_del_station_addr(enic);
1023         if (err)
1024                 return err;
1025
1026         err = enic_set_mac_addr(netdev, addr);
1027         if (err)
1028                 return err;
1029
1030         return enic_dev_add_station_addr(enic);
1031 }
1032
1033 /* netif_tx_lock held, BHs disabled */
1034 static void enic_set_rx_mode(struct net_device *netdev)
1035 {
1036         struct enic *enic = netdev_priv(netdev);
1037         int directed = 1;
1038         int multicast = (netdev->flags & IFF_MULTICAST) ? 1 : 0;
1039         int broadcast = (netdev->flags & IFF_BROADCAST) ? 1 : 0;
1040         int promisc = (netdev->flags & IFF_PROMISC) ||
1041                 netdev_uc_count(netdev) > ENIC_UNICAST_PERFECT_FILTERS;
1042         int allmulti = (netdev->flags & IFF_ALLMULTI) ||
1043                 netdev_mc_count(netdev) > ENIC_MULTICAST_PERFECT_FILTERS;
1044         unsigned int flags = netdev->flags |
1045                 (allmulti ? IFF_ALLMULTI : 0) |
1046                 (promisc ? IFF_PROMISC : 0);
1047
1048         if (enic->flags != flags) {
1049                 enic->flags = flags;
1050                 enic_dev_packet_filter(enic, directed,
1051                         multicast, broadcast, promisc, allmulti);
1052         }
1053
1054         if (!promisc) {
1055                 __dev_uc_sync(netdev, enic_uc_sync, enic_uc_unsync);
1056                 if (!allmulti)
1057                         __dev_mc_sync(netdev, enic_mc_sync, enic_mc_unsync);
1058         }
1059 }
1060
1061 /* netif_tx_lock held, BHs disabled */
1062 static void enic_tx_timeout(struct net_device *netdev, unsigned int txqueue)
1063 {
1064         struct enic *enic = netdev_priv(netdev);
1065         schedule_work(&enic->tx_hang_reset);
1066 }
1067
1068 static int enic_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
1069 {
1070         struct enic *enic = netdev_priv(netdev);
1071         struct enic_port_profile *pp;
1072         int err;
1073
1074         ENIC_PP_BY_INDEX(enic, vf, pp, &err);
1075         if (err)
1076                 return err;
1077
1078         if (is_valid_ether_addr(mac) || is_zero_ether_addr(mac)) {
1079                 if (vf == PORT_SELF_VF) {
1080                         memcpy(pp->vf_mac, mac, ETH_ALEN);
1081                         return 0;
1082                 } else {
1083                         /*
1084                          * For sriov vf's set the mac in hw
1085                          */
1086                         ENIC_DEVCMD_PROXY_BY_INDEX(vf, err, enic,
1087                                 vnic_dev_set_mac_addr, mac);
1088                         return enic_dev_status_to_errno(err);
1089                 }
1090         } else
1091                 return -EINVAL;
1092 }
1093
1094 static int enic_set_vf_port(struct net_device *netdev, int vf,
1095         struct nlattr *port[])
1096 {
1097         struct enic *enic = netdev_priv(netdev);
1098         struct enic_port_profile prev_pp;
1099         struct enic_port_profile *pp;
1100         int err = 0, restore_pp = 1;
1101
1102         ENIC_PP_BY_INDEX(enic, vf, pp, &err);
1103         if (err)
1104                 return err;
1105
1106         if (!port[IFLA_PORT_REQUEST])
1107                 return -EOPNOTSUPP;
1108
1109         memcpy(&prev_pp, pp, sizeof(*enic->pp));
1110         memset(pp, 0, sizeof(*enic->pp));
1111
1112         pp->set |= ENIC_SET_REQUEST;
1113         pp->request = nla_get_u8(port[IFLA_PORT_REQUEST]);
1114
1115         if (port[IFLA_PORT_PROFILE]) {
1116                 pp->set |= ENIC_SET_NAME;
1117                 memcpy(pp->name, nla_data(port[IFLA_PORT_PROFILE]),
1118                         PORT_PROFILE_MAX);
1119         }
1120
1121         if (port[IFLA_PORT_INSTANCE_UUID]) {
1122                 pp->set |= ENIC_SET_INSTANCE;
1123                 memcpy(pp->instance_uuid,
1124                         nla_data(port[IFLA_PORT_INSTANCE_UUID]), PORT_UUID_MAX);
1125         }
1126
1127         if (port[IFLA_PORT_HOST_UUID]) {
1128                 pp->set |= ENIC_SET_HOST;
1129                 memcpy(pp->host_uuid,
1130                         nla_data(port[IFLA_PORT_HOST_UUID]), PORT_UUID_MAX);
1131         }
1132
1133         if (vf == PORT_SELF_VF) {
1134                 /* Special case handling: mac came from IFLA_VF_MAC */
1135                 if (!is_zero_ether_addr(prev_pp.vf_mac))
1136                         memcpy(pp->mac_addr, prev_pp.vf_mac, ETH_ALEN);
1137
1138                 if (is_zero_ether_addr(netdev->dev_addr))
1139                         eth_hw_addr_random(netdev);
1140         } else {
1141                 /* SR-IOV VF: get mac from adapter */
1142                 ENIC_DEVCMD_PROXY_BY_INDEX(vf, err, enic,
1143                         vnic_dev_get_mac_addr, pp->mac_addr);
1144                 if (err) {
1145                         netdev_err(netdev, "Error getting mac for vf %d\n", vf);
1146                         memcpy(pp, &prev_pp, sizeof(*pp));
1147                         return enic_dev_status_to_errno(err);
1148                 }
1149         }
1150
1151         err = enic_process_set_pp_request(enic, vf, &prev_pp, &restore_pp);
1152         if (err) {
1153                 if (restore_pp) {
1154                         /* Things are still the way they were: Implicit
1155                          * DISASSOCIATE failed
1156                          */
1157                         memcpy(pp, &prev_pp, sizeof(*pp));
1158                 } else {
1159                         memset(pp, 0, sizeof(*pp));
1160                         if (vf == PORT_SELF_VF)
1161                                 eth_zero_addr(netdev->dev_addr);
1162                 }
1163         } else {
1164                 /* Set flag to indicate that the port assoc/disassoc
1165                  * request has been sent out to fw
1166                  */
1167                 pp->set |= ENIC_PORT_REQUEST_APPLIED;
1168
1169                 /* If DISASSOCIATE, clean up all assigned/saved macaddresses */
1170                 if (pp->request == PORT_REQUEST_DISASSOCIATE) {
1171                         eth_zero_addr(pp->mac_addr);
1172                         if (vf == PORT_SELF_VF)
1173                                 eth_zero_addr(netdev->dev_addr);
1174                 }
1175         }
1176
1177         if (vf == PORT_SELF_VF)
1178                 eth_zero_addr(pp->vf_mac);
1179
1180         return err;
1181 }
1182
1183 static int enic_get_vf_port(struct net_device *netdev, int vf,
1184         struct sk_buff *skb)
1185 {
1186         struct enic *enic = netdev_priv(netdev);
1187         u16 response = PORT_PROFILE_RESPONSE_SUCCESS;
1188         struct enic_port_profile *pp;
1189         int err;
1190
1191         ENIC_PP_BY_INDEX(enic, vf, pp, &err);
1192         if (err)
1193                 return err;
1194
1195         if (!(pp->set & ENIC_PORT_REQUEST_APPLIED))
1196                 return -ENODATA;
1197
1198         err = enic_process_get_pp_request(enic, vf, pp->request, &response);
1199         if (err)
1200                 return err;
1201
1202         if (nla_put_u16(skb, IFLA_PORT_REQUEST, pp->request) ||
1203             nla_put_u16(skb, IFLA_PORT_RESPONSE, response) ||
1204             ((pp->set & ENIC_SET_NAME) &&
1205              nla_put(skb, IFLA_PORT_PROFILE, PORT_PROFILE_MAX, pp->name)) ||
1206             ((pp->set & ENIC_SET_INSTANCE) &&
1207              nla_put(skb, IFLA_PORT_INSTANCE_UUID, PORT_UUID_MAX,
1208                      pp->instance_uuid)) ||
1209             ((pp->set & ENIC_SET_HOST) &&
1210              nla_put(skb, IFLA_PORT_HOST_UUID, PORT_UUID_MAX, pp->host_uuid)))
1211                 goto nla_put_failure;
1212         return 0;
1213
1214 nla_put_failure:
1215         return -EMSGSIZE;
1216 }
1217
1218 static void enic_free_rq_buf(struct vnic_rq *rq, struct vnic_rq_buf *buf)
1219 {
1220         struct enic *enic = vnic_dev_priv(rq->vdev);
1221
1222         if (!buf->os_buf)
1223                 return;
1224
1225         pci_unmap_single(enic->pdev, buf->dma_addr,
1226                 buf->len, PCI_DMA_FROMDEVICE);
1227         dev_kfree_skb_any(buf->os_buf);
1228         buf->os_buf = NULL;
1229 }
1230
1231 static int enic_rq_alloc_buf(struct vnic_rq *rq)
1232 {
1233         struct enic *enic = vnic_dev_priv(rq->vdev);
1234         struct net_device *netdev = enic->netdev;
1235         struct sk_buff *skb;
1236         unsigned int len = netdev->mtu + VLAN_ETH_HLEN;
1237         unsigned int os_buf_index = 0;
1238         dma_addr_t dma_addr;
1239         struct vnic_rq_buf *buf = rq->to_use;
1240
1241         if (buf->os_buf) {
1242                 enic_queue_rq_desc(rq, buf->os_buf, os_buf_index, buf->dma_addr,
1243                                    buf->len);
1244
1245                 return 0;
1246         }
1247         skb = netdev_alloc_skb_ip_align(netdev, len);
1248         if (!skb)
1249                 return -ENOMEM;
1250
1251         dma_addr = pci_map_single(enic->pdev, skb->data, len,
1252                                   PCI_DMA_FROMDEVICE);
1253         if (unlikely(enic_dma_map_check(enic, dma_addr))) {
1254                 dev_kfree_skb(skb);
1255                 return -ENOMEM;
1256         }
1257
1258         enic_queue_rq_desc(rq, skb, os_buf_index,
1259                 dma_addr, len);
1260
1261         return 0;
1262 }
1263
1264 static void enic_intr_update_pkt_size(struct vnic_rx_bytes_counter *pkt_size,
1265                                       u32 pkt_len)
1266 {
1267         if (ENIC_LARGE_PKT_THRESHOLD <= pkt_len)
1268                 pkt_size->large_pkt_bytes_cnt += pkt_len;
1269         else
1270                 pkt_size->small_pkt_bytes_cnt += pkt_len;
1271 }
1272
1273 static bool enic_rxcopybreak(struct net_device *netdev, struct sk_buff **skb,
1274                              struct vnic_rq_buf *buf, u16 len)
1275 {
1276         struct enic *enic = netdev_priv(netdev);
1277         struct sk_buff *new_skb;
1278
1279         if (len > enic->rx_copybreak)
1280                 return false;
1281         new_skb = netdev_alloc_skb_ip_align(netdev, len);
1282         if (!new_skb)
1283                 return false;
1284         pci_dma_sync_single_for_cpu(enic->pdev, buf->dma_addr, len,
1285                                     DMA_FROM_DEVICE);
1286         memcpy(new_skb->data, (*skb)->data, len);
1287         *skb = new_skb;
1288
1289         return true;
1290 }
1291
1292 static void enic_rq_indicate_buf(struct vnic_rq *rq,
1293         struct cq_desc *cq_desc, struct vnic_rq_buf *buf,
1294         int skipped, void *opaque)
1295 {
1296         struct enic *enic = vnic_dev_priv(rq->vdev);
1297         struct net_device *netdev = enic->netdev;
1298         struct sk_buff *skb;
1299         struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
1300
1301         u8 type, color, eop, sop, ingress_port, vlan_stripped;
1302         u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof;
1303         u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok;
1304         u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc;
1305         u8 packet_error;
1306         u16 q_number, completed_index, bytes_written, vlan_tci, checksum;
1307         u32 rss_hash;
1308         bool outer_csum_ok = true, encap = false;
1309
1310         if (skipped)
1311                 return;
1312
1313         skb = buf->os_buf;
1314
1315         cq_enet_rq_desc_dec((struct cq_enet_rq_desc *)cq_desc,
1316                 &type, &color, &q_number, &completed_index,
1317                 &ingress_port, &fcoe, &eop, &sop, &rss_type,
1318                 &csum_not_calc, &rss_hash, &bytes_written,
1319                 &packet_error, &vlan_stripped, &vlan_tci, &checksum,
1320                 &fcoe_sof, &fcoe_fc_crc_ok, &fcoe_enc_error,
1321                 &fcoe_eof, &tcp_udp_csum_ok, &udp, &tcp,
1322                 &ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment,
1323                 &fcs_ok);
1324
1325         if (packet_error) {
1326
1327                 if (!fcs_ok) {
1328                         if (bytes_written > 0)
1329                                 enic->rq_bad_fcs++;
1330                         else if (bytes_written == 0)
1331                                 enic->rq_truncated_pkts++;
1332                 }
1333
1334                 pci_unmap_single(enic->pdev, buf->dma_addr, buf->len,
1335                                  PCI_DMA_FROMDEVICE);
1336                 dev_kfree_skb_any(skb);
1337                 buf->os_buf = NULL;
1338
1339                 return;
1340         }
1341
1342         if (eop && bytes_written > 0) {
1343
1344                 /* Good receive
1345                  */
1346
1347                 if (!enic_rxcopybreak(netdev, &skb, buf, bytes_written)) {
1348                         buf->os_buf = NULL;
1349                         pci_unmap_single(enic->pdev, buf->dma_addr, buf->len,
1350                                          PCI_DMA_FROMDEVICE);
1351                 }
1352                 prefetch(skb->data - NET_IP_ALIGN);
1353
1354                 skb_put(skb, bytes_written);
1355                 skb->protocol = eth_type_trans(skb, netdev);
1356                 skb_record_rx_queue(skb, q_number);
1357                 if ((netdev->features & NETIF_F_RXHASH) && rss_hash &&
1358                     (type == 3)) {
1359                         switch (rss_type) {
1360                         case CQ_ENET_RQ_DESC_RSS_TYPE_TCP_IPv4:
1361                         case CQ_ENET_RQ_DESC_RSS_TYPE_TCP_IPv6:
1362                         case CQ_ENET_RQ_DESC_RSS_TYPE_TCP_IPv6_EX:
1363                                 skb_set_hash(skb, rss_hash, PKT_HASH_TYPE_L4);
1364                                 break;
1365                         case CQ_ENET_RQ_DESC_RSS_TYPE_IPv4:
1366                         case CQ_ENET_RQ_DESC_RSS_TYPE_IPv6:
1367                         case CQ_ENET_RQ_DESC_RSS_TYPE_IPv6_EX:
1368                                 skb_set_hash(skb, rss_hash, PKT_HASH_TYPE_L3);
1369                                 break;
1370                         }
1371                 }
1372                 if (enic->vxlan.vxlan_udp_port_number) {
1373                         switch (enic->vxlan.patch_level) {
1374                         case 0:
1375                                 if (fcoe) {
1376                                         encap = true;
1377                                         outer_csum_ok = fcoe_fc_crc_ok;
1378                                 }
1379                                 break;
1380                         case 2:
1381                                 if ((type == 7) &&
1382                                     (rss_hash & BIT(0))) {
1383                                         encap = true;
1384                                         outer_csum_ok = (rss_hash & BIT(1)) &&
1385                                                         (rss_hash & BIT(2));
1386                                 }
1387                                 break;
1388                         }
1389                 }
1390
1391                 /* Hardware does not provide whole packet checksum. It only
1392                  * provides pseudo checksum. Since hw validates the packet
1393                  * checksum but not provide us the checksum value. use
1394                  * CHECSUM_UNNECESSARY.
1395                  *
1396                  * In case of encap pkt tcp_udp_csum_ok/tcp_udp_csum_ok is
1397                  * inner csum_ok. outer_csum_ok is set by hw when outer udp
1398                  * csum is correct or is zero.
1399                  */
1400                 if ((netdev->features & NETIF_F_RXCSUM) && !csum_not_calc &&
1401                     tcp_udp_csum_ok && outer_csum_ok &&
1402                     (ipv4_csum_ok || ipv6)) {
1403                         skb->ip_summed = CHECKSUM_UNNECESSARY;
1404                         skb->csum_level = encap;
1405                 }
1406
1407                 if (vlan_stripped)
1408                         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tci);
1409
1410                 skb_mark_napi_id(skb, &enic->napi[rq->index]);
1411                 if (!(netdev->features & NETIF_F_GRO))
1412                         netif_receive_skb(skb);
1413                 else
1414                         napi_gro_receive(&enic->napi[q_number], skb);
1415                 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1416                         enic_intr_update_pkt_size(&cq->pkt_size_counter,
1417                                                   bytes_written);
1418         } else {
1419
1420                 /* Buffer overflow
1421                  */
1422
1423                 pci_unmap_single(enic->pdev, buf->dma_addr, buf->len,
1424                                  PCI_DMA_FROMDEVICE);
1425                 dev_kfree_skb_any(skb);
1426                 buf->os_buf = NULL;
1427         }
1428 }
1429
1430 static int enic_rq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
1431         u8 type, u16 q_number, u16 completed_index, void *opaque)
1432 {
1433         struct enic *enic = vnic_dev_priv(vdev);
1434
1435         vnic_rq_service(&enic->rq[q_number], cq_desc,
1436                 completed_index, VNIC_RQ_RETURN_DESC,
1437                 enic_rq_indicate_buf, opaque);
1438
1439         return 0;
1440 }
1441
1442 static void enic_set_int_moderation(struct enic *enic, struct vnic_rq *rq)
1443 {
1444         unsigned int intr = enic_msix_rq_intr(enic, rq->index);
1445         struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
1446         u32 timer = cq->tobe_rx_coal_timeval;
1447
1448         if (cq->tobe_rx_coal_timeval != cq->cur_rx_coal_timeval) {
1449                 vnic_intr_coalescing_timer_set(&enic->intr[intr], timer);
1450                 cq->cur_rx_coal_timeval = cq->tobe_rx_coal_timeval;
1451         }
1452 }
1453
1454 static void enic_calc_int_moderation(struct enic *enic, struct vnic_rq *rq)
1455 {
1456         struct enic_rx_coal *rx_coal = &enic->rx_coalesce_setting;
1457         struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
1458         struct vnic_rx_bytes_counter *pkt_size_counter = &cq->pkt_size_counter;
1459         int index;
1460         u32 timer;
1461         u32 range_start;
1462         u32 traffic;
1463         u64 delta;
1464         ktime_t now = ktime_get();
1465
1466         delta = ktime_us_delta(now, cq->prev_ts);
1467         if (delta < ENIC_AIC_TS_BREAK)
1468                 return;
1469         cq->prev_ts = now;
1470
1471         traffic = pkt_size_counter->large_pkt_bytes_cnt +
1472                   pkt_size_counter->small_pkt_bytes_cnt;
1473         /* The table takes Mbps
1474          * traffic *= 8    => bits
1475          * traffic *= (10^6 / delta)    => bps
1476          * traffic /= 10^6     => Mbps
1477          *
1478          * Combining, traffic *= (8 / delta)
1479          */
1480
1481         traffic <<= 3;
1482         traffic = delta > UINT_MAX ? 0 : traffic / (u32)delta;
1483
1484         for (index = 0; index < ENIC_MAX_COALESCE_TIMERS; index++)
1485                 if (traffic < mod_table[index].rx_rate)
1486                         break;
1487         range_start = (pkt_size_counter->small_pkt_bytes_cnt >
1488                        pkt_size_counter->large_pkt_bytes_cnt << 1) ?
1489                       rx_coal->small_pkt_range_start :
1490                       rx_coal->large_pkt_range_start;
1491         timer = range_start + ((rx_coal->range_end - range_start) *
1492                                mod_table[index].range_percent / 100);
1493         /* Damping */
1494         cq->tobe_rx_coal_timeval = (timer + cq->tobe_rx_coal_timeval) >> 1;
1495
1496         pkt_size_counter->large_pkt_bytes_cnt = 0;
1497         pkt_size_counter->small_pkt_bytes_cnt = 0;
1498 }
1499
1500 static int enic_poll(struct napi_struct *napi, int budget)
1501 {
1502         struct net_device *netdev = napi->dev;
1503         struct enic *enic = netdev_priv(netdev);
1504         unsigned int cq_rq = enic_cq_rq(enic, 0);
1505         unsigned int cq_wq = enic_cq_wq(enic, 0);
1506         unsigned int intr = enic_legacy_io_intr();
1507         unsigned int rq_work_to_do = budget;
1508         unsigned int wq_work_to_do = ENIC_WQ_NAPI_BUDGET;
1509         unsigned int  work_done, rq_work_done = 0, wq_work_done;
1510         int err;
1511
1512         wq_work_done = vnic_cq_service(&enic->cq[cq_wq], wq_work_to_do,
1513                                        enic_wq_service, NULL);
1514
1515         if (budget > 0)
1516                 rq_work_done = vnic_cq_service(&enic->cq[cq_rq],
1517                         rq_work_to_do, enic_rq_service, NULL);
1518
1519         /* Accumulate intr event credits for this polling
1520          * cycle.  An intr event is the completion of a
1521          * a WQ or RQ packet.
1522          */
1523
1524         work_done = rq_work_done + wq_work_done;
1525
1526         if (work_done > 0)
1527                 vnic_intr_return_credits(&enic->intr[intr],
1528                         work_done,
1529                         0 /* don't unmask intr */,
1530                         0 /* don't reset intr timer */);
1531
1532         err = vnic_rq_fill(&enic->rq[0], enic_rq_alloc_buf);
1533
1534         /* Buffer allocation failed. Stay in polling
1535          * mode so we can try to fill the ring again.
1536          */
1537
1538         if (err)
1539                 rq_work_done = rq_work_to_do;
1540         if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1541                 /* Call the function which refreshes the intr coalescing timer
1542                  * value based on the traffic.
1543                  */
1544                 enic_calc_int_moderation(enic, &enic->rq[0]);
1545
1546         if ((rq_work_done < budget) && napi_complete_done(napi, rq_work_done)) {
1547
1548                 /* Some work done, but not enough to stay in polling,
1549                  * exit polling
1550                  */
1551
1552                 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1553                         enic_set_int_moderation(enic, &enic->rq[0]);
1554                 vnic_intr_unmask(&enic->intr[intr]);
1555         }
1556
1557         return rq_work_done;
1558 }
1559
1560 #ifdef CONFIG_RFS_ACCEL
1561 static void enic_free_rx_cpu_rmap(struct enic *enic)
1562 {
1563         free_irq_cpu_rmap(enic->netdev->rx_cpu_rmap);
1564         enic->netdev->rx_cpu_rmap = NULL;
1565 }
1566
1567 static void enic_set_rx_cpu_rmap(struct enic *enic)
1568 {
1569         int i, res;
1570
1571         if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX) {
1572                 enic->netdev->rx_cpu_rmap = alloc_irq_cpu_rmap(enic->rq_count);
1573                 if (unlikely(!enic->netdev->rx_cpu_rmap))
1574                         return;
1575                 for (i = 0; i < enic->rq_count; i++) {
1576                         res = irq_cpu_rmap_add(enic->netdev->rx_cpu_rmap,
1577                                                enic->msix_entry[i].vector);
1578                         if (unlikely(res)) {
1579                                 enic_free_rx_cpu_rmap(enic);
1580                                 return;
1581                         }
1582                 }
1583         }
1584 }
1585
1586 #else
1587
1588 static void enic_free_rx_cpu_rmap(struct enic *enic)
1589 {
1590 }
1591
1592 static void enic_set_rx_cpu_rmap(struct enic *enic)
1593 {
1594 }
1595
1596 #endif /* CONFIG_RFS_ACCEL */
1597
1598 static int enic_poll_msix_wq(struct napi_struct *napi, int budget)
1599 {
1600         struct net_device *netdev = napi->dev;
1601         struct enic *enic = netdev_priv(netdev);
1602         unsigned int wq_index = (napi - &enic->napi[0]) - enic->rq_count;
1603         struct vnic_wq *wq = &enic->wq[wq_index];
1604         unsigned int cq;
1605         unsigned int intr;
1606         unsigned int wq_work_to_do = ENIC_WQ_NAPI_BUDGET;
1607         unsigned int wq_work_done;
1608         unsigned int wq_irq;
1609
1610         wq_irq = wq->index;
1611         cq = enic_cq_wq(enic, wq_irq);
1612         intr = enic_msix_wq_intr(enic, wq_irq);
1613         wq_work_done = vnic_cq_service(&enic->cq[cq], wq_work_to_do,
1614                                        enic_wq_service, NULL);
1615
1616         vnic_intr_return_credits(&enic->intr[intr], wq_work_done,
1617                                  0 /* don't unmask intr */,
1618                                  1 /* reset intr timer */);
1619         if (!wq_work_done) {
1620                 napi_complete(napi);
1621                 vnic_intr_unmask(&enic->intr[intr]);
1622                 return 0;
1623         }
1624
1625         return budget;
1626 }
1627
1628 static int enic_poll_msix_rq(struct napi_struct *napi, int budget)
1629 {
1630         struct net_device *netdev = napi->dev;
1631         struct enic *enic = netdev_priv(netdev);
1632         unsigned int rq = (napi - &enic->napi[0]);
1633         unsigned int cq = enic_cq_rq(enic, rq);
1634         unsigned int intr = enic_msix_rq_intr(enic, rq);
1635         unsigned int work_to_do = budget;
1636         unsigned int work_done = 0;
1637         int err;
1638
1639         /* Service RQ
1640          */
1641
1642         if (budget > 0)
1643                 work_done = vnic_cq_service(&enic->cq[cq],
1644                         work_to_do, enic_rq_service, NULL);
1645
1646         /* Return intr event credits for this polling
1647          * cycle.  An intr event is the completion of a
1648          * RQ packet.
1649          */
1650
1651         if (work_done > 0)
1652                 vnic_intr_return_credits(&enic->intr[intr],
1653                         work_done,
1654                         0 /* don't unmask intr */,
1655                         0 /* don't reset intr timer */);
1656
1657         err = vnic_rq_fill(&enic->rq[rq], enic_rq_alloc_buf);
1658
1659         /* Buffer allocation failed. Stay in polling mode
1660          * so we can try to fill the ring again.
1661          */
1662
1663         if (err)
1664                 work_done = work_to_do;
1665         if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1666                 /* Call the function which refreshes the intr coalescing timer
1667                  * value based on the traffic.
1668                  */
1669                 enic_calc_int_moderation(enic, &enic->rq[rq]);
1670
1671         if ((work_done < budget) && napi_complete_done(napi, work_done)) {
1672
1673                 /* Some work done, but not enough to stay in polling,
1674                  * exit polling
1675                  */
1676
1677                 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1678                         enic_set_int_moderation(enic, &enic->rq[rq]);
1679                 vnic_intr_unmask(&enic->intr[intr]);
1680         }
1681
1682         return work_done;
1683 }
1684
1685 static void enic_notify_timer(struct timer_list *t)
1686 {
1687         struct enic *enic = from_timer(enic, t, notify_timer);
1688
1689         enic_notify_check(enic);
1690
1691         mod_timer(&enic->notify_timer,
1692                 round_jiffies(jiffies + ENIC_NOTIFY_TIMER_PERIOD));
1693 }
1694
1695 static void enic_free_intr(struct enic *enic)
1696 {
1697         struct net_device *netdev = enic->netdev;
1698         unsigned int i;
1699
1700         enic_free_rx_cpu_rmap(enic);
1701         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1702         case VNIC_DEV_INTR_MODE_INTX:
1703                 free_irq(enic->pdev->irq, netdev);
1704                 break;
1705         case VNIC_DEV_INTR_MODE_MSI:
1706                 free_irq(enic->pdev->irq, enic);
1707                 break;
1708         case VNIC_DEV_INTR_MODE_MSIX:
1709                 for (i = 0; i < ARRAY_SIZE(enic->msix); i++)
1710                         if (enic->msix[i].requested)
1711                                 free_irq(enic->msix_entry[i].vector,
1712                                         enic->msix[i].devid);
1713                 break;
1714         default:
1715                 break;
1716         }
1717 }
1718
1719 static int enic_request_intr(struct enic *enic)
1720 {
1721         struct net_device *netdev = enic->netdev;
1722         unsigned int i, intr;
1723         int err = 0;
1724
1725         enic_set_rx_cpu_rmap(enic);
1726         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1727
1728         case VNIC_DEV_INTR_MODE_INTX:
1729
1730                 err = request_irq(enic->pdev->irq, enic_isr_legacy,
1731                         IRQF_SHARED, netdev->name, netdev);
1732                 break;
1733
1734         case VNIC_DEV_INTR_MODE_MSI:
1735
1736                 err = request_irq(enic->pdev->irq, enic_isr_msi,
1737                         0, netdev->name, enic);
1738                 break;
1739
1740         case VNIC_DEV_INTR_MODE_MSIX:
1741
1742                 for (i = 0; i < enic->rq_count; i++) {
1743                         intr = enic_msix_rq_intr(enic, i);
1744                         snprintf(enic->msix[intr].devname,
1745                                 sizeof(enic->msix[intr].devname),
1746                                 "%s-rx-%u", netdev->name, i);
1747                         enic->msix[intr].isr = enic_isr_msix;
1748                         enic->msix[intr].devid = &enic->napi[i];
1749                 }
1750
1751                 for (i = 0; i < enic->wq_count; i++) {
1752                         int wq = enic_cq_wq(enic, i);
1753
1754                         intr = enic_msix_wq_intr(enic, i);
1755                         snprintf(enic->msix[intr].devname,
1756                                 sizeof(enic->msix[intr].devname),
1757                                 "%s-tx-%u", netdev->name, i);
1758                         enic->msix[intr].isr = enic_isr_msix;
1759                         enic->msix[intr].devid = &enic->napi[wq];
1760                 }
1761
1762                 intr = enic_msix_err_intr(enic);
1763                 snprintf(enic->msix[intr].devname,
1764                         sizeof(enic->msix[intr].devname),
1765                         "%s-err", netdev->name);
1766                 enic->msix[intr].isr = enic_isr_msix_err;
1767                 enic->msix[intr].devid = enic;
1768
1769                 intr = enic_msix_notify_intr(enic);
1770                 snprintf(enic->msix[intr].devname,
1771                         sizeof(enic->msix[intr].devname),
1772                         "%s-notify", netdev->name);
1773                 enic->msix[intr].isr = enic_isr_msix_notify;
1774                 enic->msix[intr].devid = enic;
1775
1776                 for (i = 0; i < ARRAY_SIZE(enic->msix); i++)
1777                         enic->msix[i].requested = 0;
1778
1779                 for (i = 0; i < enic->intr_count; i++) {
1780                         err = request_irq(enic->msix_entry[i].vector,
1781                                 enic->msix[i].isr, 0,
1782                                 enic->msix[i].devname,
1783                                 enic->msix[i].devid);
1784                         if (err) {
1785                                 enic_free_intr(enic);
1786                                 break;
1787                         }
1788                         enic->msix[i].requested = 1;
1789                 }
1790
1791                 break;
1792
1793         default:
1794                 break;
1795         }
1796
1797         return err;
1798 }
1799
1800 static void enic_synchronize_irqs(struct enic *enic)
1801 {
1802         unsigned int i;
1803
1804         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1805         case VNIC_DEV_INTR_MODE_INTX:
1806         case VNIC_DEV_INTR_MODE_MSI:
1807                 synchronize_irq(enic->pdev->irq);
1808                 break;
1809         case VNIC_DEV_INTR_MODE_MSIX:
1810                 for (i = 0; i < enic->intr_count; i++)
1811                         synchronize_irq(enic->msix_entry[i].vector);
1812                 break;
1813         default:
1814                 break;
1815         }
1816 }
1817
1818 static void enic_set_rx_coal_setting(struct enic *enic)
1819 {
1820         unsigned int speed;
1821         int index = -1;
1822         struct enic_rx_coal *rx_coal = &enic->rx_coalesce_setting;
1823
1824         /* 1. Read the link speed from fw
1825          * 2. Pick the default range for the speed
1826          * 3. Update it in enic->rx_coalesce_setting
1827          */
1828         speed = vnic_dev_port_speed(enic->vdev);
1829         if (ENIC_LINK_SPEED_10G < speed)
1830                 index = ENIC_LINK_40G_INDEX;
1831         else if (ENIC_LINK_SPEED_4G < speed)
1832                 index = ENIC_LINK_10G_INDEX;
1833         else
1834                 index = ENIC_LINK_4G_INDEX;
1835
1836         rx_coal->small_pkt_range_start = mod_range[index].small_pkt_range_start;
1837         rx_coal->large_pkt_range_start = mod_range[index].large_pkt_range_start;
1838         rx_coal->range_end = ENIC_RX_COALESCE_RANGE_END;
1839
1840         /* Start with the value provided by UCSM */
1841         for (index = 0; index < enic->rq_count; index++)
1842                 enic->cq[index].cur_rx_coal_timeval =
1843                                 enic->config.intr_timer_usec;
1844
1845         rx_coal->use_adaptive_rx_coalesce = 1;
1846 }
1847
1848 static int enic_dev_notify_set(struct enic *enic)
1849 {
1850         int err;
1851
1852         spin_lock_bh(&enic->devcmd_lock);
1853         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1854         case VNIC_DEV_INTR_MODE_INTX:
1855                 err = vnic_dev_notify_set(enic->vdev,
1856                         enic_legacy_notify_intr());
1857                 break;
1858         case VNIC_DEV_INTR_MODE_MSIX:
1859                 err = vnic_dev_notify_set(enic->vdev,
1860                         enic_msix_notify_intr(enic));
1861                 break;
1862         default:
1863                 err = vnic_dev_notify_set(enic->vdev, -1 /* no intr */);
1864                 break;
1865         }
1866         spin_unlock_bh(&enic->devcmd_lock);
1867
1868         return err;
1869 }
1870
1871 static void enic_notify_timer_start(struct enic *enic)
1872 {
1873         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1874         case VNIC_DEV_INTR_MODE_MSI:
1875                 mod_timer(&enic->notify_timer, jiffies);
1876                 break;
1877         default:
1878                 /* Using intr for notification for INTx/MSI-X */
1879                 break;
1880         }
1881 }
1882
1883 /* rtnl lock is held, process context */
1884 static int enic_open(struct net_device *netdev)
1885 {
1886         struct enic *enic = netdev_priv(netdev);
1887         unsigned int i;
1888         int err, ret;
1889
1890         err = enic_request_intr(enic);
1891         if (err) {
1892                 netdev_err(netdev, "Unable to request irq.\n");
1893                 return err;
1894         }
1895         enic_init_affinity_hint(enic);
1896         enic_set_affinity_hint(enic);
1897
1898         err = enic_dev_notify_set(enic);
1899         if (err) {
1900                 netdev_err(netdev,
1901                         "Failed to alloc notify buffer, aborting.\n");
1902                 goto err_out_free_intr;
1903         }
1904
1905         for (i = 0; i < enic->rq_count; i++) {
1906                 /* enable rq before updating rq desc */
1907                 vnic_rq_enable(&enic->rq[i]);
1908                 vnic_rq_fill(&enic->rq[i], enic_rq_alloc_buf);
1909                 /* Need at least one buffer on ring to get going */
1910                 if (vnic_rq_desc_used(&enic->rq[i]) == 0) {
1911                         netdev_err(netdev, "Unable to alloc receive buffers\n");
1912                         err = -ENOMEM;
1913                         goto err_out_free_rq;
1914                 }
1915         }
1916
1917         for (i = 0; i < enic->wq_count; i++)
1918                 vnic_wq_enable(&enic->wq[i]);
1919
1920         if (!enic_is_dynamic(enic) && !enic_is_sriov_vf(enic))
1921                 enic_dev_add_station_addr(enic);
1922
1923         enic_set_rx_mode(netdev);
1924
1925         netif_tx_wake_all_queues(netdev);
1926
1927         for (i = 0; i < enic->rq_count; i++)
1928                 napi_enable(&enic->napi[i]);
1929
1930         if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX)
1931                 for (i = 0; i < enic->wq_count; i++)
1932                         napi_enable(&enic->napi[enic_cq_wq(enic, i)]);
1933         enic_dev_enable(enic);
1934
1935         for (i = 0; i < enic->intr_count; i++)
1936                 vnic_intr_unmask(&enic->intr[i]);
1937
1938         enic_notify_timer_start(enic);
1939         enic_rfs_timer_start(enic);
1940
1941         return 0;
1942
1943 err_out_free_rq:
1944         for (i = 0; i < enic->rq_count; i++) {
1945                 ret = vnic_rq_disable(&enic->rq[i]);
1946                 if (!ret)
1947                         vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
1948         }
1949         enic_dev_notify_unset(enic);
1950 err_out_free_intr:
1951         enic_unset_affinity_hint(enic);
1952         enic_free_intr(enic);
1953
1954         return err;
1955 }
1956
1957 /* rtnl lock is held, process context */
1958 static int enic_stop(struct net_device *netdev)
1959 {
1960         struct enic *enic = netdev_priv(netdev);
1961         unsigned int i;
1962         int err;
1963
1964         for (i = 0; i < enic->intr_count; i++) {
1965                 vnic_intr_mask(&enic->intr[i]);
1966                 (void)vnic_intr_masked(&enic->intr[i]); /* flush write */
1967         }
1968
1969         enic_synchronize_irqs(enic);
1970
1971         del_timer_sync(&enic->notify_timer);
1972         enic_rfs_flw_tbl_free(enic);
1973
1974         enic_dev_disable(enic);
1975
1976         for (i = 0; i < enic->rq_count; i++)
1977                 napi_disable(&enic->napi[i]);
1978
1979         netif_carrier_off(netdev);
1980         if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX)
1981                 for (i = 0; i < enic->wq_count; i++)
1982                         napi_disable(&enic->napi[enic_cq_wq(enic, i)]);
1983         netif_tx_disable(netdev);
1984
1985         if (!enic_is_dynamic(enic) && !enic_is_sriov_vf(enic))
1986                 enic_dev_del_station_addr(enic);
1987
1988         for (i = 0; i < enic->wq_count; i++) {
1989                 err = vnic_wq_disable(&enic->wq[i]);
1990                 if (err)
1991                         return err;
1992         }
1993         for (i = 0; i < enic->rq_count; i++) {
1994                 err = vnic_rq_disable(&enic->rq[i]);
1995                 if (err)
1996                         return err;
1997         }
1998
1999         enic_dev_notify_unset(enic);
2000         enic_unset_affinity_hint(enic);
2001         enic_free_intr(enic);
2002
2003         for (i = 0; i < enic->wq_count; i++)
2004                 vnic_wq_clean(&enic->wq[i], enic_free_wq_buf);
2005         for (i = 0; i < enic->rq_count; i++)
2006                 vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
2007         for (i = 0; i < enic->cq_count; i++)
2008                 vnic_cq_clean(&enic->cq[i]);
2009         for (i = 0; i < enic->intr_count; i++)
2010                 vnic_intr_clean(&enic->intr[i]);
2011
2012         return 0;
2013 }
2014
2015 static int _enic_change_mtu(struct net_device *netdev, int new_mtu)
2016 {
2017         bool running = netif_running(netdev);
2018         int err = 0;
2019
2020         ASSERT_RTNL();
2021         if (running) {
2022                 err = enic_stop(netdev);
2023                 if (err)
2024                         return err;
2025         }
2026
2027         netdev->mtu = new_mtu;
2028
2029         if (running) {
2030                 err = enic_open(netdev);
2031                 if (err)
2032                         return err;
2033         }
2034
2035         return 0;
2036 }
2037
2038 static int enic_change_mtu(struct net_device *netdev, int new_mtu)
2039 {
2040         struct enic *enic = netdev_priv(netdev);
2041
2042         if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic))
2043                 return -EOPNOTSUPP;
2044
2045         if (netdev->mtu > enic->port_mtu)
2046                 netdev_warn(netdev,
2047                             "interface MTU (%d) set higher than port MTU (%d)\n",
2048                             netdev->mtu, enic->port_mtu);
2049
2050         return _enic_change_mtu(netdev, new_mtu);
2051 }
2052
2053 static void enic_change_mtu_work(struct work_struct *work)
2054 {
2055         struct enic *enic = container_of(work, struct enic, change_mtu_work);
2056         struct net_device *netdev = enic->netdev;
2057         int new_mtu = vnic_dev_mtu(enic->vdev);
2058
2059         rtnl_lock();
2060         (void)_enic_change_mtu(netdev, new_mtu);
2061         rtnl_unlock();
2062
2063         netdev_info(netdev, "interface MTU set as %d\n", netdev->mtu);
2064 }
2065
2066 #ifdef CONFIG_NET_POLL_CONTROLLER
2067 static void enic_poll_controller(struct net_device *netdev)
2068 {
2069         struct enic *enic = netdev_priv(netdev);
2070         struct vnic_dev *vdev = enic->vdev;
2071         unsigned int i, intr;
2072
2073         switch (vnic_dev_get_intr_mode(vdev)) {
2074         case VNIC_DEV_INTR_MODE_MSIX:
2075                 for (i = 0; i < enic->rq_count; i++) {
2076                         intr = enic_msix_rq_intr(enic, i);
2077                         enic_isr_msix(enic->msix_entry[intr].vector,
2078                                       &enic->napi[i]);
2079                 }
2080
2081                 for (i = 0; i < enic->wq_count; i++) {
2082                         intr = enic_msix_wq_intr(enic, i);
2083                         enic_isr_msix(enic->msix_entry[intr].vector,
2084                                       &enic->napi[enic_cq_wq(enic, i)]);
2085                 }
2086
2087                 break;
2088         case VNIC_DEV_INTR_MODE_MSI:
2089                 enic_isr_msi(enic->pdev->irq, enic);
2090                 break;
2091         case VNIC_DEV_INTR_MODE_INTX:
2092                 enic_isr_legacy(enic->pdev->irq, netdev);
2093                 break;
2094         default:
2095                 break;
2096         }
2097 }
2098 #endif
2099
2100 static int enic_dev_wait(struct vnic_dev *vdev,
2101         int (*start)(struct vnic_dev *, int),
2102         int (*finished)(struct vnic_dev *, int *),
2103         int arg)
2104 {
2105         unsigned long time;
2106         int done;
2107         int err;
2108
2109         BUG_ON(in_interrupt());
2110
2111         err = start(vdev, arg);
2112         if (err)
2113                 return err;
2114
2115         /* Wait for func to complete...2 seconds max
2116          */
2117
2118         time = jiffies + (HZ * 2);
2119         do {
2120
2121                 err = finished(vdev, &done);
2122                 if (err)
2123                         return err;
2124
2125                 if (done)
2126                         return 0;
2127
2128                 schedule_timeout_uninterruptible(HZ / 10);
2129
2130         } while (time_after(time, jiffies));
2131
2132         return -ETIMEDOUT;
2133 }
2134
2135 static int enic_dev_open(struct enic *enic)
2136 {
2137         int err;
2138         u32 flags = CMD_OPENF_IG_DESCCACHE;
2139
2140         err = enic_dev_wait(enic->vdev, vnic_dev_open,
2141                 vnic_dev_open_done, flags);
2142         if (err)
2143                 dev_err(enic_get_dev(enic), "vNIC device open failed, err %d\n",
2144                         err);
2145
2146         return err;
2147 }
2148
2149 static int enic_dev_soft_reset(struct enic *enic)
2150 {
2151         int err;
2152
2153         err = enic_dev_wait(enic->vdev, vnic_dev_soft_reset,
2154                             vnic_dev_soft_reset_done, 0);
2155         if (err)
2156                 netdev_err(enic->netdev, "vNIC soft reset failed, err %d\n",
2157                            err);
2158
2159         return err;
2160 }
2161
2162 static int enic_dev_hang_reset(struct enic *enic)
2163 {
2164         int err;
2165
2166         err = enic_dev_wait(enic->vdev, vnic_dev_hang_reset,
2167                 vnic_dev_hang_reset_done, 0);
2168         if (err)
2169                 netdev_err(enic->netdev, "vNIC hang reset failed, err %d\n",
2170                         err);
2171
2172         return err;
2173 }
2174
2175 int __enic_set_rsskey(struct enic *enic)
2176 {
2177         union vnic_rss_key *rss_key_buf_va;
2178         dma_addr_t rss_key_buf_pa;
2179         int i, kidx, bidx, err;
2180
2181         rss_key_buf_va = pci_zalloc_consistent(enic->pdev,
2182                                                sizeof(union vnic_rss_key),
2183                                                &rss_key_buf_pa);
2184         if (!rss_key_buf_va)
2185                 return -ENOMEM;
2186
2187         for (i = 0; i < ENIC_RSS_LEN; i++) {
2188                 kidx = i / ENIC_RSS_BYTES_PER_KEY;
2189                 bidx = i % ENIC_RSS_BYTES_PER_KEY;
2190                 rss_key_buf_va->key[kidx].b[bidx] = enic->rss_key[i];
2191         }
2192         spin_lock_bh(&enic->devcmd_lock);
2193         err = enic_set_rss_key(enic,
2194                 rss_key_buf_pa,
2195                 sizeof(union vnic_rss_key));
2196         spin_unlock_bh(&enic->devcmd_lock);
2197
2198         pci_free_consistent(enic->pdev, sizeof(union vnic_rss_key),
2199                 rss_key_buf_va, rss_key_buf_pa);
2200
2201         return err;
2202 }
2203
2204 static int enic_set_rsskey(struct enic *enic)
2205 {
2206         netdev_rss_key_fill(enic->rss_key, ENIC_RSS_LEN);
2207
2208         return __enic_set_rsskey(enic);
2209 }
2210
2211 static int enic_set_rsscpu(struct enic *enic, u8 rss_hash_bits)
2212 {
2213         dma_addr_t rss_cpu_buf_pa;
2214         union vnic_rss_cpu *rss_cpu_buf_va = NULL;
2215         unsigned int i;
2216         int err;
2217
2218         rss_cpu_buf_va = pci_alloc_consistent(enic->pdev,
2219                 sizeof(union vnic_rss_cpu), &rss_cpu_buf_pa);
2220         if (!rss_cpu_buf_va)
2221                 return -ENOMEM;
2222
2223         for (i = 0; i < (1 << rss_hash_bits); i++)
2224                 (*rss_cpu_buf_va).cpu[i/4].b[i%4] = i % enic->rq_count;
2225
2226         spin_lock_bh(&enic->devcmd_lock);
2227         err = enic_set_rss_cpu(enic,
2228                 rss_cpu_buf_pa,
2229                 sizeof(union vnic_rss_cpu));
2230         spin_unlock_bh(&enic->devcmd_lock);
2231
2232         pci_free_consistent(enic->pdev, sizeof(union vnic_rss_cpu),
2233                 rss_cpu_buf_va, rss_cpu_buf_pa);
2234
2235         return err;
2236 }
2237
2238 static int enic_set_niccfg(struct enic *enic, u8 rss_default_cpu,
2239         u8 rss_hash_type, u8 rss_hash_bits, u8 rss_base_cpu, u8 rss_enable)
2240 {
2241         const u8 tso_ipid_split_en = 0;
2242         const u8 ig_vlan_strip_en = 1;
2243         int err;
2244
2245         /* Enable VLAN tag stripping.
2246         */
2247
2248         spin_lock_bh(&enic->devcmd_lock);
2249         err = enic_set_nic_cfg(enic,
2250                 rss_default_cpu, rss_hash_type,
2251                 rss_hash_bits, rss_base_cpu,
2252                 rss_enable, tso_ipid_split_en,
2253                 ig_vlan_strip_en);
2254         spin_unlock_bh(&enic->devcmd_lock);
2255
2256         return err;
2257 }
2258
2259 static int enic_set_rss_nic_cfg(struct enic *enic)
2260 {
2261         struct device *dev = enic_get_dev(enic);
2262         const u8 rss_default_cpu = 0;
2263         const u8 rss_hash_bits = 7;
2264         const u8 rss_base_cpu = 0;
2265         u8 rss_hash_type;
2266         int res;
2267         u8 rss_enable = ENIC_SETTING(enic, RSS) && (enic->rq_count > 1);
2268
2269         spin_lock_bh(&enic->devcmd_lock);
2270         res = vnic_dev_capable_rss_hash_type(enic->vdev, &rss_hash_type);
2271         spin_unlock_bh(&enic->devcmd_lock);
2272         if (res) {
2273                 /* defaults for old adapters
2274                  */
2275                 rss_hash_type = NIC_CFG_RSS_HASH_TYPE_IPV4      |
2276                                 NIC_CFG_RSS_HASH_TYPE_TCP_IPV4  |
2277                                 NIC_CFG_RSS_HASH_TYPE_IPV6      |
2278                                 NIC_CFG_RSS_HASH_TYPE_TCP_IPV6;
2279         }
2280
2281         if (rss_enable) {
2282                 if (!enic_set_rsskey(enic)) {
2283                         if (enic_set_rsscpu(enic, rss_hash_bits)) {
2284                                 rss_enable = 0;
2285                                 dev_warn(dev, "RSS disabled, "
2286                                         "Failed to set RSS cpu indirection table.");
2287                         }
2288                 } else {
2289                         rss_enable = 0;
2290                         dev_warn(dev, "RSS disabled, Failed to set RSS key.\n");
2291                 }
2292         }
2293
2294         return enic_set_niccfg(enic, rss_default_cpu, rss_hash_type,
2295                 rss_hash_bits, rss_base_cpu, rss_enable);
2296 }
2297
2298 static void enic_reset(struct work_struct *work)
2299 {
2300         struct enic *enic = container_of(work, struct enic, reset);
2301
2302         if (!netif_running(enic->netdev))
2303                 return;
2304
2305         rtnl_lock();
2306
2307         spin_lock(&enic->enic_api_lock);
2308         enic_stop(enic->netdev);
2309         enic_dev_soft_reset(enic);
2310         enic_reset_addr_lists(enic);
2311         enic_init_vnic_resources(enic);
2312         enic_set_rss_nic_cfg(enic);
2313         enic_dev_set_ig_vlan_rewrite_mode(enic);
2314         enic_open(enic->netdev);
2315         spin_unlock(&enic->enic_api_lock);
2316         call_netdevice_notifiers(NETDEV_REBOOT, enic->netdev);
2317
2318         rtnl_unlock();
2319 }
2320
2321 static void enic_tx_hang_reset(struct work_struct *work)
2322 {
2323         struct enic *enic = container_of(work, struct enic, tx_hang_reset);
2324
2325         rtnl_lock();
2326
2327         spin_lock(&enic->enic_api_lock);
2328         enic_dev_hang_notify(enic);
2329         enic_stop(enic->netdev);
2330         enic_dev_hang_reset(enic);
2331         enic_reset_addr_lists(enic);
2332         enic_init_vnic_resources(enic);
2333         enic_set_rss_nic_cfg(enic);
2334         enic_dev_set_ig_vlan_rewrite_mode(enic);
2335         enic_open(enic->netdev);
2336         spin_unlock(&enic->enic_api_lock);
2337         call_netdevice_notifiers(NETDEV_REBOOT, enic->netdev);
2338
2339         rtnl_unlock();
2340 }
2341
2342 static int enic_set_intr_mode(struct enic *enic)
2343 {
2344         unsigned int n = min_t(unsigned int, enic->rq_count, ENIC_RQ_MAX);
2345         unsigned int m = min_t(unsigned int, enic->wq_count, ENIC_WQ_MAX);
2346         unsigned int i;
2347
2348         /* Set interrupt mode (INTx, MSI, MSI-X) depending
2349          * on system capabilities.
2350          *
2351          * Try MSI-X first
2352          *
2353          * We need n RQs, m WQs, n+m CQs, and n+m+2 INTRs
2354          * (the second to last INTR is used for WQ/RQ errors)
2355          * (the last INTR is used for notifications)
2356          */
2357
2358         BUG_ON(ARRAY_SIZE(enic->msix_entry) < n + m + 2);
2359         for (i = 0; i < n + m + 2; i++)
2360                 enic->msix_entry[i].entry = i;
2361
2362         /* Use multiple RQs if RSS is enabled
2363          */
2364
2365         if (ENIC_SETTING(enic, RSS) &&
2366             enic->config.intr_mode < 1 &&
2367             enic->rq_count >= n &&
2368             enic->wq_count >= m &&
2369             enic->cq_count >= n + m &&
2370             enic->intr_count >= n + m + 2) {
2371
2372                 if (pci_enable_msix_range(enic->pdev, enic->msix_entry,
2373                                           n + m + 2, n + m + 2) > 0) {
2374
2375                         enic->rq_count = n;
2376                         enic->wq_count = m;
2377                         enic->cq_count = n + m;
2378                         enic->intr_count = n + m + 2;
2379
2380                         vnic_dev_set_intr_mode(enic->vdev,
2381                                 VNIC_DEV_INTR_MODE_MSIX);
2382
2383                         return 0;
2384                 }
2385         }
2386
2387         if (enic->config.intr_mode < 1 &&
2388             enic->rq_count >= 1 &&
2389             enic->wq_count >= m &&
2390             enic->cq_count >= 1 + m &&
2391             enic->intr_count >= 1 + m + 2) {
2392                 if (pci_enable_msix_range(enic->pdev, enic->msix_entry,
2393                                           1 + m + 2, 1 + m + 2) > 0) {
2394
2395                         enic->rq_count = 1;
2396                         enic->wq_count = m;
2397                         enic->cq_count = 1 + m;
2398                         enic->intr_count = 1 + 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         /* Next try MSI
2408          *
2409          * We need 1 RQ, 1 WQ, 2 CQs, and 1 INTR
2410          */
2411
2412         if (enic->config.intr_mode < 2 &&
2413             enic->rq_count >= 1 &&
2414             enic->wq_count >= 1 &&
2415             enic->cq_count >= 2 &&
2416             enic->intr_count >= 1 &&
2417             !pci_enable_msi(enic->pdev)) {
2418
2419                 enic->rq_count = 1;
2420                 enic->wq_count = 1;
2421                 enic->cq_count = 2;
2422                 enic->intr_count = 1;
2423
2424                 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSI);
2425
2426                 return 0;
2427         }
2428
2429         /* Next try INTx
2430          *
2431          * We need 1 RQ, 1 WQ, 2 CQs, and 3 INTRs
2432          * (the first INTR is used for WQ/RQ)
2433          * (the second INTR is used for WQ/RQ errors)
2434          * (the last INTR is used for notifications)
2435          */
2436
2437         if (enic->config.intr_mode < 3 &&
2438             enic->rq_count >= 1 &&
2439             enic->wq_count >= 1 &&
2440             enic->cq_count >= 2 &&
2441             enic->intr_count >= 3) {
2442
2443                 enic->rq_count = 1;
2444                 enic->wq_count = 1;
2445                 enic->cq_count = 2;
2446                 enic->intr_count = 3;
2447
2448                 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_INTX);
2449
2450                 return 0;
2451         }
2452
2453         vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
2454
2455         return -EINVAL;
2456 }
2457
2458 static void enic_clear_intr_mode(struct enic *enic)
2459 {
2460         switch (vnic_dev_get_intr_mode(enic->vdev)) {
2461         case VNIC_DEV_INTR_MODE_MSIX:
2462                 pci_disable_msix(enic->pdev);
2463                 break;
2464         case VNIC_DEV_INTR_MODE_MSI:
2465                 pci_disable_msi(enic->pdev);
2466                 break;
2467         default:
2468                 break;
2469         }
2470
2471         vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
2472 }
2473
2474 static const struct net_device_ops enic_netdev_dynamic_ops = {
2475         .ndo_open               = enic_open,
2476         .ndo_stop               = enic_stop,
2477         .ndo_start_xmit         = enic_hard_start_xmit,
2478         .ndo_get_stats64        = enic_get_stats,
2479         .ndo_validate_addr      = eth_validate_addr,
2480         .ndo_set_rx_mode        = enic_set_rx_mode,
2481         .ndo_set_mac_address    = enic_set_mac_address_dynamic,
2482         .ndo_change_mtu         = enic_change_mtu,
2483         .ndo_vlan_rx_add_vid    = enic_vlan_rx_add_vid,
2484         .ndo_vlan_rx_kill_vid   = enic_vlan_rx_kill_vid,
2485         .ndo_tx_timeout         = enic_tx_timeout,
2486         .ndo_set_vf_port        = enic_set_vf_port,
2487         .ndo_get_vf_port        = enic_get_vf_port,
2488         .ndo_set_vf_mac         = enic_set_vf_mac,
2489 #ifdef CONFIG_NET_POLL_CONTROLLER
2490         .ndo_poll_controller    = enic_poll_controller,
2491 #endif
2492 #ifdef CONFIG_RFS_ACCEL
2493         .ndo_rx_flow_steer      = enic_rx_flow_steer,
2494 #endif
2495         .ndo_udp_tunnel_add     = udp_tunnel_nic_add_port,
2496         .ndo_udp_tunnel_del     = udp_tunnel_nic_del_port,
2497         .ndo_features_check     = enic_features_check,
2498 };
2499
2500 static const struct net_device_ops enic_netdev_ops = {
2501         .ndo_open               = enic_open,
2502         .ndo_stop               = enic_stop,
2503         .ndo_start_xmit         = enic_hard_start_xmit,
2504         .ndo_get_stats64        = enic_get_stats,
2505         .ndo_validate_addr      = eth_validate_addr,
2506         .ndo_set_mac_address    = enic_set_mac_address,
2507         .ndo_set_rx_mode        = enic_set_rx_mode,
2508         .ndo_change_mtu         = enic_change_mtu,
2509         .ndo_vlan_rx_add_vid    = enic_vlan_rx_add_vid,
2510         .ndo_vlan_rx_kill_vid   = enic_vlan_rx_kill_vid,
2511         .ndo_tx_timeout         = enic_tx_timeout,
2512         .ndo_set_vf_port        = enic_set_vf_port,
2513         .ndo_get_vf_port        = enic_get_vf_port,
2514         .ndo_set_vf_mac         = enic_set_vf_mac,
2515 #ifdef CONFIG_NET_POLL_CONTROLLER
2516         .ndo_poll_controller    = enic_poll_controller,
2517 #endif
2518 #ifdef CONFIG_RFS_ACCEL
2519         .ndo_rx_flow_steer      = enic_rx_flow_steer,
2520 #endif
2521         .ndo_udp_tunnel_add     = udp_tunnel_nic_add_port,
2522         .ndo_udp_tunnel_del     = udp_tunnel_nic_del_port,
2523         .ndo_features_check     = enic_features_check,
2524 };
2525
2526 static void enic_dev_deinit(struct enic *enic)
2527 {
2528         unsigned int i;
2529
2530         for (i = 0; i < enic->rq_count; i++) {
2531                 napi_hash_del(&enic->napi[i]);
2532                 netif_napi_del(&enic->napi[i]);
2533         }
2534         if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX)
2535                 for (i = 0; i < enic->wq_count; i++)
2536                         netif_napi_del(&enic->napi[enic_cq_wq(enic, i)]);
2537
2538         enic_free_vnic_resources(enic);
2539         enic_clear_intr_mode(enic);
2540         enic_free_affinity_hint(enic);
2541 }
2542
2543 static void enic_kdump_kernel_config(struct enic *enic)
2544 {
2545         if (is_kdump_kernel()) {
2546                 dev_info(enic_get_dev(enic), "Running from within kdump kernel. Using minimal resources\n");
2547                 enic->rq_count = 1;
2548                 enic->wq_count = 1;
2549                 enic->config.rq_desc_count = ENIC_MIN_RQ_DESCS;
2550                 enic->config.wq_desc_count = ENIC_MIN_WQ_DESCS;
2551                 enic->config.mtu = min_t(u16, 1500, enic->config.mtu);
2552         }
2553 }
2554
2555 static int enic_dev_init(struct enic *enic)
2556 {
2557         struct device *dev = enic_get_dev(enic);
2558         struct net_device *netdev = enic->netdev;
2559         unsigned int i;
2560         int err;
2561
2562         /* Get interrupt coalesce timer info */
2563         err = enic_dev_intr_coal_timer_info(enic);
2564         if (err) {
2565                 dev_warn(dev, "Using default conversion factor for "
2566                         "interrupt coalesce timer\n");
2567                 vnic_dev_intr_coal_timer_info_default(enic->vdev);
2568         }
2569
2570         /* Get vNIC configuration
2571          */
2572
2573         err = enic_get_vnic_config(enic);
2574         if (err) {
2575                 dev_err(dev, "Get vNIC configuration failed, aborting\n");
2576                 return err;
2577         }
2578
2579         /* Get available resource counts
2580          */
2581
2582         enic_get_res_counts(enic);
2583
2584         /* modify resource count if we are in kdump_kernel
2585          */
2586         enic_kdump_kernel_config(enic);
2587
2588         /* Set interrupt mode based on resource counts and system
2589          * capabilities
2590          */
2591
2592         err = enic_set_intr_mode(enic);
2593         if (err) {
2594                 dev_err(dev, "Failed to set intr mode based on resource "
2595                         "counts and system capabilities, aborting\n");
2596                 return err;
2597         }
2598
2599         /* Allocate and configure vNIC resources
2600          */
2601
2602         err = enic_alloc_vnic_resources(enic);
2603         if (err) {
2604                 dev_err(dev, "Failed to alloc vNIC resources, aborting\n");
2605                 goto err_out_free_vnic_resources;
2606         }
2607
2608         enic_init_vnic_resources(enic);
2609
2610         err = enic_set_rss_nic_cfg(enic);
2611         if (err) {
2612                 dev_err(dev, "Failed to config nic, aborting\n");
2613                 goto err_out_free_vnic_resources;
2614         }
2615
2616         switch (vnic_dev_get_intr_mode(enic->vdev)) {
2617         default:
2618                 netif_napi_add(netdev, &enic->napi[0], enic_poll, 64);
2619                 break;
2620         case VNIC_DEV_INTR_MODE_MSIX:
2621                 for (i = 0; i < enic->rq_count; i++) {
2622                         netif_napi_add(netdev, &enic->napi[i],
2623                                 enic_poll_msix_rq, NAPI_POLL_WEIGHT);
2624                 }
2625                 for (i = 0; i < enic->wq_count; i++)
2626                         netif_napi_add(netdev, &enic->napi[enic_cq_wq(enic, i)],
2627                                        enic_poll_msix_wq, NAPI_POLL_WEIGHT);
2628                 break;
2629         }
2630
2631         return 0;
2632
2633 err_out_free_vnic_resources:
2634         enic_free_affinity_hint(enic);
2635         enic_clear_intr_mode(enic);
2636         enic_free_vnic_resources(enic);
2637
2638         return err;
2639 }
2640
2641 static void enic_iounmap(struct enic *enic)
2642 {
2643         unsigned int i;
2644
2645         for (i = 0; i < ARRAY_SIZE(enic->bar); i++)
2646                 if (enic->bar[i].vaddr)
2647                         iounmap(enic->bar[i].vaddr);
2648 }
2649
2650 static int enic_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2651 {
2652         struct device *dev = &pdev->dev;
2653         struct net_device *netdev;
2654         struct enic *enic;
2655         int using_dac = 0;
2656         unsigned int i;
2657         int err;
2658 #ifdef CONFIG_PCI_IOV
2659         int pos = 0;
2660 #endif
2661         int num_pps = 1;
2662
2663         /* Allocate net device structure and initialize.  Private
2664          * instance data is initialized to zero.
2665          */
2666
2667         netdev = alloc_etherdev_mqs(sizeof(struct enic),
2668                                     ENIC_RQ_MAX, ENIC_WQ_MAX);
2669         if (!netdev)
2670                 return -ENOMEM;
2671
2672         pci_set_drvdata(pdev, netdev);
2673
2674         SET_NETDEV_DEV(netdev, &pdev->dev);
2675
2676         enic = netdev_priv(netdev);
2677         enic->netdev = netdev;
2678         enic->pdev = pdev;
2679
2680         /* Setup PCI resources
2681          */
2682
2683         err = pci_enable_device_mem(pdev);
2684         if (err) {
2685                 dev_err(dev, "Cannot enable PCI device, aborting\n");
2686                 goto err_out_free_netdev;
2687         }
2688
2689         err = pci_request_regions(pdev, DRV_NAME);
2690         if (err) {
2691                 dev_err(dev, "Cannot request PCI regions, aborting\n");
2692                 goto err_out_disable_device;
2693         }
2694
2695         pci_set_master(pdev);
2696
2697         /* Query PCI controller on system for DMA addressing
2698          * limitation for the device.  Try 47-bit first, and
2699          * fail to 32-bit.
2700          */
2701
2702         err = pci_set_dma_mask(pdev, DMA_BIT_MASK(47));
2703         if (err) {
2704                 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
2705                 if (err) {
2706                         dev_err(dev, "No usable DMA configuration, aborting\n");
2707                         goto err_out_release_regions;
2708                 }
2709                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
2710                 if (err) {
2711                         dev_err(dev, "Unable to obtain %u-bit DMA "
2712                                 "for consistent allocations, aborting\n", 32);
2713                         goto err_out_release_regions;
2714                 }
2715         } else {
2716                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(47));
2717                 if (err) {
2718                         dev_err(dev, "Unable to obtain %u-bit DMA "
2719                                 "for consistent allocations, aborting\n", 47);
2720                         goto err_out_release_regions;
2721                 }
2722                 using_dac = 1;
2723         }
2724
2725         /* Map vNIC resources from BAR0-5
2726          */
2727
2728         for (i = 0; i < ARRAY_SIZE(enic->bar); i++) {
2729                 if (!(pci_resource_flags(pdev, i) & IORESOURCE_MEM))
2730                         continue;
2731                 enic->bar[i].len = pci_resource_len(pdev, i);
2732                 enic->bar[i].vaddr = pci_iomap(pdev, i, enic->bar[i].len);
2733                 if (!enic->bar[i].vaddr) {
2734                         dev_err(dev, "Cannot memory-map BAR %d, aborting\n", i);
2735                         err = -ENODEV;
2736                         goto err_out_iounmap;
2737                 }
2738                 enic->bar[i].bus_addr = pci_resource_start(pdev, i);
2739         }
2740
2741         /* Register vNIC device
2742          */
2743
2744         enic->vdev = vnic_dev_register(NULL, enic, pdev, enic->bar,
2745                 ARRAY_SIZE(enic->bar));
2746         if (!enic->vdev) {
2747                 dev_err(dev, "vNIC registration failed, aborting\n");
2748                 err = -ENODEV;
2749                 goto err_out_iounmap;
2750         }
2751
2752         err = vnic_devcmd_init(enic->vdev);
2753
2754         if (err)
2755                 goto err_out_vnic_unregister;
2756
2757 #ifdef CONFIG_PCI_IOV
2758         /* Get number of subvnics */
2759         pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_SRIOV);
2760         if (pos) {
2761                 pci_read_config_word(pdev, pos + PCI_SRIOV_TOTAL_VF,
2762                         &enic->num_vfs);
2763                 if (enic->num_vfs) {
2764                         err = pci_enable_sriov(pdev, enic->num_vfs);
2765                         if (err) {
2766                                 dev_err(dev, "SRIOV enable failed, aborting."
2767                                         " pci_enable_sriov() returned %d\n",
2768                                         err);
2769                                 goto err_out_vnic_unregister;
2770                         }
2771                         enic->priv_flags |= ENIC_SRIOV_ENABLED;
2772                         num_pps = enic->num_vfs;
2773                 }
2774         }
2775 #endif
2776
2777         /* Allocate structure for port profiles */
2778         enic->pp = kcalloc(num_pps, sizeof(*enic->pp), GFP_KERNEL);
2779         if (!enic->pp) {
2780                 err = -ENOMEM;
2781                 goto err_out_disable_sriov_pp;
2782         }
2783
2784         /* Issue device open to get device in known state
2785          */
2786
2787         err = enic_dev_open(enic);
2788         if (err) {
2789                 dev_err(dev, "vNIC dev open failed, aborting\n");
2790                 goto err_out_disable_sriov;
2791         }
2792
2793         /* Setup devcmd lock
2794          */
2795
2796         spin_lock_init(&enic->devcmd_lock);
2797         spin_lock_init(&enic->enic_api_lock);
2798
2799         /*
2800          * Set ingress vlan rewrite mode before vnic initialization
2801          */
2802
2803         err = enic_dev_set_ig_vlan_rewrite_mode(enic);
2804         if (err) {
2805                 dev_err(dev,
2806                         "Failed to set ingress vlan rewrite mode, aborting.\n");
2807                 goto err_out_dev_close;
2808         }
2809
2810         /* Issue device init to initialize the vnic-to-switch link.
2811          * We'll start with carrier off and wait for link UP
2812          * notification later to turn on carrier.  We don't need
2813          * to wait here for the vnic-to-switch link initialization
2814          * to complete; link UP notification is the indication that
2815          * the process is complete.
2816          */
2817
2818         netif_carrier_off(netdev);
2819
2820         /* Do not call dev_init for a dynamic vnic.
2821          * For a dynamic vnic, init_prov_info will be
2822          * called later by an upper layer.
2823          */
2824
2825         if (!enic_is_dynamic(enic)) {
2826                 err = vnic_dev_init(enic->vdev, 0);
2827                 if (err) {
2828                         dev_err(dev, "vNIC dev init failed, aborting\n");
2829                         goto err_out_dev_close;
2830                 }
2831         }
2832
2833         err = enic_dev_init(enic);
2834         if (err) {
2835                 dev_err(dev, "Device initialization failed, aborting\n");
2836                 goto err_out_dev_close;
2837         }
2838
2839         netif_set_real_num_tx_queues(netdev, enic->wq_count);
2840         netif_set_real_num_rx_queues(netdev, enic->rq_count);
2841
2842         /* Setup notification timer, HW reset task, and wq locks
2843          */
2844
2845         timer_setup(&enic->notify_timer, enic_notify_timer, 0);
2846
2847         enic_rfs_flw_tbl_init(enic);
2848         enic_set_rx_coal_setting(enic);
2849         INIT_WORK(&enic->reset, enic_reset);
2850         INIT_WORK(&enic->tx_hang_reset, enic_tx_hang_reset);
2851         INIT_WORK(&enic->change_mtu_work, enic_change_mtu_work);
2852
2853         for (i = 0; i < enic->wq_count; i++)
2854                 spin_lock_init(&enic->wq_lock[i]);
2855
2856         /* Register net device
2857          */
2858
2859         enic->port_mtu = enic->config.mtu;
2860
2861         err = enic_set_mac_addr(netdev, enic->mac_addr);
2862         if (err) {
2863                 dev_err(dev, "Invalid MAC address, aborting\n");
2864                 goto err_out_dev_deinit;
2865         }
2866
2867         enic->tx_coalesce_usecs = enic->config.intr_timer_usec;
2868         /* rx coalesce time already got initialized. This gets used
2869          * if adaptive coal is turned off
2870          */
2871         enic->rx_coalesce_usecs = enic->tx_coalesce_usecs;
2872
2873         if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic))
2874                 netdev->netdev_ops = &enic_netdev_dynamic_ops;
2875         else
2876                 netdev->netdev_ops = &enic_netdev_ops;
2877
2878         netdev->watchdog_timeo = 2 * HZ;
2879         enic_set_ethtool_ops(netdev);
2880
2881         netdev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
2882         if (ENIC_SETTING(enic, LOOP)) {
2883                 netdev->features &= ~NETIF_F_HW_VLAN_CTAG_TX;
2884                 enic->loop_enable = 1;
2885                 enic->loop_tag = enic->config.loop_tag;
2886                 dev_info(dev, "loopback tag=0x%04x\n", enic->loop_tag);
2887         }
2888         if (ENIC_SETTING(enic, TXCSUM))
2889                 netdev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM;
2890         if (ENIC_SETTING(enic, TSO))
2891                 netdev->hw_features |= NETIF_F_TSO |
2892                         NETIF_F_TSO6 | NETIF_F_TSO_ECN;
2893         if (ENIC_SETTING(enic, RSS))
2894                 netdev->hw_features |= NETIF_F_RXHASH;
2895         if (ENIC_SETTING(enic, RXCSUM))
2896                 netdev->hw_features |= NETIF_F_RXCSUM;
2897         if (ENIC_SETTING(enic, VXLAN)) {
2898                 u64 patch_level;
2899                 u64 a1 = 0;
2900
2901                 netdev->hw_enc_features |= NETIF_F_RXCSUM               |
2902                                            NETIF_F_TSO                  |
2903                                            NETIF_F_TSO6                 |
2904                                            NETIF_F_TSO_ECN              |
2905                                            NETIF_F_GSO_UDP_TUNNEL       |
2906                                            NETIF_F_HW_CSUM              |
2907                                            NETIF_F_GSO_UDP_TUNNEL_CSUM;
2908                 netdev->hw_features |= netdev->hw_enc_features;
2909                 /* get bit mask from hw about supported offload bit level
2910                  * BIT(0) = fw supports patch_level 0
2911                  *          fcoe bit = encap
2912                  *          fcoe_fc_crc_ok = outer csum ok
2913                  * BIT(1) = always set by fw
2914                  * BIT(2) = fw supports patch_level 2
2915                  *          BIT(0) in rss_hash = encap
2916                  *          BIT(1,2) in rss_hash = outer_ip_csum_ok/
2917                  *                                 outer_tcp_csum_ok
2918                  * used in enic_rq_indicate_buf
2919                  */
2920                 err = vnic_dev_get_supported_feature_ver(enic->vdev,
2921                                                          VIC_FEATURE_VXLAN,
2922                                                          &patch_level, &a1);
2923                 if (err)
2924                         patch_level = 0;
2925                 enic->vxlan.flags = (u8)a1;
2926                 /* mask bits that are supported by driver
2927                  */
2928                 patch_level &= BIT_ULL(0) | BIT_ULL(2);
2929                 patch_level = fls(patch_level);
2930                 patch_level = patch_level ? patch_level - 1 : 0;
2931                 enic->vxlan.patch_level = patch_level;
2932
2933                 if (vnic_dev_get_res_count(enic->vdev, RES_TYPE_WQ) == 1 ||
2934                     enic->vxlan.flags & ENIC_VXLAN_MULTI_WQ) {
2935                         netdev->udp_tunnel_nic_info = &enic_udp_tunnels_v4;
2936                         if (enic->vxlan.flags & ENIC_VXLAN_OUTER_IPV6)
2937                                 netdev->udp_tunnel_nic_info = &enic_udp_tunnels;
2938                 }
2939         }
2940
2941         netdev->features |= netdev->hw_features;
2942         netdev->vlan_features |= netdev->features;
2943
2944 #ifdef CONFIG_RFS_ACCEL
2945         netdev->hw_features |= NETIF_F_NTUPLE;
2946 #endif
2947
2948         if (using_dac)
2949                 netdev->features |= NETIF_F_HIGHDMA;
2950
2951         netdev->priv_flags |= IFF_UNICAST_FLT;
2952
2953         /* MTU range: 68 - 9000 */
2954         netdev->min_mtu = ENIC_MIN_MTU;
2955         netdev->max_mtu = ENIC_MAX_MTU;
2956         netdev->mtu     = enic->port_mtu;
2957
2958         err = register_netdev(netdev);
2959         if (err) {
2960                 dev_err(dev, "Cannot register net device, aborting\n");
2961                 goto err_out_dev_deinit;
2962         }
2963         enic->rx_copybreak = RX_COPYBREAK_DEFAULT;
2964
2965         return 0;
2966
2967 err_out_dev_deinit:
2968         enic_dev_deinit(enic);
2969 err_out_dev_close:
2970         vnic_dev_close(enic->vdev);
2971 err_out_disable_sriov:
2972         kfree(enic->pp);
2973 err_out_disable_sriov_pp:
2974 #ifdef CONFIG_PCI_IOV
2975         if (enic_sriov_enabled(enic)) {
2976                 pci_disable_sriov(pdev);
2977                 enic->priv_flags &= ~ENIC_SRIOV_ENABLED;
2978         }
2979 #endif
2980 err_out_vnic_unregister:
2981         vnic_dev_unregister(enic->vdev);
2982 err_out_iounmap:
2983         enic_iounmap(enic);
2984 err_out_release_regions:
2985         pci_release_regions(pdev);
2986 err_out_disable_device:
2987         pci_disable_device(pdev);
2988 err_out_free_netdev:
2989         free_netdev(netdev);
2990
2991         return err;
2992 }
2993
2994 static void enic_remove(struct pci_dev *pdev)
2995 {
2996         struct net_device *netdev = pci_get_drvdata(pdev);
2997
2998         if (netdev) {
2999                 struct enic *enic = netdev_priv(netdev);
3000
3001                 cancel_work_sync(&enic->reset);
3002                 cancel_work_sync(&enic->change_mtu_work);
3003                 unregister_netdev(netdev);
3004                 enic_dev_deinit(enic);
3005                 vnic_dev_close(enic->vdev);
3006 #ifdef CONFIG_PCI_IOV
3007                 if (enic_sriov_enabled(enic)) {
3008                         pci_disable_sriov(pdev);
3009                         enic->priv_flags &= ~ENIC_SRIOV_ENABLED;
3010                 }
3011 #endif
3012                 kfree(enic->pp);
3013                 vnic_dev_unregister(enic->vdev);
3014                 enic_iounmap(enic);
3015                 pci_release_regions(pdev);
3016                 pci_disable_device(pdev);
3017                 free_netdev(netdev);
3018         }
3019 }
3020
3021 static struct pci_driver enic_driver = {
3022         .name = DRV_NAME,
3023         .id_table = enic_id_table,
3024         .probe = enic_probe,
3025         .remove = enic_remove,
3026 };
3027
3028 static int __init enic_init_module(void)
3029 {
3030         return pci_register_driver(&enic_driver);
3031 }
3032
3033 static void __exit enic_cleanup_module(void)
3034 {
3035         pci_unregister_driver(&enic_driver);
3036 }
3037
3038 module_init(enic_init_module);
3039 module_exit(enic_cleanup_module);