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 #ifdef CONFIG_NET_RX_BUSY_POLL
  47 #include <net/busy_poll.h>
  48 #endif
  49 #include <linux/crash_dump.h>
  50
  51 #include "cq_enet_desc.h"
  52 #include "vnic_dev.h"
  53 #include "vnic_intr.h"
  54 #include "vnic_stats.h"
  55 #include "vnic_vic.h"
  56 #include "enic_res.h"
  57 #include "enic.h"
  58 #include "enic_dev.h"
  59 #include "enic_pp.h"
  60 #include "enic_clsf.h"
  61
  62 #define ENIC_NOTIFY_TIMER_PERIOD        (2 * HZ)
  63 #define WQ_ENET_MAX_DESC_LEN            (1 << WQ_ENET_LEN_BITS)
  64 #define MAX_TSO                         (1 << 16)
  65 #define ENIC_DESC_MAX_SPLITS            (MAX_TSO / WQ_ENET_MAX_DESC_LEN + 1)
  66
  67 #define PCI_DEVICE_ID_CISCO_VIC_ENET         0x0043  /* ethernet vnic */
  68 #define PCI_DEVICE_ID_CISCO_VIC_ENET_DYN     0x0044  /* enet dynamic vnic */
  69 #define PCI_DEVICE_ID_CISCO_VIC_ENET_VF      0x0071  /* enet SRIOV VF */
  70
  71 #define RX_COPYBREAK_DEFAULT            256
  72
  73 /* Supported devices */
  74 static const struct pci_device_id enic_id_table[] = {
  75         { PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET) },
  76         { PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET_DYN) },
  77         { PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET_VF) },
  78         { 0, }  /* end of table */
  79 };
  80
  81 MODULE_DESCRIPTION(DRV_DESCRIPTION);
  82 MODULE_AUTHOR("Scott Feldman <scofeldm@cisco.com>");
  83 MODULE_LICENSE("GPL");
  84 MODULE_VERSION(DRV_VERSION);
  85 MODULE_DEVICE_TABLE(pci, enic_id_table);
  86
  87 #define ENIC_LARGE_PKT_THRESHOLD                1000
  88 #define ENIC_MAX_COALESCE_TIMERS                10
  89 /*  Interrupt moderation table, which will be used to decide the
  90  *  coalescing timer values
  91  *  {rx_rate in Mbps, mapping percentage of the range}
  92  */
  93 static struct enic_intr_mod_table mod_table[ENIC_MAX_COALESCE_TIMERS + 1] = {
  94         {4000,  0},
  95         {4400, 10},
  96         {5060, 20},
  97         {5230, 30},
  98         {5540, 40},
  99         {5820, 50},
 100         {6120, 60},
 101         {6435, 70},
 102         {6745, 80},
 103         {7000, 90},
 104         {0xFFFFFFFF, 100}
 105 };
 106
 107 /* This table helps the driver to pick different ranges for rx coalescing
 108  * timer depending on the link speed.
 109  */
 110 static struct enic_intr_mod_range mod_range[ENIC_MAX_LINK_SPEEDS] = {
 111         {0,  0}, /* 0  - 4  Gbps */
 112         {0,  3}, /* 4  - 10 Gbps */
 113         {3,  6}, /* 10 - 40 Gbps */
 114 };
 115
 116 static void enic_init_affinity_hint(struct enic *enic)
 117 {
 118         int numa_node = dev_to_node(&enic->pdev->dev);
 119         int i;
 120
 121         for (i = 0; i < enic->intr_count; i++) {
 122                 if (enic_is_err_intr(enic, i) || enic_is_notify_intr(enic, i) ||
 123                     (cpumask_available(enic->msix[i].affinity_mask) &&
 124                      !cpumask_empty(enic->msix[i].affinity_mask)))
 125                         continue;
 126                 if (zalloc_cpumask_var(&enic->msix[i].affinity_mask,
 127                                        GFP_KERNEL))
 128                         cpumask_set_cpu(cpumask_local_spread(i, numa_node),
 129                                         enic->msix[i].affinity_mask);
 130         }
 131 }
 132
 133 static void enic_free_affinity_hint(struct enic *enic)
 134 {
 135         int i;
 136
 137         for (i = 0; i < enic->intr_count; i++) {
 138                 if (enic_is_err_intr(enic, i) || enic_is_notify_intr(enic, i))
 139                         continue;
 140                 free_cpumask_var(enic->msix[i].affinity_mask);
 141         }
 142 }
 143
 144 static void enic_set_affinity_hint(struct enic *enic)
 145 {
 146         int i;
 147         int err;
 148
 149         for (i = 0; i < enic->intr_count; i++) {
 150                 if (enic_is_err_intr(enic, i)           ||
 151                     enic_is_notify_intr(enic, i)        ||
 152                     !cpumask_available(enic->msix[i].affinity_mask) ||
 153                     cpumask_empty(enic->msix[i].affinity_mask))
 154                         continue;
 155                 err = irq_set_affinity_hint(enic->msix_entry[i].vector,
 156                                             enic->msix[i].affinity_mask);
 157                 if (err)
 158                         netdev_warn(enic->netdev, "irq_set_affinity_hint failed, err %d\n",
 159                                     err);
 160         }
 161
 162         for (i = 0; i < enic->wq_count; i++) {
 163                 int wq_intr = enic_msix_wq_intr(enic, i);
 164
 165                 if (cpumask_available(enic->msix[wq_intr].affinity_mask) &&
 166                     !cpumask_empty(enic->msix[wq_intr].affinity_mask))
 167                         netif_set_xps_queue(enic->netdev,
 168                                             enic->msix[wq_intr].affinity_mask,
 169                                             i);
 170         }
 171 }
 172
 173 static void enic_unset_affinity_hint(struct enic *enic)
 174 {
 175         int i;
 176
 177         for (i = 0; i < enic->intr_count; i++)
 178                 irq_set_affinity_hint(enic->msix_entry[i].vector, NULL);
 179 }
 180
 181 int enic_is_dynamic(struct enic *enic)
 182 {
 183         return enic->pdev->device == PCI_DEVICE_ID_CISCO_VIC_ENET_DYN;
 184 }
 185
 186 int enic_sriov_enabled(struct enic *enic)
 187 {
 188         return (enic->priv_flags & ENIC_SRIOV_ENABLED) ? 1 : 0;
 189 }
 190
 191 static int enic_is_sriov_vf(struct enic *enic)
 192 {
 193         return enic->pdev->device == PCI_DEVICE_ID_CISCO_VIC_ENET_VF;
 194 }
 195
 196 int enic_is_valid_vf(struct enic *enic, int vf)
 197 {
 198 #ifdef CONFIG_PCI_IOV
 199         return vf >= 0 && vf < enic->num_vfs;
 200 #else
 201         return 0;
 202 #endif
 203 }
 204
 205 static void enic_free_wq_buf(struct vnic_wq *wq, struct vnic_wq_buf *buf)
 206 {
 207         struct enic *enic = vnic_dev_priv(wq->vdev);
 208
 209         if (buf->sop)
 210                 pci_unmap_single(enic->pdev, buf->dma_addr,
 211                         buf->len, PCI_DMA_TODEVICE);
 212         else
 213                 pci_unmap_page(enic->pdev, buf->dma_addr,
 214                         buf->len, PCI_DMA_TODEVICE);
 215
 216         if (buf->os_buf)
 217                 dev_kfree_skb_any(buf->os_buf);
 218 }
 219
 220 static void enic_wq_free_buf(struct vnic_wq *wq,
 221         struct cq_desc *cq_desc, struct vnic_wq_buf *buf, void *opaque)
 222 {
 223         enic_free_wq_buf(wq, buf);
 224 }
 225
 226 static int enic_wq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
 227         u8 type, u16 q_number, u16 completed_index, void *opaque)
 228 {
 229         struct enic *enic = vnic_dev_priv(vdev);
 230
 231         spin_lock(&enic->wq_lock[q_number]);
 232
 233         vnic_wq_service(&enic->wq[q_number], cq_desc,
 234                 completed_index, enic_wq_free_buf,
 235                 opaque);
 236
 237         if (netif_tx_queue_stopped(netdev_get_tx_queue(enic->netdev, q_number)) &&
 238             vnic_wq_desc_avail(&enic->wq[q_number]) >=
 239             (MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS))
 240                 netif_wake_subqueue(enic->netdev, q_number);
 241
 242         spin_unlock(&enic->wq_lock[q_number]);
 243
 244         return 0;
 245 }
 246
 247 static bool enic_log_q_error(struct enic *enic)
 248 {
 249         unsigned int i;
 250         u32 error_status;
 251         bool err = false;
 252
 253         for (i = 0; i < enic->wq_count; i++) {
 254                 error_status = vnic_wq_error_status(&enic->wq[i]);
 255                 err |= error_status;
 256                 if (error_status)
 257                         netdev_err(enic->netdev, "WQ[%d] error_status %d\n",
 258                                 i, error_status);
 259         }
 260
 261         for (i = 0; i < enic->rq_count; i++) {
 262                 error_status = vnic_rq_error_status(&enic->rq[i]);
 263                 err |= error_status;
 264                 if (error_status)
 265                         netdev_err(enic->netdev, "RQ[%d] error_status %d\n",
 266                                 i, error_status);
 267         }
 268
 269         return err;
 270 }
 271
 272 static void enic_msglvl_check(struct enic *enic)
 273 {
 274         u32 msg_enable = vnic_dev_msg_lvl(enic->vdev);
 275
 276         if (msg_enable != enic->msg_enable) {
 277                 netdev_info(enic->netdev, "msg lvl changed from 0x%x to 0x%x\n",
 278                         enic->msg_enable, msg_enable);
 279                 enic->msg_enable = msg_enable;
 280         }
 281 }
 282
 283 static void enic_mtu_check(struct enic *enic)
 284 {
 285         u32 mtu = vnic_dev_mtu(enic->vdev);
 286         struct net_device *netdev = enic->netdev;
 287
 288         if (mtu && mtu != enic->port_mtu) {
 289                 enic->port_mtu = mtu;
 290                 if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic)) {
 291                         mtu = max_t(int, ENIC_MIN_MTU,
 292                                 min_t(int, ENIC_MAX_MTU, mtu));
 293                         if (mtu != netdev->mtu)
 294                                 schedule_work(&enic->change_mtu_work);
 295                 } else {
 296                         if (mtu < netdev->mtu)
 297                                 netdev_warn(netdev,
 298                                         "interface MTU (%d) set higher "
 299                                         "than switch port MTU (%d)\n",
 300                                         netdev->mtu, mtu);
 301                 }
 302         }
 303 }
 304
 305 static void enic_link_check(struct enic *enic)
 306 {
 307         int link_status = vnic_dev_link_status(enic->vdev);
 308         int carrier_ok = netif_carrier_ok(enic->netdev);
 309
 310         if (link_status && !carrier_ok) {
 311                 netdev_info(enic->netdev, "Link UP\n");
 312                 netif_carrier_on(enic->netdev);
 313         } else if (!link_status && carrier_ok) {
 314                 netdev_info(enic->netdev, "Link DOWN\n");
 315                 netif_carrier_off(enic->netdev);
 316         }
 317 }
 318
 319 static void enic_notify_check(struct enic *enic)
 320 {
 321         enic_msglvl_check(enic);
 322         enic_mtu_check(enic);
 323         enic_link_check(enic);
 324 }
 325
 326 #define ENIC_TEST_INTR(pba, i) (pba & (1 << i))
 327
 328 static irqreturn_t enic_isr_legacy(int irq, void *data)
 329 {
 330         struct net_device *netdev = data;
 331         struct enic *enic = netdev_priv(netdev);
 332         unsigned int io_intr = enic_legacy_io_intr();
 333         unsigned int err_intr = enic_legacy_err_intr();
 334         unsigned int notify_intr = enic_legacy_notify_intr();
 335         u32 pba;
 336
 337         vnic_intr_mask(&enic->intr[io_intr]);
 338
 339         pba = vnic_intr_legacy_pba(enic->legacy_pba);
 340         if (!pba) {
 341                 vnic_intr_unmask(&enic->intr[io_intr]);
 342                 return IRQ_NONE;        /* not our interrupt */
 343         }
 344
 345         if (ENIC_TEST_INTR(pba, notify_intr)) {
 346                 enic_notify_check(enic);
 347                 vnic_intr_return_all_credits(&enic->intr[notify_intr]);
 348         }
 349
 350         if (ENIC_TEST_INTR(pba, err_intr)) {
 351                 vnic_intr_return_all_credits(&enic->intr[err_intr]);
 352                 enic_log_q_error(enic);
 353                 /* schedule recovery from WQ/RQ error */
 354                 schedule_work(&enic->reset);
 355                 return IRQ_HANDLED;
 356         }
 357
 358         if (ENIC_TEST_INTR(pba, io_intr))
 359                 napi_schedule_irqoff(&enic->napi[0]);
 360         else
 361                 vnic_intr_unmask(&enic->intr[io_intr]);
 362
 363         return IRQ_HANDLED;
 364 }
 365
 366 static irqreturn_t enic_isr_msi(int irq, void *data)
 367 {
 368         struct enic *enic = data;
 369
 370         /* With MSI, there is no sharing of interrupts, so this is
 371          * our interrupt and there is no need to ack it.  The device
 372          * is not providing per-vector masking, so the OS will not
 373          * write to PCI config space to mask/unmask the interrupt.
 374          * We're using mask_on_assertion for MSI, so the device
 375          * automatically masks the interrupt when the interrupt is
 376          * generated.  Later, when exiting polling, the interrupt
 377          * will be unmasked (see enic_poll).
 378          *
 379          * Also, the device uses the same PCIe Traffic Class (TC)
 380          * for Memory Write data and MSI, so there are no ordering
 381          * issues; the MSI will always arrive at the Root Complex
 382          * _after_ corresponding Memory Writes (i.e. descriptor
 383          * writes).
 384          */
 385
 386         napi_schedule_irqoff(&enic->napi[0]);
 387
 388         return IRQ_HANDLED;
 389 }
 390
 391 static irqreturn_t enic_isr_msix(int irq, void *data)
 392 {
 393         struct napi_struct *napi = data;
 394
 395         napi_schedule_irqoff(napi);
 396
 397         return IRQ_HANDLED;
 398 }
 399
 400 static irqreturn_t enic_isr_msix_err(int irq, void *data)
 401 {
 402         struct enic *enic = data;
 403         unsigned int intr = enic_msix_err_intr(enic);
 404
 405         vnic_intr_return_all_credits(&enic->intr[intr]);
 406
 407         if (enic_log_q_error(enic))
 408                 /* schedule recovery from WQ/RQ error */
 409                 schedule_work(&enic->reset);
 410
 411         return IRQ_HANDLED;
 412 }
 413
 414 static irqreturn_t enic_isr_msix_notify(int irq, void *data)
 415 {
 416         struct enic *enic = data;
 417         unsigned int intr = enic_msix_notify_intr(enic);
 418
 419         enic_notify_check(enic);
 420         vnic_intr_return_all_credits(&enic->intr[intr]);
 421
 422         return IRQ_HANDLED;
 423 }
 424
 425 static int enic_queue_wq_skb_cont(struct enic *enic, struct vnic_wq *wq,
 426                                   struct sk_buff *skb, unsigned int len_left,
 427                                   int loopback)
 428 {
 429         const skb_frag_t *frag;
 430         dma_addr_t dma_addr;
 431
 432         /* Queue additional data fragments */
 433         for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
 434                 len_left -= skb_frag_size(frag);
 435                 dma_addr = skb_frag_dma_map(&enic->pdev->dev, frag, 0,
 436                                             skb_frag_size(frag),
 437                                             DMA_TO_DEVICE);
 438                 if (unlikely(enic_dma_map_check(enic, dma_addr)))
 439                         return -ENOMEM;
 440                 enic_queue_wq_desc_cont(wq, skb, dma_addr, skb_frag_size(frag),
 441                                         (len_left == 0),        /* EOP? */
 442                                         loopback);
 443         }
 444
 445         return 0;
 446 }
 447
 448 static int enic_queue_wq_skb_vlan(struct enic *enic, struct vnic_wq *wq,
 449                                   struct sk_buff *skb, int vlan_tag_insert,
 450                                   unsigned int vlan_tag, int loopback)
 451 {
 452         unsigned int head_len = skb_headlen(skb);
 453         unsigned int len_left = skb->len - head_len;
 454         int eop = (len_left == 0);
 455         dma_addr_t dma_addr;
 456         int err = 0;
 457
 458         dma_addr = pci_map_single(enic->pdev, skb->data, head_len,
 459                                   PCI_DMA_TODEVICE);
 460         if (unlikely(enic_dma_map_check(enic, dma_addr)))
 461                 return -ENOMEM;
 462
 463         /* Queue the main skb fragment. The fragments are no larger
 464          * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less
 465          * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor
 466          * per fragment is queued.
 467          */
 468         enic_queue_wq_desc(wq, skb, dma_addr, head_len, vlan_tag_insert,
 469                            vlan_tag, eop, loopback);
 470
 471         if (!eop)
 472                 err = enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback);
 473
 474         return err;
 475 }
 476
 477 static int enic_queue_wq_skb_csum_l4(struct enic *enic, struct vnic_wq *wq,
 478                                      struct sk_buff *skb, int vlan_tag_insert,
 479                                      unsigned int vlan_tag, int loopback)
 480 {
 481         unsigned int head_len = skb_headlen(skb);
 482         unsigned int len_left = skb->len - head_len;
 483         unsigned int hdr_len = skb_checksum_start_offset(skb);
 484         unsigned int csum_offset = hdr_len + skb->csum_offset;
 485         int eop = (len_left == 0);
 486         dma_addr_t dma_addr;
 487         int err = 0;
 488
 489         dma_addr = pci_map_single(enic->pdev, skb->data, head_len,
 490                                   PCI_DMA_TODEVICE);
 491         if (unlikely(enic_dma_map_check(enic, dma_addr)))
 492                 return -ENOMEM;
 493
 494         /* Queue the main skb fragment. The fragments are no larger
 495          * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less
 496          * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor
 497          * per fragment is queued.
 498          */
 499         enic_queue_wq_desc_csum_l4(wq, skb, dma_addr, head_len, csum_offset,
 500                                    hdr_len, vlan_tag_insert, vlan_tag, eop,
 501                                    loopback);
 502
 503         if (!eop)
 504                 err = enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback);
 505
 506         return err;
 507 }
 508
 509 static int enic_queue_wq_skb_tso(struct enic *enic, struct vnic_wq *wq,
 510                                  struct sk_buff *skb, unsigned int mss,
 511                                  int vlan_tag_insert, unsigned int vlan_tag,
 512                                  int loopback)
 513 {
 514         unsigned int frag_len_left = skb_headlen(skb);
 515         unsigned int len_left = skb->len - frag_len_left;
 516         unsigned int hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
 517         int eop = (len_left == 0);
 518         unsigned int len;
 519         dma_addr_t dma_addr;
 520         unsigned int offset = 0;
 521         skb_frag_t *frag;
 522
 523         /* Preload TCP csum field with IP pseudo hdr calculated
 524          * with IP length set to zero.  HW will later add in length
 525          * to each TCP segment resulting from the TSO.
 526          */
 527
 528         if (skb->protocol == cpu_to_be16(ETH_P_IP)) {
 529                 ip_hdr(skb)->check = 0;
 530                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
 531                         ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 532         } else if (skb->protocol == cpu_to_be16(ETH_P_IPV6)) {
 533                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
 534                         &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 535         }
 536
 537         /* Queue WQ_ENET_MAX_DESC_LEN length descriptors
 538          * for the main skb fragment
 539          */
 540         while (frag_len_left) {
 541                 len = min(frag_len_left, (unsigned int)WQ_ENET_MAX_DESC_LEN);
 542                 dma_addr = pci_map_single(enic->pdev, skb->data + offset, len,
 543                                           PCI_DMA_TODEVICE);
 544                 if (unlikely(enic_dma_map_check(enic, dma_addr)))
 545                         return -ENOMEM;
 546                 enic_queue_wq_desc_tso(wq, skb, dma_addr, len, mss, hdr_len,
 547                                        vlan_tag_insert, vlan_tag,
 548                                        eop && (len == frag_len_left), loopback);
 549                 frag_len_left -= len;
 550                 offset += len;
 551         }
 552
 553         if (eop)
 554                 return 0;
 555
 556         /* Queue WQ_ENET_MAX_DESC_LEN length descriptors
 557          * for additional data fragments
 558          */
 559         for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
 560                 len_left -= skb_frag_size(frag);
 561                 frag_len_left = skb_frag_size(frag);
 562                 offset = 0;
 563
 564                 while (frag_len_left) {
 565                         len = min(frag_len_left,
 566                                 (unsigned int)WQ_ENET_MAX_DESC_LEN);
 567                         dma_addr = skb_frag_dma_map(&enic->pdev->dev, frag,
 568                                                     offset, len,
 569                                                     DMA_TO_DEVICE);
 570                         if (unlikely(enic_dma_map_check(enic, dma_addr)))
 571                                 return -ENOMEM;
 572                         enic_queue_wq_desc_cont(wq, skb, dma_addr, len,
 573                                                 (len_left == 0) &&
 574                                                  (len == frag_len_left),/*EOP*/
 575                                                 loopback);
 576                         frag_len_left -= len;
 577                         offset += len;
 578                 }
 579         }
 580
 581         return 0;
 582 }
 583
 584 static inline void enic_queue_wq_skb(struct enic *enic,
 585         struct vnic_wq *wq, struct sk_buff *skb)
 586 {
 587         unsigned int mss = skb_shinfo(skb)->gso_size;
 588         unsigned int vlan_tag = 0;
 589         int vlan_tag_insert = 0;
 590         int loopback = 0;
 591         int err;
 592
 593         if (skb_vlan_tag_present(skb)) {
 594                 /* VLAN tag from trunking driver */
 595                 vlan_tag_insert = 1;
 596                 vlan_tag = skb_vlan_tag_get(skb);
 597         } else if (enic->loop_enable) {
 598                 vlan_tag = enic->loop_tag;
 599                 loopback = 1;
 600         }
 601
 602         if (mss)
 603                 err = enic_queue_wq_skb_tso(enic, wq, skb, mss,
 604                                             vlan_tag_insert, vlan_tag,
 605                                             loopback);
 606         else if (skb->ip_summed == CHECKSUM_PARTIAL)
 607                 err = enic_queue_wq_skb_csum_l4(enic, wq, skb, vlan_tag_insert,
 608                                                 vlan_tag, loopback);
 609         else
 610                 err = enic_queue_wq_skb_vlan(enic, wq, skb, vlan_tag_insert,
 611                                              vlan_tag, loopback);
 612         if (unlikely(err)) {
 613                 struct vnic_wq_buf *buf;
 614
 615                 buf = wq->to_use->prev;
 616                 /* while not EOP of previous pkt && queue not empty.
 617                  * For all non EOP bufs, os_buf is NULL.
 618                  */
 619                 while (!buf->os_buf && (buf->next != wq->to_clean)) {
 620                         enic_free_wq_buf(wq, buf);
 621                         wq->ring.desc_avail++;
 622                         buf = buf->prev;
 623                 }
 624                 wq->to_use = buf->next;
 625                 dev_kfree_skb(skb);
 626         }
 627 }
 628
 629 /* netif_tx_lock held, process context with BHs disabled, or BH */
 630 static netdev_tx_t enic_hard_start_xmit(struct sk_buff *skb,
 631         struct net_device *netdev)
 632 {
 633         struct enic *enic = netdev_priv(netdev);
 634         struct vnic_wq *wq;
 635         unsigned int txq_map;
 636         struct netdev_queue *txq;
 637
 638         if (skb->len <= 0) {
 639                 dev_kfree_skb_any(skb);
 640                 return NETDEV_TX_OK;
 641         }
 642
 643         txq_map = skb_get_queue_mapping(skb) % enic->wq_count;
 644         wq = &enic->wq[txq_map];
 645         txq = netdev_get_tx_queue(netdev, txq_map);
 646
 647         /* Non-TSO sends must fit within ENIC_NON_TSO_MAX_DESC descs,
 648          * which is very likely.  In the off chance it's going to take
 649          * more than * ENIC_NON_TSO_MAX_DESC, linearize the skb.
 650          */
 651
 652         if (skb_shinfo(skb)->gso_size == 0 &&
 653             skb_shinfo(skb)->nr_frags + 1 > ENIC_NON_TSO_MAX_DESC &&
 654             skb_linearize(skb)) {
 655                 dev_kfree_skb_any(skb);
 656                 return NETDEV_TX_OK;
 657         }
 658
 659         spin_lock(&enic->wq_lock[txq_map]);
 660
 661         if (vnic_wq_desc_avail(wq) <
 662             skb_shinfo(skb)->nr_frags + ENIC_DESC_MAX_SPLITS) {
 663                 netif_tx_stop_queue(txq);
 664                 /* This is a hard error, log it */
 665                 netdev_err(netdev, "BUG! Tx ring full when queue awake!\n");
 666                 spin_unlock(&enic->wq_lock[txq_map]);
 667                 return NETDEV_TX_BUSY;
 668         }
 669
 670         enic_queue_wq_skb(enic, wq, skb);
 671
 672         if (vnic_wq_desc_avail(wq) < MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS)
 673                 netif_tx_stop_queue(txq);
 674         if (!skb->xmit_more || netif_xmit_stopped(txq))
 675                 vnic_wq_doorbell(wq);
 676
 677         spin_unlock(&enic->wq_lock[txq_map]);
 678
 679         return NETDEV_TX_OK;
 680 }
 681
 682 /* dev_base_lock rwlock held, nominally process context */
 683 static struct rtnl_link_stats64 *enic_get_stats(struct net_device *netdev,
 684                                                 struct rtnl_link_stats64 *net_stats)
 685 {
 686         struct enic *enic = netdev_priv(netdev);
 687         struct vnic_stats *stats;
 688         int err;
 689
 690         err = enic_dev_stats_dump(enic, &stats);
 691         /* return only when pci_zalloc_consistent fails in vnic_dev_stats_dump
 692          * For other failures, like devcmd failure, we return previously
 693          * recorded stats.
 694          */
 695         if (err == -ENOMEM)
 696                 return net_stats;
 697
 698         net_stats->tx_packets = stats->tx.tx_frames_ok;
 699         net_stats->tx_bytes = stats->tx.tx_bytes_ok;
 700         net_stats->tx_errors = stats->tx.tx_errors;
 701         net_stats->tx_dropped = stats->tx.tx_drops;
 702
 703         net_stats->rx_packets = stats->rx.rx_frames_ok;
 704         net_stats->rx_bytes = stats->rx.rx_bytes_ok;
 705         net_stats->rx_errors = stats->rx.rx_errors;
 706         net_stats->multicast = stats->rx.rx_multicast_frames_ok;
 707         net_stats->rx_over_errors = enic->rq_truncated_pkts;
 708         net_stats->rx_crc_errors = enic->rq_bad_fcs;
 709         net_stats->rx_dropped = stats->rx.rx_no_bufs + stats->rx.rx_drop;
 710
 711         return net_stats;
 712 }
 713
 714 static int enic_mc_sync(struct net_device *netdev, const u8 *mc_addr)
 715 {
 716         struct enic *enic = netdev_priv(netdev);
 717
 718         if (enic->mc_count == ENIC_MULTICAST_PERFECT_FILTERS) {
 719                 unsigned int mc_count = netdev_mc_count(netdev);
 720
 721                 netdev_warn(netdev, "Registering only %d out of %d multicast addresses\n",
 722                             ENIC_MULTICAST_PERFECT_FILTERS, mc_count);
 723
 724                 return -ENOSPC;
 725         }
 726
 727         enic_dev_add_addr(enic, mc_addr);
 728         enic->mc_count++;
 729
 730         return 0;
 731 }
 732
 733 static int enic_mc_unsync(struct net_device *netdev, const u8 *mc_addr)
 734 {
 735         struct enic *enic = netdev_priv(netdev);
 736
 737         enic_dev_del_addr(enic, mc_addr);
 738         enic->mc_count--;
 739
 740         return 0;
 741 }
 742
 743 static int enic_uc_sync(struct net_device *netdev, const u8 *uc_addr)
 744 {
 745         struct enic *enic = netdev_priv(netdev);
 746
 747         if (enic->uc_count == ENIC_UNICAST_PERFECT_FILTERS) {
 748                 unsigned int uc_count = netdev_uc_count(netdev);
 749
 750                 netdev_warn(netdev, "Registering only %d out of %d unicast addresses\n",
 751                             ENIC_UNICAST_PERFECT_FILTERS, uc_count);
 752
 753                 return -ENOSPC;
 754         }
 755
 756         enic_dev_add_addr(enic, uc_addr);
 757         enic->uc_count++;
 758
 759         return 0;
 760 }
 761
 762 static int enic_uc_unsync(struct net_device *netdev, const u8 *uc_addr)
 763 {
 764         struct enic *enic = netdev_priv(netdev);
 765
 766         enic_dev_del_addr(enic, uc_addr);
 767         enic->uc_count--;
 768
 769         return 0;
 770 }
 771
 772 void enic_reset_addr_lists(struct enic *enic)
 773 {
 774         struct net_device *netdev = enic->netdev;
 775
 776         __dev_uc_unsync(netdev, NULL);
 777         __dev_mc_unsync(netdev, NULL);
 778
 779         enic->mc_count = 0;
 780         enic->uc_count = 0;
 781         enic->flags = 0;
 782 }
 783
 784 static int enic_set_mac_addr(struct net_device *netdev, char *addr)
 785 {
 786         struct enic *enic = netdev_priv(netdev);
 787
 788         if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic)) {
 789                 if (!is_valid_ether_addr(addr) && !is_zero_ether_addr(addr))
 790                         return -EADDRNOTAVAIL;
 791         } else {
 792                 if (!is_valid_ether_addr(addr))
 793                         return -EADDRNOTAVAIL;
 794         }
 795
 796         memcpy(netdev->dev_addr, addr, netdev->addr_len);
 797
 798         return 0;
 799 }
 800
 801 static int enic_set_mac_address_dynamic(struct net_device *netdev, void *p)
 802 {
 803         struct enic *enic = netdev_priv(netdev);
 804         struct sockaddr *saddr = p;
 805         char *addr = saddr->sa_data;
 806         int err;
 807
 808         if (netif_running(enic->netdev)) {
 809                 err = enic_dev_del_station_addr(enic);
 810                 if (err)
 811                         return err;
 812         }
 813
 814         err = enic_set_mac_addr(netdev, addr);
 815         if (err)
 816                 return err;
 817
 818         if (netif_running(enic->netdev)) {
 819                 err = enic_dev_add_station_addr(enic);
 820                 if (err)
 821                         return err;
 822         }
 823
 824         return err;
 825 }
 826
 827 static int enic_set_mac_address(struct net_device *netdev, void *p)
 828 {
 829         struct sockaddr *saddr = p;
 830         char *addr = saddr->sa_data;
 831         struct enic *enic = netdev_priv(netdev);
 832         int err;
 833
 834         err = enic_dev_del_station_addr(enic);
 835         if (err)
 836                 return err;
 837
 838         err = enic_set_mac_addr(netdev, addr);
 839         if (err)
 840                 return err;
 841
 842         return enic_dev_add_station_addr(enic);
 843 }
 844
 845 /* netif_tx_lock held, BHs disabled */
 846 static void enic_set_rx_mode(struct net_device *netdev)
 847 {
 848         struct enic *enic = netdev_priv(netdev);
 849         int directed = 1;
 850         int multicast = (netdev->flags & IFF_MULTICAST) ? 1 : 0;
 851         int broadcast = (netdev->flags & IFF_BROADCAST) ? 1 : 0;
 852         int promisc = (netdev->flags & IFF_PROMISC) ||
 853                 netdev_uc_count(netdev) > ENIC_UNICAST_PERFECT_FILTERS;
 854         int allmulti = (netdev->flags & IFF_ALLMULTI) ||
 855                 netdev_mc_count(netdev) > ENIC_MULTICAST_PERFECT_FILTERS;
 856         unsigned int flags = netdev->flags |
 857                 (allmulti ? IFF_ALLMULTI : 0) |
 858                 (promisc ? IFF_PROMISC : 0);
 859
 860         if (enic->flags != flags) {
 861                 enic->flags = flags;
 862                 enic_dev_packet_filter(enic, directed,
 863                         multicast, broadcast, promisc, allmulti);
 864         }
 865
 866         if (!promisc) {
 867                 __dev_uc_sync(netdev, enic_uc_sync, enic_uc_unsync);
 868                 if (!allmulti)
 869                         __dev_mc_sync(netdev, enic_mc_sync, enic_mc_unsync);
 870         }
 871 }
 872
 873 /* netif_tx_lock held, BHs disabled */
 874 static void enic_tx_timeout(struct net_device *netdev)
 875 {
 876         struct enic *enic = netdev_priv(netdev);
 877         schedule_work(&enic->tx_hang_reset);
 878 }
 879
 880 static int enic_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
 881 {
 882         struct enic *enic = netdev_priv(netdev);
 883         struct enic_port_profile *pp;
 884         int err;
 885
 886         ENIC_PP_BY_INDEX(enic, vf, pp, &err);
 887         if (err)
 888                 return err;
 889
 890         if (is_valid_ether_addr(mac) || is_zero_ether_addr(mac)) {
 891                 if (vf == PORT_SELF_VF) {
 892                         memcpy(pp->vf_mac, mac, ETH_ALEN);
 893                         return 0;
 894                 } else {
 895                         /*
 896                          * For sriov vf's set the mac in hw
 897                          */
 898                         ENIC_DEVCMD_PROXY_BY_INDEX(vf, err, enic,
 899                                 vnic_dev_set_mac_addr, mac);
 900                         return enic_dev_status_to_errno(err);
 901                 }
 902         } else
 903                 return -EINVAL;
 904 }
 905
 906 static int enic_set_vf_port(struct net_device *netdev, int vf,
 907         struct nlattr *port[])
 908 {
 909         struct enic *enic = netdev_priv(netdev);
 910         struct enic_port_profile prev_pp;
 911         struct enic_port_profile *pp;
 912         int err = 0, restore_pp = 1;
 913
 914         ENIC_PP_BY_INDEX(enic, vf, pp, &err);
 915         if (err)
 916                 return err;
 917
 918         if (!port[IFLA_PORT_REQUEST])
 919                 return -EOPNOTSUPP;
 920
 921         memcpy(&prev_pp, pp, sizeof(*enic->pp));
 922         memset(pp, 0, sizeof(*enic->pp));
 923
 924         pp->set |= ENIC_SET_REQUEST;
 925         pp->request = nla_get_u8(port[IFLA_PORT_REQUEST]);
 926
 927         if (port[IFLA_PORT_PROFILE]) {
 928                 pp->set |= ENIC_SET_NAME;
 929                 memcpy(pp->name, nla_data(port[IFLA_PORT_PROFILE]),
 930                         PORT_PROFILE_MAX);
 931         }
 932
 933         if (port[IFLA_PORT_INSTANCE_UUID]) {
 934                 pp->set |= ENIC_SET_INSTANCE;
 935                 memcpy(pp->instance_uuid,
 936                         nla_data(port[IFLA_PORT_INSTANCE_UUID]), PORT_UUID_MAX);
 937         }
 938
 939         if (port[IFLA_PORT_HOST_UUID]) {
 940                 pp->set |= ENIC_SET_HOST;
 941                 memcpy(pp->host_uuid,
 942                         nla_data(port[IFLA_PORT_HOST_UUID]), PORT_UUID_MAX);
 943         }
 944
 945         if (vf == PORT_SELF_VF) {
 946                 /* Special case handling: mac came from IFLA_VF_MAC */
 947                 if (!is_zero_ether_addr(prev_pp.vf_mac))
 948                         memcpy(pp->mac_addr, prev_pp.vf_mac, ETH_ALEN);
 949
 950                 if (is_zero_ether_addr(netdev->dev_addr))
 951                         eth_hw_addr_random(netdev);
 952         } else {
 953                 /* SR-IOV VF: get mac from adapter */
 954                 ENIC_DEVCMD_PROXY_BY_INDEX(vf, err, enic,
 955                         vnic_dev_get_mac_addr, pp->mac_addr);
 956                 if (err) {
 957                         netdev_err(netdev, "Error getting mac for vf %d\n", vf);
 958                         memcpy(pp, &prev_pp, sizeof(*pp));
 959                         return enic_dev_status_to_errno(err);
 960                 }
 961         }
 962
 963         err = enic_process_set_pp_request(enic, vf, &prev_pp, &restore_pp);
 964         if (err) {
 965                 if (restore_pp) {
 966                         /* Things are still the way they were: Implicit
 967                          * DISASSOCIATE failed
 968                          */
 969                         memcpy(pp, &prev_pp, sizeof(*pp));
 970                 } else {
 971                         memset(pp, 0, sizeof(*pp));
 972                         if (vf == PORT_SELF_VF)
 973                                 eth_zero_addr(netdev->dev_addr);
 974                 }
 975         } else {
 976                 /* Set flag to indicate that the port assoc/disassoc
 977                  * request has been sent out to fw
 978                  */
 979                 pp->set |= ENIC_PORT_REQUEST_APPLIED;
 980
 981                 /* If DISASSOCIATE, clean up all assigned/saved macaddresses */
 982                 if (pp->request == PORT_REQUEST_DISASSOCIATE) {
 983                         eth_zero_addr(pp->mac_addr);
 984                         if (vf == PORT_SELF_VF)
 985                                 eth_zero_addr(netdev->dev_addr);
 986                 }
 987         }
 988
 989         if (vf == PORT_SELF_VF)
 990                 eth_zero_addr(pp->vf_mac);
 991
 992         return err;
 993 }
 994
 995 static int enic_get_vf_port(struct net_device *netdev, int vf,
 996         struct sk_buff *skb)
 997 {
 998         struct enic *enic = netdev_priv(netdev);
 999         u16 response = PORT_PROFILE_RESPONSE_SUCCESS;
1000         struct enic_port_profile *pp;
1001         int err;
1002
1003         ENIC_PP_BY_INDEX(enic, vf, pp, &err);
1004         if (err)
1005                 return err;
1006
1007         if (!(pp->set & ENIC_PORT_REQUEST_APPLIED))
1008                 return -ENODATA;
1009
1010         err = enic_process_get_pp_request(enic, vf, pp->request, &response);
1011         if (err)
1012                 return err;
1013
1014         if (nla_put_u16(skb, IFLA_PORT_REQUEST, pp->request) ||
1015             nla_put_u16(skb, IFLA_PORT_RESPONSE, response) ||
1016             ((pp->set & ENIC_SET_NAME) &&
1017              nla_put(skb, IFLA_PORT_PROFILE, PORT_PROFILE_MAX, pp->name)) ||
1018             ((pp->set & ENIC_SET_INSTANCE) &&
1019              nla_put(skb, IFLA_PORT_INSTANCE_UUID, PORT_UUID_MAX,
1020                      pp->instance_uuid)) ||
1021             ((pp->set & ENIC_SET_HOST) &&
1022              nla_put(skb, IFLA_PORT_HOST_UUID, PORT_UUID_MAX, pp->host_uuid)))
1023                 goto nla_put_failure;
1024         return 0;
1025
1026 nla_put_failure:
1027         return -EMSGSIZE;
1028 }
1029
1030 static void enic_free_rq_buf(struct vnic_rq *rq, struct vnic_rq_buf *buf)
1031 {
1032         struct enic *enic = vnic_dev_priv(rq->vdev);
1033
1034         if (!buf->os_buf)
1035                 return;
1036
1037         pci_unmap_single(enic->pdev, buf->dma_addr,
1038                 buf->len, PCI_DMA_FROMDEVICE);
1039         dev_kfree_skb_any(buf->os_buf);
1040         buf->os_buf = NULL;
1041 }
1042
1043 static int enic_rq_alloc_buf(struct vnic_rq *rq)
1044 {
1045         struct enic *enic = vnic_dev_priv(rq->vdev);
1046         struct net_device *netdev = enic->netdev;
1047         struct sk_buff *skb;
1048         unsigned int len = netdev->mtu + VLAN_ETH_HLEN;
1049         unsigned int os_buf_index = 0;
1050         dma_addr_t dma_addr;
1051         struct vnic_rq_buf *buf = rq->to_use;
1052
1053         if (buf->os_buf) {
1054                 enic_queue_rq_desc(rq, buf->os_buf, os_buf_index, buf->dma_addr,
1055                                    buf->len);
1056
1057                 return 0;
1058         }
1059         skb = netdev_alloc_skb_ip_align(netdev, len);
1060         if (!skb)
1061                 return -ENOMEM;
1062
1063         dma_addr = pci_map_single(enic->pdev, skb->data, len,
1064                                   PCI_DMA_FROMDEVICE);
1065         if (unlikely(enic_dma_map_check(enic, dma_addr))) {
1066                 dev_kfree_skb(skb);
1067                 return -ENOMEM;
1068         }
1069
1070         enic_queue_rq_desc(rq, skb, os_buf_index,
1071                 dma_addr, len);
1072
1073         return 0;
1074 }
1075
1076 static void enic_intr_update_pkt_size(struct vnic_rx_bytes_counter *pkt_size,
1077                                       u32 pkt_len)
1078 {
1079         if (ENIC_LARGE_PKT_THRESHOLD <= pkt_len)
1080                 pkt_size->large_pkt_bytes_cnt += pkt_len;
1081         else
1082                 pkt_size->small_pkt_bytes_cnt += pkt_len;
1083 }
1084
1085 static bool enic_rxcopybreak(struct net_device *netdev, struct sk_buff **skb,
1086                              struct vnic_rq_buf *buf, u16 len)
1087 {
1088         struct enic *enic = netdev_priv(netdev);
1089         struct sk_buff *new_skb;
1090
1091         if (len > enic->rx_copybreak)
1092                 return false;
1093         new_skb = netdev_alloc_skb_ip_align(netdev, len);
1094         if (!new_skb)
1095                 return false;
1096         pci_dma_sync_single_for_cpu(enic->pdev, buf->dma_addr, len,
1097                                     DMA_FROM_DEVICE);
1098         memcpy(new_skb->data, (*skb)->data, len);
1099         *skb = new_skb;
1100
1101         return true;
1102 }
1103
1104 static void enic_rq_indicate_buf(struct vnic_rq *rq,
1105         struct cq_desc *cq_desc, struct vnic_rq_buf *buf,
1106         int skipped, void *opaque)
1107 {
1108         struct enic *enic = vnic_dev_priv(rq->vdev);
1109         struct net_device *netdev = enic->netdev;
1110         struct sk_buff *skb;
1111         struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
1112
1113         u8 type, color, eop, sop, ingress_port, vlan_stripped;
1114         u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof;
1115         u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok;
1116         u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc;
1117         u8 packet_error;
1118         u16 q_number, completed_index, bytes_written, vlan_tci, checksum;
1119         u32 rss_hash;
1120
1121         if (skipped)
1122                 return;
1123
1124         skb = buf->os_buf;
1125
1126         cq_enet_rq_desc_dec((struct cq_enet_rq_desc *)cq_desc,
1127                 &type, &color, &q_number, &completed_index,
1128                 &ingress_port, &fcoe, &eop, &sop, &rss_type,
1129                 &csum_not_calc, &rss_hash, &bytes_written,
1130                 &packet_error, &vlan_stripped, &vlan_tci, &checksum,
1131                 &fcoe_sof, &fcoe_fc_crc_ok, &fcoe_enc_error,
1132                 &fcoe_eof, &tcp_udp_csum_ok, &udp, &tcp,
1133                 &ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment,
1134                 &fcs_ok);
1135
1136         if (packet_error) {
1137
1138                 if (!fcs_ok) {
1139                         if (bytes_written > 0)
1140                                 enic->rq_bad_fcs++;
1141                         else if (bytes_written == 0)
1142                                 enic->rq_truncated_pkts++;
1143                 }
1144
1145                 pci_unmap_single(enic->pdev, buf->dma_addr, buf->len,
1146                                  PCI_DMA_FROMDEVICE);
1147                 dev_kfree_skb_any(skb);
1148                 buf->os_buf = NULL;
1149
1150                 return;
1151         }
1152
1153         if (eop && bytes_written > 0) {
1154
1155                 /* Good receive
1156                  */
1157
1158                 if (!enic_rxcopybreak(netdev, &skb, buf, bytes_written)) {
1159                         buf->os_buf = NULL;
1160                         pci_unmap_single(enic->pdev, buf->dma_addr, buf->len,
1161                                          PCI_DMA_FROMDEVICE);
1162                 }
1163                 prefetch(skb->data - NET_IP_ALIGN);
1164
1165                 skb_put(skb, bytes_written);
1166                 skb->protocol = eth_type_trans(skb, netdev);
1167                 skb_record_rx_queue(skb, q_number);
1168                 if (netdev->features & NETIF_F_RXHASH) {
1169                         skb_set_hash(skb, rss_hash,
1170                                      (rss_type &
1171                                       (NIC_CFG_RSS_HASH_TYPE_TCP_IPV6_EX |
1172                                        NIC_CFG_RSS_HASH_TYPE_TCP_IPV6 |
1173                                        NIC_CFG_RSS_HASH_TYPE_TCP_IPV4)) ?
1174                                      PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3);
1175                 }
1176
1177                 /* Hardware does not provide whole packet checksum. It only
1178                  * provides pseudo checksum. Since hw validates the packet
1179                  * checksum but not provide us the checksum value. use
1180                  * CHECSUM_UNNECESSARY.
1181                  */
1182                 if ((netdev->features & NETIF_F_RXCSUM) && tcp_udp_csum_ok &&
1183                     (ipv4_csum_ok || ipv6))
1184                         skb->ip_summed = CHECKSUM_UNNECESSARY;
1185
1186                 if (vlan_stripped)
1187                         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tci);
1188
1189                 skb_mark_napi_id(skb, &enic->napi[rq->index]);
1190                 if (enic_poll_busy_polling(rq) ||
1191                     !(netdev->features & NETIF_F_GRO))
1192                         netif_receive_skb(skb);
1193                 else
1194                         napi_gro_receive(&enic->napi[q_number], skb);
1195                 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1196                         enic_intr_update_pkt_size(&cq->pkt_size_counter,
1197                                                   bytes_written);
1198         } else {
1199
1200                 /* Buffer overflow
1201                  */
1202
1203                 pci_unmap_single(enic->pdev, buf->dma_addr, buf->len,
1204                                  PCI_DMA_FROMDEVICE);
1205                 dev_kfree_skb_any(skb);
1206                 buf->os_buf = NULL;
1207         }
1208 }
1209
1210 static int enic_rq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
1211         u8 type, u16 q_number, u16 completed_index, void *opaque)
1212 {
1213         struct enic *enic = vnic_dev_priv(vdev);
1214
1215         vnic_rq_service(&enic->rq[q_number], cq_desc,
1216                 completed_index, VNIC_RQ_RETURN_DESC,
1217                 enic_rq_indicate_buf, opaque);
1218
1219         return 0;
1220 }
1221
1222 static void enic_set_int_moderation(struct enic *enic, struct vnic_rq *rq)
1223 {
1224         unsigned int intr = enic_msix_rq_intr(enic, rq->index);
1225         struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
1226         u32 timer = cq->tobe_rx_coal_timeval;
1227
1228         if (cq->tobe_rx_coal_timeval != cq->cur_rx_coal_timeval) {
1229                 vnic_intr_coalescing_timer_set(&enic->intr[intr], timer);
1230                 cq->cur_rx_coal_timeval = cq->tobe_rx_coal_timeval;
1231         }
1232 }
1233
1234 static void enic_calc_int_moderation(struct enic *enic, struct vnic_rq *rq)
1235 {
1236         struct enic_rx_coal *rx_coal = &enic->rx_coalesce_setting;
1237         struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
1238         struct vnic_rx_bytes_counter *pkt_size_counter = &cq->pkt_size_counter;
1239         int index;
1240         u32 timer;
1241         u32 range_start;
1242         u32 traffic;
1243         u64 delta;
1244         ktime_t now = ktime_get();
1245
1246         delta = ktime_us_delta(now, cq->prev_ts);
1247         if (delta < ENIC_AIC_TS_BREAK)
1248                 return;
1249         cq->prev_ts = now;
1250
1251         traffic = pkt_size_counter->large_pkt_bytes_cnt +
1252                   pkt_size_counter->small_pkt_bytes_cnt;
1253         /* The table takes Mbps
1254          * traffic *= 8    => bits
1255          * traffic *= (10^6 / delta)    => bps
1256          * traffic /= 10^6     => Mbps
1257          *
1258          * Combining, traffic *= (8 / delta)
1259          */
1260
1261         traffic <<= 3;
1262         traffic = delta > UINT_MAX ? 0 : traffic / (u32)delta;
1263
1264         for (index = 0; index < ENIC_MAX_COALESCE_TIMERS; index++)
1265                 if (traffic < mod_table[index].rx_rate)
1266                         break;
1267         range_start = (pkt_size_counter->small_pkt_bytes_cnt >
1268                        pkt_size_counter->large_pkt_bytes_cnt << 1) ?
1269                       rx_coal->small_pkt_range_start :
1270                       rx_coal->large_pkt_range_start;
1271         timer = range_start + ((rx_coal->range_end - range_start) *
1272                                mod_table[index].range_percent / 100);
1273         /* Damping */
1274         cq->tobe_rx_coal_timeval = (timer + cq->tobe_rx_coal_timeval) >> 1;
1275
1276         pkt_size_counter->large_pkt_bytes_cnt = 0;
1277         pkt_size_counter->small_pkt_bytes_cnt = 0;
1278 }
1279
1280 static int enic_poll(struct napi_struct *napi, int budget)
1281 {
1282         struct net_device *netdev = napi->dev;
1283         struct enic *enic = netdev_priv(netdev);
1284         unsigned int cq_rq = enic_cq_rq(enic, 0);
1285         unsigned int cq_wq = enic_cq_wq(enic, 0);
1286         unsigned int intr = enic_legacy_io_intr();
1287         unsigned int rq_work_to_do = budget;
1288         unsigned int wq_work_to_do = -1; /* no limit */
1289         unsigned int  work_done, rq_work_done = 0, wq_work_done;
1290         int err;
1291
1292         wq_work_done = vnic_cq_service(&enic->cq[cq_wq], wq_work_to_do,
1293                                        enic_wq_service, NULL);
1294
1295         if (!enic_poll_lock_napi(&enic->rq[cq_rq])) {
1296                 if (wq_work_done > 0)
1297                         vnic_intr_return_credits(&enic->intr[intr],
1298                                                  wq_work_done,
1299                                                  0 /* dont unmask intr */,
1300                                                  0 /* dont reset intr timer */);
1301                 return budget;
1302         }
1303
1304         if (budget > 0)
1305                 rq_work_done = vnic_cq_service(&enic->cq[cq_rq],
1306                         rq_work_to_do, enic_rq_service, NULL);
1307
1308         /* Accumulate intr event credits for this polling
1309          * cycle.  An intr event is the completion of a
1310          * a WQ or RQ packet.
1311          */
1312
1313         work_done = rq_work_done + wq_work_done;
1314
1315         if (work_done > 0)
1316                 vnic_intr_return_credits(&enic->intr[intr],
1317                         work_done,
1318                         0 /* don't unmask intr */,
1319                         0 /* don't reset intr timer */);
1320
1321         err = vnic_rq_fill(&enic->rq[0], enic_rq_alloc_buf);
1322         enic_poll_unlock_napi(&enic->rq[cq_rq], napi);
1323
1324         /* Buffer allocation failed. Stay in polling
1325          * mode so we can try to fill the ring again.
1326          */
1327
1328         if (err)
1329                 rq_work_done = rq_work_to_do;
1330         if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1331                 /* Call the function which refreshes the intr coalescing timer
1332                  * value based on the traffic.
1333                  */
1334                 enic_calc_int_moderation(enic, &enic->rq[0]);
1335
1336         if (rq_work_done < rq_work_to_do) {
1337
1338                 /* Some work done, but not enough to stay in polling,
1339                  * exit polling
1340                  */
1341
1342                 napi_complete(napi);
1343                 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1344                         enic_set_int_moderation(enic, &enic->rq[0]);
1345                 vnic_intr_unmask(&enic->intr[intr]);
1346         }
1347
1348         return rq_work_done;
1349 }
1350
1351 #ifdef CONFIG_RFS_ACCEL
1352 static void enic_free_rx_cpu_rmap(struct enic *enic)
1353 {
1354         free_irq_cpu_rmap(enic->netdev->rx_cpu_rmap);
1355         enic->netdev->rx_cpu_rmap = NULL;
1356 }
1357
1358 static void enic_set_rx_cpu_rmap(struct enic *enic)
1359 {
1360         int i, res;
1361
1362         if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX) {
1363                 enic->netdev->rx_cpu_rmap = alloc_irq_cpu_rmap(enic->rq_count);
1364                 if (unlikely(!enic->netdev->rx_cpu_rmap))
1365                         return;
1366                 for (i = 0; i < enic->rq_count; i++) {
1367                         res = irq_cpu_rmap_add(enic->netdev->rx_cpu_rmap,
1368                                                enic->msix_entry[i].vector);
1369                         if (unlikely(res)) {
1370                                 enic_free_rx_cpu_rmap(enic);
1371                                 return;
1372                         }
1373                 }
1374         }
1375 }
1376
1377 #else
1378
1379 static void enic_free_rx_cpu_rmap(struct enic *enic)
1380 {
1381 }
1382
1383 static void enic_set_rx_cpu_rmap(struct enic *enic)
1384 {
1385 }
1386
1387 #endif /* CONFIG_RFS_ACCEL */
1388
1389 #ifdef CONFIG_NET_RX_BUSY_POLL
1390 static int enic_busy_poll(struct napi_struct *napi)
1391 {
1392         struct net_device *netdev = napi->dev;
1393         struct enic *enic = netdev_priv(netdev);
1394         unsigned int rq = (napi - &enic->napi[0]);
1395         unsigned int cq = enic_cq_rq(enic, rq);
1396         unsigned int intr = enic_msix_rq_intr(enic, rq);
1397         unsigned int work_to_do = -1; /* clean all pkts possible */
1398         unsigned int work_done;
1399
1400         if (!enic_poll_lock_poll(&enic->rq[rq]))
1401                 return LL_FLUSH_BUSY;
1402         work_done = vnic_cq_service(&enic->cq[cq], work_to_do,
1403                                     enic_rq_service, NULL);
1404
1405         if (work_done > 0)
1406                 vnic_intr_return_credits(&enic->intr[intr],
1407                                          work_done, 0, 0);
1408         vnic_rq_fill(&enic->rq[rq], enic_rq_alloc_buf);
1409         if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1410                 enic_calc_int_moderation(enic, &enic->rq[rq]);
1411         enic_poll_unlock_poll(&enic->rq[rq]);
1412
1413         return work_done;
1414 }
1415 #endif /* CONFIG_NET_RX_BUSY_POLL */
1416
1417 static int enic_poll_msix_wq(struct napi_struct *napi, int budget)
1418 {
1419         struct net_device *netdev = napi->dev;
1420         struct enic *enic = netdev_priv(netdev);
1421         unsigned int wq_index = (napi - &enic->napi[0]) - enic->rq_count;
1422         struct vnic_wq *wq = &enic->wq[wq_index];
1423         unsigned int cq;
1424         unsigned int intr;
1425         unsigned int wq_work_to_do = -1; /* clean all desc possible */
1426         unsigned int wq_work_done;
1427         unsigned int wq_irq;
1428
1429         wq_irq = wq->index;
1430         cq = enic_cq_wq(enic, wq_irq);
1431         intr = enic_msix_wq_intr(enic, wq_irq);
1432         wq_work_done = vnic_cq_service(&enic->cq[cq], wq_work_to_do,
1433                                        enic_wq_service, NULL);
1434
1435         vnic_intr_return_credits(&enic->intr[intr], wq_work_done,
1436                                  0 /* don't unmask intr */,
1437                                  1 /* reset intr timer */);
1438         if (!wq_work_done) {
1439                 napi_complete(napi);
1440                 vnic_intr_unmask(&enic->intr[intr]);
1441                 return 0;
1442         }
1443
1444         return budget;
1445 }
1446
1447 static int enic_poll_msix_rq(struct napi_struct *napi, int budget)
1448 {
1449         struct net_device *netdev = napi->dev;
1450         struct enic *enic = netdev_priv(netdev);
1451         unsigned int rq = (napi - &enic->napi[0]);
1452         unsigned int cq = enic_cq_rq(enic, rq);
1453         unsigned int intr = enic_msix_rq_intr(enic, rq);
1454         unsigned int work_to_do = budget;
1455         unsigned int work_done = 0;
1456         int err;
1457
1458         if (!enic_poll_lock_napi(&enic->rq[rq]))
1459                 return budget;
1460         /* Service RQ
1461          */
1462
1463         if (budget > 0)
1464                 work_done = vnic_cq_service(&enic->cq[cq],
1465                         work_to_do, enic_rq_service, NULL);
1466
1467         /* Return intr event credits for this polling
1468          * cycle.  An intr event is the completion of a
1469          * RQ packet.
1470          */
1471
1472         if (work_done > 0)
1473                 vnic_intr_return_credits(&enic->intr[intr],
1474                         work_done,
1475                         0 /* don't unmask intr */,
1476                         0 /* don't reset intr timer */);
1477
1478         err = vnic_rq_fill(&enic->rq[rq], enic_rq_alloc_buf);
1479
1480         /* Buffer allocation failed. Stay in polling mode
1481          * so we can try to fill the ring again.
1482          */
1483
1484         if (err)
1485                 work_done = work_to_do;
1486         if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1487                 /* Call the function which refreshes the intr coalescing timer
1488                  * value based on the traffic.
1489                  */
1490                 enic_calc_int_moderation(enic, &enic->rq[rq]);
1491
1492         enic_poll_unlock_napi(&enic->rq[rq], napi);
1493         if (work_done < work_to_do) {
1494
1495                 /* Some work done, but not enough to stay in polling,
1496                  * exit polling
1497                  */
1498
1499                 napi_complete(napi);
1500                 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1501                         enic_set_int_moderation(enic, &enic->rq[rq]);
1502                 vnic_intr_unmask(&enic->intr[intr]);
1503         }
1504
1505         return work_done;
1506 }
1507
1508 static void enic_notify_timer(unsigned long data)
1509 {
1510         struct enic *enic = (struct enic *)data;
1511
1512         enic_notify_check(enic);
1513
1514         mod_timer(&enic->notify_timer,
1515                 round_jiffies(jiffies + ENIC_NOTIFY_TIMER_PERIOD));
1516 }
1517
1518 static void enic_free_intr(struct enic *enic)
1519 {
1520         struct net_device *netdev = enic->netdev;
1521         unsigned int i;
1522
1523         enic_free_rx_cpu_rmap(enic);
1524         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1525         case VNIC_DEV_INTR_MODE_INTX:
1526                 free_irq(enic->pdev->irq, netdev);
1527                 break;
1528         case VNIC_DEV_INTR_MODE_MSI:
1529                 free_irq(enic->pdev->irq, enic);
1530                 break;
1531         case VNIC_DEV_INTR_MODE_MSIX:
1532                 for (i = 0; i < ARRAY_SIZE(enic->msix); i++)
1533                         if (enic->msix[i].requested)
1534                                 free_irq(enic->msix_entry[i].vector,
1535                                         enic->msix[i].devid);
1536                 break;
1537         default:
1538                 break;
1539         }
1540 }
1541
1542 static int enic_request_intr(struct enic *enic)
1543 {
1544         struct net_device *netdev = enic->netdev;
1545         unsigned int i, intr;
1546         int err = 0;
1547
1548         enic_set_rx_cpu_rmap(enic);
1549         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1550
1551         case VNIC_DEV_INTR_MODE_INTX:
1552
1553                 err = request_irq(enic->pdev->irq, enic_isr_legacy,
1554                         IRQF_SHARED, netdev->name, netdev);
1555                 break;
1556
1557         case VNIC_DEV_INTR_MODE_MSI:
1558
1559                 err = request_irq(enic->pdev->irq, enic_isr_msi,
1560                         0, netdev->name, enic);
1561                 break;
1562
1563         case VNIC_DEV_INTR_MODE_MSIX:
1564
1565                 for (i = 0; i < enic->rq_count; i++) {
1566                         intr = enic_msix_rq_intr(enic, i);
1567                         snprintf(enic->msix[intr].devname,
1568                                 sizeof(enic->msix[intr].devname),
1569                                 "%.11s-rx-%d", netdev->name, i);
1570                         enic->msix[intr].isr = enic_isr_msix;
1571                         enic->msix[intr].devid = &enic->napi[i];
1572                 }
1573
1574                 for (i = 0; i < enic->wq_count; i++) {
1575                         int wq = enic_cq_wq(enic, i);
1576
1577                         intr = enic_msix_wq_intr(enic, i);
1578                         snprintf(enic->msix[intr].devname,
1579                                 sizeof(enic->msix[intr].devname),
1580                                 "%.11s-tx-%d", netdev->name, i);
1581                         enic->msix[intr].isr = enic_isr_msix;
1582                         enic->msix[intr].devid = &enic->napi[wq];
1583                 }
1584
1585                 intr = enic_msix_err_intr(enic);
1586                 snprintf(enic->msix[intr].devname,
1587                         sizeof(enic->msix[intr].devname),
1588                         "%.11s-err", netdev->name);
1589                 enic->msix[intr].isr = enic_isr_msix_err;
1590                 enic->msix[intr].devid = enic;
1591
1592                 intr = enic_msix_notify_intr(enic);
1593                 snprintf(enic->msix[intr].devname,
1594                         sizeof(enic->msix[intr].devname),
1595                         "%.11s-notify", netdev->name);
1596                 enic->msix[intr].isr = enic_isr_msix_notify;
1597                 enic->msix[intr].devid = enic;
1598
1599                 for (i = 0; i < ARRAY_SIZE(enic->msix); i++)
1600                         enic->msix[i].requested = 0;
1601
1602                 for (i = 0; i < enic->intr_count; i++) {
1603                         err = request_irq(enic->msix_entry[i].vector,
1604                                 enic->msix[i].isr, 0,
1605                                 enic->msix[i].devname,
1606                                 enic->msix[i].devid);
1607                         if (err) {
1608                                 enic_free_intr(enic);
1609                                 break;
1610                         }
1611                         enic->msix[i].requested = 1;
1612                 }
1613
1614                 break;
1615
1616         default:
1617                 break;
1618         }
1619
1620         return err;
1621 }
1622
1623 static void enic_synchronize_irqs(struct enic *enic)
1624 {
1625         unsigned int i;
1626
1627         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1628         case VNIC_DEV_INTR_MODE_INTX:
1629         case VNIC_DEV_INTR_MODE_MSI:
1630                 synchronize_irq(enic->pdev->irq);
1631                 break;
1632         case VNIC_DEV_INTR_MODE_MSIX:
1633                 for (i = 0; i < enic->intr_count; i++)
1634                         synchronize_irq(enic->msix_entry[i].vector);
1635                 break;
1636         default:
1637                 break;
1638         }
1639 }
1640
1641 static void enic_set_rx_coal_setting(struct enic *enic)
1642 {
1643         unsigned int speed;
1644         int index = -1;
1645         struct enic_rx_coal *rx_coal = &enic->rx_coalesce_setting;
1646
1647         /* 1. Read the link speed from fw
1648          * 2. Pick the default range for the speed
1649          * 3. Update it in enic->rx_coalesce_setting
1650          */
1651         speed = vnic_dev_port_speed(enic->vdev);
1652         if (ENIC_LINK_SPEED_10G < speed)
1653                 index = ENIC_LINK_40G_INDEX;
1654         else if (ENIC_LINK_SPEED_4G < speed)
1655                 index = ENIC_LINK_10G_INDEX;
1656         else
1657                 index = ENIC_LINK_4G_INDEX;
1658
1659         rx_coal->small_pkt_range_start = mod_range[index].small_pkt_range_start;
1660         rx_coal->large_pkt_range_start = mod_range[index].large_pkt_range_start;
1661         rx_coal->range_end = ENIC_RX_COALESCE_RANGE_END;
1662
1663         /* Start with the value provided by UCSM */
1664         for (index = 0; index < enic->rq_count; index++)
1665                 enic->cq[index].cur_rx_coal_timeval =
1666                                 enic->config.intr_timer_usec;
1667
1668         rx_coal->use_adaptive_rx_coalesce = 1;
1669 }
1670
1671 static int enic_dev_notify_set(struct enic *enic)
1672 {
1673         int err;
1674
1675         spin_lock_bh(&enic->devcmd_lock);
1676         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1677         case VNIC_DEV_INTR_MODE_INTX:
1678                 err = vnic_dev_notify_set(enic->vdev,
1679                         enic_legacy_notify_intr());
1680                 break;
1681         case VNIC_DEV_INTR_MODE_MSIX:
1682                 err = vnic_dev_notify_set(enic->vdev,
1683                         enic_msix_notify_intr(enic));
1684                 break;
1685         default:
1686                 err = vnic_dev_notify_set(enic->vdev, -1 /* no intr */);
1687                 break;
1688         }
1689         spin_unlock_bh(&enic->devcmd_lock);
1690
1691         return err;
1692 }
1693
1694 static void enic_notify_timer_start(struct enic *enic)
1695 {
1696         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1697         case VNIC_DEV_INTR_MODE_MSI:
1698                 mod_timer(&enic->notify_timer, jiffies);
1699                 break;
1700         default:
1701                 /* Using intr for notification for INTx/MSI-X */
1702                 break;
1703         }
1704 }
1705
1706 /* rtnl lock is held, process context */
1707 static int enic_open(struct net_device *netdev)
1708 {
1709         struct enic *enic = netdev_priv(netdev);
1710         unsigned int i;
1711         int err, ret;
1712
1713         err = enic_request_intr(enic);
1714         if (err) {
1715                 netdev_err(netdev, "Unable to request irq.\n");
1716                 return err;
1717         }
1718         enic_init_affinity_hint(enic);
1719         enic_set_affinity_hint(enic);
1720
1721         err = enic_dev_notify_set(enic);
1722         if (err) {
1723                 netdev_err(netdev,
1724                         "Failed to alloc notify buffer, aborting.\n");
1725                 goto err_out_free_intr;
1726         }
1727
1728         for (i = 0; i < enic->rq_count; i++) {
1729                 /* enable rq before updating rq desc */
1730                 vnic_rq_enable(&enic->rq[i]);
1731                 vnic_rq_fill(&enic->rq[i], enic_rq_alloc_buf);
1732                 /* Need at least one buffer on ring to get going */
1733                 if (vnic_rq_desc_used(&enic->rq[i]) == 0) {
1734                         netdev_err(netdev, "Unable to alloc receive buffers\n");
1735                         err = -ENOMEM;
1736                         goto err_out_free_rq;
1737                 }
1738         }
1739
1740         for (i = 0; i < enic->wq_count; i++)
1741                 vnic_wq_enable(&enic->wq[i]);
1742
1743         if (!enic_is_dynamic(enic) && !enic_is_sriov_vf(enic))
1744                 enic_dev_add_station_addr(enic);
1745
1746         enic_set_rx_mode(netdev);
1747
1748         netif_tx_wake_all_queues(netdev);
1749
1750         for (i = 0; i < enic->rq_count; i++) {
1751                 enic_busy_poll_init_lock(&enic->rq[i]);
1752                 napi_enable(&enic->napi[i]);
1753         }
1754         if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX)
1755                 for (i = 0; i < enic->wq_count; i++)
1756                         napi_enable(&enic->napi[enic_cq_wq(enic, i)]);
1757         enic_dev_enable(enic);
1758
1759         for (i = 0; i < enic->intr_count; i++)
1760                 vnic_intr_unmask(&enic->intr[i]);
1761
1762         enic_notify_timer_start(enic);
1763         enic_rfs_timer_start(enic);
1764
1765         return 0;
1766
1767 err_out_free_rq:
1768         for (i = 0; i < enic->rq_count; i++) {
1769                 ret = vnic_rq_disable(&enic->rq[i]);
1770                 if (!ret)
1771                         vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
1772         }
1773         enic_dev_notify_unset(enic);
1774 err_out_free_intr:
1775         enic_unset_affinity_hint(enic);
1776         enic_free_intr(enic);
1777
1778         return err;
1779 }
1780
1781 /* rtnl lock is held, process context */
1782 static int enic_stop(struct net_device *netdev)
1783 {
1784         struct enic *enic = netdev_priv(netdev);
1785         unsigned int i;
1786         int err;
1787
1788         for (i = 0; i < enic->intr_count; i++) {
1789                 vnic_intr_mask(&enic->intr[i]);
1790                 (void)vnic_intr_masked(&enic->intr[i]); /* flush write */
1791         }
1792
1793         enic_synchronize_irqs(enic);
1794
1795         del_timer_sync(&enic->notify_timer);
1796         enic_rfs_flw_tbl_free(enic);
1797
1798         enic_dev_disable(enic);
1799
1800         for (i = 0; i < enic->rq_count; i++) {
1801                 napi_disable(&enic->napi[i]);
1802                 local_bh_disable();
1803                 while (!enic_poll_lock_napi(&enic->rq[i]))
1804                         mdelay(1);
1805                 local_bh_enable();
1806         }
1807
1808         netif_carrier_off(netdev);
1809         if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX)
1810                 for (i = 0; i < enic->wq_count; i++)
1811                         napi_disable(&enic->napi[enic_cq_wq(enic, i)]);
1812         netif_tx_disable(netdev);
1813
1814         if (!enic_is_dynamic(enic) && !enic_is_sriov_vf(enic))
1815                 enic_dev_del_station_addr(enic);
1816
1817         for (i = 0; i < enic->wq_count; i++) {
1818                 err = vnic_wq_disable(&enic->wq[i]);
1819                 if (err)
1820                         return err;
1821         }
1822         for (i = 0; i < enic->rq_count; i++) {
1823                 err = vnic_rq_disable(&enic->rq[i]);
1824                 if (err)
1825                         return err;
1826         }
1827
1828         enic_dev_notify_unset(enic);
1829         enic_unset_affinity_hint(enic);
1830         enic_free_intr(enic);
1831
1832         for (i = 0; i < enic->wq_count; i++)
1833                 vnic_wq_clean(&enic->wq[i], enic_free_wq_buf);
1834         for (i = 0; i < enic->rq_count; i++)
1835                 vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
1836         for (i = 0; i < enic->cq_count; i++)
1837                 vnic_cq_clean(&enic->cq[i]);
1838         for (i = 0; i < enic->intr_count; i++)
1839                 vnic_intr_clean(&enic->intr[i]);
1840
1841         return 0;
1842 }
1843
1844 static int _enic_change_mtu(struct net_device *netdev, int new_mtu)
1845 {
1846         bool running = netif_running(netdev);
1847         int err = 0;
1848
1849         ASSERT_RTNL();
1850         if (running) {
1851                 err = enic_stop(netdev);
1852                 if (err)
1853                         return err;
1854         }
1855
1856         netdev->mtu = new_mtu;
1857
1858         if (running) {
1859                 err = enic_open(netdev);
1860                 if (err)
1861                         return err;
1862         }
1863
1864         return 0;
1865 }
1866
1867 static int enic_change_mtu(struct net_device *netdev, int new_mtu)
1868 {
1869         struct enic *enic = netdev_priv(netdev);
1870
1871         if (new_mtu < ENIC_MIN_MTU || new_mtu > ENIC_MAX_MTU)
1872                 return -EINVAL;
1873
1874         if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic))
1875                 return -EOPNOTSUPP;
1876
1877         if (netdev->mtu > enic->port_mtu)
1878                 netdev_warn(netdev,
1879                             "interface MTU (%d) set higher than port MTU (%d)\n",
1880                             netdev->mtu, enic->port_mtu);
1881
1882         return _enic_change_mtu(netdev, new_mtu);
1883 }
1884
1885 static void enic_change_mtu_work(struct work_struct *work)
1886 {
1887         struct enic *enic = container_of(work, struct enic, change_mtu_work);
1888         struct net_device *netdev = enic->netdev;
1889         int new_mtu = vnic_dev_mtu(enic->vdev);
1890
1891         rtnl_lock();
1892         (void)_enic_change_mtu(netdev, new_mtu);
1893         rtnl_unlock();
1894
1895         netdev_info(netdev, "interface MTU set as %d\n", netdev->mtu);
1896 }
1897
1898 #ifdef CONFIG_NET_POLL_CONTROLLER
1899 static void enic_poll_controller(struct net_device *netdev)
1900 {
1901         struct enic *enic = netdev_priv(netdev);
1902         struct vnic_dev *vdev = enic->vdev;
1903         unsigned int i, intr;
1904
1905         switch (vnic_dev_get_intr_mode(vdev)) {
1906         case VNIC_DEV_INTR_MODE_MSIX:
1907                 for (i = 0; i < enic->rq_count; i++) {
1908                         intr = enic_msix_rq_intr(enic, i);
1909                         enic_isr_msix(enic->msix_entry[intr].vector,
1910                                       &enic->napi[i]);
1911                 }
1912
1913                 for (i = 0; i < enic->wq_count; i++) {
1914                         intr = enic_msix_wq_intr(enic, i);
1915                         enic_isr_msix(enic->msix_entry[intr].vector,
1916                                       &enic->napi[enic_cq_wq(enic, i)]);
1917                 }
1918
1919                 break;
1920         case VNIC_DEV_INTR_MODE_MSI:
1921                 enic_isr_msi(enic->pdev->irq, enic);
1922                 break;
1923         case VNIC_DEV_INTR_MODE_INTX:
1924                 enic_isr_legacy(enic->pdev->irq, netdev);
1925                 break;
1926         default:
1927                 break;
1928         }
1929 }
1930 #endif
1931
1932 static int enic_dev_wait(struct vnic_dev *vdev,
1933         int (*start)(struct vnic_dev *, int),
1934         int (*finished)(struct vnic_dev *, int *),
1935         int arg)
1936 {
1937         unsigned long time;
1938         int done;
1939         int err;
1940
1941         err = start(vdev, arg);
1942         if (err)
1943                 return err;
1944
1945         /* Wait for func to complete...2 seconds max
1946          */
1947
1948         time = jiffies + (HZ * 2);
1949         do {
1950
1951                 err = finished(vdev, &done);
1952                 if (err)
1953                         return err;
1954
1955                 if (done)
1956                         return 0;
1957
1958                 schedule_timeout_uninterruptible(HZ / 10);
1959
1960         } while (time_after(time, jiffies));
1961
1962         return -ETIMEDOUT;
1963 }
1964
1965 static int enic_dev_open(struct enic *enic)
1966 {
1967         int err;
1968
1969         err = enic_dev_wait(enic->vdev, vnic_dev_open,
1970                 vnic_dev_open_done, 0);
1971         if (err)
1972                 dev_err(enic_get_dev(enic), "vNIC device open failed, err %d\n",
1973                         err);
1974
1975         return err;
1976 }
1977
1978 static int enic_dev_soft_reset(struct enic *enic)
1979 {
1980         int err;
1981
1982         err = enic_dev_wait(enic->vdev, vnic_dev_soft_reset,
1983                             vnic_dev_soft_reset_done, 0);
1984         if (err)
1985                 netdev_err(enic->netdev, "vNIC soft reset failed, err %d\n",
1986                            err);
1987
1988         return err;
1989 }
1990
1991 static int enic_dev_hang_reset(struct enic *enic)
1992 {
1993         int err;
1994
1995         err = enic_dev_wait(enic->vdev, vnic_dev_hang_reset,
1996                 vnic_dev_hang_reset_done, 0);
1997         if (err)
1998                 netdev_err(enic->netdev, "vNIC hang reset failed, err %d\n",
1999                         err);
2000
2001         return err;
2002 }
2003
2004 int __enic_set_rsskey(struct enic *enic)
2005 {
2006         union vnic_rss_key *rss_key_buf_va;
2007         dma_addr_t rss_key_buf_pa;
2008         int i, kidx, bidx, err;
2009
2010         rss_key_buf_va = pci_zalloc_consistent(enic->pdev,
2011                                                sizeof(union vnic_rss_key),
2012                                                &rss_key_buf_pa);
2013         if (!rss_key_buf_va)
2014                 return -ENOMEM;
2015
2016         for (i = 0; i < ENIC_RSS_LEN; i++) {
2017                 kidx = i / ENIC_RSS_BYTES_PER_KEY;
2018                 bidx = i % ENIC_RSS_BYTES_PER_KEY;
2019                 rss_key_buf_va->key[kidx].b[bidx] = enic->rss_key[i];
2020         }
2021         spin_lock_bh(&enic->devcmd_lock);
2022         err = enic_set_rss_key(enic,
2023                 rss_key_buf_pa,
2024                 sizeof(union vnic_rss_key));
2025         spin_unlock_bh(&enic->devcmd_lock);
2026
2027         pci_free_consistent(enic->pdev, sizeof(union vnic_rss_key),
2028                 rss_key_buf_va, rss_key_buf_pa);
2029
2030         return err;
2031 }
2032
2033 static int enic_set_rsskey(struct enic *enic)
2034 {
2035         netdev_rss_key_fill(enic->rss_key, ENIC_RSS_LEN);
2036
2037         return __enic_set_rsskey(enic);
2038 }
2039
2040 static int enic_set_rsscpu(struct enic *enic, u8 rss_hash_bits)
2041 {
2042         dma_addr_t rss_cpu_buf_pa;
2043         union vnic_rss_cpu *rss_cpu_buf_va = NULL;
2044         unsigned int i;
2045         int err;
2046
2047         rss_cpu_buf_va = pci_alloc_consistent(enic->pdev,
2048                 sizeof(union vnic_rss_cpu), &rss_cpu_buf_pa);
2049         if (!rss_cpu_buf_va)
2050                 return -ENOMEM;
2051
2052         for (i = 0; i < (1 << rss_hash_bits); i++)
2053                 (*rss_cpu_buf_va).cpu[i/4].b[i%4] = i % enic->rq_count;
2054
2055         spin_lock_bh(&enic->devcmd_lock);
2056         err = enic_set_rss_cpu(enic,
2057                 rss_cpu_buf_pa,
2058                 sizeof(union vnic_rss_cpu));
2059         spin_unlock_bh(&enic->devcmd_lock);
2060
2061         pci_free_consistent(enic->pdev, sizeof(union vnic_rss_cpu),
2062                 rss_cpu_buf_va, rss_cpu_buf_pa);
2063
2064         return err;
2065 }
2066
2067 static int enic_set_niccfg(struct enic *enic, u8 rss_default_cpu,
2068         u8 rss_hash_type, u8 rss_hash_bits, u8 rss_base_cpu, u8 rss_enable)
2069 {
2070         const u8 tso_ipid_split_en = 0;
2071         const u8 ig_vlan_strip_en = 1;
2072         int err;
2073
2074         /* Enable VLAN tag stripping.
2075         */
2076
2077         spin_lock_bh(&enic->devcmd_lock);
2078         err = enic_set_nic_cfg(enic,
2079                 rss_default_cpu, rss_hash_type,
2080                 rss_hash_bits, rss_base_cpu,
2081                 rss_enable, tso_ipid_split_en,
2082                 ig_vlan_strip_en);
2083         spin_unlock_bh(&enic->devcmd_lock);
2084
2085         return err;
2086 }
2087
2088 static int enic_set_rss_nic_cfg(struct enic *enic)
2089 {
2090         struct device *dev = enic_get_dev(enic);
2091         const u8 rss_default_cpu = 0;
2092         const u8 rss_hash_type = NIC_CFG_RSS_HASH_TYPE_IPV4 |
2093                 NIC_CFG_RSS_HASH_TYPE_TCP_IPV4 |
2094                 NIC_CFG_RSS_HASH_TYPE_IPV6 |
2095                 NIC_CFG_RSS_HASH_TYPE_TCP_IPV6;
2096         const u8 rss_hash_bits = 7;
2097         const u8 rss_base_cpu = 0;
2098         u8 rss_enable = ENIC_SETTING(enic, RSS) && (enic->rq_count > 1);
2099
2100         if (rss_enable) {
2101                 if (!enic_set_rsskey(enic)) {
2102                         if (enic_set_rsscpu(enic, rss_hash_bits)) {
2103                                 rss_enable = 0;
2104                                 dev_warn(dev, "RSS disabled, "
2105                                         "Failed to set RSS cpu indirection table.");
2106                         }
2107                 } else {
2108                         rss_enable = 0;
2109                         dev_warn(dev, "RSS disabled, Failed to set RSS key.\n");
2110                 }
2111         }
2112
2113         return enic_set_niccfg(enic, rss_default_cpu, rss_hash_type,
2114                 rss_hash_bits, rss_base_cpu, rss_enable);
2115 }
2116
2117 static void enic_set_api_busy(struct enic *enic, bool busy)
2118 {
2119         spin_lock(&enic->enic_api_lock);
2120         enic->enic_api_busy = busy;
2121         spin_unlock(&enic->enic_api_lock);
2122 }
2123
2124 static void enic_reset(struct work_struct *work)
2125 {
2126         struct enic *enic = container_of(work, struct enic, reset);
2127
2128         if (!netif_running(enic->netdev))
2129                 return;
2130
2131         rtnl_lock();
2132
2133         /* Stop any activity from infiniband */
2134         enic_set_api_busy(enic, true);
2135
2136         enic_stop(enic->netdev);
2137         enic_dev_soft_reset(enic);
2138         enic_reset_addr_lists(enic);
2139         enic_init_vnic_resources(enic);
2140         enic_set_rss_nic_cfg(enic);
2141         enic_dev_set_ig_vlan_rewrite_mode(enic);
2142         enic_open(enic->netdev);
2143
2144         /* Allow infiniband to fiddle with the device again */
2145         enic_set_api_busy(enic, false);
2146
2147         call_netdevice_notifiers(NETDEV_REBOOT, enic->netdev);
2148
2149         rtnl_unlock();
2150 }
2151
2152 static void enic_tx_hang_reset(struct work_struct *work)
2153 {
2154         struct enic *enic = container_of(work, struct enic, tx_hang_reset);
2155
2156         rtnl_lock();
2157
2158         /* Stop any activity from infiniband */
2159         enic_set_api_busy(enic, true);
2160
2161         enic_dev_hang_notify(enic);
2162         enic_stop(enic->netdev);
2163         enic_dev_hang_reset(enic);
2164         enic_reset_addr_lists(enic);
2165         enic_init_vnic_resources(enic);
2166         enic_set_rss_nic_cfg(enic);
2167         enic_dev_set_ig_vlan_rewrite_mode(enic);
2168         enic_open(enic->netdev);
2169
2170         /* Allow infiniband to fiddle with the device again */
2171         enic_set_api_busy(enic, false);
2172
2173         call_netdevice_notifiers(NETDEV_REBOOT, enic->netdev);
2174
2175         rtnl_unlock();
2176 }
2177
2178 static int enic_set_intr_mode(struct enic *enic)
2179 {
2180         unsigned int n = min_t(unsigned int, enic->rq_count, ENIC_RQ_MAX);
2181         unsigned int m = min_t(unsigned int, enic->wq_count, ENIC_WQ_MAX);
2182         unsigned int i;
2183
2184         /* Set interrupt mode (INTx, MSI, MSI-X) depending
2185          * on system capabilities.
2186          *
2187          * Try MSI-X first
2188          *
2189          * We need n RQs, m WQs, n+m CQs, and n+m+2 INTRs
2190          * (the second to last INTR is used for WQ/RQ errors)
2191          * (the last INTR is used for notifications)
2192          */
2193
2194         BUG_ON(ARRAY_SIZE(enic->msix_entry) < n + m + 2);
2195         for (i = 0; i < n + m + 2; i++)
2196                 enic->msix_entry[i].entry = i;
2197
2198         /* Use multiple RQs if RSS is enabled
2199          */
2200
2201         if (ENIC_SETTING(enic, RSS) &&
2202             enic->config.intr_mode < 1 &&
2203             enic->rq_count >= n &&
2204             enic->wq_count >= m &&
2205             enic->cq_count >= n + m &&
2206             enic->intr_count >= n + m + 2) {
2207
2208                 if (pci_enable_msix_range(enic->pdev, enic->msix_entry,
2209                                           n + m + 2, n + m + 2) > 0) {
2210
2211                         enic->rq_count = n;
2212                         enic->wq_count = m;
2213                         enic->cq_count = n + m;
2214                         enic->intr_count = n + m + 2;
2215
2216                         vnic_dev_set_intr_mode(enic->vdev,
2217                                 VNIC_DEV_INTR_MODE_MSIX);
2218
2219                         return 0;
2220                 }
2221         }
2222
2223         if (enic->config.intr_mode < 1 &&
2224             enic->rq_count >= 1 &&
2225             enic->wq_count >= m &&
2226             enic->cq_count >= 1 + m &&
2227             enic->intr_count >= 1 + m + 2) {
2228                 if (pci_enable_msix_range(enic->pdev, enic->msix_entry,
2229                                           1 + m + 2, 1 + m + 2) > 0) {
2230
2231                         enic->rq_count = 1;
2232                         enic->wq_count = m;
2233                         enic->cq_count = 1 + m;
2234                         enic->intr_count = 1 + m + 2;
2235
2236                         vnic_dev_set_intr_mode(enic->vdev,
2237                                 VNIC_DEV_INTR_MODE_MSIX);
2238
2239                         return 0;
2240                 }
2241         }
2242
2243         /* Next try MSI
2244          *
2245          * We need 1 RQ, 1 WQ, 2 CQs, and 1 INTR
2246          */
2247
2248         if (enic->config.intr_mode < 2 &&
2249             enic->rq_count >= 1 &&
2250             enic->wq_count >= 1 &&
2251             enic->cq_count >= 2 &&
2252             enic->intr_count >= 1 &&
2253             !pci_enable_msi(enic->pdev)) {
2254
2255                 enic->rq_count = 1;
2256                 enic->wq_count = 1;
2257                 enic->cq_count = 2;
2258                 enic->intr_count = 1;
2259
2260                 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSI);
2261
2262                 return 0;
2263         }
2264
2265         /* Next try INTx
2266          *
2267          * We need 1 RQ, 1 WQ, 2 CQs, and 3 INTRs
2268          * (the first INTR is used for WQ/RQ)
2269          * (the second INTR is used for WQ/RQ errors)
2270          * (the last INTR is used for notifications)
2271          */
2272
2273         if (enic->config.intr_mode < 3 &&
2274             enic->rq_count >= 1 &&
2275             enic->wq_count >= 1 &&
2276             enic->cq_count >= 2 &&
2277             enic->intr_count >= 3) {
2278
2279                 enic->rq_count = 1;
2280                 enic->wq_count = 1;
2281                 enic->cq_count = 2;
2282                 enic->intr_count = 3;
2283
2284                 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_INTX);
2285
2286                 return 0;
2287         }
2288
2289         vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
2290
2291         return -EINVAL;
2292 }
2293
2294 static void enic_clear_intr_mode(struct enic *enic)
2295 {
2296         switch (vnic_dev_get_intr_mode(enic->vdev)) {
2297         case VNIC_DEV_INTR_MODE_MSIX:
2298                 pci_disable_msix(enic->pdev);
2299                 break;
2300         case VNIC_DEV_INTR_MODE_MSI:
2301                 pci_disable_msi(enic->pdev);
2302                 break;
2303         default:
2304                 break;
2305         }
2306
2307         vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
2308 }
2309
2310 static const struct net_device_ops enic_netdev_dynamic_ops = {
2311         .ndo_open               = enic_open,
2312         .ndo_stop               = enic_stop,
2313         .ndo_start_xmit         = enic_hard_start_xmit,
2314         .ndo_get_stats64        = enic_get_stats,
2315         .ndo_validate_addr      = eth_validate_addr,
2316         .ndo_set_rx_mode        = enic_set_rx_mode,
2317         .ndo_set_mac_address    = enic_set_mac_address_dynamic,
2318         .ndo_change_mtu         = enic_change_mtu,
2319         .ndo_vlan_rx_add_vid    = enic_vlan_rx_add_vid,
2320         .ndo_vlan_rx_kill_vid   = enic_vlan_rx_kill_vid,
2321         .ndo_tx_timeout         = enic_tx_timeout,
2322         .ndo_set_vf_port        = enic_set_vf_port,
2323         .ndo_get_vf_port        = enic_get_vf_port,
2324         .ndo_set_vf_mac         = enic_set_vf_mac,
2325 #ifdef CONFIG_NET_POLL_CONTROLLER
2326         .ndo_poll_controller    = enic_poll_controller,
2327 #endif
2328 #ifdef CONFIG_RFS_ACCEL
2329         .ndo_rx_flow_steer      = enic_rx_flow_steer,
2330 #endif
2331 #ifdef CONFIG_NET_RX_BUSY_POLL
2332         .ndo_busy_poll          = enic_busy_poll,
2333 #endif
2334 };
2335
2336 static const struct net_device_ops enic_netdev_ops = {
2337         .ndo_open               = enic_open,
2338         .ndo_stop               = enic_stop,
2339         .ndo_start_xmit         = enic_hard_start_xmit,
2340         .ndo_get_stats64        = enic_get_stats,
2341         .ndo_validate_addr      = eth_validate_addr,
2342         .ndo_set_mac_address    = enic_set_mac_address,
2343         .ndo_set_rx_mode        = enic_set_rx_mode,
2344         .ndo_change_mtu         = enic_change_mtu,
2345         .ndo_vlan_rx_add_vid    = enic_vlan_rx_add_vid,
2346         .ndo_vlan_rx_kill_vid   = enic_vlan_rx_kill_vid,
2347         .ndo_tx_timeout         = enic_tx_timeout,
2348         .ndo_set_vf_port        = enic_set_vf_port,
2349         .ndo_get_vf_port        = enic_get_vf_port,
2350         .ndo_set_vf_mac         = enic_set_vf_mac,
2351 #ifdef CONFIG_NET_POLL_CONTROLLER
2352         .ndo_poll_controller    = enic_poll_controller,
2353 #endif
2354 #ifdef CONFIG_RFS_ACCEL
2355         .ndo_rx_flow_steer      = enic_rx_flow_steer,
2356 #endif
2357 #ifdef CONFIG_NET_RX_BUSY_POLL
2358         .ndo_busy_poll          = enic_busy_poll,
2359 #endif
2360 };
2361
2362 static void enic_dev_deinit(struct enic *enic)
2363 {
2364         unsigned int i;
2365
2366         for (i = 0; i < enic->rq_count; i++) {
2367                 napi_hash_del(&enic->napi[i]);
2368                 netif_napi_del(&enic->napi[i]);
2369         }
2370         if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX)
2371                 for (i = 0; i < enic->wq_count; i++)
2372                         netif_napi_del(&enic->napi[enic_cq_wq(enic, i)]);
2373
2374         enic_free_vnic_resources(enic);
2375         enic_clear_intr_mode(enic);
2376         enic_free_affinity_hint(enic);
2377 }
2378
2379 static void enic_kdump_kernel_config(struct enic *enic)
2380 {
2381         if (is_kdump_kernel()) {
2382                 dev_info(enic_get_dev(enic), "Running from within kdump kernel. Using minimal resources\n");
2383                 enic->rq_count = 1;
2384                 enic->wq_count = 1;
2385                 enic->config.rq_desc_count = ENIC_MIN_RQ_DESCS;
2386                 enic->config.wq_desc_count = ENIC_MIN_WQ_DESCS;
2387                 enic->config.mtu = min_t(u16, 1500, enic->config.mtu);
2388         }
2389 }
2390
2391 static int enic_dev_init(struct enic *enic)
2392 {
2393         struct device *dev = enic_get_dev(enic);
2394         struct net_device *netdev = enic->netdev;
2395         unsigned int i;
2396         int err;
2397
2398         /* Get interrupt coalesce timer info */
2399         err = enic_dev_intr_coal_timer_info(enic);
2400         if (err) {
2401                 dev_warn(dev, "Using default conversion factor for "
2402                         "interrupt coalesce timer\n");
2403                 vnic_dev_intr_coal_timer_info_default(enic->vdev);
2404         }
2405
2406         /* Get vNIC configuration
2407          */
2408
2409         err = enic_get_vnic_config(enic);
2410         if (err) {
2411                 dev_err(dev, "Get vNIC configuration failed, aborting\n");
2412                 return err;
2413         }
2414
2415         /* Get available resource counts
2416          */
2417
2418         enic_get_res_counts(enic);
2419
2420         /* modify resource count if we are in kdump_kernel
2421          */
2422         enic_kdump_kernel_config(enic);
2423
2424         /* Set interrupt mode based on resource counts and system
2425          * capabilities
2426          */
2427
2428         err = enic_set_intr_mode(enic);
2429         if (err) {
2430                 dev_err(dev, "Failed to set intr mode based on resource "
2431                         "counts and system capabilities, aborting\n");
2432                 return err;
2433         }
2434
2435         /* Allocate and configure vNIC resources
2436          */
2437
2438         err = enic_alloc_vnic_resources(enic);
2439         if (err) {
2440                 dev_err(dev, "Failed to alloc vNIC resources, aborting\n");
2441                 goto err_out_free_vnic_resources;
2442         }
2443
2444         enic_init_vnic_resources(enic);
2445
2446         err = enic_set_rss_nic_cfg(enic);
2447         if (err) {
2448                 dev_err(dev, "Failed to config nic, aborting\n");
2449                 goto err_out_free_vnic_resources;
2450         }
2451
2452         switch (vnic_dev_get_intr_mode(enic->vdev)) {
2453         default:
2454                 netif_napi_add(netdev, &enic->napi[0], enic_poll, 64);
2455                 napi_hash_add(&enic->napi[0]);
2456                 break;
2457         case VNIC_DEV_INTR_MODE_MSIX:
2458                 for (i = 0; i < enic->rq_count; i++) {
2459                         netif_napi_add(netdev, &enic->napi[i],
2460                                 enic_poll_msix_rq, NAPI_POLL_WEIGHT);
2461                         napi_hash_add(&enic->napi[i]);
2462                 }
2463                 for (i = 0; i < enic->wq_count; i++)
2464                         netif_napi_add(netdev, &enic->napi[enic_cq_wq(enic, i)],
2465                                        enic_poll_msix_wq, NAPI_POLL_WEIGHT);
2466                 break;
2467         }
2468
2469         return 0;
2470
2471 err_out_free_vnic_resources:
2472         enic_free_affinity_hint(enic);
2473         enic_clear_intr_mode(enic);
2474         enic_free_vnic_resources(enic);
2475
2476         return err;
2477 }
2478
2479 static void enic_iounmap(struct enic *enic)
2480 {
2481         unsigned int i;
2482
2483         for (i = 0; i < ARRAY_SIZE(enic->bar); i++)
2484                 if (enic->bar[i].vaddr)
2485                         iounmap(enic->bar[i].vaddr);
2486 }
2487
2488 static int enic_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2489 {
2490         struct device *dev = &pdev->dev;
2491         struct net_device *netdev;
2492         struct enic *enic;
2493         int using_dac = 0;
2494         unsigned int i;
2495         int err;
2496 #ifdef CONFIG_PCI_IOV
2497         int pos = 0;
2498 #endif
2499         int num_pps = 1;
2500
2501         /* Allocate net device structure and initialize.  Private
2502          * instance data is initialized to zero.
2503          */
2504
2505         netdev = alloc_etherdev_mqs(sizeof(struct enic),
2506                                     ENIC_RQ_MAX, ENIC_WQ_MAX);
2507         if (!netdev)
2508                 return -ENOMEM;
2509
2510         pci_set_drvdata(pdev, netdev);
2511
2512         SET_NETDEV_DEV(netdev, &pdev->dev);
2513
2514         enic = netdev_priv(netdev);
2515         enic->netdev = netdev;
2516         enic->pdev = pdev;
2517
2518         /* Setup PCI resources
2519          */
2520
2521         err = pci_enable_device_mem(pdev);
2522         if (err) {
2523                 dev_err(dev, "Cannot enable PCI device, aborting\n");
2524                 goto err_out_free_netdev;
2525         }
2526
2527         err = pci_request_regions(pdev, DRV_NAME);
2528         if (err) {
2529                 dev_err(dev, "Cannot request PCI regions, aborting\n");
2530                 goto err_out_disable_device;
2531         }
2532
2533         pci_set_master(pdev);
2534
2535         /* Query PCI controller on system for DMA addressing
2536          * limitation for the device.  Try 47-bit first, and
2537          * fail to 32-bit.
2538          */
2539
2540         err = pci_set_dma_mask(pdev, DMA_BIT_MASK(47));
2541         if (err) {
2542                 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
2543                 if (err) {
2544                         dev_err(dev, "No usable DMA configuration, aborting\n");
2545                         goto err_out_release_regions;
2546                 }
2547                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
2548                 if (err) {
2549                         dev_err(dev, "Unable to obtain %u-bit DMA "
2550                                 "for consistent allocations, aborting\n", 32);
2551                         goto err_out_release_regions;
2552                 }
2553         } else {
2554                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(47));
2555                 if (err) {
2556                         dev_err(dev, "Unable to obtain %u-bit DMA "
2557                                 "for consistent allocations, aborting\n", 47);
2558                         goto err_out_release_regions;
2559                 }
2560                 using_dac = 1;
2561         }
2562
2563         /* Map vNIC resources from BAR0-5
2564          */
2565
2566         for (i = 0; i < ARRAY_SIZE(enic->bar); i++) {
2567                 if (!(pci_resource_flags(pdev, i) & IORESOURCE_MEM))
2568                         continue;
2569                 enic->bar[i].len = pci_resource_len(pdev, i);
2570                 enic->bar[i].vaddr = pci_iomap(pdev, i, enic->bar[i].len);
2571                 if (!enic->bar[i].vaddr) {
2572                         dev_err(dev, "Cannot memory-map BAR %d, aborting\n", i);
2573                         err = -ENODEV;
2574                         goto err_out_iounmap;
2575                 }
2576                 enic->bar[i].bus_addr = pci_resource_start(pdev, i);
2577         }
2578
2579         /* Register vNIC device
2580          */
2581
2582         enic->vdev = vnic_dev_register(NULL, enic, pdev, enic->bar,
2583                 ARRAY_SIZE(enic->bar));
2584         if (!enic->vdev) {
2585                 dev_err(dev, "vNIC registration failed, aborting\n");
2586                 err = -ENODEV;
2587                 goto err_out_iounmap;
2588         }
2589
2590         err = vnic_devcmd_init(enic->vdev);
2591
2592         if (err)
2593                 goto err_out_vnic_unregister;
2594
2595 #ifdef CONFIG_PCI_IOV
2596         /* Get number of subvnics */
2597         pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_SRIOV);
2598         if (pos) {
2599                 pci_read_config_word(pdev, pos + PCI_SRIOV_TOTAL_VF,
2600                         &enic->num_vfs);
2601                 if (enic->num_vfs) {
2602                         err = pci_enable_sriov(pdev, enic->num_vfs);
2603                         if (err) {
2604                                 dev_err(dev, "SRIOV enable failed, aborting."
2605                                         " pci_enable_sriov() returned %d\n",
2606                                         err);
2607                                 goto err_out_vnic_unregister;
2608                         }
2609                         enic->priv_flags |= ENIC_SRIOV_ENABLED;
2610                         num_pps = enic->num_vfs;
2611                 }
2612         }
2613 #endif
2614
2615         /* Allocate structure for port profiles */
2616         enic->pp = kcalloc(num_pps, sizeof(*enic->pp), GFP_KERNEL);
2617         if (!enic->pp) {
2618                 err = -ENOMEM;
2619                 goto err_out_disable_sriov_pp;
2620         }
2621
2622         /* Issue device open to get device in known state
2623          */
2624
2625         err = enic_dev_open(enic);
2626         if (err) {
2627                 dev_err(dev, "vNIC dev open failed, aborting\n");
2628                 goto err_out_disable_sriov;
2629         }
2630
2631         /* Setup devcmd lock
2632          */
2633
2634         spin_lock_init(&enic->devcmd_lock);
2635         spin_lock_init(&enic->enic_api_lock);
2636
2637         /*
2638          * Set ingress vlan rewrite mode before vnic initialization
2639          */
2640
2641         err = enic_dev_set_ig_vlan_rewrite_mode(enic);
2642         if (err) {
2643                 dev_err(dev,
2644                         "Failed to set ingress vlan rewrite mode, aborting.\n");
2645                 goto err_out_dev_close;
2646         }
2647
2648         /* Issue device init to initialize the vnic-to-switch link.
2649          * We'll start with carrier off and wait for link UP
2650          * notification later to turn on carrier.  We don't need
2651          * to wait here for the vnic-to-switch link initialization
2652          * to complete; link UP notification is the indication that
2653          * the process is complete.
2654          */
2655
2656         netif_carrier_off(netdev);
2657
2658         /* Do not call dev_init for a dynamic vnic.
2659          * For a dynamic vnic, init_prov_info will be
2660          * called later by an upper layer.
2661          */
2662
2663         if (!enic_is_dynamic(enic)) {
2664                 err = vnic_dev_init(enic->vdev, 0);
2665                 if (err) {
2666                         dev_err(dev, "vNIC dev init failed, aborting\n");
2667                         goto err_out_dev_close;
2668                 }
2669         }
2670
2671         err = enic_dev_init(enic);
2672         if (err) {
2673                 dev_err(dev, "Device initialization failed, aborting\n");
2674                 goto err_out_dev_close;
2675         }
2676
2677         netif_set_real_num_tx_queues(netdev, enic->wq_count);
2678         netif_set_real_num_rx_queues(netdev, enic->rq_count);
2679
2680         /* Setup notification timer, HW reset task, and wq locks
2681          */
2682
2683         init_timer(&enic->notify_timer);
2684         enic->notify_timer.function = enic_notify_timer;
2685         enic->notify_timer.data = (unsigned long)enic;
2686
2687         enic_rfs_flw_tbl_init(enic);
2688         enic_set_rx_coal_setting(enic);
2689         INIT_WORK(&enic->reset, enic_reset);
2690         INIT_WORK(&enic->tx_hang_reset, enic_tx_hang_reset);
2691         INIT_WORK(&enic->change_mtu_work, enic_change_mtu_work);
2692
2693         for (i = 0; i < enic->wq_count; i++)
2694                 spin_lock_init(&enic->wq_lock[i]);
2695
2696         /* Register net device
2697          */
2698
2699         enic->port_mtu = enic->config.mtu;
2700
2701         err = enic_set_mac_addr(netdev, enic->mac_addr);
2702         if (err) {
2703                 dev_err(dev, "Invalid MAC address, aborting\n");
2704                 goto err_out_dev_deinit;
2705         }
2706
2707         enic->tx_coalesce_usecs = enic->config.intr_timer_usec;
2708         /* rx coalesce time already got initialized. This gets used
2709          * if adaptive coal is turned off
2710          */
2711         enic->rx_coalesce_usecs = enic->tx_coalesce_usecs;
2712
2713         if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic))
2714                 netdev->netdev_ops = &enic_netdev_dynamic_ops;
2715         else
2716                 netdev->netdev_ops = &enic_netdev_ops;
2717
2718         netdev->watchdog_timeo = 2 * HZ;
2719         enic_set_ethtool_ops(netdev);
2720
2721         netdev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
2722         if (ENIC_SETTING(enic, LOOP)) {
2723                 netdev->features &= ~NETIF_F_HW_VLAN_CTAG_TX;
2724                 enic->loop_enable = 1;
2725                 enic->loop_tag = enic->config.loop_tag;
2726                 dev_info(dev, "loopback tag=0x%04x\n", enic->loop_tag);
2727         }
2728         if (ENIC_SETTING(enic, TXCSUM))
2729                 netdev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM;
2730         if (ENIC_SETTING(enic, TSO))
2731                 netdev->hw_features |= NETIF_F_TSO |
2732                         NETIF_F_TSO6 | NETIF_F_TSO_ECN;
2733         if (ENIC_SETTING(enic, RSS))
2734                 netdev->hw_features |= NETIF_F_RXHASH;
2735         if (ENIC_SETTING(enic, RXCSUM))
2736                 netdev->hw_features |= NETIF_F_RXCSUM;
2737
2738         netdev->features |= netdev->hw_features;
2739
2740 #ifdef CONFIG_RFS_ACCEL
2741         netdev->hw_features |= NETIF_F_NTUPLE;
2742 #endif
2743
2744         if (using_dac)
2745                 netdev->features |= NETIF_F_HIGHDMA;
2746
2747         netdev->priv_flags |= IFF_UNICAST_FLT;
2748         netdev->mtu = enic->port_mtu;
2749
2750         err = register_netdev(netdev);
2751         if (err) {
2752                 dev_err(dev, "Cannot register net device, aborting\n");
2753                 goto err_out_dev_deinit;
2754         }
2755         enic->rx_copybreak = RX_COPYBREAK_DEFAULT;
2756
2757         return 0;
2758
2759 err_out_dev_deinit:
2760         enic_dev_deinit(enic);
2761 err_out_dev_close:
2762         vnic_dev_close(enic->vdev);
2763 err_out_disable_sriov:
2764         kfree(enic->pp);
2765 err_out_disable_sriov_pp:
2766 #ifdef CONFIG_PCI_IOV
2767         if (enic_sriov_enabled(enic)) {
2768                 pci_disable_sriov(pdev);
2769                 enic->priv_flags &= ~ENIC_SRIOV_ENABLED;
2770         }
2771 #endif
2772 err_out_vnic_unregister:
2773         vnic_dev_unregister(enic->vdev);
2774 err_out_iounmap:
2775         enic_iounmap(enic);
2776 err_out_release_regions:
2777         pci_release_regions(pdev);
2778 err_out_disable_device:
2779         pci_disable_device(pdev);
2780 err_out_free_netdev:
2781         free_netdev(netdev);
2782
2783         return err;
2784 }
2785
2786 static void enic_remove(struct pci_dev *pdev)
2787 {
2788         struct net_device *netdev = pci_get_drvdata(pdev);
2789
2790         if (netdev) {
2791                 struct enic *enic = netdev_priv(netdev);
2792
2793                 cancel_work_sync(&enic->reset);
2794                 cancel_work_sync(&enic->change_mtu_work);
2795                 unregister_netdev(netdev);
2796                 enic_dev_deinit(enic);
2797                 vnic_dev_close(enic->vdev);
2798 #ifdef CONFIG_PCI_IOV
2799                 if (enic_sriov_enabled(enic)) {
2800                         pci_disable_sriov(pdev);
2801                         enic->priv_flags &= ~ENIC_SRIOV_ENABLED;
2802                 }
2803 #endif
2804                 kfree(enic->pp);
2805                 vnic_dev_unregister(enic->vdev);
2806                 enic_iounmap(enic);
2807                 pci_release_regions(pdev);
2808                 pci_disable_device(pdev);
2809                 free_netdev(netdev);
2810         }
2811 }
2812
2813 static struct pci_driver enic_driver = {
2814         .name = DRV_NAME,
2815         .id_table = enic_id_table,
2816         .probe = enic_probe,
2817         .remove = enic_remove,
2818 };
2819
2820 static int __init enic_init_module(void)
2821 {
2822         pr_info("%s, ver %s\n", DRV_DESCRIPTION, DRV_VERSION);
2823
2824         return pci_register_driver(&enic_driver);
2825 }
2826
2827 static void __exit enic_cleanup_module(void)
2828 {
2829         pci_unregister_driver(&enic_driver);
2830 }
2831
2832 module_init(enic_init_module);
2833 module_exit(enic_cleanup_module);