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 #ifdef CONFIG_RFS_ACCEL
  43 #include <linux/cpu_rmap.h>
  44 #endif
  45
  46 #include "cq_enet_desc.h"
  47 #include "vnic_dev.h"
  48 #include "vnic_intr.h"
  49 #include "vnic_stats.h"
  50 #include "vnic_vic.h"
  51 #include "enic_res.h"
  52 #include "enic.h"
  53 #include "enic_dev.h"
  54 #include "enic_pp.h"
  55 #include "enic_clsf.h"
  56
  57 #define ENIC_NOTIFY_TIMER_PERIOD        (2 * HZ)
  58 #define WQ_ENET_MAX_DESC_LEN            (1 << WQ_ENET_LEN_BITS)
  59 #define MAX_TSO                         (1 << 16)
  60 #define ENIC_DESC_MAX_SPLITS            (MAX_TSO / WQ_ENET_MAX_DESC_LEN + 1)
  61
  62 #define PCI_DEVICE_ID_CISCO_VIC_ENET         0x0043  /* ethernet vnic */
  63 #define PCI_DEVICE_ID_CISCO_VIC_ENET_DYN     0x0044  /* enet dynamic vnic */
  64 #define PCI_DEVICE_ID_CISCO_VIC_ENET_VF      0x0071  /* enet SRIOV VF */
  65
  66 /* Supported devices */
  67 static DEFINE_PCI_DEVICE_TABLE(enic_id_table) = {
  68         { PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET) },
  69         { PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET_DYN) },
  70         { PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET_VF) },
  71         { 0, }  /* end of table */
  72 };
  73
  74 MODULE_DESCRIPTION(DRV_DESCRIPTION);
  75 MODULE_AUTHOR("Scott Feldman <scofeldm@cisco.com>");
  76 MODULE_LICENSE("GPL");
  77 MODULE_VERSION(DRV_VERSION);
  78 MODULE_DEVICE_TABLE(pci, enic_id_table);
  79
  80 #define ENIC_LARGE_PKT_THRESHOLD                1000
  81 #define ENIC_MAX_COALESCE_TIMERS                10
  82 /*  Interrupt moderation table, which will be used to decide the
  83  *  coalescing timer values
  84  *  {rx_rate in Mbps, mapping percentage of the range}
  85  */
  86 struct enic_intr_mod_table mod_table[ENIC_MAX_COALESCE_TIMERS + 1] = {
  87         {4000,  0},
  88         {4400, 10},
  89         {5060, 20},
  90         {5230, 30},
  91         {5540, 40},
  92         {5820, 50},
  93         {6120, 60},
  94         {6435, 70},
  95         {6745, 80},
  96         {7000, 90},
  97         {0xFFFFFFFF, 100}
  98 };
  99
 100 /* This table helps the driver to pick different ranges for rx coalescing
 101  * timer depending on the link speed.
 102  */
 103 struct enic_intr_mod_range mod_range[ENIC_MAX_LINK_SPEEDS] = {
 104         {0,  0}, /* 0  - 4  Gbps */
 105         {0,  3}, /* 4  - 10 Gbps */
 106         {3,  6}, /* 10 - 40 Gbps */
 107 };
 108
 109 int enic_is_dynamic(struct enic *enic)
 110 {
 111         return enic->pdev->device == PCI_DEVICE_ID_CISCO_VIC_ENET_DYN;
 112 }
 113
 114 int enic_sriov_enabled(struct enic *enic)
 115 {
 116         return (enic->priv_flags & ENIC_SRIOV_ENABLED) ? 1 : 0;
 117 }
 118
 119 static int enic_is_sriov_vf(struct enic *enic)
 120 {
 121         return enic->pdev->device == PCI_DEVICE_ID_CISCO_VIC_ENET_VF;
 122 }
 123
 124 int enic_is_valid_vf(struct enic *enic, int vf)
 125 {
 126 #ifdef CONFIG_PCI_IOV
 127         return vf >= 0 && vf < enic->num_vfs;
 128 #else
 129         return 0;
 130 #endif
 131 }
 132
 133 static void enic_free_wq_buf(struct vnic_wq *wq, struct vnic_wq_buf *buf)
 134 {
 135         struct enic *enic = vnic_dev_priv(wq->vdev);
 136
 137         if (buf->sop)
 138                 pci_unmap_single(enic->pdev, buf->dma_addr,
 139                         buf->len, PCI_DMA_TODEVICE);
 140         else
 141                 pci_unmap_page(enic->pdev, buf->dma_addr,
 142                         buf->len, PCI_DMA_TODEVICE);
 143
 144         if (buf->os_buf)
 145                 dev_kfree_skb_any(buf->os_buf);
 146 }
 147
 148 static void enic_wq_free_buf(struct vnic_wq *wq,
 149         struct cq_desc *cq_desc, struct vnic_wq_buf *buf, void *opaque)
 150 {
 151         enic_free_wq_buf(wq, buf);
 152 }
 153
 154 static int enic_wq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
 155         u8 type, u16 q_number, u16 completed_index, void *opaque)
 156 {
 157         struct enic *enic = vnic_dev_priv(vdev);
 158
 159         spin_lock(&enic->wq_lock[q_number]);
 160
 161         vnic_wq_service(&enic->wq[q_number], cq_desc,
 162                 completed_index, enic_wq_free_buf,
 163                 opaque);
 164
 165         if (netif_tx_queue_stopped(netdev_get_tx_queue(enic->netdev, q_number)) &&
 166             vnic_wq_desc_avail(&enic->wq[q_number]) >=
 167             (MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS))
 168                 netif_wake_subqueue(enic->netdev, q_number);
 169
 170         spin_unlock(&enic->wq_lock[q_number]);
 171
 172         return 0;
 173 }
 174
 175 static void enic_log_q_error(struct enic *enic)
 176 {
 177         unsigned int i;
 178         u32 error_status;
 179
 180         for (i = 0; i < enic->wq_count; i++) {
 181                 error_status = vnic_wq_error_status(&enic->wq[i]);
 182                 if (error_status)
 183                         netdev_err(enic->netdev, "WQ[%d] error_status %d\n",
 184                                 i, error_status);
 185         }
 186
 187         for (i = 0; i < enic->rq_count; i++) {
 188                 error_status = vnic_rq_error_status(&enic->rq[i]);
 189                 if (error_status)
 190                         netdev_err(enic->netdev, "RQ[%d] error_status %d\n",
 191                                 i, error_status);
 192         }
 193 }
 194
 195 static void enic_msglvl_check(struct enic *enic)
 196 {
 197         u32 msg_enable = vnic_dev_msg_lvl(enic->vdev);
 198
 199         if (msg_enable != enic->msg_enable) {
 200                 netdev_info(enic->netdev, "msg lvl changed from 0x%x to 0x%x\n",
 201                         enic->msg_enable, msg_enable);
 202                 enic->msg_enable = msg_enable;
 203         }
 204 }
 205
 206 static void enic_mtu_check(struct enic *enic)
 207 {
 208         u32 mtu = vnic_dev_mtu(enic->vdev);
 209         struct net_device *netdev = enic->netdev;
 210
 211         if (mtu && mtu != enic->port_mtu) {
 212                 enic->port_mtu = mtu;
 213                 if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic)) {
 214                         mtu = max_t(int, ENIC_MIN_MTU,
 215                                 min_t(int, ENIC_MAX_MTU, mtu));
 216                         if (mtu != netdev->mtu)
 217                                 schedule_work(&enic->change_mtu_work);
 218                 } else {
 219                         if (mtu < netdev->mtu)
 220                                 netdev_warn(netdev,
 221                                         "interface MTU (%d) set higher "
 222                                         "than switch port MTU (%d)\n",
 223                                         netdev->mtu, mtu);
 224                 }
 225         }
 226 }
 227
 228 static void enic_link_check(struct enic *enic)
 229 {
 230         int link_status = vnic_dev_link_status(enic->vdev);
 231         int carrier_ok = netif_carrier_ok(enic->netdev);
 232
 233         if (link_status && !carrier_ok) {
 234                 netdev_info(enic->netdev, "Link UP\n");
 235                 netif_carrier_on(enic->netdev);
 236         } else if (!link_status && carrier_ok) {
 237                 netdev_info(enic->netdev, "Link DOWN\n");
 238                 netif_carrier_off(enic->netdev);
 239         }
 240 }
 241
 242 static void enic_notify_check(struct enic *enic)
 243 {
 244         enic_msglvl_check(enic);
 245         enic_mtu_check(enic);
 246         enic_link_check(enic);
 247 }
 248
 249 #define ENIC_TEST_INTR(pba, i) (pba & (1 << i))
 250
 251 static irqreturn_t enic_isr_legacy(int irq, void *data)
 252 {
 253         struct net_device *netdev = data;
 254         struct enic *enic = netdev_priv(netdev);
 255         unsigned int io_intr = enic_legacy_io_intr();
 256         unsigned int err_intr = enic_legacy_err_intr();
 257         unsigned int notify_intr = enic_legacy_notify_intr();
 258         u32 pba;
 259
 260         vnic_intr_mask(&enic->intr[io_intr]);
 261
 262         pba = vnic_intr_legacy_pba(enic->legacy_pba);
 263         if (!pba) {
 264                 vnic_intr_unmask(&enic->intr[io_intr]);
 265                 return IRQ_NONE;        /* not our interrupt */
 266         }
 267
 268         if (ENIC_TEST_INTR(pba, notify_intr)) {
 269                 vnic_intr_return_all_credits(&enic->intr[notify_intr]);
 270                 enic_notify_check(enic);
 271         }
 272
 273         if (ENIC_TEST_INTR(pba, err_intr)) {
 274                 vnic_intr_return_all_credits(&enic->intr[err_intr]);
 275                 enic_log_q_error(enic);
 276                 /* schedule recovery from WQ/RQ error */
 277                 schedule_work(&enic->reset);
 278                 return IRQ_HANDLED;
 279         }
 280
 281         if (ENIC_TEST_INTR(pba, io_intr)) {
 282                 if (napi_schedule_prep(&enic->napi[0]))
 283                         __napi_schedule(&enic->napi[0]);
 284         } else {
 285                 vnic_intr_unmask(&enic->intr[io_intr]);
 286         }
 287
 288         return IRQ_HANDLED;
 289 }
 290
 291 static irqreturn_t enic_isr_msi(int irq, void *data)
 292 {
 293         struct enic *enic = data;
 294
 295         /* With MSI, there is no sharing of interrupts, so this is
 296          * our interrupt and there is no need to ack it.  The device
 297          * is not providing per-vector masking, so the OS will not
 298          * write to PCI config space to mask/unmask the interrupt.
 299          * We're using mask_on_assertion for MSI, so the device
 300          * automatically masks the interrupt when the interrupt is
 301          * generated.  Later, when exiting polling, the interrupt
 302          * will be unmasked (see enic_poll).
 303          *
 304          * Also, the device uses the same PCIe Traffic Class (TC)
 305          * for Memory Write data and MSI, so there are no ordering
 306          * issues; the MSI will always arrive at the Root Complex
 307          * _after_ corresponding Memory Writes (i.e. descriptor
 308          * writes).
 309          */
 310
 311         napi_schedule(&enic->napi[0]);
 312
 313         return IRQ_HANDLED;
 314 }
 315
 316 static irqreturn_t enic_isr_msix_rq(int irq, void *data)
 317 {
 318         struct napi_struct *napi = data;
 319
 320         /* schedule NAPI polling for RQ cleanup */
 321         napi_schedule(napi);
 322
 323         return IRQ_HANDLED;
 324 }
 325
 326 static irqreturn_t enic_isr_msix_wq(int irq, void *data)
 327 {
 328         struct enic *enic = data;
 329         unsigned int cq;
 330         unsigned int intr;
 331         unsigned int wq_work_to_do = -1; /* no limit */
 332         unsigned int wq_work_done;
 333         unsigned int wq_irq;
 334
 335         wq_irq = (u32)irq - enic->msix_entry[enic_msix_wq_intr(enic, 0)].vector;
 336         cq = enic_cq_wq(enic, wq_irq);
 337         intr = enic_msix_wq_intr(enic, wq_irq);
 338
 339         wq_work_done = vnic_cq_service(&enic->cq[cq],
 340                 wq_work_to_do, enic_wq_service, NULL);
 341
 342         vnic_intr_return_credits(&enic->intr[intr],
 343                 wq_work_done,
 344                 1 /* unmask intr */,
 345                 1 /* reset intr timer */);
 346
 347         return IRQ_HANDLED;
 348 }
 349
 350 static irqreturn_t enic_isr_msix_err(int irq, void *data)
 351 {
 352         struct enic *enic = data;
 353         unsigned int intr = enic_msix_err_intr(enic);
 354
 355         vnic_intr_return_all_credits(&enic->intr[intr]);
 356
 357         enic_log_q_error(enic);
 358
 359         /* schedule recovery from WQ/RQ error */
 360         schedule_work(&enic->reset);
 361
 362         return IRQ_HANDLED;
 363 }
 364
 365 static irqreturn_t enic_isr_msix_notify(int irq, void *data)
 366 {
 367         struct enic *enic = data;
 368         unsigned int intr = enic_msix_notify_intr(enic);
 369
 370         vnic_intr_return_all_credits(&enic->intr[intr]);
 371         enic_notify_check(enic);
 372
 373         return IRQ_HANDLED;
 374 }
 375
 376 static inline void enic_queue_wq_skb_cont(struct enic *enic,
 377         struct vnic_wq *wq, struct sk_buff *skb,
 378         unsigned int len_left, int loopback)
 379 {
 380         const skb_frag_t *frag;
 381
 382         /* Queue additional data fragments */
 383         for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
 384                 len_left -= skb_frag_size(frag);
 385                 enic_queue_wq_desc_cont(wq, skb,
 386                         skb_frag_dma_map(&enic->pdev->dev,
 387                                          frag, 0, skb_frag_size(frag),
 388                                          DMA_TO_DEVICE),
 389                         skb_frag_size(frag),
 390                         (len_left == 0),        /* EOP? */
 391                         loopback);
 392         }
 393 }
 394
 395 static inline void enic_queue_wq_skb_vlan(struct enic *enic,
 396         struct vnic_wq *wq, struct sk_buff *skb,
 397         int vlan_tag_insert, unsigned int vlan_tag, int loopback)
 398 {
 399         unsigned int head_len = skb_headlen(skb);
 400         unsigned int len_left = skb->len - head_len;
 401         int eop = (len_left == 0);
 402
 403         /* Queue the main skb fragment. The fragments are no larger
 404          * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less
 405          * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor
 406          * per fragment is queued.
 407          */
 408         enic_queue_wq_desc(wq, skb,
 409                 pci_map_single(enic->pdev, skb->data,
 410                         head_len, PCI_DMA_TODEVICE),
 411                 head_len,
 412                 vlan_tag_insert, vlan_tag,
 413                 eop, loopback);
 414
 415         if (!eop)
 416                 enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback);
 417 }
 418
 419 static inline void enic_queue_wq_skb_csum_l4(struct enic *enic,
 420         struct vnic_wq *wq, struct sk_buff *skb,
 421         int vlan_tag_insert, unsigned int vlan_tag, int loopback)
 422 {
 423         unsigned int head_len = skb_headlen(skb);
 424         unsigned int len_left = skb->len - head_len;
 425         unsigned int hdr_len = skb_checksum_start_offset(skb);
 426         unsigned int csum_offset = hdr_len + skb->csum_offset;
 427         int eop = (len_left == 0);
 428
 429         /* Queue the main skb fragment. The fragments are no larger
 430          * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less
 431          * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor
 432          * per fragment is queued.
 433          */
 434         enic_queue_wq_desc_csum_l4(wq, skb,
 435                 pci_map_single(enic->pdev, skb->data,
 436                         head_len, PCI_DMA_TODEVICE),
 437                 head_len,
 438                 csum_offset,
 439                 hdr_len,
 440                 vlan_tag_insert, vlan_tag,
 441                 eop, loopback);
 442
 443         if (!eop)
 444                 enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback);
 445 }
 446
 447 static inline void enic_queue_wq_skb_tso(struct enic *enic,
 448         struct vnic_wq *wq, struct sk_buff *skb, unsigned int mss,
 449         int vlan_tag_insert, unsigned int vlan_tag, int loopback)
 450 {
 451         unsigned int frag_len_left = skb_headlen(skb);
 452         unsigned int len_left = skb->len - frag_len_left;
 453         unsigned int hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
 454         int eop = (len_left == 0);
 455         unsigned int len;
 456         dma_addr_t dma_addr;
 457         unsigned int offset = 0;
 458         skb_frag_t *frag;
 459
 460         /* Preload TCP csum field with IP pseudo hdr calculated
 461          * with IP length set to zero.  HW will later add in length
 462          * to each TCP segment resulting from the TSO.
 463          */
 464
 465         if (skb->protocol == cpu_to_be16(ETH_P_IP)) {
 466                 ip_hdr(skb)->check = 0;
 467                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
 468                         ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 469         } else if (skb->protocol == cpu_to_be16(ETH_P_IPV6)) {
 470                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
 471                         &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 472         }
 473
 474         /* Queue WQ_ENET_MAX_DESC_LEN length descriptors
 475          * for the main skb fragment
 476          */
 477         while (frag_len_left) {
 478                 len = min(frag_len_left, (unsigned int)WQ_ENET_MAX_DESC_LEN);
 479                 dma_addr = pci_map_single(enic->pdev, skb->data + offset,
 480                                 len, PCI_DMA_TODEVICE);
 481                 enic_queue_wq_desc_tso(wq, skb,
 482                         dma_addr,
 483                         len,
 484                         mss, hdr_len,
 485                         vlan_tag_insert, vlan_tag,
 486                         eop && (len == frag_len_left), loopback);
 487                 frag_len_left -= len;
 488                 offset += len;
 489         }
 490
 491         if (eop)
 492                 return;
 493
 494         /* Queue WQ_ENET_MAX_DESC_LEN length descriptors
 495          * for additional data fragments
 496          */
 497         for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
 498                 len_left -= skb_frag_size(frag);
 499                 frag_len_left = skb_frag_size(frag);
 500                 offset = 0;
 501
 502                 while (frag_len_left) {
 503                         len = min(frag_len_left,
 504                                 (unsigned int)WQ_ENET_MAX_DESC_LEN);
 505                         dma_addr = skb_frag_dma_map(&enic->pdev->dev, frag,
 506                                                     offset, len,
 507                                                     DMA_TO_DEVICE);
 508                         enic_queue_wq_desc_cont(wq, skb,
 509                                 dma_addr,
 510                                 len,
 511                                 (len_left == 0) &&
 512                                 (len == frag_len_left),         /* EOP? */
 513                                 loopback);
 514                         frag_len_left -= len;
 515                         offset += len;
 516                 }
 517         }
 518 }
 519
 520 static inline void enic_queue_wq_skb(struct enic *enic,
 521         struct vnic_wq *wq, struct sk_buff *skb)
 522 {
 523         unsigned int mss = skb_shinfo(skb)->gso_size;
 524         unsigned int vlan_tag = 0;
 525         int vlan_tag_insert = 0;
 526         int loopback = 0;
 527
 528         if (vlan_tx_tag_present(skb)) {
 529                 /* VLAN tag from trunking driver */
 530                 vlan_tag_insert = 1;
 531                 vlan_tag = vlan_tx_tag_get(skb);
 532         } else if (enic->loop_enable) {
 533                 vlan_tag = enic->loop_tag;
 534                 loopback = 1;
 535         }
 536
 537         if (mss)
 538                 enic_queue_wq_skb_tso(enic, wq, skb, mss,
 539                         vlan_tag_insert, vlan_tag, loopback);
 540         else if (skb->ip_summed == CHECKSUM_PARTIAL)
 541                 enic_queue_wq_skb_csum_l4(enic, wq, skb,
 542                         vlan_tag_insert, vlan_tag, loopback);
 543         else
 544                 enic_queue_wq_skb_vlan(enic, wq, skb,
 545                         vlan_tag_insert, vlan_tag, loopback);
 546 }
 547
 548 /* netif_tx_lock held, process context with BHs disabled, or BH */
 549 static netdev_tx_t enic_hard_start_xmit(struct sk_buff *skb,
 550         struct net_device *netdev)
 551 {
 552         struct enic *enic = netdev_priv(netdev);
 553         struct vnic_wq *wq;
 554         unsigned long flags;
 555         unsigned int txq_map;
 556
 557         if (skb->len <= 0) {
 558                 dev_kfree_skb_any(skb);
 559                 return NETDEV_TX_OK;
 560         }
 561
 562         txq_map = skb_get_queue_mapping(skb) % enic->wq_count;
 563         wq = &enic->wq[txq_map];
 564
 565         /* Non-TSO sends must fit within ENIC_NON_TSO_MAX_DESC descs,
 566          * which is very likely.  In the off chance it's going to take
 567          * more than * ENIC_NON_TSO_MAX_DESC, linearize the skb.
 568          */
 569
 570         if (skb_shinfo(skb)->gso_size == 0 &&
 571             skb_shinfo(skb)->nr_frags + 1 > ENIC_NON_TSO_MAX_DESC &&
 572             skb_linearize(skb)) {
 573                 dev_kfree_skb_any(skb);
 574                 return NETDEV_TX_OK;
 575         }
 576
 577         spin_lock_irqsave(&enic->wq_lock[txq_map], flags);
 578
 579         if (vnic_wq_desc_avail(wq) <
 580             skb_shinfo(skb)->nr_frags + ENIC_DESC_MAX_SPLITS) {
 581                 netif_tx_stop_queue(netdev_get_tx_queue(netdev, txq_map));
 582                 /* This is a hard error, log it */
 583                 netdev_err(netdev, "BUG! Tx ring full when queue awake!\n");
 584                 spin_unlock_irqrestore(&enic->wq_lock[txq_map], flags);
 585                 return NETDEV_TX_BUSY;
 586         }
 587
 588         enic_queue_wq_skb(enic, wq, skb);
 589
 590         if (vnic_wq_desc_avail(wq) < MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS)
 591                 netif_tx_stop_queue(netdev_get_tx_queue(netdev, txq_map));
 592
 593         spin_unlock_irqrestore(&enic->wq_lock[txq_map], flags);
 594
 595         return NETDEV_TX_OK;
 596 }
 597
 598 /* dev_base_lock rwlock held, nominally process context */
 599 static struct rtnl_link_stats64 *enic_get_stats(struct net_device *netdev,
 600                                                 struct rtnl_link_stats64 *net_stats)
 601 {
 602         struct enic *enic = netdev_priv(netdev);
 603         struct vnic_stats *stats;
 604
 605         enic_dev_stats_dump(enic, &stats);
 606
 607         net_stats->tx_packets = stats->tx.tx_frames_ok;
 608         net_stats->tx_bytes = stats->tx.tx_bytes_ok;
 609         net_stats->tx_errors = stats->tx.tx_errors;
 610         net_stats->tx_dropped = stats->tx.tx_drops;
 611
 612         net_stats->rx_packets = stats->rx.rx_frames_ok;
 613         net_stats->rx_bytes = stats->rx.rx_bytes_ok;
 614         net_stats->rx_errors = stats->rx.rx_errors;
 615         net_stats->multicast = stats->rx.rx_multicast_frames_ok;
 616         net_stats->rx_over_errors = enic->rq_truncated_pkts;
 617         net_stats->rx_crc_errors = enic->rq_bad_fcs;
 618         net_stats->rx_dropped = stats->rx.rx_no_bufs + stats->rx.rx_drop;
 619
 620         return net_stats;
 621 }
 622
 623 static int enic_mc_sync(struct net_device *netdev, const u8 *mc_addr)
 624 {
 625         struct enic *enic = netdev_priv(netdev);
 626
 627         if (enic->mc_count == ENIC_MULTICAST_PERFECT_FILTERS) {
 628                 unsigned int mc_count = netdev_mc_count(netdev);
 629
 630                 netdev_warn(netdev, "Registering only %d out of %d multicast addresses\n",
 631                             ENIC_MULTICAST_PERFECT_FILTERS, mc_count);
 632
 633                 return -ENOSPC;
 634         }
 635
 636         enic_dev_add_addr(enic, mc_addr);
 637         enic->mc_count++;
 638
 639         return 0;
 640 }
 641
 642 static int enic_mc_unsync(struct net_device *netdev, const u8 *mc_addr)
 643 {
 644         struct enic *enic = netdev_priv(netdev);
 645
 646         enic_dev_del_addr(enic, mc_addr);
 647         enic->mc_count--;
 648
 649         return 0;
 650 }
 651
 652 static int enic_uc_sync(struct net_device *netdev, const u8 *uc_addr)
 653 {
 654         struct enic *enic = netdev_priv(netdev);
 655
 656         if (enic->uc_count == ENIC_UNICAST_PERFECT_FILTERS) {
 657                 unsigned int uc_count = netdev_uc_count(netdev);
 658
 659                 netdev_warn(netdev, "Registering only %d out of %d unicast addresses\n",
 660                             ENIC_UNICAST_PERFECT_FILTERS, uc_count);
 661
 662                 return -ENOSPC;
 663         }
 664
 665         enic_dev_add_addr(enic, uc_addr);
 666         enic->uc_count++;
 667
 668         return 0;
 669 }
 670
 671 static int enic_uc_unsync(struct net_device *netdev, const u8 *uc_addr)
 672 {
 673         struct enic *enic = netdev_priv(netdev);
 674
 675         enic_dev_del_addr(enic, uc_addr);
 676         enic->uc_count--;
 677
 678         return 0;
 679 }
 680
 681 void enic_reset_addr_lists(struct enic *enic)
 682 {
 683         struct net_device *netdev = enic->netdev;
 684
 685         __dev_uc_unsync(netdev, NULL);
 686         __dev_mc_unsync(netdev, NULL);
 687
 688         enic->mc_count = 0;
 689         enic->uc_count = 0;
 690         enic->flags = 0;
 691 }
 692
 693 static int enic_set_mac_addr(struct net_device *netdev, char *addr)
 694 {
 695         struct enic *enic = netdev_priv(netdev);
 696
 697         if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic)) {
 698                 if (!is_valid_ether_addr(addr) && !is_zero_ether_addr(addr))
 699                         return -EADDRNOTAVAIL;
 700         } else {
 701                 if (!is_valid_ether_addr(addr))
 702                         return -EADDRNOTAVAIL;
 703         }
 704
 705         memcpy(netdev->dev_addr, addr, netdev->addr_len);
 706
 707         return 0;
 708 }
 709
 710 static int enic_set_mac_address_dynamic(struct net_device *netdev, void *p)
 711 {
 712         struct enic *enic = netdev_priv(netdev);
 713         struct sockaddr *saddr = p;
 714         char *addr = saddr->sa_data;
 715         int err;
 716
 717         if (netif_running(enic->netdev)) {
 718                 err = enic_dev_del_station_addr(enic);
 719                 if (err)
 720                         return err;
 721         }
 722
 723         err = enic_set_mac_addr(netdev, addr);
 724         if (err)
 725                 return err;
 726
 727         if (netif_running(enic->netdev)) {
 728                 err = enic_dev_add_station_addr(enic);
 729                 if (err)
 730                         return err;
 731         }
 732
 733         return err;
 734 }
 735
 736 static int enic_set_mac_address(struct net_device *netdev, void *p)
 737 {
 738         struct sockaddr *saddr = p;
 739         char *addr = saddr->sa_data;
 740         struct enic *enic = netdev_priv(netdev);
 741         int err;
 742
 743         err = enic_dev_del_station_addr(enic);
 744         if (err)
 745                 return err;
 746
 747         err = enic_set_mac_addr(netdev, addr);
 748         if (err)
 749                 return err;
 750
 751         return enic_dev_add_station_addr(enic);
 752 }
 753
 754 /* netif_tx_lock held, BHs disabled */
 755 static void enic_set_rx_mode(struct net_device *netdev)
 756 {
 757         struct enic *enic = netdev_priv(netdev);
 758         int directed = 1;
 759         int multicast = (netdev->flags & IFF_MULTICAST) ? 1 : 0;
 760         int broadcast = (netdev->flags & IFF_BROADCAST) ? 1 : 0;
 761         int promisc = (netdev->flags & IFF_PROMISC) ||
 762                 netdev_uc_count(netdev) > ENIC_UNICAST_PERFECT_FILTERS;
 763         int allmulti = (netdev->flags & IFF_ALLMULTI) ||
 764                 netdev_mc_count(netdev) > ENIC_MULTICAST_PERFECT_FILTERS;
 765         unsigned int flags = netdev->flags |
 766                 (allmulti ? IFF_ALLMULTI : 0) |
 767                 (promisc ? IFF_PROMISC : 0);
 768
 769         if (enic->flags != flags) {
 770                 enic->flags = flags;
 771                 enic_dev_packet_filter(enic, directed,
 772                         multicast, broadcast, promisc, allmulti);
 773         }
 774
 775         if (!promisc) {
 776                 __dev_uc_sync(netdev, enic_uc_sync, enic_uc_unsync);
 777                 if (!allmulti)
 778                         __dev_mc_sync(netdev, enic_mc_sync, enic_mc_unsync);
 779         }
 780 }
 781
 782 /* netif_tx_lock held, BHs disabled */
 783 static void enic_tx_timeout(struct net_device *netdev)
 784 {
 785         struct enic *enic = netdev_priv(netdev);
 786         schedule_work(&enic->reset);
 787 }
 788
 789 static int enic_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
 790 {
 791         struct enic *enic = netdev_priv(netdev);
 792         struct enic_port_profile *pp;
 793         int err;
 794
 795         ENIC_PP_BY_INDEX(enic, vf, pp, &err);
 796         if (err)
 797                 return err;
 798
 799         if (is_valid_ether_addr(mac) || is_zero_ether_addr(mac)) {
 800                 if (vf == PORT_SELF_VF) {
 801                         memcpy(pp->vf_mac, mac, ETH_ALEN);
 802                         return 0;
 803                 } else {
 804                         /*
 805                          * For sriov vf's set the mac in hw
 806                          */
 807                         ENIC_DEVCMD_PROXY_BY_INDEX(vf, err, enic,
 808                                 vnic_dev_set_mac_addr, mac);
 809                         return enic_dev_status_to_errno(err);
 810                 }
 811         } else
 812                 return -EINVAL;
 813 }
 814
 815 static int enic_set_vf_port(struct net_device *netdev, int vf,
 816         struct nlattr *port[])
 817 {
 818         struct enic *enic = netdev_priv(netdev);
 819         struct enic_port_profile prev_pp;
 820         struct enic_port_profile *pp;
 821         int err = 0, restore_pp = 1;
 822
 823         ENIC_PP_BY_INDEX(enic, vf, pp, &err);
 824         if (err)
 825                 return err;
 826
 827         if (!port[IFLA_PORT_REQUEST])
 828                 return -EOPNOTSUPP;
 829
 830         memcpy(&prev_pp, pp, sizeof(*enic->pp));
 831         memset(pp, 0, sizeof(*enic->pp));
 832
 833         pp->set |= ENIC_SET_REQUEST;
 834         pp->request = nla_get_u8(port[IFLA_PORT_REQUEST]);
 835
 836         if (port[IFLA_PORT_PROFILE]) {
 837                 pp->set |= ENIC_SET_NAME;
 838                 memcpy(pp->name, nla_data(port[IFLA_PORT_PROFILE]),
 839                         PORT_PROFILE_MAX);
 840         }
 841
 842         if (port[IFLA_PORT_INSTANCE_UUID]) {
 843                 pp->set |= ENIC_SET_INSTANCE;
 844                 memcpy(pp->instance_uuid,
 845                         nla_data(port[IFLA_PORT_INSTANCE_UUID]), PORT_UUID_MAX);
 846         }
 847
 848         if (port[IFLA_PORT_HOST_UUID]) {
 849                 pp->set |= ENIC_SET_HOST;
 850                 memcpy(pp->host_uuid,
 851                         nla_data(port[IFLA_PORT_HOST_UUID]), PORT_UUID_MAX);
 852         }
 853
 854         if (vf == PORT_SELF_VF) {
 855                 /* Special case handling: mac came from IFLA_VF_MAC */
 856                 if (!is_zero_ether_addr(prev_pp.vf_mac))
 857                         memcpy(pp->mac_addr, prev_pp.vf_mac, ETH_ALEN);
 858
 859                 if (is_zero_ether_addr(netdev->dev_addr))
 860                         eth_hw_addr_random(netdev);
 861         } else {
 862                 /* SR-IOV VF: get mac from adapter */
 863                 ENIC_DEVCMD_PROXY_BY_INDEX(vf, err, enic,
 864                         vnic_dev_get_mac_addr, pp->mac_addr);
 865                 if (err) {
 866                         netdev_err(netdev, "Error getting mac for vf %d\n", vf);
 867                         memcpy(pp, &prev_pp, sizeof(*pp));
 868                         return enic_dev_status_to_errno(err);
 869                 }
 870         }
 871
 872         err = enic_process_set_pp_request(enic, vf, &prev_pp, &restore_pp);
 873         if (err) {
 874                 if (restore_pp) {
 875                         /* Things are still the way they were: Implicit
 876                          * DISASSOCIATE failed
 877                          */
 878                         memcpy(pp, &prev_pp, sizeof(*pp));
 879                 } else {
 880                         memset(pp, 0, sizeof(*pp));
 881                         if (vf == PORT_SELF_VF)
 882                                 memset(netdev->dev_addr, 0, ETH_ALEN);
 883                 }
 884         } else {
 885                 /* Set flag to indicate that the port assoc/disassoc
 886                  * request has been sent out to fw
 887                  */
 888                 pp->set |= ENIC_PORT_REQUEST_APPLIED;
 889
 890                 /* If DISASSOCIATE, clean up all assigned/saved macaddresses */
 891                 if (pp->request == PORT_REQUEST_DISASSOCIATE) {
 892                         memset(pp->mac_addr, 0, ETH_ALEN);
 893                         if (vf == PORT_SELF_VF)
 894                                 memset(netdev->dev_addr, 0, ETH_ALEN);
 895                 }
 896         }
 897
 898         if (vf == PORT_SELF_VF)
 899                 memset(pp->vf_mac, 0, ETH_ALEN);
 900
 901         return err;
 902 }
 903
 904 static int enic_get_vf_port(struct net_device *netdev, int vf,
 905         struct sk_buff *skb)
 906 {
 907         struct enic *enic = netdev_priv(netdev);
 908         u16 response = PORT_PROFILE_RESPONSE_SUCCESS;
 909         struct enic_port_profile *pp;
 910         int err;
 911
 912         ENIC_PP_BY_INDEX(enic, vf, pp, &err);
 913         if (err)
 914                 return err;
 915
 916         if (!(pp->set & ENIC_PORT_REQUEST_APPLIED))
 917                 return -ENODATA;
 918
 919         err = enic_process_get_pp_request(enic, vf, pp->request, &response);
 920         if (err)
 921                 return err;
 922
 923         if (nla_put_u16(skb, IFLA_PORT_REQUEST, pp->request) ||
 924             nla_put_u16(skb, IFLA_PORT_RESPONSE, response) ||
 925             ((pp->set & ENIC_SET_NAME) &&
 926              nla_put(skb, IFLA_PORT_PROFILE, PORT_PROFILE_MAX, pp->name)) ||
 927             ((pp->set & ENIC_SET_INSTANCE) &&
 928              nla_put(skb, IFLA_PORT_INSTANCE_UUID, PORT_UUID_MAX,
 929                      pp->instance_uuid)) ||
 930             ((pp->set & ENIC_SET_HOST) &&
 931              nla_put(skb, IFLA_PORT_HOST_UUID, PORT_UUID_MAX, pp->host_uuid)))
 932                 goto nla_put_failure;
 933         return 0;
 934
 935 nla_put_failure:
 936         return -EMSGSIZE;
 937 }
 938
 939 static void enic_free_rq_buf(struct vnic_rq *rq, struct vnic_rq_buf *buf)
 940 {
 941         struct enic *enic = vnic_dev_priv(rq->vdev);
 942
 943         if (!buf->os_buf)
 944                 return;
 945
 946         pci_unmap_single(enic->pdev, buf->dma_addr,
 947                 buf->len, PCI_DMA_FROMDEVICE);
 948         dev_kfree_skb_any(buf->os_buf);
 949 }
 950
 951 static int enic_rq_alloc_buf(struct vnic_rq *rq)
 952 {
 953         struct enic *enic = vnic_dev_priv(rq->vdev);
 954         struct net_device *netdev = enic->netdev;
 955         struct sk_buff *skb;
 956         unsigned int len = netdev->mtu + VLAN_ETH_HLEN;
 957         unsigned int os_buf_index = 0;
 958         dma_addr_t dma_addr;
 959
 960         skb = netdev_alloc_skb_ip_align(netdev, len);
 961         if (!skb)
 962                 return -ENOMEM;
 963
 964         dma_addr = pci_map_single(enic->pdev, skb->data,
 965                 len, PCI_DMA_FROMDEVICE);
 966
 967         enic_queue_rq_desc(rq, skb, os_buf_index,
 968                 dma_addr, len);
 969
 970         return 0;
 971 }
 972
 973 static void enic_intr_update_pkt_size(struct vnic_rx_bytes_counter *pkt_size,
 974                                       u32 pkt_len)
 975 {
 976         if (ENIC_LARGE_PKT_THRESHOLD <= pkt_len)
 977                 pkt_size->large_pkt_bytes_cnt += pkt_len;
 978         else
 979                 pkt_size->small_pkt_bytes_cnt += pkt_len;
 980 }
 981
 982 static void enic_rq_indicate_buf(struct vnic_rq *rq,
 983         struct cq_desc *cq_desc, struct vnic_rq_buf *buf,
 984         int skipped, void *opaque)
 985 {
 986         struct enic *enic = vnic_dev_priv(rq->vdev);
 987         struct net_device *netdev = enic->netdev;
 988         struct sk_buff *skb;
 989         struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
 990
 991         u8 type, color, eop, sop, ingress_port, vlan_stripped;
 992         u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof;
 993         u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok;
 994         u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc;
 995         u8 packet_error;
 996         u16 q_number, completed_index, bytes_written, vlan_tci, checksum;
 997         u32 rss_hash;
 998
 999         if (skipped)
1000                 return;
1001
1002         skb = buf->os_buf;
1003         prefetch(skb->data - NET_IP_ALIGN);
1004         pci_unmap_single(enic->pdev, buf->dma_addr,
1005                 buf->len, PCI_DMA_FROMDEVICE);
1006
1007         cq_enet_rq_desc_dec((struct cq_enet_rq_desc *)cq_desc,
1008                 &type, &color, &q_number, &completed_index,
1009                 &ingress_port, &fcoe, &eop, &sop, &rss_type,
1010                 &csum_not_calc, &rss_hash, &bytes_written,
1011                 &packet_error, &vlan_stripped, &vlan_tci, &checksum,
1012                 &fcoe_sof, &fcoe_fc_crc_ok, &fcoe_enc_error,
1013                 &fcoe_eof, &tcp_udp_csum_ok, &udp, &tcp,
1014                 &ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment,
1015                 &fcs_ok);
1016
1017         if (packet_error) {
1018
1019                 if (!fcs_ok) {
1020                         if (bytes_written > 0)
1021                                 enic->rq_bad_fcs++;
1022                         else if (bytes_written == 0)
1023                                 enic->rq_truncated_pkts++;
1024                 }
1025
1026                 dev_kfree_skb_any(skb);
1027
1028                 return;
1029         }
1030
1031         if (eop && bytes_written > 0) {
1032
1033                 /* Good receive
1034                  */
1035
1036                 skb_put(skb, bytes_written);
1037                 skb->protocol = eth_type_trans(skb, netdev);
1038                 skb_record_rx_queue(skb, q_number);
1039                 if (netdev->features & NETIF_F_RXHASH) {
1040                         skb_set_hash(skb, rss_hash,
1041                                      (rss_type &
1042                                       (NIC_CFG_RSS_HASH_TYPE_TCP_IPV6_EX |
1043                                        NIC_CFG_RSS_HASH_TYPE_TCP_IPV6 |
1044                                        NIC_CFG_RSS_HASH_TYPE_TCP_IPV4)) ?
1045                                      PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3);
1046                 }
1047
1048                 if ((netdev->features & NETIF_F_RXCSUM) && !csum_not_calc) {
1049                         skb->csum = htons(checksum);
1050                         skb->ip_summed = CHECKSUM_COMPLETE;
1051                 }
1052
1053                 if (vlan_stripped)
1054                         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tci);
1055
1056                 if (netdev->features & NETIF_F_GRO)
1057                         napi_gro_receive(&enic->napi[q_number], skb);
1058                 else
1059                         netif_receive_skb(skb);
1060                 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1061                         enic_intr_update_pkt_size(&cq->pkt_size_counter,
1062                                                   bytes_written);
1063         } else {
1064
1065                 /* Buffer overflow
1066                  */
1067
1068                 dev_kfree_skb_any(skb);
1069         }
1070 }
1071
1072 static int enic_rq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
1073         u8 type, u16 q_number, u16 completed_index, void *opaque)
1074 {
1075         struct enic *enic = vnic_dev_priv(vdev);
1076
1077         vnic_rq_service(&enic->rq[q_number], cq_desc,
1078                 completed_index, VNIC_RQ_RETURN_DESC,
1079                 enic_rq_indicate_buf, opaque);
1080
1081         return 0;
1082 }
1083
1084 static int enic_poll(struct napi_struct *napi, int budget)
1085 {
1086         struct net_device *netdev = napi->dev;
1087         struct enic *enic = netdev_priv(netdev);
1088         unsigned int cq_rq = enic_cq_rq(enic, 0);
1089         unsigned int cq_wq = enic_cq_wq(enic, 0);
1090         unsigned int intr = enic_legacy_io_intr();
1091         unsigned int rq_work_to_do = budget;
1092         unsigned int wq_work_to_do = -1; /* no limit */
1093         unsigned int  work_done, rq_work_done = 0, wq_work_done;
1094         int err;
1095
1096         /* Service RQ (first) and WQ
1097          */
1098
1099         if (budget > 0)
1100                 rq_work_done = vnic_cq_service(&enic->cq[cq_rq],
1101                         rq_work_to_do, enic_rq_service, NULL);
1102
1103         wq_work_done = vnic_cq_service(&enic->cq[cq_wq],
1104                 wq_work_to_do, enic_wq_service, NULL);
1105
1106         /* Accumulate intr event credits for this polling
1107          * cycle.  An intr event is the completion of a
1108          * a WQ or RQ packet.
1109          */
1110
1111         work_done = rq_work_done + wq_work_done;
1112
1113         if (work_done > 0)
1114                 vnic_intr_return_credits(&enic->intr[intr],
1115                         work_done,
1116                         0 /* don't unmask intr */,
1117                         0 /* don't reset intr timer */);
1118
1119         err = vnic_rq_fill(&enic->rq[0], enic_rq_alloc_buf);
1120
1121         /* Buffer allocation failed. Stay in polling
1122          * mode so we can try to fill the ring again.
1123          */
1124
1125         if (err)
1126                 rq_work_done = rq_work_to_do;
1127
1128         if (rq_work_done < rq_work_to_do) {
1129
1130                 /* Some work done, but not enough to stay in polling,
1131                  * exit polling
1132                  */
1133
1134                 napi_complete(napi);
1135                 vnic_intr_unmask(&enic->intr[intr]);
1136         }
1137
1138         return rq_work_done;
1139 }
1140
1141 static void enic_set_int_moderation(struct enic *enic, struct vnic_rq *rq)
1142 {
1143         unsigned int intr = enic_msix_rq_intr(enic, rq->index);
1144         struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
1145         u32 timer = cq->tobe_rx_coal_timeval;
1146
1147         if (cq->tobe_rx_coal_timeval != cq->cur_rx_coal_timeval) {
1148                 vnic_intr_coalescing_timer_set(&enic->intr[intr], timer);
1149                 cq->cur_rx_coal_timeval = cq->tobe_rx_coal_timeval;
1150         }
1151 }
1152
1153 static void enic_calc_int_moderation(struct enic *enic, struct vnic_rq *rq)
1154 {
1155         struct enic_rx_coal *rx_coal = &enic->rx_coalesce_setting;
1156         struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
1157         struct vnic_rx_bytes_counter *pkt_size_counter = &cq->pkt_size_counter;
1158         int index;
1159         u32 timer;
1160         u32 range_start;
1161         u32 traffic;
1162         u64 delta;
1163         ktime_t now = ktime_get();
1164
1165         delta = ktime_us_delta(now, cq->prev_ts);
1166         if (delta < ENIC_AIC_TS_BREAK)
1167                 return;
1168         cq->prev_ts = now;
1169
1170         traffic = pkt_size_counter->large_pkt_bytes_cnt +
1171                   pkt_size_counter->small_pkt_bytes_cnt;
1172         /* The table takes Mbps
1173          * traffic *= 8    => bits
1174          * traffic *= (10^6 / delta)    => bps
1175          * traffic /= 10^6     => Mbps
1176          *
1177          * Combining, traffic *= (8 / delta)
1178          */
1179
1180         traffic <<= 3;
1181         traffic = delta > UINT_MAX ? 0 : traffic / (u32)delta;
1182
1183         for (index = 0; index < ENIC_MAX_COALESCE_TIMERS; index++)
1184                 if (traffic < mod_table[index].rx_rate)
1185                         break;
1186         range_start = (pkt_size_counter->small_pkt_bytes_cnt >
1187                        pkt_size_counter->large_pkt_bytes_cnt << 1) ?
1188                       rx_coal->small_pkt_range_start :
1189                       rx_coal->large_pkt_range_start;
1190         timer = range_start + ((rx_coal->range_end - range_start) *
1191                                mod_table[index].range_percent / 100);
1192         /* Damping */
1193         cq->tobe_rx_coal_timeval = (timer + cq->tobe_rx_coal_timeval) >> 1;
1194
1195         pkt_size_counter->large_pkt_bytes_cnt = 0;
1196         pkt_size_counter->small_pkt_bytes_cnt = 0;
1197 }
1198
1199 #ifdef CONFIG_RFS_ACCEL
1200 static void enic_free_rx_cpu_rmap(struct enic *enic)
1201 {
1202         free_irq_cpu_rmap(enic->netdev->rx_cpu_rmap);
1203         enic->netdev->rx_cpu_rmap = NULL;
1204 }
1205
1206 static void enic_set_rx_cpu_rmap(struct enic *enic)
1207 {
1208         int i, res;
1209
1210         if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX) {
1211                 enic->netdev->rx_cpu_rmap = alloc_irq_cpu_rmap(enic->rq_count);
1212                 if (unlikely(!enic->netdev->rx_cpu_rmap))
1213                         return;
1214                 for (i = 0; i < enic->rq_count; i++) {
1215                         res = irq_cpu_rmap_add(enic->netdev->rx_cpu_rmap,
1216                                                enic->msix_entry[i].vector);
1217                         if (unlikely(res)) {
1218                                 enic_free_rx_cpu_rmap(enic);
1219                                 return;
1220                         }
1221                 }
1222         }
1223 }
1224
1225 #else
1226
1227 static void enic_free_rx_cpu_rmap(struct enic *enic)
1228 {
1229 }
1230
1231 static void enic_set_rx_cpu_rmap(struct enic *enic)
1232 {
1233 }
1234
1235 #endif /* CONFIG_RFS_ACCEL */
1236
1237 static int enic_poll_msix(struct napi_struct *napi, int budget)
1238 {
1239         struct net_device *netdev = napi->dev;
1240         struct enic *enic = netdev_priv(netdev);
1241         unsigned int rq = (napi - &enic->napi[0]);
1242         unsigned int cq = enic_cq_rq(enic, rq);
1243         unsigned int intr = enic_msix_rq_intr(enic, rq);
1244         unsigned int work_to_do = budget;
1245         unsigned int work_done = 0;
1246         int err;
1247
1248         /* Service RQ
1249          */
1250
1251         if (budget > 0)
1252                 work_done = vnic_cq_service(&enic->cq[cq],
1253                         work_to_do, enic_rq_service, NULL);
1254
1255         /* Return intr event credits for this polling
1256          * cycle.  An intr event is the completion of a
1257          * RQ packet.
1258          */
1259
1260         if (work_done > 0)
1261                 vnic_intr_return_credits(&enic->intr[intr],
1262                         work_done,
1263                         0 /* don't unmask intr */,
1264                         0 /* don't reset intr timer */);
1265
1266         err = vnic_rq_fill(&enic->rq[rq], enic_rq_alloc_buf);
1267
1268         /* Buffer allocation failed. Stay in polling mode
1269          * so we can try to fill the ring again.
1270          */
1271
1272         if (err)
1273                 work_done = work_to_do;
1274         if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1275                 /* Call the function which refreshes
1276                  * the intr coalescing timer value based on
1277                  * the traffic.  This is supported only in
1278                  * the case of MSI-x mode
1279                  */
1280                 enic_calc_int_moderation(enic, &enic->rq[rq]);
1281
1282         if (work_done < work_to_do) {
1283
1284                 /* Some work done, but not enough to stay in polling,
1285                  * exit polling
1286                  */
1287
1288                 napi_complete(napi);
1289                 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1290                         enic_set_int_moderation(enic, &enic->rq[rq]);
1291                 vnic_intr_unmask(&enic->intr[intr]);
1292         }
1293
1294         return work_done;
1295 }
1296
1297 static void enic_notify_timer(unsigned long data)
1298 {
1299         struct enic *enic = (struct enic *)data;
1300
1301         enic_notify_check(enic);
1302
1303         mod_timer(&enic->notify_timer,
1304                 round_jiffies(jiffies + ENIC_NOTIFY_TIMER_PERIOD));
1305 }
1306
1307 static void enic_free_intr(struct enic *enic)
1308 {
1309         struct net_device *netdev = enic->netdev;
1310         unsigned int i;
1311
1312         enic_free_rx_cpu_rmap(enic);
1313         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1314         case VNIC_DEV_INTR_MODE_INTX:
1315                 free_irq(enic->pdev->irq, netdev);
1316                 break;
1317         case VNIC_DEV_INTR_MODE_MSI:
1318                 free_irq(enic->pdev->irq, enic);
1319                 break;
1320         case VNIC_DEV_INTR_MODE_MSIX:
1321                 for (i = 0; i < ARRAY_SIZE(enic->msix); i++)
1322                         if (enic->msix[i].requested)
1323                                 free_irq(enic->msix_entry[i].vector,
1324                                         enic->msix[i].devid);
1325                 break;
1326         default:
1327                 break;
1328         }
1329 }
1330
1331 static int enic_request_intr(struct enic *enic)
1332 {
1333         struct net_device *netdev = enic->netdev;
1334         unsigned int i, intr;
1335         int err = 0;
1336
1337         enic_set_rx_cpu_rmap(enic);
1338         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1339
1340         case VNIC_DEV_INTR_MODE_INTX:
1341
1342                 err = request_irq(enic->pdev->irq, enic_isr_legacy,
1343                         IRQF_SHARED, netdev->name, netdev);
1344                 break;
1345
1346         case VNIC_DEV_INTR_MODE_MSI:
1347
1348                 err = request_irq(enic->pdev->irq, enic_isr_msi,
1349                         0, netdev->name, enic);
1350                 break;
1351
1352         case VNIC_DEV_INTR_MODE_MSIX:
1353
1354                 for (i = 0; i < enic->rq_count; i++) {
1355                         intr = enic_msix_rq_intr(enic, i);
1356                         snprintf(enic->msix[intr].devname,
1357                                 sizeof(enic->msix[intr].devname),
1358                                 "%.11s-rx-%d", netdev->name, i);
1359                         enic->msix[intr].isr = enic_isr_msix_rq;
1360                         enic->msix[intr].devid = &enic->napi[i];
1361                 }
1362
1363                 for (i = 0; i < enic->wq_count; i++) {
1364                         intr = enic_msix_wq_intr(enic, i);
1365                         snprintf(enic->msix[intr].devname,
1366                                 sizeof(enic->msix[intr].devname),
1367                                 "%.11s-tx-%d", netdev->name, i);
1368                         enic->msix[intr].isr = enic_isr_msix_wq;
1369                         enic->msix[intr].devid = enic;
1370                 }
1371
1372                 intr = enic_msix_err_intr(enic);
1373                 snprintf(enic->msix[intr].devname,
1374                         sizeof(enic->msix[intr].devname),
1375                         "%.11s-err", netdev->name);
1376                 enic->msix[intr].isr = enic_isr_msix_err;
1377                 enic->msix[intr].devid = enic;
1378
1379                 intr = enic_msix_notify_intr(enic);
1380                 snprintf(enic->msix[intr].devname,
1381                         sizeof(enic->msix[intr].devname),
1382                         "%.11s-notify", netdev->name);
1383                 enic->msix[intr].isr = enic_isr_msix_notify;
1384                 enic->msix[intr].devid = enic;
1385
1386                 for (i = 0; i < ARRAY_SIZE(enic->msix); i++)
1387                         enic->msix[i].requested = 0;
1388
1389                 for (i = 0; i < enic->intr_count; i++) {
1390                         err = request_irq(enic->msix_entry[i].vector,
1391                                 enic->msix[i].isr, 0,
1392                                 enic->msix[i].devname,
1393                                 enic->msix[i].devid);
1394                         if (err) {
1395                                 enic_free_intr(enic);
1396                                 break;
1397                         }
1398                         enic->msix[i].requested = 1;
1399                 }
1400
1401                 break;
1402
1403         default:
1404                 break;
1405         }
1406
1407         return err;
1408 }
1409
1410 static void enic_synchronize_irqs(struct enic *enic)
1411 {
1412         unsigned int i;
1413
1414         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1415         case VNIC_DEV_INTR_MODE_INTX:
1416         case VNIC_DEV_INTR_MODE_MSI:
1417                 synchronize_irq(enic->pdev->irq);
1418                 break;
1419         case VNIC_DEV_INTR_MODE_MSIX:
1420                 for (i = 0; i < enic->intr_count; i++)
1421                         synchronize_irq(enic->msix_entry[i].vector);
1422                 break;
1423         default:
1424                 break;
1425         }
1426 }
1427
1428 static void enic_set_rx_coal_setting(struct enic *enic)
1429 {
1430         unsigned int speed;
1431         int index = -1;
1432         struct enic_rx_coal *rx_coal = &enic->rx_coalesce_setting;
1433
1434         /* If intr mode is not MSIX, do not do adaptive coalescing */
1435         if (VNIC_DEV_INTR_MODE_MSIX != vnic_dev_get_intr_mode(enic->vdev)) {
1436                 netdev_info(enic->netdev, "INTR mode is not MSIX, Not initializing adaptive coalescing");
1437                 return;
1438         }
1439
1440         /* 1. Read the link speed from fw
1441          * 2. Pick the default range for the speed
1442          * 3. Update it in enic->rx_coalesce_setting
1443          */
1444         speed = vnic_dev_port_speed(enic->vdev);
1445         if (ENIC_LINK_SPEED_10G < speed)
1446                 index = ENIC_LINK_40G_INDEX;
1447         else if (ENIC_LINK_SPEED_4G < speed)
1448                 index = ENIC_LINK_10G_INDEX;
1449         else
1450                 index = ENIC_LINK_4G_INDEX;
1451
1452         rx_coal->small_pkt_range_start = mod_range[index].small_pkt_range_start;
1453         rx_coal->large_pkt_range_start = mod_range[index].large_pkt_range_start;
1454         rx_coal->range_end = ENIC_RX_COALESCE_RANGE_END;
1455
1456         /* Start with the value provided by UCSM */
1457         for (index = 0; index < enic->rq_count; index++)
1458                 enic->cq[index].cur_rx_coal_timeval =
1459                                 enic->config.intr_timer_usec;
1460
1461         rx_coal->use_adaptive_rx_coalesce = 1;
1462 }
1463
1464 static int enic_dev_notify_set(struct enic *enic)
1465 {
1466         int err;
1467
1468         spin_lock_bh(&enic->devcmd_lock);
1469         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1470         case VNIC_DEV_INTR_MODE_INTX:
1471                 err = vnic_dev_notify_set(enic->vdev,
1472                         enic_legacy_notify_intr());
1473                 break;
1474         case VNIC_DEV_INTR_MODE_MSIX:
1475                 err = vnic_dev_notify_set(enic->vdev,
1476                         enic_msix_notify_intr(enic));
1477                 break;
1478         default:
1479                 err = vnic_dev_notify_set(enic->vdev, -1 /* no intr */);
1480                 break;
1481         }
1482         spin_unlock_bh(&enic->devcmd_lock);
1483
1484         return err;
1485 }
1486
1487 static void enic_notify_timer_start(struct enic *enic)
1488 {
1489         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1490         case VNIC_DEV_INTR_MODE_MSI:
1491                 mod_timer(&enic->notify_timer, jiffies);
1492                 break;
1493         default:
1494                 /* Using intr for notification for INTx/MSI-X */
1495                 break;
1496         }
1497 }
1498
1499 /* rtnl lock is held, process context */
1500 static int enic_open(struct net_device *netdev)
1501 {
1502         struct enic *enic = netdev_priv(netdev);
1503         unsigned int i;
1504         int err;
1505
1506         err = enic_request_intr(enic);
1507         if (err) {
1508                 netdev_err(netdev, "Unable to request irq.\n");
1509                 return err;
1510         }
1511
1512         err = enic_dev_notify_set(enic);
1513         if (err) {
1514                 netdev_err(netdev,
1515                         "Failed to alloc notify buffer, aborting.\n");
1516                 goto err_out_free_intr;
1517         }
1518
1519         for (i = 0; i < enic->rq_count; i++) {
1520                 vnic_rq_fill(&enic->rq[i], enic_rq_alloc_buf);
1521                 /* Need at least one buffer on ring to get going */
1522                 if (vnic_rq_desc_used(&enic->rq[i]) == 0) {
1523                         netdev_err(netdev, "Unable to alloc receive buffers\n");
1524                         err = -ENOMEM;
1525                         goto err_out_notify_unset;
1526                 }
1527         }
1528
1529         for (i = 0; i < enic->wq_count; i++)
1530                 vnic_wq_enable(&enic->wq[i]);
1531         for (i = 0; i < enic->rq_count; i++)
1532                 vnic_rq_enable(&enic->rq[i]);
1533
1534         if (!enic_is_dynamic(enic) && !enic_is_sriov_vf(enic))
1535                 enic_dev_add_station_addr(enic);
1536
1537         enic_set_rx_mode(netdev);
1538
1539         netif_tx_wake_all_queues(netdev);
1540
1541         for (i = 0; i < enic->rq_count; i++)
1542                 napi_enable(&enic->napi[i]);
1543
1544         enic_dev_enable(enic);
1545
1546         for (i = 0; i < enic->intr_count; i++)
1547                 vnic_intr_unmask(&enic->intr[i]);
1548
1549         enic_notify_timer_start(enic);
1550         enic_rfs_flw_tbl_init(enic);
1551
1552         return 0;
1553
1554 err_out_notify_unset:
1555         enic_dev_notify_unset(enic);
1556 err_out_free_intr:
1557         enic_free_intr(enic);
1558
1559         return err;
1560 }
1561
1562 /* rtnl lock is held, process context */
1563 static int enic_stop(struct net_device *netdev)
1564 {
1565         struct enic *enic = netdev_priv(netdev);
1566         unsigned int i;
1567         int err;
1568
1569         for (i = 0; i < enic->intr_count; i++) {
1570                 vnic_intr_mask(&enic->intr[i]);
1571                 (void)vnic_intr_masked(&enic->intr[i]); /* flush write */
1572         }
1573
1574         enic_synchronize_irqs(enic);
1575
1576         del_timer_sync(&enic->notify_timer);
1577         enic_rfs_flw_tbl_free(enic);
1578
1579         enic_dev_disable(enic);
1580
1581         for (i = 0; i < enic->rq_count; i++)
1582                 napi_disable(&enic->napi[i]);
1583
1584         netif_carrier_off(netdev);
1585         netif_tx_disable(netdev);
1586
1587         if (!enic_is_dynamic(enic) && !enic_is_sriov_vf(enic))
1588                 enic_dev_del_station_addr(enic);
1589
1590         for (i = 0; i < enic->wq_count; i++) {
1591                 err = vnic_wq_disable(&enic->wq[i]);
1592                 if (err)
1593                         return err;
1594         }
1595         for (i = 0; i < enic->rq_count; i++) {
1596                 err = vnic_rq_disable(&enic->rq[i]);
1597                 if (err)
1598                         return err;
1599         }
1600
1601         enic_dev_notify_unset(enic);
1602         enic_free_intr(enic);
1603
1604         for (i = 0; i < enic->wq_count; i++)
1605                 vnic_wq_clean(&enic->wq[i], enic_free_wq_buf);
1606         for (i = 0; i < enic->rq_count; i++)
1607                 vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
1608         for (i = 0; i < enic->cq_count; i++)
1609                 vnic_cq_clean(&enic->cq[i]);
1610         for (i = 0; i < enic->intr_count; i++)
1611                 vnic_intr_clean(&enic->intr[i]);
1612
1613         return 0;
1614 }
1615
1616 static int enic_change_mtu(struct net_device *netdev, int new_mtu)
1617 {
1618         struct enic *enic = netdev_priv(netdev);
1619         int running = netif_running(netdev);
1620
1621         if (new_mtu < ENIC_MIN_MTU || new_mtu > ENIC_MAX_MTU)
1622                 return -EINVAL;
1623
1624         if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic))
1625                 return -EOPNOTSUPP;
1626
1627         if (running)
1628                 enic_stop(netdev);
1629
1630         netdev->mtu = new_mtu;
1631
1632         if (netdev->mtu > enic->port_mtu)
1633                 netdev_warn(netdev,
1634                         "interface MTU (%d) set higher than port MTU (%d)\n",
1635                         netdev->mtu, enic->port_mtu);
1636
1637         if (running)
1638                 enic_open(netdev);
1639
1640         return 0;
1641 }
1642
1643 static void enic_change_mtu_work(struct work_struct *work)
1644 {
1645         struct enic *enic = container_of(work, struct enic, change_mtu_work);
1646         struct net_device *netdev = enic->netdev;
1647         int new_mtu = vnic_dev_mtu(enic->vdev);
1648         int err;
1649         unsigned int i;
1650
1651         new_mtu = max_t(int, ENIC_MIN_MTU, min_t(int, ENIC_MAX_MTU, new_mtu));
1652
1653         rtnl_lock();
1654
1655         /* Stop RQ */
1656         del_timer_sync(&enic->notify_timer);
1657
1658         for (i = 0; i < enic->rq_count; i++)
1659                 napi_disable(&enic->napi[i]);
1660
1661         vnic_intr_mask(&enic->intr[0]);
1662         enic_synchronize_irqs(enic);
1663         err = vnic_rq_disable(&enic->rq[0]);
1664         if (err) {
1665                 rtnl_unlock();
1666                 netdev_err(netdev, "Unable to disable RQ.\n");
1667                 return;
1668         }
1669         vnic_rq_clean(&enic->rq[0], enic_free_rq_buf);
1670         vnic_cq_clean(&enic->cq[0]);
1671         vnic_intr_clean(&enic->intr[0]);
1672
1673         /* Fill RQ with new_mtu-sized buffers */
1674         netdev->mtu = new_mtu;
1675         vnic_rq_fill(&enic->rq[0], enic_rq_alloc_buf);
1676         /* Need at least one buffer on ring to get going */
1677         if (vnic_rq_desc_used(&enic->rq[0]) == 0) {
1678                 rtnl_unlock();
1679                 netdev_err(netdev, "Unable to alloc receive buffers.\n");
1680                 return;
1681         }
1682
1683         /* Start RQ */
1684         vnic_rq_enable(&enic->rq[0]);
1685         napi_enable(&enic->napi[0]);
1686         vnic_intr_unmask(&enic->intr[0]);
1687         enic_notify_timer_start(enic);
1688
1689         rtnl_unlock();
1690
1691         netdev_info(netdev, "interface MTU set as %d\n", netdev->mtu);
1692 }
1693
1694 #ifdef CONFIG_NET_POLL_CONTROLLER
1695 static void enic_poll_controller(struct net_device *netdev)
1696 {
1697         struct enic *enic = netdev_priv(netdev);
1698         struct vnic_dev *vdev = enic->vdev;
1699         unsigned int i, intr;
1700
1701         switch (vnic_dev_get_intr_mode(vdev)) {
1702         case VNIC_DEV_INTR_MODE_MSIX:
1703                 for (i = 0; i < enic->rq_count; i++) {
1704                         intr = enic_msix_rq_intr(enic, i);
1705                         enic_isr_msix_rq(enic->msix_entry[intr].vector,
1706                                 &enic->napi[i]);
1707                 }
1708
1709                 for (i = 0; i < enic->wq_count; i++) {
1710                         intr = enic_msix_wq_intr(enic, i);
1711                         enic_isr_msix_wq(enic->msix_entry[intr].vector, enic);
1712                 }
1713
1714                 break;
1715         case VNIC_DEV_INTR_MODE_MSI:
1716                 enic_isr_msi(enic->pdev->irq, enic);
1717                 break;
1718         case VNIC_DEV_INTR_MODE_INTX:
1719                 enic_isr_legacy(enic->pdev->irq, netdev);
1720                 break;
1721         default:
1722                 break;
1723         }
1724 }
1725 #endif
1726
1727 static int enic_dev_wait(struct vnic_dev *vdev,
1728         int (*start)(struct vnic_dev *, int),
1729         int (*finished)(struct vnic_dev *, int *),
1730         int arg)
1731 {
1732         unsigned long time;
1733         int done;
1734         int err;
1735
1736         BUG_ON(in_interrupt());
1737
1738         err = start(vdev, arg);
1739         if (err)
1740                 return err;
1741
1742         /* Wait for func to complete...2 seconds max
1743          */
1744
1745         time = jiffies + (HZ * 2);
1746         do {
1747
1748                 err = finished(vdev, &done);
1749                 if (err)
1750                         return err;
1751
1752                 if (done)
1753                         return 0;
1754
1755                 schedule_timeout_uninterruptible(HZ / 10);
1756
1757         } while (time_after(time, jiffies));
1758
1759         return -ETIMEDOUT;
1760 }
1761
1762 static int enic_dev_open(struct enic *enic)
1763 {
1764         int err;
1765
1766         err = enic_dev_wait(enic->vdev, vnic_dev_open,
1767                 vnic_dev_open_done, 0);
1768         if (err)
1769                 dev_err(enic_get_dev(enic), "vNIC device open failed, err %d\n",
1770                         err);
1771
1772         return err;
1773 }
1774
1775 static int enic_dev_hang_reset(struct enic *enic)
1776 {
1777         int err;
1778
1779         err = enic_dev_wait(enic->vdev, vnic_dev_hang_reset,
1780                 vnic_dev_hang_reset_done, 0);
1781         if (err)
1782                 netdev_err(enic->netdev, "vNIC hang reset failed, err %d\n",
1783                         err);
1784
1785         return err;
1786 }
1787
1788 static int enic_set_rsskey(struct enic *enic)
1789 {
1790         dma_addr_t rss_key_buf_pa;
1791         union vnic_rss_key *rss_key_buf_va = NULL;
1792         union vnic_rss_key rss_key = {
1793                 .key[0].b = {85, 67, 83, 97, 119, 101, 115, 111, 109, 101},
1794                 .key[1].b = {80, 65, 76, 79, 117, 110, 105, 113, 117, 101},
1795                 .key[2].b = {76, 73, 78, 85, 88, 114, 111, 99, 107, 115},
1796                 .key[3].b = {69, 78, 73, 67, 105, 115, 99, 111, 111, 108},
1797         };
1798         int err;
1799
1800         rss_key_buf_va = pci_alloc_consistent(enic->pdev,
1801                 sizeof(union vnic_rss_key), &rss_key_buf_pa);
1802         if (!rss_key_buf_va)
1803                 return -ENOMEM;
1804
1805         memcpy(rss_key_buf_va, &rss_key, sizeof(union vnic_rss_key));
1806
1807         spin_lock_bh(&enic->devcmd_lock);
1808         err = enic_set_rss_key(enic,
1809                 rss_key_buf_pa,
1810                 sizeof(union vnic_rss_key));
1811         spin_unlock_bh(&enic->devcmd_lock);
1812
1813         pci_free_consistent(enic->pdev, sizeof(union vnic_rss_key),
1814                 rss_key_buf_va, rss_key_buf_pa);
1815
1816         return err;
1817 }
1818
1819 static int enic_set_rsscpu(struct enic *enic, u8 rss_hash_bits)
1820 {
1821         dma_addr_t rss_cpu_buf_pa;
1822         union vnic_rss_cpu *rss_cpu_buf_va = NULL;
1823         unsigned int i;
1824         int err;
1825
1826         rss_cpu_buf_va = pci_alloc_consistent(enic->pdev,
1827                 sizeof(union vnic_rss_cpu), &rss_cpu_buf_pa);
1828         if (!rss_cpu_buf_va)
1829                 return -ENOMEM;
1830
1831         for (i = 0; i < (1 << rss_hash_bits); i++)
1832                 (*rss_cpu_buf_va).cpu[i/4].b[i%4] = i % enic->rq_count;
1833
1834         spin_lock_bh(&enic->devcmd_lock);
1835         err = enic_set_rss_cpu(enic,
1836                 rss_cpu_buf_pa,
1837                 sizeof(union vnic_rss_cpu));
1838         spin_unlock_bh(&enic->devcmd_lock);
1839
1840         pci_free_consistent(enic->pdev, sizeof(union vnic_rss_cpu),
1841                 rss_cpu_buf_va, rss_cpu_buf_pa);
1842
1843         return err;
1844 }
1845
1846 static int enic_set_niccfg(struct enic *enic, u8 rss_default_cpu,
1847         u8 rss_hash_type, u8 rss_hash_bits, u8 rss_base_cpu, u8 rss_enable)
1848 {
1849         const u8 tso_ipid_split_en = 0;
1850         const u8 ig_vlan_strip_en = 1;
1851         int err;
1852
1853         /* Enable VLAN tag stripping.
1854         */
1855
1856         spin_lock_bh(&enic->devcmd_lock);
1857         err = enic_set_nic_cfg(enic,
1858                 rss_default_cpu, rss_hash_type,
1859                 rss_hash_bits, rss_base_cpu,
1860                 rss_enable, tso_ipid_split_en,
1861                 ig_vlan_strip_en);
1862         spin_unlock_bh(&enic->devcmd_lock);
1863
1864         return err;
1865 }
1866
1867 static int enic_set_rss_nic_cfg(struct enic *enic)
1868 {
1869         struct device *dev = enic_get_dev(enic);
1870         const u8 rss_default_cpu = 0;
1871         const u8 rss_hash_type = NIC_CFG_RSS_HASH_TYPE_IPV4 |
1872                 NIC_CFG_RSS_HASH_TYPE_TCP_IPV4 |
1873                 NIC_CFG_RSS_HASH_TYPE_IPV6 |
1874                 NIC_CFG_RSS_HASH_TYPE_TCP_IPV6;
1875         const u8 rss_hash_bits = 7;
1876         const u8 rss_base_cpu = 0;
1877         u8 rss_enable = ENIC_SETTING(enic, RSS) && (enic->rq_count > 1);
1878
1879         if (rss_enable) {
1880                 if (!enic_set_rsskey(enic)) {
1881                         if (enic_set_rsscpu(enic, rss_hash_bits)) {
1882                                 rss_enable = 0;
1883                                 dev_warn(dev, "RSS disabled, "
1884                                         "Failed to set RSS cpu indirection table.");
1885                         }
1886                 } else {
1887                         rss_enable = 0;
1888                         dev_warn(dev, "RSS disabled, Failed to set RSS key.\n");
1889                 }
1890         }
1891
1892         return enic_set_niccfg(enic, rss_default_cpu, rss_hash_type,
1893                 rss_hash_bits, rss_base_cpu, rss_enable);
1894 }
1895
1896 static void enic_reset(struct work_struct *work)
1897 {
1898         struct enic *enic = container_of(work, struct enic, reset);
1899
1900         if (!netif_running(enic->netdev))
1901                 return;
1902
1903         rtnl_lock();
1904
1905         spin_lock(&enic->enic_api_lock);
1906         enic_dev_hang_notify(enic);
1907         enic_stop(enic->netdev);
1908         enic_dev_hang_reset(enic);
1909         enic_reset_addr_lists(enic);
1910         enic_init_vnic_resources(enic);
1911         enic_set_rss_nic_cfg(enic);
1912         enic_dev_set_ig_vlan_rewrite_mode(enic);
1913         enic_open(enic->netdev);
1914         spin_unlock(&enic->enic_api_lock);
1915         call_netdevice_notifiers(NETDEV_REBOOT, enic->netdev);
1916
1917         rtnl_unlock();
1918 }
1919
1920 static int enic_set_intr_mode(struct enic *enic)
1921 {
1922         unsigned int n = min_t(unsigned int, enic->rq_count, ENIC_RQ_MAX);
1923         unsigned int m = min_t(unsigned int, enic->wq_count, ENIC_WQ_MAX);
1924         unsigned int i;
1925
1926         /* Set interrupt mode (INTx, MSI, MSI-X) depending
1927          * on system capabilities.
1928          *
1929          * Try MSI-X first
1930          *
1931          * We need n RQs, m WQs, n+m CQs, and n+m+2 INTRs
1932          * (the second to last INTR is used for WQ/RQ errors)
1933          * (the last INTR is used for notifications)
1934          */
1935
1936         BUG_ON(ARRAY_SIZE(enic->msix_entry) < n + m + 2);
1937         for (i = 0; i < n + m + 2; i++)
1938                 enic->msix_entry[i].entry = i;
1939
1940         /* Use multiple RQs if RSS is enabled
1941          */
1942
1943         if (ENIC_SETTING(enic, RSS) &&
1944             enic->config.intr_mode < 1 &&
1945             enic->rq_count >= n &&
1946             enic->wq_count >= m &&
1947             enic->cq_count >= n + m &&
1948             enic->intr_count >= n + m + 2) {
1949
1950                 if (pci_enable_msix_range(enic->pdev, enic->msix_entry,
1951                                           n + m + 2, n + m + 2) > 0) {
1952
1953                         enic->rq_count = n;
1954                         enic->wq_count = m;
1955                         enic->cq_count = n + m;
1956                         enic->intr_count = n + m + 2;
1957
1958                         vnic_dev_set_intr_mode(enic->vdev,
1959                                 VNIC_DEV_INTR_MODE_MSIX);
1960
1961                         return 0;
1962                 }
1963         }
1964
1965         if (enic->config.intr_mode < 1 &&
1966             enic->rq_count >= 1 &&
1967             enic->wq_count >= m &&
1968             enic->cq_count >= 1 + m &&
1969             enic->intr_count >= 1 + m + 2) {
1970                 if (pci_enable_msix_range(enic->pdev, enic->msix_entry,
1971                                           1 + m + 2, 1 + m + 2) > 0) {
1972
1973                         enic->rq_count = 1;
1974                         enic->wq_count = m;
1975                         enic->cq_count = 1 + m;
1976                         enic->intr_count = 1 + m + 2;
1977
1978                         vnic_dev_set_intr_mode(enic->vdev,
1979                                 VNIC_DEV_INTR_MODE_MSIX);
1980
1981                         return 0;
1982                 }
1983         }
1984
1985         /* Next try MSI
1986          *
1987          * We need 1 RQ, 1 WQ, 2 CQs, and 1 INTR
1988          */
1989
1990         if (enic->config.intr_mode < 2 &&
1991             enic->rq_count >= 1 &&
1992             enic->wq_count >= 1 &&
1993             enic->cq_count >= 2 &&
1994             enic->intr_count >= 1 &&
1995             !pci_enable_msi(enic->pdev)) {
1996
1997                 enic->rq_count = 1;
1998                 enic->wq_count = 1;
1999                 enic->cq_count = 2;
2000                 enic->intr_count = 1;
2001
2002                 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSI);
2003
2004                 return 0;
2005         }
2006
2007         /* Next try INTx
2008          *
2009          * We need 1 RQ, 1 WQ, 2 CQs, and 3 INTRs
2010          * (the first INTR is used for WQ/RQ)
2011          * (the second INTR is used for WQ/RQ errors)
2012          * (the last INTR is used for notifications)
2013          */
2014
2015         if (enic->config.intr_mode < 3 &&
2016             enic->rq_count >= 1 &&
2017             enic->wq_count >= 1 &&
2018             enic->cq_count >= 2 &&
2019             enic->intr_count >= 3) {
2020
2021                 enic->rq_count = 1;
2022                 enic->wq_count = 1;
2023                 enic->cq_count = 2;
2024                 enic->intr_count = 3;
2025
2026                 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_INTX);
2027
2028                 return 0;
2029         }
2030
2031         vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
2032
2033         return -EINVAL;
2034 }
2035
2036 static void enic_clear_intr_mode(struct enic *enic)
2037 {
2038         switch (vnic_dev_get_intr_mode(enic->vdev)) {
2039         case VNIC_DEV_INTR_MODE_MSIX:
2040                 pci_disable_msix(enic->pdev);
2041                 break;
2042         case VNIC_DEV_INTR_MODE_MSI:
2043                 pci_disable_msi(enic->pdev);
2044                 break;
2045         default:
2046                 break;
2047         }
2048
2049         vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
2050 }
2051
2052 static const struct net_device_ops enic_netdev_dynamic_ops = {
2053         .ndo_open               = enic_open,
2054         .ndo_stop               = enic_stop,
2055         .ndo_start_xmit         = enic_hard_start_xmit,
2056         .ndo_get_stats64        = enic_get_stats,
2057         .ndo_validate_addr      = eth_validate_addr,
2058         .ndo_set_rx_mode        = enic_set_rx_mode,
2059         .ndo_set_mac_address    = enic_set_mac_address_dynamic,
2060         .ndo_change_mtu         = enic_change_mtu,
2061         .ndo_vlan_rx_add_vid    = enic_vlan_rx_add_vid,
2062         .ndo_vlan_rx_kill_vid   = enic_vlan_rx_kill_vid,
2063         .ndo_tx_timeout         = enic_tx_timeout,
2064         .ndo_set_vf_port        = enic_set_vf_port,
2065         .ndo_get_vf_port        = enic_get_vf_port,
2066         .ndo_set_vf_mac         = enic_set_vf_mac,
2067 #ifdef CONFIG_NET_POLL_CONTROLLER
2068         .ndo_poll_controller    = enic_poll_controller,
2069 #endif
2070 #ifdef CONFIG_RFS_ACCEL
2071         .ndo_rx_flow_steer      = enic_rx_flow_steer,
2072 #endif
2073 };
2074
2075 static const struct net_device_ops enic_netdev_ops = {
2076         .ndo_open               = enic_open,
2077         .ndo_stop               = enic_stop,
2078         .ndo_start_xmit         = enic_hard_start_xmit,
2079         .ndo_get_stats64        = enic_get_stats,
2080         .ndo_validate_addr      = eth_validate_addr,
2081         .ndo_set_mac_address    = enic_set_mac_address,
2082         .ndo_set_rx_mode        = enic_set_rx_mode,
2083         .ndo_change_mtu         = enic_change_mtu,
2084         .ndo_vlan_rx_add_vid    = enic_vlan_rx_add_vid,
2085         .ndo_vlan_rx_kill_vid   = enic_vlan_rx_kill_vid,
2086         .ndo_tx_timeout         = enic_tx_timeout,
2087         .ndo_set_vf_port        = enic_set_vf_port,
2088         .ndo_get_vf_port        = enic_get_vf_port,
2089         .ndo_set_vf_mac         = enic_set_vf_mac,
2090 #ifdef CONFIG_NET_POLL_CONTROLLER
2091         .ndo_poll_controller    = enic_poll_controller,
2092 #endif
2093 #ifdef CONFIG_RFS_ACCEL
2094         .ndo_rx_flow_steer      = enic_rx_flow_steer,
2095 #endif
2096 };
2097
2098 static void enic_dev_deinit(struct enic *enic)
2099 {
2100         unsigned int i;
2101
2102         for (i = 0; i < enic->rq_count; i++)
2103                 netif_napi_del(&enic->napi[i]);
2104
2105         enic_free_vnic_resources(enic);
2106         enic_clear_intr_mode(enic);
2107 }
2108
2109 static int enic_dev_init(struct enic *enic)
2110 {
2111         struct device *dev = enic_get_dev(enic);
2112         struct net_device *netdev = enic->netdev;
2113         unsigned int i;
2114         int err;
2115
2116         /* Get interrupt coalesce timer info */
2117         err = enic_dev_intr_coal_timer_info(enic);
2118         if (err) {
2119                 dev_warn(dev, "Using default conversion factor for "
2120                         "interrupt coalesce timer\n");
2121                 vnic_dev_intr_coal_timer_info_default(enic->vdev);
2122         }
2123
2124         /* Get vNIC configuration
2125          */
2126
2127         err = enic_get_vnic_config(enic);
2128         if (err) {
2129                 dev_err(dev, "Get vNIC configuration failed, aborting\n");
2130                 return err;
2131         }
2132
2133         /* Get available resource counts
2134          */
2135
2136         enic_get_res_counts(enic);
2137
2138         /* Set interrupt mode based on resource counts and system
2139          * capabilities
2140          */
2141
2142         err = enic_set_intr_mode(enic);
2143         if (err) {
2144                 dev_err(dev, "Failed to set intr mode based on resource "
2145                         "counts and system capabilities, aborting\n");
2146                 return err;
2147         }
2148
2149         /* Allocate and configure vNIC resources
2150          */
2151
2152         err = enic_alloc_vnic_resources(enic);
2153         if (err) {
2154                 dev_err(dev, "Failed to alloc vNIC resources, aborting\n");
2155                 goto err_out_free_vnic_resources;
2156         }
2157
2158         enic_init_vnic_resources(enic);
2159
2160         err = enic_set_rss_nic_cfg(enic);
2161         if (err) {
2162                 dev_err(dev, "Failed to config nic, aborting\n");
2163                 goto err_out_free_vnic_resources;
2164         }
2165
2166         switch (vnic_dev_get_intr_mode(enic->vdev)) {
2167         default:
2168                 netif_napi_add(netdev, &enic->napi[0], enic_poll, 64);
2169                 break;
2170         case VNIC_DEV_INTR_MODE_MSIX:
2171                 for (i = 0; i < enic->rq_count; i++)
2172                         netif_napi_add(netdev, &enic->napi[i],
2173                                 enic_poll_msix, 64);
2174                 break;
2175         }
2176
2177         return 0;
2178
2179 err_out_free_vnic_resources:
2180         enic_clear_intr_mode(enic);
2181         enic_free_vnic_resources(enic);
2182
2183         return err;
2184 }
2185
2186 static void enic_iounmap(struct enic *enic)
2187 {
2188         unsigned int i;
2189
2190         for (i = 0; i < ARRAY_SIZE(enic->bar); i++)
2191                 if (enic->bar[i].vaddr)
2192                         iounmap(enic->bar[i].vaddr);
2193 }
2194
2195 static int enic_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2196 {
2197         struct device *dev = &pdev->dev;
2198         struct net_device *netdev;
2199         struct enic *enic;
2200         int using_dac = 0;
2201         unsigned int i;
2202         int err;
2203 #ifdef CONFIG_PCI_IOV
2204         int pos = 0;
2205 #endif
2206         int num_pps = 1;
2207
2208         /* Allocate net device structure and initialize.  Private
2209          * instance data is initialized to zero.
2210          */
2211
2212         netdev = alloc_etherdev_mqs(sizeof(struct enic),
2213                                     ENIC_RQ_MAX, ENIC_WQ_MAX);
2214         if (!netdev)
2215                 return -ENOMEM;
2216
2217         pci_set_drvdata(pdev, netdev);
2218
2219         SET_NETDEV_DEV(netdev, &pdev->dev);
2220
2221         enic = netdev_priv(netdev);
2222         enic->netdev = netdev;
2223         enic->pdev = pdev;
2224
2225         /* Setup PCI resources
2226          */
2227
2228         err = pci_enable_device_mem(pdev);
2229         if (err) {
2230                 dev_err(dev, "Cannot enable PCI device, aborting\n");
2231                 goto err_out_free_netdev;
2232         }
2233
2234         err = pci_request_regions(pdev, DRV_NAME);
2235         if (err) {
2236                 dev_err(dev, "Cannot request PCI regions, aborting\n");
2237                 goto err_out_disable_device;
2238         }
2239
2240         pci_set_master(pdev);
2241
2242         /* Query PCI controller on system for DMA addressing
2243          * limitation for the device.  Try 64-bit first, and
2244          * fail to 32-bit.
2245          */
2246
2247         err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
2248         if (err) {
2249                 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
2250                 if (err) {
2251                         dev_err(dev, "No usable DMA configuration, aborting\n");
2252                         goto err_out_release_regions;
2253                 }
2254                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
2255                 if (err) {
2256                         dev_err(dev, "Unable to obtain %u-bit DMA "
2257                                 "for consistent allocations, aborting\n", 32);
2258                         goto err_out_release_regions;
2259                 }
2260         } else {
2261                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
2262                 if (err) {
2263                         dev_err(dev, "Unable to obtain %u-bit DMA "
2264                                 "for consistent allocations, aborting\n", 64);
2265                         goto err_out_release_regions;
2266                 }
2267                 using_dac = 1;
2268         }
2269
2270         /* Map vNIC resources from BAR0-5
2271          */
2272
2273         for (i = 0; i < ARRAY_SIZE(enic->bar); i++) {
2274                 if (!(pci_resource_flags(pdev, i) & IORESOURCE_MEM))
2275                         continue;
2276                 enic->bar[i].len = pci_resource_len(pdev, i);
2277                 enic->bar[i].vaddr = pci_iomap(pdev, i, enic->bar[i].len);
2278                 if (!enic->bar[i].vaddr) {
2279                         dev_err(dev, "Cannot memory-map BAR %d, aborting\n", i);
2280                         err = -ENODEV;
2281                         goto err_out_iounmap;
2282                 }
2283                 enic->bar[i].bus_addr = pci_resource_start(pdev, i);
2284         }
2285
2286         /* Register vNIC device
2287          */
2288
2289         enic->vdev = vnic_dev_register(NULL, enic, pdev, enic->bar,
2290                 ARRAY_SIZE(enic->bar));
2291         if (!enic->vdev) {
2292                 dev_err(dev, "vNIC registration failed, aborting\n");
2293                 err = -ENODEV;
2294                 goto err_out_iounmap;
2295         }
2296
2297 #ifdef CONFIG_PCI_IOV
2298         /* Get number of subvnics */
2299         pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_SRIOV);
2300         if (pos) {
2301                 pci_read_config_word(pdev, pos + PCI_SRIOV_TOTAL_VF,
2302                         &enic->num_vfs);
2303                 if (enic->num_vfs) {
2304                         err = pci_enable_sriov(pdev, enic->num_vfs);
2305                         if (err) {
2306                                 dev_err(dev, "SRIOV enable failed, aborting."
2307                                         " pci_enable_sriov() returned %d\n",
2308                                         err);
2309                                 goto err_out_vnic_unregister;
2310                         }
2311                         enic->priv_flags |= ENIC_SRIOV_ENABLED;
2312                         num_pps = enic->num_vfs;
2313                 }
2314         }
2315 #endif
2316
2317         /* Allocate structure for port profiles */
2318         enic->pp = kcalloc(num_pps, sizeof(*enic->pp), GFP_KERNEL);
2319         if (!enic->pp) {
2320                 err = -ENOMEM;
2321                 goto err_out_disable_sriov_pp;
2322         }
2323
2324         /* Issue device open to get device in known state
2325          */
2326
2327         err = enic_dev_open(enic);
2328         if (err) {
2329                 dev_err(dev, "vNIC dev open failed, aborting\n");
2330                 goto err_out_disable_sriov;
2331         }
2332
2333         /* Setup devcmd lock
2334          */
2335
2336         spin_lock_init(&enic->devcmd_lock);
2337         spin_lock_init(&enic->enic_api_lock);
2338
2339         /*
2340          * Set ingress vlan rewrite mode before vnic initialization
2341          */
2342
2343         err = enic_dev_set_ig_vlan_rewrite_mode(enic);
2344         if (err) {
2345                 dev_err(dev,
2346                         "Failed to set ingress vlan rewrite mode, aborting.\n");
2347                 goto err_out_dev_close;
2348         }
2349
2350         /* Issue device init to initialize the vnic-to-switch link.
2351          * We'll start with carrier off and wait for link UP
2352          * notification later to turn on carrier.  We don't need
2353          * to wait here for the vnic-to-switch link initialization
2354          * to complete; link UP notification is the indication that
2355          * the process is complete.
2356          */
2357
2358         netif_carrier_off(netdev);
2359
2360         /* Do not call dev_init for a dynamic vnic.
2361          * For a dynamic vnic, init_prov_info will be
2362          * called later by an upper layer.
2363          */
2364
2365         if (!enic_is_dynamic(enic)) {
2366                 err = vnic_dev_init(enic->vdev, 0);
2367                 if (err) {
2368                         dev_err(dev, "vNIC dev init failed, aborting\n");
2369                         goto err_out_dev_close;
2370                 }
2371         }
2372
2373         err = enic_dev_init(enic);
2374         if (err) {
2375                 dev_err(dev, "Device initialization failed, aborting\n");
2376                 goto err_out_dev_close;
2377         }
2378
2379         netif_set_real_num_tx_queues(netdev, enic->wq_count);
2380         netif_set_real_num_rx_queues(netdev, enic->rq_count);
2381
2382         /* Setup notification timer, HW reset task, and wq locks
2383          */
2384
2385         init_timer(&enic->notify_timer);
2386         enic->notify_timer.function = enic_notify_timer;
2387         enic->notify_timer.data = (unsigned long)enic;
2388
2389         enic_set_rx_coal_setting(enic);
2390         INIT_WORK(&enic->reset, enic_reset);
2391         INIT_WORK(&enic->change_mtu_work, enic_change_mtu_work);
2392
2393         for (i = 0; i < enic->wq_count; i++)
2394                 spin_lock_init(&enic->wq_lock[i]);
2395
2396         /* Register net device
2397          */
2398
2399         enic->port_mtu = enic->config.mtu;
2400         (void)enic_change_mtu(netdev, enic->port_mtu);
2401
2402         err = enic_set_mac_addr(netdev, enic->mac_addr);
2403         if (err) {
2404                 dev_err(dev, "Invalid MAC address, aborting\n");
2405                 goto err_out_dev_deinit;
2406         }
2407
2408         enic->tx_coalesce_usecs = enic->config.intr_timer_usec;
2409         /* rx coalesce time already got initialized. This gets used
2410          * if adaptive coal is turned off
2411          */
2412         enic->rx_coalesce_usecs = enic->tx_coalesce_usecs;
2413
2414         if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic))
2415                 netdev->netdev_ops = &enic_netdev_dynamic_ops;
2416         else
2417                 netdev->netdev_ops = &enic_netdev_ops;
2418
2419         netdev->watchdog_timeo = 2 * HZ;
2420         enic_set_ethtool_ops(netdev);
2421
2422         netdev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
2423         if (ENIC_SETTING(enic, LOOP)) {
2424                 netdev->features &= ~NETIF_F_HW_VLAN_CTAG_TX;
2425                 enic->loop_enable = 1;
2426                 enic->loop_tag = enic->config.loop_tag;
2427                 dev_info(dev, "loopback tag=0x%04x\n", enic->loop_tag);
2428         }
2429         if (ENIC_SETTING(enic, TXCSUM))
2430                 netdev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM;
2431         if (ENIC_SETTING(enic, TSO))
2432                 netdev->hw_features |= NETIF_F_TSO |
2433                         NETIF_F_TSO6 | NETIF_F_TSO_ECN;
2434         if (ENIC_SETTING(enic, RSS))
2435                 netdev->hw_features |= NETIF_F_RXHASH;
2436         if (ENIC_SETTING(enic, RXCSUM))
2437                 netdev->hw_features |= NETIF_F_RXCSUM;
2438
2439         netdev->features |= netdev->hw_features;
2440
2441 #ifdef CONFIG_RFS_ACCEL
2442         netdev->hw_features |= NETIF_F_NTUPLE;
2443 #endif
2444
2445         if (using_dac)
2446                 netdev->features |= NETIF_F_HIGHDMA;
2447
2448         netdev->priv_flags |= IFF_UNICAST_FLT;
2449
2450         err = register_netdev(netdev);
2451         if (err) {
2452                 dev_err(dev, "Cannot register net device, aborting\n");
2453                 goto err_out_dev_deinit;
2454         }
2455
2456         return 0;
2457
2458 err_out_dev_deinit:
2459         enic_dev_deinit(enic);
2460 err_out_dev_close:
2461         vnic_dev_close(enic->vdev);
2462 err_out_disable_sriov:
2463         kfree(enic->pp);
2464 err_out_disable_sriov_pp:
2465 #ifdef CONFIG_PCI_IOV
2466         if (enic_sriov_enabled(enic)) {
2467                 pci_disable_sriov(pdev);
2468                 enic->priv_flags &= ~ENIC_SRIOV_ENABLED;
2469         }
2470 err_out_vnic_unregister:
2471 #endif
2472         vnic_dev_unregister(enic->vdev);
2473 err_out_iounmap:
2474         enic_iounmap(enic);
2475 err_out_release_regions:
2476         pci_release_regions(pdev);
2477 err_out_disable_device:
2478         pci_disable_device(pdev);
2479 err_out_free_netdev:
2480         free_netdev(netdev);
2481
2482         return err;
2483 }
2484
2485 static void enic_remove(struct pci_dev *pdev)
2486 {
2487         struct net_device *netdev = pci_get_drvdata(pdev);
2488
2489         if (netdev) {
2490                 struct enic *enic = netdev_priv(netdev);
2491
2492                 cancel_work_sync(&enic->reset);
2493                 cancel_work_sync(&enic->change_mtu_work);
2494                 unregister_netdev(netdev);
2495                 enic_dev_deinit(enic);
2496                 vnic_dev_close(enic->vdev);
2497 #ifdef CONFIG_PCI_IOV
2498                 if (enic_sriov_enabled(enic)) {
2499                         pci_disable_sriov(pdev);
2500                         enic->priv_flags &= ~ENIC_SRIOV_ENABLED;
2501                 }
2502 #endif
2503                 kfree(enic->pp);
2504                 vnic_dev_unregister(enic->vdev);
2505                 enic_iounmap(enic);
2506                 pci_release_regions(pdev);
2507                 pci_disable_device(pdev);
2508                 free_netdev(netdev);
2509         }
2510 }
2511
2512 static struct pci_driver enic_driver = {
2513         .name = DRV_NAME,
2514         .id_table = enic_id_table,
2515         .probe = enic_probe,
2516         .remove = enic_remove,
2517 };
2518
2519 static int __init enic_init_module(void)
2520 {
2521         pr_info("%s, ver %s\n", DRV_DESCRIPTION, DRV_VERSION);
2522
2523         return pci_register_driver(&enic_driver);
2524 }
2525
2526 static void __exit enic_cleanup_module(void)
2527 {
2528         pci_unregister_driver(&enic_driver);
2529 }
2530
2531 module_init(enic_init_module);
2532 module_exit(enic_cleanup_module);