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