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