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