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