drivers/net/ethernet/qlogic/qede/qede_main.c

   1 /* QLogic qede NIC Driver
   2 * Copyright (c) 2015 QLogic Corporation
   3 *
   4 * This software is available under the terms of the GNU General Public License
   5 * (GPL) Version 2, available from the file COPYING in the main directory of
   6 * this source tree.
   7 */
   8
   9 #include <linux/module.h>
  10 #include <linux/pci.h>
  11 #include <linux/version.h>
  12 #include <linux/device.h>
  13 #include <linux/netdevice.h>
  14 #include <linux/etherdevice.h>
  15 #include <linux/skbuff.h>
  16 #include <linux/errno.h>
  17 #include <linux/list.h>
  18 #include <linux/string.h>
  19 #include <linux/dma-mapping.h>
  20 #include <linux/interrupt.h>
  21 #include <asm/byteorder.h>
  22 #include <asm/param.h>
  23 #include <linux/io.h>
  24 #include <linux/netdev_features.h>
  25 #include <linux/udp.h>
  26 #include <linux/tcp.h>
  27 #include <net/vxlan.h>
  28 #include <linux/ip.h>
  29 #include <net/ipv6.h>
  30 #include <net/tcp.h>
  31 #include <linux/if_ether.h>
  32 #include <linux/if_vlan.h>
  33 #include <linux/pkt_sched.h>
  34 #include <linux/ethtool.h>
  35 #include <linux/in.h>
  36 #include <linux/random.h>
  37 #include <net/ip6_checksum.h>
  38 #include <linux/bitops.h>
  39
  40 #include "qede.h"
  41
  42 static char version[] =
  43         "QLogic FastLinQ 4xxxx Ethernet Driver qede " DRV_MODULE_VERSION "\n";
  44
  45 MODULE_DESCRIPTION("QLogic FastLinQ 4xxxx Ethernet Driver");
  46 MODULE_LICENSE("GPL");
  47 MODULE_VERSION(DRV_MODULE_VERSION);
  48
  49 static uint debug;
  50 module_param(debug, uint, 0);
  51 MODULE_PARM_DESC(debug, " Default debug msglevel");
  52
  53 static const struct qed_eth_ops *qed_ops;
  54
  55 #define CHIP_NUM_57980S_40              0x1634
  56 #define CHIP_NUM_57980S_10              0x1666
  57 #define CHIP_NUM_57980S_MF              0x1636
  58 #define CHIP_NUM_57980S_100             0x1644
  59 #define CHIP_NUM_57980S_50              0x1654
  60 #define CHIP_NUM_57980S_25              0x1656
  61
  62 #ifndef PCI_DEVICE_ID_NX2_57980E
  63 #define PCI_DEVICE_ID_57980S_40         CHIP_NUM_57980S_40
  64 #define PCI_DEVICE_ID_57980S_10         CHIP_NUM_57980S_10
  65 #define PCI_DEVICE_ID_57980S_MF         CHIP_NUM_57980S_MF
  66 #define PCI_DEVICE_ID_57980S_100        CHIP_NUM_57980S_100
  67 #define PCI_DEVICE_ID_57980S_50         CHIP_NUM_57980S_50
  68 #define PCI_DEVICE_ID_57980S_25         CHIP_NUM_57980S_25
  69 #endif
  70
  71 static const struct pci_device_id qede_pci_tbl[] = {
  72         { PCI_VDEVICE(QLOGIC, PCI_DEVICE_ID_57980S_40), 0 },
  73         { PCI_VDEVICE(QLOGIC, PCI_DEVICE_ID_57980S_10), 0 },
  74         { PCI_VDEVICE(QLOGIC, PCI_DEVICE_ID_57980S_MF), 0 },
  75         { PCI_VDEVICE(QLOGIC, PCI_DEVICE_ID_57980S_100), 0 },
  76         { PCI_VDEVICE(QLOGIC, PCI_DEVICE_ID_57980S_50), 0 },
  77         { PCI_VDEVICE(QLOGIC, PCI_DEVICE_ID_57980S_25), 0 },
  78         { 0 }
  79 };
  80
  81 MODULE_DEVICE_TABLE(pci, qede_pci_tbl);
  82
  83 static int qede_probe(struct pci_dev *pdev, const struct pci_device_id *id);
  84
  85 #define TX_TIMEOUT              (5 * HZ)
  86
  87 static void qede_remove(struct pci_dev *pdev);
  88 static int qede_alloc_rx_buffer(struct qede_dev *edev,
  89                                 struct qede_rx_queue *rxq);
  90 static void qede_link_update(void *dev, struct qed_link_output *link);
  91
  92 static struct pci_driver qede_pci_driver = {
  93         .name = "qede",
  94         .id_table = qede_pci_tbl,
  95         .probe = qede_probe,
  96         .remove = qede_remove,
  97 };
  98
  99 static struct qed_eth_cb_ops qede_ll_ops = {
 100         {
 101                 .link_update = qede_link_update,
 102         },
 103 };
 104
 105 static int qede_netdev_event(struct notifier_block *this, unsigned long event,
 106                              void *ptr)
 107 {
 108         struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
 109         struct ethtool_drvinfo drvinfo;
 110         struct qede_dev *edev;
 111
 112         /* Currently only support name change */
 113         if (event != NETDEV_CHANGENAME)
 114                 goto done;
 115
 116         /* Check whether this is a qede device */
 117         if (!ndev || !ndev->ethtool_ops || !ndev->ethtool_ops->get_drvinfo)
 118                 goto done;
 119
 120         memset(&drvinfo, 0, sizeof(drvinfo));
 121         ndev->ethtool_ops->get_drvinfo(ndev, &drvinfo);
 122         if (strcmp(drvinfo.driver, "qede"))
 123                 goto done;
 124         edev = netdev_priv(ndev);
 125
 126         /* Notify qed of the name change */
 127         if (!edev->ops || !edev->ops->common)
 128                 goto done;
 129         edev->ops->common->set_id(edev->cdev, edev->ndev->name,
 130                                   "qede");
 131
 132 done:
 133         return NOTIFY_DONE;
 134 }
 135
 136 static struct notifier_block qede_netdev_notifier = {
 137         .notifier_call = qede_netdev_event,
 138 };
 139
 140 static
 141 int __init qede_init(void)
 142 {
 143         int ret;
 144         u32 qed_ver;
 145
 146         pr_notice("qede_init: %s\n", version);
 147
 148         qed_ver = qed_get_protocol_version(QED_PROTOCOL_ETH);
 149         if (qed_ver !=  QEDE_ETH_INTERFACE_VERSION) {
 150                 pr_notice("Version mismatch [%08x != %08x]\n",
 151                           qed_ver,
 152                           QEDE_ETH_INTERFACE_VERSION);
 153                 return -EINVAL;
 154         }
 155
 156         qed_ops = qed_get_eth_ops(QEDE_ETH_INTERFACE_VERSION);
 157         if (!qed_ops) {
 158                 pr_notice("Failed to get qed ethtool operations\n");
 159                 return -EINVAL;
 160         }
 161
 162         /* Must register notifier before pci ops, since we might miss
 163          * interface rename after pci probe and netdev registeration.
 164          */
 165         ret = register_netdevice_notifier(&qede_netdev_notifier);
 166         if (ret) {
 167                 pr_notice("Failed to register netdevice_notifier\n");
 168                 qed_put_eth_ops();
 169                 return -EINVAL;
 170         }
 171
 172         ret = pci_register_driver(&qede_pci_driver);
 173         if (ret) {
 174                 pr_notice("Failed to register driver\n");
 175                 unregister_netdevice_notifier(&qede_netdev_notifier);
 176                 qed_put_eth_ops();
 177                 return -EINVAL;
 178         }
 179
 180         return 0;
 181 }
 182
 183 static void __exit qede_cleanup(void)
 184 {
 185         pr_notice("qede_cleanup called\n");
 186
 187         unregister_netdevice_notifier(&qede_netdev_notifier);
 188         pci_unregister_driver(&qede_pci_driver);
 189         qed_put_eth_ops();
 190 }
 191
 192 module_init(qede_init);
 193 module_exit(qede_cleanup);
 194
 195 /* -------------------------------------------------------------------------
 196  * START OF FAST-PATH
 197  * -------------------------------------------------------------------------
 198  */
 199
 200 /* Unmap the data and free skb */
 201 static int qede_free_tx_pkt(struct qede_dev *edev,
 202                             struct qede_tx_queue *txq,
 203                             int *len)
 204 {
 205         u16 idx = txq->sw_tx_cons & NUM_TX_BDS_MAX;
 206         struct sk_buff *skb = txq->sw_tx_ring[idx].skb;
 207         struct eth_tx_1st_bd *first_bd;
 208         struct eth_tx_bd *tx_data_bd;
 209         int bds_consumed = 0;
 210         int nbds;
 211         bool data_split = txq->sw_tx_ring[idx].flags & QEDE_TSO_SPLIT_BD;
 212         int i, split_bd_len = 0;
 213
 214         if (unlikely(!skb)) {
 215                 DP_ERR(edev,
 216                        "skb is null for txq idx=%d txq->sw_tx_cons=%d txq->sw_tx_prod=%d\n",
 217                        idx, txq->sw_tx_cons, txq->sw_tx_prod);
 218                 return -1;
 219         }
 220
 221         *len = skb->len;
 222
 223         first_bd = (struct eth_tx_1st_bd *)qed_chain_consume(&txq->tx_pbl);
 224
 225         bds_consumed++;
 226
 227         nbds = first_bd->data.nbds;
 228
 229         if (data_split) {
 230                 struct eth_tx_bd *split = (struct eth_tx_bd *)
 231                         qed_chain_consume(&txq->tx_pbl);
 232                 split_bd_len = BD_UNMAP_LEN(split);
 233                 bds_consumed++;
 234         }
 235         dma_unmap_page(&edev->pdev->dev, BD_UNMAP_ADDR(first_bd),
 236                        BD_UNMAP_LEN(first_bd) + split_bd_len, DMA_TO_DEVICE);
 237
 238         /* Unmap the data of the skb frags */
 239         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++, bds_consumed++) {
 240                 tx_data_bd = (struct eth_tx_bd *)
 241                         qed_chain_consume(&txq->tx_pbl);
 242                 dma_unmap_page(&edev->pdev->dev, BD_UNMAP_ADDR(tx_data_bd),
 243                                BD_UNMAP_LEN(tx_data_bd), DMA_TO_DEVICE);
 244         }
 245
 246         while (bds_consumed++ < nbds)
 247                 qed_chain_consume(&txq->tx_pbl);
 248
 249         /* Free skb */
 250         dev_kfree_skb_any(skb);
 251         txq->sw_tx_ring[idx].skb = NULL;
 252         txq->sw_tx_ring[idx].flags = 0;
 253
 254         return 0;
 255 }
 256
 257 /* Unmap the data and free skb when mapping failed during start_xmit */
 258 static void qede_free_failed_tx_pkt(struct qede_dev *edev,
 259                                     struct qede_tx_queue *txq,
 260                                     struct eth_tx_1st_bd *first_bd,
 261                                     int nbd,
 262                                     bool data_split)
 263 {
 264         u16 idx = txq->sw_tx_prod & NUM_TX_BDS_MAX;
 265         struct sk_buff *skb = txq->sw_tx_ring[idx].skb;
 266         struct eth_tx_bd *tx_data_bd;
 267         int i, split_bd_len = 0;
 268
 269         /* Return prod to its position before this skb was handled */
 270         qed_chain_set_prod(&txq->tx_pbl,
 271                            le16_to_cpu(txq->tx_db.data.bd_prod),
 272                            first_bd);
 273
 274         first_bd = (struct eth_tx_1st_bd *)qed_chain_produce(&txq->tx_pbl);
 275
 276         if (data_split) {
 277                 struct eth_tx_bd *split = (struct eth_tx_bd *)
 278                                           qed_chain_produce(&txq->tx_pbl);
 279                 split_bd_len = BD_UNMAP_LEN(split);
 280                 nbd--;
 281         }
 282
 283         dma_unmap_page(&edev->pdev->dev, BD_UNMAP_ADDR(first_bd),
 284                        BD_UNMAP_LEN(first_bd) + split_bd_len, DMA_TO_DEVICE);
 285
 286         /* Unmap the data of the skb frags */
 287         for (i = 0; i < nbd; i++) {
 288                 tx_data_bd = (struct eth_tx_bd *)
 289                         qed_chain_produce(&txq->tx_pbl);
 290                 if (tx_data_bd->nbytes)
 291                         dma_unmap_page(&edev->pdev->dev,
 292                                        BD_UNMAP_ADDR(tx_data_bd),
 293                                        BD_UNMAP_LEN(tx_data_bd), DMA_TO_DEVICE);
 294         }
 295
 296         /* Return again prod to its position before this skb was handled */
 297         qed_chain_set_prod(&txq->tx_pbl,
 298                            le16_to_cpu(txq->tx_db.data.bd_prod),
 299                            first_bd);
 300
 301         /* Free skb */
 302         dev_kfree_skb_any(skb);
 303         txq->sw_tx_ring[idx].skb = NULL;
 304         txq->sw_tx_ring[idx].flags = 0;
 305 }
 306
 307 static u32 qede_xmit_type(struct qede_dev *edev,
 308                           struct sk_buff *skb,
 309                           int *ipv6_ext)
 310 {
 311         u32 rc = XMIT_L4_CSUM;
 312         __be16 l3_proto;
 313
 314         if (skb->ip_summed != CHECKSUM_PARTIAL)
 315                 return XMIT_PLAIN;
 316
 317         l3_proto = vlan_get_protocol(skb);
 318         if (l3_proto == htons(ETH_P_IPV6) &&
 319             (ipv6_hdr(skb)->nexthdr == NEXTHDR_IPV6))
 320                 *ipv6_ext = 1;
 321
 322         if (skb_is_gso(skb))
 323                 rc |= XMIT_LSO;
 324
 325         return rc;
 326 }
 327
 328 static void qede_set_params_for_ipv6_ext(struct sk_buff *skb,
 329                                          struct eth_tx_2nd_bd *second_bd,
 330                                          struct eth_tx_3rd_bd *third_bd)
 331 {
 332         u8 l4_proto;
 333         u16 bd2_bits1 = 0, bd2_bits2 = 0;
 334
 335         bd2_bits1 |= (1 << ETH_TX_DATA_2ND_BD_IPV6_EXT_SHIFT);
 336
 337         bd2_bits2 |= ((((u8 *)skb_transport_header(skb) - skb->data) >> 1) &
 338                      ETH_TX_DATA_2ND_BD_L4_HDR_START_OFFSET_W_MASK)
 339                     << ETH_TX_DATA_2ND_BD_L4_HDR_START_OFFSET_W_SHIFT;
 340
 341         bd2_bits1 |= (ETH_L4_PSEUDO_CSUM_CORRECT_LENGTH <<
 342                       ETH_TX_DATA_2ND_BD_L4_PSEUDO_CSUM_MODE_SHIFT);
 343
 344         if (vlan_get_protocol(skb) == htons(ETH_P_IPV6))
 345                 l4_proto = ipv6_hdr(skb)->nexthdr;
 346         else
 347                 l4_proto = ip_hdr(skb)->protocol;
 348
 349         if (l4_proto == IPPROTO_UDP)
 350                 bd2_bits1 |= 1 << ETH_TX_DATA_2ND_BD_L4_UDP_SHIFT;
 351
 352         if (third_bd)
 353                 third_bd->data.bitfields |=
 354                         cpu_to_le16(((tcp_hdrlen(skb) / 4) &
 355                                 ETH_TX_DATA_3RD_BD_TCP_HDR_LEN_DW_MASK) <<
 356                                 ETH_TX_DATA_3RD_BD_TCP_HDR_LEN_DW_SHIFT);
 357
 358         second_bd->data.bitfields1 = cpu_to_le16(bd2_bits1);
 359         second_bd->data.bitfields2 = cpu_to_le16(bd2_bits2);
 360 }
 361
 362 static int map_frag_to_bd(struct qede_dev *edev,
 363                           skb_frag_t *frag,
 364                           struct eth_tx_bd *bd)
 365 {
 366         dma_addr_t mapping;
 367
 368         /* Map skb non-linear frag data for DMA */
 369         mapping = skb_frag_dma_map(&edev->pdev->dev, frag, 0,
 370                                    skb_frag_size(frag),
 371                                    DMA_TO_DEVICE);
 372         if (unlikely(dma_mapping_error(&edev->pdev->dev, mapping))) {
 373                 DP_NOTICE(edev, "Unable to map frag - dropping packet\n");
 374                 return -ENOMEM;
 375         }
 376
 377         /* Setup the data pointer of the frag data */
 378         BD_SET_UNMAP_ADDR_LEN(bd, mapping, skb_frag_size(frag));
 379
 380         return 0;
 381 }
 382
 383 /* +2 for 1st BD for headers and 2nd BD for headlen (if required) */
 384 #if ((MAX_SKB_FRAGS + 2) > ETH_TX_MAX_BDS_PER_NON_LSO_PACKET)
 385 static bool qede_pkt_req_lin(struct qede_dev *edev, struct sk_buff *skb,
 386                              u8 xmit_type)
 387 {
 388         int allowed_frags = ETH_TX_MAX_BDS_PER_NON_LSO_PACKET - 1;
 389
 390         if (xmit_type & XMIT_LSO) {
 391                 int hlen;
 392
 393                 hlen = skb_transport_header(skb) +
 394                        tcp_hdrlen(skb) - skb->data;
 395
 396                 /* linear payload would require its own BD */
 397                 if (skb_headlen(skb) > hlen)
 398                         allowed_frags--;
 399         }
 400
 401         return (skb_shinfo(skb)->nr_frags > allowed_frags);
 402 }
 403 #endif
 404
 405 /* Main transmit function */
 406 static
 407 netdev_tx_t qede_start_xmit(struct sk_buff *skb,
 408                             struct net_device *ndev)
 409 {
 410         struct qede_dev *edev = netdev_priv(ndev);
 411         struct netdev_queue *netdev_txq;
 412         struct qede_tx_queue *txq;
 413         struct eth_tx_1st_bd *first_bd;
 414         struct eth_tx_2nd_bd *second_bd = NULL;
 415         struct eth_tx_3rd_bd *third_bd = NULL;
 416         struct eth_tx_bd *tx_data_bd = NULL;
 417         u16 txq_index;
 418         u8 nbd = 0;
 419         dma_addr_t mapping;
 420         int rc, frag_idx = 0, ipv6_ext = 0;
 421         u8 xmit_type;
 422         u16 idx;
 423         u16 hlen;
 424         bool data_split;
 425
 426         /* Get tx-queue context and netdev index */
 427         txq_index = skb_get_queue_mapping(skb);
 428         WARN_ON(txq_index >= QEDE_TSS_CNT(edev));
 429         txq = QEDE_TX_QUEUE(edev, txq_index);
 430         netdev_txq = netdev_get_tx_queue(ndev, txq_index);
 431
 432         WARN_ON(qed_chain_get_elem_left(&txq->tx_pbl) <
 433                                (MAX_SKB_FRAGS + 1));
 434
 435         xmit_type = qede_xmit_type(edev, skb, &ipv6_ext);
 436
 437 #if ((MAX_SKB_FRAGS + 2) > ETH_TX_MAX_BDS_PER_NON_LSO_PACKET)
 438         if (qede_pkt_req_lin(edev, skb, xmit_type)) {
 439                 if (skb_linearize(skb)) {
 440                         DP_NOTICE(edev,
 441                                   "SKB linearization failed - silently dropping this SKB\n");
 442                         dev_kfree_skb_any(skb);
 443                         return NETDEV_TX_OK;
 444                 }
 445         }
 446 #endif
 447
 448         /* Fill the entry in the SW ring and the BDs in the FW ring */
 449         idx = txq->sw_tx_prod & NUM_TX_BDS_MAX;
 450         txq->sw_tx_ring[idx].skb = skb;
 451         first_bd = (struct eth_tx_1st_bd *)
 452                    qed_chain_produce(&txq->tx_pbl);
 453         memset(first_bd, 0, sizeof(*first_bd));
 454         first_bd->data.bd_flags.bitfields =
 455                 1 << ETH_TX_1ST_BD_FLAGS_START_BD_SHIFT;
 456
 457         /* Map skb linear data for DMA and set in the first BD */
 458         mapping = dma_map_single(&edev->pdev->dev, skb->data,
 459                                  skb_headlen(skb), DMA_TO_DEVICE);
 460         if (unlikely(dma_mapping_error(&edev->pdev->dev, mapping))) {
 461                 DP_NOTICE(edev, "SKB mapping failed\n");
 462                 qede_free_failed_tx_pkt(edev, txq, first_bd, 0, false);
 463                 return NETDEV_TX_OK;
 464         }
 465         nbd++;
 466         BD_SET_UNMAP_ADDR_LEN(first_bd, mapping, skb_headlen(skb));
 467
 468         /* In case there is IPv6 with extension headers or LSO we need 2nd and
 469          * 3rd BDs.
 470          */
 471         if (unlikely((xmit_type & XMIT_LSO) | ipv6_ext)) {
 472                 second_bd = (struct eth_tx_2nd_bd *)
 473                         qed_chain_produce(&txq->tx_pbl);
 474                 memset(second_bd, 0, sizeof(*second_bd));
 475
 476                 nbd++;
 477                 third_bd = (struct eth_tx_3rd_bd *)
 478                         qed_chain_produce(&txq->tx_pbl);
 479                 memset(third_bd, 0, sizeof(*third_bd));
 480
 481                 nbd++;
 482                 /* We need to fill in additional data in second_bd... */
 483                 tx_data_bd = (struct eth_tx_bd *)second_bd;
 484         }
 485
 486         if (skb_vlan_tag_present(skb)) {
 487                 first_bd->data.vlan = cpu_to_le16(skb_vlan_tag_get(skb));
 488                 first_bd->data.bd_flags.bitfields |=
 489                         1 << ETH_TX_1ST_BD_FLAGS_VLAN_INSERTION_SHIFT;
 490         }
 491
 492         /* Fill the parsing flags & params according to the requested offload */
 493         if (xmit_type & XMIT_L4_CSUM) {
 494                 u16 temp = 1 << ETH_TX_DATA_1ST_BD_TUNN_CFG_OVERRIDE_SHIFT;
 495
 496                 /* We don't re-calculate IP checksum as it is already done by
 497                  * the upper stack
 498                  */
 499                 first_bd->data.bd_flags.bitfields |=
 500                         1 << ETH_TX_1ST_BD_FLAGS_L4_CSUM_SHIFT;
 501
 502                 first_bd->data.bitfields |= cpu_to_le16(temp);
 503
 504                 /* If the packet is IPv6 with extension header, indicate that
 505                  * to FW and pass few params, since the device cracker doesn't
 506                  * support parsing IPv6 with extension header/s.
 507                  */
 508                 if (unlikely(ipv6_ext))
 509                         qede_set_params_for_ipv6_ext(skb, second_bd, third_bd);
 510         }
 511
 512         if (xmit_type & XMIT_LSO) {
 513                 first_bd->data.bd_flags.bitfields |=
 514                         (1 << ETH_TX_1ST_BD_FLAGS_LSO_SHIFT);
 515                 third_bd->data.lso_mss =
 516                         cpu_to_le16(skb_shinfo(skb)->gso_size);
 517
 518                 first_bd->data.bd_flags.bitfields |=
 519                 1 << ETH_TX_1ST_BD_FLAGS_IP_CSUM_SHIFT;
 520                 hlen = skb_transport_header(skb) +
 521                        tcp_hdrlen(skb) - skb->data;
 522
 523                 /* @@@TBD - if will not be removed need to check */
 524                 third_bd->data.bitfields |=
 525                         cpu_to_le16((1 << ETH_TX_DATA_3RD_BD_HDR_NBD_SHIFT));
 526
 527                 /* Make life easier for FW guys who can't deal with header and
 528                  * data on same BD. If we need to split, use the second bd...
 529                  */
 530                 if (unlikely(skb_headlen(skb) > hlen)) {
 531                         DP_VERBOSE(edev, NETIF_MSG_TX_QUEUED,
 532                                    "TSO split header size is %d (%x:%x)\n",
 533                                    first_bd->nbytes, first_bd->addr.hi,
 534                                    first_bd->addr.lo);
 535
 536                         mapping = HILO_U64(le32_to_cpu(first_bd->addr.hi),
 537                                            le32_to_cpu(first_bd->addr.lo)) +
 538                                            hlen;
 539
 540                         BD_SET_UNMAP_ADDR_LEN(tx_data_bd, mapping,
 541                                               le16_to_cpu(first_bd->nbytes) -
 542                                               hlen);
 543
 544                         /* this marks the BD as one that has no
 545                          * individual mapping
 546                          */
 547                         txq->sw_tx_ring[idx].flags |= QEDE_TSO_SPLIT_BD;
 548
 549                         first_bd->nbytes = cpu_to_le16(hlen);
 550
 551                         tx_data_bd = (struct eth_tx_bd *)third_bd;
 552                         data_split = true;
 553                 }
 554         }
 555
 556         /* Handle fragmented skb */
 557         /* special handle for frags inside 2nd and 3rd bds.. */
 558         while (tx_data_bd && frag_idx < skb_shinfo(skb)->nr_frags) {
 559                 rc = map_frag_to_bd(edev,
 560                                     &skb_shinfo(skb)->frags[frag_idx],
 561                                     tx_data_bd);
 562                 if (rc) {
 563                         qede_free_failed_tx_pkt(edev, txq, first_bd, nbd,
 564                                                 data_split);
 565                         return NETDEV_TX_OK;
 566                 }
 567
 568                 if (tx_data_bd == (struct eth_tx_bd *)second_bd)
 569                         tx_data_bd = (struct eth_tx_bd *)third_bd;
 570                 else
 571                         tx_data_bd = NULL;
 572
 573                 frag_idx++;
 574         }
 575
 576         /* map last frags into 4th, 5th .... */
 577         for (; frag_idx < skb_shinfo(skb)->nr_frags; frag_idx++, nbd++) {
 578                 tx_data_bd = (struct eth_tx_bd *)
 579                              qed_chain_produce(&txq->tx_pbl);
 580
 581                 memset(tx_data_bd, 0, sizeof(*tx_data_bd));
 582
 583                 rc = map_frag_to_bd(edev,
 584                                     &skb_shinfo(skb)->frags[frag_idx],
 585                                     tx_data_bd);
 586                 if (rc) {
 587                         qede_free_failed_tx_pkt(edev, txq, first_bd, nbd,
 588                                                 data_split);
 589                         return NETDEV_TX_OK;
 590                 }
 591         }
 592
 593         /* update the first BD with the actual num BDs */
 594         first_bd->data.nbds = nbd;
 595
 596         netdev_tx_sent_queue(netdev_txq, skb->len);
 597
 598         skb_tx_timestamp(skb);
 599
 600         /* Advance packet producer only before sending the packet since mapping
 601          * of pages may fail.
 602          */
 603         txq->sw_tx_prod++;
 604
 605         /* 'next page' entries are counted in the producer value */
 606         txq->tx_db.data.bd_prod =
 607                 cpu_to_le16(qed_chain_get_prod_idx(&txq->tx_pbl));
 608
 609         /* wmb makes sure that the BDs data is updated before updating the
 610          * producer, otherwise FW may read old data from the BDs.
 611          */
 612         wmb();
 613         barrier();
 614         writel(txq->tx_db.raw, txq->doorbell_addr);
 615
 616         /* mmiowb is needed to synchronize doorbell writes from more than one
 617          * processor. It guarantees that the write arrives to the device before
 618          * the queue lock is released and another start_xmit is called (possibly
 619          * on another CPU). Without this barrier, the next doorbell can bypass
 620          * this doorbell. This is applicable to IA64/Altix systems.
 621          */
 622         mmiowb();
 623
 624         if (unlikely(qed_chain_get_elem_left(&txq->tx_pbl)
 625                       < (MAX_SKB_FRAGS + 1))) {
 626                 netif_tx_stop_queue(netdev_txq);
 627                 DP_VERBOSE(edev, NETIF_MSG_TX_QUEUED,
 628                            "Stop queue was called\n");
 629                 /* paired memory barrier is in qede_tx_int(), we have to keep
 630                  * ordering of set_bit() in netif_tx_stop_queue() and read of
 631                  * fp->bd_tx_cons
 632                  */
 633                 smp_mb();
 634
 635                 if (qed_chain_get_elem_left(&txq->tx_pbl)
 636                      >= (MAX_SKB_FRAGS + 1) &&
 637                     (edev->state == QEDE_STATE_OPEN)) {
 638                         netif_tx_wake_queue(netdev_txq);
 639                         DP_VERBOSE(edev, NETIF_MSG_TX_QUEUED,
 640                                    "Wake queue was called\n");
 641                 }
 642         }
 643
 644         return NETDEV_TX_OK;
 645 }
 646
 647 static int qede_txq_has_work(struct qede_tx_queue *txq)
 648 {
 649         u16 hw_bd_cons;
 650
 651         /* Tell compiler that consumer and producer can change */
 652         barrier();
 653         hw_bd_cons = le16_to_cpu(*txq->hw_cons_ptr);
 654         if (qed_chain_get_cons_idx(&txq->tx_pbl) == hw_bd_cons + 1)
 655                 return 0;
 656
 657         return hw_bd_cons != qed_chain_get_cons_idx(&txq->tx_pbl);
 658 }
 659
 660 static int qede_tx_int(struct qede_dev *edev,
 661                        struct qede_tx_queue *txq)
 662 {
 663         struct netdev_queue *netdev_txq;
 664         u16 hw_bd_cons;
 665         unsigned int pkts_compl = 0, bytes_compl = 0;
 666         int rc;
 667
 668         netdev_txq = netdev_get_tx_queue(edev->ndev, txq->index);
 669
 670         hw_bd_cons = le16_to_cpu(*txq->hw_cons_ptr);
 671         barrier();
 672
 673         while (hw_bd_cons != qed_chain_get_cons_idx(&txq->tx_pbl)) {
 674                 int len = 0;
 675
 676                 rc = qede_free_tx_pkt(edev, txq, &len);
 677                 if (rc) {
 678                         DP_NOTICE(edev, "hw_bd_cons = %d, chain_cons=%d\n",
 679                                   hw_bd_cons,
 680                                   qed_chain_get_cons_idx(&txq->tx_pbl));
 681                         break;
 682                 }
 683
 684                 bytes_compl += len;
 685                 pkts_compl++;
 686                 txq->sw_tx_cons++;
 687         }
 688
 689         netdev_tx_completed_queue(netdev_txq, pkts_compl, bytes_compl);
 690
 691         /* Need to make the tx_bd_cons update visible to start_xmit()
 692          * before checking for netif_tx_queue_stopped().  Without the
 693          * memory barrier, there is a small possibility that
 694          * start_xmit() will miss it and cause the queue to be stopped
 695          * forever.
 696          * On the other hand we need an rmb() here to ensure the proper
 697          * ordering of bit testing in the following
 698          * netif_tx_queue_stopped(txq) call.
 699          */
 700         smp_mb();
 701
 702         if (unlikely(netif_tx_queue_stopped(netdev_txq))) {
 703                 /* Taking tx_lock is needed to prevent reenabling the queue
 704                  * while it's empty. This could have happen if rx_action() gets
 705                  * suspended in qede_tx_int() after the condition before
 706                  * netif_tx_wake_queue(), while tx_action (qede_start_xmit()):
 707                  *
 708                  * stops the queue->sees fresh tx_bd_cons->releases the queue->
 709                  * sends some packets consuming the whole queue again->
 710                  * stops the queue
 711                  */
 712
 713                 __netif_tx_lock(netdev_txq, smp_processor_id());
 714
 715                 if ((netif_tx_queue_stopped(netdev_txq)) &&
 716                     (edev->state == QEDE_STATE_OPEN) &&
 717                     (qed_chain_get_elem_left(&txq->tx_pbl)
 718                       >= (MAX_SKB_FRAGS + 1))) {
 719                         netif_tx_wake_queue(netdev_txq);
 720                         DP_VERBOSE(edev, NETIF_MSG_TX_DONE,
 721                                    "Wake queue was called\n");
 722                 }
 723
 724                 __netif_tx_unlock(netdev_txq);
 725         }
 726
 727         return 0;
 728 }
 729
 730 static bool qede_has_rx_work(struct qede_rx_queue *rxq)
 731 {
 732         u16 hw_comp_cons, sw_comp_cons;
 733
 734         /* Tell compiler that status block fields can change */
 735         barrier();
 736
 737         hw_comp_cons = le16_to_cpu(*rxq->hw_cons_ptr);
 738         sw_comp_cons = qed_chain_get_cons_idx(&rxq->rx_comp_ring);
 739
 740         return hw_comp_cons != sw_comp_cons;
 741 }
 742
 743 static bool qede_has_tx_work(struct qede_fastpath *fp)
 744 {
 745         u8 tc;
 746
 747         for (tc = 0; tc < fp->edev->num_tc; tc++)
 748                 if (qede_txq_has_work(&fp->txqs[tc]))
 749                         return true;
 750         return false;
 751 }
 752
 753 static inline void qede_rx_bd_ring_consume(struct qede_rx_queue *rxq)
 754 {
 755         qed_chain_consume(&rxq->rx_bd_ring);
 756         rxq->sw_rx_cons++;
 757 }
 758
 759 /* This function reuses the buffer(from an offset) from
 760  * consumer index to producer index in the bd ring
 761  */
 762 static inline void qede_reuse_page(struct qede_dev *edev,
 763                                    struct qede_rx_queue *rxq,
 764                                    struct sw_rx_data *curr_cons)
 765 {
 766         struct eth_rx_bd *rx_bd_prod = qed_chain_produce(&rxq->rx_bd_ring);
 767         struct sw_rx_data *curr_prod;
 768         dma_addr_t new_mapping;
 769
 770         curr_prod = &rxq->sw_rx_ring[rxq->sw_rx_prod & NUM_RX_BDS_MAX];
 771         *curr_prod = *curr_cons;
 772
 773         new_mapping = curr_prod->mapping + curr_prod->page_offset;
 774
 775         rx_bd_prod->addr.hi = cpu_to_le32(upper_32_bits(new_mapping));
 776         rx_bd_prod->addr.lo = cpu_to_le32(lower_32_bits(new_mapping));
 777
 778         rxq->sw_rx_prod++;
 779         curr_cons->data = NULL;
 780 }
 781
 782 /* In case of allocation failures reuse buffers
 783  * from consumer index to produce buffers for firmware
 784  */
 785 static void qede_recycle_rx_bd_ring(struct qede_rx_queue *rxq,
 786                                     struct qede_dev *edev, u8 count)
 787 {
 788         struct sw_rx_data *curr_cons;
 789
 790         for (; count > 0; count--) {
 791                 curr_cons = &rxq->sw_rx_ring[rxq->sw_rx_cons & NUM_RX_BDS_MAX];
 792                 qede_reuse_page(edev, rxq, curr_cons);
 793                 qede_rx_bd_ring_consume(rxq);
 794         }
 795 }
 796
 797 static inline int qede_realloc_rx_buffer(struct qede_dev *edev,
 798                                          struct qede_rx_queue *rxq,
 799                                          struct sw_rx_data *curr_cons)
 800 {
 801         /* Move to the next segment in the page */
 802         curr_cons->page_offset += rxq->rx_buf_seg_size;
 803
 804         if (curr_cons->page_offset == PAGE_SIZE) {
 805                 if (unlikely(qede_alloc_rx_buffer(edev, rxq))) {
 806                         /* Since we failed to allocate new buffer
 807                          * current buffer can be used again.
 808                          */
 809                         curr_cons->page_offset -= rxq->rx_buf_seg_size;
 810
 811                         return -ENOMEM;
 812                 }
 813
 814                 dma_unmap_page(&edev->pdev->dev, curr_cons->mapping,
 815                                PAGE_SIZE, DMA_FROM_DEVICE);
 816         } else {
 817                 /* Increment refcount of the page as we don't want
 818                  * network stack to take the ownership of the page
 819                  * which can be recycled multiple times by the driver.
 820                  */
 821                 atomic_inc(&curr_cons->data->_count);
 822                 qede_reuse_page(edev, rxq, curr_cons);
 823         }
 824
 825         return 0;
 826 }
 827
 828 static inline void qede_update_rx_prod(struct qede_dev *edev,
 829                                        struct qede_rx_queue *rxq)
 830 {
 831         u16 bd_prod = qed_chain_get_prod_idx(&rxq->rx_bd_ring);
 832         u16 cqe_prod = qed_chain_get_prod_idx(&rxq->rx_comp_ring);
 833         struct eth_rx_prod_data rx_prods = {0};
 834
 835         /* Update producers */
 836         rx_prods.bd_prod = cpu_to_le16(bd_prod);
 837         rx_prods.cqe_prod = cpu_to_le16(cqe_prod);
 838
 839         /* Make sure that the BD and SGE data is updated before updating the
 840          * producers since FW might read the BD/SGE right after the producer
 841          * is updated.
 842          */
 843         wmb();
 844
 845         internal_ram_wr(rxq->hw_rxq_prod_addr, sizeof(rx_prods),
 846                         (u32 *)&rx_prods);
 847
 848         /* mmiowb is needed to synchronize doorbell writes from more than one
 849          * processor. It guarantees that the write arrives to the device before
 850          * the napi lock is released and another qede_poll is called (possibly
 851          * on another CPU). Without this barrier, the next doorbell can bypass
 852          * this doorbell. This is applicable to IA64/Altix systems.
 853          */
 854         mmiowb();
 855 }
 856
 857 static u32 qede_get_rxhash(struct qede_dev *edev,
 858                            u8 bitfields,
 859                            __le32 rss_hash,
 860                            enum pkt_hash_types *rxhash_type)
 861 {
 862         enum rss_hash_type htype;
 863
 864         htype = GET_FIELD(bitfields, ETH_FAST_PATH_RX_REG_CQE_RSS_HASH_TYPE);
 865
 866         if ((edev->ndev->features & NETIF_F_RXHASH) && htype) {
 867                 *rxhash_type = ((htype == RSS_HASH_TYPE_IPV4) ||
 868                                 (htype == RSS_HASH_TYPE_IPV6)) ?
 869                                 PKT_HASH_TYPE_L3 : PKT_HASH_TYPE_L4;
 870                 return le32_to_cpu(rss_hash);
 871         }
 872         *rxhash_type = PKT_HASH_TYPE_NONE;
 873         return 0;
 874 }
 875
 876 static void qede_set_skb_csum(struct sk_buff *skb, u8 csum_flag)
 877 {
 878         skb_checksum_none_assert(skb);
 879
 880         if (csum_flag & QEDE_CSUM_UNNECESSARY)
 881                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 882 }
 883
 884 static inline void qede_skb_receive(struct qede_dev *edev,
 885                                     struct qede_fastpath *fp,
 886                                     struct sk_buff *skb,
 887                                     u16 vlan_tag)
 888 {
 889         if (vlan_tag)
 890                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
 891                                        vlan_tag);
 892
 893         napi_gro_receive(&fp->napi, skb);
 894 }
 895
 896 static void qede_set_gro_params(struct qede_dev *edev,
 897                                 struct sk_buff *skb,
 898                                 struct eth_fast_path_rx_tpa_start_cqe *cqe)
 899 {
 900         u16 parsing_flags = le16_to_cpu(cqe->pars_flags.flags);
 901
 902         if (((parsing_flags >> PARSING_AND_ERR_FLAGS_L3TYPE_SHIFT) &
 903             PARSING_AND_ERR_FLAGS_L3TYPE_MASK) == 2)
 904                 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
 905         else
 906                 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
 907
 908         skb_shinfo(skb)->gso_size = __le16_to_cpu(cqe->len_on_first_bd) -
 909                                         cqe->header_len;
 910 }
 911
 912 static int qede_fill_frag_skb(struct qede_dev *edev,
 913                               struct qede_rx_queue *rxq,
 914                               u8 tpa_agg_index,
 915                               u16 len_on_bd)
 916 {
 917         struct sw_rx_data *current_bd = &rxq->sw_rx_ring[rxq->sw_rx_cons &
 918                                                          NUM_RX_BDS_MAX];
 919         struct qede_agg_info *tpa_info = &rxq->tpa_info[tpa_agg_index];
 920         struct sk_buff *skb = tpa_info->skb;
 921
 922         if (unlikely(tpa_info->agg_state != QEDE_AGG_STATE_START))
 923                 goto out;
 924
 925         /* Add one frag and update the appropriate fields in the skb */
 926         skb_fill_page_desc(skb, tpa_info->frag_id++,
 927                            current_bd->data, current_bd->page_offset,
 928                            len_on_bd);
 929
 930         if (unlikely(qede_realloc_rx_buffer(edev, rxq, current_bd))) {
 931                 /* Incr page ref count to reuse on allocation failure
 932                  * so that it doesn't get freed while freeing SKB.
 933                  */
 934                 atomic_inc(&current_bd->data->_count);
 935                 goto out;
 936         }
 937
 938         qed_chain_consume(&rxq->rx_bd_ring);
 939         rxq->sw_rx_cons++;
 940
 941         skb->data_len += len_on_bd;
 942         skb->truesize += rxq->rx_buf_seg_size;
 943         skb->len += len_on_bd;
 944
 945         return 0;
 946
 947 out:
 948         tpa_info->agg_state = QEDE_AGG_STATE_ERROR;
 949         qede_recycle_rx_bd_ring(rxq, edev, 1);
 950         return -ENOMEM;
 951 }
 952
 953 static void qede_tpa_start(struct qede_dev *edev,
 954                            struct qede_rx_queue *rxq,
 955                            struct eth_fast_path_rx_tpa_start_cqe *cqe)
 956 {
 957         struct qede_agg_info *tpa_info = &rxq->tpa_info[cqe->tpa_agg_index];
 958         struct eth_rx_bd *rx_bd_cons = qed_chain_consume(&rxq->rx_bd_ring);
 959         struct eth_rx_bd *rx_bd_prod = qed_chain_produce(&rxq->rx_bd_ring);
 960         struct sw_rx_data *replace_buf = &tpa_info->replace_buf;
 961         dma_addr_t mapping = tpa_info->replace_buf_mapping;
 962         struct sw_rx_data *sw_rx_data_cons;
 963         struct sw_rx_data *sw_rx_data_prod;
 964         enum pkt_hash_types rxhash_type;
 965         u32 rxhash;
 966
 967         sw_rx_data_cons = &rxq->sw_rx_ring[rxq->sw_rx_cons & NUM_RX_BDS_MAX];
 968         sw_rx_data_prod = &rxq->sw_rx_ring[rxq->sw_rx_prod & NUM_RX_BDS_MAX];
 969
 970         /* Use pre-allocated replacement buffer - we can't release the agg.
 971          * start until its over and we don't want to risk allocation failing
 972          * here, so re-allocate when aggregation will be over.
 973          */
 974         dma_unmap_addr_set(sw_rx_data_prod, mapping,
 975                            dma_unmap_addr(replace_buf, mapping));
 976
 977         sw_rx_data_prod->data = replace_buf->data;
 978         rx_bd_prod->addr.hi = cpu_to_le32(upper_32_bits(mapping));
 979         rx_bd_prod->addr.lo = cpu_to_le32(lower_32_bits(mapping));
 980         sw_rx_data_prod->page_offset = replace_buf->page_offset;
 981
 982         rxq->sw_rx_prod++;
 983
 984         /* move partial skb from cons to pool (don't unmap yet)
 985          * save mapping, incase we drop the packet later on.
 986          */
 987         tpa_info->start_buf = *sw_rx_data_cons;
 988         mapping = HILO_U64(le32_to_cpu(rx_bd_cons->addr.hi),
 989                            le32_to_cpu(rx_bd_cons->addr.lo));
 990
 991         tpa_info->start_buf_mapping = mapping;
 992         rxq->sw_rx_cons++;
 993
 994         /* set tpa state to start only if we are able to allocate skb
 995          * for this aggregation, otherwise mark as error and aggregation will
 996          * be dropped
 997          */
 998         tpa_info->skb = netdev_alloc_skb(edev->ndev,
 999                                          le16_to_cpu(cqe->len_on_first_bd));
1000         if (unlikely(!tpa_info->skb)) {
1001                 DP_NOTICE(edev, "Failed to allocate SKB for gro\n");
1002                 tpa_info->agg_state = QEDE_AGG_STATE_ERROR;
1003                 goto cons_buf;
1004         }
1005
1006         skb_put(tpa_info->skb, le16_to_cpu(cqe->len_on_first_bd));
1007         memcpy(&tpa_info->start_cqe, cqe, sizeof(tpa_info->start_cqe));
1008
1009         /* Start filling in the aggregation info */
1010         tpa_info->frag_id = 0;
1011         tpa_info->agg_state = QEDE_AGG_STATE_START;
1012
1013         rxhash = qede_get_rxhash(edev, cqe->bitfields,
1014                                  cqe->rss_hash, &rxhash_type);
1015         skb_set_hash(tpa_info->skb, rxhash, rxhash_type);
1016         if ((le16_to_cpu(cqe->pars_flags.flags) >>
1017              PARSING_AND_ERR_FLAGS_TAG8021QEXIST_SHIFT) &
1018                     PARSING_AND_ERR_FLAGS_TAG8021QEXIST_MASK)
1019                 tpa_info->vlan_tag = le16_to_cpu(cqe->vlan_tag);
1020         else
1021                 tpa_info->vlan_tag = 0;
1022
1023         /* This is needed in order to enable forwarding support */
1024         qede_set_gro_params(edev, tpa_info->skb, cqe);
1025
1026 cons_buf: /* We still need to handle bd_len_list to consume buffers */
1027         if (likely(cqe->ext_bd_len_list[0]))
1028                 qede_fill_frag_skb(edev, rxq, cqe->tpa_agg_index,
1029                                    le16_to_cpu(cqe->ext_bd_len_list[0]));
1030
1031         if (unlikely(cqe->ext_bd_len_list[1])) {
1032                 DP_ERR(edev,
1033                        "Unlikely - got a TPA aggregation with more than one ext_bd_len_list entry in the TPA start\n");
1034                 tpa_info->agg_state = QEDE_AGG_STATE_ERROR;
1035         }
1036 }
1037
1038 #ifdef CONFIG_INET
1039 static void qede_gro_ip_csum(struct sk_buff *skb)
1040 {
1041         const struct iphdr *iph = ip_hdr(skb);
1042         struct tcphdr *th;
1043
1044         skb_set_transport_header(skb, sizeof(struct iphdr));
1045         th = tcp_hdr(skb);
1046
1047         th->check = ~tcp_v4_check(skb->len - skb_transport_offset(skb),
1048                                   iph->saddr, iph->daddr, 0);
1049
1050         tcp_gro_complete(skb);
1051 }
1052
1053 static void qede_gro_ipv6_csum(struct sk_buff *skb)
1054 {
1055         struct ipv6hdr *iph = ipv6_hdr(skb);
1056         struct tcphdr *th;
1057
1058         skb_set_transport_header(skb, sizeof(struct ipv6hdr));
1059         th = tcp_hdr(skb);
1060
1061         th->check = ~tcp_v6_check(skb->len - skb_transport_offset(skb),
1062                                   &iph->saddr, &iph->daddr, 0);
1063         tcp_gro_complete(skb);
1064 }
1065 #endif
1066
1067 static void qede_gro_receive(struct qede_dev *edev,
1068                              struct qede_fastpath *fp,
1069                              struct sk_buff *skb,
1070                              u16 vlan_tag)
1071 {
1072         /* FW can send a single MTU sized packet from gro flow
1073          * due to aggregation timeout/last segment etc. which
1074          * is not expected to be a gro packet. If a skb has zero
1075          * frags then simply push it in the stack as non gso skb.
1076          */
1077         if (unlikely(!skb->data_len)) {
1078                 skb_shinfo(skb)->gso_type = 0;
1079                 skb_shinfo(skb)->gso_size = 0;
1080                 goto send_skb;
1081         }
1082
1083 #ifdef CONFIG_INET
1084         if (skb_shinfo(skb)->gso_size) {
1085                 skb_set_network_header(skb, 0);
1086
1087                 switch (skb->protocol) {
1088                 case htons(ETH_P_IP):
1089                         qede_gro_ip_csum(skb);
1090                         break;
1091                 case htons(ETH_P_IPV6):
1092                         qede_gro_ipv6_csum(skb);
1093                         break;
1094                 default:
1095                         DP_ERR(edev,
1096                                "Error: FW GRO supports only IPv4/IPv6, not 0x%04x\n",
1097                                ntohs(skb->protocol));
1098                 }
1099         }
1100 #endif
1101
1102 send_skb:
1103         skb_record_rx_queue(skb, fp->rss_id);
1104         qede_skb_receive(edev, fp, skb, vlan_tag);
1105 }
1106
1107 static inline void qede_tpa_cont(struct qede_dev *edev,
1108                                  struct qede_rx_queue *rxq,
1109                                  struct eth_fast_path_rx_tpa_cont_cqe *cqe)
1110 {
1111         int i;
1112
1113         for (i = 0; cqe->len_list[i]; i++)
1114                 qede_fill_frag_skb(edev, rxq, cqe->tpa_agg_index,
1115                                    le16_to_cpu(cqe->len_list[i]));
1116
1117         if (unlikely(i > 1))
1118                 DP_ERR(edev,
1119                        "Strange - TPA cont with more than a single len_list entry\n");
1120 }
1121
1122 static void qede_tpa_end(struct qede_dev *edev,
1123                          struct qede_fastpath *fp,
1124                          struct eth_fast_path_rx_tpa_end_cqe *cqe)
1125 {
1126         struct qede_rx_queue *rxq = fp->rxq;
1127         struct qede_agg_info *tpa_info;
1128         struct sk_buff *skb;
1129         int i;
1130
1131         tpa_info = &rxq->tpa_info[cqe->tpa_agg_index];
1132         skb = tpa_info->skb;
1133
1134         for (i = 0; cqe->len_list[i]; i++)
1135                 qede_fill_frag_skb(edev, rxq, cqe->tpa_agg_index,
1136                                    le16_to_cpu(cqe->len_list[i]));
1137         if (unlikely(i > 1))
1138                 DP_ERR(edev,
1139                        "Strange - TPA emd with more than a single len_list entry\n");
1140
1141         if (unlikely(tpa_info->agg_state != QEDE_AGG_STATE_START))
1142                 goto err;
1143
1144         /* Sanity */
1145         if (unlikely(cqe->num_of_bds != tpa_info->frag_id + 1))
1146                 DP_ERR(edev,
1147                        "Strange - TPA had %02x BDs, but SKB has only %d frags\n",
1148                        cqe->num_of_bds, tpa_info->frag_id);
1149         if (unlikely(skb->len != le16_to_cpu(cqe->total_packet_len)))
1150                 DP_ERR(edev,
1151                        "Strange - total packet len [cqe] is %4x but SKB has len %04x\n",
1152                        le16_to_cpu(cqe->total_packet_len), skb->len);
1153
1154         memcpy(skb->data,
1155                page_address(tpa_info->start_buf.data) +
1156                 tpa_info->start_cqe.placement_offset +
1157                 tpa_info->start_buf.page_offset,
1158                le16_to_cpu(tpa_info->start_cqe.len_on_first_bd));
1159
1160         /* Recycle [mapped] start buffer for the next replacement */
1161         tpa_info->replace_buf = tpa_info->start_buf;
1162         tpa_info->replace_buf_mapping = tpa_info->start_buf_mapping;
1163
1164         /* Finalize the SKB */
1165         skb->protocol = eth_type_trans(skb, edev->ndev);
1166         skb->ip_summed = CHECKSUM_UNNECESSARY;
1167
1168         /* tcp_gro_complete() will copy NAPI_GRO_CB(skb)->count
1169          * to skb_shinfo(skb)->gso_segs
1170          */
1171         NAPI_GRO_CB(skb)->count = le16_to_cpu(cqe->num_of_coalesced_segs);
1172
1173         qede_gro_receive(edev, fp, skb, tpa_info->vlan_tag);
1174
1175         tpa_info->agg_state = QEDE_AGG_STATE_NONE;
1176
1177         return;
1178 err:
1179         /* The BD starting the aggregation is still mapped; Re-use it for
1180          * future aggregations [as replacement buffer]
1181          */
1182         memcpy(&tpa_info->replace_buf, &tpa_info->start_buf,
1183                sizeof(struct sw_rx_data));
1184         tpa_info->replace_buf_mapping = tpa_info->start_buf_mapping;
1185         tpa_info->start_buf.data = NULL;
1186         tpa_info->agg_state = QEDE_AGG_STATE_NONE;
1187         dev_kfree_skb_any(tpa_info->skb);
1188         tpa_info->skb = NULL;
1189 }
1190
1191 static u8 qede_check_csum(u16 flag)
1192 {
1193         u16 csum_flag = 0;
1194         u8 csum = 0;
1195
1196         if ((PARSING_AND_ERR_FLAGS_L4CHKSMWASCALCULATED_MASK <<
1197              PARSING_AND_ERR_FLAGS_L4CHKSMWASCALCULATED_SHIFT) & flag) {
1198                 csum_flag |= PARSING_AND_ERR_FLAGS_L4CHKSMERROR_MASK <<
1199                              PARSING_AND_ERR_FLAGS_L4CHKSMERROR_SHIFT;
1200                 csum = QEDE_CSUM_UNNECESSARY;
1201         }
1202
1203         csum_flag |= PARSING_AND_ERR_FLAGS_IPHDRERROR_MASK <<
1204                      PARSING_AND_ERR_FLAGS_IPHDRERROR_SHIFT;
1205
1206         if (csum_flag & flag)
1207                 return QEDE_CSUM_ERROR;
1208
1209         return csum;
1210 }
1211
1212 static int qede_rx_int(struct qede_fastpath *fp, int budget)
1213 {
1214         struct qede_dev *edev = fp->edev;
1215         struct qede_rx_queue *rxq = fp->rxq;
1216
1217         u16 hw_comp_cons, sw_comp_cons, sw_rx_index, parse_flag;
1218         int rx_pkt = 0;
1219         u8 csum_flag;
1220
1221         hw_comp_cons = le16_to_cpu(*rxq->hw_cons_ptr);
1222         sw_comp_cons = qed_chain_get_cons_idx(&rxq->rx_comp_ring);
1223
1224         /* Memory barrier to prevent the CPU from doing speculative reads of CQE
1225          * / BD in the while-loop before reading hw_comp_cons. If the CQE is
1226          * read before it is written by FW, then FW writes CQE and SB, and then
1227          * the CPU reads the hw_comp_cons, it will use an old CQE.
1228          */
1229         rmb();
1230
1231         /* Loop to complete all indicated BDs */
1232         while (sw_comp_cons != hw_comp_cons) {
1233                 struct eth_fast_path_rx_reg_cqe *fp_cqe;
1234                 enum pkt_hash_types rxhash_type;
1235                 enum eth_rx_cqe_type cqe_type;
1236                 struct sw_rx_data *sw_rx_data;
1237                 union eth_rx_cqe *cqe;
1238                 struct sk_buff *skb;
1239                 struct page *data;
1240                 __le16 flags;
1241                 u16 len, pad;
1242                 u32 rx_hash;
1243
1244                 /* Get the CQE from the completion ring */
1245                 cqe = (union eth_rx_cqe *)
1246                         qed_chain_consume(&rxq->rx_comp_ring);
1247                 cqe_type = cqe->fast_path_regular.type;
1248
1249                 if (unlikely(cqe_type == ETH_RX_CQE_TYPE_SLOW_PATH)) {
1250                         edev->ops->eth_cqe_completion(
1251                                         edev->cdev, fp->rss_id,
1252                                         (struct eth_slow_path_rx_cqe *)cqe);
1253                         goto next_cqe;
1254                 }
1255
1256                 if (cqe_type != ETH_RX_CQE_TYPE_REGULAR) {
1257                         switch (cqe_type) {
1258                         case ETH_RX_CQE_TYPE_TPA_START:
1259                                 qede_tpa_start(edev, rxq,
1260                                                &cqe->fast_path_tpa_start);
1261                                 goto next_cqe;
1262                         case ETH_RX_CQE_TYPE_TPA_CONT:
1263                                 qede_tpa_cont(edev, rxq,
1264                                               &cqe->fast_path_tpa_cont);
1265                                 goto next_cqe;
1266                         case ETH_RX_CQE_TYPE_TPA_END:
1267                                 qede_tpa_end(edev, fp,
1268                                              &cqe->fast_path_tpa_end);
1269                                 goto next_rx_only;
1270                         default:
1271                                 break;
1272                         }
1273                 }
1274
1275                 /* Get the data from the SW ring */
1276                 sw_rx_index = rxq->sw_rx_cons & NUM_RX_BDS_MAX;
1277                 sw_rx_data = &rxq->sw_rx_ring[sw_rx_index];
1278                 data = sw_rx_data->data;
1279
1280                 fp_cqe = &cqe->fast_path_regular;
1281                 len =  le16_to_cpu(fp_cqe->len_on_first_bd);
1282                 pad = fp_cqe->placement_offset;
1283                 flags = cqe->fast_path_regular.pars_flags.flags;
1284
1285                 /* If this is an error packet then drop it */
1286                 parse_flag = le16_to_cpu(flags);
1287
1288                 csum_flag = qede_check_csum(parse_flag);
1289                 if (unlikely(csum_flag == QEDE_CSUM_ERROR)) {
1290                         DP_NOTICE(edev,
1291                                   "CQE in CONS = %u has error, flags = %x, dropping incoming packet\n",
1292                                   sw_comp_cons, parse_flag);
1293                         rxq->rx_hw_errors++;
1294                         qede_recycle_rx_bd_ring(rxq, edev, fp_cqe->bd_num);
1295                         goto next_cqe;
1296                 }
1297
1298                 skb = netdev_alloc_skb(edev->ndev, QEDE_RX_HDR_SIZE);
1299                 if (unlikely(!skb)) {
1300                         DP_NOTICE(edev,
1301                                   "Build_skb failed, dropping incoming packet\n");
1302                         qede_recycle_rx_bd_ring(rxq, edev, fp_cqe->bd_num);
1303                         rxq->rx_alloc_errors++;
1304                         goto next_cqe;
1305                 }
1306
1307                 /* Copy data into SKB */
1308                 if (len + pad <= QEDE_RX_HDR_SIZE) {
1309                         memcpy(skb_put(skb, len),
1310                                page_address(data) + pad +
1311                                 sw_rx_data->page_offset, len);
1312                         qede_reuse_page(edev, rxq, sw_rx_data);
1313                 } else {
1314                         struct skb_frag_struct *frag;
1315                         unsigned int pull_len;
1316                         unsigned char *va;
1317
1318                         frag = &skb_shinfo(skb)->frags[0];
1319
1320                         skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, data,
1321                                         pad + sw_rx_data->page_offset,
1322                                         len, rxq->rx_buf_seg_size);
1323
1324                         va = skb_frag_address(frag);
1325                         pull_len = eth_get_headlen(va, QEDE_RX_HDR_SIZE);
1326
1327                         /* Align the pull_len to optimize memcpy */
1328                         memcpy(skb->data, va, ALIGN(pull_len, sizeof(long)));
1329
1330                         skb_frag_size_sub(frag, pull_len);
1331                         frag->page_offset += pull_len;
1332                         skb->data_len -= pull_len;
1333                         skb->tail += pull_len;
1334
1335                         if (unlikely(qede_realloc_rx_buffer(edev, rxq,
1336                                                             sw_rx_data))) {
1337                                 DP_ERR(edev, "Failed to allocate rx buffer\n");
1338                                 /* Incr page ref count to reuse on allocation
1339                                  * failure so that it doesn't get freed while
1340                                  * freeing SKB.
1341                                  */
1342
1343                                 atomic_inc(&sw_rx_data->data->_count);
1344                                 rxq->rx_alloc_errors++;
1345                                 qede_recycle_rx_bd_ring(rxq, edev,
1346                                                         fp_cqe->bd_num);
1347                                 dev_kfree_skb_any(skb);
1348                                 goto next_cqe;
1349                         }
1350                 }
1351
1352                 qede_rx_bd_ring_consume(rxq);
1353
1354                 if (fp_cqe->bd_num != 1) {
1355                         u16 pkt_len = le16_to_cpu(fp_cqe->pkt_len);
1356                         u8 num_frags;
1357
1358                         pkt_len -= len;
1359
1360                         for (num_frags = fp_cqe->bd_num - 1; num_frags > 0;
1361                              num_frags--) {
1362                                 u16 cur_size = pkt_len > rxq->rx_buf_size ?
1363                                                 rxq->rx_buf_size : pkt_len;
1364                                 if (unlikely(!cur_size)) {
1365                                         DP_ERR(edev,
1366                                                "Still got %d BDs for mapping jumbo, but length became 0\n",
1367                                                num_frags);
1368                                         qede_recycle_rx_bd_ring(rxq, edev,
1369                                                                 num_frags);
1370                                         dev_kfree_skb_any(skb);
1371                                         goto next_cqe;
1372                                 }
1373
1374                                 if (unlikely(qede_alloc_rx_buffer(edev, rxq))) {
1375                                         qede_recycle_rx_bd_ring(rxq, edev,
1376                                                                 num_frags);
1377                                         dev_kfree_skb_any(skb);
1378                                         goto next_cqe;
1379                                 }
1380
1381                                 sw_rx_index = rxq->sw_rx_cons & NUM_RX_BDS_MAX;
1382                                 sw_rx_data = &rxq->sw_rx_ring[sw_rx_index];
1383                                 qede_rx_bd_ring_consume(rxq);
1384
1385                                 dma_unmap_page(&edev->pdev->dev,
1386                                                sw_rx_data->mapping,
1387                                                PAGE_SIZE, DMA_FROM_DEVICE);
1388
1389                                 skb_fill_page_desc(skb,
1390                                                    skb_shinfo(skb)->nr_frags++,
1391                                                    sw_rx_data->data, 0,
1392                                                    cur_size);
1393
1394                                 skb->truesize += PAGE_SIZE;
1395                                 skb->data_len += cur_size;
1396                                 skb->len += cur_size;
1397                                 pkt_len -= cur_size;
1398                         }
1399
1400                         if (unlikely(pkt_len))
1401                                 DP_ERR(edev,
1402                                        "Mapped all BDs of jumbo, but still have %d bytes\n",
1403                                        pkt_len);
1404                 }
1405
1406                 skb->protocol = eth_type_trans(skb, edev->ndev);
1407
1408                 rx_hash = qede_get_rxhash(edev, fp_cqe->bitfields,
1409                                           fp_cqe->rss_hash,
1410                                           &rxhash_type);
1411
1412                 skb_set_hash(skb, rx_hash, rxhash_type);
1413
1414                 qede_set_skb_csum(skb, csum_flag);
1415
1416                 skb_record_rx_queue(skb, fp->rss_id);
1417
1418                 qede_skb_receive(edev, fp, skb, le16_to_cpu(fp_cqe->vlan_tag));
1419 next_rx_only:
1420                 rx_pkt++;
1421
1422 next_cqe: /* don't consume bd rx buffer */
1423                 qed_chain_recycle_consumed(&rxq->rx_comp_ring);
1424                 sw_comp_cons = qed_chain_get_cons_idx(&rxq->rx_comp_ring);
1425                 /* CR TPA - revisit how to handle budget in TPA perhaps
1426                  * increase on "end"
1427                  */
1428                 if (rx_pkt == budget)
1429                         break;
1430         } /* repeat while sw_comp_cons != hw_comp_cons... */
1431
1432         /* Update producers */
1433         qede_update_rx_prod(edev, rxq);
1434
1435         return rx_pkt;
1436 }
1437
1438 static int qede_poll(struct napi_struct *napi, int budget)
1439 {
1440         int work_done = 0;
1441         struct qede_fastpath *fp = container_of(napi, struct qede_fastpath,
1442                                                  napi);
1443         struct qede_dev *edev = fp->edev;
1444
1445         while (1) {
1446                 u8 tc;
1447
1448                 for (tc = 0; tc < edev->num_tc; tc++)
1449                         if (qede_txq_has_work(&fp->txqs[tc]))
1450                                 qede_tx_int(edev, &fp->txqs[tc]);
1451
1452                 if (qede_has_rx_work(fp->rxq)) {
1453                         work_done += qede_rx_int(fp, budget - work_done);
1454
1455                         /* must not complete if we consumed full budget */
1456                         if (work_done >= budget)
1457                                 break;
1458                 }
1459
1460                 /* Fall out from the NAPI loop if needed */
1461                 if (!(qede_has_rx_work(fp->rxq) || qede_has_tx_work(fp))) {
1462                         qed_sb_update_sb_idx(fp->sb_info);
1463                         /* *_has_*_work() reads the status block,
1464                          * thus we need to ensure that status block indices
1465                          * have been actually read (qed_sb_update_sb_idx)
1466                          * prior to this check (*_has_*_work) so that
1467                          * we won't write the "newer" value of the status block
1468                          * to HW (if there was a DMA right after
1469                          * qede_has_rx_work and if there is no rmb, the memory
1470                          * reading (qed_sb_update_sb_idx) may be postponed
1471                          * to right before *_ack_sb). In this case there
1472                          * will never be another interrupt until there is
1473                          * another update of the status block, while there
1474                          * is still unhandled work.
1475                          */
1476                         rmb();
1477
1478                         if (!(qede_has_rx_work(fp->rxq) ||
1479                               qede_has_tx_work(fp))) {
1480                                 napi_complete(napi);
1481                                 /* Update and reenable interrupts */
1482                                 qed_sb_ack(fp->sb_info, IGU_INT_ENABLE,
1483                                            1 /*update*/);
1484                                 break;
1485                         }
1486                 }
1487         }
1488
1489         return work_done;
1490 }
1491
1492 static irqreturn_t qede_msix_fp_int(int irq, void *fp_cookie)
1493 {
1494         struct qede_fastpath *fp = fp_cookie;
1495
1496         qed_sb_ack(fp->sb_info, IGU_INT_DISABLE, 0 /*do not update*/);
1497
1498         napi_schedule_irqoff(&fp->napi);
1499         return IRQ_HANDLED;
1500 }
1501
1502 /* -------------------------------------------------------------------------
1503  * END OF FAST-PATH
1504  * -------------------------------------------------------------------------
1505  */
1506
1507 static int qede_open(struct net_device *ndev);
1508 static int qede_close(struct net_device *ndev);
1509 static int qede_set_mac_addr(struct net_device *ndev, void *p);
1510 static void qede_set_rx_mode(struct net_device *ndev);
1511 static void qede_config_rx_mode(struct net_device *ndev);
1512
1513 static int qede_set_ucast_rx_mac(struct qede_dev *edev,
1514                                  enum qed_filter_xcast_params_type opcode,
1515                                  unsigned char mac[ETH_ALEN])
1516 {
1517         struct qed_filter_params filter_cmd;
1518
1519         memset(&filter_cmd, 0, sizeof(filter_cmd));
1520         filter_cmd.type = QED_FILTER_TYPE_UCAST;
1521         filter_cmd.filter.ucast.type = opcode;
1522         filter_cmd.filter.ucast.mac_valid = 1;
1523         ether_addr_copy(filter_cmd.filter.ucast.mac, mac);
1524
1525         return edev->ops->filter_config(edev->cdev, &filter_cmd);
1526 }
1527
1528 static int qede_set_ucast_rx_vlan(struct qede_dev *edev,
1529                                   enum qed_filter_xcast_params_type opcode,
1530                                   u16 vid)
1531 {
1532         struct qed_filter_params filter_cmd;
1533
1534         memset(&filter_cmd, 0, sizeof(filter_cmd));
1535         filter_cmd.type = QED_FILTER_TYPE_UCAST;
1536         filter_cmd.filter.ucast.type = opcode;
1537         filter_cmd.filter.ucast.vlan_valid = 1;
1538         filter_cmd.filter.ucast.vlan = vid;
1539
1540         return edev->ops->filter_config(edev->cdev, &filter_cmd);
1541 }
1542
1543 void qede_fill_by_demand_stats(struct qede_dev *edev)
1544 {
1545         struct qed_eth_stats stats;
1546
1547         edev->ops->get_vport_stats(edev->cdev, &stats);
1548         edev->stats.no_buff_discards = stats.no_buff_discards;
1549         edev->stats.rx_ucast_bytes = stats.rx_ucast_bytes;
1550         edev->stats.rx_mcast_bytes = stats.rx_mcast_bytes;
1551         edev->stats.rx_bcast_bytes = stats.rx_bcast_bytes;
1552         edev->stats.rx_ucast_pkts = stats.rx_ucast_pkts;
1553         edev->stats.rx_mcast_pkts = stats.rx_mcast_pkts;
1554         edev->stats.rx_bcast_pkts = stats.rx_bcast_pkts;
1555         edev->stats.mftag_filter_discards = stats.mftag_filter_discards;
1556         edev->stats.mac_filter_discards = stats.mac_filter_discards;
1557
1558         edev->stats.tx_ucast_bytes = stats.tx_ucast_bytes;
1559         edev->stats.tx_mcast_bytes = stats.tx_mcast_bytes;
1560         edev->stats.tx_bcast_bytes = stats.tx_bcast_bytes;
1561         edev->stats.tx_ucast_pkts = stats.tx_ucast_pkts;
1562         edev->stats.tx_mcast_pkts = stats.tx_mcast_pkts;
1563         edev->stats.tx_bcast_pkts = stats.tx_bcast_pkts;
1564         edev->stats.tx_err_drop_pkts = stats.tx_err_drop_pkts;
1565         edev->stats.coalesced_pkts = stats.tpa_coalesced_pkts;
1566         edev->stats.coalesced_events = stats.tpa_coalesced_events;
1567         edev->stats.coalesced_aborts_num = stats.tpa_aborts_num;
1568         edev->stats.non_coalesced_pkts = stats.tpa_not_coalesced_pkts;
1569         edev->stats.coalesced_bytes = stats.tpa_coalesced_bytes;
1570
1571         edev->stats.rx_64_byte_packets = stats.rx_64_byte_packets;
1572         edev->stats.rx_127_byte_packets = stats.rx_127_byte_packets;
1573         edev->stats.rx_255_byte_packets = stats.rx_255_byte_packets;
1574         edev->stats.rx_511_byte_packets = stats.rx_511_byte_packets;
1575         edev->stats.rx_1023_byte_packets = stats.rx_1023_byte_packets;
1576         edev->stats.rx_1518_byte_packets = stats.rx_1518_byte_packets;
1577         edev->stats.rx_1522_byte_packets = stats.rx_1522_byte_packets;
1578         edev->stats.rx_2047_byte_packets = stats.rx_2047_byte_packets;
1579         edev->stats.rx_4095_byte_packets = stats.rx_4095_byte_packets;
1580         edev->stats.rx_9216_byte_packets = stats.rx_9216_byte_packets;
1581         edev->stats.rx_16383_byte_packets = stats.rx_16383_byte_packets;
1582         edev->stats.rx_crc_errors = stats.rx_crc_errors;
1583         edev->stats.rx_mac_crtl_frames = stats.rx_mac_crtl_frames;
1584         edev->stats.rx_pause_frames = stats.rx_pause_frames;
1585         edev->stats.rx_pfc_frames = stats.rx_pfc_frames;
1586         edev->stats.rx_align_errors = stats.rx_align_errors;
1587         edev->stats.rx_carrier_errors = stats.rx_carrier_errors;
1588         edev->stats.rx_oversize_packets = stats.rx_oversize_packets;
1589         edev->stats.rx_jabbers = stats.rx_jabbers;
1590         edev->stats.rx_undersize_packets = stats.rx_undersize_packets;
1591         edev->stats.rx_fragments = stats.rx_fragments;
1592         edev->stats.tx_64_byte_packets = stats.tx_64_byte_packets;
1593         edev->stats.tx_65_to_127_byte_packets = stats.tx_65_to_127_byte_packets;
1594         edev->stats.tx_128_to_255_byte_packets =
1595                                 stats.tx_128_to_255_byte_packets;
1596         edev->stats.tx_256_to_511_byte_packets =
1597                                 stats.tx_256_to_511_byte_packets;
1598         edev->stats.tx_512_to_1023_byte_packets =
1599                                 stats.tx_512_to_1023_byte_packets;
1600         edev->stats.tx_1024_to_1518_byte_packets =
1601                                 stats.tx_1024_to_1518_byte_packets;
1602         edev->stats.tx_1519_to_2047_byte_packets =
1603                                 stats.tx_1519_to_2047_byte_packets;
1604         edev->stats.tx_2048_to_4095_byte_packets =
1605                                 stats.tx_2048_to_4095_byte_packets;
1606         edev->stats.tx_4096_to_9216_byte_packets =
1607                                 stats.tx_4096_to_9216_byte_packets;
1608         edev->stats.tx_9217_to_16383_byte_packets =
1609                                 stats.tx_9217_to_16383_byte_packets;
1610         edev->stats.tx_pause_frames = stats.tx_pause_frames;
1611         edev->stats.tx_pfc_frames = stats.tx_pfc_frames;
1612         edev->stats.tx_lpi_entry_count = stats.tx_lpi_entry_count;
1613         edev->stats.tx_total_collisions = stats.tx_total_collisions;
1614         edev->stats.brb_truncates = stats.brb_truncates;
1615         edev->stats.brb_discards = stats.brb_discards;
1616         edev->stats.tx_mac_ctrl_frames = stats.tx_mac_ctrl_frames;
1617 }
1618
1619 static struct rtnl_link_stats64 *qede_get_stats64(
1620                             struct net_device *dev,
1621                             struct rtnl_link_stats64 *stats)
1622 {
1623         struct qede_dev *edev = netdev_priv(dev);
1624
1625         qede_fill_by_demand_stats(edev);
1626
1627         stats->rx_packets = edev->stats.rx_ucast_pkts +
1628                             edev->stats.rx_mcast_pkts +
1629                             edev->stats.rx_bcast_pkts;
1630         stats->tx_packets = edev->stats.tx_ucast_pkts +
1631                             edev->stats.tx_mcast_pkts +
1632                             edev->stats.tx_bcast_pkts;
1633
1634         stats->rx_bytes = edev->stats.rx_ucast_bytes +
1635                           edev->stats.rx_mcast_bytes +
1636                           edev->stats.rx_bcast_bytes;
1637
1638         stats->tx_bytes = edev->stats.tx_ucast_bytes +
1639                           edev->stats.tx_mcast_bytes +
1640                           edev->stats.tx_bcast_bytes;
1641
1642         stats->tx_errors = edev->stats.tx_err_drop_pkts;
1643         stats->multicast = edev->stats.rx_mcast_pkts +
1644                            edev->stats.rx_bcast_pkts;
1645
1646         stats->rx_fifo_errors = edev->stats.no_buff_discards;
1647
1648         stats->collisions = edev->stats.tx_total_collisions;
1649         stats->rx_crc_errors = edev->stats.rx_crc_errors;
1650         stats->rx_frame_errors = edev->stats.rx_align_errors;
1651
1652         return stats;
1653 }
1654
1655 static void qede_config_accept_any_vlan(struct qede_dev *edev, bool action)
1656 {
1657         struct qed_update_vport_params params;
1658         int rc;
1659
1660         /* Proceed only if action actually needs to be performed */
1661         if (edev->accept_any_vlan == action)
1662                 return;
1663
1664         memset(&params, 0, sizeof(params));
1665
1666         params.vport_id = 0;
1667         params.accept_any_vlan = action;
1668         params.update_accept_any_vlan_flg = 1;
1669
1670         rc = edev->ops->vport_update(edev->cdev, &params);
1671         if (rc) {
1672                 DP_ERR(edev, "Failed to %s accept-any-vlan\n",
1673                        action ? "enable" : "disable");
1674         } else {
1675                 DP_INFO(edev, "%s accept-any-vlan\n",
1676                         action ? "enabled" : "disabled");
1677                 edev->accept_any_vlan = action;
1678         }
1679 }
1680
1681 static int qede_vlan_rx_add_vid(struct net_device *dev, __be16 proto, u16 vid)
1682 {
1683         struct qede_dev *edev = netdev_priv(dev);
1684         struct qede_vlan *vlan, *tmp;
1685         int rc;
1686
1687         DP_VERBOSE(edev, NETIF_MSG_IFUP, "Adding vlan 0x%04x\n", vid);
1688
1689         vlan = kzalloc(sizeof(*vlan), GFP_KERNEL);
1690         if (!vlan) {
1691                 DP_INFO(edev, "Failed to allocate struct for vlan\n");
1692                 return -ENOMEM;
1693         }
1694         INIT_LIST_HEAD(&vlan->list);
1695         vlan->vid = vid;
1696         vlan->configured = false;
1697
1698         /* Verify vlan isn't already configured */
1699         list_for_each_entry(tmp, &edev->vlan_list, list) {
1700                 if (tmp->vid == vlan->vid) {
1701                         DP_VERBOSE(edev, (NETIF_MSG_IFUP | NETIF_MSG_IFDOWN),
1702                                    "vlan already configured\n");
1703                         kfree(vlan);
1704                         return -EEXIST;
1705                 }
1706         }
1707
1708         /* If interface is down, cache this VLAN ID and return */
1709         if (edev->state != QEDE_STATE_OPEN) {
1710                 DP_VERBOSE(edev, NETIF_MSG_IFDOWN,
1711                            "Interface is down, VLAN %d will be configured when interface is up\n",
1712                            vid);
1713                 if (vid != 0)
1714                         edev->non_configured_vlans++;
1715                 list_add(&vlan->list, &edev->vlan_list);
1716
1717                 return 0;
1718         }
1719
1720         /* Check for the filter limit.
1721          * Note - vlan0 has a reserved filter and can be added without
1722          * worrying about quota
1723          */
1724         if ((edev->configured_vlans < edev->dev_info.num_vlan_filters) ||
1725             (vlan->vid == 0)) {
1726                 rc = qede_set_ucast_rx_vlan(edev,
1727                                             QED_FILTER_XCAST_TYPE_ADD,
1728                                             vlan->vid);
1729                 if (rc) {
1730                         DP_ERR(edev, "Failed to configure VLAN %d\n",
1731                                vlan->vid);
1732                         kfree(vlan);
1733                         return -EINVAL;
1734                 }
1735                 vlan->configured = true;
1736
1737                 /* vlan0 filter isn't consuming out of our quota */
1738                 if (vlan->vid != 0)
1739                         edev->configured_vlans++;
1740         } else {
1741                 /* Out of quota; Activate accept-any-VLAN mode */
1742                 if (!edev->non_configured_vlans)
1743                         qede_config_accept_any_vlan(edev, true);
1744
1745                 edev->non_configured_vlans++;
1746         }
1747
1748         list_add(&vlan->list, &edev->vlan_list);
1749
1750         return 0;
1751 }
1752
1753 static void qede_del_vlan_from_list(struct qede_dev *edev,
1754                                     struct qede_vlan *vlan)
1755 {
1756         /* vlan0 filter isn't consuming out of our quota */
1757         if (vlan->vid != 0) {
1758                 if (vlan->configured)
1759                         edev->configured_vlans--;
1760                 else
1761                         edev->non_configured_vlans--;
1762         }
1763
1764         list_del(&vlan->list);
1765         kfree(vlan);
1766 }
1767
1768 static int qede_configure_vlan_filters(struct qede_dev *edev)
1769 {
1770         int rc = 0, real_rc = 0, accept_any_vlan = 0;
1771         struct qed_dev_eth_info *dev_info;
1772         struct qede_vlan *vlan = NULL;
1773
1774         if (list_empty(&edev->vlan_list))
1775                 return 0;
1776
1777         dev_info = &edev->dev_info;
1778
1779         /* Configure non-configured vlans */
1780         list_for_each_entry(vlan, &edev->vlan_list, list) {
1781                 if (vlan->configured)
1782                         continue;
1783
1784                 /* We have used all our credits, now enable accept_any_vlan */
1785                 if ((vlan->vid != 0) &&
1786                     (edev->configured_vlans == dev_info->num_vlan_filters)) {
1787                         accept_any_vlan = 1;
1788                         continue;
1789                 }
1790
1791                 DP_VERBOSE(edev, NETIF_MSG_IFUP, "Adding vlan %d\n", vlan->vid);
1792
1793                 rc = qede_set_ucast_rx_vlan(edev, QED_FILTER_XCAST_TYPE_ADD,
1794                                             vlan->vid);
1795                 if (rc) {
1796                         DP_ERR(edev, "Failed to configure VLAN %u\n",
1797                                vlan->vid);
1798                         real_rc = rc;
1799                         continue;
1800                 }
1801
1802                 vlan->configured = true;
1803                 /* vlan0 filter doesn't consume our VLAN filter's quota */
1804                 if (vlan->vid != 0) {
1805                         edev->non_configured_vlans--;
1806                         edev->configured_vlans++;
1807                 }
1808         }
1809
1810         /* enable accept_any_vlan mode if we have more VLANs than credits,
1811          * or remove accept_any_vlan mode if we've actually removed
1812          * a non-configured vlan, and all remaining vlans are truly configured.
1813          */
1814
1815         if (accept_any_vlan)
1816                 qede_config_accept_any_vlan(edev, true);
1817         else if (!edev->non_configured_vlans)
1818                 qede_config_accept_any_vlan(edev, false);
1819
1820         return real_rc;
1821 }
1822
1823 static int qede_vlan_rx_kill_vid(struct net_device *dev, __be16 proto, u16 vid)
1824 {
1825         struct qede_dev *edev = netdev_priv(dev);
1826         struct qede_vlan *vlan = NULL;
1827         int rc;
1828
1829         DP_VERBOSE(edev, NETIF_MSG_IFDOWN, "Removing vlan 0x%04x\n", vid);
1830
1831         /* Find whether entry exists */
1832         list_for_each_entry(vlan, &edev->vlan_list, list)
1833                 if (vlan->vid == vid)
1834                         break;
1835
1836         if (!vlan || (vlan->vid != vid)) {
1837                 DP_VERBOSE(edev, (NETIF_MSG_IFUP | NETIF_MSG_IFDOWN),
1838                            "Vlan isn't configured\n");
1839                 return 0;
1840         }
1841
1842         if (edev->state != QEDE_STATE_OPEN) {
1843                 /* As interface is already down, we don't have a VPORT
1844                  * instance to remove vlan filter. So just update vlan list
1845                  */
1846                 DP_VERBOSE(edev, NETIF_MSG_IFDOWN,
1847                            "Interface is down, removing VLAN from list only\n");
1848                 qede_del_vlan_from_list(edev, vlan);
1849                 return 0;
1850         }
1851
1852         /* Remove vlan */
1853         rc = qede_set_ucast_rx_vlan(edev, QED_FILTER_XCAST_TYPE_DEL, vid);
1854         if (rc) {
1855                 DP_ERR(edev, "Failed to remove VLAN %d\n", vid);
1856                 return -EINVAL;
1857         }
1858
1859         qede_del_vlan_from_list(edev, vlan);
1860
1861         /* We have removed a VLAN - try to see if we can
1862          * configure non-configured VLAN from the list.
1863          */
1864         rc = qede_configure_vlan_filters(edev);
1865
1866         return rc;
1867 }
1868
1869 static void qede_vlan_mark_nonconfigured(struct qede_dev *edev)
1870 {
1871         struct qede_vlan *vlan = NULL;
1872
1873         if (list_empty(&edev->vlan_list))
1874                 return;
1875
1876         list_for_each_entry(vlan, &edev->vlan_list, list) {
1877                 if (!vlan->configured)
1878                         continue;
1879
1880                 vlan->configured = false;
1881
1882                 /* vlan0 filter isn't consuming out of our quota */
1883                 if (vlan->vid != 0) {
1884                         edev->non_configured_vlans++;
1885                         edev->configured_vlans--;
1886                 }
1887
1888                 DP_VERBOSE(edev, NETIF_MSG_IFDOWN,
1889                            "marked vlan %d as non-configured\n",
1890                            vlan->vid);
1891         }
1892
1893         edev->accept_any_vlan = false;
1894 }
1895
1896 static const struct net_device_ops qede_netdev_ops = {
1897         .ndo_open = qede_open,
1898         .ndo_stop = qede_close,
1899         .ndo_start_xmit = qede_start_xmit,
1900         .ndo_set_rx_mode = qede_set_rx_mode,
1901         .ndo_set_mac_address = qede_set_mac_addr,
1902         .ndo_validate_addr = eth_validate_addr,
1903         .ndo_change_mtu = qede_change_mtu,
1904         .ndo_vlan_rx_add_vid = qede_vlan_rx_add_vid,
1905         .ndo_vlan_rx_kill_vid = qede_vlan_rx_kill_vid,
1906         .ndo_get_stats64 = qede_get_stats64,
1907 };
1908
1909 /* -------------------------------------------------------------------------
1910  * START OF PROBE / REMOVE
1911  * -------------------------------------------------------------------------
1912  */
1913
1914 static struct qede_dev *qede_alloc_etherdev(struct qed_dev *cdev,
1915                                             struct pci_dev *pdev,
1916                                             struct qed_dev_eth_info *info,
1917                                             u32 dp_module,
1918                                             u8 dp_level)
1919 {
1920         struct net_device *ndev;
1921         struct qede_dev *edev;
1922
1923         ndev = alloc_etherdev_mqs(sizeof(*edev),
1924                                   info->num_queues,
1925                                   info->num_queues);
1926         if (!ndev) {
1927                 pr_err("etherdev allocation failed\n");
1928                 return NULL;
1929         }
1930
1931         edev = netdev_priv(ndev);
1932         edev->ndev = ndev;
1933         edev->cdev = cdev;
1934         edev->pdev = pdev;
1935         edev->dp_module = dp_module;
1936         edev->dp_level = dp_level;
1937         edev->ops = qed_ops;
1938         edev->q_num_rx_buffers = NUM_RX_BDS_DEF;
1939         edev->q_num_tx_buffers = NUM_TX_BDS_DEF;
1940
1941         DP_INFO(edev, "Allocated netdev with 64 tx queues and 64 rx queues\n");
1942
1943         SET_NETDEV_DEV(ndev, &pdev->dev);
1944
1945         memset(&edev->stats, 0, sizeof(edev->stats));
1946         memcpy(&edev->dev_info, info, sizeof(*info));
1947
1948         edev->num_tc = edev->dev_info.num_tc;
1949
1950         INIT_LIST_HEAD(&edev->vlan_list);
1951
1952         return edev;
1953 }
1954
1955 static void qede_init_ndev(struct qede_dev *edev)
1956 {
1957         struct net_device *ndev = edev->ndev;
1958         struct pci_dev *pdev = edev->pdev;
1959         u32 hw_features;
1960
1961         pci_set_drvdata(pdev, ndev);
1962
1963         ndev->mem_start = edev->dev_info.common.pci_mem_start;
1964         ndev->base_addr = ndev->mem_start;
1965         ndev->mem_end = edev->dev_info.common.pci_mem_end;
1966         ndev->irq = edev->dev_info.common.pci_irq;
1967
1968         ndev->watchdog_timeo = TX_TIMEOUT;
1969
1970         ndev->netdev_ops = &qede_netdev_ops;
1971
1972         qede_set_ethtool_ops(ndev);
1973
1974         /* user-changeble features */
1975         hw_features = NETIF_F_GRO | NETIF_F_SG |
1976                       NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
1977                       NETIF_F_TSO | NETIF_F_TSO6;
1978
1979         ndev->vlan_features = hw_features | NETIF_F_RXHASH | NETIF_F_RXCSUM |
1980                               NETIF_F_HIGHDMA;
1981         ndev->features = hw_features | NETIF_F_RXHASH | NETIF_F_RXCSUM |
1982                          NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_HIGHDMA |
1983                          NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_CTAG_TX;
1984
1985         ndev->hw_features = hw_features;
1986
1987         /* Set network device HW mac */
1988         ether_addr_copy(edev->ndev->dev_addr, edev->dev_info.common.hw_mac);
1989 }
1990
1991 /* This function converts from 32b param to two params of level and module
1992  * Input 32b decoding:
1993  * b31 - enable all NOTICE prints. NOTICE prints are for deviation from the
1994  * 'happy' flow, e.g. memory allocation failed.
1995  * b30 - enable all INFO prints. INFO prints are for major steps in the flow
1996  * and provide important parameters.
1997  * b29-b0 - per-module bitmap, where each bit enables VERBOSE prints of that
1998  * module. VERBOSE prints are for tracking the specific flow in low level.
1999  *
2000  * Notice that the level should be that of the lowest required logs.
2001  */
2002 void qede_config_debug(uint debug, u32 *p_dp_module, u8 *p_dp_level)
2003 {
2004         *p_dp_level = QED_LEVEL_NOTICE;
2005         *p_dp_module = 0;
2006
2007         if (debug & QED_LOG_VERBOSE_MASK) {
2008                 *p_dp_level = QED_LEVEL_VERBOSE;
2009                 *p_dp_module = (debug & 0x3FFFFFFF);
2010         } else if (debug & QED_LOG_INFO_MASK) {
2011                 *p_dp_level = QED_LEVEL_INFO;
2012         } else if (debug & QED_LOG_NOTICE_MASK) {
2013                 *p_dp_level = QED_LEVEL_NOTICE;
2014         }
2015 }
2016
2017 static void qede_free_fp_array(struct qede_dev *edev)
2018 {
2019         if (edev->fp_array) {
2020                 struct qede_fastpath *fp;
2021                 int i;
2022
2023                 for_each_rss(i) {
2024                         fp = &edev->fp_array[i];
2025
2026                         kfree(fp->sb_info);
2027                         kfree(fp->rxq);
2028                         kfree(fp->txqs);
2029                 }
2030                 kfree(edev->fp_array);
2031         }
2032         edev->num_rss = 0;
2033 }
2034
2035 static int qede_alloc_fp_array(struct qede_dev *edev)
2036 {
2037         struct qede_fastpath *fp;
2038         int i;
2039
2040         edev->fp_array = kcalloc(QEDE_RSS_CNT(edev),
2041                                  sizeof(*edev->fp_array), GFP_KERNEL);
2042         if (!edev->fp_array) {
2043                 DP_NOTICE(edev, "fp array allocation failed\n");
2044                 goto err;
2045         }
2046
2047         for_each_rss(i) {
2048                 fp = &edev->fp_array[i];
2049
2050                 fp->sb_info = kcalloc(1, sizeof(*fp->sb_info), GFP_KERNEL);
2051                 if (!fp->sb_info) {
2052                         DP_NOTICE(edev, "sb info struct allocation failed\n");
2053                         goto err;
2054                 }
2055
2056                 fp->rxq = kcalloc(1, sizeof(*fp->rxq), GFP_KERNEL);
2057                 if (!fp->rxq) {
2058                         DP_NOTICE(edev, "RXQ struct allocation failed\n");
2059                         goto err;
2060                 }
2061
2062                 fp->txqs = kcalloc(edev->num_tc, sizeof(*fp->txqs), GFP_KERNEL);
2063                 if (!fp->txqs) {
2064                         DP_NOTICE(edev, "TXQ array allocation failed\n");
2065                         goto err;
2066                 }
2067         }
2068
2069         return 0;
2070 err:
2071         qede_free_fp_array(edev);
2072         return -ENOMEM;
2073 }
2074
2075 static void qede_sp_task(struct work_struct *work)
2076 {
2077         struct qede_dev *edev = container_of(work, struct qede_dev,
2078                                              sp_task.work);
2079         mutex_lock(&edev->qede_lock);
2080
2081         if (edev->state == QEDE_STATE_OPEN) {
2082                 if (test_and_clear_bit(QEDE_SP_RX_MODE, &edev->sp_flags))
2083                         qede_config_rx_mode(edev->ndev);
2084         }
2085
2086         mutex_unlock(&edev->qede_lock);
2087 }
2088
2089 static void qede_update_pf_params(struct qed_dev *cdev)
2090 {
2091         struct qed_pf_params pf_params;
2092
2093         /* 16 rx + 16 tx */
2094         memset(&pf_params, 0, sizeof(struct qed_pf_params));
2095         pf_params.eth_pf_params.num_cons = 32;
2096         qed_ops->common->update_pf_params(cdev, &pf_params);
2097 }
2098
2099 enum qede_probe_mode {
2100         QEDE_PROBE_NORMAL,
2101 };
2102
2103 static int __qede_probe(struct pci_dev *pdev, u32 dp_module, u8 dp_level,
2104                         enum qede_probe_mode mode)
2105 {
2106         struct qed_slowpath_params params;
2107         struct qed_dev_eth_info dev_info;
2108         struct qede_dev *edev;
2109         struct qed_dev *cdev;
2110         int rc;
2111
2112         if (unlikely(dp_level & QED_LEVEL_INFO))
2113                 pr_notice("Starting qede probe\n");
2114
2115         cdev = qed_ops->common->probe(pdev, QED_PROTOCOL_ETH,
2116                                       dp_module, dp_level);
2117         if (!cdev) {
2118                 rc = -ENODEV;
2119                 goto err0;
2120         }
2121
2122         qede_update_pf_params(cdev);
2123
2124         /* Start the Slowpath-process */
2125         memset(&params, 0, sizeof(struct qed_slowpath_params));
2126         params.int_mode = QED_INT_MODE_MSIX;
2127         params.drv_major = QEDE_MAJOR_VERSION;
2128         params.drv_minor = QEDE_MINOR_VERSION;
2129         params.drv_rev = QEDE_REVISION_VERSION;
2130         params.drv_eng = QEDE_ENGINEERING_VERSION;
2131         strlcpy(params.name, "qede LAN", QED_DRV_VER_STR_SIZE);
2132         rc = qed_ops->common->slowpath_start(cdev, &params);
2133         if (rc) {
2134                 pr_notice("Cannot start slowpath\n");
2135                 goto err1;
2136         }
2137
2138         /* Learn information crucial for qede to progress */
2139         rc = qed_ops->fill_dev_info(cdev, &dev_info);
2140         if (rc)
2141                 goto err2;
2142
2143         edev = qede_alloc_etherdev(cdev, pdev, &dev_info, dp_module,
2144                                    dp_level);
2145         if (!edev) {
2146                 rc = -ENOMEM;
2147                 goto err2;
2148         }
2149
2150         qede_init_ndev(edev);
2151
2152         rc = register_netdev(edev->ndev);
2153         if (rc) {
2154                 DP_NOTICE(edev, "Cannot register net-device\n");
2155                 goto err3;
2156         }
2157
2158         edev->ops->common->set_id(cdev, edev->ndev->name, DRV_MODULE_VERSION);
2159
2160         edev->ops->register_ops(cdev, &qede_ll_ops, edev);
2161
2162         INIT_DELAYED_WORK(&edev->sp_task, qede_sp_task);
2163         mutex_init(&edev->qede_lock);
2164
2165         DP_INFO(edev, "Ending successfully qede probe\n");
2166
2167         return 0;
2168
2169 err3:
2170         free_netdev(edev->ndev);
2171 err2:
2172         qed_ops->common->slowpath_stop(cdev);
2173 err1:
2174         qed_ops->common->remove(cdev);
2175 err0:
2176         return rc;
2177 }
2178
2179 static int qede_probe(struct pci_dev *pdev, const struct pci_device_id *id)
2180 {
2181         u32 dp_module = 0;
2182         u8 dp_level = 0;
2183
2184         qede_config_debug(debug, &dp_module, &dp_level);
2185
2186         return __qede_probe(pdev, dp_module, dp_level,
2187                             QEDE_PROBE_NORMAL);
2188 }
2189
2190 enum qede_remove_mode {
2191         QEDE_REMOVE_NORMAL,
2192 };
2193
2194 static void __qede_remove(struct pci_dev *pdev, enum qede_remove_mode mode)
2195 {
2196         struct net_device *ndev = pci_get_drvdata(pdev);
2197         struct qede_dev *edev = netdev_priv(ndev);
2198         struct qed_dev *cdev = edev->cdev;
2199
2200         DP_INFO(edev, "Starting qede_remove\n");
2201
2202         cancel_delayed_work_sync(&edev->sp_task);
2203         unregister_netdev(ndev);
2204
2205         edev->ops->common->set_power_state(cdev, PCI_D0);
2206
2207         pci_set_drvdata(pdev, NULL);
2208
2209         free_netdev(ndev);
2210
2211         /* Use global ops since we've freed edev */
2212         qed_ops->common->slowpath_stop(cdev);
2213         qed_ops->common->remove(cdev);
2214
2215         pr_notice("Ending successfully qede_remove\n");
2216 }
2217
2218 static void qede_remove(struct pci_dev *pdev)
2219 {
2220         __qede_remove(pdev, QEDE_REMOVE_NORMAL);
2221 }
2222
2223 /* -------------------------------------------------------------------------
2224  * START OF LOAD / UNLOAD
2225  * -------------------------------------------------------------------------
2226  */
2227
2228 static int qede_set_num_queues(struct qede_dev *edev)
2229 {
2230         int rc;
2231         u16 rss_num;
2232
2233         /* Setup queues according to possible resources*/
2234         if (edev->req_rss)
2235                 rss_num = edev->req_rss;
2236         else
2237                 rss_num = netif_get_num_default_rss_queues() *
2238                           edev->dev_info.common.num_hwfns;
2239
2240         rss_num = min_t(u16, QEDE_MAX_RSS_CNT(edev), rss_num);
2241
2242         rc = edev->ops->common->set_fp_int(edev->cdev, rss_num);
2243         if (rc > 0) {
2244                 /* Managed to request interrupts for our queues */
2245                 edev->num_rss = rc;
2246                 DP_INFO(edev, "Managed %d [of %d] RSS queues\n",
2247                         QEDE_RSS_CNT(edev), rss_num);
2248                 rc = 0;
2249         }
2250         return rc;
2251 }
2252
2253 static void qede_free_mem_sb(struct qede_dev *edev,
2254                              struct qed_sb_info *sb_info)
2255 {
2256         if (sb_info->sb_virt)
2257                 dma_free_coherent(&edev->pdev->dev, sizeof(*sb_info->sb_virt),
2258                                   (void *)sb_info->sb_virt, sb_info->sb_phys);
2259 }
2260
2261 /* This function allocates fast-path status block memory */
2262 static int qede_alloc_mem_sb(struct qede_dev *edev,
2263                              struct qed_sb_info *sb_info,
2264                              u16 sb_id)
2265 {
2266         struct status_block *sb_virt;
2267         dma_addr_t sb_phys;
2268         int rc;
2269
2270         sb_virt = dma_alloc_coherent(&edev->pdev->dev,
2271                                      sizeof(*sb_virt),
2272                                      &sb_phys, GFP_KERNEL);
2273         if (!sb_virt) {
2274                 DP_ERR(edev, "Status block allocation failed\n");
2275                 return -ENOMEM;
2276         }
2277
2278         rc = edev->ops->common->sb_init(edev->cdev, sb_info,
2279                                         sb_virt, sb_phys, sb_id,
2280                                         QED_SB_TYPE_L2_QUEUE);
2281         if (rc) {
2282                 DP_ERR(edev, "Status block initialization failed\n");
2283                 dma_free_coherent(&edev->pdev->dev, sizeof(*sb_virt),
2284                                   sb_virt, sb_phys);
2285                 return rc;
2286         }
2287
2288         return 0;
2289 }
2290
2291 static void qede_free_rx_buffers(struct qede_dev *edev,
2292                                  struct qede_rx_queue *rxq)
2293 {
2294         u16 i;
2295
2296         for (i = rxq->sw_rx_cons; i != rxq->sw_rx_prod; i++) {
2297                 struct sw_rx_data *rx_buf;
2298                 struct page *data;
2299
2300                 rx_buf = &rxq->sw_rx_ring[i & NUM_RX_BDS_MAX];
2301                 data = rx_buf->data;
2302
2303                 dma_unmap_page(&edev->pdev->dev,
2304                                rx_buf->mapping,
2305                                PAGE_SIZE, DMA_FROM_DEVICE);
2306
2307                 rx_buf->data = NULL;
2308                 __free_page(data);
2309         }
2310 }
2311
2312 static void qede_free_sge_mem(struct qede_dev *edev,
2313                               struct qede_rx_queue *rxq) {
2314         int i;
2315
2316         if (edev->gro_disable)
2317                 return;
2318
2319         for (i = 0; i < ETH_TPA_MAX_AGGS_NUM; i++) {
2320                 struct qede_agg_info *tpa_info = &rxq->tpa_info[i];
2321                 struct sw_rx_data *replace_buf = &tpa_info->replace_buf;
2322
2323                 if (replace_buf->data) {
2324                         dma_unmap_page(&edev->pdev->dev,
2325                                        dma_unmap_addr(replace_buf, mapping),
2326                                        PAGE_SIZE, DMA_FROM_DEVICE);
2327                         __free_page(replace_buf->data);
2328                 }
2329         }
2330 }
2331
2332 static void qede_free_mem_rxq(struct qede_dev *edev,
2333                               struct qede_rx_queue *rxq)
2334 {
2335         qede_free_sge_mem(edev, rxq);
2336
2337         /* Free rx buffers */
2338         qede_free_rx_buffers(edev, rxq);
2339
2340         /* Free the parallel SW ring */
2341         kfree(rxq->sw_rx_ring);
2342
2343         /* Free the real RQ ring used by FW */
2344         edev->ops->common->chain_free(edev->cdev, &rxq->rx_bd_ring);
2345         edev->ops->common->chain_free(edev->cdev, &rxq->rx_comp_ring);
2346 }
2347
2348 static int qede_alloc_rx_buffer(struct qede_dev *edev,
2349                                 struct qede_rx_queue *rxq)
2350 {
2351         struct sw_rx_data *sw_rx_data;
2352         struct eth_rx_bd *rx_bd;
2353         dma_addr_t mapping;
2354         struct page *data;
2355         u16 rx_buf_size;
2356
2357         rx_buf_size = rxq->rx_buf_size;
2358
2359         data = alloc_pages(GFP_ATOMIC, 0);
2360         if (unlikely(!data)) {
2361                 DP_NOTICE(edev, "Failed to allocate Rx data [page]\n");
2362                 return -ENOMEM;
2363         }
2364
2365         /* Map the entire page as it would be used
2366          * for multiple RX buffer segment size mapping.
2367          */
2368         mapping = dma_map_page(&edev->pdev->dev, data, 0,
2369                                PAGE_SIZE, DMA_FROM_DEVICE);
2370         if (unlikely(dma_mapping_error(&edev->pdev->dev, mapping))) {
2371                 __free_page(data);
2372                 DP_NOTICE(edev, "Failed to map Rx buffer\n");
2373                 return -ENOMEM;
2374         }
2375
2376         sw_rx_data = &rxq->sw_rx_ring[rxq->sw_rx_prod & NUM_RX_BDS_MAX];
2377         sw_rx_data->page_offset = 0;
2378         sw_rx_data->data = data;
2379         sw_rx_data->mapping = mapping;
2380
2381         /* Advance PROD and get BD pointer */
2382         rx_bd = (struct eth_rx_bd *)qed_chain_produce(&rxq->rx_bd_ring);
2383         WARN_ON(!rx_bd);
2384         rx_bd->addr.hi = cpu_to_le32(upper_32_bits(mapping));
2385         rx_bd->addr.lo = cpu_to_le32(lower_32_bits(mapping));
2386
2387         rxq->sw_rx_prod++;
2388
2389         return 0;
2390 }
2391
2392 static int qede_alloc_sge_mem(struct qede_dev *edev,
2393                               struct qede_rx_queue *rxq)
2394 {
2395         dma_addr_t mapping;
2396         int i;
2397
2398         if (edev->gro_disable)
2399                 return 0;
2400
2401         if (edev->ndev->mtu > PAGE_SIZE) {
2402                 edev->gro_disable = 1;
2403                 return 0;
2404         }
2405
2406         for (i = 0; i < ETH_TPA_MAX_AGGS_NUM; i++) {
2407                 struct qede_agg_info *tpa_info = &rxq->tpa_info[i];
2408                 struct sw_rx_data *replace_buf = &tpa_info->replace_buf;
2409
2410                 replace_buf->data = alloc_pages(GFP_ATOMIC, 0);
2411                 if (unlikely(!replace_buf->data)) {
2412                         DP_NOTICE(edev,
2413                                   "Failed to allocate TPA skb pool [replacement buffer]\n");
2414                         goto err;
2415                 }
2416
2417                 mapping = dma_map_page(&edev->pdev->dev, replace_buf->data, 0,
2418                                        rxq->rx_buf_size, DMA_FROM_DEVICE);
2419                 if (unlikely(dma_mapping_error(&edev->pdev->dev, mapping))) {
2420                         DP_NOTICE(edev,
2421                                   "Failed to map TPA replacement buffer\n");
2422                         goto err;
2423                 }
2424
2425                 dma_unmap_addr_set(replace_buf, mapping, mapping);
2426                 tpa_info->replace_buf.page_offset = 0;
2427
2428                 tpa_info->replace_buf_mapping = mapping;
2429                 tpa_info->agg_state = QEDE_AGG_STATE_NONE;
2430         }
2431
2432         return 0;
2433 err:
2434         qede_free_sge_mem(edev, rxq);
2435         edev->gro_disable = 1;
2436         return -ENOMEM;
2437 }
2438
2439 /* This function allocates all memory needed per Rx queue */
2440 static int qede_alloc_mem_rxq(struct qede_dev *edev,
2441                               struct qede_rx_queue *rxq)
2442 {
2443         int i, rc, size;
2444
2445         rxq->num_rx_buffers = edev->q_num_rx_buffers;
2446
2447         rxq->rx_buf_size = NET_IP_ALIGN + ETH_OVERHEAD +
2448                            edev->ndev->mtu;
2449         if (rxq->rx_buf_size > PAGE_SIZE)
2450                 rxq->rx_buf_size = PAGE_SIZE;
2451
2452         /* Segment size to spilt a page in multiple equal parts */
2453         rxq->rx_buf_seg_size = roundup_pow_of_two(rxq->rx_buf_size);
2454
2455         /* Allocate the parallel driver ring for Rx buffers */
2456         size = sizeof(*rxq->sw_rx_ring) * RX_RING_SIZE;
2457         rxq->sw_rx_ring = kzalloc(size, GFP_KERNEL);
2458         if (!rxq->sw_rx_ring) {
2459                 DP_ERR(edev, "Rx buffers ring allocation failed\n");
2460                 rc = -ENOMEM;
2461                 goto err;
2462         }
2463
2464         /* Allocate FW Rx ring  */
2465         rc = edev->ops->common->chain_alloc(edev->cdev,
2466                                             QED_CHAIN_USE_TO_CONSUME_PRODUCE,
2467                                             QED_CHAIN_MODE_NEXT_PTR,
2468                                             RX_RING_SIZE,
2469                                             sizeof(struct eth_rx_bd),
2470                                             &rxq->rx_bd_ring);
2471
2472         if (rc)
2473                 goto err;
2474
2475         /* Allocate FW completion ring */
2476         rc = edev->ops->common->chain_alloc(edev->cdev,
2477                                             QED_CHAIN_USE_TO_CONSUME,
2478                                             QED_CHAIN_MODE_PBL,
2479                                             RX_RING_SIZE,
2480                                             sizeof(union eth_rx_cqe),
2481                                             &rxq->rx_comp_ring);
2482         if (rc)
2483                 goto err;
2484
2485         /* Allocate buffers for the Rx ring */
2486         for (i = 0; i < rxq->num_rx_buffers; i++) {
2487                 rc = qede_alloc_rx_buffer(edev, rxq);
2488                 if (rc) {
2489                         DP_ERR(edev,
2490                                "Rx buffers allocation failed at index %d\n", i);
2491                         goto err;
2492                 }
2493         }
2494
2495         rc = qede_alloc_sge_mem(edev, rxq);
2496 err:
2497         return rc;
2498 }
2499
2500 static void qede_free_mem_txq(struct qede_dev *edev,
2501                               struct qede_tx_queue *txq)
2502 {
2503         /* Free the parallel SW ring */
2504         kfree(txq->sw_tx_ring);
2505
2506         /* Free the real RQ ring used by FW */
2507         edev->ops->common->chain_free(edev->cdev, &txq->tx_pbl);
2508 }
2509
2510 /* This function allocates all memory needed per Tx queue */
2511 static int qede_alloc_mem_txq(struct qede_dev *edev,
2512                               struct qede_tx_queue *txq)
2513 {
2514         int size, rc;
2515         union eth_tx_bd_types *p_virt;
2516
2517         txq->num_tx_buffers = edev->q_num_tx_buffers;
2518
2519         /* Allocate the parallel driver ring for Tx buffers */
2520         size = sizeof(*txq->sw_tx_ring) * NUM_TX_BDS_MAX;
2521         txq->sw_tx_ring = kzalloc(size, GFP_KERNEL);
2522         if (!txq->sw_tx_ring) {
2523                 DP_NOTICE(edev, "Tx buffers ring allocation failed\n");
2524                 goto err;
2525         }
2526
2527         rc = edev->ops->common->chain_alloc(edev->cdev,
2528                                             QED_CHAIN_USE_TO_CONSUME_PRODUCE,
2529                                             QED_CHAIN_MODE_PBL,
2530                                             NUM_TX_BDS_MAX,
2531                                             sizeof(*p_virt),
2532                                             &txq->tx_pbl);
2533         if (rc)
2534                 goto err;
2535
2536         return 0;
2537
2538 err:
2539         qede_free_mem_txq(edev, txq);
2540         return -ENOMEM;
2541 }
2542
2543 /* This function frees all memory of a single fp */
2544 static void qede_free_mem_fp(struct qede_dev *edev,
2545                              struct qede_fastpath *fp)
2546 {
2547         int tc;
2548
2549         qede_free_mem_sb(edev, fp->sb_info);
2550
2551         qede_free_mem_rxq(edev, fp->rxq);
2552
2553         for (tc = 0; tc < edev->num_tc; tc++)
2554                 qede_free_mem_txq(edev, &fp->txqs[tc]);
2555 }
2556
2557 /* This function allocates all memory needed for a single fp (i.e. an entity
2558  * which contains status block, one rx queue and multiple per-TC tx queues.
2559  */
2560 static int qede_alloc_mem_fp(struct qede_dev *edev,
2561                              struct qede_fastpath *fp)
2562 {
2563         int rc, tc;
2564
2565         rc = qede_alloc_mem_sb(edev, fp->sb_info, fp->rss_id);
2566         if (rc)
2567                 goto err;
2568
2569         rc = qede_alloc_mem_rxq(edev, fp->rxq);
2570         if (rc)
2571                 goto err;
2572
2573         for (tc = 0; tc < edev->num_tc; tc++) {
2574                 rc = qede_alloc_mem_txq(edev, &fp->txqs[tc]);
2575                 if (rc)
2576                         goto err;
2577         }
2578
2579         return 0;
2580 err:
2581         return rc;
2582 }
2583
2584 static void qede_free_mem_load(struct qede_dev *edev)
2585 {
2586         int i;
2587
2588         for_each_rss(i) {
2589                 struct qede_fastpath *fp = &edev->fp_array[i];
2590
2591                 qede_free_mem_fp(edev, fp);
2592         }
2593 }
2594
2595 /* This function allocates all qede memory at NIC load. */
2596 static int qede_alloc_mem_load(struct qede_dev *edev)
2597 {
2598         int rc = 0, rss_id;
2599
2600         for (rss_id = 0; rss_id < QEDE_RSS_CNT(edev); rss_id++) {
2601                 struct qede_fastpath *fp = &edev->fp_array[rss_id];
2602
2603                 rc = qede_alloc_mem_fp(edev, fp);
2604                 if (rc) {
2605                         DP_ERR(edev,
2606                                "Failed to allocate memory for fastpath - rss id = %d\n",
2607                                rss_id);
2608                         qede_free_mem_load(edev);
2609                         return rc;
2610                 }
2611         }
2612
2613         return 0;
2614 }
2615
2616 /* This function inits fp content and resets the SB, RXQ and TXQ structures */
2617 static void qede_init_fp(struct qede_dev *edev)
2618 {
2619         int rss_id, txq_index, tc;
2620         struct qede_fastpath *fp;
2621
2622         for_each_rss(rss_id) {
2623                 fp = &edev->fp_array[rss_id];
2624
2625                 fp->edev = edev;
2626                 fp->rss_id = rss_id;
2627
2628                 memset((void *)&fp->napi, 0, sizeof(fp->napi));
2629
2630                 memset((void *)fp->sb_info, 0, sizeof(*fp->sb_info));
2631
2632                 memset((void *)fp->rxq, 0, sizeof(*fp->rxq));
2633                 fp->rxq->rxq_id = rss_id;
2634
2635                 memset((void *)fp->txqs, 0, (edev->num_tc * sizeof(*fp->txqs)));
2636                 for (tc = 0; tc < edev->num_tc; tc++) {
2637                         txq_index = tc * QEDE_RSS_CNT(edev) + rss_id;
2638                         fp->txqs[tc].index = txq_index;
2639                 }
2640
2641                 snprintf(fp->name, sizeof(fp->name), "%s-fp-%d",
2642                          edev->ndev->name, rss_id);
2643         }
2644
2645         edev->gro_disable = !(edev->ndev->features & NETIF_F_GRO);
2646 }
2647
2648 static int qede_set_real_num_queues(struct qede_dev *edev)
2649 {
2650         int rc = 0;
2651
2652         rc = netif_set_real_num_tx_queues(edev->ndev, QEDE_TSS_CNT(edev));
2653         if (rc) {
2654                 DP_NOTICE(edev, "Failed to set real number of Tx queues\n");
2655                 return rc;
2656         }
2657         rc = netif_set_real_num_rx_queues(edev->ndev, QEDE_RSS_CNT(edev));
2658         if (rc) {
2659                 DP_NOTICE(edev, "Failed to set real number of Rx queues\n");
2660                 return rc;
2661         }
2662
2663         return 0;
2664 }
2665
2666 static void qede_napi_disable_remove(struct qede_dev *edev)
2667 {
2668         int i;
2669
2670         for_each_rss(i) {
2671                 napi_disable(&edev->fp_array[i].napi);
2672
2673                 netif_napi_del(&edev->fp_array[i].napi);
2674         }
2675 }
2676
2677 static void qede_napi_add_enable(struct qede_dev *edev)
2678 {
2679         int i;
2680
2681         /* Add NAPI objects */
2682         for_each_rss(i) {
2683                 netif_napi_add(edev->ndev, &edev->fp_array[i].napi,
2684                                qede_poll, NAPI_POLL_WEIGHT);
2685                 napi_enable(&edev->fp_array[i].napi);
2686         }
2687 }
2688
2689 static void qede_sync_free_irqs(struct qede_dev *edev)
2690 {
2691         int i;
2692
2693         for (i = 0; i < edev->int_info.used_cnt; i++) {
2694                 if (edev->int_info.msix_cnt) {
2695                         synchronize_irq(edev->int_info.msix[i].vector);
2696                         free_irq(edev->int_info.msix[i].vector,
2697                                  &edev->fp_array[i]);
2698                 } else {
2699                         edev->ops->common->simd_handler_clean(edev->cdev, i);
2700                 }
2701         }
2702
2703         edev->int_info.used_cnt = 0;
2704 }
2705
2706 static int qede_req_msix_irqs(struct qede_dev *edev)
2707 {
2708         int i, rc;
2709
2710         /* Sanitize number of interrupts == number of prepared RSS queues */
2711         if (QEDE_RSS_CNT(edev) > edev->int_info.msix_cnt) {
2712                 DP_ERR(edev,
2713                        "Interrupt mismatch: %d RSS queues > %d MSI-x vectors\n",
2714                        QEDE_RSS_CNT(edev), edev->int_info.msix_cnt);
2715                 return -EINVAL;
2716         }
2717
2718         for (i = 0; i < QEDE_RSS_CNT(edev); i++) {
2719                 rc = request_irq(edev->int_info.msix[i].vector,
2720                                  qede_msix_fp_int, 0, edev->fp_array[i].name,
2721                                  &edev->fp_array[i]);
2722                 if (rc) {
2723                         DP_ERR(edev, "Request fp %d irq failed\n", i);
2724                         qede_sync_free_irqs(edev);
2725                         return rc;
2726                 }
2727                 DP_VERBOSE(edev, NETIF_MSG_INTR,
2728                            "Requested fp irq for %s [entry %d]. Cookie is at %p\n",
2729                            edev->fp_array[i].name, i,
2730                            &edev->fp_array[i]);
2731                 edev->int_info.used_cnt++;
2732         }
2733
2734         return 0;
2735 }
2736
2737 static void qede_simd_fp_handler(void *cookie)
2738 {
2739         struct qede_fastpath *fp = (struct qede_fastpath *)cookie;
2740
2741         napi_schedule_irqoff(&fp->napi);
2742 }
2743
2744 static int qede_setup_irqs(struct qede_dev *edev)
2745 {
2746         int i, rc = 0;
2747
2748         /* Learn Interrupt configuration */
2749         rc = edev->ops->common->get_fp_int(edev->cdev, &edev->int_info);
2750         if (rc)
2751                 return rc;
2752
2753         if (edev->int_info.msix_cnt) {
2754                 rc = qede_req_msix_irqs(edev);
2755                 if (rc)
2756                         return rc;
2757                 edev->ndev->irq = edev->int_info.msix[0].vector;
2758         } else {
2759                 const struct qed_common_ops *ops;
2760
2761                 /* qed should learn receive the RSS ids and callbacks */
2762                 ops = edev->ops->common;
2763                 for (i = 0; i < QEDE_RSS_CNT(edev); i++)
2764                         ops->simd_handler_config(edev->cdev,
2765                                                  &edev->fp_array[i], i,
2766                                                  qede_simd_fp_handler);
2767                 edev->int_info.used_cnt = QEDE_RSS_CNT(edev);
2768         }
2769         return 0;
2770 }
2771
2772 static int qede_drain_txq(struct qede_dev *edev,
2773                           struct qede_tx_queue *txq,
2774                           bool allow_drain)
2775 {
2776         int rc, cnt = 1000;
2777
2778         while (txq->sw_tx_cons != txq->sw_tx_prod) {
2779                 if (!cnt) {
2780                         if (allow_drain) {
2781                                 DP_NOTICE(edev,
2782                                           "Tx queue[%d] is stuck, requesting MCP to drain\n",
2783                                           txq->index);
2784                                 rc = edev->ops->common->drain(edev->cdev);
2785                                 if (rc)
2786                                         return rc;
2787                                 return qede_drain_txq(edev, txq, false);
2788                         }
2789                         DP_NOTICE(edev,
2790                                   "Timeout waiting for tx queue[%d]: PROD=%d, CONS=%d\n",
2791                                   txq->index, txq->sw_tx_prod,
2792                                   txq->sw_tx_cons);
2793                         return -ENODEV;
2794                 }
2795                 cnt--;
2796                 usleep_range(1000, 2000);
2797                 barrier();
2798         }
2799
2800         /* FW finished processing, wait for HW to transmit all tx packets */
2801         usleep_range(1000, 2000);
2802
2803         return 0;
2804 }
2805
2806 static int qede_stop_queues(struct qede_dev *edev)
2807 {
2808         struct qed_update_vport_params vport_update_params;
2809         struct qed_dev *cdev = edev->cdev;
2810         int rc, tc, i;
2811
2812         /* Disable the vport */
2813         memset(&vport_update_params, 0, sizeof(vport_update_params));
2814         vport_update_params.vport_id = 0;
2815         vport_update_params.update_vport_active_flg = 1;
2816         vport_update_params.vport_active_flg = 0;
2817         vport_update_params.update_rss_flg = 0;
2818
2819         rc = edev->ops->vport_update(cdev, &vport_update_params);
2820         if (rc) {
2821                 DP_ERR(edev, "Failed to update vport\n");
2822                 return rc;
2823         }
2824
2825         /* Flush Tx queues. If needed, request drain from MCP */
2826         for_each_rss(i) {
2827                 struct qede_fastpath *fp = &edev->fp_array[i];
2828
2829                 for (tc = 0; tc < edev->num_tc; tc++) {
2830                         struct qede_tx_queue *txq = &fp->txqs[tc];
2831
2832                         rc = qede_drain_txq(edev, txq, true);
2833                         if (rc)
2834                                 return rc;
2835                 }
2836         }
2837
2838         /* Stop all Queues in reverse order*/
2839         for (i = QEDE_RSS_CNT(edev) - 1; i >= 0; i--) {
2840                 struct qed_stop_rxq_params rx_params;
2841
2842                 /* Stop the Tx Queue(s)*/
2843                 for (tc = 0; tc < edev->num_tc; tc++) {
2844                         struct qed_stop_txq_params tx_params;
2845
2846                         tx_params.rss_id = i;
2847                         tx_params.tx_queue_id = tc * QEDE_RSS_CNT(edev) + i;
2848                         rc = edev->ops->q_tx_stop(cdev, &tx_params);
2849                         if (rc) {
2850                                 DP_ERR(edev, "Failed to stop TXQ #%d\n",
2851                                        tx_params.tx_queue_id);
2852                                 return rc;
2853                         }
2854                 }
2855
2856                 /* Stop the Rx Queue*/
2857                 memset(&rx_params, 0, sizeof(rx_params));
2858                 rx_params.rss_id = i;
2859                 rx_params.rx_queue_id = i;
2860
2861                 rc = edev->ops->q_rx_stop(cdev, &rx_params);
2862                 if (rc) {
2863                         DP_ERR(edev, "Failed to stop RXQ #%d\n", i);
2864                         return rc;
2865                 }
2866         }
2867
2868         /* Stop the vport */
2869         rc = edev->ops->vport_stop(cdev, 0);
2870         if (rc)
2871                 DP_ERR(edev, "Failed to stop VPORT\n");
2872
2873         return rc;
2874 }
2875
2876 static int qede_start_queues(struct qede_dev *edev)
2877 {
2878         int rc, tc, i;
2879         int vlan_removal_en = 1;
2880         struct qed_dev *cdev = edev->cdev;
2881         struct qed_update_vport_rss_params *rss_params = &edev->rss_params;
2882         struct qed_update_vport_params vport_update_params;
2883         struct qed_queue_start_common_params q_params;
2884         struct qed_start_vport_params start = {0};
2885
2886         if (!edev->num_rss) {
2887                 DP_ERR(edev,
2888                        "Cannot update V-VPORT as active as there are no Rx queues\n");
2889                 return -EINVAL;
2890         }
2891
2892         start.gro_enable = !edev->gro_disable;
2893         start.mtu = edev->ndev->mtu;
2894         start.vport_id = 0;
2895         start.drop_ttl0 = true;
2896         start.remove_inner_vlan = vlan_removal_en;
2897
2898         rc = edev->ops->vport_start(cdev, &start);
2899
2900         if (rc) {
2901                 DP_ERR(edev, "Start V-PORT failed %d\n", rc);
2902                 return rc;
2903         }
2904
2905         DP_VERBOSE(edev, NETIF_MSG_IFUP,
2906                    "Start vport ramrod passed, vport_id = %d, MTU = %d, vlan_removal_en = %d\n",
2907                    start.vport_id, edev->ndev->mtu + 0xe, vlan_removal_en);
2908
2909         for_each_rss(i) {
2910                 struct qede_fastpath *fp = &edev->fp_array[i];
2911                 dma_addr_t phys_table = fp->rxq->rx_comp_ring.pbl.p_phys_table;
2912
2913                 memset(&q_params, 0, sizeof(q_params));
2914                 q_params.rss_id = i;
2915                 q_params.queue_id = i;
2916                 q_params.vport_id = 0;
2917                 q_params.sb = fp->sb_info->igu_sb_id;
2918                 q_params.sb_idx = RX_PI;
2919
2920                 rc = edev->ops->q_rx_start(cdev, &q_params,
2921                                            fp->rxq->rx_buf_size,
2922                                            fp->rxq->rx_bd_ring.p_phys_addr,
2923                                            phys_table,
2924                                            fp->rxq->rx_comp_ring.page_cnt,
2925                                            &fp->rxq->hw_rxq_prod_addr);
2926                 if (rc) {
2927                         DP_ERR(edev, "Start RXQ #%d failed %d\n", i, rc);
2928                         return rc;
2929                 }
2930
2931                 fp->rxq->hw_cons_ptr = &fp->sb_info->sb_virt->pi_array[RX_PI];
2932
2933                 qede_update_rx_prod(edev, fp->rxq);
2934
2935                 for (tc = 0; tc < edev->num_tc; tc++) {
2936                         struct qede_tx_queue *txq = &fp->txqs[tc];
2937                         int txq_index = tc * QEDE_RSS_CNT(edev) + i;
2938
2939                         memset(&q_params, 0, sizeof(q_params));
2940                         q_params.rss_id = i;
2941                         q_params.queue_id = txq_index;
2942                         q_params.vport_id = 0;
2943                         q_params.sb = fp->sb_info->igu_sb_id;
2944                         q_params.sb_idx = TX_PI(tc);
2945
2946                         rc = edev->ops->q_tx_start(cdev, &q_params,
2947                                                    txq->tx_pbl.pbl.p_phys_table,
2948                                                    txq->tx_pbl.page_cnt,
2949                                                    &txq->doorbell_addr);
2950                         if (rc) {
2951                                 DP_ERR(edev, "Start TXQ #%d failed %d\n",
2952                                        txq_index, rc);
2953                                 return rc;
2954                         }
2955
2956                         txq->hw_cons_ptr =
2957                                 &fp->sb_info->sb_virt->pi_array[TX_PI(tc)];
2958                         SET_FIELD(txq->tx_db.data.params,
2959                                   ETH_DB_DATA_DEST, DB_DEST_XCM);
2960                         SET_FIELD(txq->tx_db.data.params, ETH_DB_DATA_AGG_CMD,
2961                                   DB_AGG_CMD_SET);
2962                         SET_FIELD(txq->tx_db.data.params,
2963                                   ETH_DB_DATA_AGG_VAL_SEL,
2964                                   DQ_XCM_ETH_TX_BD_PROD_CMD);
2965
2966                         txq->tx_db.data.agg_flags = DQ_XCM_ETH_DQ_CF_CMD;
2967                 }
2968         }
2969
2970         /* Prepare and send the vport enable */
2971         memset(&vport_update_params, 0, sizeof(vport_update_params));
2972         vport_update_params.vport_id = start.vport_id;
2973         vport_update_params.update_vport_active_flg = 1;
2974         vport_update_params.vport_active_flg = 1;
2975
2976         /* Fill struct with RSS params */
2977         if (QEDE_RSS_CNT(edev) > 1) {
2978                 vport_update_params.update_rss_flg = 1;
2979                 for (i = 0; i < 128; i++)
2980                         rss_params->rss_ind_table[i] =
2981                         ethtool_rxfh_indir_default(i, QEDE_RSS_CNT(edev));
2982                 netdev_rss_key_fill(rss_params->rss_key,
2983                                     sizeof(rss_params->rss_key));
2984         } else {
2985                 memset(rss_params, 0, sizeof(*rss_params));
2986         }
2987         memcpy(&vport_update_params.rss_params, rss_params,
2988                sizeof(*rss_params));
2989
2990         rc = edev->ops->vport_update(cdev, &vport_update_params);
2991         if (rc) {
2992                 DP_ERR(edev, "Update V-PORT failed %d\n", rc);
2993                 return rc;
2994         }
2995
2996         return 0;
2997 }
2998
2999 static int qede_set_mcast_rx_mac(struct qede_dev *edev,
3000                                  enum qed_filter_xcast_params_type opcode,
3001                                  unsigned char *mac, int num_macs)
3002 {
3003         struct qed_filter_params filter_cmd;
3004         int i;
3005
3006         memset(&filter_cmd, 0, sizeof(filter_cmd));
3007         filter_cmd.type = QED_FILTER_TYPE_MCAST;
3008         filter_cmd.filter.mcast.type = opcode;
3009         filter_cmd.filter.mcast.num = num_macs;
3010
3011         for (i = 0; i < num_macs; i++, mac += ETH_ALEN)
3012                 ether_addr_copy(filter_cmd.filter.mcast.mac[i], mac);
3013
3014         return edev->ops->filter_config(edev->cdev, &filter_cmd);
3015 }
3016
3017 enum qede_unload_mode {
3018         QEDE_UNLOAD_NORMAL,
3019 };
3020
3021 static void qede_unload(struct qede_dev *edev, enum qede_unload_mode mode)
3022 {
3023         struct qed_link_params link_params;
3024         int rc;
3025
3026         DP_INFO(edev, "Starting qede unload\n");
3027
3028         mutex_lock(&edev->qede_lock);
3029         edev->state = QEDE_STATE_CLOSED;
3030
3031         /* Close OS Tx */
3032         netif_tx_disable(edev->ndev);
3033         netif_carrier_off(edev->ndev);
3034
3035         /* Reset the link */
3036         memset(&link_params, 0, sizeof(link_params));
3037         link_params.link_up = false;
3038         edev->ops->common->set_link(edev->cdev, &link_params);
3039         rc = qede_stop_queues(edev);
3040         if (rc) {
3041                 qede_sync_free_irqs(edev);
3042                 goto out;
3043         }
3044
3045         DP_INFO(edev, "Stopped Queues\n");
3046
3047         qede_vlan_mark_nonconfigured(edev);
3048         edev->ops->fastpath_stop(edev->cdev);
3049
3050         /* Release the interrupts */
3051         qede_sync_free_irqs(edev);
3052         edev->ops->common->set_fp_int(edev->cdev, 0);
3053
3054         qede_napi_disable_remove(edev);
3055
3056         qede_free_mem_load(edev);
3057         qede_free_fp_array(edev);
3058
3059 out:
3060         mutex_unlock(&edev->qede_lock);
3061         DP_INFO(edev, "Ending qede unload\n");
3062 }
3063
3064 enum qede_load_mode {
3065         QEDE_LOAD_NORMAL,
3066 };
3067
3068 static int qede_load(struct qede_dev *edev, enum qede_load_mode mode)
3069 {
3070         struct qed_link_params link_params;
3071         struct qed_link_output link_output;
3072         int rc;
3073
3074         DP_INFO(edev, "Starting qede load\n");
3075
3076         rc = qede_set_num_queues(edev);
3077         if (rc)
3078                 goto err0;
3079
3080         rc = qede_alloc_fp_array(edev);
3081         if (rc)
3082                 goto err0;
3083
3084         qede_init_fp(edev);
3085
3086         rc = qede_alloc_mem_load(edev);
3087         if (rc)
3088                 goto err1;
3089         DP_INFO(edev, "Allocated %d RSS queues on %d TC/s\n",
3090                 QEDE_RSS_CNT(edev), edev->num_tc);
3091
3092         rc = qede_set_real_num_queues(edev);
3093         if (rc)
3094                 goto err2;
3095
3096         qede_napi_add_enable(edev);
3097         DP_INFO(edev, "Napi added and enabled\n");
3098
3099         rc = qede_setup_irqs(edev);
3100         if (rc)
3101                 goto err3;
3102         DP_INFO(edev, "Setup IRQs succeeded\n");
3103
3104         rc = qede_start_queues(edev);
3105         if (rc)
3106                 goto err4;
3107         DP_INFO(edev, "Start VPORT, RXQ and TXQ succeeded\n");
3108
3109         /* Add primary mac and set Rx filters */
3110         ether_addr_copy(edev->primary_mac, edev->ndev->dev_addr);
3111
3112         mutex_lock(&edev->qede_lock);
3113         edev->state = QEDE_STATE_OPEN;
3114         mutex_unlock(&edev->qede_lock);
3115
3116         /* Program un-configured VLANs */
3117         qede_configure_vlan_filters(edev);
3118
3119         /* Ask for link-up using current configuration */
3120         memset(&link_params, 0, sizeof(link_params));
3121         link_params.link_up = true;
3122         edev->ops->common->set_link(edev->cdev, &link_params);
3123
3124         /* Query whether link is already-up */
3125         memset(&link_output, 0, sizeof(link_output));
3126         edev->ops->common->get_link(edev->cdev, &link_output);
3127         qede_link_update(edev, &link_output);
3128
3129         DP_INFO(edev, "Ending successfully qede load\n");
3130
3131         return 0;
3132
3133 err4:
3134         qede_sync_free_irqs(edev);
3135         memset(&edev->int_info.msix_cnt, 0, sizeof(struct qed_int_info));
3136 err3:
3137         qede_napi_disable_remove(edev);
3138 err2:
3139         qede_free_mem_load(edev);
3140 err1:
3141         edev->ops->common->set_fp_int(edev->cdev, 0);
3142         qede_free_fp_array(edev);
3143         edev->num_rss = 0;
3144 err0:
3145         return rc;
3146 }
3147
3148 void qede_reload(struct qede_dev *edev,
3149                  void (*func)(struct qede_dev *, union qede_reload_args *),
3150                  union qede_reload_args *args)
3151 {
3152         qede_unload(edev, QEDE_UNLOAD_NORMAL);
3153         /* Call function handler to update parameters
3154          * needed for function load.
3155          */
3156         if (func)
3157                 func(edev, args);
3158
3159         qede_load(edev, QEDE_LOAD_NORMAL);
3160
3161         mutex_lock(&edev->qede_lock);
3162         qede_config_rx_mode(edev->ndev);
3163         mutex_unlock(&edev->qede_lock);
3164 }
3165
3166 /* called with rtnl_lock */
3167 static int qede_open(struct net_device *ndev)
3168 {
3169         struct qede_dev *edev = netdev_priv(ndev);
3170
3171         netif_carrier_off(ndev);
3172
3173         edev->ops->common->set_power_state(edev->cdev, PCI_D0);
3174
3175         return qede_load(edev, QEDE_LOAD_NORMAL);
3176 }
3177
3178 static int qede_close(struct net_device *ndev)
3179 {
3180         struct qede_dev *edev = netdev_priv(ndev);
3181
3182         qede_unload(edev, QEDE_UNLOAD_NORMAL);
3183
3184         return 0;
3185 }
3186
3187 static void qede_link_update(void *dev, struct qed_link_output *link)
3188 {
3189         struct qede_dev *edev = dev;
3190
3191         if (!netif_running(edev->ndev)) {
3192                 DP_VERBOSE(edev, NETIF_MSG_LINK, "Interface is not running\n");
3193                 return;
3194         }
3195
3196         if (link->link_up) {
3197                 if (!netif_carrier_ok(edev->ndev)) {
3198                         DP_NOTICE(edev, "Link is up\n");
3199                         netif_tx_start_all_queues(edev->ndev);
3200                         netif_carrier_on(edev->ndev);
3201                 }
3202         } else {
3203                 if (netif_carrier_ok(edev->ndev)) {
3204                         DP_NOTICE(edev, "Link is down\n");
3205                         netif_tx_disable(edev->ndev);
3206                         netif_carrier_off(edev->ndev);
3207                 }
3208         }
3209 }
3210
3211 static int qede_set_mac_addr(struct net_device *ndev, void *p)
3212 {
3213         struct qede_dev *edev = netdev_priv(ndev);
3214         struct sockaddr *addr = p;
3215         int rc;
3216
3217         ASSERT_RTNL(); /* @@@TBD To be removed */
3218
3219         DP_INFO(edev, "Set_mac_addr called\n");
3220
3221         if (!is_valid_ether_addr(addr->sa_data)) {
3222                 DP_NOTICE(edev, "The MAC address is not valid\n");
3223                 return -EFAULT;
3224         }
3225
3226         ether_addr_copy(ndev->dev_addr, addr->sa_data);
3227
3228         if (!netif_running(ndev))  {
3229                 DP_NOTICE(edev, "The device is currently down\n");
3230                 return 0;
3231         }
3232
3233         /* Remove the previous primary mac */
3234         rc = qede_set_ucast_rx_mac(edev, QED_FILTER_XCAST_TYPE_DEL,
3235                                    edev->primary_mac);
3236         if (rc)
3237                 return rc;
3238
3239         /* Add MAC filter according to the new unicast HW MAC address */
3240         ether_addr_copy(edev->primary_mac, ndev->dev_addr);
3241         return qede_set_ucast_rx_mac(edev, QED_FILTER_XCAST_TYPE_ADD,
3242                                       edev->primary_mac);
3243 }
3244
3245 static int
3246 qede_configure_mcast_filtering(struct net_device *ndev,
3247                                enum qed_filter_rx_mode_type *accept_flags)
3248 {
3249         struct qede_dev *edev = netdev_priv(ndev);
3250         unsigned char *mc_macs, *temp;
3251         struct netdev_hw_addr *ha;
3252         int rc = 0, mc_count;
3253         size_t size;
3254
3255         size = 64 * ETH_ALEN;
3256
3257         mc_macs = kzalloc(size, GFP_KERNEL);
3258         if (!mc_macs) {
3259                 DP_NOTICE(edev,
3260                           "Failed to allocate memory for multicast MACs\n");
3261                 rc = -ENOMEM;
3262                 goto exit;
3263         }
3264
3265         temp = mc_macs;
3266
3267         /* Remove all previously configured MAC filters */
3268         rc = qede_set_mcast_rx_mac(edev, QED_FILTER_XCAST_TYPE_DEL,
3269                                    mc_macs, 1);
3270         if (rc)
3271                 goto exit;
3272
3273         netif_addr_lock_bh(ndev);
3274
3275         mc_count = netdev_mc_count(ndev);
3276         if (mc_count < 64) {
3277                 netdev_for_each_mc_addr(ha, ndev) {
3278                         ether_addr_copy(temp, ha->addr);
3279                         temp += ETH_ALEN;
3280                 }
3281         }
3282
3283         netif_addr_unlock_bh(ndev);
3284
3285         /* Check for all multicast @@@TBD resource allocation */
3286         if ((ndev->flags & IFF_ALLMULTI) ||
3287             (mc_count > 64)) {
3288                 if (*accept_flags == QED_FILTER_RX_MODE_TYPE_REGULAR)
3289                         *accept_flags = QED_FILTER_RX_MODE_TYPE_MULTI_PROMISC;
3290         } else {
3291                 /* Add all multicast MAC filters */
3292                 rc = qede_set_mcast_rx_mac(edev, QED_FILTER_XCAST_TYPE_ADD,
3293                                            mc_macs, mc_count);
3294         }
3295
3296 exit:
3297         kfree(mc_macs);
3298         return rc;
3299 }
3300
3301 static void qede_set_rx_mode(struct net_device *ndev)
3302 {
3303         struct qede_dev *edev = netdev_priv(ndev);
3304
3305         DP_INFO(edev, "qede_set_rx_mode called\n");
3306
3307         if (edev->state != QEDE_STATE_OPEN) {
3308                 DP_INFO(edev,
3309                         "qede_set_rx_mode called while interface is down\n");
3310         } else {
3311                 set_bit(QEDE_SP_RX_MODE, &edev->sp_flags);
3312                 schedule_delayed_work(&edev->sp_task, 0);
3313         }
3314 }
3315
3316 /* Must be called with qede_lock held */
3317 static void qede_config_rx_mode(struct net_device *ndev)
3318 {
3319         enum qed_filter_rx_mode_type accept_flags = QED_FILTER_TYPE_UCAST;
3320         struct qede_dev *edev = netdev_priv(ndev);
3321         struct qed_filter_params rx_mode;
3322         unsigned char *uc_macs, *temp;
3323         struct netdev_hw_addr *ha;
3324         int rc, uc_count;
3325         size_t size;
3326
3327         netif_addr_lock_bh(ndev);
3328
3329         uc_count = netdev_uc_count(ndev);
3330         size = uc_count * ETH_ALEN;
3331
3332         uc_macs = kzalloc(size, GFP_ATOMIC);
3333         if (!uc_macs) {
3334                 DP_NOTICE(edev, "Failed to allocate memory for unicast MACs\n");
3335                 netif_addr_unlock_bh(ndev);
3336                 return;
3337         }
3338
3339         temp = uc_macs;
3340         netdev_for_each_uc_addr(ha, ndev) {
3341                 ether_addr_copy(temp, ha->addr);
3342                 temp += ETH_ALEN;
3343         }
3344
3345         netif_addr_unlock_bh(ndev);
3346
3347         /* Configure the struct for the Rx mode */
3348         memset(&rx_mode, 0, sizeof(struct qed_filter_params));
3349         rx_mode.type = QED_FILTER_TYPE_RX_MODE;
3350
3351         /* Remove all previous unicast secondary macs and multicast macs
3352          * (configrue / leave the primary mac)
3353          */
3354         rc = qede_set_ucast_rx_mac(edev, QED_FILTER_XCAST_TYPE_REPLACE,
3355                                    edev->primary_mac);
3356         if (rc)
3357                 goto out;
3358
3359         /* Check for promiscuous */
3360         if ((ndev->flags & IFF_PROMISC) ||
3361             (uc_count > 15)) { /* @@@TBD resource allocation - 1 */
3362                 accept_flags = QED_FILTER_RX_MODE_TYPE_PROMISC;
3363         } else {
3364                 /* Add MAC filters according to the unicast secondary macs */
3365                 int i;
3366
3367                 temp = uc_macs;
3368                 for (i = 0; i < uc_count; i++) {
3369                         rc = qede_set_ucast_rx_mac(edev,
3370                                                    QED_FILTER_XCAST_TYPE_ADD,
3371                                                    temp);
3372                         if (rc)
3373                                 goto out;
3374
3375                         temp += ETH_ALEN;
3376                 }
3377
3378                 rc = qede_configure_mcast_filtering(ndev, &accept_flags);
3379                 if (rc)
3380                         goto out;
3381         }
3382
3383         /* take care of VLAN mode */
3384         if (ndev->flags & IFF_PROMISC) {
3385                 qede_config_accept_any_vlan(edev, true);
3386         } else if (!edev->non_configured_vlans) {
3387                 /* It's possible that accept_any_vlan mode is set due to a
3388                  * previous setting of IFF_PROMISC. If vlan credits are
3389                  * sufficient, disable accept_any_vlan.
3390                  */
3391                 qede_config_accept_any_vlan(edev, false);
3392         }
3393
3394         rx_mode.filter.accept_flags = accept_flags;
3395         edev->ops->filter_config(edev->cdev, &rx_mode);
3396 out:
3397         kfree(uc_macs);
3398 }