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 #ifdef CONFIG_QEDE_VXLAN
  28 #include <net/vxlan.h>
  29 #endif
  30 #include <linux/ip.h>
  31 #include <net/ipv6.h>
  32 #include <net/tcp.h>
  33 #include <linux/if_ether.h>
  34 #include <linux/if_vlan.h>
  35 #include <linux/pkt_sched.h>
  36 #include <linux/ethtool.h>
  37 #include <linux/in.h>
  38 #include <linux/random.h>
  39 #include <net/ip6_checksum.h>
  40 #include <linux/bitops.h>
  41
  42 #include "qede.h"
  43
  44 static char version[] =
  45         "QLogic FastLinQ 4xxxx Ethernet Driver qede " DRV_MODULE_VERSION "\n";
  46
  47 MODULE_DESCRIPTION("QLogic FastLinQ 4xxxx Ethernet Driver");
  48 MODULE_LICENSE("GPL");
  49 MODULE_VERSION(DRV_MODULE_VERSION);
  50
  51 static uint debug;
  52 module_param(debug, uint, 0);
  53 MODULE_PARM_DESC(debug, " Default debug msglevel");
  54
  55 static const struct qed_eth_ops *qed_ops;
  56
  57 #define CHIP_NUM_57980S_40              0x1634
  58 #define CHIP_NUM_57980S_10              0x1666
  59 #define CHIP_NUM_57980S_MF              0x1636
  60 #define CHIP_NUM_57980S_100             0x1644
  61 #define CHIP_NUM_57980S_50              0x1654
  62 #define CHIP_NUM_57980S_25              0x1656
  63
  64 #ifndef PCI_DEVICE_ID_NX2_57980E
  65 #define PCI_DEVICE_ID_57980S_40         CHIP_NUM_57980S_40
  66 #define PCI_DEVICE_ID_57980S_10         CHIP_NUM_57980S_10
  67 #define PCI_DEVICE_ID_57980S_MF         CHIP_NUM_57980S_MF
  68 #define PCI_DEVICE_ID_57980S_100        CHIP_NUM_57980S_100
  69 #define PCI_DEVICE_ID_57980S_50         CHIP_NUM_57980S_50
  70 #define PCI_DEVICE_ID_57980S_25         CHIP_NUM_57980S_25
  71 #endif
  72
  73 static const struct pci_device_id qede_pci_tbl[] = {
  74         { PCI_VDEVICE(QLOGIC, PCI_DEVICE_ID_57980S_40), 0 },
  75         { PCI_VDEVICE(QLOGIC, PCI_DEVICE_ID_57980S_10), 0 },
  76         { PCI_VDEVICE(QLOGIC, PCI_DEVICE_ID_57980S_MF), 0 },
  77         { PCI_VDEVICE(QLOGIC, PCI_DEVICE_ID_57980S_100), 0 },
  78         { PCI_VDEVICE(QLOGIC, PCI_DEVICE_ID_57980S_50), 0 },
  79         { PCI_VDEVICE(QLOGIC, PCI_DEVICE_ID_57980S_25), 0 },
  80         { 0 }
  81 };
  82
  83 MODULE_DEVICE_TABLE(pci, qede_pci_tbl);
  84
  85 static int qede_probe(struct pci_dev *pdev, const struct pci_device_id *id);
  86
  87 #define TX_TIMEOUT              (5 * HZ)
  88
  89 static void qede_remove(struct pci_dev *pdev);
  90 static int qede_alloc_rx_buffer(struct qede_dev *edev,
  91                                 struct qede_rx_queue *rxq);
  92 static void qede_link_update(void *dev, struct qed_link_output *link);
  93
  94 static struct pci_driver qede_pci_driver = {
  95         .name = "qede",
  96         .id_table = qede_pci_tbl,
  97         .probe = qede_probe,
  98         .remove = qede_remove,
  99 };
 100
 101 static struct qed_eth_cb_ops qede_ll_ops = {
 102         {
 103                 .link_update = qede_link_update,
 104         },
 105 };
 106
 107 static int qede_netdev_event(struct notifier_block *this, unsigned long event,
 108                              void *ptr)
 109 {
 110         struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
 111         struct ethtool_drvinfo drvinfo;
 112         struct qede_dev *edev;
 113
 114         /* Currently only support name change */
 115         if (event != NETDEV_CHANGENAME)
 116                 goto done;
 117
 118         /* Check whether this is a qede device */
 119         if (!ndev || !ndev->ethtool_ops || !ndev->ethtool_ops->get_drvinfo)
 120                 goto done;
 121
 122         memset(&drvinfo, 0, sizeof(drvinfo));
 123         ndev->ethtool_ops->get_drvinfo(ndev, &drvinfo);
 124         if (strcmp(drvinfo.driver, "qede"))
 125                 goto done;
 126         edev = netdev_priv(ndev);
 127
 128         /* Notify qed of the name change */
 129         if (!edev->ops || !edev->ops->common)
 130                 goto done;
 131         edev->ops->common->set_id(edev->cdev, edev->ndev->name,
 132                                   "qede");
 133
 134 done:
 135         return NOTIFY_DONE;
 136 }
 137
 138 static struct notifier_block qede_netdev_notifier = {
 139         .notifier_call = qede_netdev_event,
 140 };
 141
 142 static
 143 int __init qede_init(void)
 144 {
 145         int ret;
 146
 147         pr_notice("qede_init: %s\n", version);
 148
 149         qed_ops = qed_get_eth_ops();
 150         if (!qed_ops) {
 151                 pr_notice("Failed to get qed ethtool operations\n");
 152                 return -EINVAL;
 153         }
 154
 155         /* Must register notifier before pci ops, since we might miss
 156          * interface rename after pci probe and netdev registeration.
 157          */
 158         ret = register_netdevice_notifier(&qede_netdev_notifier);
 159         if (ret) {
 160                 pr_notice("Failed to register netdevice_notifier\n");
 161                 qed_put_eth_ops();
 162                 return -EINVAL;
 163         }
 164
 165         ret = pci_register_driver(&qede_pci_driver);
 166         if (ret) {
 167                 pr_notice("Failed to register driver\n");
 168                 unregister_netdevice_notifier(&qede_netdev_notifier);
 169                 qed_put_eth_ops();
 170                 return -EINVAL;
 171         }
 172
 173         return 0;
 174 }
 175
 176 static void __exit qede_cleanup(void)
 177 {
 178         pr_notice("qede_cleanup called\n");
 179
 180         unregister_netdevice_notifier(&qede_netdev_notifier);
 181         pci_unregister_driver(&qede_pci_driver);
 182         qed_put_eth_ops();
 183 }
 184
 185 module_init(qede_init);
 186 module_exit(qede_cleanup);
 187
 188 /* -------------------------------------------------------------------------
 189  * START OF FAST-PATH
 190  * -------------------------------------------------------------------------
 191  */
 192
 193 /* Unmap the data and free skb */
 194 static int qede_free_tx_pkt(struct qede_dev *edev,
 195                             struct qede_tx_queue *txq,
 196                             int *len)
 197 {
 198         u16 idx = txq->sw_tx_cons & NUM_TX_BDS_MAX;
 199         struct sk_buff *skb = txq->sw_tx_ring[idx].skb;
 200         struct eth_tx_1st_bd *first_bd;
 201         struct eth_tx_bd *tx_data_bd;
 202         int bds_consumed = 0;
 203         int nbds;
 204         bool data_split = txq->sw_tx_ring[idx].flags & QEDE_TSO_SPLIT_BD;
 205         int i, split_bd_len = 0;
 206
 207         if (unlikely(!skb)) {
 208                 DP_ERR(edev,
 209                        "skb is null for txq idx=%d txq->sw_tx_cons=%d txq->sw_tx_prod=%d\n",
 210                        idx, txq->sw_tx_cons, txq->sw_tx_prod);
 211                 return -1;
 212         }
 213
 214         *len = skb->len;
 215
 216         first_bd = (struct eth_tx_1st_bd *)qed_chain_consume(&txq->tx_pbl);
 217
 218         bds_consumed++;
 219
 220         nbds = first_bd->data.nbds;
 221
 222         if (data_split) {
 223                 struct eth_tx_bd *split = (struct eth_tx_bd *)
 224                         qed_chain_consume(&txq->tx_pbl);
 225                 split_bd_len = BD_UNMAP_LEN(split);
 226                 bds_consumed++;
 227         }
 228         dma_unmap_page(&edev->pdev->dev, BD_UNMAP_ADDR(first_bd),
 229                        BD_UNMAP_LEN(first_bd) + split_bd_len, DMA_TO_DEVICE);
 230
 231         /* Unmap the data of the skb frags */
 232         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++, bds_consumed++) {
 233                 tx_data_bd = (struct eth_tx_bd *)
 234                         qed_chain_consume(&txq->tx_pbl);
 235                 dma_unmap_page(&edev->pdev->dev, BD_UNMAP_ADDR(tx_data_bd),
 236                                BD_UNMAP_LEN(tx_data_bd), DMA_TO_DEVICE);
 237         }
 238
 239         while (bds_consumed++ < nbds)
 240                 qed_chain_consume(&txq->tx_pbl);
 241
 242         /* Free skb */
 243         dev_kfree_skb_any(skb);
 244         txq->sw_tx_ring[idx].skb = NULL;
 245         txq->sw_tx_ring[idx].flags = 0;
 246
 247         return 0;
 248 }
 249
 250 /* Unmap the data and free skb when mapping failed during start_xmit */
 251 static void qede_free_failed_tx_pkt(struct qede_dev *edev,
 252                                     struct qede_tx_queue *txq,
 253                                     struct eth_tx_1st_bd *first_bd,
 254                                     int nbd,
 255                                     bool data_split)
 256 {
 257         u16 idx = txq->sw_tx_prod & NUM_TX_BDS_MAX;
 258         struct sk_buff *skb = txq->sw_tx_ring[idx].skb;
 259         struct eth_tx_bd *tx_data_bd;
 260         int i, split_bd_len = 0;
 261
 262         /* Return prod to its position before this skb was handled */
 263         qed_chain_set_prod(&txq->tx_pbl,
 264                            le16_to_cpu(txq->tx_db.data.bd_prod),
 265                            first_bd);
 266
 267         first_bd = (struct eth_tx_1st_bd *)qed_chain_produce(&txq->tx_pbl);
 268
 269         if (data_split) {
 270                 struct eth_tx_bd *split = (struct eth_tx_bd *)
 271                                           qed_chain_produce(&txq->tx_pbl);
 272                 split_bd_len = BD_UNMAP_LEN(split);
 273                 nbd--;
 274         }
 275
 276         dma_unmap_page(&edev->pdev->dev, BD_UNMAP_ADDR(first_bd),
 277                        BD_UNMAP_LEN(first_bd) + split_bd_len, DMA_TO_DEVICE);
 278
 279         /* Unmap the data of the skb frags */
 280         for (i = 0; i < nbd; i++) {
 281                 tx_data_bd = (struct eth_tx_bd *)
 282                         qed_chain_produce(&txq->tx_pbl);
 283                 if (tx_data_bd->nbytes)
 284                         dma_unmap_page(&edev->pdev->dev,
 285                                        BD_UNMAP_ADDR(tx_data_bd),
 286                                        BD_UNMAP_LEN(tx_data_bd), DMA_TO_DEVICE);
 287         }
 288
 289         /* Return again prod to its position before this skb was handled */
 290         qed_chain_set_prod(&txq->tx_pbl,
 291                            le16_to_cpu(txq->tx_db.data.bd_prod),
 292                            first_bd);
 293
 294         /* Free skb */
 295         dev_kfree_skb_any(skb);
 296         txq->sw_tx_ring[idx].skb = NULL;
 297         txq->sw_tx_ring[idx].flags = 0;
 298 }
 299
 300 static u32 qede_xmit_type(struct qede_dev *edev,
 301                           struct sk_buff *skb,
 302                           int *ipv6_ext)
 303 {
 304         u32 rc = XMIT_L4_CSUM;
 305         __be16 l3_proto;
 306
 307         if (skb->ip_summed != CHECKSUM_PARTIAL)
 308                 return XMIT_PLAIN;
 309
 310         l3_proto = vlan_get_protocol(skb);
 311         if (l3_proto == htons(ETH_P_IPV6) &&
 312             (ipv6_hdr(skb)->nexthdr == NEXTHDR_IPV6))
 313                 *ipv6_ext = 1;
 314
 315         if (skb_is_gso(skb))
 316                 rc |= XMIT_LSO;
 317
 318         return rc;
 319 }
 320
 321 static void qede_set_params_for_ipv6_ext(struct sk_buff *skb,
 322                                          struct eth_tx_2nd_bd *second_bd,
 323                                          struct eth_tx_3rd_bd *third_bd)
 324 {
 325         u8 l4_proto;
 326         u16 bd2_bits1 = 0, bd2_bits2 = 0;
 327
 328         bd2_bits1 |= (1 << ETH_TX_DATA_2ND_BD_IPV6_EXT_SHIFT);
 329
 330         bd2_bits2 |= ((((u8 *)skb_transport_header(skb) - skb->data) >> 1) &
 331                      ETH_TX_DATA_2ND_BD_L4_HDR_START_OFFSET_W_MASK)
 332                     << ETH_TX_DATA_2ND_BD_L4_HDR_START_OFFSET_W_SHIFT;
 333
 334         bd2_bits1 |= (ETH_L4_PSEUDO_CSUM_CORRECT_LENGTH <<
 335                       ETH_TX_DATA_2ND_BD_L4_PSEUDO_CSUM_MODE_SHIFT);
 336
 337         if (vlan_get_protocol(skb) == htons(ETH_P_IPV6))
 338                 l4_proto = ipv6_hdr(skb)->nexthdr;
 339         else
 340                 l4_proto = ip_hdr(skb)->protocol;
 341
 342         if (l4_proto == IPPROTO_UDP)
 343                 bd2_bits1 |= 1 << ETH_TX_DATA_2ND_BD_L4_UDP_SHIFT;
 344
 345         if (third_bd)
 346                 third_bd->data.bitfields |=
 347                         cpu_to_le16(((tcp_hdrlen(skb) / 4) &
 348                                 ETH_TX_DATA_3RD_BD_TCP_HDR_LEN_DW_MASK) <<
 349                                 ETH_TX_DATA_3RD_BD_TCP_HDR_LEN_DW_SHIFT);
 350
 351         second_bd->data.bitfields1 = cpu_to_le16(bd2_bits1);
 352         second_bd->data.bitfields2 = cpu_to_le16(bd2_bits2);
 353 }
 354
 355 static int map_frag_to_bd(struct qede_dev *edev,
 356                           skb_frag_t *frag,
 357                           struct eth_tx_bd *bd)
 358 {
 359         dma_addr_t mapping;
 360
 361         /* Map skb non-linear frag data for DMA */
 362         mapping = skb_frag_dma_map(&edev->pdev->dev, frag, 0,
 363                                    skb_frag_size(frag),
 364                                    DMA_TO_DEVICE);
 365         if (unlikely(dma_mapping_error(&edev->pdev->dev, mapping))) {
 366                 DP_NOTICE(edev, "Unable to map frag - dropping packet\n");
 367                 return -ENOMEM;
 368         }
 369
 370         /* Setup the data pointer of the frag data */
 371         BD_SET_UNMAP_ADDR_LEN(bd, mapping, skb_frag_size(frag));
 372
 373         return 0;
 374 }
 375
 376 /* +2 for 1st BD for headers and 2nd BD for headlen (if required) */
 377 #if ((MAX_SKB_FRAGS + 2) > ETH_TX_MAX_BDS_PER_NON_LSO_PACKET)
 378 static bool qede_pkt_req_lin(struct qede_dev *edev, struct sk_buff *skb,
 379                              u8 xmit_type)
 380 {
 381         int allowed_frags = ETH_TX_MAX_BDS_PER_NON_LSO_PACKET - 1;
 382
 383         if (xmit_type & XMIT_LSO) {
 384                 int hlen;
 385
 386                 hlen = skb_transport_header(skb) +
 387                        tcp_hdrlen(skb) - skb->data;
 388
 389                 /* linear payload would require its own BD */
 390                 if (skb_headlen(skb) > hlen)
 391                         allowed_frags--;
 392         }
 393
 394         return (skb_shinfo(skb)->nr_frags > allowed_frags);
 395 }
 396 #endif
 397
 398 /* Main transmit function */
 399 static
 400 netdev_tx_t qede_start_xmit(struct sk_buff *skb,
 401                             struct net_device *ndev)
 402 {
 403         struct qede_dev *edev = netdev_priv(ndev);
 404         struct netdev_queue *netdev_txq;
 405         struct qede_tx_queue *txq;
 406         struct eth_tx_1st_bd *first_bd;
 407         struct eth_tx_2nd_bd *second_bd = NULL;
 408         struct eth_tx_3rd_bd *third_bd = NULL;
 409         struct eth_tx_bd *tx_data_bd = NULL;
 410         u16 txq_index;
 411         u8 nbd = 0;
 412         dma_addr_t mapping;
 413         int rc, frag_idx = 0, ipv6_ext = 0;
 414         u8 xmit_type;
 415         u16 idx;
 416         u16 hlen;
 417         bool data_split;
 418
 419         /* Get tx-queue context and netdev index */
 420         txq_index = skb_get_queue_mapping(skb);
 421         WARN_ON(txq_index >= QEDE_TSS_CNT(edev));
 422         txq = QEDE_TX_QUEUE(edev, txq_index);
 423         netdev_txq = netdev_get_tx_queue(ndev, txq_index);
 424
 425         WARN_ON(qed_chain_get_elem_left(&txq->tx_pbl) <
 426                                (MAX_SKB_FRAGS + 1));
 427
 428         xmit_type = qede_xmit_type(edev, skb, &ipv6_ext);
 429
 430 #if ((MAX_SKB_FRAGS + 2) > ETH_TX_MAX_BDS_PER_NON_LSO_PACKET)
 431         if (qede_pkt_req_lin(edev, skb, xmit_type)) {
 432                 if (skb_linearize(skb)) {
 433                         DP_NOTICE(edev,
 434                                   "SKB linearization failed - silently dropping this SKB\n");
 435                         dev_kfree_skb_any(skb);
 436                         return NETDEV_TX_OK;
 437                 }
 438         }
 439 #endif
 440
 441         /* Fill the entry in the SW ring and the BDs in the FW ring */
 442         idx = txq->sw_tx_prod & NUM_TX_BDS_MAX;
 443         txq->sw_tx_ring[idx].skb = skb;
 444         first_bd = (struct eth_tx_1st_bd *)
 445                    qed_chain_produce(&txq->tx_pbl);
 446         memset(first_bd, 0, sizeof(*first_bd));
 447         first_bd->data.bd_flags.bitfields =
 448                 1 << ETH_TX_1ST_BD_FLAGS_START_BD_SHIFT;
 449
 450         /* Map skb linear data for DMA and set in the first BD */
 451         mapping = dma_map_single(&edev->pdev->dev, skb->data,
 452                                  skb_headlen(skb), DMA_TO_DEVICE);
 453         if (unlikely(dma_mapping_error(&edev->pdev->dev, mapping))) {
 454                 DP_NOTICE(edev, "SKB mapping failed\n");
 455                 qede_free_failed_tx_pkt(edev, txq, first_bd, 0, false);
 456                 return NETDEV_TX_OK;
 457         }
 458         nbd++;
 459         BD_SET_UNMAP_ADDR_LEN(first_bd, mapping, skb_headlen(skb));
 460
 461         /* In case there is IPv6 with extension headers or LSO we need 2nd and
 462          * 3rd BDs.
 463          */
 464         if (unlikely((xmit_type & XMIT_LSO) | ipv6_ext)) {
 465                 second_bd = (struct eth_tx_2nd_bd *)
 466                         qed_chain_produce(&txq->tx_pbl);
 467                 memset(second_bd, 0, sizeof(*second_bd));
 468
 469                 nbd++;
 470                 third_bd = (struct eth_tx_3rd_bd *)
 471                         qed_chain_produce(&txq->tx_pbl);
 472                 memset(third_bd, 0, sizeof(*third_bd));
 473
 474                 nbd++;
 475                 /* We need to fill in additional data in second_bd... */
 476                 tx_data_bd = (struct eth_tx_bd *)second_bd;
 477         }
 478
 479         if (skb_vlan_tag_present(skb)) {
 480                 first_bd->data.vlan = cpu_to_le16(skb_vlan_tag_get(skb));
 481                 first_bd->data.bd_flags.bitfields |=
 482                         1 << ETH_TX_1ST_BD_FLAGS_VLAN_INSERTION_SHIFT;
 483         }
 484
 485         /* Fill the parsing flags & params according to the requested offload */
 486         if (xmit_type & XMIT_L4_CSUM) {
 487                 u16 temp = 1 << ETH_TX_DATA_1ST_BD_TUNN_CFG_OVERRIDE_SHIFT;
 488
 489                 /* We don't re-calculate IP checksum as it is already done by
 490                  * the upper stack
 491                  */
 492                 first_bd->data.bd_flags.bitfields |=
 493                         1 << ETH_TX_1ST_BD_FLAGS_L4_CSUM_SHIFT;
 494
 495                 first_bd->data.bitfields |= cpu_to_le16(temp);
 496
 497                 /* If the packet is IPv6 with extension header, indicate that
 498                  * to FW and pass few params, since the device cracker doesn't
 499                  * support parsing IPv6 with extension header/s.
 500                  */
 501                 if (unlikely(ipv6_ext))
 502                         qede_set_params_for_ipv6_ext(skb, second_bd, third_bd);
 503         }
 504
 505         if (xmit_type & XMIT_LSO) {
 506                 first_bd->data.bd_flags.bitfields |=
 507                         (1 << ETH_TX_1ST_BD_FLAGS_LSO_SHIFT);
 508                 third_bd->data.lso_mss =
 509                         cpu_to_le16(skb_shinfo(skb)->gso_size);
 510
 511                 first_bd->data.bd_flags.bitfields |=
 512                 1 << ETH_TX_1ST_BD_FLAGS_IP_CSUM_SHIFT;
 513                 hlen = skb_transport_header(skb) +
 514                        tcp_hdrlen(skb) - skb->data;
 515
 516                 /* @@@TBD - if will not be removed need to check */
 517                 third_bd->data.bitfields |=
 518                         cpu_to_le16((1 << ETH_TX_DATA_3RD_BD_HDR_NBD_SHIFT));
 519
 520                 /* Make life easier for FW guys who can't deal with header and
 521                  * data on same BD. If we need to split, use the second bd...
 522                  */
 523                 if (unlikely(skb_headlen(skb) > hlen)) {
 524                         DP_VERBOSE(edev, NETIF_MSG_TX_QUEUED,
 525                                    "TSO split header size is %d (%x:%x)\n",
 526                                    first_bd->nbytes, first_bd->addr.hi,
 527                                    first_bd->addr.lo);
 528
 529                         mapping = HILO_U64(le32_to_cpu(first_bd->addr.hi),
 530                                            le32_to_cpu(first_bd->addr.lo)) +
 531                                            hlen;
 532
 533                         BD_SET_UNMAP_ADDR_LEN(tx_data_bd, mapping,
 534                                               le16_to_cpu(first_bd->nbytes) -
 535                                               hlen);
 536
 537                         /* this marks the BD as one that has no
 538                          * individual mapping
 539                          */
 540                         txq->sw_tx_ring[idx].flags |= QEDE_TSO_SPLIT_BD;
 541
 542                         first_bd->nbytes = cpu_to_le16(hlen);
 543
 544                         tx_data_bd = (struct eth_tx_bd *)third_bd;
 545                         data_split = true;
 546                 }
 547         }
 548
 549         /* Handle fragmented skb */
 550         /* special handle for frags inside 2nd and 3rd bds.. */
 551         while (tx_data_bd && frag_idx < skb_shinfo(skb)->nr_frags) {
 552                 rc = map_frag_to_bd(edev,
 553                                     &skb_shinfo(skb)->frags[frag_idx],
 554                                     tx_data_bd);
 555                 if (rc) {
 556                         qede_free_failed_tx_pkt(edev, txq, first_bd, nbd,
 557                                                 data_split);
 558                         return NETDEV_TX_OK;
 559                 }
 560
 561                 if (tx_data_bd == (struct eth_tx_bd *)second_bd)
 562                         tx_data_bd = (struct eth_tx_bd *)third_bd;
 563                 else
 564                         tx_data_bd = NULL;
 565
 566                 frag_idx++;
 567         }
 568
 569         /* map last frags into 4th, 5th .... */
 570         for (; frag_idx < skb_shinfo(skb)->nr_frags; frag_idx++, nbd++) {
 571                 tx_data_bd = (struct eth_tx_bd *)
 572                              qed_chain_produce(&txq->tx_pbl);
 573
 574                 memset(tx_data_bd, 0, sizeof(*tx_data_bd));
 575
 576                 rc = map_frag_to_bd(edev,
 577                                     &skb_shinfo(skb)->frags[frag_idx],
 578                                     tx_data_bd);
 579                 if (rc) {
 580                         qede_free_failed_tx_pkt(edev, txq, first_bd, nbd,
 581                                                 data_split);
 582                         return NETDEV_TX_OK;
 583                 }
 584         }
 585
 586         /* update the first BD with the actual num BDs */
 587         first_bd->data.nbds = nbd;
 588
 589         netdev_tx_sent_queue(netdev_txq, skb->len);
 590
 591         skb_tx_timestamp(skb);
 592
 593         /* Advance packet producer only before sending the packet since mapping
 594          * of pages may fail.
 595          */
 596         txq->sw_tx_prod++;
 597
 598         /* 'next page' entries are counted in the producer value */
 599         txq->tx_db.data.bd_prod =
 600                 cpu_to_le16(qed_chain_get_prod_idx(&txq->tx_pbl));
 601
 602         /* wmb makes sure that the BDs data is updated before updating the
 603          * producer, otherwise FW may read old data from the BDs.
 604          */
 605         wmb();
 606         barrier();
 607         writel(txq->tx_db.raw, txq->doorbell_addr);
 608
 609         /* mmiowb is needed to synchronize doorbell writes from more than one
 610          * processor. It guarantees that the write arrives to the device before
 611          * the queue lock is released and another start_xmit is called (possibly
 612          * on another CPU). Without this barrier, the next doorbell can bypass
 613          * this doorbell. This is applicable to IA64/Altix systems.
 614          */
 615         mmiowb();
 616
 617         if (unlikely(qed_chain_get_elem_left(&txq->tx_pbl)
 618                       < (MAX_SKB_FRAGS + 1))) {
 619                 netif_tx_stop_queue(netdev_txq);
 620                 DP_VERBOSE(edev, NETIF_MSG_TX_QUEUED,
 621                            "Stop queue was called\n");
 622                 /* paired memory barrier is in qede_tx_int(), we have to keep
 623                  * ordering of set_bit() in netif_tx_stop_queue() and read of
 624                  * fp->bd_tx_cons
 625                  */
 626                 smp_mb();
 627
 628                 if (qed_chain_get_elem_left(&txq->tx_pbl)
 629                      >= (MAX_SKB_FRAGS + 1) &&
 630                     (edev->state == QEDE_STATE_OPEN)) {
 631                         netif_tx_wake_queue(netdev_txq);
 632                         DP_VERBOSE(edev, NETIF_MSG_TX_QUEUED,
 633                                    "Wake queue was called\n");
 634                 }
 635         }
 636
 637         return NETDEV_TX_OK;
 638 }
 639
 640 static int qede_txq_has_work(struct qede_tx_queue *txq)
 641 {
 642         u16 hw_bd_cons;
 643
 644         /* Tell compiler that consumer and producer can change */
 645         barrier();
 646         hw_bd_cons = le16_to_cpu(*txq->hw_cons_ptr);
 647         if (qed_chain_get_cons_idx(&txq->tx_pbl) == hw_bd_cons + 1)
 648                 return 0;
 649
 650         return hw_bd_cons != qed_chain_get_cons_idx(&txq->tx_pbl);
 651 }
 652
 653 static int qede_tx_int(struct qede_dev *edev,
 654                        struct qede_tx_queue *txq)
 655 {
 656         struct netdev_queue *netdev_txq;
 657         u16 hw_bd_cons;
 658         unsigned int pkts_compl = 0, bytes_compl = 0;
 659         int rc;
 660
 661         netdev_txq = netdev_get_tx_queue(edev->ndev, txq->index);
 662
 663         hw_bd_cons = le16_to_cpu(*txq->hw_cons_ptr);
 664         barrier();
 665
 666         while (hw_bd_cons != qed_chain_get_cons_idx(&txq->tx_pbl)) {
 667                 int len = 0;
 668
 669                 rc = qede_free_tx_pkt(edev, txq, &len);
 670                 if (rc) {
 671                         DP_NOTICE(edev, "hw_bd_cons = %d, chain_cons=%d\n",
 672                                   hw_bd_cons,
 673                                   qed_chain_get_cons_idx(&txq->tx_pbl));
 674                         break;
 675                 }
 676
 677                 bytes_compl += len;
 678                 pkts_compl++;
 679                 txq->sw_tx_cons++;
 680         }
 681
 682         netdev_tx_completed_queue(netdev_txq, pkts_compl, bytes_compl);
 683
 684         /* Need to make the tx_bd_cons update visible to start_xmit()
 685          * before checking for netif_tx_queue_stopped().  Without the
 686          * memory barrier, there is a small possibility that
 687          * start_xmit() will miss it and cause the queue to be stopped
 688          * forever.
 689          * On the other hand we need an rmb() here to ensure the proper
 690          * ordering of bit testing in the following
 691          * netif_tx_queue_stopped(txq) call.
 692          */
 693         smp_mb();
 694
 695         if (unlikely(netif_tx_queue_stopped(netdev_txq))) {
 696                 /* Taking tx_lock is needed to prevent reenabling the queue
 697                  * while it's empty. This could have happen if rx_action() gets
 698                  * suspended in qede_tx_int() after the condition before
 699                  * netif_tx_wake_queue(), while tx_action (qede_start_xmit()):
 700                  *
 701                  * stops the queue->sees fresh tx_bd_cons->releases the queue->
 702                  * sends some packets consuming the whole queue again->
 703                  * stops the queue
 704                  */
 705
 706                 __netif_tx_lock(netdev_txq, smp_processor_id());
 707
 708                 if ((netif_tx_queue_stopped(netdev_txq)) &&
 709                     (edev->state == QEDE_STATE_OPEN) &&
 710                     (qed_chain_get_elem_left(&txq->tx_pbl)
 711                       >= (MAX_SKB_FRAGS + 1))) {
 712                         netif_tx_wake_queue(netdev_txq);
 713                         DP_VERBOSE(edev, NETIF_MSG_TX_DONE,
 714                                    "Wake queue was called\n");
 715                 }
 716
 717                 __netif_tx_unlock(netdev_txq);
 718         }
 719
 720         return 0;
 721 }
 722
 723 static bool qede_has_rx_work(struct qede_rx_queue *rxq)
 724 {
 725         u16 hw_comp_cons, sw_comp_cons;
 726
 727         /* Tell compiler that status block fields can change */
 728         barrier();
 729
 730         hw_comp_cons = le16_to_cpu(*rxq->hw_cons_ptr);
 731         sw_comp_cons = qed_chain_get_cons_idx(&rxq->rx_comp_ring);
 732
 733         return hw_comp_cons != sw_comp_cons;
 734 }
 735
 736 static bool qede_has_tx_work(struct qede_fastpath *fp)
 737 {
 738         u8 tc;
 739
 740         for (tc = 0; tc < fp->edev->num_tc; tc++)
 741                 if (qede_txq_has_work(&fp->txqs[tc]))
 742                         return true;
 743         return false;
 744 }
 745
 746 /* This function reuses the buffer(from an offset) from
 747  * consumer index to producer index in the bd ring
 748  */
 749 static inline void qede_reuse_page(struct qede_dev *edev,
 750                                    struct qede_rx_queue *rxq,
 751                                    struct sw_rx_data *curr_cons)
 752 {
 753         struct eth_rx_bd *rx_bd_prod = qed_chain_produce(&rxq->rx_bd_ring);
 754         struct sw_rx_data *curr_prod;
 755         dma_addr_t new_mapping;
 756
 757         curr_prod = &rxq->sw_rx_ring[rxq->sw_rx_prod & NUM_RX_BDS_MAX];
 758         *curr_prod = *curr_cons;
 759
 760         new_mapping = curr_prod->mapping + curr_prod->page_offset;
 761
 762         rx_bd_prod->addr.hi = cpu_to_le32(upper_32_bits(new_mapping));
 763         rx_bd_prod->addr.lo = cpu_to_le32(lower_32_bits(new_mapping));
 764
 765         rxq->sw_rx_prod++;
 766         curr_cons->data = NULL;
 767 }
 768
 769 static inline int qede_realloc_rx_buffer(struct qede_dev *edev,
 770                                          struct qede_rx_queue *rxq,
 771                                          struct sw_rx_data *curr_cons)
 772 {
 773         /* Move to the next segment in the page */
 774         curr_cons->page_offset += rxq->rx_buf_seg_size;
 775
 776         if (curr_cons->page_offset == PAGE_SIZE) {
 777                 if (unlikely(qede_alloc_rx_buffer(edev, rxq)))
 778                         return -ENOMEM;
 779
 780                 dma_unmap_page(&edev->pdev->dev, curr_cons->mapping,
 781                                PAGE_SIZE, DMA_FROM_DEVICE);
 782         } else {
 783                 /* Increment refcount of the page as we don't want
 784                  * network stack to take the ownership of the page
 785                  * which can be recycled multiple times by the driver.
 786                  */
 787                 atomic_inc(&curr_cons->data->_count);
 788                 qede_reuse_page(edev, rxq, curr_cons);
 789         }
 790
 791         return 0;
 792 }
 793
 794 static inline void qede_update_rx_prod(struct qede_dev *edev,
 795                                        struct qede_rx_queue *rxq)
 796 {
 797         u16 bd_prod = qed_chain_get_prod_idx(&rxq->rx_bd_ring);
 798         u16 cqe_prod = qed_chain_get_prod_idx(&rxq->rx_comp_ring);
 799         struct eth_rx_prod_data rx_prods = {0};
 800
 801         /* Update producers */
 802         rx_prods.bd_prod = cpu_to_le16(bd_prod);
 803         rx_prods.cqe_prod = cpu_to_le16(cqe_prod);
 804
 805         /* Make sure that the BD and SGE data is updated before updating the
 806          * producers since FW might read the BD/SGE right after the producer
 807          * is updated.
 808          */
 809         wmb();
 810
 811         internal_ram_wr(rxq->hw_rxq_prod_addr, sizeof(rx_prods),
 812                         (u32 *)&rx_prods);
 813
 814         /* mmiowb is needed to synchronize doorbell writes from more than one
 815          * processor. It guarantees that the write arrives to the device before
 816          * the napi lock is released and another qede_poll is called (possibly
 817          * on another CPU). Without this barrier, the next doorbell can bypass
 818          * this doorbell. This is applicable to IA64/Altix systems.
 819          */
 820         mmiowb();
 821 }
 822
 823 static u32 qede_get_rxhash(struct qede_dev *edev,
 824                            u8 bitfields,
 825                            __le32 rss_hash,
 826                            enum pkt_hash_types *rxhash_type)
 827 {
 828         enum rss_hash_type htype;
 829
 830         htype = GET_FIELD(bitfields, ETH_FAST_PATH_RX_REG_CQE_RSS_HASH_TYPE);
 831
 832         if ((edev->ndev->features & NETIF_F_RXHASH) && htype) {
 833                 *rxhash_type = ((htype == RSS_HASH_TYPE_IPV4) ||
 834                                 (htype == RSS_HASH_TYPE_IPV6)) ?
 835                                 PKT_HASH_TYPE_L3 : PKT_HASH_TYPE_L4;
 836                 return le32_to_cpu(rss_hash);
 837         }
 838         *rxhash_type = PKT_HASH_TYPE_NONE;
 839         return 0;
 840 }
 841
 842 static void qede_set_skb_csum(struct sk_buff *skb, u8 csum_flag)
 843 {
 844         skb_checksum_none_assert(skb);
 845
 846         if (csum_flag & QEDE_CSUM_UNNECESSARY)
 847                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 848 }
 849
 850 static inline void qede_skb_receive(struct qede_dev *edev,
 851                                     struct qede_fastpath *fp,
 852                                     struct sk_buff *skb,
 853                                     u16 vlan_tag)
 854 {
 855         if (vlan_tag)
 856                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
 857                                        vlan_tag);
 858
 859         napi_gro_receive(&fp->napi, skb);
 860 }
 861
 862 static void qede_set_gro_params(struct qede_dev *edev,
 863                                 struct sk_buff *skb,
 864                                 struct eth_fast_path_rx_tpa_start_cqe *cqe)
 865 {
 866         u16 parsing_flags = le16_to_cpu(cqe->pars_flags.flags);
 867
 868         if (((parsing_flags >> PARSING_AND_ERR_FLAGS_L3TYPE_SHIFT) &
 869             PARSING_AND_ERR_FLAGS_L3TYPE_MASK) == 2)
 870                 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
 871         else
 872                 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
 873
 874         skb_shinfo(skb)->gso_size = __le16_to_cpu(cqe->len_on_first_bd) -
 875                                         cqe->header_len;
 876 }
 877
 878 static int qede_fill_frag_skb(struct qede_dev *edev,
 879                               struct qede_rx_queue *rxq,
 880                               u8 tpa_agg_index,
 881                               u16 len_on_bd)
 882 {
 883         struct sw_rx_data *current_bd = &rxq->sw_rx_ring[rxq->sw_rx_cons &
 884                                                          NUM_RX_BDS_MAX];
 885         struct qede_agg_info *tpa_info = &rxq->tpa_info[tpa_agg_index];
 886         struct sk_buff *skb = tpa_info->skb;
 887
 888         if (unlikely(tpa_info->agg_state != QEDE_AGG_STATE_START))
 889                 goto out;
 890
 891         /* Add one frag and update the appropriate fields in the skb */
 892         skb_fill_page_desc(skb, tpa_info->frag_id++,
 893                            current_bd->data, current_bd->page_offset,
 894                            len_on_bd);
 895
 896         if (unlikely(qede_realloc_rx_buffer(edev, rxq, current_bd))) {
 897                 tpa_info->agg_state = QEDE_AGG_STATE_ERROR;
 898                 goto out;
 899         }
 900
 901         qed_chain_consume(&rxq->rx_bd_ring);
 902         rxq->sw_rx_cons++;
 903
 904         skb->data_len += len_on_bd;
 905         skb->truesize += rxq->rx_buf_seg_size;
 906         skb->len += len_on_bd;
 907
 908         return 0;
 909
 910 out:
 911         return -ENOMEM;
 912 }
 913
 914 static void qede_tpa_start(struct qede_dev *edev,
 915                            struct qede_rx_queue *rxq,
 916                            struct eth_fast_path_rx_tpa_start_cqe *cqe)
 917 {
 918         struct qede_agg_info *tpa_info = &rxq->tpa_info[cqe->tpa_agg_index];
 919         struct eth_rx_bd *rx_bd_cons = qed_chain_consume(&rxq->rx_bd_ring);
 920         struct eth_rx_bd *rx_bd_prod = qed_chain_produce(&rxq->rx_bd_ring);
 921         struct sw_rx_data *replace_buf = &tpa_info->replace_buf;
 922         dma_addr_t mapping = tpa_info->replace_buf_mapping;
 923         struct sw_rx_data *sw_rx_data_cons;
 924         struct sw_rx_data *sw_rx_data_prod;
 925         enum pkt_hash_types rxhash_type;
 926         u32 rxhash;
 927
 928         sw_rx_data_cons = &rxq->sw_rx_ring[rxq->sw_rx_cons & NUM_RX_BDS_MAX];
 929         sw_rx_data_prod = &rxq->sw_rx_ring[rxq->sw_rx_prod & NUM_RX_BDS_MAX];
 930
 931         /* Use pre-allocated replacement buffer - we can't release the agg.
 932          * start until its over and we don't want to risk allocation failing
 933          * here, so re-allocate when aggregation will be over.
 934          */
 935         dma_unmap_addr_set(sw_rx_data_prod, mapping,
 936                            dma_unmap_addr(replace_buf, mapping));
 937
 938         sw_rx_data_prod->data = replace_buf->data;
 939         rx_bd_prod->addr.hi = cpu_to_le32(upper_32_bits(mapping));
 940         rx_bd_prod->addr.lo = cpu_to_le32(lower_32_bits(mapping));
 941         sw_rx_data_prod->page_offset = replace_buf->page_offset;
 942
 943         rxq->sw_rx_prod++;
 944
 945         /* move partial skb from cons to pool (don't unmap yet)
 946          * save mapping, incase we drop the packet later on.
 947          */
 948         tpa_info->start_buf = *sw_rx_data_cons;
 949         mapping = HILO_U64(le32_to_cpu(rx_bd_cons->addr.hi),
 950                            le32_to_cpu(rx_bd_cons->addr.lo));
 951
 952         tpa_info->start_buf_mapping = mapping;
 953         rxq->sw_rx_cons++;
 954
 955         /* set tpa state to start only if we are able to allocate skb
 956          * for this aggregation, otherwise mark as error and aggregation will
 957          * be dropped
 958          */
 959         tpa_info->skb = netdev_alloc_skb(edev->ndev,
 960                                          le16_to_cpu(cqe->len_on_first_bd));
 961         if (unlikely(!tpa_info->skb)) {
 962                 tpa_info->agg_state = QEDE_AGG_STATE_ERROR;
 963                 return;
 964         }
 965
 966         skb_put(tpa_info->skb, le16_to_cpu(cqe->len_on_first_bd));
 967         memcpy(&tpa_info->start_cqe, cqe, sizeof(tpa_info->start_cqe));
 968
 969         /* Start filling in the aggregation info */
 970         tpa_info->frag_id = 0;
 971         tpa_info->agg_state = QEDE_AGG_STATE_START;
 972
 973         rxhash = qede_get_rxhash(edev, cqe->bitfields,
 974                                  cqe->rss_hash, &rxhash_type);
 975         skb_set_hash(tpa_info->skb, rxhash, rxhash_type);
 976         if ((le16_to_cpu(cqe->pars_flags.flags) >>
 977              PARSING_AND_ERR_FLAGS_TAG8021QEXIST_SHIFT) &
 978                     PARSING_AND_ERR_FLAGS_TAG8021QEXIST_MASK)
 979                 tpa_info->vlan_tag = le16_to_cpu(cqe->vlan_tag);
 980         else
 981                 tpa_info->vlan_tag = 0;
 982
 983         /* This is needed in order to enable forwarding support */
 984         qede_set_gro_params(edev, tpa_info->skb, cqe);
 985
 986         if (likely(cqe->ext_bd_len_list[0]))
 987                 qede_fill_frag_skb(edev, rxq, cqe->tpa_agg_index,
 988                                    le16_to_cpu(cqe->ext_bd_len_list[0]));
 989
 990         if (unlikely(cqe->ext_bd_len_list[1])) {
 991                 DP_ERR(edev,
 992                        "Unlikely - got a TPA aggregation with more than one ext_bd_len_list entry in the TPA start\n");
 993                 tpa_info->agg_state = QEDE_AGG_STATE_ERROR;
 994         }
 995 }
 996
 997 #ifdef CONFIG_INET
 998 static void qede_gro_ip_csum(struct sk_buff *skb)
 999 {
1000         const struct iphdr *iph = ip_hdr(skb);
1001         struct tcphdr *th;
1002
1003         skb_set_network_header(skb, 0);
1004         skb_set_transport_header(skb, sizeof(struct iphdr));
1005         th = tcp_hdr(skb);
1006
1007         th->check = ~tcp_v4_check(skb->len - skb_transport_offset(skb),
1008                                   iph->saddr, iph->daddr, 0);
1009
1010         tcp_gro_complete(skb);
1011 }
1012
1013 static void qede_gro_ipv6_csum(struct sk_buff *skb)
1014 {
1015         struct ipv6hdr *iph = ipv6_hdr(skb);
1016         struct tcphdr *th;
1017
1018         skb_set_network_header(skb, 0);
1019         skb_set_transport_header(skb, sizeof(struct ipv6hdr));
1020         th = tcp_hdr(skb);
1021
1022         th->check = ~tcp_v6_check(skb->len - skb_transport_offset(skb),
1023                                   &iph->saddr, &iph->daddr, 0);
1024         tcp_gro_complete(skb);
1025 }
1026 #endif
1027
1028 static void qede_gro_receive(struct qede_dev *edev,
1029                              struct qede_fastpath *fp,
1030                              struct sk_buff *skb,
1031                              u16 vlan_tag)
1032 {
1033 #ifdef CONFIG_INET
1034         if (skb_shinfo(skb)->gso_size) {
1035                 switch (skb->protocol) {
1036                 case htons(ETH_P_IP):
1037                         qede_gro_ip_csum(skb);
1038                         break;
1039                 case htons(ETH_P_IPV6):
1040                         qede_gro_ipv6_csum(skb);
1041                         break;
1042                 default:
1043                         DP_ERR(edev,
1044                                "Error: FW GRO supports only IPv4/IPv6, not 0x%04x\n",
1045                                ntohs(skb->protocol));
1046                 }
1047         }
1048 #endif
1049         skb_record_rx_queue(skb, fp->rss_id);
1050         qede_skb_receive(edev, fp, skb, vlan_tag);
1051 }
1052
1053 static inline void qede_tpa_cont(struct qede_dev *edev,
1054                                  struct qede_rx_queue *rxq,
1055                                  struct eth_fast_path_rx_tpa_cont_cqe *cqe)
1056 {
1057         int i;
1058
1059         for (i = 0; cqe->len_list[i]; i++)
1060                 qede_fill_frag_skb(edev, rxq, cqe->tpa_agg_index,
1061                                    le16_to_cpu(cqe->len_list[i]));
1062
1063         if (unlikely(i > 1))
1064                 DP_ERR(edev,
1065                        "Strange - TPA cont with more than a single len_list entry\n");
1066 }
1067
1068 static void qede_tpa_end(struct qede_dev *edev,
1069                          struct qede_fastpath *fp,
1070                          struct eth_fast_path_rx_tpa_end_cqe *cqe)
1071 {
1072         struct qede_rx_queue *rxq = fp->rxq;
1073         struct qede_agg_info *tpa_info;
1074         struct sk_buff *skb;
1075         int i;
1076
1077         tpa_info = &rxq->tpa_info[cqe->tpa_agg_index];
1078         skb = tpa_info->skb;
1079
1080         for (i = 0; cqe->len_list[i]; i++)
1081                 qede_fill_frag_skb(edev, rxq, cqe->tpa_agg_index,
1082                                    le16_to_cpu(cqe->len_list[i]));
1083         if (unlikely(i > 1))
1084                 DP_ERR(edev,
1085                        "Strange - TPA emd with more than a single len_list entry\n");
1086
1087         if (unlikely(tpa_info->agg_state != QEDE_AGG_STATE_START))
1088                 goto err;
1089
1090         /* Sanity */
1091         if (unlikely(cqe->num_of_bds != tpa_info->frag_id + 1))
1092                 DP_ERR(edev,
1093                        "Strange - TPA had %02x BDs, but SKB has only %d frags\n",
1094                        cqe->num_of_bds, tpa_info->frag_id);
1095         if (unlikely(skb->len != le16_to_cpu(cqe->total_packet_len)))
1096                 DP_ERR(edev,
1097                        "Strange - total packet len [cqe] is %4x but SKB has len %04x\n",
1098                        le16_to_cpu(cqe->total_packet_len), skb->len);
1099
1100         memcpy(skb->data,
1101                page_address(tpa_info->start_buf.data) +
1102                 tpa_info->start_cqe.placement_offset +
1103                 tpa_info->start_buf.page_offset,
1104                le16_to_cpu(tpa_info->start_cqe.len_on_first_bd));
1105
1106         /* Recycle [mapped] start buffer for the next replacement */
1107         tpa_info->replace_buf = tpa_info->start_buf;
1108         tpa_info->replace_buf_mapping = tpa_info->start_buf_mapping;
1109
1110         /* Finalize the SKB */
1111         skb->protocol = eth_type_trans(skb, edev->ndev);
1112         skb->ip_summed = CHECKSUM_UNNECESSARY;
1113
1114         /* tcp_gro_complete() will copy NAPI_GRO_CB(skb)->count
1115          * to skb_shinfo(skb)->gso_segs
1116          */
1117         NAPI_GRO_CB(skb)->count = le16_to_cpu(cqe->num_of_coalesced_segs);
1118
1119         qede_gro_receive(edev, fp, skb, tpa_info->vlan_tag);
1120
1121         tpa_info->agg_state = QEDE_AGG_STATE_NONE;
1122
1123         return;
1124 err:
1125         /* The BD starting the aggregation is still mapped; Re-use it for
1126          * future aggregations [as replacement buffer]
1127          */
1128         memcpy(&tpa_info->replace_buf, &tpa_info->start_buf,
1129                sizeof(struct sw_rx_data));
1130         tpa_info->replace_buf_mapping = tpa_info->start_buf_mapping;
1131         tpa_info->start_buf.data = NULL;
1132         tpa_info->agg_state = QEDE_AGG_STATE_NONE;
1133         dev_kfree_skb_any(tpa_info->skb);
1134         tpa_info->skb = NULL;
1135 }
1136
1137 static u8 qede_check_csum(u16 flag)
1138 {
1139         u16 csum_flag = 0;
1140         u8 csum = 0;
1141
1142         if ((PARSING_AND_ERR_FLAGS_L4CHKSMWASCALCULATED_MASK <<
1143              PARSING_AND_ERR_FLAGS_L4CHKSMWASCALCULATED_SHIFT) & flag) {
1144                 csum_flag |= PARSING_AND_ERR_FLAGS_L4CHKSMERROR_MASK <<
1145                              PARSING_AND_ERR_FLAGS_L4CHKSMERROR_SHIFT;
1146                 csum = QEDE_CSUM_UNNECESSARY;
1147         }
1148
1149         csum_flag |= PARSING_AND_ERR_FLAGS_IPHDRERROR_MASK <<
1150                      PARSING_AND_ERR_FLAGS_IPHDRERROR_SHIFT;
1151
1152         if (csum_flag & flag)
1153                 return QEDE_CSUM_ERROR;
1154
1155         return csum;
1156 }
1157
1158 static int qede_rx_int(struct qede_fastpath *fp, int budget)
1159 {
1160         struct qede_dev *edev = fp->edev;
1161         struct qede_rx_queue *rxq = fp->rxq;
1162
1163         u16 hw_comp_cons, sw_comp_cons, sw_rx_index, parse_flag;
1164         int rx_pkt = 0;
1165         u8 csum_flag;
1166
1167         hw_comp_cons = le16_to_cpu(*rxq->hw_cons_ptr);
1168         sw_comp_cons = qed_chain_get_cons_idx(&rxq->rx_comp_ring);
1169
1170         /* Memory barrier to prevent the CPU from doing speculative reads of CQE
1171          * / BD in the while-loop before reading hw_comp_cons. If the CQE is
1172          * read before it is written by FW, then FW writes CQE and SB, and then
1173          * the CPU reads the hw_comp_cons, it will use an old CQE.
1174          */
1175         rmb();
1176
1177         /* Loop to complete all indicated BDs */
1178         while (sw_comp_cons != hw_comp_cons) {
1179                 struct eth_fast_path_rx_reg_cqe *fp_cqe;
1180                 enum pkt_hash_types rxhash_type;
1181                 enum eth_rx_cqe_type cqe_type;
1182                 struct sw_rx_data *sw_rx_data;
1183                 union eth_rx_cqe *cqe;
1184                 struct sk_buff *skb;
1185                 struct page *data;
1186                 __le16 flags;
1187                 u16 len, pad;
1188                 u32 rx_hash;
1189
1190                 /* Get the CQE from the completion ring */
1191                 cqe = (union eth_rx_cqe *)
1192                         qed_chain_consume(&rxq->rx_comp_ring);
1193                 cqe_type = cqe->fast_path_regular.type;
1194
1195                 if (unlikely(cqe_type == ETH_RX_CQE_TYPE_SLOW_PATH)) {
1196                         edev->ops->eth_cqe_completion(
1197                                         edev->cdev, fp->rss_id,
1198                                         (struct eth_slow_path_rx_cqe *)cqe);
1199                         goto next_cqe;
1200                 }
1201
1202                 if (cqe_type != ETH_RX_CQE_TYPE_REGULAR) {
1203                         switch (cqe_type) {
1204                         case ETH_RX_CQE_TYPE_TPA_START:
1205                                 qede_tpa_start(edev, rxq,
1206                                                &cqe->fast_path_tpa_start);
1207                                 goto next_cqe;
1208                         case ETH_RX_CQE_TYPE_TPA_CONT:
1209                                 qede_tpa_cont(edev, rxq,
1210                                               &cqe->fast_path_tpa_cont);
1211                                 goto next_cqe;
1212                         case ETH_RX_CQE_TYPE_TPA_END:
1213                                 qede_tpa_end(edev, fp,
1214                                              &cqe->fast_path_tpa_end);
1215                                 goto next_rx_only;
1216                         default:
1217                                 break;
1218                         }
1219                 }
1220
1221                 /* Get the data from the SW ring */
1222                 sw_rx_index = rxq->sw_rx_cons & NUM_RX_BDS_MAX;
1223                 sw_rx_data = &rxq->sw_rx_ring[sw_rx_index];
1224                 data = sw_rx_data->data;
1225
1226                 fp_cqe = &cqe->fast_path_regular;
1227                 len =  le16_to_cpu(fp_cqe->len_on_first_bd);
1228                 pad = fp_cqe->placement_offset;
1229                 flags = cqe->fast_path_regular.pars_flags.flags;
1230
1231                 /* If this is an error packet then drop it */
1232                 parse_flag = le16_to_cpu(flags);
1233
1234                 csum_flag = qede_check_csum(parse_flag);
1235                 if (unlikely(csum_flag == QEDE_CSUM_ERROR)) {
1236                         DP_NOTICE(edev,
1237                                   "CQE in CONS = %u has error, flags = %x, dropping incoming packet\n",
1238                                   sw_comp_cons, parse_flag);
1239                         rxq->rx_hw_errors++;
1240                         qede_reuse_page(edev, rxq, sw_rx_data);
1241                         goto next_rx;
1242                 }
1243
1244                 skb = netdev_alloc_skb(edev->ndev, QEDE_RX_HDR_SIZE);
1245                 if (unlikely(!skb)) {
1246                         DP_NOTICE(edev,
1247                                   "Build_skb failed, dropping incoming packet\n");
1248                         qede_reuse_page(edev, rxq, sw_rx_data);
1249                         rxq->rx_alloc_errors++;
1250                         goto next_rx;
1251                 }
1252
1253                 /* Copy data into SKB */
1254                 if (len + pad <= QEDE_RX_HDR_SIZE) {
1255                         memcpy(skb_put(skb, len),
1256                                page_address(data) + pad +
1257                                 sw_rx_data->page_offset, len);
1258                         qede_reuse_page(edev, rxq, sw_rx_data);
1259                 } else {
1260                         struct skb_frag_struct *frag;
1261                         unsigned int pull_len;
1262                         unsigned char *va;
1263
1264                         frag = &skb_shinfo(skb)->frags[0];
1265
1266                         skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, data,
1267                                         pad + sw_rx_data->page_offset,
1268                                         len, rxq->rx_buf_seg_size);
1269
1270                         va = skb_frag_address(frag);
1271                         pull_len = eth_get_headlen(va, QEDE_RX_HDR_SIZE);
1272
1273                         /* Align the pull_len to optimize memcpy */
1274                         memcpy(skb->data, va, ALIGN(pull_len, sizeof(long)));
1275
1276                         skb_frag_size_sub(frag, pull_len);
1277                         frag->page_offset += pull_len;
1278                         skb->data_len -= pull_len;
1279                         skb->tail += pull_len;
1280
1281                         if (unlikely(qede_realloc_rx_buffer(edev, rxq,
1282                                                             sw_rx_data))) {
1283                                 DP_ERR(edev, "Failed to allocate rx buffer\n");
1284                                 rxq->rx_alloc_errors++;
1285                                 goto next_cqe;
1286                         }
1287                 }
1288
1289                 if (fp_cqe->bd_num != 1) {
1290                         u16 pkt_len = le16_to_cpu(fp_cqe->pkt_len);
1291                         u8 num_frags;
1292
1293                         pkt_len -= len;
1294
1295                         for (num_frags = fp_cqe->bd_num - 1; num_frags > 0;
1296                              num_frags--) {
1297                                 u16 cur_size = pkt_len > rxq->rx_buf_size ?
1298                                                 rxq->rx_buf_size : pkt_len;
1299
1300                                 WARN_ONCE(!cur_size,
1301                                           "Still got %d BDs for mapping jumbo, but length became 0\n",
1302                                           num_frags);
1303
1304                                 if (unlikely(qede_alloc_rx_buffer(edev, rxq)))
1305                                         goto next_cqe;
1306
1307                                 rxq->sw_rx_cons++;
1308                                 sw_rx_index = rxq->sw_rx_cons & NUM_RX_BDS_MAX;
1309                                 sw_rx_data = &rxq->sw_rx_ring[sw_rx_index];
1310                                 qed_chain_consume(&rxq->rx_bd_ring);
1311                                 dma_unmap_page(&edev->pdev->dev,
1312                                                sw_rx_data->mapping,
1313                                                PAGE_SIZE, DMA_FROM_DEVICE);
1314
1315                                 skb_fill_page_desc(skb,
1316                                                    skb_shinfo(skb)->nr_frags++,
1317                                                    sw_rx_data->data, 0,
1318                                                    cur_size);
1319
1320                                 skb->truesize += PAGE_SIZE;
1321                                 skb->data_len += cur_size;
1322                                 skb->len += cur_size;
1323                                 pkt_len -= cur_size;
1324                         }
1325
1326                         if (pkt_len)
1327                                 DP_ERR(edev,
1328                                        "Mapped all BDs of jumbo, but still have %d bytes\n",
1329                                        pkt_len);
1330                 }
1331
1332                 skb->protocol = eth_type_trans(skb, edev->ndev);
1333
1334                 rx_hash = qede_get_rxhash(edev, fp_cqe->bitfields,
1335                                           fp_cqe->rss_hash,
1336                                           &rxhash_type);
1337
1338                 skb_set_hash(skb, rx_hash, rxhash_type);
1339
1340                 qede_set_skb_csum(skb, csum_flag);
1341
1342                 skb_record_rx_queue(skb, fp->rss_id);
1343
1344                 qede_skb_receive(edev, fp, skb, le16_to_cpu(fp_cqe->vlan_tag));
1345
1346                 qed_chain_consume(&rxq->rx_bd_ring);
1347 next_rx:
1348                 rxq->sw_rx_cons++;
1349 next_rx_only:
1350                 rx_pkt++;
1351
1352 next_cqe: /* don't consume bd rx buffer */
1353                 qed_chain_recycle_consumed(&rxq->rx_comp_ring);
1354                 sw_comp_cons = qed_chain_get_cons_idx(&rxq->rx_comp_ring);
1355                 /* CR TPA - revisit how to handle budget in TPA perhaps
1356                  * increase on "end"
1357                  */
1358                 if (rx_pkt == budget)
1359                         break;
1360         } /* repeat while sw_comp_cons != hw_comp_cons... */
1361
1362         /* Update producers */
1363         qede_update_rx_prod(edev, rxq);
1364
1365         return rx_pkt;
1366 }
1367
1368 static int qede_poll(struct napi_struct *napi, int budget)
1369 {
1370         int work_done = 0;
1371         struct qede_fastpath *fp = container_of(napi, struct qede_fastpath,
1372                                                  napi);
1373         struct qede_dev *edev = fp->edev;
1374
1375         while (1) {
1376                 u8 tc;
1377
1378                 for (tc = 0; tc < edev->num_tc; tc++)
1379                         if (qede_txq_has_work(&fp->txqs[tc]))
1380                                 qede_tx_int(edev, &fp->txqs[tc]);
1381
1382                 if (qede_has_rx_work(fp->rxq)) {
1383                         work_done += qede_rx_int(fp, budget - work_done);
1384
1385                         /* must not complete if we consumed full budget */
1386                         if (work_done >= budget)
1387                                 break;
1388                 }
1389
1390                 /* Fall out from the NAPI loop if needed */
1391                 if (!(qede_has_rx_work(fp->rxq) || qede_has_tx_work(fp))) {
1392                         qed_sb_update_sb_idx(fp->sb_info);
1393                         /* *_has_*_work() reads the status block,
1394                          * thus we need to ensure that status block indices
1395                          * have been actually read (qed_sb_update_sb_idx)
1396                          * prior to this check (*_has_*_work) so that
1397                          * we won't write the "newer" value of the status block
1398                          * to HW (if there was a DMA right after
1399                          * qede_has_rx_work and if there is no rmb, the memory
1400                          * reading (qed_sb_update_sb_idx) may be postponed
1401                          * to right before *_ack_sb). In this case there
1402                          * will never be another interrupt until there is
1403                          * another update of the status block, while there
1404                          * is still unhandled work.
1405                          */
1406                         rmb();
1407
1408                         if (!(qede_has_rx_work(fp->rxq) ||
1409                               qede_has_tx_work(fp))) {
1410                                 napi_complete(napi);
1411                                 /* Update and reenable interrupts */
1412                                 qed_sb_ack(fp->sb_info, IGU_INT_ENABLE,
1413                                            1 /*update*/);
1414                                 break;
1415                         }
1416                 }
1417         }
1418
1419         return work_done;
1420 }
1421
1422 static irqreturn_t qede_msix_fp_int(int irq, void *fp_cookie)
1423 {
1424         struct qede_fastpath *fp = fp_cookie;
1425
1426         qed_sb_ack(fp->sb_info, IGU_INT_DISABLE, 0 /*do not update*/);
1427
1428         napi_schedule_irqoff(&fp->napi);
1429         return IRQ_HANDLED;
1430 }
1431
1432 /* -------------------------------------------------------------------------
1433  * END OF FAST-PATH
1434  * -------------------------------------------------------------------------
1435  */
1436
1437 static int qede_open(struct net_device *ndev);
1438 static int qede_close(struct net_device *ndev);
1439 static int qede_set_mac_addr(struct net_device *ndev, void *p);
1440 static void qede_set_rx_mode(struct net_device *ndev);
1441 static void qede_config_rx_mode(struct net_device *ndev);
1442
1443 static int qede_set_ucast_rx_mac(struct qede_dev *edev,
1444                                  enum qed_filter_xcast_params_type opcode,
1445                                  unsigned char mac[ETH_ALEN])
1446 {
1447         struct qed_filter_params filter_cmd;
1448
1449         memset(&filter_cmd, 0, sizeof(filter_cmd));
1450         filter_cmd.type = QED_FILTER_TYPE_UCAST;
1451         filter_cmd.filter.ucast.type = opcode;
1452         filter_cmd.filter.ucast.mac_valid = 1;
1453         ether_addr_copy(filter_cmd.filter.ucast.mac, mac);
1454
1455         return edev->ops->filter_config(edev->cdev, &filter_cmd);
1456 }
1457
1458 static int qede_set_ucast_rx_vlan(struct qede_dev *edev,
1459                                   enum qed_filter_xcast_params_type opcode,
1460                                   u16 vid)
1461 {
1462         struct qed_filter_params filter_cmd;
1463
1464         memset(&filter_cmd, 0, sizeof(filter_cmd));
1465         filter_cmd.type = QED_FILTER_TYPE_UCAST;
1466         filter_cmd.filter.ucast.type = opcode;
1467         filter_cmd.filter.ucast.vlan_valid = 1;
1468         filter_cmd.filter.ucast.vlan = vid;
1469
1470         return edev->ops->filter_config(edev->cdev, &filter_cmd);
1471 }
1472
1473 void qede_fill_by_demand_stats(struct qede_dev *edev)
1474 {
1475         struct qed_eth_stats stats;
1476
1477         edev->ops->get_vport_stats(edev->cdev, &stats);
1478         edev->stats.no_buff_discards = stats.no_buff_discards;
1479         edev->stats.rx_ucast_bytes = stats.rx_ucast_bytes;
1480         edev->stats.rx_mcast_bytes = stats.rx_mcast_bytes;
1481         edev->stats.rx_bcast_bytes = stats.rx_bcast_bytes;
1482         edev->stats.rx_ucast_pkts = stats.rx_ucast_pkts;
1483         edev->stats.rx_mcast_pkts = stats.rx_mcast_pkts;
1484         edev->stats.rx_bcast_pkts = stats.rx_bcast_pkts;
1485         edev->stats.mftag_filter_discards = stats.mftag_filter_discards;
1486         edev->stats.mac_filter_discards = stats.mac_filter_discards;
1487
1488         edev->stats.tx_ucast_bytes = stats.tx_ucast_bytes;
1489         edev->stats.tx_mcast_bytes = stats.tx_mcast_bytes;
1490         edev->stats.tx_bcast_bytes = stats.tx_bcast_bytes;
1491         edev->stats.tx_ucast_pkts = stats.tx_ucast_pkts;
1492         edev->stats.tx_mcast_pkts = stats.tx_mcast_pkts;
1493         edev->stats.tx_bcast_pkts = stats.tx_bcast_pkts;
1494         edev->stats.tx_err_drop_pkts = stats.tx_err_drop_pkts;
1495         edev->stats.coalesced_pkts = stats.tpa_coalesced_pkts;
1496         edev->stats.coalesced_events = stats.tpa_coalesced_events;
1497         edev->stats.coalesced_aborts_num = stats.tpa_aborts_num;
1498         edev->stats.non_coalesced_pkts = stats.tpa_not_coalesced_pkts;
1499         edev->stats.coalesced_bytes = stats.tpa_coalesced_bytes;
1500
1501         edev->stats.rx_64_byte_packets = stats.rx_64_byte_packets;
1502         edev->stats.rx_127_byte_packets = stats.rx_127_byte_packets;
1503         edev->stats.rx_255_byte_packets = stats.rx_255_byte_packets;
1504         edev->stats.rx_511_byte_packets = stats.rx_511_byte_packets;
1505         edev->stats.rx_1023_byte_packets = stats.rx_1023_byte_packets;
1506         edev->stats.rx_1518_byte_packets = stats.rx_1518_byte_packets;
1507         edev->stats.rx_1522_byte_packets = stats.rx_1522_byte_packets;
1508         edev->stats.rx_2047_byte_packets = stats.rx_2047_byte_packets;
1509         edev->stats.rx_4095_byte_packets = stats.rx_4095_byte_packets;
1510         edev->stats.rx_9216_byte_packets = stats.rx_9216_byte_packets;
1511         edev->stats.rx_16383_byte_packets = stats.rx_16383_byte_packets;
1512         edev->stats.rx_crc_errors = stats.rx_crc_errors;
1513         edev->stats.rx_mac_crtl_frames = stats.rx_mac_crtl_frames;
1514         edev->stats.rx_pause_frames = stats.rx_pause_frames;
1515         edev->stats.rx_pfc_frames = stats.rx_pfc_frames;
1516         edev->stats.rx_align_errors = stats.rx_align_errors;
1517         edev->stats.rx_carrier_errors = stats.rx_carrier_errors;
1518         edev->stats.rx_oversize_packets = stats.rx_oversize_packets;
1519         edev->stats.rx_jabbers = stats.rx_jabbers;
1520         edev->stats.rx_undersize_packets = stats.rx_undersize_packets;
1521         edev->stats.rx_fragments = stats.rx_fragments;
1522         edev->stats.tx_64_byte_packets = stats.tx_64_byte_packets;
1523         edev->stats.tx_65_to_127_byte_packets = stats.tx_65_to_127_byte_packets;
1524         edev->stats.tx_128_to_255_byte_packets =
1525                                 stats.tx_128_to_255_byte_packets;
1526         edev->stats.tx_256_to_511_byte_packets =
1527                                 stats.tx_256_to_511_byte_packets;
1528         edev->stats.tx_512_to_1023_byte_packets =
1529                                 stats.tx_512_to_1023_byte_packets;
1530         edev->stats.tx_1024_to_1518_byte_packets =
1531                                 stats.tx_1024_to_1518_byte_packets;
1532         edev->stats.tx_1519_to_2047_byte_packets =
1533                                 stats.tx_1519_to_2047_byte_packets;
1534         edev->stats.tx_2048_to_4095_byte_packets =
1535                                 stats.tx_2048_to_4095_byte_packets;
1536         edev->stats.tx_4096_to_9216_byte_packets =
1537                                 stats.tx_4096_to_9216_byte_packets;
1538         edev->stats.tx_9217_to_16383_byte_packets =
1539                                 stats.tx_9217_to_16383_byte_packets;
1540         edev->stats.tx_pause_frames = stats.tx_pause_frames;
1541         edev->stats.tx_pfc_frames = stats.tx_pfc_frames;
1542         edev->stats.tx_lpi_entry_count = stats.tx_lpi_entry_count;
1543         edev->stats.tx_total_collisions = stats.tx_total_collisions;
1544         edev->stats.brb_truncates = stats.brb_truncates;
1545         edev->stats.brb_discards = stats.brb_discards;
1546         edev->stats.tx_mac_ctrl_frames = stats.tx_mac_ctrl_frames;
1547 }
1548
1549 static struct rtnl_link_stats64 *qede_get_stats64(
1550                             struct net_device *dev,
1551                             struct rtnl_link_stats64 *stats)
1552 {
1553         struct qede_dev *edev = netdev_priv(dev);
1554
1555         qede_fill_by_demand_stats(edev);
1556
1557         stats->rx_packets = edev->stats.rx_ucast_pkts +
1558                             edev->stats.rx_mcast_pkts +
1559                             edev->stats.rx_bcast_pkts;
1560         stats->tx_packets = edev->stats.tx_ucast_pkts +
1561                             edev->stats.tx_mcast_pkts +
1562                             edev->stats.tx_bcast_pkts;
1563
1564         stats->rx_bytes = edev->stats.rx_ucast_bytes +
1565                           edev->stats.rx_mcast_bytes +
1566                           edev->stats.rx_bcast_bytes;
1567
1568         stats->tx_bytes = edev->stats.tx_ucast_bytes +
1569                           edev->stats.tx_mcast_bytes +
1570                           edev->stats.tx_bcast_bytes;
1571
1572         stats->tx_errors = edev->stats.tx_err_drop_pkts;
1573         stats->multicast = edev->stats.rx_mcast_pkts +
1574                            edev->stats.rx_bcast_pkts;
1575
1576         stats->rx_fifo_errors = edev->stats.no_buff_discards;
1577
1578         stats->collisions = edev->stats.tx_total_collisions;
1579         stats->rx_crc_errors = edev->stats.rx_crc_errors;
1580         stats->rx_frame_errors = edev->stats.rx_align_errors;
1581
1582         return stats;
1583 }
1584
1585 static void qede_config_accept_any_vlan(struct qede_dev *edev, bool action)
1586 {
1587         struct qed_update_vport_params params;
1588         int rc;
1589
1590         /* Proceed only if action actually needs to be performed */
1591         if (edev->accept_any_vlan == action)
1592                 return;
1593
1594         memset(&params, 0, sizeof(params));
1595
1596         params.vport_id = 0;
1597         params.accept_any_vlan = action;
1598         params.update_accept_any_vlan_flg = 1;
1599
1600         rc = edev->ops->vport_update(edev->cdev, &params);
1601         if (rc) {
1602                 DP_ERR(edev, "Failed to %s accept-any-vlan\n",
1603                        action ? "enable" : "disable");
1604         } else {
1605                 DP_INFO(edev, "%s accept-any-vlan\n",
1606                         action ? "enabled" : "disabled");
1607                 edev->accept_any_vlan = action;
1608         }
1609 }
1610
1611 static int qede_vlan_rx_add_vid(struct net_device *dev, __be16 proto, u16 vid)
1612 {
1613         struct qede_dev *edev = netdev_priv(dev);
1614         struct qede_vlan *vlan, *tmp;
1615         int rc;
1616
1617         DP_VERBOSE(edev, NETIF_MSG_IFUP, "Adding vlan 0x%04x\n", vid);
1618
1619         vlan = kzalloc(sizeof(*vlan), GFP_KERNEL);
1620         if (!vlan) {
1621                 DP_INFO(edev, "Failed to allocate struct for vlan\n");
1622                 return -ENOMEM;
1623         }
1624         INIT_LIST_HEAD(&vlan->list);
1625         vlan->vid = vid;
1626         vlan->configured = false;
1627
1628         /* Verify vlan isn't already configured */
1629         list_for_each_entry(tmp, &edev->vlan_list, list) {
1630                 if (tmp->vid == vlan->vid) {
1631                         DP_VERBOSE(edev, (NETIF_MSG_IFUP | NETIF_MSG_IFDOWN),
1632                                    "vlan already configured\n");
1633                         kfree(vlan);
1634                         return -EEXIST;
1635                 }
1636         }
1637
1638         /* If interface is down, cache this VLAN ID and return */
1639         if (edev->state != QEDE_STATE_OPEN) {
1640                 DP_VERBOSE(edev, NETIF_MSG_IFDOWN,
1641                            "Interface is down, VLAN %d will be configured when interface is up\n",
1642                            vid);
1643                 if (vid != 0)
1644                         edev->non_configured_vlans++;
1645                 list_add(&vlan->list, &edev->vlan_list);
1646
1647                 return 0;
1648         }
1649
1650         /* Check for the filter limit.
1651          * Note - vlan0 has a reserved filter and can be added without
1652          * worrying about quota
1653          */
1654         if ((edev->configured_vlans < edev->dev_info.num_vlan_filters) ||
1655             (vlan->vid == 0)) {
1656                 rc = qede_set_ucast_rx_vlan(edev,
1657                                             QED_FILTER_XCAST_TYPE_ADD,
1658                                             vlan->vid);
1659                 if (rc) {
1660                         DP_ERR(edev, "Failed to configure VLAN %d\n",
1661                                vlan->vid);
1662                         kfree(vlan);
1663                         return -EINVAL;
1664                 }
1665                 vlan->configured = true;
1666
1667                 /* vlan0 filter isn't consuming out of our quota */
1668                 if (vlan->vid != 0)
1669                         edev->configured_vlans++;
1670         } else {
1671                 /* Out of quota; Activate accept-any-VLAN mode */
1672                 if (!edev->non_configured_vlans)
1673                         qede_config_accept_any_vlan(edev, true);
1674
1675                 edev->non_configured_vlans++;
1676         }
1677
1678         list_add(&vlan->list, &edev->vlan_list);
1679
1680         return 0;
1681 }
1682
1683 static void qede_del_vlan_from_list(struct qede_dev *edev,
1684                                     struct qede_vlan *vlan)
1685 {
1686         /* vlan0 filter isn't consuming out of our quota */
1687         if (vlan->vid != 0) {
1688                 if (vlan->configured)
1689                         edev->configured_vlans--;
1690                 else
1691                         edev->non_configured_vlans--;
1692         }
1693
1694         list_del(&vlan->list);
1695         kfree(vlan);
1696 }
1697
1698 static int qede_configure_vlan_filters(struct qede_dev *edev)
1699 {
1700         int rc = 0, real_rc = 0, accept_any_vlan = 0;
1701         struct qed_dev_eth_info *dev_info;
1702         struct qede_vlan *vlan = NULL;
1703
1704         if (list_empty(&edev->vlan_list))
1705                 return 0;
1706
1707         dev_info = &edev->dev_info;
1708
1709         /* Configure non-configured vlans */
1710         list_for_each_entry(vlan, &edev->vlan_list, list) {
1711                 if (vlan->configured)
1712                         continue;
1713
1714                 /* We have used all our credits, now enable accept_any_vlan */
1715                 if ((vlan->vid != 0) &&
1716                     (edev->configured_vlans == dev_info->num_vlan_filters)) {
1717                         accept_any_vlan = 1;
1718                         continue;
1719                 }
1720
1721                 DP_VERBOSE(edev, NETIF_MSG_IFUP, "Adding vlan %d\n", vlan->vid);
1722
1723                 rc = qede_set_ucast_rx_vlan(edev, QED_FILTER_XCAST_TYPE_ADD,
1724                                             vlan->vid);
1725                 if (rc) {
1726                         DP_ERR(edev, "Failed to configure VLAN %u\n",
1727                                vlan->vid);
1728                         real_rc = rc;
1729                         continue;
1730                 }
1731
1732                 vlan->configured = true;
1733                 /* vlan0 filter doesn't consume our VLAN filter's quota */
1734                 if (vlan->vid != 0) {
1735                         edev->non_configured_vlans--;
1736                         edev->configured_vlans++;
1737                 }
1738         }
1739
1740         /* enable accept_any_vlan mode if we have more VLANs than credits,
1741          * or remove accept_any_vlan mode if we've actually removed
1742          * a non-configured vlan, and all remaining vlans are truly configured.
1743          */
1744
1745         if (accept_any_vlan)
1746                 qede_config_accept_any_vlan(edev, true);
1747         else if (!edev->non_configured_vlans)
1748                 qede_config_accept_any_vlan(edev, false);
1749
1750         return real_rc;
1751 }
1752
1753 static int qede_vlan_rx_kill_vid(struct net_device *dev, __be16 proto, u16 vid)
1754 {
1755         struct qede_dev *edev = netdev_priv(dev);
1756         struct qede_vlan *vlan = NULL;
1757         int rc;
1758
1759         DP_VERBOSE(edev, NETIF_MSG_IFDOWN, "Removing vlan 0x%04x\n", vid);
1760
1761         /* Find whether entry exists */
1762         list_for_each_entry(vlan, &edev->vlan_list, list)
1763                 if (vlan->vid == vid)
1764                         break;
1765
1766         if (!vlan || (vlan->vid != vid)) {
1767                 DP_VERBOSE(edev, (NETIF_MSG_IFUP | NETIF_MSG_IFDOWN),
1768                            "Vlan isn't configured\n");
1769                 return 0;
1770         }
1771
1772         if (edev->state != QEDE_STATE_OPEN) {
1773                 /* As interface is already down, we don't have a VPORT
1774                  * instance to remove vlan filter. So just update vlan list
1775                  */
1776                 DP_VERBOSE(edev, NETIF_MSG_IFDOWN,
1777                            "Interface is down, removing VLAN from list only\n");
1778                 qede_del_vlan_from_list(edev, vlan);
1779                 return 0;
1780         }
1781
1782         /* Remove vlan */
1783         rc = qede_set_ucast_rx_vlan(edev, QED_FILTER_XCAST_TYPE_DEL, vid);
1784         if (rc) {
1785                 DP_ERR(edev, "Failed to remove VLAN %d\n", vid);
1786                 return -EINVAL;
1787         }
1788
1789         qede_del_vlan_from_list(edev, vlan);
1790
1791         /* We have removed a VLAN - try to see if we can
1792          * configure non-configured VLAN from the list.
1793          */
1794         rc = qede_configure_vlan_filters(edev);
1795
1796         return rc;
1797 }
1798
1799 static void qede_vlan_mark_nonconfigured(struct qede_dev *edev)
1800 {
1801         struct qede_vlan *vlan = NULL;
1802
1803         if (list_empty(&edev->vlan_list))
1804                 return;
1805
1806         list_for_each_entry(vlan, &edev->vlan_list, list) {
1807                 if (!vlan->configured)
1808                         continue;
1809
1810                 vlan->configured = false;
1811
1812                 /* vlan0 filter isn't consuming out of our quota */
1813                 if (vlan->vid != 0) {
1814                         edev->non_configured_vlans++;
1815                         edev->configured_vlans--;
1816                 }
1817
1818                 DP_VERBOSE(edev, NETIF_MSG_IFDOWN,
1819                            "marked vlan %d as non-configured\n",
1820                            vlan->vid);
1821         }
1822
1823         edev->accept_any_vlan = false;
1824 }
1825
1826 #ifdef CONFIG_QEDE_VXLAN
1827 static void qede_add_vxlan_port(struct net_device *dev,
1828                                 sa_family_t sa_family, __be16 port)
1829 {
1830         struct qede_dev *edev = netdev_priv(dev);
1831         u16 t_port = ntohs(port);
1832
1833         if (edev->vxlan_dst_port)
1834                 return;
1835
1836         edev->vxlan_dst_port = t_port;
1837
1838         DP_VERBOSE(edev, QED_MSG_DEBUG, "Added vxlan port=%d", t_port);
1839
1840         set_bit(QEDE_SP_VXLAN_PORT_CONFIG, &edev->sp_flags);
1841         schedule_delayed_work(&edev->sp_task, 0);
1842 }
1843
1844 static void qede_del_vxlan_port(struct net_device *dev,
1845                                 sa_family_t sa_family, __be16 port)
1846 {
1847         struct qede_dev *edev = netdev_priv(dev);
1848         u16 t_port = ntohs(port);
1849
1850         if (t_port != edev->vxlan_dst_port)
1851                 return;
1852
1853         edev->vxlan_dst_port = 0;
1854
1855         DP_VERBOSE(edev, QED_MSG_DEBUG, "Deleted vxlan port=%d", t_port);
1856
1857         set_bit(QEDE_SP_VXLAN_PORT_CONFIG, &edev->sp_flags);
1858         schedule_delayed_work(&edev->sp_task, 0);
1859 }
1860 #endif
1861
1862 static const struct net_device_ops qede_netdev_ops = {
1863         .ndo_open = qede_open,
1864         .ndo_stop = qede_close,
1865         .ndo_start_xmit = qede_start_xmit,
1866         .ndo_set_rx_mode = qede_set_rx_mode,
1867         .ndo_set_mac_address = qede_set_mac_addr,
1868         .ndo_validate_addr = eth_validate_addr,
1869         .ndo_change_mtu = qede_change_mtu,
1870         .ndo_vlan_rx_add_vid = qede_vlan_rx_add_vid,
1871         .ndo_vlan_rx_kill_vid = qede_vlan_rx_kill_vid,
1872         .ndo_get_stats64 = qede_get_stats64,
1873 #ifdef CONFIG_QEDE_VXLAN
1874         .ndo_add_vxlan_port = qede_add_vxlan_port,
1875         .ndo_del_vxlan_port = qede_del_vxlan_port,
1876 #endif
1877 };
1878
1879 /* -------------------------------------------------------------------------
1880  * START OF PROBE / REMOVE
1881  * -------------------------------------------------------------------------
1882  */
1883
1884 static struct qede_dev *qede_alloc_etherdev(struct qed_dev *cdev,
1885                                             struct pci_dev *pdev,
1886                                             struct qed_dev_eth_info *info,
1887                                             u32 dp_module,
1888                                             u8 dp_level)
1889 {
1890         struct net_device *ndev;
1891         struct qede_dev *edev;
1892
1893         ndev = alloc_etherdev_mqs(sizeof(*edev),
1894                                   info->num_queues,
1895                                   info->num_queues);
1896         if (!ndev) {
1897                 pr_err("etherdev allocation failed\n");
1898                 return NULL;
1899         }
1900
1901         edev = netdev_priv(ndev);
1902         edev->ndev = ndev;
1903         edev->cdev = cdev;
1904         edev->pdev = pdev;
1905         edev->dp_module = dp_module;
1906         edev->dp_level = dp_level;
1907         edev->ops = qed_ops;
1908         edev->q_num_rx_buffers = NUM_RX_BDS_DEF;
1909         edev->q_num_tx_buffers = NUM_TX_BDS_DEF;
1910
1911         DP_INFO(edev, "Allocated netdev with 64 tx queues and 64 rx queues\n");
1912
1913         SET_NETDEV_DEV(ndev, &pdev->dev);
1914
1915         memset(&edev->stats, 0, sizeof(edev->stats));
1916         memcpy(&edev->dev_info, info, sizeof(*info));
1917
1918         edev->num_tc = edev->dev_info.num_tc;
1919
1920         INIT_LIST_HEAD(&edev->vlan_list);
1921
1922         return edev;
1923 }
1924
1925 static void qede_init_ndev(struct qede_dev *edev)
1926 {
1927         struct net_device *ndev = edev->ndev;
1928         struct pci_dev *pdev = edev->pdev;
1929         u32 hw_features;
1930
1931         pci_set_drvdata(pdev, ndev);
1932
1933         ndev->mem_start = edev->dev_info.common.pci_mem_start;
1934         ndev->base_addr = ndev->mem_start;
1935         ndev->mem_end = edev->dev_info.common.pci_mem_end;
1936         ndev->irq = edev->dev_info.common.pci_irq;
1937
1938         ndev->watchdog_timeo = TX_TIMEOUT;
1939
1940         ndev->netdev_ops = &qede_netdev_ops;
1941
1942         qede_set_ethtool_ops(ndev);
1943
1944         /* user-changeble features */
1945         hw_features = NETIF_F_GRO | NETIF_F_SG |
1946                       NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
1947                       NETIF_F_TSO | NETIF_F_TSO6;
1948
1949         ndev->vlan_features = hw_features | NETIF_F_RXHASH | NETIF_F_RXCSUM |
1950                               NETIF_F_HIGHDMA;
1951         ndev->features = hw_features | NETIF_F_RXHASH | NETIF_F_RXCSUM |
1952                          NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_HIGHDMA |
1953                          NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_CTAG_TX;
1954
1955         ndev->hw_features = hw_features;
1956
1957         /* Set network device HW mac */
1958         ether_addr_copy(edev->ndev->dev_addr, edev->dev_info.common.hw_mac);
1959 }
1960
1961 /* This function converts from 32b param to two params of level and module
1962  * Input 32b decoding:
1963  * b31 - enable all NOTICE prints. NOTICE prints are for deviation from the
1964  * 'happy' flow, e.g. memory allocation failed.
1965  * b30 - enable all INFO prints. INFO prints are for major steps in the flow
1966  * and provide important parameters.
1967  * b29-b0 - per-module bitmap, where each bit enables VERBOSE prints of that
1968  * module. VERBOSE prints are for tracking the specific flow in low level.
1969  *
1970  * Notice that the level should be that of the lowest required logs.
1971  */
1972 void qede_config_debug(uint debug, u32 *p_dp_module, u8 *p_dp_level)
1973 {
1974         *p_dp_level = QED_LEVEL_NOTICE;
1975         *p_dp_module = 0;
1976
1977         if (debug & QED_LOG_VERBOSE_MASK) {
1978                 *p_dp_level = QED_LEVEL_VERBOSE;
1979                 *p_dp_module = (debug & 0x3FFFFFFF);
1980         } else if (debug & QED_LOG_INFO_MASK) {
1981                 *p_dp_level = QED_LEVEL_INFO;
1982         } else if (debug & QED_LOG_NOTICE_MASK) {
1983                 *p_dp_level = QED_LEVEL_NOTICE;
1984         }
1985 }
1986
1987 static void qede_free_fp_array(struct qede_dev *edev)
1988 {
1989         if (edev->fp_array) {
1990                 struct qede_fastpath *fp;
1991                 int i;
1992
1993                 for_each_rss(i) {
1994                         fp = &edev->fp_array[i];
1995
1996                         kfree(fp->sb_info);
1997                         kfree(fp->rxq);
1998                         kfree(fp->txqs);
1999                 }
2000                 kfree(edev->fp_array);
2001         }
2002         edev->num_rss = 0;
2003 }
2004
2005 static int qede_alloc_fp_array(struct qede_dev *edev)
2006 {
2007         struct qede_fastpath *fp;
2008         int i;
2009
2010         edev->fp_array = kcalloc(QEDE_RSS_CNT(edev),
2011                                  sizeof(*edev->fp_array), GFP_KERNEL);
2012         if (!edev->fp_array) {
2013                 DP_NOTICE(edev, "fp array allocation failed\n");
2014                 goto err;
2015         }
2016
2017         for_each_rss(i) {
2018                 fp = &edev->fp_array[i];
2019
2020                 fp->sb_info = kcalloc(1, sizeof(*fp->sb_info), GFP_KERNEL);
2021                 if (!fp->sb_info) {
2022                         DP_NOTICE(edev, "sb info struct allocation failed\n");
2023                         goto err;
2024                 }
2025
2026                 fp->rxq = kcalloc(1, sizeof(*fp->rxq), GFP_KERNEL);
2027                 if (!fp->rxq) {
2028                         DP_NOTICE(edev, "RXQ struct allocation failed\n");
2029                         goto err;
2030                 }
2031
2032                 fp->txqs = kcalloc(edev->num_tc, sizeof(*fp->txqs), GFP_KERNEL);
2033                 if (!fp->txqs) {
2034                         DP_NOTICE(edev, "TXQ array allocation failed\n");
2035                         goto err;
2036                 }
2037         }
2038
2039         return 0;
2040 err:
2041         qede_free_fp_array(edev);
2042         return -ENOMEM;
2043 }
2044
2045 static void qede_sp_task(struct work_struct *work)
2046 {
2047         struct qede_dev *edev = container_of(work, struct qede_dev,
2048                                              sp_task.work);
2049         struct qed_dev *cdev = edev->cdev;
2050
2051         mutex_lock(&edev->qede_lock);
2052
2053         if (edev->state == QEDE_STATE_OPEN) {
2054                 if (test_and_clear_bit(QEDE_SP_RX_MODE, &edev->sp_flags))
2055                         qede_config_rx_mode(edev->ndev);
2056         }
2057
2058         if (test_and_clear_bit(QEDE_SP_VXLAN_PORT_CONFIG, &edev->sp_flags)) {
2059                 struct qed_tunn_params tunn_params;
2060
2061                 memset(&tunn_params, 0, sizeof(tunn_params));
2062                 tunn_params.update_vxlan_port = 1;
2063                 tunn_params.vxlan_port = edev->vxlan_dst_port;
2064                 qed_ops->tunn_config(cdev, &tunn_params);
2065         }
2066
2067         mutex_unlock(&edev->qede_lock);
2068 }
2069
2070 static void qede_update_pf_params(struct qed_dev *cdev)
2071 {
2072         struct qed_pf_params pf_params;
2073
2074         /* 16 rx + 16 tx */
2075         memset(&pf_params, 0, sizeof(struct qed_pf_params));
2076         pf_params.eth_pf_params.num_cons = 32;
2077         qed_ops->common->update_pf_params(cdev, &pf_params);
2078 }
2079
2080 enum qede_probe_mode {
2081         QEDE_PROBE_NORMAL,
2082 };
2083
2084 static int __qede_probe(struct pci_dev *pdev, u32 dp_module, u8 dp_level,
2085                         enum qede_probe_mode mode)
2086 {
2087         struct qed_slowpath_params params;
2088         struct qed_dev_eth_info dev_info;
2089         struct qede_dev *edev;
2090         struct qed_dev *cdev;
2091         int rc;
2092
2093         if (unlikely(dp_level & QED_LEVEL_INFO))
2094                 pr_notice("Starting qede probe\n");
2095
2096         cdev = qed_ops->common->probe(pdev, QED_PROTOCOL_ETH,
2097                                       dp_module, dp_level);
2098         if (!cdev) {
2099                 rc = -ENODEV;
2100                 goto err0;
2101         }
2102
2103         qede_update_pf_params(cdev);
2104
2105         /* Start the Slowpath-process */
2106         memset(&params, 0, sizeof(struct qed_slowpath_params));
2107         params.int_mode = QED_INT_MODE_MSIX;
2108         params.drv_major = QEDE_MAJOR_VERSION;
2109         params.drv_minor = QEDE_MINOR_VERSION;
2110         params.drv_rev = QEDE_REVISION_VERSION;
2111         params.drv_eng = QEDE_ENGINEERING_VERSION;
2112         strlcpy(params.name, "qede LAN", QED_DRV_VER_STR_SIZE);
2113         rc = qed_ops->common->slowpath_start(cdev, &params);
2114         if (rc) {
2115                 pr_notice("Cannot start slowpath\n");
2116                 goto err1;
2117         }
2118
2119         /* Learn information crucial for qede to progress */
2120         rc = qed_ops->fill_dev_info(cdev, &dev_info);
2121         if (rc)
2122                 goto err2;
2123
2124         edev = qede_alloc_etherdev(cdev, pdev, &dev_info, dp_module,
2125                                    dp_level);
2126         if (!edev) {
2127                 rc = -ENOMEM;
2128                 goto err2;
2129         }
2130
2131         qede_init_ndev(edev);
2132
2133         rc = register_netdev(edev->ndev);
2134         if (rc) {
2135                 DP_NOTICE(edev, "Cannot register net-device\n");
2136                 goto err3;
2137         }
2138
2139         edev->ops->common->set_id(cdev, edev->ndev->name, DRV_MODULE_VERSION);
2140
2141         edev->ops->register_ops(cdev, &qede_ll_ops, edev);
2142
2143         INIT_DELAYED_WORK(&edev->sp_task, qede_sp_task);
2144         mutex_init(&edev->qede_lock);
2145
2146         DP_INFO(edev, "Ending successfully qede probe\n");
2147
2148         return 0;
2149
2150 err3:
2151         free_netdev(edev->ndev);
2152 err2:
2153         qed_ops->common->slowpath_stop(cdev);
2154 err1:
2155         qed_ops->common->remove(cdev);
2156 err0:
2157         return rc;
2158 }
2159
2160 static int qede_probe(struct pci_dev *pdev, const struct pci_device_id *id)
2161 {
2162         u32 dp_module = 0;
2163         u8 dp_level = 0;
2164
2165         qede_config_debug(debug, &dp_module, &dp_level);
2166
2167         return __qede_probe(pdev, dp_module, dp_level,
2168                             QEDE_PROBE_NORMAL);
2169 }
2170
2171 enum qede_remove_mode {
2172         QEDE_REMOVE_NORMAL,
2173 };
2174
2175 static void __qede_remove(struct pci_dev *pdev, enum qede_remove_mode mode)
2176 {
2177         struct net_device *ndev = pci_get_drvdata(pdev);
2178         struct qede_dev *edev = netdev_priv(ndev);
2179         struct qed_dev *cdev = edev->cdev;
2180
2181         DP_INFO(edev, "Starting qede_remove\n");
2182
2183         cancel_delayed_work_sync(&edev->sp_task);
2184         unregister_netdev(ndev);
2185
2186         edev->ops->common->set_power_state(cdev, PCI_D0);
2187
2188         pci_set_drvdata(pdev, NULL);
2189
2190         free_netdev(ndev);
2191
2192         /* Use global ops since we've freed edev */
2193         qed_ops->common->slowpath_stop(cdev);
2194         qed_ops->common->remove(cdev);
2195
2196         pr_notice("Ending successfully qede_remove\n");
2197 }
2198
2199 static void qede_remove(struct pci_dev *pdev)
2200 {
2201         __qede_remove(pdev, QEDE_REMOVE_NORMAL);
2202 }
2203
2204 /* -------------------------------------------------------------------------
2205  * START OF LOAD / UNLOAD
2206  * -------------------------------------------------------------------------
2207  */
2208
2209 static int qede_set_num_queues(struct qede_dev *edev)
2210 {
2211         int rc;
2212         u16 rss_num;
2213
2214         /* Setup queues according to possible resources*/
2215         if (edev->req_rss)
2216                 rss_num = edev->req_rss;
2217         else
2218                 rss_num = netif_get_num_default_rss_queues() *
2219                           edev->dev_info.common.num_hwfns;
2220
2221         rss_num = min_t(u16, QEDE_MAX_RSS_CNT(edev), rss_num);
2222
2223         rc = edev->ops->common->set_fp_int(edev->cdev, rss_num);
2224         if (rc > 0) {
2225                 /* Managed to request interrupts for our queues */
2226                 edev->num_rss = rc;
2227                 DP_INFO(edev, "Managed %d [of %d] RSS queues\n",
2228                         QEDE_RSS_CNT(edev), rss_num);
2229                 rc = 0;
2230         }
2231         return rc;
2232 }
2233
2234 static void qede_free_mem_sb(struct qede_dev *edev,
2235                              struct qed_sb_info *sb_info)
2236 {
2237         if (sb_info->sb_virt)
2238                 dma_free_coherent(&edev->pdev->dev, sizeof(*sb_info->sb_virt),
2239                                   (void *)sb_info->sb_virt, sb_info->sb_phys);
2240 }
2241
2242 /* This function allocates fast-path status block memory */
2243 static int qede_alloc_mem_sb(struct qede_dev *edev,
2244                              struct qed_sb_info *sb_info,
2245                              u16 sb_id)
2246 {
2247         struct status_block *sb_virt;
2248         dma_addr_t sb_phys;
2249         int rc;
2250
2251         sb_virt = dma_alloc_coherent(&edev->pdev->dev,
2252                                      sizeof(*sb_virt),
2253                                      &sb_phys, GFP_KERNEL);
2254         if (!sb_virt) {
2255                 DP_ERR(edev, "Status block allocation failed\n");
2256                 return -ENOMEM;
2257         }
2258
2259         rc = edev->ops->common->sb_init(edev->cdev, sb_info,
2260                                         sb_virt, sb_phys, sb_id,
2261                                         QED_SB_TYPE_L2_QUEUE);
2262         if (rc) {
2263                 DP_ERR(edev, "Status block initialization failed\n");
2264                 dma_free_coherent(&edev->pdev->dev, sizeof(*sb_virt),
2265                                   sb_virt, sb_phys);
2266                 return rc;
2267         }
2268
2269         return 0;
2270 }
2271
2272 static void qede_free_rx_buffers(struct qede_dev *edev,
2273                                  struct qede_rx_queue *rxq)
2274 {
2275         u16 i;
2276
2277         for (i = rxq->sw_rx_cons; i != rxq->sw_rx_prod; i++) {
2278                 struct sw_rx_data *rx_buf;
2279                 struct page *data;
2280
2281                 rx_buf = &rxq->sw_rx_ring[i & NUM_RX_BDS_MAX];
2282                 data = rx_buf->data;
2283
2284                 dma_unmap_page(&edev->pdev->dev,
2285                                rx_buf->mapping,
2286                                PAGE_SIZE, DMA_FROM_DEVICE);
2287
2288                 rx_buf->data = NULL;
2289                 __free_page(data);
2290         }
2291 }
2292
2293 static void qede_free_sge_mem(struct qede_dev *edev,
2294                               struct qede_rx_queue *rxq) {
2295         int i;
2296
2297         if (edev->gro_disable)
2298                 return;
2299
2300         for (i = 0; i < ETH_TPA_MAX_AGGS_NUM; i++) {
2301                 struct qede_agg_info *tpa_info = &rxq->tpa_info[i];
2302                 struct sw_rx_data *replace_buf = &tpa_info->replace_buf;
2303
2304                 if (replace_buf) {
2305                         dma_unmap_page(&edev->pdev->dev,
2306                                        dma_unmap_addr(replace_buf, mapping),
2307                                        PAGE_SIZE, DMA_FROM_DEVICE);
2308                         __free_page(replace_buf->data);
2309                 }
2310         }
2311 }
2312
2313 static void qede_free_mem_rxq(struct qede_dev *edev,
2314                               struct qede_rx_queue *rxq)
2315 {
2316         qede_free_sge_mem(edev, rxq);
2317
2318         /* Free rx buffers */
2319         qede_free_rx_buffers(edev, rxq);
2320
2321         /* Free the parallel SW ring */
2322         kfree(rxq->sw_rx_ring);
2323
2324         /* Free the real RQ ring used by FW */
2325         edev->ops->common->chain_free(edev->cdev, &rxq->rx_bd_ring);
2326         edev->ops->common->chain_free(edev->cdev, &rxq->rx_comp_ring);
2327 }
2328
2329 static int qede_alloc_rx_buffer(struct qede_dev *edev,
2330                                 struct qede_rx_queue *rxq)
2331 {
2332         struct sw_rx_data *sw_rx_data;
2333         struct eth_rx_bd *rx_bd;
2334         dma_addr_t mapping;
2335         struct page *data;
2336         u16 rx_buf_size;
2337
2338         rx_buf_size = rxq->rx_buf_size;
2339
2340         data = alloc_pages(GFP_ATOMIC, 0);
2341         if (unlikely(!data)) {
2342                 DP_NOTICE(edev, "Failed to allocate Rx data [page]\n");
2343                 return -ENOMEM;
2344         }
2345
2346         /* Map the entire page as it would be used
2347          * for multiple RX buffer segment size mapping.
2348          */
2349         mapping = dma_map_page(&edev->pdev->dev, data, 0,
2350                                PAGE_SIZE, DMA_FROM_DEVICE);
2351         if (unlikely(dma_mapping_error(&edev->pdev->dev, mapping))) {
2352                 __free_page(data);
2353                 DP_NOTICE(edev, "Failed to map Rx buffer\n");
2354                 return -ENOMEM;
2355         }
2356
2357         sw_rx_data = &rxq->sw_rx_ring[rxq->sw_rx_prod & NUM_RX_BDS_MAX];
2358         sw_rx_data->page_offset = 0;
2359         sw_rx_data->data = data;
2360         sw_rx_data->mapping = mapping;
2361
2362         /* Advance PROD and get BD pointer */
2363         rx_bd = (struct eth_rx_bd *)qed_chain_produce(&rxq->rx_bd_ring);
2364         WARN_ON(!rx_bd);
2365         rx_bd->addr.hi = cpu_to_le32(upper_32_bits(mapping));
2366         rx_bd->addr.lo = cpu_to_le32(lower_32_bits(mapping));
2367
2368         rxq->sw_rx_prod++;
2369
2370         return 0;
2371 }
2372
2373 static int qede_alloc_sge_mem(struct qede_dev *edev,
2374                               struct qede_rx_queue *rxq)
2375 {
2376         dma_addr_t mapping;
2377         int i;
2378
2379         if (edev->gro_disable)
2380                 return 0;
2381
2382         if (edev->ndev->mtu > PAGE_SIZE) {
2383                 edev->gro_disable = 1;
2384                 return 0;
2385         }
2386
2387         for (i = 0; i < ETH_TPA_MAX_AGGS_NUM; i++) {
2388                 struct qede_agg_info *tpa_info = &rxq->tpa_info[i];
2389                 struct sw_rx_data *replace_buf = &tpa_info->replace_buf;
2390
2391                 replace_buf->data = alloc_pages(GFP_ATOMIC, 0);
2392                 if (unlikely(!replace_buf->data)) {
2393                         DP_NOTICE(edev,
2394                                   "Failed to allocate TPA skb pool [replacement buffer]\n");
2395                         goto err;
2396                 }
2397
2398                 mapping = dma_map_page(&edev->pdev->dev, replace_buf->data, 0,
2399                                        rxq->rx_buf_size, DMA_FROM_DEVICE);
2400                 if (unlikely(dma_mapping_error(&edev->pdev->dev, mapping))) {
2401                         DP_NOTICE(edev,
2402                                   "Failed to map TPA replacement buffer\n");
2403                         goto err;
2404                 }
2405
2406                 dma_unmap_addr_set(replace_buf, mapping, mapping);
2407                 tpa_info->replace_buf.page_offset = 0;
2408
2409                 tpa_info->replace_buf_mapping = mapping;
2410                 tpa_info->agg_state = QEDE_AGG_STATE_NONE;
2411         }
2412
2413         return 0;
2414 err:
2415         qede_free_sge_mem(edev, rxq);
2416         edev->gro_disable = 1;
2417         return -ENOMEM;
2418 }
2419
2420 /* This function allocates all memory needed per Rx queue */
2421 static int qede_alloc_mem_rxq(struct qede_dev *edev,
2422                               struct qede_rx_queue *rxq)
2423 {
2424         int i, rc, size, num_allocated;
2425
2426         rxq->num_rx_buffers = edev->q_num_rx_buffers;
2427
2428         rxq->rx_buf_size = NET_IP_ALIGN + ETH_OVERHEAD +
2429                            edev->ndev->mtu;
2430         if (rxq->rx_buf_size > PAGE_SIZE)
2431                 rxq->rx_buf_size = PAGE_SIZE;
2432
2433         /* Segment size to spilt a page in multiple equal parts */
2434         rxq->rx_buf_seg_size = roundup_pow_of_two(rxq->rx_buf_size);
2435
2436         /* Allocate the parallel driver ring for Rx buffers */
2437         size = sizeof(*rxq->sw_rx_ring) * RX_RING_SIZE;
2438         rxq->sw_rx_ring = kzalloc(size, GFP_KERNEL);
2439         if (!rxq->sw_rx_ring) {
2440                 DP_ERR(edev, "Rx buffers ring allocation failed\n");
2441                 goto err;
2442         }
2443
2444         /* Allocate FW Rx ring  */
2445         rc = edev->ops->common->chain_alloc(edev->cdev,
2446                                             QED_CHAIN_USE_TO_CONSUME_PRODUCE,
2447                                             QED_CHAIN_MODE_NEXT_PTR,
2448                                             RX_RING_SIZE,
2449                                             sizeof(struct eth_rx_bd),
2450                                             &rxq->rx_bd_ring);
2451
2452         if (rc)
2453                 goto err;
2454
2455         /* Allocate FW completion ring */
2456         rc = edev->ops->common->chain_alloc(edev->cdev,
2457                                             QED_CHAIN_USE_TO_CONSUME,
2458                                             QED_CHAIN_MODE_PBL,
2459                                             RX_RING_SIZE,
2460                                             sizeof(union eth_rx_cqe),
2461                                             &rxq->rx_comp_ring);
2462         if (rc)
2463                 goto err;
2464
2465         /* Allocate buffers for the Rx ring */
2466         for (i = 0; i < rxq->num_rx_buffers; i++) {
2467                 rc = qede_alloc_rx_buffer(edev, rxq);
2468                 if (rc)
2469                         break;
2470         }
2471         num_allocated = i;
2472         if (!num_allocated) {
2473                 DP_ERR(edev, "Rx buffers allocation failed\n");
2474                 goto err;
2475         } else if (num_allocated < rxq->num_rx_buffers) {
2476                 DP_NOTICE(edev,
2477                           "Allocated less buffers than desired (%d allocated)\n",
2478                           num_allocated);
2479         }
2480
2481         qede_alloc_sge_mem(edev, rxq);
2482
2483         return 0;
2484
2485 err:
2486         qede_free_mem_rxq(edev, rxq);
2487         return -ENOMEM;
2488 }
2489
2490 static void qede_free_mem_txq(struct qede_dev *edev,
2491                               struct qede_tx_queue *txq)
2492 {
2493         /* Free the parallel SW ring */
2494         kfree(txq->sw_tx_ring);
2495
2496         /* Free the real RQ ring used by FW */
2497         edev->ops->common->chain_free(edev->cdev, &txq->tx_pbl);
2498 }
2499
2500 /* This function allocates all memory needed per Tx queue */
2501 static int qede_alloc_mem_txq(struct qede_dev *edev,
2502                               struct qede_tx_queue *txq)
2503 {
2504         int size, rc;
2505         union eth_tx_bd_types *p_virt;
2506
2507         txq->num_tx_buffers = edev->q_num_tx_buffers;
2508
2509         /* Allocate the parallel driver ring for Tx buffers */
2510         size = sizeof(*txq->sw_tx_ring) * NUM_TX_BDS_MAX;
2511         txq->sw_tx_ring = kzalloc(size, GFP_KERNEL);
2512         if (!txq->sw_tx_ring) {
2513                 DP_NOTICE(edev, "Tx buffers ring allocation failed\n");
2514                 goto err;
2515         }
2516
2517         rc = edev->ops->common->chain_alloc(edev->cdev,
2518                                             QED_CHAIN_USE_TO_CONSUME_PRODUCE,
2519                                             QED_CHAIN_MODE_PBL,
2520                                             NUM_TX_BDS_MAX,
2521                                             sizeof(*p_virt),
2522                                             &txq->tx_pbl);
2523         if (rc)
2524                 goto err;
2525
2526         return 0;
2527
2528 err:
2529         qede_free_mem_txq(edev, txq);
2530         return -ENOMEM;
2531 }
2532
2533 /* This function frees all memory of a single fp */
2534 static void qede_free_mem_fp(struct qede_dev *edev,
2535                              struct qede_fastpath *fp)
2536 {
2537         int tc;
2538
2539         qede_free_mem_sb(edev, fp->sb_info);
2540
2541         qede_free_mem_rxq(edev, fp->rxq);
2542
2543         for (tc = 0; tc < edev->num_tc; tc++)
2544                 qede_free_mem_txq(edev, &fp->txqs[tc]);
2545 }
2546
2547 /* This function allocates all memory needed for a single fp (i.e. an entity
2548  * which contains status block, one rx queue and multiple per-TC tx queues.
2549  */
2550 static int qede_alloc_mem_fp(struct qede_dev *edev,
2551                              struct qede_fastpath *fp)
2552 {
2553         int rc, tc;
2554
2555         rc = qede_alloc_mem_sb(edev, fp->sb_info, fp->rss_id);
2556         if (rc)
2557                 goto err;
2558
2559         rc = qede_alloc_mem_rxq(edev, fp->rxq);
2560         if (rc)
2561                 goto err;
2562
2563         for (tc = 0; tc < edev->num_tc; tc++) {
2564                 rc = qede_alloc_mem_txq(edev, &fp->txqs[tc]);
2565                 if (rc)
2566                         goto err;
2567         }
2568
2569         return 0;
2570
2571 err:
2572         qede_free_mem_fp(edev, fp);
2573         return -ENOMEM;
2574 }
2575
2576 static void qede_free_mem_load(struct qede_dev *edev)
2577 {
2578         int i;
2579
2580         for_each_rss(i) {
2581                 struct qede_fastpath *fp = &edev->fp_array[i];
2582
2583                 qede_free_mem_fp(edev, fp);
2584         }
2585 }
2586
2587 /* This function allocates all qede memory at NIC load. */
2588 static int qede_alloc_mem_load(struct qede_dev *edev)
2589 {
2590         int rc = 0, rss_id;
2591
2592         for (rss_id = 0; rss_id < QEDE_RSS_CNT(edev); rss_id++) {
2593                 struct qede_fastpath *fp = &edev->fp_array[rss_id];
2594
2595                 rc = qede_alloc_mem_fp(edev, fp);
2596                 if (rc)
2597                         break;
2598         }
2599
2600         if (rss_id != QEDE_RSS_CNT(edev)) {
2601                 /* Failed allocating memory for all the queues */
2602                 if (!rss_id) {
2603                         DP_ERR(edev,
2604                                "Failed to allocate memory for the leading queue\n");
2605                         rc = -ENOMEM;
2606                 } else {
2607                         DP_NOTICE(edev,
2608                                   "Failed to allocate memory for all of RSS queues\n Desired: %d queues, allocated: %d queues\n",
2609                                   QEDE_RSS_CNT(edev), rss_id);
2610                 }
2611                 edev->num_rss = rss_id;
2612         }
2613
2614         return 0;
2615 }
2616
2617 /* This function inits fp content and resets the SB, RXQ and TXQ structures */
2618 static void qede_init_fp(struct qede_dev *edev)
2619 {
2620         int rss_id, txq_index, tc;
2621         struct qede_fastpath *fp;
2622
2623         for_each_rss(rss_id) {
2624                 fp = &edev->fp_array[rss_id];
2625
2626                 fp->edev = edev;
2627                 fp->rss_id = rss_id;
2628
2629                 memset((void *)&fp->napi, 0, sizeof(fp->napi));
2630
2631                 memset((void *)fp->sb_info, 0, sizeof(*fp->sb_info));
2632
2633                 memset((void *)fp->rxq, 0, sizeof(*fp->rxq));
2634                 fp->rxq->rxq_id = rss_id;
2635
2636                 memset((void *)fp->txqs, 0, (edev->num_tc * sizeof(*fp->txqs)));
2637                 for (tc = 0; tc < edev->num_tc; tc++) {
2638                         txq_index = tc * QEDE_RSS_CNT(edev) + rss_id;
2639                         fp->txqs[tc].index = txq_index;
2640                 }
2641
2642                 snprintf(fp->name, sizeof(fp->name), "%s-fp-%d",
2643                          edev->ndev->name, rss_id);
2644         }
2645
2646         edev->gro_disable = !(edev->ndev->features & NETIF_F_GRO);
2647 }
2648
2649 static int qede_set_real_num_queues(struct qede_dev *edev)
2650 {
2651         int rc = 0;
2652
2653         rc = netif_set_real_num_tx_queues(edev->ndev, QEDE_TSS_CNT(edev));
2654         if (rc) {
2655                 DP_NOTICE(edev, "Failed to set real number of Tx queues\n");
2656                 return rc;
2657         }
2658         rc = netif_set_real_num_rx_queues(edev->ndev, QEDE_RSS_CNT(edev));
2659         if (rc) {
2660                 DP_NOTICE(edev, "Failed to set real number of Rx queues\n");
2661                 return rc;
2662         }
2663
2664         return 0;
2665 }
2666
2667 static void qede_napi_disable_remove(struct qede_dev *edev)
2668 {
2669         int i;
2670
2671         for_each_rss(i) {
2672                 napi_disable(&edev->fp_array[i].napi);
2673
2674                 netif_napi_del(&edev->fp_array[i].napi);
2675         }
2676 }
2677
2678 static void qede_napi_add_enable(struct qede_dev *edev)
2679 {
2680         int i;
2681
2682         /* Add NAPI objects */
2683         for_each_rss(i) {
2684                 netif_napi_add(edev->ndev, &edev->fp_array[i].napi,
2685                                qede_poll, NAPI_POLL_WEIGHT);
2686                 napi_enable(&edev->fp_array[i].napi);
2687         }
2688 }
2689
2690 static void qede_sync_free_irqs(struct qede_dev *edev)
2691 {
2692         int i;
2693
2694         for (i = 0; i < edev->int_info.used_cnt; i++) {
2695                 if (edev->int_info.msix_cnt) {
2696                         synchronize_irq(edev->int_info.msix[i].vector);
2697                         free_irq(edev->int_info.msix[i].vector,
2698                                  &edev->fp_array[i]);
2699                 } else {
2700                         edev->ops->common->simd_handler_clean(edev->cdev, i);
2701                 }
2702         }
2703
2704         edev->int_info.used_cnt = 0;
2705 }
2706
2707 static int qede_req_msix_irqs(struct qede_dev *edev)
2708 {
2709         int i, rc;
2710
2711         /* Sanitize number of interrupts == number of prepared RSS queues */
2712         if (QEDE_RSS_CNT(edev) > edev->int_info.msix_cnt) {
2713                 DP_ERR(edev,
2714                        "Interrupt mismatch: %d RSS queues > %d MSI-x vectors\n",
2715                        QEDE_RSS_CNT(edev), edev->int_info.msix_cnt);
2716                 return -EINVAL;
2717         }
2718
2719         for (i = 0; i < QEDE_RSS_CNT(edev); i++) {
2720                 rc = request_irq(edev->int_info.msix[i].vector,
2721                                  qede_msix_fp_int, 0, edev->fp_array[i].name,
2722                                  &edev->fp_array[i]);
2723                 if (rc) {
2724                         DP_ERR(edev, "Request fp %d irq failed\n", i);
2725                         qede_sync_free_irqs(edev);
2726                         return rc;
2727                 }
2728                 DP_VERBOSE(edev, NETIF_MSG_INTR,
2729                            "Requested fp irq for %s [entry %d]. Cookie is at %p\n",
2730                            edev->fp_array[i].name, i,
2731                            &edev->fp_array[i]);
2732                 edev->int_info.used_cnt++;
2733         }
2734
2735         return 0;
2736 }
2737
2738 static void qede_simd_fp_handler(void *cookie)
2739 {
2740         struct qede_fastpath *fp = (struct qede_fastpath *)cookie;
2741
2742         napi_schedule_irqoff(&fp->napi);
2743 }
2744
2745 static int qede_setup_irqs(struct qede_dev *edev)
2746 {
2747         int i, rc = 0;
2748
2749         /* Learn Interrupt configuration */
2750         rc = edev->ops->common->get_fp_int(edev->cdev, &edev->int_info);
2751         if (rc)
2752                 return rc;
2753
2754         if (edev->int_info.msix_cnt) {
2755                 rc = qede_req_msix_irqs(edev);
2756                 if (rc)
2757                         return rc;
2758                 edev->ndev->irq = edev->int_info.msix[0].vector;
2759         } else {
2760                 const struct qed_common_ops *ops;
2761
2762                 /* qed should learn receive the RSS ids and callbacks */
2763                 ops = edev->ops->common;
2764                 for (i = 0; i < QEDE_RSS_CNT(edev); i++)
2765                         ops->simd_handler_config(edev->cdev,
2766                                                  &edev->fp_array[i], i,
2767                                                  qede_simd_fp_handler);
2768                 edev->int_info.used_cnt = QEDE_RSS_CNT(edev);
2769         }
2770         return 0;
2771 }
2772
2773 static int qede_drain_txq(struct qede_dev *edev,
2774                           struct qede_tx_queue *txq,
2775                           bool allow_drain)
2776 {
2777         int rc, cnt = 1000;
2778
2779         while (txq->sw_tx_cons != txq->sw_tx_prod) {
2780                 if (!cnt) {
2781                         if (allow_drain) {
2782                                 DP_NOTICE(edev,
2783                                           "Tx queue[%d] is stuck, requesting MCP to drain\n",
2784                                           txq->index);
2785                                 rc = edev->ops->common->drain(edev->cdev);
2786                                 if (rc)
2787                                         return rc;
2788                                 return qede_drain_txq(edev, txq, false);
2789                         }
2790                         DP_NOTICE(edev,
2791                                   "Timeout waiting for tx queue[%d]: PROD=%d, CONS=%d\n",
2792                                   txq->index, txq->sw_tx_prod,
2793                                   txq->sw_tx_cons);
2794                         return -ENODEV;
2795                 }
2796                 cnt--;
2797                 usleep_range(1000, 2000);
2798                 barrier();
2799         }
2800
2801         /* FW finished processing, wait for HW to transmit all tx packets */
2802         usleep_range(1000, 2000);
2803
2804         return 0;
2805 }
2806
2807 static int qede_stop_queues(struct qede_dev *edev)
2808 {
2809         struct qed_update_vport_params vport_update_params;
2810         struct qed_dev *cdev = edev->cdev;
2811         int rc, tc, i;
2812
2813         /* Disable the vport */
2814         memset(&vport_update_params, 0, sizeof(vport_update_params));
2815         vport_update_params.vport_id = 0;
2816         vport_update_params.update_vport_active_flg = 1;
2817         vport_update_params.vport_active_flg = 0;
2818         vport_update_params.update_rss_flg = 0;
2819
2820         rc = edev->ops->vport_update(cdev, &vport_update_params);
2821         if (rc) {
2822                 DP_ERR(edev, "Failed to update vport\n");
2823                 return rc;
2824         }
2825
2826         /* Flush Tx queues. If needed, request drain from MCP */
2827         for_each_rss(i) {
2828                 struct qede_fastpath *fp = &edev->fp_array[i];
2829
2830                 for (tc = 0; tc < edev->num_tc; tc++) {
2831                         struct qede_tx_queue *txq = &fp->txqs[tc];
2832
2833                         rc = qede_drain_txq(edev, txq, true);
2834                         if (rc)
2835                                 return rc;
2836                 }
2837         }
2838
2839         /* Stop all Queues in reverse order*/
2840         for (i = QEDE_RSS_CNT(edev) - 1; i >= 0; i--) {
2841                 struct qed_stop_rxq_params rx_params;
2842
2843                 /* Stop the Tx Queue(s)*/
2844                 for (tc = 0; tc < edev->num_tc; tc++) {
2845                         struct qed_stop_txq_params tx_params;
2846
2847                         tx_params.rss_id = i;
2848                         tx_params.tx_queue_id = tc * QEDE_RSS_CNT(edev) + i;
2849                         rc = edev->ops->q_tx_stop(cdev, &tx_params);
2850                         if (rc) {
2851                                 DP_ERR(edev, "Failed to stop TXQ #%d\n",
2852                                        tx_params.tx_queue_id);
2853                                 return rc;
2854                         }
2855                 }
2856
2857                 /* Stop the Rx Queue*/
2858                 memset(&rx_params, 0, sizeof(rx_params));
2859                 rx_params.rss_id = i;
2860                 rx_params.rx_queue_id = i;
2861
2862                 rc = edev->ops->q_rx_stop(cdev, &rx_params);
2863                 if (rc) {
2864                         DP_ERR(edev, "Failed to stop RXQ #%d\n", i);
2865                         return rc;
2866                 }
2867         }
2868
2869         /* Stop the vport */
2870         rc = edev->ops->vport_stop(cdev, 0);
2871         if (rc)
2872                 DP_ERR(edev, "Failed to stop VPORT\n");
2873
2874         return rc;
2875 }
2876
2877 static int qede_start_queues(struct qede_dev *edev)
2878 {
2879         int rc, tc, i;
2880         int vlan_removal_en = 1;
2881         struct qed_dev *cdev = edev->cdev;
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         bool reset_rss_indir = false;
2886
2887         if (!edev->num_rss) {
2888                 DP_ERR(edev,
2889                        "Cannot update V-VPORT as active as there are no Rx queues\n");
2890                 return -EINVAL;
2891         }
2892
2893         start.gro_enable = !edev->gro_disable;
2894         start.mtu = edev->ndev->mtu;
2895         start.vport_id = 0;
2896         start.drop_ttl0 = true;
2897         start.remove_inner_vlan = vlan_removal_en;
2898
2899         rc = edev->ops->vport_start(cdev, &start);
2900
2901         if (rc) {
2902                 DP_ERR(edev, "Start V-PORT failed %d\n", rc);
2903                 return rc;
2904         }
2905
2906         DP_VERBOSE(edev, NETIF_MSG_IFUP,
2907                    "Start vport ramrod passed, vport_id = %d, MTU = %d, vlan_removal_en = %d\n",
2908                    start.vport_id, edev->ndev->mtu + 0xe, vlan_removal_en);
2909
2910         for_each_rss(i) {
2911                 struct qede_fastpath *fp = &edev->fp_array[i];
2912                 dma_addr_t phys_table = fp->rxq->rx_comp_ring.pbl.p_phys_table;
2913
2914                 memset(&q_params, 0, sizeof(q_params));
2915                 q_params.rss_id = i;
2916                 q_params.queue_id = i;
2917                 q_params.vport_id = 0;
2918                 q_params.sb = fp->sb_info->igu_sb_id;
2919                 q_params.sb_idx = RX_PI;
2920
2921                 rc = edev->ops->q_rx_start(cdev, &q_params,
2922                                            fp->rxq->rx_buf_size,
2923                                            fp->rxq->rx_bd_ring.p_phys_addr,
2924                                            phys_table,
2925                                            fp->rxq->rx_comp_ring.page_cnt,
2926                                            &fp->rxq->hw_rxq_prod_addr);
2927                 if (rc) {
2928                         DP_ERR(edev, "Start RXQ #%d failed %d\n", i, rc);
2929                         return rc;
2930                 }
2931
2932                 fp->rxq->hw_cons_ptr = &fp->sb_info->sb_virt->pi_array[RX_PI];
2933
2934                 qede_update_rx_prod(edev, fp->rxq);
2935
2936                 for (tc = 0; tc < edev->num_tc; tc++) {
2937                         struct qede_tx_queue *txq = &fp->txqs[tc];
2938                         int txq_index = tc * QEDE_RSS_CNT(edev) + i;
2939
2940                         memset(&q_params, 0, sizeof(q_params));
2941                         q_params.rss_id = i;
2942                         q_params.queue_id = txq_index;
2943                         q_params.vport_id = 0;
2944                         q_params.sb = fp->sb_info->igu_sb_id;
2945                         q_params.sb_idx = TX_PI(tc);
2946
2947                         rc = edev->ops->q_tx_start(cdev, &q_params,
2948                                                    txq->tx_pbl.pbl.p_phys_table,
2949                                                    txq->tx_pbl.page_cnt,
2950                                                    &txq->doorbell_addr);
2951                         if (rc) {
2952                                 DP_ERR(edev, "Start TXQ #%d failed %d\n",
2953                                        txq_index, rc);
2954                                 return rc;
2955                         }
2956
2957                         txq->hw_cons_ptr =
2958                                 &fp->sb_info->sb_virt->pi_array[TX_PI(tc)];
2959                         SET_FIELD(txq->tx_db.data.params,
2960                                   ETH_DB_DATA_DEST, DB_DEST_XCM);
2961                         SET_FIELD(txq->tx_db.data.params, ETH_DB_DATA_AGG_CMD,
2962                                   DB_AGG_CMD_SET);
2963                         SET_FIELD(txq->tx_db.data.params,
2964                                   ETH_DB_DATA_AGG_VAL_SEL,
2965                                   DQ_XCM_ETH_TX_BD_PROD_CMD);
2966
2967                         txq->tx_db.data.agg_flags = DQ_XCM_ETH_DQ_CF_CMD;
2968                 }
2969         }
2970
2971         /* Prepare and send the vport enable */
2972         memset(&vport_update_params, 0, sizeof(vport_update_params));
2973         vport_update_params.vport_id = start.vport_id;
2974         vport_update_params.update_vport_active_flg = 1;
2975         vport_update_params.vport_active_flg = 1;
2976
2977         /* Fill struct with RSS params */
2978         if (QEDE_RSS_CNT(edev) > 1) {
2979                 vport_update_params.update_rss_flg = 1;
2980
2981                 /* Need to validate current RSS config uses valid entries */
2982                 for (i = 0; i < QED_RSS_IND_TABLE_SIZE; i++) {
2983                         if (edev->rss_params.rss_ind_table[i] >=
2984                             edev->num_rss) {
2985                                 reset_rss_indir = true;
2986                                 break;
2987                         }
2988                 }
2989
2990                 if (!(edev->rss_params_inited & QEDE_RSS_INDIR_INITED) ||
2991                     reset_rss_indir) {
2992                         u16 val;
2993
2994                         for (i = 0; i < QED_RSS_IND_TABLE_SIZE; i++) {
2995                                 u16 indir_val;
2996
2997                                 val = QEDE_RSS_CNT(edev);
2998                                 indir_val = ethtool_rxfh_indir_default(i, val);
2999                                 edev->rss_params.rss_ind_table[i] = indir_val;
3000                         }
3001                         edev->rss_params_inited |= QEDE_RSS_INDIR_INITED;
3002                 }
3003
3004                 if (!(edev->rss_params_inited & QEDE_RSS_KEY_INITED)) {
3005                         netdev_rss_key_fill(edev->rss_params.rss_key,
3006                                             sizeof(edev->rss_params.rss_key));
3007                         edev->rss_params_inited |= QEDE_RSS_KEY_INITED;
3008                 }
3009
3010                 if (!(edev->rss_params_inited & QEDE_RSS_CAPS_INITED)) {
3011                         edev->rss_params.rss_caps = QED_RSS_IPV4 |
3012                                                     QED_RSS_IPV6 |
3013                                                     QED_RSS_IPV4_TCP |
3014                                                     QED_RSS_IPV6_TCP;
3015                         edev->rss_params_inited |= QEDE_RSS_CAPS_INITED;
3016                 }
3017
3018                 memcpy(&vport_update_params.rss_params, &edev->rss_params,
3019                        sizeof(vport_update_params.rss_params));
3020         } else {
3021                 memset(&vport_update_params.rss_params, 0,
3022                        sizeof(vport_update_params.rss_params));
3023         }
3024
3025         rc = edev->ops->vport_update(cdev, &vport_update_params);
3026         if (rc) {
3027                 DP_ERR(edev, "Update V-PORT failed %d\n", rc);
3028                 return rc;
3029         }
3030
3031         return 0;
3032 }
3033
3034 static int qede_set_mcast_rx_mac(struct qede_dev *edev,
3035                                  enum qed_filter_xcast_params_type opcode,
3036                                  unsigned char *mac, int num_macs)
3037 {
3038         struct qed_filter_params filter_cmd;
3039         int i;
3040
3041         memset(&filter_cmd, 0, sizeof(filter_cmd));
3042         filter_cmd.type = QED_FILTER_TYPE_MCAST;
3043         filter_cmd.filter.mcast.type = opcode;
3044         filter_cmd.filter.mcast.num = num_macs;
3045
3046         for (i = 0; i < num_macs; i++, mac += ETH_ALEN)
3047                 ether_addr_copy(filter_cmd.filter.mcast.mac[i], mac);
3048
3049         return edev->ops->filter_config(edev->cdev, &filter_cmd);
3050 }
3051
3052 enum qede_unload_mode {
3053         QEDE_UNLOAD_NORMAL,
3054 };
3055
3056 static void qede_unload(struct qede_dev *edev, enum qede_unload_mode mode)
3057 {
3058         struct qed_link_params link_params;
3059         int rc;
3060
3061         DP_INFO(edev, "Starting qede unload\n");
3062
3063         mutex_lock(&edev->qede_lock);
3064         edev->state = QEDE_STATE_CLOSED;
3065
3066         /* Close OS Tx */
3067         netif_tx_disable(edev->ndev);
3068         netif_carrier_off(edev->ndev);
3069
3070         /* Reset the link */
3071         memset(&link_params, 0, sizeof(link_params));
3072         link_params.link_up = false;
3073         edev->ops->common->set_link(edev->cdev, &link_params);
3074         rc = qede_stop_queues(edev);
3075         if (rc) {
3076                 qede_sync_free_irqs(edev);
3077                 goto out;
3078         }
3079
3080         DP_INFO(edev, "Stopped Queues\n");
3081
3082         qede_vlan_mark_nonconfigured(edev);
3083         edev->ops->fastpath_stop(edev->cdev);
3084
3085         /* Release the interrupts */
3086         qede_sync_free_irqs(edev);
3087         edev->ops->common->set_fp_int(edev->cdev, 0);
3088
3089         qede_napi_disable_remove(edev);
3090
3091         qede_free_mem_load(edev);
3092         qede_free_fp_array(edev);
3093
3094 out:
3095         mutex_unlock(&edev->qede_lock);
3096         DP_INFO(edev, "Ending qede unload\n");
3097 }
3098
3099 enum qede_load_mode {
3100         QEDE_LOAD_NORMAL,
3101 };
3102
3103 static int qede_load(struct qede_dev *edev, enum qede_load_mode mode)
3104 {
3105         struct qed_link_params link_params;
3106         struct qed_link_output link_output;
3107         int rc;
3108
3109         DP_INFO(edev, "Starting qede load\n");
3110
3111         rc = qede_set_num_queues(edev);
3112         if (rc)
3113                 goto err0;
3114
3115         rc = qede_alloc_fp_array(edev);
3116         if (rc)
3117                 goto err0;
3118
3119         qede_init_fp(edev);
3120
3121         rc = qede_alloc_mem_load(edev);
3122         if (rc)
3123                 goto err1;
3124         DP_INFO(edev, "Allocated %d RSS queues on %d TC/s\n",
3125                 QEDE_RSS_CNT(edev), edev->num_tc);
3126
3127         rc = qede_set_real_num_queues(edev);
3128         if (rc)
3129                 goto err2;
3130
3131         qede_napi_add_enable(edev);
3132         DP_INFO(edev, "Napi added and enabled\n");
3133
3134         rc = qede_setup_irqs(edev);
3135         if (rc)
3136                 goto err3;
3137         DP_INFO(edev, "Setup IRQs succeeded\n");
3138
3139         rc = qede_start_queues(edev);
3140         if (rc)
3141                 goto err4;
3142         DP_INFO(edev, "Start VPORT, RXQ and TXQ succeeded\n");
3143
3144         /* Add primary mac and set Rx filters */
3145         ether_addr_copy(edev->primary_mac, edev->ndev->dev_addr);
3146
3147         mutex_lock(&edev->qede_lock);
3148         edev->state = QEDE_STATE_OPEN;
3149         mutex_unlock(&edev->qede_lock);
3150
3151         /* Program un-configured VLANs */
3152         qede_configure_vlan_filters(edev);
3153
3154         /* Ask for link-up using current configuration */
3155         memset(&link_params, 0, sizeof(link_params));
3156         link_params.link_up = true;
3157         edev->ops->common->set_link(edev->cdev, &link_params);
3158
3159         /* Query whether link is already-up */
3160         memset(&link_output, 0, sizeof(link_output));
3161         edev->ops->common->get_link(edev->cdev, &link_output);
3162         qede_link_update(edev, &link_output);
3163
3164         DP_INFO(edev, "Ending successfully qede load\n");
3165
3166         return 0;
3167
3168 err4:
3169         qede_sync_free_irqs(edev);
3170         memset(&edev->int_info.msix_cnt, 0, sizeof(struct qed_int_info));
3171 err3:
3172         qede_napi_disable_remove(edev);
3173 err2:
3174         qede_free_mem_load(edev);
3175 err1:
3176         edev->ops->common->set_fp_int(edev->cdev, 0);
3177         qede_free_fp_array(edev);
3178         edev->num_rss = 0;
3179 err0:
3180         return rc;
3181 }
3182
3183 void qede_reload(struct qede_dev *edev,
3184                  void (*func)(struct qede_dev *, union qede_reload_args *),
3185                  union qede_reload_args *args)
3186 {
3187         qede_unload(edev, QEDE_UNLOAD_NORMAL);
3188         /* Call function handler to update parameters
3189          * needed for function load.
3190          */
3191         if (func)
3192                 func(edev, args);
3193
3194         qede_load(edev, QEDE_LOAD_NORMAL);
3195
3196         mutex_lock(&edev->qede_lock);
3197         qede_config_rx_mode(edev->ndev);
3198         mutex_unlock(&edev->qede_lock);
3199 }
3200
3201 /* called with rtnl_lock */
3202 static int qede_open(struct net_device *ndev)
3203 {
3204         struct qede_dev *edev = netdev_priv(ndev);
3205         int rc;
3206
3207         netif_carrier_off(ndev);
3208
3209         edev->ops->common->set_power_state(edev->cdev, PCI_D0);
3210
3211         rc = qede_load(edev, QEDE_LOAD_NORMAL);
3212
3213         if (rc)
3214                 return rc;
3215
3216 #ifdef CONFIG_QEDE_VXLAN
3217         vxlan_get_rx_port(ndev);
3218 #endif
3219         return 0;
3220 }
3221
3222 static int qede_close(struct net_device *ndev)
3223 {
3224         struct qede_dev *edev = netdev_priv(ndev);
3225
3226         qede_unload(edev, QEDE_UNLOAD_NORMAL);
3227
3228         return 0;
3229 }
3230
3231 static void qede_link_update(void *dev, struct qed_link_output *link)
3232 {
3233         struct qede_dev *edev = dev;
3234
3235         if (!netif_running(edev->ndev)) {
3236                 DP_VERBOSE(edev, NETIF_MSG_LINK, "Interface is not running\n");
3237                 return;
3238         }
3239
3240         if (link->link_up) {
3241                 if (!netif_carrier_ok(edev->ndev)) {
3242                         DP_NOTICE(edev, "Link is up\n");
3243                         netif_tx_start_all_queues(edev->ndev);
3244                         netif_carrier_on(edev->ndev);
3245                 }
3246         } else {
3247                 if (netif_carrier_ok(edev->ndev)) {
3248                         DP_NOTICE(edev, "Link is down\n");
3249                         netif_tx_disable(edev->ndev);
3250                         netif_carrier_off(edev->ndev);
3251                 }
3252         }
3253 }
3254
3255 static int qede_set_mac_addr(struct net_device *ndev, void *p)
3256 {
3257         struct qede_dev *edev = netdev_priv(ndev);
3258         struct sockaddr *addr = p;
3259         int rc;
3260
3261         ASSERT_RTNL(); /* @@@TBD To be removed */
3262
3263         DP_INFO(edev, "Set_mac_addr called\n");
3264
3265         if (!is_valid_ether_addr(addr->sa_data)) {
3266                 DP_NOTICE(edev, "The MAC address is not valid\n");
3267                 return -EFAULT;
3268         }
3269
3270         ether_addr_copy(ndev->dev_addr, addr->sa_data);
3271
3272         if (!netif_running(ndev))  {
3273                 DP_NOTICE(edev, "The device is currently down\n");
3274                 return 0;
3275         }
3276
3277         /* Remove the previous primary mac */
3278         rc = qede_set_ucast_rx_mac(edev, QED_FILTER_XCAST_TYPE_DEL,
3279                                    edev->primary_mac);
3280         if (rc)
3281                 return rc;
3282
3283         /* Add MAC filter according to the new unicast HW MAC address */
3284         ether_addr_copy(edev->primary_mac, ndev->dev_addr);
3285         return qede_set_ucast_rx_mac(edev, QED_FILTER_XCAST_TYPE_ADD,
3286                                       edev->primary_mac);
3287 }
3288
3289 static int
3290 qede_configure_mcast_filtering(struct net_device *ndev,
3291                                enum qed_filter_rx_mode_type *accept_flags)
3292 {
3293         struct qede_dev *edev = netdev_priv(ndev);
3294         unsigned char *mc_macs, *temp;
3295         struct netdev_hw_addr *ha;
3296         int rc = 0, mc_count;
3297         size_t size;
3298
3299         size = 64 * ETH_ALEN;
3300
3301         mc_macs = kzalloc(size, GFP_KERNEL);
3302         if (!mc_macs) {
3303                 DP_NOTICE(edev,
3304                           "Failed to allocate memory for multicast MACs\n");
3305                 rc = -ENOMEM;
3306                 goto exit;
3307         }
3308
3309         temp = mc_macs;
3310
3311         /* Remove all previously configured MAC filters */
3312         rc = qede_set_mcast_rx_mac(edev, QED_FILTER_XCAST_TYPE_DEL,
3313                                    mc_macs, 1);
3314         if (rc)
3315                 goto exit;
3316
3317         netif_addr_lock_bh(ndev);
3318
3319         mc_count = netdev_mc_count(ndev);
3320         if (mc_count < 64) {
3321                 netdev_for_each_mc_addr(ha, ndev) {
3322                         ether_addr_copy(temp, ha->addr);
3323                         temp += ETH_ALEN;
3324                 }
3325         }
3326
3327         netif_addr_unlock_bh(ndev);
3328
3329         /* Check for all multicast @@@TBD resource allocation */
3330         if ((ndev->flags & IFF_ALLMULTI) ||
3331             (mc_count > 64)) {
3332                 if (*accept_flags == QED_FILTER_RX_MODE_TYPE_REGULAR)
3333                         *accept_flags = QED_FILTER_RX_MODE_TYPE_MULTI_PROMISC;
3334         } else {
3335                 /* Add all multicast MAC filters */
3336                 rc = qede_set_mcast_rx_mac(edev, QED_FILTER_XCAST_TYPE_ADD,
3337                                            mc_macs, mc_count);
3338         }
3339
3340 exit:
3341         kfree(mc_macs);
3342         return rc;
3343 }
3344
3345 static void qede_set_rx_mode(struct net_device *ndev)
3346 {
3347         struct qede_dev *edev = netdev_priv(ndev);
3348
3349         DP_INFO(edev, "qede_set_rx_mode called\n");
3350
3351         if (edev->state != QEDE_STATE_OPEN) {
3352                 DP_INFO(edev,
3353                         "qede_set_rx_mode called while interface is down\n");
3354         } else {
3355                 set_bit(QEDE_SP_RX_MODE, &edev->sp_flags);
3356                 schedule_delayed_work(&edev->sp_task, 0);
3357         }
3358 }
3359
3360 /* Must be called with qede_lock held */
3361 static void qede_config_rx_mode(struct net_device *ndev)
3362 {
3363         enum qed_filter_rx_mode_type accept_flags = QED_FILTER_TYPE_UCAST;
3364         struct qede_dev *edev = netdev_priv(ndev);
3365         struct qed_filter_params rx_mode;
3366         unsigned char *uc_macs, *temp;
3367         struct netdev_hw_addr *ha;
3368         int rc, uc_count;
3369         size_t size;
3370
3371         netif_addr_lock_bh(ndev);
3372
3373         uc_count = netdev_uc_count(ndev);
3374         size = uc_count * ETH_ALEN;
3375
3376         uc_macs = kzalloc(size, GFP_ATOMIC);
3377         if (!uc_macs) {
3378                 DP_NOTICE(edev, "Failed to allocate memory for unicast MACs\n");
3379                 netif_addr_unlock_bh(ndev);
3380                 return;
3381         }
3382
3383         temp = uc_macs;
3384         netdev_for_each_uc_addr(ha, ndev) {
3385                 ether_addr_copy(temp, ha->addr);
3386                 temp += ETH_ALEN;
3387         }
3388
3389         netif_addr_unlock_bh(ndev);
3390
3391         /* Configure the struct for the Rx mode */
3392         memset(&rx_mode, 0, sizeof(struct qed_filter_params));
3393         rx_mode.type = QED_FILTER_TYPE_RX_MODE;
3394
3395         /* Remove all previous unicast secondary macs and multicast macs
3396          * (configrue / leave the primary mac)
3397          */
3398         rc = qede_set_ucast_rx_mac(edev, QED_FILTER_XCAST_TYPE_REPLACE,
3399                                    edev->primary_mac);
3400         if (rc)
3401                 goto out;
3402
3403         /* Check for promiscuous */
3404         if ((ndev->flags & IFF_PROMISC) ||
3405             (uc_count > 15)) { /* @@@TBD resource allocation - 1 */
3406                 accept_flags = QED_FILTER_RX_MODE_TYPE_PROMISC;
3407         } else {
3408                 /* Add MAC filters according to the unicast secondary macs */
3409                 int i;
3410
3411                 temp = uc_macs;
3412                 for (i = 0; i < uc_count; i++) {
3413                         rc = qede_set_ucast_rx_mac(edev,
3414                                                    QED_FILTER_XCAST_TYPE_ADD,
3415                                                    temp);
3416                         if (rc)
3417                                 goto out;
3418
3419                         temp += ETH_ALEN;
3420                 }
3421
3422                 rc = qede_configure_mcast_filtering(ndev, &accept_flags);
3423                 if (rc)
3424                         goto out;
3425         }
3426
3427         /* take care of VLAN mode */
3428         if (ndev->flags & IFF_PROMISC) {
3429                 qede_config_accept_any_vlan(edev, true);
3430         } else if (!edev->non_configured_vlans) {
3431                 /* It's possible that accept_any_vlan mode is set due to a
3432                  * previous setting of IFF_PROMISC. If vlan credits are
3433                  * sufficient, disable accept_any_vlan.
3434                  */
3435                 qede_config_accept_any_vlan(edev, false);
3436         }
3437
3438         rx_mode.filter.accept_flags = accept_flags;
3439         edev->ops->filter_config(edev->cdev, &rx_mode);
3440 out:
3441         kfree(uc_macs);
3442 }