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

   1 /* QLogic qede NIC Driver
   2 * Copyright (c) 2015 QLogic Corporation
   3 *
   4 * This software is available under the terms of the GNU General Public License
   5 * (GPL) Version 2, available from the file COPYING in the main directory of
   6 * this source tree.
   7 */
   8
   9 #include <linux/module.h>
  10 #include <linux/pci.h>
  11 #include <linux/version.h>
  12 #include <linux/device.h>
  13 #include <linux/netdevice.h>
  14 #include <linux/etherdevice.h>
  15 #include <linux/skbuff.h>
  16 #include <linux/errno.h>
  17 #include <linux/list.h>
  18 #include <linux/string.h>
  19 #include <linux/dma-mapping.h>
  20 #include <linux/interrupt.h>
  21 #include <asm/byteorder.h>
  22 #include <asm/param.h>
  23 #include <linux/io.h>
  24 #include <linux/netdev_features.h>
  25 #include <linux/udp.h>
  26 #include <linux/tcp.h>
  27 #include <net/udp_tunnel.h>
  28 #include <linux/ip.h>
  29 #include <net/ipv6.h>
  30 #include <net/tcp.h>
  31 #include <linux/if_ether.h>
  32 #include <linux/if_vlan.h>
  33 #include <linux/pkt_sched.h>
  34 #include <linux/ethtool.h>
  35 #include <linux/in.h>
  36 #include <linux/random.h>
  37 #include <net/ip6_checksum.h>
  38 #include <linux/bitops.h>
  39 #include <linux/qed/qede_roce.h>
  40 #include "qede.h"
  41
  42 static char version[] =
  43         "QLogic FastLinQ 4xxxx Ethernet Driver qede " DRV_MODULE_VERSION "\n";
  44
  45 MODULE_DESCRIPTION("QLogic FastLinQ 4xxxx Ethernet Driver");
  46 MODULE_LICENSE("GPL");
  47 MODULE_VERSION(DRV_MODULE_VERSION);
  48
  49 static uint debug;
  50 module_param(debug, uint, 0);
  51 MODULE_PARM_DESC(debug, " Default debug msglevel");
  52
  53 static const struct qed_eth_ops *qed_ops;
  54
  55 #define CHIP_NUM_57980S_40              0x1634
  56 #define CHIP_NUM_57980S_10              0x1666
  57 #define CHIP_NUM_57980S_MF              0x1636
  58 #define CHIP_NUM_57980S_100             0x1644
  59 #define CHIP_NUM_57980S_50              0x1654
  60 #define CHIP_NUM_57980S_25              0x1656
  61 #define CHIP_NUM_57980S_IOV             0x1664
  62
  63 #ifndef PCI_DEVICE_ID_NX2_57980E
  64 #define PCI_DEVICE_ID_57980S_40         CHIP_NUM_57980S_40
  65 #define PCI_DEVICE_ID_57980S_10         CHIP_NUM_57980S_10
  66 #define PCI_DEVICE_ID_57980S_MF         CHIP_NUM_57980S_MF
  67 #define PCI_DEVICE_ID_57980S_100        CHIP_NUM_57980S_100
  68 #define PCI_DEVICE_ID_57980S_50         CHIP_NUM_57980S_50
  69 #define PCI_DEVICE_ID_57980S_25         CHIP_NUM_57980S_25
  70 #define PCI_DEVICE_ID_57980S_IOV        CHIP_NUM_57980S_IOV
  71 #endif
  72
  73 enum qede_pci_private {
  74         QEDE_PRIVATE_PF,
  75         QEDE_PRIVATE_VF
  76 };
  77
  78 static const struct pci_device_id qede_pci_tbl[] = {
  79         {PCI_VDEVICE(QLOGIC, PCI_DEVICE_ID_57980S_40), QEDE_PRIVATE_PF},
  80         {PCI_VDEVICE(QLOGIC, PCI_DEVICE_ID_57980S_10), QEDE_PRIVATE_PF},
  81         {PCI_VDEVICE(QLOGIC, PCI_DEVICE_ID_57980S_MF), QEDE_PRIVATE_PF},
  82         {PCI_VDEVICE(QLOGIC, PCI_DEVICE_ID_57980S_100), QEDE_PRIVATE_PF},
  83         {PCI_VDEVICE(QLOGIC, PCI_DEVICE_ID_57980S_50), QEDE_PRIVATE_PF},
  84         {PCI_VDEVICE(QLOGIC, PCI_DEVICE_ID_57980S_25), QEDE_PRIVATE_PF},
  85 #ifdef CONFIG_QED_SRIOV
  86         {PCI_VDEVICE(QLOGIC, PCI_DEVICE_ID_57980S_IOV), QEDE_PRIVATE_VF},
  87 #endif
  88         { 0 }
  89 };
  90
  91 MODULE_DEVICE_TABLE(pci, qede_pci_tbl);
  92
  93 static int qede_probe(struct pci_dev *pdev, const struct pci_device_id *id);
  94
  95 #define TX_TIMEOUT              (5 * HZ)
  96
  97 static void qede_remove(struct pci_dev *pdev);
  98 static int qede_alloc_rx_buffer(struct qede_dev *edev,
  99                                 struct qede_rx_queue *rxq);
 100 static void qede_link_update(void *dev, struct qed_link_output *link);
 101
 102 #ifdef CONFIG_QED_SRIOV
 103 static int qede_set_vf_vlan(struct net_device *ndev, int vf, u16 vlan, u8 qos,
 104                             __be16 vlan_proto)
 105 {
 106         struct qede_dev *edev = netdev_priv(ndev);
 107
 108         if (vlan > 4095) {
 109                 DP_NOTICE(edev, "Illegal vlan value %d\n", vlan);
 110                 return -EINVAL;
 111         }
 112
 113         if (vlan_proto != htons(ETH_P_8021Q))
 114                 return -EPROTONOSUPPORT;
 115
 116         DP_VERBOSE(edev, QED_MSG_IOV, "Setting Vlan 0x%04x to VF [%d]\n",
 117                    vlan, vf);
 118
 119         return edev->ops->iov->set_vlan(edev->cdev, vlan, vf);
 120 }
 121
 122 static int qede_set_vf_mac(struct net_device *ndev, int vfidx, u8 *mac)
 123 {
 124         struct qede_dev *edev = netdev_priv(ndev);
 125
 126         DP_VERBOSE(edev, QED_MSG_IOV,
 127                    "Setting MAC %02x:%02x:%02x:%02x:%02x:%02x to VF [%d]\n",
 128                    mac[0], mac[1], mac[2], mac[3], mac[4], mac[5], vfidx);
 129
 130         if (!is_valid_ether_addr(mac)) {
 131                 DP_VERBOSE(edev, QED_MSG_IOV, "MAC address isn't valid\n");
 132                 return -EINVAL;
 133         }
 134
 135         return edev->ops->iov->set_mac(edev->cdev, mac, vfidx);
 136 }
 137
 138 static int qede_sriov_configure(struct pci_dev *pdev, int num_vfs_param)
 139 {
 140         struct qede_dev *edev = netdev_priv(pci_get_drvdata(pdev));
 141         struct qed_dev_info *qed_info = &edev->dev_info.common;
 142         int rc;
 143
 144         DP_VERBOSE(edev, QED_MSG_IOV, "Requested %d VFs\n", num_vfs_param);
 145
 146         rc = edev->ops->iov->configure(edev->cdev, num_vfs_param);
 147
 148         /* Enable/Disable Tx switching for PF */
 149         if ((rc == num_vfs_param) && netif_running(edev->ndev) &&
 150             qed_info->mf_mode != QED_MF_NPAR && qed_info->tx_switching) {
 151                 struct qed_update_vport_params params;
 152
 153                 memset(&params, 0, sizeof(params));
 154                 params.vport_id = 0;
 155                 params.update_tx_switching_flg = 1;
 156                 params.tx_switching_flg = num_vfs_param ? 1 : 0;
 157                 edev->ops->vport_update(edev->cdev, &params);
 158         }
 159
 160         return rc;
 161 }
 162 #endif
 163
 164 static struct pci_driver qede_pci_driver = {
 165         .name = "qede",
 166         .id_table = qede_pci_tbl,
 167         .probe = qede_probe,
 168         .remove = qede_remove,
 169 #ifdef CONFIG_QED_SRIOV
 170         .sriov_configure = qede_sriov_configure,
 171 #endif
 172 };
 173
 174 static void qede_force_mac(void *dev, u8 *mac)
 175 {
 176         struct qede_dev *edev = dev;
 177
 178         ether_addr_copy(edev->ndev->dev_addr, mac);
 179         ether_addr_copy(edev->primary_mac, mac);
 180 }
 181
 182 static struct qed_eth_cb_ops qede_ll_ops = {
 183         {
 184                 .link_update = qede_link_update,
 185         },
 186         .force_mac = qede_force_mac,
 187 };
 188
 189 static int qede_netdev_event(struct notifier_block *this, unsigned long event,
 190                              void *ptr)
 191 {
 192         struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
 193         struct ethtool_drvinfo drvinfo;
 194         struct qede_dev *edev;
 195
 196         if (event != NETDEV_CHANGENAME && event != NETDEV_CHANGEADDR)
 197                 goto done;
 198
 199         /* Check whether this is a qede device */
 200         if (!ndev || !ndev->ethtool_ops || !ndev->ethtool_ops->get_drvinfo)
 201                 goto done;
 202
 203         memset(&drvinfo, 0, sizeof(drvinfo));
 204         ndev->ethtool_ops->get_drvinfo(ndev, &drvinfo);
 205         if (strcmp(drvinfo.driver, "qede"))
 206                 goto done;
 207         edev = netdev_priv(ndev);
 208
 209         switch (event) {
 210         case NETDEV_CHANGENAME:
 211                 /* Notify qed of the name change */
 212                 if (!edev->ops || !edev->ops->common)
 213                         goto done;
 214                 edev->ops->common->set_id(edev->cdev, edev->ndev->name, "qede");
 215                 break;
 216         case NETDEV_CHANGEADDR:
 217                 edev = netdev_priv(ndev);
 218                 qede_roce_event_changeaddr(edev);
 219                 break;
 220         }
 221
 222 done:
 223         return NOTIFY_DONE;
 224 }
 225
 226 static struct notifier_block qede_netdev_notifier = {
 227         .notifier_call = qede_netdev_event,
 228 };
 229
 230 static
 231 int __init qede_init(void)
 232 {
 233         int ret;
 234
 235         pr_info("qede_init: %s\n", version);
 236
 237         qed_ops = qed_get_eth_ops();
 238         if (!qed_ops) {
 239                 pr_notice("Failed to get qed ethtool operations\n");
 240                 return -EINVAL;
 241         }
 242
 243         /* Must register notifier before pci ops, since we might miss
 244          * interface rename after pci probe and netdev registeration.
 245          */
 246         ret = register_netdevice_notifier(&qede_netdev_notifier);
 247         if (ret) {
 248                 pr_notice("Failed to register netdevice_notifier\n");
 249                 qed_put_eth_ops();
 250                 return -EINVAL;
 251         }
 252
 253         ret = pci_register_driver(&qede_pci_driver);
 254         if (ret) {
 255                 pr_notice("Failed to register driver\n");
 256                 unregister_netdevice_notifier(&qede_netdev_notifier);
 257                 qed_put_eth_ops();
 258                 return -EINVAL;
 259         }
 260
 261         return 0;
 262 }
 263
 264 static void __exit qede_cleanup(void)
 265 {
 266         if (debug & QED_LOG_INFO_MASK)
 267                 pr_info("qede_cleanup called\n");
 268
 269         unregister_netdevice_notifier(&qede_netdev_notifier);
 270         pci_unregister_driver(&qede_pci_driver);
 271         qed_put_eth_ops();
 272 }
 273
 274 module_init(qede_init);
 275 module_exit(qede_cleanup);
 276
 277 /* -------------------------------------------------------------------------
 278  * START OF FAST-PATH
 279  * -------------------------------------------------------------------------
 280  */
 281
 282 /* Unmap the data and free skb */
 283 static int qede_free_tx_pkt(struct qede_dev *edev,
 284                             struct qede_tx_queue *txq, int *len)
 285 {
 286         u16 idx = txq->sw_tx_cons & NUM_TX_BDS_MAX;
 287         struct sk_buff *skb = txq->sw_tx_ring[idx].skb;
 288         struct eth_tx_1st_bd *first_bd;
 289         struct eth_tx_bd *tx_data_bd;
 290         int bds_consumed = 0;
 291         int nbds;
 292         bool data_split = txq->sw_tx_ring[idx].flags & QEDE_TSO_SPLIT_BD;
 293         int i, split_bd_len = 0;
 294
 295         if (unlikely(!skb)) {
 296                 DP_ERR(edev,
 297                        "skb is null for txq idx=%d txq->sw_tx_cons=%d txq->sw_tx_prod=%d\n",
 298                        idx, txq->sw_tx_cons, txq->sw_tx_prod);
 299                 return -1;
 300         }
 301
 302         *len = skb->len;
 303
 304         first_bd = (struct eth_tx_1st_bd *)qed_chain_consume(&txq->tx_pbl);
 305
 306         bds_consumed++;
 307
 308         nbds = first_bd->data.nbds;
 309
 310         if (data_split) {
 311                 struct eth_tx_bd *split = (struct eth_tx_bd *)
 312                         qed_chain_consume(&txq->tx_pbl);
 313                 split_bd_len = BD_UNMAP_LEN(split);
 314                 bds_consumed++;
 315         }
 316         dma_unmap_single(&edev->pdev->dev, BD_UNMAP_ADDR(first_bd),
 317                          BD_UNMAP_LEN(first_bd) + split_bd_len, DMA_TO_DEVICE);
 318
 319         /* Unmap the data of the skb frags */
 320         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++, bds_consumed++) {
 321                 tx_data_bd = (struct eth_tx_bd *)
 322                         qed_chain_consume(&txq->tx_pbl);
 323                 dma_unmap_page(&edev->pdev->dev, BD_UNMAP_ADDR(tx_data_bd),
 324                                BD_UNMAP_LEN(tx_data_bd), DMA_TO_DEVICE);
 325         }
 326
 327         while (bds_consumed++ < nbds)
 328                 qed_chain_consume(&txq->tx_pbl);
 329
 330         /* Free skb */
 331         dev_kfree_skb_any(skb);
 332         txq->sw_tx_ring[idx].skb = NULL;
 333         txq->sw_tx_ring[idx].flags = 0;
 334
 335         return 0;
 336 }
 337
 338 /* Unmap the data and free skb when mapping failed during start_xmit */
 339 static void qede_free_failed_tx_pkt(struct qede_dev *edev,
 340                                     struct qede_tx_queue *txq,
 341                                     struct eth_tx_1st_bd *first_bd,
 342                                     int nbd, bool data_split)
 343 {
 344         u16 idx = txq->sw_tx_prod & NUM_TX_BDS_MAX;
 345         struct sk_buff *skb = txq->sw_tx_ring[idx].skb;
 346         struct eth_tx_bd *tx_data_bd;
 347         int i, split_bd_len = 0;
 348
 349         /* Return prod to its position before this skb was handled */
 350         qed_chain_set_prod(&txq->tx_pbl,
 351                            le16_to_cpu(txq->tx_db.data.bd_prod), first_bd);
 352
 353         first_bd = (struct eth_tx_1st_bd *)qed_chain_produce(&txq->tx_pbl);
 354
 355         if (data_split) {
 356                 struct eth_tx_bd *split = (struct eth_tx_bd *)
 357                                           qed_chain_produce(&txq->tx_pbl);
 358                 split_bd_len = BD_UNMAP_LEN(split);
 359                 nbd--;
 360         }
 361
 362         dma_unmap_single(&edev->pdev->dev, BD_UNMAP_ADDR(first_bd),
 363                          BD_UNMAP_LEN(first_bd) + split_bd_len, DMA_TO_DEVICE);
 364
 365         /* Unmap the data of the skb frags */
 366         for (i = 0; i < nbd; i++) {
 367                 tx_data_bd = (struct eth_tx_bd *)
 368                         qed_chain_produce(&txq->tx_pbl);
 369                 if (tx_data_bd->nbytes)
 370                         dma_unmap_page(&edev->pdev->dev,
 371                                        BD_UNMAP_ADDR(tx_data_bd),
 372                                        BD_UNMAP_LEN(tx_data_bd), DMA_TO_DEVICE);
 373         }
 374
 375         /* Return again prod to its position before this skb was handled */
 376         qed_chain_set_prod(&txq->tx_pbl,
 377                            le16_to_cpu(txq->tx_db.data.bd_prod), first_bd);
 378
 379         /* Free skb */
 380         dev_kfree_skb_any(skb);
 381         txq->sw_tx_ring[idx].skb = NULL;
 382         txq->sw_tx_ring[idx].flags = 0;
 383 }
 384
 385 static u32 qede_xmit_type(struct qede_dev *edev,
 386                           struct sk_buff *skb, int *ipv6_ext)
 387 {
 388         u32 rc = XMIT_L4_CSUM;
 389         __be16 l3_proto;
 390
 391         if (skb->ip_summed != CHECKSUM_PARTIAL)
 392                 return XMIT_PLAIN;
 393
 394         l3_proto = vlan_get_protocol(skb);
 395         if (l3_proto == htons(ETH_P_IPV6) &&
 396             (ipv6_hdr(skb)->nexthdr == NEXTHDR_IPV6))
 397                 *ipv6_ext = 1;
 398
 399         if (skb->encapsulation)
 400                 rc |= XMIT_ENC;
 401
 402         if (skb_is_gso(skb))
 403                 rc |= XMIT_LSO;
 404
 405         return rc;
 406 }
 407
 408 static void qede_set_params_for_ipv6_ext(struct sk_buff *skb,
 409                                          struct eth_tx_2nd_bd *second_bd,
 410                                          struct eth_tx_3rd_bd *third_bd)
 411 {
 412         u8 l4_proto;
 413         u16 bd2_bits1 = 0, bd2_bits2 = 0;
 414
 415         bd2_bits1 |= (1 << ETH_TX_DATA_2ND_BD_IPV6_EXT_SHIFT);
 416
 417         bd2_bits2 |= ((((u8 *)skb_transport_header(skb) - skb->data) >> 1) &
 418                      ETH_TX_DATA_2ND_BD_L4_HDR_START_OFFSET_W_MASK)
 419                     << ETH_TX_DATA_2ND_BD_L4_HDR_START_OFFSET_W_SHIFT;
 420
 421         bd2_bits1 |= (ETH_L4_PSEUDO_CSUM_CORRECT_LENGTH <<
 422                       ETH_TX_DATA_2ND_BD_L4_PSEUDO_CSUM_MODE_SHIFT);
 423
 424         if (vlan_get_protocol(skb) == htons(ETH_P_IPV6))
 425                 l4_proto = ipv6_hdr(skb)->nexthdr;
 426         else
 427                 l4_proto = ip_hdr(skb)->protocol;
 428
 429         if (l4_proto == IPPROTO_UDP)
 430                 bd2_bits1 |= 1 << ETH_TX_DATA_2ND_BD_L4_UDP_SHIFT;
 431
 432         if (third_bd)
 433                 third_bd->data.bitfields |=
 434                         cpu_to_le16(((tcp_hdrlen(skb) / 4) &
 435                                 ETH_TX_DATA_3RD_BD_TCP_HDR_LEN_DW_MASK) <<
 436                                 ETH_TX_DATA_3RD_BD_TCP_HDR_LEN_DW_SHIFT);
 437
 438         second_bd->data.bitfields1 = cpu_to_le16(bd2_bits1);
 439         second_bd->data.bitfields2 = cpu_to_le16(bd2_bits2);
 440 }
 441
 442 static int map_frag_to_bd(struct qede_dev *edev,
 443                           skb_frag_t *frag, struct eth_tx_bd *bd)
 444 {
 445         dma_addr_t mapping;
 446
 447         /* Map skb non-linear frag data for DMA */
 448         mapping = skb_frag_dma_map(&edev->pdev->dev, frag, 0,
 449                                    skb_frag_size(frag), DMA_TO_DEVICE);
 450         if (unlikely(dma_mapping_error(&edev->pdev->dev, mapping))) {
 451                 DP_NOTICE(edev, "Unable to map frag - dropping packet\n");
 452                 return -ENOMEM;
 453         }
 454
 455         /* Setup the data pointer of the frag data */
 456         BD_SET_UNMAP_ADDR_LEN(bd, mapping, skb_frag_size(frag));
 457
 458         return 0;
 459 }
 460
 461 static u16 qede_get_skb_hlen(struct sk_buff *skb, bool is_encap_pkt)
 462 {
 463         if (is_encap_pkt)
 464                 return (skb_inner_transport_header(skb) +
 465                         inner_tcp_hdrlen(skb) - skb->data);
 466         else
 467                 return (skb_transport_header(skb) +
 468                         tcp_hdrlen(skb) - skb->data);
 469 }
 470
 471 /* +2 for 1st BD for headers and 2nd BD for headlen (if required) */
 472 #if ((MAX_SKB_FRAGS + 2) > ETH_TX_MAX_BDS_PER_NON_LSO_PACKET)
 473 static bool qede_pkt_req_lin(struct qede_dev *edev, struct sk_buff *skb,
 474                              u8 xmit_type)
 475 {
 476         int allowed_frags = ETH_TX_MAX_BDS_PER_NON_LSO_PACKET - 1;
 477
 478         if (xmit_type & XMIT_LSO) {
 479                 int hlen;
 480
 481                 hlen = qede_get_skb_hlen(skb, xmit_type & XMIT_ENC);
 482
 483                 /* linear payload would require its own BD */
 484                 if (skb_headlen(skb) > hlen)
 485                         allowed_frags--;
 486         }
 487
 488         return (skb_shinfo(skb)->nr_frags > allowed_frags);
 489 }
 490 #endif
 491
 492 static inline void qede_update_tx_producer(struct qede_tx_queue *txq)
 493 {
 494         /* wmb makes sure that the BDs data is updated before updating the
 495          * producer, otherwise FW may read old data from the BDs.
 496          */
 497         wmb();
 498         barrier();
 499         writel(txq->tx_db.raw, txq->doorbell_addr);
 500
 501         /* mmiowb is needed to synchronize doorbell writes from more than one
 502          * processor. It guarantees that the write arrives to the device before
 503          * the queue lock is released and another start_xmit is called (possibly
 504          * on another CPU). Without this barrier, the next doorbell can bypass
 505          * this doorbell. This is applicable to IA64/Altix systems.
 506          */
 507         mmiowb();
 508 }
 509
 510 /* Main transmit function */
 511 static netdev_tx_t qede_start_xmit(struct sk_buff *skb,
 512                                    struct net_device *ndev)
 513 {
 514         struct qede_dev *edev = netdev_priv(ndev);
 515         struct netdev_queue *netdev_txq;
 516         struct qede_tx_queue *txq;
 517         struct eth_tx_1st_bd *first_bd;
 518         struct eth_tx_2nd_bd *second_bd = NULL;
 519         struct eth_tx_3rd_bd *third_bd = NULL;
 520         struct eth_tx_bd *tx_data_bd = NULL;
 521         u16 txq_index;
 522         u8 nbd = 0;
 523         dma_addr_t mapping;
 524         int rc, frag_idx = 0, ipv6_ext = 0;
 525         u8 xmit_type;
 526         u16 idx;
 527         u16 hlen;
 528         bool data_split = false;
 529
 530         /* Get tx-queue context and netdev index */
 531         txq_index = skb_get_queue_mapping(skb);
 532         WARN_ON(txq_index >= QEDE_TSS_COUNT(edev));
 533         txq = QEDE_TX_QUEUE(edev, txq_index);
 534         netdev_txq = netdev_get_tx_queue(ndev, txq_index);
 535
 536         WARN_ON(qed_chain_get_elem_left(&txq->tx_pbl) < (MAX_SKB_FRAGS + 1));
 537
 538         xmit_type = qede_xmit_type(edev, skb, &ipv6_ext);
 539
 540 #if ((MAX_SKB_FRAGS + 2) > ETH_TX_MAX_BDS_PER_NON_LSO_PACKET)
 541         if (qede_pkt_req_lin(edev, skb, xmit_type)) {
 542                 if (skb_linearize(skb)) {
 543                         DP_NOTICE(edev,
 544                                   "SKB linearization failed - silently dropping this SKB\n");
 545                         dev_kfree_skb_any(skb);
 546                         return NETDEV_TX_OK;
 547                 }
 548         }
 549 #endif
 550
 551         /* Fill the entry in the SW ring and the BDs in the FW ring */
 552         idx = txq->sw_tx_prod & NUM_TX_BDS_MAX;
 553         txq->sw_tx_ring[idx].skb = skb;
 554         first_bd = (struct eth_tx_1st_bd *)
 555                    qed_chain_produce(&txq->tx_pbl);
 556         memset(first_bd, 0, sizeof(*first_bd));
 557         first_bd->data.bd_flags.bitfields =
 558                 1 << ETH_TX_1ST_BD_FLAGS_START_BD_SHIFT;
 559
 560         /* Map skb linear data for DMA and set in the first BD */
 561         mapping = dma_map_single(&edev->pdev->dev, skb->data,
 562                                  skb_headlen(skb), DMA_TO_DEVICE);
 563         if (unlikely(dma_mapping_error(&edev->pdev->dev, mapping))) {
 564                 DP_NOTICE(edev, "SKB mapping failed\n");
 565                 qede_free_failed_tx_pkt(edev, txq, first_bd, 0, false);
 566                 qede_update_tx_producer(txq);
 567                 return NETDEV_TX_OK;
 568         }
 569         nbd++;
 570         BD_SET_UNMAP_ADDR_LEN(first_bd, mapping, skb_headlen(skb));
 571
 572         /* In case there is IPv6 with extension headers or LSO we need 2nd and
 573          * 3rd BDs.
 574          */
 575         if (unlikely((xmit_type & XMIT_LSO) | ipv6_ext)) {
 576                 second_bd = (struct eth_tx_2nd_bd *)
 577                         qed_chain_produce(&txq->tx_pbl);
 578                 memset(second_bd, 0, sizeof(*second_bd));
 579
 580                 nbd++;
 581                 third_bd = (struct eth_tx_3rd_bd *)
 582                         qed_chain_produce(&txq->tx_pbl);
 583                 memset(third_bd, 0, sizeof(*third_bd));
 584
 585                 nbd++;
 586                 /* We need to fill in additional data in second_bd... */
 587                 tx_data_bd = (struct eth_tx_bd *)second_bd;
 588         }
 589
 590         if (skb_vlan_tag_present(skb)) {
 591                 first_bd->data.vlan = cpu_to_le16(skb_vlan_tag_get(skb));
 592                 first_bd->data.bd_flags.bitfields |=
 593                         1 << ETH_TX_1ST_BD_FLAGS_VLAN_INSERTION_SHIFT;
 594         }
 595
 596         /* Fill the parsing flags & params according to the requested offload */
 597         if (xmit_type & XMIT_L4_CSUM) {
 598                 /* We don't re-calculate IP checksum as it is already done by
 599                  * the upper stack
 600                  */
 601                 first_bd->data.bd_flags.bitfields |=
 602                         1 << ETH_TX_1ST_BD_FLAGS_L4_CSUM_SHIFT;
 603
 604                 if (xmit_type & XMIT_ENC) {
 605                         first_bd->data.bd_flags.bitfields |=
 606                                 1 << ETH_TX_1ST_BD_FLAGS_IP_CSUM_SHIFT;
 607                         first_bd->data.bitfields |=
 608                             1 << ETH_TX_DATA_1ST_BD_TUNN_FLAG_SHIFT;
 609                 }
 610
 611                 /* Legacy FW had flipped behavior in regard to this bit -
 612                  * I.e., needed to set to prevent FW from touching encapsulated
 613                  * packets when it didn't need to.
 614                  */
 615                 if (unlikely(txq->is_legacy))
 616                         first_bd->data.bitfields ^=
 617                             1 << ETH_TX_DATA_1ST_BD_TUNN_FLAG_SHIFT;
 618
 619                 /* If the packet is IPv6 with extension header, indicate that
 620                  * to FW and pass few params, since the device cracker doesn't
 621                  * support parsing IPv6 with extension header/s.
 622                  */
 623                 if (unlikely(ipv6_ext))
 624                         qede_set_params_for_ipv6_ext(skb, second_bd, third_bd);
 625         }
 626
 627         if (xmit_type & XMIT_LSO) {
 628                 first_bd->data.bd_flags.bitfields |=
 629                         (1 << ETH_TX_1ST_BD_FLAGS_LSO_SHIFT);
 630                 third_bd->data.lso_mss =
 631                         cpu_to_le16(skb_shinfo(skb)->gso_size);
 632
 633                 if (unlikely(xmit_type & XMIT_ENC)) {
 634                         first_bd->data.bd_flags.bitfields |=
 635                                 1 << ETH_TX_1ST_BD_FLAGS_TUNN_IP_CSUM_SHIFT;
 636                         hlen = qede_get_skb_hlen(skb, true);
 637                 } else {
 638                         first_bd->data.bd_flags.bitfields |=
 639                                 1 << ETH_TX_1ST_BD_FLAGS_IP_CSUM_SHIFT;
 640                         hlen = qede_get_skb_hlen(skb, false);
 641                 }
 642
 643                 /* @@@TBD - if will not be removed need to check */
 644                 third_bd->data.bitfields |=
 645                         cpu_to_le16((1 << ETH_TX_DATA_3RD_BD_HDR_NBD_SHIFT));
 646
 647                 /* Make life easier for FW guys who can't deal with header and
 648                  * data on same BD. If we need to split, use the second bd...
 649                  */
 650                 if (unlikely(skb_headlen(skb) > hlen)) {
 651                         DP_VERBOSE(edev, NETIF_MSG_TX_QUEUED,
 652                                    "TSO split header size is %d (%x:%x)\n",
 653                                    first_bd->nbytes, first_bd->addr.hi,
 654                                    first_bd->addr.lo);
 655
 656                         mapping = HILO_U64(le32_to_cpu(first_bd->addr.hi),
 657                                            le32_to_cpu(first_bd->addr.lo)) +
 658                                            hlen;
 659
 660                         BD_SET_UNMAP_ADDR_LEN(tx_data_bd, mapping,
 661                                               le16_to_cpu(first_bd->nbytes) -
 662                                               hlen);
 663
 664                         /* this marks the BD as one that has no
 665                          * individual mapping
 666                          */
 667                         txq->sw_tx_ring[idx].flags |= QEDE_TSO_SPLIT_BD;
 668
 669                         first_bd->nbytes = cpu_to_le16(hlen);
 670
 671                         tx_data_bd = (struct eth_tx_bd *)third_bd;
 672                         data_split = true;
 673                 }
 674         } else {
 675                 first_bd->data.bitfields |=
 676                     (skb->len & ETH_TX_DATA_1ST_BD_PKT_LEN_MASK) <<
 677                     ETH_TX_DATA_1ST_BD_PKT_LEN_SHIFT;
 678         }
 679
 680         /* Handle fragmented skb */
 681         /* special handle for frags inside 2nd and 3rd bds.. */
 682         while (tx_data_bd && frag_idx < skb_shinfo(skb)->nr_frags) {
 683                 rc = map_frag_to_bd(edev,
 684                                     &skb_shinfo(skb)->frags[frag_idx],
 685                                     tx_data_bd);
 686                 if (rc) {
 687                         qede_free_failed_tx_pkt(edev, txq, first_bd, nbd,
 688                                                 data_split);
 689                         qede_update_tx_producer(txq);
 690                         return NETDEV_TX_OK;
 691                 }
 692
 693                 if (tx_data_bd == (struct eth_tx_bd *)second_bd)
 694                         tx_data_bd = (struct eth_tx_bd *)third_bd;
 695                 else
 696                         tx_data_bd = NULL;
 697
 698                 frag_idx++;
 699         }
 700
 701         /* map last frags into 4th, 5th .... */
 702         for (; frag_idx < skb_shinfo(skb)->nr_frags; frag_idx++, nbd++) {
 703                 tx_data_bd = (struct eth_tx_bd *)
 704                              qed_chain_produce(&txq->tx_pbl);
 705
 706                 memset(tx_data_bd, 0, sizeof(*tx_data_bd));
 707
 708                 rc = map_frag_to_bd(edev,
 709                                     &skb_shinfo(skb)->frags[frag_idx],
 710                                     tx_data_bd);
 711                 if (rc) {
 712                         qede_free_failed_tx_pkt(edev, txq, first_bd, nbd,
 713                                                 data_split);
 714                         qede_update_tx_producer(txq);
 715                         return NETDEV_TX_OK;
 716                 }
 717         }
 718
 719         /* update the first BD with the actual num BDs */
 720         first_bd->data.nbds = nbd;
 721
 722         netdev_tx_sent_queue(netdev_txq, skb->len);
 723
 724         skb_tx_timestamp(skb);
 725
 726         /* Advance packet producer only before sending the packet since mapping
 727          * of pages may fail.
 728          */
 729         txq->sw_tx_prod++;
 730
 731         /* 'next page' entries are counted in the producer value */
 732         txq->tx_db.data.bd_prod =
 733                 cpu_to_le16(qed_chain_get_prod_idx(&txq->tx_pbl));
 734
 735         if (!skb->xmit_more || netif_xmit_stopped(netdev_txq))
 736                 qede_update_tx_producer(txq);
 737
 738         if (unlikely(qed_chain_get_elem_left(&txq->tx_pbl)
 739                       < (MAX_SKB_FRAGS + 1))) {
 740                 if (skb->xmit_more)
 741                         qede_update_tx_producer(txq);
 742
 743                 netif_tx_stop_queue(netdev_txq);
 744                 txq->stopped_cnt++;
 745                 DP_VERBOSE(edev, NETIF_MSG_TX_QUEUED,
 746                            "Stop queue was called\n");
 747                 /* paired memory barrier is in qede_tx_int(), we have to keep
 748                  * ordering of set_bit() in netif_tx_stop_queue() and read of
 749                  * fp->bd_tx_cons
 750                  */
 751                 smp_mb();
 752
 753                 if (qed_chain_get_elem_left(&txq->tx_pbl)
 754                      >= (MAX_SKB_FRAGS + 1) &&
 755                     (edev->state == QEDE_STATE_OPEN)) {
 756                         netif_tx_wake_queue(netdev_txq);
 757                         DP_VERBOSE(edev, NETIF_MSG_TX_QUEUED,
 758                                    "Wake queue was called\n");
 759                 }
 760         }
 761
 762         return NETDEV_TX_OK;
 763 }
 764
 765 int qede_txq_has_work(struct qede_tx_queue *txq)
 766 {
 767         u16 hw_bd_cons;
 768
 769         /* Tell compiler that consumer and producer can change */
 770         barrier();
 771         hw_bd_cons = le16_to_cpu(*txq->hw_cons_ptr);
 772         if (qed_chain_get_cons_idx(&txq->tx_pbl) == hw_bd_cons + 1)
 773                 return 0;
 774
 775         return hw_bd_cons != qed_chain_get_cons_idx(&txq->tx_pbl);
 776 }
 777
 778 static int qede_tx_int(struct qede_dev *edev, struct qede_tx_queue *txq)
 779 {
 780         struct netdev_queue *netdev_txq;
 781         u16 hw_bd_cons;
 782         unsigned int pkts_compl = 0, bytes_compl = 0;
 783         int rc;
 784
 785         netdev_txq = netdev_get_tx_queue(edev->ndev, txq->index);
 786
 787         hw_bd_cons = le16_to_cpu(*txq->hw_cons_ptr);
 788         barrier();
 789
 790         while (hw_bd_cons != qed_chain_get_cons_idx(&txq->tx_pbl)) {
 791                 int len = 0;
 792
 793                 rc = qede_free_tx_pkt(edev, txq, &len);
 794                 if (rc) {
 795                         DP_NOTICE(edev, "hw_bd_cons = %d, chain_cons=%d\n",
 796                                   hw_bd_cons,
 797                                   qed_chain_get_cons_idx(&txq->tx_pbl));
 798                         break;
 799                 }
 800
 801                 bytes_compl += len;
 802                 pkts_compl++;
 803                 txq->sw_tx_cons++;
 804                 txq->xmit_pkts++;
 805         }
 806
 807         netdev_tx_completed_queue(netdev_txq, pkts_compl, bytes_compl);
 808
 809         /* Need to make the tx_bd_cons update visible to start_xmit()
 810          * before checking for netif_tx_queue_stopped().  Without the
 811          * memory barrier, there is a small possibility that
 812          * start_xmit() will miss it and cause the queue to be stopped
 813          * forever.
 814          * On the other hand we need an rmb() here to ensure the proper
 815          * ordering of bit testing in the following
 816          * netif_tx_queue_stopped(txq) call.
 817          */
 818         smp_mb();
 819
 820         if (unlikely(netif_tx_queue_stopped(netdev_txq))) {
 821                 /* Taking tx_lock is needed to prevent reenabling the queue
 822                  * while it's empty. This could have happen if rx_action() gets
 823                  * suspended in qede_tx_int() after the condition before
 824                  * netif_tx_wake_queue(), while tx_action (qede_start_xmit()):
 825                  *
 826                  * stops the queue->sees fresh tx_bd_cons->releases the queue->
 827                  * sends some packets consuming the whole queue again->
 828                  * stops the queue
 829                  */
 830
 831                 __netif_tx_lock(netdev_txq, smp_processor_id());
 832
 833                 if ((netif_tx_queue_stopped(netdev_txq)) &&
 834                     (edev->state == QEDE_STATE_OPEN) &&
 835                     (qed_chain_get_elem_left(&txq->tx_pbl)
 836                       >= (MAX_SKB_FRAGS + 1))) {
 837                         netif_tx_wake_queue(netdev_txq);
 838                         DP_VERBOSE(edev, NETIF_MSG_TX_DONE,
 839                                    "Wake queue was called\n");
 840                 }
 841
 842                 __netif_tx_unlock(netdev_txq);
 843         }
 844
 845         return 0;
 846 }
 847
 848 bool qede_has_rx_work(struct qede_rx_queue *rxq)
 849 {
 850         u16 hw_comp_cons, sw_comp_cons;
 851
 852         /* Tell compiler that status block fields can change */
 853         barrier();
 854
 855         hw_comp_cons = le16_to_cpu(*rxq->hw_cons_ptr);
 856         sw_comp_cons = qed_chain_get_cons_idx(&rxq->rx_comp_ring);
 857
 858         return hw_comp_cons != sw_comp_cons;
 859 }
 860
 861 static bool qede_has_tx_work(struct qede_fastpath *fp)
 862 {
 863         u8 tc;
 864
 865         for (tc = 0; tc < fp->edev->num_tc; tc++)
 866                 if (qede_txq_has_work(&fp->txqs[tc]))
 867                         return true;
 868         return false;
 869 }
 870
 871 static inline void qede_rx_bd_ring_consume(struct qede_rx_queue *rxq)
 872 {
 873         qed_chain_consume(&rxq->rx_bd_ring);
 874         rxq->sw_rx_cons++;
 875 }
 876
 877 /* This function reuses the buffer(from an offset) from
 878  * consumer index to producer index in the bd ring
 879  */
 880 static inline void qede_reuse_page(struct qede_dev *edev,
 881                                    struct qede_rx_queue *rxq,
 882                                    struct sw_rx_data *curr_cons)
 883 {
 884         struct eth_rx_bd *rx_bd_prod = qed_chain_produce(&rxq->rx_bd_ring);
 885         struct sw_rx_data *curr_prod;
 886         dma_addr_t new_mapping;
 887
 888         curr_prod = &rxq->sw_rx_ring[rxq->sw_rx_prod & NUM_RX_BDS_MAX];
 889         *curr_prod = *curr_cons;
 890
 891         new_mapping = curr_prod->mapping + curr_prod->page_offset;
 892
 893         rx_bd_prod->addr.hi = cpu_to_le32(upper_32_bits(new_mapping));
 894         rx_bd_prod->addr.lo = cpu_to_le32(lower_32_bits(new_mapping));
 895
 896         rxq->sw_rx_prod++;
 897         curr_cons->data = NULL;
 898 }
 899
 900 /* In case of allocation failures reuse buffers
 901  * from consumer index to produce buffers for firmware
 902  */
 903 void qede_recycle_rx_bd_ring(struct qede_rx_queue *rxq,
 904                              struct qede_dev *edev, u8 count)
 905 {
 906         struct sw_rx_data *curr_cons;
 907
 908         for (; count > 0; count--) {
 909                 curr_cons = &rxq->sw_rx_ring[rxq->sw_rx_cons & NUM_RX_BDS_MAX];
 910                 qede_reuse_page(edev, rxq, curr_cons);
 911                 qede_rx_bd_ring_consume(rxq);
 912         }
 913 }
 914
 915 static inline int qede_realloc_rx_buffer(struct qede_dev *edev,
 916                                          struct qede_rx_queue *rxq,
 917                                          struct sw_rx_data *curr_cons)
 918 {
 919         /* Move to the next segment in the page */
 920         curr_cons->page_offset += rxq->rx_buf_seg_size;
 921
 922         if (curr_cons->page_offset == PAGE_SIZE) {
 923                 if (unlikely(qede_alloc_rx_buffer(edev, rxq))) {
 924                         /* Since we failed to allocate new buffer
 925                          * current buffer can be used again.
 926                          */
 927                         curr_cons->page_offset -= rxq->rx_buf_seg_size;
 928
 929                         return -ENOMEM;
 930                 }
 931
 932                 dma_unmap_page(&edev->pdev->dev, curr_cons->mapping,
 933                                PAGE_SIZE, DMA_FROM_DEVICE);
 934         } else {
 935                 /* Increment refcount of the page as we don't want
 936                  * network stack to take the ownership of the page
 937                  * which can be recycled multiple times by the driver.
 938                  */
 939                 page_ref_inc(curr_cons->data);
 940                 qede_reuse_page(edev, rxq, curr_cons);
 941         }
 942
 943         return 0;
 944 }
 945
 946 void qede_update_rx_prod(struct qede_dev *edev, struct qede_rx_queue *rxq)
 947 {
 948         u16 bd_prod = qed_chain_get_prod_idx(&rxq->rx_bd_ring);
 949         u16 cqe_prod = qed_chain_get_prod_idx(&rxq->rx_comp_ring);
 950         struct eth_rx_prod_data rx_prods = {0};
 951
 952         /* Update producers */
 953         rx_prods.bd_prod = cpu_to_le16(bd_prod);
 954         rx_prods.cqe_prod = cpu_to_le16(cqe_prod);
 955
 956         /* Make sure that the BD and SGE data is updated before updating the
 957          * producers since FW might read the BD/SGE right after the producer
 958          * is updated.
 959          */
 960         wmb();
 961
 962         internal_ram_wr(rxq->hw_rxq_prod_addr, sizeof(rx_prods),
 963                         (u32 *)&rx_prods);
 964
 965         /* mmiowb is needed to synchronize doorbell writes from more than one
 966          * processor. It guarantees that the write arrives to the device before
 967          * the napi lock is released and another qede_poll is called (possibly
 968          * on another CPU). Without this barrier, the next doorbell can bypass
 969          * this doorbell. This is applicable to IA64/Altix systems.
 970          */
 971         mmiowb();
 972 }
 973
 974 static u32 qede_get_rxhash(struct qede_dev *edev,
 975                            u8 bitfields,
 976                            __le32 rss_hash, enum pkt_hash_types *rxhash_type)
 977 {
 978         enum rss_hash_type htype;
 979
 980         htype = GET_FIELD(bitfields, ETH_FAST_PATH_RX_REG_CQE_RSS_HASH_TYPE);
 981
 982         if ((edev->ndev->features & NETIF_F_RXHASH) && htype) {
 983                 *rxhash_type = ((htype == RSS_HASH_TYPE_IPV4) ||
 984                                 (htype == RSS_HASH_TYPE_IPV6)) ?
 985                                 PKT_HASH_TYPE_L3 : PKT_HASH_TYPE_L4;
 986                 return le32_to_cpu(rss_hash);
 987         }
 988         *rxhash_type = PKT_HASH_TYPE_NONE;
 989         return 0;
 990 }
 991
 992 static void qede_set_skb_csum(struct sk_buff *skb, u8 csum_flag)
 993 {
 994         skb_checksum_none_assert(skb);
 995
 996         if (csum_flag & QEDE_CSUM_UNNECESSARY)
 997                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 998
 999         if (csum_flag & QEDE_TUNN_CSUM_UNNECESSARY)
1000                 skb->csum_level = 1;
1001 }
1002
1003 static inline void qede_skb_receive(struct qede_dev *edev,
1004                                     struct qede_fastpath *fp,
1005                                     struct sk_buff *skb, u16 vlan_tag)
1006 {
1007         if (vlan_tag)
1008                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tag);
1009
1010         napi_gro_receive(&fp->napi, skb);
1011 }
1012
1013 static void qede_set_gro_params(struct qede_dev *edev,
1014                                 struct sk_buff *skb,
1015                                 struct eth_fast_path_rx_tpa_start_cqe *cqe)
1016 {
1017         u16 parsing_flags = le16_to_cpu(cqe->pars_flags.flags);
1018
1019         if (((parsing_flags >> PARSING_AND_ERR_FLAGS_L3TYPE_SHIFT) &
1020             PARSING_AND_ERR_FLAGS_L3TYPE_MASK) == 2)
1021                 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1022         else
1023                 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1024
1025         skb_shinfo(skb)->gso_size = __le16_to_cpu(cqe->len_on_first_bd) -
1026                                         cqe->header_len;
1027 }
1028
1029 static int qede_fill_frag_skb(struct qede_dev *edev,
1030                               struct qede_rx_queue *rxq,
1031                               u8 tpa_agg_index, u16 len_on_bd)
1032 {
1033         struct sw_rx_data *current_bd = &rxq->sw_rx_ring[rxq->sw_rx_cons &
1034                                                          NUM_RX_BDS_MAX];
1035         struct qede_agg_info *tpa_info = &rxq->tpa_info[tpa_agg_index];
1036         struct sk_buff *skb = tpa_info->skb;
1037
1038         if (unlikely(tpa_info->agg_state != QEDE_AGG_STATE_START))
1039                 goto out;
1040
1041         /* Add one frag and update the appropriate fields in the skb */
1042         skb_fill_page_desc(skb, tpa_info->frag_id++,
1043                            current_bd->data, current_bd->page_offset,
1044                            len_on_bd);
1045
1046         if (unlikely(qede_realloc_rx_buffer(edev, rxq, current_bd))) {
1047                 /* Incr page ref count to reuse on allocation failure
1048                  * so that it doesn't get freed while freeing SKB.
1049                  */
1050                 page_ref_inc(current_bd->data);
1051                 goto out;
1052         }
1053
1054         qed_chain_consume(&rxq->rx_bd_ring);
1055         rxq->sw_rx_cons++;
1056
1057         skb->data_len += len_on_bd;
1058         skb->truesize += rxq->rx_buf_seg_size;
1059         skb->len += len_on_bd;
1060
1061         return 0;
1062
1063 out:
1064         tpa_info->agg_state = QEDE_AGG_STATE_ERROR;
1065         qede_recycle_rx_bd_ring(rxq, edev, 1);
1066         return -ENOMEM;
1067 }
1068
1069 static void qede_tpa_start(struct qede_dev *edev,
1070                            struct qede_rx_queue *rxq,
1071                            struct eth_fast_path_rx_tpa_start_cqe *cqe)
1072 {
1073         struct qede_agg_info *tpa_info = &rxq->tpa_info[cqe->tpa_agg_index];
1074         struct eth_rx_bd *rx_bd_cons = qed_chain_consume(&rxq->rx_bd_ring);
1075         struct eth_rx_bd *rx_bd_prod = qed_chain_produce(&rxq->rx_bd_ring);
1076         struct sw_rx_data *replace_buf = &tpa_info->replace_buf;
1077         dma_addr_t mapping = tpa_info->replace_buf_mapping;
1078         struct sw_rx_data *sw_rx_data_cons;
1079         struct sw_rx_data *sw_rx_data_prod;
1080         enum pkt_hash_types rxhash_type;
1081         u32 rxhash;
1082
1083         sw_rx_data_cons = &rxq->sw_rx_ring[rxq->sw_rx_cons & NUM_RX_BDS_MAX];
1084         sw_rx_data_prod = &rxq->sw_rx_ring[rxq->sw_rx_prod & NUM_RX_BDS_MAX];
1085
1086         /* Use pre-allocated replacement buffer - we can't release the agg.
1087          * start until its over and we don't want to risk allocation failing
1088          * here, so re-allocate when aggregation will be over.
1089          */
1090         sw_rx_data_prod->mapping = replace_buf->mapping;
1091
1092         sw_rx_data_prod->data = replace_buf->data;
1093         rx_bd_prod->addr.hi = cpu_to_le32(upper_32_bits(mapping));
1094         rx_bd_prod->addr.lo = cpu_to_le32(lower_32_bits(mapping));
1095         sw_rx_data_prod->page_offset = replace_buf->page_offset;
1096
1097         rxq->sw_rx_prod++;
1098
1099         /* move partial skb from cons to pool (don't unmap yet)
1100          * save mapping, incase we drop the packet later on.
1101          */
1102         tpa_info->start_buf = *sw_rx_data_cons;
1103         mapping = HILO_U64(le32_to_cpu(rx_bd_cons->addr.hi),
1104                            le32_to_cpu(rx_bd_cons->addr.lo));
1105
1106         tpa_info->start_buf_mapping = mapping;
1107         rxq->sw_rx_cons++;
1108
1109         /* set tpa state to start only if we are able to allocate skb
1110          * for this aggregation, otherwise mark as error and aggregation will
1111          * be dropped
1112          */
1113         tpa_info->skb = netdev_alloc_skb(edev->ndev,
1114                                          le16_to_cpu(cqe->len_on_first_bd));
1115         if (unlikely(!tpa_info->skb)) {
1116                 DP_NOTICE(edev, "Failed to allocate SKB for gro\n");
1117                 tpa_info->agg_state = QEDE_AGG_STATE_ERROR;
1118                 goto cons_buf;
1119         }
1120
1121         skb_put(tpa_info->skb, le16_to_cpu(cqe->len_on_first_bd));
1122         memcpy(&tpa_info->start_cqe, cqe, sizeof(tpa_info->start_cqe));
1123
1124         /* Start filling in the aggregation info */
1125         tpa_info->frag_id = 0;
1126         tpa_info->agg_state = QEDE_AGG_STATE_START;
1127
1128         rxhash = qede_get_rxhash(edev, cqe->bitfields,
1129                                  cqe->rss_hash, &rxhash_type);
1130         skb_set_hash(tpa_info->skb, rxhash, rxhash_type);
1131         if ((le16_to_cpu(cqe->pars_flags.flags) >>
1132              PARSING_AND_ERR_FLAGS_TAG8021QEXIST_SHIFT) &
1133                     PARSING_AND_ERR_FLAGS_TAG8021QEXIST_MASK)
1134                 tpa_info->vlan_tag = le16_to_cpu(cqe->vlan_tag);
1135         else
1136                 tpa_info->vlan_tag = 0;
1137
1138         /* This is needed in order to enable forwarding support */
1139         qede_set_gro_params(edev, tpa_info->skb, cqe);
1140
1141 cons_buf: /* We still need to handle bd_len_list to consume buffers */
1142         if (likely(cqe->ext_bd_len_list[0]))
1143                 qede_fill_frag_skb(edev, rxq, cqe->tpa_agg_index,
1144                                    le16_to_cpu(cqe->ext_bd_len_list[0]));
1145
1146         if (unlikely(cqe->ext_bd_len_list[1])) {
1147                 DP_ERR(edev,
1148                        "Unlikely - got a TPA aggregation with more than one ext_bd_len_list entry in the TPA start\n");
1149                 tpa_info->agg_state = QEDE_AGG_STATE_ERROR;
1150         }
1151 }
1152
1153 #ifdef CONFIG_INET
1154 static void qede_gro_ip_csum(struct sk_buff *skb)
1155 {
1156         const struct iphdr *iph = ip_hdr(skb);
1157         struct tcphdr *th;
1158
1159         skb_set_transport_header(skb, sizeof(struct iphdr));
1160         th = tcp_hdr(skb);
1161
1162         th->check = ~tcp_v4_check(skb->len - skb_transport_offset(skb),
1163                                   iph->saddr, iph->daddr, 0);
1164
1165         tcp_gro_complete(skb);
1166 }
1167
1168 static void qede_gro_ipv6_csum(struct sk_buff *skb)
1169 {
1170         struct ipv6hdr *iph = ipv6_hdr(skb);
1171         struct tcphdr *th;
1172
1173         skb_set_transport_header(skb, sizeof(struct ipv6hdr));
1174         th = tcp_hdr(skb);
1175
1176         th->check = ~tcp_v6_check(skb->len - skb_transport_offset(skb),
1177                                   &iph->saddr, &iph->daddr, 0);
1178         tcp_gro_complete(skb);
1179 }
1180 #endif
1181
1182 static void qede_gro_receive(struct qede_dev *edev,
1183                              struct qede_fastpath *fp,
1184                              struct sk_buff *skb,
1185                              u16 vlan_tag)
1186 {
1187         /* FW can send a single MTU sized packet from gro flow
1188          * due to aggregation timeout/last segment etc. which
1189          * is not expected to be a gro packet. If a skb has zero
1190          * frags then simply push it in the stack as non gso skb.
1191          */
1192         if (unlikely(!skb->data_len)) {
1193                 skb_shinfo(skb)->gso_type = 0;
1194                 skb_shinfo(skb)->gso_size = 0;
1195                 goto send_skb;
1196         }
1197
1198 #ifdef CONFIG_INET
1199         if (skb_shinfo(skb)->gso_size) {
1200                 skb_set_network_header(skb, 0);
1201
1202                 switch (skb->protocol) {
1203                 case htons(ETH_P_IP):
1204                         qede_gro_ip_csum(skb);
1205                         break;
1206                 case htons(ETH_P_IPV6):
1207                         qede_gro_ipv6_csum(skb);
1208                         break;
1209                 default:
1210                         DP_ERR(edev,
1211                                "Error: FW GRO supports only IPv4/IPv6, not 0x%04x\n",
1212                                ntohs(skb->protocol));
1213                 }
1214         }
1215 #endif
1216
1217 send_skb:
1218         skb_record_rx_queue(skb, fp->rxq->rxq_id);
1219         qede_skb_receive(edev, fp, skb, vlan_tag);
1220 }
1221
1222 static inline void qede_tpa_cont(struct qede_dev *edev,
1223                                  struct qede_rx_queue *rxq,
1224                                  struct eth_fast_path_rx_tpa_cont_cqe *cqe)
1225 {
1226         int i;
1227
1228         for (i = 0; cqe->len_list[i]; i++)
1229                 qede_fill_frag_skb(edev, rxq, cqe->tpa_agg_index,
1230                                    le16_to_cpu(cqe->len_list[i]));
1231
1232         if (unlikely(i > 1))
1233                 DP_ERR(edev,
1234                        "Strange - TPA cont with more than a single len_list entry\n");
1235 }
1236
1237 static void qede_tpa_end(struct qede_dev *edev,
1238                          struct qede_fastpath *fp,
1239                          struct eth_fast_path_rx_tpa_end_cqe *cqe)
1240 {
1241         struct qede_rx_queue *rxq = fp->rxq;
1242         struct qede_agg_info *tpa_info;
1243         struct sk_buff *skb;
1244         int i;
1245
1246         tpa_info = &rxq->tpa_info[cqe->tpa_agg_index];
1247         skb = tpa_info->skb;
1248
1249         for (i = 0; cqe->len_list[i]; i++)
1250                 qede_fill_frag_skb(edev, rxq, cqe->tpa_agg_index,
1251                                    le16_to_cpu(cqe->len_list[i]));
1252         if (unlikely(i > 1))
1253                 DP_ERR(edev,
1254                        "Strange - TPA emd with more than a single len_list entry\n");
1255
1256         if (unlikely(tpa_info->agg_state != QEDE_AGG_STATE_START))
1257                 goto err;
1258
1259         /* Sanity */
1260         if (unlikely(cqe->num_of_bds != tpa_info->frag_id + 1))
1261                 DP_ERR(edev,
1262                        "Strange - TPA had %02x BDs, but SKB has only %d frags\n",
1263                        cqe->num_of_bds, tpa_info->frag_id);
1264         if (unlikely(skb->len != le16_to_cpu(cqe->total_packet_len)))
1265                 DP_ERR(edev,
1266                        "Strange - total packet len [cqe] is %4x but SKB has len %04x\n",
1267                        le16_to_cpu(cqe->total_packet_len), skb->len);
1268
1269         memcpy(skb->data,
1270                page_address(tpa_info->start_buf.data) +
1271                 tpa_info->start_cqe.placement_offset +
1272                 tpa_info->start_buf.page_offset,
1273                le16_to_cpu(tpa_info->start_cqe.len_on_first_bd));
1274
1275         /* Recycle [mapped] start buffer for the next replacement */
1276         tpa_info->replace_buf = tpa_info->start_buf;
1277         tpa_info->replace_buf_mapping = tpa_info->start_buf_mapping;
1278
1279         /* Finalize the SKB */
1280         skb->protocol = eth_type_trans(skb, edev->ndev);
1281         skb->ip_summed = CHECKSUM_UNNECESSARY;
1282
1283         /* tcp_gro_complete() will copy NAPI_GRO_CB(skb)->count
1284          * to skb_shinfo(skb)->gso_segs
1285          */
1286         NAPI_GRO_CB(skb)->count = le16_to_cpu(cqe->num_of_coalesced_segs);
1287
1288         qede_gro_receive(edev, fp, skb, tpa_info->vlan_tag);
1289
1290         tpa_info->agg_state = QEDE_AGG_STATE_NONE;
1291
1292         return;
1293 err:
1294         /* The BD starting the aggregation is still mapped; Re-use it for
1295          * future aggregations [as replacement buffer]
1296          */
1297         memcpy(&tpa_info->replace_buf, &tpa_info->start_buf,
1298                sizeof(struct sw_rx_data));
1299         tpa_info->replace_buf_mapping = tpa_info->start_buf_mapping;
1300         tpa_info->start_buf.data = NULL;
1301         tpa_info->agg_state = QEDE_AGG_STATE_NONE;
1302         dev_kfree_skb_any(tpa_info->skb);
1303         tpa_info->skb = NULL;
1304 }
1305
1306 static bool qede_tunn_exist(u16 flag)
1307 {
1308         return !!(flag & (PARSING_AND_ERR_FLAGS_TUNNELEXIST_MASK <<
1309                           PARSING_AND_ERR_FLAGS_TUNNELEXIST_SHIFT));
1310 }
1311
1312 static u8 qede_check_tunn_csum(u16 flag)
1313 {
1314         u16 csum_flag = 0;
1315         u8 tcsum = 0;
1316
1317         if (flag & (PARSING_AND_ERR_FLAGS_TUNNELL4CHKSMWASCALCULATED_MASK <<
1318                     PARSING_AND_ERR_FLAGS_TUNNELL4CHKSMWASCALCULATED_SHIFT))
1319                 csum_flag |= PARSING_AND_ERR_FLAGS_TUNNELL4CHKSMERROR_MASK <<
1320                              PARSING_AND_ERR_FLAGS_TUNNELL4CHKSMERROR_SHIFT;
1321
1322         if (flag & (PARSING_AND_ERR_FLAGS_L4CHKSMWASCALCULATED_MASK <<
1323                     PARSING_AND_ERR_FLAGS_L4CHKSMWASCALCULATED_SHIFT)) {
1324                 csum_flag |= PARSING_AND_ERR_FLAGS_L4CHKSMERROR_MASK <<
1325                              PARSING_AND_ERR_FLAGS_L4CHKSMERROR_SHIFT;
1326                 tcsum = QEDE_TUNN_CSUM_UNNECESSARY;
1327         }
1328
1329         csum_flag |= PARSING_AND_ERR_FLAGS_TUNNELIPHDRERROR_MASK <<
1330                      PARSING_AND_ERR_FLAGS_TUNNELIPHDRERROR_SHIFT |
1331                      PARSING_AND_ERR_FLAGS_IPHDRERROR_MASK <<
1332                      PARSING_AND_ERR_FLAGS_IPHDRERROR_SHIFT;
1333
1334         if (csum_flag & flag)
1335                 return QEDE_CSUM_ERROR;
1336
1337         return QEDE_CSUM_UNNECESSARY | tcsum;
1338 }
1339
1340 static u8 qede_check_notunn_csum(u16 flag)
1341 {
1342         u16 csum_flag = 0;
1343         u8 csum = 0;
1344
1345         if (flag & (PARSING_AND_ERR_FLAGS_L4CHKSMWASCALCULATED_MASK <<
1346                     PARSING_AND_ERR_FLAGS_L4CHKSMWASCALCULATED_SHIFT)) {
1347                 csum_flag |= PARSING_AND_ERR_FLAGS_L4CHKSMERROR_MASK <<
1348                              PARSING_AND_ERR_FLAGS_L4CHKSMERROR_SHIFT;
1349                 csum = QEDE_CSUM_UNNECESSARY;
1350         }
1351
1352         csum_flag |= PARSING_AND_ERR_FLAGS_IPHDRERROR_MASK <<
1353                      PARSING_AND_ERR_FLAGS_IPHDRERROR_SHIFT;
1354
1355         if (csum_flag & flag)
1356                 return QEDE_CSUM_ERROR;
1357
1358         return csum;
1359 }
1360
1361 static u8 qede_check_csum(u16 flag)
1362 {
1363         if (!qede_tunn_exist(flag))
1364                 return qede_check_notunn_csum(flag);
1365         else
1366                 return qede_check_tunn_csum(flag);
1367 }
1368
1369 static bool qede_pkt_is_ip_fragmented(struct eth_fast_path_rx_reg_cqe *cqe,
1370                                       u16 flag)
1371 {
1372         u8 tun_pars_flg = cqe->tunnel_pars_flags.flags;
1373
1374         if ((tun_pars_flg & (ETH_TUNNEL_PARSING_FLAGS_IPV4_FRAGMENT_MASK <<
1375                              ETH_TUNNEL_PARSING_FLAGS_IPV4_FRAGMENT_SHIFT)) ||
1376             (flag & (PARSING_AND_ERR_FLAGS_IPV4FRAG_MASK <<
1377                      PARSING_AND_ERR_FLAGS_IPV4FRAG_SHIFT)))
1378                 return true;
1379
1380         return false;
1381 }
1382
1383 static int qede_rx_int(struct qede_fastpath *fp, int budget)
1384 {
1385         struct qede_dev *edev = fp->edev;
1386         struct qede_rx_queue *rxq = fp->rxq;
1387
1388         u16 hw_comp_cons, sw_comp_cons, sw_rx_index, parse_flag;
1389         int rx_pkt = 0;
1390         u8 csum_flag;
1391
1392         hw_comp_cons = le16_to_cpu(*rxq->hw_cons_ptr);
1393         sw_comp_cons = qed_chain_get_cons_idx(&rxq->rx_comp_ring);
1394
1395         /* Memory barrier to prevent the CPU from doing speculative reads of CQE
1396          * / BD in the while-loop before reading hw_comp_cons. If the CQE is
1397          * read before it is written by FW, then FW writes CQE and SB, and then
1398          * the CPU reads the hw_comp_cons, it will use an old CQE.
1399          */
1400         rmb();
1401
1402         /* Loop to complete all indicated BDs */
1403         while (sw_comp_cons != hw_comp_cons) {
1404                 struct eth_fast_path_rx_reg_cqe *fp_cqe;
1405                 enum pkt_hash_types rxhash_type;
1406                 enum eth_rx_cqe_type cqe_type;
1407                 struct sw_rx_data *sw_rx_data;
1408                 union eth_rx_cqe *cqe;
1409                 struct sk_buff *skb;
1410                 struct page *data;
1411                 __le16 flags;
1412                 u16 len, pad;
1413                 u32 rx_hash;
1414
1415                 /* Get the CQE from the completion ring */
1416                 cqe = (union eth_rx_cqe *)
1417                         qed_chain_consume(&rxq->rx_comp_ring);
1418                 cqe_type = cqe->fast_path_regular.type;
1419
1420                 if (unlikely(cqe_type == ETH_RX_CQE_TYPE_SLOW_PATH)) {
1421                         edev->ops->eth_cqe_completion(
1422                                         edev->cdev, fp->id,
1423                                         (struct eth_slow_path_rx_cqe *)cqe);
1424                         goto next_cqe;
1425                 }
1426
1427                 if (cqe_type != ETH_RX_CQE_TYPE_REGULAR) {
1428                         switch (cqe_type) {
1429                         case ETH_RX_CQE_TYPE_TPA_START:
1430                                 qede_tpa_start(edev, rxq,
1431                                                &cqe->fast_path_tpa_start);
1432                                 goto next_cqe;
1433                         case ETH_RX_CQE_TYPE_TPA_CONT:
1434                                 qede_tpa_cont(edev, rxq,
1435                                               &cqe->fast_path_tpa_cont);
1436                                 goto next_cqe;
1437                         case ETH_RX_CQE_TYPE_TPA_END:
1438                                 qede_tpa_end(edev, fp,
1439                                              &cqe->fast_path_tpa_end);
1440                                 goto next_rx_only;
1441                         default:
1442                                 break;
1443                         }
1444                 }
1445
1446                 /* Get the data from the SW ring */
1447                 sw_rx_index = rxq->sw_rx_cons & NUM_RX_BDS_MAX;
1448                 sw_rx_data = &rxq->sw_rx_ring[sw_rx_index];
1449                 data = sw_rx_data->data;
1450
1451                 fp_cqe = &cqe->fast_path_regular;
1452                 len =  le16_to_cpu(fp_cqe->len_on_first_bd);
1453                 pad = fp_cqe->placement_offset;
1454                 flags = cqe->fast_path_regular.pars_flags.flags;
1455
1456                 /* If this is an error packet then drop it */
1457                 parse_flag = le16_to_cpu(flags);
1458
1459                 csum_flag = qede_check_csum(parse_flag);
1460                 if (unlikely(csum_flag == QEDE_CSUM_ERROR)) {
1461                         if (qede_pkt_is_ip_fragmented(&cqe->fast_path_regular,
1462                                                       parse_flag)) {
1463                                 rxq->rx_ip_frags++;
1464                                 goto alloc_skb;
1465                         }
1466
1467                         DP_NOTICE(edev,
1468                                   "CQE in CONS = %u has error, flags = %x, dropping incoming packet\n",
1469                                   sw_comp_cons, parse_flag);
1470                         rxq->rx_hw_errors++;
1471                         qede_recycle_rx_bd_ring(rxq, edev, fp_cqe->bd_num);
1472                         goto next_cqe;
1473                 }
1474
1475 alloc_skb:
1476                 skb = netdev_alloc_skb(edev->ndev, QEDE_RX_HDR_SIZE);
1477                 if (unlikely(!skb)) {
1478                         DP_NOTICE(edev,
1479                                   "skb allocation failed, dropping incoming packet\n");
1480                         qede_recycle_rx_bd_ring(rxq, edev, fp_cqe->bd_num);
1481                         rxq->rx_alloc_errors++;
1482                         goto next_cqe;
1483                 }
1484
1485                 /* Copy data into SKB */
1486                 if (len + pad <= edev->rx_copybreak) {
1487                         memcpy(skb_put(skb, len),
1488                                page_address(data) + pad +
1489                                 sw_rx_data->page_offset, len);
1490                         qede_reuse_page(edev, rxq, sw_rx_data);
1491                 } else {
1492                         struct skb_frag_struct *frag;
1493                         unsigned int pull_len;
1494                         unsigned char *va;
1495
1496                         frag = &skb_shinfo(skb)->frags[0];
1497
1498                         skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, data,
1499                                         pad + sw_rx_data->page_offset,
1500                                         len, rxq->rx_buf_seg_size);
1501
1502                         va = skb_frag_address(frag);
1503                         pull_len = eth_get_headlen(va, QEDE_RX_HDR_SIZE);
1504
1505                         /* Align the pull_len to optimize memcpy */
1506                         memcpy(skb->data, va, ALIGN(pull_len, sizeof(long)));
1507
1508                         skb_frag_size_sub(frag, pull_len);
1509                         frag->page_offset += pull_len;
1510                         skb->data_len -= pull_len;
1511                         skb->tail += pull_len;
1512
1513                         if (unlikely(qede_realloc_rx_buffer(edev, rxq,
1514                                                             sw_rx_data))) {
1515                                 DP_ERR(edev, "Failed to allocate rx buffer\n");
1516                                 /* Incr page ref count to reuse on allocation
1517                                  * failure so that it doesn't get freed while
1518                                  * freeing SKB.
1519                                  */
1520
1521                                 page_ref_inc(sw_rx_data->data);
1522                                 rxq->rx_alloc_errors++;
1523                                 qede_recycle_rx_bd_ring(rxq, edev,
1524                                                         fp_cqe->bd_num);
1525                                 dev_kfree_skb_any(skb);
1526                                 goto next_cqe;
1527                         }
1528                 }
1529
1530                 qede_rx_bd_ring_consume(rxq);
1531
1532                 if (fp_cqe->bd_num != 1) {
1533                         u16 pkt_len = le16_to_cpu(fp_cqe->pkt_len);
1534                         u8 num_frags;
1535
1536                         pkt_len -= len;
1537
1538                         for (num_frags = fp_cqe->bd_num - 1; num_frags > 0;
1539                              num_frags--) {
1540                                 u16 cur_size = pkt_len > rxq->rx_buf_size ?
1541                                                 rxq->rx_buf_size : pkt_len;
1542                                 if (unlikely(!cur_size)) {
1543                                         DP_ERR(edev,
1544                                                "Still got %d BDs for mapping jumbo, but length became 0\n",
1545                                                num_frags);
1546                                         qede_recycle_rx_bd_ring(rxq, edev,
1547                                                                 num_frags);
1548                                         dev_kfree_skb_any(skb);
1549                                         goto next_cqe;
1550                                 }
1551
1552                                 if (unlikely(qede_alloc_rx_buffer(edev, rxq))) {
1553                                         qede_recycle_rx_bd_ring(rxq, edev,
1554                                                                 num_frags);
1555                                         dev_kfree_skb_any(skb);
1556                                         goto next_cqe;
1557                                 }
1558
1559                                 sw_rx_index = rxq->sw_rx_cons & NUM_RX_BDS_MAX;
1560                                 sw_rx_data = &rxq->sw_rx_ring[sw_rx_index];
1561                                 qede_rx_bd_ring_consume(rxq);
1562
1563                                 dma_unmap_page(&edev->pdev->dev,
1564                                                sw_rx_data->mapping,
1565                                                PAGE_SIZE, DMA_FROM_DEVICE);
1566
1567                                 skb_fill_page_desc(skb,
1568                                                    skb_shinfo(skb)->nr_frags++,
1569                                                    sw_rx_data->data, 0,
1570                                                    cur_size);
1571
1572                                 skb->truesize += PAGE_SIZE;
1573                                 skb->data_len += cur_size;
1574                                 skb->len += cur_size;
1575                                 pkt_len -= cur_size;
1576                         }
1577
1578                         if (unlikely(pkt_len))
1579                                 DP_ERR(edev,
1580                                        "Mapped all BDs of jumbo, but still have %d bytes\n",
1581                                        pkt_len);
1582                 }
1583
1584                 skb->protocol = eth_type_trans(skb, edev->ndev);
1585
1586                 rx_hash = qede_get_rxhash(edev, fp_cqe->bitfields,
1587                                           fp_cqe->rss_hash, &rxhash_type);
1588
1589                 skb_set_hash(skb, rx_hash, rxhash_type);
1590
1591                 qede_set_skb_csum(skb, csum_flag);
1592
1593                 skb_record_rx_queue(skb, fp->rxq->rxq_id);
1594
1595                 qede_skb_receive(edev, fp, skb, le16_to_cpu(fp_cqe->vlan_tag));
1596 next_rx_only:
1597                 rx_pkt++;
1598
1599 next_cqe: /* don't consume bd rx buffer */
1600                 qed_chain_recycle_consumed(&rxq->rx_comp_ring);
1601                 sw_comp_cons = qed_chain_get_cons_idx(&rxq->rx_comp_ring);
1602                 /* CR TPA - revisit how to handle budget in TPA perhaps
1603                  * increase on "end"
1604                  */
1605                 if (rx_pkt == budget)
1606                         break;
1607         } /* repeat while sw_comp_cons != hw_comp_cons... */
1608
1609         /* Update producers */
1610         qede_update_rx_prod(edev, rxq);
1611
1612         rxq->rcv_pkts += rx_pkt;
1613
1614         return rx_pkt;
1615 }
1616
1617 static int qede_poll(struct napi_struct *napi, int budget)
1618 {
1619         struct qede_fastpath *fp = container_of(napi, struct qede_fastpath,
1620                                                 napi);
1621         struct qede_dev *edev = fp->edev;
1622         int rx_work_done = 0;
1623         u8 tc;
1624
1625         for (tc = 0; tc < edev->num_tc; tc++)
1626                 if (likely(fp->type & QEDE_FASTPATH_TX) &&
1627                     qede_txq_has_work(&fp->txqs[tc]))
1628                         qede_tx_int(edev, &fp->txqs[tc]);
1629
1630         rx_work_done = (likely(fp->type & QEDE_FASTPATH_RX) &&
1631                         qede_has_rx_work(fp->rxq)) ?
1632                         qede_rx_int(fp, budget) : 0;
1633         if (rx_work_done < budget) {
1634                 qed_sb_update_sb_idx(fp->sb_info);
1635                 /* *_has_*_work() reads the status block,
1636                  * thus we need to ensure that status block indices
1637                  * have been actually read (qed_sb_update_sb_idx)
1638                  * prior to this check (*_has_*_work) so that
1639                  * we won't write the "newer" value of the status block
1640                  * to HW (if there was a DMA right after
1641                  * qede_has_rx_work and if there is no rmb, the memory
1642                  * reading (qed_sb_update_sb_idx) may be postponed
1643                  * to right before *_ack_sb). In this case there
1644                  * will never be another interrupt until there is
1645                  * another update of the status block, while there
1646                  * is still unhandled work.
1647                  */
1648                 rmb();
1649
1650                 /* Fall out from the NAPI loop if needed */
1651                 if (!((likely(fp->type & QEDE_FASTPATH_RX) &&
1652                        qede_has_rx_work(fp->rxq)) ||
1653                       (likely(fp->type & QEDE_FASTPATH_TX) &&
1654                        qede_has_tx_work(fp)))) {
1655                         napi_complete(napi);
1656
1657                         /* Update and reenable interrupts */
1658                         qed_sb_ack(fp->sb_info, IGU_INT_ENABLE,
1659                                    1 /*update*/);
1660                 } else {
1661                         rx_work_done = budget;
1662                 }
1663         }
1664
1665         return rx_work_done;
1666 }
1667
1668 static irqreturn_t qede_msix_fp_int(int irq, void *fp_cookie)
1669 {
1670         struct qede_fastpath *fp = fp_cookie;
1671
1672         qed_sb_ack(fp->sb_info, IGU_INT_DISABLE, 0 /*do not update*/);
1673
1674         napi_schedule_irqoff(&fp->napi);
1675         return IRQ_HANDLED;
1676 }
1677
1678 /* -------------------------------------------------------------------------
1679  * END OF FAST-PATH
1680  * -------------------------------------------------------------------------
1681  */
1682
1683 static int qede_open(struct net_device *ndev);
1684 static int qede_close(struct net_device *ndev);
1685 static int qede_set_mac_addr(struct net_device *ndev, void *p);
1686 static void qede_set_rx_mode(struct net_device *ndev);
1687 static void qede_config_rx_mode(struct net_device *ndev);
1688
1689 static int qede_set_ucast_rx_mac(struct qede_dev *edev,
1690                                  enum qed_filter_xcast_params_type opcode,
1691                                  unsigned char mac[ETH_ALEN])
1692 {
1693         struct qed_filter_params filter_cmd;
1694
1695         memset(&filter_cmd, 0, sizeof(filter_cmd));
1696         filter_cmd.type = QED_FILTER_TYPE_UCAST;
1697         filter_cmd.filter.ucast.type = opcode;
1698         filter_cmd.filter.ucast.mac_valid = 1;
1699         ether_addr_copy(filter_cmd.filter.ucast.mac, mac);
1700
1701         return edev->ops->filter_config(edev->cdev, &filter_cmd);
1702 }
1703
1704 static int qede_set_ucast_rx_vlan(struct qede_dev *edev,
1705                                   enum qed_filter_xcast_params_type opcode,
1706                                   u16 vid)
1707 {
1708         struct qed_filter_params filter_cmd;
1709
1710         memset(&filter_cmd, 0, sizeof(filter_cmd));
1711         filter_cmd.type = QED_FILTER_TYPE_UCAST;
1712         filter_cmd.filter.ucast.type = opcode;
1713         filter_cmd.filter.ucast.vlan_valid = 1;
1714         filter_cmd.filter.ucast.vlan = vid;
1715
1716         return edev->ops->filter_config(edev->cdev, &filter_cmd);
1717 }
1718
1719 void qede_fill_by_demand_stats(struct qede_dev *edev)
1720 {
1721         struct qed_eth_stats stats;
1722
1723         edev->ops->get_vport_stats(edev->cdev, &stats);
1724         edev->stats.no_buff_discards = stats.no_buff_discards;
1725         edev->stats.packet_too_big_discard = stats.packet_too_big_discard;
1726         edev->stats.ttl0_discard = stats.ttl0_discard;
1727         edev->stats.rx_ucast_bytes = stats.rx_ucast_bytes;
1728         edev->stats.rx_mcast_bytes = stats.rx_mcast_bytes;
1729         edev->stats.rx_bcast_bytes = stats.rx_bcast_bytes;
1730         edev->stats.rx_ucast_pkts = stats.rx_ucast_pkts;
1731         edev->stats.rx_mcast_pkts = stats.rx_mcast_pkts;
1732         edev->stats.rx_bcast_pkts = stats.rx_bcast_pkts;
1733         edev->stats.mftag_filter_discards = stats.mftag_filter_discards;
1734         edev->stats.mac_filter_discards = stats.mac_filter_discards;
1735
1736         edev->stats.tx_ucast_bytes = stats.tx_ucast_bytes;
1737         edev->stats.tx_mcast_bytes = stats.tx_mcast_bytes;
1738         edev->stats.tx_bcast_bytes = stats.tx_bcast_bytes;
1739         edev->stats.tx_ucast_pkts = stats.tx_ucast_pkts;
1740         edev->stats.tx_mcast_pkts = stats.tx_mcast_pkts;
1741         edev->stats.tx_bcast_pkts = stats.tx_bcast_pkts;
1742         edev->stats.tx_err_drop_pkts = stats.tx_err_drop_pkts;
1743         edev->stats.coalesced_pkts = stats.tpa_coalesced_pkts;
1744         edev->stats.coalesced_events = stats.tpa_coalesced_events;
1745         edev->stats.coalesced_aborts_num = stats.tpa_aborts_num;
1746         edev->stats.non_coalesced_pkts = stats.tpa_not_coalesced_pkts;
1747         edev->stats.coalesced_bytes = stats.tpa_coalesced_bytes;
1748
1749         edev->stats.rx_64_byte_packets = stats.rx_64_byte_packets;
1750         edev->stats.rx_65_to_127_byte_packets = stats.rx_65_to_127_byte_packets;
1751         edev->stats.rx_128_to_255_byte_packets =
1752                                 stats.rx_128_to_255_byte_packets;
1753         edev->stats.rx_256_to_511_byte_packets =
1754                                 stats.rx_256_to_511_byte_packets;
1755         edev->stats.rx_512_to_1023_byte_packets =
1756                                 stats.rx_512_to_1023_byte_packets;
1757         edev->stats.rx_1024_to_1518_byte_packets =
1758                                 stats.rx_1024_to_1518_byte_packets;
1759         edev->stats.rx_1519_to_1522_byte_packets =
1760                                 stats.rx_1519_to_1522_byte_packets;
1761         edev->stats.rx_1519_to_2047_byte_packets =
1762                                 stats.rx_1519_to_2047_byte_packets;
1763         edev->stats.rx_2048_to_4095_byte_packets =
1764                                 stats.rx_2048_to_4095_byte_packets;
1765         edev->stats.rx_4096_to_9216_byte_packets =
1766                                 stats.rx_4096_to_9216_byte_packets;
1767         edev->stats.rx_9217_to_16383_byte_packets =
1768                                 stats.rx_9217_to_16383_byte_packets;
1769         edev->stats.rx_crc_errors = stats.rx_crc_errors;
1770         edev->stats.rx_mac_crtl_frames = stats.rx_mac_crtl_frames;
1771         edev->stats.rx_pause_frames = stats.rx_pause_frames;
1772         edev->stats.rx_pfc_frames = stats.rx_pfc_frames;
1773         edev->stats.rx_align_errors = stats.rx_align_errors;
1774         edev->stats.rx_carrier_errors = stats.rx_carrier_errors;
1775         edev->stats.rx_oversize_packets = stats.rx_oversize_packets;
1776         edev->stats.rx_jabbers = stats.rx_jabbers;
1777         edev->stats.rx_undersize_packets = stats.rx_undersize_packets;
1778         edev->stats.rx_fragments = stats.rx_fragments;
1779         edev->stats.tx_64_byte_packets = stats.tx_64_byte_packets;
1780         edev->stats.tx_65_to_127_byte_packets = stats.tx_65_to_127_byte_packets;
1781         edev->stats.tx_128_to_255_byte_packets =
1782                                 stats.tx_128_to_255_byte_packets;
1783         edev->stats.tx_256_to_511_byte_packets =
1784                                 stats.tx_256_to_511_byte_packets;
1785         edev->stats.tx_512_to_1023_byte_packets =
1786                                 stats.tx_512_to_1023_byte_packets;
1787         edev->stats.tx_1024_to_1518_byte_packets =
1788                                 stats.tx_1024_to_1518_byte_packets;
1789         edev->stats.tx_1519_to_2047_byte_packets =
1790                                 stats.tx_1519_to_2047_byte_packets;
1791         edev->stats.tx_2048_to_4095_byte_packets =
1792                                 stats.tx_2048_to_4095_byte_packets;
1793         edev->stats.tx_4096_to_9216_byte_packets =
1794                                 stats.tx_4096_to_9216_byte_packets;
1795         edev->stats.tx_9217_to_16383_byte_packets =
1796                                 stats.tx_9217_to_16383_byte_packets;
1797         edev->stats.tx_pause_frames = stats.tx_pause_frames;
1798         edev->stats.tx_pfc_frames = stats.tx_pfc_frames;
1799         edev->stats.tx_lpi_entry_count = stats.tx_lpi_entry_count;
1800         edev->stats.tx_total_collisions = stats.tx_total_collisions;
1801         edev->stats.brb_truncates = stats.brb_truncates;
1802         edev->stats.brb_discards = stats.brb_discards;
1803         edev->stats.tx_mac_ctrl_frames = stats.tx_mac_ctrl_frames;
1804 }
1805
1806 static
1807 struct rtnl_link_stats64 *qede_get_stats64(struct net_device *dev,
1808                                            struct rtnl_link_stats64 *stats)
1809 {
1810         struct qede_dev *edev = netdev_priv(dev);
1811
1812         qede_fill_by_demand_stats(edev);
1813
1814         stats->rx_packets = edev->stats.rx_ucast_pkts +
1815                             edev->stats.rx_mcast_pkts +
1816                             edev->stats.rx_bcast_pkts;
1817         stats->tx_packets = edev->stats.tx_ucast_pkts +
1818                             edev->stats.tx_mcast_pkts +
1819                             edev->stats.tx_bcast_pkts;
1820
1821         stats->rx_bytes = edev->stats.rx_ucast_bytes +
1822                           edev->stats.rx_mcast_bytes +
1823                           edev->stats.rx_bcast_bytes;
1824
1825         stats->tx_bytes = edev->stats.tx_ucast_bytes +
1826                           edev->stats.tx_mcast_bytes +
1827                           edev->stats.tx_bcast_bytes;
1828
1829         stats->tx_errors = edev->stats.tx_err_drop_pkts;
1830         stats->multicast = edev->stats.rx_mcast_pkts +
1831                            edev->stats.rx_bcast_pkts;
1832
1833         stats->rx_fifo_errors = edev->stats.no_buff_discards;
1834
1835         stats->collisions = edev->stats.tx_total_collisions;
1836         stats->rx_crc_errors = edev->stats.rx_crc_errors;
1837         stats->rx_frame_errors = edev->stats.rx_align_errors;
1838
1839         return stats;
1840 }
1841
1842 #ifdef CONFIG_QED_SRIOV
1843 static int qede_get_vf_config(struct net_device *dev, int vfidx,
1844                               struct ifla_vf_info *ivi)
1845 {
1846         struct qede_dev *edev = netdev_priv(dev);
1847
1848         if (!edev->ops)
1849                 return -EINVAL;
1850
1851         return edev->ops->iov->get_config(edev->cdev, vfidx, ivi);
1852 }
1853
1854 static int qede_set_vf_rate(struct net_device *dev, int vfidx,
1855                             int min_tx_rate, int max_tx_rate)
1856 {
1857         struct qede_dev *edev = netdev_priv(dev);
1858
1859         return edev->ops->iov->set_rate(edev->cdev, vfidx, min_tx_rate,
1860                                         max_tx_rate);
1861 }
1862
1863 static int qede_set_vf_spoofchk(struct net_device *dev, int vfidx, bool val)
1864 {
1865         struct qede_dev *edev = netdev_priv(dev);
1866
1867         if (!edev->ops)
1868                 return -EINVAL;
1869
1870         return edev->ops->iov->set_spoof(edev->cdev, vfidx, val);
1871 }
1872
1873 static int qede_set_vf_link_state(struct net_device *dev, int vfidx,
1874                                   int link_state)
1875 {
1876         struct qede_dev *edev = netdev_priv(dev);
1877
1878         if (!edev->ops)
1879                 return -EINVAL;
1880
1881         return edev->ops->iov->set_link_state(edev->cdev, vfidx, link_state);
1882 }
1883 #endif
1884
1885 static void qede_config_accept_any_vlan(struct qede_dev *edev, bool action)
1886 {
1887         struct qed_update_vport_params params;
1888         int rc;
1889
1890         /* Proceed only if action actually needs to be performed */
1891         if (edev->accept_any_vlan == action)
1892                 return;
1893
1894         memset(&params, 0, sizeof(params));
1895
1896         params.vport_id = 0;
1897         params.accept_any_vlan = action;
1898         params.update_accept_any_vlan_flg = 1;
1899
1900         rc = edev->ops->vport_update(edev->cdev, &params);
1901         if (rc) {
1902                 DP_ERR(edev, "Failed to %s accept-any-vlan\n",
1903                        action ? "enable" : "disable");
1904         } else {
1905                 DP_INFO(edev, "%s accept-any-vlan\n",
1906                         action ? "enabled" : "disabled");
1907                 edev->accept_any_vlan = action;
1908         }
1909 }
1910
1911 static int qede_vlan_rx_add_vid(struct net_device *dev, __be16 proto, u16 vid)
1912 {
1913         struct qede_dev *edev = netdev_priv(dev);
1914         struct qede_vlan *vlan, *tmp;
1915         int rc;
1916
1917         DP_VERBOSE(edev, NETIF_MSG_IFUP, "Adding vlan 0x%04x\n", vid);
1918
1919         vlan = kzalloc(sizeof(*vlan), GFP_KERNEL);
1920         if (!vlan) {
1921                 DP_INFO(edev, "Failed to allocate struct for vlan\n");
1922                 return -ENOMEM;
1923         }
1924         INIT_LIST_HEAD(&vlan->list);
1925         vlan->vid = vid;
1926         vlan->configured = false;
1927
1928         /* Verify vlan isn't already configured */
1929         list_for_each_entry(tmp, &edev->vlan_list, list) {
1930                 if (tmp->vid == vlan->vid) {
1931                         DP_VERBOSE(edev, (NETIF_MSG_IFUP | NETIF_MSG_IFDOWN),
1932                                    "vlan already configured\n");
1933                         kfree(vlan);
1934                         return -EEXIST;
1935                 }
1936         }
1937
1938         /* If interface is down, cache this VLAN ID and return */
1939         if (edev->state != QEDE_STATE_OPEN) {
1940                 DP_VERBOSE(edev, NETIF_MSG_IFDOWN,
1941                            "Interface is down, VLAN %d will be configured when interface is up\n",
1942                            vid);
1943                 if (vid != 0)
1944                         edev->non_configured_vlans++;
1945                 list_add(&vlan->list, &edev->vlan_list);
1946
1947                 return 0;
1948         }
1949
1950         /* Check for the filter limit.
1951          * Note - vlan0 has a reserved filter and can be added without
1952          * worrying about quota
1953          */
1954         if ((edev->configured_vlans < edev->dev_info.num_vlan_filters) ||
1955             (vlan->vid == 0)) {
1956                 rc = qede_set_ucast_rx_vlan(edev,
1957                                             QED_FILTER_XCAST_TYPE_ADD,
1958                                             vlan->vid);
1959                 if (rc) {
1960                         DP_ERR(edev, "Failed to configure VLAN %d\n",
1961                                vlan->vid);
1962                         kfree(vlan);
1963                         return -EINVAL;
1964                 }
1965                 vlan->configured = true;
1966
1967                 /* vlan0 filter isn't consuming out of our quota */
1968                 if (vlan->vid != 0)
1969                         edev->configured_vlans++;
1970         } else {
1971                 /* Out of quota; Activate accept-any-VLAN mode */
1972                 if (!edev->non_configured_vlans)
1973                         qede_config_accept_any_vlan(edev, true);
1974
1975                 edev->non_configured_vlans++;
1976         }
1977
1978         list_add(&vlan->list, &edev->vlan_list);
1979
1980         return 0;
1981 }
1982
1983 static void qede_del_vlan_from_list(struct qede_dev *edev,
1984                                     struct qede_vlan *vlan)
1985 {
1986         /* vlan0 filter isn't consuming out of our quota */
1987         if (vlan->vid != 0) {
1988                 if (vlan->configured)
1989                         edev->configured_vlans--;
1990                 else
1991                         edev->non_configured_vlans--;
1992         }
1993
1994         list_del(&vlan->list);
1995         kfree(vlan);
1996 }
1997
1998 static int qede_configure_vlan_filters(struct qede_dev *edev)
1999 {
2000         int rc = 0, real_rc = 0, accept_any_vlan = 0;
2001         struct qed_dev_eth_info *dev_info;
2002         struct qede_vlan *vlan = NULL;
2003
2004         if (list_empty(&edev->vlan_list))
2005                 return 0;
2006
2007         dev_info = &edev->dev_info;
2008
2009         /* Configure non-configured vlans */
2010         list_for_each_entry(vlan, &edev->vlan_list, list) {
2011                 if (vlan->configured)
2012                         continue;
2013
2014                 /* We have used all our credits, now enable accept_any_vlan */
2015                 if ((vlan->vid != 0) &&
2016                     (edev->configured_vlans == dev_info->num_vlan_filters)) {
2017                         accept_any_vlan = 1;
2018                         continue;
2019                 }
2020
2021                 DP_VERBOSE(edev, NETIF_MSG_IFUP, "Adding vlan %d\n", vlan->vid);
2022
2023                 rc = qede_set_ucast_rx_vlan(edev, QED_FILTER_XCAST_TYPE_ADD,
2024                                             vlan->vid);
2025                 if (rc) {
2026                         DP_ERR(edev, "Failed to configure VLAN %u\n",
2027                                vlan->vid);
2028                         real_rc = rc;
2029                         continue;
2030                 }
2031
2032                 vlan->configured = true;
2033                 /* vlan0 filter doesn't consume our VLAN filter's quota */
2034                 if (vlan->vid != 0) {
2035                         edev->non_configured_vlans--;
2036                         edev->configured_vlans++;
2037                 }
2038         }
2039
2040         /* enable accept_any_vlan mode if we have more VLANs than credits,
2041          * or remove accept_any_vlan mode if we've actually removed
2042          * a non-configured vlan, and all remaining vlans are truly configured.
2043          */
2044
2045         if (accept_any_vlan)
2046                 qede_config_accept_any_vlan(edev, true);
2047         else if (!edev->non_configured_vlans)
2048                 qede_config_accept_any_vlan(edev, false);
2049
2050         return real_rc;
2051 }
2052
2053 static int qede_vlan_rx_kill_vid(struct net_device *dev, __be16 proto, u16 vid)
2054 {
2055         struct qede_dev *edev = netdev_priv(dev);
2056         struct qede_vlan *vlan = NULL;
2057         int rc;
2058
2059         DP_VERBOSE(edev, NETIF_MSG_IFDOWN, "Removing vlan 0x%04x\n", vid);
2060
2061         /* Find whether entry exists */
2062         list_for_each_entry(vlan, &edev->vlan_list, list)
2063                 if (vlan->vid == vid)
2064                         break;
2065
2066         if (!vlan || (vlan->vid != vid)) {
2067                 DP_VERBOSE(edev, (NETIF_MSG_IFUP | NETIF_MSG_IFDOWN),
2068                            "Vlan isn't configured\n");
2069                 return 0;
2070         }
2071
2072         if (edev->state != QEDE_STATE_OPEN) {
2073                 /* As interface is already down, we don't have a VPORT
2074                  * instance to remove vlan filter. So just update vlan list
2075                  */
2076                 DP_VERBOSE(edev, NETIF_MSG_IFDOWN,
2077                            "Interface is down, removing VLAN from list only\n");
2078                 qede_del_vlan_from_list(edev, vlan);
2079                 return 0;
2080         }
2081
2082         /* Remove vlan */
2083         if (vlan->configured) {
2084                 rc = qede_set_ucast_rx_vlan(edev, QED_FILTER_XCAST_TYPE_DEL,
2085                                             vid);
2086                 if (rc) {
2087                         DP_ERR(edev, "Failed to remove VLAN %d\n", vid);
2088                         return -EINVAL;
2089                 }
2090         }
2091
2092         qede_del_vlan_from_list(edev, vlan);
2093
2094         /* We have removed a VLAN - try to see if we can
2095          * configure non-configured VLAN from the list.
2096          */
2097         rc = qede_configure_vlan_filters(edev);
2098
2099         return rc;
2100 }
2101
2102 static void qede_vlan_mark_nonconfigured(struct qede_dev *edev)
2103 {
2104         struct qede_vlan *vlan = NULL;
2105
2106         if (list_empty(&edev->vlan_list))
2107                 return;
2108
2109         list_for_each_entry(vlan, &edev->vlan_list, list) {
2110                 if (!vlan->configured)
2111                         continue;
2112
2113                 vlan->configured = false;
2114
2115                 /* vlan0 filter isn't consuming out of our quota */
2116                 if (vlan->vid != 0) {
2117                         edev->non_configured_vlans++;
2118                         edev->configured_vlans--;
2119                 }
2120
2121                 DP_VERBOSE(edev, NETIF_MSG_IFDOWN,
2122                            "marked vlan %d as non-configured\n", vlan->vid);
2123         }
2124
2125         edev->accept_any_vlan = false;
2126 }
2127
2128 static int qede_set_features(struct net_device *dev, netdev_features_t features)
2129 {
2130         struct qede_dev *edev = netdev_priv(dev);
2131         netdev_features_t changes = features ^ dev->features;
2132         bool need_reload = false;
2133
2134         /* No action needed if hardware GRO is disabled during driver load */
2135         if (changes & NETIF_F_GRO) {
2136                 if (dev->features & NETIF_F_GRO)
2137                         need_reload = !edev->gro_disable;
2138                 else
2139                         need_reload = edev->gro_disable;
2140         }
2141
2142         if (need_reload && netif_running(edev->ndev)) {
2143                 dev->features = features;
2144                 qede_reload(edev, NULL, NULL);
2145                 return 1;
2146         }
2147
2148         return 0;
2149 }
2150
2151 static void qede_udp_tunnel_add(struct net_device *dev,
2152                                 struct udp_tunnel_info *ti)
2153 {
2154         struct qede_dev *edev = netdev_priv(dev);
2155         u16 t_port = ntohs(ti->port);
2156
2157         switch (ti->type) {
2158         case UDP_TUNNEL_TYPE_VXLAN:
2159                 if (edev->vxlan_dst_port)
2160                         return;
2161
2162                 edev->vxlan_dst_port = t_port;
2163
2164                 DP_VERBOSE(edev, QED_MSG_DEBUG, "Added vxlan port=%d\n",
2165                            t_port);
2166
2167                 set_bit(QEDE_SP_VXLAN_PORT_CONFIG, &edev->sp_flags);
2168                 break;
2169         case UDP_TUNNEL_TYPE_GENEVE:
2170                 if (edev->geneve_dst_port)
2171                         return;
2172
2173                 edev->geneve_dst_port = t_port;
2174
2175                 DP_VERBOSE(edev, QED_MSG_DEBUG, "Added geneve port=%d\n",
2176                            t_port);
2177                 set_bit(QEDE_SP_GENEVE_PORT_CONFIG, &edev->sp_flags);
2178                 break;
2179         default:
2180                 return;
2181         }
2182
2183         schedule_delayed_work(&edev->sp_task, 0);
2184 }
2185
2186 static void qede_udp_tunnel_del(struct net_device *dev,
2187                                 struct udp_tunnel_info *ti)
2188 {
2189         struct qede_dev *edev = netdev_priv(dev);
2190         u16 t_port = ntohs(ti->port);
2191
2192         switch (ti->type) {
2193         case UDP_TUNNEL_TYPE_VXLAN:
2194                 if (t_port != edev->vxlan_dst_port)
2195                         return;
2196
2197                 edev->vxlan_dst_port = 0;
2198
2199                 DP_VERBOSE(edev, QED_MSG_DEBUG, "Deleted vxlan port=%d\n",
2200                            t_port);
2201
2202                 set_bit(QEDE_SP_VXLAN_PORT_CONFIG, &edev->sp_flags);
2203                 break;
2204         case UDP_TUNNEL_TYPE_GENEVE:
2205                 if (t_port != edev->geneve_dst_port)
2206                         return;
2207
2208                 edev->geneve_dst_port = 0;
2209
2210                 DP_VERBOSE(edev, QED_MSG_DEBUG, "Deleted geneve port=%d\n",
2211                            t_port);
2212                 set_bit(QEDE_SP_GENEVE_PORT_CONFIG, &edev->sp_flags);
2213                 break;
2214         default:
2215                 return;
2216         }
2217
2218         schedule_delayed_work(&edev->sp_task, 0);
2219 }
2220
2221 static const struct net_device_ops qede_netdev_ops = {
2222         .ndo_open = qede_open,
2223         .ndo_stop = qede_close,
2224         .ndo_start_xmit = qede_start_xmit,
2225         .ndo_set_rx_mode = qede_set_rx_mode,
2226         .ndo_set_mac_address = qede_set_mac_addr,
2227         .ndo_validate_addr = eth_validate_addr,
2228         .ndo_change_mtu = qede_change_mtu,
2229 #ifdef CONFIG_QED_SRIOV
2230         .ndo_set_vf_mac = qede_set_vf_mac,
2231         .ndo_set_vf_vlan = qede_set_vf_vlan,
2232 #endif
2233         .ndo_vlan_rx_add_vid = qede_vlan_rx_add_vid,
2234         .ndo_vlan_rx_kill_vid = qede_vlan_rx_kill_vid,
2235         .ndo_set_features = qede_set_features,
2236         .ndo_get_stats64 = qede_get_stats64,
2237 #ifdef CONFIG_QED_SRIOV
2238         .ndo_set_vf_link_state = qede_set_vf_link_state,
2239         .ndo_set_vf_spoofchk = qede_set_vf_spoofchk,
2240         .ndo_get_vf_config = qede_get_vf_config,
2241         .ndo_set_vf_rate = qede_set_vf_rate,
2242 #endif
2243         .ndo_udp_tunnel_add = qede_udp_tunnel_add,
2244         .ndo_udp_tunnel_del = qede_udp_tunnel_del,
2245 };
2246
2247 /* -------------------------------------------------------------------------
2248  * START OF PROBE / REMOVE
2249  * -------------------------------------------------------------------------
2250  */
2251
2252 static struct qede_dev *qede_alloc_etherdev(struct qed_dev *cdev,
2253                                             struct pci_dev *pdev,
2254                                             struct qed_dev_eth_info *info,
2255                                             u32 dp_module, u8 dp_level)
2256 {
2257         struct net_device *ndev;
2258         struct qede_dev *edev;
2259
2260         ndev = alloc_etherdev_mqs(sizeof(*edev),
2261                                   info->num_queues, info->num_queues);
2262         if (!ndev) {
2263                 pr_err("etherdev allocation failed\n");
2264                 return NULL;
2265         }
2266
2267         edev = netdev_priv(ndev);
2268         edev->ndev = ndev;
2269         edev->cdev = cdev;
2270         edev->pdev = pdev;
2271         edev->dp_module = dp_module;
2272         edev->dp_level = dp_level;
2273         edev->ops = qed_ops;
2274         edev->q_num_rx_buffers = NUM_RX_BDS_DEF;
2275         edev->q_num_tx_buffers = NUM_TX_BDS_DEF;
2276
2277         DP_INFO(edev, "Allocated netdev with %d tx queues and %d rx queues\n",
2278                 info->num_queues, info->num_queues);
2279
2280         SET_NETDEV_DEV(ndev, &pdev->dev);
2281
2282         memset(&edev->stats, 0, sizeof(edev->stats));
2283         memcpy(&edev->dev_info, info, sizeof(*info));
2284
2285         edev->num_tc = edev->dev_info.num_tc;
2286
2287         INIT_LIST_HEAD(&edev->vlan_list);
2288
2289         return edev;
2290 }
2291
2292 static void qede_init_ndev(struct qede_dev *edev)
2293 {
2294         struct net_device *ndev = edev->ndev;
2295         struct pci_dev *pdev = edev->pdev;
2296         u32 hw_features;
2297
2298         pci_set_drvdata(pdev, ndev);
2299
2300         ndev->mem_start = edev->dev_info.common.pci_mem_start;
2301         ndev->base_addr = ndev->mem_start;
2302         ndev->mem_end = edev->dev_info.common.pci_mem_end;
2303         ndev->irq = edev->dev_info.common.pci_irq;
2304
2305         ndev->watchdog_timeo = TX_TIMEOUT;
2306
2307         ndev->netdev_ops = &qede_netdev_ops;
2308
2309         qede_set_ethtool_ops(ndev);
2310
2311         /* user-changeble features */
2312         hw_features = NETIF_F_GRO | NETIF_F_SG |
2313                       NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2314                       NETIF_F_TSO | NETIF_F_TSO6;
2315
2316         /* Encap features*/
2317         hw_features |= NETIF_F_GSO_GRE | NETIF_F_GSO_UDP_TUNNEL |
2318                        NETIF_F_TSO_ECN;
2319         ndev->hw_enc_features = NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2320                                 NETIF_F_SG | NETIF_F_TSO | NETIF_F_TSO_ECN |
2321                                 NETIF_F_TSO6 | NETIF_F_GSO_GRE |
2322                                 NETIF_F_GSO_UDP_TUNNEL | NETIF_F_RXCSUM;
2323
2324         ndev->vlan_features = hw_features | NETIF_F_RXHASH | NETIF_F_RXCSUM |
2325                               NETIF_F_HIGHDMA;
2326         ndev->features = hw_features | NETIF_F_RXHASH | NETIF_F_RXCSUM |
2327                          NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_HIGHDMA |
2328                          NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_CTAG_TX;
2329
2330         ndev->hw_features = hw_features;
2331
2332         /* Set network device HW mac */
2333         ether_addr_copy(edev->ndev->dev_addr, edev->dev_info.common.hw_mac);
2334 }
2335
2336 /* This function converts from 32b param to two params of level and module
2337  * Input 32b decoding:
2338  * b31 - enable all NOTICE prints. NOTICE prints are for deviation from the
2339  * 'happy' flow, e.g. memory allocation failed.
2340  * b30 - enable all INFO prints. INFO prints are for major steps in the flow
2341  * and provide important parameters.
2342  * b29-b0 - per-module bitmap, where each bit enables VERBOSE prints of that
2343  * module. VERBOSE prints are for tracking the specific flow in low level.
2344  *
2345  * Notice that the level should be that of the lowest required logs.
2346  */
2347 void qede_config_debug(uint debug, u32 *p_dp_module, u8 *p_dp_level)
2348 {
2349         *p_dp_level = QED_LEVEL_NOTICE;
2350         *p_dp_module = 0;
2351
2352         if (debug & QED_LOG_VERBOSE_MASK) {
2353                 *p_dp_level = QED_LEVEL_VERBOSE;
2354                 *p_dp_module = (debug & 0x3FFFFFFF);
2355         } else if (debug & QED_LOG_INFO_MASK) {
2356                 *p_dp_level = QED_LEVEL_INFO;
2357         } else if (debug & QED_LOG_NOTICE_MASK) {
2358                 *p_dp_level = QED_LEVEL_NOTICE;
2359         }
2360 }
2361
2362 static void qede_free_fp_array(struct qede_dev *edev)
2363 {
2364         if (edev->fp_array) {
2365                 struct qede_fastpath *fp;
2366                 int i;
2367
2368                 for_each_queue(i) {
2369                         fp = &edev->fp_array[i];
2370
2371                         kfree(fp->sb_info);
2372                         kfree(fp->rxq);
2373                         kfree(fp->txqs);
2374                 }
2375                 kfree(edev->fp_array);
2376         }
2377
2378         edev->num_queues = 0;
2379         edev->fp_num_tx = 0;
2380         edev->fp_num_rx = 0;
2381 }
2382
2383 static int qede_alloc_fp_array(struct qede_dev *edev)
2384 {
2385         u8 fp_combined, fp_rx = edev->fp_num_rx;
2386         struct qede_fastpath *fp;
2387         int i;
2388
2389         edev->fp_array = kcalloc(QEDE_QUEUE_CNT(edev),
2390                                  sizeof(*edev->fp_array), GFP_KERNEL);
2391         if (!edev->fp_array) {
2392                 DP_NOTICE(edev, "fp array allocation failed\n");
2393                 goto err;
2394         }
2395
2396         fp_combined = QEDE_QUEUE_CNT(edev) - fp_rx - edev->fp_num_tx;
2397
2398         /* Allocate the FP elements for Rx queues followed by combined and then
2399          * the Tx. This ordering should be maintained so that the respective
2400          * queues (Rx or Tx) will be together in the fastpath array and the
2401          * associated ids will be sequential.
2402          */
2403         for_each_queue(i) {
2404                 fp = &edev->fp_array[i];
2405
2406                 fp->sb_info = kcalloc(1, sizeof(*fp->sb_info), GFP_KERNEL);
2407                 if (!fp->sb_info) {
2408                         DP_NOTICE(edev, "sb info struct allocation failed\n");
2409                         goto err;
2410                 }
2411
2412                 if (fp_rx) {
2413                         fp->type = QEDE_FASTPATH_RX;
2414                         fp_rx--;
2415                 } else if (fp_combined) {
2416                         fp->type = QEDE_FASTPATH_COMBINED;
2417                         fp_combined--;
2418                 } else {
2419                         fp->type = QEDE_FASTPATH_TX;
2420                 }
2421
2422                 if (fp->type & QEDE_FASTPATH_TX) {
2423                         fp->txqs = kcalloc(edev->num_tc, sizeof(*fp->txqs),
2424                                            GFP_KERNEL);
2425                         if (!fp->txqs) {
2426                                 DP_NOTICE(edev,
2427                                           "TXQ array allocation failed\n");
2428                                 goto err;
2429                         }
2430                 }
2431
2432                 if (fp->type & QEDE_FASTPATH_RX) {
2433                         fp->rxq = kcalloc(1, sizeof(*fp->rxq), GFP_KERNEL);
2434                         if (!fp->rxq) {
2435                                 DP_NOTICE(edev,
2436                                           "RXQ struct allocation failed\n");
2437                                 goto err;
2438                         }
2439                 }
2440         }
2441
2442         return 0;
2443 err:
2444         qede_free_fp_array(edev);
2445         return -ENOMEM;
2446 }
2447
2448 static void qede_sp_task(struct work_struct *work)
2449 {
2450         struct qede_dev *edev = container_of(work, struct qede_dev,
2451                                              sp_task.work);
2452         struct qed_dev *cdev = edev->cdev;
2453
2454         mutex_lock(&edev->qede_lock);
2455
2456         if (edev->state == QEDE_STATE_OPEN) {
2457                 if (test_and_clear_bit(QEDE_SP_RX_MODE, &edev->sp_flags))
2458                         qede_config_rx_mode(edev->ndev);
2459         }
2460
2461         if (test_and_clear_bit(QEDE_SP_VXLAN_PORT_CONFIG, &edev->sp_flags)) {
2462                 struct qed_tunn_params tunn_params;
2463
2464                 memset(&tunn_params, 0, sizeof(tunn_params));
2465                 tunn_params.update_vxlan_port = 1;
2466                 tunn_params.vxlan_port = edev->vxlan_dst_port;
2467                 qed_ops->tunn_config(cdev, &tunn_params);
2468         }
2469
2470         if (test_and_clear_bit(QEDE_SP_GENEVE_PORT_CONFIG, &edev->sp_flags)) {
2471                 struct qed_tunn_params tunn_params;
2472
2473                 memset(&tunn_params, 0, sizeof(tunn_params));
2474                 tunn_params.update_geneve_port = 1;
2475                 tunn_params.geneve_port = edev->geneve_dst_port;
2476                 qed_ops->tunn_config(cdev, &tunn_params);
2477         }
2478
2479         mutex_unlock(&edev->qede_lock);
2480 }
2481
2482 static void qede_update_pf_params(struct qed_dev *cdev)
2483 {
2484         struct qed_pf_params pf_params;
2485
2486         /* 64 rx + 64 tx */
2487         memset(&pf_params, 0, sizeof(struct qed_pf_params));
2488         pf_params.eth_pf_params.num_cons = 128;
2489         qed_ops->common->update_pf_params(cdev, &pf_params);
2490 }
2491
2492 enum qede_probe_mode {
2493         QEDE_PROBE_NORMAL,
2494 };
2495
2496 static int __qede_probe(struct pci_dev *pdev, u32 dp_module, u8 dp_level,
2497                         bool is_vf, enum qede_probe_mode mode)
2498 {
2499         struct qed_probe_params probe_params;
2500         struct qed_slowpath_params sp_params;
2501         struct qed_dev_eth_info dev_info;
2502         struct qede_dev *edev;
2503         struct qed_dev *cdev;
2504         int rc;
2505
2506         if (unlikely(dp_level & QED_LEVEL_INFO))
2507                 pr_notice("Starting qede probe\n");
2508
2509         memset(&probe_params, 0, sizeof(probe_params));
2510         probe_params.protocol = QED_PROTOCOL_ETH;
2511         probe_params.dp_module = dp_module;
2512         probe_params.dp_level = dp_level;
2513         probe_params.is_vf = is_vf;
2514         cdev = qed_ops->common->probe(pdev, &probe_params);
2515         if (!cdev) {
2516                 rc = -ENODEV;
2517                 goto err0;
2518         }
2519
2520         qede_update_pf_params(cdev);
2521
2522         /* Start the Slowpath-process */
2523         memset(&sp_params, 0, sizeof(sp_params));
2524         sp_params.int_mode = QED_INT_MODE_MSIX;
2525         sp_params.drv_major = QEDE_MAJOR_VERSION;
2526         sp_params.drv_minor = QEDE_MINOR_VERSION;
2527         sp_params.drv_rev = QEDE_REVISION_VERSION;
2528         sp_params.drv_eng = QEDE_ENGINEERING_VERSION;
2529         strlcpy(sp_params.name, "qede LAN", QED_DRV_VER_STR_SIZE);
2530         rc = qed_ops->common->slowpath_start(cdev, &sp_params);
2531         if (rc) {
2532                 pr_notice("Cannot start slowpath\n");
2533                 goto err1;
2534         }
2535
2536         /* Learn information crucial for qede to progress */
2537         rc = qed_ops->fill_dev_info(cdev, &dev_info);
2538         if (rc)
2539                 goto err2;
2540
2541         edev = qede_alloc_etherdev(cdev, pdev, &dev_info, dp_module,
2542                                    dp_level);
2543         if (!edev) {
2544                 rc = -ENOMEM;
2545                 goto err2;
2546         }
2547
2548         if (is_vf)
2549                 edev->flags |= QEDE_FLAG_IS_VF;
2550
2551         qede_init_ndev(edev);
2552
2553         rc = qede_roce_dev_add(edev);
2554         if (rc)
2555                 goto err3;
2556
2557         rc = register_netdev(edev->ndev);
2558         if (rc) {
2559                 DP_NOTICE(edev, "Cannot register net-device\n");
2560                 goto err4;
2561         }
2562
2563         edev->ops->common->set_id(cdev, edev->ndev->name, DRV_MODULE_VERSION);
2564
2565         edev->ops->register_ops(cdev, &qede_ll_ops, edev);
2566
2567 #ifdef CONFIG_DCB
2568         if (!IS_VF(edev))
2569                 qede_set_dcbnl_ops(edev->ndev);
2570 #endif
2571
2572         INIT_DELAYED_WORK(&edev->sp_task, qede_sp_task);
2573         mutex_init(&edev->qede_lock);
2574         edev->rx_copybreak = QEDE_RX_HDR_SIZE;
2575
2576         DP_INFO(edev, "Ending successfully qede probe\n");
2577
2578         return 0;
2579
2580 err4:
2581         qede_roce_dev_remove(edev);
2582 err3:
2583         free_netdev(edev->ndev);
2584 err2:
2585         qed_ops->common->slowpath_stop(cdev);
2586 err1:
2587         qed_ops->common->remove(cdev);
2588 err0:
2589         return rc;
2590 }
2591
2592 static int qede_probe(struct pci_dev *pdev, const struct pci_device_id *id)
2593 {
2594         bool is_vf = false;
2595         u32 dp_module = 0;
2596         u8 dp_level = 0;
2597
2598         switch ((enum qede_pci_private)id->driver_data) {
2599         case QEDE_PRIVATE_VF:
2600                 if (debug & QED_LOG_VERBOSE_MASK)
2601                         dev_err(&pdev->dev, "Probing a VF\n");
2602                 is_vf = true;
2603                 break;
2604         default:
2605                 if (debug & QED_LOG_VERBOSE_MASK)
2606                         dev_err(&pdev->dev, "Probing a PF\n");
2607         }
2608
2609         qede_config_debug(debug, &dp_module, &dp_level);
2610
2611         return __qede_probe(pdev, dp_module, dp_level, is_vf,
2612                             QEDE_PROBE_NORMAL);
2613 }
2614
2615 enum qede_remove_mode {
2616         QEDE_REMOVE_NORMAL,
2617 };
2618
2619 static void __qede_remove(struct pci_dev *pdev, enum qede_remove_mode mode)
2620 {
2621         struct net_device *ndev = pci_get_drvdata(pdev);
2622         struct qede_dev *edev = netdev_priv(ndev);
2623         struct qed_dev *cdev = edev->cdev;
2624
2625         DP_INFO(edev, "Starting qede_remove\n");
2626
2627         cancel_delayed_work_sync(&edev->sp_task);
2628
2629         unregister_netdev(ndev);
2630
2631         qede_roce_dev_remove(edev);
2632
2633         edev->ops->common->set_power_state(cdev, PCI_D0);
2634
2635         pci_set_drvdata(pdev, NULL);
2636
2637         free_netdev(ndev);
2638
2639         /* Use global ops since we've freed edev */
2640         qed_ops->common->slowpath_stop(cdev);
2641         qed_ops->common->remove(cdev);
2642
2643         dev_info(&pdev->dev, "Ending qede_remove successfully\n");
2644 }
2645
2646 static void qede_remove(struct pci_dev *pdev)
2647 {
2648         __qede_remove(pdev, QEDE_REMOVE_NORMAL);
2649 }
2650
2651 /* -------------------------------------------------------------------------
2652  * START OF LOAD / UNLOAD
2653  * -------------------------------------------------------------------------
2654  */
2655
2656 static int qede_set_num_queues(struct qede_dev *edev)
2657 {
2658         int rc;
2659         u16 rss_num;
2660
2661         /* Setup queues according to possible resources*/
2662         if (edev->req_queues)
2663                 rss_num = edev->req_queues;
2664         else
2665                 rss_num = netif_get_num_default_rss_queues() *
2666                           edev->dev_info.common.num_hwfns;
2667
2668         rss_num = min_t(u16, QEDE_MAX_RSS_CNT(edev), rss_num);
2669
2670         rc = edev->ops->common->set_fp_int(edev->cdev, rss_num);
2671         if (rc > 0) {
2672                 /* Managed to request interrupts for our queues */
2673                 edev->num_queues = rc;
2674                 DP_INFO(edev, "Managed %d [of %d] RSS queues\n",
2675                         QEDE_QUEUE_CNT(edev), rss_num);
2676                 rc = 0;
2677         }
2678
2679         edev->fp_num_tx = edev->req_num_tx;
2680         edev->fp_num_rx = edev->req_num_rx;
2681
2682         return rc;
2683 }
2684
2685 static void qede_free_mem_sb(struct qede_dev *edev,
2686                              struct qed_sb_info *sb_info)
2687 {
2688         if (sb_info->sb_virt)
2689                 dma_free_coherent(&edev->pdev->dev, sizeof(*sb_info->sb_virt),
2690                                   (void *)sb_info->sb_virt, sb_info->sb_phys);
2691 }
2692
2693 /* This function allocates fast-path status block memory */
2694 static int qede_alloc_mem_sb(struct qede_dev *edev,
2695                              struct qed_sb_info *sb_info, u16 sb_id)
2696 {
2697         struct status_block *sb_virt;
2698         dma_addr_t sb_phys;
2699         int rc;
2700
2701         sb_virt = dma_alloc_coherent(&edev->pdev->dev,
2702                                      sizeof(*sb_virt), &sb_phys, GFP_KERNEL);
2703         if (!sb_virt) {
2704                 DP_ERR(edev, "Status block allocation failed\n");
2705                 return -ENOMEM;
2706         }
2707
2708         rc = edev->ops->common->sb_init(edev->cdev, sb_info,
2709                                         sb_virt, sb_phys, sb_id,
2710                                         QED_SB_TYPE_L2_QUEUE);
2711         if (rc) {
2712                 DP_ERR(edev, "Status block initialization failed\n");
2713                 dma_free_coherent(&edev->pdev->dev, sizeof(*sb_virt),
2714                                   sb_virt, sb_phys);
2715                 return rc;
2716         }
2717
2718         return 0;
2719 }
2720
2721 static void qede_free_rx_buffers(struct qede_dev *edev,
2722                                  struct qede_rx_queue *rxq)
2723 {
2724         u16 i;
2725
2726         for (i = rxq->sw_rx_cons; i != rxq->sw_rx_prod; i++) {
2727                 struct sw_rx_data *rx_buf;
2728                 struct page *data;
2729
2730                 rx_buf = &rxq->sw_rx_ring[i & NUM_RX_BDS_MAX];
2731                 data = rx_buf->data;
2732
2733                 dma_unmap_page(&edev->pdev->dev,
2734                                rx_buf->mapping, PAGE_SIZE, DMA_FROM_DEVICE);
2735
2736                 rx_buf->data = NULL;
2737                 __free_page(data);
2738         }
2739 }
2740
2741 static void qede_free_sge_mem(struct qede_dev *edev, struct qede_rx_queue *rxq)
2742 {
2743         int i;
2744
2745         if (edev->gro_disable)
2746                 return;
2747
2748         for (i = 0; i < ETH_TPA_MAX_AGGS_NUM; i++) {
2749                 struct qede_agg_info *tpa_info = &rxq->tpa_info[i];
2750                 struct sw_rx_data *replace_buf = &tpa_info->replace_buf;
2751
2752                 if (replace_buf->data) {
2753                         dma_unmap_page(&edev->pdev->dev,
2754                                        replace_buf->mapping,
2755                                        PAGE_SIZE, DMA_FROM_DEVICE);
2756                         __free_page(replace_buf->data);
2757                 }
2758         }
2759 }
2760
2761 static void qede_free_mem_rxq(struct qede_dev *edev, struct qede_rx_queue *rxq)
2762 {
2763         qede_free_sge_mem(edev, rxq);
2764
2765         /* Free rx buffers */
2766         qede_free_rx_buffers(edev, rxq);
2767
2768         /* Free the parallel SW ring */
2769         kfree(rxq->sw_rx_ring);
2770
2771         /* Free the real RQ ring used by FW */
2772         edev->ops->common->chain_free(edev->cdev, &rxq->rx_bd_ring);
2773         edev->ops->common->chain_free(edev->cdev, &rxq->rx_comp_ring);
2774 }
2775
2776 static int qede_alloc_rx_buffer(struct qede_dev *edev,
2777                                 struct qede_rx_queue *rxq)
2778 {
2779         struct sw_rx_data *sw_rx_data;
2780         struct eth_rx_bd *rx_bd;
2781         dma_addr_t mapping;
2782         struct page *data;
2783
2784         data = alloc_pages(GFP_ATOMIC, 0);
2785         if (unlikely(!data)) {
2786                 DP_NOTICE(edev, "Failed to allocate Rx data [page]\n");
2787                 return -ENOMEM;
2788         }
2789
2790         /* Map the entire page as it would be used
2791          * for multiple RX buffer segment size mapping.
2792          */
2793         mapping = dma_map_page(&edev->pdev->dev, data, 0,
2794                                PAGE_SIZE, DMA_FROM_DEVICE);
2795         if (unlikely(dma_mapping_error(&edev->pdev->dev, mapping))) {
2796                 __free_page(data);
2797                 DP_NOTICE(edev, "Failed to map Rx buffer\n");
2798                 return -ENOMEM;
2799         }
2800
2801         sw_rx_data = &rxq->sw_rx_ring[rxq->sw_rx_prod & NUM_RX_BDS_MAX];
2802         sw_rx_data->page_offset = 0;
2803         sw_rx_data->data = data;
2804         sw_rx_data->mapping = mapping;
2805
2806         /* Advance PROD and get BD pointer */
2807         rx_bd = (struct eth_rx_bd *)qed_chain_produce(&rxq->rx_bd_ring);
2808         WARN_ON(!rx_bd);
2809         rx_bd->addr.hi = cpu_to_le32(upper_32_bits(mapping));
2810         rx_bd->addr.lo = cpu_to_le32(lower_32_bits(mapping));
2811
2812         rxq->sw_rx_prod++;
2813
2814         return 0;
2815 }
2816
2817 static int qede_alloc_sge_mem(struct qede_dev *edev, struct qede_rx_queue *rxq)
2818 {
2819         dma_addr_t mapping;
2820         int i;
2821
2822         if (edev->gro_disable)
2823                 return 0;
2824
2825         if (edev->ndev->mtu > PAGE_SIZE) {
2826                 edev->gro_disable = 1;
2827                 return 0;
2828         }
2829
2830         for (i = 0; i < ETH_TPA_MAX_AGGS_NUM; i++) {
2831                 struct qede_agg_info *tpa_info = &rxq->tpa_info[i];
2832                 struct sw_rx_data *replace_buf = &tpa_info->replace_buf;
2833
2834                 replace_buf->data = alloc_pages(GFP_ATOMIC, 0);
2835                 if (unlikely(!replace_buf->data)) {
2836                         DP_NOTICE(edev,
2837                                   "Failed to allocate TPA skb pool [replacement buffer]\n");
2838                         goto err;
2839                 }
2840
2841                 mapping = dma_map_page(&edev->pdev->dev, replace_buf->data, 0,
2842                                        rxq->rx_buf_size, DMA_FROM_DEVICE);
2843                 if (unlikely(dma_mapping_error(&edev->pdev->dev, mapping))) {
2844                         DP_NOTICE(edev,
2845                                   "Failed to map TPA replacement buffer\n");
2846                         goto err;
2847                 }
2848
2849                 replace_buf->mapping = mapping;
2850                 tpa_info->replace_buf.page_offset = 0;
2851
2852                 tpa_info->replace_buf_mapping = mapping;
2853                 tpa_info->agg_state = QEDE_AGG_STATE_NONE;
2854         }
2855
2856         return 0;
2857 err:
2858         qede_free_sge_mem(edev, rxq);
2859         edev->gro_disable = 1;
2860         return -ENOMEM;
2861 }
2862
2863 /* This function allocates all memory needed per Rx queue */
2864 static int qede_alloc_mem_rxq(struct qede_dev *edev, struct qede_rx_queue *rxq)
2865 {
2866         int i, rc, size;
2867
2868         rxq->num_rx_buffers = edev->q_num_rx_buffers;
2869
2870         rxq->rx_buf_size = NET_IP_ALIGN + ETH_OVERHEAD + edev->ndev->mtu;
2871
2872         if (rxq->rx_buf_size > PAGE_SIZE)
2873                 rxq->rx_buf_size = PAGE_SIZE;
2874
2875         /* Segment size to spilt a page in multiple equal parts */
2876         rxq->rx_buf_seg_size = roundup_pow_of_two(rxq->rx_buf_size);
2877
2878         /* Allocate the parallel driver ring for Rx buffers */
2879         size = sizeof(*rxq->sw_rx_ring) * RX_RING_SIZE;
2880         rxq->sw_rx_ring = kzalloc(size, GFP_KERNEL);
2881         if (!rxq->sw_rx_ring) {
2882                 DP_ERR(edev, "Rx buffers ring allocation failed\n");
2883                 rc = -ENOMEM;
2884                 goto err;
2885         }
2886
2887         /* Allocate FW Rx ring  */
2888         rc = edev->ops->common->chain_alloc(edev->cdev,
2889                                             QED_CHAIN_USE_TO_CONSUME_PRODUCE,
2890                                             QED_CHAIN_MODE_NEXT_PTR,
2891                                             QED_CHAIN_CNT_TYPE_U16,
2892                                             RX_RING_SIZE,
2893                                             sizeof(struct eth_rx_bd),
2894                                             &rxq->rx_bd_ring);
2895
2896         if (rc)
2897                 goto err;
2898
2899         /* Allocate FW completion ring */
2900         rc = edev->ops->common->chain_alloc(edev->cdev,
2901                                             QED_CHAIN_USE_TO_CONSUME,
2902                                             QED_CHAIN_MODE_PBL,
2903                                             QED_CHAIN_CNT_TYPE_U16,
2904                                             RX_RING_SIZE,
2905                                             sizeof(union eth_rx_cqe),
2906                                             &rxq->rx_comp_ring);
2907         if (rc)
2908                 goto err;
2909
2910         /* Allocate buffers for the Rx ring */
2911         for (i = 0; i < rxq->num_rx_buffers; i++) {
2912                 rc = qede_alloc_rx_buffer(edev, rxq);
2913                 if (rc) {
2914                         DP_ERR(edev,
2915                                "Rx buffers allocation failed at index %d\n", i);
2916                         goto err;
2917                 }
2918         }
2919
2920         rc = qede_alloc_sge_mem(edev, rxq);
2921 err:
2922         return rc;
2923 }
2924
2925 static void qede_free_mem_txq(struct qede_dev *edev, struct qede_tx_queue *txq)
2926 {
2927         /* Free the parallel SW ring */
2928         kfree(txq->sw_tx_ring);
2929
2930         /* Free the real RQ ring used by FW */
2931         edev->ops->common->chain_free(edev->cdev, &txq->tx_pbl);
2932 }
2933
2934 /* This function allocates all memory needed per Tx queue */
2935 static int qede_alloc_mem_txq(struct qede_dev *edev, struct qede_tx_queue *txq)
2936 {
2937         int size, rc;
2938         union eth_tx_bd_types *p_virt;
2939
2940         txq->num_tx_buffers = edev->q_num_tx_buffers;
2941
2942         /* Allocate the parallel driver ring for Tx buffers */
2943         size = sizeof(*txq->sw_tx_ring) * TX_RING_SIZE;
2944         txq->sw_tx_ring = kzalloc(size, GFP_KERNEL);
2945         if (!txq->sw_tx_ring) {
2946                 DP_NOTICE(edev, "Tx buffers ring allocation failed\n");
2947                 goto err;
2948         }
2949
2950         rc = edev->ops->common->chain_alloc(edev->cdev,
2951                                             QED_CHAIN_USE_TO_CONSUME_PRODUCE,
2952                                             QED_CHAIN_MODE_PBL,
2953                                             QED_CHAIN_CNT_TYPE_U16,
2954                                             TX_RING_SIZE,
2955                                             sizeof(*p_virt), &txq->tx_pbl);
2956         if (rc)
2957                 goto err;
2958
2959         return 0;
2960
2961 err:
2962         qede_free_mem_txq(edev, txq);
2963         return -ENOMEM;
2964 }
2965
2966 /* This function frees all memory of a single fp */
2967 static void qede_free_mem_fp(struct qede_dev *edev, struct qede_fastpath *fp)
2968 {
2969         int tc;
2970
2971         qede_free_mem_sb(edev, fp->sb_info);
2972
2973         if (fp->type & QEDE_FASTPATH_RX)
2974                 qede_free_mem_rxq(edev, fp->rxq);
2975
2976         if (fp->type & QEDE_FASTPATH_TX)
2977                 for (tc = 0; tc < edev->num_tc; tc++)
2978                         qede_free_mem_txq(edev, &fp->txqs[tc]);
2979 }
2980
2981 /* This function allocates all memory needed for a single fp (i.e. an entity
2982  * which contains status block, one rx queue and/or multiple per-TC tx queues.
2983  */
2984 static int qede_alloc_mem_fp(struct qede_dev *edev, struct qede_fastpath *fp)
2985 {
2986         int rc, tc;
2987
2988         rc = qede_alloc_mem_sb(edev, fp->sb_info, fp->id);
2989         if (rc)
2990                 goto err;
2991
2992         if (fp->type & QEDE_FASTPATH_RX) {
2993                 rc = qede_alloc_mem_rxq(edev, fp->rxq);
2994                 if (rc)
2995                         goto err;
2996         }
2997
2998         if (fp->type & QEDE_FASTPATH_TX) {
2999                 for (tc = 0; tc < edev->num_tc; tc++) {
3000                         rc = qede_alloc_mem_txq(edev, &fp->txqs[tc]);
3001                         if (rc)
3002                                 goto err;
3003                 }
3004         }
3005
3006         return 0;
3007 err:
3008         return rc;
3009 }
3010
3011 static void qede_free_mem_load(struct qede_dev *edev)
3012 {
3013         int i;
3014
3015         for_each_queue(i) {
3016                 struct qede_fastpath *fp = &edev->fp_array[i];
3017
3018                 qede_free_mem_fp(edev, fp);
3019         }
3020 }
3021
3022 /* This function allocates all qede memory at NIC load. */
3023 static int qede_alloc_mem_load(struct qede_dev *edev)
3024 {
3025         int rc = 0, queue_id;
3026
3027         for (queue_id = 0; queue_id < QEDE_QUEUE_CNT(edev); queue_id++) {
3028                 struct qede_fastpath *fp = &edev->fp_array[queue_id];
3029
3030                 rc = qede_alloc_mem_fp(edev, fp);
3031                 if (rc) {
3032                         DP_ERR(edev,
3033                                "Failed to allocate memory for fastpath - rss id = %d\n",
3034                                queue_id);
3035                         qede_free_mem_load(edev);
3036                         return rc;
3037                 }
3038         }
3039
3040         return 0;
3041 }
3042
3043 /* This function inits fp content and resets the SB, RXQ and TXQ structures */
3044 static void qede_init_fp(struct qede_dev *edev)
3045 {
3046         int queue_id, rxq_index = 0, txq_index = 0, tc;
3047         struct qede_fastpath *fp;
3048
3049         for_each_queue(queue_id) {
3050                 fp = &edev->fp_array[queue_id];
3051
3052                 fp->edev = edev;
3053                 fp->id = queue_id;
3054
3055                 memset((void *)&fp->napi, 0, sizeof(fp->napi));
3056
3057                 memset((void *)fp->sb_info, 0, sizeof(*fp->sb_info));
3058
3059                 if (fp->type & QEDE_FASTPATH_RX) {
3060                         memset((void *)fp->rxq, 0, sizeof(*fp->rxq));
3061                         fp->rxq->rxq_id = rxq_index++;
3062                 }
3063
3064                 if (fp->type & QEDE_FASTPATH_TX) {
3065                         memset((void *)fp->txqs, 0,
3066                                (edev->num_tc * sizeof(*fp->txqs)));
3067                         for (tc = 0; tc < edev->num_tc; tc++) {
3068                                 fp->txqs[tc].index = txq_index +
3069                                     tc * QEDE_TSS_COUNT(edev);
3070                                 if (edev->dev_info.is_legacy)
3071                                         fp->txqs[tc].is_legacy = true;
3072                         }
3073                         txq_index++;
3074                 }
3075
3076                 snprintf(fp->name, sizeof(fp->name), "%s-fp-%d",
3077                          edev->ndev->name, queue_id);
3078         }
3079
3080         edev->gro_disable = !(edev->ndev->features & NETIF_F_GRO);
3081 }
3082
3083 static int qede_set_real_num_queues(struct qede_dev *edev)
3084 {
3085         int rc = 0;
3086
3087         rc = netif_set_real_num_tx_queues(edev->ndev, QEDE_TSS_COUNT(edev));
3088         if (rc) {
3089                 DP_NOTICE(edev, "Failed to set real number of Tx queues\n");
3090                 return rc;
3091         }
3092
3093         rc = netif_set_real_num_rx_queues(edev->ndev, QEDE_RSS_COUNT(edev));
3094         if (rc) {
3095                 DP_NOTICE(edev, "Failed to set real number of Rx queues\n");
3096                 return rc;
3097         }
3098
3099         return 0;
3100 }
3101
3102 static void qede_napi_disable_remove(struct qede_dev *edev)
3103 {
3104         int i;
3105
3106         for_each_queue(i) {
3107                 napi_disable(&edev->fp_array[i].napi);
3108
3109                 netif_napi_del(&edev->fp_array[i].napi);
3110         }
3111 }
3112
3113 static void qede_napi_add_enable(struct qede_dev *edev)
3114 {
3115         int i;
3116
3117         /* Add NAPI objects */
3118         for_each_queue(i) {
3119                 netif_napi_add(edev->ndev, &edev->fp_array[i].napi,
3120                                qede_poll, NAPI_POLL_WEIGHT);
3121                 napi_enable(&edev->fp_array[i].napi);
3122         }
3123 }
3124
3125 static void qede_sync_free_irqs(struct qede_dev *edev)
3126 {
3127         int i;
3128
3129         for (i = 0; i < edev->int_info.used_cnt; i++) {
3130                 if (edev->int_info.msix_cnt) {
3131                         synchronize_irq(edev->int_info.msix[i].vector);
3132                         free_irq(edev->int_info.msix[i].vector,
3133                                  &edev->fp_array[i]);
3134                 } else {
3135                         edev->ops->common->simd_handler_clean(edev->cdev, i);
3136                 }
3137         }
3138
3139         edev->int_info.used_cnt = 0;
3140 }
3141
3142 static int qede_req_msix_irqs(struct qede_dev *edev)
3143 {
3144         int i, rc;
3145
3146         /* Sanitize number of interrupts == number of prepared RSS queues */
3147         if (QEDE_QUEUE_CNT(edev) > edev->int_info.msix_cnt) {
3148                 DP_ERR(edev,
3149                        "Interrupt mismatch: %d RSS queues > %d MSI-x vectors\n",
3150                        QEDE_QUEUE_CNT(edev), edev->int_info.msix_cnt);
3151                 return -EINVAL;
3152         }
3153
3154         for (i = 0; i < QEDE_QUEUE_CNT(edev); i++) {
3155                 rc = request_irq(edev->int_info.msix[i].vector,
3156                                  qede_msix_fp_int, 0, edev->fp_array[i].name,
3157                                  &edev->fp_array[i]);
3158                 if (rc) {
3159                         DP_ERR(edev, "Request fp %d irq failed\n", i);
3160                         qede_sync_free_irqs(edev);
3161                         return rc;
3162                 }
3163                 DP_VERBOSE(edev, NETIF_MSG_INTR,
3164                            "Requested fp irq for %s [entry %d]. Cookie is at %p\n",
3165                            edev->fp_array[i].name, i,
3166                            &edev->fp_array[i]);
3167                 edev->int_info.used_cnt++;
3168         }
3169
3170         return 0;
3171 }
3172
3173 static void qede_simd_fp_handler(void *cookie)
3174 {
3175         struct qede_fastpath *fp = (struct qede_fastpath *)cookie;
3176
3177         napi_schedule_irqoff(&fp->napi);
3178 }
3179
3180 static int qede_setup_irqs(struct qede_dev *edev)
3181 {
3182         int i, rc = 0;
3183
3184         /* Learn Interrupt configuration */
3185         rc = edev->ops->common->get_fp_int(edev->cdev, &edev->int_info);
3186         if (rc)
3187                 return rc;
3188
3189         if (edev->int_info.msix_cnt) {
3190                 rc = qede_req_msix_irqs(edev);
3191                 if (rc)
3192                         return rc;
3193                 edev->ndev->irq = edev->int_info.msix[0].vector;
3194         } else {
3195                 const struct qed_common_ops *ops;
3196
3197                 /* qed should learn receive the RSS ids and callbacks */
3198                 ops = edev->ops->common;
3199                 for (i = 0; i < QEDE_QUEUE_CNT(edev); i++)
3200                         ops->simd_handler_config(edev->cdev,
3201                                                  &edev->fp_array[i], i,
3202                                                  qede_simd_fp_handler);
3203                 edev->int_info.used_cnt = QEDE_QUEUE_CNT(edev);
3204         }
3205         return 0;
3206 }
3207
3208 static int qede_drain_txq(struct qede_dev *edev,
3209                           struct qede_tx_queue *txq, bool allow_drain)
3210 {
3211         int rc, cnt = 1000;
3212
3213         while (txq->sw_tx_cons != txq->sw_tx_prod) {
3214                 if (!cnt) {
3215                         if (allow_drain) {
3216                                 DP_NOTICE(edev,
3217                                           "Tx queue[%d] is stuck, requesting MCP to drain\n",
3218                                           txq->index);
3219                                 rc = edev->ops->common->drain(edev->cdev);
3220                                 if (rc)
3221                                         return rc;
3222                                 return qede_drain_txq(edev, txq, false);
3223                         }
3224                         DP_NOTICE(edev,
3225                                   "Timeout waiting for tx queue[%d]: PROD=%d, CONS=%d\n",
3226                                   txq->index, txq->sw_tx_prod,
3227                                   txq->sw_tx_cons);
3228                         return -ENODEV;
3229                 }
3230                 cnt--;
3231                 usleep_range(1000, 2000);
3232                 barrier();
3233         }
3234
3235         /* FW finished processing, wait for HW to transmit all tx packets */
3236         usleep_range(1000, 2000);
3237
3238         return 0;
3239 }
3240
3241 static int qede_stop_queues(struct qede_dev *edev)
3242 {
3243         struct qed_update_vport_params vport_update_params;
3244         struct qed_dev *cdev = edev->cdev;
3245         int rc, tc, i;
3246
3247         /* Disable the vport */
3248         memset(&vport_update_params, 0, sizeof(vport_update_params));
3249         vport_update_params.vport_id = 0;
3250         vport_update_params.update_vport_active_flg = 1;
3251         vport_update_params.vport_active_flg = 0;
3252         vport_update_params.update_rss_flg = 0;
3253
3254         rc = edev->ops->vport_update(cdev, &vport_update_params);
3255         if (rc) {
3256                 DP_ERR(edev, "Failed to update vport\n");
3257                 return rc;
3258         }
3259
3260         /* Flush Tx queues. If needed, request drain from MCP */
3261         for_each_queue(i) {
3262                 struct qede_fastpath *fp = &edev->fp_array[i];
3263
3264                 if (fp->type & QEDE_FASTPATH_TX) {
3265                         for (tc = 0; tc < edev->num_tc; tc++) {
3266                                 struct qede_tx_queue *txq = &fp->txqs[tc];
3267
3268                                 rc = qede_drain_txq(edev, txq, true);
3269                                 if (rc)
3270                                         return rc;
3271                         }
3272                 }
3273         }
3274
3275         /* Stop all Queues in reverse order */
3276         for (i = QEDE_QUEUE_CNT(edev) - 1; i >= 0; i--) {
3277                 struct qed_stop_rxq_params rx_params;
3278
3279                 /* Stop the Tx Queue(s) */
3280                 if (edev->fp_array[i].type & QEDE_FASTPATH_TX) {
3281                         for (tc = 0; tc < edev->num_tc; tc++) {
3282                                 struct qed_stop_txq_params tx_params;
3283                                 u8 val;
3284
3285                                 tx_params.rss_id = i;
3286                                 val = edev->fp_array[i].txqs[tc].index;
3287                                 tx_params.tx_queue_id = val;
3288                                 rc = edev->ops->q_tx_stop(cdev, &tx_params);
3289                                 if (rc) {
3290                                         DP_ERR(edev, "Failed to stop TXQ #%d\n",
3291                                                tx_params.tx_queue_id);
3292                                         return rc;
3293                                 }
3294                         }
3295                 }
3296
3297                 /* Stop the Rx Queue */
3298                 if (edev->fp_array[i].type & QEDE_FASTPATH_RX) {
3299                         memset(&rx_params, 0, sizeof(rx_params));
3300                         rx_params.rss_id = i;
3301                         rx_params.rx_queue_id = edev->fp_array[i].rxq->rxq_id;
3302
3303                         rc = edev->ops->q_rx_stop(cdev, &rx_params);
3304                         if (rc) {
3305                                 DP_ERR(edev, "Failed to stop RXQ #%d\n", i);
3306                                 return rc;
3307                         }
3308                 }
3309         }
3310
3311         /* Stop the vport */
3312         rc = edev->ops->vport_stop(cdev, 0);
3313         if (rc)
3314                 DP_ERR(edev, "Failed to stop VPORT\n");
3315
3316         return rc;
3317 }
3318
3319 static int qede_start_queues(struct qede_dev *edev, bool clear_stats)
3320 {
3321         int rc, tc, i;
3322         int vlan_removal_en = 1;
3323         struct qed_dev *cdev = edev->cdev;
3324         struct qed_update_vport_params vport_update_params;
3325         struct qed_queue_start_common_params q_params;
3326         struct qed_dev_info *qed_info = &edev->dev_info.common;
3327         struct qed_start_vport_params start = {0};
3328         bool reset_rss_indir = false;
3329
3330         if (!edev->num_queues) {
3331                 DP_ERR(edev,
3332                        "Cannot update V-VPORT as active as there are no Rx queues\n");
3333                 return -EINVAL;
3334         }
3335
3336         start.gro_enable = !edev->gro_disable;
3337         start.mtu = edev->ndev->mtu;
3338         start.vport_id = 0;
3339         start.drop_ttl0 = true;
3340         start.remove_inner_vlan = vlan_removal_en;
3341         start.clear_stats = clear_stats;
3342
3343         rc = edev->ops->vport_start(cdev, &start);
3344
3345         if (rc) {
3346                 DP_ERR(edev, "Start V-PORT failed %d\n", rc);
3347                 return rc;
3348         }
3349
3350         DP_VERBOSE(edev, NETIF_MSG_IFUP,
3351                    "Start vport ramrod passed, vport_id = %d, MTU = %d, vlan_removal_en = %d\n",
3352                    start.vport_id, edev->ndev->mtu + 0xe, vlan_removal_en);
3353
3354         for_each_queue(i) {
3355                 struct qede_fastpath *fp = &edev->fp_array[i];
3356                 dma_addr_t p_phys_table;
3357                 u32 page_cnt;
3358
3359                 if (fp->type & QEDE_FASTPATH_RX) {
3360                         struct qede_rx_queue *rxq = fp->rxq;
3361                         __le16 *val;
3362
3363                         memset(&q_params, 0, sizeof(q_params));
3364                         q_params.rss_id = i;
3365                         q_params.queue_id = rxq->rxq_id;
3366                         q_params.vport_id = 0;
3367                         q_params.sb = fp->sb_info->igu_sb_id;
3368                         q_params.sb_idx = RX_PI;
3369
3370                         p_phys_table =
3371                             qed_chain_get_pbl_phys(&rxq->rx_comp_ring);
3372                         page_cnt = qed_chain_get_page_cnt(&rxq->rx_comp_ring);
3373
3374                         rc = edev->ops->q_rx_start(cdev, &q_params,
3375                                                    rxq->rx_buf_size,
3376                                                    rxq->rx_bd_ring.p_phys_addr,
3377                                                    p_phys_table,
3378                                                    page_cnt,
3379                                                    &rxq->hw_rxq_prod_addr);
3380                         if (rc) {
3381                                 DP_ERR(edev, "Start RXQ #%d failed %d\n", i,
3382                                        rc);
3383                                 return rc;
3384                         }
3385
3386                         val = &fp->sb_info->sb_virt->pi_array[RX_PI];
3387                         rxq->hw_cons_ptr = val;
3388
3389                         qede_update_rx_prod(edev, rxq);
3390                 }
3391
3392                 if (!(fp->type & QEDE_FASTPATH_TX))
3393                         continue;
3394
3395                 for (tc = 0; tc < edev->num_tc; tc++) {
3396                         struct qede_tx_queue *txq = &fp->txqs[tc];
3397
3398                         p_phys_table = qed_chain_get_pbl_phys(&txq->tx_pbl);
3399                         page_cnt = qed_chain_get_page_cnt(&txq->tx_pbl);
3400
3401                         memset(&q_params, 0, sizeof(q_params));
3402                         q_params.rss_id = i;
3403                         q_params.queue_id = txq->index;
3404                         q_params.vport_id = 0;
3405                         q_params.sb = fp->sb_info->igu_sb_id;
3406                         q_params.sb_idx = TX_PI(tc);
3407
3408                         rc = edev->ops->q_tx_start(cdev, &q_params,
3409                                                    p_phys_table, page_cnt,
3410                                                    &txq->doorbell_addr);
3411                         if (rc) {
3412                                 DP_ERR(edev, "Start TXQ #%d failed %d\n",
3413                                        txq->index, rc);
3414                                 return rc;
3415                         }
3416
3417                         txq->hw_cons_ptr =
3418                                 &fp->sb_info->sb_virt->pi_array[TX_PI(tc)];
3419                         SET_FIELD(txq->tx_db.data.params,
3420                                   ETH_DB_DATA_DEST, DB_DEST_XCM);
3421                         SET_FIELD(txq->tx_db.data.params, ETH_DB_DATA_AGG_CMD,
3422                                   DB_AGG_CMD_SET);
3423                         SET_FIELD(txq->tx_db.data.params,
3424                                   ETH_DB_DATA_AGG_VAL_SEL,
3425                                   DQ_XCM_ETH_TX_BD_PROD_CMD);
3426
3427                         txq->tx_db.data.agg_flags = DQ_XCM_ETH_DQ_CF_CMD;
3428                 }
3429         }
3430
3431         /* Prepare and send the vport enable */
3432         memset(&vport_update_params, 0, sizeof(vport_update_params));
3433         vport_update_params.vport_id = start.vport_id;
3434         vport_update_params.update_vport_active_flg = 1;
3435         vport_update_params.vport_active_flg = 1;
3436
3437         if ((qed_info->mf_mode == QED_MF_NPAR || pci_num_vf(edev->pdev)) &&
3438             qed_info->tx_switching) {
3439                 vport_update_params.update_tx_switching_flg = 1;
3440                 vport_update_params.tx_switching_flg = 1;
3441         }
3442
3443         /* Fill struct with RSS params */
3444         if (QEDE_RSS_COUNT(edev) > 1) {
3445                 vport_update_params.update_rss_flg = 1;
3446
3447                 /* Need to validate current RSS config uses valid entries */
3448                 for (i = 0; i < QED_RSS_IND_TABLE_SIZE; i++) {
3449                         if (edev->rss_params.rss_ind_table[i] >=
3450                             QEDE_RSS_COUNT(edev)) {
3451                                 reset_rss_indir = true;
3452                                 break;
3453                         }
3454                 }
3455
3456                 if (!(edev->rss_params_inited & QEDE_RSS_INDIR_INITED) ||
3457                     reset_rss_indir) {
3458                         u16 val;
3459
3460                         for (i = 0; i < QED_RSS_IND_TABLE_SIZE; i++) {
3461                                 u16 indir_val;
3462
3463                                 val = QEDE_RSS_COUNT(edev);
3464                                 indir_val = ethtool_rxfh_indir_default(i, val);
3465                                 edev->rss_params.rss_ind_table[i] = indir_val;
3466                         }
3467                         edev->rss_params_inited |= QEDE_RSS_INDIR_INITED;
3468                 }
3469
3470                 if (!(edev->rss_params_inited & QEDE_RSS_KEY_INITED)) {
3471                         netdev_rss_key_fill(edev->rss_params.rss_key,
3472                                             sizeof(edev->rss_params.rss_key));
3473                         edev->rss_params_inited |= QEDE_RSS_KEY_INITED;
3474                 }
3475
3476                 if (!(edev->rss_params_inited & QEDE_RSS_CAPS_INITED)) {
3477                         edev->rss_params.rss_caps = QED_RSS_IPV4 |
3478                                                     QED_RSS_IPV6 |
3479                                                     QED_RSS_IPV4_TCP |
3480                                                     QED_RSS_IPV6_TCP;
3481                         edev->rss_params_inited |= QEDE_RSS_CAPS_INITED;
3482                 }
3483
3484                 memcpy(&vport_update_params.rss_params, &edev->rss_params,
3485                        sizeof(vport_update_params.rss_params));
3486         } else {
3487                 memset(&vport_update_params.rss_params, 0,
3488                        sizeof(vport_update_params.rss_params));
3489         }
3490
3491         rc = edev->ops->vport_update(cdev, &vport_update_params);
3492         if (rc) {
3493                 DP_ERR(edev, "Update V-PORT failed %d\n", rc);
3494                 return rc;
3495         }
3496
3497         return 0;
3498 }
3499
3500 static int qede_set_mcast_rx_mac(struct qede_dev *edev,
3501                                  enum qed_filter_xcast_params_type opcode,
3502                                  unsigned char *mac, int num_macs)
3503 {
3504         struct qed_filter_params filter_cmd;
3505         int i;
3506
3507         memset(&filter_cmd, 0, sizeof(filter_cmd));
3508         filter_cmd.type = QED_FILTER_TYPE_MCAST;
3509         filter_cmd.filter.mcast.type = opcode;
3510         filter_cmd.filter.mcast.num = num_macs;
3511
3512         for (i = 0; i < num_macs; i++, mac += ETH_ALEN)
3513                 ether_addr_copy(filter_cmd.filter.mcast.mac[i], mac);
3514
3515         return edev->ops->filter_config(edev->cdev, &filter_cmd);
3516 }
3517
3518 enum qede_unload_mode {
3519         QEDE_UNLOAD_NORMAL,
3520 };
3521
3522 static void qede_unload(struct qede_dev *edev, enum qede_unload_mode mode)
3523 {
3524         struct qed_link_params link_params;
3525         int rc;
3526
3527         DP_INFO(edev, "Starting qede unload\n");
3528
3529         qede_roce_dev_event_close(edev);
3530         mutex_lock(&edev->qede_lock);
3531         edev->state = QEDE_STATE_CLOSED;
3532
3533         /* Close OS Tx */
3534         netif_tx_disable(edev->ndev);
3535         netif_carrier_off(edev->ndev);
3536
3537         /* Reset the link */
3538         memset(&link_params, 0, sizeof(link_params));
3539         link_params.link_up = false;
3540         edev->ops->common->set_link(edev->cdev, &link_params);
3541         rc = qede_stop_queues(edev);
3542         if (rc) {
3543                 qede_sync_free_irqs(edev);
3544                 goto out;
3545         }
3546
3547         DP_INFO(edev, "Stopped Queues\n");
3548
3549         qede_vlan_mark_nonconfigured(edev);
3550         edev->ops->fastpath_stop(edev->cdev);
3551
3552         /* Release the interrupts */
3553         qede_sync_free_irqs(edev);
3554         edev->ops->common->set_fp_int(edev->cdev, 0);
3555
3556         qede_napi_disable_remove(edev);
3557
3558         qede_free_mem_load(edev);
3559         qede_free_fp_array(edev);
3560
3561 out:
3562         mutex_unlock(&edev->qede_lock);
3563         DP_INFO(edev, "Ending qede unload\n");
3564 }
3565
3566 enum qede_load_mode {
3567         QEDE_LOAD_NORMAL,
3568         QEDE_LOAD_RELOAD,
3569 };
3570
3571 static int qede_load(struct qede_dev *edev, enum qede_load_mode mode)
3572 {
3573         struct qed_link_params link_params;
3574         struct qed_link_output link_output;
3575         int rc;
3576
3577         DP_INFO(edev, "Starting qede load\n");
3578
3579         rc = qede_set_num_queues(edev);
3580         if (rc)
3581                 goto err0;
3582
3583         rc = qede_alloc_fp_array(edev);
3584         if (rc)
3585                 goto err0;
3586
3587         qede_init_fp(edev);
3588
3589         rc = qede_alloc_mem_load(edev);
3590         if (rc)
3591                 goto err1;
3592         DP_INFO(edev, "Allocated %d RSS queues on %d TC/s\n",
3593                 QEDE_QUEUE_CNT(edev), edev->num_tc);
3594
3595         rc = qede_set_real_num_queues(edev);
3596         if (rc)
3597                 goto err2;
3598
3599         qede_napi_add_enable(edev);
3600         DP_INFO(edev, "Napi added and enabled\n");
3601
3602         rc = qede_setup_irqs(edev);
3603         if (rc)
3604                 goto err3;
3605         DP_INFO(edev, "Setup IRQs succeeded\n");
3606
3607         rc = qede_start_queues(edev, mode != QEDE_LOAD_RELOAD);
3608         if (rc)
3609                 goto err4;
3610         DP_INFO(edev, "Start VPORT, RXQ and TXQ succeeded\n");
3611
3612         /* Add primary mac and set Rx filters */
3613         ether_addr_copy(edev->primary_mac, edev->ndev->dev_addr);
3614
3615         mutex_lock(&edev->qede_lock);
3616         edev->state = QEDE_STATE_OPEN;
3617         mutex_unlock(&edev->qede_lock);
3618
3619         /* Program un-configured VLANs */
3620         qede_configure_vlan_filters(edev);
3621
3622         /* Ask for link-up using current configuration */
3623         memset(&link_params, 0, sizeof(link_params));
3624         link_params.link_up = true;
3625         edev->ops->common->set_link(edev->cdev, &link_params);
3626
3627         /* Query whether link is already-up */
3628         memset(&link_output, 0, sizeof(link_output));
3629         edev->ops->common->get_link(edev->cdev, &link_output);
3630         qede_roce_dev_event_open(edev);
3631         qede_link_update(edev, &link_output);
3632
3633         DP_INFO(edev, "Ending successfully qede load\n");
3634
3635         return 0;
3636
3637 err4:
3638         qede_sync_free_irqs(edev);
3639         memset(&edev->int_info.msix_cnt, 0, sizeof(struct qed_int_info));
3640 err3:
3641         qede_napi_disable_remove(edev);
3642 err2:
3643         qede_free_mem_load(edev);
3644 err1:
3645         edev->ops->common->set_fp_int(edev->cdev, 0);
3646         qede_free_fp_array(edev);
3647         edev->num_queues = 0;
3648         edev->fp_num_tx = 0;
3649         edev->fp_num_rx = 0;
3650 err0:
3651         return rc;
3652 }
3653
3654 void qede_reload(struct qede_dev *edev,
3655                  void (*func)(struct qede_dev *, union qede_reload_args *),
3656                  union qede_reload_args *args)
3657 {
3658         qede_unload(edev, QEDE_UNLOAD_NORMAL);
3659         /* Call function handler to update parameters
3660          * needed for function load.
3661          */
3662         if (func)
3663                 func(edev, args);
3664
3665         qede_load(edev, QEDE_LOAD_RELOAD);
3666
3667         mutex_lock(&edev->qede_lock);
3668         qede_config_rx_mode(edev->ndev);
3669         mutex_unlock(&edev->qede_lock);
3670 }
3671
3672 /* called with rtnl_lock */
3673 static int qede_open(struct net_device *ndev)
3674 {
3675         struct qede_dev *edev = netdev_priv(ndev);
3676         int rc;
3677
3678         netif_carrier_off(ndev);
3679
3680         edev->ops->common->set_power_state(edev->cdev, PCI_D0);
3681
3682         rc = qede_load(edev, QEDE_LOAD_NORMAL);
3683
3684         if (rc)
3685                 return rc;
3686
3687         udp_tunnel_get_rx_info(ndev);
3688
3689         return 0;
3690 }
3691
3692 static int qede_close(struct net_device *ndev)
3693 {
3694         struct qede_dev *edev = netdev_priv(ndev);
3695
3696         qede_unload(edev, QEDE_UNLOAD_NORMAL);
3697
3698         return 0;
3699 }
3700
3701 static void qede_link_update(void *dev, struct qed_link_output *link)
3702 {
3703         struct qede_dev *edev = dev;
3704
3705         if (!netif_running(edev->ndev)) {
3706                 DP_VERBOSE(edev, NETIF_MSG_LINK, "Interface is not running\n");
3707                 return;
3708         }
3709
3710         if (link->link_up) {
3711                 if (!netif_carrier_ok(edev->ndev)) {
3712                         DP_NOTICE(edev, "Link is up\n");
3713                         netif_tx_start_all_queues(edev->ndev);
3714                         netif_carrier_on(edev->ndev);
3715                 }
3716         } else {
3717                 if (netif_carrier_ok(edev->ndev)) {
3718                         DP_NOTICE(edev, "Link is down\n");
3719                         netif_tx_disable(edev->ndev);
3720                         netif_carrier_off(edev->ndev);
3721                 }
3722         }
3723 }
3724
3725 static int qede_set_mac_addr(struct net_device *ndev, void *p)
3726 {
3727         struct qede_dev *edev = netdev_priv(ndev);
3728         struct sockaddr *addr = p;
3729         int rc;
3730
3731         ASSERT_RTNL(); /* @@@TBD To be removed */
3732
3733         DP_INFO(edev, "Set_mac_addr called\n");
3734
3735         if (!is_valid_ether_addr(addr->sa_data)) {
3736                 DP_NOTICE(edev, "The MAC address is not valid\n");
3737                 return -EFAULT;
3738         }
3739
3740         if (!edev->ops->check_mac(edev->cdev, addr->sa_data)) {
3741                 DP_NOTICE(edev, "qed prevents setting MAC\n");
3742                 return -EINVAL;
3743         }
3744
3745         ether_addr_copy(ndev->dev_addr, addr->sa_data);
3746
3747         if (!netif_running(ndev))  {
3748                 DP_NOTICE(edev, "The device is currently down\n");
3749                 return 0;
3750         }
3751
3752         /* Remove the previous primary mac */
3753         rc = qede_set_ucast_rx_mac(edev, QED_FILTER_XCAST_TYPE_DEL,
3754                                    edev->primary_mac);
3755         if (rc)
3756                 return rc;
3757
3758         /* Add MAC filter according to the new unicast HW MAC address */
3759         ether_addr_copy(edev->primary_mac, ndev->dev_addr);
3760         return qede_set_ucast_rx_mac(edev, QED_FILTER_XCAST_TYPE_ADD,
3761                                       edev->primary_mac);
3762 }
3763
3764 static int
3765 qede_configure_mcast_filtering(struct net_device *ndev,
3766                                enum qed_filter_rx_mode_type *accept_flags)
3767 {
3768         struct qede_dev *edev = netdev_priv(ndev);
3769         unsigned char *mc_macs, *temp;
3770         struct netdev_hw_addr *ha;
3771         int rc = 0, mc_count;
3772         size_t size;
3773
3774         size = 64 * ETH_ALEN;
3775
3776         mc_macs = kzalloc(size, GFP_KERNEL);
3777         if (!mc_macs) {
3778                 DP_NOTICE(edev,
3779                           "Failed to allocate memory for multicast MACs\n");
3780                 rc = -ENOMEM;
3781                 goto exit;
3782         }
3783
3784         temp = mc_macs;
3785
3786         /* Remove all previously configured MAC filters */
3787         rc = qede_set_mcast_rx_mac(edev, QED_FILTER_XCAST_TYPE_DEL,
3788                                    mc_macs, 1);
3789         if (rc)
3790                 goto exit;
3791
3792         netif_addr_lock_bh(ndev);
3793
3794         mc_count = netdev_mc_count(ndev);
3795         if (mc_count < 64) {
3796                 netdev_for_each_mc_addr(ha, ndev) {
3797                         ether_addr_copy(temp, ha->addr);
3798                         temp += ETH_ALEN;
3799                 }
3800         }
3801
3802         netif_addr_unlock_bh(ndev);
3803
3804         /* Check for all multicast @@@TBD resource allocation */
3805         if ((ndev->flags & IFF_ALLMULTI) ||
3806             (mc_count > 64)) {
3807                 if (*accept_flags == QED_FILTER_RX_MODE_TYPE_REGULAR)
3808                         *accept_flags = QED_FILTER_RX_MODE_TYPE_MULTI_PROMISC;
3809         } else {
3810                 /* Add all multicast MAC filters */
3811                 rc = qede_set_mcast_rx_mac(edev, QED_FILTER_XCAST_TYPE_ADD,
3812                                            mc_macs, mc_count);
3813         }
3814
3815 exit:
3816         kfree(mc_macs);
3817         return rc;
3818 }
3819
3820 static void qede_set_rx_mode(struct net_device *ndev)
3821 {
3822         struct qede_dev *edev = netdev_priv(ndev);
3823
3824         DP_INFO(edev, "qede_set_rx_mode called\n");
3825
3826         if (edev->state != QEDE_STATE_OPEN) {
3827                 DP_INFO(edev,
3828                         "qede_set_rx_mode called while interface is down\n");
3829         } else {
3830                 set_bit(QEDE_SP_RX_MODE, &edev->sp_flags);
3831                 schedule_delayed_work(&edev->sp_task, 0);
3832         }
3833 }
3834
3835 /* Must be called with qede_lock held */
3836 static void qede_config_rx_mode(struct net_device *ndev)
3837 {
3838         enum qed_filter_rx_mode_type accept_flags = QED_FILTER_TYPE_UCAST;
3839         struct qede_dev *edev = netdev_priv(ndev);
3840         struct qed_filter_params rx_mode;
3841         unsigned char *uc_macs, *temp;
3842         struct netdev_hw_addr *ha;
3843         int rc, uc_count;
3844         size_t size;
3845
3846         netif_addr_lock_bh(ndev);
3847
3848         uc_count = netdev_uc_count(ndev);
3849         size = uc_count * ETH_ALEN;
3850
3851         uc_macs = kzalloc(size, GFP_ATOMIC);
3852         if (!uc_macs) {
3853                 DP_NOTICE(edev, "Failed to allocate memory for unicast MACs\n");
3854                 netif_addr_unlock_bh(ndev);
3855                 return;
3856         }
3857
3858         temp = uc_macs;
3859         netdev_for_each_uc_addr(ha, ndev) {
3860                 ether_addr_copy(temp, ha->addr);
3861                 temp += ETH_ALEN;
3862         }
3863
3864         netif_addr_unlock_bh(ndev);
3865
3866         /* Configure the struct for the Rx mode */
3867         memset(&rx_mode, 0, sizeof(struct qed_filter_params));
3868         rx_mode.type = QED_FILTER_TYPE_RX_MODE;
3869
3870         /* Remove all previous unicast secondary macs and multicast macs
3871          * (configrue / leave the primary mac)
3872          */
3873         rc = qede_set_ucast_rx_mac(edev, QED_FILTER_XCAST_TYPE_REPLACE,
3874                                    edev->primary_mac);
3875         if (rc)
3876                 goto out;
3877
3878         /* Check for promiscuous */
3879         if ((ndev->flags & IFF_PROMISC) ||
3880             (uc_count > 15)) { /* @@@TBD resource allocation - 1 */
3881                 accept_flags = QED_FILTER_RX_MODE_TYPE_PROMISC;
3882         } else {
3883                 /* Add MAC filters according to the unicast secondary macs */
3884                 int i;
3885
3886                 temp = uc_macs;
3887                 for (i = 0; i < uc_count; i++) {
3888                         rc = qede_set_ucast_rx_mac(edev,
3889                                                    QED_FILTER_XCAST_TYPE_ADD,
3890                                                    temp);
3891                         if (rc)
3892                                 goto out;
3893
3894                         temp += ETH_ALEN;
3895                 }
3896
3897                 rc = qede_configure_mcast_filtering(ndev, &accept_flags);
3898                 if (rc)
3899                         goto out;
3900         }
3901
3902         /* take care of VLAN mode */
3903         if (ndev->flags & IFF_PROMISC) {
3904                 qede_config_accept_any_vlan(edev, true);
3905         } else if (!edev->non_configured_vlans) {
3906                 /* It's possible that accept_any_vlan mode is set due to a
3907                  * previous setting of IFF_PROMISC. If vlan credits are
3908                  * sufficient, disable accept_any_vlan.
3909                  */
3910                 qede_config_accept_any_vlan(edev, false);
3911         }
3912
3913         rx_mode.filter.accept_flags = accept_flags;
3914         edev->ops->filter_config(edev->cdev, &rx_mode);
3915 out:
3916         kfree(uc_macs);
3917 }