Pileus Git - ~andy/linux/blob - drivers/net/ethernet/broadcom/bnx2x/bnx2x_cmn.c

   1 /* bnx2x_cmn.c: Broadcom Everest network driver.
   2  *
   3  * Copyright (c) 2007-2013 Broadcom Corporation
   4  *
   5  * This program is free software; you can redistribute it and/or modify
   6  * it under the terms of the GNU General Public License as published by
   7  * the Free Software Foundation.
   8  *
   9  * Maintained by: Eilon Greenstein <eilong@broadcom.com>
  10  * Written by: Eliezer Tamir
  11  * Based on code from Michael Chan's bnx2 driver
  12  * UDP CSUM errata workaround by Arik Gendelman
  13  * Slowpath and fastpath rework by Vladislav Zolotarov
  14  * Statistics and Link management by Yitchak Gertner
  15  *
  16  */
  17
  18 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  19
  20 #include <linux/etherdevice.h>
  21 #include <linux/if_vlan.h>
  22 #include <linux/interrupt.h>
  23 #include <linux/ip.h>
  24 #include <net/tcp.h>
  25 #include <net/ipv6.h>
  26 #include <net/ip6_checksum.h>
  27 #include <net/busy_poll.h>
  28 #include <linux/prefetch.h>
  29 #include "bnx2x_cmn.h"
  30 #include "bnx2x_init.h"
  31 #include "bnx2x_sp.h"
  32
  33 /**
  34  * bnx2x_move_fp - move content of the fastpath structure.
  35  *
  36  * @bp:         driver handle
  37  * @from:       source FP index
  38  * @to:         destination FP index
  39  *
  40  * Makes sure the contents of the bp->fp[to].napi is kept
  41  * intact. This is done by first copying the napi struct from
  42  * the target to the source, and then mem copying the entire
  43  * source onto the target. Update txdata pointers and related
  44  * content.
  45  */
  46 static inline void bnx2x_move_fp(struct bnx2x *bp, int from, int to)
  47 {
  48         struct bnx2x_fastpath *from_fp = &bp->fp[from];
  49         struct bnx2x_fastpath *to_fp = &bp->fp[to];
  50         struct bnx2x_sp_objs *from_sp_objs = &bp->sp_objs[from];
  51         struct bnx2x_sp_objs *to_sp_objs = &bp->sp_objs[to];
  52         struct bnx2x_fp_stats *from_fp_stats = &bp->fp_stats[from];
  53         struct bnx2x_fp_stats *to_fp_stats = &bp->fp_stats[to];
  54         int old_max_eth_txqs, new_max_eth_txqs;
  55         int old_txdata_index = 0, new_txdata_index = 0;
  56         struct bnx2x_agg_info *old_tpa_info = to_fp->tpa_info;
  57
  58         /* Copy the NAPI object as it has been already initialized */
  59         from_fp->napi = to_fp->napi;
  60
  61         /* Move bnx2x_fastpath contents */
  62         memcpy(to_fp, from_fp, sizeof(*to_fp));
  63         to_fp->index = to;
  64
  65         /* Retain the tpa_info of the original `to' version as we don't want
  66          * 2 FPs to contain the same tpa_info pointer.
  67          */
  68         to_fp->tpa_info = old_tpa_info;
  69
  70         /* move sp_objs contents as well, as their indices match fp ones */
  71         memcpy(to_sp_objs, from_sp_objs, sizeof(*to_sp_objs));
  72
  73         /* move fp_stats contents as well, as their indices match fp ones */
  74         memcpy(to_fp_stats, from_fp_stats, sizeof(*to_fp_stats));
  75
  76         /* Update txdata pointers in fp and move txdata content accordingly:
  77          * Each fp consumes 'max_cos' txdata structures, so the index should be
  78          * decremented by max_cos x delta.
  79          */
  80
  81         old_max_eth_txqs = BNX2X_NUM_ETH_QUEUES(bp) * (bp)->max_cos;
  82         new_max_eth_txqs = (BNX2X_NUM_ETH_QUEUES(bp) - from + to) *
  83                                 (bp)->max_cos;
  84         if (from == FCOE_IDX(bp)) {
  85                 old_txdata_index = old_max_eth_txqs + FCOE_TXQ_IDX_OFFSET;
  86                 new_txdata_index = new_max_eth_txqs + FCOE_TXQ_IDX_OFFSET;
  87         }
  88
  89         memcpy(&bp->bnx2x_txq[new_txdata_index],
  90                &bp->bnx2x_txq[old_txdata_index],
  91                sizeof(struct bnx2x_fp_txdata));
  92         to_fp->txdata_ptr[0] = &bp->bnx2x_txq[new_txdata_index];
  93 }
  94
  95 /**
  96  * bnx2x_fill_fw_str - Fill buffer with FW version string.
  97  *
  98  * @bp:        driver handle
  99  * @buf:       character buffer to fill with the fw name
 100  * @buf_len:   length of the above buffer
 101  *
 102  */
 103 void bnx2x_fill_fw_str(struct bnx2x *bp, char *buf, size_t buf_len)
 104 {
 105         if (IS_PF(bp)) {
 106                 u8 phy_fw_ver[PHY_FW_VER_LEN];
 107
 108                 phy_fw_ver[0] = '\0';
 109                 bnx2x_get_ext_phy_fw_version(&bp->link_params,
 110                                              phy_fw_ver, PHY_FW_VER_LEN);
 111                 strlcpy(buf, bp->fw_ver, buf_len);
 112                 snprintf(buf + strlen(bp->fw_ver), 32 - strlen(bp->fw_ver),
 113                          "bc %d.%d.%d%s%s",
 114                          (bp->common.bc_ver & 0xff0000) >> 16,
 115                          (bp->common.bc_ver & 0xff00) >> 8,
 116                          (bp->common.bc_ver & 0xff),
 117                          ((phy_fw_ver[0] != '\0') ? " phy " : ""), phy_fw_ver);
 118         } else {
 119                 bnx2x_vf_fill_fw_str(bp, buf, buf_len);
 120         }
 121 }
 122
 123 /**
 124  * bnx2x_shrink_eth_fp - guarantees fastpath structures stay intact
 125  *
 126  * @bp: driver handle
 127  * @delta:      number of eth queues which were not allocated
 128  */
 129 static void bnx2x_shrink_eth_fp(struct bnx2x *bp, int delta)
 130 {
 131         int i, cos, old_eth_num = BNX2X_NUM_ETH_QUEUES(bp);
 132
 133         /* Queue pointer cannot be re-set on an fp-basis, as moving pointer
 134          * backward along the array could cause memory to be overridden
 135          */
 136         for (cos = 1; cos < bp->max_cos; cos++) {
 137                 for (i = 0; i < old_eth_num - delta; i++) {
 138                         struct bnx2x_fastpath *fp = &bp->fp[i];
 139                         int new_idx = cos * (old_eth_num - delta) + i;
 140
 141                         memcpy(&bp->bnx2x_txq[new_idx], fp->txdata_ptr[cos],
 142                                sizeof(struct bnx2x_fp_txdata));
 143                         fp->txdata_ptr[cos] = &bp->bnx2x_txq[new_idx];
 144                 }
 145         }
 146 }
 147
 148 int load_count[2][3] = { {0} }; /* per-path: 0-common, 1-port0, 2-port1 */
 149
 150 /* free skb in the packet ring at pos idx
 151  * return idx of last bd freed
 152  */
 153 static u16 bnx2x_free_tx_pkt(struct bnx2x *bp, struct bnx2x_fp_txdata *txdata,
 154                              u16 idx, unsigned int *pkts_compl,
 155                              unsigned int *bytes_compl)
 156 {
 157         struct sw_tx_bd *tx_buf = &txdata->tx_buf_ring[idx];
 158         struct eth_tx_start_bd *tx_start_bd;
 159         struct eth_tx_bd *tx_data_bd;
 160         struct sk_buff *skb = tx_buf->skb;
 161         u16 bd_idx = TX_BD(tx_buf->first_bd), new_cons;
 162         int nbd;
 163
 164         /* prefetch skb end pointer to speedup dev_kfree_skb() */
 165         prefetch(&skb->end);
 166
 167         DP(NETIF_MSG_TX_DONE, "fp[%d]: pkt_idx %d  buff @(%p)->skb %p\n",
 168            txdata->txq_index, idx, tx_buf, skb);
 169
 170         /* unmap first bd */
 171         tx_start_bd = &txdata->tx_desc_ring[bd_idx].start_bd;
 172         dma_unmap_single(&bp->pdev->dev, BD_UNMAP_ADDR(tx_start_bd),
 173                          BD_UNMAP_LEN(tx_start_bd), DMA_TO_DEVICE);
 174
 175         nbd = le16_to_cpu(tx_start_bd->nbd) - 1;
 176 #ifdef BNX2X_STOP_ON_ERROR
 177         if ((nbd - 1) > (MAX_SKB_FRAGS + 2)) {
 178                 BNX2X_ERR("BAD nbd!\n");
 179                 bnx2x_panic();
 180         }
 181 #endif
 182         new_cons = nbd + tx_buf->first_bd;
 183
 184         /* Get the next bd */
 185         bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
 186
 187         /* Skip a parse bd... */
 188         --nbd;
 189         bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
 190
 191         /* ...and the TSO split header bd since they have no mapping */
 192         if (tx_buf->flags & BNX2X_TSO_SPLIT_BD) {
 193                 --nbd;
 194                 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
 195         }
 196
 197         /* now free frags */
 198         while (nbd > 0) {
 199
 200                 tx_data_bd = &txdata->tx_desc_ring[bd_idx].reg_bd;
 201                 dma_unmap_page(&bp->pdev->dev, BD_UNMAP_ADDR(tx_data_bd),
 202                                BD_UNMAP_LEN(tx_data_bd), DMA_TO_DEVICE);
 203                 if (--nbd)
 204                         bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
 205         }
 206
 207         /* release skb */
 208         WARN_ON(!skb);
 209         if (likely(skb)) {
 210                 (*pkts_compl)++;
 211                 (*bytes_compl) += skb->len;
 212         }
 213
 214         dev_kfree_skb_any(skb);
 215         tx_buf->first_bd = 0;
 216         tx_buf->skb = NULL;
 217
 218         return new_cons;
 219 }
 220
 221 int bnx2x_tx_int(struct bnx2x *bp, struct bnx2x_fp_txdata *txdata)
 222 {
 223         struct netdev_queue *txq;
 224         u16 hw_cons, sw_cons, bd_cons = txdata->tx_bd_cons;
 225         unsigned int pkts_compl = 0, bytes_compl = 0;
 226
 227 #ifdef BNX2X_STOP_ON_ERROR
 228         if (unlikely(bp->panic))
 229                 return -1;
 230 #endif
 231
 232         txq = netdev_get_tx_queue(bp->dev, txdata->txq_index);
 233         hw_cons = le16_to_cpu(*txdata->tx_cons_sb);
 234         sw_cons = txdata->tx_pkt_cons;
 235
 236         while (sw_cons != hw_cons) {
 237                 u16 pkt_cons;
 238
 239                 pkt_cons = TX_BD(sw_cons);
 240
 241                 DP(NETIF_MSG_TX_DONE,
 242                    "queue[%d]: hw_cons %u  sw_cons %u  pkt_cons %u\n",
 243                    txdata->txq_index, hw_cons, sw_cons, pkt_cons);
 244
 245                 bd_cons = bnx2x_free_tx_pkt(bp, txdata, pkt_cons,
 246                                             &pkts_compl, &bytes_compl);
 247
 248                 sw_cons++;
 249         }
 250
 251         netdev_tx_completed_queue(txq, pkts_compl, bytes_compl);
 252
 253         txdata->tx_pkt_cons = sw_cons;
 254         txdata->tx_bd_cons = bd_cons;
 255
 256         /* Need to make the tx_bd_cons update visible to start_xmit()
 257          * before checking for netif_tx_queue_stopped().  Without the
 258          * memory barrier, there is a small possibility that
 259          * start_xmit() will miss it and cause the queue to be stopped
 260          * forever.
 261          * On the other hand we need an rmb() here to ensure the proper
 262          * ordering of bit testing in the following
 263          * netif_tx_queue_stopped(txq) call.
 264          */
 265         smp_mb();
 266
 267         if (unlikely(netif_tx_queue_stopped(txq))) {
 268                 /* Taking tx_lock() is needed to prevent re-enabling the queue
 269                  * while it's empty. This could have happen if rx_action() gets
 270                  * suspended in bnx2x_tx_int() after the condition before
 271                  * netif_tx_wake_queue(), while tx_action (bnx2x_start_xmit()):
 272                  *
 273                  * stops the queue->sees fresh tx_bd_cons->releases the queue->
 274                  * sends some packets consuming the whole queue again->
 275                  * stops the queue
 276                  */
 277
 278                 __netif_tx_lock(txq, smp_processor_id());
 279
 280                 if ((netif_tx_queue_stopped(txq)) &&
 281                     (bp->state == BNX2X_STATE_OPEN) &&
 282                     (bnx2x_tx_avail(bp, txdata) >= MAX_DESC_PER_TX_PKT))
 283                         netif_tx_wake_queue(txq);
 284
 285                 __netif_tx_unlock(txq);
 286         }
 287         return 0;
 288 }
 289
 290 static inline void bnx2x_update_last_max_sge(struct bnx2x_fastpath *fp,
 291                                              u16 idx)
 292 {
 293         u16 last_max = fp->last_max_sge;
 294
 295         if (SUB_S16(idx, last_max) > 0)
 296                 fp->last_max_sge = idx;
 297 }
 298
 299 static inline void bnx2x_update_sge_prod(struct bnx2x_fastpath *fp,
 300                                          u16 sge_len,
 301                                          struct eth_end_agg_rx_cqe *cqe)
 302 {
 303         struct bnx2x *bp = fp->bp;
 304         u16 last_max, last_elem, first_elem;
 305         u16 delta = 0;
 306         u16 i;
 307
 308         if (!sge_len)
 309                 return;
 310
 311         /* First mark all used pages */
 312         for (i = 0; i < sge_len; i++)
 313                 BIT_VEC64_CLEAR_BIT(fp->sge_mask,
 314                         RX_SGE(le16_to_cpu(cqe->sgl_or_raw_data.sgl[i])));
 315
 316         DP(NETIF_MSG_RX_STATUS, "fp_cqe->sgl[%d] = %d\n",
 317            sge_len - 1, le16_to_cpu(cqe->sgl_or_raw_data.sgl[sge_len - 1]));
 318
 319         /* Here we assume that the last SGE index is the biggest */
 320         prefetch((void *)(fp->sge_mask));
 321         bnx2x_update_last_max_sge(fp,
 322                 le16_to_cpu(cqe->sgl_or_raw_data.sgl[sge_len - 1]));
 323
 324         last_max = RX_SGE(fp->last_max_sge);
 325         last_elem = last_max >> BIT_VEC64_ELEM_SHIFT;
 326         first_elem = RX_SGE(fp->rx_sge_prod) >> BIT_VEC64_ELEM_SHIFT;
 327
 328         /* If ring is not full */
 329         if (last_elem + 1 != first_elem)
 330                 last_elem++;
 331
 332         /* Now update the prod */
 333         for (i = first_elem; i != last_elem; i = NEXT_SGE_MASK_ELEM(i)) {
 334                 if (likely(fp->sge_mask[i]))
 335                         break;
 336
 337                 fp->sge_mask[i] = BIT_VEC64_ELEM_ONE_MASK;
 338                 delta += BIT_VEC64_ELEM_SZ;
 339         }
 340
 341         if (delta > 0) {
 342                 fp->rx_sge_prod += delta;
 343                 /* clear page-end entries */
 344                 bnx2x_clear_sge_mask_next_elems(fp);
 345         }
 346
 347         DP(NETIF_MSG_RX_STATUS,
 348            "fp->last_max_sge = %d  fp->rx_sge_prod = %d\n",
 349            fp->last_max_sge, fp->rx_sge_prod);
 350 }
 351
 352 /* Get Toeplitz hash value in the skb using the value from the
 353  * CQE (calculated by HW).
 354  */
 355 static u32 bnx2x_get_rxhash(const struct bnx2x *bp,
 356                             const struct eth_fast_path_rx_cqe *cqe,
 357                             bool *l4_rxhash)
 358 {
 359         /* Get Toeplitz hash from CQE */
 360         if ((bp->dev->features & NETIF_F_RXHASH) &&
 361             (cqe->status_flags & ETH_FAST_PATH_RX_CQE_RSS_HASH_FLG)) {
 362                 enum eth_rss_hash_type htype;
 363
 364                 htype = cqe->status_flags & ETH_FAST_PATH_RX_CQE_RSS_HASH_TYPE;
 365                 *l4_rxhash = (htype == TCP_IPV4_HASH_TYPE) ||
 366                              (htype == TCP_IPV6_HASH_TYPE);
 367                 return le32_to_cpu(cqe->rss_hash_result);
 368         }
 369         *l4_rxhash = false;
 370         return 0;
 371 }
 372
 373 static void bnx2x_tpa_start(struct bnx2x_fastpath *fp, u16 queue,
 374                             u16 cons, u16 prod,
 375                             struct eth_fast_path_rx_cqe *cqe)
 376 {
 377         struct bnx2x *bp = fp->bp;
 378         struct sw_rx_bd *cons_rx_buf = &fp->rx_buf_ring[cons];
 379         struct sw_rx_bd *prod_rx_buf = &fp->rx_buf_ring[prod];
 380         struct eth_rx_bd *prod_bd = &fp->rx_desc_ring[prod];
 381         dma_addr_t mapping;
 382         struct bnx2x_agg_info *tpa_info = &fp->tpa_info[queue];
 383         struct sw_rx_bd *first_buf = &tpa_info->first_buf;
 384
 385         /* print error if current state != stop */
 386         if (tpa_info->tpa_state != BNX2X_TPA_STOP)
 387                 BNX2X_ERR("start of bin not in stop [%d]\n", queue);
 388
 389         /* Try to map an empty data buffer from the aggregation info  */
 390         mapping = dma_map_single(&bp->pdev->dev,
 391                                  first_buf->data + NET_SKB_PAD,
 392                                  fp->rx_buf_size, DMA_FROM_DEVICE);
 393         /*
 394          *  ...if it fails - move the skb from the consumer to the producer
 395          *  and set the current aggregation state as ERROR to drop it
 396          *  when TPA_STOP arrives.
 397          */
 398
 399         if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
 400                 /* Move the BD from the consumer to the producer */
 401                 bnx2x_reuse_rx_data(fp, cons, prod);
 402                 tpa_info->tpa_state = BNX2X_TPA_ERROR;
 403                 return;
 404         }
 405
 406         /* move empty data from pool to prod */
 407         prod_rx_buf->data = first_buf->data;
 408         dma_unmap_addr_set(prod_rx_buf, mapping, mapping);
 409         /* point prod_bd to new data */
 410         prod_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
 411         prod_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
 412
 413         /* move partial skb from cons to pool (don't unmap yet) */
 414         *first_buf = *cons_rx_buf;
 415
 416         /* mark bin state as START */
 417         tpa_info->parsing_flags =
 418                 le16_to_cpu(cqe->pars_flags.flags);
 419         tpa_info->vlan_tag = le16_to_cpu(cqe->vlan_tag);
 420         tpa_info->tpa_state = BNX2X_TPA_START;
 421         tpa_info->len_on_bd = le16_to_cpu(cqe->len_on_bd);
 422         tpa_info->placement_offset = cqe->placement_offset;
 423         tpa_info->rxhash = bnx2x_get_rxhash(bp, cqe, &tpa_info->l4_rxhash);
 424         if (fp->mode == TPA_MODE_GRO) {
 425                 u16 gro_size = le16_to_cpu(cqe->pkt_len_or_gro_seg_len);
 426                 tpa_info->full_page = SGE_PAGES / gro_size * gro_size;
 427                 tpa_info->gro_size = gro_size;
 428         }
 429
 430 #ifdef BNX2X_STOP_ON_ERROR
 431         fp->tpa_queue_used |= (1 << queue);
 432 #ifdef _ASM_GENERIC_INT_L64_H
 433         DP(NETIF_MSG_RX_STATUS, "fp->tpa_queue_used = 0x%lx\n",
 434 #else
 435         DP(NETIF_MSG_RX_STATUS, "fp->tpa_queue_used = 0x%llx\n",
 436 #endif
 437            fp->tpa_queue_used);
 438 #endif
 439 }
 440
 441 /* Timestamp option length allowed for TPA aggregation:
 442  *
 443  *              nop nop kind length echo val
 444  */
 445 #define TPA_TSTAMP_OPT_LEN      12
 446 /**
 447  * bnx2x_set_gro_params - compute GRO values
 448  *
 449  * @skb:                packet skb
 450  * @parsing_flags:      parsing flags from the START CQE
 451  * @len_on_bd:          total length of the first packet for the
 452  *                      aggregation.
 453  * @pkt_len:            length of all segments
 454  *
 455  * Approximate value of the MSS for this aggregation calculated using
 456  * the first packet of it.
 457  * Compute number of aggregated segments, and gso_type.
 458  */
 459 static void bnx2x_set_gro_params(struct sk_buff *skb, u16 parsing_flags,
 460                                  u16 len_on_bd, unsigned int pkt_len,
 461                                  u16 num_of_coalesced_segs)
 462 {
 463         /* TPA aggregation won't have either IP options or TCP options
 464          * other than timestamp or IPv6 extension headers.
 465          */
 466         u16 hdrs_len = ETH_HLEN + sizeof(struct tcphdr);
 467
 468         if (GET_FLAG(parsing_flags, PARSING_FLAGS_OVER_ETHERNET_PROTOCOL) ==
 469             PRS_FLAG_OVERETH_IPV6) {
 470                 hdrs_len += sizeof(struct ipv6hdr);
 471                 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
 472         } else {
 473                 hdrs_len += sizeof(struct iphdr);
 474                 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
 475         }
 476
 477         /* Check if there was a TCP timestamp, if there is it's will
 478          * always be 12 bytes length: nop nop kind length echo val.
 479          *
 480          * Otherwise FW would close the aggregation.
 481          */
 482         if (parsing_flags & PARSING_FLAGS_TIME_STAMP_EXIST_FLAG)
 483                 hdrs_len += TPA_TSTAMP_OPT_LEN;
 484
 485         skb_shinfo(skb)->gso_size = len_on_bd - hdrs_len;
 486
 487         /* tcp_gro_complete() will copy NAPI_GRO_CB(skb)->count
 488          * to skb_shinfo(skb)->gso_segs
 489          */
 490         NAPI_GRO_CB(skb)->count = num_of_coalesced_segs;
 491 }
 492
 493 static int bnx2x_alloc_rx_sge(struct bnx2x *bp, struct bnx2x_fastpath *fp,
 494                               u16 index, gfp_t gfp_mask)
 495 {
 496         struct page *page = alloc_pages(gfp_mask, PAGES_PER_SGE_SHIFT);
 497         struct sw_rx_page *sw_buf = &fp->rx_page_ring[index];
 498         struct eth_rx_sge *sge = &fp->rx_sge_ring[index];
 499         dma_addr_t mapping;
 500
 501         if (unlikely(page == NULL)) {
 502                 BNX2X_ERR("Can't alloc sge\n");
 503                 return -ENOMEM;
 504         }
 505
 506         mapping = dma_map_page(&bp->pdev->dev, page, 0,
 507                                SGE_PAGES, DMA_FROM_DEVICE);
 508         if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
 509                 __free_pages(page, PAGES_PER_SGE_SHIFT);
 510                 BNX2X_ERR("Can't map sge\n");
 511                 return -ENOMEM;
 512         }
 513
 514         sw_buf->page = page;
 515         dma_unmap_addr_set(sw_buf, mapping, mapping);
 516
 517         sge->addr_hi = cpu_to_le32(U64_HI(mapping));
 518         sge->addr_lo = cpu_to_le32(U64_LO(mapping));
 519
 520         return 0;
 521 }
 522
 523 static int bnx2x_fill_frag_skb(struct bnx2x *bp, struct bnx2x_fastpath *fp,
 524                                struct bnx2x_agg_info *tpa_info,
 525                                u16 pages,
 526                                struct sk_buff *skb,
 527                                struct eth_end_agg_rx_cqe *cqe,
 528                                u16 cqe_idx)
 529 {
 530         struct sw_rx_page *rx_pg, old_rx_pg;
 531         u32 i, frag_len, frag_size;
 532         int err, j, frag_id = 0;
 533         u16 len_on_bd = tpa_info->len_on_bd;
 534         u16 full_page = 0, gro_size = 0;
 535
 536         frag_size = le16_to_cpu(cqe->pkt_len) - len_on_bd;
 537
 538         if (fp->mode == TPA_MODE_GRO) {
 539                 gro_size = tpa_info->gro_size;
 540                 full_page = tpa_info->full_page;
 541         }
 542
 543         /* This is needed in order to enable forwarding support */
 544         if (frag_size)
 545                 bnx2x_set_gro_params(skb, tpa_info->parsing_flags, len_on_bd,
 546                                      le16_to_cpu(cqe->pkt_len),
 547                                      le16_to_cpu(cqe->num_of_coalesced_segs));
 548
 549 #ifdef BNX2X_STOP_ON_ERROR
 550         if (pages > min_t(u32, 8, MAX_SKB_FRAGS) * SGE_PAGES) {
 551                 BNX2X_ERR("SGL length is too long: %d. CQE index is %d\n",
 552                           pages, cqe_idx);
 553                 BNX2X_ERR("cqe->pkt_len = %d\n", cqe->pkt_len);
 554                 bnx2x_panic();
 555                 return -EINVAL;
 556         }
 557 #endif
 558
 559         /* Run through the SGL and compose the fragmented skb */
 560         for (i = 0, j = 0; i < pages; i += PAGES_PER_SGE, j++) {
 561                 u16 sge_idx = RX_SGE(le16_to_cpu(cqe->sgl_or_raw_data.sgl[j]));
 562
 563                 /* FW gives the indices of the SGE as if the ring is an array
 564                    (meaning that "next" element will consume 2 indices) */
 565                 if (fp->mode == TPA_MODE_GRO)
 566                         frag_len = min_t(u32, frag_size, (u32)full_page);
 567                 else /* LRO */
 568                         frag_len = min_t(u32, frag_size, (u32)SGE_PAGES);
 569
 570                 rx_pg = &fp->rx_page_ring[sge_idx];
 571                 old_rx_pg = *rx_pg;
 572
 573                 /* If we fail to allocate a substitute page, we simply stop
 574                    where we are and drop the whole packet */
 575                 err = bnx2x_alloc_rx_sge(bp, fp, sge_idx, GFP_ATOMIC);
 576                 if (unlikely(err)) {
 577                         bnx2x_fp_qstats(bp, fp)->rx_skb_alloc_failed++;
 578                         return err;
 579                 }
 580
 581                 /* Unmap the page as we're going to pass it to the stack */
 582                 dma_unmap_page(&bp->pdev->dev,
 583                                dma_unmap_addr(&old_rx_pg, mapping),
 584                                SGE_PAGES, DMA_FROM_DEVICE);
 585                 /* Add one frag and update the appropriate fields in the skb */
 586                 if (fp->mode == TPA_MODE_LRO)
 587                         skb_fill_page_desc(skb, j, old_rx_pg.page, 0, frag_len);
 588                 else { /* GRO */
 589                         int rem;
 590                         int offset = 0;
 591                         for (rem = frag_len; rem > 0; rem -= gro_size) {
 592                                 int len = rem > gro_size ? gro_size : rem;
 593                                 skb_fill_page_desc(skb, frag_id++,
 594                                                    old_rx_pg.page, offset, len);
 595                                 if (offset)
 596                                         get_page(old_rx_pg.page);
 597                                 offset += len;
 598                         }
 599                 }
 600
 601                 skb->data_len += frag_len;
 602                 skb->truesize += SGE_PAGES;
 603                 skb->len += frag_len;
 604
 605                 frag_size -= frag_len;
 606         }
 607
 608         return 0;
 609 }
 610
 611 static void bnx2x_frag_free(const struct bnx2x_fastpath *fp, void *data)
 612 {
 613         if (fp->rx_frag_size)
 614                 put_page(virt_to_head_page(data));
 615         else
 616                 kfree(data);
 617 }
 618
 619 static void *bnx2x_frag_alloc(const struct bnx2x_fastpath *fp, gfp_t gfp_mask)
 620 {
 621         if (fp->rx_frag_size) {
 622                 /* GFP_KERNEL allocations are used only during initialization */
 623                 if (unlikely(gfp_mask & __GFP_WAIT))
 624                         return (void *)__get_free_page(gfp_mask);
 625
 626                 return netdev_alloc_frag(fp->rx_frag_size);
 627         }
 628
 629         return kmalloc(fp->rx_buf_size + NET_SKB_PAD, gfp_mask);
 630 }
 631
 632 #ifdef CONFIG_INET
 633 static void bnx2x_gro_ip_csum(struct bnx2x *bp, struct sk_buff *skb)
 634 {
 635         const struct iphdr *iph = ip_hdr(skb);
 636         struct tcphdr *th;
 637
 638         skb_set_transport_header(skb, sizeof(struct iphdr));
 639         th = tcp_hdr(skb);
 640
 641         th->check = ~tcp_v4_check(skb->len - skb_transport_offset(skb),
 642                                   iph->saddr, iph->daddr, 0);
 643 }
 644
 645 static void bnx2x_gro_ipv6_csum(struct bnx2x *bp, struct sk_buff *skb)
 646 {
 647         struct ipv6hdr *iph = ipv6_hdr(skb);
 648         struct tcphdr *th;
 649
 650         skb_set_transport_header(skb, sizeof(struct ipv6hdr));
 651         th = tcp_hdr(skb);
 652
 653         th->check = ~tcp_v6_check(skb->len - skb_transport_offset(skb),
 654                                   &iph->saddr, &iph->daddr, 0);
 655 }
 656
 657 static void bnx2x_gro_csum(struct bnx2x *bp, struct sk_buff *skb,
 658                             void (*gro_func)(struct bnx2x*, struct sk_buff*))
 659 {
 660         skb_set_network_header(skb, 0);
 661         gro_func(bp, skb);
 662         tcp_gro_complete(skb);
 663 }
 664 #endif
 665
 666 static void bnx2x_gro_receive(struct bnx2x *bp, struct bnx2x_fastpath *fp,
 667                                struct sk_buff *skb)
 668 {
 669 #ifdef CONFIG_INET
 670         if (skb_shinfo(skb)->gso_size) {
 671                 switch (be16_to_cpu(skb->protocol)) {
 672                 case ETH_P_IP:
 673                         bnx2x_gro_csum(bp, skb, bnx2x_gro_ip_csum);
 674                         break;
 675                 case ETH_P_IPV6:
 676                         bnx2x_gro_csum(bp, skb, bnx2x_gro_ipv6_csum);
 677                         break;
 678                 default:
 679                         BNX2X_ERR("Error: FW GRO supports only IPv4/IPv6, not 0x%04x\n",
 680                                   be16_to_cpu(skb->protocol));
 681                 }
 682         }
 683 #endif
 684         skb_record_rx_queue(skb, fp->rx_queue);
 685         napi_gro_receive(&fp->napi, skb);
 686 }
 687
 688 static void bnx2x_tpa_stop(struct bnx2x *bp, struct bnx2x_fastpath *fp,
 689                            struct bnx2x_agg_info *tpa_info,
 690                            u16 pages,
 691                            struct eth_end_agg_rx_cqe *cqe,
 692                            u16 cqe_idx)
 693 {
 694         struct sw_rx_bd *rx_buf = &tpa_info->first_buf;
 695         u8 pad = tpa_info->placement_offset;
 696         u16 len = tpa_info->len_on_bd;
 697         struct sk_buff *skb = NULL;
 698         u8 *new_data, *data = rx_buf->data;
 699         u8 old_tpa_state = tpa_info->tpa_state;
 700
 701         tpa_info->tpa_state = BNX2X_TPA_STOP;
 702
 703         /* If we there was an error during the handling of the TPA_START -
 704          * drop this aggregation.
 705          */
 706         if (old_tpa_state == BNX2X_TPA_ERROR)
 707                 goto drop;
 708
 709         /* Try to allocate the new data */
 710         new_data = bnx2x_frag_alloc(fp, GFP_ATOMIC);
 711         /* Unmap skb in the pool anyway, as we are going to change
 712            pool entry status to BNX2X_TPA_STOP even if new skb allocation
 713            fails. */
 714         dma_unmap_single(&bp->pdev->dev, dma_unmap_addr(rx_buf, mapping),
 715                          fp->rx_buf_size, DMA_FROM_DEVICE);
 716         if (likely(new_data))
 717                 skb = build_skb(data, fp->rx_frag_size);
 718
 719         if (likely(skb)) {
 720 #ifdef BNX2X_STOP_ON_ERROR
 721                 if (pad + len > fp->rx_buf_size) {
 722                         BNX2X_ERR("skb_put is about to fail...  pad %d  len %d  rx_buf_size %d\n",
 723                                   pad, len, fp->rx_buf_size);
 724                         bnx2x_panic();
 725                         return;
 726                 }
 727 #endif
 728
 729                 skb_reserve(skb, pad + NET_SKB_PAD);
 730                 skb_put(skb, len);
 731                 skb->rxhash = tpa_info->rxhash;
 732                 skb->l4_rxhash = tpa_info->l4_rxhash;
 733
 734                 skb->protocol = eth_type_trans(skb, bp->dev);
 735                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 736
 737                 if (!bnx2x_fill_frag_skb(bp, fp, tpa_info, pages,
 738                                          skb, cqe, cqe_idx)) {
 739                         if (tpa_info->parsing_flags & PARSING_FLAGS_VLAN)
 740                                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), tpa_info->vlan_tag);
 741                         bnx2x_gro_receive(bp, fp, skb);
 742                 } else {
 743                         DP(NETIF_MSG_RX_STATUS,
 744                            "Failed to allocate new pages - dropping packet!\n");
 745                         dev_kfree_skb_any(skb);
 746                 }
 747
 748                 /* put new data in bin */
 749                 rx_buf->data = new_data;
 750
 751                 return;
 752         }
 753         bnx2x_frag_free(fp, new_data);
 754 drop:
 755         /* drop the packet and keep the buffer in the bin */
 756         DP(NETIF_MSG_RX_STATUS,
 757            "Failed to allocate or map a new skb - dropping packet!\n");
 758         bnx2x_fp_stats(bp, fp)->eth_q_stats.rx_skb_alloc_failed++;
 759 }
 760
 761 static int bnx2x_alloc_rx_data(struct bnx2x *bp, struct bnx2x_fastpath *fp,
 762                                u16 index, gfp_t gfp_mask)
 763 {
 764         u8 *data;
 765         struct sw_rx_bd *rx_buf = &fp->rx_buf_ring[index];
 766         struct eth_rx_bd *rx_bd = &fp->rx_desc_ring[index];
 767         dma_addr_t mapping;
 768
 769         data = bnx2x_frag_alloc(fp, gfp_mask);
 770         if (unlikely(data == NULL))
 771                 return -ENOMEM;
 772
 773         mapping = dma_map_single(&bp->pdev->dev, data + NET_SKB_PAD,
 774                                  fp->rx_buf_size,
 775                                  DMA_FROM_DEVICE);
 776         if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
 777                 bnx2x_frag_free(fp, data);
 778                 BNX2X_ERR("Can't map rx data\n");
 779                 return -ENOMEM;
 780         }
 781
 782         rx_buf->data = data;
 783         dma_unmap_addr_set(rx_buf, mapping, mapping);
 784
 785         rx_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
 786         rx_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
 787
 788         return 0;
 789 }
 790
 791 static
 792 void bnx2x_csum_validate(struct sk_buff *skb, union eth_rx_cqe *cqe,
 793                                  struct bnx2x_fastpath *fp,
 794                                  struct bnx2x_eth_q_stats *qstats)
 795 {
 796         /* Do nothing if no L4 csum validation was done.
 797          * We do not check whether IP csum was validated. For IPv4 we assume
 798          * that if the card got as far as validating the L4 csum, it also
 799          * validated the IP csum. IPv6 has no IP csum.
 800          */
 801         if (cqe->fast_path_cqe.status_flags &
 802             ETH_FAST_PATH_RX_CQE_L4_XSUM_NO_VALIDATION_FLG)
 803                 return;
 804
 805         /* If L4 validation was done, check if an error was found. */
 806
 807         if (cqe->fast_path_cqe.type_error_flags &
 808             (ETH_FAST_PATH_RX_CQE_IP_BAD_XSUM_FLG |
 809              ETH_FAST_PATH_RX_CQE_L4_BAD_XSUM_FLG))
 810                 qstats->hw_csum_err++;
 811         else
 812                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 813 }
 814
 815 int bnx2x_rx_int(struct bnx2x_fastpath *fp, int budget)
 816 {
 817         struct bnx2x *bp = fp->bp;
 818         u16 bd_cons, bd_prod, bd_prod_fw, comp_ring_cons;
 819         u16 sw_comp_cons, sw_comp_prod;
 820         int rx_pkt = 0;
 821         union eth_rx_cqe *cqe;
 822         struct eth_fast_path_rx_cqe *cqe_fp;
 823
 824 #ifdef BNX2X_STOP_ON_ERROR
 825         if (unlikely(bp->panic))
 826                 return 0;
 827 #endif
 828
 829         bd_cons = fp->rx_bd_cons;
 830         bd_prod = fp->rx_bd_prod;
 831         bd_prod_fw = bd_prod;
 832         sw_comp_cons = fp->rx_comp_cons;
 833         sw_comp_prod = fp->rx_comp_prod;
 834
 835         comp_ring_cons = RCQ_BD(sw_comp_cons);
 836         cqe = &fp->rx_comp_ring[comp_ring_cons];
 837         cqe_fp = &cqe->fast_path_cqe;
 838
 839         DP(NETIF_MSG_RX_STATUS,
 840            "queue[%d]: sw_comp_cons %u\n", fp->index, sw_comp_cons);
 841
 842         while (BNX2X_IS_CQE_COMPLETED(cqe_fp)) {
 843                 struct sw_rx_bd *rx_buf = NULL;
 844                 struct sk_buff *skb;
 845                 u8 cqe_fp_flags;
 846                 enum eth_rx_cqe_type cqe_fp_type;
 847                 u16 len, pad, queue;
 848                 u8 *data;
 849                 bool l4_rxhash;
 850
 851 #ifdef BNX2X_STOP_ON_ERROR
 852                 if (unlikely(bp->panic))
 853                         return 0;
 854 #endif
 855
 856                 bd_prod = RX_BD(bd_prod);
 857                 bd_cons = RX_BD(bd_cons);
 858
 859                 cqe_fp_flags = cqe_fp->type_error_flags;
 860                 cqe_fp_type = cqe_fp_flags & ETH_FAST_PATH_RX_CQE_TYPE;
 861
 862                 DP(NETIF_MSG_RX_STATUS,
 863                    "CQE type %x  err %x  status %x  queue %x  vlan %x  len %u\n",
 864                    CQE_TYPE(cqe_fp_flags),
 865                    cqe_fp_flags, cqe_fp->status_flags,
 866                    le32_to_cpu(cqe_fp->rss_hash_result),
 867                    le16_to_cpu(cqe_fp->vlan_tag),
 868                    le16_to_cpu(cqe_fp->pkt_len_or_gro_seg_len));
 869
 870                 /* is this a slowpath msg? */
 871                 if (unlikely(CQE_TYPE_SLOW(cqe_fp_type))) {
 872                         bnx2x_sp_event(fp, cqe);
 873                         goto next_cqe;
 874                 }
 875
 876                 rx_buf = &fp->rx_buf_ring[bd_cons];
 877                 data = rx_buf->data;
 878
 879                 if (!CQE_TYPE_FAST(cqe_fp_type)) {
 880                         struct bnx2x_agg_info *tpa_info;
 881                         u16 frag_size, pages;
 882 #ifdef BNX2X_STOP_ON_ERROR
 883                         /* sanity check */
 884                         if (fp->disable_tpa &&
 885                             (CQE_TYPE_START(cqe_fp_type) ||
 886                              CQE_TYPE_STOP(cqe_fp_type)))
 887                                 BNX2X_ERR("START/STOP packet while disable_tpa type %x\n",
 888                                           CQE_TYPE(cqe_fp_type));
 889 #endif
 890
 891                         if (CQE_TYPE_START(cqe_fp_type)) {
 892                                 u16 queue = cqe_fp->queue_index;
 893                                 DP(NETIF_MSG_RX_STATUS,
 894                                    "calling tpa_start on queue %d\n",
 895                                    queue);
 896
 897                                 bnx2x_tpa_start(fp, queue,
 898                                                 bd_cons, bd_prod,
 899                                                 cqe_fp);
 900
 901                                 goto next_rx;
 902                         }
 903                         queue = cqe->end_agg_cqe.queue_index;
 904                         tpa_info = &fp->tpa_info[queue];
 905                         DP(NETIF_MSG_RX_STATUS,
 906                            "calling tpa_stop on queue %d\n",
 907                            queue);
 908
 909                         frag_size = le16_to_cpu(cqe->end_agg_cqe.pkt_len) -
 910                                     tpa_info->len_on_bd;
 911
 912                         if (fp->mode == TPA_MODE_GRO)
 913                                 pages = (frag_size + tpa_info->full_page - 1) /
 914                                          tpa_info->full_page;
 915                         else
 916                                 pages = SGE_PAGE_ALIGN(frag_size) >>
 917                                         SGE_PAGE_SHIFT;
 918
 919                         bnx2x_tpa_stop(bp, fp, tpa_info, pages,
 920                                        &cqe->end_agg_cqe, comp_ring_cons);
 921 #ifdef BNX2X_STOP_ON_ERROR
 922                         if (bp->panic)
 923                                 return 0;
 924 #endif
 925
 926                         bnx2x_update_sge_prod(fp, pages, &cqe->end_agg_cqe);
 927                         goto next_cqe;
 928                 }
 929                 /* non TPA */
 930                 len = le16_to_cpu(cqe_fp->pkt_len_or_gro_seg_len);
 931                 pad = cqe_fp->placement_offset;
 932                 dma_sync_single_for_cpu(&bp->pdev->dev,
 933                                         dma_unmap_addr(rx_buf, mapping),
 934                                         pad + RX_COPY_THRESH,
 935                                         DMA_FROM_DEVICE);
 936                 pad += NET_SKB_PAD;
 937                 prefetch(data + pad); /* speedup eth_type_trans() */
 938                 /* is this an error packet? */
 939                 if (unlikely(cqe_fp_flags & ETH_RX_ERROR_FALGS)) {
 940                         DP(NETIF_MSG_RX_ERR | NETIF_MSG_RX_STATUS,
 941                            "ERROR  flags %x  rx packet %u\n",
 942                            cqe_fp_flags, sw_comp_cons);
 943                         bnx2x_fp_qstats(bp, fp)->rx_err_discard_pkt++;
 944                         goto reuse_rx;
 945                 }
 946
 947                 /* Since we don't have a jumbo ring
 948                  * copy small packets if mtu > 1500
 949                  */
 950                 if ((bp->dev->mtu > ETH_MAX_PACKET_SIZE) &&
 951                     (len <= RX_COPY_THRESH)) {
 952                         skb = netdev_alloc_skb_ip_align(bp->dev, len);
 953                         if (skb == NULL) {
 954                                 DP(NETIF_MSG_RX_ERR | NETIF_MSG_RX_STATUS,
 955                                    "ERROR  packet dropped because of alloc failure\n");
 956                                 bnx2x_fp_qstats(bp, fp)->rx_skb_alloc_failed++;
 957                                 goto reuse_rx;
 958                         }
 959                         memcpy(skb->data, data + pad, len);
 960                         bnx2x_reuse_rx_data(fp, bd_cons, bd_prod);
 961                 } else {
 962                         if (likely(bnx2x_alloc_rx_data(bp, fp, bd_prod,
 963                                                        GFP_ATOMIC) == 0)) {
 964                                 dma_unmap_single(&bp->pdev->dev,
 965                                                  dma_unmap_addr(rx_buf, mapping),
 966                                                  fp->rx_buf_size,
 967                                                  DMA_FROM_DEVICE);
 968                                 skb = build_skb(data, fp->rx_frag_size);
 969                                 if (unlikely(!skb)) {
 970                                         bnx2x_frag_free(fp, data);
 971                                         bnx2x_fp_qstats(bp, fp)->
 972                                                         rx_skb_alloc_failed++;
 973                                         goto next_rx;
 974                                 }
 975                                 skb_reserve(skb, pad);
 976                         } else {
 977                                 DP(NETIF_MSG_RX_ERR | NETIF_MSG_RX_STATUS,
 978                                    "ERROR  packet dropped because of alloc failure\n");
 979                                 bnx2x_fp_qstats(bp, fp)->rx_skb_alloc_failed++;
 980 reuse_rx:
 981                                 bnx2x_reuse_rx_data(fp, bd_cons, bd_prod);
 982                                 goto next_rx;
 983                         }
 984                 }
 985
 986                 skb_put(skb, len);
 987                 skb->protocol = eth_type_trans(skb, bp->dev);
 988
 989                 /* Set Toeplitz hash for a none-LRO skb */
 990                 skb->rxhash = bnx2x_get_rxhash(bp, cqe_fp, &l4_rxhash);
 991                 skb->l4_rxhash = l4_rxhash;
 992
 993                 skb_checksum_none_assert(skb);
 994
 995                 if (bp->dev->features & NETIF_F_RXCSUM)
 996                         bnx2x_csum_validate(skb, cqe, fp,
 997                                             bnx2x_fp_qstats(bp, fp));
 998
 999                 skb_record_rx_queue(skb, fp->rx_queue);
1000
1001                 if (le16_to_cpu(cqe_fp->pars_flags.flags) &
1002                     PARSING_FLAGS_VLAN)
1003                         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
1004                                                le16_to_cpu(cqe_fp->vlan_tag));
1005
1006                 skb_mark_napi_id(skb, &fp->napi);
1007
1008                 if (bnx2x_fp_ll_polling(fp))
1009                         netif_receive_skb(skb);
1010                 else
1011                         napi_gro_receive(&fp->napi, skb);
1012 next_rx:
1013                 rx_buf->data = NULL;
1014
1015                 bd_cons = NEXT_RX_IDX(bd_cons);
1016                 bd_prod = NEXT_RX_IDX(bd_prod);
1017                 bd_prod_fw = NEXT_RX_IDX(bd_prod_fw);
1018                 rx_pkt++;
1019 next_cqe:
1020                 sw_comp_prod = NEXT_RCQ_IDX(sw_comp_prod);
1021                 sw_comp_cons = NEXT_RCQ_IDX(sw_comp_cons);
1022
1023                 /* mark CQE as free */
1024                 BNX2X_SEED_CQE(cqe_fp);
1025
1026                 if (rx_pkt == budget)
1027                         break;
1028
1029                 comp_ring_cons = RCQ_BD(sw_comp_cons);
1030                 cqe = &fp->rx_comp_ring[comp_ring_cons];
1031                 cqe_fp = &cqe->fast_path_cqe;
1032         } /* while */
1033
1034         fp->rx_bd_cons = bd_cons;
1035         fp->rx_bd_prod = bd_prod_fw;
1036         fp->rx_comp_cons = sw_comp_cons;
1037         fp->rx_comp_prod = sw_comp_prod;
1038
1039         /* Update producers */
1040         bnx2x_update_rx_prod(bp, fp, bd_prod_fw, sw_comp_prod,
1041                              fp->rx_sge_prod);
1042
1043         fp->rx_pkt += rx_pkt;
1044         fp->rx_calls++;
1045
1046         return rx_pkt;
1047 }
1048
1049 static irqreturn_t bnx2x_msix_fp_int(int irq, void *fp_cookie)
1050 {
1051         struct bnx2x_fastpath *fp = fp_cookie;
1052         struct bnx2x *bp = fp->bp;
1053         u8 cos;
1054
1055         DP(NETIF_MSG_INTR,
1056            "got an MSI-X interrupt on IDX:SB [fp %d fw_sd %d igusb %d]\n",
1057            fp->index, fp->fw_sb_id, fp->igu_sb_id);
1058
1059         bnx2x_ack_sb(bp, fp->igu_sb_id, USTORM_ID, 0, IGU_INT_DISABLE, 0);
1060
1061 #ifdef BNX2X_STOP_ON_ERROR
1062         if (unlikely(bp->panic))
1063                 return IRQ_HANDLED;
1064 #endif
1065
1066         /* Handle Rx and Tx according to MSI-X vector */
1067         for_each_cos_in_tx_queue(fp, cos)
1068                 prefetch(fp->txdata_ptr[cos]->tx_cons_sb);
1069
1070         prefetch(&fp->sb_running_index[SM_RX_ID]);
1071         napi_schedule(&bnx2x_fp(bp, fp->index, napi));
1072
1073         return IRQ_HANDLED;
1074 }
1075
1076 /* HW Lock for shared dual port PHYs */
1077 void bnx2x_acquire_phy_lock(struct bnx2x *bp)
1078 {
1079         mutex_lock(&bp->port.phy_mutex);
1080
1081         bnx2x_acquire_hw_lock(bp, HW_LOCK_RESOURCE_MDIO);
1082 }
1083
1084 void bnx2x_release_phy_lock(struct bnx2x *bp)
1085 {
1086         bnx2x_release_hw_lock(bp, HW_LOCK_RESOURCE_MDIO);
1087
1088         mutex_unlock(&bp->port.phy_mutex);
1089 }
1090
1091 /* calculates MF speed according to current linespeed and MF configuration */
1092 u16 bnx2x_get_mf_speed(struct bnx2x *bp)
1093 {
1094         u16 line_speed = bp->link_vars.line_speed;
1095         if (IS_MF(bp)) {
1096                 u16 maxCfg = bnx2x_extract_max_cfg(bp,
1097                                                    bp->mf_config[BP_VN(bp)]);
1098
1099                 /* Calculate the current MAX line speed limit for the MF
1100                  * devices
1101                  */
1102                 if (IS_MF_SI(bp))
1103                         line_speed = (line_speed * maxCfg) / 100;
1104                 else { /* SD mode */
1105                         u16 vn_max_rate = maxCfg * 100;
1106
1107                         if (vn_max_rate < line_speed)
1108                                 line_speed = vn_max_rate;
1109                 }
1110         }
1111
1112         return line_speed;
1113 }
1114
1115 /**
1116  * bnx2x_fill_report_data - fill link report data to report
1117  *
1118  * @bp:         driver handle
1119  * @data:       link state to update
1120  *
1121  * It uses a none-atomic bit operations because is called under the mutex.
1122  */
1123 static void bnx2x_fill_report_data(struct bnx2x *bp,
1124                                    struct bnx2x_link_report_data *data)
1125 {
1126         u16 line_speed = bnx2x_get_mf_speed(bp);
1127
1128         memset(data, 0, sizeof(*data));
1129
1130         /* Fill the report data: effective line speed */
1131         data->line_speed = line_speed;
1132
1133         /* Link is down */
1134         if (!bp->link_vars.link_up || (bp->flags & MF_FUNC_DIS))
1135                 __set_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1136                           &data->link_report_flags);
1137
1138         /* Full DUPLEX */
1139         if (bp->link_vars.duplex == DUPLEX_FULL)
1140                 __set_bit(BNX2X_LINK_REPORT_FD, &data->link_report_flags);
1141
1142         /* Rx Flow Control is ON */
1143         if (bp->link_vars.flow_ctrl & BNX2X_FLOW_CTRL_RX)
1144                 __set_bit(BNX2X_LINK_REPORT_RX_FC_ON, &data->link_report_flags);
1145
1146         /* Tx Flow Control is ON */
1147         if (bp->link_vars.flow_ctrl & BNX2X_FLOW_CTRL_TX)
1148                 __set_bit(BNX2X_LINK_REPORT_TX_FC_ON, &data->link_report_flags);
1149 }
1150
1151 /**
1152  * bnx2x_link_report - report link status to OS.
1153  *
1154  * @bp:         driver handle
1155  *
1156  * Calls the __bnx2x_link_report() under the same locking scheme
1157  * as a link/PHY state managing code to ensure a consistent link
1158  * reporting.
1159  */
1160
1161 void bnx2x_link_report(struct bnx2x *bp)
1162 {
1163         bnx2x_acquire_phy_lock(bp);
1164         __bnx2x_link_report(bp);
1165         bnx2x_release_phy_lock(bp);
1166 }
1167
1168 /**
1169  * __bnx2x_link_report - report link status to OS.
1170  *
1171  * @bp:         driver handle
1172  *
1173  * None atomic implementation.
1174  * Should be called under the phy_lock.
1175  */
1176 void __bnx2x_link_report(struct bnx2x *bp)
1177 {
1178         struct bnx2x_link_report_data cur_data;
1179
1180         /* reread mf_cfg */
1181         if (IS_PF(bp) && !CHIP_IS_E1(bp))
1182                 bnx2x_read_mf_cfg(bp);
1183
1184         /* Read the current link report info */
1185         bnx2x_fill_report_data(bp, &cur_data);
1186
1187         /* Don't report link down or exactly the same link status twice */
1188         if (!memcmp(&cur_data, &bp->last_reported_link, sizeof(cur_data)) ||
1189             (test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1190                       &bp->last_reported_link.link_report_flags) &&
1191              test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1192                       &cur_data.link_report_flags)))
1193                 return;
1194
1195         bp->link_cnt++;
1196
1197         /* We are going to report a new link parameters now -
1198          * remember the current data for the next time.
1199          */
1200         memcpy(&bp->last_reported_link, &cur_data, sizeof(cur_data));
1201
1202         if (test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1203                      &cur_data.link_report_flags)) {
1204                 netif_carrier_off(bp->dev);
1205                 netdev_err(bp->dev, "NIC Link is Down\n");
1206                 return;
1207         } else {
1208                 const char *duplex;
1209                 const char *flow;
1210
1211                 netif_carrier_on(bp->dev);
1212
1213                 if (test_and_clear_bit(BNX2X_LINK_REPORT_FD,
1214                                        &cur_data.link_report_flags))
1215                         duplex = "full";
1216                 else
1217                         duplex = "half";
1218
1219                 /* Handle the FC at the end so that only these flags would be
1220                  * possibly set. This way we may easily check if there is no FC
1221                  * enabled.
1222                  */
1223                 if (cur_data.link_report_flags) {
1224                         if (test_bit(BNX2X_LINK_REPORT_RX_FC_ON,
1225                                      &cur_data.link_report_flags)) {
1226                                 if (test_bit(BNX2X_LINK_REPORT_TX_FC_ON,
1227                                      &cur_data.link_report_flags))
1228                                         flow = "ON - receive & transmit";
1229                                 else
1230                                         flow = "ON - receive";
1231                         } else {
1232                                 flow = "ON - transmit";
1233                         }
1234                 } else {
1235                         flow = "none";
1236                 }
1237                 netdev_info(bp->dev, "NIC Link is Up, %d Mbps %s duplex, Flow control: %s\n",
1238                             cur_data.line_speed, duplex, flow);
1239         }
1240 }
1241
1242 static void bnx2x_set_next_page_sgl(struct bnx2x_fastpath *fp)
1243 {
1244         int i;
1245
1246         for (i = 1; i <= NUM_RX_SGE_PAGES; i++) {
1247                 struct eth_rx_sge *sge;
1248
1249                 sge = &fp->rx_sge_ring[RX_SGE_CNT * i - 2];
1250                 sge->addr_hi =
1251                         cpu_to_le32(U64_HI(fp->rx_sge_mapping +
1252                         BCM_PAGE_SIZE*(i % NUM_RX_SGE_PAGES)));
1253
1254                 sge->addr_lo =
1255                         cpu_to_le32(U64_LO(fp->rx_sge_mapping +
1256                         BCM_PAGE_SIZE*(i % NUM_RX_SGE_PAGES)));
1257         }
1258 }
1259
1260 static void bnx2x_free_tpa_pool(struct bnx2x *bp,
1261                                 struct bnx2x_fastpath *fp, int last)
1262 {
1263         int i;
1264
1265         for (i = 0; i < last; i++) {
1266                 struct bnx2x_agg_info *tpa_info = &fp->tpa_info[i];
1267                 struct sw_rx_bd *first_buf = &tpa_info->first_buf;
1268                 u8 *data = first_buf->data;
1269
1270                 if (data == NULL) {
1271                         DP(NETIF_MSG_IFDOWN, "tpa bin %d empty on free\n", i);
1272                         continue;
1273                 }
1274                 if (tpa_info->tpa_state == BNX2X_TPA_START)
1275                         dma_unmap_single(&bp->pdev->dev,
1276                                          dma_unmap_addr(first_buf, mapping),
1277                                          fp->rx_buf_size, DMA_FROM_DEVICE);
1278                 bnx2x_frag_free(fp, data);
1279                 first_buf->data = NULL;
1280         }
1281 }
1282
1283 void bnx2x_init_rx_rings_cnic(struct bnx2x *bp)
1284 {
1285         int j;
1286
1287         for_each_rx_queue_cnic(bp, j) {
1288                 struct bnx2x_fastpath *fp = &bp->fp[j];
1289
1290                 fp->rx_bd_cons = 0;
1291
1292                 /* Activate BD ring */
1293                 /* Warning!
1294                  * this will generate an interrupt (to the TSTORM)
1295                  * must only be done after chip is initialized
1296                  */
1297                 bnx2x_update_rx_prod(bp, fp, fp->rx_bd_prod, fp->rx_comp_prod,
1298                                      fp->rx_sge_prod);
1299         }
1300 }
1301
1302 void bnx2x_init_rx_rings(struct bnx2x *bp)
1303 {
1304         int func = BP_FUNC(bp);
1305         u16 ring_prod;
1306         int i, j;
1307
1308         /* Allocate TPA resources */
1309         for_each_eth_queue(bp, j) {
1310                 struct bnx2x_fastpath *fp = &bp->fp[j];
1311
1312                 DP(NETIF_MSG_IFUP,
1313                    "mtu %d  rx_buf_size %d\n", bp->dev->mtu, fp->rx_buf_size);
1314
1315                 if (!fp->disable_tpa) {
1316                         /* Fill the per-aggregation pool */
1317                         for (i = 0; i < MAX_AGG_QS(bp); i++) {
1318                                 struct bnx2x_agg_info *tpa_info =
1319                                         &fp->tpa_info[i];
1320                                 struct sw_rx_bd *first_buf =
1321                                         &tpa_info->first_buf;
1322
1323                                 first_buf->data =
1324                                         bnx2x_frag_alloc(fp, GFP_KERNEL);
1325                                 if (!first_buf->data) {
1326                                         BNX2X_ERR("Failed to allocate TPA skb pool for queue[%d] - disabling TPA on this queue!\n",
1327                                                   j);
1328                                         bnx2x_free_tpa_pool(bp, fp, i);
1329                                         fp->disable_tpa = 1;
1330                                         break;
1331                                 }
1332                                 dma_unmap_addr_set(first_buf, mapping, 0);
1333                                 tpa_info->tpa_state = BNX2X_TPA_STOP;
1334                         }
1335
1336                         /* "next page" elements initialization */
1337                         bnx2x_set_next_page_sgl(fp);
1338
1339                         /* set SGEs bit mask */
1340                         bnx2x_init_sge_ring_bit_mask(fp);
1341
1342                         /* Allocate SGEs and initialize the ring elements */
1343                         for (i = 0, ring_prod = 0;
1344                              i < MAX_RX_SGE_CNT*NUM_RX_SGE_PAGES; i++) {
1345
1346                                 if (bnx2x_alloc_rx_sge(bp, fp, ring_prod,
1347                                                        GFP_KERNEL) < 0) {
1348                                         BNX2X_ERR("was only able to allocate %d rx sges\n",
1349                                                   i);
1350                                         BNX2X_ERR("disabling TPA for queue[%d]\n",
1351                                                   j);
1352                                         /* Cleanup already allocated elements */
1353                                         bnx2x_free_rx_sge_range(bp, fp,
1354                                                                 ring_prod);
1355                                         bnx2x_free_tpa_pool(bp, fp,
1356                                                             MAX_AGG_QS(bp));
1357                                         fp->disable_tpa = 1;
1358                                         ring_prod = 0;
1359                                         break;
1360                                 }
1361                                 ring_prod = NEXT_SGE_IDX(ring_prod);
1362                         }
1363
1364                         fp->rx_sge_prod = ring_prod;
1365                 }
1366         }
1367
1368         for_each_eth_queue(bp, j) {
1369                 struct bnx2x_fastpath *fp = &bp->fp[j];
1370
1371                 fp->rx_bd_cons = 0;
1372
1373                 /* Activate BD ring */
1374                 /* Warning!
1375                  * this will generate an interrupt (to the TSTORM)
1376                  * must only be done after chip is initialized
1377                  */
1378                 bnx2x_update_rx_prod(bp, fp, fp->rx_bd_prod, fp->rx_comp_prod,
1379                                      fp->rx_sge_prod);
1380
1381                 if (j != 0)
1382                         continue;
1383
1384                 if (CHIP_IS_E1(bp)) {
1385                         REG_WR(bp, BAR_USTRORM_INTMEM +
1386                                USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func),
1387                                U64_LO(fp->rx_comp_mapping));
1388                         REG_WR(bp, BAR_USTRORM_INTMEM +
1389                                USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func) + 4,
1390                                U64_HI(fp->rx_comp_mapping));
1391                 }
1392         }
1393 }
1394
1395 static void bnx2x_free_tx_skbs_queue(struct bnx2x_fastpath *fp)
1396 {
1397         u8 cos;
1398         struct bnx2x *bp = fp->bp;
1399
1400         for_each_cos_in_tx_queue(fp, cos) {
1401                 struct bnx2x_fp_txdata *txdata = fp->txdata_ptr[cos];
1402                 unsigned pkts_compl = 0, bytes_compl = 0;
1403
1404                 u16 sw_prod = txdata->tx_pkt_prod;
1405                 u16 sw_cons = txdata->tx_pkt_cons;
1406
1407                 while (sw_cons != sw_prod) {
1408                         bnx2x_free_tx_pkt(bp, txdata, TX_BD(sw_cons),
1409                                           &pkts_compl, &bytes_compl);
1410                         sw_cons++;
1411                 }
1412
1413                 netdev_tx_reset_queue(
1414                         netdev_get_tx_queue(bp->dev,
1415                                             txdata->txq_index));
1416         }
1417 }
1418
1419 static void bnx2x_free_tx_skbs_cnic(struct bnx2x *bp)
1420 {
1421         int i;
1422
1423         for_each_tx_queue_cnic(bp, i) {
1424                 bnx2x_free_tx_skbs_queue(&bp->fp[i]);
1425         }
1426 }
1427
1428 static void bnx2x_free_tx_skbs(struct bnx2x *bp)
1429 {
1430         int i;
1431
1432         for_each_eth_queue(bp, i) {
1433                 bnx2x_free_tx_skbs_queue(&bp->fp[i]);
1434         }
1435 }
1436
1437 static void bnx2x_free_rx_bds(struct bnx2x_fastpath *fp)
1438 {
1439         struct bnx2x *bp = fp->bp;
1440         int i;
1441
1442         /* ring wasn't allocated */
1443         if (fp->rx_buf_ring == NULL)
1444                 return;
1445
1446         for (i = 0; i < NUM_RX_BD; i++) {
1447                 struct sw_rx_bd *rx_buf = &fp->rx_buf_ring[i];
1448                 u8 *data = rx_buf->data;
1449
1450                 if (data == NULL)
1451                         continue;
1452                 dma_unmap_single(&bp->pdev->dev,
1453                                  dma_unmap_addr(rx_buf, mapping),
1454                                  fp->rx_buf_size, DMA_FROM_DEVICE);
1455
1456                 rx_buf->data = NULL;
1457                 bnx2x_frag_free(fp, data);
1458         }
1459 }
1460
1461 static void bnx2x_free_rx_skbs_cnic(struct bnx2x *bp)
1462 {
1463         int j;
1464
1465         for_each_rx_queue_cnic(bp, j) {
1466                 bnx2x_free_rx_bds(&bp->fp[j]);
1467         }
1468 }
1469
1470 static void bnx2x_free_rx_skbs(struct bnx2x *bp)
1471 {
1472         int j;
1473
1474         for_each_eth_queue(bp, j) {
1475                 struct bnx2x_fastpath *fp = &bp->fp[j];
1476
1477                 bnx2x_free_rx_bds(fp);
1478
1479                 if (!fp->disable_tpa)
1480                         bnx2x_free_tpa_pool(bp, fp, MAX_AGG_QS(bp));
1481         }
1482 }
1483
1484 void bnx2x_free_skbs_cnic(struct bnx2x *bp)
1485 {
1486         bnx2x_free_tx_skbs_cnic(bp);
1487         bnx2x_free_rx_skbs_cnic(bp);
1488 }
1489
1490 void bnx2x_free_skbs(struct bnx2x *bp)
1491 {
1492         bnx2x_free_tx_skbs(bp);
1493         bnx2x_free_rx_skbs(bp);
1494 }
1495
1496 void bnx2x_update_max_mf_config(struct bnx2x *bp, u32 value)
1497 {
1498         /* load old values */
1499         u32 mf_cfg = bp->mf_config[BP_VN(bp)];
1500
1501         if (value != bnx2x_extract_max_cfg(bp, mf_cfg)) {
1502                 /* leave all but MAX value */
1503                 mf_cfg &= ~FUNC_MF_CFG_MAX_BW_MASK;
1504
1505                 /* set new MAX value */
1506                 mf_cfg |= (value << FUNC_MF_CFG_MAX_BW_SHIFT)
1507                                 & FUNC_MF_CFG_MAX_BW_MASK;
1508
1509                 bnx2x_fw_command(bp, DRV_MSG_CODE_SET_MF_BW, mf_cfg);
1510         }
1511 }
1512
1513 /**
1514  * bnx2x_free_msix_irqs - free previously requested MSI-X IRQ vectors
1515  *
1516  * @bp:         driver handle
1517  * @nvecs:      number of vectors to be released
1518  */
1519 static void bnx2x_free_msix_irqs(struct bnx2x *bp, int nvecs)
1520 {
1521         int i, offset = 0;
1522
1523         if (nvecs == offset)
1524                 return;
1525
1526         /* VFs don't have a default SB */
1527         if (IS_PF(bp)) {
1528                 free_irq(bp->msix_table[offset].vector, bp->dev);
1529                 DP(NETIF_MSG_IFDOWN, "released sp irq (%d)\n",
1530                    bp->msix_table[offset].vector);
1531                 offset++;
1532         }
1533
1534         if (CNIC_SUPPORT(bp)) {
1535                 if (nvecs == offset)
1536                         return;
1537                 offset++;
1538         }
1539
1540         for_each_eth_queue(bp, i) {
1541                 if (nvecs == offset)
1542                         return;
1543                 DP(NETIF_MSG_IFDOWN, "about to release fp #%d->%d irq\n",
1544                    i, bp->msix_table[offset].vector);
1545
1546                 free_irq(bp->msix_table[offset++].vector, &bp->fp[i]);
1547         }
1548 }
1549
1550 void bnx2x_free_irq(struct bnx2x *bp)
1551 {
1552         if (bp->flags & USING_MSIX_FLAG &&
1553             !(bp->flags & USING_SINGLE_MSIX_FLAG)) {
1554                 int nvecs = BNX2X_NUM_ETH_QUEUES(bp) + CNIC_SUPPORT(bp);
1555
1556                 /* vfs don't have a default status block */
1557                 if (IS_PF(bp))
1558                         nvecs++;
1559
1560                 bnx2x_free_msix_irqs(bp, nvecs);
1561         } else {
1562                 free_irq(bp->dev->irq, bp->dev);
1563         }
1564 }
1565
1566 int bnx2x_enable_msix(struct bnx2x *bp)
1567 {
1568         int msix_vec = 0, i, rc;
1569
1570         /* VFs don't have a default status block */
1571         if (IS_PF(bp)) {
1572                 bp->msix_table[msix_vec].entry = msix_vec;
1573                 BNX2X_DEV_INFO("msix_table[0].entry = %d (slowpath)\n",
1574                                bp->msix_table[0].entry);
1575                 msix_vec++;
1576         }
1577
1578         /* Cnic requires an msix vector for itself */
1579         if (CNIC_SUPPORT(bp)) {
1580                 bp->msix_table[msix_vec].entry = msix_vec;
1581                 BNX2X_DEV_INFO("msix_table[%d].entry = %d (CNIC)\n",
1582                                msix_vec, bp->msix_table[msix_vec].entry);
1583                 msix_vec++;
1584         }
1585
1586         /* We need separate vectors for ETH queues only (not FCoE) */
1587         for_each_eth_queue(bp, i) {
1588                 bp->msix_table[msix_vec].entry = msix_vec;
1589                 BNX2X_DEV_INFO("msix_table[%d].entry = %d (fastpath #%u)\n",
1590                                msix_vec, msix_vec, i);
1591                 msix_vec++;
1592         }
1593
1594         DP(BNX2X_MSG_SP, "about to request enable msix with %d vectors\n",
1595            msix_vec);
1596
1597         rc = pci_enable_msix(bp->pdev, &bp->msix_table[0], msix_vec);
1598
1599         /*
1600          * reconfigure number of tx/rx queues according to available
1601          * MSI-X vectors
1602          */
1603         if (rc >= BNX2X_MIN_MSIX_VEC_CNT(bp)) {
1604                 /* how less vectors we will have? */
1605                 int diff = msix_vec - rc;
1606
1607                 BNX2X_DEV_INFO("Trying to use less MSI-X vectors: %d\n", rc);
1608
1609                 rc = pci_enable_msix(bp->pdev, &bp->msix_table[0], rc);
1610
1611                 if (rc) {
1612                         BNX2X_DEV_INFO("MSI-X is not attainable rc %d\n", rc);
1613                         goto no_msix;
1614                 }
1615                 /*
1616                  * decrease number of queues by number of unallocated entries
1617                  */
1618                 bp->num_ethernet_queues -= diff;
1619                 bp->num_queues = bp->num_ethernet_queues + bp->num_cnic_queues;
1620
1621                 BNX2X_DEV_INFO("New queue configuration set: %d\n",
1622                                bp->num_queues);
1623         } else if (rc > 0) {
1624                 /* Get by with single vector */
1625                 rc = pci_enable_msix(bp->pdev, &bp->msix_table[0], 1);
1626                 if (rc) {
1627                         BNX2X_DEV_INFO("Single MSI-X is not attainable rc %d\n",
1628                                        rc);
1629                         goto no_msix;
1630                 }
1631
1632                 BNX2X_DEV_INFO("Using single MSI-X vector\n");
1633                 bp->flags |= USING_SINGLE_MSIX_FLAG;
1634
1635                 BNX2X_DEV_INFO("set number of queues to 1\n");
1636                 bp->num_ethernet_queues = 1;
1637                 bp->num_queues = bp->num_ethernet_queues + bp->num_cnic_queues;
1638         } else if (rc < 0) {
1639                 BNX2X_DEV_INFO("MSI-X is not attainable  rc %d\n", rc);
1640                 goto no_msix;
1641         }
1642
1643         bp->flags |= USING_MSIX_FLAG;
1644
1645         return 0;
1646
1647 no_msix:
1648         /* fall to INTx if not enough memory */
1649         if (rc == -ENOMEM)
1650                 bp->flags |= DISABLE_MSI_FLAG;
1651
1652         return rc;
1653 }
1654
1655 static int bnx2x_req_msix_irqs(struct bnx2x *bp)
1656 {
1657         int i, rc, offset = 0;
1658
1659         /* no default status block for vf */
1660         if (IS_PF(bp)) {
1661                 rc = request_irq(bp->msix_table[offset++].vector,
1662                                  bnx2x_msix_sp_int, 0,
1663                                  bp->dev->name, bp->dev);
1664                 if (rc) {
1665                         BNX2X_ERR("request sp irq failed\n");
1666                         return -EBUSY;
1667                 }
1668         }
1669
1670         if (CNIC_SUPPORT(bp))
1671                 offset++;
1672
1673         for_each_eth_queue(bp, i) {
1674                 struct bnx2x_fastpath *fp = &bp->fp[i];
1675                 snprintf(fp->name, sizeof(fp->name), "%s-fp-%d",
1676                          bp->dev->name, i);
1677
1678                 rc = request_irq(bp->msix_table[offset].vector,
1679                                  bnx2x_msix_fp_int, 0, fp->name, fp);
1680                 if (rc) {
1681                         BNX2X_ERR("request fp #%d irq (%d) failed  rc %d\n", i,
1682                               bp->msix_table[offset].vector, rc);
1683                         bnx2x_free_msix_irqs(bp, offset);
1684                         return -EBUSY;
1685                 }
1686
1687                 offset++;
1688         }
1689
1690         i = BNX2X_NUM_ETH_QUEUES(bp);
1691         if (IS_PF(bp)) {
1692                 offset = 1 + CNIC_SUPPORT(bp);
1693                 netdev_info(bp->dev,
1694                             "using MSI-X  IRQs: sp %d  fp[%d] %d ... fp[%d] %d\n",
1695                             bp->msix_table[0].vector,
1696                             0, bp->msix_table[offset].vector,
1697                             i - 1, bp->msix_table[offset + i - 1].vector);
1698         } else {
1699                 offset = CNIC_SUPPORT(bp);
1700                 netdev_info(bp->dev,
1701                             "using MSI-X  IRQs: fp[%d] %d ... fp[%d] %d\n",
1702                             0, bp->msix_table[offset].vector,
1703                             i - 1, bp->msix_table[offset + i - 1].vector);
1704         }
1705         return 0;
1706 }
1707
1708 int bnx2x_enable_msi(struct bnx2x *bp)
1709 {
1710         int rc;
1711
1712         rc = pci_enable_msi(bp->pdev);
1713         if (rc) {
1714                 BNX2X_DEV_INFO("MSI is not attainable\n");
1715                 return -1;
1716         }
1717         bp->flags |= USING_MSI_FLAG;
1718
1719         return 0;
1720 }
1721
1722 static int bnx2x_req_irq(struct bnx2x *bp)
1723 {
1724         unsigned long flags;
1725         unsigned int irq;
1726
1727         if (bp->flags & (USING_MSI_FLAG | USING_MSIX_FLAG))
1728                 flags = 0;
1729         else
1730                 flags = IRQF_SHARED;
1731
1732         if (bp->flags & USING_MSIX_FLAG)
1733                 irq = bp->msix_table[0].vector;
1734         else
1735                 irq = bp->pdev->irq;
1736
1737         return request_irq(irq, bnx2x_interrupt, flags, bp->dev->name, bp->dev);
1738 }
1739
1740 static int bnx2x_setup_irqs(struct bnx2x *bp)
1741 {
1742         int rc = 0;
1743         if (bp->flags & USING_MSIX_FLAG &&
1744             !(bp->flags & USING_SINGLE_MSIX_FLAG)) {
1745                 rc = bnx2x_req_msix_irqs(bp);
1746                 if (rc)
1747                         return rc;
1748         } else {
1749                 rc = bnx2x_req_irq(bp);
1750                 if (rc) {
1751                         BNX2X_ERR("IRQ request failed  rc %d, aborting\n", rc);
1752                         return rc;
1753                 }
1754                 if (bp->flags & USING_MSI_FLAG) {
1755                         bp->dev->irq = bp->pdev->irq;
1756                         netdev_info(bp->dev, "using MSI IRQ %d\n",
1757                                     bp->dev->irq);
1758                 }
1759                 if (bp->flags & USING_MSIX_FLAG) {
1760                         bp->dev->irq = bp->msix_table[0].vector;
1761                         netdev_info(bp->dev, "using MSIX IRQ %d\n",
1762                                     bp->dev->irq);
1763                 }
1764         }
1765
1766         return 0;
1767 }
1768
1769 static void bnx2x_napi_enable_cnic(struct bnx2x *bp)
1770 {
1771         int i;
1772
1773         for_each_rx_queue_cnic(bp, i) {
1774                 bnx2x_fp_init_lock(&bp->fp[i]);
1775                 napi_enable(&bnx2x_fp(bp, i, napi));
1776         }
1777 }
1778
1779 static void bnx2x_napi_enable(struct bnx2x *bp)
1780 {
1781         int i;
1782
1783         for_each_eth_queue(bp, i) {
1784                 bnx2x_fp_init_lock(&bp->fp[i]);
1785                 napi_enable(&bnx2x_fp(bp, i, napi));
1786         }
1787 }
1788
1789 static void bnx2x_napi_disable_cnic(struct bnx2x *bp)
1790 {
1791         int i;
1792
1793         for_each_rx_queue_cnic(bp, i) {
1794                 napi_disable(&bnx2x_fp(bp, i, napi));
1795                 while (!bnx2x_fp_ll_disable(&bp->fp[i]))
1796                         usleep_range(1000, 2000);
1797         }
1798 }
1799
1800 static void bnx2x_napi_disable(struct bnx2x *bp)
1801 {
1802         int i;
1803
1804         for_each_eth_queue(bp, i) {
1805                 napi_disable(&bnx2x_fp(bp, i, napi));
1806                 while (!bnx2x_fp_ll_disable(&bp->fp[i]))
1807                         usleep_range(1000, 2000);
1808         }
1809 }
1810
1811 void bnx2x_netif_start(struct bnx2x *bp)
1812 {
1813         if (netif_running(bp->dev)) {
1814                 bnx2x_napi_enable(bp);
1815                 if (CNIC_LOADED(bp))
1816                         bnx2x_napi_enable_cnic(bp);
1817                 bnx2x_int_enable(bp);
1818                 if (bp->state == BNX2X_STATE_OPEN)
1819                         netif_tx_wake_all_queues(bp->dev);
1820         }
1821 }
1822
1823 void bnx2x_netif_stop(struct bnx2x *bp, int disable_hw)
1824 {
1825         bnx2x_int_disable_sync(bp, disable_hw);
1826         bnx2x_napi_disable(bp);
1827         if (CNIC_LOADED(bp))
1828                 bnx2x_napi_disable_cnic(bp);
1829 }
1830
1831 u16 bnx2x_select_queue(struct net_device *dev, struct sk_buff *skb)
1832 {
1833         struct bnx2x *bp = netdev_priv(dev);
1834
1835         if (CNIC_LOADED(bp) && !NO_FCOE(bp)) {
1836                 struct ethhdr *hdr = (struct ethhdr *)skb->data;
1837                 u16 ether_type = ntohs(hdr->h_proto);
1838
1839                 /* Skip VLAN tag if present */
1840                 if (ether_type == ETH_P_8021Q) {
1841                         struct vlan_ethhdr *vhdr =
1842                                 (struct vlan_ethhdr *)skb->data;
1843
1844                         ether_type = ntohs(vhdr->h_vlan_encapsulated_proto);
1845                 }
1846
1847                 /* If ethertype is FCoE or FIP - use FCoE ring */
1848                 if ((ether_type == ETH_P_FCOE) || (ether_type == ETH_P_FIP))
1849                         return bnx2x_fcoe_tx(bp, txq_index);
1850         }
1851
1852         /* select a non-FCoE queue */
1853         return __netdev_pick_tx(dev, skb) % BNX2X_NUM_ETH_QUEUES(bp);
1854 }
1855
1856 void bnx2x_set_num_queues(struct bnx2x *bp)
1857 {
1858         /* RSS queues */
1859         bp->num_ethernet_queues = bnx2x_calc_num_queues(bp);
1860
1861         /* override in STORAGE SD modes */
1862         if (IS_MF_STORAGE_SD(bp) || IS_MF_FCOE_AFEX(bp))
1863                 bp->num_ethernet_queues = 1;
1864
1865         /* Add special queues */
1866         bp->num_cnic_queues = CNIC_SUPPORT(bp); /* For FCOE */
1867         bp->num_queues = bp->num_ethernet_queues + bp->num_cnic_queues;
1868
1869         BNX2X_DEV_INFO("set number of queues to %d\n", bp->num_queues);
1870 }
1871
1872 /**
1873  * bnx2x_set_real_num_queues - configure netdev->real_num_[tx,rx]_queues
1874  *
1875  * @bp:         Driver handle
1876  *
1877  * We currently support for at most 16 Tx queues for each CoS thus we will
1878  * allocate a multiple of 16 for ETH L2 rings according to the value of the
1879  * bp->max_cos.
1880  *
1881  * If there is an FCoE L2 queue the appropriate Tx queue will have the next
1882  * index after all ETH L2 indices.
1883  *
1884  * If the actual number of Tx queues (for each CoS) is less than 16 then there
1885  * will be the holes at the end of each group of 16 ETh L2 indices (0..15,
1886  * 16..31,...) with indices that are not coupled with any real Tx queue.
1887  *
1888  * The proper configuration of skb->queue_mapping is handled by
1889  * bnx2x_select_queue() and __skb_tx_hash().
1890  *
1891  * bnx2x_setup_tc() takes care of the proper TC mappings so that __skb_tx_hash()
1892  * will return a proper Tx index if TC is enabled (netdev->num_tc > 0).
1893  */
1894 static int bnx2x_set_real_num_queues(struct bnx2x *bp, int include_cnic)
1895 {
1896         int rc, tx, rx;
1897
1898         tx = BNX2X_NUM_ETH_QUEUES(bp) * bp->max_cos;
1899         rx = BNX2X_NUM_ETH_QUEUES(bp);
1900
1901 /* account for fcoe queue */
1902         if (include_cnic && !NO_FCOE(bp)) {
1903                 rx++;
1904                 tx++;
1905         }
1906
1907         rc = netif_set_real_num_tx_queues(bp->dev, tx);
1908         if (rc) {
1909                 BNX2X_ERR("Failed to set real number of Tx queues: %d\n", rc);
1910                 return rc;
1911         }
1912         rc = netif_set_real_num_rx_queues(bp->dev, rx);
1913         if (rc) {
1914                 BNX2X_ERR("Failed to set real number of Rx queues: %d\n", rc);
1915                 return rc;
1916         }
1917
1918         DP(NETIF_MSG_IFUP, "Setting real num queues to (tx, rx) (%d, %d)\n",
1919                           tx, rx);
1920
1921         return rc;
1922 }
1923
1924 static void bnx2x_set_rx_buf_size(struct bnx2x *bp)
1925 {
1926         int i;
1927
1928         for_each_queue(bp, i) {
1929                 struct bnx2x_fastpath *fp = &bp->fp[i];
1930                 u32 mtu;
1931
1932                 /* Always use a mini-jumbo MTU for the FCoE L2 ring */
1933                 if (IS_FCOE_IDX(i))
1934                         /*
1935                          * Although there are no IP frames expected to arrive to
1936                          * this ring we still want to add an
1937                          * IP_HEADER_ALIGNMENT_PADDING to prevent a buffer
1938                          * overrun attack.
1939                          */
1940                         mtu = BNX2X_FCOE_MINI_JUMBO_MTU;
1941                 else
1942                         mtu = bp->dev->mtu;
1943                 fp->rx_buf_size = BNX2X_FW_RX_ALIGN_START +
1944                                   IP_HEADER_ALIGNMENT_PADDING +
1945                                   ETH_OVREHEAD +
1946                                   mtu +
1947                                   BNX2X_FW_RX_ALIGN_END;
1948                 /* Note : rx_buf_size doesn't take into account NET_SKB_PAD */
1949                 if (fp->rx_buf_size + NET_SKB_PAD <= PAGE_SIZE)
1950                         fp->rx_frag_size = fp->rx_buf_size + NET_SKB_PAD;
1951                 else
1952                         fp->rx_frag_size = 0;
1953         }
1954 }
1955
1956 static int bnx2x_init_rss(struct bnx2x *bp)
1957 {
1958         int i;
1959         u8 num_eth_queues = BNX2X_NUM_ETH_QUEUES(bp);
1960
1961         /* Prepare the initial contents for the indirection table if RSS is
1962          * enabled
1963          */
1964         for (i = 0; i < sizeof(bp->rss_conf_obj.ind_table); i++)
1965                 bp->rss_conf_obj.ind_table[i] =
1966                         bp->fp->cl_id +
1967                         ethtool_rxfh_indir_default(i, num_eth_queues);
1968
1969         /*
1970          * For 57710 and 57711 SEARCHER configuration (rss_keys) is
1971          * per-port, so if explicit configuration is needed , do it only
1972          * for a PMF.
1973          *
1974          * For 57712 and newer on the other hand it's a per-function
1975          * configuration.
1976          */
1977         return bnx2x_config_rss_eth(bp, bp->port.pmf || !CHIP_IS_E1x(bp));
1978 }
1979
1980 int bnx2x_rss(struct bnx2x *bp, struct bnx2x_rss_config_obj *rss_obj,
1981               bool config_hash, bool enable)
1982 {
1983         struct bnx2x_config_rss_params params = {NULL};
1984
1985         /* Although RSS is meaningless when there is a single HW queue we
1986          * still need it enabled in order to have HW Rx hash generated.
1987          *
1988          * if (!is_eth_multi(bp))
1989          *      bp->multi_mode = ETH_RSS_MODE_DISABLED;
1990          */
1991
1992         params.rss_obj = rss_obj;
1993
1994         __set_bit(RAMROD_COMP_WAIT, &params.ramrod_flags);
1995
1996         if (enable) {
1997                 __set_bit(BNX2X_RSS_MODE_REGULAR, &params.rss_flags);
1998
1999                 /* RSS configuration */
2000                 __set_bit(BNX2X_RSS_IPV4, &params.rss_flags);
2001                 __set_bit(BNX2X_RSS_IPV4_TCP, &params.rss_flags);
2002                 __set_bit(BNX2X_RSS_IPV6, &params.rss_flags);
2003                 __set_bit(BNX2X_RSS_IPV6_TCP, &params.rss_flags);
2004                 if (rss_obj->udp_rss_v4)
2005                         __set_bit(BNX2X_RSS_IPV4_UDP, &params.rss_flags);
2006                 if (rss_obj->udp_rss_v6)
2007                         __set_bit(BNX2X_RSS_IPV6_UDP, &params.rss_flags);
2008         } else {
2009                 __set_bit(BNX2X_RSS_MODE_DISABLED, &params.rss_flags);
2010         }
2011
2012         /* Hash bits */
2013         params.rss_result_mask = MULTI_MASK;
2014
2015         memcpy(params.ind_table, rss_obj->ind_table, sizeof(params.ind_table));
2016
2017         if (config_hash) {
2018                 /* RSS keys */
2019                 prandom_bytes(params.rss_key, T_ETH_RSS_KEY * 4);
2020                 __set_bit(BNX2X_RSS_SET_SRCH, &params.rss_flags);
2021         }
2022
2023         if (IS_PF(bp))
2024                 return bnx2x_config_rss(bp, &params);
2025         else
2026                 return bnx2x_vfpf_config_rss(bp, &params);
2027 }
2028
2029 static int bnx2x_init_hw(struct bnx2x *bp, u32 load_code)
2030 {
2031         struct bnx2x_func_state_params func_params = {NULL};
2032
2033         /* Prepare parameters for function state transitions */
2034         __set_bit(RAMROD_COMP_WAIT, &func_params.ramrod_flags);
2035
2036         func_params.f_obj = &bp->func_obj;
2037         func_params.cmd = BNX2X_F_CMD_HW_INIT;
2038
2039         func_params.params.hw_init.load_phase = load_code;
2040
2041         return bnx2x_func_state_change(bp, &func_params);
2042 }
2043
2044 /*
2045  * Cleans the object that have internal lists without sending
2046  * ramrods. Should be run when interrupts are disabled.
2047  */
2048 void bnx2x_squeeze_objects(struct bnx2x *bp)
2049 {
2050         int rc;
2051         unsigned long ramrod_flags = 0, vlan_mac_flags = 0;
2052         struct bnx2x_mcast_ramrod_params rparam = {NULL};
2053         struct bnx2x_vlan_mac_obj *mac_obj = &bp->sp_objs->mac_obj;
2054
2055         /***************** Cleanup MACs' object first *************************/
2056
2057         /* Wait for completion of requested */
2058         __set_bit(RAMROD_COMP_WAIT, &ramrod_flags);
2059         /* Perform a dry cleanup */
2060         __set_bit(RAMROD_DRV_CLR_ONLY, &ramrod_flags);
2061
2062         /* Clean ETH primary MAC */
2063         __set_bit(BNX2X_ETH_MAC, &vlan_mac_flags);
2064         rc = mac_obj->delete_all(bp, &bp->sp_objs->mac_obj, &vlan_mac_flags,
2065                                  &ramrod_flags);
2066         if (rc != 0)
2067                 BNX2X_ERR("Failed to clean ETH MACs: %d\n", rc);
2068
2069         /* Cleanup UC list */
2070         vlan_mac_flags = 0;
2071         __set_bit(BNX2X_UC_LIST_MAC, &vlan_mac_flags);
2072         rc = mac_obj->delete_all(bp, mac_obj, &vlan_mac_flags,
2073                                  &ramrod_flags);
2074         if (rc != 0)
2075                 BNX2X_ERR("Failed to clean UC list MACs: %d\n", rc);
2076
2077         /***************** Now clean mcast object *****************************/
2078         rparam.mcast_obj = &bp->mcast_obj;
2079         __set_bit(RAMROD_DRV_CLR_ONLY, &rparam.ramrod_flags);
2080
2081         /* Add a DEL command... - Since we're doing a driver cleanup only,
2082          * we take a lock surrounding both the initial send and the CONTs,
2083          * as we don't want a true completion to disrupt us in the middle.
2084          */
2085         netif_addr_lock_bh(bp->dev);
2086         rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_DEL);
2087         if (rc < 0)
2088                 BNX2X_ERR("Failed to add a new DEL command to a multi-cast object: %d\n",
2089                           rc);
2090
2091         /* ...and wait until all pending commands are cleared */
2092         rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_CONT);
2093         while (rc != 0) {
2094                 if (rc < 0) {
2095                         BNX2X_ERR("Failed to clean multi-cast object: %d\n",
2096                                   rc);
2097                         netif_addr_unlock_bh(bp->dev);
2098                         return;
2099                 }
2100
2101                 rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_CONT);
2102         }
2103         netif_addr_unlock_bh(bp->dev);
2104 }
2105
2106 #ifndef BNX2X_STOP_ON_ERROR
2107 #define LOAD_ERROR_EXIT(bp, label) \
2108         do { \
2109                 (bp)->state = BNX2X_STATE_ERROR; \
2110                 goto label; \
2111         } while (0)
2112
2113 #define LOAD_ERROR_EXIT_CNIC(bp, label) \
2114         do { \
2115                 bp->cnic_loaded = false; \
2116                 goto label; \
2117         } while (0)
2118 #else /*BNX2X_STOP_ON_ERROR*/
2119 #define LOAD_ERROR_EXIT(bp, label) \
2120         do { \
2121                 (bp)->state = BNX2X_STATE_ERROR; \
2122                 (bp)->panic = 1; \
2123                 return -EBUSY; \
2124         } while (0)
2125 #define LOAD_ERROR_EXIT_CNIC(bp, label) \
2126         do { \
2127                 bp->cnic_loaded = false; \
2128                 (bp)->panic = 1; \
2129                 return -EBUSY; \
2130         } while (0)
2131 #endif /*BNX2X_STOP_ON_ERROR*/
2132
2133 static void bnx2x_free_fw_stats_mem(struct bnx2x *bp)
2134 {
2135         BNX2X_PCI_FREE(bp->fw_stats, bp->fw_stats_mapping,
2136                        bp->fw_stats_data_sz + bp->fw_stats_req_sz);
2137         return;
2138 }
2139
2140 static int bnx2x_alloc_fw_stats_mem(struct bnx2x *bp)
2141 {
2142         int num_groups, vf_headroom = 0;
2143         int is_fcoe_stats = NO_FCOE(bp) ? 0 : 1;
2144
2145         /* number of queues for statistics is number of eth queues + FCoE */
2146         u8 num_queue_stats = BNX2X_NUM_ETH_QUEUES(bp) + is_fcoe_stats;
2147
2148         /* Total number of FW statistics requests =
2149          * 1 for port stats + 1 for PF stats + potential 2 for FCoE (fcoe proper
2150          * and fcoe l2 queue) stats + num of queues (which includes another 1
2151          * for fcoe l2 queue if applicable)
2152          */
2153         bp->fw_stats_num = 2 + is_fcoe_stats + num_queue_stats;
2154
2155         /* vf stats appear in the request list, but their data is allocated by
2156          * the VFs themselves. We don't include them in the bp->fw_stats_num as
2157          * it is used to determine where to place the vf stats queries in the
2158          * request struct
2159          */
2160         if (IS_SRIOV(bp))
2161                 vf_headroom = bnx2x_vf_headroom(bp);
2162
2163         /* Request is built from stats_query_header and an array of
2164          * stats_query_cmd_group each of which contains
2165          * STATS_QUERY_CMD_COUNT rules. The real number or requests is
2166          * configured in the stats_query_header.
2167          */
2168         num_groups =
2169                 (((bp->fw_stats_num + vf_headroom) / STATS_QUERY_CMD_COUNT) +
2170                  (((bp->fw_stats_num + vf_headroom) % STATS_QUERY_CMD_COUNT) ?
2171                  1 : 0));
2172
2173         DP(BNX2X_MSG_SP, "stats fw_stats_num %d, vf headroom %d, num_groups %d\n",
2174            bp->fw_stats_num, vf_headroom, num_groups);
2175         bp->fw_stats_req_sz = sizeof(struct stats_query_header) +
2176                 num_groups * sizeof(struct stats_query_cmd_group);
2177
2178         /* Data for statistics requests + stats_counter
2179          * stats_counter holds per-STORM counters that are incremented
2180          * when STORM has finished with the current request.
2181          * memory for FCoE offloaded statistics are counted anyway,
2182          * even if they will not be sent.
2183          * VF stats are not accounted for here as the data of VF stats is stored
2184          * in memory allocated by the VF, not here.
2185          */
2186         bp->fw_stats_data_sz = sizeof(struct per_port_stats) +
2187                 sizeof(struct per_pf_stats) +
2188                 sizeof(struct fcoe_statistics_params) +
2189                 sizeof(struct per_queue_stats) * num_queue_stats +
2190                 sizeof(struct stats_counter);
2191
2192         BNX2X_PCI_ALLOC(bp->fw_stats, &bp->fw_stats_mapping,
2193                         bp->fw_stats_data_sz + bp->fw_stats_req_sz);
2194
2195         /* Set shortcuts */
2196         bp->fw_stats_req = (struct bnx2x_fw_stats_req *)bp->fw_stats;
2197         bp->fw_stats_req_mapping = bp->fw_stats_mapping;
2198         bp->fw_stats_data = (struct bnx2x_fw_stats_data *)
2199                 ((u8 *)bp->fw_stats + bp->fw_stats_req_sz);
2200         bp->fw_stats_data_mapping = bp->fw_stats_mapping +
2201                 bp->fw_stats_req_sz;
2202
2203         DP(BNX2X_MSG_SP, "statistics request base address set to %x %x\n",
2204            U64_HI(bp->fw_stats_req_mapping),
2205            U64_LO(bp->fw_stats_req_mapping));
2206         DP(BNX2X_MSG_SP, "statistics data base address set to %x %x\n",
2207            U64_HI(bp->fw_stats_data_mapping),
2208            U64_LO(bp->fw_stats_data_mapping));
2209         return 0;
2210
2211 alloc_mem_err:
2212         bnx2x_free_fw_stats_mem(bp);
2213         BNX2X_ERR("Can't allocate FW stats memory\n");
2214         return -ENOMEM;
2215 }
2216
2217 /* send load request to mcp and analyze response */
2218 static int bnx2x_nic_load_request(struct bnx2x *bp, u32 *load_code)
2219 {
2220         u32 param;
2221
2222         /* init fw_seq */
2223         bp->fw_seq =
2224                 (SHMEM_RD(bp, func_mb[BP_FW_MB_IDX(bp)].drv_mb_header) &
2225                  DRV_MSG_SEQ_NUMBER_MASK);
2226         BNX2X_DEV_INFO("fw_seq 0x%08x\n", bp->fw_seq);
2227
2228         /* Get current FW pulse sequence */
2229         bp->fw_drv_pulse_wr_seq =
2230                 (SHMEM_RD(bp, func_mb[BP_FW_MB_IDX(bp)].drv_pulse_mb) &
2231                  DRV_PULSE_SEQ_MASK);
2232         BNX2X_DEV_INFO("drv_pulse 0x%x\n", bp->fw_drv_pulse_wr_seq);
2233
2234         param = DRV_MSG_CODE_LOAD_REQ_WITH_LFA;
2235
2236         if (IS_MF_SD(bp) && bnx2x_port_after_undi(bp))
2237                 param |= DRV_MSG_CODE_LOAD_REQ_FORCE_LFA;
2238
2239         /* load request */
2240         (*load_code) = bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_REQ, param);
2241
2242         /* if mcp fails to respond we must abort */
2243         if (!(*load_code)) {
2244                 BNX2X_ERR("MCP response failure, aborting\n");
2245                 return -EBUSY;
2246         }
2247
2248         /* If mcp refused (e.g. other port is in diagnostic mode) we
2249          * must abort
2250          */
2251         if ((*load_code) == FW_MSG_CODE_DRV_LOAD_REFUSED) {
2252                 BNX2X_ERR("MCP refused load request, aborting\n");
2253                 return -EBUSY;
2254         }
2255         return 0;
2256 }
2257
2258 /* check whether another PF has already loaded FW to chip. In
2259  * virtualized environments a pf from another VM may have already
2260  * initialized the device including loading FW
2261  */
2262 int bnx2x_nic_load_analyze_req(struct bnx2x *bp, u32 load_code)
2263 {
2264         /* is another pf loaded on this engine? */
2265         if (load_code != FW_MSG_CODE_DRV_LOAD_COMMON_CHIP &&
2266             load_code != FW_MSG_CODE_DRV_LOAD_COMMON) {
2267                 /* build my FW version dword */
2268                 u32 my_fw = (BCM_5710_FW_MAJOR_VERSION) +
2269                         (BCM_5710_FW_MINOR_VERSION << 8) +
2270                         (BCM_5710_FW_REVISION_VERSION << 16) +
2271                         (BCM_5710_FW_ENGINEERING_VERSION << 24);
2272
2273                 /* read loaded FW from chip */
2274                 u32 loaded_fw = REG_RD(bp, XSEM_REG_PRAM);
2275
2276                 DP(BNX2X_MSG_SP, "loaded fw %x, my fw %x\n",
2277                    loaded_fw, my_fw);
2278
2279                 /* abort nic load if version mismatch */
2280                 if (my_fw != loaded_fw) {
2281                         BNX2X_ERR("bnx2x with FW %x was already loaded which mismatches my %x FW. Aborting\n",
2282                                   loaded_fw, my_fw);
2283                         return -EBUSY;
2284                 }
2285         }
2286         return 0;
2287 }
2288
2289 /* returns the "mcp load_code" according to global load_count array */
2290 static int bnx2x_nic_load_no_mcp(struct bnx2x *bp, int port)
2291 {
2292         int path = BP_PATH(bp);
2293
2294         DP(NETIF_MSG_IFUP, "NO MCP - load counts[%d]      %d, %d, %d\n",
2295            path, load_count[path][0], load_count[path][1],
2296            load_count[path][2]);
2297         load_count[path][0]++;
2298         load_count[path][1 + port]++;
2299         DP(NETIF_MSG_IFUP, "NO MCP - new load counts[%d]  %d, %d, %d\n",
2300            path, load_count[path][0], load_count[path][1],
2301            load_count[path][2]);
2302         if (load_count[path][0] == 1)
2303                 return FW_MSG_CODE_DRV_LOAD_COMMON;
2304         else if (load_count[path][1 + port] == 1)
2305                 return FW_MSG_CODE_DRV_LOAD_PORT;
2306         else
2307                 return FW_MSG_CODE_DRV_LOAD_FUNCTION;
2308 }
2309
2310 /* mark PMF if applicable */
2311 static void bnx2x_nic_load_pmf(struct bnx2x *bp, u32 load_code)
2312 {
2313         if ((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
2314             (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP) ||
2315             (load_code == FW_MSG_CODE_DRV_LOAD_PORT)) {
2316                 bp->port.pmf = 1;
2317                 /* We need the barrier to ensure the ordering between the
2318                  * writing to bp->port.pmf here and reading it from the
2319                  * bnx2x_periodic_task().
2320                  */
2321                 smp_mb();
2322         } else {
2323                 bp->port.pmf = 0;
2324         }
2325
2326         DP(NETIF_MSG_LINK, "pmf %d\n", bp->port.pmf);
2327 }
2328
2329 static void bnx2x_nic_load_afex_dcc(struct bnx2x *bp, int load_code)
2330 {
2331         if (((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
2332              (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP)) &&
2333             (bp->common.shmem2_base)) {
2334                 if (SHMEM2_HAS(bp, dcc_support))
2335                         SHMEM2_WR(bp, dcc_support,
2336                                   (SHMEM_DCC_SUPPORT_DISABLE_ENABLE_PF_TLV |
2337                                    SHMEM_DCC_SUPPORT_BANDWIDTH_ALLOCATION_TLV));
2338                 if (SHMEM2_HAS(bp, afex_driver_support))
2339                         SHMEM2_WR(bp, afex_driver_support,
2340                                   SHMEM_AFEX_SUPPORTED_VERSION_ONE);
2341         }
2342
2343         /* Set AFEX default VLAN tag to an invalid value */
2344         bp->afex_def_vlan_tag = -1;
2345 }
2346
2347 /**
2348  * bnx2x_bz_fp - zero content of the fastpath structure.
2349  *
2350  * @bp:         driver handle
2351  * @index:      fastpath index to be zeroed
2352  *
2353  * Makes sure the contents of the bp->fp[index].napi is kept
2354  * intact.
2355  */
2356 static void bnx2x_bz_fp(struct bnx2x *bp, int index)
2357 {
2358         struct bnx2x_fastpath *fp = &bp->fp[index];
2359         int cos;
2360         struct napi_struct orig_napi = fp->napi;
2361         struct bnx2x_agg_info *orig_tpa_info = fp->tpa_info;
2362
2363         /* bzero bnx2x_fastpath contents */
2364         if (fp->tpa_info)
2365                 memset(fp->tpa_info, 0, ETH_MAX_AGGREGATION_QUEUES_E1H_E2 *
2366                        sizeof(struct bnx2x_agg_info));
2367         memset(fp, 0, sizeof(*fp));
2368
2369         /* Restore the NAPI object as it has been already initialized */
2370         fp->napi = orig_napi;
2371         fp->tpa_info = orig_tpa_info;
2372         fp->bp = bp;
2373         fp->index = index;
2374         if (IS_ETH_FP(fp))
2375                 fp->max_cos = bp->max_cos;
2376         else
2377                 /* Special queues support only one CoS */
2378                 fp->max_cos = 1;
2379
2380         /* Init txdata pointers */
2381         if (IS_FCOE_FP(fp))
2382                 fp->txdata_ptr[0] = &bp->bnx2x_txq[FCOE_TXQ_IDX(bp)];
2383         if (IS_ETH_FP(fp))
2384                 for_each_cos_in_tx_queue(fp, cos)
2385                         fp->txdata_ptr[cos] = &bp->bnx2x_txq[cos *
2386                                 BNX2X_NUM_ETH_QUEUES(bp) + index];
2387
2388         /* set the tpa flag for each queue. The tpa flag determines the queue
2389          * minimal size so it must be set prior to queue memory allocation
2390          */
2391         fp->disable_tpa = !(bp->flags & TPA_ENABLE_FLAG ||
2392                                   (bp->flags & GRO_ENABLE_FLAG &&
2393                                    bnx2x_mtu_allows_gro(bp->dev->mtu)));
2394         if (bp->flags & TPA_ENABLE_FLAG)
2395                 fp->mode = TPA_MODE_LRO;
2396         else if (bp->flags & GRO_ENABLE_FLAG)
2397                 fp->mode = TPA_MODE_GRO;
2398
2399         /* We don't want TPA on an FCoE L2 ring */
2400         if (IS_FCOE_FP(fp))
2401                 fp->disable_tpa = 1;
2402 }
2403
2404 int bnx2x_load_cnic(struct bnx2x *bp)
2405 {
2406         int i, rc, port = BP_PORT(bp);
2407
2408         DP(NETIF_MSG_IFUP, "Starting CNIC-related load\n");
2409
2410         mutex_init(&bp->cnic_mutex);
2411
2412         if (IS_PF(bp)) {
2413                 rc = bnx2x_alloc_mem_cnic(bp);
2414                 if (rc) {
2415                         BNX2X_ERR("Unable to allocate bp memory for cnic\n");
2416                         LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic0);
2417                 }
2418         }
2419
2420         rc = bnx2x_alloc_fp_mem_cnic(bp);
2421         if (rc) {
2422                 BNX2X_ERR("Unable to allocate memory for cnic fps\n");
2423                 LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic0);
2424         }
2425
2426         /* Update the number of queues with the cnic queues */
2427         rc = bnx2x_set_real_num_queues(bp, 1);
2428         if (rc) {
2429                 BNX2X_ERR("Unable to set real_num_queues including cnic\n");
2430                 LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic0);
2431         }
2432
2433         /* Add all CNIC NAPI objects */
2434         bnx2x_add_all_napi_cnic(bp);
2435         DP(NETIF_MSG_IFUP, "cnic napi added\n");
2436         bnx2x_napi_enable_cnic(bp);
2437
2438         rc = bnx2x_init_hw_func_cnic(bp);
2439         if (rc)
2440                 LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic1);
2441
2442         bnx2x_nic_init_cnic(bp);
2443
2444         if (IS_PF(bp)) {
2445                 /* Enable Timer scan */
2446                 REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + port*4, 1);
2447
2448                 /* setup cnic queues */
2449                 for_each_cnic_queue(bp, i) {
2450                         rc = bnx2x_setup_queue(bp, &bp->fp[i], 0);
2451                         if (rc) {
2452                                 BNX2X_ERR("Queue setup failed\n");
2453                                 LOAD_ERROR_EXIT(bp, load_error_cnic2);
2454                         }
2455                 }
2456         }
2457
2458         /* Initialize Rx filter. */
2459         bnx2x_set_rx_mode_inner(bp);
2460
2461         /* re-read iscsi info */
2462         bnx2x_get_iscsi_info(bp);
2463         bnx2x_setup_cnic_irq_info(bp);
2464         bnx2x_setup_cnic_info(bp);
2465         bp->cnic_loaded = true;
2466         if (bp->state == BNX2X_STATE_OPEN)
2467                 bnx2x_cnic_notify(bp, CNIC_CTL_START_CMD);
2468
2469         DP(NETIF_MSG_IFUP, "Ending successfully CNIC-related load\n");
2470
2471         return 0;
2472
2473 #ifndef BNX2X_STOP_ON_ERROR
2474 load_error_cnic2:
2475         /* Disable Timer scan */
2476         REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + port*4, 0);
2477
2478 load_error_cnic1:
2479         bnx2x_napi_disable_cnic(bp);
2480         /* Update the number of queues without the cnic queues */
2481         if (bnx2x_set_real_num_queues(bp, 0))
2482                 BNX2X_ERR("Unable to set real_num_queues not including cnic\n");
2483 load_error_cnic0:
2484         BNX2X_ERR("CNIC-related load failed\n");
2485         bnx2x_free_fp_mem_cnic(bp);
2486         bnx2x_free_mem_cnic(bp);
2487         return rc;
2488 #endif /* ! BNX2X_STOP_ON_ERROR */
2489 }
2490
2491 /* must be called with rtnl_lock */
2492 int bnx2x_nic_load(struct bnx2x *bp, int load_mode)
2493 {
2494         int port = BP_PORT(bp);
2495         int i, rc = 0, load_code = 0;
2496
2497         DP(NETIF_MSG_IFUP, "Starting NIC load\n");
2498         DP(NETIF_MSG_IFUP,
2499            "CNIC is %s\n", CNIC_ENABLED(bp) ? "enabled" : "disabled");
2500
2501 #ifdef BNX2X_STOP_ON_ERROR
2502         if (unlikely(bp->panic)) {
2503                 BNX2X_ERR("Can't load NIC when there is panic\n");
2504                 return -EPERM;
2505         }
2506 #endif
2507
2508         bp->state = BNX2X_STATE_OPENING_WAIT4_LOAD;
2509
2510         /* zero the structure w/o any lock, before SP handler is initialized */
2511         memset(&bp->last_reported_link, 0, sizeof(bp->last_reported_link));
2512         __set_bit(BNX2X_LINK_REPORT_LINK_DOWN,
2513                 &bp->last_reported_link.link_report_flags);
2514
2515         if (IS_PF(bp))
2516                 /* must be called before memory allocation and HW init */
2517                 bnx2x_ilt_set_info(bp);
2518
2519         /*
2520          * Zero fastpath structures preserving invariants like napi, which are
2521          * allocated only once, fp index, max_cos, bp pointer.
2522          * Also set fp->disable_tpa and txdata_ptr.
2523          */
2524         DP(NETIF_MSG_IFUP, "num queues: %d", bp->num_queues);
2525         for_each_queue(bp, i)
2526                 bnx2x_bz_fp(bp, i);
2527         memset(bp->bnx2x_txq, 0, (BNX2X_MAX_RSS_COUNT(bp) * BNX2X_MULTI_TX_COS +
2528                                   bp->num_cnic_queues) *
2529                                   sizeof(struct bnx2x_fp_txdata));
2530
2531         bp->fcoe_init = false;
2532
2533         /* Set the receive queues buffer size */
2534         bnx2x_set_rx_buf_size(bp);
2535
2536         if (IS_PF(bp)) {
2537                 rc = bnx2x_alloc_mem(bp);
2538                 if (rc) {
2539                         BNX2X_ERR("Unable to allocate bp memory\n");
2540                         return rc;
2541                 }
2542         }
2543
2544         /* need to be done after alloc mem, since it's self adjusting to amount
2545          * of memory available for RSS queues
2546          */
2547         rc = bnx2x_alloc_fp_mem(bp);
2548         if (rc) {
2549                 BNX2X_ERR("Unable to allocate memory for fps\n");
2550                 LOAD_ERROR_EXIT(bp, load_error0);
2551         }
2552
2553         /* Allocated memory for FW statistics  */
2554         if (bnx2x_alloc_fw_stats_mem(bp))
2555                 LOAD_ERROR_EXIT(bp, load_error0);
2556
2557         /* request pf to initialize status blocks */
2558         if (IS_VF(bp)) {
2559                 rc = bnx2x_vfpf_init(bp);
2560                 if (rc)
2561                         LOAD_ERROR_EXIT(bp, load_error0);
2562         }
2563
2564         /* As long as bnx2x_alloc_mem() may possibly update
2565          * bp->num_queues, bnx2x_set_real_num_queues() should always
2566          * come after it. At this stage cnic queues are not counted.
2567          */
2568         rc = bnx2x_set_real_num_queues(bp, 0);
2569         if (rc) {
2570                 BNX2X_ERR("Unable to set real_num_queues\n");
2571                 LOAD_ERROR_EXIT(bp, load_error0);
2572         }
2573
2574         /* configure multi cos mappings in kernel.
2575          * this configuration may be overridden by a multi class queue
2576          * discipline or by a dcbx negotiation result.
2577          */
2578         bnx2x_setup_tc(bp->dev, bp->max_cos);
2579
2580         /* Add all NAPI objects */
2581         bnx2x_add_all_napi(bp);
2582         DP(NETIF_MSG_IFUP, "napi added\n");
2583         bnx2x_napi_enable(bp);
2584
2585         if (IS_PF(bp)) {
2586                 /* set pf load just before approaching the MCP */
2587                 bnx2x_set_pf_load(bp);
2588
2589                 /* if mcp exists send load request and analyze response */
2590                 if (!BP_NOMCP(bp)) {
2591                         /* attempt to load pf */
2592                         rc = bnx2x_nic_load_request(bp, &load_code);
2593                         if (rc)
2594                                 LOAD_ERROR_EXIT(bp, load_error1);
2595
2596                         /* what did mcp say? */
2597                         rc = bnx2x_nic_load_analyze_req(bp, load_code);
2598                         if (rc) {
2599                                 bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2600                                 LOAD_ERROR_EXIT(bp, load_error2);
2601                         }
2602                 } else {
2603                         load_code = bnx2x_nic_load_no_mcp(bp, port);
2604                 }
2605
2606                 /* mark pmf if applicable */
2607                 bnx2x_nic_load_pmf(bp, load_code);
2608
2609                 /* Init Function state controlling object */
2610                 bnx2x__init_func_obj(bp);
2611
2612                 /* Initialize HW */
2613                 rc = bnx2x_init_hw(bp, load_code);
2614                 if (rc) {
2615                         BNX2X_ERR("HW init failed, aborting\n");
2616                         bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2617                         LOAD_ERROR_EXIT(bp, load_error2);
2618                 }
2619         }
2620
2621         bnx2x_pre_irq_nic_init(bp);
2622
2623         /* Connect to IRQs */
2624         rc = bnx2x_setup_irqs(bp);
2625         if (rc) {
2626                 BNX2X_ERR("setup irqs failed\n");
2627                 if (IS_PF(bp))
2628                         bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2629                 LOAD_ERROR_EXIT(bp, load_error2);
2630         }
2631
2632         /* Init per-function objects */
2633         if (IS_PF(bp)) {
2634                 /* Setup NIC internals and enable interrupts */
2635                 bnx2x_post_irq_nic_init(bp, load_code);
2636
2637                 bnx2x_init_bp_objs(bp);
2638                 bnx2x_iov_nic_init(bp);
2639
2640                 /* Set AFEX default VLAN tag to an invalid value */
2641                 bp->afex_def_vlan_tag = -1;
2642                 bnx2x_nic_load_afex_dcc(bp, load_code);
2643                 bp->state = BNX2X_STATE_OPENING_WAIT4_PORT;
2644                 rc = bnx2x_func_start(bp);
2645                 if (rc) {
2646                         BNX2X_ERR("Function start failed!\n");
2647                         bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2648
2649                         LOAD_ERROR_EXIT(bp, load_error3);
2650                 }
2651
2652                 /* Send LOAD_DONE command to MCP */
2653                 if (!BP_NOMCP(bp)) {
2654                         load_code = bnx2x_fw_command(bp,
2655                                                      DRV_MSG_CODE_LOAD_DONE, 0);
2656                         if (!load_code) {
2657                                 BNX2X_ERR("MCP response failure, aborting\n");
2658                                 rc = -EBUSY;
2659                                 LOAD_ERROR_EXIT(bp, load_error3);
2660                         }
2661                 }
2662
2663                 /* initialize FW coalescing state machines in RAM */
2664                 bnx2x_update_coalesce(bp);
2665         }
2666
2667         /* setup the leading queue */
2668         rc = bnx2x_setup_leading(bp);
2669         if (rc) {
2670                 BNX2X_ERR("Setup leading failed!\n");
2671                 LOAD_ERROR_EXIT(bp, load_error3);
2672         }
2673
2674         /* set up the rest of the queues */
2675         for_each_nondefault_eth_queue(bp, i) {
2676                 if (IS_PF(bp))
2677                         rc = bnx2x_setup_queue(bp, &bp->fp[i], false);
2678                 else /* VF */
2679                         rc = bnx2x_vfpf_setup_q(bp, &bp->fp[i], false);
2680                 if (rc) {
2681                         BNX2X_ERR("Queue %d setup failed\n", i);
2682                         LOAD_ERROR_EXIT(bp, load_error3);
2683                 }
2684         }
2685
2686         /* setup rss */
2687         rc = bnx2x_init_rss(bp);
2688         if (rc) {
2689                 BNX2X_ERR("PF RSS init failed\n");
2690                 LOAD_ERROR_EXIT(bp, load_error3);
2691         }
2692
2693         /* Now when Clients are configured we are ready to work */
2694         bp->state = BNX2X_STATE_OPEN;
2695
2696         /* Configure a ucast MAC */
2697         if (IS_PF(bp))
2698                 rc = bnx2x_set_eth_mac(bp, true);
2699         else /* vf */
2700                 rc = bnx2x_vfpf_config_mac(bp, bp->dev->dev_addr, bp->fp->index,
2701                                            true);
2702         if (rc) {
2703                 BNX2X_ERR("Setting Ethernet MAC failed\n");
2704                 LOAD_ERROR_EXIT(bp, load_error3);
2705         }
2706
2707         if (IS_PF(bp) && bp->pending_max) {
2708                 bnx2x_update_max_mf_config(bp, bp->pending_max);
2709                 bp->pending_max = 0;
2710         }
2711
2712         if (bp->port.pmf) {
2713                 rc = bnx2x_initial_phy_init(bp, load_mode);
2714                 if (rc)
2715                         LOAD_ERROR_EXIT(bp, load_error3);
2716         }
2717         bp->link_params.feature_config_flags &= ~FEATURE_CONFIG_BOOT_FROM_SAN;
2718
2719         /* Start fast path */
2720
2721         /* Initialize Rx filter. */
2722         bnx2x_set_rx_mode_inner(bp);
2723
2724         /* Start the Tx */
2725         switch (load_mode) {
2726         case LOAD_NORMAL:
2727                 /* Tx queue should be only re-enabled */
2728                 netif_tx_wake_all_queues(bp->dev);
2729                 break;
2730
2731         case LOAD_OPEN:
2732                 netif_tx_start_all_queues(bp->dev);
2733                 smp_mb__after_clear_bit();
2734                 break;
2735
2736         case LOAD_DIAG:
2737         case LOAD_LOOPBACK_EXT:
2738                 bp->state = BNX2X_STATE_DIAG;
2739                 break;
2740
2741         default:
2742                 break;
2743         }
2744
2745         if (bp->port.pmf)
2746                 bnx2x_update_drv_flags(bp, 1 << DRV_FLAGS_PORT_MASK, 0);
2747         else
2748                 bnx2x__link_status_update(bp);
2749
2750         /* start the timer */
2751         mod_timer(&bp->timer, jiffies + bp->current_interval);
2752
2753         if (CNIC_ENABLED(bp))
2754                 bnx2x_load_cnic(bp);
2755
2756         if (IS_PF(bp) && SHMEM2_HAS(bp, drv_capabilities_flag)) {
2757                 /* mark driver is loaded in shmem2 */
2758                 u32 val;
2759                 val = SHMEM2_RD(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)]);
2760                 SHMEM2_WR(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)],
2761                           val | DRV_FLAGS_CAPABILITIES_LOADED_SUPPORTED |
2762                           DRV_FLAGS_CAPABILITIES_LOADED_L2);
2763         }
2764
2765         /* Wait for all pending SP commands to complete */
2766         if (IS_PF(bp) && !bnx2x_wait_sp_comp(bp, ~0x0UL)) {
2767                 BNX2X_ERR("Timeout waiting for SP elements to complete\n");
2768                 bnx2x_nic_unload(bp, UNLOAD_CLOSE, false);
2769                 return -EBUSY;
2770         }
2771
2772         /* If PMF - send ADMIN DCBX msg to MFW to initiate DCBX FSM */
2773         if (bp->port.pmf && (bp->state != BNX2X_STATE_DIAG))
2774                 bnx2x_dcbx_init(bp, false);
2775
2776         DP(NETIF_MSG_IFUP, "Ending successfully NIC load\n");
2777
2778         return 0;
2779
2780 #ifndef BNX2X_STOP_ON_ERROR
2781 load_error3:
2782         if (IS_PF(bp)) {
2783                 bnx2x_int_disable_sync(bp, 1);
2784
2785                 /* Clean queueable objects */
2786                 bnx2x_squeeze_objects(bp);
2787         }
2788
2789         /* Free SKBs, SGEs, TPA pool and driver internals */
2790         bnx2x_free_skbs(bp);
2791         for_each_rx_queue(bp, i)
2792                 bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
2793
2794         /* Release IRQs */
2795         bnx2x_free_irq(bp);
2796 load_error2:
2797         if (IS_PF(bp) && !BP_NOMCP(bp)) {
2798                 bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_REQ_WOL_MCP, 0);
2799                 bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_DONE, 0);
2800         }
2801
2802         bp->port.pmf = 0;
2803 load_error1:
2804         bnx2x_napi_disable(bp);
2805         bnx2x_del_all_napi(bp);
2806
2807         /* clear pf_load status, as it was already set */
2808         if (IS_PF(bp))
2809                 bnx2x_clear_pf_load(bp);
2810 load_error0:
2811         bnx2x_free_fw_stats_mem(bp);
2812         bnx2x_free_fp_mem(bp);
2813         bnx2x_free_mem(bp);
2814
2815         return rc;
2816 #endif /* ! BNX2X_STOP_ON_ERROR */
2817 }
2818
2819 int bnx2x_drain_tx_queues(struct bnx2x *bp)
2820 {
2821         u8 rc = 0, cos, i;
2822
2823         /* Wait until tx fastpath tasks complete */
2824         for_each_tx_queue(bp, i) {
2825                 struct bnx2x_fastpath *fp = &bp->fp[i];
2826
2827                 for_each_cos_in_tx_queue(fp, cos)
2828                         rc = bnx2x_clean_tx_queue(bp, fp->txdata_ptr[cos]);
2829                 if (rc)
2830                         return rc;
2831         }
2832         return 0;
2833 }
2834
2835 /* must be called with rtnl_lock */
2836 int bnx2x_nic_unload(struct bnx2x *bp, int unload_mode, bool keep_link)
2837 {
2838         int i;
2839         bool global = false;
2840
2841         DP(NETIF_MSG_IFUP, "Starting NIC unload\n");
2842
2843         /* mark driver is unloaded in shmem2 */
2844         if (IS_PF(bp) && SHMEM2_HAS(bp, drv_capabilities_flag)) {
2845                 u32 val;
2846                 val = SHMEM2_RD(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)]);
2847                 SHMEM2_WR(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)],
2848                           val & ~DRV_FLAGS_CAPABILITIES_LOADED_L2);
2849         }
2850
2851         if (IS_PF(bp) && bp->recovery_state != BNX2X_RECOVERY_DONE &&
2852             (bp->state == BNX2X_STATE_CLOSED ||
2853              bp->state == BNX2X_STATE_ERROR)) {
2854                 /* We can get here if the driver has been unloaded
2855                  * during parity error recovery and is either waiting for a
2856                  * leader to complete or for other functions to unload and
2857                  * then ifdown has been issued. In this case we want to
2858                  * unload and let other functions to complete a recovery
2859                  * process.
2860                  */
2861                 bp->recovery_state = BNX2X_RECOVERY_DONE;
2862                 bp->is_leader = 0;
2863                 bnx2x_release_leader_lock(bp);
2864                 smp_mb();
2865
2866                 DP(NETIF_MSG_IFDOWN, "Releasing a leadership...\n");
2867                 BNX2X_ERR("Can't unload in closed or error state\n");
2868                 return -EINVAL;
2869         }
2870
2871         /* Nothing to do during unload if previous bnx2x_nic_load()
2872          * have not completed successfully - all resources are released.
2873          *
2874          * we can get here only after unsuccessful ndo_* callback, during which
2875          * dev->IFF_UP flag is still on.
2876          */
2877         if (bp->state == BNX2X_STATE_CLOSED || bp->state == BNX2X_STATE_ERROR)
2878                 return 0;
2879
2880         /* It's important to set the bp->state to the value different from
2881          * BNX2X_STATE_OPEN and only then stop the Tx. Otherwise bnx2x_tx_int()
2882          * may restart the Tx from the NAPI context (see bnx2x_tx_int()).
2883          */
2884         bp->state = BNX2X_STATE_CLOSING_WAIT4_HALT;
2885         smp_mb();
2886
2887         /* indicate to VFs that the PF is going down */
2888         bnx2x_iov_channel_down(bp);
2889
2890         if (CNIC_LOADED(bp))
2891                 bnx2x_cnic_notify(bp, CNIC_CTL_STOP_CMD);
2892
2893         /* Stop Tx */
2894         bnx2x_tx_disable(bp);
2895         netdev_reset_tc(bp->dev);
2896
2897         bp->rx_mode = BNX2X_RX_MODE_NONE;
2898
2899         del_timer_sync(&bp->timer);
2900
2901         if (IS_PF(bp)) {
2902                 /* Set ALWAYS_ALIVE bit in shmem */
2903                 bp->fw_drv_pulse_wr_seq |= DRV_PULSE_ALWAYS_ALIVE;
2904                 bnx2x_drv_pulse(bp);
2905                 bnx2x_stats_handle(bp, STATS_EVENT_STOP);
2906                 bnx2x_save_statistics(bp);
2907         }
2908
2909         /* wait till consumers catch up with producers in all queues */
2910         bnx2x_drain_tx_queues(bp);
2911
2912         /* if VF indicate to PF this function is going down (PF will delete sp
2913          * elements and clear initializations
2914          */
2915         if (IS_VF(bp))
2916                 bnx2x_vfpf_close_vf(bp);
2917         else if (unload_mode != UNLOAD_RECOVERY)
2918                 /* if this is a normal/close unload need to clean up chip*/
2919                 bnx2x_chip_cleanup(bp, unload_mode, keep_link);
2920         else {
2921                 /* Send the UNLOAD_REQUEST to the MCP */
2922                 bnx2x_send_unload_req(bp, unload_mode);
2923
2924                 /* Prevent transactions to host from the functions on the
2925                  * engine that doesn't reset global blocks in case of global
2926                  * attention once global blocks are reset and gates are opened
2927                  * (the engine which leader will perform the recovery
2928                  * last).
2929                  */
2930                 if (!CHIP_IS_E1x(bp))
2931                         bnx2x_pf_disable(bp);
2932
2933                 /* Disable HW interrupts, NAPI */
2934                 bnx2x_netif_stop(bp, 1);
2935                 /* Delete all NAPI objects */
2936                 bnx2x_del_all_napi(bp);
2937                 if (CNIC_LOADED(bp))
2938                         bnx2x_del_all_napi_cnic(bp);
2939                 /* Release IRQs */
2940                 bnx2x_free_irq(bp);
2941
2942                 /* Report UNLOAD_DONE to MCP */
2943                 bnx2x_send_unload_done(bp, false);
2944         }
2945
2946         /*
2947          * At this stage no more interrupts will arrive so we may safely clean
2948          * the queueable objects here in case they failed to get cleaned so far.
2949          */
2950         if (IS_PF(bp))
2951                 bnx2x_squeeze_objects(bp);
2952
2953         /* There should be no more pending SP commands at this stage */
2954         bp->sp_state = 0;
2955
2956         bp->port.pmf = 0;
2957
2958         /* clear pending work in rtnl task */
2959         bp->sp_rtnl_state = 0;
2960         smp_mb();
2961
2962         /* Free SKBs, SGEs, TPA pool and driver internals */
2963         bnx2x_free_skbs(bp);
2964         if (CNIC_LOADED(bp))
2965                 bnx2x_free_skbs_cnic(bp);
2966         for_each_rx_queue(bp, i)
2967                 bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
2968
2969         bnx2x_free_fp_mem(bp);
2970         if (CNIC_LOADED(bp))
2971                 bnx2x_free_fp_mem_cnic(bp);
2972
2973         if (IS_PF(bp)) {
2974                 if (CNIC_LOADED(bp))
2975                         bnx2x_free_mem_cnic(bp);
2976         }
2977         bnx2x_free_mem(bp);
2978
2979         bp->state = BNX2X_STATE_CLOSED;
2980         bp->cnic_loaded = false;
2981
2982         /* Check if there are pending parity attentions. If there are - set
2983          * RECOVERY_IN_PROGRESS.
2984          */
2985         if (IS_PF(bp) && bnx2x_chk_parity_attn(bp, &global, false)) {
2986                 bnx2x_set_reset_in_progress(bp);
2987
2988                 /* Set RESET_IS_GLOBAL if needed */
2989                 if (global)
2990                         bnx2x_set_reset_global(bp);
2991         }
2992
2993         /* The last driver must disable a "close the gate" if there is no
2994          * parity attention or "process kill" pending.
2995          */
2996         if (IS_PF(bp) &&
2997             !bnx2x_clear_pf_load(bp) &&
2998             bnx2x_reset_is_done(bp, BP_PATH(bp)))
2999                 bnx2x_disable_close_the_gate(bp);
3000
3001         DP(NETIF_MSG_IFUP, "Ending NIC unload\n");
3002
3003         return 0;
3004 }
3005
3006 int bnx2x_set_power_state(struct bnx2x *bp, pci_power_t state)
3007 {
3008         u16 pmcsr;
3009
3010         /* If there is no power capability, silently succeed */
3011         if (!bp->pdev->pm_cap) {
3012                 BNX2X_DEV_INFO("No power capability. Breaking.\n");
3013                 return 0;
3014         }
3015
3016         pci_read_config_word(bp->pdev, bp->pdev->pm_cap + PCI_PM_CTRL, &pmcsr);
3017
3018         switch (state) {
3019         case PCI_D0:
3020                 pci_write_config_word(bp->pdev, bp->pdev->pm_cap + PCI_PM_CTRL,
3021                                       ((pmcsr & ~PCI_PM_CTRL_STATE_MASK) |
3022                                        PCI_PM_CTRL_PME_STATUS));
3023
3024                 if (pmcsr & PCI_PM_CTRL_STATE_MASK)
3025                         /* delay required during transition out of D3hot */
3026                         msleep(20);
3027                 break;
3028
3029         case PCI_D3hot:
3030                 /* If there are other clients above don't
3031                    shut down the power */
3032                 if (atomic_read(&bp->pdev->enable_cnt) != 1)
3033                         return 0;
3034                 /* Don't shut down the power for emulation and FPGA */
3035                 if (CHIP_REV_IS_SLOW(bp))
3036                         return 0;
3037
3038                 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
3039                 pmcsr |= 3;
3040
3041                 if (bp->wol)
3042                         pmcsr |= PCI_PM_CTRL_PME_ENABLE;
3043
3044                 pci_write_config_word(bp->pdev, bp->pdev->pm_cap + PCI_PM_CTRL,
3045                                       pmcsr);
3046
3047                 /* No more memory access after this point until
3048                 * device is brought back to D0.
3049                 */
3050                 break;
3051
3052         default:
3053                 dev_err(&bp->pdev->dev, "Can't support state = %d\n", state);
3054                 return -EINVAL;
3055         }
3056         return 0;
3057 }
3058
3059 /*
3060  * net_device service functions
3061  */
3062 int bnx2x_poll(struct napi_struct *napi, int budget)
3063 {
3064         int work_done = 0;
3065         u8 cos;
3066         struct bnx2x_fastpath *fp = container_of(napi, struct bnx2x_fastpath,
3067                                                  napi);
3068         struct bnx2x *bp = fp->bp;
3069
3070         while (1) {
3071 #ifdef BNX2X_STOP_ON_ERROR
3072                 if (unlikely(bp->panic)) {
3073                         napi_complete(napi);
3074                         return 0;
3075                 }
3076 #endif
3077                 if (!bnx2x_fp_lock_napi(fp))
3078                         return work_done;
3079
3080                 for_each_cos_in_tx_queue(fp, cos)
3081                         if (bnx2x_tx_queue_has_work(fp->txdata_ptr[cos]))
3082                                 bnx2x_tx_int(bp, fp->txdata_ptr[cos]);
3083
3084                 if (bnx2x_has_rx_work(fp)) {
3085                         work_done += bnx2x_rx_int(fp, budget - work_done);
3086
3087                         /* must not complete if we consumed full budget */
3088                         if (work_done >= budget) {
3089                                 bnx2x_fp_unlock_napi(fp);
3090                                 break;
3091                         }
3092                 }
3093
3094                 /* Fall out from the NAPI loop if needed */
3095                 if (!bnx2x_fp_unlock_napi(fp) &&
3096                     !(bnx2x_has_rx_work(fp) || bnx2x_has_tx_work(fp))) {
3097
3098                         /* No need to update SB for FCoE L2 ring as long as
3099                          * it's connected to the default SB and the SB
3100                          * has been updated when NAPI was scheduled.
3101                          */
3102                         if (IS_FCOE_FP(fp)) {
3103                                 napi_complete(napi);
3104                                 break;
3105                         }
3106                         bnx2x_update_fpsb_idx(fp);
3107                         /* bnx2x_has_rx_work() reads the status block,
3108                          * thus we need to ensure that status block indices
3109                          * have been actually read (bnx2x_update_fpsb_idx)
3110                          * prior to this check (bnx2x_has_rx_work) so that
3111                          * we won't write the "newer" value of the status block
3112                          * to IGU (if there was a DMA right after
3113                          * bnx2x_has_rx_work and if there is no rmb, the memory
3114                          * reading (bnx2x_update_fpsb_idx) may be postponed
3115                          * to right before bnx2x_ack_sb). In this case there
3116                          * will never be another interrupt until there is
3117                          * another update of the status block, while there
3118                          * is still unhandled work.
3119                          */
3120                         rmb();
3121
3122                         if (!(bnx2x_has_rx_work(fp) || bnx2x_has_tx_work(fp))) {
3123                                 napi_complete(napi);
3124                                 /* Re-enable interrupts */
3125                                 DP(NETIF_MSG_RX_STATUS,
3126                                    "Update index to %d\n", fp->fp_hc_idx);
3127                                 bnx2x_ack_sb(bp, fp->igu_sb_id, USTORM_ID,
3128                                              le16_to_cpu(fp->fp_hc_idx),
3129                                              IGU_INT_ENABLE, 1);
3130                                 break;
3131                         }
3132                 }
3133         }
3134
3135         return work_done;
3136 }
3137
3138 #ifdef CONFIG_NET_RX_BUSY_POLL
3139 /* must be called with local_bh_disable()d */
3140 int bnx2x_low_latency_recv(struct napi_struct *napi)
3141 {
3142         struct bnx2x_fastpath *fp = container_of(napi, struct bnx2x_fastpath,
3143                                                  napi);
3144         struct bnx2x *bp = fp->bp;
3145         int found = 0;
3146
3147         if ((bp->state == BNX2X_STATE_CLOSED) ||
3148             (bp->state == BNX2X_STATE_ERROR) ||
3149             (bp->flags & (TPA_ENABLE_FLAG | GRO_ENABLE_FLAG)))
3150                 return LL_FLUSH_FAILED;
3151
3152         if (!bnx2x_fp_lock_poll(fp))
3153                 return LL_FLUSH_BUSY;
3154
3155         if (bnx2x_has_rx_work(fp))
3156                 found = bnx2x_rx_int(fp, 4);
3157
3158         bnx2x_fp_unlock_poll(fp);
3159
3160         return found;
3161 }
3162 #endif
3163
3164 /* we split the first BD into headers and data BDs
3165  * to ease the pain of our fellow microcode engineers
3166  * we use one mapping for both BDs
3167  */
3168 static u16 bnx2x_tx_split(struct bnx2x *bp,
3169                           struct bnx2x_fp_txdata *txdata,
3170                           struct sw_tx_bd *tx_buf,
3171                           struct eth_tx_start_bd **tx_bd, u16 hlen,
3172                           u16 bd_prod)
3173 {
3174         struct eth_tx_start_bd *h_tx_bd = *tx_bd;
3175         struct eth_tx_bd *d_tx_bd;
3176         dma_addr_t mapping;
3177         int old_len = le16_to_cpu(h_tx_bd->nbytes);
3178
3179         /* first fix first BD */
3180         h_tx_bd->nbytes = cpu_to_le16(hlen);
3181
3182         DP(NETIF_MSG_TX_QUEUED, "TSO split header size is %d (%x:%x)\n",
3183            h_tx_bd->nbytes, h_tx_bd->addr_hi, h_tx_bd->addr_lo);
3184
3185         /* now get a new data BD
3186          * (after the pbd) and fill it */
3187         bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3188         d_tx_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
3189
3190         mapping = HILO_U64(le32_to_cpu(h_tx_bd->addr_hi),
3191                            le32_to_cpu(h_tx_bd->addr_lo)) + hlen;
3192
3193         d_tx_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
3194         d_tx_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
3195         d_tx_bd->nbytes = cpu_to_le16(old_len - hlen);
3196
3197         /* this marks the BD as one that has no individual mapping */
3198         tx_buf->flags |= BNX2X_TSO_SPLIT_BD;
3199
3200         DP(NETIF_MSG_TX_QUEUED,
3201            "TSO split data size is %d (%x:%x)\n",
3202            d_tx_bd->nbytes, d_tx_bd->addr_hi, d_tx_bd->addr_lo);
3203
3204         /* update tx_bd */
3205         *tx_bd = (struct eth_tx_start_bd *)d_tx_bd;
3206
3207         return bd_prod;
3208 }
3209
3210 #define bswab32(b32) ((__force __le32) swab32((__force __u32) (b32)))
3211 #define bswab16(b16) ((__force __le16) swab16((__force __u16) (b16)))
3212 static __le16 bnx2x_csum_fix(unsigned char *t_header, u16 csum, s8 fix)
3213 {
3214         __sum16 tsum = (__force __sum16) csum;
3215
3216         if (fix > 0)
3217                 tsum = ~csum_fold(csum_sub((__force __wsum) csum,
3218                                   csum_partial(t_header - fix, fix, 0)));
3219
3220         else if (fix < 0)
3221                 tsum = ~csum_fold(csum_add((__force __wsum) csum,
3222                                   csum_partial(t_header, -fix, 0)));
3223
3224         return bswab16(tsum);
3225 }
3226
3227 static u32 bnx2x_xmit_type(struct bnx2x *bp, struct sk_buff *skb)
3228 {
3229         u32 rc;
3230         __u8 prot = 0;
3231         __be16 protocol;
3232
3233         if (skb->ip_summed != CHECKSUM_PARTIAL)
3234                 return XMIT_PLAIN;
3235
3236         protocol = vlan_get_protocol(skb);
3237         if (protocol == htons(ETH_P_IPV6)) {
3238                 rc = XMIT_CSUM_V6;
3239                 prot = ipv6_hdr(skb)->nexthdr;
3240         } else {
3241                 rc = XMIT_CSUM_V4;
3242                 prot = ip_hdr(skb)->protocol;
3243         }
3244
3245         if (!CHIP_IS_E1x(bp) && skb->encapsulation) {
3246                 if (inner_ip_hdr(skb)->version == 6) {
3247                         rc |= XMIT_CSUM_ENC_V6;
3248                         if (inner_ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3249                                 rc |= XMIT_CSUM_TCP;
3250                 } else {
3251                         rc |= XMIT_CSUM_ENC_V4;
3252                         if (inner_ip_hdr(skb)->protocol == IPPROTO_TCP)
3253                                 rc |= XMIT_CSUM_TCP;
3254                 }
3255         }
3256         if (prot == IPPROTO_TCP)
3257                 rc |= XMIT_CSUM_TCP;
3258
3259         if (skb_is_gso(skb)) {
3260                 if (skb_is_gso_v6(skb)) {
3261                         rc |= (XMIT_GSO_V6 | XMIT_CSUM_TCP);
3262                         if (rc & XMIT_CSUM_ENC)
3263                                 rc |= XMIT_GSO_ENC_V6;
3264                 } else {
3265                         rc |= (XMIT_GSO_V4 | XMIT_CSUM_TCP);
3266                         if (rc & XMIT_CSUM_ENC)
3267                                 rc |= XMIT_GSO_ENC_V4;
3268                 }
3269         }
3270
3271         return rc;
3272 }
3273
3274 #if (MAX_SKB_FRAGS >= MAX_FETCH_BD - 3)
3275 /* check if packet requires linearization (packet is too fragmented)
3276    no need to check fragmentation if page size > 8K (there will be no
3277    violation to FW restrictions) */
3278 static int bnx2x_pkt_req_lin(struct bnx2x *bp, struct sk_buff *skb,
3279                              u32 xmit_type)
3280 {
3281         int to_copy = 0;
3282         int hlen = 0;
3283         int first_bd_sz = 0;
3284
3285         /* 3 = 1 (for linear data BD) + 2 (for PBD and last BD) */
3286         if (skb_shinfo(skb)->nr_frags >= (MAX_FETCH_BD - 3)) {
3287
3288                 if (xmit_type & XMIT_GSO) {
3289                         unsigned short lso_mss = skb_shinfo(skb)->gso_size;
3290                         /* Check if LSO packet needs to be copied:
3291                            3 = 1 (for headers BD) + 2 (for PBD and last BD) */
3292                         int wnd_size = MAX_FETCH_BD - 3;
3293                         /* Number of windows to check */
3294                         int num_wnds = skb_shinfo(skb)->nr_frags - wnd_size;
3295                         int wnd_idx = 0;
3296                         int frag_idx = 0;
3297                         u32 wnd_sum = 0;
3298
3299                         /* Headers length */
3300                         hlen = (int)(skb_transport_header(skb) - skb->data) +
3301                                 tcp_hdrlen(skb);
3302
3303                         /* Amount of data (w/o headers) on linear part of SKB*/
3304                         first_bd_sz = skb_headlen(skb) - hlen;
3305
3306                         wnd_sum  = first_bd_sz;
3307
3308                         /* Calculate the first sum - it's special */
3309                         for (frag_idx = 0; frag_idx < wnd_size - 1; frag_idx++)
3310                                 wnd_sum +=
3311                                         skb_frag_size(&skb_shinfo(skb)->frags[frag_idx]);
3312
3313                         /* If there was data on linear skb data - check it */
3314                         if (first_bd_sz > 0) {
3315                                 if (unlikely(wnd_sum < lso_mss)) {
3316                                         to_copy = 1;
3317                                         goto exit_lbl;
3318                                 }
3319
3320                                 wnd_sum -= first_bd_sz;
3321                         }
3322
3323                         /* Others are easier: run through the frag list and
3324                            check all windows */
3325                         for (wnd_idx = 0; wnd_idx <= num_wnds; wnd_idx++) {
3326                                 wnd_sum +=
3327                           skb_frag_size(&skb_shinfo(skb)->frags[wnd_idx + wnd_size - 1]);
3328
3329                                 if (unlikely(wnd_sum < lso_mss)) {
3330                                         to_copy = 1;
3331                                         break;
3332                                 }
3333                                 wnd_sum -=
3334                                         skb_frag_size(&skb_shinfo(skb)->frags[wnd_idx]);
3335                         }
3336                 } else {
3337                         /* in non-LSO too fragmented packet should always
3338                            be linearized */
3339                         to_copy = 1;
3340                 }
3341         }
3342
3343 exit_lbl:
3344         if (unlikely(to_copy))
3345                 DP(NETIF_MSG_TX_QUEUED,
3346                    "Linearization IS REQUIRED for %s packet. num_frags %d  hlen %d  first_bd_sz %d\n",
3347                    (xmit_type & XMIT_GSO) ? "LSO" : "non-LSO",
3348                    skb_shinfo(skb)->nr_frags, hlen, first_bd_sz);
3349
3350         return to_copy;
3351 }
3352 #endif
3353
3354 static void bnx2x_set_pbd_gso_e2(struct sk_buff *skb, u32 *parsing_data,
3355                                  u32 xmit_type)
3356 {
3357         struct ipv6hdr *ipv6;
3358
3359         *parsing_data |= (skb_shinfo(skb)->gso_size <<
3360                               ETH_TX_PARSE_BD_E2_LSO_MSS_SHIFT) &
3361                               ETH_TX_PARSE_BD_E2_LSO_MSS;
3362
3363         if (xmit_type & XMIT_GSO_ENC_V6)
3364                 ipv6 = inner_ipv6_hdr(skb);
3365         else if (xmit_type & XMIT_GSO_V6)
3366                 ipv6 = ipv6_hdr(skb);
3367         else
3368                 ipv6 = NULL;
3369
3370         if (ipv6 && ipv6->nexthdr == NEXTHDR_IPV6)
3371                 *parsing_data |= ETH_TX_PARSE_BD_E2_IPV6_WITH_EXT_HDR;
3372 }
3373
3374 /**
3375  * bnx2x_set_pbd_gso - update PBD in GSO case.
3376  *
3377  * @skb:        packet skb
3378  * @pbd:        parse BD
3379  * @xmit_type:  xmit flags
3380  */
3381 static void bnx2x_set_pbd_gso(struct sk_buff *skb,
3382                               struct eth_tx_parse_bd_e1x *pbd,
3383                               struct eth_tx_start_bd *tx_start_bd,
3384                               u32 xmit_type)
3385 {
3386         pbd->lso_mss = cpu_to_le16(skb_shinfo(skb)->gso_size);
3387         pbd->tcp_send_seq = bswab32(tcp_hdr(skb)->seq);
3388         pbd->tcp_flags = pbd_tcp_flags(tcp_hdr(skb));
3389
3390         if (xmit_type & XMIT_GSO_V4) {
3391                 pbd->ip_id = bswab16(ip_hdr(skb)->id);
3392                 pbd->tcp_pseudo_csum =
3393                         bswab16(~csum_tcpudp_magic(ip_hdr(skb)->saddr,
3394                                                    ip_hdr(skb)->daddr,
3395                                                    0, IPPROTO_TCP, 0));
3396
3397                 /* GSO on 57710/57711 needs FW to calculate IP checksum */
3398                 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_IP_CSUM;
3399         } else {
3400                 pbd->tcp_pseudo_csum =
3401                         bswab16(~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3402                                                  &ipv6_hdr(skb)->daddr,
3403                                                  0, IPPROTO_TCP, 0));
3404         }
3405
3406         pbd->global_data |=
3407                 cpu_to_le16(ETH_TX_PARSE_BD_E1X_PSEUDO_CS_WITHOUT_LEN);
3408 }
3409
3410 /**
3411  * bnx2x_set_pbd_csum_enc - update PBD with checksum and return header length
3412  *
3413  * @bp:                 driver handle
3414  * @skb:                packet skb
3415  * @parsing_data:       data to be updated
3416  * @xmit_type:          xmit flags
3417  *
3418  * 57712/578xx related, when skb has encapsulation
3419  */
3420 static u8 bnx2x_set_pbd_csum_enc(struct bnx2x *bp, struct sk_buff *skb,
3421                                  u32 *parsing_data, u32 xmit_type)
3422 {
3423         *parsing_data |=
3424                 ((((u8 *)skb_inner_transport_header(skb) - skb->data) >> 1) <<
3425                 ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W_SHIFT) &
3426                 ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W;
3427
3428         if (xmit_type & XMIT_CSUM_TCP) {
3429                 *parsing_data |= ((inner_tcp_hdrlen(skb) / 4) <<
3430                         ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW_SHIFT) &
3431                         ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW;
3432
3433                 return skb_inner_transport_header(skb) +
3434                         inner_tcp_hdrlen(skb) - skb->data;
3435         }
3436
3437         /* We support checksum offload for TCP and UDP only.
3438          * No need to pass the UDP header length - it's a constant.
3439          */
3440         return skb_inner_transport_header(skb) +
3441                 sizeof(struct udphdr) - skb->data;
3442 }
3443
3444 /**
3445  * bnx2x_set_pbd_csum_e2 - update PBD with checksum and return header length
3446  *
3447  * @bp:                 driver handle
3448  * @skb:                packet skb
3449  * @parsing_data:       data to be updated
3450  * @xmit_type:          xmit flags
3451  *
3452  * 57712/578xx related
3453  */
3454 static u8 bnx2x_set_pbd_csum_e2(struct bnx2x *bp, struct sk_buff *skb,
3455                                 u32 *parsing_data, u32 xmit_type)
3456 {
3457         *parsing_data |=
3458                 ((((u8 *)skb_transport_header(skb) - skb->data) >> 1) <<
3459                 ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W_SHIFT) &
3460                 ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W;
3461
3462         if (xmit_type & XMIT_CSUM_TCP) {
3463                 *parsing_data |= ((tcp_hdrlen(skb) / 4) <<
3464                         ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW_SHIFT) &
3465                         ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW;
3466
3467                 return skb_transport_header(skb) + tcp_hdrlen(skb) - skb->data;
3468         }
3469         /* We support checksum offload for TCP and UDP only.
3470          * No need to pass the UDP header length - it's a constant.
3471          */
3472         return skb_transport_header(skb) + sizeof(struct udphdr) - skb->data;
3473 }
3474
3475 /* set FW indication according to inner or outer protocols if tunneled */
3476 static void bnx2x_set_sbd_csum(struct bnx2x *bp, struct sk_buff *skb,
3477                                struct eth_tx_start_bd *tx_start_bd,
3478                                u32 xmit_type)
3479 {
3480         tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_L4_CSUM;
3481
3482         if (xmit_type & (XMIT_CSUM_ENC_V6 | XMIT_CSUM_V6))
3483                 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_IPV6;
3484
3485         if (!(xmit_type & XMIT_CSUM_TCP))
3486                 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_IS_UDP;
3487 }
3488
3489 /**
3490  * bnx2x_set_pbd_csum - update PBD with checksum and return header length
3491  *
3492  * @bp:         driver handle
3493  * @skb:        packet skb
3494  * @pbd:        parse BD to be updated
3495  * @xmit_type:  xmit flags
3496  */
3497 static u8 bnx2x_set_pbd_csum(struct bnx2x *bp, struct sk_buff *skb,
3498                              struct eth_tx_parse_bd_e1x *pbd,
3499                              u32 xmit_type)
3500 {
3501         u8 hlen = (skb_network_header(skb) - skb->data) >> 1;
3502
3503         /* for now NS flag is not used in Linux */
3504         pbd->global_data =
3505                 cpu_to_le16(hlen |
3506                             ((skb->protocol == cpu_to_be16(ETH_P_8021Q)) <<
3507                              ETH_TX_PARSE_BD_E1X_LLC_SNAP_EN_SHIFT));
3508
3509         pbd->ip_hlen_w = (skb_transport_header(skb) -
3510                         skb_network_header(skb)) >> 1;
3511
3512         hlen += pbd->ip_hlen_w;
3513
3514         /* We support checksum offload for TCP and UDP only */
3515         if (xmit_type & XMIT_CSUM_TCP)
3516                 hlen += tcp_hdrlen(skb) / 2;
3517         else
3518                 hlen += sizeof(struct udphdr) / 2;
3519
3520         pbd->total_hlen_w = cpu_to_le16(hlen);
3521         hlen = hlen*2;
3522
3523         if (xmit_type & XMIT_CSUM_TCP) {
3524                 pbd->tcp_pseudo_csum = bswab16(tcp_hdr(skb)->check);
3525
3526         } else {
3527                 s8 fix = SKB_CS_OFF(skb); /* signed! */
3528
3529                 DP(NETIF_MSG_TX_QUEUED,
3530                    "hlen %d  fix %d  csum before fix %x\n",
3531                    le16_to_cpu(pbd->total_hlen_w), fix, SKB_CS(skb));
3532
3533                 /* HW bug: fixup the CSUM */
3534                 pbd->tcp_pseudo_csum =
3535                         bnx2x_csum_fix(skb_transport_header(skb),
3536                                        SKB_CS(skb), fix);
3537
3538                 DP(NETIF_MSG_TX_QUEUED, "csum after fix %x\n",
3539                    pbd->tcp_pseudo_csum);
3540         }
3541
3542         return hlen;
3543 }
3544
3545 static void bnx2x_update_pbds_gso_enc(struct sk_buff *skb,
3546                                       struct eth_tx_parse_bd_e2 *pbd_e2,
3547                                       struct eth_tx_parse_2nd_bd *pbd2,
3548                                       u16 *global_data,
3549                                       u32 xmit_type)
3550 {
3551         u16 hlen_w = 0;
3552         u8 outerip_off, outerip_len = 0;
3553
3554         /* from outer IP to transport */
3555         hlen_w = (skb_inner_transport_header(skb) -
3556                   skb_network_header(skb)) >> 1;
3557
3558         /* transport len */
3559         hlen_w += inner_tcp_hdrlen(skb) >> 1;
3560
3561         pbd2->fw_ip_hdr_to_payload_w = hlen_w;
3562
3563         /* outer IP header info */
3564         if (xmit_type & XMIT_CSUM_V4) {
3565                 struct iphdr *iph = ip_hdr(skb);
3566                 u32 csum = (__force u32)(~iph->check) -
3567                            (__force u32)iph->tot_len -
3568                            (__force u32)iph->frag_off;
3569
3570                 pbd2->fw_ip_csum_wo_len_flags_frag =
3571                         bswab16(csum_fold((__force __wsum)csum));
3572         } else {
3573                 pbd2->fw_ip_hdr_to_payload_w =
3574                         hlen_w - ((sizeof(struct ipv6hdr)) >> 1);
3575         }
3576
3577         pbd2->tcp_send_seq = bswab32(inner_tcp_hdr(skb)->seq);
3578
3579         pbd2->tcp_flags = pbd_tcp_flags(inner_tcp_hdr(skb));
3580
3581         if (xmit_type & XMIT_GSO_V4) {
3582                 pbd2->hw_ip_id = bswab16(inner_ip_hdr(skb)->id);
3583
3584                 pbd_e2->data.tunnel_data.pseudo_csum =
3585                         bswab16(~csum_tcpudp_magic(
3586                                         inner_ip_hdr(skb)->saddr,
3587                                         inner_ip_hdr(skb)->daddr,
3588                                         0, IPPROTO_TCP, 0));
3589
3590                 outerip_len = ip_hdr(skb)->ihl << 1;
3591         } else {
3592                 pbd_e2->data.tunnel_data.pseudo_csum =
3593                         bswab16(~csum_ipv6_magic(
3594                                         &inner_ipv6_hdr(skb)->saddr,
3595                                         &inner_ipv6_hdr(skb)->daddr,
3596                                         0, IPPROTO_TCP, 0));
3597         }
3598
3599         outerip_off = (skb_network_header(skb) - skb->data) >> 1;
3600
3601         *global_data |=
3602                 outerip_off |
3603                 (!!(xmit_type & XMIT_CSUM_V6) <<
3604                         ETH_TX_PARSE_2ND_BD_IP_HDR_TYPE_OUTER_SHIFT) |
3605                 (outerip_len <<
3606                         ETH_TX_PARSE_2ND_BD_IP_HDR_LEN_OUTER_W_SHIFT) |
3607                 ((skb->protocol == cpu_to_be16(ETH_P_8021Q)) <<
3608                         ETH_TX_PARSE_2ND_BD_LLC_SNAP_EN_SHIFT);
3609
3610         if (ip_hdr(skb)->protocol == IPPROTO_UDP) {
3611                 SET_FLAG(*global_data, ETH_TX_PARSE_2ND_BD_TUNNEL_UDP_EXIST, 1);
3612                 pbd2->tunnel_udp_hdr_start_w = skb_transport_offset(skb) >> 1;
3613         }
3614 }
3615
3616 /* called with netif_tx_lock
3617  * bnx2x_tx_int() runs without netif_tx_lock unless it needs to call
3618  * netif_wake_queue()
3619  */
3620 netdev_tx_t bnx2x_start_xmit(struct sk_buff *skb, struct net_device *dev)
3621 {
3622         struct bnx2x *bp = netdev_priv(dev);
3623
3624         struct netdev_queue *txq;
3625         struct bnx2x_fp_txdata *txdata;
3626         struct sw_tx_bd *tx_buf;
3627         struct eth_tx_start_bd *tx_start_bd, *first_bd;
3628         struct eth_tx_bd *tx_data_bd, *total_pkt_bd = NULL;
3629         struct eth_tx_parse_bd_e1x *pbd_e1x = NULL;
3630         struct eth_tx_parse_bd_e2 *pbd_e2 = NULL;
3631         struct eth_tx_parse_2nd_bd *pbd2 = NULL;
3632         u32 pbd_e2_parsing_data = 0;
3633         u16 pkt_prod, bd_prod;
3634         int nbd, txq_index;
3635         dma_addr_t mapping;
3636         u32 xmit_type = bnx2x_xmit_type(bp, skb);
3637         int i;
3638         u8 hlen = 0;
3639         __le16 pkt_size = 0;
3640         struct ethhdr *eth;
3641         u8 mac_type = UNICAST_ADDRESS;
3642
3643 #ifdef BNX2X_STOP_ON_ERROR
3644         if (unlikely(bp->panic))
3645                 return NETDEV_TX_BUSY;
3646 #endif
3647
3648         txq_index = skb_get_queue_mapping(skb);
3649         txq = netdev_get_tx_queue(dev, txq_index);
3650
3651         BUG_ON(txq_index >= MAX_ETH_TXQ_IDX(bp) + (CNIC_LOADED(bp) ? 1 : 0));
3652
3653         txdata = &bp->bnx2x_txq[txq_index];
3654
3655         /* enable this debug print to view the transmission queue being used
3656         DP(NETIF_MSG_TX_QUEUED, "indices: txq %d, fp %d, txdata %d\n",
3657            txq_index, fp_index, txdata_index); */
3658
3659         /* enable this debug print to view the transmission details
3660         DP(NETIF_MSG_TX_QUEUED,
3661            "transmitting packet cid %d fp index %d txdata_index %d tx_data ptr %p fp pointer %p\n",
3662            txdata->cid, fp_index, txdata_index, txdata, fp); */
3663
3664         if (unlikely(bnx2x_tx_avail(bp, txdata) <
3665                         skb_shinfo(skb)->nr_frags +
3666                         BDS_PER_TX_PKT +
3667                         NEXT_CNT_PER_TX_PKT(MAX_BDS_PER_TX_PKT))) {
3668                 /* Handle special storage cases separately */
3669                 if (txdata->tx_ring_size == 0) {
3670                         struct bnx2x_eth_q_stats *q_stats =
3671                                 bnx2x_fp_qstats(bp, txdata->parent_fp);
3672                         q_stats->driver_filtered_tx_pkt++;
3673                         dev_kfree_skb(skb);
3674                         return NETDEV_TX_OK;
3675                 }
3676                 bnx2x_fp_qstats(bp, txdata->parent_fp)->driver_xoff++;
3677                 netif_tx_stop_queue(txq);
3678                 BNX2X_ERR("BUG! Tx ring full when queue awake!\n");
3679
3680                 return NETDEV_TX_BUSY;
3681         }
3682
3683         DP(NETIF_MSG_TX_QUEUED,
3684            "queue[%d]: SKB: summed %x  protocol %x protocol(%x,%x) gso type %x  xmit_type %x len %d\n",
3685            txq_index, skb->ip_summed, skb->protocol, ipv6_hdr(skb)->nexthdr,
3686            ip_hdr(skb)->protocol, skb_shinfo(skb)->gso_type, xmit_type,
3687            skb->len);
3688
3689         eth = (struct ethhdr *)skb->data;
3690
3691         /* set flag according to packet type (UNICAST_ADDRESS is default)*/
3692         if (unlikely(is_multicast_ether_addr(eth->h_dest))) {
3693                 if (is_broadcast_ether_addr(eth->h_dest))
3694                         mac_type = BROADCAST_ADDRESS;
3695                 else
3696                         mac_type = MULTICAST_ADDRESS;
3697         }
3698
3699 #if (MAX_SKB_FRAGS >= MAX_FETCH_BD - BDS_PER_TX_PKT)
3700         /* First, check if we need to linearize the skb (due to FW
3701            restrictions). No need to check fragmentation if page size > 8K
3702            (there will be no violation to FW restrictions) */
3703         if (bnx2x_pkt_req_lin(bp, skb, xmit_type)) {
3704                 /* Statistics of linearization */
3705                 bp->lin_cnt++;
3706                 if (skb_linearize(skb) != 0) {
3707                         DP(NETIF_MSG_TX_QUEUED,
3708                            "SKB linearization failed - silently dropping this SKB\n");
3709                         dev_kfree_skb_any(skb);
3710                         return NETDEV_TX_OK;
3711                 }
3712         }
3713 #endif
3714         /* Map skb linear data for DMA */
3715         mapping = dma_map_single(&bp->pdev->dev, skb->data,
3716                                  skb_headlen(skb), DMA_TO_DEVICE);
3717         if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
3718                 DP(NETIF_MSG_TX_QUEUED,
3719                    "SKB mapping failed - silently dropping this SKB\n");
3720                 dev_kfree_skb_any(skb);
3721                 return NETDEV_TX_OK;
3722         }
3723         /*
3724         Please read carefully. First we use one BD which we mark as start,
3725         then we have a parsing info BD (used for TSO or xsum),
3726         and only then we have the rest of the TSO BDs.
3727         (don't forget to mark the last one as last,
3728         and to unmap only AFTER you write to the BD ...)
3729         And above all, all pdb sizes are in words - NOT DWORDS!
3730         */
3731
3732         /* get current pkt produced now - advance it just before sending packet
3733          * since mapping of pages may fail and cause packet to be dropped
3734          */
3735         pkt_prod = txdata->tx_pkt_prod;
3736         bd_prod = TX_BD(txdata->tx_bd_prod);
3737
3738         /* get a tx_buf and first BD
3739          * tx_start_bd may be changed during SPLIT,
3740          * but first_bd will always stay first
3741          */
3742         tx_buf = &txdata->tx_buf_ring[TX_BD(pkt_prod)];
3743         tx_start_bd = &txdata->tx_desc_ring[bd_prod].start_bd;
3744         first_bd = tx_start_bd;
3745
3746         tx_start_bd->bd_flags.as_bitfield = ETH_TX_BD_FLAGS_START_BD;
3747
3748         /* header nbd: indirectly zero other flags! */
3749         tx_start_bd->general_data = 1 << ETH_TX_START_BD_HDR_NBDS_SHIFT;
3750
3751         /* remember the first BD of the packet */
3752         tx_buf->first_bd = txdata->tx_bd_prod;
3753         tx_buf->skb = skb;
3754         tx_buf->flags = 0;
3755
3756         DP(NETIF_MSG_TX_QUEUED,
3757            "sending pkt %u @%p  next_idx %u  bd %u @%p\n",
3758            pkt_prod, tx_buf, txdata->tx_pkt_prod, bd_prod, tx_start_bd);
3759
3760         if (vlan_tx_tag_present(skb)) {
3761                 tx_start_bd->vlan_or_ethertype =
3762                     cpu_to_le16(vlan_tx_tag_get(skb));
3763                 tx_start_bd->bd_flags.as_bitfield |=
3764                     (X_ETH_OUTBAND_VLAN << ETH_TX_BD_FLAGS_VLAN_MODE_SHIFT);
3765         } else {
3766                 /* when transmitting in a vf, start bd must hold the ethertype
3767                  * for fw to enforce it
3768                  */
3769                 if (IS_VF(bp))
3770                         tx_start_bd->vlan_or_ethertype =
3771                                 cpu_to_le16(ntohs(eth->h_proto));
3772                 else
3773                         /* used by FW for packet accounting */
3774                         tx_start_bd->vlan_or_ethertype = cpu_to_le16(pkt_prod);
3775         }
3776
3777         nbd = 2; /* start_bd + pbd + frags (updated when pages are mapped) */
3778
3779         /* turn on parsing and get a BD */
3780         bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3781
3782         if (xmit_type & XMIT_CSUM)
3783                 bnx2x_set_sbd_csum(bp, skb, tx_start_bd, xmit_type);
3784
3785         if (!CHIP_IS_E1x(bp)) {
3786                 pbd_e2 = &txdata->tx_desc_ring[bd_prod].parse_bd_e2;
3787                 memset(pbd_e2, 0, sizeof(struct eth_tx_parse_bd_e2));
3788
3789                 if (xmit_type & XMIT_CSUM_ENC) {
3790                         u16 global_data = 0;
3791
3792                         /* Set PBD in enc checksum offload case */
3793                         hlen = bnx2x_set_pbd_csum_enc(bp, skb,
3794                                                       &pbd_e2_parsing_data,
3795                                                       xmit_type);
3796
3797                         /* turn on 2nd parsing and get a BD */
3798                         bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3799
3800                         pbd2 = &txdata->tx_desc_ring[bd_prod].parse_2nd_bd;
3801
3802                         memset(pbd2, 0, sizeof(*pbd2));
3803
3804                         pbd_e2->data.tunnel_data.ip_hdr_start_inner_w =
3805                                 (skb_inner_network_header(skb) -
3806                                  skb->data) >> 1;
3807
3808                         if (xmit_type & XMIT_GSO_ENC)
3809                                 bnx2x_update_pbds_gso_enc(skb, pbd_e2, pbd2,
3810                                                           &global_data,
3811                                                           xmit_type);
3812
3813                         pbd2->global_data = cpu_to_le16(global_data);
3814
3815                         /* add addition parse BD indication to start BD */
3816                         SET_FLAG(tx_start_bd->general_data,
3817                                  ETH_TX_START_BD_PARSE_NBDS, 1);
3818                         /* set encapsulation flag in start BD */
3819                         SET_FLAG(tx_start_bd->general_data,
3820                                  ETH_TX_START_BD_TUNNEL_EXIST, 1);
3821                         nbd++;
3822                 } else if (xmit_type & XMIT_CSUM) {
3823                         /* Set PBD in checksum offload case w/o encapsulation */
3824                         hlen = bnx2x_set_pbd_csum_e2(bp, skb,
3825                                                      &pbd_e2_parsing_data,
3826                                                      xmit_type);
3827                 }
3828
3829                 /* Add the macs to the parsing BD this is a vf */
3830                 if (IS_VF(bp)) {
3831                         /* override GRE parameters in BD */
3832                         bnx2x_set_fw_mac_addr(&pbd_e2->data.mac_addr.src_hi,
3833                                               &pbd_e2->data.mac_addr.src_mid,
3834                                               &pbd_e2->data.mac_addr.src_lo,
3835                                               eth->h_source);
3836
3837                         bnx2x_set_fw_mac_addr(&pbd_e2->data.mac_addr.dst_hi,
3838                                               &pbd_e2->data.mac_addr.dst_mid,
3839                                               &pbd_e2->data.mac_addr.dst_lo,
3840                                               eth->h_dest);
3841                 }
3842
3843                 SET_FLAG(pbd_e2_parsing_data,
3844                          ETH_TX_PARSE_BD_E2_ETH_ADDR_TYPE, mac_type);
3845         } else {
3846                 u16 global_data = 0;
3847                 pbd_e1x = &txdata->tx_desc_ring[bd_prod].parse_bd_e1x;
3848                 memset(pbd_e1x, 0, sizeof(struct eth_tx_parse_bd_e1x));
3849                 /* Set PBD in checksum offload case */
3850                 if (xmit_type & XMIT_CSUM)
3851                         hlen = bnx2x_set_pbd_csum(bp, skb, pbd_e1x, xmit_type);
3852
3853                 SET_FLAG(global_data,
3854                          ETH_TX_PARSE_BD_E1X_ETH_ADDR_TYPE, mac_type);
3855                 pbd_e1x->global_data |= cpu_to_le16(global_data);
3856         }
3857
3858         /* Setup the data pointer of the first BD of the packet */
3859         tx_start_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
3860         tx_start_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
3861         tx_start_bd->nbytes = cpu_to_le16(skb_headlen(skb));
3862         pkt_size = tx_start_bd->nbytes;
3863
3864         DP(NETIF_MSG_TX_QUEUED,
3865            "first bd @%p  addr (%x:%x)  nbytes %d  flags %x  vlan %x\n",
3866            tx_start_bd, tx_start_bd->addr_hi, tx_start_bd->addr_lo,
3867            le16_to_cpu(tx_start_bd->nbytes),
3868            tx_start_bd->bd_flags.as_bitfield,
3869            le16_to_cpu(tx_start_bd->vlan_or_ethertype));
3870
3871         if (xmit_type & XMIT_GSO) {
3872
3873                 DP(NETIF_MSG_TX_QUEUED,
3874                    "TSO packet len %d  hlen %d  total len %d  tso size %d\n",
3875                    skb->len, hlen, skb_headlen(skb),
3876                    skb_shinfo(skb)->gso_size);
3877
3878                 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_SW_LSO;
3879
3880                 if (unlikely(skb_headlen(skb) > hlen)) {
3881                         nbd++;
3882                         bd_prod = bnx2x_tx_split(bp, txdata, tx_buf,
3883                                                  &tx_start_bd, hlen,
3884                                                  bd_prod);
3885                 }
3886                 if (!CHIP_IS_E1x(bp))
3887                         bnx2x_set_pbd_gso_e2(skb, &pbd_e2_parsing_data,
3888                                              xmit_type);
3889                 else
3890                         bnx2x_set_pbd_gso(skb, pbd_e1x, first_bd, xmit_type);
3891         }
3892
3893         /* Set the PBD's parsing_data field if not zero
3894          * (for the chips newer than 57711).
3895          */
3896         if (pbd_e2_parsing_data)
3897                 pbd_e2->parsing_data = cpu_to_le32(pbd_e2_parsing_data);
3898
3899         tx_data_bd = (struct eth_tx_bd *)tx_start_bd;
3900
3901         /* Handle fragmented skb */
3902         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3903                 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3904
3905                 mapping = skb_frag_dma_map(&bp->pdev->dev, frag, 0,
3906                                            skb_frag_size(frag), DMA_TO_DEVICE);
3907                 if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
3908                         unsigned int pkts_compl = 0, bytes_compl = 0;
3909
3910                         DP(NETIF_MSG_TX_QUEUED,
3911                            "Unable to map page - dropping packet...\n");
3912
3913                         /* we need unmap all buffers already mapped
3914                          * for this SKB;
3915                          * first_bd->nbd need to be properly updated
3916                          * before call to bnx2x_free_tx_pkt
3917                          */
3918                         first_bd->nbd = cpu_to_le16(nbd);
3919                         bnx2x_free_tx_pkt(bp, txdata,
3920                                           TX_BD(txdata->tx_pkt_prod),
3921                                           &pkts_compl, &bytes_compl);
3922                         return NETDEV_TX_OK;
3923                 }
3924
3925                 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3926                 tx_data_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
3927                 if (total_pkt_bd == NULL)
3928                         total_pkt_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
3929
3930                 tx_data_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
3931                 tx_data_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
3932                 tx_data_bd->nbytes = cpu_to_le16(skb_frag_size(frag));
3933                 le16_add_cpu(&pkt_size, skb_frag_size(frag));
3934                 nbd++;
3935
3936                 DP(NETIF_MSG_TX_QUEUED,
3937                    "frag %d  bd @%p  addr (%x:%x)  nbytes %d\n",
3938                    i, tx_data_bd, tx_data_bd->addr_hi, tx_data_bd->addr_lo,
3939                    le16_to_cpu(tx_data_bd->nbytes));
3940         }
3941
3942         DP(NETIF_MSG_TX_QUEUED, "last bd @%p\n", tx_data_bd);
3943
3944         /* update with actual num BDs */
3945         first_bd->nbd = cpu_to_le16(nbd);
3946
3947         bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3948
3949         /* now send a tx doorbell, counting the next BD
3950          * if the packet contains or ends with it
3951          */
3952         if (TX_BD_POFF(bd_prod) < nbd)
3953                 nbd++;
3954
3955         /* total_pkt_bytes should be set on the first data BD if
3956          * it's not an LSO packet and there is more than one
3957          * data BD. In this case pkt_size is limited by an MTU value.
3958          * However we prefer to set it for an LSO packet (while we don't
3959          * have to) in order to save some CPU cycles in a none-LSO
3960          * case, when we much more care about them.
3961          */
3962         if (total_pkt_bd != NULL)
3963                 total_pkt_bd->total_pkt_bytes = pkt_size;
3964
3965         if (pbd_e1x)
3966                 DP(NETIF_MSG_TX_QUEUED,
3967                    "PBD (E1X) @%p  ip_data %x  ip_hlen %u  ip_id %u  lso_mss %u  tcp_flags %x  xsum %x  seq %u  hlen %u\n",
3968                    pbd_e1x, pbd_e1x->global_data, pbd_e1x->ip_hlen_w,
3969                    pbd_e1x->ip_id, pbd_e1x->lso_mss, pbd_e1x->tcp_flags,
3970                    pbd_e1x->tcp_pseudo_csum, pbd_e1x->tcp_send_seq,
3971                     le16_to_cpu(pbd_e1x->total_hlen_w));
3972         if (pbd_e2)
3973                 DP(NETIF_MSG_TX_QUEUED,
3974                    "PBD (E2) @%p  dst %x %x %x src %x %x %x parsing_data %x\n",
3975                    pbd_e2,
3976                    pbd_e2->data.mac_addr.dst_hi,
3977                    pbd_e2->data.mac_addr.dst_mid,
3978                    pbd_e2->data.mac_addr.dst_lo,
3979                    pbd_e2->data.mac_addr.src_hi,
3980                    pbd_e2->data.mac_addr.src_mid,
3981                    pbd_e2->data.mac_addr.src_lo,
3982                    pbd_e2->parsing_data);
3983         DP(NETIF_MSG_TX_QUEUED, "doorbell: nbd %d  bd %u\n", nbd, bd_prod);
3984
3985         netdev_tx_sent_queue(txq, skb->len);
3986
3987         skb_tx_timestamp(skb);
3988
3989         txdata->tx_pkt_prod++;
3990         /*
3991          * Make sure that the BD data is updated before updating the producer
3992          * since FW might read the BD right after the producer is updated.
3993          * This is only applicable for weak-ordered memory model archs such
3994          * as IA-64. The following barrier is also mandatory since FW will
3995          * assumes packets must have BDs.
3996          */
3997         wmb();
3998
3999         txdata->tx_db.data.prod += nbd;
4000         barrier();
4001
4002         DOORBELL(bp, txdata->cid, txdata->tx_db.raw);
4003
4004         mmiowb();
4005
4006         txdata->tx_bd_prod += nbd;
4007
4008         if (unlikely(bnx2x_tx_avail(bp, txdata) < MAX_DESC_PER_TX_PKT)) {
4009                 netif_tx_stop_queue(txq);
4010
4011                 /* paired memory barrier is in bnx2x_tx_int(), we have to keep
4012                  * ordering of set_bit() in netif_tx_stop_queue() and read of
4013                  * fp->bd_tx_cons */
4014                 smp_mb();
4015
4016                 bnx2x_fp_qstats(bp, txdata->parent_fp)->driver_xoff++;
4017                 if (bnx2x_tx_avail(bp, txdata) >= MAX_DESC_PER_TX_PKT)
4018                         netif_tx_wake_queue(txq);
4019         }
4020         txdata->tx_pkt++;
4021
4022         return NETDEV_TX_OK;
4023 }
4024
4025 /**
4026  * bnx2x_setup_tc - routine to configure net_device for multi tc
4027  *
4028  * @netdev: net device to configure
4029  * @tc: number of traffic classes to enable
4030  *
4031  * callback connected to the ndo_setup_tc function pointer
4032  */
4033 int bnx2x_setup_tc(struct net_device *dev, u8 num_tc)
4034 {
4035         int cos, prio, count, offset;
4036         struct bnx2x *bp = netdev_priv(dev);
4037
4038         /* setup tc must be called under rtnl lock */
4039         ASSERT_RTNL();
4040
4041         /* no traffic classes requested. Aborting */
4042         if (!num_tc) {
4043                 netdev_reset_tc(dev);
4044                 return 0;
4045         }
4046
4047         /* requested to support too many traffic classes */
4048         if (num_tc > bp->max_cos) {
4049                 BNX2X_ERR("support for too many traffic classes requested: %d. Max supported is %d\n",
4050                           num_tc, bp->max_cos);
4051                 return -EINVAL;
4052         }
4053
4054         /* declare amount of supported traffic classes */
4055         if (netdev_set_num_tc(dev, num_tc)) {
4056                 BNX2X_ERR("failed to declare %d traffic classes\n", num_tc);
4057                 return -EINVAL;
4058         }
4059
4060         /* configure priority to traffic class mapping */
4061         for (prio = 0; prio < BNX2X_MAX_PRIORITY; prio++) {
4062                 netdev_set_prio_tc_map(dev, prio, bp->prio_to_cos[prio]);
4063                 DP(BNX2X_MSG_SP | NETIF_MSG_IFUP,
4064                    "mapping priority %d to tc %d\n",
4065                    prio, bp->prio_to_cos[prio]);
4066         }
4067
4068         /* Use this configuration to differentiate tc0 from other COSes
4069            This can be used for ets or pfc, and save the effort of setting
4070            up a multio class queue disc or negotiating DCBX with a switch
4071         netdev_set_prio_tc_map(dev, 0, 0);
4072         DP(BNX2X_MSG_SP, "mapping priority %d to tc %d\n", 0, 0);
4073         for (prio = 1; prio < 16; prio++) {
4074                 netdev_set_prio_tc_map(dev, prio, 1);
4075                 DP(BNX2X_MSG_SP, "mapping priority %d to tc %d\n", prio, 1);
4076         } */
4077
4078         /* configure traffic class to transmission queue mapping */
4079         for (cos = 0; cos < bp->max_cos; cos++) {
4080                 count = BNX2X_NUM_ETH_QUEUES(bp);
4081                 offset = cos * BNX2X_NUM_NON_CNIC_QUEUES(bp);
4082                 netdev_set_tc_queue(dev, cos, count, offset);
4083                 DP(BNX2X_MSG_SP | NETIF_MSG_IFUP,
4084                    "mapping tc %d to offset %d count %d\n",
4085                    cos, offset, count);
4086         }
4087
4088         return 0;
4089 }
4090
4091 /* called with rtnl_lock */
4092 int bnx2x_change_mac_addr(struct net_device *dev, void *p)
4093 {
4094         struct sockaddr *addr = p;
4095         struct bnx2x *bp = netdev_priv(dev);
4096         int rc = 0;
4097
4098         if (!bnx2x_is_valid_ether_addr(bp, addr->sa_data)) {
4099                 BNX2X_ERR("Requested MAC address is not valid\n");
4100                 return -EINVAL;
4101         }
4102
4103         if ((IS_MF_STORAGE_SD(bp) || IS_MF_FCOE_AFEX(bp)) &&
4104             !is_zero_ether_addr(addr->sa_data)) {
4105                 BNX2X_ERR("Can't configure non-zero address on iSCSI or FCoE functions in MF-SD mode\n");
4106                 return -EINVAL;
4107         }
4108
4109         if (netif_running(dev))  {
4110                 rc = bnx2x_set_eth_mac(bp, false);
4111                 if (rc)
4112                         return rc;
4113         }
4114
4115         memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
4116
4117         if (netif_running(dev))
4118                 rc = bnx2x_set_eth_mac(bp, true);
4119
4120         return rc;
4121 }
4122
4123 static void bnx2x_free_fp_mem_at(struct bnx2x *bp, int fp_index)
4124 {
4125         union host_hc_status_block *sb = &bnx2x_fp(bp, fp_index, status_blk);
4126         struct bnx2x_fastpath *fp = &bp->fp[fp_index];
4127         u8 cos;
4128
4129         /* Common */
4130
4131         if (IS_FCOE_IDX(fp_index)) {
4132                 memset(sb, 0, sizeof(union host_hc_status_block));
4133                 fp->status_blk_mapping = 0;
4134         } else {
4135                 /* status blocks */
4136                 if (!CHIP_IS_E1x(bp))
4137                         BNX2X_PCI_FREE(sb->e2_sb,
4138                                        bnx2x_fp(bp, fp_index,
4139                                                 status_blk_mapping),
4140                                        sizeof(struct host_hc_status_block_e2));
4141                 else
4142                         BNX2X_PCI_FREE(sb->e1x_sb,
4143                                        bnx2x_fp(bp, fp_index,
4144                                                 status_blk_mapping),
4145                                        sizeof(struct host_hc_status_block_e1x));
4146         }
4147
4148         /* Rx */
4149         if (!skip_rx_queue(bp, fp_index)) {
4150                 bnx2x_free_rx_bds(fp);
4151
4152                 /* fastpath rx rings: rx_buf rx_desc rx_comp */
4153                 BNX2X_FREE(bnx2x_fp(bp, fp_index, rx_buf_ring));
4154                 BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_desc_ring),
4155                                bnx2x_fp(bp, fp_index, rx_desc_mapping),
4156                                sizeof(struct eth_rx_bd) * NUM_RX_BD);
4157
4158                 BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_comp_ring),
4159                                bnx2x_fp(bp, fp_index, rx_comp_mapping),
4160                                sizeof(struct eth_fast_path_rx_cqe) *
4161                                NUM_RCQ_BD);
4162
4163                 /* SGE ring */
4164                 BNX2X_FREE(bnx2x_fp(bp, fp_index, rx_page_ring));
4165                 BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_sge_ring),
4166                                bnx2x_fp(bp, fp_index, rx_sge_mapping),
4167                                BCM_PAGE_SIZE * NUM_RX_SGE_PAGES);
4168         }
4169
4170         /* Tx */
4171         if (!skip_tx_queue(bp, fp_index)) {
4172                 /* fastpath tx rings: tx_buf tx_desc */
4173                 for_each_cos_in_tx_queue(fp, cos) {
4174                         struct bnx2x_fp_txdata *txdata = fp->txdata_ptr[cos];
4175
4176                         DP(NETIF_MSG_IFDOWN,
4177                            "freeing tx memory of fp %d cos %d cid %d\n",
4178                            fp_index, cos, txdata->cid);
4179
4180                         BNX2X_FREE(txdata->tx_buf_ring);
4181                         BNX2X_PCI_FREE(txdata->tx_desc_ring,
4182                                 txdata->tx_desc_mapping,
4183                                 sizeof(union eth_tx_bd_types) * NUM_TX_BD);
4184                 }
4185         }
4186         /* end of fastpath */
4187 }
4188
4189 void bnx2x_free_fp_mem_cnic(struct bnx2x *bp)
4190 {
4191         int i;
4192         for_each_cnic_queue(bp, i)
4193                 bnx2x_free_fp_mem_at(bp, i);
4194 }
4195
4196 void bnx2x_free_fp_mem(struct bnx2x *bp)
4197 {
4198         int i;
4199         for_each_eth_queue(bp, i)
4200                 bnx2x_free_fp_mem_at(bp, i);
4201 }
4202
4203 static void set_sb_shortcuts(struct bnx2x *bp, int index)
4204 {
4205         union host_hc_status_block status_blk = bnx2x_fp(bp, index, status_blk);
4206         if (!CHIP_IS_E1x(bp)) {
4207                 bnx2x_fp(bp, index, sb_index_values) =
4208                         (__le16 *)status_blk.e2_sb->sb.index_values;
4209                 bnx2x_fp(bp, index, sb_running_index) =
4210                         (__le16 *)status_blk.e2_sb->sb.running_index;
4211         } else {
4212                 bnx2x_fp(bp, index, sb_index_values) =
4213                         (__le16 *)status_blk.e1x_sb->sb.index_values;
4214                 bnx2x_fp(bp, index, sb_running_index) =
4215                         (__le16 *)status_blk.e1x_sb->sb.running_index;
4216         }
4217 }
4218
4219 /* Returns the number of actually allocated BDs */
4220 static int bnx2x_alloc_rx_bds(struct bnx2x_fastpath *fp,
4221                               int rx_ring_size)
4222 {
4223         struct bnx2x *bp = fp->bp;
4224         u16 ring_prod, cqe_ring_prod;
4225         int i, failure_cnt = 0;
4226
4227         fp->rx_comp_cons = 0;
4228         cqe_ring_prod = ring_prod = 0;
4229
4230         /* This routine is called only during fo init so
4231          * fp->eth_q_stats.rx_skb_alloc_failed = 0
4232          */
4233         for (i = 0; i < rx_ring_size; i++) {
4234                 if (bnx2x_alloc_rx_data(bp, fp, ring_prod, GFP_KERNEL) < 0) {
4235                         failure_cnt++;
4236                         continue;
4237                 }
4238                 ring_prod = NEXT_RX_IDX(ring_prod);
4239                 cqe_ring_prod = NEXT_RCQ_IDX(cqe_ring_prod);
4240                 WARN_ON(ring_prod <= (i - failure_cnt));
4241         }
4242
4243         if (failure_cnt)
4244                 BNX2X_ERR("was only able to allocate %d rx skbs on queue[%d]\n",
4245                           i - failure_cnt, fp->index);
4246
4247         fp->rx_bd_prod = ring_prod;
4248         /* Limit the CQE producer by the CQE ring size */
4249         fp->rx_comp_prod = min_t(u16, NUM_RCQ_RINGS*RCQ_DESC_CNT,
4250                                cqe_ring_prod);
4251         fp->rx_pkt = fp->rx_calls = 0;
4252
4253         bnx2x_fp_stats(bp, fp)->eth_q_stats.rx_skb_alloc_failed += failure_cnt;
4254
4255         return i - failure_cnt;
4256 }
4257
4258 static void bnx2x_set_next_page_rx_cq(struct bnx2x_fastpath *fp)
4259 {
4260         int i;
4261
4262         for (i = 1; i <= NUM_RCQ_RINGS; i++) {
4263                 struct eth_rx_cqe_next_page *nextpg;
4264
4265                 nextpg = (struct eth_rx_cqe_next_page *)
4266                         &fp->rx_comp_ring[RCQ_DESC_CNT * i - 1];
4267                 nextpg->addr_hi =
4268                         cpu_to_le32(U64_HI(fp->rx_comp_mapping +
4269                                    BCM_PAGE_SIZE*(i % NUM_RCQ_RINGS)));
4270                 nextpg->addr_lo =
4271                         cpu_to_le32(U64_LO(fp->rx_comp_mapping +
4272                                    BCM_PAGE_SIZE*(i % NUM_RCQ_RINGS)));
4273         }
4274 }
4275
4276 static int bnx2x_alloc_fp_mem_at(struct bnx2x *bp, int index)
4277 {
4278         union host_hc_status_block *sb;
4279         struct bnx2x_fastpath *fp = &bp->fp[index];
4280         int ring_size = 0;
4281         u8 cos;
4282         int rx_ring_size = 0;
4283
4284         if (!bp->rx_ring_size &&
4285             (IS_MF_STORAGE_SD(bp) || IS_MF_FCOE_AFEX(bp))) {
4286                 rx_ring_size = MIN_RX_SIZE_NONTPA;
4287                 bp->rx_ring_size = rx_ring_size;
4288         } else if (!bp->rx_ring_size) {
4289                 rx_ring_size = MAX_RX_AVAIL/BNX2X_NUM_RX_QUEUES(bp);
4290
4291                 if (CHIP_IS_E3(bp)) {
4292                         u32 cfg = SHMEM_RD(bp,
4293                                            dev_info.port_hw_config[BP_PORT(bp)].
4294                                            default_cfg);
4295
4296                         /* Decrease ring size for 1G functions */
4297                         if ((cfg & PORT_HW_CFG_NET_SERDES_IF_MASK) ==
4298                             PORT_HW_CFG_NET_SERDES_IF_SGMII)
4299                                 rx_ring_size /= 10;
4300                 }
4301
4302                 /* allocate at least number of buffers required by FW */
4303                 rx_ring_size = max_t(int, bp->disable_tpa ? MIN_RX_SIZE_NONTPA :
4304                                      MIN_RX_SIZE_TPA, rx_ring_size);
4305
4306                 bp->rx_ring_size = rx_ring_size;
4307         } else /* if rx_ring_size specified - use it */
4308                 rx_ring_size = bp->rx_ring_size;
4309
4310         DP(BNX2X_MSG_SP, "calculated rx_ring_size %d\n", rx_ring_size);
4311
4312         /* Common */
4313         sb = &bnx2x_fp(bp, index, status_blk);
4314
4315         if (!IS_FCOE_IDX(index)) {
4316                 /* status blocks */
4317                 if (!CHIP_IS_E1x(bp))
4318                         BNX2X_PCI_ALLOC(sb->e2_sb,
4319                                 &bnx2x_fp(bp, index, status_blk_mapping),
4320                                 sizeof(struct host_hc_status_block_e2));
4321                 else
4322                         BNX2X_PCI_ALLOC(sb->e1x_sb,
4323                                 &bnx2x_fp(bp, index, status_blk_mapping),
4324                             sizeof(struct host_hc_status_block_e1x));
4325         }
4326
4327         /* FCoE Queue uses Default SB and doesn't ACK the SB, thus no need to
4328          * set shortcuts for it.
4329          */
4330         if (!IS_FCOE_IDX(index))
4331                 set_sb_shortcuts(bp, index);
4332
4333         /* Tx */
4334         if (!skip_tx_queue(bp, index)) {
4335                 /* fastpath tx rings: tx_buf tx_desc */
4336                 for_each_cos_in_tx_queue(fp, cos) {
4337                         struct bnx2x_fp_txdata *txdata = fp->txdata_ptr[cos];
4338
4339                         DP(NETIF_MSG_IFUP,
4340                            "allocating tx memory of fp %d cos %d\n",
4341                            index, cos);
4342
4343                         BNX2X_ALLOC(txdata->tx_buf_ring,
4344                                 sizeof(struct sw_tx_bd) * NUM_TX_BD);
4345                         BNX2X_PCI_ALLOC(txdata->tx_desc_ring,
4346                                 &txdata->tx_desc_mapping,
4347                                 sizeof(union eth_tx_bd_types) * NUM_TX_BD);
4348                 }
4349         }
4350
4351         /* Rx */
4352         if (!skip_rx_queue(bp, index)) {
4353                 /* fastpath rx rings: rx_buf rx_desc rx_comp */
4354                 BNX2X_ALLOC(bnx2x_fp(bp, index, rx_buf_ring),
4355                                 sizeof(struct sw_rx_bd) * NUM_RX_BD);
4356                 BNX2X_PCI_ALLOC(bnx2x_fp(bp, index, rx_desc_ring),
4357                                 &bnx2x_fp(bp, index, rx_desc_mapping),
4358                                 sizeof(struct eth_rx_bd) * NUM_RX_BD);
4359
4360                 /* Seed all CQEs by 1s */
4361                 BNX2X_PCI_FALLOC(bnx2x_fp(bp, index, rx_comp_ring),
4362                                  &bnx2x_fp(bp, index, rx_comp_mapping),
4363                                  sizeof(struct eth_fast_path_rx_cqe) *
4364                                  NUM_RCQ_BD);
4365
4366                 /* SGE ring */
4367                 BNX2X_ALLOC(bnx2x_fp(bp, index, rx_page_ring),
4368                                 sizeof(struct sw_rx_page) * NUM_RX_SGE);
4369                 BNX2X_PCI_ALLOC(bnx2x_fp(bp, index, rx_sge_ring),
4370                                 &bnx2x_fp(bp, index, rx_sge_mapping),
4371                                 BCM_PAGE_SIZE * NUM_RX_SGE_PAGES);
4372                 /* RX BD ring */
4373                 bnx2x_set_next_page_rx_bd(fp);
4374
4375                 /* CQ ring */
4376                 bnx2x_set_next_page_rx_cq(fp);
4377
4378                 /* BDs */
4379                 ring_size = bnx2x_alloc_rx_bds(fp, rx_ring_size);
4380                 if (ring_size < rx_ring_size)
4381                         goto alloc_mem_err;
4382         }
4383
4384         return 0;
4385
4386 /* handles low memory cases */
4387 alloc_mem_err:
4388         BNX2X_ERR("Unable to allocate full memory for queue %d (size %d)\n",
4389                                                 index, ring_size);
4390         /* FW will drop all packets if queue is not big enough,
4391          * In these cases we disable the queue
4392          * Min size is different for OOO, TPA and non-TPA queues
4393          */
4394         if (ring_size < (fp->disable_tpa ?
4395                                 MIN_RX_SIZE_NONTPA : MIN_RX_SIZE_TPA)) {
4396                         /* release memory allocated for this queue */
4397                         bnx2x_free_fp_mem_at(bp, index);
4398                         return -ENOMEM;
4399         }
4400         return 0;
4401 }
4402
4403 int bnx2x_alloc_fp_mem_cnic(struct bnx2x *bp)
4404 {
4405         if (!NO_FCOE(bp))
4406                 /* FCoE */
4407                 if (bnx2x_alloc_fp_mem_at(bp, FCOE_IDX(bp)))
4408                         /* we will fail load process instead of mark
4409                          * NO_FCOE_FLAG
4410                          */
4411                         return -ENOMEM;
4412
4413         return 0;
4414 }
4415
4416 int bnx2x_alloc_fp_mem(struct bnx2x *bp)
4417 {
4418         int i;
4419
4420         /* 1. Allocate FP for leading - fatal if error
4421          * 2. Allocate RSS - fix number of queues if error
4422          */
4423
4424         /* leading */
4425         if (bnx2x_alloc_fp_mem_at(bp, 0))
4426                 return -ENOMEM;
4427
4428         /* RSS */
4429         for_each_nondefault_eth_queue(bp, i)
4430                 if (bnx2x_alloc_fp_mem_at(bp, i))
4431                         break;
4432
4433         /* handle memory failures */
4434         if (i != BNX2X_NUM_ETH_QUEUES(bp)) {
4435                 int delta = BNX2X_NUM_ETH_QUEUES(bp) - i;
4436
4437                 WARN_ON(delta < 0);
4438                 bnx2x_shrink_eth_fp(bp, delta);
4439                 if (CNIC_SUPPORT(bp))
4440                         /* move non eth FPs next to last eth FP
4441                          * must be done in that order
4442                          * FCOE_IDX < FWD_IDX < OOO_IDX
4443                          */
4444
4445                         /* move FCoE fp even NO_FCOE_FLAG is on */
4446                         bnx2x_move_fp(bp, FCOE_IDX(bp), FCOE_IDX(bp) - delta);
4447                 bp->num_ethernet_queues -= delta;
4448                 bp->num_queues = bp->num_ethernet_queues +
4449                                  bp->num_cnic_queues;
4450                 BNX2X_ERR("Adjusted num of queues from %d to %d\n",
4451                           bp->num_queues + delta, bp->num_queues);
4452         }
4453
4454         return 0;
4455 }
4456
4457 void bnx2x_free_mem_bp(struct bnx2x *bp)
4458 {
4459         int i;
4460
4461         for (i = 0; i < bp->fp_array_size; i++)
4462                 kfree(bp->fp[i].tpa_info);
4463         kfree(bp->fp);
4464         kfree(bp->sp_objs);
4465         kfree(bp->fp_stats);
4466         kfree(bp->bnx2x_txq);
4467         kfree(bp->msix_table);
4468         kfree(bp->ilt);
4469 }
4470
4471 int bnx2x_alloc_mem_bp(struct bnx2x *bp)
4472 {
4473         struct bnx2x_fastpath *fp;
4474         struct msix_entry *tbl;
4475         struct bnx2x_ilt *ilt;
4476         int msix_table_size = 0;
4477         int fp_array_size, txq_array_size;
4478         int i;
4479
4480         /*
4481          * The biggest MSI-X table we might need is as a maximum number of fast
4482          * path IGU SBs plus default SB (for PF only).
4483          */
4484         msix_table_size = bp->igu_sb_cnt;
4485         if (IS_PF(bp))
4486                 msix_table_size++;
4487         BNX2X_DEV_INFO("msix_table_size %d\n", msix_table_size);
4488
4489         /* fp array: RSS plus CNIC related L2 queues */
4490         fp_array_size = BNX2X_MAX_RSS_COUNT(bp) + CNIC_SUPPORT(bp);
4491         bp->fp_array_size = fp_array_size;
4492         BNX2X_DEV_INFO("fp_array_size %d\n", bp->fp_array_size);
4493
4494         fp = kcalloc(bp->fp_array_size, sizeof(*fp), GFP_KERNEL);
4495         if (!fp)
4496                 goto alloc_err;
4497         for (i = 0; i < bp->fp_array_size; i++) {
4498                 fp[i].tpa_info =
4499                         kcalloc(ETH_MAX_AGGREGATION_QUEUES_E1H_E2,
4500                                 sizeof(struct bnx2x_agg_info), GFP_KERNEL);
4501                 if (!(fp[i].tpa_info))
4502                         goto alloc_err;
4503         }
4504
4505         bp->fp = fp;
4506
4507         /* allocate sp objs */
4508         bp->sp_objs = kcalloc(bp->fp_array_size, sizeof(struct bnx2x_sp_objs),
4509                               GFP_KERNEL);
4510         if (!bp->sp_objs)
4511                 goto alloc_err;
4512
4513         /* allocate fp_stats */
4514         bp->fp_stats = kcalloc(bp->fp_array_size, sizeof(struct bnx2x_fp_stats),
4515                                GFP_KERNEL);
4516         if (!bp->fp_stats)
4517                 goto alloc_err;
4518
4519         /* Allocate memory for the transmission queues array */
4520         txq_array_size =
4521                 BNX2X_MAX_RSS_COUNT(bp) * BNX2X_MULTI_TX_COS + CNIC_SUPPORT(bp);
4522         BNX2X_DEV_INFO("txq_array_size %d", txq_array_size);
4523
4524         bp->bnx2x_txq = kcalloc(txq_array_size, sizeof(struct bnx2x_fp_txdata),
4525                                 GFP_KERNEL);
4526         if (!bp->bnx2x_txq)
4527                 goto alloc_err;
4528
4529         /* msix table */
4530         tbl = kcalloc(msix_table_size, sizeof(*tbl), GFP_KERNEL);
4531         if (!tbl)
4532                 goto alloc_err;
4533         bp->msix_table = tbl;
4534
4535         /* ilt */
4536         ilt = kzalloc(sizeof(*ilt), GFP_KERNEL);
4537         if (!ilt)
4538                 goto alloc_err;
4539         bp->ilt = ilt;
4540
4541         return 0;
4542 alloc_err:
4543         bnx2x_free_mem_bp(bp);
4544         return -ENOMEM;
4545 }
4546
4547 int bnx2x_reload_if_running(struct net_device *dev)
4548 {
4549         struct bnx2x *bp = netdev_priv(dev);
4550
4551         if (unlikely(!netif_running(dev)))
4552                 return 0;
4553
4554         bnx2x_nic_unload(bp, UNLOAD_NORMAL, true);
4555         return bnx2x_nic_load(bp, LOAD_NORMAL);
4556 }
4557
4558 int bnx2x_get_cur_phy_idx(struct bnx2x *bp)
4559 {
4560         u32 sel_phy_idx = 0;
4561         if (bp->link_params.num_phys <= 1)
4562                 return INT_PHY;
4563
4564         if (bp->link_vars.link_up) {
4565                 sel_phy_idx = EXT_PHY1;
4566                 /* In case link is SERDES, check if the EXT_PHY2 is the one */
4567                 if ((bp->link_vars.link_status & LINK_STATUS_SERDES_LINK) &&
4568                     (bp->link_params.phy[EXT_PHY2].supported & SUPPORTED_FIBRE))
4569                         sel_phy_idx = EXT_PHY2;
4570         } else {
4571
4572                 switch (bnx2x_phy_selection(&bp->link_params)) {
4573                 case PORT_HW_CFG_PHY_SELECTION_HARDWARE_DEFAULT:
4574                 case PORT_HW_CFG_PHY_SELECTION_FIRST_PHY:
4575                 case PORT_HW_CFG_PHY_SELECTION_FIRST_PHY_PRIORITY:
4576                        sel_phy_idx = EXT_PHY1;
4577                        break;
4578                 case PORT_HW_CFG_PHY_SELECTION_SECOND_PHY:
4579                 case PORT_HW_CFG_PHY_SELECTION_SECOND_PHY_PRIORITY:
4580                        sel_phy_idx = EXT_PHY2;
4581                        break;
4582                 }
4583         }
4584
4585         return sel_phy_idx;
4586 }
4587 int bnx2x_get_link_cfg_idx(struct bnx2x *bp)
4588 {
4589         u32 sel_phy_idx = bnx2x_get_cur_phy_idx(bp);
4590         /*
4591          * The selected activated PHY is always after swapping (in case PHY
4592          * swapping is enabled). So when swapping is enabled, we need to reverse
4593          * the configuration
4594          */
4595
4596         if (bp->link_params.multi_phy_config &
4597             PORT_HW_CFG_PHY_SWAPPED_ENABLED) {
4598                 if (sel_phy_idx == EXT_PHY1)
4599                         sel_phy_idx = EXT_PHY2;
4600                 else if (sel_phy_idx == EXT_PHY2)
4601                         sel_phy_idx = EXT_PHY1;
4602         }
4603         return LINK_CONFIG_IDX(sel_phy_idx);
4604 }
4605
4606 #ifdef NETDEV_FCOE_WWNN
4607 int bnx2x_fcoe_get_wwn(struct net_device *dev, u64 *wwn, int type)
4608 {
4609         struct bnx2x *bp = netdev_priv(dev);
4610         struct cnic_eth_dev *cp = &bp->cnic_eth_dev;
4611
4612         switch (type) {
4613         case NETDEV_FCOE_WWNN:
4614                 *wwn = HILO_U64(cp->fcoe_wwn_node_name_hi,
4615                                 cp->fcoe_wwn_node_name_lo);
4616                 break;
4617         case NETDEV_FCOE_WWPN:
4618                 *wwn = HILO_U64(cp->fcoe_wwn_port_name_hi,
4619                                 cp->fcoe_wwn_port_name_lo);
4620                 break;
4621         default:
4622                 BNX2X_ERR("Wrong WWN type requested - %d\n", type);
4623                 return -EINVAL;
4624         }
4625
4626         return 0;
4627 }
4628 #endif
4629
4630 /* called with rtnl_lock */
4631 int bnx2x_change_mtu(struct net_device *dev, int new_mtu)
4632 {
4633         struct bnx2x *bp = netdev_priv(dev);
4634
4635         if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
4636                 BNX2X_ERR("Can't perform change MTU during parity recovery\n");
4637                 return -EAGAIN;
4638         }
4639
4640         if ((new_mtu > ETH_MAX_JUMBO_PACKET_SIZE) ||
4641             ((new_mtu + ETH_HLEN) < ETH_MIN_PACKET_SIZE)) {
4642                 BNX2X_ERR("Can't support requested MTU size\n");
4643                 return -EINVAL;
4644         }
4645
4646         /* This does not race with packet allocation
4647          * because the actual alloc size is
4648          * only updated as part of load
4649          */
4650         dev->mtu = new_mtu;
4651
4652         return bnx2x_reload_if_running(dev);
4653 }
4654
4655 netdev_features_t bnx2x_fix_features(struct net_device *dev,
4656                                      netdev_features_t features)
4657 {
4658         struct bnx2x *bp = netdev_priv(dev);
4659
4660         /* TPA requires Rx CSUM offloading */
4661         if (!(features & NETIF_F_RXCSUM) || bp->disable_tpa) {
4662                 features &= ~NETIF_F_LRO;
4663                 features &= ~NETIF_F_GRO;
4664         }
4665
4666         return features;
4667 }
4668
4669 int bnx2x_set_features(struct net_device *dev, netdev_features_t features)
4670 {
4671         struct bnx2x *bp = netdev_priv(dev);
4672         u32 flags = bp->flags;
4673         u32 changes;
4674         bool bnx2x_reload = false;
4675
4676         if (features & NETIF_F_LRO)
4677                 flags |= TPA_ENABLE_FLAG;
4678         else
4679                 flags &= ~TPA_ENABLE_FLAG;
4680
4681         if (features & NETIF_F_GRO)
4682                 flags |= GRO_ENABLE_FLAG;
4683         else
4684                 flags &= ~GRO_ENABLE_FLAG;
4685
4686         if (features & NETIF_F_LOOPBACK) {
4687                 if (bp->link_params.loopback_mode != LOOPBACK_BMAC) {
4688                         bp->link_params.loopback_mode = LOOPBACK_BMAC;
4689                         bnx2x_reload = true;
4690                 }
4691         } else {
4692                 if (bp->link_params.loopback_mode != LOOPBACK_NONE) {
4693                         bp->link_params.loopback_mode = LOOPBACK_NONE;
4694                         bnx2x_reload = true;
4695                 }
4696         }
4697
4698         changes = flags ^ bp->flags;
4699
4700         /* if GRO is changed while LRO is enabled, don't force a reload */
4701         if ((changes & GRO_ENABLE_FLAG) && (flags & TPA_ENABLE_FLAG))
4702                 changes &= ~GRO_ENABLE_FLAG;
4703
4704         if (changes)
4705                 bnx2x_reload = true;
4706
4707         bp->flags = flags;
4708
4709         if (bnx2x_reload) {
4710                 if (bp->recovery_state == BNX2X_RECOVERY_DONE)
4711                         return bnx2x_reload_if_running(dev);
4712                 /* else: bnx2x_nic_load() will be called at end of recovery */
4713         }
4714
4715         return 0;
4716 }
4717
4718 void bnx2x_tx_timeout(struct net_device *dev)
4719 {
4720         struct bnx2x *bp = netdev_priv(dev);
4721
4722 #ifdef BNX2X_STOP_ON_ERROR
4723         if (!bp->panic)
4724                 bnx2x_panic();
4725 #endif
4726
4727         smp_mb__before_clear_bit();
4728         set_bit(BNX2X_SP_RTNL_TX_TIMEOUT, &bp->sp_rtnl_state);
4729         smp_mb__after_clear_bit();
4730
4731         /* This allows the netif to be shutdown gracefully before resetting */
4732         schedule_delayed_work(&bp->sp_rtnl_task, 0);
4733 }
4734
4735 int bnx2x_suspend(struct pci_dev *pdev, pm_message_t state)
4736 {
4737         struct net_device *dev = pci_get_drvdata(pdev);
4738         struct bnx2x *bp;
4739
4740         if (!dev) {
4741                 dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
4742                 return -ENODEV;
4743         }
4744         bp = netdev_priv(dev);
4745
4746         rtnl_lock();
4747
4748         pci_save_state(pdev);
4749
4750         if (!netif_running(dev)) {
4751                 rtnl_unlock();
4752                 return 0;
4753         }
4754
4755         netif_device_detach(dev);
4756
4757         bnx2x_nic_unload(bp, UNLOAD_CLOSE, false);
4758
4759         bnx2x_set_power_state(bp, pci_choose_state(pdev, state));
4760
4761         rtnl_unlock();
4762
4763         return 0;
4764 }
4765
4766 int bnx2x_resume(struct pci_dev *pdev)
4767 {
4768         struct net_device *dev = pci_get_drvdata(pdev);
4769         struct bnx2x *bp;
4770         int rc;
4771
4772         if (!dev) {
4773                 dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
4774                 return -ENODEV;
4775         }
4776         bp = netdev_priv(dev);
4777
4778         if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
4779                 BNX2X_ERR("Handling parity error recovery. Try again later\n");
4780                 return -EAGAIN;
4781         }
4782
4783         rtnl_lock();
4784
4785         pci_restore_state(pdev);
4786
4787         if (!netif_running(dev)) {
4788                 rtnl_unlock();
4789                 return 0;
4790         }
4791
4792         bnx2x_set_power_state(bp, PCI_D0);
4793         netif_device_attach(dev);
4794
4795         rc = bnx2x_nic_load(bp, LOAD_OPEN);
4796
4797         rtnl_unlock();
4798
4799         return rc;
4800 }
4801
4802 void bnx2x_set_ctx_validation(struct bnx2x *bp, struct eth_context *cxt,
4803                               u32 cid)
4804 {
4805         if (!cxt) {
4806                 BNX2X_ERR("bad context pointer %p\n", cxt);
4807                 return;
4808         }
4809
4810         /* ustorm cxt validation */
4811         cxt->ustorm_ag_context.cdu_usage =
4812                 CDU_RSRVD_VALUE_TYPE_A(HW_CID(bp, cid),
4813                         CDU_REGION_NUMBER_UCM_AG, ETH_CONNECTION_TYPE);
4814         /* xcontext validation */
4815         cxt->xstorm_ag_context.cdu_reserved =
4816                 CDU_RSRVD_VALUE_TYPE_A(HW_CID(bp, cid),
4817                         CDU_REGION_NUMBER_XCM_AG, ETH_CONNECTION_TYPE);
4818 }
4819
4820 static void storm_memset_hc_timeout(struct bnx2x *bp, u8 port,
4821                                     u8 fw_sb_id, u8 sb_index,
4822                                     u8 ticks)
4823 {
4824         u32 addr = BAR_CSTRORM_INTMEM +
4825                    CSTORM_STATUS_BLOCK_DATA_TIMEOUT_OFFSET(fw_sb_id, sb_index);
4826         REG_WR8(bp, addr, ticks);
4827         DP(NETIF_MSG_IFUP,
4828            "port %x fw_sb_id %d sb_index %d ticks %d\n",
4829            port, fw_sb_id, sb_index, ticks);
4830 }
4831
4832 static void storm_memset_hc_disable(struct bnx2x *bp, u8 port,
4833                                     u16 fw_sb_id, u8 sb_index,
4834                                     u8 disable)
4835 {
4836         u32 enable_flag = disable ? 0 : (1 << HC_INDEX_DATA_HC_ENABLED_SHIFT);
4837         u32 addr = BAR_CSTRORM_INTMEM +
4838                    CSTORM_STATUS_BLOCK_DATA_FLAGS_OFFSET(fw_sb_id, sb_index);
4839         u8 flags = REG_RD8(bp, addr);
4840         /* clear and set */
4841         flags &= ~HC_INDEX_DATA_HC_ENABLED;
4842         flags |= enable_flag;
4843         REG_WR8(bp, addr, flags);
4844         DP(NETIF_MSG_IFUP,
4845            "port %x fw_sb_id %d sb_index %d disable %d\n",
4846            port, fw_sb_id, sb_index, disable);
4847 }
4848
4849 void bnx2x_update_coalesce_sb_index(struct bnx2x *bp, u8 fw_sb_id,
4850                                     u8 sb_index, u8 disable, u16 usec)
4851 {
4852         int port = BP_PORT(bp);
4853         u8 ticks = usec / BNX2X_BTR;
4854
4855         storm_memset_hc_timeout(bp, port, fw_sb_id, sb_index, ticks);
4856
4857         disable = disable ? 1 : (usec ? 0 : 1);
4858         storm_memset_hc_disable(bp, port, fw_sb_id, sb_index, disable);
4859 }