1 /* bnx2x_cmn.c: QLogic Everest network driver.
3 * Copyright (c) 2007-2013 Broadcom Corporation
4 * Copyright (c) 2014 QLogic Corporation
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation.
11 * Maintained by: Ariel Elior <ariel.elior@qlogic.com>
12 * Written by: Eliezer Tamir
13 * Based on code from Michael Chan's bnx2 driver
14 * UDP CSUM errata workaround by Arik Gendelman
15 * Slowpath and fastpath rework by Vladislav Zolotarov
16 * Statistics and Link management by Yitchak Gertner
20 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
22 #include <linux/etherdevice.h>
23 #include <linux/if_vlan.h>
24 #include <linux/interrupt.h>
26 #include <linux/crash_dump.h>
29 #include <net/ip6_checksum.h>
30 #include <linux/prefetch.h>
31 #include "bnx2x_cmn.h"
32 #include "bnx2x_init.h"
35 static void bnx2x_free_fp_mem_cnic(struct bnx2x *bp);
36 static int bnx2x_alloc_fp_mem_cnic(struct bnx2x *bp);
37 static int bnx2x_alloc_fp_mem(struct bnx2x *bp);
38 static int bnx2x_poll(struct napi_struct *napi, int budget);
40 static void bnx2x_add_all_napi_cnic(struct bnx2x *bp)
44 /* Add NAPI objects */
45 for_each_rx_queue_cnic(bp, i) {
46 netif_napi_add(bp->dev, &bnx2x_fp(bp, i, napi),
47 bnx2x_poll, NAPI_POLL_WEIGHT);
51 static void bnx2x_add_all_napi(struct bnx2x *bp)
55 /* Add NAPI objects */
56 for_each_eth_queue(bp, i) {
57 netif_napi_add(bp->dev, &bnx2x_fp(bp, i, napi),
58 bnx2x_poll, NAPI_POLL_WEIGHT);
62 static int bnx2x_calc_num_queues(struct bnx2x *bp)
64 int nq = bnx2x_num_queues ? : netif_get_num_default_rss_queues();
66 /* Reduce memory usage in kdump environment by using only one queue */
67 if (is_kdump_kernel())
70 nq = clamp(nq, 1, BNX2X_MAX_QUEUES(bp));
75 * bnx2x_move_fp - move content of the fastpath structure.
78 * @from: source FP index
79 * @to: destination FP index
81 * Makes sure the contents of the bp->fp[to].napi is kept
82 * intact. This is done by first copying the napi struct from
83 * the target to the source, and then mem copying the entire
84 * source onto the target. Update txdata pointers and related
87 static inline void bnx2x_move_fp(struct bnx2x *bp, int from, int to)
89 struct bnx2x_fastpath *from_fp = &bp->fp[from];
90 struct bnx2x_fastpath *to_fp = &bp->fp[to];
91 struct bnx2x_sp_objs *from_sp_objs = &bp->sp_objs[from];
92 struct bnx2x_sp_objs *to_sp_objs = &bp->sp_objs[to];
93 struct bnx2x_fp_stats *from_fp_stats = &bp->fp_stats[from];
94 struct bnx2x_fp_stats *to_fp_stats = &bp->fp_stats[to];
95 int old_max_eth_txqs, new_max_eth_txqs;
96 int old_txdata_index = 0, new_txdata_index = 0;
97 struct bnx2x_agg_info *old_tpa_info = to_fp->tpa_info;
99 /* Copy the NAPI object as it has been already initialized */
100 from_fp->napi = to_fp->napi;
102 /* Move bnx2x_fastpath contents */
103 memcpy(to_fp, from_fp, sizeof(*to_fp));
106 /* Retain the tpa_info of the original `to' version as we don't want
107 * 2 FPs to contain the same tpa_info pointer.
109 to_fp->tpa_info = old_tpa_info;
111 /* move sp_objs contents as well, as their indices match fp ones */
112 memcpy(to_sp_objs, from_sp_objs, sizeof(*to_sp_objs));
114 /* move fp_stats contents as well, as their indices match fp ones */
115 memcpy(to_fp_stats, from_fp_stats, sizeof(*to_fp_stats));
117 /* Update txdata pointers in fp and move txdata content accordingly:
118 * Each fp consumes 'max_cos' txdata structures, so the index should be
119 * decremented by max_cos x delta.
122 old_max_eth_txqs = BNX2X_NUM_ETH_QUEUES(bp) * (bp)->max_cos;
123 new_max_eth_txqs = (BNX2X_NUM_ETH_QUEUES(bp) - from + to) *
125 if (from == FCOE_IDX(bp)) {
126 old_txdata_index = old_max_eth_txqs + FCOE_TXQ_IDX_OFFSET;
127 new_txdata_index = new_max_eth_txqs + FCOE_TXQ_IDX_OFFSET;
130 memcpy(&bp->bnx2x_txq[new_txdata_index],
131 &bp->bnx2x_txq[old_txdata_index],
132 sizeof(struct bnx2x_fp_txdata));
133 to_fp->txdata_ptr[0] = &bp->bnx2x_txq[new_txdata_index];
137 * bnx2x_fill_fw_str - Fill buffer with FW version string.
140 * @buf: character buffer to fill with the fw name
141 * @buf_len: length of the above buffer
144 void bnx2x_fill_fw_str(struct bnx2x *bp, char *buf, size_t buf_len)
147 u8 phy_fw_ver[PHY_FW_VER_LEN];
149 phy_fw_ver[0] = '\0';
150 bnx2x_get_ext_phy_fw_version(&bp->link_params,
151 phy_fw_ver, PHY_FW_VER_LEN);
152 strlcpy(buf, bp->fw_ver, buf_len);
153 snprintf(buf + strlen(bp->fw_ver), 32 - strlen(bp->fw_ver),
155 (bp->common.bc_ver & 0xff0000) >> 16,
156 (bp->common.bc_ver & 0xff00) >> 8,
157 (bp->common.bc_ver & 0xff),
158 ((phy_fw_ver[0] != '\0') ? " phy " : ""), phy_fw_ver);
160 bnx2x_vf_fill_fw_str(bp, buf, buf_len);
165 * bnx2x_shrink_eth_fp - guarantees fastpath structures stay intact
168 * @delta: number of eth queues which were not allocated
170 static void bnx2x_shrink_eth_fp(struct bnx2x *bp, int delta)
172 int i, cos, old_eth_num = BNX2X_NUM_ETH_QUEUES(bp);
174 /* Queue pointer cannot be re-set on an fp-basis, as moving pointer
175 * backward along the array could cause memory to be overridden
177 for (cos = 1; cos < bp->max_cos; cos++) {
178 for (i = 0; i < old_eth_num - delta; i++) {
179 struct bnx2x_fastpath *fp = &bp->fp[i];
180 int new_idx = cos * (old_eth_num - delta) + i;
182 memcpy(&bp->bnx2x_txq[new_idx], fp->txdata_ptr[cos],
183 sizeof(struct bnx2x_fp_txdata));
184 fp->txdata_ptr[cos] = &bp->bnx2x_txq[new_idx];
189 int bnx2x_load_count[2][3] = { {0} }; /* per-path: 0-common, 1-port0, 2-port1 */
191 /* free skb in the packet ring at pos idx
192 * return idx of last bd freed
194 static u16 bnx2x_free_tx_pkt(struct bnx2x *bp, struct bnx2x_fp_txdata *txdata,
195 u16 idx, unsigned int *pkts_compl,
196 unsigned int *bytes_compl)
198 struct sw_tx_bd *tx_buf = &txdata->tx_buf_ring[idx];
199 struct eth_tx_start_bd *tx_start_bd;
200 struct eth_tx_bd *tx_data_bd;
201 struct sk_buff *skb = tx_buf->skb;
202 u16 bd_idx = TX_BD(tx_buf->first_bd), new_cons;
204 u16 split_bd_len = 0;
206 /* prefetch skb end pointer to speedup dev_kfree_skb() */
209 DP(NETIF_MSG_TX_DONE, "fp[%d]: pkt_idx %d buff @(%p)->skb %p\n",
210 txdata->txq_index, idx, tx_buf, skb);
212 tx_start_bd = &txdata->tx_desc_ring[bd_idx].start_bd;
214 nbd = le16_to_cpu(tx_start_bd->nbd) - 1;
215 #ifdef BNX2X_STOP_ON_ERROR
216 if ((nbd - 1) > (MAX_SKB_FRAGS + 2)) {
217 BNX2X_ERR("BAD nbd!\n");
221 new_cons = nbd + tx_buf->first_bd;
223 /* Get the next bd */
224 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
226 /* Skip a parse bd... */
228 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
230 if (tx_buf->flags & BNX2X_HAS_SECOND_PBD) {
231 /* Skip second parse bd... */
233 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
236 /* TSO headers+data bds share a common mapping. See bnx2x_tx_split() */
237 if (tx_buf->flags & BNX2X_TSO_SPLIT_BD) {
238 tx_data_bd = &txdata->tx_desc_ring[bd_idx].reg_bd;
239 split_bd_len = BD_UNMAP_LEN(tx_data_bd);
241 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
245 dma_unmap_single(&bp->pdev->dev, BD_UNMAP_ADDR(tx_start_bd),
246 BD_UNMAP_LEN(tx_start_bd) + split_bd_len,
252 tx_data_bd = &txdata->tx_desc_ring[bd_idx].reg_bd;
253 dma_unmap_page(&bp->pdev->dev, BD_UNMAP_ADDR(tx_data_bd),
254 BD_UNMAP_LEN(tx_data_bd), DMA_TO_DEVICE);
256 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
263 (*bytes_compl) += skb->len;
264 dev_kfree_skb_any(skb);
267 tx_buf->first_bd = 0;
273 int bnx2x_tx_int(struct bnx2x *bp, struct bnx2x_fp_txdata *txdata)
275 struct netdev_queue *txq;
276 u16 hw_cons, sw_cons, bd_cons = txdata->tx_bd_cons;
277 unsigned int pkts_compl = 0, bytes_compl = 0;
279 #ifdef BNX2X_STOP_ON_ERROR
280 if (unlikely(bp->panic))
284 txq = netdev_get_tx_queue(bp->dev, txdata->txq_index);
285 hw_cons = le16_to_cpu(*txdata->tx_cons_sb);
286 sw_cons = txdata->tx_pkt_cons;
288 while (sw_cons != hw_cons) {
291 pkt_cons = TX_BD(sw_cons);
293 DP(NETIF_MSG_TX_DONE,
294 "queue[%d]: hw_cons %u sw_cons %u pkt_cons %u\n",
295 txdata->txq_index, hw_cons, sw_cons, pkt_cons);
297 bd_cons = bnx2x_free_tx_pkt(bp, txdata, pkt_cons,
298 &pkts_compl, &bytes_compl);
303 netdev_tx_completed_queue(txq, pkts_compl, bytes_compl);
305 txdata->tx_pkt_cons = sw_cons;
306 txdata->tx_bd_cons = bd_cons;
308 /* Need to make the tx_bd_cons update visible to start_xmit()
309 * before checking for netif_tx_queue_stopped(). Without the
310 * memory barrier, there is a small possibility that
311 * start_xmit() will miss it and cause the queue to be stopped
313 * On the other hand we need an rmb() here to ensure the proper
314 * ordering of bit testing in the following
315 * netif_tx_queue_stopped(txq) call.
319 if (unlikely(netif_tx_queue_stopped(txq))) {
320 /* Taking tx_lock() is needed to prevent re-enabling the queue
321 * while it's empty. This could have happen if rx_action() gets
322 * suspended in bnx2x_tx_int() after the condition before
323 * netif_tx_wake_queue(), while tx_action (bnx2x_start_xmit()):
325 * stops the queue->sees fresh tx_bd_cons->releases the queue->
326 * sends some packets consuming the whole queue again->
330 __netif_tx_lock(txq, smp_processor_id());
332 if ((netif_tx_queue_stopped(txq)) &&
333 (bp->state == BNX2X_STATE_OPEN) &&
334 (bnx2x_tx_avail(bp, txdata) >= MAX_DESC_PER_TX_PKT))
335 netif_tx_wake_queue(txq);
337 __netif_tx_unlock(txq);
342 static inline void bnx2x_update_last_max_sge(struct bnx2x_fastpath *fp,
345 u16 last_max = fp->last_max_sge;
347 if (SUB_S16(idx, last_max) > 0)
348 fp->last_max_sge = idx;
351 static inline void bnx2x_update_sge_prod(struct bnx2x_fastpath *fp,
353 struct eth_end_agg_rx_cqe *cqe)
355 struct bnx2x *bp = fp->bp;
356 u16 last_max, last_elem, first_elem;
363 /* First mark all used pages */
364 for (i = 0; i < sge_len; i++)
365 BIT_VEC64_CLEAR_BIT(fp->sge_mask,
366 RX_SGE(le16_to_cpu(cqe->sgl_or_raw_data.sgl[i])));
368 DP(NETIF_MSG_RX_STATUS, "fp_cqe->sgl[%d] = %d\n",
369 sge_len - 1, le16_to_cpu(cqe->sgl_or_raw_data.sgl[sge_len - 1]));
371 /* Here we assume that the last SGE index is the biggest */
372 prefetch((void *)(fp->sge_mask));
373 bnx2x_update_last_max_sge(fp,
374 le16_to_cpu(cqe->sgl_or_raw_data.sgl[sge_len - 1]));
376 last_max = RX_SGE(fp->last_max_sge);
377 last_elem = last_max >> BIT_VEC64_ELEM_SHIFT;
378 first_elem = RX_SGE(fp->rx_sge_prod) >> BIT_VEC64_ELEM_SHIFT;
380 /* If ring is not full */
381 if (last_elem + 1 != first_elem)
384 /* Now update the prod */
385 for (i = first_elem; i != last_elem; i = NEXT_SGE_MASK_ELEM(i)) {
386 if (likely(fp->sge_mask[i]))
389 fp->sge_mask[i] = BIT_VEC64_ELEM_ONE_MASK;
390 delta += BIT_VEC64_ELEM_SZ;
394 fp->rx_sge_prod += delta;
395 /* clear page-end entries */
396 bnx2x_clear_sge_mask_next_elems(fp);
399 DP(NETIF_MSG_RX_STATUS,
400 "fp->last_max_sge = %d fp->rx_sge_prod = %d\n",
401 fp->last_max_sge, fp->rx_sge_prod);
404 /* Get Toeplitz hash value in the skb using the value from the
405 * CQE (calculated by HW).
407 static u32 bnx2x_get_rxhash(const struct bnx2x *bp,
408 const struct eth_fast_path_rx_cqe *cqe,
409 enum pkt_hash_types *rxhash_type)
411 /* Get Toeplitz hash from CQE */
412 if ((bp->dev->features & NETIF_F_RXHASH) &&
413 (cqe->status_flags & ETH_FAST_PATH_RX_CQE_RSS_HASH_FLG)) {
414 enum eth_rss_hash_type htype;
416 htype = cqe->status_flags & ETH_FAST_PATH_RX_CQE_RSS_HASH_TYPE;
417 *rxhash_type = ((htype == TCP_IPV4_HASH_TYPE) ||
418 (htype == TCP_IPV6_HASH_TYPE)) ?
419 PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3;
421 return le32_to_cpu(cqe->rss_hash_result);
423 *rxhash_type = PKT_HASH_TYPE_NONE;
427 static void bnx2x_tpa_start(struct bnx2x_fastpath *fp, u16 queue,
429 struct eth_fast_path_rx_cqe *cqe)
431 struct bnx2x *bp = fp->bp;
432 struct sw_rx_bd *cons_rx_buf = &fp->rx_buf_ring[cons];
433 struct sw_rx_bd *prod_rx_buf = &fp->rx_buf_ring[prod];
434 struct eth_rx_bd *prod_bd = &fp->rx_desc_ring[prod];
436 struct bnx2x_agg_info *tpa_info = &fp->tpa_info[queue];
437 struct sw_rx_bd *first_buf = &tpa_info->first_buf;
439 /* print error if current state != stop */
440 if (tpa_info->tpa_state != BNX2X_TPA_STOP)
441 BNX2X_ERR("start of bin not in stop [%d]\n", queue);
443 /* Try to map an empty data buffer from the aggregation info */
444 mapping = dma_map_single(&bp->pdev->dev,
445 first_buf->data + NET_SKB_PAD,
446 fp->rx_buf_size, DMA_FROM_DEVICE);
448 * ...if it fails - move the skb from the consumer to the producer
449 * and set the current aggregation state as ERROR to drop it
450 * when TPA_STOP arrives.
453 if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
454 /* Move the BD from the consumer to the producer */
455 bnx2x_reuse_rx_data(fp, cons, prod);
456 tpa_info->tpa_state = BNX2X_TPA_ERROR;
460 /* move empty data from pool to prod */
461 prod_rx_buf->data = first_buf->data;
462 dma_unmap_addr_set(prod_rx_buf, mapping, mapping);
463 /* point prod_bd to new data */
464 prod_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
465 prod_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
467 /* move partial skb from cons to pool (don't unmap yet) */
468 *first_buf = *cons_rx_buf;
470 /* mark bin state as START */
471 tpa_info->parsing_flags =
472 le16_to_cpu(cqe->pars_flags.flags);
473 tpa_info->vlan_tag = le16_to_cpu(cqe->vlan_tag);
474 tpa_info->tpa_state = BNX2X_TPA_START;
475 tpa_info->len_on_bd = le16_to_cpu(cqe->len_on_bd);
476 tpa_info->placement_offset = cqe->placement_offset;
477 tpa_info->rxhash = bnx2x_get_rxhash(bp, cqe, &tpa_info->rxhash_type);
478 if (fp->mode == TPA_MODE_GRO) {
479 u16 gro_size = le16_to_cpu(cqe->pkt_len_or_gro_seg_len);
480 tpa_info->full_page = SGE_PAGES / gro_size * gro_size;
481 tpa_info->gro_size = gro_size;
484 #ifdef BNX2X_STOP_ON_ERROR
485 fp->tpa_queue_used |= (1 << queue);
486 DP(NETIF_MSG_RX_STATUS, "fp->tpa_queue_used = 0x%llx\n",
491 /* Timestamp option length allowed for TPA aggregation:
493 * nop nop kind length echo val
495 #define TPA_TSTAMP_OPT_LEN 12
497 * bnx2x_set_gro_params - compute GRO values
500 * @parsing_flags: parsing flags from the START CQE
501 * @len_on_bd: total length of the first packet for the
503 * @pkt_len: length of all segments
505 * Approximate value of the MSS for this aggregation calculated using
506 * the first packet of it.
507 * Compute number of aggregated segments, and gso_type.
509 static void bnx2x_set_gro_params(struct sk_buff *skb, u16 parsing_flags,
510 u16 len_on_bd, unsigned int pkt_len,
511 u16 num_of_coalesced_segs)
513 /* TPA aggregation won't have either IP options or TCP options
514 * other than timestamp or IPv6 extension headers.
516 u16 hdrs_len = ETH_HLEN + sizeof(struct tcphdr);
518 if (GET_FLAG(parsing_flags, PARSING_FLAGS_OVER_ETHERNET_PROTOCOL) ==
519 PRS_FLAG_OVERETH_IPV6) {
520 hdrs_len += sizeof(struct ipv6hdr);
521 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
523 hdrs_len += sizeof(struct iphdr);
524 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
527 /* Check if there was a TCP timestamp, if there is it's will
528 * always be 12 bytes length: nop nop kind length echo val.
530 * Otherwise FW would close the aggregation.
532 if (parsing_flags & PARSING_FLAGS_TIME_STAMP_EXIST_FLAG)
533 hdrs_len += TPA_TSTAMP_OPT_LEN;
535 skb_shinfo(skb)->gso_size = len_on_bd - hdrs_len;
537 /* tcp_gro_complete() will copy NAPI_GRO_CB(skb)->count
538 * to skb_shinfo(skb)->gso_segs
540 NAPI_GRO_CB(skb)->count = num_of_coalesced_segs;
543 static int bnx2x_alloc_rx_sge(struct bnx2x *bp, struct bnx2x_fastpath *fp,
544 u16 index, gfp_t gfp_mask)
546 struct sw_rx_page *sw_buf = &fp->rx_page_ring[index];
547 struct eth_rx_sge *sge = &fp->rx_sge_ring[index];
548 struct bnx2x_alloc_pool *pool = &fp->page_pool;
552 pool->page = alloc_pages(gfp_mask, PAGES_PER_SGE_SHIFT);
553 if (unlikely(!pool->page))
559 mapping = dma_map_page(&bp->pdev->dev, pool->page,
560 pool->offset, SGE_PAGE_SIZE, DMA_FROM_DEVICE);
561 if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
562 BNX2X_ERR("Can't map sge\n");
566 sw_buf->page = pool->page;
567 sw_buf->offset = pool->offset;
569 dma_unmap_addr_set(sw_buf, mapping, mapping);
571 sge->addr_hi = cpu_to_le32(U64_HI(mapping));
572 sge->addr_lo = cpu_to_le32(U64_LO(mapping));
574 pool->offset += SGE_PAGE_SIZE;
575 if (PAGE_SIZE - pool->offset >= SGE_PAGE_SIZE)
576 get_page(pool->page);
582 static int bnx2x_fill_frag_skb(struct bnx2x *bp, struct bnx2x_fastpath *fp,
583 struct bnx2x_agg_info *tpa_info,
586 struct eth_end_agg_rx_cqe *cqe,
589 struct sw_rx_page *rx_pg, old_rx_pg;
590 u32 i, frag_len, frag_size;
591 int err, j, frag_id = 0;
592 u16 len_on_bd = tpa_info->len_on_bd;
593 u16 full_page = 0, gro_size = 0;
595 frag_size = le16_to_cpu(cqe->pkt_len) - len_on_bd;
597 if (fp->mode == TPA_MODE_GRO) {
598 gro_size = tpa_info->gro_size;
599 full_page = tpa_info->full_page;
602 /* This is needed in order to enable forwarding support */
604 bnx2x_set_gro_params(skb, tpa_info->parsing_flags, len_on_bd,
605 le16_to_cpu(cqe->pkt_len),
606 le16_to_cpu(cqe->num_of_coalesced_segs));
608 #ifdef BNX2X_STOP_ON_ERROR
609 if (pages > min_t(u32, 8, MAX_SKB_FRAGS) * SGE_PAGES) {
610 BNX2X_ERR("SGL length is too long: %d. CQE index is %d\n",
612 BNX2X_ERR("cqe->pkt_len = %d\n", cqe->pkt_len);
618 /* Run through the SGL and compose the fragmented skb */
619 for (i = 0, j = 0; i < pages; i += PAGES_PER_SGE, j++) {
620 u16 sge_idx = RX_SGE(le16_to_cpu(cqe->sgl_or_raw_data.sgl[j]));
622 /* FW gives the indices of the SGE as if the ring is an array
623 (meaning that "next" element will consume 2 indices) */
624 if (fp->mode == TPA_MODE_GRO)
625 frag_len = min_t(u32, frag_size, (u32)full_page);
627 frag_len = min_t(u32, frag_size, (u32)SGE_PAGES);
629 rx_pg = &fp->rx_page_ring[sge_idx];
632 /* If we fail to allocate a substitute page, we simply stop
633 where we are and drop the whole packet */
634 err = bnx2x_alloc_rx_sge(bp, fp, sge_idx, GFP_ATOMIC);
636 bnx2x_fp_qstats(bp, fp)->rx_skb_alloc_failed++;
640 dma_unmap_page(&bp->pdev->dev,
641 dma_unmap_addr(&old_rx_pg, mapping),
642 SGE_PAGE_SIZE, DMA_FROM_DEVICE);
643 /* Add one frag and update the appropriate fields in the skb */
644 if (fp->mode == TPA_MODE_LRO)
645 skb_fill_page_desc(skb, j, old_rx_pg.page,
646 old_rx_pg.offset, frag_len);
650 for (rem = frag_len; rem > 0; rem -= gro_size) {
651 int len = rem > gro_size ? gro_size : rem;
652 skb_fill_page_desc(skb, frag_id++,
654 old_rx_pg.offset + offset,
657 get_page(old_rx_pg.page);
662 skb->data_len += frag_len;
663 skb->truesize += SGE_PAGES;
664 skb->len += frag_len;
666 frag_size -= frag_len;
672 static void bnx2x_frag_free(const struct bnx2x_fastpath *fp, void *data)
674 if (fp->rx_frag_size)
680 static void *bnx2x_frag_alloc(const struct bnx2x_fastpath *fp, gfp_t gfp_mask)
682 if (fp->rx_frag_size) {
683 /* GFP_KERNEL allocations are used only during initialization */
684 if (unlikely(gfpflags_allow_blocking(gfp_mask)))
685 return (void *)__get_free_page(gfp_mask);
687 return napi_alloc_frag(fp->rx_frag_size);
690 return kmalloc(fp->rx_buf_size + NET_SKB_PAD, gfp_mask);
694 static void bnx2x_gro_ip_csum(struct bnx2x *bp, struct sk_buff *skb)
696 const struct iphdr *iph = ip_hdr(skb);
699 skb_set_transport_header(skb, sizeof(struct iphdr));
702 th->check = ~tcp_v4_check(skb->len - skb_transport_offset(skb),
703 iph->saddr, iph->daddr, 0);
706 static void bnx2x_gro_ipv6_csum(struct bnx2x *bp, struct sk_buff *skb)
708 struct ipv6hdr *iph = ipv6_hdr(skb);
711 skb_set_transport_header(skb, sizeof(struct ipv6hdr));
714 th->check = ~tcp_v6_check(skb->len - skb_transport_offset(skb),
715 &iph->saddr, &iph->daddr, 0);
718 static void bnx2x_gro_csum(struct bnx2x *bp, struct sk_buff *skb,
719 void (*gro_func)(struct bnx2x*, struct sk_buff*))
721 skb_reset_network_header(skb);
723 tcp_gro_complete(skb);
727 static void bnx2x_gro_receive(struct bnx2x *bp, struct bnx2x_fastpath *fp,
731 if (skb_shinfo(skb)->gso_size) {
732 switch (be16_to_cpu(skb->protocol)) {
734 bnx2x_gro_csum(bp, skb, bnx2x_gro_ip_csum);
737 bnx2x_gro_csum(bp, skb, bnx2x_gro_ipv6_csum);
740 netdev_WARN_ONCE(bp->dev,
741 "Error: FW GRO supports only IPv4/IPv6, not 0x%04x\n",
742 be16_to_cpu(skb->protocol));
746 skb_record_rx_queue(skb, fp->rx_queue);
747 napi_gro_receive(&fp->napi, skb);
750 static void bnx2x_tpa_stop(struct bnx2x *bp, struct bnx2x_fastpath *fp,
751 struct bnx2x_agg_info *tpa_info,
753 struct eth_end_agg_rx_cqe *cqe,
756 struct sw_rx_bd *rx_buf = &tpa_info->first_buf;
757 u8 pad = tpa_info->placement_offset;
758 u16 len = tpa_info->len_on_bd;
759 struct sk_buff *skb = NULL;
760 u8 *new_data, *data = rx_buf->data;
761 u8 old_tpa_state = tpa_info->tpa_state;
763 tpa_info->tpa_state = BNX2X_TPA_STOP;
765 /* If we there was an error during the handling of the TPA_START -
766 * drop this aggregation.
768 if (old_tpa_state == BNX2X_TPA_ERROR)
771 /* Try to allocate the new data */
772 new_data = bnx2x_frag_alloc(fp, GFP_ATOMIC);
773 /* Unmap skb in the pool anyway, as we are going to change
774 pool entry status to BNX2X_TPA_STOP even if new skb allocation
776 dma_unmap_single(&bp->pdev->dev, dma_unmap_addr(rx_buf, mapping),
777 fp->rx_buf_size, DMA_FROM_DEVICE);
778 if (likely(new_data))
779 skb = build_skb(data, fp->rx_frag_size);
782 #ifdef BNX2X_STOP_ON_ERROR
783 if (pad + len > fp->rx_buf_size) {
784 BNX2X_ERR("skb_put is about to fail... pad %d len %d rx_buf_size %d\n",
785 pad, len, fp->rx_buf_size);
791 skb_reserve(skb, pad + NET_SKB_PAD);
793 skb_set_hash(skb, tpa_info->rxhash, tpa_info->rxhash_type);
795 skb->protocol = eth_type_trans(skb, bp->dev);
796 skb->ip_summed = CHECKSUM_UNNECESSARY;
798 if (!bnx2x_fill_frag_skb(bp, fp, tpa_info, pages,
799 skb, cqe, cqe_idx)) {
800 if (tpa_info->parsing_flags & PARSING_FLAGS_VLAN)
801 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), tpa_info->vlan_tag);
802 bnx2x_gro_receive(bp, fp, skb);
804 DP(NETIF_MSG_RX_STATUS,
805 "Failed to allocate new pages - dropping packet!\n");
806 dev_kfree_skb_any(skb);
809 /* put new data in bin */
810 rx_buf->data = new_data;
815 bnx2x_frag_free(fp, new_data);
817 /* drop the packet and keep the buffer in the bin */
818 DP(NETIF_MSG_RX_STATUS,
819 "Failed to allocate or map a new skb - dropping packet!\n");
820 bnx2x_fp_stats(bp, fp)->eth_q_stats.rx_skb_alloc_failed++;
823 static int bnx2x_alloc_rx_data(struct bnx2x *bp, struct bnx2x_fastpath *fp,
824 u16 index, gfp_t gfp_mask)
827 struct sw_rx_bd *rx_buf = &fp->rx_buf_ring[index];
828 struct eth_rx_bd *rx_bd = &fp->rx_desc_ring[index];
831 data = bnx2x_frag_alloc(fp, gfp_mask);
832 if (unlikely(data == NULL))
835 mapping = dma_map_single(&bp->pdev->dev, data + NET_SKB_PAD,
838 if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
839 bnx2x_frag_free(fp, data);
840 BNX2X_ERR("Can't map rx data\n");
845 dma_unmap_addr_set(rx_buf, mapping, mapping);
847 rx_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
848 rx_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
854 void bnx2x_csum_validate(struct sk_buff *skb, union eth_rx_cqe *cqe,
855 struct bnx2x_fastpath *fp,
856 struct bnx2x_eth_q_stats *qstats)
858 /* Do nothing if no L4 csum validation was done.
859 * We do not check whether IP csum was validated. For IPv4 we assume
860 * that if the card got as far as validating the L4 csum, it also
861 * validated the IP csum. IPv6 has no IP csum.
863 if (cqe->fast_path_cqe.status_flags &
864 ETH_FAST_PATH_RX_CQE_L4_XSUM_NO_VALIDATION_FLG)
867 /* If L4 validation was done, check if an error was found. */
869 if (cqe->fast_path_cqe.type_error_flags &
870 (ETH_FAST_PATH_RX_CQE_IP_BAD_XSUM_FLG |
871 ETH_FAST_PATH_RX_CQE_L4_BAD_XSUM_FLG))
872 qstats->hw_csum_err++;
874 skb->ip_summed = CHECKSUM_UNNECESSARY;
877 static int bnx2x_rx_int(struct bnx2x_fastpath *fp, int budget)
879 struct bnx2x *bp = fp->bp;
880 u16 bd_cons, bd_prod, bd_prod_fw, comp_ring_cons;
881 u16 sw_comp_cons, sw_comp_prod;
883 union eth_rx_cqe *cqe;
884 struct eth_fast_path_rx_cqe *cqe_fp;
886 #ifdef BNX2X_STOP_ON_ERROR
887 if (unlikely(bp->panic))
893 bd_cons = fp->rx_bd_cons;
894 bd_prod = fp->rx_bd_prod;
895 bd_prod_fw = bd_prod;
896 sw_comp_cons = fp->rx_comp_cons;
897 sw_comp_prod = fp->rx_comp_prod;
899 comp_ring_cons = RCQ_BD(sw_comp_cons);
900 cqe = &fp->rx_comp_ring[comp_ring_cons];
901 cqe_fp = &cqe->fast_path_cqe;
903 DP(NETIF_MSG_RX_STATUS,
904 "queue[%d]: sw_comp_cons %u\n", fp->index, sw_comp_cons);
906 while (BNX2X_IS_CQE_COMPLETED(cqe_fp)) {
907 struct sw_rx_bd *rx_buf = NULL;
910 enum eth_rx_cqe_type cqe_fp_type;
914 enum pkt_hash_types rxhash_type;
916 #ifdef BNX2X_STOP_ON_ERROR
917 if (unlikely(bp->panic))
921 bd_prod = RX_BD(bd_prod);
922 bd_cons = RX_BD(bd_cons);
924 /* A rmb() is required to ensure that the CQE is not read
925 * before it is written by the adapter DMA. PCI ordering
926 * rules will make sure the other fields are written before
927 * the marker at the end of struct eth_fast_path_rx_cqe
928 * but without rmb() a weakly ordered processor can process
929 * stale data. Without the barrier TPA state-machine might
930 * enter inconsistent state and kernel stack might be
931 * provided with incorrect packet description - these lead
932 * to various kernel crashed.
936 cqe_fp_flags = cqe_fp->type_error_flags;
937 cqe_fp_type = cqe_fp_flags & ETH_FAST_PATH_RX_CQE_TYPE;
939 DP(NETIF_MSG_RX_STATUS,
940 "CQE type %x err %x status %x queue %x vlan %x len %u\n",
941 CQE_TYPE(cqe_fp_flags),
942 cqe_fp_flags, cqe_fp->status_flags,
943 le32_to_cpu(cqe_fp->rss_hash_result),
944 le16_to_cpu(cqe_fp->vlan_tag),
945 le16_to_cpu(cqe_fp->pkt_len_or_gro_seg_len));
947 /* is this a slowpath msg? */
948 if (unlikely(CQE_TYPE_SLOW(cqe_fp_type))) {
949 bnx2x_sp_event(fp, cqe);
953 rx_buf = &fp->rx_buf_ring[bd_cons];
956 if (!CQE_TYPE_FAST(cqe_fp_type)) {
957 struct bnx2x_agg_info *tpa_info;
958 u16 frag_size, pages;
959 #ifdef BNX2X_STOP_ON_ERROR
961 if (fp->mode == TPA_MODE_DISABLED &&
962 (CQE_TYPE_START(cqe_fp_type) ||
963 CQE_TYPE_STOP(cqe_fp_type)))
964 BNX2X_ERR("START/STOP packet while TPA disabled, type %x\n",
965 CQE_TYPE(cqe_fp_type));
968 if (CQE_TYPE_START(cqe_fp_type)) {
969 u16 queue = cqe_fp->queue_index;
970 DP(NETIF_MSG_RX_STATUS,
971 "calling tpa_start on queue %d\n",
974 bnx2x_tpa_start(fp, queue,
980 queue = cqe->end_agg_cqe.queue_index;
981 tpa_info = &fp->tpa_info[queue];
982 DP(NETIF_MSG_RX_STATUS,
983 "calling tpa_stop on queue %d\n",
986 frag_size = le16_to_cpu(cqe->end_agg_cqe.pkt_len) -
989 if (fp->mode == TPA_MODE_GRO)
990 pages = (frag_size + tpa_info->full_page - 1) /
993 pages = SGE_PAGE_ALIGN(frag_size) >>
996 bnx2x_tpa_stop(bp, fp, tpa_info, pages,
997 &cqe->end_agg_cqe, comp_ring_cons);
998 #ifdef BNX2X_STOP_ON_ERROR
1003 bnx2x_update_sge_prod(fp, pages, &cqe->end_agg_cqe);
1007 len = le16_to_cpu(cqe_fp->pkt_len_or_gro_seg_len);
1008 pad = cqe_fp->placement_offset;
1009 dma_sync_single_for_cpu(&bp->pdev->dev,
1010 dma_unmap_addr(rx_buf, mapping),
1011 pad + RX_COPY_THRESH,
1014 prefetch(data + pad); /* speedup eth_type_trans() */
1015 /* is this an error packet? */
1016 if (unlikely(cqe_fp_flags & ETH_RX_ERROR_FALGS)) {
1017 DP(NETIF_MSG_RX_ERR | NETIF_MSG_RX_STATUS,
1018 "ERROR flags %x rx packet %u\n",
1019 cqe_fp_flags, sw_comp_cons);
1020 bnx2x_fp_qstats(bp, fp)->rx_err_discard_pkt++;
1024 /* Since we don't have a jumbo ring
1025 * copy small packets if mtu > 1500
1027 if ((bp->dev->mtu > ETH_MAX_PACKET_SIZE) &&
1028 (len <= RX_COPY_THRESH)) {
1029 skb = napi_alloc_skb(&fp->napi, len);
1031 DP(NETIF_MSG_RX_ERR | NETIF_MSG_RX_STATUS,
1032 "ERROR packet dropped because of alloc failure\n");
1033 bnx2x_fp_qstats(bp, fp)->rx_skb_alloc_failed++;
1036 memcpy(skb->data, data + pad, len);
1037 bnx2x_reuse_rx_data(fp, bd_cons, bd_prod);
1039 if (likely(bnx2x_alloc_rx_data(bp, fp, bd_prod,
1040 GFP_ATOMIC) == 0)) {
1041 dma_unmap_single(&bp->pdev->dev,
1042 dma_unmap_addr(rx_buf, mapping),
1045 skb = build_skb(data, fp->rx_frag_size);
1046 if (unlikely(!skb)) {
1047 bnx2x_frag_free(fp, data);
1048 bnx2x_fp_qstats(bp, fp)->
1049 rx_skb_alloc_failed++;
1052 skb_reserve(skb, pad);
1054 DP(NETIF_MSG_RX_ERR | NETIF_MSG_RX_STATUS,
1055 "ERROR packet dropped because of alloc failure\n");
1056 bnx2x_fp_qstats(bp, fp)->rx_skb_alloc_failed++;
1058 bnx2x_reuse_rx_data(fp, bd_cons, bd_prod);
1064 skb->protocol = eth_type_trans(skb, bp->dev);
1066 /* Set Toeplitz hash for a none-LRO skb */
1067 rxhash = bnx2x_get_rxhash(bp, cqe_fp, &rxhash_type);
1068 skb_set_hash(skb, rxhash, rxhash_type);
1070 skb_checksum_none_assert(skb);
1072 if (bp->dev->features & NETIF_F_RXCSUM)
1073 bnx2x_csum_validate(skb, cqe, fp,
1074 bnx2x_fp_qstats(bp, fp));
1076 skb_record_rx_queue(skb, fp->rx_queue);
1078 /* Check if this packet was timestamped */
1079 if (unlikely(cqe->fast_path_cqe.type_error_flags &
1080 (1 << ETH_FAST_PATH_RX_CQE_PTP_PKT_SHIFT)))
1081 bnx2x_set_rx_ts(bp, skb);
1083 if (le16_to_cpu(cqe_fp->pars_flags.flags) &
1085 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
1086 le16_to_cpu(cqe_fp->vlan_tag));
1088 napi_gro_receive(&fp->napi, skb);
1090 rx_buf->data = NULL;
1092 bd_cons = NEXT_RX_IDX(bd_cons);
1093 bd_prod = NEXT_RX_IDX(bd_prod);
1094 bd_prod_fw = NEXT_RX_IDX(bd_prod_fw);
1097 sw_comp_prod = NEXT_RCQ_IDX(sw_comp_prod);
1098 sw_comp_cons = NEXT_RCQ_IDX(sw_comp_cons);
1100 /* mark CQE as free */
1101 BNX2X_SEED_CQE(cqe_fp);
1103 if (rx_pkt == budget)
1106 comp_ring_cons = RCQ_BD(sw_comp_cons);
1107 cqe = &fp->rx_comp_ring[comp_ring_cons];
1108 cqe_fp = &cqe->fast_path_cqe;
1111 fp->rx_bd_cons = bd_cons;
1112 fp->rx_bd_prod = bd_prod_fw;
1113 fp->rx_comp_cons = sw_comp_cons;
1114 fp->rx_comp_prod = sw_comp_prod;
1116 /* Update producers */
1117 bnx2x_update_rx_prod(bp, fp, bd_prod_fw, sw_comp_prod,
1123 static irqreturn_t bnx2x_msix_fp_int(int irq, void *fp_cookie)
1125 struct bnx2x_fastpath *fp = fp_cookie;
1126 struct bnx2x *bp = fp->bp;
1130 "got an MSI-X interrupt on IDX:SB [fp %d fw_sd %d igusb %d]\n",
1131 fp->index, fp->fw_sb_id, fp->igu_sb_id);
1133 bnx2x_ack_sb(bp, fp->igu_sb_id, USTORM_ID, 0, IGU_INT_DISABLE, 0);
1135 #ifdef BNX2X_STOP_ON_ERROR
1136 if (unlikely(bp->panic))
1140 /* Handle Rx and Tx according to MSI-X vector */
1141 for_each_cos_in_tx_queue(fp, cos)
1142 prefetch(fp->txdata_ptr[cos]->tx_cons_sb);
1144 prefetch(&fp->sb_running_index[SM_RX_ID]);
1145 napi_schedule_irqoff(&bnx2x_fp(bp, fp->index, napi));
1150 /* HW Lock for shared dual port PHYs */
1151 void bnx2x_acquire_phy_lock(struct bnx2x *bp)
1153 mutex_lock(&bp->port.phy_mutex);
1155 bnx2x_acquire_hw_lock(bp, HW_LOCK_RESOURCE_MDIO);
1158 void bnx2x_release_phy_lock(struct bnx2x *bp)
1160 bnx2x_release_hw_lock(bp, HW_LOCK_RESOURCE_MDIO);
1162 mutex_unlock(&bp->port.phy_mutex);
1165 /* calculates MF speed according to current linespeed and MF configuration */
1166 u16 bnx2x_get_mf_speed(struct bnx2x *bp)
1168 u16 line_speed = bp->link_vars.line_speed;
1170 u16 maxCfg = bnx2x_extract_max_cfg(bp,
1171 bp->mf_config[BP_VN(bp)]);
1173 /* Calculate the current MAX line speed limit for the MF
1176 if (IS_MF_PERCENT_BW(bp))
1177 line_speed = (line_speed * maxCfg) / 100;
1178 else { /* SD mode */
1179 u16 vn_max_rate = maxCfg * 100;
1181 if (vn_max_rate < line_speed)
1182 line_speed = vn_max_rate;
1190 * bnx2x_fill_report_data - fill link report data to report
1192 * @bp: driver handle
1193 * @data: link state to update
1195 * It uses a none-atomic bit operations because is called under the mutex.
1197 static void bnx2x_fill_report_data(struct bnx2x *bp,
1198 struct bnx2x_link_report_data *data)
1200 memset(data, 0, sizeof(*data));
1203 /* Fill the report data: effective line speed */
1204 data->line_speed = bnx2x_get_mf_speed(bp);
1207 if (!bp->link_vars.link_up || (bp->flags & MF_FUNC_DIS))
1208 __set_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1209 &data->link_report_flags);
1211 if (!BNX2X_NUM_ETH_QUEUES(bp))
1212 __set_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1213 &data->link_report_flags);
1216 if (bp->link_vars.duplex == DUPLEX_FULL)
1217 __set_bit(BNX2X_LINK_REPORT_FD,
1218 &data->link_report_flags);
1220 /* Rx Flow Control is ON */
1221 if (bp->link_vars.flow_ctrl & BNX2X_FLOW_CTRL_RX)
1222 __set_bit(BNX2X_LINK_REPORT_RX_FC_ON,
1223 &data->link_report_flags);
1225 /* Tx Flow Control is ON */
1226 if (bp->link_vars.flow_ctrl & BNX2X_FLOW_CTRL_TX)
1227 __set_bit(BNX2X_LINK_REPORT_TX_FC_ON,
1228 &data->link_report_flags);
1230 *data = bp->vf_link_vars;
1235 * bnx2x_link_report - report link status to OS.
1237 * @bp: driver handle
1239 * Calls the __bnx2x_link_report() under the same locking scheme
1240 * as a link/PHY state managing code to ensure a consistent link
1244 void bnx2x_link_report(struct bnx2x *bp)
1246 bnx2x_acquire_phy_lock(bp);
1247 __bnx2x_link_report(bp);
1248 bnx2x_release_phy_lock(bp);
1252 * __bnx2x_link_report - report link status to OS.
1254 * @bp: driver handle
1256 * None atomic implementation.
1257 * Should be called under the phy_lock.
1259 void __bnx2x_link_report(struct bnx2x *bp)
1261 struct bnx2x_link_report_data cur_data;
1263 if (bp->force_link_down) {
1264 bp->link_vars.link_up = 0;
1269 if (IS_PF(bp) && !CHIP_IS_E1(bp))
1270 bnx2x_read_mf_cfg(bp);
1272 /* Read the current link report info */
1273 bnx2x_fill_report_data(bp, &cur_data);
1275 /* Don't report link down or exactly the same link status twice */
1276 if (!memcmp(&cur_data, &bp->last_reported_link, sizeof(cur_data)) ||
1277 (test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1278 &bp->last_reported_link.link_report_flags) &&
1279 test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1280 &cur_data.link_report_flags)))
1285 /* We are going to report a new link parameters now -
1286 * remember the current data for the next time.
1288 memcpy(&bp->last_reported_link, &cur_data, sizeof(cur_data));
1290 /* propagate status to VFs */
1292 bnx2x_iov_link_update(bp);
1294 if (test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1295 &cur_data.link_report_flags)) {
1296 netif_carrier_off(bp->dev);
1297 netdev_err(bp->dev, "NIC Link is Down\n");
1303 netif_carrier_on(bp->dev);
1305 if (test_and_clear_bit(BNX2X_LINK_REPORT_FD,
1306 &cur_data.link_report_flags))
1311 /* Handle the FC at the end so that only these flags would be
1312 * possibly set. This way we may easily check if there is no FC
1315 if (cur_data.link_report_flags) {
1316 if (test_bit(BNX2X_LINK_REPORT_RX_FC_ON,
1317 &cur_data.link_report_flags)) {
1318 if (test_bit(BNX2X_LINK_REPORT_TX_FC_ON,
1319 &cur_data.link_report_flags))
1320 flow = "ON - receive & transmit";
1322 flow = "ON - receive";
1324 flow = "ON - transmit";
1329 netdev_info(bp->dev, "NIC Link is Up, %d Mbps %s duplex, Flow control: %s\n",
1330 cur_data.line_speed, duplex, flow);
1334 static void bnx2x_set_next_page_sgl(struct bnx2x_fastpath *fp)
1338 for (i = 1; i <= NUM_RX_SGE_PAGES; i++) {
1339 struct eth_rx_sge *sge;
1341 sge = &fp->rx_sge_ring[RX_SGE_CNT * i - 2];
1343 cpu_to_le32(U64_HI(fp->rx_sge_mapping +
1344 BCM_PAGE_SIZE*(i % NUM_RX_SGE_PAGES)));
1347 cpu_to_le32(U64_LO(fp->rx_sge_mapping +
1348 BCM_PAGE_SIZE*(i % NUM_RX_SGE_PAGES)));
1352 static void bnx2x_free_tpa_pool(struct bnx2x *bp,
1353 struct bnx2x_fastpath *fp, int last)
1357 for (i = 0; i < last; i++) {
1358 struct bnx2x_agg_info *tpa_info = &fp->tpa_info[i];
1359 struct sw_rx_bd *first_buf = &tpa_info->first_buf;
1360 u8 *data = first_buf->data;
1363 DP(NETIF_MSG_IFDOWN, "tpa bin %d empty on free\n", i);
1366 if (tpa_info->tpa_state == BNX2X_TPA_START)
1367 dma_unmap_single(&bp->pdev->dev,
1368 dma_unmap_addr(first_buf, mapping),
1369 fp->rx_buf_size, DMA_FROM_DEVICE);
1370 bnx2x_frag_free(fp, data);
1371 first_buf->data = NULL;
1375 void bnx2x_init_rx_rings_cnic(struct bnx2x *bp)
1379 for_each_rx_queue_cnic(bp, j) {
1380 struct bnx2x_fastpath *fp = &bp->fp[j];
1384 /* Activate BD ring */
1386 * this will generate an interrupt (to the TSTORM)
1387 * must only be done after chip is initialized
1389 bnx2x_update_rx_prod(bp, fp, fp->rx_bd_prod, fp->rx_comp_prod,
1394 void bnx2x_init_rx_rings(struct bnx2x *bp)
1396 int func = BP_FUNC(bp);
1400 /* Allocate TPA resources */
1401 for_each_eth_queue(bp, j) {
1402 struct bnx2x_fastpath *fp = &bp->fp[j];
1405 "mtu %d rx_buf_size %d\n", bp->dev->mtu, fp->rx_buf_size);
1407 if (fp->mode != TPA_MODE_DISABLED) {
1408 /* Fill the per-aggregation pool */
1409 for (i = 0; i < MAX_AGG_QS(bp); i++) {
1410 struct bnx2x_agg_info *tpa_info =
1412 struct sw_rx_bd *first_buf =
1413 &tpa_info->first_buf;
1416 bnx2x_frag_alloc(fp, GFP_KERNEL);
1417 if (!first_buf->data) {
1418 BNX2X_ERR("Failed to allocate TPA skb pool for queue[%d] - disabling TPA on this queue!\n",
1420 bnx2x_free_tpa_pool(bp, fp, i);
1421 fp->mode = TPA_MODE_DISABLED;
1424 dma_unmap_addr_set(first_buf, mapping, 0);
1425 tpa_info->tpa_state = BNX2X_TPA_STOP;
1428 /* "next page" elements initialization */
1429 bnx2x_set_next_page_sgl(fp);
1431 /* set SGEs bit mask */
1432 bnx2x_init_sge_ring_bit_mask(fp);
1434 /* Allocate SGEs and initialize the ring elements */
1435 for (i = 0, ring_prod = 0;
1436 i < MAX_RX_SGE_CNT*NUM_RX_SGE_PAGES; i++) {
1438 if (bnx2x_alloc_rx_sge(bp, fp, ring_prod,
1440 BNX2X_ERR("was only able to allocate %d rx sges\n",
1442 BNX2X_ERR("disabling TPA for queue[%d]\n",
1444 /* Cleanup already allocated elements */
1445 bnx2x_free_rx_sge_range(bp, fp,
1447 bnx2x_free_tpa_pool(bp, fp,
1449 fp->mode = TPA_MODE_DISABLED;
1453 ring_prod = NEXT_SGE_IDX(ring_prod);
1456 fp->rx_sge_prod = ring_prod;
1460 for_each_eth_queue(bp, j) {
1461 struct bnx2x_fastpath *fp = &bp->fp[j];
1465 /* Activate BD ring */
1467 * this will generate an interrupt (to the TSTORM)
1468 * must only be done after chip is initialized
1470 bnx2x_update_rx_prod(bp, fp, fp->rx_bd_prod, fp->rx_comp_prod,
1476 if (CHIP_IS_E1(bp)) {
1477 REG_WR(bp, BAR_USTRORM_INTMEM +
1478 USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func),
1479 U64_LO(fp->rx_comp_mapping));
1480 REG_WR(bp, BAR_USTRORM_INTMEM +
1481 USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func) + 4,
1482 U64_HI(fp->rx_comp_mapping));
1487 static void bnx2x_free_tx_skbs_queue(struct bnx2x_fastpath *fp)
1490 struct bnx2x *bp = fp->bp;
1492 for_each_cos_in_tx_queue(fp, cos) {
1493 struct bnx2x_fp_txdata *txdata = fp->txdata_ptr[cos];
1494 unsigned pkts_compl = 0, bytes_compl = 0;
1496 u16 sw_prod = txdata->tx_pkt_prod;
1497 u16 sw_cons = txdata->tx_pkt_cons;
1499 while (sw_cons != sw_prod) {
1500 bnx2x_free_tx_pkt(bp, txdata, TX_BD(sw_cons),
1501 &pkts_compl, &bytes_compl);
1505 netdev_tx_reset_queue(
1506 netdev_get_tx_queue(bp->dev,
1507 txdata->txq_index));
1511 static void bnx2x_free_tx_skbs_cnic(struct bnx2x *bp)
1515 for_each_tx_queue_cnic(bp, i) {
1516 bnx2x_free_tx_skbs_queue(&bp->fp[i]);
1520 static void bnx2x_free_tx_skbs(struct bnx2x *bp)
1524 for_each_eth_queue(bp, i) {
1525 bnx2x_free_tx_skbs_queue(&bp->fp[i]);
1529 static void bnx2x_free_rx_bds(struct bnx2x_fastpath *fp)
1531 struct bnx2x *bp = fp->bp;
1534 /* ring wasn't allocated */
1535 if (fp->rx_buf_ring == NULL)
1538 for (i = 0; i < NUM_RX_BD; i++) {
1539 struct sw_rx_bd *rx_buf = &fp->rx_buf_ring[i];
1540 u8 *data = rx_buf->data;
1544 dma_unmap_single(&bp->pdev->dev,
1545 dma_unmap_addr(rx_buf, mapping),
1546 fp->rx_buf_size, DMA_FROM_DEVICE);
1548 rx_buf->data = NULL;
1549 bnx2x_frag_free(fp, data);
1553 static void bnx2x_free_rx_skbs_cnic(struct bnx2x *bp)
1557 for_each_rx_queue_cnic(bp, j) {
1558 bnx2x_free_rx_bds(&bp->fp[j]);
1562 static void bnx2x_free_rx_skbs(struct bnx2x *bp)
1566 for_each_eth_queue(bp, j) {
1567 struct bnx2x_fastpath *fp = &bp->fp[j];
1569 bnx2x_free_rx_bds(fp);
1571 if (fp->mode != TPA_MODE_DISABLED)
1572 bnx2x_free_tpa_pool(bp, fp, MAX_AGG_QS(bp));
1576 static void bnx2x_free_skbs_cnic(struct bnx2x *bp)
1578 bnx2x_free_tx_skbs_cnic(bp);
1579 bnx2x_free_rx_skbs_cnic(bp);
1582 void bnx2x_free_skbs(struct bnx2x *bp)
1584 bnx2x_free_tx_skbs(bp);
1585 bnx2x_free_rx_skbs(bp);
1588 void bnx2x_update_max_mf_config(struct bnx2x *bp, u32 value)
1590 /* load old values */
1591 u32 mf_cfg = bp->mf_config[BP_VN(bp)];
1593 if (value != bnx2x_extract_max_cfg(bp, mf_cfg)) {
1594 /* leave all but MAX value */
1595 mf_cfg &= ~FUNC_MF_CFG_MAX_BW_MASK;
1597 /* set new MAX value */
1598 mf_cfg |= (value << FUNC_MF_CFG_MAX_BW_SHIFT)
1599 & FUNC_MF_CFG_MAX_BW_MASK;
1601 bnx2x_fw_command(bp, DRV_MSG_CODE_SET_MF_BW, mf_cfg);
1606 * bnx2x_free_msix_irqs - free previously requested MSI-X IRQ vectors
1608 * @bp: driver handle
1609 * @nvecs: number of vectors to be released
1611 static void bnx2x_free_msix_irqs(struct bnx2x *bp, int nvecs)
1615 if (nvecs == offset)
1618 /* VFs don't have a default SB */
1620 free_irq(bp->msix_table[offset].vector, bp->dev);
1621 DP(NETIF_MSG_IFDOWN, "released sp irq (%d)\n",
1622 bp->msix_table[offset].vector);
1626 if (CNIC_SUPPORT(bp)) {
1627 if (nvecs == offset)
1632 for_each_eth_queue(bp, i) {
1633 if (nvecs == offset)
1635 DP(NETIF_MSG_IFDOWN, "about to release fp #%d->%d irq\n",
1636 i, bp->msix_table[offset].vector);
1638 free_irq(bp->msix_table[offset++].vector, &bp->fp[i]);
1642 void bnx2x_free_irq(struct bnx2x *bp)
1644 if (bp->flags & USING_MSIX_FLAG &&
1645 !(bp->flags & USING_SINGLE_MSIX_FLAG)) {
1646 int nvecs = BNX2X_NUM_ETH_QUEUES(bp) + CNIC_SUPPORT(bp);
1648 /* vfs don't have a default status block */
1652 bnx2x_free_msix_irqs(bp, nvecs);
1654 free_irq(bp->dev->irq, bp->dev);
1658 int bnx2x_enable_msix(struct bnx2x *bp)
1660 int msix_vec = 0, i, rc;
1662 /* VFs don't have a default status block */
1664 bp->msix_table[msix_vec].entry = msix_vec;
1665 BNX2X_DEV_INFO("msix_table[0].entry = %d (slowpath)\n",
1666 bp->msix_table[0].entry);
1670 /* Cnic requires an msix vector for itself */
1671 if (CNIC_SUPPORT(bp)) {
1672 bp->msix_table[msix_vec].entry = msix_vec;
1673 BNX2X_DEV_INFO("msix_table[%d].entry = %d (CNIC)\n",
1674 msix_vec, bp->msix_table[msix_vec].entry);
1678 /* We need separate vectors for ETH queues only (not FCoE) */
1679 for_each_eth_queue(bp, i) {
1680 bp->msix_table[msix_vec].entry = msix_vec;
1681 BNX2X_DEV_INFO("msix_table[%d].entry = %d (fastpath #%u)\n",
1682 msix_vec, msix_vec, i);
1686 DP(BNX2X_MSG_SP, "about to request enable msix with %d vectors\n",
1689 rc = pci_enable_msix_range(bp->pdev, &bp->msix_table[0],
1690 BNX2X_MIN_MSIX_VEC_CNT(bp), msix_vec);
1692 * reconfigure number of tx/rx queues according to available
1695 if (rc == -ENOSPC) {
1696 /* Get by with single vector */
1697 rc = pci_enable_msix_range(bp->pdev, &bp->msix_table[0], 1, 1);
1699 BNX2X_DEV_INFO("Single MSI-X is not attainable rc %d\n",
1704 BNX2X_DEV_INFO("Using single MSI-X vector\n");
1705 bp->flags |= USING_SINGLE_MSIX_FLAG;
1707 BNX2X_DEV_INFO("set number of queues to 1\n");
1708 bp->num_ethernet_queues = 1;
1709 bp->num_queues = bp->num_ethernet_queues + bp->num_cnic_queues;
1710 } else if (rc < 0) {
1711 BNX2X_DEV_INFO("MSI-X is not attainable rc %d\n", rc);
1713 } else if (rc < msix_vec) {
1714 /* how less vectors we will have? */
1715 int diff = msix_vec - rc;
1717 BNX2X_DEV_INFO("Trying to use less MSI-X vectors: %d\n", rc);
1720 * decrease number of queues by number of unallocated entries
1722 bp->num_ethernet_queues -= diff;
1723 bp->num_queues = bp->num_ethernet_queues + bp->num_cnic_queues;
1725 BNX2X_DEV_INFO("New queue configuration set: %d\n",
1729 bp->flags |= USING_MSIX_FLAG;
1734 /* fall to INTx if not enough memory */
1736 bp->flags |= DISABLE_MSI_FLAG;
1741 static int bnx2x_req_msix_irqs(struct bnx2x *bp)
1743 int i, rc, offset = 0;
1745 /* no default status block for vf */
1747 rc = request_irq(bp->msix_table[offset++].vector,
1748 bnx2x_msix_sp_int, 0,
1749 bp->dev->name, bp->dev);
1751 BNX2X_ERR("request sp irq failed\n");
1756 if (CNIC_SUPPORT(bp))
1759 for_each_eth_queue(bp, i) {
1760 struct bnx2x_fastpath *fp = &bp->fp[i];
1761 snprintf(fp->name, sizeof(fp->name), "%s-fp-%d",
1764 rc = request_irq(bp->msix_table[offset].vector,
1765 bnx2x_msix_fp_int, 0, fp->name, fp);
1767 BNX2X_ERR("request fp #%d irq (%d) failed rc %d\n", i,
1768 bp->msix_table[offset].vector, rc);
1769 bnx2x_free_msix_irqs(bp, offset);
1776 i = BNX2X_NUM_ETH_QUEUES(bp);
1778 offset = 1 + CNIC_SUPPORT(bp);
1779 netdev_info(bp->dev,
1780 "using MSI-X IRQs: sp %d fp[%d] %d ... fp[%d] %d\n",
1781 bp->msix_table[0].vector,
1782 0, bp->msix_table[offset].vector,
1783 i - 1, bp->msix_table[offset + i - 1].vector);
1785 offset = CNIC_SUPPORT(bp);
1786 netdev_info(bp->dev,
1787 "using MSI-X IRQs: fp[%d] %d ... fp[%d] %d\n",
1788 0, bp->msix_table[offset].vector,
1789 i - 1, bp->msix_table[offset + i - 1].vector);
1794 int bnx2x_enable_msi(struct bnx2x *bp)
1798 rc = pci_enable_msi(bp->pdev);
1800 BNX2X_DEV_INFO("MSI is not attainable\n");
1803 bp->flags |= USING_MSI_FLAG;
1808 static int bnx2x_req_irq(struct bnx2x *bp)
1810 unsigned long flags;
1813 if (bp->flags & (USING_MSI_FLAG | USING_MSIX_FLAG))
1816 flags = IRQF_SHARED;
1818 if (bp->flags & USING_MSIX_FLAG)
1819 irq = bp->msix_table[0].vector;
1821 irq = bp->pdev->irq;
1823 return request_irq(irq, bnx2x_interrupt, flags, bp->dev->name, bp->dev);
1826 static int bnx2x_setup_irqs(struct bnx2x *bp)
1829 if (bp->flags & USING_MSIX_FLAG &&
1830 !(bp->flags & USING_SINGLE_MSIX_FLAG)) {
1831 rc = bnx2x_req_msix_irqs(bp);
1835 rc = bnx2x_req_irq(bp);
1837 BNX2X_ERR("IRQ request failed rc %d, aborting\n", rc);
1840 if (bp->flags & USING_MSI_FLAG) {
1841 bp->dev->irq = bp->pdev->irq;
1842 netdev_info(bp->dev, "using MSI IRQ %d\n",
1845 if (bp->flags & USING_MSIX_FLAG) {
1846 bp->dev->irq = bp->msix_table[0].vector;
1847 netdev_info(bp->dev, "using MSIX IRQ %d\n",
1855 static void bnx2x_napi_enable_cnic(struct bnx2x *bp)
1859 for_each_rx_queue_cnic(bp, i) {
1860 napi_enable(&bnx2x_fp(bp, i, napi));
1864 static void bnx2x_napi_enable(struct bnx2x *bp)
1868 for_each_eth_queue(bp, i) {
1869 napi_enable(&bnx2x_fp(bp, i, napi));
1873 static void bnx2x_napi_disable_cnic(struct bnx2x *bp)
1877 for_each_rx_queue_cnic(bp, i) {
1878 napi_disable(&bnx2x_fp(bp, i, napi));
1882 static void bnx2x_napi_disable(struct bnx2x *bp)
1886 for_each_eth_queue(bp, i) {
1887 napi_disable(&bnx2x_fp(bp, i, napi));
1891 void bnx2x_netif_start(struct bnx2x *bp)
1893 if (netif_running(bp->dev)) {
1894 bnx2x_napi_enable(bp);
1895 if (CNIC_LOADED(bp))
1896 bnx2x_napi_enable_cnic(bp);
1897 bnx2x_int_enable(bp);
1898 if (bp->state == BNX2X_STATE_OPEN)
1899 netif_tx_wake_all_queues(bp->dev);
1903 void bnx2x_netif_stop(struct bnx2x *bp, int disable_hw)
1905 bnx2x_int_disable_sync(bp, disable_hw);
1906 bnx2x_napi_disable(bp);
1907 if (CNIC_LOADED(bp))
1908 bnx2x_napi_disable_cnic(bp);
1911 u16 bnx2x_select_queue(struct net_device *dev, struct sk_buff *skb,
1912 struct net_device *sb_dev)
1914 struct bnx2x *bp = netdev_priv(dev);
1916 if (CNIC_LOADED(bp) && !NO_FCOE(bp)) {
1917 struct ethhdr *hdr = (struct ethhdr *)skb->data;
1918 u16 ether_type = ntohs(hdr->h_proto);
1920 /* Skip VLAN tag if present */
1921 if (ether_type == ETH_P_8021Q) {
1922 struct vlan_ethhdr *vhdr =
1923 (struct vlan_ethhdr *)skb->data;
1925 ether_type = ntohs(vhdr->h_vlan_encapsulated_proto);
1928 /* If ethertype is FCoE or FIP - use FCoE ring */
1929 if ((ether_type == ETH_P_FCOE) || (ether_type == ETH_P_FIP))
1930 return bnx2x_fcoe_tx(bp, txq_index);
1933 /* select a non-FCoE queue */
1934 return netdev_pick_tx(dev, skb, NULL) %
1935 (BNX2X_NUM_ETH_QUEUES(bp) * bp->max_cos);
1938 void bnx2x_set_num_queues(struct bnx2x *bp)
1941 bp->num_ethernet_queues = bnx2x_calc_num_queues(bp);
1943 /* override in STORAGE SD modes */
1944 if (IS_MF_STORAGE_ONLY(bp))
1945 bp->num_ethernet_queues = 1;
1947 /* Add special queues */
1948 bp->num_cnic_queues = CNIC_SUPPORT(bp); /* For FCOE */
1949 bp->num_queues = bp->num_ethernet_queues + bp->num_cnic_queues;
1951 BNX2X_DEV_INFO("set number of queues to %d\n", bp->num_queues);
1955 * bnx2x_set_real_num_queues - configure netdev->real_num_[tx,rx]_queues
1957 * @bp: Driver handle
1959 * We currently support for at most 16 Tx queues for each CoS thus we will
1960 * allocate a multiple of 16 for ETH L2 rings according to the value of the
1963 * If there is an FCoE L2 queue the appropriate Tx queue will have the next
1964 * index after all ETH L2 indices.
1966 * If the actual number of Tx queues (for each CoS) is less than 16 then there
1967 * will be the holes at the end of each group of 16 ETh L2 indices (0..15,
1968 * 16..31,...) with indices that are not coupled with any real Tx queue.
1970 * The proper configuration of skb->queue_mapping is handled by
1971 * bnx2x_select_queue() and __skb_tx_hash().
1973 * bnx2x_setup_tc() takes care of the proper TC mappings so that __skb_tx_hash()
1974 * will return a proper Tx index if TC is enabled (netdev->num_tc > 0).
1976 static int bnx2x_set_real_num_queues(struct bnx2x *bp, int include_cnic)
1980 tx = BNX2X_NUM_ETH_QUEUES(bp) * bp->max_cos;
1981 rx = BNX2X_NUM_ETH_QUEUES(bp);
1983 /* account for fcoe queue */
1984 if (include_cnic && !NO_FCOE(bp)) {
1989 rc = netif_set_real_num_tx_queues(bp->dev, tx);
1991 BNX2X_ERR("Failed to set real number of Tx queues: %d\n", rc);
1994 rc = netif_set_real_num_rx_queues(bp->dev, rx);
1996 BNX2X_ERR("Failed to set real number of Rx queues: %d\n", rc);
2000 DP(NETIF_MSG_IFUP, "Setting real num queues to (tx, rx) (%d, %d)\n",
2006 static void bnx2x_set_rx_buf_size(struct bnx2x *bp)
2010 for_each_queue(bp, i) {
2011 struct bnx2x_fastpath *fp = &bp->fp[i];
2014 /* Always use a mini-jumbo MTU for the FCoE L2 ring */
2017 * Although there are no IP frames expected to arrive to
2018 * this ring we still want to add an
2019 * IP_HEADER_ALIGNMENT_PADDING to prevent a buffer
2022 mtu = BNX2X_FCOE_MINI_JUMBO_MTU;
2025 fp->rx_buf_size = BNX2X_FW_RX_ALIGN_START +
2026 IP_HEADER_ALIGNMENT_PADDING +
2029 BNX2X_FW_RX_ALIGN_END;
2030 fp->rx_buf_size = SKB_DATA_ALIGN(fp->rx_buf_size);
2031 /* Note : rx_buf_size doesn't take into account NET_SKB_PAD */
2032 if (fp->rx_buf_size + NET_SKB_PAD <= PAGE_SIZE)
2033 fp->rx_frag_size = fp->rx_buf_size + NET_SKB_PAD;
2035 fp->rx_frag_size = 0;
2039 static int bnx2x_init_rss(struct bnx2x *bp)
2042 u8 num_eth_queues = BNX2X_NUM_ETH_QUEUES(bp);
2044 /* Prepare the initial contents for the indirection table if RSS is
2047 for (i = 0; i < sizeof(bp->rss_conf_obj.ind_table); i++)
2048 bp->rss_conf_obj.ind_table[i] =
2050 ethtool_rxfh_indir_default(i, num_eth_queues);
2053 * For 57710 and 57711 SEARCHER configuration (rss_keys) is
2054 * per-port, so if explicit configuration is needed , do it only
2057 * For 57712 and newer on the other hand it's a per-function
2060 return bnx2x_config_rss_eth(bp, bp->port.pmf || !CHIP_IS_E1x(bp));
2063 int bnx2x_rss(struct bnx2x *bp, struct bnx2x_rss_config_obj *rss_obj,
2064 bool config_hash, bool enable)
2066 struct bnx2x_config_rss_params params = {NULL};
2068 /* Although RSS is meaningless when there is a single HW queue we
2069 * still need it enabled in order to have HW Rx hash generated.
2071 * if (!is_eth_multi(bp))
2072 * bp->multi_mode = ETH_RSS_MODE_DISABLED;
2075 params.rss_obj = rss_obj;
2077 __set_bit(RAMROD_COMP_WAIT, ¶ms.ramrod_flags);
2080 __set_bit(BNX2X_RSS_MODE_REGULAR, ¶ms.rss_flags);
2082 /* RSS configuration */
2083 __set_bit(BNX2X_RSS_IPV4, ¶ms.rss_flags);
2084 __set_bit(BNX2X_RSS_IPV4_TCP, ¶ms.rss_flags);
2085 __set_bit(BNX2X_RSS_IPV6, ¶ms.rss_flags);
2086 __set_bit(BNX2X_RSS_IPV6_TCP, ¶ms.rss_flags);
2087 if (rss_obj->udp_rss_v4)
2088 __set_bit(BNX2X_RSS_IPV4_UDP, ¶ms.rss_flags);
2089 if (rss_obj->udp_rss_v6)
2090 __set_bit(BNX2X_RSS_IPV6_UDP, ¶ms.rss_flags);
2092 if (!CHIP_IS_E1x(bp)) {
2093 /* valid only for TUNN_MODE_VXLAN tunnel mode */
2094 __set_bit(BNX2X_RSS_IPV4_VXLAN, ¶ms.rss_flags);
2095 __set_bit(BNX2X_RSS_IPV6_VXLAN, ¶ms.rss_flags);
2097 /* valid only for TUNN_MODE_GRE tunnel mode */
2098 __set_bit(BNX2X_RSS_TUNN_INNER_HDRS, ¶ms.rss_flags);
2101 __set_bit(BNX2X_RSS_MODE_DISABLED, ¶ms.rss_flags);
2105 params.rss_result_mask = MULTI_MASK;
2107 memcpy(params.ind_table, rss_obj->ind_table, sizeof(params.ind_table));
2111 netdev_rss_key_fill(params.rss_key, T_ETH_RSS_KEY * 4);
2112 __set_bit(BNX2X_RSS_SET_SRCH, ¶ms.rss_flags);
2116 return bnx2x_config_rss(bp, ¶ms);
2118 return bnx2x_vfpf_config_rss(bp, ¶ms);
2121 static int bnx2x_init_hw(struct bnx2x *bp, u32 load_code)
2123 struct bnx2x_func_state_params func_params = {NULL};
2125 /* Prepare parameters for function state transitions */
2126 __set_bit(RAMROD_COMP_WAIT, &func_params.ramrod_flags);
2128 func_params.f_obj = &bp->func_obj;
2129 func_params.cmd = BNX2X_F_CMD_HW_INIT;
2131 func_params.params.hw_init.load_phase = load_code;
2133 return bnx2x_func_state_change(bp, &func_params);
2137 * Cleans the object that have internal lists without sending
2138 * ramrods. Should be run when interrupts are disabled.
2140 void bnx2x_squeeze_objects(struct bnx2x *bp)
2143 unsigned long ramrod_flags = 0, vlan_mac_flags = 0;
2144 struct bnx2x_mcast_ramrod_params rparam = {NULL};
2145 struct bnx2x_vlan_mac_obj *mac_obj = &bp->sp_objs->mac_obj;
2147 /***************** Cleanup MACs' object first *************************/
2149 /* Wait for completion of requested */
2150 __set_bit(RAMROD_COMP_WAIT, &ramrod_flags);
2151 /* Perform a dry cleanup */
2152 __set_bit(RAMROD_DRV_CLR_ONLY, &ramrod_flags);
2154 /* Clean ETH primary MAC */
2155 __set_bit(BNX2X_ETH_MAC, &vlan_mac_flags);
2156 rc = mac_obj->delete_all(bp, &bp->sp_objs->mac_obj, &vlan_mac_flags,
2159 BNX2X_ERR("Failed to clean ETH MACs: %d\n", rc);
2161 /* Cleanup UC list */
2163 __set_bit(BNX2X_UC_LIST_MAC, &vlan_mac_flags);
2164 rc = mac_obj->delete_all(bp, mac_obj, &vlan_mac_flags,
2167 BNX2X_ERR("Failed to clean UC list MACs: %d\n", rc);
2169 /***************** Now clean mcast object *****************************/
2170 rparam.mcast_obj = &bp->mcast_obj;
2171 __set_bit(RAMROD_DRV_CLR_ONLY, &rparam.ramrod_flags);
2173 /* Add a DEL command... - Since we're doing a driver cleanup only,
2174 * we take a lock surrounding both the initial send and the CONTs,
2175 * as we don't want a true completion to disrupt us in the middle.
2177 netif_addr_lock_bh(bp->dev);
2178 rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_DEL);
2180 BNX2X_ERR("Failed to add a new DEL command to a multi-cast object: %d\n",
2183 /* ...and wait until all pending commands are cleared */
2184 rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_CONT);
2187 BNX2X_ERR("Failed to clean multi-cast object: %d\n",
2189 netif_addr_unlock_bh(bp->dev);
2193 rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_CONT);
2195 netif_addr_unlock_bh(bp->dev);
2198 #ifndef BNX2X_STOP_ON_ERROR
2199 #define LOAD_ERROR_EXIT(bp, label) \
2201 (bp)->state = BNX2X_STATE_ERROR; \
2205 #define LOAD_ERROR_EXIT_CNIC(bp, label) \
2207 bp->cnic_loaded = false; \
2210 #else /*BNX2X_STOP_ON_ERROR*/
2211 #define LOAD_ERROR_EXIT(bp, label) \
2213 (bp)->state = BNX2X_STATE_ERROR; \
2217 #define LOAD_ERROR_EXIT_CNIC(bp, label) \
2219 bp->cnic_loaded = false; \
2223 #endif /*BNX2X_STOP_ON_ERROR*/
2225 static void bnx2x_free_fw_stats_mem(struct bnx2x *bp)
2227 BNX2X_PCI_FREE(bp->fw_stats, bp->fw_stats_mapping,
2228 bp->fw_stats_data_sz + bp->fw_stats_req_sz);
2232 static int bnx2x_alloc_fw_stats_mem(struct bnx2x *bp)
2234 int num_groups, vf_headroom = 0;
2235 int is_fcoe_stats = NO_FCOE(bp) ? 0 : 1;
2237 /* number of queues for statistics is number of eth queues + FCoE */
2238 u8 num_queue_stats = BNX2X_NUM_ETH_QUEUES(bp) + is_fcoe_stats;
2240 /* Total number of FW statistics requests =
2241 * 1 for port stats + 1 for PF stats + potential 2 for FCoE (fcoe proper
2242 * and fcoe l2 queue) stats + num of queues (which includes another 1
2243 * for fcoe l2 queue if applicable)
2245 bp->fw_stats_num = 2 + is_fcoe_stats + num_queue_stats;
2247 /* vf stats appear in the request list, but their data is allocated by
2248 * the VFs themselves. We don't include them in the bp->fw_stats_num as
2249 * it is used to determine where to place the vf stats queries in the
2253 vf_headroom = bnx2x_vf_headroom(bp);
2255 /* Request is built from stats_query_header and an array of
2256 * stats_query_cmd_group each of which contains
2257 * STATS_QUERY_CMD_COUNT rules. The real number or requests is
2258 * configured in the stats_query_header.
2261 (((bp->fw_stats_num + vf_headroom) / STATS_QUERY_CMD_COUNT) +
2262 (((bp->fw_stats_num + vf_headroom) % STATS_QUERY_CMD_COUNT) ?
2265 DP(BNX2X_MSG_SP, "stats fw_stats_num %d, vf headroom %d, num_groups %d\n",
2266 bp->fw_stats_num, vf_headroom, num_groups);
2267 bp->fw_stats_req_sz = sizeof(struct stats_query_header) +
2268 num_groups * sizeof(struct stats_query_cmd_group);
2270 /* Data for statistics requests + stats_counter
2271 * stats_counter holds per-STORM counters that are incremented
2272 * when STORM has finished with the current request.
2273 * memory for FCoE offloaded statistics are counted anyway,
2274 * even if they will not be sent.
2275 * VF stats are not accounted for here as the data of VF stats is stored
2276 * in memory allocated by the VF, not here.
2278 bp->fw_stats_data_sz = sizeof(struct per_port_stats) +
2279 sizeof(struct per_pf_stats) +
2280 sizeof(struct fcoe_statistics_params) +
2281 sizeof(struct per_queue_stats) * num_queue_stats +
2282 sizeof(struct stats_counter);
2284 bp->fw_stats = BNX2X_PCI_ALLOC(&bp->fw_stats_mapping,
2285 bp->fw_stats_data_sz + bp->fw_stats_req_sz);
2290 bp->fw_stats_req = (struct bnx2x_fw_stats_req *)bp->fw_stats;
2291 bp->fw_stats_req_mapping = bp->fw_stats_mapping;
2292 bp->fw_stats_data = (struct bnx2x_fw_stats_data *)
2293 ((u8 *)bp->fw_stats + bp->fw_stats_req_sz);
2294 bp->fw_stats_data_mapping = bp->fw_stats_mapping +
2295 bp->fw_stats_req_sz;
2297 DP(BNX2X_MSG_SP, "statistics request base address set to %x %x\n",
2298 U64_HI(bp->fw_stats_req_mapping),
2299 U64_LO(bp->fw_stats_req_mapping));
2300 DP(BNX2X_MSG_SP, "statistics data base address set to %x %x\n",
2301 U64_HI(bp->fw_stats_data_mapping),
2302 U64_LO(bp->fw_stats_data_mapping));
2306 bnx2x_free_fw_stats_mem(bp);
2307 BNX2X_ERR("Can't allocate FW stats memory\n");
2311 /* send load request to mcp and analyze response */
2312 static int bnx2x_nic_load_request(struct bnx2x *bp, u32 *load_code)
2318 (SHMEM_RD(bp, func_mb[BP_FW_MB_IDX(bp)].drv_mb_header) &
2319 DRV_MSG_SEQ_NUMBER_MASK);
2320 BNX2X_DEV_INFO("fw_seq 0x%08x\n", bp->fw_seq);
2322 /* Get current FW pulse sequence */
2323 bp->fw_drv_pulse_wr_seq =
2324 (SHMEM_RD(bp, func_mb[BP_FW_MB_IDX(bp)].drv_pulse_mb) &
2325 DRV_PULSE_SEQ_MASK);
2326 BNX2X_DEV_INFO("drv_pulse 0x%x\n", bp->fw_drv_pulse_wr_seq);
2328 param = DRV_MSG_CODE_LOAD_REQ_WITH_LFA;
2330 if (IS_MF_SD(bp) && bnx2x_port_after_undi(bp))
2331 param |= DRV_MSG_CODE_LOAD_REQ_FORCE_LFA;
2334 (*load_code) = bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_REQ, param);
2336 /* if mcp fails to respond we must abort */
2337 if (!(*load_code)) {
2338 BNX2X_ERR("MCP response failure, aborting\n");
2342 /* If mcp refused (e.g. other port is in diagnostic mode) we
2345 if ((*load_code) == FW_MSG_CODE_DRV_LOAD_REFUSED) {
2346 BNX2X_ERR("MCP refused load request, aborting\n");
2352 /* check whether another PF has already loaded FW to chip. In
2353 * virtualized environments a pf from another VM may have already
2354 * initialized the device including loading FW
2356 int bnx2x_compare_fw_ver(struct bnx2x *bp, u32 load_code, bool print_err)
2358 /* is another pf loaded on this engine? */
2359 if (load_code != FW_MSG_CODE_DRV_LOAD_COMMON_CHIP &&
2360 load_code != FW_MSG_CODE_DRV_LOAD_COMMON) {
2361 /* build my FW version dword */
2362 u32 my_fw = (BCM_5710_FW_MAJOR_VERSION) +
2363 (BCM_5710_FW_MINOR_VERSION << 8) +
2364 (BCM_5710_FW_REVISION_VERSION << 16) +
2365 (BCM_5710_FW_ENGINEERING_VERSION << 24);
2367 /* read loaded FW from chip */
2368 u32 loaded_fw = REG_RD(bp, XSEM_REG_PRAM);
2370 DP(BNX2X_MSG_SP, "loaded fw %x, my fw %x\n",
2373 /* abort nic load if version mismatch */
2374 if (my_fw != loaded_fw) {
2376 BNX2X_ERR("bnx2x with FW %x was already loaded which mismatches my %x FW. Aborting\n",
2379 BNX2X_DEV_INFO("bnx2x with FW %x was already loaded which mismatches my %x FW, possibly due to MF UNDI\n",
2387 /* returns the "mcp load_code" according to global load_count array */
2388 static int bnx2x_nic_load_no_mcp(struct bnx2x *bp, int port)
2390 int path = BP_PATH(bp);
2392 DP(NETIF_MSG_IFUP, "NO MCP - load counts[%d] %d, %d, %d\n",
2393 path, bnx2x_load_count[path][0], bnx2x_load_count[path][1],
2394 bnx2x_load_count[path][2]);
2395 bnx2x_load_count[path][0]++;
2396 bnx2x_load_count[path][1 + port]++;
2397 DP(NETIF_MSG_IFUP, "NO MCP - new load counts[%d] %d, %d, %d\n",
2398 path, bnx2x_load_count[path][0], bnx2x_load_count[path][1],
2399 bnx2x_load_count[path][2]);
2400 if (bnx2x_load_count[path][0] == 1)
2401 return FW_MSG_CODE_DRV_LOAD_COMMON;
2402 else if (bnx2x_load_count[path][1 + port] == 1)
2403 return FW_MSG_CODE_DRV_LOAD_PORT;
2405 return FW_MSG_CODE_DRV_LOAD_FUNCTION;
2408 /* mark PMF if applicable */
2409 static void bnx2x_nic_load_pmf(struct bnx2x *bp, u32 load_code)
2411 if ((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
2412 (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP) ||
2413 (load_code == FW_MSG_CODE_DRV_LOAD_PORT)) {
2415 /* We need the barrier to ensure the ordering between the
2416 * writing to bp->port.pmf here and reading it from the
2417 * bnx2x_periodic_task().
2424 DP(NETIF_MSG_LINK, "pmf %d\n", bp->port.pmf);
2427 static void bnx2x_nic_load_afex_dcc(struct bnx2x *bp, int load_code)
2429 if (((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
2430 (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP)) &&
2431 (bp->common.shmem2_base)) {
2432 if (SHMEM2_HAS(bp, dcc_support))
2433 SHMEM2_WR(bp, dcc_support,
2434 (SHMEM_DCC_SUPPORT_DISABLE_ENABLE_PF_TLV |
2435 SHMEM_DCC_SUPPORT_BANDWIDTH_ALLOCATION_TLV));
2436 if (SHMEM2_HAS(bp, afex_driver_support))
2437 SHMEM2_WR(bp, afex_driver_support,
2438 SHMEM_AFEX_SUPPORTED_VERSION_ONE);
2441 /* Set AFEX default VLAN tag to an invalid value */
2442 bp->afex_def_vlan_tag = -1;
2446 * bnx2x_bz_fp - zero content of the fastpath structure.
2448 * @bp: driver handle
2449 * @index: fastpath index to be zeroed
2451 * Makes sure the contents of the bp->fp[index].napi is kept
2454 static void bnx2x_bz_fp(struct bnx2x *bp, int index)
2456 struct bnx2x_fastpath *fp = &bp->fp[index];
2458 struct napi_struct orig_napi = fp->napi;
2459 struct bnx2x_agg_info *orig_tpa_info = fp->tpa_info;
2461 /* bzero bnx2x_fastpath contents */
2463 memset(fp->tpa_info, 0, ETH_MAX_AGGREGATION_QUEUES_E1H_E2 *
2464 sizeof(struct bnx2x_agg_info));
2465 memset(fp, 0, sizeof(*fp));
2467 /* Restore the NAPI object as it has been already initialized */
2468 fp->napi = orig_napi;
2469 fp->tpa_info = orig_tpa_info;
2473 fp->max_cos = bp->max_cos;
2475 /* Special queues support only one CoS */
2478 /* Init txdata pointers */
2480 fp->txdata_ptr[0] = &bp->bnx2x_txq[FCOE_TXQ_IDX(bp)];
2482 for_each_cos_in_tx_queue(fp, cos)
2483 fp->txdata_ptr[cos] = &bp->bnx2x_txq[cos *
2484 BNX2X_NUM_ETH_QUEUES(bp) + index];
2486 /* set the tpa flag for each queue. The tpa flag determines the queue
2487 * minimal size so it must be set prior to queue memory allocation
2489 if (bp->dev->features & NETIF_F_LRO)
2490 fp->mode = TPA_MODE_LRO;
2491 else if (bp->dev->features & NETIF_F_GRO_HW)
2492 fp->mode = TPA_MODE_GRO;
2494 fp->mode = TPA_MODE_DISABLED;
2496 /* We don't want TPA if it's disabled in bp
2497 * or if this is an FCoE L2 ring.
2499 if (bp->disable_tpa || IS_FCOE_FP(fp))
2500 fp->mode = TPA_MODE_DISABLED;
2503 void bnx2x_set_os_driver_state(struct bnx2x *bp, u32 state)
2507 if (!IS_MF_BD(bp) || !SHMEM2_HAS(bp, os_driver_state) || IS_VF(bp))
2510 cur = SHMEM2_RD(bp, os_driver_state[BP_FW_MB_IDX(bp)]);
2511 DP(NETIF_MSG_IFUP, "Driver state %08x-->%08x\n",
2514 SHMEM2_WR(bp, os_driver_state[BP_FW_MB_IDX(bp)], state);
2517 int bnx2x_load_cnic(struct bnx2x *bp)
2519 int i, rc, port = BP_PORT(bp);
2521 DP(NETIF_MSG_IFUP, "Starting CNIC-related load\n");
2523 mutex_init(&bp->cnic_mutex);
2526 rc = bnx2x_alloc_mem_cnic(bp);
2528 BNX2X_ERR("Unable to allocate bp memory for cnic\n");
2529 LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic0);
2533 rc = bnx2x_alloc_fp_mem_cnic(bp);
2535 BNX2X_ERR("Unable to allocate memory for cnic fps\n");
2536 LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic0);
2539 /* Update the number of queues with the cnic queues */
2540 rc = bnx2x_set_real_num_queues(bp, 1);
2542 BNX2X_ERR("Unable to set real_num_queues including cnic\n");
2543 LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic0);
2546 /* Add all CNIC NAPI objects */
2547 bnx2x_add_all_napi_cnic(bp);
2548 DP(NETIF_MSG_IFUP, "cnic napi added\n");
2549 bnx2x_napi_enable_cnic(bp);
2551 rc = bnx2x_init_hw_func_cnic(bp);
2553 LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic1);
2555 bnx2x_nic_init_cnic(bp);
2558 /* Enable Timer scan */
2559 REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + port*4, 1);
2561 /* setup cnic queues */
2562 for_each_cnic_queue(bp, i) {
2563 rc = bnx2x_setup_queue(bp, &bp->fp[i], 0);
2565 BNX2X_ERR("Queue setup failed\n");
2566 LOAD_ERROR_EXIT(bp, load_error_cnic2);
2571 /* Initialize Rx filter. */
2572 bnx2x_set_rx_mode_inner(bp);
2574 /* re-read iscsi info */
2575 bnx2x_get_iscsi_info(bp);
2576 bnx2x_setup_cnic_irq_info(bp);
2577 bnx2x_setup_cnic_info(bp);
2578 bp->cnic_loaded = true;
2579 if (bp->state == BNX2X_STATE_OPEN)
2580 bnx2x_cnic_notify(bp, CNIC_CTL_START_CMD);
2582 DP(NETIF_MSG_IFUP, "Ending successfully CNIC-related load\n");
2586 #ifndef BNX2X_STOP_ON_ERROR
2588 /* Disable Timer scan */
2589 REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + port*4, 0);
2592 bnx2x_napi_disable_cnic(bp);
2593 /* Update the number of queues without the cnic queues */
2594 if (bnx2x_set_real_num_queues(bp, 0))
2595 BNX2X_ERR("Unable to set real_num_queues not including cnic\n");
2597 BNX2X_ERR("CNIC-related load failed\n");
2598 bnx2x_free_fp_mem_cnic(bp);
2599 bnx2x_free_mem_cnic(bp);
2601 #endif /* ! BNX2X_STOP_ON_ERROR */
2604 /* must be called with rtnl_lock */
2605 int bnx2x_nic_load(struct bnx2x *bp, int load_mode)
2607 int port = BP_PORT(bp);
2608 int i, rc = 0, load_code = 0;
2610 DP(NETIF_MSG_IFUP, "Starting NIC load\n");
2612 "CNIC is %s\n", CNIC_ENABLED(bp) ? "enabled" : "disabled");
2614 #ifdef BNX2X_STOP_ON_ERROR
2615 if (unlikely(bp->panic)) {
2616 BNX2X_ERR("Can't load NIC when there is panic\n");
2621 bp->state = BNX2X_STATE_OPENING_WAIT4_LOAD;
2623 /* zero the structure w/o any lock, before SP handler is initialized */
2624 memset(&bp->last_reported_link, 0, sizeof(bp->last_reported_link));
2625 __set_bit(BNX2X_LINK_REPORT_LINK_DOWN,
2626 &bp->last_reported_link.link_report_flags);
2629 /* must be called before memory allocation and HW init */
2630 bnx2x_ilt_set_info(bp);
2633 * Zero fastpath structures preserving invariants like napi, which are
2634 * allocated only once, fp index, max_cos, bp pointer.
2635 * Also set fp->mode and txdata_ptr.
2637 DP(NETIF_MSG_IFUP, "num queues: %d", bp->num_queues);
2638 for_each_queue(bp, i)
2640 memset(bp->bnx2x_txq, 0, (BNX2X_MAX_RSS_COUNT(bp) * BNX2X_MULTI_TX_COS +
2641 bp->num_cnic_queues) *
2642 sizeof(struct bnx2x_fp_txdata));
2644 bp->fcoe_init = false;
2646 /* Set the receive queues buffer size */
2647 bnx2x_set_rx_buf_size(bp);
2650 rc = bnx2x_alloc_mem(bp);
2652 BNX2X_ERR("Unable to allocate bp memory\n");
2657 /* need to be done after alloc mem, since it's self adjusting to amount
2658 * of memory available for RSS queues
2660 rc = bnx2x_alloc_fp_mem(bp);
2662 BNX2X_ERR("Unable to allocate memory for fps\n");
2663 LOAD_ERROR_EXIT(bp, load_error0);
2666 /* Allocated memory for FW statistics */
2667 if (bnx2x_alloc_fw_stats_mem(bp))
2668 LOAD_ERROR_EXIT(bp, load_error0);
2670 /* request pf to initialize status blocks */
2672 rc = bnx2x_vfpf_init(bp);
2674 LOAD_ERROR_EXIT(bp, load_error0);
2677 /* As long as bnx2x_alloc_mem() may possibly update
2678 * bp->num_queues, bnx2x_set_real_num_queues() should always
2679 * come after it. At this stage cnic queues are not counted.
2681 rc = bnx2x_set_real_num_queues(bp, 0);
2683 BNX2X_ERR("Unable to set real_num_queues\n");
2684 LOAD_ERROR_EXIT(bp, load_error0);
2687 /* configure multi cos mappings in kernel.
2688 * this configuration may be overridden by a multi class queue
2689 * discipline or by a dcbx negotiation result.
2691 bnx2x_setup_tc(bp->dev, bp->max_cos);
2693 /* Add all NAPI objects */
2694 bnx2x_add_all_napi(bp);
2695 DP(NETIF_MSG_IFUP, "napi added\n");
2696 bnx2x_napi_enable(bp);
2699 /* set pf load just before approaching the MCP */
2700 bnx2x_set_pf_load(bp);
2702 /* if mcp exists send load request and analyze response */
2703 if (!BP_NOMCP(bp)) {
2704 /* attempt to load pf */
2705 rc = bnx2x_nic_load_request(bp, &load_code);
2707 LOAD_ERROR_EXIT(bp, load_error1);
2709 /* what did mcp say? */
2710 rc = bnx2x_compare_fw_ver(bp, load_code, true);
2712 bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2713 LOAD_ERROR_EXIT(bp, load_error2);
2716 load_code = bnx2x_nic_load_no_mcp(bp, port);
2719 /* mark pmf if applicable */
2720 bnx2x_nic_load_pmf(bp, load_code);
2722 /* Init Function state controlling object */
2723 bnx2x__init_func_obj(bp);
2726 rc = bnx2x_init_hw(bp, load_code);
2728 BNX2X_ERR("HW init failed, aborting\n");
2729 bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2730 LOAD_ERROR_EXIT(bp, load_error2);
2734 bnx2x_pre_irq_nic_init(bp);
2736 /* Connect to IRQs */
2737 rc = bnx2x_setup_irqs(bp);
2739 BNX2X_ERR("setup irqs failed\n");
2741 bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2742 LOAD_ERROR_EXIT(bp, load_error2);
2745 /* Init per-function objects */
2747 /* Setup NIC internals and enable interrupts */
2748 bnx2x_post_irq_nic_init(bp, load_code);
2750 bnx2x_init_bp_objs(bp);
2751 bnx2x_iov_nic_init(bp);
2753 /* Set AFEX default VLAN tag to an invalid value */
2754 bp->afex_def_vlan_tag = -1;
2755 bnx2x_nic_load_afex_dcc(bp, load_code);
2756 bp->state = BNX2X_STATE_OPENING_WAIT4_PORT;
2757 rc = bnx2x_func_start(bp);
2759 BNX2X_ERR("Function start failed!\n");
2760 bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2762 LOAD_ERROR_EXIT(bp, load_error3);
2765 /* Send LOAD_DONE command to MCP */
2766 if (!BP_NOMCP(bp)) {
2767 load_code = bnx2x_fw_command(bp,
2768 DRV_MSG_CODE_LOAD_DONE, 0);
2770 BNX2X_ERR("MCP response failure, aborting\n");
2772 LOAD_ERROR_EXIT(bp, load_error3);
2776 /* initialize FW coalescing state machines in RAM */
2777 bnx2x_update_coalesce(bp);
2780 /* setup the leading queue */
2781 rc = bnx2x_setup_leading(bp);
2783 BNX2X_ERR("Setup leading failed!\n");
2784 LOAD_ERROR_EXIT(bp, load_error3);
2787 /* set up the rest of the queues */
2788 for_each_nondefault_eth_queue(bp, i) {
2790 rc = bnx2x_setup_queue(bp, &bp->fp[i], false);
2792 rc = bnx2x_vfpf_setup_q(bp, &bp->fp[i], false);
2794 BNX2X_ERR("Queue %d setup failed\n", i);
2795 LOAD_ERROR_EXIT(bp, load_error3);
2800 rc = bnx2x_init_rss(bp);
2802 BNX2X_ERR("PF RSS init failed\n");
2803 LOAD_ERROR_EXIT(bp, load_error3);
2806 /* Now when Clients are configured we are ready to work */
2807 bp->state = BNX2X_STATE_OPEN;
2809 /* Configure a ucast MAC */
2811 rc = bnx2x_set_eth_mac(bp, true);
2813 rc = bnx2x_vfpf_config_mac(bp, bp->dev->dev_addr, bp->fp->index,
2816 BNX2X_ERR("Setting Ethernet MAC failed\n");
2817 LOAD_ERROR_EXIT(bp, load_error3);
2820 if (IS_PF(bp) && bp->pending_max) {
2821 bnx2x_update_max_mf_config(bp, bp->pending_max);
2822 bp->pending_max = 0;
2825 bp->force_link_down = false;
2827 rc = bnx2x_initial_phy_init(bp, load_mode);
2829 LOAD_ERROR_EXIT(bp, load_error3);
2831 bp->link_params.feature_config_flags &= ~FEATURE_CONFIG_BOOT_FROM_SAN;
2833 /* Start fast path */
2835 /* Re-configure vlan filters */
2836 rc = bnx2x_vlan_reconfigure_vid(bp);
2838 LOAD_ERROR_EXIT(bp, load_error3);
2840 /* Initialize Rx filter. */
2841 bnx2x_set_rx_mode_inner(bp);
2843 if (bp->flags & PTP_SUPPORTED) {
2844 bnx2x_register_phc(bp);
2846 bnx2x_configure_ptp_filters(bp);
2849 switch (load_mode) {
2851 /* Tx queue should be only re-enabled */
2852 netif_tx_wake_all_queues(bp->dev);
2856 netif_tx_start_all_queues(bp->dev);
2857 smp_mb__after_atomic();
2861 case LOAD_LOOPBACK_EXT:
2862 bp->state = BNX2X_STATE_DIAG;
2870 bnx2x_update_drv_flags(bp, 1 << DRV_FLAGS_PORT_MASK, 0);
2872 bnx2x__link_status_update(bp);
2874 /* start the timer */
2875 mod_timer(&bp->timer, jiffies + bp->current_interval);
2877 if (CNIC_ENABLED(bp))
2878 bnx2x_load_cnic(bp);
2881 bnx2x_schedule_sp_rtnl(bp, BNX2X_SP_RTNL_GET_DRV_VERSION, 0);
2883 if (IS_PF(bp) && SHMEM2_HAS(bp, drv_capabilities_flag)) {
2884 /* mark driver is loaded in shmem2 */
2886 val = SHMEM2_RD(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)]);
2887 val &= ~DRV_FLAGS_MTU_MASK;
2888 val |= (bp->dev->mtu << DRV_FLAGS_MTU_SHIFT);
2889 SHMEM2_WR(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)],
2890 val | DRV_FLAGS_CAPABILITIES_LOADED_SUPPORTED |
2891 DRV_FLAGS_CAPABILITIES_LOADED_L2);
2894 /* Wait for all pending SP commands to complete */
2895 if (IS_PF(bp) && !bnx2x_wait_sp_comp(bp, ~0x0UL)) {
2896 BNX2X_ERR("Timeout waiting for SP elements to complete\n");
2897 bnx2x_nic_unload(bp, UNLOAD_CLOSE, false);
2901 /* Update driver data for On-Chip MFW dump. */
2903 bnx2x_update_mfw_dump(bp);
2905 /* If PMF - send ADMIN DCBX msg to MFW to initiate DCBX FSM */
2906 if (bp->port.pmf && (bp->state != BNX2X_STATE_DIAG))
2907 bnx2x_dcbx_init(bp, false);
2909 if (!IS_MF_SD_STORAGE_PERSONALITY_ONLY(bp))
2910 bnx2x_set_os_driver_state(bp, OS_DRIVER_STATE_ACTIVE);
2912 DP(NETIF_MSG_IFUP, "Ending successfully NIC load\n");
2916 #ifndef BNX2X_STOP_ON_ERROR
2919 bnx2x_int_disable_sync(bp, 1);
2921 /* Clean queueable objects */
2922 bnx2x_squeeze_objects(bp);
2925 /* Free SKBs, SGEs, TPA pool and driver internals */
2926 bnx2x_free_skbs(bp);
2927 for_each_rx_queue(bp, i)
2928 bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
2933 if (IS_PF(bp) && !BP_NOMCP(bp)) {
2934 bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_REQ_WOL_MCP, 0);
2935 bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_DONE, 0);
2940 bnx2x_napi_disable(bp);
2941 bnx2x_del_all_napi(bp);
2943 /* clear pf_load status, as it was already set */
2945 bnx2x_clear_pf_load(bp);
2947 bnx2x_free_fw_stats_mem(bp);
2948 bnx2x_free_fp_mem(bp);
2952 #endif /* ! BNX2X_STOP_ON_ERROR */
2955 int bnx2x_drain_tx_queues(struct bnx2x *bp)
2959 /* Wait until tx fastpath tasks complete */
2960 for_each_tx_queue(bp, i) {
2961 struct bnx2x_fastpath *fp = &bp->fp[i];
2963 for_each_cos_in_tx_queue(fp, cos)
2964 rc = bnx2x_clean_tx_queue(bp, fp->txdata_ptr[cos]);
2971 /* must be called with rtnl_lock */
2972 int bnx2x_nic_unload(struct bnx2x *bp, int unload_mode, bool keep_link)
2975 bool global = false;
2977 DP(NETIF_MSG_IFUP, "Starting NIC unload\n");
2979 if (!IS_MF_SD_STORAGE_PERSONALITY_ONLY(bp))
2980 bnx2x_set_os_driver_state(bp, OS_DRIVER_STATE_DISABLED);
2982 /* mark driver is unloaded in shmem2 */
2983 if (IS_PF(bp) && SHMEM2_HAS(bp, drv_capabilities_flag)) {
2985 val = SHMEM2_RD(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)]);
2986 SHMEM2_WR(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)],
2987 val & ~DRV_FLAGS_CAPABILITIES_LOADED_L2);
2990 if (IS_PF(bp) && bp->recovery_state != BNX2X_RECOVERY_DONE &&
2991 (bp->state == BNX2X_STATE_CLOSED ||
2992 bp->state == BNX2X_STATE_ERROR)) {
2993 /* We can get here if the driver has been unloaded
2994 * during parity error recovery and is either waiting for a
2995 * leader to complete or for other functions to unload and
2996 * then ifdown has been issued. In this case we want to
2997 * unload and let other functions to complete a recovery
3000 bp->recovery_state = BNX2X_RECOVERY_DONE;
3002 bnx2x_release_leader_lock(bp);
3005 DP(NETIF_MSG_IFDOWN, "Releasing a leadership...\n");
3006 BNX2X_ERR("Can't unload in closed or error state\n");
3010 /* Nothing to do during unload if previous bnx2x_nic_load()
3011 * have not completed successfully - all resources are released.
3013 * we can get here only after unsuccessful ndo_* callback, during which
3014 * dev->IFF_UP flag is still on.
3016 if (bp->state == BNX2X_STATE_CLOSED || bp->state == BNX2X_STATE_ERROR)
3019 /* It's important to set the bp->state to the value different from
3020 * BNX2X_STATE_OPEN and only then stop the Tx. Otherwise bnx2x_tx_int()
3021 * may restart the Tx from the NAPI context (see bnx2x_tx_int()).
3023 bp->state = BNX2X_STATE_CLOSING_WAIT4_HALT;
3026 /* indicate to VFs that the PF is going down */
3027 bnx2x_iov_channel_down(bp);
3029 if (CNIC_LOADED(bp))
3030 bnx2x_cnic_notify(bp, CNIC_CTL_STOP_CMD);
3033 bnx2x_tx_disable(bp);
3034 netdev_reset_tc(bp->dev);
3036 bp->rx_mode = BNX2X_RX_MODE_NONE;
3038 del_timer_sync(&bp->timer);
3040 if (IS_PF(bp) && !BP_NOMCP(bp)) {
3041 /* Set ALWAYS_ALIVE bit in shmem */
3042 bp->fw_drv_pulse_wr_seq |= DRV_PULSE_ALWAYS_ALIVE;
3043 bnx2x_drv_pulse(bp);
3044 bnx2x_stats_handle(bp, STATS_EVENT_STOP);
3045 bnx2x_save_statistics(bp);
3048 /* wait till consumers catch up with producers in all queues.
3049 * If we're recovering, FW can't write to host so no reason
3050 * to wait for the queues to complete all Tx.
3052 if (unload_mode != UNLOAD_RECOVERY)
3053 bnx2x_drain_tx_queues(bp);
3055 /* if VF indicate to PF this function is going down (PF will delete sp
3056 * elements and clear initializations
3059 bnx2x_vfpf_close_vf(bp);
3060 else if (unload_mode != UNLOAD_RECOVERY)
3061 /* if this is a normal/close unload need to clean up chip*/
3062 bnx2x_chip_cleanup(bp, unload_mode, keep_link);
3064 /* Send the UNLOAD_REQUEST to the MCP */
3065 bnx2x_send_unload_req(bp, unload_mode);
3067 /* Prevent transactions to host from the functions on the
3068 * engine that doesn't reset global blocks in case of global
3069 * attention once global blocks are reset and gates are opened
3070 * (the engine which leader will perform the recovery
3073 if (!CHIP_IS_E1x(bp))
3074 bnx2x_pf_disable(bp);
3076 /* Disable HW interrupts, NAPI */
3077 bnx2x_netif_stop(bp, 1);
3078 /* Delete all NAPI objects */
3079 bnx2x_del_all_napi(bp);
3080 if (CNIC_LOADED(bp))
3081 bnx2x_del_all_napi_cnic(bp);
3085 /* Report UNLOAD_DONE to MCP */
3086 bnx2x_send_unload_done(bp, false);
3090 * At this stage no more interrupts will arrive so we may safely clean
3091 * the queueable objects here in case they failed to get cleaned so far.
3094 bnx2x_squeeze_objects(bp);
3096 /* There should be no more pending SP commands at this stage */
3101 /* clear pending work in rtnl task */
3102 bp->sp_rtnl_state = 0;
3105 /* Free SKBs, SGEs, TPA pool and driver internals */
3106 bnx2x_free_skbs(bp);
3107 if (CNIC_LOADED(bp))
3108 bnx2x_free_skbs_cnic(bp);
3109 for_each_rx_queue(bp, i)
3110 bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
3112 bnx2x_free_fp_mem(bp);
3113 if (CNIC_LOADED(bp))
3114 bnx2x_free_fp_mem_cnic(bp);
3117 if (CNIC_LOADED(bp))
3118 bnx2x_free_mem_cnic(bp);
3122 bp->state = BNX2X_STATE_CLOSED;
3123 bp->cnic_loaded = false;
3125 /* Clear driver version indication in shmem */
3126 if (IS_PF(bp) && !BP_NOMCP(bp))
3127 bnx2x_update_mng_version(bp);
3129 /* Check if there are pending parity attentions. If there are - set
3130 * RECOVERY_IN_PROGRESS.
3132 if (IS_PF(bp) && bnx2x_chk_parity_attn(bp, &global, false)) {
3133 bnx2x_set_reset_in_progress(bp);
3135 /* Set RESET_IS_GLOBAL if needed */
3137 bnx2x_set_reset_global(bp);
3140 /* The last driver must disable a "close the gate" if there is no
3141 * parity attention or "process kill" pending.
3144 !bnx2x_clear_pf_load(bp) &&
3145 bnx2x_reset_is_done(bp, BP_PATH(bp)))
3146 bnx2x_disable_close_the_gate(bp);
3148 DP(NETIF_MSG_IFUP, "Ending NIC unload\n");
3153 int bnx2x_set_power_state(struct bnx2x *bp, pci_power_t state)
3157 /* If there is no power capability, silently succeed */
3158 if (!bp->pdev->pm_cap) {
3159 BNX2X_DEV_INFO("No power capability. Breaking.\n");
3163 pci_read_config_word(bp->pdev, bp->pdev->pm_cap + PCI_PM_CTRL, &pmcsr);
3167 pci_write_config_word(bp->pdev, bp->pdev->pm_cap + PCI_PM_CTRL,
3168 ((pmcsr & ~PCI_PM_CTRL_STATE_MASK) |
3169 PCI_PM_CTRL_PME_STATUS));
3171 if (pmcsr & PCI_PM_CTRL_STATE_MASK)
3172 /* delay required during transition out of D3hot */
3177 /* If there are other clients above don't
3178 shut down the power */
3179 if (atomic_read(&bp->pdev->enable_cnt) != 1)
3181 /* Don't shut down the power for emulation and FPGA */
3182 if (CHIP_REV_IS_SLOW(bp))
3185 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
3189 pmcsr |= PCI_PM_CTRL_PME_ENABLE;
3191 pci_write_config_word(bp->pdev, bp->pdev->pm_cap + PCI_PM_CTRL,
3194 /* No more memory access after this point until
3195 * device is brought back to D0.
3200 dev_err(&bp->pdev->dev, "Can't support state = %d\n", state);
3207 * net_device service functions
3209 static int bnx2x_poll(struct napi_struct *napi, int budget)
3211 struct bnx2x_fastpath *fp = container_of(napi, struct bnx2x_fastpath,
3213 struct bnx2x *bp = fp->bp;
3217 #ifdef BNX2X_STOP_ON_ERROR
3218 if (unlikely(bp->panic)) {
3219 napi_complete(napi);
3223 for_each_cos_in_tx_queue(fp, cos)
3224 if (bnx2x_tx_queue_has_work(fp->txdata_ptr[cos]))
3225 bnx2x_tx_int(bp, fp->txdata_ptr[cos]);
3227 rx_work_done = (bnx2x_has_rx_work(fp)) ? bnx2x_rx_int(fp, budget) : 0;
3229 if (rx_work_done < budget) {
3230 /* No need to update SB for FCoE L2 ring as long as
3231 * it's connected to the default SB and the SB
3232 * has been updated when NAPI was scheduled.
3234 if (IS_FCOE_FP(fp)) {
3235 napi_complete_done(napi, rx_work_done);
3237 bnx2x_update_fpsb_idx(fp);
3238 /* bnx2x_has_rx_work() reads the status block,
3239 * thus we need to ensure that status block indices
3240 * have been actually read (bnx2x_update_fpsb_idx)
3241 * prior to this check (bnx2x_has_rx_work) so that
3242 * we won't write the "newer" value of the status block
3243 * to IGU (if there was a DMA right after
3244 * bnx2x_has_rx_work and if there is no rmb, the memory
3245 * reading (bnx2x_update_fpsb_idx) may be postponed
3246 * to right before bnx2x_ack_sb). In this case there
3247 * will never be another interrupt until there is
3248 * another update of the status block, while there
3249 * is still unhandled work.
3253 if (!(bnx2x_has_rx_work(fp) || bnx2x_has_tx_work(fp))) {
3254 if (napi_complete_done(napi, rx_work_done)) {
3255 /* Re-enable interrupts */
3256 DP(NETIF_MSG_RX_STATUS,
3257 "Update index to %d\n", fp->fp_hc_idx);
3258 bnx2x_ack_sb(bp, fp->igu_sb_id, USTORM_ID,
3259 le16_to_cpu(fp->fp_hc_idx),
3263 rx_work_done = budget;
3268 return rx_work_done;
3271 /* we split the first BD into headers and data BDs
3272 * to ease the pain of our fellow microcode engineers
3273 * we use one mapping for both BDs
3275 static u16 bnx2x_tx_split(struct bnx2x *bp,
3276 struct bnx2x_fp_txdata *txdata,
3277 struct sw_tx_bd *tx_buf,
3278 struct eth_tx_start_bd **tx_bd, u16 hlen,
3281 struct eth_tx_start_bd *h_tx_bd = *tx_bd;
3282 struct eth_tx_bd *d_tx_bd;
3284 int old_len = le16_to_cpu(h_tx_bd->nbytes);
3286 /* first fix first BD */
3287 h_tx_bd->nbytes = cpu_to_le16(hlen);
3289 DP(NETIF_MSG_TX_QUEUED, "TSO split header size is %d (%x:%x)\n",
3290 h_tx_bd->nbytes, h_tx_bd->addr_hi, h_tx_bd->addr_lo);
3292 /* now get a new data BD
3293 * (after the pbd) and fill it */
3294 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3295 d_tx_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
3297 mapping = HILO_U64(le32_to_cpu(h_tx_bd->addr_hi),
3298 le32_to_cpu(h_tx_bd->addr_lo)) + hlen;
3300 d_tx_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
3301 d_tx_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
3302 d_tx_bd->nbytes = cpu_to_le16(old_len - hlen);
3304 /* this marks the BD as one that has no individual mapping */
3305 tx_buf->flags |= BNX2X_TSO_SPLIT_BD;
3307 DP(NETIF_MSG_TX_QUEUED,
3308 "TSO split data size is %d (%x:%x)\n",
3309 d_tx_bd->nbytes, d_tx_bd->addr_hi, d_tx_bd->addr_lo);
3312 *tx_bd = (struct eth_tx_start_bd *)d_tx_bd;
3317 #define bswab32(b32) ((__force __le32) swab32((__force __u32) (b32)))
3318 #define bswab16(b16) ((__force __le16) swab16((__force __u16) (b16)))
3319 static __le16 bnx2x_csum_fix(unsigned char *t_header, u16 csum, s8 fix)
3321 __sum16 tsum = (__force __sum16) csum;
3324 tsum = ~csum_fold(csum_sub((__force __wsum) csum,
3325 csum_partial(t_header - fix, fix, 0)));
3328 tsum = ~csum_fold(csum_add((__force __wsum) csum,
3329 csum_partial(t_header, -fix, 0)));
3331 return bswab16(tsum);
3334 static u32 bnx2x_xmit_type(struct bnx2x *bp, struct sk_buff *skb)
3340 if (skb->ip_summed != CHECKSUM_PARTIAL)
3343 protocol = vlan_get_protocol(skb);
3344 if (protocol == htons(ETH_P_IPV6)) {
3346 prot = ipv6_hdr(skb)->nexthdr;
3349 prot = ip_hdr(skb)->protocol;
3352 if (!CHIP_IS_E1x(bp) && skb->encapsulation) {
3353 if (inner_ip_hdr(skb)->version == 6) {
3354 rc |= XMIT_CSUM_ENC_V6;
3355 if (inner_ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3356 rc |= XMIT_CSUM_TCP;
3358 rc |= XMIT_CSUM_ENC_V4;
3359 if (inner_ip_hdr(skb)->protocol == IPPROTO_TCP)
3360 rc |= XMIT_CSUM_TCP;
3363 if (prot == IPPROTO_TCP)
3364 rc |= XMIT_CSUM_TCP;
3366 if (skb_is_gso(skb)) {
3367 if (skb_is_gso_v6(skb)) {
3368 rc |= (XMIT_GSO_V6 | XMIT_CSUM_TCP);
3369 if (rc & XMIT_CSUM_ENC)
3370 rc |= XMIT_GSO_ENC_V6;
3372 rc |= (XMIT_GSO_V4 | XMIT_CSUM_TCP);
3373 if (rc & XMIT_CSUM_ENC)
3374 rc |= XMIT_GSO_ENC_V4;
3381 /* VXLAN: 4 = 1 (for linear data BD) + 3 (2 for PBD and last BD) */
3382 #define BNX2X_NUM_VXLAN_TSO_WIN_SUB_BDS 4
3384 /* Regular: 3 = 1 (for linear data BD) + 2 (for PBD and last BD) */
3385 #define BNX2X_NUM_TSO_WIN_SUB_BDS 3
3387 #if (MAX_SKB_FRAGS >= MAX_FETCH_BD - BDS_PER_TX_PKT)
3388 /* check if packet requires linearization (packet is too fragmented)
3389 no need to check fragmentation if page size > 8K (there will be no
3390 violation to FW restrictions) */
3391 static int bnx2x_pkt_req_lin(struct bnx2x *bp, struct sk_buff *skb,
3394 int first_bd_sz = 0, num_tso_win_sub = BNX2X_NUM_TSO_WIN_SUB_BDS;
3395 int to_copy = 0, hlen = 0;
3397 if (xmit_type & XMIT_GSO_ENC)
3398 num_tso_win_sub = BNX2X_NUM_VXLAN_TSO_WIN_SUB_BDS;
3400 if (skb_shinfo(skb)->nr_frags >= (MAX_FETCH_BD - num_tso_win_sub)) {
3401 if (xmit_type & XMIT_GSO) {
3402 unsigned short lso_mss = skb_shinfo(skb)->gso_size;
3403 int wnd_size = MAX_FETCH_BD - num_tso_win_sub;
3404 /* Number of windows to check */
3405 int num_wnds = skb_shinfo(skb)->nr_frags - wnd_size;
3410 /* Headers length */
3411 if (xmit_type & XMIT_GSO_ENC)
3412 hlen = (int)(skb_inner_transport_header(skb) -
3414 inner_tcp_hdrlen(skb);
3416 hlen = (int)(skb_transport_header(skb) -
3417 skb->data) + tcp_hdrlen(skb);
3419 /* Amount of data (w/o headers) on linear part of SKB*/
3420 first_bd_sz = skb_headlen(skb) - hlen;
3422 wnd_sum = first_bd_sz;
3424 /* Calculate the first sum - it's special */
3425 for (frag_idx = 0; frag_idx < wnd_size - 1; frag_idx++)
3427 skb_frag_size(&skb_shinfo(skb)->frags[frag_idx]);
3429 /* If there was data on linear skb data - check it */
3430 if (first_bd_sz > 0) {
3431 if (unlikely(wnd_sum < lso_mss)) {
3436 wnd_sum -= first_bd_sz;
3439 /* Others are easier: run through the frag list and
3440 check all windows */
3441 for (wnd_idx = 0; wnd_idx <= num_wnds; wnd_idx++) {
3443 skb_frag_size(&skb_shinfo(skb)->frags[wnd_idx + wnd_size - 1]);
3445 if (unlikely(wnd_sum < lso_mss)) {
3450 skb_frag_size(&skb_shinfo(skb)->frags[wnd_idx]);
3453 /* in non-LSO too fragmented packet should always
3460 if (unlikely(to_copy))
3461 DP(NETIF_MSG_TX_QUEUED,
3462 "Linearization IS REQUIRED for %s packet. num_frags %d hlen %d first_bd_sz %d\n",
3463 (xmit_type & XMIT_GSO) ? "LSO" : "non-LSO",
3464 skb_shinfo(skb)->nr_frags, hlen, first_bd_sz);
3471 * bnx2x_set_pbd_gso - update PBD in GSO case.
3475 * @xmit_type: xmit flags
3477 static void bnx2x_set_pbd_gso(struct sk_buff *skb,
3478 struct eth_tx_parse_bd_e1x *pbd,
3481 pbd->lso_mss = cpu_to_le16(skb_shinfo(skb)->gso_size);
3482 pbd->tcp_send_seq = bswab32(tcp_hdr(skb)->seq);
3483 pbd->tcp_flags = pbd_tcp_flags(tcp_hdr(skb));
3485 if (xmit_type & XMIT_GSO_V4) {
3486 pbd->ip_id = bswab16(ip_hdr(skb)->id);
3487 pbd->tcp_pseudo_csum =
3488 bswab16(~csum_tcpudp_magic(ip_hdr(skb)->saddr,
3490 0, IPPROTO_TCP, 0));
3492 pbd->tcp_pseudo_csum =
3493 bswab16(~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3494 &ipv6_hdr(skb)->daddr,
3495 0, IPPROTO_TCP, 0));
3499 cpu_to_le16(ETH_TX_PARSE_BD_E1X_PSEUDO_CS_WITHOUT_LEN);
3503 * bnx2x_set_pbd_csum_enc - update PBD with checksum and return header length
3505 * @bp: driver handle
3507 * @parsing_data: data to be updated
3508 * @xmit_type: xmit flags
3510 * 57712/578xx related, when skb has encapsulation
3512 static u8 bnx2x_set_pbd_csum_enc(struct bnx2x *bp, struct sk_buff *skb,
3513 u32 *parsing_data, u32 xmit_type)
3516 ((((u8 *)skb_inner_transport_header(skb) - skb->data) >> 1) <<
3517 ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W_SHIFT) &
3518 ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W;
3520 if (xmit_type & XMIT_CSUM_TCP) {
3521 *parsing_data |= ((inner_tcp_hdrlen(skb) / 4) <<
3522 ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW_SHIFT) &
3523 ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW;
3525 return skb_inner_transport_header(skb) +
3526 inner_tcp_hdrlen(skb) - skb->data;
3529 /* We support checksum offload for TCP and UDP only.
3530 * No need to pass the UDP header length - it's a constant.
3532 return skb_inner_transport_header(skb) +
3533 sizeof(struct udphdr) - skb->data;
3537 * bnx2x_set_pbd_csum_e2 - update PBD with checksum and return header length
3539 * @bp: driver handle
3541 * @parsing_data: data to be updated
3542 * @xmit_type: xmit flags
3544 * 57712/578xx related
3546 static u8 bnx2x_set_pbd_csum_e2(struct bnx2x *bp, struct sk_buff *skb,
3547 u32 *parsing_data, u32 xmit_type)
3550 ((((u8 *)skb_transport_header(skb) - skb->data) >> 1) <<
3551 ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W_SHIFT) &
3552 ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W;
3554 if (xmit_type & XMIT_CSUM_TCP) {
3555 *parsing_data |= ((tcp_hdrlen(skb) / 4) <<
3556 ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW_SHIFT) &
3557 ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW;
3559 return skb_transport_header(skb) + tcp_hdrlen(skb) - skb->data;
3561 /* We support checksum offload for TCP and UDP only.
3562 * No need to pass the UDP header length - it's a constant.
3564 return skb_transport_header(skb) + sizeof(struct udphdr) - skb->data;
3567 /* set FW indication according to inner or outer protocols if tunneled */
3568 static void bnx2x_set_sbd_csum(struct bnx2x *bp, struct sk_buff *skb,
3569 struct eth_tx_start_bd *tx_start_bd,
3572 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_L4_CSUM;
3574 if (xmit_type & (XMIT_CSUM_ENC_V6 | XMIT_CSUM_V6))
3575 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_IPV6;
3577 if (!(xmit_type & XMIT_CSUM_TCP))
3578 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_IS_UDP;
3582 * bnx2x_set_pbd_csum - update PBD with checksum and return header length
3584 * @bp: driver handle
3586 * @pbd: parse BD to be updated
3587 * @xmit_type: xmit flags
3589 static u8 bnx2x_set_pbd_csum(struct bnx2x *bp, struct sk_buff *skb,
3590 struct eth_tx_parse_bd_e1x *pbd,
3593 u8 hlen = (skb_network_header(skb) - skb->data) >> 1;
3595 /* for now NS flag is not used in Linux */
3598 ((skb->protocol == cpu_to_be16(ETH_P_8021Q)) <<
3599 ETH_TX_PARSE_BD_E1X_LLC_SNAP_EN_SHIFT));
3601 pbd->ip_hlen_w = (skb_transport_header(skb) -
3602 skb_network_header(skb)) >> 1;
3604 hlen += pbd->ip_hlen_w;
3606 /* We support checksum offload for TCP and UDP only */
3607 if (xmit_type & XMIT_CSUM_TCP)
3608 hlen += tcp_hdrlen(skb) / 2;
3610 hlen += sizeof(struct udphdr) / 2;
3612 pbd->total_hlen_w = cpu_to_le16(hlen);
3615 if (xmit_type & XMIT_CSUM_TCP) {
3616 pbd->tcp_pseudo_csum = bswab16(tcp_hdr(skb)->check);
3619 s8 fix = SKB_CS_OFF(skb); /* signed! */
3621 DP(NETIF_MSG_TX_QUEUED,
3622 "hlen %d fix %d csum before fix %x\n",
3623 le16_to_cpu(pbd->total_hlen_w), fix, SKB_CS(skb));
3625 /* HW bug: fixup the CSUM */
3626 pbd->tcp_pseudo_csum =
3627 bnx2x_csum_fix(skb_transport_header(skb),
3630 DP(NETIF_MSG_TX_QUEUED, "csum after fix %x\n",
3631 pbd->tcp_pseudo_csum);
3637 static void bnx2x_update_pbds_gso_enc(struct sk_buff *skb,
3638 struct eth_tx_parse_bd_e2 *pbd_e2,
3639 struct eth_tx_parse_2nd_bd *pbd2,
3644 u8 outerip_off, outerip_len = 0;
3646 /* from outer IP to transport */
3647 hlen_w = (skb_inner_transport_header(skb) -
3648 skb_network_header(skb)) >> 1;
3651 hlen_w += inner_tcp_hdrlen(skb) >> 1;
3653 pbd2->fw_ip_hdr_to_payload_w = hlen_w;
3655 /* outer IP header info */
3656 if (xmit_type & XMIT_CSUM_V4) {
3657 struct iphdr *iph = ip_hdr(skb);
3658 u32 csum = (__force u32)(~iph->check) -
3659 (__force u32)iph->tot_len -
3660 (__force u32)iph->frag_off;
3662 outerip_len = iph->ihl << 1;
3664 pbd2->fw_ip_csum_wo_len_flags_frag =
3665 bswab16(csum_fold((__force __wsum)csum));
3667 pbd2->fw_ip_hdr_to_payload_w =
3668 hlen_w - ((sizeof(struct ipv6hdr)) >> 1);
3669 pbd_e2->data.tunnel_data.flags |=
3670 ETH_TUNNEL_DATA_IPV6_OUTER;
3673 pbd2->tcp_send_seq = bswab32(inner_tcp_hdr(skb)->seq);
3675 pbd2->tcp_flags = pbd_tcp_flags(inner_tcp_hdr(skb));
3677 /* inner IP header info */
3678 if (xmit_type & XMIT_CSUM_ENC_V4) {
3679 pbd2->hw_ip_id = bswab16(inner_ip_hdr(skb)->id);
3681 pbd_e2->data.tunnel_data.pseudo_csum =
3682 bswab16(~csum_tcpudp_magic(
3683 inner_ip_hdr(skb)->saddr,
3684 inner_ip_hdr(skb)->daddr,
3685 0, IPPROTO_TCP, 0));
3687 pbd_e2->data.tunnel_data.pseudo_csum =
3688 bswab16(~csum_ipv6_magic(
3689 &inner_ipv6_hdr(skb)->saddr,
3690 &inner_ipv6_hdr(skb)->daddr,
3691 0, IPPROTO_TCP, 0));
3694 outerip_off = (skb_network_header(skb) - skb->data) >> 1;
3699 ETH_TX_PARSE_2ND_BD_IP_HDR_LEN_OUTER_W_SHIFT) |
3700 ((skb->protocol == cpu_to_be16(ETH_P_8021Q)) <<
3701 ETH_TX_PARSE_2ND_BD_LLC_SNAP_EN_SHIFT);
3703 if (ip_hdr(skb)->protocol == IPPROTO_UDP) {
3704 SET_FLAG(*global_data, ETH_TX_PARSE_2ND_BD_TUNNEL_UDP_EXIST, 1);
3705 pbd2->tunnel_udp_hdr_start_w = skb_transport_offset(skb) >> 1;
3709 static inline void bnx2x_set_ipv6_ext_e2(struct sk_buff *skb, u32 *parsing_data,
3712 struct ipv6hdr *ipv6;
3714 if (!(xmit_type & (XMIT_GSO_ENC_V6 | XMIT_GSO_V6)))
3717 if (xmit_type & XMIT_GSO_ENC_V6)
3718 ipv6 = inner_ipv6_hdr(skb);
3719 else /* XMIT_GSO_V6 */
3720 ipv6 = ipv6_hdr(skb);
3722 if (ipv6->nexthdr == NEXTHDR_IPV6)
3723 *parsing_data |= ETH_TX_PARSE_BD_E2_IPV6_WITH_EXT_HDR;
3726 /* called with netif_tx_lock
3727 * bnx2x_tx_int() runs without netif_tx_lock unless it needs to call
3728 * netif_wake_queue()
3730 netdev_tx_t bnx2x_start_xmit(struct sk_buff *skb, struct net_device *dev)
3732 struct bnx2x *bp = netdev_priv(dev);
3734 struct netdev_queue *txq;
3735 struct bnx2x_fp_txdata *txdata;
3736 struct sw_tx_bd *tx_buf;
3737 struct eth_tx_start_bd *tx_start_bd, *first_bd;
3738 struct eth_tx_bd *tx_data_bd, *total_pkt_bd = NULL;
3739 struct eth_tx_parse_bd_e1x *pbd_e1x = NULL;
3740 struct eth_tx_parse_bd_e2 *pbd_e2 = NULL;
3741 struct eth_tx_parse_2nd_bd *pbd2 = NULL;
3742 u32 pbd_e2_parsing_data = 0;
3743 u16 pkt_prod, bd_prod;
3746 u32 xmit_type = bnx2x_xmit_type(bp, skb);
3749 __le16 pkt_size = 0;
3751 u8 mac_type = UNICAST_ADDRESS;
3753 #ifdef BNX2X_STOP_ON_ERROR
3754 if (unlikely(bp->panic))
3755 return NETDEV_TX_BUSY;
3758 txq_index = skb_get_queue_mapping(skb);
3759 txq = netdev_get_tx_queue(dev, txq_index);
3761 BUG_ON(txq_index >= MAX_ETH_TXQ_IDX(bp) + (CNIC_LOADED(bp) ? 1 : 0));
3763 txdata = &bp->bnx2x_txq[txq_index];
3765 /* enable this debug print to view the transmission queue being used
3766 DP(NETIF_MSG_TX_QUEUED, "indices: txq %d, fp %d, txdata %d\n",
3767 txq_index, fp_index, txdata_index); */
3769 /* enable this debug print to view the transmission details
3770 DP(NETIF_MSG_TX_QUEUED,
3771 "transmitting packet cid %d fp index %d txdata_index %d tx_data ptr %p fp pointer %p\n",
3772 txdata->cid, fp_index, txdata_index, txdata, fp); */
3774 if (unlikely(bnx2x_tx_avail(bp, txdata) <
3775 skb_shinfo(skb)->nr_frags +
3777 NEXT_CNT_PER_TX_PKT(MAX_BDS_PER_TX_PKT))) {
3778 /* Handle special storage cases separately */
3779 if (txdata->tx_ring_size == 0) {
3780 struct bnx2x_eth_q_stats *q_stats =
3781 bnx2x_fp_qstats(bp, txdata->parent_fp);
3782 q_stats->driver_filtered_tx_pkt++;
3784 return NETDEV_TX_OK;
3786 bnx2x_fp_qstats(bp, txdata->parent_fp)->driver_xoff++;
3787 netif_tx_stop_queue(txq);
3788 BNX2X_ERR("BUG! Tx ring full when queue awake!\n");
3790 return NETDEV_TX_BUSY;
3793 DP(NETIF_MSG_TX_QUEUED,
3794 "queue[%d]: SKB: summed %x protocol %x protocol(%x,%x) gso type %x xmit_type %x len %d\n",
3795 txq_index, skb->ip_summed, skb->protocol, ipv6_hdr(skb)->nexthdr,
3796 ip_hdr(skb)->protocol, skb_shinfo(skb)->gso_type, xmit_type,
3799 eth = (struct ethhdr *)skb->data;
3801 /* set flag according to packet type (UNICAST_ADDRESS is default)*/
3802 if (unlikely(is_multicast_ether_addr(eth->h_dest))) {
3803 if (is_broadcast_ether_addr(eth->h_dest))
3804 mac_type = BROADCAST_ADDRESS;
3806 mac_type = MULTICAST_ADDRESS;
3809 #if (MAX_SKB_FRAGS >= MAX_FETCH_BD - BDS_PER_TX_PKT)
3810 /* First, check if we need to linearize the skb (due to FW
3811 restrictions). No need to check fragmentation if page size > 8K
3812 (there will be no violation to FW restrictions) */
3813 if (bnx2x_pkt_req_lin(bp, skb, xmit_type)) {
3814 /* Statistics of linearization */
3816 if (skb_linearize(skb) != 0) {
3817 DP(NETIF_MSG_TX_QUEUED,
3818 "SKB linearization failed - silently dropping this SKB\n");
3819 dev_kfree_skb_any(skb);
3820 return NETDEV_TX_OK;
3824 /* Map skb linear data for DMA */
3825 mapping = dma_map_single(&bp->pdev->dev, skb->data,
3826 skb_headlen(skb), DMA_TO_DEVICE);
3827 if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
3828 DP(NETIF_MSG_TX_QUEUED,
3829 "SKB mapping failed - silently dropping this SKB\n");
3830 dev_kfree_skb_any(skb);
3831 return NETDEV_TX_OK;
3834 Please read carefully. First we use one BD which we mark as start,
3835 then we have a parsing info BD (used for TSO or xsum),
3836 and only then we have the rest of the TSO BDs.
3837 (don't forget to mark the last one as last,
3838 and to unmap only AFTER you write to the BD ...)
3839 And above all, all pdb sizes are in words - NOT DWORDS!
3842 /* get current pkt produced now - advance it just before sending packet
3843 * since mapping of pages may fail and cause packet to be dropped
3845 pkt_prod = txdata->tx_pkt_prod;
3846 bd_prod = TX_BD(txdata->tx_bd_prod);
3848 /* get a tx_buf and first BD
3849 * tx_start_bd may be changed during SPLIT,
3850 * but first_bd will always stay first
3852 tx_buf = &txdata->tx_buf_ring[TX_BD(pkt_prod)];
3853 tx_start_bd = &txdata->tx_desc_ring[bd_prod].start_bd;
3854 first_bd = tx_start_bd;
3856 tx_start_bd->bd_flags.as_bitfield = ETH_TX_BD_FLAGS_START_BD;
3858 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) {
3859 if (!(bp->flags & TX_TIMESTAMPING_EN)) {
3860 BNX2X_ERR("Tx timestamping was not enabled, this packet will not be timestamped\n");
3861 } else if (bp->ptp_tx_skb) {
3862 BNX2X_ERR("The device supports only a single outstanding packet to timestamp, this packet will not be timestamped\n");
3864 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
3865 /* schedule check for Tx timestamp */
3866 bp->ptp_tx_skb = skb_get(skb);
3867 bp->ptp_tx_start = jiffies;
3868 schedule_work(&bp->ptp_task);
3872 /* header nbd: indirectly zero other flags! */
3873 tx_start_bd->general_data = 1 << ETH_TX_START_BD_HDR_NBDS_SHIFT;
3875 /* remember the first BD of the packet */
3876 tx_buf->first_bd = txdata->tx_bd_prod;
3880 DP(NETIF_MSG_TX_QUEUED,
3881 "sending pkt %u @%p next_idx %u bd %u @%p\n",
3882 pkt_prod, tx_buf, txdata->tx_pkt_prod, bd_prod, tx_start_bd);
3884 if (skb_vlan_tag_present(skb)) {
3885 tx_start_bd->vlan_or_ethertype =
3886 cpu_to_le16(skb_vlan_tag_get(skb));
3887 tx_start_bd->bd_flags.as_bitfield |=
3888 (X_ETH_OUTBAND_VLAN << ETH_TX_BD_FLAGS_VLAN_MODE_SHIFT);
3890 /* when transmitting in a vf, start bd must hold the ethertype
3891 * for fw to enforce it
3894 #ifndef BNX2X_STOP_ON_ERROR
3897 /* Still need to consider inband vlan for enforced */
3898 if (__vlan_get_tag(skb, &vlan_tci)) {
3899 tx_start_bd->vlan_or_ethertype =
3900 cpu_to_le16(ntohs(eth->h_proto));
3902 tx_start_bd->bd_flags.as_bitfield |=
3903 (X_ETH_INBAND_VLAN <<
3904 ETH_TX_BD_FLAGS_VLAN_MODE_SHIFT);
3905 tx_start_bd->vlan_or_ethertype =
3906 cpu_to_le16(vlan_tci);
3908 #ifndef BNX2X_STOP_ON_ERROR
3910 /* used by FW for packet accounting */
3911 tx_start_bd->vlan_or_ethertype = cpu_to_le16(pkt_prod);
3916 nbd = 2; /* start_bd + pbd + frags (updated when pages are mapped) */
3918 /* turn on parsing and get a BD */
3919 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3921 if (xmit_type & XMIT_CSUM)
3922 bnx2x_set_sbd_csum(bp, skb, tx_start_bd, xmit_type);
3924 if (!CHIP_IS_E1x(bp)) {
3925 pbd_e2 = &txdata->tx_desc_ring[bd_prod].parse_bd_e2;
3926 memset(pbd_e2, 0, sizeof(struct eth_tx_parse_bd_e2));
3928 if (xmit_type & XMIT_CSUM_ENC) {
3929 u16 global_data = 0;
3931 /* Set PBD in enc checksum offload case */
3932 hlen = bnx2x_set_pbd_csum_enc(bp, skb,
3933 &pbd_e2_parsing_data,
3936 /* turn on 2nd parsing and get a BD */
3937 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3939 pbd2 = &txdata->tx_desc_ring[bd_prod].parse_2nd_bd;
3941 memset(pbd2, 0, sizeof(*pbd2));
3943 pbd_e2->data.tunnel_data.ip_hdr_start_inner_w =
3944 (skb_inner_network_header(skb) -
3947 if (xmit_type & XMIT_GSO_ENC)
3948 bnx2x_update_pbds_gso_enc(skb, pbd_e2, pbd2,
3952 pbd2->global_data = cpu_to_le16(global_data);
3954 /* add addition parse BD indication to start BD */
3955 SET_FLAG(tx_start_bd->general_data,
3956 ETH_TX_START_BD_PARSE_NBDS, 1);
3957 /* set encapsulation flag in start BD */
3958 SET_FLAG(tx_start_bd->general_data,
3959 ETH_TX_START_BD_TUNNEL_EXIST, 1);
3961 tx_buf->flags |= BNX2X_HAS_SECOND_PBD;
3964 } else if (xmit_type & XMIT_CSUM) {
3965 /* Set PBD in checksum offload case w/o encapsulation */
3966 hlen = bnx2x_set_pbd_csum_e2(bp, skb,
3967 &pbd_e2_parsing_data,
3971 bnx2x_set_ipv6_ext_e2(skb, &pbd_e2_parsing_data, xmit_type);
3972 /* Add the macs to the parsing BD if this is a vf or if
3973 * Tx Switching is enabled.
3976 /* override GRE parameters in BD */
3977 bnx2x_set_fw_mac_addr(&pbd_e2->data.mac_addr.src_hi,
3978 &pbd_e2->data.mac_addr.src_mid,
3979 &pbd_e2->data.mac_addr.src_lo,
3982 bnx2x_set_fw_mac_addr(&pbd_e2->data.mac_addr.dst_hi,
3983 &pbd_e2->data.mac_addr.dst_mid,
3984 &pbd_e2->data.mac_addr.dst_lo,
3987 if (bp->flags & TX_SWITCHING)
3988 bnx2x_set_fw_mac_addr(
3989 &pbd_e2->data.mac_addr.dst_hi,
3990 &pbd_e2->data.mac_addr.dst_mid,
3991 &pbd_e2->data.mac_addr.dst_lo,
3993 #ifdef BNX2X_STOP_ON_ERROR
3994 /* Enforce security is always set in Stop on Error -
3995 * source mac should be present in the parsing BD
3997 bnx2x_set_fw_mac_addr(&pbd_e2->data.mac_addr.src_hi,
3998 &pbd_e2->data.mac_addr.src_mid,
3999 &pbd_e2->data.mac_addr.src_lo,
4004 SET_FLAG(pbd_e2_parsing_data,
4005 ETH_TX_PARSE_BD_E2_ETH_ADDR_TYPE, mac_type);
4007 u16 global_data = 0;
4008 pbd_e1x = &txdata->tx_desc_ring[bd_prod].parse_bd_e1x;
4009 memset(pbd_e1x, 0, sizeof(struct eth_tx_parse_bd_e1x));
4010 /* Set PBD in checksum offload case */
4011 if (xmit_type & XMIT_CSUM)
4012 hlen = bnx2x_set_pbd_csum(bp, skb, pbd_e1x, xmit_type);
4014 SET_FLAG(global_data,
4015 ETH_TX_PARSE_BD_E1X_ETH_ADDR_TYPE, mac_type);
4016 pbd_e1x->global_data |= cpu_to_le16(global_data);
4019 /* Setup the data pointer of the first BD of the packet */
4020 tx_start_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
4021 tx_start_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
4022 tx_start_bd->nbytes = cpu_to_le16(skb_headlen(skb));
4023 pkt_size = tx_start_bd->nbytes;
4025 DP(NETIF_MSG_TX_QUEUED,
4026 "first bd @%p addr (%x:%x) nbytes %d flags %x vlan %x\n",
4027 tx_start_bd, tx_start_bd->addr_hi, tx_start_bd->addr_lo,
4028 le16_to_cpu(tx_start_bd->nbytes),
4029 tx_start_bd->bd_flags.as_bitfield,
4030 le16_to_cpu(tx_start_bd->vlan_or_ethertype));
4032 if (xmit_type & XMIT_GSO) {
4034 DP(NETIF_MSG_TX_QUEUED,
4035 "TSO packet len %d hlen %d total len %d tso size %d\n",
4036 skb->len, hlen, skb_headlen(skb),
4037 skb_shinfo(skb)->gso_size);
4039 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_SW_LSO;
4041 if (unlikely(skb_headlen(skb) > hlen)) {
4043 bd_prod = bnx2x_tx_split(bp, txdata, tx_buf,
4047 if (!CHIP_IS_E1x(bp))
4048 pbd_e2_parsing_data |=
4049 (skb_shinfo(skb)->gso_size <<
4050 ETH_TX_PARSE_BD_E2_LSO_MSS_SHIFT) &
4051 ETH_TX_PARSE_BD_E2_LSO_MSS;
4053 bnx2x_set_pbd_gso(skb, pbd_e1x, xmit_type);
4056 /* Set the PBD's parsing_data field if not zero
4057 * (for the chips newer than 57711).
4059 if (pbd_e2_parsing_data)
4060 pbd_e2->parsing_data = cpu_to_le32(pbd_e2_parsing_data);
4062 tx_data_bd = (struct eth_tx_bd *)tx_start_bd;
4064 /* Handle fragmented skb */
4065 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
4066 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
4068 mapping = skb_frag_dma_map(&bp->pdev->dev, frag, 0,
4069 skb_frag_size(frag), DMA_TO_DEVICE);
4070 if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
4071 unsigned int pkts_compl = 0, bytes_compl = 0;
4073 DP(NETIF_MSG_TX_QUEUED,
4074 "Unable to map page - dropping packet...\n");
4076 /* we need unmap all buffers already mapped
4078 * first_bd->nbd need to be properly updated
4079 * before call to bnx2x_free_tx_pkt
4081 first_bd->nbd = cpu_to_le16(nbd);
4082 bnx2x_free_tx_pkt(bp, txdata,
4083 TX_BD(txdata->tx_pkt_prod),
4084 &pkts_compl, &bytes_compl);
4085 return NETDEV_TX_OK;
4088 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
4089 tx_data_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
4090 if (total_pkt_bd == NULL)
4091 total_pkt_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
4093 tx_data_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
4094 tx_data_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
4095 tx_data_bd->nbytes = cpu_to_le16(skb_frag_size(frag));
4096 le16_add_cpu(&pkt_size, skb_frag_size(frag));
4099 DP(NETIF_MSG_TX_QUEUED,
4100 "frag %d bd @%p addr (%x:%x) nbytes %d\n",
4101 i, tx_data_bd, tx_data_bd->addr_hi, tx_data_bd->addr_lo,
4102 le16_to_cpu(tx_data_bd->nbytes));
4105 DP(NETIF_MSG_TX_QUEUED, "last bd @%p\n", tx_data_bd);
4107 /* update with actual num BDs */
4108 first_bd->nbd = cpu_to_le16(nbd);
4110 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
4112 /* now send a tx doorbell, counting the next BD
4113 * if the packet contains or ends with it
4115 if (TX_BD_POFF(bd_prod) < nbd)
4118 /* total_pkt_bytes should be set on the first data BD if
4119 * it's not an LSO packet and there is more than one
4120 * data BD. In this case pkt_size is limited by an MTU value.
4121 * However we prefer to set it for an LSO packet (while we don't
4122 * have to) in order to save some CPU cycles in a none-LSO
4123 * case, when we much more care about them.
4125 if (total_pkt_bd != NULL)
4126 total_pkt_bd->total_pkt_bytes = pkt_size;
4129 DP(NETIF_MSG_TX_QUEUED,
4130 "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",
4131 pbd_e1x, pbd_e1x->global_data, pbd_e1x->ip_hlen_w,
4132 pbd_e1x->ip_id, pbd_e1x->lso_mss, pbd_e1x->tcp_flags,
4133 pbd_e1x->tcp_pseudo_csum, pbd_e1x->tcp_send_seq,
4134 le16_to_cpu(pbd_e1x->total_hlen_w));
4136 DP(NETIF_MSG_TX_QUEUED,
4137 "PBD (E2) @%p dst %x %x %x src %x %x %x parsing_data %x\n",
4139 pbd_e2->data.mac_addr.dst_hi,
4140 pbd_e2->data.mac_addr.dst_mid,
4141 pbd_e2->data.mac_addr.dst_lo,
4142 pbd_e2->data.mac_addr.src_hi,
4143 pbd_e2->data.mac_addr.src_mid,
4144 pbd_e2->data.mac_addr.src_lo,
4145 pbd_e2->parsing_data);
4146 DP(NETIF_MSG_TX_QUEUED, "doorbell: nbd %d bd %u\n", nbd, bd_prod);
4148 netdev_tx_sent_queue(txq, skb->len);
4150 skb_tx_timestamp(skb);
4152 txdata->tx_pkt_prod++;
4154 * Make sure that the BD data is updated before updating the producer
4155 * since FW might read the BD right after the producer is updated.
4156 * This is only applicable for weak-ordered memory model archs such
4157 * as IA-64. The following barrier is also mandatory since FW will
4158 * assumes packets must have BDs.
4162 txdata->tx_db.data.prod += nbd;
4163 /* make sure descriptor update is observed by HW */
4166 DOORBELL_RELAXED(bp, txdata->cid, txdata->tx_db.raw);
4168 txdata->tx_bd_prod += nbd;
4170 if (unlikely(bnx2x_tx_avail(bp, txdata) < MAX_DESC_PER_TX_PKT)) {
4171 netif_tx_stop_queue(txq);
4173 /* paired memory barrier is in bnx2x_tx_int(), we have to keep
4174 * ordering of set_bit() in netif_tx_stop_queue() and read of
4178 bnx2x_fp_qstats(bp, txdata->parent_fp)->driver_xoff++;
4179 if (bnx2x_tx_avail(bp, txdata) >= MAX_DESC_PER_TX_PKT)
4180 netif_tx_wake_queue(txq);
4184 return NETDEV_TX_OK;
4187 void bnx2x_get_c2s_mapping(struct bnx2x *bp, u8 *c2s_map, u8 *c2s_default)
4189 int mfw_vn = BP_FW_MB_IDX(bp);
4192 /* If the shmem shouldn't affect configuration, reflect */
4193 if (!IS_MF_BD(bp)) {
4196 for (i = 0; i < BNX2X_MAX_PRIORITY; i++)
4203 tmp = SHMEM2_RD(bp, c2s_pcp_map_lower[mfw_vn]);
4204 tmp = (__force u32)be32_to_cpu((__force __be32)tmp);
4205 c2s_map[0] = tmp & 0xff;
4206 c2s_map[1] = (tmp >> 8) & 0xff;
4207 c2s_map[2] = (tmp >> 16) & 0xff;
4208 c2s_map[3] = (tmp >> 24) & 0xff;
4210 tmp = SHMEM2_RD(bp, c2s_pcp_map_upper[mfw_vn]);
4211 tmp = (__force u32)be32_to_cpu((__force __be32)tmp);
4212 c2s_map[4] = tmp & 0xff;
4213 c2s_map[5] = (tmp >> 8) & 0xff;
4214 c2s_map[6] = (tmp >> 16) & 0xff;
4215 c2s_map[7] = (tmp >> 24) & 0xff;
4217 tmp = SHMEM2_RD(bp, c2s_pcp_map_default[mfw_vn]);
4218 tmp = (__force u32)be32_to_cpu((__force __be32)tmp);
4219 *c2s_default = (tmp >> (8 * mfw_vn)) & 0xff;
4223 * bnx2x_setup_tc - routine to configure net_device for multi tc
4225 * @netdev: net device to configure
4226 * @tc: number of traffic classes to enable
4228 * callback connected to the ndo_setup_tc function pointer
4230 int bnx2x_setup_tc(struct net_device *dev, u8 num_tc)
4232 struct bnx2x *bp = netdev_priv(dev);
4233 u8 c2s_map[BNX2X_MAX_PRIORITY], c2s_def;
4234 int cos, prio, count, offset;
4236 /* setup tc must be called under rtnl lock */
4239 /* no traffic classes requested. Aborting */
4241 netdev_reset_tc(dev);
4245 /* requested to support too many traffic classes */
4246 if (num_tc > bp->max_cos) {
4247 BNX2X_ERR("support for too many traffic classes requested: %d. Max supported is %d\n",
4248 num_tc, bp->max_cos);
4252 /* declare amount of supported traffic classes */
4253 if (netdev_set_num_tc(dev, num_tc)) {
4254 BNX2X_ERR("failed to declare %d traffic classes\n", num_tc);
4258 bnx2x_get_c2s_mapping(bp, c2s_map, &c2s_def);
4260 /* configure priority to traffic class mapping */
4261 for (prio = 0; prio < BNX2X_MAX_PRIORITY; prio++) {
4262 int outer_prio = c2s_map[prio];
4264 netdev_set_prio_tc_map(dev, prio, bp->prio_to_cos[outer_prio]);
4265 DP(BNX2X_MSG_SP | NETIF_MSG_IFUP,
4266 "mapping priority %d to tc %d\n",
4267 outer_prio, bp->prio_to_cos[outer_prio]);
4270 /* Use this configuration to differentiate tc0 from other COSes
4271 This can be used for ets or pfc, and save the effort of setting
4272 up a multio class queue disc or negotiating DCBX with a switch
4273 netdev_set_prio_tc_map(dev, 0, 0);
4274 DP(BNX2X_MSG_SP, "mapping priority %d to tc %d\n", 0, 0);
4275 for (prio = 1; prio < 16; prio++) {
4276 netdev_set_prio_tc_map(dev, prio, 1);
4277 DP(BNX2X_MSG_SP, "mapping priority %d to tc %d\n", prio, 1);
4280 /* configure traffic class to transmission queue mapping */
4281 for (cos = 0; cos < bp->max_cos; cos++) {
4282 count = BNX2X_NUM_ETH_QUEUES(bp);
4283 offset = cos * BNX2X_NUM_NON_CNIC_QUEUES(bp);
4284 netdev_set_tc_queue(dev, cos, count, offset);
4285 DP(BNX2X_MSG_SP | NETIF_MSG_IFUP,
4286 "mapping tc %d to offset %d count %d\n",
4287 cos, offset, count);
4293 int __bnx2x_setup_tc(struct net_device *dev, enum tc_setup_type type,
4296 struct tc_mqprio_qopt *mqprio = type_data;
4298 if (type != TC_SETUP_QDISC_MQPRIO)
4301 mqprio->hw = TC_MQPRIO_HW_OFFLOAD_TCS;
4303 return bnx2x_setup_tc(dev, mqprio->num_tc);
4306 /* called with rtnl_lock */
4307 int bnx2x_change_mac_addr(struct net_device *dev, void *p)
4309 struct sockaddr *addr = p;
4310 struct bnx2x *bp = netdev_priv(dev);
4313 if (!is_valid_ether_addr(addr->sa_data)) {
4314 BNX2X_ERR("Requested MAC address is not valid\n");
4318 if (IS_MF_STORAGE_ONLY(bp)) {
4319 BNX2X_ERR("Can't change address on STORAGE ONLY function\n");
4323 if (netif_running(dev)) {
4324 rc = bnx2x_set_eth_mac(bp, false);
4329 memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
4331 if (netif_running(dev))
4332 rc = bnx2x_set_eth_mac(bp, true);
4334 if (IS_PF(bp) && SHMEM2_HAS(bp, curr_cfg))
4335 SHMEM2_WR(bp, curr_cfg, CURR_CFG_MET_OS);
4340 static void bnx2x_free_fp_mem_at(struct bnx2x *bp, int fp_index)
4342 union host_hc_status_block *sb = &bnx2x_fp(bp, fp_index, status_blk);
4343 struct bnx2x_fastpath *fp = &bp->fp[fp_index];
4348 if (IS_FCOE_IDX(fp_index)) {
4349 memset(sb, 0, sizeof(union host_hc_status_block));
4350 fp->status_blk_mapping = 0;
4353 if (!CHIP_IS_E1x(bp))
4354 BNX2X_PCI_FREE(sb->e2_sb,
4355 bnx2x_fp(bp, fp_index,
4356 status_blk_mapping),
4357 sizeof(struct host_hc_status_block_e2));
4359 BNX2X_PCI_FREE(sb->e1x_sb,
4360 bnx2x_fp(bp, fp_index,
4361 status_blk_mapping),
4362 sizeof(struct host_hc_status_block_e1x));
4366 if (!skip_rx_queue(bp, fp_index)) {
4367 bnx2x_free_rx_bds(fp);
4369 /* fastpath rx rings: rx_buf rx_desc rx_comp */
4370 BNX2X_FREE(bnx2x_fp(bp, fp_index, rx_buf_ring));
4371 BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_desc_ring),
4372 bnx2x_fp(bp, fp_index, rx_desc_mapping),
4373 sizeof(struct eth_rx_bd) * NUM_RX_BD);
4375 BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_comp_ring),
4376 bnx2x_fp(bp, fp_index, rx_comp_mapping),
4377 sizeof(struct eth_fast_path_rx_cqe) *
4381 BNX2X_FREE(bnx2x_fp(bp, fp_index, rx_page_ring));
4382 BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_sge_ring),
4383 bnx2x_fp(bp, fp_index, rx_sge_mapping),
4384 BCM_PAGE_SIZE * NUM_RX_SGE_PAGES);
4388 if (!skip_tx_queue(bp, fp_index)) {
4389 /* fastpath tx rings: tx_buf tx_desc */
4390 for_each_cos_in_tx_queue(fp, cos) {
4391 struct bnx2x_fp_txdata *txdata = fp->txdata_ptr[cos];
4393 DP(NETIF_MSG_IFDOWN,
4394 "freeing tx memory of fp %d cos %d cid %d\n",
4395 fp_index, cos, txdata->cid);
4397 BNX2X_FREE(txdata->tx_buf_ring);
4398 BNX2X_PCI_FREE(txdata->tx_desc_ring,
4399 txdata->tx_desc_mapping,
4400 sizeof(union eth_tx_bd_types) * NUM_TX_BD);
4403 /* end of fastpath */
4406 static void bnx2x_free_fp_mem_cnic(struct bnx2x *bp)
4409 for_each_cnic_queue(bp, i)
4410 bnx2x_free_fp_mem_at(bp, i);
4413 void bnx2x_free_fp_mem(struct bnx2x *bp)
4416 for_each_eth_queue(bp, i)
4417 bnx2x_free_fp_mem_at(bp, i);
4420 static void set_sb_shortcuts(struct bnx2x *bp, int index)
4422 union host_hc_status_block status_blk = bnx2x_fp(bp, index, status_blk);
4423 if (!CHIP_IS_E1x(bp)) {
4424 bnx2x_fp(bp, index, sb_index_values) =
4425 (__le16 *)status_blk.e2_sb->sb.index_values;
4426 bnx2x_fp(bp, index, sb_running_index) =
4427 (__le16 *)status_blk.e2_sb->sb.running_index;
4429 bnx2x_fp(bp, index, sb_index_values) =
4430 (__le16 *)status_blk.e1x_sb->sb.index_values;
4431 bnx2x_fp(bp, index, sb_running_index) =
4432 (__le16 *)status_blk.e1x_sb->sb.running_index;
4436 /* Returns the number of actually allocated BDs */
4437 static int bnx2x_alloc_rx_bds(struct bnx2x_fastpath *fp,
4440 struct bnx2x *bp = fp->bp;
4441 u16 ring_prod, cqe_ring_prod;
4442 int i, failure_cnt = 0;
4444 fp->rx_comp_cons = 0;
4445 cqe_ring_prod = ring_prod = 0;
4447 /* This routine is called only during fo init so
4448 * fp->eth_q_stats.rx_skb_alloc_failed = 0
4450 for (i = 0; i < rx_ring_size; i++) {
4451 if (bnx2x_alloc_rx_data(bp, fp, ring_prod, GFP_KERNEL) < 0) {
4455 ring_prod = NEXT_RX_IDX(ring_prod);
4456 cqe_ring_prod = NEXT_RCQ_IDX(cqe_ring_prod);
4457 WARN_ON(ring_prod <= (i - failure_cnt));
4461 BNX2X_ERR("was only able to allocate %d rx skbs on queue[%d]\n",
4462 i - failure_cnt, fp->index);
4464 fp->rx_bd_prod = ring_prod;
4465 /* Limit the CQE producer by the CQE ring size */
4466 fp->rx_comp_prod = min_t(u16, NUM_RCQ_RINGS*RCQ_DESC_CNT,
4469 bnx2x_fp_stats(bp, fp)->eth_q_stats.rx_skb_alloc_failed += failure_cnt;
4471 return i - failure_cnt;
4474 static void bnx2x_set_next_page_rx_cq(struct bnx2x_fastpath *fp)
4478 for (i = 1; i <= NUM_RCQ_RINGS; i++) {
4479 struct eth_rx_cqe_next_page *nextpg;
4481 nextpg = (struct eth_rx_cqe_next_page *)
4482 &fp->rx_comp_ring[RCQ_DESC_CNT * i - 1];
4484 cpu_to_le32(U64_HI(fp->rx_comp_mapping +
4485 BCM_PAGE_SIZE*(i % NUM_RCQ_RINGS)));
4487 cpu_to_le32(U64_LO(fp->rx_comp_mapping +
4488 BCM_PAGE_SIZE*(i % NUM_RCQ_RINGS)));
4492 static int bnx2x_alloc_fp_mem_at(struct bnx2x *bp, int index)
4494 union host_hc_status_block *sb;
4495 struct bnx2x_fastpath *fp = &bp->fp[index];
4498 int rx_ring_size = 0;
4500 if (!bp->rx_ring_size && IS_MF_STORAGE_ONLY(bp)) {
4501 rx_ring_size = MIN_RX_SIZE_NONTPA;
4502 bp->rx_ring_size = rx_ring_size;
4503 } else if (!bp->rx_ring_size) {
4504 rx_ring_size = MAX_RX_AVAIL/BNX2X_NUM_RX_QUEUES(bp);
4506 if (CHIP_IS_E3(bp)) {
4507 u32 cfg = SHMEM_RD(bp,
4508 dev_info.port_hw_config[BP_PORT(bp)].
4511 /* Decrease ring size for 1G functions */
4512 if ((cfg & PORT_HW_CFG_NET_SERDES_IF_MASK) ==
4513 PORT_HW_CFG_NET_SERDES_IF_SGMII)
4517 /* allocate at least number of buffers required by FW */
4518 rx_ring_size = max_t(int, bp->disable_tpa ? MIN_RX_SIZE_NONTPA :
4519 MIN_RX_SIZE_TPA, rx_ring_size);
4521 bp->rx_ring_size = rx_ring_size;
4522 } else /* if rx_ring_size specified - use it */
4523 rx_ring_size = bp->rx_ring_size;
4525 DP(BNX2X_MSG_SP, "calculated rx_ring_size %d\n", rx_ring_size);
4528 sb = &bnx2x_fp(bp, index, status_blk);
4530 if (!IS_FCOE_IDX(index)) {
4532 if (!CHIP_IS_E1x(bp)) {
4533 sb->e2_sb = BNX2X_PCI_ALLOC(&bnx2x_fp(bp, index, status_blk_mapping),
4534 sizeof(struct host_hc_status_block_e2));
4538 sb->e1x_sb = BNX2X_PCI_ALLOC(&bnx2x_fp(bp, index, status_blk_mapping),
4539 sizeof(struct host_hc_status_block_e1x));
4545 /* FCoE Queue uses Default SB and doesn't ACK the SB, thus no need to
4546 * set shortcuts for it.
4548 if (!IS_FCOE_IDX(index))
4549 set_sb_shortcuts(bp, index);
4552 if (!skip_tx_queue(bp, index)) {
4553 /* fastpath tx rings: tx_buf tx_desc */
4554 for_each_cos_in_tx_queue(fp, cos) {
4555 struct bnx2x_fp_txdata *txdata = fp->txdata_ptr[cos];
4558 "allocating tx memory of fp %d cos %d\n",
4561 txdata->tx_buf_ring = kcalloc(NUM_TX_BD,
4562 sizeof(struct sw_tx_bd),
4564 if (!txdata->tx_buf_ring)
4566 txdata->tx_desc_ring = BNX2X_PCI_ALLOC(&txdata->tx_desc_mapping,
4567 sizeof(union eth_tx_bd_types) * NUM_TX_BD);
4568 if (!txdata->tx_desc_ring)
4574 if (!skip_rx_queue(bp, index)) {
4575 /* fastpath rx rings: rx_buf rx_desc rx_comp */
4576 bnx2x_fp(bp, index, rx_buf_ring) =
4577 kcalloc(NUM_RX_BD, sizeof(struct sw_rx_bd), GFP_KERNEL);
4578 if (!bnx2x_fp(bp, index, rx_buf_ring))
4580 bnx2x_fp(bp, index, rx_desc_ring) =
4581 BNX2X_PCI_ALLOC(&bnx2x_fp(bp, index, rx_desc_mapping),
4582 sizeof(struct eth_rx_bd) * NUM_RX_BD);
4583 if (!bnx2x_fp(bp, index, rx_desc_ring))
4586 /* Seed all CQEs by 1s */
4587 bnx2x_fp(bp, index, rx_comp_ring) =
4588 BNX2X_PCI_FALLOC(&bnx2x_fp(bp, index, rx_comp_mapping),
4589 sizeof(struct eth_fast_path_rx_cqe) * NUM_RCQ_BD);
4590 if (!bnx2x_fp(bp, index, rx_comp_ring))
4594 bnx2x_fp(bp, index, rx_page_ring) =
4595 kcalloc(NUM_RX_SGE, sizeof(struct sw_rx_page),
4597 if (!bnx2x_fp(bp, index, rx_page_ring))
4599 bnx2x_fp(bp, index, rx_sge_ring) =
4600 BNX2X_PCI_ALLOC(&bnx2x_fp(bp, index, rx_sge_mapping),
4601 BCM_PAGE_SIZE * NUM_RX_SGE_PAGES);
4602 if (!bnx2x_fp(bp, index, rx_sge_ring))
4605 bnx2x_set_next_page_rx_bd(fp);
4608 bnx2x_set_next_page_rx_cq(fp);
4611 ring_size = bnx2x_alloc_rx_bds(fp, rx_ring_size);
4612 if (ring_size < rx_ring_size)
4618 /* handles low memory cases */
4620 BNX2X_ERR("Unable to allocate full memory for queue %d (size %d)\n",
4622 /* FW will drop all packets if queue is not big enough,
4623 * In these cases we disable the queue
4624 * Min size is different for OOO, TPA and non-TPA queues
4626 if (ring_size < (fp->mode == TPA_MODE_DISABLED ?
4627 MIN_RX_SIZE_NONTPA : MIN_RX_SIZE_TPA)) {
4628 /* release memory allocated for this queue */
4629 bnx2x_free_fp_mem_at(bp, index);
4635 static int bnx2x_alloc_fp_mem_cnic(struct bnx2x *bp)
4639 if (bnx2x_alloc_fp_mem_at(bp, FCOE_IDX(bp)))
4640 /* we will fail load process instead of mark
4648 static int bnx2x_alloc_fp_mem(struct bnx2x *bp)
4652 /* 1. Allocate FP for leading - fatal if error
4653 * 2. Allocate RSS - fix number of queues if error
4657 if (bnx2x_alloc_fp_mem_at(bp, 0))
4661 for_each_nondefault_eth_queue(bp, i)
4662 if (bnx2x_alloc_fp_mem_at(bp, i))
4665 /* handle memory failures */
4666 if (i != BNX2X_NUM_ETH_QUEUES(bp)) {
4667 int delta = BNX2X_NUM_ETH_QUEUES(bp) - i;
4670 bnx2x_shrink_eth_fp(bp, delta);
4671 if (CNIC_SUPPORT(bp))
4672 /* move non eth FPs next to last eth FP
4673 * must be done in that order
4674 * FCOE_IDX < FWD_IDX < OOO_IDX
4677 /* move FCoE fp even NO_FCOE_FLAG is on */
4678 bnx2x_move_fp(bp, FCOE_IDX(bp), FCOE_IDX(bp) - delta);
4679 bp->num_ethernet_queues -= delta;
4680 bp->num_queues = bp->num_ethernet_queues +
4681 bp->num_cnic_queues;
4682 BNX2X_ERR("Adjusted num of queues from %d to %d\n",
4683 bp->num_queues + delta, bp->num_queues);
4689 void bnx2x_free_mem_bp(struct bnx2x *bp)
4693 for (i = 0; i < bp->fp_array_size; i++)
4694 kfree(bp->fp[i].tpa_info);
4697 kfree(bp->fp_stats);
4698 kfree(bp->bnx2x_txq);
4699 kfree(bp->msix_table);
4703 int bnx2x_alloc_mem_bp(struct bnx2x *bp)
4705 struct bnx2x_fastpath *fp;
4706 struct msix_entry *tbl;
4707 struct bnx2x_ilt *ilt;
4708 int msix_table_size = 0;
4709 int fp_array_size, txq_array_size;
4713 * The biggest MSI-X table we might need is as a maximum number of fast
4714 * path IGU SBs plus default SB (for PF only).
4716 msix_table_size = bp->igu_sb_cnt;
4719 BNX2X_DEV_INFO("msix_table_size %d\n", msix_table_size);
4721 /* fp array: RSS plus CNIC related L2 queues */
4722 fp_array_size = BNX2X_MAX_RSS_COUNT(bp) + CNIC_SUPPORT(bp);
4723 bp->fp_array_size = fp_array_size;
4724 BNX2X_DEV_INFO("fp_array_size %d\n", bp->fp_array_size);
4726 fp = kcalloc(bp->fp_array_size, sizeof(*fp), GFP_KERNEL);
4729 for (i = 0; i < bp->fp_array_size; i++) {
4731 kcalloc(ETH_MAX_AGGREGATION_QUEUES_E1H_E2,
4732 sizeof(struct bnx2x_agg_info), GFP_KERNEL);
4733 if (!(fp[i].tpa_info))
4739 /* allocate sp objs */
4740 bp->sp_objs = kcalloc(bp->fp_array_size, sizeof(struct bnx2x_sp_objs),
4745 /* allocate fp_stats */
4746 bp->fp_stats = kcalloc(bp->fp_array_size, sizeof(struct bnx2x_fp_stats),
4751 /* Allocate memory for the transmission queues array */
4753 BNX2X_MAX_RSS_COUNT(bp) * BNX2X_MULTI_TX_COS + CNIC_SUPPORT(bp);
4754 BNX2X_DEV_INFO("txq_array_size %d", txq_array_size);
4756 bp->bnx2x_txq = kcalloc(txq_array_size, sizeof(struct bnx2x_fp_txdata),
4762 tbl = kcalloc(msix_table_size, sizeof(*tbl), GFP_KERNEL);
4765 bp->msix_table = tbl;
4768 ilt = kzalloc(sizeof(*ilt), GFP_KERNEL);
4775 bnx2x_free_mem_bp(bp);
4779 int bnx2x_reload_if_running(struct net_device *dev)
4781 struct bnx2x *bp = netdev_priv(dev);
4783 if (unlikely(!netif_running(dev)))
4786 bnx2x_nic_unload(bp, UNLOAD_NORMAL, true);
4787 return bnx2x_nic_load(bp, LOAD_NORMAL);
4790 int bnx2x_get_cur_phy_idx(struct bnx2x *bp)
4792 u32 sel_phy_idx = 0;
4793 if (bp->link_params.num_phys <= 1)
4796 if (bp->link_vars.link_up) {
4797 sel_phy_idx = EXT_PHY1;
4798 /* In case link is SERDES, check if the EXT_PHY2 is the one */
4799 if ((bp->link_vars.link_status & LINK_STATUS_SERDES_LINK) &&
4800 (bp->link_params.phy[EXT_PHY2].supported & SUPPORTED_FIBRE))
4801 sel_phy_idx = EXT_PHY2;
4804 switch (bnx2x_phy_selection(&bp->link_params)) {
4805 case PORT_HW_CFG_PHY_SELECTION_HARDWARE_DEFAULT:
4806 case PORT_HW_CFG_PHY_SELECTION_FIRST_PHY:
4807 case PORT_HW_CFG_PHY_SELECTION_FIRST_PHY_PRIORITY:
4808 sel_phy_idx = EXT_PHY1;
4810 case PORT_HW_CFG_PHY_SELECTION_SECOND_PHY:
4811 case PORT_HW_CFG_PHY_SELECTION_SECOND_PHY_PRIORITY:
4812 sel_phy_idx = EXT_PHY2;
4819 int bnx2x_get_link_cfg_idx(struct bnx2x *bp)
4821 u32 sel_phy_idx = bnx2x_get_cur_phy_idx(bp);
4823 * The selected activated PHY is always after swapping (in case PHY
4824 * swapping is enabled). So when swapping is enabled, we need to reverse
4828 if (bp->link_params.multi_phy_config &
4829 PORT_HW_CFG_PHY_SWAPPED_ENABLED) {
4830 if (sel_phy_idx == EXT_PHY1)
4831 sel_phy_idx = EXT_PHY2;
4832 else if (sel_phy_idx == EXT_PHY2)
4833 sel_phy_idx = EXT_PHY1;
4835 return LINK_CONFIG_IDX(sel_phy_idx);
4838 #ifdef NETDEV_FCOE_WWNN
4839 int bnx2x_fcoe_get_wwn(struct net_device *dev, u64 *wwn, int type)
4841 struct bnx2x *bp = netdev_priv(dev);
4842 struct cnic_eth_dev *cp = &bp->cnic_eth_dev;
4845 case NETDEV_FCOE_WWNN:
4846 *wwn = HILO_U64(cp->fcoe_wwn_node_name_hi,
4847 cp->fcoe_wwn_node_name_lo);
4849 case NETDEV_FCOE_WWPN:
4850 *wwn = HILO_U64(cp->fcoe_wwn_port_name_hi,
4851 cp->fcoe_wwn_port_name_lo);
4854 BNX2X_ERR("Wrong WWN type requested - %d\n", type);
4862 /* called with rtnl_lock */
4863 int bnx2x_change_mtu(struct net_device *dev, int new_mtu)
4865 struct bnx2x *bp = netdev_priv(dev);
4867 if (pci_num_vf(bp->pdev)) {
4868 DP(BNX2X_MSG_IOV, "VFs are enabled, can not change MTU\n");
4872 if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
4873 BNX2X_ERR("Can't perform change MTU during parity recovery\n");
4877 /* This does not race with packet allocation
4878 * because the actual alloc size is
4879 * only updated as part of load
4883 if (!bnx2x_mtu_allows_gro(new_mtu))
4884 dev->features &= ~NETIF_F_GRO_HW;
4886 if (IS_PF(bp) && SHMEM2_HAS(bp, curr_cfg))
4887 SHMEM2_WR(bp, curr_cfg, CURR_CFG_MET_OS);
4889 return bnx2x_reload_if_running(dev);
4892 netdev_features_t bnx2x_fix_features(struct net_device *dev,
4893 netdev_features_t features)
4895 struct bnx2x *bp = netdev_priv(dev);
4897 if (pci_num_vf(bp->pdev)) {
4898 netdev_features_t changed = dev->features ^ features;
4900 /* Revert the requested changes in features if they
4901 * would require internal reload of PF in bnx2x_set_features().
4903 if (!(features & NETIF_F_RXCSUM) && !bp->disable_tpa) {
4904 features &= ~NETIF_F_RXCSUM;
4905 features |= dev->features & NETIF_F_RXCSUM;
4908 if (changed & NETIF_F_LOOPBACK) {
4909 features &= ~NETIF_F_LOOPBACK;
4910 features |= dev->features & NETIF_F_LOOPBACK;
4914 /* TPA requires Rx CSUM offloading */
4915 if (!(features & NETIF_F_RXCSUM))
4916 features &= ~NETIF_F_LRO;
4918 if (!(features & NETIF_F_GRO) || !bnx2x_mtu_allows_gro(dev->mtu))
4919 features &= ~NETIF_F_GRO_HW;
4920 if (features & NETIF_F_GRO_HW)
4921 features &= ~NETIF_F_LRO;
4926 int bnx2x_set_features(struct net_device *dev, netdev_features_t features)
4928 struct bnx2x *bp = netdev_priv(dev);
4929 netdev_features_t changes = features ^ dev->features;
4930 bool bnx2x_reload = false;
4933 /* VFs or non SRIOV PFs should be able to change loopback feature */
4934 if (!pci_num_vf(bp->pdev)) {
4935 if (features & NETIF_F_LOOPBACK) {
4936 if (bp->link_params.loopback_mode != LOOPBACK_BMAC) {
4937 bp->link_params.loopback_mode = LOOPBACK_BMAC;
4938 bnx2x_reload = true;
4941 if (bp->link_params.loopback_mode != LOOPBACK_NONE) {
4942 bp->link_params.loopback_mode = LOOPBACK_NONE;
4943 bnx2x_reload = true;
4948 /* Don't care about GRO changes */
4949 changes &= ~NETIF_F_GRO;
4952 bnx2x_reload = true;
4955 if (bp->recovery_state == BNX2X_RECOVERY_DONE) {
4956 dev->features = features;
4957 rc = bnx2x_reload_if_running(dev);
4960 /* else: bnx2x_nic_load() will be called at end of recovery */
4966 void bnx2x_tx_timeout(struct net_device *dev)
4968 struct bnx2x *bp = netdev_priv(dev);
4970 /* We want the information of the dump logged,
4971 * but calling bnx2x_panic() would kill all chances of recovery.
4974 #ifndef BNX2X_STOP_ON_ERROR
4975 bnx2x_panic_dump(bp, false);
4980 /* This allows the netif to be shutdown gracefully before resetting */
4981 bnx2x_schedule_sp_rtnl(bp, BNX2X_SP_RTNL_TX_TIMEOUT, 0);
4984 int bnx2x_suspend(struct pci_dev *pdev, pm_message_t state)
4986 struct net_device *dev = pci_get_drvdata(pdev);
4990 dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
4993 bp = netdev_priv(dev);
4997 pci_save_state(pdev);
4999 if (!netif_running(dev)) {
5004 netif_device_detach(dev);
5006 bnx2x_nic_unload(bp, UNLOAD_CLOSE, false);
5008 bnx2x_set_power_state(bp, pci_choose_state(pdev, state));
5015 int bnx2x_resume(struct pci_dev *pdev)
5017 struct net_device *dev = pci_get_drvdata(pdev);
5022 dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
5025 bp = netdev_priv(dev);
5027 if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
5028 BNX2X_ERR("Handling parity error recovery. Try again later\n");
5034 pci_restore_state(pdev);
5036 if (!netif_running(dev)) {
5041 bnx2x_set_power_state(bp, PCI_D0);
5042 netif_device_attach(dev);
5044 rc = bnx2x_nic_load(bp, LOAD_OPEN);
5051 void bnx2x_set_ctx_validation(struct bnx2x *bp, struct eth_context *cxt,
5055 BNX2X_ERR("bad context pointer %p\n", cxt);
5059 /* ustorm cxt validation */
5060 cxt->ustorm_ag_context.cdu_usage =
5061 CDU_RSRVD_VALUE_TYPE_A(HW_CID(bp, cid),
5062 CDU_REGION_NUMBER_UCM_AG, ETH_CONNECTION_TYPE);
5063 /* xcontext validation */
5064 cxt->xstorm_ag_context.cdu_reserved =
5065 CDU_RSRVD_VALUE_TYPE_A(HW_CID(bp, cid),
5066 CDU_REGION_NUMBER_XCM_AG, ETH_CONNECTION_TYPE);
5069 static void storm_memset_hc_timeout(struct bnx2x *bp, u8 port,
5070 u8 fw_sb_id, u8 sb_index,
5073 u32 addr = BAR_CSTRORM_INTMEM +
5074 CSTORM_STATUS_BLOCK_DATA_TIMEOUT_OFFSET(fw_sb_id, sb_index);
5075 REG_WR8(bp, addr, ticks);
5077 "port %x fw_sb_id %d sb_index %d ticks %d\n",
5078 port, fw_sb_id, sb_index, ticks);
5081 static void storm_memset_hc_disable(struct bnx2x *bp, u8 port,
5082 u16 fw_sb_id, u8 sb_index,
5085 u32 enable_flag = disable ? 0 : (1 << HC_INDEX_DATA_HC_ENABLED_SHIFT);
5086 u32 addr = BAR_CSTRORM_INTMEM +
5087 CSTORM_STATUS_BLOCK_DATA_FLAGS_OFFSET(fw_sb_id, sb_index);
5088 u8 flags = REG_RD8(bp, addr);
5090 flags &= ~HC_INDEX_DATA_HC_ENABLED;
5091 flags |= enable_flag;
5092 REG_WR8(bp, addr, flags);
5094 "port %x fw_sb_id %d sb_index %d disable %d\n",
5095 port, fw_sb_id, sb_index, disable);
5098 void bnx2x_update_coalesce_sb_index(struct bnx2x *bp, u8 fw_sb_id,
5099 u8 sb_index, u8 disable, u16 usec)
5101 int port = BP_PORT(bp);
5102 u8 ticks = usec / BNX2X_BTR;
5104 storm_memset_hc_timeout(bp, port, fw_sb_id, sb_index, ticks);
5106 disable = disable ? 1 : (usec ? 0 : 1);
5107 storm_memset_hc_disable(bp, port, fw_sb_id, sb_index, disable);
5110 void bnx2x_schedule_sp_rtnl(struct bnx2x *bp, enum sp_rtnl_flag flag,
5113 smp_mb__before_atomic();
5114 set_bit(flag, &bp->sp_rtnl_state);
5115 smp_mb__after_atomic();
5116 DP((BNX2X_MSG_SP | verbose), "Scheduling sp_rtnl task [Flag: %d]\n",
5118 schedule_delayed_work(&bp->sp_rtnl_task, 0);