1 /* bnx2x_cmn.c: Broadcom Everest network driver.
3 * Copyright (c) 2007-2013 Broadcom Corporation
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation.
9 * Maintained by: Ariel Elior <ariel.elior@qlogic.com>
10 * Written by: Eliezer Tamir
11 * Based on code from Michael Chan's bnx2 driver
12 * UDP CSUM errata workaround by Arik Gendelman
13 * Slowpath and fastpath rework by Vladislav Zolotarov
14 * Statistics and Link management by Yitchak Gertner
18 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
20 #include <linux/etherdevice.h>
21 #include <linux/if_vlan.h>
22 #include <linux/interrupt.h>
26 #include <net/ip6_checksum.h>
27 #include <net/busy_poll.h>
28 #include <linux/prefetch.h>
29 #include "bnx2x_cmn.h"
30 #include "bnx2x_init.h"
33 static void bnx2x_free_fp_mem_cnic(struct bnx2x *bp);
34 static int bnx2x_alloc_fp_mem_cnic(struct bnx2x *bp);
35 static int bnx2x_alloc_fp_mem(struct bnx2x *bp);
36 static int bnx2x_poll(struct napi_struct *napi, int budget);
38 static void bnx2x_add_all_napi_cnic(struct bnx2x *bp)
42 /* Add NAPI objects */
43 for_each_rx_queue_cnic(bp, i) {
44 netif_napi_add(bp->dev, &bnx2x_fp(bp, i, napi),
45 bnx2x_poll, NAPI_POLL_WEIGHT);
46 napi_hash_add(&bnx2x_fp(bp, i, napi));
50 static void bnx2x_add_all_napi(struct bnx2x *bp)
54 /* Add NAPI objects */
55 for_each_eth_queue(bp, i) {
56 netif_napi_add(bp->dev, &bnx2x_fp(bp, i, napi),
57 bnx2x_poll, NAPI_POLL_WEIGHT);
58 napi_hash_add(&bnx2x_fp(bp, i, napi));
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 */
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;
266 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 #ifdef _ASM_GENERIC_INT_L64_H
487 DP(NETIF_MSG_RX_STATUS, "fp->tpa_queue_used = 0x%lx\n",
489 DP(NETIF_MSG_RX_STATUS, "fp->tpa_queue_used = 0x%llx\n",
495 /* Timestamp option length allowed for TPA aggregation:
497 * nop nop kind length echo val
499 #define TPA_TSTAMP_OPT_LEN 12
501 * bnx2x_set_gro_params - compute GRO values
504 * @parsing_flags: parsing flags from the START CQE
505 * @len_on_bd: total length of the first packet for the
507 * @pkt_len: length of all segments
509 * Approximate value of the MSS for this aggregation calculated using
510 * the first packet of it.
511 * Compute number of aggregated segments, and gso_type.
513 static void bnx2x_set_gro_params(struct sk_buff *skb, u16 parsing_flags,
514 u16 len_on_bd, unsigned int pkt_len,
515 u16 num_of_coalesced_segs)
517 /* TPA aggregation won't have either IP options or TCP options
518 * other than timestamp or IPv6 extension headers.
520 u16 hdrs_len = ETH_HLEN + sizeof(struct tcphdr);
522 if (GET_FLAG(parsing_flags, PARSING_FLAGS_OVER_ETHERNET_PROTOCOL) ==
523 PRS_FLAG_OVERETH_IPV6) {
524 hdrs_len += sizeof(struct ipv6hdr);
525 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
527 hdrs_len += sizeof(struct iphdr);
528 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
531 /* Check if there was a TCP timestamp, if there is it's will
532 * always be 12 bytes length: nop nop kind length echo val.
534 * Otherwise FW would close the aggregation.
536 if (parsing_flags & PARSING_FLAGS_TIME_STAMP_EXIST_FLAG)
537 hdrs_len += TPA_TSTAMP_OPT_LEN;
539 skb_shinfo(skb)->gso_size = len_on_bd - hdrs_len;
541 /* tcp_gro_complete() will copy NAPI_GRO_CB(skb)->count
542 * to skb_shinfo(skb)->gso_segs
544 NAPI_GRO_CB(skb)->count = num_of_coalesced_segs;
547 static int bnx2x_alloc_rx_sge(struct bnx2x *bp, struct bnx2x_fastpath *fp,
548 u16 index, gfp_t gfp_mask)
550 struct page *page = alloc_pages(gfp_mask, PAGES_PER_SGE_SHIFT);
551 struct sw_rx_page *sw_buf = &fp->rx_page_ring[index];
552 struct eth_rx_sge *sge = &fp->rx_sge_ring[index];
555 if (unlikely(page == NULL)) {
556 BNX2X_ERR("Can't alloc sge\n");
560 mapping = dma_map_page(&bp->pdev->dev, page, 0,
561 SGE_PAGES, DMA_FROM_DEVICE);
562 if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
563 __free_pages(page, PAGES_PER_SGE_SHIFT);
564 BNX2X_ERR("Can't map sge\n");
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));
577 static int bnx2x_fill_frag_skb(struct bnx2x *bp, struct bnx2x_fastpath *fp,
578 struct bnx2x_agg_info *tpa_info,
581 struct eth_end_agg_rx_cqe *cqe,
584 struct sw_rx_page *rx_pg, old_rx_pg;
585 u32 i, frag_len, frag_size;
586 int err, j, frag_id = 0;
587 u16 len_on_bd = tpa_info->len_on_bd;
588 u16 full_page = 0, gro_size = 0;
590 frag_size = le16_to_cpu(cqe->pkt_len) - len_on_bd;
592 if (fp->mode == TPA_MODE_GRO) {
593 gro_size = tpa_info->gro_size;
594 full_page = tpa_info->full_page;
597 /* This is needed in order to enable forwarding support */
599 bnx2x_set_gro_params(skb, tpa_info->parsing_flags, len_on_bd,
600 le16_to_cpu(cqe->pkt_len),
601 le16_to_cpu(cqe->num_of_coalesced_segs));
603 #ifdef BNX2X_STOP_ON_ERROR
604 if (pages > min_t(u32, 8, MAX_SKB_FRAGS) * SGE_PAGES) {
605 BNX2X_ERR("SGL length is too long: %d. CQE index is %d\n",
607 BNX2X_ERR("cqe->pkt_len = %d\n", cqe->pkt_len);
613 /* Run through the SGL and compose the fragmented skb */
614 for (i = 0, j = 0; i < pages; i += PAGES_PER_SGE, j++) {
615 u16 sge_idx = RX_SGE(le16_to_cpu(cqe->sgl_or_raw_data.sgl[j]));
617 /* FW gives the indices of the SGE as if the ring is an array
618 (meaning that "next" element will consume 2 indices) */
619 if (fp->mode == TPA_MODE_GRO)
620 frag_len = min_t(u32, frag_size, (u32)full_page);
622 frag_len = min_t(u32, frag_size, (u32)SGE_PAGES);
624 rx_pg = &fp->rx_page_ring[sge_idx];
627 /* If we fail to allocate a substitute page, we simply stop
628 where we are and drop the whole packet */
629 err = bnx2x_alloc_rx_sge(bp, fp, sge_idx, GFP_ATOMIC);
631 bnx2x_fp_qstats(bp, fp)->rx_skb_alloc_failed++;
635 /* Unmap the page as we're going to pass it to the stack */
636 dma_unmap_page(&bp->pdev->dev,
637 dma_unmap_addr(&old_rx_pg, mapping),
638 SGE_PAGES, DMA_FROM_DEVICE);
639 /* Add one frag and update the appropriate fields in the skb */
640 if (fp->mode == TPA_MODE_LRO)
641 skb_fill_page_desc(skb, j, old_rx_pg.page, 0, frag_len);
645 for (rem = frag_len; rem > 0; rem -= gro_size) {
646 int len = rem > gro_size ? gro_size : rem;
647 skb_fill_page_desc(skb, frag_id++,
648 old_rx_pg.page, offset, len);
650 get_page(old_rx_pg.page);
655 skb->data_len += frag_len;
656 skb->truesize += SGE_PAGES;
657 skb->len += frag_len;
659 frag_size -= frag_len;
665 static void bnx2x_frag_free(const struct bnx2x_fastpath *fp, void *data)
667 if (fp->rx_frag_size)
668 put_page(virt_to_head_page(data));
673 static void *bnx2x_frag_alloc(const struct bnx2x_fastpath *fp, gfp_t gfp_mask)
675 if (fp->rx_frag_size) {
676 /* GFP_KERNEL allocations are used only during initialization */
677 if (unlikely(gfp_mask & __GFP_WAIT))
678 return (void *)__get_free_page(gfp_mask);
680 return netdev_alloc_frag(fp->rx_frag_size);
683 return kmalloc(fp->rx_buf_size + NET_SKB_PAD, gfp_mask);
687 static void bnx2x_gro_ip_csum(struct bnx2x *bp, struct sk_buff *skb)
689 const struct iphdr *iph = ip_hdr(skb);
692 skb_set_transport_header(skb, sizeof(struct iphdr));
695 th->check = ~tcp_v4_check(skb->len - skb_transport_offset(skb),
696 iph->saddr, iph->daddr, 0);
699 static void bnx2x_gro_ipv6_csum(struct bnx2x *bp, struct sk_buff *skb)
701 struct ipv6hdr *iph = ipv6_hdr(skb);
704 skb_set_transport_header(skb, sizeof(struct ipv6hdr));
707 th->check = ~tcp_v6_check(skb->len - skb_transport_offset(skb),
708 &iph->saddr, &iph->daddr, 0);
711 static void bnx2x_gro_csum(struct bnx2x *bp, struct sk_buff *skb,
712 void (*gro_func)(struct bnx2x*, struct sk_buff*))
714 skb_set_network_header(skb, 0);
716 tcp_gro_complete(skb);
720 static void bnx2x_gro_receive(struct bnx2x *bp, struct bnx2x_fastpath *fp,
724 if (skb_shinfo(skb)->gso_size) {
725 switch (be16_to_cpu(skb->protocol)) {
727 bnx2x_gro_csum(bp, skb, bnx2x_gro_ip_csum);
730 bnx2x_gro_csum(bp, skb, bnx2x_gro_ipv6_csum);
733 BNX2X_ERR("Error: FW GRO supports only IPv4/IPv6, not 0x%04x\n",
734 be16_to_cpu(skb->protocol));
738 skb_record_rx_queue(skb, fp->rx_queue);
739 napi_gro_receive(&fp->napi, skb);
742 static void bnx2x_tpa_stop(struct bnx2x *bp, struct bnx2x_fastpath *fp,
743 struct bnx2x_agg_info *tpa_info,
745 struct eth_end_agg_rx_cqe *cqe,
748 struct sw_rx_bd *rx_buf = &tpa_info->first_buf;
749 u8 pad = tpa_info->placement_offset;
750 u16 len = tpa_info->len_on_bd;
751 struct sk_buff *skb = NULL;
752 u8 *new_data, *data = rx_buf->data;
753 u8 old_tpa_state = tpa_info->tpa_state;
755 tpa_info->tpa_state = BNX2X_TPA_STOP;
757 /* If we there was an error during the handling of the TPA_START -
758 * drop this aggregation.
760 if (old_tpa_state == BNX2X_TPA_ERROR)
763 /* Try to allocate the new data */
764 new_data = bnx2x_frag_alloc(fp, GFP_ATOMIC);
765 /* Unmap skb in the pool anyway, as we are going to change
766 pool entry status to BNX2X_TPA_STOP even if new skb allocation
768 dma_unmap_single(&bp->pdev->dev, dma_unmap_addr(rx_buf, mapping),
769 fp->rx_buf_size, DMA_FROM_DEVICE);
770 if (likely(new_data))
771 skb = build_skb(data, fp->rx_frag_size);
774 #ifdef BNX2X_STOP_ON_ERROR
775 if (pad + len > fp->rx_buf_size) {
776 BNX2X_ERR("skb_put is about to fail... pad %d len %d rx_buf_size %d\n",
777 pad, len, fp->rx_buf_size);
783 skb_reserve(skb, pad + NET_SKB_PAD);
785 skb_set_hash(skb, tpa_info->rxhash, tpa_info->rxhash_type);
787 skb->protocol = eth_type_trans(skb, bp->dev);
788 skb->ip_summed = CHECKSUM_UNNECESSARY;
790 if (!bnx2x_fill_frag_skb(bp, fp, tpa_info, pages,
791 skb, cqe, cqe_idx)) {
792 if (tpa_info->parsing_flags & PARSING_FLAGS_VLAN)
793 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), tpa_info->vlan_tag);
794 bnx2x_gro_receive(bp, fp, skb);
796 DP(NETIF_MSG_RX_STATUS,
797 "Failed to allocate new pages - dropping packet!\n");
798 dev_kfree_skb_any(skb);
801 /* put new data in bin */
802 rx_buf->data = new_data;
807 bnx2x_frag_free(fp, new_data);
809 /* drop the packet and keep the buffer in the bin */
810 DP(NETIF_MSG_RX_STATUS,
811 "Failed to allocate or map a new skb - dropping packet!\n");
812 bnx2x_fp_stats(bp, fp)->eth_q_stats.rx_skb_alloc_failed++;
815 static int bnx2x_alloc_rx_data(struct bnx2x *bp, struct bnx2x_fastpath *fp,
816 u16 index, gfp_t gfp_mask)
819 struct sw_rx_bd *rx_buf = &fp->rx_buf_ring[index];
820 struct eth_rx_bd *rx_bd = &fp->rx_desc_ring[index];
823 data = bnx2x_frag_alloc(fp, gfp_mask);
824 if (unlikely(data == NULL))
827 mapping = dma_map_single(&bp->pdev->dev, data + NET_SKB_PAD,
830 if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
831 bnx2x_frag_free(fp, data);
832 BNX2X_ERR("Can't map rx data\n");
837 dma_unmap_addr_set(rx_buf, mapping, mapping);
839 rx_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
840 rx_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
846 void bnx2x_csum_validate(struct sk_buff *skb, union eth_rx_cqe *cqe,
847 struct bnx2x_fastpath *fp,
848 struct bnx2x_eth_q_stats *qstats)
850 /* Do nothing if no L4 csum validation was done.
851 * We do not check whether IP csum was validated. For IPv4 we assume
852 * that if the card got as far as validating the L4 csum, it also
853 * validated the IP csum. IPv6 has no IP csum.
855 if (cqe->fast_path_cqe.status_flags &
856 ETH_FAST_PATH_RX_CQE_L4_XSUM_NO_VALIDATION_FLG)
859 /* If L4 validation was done, check if an error was found. */
861 if (cqe->fast_path_cqe.type_error_flags &
862 (ETH_FAST_PATH_RX_CQE_IP_BAD_XSUM_FLG |
863 ETH_FAST_PATH_RX_CQE_L4_BAD_XSUM_FLG))
864 qstats->hw_csum_err++;
866 skb->ip_summed = CHECKSUM_UNNECESSARY;
869 static int bnx2x_rx_int(struct bnx2x_fastpath *fp, int budget)
871 struct bnx2x *bp = fp->bp;
872 u16 bd_cons, bd_prod, bd_prod_fw, comp_ring_cons;
873 u16 sw_comp_cons, sw_comp_prod;
875 union eth_rx_cqe *cqe;
876 struct eth_fast_path_rx_cqe *cqe_fp;
878 #ifdef BNX2X_STOP_ON_ERROR
879 if (unlikely(bp->panic))
885 bd_cons = fp->rx_bd_cons;
886 bd_prod = fp->rx_bd_prod;
887 bd_prod_fw = bd_prod;
888 sw_comp_cons = fp->rx_comp_cons;
889 sw_comp_prod = fp->rx_comp_prod;
891 comp_ring_cons = RCQ_BD(sw_comp_cons);
892 cqe = &fp->rx_comp_ring[comp_ring_cons];
893 cqe_fp = &cqe->fast_path_cqe;
895 DP(NETIF_MSG_RX_STATUS,
896 "queue[%d]: sw_comp_cons %u\n", fp->index, sw_comp_cons);
898 while (BNX2X_IS_CQE_COMPLETED(cqe_fp)) {
899 struct sw_rx_bd *rx_buf = NULL;
902 enum eth_rx_cqe_type cqe_fp_type;
906 enum pkt_hash_types rxhash_type;
908 #ifdef BNX2X_STOP_ON_ERROR
909 if (unlikely(bp->panic))
913 bd_prod = RX_BD(bd_prod);
914 bd_cons = RX_BD(bd_cons);
916 /* A rmb() is required to ensure that the CQE is not read
917 * before it is written by the adapter DMA. PCI ordering
918 * rules will make sure the other fields are written before
919 * the marker at the end of struct eth_fast_path_rx_cqe
920 * but without rmb() a weakly ordered processor can process
921 * stale data. Without the barrier TPA state-machine might
922 * enter inconsistent state and kernel stack might be
923 * provided with incorrect packet description - these lead
924 * to various kernel crashed.
928 cqe_fp_flags = cqe_fp->type_error_flags;
929 cqe_fp_type = cqe_fp_flags & ETH_FAST_PATH_RX_CQE_TYPE;
931 DP(NETIF_MSG_RX_STATUS,
932 "CQE type %x err %x status %x queue %x vlan %x len %u\n",
933 CQE_TYPE(cqe_fp_flags),
934 cqe_fp_flags, cqe_fp->status_flags,
935 le32_to_cpu(cqe_fp->rss_hash_result),
936 le16_to_cpu(cqe_fp->vlan_tag),
937 le16_to_cpu(cqe_fp->pkt_len_or_gro_seg_len));
939 /* is this a slowpath msg? */
940 if (unlikely(CQE_TYPE_SLOW(cqe_fp_type))) {
941 bnx2x_sp_event(fp, cqe);
945 rx_buf = &fp->rx_buf_ring[bd_cons];
948 if (!CQE_TYPE_FAST(cqe_fp_type)) {
949 struct bnx2x_agg_info *tpa_info;
950 u16 frag_size, pages;
951 #ifdef BNX2X_STOP_ON_ERROR
953 if (fp->disable_tpa &&
954 (CQE_TYPE_START(cqe_fp_type) ||
955 CQE_TYPE_STOP(cqe_fp_type)))
956 BNX2X_ERR("START/STOP packet while disable_tpa type %x\n",
957 CQE_TYPE(cqe_fp_type));
960 if (CQE_TYPE_START(cqe_fp_type)) {
961 u16 queue = cqe_fp->queue_index;
962 DP(NETIF_MSG_RX_STATUS,
963 "calling tpa_start on queue %d\n",
966 bnx2x_tpa_start(fp, queue,
972 queue = cqe->end_agg_cqe.queue_index;
973 tpa_info = &fp->tpa_info[queue];
974 DP(NETIF_MSG_RX_STATUS,
975 "calling tpa_stop on queue %d\n",
978 frag_size = le16_to_cpu(cqe->end_agg_cqe.pkt_len) -
981 if (fp->mode == TPA_MODE_GRO)
982 pages = (frag_size + tpa_info->full_page - 1) /
985 pages = SGE_PAGE_ALIGN(frag_size) >>
988 bnx2x_tpa_stop(bp, fp, tpa_info, pages,
989 &cqe->end_agg_cqe, comp_ring_cons);
990 #ifdef BNX2X_STOP_ON_ERROR
995 bnx2x_update_sge_prod(fp, pages, &cqe->end_agg_cqe);
999 len = le16_to_cpu(cqe_fp->pkt_len_or_gro_seg_len);
1000 pad = cqe_fp->placement_offset;
1001 dma_sync_single_for_cpu(&bp->pdev->dev,
1002 dma_unmap_addr(rx_buf, mapping),
1003 pad + RX_COPY_THRESH,
1006 prefetch(data + pad); /* speedup eth_type_trans() */
1007 /* is this an error packet? */
1008 if (unlikely(cqe_fp_flags & ETH_RX_ERROR_FALGS)) {
1009 DP(NETIF_MSG_RX_ERR | NETIF_MSG_RX_STATUS,
1010 "ERROR flags %x rx packet %u\n",
1011 cqe_fp_flags, sw_comp_cons);
1012 bnx2x_fp_qstats(bp, fp)->rx_err_discard_pkt++;
1016 /* Since we don't have a jumbo ring
1017 * copy small packets if mtu > 1500
1019 if ((bp->dev->mtu > ETH_MAX_PACKET_SIZE) &&
1020 (len <= RX_COPY_THRESH)) {
1021 skb = netdev_alloc_skb_ip_align(bp->dev, len);
1023 DP(NETIF_MSG_RX_ERR | NETIF_MSG_RX_STATUS,
1024 "ERROR packet dropped because of alloc failure\n");
1025 bnx2x_fp_qstats(bp, fp)->rx_skb_alloc_failed++;
1028 memcpy(skb->data, data + pad, len);
1029 bnx2x_reuse_rx_data(fp, bd_cons, bd_prod);
1031 if (likely(bnx2x_alloc_rx_data(bp, fp, bd_prod,
1032 GFP_ATOMIC) == 0)) {
1033 dma_unmap_single(&bp->pdev->dev,
1034 dma_unmap_addr(rx_buf, mapping),
1037 skb = build_skb(data, fp->rx_frag_size);
1038 if (unlikely(!skb)) {
1039 bnx2x_frag_free(fp, data);
1040 bnx2x_fp_qstats(bp, fp)->
1041 rx_skb_alloc_failed++;
1044 skb_reserve(skb, pad);
1046 DP(NETIF_MSG_RX_ERR | NETIF_MSG_RX_STATUS,
1047 "ERROR packet dropped because of alloc failure\n");
1048 bnx2x_fp_qstats(bp, fp)->rx_skb_alloc_failed++;
1050 bnx2x_reuse_rx_data(fp, bd_cons, bd_prod);
1056 skb->protocol = eth_type_trans(skb, bp->dev);
1058 /* Set Toeplitz hash for a none-LRO skb */
1059 rxhash = bnx2x_get_rxhash(bp, cqe_fp, &rxhash_type);
1060 skb_set_hash(skb, rxhash, rxhash_type);
1062 skb_checksum_none_assert(skb);
1064 if (bp->dev->features & NETIF_F_RXCSUM)
1065 bnx2x_csum_validate(skb, cqe, fp,
1066 bnx2x_fp_qstats(bp, fp));
1068 skb_record_rx_queue(skb, fp->rx_queue);
1070 if (le16_to_cpu(cqe_fp->pars_flags.flags) &
1072 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
1073 le16_to_cpu(cqe_fp->vlan_tag));
1075 skb_mark_napi_id(skb, &fp->napi);
1077 if (bnx2x_fp_ll_polling(fp))
1078 netif_receive_skb(skb);
1080 napi_gro_receive(&fp->napi, skb);
1082 rx_buf->data = NULL;
1084 bd_cons = NEXT_RX_IDX(bd_cons);
1085 bd_prod = NEXT_RX_IDX(bd_prod);
1086 bd_prod_fw = NEXT_RX_IDX(bd_prod_fw);
1089 sw_comp_prod = NEXT_RCQ_IDX(sw_comp_prod);
1090 sw_comp_cons = NEXT_RCQ_IDX(sw_comp_cons);
1092 /* mark CQE as free */
1093 BNX2X_SEED_CQE(cqe_fp);
1095 if (rx_pkt == budget)
1098 comp_ring_cons = RCQ_BD(sw_comp_cons);
1099 cqe = &fp->rx_comp_ring[comp_ring_cons];
1100 cqe_fp = &cqe->fast_path_cqe;
1103 fp->rx_bd_cons = bd_cons;
1104 fp->rx_bd_prod = bd_prod_fw;
1105 fp->rx_comp_cons = sw_comp_cons;
1106 fp->rx_comp_prod = sw_comp_prod;
1108 /* Update producers */
1109 bnx2x_update_rx_prod(bp, fp, bd_prod_fw, sw_comp_prod,
1112 fp->rx_pkt += rx_pkt;
1118 static irqreturn_t bnx2x_msix_fp_int(int irq, void *fp_cookie)
1120 struct bnx2x_fastpath *fp = fp_cookie;
1121 struct bnx2x *bp = fp->bp;
1125 "got an MSI-X interrupt on IDX:SB [fp %d fw_sd %d igusb %d]\n",
1126 fp->index, fp->fw_sb_id, fp->igu_sb_id);
1128 bnx2x_ack_sb(bp, fp->igu_sb_id, USTORM_ID, 0, IGU_INT_DISABLE, 0);
1130 #ifdef BNX2X_STOP_ON_ERROR
1131 if (unlikely(bp->panic))
1135 /* Handle Rx and Tx according to MSI-X vector */
1136 for_each_cos_in_tx_queue(fp, cos)
1137 prefetch(fp->txdata_ptr[cos]->tx_cons_sb);
1139 prefetch(&fp->sb_running_index[SM_RX_ID]);
1140 napi_schedule(&bnx2x_fp(bp, fp->index, napi));
1145 /* HW Lock for shared dual port PHYs */
1146 void bnx2x_acquire_phy_lock(struct bnx2x *bp)
1148 mutex_lock(&bp->port.phy_mutex);
1150 bnx2x_acquire_hw_lock(bp, HW_LOCK_RESOURCE_MDIO);
1153 void bnx2x_release_phy_lock(struct bnx2x *bp)
1155 bnx2x_release_hw_lock(bp, HW_LOCK_RESOURCE_MDIO);
1157 mutex_unlock(&bp->port.phy_mutex);
1160 /* calculates MF speed according to current linespeed and MF configuration */
1161 u16 bnx2x_get_mf_speed(struct bnx2x *bp)
1163 u16 line_speed = bp->link_vars.line_speed;
1165 u16 maxCfg = bnx2x_extract_max_cfg(bp,
1166 bp->mf_config[BP_VN(bp)]);
1168 /* Calculate the current MAX line speed limit for the MF
1172 line_speed = (line_speed * maxCfg) / 100;
1173 else { /* SD mode */
1174 u16 vn_max_rate = maxCfg * 100;
1176 if (vn_max_rate < line_speed)
1177 line_speed = vn_max_rate;
1185 * bnx2x_fill_report_data - fill link report data to report
1187 * @bp: driver handle
1188 * @data: link state to update
1190 * It uses a none-atomic bit operations because is called under the mutex.
1192 static void bnx2x_fill_report_data(struct bnx2x *bp,
1193 struct bnx2x_link_report_data *data)
1195 u16 line_speed = bnx2x_get_mf_speed(bp);
1197 memset(data, 0, sizeof(*data));
1199 /* Fill the report data: effective line speed */
1200 data->line_speed = line_speed;
1203 if (!bp->link_vars.link_up || (bp->flags & MF_FUNC_DIS))
1204 __set_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1205 &data->link_report_flags);
1208 if (bp->link_vars.duplex == DUPLEX_FULL)
1209 __set_bit(BNX2X_LINK_REPORT_FD, &data->link_report_flags);
1211 /* Rx Flow Control is ON */
1212 if (bp->link_vars.flow_ctrl & BNX2X_FLOW_CTRL_RX)
1213 __set_bit(BNX2X_LINK_REPORT_RX_FC_ON, &data->link_report_flags);
1215 /* Tx Flow Control is ON */
1216 if (bp->link_vars.flow_ctrl & BNX2X_FLOW_CTRL_TX)
1217 __set_bit(BNX2X_LINK_REPORT_TX_FC_ON, &data->link_report_flags);
1221 * bnx2x_link_report - report link status to OS.
1223 * @bp: driver handle
1225 * Calls the __bnx2x_link_report() under the same locking scheme
1226 * as a link/PHY state managing code to ensure a consistent link
1230 void bnx2x_link_report(struct bnx2x *bp)
1232 bnx2x_acquire_phy_lock(bp);
1233 __bnx2x_link_report(bp);
1234 bnx2x_release_phy_lock(bp);
1238 * __bnx2x_link_report - report link status to OS.
1240 * @bp: driver handle
1242 * None atomic implementation.
1243 * Should be called under the phy_lock.
1245 void __bnx2x_link_report(struct bnx2x *bp)
1247 struct bnx2x_link_report_data cur_data;
1250 if (IS_PF(bp) && !CHIP_IS_E1(bp))
1251 bnx2x_read_mf_cfg(bp);
1253 /* Read the current link report info */
1254 bnx2x_fill_report_data(bp, &cur_data);
1256 /* Don't report link down or exactly the same link status twice */
1257 if (!memcmp(&cur_data, &bp->last_reported_link, sizeof(cur_data)) ||
1258 (test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1259 &bp->last_reported_link.link_report_flags) &&
1260 test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1261 &cur_data.link_report_flags)))
1266 /* We are going to report a new link parameters now -
1267 * remember the current data for the next time.
1269 memcpy(&bp->last_reported_link, &cur_data, sizeof(cur_data));
1271 if (test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1272 &cur_data.link_report_flags)) {
1273 netif_carrier_off(bp->dev);
1274 netdev_err(bp->dev, "NIC Link is Down\n");
1280 netif_carrier_on(bp->dev);
1282 if (test_and_clear_bit(BNX2X_LINK_REPORT_FD,
1283 &cur_data.link_report_flags))
1288 /* Handle the FC at the end so that only these flags would be
1289 * possibly set. This way we may easily check if there is no FC
1292 if (cur_data.link_report_flags) {
1293 if (test_bit(BNX2X_LINK_REPORT_RX_FC_ON,
1294 &cur_data.link_report_flags)) {
1295 if (test_bit(BNX2X_LINK_REPORT_TX_FC_ON,
1296 &cur_data.link_report_flags))
1297 flow = "ON - receive & transmit";
1299 flow = "ON - receive";
1301 flow = "ON - transmit";
1306 netdev_info(bp->dev, "NIC Link is Up, %d Mbps %s duplex, Flow control: %s\n",
1307 cur_data.line_speed, duplex, flow);
1311 static void bnx2x_set_next_page_sgl(struct bnx2x_fastpath *fp)
1315 for (i = 1; i <= NUM_RX_SGE_PAGES; i++) {
1316 struct eth_rx_sge *sge;
1318 sge = &fp->rx_sge_ring[RX_SGE_CNT * i - 2];
1320 cpu_to_le32(U64_HI(fp->rx_sge_mapping +
1321 BCM_PAGE_SIZE*(i % NUM_RX_SGE_PAGES)));
1324 cpu_to_le32(U64_LO(fp->rx_sge_mapping +
1325 BCM_PAGE_SIZE*(i % NUM_RX_SGE_PAGES)));
1329 static void bnx2x_free_tpa_pool(struct bnx2x *bp,
1330 struct bnx2x_fastpath *fp, int last)
1334 for (i = 0; i < last; i++) {
1335 struct bnx2x_agg_info *tpa_info = &fp->tpa_info[i];
1336 struct sw_rx_bd *first_buf = &tpa_info->first_buf;
1337 u8 *data = first_buf->data;
1340 DP(NETIF_MSG_IFDOWN, "tpa bin %d empty on free\n", i);
1343 if (tpa_info->tpa_state == BNX2X_TPA_START)
1344 dma_unmap_single(&bp->pdev->dev,
1345 dma_unmap_addr(first_buf, mapping),
1346 fp->rx_buf_size, DMA_FROM_DEVICE);
1347 bnx2x_frag_free(fp, data);
1348 first_buf->data = NULL;
1352 void bnx2x_init_rx_rings_cnic(struct bnx2x *bp)
1356 for_each_rx_queue_cnic(bp, j) {
1357 struct bnx2x_fastpath *fp = &bp->fp[j];
1361 /* Activate BD ring */
1363 * this will generate an interrupt (to the TSTORM)
1364 * must only be done after chip is initialized
1366 bnx2x_update_rx_prod(bp, fp, fp->rx_bd_prod, fp->rx_comp_prod,
1371 void bnx2x_init_rx_rings(struct bnx2x *bp)
1373 int func = BP_FUNC(bp);
1377 /* Allocate TPA resources */
1378 for_each_eth_queue(bp, j) {
1379 struct bnx2x_fastpath *fp = &bp->fp[j];
1382 "mtu %d rx_buf_size %d\n", bp->dev->mtu, fp->rx_buf_size);
1384 if (!fp->disable_tpa) {
1385 /* Fill the per-aggregation pool */
1386 for (i = 0; i < MAX_AGG_QS(bp); i++) {
1387 struct bnx2x_agg_info *tpa_info =
1389 struct sw_rx_bd *first_buf =
1390 &tpa_info->first_buf;
1393 bnx2x_frag_alloc(fp, GFP_KERNEL);
1394 if (!first_buf->data) {
1395 BNX2X_ERR("Failed to allocate TPA skb pool for queue[%d] - disabling TPA on this queue!\n",
1397 bnx2x_free_tpa_pool(bp, fp, i);
1398 fp->disable_tpa = 1;
1401 dma_unmap_addr_set(first_buf, mapping, 0);
1402 tpa_info->tpa_state = BNX2X_TPA_STOP;
1405 /* "next page" elements initialization */
1406 bnx2x_set_next_page_sgl(fp);
1408 /* set SGEs bit mask */
1409 bnx2x_init_sge_ring_bit_mask(fp);
1411 /* Allocate SGEs and initialize the ring elements */
1412 for (i = 0, ring_prod = 0;
1413 i < MAX_RX_SGE_CNT*NUM_RX_SGE_PAGES; i++) {
1415 if (bnx2x_alloc_rx_sge(bp, fp, ring_prod,
1417 BNX2X_ERR("was only able to allocate %d rx sges\n",
1419 BNX2X_ERR("disabling TPA for queue[%d]\n",
1421 /* Cleanup already allocated elements */
1422 bnx2x_free_rx_sge_range(bp, fp,
1424 bnx2x_free_tpa_pool(bp, fp,
1426 fp->disable_tpa = 1;
1430 ring_prod = NEXT_SGE_IDX(ring_prod);
1433 fp->rx_sge_prod = ring_prod;
1437 for_each_eth_queue(bp, j) {
1438 struct bnx2x_fastpath *fp = &bp->fp[j];
1442 /* Activate BD ring */
1444 * this will generate an interrupt (to the TSTORM)
1445 * must only be done after chip is initialized
1447 bnx2x_update_rx_prod(bp, fp, fp->rx_bd_prod, fp->rx_comp_prod,
1453 if (CHIP_IS_E1(bp)) {
1454 REG_WR(bp, BAR_USTRORM_INTMEM +
1455 USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func),
1456 U64_LO(fp->rx_comp_mapping));
1457 REG_WR(bp, BAR_USTRORM_INTMEM +
1458 USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func) + 4,
1459 U64_HI(fp->rx_comp_mapping));
1464 static void bnx2x_free_tx_skbs_queue(struct bnx2x_fastpath *fp)
1467 struct bnx2x *bp = fp->bp;
1469 for_each_cos_in_tx_queue(fp, cos) {
1470 struct bnx2x_fp_txdata *txdata = fp->txdata_ptr[cos];
1471 unsigned pkts_compl = 0, bytes_compl = 0;
1473 u16 sw_prod = txdata->tx_pkt_prod;
1474 u16 sw_cons = txdata->tx_pkt_cons;
1476 while (sw_cons != sw_prod) {
1477 bnx2x_free_tx_pkt(bp, txdata, TX_BD(sw_cons),
1478 &pkts_compl, &bytes_compl);
1482 netdev_tx_reset_queue(
1483 netdev_get_tx_queue(bp->dev,
1484 txdata->txq_index));
1488 static void bnx2x_free_tx_skbs_cnic(struct bnx2x *bp)
1492 for_each_tx_queue_cnic(bp, i) {
1493 bnx2x_free_tx_skbs_queue(&bp->fp[i]);
1497 static void bnx2x_free_tx_skbs(struct bnx2x *bp)
1501 for_each_eth_queue(bp, i) {
1502 bnx2x_free_tx_skbs_queue(&bp->fp[i]);
1506 static void bnx2x_free_rx_bds(struct bnx2x_fastpath *fp)
1508 struct bnx2x *bp = fp->bp;
1511 /* ring wasn't allocated */
1512 if (fp->rx_buf_ring == NULL)
1515 for (i = 0; i < NUM_RX_BD; i++) {
1516 struct sw_rx_bd *rx_buf = &fp->rx_buf_ring[i];
1517 u8 *data = rx_buf->data;
1521 dma_unmap_single(&bp->pdev->dev,
1522 dma_unmap_addr(rx_buf, mapping),
1523 fp->rx_buf_size, DMA_FROM_DEVICE);
1525 rx_buf->data = NULL;
1526 bnx2x_frag_free(fp, data);
1530 static void bnx2x_free_rx_skbs_cnic(struct bnx2x *bp)
1534 for_each_rx_queue_cnic(bp, j) {
1535 bnx2x_free_rx_bds(&bp->fp[j]);
1539 static void bnx2x_free_rx_skbs(struct bnx2x *bp)
1543 for_each_eth_queue(bp, j) {
1544 struct bnx2x_fastpath *fp = &bp->fp[j];
1546 bnx2x_free_rx_bds(fp);
1548 if (!fp->disable_tpa)
1549 bnx2x_free_tpa_pool(bp, fp, MAX_AGG_QS(bp));
1553 static void bnx2x_free_skbs_cnic(struct bnx2x *bp)
1555 bnx2x_free_tx_skbs_cnic(bp);
1556 bnx2x_free_rx_skbs_cnic(bp);
1559 void bnx2x_free_skbs(struct bnx2x *bp)
1561 bnx2x_free_tx_skbs(bp);
1562 bnx2x_free_rx_skbs(bp);
1565 void bnx2x_update_max_mf_config(struct bnx2x *bp, u32 value)
1567 /* load old values */
1568 u32 mf_cfg = bp->mf_config[BP_VN(bp)];
1570 if (value != bnx2x_extract_max_cfg(bp, mf_cfg)) {
1571 /* leave all but MAX value */
1572 mf_cfg &= ~FUNC_MF_CFG_MAX_BW_MASK;
1574 /* set new MAX value */
1575 mf_cfg |= (value << FUNC_MF_CFG_MAX_BW_SHIFT)
1576 & FUNC_MF_CFG_MAX_BW_MASK;
1578 bnx2x_fw_command(bp, DRV_MSG_CODE_SET_MF_BW, mf_cfg);
1583 * bnx2x_free_msix_irqs - free previously requested MSI-X IRQ vectors
1585 * @bp: driver handle
1586 * @nvecs: number of vectors to be released
1588 static void bnx2x_free_msix_irqs(struct bnx2x *bp, int nvecs)
1592 if (nvecs == offset)
1595 /* VFs don't have a default SB */
1597 free_irq(bp->msix_table[offset].vector, bp->dev);
1598 DP(NETIF_MSG_IFDOWN, "released sp irq (%d)\n",
1599 bp->msix_table[offset].vector);
1603 if (CNIC_SUPPORT(bp)) {
1604 if (nvecs == offset)
1609 for_each_eth_queue(bp, i) {
1610 if (nvecs == offset)
1612 DP(NETIF_MSG_IFDOWN, "about to release fp #%d->%d irq\n",
1613 i, bp->msix_table[offset].vector);
1615 free_irq(bp->msix_table[offset++].vector, &bp->fp[i]);
1619 void bnx2x_free_irq(struct bnx2x *bp)
1621 if (bp->flags & USING_MSIX_FLAG &&
1622 !(bp->flags & USING_SINGLE_MSIX_FLAG)) {
1623 int nvecs = BNX2X_NUM_ETH_QUEUES(bp) + CNIC_SUPPORT(bp);
1625 /* vfs don't have a default status block */
1629 bnx2x_free_msix_irqs(bp, nvecs);
1631 free_irq(bp->dev->irq, bp->dev);
1635 int bnx2x_enable_msix(struct bnx2x *bp)
1637 int msix_vec = 0, i, rc;
1639 /* VFs don't have a default status block */
1641 bp->msix_table[msix_vec].entry = msix_vec;
1642 BNX2X_DEV_INFO("msix_table[0].entry = %d (slowpath)\n",
1643 bp->msix_table[0].entry);
1647 /* Cnic requires an msix vector for itself */
1648 if (CNIC_SUPPORT(bp)) {
1649 bp->msix_table[msix_vec].entry = msix_vec;
1650 BNX2X_DEV_INFO("msix_table[%d].entry = %d (CNIC)\n",
1651 msix_vec, bp->msix_table[msix_vec].entry);
1655 /* We need separate vectors for ETH queues only (not FCoE) */
1656 for_each_eth_queue(bp, i) {
1657 bp->msix_table[msix_vec].entry = msix_vec;
1658 BNX2X_DEV_INFO("msix_table[%d].entry = %d (fastpath #%u)\n",
1659 msix_vec, msix_vec, i);
1663 DP(BNX2X_MSG_SP, "about to request enable msix with %d vectors\n",
1666 rc = pci_enable_msix_range(bp->pdev, &bp->msix_table[0],
1667 BNX2X_MIN_MSIX_VEC_CNT(bp), msix_vec);
1669 * reconfigure number of tx/rx queues according to available
1672 if (rc == -ENOSPC) {
1673 /* Get by with single vector */
1674 rc = pci_enable_msix_range(bp->pdev, &bp->msix_table[0], 1, 1);
1676 BNX2X_DEV_INFO("Single MSI-X is not attainable rc %d\n",
1681 BNX2X_DEV_INFO("Using single MSI-X vector\n");
1682 bp->flags |= USING_SINGLE_MSIX_FLAG;
1684 BNX2X_DEV_INFO("set number of queues to 1\n");
1685 bp->num_ethernet_queues = 1;
1686 bp->num_queues = bp->num_ethernet_queues + bp->num_cnic_queues;
1687 } else if (rc < 0) {
1688 BNX2X_DEV_INFO("MSI-X is not attainable rc %d\n", rc);
1690 } else if (rc < msix_vec) {
1691 /* how less vectors we will have? */
1692 int diff = msix_vec - rc;
1694 BNX2X_DEV_INFO("Trying to use less MSI-X vectors: %d\n", rc);
1697 * decrease number of queues by number of unallocated entries
1699 bp->num_ethernet_queues -= diff;
1700 bp->num_queues = bp->num_ethernet_queues + bp->num_cnic_queues;
1702 BNX2X_DEV_INFO("New queue configuration set: %d\n",
1706 bp->flags |= USING_MSIX_FLAG;
1711 /* fall to INTx if not enough memory */
1713 bp->flags |= DISABLE_MSI_FLAG;
1718 static int bnx2x_req_msix_irqs(struct bnx2x *bp)
1720 int i, rc, offset = 0;
1722 /* no default status block for vf */
1724 rc = request_irq(bp->msix_table[offset++].vector,
1725 bnx2x_msix_sp_int, 0,
1726 bp->dev->name, bp->dev);
1728 BNX2X_ERR("request sp irq failed\n");
1733 if (CNIC_SUPPORT(bp))
1736 for_each_eth_queue(bp, i) {
1737 struct bnx2x_fastpath *fp = &bp->fp[i];
1738 snprintf(fp->name, sizeof(fp->name), "%s-fp-%d",
1741 rc = request_irq(bp->msix_table[offset].vector,
1742 bnx2x_msix_fp_int, 0, fp->name, fp);
1744 BNX2X_ERR("request fp #%d irq (%d) failed rc %d\n", i,
1745 bp->msix_table[offset].vector, rc);
1746 bnx2x_free_msix_irqs(bp, offset);
1753 i = BNX2X_NUM_ETH_QUEUES(bp);
1755 offset = 1 + CNIC_SUPPORT(bp);
1756 netdev_info(bp->dev,
1757 "using MSI-X IRQs: sp %d fp[%d] %d ... fp[%d] %d\n",
1758 bp->msix_table[0].vector,
1759 0, bp->msix_table[offset].vector,
1760 i - 1, bp->msix_table[offset + i - 1].vector);
1762 offset = CNIC_SUPPORT(bp);
1763 netdev_info(bp->dev,
1764 "using MSI-X IRQs: fp[%d] %d ... fp[%d] %d\n",
1765 0, bp->msix_table[offset].vector,
1766 i - 1, bp->msix_table[offset + i - 1].vector);
1771 int bnx2x_enable_msi(struct bnx2x *bp)
1775 rc = pci_enable_msi(bp->pdev);
1777 BNX2X_DEV_INFO("MSI is not attainable\n");
1780 bp->flags |= USING_MSI_FLAG;
1785 static int bnx2x_req_irq(struct bnx2x *bp)
1787 unsigned long flags;
1790 if (bp->flags & (USING_MSI_FLAG | USING_MSIX_FLAG))
1793 flags = IRQF_SHARED;
1795 if (bp->flags & USING_MSIX_FLAG)
1796 irq = bp->msix_table[0].vector;
1798 irq = bp->pdev->irq;
1800 return request_irq(irq, bnx2x_interrupt, flags, bp->dev->name, bp->dev);
1803 static int bnx2x_setup_irqs(struct bnx2x *bp)
1806 if (bp->flags & USING_MSIX_FLAG &&
1807 !(bp->flags & USING_SINGLE_MSIX_FLAG)) {
1808 rc = bnx2x_req_msix_irqs(bp);
1812 rc = bnx2x_req_irq(bp);
1814 BNX2X_ERR("IRQ request failed rc %d, aborting\n", rc);
1817 if (bp->flags & USING_MSI_FLAG) {
1818 bp->dev->irq = bp->pdev->irq;
1819 netdev_info(bp->dev, "using MSI IRQ %d\n",
1822 if (bp->flags & USING_MSIX_FLAG) {
1823 bp->dev->irq = bp->msix_table[0].vector;
1824 netdev_info(bp->dev, "using MSIX IRQ %d\n",
1832 static void bnx2x_napi_enable_cnic(struct bnx2x *bp)
1836 for_each_rx_queue_cnic(bp, i) {
1837 bnx2x_fp_init_lock(&bp->fp[i]);
1838 napi_enable(&bnx2x_fp(bp, i, napi));
1842 static void bnx2x_napi_enable(struct bnx2x *bp)
1846 for_each_eth_queue(bp, i) {
1847 bnx2x_fp_init_lock(&bp->fp[i]);
1848 napi_enable(&bnx2x_fp(bp, i, napi));
1852 static void bnx2x_napi_disable_cnic(struct bnx2x *bp)
1856 for_each_rx_queue_cnic(bp, i) {
1857 napi_disable(&bnx2x_fp(bp, i, napi));
1858 while (!bnx2x_fp_ll_disable(&bp->fp[i]))
1859 usleep_range(1000, 2000);
1863 static void bnx2x_napi_disable(struct bnx2x *bp)
1867 for_each_eth_queue(bp, i) {
1868 napi_disable(&bnx2x_fp(bp, i, napi));
1869 while (!bnx2x_fp_ll_disable(&bp->fp[i]))
1870 usleep_range(1000, 2000);
1874 void bnx2x_netif_start(struct bnx2x *bp)
1876 if (netif_running(bp->dev)) {
1877 bnx2x_napi_enable(bp);
1878 if (CNIC_LOADED(bp))
1879 bnx2x_napi_enable_cnic(bp);
1880 bnx2x_int_enable(bp);
1881 if (bp->state == BNX2X_STATE_OPEN)
1882 netif_tx_wake_all_queues(bp->dev);
1886 void bnx2x_netif_stop(struct bnx2x *bp, int disable_hw)
1888 bnx2x_int_disable_sync(bp, disable_hw);
1889 bnx2x_napi_disable(bp);
1890 if (CNIC_LOADED(bp))
1891 bnx2x_napi_disable_cnic(bp);
1894 u16 bnx2x_select_queue(struct net_device *dev, struct sk_buff *skb,
1895 void *accel_priv, select_queue_fallback_t fallback)
1897 struct bnx2x *bp = netdev_priv(dev);
1899 if (CNIC_LOADED(bp) && !NO_FCOE(bp)) {
1900 struct ethhdr *hdr = (struct ethhdr *)skb->data;
1901 u16 ether_type = ntohs(hdr->h_proto);
1903 /* Skip VLAN tag if present */
1904 if (ether_type == ETH_P_8021Q) {
1905 struct vlan_ethhdr *vhdr =
1906 (struct vlan_ethhdr *)skb->data;
1908 ether_type = ntohs(vhdr->h_vlan_encapsulated_proto);
1911 /* If ethertype is FCoE or FIP - use FCoE ring */
1912 if ((ether_type == ETH_P_FCOE) || (ether_type == ETH_P_FIP))
1913 return bnx2x_fcoe_tx(bp, txq_index);
1916 /* select a non-FCoE queue */
1917 return fallback(dev, skb) % BNX2X_NUM_ETH_QUEUES(bp);
1920 void bnx2x_set_num_queues(struct bnx2x *bp)
1923 bp->num_ethernet_queues = bnx2x_calc_num_queues(bp);
1925 /* override in STORAGE SD modes */
1926 if (IS_MF_STORAGE_SD(bp) || IS_MF_FCOE_AFEX(bp))
1927 bp->num_ethernet_queues = 1;
1929 /* Add special queues */
1930 bp->num_cnic_queues = CNIC_SUPPORT(bp); /* For FCOE */
1931 bp->num_queues = bp->num_ethernet_queues + bp->num_cnic_queues;
1933 BNX2X_DEV_INFO("set number of queues to %d\n", bp->num_queues);
1937 * bnx2x_set_real_num_queues - configure netdev->real_num_[tx,rx]_queues
1939 * @bp: Driver handle
1941 * We currently support for at most 16 Tx queues for each CoS thus we will
1942 * allocate a multiple of 16 for ETH L2 rings according to the value of the
1945 * If there is an FCoE L2 queue the appropriate Tx queue will have the next
1946 * index after all ETH L2 indices.
1948 * If the actual number of Tx queues (for each CoS) is less than 16 then there
1949 * will be the holes at the end of each group of 16 ETh L2 indices (0..15,
1950 * 16..31,...) with indices that are not coupled with any real Tx queue.
1952 * The proper configuration of skb->queue_mapping is handled by
1953 * bnx2x_select_queue() and __skb_tx_hash().
1955 * bnx2x_setup_tc() takes care of the proper TC mappings so that __skb_tx_hash()
1956 * will return a proper Tx index if TC is enabled (netdev->num_tc > 0).
1958 static int bnx2x_set_real_num_queues(struct bnx2x *bp, int include_cnic)
1962 tx = BNX2X_NUM_ETH_QUEUES(bp) * bp->max_cos;
1963 rx = BNX2X_NUM_ETH_QUEUES(bp);
1965 /* account for fcoe queue */
1966 if (include_cnic && !NO_FCOE(bp)) {
1971 rc = netif_set_real_num_tx_queues(bp->dev, tx);
1973 BNX2X_ERR("Failed to set real number of Tx queues: %d\n", rc);
1976 rc = netif_set_real_num_rx_queues(bp->dev, rx);
1978 BNX2X_ERR("Failed to set real number of Rx queues: %d\n", rc);
1982 DP(NETIF_MSG_IFUP, "Setting real num queues to (tx, rx) (%d, %d)\n",
1988 static void bnx2x_set_rx_buf_size(struct bnx2x *bp)
1992 for_each_queue(bp, i) {
1993 struct bnx2x_fastpath *fp = &bp->fp[i];
1996 /* Always use a mini-jumbo MTU for the FCoE L2 ring */
1999 * Although there are no IP frames expected to arrive to
2000 * this ring we still want to add an
2001 * IP_HEADER_ALIGNMENT_PADDING to prevent a buffer
2004 mtu = BNX2X_FCOE_MINI_JUMBO_MTU;
2007 fp->rx_buf_size = BNX2X_FW_RX_ALIGN_START +
2008 IP_HEADER_ALIGNMENT_PADDING +
2011 BNX2X_FW_RX_ALIGN_END;
2012 /* Note : rx_buf_size doesn't take into account NET_SKB_PAD */
2013 if (fp->rx_buf_size + NET_SKB_PAD <= PAGE_SIZE)
2014 fp->rx_frag_size = fp->rx_buf_size + NET_SKB_PAD;
2016 fp->rx_frag_size = 0;
2020 static int bnx2x_init_rss(struct bnx2x *bp)
2023 u8 num_eth_queues = BNX2X_NUM_ETH_QUEUES(bp);
2025 /* Prepare the initial contents for the indirection table if RSS is
2028 for (i = 0; i < sizeof(bp->rss_conf_obj.ind_table); i++)
2029 bp->rss_conf_obj.ind_table[i] =
2031 ethtool_rxfh_indir_default(i, num_eth_queues);
2034 * For 57710 and 57711 SEARCHER configuration (rss_keys) is
2035 * per-port, so if explicit configuration is needed , do it only
2038 * For 57712 and newer on the other hand it's a per-function
2041 return bnx2x_config_rss_eth(bp, bp->port.pmf || !CHIP_IS_E1x(bp));
2044 int bnx2x_rss(struct bnx2x *bp, struct bnx2x_rss_config_obj *rss_obj,
2045 bool config_hash, bool enable)
2047 struct bnx2x_config_rss_params params = {NULL};
2049 /* Although RSS is meaningless when there is a single HW queue we
2050 * still need it enabled in order to have HW Rx hash generated.
2052 * if (!is_eth_multi(bp))
2053 * bp->multi_mode = ETH_RSS_MODE_DISABLED;
2056 params.rss_obj = rss_obj;
2058 __set_bit(RAMROD_COMP_WAIT, ¶ms.ramrod_flags);
2061 __set_bit(BNX2X_RSS_MODE_REGULAR, ¶ms.rss_flags);
2063 /* RSS configuration */
2064 __set_bit(BNX2X_RSS_IPV4, ¶ms.rss_flags);
2065 __set_bit(BNX2X_RSS_IPV4_TCP, ¶ms.rss_flags);
2066 __set_bit(BNX2X_RSS_IPV6, ¶ms.rss_flags);
2067 __set_bit(BNX2X_RSS_IPV6_TCP, ¶ms.rss_flags);
2068 if (rss_obj->udp_rss_v4)
2069 __set_bit(BNX2X_RSS_IPV4_UDP, ¶ms.rss_flags);
2070 if (rss_obj->udp_rss_v6)
2071 __set_bit(BNX2X_RSS_IPV6_UDP, ¶ms.rss_flags);
2073 __set_bit(BNX2X_RSS_MODE_DISABLED, ¶ms.rss_flags);
2077 params.rss_result_mask = MULTI_MASK;
2079 memcpy(params.ind_table, rss_obj->ind_table, sizeof(params.ind_table));
2083 prandom_bytes(params.rss_key, T_ETH_RSS_KEY * 4);
2084 __set_bit(BNX2X_RSS_SET_SRCH, ¶ms.rss_flags);
2088 return bnx2x_config_rss(bp, ¶ms);
2090 return bnx2x_vfpf_config_rss(bp, ¶ms);
2093 static int bnx2x_init_hw(struct bnx2x *bp, u32 load_code)
2095 struct bnx2x_func_state_params func_params = {NULL};
2097 /* Prepare parameters for function state transitions */
2098 __set_bit(RAMROD_COMP_WAIT, &func_params.ramrod_flags);
2100 func_params.f_obj = &bp->func_obj;
2101 func_params.cmd = BNX2X_F_CMD_HW_INIT;
2103 func_params.params.hw_init.load_phase = load_code;
2105 return bnx2x_func_state_change(bp, &func_params);
2109 * Cleans the object that have internal lists without sending
2110 * ramrods. Should be run when interrupts are disabled.
2112 void bnx2x_squeeze_objects(struct bnx2x *bp)
2115 unsigned long ramrod_flags = 0, vlan_mac_flags = 0;
2116 struct bnx2x_mcast_ramrod_params rparam = {NULL};
2117 struct bnx2x_vlan_mac_obj *mac_obj = &bp->sp_objs->mac_obj;
2119 /***************** Cleanup MACs' object first *************************/
2121 /* Wait for completion of requested */
2122 __set_bit(RAMROD_COMP_WAIT, &ramrod_flags);
2123 /* Perform a dry cleanup */
2124 __set_bit(RAMROD_DRV_CLR_ONLY, &ramrod_flags);
2126 /* Clean ETH primary MAC */
2127 __set_bit(BNX2X_ETH_MAC, &vlan_mac_flags);
2128 rc = mac_obj->delete_all(bp, &bp->sp_objs->mac_obj, &vlan_mac_flags,
2131 BNX2X_ERR("Failed to clean ETH MACs: %d\n", rc);
2133 /* Cleanup UC list */
2135 __set_bit(BNX2X_UC_LIST_MAC, &vlan_mac_flags);
2136 rc = mac_obj->delete_all(bp, mac_obj, &vlan_mac_flags,
2139 BNX2X_ERR("Failed to clean UC list MACs: %d\n", rc);
2141 /***************** Now clean mcast object *****************************/
2142 rparam.mcast_obj = &bp->mcast_obj;
2143 __set_bit(RAMROD_DRV_CLR_ONLY, &rparam.ramrod_flags);
2145 /* Add a DEL command... - Since we're doing a driver cleanup only,
2146 * we take a lock surrounding both the initial send and the CONTs,
2147 * as we don't want a true completion to disrupt us in the middle.
2149 netif_addr_lock_bh(bp->dev);
2150 rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_DEL);
2152 BNX2X_ERR("Failed to add a new DEL command to a multi-cast object: %d\n",
2155 /* ...and wait until all pending commands are cleared */
2156 rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_CONT);
2159 BNX2X_ERR("Failed to clean multi-cast object: %d\n",
2161 netif_addr_unlock_bh(bp->dev);
2165 rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_CONT);
2167 netif_addr_unlock_bh(bp->dev);
2170 #ifndef BNX2X_STOP_ON_ERROR
2171 #define LOAD_ERROR_EXIT(bp, label) \
2173 (bp)->state = BNX2X_STATE_ERROR; \
2177 #define LOAD_ERROR_EXIT_CNIC(bp, label) \
2179 bp->cnic_loaded = false; \
2182 #else /*BNX2X_STOP_ON_ERROR*/
2183 #define LOAD_ERROR_EXIT(bp, label) \
2185 (bp)->state = BNX2X_STATE_ERROR; \
2189 #define LOAD_ERROR_EXIT_CNIC(bp, label) \
2191 bp->cnic_loaded = false; \
2195 #endif /*BNX2X_STOP_ON_ERROR*/
2197 static void bnx2x_free_fw_stats_mem(struct bnx2x *bp)
2199 BNX2X_PCI_FREE(bp->fw_stats, bp->fw_stats_mapping,
2200 bp->fw_stats_data_sz + bp->fw_stats_req_sz);
2204 static int bnx2x_alloc_fw_stats_mem(struct bnx2x *bp)
2206 int num_groups, vf_headroom = 0;
2207 int is_fcoe_stats = NO_FCOE(bp) ? 0 : 1;
2209 /* number of queues for statistics is number of eth queues + FCoE */
2210 u8 num_queue_stats = BNX2X_NUM_ETH_QUEUES(bp) + is_fcoe_stats;
2212 /* Total number of FW statistics requests =
2213 * 1 for port stats + 1 for PF stats + potential 2 for FCoE (fcoe proper
2214 * and fcoe l2 queue) stats + num of queues (which includes another 1
2215 * for fcoe l2 queue if applicable)
2217 bp->fw_stats_num = 2 + is_fcoe_stats + num_queue_stats;
2219 /* vf stats appear in the request list, but their data is allocated by
2220 * the VFs themselves. We don't include them in the bp->fw_stats_num as
2221 * it is used to determine where to place the vf stats queries in the
2225 vf_headroom = bnx2x_vf_headroom(bp);
2227 /* Request is built from stats_query_header and an array of
2228 * stats_query_cmd_group each of which contains
2229 * STATS_QUERY_CMD_COUNT rules. The real number or requests is
2230 * configured in the stats_query_header.
2233 (((bp->fw_stats_num + vf_headroom) / STATS_QUERY_CMD_COUNT) +
2234 (((bp->fw_stats_num + vf_headroom) % STATS_QUERY_CMD_COUNT) ?
2237 DP(BNX2X_MSG_SP, "stats fw_stats_num %d, vf headroom %d, num_groups %d\n",
2238 bp->fw_stats_num, vf_headroom, num_groups);
2239 bp->fw_stats_req_sz = sizeof(struct stats_query_header) +
2240 num_groups * sizeof(struct stats_query_cmd_group);
2242 /* Data for statistics requests + stats_counter
2243 * stats_counter holds per-STORM counters that are incremented
2244 * when STORM has finished with the current request.
2245 * memory for FCoE offloaded statistics are counted anyway,
2246 * even if they will not be sent.
2247 * VF stats are not accounted for here as the data of VF stats is stored
2248 * in memory allocated by the VF, not here.
2250 bp->fw_stats_data_sz = sizeof(struct per_port_stats) +
2251 sizeof(struct per_pf_stats) +
2252 sizeof(struct fcoe_statistics_params) +
2253 sizeof(struct per_queue_stats) * num_queue_stats +
2254 sizeof(struct stats_counter);
2256 bp->fw_stats = BNX2X_PCI_ALLOC(&bp->fw_stats_mapping,
2257 bp->fw_stats_data_sz + bp->fw_stats_req_sz);
2262 bp->fw_stats_req = (struct bnx2x_fw_stats_req *)bp->fw_stats;
2263 bp->fw_stats_req_mapping = bp->fw_stats_mapping;
2264 bp->fw_stats_data = (struct bnx2x_fw_stats_data *)
2265 ((u8 *)bp->fw_stats + bp->fw_stats_req_sz);
2266 bp->fw_stats_data_mapping = bp->fw_stats_mapping +
2267 bp->fw_stats_req_sz;
2269 DP(BNX2X_MSG_SP, "statistics request base address set to %x %x\n",
2270 U64_HI(bp->fw_stats_req_mapping),
2271 U64_LO(bp->fw_stats_req_mapping));
2272 DP(BNX2X_MSG_SP, "statistics data base address set to %x %x\n",
2273 U64_HI(bp->fw_stats_data_mapping),
2274 U64_LO(bp->fw_stats_data_mapping));
2278 bnx2x_free_fw_stats_mem(bp);
2279 BNX2X_ERR("Can't allocate FW stats memory\n");
2283 /* send load request to mcp and analyze response */
2284 static int bnx2x_nic_load_request(struct bnx2x *bp, u32 *load_code)
2290 (SHMEM_RD(bp, func_mb[BP_FW_MB_IDX(bp)].drv_mb_header) &
2291 DRV_MSG_SEQ_NUMBER_MASK);
2292 BNX2X_DEV_INFO("fw_seq 0x%08x\n", bp->fw_seq);
2294 /* Get current FW pulse sequence */
2295 bp->fw_drv_pulse_wr_seq =
2296 (SHMEM_RD(bp, func_mb[BP_FW_MB_IDX(bp)].drv_pulse_mb) &
2297 DRV_PULSE_SEQ_MASK);
2298 BNX2X_DEV_INFO("drv_pulse 0x%x\n", bp->fw_drv_pulse_wr_seq);
2300 param = DRV_MSG_CODE_LOAD_REQ_WITH_LFA;
2302 if (IS_MF_SD(bp) && bnx2x_port_after_undi(bp))
2303 param |= DRV_MSG_CODE_LOAD_REQ_FORCE_LFA;
2306 (*load_code) = bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_REQ, param);
2308 /* if mcp fails to respond we must abort */
2309 if (!(*load_code)) {
2310 BNX2X_ERR("MCP response failure, aborting\n");
2314 /* If mcp refused (e.g. other port is in diagnostic mode) we
2317 if ((*load_code) == FW_MSG_CODE_DRV_LOAD_REFUSED) {
2318 BNX2X_ERR("MCP refused load request, aborting\n");
2324 /* check whether another PF has already loaded FW to chip. In
2325 * virtualized environments a pf from another VM may have already
2326 * initialized the device including loading FW
2328 int bnx2x_compare_fw_ver(struct bnx2x *bp, u32 load_code, bool print_err)
2330 /* is another pf loaded on this engine? */
2331 if (load_code != FW_MSG_CODE_DRV_LOAD_COMMON_CHIP &&
2332 load_code != FW_MSG_CODE_DRV_LOAD_COMMON) {
2333 /* build my FW version dword */
2334 u32 my_fw = (BCM_5710_FW_MAJOR_VERSION) +
2335 (BCM_5710_FW_MINOR_VERSION << 8) +
2336 (BCM_5710_FW_REVISION_VERSION << 16) +
2337 (BCM_5710_FW_ENGINEERING_VERSION << 24);
2339 /* read loaded FW from chip */
2340 u32 loaded_fw = REG_RD(bp, XSEM_REG_PRAM);
2342 DP(BNX2X_MSG_SP, "loaded fw %x, my fw %x\n",
2345 /* abort nic load if version mismatch */
2346 if (my_fw != loaded_fw) {
2348 BNX2X_ERR("bnx2x with FW %x was already loaded which mismatches my %x FW. Aborting\n",
2351 BNX2X_DEV_INFO("bnx2x with FW %x was already loaded which mismatches my %x FW, possibly due to MF UNDI\n",
2359 /* returns the "mcp load_code" according to global load_count array */
2360 static int bnx2x_nic_load_no_mcp(struct bnx2x *bp, int port)
2362 int path = BP_PATH(bp);
2364 DP(NETIF_MSG_IFUP, "NO MCP - load counts[%d] %d, %d, %d\n",
2365 path, bnx2x_load_count[path][0], bnx2x_load_count[path][1],
2366 bnx2x_load_count[path][2]);
2367 bnx2x_load_count[path][0]++;
2368 bnx2x_load_count[path][1 + port]++;
2369 DP(NETIF_MSG_IFUP, "NO MCP - new load counts[%d] %d, %d, %d\n",
2370 path, bnx2x_load_count[path][0], bnx2x_load_count[path][1],
2371 bnx2x_load_count[path][2]);
2372 if (bnx2x_load_count[path][0] == 1)
2373 return FW_MSG_CODE_DRV_LOAD_COMMON;
2374 else if (bnx2x_load_count[path][1 + port] == 1)
2375 return FW_MSG_CODE_DRV_LOAD_PORT;
2377 return FW_MSG_CODE_DRV_LOAD_FUNCTION;
2380 /* mark PMF if applicable */
2381 static void bnx2x_nic_load_pmf(struct bnx2x *bp, u32 load_code)
2383 if ((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
2384 (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP) ||
2385 (load_code == FW_MSG_CODE_DRV_LOAD_PORT)) {
2387 /* We need the barrier to ensure the ordering between the
2388 * writing to bp->port.pmf here and reading it from the
2389 * bnx2x_periodic_task().
2396 DP(NETIF_MSG_LINK, "pmf %d\n", bp->port.pmf);
2399 static void bnx2x_nic_load_afex_dcc(struct bnx2x *bp, int load_code)
2401 if (((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
2402 (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP)) &&
2403 (bp->common.shmem2_base)) {
2404 if (SHMEM2_HAS(bp, dcc_support))
2405 SHMEM2_WR(bp, dcc_support,
2406 (SHMEM_DCC_SUPPORT_DISABLE_ENABLE_PF_TLV |
2407 SHMEM_DCC_SUPPORT_BANDWIDTH_ALLOCATION_TLV));
2408 if (SHMEM2_HAS(bp, afex_driver_support))
2409 SHMEM2_WR(bp, afex_driver_support,
2410 SHMEM_AFEX_SUPPORTED_VERSION_ONE);
2413 /* Set AFEX default VLAN tag to an invalid value */
2414 bp->afex_def_vlan_tag = -1;
2418 * bnx2x_bz_fp - zero content of the fastpath structure.
2420 * @bp: driver handle
2421 * @index: fastpath index to be zeroed
2423 * Makes sure the contents of the bp->fp[index].napi is kept
2426 static void bnx2x_bz_fp(struct bnx2x *bp, int index)
2428 struct bnx2x_fastpath *fp = &bp->fp[index];
2430 struct napi_struct orig_napi = fp->napi;
2431 struct bnx2x_agg_info *orig_tpa_info = fp->tpa_info;
2433 /* bzero bnx2x_fastpath contents */
2435 memset(fp->tpa_info, 0, ETH_MAX_AGGREGATION_QUEUES_E1H_E2 *
2436 sizeof(struct bnx2x_agg_info));
2437 memset(fp, 0, sizeof(*fp));
2439 /* Restore the NAPI object as it has been already initialized */
2440 fp->napi = orig_napi;
2441 fp->tpa_info = orig_tpa_info;
2445 fp->max_cos = bp->max_cos;
2447 /* Special queues support only one CoS */
2450 /* Init txdata pointers */
2452 fp->txdata_ptr[0] = &bp->bnx2x_txq[FCOE_TXQ_IDX(bp)];
2454 for_each_cos_in_tx_queue(fp, cos)
2455 fp->txdata_ptr[cos] = &bp->bnx2x_txq[cos *
2456 BNX2X_NUM_ETH_QUEUES(bp) + index];
2458 /* set the tpa flag for each queue. The tpa flag determines the queue
2459 * minimal size so it must be set prior to queue memory allocation
2461 fp->disable_tpa = !(bp->flags & TPA_ENABLE_FLAG ||
2462 (bp->flags & GRO_ENABLE_FLAG &&
2463 bnx2x_mtu_allows_gro(bp->dev->mtu)));
2464 if (bp->flags & TPA_ENABLE_FLAG)
2465 fp->mode = TPA_MODE_LRO;
2466 else if (bp->flags & GRO_ENABLE_FLAG)
2467 fp->mode = TPA_MODE_GRO;
2469 /* We don't want TPA on an FCoE L2 ring */
2471 fp->disable_tpa = 1;
2474 int bnx2x_load_cnic(struct bnx2x *bp)
2476 int i, rc, port = BP_PORT(bp);
2478 DP(NETIF_MSG_IFUP, "Starting CNIC-related load\n");
2480 mutex_init(&bp->cnic_mutex);
2483 rc = bnx2x_alloc_mem_cnic(bp);
2485 BNX2X_ERR("Unable to allocate bp memory for cnic\n");
2486 LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic0);
2490 rc = bnx2x_alloc_fp_mem_cnic(bp);
2492 BNX2X_ERR("Unable to allocate memory for cnic fps\n");
2493 LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic0);
2496 /* Update the number of queues with the cnic queues */
2497 rc = bnx2x_set_real_num_queues(bp, 1);
2499 BNX2X_ERR("Unable to set real_num_queues including cnic\n");
2500 LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic0);
2503 /* Add all CNIC NAPI objects */
2504 bnx2x_add_all_napi_cnic(bp);
2505 DP(NETIF_MSG_IFUP, "cnic napi added\n");
2506 bnx2x_napi_enable_cnic(bp);
2508 rc = bnx2x_init_hw_func_cnic(bp);
2510 LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic1);
2512 bnx2x_nic_init_cnic(bp);
2515 /* Enable Timer scan */
2516 REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + port*4, 1);
2518 /* setup cnic queues */
2519 for_each_cnic_queue(bp, i) {
2520 rc = bnx2x_setup_queue(bp, &bp->fp[i], 0);
2522 BNX2X_ERR("Queue setup failed\n");
2523 LOAD_ERROR_EXIT(bp, load_error_cnic2);
2528 /* Initialize Rx filter. */
2529 bnx2x_set_rx_mode_inner(bp);
2531 /* re-read iscsi info */
2532 bnx2x_get_iscsi_info(bp);
2533 bnx2x_setup_cnic_irq_info(bp);
2534 bnx2x_setup_cnic_info(bp);
2535 bp->cnic_loaded = true;
2536 if (bp->state == BNX2X_STATE_OPEN)
2537 bnx2x_cnic_notify(bp, CNIC_CTL_START_CMD);
2539 DP(NETIF_MSG_IFUP, "Ending successfully CNIC-related load\n");
2543 #ifndef BNX2X_STOP_ON_ERROR
2545 /* Disable Timer scan */
2546 REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + port*4, 0);
2549 bnx2x_napi_disable_cnic(bp);
2550 /* Update the number of queues without the cnic queues */
2551 if (bnx2x_set_real_num_queues(bp, 0))
2552 BNX2X_ERR("Unable to set real_num_queues not including cnic\n");
2554 BNX2X_ERR("CNIC-related load failed\n");
2555 bnx2x_free_fp_mem_cnic(bp);
2556 bnx2x_free_mem_cnic(bp);
2558 #endif /* ! BNX2X_STOP_ON_ERROR */
2561 /* must be called with rtnl_lock */
2562 int bnx2x_nic_load(struct bnx2x *bp, int load_mode)
2564 int port = BP_PORT(bp);
2565 int i, rc = 0, load_code = 0;
2567 DP(NETIF_MSG_IFUP, "Starting NIC load\n");
2569 "CNIC is %s\n", CNIC_ENABLED(bp) ? "enabled" : "disabled");
2571 #ifdef BNX2X_STOP_ON_ERROR
2572 if (unlikely(bp->panic)) {
2573 BNX2X_ERR("Can't load NIC when there is panic\n");
2578 bp->state = BNX2X_STATE_OPENING_WAIT4_LOAD;
2580 /* zero the structure w/o any lock, before SP handler is initialized */
2581 memset(&bp->last_reported_link, 0, sizeof(bp->last_reported_link));
2582 __set_bit(BNX2X_LINK_REPORT_LINK_DOWN,
2583 &bp->last_reported_link.link_report_flags);
2586 /* must be called before memory allocation and HW init */
2587 bnx2x_ilt_set_info(bp);
2590 * Zero fastpath structures preserving invariants like napi, which are
2591 * allocated only once, fp index, max_cos, bp pointer.
2592 * Also set fp->disable_tpa and txdata_ptr.
2594 DP(NETIF_MSG_IFUP, "num queues: %d", bp->num_queues);
2595 for_each_queue(bp, i)
2597 memset(bp->bnx2x_txq, 0, (BNX2X_MAX_RSS_COUNT(bp) * BNX2X_MULTI_TX_COS +
2598 bp->num_cnic_queues) *
2599 sizeof(struct bnx2x_fp_txdata));
2601 bp->fcoe_init = false;
2603 /* Set the receive queues buffer size */
2604 bnx2x_set_rx_buf_size(bp);
2607 rc = bnx2x_alloc_mem(bp);
2609 BNX2X_ERR("Unable to allocate bp memory\n");
2614 /* need to be done after alloc mem, since it's self adjusting to amount
2615 * of memory available for RSS queues
2617 rc = bnx2x_alloc_fp_mem(bp);
2619 BNX2X_ERR("Unable to allocate memory for fps\n");
2620 LOAD_ERROR_EXIT(bp, load_error0);
2623 /* Allocated memory for FW statistics */
2624 if (bnx2x_alloc_fw_stats_mem(bp))
2625 LOAD_ERROR_EXIT(bp, load_error0);
2627 /* request pf to initialize status blocks */
2629 rc = bnx2x_vfpf_init(bp);
2631 LOAD_ERROR_EXIT(bp, load_error0);
2634 /* As long as bnx2x_alloc_mem() may possibly update
2635 * bp->num_queues, bnx2x_set_real_num_queues() should always
2636 * come after it. At this stage cnic queues are not counted.
2638 rc = bnx2x_set_real_num_queues(bp, 0);
2640 BNX2X_ERR("Unable to set real_num_queues\n");
2641 LOAD_ERROR_EXIT(bp, load_error0);
2644 /* configure multi cos mappings in kernel.
2645 * this configuration may be overridden by a multi class queue
2646 * discipline or by a dcbx negotiation result.
2648 bnx2x_setup_tc(bp->dev, bp->max_cos);
2650 /* Add all NAPI objects */
2651 bnx2x_add_all_napi(bp);
2652 DP(NETIF_MSG_IFUP, "napi added\n");
2653 bnx2x_napi_enable(bp);
2656 /* set pf load just before approaching the MCP */
2657 bnx2x_set_pf_load(bp);
2659 /* if mcp exists send load request and analyze response */
2660 if (!BP_NOMCP(bp)) {
2661 /* attempt to load pf */
2662 rc = bnx2x_nic_load_request(bp, &load_code);
2664 LOAD_ERROR_EXIT(bp, load_error1);
2666 /* what did mcp say? */
2667 rc = bnx2x_compare_fw_ver(bp, load_code, true);
2669 bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2670 LOAD_ERROR_EXIT(bp, load_error2);
2673 load_code = bnx2x_nic_load_no_mcp(bp, port);
2676 /* mark pmf if applicable */
2677 bnx2x_nic_load_pmf(bp, load_code);
2679 /* Init Function state controlling object */
2680 bnx2x__init_func_obj(bp);
2683 rc = bnx2x_init_hw(bp, load_code);
2685 BNX2X_ERR("HW init failed, aborting\n");
2686 bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2687 LOAD_ERROR_EXIT(bp, load_error2);
2691 bnx2x_pre_irq_nic_init(bp);
2693 /* Connect to IRQs */
2694 rc = bnx2x_setup_irqs(bp);
2696 BNX2X_ERR("setup irqs failed\n");
2698 bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2699 LOAD_ERROR_EXIT(bp, load_error2);
2702 /* Init per-function objects */
2704 /* Setup NIC internals and enable interrupts */
2705 bnx2x_post_irq_nic_init(bp, load_code);
2707 bnx2x_init_bp_objs(bp);
2708 bnx2x_iov_nic_init(bp);
2710 /* Set AFEX default VLAN tag to an invalid value */
2711 bp->afex_def_vlan_tag = -1;
2712 bnx2x_nic_load_afex_dcc(bp, load_code);
2713 bp->state = BNX2X_STATE_OPENING_WAIT4_PORT;
2714 rc = bnx2x_func_start(bp);
2716 BNX2X_ERR("Function start failed!\n");
2717 bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2719 LOAD_ERROR_EXIT(bp, load_error3);
2722 /* Send LOAD_DONE command to MCP */
2723 if (!BP_NOMCP(bp)) {
2724 load_code = bnx2x_fw_command(bp,
2725 DRV_MSG_CODE_LOAD_DONE, 0);
2727 BNX2X_ERR("MCP response failure, aborting\n");
2729 LOAD_ERROR_EXIT(bp, load_error3);
2733 /* initialize FW coalescing state machines in RAM */
2734 bnx2x_update_coalesce(bp);
2737 /* setup the leading queue */
2738 rc = bnx2x_setup_leading(bp);
2740 BNX2X_ERR("Setup leading failed!\n");
2741 LOAD_ERROR_EXIT(bp, load_error3);
2744 /* set up the rest of the queues */
2745 for_each_nondefault_eth_queue(bp, i) {
2747 rc = bnx2x_setup_queue(bp, &bp->fp[i], false);
2749 rc = bnx2x_vfpf_setup_q(bp, &bp->fp[i], false);
2751 BNX2X_ERR("Queue %d setup failed\n", i);
2752 LOAD_ERROR_EXIT(bp, load_error3);
2757 rc = bnx2x_init_rss(bp);
2759 BNX2X_ERR("PF RSS init failed\n");
2760 LOAD_ERROR_EXIT(bp, load_error3);
2763 /* Now when Clients are configured we are ready to work */
2764 bp->state = BNX2X_STATE_OPEN;
2766 /* Configure a ucast MAC */
2768 rc = bnx2x_set_eth_mac(bp, true);
2770 rc = bnx2x_vfpf_config_mac(bp, bp->dev->dev_addr, bp->fp->index,
2773 BNX2X_ERR("Setting Ethernet MAC failed\n");
2774 LOAD_ERROR_EXIT(bp, load_error3);
2777 if (IS_PF(bp) && bp->pending_max) {
2778 bnx2x_update_max_mf_config(bp, bp->pending_max);
2779 bp->pending_max = 0;
2783 rc = bnx2x_initial_phy_init(bp, load_mode);
2785 LOAD_ERROR_EXIT(bp, load_error3);
2787 bp->link_params.feature_config_flags &= ~FEATURE_CONFIG_BOOT_FROM_SAN;
2789 /* Start fast path */
2791 /* Initialize Rx filter. */
2792 bnx2x_set_rx_mode_inner(bp);
2795 switch (load_mode) {
2797 /* Tx queue should be only re-enabled */
2798 netif_tx_wake_all_queues(bp->dev);
2802 netif_tx_start_all_queues(bp->dev);
2803 smp_mb__after_atomic();
2807 case LOAD_LOOPBACK_EXT:
2808 bp->state = BNX2X_STATE_DIAG;
2816 bnx2x_update_drv_flags(bp, 1 << DRV_FLAGS_PORT_MASK, 0);
2818 bnx2x__link_status_update(bp);
2820 /* start the timer */
2821 mod_timer(&bp->timer, jiffies + bp->current_interval);
2823 if (CNIC_ENABLED(bp))
2824 bnx2x_load_cnic(bp);
2827 bnx2x_schedule_sp_rtnl(bp, BNX2X_SP_RTNL_GET_DRV_VERSION, 0);
2829 if (IS_PF(bp) && SHMEM2_HAS(bp, drv_capabilities_flag)) {
2830 /* mark driver is loaded in shmem2 */
2832 val = SHMEM2_RD(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)]);
2833 SHMEM2_WR(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)],
2834 val | DRV_FLAGS_CAPABILITIES_LOADED_SUPPORTED |
2835 DRV_FLAGS_CAPABILITIES_LOADED_L2);
2838 /* Wait for all pending SP commands to complete */
2839 if (IS_PF(bp) && !bnx2x_wait_sp_comp(bp, ~0x0UL)) {
2840 BNX2X_ERR("Timeout waiting for SP elements to complete\n");
2841 bnx2x_nic_unload(bp, UNLOAD_CLOSE, false);
2845 /* If PMF - send ADMIN DCBX msg to MFW to initiate DCBX FSM */
2846 if (bp->port.pmf && (bp->state != BNX2X_STATE_DIAG))
2847 bnx2x_dcbx_init(bp, false);
2849 DP(NETIF_MSG_IFUP, "Ending successfully NIC load\n");
2853 #ifndef BNX2X_STOP_ON_ERROR
2856 bnx2x_int_disable_sync(bp, 1);
2858 /* Clean queueable objects */
2859 bnx2x_squeeze_objects(bp);
2862 /* Free SKBs, SGEs, TPA pool and driver internals */
2863 bnx2x_free_skbs(bp);
2864 for_each_rx_queue(bp, i)
2865 bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
2870 if (IS_PF(bp) && !BP_NOMCP(bp)) {
2871 bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_REQ_WOL_MCP, 0);
2872 bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_DONE, 0);
2877 bnx2x_napi_disable(bp);
2878 bnx2x_del_all_napi(bp);
2880 /* clear pf_load status, as it was already set */
2882 bnx2x_clear_pf_load(bp);
2884 bnx2x_free_fw_stats_mem(bp);
2885 bnx2x_free_fp_mem(bp);
2889 #endif /* ! BNX2X_STOP_ON_ERROR */
2892 int bnx2x_drain_tx_queues(struct bnx2x *bp)
2896 /* Wait until tx fastpath tasks complete */
2897 for_each_tx_queue(bp, i) {
2898 struct bnx2x_fastpath *fp = &bp->fp[i];
2900 for_each_cos_in_tx_queue(fp, cos)
2901 rc = bnx2x_clean_tx_queue(bp, fp->txdata_ptr[cos]);
2908 /* must be called with rtnl_lock */
2909 int bnx2x_nic_unload(struct bnx2x *bp, int unload_mode, bool keep_link)
2912 bool global = false;
2914 DP(NETIF_MSG_IFUP, "Starting NIC unload\n");
2916 /* mark driver is unloaded in shmem2 */
2917 if (IS_PF(bp) && SHMEM2_HAS(bp, drv_capabilities_flag)) {
2919 val = SHMEM2_RD(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)]);
2920 SHMEM2_WR(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)],
2921 val & ~DRV_FLAGS_CAPABILITIES_LOADED_L2);
2924 if (IS_PF(bp) && bp->recovery_state != BNX2X_RECOVERY_DONE &&
2925 (bp->state == BNX2X_STATE_CLOSED ||
2926 bp->state == BNX2X_STATE_ERROR)) {
2927 /* We can get here if the driver has been unloaded
2928 * during parity error recovery and is either waiting for a
2929 * leader to complete or for other functions to unload and
2930 * then ifdown has been issued. In this case we want to
2931 * unload and let other functions to complete a recovery
2934 bp->recovery_state = BNX2X_RECOVERY_DONE;
2936 bnx2x_release_leader_lock(bp);
2939 DP(NETIF_MSG_IFDOWN, "Releasing a leadership...\n");
2940 BNX2X_ERR("Can't unload in closed or error state\n");
2944 /* Nothing to do during unload if previous bnx2x_nic_load()
2945 * have not completed successfully - all resources are released.
2947 * we can get here only after unsuccessful ndo_* callback, during which
2948 * dev->IFF_UP flag is still on.
2950 if (bp->state == BNX2X_STATE_CLOSED || bp->state == BNX2X_STATE_ERROR)
2953 /* It's important to set the bp->state to the value different from
2954 * BNX2X_STATE_OPEN and only then stop the Tx. Otherwise bnx2x_tx_int()
2955 * may restart the Tx from the NAPI context (see bnx2x_tx_int()).
2957 bp->state = BNX2X_STATE_CLOSING_WAIT4_HALT;
2960 /* indicate to VFs that the PF is going down */
2961 bnx2x_iov_channel_down(bp);
2963 if (CNIC_LOADED(bp))
2964 bnx2x_cnic_notify(bp, CNIC_CTL_STOP_CMD);
2967 bnx2x_tx_disable(bp);
2968 netdev_reset_tc(bp->dev);
2970 bp->rx_mode = BNX2X_RX_MODE_NONE;
2972 del_timer_sync(&bp->timer);
2975 /* Set ALWAYS_ALIVE bit in shmem */
2976 bp->fw_drv_pulse_wr_seq |= DRV_PULSE_ALWAYS_ALIVE;
2977 bnx2x_drv_pulse(bp);
2978 bnx2x_stats_handle(bp, STATS_EVENT_STOP);
2979 bnx2x_save_statistics(bp);
2982 /* wait till consumers catch up with producers in all queues */
2983 bnx2x_drain_tx_queues(bp);
2985 /* if VF indicate to PF this function is going down (PF will delete sp
2986 * elements and clear initializations
2989 bnx2x_vfpf_close_vf(bp);
2990 else if (unload_mode != UNLOAD_RECOVERY)
2991 /* if this is a normal/close unload need to clean up chip*/
2992 bnx2x_chip_cleanup(bp, unload_mode, keep_link);
2994 /* Send the UNLOAD_REQUEST to the MCP */
2995 bnx2x_send_unload_req(bp, unload_mode);
2997 /* Prevent transactions to host from the functions on the
2998 * engine that doesn't reset global blocks in case of global
2999 * attention once global blocks are reset and gates are opened
3000 * (the engine which leader will perform the recovery
3003 if (!CHIP_IS_E1x(bp))
3004 bnx2x_pf_disable(bp);
3006 /* Disable HW interrupts, NAPI */
3007 bnx2x_netif_stop(bp, 1);
3008 /* Delete all NAPI objects */
3009 bnx2x_del_all_napi(bp);
3010 if (CNIC_LOADED(bp))
3011 bnx2x_del_all_napi_cnic(bp);
3015 /* Report UNLOAD_DONE to MCP */
3016 bnx2x_send_unload_done(bp, false);
3020 * At this stage no more interrupts will arrive so we may safely clean
3021 * the queueable objects here in case they failed to get cleaned so far.
3024 bnx2x_squeeze_objects(bp);
3026 /* There should be no more pending SP commands at this stage */
3031 /* clear pending work in rtnl task */
3032 bp->sp_rtnl_state = 0;
3035 /* Free SKBs, SGEs, TPA pool and driver internals */
3036 bnx2x_free_skbs(bp);
3037 if (CNIC_LOADED(bp))
3038 bnx2x_free_skbs_cnic(bp);
3039 for_each_rx_queue(bp, i)
3040 bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
3042 bnx2x_free_fp_mem(bp);
3043 if (CNIC_LOADED(bp))
3044 bnx2x_free_fp_mem_cnic(bp);
3047 if (CNIC_LOADED(bp))
3048 bnx2x_free_mem_cnic(bp);
3052 bp->state = BNX2X_STATE_CLOSED;
3053 bp->cnic_loaded = false;
3055 /* Clear driver version indication in shmem */
3057 bnx2x_update_mng_version(bp);
3059 /* Check if there are pending parity attentions. If there are - set
3060 * RECOVERY_IN_PROGRESS.
3062 if (IS_PF(bp) && bnx2x_chk_parity_attn(bp, &global, false)) {
3063 bnx2x_set_reset_in_progress(bp);
3065 /* Set RESET_IS_GLOBAL if needed */
3067 bnx2x_set_reset_global(bp);
3070 /* The last driver must disable a "close the gate" if there is no
3071 * parity attention or "process kill" pending.
3074 !bnx2x_clear_pf_load(bp) &&
3075 bnx2x_reset_is_done(bp, BP_PATH(bp)))
3076 bnx2x_disable_close_the_gate(bp);
3078 DP(NETIF_MSG_IFUP, "Ending NIC unload\n");
3083 int bnx2x_set_power_state(struct bnx2x *bp, pci_power_t state)
3087 /* If there is no power capability, silently succeed */
3088 if (!bp->pdev->pm_cap) {
3089 BNX2X_DEV_INFO("No power capability. Breaking.\n");
3093 pci_read_config_word(bp->pdev, bp->pdev->pm_cap + PCI_PM_CTRL, &pmcsr);
3097 pci_write_config_word(bp->pdev, bp->pdev->pm_cap + PCI_PM_CTRL,
3098 ((pmcsr & ~PCI_PM_CTRL_STATE_MASK) |
3099 PCI_PM_CTRL_PME_STATUS));
3101 if (pmcsr & PCI_PM_CTRL_STATE_MASK)
3102 /* delay required during transition out of D3hot */
3107 /* If there are other clients above don't
3108 shut down the power */
3109 if (atomic_read(&bp->pdev->enable_cnt) != 1)
3111 /* Don't shut down the power for emulation and FPGA */
3112 if (CHIP_REV_IS_SLOW(bp))
3115 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
3119 pmcsr |= PCI_PM_CTRL_PME_ENABLE;
3121 pci_write_config_word(bp->pdev, bp->pdev->pm_cap + PCI_PM_CTRL,
3124 /* No more memory access after this point until
3125 * device is brought back to D0.
3130 dev_err(&bp->pdev->dev, "Can't support state = %d\n", state);
3137 * net_device service functions
3139 static int bnx2x_poll(struct napi_struct *napi, int budget)
3143 struct bnx2x_fastpath *fp = container_of(napi, struct bnx2x_fastpath,
3145 struct bnx2x *bp = fp->bp;
3148 #ifdef BNX2X_STOP_ON_ERROR
3149 if (unlikely(bp->panic)) {
3150 napi_complete(napi);
3154 if (!bnx2x_fp_lock_napi(fp))
3157 for_each_cos_in_tx_queue(fp, cos)
3158 if (bnx2x_tx_queue_has_work(fp->txdata_ptr[cos]))
3159 bnx2x_tx_int(bp, fp->txdata_ptr[cos]);
3161 if (bnx2x_has_rx_work(fp)) {
3162 work_done += bnx2x_rx_int(fp, budget - work_done);
3164 /* must not complete if we consumed full budget */
3165 if (work_done >= budget) {
3166 bnx2x_fp_unlock_napi(fp);
3171 /* Fall out from the NAPI loop if needed */
3172 if (!bnx2x_fp_unlock_napi(fp) &&
3173 !(bnx2x_has_rx_work(fp) || bnx2x_has_tx_work(fp))) {
3175 /* No need to update SB for FCoE L2 ring as long as
3176 * it's connected to the default SB and the SB
3177 * has been updated when NAPI was scheduled.
3179 if (IS_FCOE_FP(fp)) {
3180 napi_complete(napi);
3183 bnx2x_update_fpsb_idx(fp);
3184 /* bnx2x_has_rx_work() reads the status block,
3185 * thus we need to ensure that status block indices
3186 * have been actually read (bnx2x_update_fpsb_idx)
3187 * prior to this check (bnx2x_has_rx_work) so that
3188 * we won't write the "newer" value of the status block
3189 * to IGU (if there was a DMA right after
3190 * bnx2x_has_rx_work and if there is no rmb, the memory
3191 * reading (bnx2x_update_fpsb_idx) may be postponed
3192 * to right before bnx2x_ack_sb). In this case there
3193 * will never be another interrupt until there is
3194 * another update of the status block, while there
3195 * is still unhandled work.
3199 if (!(bnx2x_has_rx_work(fp) || bnx2x_has_tx_work(fp))) {
3200 napi_complete(napi);
3201 /* Re-enable interrupts */
3202 DP(NETIF_MSG_RX_STATUS,
3203 "Update index to %d\n", fp->fp_hc_idx);
3204 bnx2x_ack_sb(bp, fp->igu_sb_id, USTORM_ID,
3205 le16_to_cpu(fp->fp_hc_idx),
3215 #ifdef CONFIG_NET_RX_BUSY_POLL
3216 /* must be called with local_bh_disable()d */
3217 int bnx2x_low_latency_recv(struct napi_struct *napi)
3219 struct bnx2x_fastpath *fp = container_of(napi, struct bnx2x_fastpath,
3221 struct bnx2x *bp = fp->bp;
3224 if ((bp->state == BNX2X_STATE_CLOSED) ||
3225 (bp->state == BNX2X_STATE_ERROR) ||
3226 (bp->flags & (TPA_ENABLE_FLAG | GRO_ENABLE_FLAG)))
3227 return LL_FLUSH_FAILED;
3229 if (!bnx2x_fp_lock_poll(fp))
3230 return LL_FLUSH_BUSY;
3232 if (bnx2x_has_rx_work(fp))
3233 found = bnx2x_rx_int(fp, 4);
3235 bnx2x_fp_unlock_poll(fp);
3241 /* we split the first BD into headers and data BDs
3242 * to ease the pain of our fellow microcode engineers
3243 * we use one mapping for both BDs
3245 static u16 bnx2x_tx_split(struct bnx2x *bp,
3246 struct bnx2x_fp_txdata *txdata,
3247 struct sw_tx_bd *tx_buf,
3248 struct eth_tx_start_bd **tx_bd, u16 hlen,
3251 struct eth_tx_start_bd *h_tx_bd = *tx_bd;
3252 struct eth_tx_bd *d_tx_bd;
3254 int old_len = le16_to_cpu(h_tx_bd->nbytes);
3256 /* first fix first BD */
3257 h_tx_bd->nbytes = cpu_to_le16(hlen);
3259 DP(NETIF_MSG_TX_QUEUED, "TSO split header size is %d (%x:%x)\n",
3260 h_tx_bd->nbytes, h_tx_bd->addr_hi, h_tx_bd->addr_lo);
3262 /* now get a new data BD
3263 * (after the pbd) and fill it */
3264 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3265 d_tx_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
3267 mapping = HILO_U64(le32_to_cpu(h_tx_bd->addr_hi),
3268 le32_to_cpu(h_tx_bd->addr_lo)) + hlen;
3270 d_tx_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
3271 d_tx_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
3272 d_tx_bd->nbytes = cpu_to_le16(old_len - hlen);
3274 /* this marks the BD as one that has no individual mapping */
3275 tx_buf->flags |= BNX2X_TSO_SPLIT_BD;
3277 DP(NETIF_MSG_TX_QUEUED,
3278 "TSO split data size is %d (%x:%x)\n",
3279 d_tx_bd->nbytes, d_tx_bd->addr_hi, d_tx_bd->addr_lo);
3282 *tx_bd = (struct eth_tx_start_bd *)d_tx_bd;
3287 #define bswab32(b32) ((__force __le32) swab32((__force __u32) (b32)))
3288 #define bswab16(b16) ((__force __le16) swab16((__force __u16) (b16)))
3289 static __le16 bnx2x_csum_fix(unsigned char *t_header, u16 csum, s8 fix)
3291 __sum16 tsum = (__force __sum16) csum;
3294 tsum = ~csum_fold(csum_sub((__force __wsum) csum,
3295 csum_partial(t_header - fix, fix, 0)));
3298 tsum = ~csum_fold(csum_add((__force __wsum) csum,
3299 csum_partial(t_header, -fix, 0)));
3301 return bswab16(tsum);
3304 static u32 bnx2x_xmit_type(struct bnx2x *bp, struct sk_buff *skb)
3310 if (skb->ip_summed != CHECKSUM_PARTIAL)
3313 protocol = vlan_get_protocol(skb);
3314 if (protocol == htons(ETH_P_IPV6)) {
3316 prot = ipv6_hdr(skb)->nexthdr;
3319 prot = ip_hdr(skb)->protocol;
3322 if (!CHIP_IS_E1x(bp) && skb->encapsulation) {
3323 if (inner_ip_hdr(skb)->version == 6) {
3324 rc |= XMIT_CSUM_ENC_V6;
3325 if (inner_ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3326 rc |= XMIT_CSUM_TCP;
3328 rc |= XMIT_CSUM_ENC_V4;
3329 if (inner_ip_hdr(skb)->protocol == IPPROTO_TCP)
3330 rc |= XMIT_CSUM_TCP;
3333 if (prot == IPPROTO_TCP)
3334 rc |= XMIT_CSUM_TCP;
3336 if (skb_is_gso(skb)) {
3337 if (skb_is_gso_v6(skb)) {
3338 rc |= (XMIT_GSO_V6 | XMIT_CSUM_TCP);
3339 if (rc & XMIT_CSUM_ENC)
3340 rc |= XMIT_GSO_ENC_V6;
3342 rc |= (XMIT_GSO_V4 | XMIT_CSUM_TCP);
3343 if (rc & XMIT_CSUM_ENC)
3344 rc |= XMIT_GSO_ENC_V4;
3351 #if (MAX_SKB_FRAGS >= MAX_FETCH_BD - 3)
3352 /* check if packet requires linearization (packet is too fragmented)
3353 no need to check fragmentation if page size > 8K (there will be no
3354 violation to FW restrictions) */
3355 static int bnx2x_pkt_req_lin(struct bnx2x *bp, struct sk_buff *skb,
3360 int first_bd_sz = 0;
3362 /* 3 = 1 (for linear data BD) + 2 (for PBD and last BD) */
3363 if (skb_shinfo(skb)->nr_frags >= (MAX_FETCH_BD - 3)) {
3365 if (xmit_type & XMIT_GSO) {
3366 unsigned short lso_mss = skb_shinfo(skb)->gso_size;
3367 /* Check if LSO packet needs to be copied:
3368 3 = 1 (for headers BD) + 2 (for PBD and last BD) */
3369 int wnd_size = MAX_FETCH_BD - 3;
3370 /* Number of windows to check */
3371 int num_wnds = skb_shinfo(skb)->nr_frags - wnd_size;
3376 /* Headers length */
3377 hlen = (int)(skb_transport_header(skb) - skb->data) +
3380 /* Amount of data (w/o headers) on linear part of SKB*/
3381 first_bd_sz = skb_headlen(skb) - hlen;
3383 wnd_sum = first_bd_sz;
3385 /* Calculate the first sum - it's special */
3386 for (frag_idx = 0; frag_idx < wnd_size - 1; frag_idx++)
3388 skb_frag_size(&skb_shinfo(skb)->frags[frag_idx]);
3390 /* If there was data on linear skb data - check it */
3391 if (first_bd_sz > 0) {
3392 if (unlikely(wnd_sum < lso_mss)) {
3397 wnd_sum -= first_bd_sz;
3400 /* Others are easier: run through the frag list and
3401 check all windows */
3402 for (wnd_idx = 0; wnd_idx <= num_wnds; wnd_idx++) {
3404 skb_frag_size(&skb_shinfo(skb)->frags[wnd_idx + wnd_size - 1]);
3406 if (unlikely(wnd_sum < lso_mss)) {
3411 skb_frag_size(&skb_shinfo(skb)->frags[wnd_idx]);
3414 /* in non-LSO too fragmented packet should always
3421 if (unlikely(to_copy))
3422 DP(NETIF_MSG_TX_QUEUED,
3423 "Linearization IS REQUIRED for %s packet. num_frags %d hlen %d first_bd_sz %d\n",
3424 (xmit_type & XMIT_GSO) ? "LSO" : "non-LSO",
3425 skb_shinfo(skb)->nr_frags, hlen, first_bd_sz);
3431 static void bnx2x_set_pbd_gso_e2(struct sk_buff *skb, u32 *parsing_data,
3434 struct ipv6hdr *ipv6;
3436 *parsing_data |= (skb_shinfo(skb)->gso_size <<
3437 ETH_TX_PARSE_BD_E2_LSO_MSS_SHIFT) &
3438 ETH_TX_PARSE_BD_E2_LSO_MSS;
3440 if (xmit_type & XMIT_GSO_ENC_V6)
3441 ipv6 = inner_ipv6_hdr(skb);
3442 else if (xmit_type & XMIT_GSO_V6)
3443 ipv6 = ipv6_hdr(skb);
3447 if (ipv6 && ipv6->nexthdr == NEXTHDR_IPV6)
3448 *parsing_data |= ETH_TX_PARSE_BD_E2_IPV6_WITH_EXT_HDR;
3452 * bnx2x_set_pbd_gso - update PBD in GSO case.
3456 * @xmit_type: xmit flags
3458 static void bnx2x_set_pbd_gso(struct sk_buff *skb,
3459 struct eth_tx_parse_bd_e1x *pbd,
3460 struct eth_tx_start_bd *tx_start_bd,
3463 pbd->lso_mss = cpu_to_le16(skb_shinfo(skb)->gso_size);
3464 pbd->tcp_send_seq = bswab32(tcp_hdr(skb)->seq);
3465 pbd->tcp_flags = pbd_tcp_flags(tcp_hdr(skb));
3467 if (xmit_type & XMIT_GSO_V4) {
3468 pbd->ip_id = bswab16(ip_hdr(skb)->id);
3469 pbd->tcp_pseudo_csum =
3470 bswab16(~csum_tcpudp_magic(ip_hdr(skb)->saddr,
3472 0, IPPROTO_TCP, 0));
3474 /* GSO on 57710/57711 needs FW to calculate IP checksum */
3475 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_IP_CSUM;
3477 pbd->tcp_pseudo_csum =
3478 bswab16(~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3479 &ipv6_hdr(skb)->daddr,
3480 0, IPPROTO_TCP, 0));
3484 cpu_to_le16(ETH_TX_PARSE_BD_E1X_PSEUDO_CS_WITHOUT_LEN);
3488 * bnx2x_set_pbd_csum_enc - update PBD with checksum and return header length
3490 * @bp: driver handle
3492 * @parsing_data: data to be updated
3493 * @xmit_type: xmit flags
3495 * 57712/578xx related, when skb has encapsulation
3497 static u8 bnx2x_set_pbd_csum_enc(struct bnx2x *bp, struct sk_buff *skb,
3498 u32 *parsing_data, u32 xmit_type)
3501 ((((u8 *)skb_inner_transport_header(skb) - skb->data) >> 1) <<
3502 ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W_SHIFT) &
3503 ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W;
3505 if (xmit_type & XMIT_CSUM_TCP) {
3506 *parsing_data |= ((inner_tcp_hdrlen(skb) / 4) <<
3507 ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW_SHIFT) &
3508 ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW;
3510 return skb_inner_transport_header(skb) +
3511 inner_tcp_hdrlen(skb) - skb->data;
3514 /* We support checksum offload for TCP and UDP only.
3515 * No need to pass the UDP header length - it's a constant.
3517 return skb_inner_transport_header(skb) +
3518 sizeof(struct udphdr) - skb->data;
3522 * bnx2x_set_pbd_csum_e2 - update PBD with checksum and return header length
3524 * @bp: driver handle
3526 * @parsing_data: data to be updated
3527 * @xmit_type: xmit flags
3529 * 57712/578xx related
3531 static u8 bnx2x_set_pbd_csum_e2(struct bnx2x *bp, struct sk_buff *skb,
3532 u32 *parsing_data, u32 xmit_type)
3535 ((((u8 *)skb_transport_header(skb) - skb->data) >> 1) <<
3536 ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W_SHIFT) &
3537 ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W;
3539 if (xmit_type & XMIT_CSUM_TCP) {
3540 *parsing_data |= ((tcp_hdrlen(skb) / 4) <<
3541 ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW_SHIFT) &
3542 ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW;
3544 return skb_transport_header(skb) + tcp_hdrlen(skb) - skb->data;
3546 /* We support checksum offload for TCP and UDP only.
3547 * No need to pass the UDP header length - it's a constant.
3549 return skb_transport_header(skb) + sizeof(struct udphdr) - skb->data;
3552 /* set FW indication according to inner or outer protocols if tunneled */
3553 static void bnx2x_set_sbd_csum(struct bnx2x *bp, struct sk_buff *skb,
3554 struct eth_tx_start_bd *tx_start_bd,
3557 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_L4_CSUM;
3559 if (xmit_type & (XMIT_CSUM_ENC_V6 | XMIT_CSUM_V6))
3560 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_IPV6;
3562 if (!(xmit_type & XMIT_CSUM_TCP))
3563 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_IS_UDP;
3567 * bnx2x_set_pbd_csum - update PBD with checksum and return header length
3569 * @bp: driver handle
3571 * @pbd: parse BD to be updated
3572 * @xmit_type: xmit flags
3574 static u8 bnx2x_set_pbd_csum(struct bnx2x *bp, struct sk_buff *skb,
3575 struct eth_tx_parse_bd_e1x *pbd,
3578 u8 hlen = (skb_network_header(skb) - skb->data) >> 1;
3580 /* for now NS flag is not used in Linux */
3583 ((skb->protocol == cpu_to_be16(ETH_P_8021Q)) <<
3584 ETH_TX_PARSE_BD_E1X_LLC_SNAP_EN_SHIFT));
3586 pbd->ip_hlen_w = (skb_transport_header(skb) -
3587 skb_network_header(skb)) >> 1;
3589 hlen += pbd->ip_hlen_w;
3591 /* We support checksum offload for TCP and UDP only */
3592 if (xmit_type & XMIT_CSUM_TCP)
3593 hlen += tcp_hdrlen(skb) / 2;
3595 hlen += sizeof(struct udphdr) / 2;
3597 pbd->total_hlen_w = cpu_to_le16(hlen);
3600 if (xmit_type & XMIT_CSUM_TCP) {
3601 pbd->tcp_pseudo_csum = bswab16(tcp_hdr(skb)->check);
3604 s8 fix = SKB_CS_OFF(skb); /* signed! */
3606 DP(NETIF_MSG_TX_QUEUED,
3607 "hlen %d fix %d csum before fix %x\n",
3608 le16_to_cpu(pbd->total_hlen_w), fix, SKB_CS(skb));
3610 /* HW bug: fixup the CSUM */
3611 pbd->tcp_pseudo_csum =
3612 bnx2x_csum_fix(skb_transport_header(skb),
3615 DP(NETIF_MSG_TX_QUEUED, "csum after fix %x\n",
3616 pbd->tcp_pseudo_csum);
3622 static void bnx2x_update_pbds_gso_enc(struct sk_buff *skb,
3623 struct eth_tx_parse_bd_e2 *pbd_e2,
3624 struct eth_tx_parse_2nd_bd *pbd2,
3629 u8 outerip_off, outerip_len = 0;
3631 /* from outer IP to transport */
3632 hlen_w = (skb_inner_transport_header(skb) -
3633 skb_network_header(skb)) >> 1;
3636 hlen_w += inner_tcp_hdrlen(skb) >> 1;
3638 pbd2->fw_ip_hdr_to_payload_w = hlen_w;
3640 /* outer IP header info */
3641 if (xmit_type & XMIT_CSUM_V4) {
3642 struct iphdr *iph = ip_hdr(skb);
3643 u32 csum = (__force u32)(~iph->check) -
3644 (__force u32)iph->tot_len -
3645 (__force u32)iph->frag_off;
3647 pbd2->fw_ip_csum_wo_len_flags_frag =
3648 bswab16(csum_fold((__force __wsum)csum));
3650 pbd2->fw_ip_hdr_to_payload_w =
3651 hlen_w - ((sizeof(struct ipv6hdr)) >> 1);
3654 pbd2->tcp_send_seq = bswab32(inner_tcp_hdr(skb)->seq);
3656 pbd2->tcp_flags = pbd_tcp_flags(inner_tcp_hdr(skb));
3658 if (xmit_type & XMIT_GSO_V4) {
3659 pbd2->hw_ip_id = bswab16(inner_ip_hdr(skb)->id);
3661 pbd_e2->data.tunnel_data.pseudo_csum =
3662 bswab16(~csum_tcpudp_magic(
3663 inner_ip_hdr(skb)->saddr,
3664 inner_ip_hdr(skb)->daddr,
3665 0, IPPROTO_TCP, 0));
3667 outerip_len = ip_hdr(skb)->ihl << 1;
3669 pbd_e2->data.tunnel_data.pseudo_csum =
3670 bswab16(~csum_ipv6_magic(
3671 &inner_ipv6_hdr(skb)->saddr,
3672 &inner_ipv6_hdr(skb)->daddr,
3673 0, IPPROTO_TCP, 0));
3676 outerip_off = (skb_network_header(skb) - skb->data) >> 1;
3680 (!!(xmit_type & XMIT_CSUM_V6) <<
3681 ETH_TX_PARSE_2ND_BD_IP_HDR_TYPE_OUTER_SHIFT) |
3683 ETH_TX_PARSE_2ND_BD_IP_HDR_LEN_OUTER_W_SHIFT) |
3684 ((skb->protocol == cpu_to_be16(ETH_P_8021Q)) <<
3685 ETH_TX_PARSE_2ND_BD_LLC_SNAP_EN_SHIFT);
3687 if (ip_hdr(skb)->protocol == IPPROTO_UDP) {
3688 SET_FLAG(*global_data, ETH_TX_PARSE_2ND_BD_TUNNEL_UDP_EXIST, 1);
3689 pbd2->tunnel_udp_hdr_start_w = skb_transport_offset(skb) >> 1;
3693 /* called with netif_tx_lock
3694 * bnx2x_tx_int() runs without netif_tx_lock unless it needs to call
3695 * netif_wake_queue()
3697 netdev_tx_t bnx2x_start_xmit(struct sk_buff *skb, struct net_device *dev)
3699 struct bnx2x *bp = netdev_priv(dev);
3701 struct netdev_queue *txq;
3702 struct bnx2x_fp_txdata *txdata;
3703 struct sw_tx_bd *tx_buf;
3704 struct eth_tx_start_bd *tx_start_bd, *first_bd;
3705 struct eth_tx_bd *tx_data_bd, *total_pkt_bd = NULL;
3706 struct eth_tx_parse_bd_e1x *pbd_e1x = NULL;
3707 struct eth_tx_parse_bd_e2 *pbd_e2 = NULL;
3708 struct eth_tx_parse_2nd_bd *pbd2 = NULL;
3709 u32 pbd_e2_parsing_data = 0;
3710 u16 pkt_prod, bd_prod;
3713 u32 xmit_type = bnx2x_xmit_type(bp, skb);
3716 __le16 pkt_size = 0;
3718 u8 mac_type = UNICAST_ADDRESS;
3720 #ifdef BNX2X_STOP_ON_ERROR
3721 if (unlikely(bp->panic))
3722 return NETDEV_TX_BUSY;
3725 txq_index = skb_get_queue_mapping(skb);
3726 txq = netdev_get_tx_queue(dev, txq_index);
3728 BUG_ON(txq_index >= MAX_ETH_TXQ_IDX(bp) + (CNIC_LOADED(bp) ? 1 : 0));
3730 txdata = &bp->bnx2x_txq[txq_index];
3732 /* enable this debug print to view the transmission queue being used
3733 DP(NETIF_MSG_TX_QUEUED, "indices: txq %d, fp %d, txdata %d\n",
3734 txq_index, fp_index, txdata_index); */
3736 /* enable this debug print to view the transmission details
3737 DP(NETIF_MSG_TX_QUEUED,
3738 "transmitting packet cid %d fp index %d txdata_index %d tx_data ptr %p fp pointer %p\n",
3739 txdata->cid, fp_index, txdata_index, txdata, fp); */
3741 if (unlikely(bnx2x_tx_avail(bp, txdata) <
3742 skb_shinfo(skb)->nr_frags +
3744 NEXT_CNT_PER_TX_PKT(MAX_BDS_PER_TX_PKT))) {
3745 /* Handle special storage cases separately */
3746 if (txdata->tx_ring_size == 0) {
3747 struct bnx2x_eth_q_stats *q_stats =
3748 bnx2x_fp_qstats(bp, txdata->parent_fp);
3749 q_stats->driver_filtered_tx_pkt++;
3751 return NETDEV_TX_OK;
3753 bnx2x_fp_qstats(bp, txdata->parent_fp)->driver_xoff++;
3754 netif_tx_stop_queue(txq);
3755 BNX2X_ERR("BUG! Tx ring full when queue awake!\n");
3757 return NETDEV_TX_BUSY;
3760 DP(NETIF_MSG_TX_QUEUED,
3761 "queue[%d]: SKB: summed %x protocol %x protocol(%x,%x) gso type %x xmit_type %x len %d\n",
3762 txq_index, skb->ip_summed, skb->protocol, ipv6_hdr(skb)->nexthdr,
3763 ip_hdr(skb)->protocol, skb_shinfo(skb)->gso_type, xmit_type,
3766 eth = (struct ethhdr *)skb->data;
3768 /* set flag according to packet type (UNICAST_ADDRESS is default)*/
3769 if (unlikely(is_multicast_ether_addr(eth->h_dest))) {
3770 if (is_broadcast_ether_addr(eth->h_dest))
3771 mac_type = BROADCAST_ADDRESS;
3773 mac_type = MULTICAST_ADDRESS;
3776 #if (MAX_SKB_FRAGS >= MAX_FETCH_BD - BDS_PER_TX_PKT)
3777 /* First, check if we need to linearize the skb (due to FW
3778 restrictions). No need to check fragmentation if page size > 8K
3779 (there will be no violation to FW restrictions) */
3780 if (bnx2x_pkt_req_lin(bp, skb, xmit_type)) {
3781 /* Statistics of linearization */
3783 if (skb_linearize(skb) != 0) {
3784 DP(NETIF_MSG_TX_QUEUED,
3785 "SKB linearization failed - silently dropping this SKB\n");
3786 dev_kfree_skb_any(skb);
3787 return NETDEV_TX_OK;
3791 /* Map skb linear data for DMA */
3792 mapping = dma_map_single(&bp->pdev->dev, skb->data,
3793 skb_headlen(skb), DMA_TO_DEVICE);
3794 if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
3795 DP(NETIF_MSG_TX_QUEUED,
3796 "SKB mapping failed - silently dropping this SKB\n");
3797 dev_kfree_skb_any(skb);
3798 return NETDEV_TX_OK;
3801 Please read carefully. First we use one BD which we mark as start,
3802 then we have a parsing info BD (used for TSO or xsum),
3803 and only then we have the rest of the TSO BDs.
3804 (don't forget to mark the last one as last,
3805 and to unmap only AFTER you write to the BD ...)
3806 And above all, all pdb sizes are in words - NOT DWORDS!
3809 /* get current pkt produced now - advance it just before sending packet
3810 * since mapping of pages may fail and cause packet to be dropped
3812 pkt_prod = txdata->tx_pkt_prod;
3813 bd_prod = TX_BD(txdata->tx_bd_prod);
3815 /* get a tx_buf and first BD
3816 * tx_start_bd may be changed during SPLIT,
3817 * but first_bd will always stay first
3819 tx_buf = &txdata->tx_buf_ring[TX_BD(pkt_prod)];
3820 tx_start_bd = &txdata->tx_desc_ring[bd_prod].start_bd;
3821 first_bd = tx_start_bd;
3823 tx_start_bd->bd_flags.as_bitfield = ETH_TX_BD_FLAGS_START_BD;
3825 /* header nbd: indirectly zero other flags! */
3826 tx_start_bd->general_data = 1 << ETH_TX_START_BD_HDR_NBDS_SHIFT;
3828 /* remember the first BD of the packet */
3829 tx_buf->first_bd = txdata->tx_bd_prod;
3833 DP(NETIF_MSG_TX_QUEUED,
3834 "sending pkt %u @%p next_idx %u bd %u @%p\n",
3835 pkt_prod, tx_buf, txdata->tx_pkt_prod, bd_prod, tx_start_bd);
3837 if (vlan_tx_tag_present(skb)) {
3838 tx_start_bd->vlan_or_ethertype =
3839 cpu_to_le16(vlan_tx_tag_get(skb));
3840 tx_start_bd->bd_flags.as_bitfield |=
3841 (X_ETH_OUTBAND_VLAN << ETH_TX_BD_FLAGS_VLAN_MODE_SHIFT);
3843 /* when transmitting in a vf, start bd must hold the ethertype
3844 * for fw to enforce it
3847 tx_start_bd->vlan_or_ethertype =
3848 cpu_to_le16(ntohs(eth->h_proto));
3850 /* used by FW for packet accounting */
3851 tx_start_bd->vlan_or_ethertype = cpu_to_le16(pkt_prod);
3854 nbd = 2; /* start_bd + pbd + frags (updated when pages are mapped) */
3856 /* turn on parsing and get a BD */
3857 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3859 if (xmit_type & XMIT_CSUM)
3860 bnx2x_set_sbd_csum(bp, skb, tx_start_bd, xmit_type);
3862 if (!CHIP_IS_E1x(bp)) {
3863 pbd_e2 = &txdata->tx_desc_ring[bd_prod].parse_bd_e2;
3864 memset(pbd_e2, 0, sizeof(struct eth_tx_parse_bd_e2));
3866 if (xmit_type & XMIT_CSUM_ENC) {
3867 u16 global_data = 0;
3869 /* Set PBD in enc checksum offload case */
3870 hlen = bnx2x_set_pbd_csum_enc(bp, skb,
3871 &pbd_e2_parsing_data,
3874 /* turn on 2nd parsing and get a BD */
3875 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3877 pbd2 = &txdata->tx_desc_ring[bd_prod].parse_2nd_bd;
3879 memset(pbd2, 0, sizeof(*pbd2));
3881 pbd_e2->data.tunnel_data.ip_hdr_start_inner_w =
3882 (skb_inner_network_header(skb) -
3885 if (xmit_type & XMIT_GSO_ENC)
3886 bnx2x_update_pbds_gso_enc(skb, pbd_e2, pbd2,
3890 pbd2->global_data = cpu_to_le16(global_data);
3892 /* add addition parse BD indication to start BD */
3893 SET_FLAG(tx_start_bd->general_data,
3894 ETH_TX_START_BD_PARSE_NBDS, 1);
3895 /* set encapsulation flag in start BD */
3896 SET_FLAG(tx_start_bd->general_data,
3897 ETH_TX_START_BD_TUNNEL_EXIST, 1);
3899 tx_buf->flags |= BNX2X_HAS_SECOND_PBD;
3902 } else if (xmit_type & XMIT_CSUM) {
3903 /* Set PBD in checksum offload case w/o encapsulation */
3904 hlen = bnx2x_set_pbd_csum_e2(bp, skb,
3905 &pbd_e2_parsing_data,
3909 /* Add the macs to the parsing BD if this is a vf or if
3910 * Tx Switching is enabled.
3913 /* override GRE parameters in BD */
3914 bnx2x_set_fw_mac_addr(&pbd_e2->data.mac_addr.src_hi,
3915 &pbd_e2->data.mac_addr.src_mid,
3916 &pbd_e2->data.mac_addr.src_lo,
3919 bnx2x_set_fw_mac_addr(&pbd_e2->data.mac_addr.dst_hi,
3920 &pbd_e2->data.mac_addr.dst_mid,
3921 &pbd_e2->data.mac_addr.dst_lo,
3923 } else if (bp->flags & TX_SWITCHING) {
3924 bnx2x_set_fw_mac_addr(&pbd_e2->data.mac_addr.dst_hi,
3925 &pbd_e2->data.mac_addr.dst_mid,
3926 &pbd_e2->data.mac_addr.dst_lo,
3930 SET_FLAG(pbd_e2_parsing_data,
3931 ETH_TX_PARSE_BD_E2_ETH_ADDR_TYPE, mac_type);
3933 u16 global_data = 0;
3934 pbd_e1x = &txdata->tx_desc_ring[bd_prod].parse_bd_e1x;
3935 memset(pbd_e1x, 0, sizeof(struct eth_tx_parse_bd_e1x));
3936 /* Set PBD in checksum offload case */
3937 if (xmit_type & XMIT_CSUM)
3938 hlen = bnx2x_set_pbd_csum(bp, skb, pbd_e1x, xmit_type);
3940 SET_FLAG(global_data,
3941 ETH_TX_PARSE_BD_E1X_ETH_ADDR_TYPE, mac_type);
3942 pbd_e1x->global_data |= cpu_to_le16(global_data);
3945 /* Setup the data pointer of the first BD of the packet */
3946 tx_start_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
3947 tx_start_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
3948 tx_start_bd->nbytes = cpu_to_le16(skb_headlen(skb));
3949 pkt_size = tx_start_bd->nbytes;
3951 DP(NETIF_MSG_TX_QUEUED,
3952 "first bd @%p addr (%x:%x) nbytes %d flags %x vlan %x\n",
3953 tx_start_bd, tx_start_bd->addr_hi, tx_start_bd->addr_lo,
3954 le16_to_cpu(tx_start_bd->nbytes),
3955 tx_start_bd->bd_flags.as_bitfield,
3956 le16_to_cpu(tx_start_bd->vlan_or_ethertype));
3958 if (xmit_type & XMIT_GSO) {
3960 DP(NETIF_MSG_TX_QUEUED,
3961 "TSO packet len %d hlen %d total len %d tso size %d\n",
3962 skb->len, hlen, skb_headlen(skb),
3963 skb_shinfo(skb)->gso_size);
3965 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_SW_LSO;
3967 if (unlikely(skb_headlen(skb) > hlen)) {
3969 bd_prod = bnx2x_tx_split(bp, txdata, tx_buf,
3973 if (!CHIP_IS_E1x(bp))
3974 bnx2x_set_pbd_gso_e2(skb, &pbd_e2_parsing_data,
3977 bnx2x_set_pbd_gso(skb, pbd_e1x, first_bd, xmit_type);
3980 /* Set the PBD's parsing_data field if not zero
3981 * (for the chips newer than 57711).
3983 if (pbd_e2_parsing_data)
3984 pbd_e2->parsing_data = cpu_to_le32(pbd_e2_parsing_data);
3986 tx_data_bd = (struct eth_tx_bd *)tx_start_bd;
3988 /* Handle fragmented skb */
3989 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3990 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3992 mapping = skb_frag_dma_map(&bp->pdev->dev, frag, 0,
3993 skb_frag_size(frag), DMA_TO_DEVICE);
3994 if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
3995 unsigned int pkts_compl = 0, bytes_compl = 0;
3997 DP(NETIF_MSG_TX_QUEUED,
3998 "Unable to map page - dropping packet...\n");
4000 /* we need unmap all buffers already mapped
4002 * first_bd->nbd need to be properly updated
4003 * before call to bnx2x_free_tx_pkt
4005 first_bd->nbd = cpu_to_le16(nbd);
4006 bnx2x_free_tx_pkt(bp, txdata,
4007 TX_BD(txdata->tx_pkt_prod),
4008 &pkts_compl, &bytes_compl);
4009 return NETDEV_TX_OK;
4012 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
4013 tx_data_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
4014 if (total_pkt_bd == NULL)
4015 total_pkt_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
4017 tx_data_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
4018 tx_data_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
4019 tx_data_bd->nbytes = cpu_to_le16(skb_frag_size(frag));
4020 le16_add_cpu(&pkt_size, skb_frag_size(frag));
4023 DP(NETIF_MSG_TX_QUEUED,
4024 "frag %d bd @%p addr (%x:%x) nbytes %d\n",
4025 i, tx_data_bd, tx_data_bd->addr_hi, tx_data_bd->addr_lo,
4026 le16_to_cpu(tx_data_bd->nbytes));
4029 DP(NETIF_MSG_TX_QUEUED, "last bd @%p\n", tx_data_bd);
4031 /* update with actual num BDs */
4032 first_bd->nbd = cpu_to_le16(nbd);
4034 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
4036 /* now send a tx doorbell, counting the next BD
4037 * if the packet contains or ends with it
4039 if (TX_BD_POFF(bd_prod) < nbd)
4042 /* total_pkt_bytes should be set on the first data BD if
4043 * it's not an LSO packet and there is more than one
4044 * data BD. In this case pkt_size is limited by an MTU value.
4045 * However we prefer to set it for an LSO packet (while we don't
4046 * have to) in order to save some CPU cycles in a none-LSO
4047 * case, when we much more care about them.
4049 if (total_pkt_bd != NULL)
4050 total_pkt_bd->total_pkt_bytes = pkt_size;
4053 DP(NETIF_MSG_TX_QUEUED,
4054 "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",
4055 pbd_e1x, pbd_e1x->global_data, pbd_e1x->ip_hlen_w,
4056 pbd_e1x->ip_id, pbd_e1x->lso_mss, pbd_e1x->tcp_flags,
4057 pbd_e1x->tcp_pseudo_csum, pbd_e1x->tcp_send_seq,
4058 le16_to_cpu(pbd_e1x->total_hlen_w));
4060 DP(NETIF_MSG_TX_QUEUED,
4061 "PBD (E2) @%p dst %x %x %x src %x %x %x parsing_data %x\n",
4063 pbd_e2->data.mac_addr.dst_hi,
4064 pbd_e2->data.mac_addr.dst_mid,
4065 pbd_e2->data.mac_addr.dst_lo,
4066 pbd_e2->data.mac_addr.src_hi,
4067 pbd_e2->data.mac_addr.src_mid,
4068 pbd_e2->data.mac_addr.src_lo,
4069 pbd_e2->parsing_data);
4070 DP(NETIF_MSG_TX_QUEUED, "doorbell: nbd %d bd %u\n", nbd, bd_prod);
4072 netdev_tx_sent_queue(txq, skb->len);
4074 skb_tx_timestamp(skb);
4076 txdata->tx_pkt_prod++;
4078 * Make sure that the BD data is updated before updating the producer
4079 * since FW might read the BD right after the producer is updated.
4080 * This is only applicable for weak-ordered memory model archs such
4081 * as IA-64. The following barrier is also mandatory since FW will
4082 * assumes packets must have BDs.
4086 txdata->tx_db.data.prod += nbd;
4089 DOORBELL(bp, txdata->cid, txdata->tx_db.raw);
4093 txdata->tx_bd_prod += nbd;
4095 if (unlikely(bnx2x_tx_avail(bp, txdata) < MAX_DESC_PER_TX_PKT)) {
4096 netif_tx_stop_queue(txq);
4098 /* paired memory barrier is in bnx2x_tx_int(), we have to keep
4099 * ordering of set_bit() in netif_tx_stop_queue() and read of
4103 bnx2x_fp_qstats(bp, txdata->parent_fp)->driver_xoff++;
4104 if (bnx2x_tx_avail(bp, txdata) >= MAX_DESC_PER_TX_PKT)
4105 netif_tx_wake_queue(txq);
4109 return NETDEV_TX_OK;
4113 * bnx2x_setup_tc - routine to configure net_device for multi tc
4115 * @netdev: net device to configure
4116 * @tc: number of traffic classes to enable
4118 * callback connected to the ndo_setup_tc function pointer
4120 int bnx2x_setup_tc(struct net_device *dev, u8 num_tc)
4122 int cos, prio, count, offset;
4123 struct bnx2x *bp = netdev_priv(dev);
4125 /* setup tc must be called under rtnl lock */
4128 /* no traffic classes requested. Aborting */
4130 netdev_reset_tc(dev);
4134 /* requested to support too many traffic classes */
4135 if (num_tc > bp->max_cos) {
4136 BNX2X_ERR("support for too many traffic classes requested: %d. Max supported is %d\n",
4137 num_tc, bp->max_cos);
4141 /* declare amount of supported traffic classes */
4142 if (netdev_set_num_tc(dev, num_tc)) {
4143 BNX2X_ERR("failed to declare %d traffic classes\n", num_tc);
4147 /* configure priority to traffic class mapping */
4148 for (prio = 0; prio < BNX2X_MAX_PRIORITY; prio++) {
4149 netdev_set_prio_tc_map(dev, prio, bp->prio_to_cos[prio]);
4150 DP(BNX2X_MSG_SP | NETIF_MSG_IFUP,
4151 "mapping priority %d to tc %d\n",
4152 prio, bp->prio_to_cos[prio]);
4155 /* Use this configuration to differentiate tc0 from other COSes
4156 This can be used for ets or pfc, and save the effort of setting
4157 up a multio class queue disc or negotiating DCBX with a switch
4158 netdev_set_prio_tc_map(dev, 0, 0);
4159 DP(BNX2X_MSG_SP, "mapping priority %d to tc %d\n", 0, 0);
4160 for (prio = 1; prio < 16; prio++) {
4161 netdev_set_prio_tc_map(dev, prio, 1);
4162 DP(BNX2X_MSG_SP, "mapping priority %d to tc %d\n", prio, 1);
4165 /* configure traffic class to transmission queue mapping */
4166 for (cos = 0; cos < bp->max_cos; cos++) {
4167 count = BNX2X_NUM_ETH_QUEUES(bp);
4168 offset = cos * BNX2X_NUM_NON_CNIC_QUEUES(bp);
4169 netdev_set_tc_queue(dev, cos, count, offset);
4170 DP(BNX2X_MSG_SP | NETIF_MSG_IFUP,
4171 "mapping tc %d to offset %d count %d\n",
4172 cos, offset, count);
4178 /* called with rtnl_lock */
4179 int bnx2x_change_mac_addr(struct net_device *dev, void *p)
4181 struct sockaddr *addr = p;
4182 struct bnx2x *bp = netdev_priv(dev);
4185 if (!bnx2x_is_valid_ether_addr(bp, addr->sa_data)) {
4186 BNX2X_ERR("Requested MAC address is not valid\n");
4190 if ((IS_MF_STORAGE_SD(bp) || IS_MF_FCOE_AFEX(bp)) &&
4191 !is_zero_ether_addr(addr->sa_data)) {
4192 BNX2X_ERR("Can't configure non-zero address on iSCSI or FCoE functions in MF-SD mode\n");
4196 if (netif_running(dev)) {
4197 rc = bnx2x_set_eth_mac(bp, false);
4202 memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
4204 if (netif_running(dev))
4205 rc = bnx2x_set_eth_mac(bp, true);
4210 static void bnx2x_free_fp_mem_at(struct bnx2x *bp, int fp_index)
4212 union host_hc_status_block *sb = &bnx2x_fp(bp, fp_index, status_blk);
4213 struct bnx2x_fastpath *fp = &bp->fp[fp_index];
4218 if (IS_FCOE_IDX(fp_index)) {
4219 memset(sb, 0, sizeof(union host_hc_status_block));
4220 fp->status_blk_mapping = 0;
4223 if (!CHIP_IS_E1x(bp))
4224 BNX2X_PCI_FREE(sb->e2_sb,
4225 bnx2x_fp(bp, fp_index,
4226 status_blk_mapping),
4227 sizeof(struct host_hc_status_block_e2));
4229 BNX2X_PCI_FREE(sb->e1x_sb,
4230 bnx2x_fp(bp, fp_index,
4231 status_blk_mapping),
4232 sizeof(struct host_hc_status_block_e1x));
4236 if (!skip_rx_queue(bp, fp_index)) {
4237 bnx2x_free_rx_bds(fp);
4239 /* fastpath rx rings: rx_buf rx_desc rx_comp */
4240 BNX2X_FREE(bnx2x_fp(bp, fp_index, rx_buf_ring));
4241 BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_desc_ring),
4242 bnx2x_fp(bp, fp_index, rx_desc_mapping),
4243 sizeof(struct eth_rx_bd) * NUM_RX_BD);
4245 BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_comp_ring),
4246 bnx2x_fp(bp, fp_index, rx_comp_mapping),
4247 sizeof(struct eth_fast_path_rx_cqe) *
4251 BNX2X_FREE(bnx2x_fp(bp, fp_index, rx_page_ring));
4252 BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_sge_ring),
4253 bnx2x_fp(bp, fp_index, rx_sge_mapping),
4254 BCM_PAGE_SIZE * NUM_RX_SGE_PAGES);
4258 if (!skip_tx_queue(bp, fp_index)) {
4259 /* fastpath tx rings: tx_buf tx_desc */
4260 for_each_cos_in_tx_queue(fp, cos) {
4261 struct bnx2x_fp_txdata *txdata = fp->txdata_ptr[cos];
4263 DP(NETIF_MSG_IFDOWN,
4264 "freeing tx memory of fp %d cos %d cid %d\n",
4265 fp_index, cos, txdata->cid);
4267 BNX2X_FREE(txdata->tx_buf_ring);
4268 BNX2X_PCI_FREE(txdata->tx_desc_ring,
4269 txdata->tx_desc_mapping,
4270 sizeof(union eth_tx_bd_types) * NUM_TX_BD);
4273 /* end of fastpath */
4276 static void bnx2x_free_fp_mem_cnic(struct bnx2x *bp)
4279 for_each_cnic_queue(bp, i)
4280 bnx2x_free_fp_mem_at(bp, i);
4283 void bnx2x_free_fp_mem(struct bnx2x *bp)
4286 for_each_eth_queue(bp, i)
4287 bnx2x_free_fp_mem_at(bp, i);
4290 static void set_sb_shortcuts(struct bnx2x *bp, int index)
4292 union host_hc_status_block status_blk = bnx2x_fp(bp, index, status_blk);
4293 if (!CHIP_IS_E1x(bp)) {
4294 bnx2x_fp(bp, index, sb_index_values) =
4295 (__le16 *)status_blk.e2_sb->sb.index_values;
4296 bnx2x_fp(bp, index, sb_running_index) =
4297 (__le16 *)status_blk.e2_sb->sb.running_index;
4299 bnx2x_fp(bp, index, sb_index_values) =
4300 (__le16 *)status_blk.e1x_sb->sb.index_values;
4301 bnx2x_fp(bp, index, sb_running_index) =
4302 (__le16 *)status_blk.e1x_sb->sb.running_index;
4306 /* Returns the number of actually allocated BDs */
4307 static int bnx2x_alloc_rx_bds(struct bnx2x_fastpath *fp,
4310 struct bnx2x *bp = fp->bp;
4311 u16 ring_prod, cqe_ring_prod;
4312 int i, failure_cnt = 0;
4314 fp->rx_comp_cons = 0;
4315 cqe_ring_prod = ring_prod = 0;
4317 /* This routine is called only during fo init so
4318 * fp->eth_q_stats.rx_skb_alloc_failed = 0
4320 for (i = 0; i < rx_ring_size; i++) {
4321 if (bnx2x_alloc_rx_data(bp, fp, ring_prod, GFP_KERNEL) < 0) {
4325 ring_prod = NEXT_RX_IDX(ring_prod);
4326 cqe_ring_prod = NEXT_RCQ_IDX(cqe_ring_prod);
4327 WARN_ON(ring_prod <= (i - failure_cnt));
4331 BNX2X_ERR("was only able to allocate %d rx skbs on queue[%d]\n",
4332 i - failure_cnt, fp->index);
4334 fp->rx_bd_prod = ring_prod;
4335 /* Limit the CQE producer by the CQE ring size */
4336 fp->rx_comp_prod = min_t(u16, NUM_RCQ_RINGS*RCQ_DESC_CNT,
4338 fp->rx_pkt = fp->rx_calls = 0;
4340 bnx2x_fp_stats(bp, fp)->eth_q_stats.rx_skb_alloc_failed += failure_cnt;
4342 return i - failure_cnt;
4345 static void bnx2x_set_next_page_rx_cq(struct bnx2x_fastpath *fp)
4349 for (i = 1; i <= NUM_RCQ_RINGS; i++) {
4350 struct eth_rx_cqe_next_page *nextpg;
4352 nextpg = (struct eth_rx_cqe_next_page *)
4353 &fp->rx_comp_ring[RCQ_DESC_CNT * i - 1];
4355 cpu_to_le32(U64_HI(fp->rx_comp_mapping +
4356 BCM_PAGE_SIZE*(i % NUM_RCQ_RINGS)));
4358 cpu_to_le32(U64_LO(fp->rx_comp_mapping +
4359 BCM_PAGE_SIZE*(i % NUM_RCQ_RINGS)));
4363 static int bnx2x_alloc_fp_mem_at(struct bnx2x *bp, int index)
4365 union host_hc_status_block *sb;
4366 struct bnx2x_fastpath *fp = &bp->fp[index];
4369 int rx_ring_size = 0;
4371 if (!bp->rx_ring_size &&
4372 (IS_MF_STORAGE_SD(bp) || IS_MF_FCOE_AFEX(bp))) {
4373 rx_ring_size = MIN_RX_SIZE_NONTPA;
4374 bp->rx_ring_size = rx_ring_size;
4375 } else if (!bp->rx_ring_size) {
4376 rx_ring_size = MAX_RX_AVAIL/BNX2X_NUM_RX_QUEUES(bp);
4378 if (CHIP_IS_E3(bp)) {
4379 u32 cfg = SHMEM_RD(bp,
4380 dev_info.port_hw_config[BP_PORT(bp)].
4383 /* Decrease ring size for 1G functions */
4384 if ((cfg & PORT_HW_CFG_NET_SERDES_IF_MASK) ==
4385 PORT_HW_CFG_NET_SERDES_IF_SGMII)
4389 /* allocate at least number of buffers required by FW */
4390 rx_ring_size = max_t(int, bp->disable_tpa ? MIN_RX_SIZE_NONTPA :
4391 MIN_RX_SIZE_TPA, rx_ring_size);
4393 bp->rx_ring_size = rx_ring_size;
4394 } else /* if rx_ring_size specified - use it */
4395 rx_ring_size = bp->rx_ring_size;
4397 DP(BNX2X_MSG_SP, "calculated rx_ring_size %d\n", rx_ring_size);
4400 sb = &bnx2x_fp(bp, index, status_blk);
4402 if (!IS_FCOE_IDX(index)) {
4404 if (!CHIP_IS_E1x(bp)) {
4405 sb->e2_sb = BNX2X_PCI_ALLOC(&bnx2x_fp(bp, index, status_blk_mapping),
4406 sizeof(struct host_hc_status_block_e2));
4410 sb->e1x_sb = BNX2X_PCI_ALLOC(&bnx2x_fp(bp, index, status_blk_mapping),
4411 sizeof(struct host_hc_status_block_e1x));
4417 /* FCoE Queue uses Default SB and doesn't ACK the SB, thus no need to
4418 * set shortcuts for it.
4420 if (!IS_FCOE_IDX(index))
4421 set_sb_shortcuts(bp, index);
4424 if (!skip_tx_queue(bp, index)) {
4425 /* fastpath tx rings: tx_buf tx_desc */
4426 for_each_cos_in_tx_queue(fp, cos) {
4427 struct bnx2x_fp_txdata *txdata = fp->txdata_ptr[cos];
4430 "allocating tx memory of fp %d cos %d\n",
4433 txdata->tx_buf_ring = kcalloc(NUM_TX_BD,
4434 sizeof(struct sw_tx_bd),
4436 if (!txdata->tx_buf_ring)
4438 txdata->tx_desc_ring = BNX2X_PCI_ALLOC(&txdata->tx_desc_mapping,
4439 sizeof(union eth_tx_bd_types) * NUM_TX_BD);
4440 if (!txdata->tx_desc_ring)
4446 if (!skip_rx_queue(bp, index)) {
4447 /* fastpath rx rings: rx_buf rx_desc rx_comp */
4448 bnx2x_fp(bp, index, rx_buf_ring) =
4449 kcalloc(NUM_RX_BD, sizeof(struct sw_rx_bd), GFP_KERNEL);
4450 if (!bnx2x_fp(bp, index, rx_buf_ring))
4452 bnx2x_fp(bp, index, rx_desc_ring) =
4453 BNX2X_PCI_ALLOC(&bnx2x_fp(bp, index, rx_desc_mapping),
4454 sizeof(struct eth_rx_bd) * NUM_RX_BD);
4455 if (!bnx2x_fp(bp, index, rx_desc_ring))
4458 /* Seed all CQEs by 1s */
4459 bnx2x_fp(bp, index, rx_comp_ring) =
4460 BNX2X_PCI_FALLOC(&bnx2x_fp(bp, index, rx_comp_mapping),
4461 sizeof(struct eth_fast_path_rx_cqe) * NUM_RCQ_BD);
4462 if (!bnx2x_fp(bp, index, rx_comp_ring))
4466 bnx2x_fp(bp, index, rx_page_ring) =
4467 kcalloc(NUM_RX_SGE, sizeof(struct sw_rx_page),
4469 if (!bnx2x_fp(bp, index, rx_page_ring))
4471 bnx2x_fp(bp, index, rx_sge_ring) =
4472 BNX2X_PCI_ALLOC(&bnx2x_fp(bp, index, rx_sge_mapping),
4473 BCM_PAGE_SIZE * NUM_RX_SGE_PAGES);
4474 if (!bnx2x_fp(bp, index, rx_sge_ring))
4477 bnx2x_set_next_page_rx_bd(fp);
4480 bnx2x_set_next_page_rx_cq(fp);
4483 ring_size = bnx2x_alloc_rx_bds(fp, rx_ring_size);
4484 if (ring_size < rx_ring_size)
4490 /* handles low memory cases */
4492 BNX2X_ERR("Unable to allocate full memory for queue %d (size %d)\n",
4494 /* FW will drop all packets if queue is not big enough,
4495 * In these cases we disable the queue
4496 * Min size is different for OOO, TPA and non-TPA queues
4498 if (ring_size < (fp->disable_tpa ?
4499 MIN_RX_SIZE_NONTPA : MIN_RX_SIZE_TPA)) {
4500 /* release memory allocated for this queue */
4501 bnx2x_free_fp_mem_at(bp, index);
4507 static int bnx2x_alloc_fp_mem_cnic(struct bnx2x *bp)
4511 if (bnx2x_alloc_fp_mem_at(bp, FCOE_IDX(bp)))
4512 /* we will fail load process instead of mark
4520 static int bnx2x_alloc_fp_mem(struct bnx2x *bp)
4524 /* 1. Allocate FP for leading - fatal if error
4525 * 2. Allocate RSS - fix number of queues if error
4529 if (bnx2x_alloc_fp_mem_at(bp, 0))
4533 for_each_nondefault_eth_queue(bp, i)
4534 if (bnx2x_alloc_fp_mem_at(bp, i))
4537 /* handle memory failures */
4538 if (i != BNX2X_NUM_ETH_QUEUES(bp)) {
4539 int delta = BNX2X_NUM_ETH_QUEUES(bp) - i;
4542 bnx2x_shrink_eth_fp(bp, delta);
4543 if (CNIC_SUPPORT(bp))
4544 /* move non eth FPs next to last eth FP
4545 * must be done in that order
4546 * FCOE_IDX < FWD_IDX < OOO_IDX
4549 /* move FCoE fp even NO_FCOE_FLAG is on */
4550 bnx2x_move_fp(bp, FCOE_IDX(bp), FCOE_IDX(bp) - delta);
4551 bp->num_ethernet_queues -= delta;
4552 bp->num_queues = bp->num_ethernet_queues +
4553 bp->num_cnic_queues;
4554 BNX2X_ERR("Adjusted num of queues from %d to %d\n",
4555 bp->num_queues + delta, bp->num_queues);
4561 void bnx2x_free_mem_bp(struct bnx2x *bp)
4565 for (i = 0; i < bp->fp_array_size; i++)
4566 kfree(bp->fp[i].tpa_info);
4569 kfree(bp->fp_stats);
4570 kfree(bp->bnx2x_txq);
4571 kfree(bp->msix_table);
4575 int bnx2x_alloc_mem_bp(struct bnx2x *bp)
4577 struct bnx2x_fastpath *fp;
4578 struct msix_entry *tbl;
4579 struct bnx2x_ilt *ilt;
4580 int msix_table_size = 0;
4581 int fp_array_size, txq_array_size;
4585 * The biggest MSI-X table we might need is as a maximum number of fast
4586 * path IGU SBs plus default SB (for PF only).
4588 msix_table_size = bp->igu_sb_cnt;
4591 BNX2X_DEV_INFO("msix_table_size %d\n", msix_table_size);
4593 /* fp array: RSS plus CNIC related L2 queues */
4594 fp_array_size = BNX2X_MAX_RSS_COUNT(bp) + CNIC_SUPPORT(bp);
4595 bp->fp_array_size = fp_array_size;
4596 BNX2X_DEV_INFO("fp_array_size %d\n", bp->fp_array_size);
4598 fp = kcalloc(bp->fp_array_size, sizeof(*fp), GFP_KERNEL);
4601 for (i = 0; i < bp->fp_array_size; i++) {
4603 kcalloc(ETH_MAX_AGGREGATION_QUEUES_E1H_E2,
4604 sizeof(struct bnx2x_agg_info), GFP_KERNEL);
4605 if (!(fp[i].tpa_info))
4611 /* allocate sp objs */
4612 bp->sp_objs = kcalloc(bp->fp_array_size, sizeof(struct bnx2x_sp_objs),
4617 /* allocate fp_stats */
4618 bp->fp_stats = kcalloc(bp->fp_array_size, sizeof(struct bnx2x_fp_stats),
4623 /* Allocate memory for the transmission queues array */
4625 BNX2X_MAX_RSS_COUNT(bp) * BNX2X_MULTI_TX_COS + CNIC_SUPPORT(bp);
4626 BNX2X_DEV_INFO("txq_array_size %d", txq_array_size);
4628 bp->bnx2x_txq = kcalloc(txq_array_size, sizeof(struct bnx2x_fp_txdata),
4634 tbl = kcalloc(msix_table_size, sizeof(*tbl), GFP_KERNEL);
4637 bp->msix_table = tbl;
4640 ilt = kzalloc(sizeof(*ilt), GFP_KERNEL);
4647 bnx2x_free_mem_bp(bp);
4651 int bnx2x_reload_if_running(struct net_device *dev)
4653 struct bnx2x *bp = netdev_priv(dev);
4655 if (unlikely(!netif_running(dev)))
4658 bnx2x_nic_unload(bp, UNLOAD_NORMAL, true);
4659 return bnx2x_nic_load(bp, LOAD_NORMAL);
4662 int bnx2x_get_cur_phy_idx(struct bnx2x *bp)
4664 u32 sel_phy_idx = 0;
4665 if (bp->link_params.num_phys <= 1)
4668 if (bp->link_vars.link_up) {
4669 sel_phy_idx = EXT_PHY1;
4670 /* In case link is SERDES, check if the EXT_PHY2 is the one */
4671 if ((bp->link_vars.link_status & LINK_STATUS_SERDES_LINK) &&
4672 (bp->link_params.phy[EXT_PHY2].supported & SUPPORTED_FIBRE))
4673 sel_phy_idx = EXT_PHY2;
4676 switch (bnx2x_phy_selection(&bp->link_params)) {
4677 case PORT_HW_CFG_PHY_SELECTION_HARDWARE_DEFAULT:
4678 case PORT_HW_CFG_PHY_SELECTION_FIRST_PHY:
4679 case PORT_HW_CFG_PHY_SELECTION_FIRST_PHY_PRIORITY:
4680 sel_phy_idx = EXT_PHY1;
4682 case PORT_HW_CFG_PHY_SELECTION_SECOND_PHY:
4683 case PORT_HW_CFG_PHY_SELECTION_SECOND_PHY_PRIORITY:
4684 sel_phy_idx = EXT_PHY2;
4691 int bnx2x_get_link_cfg_idx(struct bnx2x *bp)
4693 u32 sel_phy_idx = bnx2x_get_cur_phy_idx(bp);
4695 * The selected activated PHY is always after swapping (in case PHY
4696 * swapping is enabled). So when swapping is enabled, we need to reverse
4700 if (bp->link_params.multi_phy_config &
4701 PORT_HW_CFG_PHY_SWAPPED_ENABLED) {
4702 if (sel_phy_idx == EXT_PHY1)
4703 sel_phy_idx = EXT_PHY2;
4704 else if (sel_phy_idx == EXT_PHY2)
4705 sel_phy_idx = EXT_PHY1;
4707 return LINK_CONFIG_IDX(sel_phy_idx);
4710 #ifdef NETDEV_FCOE_WWNN
4711 int bnx2x_fcoe_get_wwn(struct net_device *dev, u64 *wwn, int type)
4713 struct bnx2x *bp = netdev_priv(dev);
4714 struct cnic_eth_dev *cp = &bp->cnic_eth_dev;
4717 case NETDEV_FCOE_WWNN:
4718 *wwn = HILO_U64(cp->fcoe_wwn_node_name_hi,
4719 cp->fcoe_wwn_node_name_lo);
4721 case NETDEV_FCOE_WWPN:
4722 *wwn = HILO_U64(cp->fcoe_wwn_port_name_hi,
4723 cp->fcoe_wwn_port_name_lo);
4726 BNX2X_ERR("Wrong WWN type requested - %d\n", type);
4734 /* called with rtnl_lock */
4735 int bnx2x_change_mtu(struct net_device *dev, int new_mtu)
4737 struct bnx2x *bp = netdev_priv(dev);
4739 if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
4740 BNX2X_ERR("Can't perform change MTU during parity recovery\n");
4744 if ((new_mtu > ETH_MAX_JUMBO_PACKET_SIZE) ||
4745 ((new_mtu + ETH_HLEN) < ETH_MIN_PACKET_SIZE)) {
4746 BNX2X_ERR("Can't support requested MTU size\n");
4750 /* This does not race with packet allocation
4751 * because the actual alloc size is
4752 * only updated as part of load
4756 return bnx2x_reload_if_running(dev);
4759 netdev_features_t bnx2x_fix_features(struct net_device *dev,
4760 netdev_features_t features)
4762 struct bnx2x *bp = netdev_priv(dev);
4764 /* TPA requires Rx CSUM offloading */
4765 if (!(features & NETIF_F_RXCSUM) || bp->disable_tpa) {
4766 features &= ~NETIF_F_LRO;
4767 features &= ~NETIF_F_GRO;
4773 int bnx2x_set_features(struct net_device *dev, netdev_features_t features)
4775 struct bnx2x *bp = netdev_priv(dev);
4776 u32 flags = bp->flags;
4778 bool bnx2x_reload = false;
4780 if (features & NETIF_F_LRO)
4781 flags |= TPA_ENABLE_FLAG;
4783 flags &= ~TPA_ENABLE_FLAG;
4785 if (features & NETIF_F_GRO)
4786 flags |= GRO_ENABLE_FLAG;
4788 flags &= ~GRO_ENABLE_FLAG;
4790 if (features & NETIF_F_LOOPBACK) {
4791 if (bp->link_params.loopback_mode != LOOPBACK_BMAC) {
4792 bp->link_params.loopback_mode = LOOPBACK_BMAC;
4793 bnx2x_reload = true;
4796 if (bp->link_params.loopback_mode != LOOPBACK_NONE) {
4797 bp->link_params.loopback_mode = LOOPBACK_NONE;
4798 bnx2x_reload = true;
4802 changes = flags ^ bp->flags;
4804 /* if GRO is changed while LRO is enabled, don't force a reload */
4805 if ((changes & GRO_ENABLE_FLAG) && (flags & TPA_ENABLE_FLAG))
4806 changes &= ~GRO_ENABLE_FLAG;
4809 bnx2x_reload = true;
4814 if (bp->recovery_state == BNX2X_RECOVERY_DONE)
4815 return bnx2x_reload_if_running(dev);
4816 /* else: bnx2x_nic_load() will be called at end of recovery */
4822 void bnx2x_tx_timeout(struct net_device *dev)
4824 struct bnx2x *bp = netdev_priv(dev);
4826 #ifdef BNX2X_STOP_ON_ERROR
4831 /* This allows the netif to be shutdown gracefully before resetting */
4832 bnx2x_schedule_sp_rtnl(bp, BNX2X_SP_RTNL_TX_TIMEOUT, 0);
4835 int bnx2x_suspend(struct pci_dev *pdev, pm_message_t state)
4837 struct net_device *dev = pci_get_drvdata(pdev);
4841 dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
4844 bp = netdev_priv(dev);
4848 pci_save_state(pdev);
4850 if (!netif_running(dev)) {
4855 netif_device_detach(dev);
4857 bnx2x_nic_unload(bp, UNLOAD_CLOSE, false);
4859 bnx2x_set_power_state(bp, pci_choose_state(pdev, state));
4866 int bnx2x_resume(struct pci_dev *pdev)
4868 struct net_device *dev = pci_get_drvdata(pdev);
4873 dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
4876 bp = netdev_priv(dev);
4878 if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
4879 BNX2X_ERR("Handling parity error recovery. Try again later\n");
4885 pci_restore_state(pdev);
4887 if (!netif_running(dev)) {
4892 bnx2x_set_power_state(bp, PCI_D0);
4893 netif_device_attach(dev);
4895 rc = bnx2x_nic_load(bp, LOAD_OPEN);
4902 void bnx2x_set_ctx_validation(struct bnx2x *bp, struct eth_context *cxt,
4906 BNX2X_ERR("bad context pointer %p\n", cxt);
4910 /* ustorm cxt validation */
4911 cxt->ustorm_ag_context.cdu_usage =
4912 CDU_RSRVD_VALUE_TYPE_A(HW_CID(bp, cid),
4913 CDU_REGION_NUMBER_UCM_AG, ETH_CONNECTION_TYPE);
4914 /* xcontext validation */
4915 cxt->xstorm_ag_context.cdu_reserved =
4916 CDU_RSRVD_VALUE_TYPE_A(HW_CID(bp, cid),
4917 CDU_REGION_NUMBER_XCM_AG, ETH_CONNECTION_TYPE);
4920 static void storm_memset_hc_timeout(struct bnx2x *bp, u8 port,
4921 u8 fw_sb_id, u8 sb_index,
4924 u32 addr = BAR_CSTRORM_INTMEM +
4925 CSTORM_STATUS_BLOCK_DATA_TIMEOUT_OFFSET(fw_sb_id, sb_index);
4926 REG_WR8(bp, addr, ticks);
4928 "port %x fw_sb_id %d sb_index %d ticks %d\n",
4929 port, fw_sb_id, sb_index, ticks);
4932 static void storm_memset_hc_disable(struct bnx2x *bp, u8 port,
4933 u16 fw_sb_id, u8 sb_index,
4936 u32 enable_flag = disable ? 0 : (1 << HC_INDEX_DATA_HC_ENABLED_SHIFT);
4937 u32 addr = BAR_CSTRORM_INTMEM +
4938 CSTORM_STATUS_BLOCK_DATA_FLAGS_OFFSET(fw_sb_id, sb_index);
4939 u8 flags = REG_RD8(bp, addr);
4941 flags &= ~HC_INDEX_DATA_HC_ENABLED;
4942 flags |= enable_flag;
4943 REG_WR8(bp, addr, flags);
4945 "port %x fw_sb_id %d sb_index %d disable %d\n",
4946 port, fw_sb_id, sb_index, disable);
4949 void bnx2x_update_coalesce_sb_index(struct bnx2x *bp, u8 fw_sb_id,
4950 u8 sb_index, u8 disable, u16 usec)
4952 int port = BP_PORT(bp);
4953 u8 ticks = usec / BNX2X_BTR;
4955 storm_memset_hc_timeout(bp, port, fw_sb_id, sb_index, ticks);
4957 disable = disable ? 1 : (usec ? 0 : 1);
4958 storm_memset_hc_disable(bp, port, fw_sb_id, sb_index, disable);
4961 void bnx2x_schedule_sp_rtnl(struct bnx2x *bp, enum sp_rtnl_flag flag,
4964 smp_mb__before_atomic();
4965 set_bit(flag, &bp->sp_rtnl_state);
4966 smp_mb__after_atomic();
4967 DP((BNX2X_MSG_SP | verbose), "Scheduling sp_rtnl task [Flag: %d]\n",
4969 schedule_delayed_work(&bp->sp_rtnl_task, 0);
4971 EXPORT_SYMBOL(bnx2x_schedule_sp_rtnl);