1 /* bnx2x_cmn.c: QLogic Everest network driver.
3 * Copyright (c) 2007-2013 Broadcom Corporation
4 * Copyright (c) 2014 QLogic Corporation
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation.
11 * Maintained by: Ariel Elior <ariel.elior@qlogic.com>
12 * Written by: Eliezer Tamir
13 * Based on code from Michael Chan's bnx2 driver
14 * UDP CSUM errata workaround by Arik Gendelman
15 * Slowpath and fastpath rework by Vladislav Zolotarov
16 * Statistics and Link management by Yitchak Gertner
20 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
22 #include <linux/etherdevice.h>
23 #include <linux/if_vlan.h>
24 #include <linux/interrupt.h>
26 #include <linux/crash_dump.h>
29 #include <net/ip6_checksum.h>
30 #include <net/busy_poll.h>
31 #include <linux/prefetch.h>
32 #include "bnx2x_cmn.h"
33 #include "bnx2x_init.h"
36 static void bnx2x_free_fp_mem_cnic(struct bnx2x *bp);
37 static int bnx2x_alloc_fp_mem_cnic(struct bnx2x *bp);
38 static int bnx2x_alloc_fp_mem(struct bnx2x *bp);
39 static int bnx2x_poll(struct napi_struct *napi, int budget);
41 static void bnx2x_add_all_napi_cnic(struct bnx2x *bp)
45 /* Add NAPI objects */
46 for_each_rx_queue_cnic(bp, i) {
47 netif_napi_add(bp->dev, &bnx2x_fp(bp, i, napi),
48 bnx2x_poll, NAPI_POLL_WEIGHT);
49 napi_hash_add(&bnx2x_fp(bp, i, napi));
53 static void bnx2x_add_all_napi(struct bnx2x *bp)
57 /* Add NAPI objects */
58 for_each_eth_queue(bp, i) {
59 netif_napi_add(bp->dev, &bnx2x_fp(bp, i, napi),
60 bnx2x_poll, NAPI_POLL_WEIGHT);
61 napi_hash_add(&bnx2x_fp(bp, i, napi));
65 static int bnx2x_calc_num_queues(struct bnx2x *bp)
67 int nq = bnx2x_num_queues ? : netif_get_num_default_rss_queues();
69 /* Reduce memory usage in kdump environment by using only one queue */
70 if (is_kdump_kernel())
73 nq = clamp(nq, 1, BNX2X_MAX_QUEUES(bp));
78 * bnx2x_move_fp - move content of the fastpath structure.
81 * @from: source FP index
82 * @to: destination FP index
84 * Makes sure the contents of the bp->fp[to].napi is kept
85 * intact. This is done by first copying the napi struct from
86 * the target to the source, and then mem copying the entire
87 * source onto the target. Update txdata pointers and related
90 static inline void bnx2x_move_fp(struct bnx2x *bp, int from, int to)
92 struct bnx2x_fastpath *from_fp = &bp->fp[from];
93 struct bnx2x_fastpath *to_fp = &bp->fp[to];
94 struct bnx2x_sp_objs *from_sp_objs = &bp->sp_objs[from];
95 struct bnx2x_sp_objs *to_sp_objs = &bp->sp_objs[to];
96 struct bnx2x_fp_stats *from_fp_stats = &bp->fp_stats[from];
97 struct bnx2x_fp_stats *to_fp_stats = &bp->fp_stats[to];
98 int old_max_eth_txqs, new_max_eth_txqs;
99 int old_txdata_index = 0, new_txdata_index = 0;
100 struct bnx2x_agg_info *old_tpa_info = to_fp->tpa_info;
102 /* Copy the NAPI object as it has been already initialized */
103 from_fp->napi = to_fp->napi;
105 /* Move bnx2x_fastpath contents */
106 memcpy(to_fp, from_fp, sizeof(*to_fp));
109 /* Retain the tpa_info of the original `to' version as we don't want
110 * 2 FPs to contain the same tpa_info pointer.
112 to_fp->tpa_info = old_tpa_info;
114 /* move sp_objs contents as well, as their indices match fp ones */
115 memcpy(to_sp_objs, from_sp_objs, sizeof(*to_sp_objs));
117 /* move fp_stats contents as well, as their indices match fp ones */
118 memcpy(to_fp_stats, from_fp_stats, sizeof(*to_fp_stats));
120 /* Update txdata pointers in fp and move txdata content accordingly:
121 * Each fp consumes 'max_cos' txdata structures, so the index should be
122 * decremented by max_cos x delta.
125 old_max_eth_txqs = BNX2X_NUM_ETH_QUEUES(bp) * (bp)->max_cos;
126 new_max_eth_txqs = (BNX2X_NUM_ETH_QUEUES(bp) - from + to) *
128 if (from == FCOE_IDX(bp)) {
129 old_txdata_index = old_max_eth_txqs + FCOE_TXQ_IDX_OFFSET;
130 new_txdata_index = new_max_eth_txqs + FCOE_TXQ_IDX_OFFSET;
133 memcpy(&bp->bnx2x_txq[new_txdata_index],
134 &bp->bnx2x_txq[old_txdata_index],
135 sizeof(struct bnx2x_fp_txdata));
136 to_fp->txdata_ptr[0] = &bp->bnx2x_txq[new_txdata_index];
140 * bnx2x_fill_fw_str - Fill buffer with FW version string.
143 * @buf: character buffer to fill with the fw name
144 * @buf_len: length of the above buffer
147 void bnx2x_fill_fw_str(struct bnx2x *bp, char *buf, size_t buf_len)
150 u8 phy_fw_ver[PHY_FW_VER_LEN];
152 phy_fw_ver[0] = '\0';
153 bnx2x_get_ext_phy_fw_version(&bp->link_params,
154 phy_fw_ver, PHY_FW_VER_LEN);
155 strlcpy(buf, bp->fw_ver, buf_len);
156 snprintf(buf + strlen(bp->fw_ver), 32 - strlen(bp->fw_ver),
158 (bp->common.bc_ver & 0xff0000) >> 16,
159 (bp->common.bc_ver & 0xff00) >> 8,
160 (bp->common.bc_ver & 0xff),
161 ((phy_fw_ver[0] != '\0') ? " phy " : ""), phy_fw_ver);
163 bnx2x_vf_fill_fw_str(bp, buf, buf_len);
168 * bnx2x_shrink_eth_fp - guarantees fastpath structures stay intact
171 * @delta: number of eth queues which were not allocated
173 static void bnx2x_shrink_eth_fp(struct bnx2x *bp, int delta)
175 int i, cos, old_eth_num = BNX2X_NUM_ETH_QUEUES(bp);
177 /* Queue pointer cannot be re-set on an fp-basis, as moving pointer
178 * backward along the array could cause memory to be overridden
180 for (cos = 1; cos < bp->max_cos; cos++) {
181 for (i = 0; i < old_eth_num - delta; i++) {
182 struct bnx2x_fastpath *fp = &bp->fp[i];
183 int new_idx = cos * (old_eth_num - delta) + i;
185 memcpy(&bp->bnx2x_txq[new_idx], fp->txdata_ptr[cos],
186 sizeof(struct bnx2x_fp_txdata));
187 fp->txdata_ptr[cos] = &bp->bnx2x_txq[new_idx];
192 int bnx2x_load_count[2][3] = { {0} }; /* per-path: 0-common, 1-port0, 2-port1 */
194 /* free skb in the packet ring at pos idx
195 * return idx of last bd freed
197 static u16 bnx2x_free_tx_pkt(struct bnx2x *bp, struct bnx2x_fp_txdata *txdata,
198 u16 idx, unsigned int *pkts_compl,
199 unsigned int *bytes_compl)
201 struct sw_tx_bd *tx_buf = &txdata->tx_buf_ring[idx];
202 struct eth_tx_start_bd *tx_start_bd;
203 struct eth_tx_bd *tx_data_bd;
204 struct sk_buff *skb = tx_buf->skb;
205 u16 bd_idx = TX_BD(tx_buf->first_bd), new_cons;
207 u16 split_bd_len = 0;
209 /* prefetch skb end pointer to speedup dev_kfree_skb() */
212 DP(NETIF_MSG_TX_DONE, "fp[%d]: pkt_idx %d buff @(%p)->skb %p\n",
213 txdata->txq_index, idx, tx_buf, skb);
215 tx_start_bd = &txdata->tx_desc_ring[bd_idx].start_bd;
217 nbd = le16_to_cpu(tx_start_bd->nbd) - 1;
218 #ifdef BNX2X_STOP_ON_ERROR
219 if ((nbd - 1) > (MAX_SKB_FRAGS + 2)) {
220 BNX2X_ERR("BAD nbd!\n");
224 new_cons = nbd + tx_buf->first_bd;
226 /* Get the next bd */
227 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
229 /* Skip a parse bd... */
231 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
233 if (tx_buf->flags & BNX2X_HAS_SECOND_PBD) {
234 /* Skip second parse bd... */
236 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
239 /* TSO headers+data bds share a common mapping. See bnx2x_tx_split() */
240 if (tx_buf->flags & BNX2X_TSO_SPLIT_BD) {
241 tx_data_bd = &txdata->tx_desc_ring[bd_idx].reg_bd;
242 split_bd_len = BD_UNMAP_LEN(tx_data_bd);
244 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
248 dma_unmap_single(&bp->pdev->dev, BD_UNMAP_ADDR(tx_start_bd),
249 BD_UNMAP_LEN(tx_start_bd) + split_bd_len,
255 tx_data_bd = &txdata->tx_desc_ring[bd_idx].reg_bd;
256 dma_unmap_page(&bp->pdev->dev, BD_UNMAP_ADDR(tx_data_bd),
257 BD_UNMAP_LEN(tx_data_bd), DMA_TO_DEVICE);
259 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
266 (*bytes_compl) += skb->len;
269 dev_kfree_skb_any(skb);
270 tx_buf->first_bd = 0;
276 int bnx2x_tx_int(struct bnx2x *bp, struct bnx2x_fp_txdata *txdata)
278 struct netdev_queue *txq;
279 u16 hw_cons, sw_cons, bd_cons = txdata->tx_bd_cons;
280 unsigned int pkts_compl = 0, bytes_compl = 0;
282 #ifdef BNX2X_STOP_ON_ERROR
283 if (unlikely(bp->panic))
287 txq = netdev_get_tx_queue(bp->dev, txdata->txq_index);
288 hw_cons = le16_to_cpu(*txdata->tx_cons_sb);
289 sw_cons = txdata->tx_pkt_cons;
291 while (sw_cons != hw_cons) {
294 pkt_cons = TX_BD(sw_cons);
296 DP(NETIF_MSG_TX_DONE,
297 "queue[%d]: hw_cons %u sw_cons %u pkt_cons %u\n",
298 txdata->txq_index, hw_cons, sw_cons, pkt_cons);
300 bd_cons = bnx2x_free_tx_pkt(bp, txdata, pkt_cons,
301 &pkts_compl, &bytes_compl);
306 netdev_tx_completed_queue(txq, pkts_compl, bytes_compl);
308 txdata->tx_pkt_cons = sw_cons;
309 txdata->tx_bd_cons = bd_cons;
311 /* Need to make the tx_bd_cons update visible to start_xmit()
312 * before checking for netif_tx_queue_stopped(). Without the
313 * memory barrier, there is a small possibility that
314 * start_xmit() will miss it and cause the queue to be stopped
316 * On the other hand we need an rmb() here to ensure the proper
317 * ordering of bit testing in the following
318 * netif_tx_queue_stopped(txq) call.
322 if (unlikely(netif_tx_queue_stopped(txq))) {
323 /* Taking tx_lock() is needed to prevent re-enabling the queue
324 * while it's empty. This could have happen if rx_action() gets
325 * suspended in bnx2x_tx_int() after the condition before
326 * netif_tx_wake_queue(), while tx_action (bnx2x_start_xmit()):
328 * stops the queue->sees fresh tx_bd_cons->releases the queue->
329 * sends some packets consuming the whole queue again->
333 __netif_tx_lock(txq, smp_processor_id());
335 if ((netif_tx_queue_stopped(txq)) &&
336 (bp->state == BNX2X_STATE_OPEN) &&
337 (bnx2x_tx_avail(bp, txdata) >= MAX_DESC_PER_TX_PKT))
338 netif_tx_wake_queue(txq);
340 __netif_tx_unlock(txq);
345 static inline void bnx2x_update_last_max_sge(struct bnx2x_fastpath *fp,
348 u16 last_max = fp->last_max_sge;
350 if (SUB_S16(idx, last_max) > 0)
351 fp->last_max_sge = idx;
354 static inline void bnx2x_update_sge_prod(struct bnx2x_fastpath *fp,
356 struct eth_end_agg_rx_cqe *cqe)
358 struct bnx2x *bp = fp->bp;
359 u16 last_max, last_elem, first_elem;
366 /* First mark all used pages */
367 for (i = 0; i < sge_len; i++)
368 BIT_VEC64_CLEAR_BIT(fp->sge_mask,
369 RX_SGE(le16_to_cpu(cqe->sgl_or_raw_data.sgl[i])));
371 DP(NETIF_MSG_RX_STATUS, "fp_cqe->sgl[%d] = %d\n",
372 sge_len - 1, le16_to_cpu(cqe->sgl_or_raw_data.sgl[sge_len - 1]));
374 /* Here we assume that the last SGE index is the biggest */
375 prefetch((void *)(fp->sge_mask));
376 bnx2x_update_last_max_sge(fp,
377 le16_to_cpu(cqe->sgl_or_raw_data.sgl[sge_len - 1]));
379 last_max = RX_SGE(fp->last_max_sge);
380 last_elem = last_max >> BIT_VEC64_ELEM_SHIFT;
381 first_elem = RX_SGE(fp->rx_sge_prod) >> BIT_VEC64_ELEM_SHIFT;
383 /* If ring is not full */
384 if (last_elem + 1 != first_elem)
387 /* Now update the prod */
388 for (i = first_elem; i != last_elem; i = NEXT_SGE_MASK_ELEM(i)) {
389 if (likely(fp->sge_mask[i]))
392 fp->sge_mask[i] = BIT_VEC64_ELEM_ONE_MASK;
393 delta += BIT_VEC64_ELEM_SZ;
397 fp->rx_sge_prod += delta;
398 /* clear page-end entries */
399 bnx2x_clear_sge_mask_next_elems(fp);
402 DP(NETIF_MSG_RX_STATUS,
403 "fp->last_max_sge = %d fp->rx_sge_prod = %d\n",
404 fp->last_max_sge, fp->rx_sge_prod);
407 /* Get Toeplitz hash value in the skb using the value from the
408 * CQE (calculated by HW).
410 static u32 bnx2x_get_rxhash(const struct bnx2x *bp,
411 const struct eth_fast_path_rx_cqe *cqe,
412 enum pkt_hash_types *rxhash_type)
414 /* Get Toeplitz hash from CQE */
415 if ((bp->dev->features & NETIF_F_RXHASH) &&
416 (cqe->status_flags & ETH_FAST_PATH_RX_CQE_RSS_HASH_FLG)) {
417 enum eth_rss_hash_type htype;
419 htype = cqe->status_flags & ETH_FAST_PATH_RX_CQE_RSS_HASH_TYPE;
420 *rxhash_type = ((htype == TCP_IPV4_HASH_TYPE) ||
421 (htype == TCP_IPV6_HASH_TYPE)) ?
422 PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3;
424 return le32_to_cpu(cqe->rss_hash_result);
426 *rxhash_type = PKT_HASH_TYPE_NONE;
430 static void bnx2x_tpa_start(struct bnx2x_fastpath *fp, u16 queue,
432 struct eth_fast_path_rx_cqe *cqe)
434 struct bnx2x *bp = fp->bp;
435 struct sw_rx_bd *cons_rx_buf = &fp->rx_buf_ring[cons];
436 struct sw_rx_bd *prod_rx_buf = &fp->rx_buf_ring[prod];
437 struct eth_rx_bd *prod_bd = &fp->rx_desc_ring[prod];
439 struct bnx2x_agg_info *tpa_info = &fp->tpa_info[queue];
440 struct sw_rx_bd *first_buf = &tpa_info->first_buf;
442 /* print error if current state != stop */
443 if (tpa_info->tpa_state != BNX2X_TPA_STOP)
444 BNX2X_ERR("start of bin not in stop [%d]\n", queue);
446 /* Try to map an empty data buffer from the aggregation info */
447 mapping = dma_map_single(&bp->pdev->dev,
448 first_buf->data + NET_SKB_PAD,
449 fp->rx_buf_size, DMA_FROM_DEVICE);
451 * ...if it fails - move the skb from the consumer to the producer
452 * and set the current aggregation state as ERROR to drop it
453 * when TPA_STOP arrives.
456 if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
457 /* Move the BD from the consumer to the producer */
458 bnx2x_reuse_rx_data(fp, cons, prod);
459 tpa_info->tpa_state = BNX2X_TPA_ERROR;
463 /* move empty data from pool to prod */
464 prod_rx_buf->data = first_buf->data;
465 dma_unmap_addr_set(prod_rx_buf, mapping, mapping);
466 /* point prod_bd to new data */
467 prod_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
468 prod_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
470 /* move partial skb from cons to pool (don't unmap yet) */
471 *first_buf = *cons_rx_buf;
473 /* mark bin state as START */
474 tpa_info->parsing_flags =
475 le16_to_cpu(cqe->pars_flags.flags);
476 tpa_info->vlan_tag = le16_to_cpu(cqe->vlan_tag);
477 tpa_info->tpa_state = BNX2X_TPA_START;
478 tpa_info->len_on_bd = le16_to_cpu(cqe->len_on_bd);
479 tpa_info->placement_offset = cqe->placement_offset;
480 tpa_info->rxhash = bnx2x_get_rxhash(bp, cqe, &tpa_info->rxhash_type);
481 if (fp->mode == TPA_MODE_GRO) {
482 u16 gro_size = le16_to_cpu(cqe->pkt_len_or_gro_seg_len);
483 tpa_info->full_page = SGE_PAGES / gro_size * gro_size;
484 tpa_info->gro_size = gro_size;
487 #ifdef BNX2X_STOP_ON_ERROR
488 fp->tpa_queue_used |= (1 << queue);
489 DP(NETIF_MSG_RX_STATUS, "fp->tpa_queue_used = 0x%llx\n",
494 /* Timestamp option length allowed for TPA aggregation:
496 * nop nop kind length echo val
498 #define TPA_TSTAMP_OPT_LEN 12
500 * bnx2x_set_gro_params - compute GRO values
503 * @parsing_flags: parsing flags from the START CQE
504 * @len_on_bd: total length of the first packet for the
506 * @pkt_len: length of all segments
508 * Approximate value of the MSS for this aggregation calculated using
509 * the first packet of it.
510 * Compute number of aggregated segments, and gso_type.
512 static void bnx2x_set_gro_params(struct sk_buff *skb, u16 parsing_flags,
513 u16 len_on_bd, unsigned int pkt_len,
514 u16 num_of_coalesced_segs)
516 /* TPA aggregation won't have either IP options or TCP options
517 * other than timestamp or IPv6 extension headers.
519 u16 hdrs_len = ETH_HLEN + sizeof(struct tcphdr);
521 if (GET_FLAG(parsing_flags, PARSING_FLAGS_OVER_ETHERNET_PROTOCOL) ==
522 PRS_FLAG_OVERETH_IPV6) {
523 hdrs_len += sizeof(struct ipv6hdr);
524 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
526 hdrs_len += sizeof(struct iphdr);
527 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
530 /* Check if there was a TCP timestamp, if there is it's will
531 * always be 12 bytes length: nop nop kind length echo val.
533 * Otherwise FW would close the aggregation.
535 if (parsing_flags & PARSING_FLAGS_TIME_STAMP_EXIST_FLAG)
536 hdrs_len += TPA_TSTAMP_OPT_LEN;
538 skb_shinfo(skb)->gso_size = len_on_bd - hdrs_len;
540 /* tcp_gro_complete() will copy NAPI_GRO_CB(skb)->count
541 * to skb_shinfo(skb)->gso_segs
543 NAPI_GRO_CB(skb)->count = num_of_coalesced_segs;
546 static int bnx2x_alloc_rx_sge(struct bnx2x *bp, struct bnx2x_fastpath *fp,
547 u16 index, gfp_t gfp_mask)
549 struct sw_rx_page *sw_buf = &fp->rx_page_ring[index];
550 struct eth_rx_sge *sge = &fp->rx_sge_ring[index];
551 struct bnx2x_alloc_pool *pool = &fp->page_pool;
554 if (!pool->page || (PAGE_SIZE - pool->offset) < SGE_PAGE_SIZE) {
556 /* put page reference used by the memory pool, since we
557 * won't be using this page as the mempool anymore.
560 put_page(pool->page);
562 pool->page = alloc_pages(gfp_mask, PAGES_PER_SGE_SHIFT);
563 if (unlikely(!pool->page)) {
564 BNX2X_ERR("Can't alloc sge\n");
571 mapping = dma_map_page(&bp->pdev->dev, pool->page,
572 pool->offset, SGE_PAGE_SIZE, DMA_FROM_DEVICE);
573 if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
574 BNX2X_ERR("Can't map sge\n");
578 get_page(pool->page);
579 sw_buf->page = pool->page;
580 sw_buf->offset = pool->offset;
582 dma_unmap_addr_set(sw_buf, mapping, mapping);
584 sge->addr_hi = cpu_to_le32(U64_HI(mapping));
585 sge->addr_lo = cpu_to_le32(U64_LO(mapping));
587 pool->offset += SGE_PAGE_SIZE;
592 static int bnx2x_fill_frag_skb(struct bnx2x *bp, struct bnx2x_fastpath *fp,
593 struct bnx2x_agg_info *tpa_info,
596 struct eth_end_agg_rx_cqe *cqe,
599 struct sw_rx_page *rx_pg, old_rx_pg;
600 u32 i, frag_len, frag_size;
601 int err, j, frag_id = 0;
602 u16 len_on_bd = tpa_info->len_on_bd;
603 u16 full_page = 0, gro_size = 0;
605 frag_size = le16_to_cpu(cqe->pkt_len) - len_on_bd;
607 if (fp->mode == TPA_MODE_GRO) {
608 gro_size = tpa_info->gro_size;
609 full_page = tpa_info->full_page;
612 /* This is needed in order to enable forwarding support */
614 bnx2x_set_gro_params(skb, tpa_info->parsing_flags, len_on_bd,
615 le16_to_cpu(cqe->pkt_len),
616 le16_to_cpu(cqe->num_of_coalesced_segs));
618 #ifdef BNX2X_STOP_ON_ERROR
619 if (pages > min_t(u32, 8, MAX_SKB_FRAGS) * SGE_PAGES) {
620 BNX2X_ERR("SGL length is too long: %d. CQE index is %d\n",
622 BNX2X_ERR("cqe->pkt_len = %d\n", cqe->pkt_len);
628 /* Run through the SGL and compose the fragmented skb */
629 for (i = 0, j = 0; i < pages; i += PAGES_PER_SGE, j++) {
630 u16 sge_idx = RX_SGE(le16_to_cpu(cqe->sgl_or_raw_data.sgl[j]));
632 /* FW gives the indices of the SGE as if the ring is an array
633 (meaning that "next" element will consume 2 indices) */
634 if (fp->mode == TPA_MODE_GRO)
635 frag_len = min_t(u32, frag_size, (u32)full_page);
637 frag_len = min_t(u32, frag_size, (u32)SGE_PAGES);
639 rx_pg = &fp->rx_page_ring[sge_idx];
642 /* If we fail to allocate a substitute page, we simply stop
643 where we are and drop the whole packet */
644 err = bnx2x_alloc_rx_sge(bp, fp, sge_idx, GFP_ATOMIC);
646 bnx2x_fp_qstats(bp, fp)->rx_skb_alloc_failed++;
650 dma_unmap_page(&bp->pdev->dev,
651 dma_unmap_addr(&old_rx_pg, mapping),
652 SGE_PAGE_SIZE, DMA_FROM_DEVICE);
653 /* Add one frag and update the appropriate fields in the skb */
654 if (fp->mode == TPA_MODE_LRO)
655 skb_fill_page_desc(skb, j, old_rx_pg.page,
656 old_rx_pg.offset, frag_len);
660 for (rem = frag_len; rem > 0; rem -= gro_size) {
661 int len = rem > gro_size ? gro_size : rem;
662 skb_fill_page_desc(skb, frag_id++,
664 old_rx_pg.offset + offset,
667 get_page(old_rx_pg.page);
672 skb->data_len += frag_len;
673 skb->truesize += SGE_PAGES;
674 skb->len += frag_len;
676 frag_size -= frag_len;
682 static void bnx2x_frag_free(const struct bnx2x_fastpath *fp, void *data)
684 if (fp->rx_frag_size)
690 static void *bnx2x_frag_alloc(const struct bnx2x_fastpath *fp, gfp_t gfp_mask)
692 if (fp->rx_frag_size) {
693 /* GFP_KERNEL allocations are used only during initialization */
694 if (unlikely(gfp_mask & __GFP_WAIT))
695 return (void *)__get_free_page(gfp_mask);
697 return netdev_alloc_frag(fp->rx_frag_size);
700 return kmalloc(fp->rx_buf_size + NET_SKB_PAD, gfp_mask);
704 static void bnx2x_gro_ip_csum(struct bnx2x *bp, struct sk_buff *skb)
706 const struct iphdr *iph = ip_hdr(skb);
709 skb_set_transport_header(skb, sizeof(struct iphdr));
712 th->check = ~tcp_v4_check(skb->len - skb_transport_offset(skb),
713 iph->saddr, iph->daddr, 0);
716 static void bnx2x_gro_ipv6_csum(struct bnx2x *bp, struct sk_buff *skb)
718 struct ipv6hdr *iph = ipv6_hdr(skb);
721 skb_set_transport_header(skb, sizeof(struct ipv6hdr));
724 th->check = ~tcp_v6_check(skb->len - skb_transport_offset(skb),
725 &iph->saddr, &iph->daddr, 0);
728 static void bnx2x_gro_csum(struct bnx2x *bp, struct sk_buff *skb,
729 void (*gro_func)(struct bnx2x*, struct sk_buff*))
731 skb_set_network_header(skb, 0);
733 tcp_gro_complete(skb);
737 static void bnx2x_gro_receive(struct bnx2x *bp, struct bnx2x_fastpath *fp,
741 if (skb_shinfo(skb)->gso_size) {
742 switch (be16_to_cpu(skb->protocol)) {
744 bnx2x_gro_csum(bp, skb, bnx2x_gro_ip_csum);
747 bnx2x_gro_csum(bp, skb, bnx2x_gro_ipv6_csum);
750 BNX2X_ERR("Error: FW GRO supports only IPv4/IPv6, not 0x%04x\n",
751 be16_to_cpu(skb->protocol));
755 skb_record_rx_queue(skb, fp->rx_queue);
756 napi_gro_receive(&fp->napi, skb);
759 static void bnx2x_tpa_stop(struct bnx2x *bp, struct bnx2x_fastpath *fp,
760 struct bnx2x_agg_info *tpa_info,
762 struct eth_end_agg_rx_cqe *cqe,
765 struct sw_rx_bd *rx_buf = &tpa_info->first_buf;
766 u8 pad = tpa_info->placement_offset;
767 u16 len = tpa_info->len_on_bd;
768 struct sk_buff *skb = NULL;
769 u8 *new_data, *data = rx_buf->data;
770 u8 old_tpa_state = tpa_info->tpa_state;
772 tpa_info->tpa_state = BNX2X_TPA_STOP;
774 /* If we there was an error during the handling of the TPA_START -
775 * drop this aggregation.
777 if (old_tpa_state == BNX2X_TPA_ERROR)
780 /* Try to allocate the new data */
781 new_data = bnx2x_frag_alloc(fp, GFP_ATOMIC);
782 /* Unmap skb in the pool anyway, as we are going to change
783 pool entry status to BNX2X_TPA_STOP even if new skb allocation
785 dma_unmap_single(&bp->pdev->dev, dma_unmap_addr(rx_buf, mapping),
786 fp->rx_buf_size, DMA_FROM_DEVICE);
787 if (likely(new_data))
788 skb = build_skb(data, fp->rx_frag_size);
791 #ifdef BNX2X_STOP_ON_ERROR
792 if (pad + len > fp->rx_buf_size) {
793 BNX2X_ERR("skb_put is about to fail... pad %d len %d rx_buf_size %d\n",
794 pad, len, fp->rx_buf_size);
800 skb_reserve(skb, pad + NET_SKB_PAD);
802 skb_set_hash(skb, tpa_info->rxhash, tpa_info->rxhash_type);
804 skb->protocol = eth_type_trans(skb, bp->dev);
805 skb->ip_summed = CHECKSUM_UNNECESSARY;
807 if (!bnx2x_fill_frag_skb(bp, fp, tpa_info, pages,
808 skb, cqe, cqe_idx)) {
809 if (tpa_info->parsing_flags & PARSING_FLAGS_VLAN)
810 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), tpa_info->vlan_tag);
811 bnx2x_gro_receive(bp, fp, skb);
813 DP(NETIF_MSG_RX_STATUS,
814 "Failed to allocate new pages - dropping packet!\n");
815 dev_kfree_skb_any(skb);
818 /* put new data in bin */
819 rx_buf->data = new_data;
824 bnx2x_frag_free(fp, new_data);
826 /* drop the packet and keep the buffer in the bin */
827 DP(NETIF_MSG_RX_STATUS,
828 "Failed to allocate or map a new skb - dropping packet!\n");
829 bnx2x_fp_stats(bp, fp)->eth_q_stats.rx_skb_alloc_failed++;
832 static int bnx2x_alloc_rx_data(struct bnx2x *bp, struct bnx2x_fastpath *fp,
833 u16 index, gfp_t gfp_mask)
836 struct sw_rx_bd *rx_buf = &fp->rx_buf_ring[index];
837 struct eth_rx_bd *rx_bd = &fp->rx_desc_ring[index];
840 data = bnx2x_frag_alloc(fp, gfp_mask);
841 if (unlikely(data == NULL))
844 mapping = dma_map_single(&bp->pdev->dev, data + NET_SKB_PAD,
847 if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
848 bnx2x_frag_free(fp, data);
849 BNX2X_ERR("Can't map rx data\n");
854 dma_unmap_addr_set(rx_buf, mapping, mapping);
856 rx_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
857 rx_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
863 void bnx2x_csum_validate(struct sk_buff *skb, union eth_rx_cqe *cqe,
864 struct bnx2x_fastpath *fp,
865 struct bnx2x_eth_q_stats *qstats)
867 /* Do nothing if no L4 csum validation was done.
868 * We do not check whether IP csum was validated. For IPv4 we assume
869 * that if the card got as far as validating the L4 csum, it also
870 * validated the IP csum. IPv6 has no IP csum.
872 if (cqe->fast_path_cqe.status_flags &
873 ETH_FAST_PATH_RX_CQE_L4_XSUM_NO_VALIDATION_FLG)
876 /* If L4 validation was done, check if an error was found. */
878 if (cqe->fast_path_cqe.type_error_flags &
879 (ETH_FAST_PATH_RX_CQE_IP_BAD_XSUM_FLG |
880 ETH_FAST_PATH_RX_CQE_L4_BAD_XSUM_FLG))
881 qstats->hw_csum_err++;
883 skb->ip_summed = CHECKSUM_UNNECESSARY;
886 static int bnx2x_rx_int(struct bnx2x_fastpath *fp, int budget)
888 struct bnx2x *bp = fp->bp;
889 u16 bd_cons, bd_prod, bd_prod_fw, comp_ring_cons;
890 u16 sw_comp_cons, sw_comp_prod;
892 union eth_rx_cqe *cqe;
893 struct eth_fast_path_rx_cqe *cqe_fp;
895 #ifdef BNX2X_STOP_ON_ERROR
896 if (unlikely(bp->panic))
902 bd_cons = fp->rx_bd_cons;
903 bd_prod = fp->rx_bd_prod;
904 bd_prod_fw = bd_prod;
905 sw_comp_cons = fp->rx_comp_cons;
906 sw_comp_prod = fp->rx_comp_prod;
908 comp_ring_cons = RCQ_BD(sw_comp_cons);
909 cqe = &fp->rx_comp_ring[comp_ring_cons];
910 cqe_fp = &cqe->fast_path_cqe;
912 DP(NETIF_MSG_RX_STATUS,
913 "queue[%d]: sw_comp_cons %u\n", fp->index, sw_comp_cons);
915 while (BNX2X_IS_CQE_COMPLETED(cqe_fp)) {
916 struct sw_rx_bd *rx_buf = NULL;
919 enum eth_rx_cqe_type cqe_fp_type;
923 enum pkt_hash_types rxhash_type;
925 #ifdef BNX2X_STOP_ON_ERROR
926 if (unlikely(bp->panic))
930 bd_prod = RX_BD(bd_prod);
931 bd_cons = RX_BD(bd_cons);
933 /* A rmb() is required to ensure that the CQE is not read
934 * before it is written by the adapter DMA. PCI ordering
935 * rules will make sure the other fields are written before
936 * the marker at the end of struct eth_fast_path_rx_cqe
937 * but without rmb() a weakly ordered processor can process
938 * stale data. Without the barrier TPA state-machine might
939 * enter inconsistent state and kernel stack might be
940 * provided with incorrect packet description - these lead
941 * to various kernel crashed.
945 cqe_fp_flags = cqe_fp->type_error_flags;
946 cqe_fp_type = cqe_fp_flags & ETH_FAST_PATH_RX_CQE_TYPE;
948 DP(NETIF_MSG_RX_STATUS,
949 "CQE type %x err %x status %x queue %x vlan %x len %u\n",
950 CQE_TYPE(cqe_fp_flags),
951 cqe_fp_flags, cqe_fp->status_flags,
952 le32_to_cpu(cqe_fp->rss_hash_result),
953 le16_to_cpu(cqe_fp->vlan_tag),
954 le16_to_cpu(cqe_fp->pkt_len_or_gro_seg_len));
956 /* is this a slowpath msg? */
957 if (unlikely(CQE_TYPE_SLOW(cqe_fp_type))) {
958 bnx2x_sp_event(fp, cqe);
962 rx_buf = &fp->rx_buf_ring[bd_cons];
965 if (!CQE_TYPE_FAST(cqe_fp_type)) {
966 struct bnx2x_agg_info *tpa_info;
967 u16 frag_size, pages;
968 #ifdef BNX2X_STOP_ON_ERROR
970 if (fp->mode == TPA_MODE_DISABLED &&
971 (CQE_TYPE_START(cqe_fp_type) ||
972 CQE_TYPE_STOP(cqe_fp_type)))
973 BNX2X_ERR("START/STOP packet while TPA disabled, type %x\n",
974 CQE_TYPE(cqe_fp_type));
977 if (CQE_TYPE_START(cqe_fp_type)) {
978 u16 queue = cqe_fp->queue_index;
979 DP(NETIF_MSG_RX_STATUS,
980 "calling tpa_start on queue %d\n",
983 bnx2x_tpa_start(fp, queue,
989 queue = cqe->end_agg_cqe.queue_index;
990 tpa_info = &fp->tpa_info[queue];
991 DP(NETIF_MSG_RX_STATUS,
992 "calling tpa_stop on queue %d\n",
995 frag_size = le16_to_cpu(cqe->end_agg_cqe.pkt_len) -
998 if (fp->mode == TPA_MODE_GRO)
999 pages = (frag_size + tpa_info->full_page - 1) /
1000 tpa_info->full_page;
1002 pages = SGE_PAGE_ALIGN(frag_size) >>
1005 bnx2x_tpa_stop(bp, fp, tpa_info, pages,
1006 &cqe->end_agg_cqe, comp_ring_cons);
1007 #ifdef BNX2X_STOP_ON_ERROR
1012 bnx2x_update_sge_prod(fp, pages, &cqe->end_agg_cqe);
1016 len = le16_to_cpu(cqe_fp->pkt_len_or_gro_seg_len);
1017 pad = cqe_fp->placement_offset;
1018 dma_sync_single_for_cpu(&bp->pdev->dev,
1019 dma_unmap_addr(rx_buf, mapping),
1020 pad + RX_COPY_THRESH,
1023 prefetch(data + pad); /* speedup eth_type_trans() */
1024 /* is this an error packet? */
1025 if (unlikely(cqe_fp_flags & ETH_RX_ERROR_FALGS)) {
1026 DP(NETIF_MSG_RX_ERR | NETIF_MSG_RX_STATUS,
1027 "ERROR flags %x rx packet %u\n",
1028 cqe_fp_flags, sw_comp_cons);
1029 bnx2x_fp_qstats(bp, fp)->rx_err_discard_pkt++;
1033 /* Since we don't have a jumbo ring
1034 * copy small packets if mtu > 1500
1036 if ((bp->dev->mtu > ETH_MAX_PACKET_SIZE) &&
1037 (len <= RX_COPY_THRESH)) {
1038 skb = napi_alloc_skb(&fp->napi, len);
1040 DP(NETIF_MSG_RX_ERR | NETIF_MSG_RX_STATUS,
1041 "ERROR packet dropped because of alloc failure\n");
1042 bnx2x_fp_qstats(bp, fp)->rx_skb_alloc_failed++;
1045 memcpy(skb->data, data + pad, len);
1046 bnx2x_reuse_rx_data(fp, bd_cons, bd_prod);
1048 if (likely(bnx2x_alloc_rx_data(bp, fp, bd_prod,
1049 GFP_ATOMIC) == 0)) {
1050 dma_unmap_single(&bp->pdev->dev,
1051 dma_unmap_addr(rx_buf, mapping),
1054 skb = build_skb(data, fp->rx_frag_size);
1055 if (unlikely(!skb)) {
1056 bnx2x_frag_free(fp, data);
1057 bnx2x_fp_qstats(bp, fp)->
1058 rx_skb_alloc_failed++;
1061 skb_reserve(skb, pad);
1063 DP(NETIF_MSG_RX_ERR | NETIF_MSG_RX_STATUS,
1064 "ERROR packet dropped because of alloc failure\n");
1065 bnx2x_fp_qstats(bp, fp)->rx_skb_alloc_failed++;
1067 bnx2x_reuse_rx_data(fp, bd_cons, bd_prod);
1073 skb->protocol = eth_type_trans(skb, bp->dev);
1075 /* Set Toeplitz hash for a none-LRO skb */
1076 rxhash = bnx2x_get_rxhash(bp, cqe_fp, &rxhash_type);
1077 skb_set_hash(skb, rxhash, rxhash_type);
1079 skb_checksum_none_assert(skb);
1081 if (bp->dev->features & NETIF_F_RXCSUM)
1082 bnx2x_csum_validate(skb, cqe, fp,
1083 bnx2x_fp_qstats(bp, fp));
1085 skb_record_rx_queue(skb, fp->rx_queue);
1087 /* Check if this packet was timestamped */
1088 if (unlikely(cqe->fast_path_cqe.type_error_flags &
1089 (1 << ETH_FAST_PATH_RX_CQE_PTP_PKT_SHIFT)))
1090 bnx2x_set_rx_ts(bp, skb);
1092 if (le16_to_cpu(cqe_fp->pars_flags.flags) &
1094 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
1095 le16_to_cpu(cqe_fp->vlan_tag));
1097 skb_mark_napi_id(skb, &fp->napi);
1099 if (bnx2x_fp_ll_polling(fp))
1100 netif_receive_skb(skb);
1102 napi_gro_receive(&fp->napi, skb);
1104 rx_buf->data = NULL;
1106 bd_cons = NEXT_RX_IDX(bd_cons);
1107 bd_prod = NEXT_RX_IDX(bd_prod);
1108 bd_prod_fw = NEXT_RX_IDX(bd_prod_fw);
1111 sw_comp_prod = NEXT_RCQ_IDX(sw_comp_prod);
1112 sw_comp_cons = NEXT_RCQ_IDX(sw_comp_cons);
1114 /* mark CQE as free */
1115 BNX2X_SEED_CQE(cqe_fp);
1117 if (rx_pkt == budget)
1120 comp_ring_cons = RCQ_BD(sw_comp_cons);
1121 cqe = &fp->rx_comp_ring[comp_ring_cons];
1122 cqe_fp = &cqe->fast_path_cqe;
1125 fp->rx_bd_cons = bd_cons;
1126 fp->rx_bd_prod = bd_prod_fw;
1127 fp->rx_comp_cons = sw_comp_cons;
1128 fp->rx_comp_prod = sw_comp_prod;
1130 /* Update producers */
1131 bnx2x_update_rx_prod(bp, fp, bd_prod_fw, sw_comp_prod,
1134 fp->rx_pkt += rx_pkt;
1140 static irqreturn_t bnx2x_msix_fp_int(int irq, void *fp_cookie)
1142 struct bnx2x_fastpath *fp = fp_cookie;
1143 struct bnx2x *bp = fp->bp;
1147 "got an MSI-X interrupt on IDX:SB [fp %d fw_sd %d igusb %d]\n",
1148 fp->index, fp->fw_sb_id, fp->igu_sb_id);
1150 bnx2x_ack_sb(bp, fp->igu_sb_id, USTORM_ID, 0, IGU_INT_DISABLE, 0);
1152 #ifdef BNX2X_STOP_ON_ERROR
1153 if (unlikely(bp->panic))
1157 /* Handle Rx and Tx according to MSI-X vector */
1158 for_each_cos_in_tx_queue(fp, cos)
1159 prefetch(fp->txdata_ptr[cos]->tx_cons_sb);
1161 prefetch(&fp->sb_running_index[SM_RX_ID]);
1162 napi_schedule_irqoff(&bnx2x_fp(bp, fp->index, napi));
1167 /* HW Lock for shared dual port PHYs */
1168 void bnx2x_acquire_phy_lock(struct bnx2x *bp)
1170 mutex_lock(&bp->port.phy_mutex);
1172 bnx2x_acquire_hw_lock(bp, HW_LOCK_RESOURCE_MDIO);
1175 void bnx2x_release_phy_lock(struct bnx2x *bp)
1177 bnx2x_release_hw_lock(bp, HW_LOCK_RESOURCE_MDIO);
1179 mutex_unlock(&bp->port.phy_mutex);
1182 /* calculates MF speed according to current linespeed and MF configuration */
1183 u16 bnx2x_get_mf_speed(struct bnx2x *bp)
1185 u16 line_speed = bp->link_vars.line_speed;
1187 u16 maxCfg = bnx2x_extract_max_cfg(bp,
1188 bp->mf_config[BP_VN(bp)]);
1190 /* Calculate the current MAX line speed limit for the MF
1194 line_speed = (line_speed * maxCfg) / 100;
1195 else { /* SD mode */
1196 u16 vn_max_rate = maxCfg * 100;
1198 if (vn_max_rate < line_speed)
1199 line_speed = vn_max_rate;
1207 * bnx2x_fill_report_data - fill link report data to report
1209 * @bp: driver handle
1210 * @data: link state to update
1212 * It uses a none-atomic bit operations because is called under the mutex.
1214 static void bnx2x_fill_report_data(struct bnx2x *bp,
1215 struct bnx2x_link_report_data *data)
1217 memset(data, 0, sizeof(*data));
1220 /* Fill the report data: effective line speed */
1221 data->line_speed = bnx2x_get_mf_speed(bp);
1224 if (!bp->link_vars.link_up || (bp->flags & MF_FUNC_DIS))
1225 __set_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1226 &data->link_report_flags);
1228 if (!BNX2X_NUM_ETH_QUEUES(bp))
1229 __set_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1230 &data->link_report_flags);
1233 if (bp->link_vars.duplex == DUPLEX_FULL)
1234 __set_bit(BNX2X_LINK_REPORT_FD,
1235 &data->link_report_flags);
1237 /* Rx Flow Control is ON */
1238 if (bp->link_vars.flow_ctrl & BNX2X_FLOW_CTRL_RX)
1239 __set_bit(BNX2X_LINK_REPORT_RX_FC_ON,
1240 &data->link_report_flags);
1242 /* Tx Flow Control is ON */
1243 if (bp->link_vars.flow_ctrl & BNX2X_FLOW_CTRL_TX)
1244 __set_bit(BNX2X_LINK_REPORT_TX_FC_ON,
1245 &data->link_report_flags);
1247 *data = bp->vf_link_vars;
1252 * bnx2x_link_report - report link status to OS.
1254 * @bp: driver handle
1256 * Calls the __bnx2x_link_report() under the same locking scheme
1257 * as a link/PHY state managing code to ensure a consistent link
1261 void bnx2x_link_report(struct bnx2x *bp)
1263 bnx2x_acquire_phy_lock(bp);
1264 __bnx2x_link_report(bp);
1265 bnx2x_release_phy_lock(bp);
1269 * __bnx2x_link_report - report link status to OS.
1271 * @bp: driver handle
1273 * None atomic implementation.
1274 * Should be called under the phy_lock.
1276 void __bnx2x_link_report(struct bnx2x *bp)
1278 struct bnx2x_link_report_data cur_data;
1281 if (IS_PF(bp) && !CHIP_IS_E1(bp))
1282 bnx2x_read_mf_cfg(bp);
1284 /* Read the current link report info */
1285 bnx2x_fill_report_data(bp, &cur_data);
1287 /* Don't report link down or exactly the same link status twice */
1288 if (!memcmp(&cur_data, &bp->last_reported_link, sizeof(cur_data)) ||
1289 (test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1290 &bp->last_reported_link.link_report_flags) &&
1291 test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1292 &cur_data.link_report_flags)))
1297 /* We are going to report a new link parameters now -
1298 * remember the current data for the next time.
1300 memcpy(&bp->last_reported_link, &cur_data, sizeof(cur_data));
1302 /* propagate status to VFs */
1304 bnx2x_iov_link_update(bp);
1306 if (test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1307 &cur_data.link_report_flags)) {
1308 netif_carrier_off(bp->dev);
1309 netdev_err(bp->dev, "NIC Link is Down\n");
1315 netif_carrier_on(bp->dev);
1317 if (test_and_clear_bit(BNX2X_LINK_REPORT_FD,
1318 &cur_data.link_report_flags))
1323 /* Handle the FC at the end so that only these flags would be
1324 * possibly set. This way we may easily check if there is no FC
1327 if (cur_data.link_report_flags) {
1328 if (test_bit(BNX2X_LINK_REPORT_RX_FC_ON,
1329 &cur_data.link_report_flags)) {
1330 if (test_bit(BNX2X_LINK_REPORT_TX_FC_ON,
1331 &cur_data.link_report_flags))
1332 flow = "ON - receive & transmit";
1334 flow = "ON - receive";
1336 flow = "ON - transmit";
1341 netdev_info(bp->dev, "NIC Link is Up, %d Mbps %s duplex, Flow control: %s\n",
1342 cur_data.line_speed, duplex, flow);
1346 static void bnx2x_set_next_page_sgl(struct bnx2x_fastpath *fp)
1350 for (i = 1; i <= NUM_RX_SGE_PAGES; i++) {
1351 struct eth_rx_sge *sge;
1353 sge = &fp->rx_sge_ring[RX_SGE_CNT * i - 2];
1355 cpu_to_le32(U64_HI(fp->rx_sge_mapping +
1356 BCM_PAGE_SIZE*(i % NUM_RX_SGE_PAGES)));
1359 cpu_to_le32(U64_LO(fp->rx_sge_mapping +
1360 BCM_PAGE_SIZE*(i % NUM_RX_SGE_PAGES)));
1364 static void bnx2x_free_tpa_pool(struct bnx2x *bp,
1365 struct bnx2x_fastpath *fp, int last)
1369 for (i = 0; i < last; i++) {
1370 struct bnx2x_agg_info *tpa_info = &fp->tpa_info[i];
1371 struct sw_rx_bd *first_buf = &tpa_info->first_buf;
1372 u8 *data = first_buf->data;
1375 DP(NETIF_MSG_IFDOWN, "tpa bin %d empty on free\n", i);
1378 if (tpa_info->tpa_state == BNX2X_TPA_START)
1379 dma_unmap_single(&bp->pdev->dev,
1380 dma_unmap_addr(first_buf, mapping),
1381 fp->rx_buf_size, DMA_FROM_DEVICE);
1382 bnx2x_frag_free(fp, data);
1383 first_buf->data = NULL;
1387 void bnx2x_init_rx_rings_cnic(struct bnx2x *bp)
1391 for_each_rx_queue_cnic(bp, j) {
1392 struct bnx2x_fastpath *fp = &bp->fp[j];
1396 /* Activate BD ring */
1398 * this will generate an interrupt (to the TSTORM)
1399 * must only be done after chip is initialized
1401 bnx2x_update_rx_prod(bp, fp, fp->rx_bd_prod, fp->rx_comp_prod,
1406 void bnx2x_init_rx_rings(struct bnx2x *bp)
1408 int func = BP_FUNC(bp);
1412 /* Allocate TPA resources */
1413 for_each_eth_queue(bp, j) {
1414 struct bnx2x_fastpath *fp = &bp->fp[j];
1417 "mtu %d rx_buf_size %d\n", bp->dev->mtu, fp->rx_buf_size);
1419 if (fp->mode != TPA_MODE_DISABLED) {
1420 /* Fill the per-aggregation pool */
1421 for (i = 0; i < MAX_AGG_QS(bp); i++) {
1422 struct bnx2x_agg_info *tpa_info =
1424 struct sw_rx_bd *first_buf =
1425 &tpa_info->first_buf;
1428 bnx2x_frag_alloc(fp, GFP_KERNEL);
1429 if (!first_buf->data) {
1430 BNX2X_ERR("Failed to allocate TPA skb pool for queue[%d] - disabling TPA on this queue!\n",
1432 bnx2x_free_tpa_pool(bp, fp, i);
1433 fp->mode = TPA_MODE_DISABLED;
1436 dma_unmap_addr_set(first_buf, mapping, 0);
1437 tpa_info->tpa_state = BNX2X_TPA_STOP;
1440 /* "next page" elements initialization */
1441 bnx2x_set_next_page_sgl(fp);
1443 /* set SGEs bit mask */
1444 bnx2x_init_sge_ring_bit_mask(fp);
1446 /* Allocate SGEs and initialize the ring elements */
1447 for (i = 0, ring_prod = 0;
1448 i < MAX_RX_SGE_CNT*NUM_RX_SGE_PAGES; i++) {
1450 if (bnx2x_alloc_rx_sge(bp, fp, ring_prod,
1452 BNX2X_ERR("was only able to allocate %d rx sges\n",
1454 BNX2X_ERR("disabling TPA for queue[%d]\n",
1456 /* Cleanup already allocated elements */
1457 bnx2x_free_rx_sge_range(bp, fp,
1459 bnx2x_free_tpa_pool(bp, fp,
1461 fp->mode = TPA_MODE_DISABLED;
1465 ring_prod = NEXT_SGE_IDX(ring_prod);
1468 fp->rx_sge_prod = ring_prod;
1472 for_each_eth_queue(bp, j) {
1473 struct bnx2x_fastpath *fp = &bp->fp[j];
1477 /* Activate BD ring */
1479 * this will generate an interrupt (to the TSTORM)
1480 * must only be done after chip is initialized
1482 bnx2x_update_rx_prod(bp, fp, fp->rx_bd_prod, fp->rx_comp_prod,
1488 if (CHIP_IS_E1(bp)) {
1489 REG_WR(bp, BAR_USTRORM_INTMEM +
1490 USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func),
1491 U64_LO(fp->rx_comp_mapping));
1492 REG_WR(bp, BAR_USTRORM_INTMEM +
1493 USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func) + 4,
1494 U64_HI(fp->rx_comp_mapping));
1499 static void bnx2x_free_tx_skbs_queue(struct bnx2x_fastpath *fp)
1502 struct bnx2x *bp = fp->bp;
1504 for_each_cos_in_tx_queue(fp, cos) {
1505 struct bnx2x_fp_txdata *txdata = fp->txdata_ptr[cos];
1506 unsigned pkts_compl = 0, bytes_compl = 0;
1508 u16 sw_prod = txdata->tx_pkt_prod;
1509 u16 sw_cons = txdata->tx_pkt_cons;
1511 while (sw_cons != sw_prod) {
1512 bnx2x_free_tx_pkt(bp, txdata, TX_BD(sw_cons),
1513 &pkts_compl, &bytes_compl);
1517 netdev_tx_reset_queue(
1518 netdev_get_tx_queue(bp->dev,
1519 txdata->txq_index));
1523 static void bnx2x_free_tx_skbs_cnic(struct bnx2x *bp)
1527 for_each_tx_queue_cnic(bp, i) {
1528 bnx2x_free_tx_skbs_queue(&bp->fp[i]);
1532 static void bnx2x_free_tx_skbs(struct bnx2x *bp)
1536 for_each_eth_queue(bp, i) {
1537 bnx2x_free_tx_skbs_queue(&bp->fp[i]);
1541 static void bnx2x_free_rx_bds(struct bnx2x_fastpath *fp)
1543 struct bnx2x *bp = fp->bp;
1546 /* ring wasn't allocated */
1547 if (fp->rx_buf_ring == NULL)
1550 for (i = 0; i < NUM_RX_BD; i++) {
1551 struct sw_rx_bd *rx_buf = &fp->rx_buf_ring[i];
1552 u8 *data = rx_buf->data;
1556 dma_unmap_single(&bp->pdev->dev,
1557 dma_unmap_addr(rx_buf, mapping),
1558 fp->rx_buf_size, DMA_FROM_DEVICE);
1560 rx_buf->data = NULL;
1561 bnx2x_frag_free(fp, data);
1565 static void bnx2x_free_rx_skbs_cnic(struct bnx2x *bp)
1569 for_each_rx_queue_cnic(bp, j) {
1570 bnx2x_free_rx_bds(&bp->fp[j]);
1574 static void bnx2x_free_rx_skbs(struct bnx2x *bp)
1578 for_each_eth_queue(bp, j) {
1579 struct bnx2x_fastpath *fp = &bp->fp[j];
1581 bnx2x_free_rx_bds(fp);
1583 if (fp->mode != TPA_MODE_DISABLED)
1584 bnx2x_free_tpa_pool(bp, fp, MAX_AGG_QS(bp));
1588 static void bnx2x_free_skbs_cnic(struct bnx2x *bp)
1590 bnx2x_free_tx_skbs_cnic(bp);
1591 bnx2x_free_rx_skbs_cnic(bp);
1594 void bnx2x_free_skbs(struct bnx2x *bp)
1596 bnx2x_free_tx_skbs(bp);
1597 bnx2x_free_rx_skbs(bp);
1600 void bnx2x_update_max_mf_config(struct bnx2x *bp, u32 value)
1602 /* load old values */
1603 u32 mf_cfg = bp->mf_config[BP_VN(bp)];
1605 if (value != bnx2x_extract_max_cfg(bp, mf_cfg)) {
1606 /* leave all but MAX value */
1607 mf_cfg &= ~FUNC_MF_CFG_MAX_BW_MASK;
1609 /* set new MAX value */
1610 mf_cfg |= (value << FUNC_MF_CFG_MAX_BW_SHIFT)
1611 & FUNC_MF_CFG_MAX_BW_MASK;
1613 bnx2x_fw_command(bp, DRV_MSG_CODE_SET_MF_BW, mf_cfg);
1618 * bnx2x_free_msix_irqs - free previously requested MSI-X IRQ vectors
1620 * @bp: driver handle
1621 * @nvecs: number of vectors to be released
1623 static void bnx2x_free_msix_irqs(struct bnx2x *bp, int nvecs)
1627 if (nvecs == offset)
1630 /* VFs don't have a default SB */
1632 free_irq(bp->msix_table[offset].vector, bp->dev);
1633 DP(NETIF_MSG_IFDOWN, "released sp irq (%d)\n",
1634 bp->msix_table[offset].vector);
1638 if (CNIC_SUPPORT(bp)) {
1639 if (nvecs == offset)
1644 for_each_eth_queue(bp, i) {
1645 if (nvecs == offset)
1647 DP(NETIF_MSG_IFDOWN, "about to release fp #%d->%d irq\n",
1648 i, bp->msix_table[offset].vector);
1650 free_irq(bp->msix_table[offset++].vector, &bp->fp[i]);
1654 void bnx2x_free_irq(struct bnx2x *bp)
1656 if (bp->flags & USING_MSIX_FLAG &&
1657 !(bp->flags & USING_SINGLE_MSIX_FLAG)) {
1658 int nvecs = BNX2X_NUM_ETH_QUEUES(bp) + CNIC_SUPPORT(bp);
1660 /* vfs don't have a default status block */
1664 bnx2x_free_msix_irqs(bp, nvecs);
1666 free_irq(bp->dev->irq, bp->dev);
1670 int bnx2x_enable_msix(struct bnx2x *bp)
1672 int msix_vec = 0, i, rc;
1674 /* VFs don't have a default status block */
1676 bp->msix_table[msix_vec].entry = msix_vec;
1677 BNX2X_DEV_INFO("msix_table[0].entry = %d (slowpath)\n",
1678 bp->msix_table[0].entry);
1682 /* Cnic requires an msix vector for itself */
1683 if (CNIC_SUPPORT(bp)) {
1684 bp->msix_table[msix_vec].entry = msix_vec;
1685 BNX2X_DEV_INFO("msix_table[%d].entry = %d (CNIC)\n",
1686 msix_vec, bp->msix_table[msix_vec].entry);
1690 /* We need separate vectors for ETH queues only (not FCoE) */
1691 for_each_eth_queue(bp, i) {
1692 bp->msix_table[msix_vec].entry = msix_vec;
1693 BNX2X_DEV_INFO("msix_table[%d].entry = %d (fastpath #%u)\n",
1694 msix_vec, msix_vec, i);
1698 DP(BNX2X_MSG_SP, "about to request enable msix with %d vectors\n",
1701 rc = pci_enable_msix_range(bp->pdev, &bp->msix_table[0],
1702 BNX2X_MIN_MSIX_VEC_CNT(bp), msix_vec);
1704 * reconfigure number of tx/rx queues according to available
1707 if (rc == -ENOSPC) {
1708 /* Get by with single vector */
1709 rc = pci_enable_msix_range(bp->pdev, &bp->msix_table[0], 1, 1);
1711 BNX2X_DEV_INFO("Single MSI-X is not attainable rc %d\n",
1716 BNX2X_DEV_INFO("Using single MSI-X vector\n");
1717 bp->flags |= USING_SINGLE_MSIX_FLAG;
1719 BNX2X_DEV_INFO("set number of queues to 1\n");
1720 bp->num_ethernet_queues = 1;
1721 bp->num_queues = bp->num_ethernet_queues + bp->num_cnic_queues;
1722 } else if (rc < 0) {
1723 BNX2X_DEV_INFO("MSI-X is not attainable rc %d\n", rc);
1725 } else if (rc < msix_vec) {
1726 /* how less vectors we will have? */
1727 int diff = msix_vec - rc;
1729 BNX2X_DEV_INFO("Trying to use less MSI-X vectors: %d\n", rc);
1732 * decrease number of queues by number of unallocated entries
1734 bp->num_ethernet_queues -= diff;
1735 bp->num_queues = bp->num_ethernet_queues + bp->num_cnic_queues;
1737 BNX2X_DEV_INFO("New queue configuration set: %d\n",
1741 bp->flags |= USING_MSIX_FLAG;
1746 /* fall to INTx if not enough memory */
1748 bp->flags |= DISABLE_MSI_FLAG;
1753 static int bnx2x_req_msix_irqs(struct bnx2x *bp)
1755 int i, rc, offset = 0;
1757 /* no default status block for vf */
1759 rc = request_irq(bp->msix_table[offset++].vector,
1760 bnx2x_msix_sp_int, 0,
1761 bp->dev->name, bp->dev);
1763 BNX2X_ERR("request sp irq failed\n");
1768 if (CNIC_SUPPORT(bp))
1771 for_each_eth_queue(bp, i) {
1772 struct bnx2x_fastpath *fp = &bp->fp[i];
1773 snprintf(fp->name, sizeof(fp->name), "%s-fp-%d",
1776 rc = request_irq(bp->msix_table[offset].vector,
1777 bnx2x_msix_fp_int, 0, fp->name, fp);
1779 BNX2X_ERR("request fp #%d irq (%d) failed rc %d\n", i,
1780 bp->msix_table[offset].vector, rc);
1781 bnx2x_free_msix_irqs(bp, offset);
1788 i = BNX2X_NUM_ETH_QUEUES(bp);
1790 offset = 1 + CNIC_SUPPORT(bp);
1791 netdev_info(bp->dev,
1792 "using MSI-X IRQs: sp %d fp[%d] %d ... fp[%d] %d\n",
1793 bp->msix_table[0].vector,
1794 0, bp->msix_table[offset].vector,
1795 i - 1, bp->msix_table[offset + i - 1].vector);
1797 offset = CNIC_SUPPORT(bp);
1798 netdev_info(bp->dev,
1799 "using MSI-X IRQs: fp[%d] %d ... fp[%d] %d\n",
1800 0, bp->msix_table[offset].vector,
1801 i - 1, bp->msix_table[offset + i - 1].vector);
1806 int bnx2x_enable_msi(struct bnx2x *bp)
1810 rc = pci_enable_msi(bp->pdev);
1812 BNX2X_DEV_INFO("MSI is not attainable\n");
1815 bp->flags |= USING_MSI_FLAG;
1820 static int bnx2x_req_irq(struct bnx2x *bp)
1822 unsigned long flags;
1825 if (bp->flags & (USING_MSI_FLAG | USING_MSIX_FLAG))
1828 flags = IRQF_SHARED;
1830 if (bp->flags & USING_MSIX_FLAG)
1831 irq = bp->msix_table[0].vector;
1833 irq = bp->pdev->irq;
1835 return request_irq(irq, bnx2x_interrupt, flags, bp->dev->name, bp->dev);
1838 static int bnx2x_setup_irqs(struct bnx2x *bp)
1841 if (bp->flags & USING_MSIX_FLAG &&
1842 !(bp->flags & USING_SINGLE_MSIX_FLAG)) {
1843 rc = bnx2x_req_msix_irqs(bp);
1847 rc = bnx2x_req_irq(bp);
1849 BNX2X_ERR("IRQ request failed rc %d, aborting\n", rc);
1852 if (bp->flags & USING_MSI_FLAG) {
1853 bp->dev->irq = bp->pdev->irq;
1854 netdev_info(bp->dev, "using MSI IRQ %d\n",
1857 if (bp->flags & USING_MSIX_FLAG) {
1858 bp->dev->irq = bp->msix_table[0].vector;
1859 netdev_info(bp->dev, "using MSIX IRQ %d\n",
1867 static void bnx2x_napi_enable_cnic(struct bnx2x *bp)
1871 for_each_rx_queue_cnic(bp, i) {
1872 bnx2x_fp_busy_poll_init(&bp->fp[i]);
1873 napi_enable(&bnx2x_fp(bp, i, napi));
1877 static void bnx2x_napi_enable(struct bnx2x *bp)
1881 for_each_eth_queue(bp, i) {
1882 bnx2x_fp_busy_poll_init(&bp->fp[i]);
1883 napi_enable(&bnx2x_fp(bp, i, napi));
1887 static void bnx2x_napi_disable_cnic(struct bnx2x *bp)
1891 for_each_rx_queue_cnic(bp, i) {
1892 napi_disable(&bnx2x_fp(bp, i, napi));
1893 while (!bnx2x_fp_ll_disable(&bp->fp[i]))
1894 usleep_range(1000, 2000);
1898 static void bnx2x_napi_disable(struct bnx2x *bp)
1902 for_each_eth_queue(bp, i) {
1903 napi_disable(&bnx2x_fp(bp, i, napi));
1904 while (!bnx2x_fp_ll_disable(&bp->fp[i]))
1905 usleep_range(1000, 2000);
1909 void bnx2x_netif_start(struct bnx2x *bp)
1911 if (netif_running(bp->dev)) {
1912 bnx2x_napi_enable(bp);
1913 if (CNIC_LOADED(bp))
1914 bnx2x_napi_enable_cnic(bp);
1915 bnx2x_int_enable(bp);
1916 if (bp->state == BNX2X_STATE_OPEN)
1917 netif_tx_wake_all_queues(bp->dev);
1921 void bnx2x_netif_stop(struct bnx2x *bp, int disable_hw)
1923 bnx2x_int_disable_sync(bp, disable_hw);
1924 bnx2x_napi_disable(bp);
1925 if (CNIC_LOADED(bp))
1926 bnx2x_napi_disable_cnic(bp);
1929 u16 bnx2x_select_queue(struct net_device *dev, struct sk_buff *skb,
1930 void *accel_priv, select_queue_fallback_t fallback)
1932 struct bnx2x *bp = netdev_priv(dev);
1934 if (CNIC_LOADED(bp) && !NO_FCOE(bp)) {
1935 struct ethhdr *hdr = (struct ethhdr *)skb->data;
1936 u16 ether_type = ntohs(hdr->h_proto);
1938 /* Skip VLAN tag if present */
1939 if (ether_type == ETH_P_8021Q) {
1940 struct vlan_ethhdr *vhdr =
1941 (struct vlan_ethhdr *)skb->data;
1943 ether_type = ntohs(vhdr->h_vlan_encapsulated_proto);
1946 /* If ethertype is FCoE or FIP - use FCoE ring */
1947 if ((ether_type == ETH_P_FCOE) || (ether_type == ETH_P_FIP))
1948 return bnx2x_fcoe_tx(bp, txq_index);
1951 /* select a non-FCoE queue */
1952 return fallback(dev, skb) % BNX2X_NUM_ETH_QUEUES(bp);
1955 void bnx2x_set_num_queues(struct bnx2x *bp)
1958 bp->num_ethernet_queues = bnx2x_calc_num_queues(bp);
1960 /* override in STORAGE SD modes */
1961 if (IS_MF_STORAGE_ONLY(bp))
1962 bp->num_ethernet_queues = 1;
1964 /* Add special queues */
1965 bp->num_cnic_queues = CNIC_SUPPORT(bp); /* For FCOE */
1966 bp->num_queues = bp->num_ethernet_queues + bp->num_cnic_queues;
1968 BNX2X_DEV_INFO("set number of queues to %d\n", bp->num_queues);
1972 * bnx2x_set_real_num_queues - configure netdev->real_num_[tx,rx]_queues
1974 * @bp: Driver handle
1976 * We currently support for at most 16 Tx queues for each CoS thus we will
1977 * allocate a multiple of 16 for ETH L2 rings according to the value of the
1980 * If there is an FCoE L2 queue the appropriate Tx queue will have the next
1981 * index after all ETH L2 indices.
1983 * If the actual number of Tx queues (for each CoS) is less than 16 then there
1984 * will be the holes at the end of each group of 16 ETh L2 indices (0..15,
1985 * 16..31,...) with indices that are not coupled with any real Tx queue.
1987 * The proper configuration of skb->queue_mapping is handled by
1988 * bnx2x_select_queue() and __skb_tx_hash().
1990 * bnx2x_setup_tc() takes care of the proper TC mappings so that __skb_tx_hash()
1991 * will return a proper Tx index if TC is enabled (netdev->num_tc > 0).
1993 static int bnx2x_set_real_num_queues(struct bnx2x *bp, int include_cnic)
1997 tx = BNX2X_NUM_ETH_QUEUES(bp) * bp->max_cos;
1998 rx = BNX2X_NUM_ETH_QUEUES(bp);
2000 /* account for fcoe queue */
2001 if (include_cnic && !NO_FCOE(bp)) {
2006 rc = netif_set_real_num_tx_queues(bp->dev, tx);
2008 BNX2X_ERR("Failed to set real number of Tx queues: %d\n", rc);
2011 rc = netif_set_real_num_rx_queues(bp->dev, rx);
2013 BNX2X_ERR("Failed to set real number of Rx queues: %d\n", rc);
2017 DP(NETIF_MSG_IFUP, "Setting real num queues to (tx, rx) (%d, %d)\n",
2023 static void bnx2x_set_rx_buf_size(struct bnx2x *bp)
2027 for_each_queue(bp, i) {
2028 struct bnx2x_fastpath *fp = &bp->fp[i];
2031 /* Always use a mini-jumbo MTU for the FCoE L2 ring */
2034 * Although there are no IP frames expected to arrive to
2035 * this ring we still want to add an
2036 * IP_HEADER_ALIGNMENT_PADDING to prevent a buffer
2039 mtu = BNX2X_FCOE_MINI_JUMBO_MTU;
2042 fp->rx_buf_size = BNX2X_FW_RX_ALIGN_START +
2043 IP_HEADER_ALIGNMENT_PADDING +
2046 BNX2X_FW_RX_ALIGN_END;
2047 /* Note : rx_buf_size doesn't take into account NET_SKB_PAD */
2048 if (fp->rx_buf_size + NET_SKB_PAD <= PAGE_SIZE)
2049 fp->rx_frag_size = fp->rx_buf_size + NET_SKB_PAD;
2051 fp->rx_frag_size = 0;
2055 static int bnx2x_init_rss(struct bnx2x *bp)
2058 u8 num_eth_queues = BNX2X_NUM_ETH_QUEUES(bp);
2060 /* Prepare the initial contents for the indirection table if RSS is
2063 for (i = 0; i < sizeof(bp->rss_conf_obj.ind_table); i++)
2064 bp->rss_conf_obj.ind_table[i] =
2066 ethtool_rxfh_indir_default(i, num_eth_queues);
2069 * For 57710 and 57711 SEARCHER configuration (rss_keys) is
2070 * per-port, so if explicit configuration is needed , do it only
2073 * For 57712 and newer on the other hand it's a per-function
2076 return bnx2x_config_rss_eth(bp, bp->port.pmf || !CHIP_IS_E1x(bp));
2079 int bnx2x_rss(struct bnx2x *bp, struct bnx2x_rss_config_obj *rss_obj,
2080 bool config_hash, bool enable)
2082 struct bnx2x_config_rss_params params = {NULL};
2084 /* Although RSS is meaningless when there is a single HW queue we
2085 * still need it enabled in order to have HW Rx hash generated.
2087 * if (!is_eth_multi(bp))
2088 * bp->multi_mode = ETH_RSS_MODE_DISABLED;
2091 params.rss_obj = rss_obj;
2093 __set_bit(RAMROD_COMP_WAIT, ¶ms.ramrod_flags);
2096 __set_bit(BNX2X_RSS_MODE_REGULAR, ¶ms.rss_flags);
2098 /* RSS configuration */
2099 __set_bit(BNX2X_RSS_IPV4, ¶ms.rss_flags);
2100 __set_bit(BNX2X_RSS_IPV4_TCP, ¶ms.rss_flags);
2101 __set_bit(BNX2X_RSS_IPV6, ¶ms.rss_flags);
2102 __set_bit(BNX2X_RSS_IPV6_TCP, ¶ms.rss_flags);
2103 if (rss_obj->udp_rss_v4)
2104 __set_bit(BNX2X_RSS_IPV4_UDP, ¶ms.rss_flags);
2105 if (rss_obj->udp_rss_v6)
2106 __set_bit(BNX2X_RSS_IPV6_UDP, ¶ms.rss_flags);
2108 if (!CHIP_IS_E1x(bp)) {
2109 /* valid only for TUNN_MODE_VXLAN tunnel mode */
2110 __set_bit(BNX2X_RSS_IPV4_VXLAN, ¶ms.rss_flags);
2111 __set_bit(BNX2X_RSS_IPV6_VXLAN, ¶ms.rss_flags);
2113 /* valid only for TUNN_MODE_GRE tunnel mode */
2114 __set_bit(BNX2X_RSS_TUNN_INNER_HDRS, ¶ms.rss_flags);
2117 __set_bit(BNX2X_RSS_MODE_DISABLED, ¶ms.rss_flags);
2121 params.rss_result_mask = MULTI_MASK;
2123 memcpy(params.ind_table, rss_obj->ind_table, sizeof(params.ind_table));
2127 netdev_rss_key_fill(params.rss_key, T_ETH_RSS_KEY * 4);
2128 __set_bit(BNX2X_RSS_SET_SRCH, ¶ms.rss_flags);
2132 return bnx2x_config_rss(bp, ¶ms);
2134 return bnx2x_vfpf_config_rss(bp, ¶ms);
2137 static int bnx2x_init_hw(struct bnx2x *bp, u32 load_code)
2139 struct bnx2x_func_state_params func_params = {NULL};
2141 /* Prepare parameters for function state transitions */
2142 __set_bit(RAMROD_COMP_WAIT, &func_params.ramrod_flags);
2144 func_params.f_obj = &bp->func_obj;
2145 func_params.cmd = BNX2X_F_CMD_HW_INIT;
2147 func_params.params.hw_init.load_phase = load_code;
2149 return bnx2x_func_state_change(bp, &func_params);
2153 * Cleans the object that have internal lists without sending
2154 * ramrods. Should be run when interrupts are disabled.
2156 void bnx2x_squeeze_objects(struct bnx2x *bp)
2159 unsigned long ramrod_flags = 0, vlan_mac_flags = 0;
2160 struct bnx2x_mcast_ramrod_params rparam = {NULL};
2161 struct bnx2x_vlan_mac_obj *mac_obj = &bp->sp_objs->mac_obj;
2163 /***************** Cleanup MACs' object first *************************/
2165 /* Wait for completion of requested */
2166 __set_bit(RAMROD_COMP_WAIT, &ramrod_flags);
2167 /* Perform a dry cleanup */
2168 __set_bit(RAMROD_DRV_CLR_ONLY, &ramrod_flags);
2170 /* Clean ETH primary MAC */
2171 __set_bit(BNX2X_ETH_MAC, &vlan_mac_flags);
2172 rc = mac_obj->delete_all(bp, &bp->sp_objs->mac_obj, &vlan_mac_flags,
2175 BNX2X_ERR("Failed to clean ETH MACs: %d\n", rc);
2177 /* Cleanup UC list */
2179 __set_bit(BNX2X_UC_LIST_MAC, &vlan_mac_flags);
2180 rc = mac_obj->delete_all(bp, mac_obj, &vlan_mac_flags,
2183 BNX2X_ERR("Failed to clean UC list MACs: %d\n", rc);
2185 /***************** Now clean mcast object *****************************/
2186 rparam.mcast_obj = &bp->mcast_obj;
2187 __set_bit(RAMROD_DRV_CLR_ONLY, &rparam.ramrod_flags);
2189 /* Add a DEL command... - Since we're doing a driver cleanup only,
2190 * we take a lock surrounding both the initial send and the CONTs,
2191 * as we don't want a true completion to disrupt us in the middle.
2193 netif_addr_lock_bh(bp->dev);
2194 rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_DEL);
2196 BNX2X_ERR("Failed to add a new DEL command to a multi-cast object: %d\n",
2199 /* ...and wait until all pending commands are cleared */
2200 rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_CONT);
2203 BNX2X_ERR("Failed to clean multi-cast object: %d\n",
2205 netif_addr_unlock_bh(bp->dev);
2209 rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_CONT);
2211 netif_addr_unlock_bh(bp->dev);
2214 #ifndef BNX2X_STOP_ON_ERROR
2215 #define LOAD_ERROR_EXIT(bp, label) \
2217 (bp)->state = BNX2X_STATE_ERROR; \
2221 #define LOAD_ERROR_EXIT_CNIC(bp, label) \
2223 bp->cnic_loaded = false; \
2226 #else /*BNX2X_STOP_ON_ERROR*/
2227 #define LOAD_ERROR_EXIT(bp, label) \
2229 (bp)->state = BNX2X_STATE_ERROR; \
2233 #define LOAD_ERROR_EXIT_CNIC(bp, label) \
2235 bp->cnic_loaded = false; \
2239 #endif /*BNX2X_STOP_ON_ERROR*/
2241 static void bnx2x_free_fw_stats_mem(struct bnx2x *bp)
2243 BNX2X_PCI_FREE(bp->fw_stats, bp->fw_stats_mapping,
2244 bp->fw_stats_data_sz + bp->fw_stats_req_sz);
2248 static int bnx2x_alloc_fw_stats_mem(struct bnx2x *bp)
2250 int num_groups, vf_headroom = 0;
2251 int is_fcoe_stats = NO_FCOE(bp) ? 0 : 1;
2253 /* number of queues for statistics is number of eth queues + FCoE */
2254 u8 num_queue_stats = BNX2X_NUM_ETH_QUEUES(bp) + is_fcoe_stats;
2256 /* Total number of FW statistics requests =
2257 * 1 for port stats + 1 for PF stats + potential 2 for FCoE (fcoe proper
2258 * and fcoe l2 queue) stats + num of queues (which includes another 1
2259 * for fcoe l2 queue if applicable)
2261 bp->fw_stats_num = 2 + is_fcoe_stats + num_queue_stats;
2263 /* vf stats appear in the request list, but their data is allocated by
2264 * the VFs themselves. We don't include them in the bp->fw_stats_num as
2265 * it is used to determine where to place the vf stats queries in the
2269 vf_headroom = bnx2x_vf_headroom(bp);
2271 /* Request is built from stats_query_header and an array of
2272 * stats_query_cmd_group each of which contains
2273 * STATS_QUERY_CMD_COUNT rules. The real number or requests is
2274 * configured in the stats_query_header.
2277 (((bp->fw_stats_num + vf_headroom) / STATS_QUERY_CMD_COUNT) +
2278 (((bp->fw_stats_num + vf_headroom) % STATS_QUERY_CMD_COUNT) ?
2281 DP(BNX2X_MSG_SP, "stats fw_stats_num %d, vf headroom %d, num_groups %d\n",
2282 bp->fw_stats_num, vf_headroom, num_groups);
2283 bp->fw_stats_req_sz = sizeof(struct stats_query_header) +
2284 num_groups * sizeof(struct stats_query_cmd_group);
2286 /* Data for statistics requests + stats_counter
2287 * stats_counter holds per-STORM counters that are incremented
2288 * when STORM has finished with the current request.
2289 * memory for FCoE offloaded statistics are counted anyway,
2290 * even if they will not be sent.
2291 * VF stats are not accounted for here as the data of VF stats is stored
2292 * in memory allocated by the VF, not here.
2294 bp->fw_stats_data_sz = sizeof(struct per_port_stats) +
2295 sizeof(struct per_pf_stats) +
2296 sizeof(struct fcoe_statistics_params) +
2297 sizeof(struct per_queue_stats) * num_queue_stats +
2298 sizeof(struct stats_counter);
2300 bp->fw_stats = BNX2X_PCI_ALLOC(&bp->fw_stats_mapping,
2301 bp->fw_stats_data_sz + bp->fw_stats_req_sz);
2306 bp->fw_stats_req = (struct bnx2x_fw_stats_req *)bp->fw_stats;
2307 bp->fw_stats_req_mapping = bp->fw_stats_mapping;
2308 bp->fw_stats_data = (struct bnx2x_fw_stats_data *)
2309 ((u8 *)bp->fw_stats + bp->fw_stats_req_sz);
2310 bp->fw_stats_data_mapping = bp->fw_stats_mapping +
2311 bp->fw_stats_req_sz;
2313 DP(BNX2X_MSG_SP, "statistics request base address set to %x %x\n",
2314 U64_HI(bp->fw_stats_req_mapping),
2315 U64_LO(bp->fw_stats_req_mapping));
2316 DP(BNX2X_MSG_SP, "statistics data base address set to %x %x\n",
2317 U64_HI(bp->fw_stats_data_mapping),
2318 U64_LO(bp->fw_stats_data_mapping));
2322 bnx2x_free_fw_stats_mem(bp);
2323 BNX2X_ERR("Can't allocate FW stats memory\n");
2327 /* send load request to mcp and analyze response */
2328 static int bnx2x_nic_load_request(struct bnx2x *bp, u32 *load_code)
2334 (SHMEM_RD(bp, func_mb[BP_FW_MB_IDX(bp)].drv_mb_header) &
2335 DRV_MSG_SEQ_NUMBER_MASK);
2336 BNX2X_DEV_INFO("fw_seq 0x%08x\n", bp->fw_seq);
2338 /* Get current FW pulse sequence */
2339 bp->fw_drv_pulse_wr_seq =
2340 (SHMEM_RD(bp, func_mb[BP_FW_MB_IDX(bp)].drv_pulse_mb) &
2341 DRV_PULSE_SEQ_MASK);
2342 BNX2X_DEV_INFO("drv_pulse 0x%x\n", bp->fw_drv_pulse_wr_seq);
2344 param = DRV_MSG_CODE_LOAD_REQ_WITH_LFA;
2346 if (IS_MF_SD(bp) && bnx2x_port_after_undi(bp))
2347 param |= DRV_MSG_CODE_LOAD_REQ_FORCE_LFA;
2350 (*load_code) = bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_REQ, param);
2352 /* if mcp fails to respond we must abort */
2353 if (!(*load_code)) {
2354 BNX2X_ERR("MCP response failure, aborting\n");
2358 /* If mcp refused (e.g. other port is in diagnostic mode) we
2361 if ((*load_code) == FW_MSG_CODE_DRV_LOAD_REFUSED) {
2362 BNX2X_ERR("MCP refused load request, aborting\n");
2368 /* check whether another PF has already loaded FW to chip. In
2369 * virtualized environments a pf from another VM may have already
2370 * initialized the device including loading FW
2372 int bnx2x_compare_fw_ver(struct bnx2x *bp, u32 load_code, bool print_err)
2374 /* is another pf loaded on this engine? */
2375 if (load_code != FW_MSG_CODE_DRV_LOAD_COMMON_CHIP &&
2376 load_code != FW_MSG_CODE_DRV_LOAD_COMMON) {
2377 /* build my FW version dword */
2378 u32 my_fw = (BCM_5710_FW_MAJOR_VERSION) +
2379 (BCM_5710_FW_MINOR_VERSION << 8) +
2380 (BCM_5710_FW_REVISION_VERSION << 16) +
2381 (BCM_5710_FW_ENGINEERING_VERSION << 24);
2383 /* read loaded FW from chip */
2384 u32 loaded_fw = REG_RD(bp, XSEM_REG_PRAM);
2386 DP(BNX2X_MSG_SP, "loaded fw %x, my fw %x\n",
2389 /* abort nic load if version mismatch */
2390 if (my_fw != loaded_fw) {
2392 BNX2X_ERR("bnx2x with FW %x was already loaded which mismatches my %x FW. Aborting\n",
2395 BNX2X_DEV_INFO("bnx2x with FW %x was already loaded which mismatches my %x FW, possibly due to MF UNDI\n",
2403 /* returns the "mcp load_code" according to global load_count array */
2404 static int bnx2x_nic_load_no_mcp(struct bnx2x *bp, int port)
2406 int path = BP_PATH(bp);
2408 DP(NETIF_MSG_IFUP, "NO MCP - load counts[%d] %d, %d, %d\n",
2409 path, bnx2x_load_count[path][0], bnx2x_load_count[path][1],
2410 bnx2x_load_count[path][2]);
2411 bnx2x_load_count[path][0]++;
2412 bnx2x_load_count[path][1 + port]++;
2413 DP(NETIF_MSG_IFUP, "NO MCP - new load counts[%d] %d, %d, %d\n",
2414 path, bnx2x_load_count[path][0], bnx2x_load_count[path][1],
2415 bnx2x_load_count[path][2]);
2416 if (bnx2x_load_count[path][0] == 1)
2417 return FW_MSG_CODE_DRV_LOAD_COMMON;
2418 else if (bnx2x_load_count[path][1 + port] == 1)
2419 return FW_MSG_CODE_DRV_LOAD_PORT;
2421 return FW_MSG_CODE_DRV_LOAD_FUNCTION;
2424 /* mark PMF if applicable */
2425 static void bnx2x_nic_load_pmf(struct bnx2x *bp, u32 load_code)
2427 if ((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
2428 (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP) ||
2429 (load_code == FW_MSG_CODE_DRV_LOAD_PORT)) {
2431 /* We need the barrier to ensure the ordering between the
2432 * writing to bp->port.pmf here and reading it from the
2433 * bnx2x_periodic_task().
2440 DP(NETIF_MSG_LINK, "pmf %d\n", bp->port.pmf);
2443 static void bnx2x_nic_load_afex_dcc(struct bnx2x *bp, int load_code)
2445 if (((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
2446 (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP)) &&
2447 (bp->common.shmem2_base)) {
2448 if (SHMEM2_HAS(bp, dcc_support))
2449 SHMEM2_WR(bp, dcc_support,
2450 (SHMEM_DCC_SUPPORT_DISABLE_ENABLE_PF_TLV |
2451 SHMEM_DCC_SUPPORT_BANDWIDTH_ALLOCATION_TLV));
2452 if (SHMEM2_HAS(bp, afex_driver_support))
2453 SHMEM2_WR(bp, afex_driver_support,
2454 SHMEM_AFEX_SUPPORTED_VERSION_ONE);
2457 /* Set AFEX default VLAN tag to an invalid value */
2458 bp->afex_def_vlan_tag = -1;
2462 * bnx2x_bz_fp - zero content of the fastpath structure.
2464 * @bp: driver handle
2465 * @index: fastpath index to be zeroed
2467 * Makes sure the contents of the bp->fp[index].napi is kept
2470 static void bnx2x_bz_fp(struct bnx2x *bp, int index)
2472 struct bnx2x_fastpath *fp = &bp->fp[index];
2474 struct napi_struct orig_napi = fp->napi;
2475 struct bnx2x_agg_info *orig_tpa_info = fp->tpa_info;
2477 /* bzero bnx2x_fastpath contents */
2479 memset(fp->tpa_info, 0, ETH_MAX_AGGREGATION_QUEUES_E1H_E2 *
2480 sizeof(struct bnx2x_agg_info));
2481 memset(fp, 0, sizeof(*fp));
2483 /* Restore the NAPI object as it has been already initialized */
2484 fp->napi = orig_napi;
2485 fp->tpa_info = orig_tpa_info;
2489 fp->max_cos = bp->max_cos;
2491 /* Special queues support only one CoS */
2494 /* Init txdata pointers */
2496 fp->txdata_ptr[0] = &bp->bnx2x_txq[FCOE_TXQ_IDX(bp)];
2498 for_each_cos_in_tx_queue(fp, cos)
2499 fp->txdata_ptr[cos] = &bp->bnx2x_txq[cos *
2500 BNX2X_NUM_ETH_QUEUES(bp) + index];
2502 /* set the tpa flag for each queue. The tpa flag determines the queue
2503 * minimal size so it must be set prior to queue memory allocation
2505 if (bp->dev->features & NETIF_F_LRO)
2506 fp->mode = TPA_MODE_LRO;
2507 else if (bp->dev->features & NETIF_F_GRO &&
2508 bnx2x_mtu_allows_gro(bp->dev->mtu))
2509 fp->mode = TPA_MODE_GRO;
2511 fp->mode = TPA_MODE_DISABLED;
2513 /* We don't want TPA if it's disabled in bp
2514 * or if this is an FCoE L2 ring.
2516 if (bp->disable_tpa || IS_FCOE_FP(fp))
2517 fp->mode = TPA_MODE_DISABLED;
2520 int bnx2x_load_cnic(struct bnx2x *bp)
2522 int i, rc, port = BP_PORT(bp);
2524 DP(NETIF_MSG_IFUP, "Starting CNIC-related load\n");
2526 mutex_init(&bp->cnic_mutex);
2529 rc = bnx2x_alloc_mem_cnic(bp);
2531 BNX2X_ERR("Unable to allocate bp memory for cnic\n");
2532 LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic0);
2536 rc = bnx2x_alloc_fp_mem_cnic(bp);
2538 BNX2X_ERR("Unable to allocate memory for cnic fps\n");
2539 LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic0);
2542 /* Update the number of queues with the cnic queues */
2543 rc = bnx2x_set_real_num_queues(bp, 1);
2545 BNX2X_ERR("Unable to set real_num_queues including cnic\n");
2546 LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic0);
2549 /* Add all CNIC NAPI objects */
2550 bnx2x_add_all_napi_cnic(bp);
2551 DP(NETIF_MSG_IFUP, "cnic napi added\n");
2552 bnx2x_napi_enable_cnic(bp);
2554 rc = bnx2x_init_hw_func_cnic(bp);
2556 LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic1);
2558 bnx2x_nic_init_cnic(bp);
2561 /* Enable Timer scan */
2562 REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + port*4, 1);
2564 /* setup cnic queues */
2565 for_each_cnic_queue(bp, i) {
2566 rc = bnx2x_setup_queue(bp, &bp->fp[i], 0);
2568 BNX2X_ERR("Queue setup failed\n");
2569 LOAD_ERROR_EXIT(bp, load_error_cnic2);
2574 /* Initialize Rx filter. */
2575 bnx2x_set_rx_mode_inner(bp);
2577 /* re-read iscsi info */
2578 bnx2x_get_iscsi_info(bp);
2579 bnx2x_setup_cnic_irq_info(bp);
2580 bnx2x_setup_cnic_info(bp);
2581 bp->cnic_loaded = true;
2582 if (bp->state == BNX2X_STATE_OPEN)
2583 bnx2x_cnic_notify(bp, CNIC_CTL_START_CMD);
2585 DP(NETIF_MSG_IFUP, "Ending successfully CNIC-related load\n");
2589 #ifndef BNX2X_STOP_ON_ERROR
2591 /* Disable Timer scan */
2592 REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + port*4, 0);
2595 bnx2x_napi_disable_cnic(bp);
2596 /* Update the number of queues without the cnic queues */
2597 if (bnx2x_set_real_num_queues(bp, 0))
2598 BNX2X_ERR("Unable to set real_num_queues not including cnic\n");
2600 BNX2X_ERR("CNIC-related load failed\n");
2601 bnx2x_free_fp_mem_cnic(bp);
2602 bnx2x_free_mem_cnic(bp);
2604 #endif /* ! BNX2X_STOP_ON_ERROR */
2607 /* must be called with rtnl_lock */
2608 int bnx2x_nic_load(struct bnx2x *bp, int load_mode)
2610 int port = BP_PORT(bp);
2611 int i, rc = 0, load_code = 0;
2613 DP(NETIF_MSG_IFUP, "Starting NIC load\n");
2615 "CNIC is %s\n", CNIC_ENABLED(bp) ? "enabled" : "disabled");
2617 #ifdef BNX2X_STOP_ON_ERROR
2618 if (unlikely(bp->panic)) {
2619 BNX2X_ERR("Can't load NIC when there is panic\n");
2624 bp->state = BNX2X_STATE_OPENING_WAIT4_LOAD;
2626 /* zero the structure w/o any lock, before SP handler is initialized */
2627 memset(&bp->last_reported_link, 0, sizeof(bp->last_reported_link));
2628 __set_bit(BNX2X_LINK_REPORT_LINK_DOWN,
2629 &bp->last_reported_link.link_report_flags);
2632 /* must be called before memory allocation and HW init */
2633 bnx2x_ilt_set_info(bp);
2636 * Zero fastpath structures preserving invariants like napi, which are
2637 * allocated only once, fp index, max_cos, bp pointer.
2638 * Also set fp->mode and txdata_ptr.
2640 DP(NETIF_MSG_IFUP, "num queues: %d", bp->num_queues);
2641 for_each_queue(bp, i)
2643 memset(bp->bnx2x_txq, 0, (BNX2X_MAX_RSS_COUNT(bp) * BNX2X_MULTI_TX_COS +
2644 bp->num_cnic_queues) *
2645 sizeof(struct bnx2x_fp_txdata));
2647 bp->fcoe_init = false;
2649 /* Set the receive queues buffer size */
2650 bnx2x_set_rx_buf_size(bp);
2653 rc = bnx2x_alloc_mem(bp);
2655 BNX2X_ERR("Unable to allocate bp memory\n");
2660 /* need to be done after alloc mem, since it's self adjusting to amount
2661 * of memory available for RSS queues
2663 rc = bnx2x_alloc_fp_mem(bp);
2665 BNX2X_ERR("Unable to allocate memory for fps\n");
2666 LOAD_ERROR_EXIT(bp, load_error0);
2669 /* Allocated memory for FW statistics */
2670 if (bnx2x_alloc_fw_stats_mem(bp))
2671 LOAD_ERROR_EXIT(bp, load_error0);
2673 /* request pf to initialize status blocks */
2675 rc = bnx2x_vfpf_init(bp);
2677 LOAD_ERROR_EXIT(bp, load_error0);
2680 /* As long as bnx2x_alloc_mem() may possibly update
2681 * bp->num_queues, bnx2x_set_real_num_queues() should always
2682 * come after it. At this stage cnic queues are not counted.
2684 rc = bnx2x_set_real_num_queues(bp, 0);
2686 BNX2X_ERR("Unable to set real_num_queues\n");
2687 LOAD_ERROR_EXIT(bp, load_error0);
2690 /* configure multi cos mappings in kernel.
2691 * this configuration may be overridden by a multi class queue
2692 * discipline or by a dcbx negotiation result.
2694 bnx2x_setup_tc(bp->dev, bp->max_cos);
2696 /* Add all NAPI objects */
2697 bnx2x_add_all_napi(bp);
2698 DP(NETIF_MSG_IFUP, "napi added\n");
2699 bnx2x_napi_enable(bp);
2702 /* set pf load just before approaching the MCP */
2703 bnx2x_set_pf_load(bp);
2705 /* if mcp exists send load request and analyze response */
2706 if (!BP_NOMCP(bp)) {
2707 /* attempt to load pf */
2708 rc = bnx2x_nic_load_request(bp, &load_code);
2710 LOAD_ERROR_EXIT(bp, load_error1);
2712 /* what did mcp say? */
2713 rc = bnx2x_compare_fw_ver(bp, load_code, true);
2715 bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2716 LOAD_ERROR_EXIT(bp, load_error2);
2719 load_code = bnx2x_nic_load_no_mcp(bp, port);
2722 /* mark pmf if applicable */
2723 bnx2x_nic_load_pmf(bp, load_code);
2725 /* Init Function state controlling object */
2726 bnx2x__init_func_obj(bp);
2729 rc = bnx2x_init_hw(bp, load_code);
2731 BNX2X_ERR("HW init failed, aborting\n");
2732 bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2733 LOAD_ERROR_EXIT(bp, load_error2);
2737 bnx2x_pre_irq_nic_init(bp);
2739 /* Connect to IRQs */
2740 rc = bnx2x_setup_irqs(bp);
2742 BNX2X_ERR("setup irqs failed\n");
2744 bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2745 LOAD_ERROR_EXIT(bp, load_error2);
2748 /* Init per-function objects */
2750 /* Setup NIC internals and enable interrupts */
2751 bnx2x_post_irq_nic_init(bp, load_code);
2753 bnx2x_init_bp_objs(bp);
2754 bnx2x_iov_nic_init(bp);
2756 /* Set AFEX default VLAN tag to an invalid value */
2757 bp->afex_def_vlan_tag = -1;
2758 bnx2x_nic_load_afex_dcc(bp, load_code);
2759 bp->state = BNX2X_STATE_OPENING_WAIT4_PORT;
2760 rc = bnx2x_func_start(bp);
2762 BNX2X_ERR("Function start failed!\n");
2763 bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2765 LOAD_ERROR_EXIT(bp, load_error3);
2768 /* Send LOAD_DONE command to MCP */
2769 if (!BP_NOMCP(bp)) {
2770 load_code = bnx2x_fw_command(bp,
2771 DRV_MSG_CODE_LOAD_DONE, 0);
2773 BNX2X_ERR("MCP response failure, aborting\n");
2775 LOAD_ERROR_EXIT(bp, load_error3);
2779 /* initialize FW coalescing state machines in RAM */
2780 bnx2x_update_coalesce(bp);
2783 /* setup the leading queue */
2784 rc = bnx2x_setup_leading(bp);
2786 BNX2X_ERR("Setup leading failed!\n");
2787 LOAD_ERROR_EXIT(bp, load_error3);
2790 /* set up the rest of the queues */
2791 for_each_nondefault_eth_queue(bp, i) {
2793 rc = bnx2x_setup_queue(bp, &bp->fp[i], false);
2795 rc = bnx2x_vfpf_setup_q(bp, &bp->fp[i], false);
2797 BNX2X_ERR("Queue %d setup failed\n", i);
2798 LOAD_ERROR_EXIT(bp, load_error3);
2803 rc = bnx2x_init_rss(bp);
2805 BNX2X_ERR("PF RSS init failed\n");
2806 LOAD_ERROR_EXIT(bp, load_error3);
2809 /* Now when Clients are configured we are ready to work */
2810 bp->state = BNX2X_STATE_OPEN;
2812 /* Configure a ucast MAC */
2814 rc = bnx2x_set_eth_mac(bp, true);
2816 rc = bnx2x_vfpf_config_mac(bp, bp->dev->dev_addr, bp->fp->index,
2819 BNX2X_ERR("Setting Ethernet MAC failed\n");
2820 LOAD_ERROR_EXIT(bp, load_error3);
2823 if (IS_PF(bp) && bp->pending_max) {
2824 bnx2x_update_max_mf_config(bp, bp->pending_max);
2825 bp->pending_max = 0;
2829 rc = bnx2x_initial_phy_init(bp, load_mode);
2831 LOAD_ERROR_EXIT(bp, load_error3);
2833 bp->link_params.feature_config_flags &= ~FEATURE_CONFIG_BOOT_FROM_SAN;
2835 /* Start fast path */
2837 /* Initialize Rx filter. */
2838 bnx2x_set_rx_mode_inner(bp);
2840 if (bp->flags & PTP_SUPPORTED) {
2842 bnx2x_configure_ptp_filters(bp);
2845 switch (load_mode) {
2847 /* Tx queue should be only re-enabled */
2848 netif_tx_wake_all_queues(bp->dev);
2852 netif_tx_start_all_queues(bp->dev);
2853 smp_mb__after_atomic();
2857 case LOAD_LOOPBACK_EXT:
2858 bp->state = BNX2X_STATE_DIAG;
2866 bnx2x_update_drv_flags(bp, 1 << DRV_FLAGS_PORT_MASK, 0);
2868 bnx2x__link_status_update(bp);
2870 /* start the timer */
2871 mod_timer(&bp->timer, jiffies + bp->current_interval);
2873 if (CNIC_ENABLED(bp))
2874 bnx2x_load_cnic(bp);
2877 bnx2x_schedule_sp_rtnl(bp, BNX2X_SP_RTNL_GET_DRV_VERSION, 0);
2879 if (IS_PF(bp) && SHMEM2_HAS(bp, drv_capabilities_flag)) {
2880 /* mark driver is loaded in shmem2 */
2882 val = SHMEM2_RD(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)]);
2883 SHMEM2_WR(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)],
2884 val | DRV_FLAGS_CAPABILITIES_LOADED_SUPPORTED |
2885 DRV_FLAGS_CAPABILITIES_LOADED_L2);
2888 /* Wait for all pending SP commands to complete */
2889 if (IS_PF(bp) && !bnx2x_wait_sp_comp(bp, ~0x0UL)) {
2890 BNX2X_ERR("Timeout waiting for SP elements to complete\n");
2891 bnx2x_nic_unload(bp, UNLOAD_CLOSE, false);
2895 /* If PMF - send ADMIN DCBX msg to MFW to initiate DCBX FSM */
2896 if (bp->port.pmf && (bp->state != BNX2X_STATE_DIAG))
2897 bnx2x_dcbx_init(bp, false);
2899 DP(NETIF_MSG_IFUP, "Ending successfully NIC load\n");
2903 #ifndef BNX2X_STOP_ON_ERROR
2906 bnx2x_int_disable_sync(bp, 1);
2908 /* Clean queueable objects */
2909 bnx2x_squeeze_objects(bp);
2912 /* Free SKBs, SGEs, TPA pool and driver internals */
2913 bnx2x_free_skbs(bp);
2914 for_each_rx_queue(bp, i)
2915 bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
2920 if (IS_PF(bp) && !BP_NOMCP(bp)) {
2921 bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_REQ_WOL_MCP, 0);
2922 bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_DONE, 0);
2927 bnx2x_napi_disable(bp);
2928 bnx2x_del_all_napi(bp);
2930 /* clear pf_load status, as it was already set */
2932 bnx2x_clear_pf_load(bp);
2934 bnx2x_free_fw_stats_mem(bp);
2935 bnx2x_free_fp_mem(bp);
2939 #endif /* ! BNX2X_STOP_ON_ERROR */
2942 int bnx2x_drain_tx_queues(struct bnx2x *bp)
2946 /* Wait until tx fastpath tasks complete */
2947 for_each_tx_queue(bp, i) {
2948 struct bnx2x_fastpath *fp = &bp->fp[i];
2950 for_each_cos_in_tx_queue(fp, cos)
2951 rc = bnx2x_clean_tx_queue(bp, fp->txdata_ptr[cos]);
2958 /* must be called with rtnl_lock */
2959 int bnx2x_nic_unload(struct bnx2x *bp, int unload_mode, bool keep_link)
2962 bool global = false;
2964 DP(NETIF_MSG_IFUP, "Starting NIC unload\n");
2966 /* mark driver is unloaded in shmem2 */
2967 if (IS_PF(bp) && SHMEM2_HAS(bp, drv_capabilities_flag)) {
2969 val = SHMEM2_RD(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)]);
2970 SHMEM2_WR(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)],
2971 val & ~DRV_FLAGS_CAPABILITIES_LOADED_L2);
2974 if (IS_PF(bp) && bp->recovery_state != BNX2X_RECOVERY_DONE &&
2975 (bp->state == BNX2X_STATE_CLOSED ||
2976 bp->state == BNX2X_STATE_ERROR)) {
2977 /* We can get here if the driver has been unloaded
2978 * during parity error recovery and is either waiting for a
2979 * leader to complete or for other functions to unload and
2980 * then ifdown has been issued. In this case we want to
2981 * unload and let other functions to complete a recovery
2984 bp->recovery_state = BNX2X_RECOVERY_DONE;
2986 bnx2x_release_leader_lock(bp);
2989 DP(NETIF_MSG_IFDOWN, "Releasing a leadership...\n");
2990 BNX2X_ERR("Can't unload in closed or error state\n");
2994 /* Nothing to do during unload if previous bnx2x_nic_load()
2995 * have not completed successfully - all resources are released.
2997 * we can get here only after unsuccessful ndo_* callback, during which
2998 * dev->IFF_UP flag is still on.
3000 if (bp->state == BNX2X_STATE_CLOSED || bp->state == BNX2X_STATE_ERROR)
3003 /* It's important to set the bp->state to the value different from
3004 * BNX2X_STATE_OPEN and only then stop the Tx. Otherwise bnx2x_tx_int()
3005 * may restart the Tx from the NAPI context (see bnx2x_tx_int()).
3007 bp->state = BNX2X_STATE_CLOSING_WAIT4_HALT;
3010 /* indicate to VFs that the PF is going down */
3011 bnx2x_iov_channel_down(bp);
3013 if (CNIC_LOADED(bp))
3014 bnx2x_cnic_notify(bp, CNIC_CTL_STOP_CMD);
3017 bnx2x_tx_disable(bp);
3018 netdev_reset_tc(bp->dev);
3020 bp->rx_mode = BNX2X_RX_MODE_NONE;
3022 del_timer_sync(&bp->timer);
3025 /* Set ALWAYS_ALIVE bit in shmem */
3026 bp->fw_drv_pulse_wr_seq |= DRV_PULSE_ALWAYS_ALIVE;
3027 bnx2x_drv_pulse(bp);
3028 bnx2x_stats_handle(bp, STATS_EVENT_STOP);
3029 bnx2x_save_statistics(bp);
3032 /* wait till consumers catch up with producers in all queues */
3033 bnx2x_drain_tx_queues(bp);
3035 /* if VF indicate to PF this function is going down (PF will delete sp
3036 * elements and clear initializations
3039 bnx2x_vfpf_close_vf(bp);
3040 else if (unload_mode != UNLOAD_RECOVERY)
3041 /* if this is a normal/close unload need to clean up chip*/
3042 bnx2x_chip_cleanup(bp, unload_mode, keep_link);
3044 /* Send the UNLOAD_REQUEST to the MCP */
3045 bnx2x_send_unload_req(bp, unload_mode);
3047 /* Prevent transactions to host from the functions on the
3048 * engine that doesn't reset global blocks in case of global
3049 * attention once global blocks are reset and gates are opened
3050 * (the engine which leader will perform the recovery
3053 if (!CHIP_IS_E1x(bp))
3054 bnx2x_pf_disable(bp);
3056 /* Disable HW interrupts, NAPI */
3057 bnx2x_netif_stop(bp, 1);
3058 /* Delete all NAPI objects */
3059 bnx2x_del_all_napi(bp);
3060 if (CNIC_LOADED(bp))
3061 bnx2x_del_all_napi_cnic(bp);
3065 /* Report UNLOAD_DONE to MCP */
3066 bnx2x_send_unload_done(bp, false);
3070 * At this stage no more interrupts will arrive so we may safely clean
3071 * the queueable objects here in case they failed to get cleaned so far.
3074 bnx2x_squeeze_objects(bp);
3076 /* There should be no more pending SP commands at this stage */
3081 /* clear pending work in rtnl task */
3082 bp->sp_rtnl_state = 0;
3085 /* Free SKBs, SGEs, TPA pool and driver internals */
3086 bnx2x_free_skbs(bp);
3087 if (CNIC_LOADED(bp))
3088 bnx2x_free_skbs_cnic(bp);
3089 for_each_rx_queue(bp, i)
3090 bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
3092 bnx2x_free_fp_mem(bp);
3093 if (CNIC_LOADED(bp))
3094 bnx2x_free_fp_mem_cnic(bp);
3097 if (CNIC_LOADED(bp))
3098 bnx2x_free_mem_cnic(bp);
3102 bp->state = BNX2X_STATE_CLOSED;
3103 bp->cnic_loaded = false;
3105 /* Clear driver version indication in shmem */
3107 bnx2x_update_mng_version(bp);
3109 /* Check if there are pending parity attentions. If there are - set
3110 * RECOVERY_IN_PROGRESS.
3112 if (IS_PF(bp) && bnx2x_chk_parity_attn(bp, &global, false)) {
3113 bnx2x_set_reset_in_progress(bp);
3115 /* Set RESET_IS_GLOBAL if needed */
3117 bnx2x_set_reset_global(bp);
3120 /* The last driver must disable a "close the gate" if there is no
3121 * parity attention or "process kill" pending.
3124 !bnx2x_clear_pf_load(bp) &&
3125 bnx2x_reset_is_done(bp, BP_PATH(bp)))
3126 bnx2x_disable_close_the_gate(bp);
3128 DP(NETIF_MSG_IFUP, "Ending NIC unload\n");
3133 int bnx2x_set_power_state(struct bnx2x *bp, pci_power_t state)
3137 /* If there is no power capability, silently succeed */
3138 if (!bp->pdev->pm_cap) {
3139 BNX2X_DEV_INFO("No power capability. Breaking.\n");
3143 pci_read_config_word(bp->pdev, bp->pdev->pm_cap + PCI_PM_CTRL, &pmcsr);
3147 pci_write_config_word(bp->pdev, bp->pdev->pm_cap + PCI_PM_CTRL,
3148 ((pmcsr & ~PCI_PM_CTRL_STATE_MASK) |
3149 PCI_PM_CTRL_PME_STATUS));
3151 if (pmcsr & PCI_PM_CTRL_STATE_MASK)
3152 /* delay required during transition out of D3hot */
3157 /* If there are other clients above don't
3158 shut down the power */
3159 if (atomic_read(&bp->pdev->enable_cnt) != 1)
3161 /* Don't shut down the power for emulation and FPGA */
3162 if (CHIP_REV_IS_SLOW(bp))
3165 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
3169 pmcsr |= PCI_PM_CTRL_PME_ENABLE;
3171 pci_write_config_word(bp->pdev, bp->pdev->pm_cap + PCI_PM_CTRL,
3174 /* No more memory access after this point until
3175 * device is brought back to D0.
3180 dev_err(&bp->pdev->dev, "Can't support state = %d\n", state);
3187 * net_device service functions
3189 static int bnx2x_poll(struct napi_struct *napi, int budget)
3193 struct bnx2x_fastpath *fp = container_of(napi, struct bnx2x_fastpath,
3195 struct bnx2x *bp = fp->bp;
3198 #ifdef BNX2X_STOP_ON_ERROR
3199 if (unlikely(bp->panic)) {
3200 napi_complete(napi);
3204 if (!bnx2x_fp_lock_napi(fp))
3207 for_each_cos_in_tx_queue(fp, cos)
3208 if (bnx2x_tx_queue_has_work(fp->txdata_ptr[cos]))
3209 bnx2x_tx_int(bp, fp->txdata_ptr[cos]);
3211 if (bnx2x_has_rx_work(fp)) {
3212 work_done += bnx2x_rx_int(fp, budget - work_done);
3214 /* must not complete if we consumed full budget */
3215 if (work_done >= budget) {
3216 bnx2x_fp_unlock_napi(fp);
3221 bnx2x_fp_unlock_napi(fp);
3223 /* Fall out from the NAPI loop if needed */
3224 if (!(bnx2x_has_rx_work(fp) || bnx2x_has_tx_work(fp))) {
3226 /* No need to update SB for FCoE L2 ring as long as
3227 * it's connected to the default SB and the SB
3228 * has been updated when NAPI was scheduled.
3230 if (IS_FCOE_FP(fp)) {
3231 napi_complete(napi);
3234 bnx2x_update_fpsb_idx(fp);
3235 /* bnx2x_has_rx_work() reads the status block,
3236 * thus we need to ensure that status block indices
3237 * have been actually read (bnx2x_update_fpsb_idx)
3238 * prior to this check (bnx2x_has_rx_work) so that
3239 * we won't write the "newer" value of the status block
3240 * to IGU (if there was a DMA right after
3241 * bnx2x_has_rx_work and if there is no rmb, the memory
3242 * reading (bnx2x_update_fpsb_idx) may be postponed
3243 * to right before bnx2x_ack_sb). In this case there
3244 * will never be another interrupt until there is
3245 * another update of the status block, while there
3246 * is still unhandled work.
3250 if (!(bnx2x_has_rx_work(fp) || bnx2x_has_tx_work(fp))) {
3251 napi_complete(napi);
3252 /* Re-enable interrupts */
3253 DP(NETIF_MSG_RX_STATUS,
3254 "Update index to %d\n", fp->fp_hc_idx);
3255 bnx2x_ack_sb(bp, fp->igu_sb_id, USTORM_ID,
3256 le16_to_cpu(fp->fp_hc_idx),
3266 #ifdef CONFIG_NET_RX_BUSY_POLL
3267 /* must be called with local_bh_disable()d */
3268 int bnx2x_low_latency_recv(struct napi_struct *napi)
3270 struct bnx2x_fastpath *fp = container_of(napi, struct bnx2x_fastpath,
3272 struct bnx2x *bp = fp->bp;
3275 if ((bp->state == BNX2X_STATE_CLOSED) ||
3276 (bp->state == BNX2X_STATE_ERROR) ||
3277 (bp->dev->features & (NETIF_F_LRO | NETIF_F_GRO)))
3278 return LL_FLUSH_FAILED;
3280 if (!bnx2x_fp_lock_poll(fp))
3281 return LL_FLUSH_BUSY;
3283 if (bnx2x_has_rx_work(fp))
3284 found = bnx2x_rx_int(fp, 4);
3286 bnx2x_fp_unlock_poll(fp);
3292 /* we split the first BD into headers and data BDs
3293 * to ease the pain of our fellow microcode engineers
3294 * we use one mapping for both BDs
3296 static u16 bnx2x_tx_split(struct bnx2x *bp,
3297 struct bnx2x_fp_txdata *txdata,
3298 struct sw_tx_bd *tx_buf,
3299 struct eth_tx_start_bd **tx_bd, u16 hlen,
3302 struct eth_tx_start_bd *h_tx_bd = *tx_bd;
3303 struct eth_tx_bd *d_tx_bd;
3305 int old_len = le16_to_cpu(h_tx_bd->nbytes);
3307 /* first fix first BD */
3308 h_tx_bd->nbytes = cpu_to_le16(hlen);
3310 DP(NETIF_MSG_TX_QUEUED, "TSO split header size is %d (%x:%x)\n",
3311 h_tx_bd->nbytes, h_tx_bd->addr_hi, h_tx_bd->addr_lo);
3313 /* now get a new data BD
3314 * (after the pbd) and fill it */
3315 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3316 d_tx_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
3318 mapping = HILO_U64(le32_to_cpu(h_tx_bd->addr_hi),
3319 le32_to_cpu(h_tx_bd->addr_lo)) + hlen;
3321 d_tx_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
3322 d_tx_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
3323 d_tx_bd->nbytes = cpu_to_le16(old_len - hlen);
3325 /* this marks the BD as one that has no individual mapping */
3326 tx_buf->flags |= BNX2X_TSO_SPLIT_BD;
3328 DP(NETIF_MSG_TX_QUEUED,
3329 "TSO split data size is %d (%x:%x)\n",
3330 d_tx_bd->nbytes, d_tx_bd->addr_hi, d_tx_bd->addr_lo);
3333 *tx_bd = (struct eth_tx_start_bd *)d_tx_bd;
3338 #define bswab32(b32) ((__force __le32) swab32((__force __u32) (b32)))
3339 #define bswab16(b16) ((__force __le16) swab16((__force __u16) (b16)))
3340 static __le16 bnx2x_csum_fix(unsigned char *t_header, u16 csum, s8 fix)
3342 __sum16 tsum = (__force __sum16) csum;
3345 tsum = ~csum_fold(csum_sub((__force __wsum) csum,
3346 csum_partial(t_header - fix, fix, 0)));
3349 tsum = ~csum_fold(csum_add((__force __wsum) csum,
3350 csum_partial(t_header, -fix, 0)));
3352 return bswab16(tsum);
3355 static u32 bnx2x_xmit_type(struct bnx2x *bp, struct sk_buff *skb)
3361 if (skb->ip_summed != CHECKSUM_PARTIAL)
3364 protocol = vlan_get_protocol(skb);
3365 if (protocol == htons(ETH_P_IPV6)) {
3367 prot = ipv6_hdr(skb)->nexthdr;
3370 prot = ip_hdr(skb)->protocol;
3373 if (!CHIP_IS_E1x(bp) && skb->encapsulation) {
3374 if (inner_ip_hdr(skb)->version == 6) {
3375 rc |= XMIT_CSUM_ENC_V6;
3376 if (inner_ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3377 rc |= XMIT_CSUM_TCP;
3379 rc |= XMIT_CSUM_ENC_V4;
3380 if (inner_ip_hdr(skb)->protocol == IPPROTO_TCP)
3381 rc |= XMIT_CSUM_TCP;
3384 if (prot == IPPROTO_TCP)
3385 rc |= XMIT_CSUM_TCP;
3387 if (skb_is_gso(skb)) {
3388 if (skb_is_gso_v6(skb)) {
3389 rc |= (XMIT_GSO_V6 | XMIT_CSUM_TCP);
3390 if (rc & XMIT_CSUM_ENC)
3391 rc |= XMIT_GSO_ENC_V6;
3393 rc |= (XMIT_GSO_V4 | XMIT_CSUM_TCP);
3394 if (rc & XMIT_CSUM_ENC)
3395 rc |= XMIT_GSO_ENC_V4;
3402 #if (MAX_SKB_FRAGS >= MAX_FETCH_BD - 3)
3403 /* check if packet requires linearization (packet is too fragmented)
3404 no need to check fragmentation if page size > 8K (there will be no
3405 violation to FW restrictions) */
3406 static int bnx2x_pkt_req_lin(struct bnx2x *bp, struct sk_buff *skb,
3411 int first_bd_sz = 0;
3413 /* 3 = 1 (for linear data BD) + 2 (for PBD and last BD) */
3414 if (skb_shinfo(skb)->nr_frags >= (MAX_FETCH_BD - 3)) {
3416 if (xmit_type & XMIT_GSO) {
3417 unsigned short lso_mss = skb_shinfo(skb)->gso_size;
3418 /* Check if LSO packet needs to be copied:
3419 3 = 1 (for headers BD) + 2 (for PBD and last BD) */
3420 int wnd_size = MAX_FETCH_BD - 3;
3421 /* Number of windows to check */
3422 int num_wnds = skb_shinfo(skb)->nr_frags - wnd_size;
3427 /* Headers length */
3428 if (xmit_type & XMIT_GSO_ENC)
3429 hlen = (int)(skb_inner_transport_header(skb) -
3431 inner_tcp_hdrlen(skb);
3433 hlen = (int)(skb_transport_header(skb) -
3434 skb->data) + tcp_hdrlen(skb);
3436 /* Amount of data (w/o headers) on linear part of SKB*/
3437 first_bd_sz = skb_headlen(skb) - hlen;
3439 wnd_sum = first_bd_sz;
3441 /* Calculate the first sum - it's special */
3442 for (frag_idx = 0; frag_idx < wnd_size - 1; frag_idx++)
3444 skb_frag_size(&skb_shinfo(skb)->frags[frag_idx]);
3446 /* If there was data on linear skb data - check it */
3447 if (first_bd_sz > 0) {
3448 if (unlikely(wnd_sum < lso_mss)) {
3453 wnd_sum -= first_bd_sz;
3456 /* Others are easier: run through the frag list and
3457 check all windows */
3458 for (wnd_idx = 0; wnd_idx <= num_wnds; wnd_idx++) {
3460 skb_frag_size(&skb_shinfo(skb)->frags[wnd_idx + wnd_size - 1]);
3462 if (unlikely(wnd_sum < lso_mss)) {
3467 skb_frag_size(&skb_shinfo(skb)->frags[wnd_idx]);
3470 /* in non-LSO too fragmented packet should always
3477 if (unlikely(to_copy))
3478 DP(NETIF_MSG_TX_QUEUED,
3479 "Linearization IS REQUIRED for %s packet. num_frags %d hlen %d first_bd_sz %d\n",
3480 (xmit_type & XMIT_GSO) ? "LSO" : "non-LSO",
3481 skb_shinfo(skb)->nr_frags, hlen, first_bd_sz);
3488 * bnx2x_set_pbd_gso - update PBD in GSO case.
3492 * @xmit_type: xmit flags
3494 static void bnx2x_set_pbd_gso(struct sk_buff *skb,
3495 struct eth_tx_parse_bd_e1x *pbd,
3498 pbd->lso_mss = cpu_to_le16(skb_shinfo(skb)->gso_size);
3499 pbd->tcp_send_seq = bswab32(tcp_hdr(skb)->seq);
3500 pbd->tcp_flags = pbd_tcp_flags(tcp_hdr(skb));
3502 if (xmit_type & XMIT_GSO_V4) {
3503 pbd->ip_id = bswab16(ip_hdr(skb)->id);
3504 pbd->tcp_pseudo_csum =
3505 bswab16(~csum_tcpudp_magic(ip_hdr(skb)->saddr,
3507 0, IPPROTO_TCP, 0));
3509 pbd->tcp_pseudo_csum =
3510 bswab16(~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3511 &ipv6_hdr(skb)->daddr,
3512 0, IPPROTO_TCP, 0));
3516 cpu_to_le16(ETH_TX_PARSE_BD_E1X_PSEUDO_CS_WITHOUT_LEN);
3520 * bnx2x_set_pbd_csum_enc - update PBD with checksum and return header length
3522 * @bp: driver handle
3524 * @parsing_data: data to be updated
3525 * @xmit_type: xmit flags
3527 * 57712/578xx related, when skb has encapsulation
3529 static u8 bnx2x_set_pbd_csum_enc(struct bnx2x *bp, struct sk_buff *skb,
3530 u32 *parsing_data, u32 xmit_type)
3533 ((((u8 *)skb_inner_transport_header(skb) - skb->data) >> 1) <<
3534 ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W_SHIFT) &
3535 ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W;
3537 if (xmit_type & XMIT_CSUM_TCP) {
3538 *parsing_data |= ((inner_tcp_hdrlen(skb) / 4) <<
3539 ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW_SHIFT) &
3540 ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW;
3542 return skb_inner_transport_header(skb) +
3543 inner_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_inner_transport_header(skb) +
3550 sizeof(struct udphdr) - skb->data;
3554 * bnx2x_set_pbd_csum_e2 - update PBD with checksum and return header length
3556 * @bp: driver handle
3558 * @parsing_data: data to be updated
3559 * @xmit_type: xmit flags
3561 * 57712/578xx related
3563 static u8 bnx2x_set_pbd_csum_e2(struct bnx2x *bp, struct sk_buff *skb,
3564 u32 *parsing_data, u32 xmit_type)
3567 ((((u8 *)skb_transport_header(skb) - skb->data) >> 1) <<
3568 ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W_SHIFT) &
3569 ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W;
3571 if (xmit_type & XMIT_CSUM_TCP) {
3572 *parsing_data |= ((tcp_hdrlen(skb) / 4) <<
3573 ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW_SHIFT) &
3574 ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW;
3576 return skb_transport_header(skb) + tcp_hdrlen(skb) - skb->data;
3578 /* We support checksum offload for TCP and UDP only.
3579 * No need to pass the UDP header length - it's a constant.
3581 return skb_transport_header(skb) + sizeof(struct udphdr) - skb->data;
3584 /* set FW indication according to inner or outer protocols if tunneled */
3585 static void bnx2x_set_sbd_csum(struct bnx2x *bp, struct sk_buff *skb,
3586 struct eth_tx_start_bd *tx_start_bd,
3589 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_L4_CSUM;
3591 if (xmit_type & (XMIT_CSUM_ENC_V6 | XMIT_CSUM_V6))
3592 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_IPV6;
3594 if (!(xmit_type & XMIT_CSUM_TCP))
3595 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_IS_UDP;
3599 * bnx2x_set_pbd_csum - update PBD with checksum and return header length
3601 * @bp: driver handle
3603 * @pbd: parse BD to be updated
3604 * @xmit_type: xmit flags
3606 static u8 bnx2x_set_pbd_csum(struct bnx2x *bp, struct sk_buff *skb,
3607 struct eth_tx_parse_bd_e1x *pbd,
3610 u8 hlen = (skb_network_header(skb) - skb->data) >> 1;
3612 /* for now NS flag is not used in Linux */
3615 ((skb->protocol == cpu_to_be16(ETH_P_8021Q)) <<
3616 ETH_TX_PARSE_BD_E1X_LLC_SNAP_EN_SHIFT));
3618 pbd->ip_hlen_w = (skb_transport_header(skb) -
3619 skb_network_header(skb)) >> 1;
3621 hlen += pbd->ip_hlen_w;
3623 /* We support checksum offload for TCP and UDP only */
3624 if (xmit_type & XMIT_CSUM_TCP)
3625 hlen += tcp_hdrlen(skb) / 2;
3627 hlen += sizeof(struct udphdr) / 2;
3629 pbd->total_hlen_w = cpu_to_le16(hlen);
3632 if (xmit_type & XMIT_CSUM_TCP) {
3633 pbd->tcp_pseudo_csum = bswab16(tcp_hdr(skb)->check);
3636 s8 fix = SKB_CS_OFF(skb); /* signed! */
3638 DP(NETIF_MSG_TX_QUEUED,
3639 "hlen %d fix %d csum before fix %x\n",
3640 le16_to_cpu(pbd->total_hlen_w), fix, SKB_CS(skb));
3642 /* HW bug: fixup the CSUM */
3643 pbd->tcp_pseudo_csum =
3644 bnx2x_csum_fix(skb_transport_header(skb),
3647 DP(NETIF_MSG_TX_QUEUED, "csum after fix %x\n",
3648 pbd->tcp_pseudo_csum);
3654 static void bnx2x_update_pbds_gso_enc(struct sk_buff *skb,
3655 struct eth_tx_parse_bd_e2 *pbd_e2,
3656 struct eth_tx_parse_2nd_bd *pbd2,
3661 u8 outerip_off, outerip_len = 0;
3663 /* from outer IP to transport */
3664 hlen_w = (skb_inner_transport_header(skb) -
3665 skb_network_header(skb)) >> 1;
3668 hlen_w += inner_tcp_hdrlen(skb) >> 1;
3670 pbd2->fw_ip_hdr_to_payload_w = hlen_w;
3672 /* outer IP header info */
3673 if (xmit_type & XMIT_CSUM_V4) {
3674 struct iphdr *iph = ip_hdr(skb);
3675 u32 csum = (__force u32)(~iph->check) -
3676 (__force u32)iph->tot_len -
3677 (__force u32)iph->frag_off;
3679 outerip_len = iph->ihl << 1;
3681 pbd2->fw_ip_csum_wo_len_flags_frag =
3682 bswab16(csum_fold((__force __wsum)csum));
3684 pbd2->fw_ip_hdr_to_payload_w =
3685 hlen_w - ((sizeof(struct ipv6hdr)) >> 1);
3686 pbd_e2->data.tunnel_data.flags |=
3687 ETH_TUNNEL_DATA_IPV6_OUTER;
3690 pbd2->tcp_send_seq = bswab32(inner_tcp_hdr(skb)->seq);
3692 pbd2->tcp_flags = pbd_tcp_flags(inner_tcp_hdr(skb));
3694 /* inner IP header info */
3695 if (xmit_type & XMIT_CSUM_ENC_V4) {
3696 pbd2->hw_ip_id = bswab16(inner_ip_hdr(skb)->id);
3698 pbd_e2->data.tunnel_data.pseudo_csum =
3699 bswab16(~csum_tcpudp_magic(
3700 inner_ip_hdr(skb)->saddr,
3701 inner_ip_hdr(skb)->daddr,
3702 0, IPPROTO_TCP, 0));
3704 pbd_e2->data.tunnel_data.pseudo_csum =
3705 bswab16(~csum_ipv6_magic(
3706 &inner_ipv6_hdr(skb)->saddr,
3707 &inner_ipv6_hdr(skb)->daddr,
3708 0, IPPROTO_TCP, 0));
3711 outerip_off = (skb_network_header(skb) - skb->data) >> 1;
3716 ETH_TX_PARSE_2ND_BD_IP_HDR_LEN_OUTER_W_SHIFT) |
3717 ((skb->protocol == cpu_to_be16(ETH_P_8021Q)) <<
3718 ETH_TX_PARSE_2ND_BD_LLC_SNAP_EN_SHIFT);
3720 if (ip_hdr(skb)->protocol == IPPROTO_UDP) {
3721 SET_FLAG(*global_data, ETH_TX_PARSE_2ND_BD_TUNNEL_UDP_EXIST, 1);
3722 pbd2->tunnel_udp_hdr_start_w = skb_transport_offset(skb) >> 1;
3726 static inline void bnx2x_set_ipv6_ext_e2(struct sk_buff *skb, u32 *parsing_data,
3729 struct ipv6hdr *ipv6;
3731 if (!(xmit_type & (XMIT_GSO_ENC_V6 | XMIT_GSO_V6)))
3734 if (xmit_type & XMIT_GSO_ENC_V6)
3735 ipv6 = inner_ipv6_hdr(skb);
3736 else /* XMIT_GSO_V6 */
3737 ipv6 = ipv6_hdr(skb);
3739 if (ipv6->nexthdr == NEXTHDR_IPV6)
3740 *parsing_data |= ETH_TX_PARSE_BD_E2_IPV6_WITH_EXT_HDR;
3743 /* called with netif_tx_lock
3744 * bnx2x_tx_int() runs without netif_tx_lock unless it needs to call
3745 * netif_wake_queue()
3747 netdev_tx_t bnx2x_start_xmit(struct sk_buff *skb, struct net_device *dev)
3749 struct bnx2x *bp = netdev_priv(dev);
3751 struct netdev_queue *txq;
3752 struct bnx2x_fp_txdata *txdata;
3753 struct sw_tx_bd *tx_buf;
3754 struct eth_tx_start_bd *tx_start_bd, *first_bd;
3755 struct eth_tx_bd *tx_data_bd, *total_pkt_bd = NULL;
3756 struct eth_tx_parse_bd_e1x *pbd_e1x = NULL;
3757 struct eth_tx_parse_bd_e2 *pbd_e2 = NULL;
3758 struct eth_tx_parse_2nd_bd *pbd2 = NULL;
3759 u32 pbd_e2_parsing_data = 0;
3760 u16 pkt_prod, bd_prod;
3763 u32 xmit_type = bnx2x_xmit_type(bp, skb);
3766 __le16 pkt_size = 0;
3768 u8 mac_type = UNICAST_ADDRESS;
3770 #ifdef BNX2X_STOP_ON_ERROR
3771 if (unlikely(bp->panic))
3772 return NETDEV_TX_BUSY;
3775 txq_index = skb_get_queue_mapping(skb);
3776 txq = netdev_get_tx_queue(dev, txq_index);
3778 BUG_ON(txq_index >= MAX_ETH_TXQ_IDX(bp) + (CNIC_LOADED(bp) ? 1 : 0));
3780 txdata = &bp->bnx2x_txq[txq_index];
3782 /* enable this debug print to view the transmission queue being used
3783 DP(NETIF_MSG_TX_QUEUED, "indices: txq %d, fp %d, txdata %d\n",
3784 txq_index, fp_index, txdata_index); */
3786 /* enable this debug print to view the transmission details
3787 DP(NETIF_MSG_TX_QUEUED,
3788 "transmitting packet cid %d fp index %d txdata_index %d tx_data ptr %p fp pointer %p\n",
3789 txdata->cid, fp_index, txdata_index, txdata, fp); */
3791 if (unlikely(bnx2x_tx_avail(bp, txdata) <
3792 skb_shinfo(skb)->nr_frags +
3794 NEXT_CNT_PER_TX_PKT(MAX_BDS_PER_TX_PKT))) {
3795 /* Handle special storage cases separately */
3796 if (txdata->tx_ring_size == 0) {
3797 struct bnx2x_eth_q_stats *q_stats =
3798 bnx2x_fp_qstats(bp, txdata->parent_fp);
3799 q_stats->driver_filtered_tx_pkt++;
3801 return NETDEV_TX_OK;
3803 bnx2x_fp_qstats(bp, txdata->parent_fp)->driver_xoff++;
3804 netif_tx_stop_queue(txq);
3805 BNX2X_ERR("BUG! Tx ring full when queue awake!\n");
3807 return NETDEV_TX_BUSY;
3810 DP(NETIF_MSG_TX_QUEUED,
3811 "queue[%d]: SKB: summed %x protocol %x protocol(%x,%x) gso type %x xmit_type %x len %d\n",
3812 txq_index, skb->ip_summed, skb->protocol, ipv6_hdr(skb)->nexthdr,
3813 ip_hdr(skb)->protocol, skb_shinfo(skb)->gso_type, xmit_type,
3816 eth = (struct ethhdr *)skb->data;
3818 /* set flag according to packet type (UNICAST_ADDRESS is default)*/
3819 if (unlikely(is_multicast_ether_addr(eth->h_dest))) {
3820 if (is_broadcast_ether_addr(eth->h_dest))
3821 mac_type = BROADCAST_ADDRESS;
3823 mac_type = MULTICAST_ADDRESS;
3826 #if (MAX_SKB_FRAGS >= MAX_FETCH_BD - BDS_PER_TX_PKT)
3827 /* First, check if we need to linearize the skb (due to FW
3828 restrictions). No need to check fragmentation if page size > 8K
3829 (there will be no violation to FW restrictions) */
3830 if (bnx2x_pkt_req_lin(bp, skb, xmit_type)) {
3831 /* Statistics of linearization */
3833 if (skb_linearize(skb) != 0) {
3834 DP(NETIF_MSG_TX_QUEUED,
3835 "SKB linearization failed - silently dropping this SKB\n");
3836 dev_kfree_skb_any(skb);
3837 return NETDEV_TX_OK;
3841 /* Map skb linear data for DMA */
3842 mapping = dma_map_single(&bp->pdev->dev, skb->data,
3843 skb_headlen(skb), DMA_TO_DEVICE);
3844 if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
3845 DP(NETIF_MSG_TX_QUEUED,
3846 "SKB mapping failed - silently dropping this SKB\n");
3847 dev_kfree_skb_any(skb);
3848 return NETDEV_TX_OK;
3851 Please read carefully. First we use one BD which we mark as start,
3852 then we have a parsing info BD (used for TSO or xsum),
3853 and only then we have the rest of the TSO BDs.
3854 (don't forget to mark the last one as last,
3855 and to unmap only AFTER you write to the BD ...)
3856 And above all, all pdb sizes are in words - NOT DWORDS!
3859 /* get current pkt produced now - advance it just before sending packet
3860 * since mapping of pages may fail and cause packet to be dropped
3862 pkt_prod = txdata->tx_pkt_prod;
3863 bd_prod = TX_BD(txdata->tx_bd_prod);
3865 /* get a tx_buf and first BD
3866 * tx_start_bd may be changed during SPLIT,
3867 * but first_bd will always stay first
3869 tx_buf = &txdata->tx_buf_ring[TX_BD(pkt_prod)];
3870 tx_start_bd = &txdata->tx_desc_ring[bd_prod].start_bd;
3871 first_bd = tx_start_bd;
3873 tx_start_bd->bd_flags.as_bitfield = ETH_TX_BD_FLAGS_START_BD;
3875 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) {
3876 if (!(bp->flags & TX_TIMESTAMPING_EN)) {
3877 BNX2X_ERR("Tx timestamping was not enabled, this packet will not be timestamped\n");
3878 } else if (bp->ptp_tx_skb) {
3879 BNX2X_ERR("The device supports only a single outstanding packet to timestamp, this packet will not be timestamped\n");
3881 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
3882 /* schedule check for Tx timestamp */
3883 bp->ptp_tx_skb = skb_get(skb);
3884 bp->ptp_tx_start = jiffies;
3885 schedule_work(&bp->ptp_task);
3889 /* header nbd: indirectly zero other flags! */
3890 tx_start_bd->general_data = 1 << ETH_TX_START_BD_HDR_NBDS_SHIFT;
3892 /* remember the first BD of the packet */
3893 tx_buf->first_bd = txdata->tx_bd_prod;
3897 DP(NETIF_MSG_TX_QUEUED,
3898 "sending pkt %u @%p next_idx %u bd %u @%p\n",
3899 pkt_prod, tx_buf, txdata->tx_pkt_prod, bd_prod, tx_start_bd);
3901 if (skb_vlan_tag_present(skb)) {
3902 tx_start_bd->vlan_or_ethertype =
3903 cpu_to_le16(skb_vlan_tag_get(skb));
3904 tx_start_bd->bd_flags.as_bitfield |=
3905 (X_ETH_OUTBAND_VLAN << ETH_TX_BD_FLAGS_VLAN_MODE_SHIFT);
3907 /* when transmitting in a vf, start bd must hold the ethertype
3908 * for fw to enforce it
3910 #ifndef BNX2X_STOP_ON_ERROR
3913 tx_start_bd->vlan_or_ethertype =
3914 cpu_to_le16(ntohs(eth->h_proto));
3915 #ifndef BNX2X_STOP_ON_ERROR
3917 /* used by FW for packet accounting */
3918 tx_start_bd->vlan_or_ethertype = cpu_to_le16(pkt_prod);
3922 nbd = 2; /* start_bd + pbd + frags (updated when pages are mapped) */
3924 /* turn on parsing and get a BD */
3925 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3927 if (xmit_type & XMIT_CSUM)
3928 bnx2x_set_sbd_csum(bp, skb, tx_start_bd, xmit_type);
3930 if (!CHIP_IS_E1x(bp)) {
3931 pbd_e2 = &txdata->tx_desc_ring[bd_prod].parse_bd_e2;
3932 memset(pbd_e2, 0, sizeof(struct eth_tx_parse_bd_e2));
3934 if (xmit_type & XMIT_CSUM_ENC) {
3935 u16 global_data = 0;
3937 /* Set PBD in enc checksum offload case */
3938 hlen = bnx2x_set_pbd_csum_enc(bp, skb,
3939 &pbd_e2_parsing_data,
3942 /* turn on 2nd parsing and get a BD */
3943 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3945 pbd2 = &txdata->tx_desc_ring[bd_prod].parse_2nd_bd;
3947 memset(pbd2, 0, sizeof(*pbd2));
3949 pbd_e2->data.tunnel_data.ip_hdr_start_inner_w =
3950 (skb_inner_network_header(skb) -
3953 if (xmit_type & XMIT_GSO_ENC)
3954 bnx2x_update_pbds_gso_enc(skb, pbd_e2, pbd2,
3958 pbd2->global_data = cpu_to_le16(global_data);
3960 /* add addition parse BD indication to start BD */
3961 SET_FLAG(tx_start_bd->general_data,
3962 ETH_TX_START_BD_PARSE_NBDS, 1);
3963 /* set encapsulation flag in start BD */
3964 SET_FLAG(tx_start_bd->general_data,
3965 ETH_TX_START_BD_TUNNEL_EXIST, 1);
3967 tx_buf->flags |= BNX2X_HAS_SECOND_PBD;
3970 } else if (xmit_type & XMIT_CSUM) {
3971 /* Set PBD in checksum offload case w/o encapsulation */
3972 hlen = bnx2x_set_pbd_csum_e2(bp, skb,
3973 &pbd_e2_parsing_data,
3977 bnx2x_set_ipv6_ext_e2(skb, &pbd_e2_parsing_data, xmit_type);
3978 /* Add the macs to the parsing BD if this is a vf or if
3979 * Tx Switching is enabled.
3982 /* override GRE parameters in BD */
3983 bnx2x_set_fw_mac_addr(&pbd_e2->data.mac_addr.src_hi,
3984 &pbd_e2->data.mac_addr.src_mid,
3985 &pbd_e2->data.mac_addr.src_lo,
3988 bnx2x_set_fw_mac_addr(&pbd_e2->data.mac_addr.dst_hi,
3989 &pbd_e2->data.mac_addr.dst_mid,
3990 &pbd_e2->data.mac_addr.dst_lo,
3993 if (bp->flags & TX_SWITCHING)
3994 bnx2x_set_fw_mac_addr(
3995 &pbd_e2->data.mac_addr.dst_hi,
3996 &pbd_e2->data.mac_addr.dst_mid,
3997 &pbd_e2->data.mac_addr.dst_lo,
3999 #ifdef BNX2X_STOP_ON_ERROR
4000 /* Enforce security is always set in Stop on Error -
4001 * source mac should be present in the parsing BD
4003 bnx2x_set_fw_mac_addr(&pbd_e2->data.mac_addr.src_hi,
4004 &pbd_e2->data.mac_addr.src_mid,
4005 &pbd_e2->data.mac_addr.src_lo,
4010 SET_FLAG(pbd_e2_parsing_data,
4011 ETH_TX_PARSE_BD_E2_ETH_ADDR_TYPE, mac_type);
4013 u16 global_data = 0;
4014 pbd_e1x = &txdata->tx_desc_ring[bd_prod].parse_bd_e1x;
4015 memset(pbd_e1x, 0, sizeof(struct eth_tx_parse_bd_e1x));
4016 /* Set PBD in checksum offload case */
4017 if (xmit_type & XMIT_CSUM)
4018 hlen = bnx2x_set_pbd_csum(bp, skb, pbd_e1x, xmit_type);
4020 SET_FLAG(global_data,
4021 ETH_TX_PARSE_BD_E1X_ETH_ADDR_TYPE, mac_type);
4022 pbd_e1x->global_data |= cpu_to_le16(global_data);
4025 /* Setup the data pointer of the first BD of the packet */
4026 tx_start_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
4027 tx_start_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
4028 tx_start_bd->nbytes = cpu_to_le16(skb_headlen(skb));
4029 pkt_size = tx_start_bd->nbytes;
4031 DP(NETIF_MSG_TX_QUEUED,
4032 "first bd @%p addr (%x:%x) nbytes %d flags %x vlan %x\n",
4033 tx_start_bd, tx_start_bd->addr_hi, tx_start_bd->addr_lo,
4034 le16_to_cpu(tx_start_bd->nbytes),
4035 tx_start_bd->bd_flags.as_bitfield,
4036 le16_to_cpu(tx_start_bd->vlan_or_ethertype));
4038 if (xmit_type & XMIT_GSO) {
4040 DP(NETIF_MSG_TX_QUEUED,
4041 "TSO packet len %d hlen %d total len %d tso size %d\n",
4042 skb->len, hlen, skb_headlen(skb),
4043 skb_shinfo(skb)->gso_size);
4045 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_SW_LSO;
4047 if (unlikely(skb_headlen(skb) > hlen)) {
4049 bd_prod = bnx2x_tx_split(bp, txdata, tx_buf,
4053 if (!CHIP_IS_E1x(bp))
4054 pbd_e2_parsing_data |=
4055 (skb_shinfo(skb)->gso_size <<
4056 ETH_TX_PARSE_BD_E2_LSO_MSS_SHIFT) &
4057 ETH_TX_PARSE_BD_E2_LSO_MSS;
4059 bnx2x_set_pbd_gso(skb, pbd_e1x, xmit_type);
4062 /* Set the PBD's parsing_data field if not zero
4063 * (for the chips newer than 57711).
4065 if (pbd_e2_parsing_data)
4066 pbd_e2->parsing_data = cpu_to_le32(pbd_e2_parsing_data);
4068 tx_data_bd = (struct eth_tx_bd *)tx_start_bd;
4070 /* Handle fragmented skb */
4071 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
4072 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
4074 mapping = skb_frag_dma_map(&bp->pdev->dev, frag, 0,
4075 skb_frag_size(frag), DMA_TO_DEVICE);
4076 if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
4077 unsigned int pkts_compl = 0, bytes_compl = 0;
4079 DP(NETIF_MSG_TX_QUEUED,
4080 "Unable to map page - dropping packet...\n");
4082 /* we need unmap all buffers already mapped
4084 * first_bd->nbd need to be properly updated
4085 * before call to bnx2x_free_tx_pkt
4087 first_bd->nbd = cpu_to_le16(nbd);
4088 bnx2x_free_tx_pkt(bp, txdata,
4089 TX_BD(txdata->tx_pkt_prod),
4090 &pkts_compl, &bytes_compl);
4091 return NETDEV_TX_OK;
4094 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
4095 tx_data_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
4096 if (total_pkt_bd == NULL)
4097 total_pkt_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
4099 tx_data_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
4100 tx_data_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
4101 tx_data_bd->nbytes = cpu_to_le16(skb_frag_size(frag));
4102 le16_add_cpu(&pkt_size, skb_frag_size(frag));
4105 DP(NETIF_MSG_TX_QUEUED,
4106 "frag %d bd @%p addr (%x:%x) nbytes %d\n",
4107 i, tx_data_bd, tx_data_bd->addr_hi, tx_data_bd->addr_lo,
4108 le16_to_cpu(tx_data_bd->nbytes));
4111 DP(NETIF_MSG_TX_QUEUED, "last bd @%p\n", tx_data_bd);
4113 /* update with actual num BDs */
4114 first_bd->nbd = cpu_to_le16(nbd);
4116 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
4118 /* now send a tx doorbell, counting the next BD
4119 * if the packet contains or ends with it
4121 if (TX_BD_POFF(bd_prod) < nbd)
4124 /* total_pkt_bytes should be set on the first data BD if
4125 * it's not an LSO packet and there is more than one
4126 * data BD. In this case pkt_size is limited by an MTU value.
4127 * However we prefer to set it for an LSO packet (while we don't
4128 * have to) in order to save some CPU cycles in a none-LSO
4129 * case, when we much more care about them.
4131 if (total_pkt_bd != NULL)
4132 total_pkt_bd->total_pkt_bytes = pkt_size;
4135 DP(NETIF_MSG_TX_QUEUED,
4136 "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",
4137 pbd_e1x, pbd_e1x->global_data, pbd_e1x->ip_hlen_w,
4138 pbd_e1x->ip_id, pbd_e1x->lso_mss, pbd_e1x->tcp_flags,
4139 pbd_e1x->tcp_pseudo_csum, pbd_e1x->tcp_send_seq,
4140 le16_to_cpu(pbd_e1x->total_hlen_w));
4142 DP(NETIF_MSG_TX_QUEUED,
4143 "PBD (E2) @%p dst %x %x %x src %x %x %x parsing_data %x\n",
4145 pbd_e2->data.mac_addr.dst_hi,
4146 pbd_e2->data.mac_addr.dst_mid,
4147 pbd_e2->data.mac_addr.dst_lo,
4148 pbd_e2->data.mac_addr.src_hi,
4149 pbd_e2->data.mac_addr.src_mid,
4150 pbd_e2->data.mac_addr.src_lo,
4151 pbd_e2->parsing_data);
4152 DP(NETIF_MSG_TX_QUEUED, "doorbell: nbd %d bd %u\n", nbd, bd_prod);
4154 netdev_tx_sent_queue(txq, skb->len);
4156 skb_tx_timestamp(skb);
4158 txdata->tx_pkt_prod++;
4160 * Make sure that the BD data is updated before updating the producer
4161 * since FW might read the BD right after the producer is updated.
4162 * This is only applicable for weak-ordered memory model archs such
4163 * as IA-64. The following barrier is also mandatory since FW will
4164 * assumes packets must have BDs.
4168 txdata->tx_db.data.prod += nbd;
4171 DOORBELL(bp, txdata->cid, txdata->tx_db.raw);
4175 txdata->tx_bd_prod += nbd;
4177 if (unlikely(bnx2x_tx_avail(bp, txdata) < MAX_DESC_PER_TX_PKT)) {
4178 netif_tx_stop_queue(txq);
4180 /* paired memory barrier is in bnx2x_tx_int(), we have to keep
4181 * ordering of set_bit() in netif_tx_stop_queue() and read of
4185 bnx2x_fp_qstats(bp, txdata->parent_fp)->driver_xoff++;
4186 if (bnx2x_tx_avail(bp, txdata) >= MAX_DESC_PER_TX_PKT)
4187 netif_tx_wake_queue(txq);
4191 return NETDEV_TX_OK;
4195 * bnx2x_setup_tc - routine to configure net_device for multi tc
4197 * @netdev: net device to configure
4198 * @tc: number of traffic classes to enable
4200 * callback connected to the ndo_setup_tc function pointer
4202 int bnx2x_setup_tc(struct net_device *dev, u8 num_tc)
4204 int cos, prio, count, offset;
4205 struct bnx2x *bp = netdev_priv(dev);
4207 /* setup tc must be called under rtnl lock */
4210 /* no traffic classes requested. Aborting */
4212 netdev_reset_tc(dev);
4216 /* requested to support too many traffic classes */
4217 if (num_tc > bp->max_cos) {
4218 BNX2X_ERR("support for too many traffic classes requested: %d. Max supported is %d\n",
4219 num_tc, bp->max_cos);
4223 /* declare amount of supported traffic classes */
4224 if (netdev_set_num_tc(dev, num_tc)) {
4225 BNX2X_ERR("failed to declare %d traffic classes\n", num_tc);
4229 /* configure priority to traffic class mapping */
4230 for (prio = 0; prio < BNX2X_MAX_PRIORITY; prio++) {
4231 netdev_set_prio_tc_map(dev, prio, bp->prio_to_cos[prio]);
4232 DP(BNX2X_MSG_SP | NETIF_MSG_IFUP,
4233 "mapping priority %d to tc %d\n",
4234 prio, bp->prio_to_cos[prio]);
4237 /* Use this configuration to differentiate tc0 from other COSes
4238 This can be used for ets or pfc, and save the effort of setting
4239 up a multio class queue disc or negotiating DCBX with a switch
4240 netdev_set_prio_tc_map(dev, 0, 0);
4241 DP(BNX2X_MSG_SP, "mapping priority %d to tc %d\n", 0, 0);
4242 for (prio = 1; prio < 16; prio++) {
4243 netdev_set_prio_tc_map(dev, prio, 1);
4244 DP(BNX2X_MSG_SP, "mapping priority %d to tc %d\n", prio, 1);
4247 /* configure traffic class to transmission queue mapping */
4248 for (cos = 0; cos < bp->max_cos; cos++) {
4249 count = BNX2X_NUM_ETH_QUEUES(bp);
4250 offset = cos * BNX2X_NUM_NON_CNIC_QUEUES(bp);
4251 netdev_set_tc_queue(dev, cos, count, offset);
4252 DP(BNX2X_MSG_SP | NETIF_MSG_IFUP,
4253 "mapping tc %d to offset %d count %d\n",
4254 cos, offset, count);
4260 /* called with rtnl_lock */
4261 int bnx2x_change_mac_addr(struct net_device *dev, void *p)
4263 struct sockaddr *addr = p;
4264 struct bnx2x *bp = netdev_priv(dev);
4267 if (!is_valid_ether_addr(addr->sa_data)) {
4268 BNX2X_ERR("Requested MAC address is not valid\n");
4272 if (IS_MF_STORAGE_ONLY(bp)) {
4273 BNX2X_ERR("Can't change address on STORAGE ONLY function\n");
4277 if (netif_running(dev)) {
4278 rc = bnx2x_set_eth_mac(bp, false);
4283 memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
4285 if (netif_running(dev))
4286 rc = bnx2x_set_eth_mac(bp, true);
4291 static void bnx2x_free_fp_mem_at(struct bnx2x *bp, int fp_index)
4293 union host_hc_status_block *sb = &bnx2x_fp(bp, fp_index, status_blk);
4294 struct bnx2x_fastpath *fp = &bp->fp[fp_index];
4299 if (IS_FCOE_IDX(fp_index)) {
4300 memset(sb, 0, sizeof(union host_hc_status_block));
4301 fp->status_blk_mapping = 0;
4304 if (!CHIP_IS_E1x(bp))
4305 BNX2X_PCI_FREE(sb->e2_sb,
4306 bnx2x_fp(bp, fp_index,
4307 status_blk_mapping),
4308 sizeof(struct host_hc_status_block_e2));
4310 BNX2X_PCI_FREE(sb->e1x_sb,
4311 bnx2x_fp(bp, fp_index,
4312 status_blk_mapping),
4313 sizeof(struct host_hc_status_block_e1x));
4317 if (!skip_rx_queue(bp, fp_index)) {
4318 bnx2x_free_rx_bds(fp);
4320 /* fastpath rx rings: rx_buf rx_desc rx_comp */
4321 BNX2X_FREE(bnx2x_fp(bp, fp_index, rx_buf_ring));
4322 BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_desc_ring),
4323 bnx2x_fp(bp, fp_index, rx_desc_mapping),
4324 sizeof(struct eth_rx_bd) * NUM_RX_BD);
4326 BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_comp_ring),
4327 bnx2x_fp(bp, fp_index, rx_comp_mapping),
4328 sizeof(struct eth_fast_path_rx_cqe) *
4332 BNX2X_FREE(bnx2x_fp(bp, fp_index, rx_page_ring));
4333 BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_sge_ring),
4334 bnx2x_fp(bp, fp_index, rx_sge_mapping),
4335 BCM_PAGE_SIZE * NUM_RX_SGE_PAGES);
4339 if (!skip_tx_queue(bp, fp_index)) {
4340 /* fastpath tx rings: tx_buf tx_desc */
4341 for_each_cos_in_tx_queue(fp, cos) {
4342 struct bnx2x_fp_txdata *txdata = fp->txdata_ptr[cos];
4344 DP(NETIF_MSG_IFDOWN,
4345 "freeing tx memory of fp %d cos %d cid %d\n",
4346 fp_index, cos, txdata->cid);
4348 BNX2X_FREE(txdata->tx_buf_ring);
4349 BNX2X_PCI_FREE(txdata->tx_desc_ring,
4350 txdata->tx_desc_mapping,
4351 sizeof(union eth_tx_bd_types) * NUM_TX_BD);
4354 /* end of fastpath */
4357 static void bnx2x_free_fp_mem_cnic(struct bnx2x *bp)
4360 for_each_cnic_queue(bp, i)
4361 bnx2x_free_fp_mem_at(bp, i);
4364 void bnx2x_free_fp_mem(struct bnx2x *bp)
4367 for_each_eth_queue(bp, i)
4368 bnx2x_free_fp_mem_at(bp, i);
4371 static void set_sb_shortcuts(struct bnx2x *bp, int index)
4373 union host_hc_status_block status_blk = bnx2x_fp(bp, index, status_blk);
4374 if (!CHIP_IS_E1x(bp)) {
4375 bnx2x_fp(bp, index, sb_index_values) =
4376 (__le16 *)status_blk.e2_sb->sb.index_values;
4377 bnx2x_fp(bp, index, sb_running_index) =
4378 (__le16 *)status_blk.e2_sb->sb.running_index;
4380 bnx2x_fp(bp, index, sb_index_values) =
4381 (__le16 *)status_blk.e1x_sb->sb.index_values;
4382 bnx2x_fp(bp, index, sb_running_index) =
4383 (__le16 *)status_blk.e1x_sb->sb.running_index;
4387 /* Returns the number of actually allocated BDs */
4388 static int bnx2x_alloc_rx_bds(struct bnx2x_fastpath *fp,
4391 struct bnx2x *bp = fp->bp;
4392 u16 ring_prod, cqe_ring_prod;
4393 int i, failure_cnt = 0;
4395 fp->rx_comp_cons = 0;
4396 cqe_ring_prod = ring_prod = 0;
4398 /* This routine is called only during fo init so
4399 * fp->eth_q_stats.rx_skb_alloc_failed = 0
4401 for (i = 0; i < rx_ring_size; i++) {
4402 if (bnx2x_alloc_rx_data(bp, fp, ring_prod, GFP_KERNEL) < 0) {
4406 ring_prod = NEXT_RX_IDX(ring_prod);
4407 cqe_ring_prod = NEXT_RCQ_IDX(cqe_ring_prod);
4408 WARN_ON(ring_prod <= (i - failure_cnt));
4412 BNX2X_ERR("was only able to allocate %d rx skbs on queue[%d]\n",
4413 i - failure_cnt, fp->index);
4415 fp->rx_bd_prod = ring_prod;
4416 /* Limit the CQE producer by the CQE ring size */
4417 fp->rx_comp_prod = min_t(u16, NUM_RCQ_RINGS*RCQ_DESC_CNT,
4419 fp->rx_pkt = fp->rx_calls = 0;
4421 bnx2x_fp_stats(bp, fp)->eth_q_stats.rx_skb_alloc_failed += failure_cnt;
4423 return i - failure_cnt;
4426 static void bnx2x_set_next_page_rx_cq(struct bnx2x_fastpath *fp)
4430 for (i = 1; i <= NUM_RCQ_RINGS; i++) {
4431 struct eth_rx_cqe_next_page *nextpg;
4433 nextpg = (struct eth_rx_cqe_next_page *)
4434 &fp->rx_comp_ring[RCQ_DESC_CNT * i - 1];
4436 cpu_to_le32(U64_HI(fp->rx_comp_mapping +
4437 BCM_PAGE_SIZE*(i % NUM_RCQ_RINGS)));
4439 cpu_to_le32(U64_LO(fp->rx_comp_mapping +
4440 BCM_PAGE_SIZE*(i % NUM_RCQ_RINGS)));
4444 static int bnx2x_alloc_fp_mem_at(struct bnx2x *bp, int index)
4446 union host_hc_status_block *sb;
4447 struct bnx2x_fastpath *fp = &bp->fp[index];
4450 int rx_ring_size = 0;
4452 if (!bp->rx_ring_size && IS_MF_STORAGE_ONLY(bp)) {
4453 rx_ring_size = MIN_RX_SIZE_NONTPA;
4454 bp->rx_ring_size = rx_ring_size;
4455 } else if (!bp->rx_ring_size) {
4456 rx_ring_size = MAX_RX_AVAIL/BNX2X_NUM_RX_QUEUES(bp);
4458 if (CHIP_IS_E3(bp)) {
4459 u32 cfg = SHMEM_RD(bp,
4460 dev_info.port_hw_config[BP_PORT(bp)].
4463 /* Decrease ring size for 1G functions */
4464 if ((cfg & PORT_HW_CFG_NET_SERDES_IF_MASK) ==
4465 PORT_HW_CFG_NET_SERDES_IF_SGMII)
4469 /* allocate at least number of buffers required by FW */
4470 rx_ring_size = max_t(int, bp->disable_tpa ? MIN_RX_SIZE_NONTPA :
4471 MIN_RX_SIZE_TPA, rx_ring_size);
4473 bp->rx_ring_size = rx_ring_size;
4474 } else /* if rx_ring_size specified - use it */
4475 rx_ring_size = bp->rx_ring_size;
4477 DP(BNX2X_MSG_SP, "calculated rx_ring_size %d\n", rx_ring_size);
4480 sb = &bnx2x_fp(bp, index, status_blk);
4482 if (!IS_FCOE_IDX(index)) {
4484 if (!CHIP_IS_E1x(bp)) {
4485 sb->e2_sb = BNX2X_PCI_ALLOC(&bnx2x_fp(bp, index, status_blk_mapping),
4486 sizeof(struct host_hc_status_block_e2));
4490 sb->e1x_sb = BNX2X_PCI_ALLOC(&bnx2x_fp(bp, index, status_blk_mapping),
4491 sizeof(struct host_hc_status_block_e1x));
4497 /* FCoE Queue uses Default SB and doesn't ACK the SB, thus no need to
4498 * set shortcuts for it.
4500 if (!IS_FCOE_IDX(index))
4501 set_sb_shortcuts(bp, index);
4504 if (!skip_tx_queue(bp, index)) {
4505 /* fastpath tx rings: tx_buf tx_desc */
4506 for_each_cos_in_tx_queue(fp, cos) {
4507 struct bnx2x_fp_txdata *txdata = fp->txdata_ptr[cos];
4510 "allocating tx memory of fp %d cos %d\n",
4513 txdata->tx_buf_ring = kcalloc(NUM_TX_BD,
4514 sizeof(struct sw_tx_bd),
4516 if (!txdata->tx_buf_ring)
4518 txdata->tx_desc_ring = BNX2X_PCI_ALLOC(&txdata->tx_desc_mapping,
4519 sizeof(union eth_tx_bd_types) * NUM_TX_BD);
4520 if (!txdata->tx_desc_ring)
4526 if (!skip_rx_queue(bp, index)) {
4527 /* fastpath rx rings: rx_buf rx_desc rx_comp */
4528 bnx2x_fp(bp, index, rx_buf_ring) =
4529 kcalloc(NUM_RX_BD, sizeof(struct sw_rx_bd), GFP_KERNEL);
4530 if (!bnx2x_fp(bp, index, rx_buf_ring))
4532 bnx2x_fp(bp, index, rx_desc_ring) =
4533 BNX2X_PCI_ALLOC(&bnx2x_fp(bp, index, rx_desc_mapping),
4534 sizeof(struct eth_rx_bd) * NUM_RX_BD);
4535 if (!bnx2x_fp(bp, index, rx_desc_ring))
4538 /* Seed all CQEs by 1s */
4539 bnx2x_fp(bp, index, rx_comp_ring) =
4540 BNX2X_PCI_FALLOC(&bnx2x_fp(bp, index, rx_comp_mapping),
4541 sizeof(struct eth_fast_path_rx_cqe) * NUM_RCQ_BD);
4542 if (!bnx2x_fp(bp, index, rx_comp_ring))
4546 bnx2x_fp(bp, index, rx_page_ring) =
4547 kcalloc(NUM_RX_SGE, sizeof(struct sw_rx_page),
4549 if (!bnx2x_fp(bp, index, rx_page_ring))
4551 bnx2x_fp(bp, index, rx_sge_ring) =
4552 BNX2X_PCI_ALLOC(&bnx2x_fp(bp, index, rx_sge_mapping),
4553 BCM_PAGE_SIZE * NUM_RX_SGE_PAGES);
4554 if (!bnx2x_fp(bp, index, rx_sge_ring))
4557 bnx2x_set_next_page_rx_bd(fp);
4560 bnx2x_set_next_page_rx_cq(fp);
4563 ring_size = bnx2x_alloc_rx_bds(fp, rx_ring_size);
4564 if (ring_size < rx_ring_size)
4570 /* handles low memory cases */
4572 BNX2X_ERR("Unable to allocate full memory for queue %d (size %d)\n",
4574 /* FW will drop all packets if queue is not big enough,
4575 * In these cases we disable the queue
4576 * Min size is different for OOO, TPA and non-TPA queues
4578 if (ring_size < (fp->mode == TPA_MODE_DISABLED ?
4579 MIN_RX_SIZE_NONTPA : MIN_RX_SIZE_TPA)) {
4580 /* release memory allocated for this queue */
4581 bnx2x_free_fp_mem_at(bp, index);
4587 static int bnx2x_alloc_fp_mem_cnic(struct bnx2x *bp)
4591 if (bnx2x_alloc_fp_mem_at(bp, FCOE_IDX(bp)))
4592 /* we will fail load process instead of mark
4600 static int bnx2x_alloc_fp_mem(struct bnx2x *bp)
4604 /* 1. Allocate FP for leading - fatal if error
4605 * 2. Allocate RSS - fix number of queues if error
4609 if (bnx2x_alloc_fp_mem_at(bp, 0))
4613 for_each_nondefault_eth_queue(bp, i)
4614 if (bnx2x_alloc_fp_mem_at(bp, i))
4617 /* handle memory failures */
4618 if (i != BNX2X_NUM_ETH_QUEUES(bp)) {
4619 int delta = BNX2X_NUM_ETH_QUEUES(bp) - i;
4622 bnx2x_shrink_eth_fp(bp, delta);
4623 if (CNIC_SUPPORT(bp))
4624 /* move non eth FPs next to last eth FP
4625 * must be done in that order
4626 * FCOE_IDX < FWD_IDX < OOO_IDX
4629 /* move FCoE fp even NO_FCOE_FLAG is on */
4630 bnx2x_move_fp(bp, FCOE_IDX(bp), FCOE_IDX(bp) - delta);
4631 bp->num_ethernet_queues -= delta;
4632 bp->num_queues = bp->num_ethernet_queues +
4633 bp->num_cnic_queues;
4634 BNX2X_ERR("Adjusted num of queues from %d to %d\n",
4635 bp->num_queues + delta, bp->num_queues);
4641 void bnx2x_free_mem_bp(struct bnx2x *bp)
4645 for (i = 0; i < bp->fp_array_size; i++)
4646 kfree(bp->fp[i].tpa_info);
4649 kfree(bp->fp_stats);
4650 kfree(bp->bnx2x_txq);
4651 kfree(bp->msix_table);
4655 int bnx2x_alloc_mem_bp(struct bnx2x *bp)
4657 struct bnx2x_fastpath *fp;
4658 struct msix_entry *tbl;
4659 struct bnx2x_ilt *ilt;
4660 int msix_table_size = 0;
4661 int fp_array_size, txq_array_size;
4665 * The biggest MSI-X table we might need is as a maximum number of fast
4666 * path IGU SBs plus default SB (for PF only).
4668 msix_table_size = bp->igu_sb_cnt;
4671 BNX2X_DEV_INFO("msix_table_size %d\n", msix_table_size);
4673 /* fp array: RSS plus CNIC related L2 queues */
4674 fp_array_size = BNX2X_MAX_RSS_COUNT(bp) + CNIC_SUPPORT(bp);
4675 bp->fp_array_size = fp_array_size;
4676 BNX2X_DEV_INFO("fp_array_size %d\n", bp->fp_array_size);
4678 fp = kcalloc(bp->fp_array_size, sizeof(*fp), GFP_KERNEL);
4681 for (i = 0; i < bp->fp_array_size; i++) {
4683 kcalloc(ETH_MAX_AGGREGATION_QUEUES_E1H_E2,
4684 sizeof(struct bnx2x_agg_info), GFP_KERNEL);
4685 if (!(fp[i].tpa_info))
4691 /* allocate sp objs */
4692 bp->sp_objs = kcalloc(bp->fp_array_size, sizeof(struct bnx2x_sp_objs),
4697 /* allocate fp_stats */
4698 bp->fp_stats = kcalloc(bp->fp_array_size, sizeof(struct bnx2x_fp_stats),
4703 /* Allocate memory for the transmission queues array */
4705 BNX2X_MAX_RSS_COUNT(bp) * BNX2X_MULTI_TX_COS + CNIC_SUPPORT(bp);
4706 BNX2X_DEV_INFO("txq_array_size %d", txq_array_size);
4708 bp->bnx2x_txq = kcalloc(txq_array_size, sizeof(struct bnx2x_fp_txdata),
4714 tbl = kcalloc(msix_table_size, sizeof(*tbl), GFP_KERNEL);
4717 bp->msix_table = tbl;
4720 ilt = kzalloc(sizeof(*ilt), GFP_KERNEL);
4727 bnx2x_free_mem_bp(bp);
4731 int bnx2x_reload_if_running(struct net_device *dev)
4733 struct bnx2x *bp = netdev_priv(dev);
4735 if (unlikely(!netif_running(dev)))
4738 bnx2x_nic_unload(bp, UNLOAD_NORMAL, true);
4739 return bnx2x_nic_load(bp, LOAD_NORMAL);
4742 int bnx2x_get_cur_phy_idx(struct bnx2x *bp)
4744 u32 sel_phy_idx = 0;
4745 if (bp->link_params.num_phys <= 1)
4748 if (bp->link_vars.link_up) {
4749 sel_phy_idx = EXT_PHY1;
4750 /* In case link is SERDES, check if the EXT_PHY2 is the one */
4751 if ((bp->link_vars.link_status & LINK_STATUS_SERDES_LINK) &&
4752 (bp->link_params.phy[EXT_PHY2].supported & SUPPORTED_FIBRE))
4753 sel_phy_idx = EXT_PHY2;
4756 switch (bnx2x_phy_selection(&bp->link_params)) {
4757 case PORT_HW_CFG_PHY_SELECTION_HARDWARE_DEFAULT:
4758 case PORT_HW_CFG_PHY_SELECTION_FIRST_PHY:
4759 case PORT_HW_CFG_PHY_SELECTION_FIRST_PHY_PRIORITY:
4760 sel_phy_idx = EXT_PHY1;
4762 case PORT_HW_CFG_PHY_SELECTION_SECOND_PHY:
4763 case PORT_HW_CFG_PHY_SELECTION_SECOND_PHY_PRIORITY:
4764 sel_phy_idx = EXT_PHY2;
4771 int bnx2x_get_link_cfg_idx(struct bnx2x *bp)
4773 u32 sel_phy_idx = bnx2x_get_cur_phy_idx(bp);
4775 * The selected activated PHY is always after swapping (in case PHY
4776 * swapping is enabled). So when swapping is enabled, we need to reverse
4780 if (bp->link_params.multi_phy_config &
4781 PORT_HW_CFG_PHY_SWAPPED_ENABLED) {
4782 if (sel_phy_idx == EXT_PHY1)
4783 sel_phy_idx = EXT_PHY2;
4784 else if (sel_phy_idx == EXT_PHY2)
4785 sel_phy_idx = EXT_PHY1;
4787 return LINK_CONFIG_IDX(sel_phy_idx);
4790 #ifdef NETDEV_FCOE_WWNN
4791 int bnx2x_fcoe_get_wwn(struct net_device *dev, u64 *wwn, int type)
4793 struct bnx2x *bp = netdev_priv(dev);
4794 struct cnic_eth_dev *cp = &bp->cnic_eth_dev;
4797 case NETDEV_FCOE_WWNN:
4798 *wwn = HILO_U64(cp->fcoe_wwn_node_name_hi,
4799 cp->fcoe_wwn_node_name_lo);
4801 case NETDEV_FCOE_WWPN:
4802 *wwn = HILO_U64(cp->fcoe_wwn_port_name_hi,
4803 cp->fcoe_wwn_port_name_lo);
4806 BNX2X_ERR("Wrong WWN type requested - %d\n", type);
4814 /* called with rtnl_lock */
4815 int bnx2x_change_mtu(struct net_device *dev, int new_mtu)
4817 struct bnx2x *bp = netdev_priv(dev);
4819 if (pci_num_vf(bp->pdev)) {
4820 DP(BNX2X_MSG_IOV, "VFs are enabled, can not change MTU\n");
4824 if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
4825 BNX2X_ERR("Can't perform change MTU during parity recovery\n");
4829 if ((new_mtu > ETH_MAX_JUMBO_PACKET_SIZE) ||
4830 ((new_mtu + ETH_HLEN) < ETH_MIN_PACKET_SIZE)) {
4831 BNX2X_ERR("Can't support requested MTU size\n");
4835 /* This does not race with packet allocation
4836 * because the actual alloc size is
4837 * only updated as part of load
4841 return bnx2x_reload_if_running(dev);
4844 netdev_features_t bnx2x_fix_features(struct net_device *dev,
4845 netdev_features_t features)
4847 struct bnx2x *bp = netdev_priv(dev);
4849 if (pci_num_vf(bp->pdev)) {
4850 netdev_features_t changed = dev->features ^ features;
4852 /* Revert the requested changes in features if they
4853 * would require internal reload of PF in bnx2x_set_features().
4855 if (!(features & NETIF_F_RXCSUM) && !bp->disable_tpa) {
4856 features &= ~NETIF_F_RXCSUM;
4857 features |= dev->features & NETIF_F_RXCSUM;
4860 if (changed & NETIF_F_LOOPBACK) {
4861 features &= ~NETIF_F_LOOPBACK;
4862 features |= dev->features & NETIF_F_LOOPBACK;
4866 /* TPA requires Rx CSUM offloading */
4867 if (!(features & NETIF_F_RXCSUM)) {
4868 features &= ~NETIF_F_LRO;
4869 features &= ~NETIF_F_GRO;
4875 int bnx2x_set_features(struct net_device *dev, netdev_features_t features)
4877 struct bnx2x *bp = netdev_priv(dev);
4878 netdev_features_t changes = features ^ dev->features;
4879 bool bnx2x_reload = false;
4882 /* VFs or non SRIOV PFs should be able to change loopback feature */
4883 if (!pci_num_vf(bp->pdev)) {
4884 if (features & NETIF_F_LOOPBACK) {
4885 if (bp->link_params.loopback_mode != LOOPBACK_BMAC) {
4886 bp->link_params.loopback_mode = LOOPBACK_BMAC;
4887 bnx2x_reload = true;
4890 if (bp->link_params.loopback_mode != LOOPBACK_NONE) {
4891 bp->link_params.loopback_mode = LOOPBACK_NONE;
4892 bnx2x_reload = true;
4897 /* if GRO is changed while LRO is enabled, don't force a reload */
4898 if ((changes & NETIF_F_GRO) && (features & NETIF_F_LRO))
4899 changes &= ~NETIF_F_GRO;
4901 /* if GRO is changed while HW TPA is off, don't force a reload */
4902 if ((changes & NETIF_F_GRO) && bp->disable_tpa)
4903 changes &= ~NETIF_F_GRO;
4906 bnx2x_reload = true;
4909 if (bp->recovery_state == BNX2X_RECOVERY_DONE) {
4910 dev->features = features;
4911 rc = bnx2x_reload_if_running(dev);
4914 /* else: bnx2x_nic_load() will be called at end of recovery */
4920 void bnx2x_tx_timeout(struct net_device *dev)
4922 struct bnx2x *bp = netdev_priv(dev);
4924 #ifdef BNX2X_STOP_ON_ERROR
4929 /* This allows the netif to be shutdown gracefully before resetting */
4930 bnx2x_schedule_sp_rtnl(bp, BNX2X_SP_RTNL_TX_TIMEOUT, 0);
4933 int bnx2x_suspend(struct pci_dev *pdev, pm_message_t state)
4935 struct net_device *dev = pci_get_drvdata(pdev);
4939 dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
4942 bp = netdev_priv(dev);
4946 pci_save_state(pdev);
4948 if (!netif_running(dev)) {
4953 netif_device_detach(dev);
4955 bnx2x_nic_unload(bp, UNLOAD_CLOSE, false);
4957 bnx2x_set_power_state(bp, pci_choose_state(pdev, state));
4964 int bnx2x_resume(struct pci_dev *pdev)
4966 struct net_device *dev = pci_get_drvdata(pdev);
4971 dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
4974 bp = netdev_priv(dev);
4976 if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
4977 BNX2X_ERR("Handling parity error recovery. Try again later\n");
4983 pci_restore_state(pdev);
4985 if (!netif_running(dev)) {
4990 bnx2x_set_power_state(bp, PCI_D0);
4991 netif_device_attach(dev);
4993 rc = bnx2x_nic_load(bp, LOAD_OPEN);
5000 void bnx2x_set_ctx_validation(struct bnx2x *bp, struct eth_context *cxt,
5004 BNX2X_ERR("bad context pointer %p\n", cxt);
5008 /* ustorm cxt validation */
5009 cxt->ustorm_ag_context.cdu_usage =
5010 CDU_RSRVD_VALUE_TYPE_A(HW_CID(bp, cid),
5011 CDU_REGION_NUMBER_UCM_AG, ETH_CONNECTION_TYPE);
5012 /* xcontext validation */
5013 cxt->xstorm_ag_context.cdu_reserved =
5014 CDU_RSRVD_VALUE_TYPE_A(HW_CID(bp, cid),
5015 CDU_REGION_NUMBER_XCM_AG, ETH_CONNECTION_TYPE);
5018 static void storm_memset_hc_timeout(struct bnx2x *bp, u8 port,
5019 u8 fw_sb_id, u8 sb_index,
5022 u32 addr = BAR_CSTRORM_INTMEM +
5023 CSTORM_STATUS_BLOCK_DATA_TIMEOUT_OFFSET(fw_sb_id, sb_index);
5024 REG_WR8(bp, addr, ticks);
5026 "port %x fw_sb_id %d sb_index %d ticks %d\n",
5027 port, fw_sb_id, sb_index, ticks);
5030 static void storm_memset_hc_disable(struct bnx2x *bp, u8 port,
5031 u16 fw_sb_id, u8 sb_index,
5034 u32 enable_flag = disable ? 0 : (1 << HC_INDEX_DATA_HC_ENABLED_SHIFT);
5035 u32 addr = BAR_CSTRORM_INTMEM +
5036 CSTORM_STATUS_BLOCK_DATA_FLAGS_OFFSET(fw_sb_id, sb_index);
5037 u8 flags = REG_RD8(bp, addr);
5039 flags &= ~HC_INDEX_DATA_HC_ENABLED;
5040 flags |= enable_flag;
5041 REG_WR8(bp, addr, flags);
5043 "port %x fw_sb_id %d sb_index %d disable %d\n",
5044 port, fw_sb_id, sb_index, disable);
5047 void bnx2x_update_coalesce_sb_index(struct bnx2x *bp, u8 fw_sb_id,
5048 u8 sb_index, u8 disable, u16 usec)
5050 int port = BP_PORT(bp);
5051 u8 ticks = usec / BNX2X_BTR;
5053 storm_memset_hc_timeout(bp, port, fw_sb_id, sb_index, ticks);
5055 disable = disable ? 1 : (usec ? 0 : 1);
5056 storm_memset_hc_disable(bp, port, fw_sb_id, sb_index, disable);
5059 void bnx2x_schedule_sp_rtnl(struct bnx2x *bp, enum sp_rtnl_flag flag,
5062 smp_mb__before_atomic();
5063 set_bit(flag, &bp->sp_rtnl_state);
5064 smp_mb__after_atomic();
5065 DP((BNX2X_MSG_SP | verbose), "Scheduling sp_rtnl task [Flag: %d]\n",
5067 schedule_delayed_work(&bp->sp_rtnl_task, 0);
5069 EXPORT_SYMBOL(bnx2x_schedule_sp_rtnl);