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);
52 static void bnx2x_add_all_napi(struct bnx2x *bp)
56 /* Add NAPI objects */
57 for_each_eth_queue(bp, i) {
58 netif_napi_add(bp->dev, &bnx2x_fp(bp, i, napi),
59 bnx2x_poll, NAPI_POLL_WEIGHT);
63 static int bnx2x_calc_num_queues(struct bnx2x *bp)
65 int nq = bnx2x_num_queues ? : netif_get_num_default_rss_queues();
67 /* Reduce memory usage in kdump environment by using only one queue */
68 if (is_kdump_kernel())
71 nq = clamp(nq, 1, BNX2X_MAX_QUEUES(bp));
76 * bnx2x_move_fp - move content of the fastpath structure.
79 * @from: source FP index
80 * @to: destination FP index
82 * Makes sure the contents of the bp->fp[to].napi is kept
83 * intact. This is done by first copying the napi struct from
84 * the target to the source, and then mem copying the entire
85 * source onto the target. Update txdata pointers and related
88 static inline void bnx2x_move_fp(struct bnx2x *bp, int from, int to)
90 struct bnx2x_fastpath *from_fp = &bp->fp[from];
91 struct bnx2x_fastpath *to_fp = &bp->fp[to];
92 struct bnx2x_sp_objs *from_sp_objs = &bp->sp_objs[from];
93 struct bnx2x_sp_objs *to_sp_objs = &bp->sp_objs[to];
94 struct bnx2x_fp_stats *from_fp_stats = &bp->fp_stats[from];
95 struct bnx2x_fp_stats *to_fp_stats = &bp->fp_stats[to];
96 int old_max_eth_txqs, new_max_eth_txqs;
97 int old_txdata_index = 0, new_txdata_index = 0;
98 struct bnx2x_agg_info *old_tpa_info = to_fp->tpa_info;
100 /* Copy the NAPI object as it has been already initialized */
101 from_fp->napi = to_fp->napi;
103 /* Move bnx2x_fastpath contents */
104 memcpy(to_fp, from_fp, sizeof(*to_fp));
107 /* Retain the tpa_info of the original `to' version as we don't want
108 * 2 FPs to contain the same tpa_info pointer.
110 to_fp->tpa_info = old_tpa_info;
112 /* move sp_objs contents as well, as their indices match fp ones */
113 memcpy(to_sp_objs, from_sp_objs, sizeof(*to_sp_objs));
115 /* move fp_stats contents as well, as their indices match fp ones */
116 memcpy(to_fp_stats, from_fp_stats, sizeof(*to_fp_stats));
118 /* Update txdata pointers in fp and move txdata content accordingly:
119 * Each fp consumes 'max_cos' txdata structures, so the index should be
120 * decremented by max_cos x delta.
123 old_max_eth_txqs = BNX2X_NUM_ETH_QUEUES(bp) * (bp)->max_cos;
124 new_max_eth_txqs = (BNX2X_NUM_ETH_QUEUES(bp) - from + to) *
126 if (from == FCOE_IDX(bp)) {
127 old_txdata_index = old_max_eth_txqs + FCOE_TXQ_IDX_OFFSET;
128 new_txdata_index = new_max_eth_txqs + FCOE_TXQ_IDX_OFFSET;
131 memcpy(&bp->bnx2x_txq[new_txdata_index],
132 &bp->bnx2x_txq[old_txdata_index],
133 sizeof(struct bnx2x_fp_txdata));
134 to_fp->txdata_ptr[0] = &bp->bnx2x_txq[new_txdata_index];
138 * bnx2x_fill_fw_str - Fill buffer with FW version string.
141 * @buf: character buffer to fill with the fw name
142 * @buf_len: length of the above buffer
145 void bnx2x_fill_fw_str(struct bnx2x *bp, char *buf, size_t buf_len)
148 u8 phy_fw_ver[PHY_FW_VER_LEN];
150 phy_fw_ver[0] = '\0';
151 bnx2x_get_ext_phy_fw_version(&bp->link_params,
152 phy_fw_ver, PHY_FW_VER_LEN);
153 strlcpy(buf, bp->fw_ver, buf_len);
154 snprintf(buf + strlen(bp->fw_ver), 32 - strlen(bp->fw_ver),
156 (bp->common.bc_ver & 0xff0000) >> 16,
157 (bp->common.bc_ver & 0xff00) >> 8,
158 (bp->common.bc_ver & 0xff),
159 ((phy_fw_ver[0] != '\0') ? " phy " : ""), phy_fw_ver);
161 bnx2x_vf_fill_fw_str(bp, buf, buf_len);
166 * bnx2x_shrink_eth_fp - guarantees fastpath structures stay intact
169 * @delta: number of eth queues which were not allocated
171 static void bnx2x_shrink_eth_fp(struct bnx2x *bp, int delta)
173 int i, cos, old_eth_num = BNX2X_NUM_ETH_QUEUES(bp);
175 /* Queue pointer cannot be re-set on an fp-basis, as moving pointer
176 * backward along the array could cause memory to be overridden
178 for (cos = 1; cos < bp->max_cos; cos++) {
179 for (i = 0; i < old_eth_num - delta; i++) {
180 struct bnx2x_fastpath *fp = &bp->fp[i];
181 int new_idx = cos * (old_eth_num - delta) + i;
183 memcpy(&bp->bnx2x_txq[new_idx], fp->txdata_ptr[cos],
184 sizeof(struct bnx2x_fp_txdata));
185 fp->txdata_ptr[cos] = &bp->bnx2x_txq[new_idx];
190 int bnx2x_load_count[2][3] = { {0} }; /* per-path: 0-common, 1-port0, 2-port1 */
192 /* free skb in the packet ring at pos idx
193 * return idx of last bd freed
195 static u16 bnx2x_free_tx_pkt(struct bnx2x *bp, struct bnx2x_fp_txdata *txdata,
196 u16 idx, unsigned int *pkts_compl,
197 unsigned int *bytes_compl)
199 struct sw_tx_bd *tx_buf = &txdata->tx_buf_ring[idx];
200 struct eth_tx_start_bd *tx_start_bd;
201 struct eth_tx_bd *tx_data_bd;
202 struct sk_buff *skb = tx_buf->skb;
203 u16 bd_idx = TX_BD(tx_buf->first_bd), new_cons;
205 u16 split_bd_len = 0;
207 /* prefetch skb end pointer to speedup dev_kfree_skb() */
210 DP(NETIF_MSG_TX_DONE, "fp[%d]: pkt_idx %d buff @(%p)->skb %p\n",
211 txdata->txq_index, idx, tx_buf, skb);
213 tx_start_bd = &txdata->tx_desc_ring[bd_idx].start_bd;
215 nbd = le16_to_cpu(tx_start_bd->nbd) - 1;
216 #ifdef BNX2X_STOP_ON_ERROR
217 if ((nbd - 1) > (MAX_SKB_FRAGS + 2)) {
218 BNX2X_ERR("BAD nbd!\n");
222 new_cons = nbd + tx_buf->first_bd;
224 /* Get the next bd */
225 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
227 /* Skip a parse bd... */
229 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
231 if (tx_buf->flags & BNX2X_HAS_SECOND_PBD) {
232 /* Skip second parse bd... */
234 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
237 /* TSO headers+data bds share a common mapping. See bnx2x_tx_split() */
238 if (tx_buf->flags & BNX2X_TSO_SPLIT_BD) {
239 tx_data_bd = &txdata->tx_desc_ring[bd_idx].reg_bd;
240 split_bd_len = BD_UNMAP_LEN(tx_data_bd);
242 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
246 dma_unmap_single(&bp->pdev->dev, BD_UNMAP_ADDR(tx_start_bd),
247 BD_UNMAP_LEN(tx_start_bd) + split_bd_len,
253 tx_data_bd = &txdata->tx_desc_ring[bd_idx].reg_bd;
254 dma_unmap_page(&bp->pdev->dev, BD_UNMAP_ADDR(tx_data_bd),
255 BD_UNMAP_LEN(tx_data_bd), DMA_TO_DEVICE);
257 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
264 (*bytes_compl) += skb->len;
265 dev_kfree_skb_any(skb);
268 tx_buf->first_bd = 0;
274 int bnx2x_tx_int(struct bnx2x *bp, struct bnx2x_fp_txdata *txdata)
276 struct netdev_queue *txq;
277 u16 hw_cons, sw_cons, bd_cons = txdata->tx_bd_cons;
278 unsigned int pkts_compl = 0, bytes_compl = 0;
280 #ifdef BNX2X_STOP_ON_ERROR
281 if (unlikely(bp->panic))
285 txq = netdev_get_tx_queue(bp->dev, txdata->txq_index);
286 hw_cons = le16_to_cpu(*txdata->tx_cons_sb);
287 sw_cons = txdata->tx_pkt_cons;
289 while (sw_cons != hw_cons) {
292 pkt_cons = TX_BD(sw_cons);
294 DP(NETIF_MSG_TX_DONE,
295 "queue[%d]: hw_cons %u sw_cons %u pkt_cons %u\n",
296 txdata->txq_index, hw_cons, sw_cons, pkt_cons);
298 bd_cons = bnx2x_free_tx_pkt(bp, txdata, pkt_cons,
299 &pkts_compl, &bytes_compl);
304 netdev_tx_completed_queue(txq, pkts_compl, bytes_compl);
306 txdata->tx_pkt_cons = sw_cons;
307 txdata->tx_bd_cons = bd_cons;
309 /* Need to make the tx_bd_cons update visible to start_xmit()
310 * before checking for netif_tx_queue_stopped(). Without the
311 * memory barrier, there is a small possibility that
312 * start_xmit() will miss it and cause the queue to be stopped
314 * On the other hand we need an rmb() here to ensure the proper
315 * ordering of bit testing in the following
316 * netif_tx_queue_stopped(txq) call.
320 if (unlikely(netif_tx_queue_stopped(txq))) {
321 /* Taking tx_lock() is needed to prevent re-enabling the queue
322 * while it's empty. This could have happen if rx_action() gets
323 * suspended in bnx2x_tx_int() after the condition before
324 * netif_tx_wake_queue(), while tx_action (bnx2x_start_xmit()):
326 * stops the queue->sees fresh tx_bd_cons->releases the queue->
327 * sends some packets consuming the whole queue again->
331 __netif_tx_lock(txq, smp_processor_id());
333 if ((netif_tx_queue_stopped(txq)) &&
334 (bp->state == BNX2X_STATE_OPEN) &&
335 (bnx2x_tx_avail(bp, txdata) >= MAX_DESC_PER_TX_PKT))
336 netif_tx_wake_queue(txq);
338 __netif_tx_unlock(txq);
343 static inline void bnx2x_update_last_max_sge(struct bnx2x_fastpath *fp,
346 u16 last_max = fp->last_max_sge;
348 if (SUB_S16(idx, last_max) > 0)
349 fp->last_max_sge = idx;
352 static inline void bnx2x_update_sge_prod(struct bnx2x_fastpath *fp,
354 struct eth_end_agg_rx_cqe *cqe)
356 struct bnx2x *bp = fp->bp;
357 u16 last_max, last_elem, first_elem;
364 /* First mark all used pages */
365 for (i = 0; i < sge_len; i++)
366 BIT_VEC64_CLEAR_BIT(fp->sge_mask,
367 RX_SGE(le16_to_cpu(cqe->sgl_or_raw_data.sgl[i])));
369 DP(NETIF_MSG_RX_STATUS, "fp_cqe->sgl[%d] = %d\n",
370 sge_len - 1, le16_to_cpu(cqe->sgl_or_raw_data.sgl[sge_len - 1]));
372 /* Here we assume that the last SGE index is the biggest */
373 prefetch((void *)(fp->sge_mask));
374 bnx2x_update_last_max_sge(fp,
375 le16_to_cpu(cqe->sgl_or_raw_data.sgl[sge_len - 1]));
377 last_max = RX_SGE(fp->last_max_sge);
378 last_elem = last_max >> BIT_VEC64_ELEM_SHIFT;
379 first_elem = RX_SGE(fp->rx_sge_prod) >> BIT_VEC64_ELEM_SHIFT;
381 /* If ring is not full */
382 if (last_elem + 1 != first_elem)
385 /* Now update the prod */
386 for (i = first_elem; i != last_elem; i = NEXT_SGE_MASK_ELEM(i)) {
387 if (likely(fp->sge_mask[i]))
390 fp->sge_mask[i] = BIT_VEC64_ELEM_ONE_MASK;
391 delta += BIT_VEC64_ELEM_SZ;
395 fp->rx_sge_prod += delta;
396 /* clear page-end entries */
397 bnx2x_clear_sge_mask_next_elems(fp);
400 DP(NETIF_MSG_RX_STATUS,
401 "fp->last_max_sge = %d fp->rx_sge_prod = %d\n",
402 fp->last_max_sge, fp->rx_sge_prod);
405 /* Get Toeplitz hash value in the skb using the value from the
406 * CQE (calculated by HW).
408 static u32 bnx2x_get_rxhash(const struct bnx2x *bp,
409 const struct eth_fast_path_rx_cqe *cqe,
410 enum pkt_hash_types *rxhash_type)
412 /* Get Toeplitz hash from CQE */
413 if ((bp->dev->features & NETIF_F_RXHASH) &&
414 (cqe->status_flags & ETH_FAST_PATH_RX_CQE_RSS_HASH_FLG)) {
415 enum eth_rss_hash_type htype;
417 htype = cqe->status_flags & ETH_FAST_PATH_RX_CQE_RSS_HASH_TYPE;
418 *rxhash_type = ((htype == TCP_IPV4_HASH_TYPE) ||
419 (htype == TCP_IPV6_HASH_TYPE)) ?
420 PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3;
422 return le32_to_cpu(cqe->rss_hash_result);
424 *rxhash_type = PKT_HASH_TYPE_NONE;
428 static void bnx2x_tpa_start(struct bnx2x_fastpath *fp, u16 queue,
430 struct eth_fast_path_rx_cqe *cqe)
432 struct bnx2x *bp = fp->bp;
433 struct sw_rx_bd *cons_rx_buf = &fp->rx_buf_ring[cons];
434 struct sw_rx_bd *prod_rx_buf = &fp->rx_buf_ring[prod];
435 struct eth_rx_bd *prod_bd = &fp->rx_desc_ring[prod];
437 struct bnx2x_agg_info *tpa_info = &fp->tpa_info[queue];
438 struct sw_rx_bd *first_buf = &tpa_info->first_buf;
440 /* print error if current state != stop */
441 if (tpa_info->tpa_state != BNX2X_TPA_STOP)
442 BNX2X_ERR("start of bin not in stop [%d]\n", queue);
444 /* Try to map an empty data buffer from the aggregation info */
445 mapping = dma_map_single(&bp->pdev->dev,
446 first_buf->data + NET_SKB_PAD,
447 fp->rx_buf_size, DMA_FROM_DEVICE);
449 * ...if it fails - move the skb from the consumer to the producer
450 * and set the current aggregation state as ERROR to drop it
451 * when TPA_STOP arrives.
454 if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
455 /* Move the BD from the consumer to the producer */
456 bnx2x_reuse_rx_data(fp, cons, prod);
457 tpa_info->tpa_state = BNX2X_TPA_ERROR;
461 /* move empty data from pool to prod */
462 prod_rx_buf->data = first_buf->data;
463 dma_unmap_addr_set(prod_rx_buf, mapping, mapping);
464 /* point prod_bd to new data */
465 prod_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
466 prod_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
468 /* move partial skb from cons to pool (don't unmap yet) */
469 *first_buf = *cons_rx_buf;
471 /* mark bin state as START */
472 tpa_info->parsing_flags =
473 le16_to_cpu(cqe->pars_flags.flags);
474 tpa_info->vlan_tag = le16_to_cpu(cqe->vlan_tag);
475 tpa_info->tpa_state = BNX2X_TPA_START;
476 tpa_info->len_on_bd = le16_to_cpu(cqe->len_on_bd);
477 tpa_info->placement_offset = cqe->placement_offset;
478 tpa_info->rxhash = bnx2x_get_rxhash(bp, cqe, &tpa_info->rxhash_type);
479 if (fp->mode == TPA_MODE_GRO) {
480 u16 gro_size = le16_to_cpu(cqe->pkt_len_or_gro_seg_len);
481 tpa_info->full_page = SGE_PAGES / gro_size * gro_size;
482 tpa_info->gro_size = gro_size;
485 #ifdef BNX2X_STOP_ON_ERROR
486 fp->tpa_queue_used |= (1 << queue);
487 DP(NETIF_MSG_RX_STATUS, "fp->tpa_queue_used = 0x%llx\n",
492 /* Timestamp option length allowed for TPA aggregation:
494 * nop nop kind length echo val
496 #define TPA_TSTAMP_OPT_LEN 12
498 * bnx2x_set_gro_params - compute GRO values
501 * @parsing_flags: parsing flags from the START CQE
502 * @len_on_bd: total length of the first packet for the
504 * @pkt_len: length of all segments
506 * Approximate value of the MSS for this aggregation calculated using
507 * the first packet of it.
508 * Compute number of aggregated segments, and gso_type.
510 static void bnx2x_set_gro_params(struct sk_buff *skb, u16 parsing_flags,
511 u16 len_on_bd, unsigned int pkt_len,
512 u16 num_of_coalesced_segs)
514 /* TPA aggregation won't have either IP options or TCP options
515 * other than timestamp or IPv6 extension headers.
517 u16 hdrs_len = ETH_HLEN + sizeof(struct tcphdr);
519 if (GET_FLAG(parsing_flags, PARSING_FLAGS_OVER_ETHERNET_PROTOCOL) ==
520 PRS_FLAG_OVERETH_IPV6) {
521 hdrs_len += sizeof(struct ipv6hdr);
522 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
524 hdrs_len += sizeof(struct iphdr);
525 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
528 /* Check if there was a TCP timestamp, if there is it's will
529 * always be 12 bytes length: nop nop kind length echo val.
531 * Otherwise FW would close the aggregation.
533 if (parsing_flags & PARSING_FLAGS_TIME_STAMP_EXIST_FLAG)
534 hdrs_len += TPA_TSTAMP_OPT_LEN;
536 skb_shinfo(skb)->gso_size = len_on_bd - hdrs_len;
538 /* tcp_gro_complete() will copy NAPI_GRO_CB(skb)->count
539 * to skb_shinfo(skb)->gso_segs
541 NAPI_GRO_CB(skb)->count = num_of_coalesced_segs;
544 static int bnx2x_alloc_rx_sge(struct bnx2x *bp, struct bnx2x_fastpath *fp,
545 u16 index, gfp_t gfp_mask)
547 struct sw_rx_page *sw_buf = &fp->rx_page_ring[index];
548 struct eth_rx_sge *sge = &fp->rx_sge_ring[index];
549 struct bnx2x_alloc_pool *pool = &fp->page_pool;
552 if (!pool->page || (PAGE_SIZE - pool->offset) < SGE_PAGE_SIZE) {
554 /* put page reference used by the memory pool, since we
555 * won't be using this page as the mempool anymore.
558 put_page(pool->page);
560 pool->page = alloc_pages(gfp_mask, PAGES_PER_SGE_SHIFT);
561 if (unlikely(!pool->page))
567 mapping = dma_map_page(&bp->pdev->dev, pool->page,
568 pool->offset, SGE_PAGE_SIZE, DMA_FROM_DEVICE);
569 if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
570 BNX2X_ERR("Can't map sge\n");
574 get_page(pool->page);
575 sw_buf->page = pool->page;
576 sw_buf->offset = pool->offset;
578 dma_unmap_addr_set(sw_buf, mapping, mapping);
580 sge->addr_hi = cpu_to_le32(U64_HI(mapping));
581 sge->addr_lo = cpu_to_le32(U64_LO(mapping));
583 pool->offset += SGE_PAGE_SIZE;
588 static int bnx2x_fill_frag_skb(struct bnx2x *bp, struct bnx2x_fastpath *fp,
589 struct bnx2x_agg_info *tpa_info,
592 struct eth_end_agg_rx_cqe *cqe,
595 struct sw_rx_page *rx_pg, old_rx_pg;
596 u32 i, frag_len, frag_size;
597 int err, j, frag_id = 0;
598 u16 len_on_bd = tpa_info->len_on_bd;
599 u16 full_page = 0, gro_size = 0;
601 frag_size = le16_to_cpu(cqe->pkt_len) - len_on_bd;
603 if (fp->mode == TPA_MODE_GRO) {
604 gro_size = tpa_info->gro_size;
605 full_page = tpa_info->full_page;
608 /* This is needed in order to enable forwarding support */
610 bnx2x_set_gro_params(skb, tpa_info->parsing_flags, len_on_bd,
611 le16_to_cpu(cqe->pkt_len),
612 le16_to_cpu(cqe->num_of_coalesced_segs));
614 #ifdef BNX2X_STOP_ON_ERROR
615 if (pages > min_t(u32, 8, MAX_SKB_FRAGS) * SGE_PAGES) {
616 BNX2X_ERR("SGL length is too long: %d. CQE index is %d\n",
618 BNX2X_ERR("cqe->pkt_len = %d\n", cqe->pkt_len);
624 /* Run through the SGL and compose the fragmented skb */
625 for (i = 0, j = 0; i < pages; i += PAGES_PER_SGE, j++) {
626 u16 sge_idx = RX_SGE(le16_to_cpu(cqe->sgl_or_raw_data.sgl[j]));
628 /* FW gives the indices of the SGE as if the ring is an array
629 (meaning that "next" element will consume 2 indices) */
630 if (fp->mode == TPA_MODE_GRO)
631 frag_len = min_t(u32, frag_size, (u32)full_page);
633 frag_len = min_t(u32, frag_size, (u32)SGE_PAGES);
635 rx_pg = &fp->rx_page_ring[sge_idx];
638 /* If we fail to allocate a substitute page, we simply stop
639 where we are and drop the whole packet */
640 err = bnx2x_alloc_rx_sge(bp, fp, sge_idx, GFP_ATOMIC);
642 bnx2x_fp_qstats(bp, fp)->rx_skb_alloc_failed++;
646 dma_unmap_page(&bp->pdev->dev,
647 dma_unmap_addr(&old_rx_pg, mapping),
648 SGE_PAGE_SIZE, DMA_FROM_DEVICE);
649 /* Add one frag and update the appropriate fields in the skb */
650 if (fp->mode == TPA_MODE_LRO)
651 skb_fill_page_desc(skb, j, old_rx_pg.page,
652 old_rx_pg.offset, frag_len);
656 for (rem = frag_len; rem > 0; rem -= gro_size) {
657 int len = rem > gro_size ? gro_size : rem;
658 skb_fill_page_desc(skb, frag_id++,
660 old_rx_pg.offset + offset,
663 get_page(old_rx_pg.page);
668 skb->data_len += frag_len;
669 skb->truesize += SGE_PAGES;
670 skb->len += frag_len;
672 frag_size -= frag_len;
678 static void bnx2x_frag_free(const struct bnx2x_fastpath *fp, void *data)
680 if (fp->rx_frag_size)
686 static void *bnx2x_frag_alloc(const struct bnx2x_fastpath *fp, gfp_t gfp_mask)
688 if (fp->rx_frag_size) {
689 /* GFP_KERNEL allocations are used only during initialization */
690 if (unlikely(gfpflags_allow_blocking(gfp_mask)))
691 return (void *)__get_free_page(gfp_mask);
693 return netdev_alloc_frag(fp->rx_frag_size);
696 return kmalloc(fp->rx_buf_size + NET_SKB_PAD, gfp_mask);
700 static void bnx2x_gro_ip_csum(struct bnx2x *bp, struct sk_buff *skb)
702 const struct iphdr *iph = ip_hdr(skb);
705 skb_set_transport_header(skb, sizeof(struct iphdr));
708 th->check = ~tcp_v4_check(skb->len - skb_transport_offset(skb),
709 iph->saddr, iph->daddr, 0);
712 static void bnx2x_gro_ipv6_csum(struct bnx2x *bp, struct sk_buff *skb)
714 struct ipv6hdr *iph = ipv6_hdr(skb);
717 skb_set_transport_header(skb, sizeof(struct ipv6hdr));
720 th->check = ~tcp_v6_check(skb->len - skb_transport_offset(skb),
721 &iph->saddr, &iph->daddr, 0);
724 static void bnx2x_gro_csum(struct bnx2x *bp, struct sk_buff *skb,
725 void (*gro_func)(struct bnx2x*, struct sk_buff*))
727 skb_set_network_header(skb, 0);
729 tcp_gro_complete(skb);
733 static void bnx2x_gro_receive(struct bnx2x *bp, struct bnx2x_fastpath *fp,
737 if (skb_shinfo(skb)->gso_size) {
738 switch (be16_to_cpu(skb->protocol)) {
740 bnx2x_gro_csum(bp, skb, bnx2x_gro_ip_csum);
743 bnx2x_gro_csum(bp, skb, bnx2x_gro_ipv6_csum);
746 WARN_ONCE(1, "Error: FW GRO supports only IPv4/IPv6, not 0x%04x\n",
747 be16_to_cpu(skb->protocol));
751 skb_record_rx_queue(skb, fp->rx_queue);
752 napi_gro_receive(&fp->napi, skb);
755 static void bnx2x_tpa_stop(struct bnx2x *bp, struct bnx2x_fastpath *fp,
756 struct bnx2x_agg_info *tpa_info,
758 struct eth_end_agg_rx_cqe *cqe,
761 struct sw_rx_bd *rx_buf = &tpa_info->first_buf;
762 u8 pad = tpa_info->placement_offset;
763 u16 len = tpa_info->len_on_bd;
764 struct sk_buff *skb = NULL;
765 u8 *new_data, *data = rx_buf->data;
766 u8 old_tpa_state = tpa_info->tpa_state;
768 tpa_info->tpa_state = BNX2X_TPA_STOP;
770 /* If we there was an error during the handling of the TPA_START -
771 * drop this aggregation.
773 if (old_tpa_state == BNX2X_TPA_ERROR)
776 /* Try to allocate the new data */
777 new_data = bnx2x_frag_alloc(fp, GFP_ATOMIC);
778 /* Unmap skb in the pool anyway, as we are going to change
779 pool entry status to BNX2X_TPA_STOP even if new skb allocation
781 dma_unmap_single(&bp->pdev->dev, dma_unmap_addr(rx_buf, mapping),
782 fp->rx_buf_size, DMA_FROM_DEVICE);
783 if (likely(new_data))
784 skb = build_skb(data, fp->rx_frag_size);
787 #ifdef BNX2X_STOP_ON_ERROR
788 if (pad + len > fp->rx_buf_size) {
789 BNX2X_ERR("skb_put is about to fail... pad %d len %d rx_buf_size %d\n",
790 pad, len, fp->rx_buf_size);
796 skb_reserve(skb, pad + NET_SKB_PAD);
798 skb_set_hash(skb, tpa_info->rxhash, tpa_info->rxhash_type);
800 skb->protocol = eth_type_trans(skb, bp->dev);
801 skb->ip_summed = CHECKSUM_UNNECESSARY;
803 if (!bnx2x_fill_frag_skb(bp, fp, tpa_info, pages,
804 skb, cqe, cqe_idx)) {
805 if (tpa_info->parsing_flags & PARSING_FLAGS_VLAN)
806 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), tpa_info->vlan_tag);
807 bnx2x_gro_receive(bp, fp, skb);
809 DP(NETIF_MSG_RX_STATUS,
810 "Failed to allocate new pages - dropping packet!\n");
811 dev_kfree_skb_any(skb);
814 /* put new data in bin */
815 rx_buf->data = new_data;
820 bnx2x_frag_free(fp, new_data);
822 /* drop the packet and keep the buffer in the bin */
823 DP(NETIF_MSG_RX_STATUS,
824 "Failed to allocate or map a new skb - dropping packet!\n");
825 bnx2x_fp_stats(bp, fp)->eth_q_stats.rx_skb_alloc_failed++;
828 static int bnx2x_alloc_rx_data(struct bnx2x *bp, struct bnx2x_fastpath *fp,
829 u16 index, gfp_t gfp_mask)
832 struct sw_rx_bd *rx_buf = &fp->rx_buf_ring[index];
833 struct eth_rx_bd *rx_bd = &fp->rx_desc_ring[index];
836 data = bnx2x_frag_alloc(fp, gfp_mask);
837 if (unlikely(data == NULL))
840 mapping = dma_map_single(&bp->pdev->dev, data + NET_SKB_PAD,
843 if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
844 bnx2x_frag_free(fp, data);
845 BNX2X_ERR("Can't map rx data\n");
850 dma_unmap_addr_set(rx_buf, mapping, mapping);
852 rx_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
853 rx_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
859 void bnx2x_csum_validate(struct sk_buff *skb, union eth_rx_cqe *cqe,
860 struct bnx2x_fastpath *fp,
861 struct bnx2x_eth_q_stats *qstats)
863 /* Do nothing if no L4 csum validation was done.
864 * We do not check whether IP csum was validated. For IPv4 we assume
865 * that if the card got as far as validating the L4 csum, it also
866 * validated the IP csum. IPv6 has no IP csum.
868 if (cqe->fast_path_cqe.status_flags &
869 ETH_FAST_PATH_RX_CQE_L4_XSUM_NO_VALIDATION_FLG)
872 /* If L4 validation was done, check if an error was found. */
874 if (cqe->fast_path_cqe.type_error_flags &
875 (ETH_FAST_PATH_RX_CQE_IP_BAD_XSUM_FLG |
876 ETH_FAST_PATH_RX_CQE_L4_BAD_XSUM_FLG))
877 qstats->hw_csum_err++;
879 skb->ip_summed = CHECKSUM_UNNECESSARY;
882 static int bnx2x_rx_int(struct bnx2x_fastpath *fp, int budget)
884 struct bnx2x *bp = fp->bp;
885 u16 bd_cons, bd_prod, bd_prod_fw, comp_ring_cons;
886 u16 sw_comp_cons, sw_comp_prod;
888 union eth_rx_cqe *cqe;
889 struct eth_fast_path_rx_cqe *cqe_fp;
891 #ifdef BNX2X_STOP_ON_ERROR
892 if (unlikely(bp->panic))
898 bd_cons = fp->rx_bd_cons;
899 bd_prod = fp->rx_bd_prod;
900 bd_prod_fw = bd_prod;
901 sw_comp_cons = fp->rx_comp_cons;
902 sw_comp_prod = fp->rx_comp_prod;
904 comp_ring_cons = RCQ_BD(sw_comp_cons);
905 cqe = &fp->rx_comp_ring[comp_ring_cons];
906 cqe_fp = &cqe->fast_path_cqe;
908 DP(NETIF_MSG_RX_STATUS,
909 "queue[%d]: sw_comp_cons %u\n", fp->index, sw_comp_cons);
911 while (BNX2X_IS_CQE_COMPLETED(cqe_fp)) {
912 struct sw_rx_bd *rx_buf = NULL;
915 enum eth_rx_cqe_type cqe_fp_type;
919 enum pkt_hash_types rxhash_type;
921 #ifdef BNX2X_STOP_ON_ERROR
922 if (unlikely(bp->panic))
926 bd_prod = RX_BD(bd_prod);
927 bd_cons = RX_BD(bd_cons);
929 /* A rmb() is required to ensure that the CQE is not read
930 * before it is written by the adapter DMA. PCI ordering
931 * rules will make sure the other fields are written before
932 * the marker at the end of struct eth_fast_path_rx_cqe
933 * but without rmb() a weakly ordered processor can process
934 * stale data. Without the barrier TPA state-machine might
935 * enter inconsistent state and kernel stack might be
936 * provided with incorrect packet description - these lead
937 * to various kernel crashed.
941 cqe_fp_flags = cqe_fp->type_error_flags;
942 cqe_fp_type = cqe_fp_flags & ETH_FAST_PATH_RX_CQE_TYPE;
944 DP(NETIF_MSG_RX_STATUS,
945 "CQE type %x err %x status %x queue %x vlan %x len %u\n",
946 CQE_TYPE(cqe_fp_flags),
947 cqe_fp_flags, cqe_fp->status_flags,
948 le32_to_cpu(cqe_fp->rss_hash_result),
949 le16_to_cpu(cqe_fp->vlan_tag),
950 le16_to_cpu(cqe_fp->pkt_len_or_gro_seg_len));
952 /* is this a slowpath msg? */
953 if (unlikely(CQE_TYPE_SLOW(cqe_fp_type))) {
954 bnx2x_sp_event(fp, cqe);
958 rx_buf = &fp->rx_buf_ring[bd_cons];
961 if (!CQE_TYPE_FAST(cqe_fp_type)) {
962 struct bnx2x_agg_info *tpa_info;
963 u16 frag_size, pages;
964 #ifdef BNX2X_STOP_ON_ERROR
966 if (fp->mode == TPA_MODE_DISABLED &&
967 (CQE_TYPE_START(cqe_fp_type) ||
968 CQE_TYPE_STOP(cqe_fp_type)))
969 BNX2X_ERR("START/STOP packet while TPA disabled, type %x\n",
970 CQE_TYPE(cqe_fp_type));
973 if (CQE_TYPE_START(cqe_fp_type)) {
974 u16 queue = cqe_fp->queue_index;
975 DP(NETIF_MSG_RX_STATUS,
976 "calling tpa_start on queue %d\n",
979 bnx2x_tpa_start(fp, queue,
985 queue = cqe->end_agg_cqe.queue_index;
986 tpa_info = &fp->tpa_info[queue];
987 DP(NETIF_MSG_RX_STATUS,
988 "calling tpa_stop on queue %d\n",
991 frag_size = le16_to_cpu(cqe->end_agg_cqe.pkt_len) -
994 if (fp->mode == TPA_MODE_GRO)
995 pages = (frag_size + tpa_info->full_page - 1) /
998 pages = SGE_PAGE_ALIGN(frag_size) >>
1001 bnx2x_tpa_stop(bp, fp, tpa_info, pages,
1002 &cqe->end_agg_cqe, comp_ring_cons);
1003 #ifdef BNX2X_STOP_ON_ERROR
1008 bnx2x_update_sge_prod(fp, pages, &cqe->end_agg_cqe);
1012 len = le16_to_cpu(cqe_fp->pkt_len_or_gro_seg_len);
1013 pad = cqe_fp->placement_offset;
1014 dma_sync_single_for_cpu(&bp->pdev->dev,
1015 dma_unmap_addr(rx_buf, mapping),
1016 pad + RX_COPY_THRESH,
1019 prefetch(data + pad); /* speedup eth_type_trans() */
1020 /* is this an error packet? */
1021 if (unlikely(cqe_fp_flags & ETH_RX_ERROR_FALGS)) {
1022 DP(NETIF_MSG_RX_ERR | NETIF_MSG_RX_STATUS,
1023 "ERROR flags %x rx packet %u\n",
1024 cqe_fp_flags, sw_comp_cons);
1025 bnx2x_fp_qstats(bp, fp)->rx_err_discard_pkt++;
1029 /* Since we don't have a jumbo ring
1030 * copy small packets if mtu > 1500
1032 if ((bp->dev->mtu > ETH_MAX_PACKET_SIZE) &&
1033 (len <= RX_COPY_THRESH)) {
1034 skb = napi_alloc_skb(&fp->napi, len);
1036 DP(NETIF_MSG_RX_ERR | NETIF_MSG_RX_STATUS,
1037 "ERROR packet dropped because of alloc failure\n");
1038 bnx2x_fp_qstats(bp, fp)->rx_skb_alloc_failed++;
1041 memcpy(skb->data, data + pad, len);
1042 bnx2x_reuse_rx_data(fp, bd_cons, bd_prod);
1044 if (likely(bnx2x_alloc_rx_data(bp, fp, bd_prod,
1045 GFP_ATOMIC) == 0)) {
1046 dma_unmap_single(&bp->pdev->dev,
1047 dma_unmap_addr(rx_buf, mapping),
1050 skb = build_skb(data, fp->rx_frag_size);
1051 if (unlikely(!skb)) {
1052 bnx2x_frag_free(fp, data);
1053 bnx2x_fp_qstats(bp, fp)->
1054 rx_skb_alloc_failed++;
1057 skb_reserve(skb, pad);
1059 DP(NETIF_MSG_RX_ERR | NETIF_MSG_RX_STATUS,
1060 "ERROR packet dropped because of alloc failure\n");
1061 bnx2x_fp_qstats(bp, fp)->rx_skb_alloc_failed++;
1063 bnx2x_reuse_rx_data(fp, bd_cons, bd_prod);
1069 skb->protocol = eth_type_trans(skb, bp->dev);
1071 /* Set Toeplitz hash for a none-LRO skb */
1072 rxhash = bnx2x_get_rxhash(bp, cqe_fp, &rxhash_type);
1073 skb_set_hash(skb, rxhash, rxhash_type);
1075 skb_checksum_none_assert(skb);
1077 if (bp->dev->features & NETIF_F_RXCSUM)
1078 bnx2x_csum_validate(skb, cqe, fp,
1079 bnx2x_fp_qstats(bp, fp));
1081 skb_record_rx_queue(skb, fp->rx_queue);
1083 /* Check if this packet was timestamped */
1084 if (unlikely(cqe->fast_path_cqe.type_error_flags &
1085 (1 << ETH_FAST_PATH_RX_CQE_PTP_PKT_SHIFT)))
1086 bnx2x_set_rx_ts(bp, skb);
1088 if (le16_to_cpu(cqe_fp->pars_flags.flags) &
1090 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
1091 le16_to_cpu(cqe_fp->vlan_tag));
1093 napi_gro_receive(&fp->napi, skb);
1095 rx_buf->data = NULL;
1097 bd_cons = NEXT_RX_IDX(bd_cons);
1098 bd_prod = NEXT_RX_IDX(bd_prod);
1099 bd_prod_fw = NEXT_RX_IDX(bd_prod_fw);
1102 sw_comp_prod = NEXT_RCQ_IDX(sw_comp_prod);
1103 sw_comp_cons = NEXT_RCQ_IDX(sw_comp_cons);
1105 /* mark CQE as free */
1106 BNX2X_SEED_CQE(cqe_fp);
1108 if (rx_pkt == budget)
1111 comp_ring_cons = RCQ_BD(sw_comp_cons);
1112 cqe = &fp->rx_comp_ring[comp_ring_cons];
1113 cqe_fp = &cqe->fast_path_cqe;
1116 fp->rx_bd_cons = bd_cons;
1117 fp->rx_bd_prod = bd_prod_fw;
1118 fp->rx_comp_cons = sw_comp_cons;
1119 fp->rx_comp_prod = sw_comp_prod;
1121 /* Update producers */
1122 bnx2x_update_rx_prod(bp, fp, bd_prod_fw, sw_comp_prod,
1128 static irqreturn_t bnx2x_msix_fp_int(int irq, void *fp_cookie)
1130 struct bnx2x_fastpath *fp = fp_cookie;
1131 struct bnx2x *bp = fp->bp;
1135 "got an MSI-X interrupt on IDX:SB [fp %d fw_sd %d igusb %d]\n",
1136 fp->index, fp->fw_sb_id, fp->igu_sb_id);
1138 bnx2x_ack_sb(bp, fp->igu_sb_id, USTORM_ID, 0, IGU_INT_DISABLE, 0);
1140 #ifdef BNX2X_STOP_ON_ERROR
1141 if (unlikely(bp->panic))
1145 /* Handle Rx and Tx according to MSI-X vector */
1146 for_each_cos_in_tx_queue(fp, cos)
1147 prefetch(fp->txdata_ptr[cos]->tx_cons_sb);
1149 prefetch(&fp->sb_running_index[SM_RX_ID]);
1150 napi_schedule_irqoff(&bnx2x_fp(bp, fp->index, napi));
1155 /* HW Lock for shared dual port PHYs */
1156 void bnx2x_acquire_phy_lock(struct bnx2x *bp)
1158 mutex_lock(&bp->port.phy_mutex);
1160 bnx2x_acquire_hw_lock(bp, HW_LOCK_RESOURCE_MDIO);
1163 void bnx2x_release_phy_lock(struct bnx2x *bp)
1165 bnx2x_release_hw_lock(bp, HW_LOCK_RESOURCE_MDIO);
1167 mutex_unlock(&bp->port.phy_mutex);
1170 /* calculates MF speed according to current linespeed and MF configuration */
1171 u16 bnx2x_get_mf_speed(struct bnx2x *bp)
1173 u16 line_speed = bp->link_vars.line_speed;
1175 u16 maxCfg = bnx2x_extract_max_cfg(bp,
1176 bp->mf_config[BP_VN(bp)]);
1178 /* Calculate the current MAX line speed limit for the MF
1181 if (IS_MF_PERCENT_BW(bp))
1182 line_speed = (line_speed * maxCfg) / 100;
1183 else { /* SD mode */
1184 u16 vn_max_rate = maxCfg * 100;
1186 if (vn_max_rate < line_speed)
1187 line_speed = vn_max_rate;
1195 * bnx2x_fill_report_data - fill link report data to report
1197 * @bp: driver handle
1198 * @data: link state to update
1200 * It uses a none-atomic bit operations because is called under the mutex.
1202 static void bnx2x_fill_report_data(struct bnx2x *bp,
1203 struct bnx2x_link_report_data *data)
1205 memset(data, 0, sizeof(*data));
1208 /* Fill the report data: effective line speed */
1209 data->line_speed = bnx2x_get_mf_speed(bp);
1212 if (!bp->link_vars.link_up || (bp->flags & MF_FUNC_DIS))
1213 __set_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1214 &data->link_report_flags);
1216 if (!BNX2X_NUM_ETH_QUEUES(bp))
1217 __set_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1218 &data->link_report_flags);
1221 if (bp->link_vars.duplex == DUPLEX_FULL)
1222 __set_bit(BNX2X_LINK_REPORT_FD,
1223 &data->link_report_flags);
1225 /* Rx Flow Control is ON */
1226 if (bp->link_vars.flow_ctrl & BNX2X_FLOW_CTRL_RX)
1227 __set_bit(BNX2X_LINK_REPORT_RX_FC_ON,
1228 &data->link_report_flags);
1230 /* Tx Flow Control is ON */
1231 if (bp->link_vars.flow_ctrl & BNX2X_FLOW_CTRL_TX)
1232 __set_bit(BNX2X_LINK_REPORT_TX_FC_ON,
1233 &data->link_report_flags);
1235 *data = bp->vf_link_vars;
1240 * bnx2x_link_report - report link status to OS.
1242 * @bp: driver handle
1244 * Calls the __bnx2x_link_report() under the same locking scheme
1245 * as a link/PHY state managing code to ensure a consistent link
1249 void bnx2x_link_report(struct bnx2x *bp)
1251 bnx2x_acquire_phy_lock(bp);
1252 __bnx2x_link_report(bp);
1253 bnx2x_release_phy_lock(bp);
1257 * __bnx2x_link_report - report link status to OS.
1259 * @bp: driver handle
1261 * None atomic implementation.
1262 * Should be called under the phy_lock.
1264 void __bnx2x_link_report(struct bnx2x *bp)
1266 struct bnx2x_link_report_data cur_data;
1269 if (IS_PF(bp) && !CHIP_IS_E1(bp))
1270 bnx2x_read_mf_cfg(bp);
1272 /* Read the current link report info */
1273 bnx2x_fill_report_data(bp, &cur_data);
1275 /* Don't report link down or exactly the same link status twice */
1276 if (!memcmp(&cur_data, &bp->last_reported_link, sizeof(cur_data)) ||
1277 (test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1278 &bp->last_reported_link.link_report_flags) &&
1279 test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1280 &cur_data.link_report_flags)))
1285 /* We are going to report a new link parameters now -
1286 * remember the current data for the next time.
1288 memcpy(&bp->last_reported_link, &cur_data, sizeof(cur_data));
1290 /* propagate status to VFs */
1292 bnx2x_iov_link_update(bp);
1294 if (test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1295 &cur_data.link_report_flags)) {
1296 netif_carrier_off(bp->dev);
1297 netdev_err(bp->dev, "NIC Link is Down\n");
1303 netif_carrier_on(bp->dev);
1305 if (test_and_clear_bit(BNX2X_LINK_REPORT_FD,
1306 &cur_data.link_report_flags))
1311 /* Handle the FC at the end so that only these flags would be
1312 * possibly set. This way we may easily check if there is no FC
1315 if (cur_data.link_report_flags) {
1316 if (test_bit(BNX2X_LINK_REPORT_RX_FC_ON,
1317 &cur_data.link_report_flags)) {
1318 if (test_bit(BNX2X_LINK_REPORT_TX_FC_ON,
1319 &cur_data.link_report_flags))
1320 flow = "ON - receive & transmit";
1322 flow = "ON - receive";
1324 flow = "ON - transmit";
1329 netdev_info(bp->dev, "NIC Link is Up, %d Mbps %s duplex, Flow control: %s\n",
1330 cur_data.line_speed, duplex, flow);
1334 static void bnx2x_set_next_page_sgl(struct bnx2x_fastpath *fp)
1338 for (i = 1; i <= NUM_RX_SGE_PAGES; i++) {
1339 struct eth_rx_sge *sge;
1341 sge = &fp->rx_sge_ring[RX_SGE_CNT * i - 2];
1343 cpu_to_le32(U64_HI(fp->rx_sge_mapping +
1344 BCM_PAGE_SIZE*(i % NUM_RX_SGE_PAGES)));
1347 cpu_to_le32(U64_LO(fp->rx_sge_mapping +
1348 BCM_PAGE_SIZE*(i % NUM_RX_SGE_PAGES)));
1352 static void bnx2x_free_tpa_pool(struct bnx2x *bp,
1353 struct bnx2x_fastpath *fp, int last)
1357 for (i = 0; i < last; i++) {
1358 struct bnx2x_agg_info *tpa_info = &fp->tpa_info[i];
1359 struct sw_rx_bd *first_buf = &tpa_info->first_buf;
1360 u8 *data = first_buf->data;
1363 DP(NETIF_MSG_IFDOWN, "tpa bin %d empty on free\n", i);
1366 if (tpa_info->tpa_state == BNX2X_TPA_START)
1367 dma_unmap_single(&bp->pdev->dev,
1368 dma_unmap_addr(first_buf, mapping),
1369 fp->rx_buf_size, DMA_FROM_DEVICE);
1370 bnx2x_frag_free(fp, data);
1371 first_buf->data = NULL;
1375 void bnx2x_init_rx_rings_cnic(struct bnx2x *bp)
1379 for_each_rx_queue_cnic(bp, j) {
1380 struct bnx2x_fastpath *fp = &bp->fp[j];
1384 /* Activate BD ring */
1386 * this will generate an interrupt (to the TSTORM)
1387 * must only be done after chip is initialized
1389 bnx2x_update_rx_prod(bp, fp, fp->rx_bd_prod, fp->rx_comp_prod,
1394 void bnx2x_init_rx_rings(struct bnx2x *bp)
1396 int func = BP_FUNC(bp);
1400 /* Allocate TPA resources */
1401 for_each_eth_queue(bp, j) {
1402 struct bnx2x_fastpath *fp = &bp->fp[j];
1405 "mtu %d rx_buf_size %d\n", bp->dev->mtu, fp->rx_buf_size);
1407 if (fp->mode != TPA_MODE_DISABLED) {
1408 /* Fill the per-aggregation pool */
1409 for (i = 0; i < MAX_AGG_QS(bp); i++) {
1410 struct bnx2x_agg_info *tpa_info =
1412 struct sw_rx_bd *first_buf =
1413 &tpa_info->first_buf;
1416 bnx2x_frag_alloc(fp, GFP_KERNEL);
1417 if (!first_buf->data) {
1418 BNX2X_ERR("Failed to allocate TPA skb pool for queue[%d] - disabling TPA on this queue!\n",
1420 bnx2x_free_tpa_pool(bp, fp, i);
1421 fp->mode = TPA_MODE_DISABLED;
1424 dma_unmap_addr_set(first_buf, mapping, 0);
1425 tpa_info->tpa_state = BNX2X_TPA_STOP;
1428 /* "next page" elements initialization */
1429 bnx2x_set_next_page_sgl(fp);
1431 /* set SGEs bit mask */
1432 bnx2x_init_sge_ring_bit_mask(fp);
1434 /* Allocate SGEs and initialize the ring elements */
1435 for (i = 0, ring_prod = 0;
1436 i < MAX_RX_SGE_CNT*NUM_RX_SGE_PAGES; i++) {
1438 if (bnx2x_alloc_rx_sge(bp, fp, ring_prod,
1440 BNX2X_ERR("was only able to allocate %d rx sges\n",
1442 BNX2X_ERR("disabling TPA for queue[%d]\n",
1444 /* Cleanup already allocated elements */
1445 bnx2x_free_rx_sge_range(bp, fp,
1447 bnx2x_free_tpa_pool(bp, fp,
1449 fp->mode = TPA_MODE_DISABLED;
1453 ring_prod = NEXT_SGE_IDX(ring_prod);
1456 fp->rx_sge_prod = ring_prod;
1460 for_each_eth_queue(bp, j) {
1461 struct bnx2x_fastpath *fp = &bp->fp[j];
1465 /* Activate BD ring */
1467 * this will generate an interrupt (to the TSTORM)
1468 * must only be done after chip is initialized
1470 bnx2x_update_rx_prod(bp, fp, fp->rx_bd_prod, fp->rx_comp_prod,
1476 if (CHIP_IS_E1(bp)) {
1477 REG_WR(bp, BAR_USTRORM_INTMEM +
1478 USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func),
1479 U64_LO(fp->rx_comp_mapping));
1480 REG_WR(bp, BAR_USTRORM_INTMEM +
1481 USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func) + 4,
1482 U64_HI(fp->rx_comp_mapping));
1487 static void bnx2x_free_tx_skbs_queue(struct bnx2x_fastpath *fp)
1490 struct bnx2x *bp = fp->bp;
1492 for_each_cos_in_tx_queue(fp, cos) {
1493 struct bnx2x_fp_txdata *txdata = fp->txdata_ptr[cos];
1494 unsigned pkts_compl = 0, bytes_compl = 0;
1496 u16 sw_prod = txdata->tx_pkt_prod;
1497 u16 sw_cons = txdata->tx_pkt_cons;
1499 while (sw_cons != sw_prod) {
1500 bnx2x_free_tx_pkt(bp, txdata, TX_BD(sw_cons),
1501 &pkts_compl, &bytes_compl);
1505 netdev_tx_reset_queue(
1506 netdev_get_tx_queue(bp->dev,
1507 txdata->txq_index));
1511 static void bnx2x_free_tx_skbs_cnic(struct bnx2x *bp)
1515 for_each_tx_queue_cnic(bp, i) {
1516 bnx2x_free_tx_skbs_queue(&bp->fp[i]);
1520 static void bnx2x_free_tx_skbs(struct bnx2x *bp)
1524 for_each_eth_queue(bp, i) {
1525 bnx2x_free_tx_skbs_queue(&bp->fp[i]);
1529 static void bnx2x_free_rx_bds(struct bnx2x_fastpath *fp)
1531 struct bnx2x *bp = fp->bp;
1534 /* ring wasn't allocated */
1535 if (fp->rx_buf_ring == NULL)
1538 for (i = 0; i < NUM_RX_BD; i++) {
1539 struct sw_rx_bd *rx_buf = &fp->rx_buf_ring[i];
1540 u8 *data = rx_buf->data;
1544 dma_unmap_single(&bp->pdev->dev,
1545 dma_unmap_addr(rx_buf, mapping),
1546 fp->rx_buf_size, DMA_FROM_DEVICE);
1548 rx_buf->data = NULL;
1549 bnx2x_frag_free(fp, data);
1553 static void bnx2x_free_rx_skbs_cnic(struct bnx2x *bp)
1557 for_each_rx_queue_cnic(bp, j) {
1558 bnx2x_free_rx_bds(&bp->fp[j]);
1562 static void bnx2x_free_rx_skbs(struct bnx2x *bp)
1566 for_each_eth_queue(bp, j) {
1567 struct bnx2x_fastpath *fp = &bp->fp[j];
1569 bnx2x_free_rx_bds(fp);
1571 if (fp->mode != TPA_MODE_DISABLED)
1572 bnx2x_free_tpa_pool(bp, fp, MAX_AGG_QS(bp));
1576 static void bnx2x_free_skbs_cnic(struct bnx2x *bp)
1578 bnx2x_free_tx_skbs_cnic(bp);
1579 bnx2x_free_rx_skbs_cnic(bp);
1582 void bnx2x_free_skbs(struct bnx2x *bp)
1584 bnx2x_free_tx_skbs(bp);
1585 bnx2x_free_rx_skbs(bp);
1588 void bnx2x_update_max_mf_config(struct bnx2x *bp, u32 value)
1590 /* load old values */
1591 u32 mf_cfg = bp->mf_config[BP_VN(bp)];
1593 if (value != bnx2x_extract_max_cfg(bp, mf_cfg)) {
1594 /* leave all but MAX value */
1595 mf_cfg &= ~FUNC_MF_CFG_MAX_BW_MASK;
1597 /* set new MAX value */
1598 mf_cfg |= (value << FUNC_MF_CFG_MAX_BW_SHIFT)
1599 & FUNC_MF_CFG_MAX_BW_MASK;
1601 bnx2x_fw_command(bp, DRV_MSG_CODE_SET_MF_BW, mf_cfg);
1606 * bnx2x_free_msix_irqs - free previously requested MSI-X IRQ vectors
1608 * @bp: driver handle
1609 * @nvecs: number of vectors to be released
1611 static void bnx2x_free_msix_irqs(struct bnx2x *bp, int nvecs)
1615 if (nvecs == offset)
1618 /* VFs don't have a default SB */
1620 free_irq(bp->msix_table[offset].vector, bp->dev);
1621 DP(NETIF_MSG_IFDOWN, "released sp irq (%d)\n",
1622 bp->msix_table[offset].vector);
1626 if (CNIC_SUPPORT(bp)) {
1627 if (nvecs == offset)
1632 for_each_eth_queue(bp, i) {
1633 if (nvecs == offset)
1635 DP(NETIF_MSG_IFDOWN, "about to release fp #%d->%d irq\n",
1636 i, bp->msix_table[offset].vector);
1638 free_irq(bp->msix_table[offset++].vector, &bp->fp[i]);
1642 void bnx2x_free_irq(struct bnx2x *bp)
1644 if (bp->flags & USING_MSIX_FLAG &&
1645 !(bp->flags & USING_SINGLE_MSIX_FLAG)) {
1646 int nvecs = BNX2X_NUM_ETH_QUEUES(bp) + CNIC_SUPPORT(bp);
1648 /* vfs don't have a default status block */
1652 bnx2x_free_msix_irqs(bp, nvecs);
1654 free_irq(bp->dev->irq, bp->dev);
1658 int bnx2x_enable_msix(struct bnx2x *bp)
1660 int msix_vec = 0, i, rc;
1662 /* VFs don't have a default status block */
1664 bp->msix_table[msix_vec].entry = msix_vec;
1665 BNX2X_DEV_INFO("msix_table[0].entry = %d (slowpath)\n",
1666 bp->msix_table[0].entry);
1670 /* Cnic requires an msix vector for itself */
1671 if (CNIC_SUPPORT(bp)) {
1672 bp->msix_table[msix_vec].entry = msix_vec;
1673 BNX2X_DEV_INFO("msix_table[%d].entry = %d (CNIC)\n",
1674 msix_vec, bp->msix_table[msix_vec].entry);
1678 /* We need separate vectors for ETH queues only (not FCoE) */
1679 for_each_eth_queue(bp, i) {
1680 bp->msix_table[msix_vec].entry = msix_vec;
1681 BNX2X_DEV_INFO("msix_table[%d].entry = %d (fastpath #%u)\n",
1682 msix_vec, msix_vec, i);
1686 DP(BNX2X_MSG_SP, "about to request enable msix with %d vectors\n",
1689 rc = pci_enable_msix_range(bp->pdev, &bp->msix_table[0],
1690 BNX2X_MIN_MSIX_VEC_CNT(bp), msix_vec);
1692 * reconfigure number of tx/rx queues according to available
1695 if (rc == -ENOSPC) {
1696 /* Get by with single vector */
1697 rc = pci_enable_msix_range(bp->pdev, &bp->msix_table[0], 1, 1);
1699 BNX2X_DEV_INFO("Single MSI-X is not attainable rc %d\n",
1704 BNX2X_DEV_INFO("Using single MSI-X vector\n");
1705 bp->flags |= USING_SINGLE_MSIX_FLAG;
1707 BNX2X_DEV_INFO("set number of queues to 1\n");
1708 bp->num_ethernet_queues = 1;
1709 bp->num_queues = bp->num_ethernet_queues + bp->num_cnic_queues;
1710 } else if (rc < 0) {
1711 BNX2X_DEV_INFO("MSI-X is not attainable rc %d\n", rc);
1713 } else if (rc < msix_vec) {
1714 /* how less vectors we will have? */
1715 int diff = msix_vec - rc;
1717 BNX2X_DEV_INFO("Trying to use less MSI-X vectors: %d\n", rc);
1720 * decrease number of queues by number of unallocated entries
1722 bp->num_ethernet_queues -= diff;
1723 bp->num_queues = bp->num_ethernet_queues + bp->num_cnic_queues;
1725 BNX2X_DEV_INFO("New queue configuration set: %d\n",
1729 bp->flags |= USING_MSIX_FLAG;
1734 /* fall to INTx if not enough memory */
1736 bp->flags |= DISABLE_MSI_FLAG;
1741 static int bnx2x_req_msix_irqs(struct bnx2x *bp)
1743 int i, rc, offset = 0;
1745 /* no default status block for vf */
1747 rc = request_irq(bp->msix_table[offset++].vector,
1748 bnx2x_msix_sp_int, 0,
1749 bp->dev->name, bp->dev);
1751 BNX2X_ERR("request sp irq failed\n");
1756 if (CNIC_SUPPORT(bp))
1759 for_each_eth_queue(bp, i) {
1760 struct bnx2x_fastpath *fp = &bp->fp[i];
1761 snprintf(fp->name, sizeof(fp->name), "%s-fp-%d",
1764 rc = request_irq(bp->msix_table[offset].vector,
1765 bnx2x_msix_fp_int, 0, fp->name, fp);
1767 BNX2X_ERR("request fp #%d irq (%d) failed rc %d\n", i,
1768 bp->msix_table[offset].vector, rc);
1769 bnx2x_free_msix_irqs(bp, offset);
1776 i = BNX2X_NUM_ETH_QUEUES(bp);
1778 offset = 1 + CNIC_SUPPORT(bp);
1779 netdev_info(bp->dev,
1780 "using MSI-X IRQs: sp %d fp[%d] %d ... fp[%d] %d\n",
1781 bp->msix_table[0].vector,
1782 0, bp->msix_table[offset].vector,
1783 i - 1, bp->msix_table[offset + i - 1].vector);
1785 offset = CNIC_SUPPORT(bp);
1786 netdev_info(bp->dev,
1787 "using MSI-X IRQs: fp[%d] %d ... fp[%d] %d\n",
1788 0, bp->msix_table[offset].vector,
1789 i - 1, bp->msix_table[offset + i - 1].vector);
1794 int bnx2x_enable_msi(struct bnx2x *bp)
1798 rc = pci_enable_msi(bp->pdev);
1800 BNX2X_DEV_INFO("MSI is not attainable\n");
1803 bp->flags |= USING_MSI_FLAG;
1808 static int bnx2x_req_irq(struct bnx2x *bp)
1810 unsigned long flags;
1813 if (bp->flags & (USING_MSI_FLAG | USING_MSIX_FLAG))
1816 flags = IRQF_SHARED;
1818 if (bp->flags & USING_MSIX_FLAG)
1819 irq = bp->msix_table[0].vector;
1821 irq = bp->pdev->irq;
1823 return request_irq(irq, bnx2x_interrupt, flags, bp->dev->name, bp->dev);
1826 static int bnx2x_setup_irqs(struct bnx2x *bp)
1829 if (bp->flags & USING_MSIX_FLAG &&
1830 !(bp->flags & USING_SINGLE_MSIX_FLAG)) {
1831 rc = bnx2x_req_msix_irqs(bp);
1835 rc = bnx2x_req_irq(bp);
1837 BNX2X_ERR("IRQ request failed rc %d, aborting\n", rc);
1840 if (bp->flags & USING_MSI_FLAG) {
1841 bp->dev->irq = bp->pdev->irq;
1842 netdev_info(bp->dev, "using MSI IRQ %d\n",
1845 if (bp->flags & USING_MSIX_FLAG) {
1846 bp->dev->irq = bp->msix_table[0].vector;
1847 netdev_info(bp->dev, "using MSIX IRQ %d\n",
1855 static void bnx2x_napi_enable_cnic(struct bnx2x *bp)
1859 for_each_rx_queue_cnic(bp, i) {
1860 napi_enable(&bnx2x_fp(bp, i, napi));
1864 static void bnx2x_napi_enable(struct bnx2x *bp)
1868 for_each_eth_queue(bp, i) {
1869 napi_enable(&bnx2x_fp(bp, i, napi));
1873 static void bnx2x_napi_disable_cnic(struct bnx2x *bp)
1877 for_each_rx_queue_cnic(bp, i) {
1878 napi_disable(&bnx2x_fp(bp, i, napi));
1882 static void bnx2x_napi_disable(struct bnx2x *bp)
1886 for_each_eth_queue(bp, i) {
1887 napi_disable(&bnx2x_fp(bp, i, napi));
1891 void bnx2x_netif_start(struct bnx2x *bp)
1893 if (netif_running(bp->dev)) {
1894 bnx2x_napi_enable(bp);
1895 if (CNIC_LOADED(bp))
1896 bnx2x_napi_enable_cnic(bp);
1897 bnx2x_int_enable(bp);
1898 if (bp->state == BNX2X_STATE_OPEN)
1899 netif_tx_wake_all_queues(bp->dev);
1903 void bnx2x_netif_stop(struct bnx2x *bp, int disable_hw)
1905 bnx2x_int_disable_sync(bp, disable_hw);
1906 bnx2x_napi_disable(bp);
1907 if (CNIC_LOADED(bp))
1908 bnx2x_napi_disable_cnic(bp);
1911 u16 bnx2x_select_queue(struct net_device *dev, struct sk_buff *skb,
1912 void *accel_priv, select_queue_fallback_t fallback)
1914 struct bnx2x *bp = netdev_priv(dev);
1916 if (CNIC_LOADED(bp) && !NO_FCOE(bp)) {
1917 struct ethhdr *hdr = (struct ethhdr *)skb->data;
1918 u16 ether_type = ntohs(hdr->h_proto);
1920 /* Skip VLAN tag if present */
1921 if (ether_type == ETH_P_8021Q) {
1922 struct vlan_ethhdr *vhdr =
1923 (struct vlan_ethhdr *)skb->data;
1925 ether_type = ntohs(vhdr->h_vlan_encapsulated_proto);
1928 /* If ethertype is FCoE or FIP - use FCoE ring */
1929 if ((ether_type == ETH_P_FCOE) || (ether_type == ETH_P_FIP))
1930 return bnx2x_fcoe_tx(bp, txq_index);
1933 /* select a non-FCoE queue */
1934 return fallback(dev, skb) % BNX2X_NUM_ETH_QUEUES(bp);
1937 void bnx2x_set_num_queues(struct bnx2x *bp)
1940 bp->num_ethernet_queues = bnx2x_calc_num_queues(bp);
1942 /* override in STORAGE SD modes */
1943 if (IS_MF_STORAGE_ONLY(bp))
1944 bp->num_ethernet_queues = 1;
1946 /* Add special queues */
1947 bp->num_cnic_queues = CNIC_SUPPORT(bp); /* For FCOE */
1948 bp->num_queues = bp->num_ethernet_queues + bp->num_cnic_queues;
1950 BNX2X_DEV_INFO("set number of queues to %d\n", bp->num_queues);
1954 * bnx2x_set_real_num_queues - configure netdev->real_num_[tx,rx]_queues
1956 * @bp: Driver handle
1958 * We currently support for at most 16 Tx queues for each CoS thus we will
1959 * allocate a multiple of 16 for ETH L2 rings according to the value of the
1962 * If there is an FCoE L2 queue the appropriate Tx queue will have the next
1963 * index after all ETH L2 indices.
1965 * If the actual number of Tx queues (for each CoS) is less than 16 then there
1966 * will be the holes at the end of each group of 16 ETh L2 indices (0..15,
1967 * 16..31,...) with indices that are not coupled with any real Tx queue.
1969 * The proper configuration of skb->queue_mapping is handled by
1970 * bnx2x_select_queue() and __skb_tx_hash().
1972 * bnx2x_setup_tc() takes care of the proper TC mappings so that __skb_tx_hash()
1973 * will return a proper Tx index if TC is enabled (netdev->num_tc > 0).
1975 static int bnx2x_set_real_num_queues(struct bnx2x *bp, int include_cnic)
1979 tx = BNX2X_NUM_ETH_QUEUES(bp) * bp->max_cos;
1980 rx = BNX2X_NUM_ETH_QUEUES(bp);
1982 /* account for fcoe queue */
1983 if (include_cnic && !NO_FCOE(bp)) {
1988 rc = netif_set_real_num_tx_queues(bp->dev, tx);
1990 BNX2X_ERR("Failed to set real number of Tx queues: %d\n", rc);
1993 rc = netif_set_real_num_rx_queues(bp->dev, rx);
1995 BNX2X_ERR("Failed to set real number of Rx queues: %d\n", rc);
1999 DP(NETIF_MSG_IFUP, "Setting real num queues to (tx, rx) (%d, %d)\n",
2005 static void bnx2x_set_rx_buf_size(struct bnx2x *bp)
2009 for_each_queue(bp, i) {
2010 struct bnx2x_fastpath *fp = &bp->fp[i];
2013 /* Always use a mini-jumbo MTU for the FCoE L2 ring */
2016 * Although there are no IP frames expected to arrive to
2017 * this ring we still want to add an
2018 * IP_HEADER_ALIGNMENT_PADDING to prevent a buffer
2021 mtu = BNX2X_FCOE_MINI_JUMBO_MTU;
2024 fp->rx_buf_size = BNX2X_FW_RX_ALIGN_START +
2025 IP_HEADER_ALIGNMENT_PADDING +
2028 BNX2X_FW_RX_ALIGN_END;
2029 /* Note : rx_buf_size doesn't take into account NET_SKB_PAD */
2030 if (fp->rx_buf_size + NET_SKB_PAD <= PAGE_SIZE)
2031 fp->rx_frag_size = fp->rx_buf_size + NET_SKB_PAD;
2033 fp->rx_frag_size = 0;
2037 static int bnx2x_init_rss(struct bnx2x *bp)
2040 u8 num_eth_queues = BNX2X_NUM_ETH_QUEUES(bp);
2042 /* Prepare the initial contents for the indirection table if RSS is
2045 for (i = 0; i < sizeof(bp->rss_conf_obj.ind_table); i++)
2046 bp->rss_conf_obj.ind_table[i] =
2048 ethtool_rxfh_indir_default(i, num_eth_queues);
2051 * For 57710 and 57711 SEARCHER configuration (rss_keys) is
2052 * per-port, so if explicit configuration is needed , do it only
2055 * For 57712 and newer on the other hand it's a per-function
2058 return bnx2x_config_rss_eth(bp, bp->port.pmf || !CHIP_IS_E1x(bp));
2061 int bnx2x_rss(struct bnx2x *bp, struct bnx2x_rss_config_obj *rss_obj,
2062 bool config_hash, bool enable)
2064 struct bnx2x_config_rss_params params = {NULL};
2066 /* Although RSS is meaningless when there is a single HW queue we
2067 * still need it enabled in order to have HW Rx hash generated.
2069 * if (!is_eth_multi(bp))
2070 * bp->multi_mode = ETH_RSS_MODE_DISABLED;
2073 params.rss_obj = rss_obj;
2075 __set_bit(RAMROD_COMP_WAIT, ¶ms.ramrod_flags);
2078 __set_bit(BNX2X_RSS_MODE_REGULAR, ¶ms.rss_flags);
2080 /* RSS configuration */
2081 __set_bit(BNX2X_RSS_IPV4, ¶ms.rss_flags);
2082 __set_bit(BNX2X_RSS_IPV4_TCP, ¶ms.rss_flags);
2083 __set_bit(BNX2X_RSS_IPV6, ¶ms.rss_flags);
2084 __set_bit(BNX2X_RSS_IPV6_TCP, ¶ms.rss_flags);
2085 if (rss_obj->udp_rss_v4)
2086 __set_bit(BNX2X_RSS_IPV4_UDP, ¶ms.rss_flags);
2087 if (rss_obj->udp_rss_v6)
2088 __set_bit(BNX2X_RSS_IPV6_UDP, ¶ms.rss_flags);
2090 if (!CHIP_IS_E1x(bp)) {
2091 /* valid only for TUNN_MODE_VXLAN tunnel mode */
2092 __set_bit(BNX2X_RSS_IPV4_VXLAN, ¶ms.rss_flags);
2093 __set_bit(BNX2X_RSS_IPV6_VXLAN, ¶ms.rss_flags);
2095 /* valid only for TUNN_MODE_GRE tunnel mode */
2096 __set_bit(BNX2X_RSS_TUNN_INNER_HDRS, ¶ms.rss_flags);
2099 __set_bit(BNX2X_RSS_MODE_DISABLED, ¶ms.rss_flags);
2103 params.rss_result_mask = MULTI_MASK;
2105 memcpy(params.ind_table, rss_obj->ind_table, sizeof(params.ind_table));
2109 netdev_rss_key_fill(params.rss_key, T_ETH_RSS_KEY * 4);
2110 __set_bit(BNX2X_RSS_SET_SRCH, ¶ms.rss_flags);
2114 return bnx2x_config_rss(bp, ¶ms);
2116 return bnx2x_vfpf_config_rss(bp, ¶ms);
2119 static int bnx2x_init_hw(struct bnx2x *bp, u32 load_code)
2121 struct bnx2x_func_state_params func_params = {NULL};
2123 /* Prepare parameters for function state transitions */
2124 __set_bit(RAMROD_COMP_WAIT, &func_params.ramrod_flags);
2126 func_params.f_obj = &bp->func_obj;
2127 func_params.cmd = BNX2X_F_CMD_HW_INIT;
2129 func_params.params.hw_init.load_phase = load_code;
2131 return bnx2x_func_state_change(bp, &func_params);
2135 * Cleans the object that have internal lists without sending
2136 * ramrods. Should be run when interrupts are disabled.
2138 void bnx2x_squeeze_objects(struct bnx2x *bp)
2141 unsigned long ramrod_flags = 0, vlan_mac_flags = 0;
2142 struct bnx2x_mcast_ramrod_params rparam = {NULL};
2143 struct bnx2x_vlan_mac_obj *mac_obj = &bp->sp_objs->mac_obj;
2145 /***************** Cleanup MACs' object first *************************/
2147 /* Wait for completion of requested */
2148 __set_bit(RAMROD_COMP_WAIT, &ramrod_flags);
2149 /* Perform a dry cleanup */
2150 __set_bit(RAMROD_DRV_CLR_ONLY, &ramrod_flags);
2152 /* Clean ETH primary MAC */
2153 __set_bit(BNX2X_ETH_MAC, &vlan_mac_flags);
2154 rc = mac_obj->delete_all(bp, &bp->sp_objs->mac_obj, &vlan_mac_flags,
2157 BNX2X_ERR("Failed to clean ETH MACs: %d\n", rc);
2159 /* Cleanup UC list */
2161 __set_bit(BNX2X_UC_LIST_MAC, &vlan_mac_flags);
2162 rc = mac_obj->delete_all(bp, mac_obj, &vlan_mac_flags,
2165 BNX2X_ERR("Failed to clean UC list MACs: %d\n", rc);
2167 /***************** Now clean mcast object *****************************/
2168 rparam.mcast_obj = &bp->mcast_obj;
2169 __set_bit(RAMROD_DRV_CLR_ONLY, &rparam.ramrod_flags);
2171 /* Add a DEL command... - Since we're doing a driver cleanup only,
2172 * we take a lock surrounding both the initial send and the CONTs,
2173 * as we don't want a true completion to disrupt us in the middle.
2175 netif_addr_lock_bh(bp->dev);
2176 rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_DEL);
2178 BNX2X_ERR("Failed to add a new DEL command to a multi-cast object: %d\n",
2181 /* ...and wait until all pending commands are cleared */
2182 rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_CONT);
2185 BNX2X_ERR("Failed to clean multi-cast object: %d\n",
2187 netif_addr_unlock_bh(bp->dev);
2191 rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_CONT);
2193 netif_addr_unlock_bh(bp->dev);
2196 #ifndef BNX2X_STOP_ON_ERROR
2197 #define LOAD_ERROR_EXIT(bp, label) \
2199 (bp)->state = BNX2X_STATE_ERROR; \
2203 #define LOAD_ERROR_EXIT_CNIC(bp, label) \
2205 bp->cnic_loaded = false; \
2208 #else /*BNX2X_STOP_ON_ERROR*/
2209 #define LOAD_ERROR_EXIT(bp, label) \
2211 (bp)->state = BNX2X_STATE_ERROR; \
2215 #define LOAD_ERROR_EXIT_CNIC(bp, label) \
2217 bp->cnic_loaded = false; \
2221 #endif /*BNX2X_STOP_ON_ERROR*/
2223 static void bnx2x_free_fw_stats_mem(struct bnx2x *bp)
2225 BNX2X_PCI_FREE(bp->fw_stats, bp->fw_stats_mapping,
2226 bp->fw_stats_data_sz + bp->fw_stats_req_sz);
2230 static int bnx2x_alloc_fw_stats_mem(struct bnx2x *bp)
2232 int num_groups, vf_headroom = 0;
2233 int is_fcoe_stats = NO_FCOE(bp) ? 0 : 1;
2235 /* number of queues for statistics is number of eth queues + FCoE */
2236 u8 num_queue_stats = BNX2X_NUM_ETH_QUEUES(bp) + is_fcoe_stats;
2238 /* Total number of FW statistics requests =
2239 * 1 for port stats + 1 for PF stats + potential 2 for FCoE (fcoe proper
2240 * and fcoe l2 queue) stats + num of queues (which includes another 1
2241 * for fcoe l2 queue if applicable)
2243 bp->fw_stats_num = 2 + is_fcoe_stats + num_queue_stats;
2245 /* vf stats appear in the request list, but their data is allocated by
2246 * the VFs themselves. We don't include them in the bp->fw_stats_num as
2247 * it is used to determine where to place the vf stats queries in the
2251 vf_headroom = bnx2x_vf_headroom(bp);
2253 /* Request is built from stats_query_header and an array of
2254 * stats_query_cmd_group each of which contains
2255 * STATS_QUERY_CMD_COUNT rules. The real number or requests is
2256 * configured in the stats_query_header.
2259 (((bp->fw_stats_num + vf_headroom) / STATS_QUERY_CMD_COUNT) +
2260 (((bp->fw_stats_num + vf_headroom) % STATS_QUERY_CMD_COUNT) ?
2263 DP(BNX2X_MSG_SP, "stats fw_stats_num %d, vf headroom %d, num_groups %d\n",
2264 bp->fw_stats_num, vf_headroom, num_groups);
2265 bp->fw_stats_req_sz = sizeof(struct stats_query_header) +
2266 num_groups * sizeof(struct stats_query_cmd_group);
2268 /* Data for statistics requests + stats_counter
2269 * stats_counter holds per-STORM counters that are incremented
2270 * when STORM has finished with the current request.
2271 * memory for FCoE offloaded statistics are counted anyway,
2272 * even if they will not be sent.
2273 * VF stats are not accounted for here as the data of VF stats is stored
2274 * in memory allocated by the VF, not here.
2276 bp->fw_stats_data_sz = sizeof(struct per_port_stats) +
2277 sizeof(struct per_pf_stats) +
2278 sizeof(struct fcoe_statistics_params) +
2279 sizeof(struct per_queue_stats) * num_queue_stats +
2280 sizeof(struct stats_counter);
2282 bp->fw_stats = BNX2X_PCI_ALLOC(&bp->fw_stats_mapping,
2283 bp->fw_stats_data_sz + bp->fw_stats_req_sz);
2288 bp->fw_stats_req = (struct bnx2x_fw_stats_req *)bp->fw_stats;
2289 bp->fw_stats_req_mapping = bp->fw_stats_mapping;
2290 bp->fw_stats_data = (struct bnx2x_fw_stats_data *)
2291 ((u8 *)bp->fw_stats + bp->fw_stats_req_sz);
2292 bp->fw_stats_data_mapping = bp->fw_stats_mapping +
2293 bp->fw_stats_req_sz;
2295 DP(BNX2X_MSG_SP, "statistics request base address set to %x %x\n",
2296 U64_HI(bp->fw_stats_req_mapping),
2297 U64_LO(bp->fw_stats_req_mapping));
2298 DP(BNX2X_MSG_SP, "statistics data base address set to %x %x\n",
2299 U64_HI(bp->fw_stats_data_mapping),
2300 U64_LO(bp->fw_stats_data_mapping));
2304 bnx2x_free_fw_stats_mem(bp);
2305 BNX2X_ERR("Can't allocate FW stats memory\n");
2309 /* send load request to mcp and analyze response */
2310 static int bnx2x_nic_load_request(struct bnx2x *bp, u32 *load_code)
2316 (SHMEM_RD(bp, func_mb[BP_FW_MB_IDX(bp)].drv_mb_header) &
2317 DRV_MSG_SEQ_NUMBER_MASK);
2318 BNX2X_DEV_INFO("fw_seq 0x%08x\n", bp->fw_seq);
2320 /* Get current FW pulse sequence */
2321 bp->fw_drv_pulse_wr_seq =
2322 (SHMEM_RD(bp, func_mb[BP_FW_MB_IDX(bp)].drv_pulse_mb) &
2323 DRV_PULSE_SEQ_MASK);
2324 BNX2X_DEV_INFO("drv_pulse 0x%x\n", bp->fw_drv_pulse_wr_seq);
2326 param = DRV_MSG_CODE_LOAD_REQ_WITH_LFA;
2328 if (IS_MF_SD(bp) && bnx2x_port_after_undi(bp))
2329 param |= DRV_MSG_CODE_LOAD_REQ_FORCE_LFA;
2332 (*load_code) = bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_REQ, param);
2334 /* if mcp fails to respond we must abort */
2335 if (!(*load_code)) {
2336 BNX2X_ERR("MCP response failure, aborting\n");
2340 /* If mcp refused (e.g. other port is in diagnostic mode) we
2343 if ((*load_code) == FW_MSG_CODE_DRV_LOAD_REFUSED) {
2344 BNX2X_ERR("MCP refused load request, aborting\n");
2350 /* check whether another PF has already loaded FW to chip. In
2351 * virtualized environments a pf from another VM may have already
2352 * initialized the device including loading FW
2354 int bnx2x_compare_fw_ver(struct bnx2x *bp, u32 load_code, bool print_err)
2356 /* is another pf loaded on this engine? */
2357 if (load_code != FW_MSG_CODE_DRV_LOAD_COMMON_CHIP &&
2358 load_code != FW_MSG_CODE_DRV_LOAD_COMMON) {
2359 /* build my FW version dword */
2362 /* read loaded FW from chip */
2363 u32 loaded_fw = REG_RD(bp, XSEM_REG_PRAM);
2365 u32 my_fw = ~loaded_fw;
2367 DP(BNX2X_MSG_SP, "loaded fw %x, my fw %x\n",
2370 /* abort nic load if version mismatch */
2371 if (my_fw != loaded_fw) {
2373 BNX2X_ERR("bnx2x with FW %x was already loaded which mismatches my %x FW. Aborting\n",
2376 BNX2X_DEV_INFO("bnx2x with FW %x was already loaded which mismatches my %x FW, possibly due to MF UNDI\n",
2384 /* returns the "mcp load_code" according to global load_count array */
2385 static int bnx2x_nic_load_no_mcp(struct bnx2x *bp, int port)
2387 int path = BP_PATH(bp);
2389 DP(NETIF_MSG_IFUP, "NO MCP - load counts[%d] %d, %d, %d\n",
2390 path, bnx2x_load_count[path][0], bnx2x_load_count[path][1],
2391 bnx2x_load_count[path][2]);
2392 bnx2x_load_count[path][0]++;
2393 bnx2x_load_count[path][1 + port]++;
2394 DP(NETIF_MSG_IFUP, "NO MCP - new load counts[%d] %d, %d, %d\n",
2395 path, bnx2x_load_count[path][0], bnx2x_load_count[path][1],
2396 bnx2x_load_count[path][2]);
2397 if (bnx2x_load_count[path][0] == 1)
2398 return FW_MSG_CODE_DRV_LOAD_COMMON;
2399 else if (bnx2x_load_count[path][1 + port] == 1)
2400 return FW_MSG_CODE_DRV_LOAD_PORT;
2402 return FW_MSG_CODE_DRV_LOAD_FUNCTION;
2405 /* mark PMF if applicable */
2406 static void bnx2x_nic_load_pmf(struct bnx2x *bp, u32 load_code)
2408 if ((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
2409 (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP) ||
2410 (load_code == FW_MSG_CODE_DRV_LOAD_PORT)) {
2412 /* We need the barrier to ensure the ordering between the
2413 * writing to bp->port.pmf here and reading it from the
2414 * bnx2x_periodic_task().
2421 DP(NETIF_MSG_LINK, "pmf %d\n", bp->port.pmf);
2424 static void bnx2x_nic_load_afex_dcc(struct bnx2x *bp, int load_code)
2426 if (((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
2427 (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP)) &&
2428 (bp->common.shmem2_base)) {
2429 if (SHMEM2_HAS(bp, dcc_support))
2430 SHMEM2_WR(bp, dcc_support,
2431 (SHMEM_DCC_SUPPORT_DISABLE_ENABLE_PF_TLV |
2432 SHMEM_DCC_SUPPORT_BANDWIDTH_ALLOCATION_TLV));
2433 if (SHMEM2_HAS(bp, afex_driver_support))
2434 SHMEM2_WR(bp, afex_driver_support,
2435 SHMEM_AFEX_SUPPORTED_VERSION_ONE);
2438 /* Set AFEX default VLAN tag to an invalid value */
2439 bp->afex_def_vlan_tag = -1;
2443 * bnx2x_bz_fp - zero content of the fastpath structure.
2445 * @bp: driver handle
2446 * @index: fastpath index to be zeroed
2448 * Makes sure the contents of the bp->fp[index].napi is kept
2451 static void bnx2x_bz_fp(struct bnx2x *bp, int index)
2453 struct bnx2x_fastpath *fp = &bp->fp[index];
2455 struct napi_struct orig_napi = fp->napi;
2456 struct bnx2x_agg_info *orig_tpa_info = fp->tpa_info;
2458 /* bzero bnx2x_fastpath contents */
2460 memset(fp->tpa_info, 0, ETH_MAX_AGGREGATION_QUEUES_E1H_E2 *
2461 sizeof(struct bnx2x_agg_info));
2462 memset(fp, 0, sizeof(*fp));
2464 /* Restore the NAPI object as it has been already initialized */
2465 fp->napi = orig_napi;
2466 fp->tpa_info = orig_tpa_info;
2470 fp->max_cos = bp->max_cos;
2472 /* Special queues support only one CoS */
2475 /* Init txdata pointers */
2477 fp->txdata_ptr[0] = &bp->bnx2x_txq[FCOE_TXQ_IDX(bp)];
2479 for_each_cos_in_tx_queue(fp, cos)
2480 fp->txdata_ptr[cos] = &bp->bnx2x_txq[cos *
2481 BNX2X_NUM_ETH_QUEUES(bp) + index];
2483 /* set the tpa flag for each queue. The tpa flag determines the queue
2484 * minimal size so it must be set prior to queue memory allocation
2486 if (bp->dev->features & NETIF_F_LRO)
2487 fp->mode = TPA_MODE_LRO;
2488 else if (bp->dev->features & NETIF_F_GRO &&
2489 bnx2x_mtu_allows_gro(bp->dev->mtu))
2490 fp->mode = TPA_MODE_GRO;
2492 fp->mode = TPA_MODE_DISABLED;
2494 /* We don't want TPA if it's disabled in bp
2495 * or if this is an FCoE L2 ring.
2497 if (bp->disable_tpa || IS_FCOE_FP(fp))
2498 fp->mode = TPA_MODE_DISABLED;
2501 void bnx2x_set_os_driver_state(struct bnx2x *bp, u32 state)
2505 if (!IS_MF_BD(bp) || !SHMEM2_HAS(bp, os_driver_state) || IS_VF(bp))
2508 cur = SHMEM2_RD(bp, os_driver_state[BP_FW_MB_IDX(bp)]);
2509 DP(NETIF_MSG_IFUP, "Driver state %08x-->%08x\n",
2512 SHMEM2_WR(bp, os_driver_state[BP_FW_MB_IDX(bp)], state);
2515 int bnx2x_load_cnic(struct bnx2x *bp)
2517 int i, rc, port = BP_PORT(bp);
2519 DP(NETIF_MSG_IFUP, "Starting CNIC-related load\n");
2521 mutex_init(&bp->cnic_mutex);
2524 rc = bnx2x_alloc_mem_cnic(bp);
2526 BNX2X_ERR("Unable to allocate bp memory for cnic\n");
2527 LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic0);
2531 rc = bnx2x_alloc_fp_mem_cnic(bp);
2533 BNX2X_ERR("Unable to allocate memory for cnic fps\n");
2534 LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic0);
2537 /* Update the number of queues with the cnic queues */
2538 rc = bnx2x_set_real_num_queues(bp, 1);
2540 BNX2X_ERR("Unable to set real_num_queues including cnic\n");
2541 LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic0);
2544 /* Add all CNIC NAPI objects */
2545 bnx2x_add_all_napi_cnic(bp);
2546 DP(NETIF_MSG_IFUP, "cnic napi added\n");
2547 bnx2x_napi_enable_cnic(bp);
2549 rc = bnx2x_init_hw_func_cnic(bp);
2551 LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic1);
2553 bnx2x_nic_init_cnic(bp);
2556 /* Enable Timer scan */
2557 REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + port*4, 1);
2559 /* setup cnic queues */
2560 for_each_cnic_queue(bp, i) {
2561 rc = bnx2x_setup_queue(bp, &bp->fp[i], 0);
2563 BNX2X_ERR("Queue setup failed\n");
2564 LOAD_ERROR_EXIT(bp, load_error_cnic2);
2569 /* Initialize Rx filter. */
2570 bnx2x_set_rx_mode_inner(bp);
2572 /* re-read iscsi info */
2573 bnx2x_get_iscsi_info(bp);
2574 bnx2x_setup_cnic_irq_info(bp);
2575 bnx2x_setup_cnic_info(bp);
2576 bp->cnic_loaded = true;
2577 if (bp->state == BNX2X_STATE_OPEN)
2578 bnx2x_cnic_notify(bp, CNIC_CTL_START_CMD);
2580 DP(NETIF_MSG_IFUP, "Ending successfully CNIC-related load\n");
2584 #ifndef BNX2X_STOP_ON_ERROR
2586 /* Disable Timer scan */
2587 REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + port*4, 0);
2590 bnx2x_napi_disable_cnic(bp);
2591 /* Update the number of queues without the cnic queues */
2592 if (bnx2x_set_real_num_queues(bp, 0))
2593 BNX2X_ERR("Unable to set real_num_queues not including cnic\n");
2595 BNX2X_ERR("CNIC-related load failed\n");
2596 bnx2x_free_fp_mem_cnic(bp);
2597 bnx2x_free_mem_cnic(bp);
2599 #endif /* ! BNX2X_STOP_ON_ERROR */
2602 /* must be called with rtnl_lock */
2603 int bnx2x_nic_load(struct bnx2x *bp, int load_mode)
2605 int port = BP_PORT(bp);
2606 int i, rc = 0, load_code = 0;
2608 DP(NETIF_MSG_IFUP, "Starting NIC load\n");
2610 "CNIC is %s\n", CNIC_ENABLED(bp) ? "enabled" : "disabled");
2612 #ifdef BNX2X_STOP_ON_ERROR
2613 if (unlikely(bp->panic)) {
2614 BNX2X_ERR("Can't load NIC when there is panic\n");
2619 bp->state = BNX2X_STATE_OPENING_WAIT4_LOAD;
2621 /* zero the structure w/o any lock, before SP handler is initialized */
2622 memset(&bp->last_reported_link, 0, sizeof(bp->last_reported_link));
2623 __set_bit(BNX2X_LINK_REPORT_LINK_DOWN,
2624 &bp->last_reported_link.link_report_flags);
2627 /* must be called before memory allocation and HW init */
2628 bnx2x_ilt_set_info(bp);
2631 * Zero fastpath structures preserving invariants like napi, which are
2632 * allocated only once, fp index, max_cos, bp pointer.
2633 * Also set fp->mode and txdata_ptr.
2635 DP(NETIF_MSG_IFUP, "num queues: %d", bp->num_queues);
2636 for_each_queue(bp, i)
2638 memset(bp->bnx2x_txq, 0, (BNX2X_MAX_RSS_COUNT(bp) * BNX2X_MULTI_TX_COS +
2639 bp->num_cnic_queues) *
2640 sizeof(struct bnx2x_fp_txdata));
2642 bp->fcoe_init = false;
2644 /* Set the receive queues buffer size */
2645 bnx2x_set_rx_buf_size(bp);
2648 rc = bnx2x_alloc_mem(bp);
2650 BNX2X_ERR("Unable to allocate bp memory\n");
2655 /* need to be done after alloc mem, since it's self adjusting to amount
2656 * of memory available for RSS queues
2658 rc = bnx2x_alloc_fp_mem(bp);
2660 BNX2X_ERR("Unable to allocate memory for fps\n");
2661 LOAD_ERROR_EXIT(bp, load_error0);
2664 /* Allocated memory for FW statistics */
2665 if (bnx2x_alloc_fw_stats_mem(bp))
2666 LOAD_ERROR_EXIT(bp, load_error0);
2668 /* request pf to initialize status blocks */
2670 rc = bnx2x_vfpf_init(bp);
2672 LOAD_ERROR_EXIT(bp, load_error0);
2675 /* As long as bnx2x_alloc_mem() may possibly update
2676 * bp->num_queues, bnx2x_set_real_num_queues() should always
2677 * come after it. At this stage cnic queues are not counted.
2679 rc = bnx2x_set_real_num_queues(bp, 0);
2681 BNX2X_ERR("Unable to set real_num_queues\n");
2682 LOAD_ERROR_EXIT(bp, load_error0);
2685 /* configure multi cos mappings in kernel.
2686 * this configuration may be overridden by a multi class queue
2687 * discipline or by a dcbx negotiation result.
2689 bnx2x_setup_tc(bp->dev, bp->max_cos);
2691 /* Add all NAPI objects */
2692 bnx2x_add_all_napi(bp);
2693 DP(NETIF_MSG_IFUP, "napi added\n");
2694 bnx2x_napi_enable(bp);
2697 /* set pf load just before approaching the MCP */
2698 bnx2x_set_pf_load(bp);
2700 /* if mcp exists send load request and analyze response */
2701 if (!BP_NOMCP(bp)) {
2702 /* attempt to load pf */
2703 rc = bnx2x_nic_load_request(bp, &load_code);
2705 LOAD_ERROR_EXIT(bp, load_error1);
2707 /* what did mcp say? */
2708 rc = bnx2x_compare_fw_ver(bp, load_code, true);
2710 bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2711 LOAD_ERROR_EXIT(bp, load_error2);
2714 load_code = bnx2x_nic_load_no_mcp(bp, port);
2717 /* mark pmf if applicable */
2718 bnx2x_nic_load_pmf(bp, load_code);
2720 /* Init Function state controlling object */
2721 bnx2x__init_func_obj(bp);
2724 rc = bnx2x_init_hw(bp, load_code);
2726 BNX2X_ERR("HW init failed, aborting\n");
2727 bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2728 LOAD_ERROR_EXIT(bp, load_error2);
2732 bnx2x_pre_irq_nic_init(bp);
2734 /* Connect to IRQs */
2735 rc = bnx2x_setup_irqs(bp);
2737 BNX2X_ERR("setup irqs failed\n");
2739 bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2740 LOAD_ERROR_EXIT(bp, load_error2);
2743 /* Init per-function objects */
2745 /* Setup NIC internals and enable interrupts */
2746 bnx2x_post_irq_nic_init(bp, load_code);
2748 bnx2x_init_bp_objs(bp);
2749 bnx2x_iov_nic_init(bp);
2751 /* Set AFEX default VLAN tag to an invalid value */
2752 bp->afex_def_vlan_tag = -1;
2753 bnx2x_nic_load_afex_dcc(bp, load_code);
2754 bp->state = BNX2X_STATE_OPENING_WAIT4_PORT;
2755 rc = bnx2x_func_start(bp);
2757 BNX2X_ERR("Function start failed!\n");
2758 bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2760 LOAD_ERROR_EXIT(bp, load_error3);
2763 /* Send LOAD_DONE command to MCP */
2764 if (!BP_NOMCP(bp)) {
2765 load_code = bnx2x_fw_command(bp,
2766 DRV_MSG_CODE_LOAD_DONE, 0);
2768 BNX2X_ERR("MCP response failure, aborting\n");
2770 LOAD_ERROR_EXIT(bp, load_error3);
2774 /* initialize FW coalescing state machines in RAM */
2775 bnx2x_update_coalesce(bp);
2778 /* setup the leading queue */
2779 rc = bnx2x_setup_leading(bp);
2781 BNX2X_ERR("Setup leading failed!\n");
2782 LOAD_ERROR_EXIT(bp, load_error3);
2785 /* set up the rest of the queues */
2786 for_each_nondefault_eth_queue(bp, i) {
2788 rc = bnx2x_setup_queue(bp, &bp->fp[i], false);
2790 rc = bnx2x_vfpf_setup_q(bp, &bp->fp[i], false);
2792 BNX2X_ERR("Queue %d setup failed\n", i);
2793 LOAD_ERROR_EXIT(bp, load_error3);
2798 rc = bnx2x_init_rss(bp);
2800 BNX2X_ERR("PF RSS init failed\n");
2801 LOAD_ERROR_EXIT(bp, load_error3);
2804 /* Now when Clients are configured we are ready to work */
2805 bp->state = BNX2X_STATE_OPEN;
2807 /* Configure a ucast MAC */
2809 rc = bnx2x_set_eth_mac(bp, true);
2811 rc = bnx2x_vfpf_config_mac(bp, bp->dev->dev_addr, bp->fp->index,
2814 BNX2X_ERR("Setting Ethernet MAC failed\n");
2815 LOAD_ERROR_EXIT(bp, load_error3);
2818 if (IS_PF(bp) && bp->pending_max) {
2819 bnx2x_update_max_mf_config(bp, bp->pending_max);
2820 bp->pending_max = 0;
2824 rc = bnx2x_initial_phy_init(bp, load_mode);
2826 LOAD_ERROR_EXIT(bp, load_error3);
2828 bp->link_params.feature_config_flags &= ~FEATURE_CONFIG_BOOT_FROM_SAN;
2830 /* Start fast path */
2832 /* Re-configure vlan filters */
2833 rc = bnx2x_vlan_reconfigure_vid(bp);
2835 LOAD_ERROR_EXIT(bp, load_error3);
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 val &= ~DRV_FLAGS_MTU_MASK;
2884 val |= (bp->dev->mtu << DRV_FLAGS_MTU_SHIFT);
2885 SHMEM2_WR(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)],
2886 val | DRV_FLAGS_CAPABILITIES_LOADED_SUPPORTED |
2887 DRV_FLAGS_CAPABILITIES_LOADED_L2);
2890 /* Wait for all pending SP commands to complete */
2891 if (IS_PF(bp) && !bnx2x_wait_sp_comp(bp, ~0x0UL)) {
2892 BNX2X_ERR("Timeout waiting for SP elements to complete\n");
2893 bnx2x_nic_unload(bp, UNLOAD_CLOSE, false);
2897 /* Update driver data for On-Chip MFW dump. */
2899 bnx2x_update_mfw_dump(bp);
2901 /* If PMF - send ADMIN DCBX msg to MFW to initiate DCBX FSM */
2902 if (bp->port.pmf && (bp->state != BNX2X_STATE_DIAG))
2903 bnx2x_dcbx_init(bp, false);
2905 if (!IS_MF_SD_STORAGE_PERSONALITY_ONLY(bp))
2906 bnx2x_set_os_driver_state(bp, OS_DRIVER_STATE_ACTIVE);
2908 DP(NETIF_MSG_IFUP, "Ending successfully NIC load\n");
2912 #ifndef BNX2X_STOP_ON_ERROR
2915 bnx2x_int_disable_sync(bp, 1);
2917 /* Clean queueable objects */
2918 bnx2x_squeeze_objects(bp);
2921 /* Free SKBs, SGEs, TPA pool and driver internals */
2922 bnx2x_free_skbs(bp);
2923 for_each_rx_queue(bp, i)
2924 bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
2929 if (IS_PF(bp) && !BP_NOMCP(bp)) {
2930 bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_REQ_WOL_MCP, 0);
2931 bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_DONE, 0);
2936 bnx2x_napi_disable(bp);
2937 bnx2x_del_all_napi(bp);
2939 /* clear pf_load status, as it was already set */
2941 bnx2x_clear_pf_load(bp);
2943 bnx2x_free_fw_stats_mem(bp);
2944 bnx2x_free_fp_mem(bp);
2948 #endif /* ! BNX2X_STOP_ON_ERROR */
2951 int bnx2x_drain_tx_queues(struct bnx2x *bp)
2955 /* Wait until tx fastpath tasks complete */
2956 for_each_tx_queue(bp, i) {
2957 struct bnx2x_fastpath *fp = &bp->fp[i];
2959 for_each_cos_in_tx_queue(fp, cos)
2960 rc = bnx2x_clean_tx_queue(bp, fp->txdata_ptr[cos]);
2967 /* must be called with rtnl_lock */
2968 int bnx2x_nic_unload(struct bnx2x *bp, int unload_mode, bool keep_link)
2971 bool global = false;
2973 DP(NETIF_MSG_IFUP, "Starting NIC unload\n");
2975 if (!IS_MF_SD_STORAGE_PERSONALITY_ONLY(bp))
2976 bnx2x_set_os_driver_state(bp, OS_DRIVER_STATE_DISABLED);
2978 /* mark driver is unloaded in shmem2 */
2979 if (IS_PF(bp) && SHMEM2_HAS(bp, drv_capabilities_flag)) {
2981 val = SHMEM2_RD(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)]);
2982 SHMEM2_WR(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)],
2983 val & ~DRV_FLAGS_CAPABILITIES_LOADED_L2);
2986 if (IS_PF(bp) && bp->recovery_state != BNX2X_RECOVERY_DONE &&
2987 (bp->state == BNX2X_STATE_CLOSED ||
2988 bp->state == BNX2X_STATE_ERROR)) {
2989 /* We can get here if the driver has been unloaded
2990 * during parity error recovery and is either waiting for a
2991 * leader to complete or for other functions to unload and
2992 * then ifdown has been issued. In this case we want to
2993 * unload and let other functions to complete a recovery
2996 bp->recovery_state = BNX2X_RECOVERY_DONE;
2998 bnx2x_release_leader_lock(bp);
3001 DP(NETIF_MSG_IFDOWN, "Releasing a leadership...\n");
3002 BNX2X_ERR("Can't unload in closed or error state\n");
3006 /* Nothing to do during unload if previous bnx2x_nic_load()
3007 * have not completed successfully - all resources are released.
3009 * we can get here only after unsuccessful ndo_* callback, during which
3010 * dev->IFF_UP flag is still on.
3012 if (bp->state == BNX2X_STATE_CLOSED || bp->state == BNX2X_STATE_ERROR)
3015 /* It's important to set the bp->state to the value different from
3016 * BNX2X_STATE_OPEN and only then stop the Tx. Otherwise bnx2x_tx_int()
3017 * may restart the Tx from the NAPI context (see bnx2x_tx_int()).
3019 bp->state = BNX2X_STATE_CLOSING_WAIT4_HALT;
3022 /* indicate to VFs that the PF is going down */
3023 bnx2x_iov_channel_down(bp);
3025 if (CNIC_LOADED(bp))
3026 bnx2x_cnic_notify(bp, CNIC_CTL_STOP_CMD);
3029 bnx2x_tx_disable(bp);
3030 netdev_reset_tc(bp->dev);
3032 bp->rx_mode = BNX2X_RX_MODE_NONE;
3034 del_timer_sync(&bp->timer);
3037 /* Set ALWAYS_ALIVE bit in shmem */
3038 bp->fw_drv_pulse_wr_seq |= DRV_PULSE_ALWAYS_ALIVE;
3039 bnx2x_drv_pulse(bp);
3040 bnx2x_stats_handle(bp, STATS_EVENT_STOP);
3041 bnx2x_save_statistics(bp);
3044 /* wait till consumers catch up with producers in all queues.
3045 * If we're recovering, FW can't write to host so no reason
3046 * to wait for the queues to complete all Tx.
3048 if (unload_mode != UNLOAD_RECOVERY)
3049 bnx2x_drain_tx_queues(bp);
3051 /* if VF indicate to PF this function is going down (PF will delete sp
3052 * elements and clear initializations
3055 bnx2x_vfpf_close_vf(bp);
3056 else if (unload_mode != UNLOAD_RECOVERY)
3057 /* if this is a normal/close unload need to clean up chip*/
3058 bnx2x_chip_cleanup(bp, unload_mode, keep_link);
3060 /* Send the UNLOAD_REQUEST to the MCP */
3061 bnx2x_send_unload_req(bp, unload_mode);
3063 /* Prevent transactions to host from the functions on the
3064 * engine that doesn't reset global blocks in case of global
3065 * attention once global blocks are reset and gates are opened
3066 * (the engine which leader will perform the recovery
3069 if (!CHIP_IS_E1x(bp))
3070 bnx2x_pf_disable(bp);
3072 /* Disable HW interrupts, NAPI */
3073 bnx2x_netif_stop(bp, 1);
3074 /* Delete all NAPI objects */
3075 bnx2x_del_all_napi(bp);
3076 if (CNIC_LOADED(bp))
3077 bnx2x_del_all_napi_cnic(bp);
3081 /* Report UNLOAD_DONE to MCP */
3082 bnx2x_send_unload_done(bp, false);
3086 * At this stage no more interrupts will arrive so we may safely clean
3087 * the queueable objects here in case they failed to get cleaned so far.
3090 bnx2x_squeeze_objects(bp);
3092 /* There should be no more pending SP commands at this stage */
3097 /* clear pending work in rtnl task */
3098 bp->sp_rtnl_state = 0;
3101 /* Free SKBs, SGEs, TPA pool and driver internals */
3102 bnx2x_free_skbs(bp);
3103 if (CNIC_LOADED(bp))
3104 bnx2x_free_skbs_cnic(bp);
3105 for_each_rx_queue(bp, i)
3106 bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
3108 bnx2x_free_fp_mem(bp);
3109 if (CNIC_LOADED(bp))
3110 bnx2x_free_fp_mem_cnic(bp);
3113 if (CNIC_LOADED(bp))
3114 bnx2x_free_mem_cnic(bp);
3118 bp->state = BNX2X_STATE_CLOSED;
3119 bp->cnic_loaded = false;
3121 /* Clear driver version indication in shmem */
3123 bnx2x_update_mng_version(bp);
3125 /* Check if there are pending parity attentions. If there are - set
3126 * RECOVERY_IN_PROGRESS.
3128 if (IS_PF(bp) && bnx2x_chk_parity_attn(bp, &global, false)) {
3129 bnx2x_set_reset_in_progress(bp);
3131 /* Set RESET_IS_GLOBAL if needed */
3133 bnx2x_set_reset_global(bp);
3136 /* The last driver must disable a "close the gate" if there is no
3137 * parity attention or "process kill" pending.
3140 !bnx2x_clear_pf_load(bp) &&
3141 bnx2x_reset_is_done(bp, BP_PATH(bp)))
3142 bnx2x_disable_close_the_gate(bp);
3144 DP(NETIF_MSG_IFUP, "Ending NIC unload\n");
3149 int bnx2x_set_power_state(struct bnx2x *bp, pci_power_t state)
3153 /* If there is no power capability, silently succeed */
3154 if (!bp->pdev->pm_cap) {
3155 BNX2X_DEV_INFO("No power capability. Breaking.\n");
3159 pci_read_config_word(bp->pdev, bp->pdev->pm_cap + PCI_PM_CTRL, &pmcsr);
3163 pci_write_config_word(bp->pdev, bp->pdev->pm_cap + PCI_PM_CTRL,
3164 ((pmcsr & ~PCI_PM_CTRL_STATE_MASK) |
3165 PCI_PM_CTRL_PME_STATUS));
3167 if (pmcsr & PCI_PM_CTRL_STATE_MASK)
3168 /* delay required during transition out of D3hot */
3173 /* If there are other clients above don't
3174 shut down the power */
3175 if (atomic_read(&bp->pdev->enable_cnt) != 1)
3177 /* Don't shut down the power for emulation and FPGA */
3178 if (CHIP_REV_IS_SLOW(bp))
3181 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
3185 pmcsr |= PCI_PM_CTRL_PME_ENABLE;
3187 pci_write_config_word(bp->pdev, bp->pdev->pm_cap + PCI_PM_CTRL,
3190 /* No more memory access after this point until
3191 * device is brought back to D0.
3196 dev_err(&bp->pdev->dev, "Can't support state = %d\n", state);
3203 * net_device service functions
3205 static int bnx2x_poll(struct napi_struct *napi, int budget)
3207 struct bnx2x_fastpath *fp = container_of(napi, struct bnx2x_fastpath,
3209 struct bnx2x *bp = fp->bp;
3213 #ifdef BNX2X_STOP_ON_ERROR
3214 if (unlikely(bp->panic)) {
3215 napi_complete(napi);
3219 for_each_cos_in_tx_queue(fp, cos)
3220 if (bnx2x_tx_queue_has_work(fp->txdata_ptr[cos]))
3221 bnx2x_tx_int(bp, fp->txdata_ptr[cos]);
3223 rx_work_done = (bnx2x_has_rx_work(fp)) ? bnx2x_rx_int(fp, budget) : 0;
3225 if (rx_work_done < budget) {
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);
3233 bnx2x_update_fpsb_idx(fp);
3234 /* bnx2x_has_rx_work() reads the status block,
3235 * thus we need to ensure that status block indices
3236 * have been actually read (bnx2x_update_fpsb_idx)
3237 * prior to this check (bnx2x_has_rx_work) so that
3238 * we won't write the "newer" value of the status block
3239 * to IGU (if there was a DMA right after
3240 * bnx2x_has_rx_work and if there is no rmb, the memory
3241 * reading (bnx2x_update_fpsb_idx) may be postponed
3242 * to right before bnx2x_ack_sb). In this case there
3243 * will never be another interrupt until there is
3244 * another update of the status block, while there
3245 * is still unhandled work.
3249 if (!(bnx2x_has_rx_work(fp) || bnx2x_has_tx_work(fp))) {
3250 napi_complete(napi);
3251 /* Re-enable interrupts */
3252 DP(NETIF_MSG_RX_STATUS,
3253 "Update index to %d\n", fp->fp_hc_idx);
3254 bnx2x_ack_sb(bp, fp->igu_sb_id, USTORM_ID,
3255 le16_to_cpu(fp->fp_hc_idx),
3258 rx_work_done = budget;
3263 return rx_work_done;
3266 /* we split the first BD into headers and data BDs
3267 * to ease the pain of our fellow microcode engineers
3268 * we use one mapping for both BDs
3270 static u16 bnx2x_tx_split(struct bnx2x *bp,
3271 struct bnx2x_fp_txdata *txdata,
3272 struct sw_tx_bd *tx_buf,
3273 struct eth_tx_start_bd **tx_bd, u16 hlen,
3276 struct eth_tx_start_bd *h_tx_bd = *tx_bd;
3277 struct eth_tx_bd *d_tx_bd;
3279 int old_len = le16_to_cpu(h_tx_bd->nbytes);
3281 /* first fix first BD */
3282 h_tx_bd->nbytes = cpu_to_le16(hlen);
3284 DP(NETIF_MSG_TX_QUEUED, "TSO split header size is %d (%x:%x)\n",
3285 h_tx_bd->nbytes, h_tx_bd->addr_hi, h_tx_bd->addr_lo);
3287 /* now get a new data BD
3288 * (after the pbd) and fill it */
3289 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3290 d_tx_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
3292 mapping = HILO_U64(le32_to_cpu(h_tx_bd->addr_hi),
3293 le32_to_cpu(h_tx_bd->addr_lo)) + hlen;
3295 d_tx_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
3296 d_tx_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
3297 d_tx_bd->nbytes = cpu_to_le16(old_len - hlen);
3299 /* this marks the BD as one that has no individual mapping */
3300 tx_buf->flags |= BNX2X_TSO_SPLIT_BD;
3302 DP(NETIF_MSG_TX_QUEUED,
3303 "TSO split data size is %d (%x:%x)\n",
3304 d_tx_bd->nbytes, d_tx_bd->addr_hi, d_tx_bd->addr_lo);
3307 *tx_bd = (struct eth_tx_start_bd *)d_tx_bd;
3312 #define bswab32(b32) ((__force __le32) swab32((__force __u32) (b32)))
3313 #define bswab16(b16) ((__force __le16) swab16((__force __u16) (b16)))
3314 static __le16 bnx2x_csum_fix(unsigned char *t_header, u16 csum, s8 fix)
3316 __sum16 tsum = (__force __sum16) csum;
3319 tsum = ~csum_fold(csum_sub((__force __wsum) csum,
3320 csum_partial(t_header - fix, fix, 0)));
3323 tsum = ~csum_fold(csum_add((__force __wsum) csum,
3324 csum_partial(t_header, -fix, 0)));
3326 return bswab16(tsum);
3329 static u32 bnx2x_xmit_type(struct bnx2x *bp, struct sk_buff *skb)
3335 if (skb->ip_summed != CHECKSUM_PARTIAL)
3338 protocol = vlan_get_protocol(skb);
3339 if (protocol == htons(ETH_P_IPV6)) {
3341 prot = ipv6_hdr(skb)->nexthdr;
3344 prot = ip_hdr(skb)->protocol;
3347 if (!CHIP_IS_E1x(bp) && skb->encapsulation) {
3348 if (inner_ip_hdr(skb)->version == 6) {
3349 rc |= XMIT_CSUM_ENC_V6;
3350 if (inner_ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3351 rc |= XMIT_CSUM_TCP;
3353 rc |= XMIT_CSUM_ENC_V4;
3354 if (inner_ip_hdr(skb)->protocol == IPPROTO_TCP)
3355 rc |= XMIT_CSUM_TCP;
3358 if (prot == IPPROTO_TCP)
3359 rc |= XMIT_CSUM_TCP;
3361 if (skb_is_gso(skb)) {
3362 if (skb_is_gso_v6(skb)) {
3363 rc |= (XMIT_GSO_V6 | XMIT_CSUM_TCP);
3364 if (rc & XMIT_CSUM_ENC)
3365 rc |= XMIT_GSO_ENC_V6;
3367 rc |= (XMIT_GSO_V4 | XMIT_CSUM_TCP);
3368 if (rc & XMIT_CSUM_ENC)
3369 rc |= XMIT_GSO_ENC_V4;
3376 /* VXLAN: 4 = 1 (for linear data BD) + 3 (2 for PBD and last BD) */
3377 #define BNX2X_NUM_VXLAN_TSO_WIN_SUB_BDS 4
3379 /* Regular: 3 = 1 (for linear data BD) + 2 (for PBD and last BD) */
3380 #define BNX2X_NUM_TSO_WIN_SUB_BDS 3
3382 #if (MAX_SKB_FRAGS >= MAX_FETCH_BD - BDS_PER_TX_PKT)
3383 /* check if packet requires linearization (packet is too fragmented)
3384 no need to check fragmentation if page size > 8K (there will be no
3385 violation to FW restrictions) */
3386 static int bnx2x_pkt_req_lin(struct bnx2x *bp, struct sk_buff *skb,
3389 int first_bd_sz = 0, num_tso_win_sub = BNX2X_NUM_TSO_WIN_SUB_BDS;
3390 int to_copy = 0, hlen = 0;
3392 if (xmit_type & XMIT_GSO_ENC)
3393 num_tso_win_sub = BNX2X_NUM_VXLAN_TSO_WIN_SUB_BDS;
3395 if (skb_shinfo(skb)->nr_frags >= (MAX_FETCH_BD - num_tso_win_sub)) {
3396 if (xmit_type & XMIT_GSO) {
3397 unsigned short lso_mss = skb_shinfo(skb)->gso_size;
3398 int wnd_size = MAX_FETCH_BD - num_tso_win_sub;
3399 /* Number of windows to check */
3400 int num_wnds = skb_shinfo(skb)->nr_frags - wnd_size;
3405 /* Headers length */
3406 if (xmit_type & XMIT_GSO_ENC)
3407 hlen = (int)(skb_inner_transport_header(skb) -
3409 inner_tcp_hdrlen(skb);
3411 hlen = (int)(skb_transport_header(skb) -
3412 skb->data) + tcp_hdrlen(skb);
3414 /* Amount of data (w/o headers) on linear part of SKB*/
3415 first_bd_sz = skb_headlen(skb) - hlen;
3417 wnd_sum = first_bd_sz;
3419 /* Calculate the first sum - it's special */
3420 for (frag_idx = 0; frag_idx < wnd_size - 1; frag_idx++)
3422 skb_frag_size(&skb_shinfo(skb)->frags[frag_idx]);
3424 /* If there was data on linear skb data - check it */
3425 if (first_bd_sz > 0) {
3426 if (unlikely(wnd_sum < lso_mss)) {
3431 wnd_sum -= first_bd_sz;
3434 /* Others are easier: run through the frag list and
3435 check all windows */
3436 for (wnd_idx = 0; wnd_idx <= num_wnds; wnd_idx++) {
3438 skb_frag_size(&skb_shinfo(skb)->frags[wnd_idx + wnd_size - 1]);
3440 if (unlikely(wnd_sum < lso_mss)) {
3445 skb_frag_size(&skb_shinfo(skb)->frags[wnd_idx]);
3448 /* in non-LSO too fragmented packet should always
3455 if (unlikely(to_copy))
3456 DP(NETIF_MSG_TX_QUEUED,
3457 "Linearization IS REQUIRED for %s packet. num_frags %d hlen %d first_bd_sz %d\n",
3458 (xmit_type & XMIT_GSO) ? "LSO" : "non-LSO",
3459 skb_shinfo(skb)->nr_frags, hlen, first_bd_sz);
3466 * bnx2x_set_pbd_gso - update PBD in GSO case.
3470 * @xmit_type: xmit flags
3472 static void bnx2x_set_pbd_gso(struct sk_buff *skb,
3473 struct eth_tx_parse_bd_e1x *pbd,
3476 pbd->lso_mss = cpu_to_le16(skb_shinfo(skb)->gso_size);
3477 pbd->tcp_send_seq = bswab32(tcp_hdr(skb)->seq);
3478 pbd->tcp_flags = pbd_tcp_flags(tcp_hdr(skb));
3480 if (xmit_type & XMIT_GSO_V4) {
3481 pbd->ip_id = bswab16(ip_hdr(skb)->id);
3482 pbd->tcp_pseudo_csum =
3483 bswab16(~csum_tcpudp_magic(ip_hdr(skb)->saddr,
3485 0, IPPROTO_TCP, 0));
3487 pbd->tcp_pseudo_csum =
3488 bswab16(~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3489 &ipv6_hdr(skb)->daddr,
3490 0, IPPROTO_TCP, 0));
3494 cpu_to_le16(ETH_TX_PARSE_BD_E1X_PSEUDO_CS_WITHOUT_LEN);
3498 * bnx2x_set_pbd_csum_enc - update PBD with checksum and return header length
3500 * @bp: driver handle
3502 * @parsing_data: data to be updated
3503 * @xmit_type: xmit flags
3505 * 57712/578xx related, when skb has encapsulation
3507 static u8 bnx2x_set_pbd_csum_enc(struct bnx2x *bp, struct sk_buff *skb,
3508 u32 *parsing_data, u32 xmit_type)
3511 ((((u8 *)skb_inner_transport_header(skb) - skb->data) >> 1) <<
3512 ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W_SHIFT) &
3513 ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W;
3515 if (xmit_type & XMIT_CSUM_TCP) {
3516 *parsing_data |= ((inner_tcp_hdrlen(skb) / 4) <<
3517 ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW_SHIFT) &
3518 ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW;
3520 return skb_inner_transport_header(skb) +
3521 inner_tcp_hdrlen(skb) - skb->data;
3524 /* We support checksum offload for TCP and UDP only.
3525 * No need to pass the UDP header length - it's a constant.
3527 return skb_inner_transport_header(skb) +
3528 sizeof(struct udphdr) - skb->data;
3532 * bnx2x_set_pbd_csum_e2 - update PBD with checksum and return header length
3534 * @bp: driver handle
3536 * @parsing_data: data to be updated
3537 * @xmit_type: xmit flags
3539 * 57712/578xx related
3541 static u8 bnx2x_set_pbd_csum_e2(struct bnx2x *bp, struct sk_buff *skb,
3542 u32 *parsing_data, u32 xmit_type)
3545 ((((u8 *)skb_transport_header(skb) - skb->data) >> 1) <<
3546 ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W_SHIFT) &
3547 ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W;
3549 if (xmit_type & XMIT_CSUM_TCP) {
3550 *parsing_data |= ((tcp_hdrlen(skb) / 4) <<
3551 ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW_SHIFT) &
3552 ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW;
3554 return skb_transport_header(skb) + tcp_hdrlen(skb) - skb->data;
3556 /* We support checksum offload for TCP and UDP only.
3557 * No need to pass the UDP header length - it's a constant.
3559 return skb_transport_header(skb) + sizeof(struct udphdr) - skb->data;
3562 /* set FW indication according to inner or outer protocols if tunneled */
3563 static void bnx2x_set_sbd_csum(struct bnx2x *bp, struct sk_buff *skb,
3564 struct eth_tx_start_bd *tx_start_bd,
3567 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_L4_CSUM;
3569 if (xmit_type & (XMIT_CSUM_ENC_V6 | XMIT_CSUM_V6))
3570 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_IPV6;
3572 if (!(xmit_type & XMIT_CSUM_TCP))
3573 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_IS_UDP;
3577 * bnx2x_set_pbd_csum - update PBD with checksum and return header length
3579 * @bp: driver handle
3581 * @pbd: parse BD to be updated
3582 * @xmit_type: xmit flags
3584 static u8 bnx2x_set_pbd_csum(struct bnx2x *bp, struct sk_buff *skb,
3585 struct eth_tx_parse_bd_e1x *pbd,
3588 u8 hlen = (skb_network_header(skb) - skb->data) >> 1;
3590 /* for now NS flag is not used in Linux */
3593 ((skb->protocol == cpu_to_be16(ETH_P_8021Q)) <<
3594 ETH_TX_PARSE_BD_E1X_LLC_SNAP_EN_SHIFT));
3596 pbd->ip_hlen_w = (skb_transport_header(skb) -
3597 skb_network_header(skb)) >> 1;
3599 hlen += pbd->ip_hlen_w;
3601 /* We support checksum offload for TCP and UDP only */
3602 if (xmit_type & XMIT_CSUM_TCP)
3603 hlen += tcp_hdrlen(skb) / 2;
3605 hlen += sizeof(struct udphdr) / 2;
3607 pbd->total_hlen_w = cpu_to_le16(hlen);
3610 if (xmit_type & XMIT_CSUM_TCP) {
3611 pbd->tcp_pseudo_csum = bswab16(tcp_hdr(skb)->check);
3614 s8 fix = SKB_CS_OFF(skb); /* signed! */
3616 DP(NETIF_MSG_TX_QUEUED,
3617 "hlen %d fix %d csum before fix %x\n",
3618 le16_to_cpu(pbd->total_hlen_w), fix, SKB_CS(skb));
3620 /* HW bug: fixup the CSUM */
3621 pbd->tcp_pseudo_csum =
3622 bnx2x_csum_fix(skb_transport_header(skb),
3625 DP(NETIF_MSG_TX_QUEUED, "csum after fix %x\n",
3626 pbd->tcp_pseudo_csum);
3632 static void bnx2x_update_pbds_gso_enc(struct sk_buff *skb,
3633 struct eth_tx_parse_bd_e2 *pbd_e2,
3634 struct eth_tx_parse_2nd_bd *pbd2,
3639 u8 outerip_off, outerip_len = 0;
3641 /* from outer IP to transport */
3642 hlen_w = (skb_inner_transport_header(skb) -
3643 skb_network_header(skb)) >> 1;
3646 hlen_w += inner_tcp_hdrlen(skb) >> 1;
3648 pbd2->fw_ip_hdr_to_payload_w = hlen_w;
3650 /* outer IP header info */
3651 if (xmit_type & XMIT_CSUM_V4) {
3652 struct iphdr *iph = ip_hdr(skb);
3653 u32 csum = (__force u32)(~iph->check) -
3654 (__force u32)iph->tot_len -
3655 (__force u32)iph->frag_off;
3657 outerip_len = iph->ihl << 1;
3659 pbd2->fw_ip_csum_wo_len_flags_frag =
3660 bswab16(csum_fold((__force __wsum)csum));
3662 pbd2->fw_ip_hdr_to_payload_w =
3663 hlen_w - ((sizeof(struct ipv6hdr)) >> 1);
3664 pbd_e2->data.tunnel_data.flags |=
3665 ETH_TUNNEL_DATA_IPV6_OUTER;
3668 pbd2->tcp_send_seq = bswab32(inner_tcp_hdr(skb)->seq);
3670 pbd2->tcp_flags = pbd_tcp_flags(inner_tcp_hdr(skb));
3672 /* inner IP header info */
3673 if (xmit_type & XMIT_CSUM_ENC_V4) {
3674 pbd2->hw_ip_id = bswab16(inner_ip_hdr(skb)->id);
3676 pbd_e2->data.tunnel_data.pseudo_csum =
3677 bswab16(~csum_tcpudp_magic(
3678 inner_ip_hdr(skb)->saddr,
3679 inner_ip_hdr(skb)->daddr,
3680 0, IPPROTO_TCP, 0));
3682 pbd_e2->data.tunnel_data.pseudo_csum =
3683 bswab16(~csum_ipv6_magic(
3684 &inner_ipv6_hdr(skb)->saddr,
3685 &inner_ipv6_hdr(skb)->daddr,
3686 0, IPPROTO_TCP, 0));
3689 outerip_off = (skb_network_header(skb) - skb->data) >> 1;
3694 ETH_TX_PARSE_2ND_BD_IP_HDR_LEN_OUTER_W_SHIFT) |
3695 ((skb->protocol == cpu_to_be16(ETH_P_8021Q)) <<
3696 ETH_TX_PARSE_2ND_BD_LLC_SNAP_EN_SHIFT);
3698 if (ip_hdr(skb)->protocol == IPPROTO_UDP) {
3699 SET_FLAG(*global_data, ETH_TX_PARSE_2ND_BD_TUNNEL_UDP_EXIST, 1);
3700 pbd2->tunnel_udp_hdr_start_w = skb_transport_offset(skb) >> 1;
3704 static inline void bnx2x_set_ipv6_ext_e2(struct sk_buff *skb, u32 *parsing_data,
3707 struct ipv6hdr *ipv6;
3709 if (!(xmit_type & (XMIT_GSO_ENC_V6 | XMIT_GSO_V6)))
3712 if (xmit_type & XMIT_GSO_ENC_V6)
3713 ipv6 = inner_ipv6_hdr(skb);
3714 else /* XMIT_GSO_V6 */
3715 ipv6 = ipv6_hdr(skb);
3717 if (ipv6->nexthdr == NEXTHDR_IPV6)
3718 *parsing_data |= ETH_TX_PARSE_BD_E2_IPV6_WITH_EXT_HDR;
3721 /* called with netif_tx_lock
3722 * bnx2x_tx_int() runs without netif_tx_lock unless it needs to call
3723 * netif_wake_queue()
3725 netdev_tx_t bnx2x_start_xmit(struct sk_buff *skb, struct net_device *dev)
3727 struct bnx2x *bp = netdev_priv(dev);
3729 struct netdev_queue *txq;
3730 struct bnx2x_fp_txdata *txdata;
3731 struct sw_tx_bd *tx_buf;
3732 struct eth_tx_start_bd *tx_start_bd, *first_bd;
3733 struct eth_tx_bd *tx_data_bd, *total_pkt_bd = NULL;
3734 struct eth_tx_parse_bd_e1x *pbd_e1x = NULL;
3735 struct eth_tx_parse_bd_e2 *pbd_e2 = NULL;
3736 struct eth_tx_parse_2nd_bd *pbd2 = NULL;
3737 u32 pbd_e2_parsing_data = 0;
3738 u16 pkt_prod, bd_prod;
3741 u32 xmit_type = bnx2x_xmit_type(bp, skb);
3744 __le16 pkt_size = 0;
3746 u8 mac_type = UNICAST_ADDRESS;
3748 #ifdef BNX2X_STOP_ON_ERROR
3749 if (unlikely(bp->panic))
3750 return NETDEV_TX_BUSY;
3753 txq_index = skb_get_queue_mapping(skb);
3754 txq = netdev_get_tx_queue(dev, txq_index);
3756 BUG_ON(txq_index >= MAX_ETH_TXQ_IDX(bp) + (CNIC_LOADED(bp) ? 1 : 0));
3758 txdata = &bp->bnx2x_txq[txq_index];
3760 /* enable this debug print to view the transmission queue being used
3761 DP(NETIF_MSG_TX_QUEUED, "indices: txq %d, fp %d, txdata %d\n",
3762 txq_index, fp_index, txdata_index); */
3764 /* enable this debug print to view the transmission details
3765 DP(NETIF_MSG_TX_QUEUED,
3766 "transmitting packet cid %d fp index %d txdata_index %d tx_data ptr %p fp pointer %p\n",
3767 txdata->cid, fp_index, txdata_index, txdata, fp); */
3769 if (unlikely(bnx2x_tx_avail(bp, txdata) <
3770 skb_shinfo(skb)->nr_frags +
3772 NEXT_CNT_PER_TX_PKT(MAX_BDS_PER_TX_PKT))) {
3773 /* Handle special storage cases separately */
3774 if (txdata->tx_ring_size == 0) {
3775 struct bnx2x_eth_q_stats *q_stats =
3776 bnx2x_fp_qstats(bp, txdata->parent_fp);
3777 q_stats->driver_filtered_tx_pkt++;
3779 return NETDEV_TX_OK;
3781 bnx2x_fp_qstats(bp, txdata->parent_fp)->driver_xoff++;
3782 netif_tx_stop_queue(txq);
3783 BNX2X_ERR("BUG! Tx ring full when queue awake!\n");
3785 return NETDEV_TX_BUSY;
3788 DP(NETIF_MSG_TX_QUEUED,
3789 "queue[%d]: SKB: summed %x protocol %x protocol(%x,%x) gso type %x xmit_type %x len %d\n",
3790 txq_index, skb->ip_summed, skb->protocol, ipv6_hdr(skb)->nexthdr,
3791 ip_hdr(skb)->protocol, skb_shinfo(skb)->gso_type, xmit_type,
3794 eth = (struct ethhdr *)skb->data;
3796 /* set flag according to packet type (UNICAST_ADDRESS is default)*/
3797 if (unlikely(is_multicast_ether_addr(eth->h_dest))) {
3798 if (is_broadcast_ether_addr(eth->h_dest))
3799 mac_type = BROADCAST_ADDRESS;
3801 mac_type = MULTICAST_ADDRESS;
3804 #if (MAX_SKB_FRAGS >= MAX_FETCH_BD - BDS_PER_TX_PKT)
3805 /* First, check if we need to linearize the skb (due to FW
3806 restrictions). No need to check fragmentation if page size > 8K
3807 (there will be no violation to FW restrictions) */
3808 if (bnx2x_pkt_req_lin(bp, skb, xmit_type)) {
3809 /* Statistics of linearization */
3811 if (skb_linearize(skb) != 0) {
3812 DP(NETIF_MSG_TX_QUEUED,
3813 "SKB linearization failed - silently dropping this SKB\n");
3814 dev_kfree_skb_any(skb);
3815 return NETDEV_TX_OK;
3819 /* Map skb linear data for DMA */
3820 mapping = dma_map_single(&bp->pdev->dev, skb->data,
3821 skb_headlen(skb), DMA_TO_DEVICE);
3822 if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
3823 DP(NETIF_MSG_TX_QUEUED,
3824 "SKB mapping failed - silently dropping this SKB\n");
3825 dev_kfree_skb_any(skb);
3826 return NETDEV_TX_OK;
3829 Please read carefully. First we use one BD which we mark as start,
3830 then we have a parsing info BD (used for TSO or xsum),
3831 and only then we have the rest of the TSO BDs.
3832 (don't forget to mark the last one as last,
3833 and to unmap only AFTER you write to the BD ...)
3834 And above all, all pdb sizes are in words - NOT DWORDS!
3837 /* get current pkt produced now - advance it just before sending packet
3838 * since mapping of pages may fail and cause packet to be dropped
3840 pkt_prod = txdata->tx_pkt_prod;
3841 bd_prod = TX_BD(txdata->tx_bd_prod);
3843 /* get a tx_buf and first BD
3844 * tx_start_bd may be changed during SPLIT,
3845 * but first_bd will always stay first
3847 tx_buf = &txdata->tx_buf_ring[TX_BD(pkt_prod)];
3848 tx_start_bd = &txdata->tx_desc_ring[bd_prod].start_bd;
3849 first_bd = tx_start_bd;
3851 tx_start_bd->bd_flags.as_bitfield = ETH_TX_BD_FLAGS_START_BD;
3853 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) {
3854 if (!(bp->flags & TX_TIMESTAMPING_EN)) {
3855 BNX2X_ERR("Tx timestamping was not enabled, this packet will not be timestamped\n");
3856 } else if (bp->ptp_tx_skb) {
3857 BNX2X_ERR("The device supports only a single outstanding packet to timestamp, this packet will not be timestamped\n");
3859 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
3860 /* schedule check for Tx timestamp */
3861 bp->ptp_tx_skb = skb_get(skb);
3862 bp->ptp_tx_start = jiffies;
3863 schedule_work(&bp->ptp_task);
3867 /* header nbd: indirectly zero other flags! */
3868 tx_start_bd->general_data = 1 << ETH_TX_START_BD_HDR_NBDS_SHIFT;
3870 /* remember the first BD of the packet */
3871 tx_buf->first_bd = txdata->tx_bd_prod;
3875 DP(NETIF_MSG_TX_QUEUED,
3876 "sending pkt %u @%p next_idx %u bd %u @%p\n",
3877 pkt_prod, tx_buf, txdata->tx_pkt_prod, bd_prod, tx_start_bd);
3879 if (skb_vlan_tag_present(skb)) {
3880 tx_start_bd->vlan_or_ethertype =
3881 cpu_to_le16(skb_vlan_tag_get(skb));
3882 tx_start_bd->bd_flags.as_bitfield |=
3883 (X_ETH_OUTBAND_VLAN << ETH_TX_BD_FLAGS_VLAN_MODE_SHIFT);
3885 /* when transmitting in a vf, start bd must hold the ethertype
3886 * for fw to enforce it
3888 #ifndef BNX2X_STOP_ON_ERROR
3891 tx_start_bd->vlan_or_ethertype =
3892 cpu_to_le16(ntohs(eth->h_proto));
3893 #ifndef BNX2X_STOP_ON_ERROR
3895 /* used by FW for packet accounting */
3896 tx_start_bd->vlan_or_ethertype = cpu_to_le16(pkt_prod);
3900 nbd = 2; /* start_bd + pbd + frags (updated when pages are mapped) */
3902 /* turn on parsing and get a BD */
3903 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3905 if (xmit_type & XMIT_CSUM)
3906 bnx2x_set_sbd_csum(bp, skb, tx_start_bd, xmit_type);
3908 if (!CHIP_IS_E1x(bp)) {
3909 pbd_e2 = &txdata->tx_desc_ring[bd_prod].parse_bd_e2;
3910 memset(pbd_e2, 0, sizeof(struct eth_tx_parse_bd_e2));
3912 if (xmit_type & XMIT_CSUM_ENC) {
3913 u16 global_data = 0;
3915 /* Set PBD in enc checksum offload case */
3916 hlen = bnx2x_set_pbd_csum_enc(bp, skb,
3917 &pbd_e2_parsing_data,
3920 /* turn on 2nd parsing and get a BD */
3921 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3923 pbd2 = &txdata->tx_desc_ring[bd_prod].parse_2nd_bd;
3925 memset(pbd2, 0, sizeof(*pbd2));
3927 pbd_e2->data.tunnel_data.ip_hdr_start_inner_w =
3928 (skb_inner_network_header(skb) -
3931 if (xmit_type & XMIT_GSO_ENC)
3932 bnx2x_update_pbds_gso_enc(skb, pbd_e2, pbd2,
3936 pbd2->global_data = cpu_to_le16(global_data);
3938 /* add addition parse BD indication to start BD */
3939 SET_FLAG(tx_start_bd->general_data,
3940 ETH_TX_START_BD_PARSE_NBDS, 1);
3941 /* set encapsulation flag in start BD */
3942 SET_FLAG(tx_start_bd->general_data,
3943 ETH_TX_START_BD_TUNNEL_EXIST, 1);
3945 tx_buf->flags |= BNX2X_HAS_SECOND_PBD;
3948 } else if (xmit_type & XMIT_CSUM) {
3949 /* Set PBD in checksum offload case w/o encapsulation */
3950 hlen = bnx2x_set_pbd_csum_e2(bp, skb,
3951 &pbd_e2_parsing_data,
3955 bnx2x_set_ipv6_ext_e2(skb, &pbd_e2_parsing_data, xmit_type);
3956 /* Add the macs to the parsing BD if this is a vf or if
3957 * Tx Switching is enabled.
3960 /* override GRE parameters in BD */
3961 bnx2x_set_fw_mac_addr(&pbd_e2->data.mac_addr.src_hi,
3962 &pbd_e2->data.mac_addr.src_mid,
3963 &pbd_e2->data.mac_addr.src_lo,
3966 bnx2x_set_fw_mac_addr(&pbd_e2->data.mac_addr.dst_hi,
3967 &pbd_e2->data.mac_addr.dst_mid,
3968 &pbd_e2->data.mac_addr.dst_lo,
3971 if (bp->flags & TX_SWITCHING)
3972 bnx2x_set_fw_mac_addr(
3973 &pbd_e2->data.mac_addr.dst_hi,
3974 &pbd_e2->data.mac_addr.dst_mid,
3975 &pbd_e2->data.mac_addr.dst_lo,
3977 #ifdef BNX2X_STOP_ON_ERROR
3978 /* Enforce security is always set in Stop on Error -
3979 * source mac should be present in the parsing BD
3981 bnx2x_set_fw_mac_addr(&pbd_e2->data.mac_addr.src_hi,
3982 &pbd_e2->data.mac_addr.src_mid,
3983 &pbd_e2->data.mac_addr.src_lo,
3988 SET_FLAG(pbd_e2_parsing_data,
3989 ETH_TX_PARSE_BD_E2_ETH_ADDR_TYPE, mac_type);
3991 u16 global_data = 0;
3992 pbd_e1x = &txdata->tx_desc_ring[bd_prod].parse_bd_e1x;
3993 memset(pbd_e1x, 0, sizeof(struct eth_tx_parse_bd_e1x));
3994 /* Set PBD in checksum offload case */
3995 if (xmit_type & XMIT_CSUM)
3996 hlen = bnx2x_set_pbd_csum(bp, skb, pbd_e1x, xmit_type);
3998 SET_FLAG(global_data,
3999 ETH_TX_PARSE_BD_E1X_ETH_ADDR_TYPE, mac_type);
4000 pbd_e1x->global_data |= cpu_to_le16(global_data);
4003 /* Setup the data pointer of the first BD of the packet */
4004 tx_start_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
4005 tx_start_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
4006 tx_start_bd->nbytes = cpu_to_le16(skb_headlen(skb));
4007 pkt_size = tx_start_bd->nbytes;
4009 DP(NETIF_MSG_TX_QUEUED,
4010 "first bd @%p addr (%x:%x) nbytes %d flags %x vlan %x\n",
4011 tx_start_bd, tx_start_bd->addr_hi, tx_start_bd->addr_lo,
4012 le16_to_cpu(tx_start_bd->nbytes),
4013 tx_start_bd->bd_flags.as_bitfield,
4014 le16_to_cpu(tx_start_bd->vlan_or_ethertype));
4016 if (xmit_type & XMIT_GSO) {
4018 DP(NETIF_MSG_TX_QUEUED,
4019 "TSO packet len %d hlen %d total len %d tso size %d\n",
4020 skb->len, hlen, skb_headlen(skb),
4021 skb_shinfo(skb)->gso_size);
4023 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_SW_LSO;
4025 if (unlikely(skb_headlen(skb) > hlen)) {
4027 bd_prod = bnx2x_tx_split(bp, txdata, tx_buf,
4031 if (!CHIP_IS_E1x(bp))
4032 pbd_e2_parsing_data |=
4033 (skb_shinfo(skb)->gso_size <<
4034 ETH_TX_PARSE_BD_E2_LSO_MSS_SHIFT) &
4035 ETH_TX_PARSE_BD_E2_LSO_MSS;
4037 bnx2x_set_pbd_gso(skb, pbd_e1x, xmit_type);
4040 /* Set the PBD's parsing_data field if not zero
4041 * (for the chips newer than 57711).
4043 if (pbd_e2_parsing_data)
4044 pbd_e2->parsing_data = cpu_to_le32(pbd_e2_parsing_data);
4046 tx_data_bd = (struct eth_tx_bd *)tx_start_bd;
4048 /* Handle fragmented skb */
4049 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
4050 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
4052 mapping = skb_frag_dma_map(&bp->pdev->dev, frag, 0,
4053 skb_frag_size(frag), DMA_TO_DEVICE);
4054 if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
4055 unsigned int pkts_compl = 0, bytes_compl = 0;
4057 DP(NETIF_MSG_TX_QUEUED,
4058 "Unable to map page - dropping packet...\n");
4060 /* we need unmap all buffers already mapped
4062 * first_bd->nbd need to be properly updated
4063 * before call to bnx2x_free_tx_pkt
4065 first_bd->nbd = cpu_to_le16(nbd);
4066 bnx2x_free_tx_pkt(bp, txdata,
4067 TX_BD(txdata->tx_pkt_prod),
4068 &pkts_compl, &bytes_compl);
4069 return NETDEV_TX_OK;
4072 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
4073 tx_data_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
4074 if (total_pkt_bd == NULL)
4075 total_pkt_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
4077 tx_data_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
4078 tx_data_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
4079 tx_data_bd->nbytes = cpu_to_le16(skb_frag_size(frag));
4080 le16_add_cpu(&pkt_size, skb_frag_size(frag));
4083 DP(NETIF_MSG_TX_QUEUED,
4084 "frag %d bd @%p addr (%x:%x) nbytes %d\n",
4085 i, tx_data_bd, tx_data_bd->addr_hi, tx_data_bd->addr_lo,
4086 le16_to_cpu(tx_data_bd->nbytes));
4089 DP(NETIF_MSG_TX_QUEUED, "last bd @%p\n", tx_data_bd);
4091 /* update with actual num BDs */
4092 first_bd->nbd = cpu_to_le16(nbd);
4094 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
4096 /* now send a tx doorbell, counting the next BD
4097 * if the packet contains or ends with it
4099 if (TX_BD_POFF(bd_prod) < nbd)
4102 /* total_pkt_bytes should be set on the first data BD if
4103 * it's not an LSO packet and there is more than one
4104 * data BD. In this case pkt_size is limited by an MTU value.
4105 * However we prefer to set it for an LSO packet (while we don't
4106 * have to) in order to save some CPU cycles in a none-LSO
4107 * case, when we much more care about them.
4109 if (total_pkt_bd != NULL)
4110 total_pkt_bd->total_pkt_bytes = pkt_size;
4113 DP(NETIF_MSG_TX_QUEUED,
4114 "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",
4115 pbd_e1x, pbd_e1x->global_data, pbd_e1x->ip_hlen_w,
4116 pbd_e1x->ip_id, pbd_e1x->lso_mss, pbd_e1x->tcp_flags,
4117 pbd_e1x->tcp_pseudo_csum, pbd_e1x->tcp_send_seq,
4118 le16_to_cpu(pbd_e1x->total_hlen_w));
4120 DP(NETIF_MSG_TX_QUEUED,
4121 "PBD (E2) @%p dst %x %x %x src %x %x %x parsing_data %x\n",
4123 pbd_e2->data.mac_addr.dst_hi,
4124 pbd_e2->data.mac_addr.dst_mid,
4125 pbd_e2->data.mac_addr.dst_lo,
4126 pbd_e2->data.mac_addr.src_hi,
4127 pbd_e2->data.mac_addr.src_mid,
4128 pbd_e2->data.mac_addr.src_lo,
4129 pbd_e2->parsing_data);
4130 DP(NETIF_MSG_TX_QUEUED, "doorbell: nbd %d bd %u\n", nbd, bd_prod);
4132 netdev_tx_sent_queue(txq, skb->len);
4134 skb_tx_timestamp(skb);
4136 txdata->tx_pkt_prod++;
4138 * Make sure that the BD data is updated before updating the producer
4139 * since FW might read the BD right after the producer is updated.
4140 * This is only applicable for weak-ordered memory model archs such
4141 * as IA-64. The following barrier is also mandatory since FW will
4142 * assumes packets must have BDs.
4146 txdata->tx_db.data.prod += nbd;
4149 DOORBELL(bp, txdata->cid, txdata->tx_db.raw);
4153 txdata->tx_bd_prod += nbd;
4155 if (unlikely(bnx2x_tx_avail(bp, txdata) < MAX_DESC_PER_TX_PKT)) {
4156 netif_tx_stop_queue(txq);
4158 /* paired memory barrier is in bnx2x_tx_int(), we have to keep
4159 * ordering of set_bit() in netif_tx_stop_queue() and read of
4163 bnx2x_fp_qstats(bp, txdata->parent_fp)->driver_xoff++;
4164 if (bnx2x_tx_avail(bp, txdata) >= MAX_DESC_PER_TX_PKT)
4165 netif_tx_wake_queue(txq);
4169 return NETDEV_TX_OK;
4172 void bnx2x_get_c2s_mapping(struct bnx2x *bp, u8 *c2s_map, u8 *c2s_default)
4174 int mfw_vn = BP_FW_MB_IDX(bp);
4177 /* If the shmem shouldn't affect configuration, reflect */
4178 if (!IS_MF_BD(bp)) {
4181 for (i = 0; i < BNX2X_MAX_PRIORITY; i++)
4188 tmp = SHMEM2_RD(bp, c2s_pcp_map_lower[mfw_vn]);
4189 tmp = (__force u32)be32_to_cpu((__force __be32)tmp);
4190 c2s_map[0] = tmp & 0xff;
4191 c2s_map[1] = (tmp >> 8) & 0xff;
4192 c2s_map[2] = (tmp >> 16) & 0xff;
4193 c2s_map[3] = (tmp >> 24) & 0xff;
4195 tmp = SHMEM2_RD(bp, c2s_pcp_map_upper[mfw_vn]);
4196 tmp = (__force u32)be32_to_cpu((__force __be32)tmp);
4197 c2s_map[4] = tmp & 0xff;
4198 c2s_map[5] = (tmp >> 8) & 0xff;
4199 c2s_map[6] = (tmp >> 16) & 0xff;
4200 c2s_map[7] = (tmp >> 24) & 0xff;
4202 tmp = SHMEM2_RD(bp, c2s_pcp_map_default[mfw_vn]);
4203 tmp = (__force u32)be32_to_cpu((__force __be32)tmp);
4204 *c2s_default = (tmp >> (8 * mfw_vn)) & 0xff;
4208 * bnx2x_setup_tc - routine to configure net_device for multi tc
4210 * @netdev: net device to configure
4211 * @tc: number of traffic classes to enable
4213 * callback connected to the ndo_setup_tc function pointer
4215 int bnx2x_setup_tc(struct net_device *dev, u8 num_tc)
4217 struct bnx2x *bp = netdev_priv(dev);
4218 u8 c2s_map[BNX2X_MAX_PRIORITY], c2s_def;
4219 int cos, prio, count, offset;
4221 /* setup tc must be called under rtnl lock */
4224 /* no traffic classes requested. Aborting */
4226 netdev_reset_tc(dev);
4230 /* requested to support too many traffic classes */
4231 if (num_tc > bp->max_cos) {
4232 BNX2X_ERR("support for too many traffic classes requested: %d. Max supported is %d\n",
4233 num_tc, bp->max_cos);
4237 /* declare amount of supported traffic classes */
4238 if (netdev_set_num_tc(dev, num_tc)) {
4239 BNX2X_ERR("failed to declare %d traffic classes\n", num_tc);
4243 bnx2x_get_c2s_mapping(bp, c2s_map, &c2s_def);
4245 /* configure priority to traffic class mapping */
4246 for (prio = 0; prio < BNX2X_MAX_PRIORITY; prio++) {
4247 int outer_prio = c2s_map[prio];
4249 netdev_set_prio_tc_map(dev, prio, bp->prio_to_cos[outer_prio]);
4250 DP(BNX2X_MSG_SP | NETIF_MSG_IFUP,
4251 "mapping priority %d to tc %d\n",
4252 outer_prio, bp->prio_to_cos[outer_prio]);
4255 /* Use this configuration to differentiate tc0 from other COSes
4256 This can be used for ets or pfc, and save the effort of setting
4257 up a multio class queue disc or negotiating DCBX with a switch
4258 netdev_set_prio_tc_map(dev, 0, 0);
4259 DP(BNX2X_MSG_SP, "mapping priority %d to tc %d\n", 0, 0);
4260 for (prio = 1; prio < 16; prio++) {
4261 netdev_set_prio_tc_map(dev, prio, 1);
4262 DP(BNX2X_MSG_SP, "mapping priority %d to tc %d\n", prio, 1);
4265 /* configure traffic class to transmission queue mapping */
4266 for (cos = 0; cos < bp->max_cos; cos++) {
4267 count = BNX2X_NUM_ETH_QUEUES(bp);
4268 offset = cos * BNX2X_NUM_NON_CNIC_QUEUES(bp);
4269 netdev_set_tc_queue(dev, cos, count, offset);
4270 DP(BNX2X_MSG_SP | NETIF_MSG_IFUP,
4271 "mapping tc %d to offset %d count %d\n",
4272 cos, offset, count);
4278 int __bnx2x_setup_tc(struct net_device *dev, u32 handle, __be16 proto,
4279 struct tc_to_netdev *tc)
4281 if (tc->type != TC_SETUP_MQPRIO)
4283 return bnx2x_setup_tc(dev, tc->tc);
4286 /* called with rtnl_lock */
4287 int bnx2x_change_mac_addr(struct net_device *dev, void *p)
4289 struct sockaddr *addr = p;
4290 struct bnx2x *bp = netdev_priv(dev);
4293 if (!is_valid_ether_addr(addr->sa_data)) {
4294 BNX2X_ERR("Requested MAC address is not valid\n");
4298 if (IS_MF_STORAGE_ONLY(bp)) {
4299 BNX2X_ERR("Can't change address on STORAGE ONLY function\n");
4303 if (netif_running(dev)) {
4304 rc = bnx2x_set_eth_mac(bp, false);
4309 memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
4311 if (netif_running(dev))
4312 rc = bnx2x_set_eth_mac(bp, true);
4314 if (IS_PF(bp) && SHMEM2_HAS(bp, curr_cfg))
4315 SHMEM2_WR(bp, curr_cfg, CURR_CFG_MET_OS);
4320 static void bnx2x_free_fp_mem_at(struct bnx2x *bp, int fp_index)
4322 union host_hc_status_block *sb = &bnx2x_fp(bp, fp_index, status_blk);
4323 struct bnx2x_fastpath *fp = &bp->fp[fp_index];
4328 if (IS_FCOE_IDX(fp_index)) {
4329 memset(sb, 0, sizeof(union host_hc_status_block));
4330 fp->status_blk_mapping = 0;
4333 if (!CHIP_IS_E1x(bp))
4334 BNX2X_PCI_FREE(sb->e2_sb,
4335 bnx2x_fp(bp, fp_index,
4336 status_blk_mapping),
4337 sizeof(struct host_hc_status_block_e2));
4339 BNX2X_PCI_FREE(sb->e1x_sb,
4340 bnx2x_fp(bp, fp_index,
4341 status_blk_mapping),
4342 sizeof(struct host_hc_status_block_e1x));
4346 if (!skip_rx_queue(bp, fp_index)) {
4347 bnx2x_free_rx_bds(fp);
4349 /* fastpath rx rings: rx_buf rx_desc rx_comp */
4350 BNX2X_FREE(bnx2x_fp(bp, fp_index, rx_buf_ring));
4351 BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_desc_ring),
4352 bnx2x_fp(bp, fp_index, rx_desc_mapping),
4353 sizeof(struct eth_rx_bd) * NUM_RX_BD);
4355 BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_comp_ring),
4356 bnx2x_fp(bp, fp_index, rx_comp_mapping),
4357 sizeof(struct eth_fast_path_rx_cqe) *
4361 BNX2X_FREE(bnx2x_fp(bp, fp_index, rx_page_ring));
4362 BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_sge_ring),
4363 bnx2x_fp(bp, fp_index, rx_sge_mapping),
4364 BCM_PAGE_SIZE * NUM_RX_SGE_PAGES);
4368 if (!skip_tx_queue(bp, fp_index)) {
4369 /* fastpath tx rings: tx_buf tx_desc */
4370 for_each_cos_in_tx_queue(fp, cos) {
4371 struct bnx2x_fp_txdata *txdata = fp->txdata_ptr[cos];
4373 DP(NETIF_MSG_IFDOWN,
4374 "freeing tx memory of fp %d cos %d cid %d\n",
4375 fp_index, cos, txdata->cid);
4377 BNX2X_FREE(txdata->tx_buf_ring);
4378 BNX2X_PCI_FREE(txdata->tx_desc_ring,
4379 txdata->tx_desc_mapping,
4380 sizeof(union eth_tx_bd_types) * NUM_TX_BD);
4383 /* end of fastpath */
4386 static void bnx2x_free_fp_mem_cnic(struct bnx2x *bp)
4389 for_each_cnic_queue(bp, i)
4390 bnx2x_free_fp_mem_at(bp, i);
4393 void bnx2x_free_fp_mem(struct bnx2x *bp)
4396 for_each_eth_queue(bp, i)
4397 bnx2x_free_fp_mem_at(bp, i);
4400 static void set_sb_shortcuts(struct bnx2x *bp, int index)
4402 union host_hc_status_block status_blk = bnx2x_fp(bp, index, status_blk);
4403 if (!CHIP_IS_E1x(bp)) {
4404 bnx2x_fp(bp, index, sb_index_values) =
4405 (__le16 *)status_blk.e2_sb->sb.index_values;
4406 bnx2x_fp(bp, index, sb_running_index) =
4407 (__le16 *)status_blk.e2_sb->sb.running_index;
4409 bnx2x_fp(bp, index, sb_index_values) =
4410 (__le16 *)status_blk.e1x_sb->sb.index_values;
4411 bnx2x_fp(bp, index, sb_running_index) =
4412 (__le16 *)status_blk.e1x_sb->sb.running_index;
4416 /* Returns the number of actually allocated BDs */
4417 static int bnx2x_alloc_rx_bds(struct bnx2x_fastpath *fp,
4420 struct bnx2x *bp = fp->bp;
4421 u16 ring_prod, cqe_ring_prod;
4422 int i, failure_cnt = 0;
4424 fp->rx_comp_cons = 0;
4425 cqe_ring_prod = ring_prod = 0;
4427 /* This routine is called only during fo init so
4428 * fp->eth_q_stats.rx_skb_alloc_failed = 0
4430 for (i = 0; i < rx_ring_size; i++) {
4431 if (bnx2x_alloc_rx_data(bp, fp, ring_prod, GFP_KERNEL) < 0) {
4435 ring_prod = NEXT_RX_IDX(ring_prod);
4436 cqe_ring_prod = NEXT_RCQ_IDX(cqe_ring_prod);
4437 WARN_ON(ring_prod <= (i - failure_cnt));
4441 BNX2X_ERR("was only able to allocate %d rx skbs on queue[%d]\n",
4442 i - failure_cnt, fp->index);
4444 fp->rx_bd_prod = ring_prod;
4445 /* Limit the CQE producer by the CQE ring size */
4446 fp->rx_comp_prod = min_t(u16, NUM_RCQ_RINGS*RCQ_DESC_CNT,
4449 bnx2x_fp_stats(bp, fp)->eth_q_stats.rx_skb_alloc_failed += failure_cnt;
4451 return i - failure_cnt;
4454 static void bnx2x_set_next_page_rx_cq(struct bnx2x_fastpath *fp)
4458 for (i = 1; i <= NUM_RCQ_RINGS; i++) {
4459 struct eth_rx_cqe_next_page *nextpg;
4461 nextpg = (struct eth_rx_cqe_next_page *)
4462 &fp->rx_comp_ring[RCQ_DESC_CNT * i - 1];
4464 cpu_to_le32(U64_HI(fp->rx_comp_mapping +
4465 BCM_PAGE_SIZE*(i % NUM_RCQ_RINGS)));
4467 cpu_to_le32(U64_LO(fp->rx_comp_mapping +
4468 BCM_PAGE_SIZE*(i % NUM_RCQ_RINGS)));
4472 static int bnx2x_alloc_fp_mem_at(struct bnx2x *bp, int index)
4474 union host_hc_status_block *sb;
4475 struct bnx2x_fastpath *fp = &bp->fp[index];
4478 int rx_ring_size = 0;
4480 if (!bp->rx_ring_size && IS_MF_STORAGE_ONLY(bp)) {
4481 rx_ring_size = MIN_RX_SIZE_NONTPA;
4482 bp->rx_ring_size = rx_ring_size;
4483 } else if (!bp->rx_ring_size) {
4484 rx_ring_size = MAX_RX_AVAIL/BNX2X_NUM_RX_QUEUES(bp);
4486 if (CHIP_IS_E3(bp)) {
4487 u32 cfg = SHMEM_RD(bp,
4488 dev_info.port_hw_config[BP_PORT(bp)].
4491 /* Decrease ring size for 1G functions */
4492 if ((cfg & PORT_HW_CFG_NET_SERDES_IF_MASK) ==
4493 PORT_HW_CFG_NET_SERDES_IF_SGMII)
4497 /* allocate at least number of buffers required by FW */
4498 rx_ring_size = max_t(int, bp->disable_tpa ? MIN_RX_SIZE_NONTPA :
4499 MIN_RX_SIZE_TPA, rx_ring_size);
4501 bp->rx_ring_size = rx_ring_size;
4502 } else /* if rx_ring_size specified - use it */
4503 rx_ring_size = bp->rx_ring_size;
4505 DP(BNX2X_MSG_SP, "calculated rx_ring_size %d\n", rx_ring_size);
4508 sb = &bnx2x_fp(bp, index, status_blk);
4510 if (!IS_FCOE_IDX(index)) {
4512 if (!CHIP_IS_E1x(bp)) {
4513 sb->e2_sb = BNX2X_PCI_ALLOC(&bnx2x_fp(bp, index, status_blk_mapping),
4514 sizeof(struct host_hc_status_block_e2));
4518 sb->e1x_sb = BNX2X_PCI_ALLOC(&bnx2x_fp(bp, index, status_blk_mapping),
4519 sizeof(struct host_hc_status_block_e1x));
4525 /* FCoE Queue uses Default SB and doesn't ACK the SB, thus no need to
4526 * set shortcuts for it.
4528 if (!IS_FCOE_IDX(index))
4529 set_sb_shortcuts(bp, index);
4532 if (!skip_tx_queue(bp, index)) {
4533 /* fastpath tx rings: tx_buf tx_desc */
4534 for_each_cos_in_tx_queue(fp, cos) {
4535 struct bnx2x_fp_txdata *txdata = fp->txdata_ptr[cos];
4538 "allocating tx memory of fp %d cos %d\n",
4541 txdata->tx_buf_ring = kcalloc(NUM_TX_BD,
4542 sizeof(struct sw_tx_bd),
4544 if (!txdata->tx_buf_ring)
4546 txdata->tx_desc_ring = BNX2X_PCI_ALLOC(&txdata->tx_desc_mapping,
4547 sizeof(union eth_tx_bd_types) * NUM_TX_BD);
4548 if (!txdata->tx_desc_ring)
4554 if (!skip_rx_queue(bp, index)) {
4555 /* fastpath rx rings: rx_buf rx_desc rx_comp */
4556 bnx2x_fp(bp, index, rx_buf_ring) =
4557 kcalloc(NUM_RX_BD, sizeof(struct sw_rx_bd), GFP_KERNEL);
4558 if (!bnx2x_fp(bp, index, rx_buf_ring))
4560 bnx2x_fp(bp, index, rx_desc_ring) =
4561 BNX2X_PCI_ALLOC(&bnx2x_fp(bp, index, rx_desc_mapping),
4562 sizeof(struct eth_rx_bd) * NUM_RX_BD);
4563 if (!bnx2x_fp(bp, index, rx_desc_ring))
4566 /* Seed all CQEs by 1s */
4567 bnx2x_fp(bp, index, rx_comp_ring) =
4568 BNX2X_PCI_FALLOC(&bnx2x_fp(bp, index, rx_comp_mapping),
4569 sizeof(struct eth_fast_path_rx_cqe) * NUM_RCQ_BD);
4570 if (!bnx2x_fp(bp, index, rx_comp_ring))
4574 bnx2x_fp(bp, index, rx_page_ring) =
4575 kcalloc(NUM_RX_SGE, sizeof(struct sw_rx_page),
4577 if (!bnx2x_fp(bp, index, rx_page_ring))
4579 bnx2x_fp(bp, index, rx_sge_ring) =
4580 BNX2X_PCI_ALLOC(&bnx2x_fp(bp, index, rx_sge_mapping),
4581 BCM_PAGE_SIZE * NUM_RX_SGE_PAGES);
4582 if (!bnx2x_fp(bp, index, rx_sge_ring))
4585 bnx2x_set_next_page_rx_bd(fp);
4588 bnx2x_set_next_page_rx_cq(fp);
4591 ring_size = bnx2x_alloc_rx_bds(fp, rx_ring_size);
4592 if (ring_size < rx_ring_size)
4598 /* handles low memory cases */
4600 BNX2X_ERR("Unable to allocate full memory for queue %d (size %d)\n",
4602 /* FW will drop all packets if queue is not big enough,
4603 * In these cases we disable the queue
4604 * Min size is different for OOO, TPA and non-TPA queues
4606 if (ring_size < (fp->mode == TPA_MODE_DISABLED ?
4607 MIN_RX_SIZE_NONTPA : MIN_RX_SIZE_TPA)) {
4608 /* release memory allocated for this queue */
4609 bnx2x_free_fp_mem_at(bp, index);
4615 static int bnx2x_alloc_fp_mem_cnic(struct bnx2x *bp)
4619 if (bnx2x_alloc_fp_mem_at(bp, FCOE_IDX(bp)))
4620 /* we will fail load process instead of mark
4628 static int bnx2x_alloc_fp_mem(struct bnx2x *bp)
4632 /* 1. Allocate FP for leading - fatal if error
4633 * 2. Allocate RSS - fix number of queues if error
4637 if (bnx2x_alloc_fp_mem_at(bp, 0))
4641 for_each_nondefault_eth_queue(bp, i)
4642 if (bnx2x_alloc_fp_mem_at(bp, i))
4645 /* handle memory failures */
4646 if (i != BNX2X_NUM_ETH_QUEUES(bp)) {
4647 int delta = BNX2X_NUM_ETH_QUEUES(bp) - i;
4650 bnx2x_shrink_eth_fp(bp, delta);
4651 if (CNIC_SUPPORT(bp))
4652 /* move non eth FPs next to last eth FP
4653 * must be done in that order
4654 * FCOE_IDX < FWD_IDX < OOO_IDX
4657 /* move FCoE fp even NO_FCOE_FLAG is on */
4658 bnx2x_move_fp(bp, FCOE_IDX(bp), FCOE_IDX(bp) - delta);
4659 bp->num_ethernet_queues -= delta;
4660 bp->num_queues = bp->num_ethernet_queues +
4661 bp->num_cnic_queues;
4662 BNX2X_ERR("Adjusted num of queues from %d to %d\n",
4663 bp->num_queues + delta, bp->num_queues);
4669 void bnx2x_free_mem_bp(struct bnx2x *bp)
4673 for (i = 0; i < bp->fp_array_size; i++)
4674 kfree(bp->fp[i].tpa_info);
4677 kfree(bp->fp_stats);
4678 kfree(bp->bnx2x_txq);
4679 kfree(bp->msix_table);
4683 int bnx2x_alloc_mem_bp(struct bnx2x *bp)
4685 struct bnx2x_fastpath *fp;
4686 struct msix_entry *tbl;
4687 struct bnx2x_ilt *ilt;
4688 int msix_table_size = 0;
4689 int fp_array_size, txq_array_size;
4693 * The biggest MSI-X table we might need is as a maximum number of fast
4694 * path IGU SBs plus default SB (for PF only).
4696 msix_table_size = bp->igu_sb_cnt;
4699 BNX2X_DEV_INFO("msix_table_size %d\n", msix_table_size);
4701 /* fp array: RSS plus CNIC related L2 queues */
4702 fp_array_size = BNX2X_MAX_RSS_COUNT(bp) + CNIC_SUPPORT(bp);
4703 bp->fp_array_size = fp_array_size;
4704 BNX2X_DEV_INFO("fp_array_size %d\n", bp->fp_array_size);
4706 fp = kcalloc(bp->fp_array_size, sizeof(*fp), GFP_KERNEL);
4709 for (i = 0; i < bp->fp_array_size; i++) {
4711 kcalloc(ETH_MAX_AGGREGATION_QUEUES_E1H_E2,
4712 sizeof(struct bnx2x_agg_info), GFP_KERNEL);
4713 if (!(fp[i].tpa_info))
4719 /* allocate sp objs */
4720 bp->sp_objs = kcalloc(bp->fp_array_size, sizeof(struct bnx2x_sp_objs),
4725 /* allocate fp_stats */
4726 bp->fp_stats = kcalloc(bp->fp_array_size, sizeof(struct bnx2x_fp_stats),
4731 /* Allocate memory for the transmission queues array */
4733 BNX2X_MAX_RSS_COUNT(bp) * BNX2X_MULTI_TX_COS + CNIC_SUPPORT(bp);
4734 BNX2X_DEV_INFO("txq_array_size %d", txq_array_size);
4736 bp->bnx2x_txq = kcalloc(txq_array_size, sizeof(struct bnx2x_fp_txdata),
4742 tbl = kcalloc(msix_table_size, sizeof(*tbl), GFP_KERNEL);
4745 bp->msix_table = tbl;
4748 ilt = kzalloc(sizeof(*ilt), GFP_KERNEL);
4755 bnx2x_free_mem_bp(bp);
4759 int bnx2x_reload_if_running(struct net_device *dev)
4761 struct bnx2x *bp = netdev_priv(dev);
4763 if (unlikely(!netif_running(dev)))
4766 bnx2x_nic_unload(bp, UNLOAD_NORMAL, true);
4767 return bnx2x_nic_load(bp, LOAD_NORMAL);
4770 int bnx2x_get_cur_phy_idx(struct bnx2x *bp)
4772 u32 sel_phy_idx = 0;
4773 if (bp->link_params.num_phys <= 1)
4776 if (bp->link_vars.link_up) {
4777 sel_phy_idx = EXT_PHY1;
4778 /* In case link is SERDES, check if the EXT_PHY2 is the one */
4779 if ((bp->link_vars.link_status & LINK_STATUS_SERDES_LINK) &&
4780 (bp->link_params.phy[EXT_PHY2].supported & SUPPORTED_FIBRE))
4781 sel_phy_idx = EXT_PHY2;
4784 switch (bnx2x_phy_selection(&bp->link_params)) {
4785 case PORT_HW_CFG_PHY_SELECTION_HARDWARE_DEFAULT:
4786 case PORT_HW_CFG_PHY_SELECTION_FIRST_PHY:
4787 case PORT_HW_CFG_PHY_SELECTION_FIRST_PHY_PRIORITY:
4788 sel_phy_idx = EXT_PHY1;
4790 case PORT_HW_CFG_PHY_SELECTION_SECOND_PHY:
4791 case PORT_HW_CFG_PHY_SELECTION_SECOND_PHY_PRIORITY:
4792 sel_phy_idx = EXT_PHY2;
4799 int bnx2x_get_link_cfg_idx(struct bnx2x *bp)
4801 u32 sel_phy_idx = bnx2x_get_cur_phy_idx(bp);
4803 * The selected activated PHY is always after swapping (in case PHY
4804 * swapping is enabled). So when swapping is enabled, we need to reverse
4808 if (bp->link_params.multi_phy_config &
4809 PORT_HW_CFG_PHY_SWAPPED_ENABLED) {
4810 if (sel_phy_idx == EXT_PHY1)
4811 sel_phy_idx = EXT_PHY2;
4812 else if (sel_phy_idx == EXT_PHY2)
4813 sel_phy_idx = EXT_PHY1;
4815 return LINK_CONFIG_IDX(sel_phy_idx);
4818 #ifdef NETDEV_FCOE_WWNN
4819 int bnx2x_fcoe_get_wwn(struct net_device *dev, u64 *wwn, int type)
4821 struct bnx2x *bp = netdev_priv(dev);
4822 struct cnic_eth_dev *cp = &bp->cnic_eth_dev;
4825 case NETDEV_FCOE_WWNN:
4826 *wwn = HILO_U64(cp->fcoe_wwn_node_name_hi,
4827 cp->fcoe_wwn_node_name_lo);
4829 case NETDEV_FCOE_WWPN:
4830 *wwn = HILO_U64(cp->fcoe_wwn_port_name_hi,
4831 cp->fcoe_wwn_port_name_lo);
4834 BNX2X_ERR("Wrong WWN type requested - %d\n", type);
4842 /* called with rtnl_lock */
4843 int bnx2x_change_mtu(struct net_device *dev, int new_mtu)
4845 struct bnx2x *bp = netdev_priv(dev);
4847 if (pci_num_vf(bp->pdev)) {
4848 DP(BNX2X_MSG_IOV, "VFs are enabled, can not change MTU\n");
4852 if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
4853 BNX2X_ERR("Can't perform change MTU during parity recovery\n");
4857 if ((new_mtu > ETH_MAX_JUMBO_PACKET_SIZE) ||
4858 ((new_mtu + ETH_HLEN) < ETH_MIN_PACKET_SIZE)) {
4859 BNX2X_ERR("Can't support requested MTU size\n");
4863 /* This does not race with packet allocation
4864 * because the actual alloc size is
4865 * only updated as part of load
4869 if (IS_PF(bp) && SHMEM2_HAS(bp, curr_cfg))
4870 SHMEM2_WR(bp, curr_cfg, CURR_CFG_MET_OS);
4872 return bnx2x_reload_if_running(dev);
4875 netdev_features_t bnx2x_fix_features(struct net_device *dev,
4876 netdev_features_t features)
4878 struct bnx2x *bp = netdev_priv(dev);
4880 if (pci_num_vf(bp->pdev)) {
4881 netdev_features_t changed = dev->features ^ features;
4883 /* Revert the requested changes in features if they
4884 * would require internal reload of PF in bnx2x_set_features().
4886 if (!(features & NETIF_F_RXCSUM) && !bp->disable_tpa) {
4887 features &= ~NETIF_F_RXCSUM;
4888 features |= dev->features & NETIF_F_RXCSUM;
4891 if (changed & NETIF_F_LOOPBACK) {
4892 features &= ~NETIF_F_LOOPBACK;
4893 features |= dev->features & NETIF_F_LOOPBACK;
4897 /* TPA requires Rx CSUM offloading */
4898 if (!(features & NETIF_F_RXCSUM)) {
4899 features &= ~NETIF_F_LRO;
4900 features &= ~NETIF_F_GRO;
4906 int bnx2x_set_features(struct net_device *dev, netdev_features_t features)
4908 struct bnx2x *bp = netdev_priv(dev);
4909 netdev_features_t changes = features ^ dev->features;
4910 bool bnx2x_reload = false;
4913 /* VFs or non SRIOV PFs should be able to change loopback feature */
4914 if (!pci_num_vf(bp->pdev)) {
4915 if (features & NETIF_F_LOOPBACK) {
4916 if (bp->link_params.loopback_mode != LOOPBACK_BMAC) {
4917 bp->link_params.loopback_mode = LOOPBACK_BMAC;
4918 bnx2x_reload = true;
4921 if (bp->link_params.loopback_mode != LOOPBACK_NONE) {
4922 bp->link_params.loopback_mode = LOOPBACK_NONE;
4923 bnx2x_reload = true;
4928 /* if GRO is changed while LRO is enabled, don't force a reload */
4929 if ((changes & NETIF_F_GRO) && (features & NETIF_F_LRO))
4930 changes &= ~NETIF_F_GRO;
4932 /* if GRO is changed while HW TPA is off, don't force a reload */
4933 if ((changes & NETIF_F_GRO) && bp->disable_tpa)
4934 changes &= ~NETIF_F_GRO;
4937 bnx2x_reload = true;
4940 if (bp->recovery_state == BNX2X_RECOVERY_DONE) {
4941 dev->features = features;
4942 rc = bnx2x_reload_if_running(dev);
4945 /* else: bnx2x_nic_load() will be called at end of recovery */
4951 void bnx2x_tx_timeout(struct net_device *dev)
4953 struct bnx2x *bp = netdev_priv(dev);
4955 #ifdef BNX2X_STOP_ON_ERROR
4960 /* This allows the netif to be shutdown gracefully before resetting */
4961 bnx2x_schedule_sp_rtnl(bp, BNX2X_SP_RTNL_TX_TIMEOUT, 0);
4964 int bnx2x_suspend(struct pci_dev *pdev, pm_message_t state)
4966 struct net_device *dev = pci_get_drvdata(pdev);
4970 dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
4973 bp = netdev_priv(dev);
4977 pci_save_state(pdev);
4979 if (!netif_running(dev)) {
4984 netif_device_detach(dev);
4986 bnx2x_nic_unload(bp, UNLOAD_CLOSE, false);
4988 bnx2x_set_power_state(bp, pci_choose_state(pdev, state));
4995 int bnx2x_resume(struct pci_dev *pdev)
4997 struct net_device *dev = pci_get_drvdata(pdev);
5002 dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
5005 bp = netdev_priv(dev);
5007 if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
5008 BNX2X_ERR("Handling parity error recovery. Try again later\n");
5014 pci_restore_state(pdev);
5016 if (!netif_running(dev)) {
5021 bnx2x_set_power_state(bp, PCI_D0);
5022 netif_device_attach(dev);
5024 rc = bnx2x_nic_load(bp, LOAD_OPEN);
5031 void bnx2x_set_ctx_validation(struct bnx2x *bp, struct eth_context *cxt,
5035 BNX2X_ERR("bad context pointer %p\n", cxt);
5039 /* ustorm cxt validation */
5040 cxt->ustorm_ag_context.cdu_usage =
5041 CDU_RSRVD_VALUE_TYPE_A(HW_CID(bp, cid),
5042 CDU_REGION_NUMBER_UCM_AG, ETH_CONNECTION_TYPE);
5043 /* xcontext validation */
5044 cxt->xstorm_ag_context.cdu_reserved =
5045 CDU_RSRVD_VALUE_TYPE_A(HW_CID(bp, cid),
5046 CDU_REGION_NUMBER_XCM_AG, ETH_CONNECTION_TYPE);
5049 static void storm_memset_hc_timeout(struct bnx2x *bp, u8 port,
5050 u8 fw_sb_id, u8 sb_index,
5053 u32 addr = BAR_CSTRORM_INTMEM +
5054 CSTORM_STATUS_BLOCK_DATA_TIMEOUT_OFFSET(fw_sb_id, sb_index);
5055 REG_WR8(bp, addr, ticks);
5057 "port %x fw_sb_id %d sb_index %d ticks %d\n",
5058 port, fw_sb_id, sb_index, ticks);
5061 static void storm_memset_hc_disable(struct bnx2x *bp, u8 port,
5062 u16 fw_sb_id, u8 sb_index,
5065 u32 enable_flag = disable ? 0 : (1 << HC_INDEX_DATA_HC_ENABLED_SHIFT);
5066 u32 addr = BAR_CSTRORM_INTMEM +
5067 CSTORM_STATUS_BLOCK_DATA_FLAGS_OFFSET(fw_sb_id, sb_index);
5068 u8 flags = REG_RD8(bp, addr);
5070 flags &= ~HC_INDEX_DATA_HC_ENABLED;
5071 flags |= enable_flag;
5072 REG_WR8(bp, addr, flags);
5074 "port %x fw_sb_id %d sb_index %d disable %d\n",
5075 port, fw_sb_id, sb_index, disable);
5078 void bnx2x_update_coalesce_sb_index(struct bnx2x *bp, u8 fw_sb_id,
5079 u8 sb_index, u8 disable, u16 usec)
5081 int port = BP_PORT(bp);
5082 u8 ticks = usec / BNX2X_BTR;
5084 storm_memset_hc_timeout(bp, port, fw_sb_id, sb_index, ticks);
5086 disable = disable ? 1 : (usec ? 0 : 1);
5087 storm_memset_hc_disable(bp, port, fw_sb_id, sb_index, disable);
5090 void bnx2x_schedule_sp_rtnl(struct bnx2x *bp, enum sp_rtnl_flag flag,
5093 smp_mb__before_atomic();
5094 set_bit(flag, &bp->sp_rtnl_state);
5095 smp_mb__after_atomic();
5096 DP((BNX2X_MSG_SP | verbose), "Scheduling sp_rtnl task [Flag: %d]\n",
5098 schedule_delayed_work(&bp->sp_rtnl_task, 0);