1 /* bnx2x_cmn.c: Broadcom Everest network driver.
3 * Copyright (c) 2007-2013 Broadcom Corporation
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation.
9 * Maintained by: Eilon Greenstein <eilong@broadcom.com>
10 * Written by: Eliezer Tamir
11 * Based on code from Michael Chan's bnx2 driver
12 * UDP CSUM errata workaround by Arik Gendelman
13 * Slowpath and fastpath rework by Vladislav Zolotarov
14 * Statistics and Link management by Yitchak Gertner
18 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
20 #include <linux/etherdevice.h>
21 #include <linux/if_vlan.h>
22 #include <linux/interrupt.h>
26 #include <net/ip6_checksum.h>
27 #include <linux/prefetch.h>
28 #include "bnx2x_cmn.h"
29 #include "bnx2x_init.h"
33 * bnx2x_move_fp - move content of the fastpath structure.
36 * @from: source FP index
37 * @to: destination FP index
39 * Makes sure the contents of the bp->fp[to].napi is kept
40 * intact. This is done by first copying the napi struct from
41 * the target to the source, and then mem copying the entire
42 * source onto the target. Update txdata pointers and related
45 static inline void bnx2x_move_fp(struct bnx2x *bp, int from, int to)
47 struct bnx2x_fastpath *from_fp = &bp->fp[from];
48 struct bnx2x_fastpath *to_fp = &bp->fp[to];
49 struct bnx2x_sp_objs *from_sp_objs = &bp->sp_objs[from];
50 struct bnx2x_sp_objs *to_sp_objs = &bp->sp_objs[to];
51 struct bnx2x_fp_stats *from_fp_stats = &bp->fp_stats[from];
52 struct bnx2x_fp_stats *to_fp_stats = &bp->fp_stats[to];
53 int old_max_eth_txqs, new_max_eth_txqs;
54 int old_txdata_index = 0, new_txdata_index = 0;
56 /* Copy the NAPI object as it has been already initialized */
57 from_fp->napi = to_fp->napi;
59 /* Move bnx2x_fastpath contents */
60 memcpy(to_fp, from_fp, sizeof(*to_fp));
63 /* move sp_objs contents as well, as their indices match fp ones */
64 memcpy(to_sp_objs, from_sp_objs, sizeof(*to_sp_objs));
66 /* move fp_stats contents as well, as their indices match fp ones */
67 memcpy(to_fp_stats, from_fp_stats, sizeof(*to_fp_stats));
69 /* Update txdata pointers in fp and move txdata content accordingly:
70 * Each fp consumes 'max_cos' txdata structures, so the index should be
71 * decremented by max_cos x delta.
74 old_max_eth_txqs = BNX2X_NUM_ETH_QUEUES(bp) * (bp)->max_cos;
75 new_max_eth_txqs = (BNX2X_NUM_ETH_QUEUES(bp) - from + to) *
77 if (from == FCOE_IDX(bp)) {
78 old_txdata_index = old_max_eth_txqs + FCOE_TXQ_IDX_OFFSET;
79 new_txdata_index = new_max_eth_txqs + FCOE_TXQ_IDX_OFFSET;
82 memcpy(&bp->bnx2x_txq[new_txdata_index],
83 &bp->bnx2x_txq[old_txdata_index],
84 sizeof(struct bnx2x_fp_txdata));
85 to_fp->txdata_ptr[0] = &bp->bnx2x_txq[new_txdata_index];
89 * bnx2x_fill_fw_str - Fill buffer with FW version string.
92 * @buf: character buffer to fill with the fw name
93 * @buf_len: length of the above buffer
96 void bnx2x_fill_fw_str(struct bnx2x *bp, char *buf, size_t buf_len)
99 u8 phy_fw_ver[PHY_FW_VER_LEN];
101 phy_fw_ver[0] = '\0';
102 bnx2x_get_ext_phy_fw_version(&bp->link_params,
103 phy_fw_ver, PHY_FW_VER_LEN);
104 strlcpy(buf, bp->fw_ver, buf_len);
105 snprintf(buf + strlen(bp->fw_ver), 32 - strlen(bp->fw_ver),
107 (bp->common.bc_ver & 0xff0000) >> 16,
108 (bp->common.bc_ver & 0xff00) >> 8,
109 (bp->common.bc_ver & 0xff),
110 ((phy_fw_ver[0] != '\0') ? " phy " : ""), phy_fw_ver);
112 bnx2x_vf_fill_fw_str(bp, buf, buf_len);
117 * bnx2x_shrink_eth_fp - guarantees fastpath structures stay intact
120 * @delta: number of eth queues which were not allocated
122 static void bnx2x_shrink_eth_fp(struct bnx2x *bp, int delta)
124 int i, cos, old_eth_num = BNX2X_NUM_ETH_QUEUES(bp);
126 /* Queue pointer cannot be re-set on an fp-basis, as moving pointer
127 * backward along the array could cause memory to be overriden
129 for (cos = 1; cos < bp->max_cos; cos++) {
130 for (i = 0; i < old_eth_num - delta; i++) {
131 struct bnx2x_fastpath *fp = &bp->fp[i];
132 int new_idx = cos * (old_eth_num - delta) + i;
134 memcpy(&bp->bnx2x_txq[new_idx], fp->txdata_ptr[cos],
135 sizeof(struct bnx2x_fp_txdata));
136 fp->txdata_ptr[cos] = &bp->bnx2x_txq[new_idx];
141 int load_count[2][3] = { {0} }; /* per-path: 0-common, 1-port0, 2-port1 */
143 /* free skb in the packet ring at pos idx
144 * return idx of last bd freed
146 static u16 bnx2x_free_tx_pkt(struct bnx2x *bp, struct bnx2x_fp_txdata *txdata,
147 u16 idx, unsigned int *pkts_compl,
148 unsigned int *bytes_compl)
150 struct sw_tx_bd *tx_buf = &txdata->tx_buf_ring[idx];
151 struct eth_tx_start_bd *tx_start_bd;
152 struct eth_tx_bd *tx_data_bd;
153 struct sk_buff *skb = tx_buf->skb;
154 u16 bd_idx = TX_BD(tx_buf->first_bd), new_cons;
157 /* prefetch skb end pointer to speedup dev_kfree_skb() */
160 DP(NETIF_MSG_TX_DONE, "fp[%d]: pkt_idx %d buff @(%p)->skb %p\n",
161 txdata->txq_index, idx, tx_buf, skb);
164 tx_start_bd = &txdata->tx_desc_ring[bd_idx].start_bd;
165 dma_unmap_single(&bp->pdev->dev, BD_UNMAP_ADDR(tx_start_bd),
166 BD_UNMAP_LEN(tx_start_bd), DMA_TO_DEVICE);
169 nbd = le16_to_cpu(tx_start_bd->nbd) - 1;
170 #ifdef BNX2X_STOP_ON_ERROR
171 if ((nbd - 1) > (MAX_SKB_FRAGS + 2)) {
172 BNX2X_ERR("BAD nbd!\n");
176 new_cons = nbd + tx_buf->first_bd;
178 /* Get the next bd */
179 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
181 /* Skip a parse bd... */
183 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
185 /* ...and the TSO split header bd since they have no mapping */
186 if (tx_buf->flags & BNX2X_TSO_SPLIT_BD) {
188 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
194 tx_data_bd = &txdata->tx_desc_ring[bd_idx].reg_bd;
195 dma_unmap_page(&bp->pdev->dev, BD_UNMAP_ADDR(tx_data_bd),
196 BD_UNMAP_LEN(tx_data_bd), DMA_TO_DEVICE);
198 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
205 (*bytes_compl) += skb->len;
208 dev_kfree_skb_any(skb);
209 tx_buf->first_bd = 0;
215 int bnx2x_tx_int(struct bnx2x *bp, struct bnx2x_fp_txdata *txdata)
217 struct netdev_queue *txq;
218 u16 hw_cons, sw_cons, bd_cons = txdata->tx_bd_cons;
219 unsigned int pkts_compl = 0, bytes_compl = 0;
221 #ifdef BNX2X_STOP_ON_ERROR
222 if (unlikely(bp->panic))
226 txq = netdev_get_tx_queue(bp->dev, txdata->txq_index);
227 hw_cons = le16_to_cpu(*txdata->tx_cons_sb);
228 sw_cons = txdata->tx_pkt_cons;
230 while (sw_cons != hw_cons) {
233 pkt_cons = TX_BD(sw_cons);
235 DP(NETIF_MSG_TX_DONE,
236 "queue[%d]: hw_cons %u sw_cons %u pkt_cons %u\n",
237 txdata->txq_index, hw_cons, sw_cons, pkt_cons);
239 bd_cons = bnx2x_free_tx_pkt(bp, txdata, pkt_cons,
240 &pkts_compl, &bytes_compl);
245 netdev_tx_completed_queue(txq, pkts_compl, bytes_compl);
247 txdata->tx_pkt_cons = sw_cons;
248 txdata->tx_bd_cons = bd_cons;
250 /* Need to make the tx_bd_cons update visible to start_xmit()
251 * before checking for netif_tx_queue_stopped(). Without the
252 * memory barrier, there is a small possibility that
253 * start_xmit() will miss it and cause the queue to be stopped
255 * On the other hand we need an rmb() here to ensure the proper
256 * ordering of bit testing in the following
257 * netif_tx_queue_stopped(txq) call.
261 if (unlikely(netif_tx_queue_stopped(txq))) {
262 /* Taking tx_lock() is needed to prevent reenabling the queue
263 * while it's empty. This could have happen if rx_action() gets
264 * suspended in bnx2x_tx_int() after the condition before
265 * netif_tx_wake_queue(), while tx_action (bnx2x_start_xmit()):
267 * stops the queue->sees fresh tx_bd_cons->releases the queue->
268 * sends some packets consuming the whole queue again->
272 __netif_tx_lock(txq, smp_processor_id());
274 if ((netif_tx_queue_stopped(txq)) &&
275 (bp->state == BNX2X_STATE_OPEN) &&
276 (bnx2x_tx_avail(bp, txdata) >= MAX_DESC_PER_TX_PKT))
277 netif_tx_wake_queue(txq);
279 __netif_tx_unlock(txq);
284 static inline void bnx2x_update_last_max_sge(struct bnx2x_fastpath *fp,
287 u16 last_max = fp->last_max_sge;
289 if (SUB_S16(idx, last_max) > 0)
290 fp->last_max_sge = idx;
293 static inline void bnx2x_update_sge_prod(struct bnx2x_fastpath *fp,
295 struct eth_end_agg_rx_cqe *cqe)
297 struct bnx2x *bp = fp->bp;
298 u16 last_max, last_elem, first_elem;
305 /* First mark all used pages */
306 for (i = 0; i < sge_len; i++)
307 BIT_VEC64_CLEAR_BIT(fp->sge_mask,
308 RX_SGE(le16_to_cpu(cqe->sgl_or_raw_data.sgl[i])));
310 DP(NETIF_MSG_RX_STATUS, "fp_cqe->sgl[%d] = %d\n",
311 sge_len - 1, le16_to_cpu(cqe->sgl_or_raw_data.sgl[sge_len - 1]));
313 /* Here we assume that the last SGE index is the biggest */
314 prefetch((void *)(fp->sge_mask));
315 bnx2x_update_last_max_sge(fp,
316 le16_to_cpu(cqe->sgl_or_raw_data.sgl[sge_len - 1]));
318 last_max = RX_SGE(fp->last_max_sge);
319 last_elem = last_max >> BIT_VEC64_ELEM_SHIFT;
320 first_elem = RX_SGE(fp->rx_sge_prod) >> BIT_VEC64_ELEM_SHIFT;
322 /* If ring is not full */
323 if (last_elem + 1 != first_elem)
326 /* Now update the prod */
327 for (i = first_elem; i != last_elem; i = NEXT_SGE_MASK_ELEM(i)) {
328 if (likely(fp->sge_mask[i]))
331 fp->sge_mask[i] = BIT_VEC64_ELEM_ONE_MASK;
332 delta += BIT_VEC64_ELEM_SZ;
336 fp->rx_sge_prod += delta;
337 /* clear page-end entries */
338 bnx2x_clear_sge_mask_next_elems(fp);
341 DP(NETIF_MSG_RX_STATUS,
342 "fp->last_max_sge = %d fp->rx_sge_prod = %d\n",
343 fp->last_max_sge, fp->rx_sge_prod);
346 /* Get Toeplitz hash value in the skb using the value from the
347 * CQE (calculated by HW).
349 static u32 bnx2x_get_rxhash(const struct bnx2x *bp,
350 const struct eth_fast_path_rx_cqe *cqe,
353 /* Get Toeplitz hash from CQE */
354 if ((bp->dev->features & NETIF_F_RXHASH) &&
355 (cqe->status_flags & ETH_FAST_PATH_RX_CQE_RSS_HASH_FLG)) {
356 enum eth_rss_hash_type htype;
358 htype = cqe->status_flags & ETH_FAST_PATH_RX_CQE_RSS_HASH_TYPE;
359 *l4_rxhash = (htype == TCP_IPV4_HASH_TYPE) ||
360 (htype == TCP_IPV6_HASH_TYPE);
361 return le32_to_cpu(cqe->rss_hash_result);
367 static void bnx2x_tpa_start(struct bnx2x_fastpath *fp, u16 queue,
369 struct eth_fast_path_rx_cqe *cqe)
371 struct bnx2x *bp = fp->bp;
372 struct sw_rx_bd *cons_rx_buf = &fp->rx_buf_ring[cons];
373 struct sw_rx_bd *prod_rx_buf = &fp->rx_buf_ring[prod];
374 struct eth_rx_bd *prod_bd = &fp->rx_desc_ring[prod];
376 struct bnx2x_agg_info *tpa_info = &fp->tpa_info[queue];
377 struct sw_rx_bd *first_buf = &tpa_info->first_buf;
379 /* print error if current state != stop */
380 if (tpa_info->tpa_state != BNX2X_TPA_STOP)
381 BNX2X_ERR("start of bin not in stop [%d]\n", queue);
383 /* Try to map an empty data buffer from the aggregation info */
384 mapping = dma_map_single(&bp->pdev->dev,
385 first_buf->data + NET_SKB_PAD,
386 fp->rx_buf_size, DMA_FROM_DEVICE);
388 * ...if it fails - move the skb from the consumer to the producer
389 * and set the current aggregation state as ERROR to drop it
390 * when TPA_STOP arrives.
393 if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
394 /* Move the BD from the consumer to the producer */
395 bnx2x_reuse_rx_data(fp, cons, prod);
396 tpa_info->tpa_state = BNX2X_TPA_ERROR;
400 /* move empty data from pool to prod */
401 prod_rx_buf->data = first_buf->data;
402 dma_unmap_addr_set(prod_rx_buf, mapping, mapping);
403 /* point prod_bd to new data */
404 prod_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
405 prod_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
407 /* move partial skb from cons to pool (don't unmap yet) */
408 *first_buf = *cons_rx_buf;
410 /* mark bin state as START */
411 tpa_info->parsing_flags =
412 le16_to_cpu(cqe->pars_flags.flags);
413 tpa_info->vlan_tag = le16_to_cpu(cqe->vlan_tag);
414 tpa_info->tpa_state = BNX2X_TPA_START;
415 tpa_info->len_on_bd = le16_to_cpu(cqe->len_on_bd);
416 tpa_info->placement_offset = cqe->placement_offset;
417 tpa_info->rxhash = bnx2x_get_rxhash(bp, cqe, &tpa_info->l4_rxhash);
418 if (fp->mode == TPA_MODE_GRO) {
419 u16 gro_size = le16_to_cpu(cqe->pkt_len_or_gro_seg_len);
420 tpa_info->full_page = SGE_PAGES / gro_size * gro_size;
421 tpa_info->gro_size = gro_size;
424 #ifdef BNX2X_STOP_ON_ERROR
425 fp->tpa_queue_used |= (1 << queue);
426 #ifdef _ASM_GENERIC_INT_L64_H
427 DP(NETIF_MSG_RX_STATUS, "fp->tpa_queue_used = 0x%lx\n",
429 DP(NETIF_MSG_RX_STATUS, "fp->tpa_queue_used = 0x%llx\n",
435 /* Timestamp option length allowed for TPA aggregation:
437 * nop nop kind length echo val
439 #define TPA_TSTAMP_OPT_LEN 12
441 * bnx2x_set_gro_params - compute GRO values
444 * @parsing_flags: parsing flags from the START CQE
445 * @len_on_bd: total length of the first packet for the
447 * @pkt_len: length of all segments
449 * Approximate value of the MSS for this aggregation calculated using
450 * the first packet of it.
451 * Compute number of aggregated segments, and gso_type.
453 static void bnx2x_set_gro_params(struct sk_buff *skb, u16 parsing_flags,
454 u16 len_on_bd, unsigned int pkt_len,
455 u16 num_of_coalesced_segs)
457 /* TPA aggregation won't have either IP options or TCP options
458 * other than timestamp or IPv6 extension headers.
460 u16 hdrs_len = ETH_HLEN + sizeof(struct tcphdr);
462 if (GET_FLAG(parsing_flags, PARSING_FLAGS_OVER_ETHERNET_PROTOCOL) ==
463 PRS_FLAG_OVERETH_IPV6) {
464 hdrs_len += sizeof(struct ipv6hdr);
465 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
467 hdrs_len += sizeof(struct iphdr);
468 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
471 /* Check if there was a TCP timestamp, if there is it's will
472 * always be 12 bytes length: nop nop kind length echo val.
474 * Otherwise FW would close the aggregation.
476 if (parsing_flags & PARSING_FLAGS_TIME_STAMP_EXIST_FLAG)
477 hdrs_len += TPA_TSTAMP_OPT_LEN;
479 skb_shinfo(skb)->gso_size = len_on_bd - hdrs_len;
481 /* tcp_gro_complete() will copy NAPI_GRO_CB(skb)->count
482 * to skb_shinfo(skb)->gso_segs
484 NAPI_GRO_CB(skb)->count = num_of_coalesced_segs;
487 static int bnx2x_alloc_rx_sge(struct bnx2x *bp,
488 struct bnx2x_fastpath *fp, u16 index)
490 struct page *page = alloc_pages(GFP_ATOMIC, PAGES_PER_SGE_SHIFT);
491 struct sw_rx_page *sw_buf = &fp->rx_page_ring[index];
492 struct eth_rx_sge *sge = &fp->rx_sge_ring[index];
495 if (unlikely(page == NULL)) {
496 BNX2X_ERR("Can't alloc sge\n");
500 mapping = dma_map_page(&bp->pdev->dev, page, 0,
501 SGE_PAGES, DMA_FROM_DEVICE);
502 if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
503 __free_pages(page, PAGES_PER_SGE_SHIFT);
504 BNX2X_ERR("Can't map sge\n");
509 dma_unmap_addr_set(sw_buf, mapping, mapping);
511 sge->addr_hi = cpu_to_le32(U64_HI(mapping));
512 sge->addr_lo = cpu_to_le32(U64_LO(mapping));
517 static int bnx2x_fill_frag_skb(struct bnx2x *bp, struct bnx2x_fastpath *fp,
518 struct bnx2x_agg_info *tpa_info,
521 struct eth_end_agg_rx_cqe *cqe,
524 struct sw_rx_page *rx_pg, old_rx_pg;
525 u32 i, frag_len, frag_size;
526 int err, j, frag_id = 0;
527 u16 len_on_bd = tpa_info->len_on_bd;
528 u16 full_page = 0, gro_size = 0;
530 frag_size = le16_to_cpu(cqe->pkt_len) - len_on_bd;
532 if (fp->mode == TPA_MODE_GRO) {
533 gro_size = tpa_info->gro_size;
534 full_page = tpa_info->full_page;
537 /* This is needed in order to enable forwarding support */
539 bnx2x_set_gro_params(skb, tpa_info->parsing_flags, len_on_bd,
540 le16_to_cpu(cqe->pkt_len),
541 le16_to_cpu(cqe->num_of_coalesced_segs));
543 #ifdef BNX2X_STOP_ON_ERROR
544 if (pages > min_t(u32, 8, MAX_SKB_FRAGS) * SGE_PAGES) {
545 BNX2X_ERR("SGL length is too long: %d. CQE index is %d\n",
547 BNX2X_ERR("cqe->pkt_len = %d\n", cqe->pkt_len);
553 /* Run through the SGL and compose the fragmented skb */
554 for (i = 0, j = 0; i < pages; i += PAGES_PER_SGE, j++) {
555 u16 sge_idx = RX_SGE(le16_to_cpu(cqe->sgl_or_raw_data.sgl[j]));
557 /* FW gives the indices of the SGE as if the ring is an array
558 (meaning that "next" element will consume 2 indices) */
559 if (fp->mode == TPA_MODE_GRO)
560 frag_len = min_t(u32, frag_size, (u32)full_page);
562 frag_len = min_t(u32, frag_size, (u32)SGE_PAGES);
564 rx_pg = &fp->rx_page_ring[sge_idx];
567 /* If we fail to allocate a substitute page, we simply stop
568 where we are and drop the whole packet */
569 err = bnx2x_alloc_rx_sge(bp, fp, sge_idx);
571 bnx2x_fp_qstats(bp, fp)->rx_skb_alloc_failed++;
575 /* Unmap the page as we r going to pass it to the stack */
576 dma_unmap_page(&bp->pdev->dev,
577 dma_unmap_addr(&old_rx_pg, mapping),
578 SGE_PAGES, DMA_FROM_DEVICE);
579 /* Add one frag and update the appropriate fields in the skb */
580 if (fp->mode == TPA_MODE_LRO)
581 skb_fill_page_desc(skb, j, old_rx_pg.page, 0, frag_len);
585 for (rem = frag_len; rem > 0; rem -= gro_size) {
586 int len = rem > gro_size ? gro_size : rem;
587 skb_fill_page_desc(skb, frag_id++,
588 old_rx_pg.page, offset, len);
590 get_page(old_rx_pg.page);
595 skb->data_len += frag_len;
596 skb->truesize += SGE_PAGES;
597 skb->len += frag_len;
599 frag_size -= frag_len;
605 static void bnx2x_frag_free(const struct bnx2x_fastpath *fp, void *data)
607 if (fp->rx_frag_size)
608 put_page(virt_to_head_page(data));
613 static void *bnx2x_frag_alloc(const struct bnx2x_fastpath *fp)
615 if (fp->rx_frag_size)
616 return netdev_alloc_frag(fp->rx_frag_size);
618 return kmalloc(fp->rx_buf_size + NET_SKB_PAD, GFP_ATOMIC);
622 static void bnx2x_gro_ip_csum(struct bnx2x *bp, struct sk_buff *skb)
624 const struct iphdr *iph = ip_hdr(skb);
627 skb_set_transport_header(skb, sizeof(struct iphdr));
630 th->check = ~tcp_v4_check(skb->len - skb_transport_offset(skb),
631 iph->saddr, iph->daddr, 0);
634 static void bnx2x_gro_ipv6_csum(struct bnx2x *bp, struct sk_buff *skb)
636 struct ipv6hdr *iph = ipv6_hdr(skb);
639 skb_set_transport_header(skb, sizeof(struct ipv6hdr));
642 th->check = ~tcp_v6_check(skb->len - skb_transport_offset(skb),
643 &iph->saddr, &iph->daddr, 0);
646 static void bnx2x_gro_csum(struct bnx2x *bp, struct sk_buff *skb,
647 void (*gro_func)(struct bnx2x*, struct sk_buff*))
649 skb_set_network_header(skb, 0);
651 tcp_gro_complete(skb);
655 static void bnx2x_gro_receive(struct bnx2x *bp, struct bnx2x_fastpath *fp,
659 if (skb_shinfo(skb)->gso_size) {
660 switch (be16_to_cpu(skb->protocol)) {
662 bnx2x_gro_csum(bp, skb, bnx2x_gro_ip_csum);
665 bnx2x_gro_csum(bp, skb, bnx2x_gro_ipv6_csum);
668 BNX2X_ERR("Error: FW GRO supports only IPv4/IPv6, not 0x%04x\n",
669 be16_to_cpu(skb->protocol));
673 napi_gro_receive(&fp->napi, skb);
676 static void bnx2x_tpa_stop(struct bnx2x *bp, struct bnx2x_fastpath *fp,
677 struct bnx2x_agg_info *tpa_info,
679 struct eth_end_agg_rx_cqe *cqe,
682 struct sw_rx_bd *rx_buf = &tpa_info->first_buf;
683 u8 pad = tpa_info->placement_offset;
684 u16 len = tpa_info->len_on_bd;
685 struct sk_buff *skb = NULL;
686 u8 *new_data, *data = rx_buf->data;
687 u8 old_tpa_state = tpa_info->tpa_state;
689 tpa_info->tpa_state = BNX2X_TPA_STOP;
691 /* If we there was an error during the handling of the TPA_START -
692 * drop this aggregation.
694 if (old_tpa_state == BNX2X_TPA_ERROR)
697 /* Try to allocate the new data */
698 new_data = bnx2x_frag_alloc(fp);
699 /* Unmap skb in the pool anyway, as we are going to change
700 pool entry status to BNX2X_TPA_STOP even if new skb allocation
702 dma_unmap_single(&bp->pdev->dev, dma_unmap_addr(rx_buf, mapping),
703 fp->rx_buf_size, DMA_FROM_DEVICE);
704 if (likely(new_data))
705 skb = build_skb(data, fp->rx_frag_size);
708 #ifdef BNX2X_STOP_ON_ERROR
709 if (pad + len > fp->rx_buf_size) {
710 BNX2X_ERR("skb_put is about to fail... pad %d len %d rx_buf_size %d\n",
711 pad, len, fp->rx_buf_size);
717 skb_reserve(skb, pad + NET_SKB_PAD);
719 skb->rxhash = tpa_info->rxhash;
720 skb->l4_rxhash = tpa_info->l4_rxhash;
722 skb->protocol = eth_type_trans(skb, bp->dev);
723 skb->ip_summed = CHECKSUM_UNNECESSARY;
725 if (!bnx2x_fill_frag_skb(bp, fp, tpa_info, pages,
726 skb, cqe, cqe_idx)) {
727 if (tpa_info->parsing_flags & PARSING_FLAGS_VLAN)
728 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), tpa_info->vlan_tag);
729 bnx2x_gro_receive(bp, fp, skb);
731 DP(NETIF_MSG_RX_STATUS,
732 "Failed to allocate new pages - dropping packet!\n");
733 dev_kfree_skb_any(skb);
737 /* put new data in bin */
738 rx_buf->data = new_data;
742 bnx2x_frag_free(fp, new_data);
744 /* drop the packet and keep the buffer in the bin */
745 DP(NETIF_MSG_RX_STATUS,
746 "Failed to allocate or map a new skb - dropping packet!\n");
747 bnx2x_fp_stats(bp, fp)->eth_q_stats.rx_skb_alloc_failed++;
750 static int bnx2x_alloc_rx_data(struct bnx2x *bp,
751 struct bnx2x_fastpath *fp, u16 index)
754 struct sw_rx_bd *rx_buf = &fp->rx_buf_ring[index];
755 struct eth_rx_bd *rx_bd = &fp->rx_desc_ring[index];
758 data = bnx2x_frag_alloc(fp);
759 if (unlikely(data == NULL))
762 mapping = dma_map_single(&bp->pdev->dev, data + NET_SKB_PAD,
765 if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
766 bnx2x_frag_free(fp, data);
767 BNX2X_ERR("Can't map rx data\n");
772 dma_unmap_addr_set(rx_buf, mapping, mapping);
774 rx_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
775 rx_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
781 void bnx2x_csum_validate(struct sk_buff *skb, union eth_rx_cqe *cqe,
782 struct bnx2x_fastpath *fp,
783 struct bnx2x_eth_q_stats *qstats)
785 /* Do nothing if no L4 csum validation was done.
786 * We do not check whether IP csum was validated. For IPv4 we assume
787 * that if the card got as far as validating the L4 csum, it also
788 * validated the IP csum. IPv6 has no IP csum.
790 if (cqe->fast_path_cqe.status_flags &
791 ETH_FAST_PATH_RX_CQE_L4_XSUM_NO_VALIDATION_FLG)
794 /* If L4 validation was done, check if an error was found. */
796 if (cqe->fast_path_cqe.type_error_flags &
797 (ETH_FAST_PATH_RX_CQE_IP_BAD_XSUM_FLG |
798 ETH_FAST_PATH_RX_CQE_L4_BAD_XSUM_FLG))
799 qstats->hw_csum_err++;
801 skb->ip_summed = CHECKSUM_UNNECESSARY;
804 int bnx2x_rx_int(struct bnx2x_fastpath *fp, int budget)
806 struct bnx2x *bp = fp->bp;
807 u16 bd_cons, bd_prod, bd_prod_fw, comp_ring_cons;
808 u16 hw_comp_cons, sw_comp_cons, sw_comp_prod;
811 #ifdef BNX2X_STOP_ON_ERROR
812 if (unlikely(bp->panic))
816 /* CQ "next element" is of the size of the regular element,
817 that's why it's ok here */
818 hw_comp_cons = le16_to_cpu(*fp->rx_cons_sb);
819 if ((hw_comp_cons & MAX_RCQ_DESC_CNT) == MAX_RCQ_DESC_CNT)
822 bd_cons = fp->rx_bd_cons;
823 bd_prod = fp->rx_bd_prod;
824 bd_prod_fw = bd_prod;
825 sw_comp_cons = fp->rx_comp_cons;
826 sw_comp_prod = fp->rx_comp_prod;
828 /* Memory barrier necessary as speculative reads of the rx
829 * buffer can be ahead of the index in the status block
833 DP(NETIF_MSG_RX_STATUS,
834 "queue[%d]: hw_comp_cons %u sw_comp_cons %u\n",
835 fp->index, hw_comp_cons, sw_comp_cons);
837 while (sw_comp_cons != hw_comp_cons) {
838 struct sw_rx_bd *rx_buf = NULL;
840 union eth_rx_cqe *cqe;
841 struct eth_fast_path_rx_cqe *cqe_fp;
843 enum eth_rx_cqe_type cqe_fp_type;
848 #ifdef BNX2X_STOP_ON_ERROR
849 if (unlikely(bp->panic))
853 comp_ring_cons = RCQ_BD(sw_comp_cons);
854 bd_prod = RX_BD(bd_prod);
855 bd_cons = RX_BD(bd_cons);
857 cqe = &fp->rx_comp_ring[comp_ring_cons];
858 cqe_fp = &cqe->fast_path_cqe;
859 cqe_fp_flags = cqe_fp->type_error_flags;
860 cqe_fp_type = cqe_fp_flags & ETH_FAST_PATH_RX_CQE_TYPE;
862 DP(NETIF_MSG_RX_STATUS,
863 "CQE type %x err %x status %x queue %x vlan %x len %u\n",
864 CQE_TYPE(cqe_fp_flags),
865 cqe_fp_flags, cqe_fp->status_flags,
866 le32_to_cpu(cqe_fp->rss_hash_result),
867 le16_to_cpu(cqe_fp->vlan_tag),
868 le16_to_cpu(cqe_fp->pkt_len_or_gro_seg_len));
870 /* is this a slowpath msg? */
871 if (unlikely(CQE_TYPE_SLOW(cqe_fp_type))) {
872 bnx2x_sp_event(fp, cqe);
876 rx_buf = &fp->rx_buf_ring[bd_cons];
879 if (!CQE_TYPE_FAST(cqe_fp_type)) {
880 struct bnx2x_agg_info *tpa_info;
881 u16 frag_size, pages;
882 #ifdef BNX2X_STOP_ON_ERROR
884 if (fp->disable_tpa &&
885 (CQE_TYPE_START(cqe_fp_type) ||
886 CQE_TYPE_STOP(cqe_fp_type)))
887 BNX2X_ERR("START/STOP packet while disable_tpa type %x\n",
888 CQE_TYPE(cqe_fp_type));
891 if (CQE_TYPE_START(cqe_fp_type)) {
892 u16 queue = cqe_fp->queue_index;
893 DP(NETIF_MSG_RX_STATUS,
894 "calling tpa_start on queue %d\n",
897 bnx2x_tpa_start(fp, queue,
904 queue = cqe->end_agg_cqe.queue_index;
905 tpa_info = &fp->tpa_info[queue];
906 DP(NETIF_MSG_RX_STATUS,
907 "calling tpa_stop on queue %d\n",
910 frag_size = le16_to_cpu(cqe->end_agg_cqe.pkt_len) -
913 if (fp->mode == TPA_MODE_GRO)
914 pages = (frag_size + tpa_info->full_page - 1) /
917 pages = SGE_PAGE_ALIGN(frag_size) >>
920 bnx2x_tpa_stop(bp, fp, tpa_info, pages,
921 &cqe->end_agg_cqe, comp_ring_cons);
922 #ifdef BNX2X_STOP_ON_ERROR
927 bnx2x_update_sge_prod(fp, pages, &cqe->end_agg_cqe);
931 len = le16_to_cpu(cqe_fp->pkt_len_or_gro_seg_len);
932 pad = cqe_fp->placement_offset;
933 dma_sync_single_for_cpu(&bp->pdev->dev,
934 dma_unmap_addr(rx_buf, mapping),
935 pad + RX_COPY_THRESH,
938 prefetch(data + pad); /* speedup eth_type_trans() */
939 /* is this an error packet? */
940 if (unlikely(cqe_fp_flags & ETH_RX_ERROR_FALGS)) {
941 DP(NETIF_MSG_RX_ERR | NETIF_MSG_RX_STATUS,
942 "ERROR flags %x rx packet %u\n",
943 cqe_fp_flags, sw_comp_cons);
944 bnx2x_fp_qstats(bp, fp)->rx_err_discard_pkt++;
948 /* Since we don't have a jumbo ring
949 * copy small packets if mtu > 1500
951 if ((bp->dev->mtu > ETH_MAX_PACKET_SIZE) &&
952 (len <= RX_COPY_THRESH)) {
953 skb = netdev_alloc_skb_ip_align(bp->dev, len);
955 DP(NETIF_MSG_RX_ERR | NETIF_MSG_RX_STATUS,
956 "ERROR packet dropped because of alloc failure\n");
957 bnx2x_fp_qstats(bp, fp)->rx_skb_alloc_failed++;
960 memcpy(skb->data, data + pad, len);
961 bnx2x_reuse_rx_data(fp, bd_cons, bd_prod);
963 if (likely(bnx2x_alloc_rx_data(bp, fp, bd_prod) == 0)) {
964 dma_unmap_single(&bp->pdev->dev,
965 dma_unmap_addr(rx_buf, mapping),
968 skb = build_skb(data, fp->rx_frag_size);
969 if (unlikely(!skb)) {
970 bnx2x_frag_free(fp, data);
971 bnx2x_fp_qstats(bp, fp)->
972 rx_skb_alloc_failed++;
975 skb_reserve(skb, pad);
977 DP(NETIF_MSG_RX_ERR | NETIF_MSG_RX_STATUS,
978 "ERROR packet dropped because of alloc failure\n");
979 bnx2x_fp_qstats(bp, fp)->rx_skb_alloc_failed++;
981 bnx2x_reuse_rx_data(fp, bd_cons, bd_prod);
987 skb->protocol = eth_type_trans(skb, bp->dev);
989 /* Set Toeplitz hash for a none-LRO skb */
990 skb->rxhash = bnx2x_get_rxhash(bp, cqe_fp, &l4_rxhash);
991 skb->l4_rxhash = l4_rxhash;
993 skb_checksum_none_assert(skb);
995 if (bp->dev->features & NETIF_F_RXCSUM)
996 bnx2x_csum_validate(skb, cqe, fp,
997 bnx2x_fp_qstats(bp, fp));
999 skb_record_rx_queue(skb, fp->rx_queue);
1001 if (le16_to_cpu(cqe_fp->pars_flags.flags) &
1003 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
1004 le16_to_cpu(cqe_fp->vlan_tag));
1005 napi_gro_receive(&fp->napi, skb);
1009 rx_buf->data = NULL;
1011 bd_cons = NEXT_RX_IDX(bd_cons);
1012 bd_prod = NEXT_RX_IDX(bd_prod);
1013 bd_prod_fw = NEXT_RX_IDX(bd_prod_fw);
1016 sw_comp_prod = NEXT_RCQ_IDX(sw_comp_prod);
1017 sw_comp_cons = NEXT_RCQ_IDX(sw_comp_cons);
1019 if (rx_pkt == budget)
1023 fp->rx_bd_cons = bd_cons;
1024 fp->rx_bd_prod = bd_prod_fw;
1025 fp->rx_comp_cons = sw_comp_cons;
1026 fp->rx_comp_prod = sw_comp_prod;
1028 /* Update producers */
1029 bnx2x_update_rx_prod(bp, fp, bd_prod_fw, sw_comp_prod,
1032 fp->rx_pkt += rx_pkt;
1038 static irqreturn_t bnx2x_msix_fp_int(int irq, void *fp_cookie)
1040 struct bnx2x_fastpath *fp = fp_cookie;
1041 struct bnx2x *bp = fp->bp;
1045 "got an MSI-X interrupt on IDX:SB [fp %d fw_sd %d igusb %d]\n",
1046 fp->index, fp->fw_sb_id, fp->igu_sb_id);
1048 bnx2x_ack_sb(bp, fp->igu_sb_id, USTORM_ID, 0, IGU_INT_DISABLE, 0);
1050 #ifdef BNX2X_STOP_ON_ERROR
1051 if (unlikely(bp->panic))
1055 /* Handle Rx and Tx according to MSI-X vector */
1056 prefetch(fp->rx_cons_sb);
1058 for_each_cos_in_tx_queue(fp, cos)
1059 prefetch(fp->txdata_ptr[cos]->tx_cons_sb);
1061 prefetch(&fp->sb_running_index[SM_RX_ID]);
1062 napi_schedule(&bnx2x_fp(bp, fp->index, napi));
1067 /* HW Lock for shared dual port PHYs */
1068 void bnx2x_acquire_phy_lock(struct bnx2x *bp)
1070 mutex_lock(&bp->port.phy_mutex);
1072 bnx2x_acquire_hw_lock(bp, HW_LOCK_RESOURCE_MDIO);
1075 void bnx2x_release_phy_lock(struct bnx2x *bp)
1077 bnx2x_release_hw_lock(bp, HW_LOCK_RESOURCE_MDIO);
1079 mutex_unlock(&bp->port.phy_mutex);
1082 /* calculates MF speed according to current linespeed and MF configuration */
1083 u16 bnx2x_get_mf_speed(struct bnx2x *bp)
1085 u16 line_speed = bp->link_vars.line_speed;
1087 u16 maxCfg = bnx2x_extract_max_cfg(bp,
1088 bp->mf_config[BP_VN(bp)]);
1090 /* Calculate the current MAX line speed limit for the MF
1094 line_speed = (line_speed * maxCfg) / 100;
1095 else { /* SD mode */
1096 u16 vn_max_rate = maxCfg * 100;
1098 if (vn_max_rate < line_speed)
1099 line_speed = vn_max_rate;
1107 * bnx2x_fill_report_data - fill link report data to report
1109 * @bp: driver handle
1110 * @data: link state to update
1112 * It uses a none-atomic bit operations because is called under the mutex.
1114 static void bnx2x_fill_report_data(struct bnx2x *bp,
1115 struct bnx2x_link_report_data *data)
1117 u16 line_speed = bnx2x_get_mf_speed(bp);
1119 memset(data, 0, sizeof(*data));
1121 /* Fill the report data: efective line speed */
1122 data->line_speed = line_speed;
1125 if (!bp->link_vars.link_up || (bp->flags & MF_FUNC_DIS))
1126 __set_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1127 &data->link_report_flags);
1130 if (bp->link_vars.duplex == DUPLEX_FULL)
1131 __set_bit(BNX2X_LINK_REPORT_FD, &data->link_report_flags);
1133 /* Rx Flow Control is ON */
1134 if (bp->link_vars.flow_ctrl & BNX2X_FLOW_CTRL_RX)
1135 __set_bit(BNX2X_LINK_REPORT_RX_FC_ON, &data->link_report_flags);
1137 /* Tx Flow Control is ON */
1138 if (bp->link_vars.flow_ctrl & BNX2X_FLOW_CTRL_TX)
1139 __set_bit(BNX2X_LINK_REPORT_TX_FC_ON, &data->link_report_flags);
1143 * bnx2x_link_report - report link status to OS.
1145 * @bp: driver handle
1147 * Calls the __bnx2x_link_report() under the same locking scheme
1148 * as a link/PHY state managing code to ensure a consistent link
1152 void bnx2x_link_report(struct bnx2x *bp)
1154 bnx2x_acquire_phy_lock(bp);
1155 __bnx2x_link_report(bp);
1156 bnx2x_release_phy_lock(bp);
1160 * __bnx2x_link_report - report link status to OS.
1162 * @bp: driver handle
1164 * None atomic inmlementation.
1165 * Should be called under the phy_lock.
1167 void __bnx2x_link_report(struct bnx2x *bp)
1169 struct bnx2x_link_report_data cur_data;
1172 if (IS_PF(bp) && !CHIP_IS_E1(bp))
1173 bnx2x_read_mf_cfg(bp);
1175 /* Read the current link report info */
1176 bnx2x_fill_report_data(bp, &cur_data);
1178 /* Don't report link down or exactly the same link status twice */
1179 if (!memcmp(&cur_data, &bp->last_reported_link, sizeof(cur_data)) ||
1180 (test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1181 &bp->last_reported_link.link_report_flags) &&
1182 test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1183 &cur_data.link_report_flags)))
1188 /* We are going to report a new link parameters now -
1189 * remember the current data for the next time.
1191 memcpy(&bp->last_reported_link, &cur_data, sizeof(cur_data));
1193 if (test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1194 &cur_data.link_report_flags)) {
1195 netif_carrier_off(bp->dev);
1196 netdev_err(bp->dev, "NIC Link is Down\n");
1202 netif_carrier_on(bp->dev);
1204 if (test_and_clear_bit(BNX2X_LINK_REPORT_FD,
1205 &cur_data.link_report_flags))
1210 /* Handle the FC at the end so that only these flags would be
1211 * possibly set. This way we may easily check if there is no FC
1214 if (cur_data.link_report_flags) {
1215 if (test_bit(BNX2X_LINK_REPORT_RX_FC_ON,
1216 &cur_data.link_report_flags)) {
1217 if (test_bit(BNX2X_LINK_REPORT_TX_FC_ON,
1218 &cur_data.link_report_flags))
1219 flow = "ON - receive & transmit";
1221 flow = "ON - receive";
1223 flow = "ON - transmit";
1228 netdev_info(bp->dev, "NIC Link is Up, %d Mbps %s duplex, Flow control: %s\n",
1229 cur_data.line_speed, duplex, flow);
1233 static void bnx2x_set_next_page_sgl(struct bnx2x_fastpath *fp)
1237 for (i = 1; i <= NUM_RX_SGE_PAGES; i++) {
1238 struct eth_rx_sge *sge;
1240 sge = &fp->rx_sge_ring[RX_SGE_CNT * i - 2];
1242 cpu_to_le32(U64_HI(fp->rx_sge_mapping +
1243 BCM_PAGE_SIZE*(i % NUM_RX_SGE_PAGES)));
1246 cpu_to_le32(U64_LO(fp->rx_sge_mapping +
1247 BCM_PAGE_SIZE*(i % NUM_RX_SGE_PAGES)));
1251 static void bnx2x_free_tpa_pool(struct bnx2x *bp,
1252 struct bnx2x_fastpath *fp, int last)
1256 for (i = 0; i < last; i++) {
1257 struct bnx2x_agg_info *tpa_info = &fp->tpa_info[i];
1258 struct sw_rx_bd *first_buf = &tpa_info->first_buf;
1259 u8 *data = first_buf->data;
1262 DP(NETIF_MSG_IFDOWN, "tpa bin %d empty on free\n", i);
1265 if (tpa_info->tpa_state == BNX2X_TPA_START)
1266 dma_unmap_single(&bp->pdev->dev,
1267 dma_unmap_addr(first_buf, mapping),
1268 fp->rx_buf_size, DMA_FROM_DEVICE);
1269 bnx2x_frag_free(fp, data);
1270 first_buf->data = NULL;
1274 void bnx2x_init_rx_rings_cnic(struct bnx2x *bp)
1278 for_each_rx_queue_cnic(bp, j) {
1279 struct bnx2x_fastpath *fp = &bp->fp[j];
1283 /* Activate BD ring */
1285 * this will generate an interrupt (to the TSTORM)
1286 * must only be done after chip is initialized
1288 bnx2x_update_rx_prod(bp, fp, fp->rx_bd_prod, fp->rx_comp_prod,
1293 void bnx2x_init_rx_rings(struct bnx2x *bp)
1295 int func = BP_FUNC(bp);
1299 /* Allocate TPA resources */
1300 for_each_eth_queue(bp, j) {
1301 struct bnx2x_fastpath *fp = &bp->fp[j];
1304 "mtu %d rx_buf_size %d\n", bp->dev->mtu, fp->rx_buf_size);
1306 if (!fp->disable_tpa) {
1307 /* Fill the per-aggregtion pool */
1308 for (i = 0; i < MAX_AGG_QS(bp); i++) {
1309 struct bnx2x_agg_info *tpa_info =
1311 struct sw_rx_bd *first_buf =
1312 &tpa_info->first_buf;
1314 first_buf->data = bnx2x_frag_alloc(fp);
1315 if (!first_buf->data) {
1316 BNX2X_ERR("Failed to allocate TPA skb pool for queue[%d] - disabling TPA on this queue!\n",
1318 bnx2x_free_tpa_pool(bp, fp, i);
1319 fp->disable_tpa = 1;
1322 dma_unmap_addr_set(first_buf, mapping, 0);
1323 tpa_info->tpa_state = BNX2X_TPA_STOP;
1326 /* "next page" elements initialization */
1327 bnx2x_set_next_page_sgl(fp);
1329 /* set SGEs bit mask */
1330 bnx2x_init_sge_ring_bit_mask(fp);
1332 /* Allocate SGEs and initialize the ring elements */
1333 for (i = 0, ring_prod = 0;
1334 i < MAX_RX_SGE_CNT*NUM_RX_SGE_PAGES; i++) {
1336 if (bnx2x_alloc_rx_sge(bp, fp, ring_prod) < 0) {
1337 BNX2X_ERR("was only able to allocate %d rx sges\n",
1339 BNX2X_ERR("disabling TPA for queue[%d]\n",
1341 /* Cleanup already allocated elements */
1342 bnx2x_free_rx_sge_range(bp, fp,
1344 bnx2x_free_tpa_pool(bp, fp,
1346 fp->disable_tpa = 1;
1350 ring_prod = NEXT_SGE_IDX(ring_prod);
1353 fp->rx_sge_prod = ring_prod;
1357 for_each_eth_queue(bp, j) {
1358 struct bnx2x_fastpath *fp = &bp->fp[j];
1362 /* Activate BD ring */
1364 * this will generate an interrupt (to the TSTORM)
1365 * must only be done after chip is initialized
1367 bnx2x_update_rx_prod(bp, fp, fp->rx_bd_prod, fp->rx_comp_prod,
1373 if (CHIP_IS_E1(bp)) {
1374 REG_WR(bp, BAR_USTRORM_INTMEM +
1375 USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func),
1376 U64_LO(fp->rx_comp_mapping));
1377 REG_WR(bp, BAR_USTRORM_INTMEM +
1378 USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func) + 4,
1379 U64_HI(fp->rx_comp_mapping));
1384 static void bnx2x_free_tx_skbs_queue(struct bnx2x_fastpath *fp)
1387 struct bnx2x *bp = fp->bp;
1389 for_each_cos_in_tx_queue(fp, cos) {
1390 struct bnx2x_fp_txdata *txdata = fp->txdata_ptr[cos];
1391 unsigned pkts_compl = 0, bytes_compl = 0;
1393 u16 sw_prod = txdata->tx_pkt_prod;
1394 u16 sw_cons = txdata->tx_pkt_cons;
1396 while (sw_cons != sw_prod) {
1397 bnx2x_free_tx_pkt(bp, txdata, TX_BD(sw_cons),
1398 &pkts_compl, &bytes_compl);
1402 netdev_tx_reset_queue(
1403 netdev_get_tx_queue(bp->dev,
1404 txdata->txq_index));
1408 static void bnx2x_free_tx_skbs_cnic(struct bnx2x *bp)
1412 for_each_tx_queue_cnic(bp, i) {
1413 bnx2x_free_tx_skbs_queue(&bp->fp[i]);
1417 static void bnx2x_free_tx_skbs(struct bnx2x *bp)
1421 for_each_eth_queue(bp, i) {
1422 bnx2x_free_tx_skbs_queue(&bp->fp[i]);
1426 static void bnx2x_free_rx_bds(struct bnx2x_fastpath *fp)
1428 struct bnx2x *bp = fp->bp;
1431 /* ring wasn't allocated */
1432 if (fp->rx_buf_ring == NULL)
1435 for (i = 0; i < NUM_RX_BD; i++) {
1436 struct sw_rx_bd *rx_buf = &fp->rx_buf_ring[i];
1437 u8 *data = rx_buf->data;
1441 dma_unmap_single(&bp->pdev->dev,
1442 dma_unmap_addr(rx_buf, mapping),
1443 fp->rx_buf_size, DMA_FROM_DEVICE);
1445 rx_buf->data = NULL;
1446 bnx2x_frag_free(fp, data);
1450 static void bnx2x_free_rx_skbs_cnic(struct bnx2x *bp)
1454 for_each_rx_queue_cnic(bp, j) {
1455 bnx2x_free_rx_bds(&bp->fp[j]);
1459 static void bnx2x_free_rx_skbs(struct bnx2x *bp)
1463 for_each_eth_queue(bp, j) {
1464 struct bnx2x_fastpath *fp = &bp->fp[j];
1466 bnx2x_free_rx_bds(fp);
1468 if (!fp->disable_tpa)
1469 bnx2x_free_tpa_pool(bp, fp, MAX_AGG_QS(bp));
1473 void bnx2x_free_skbs_cnic(struct bnx2x *bp)
1475 bnx2x_free_tx_skbs_cnic(bp);
1476 bnx2x_free_rx_skbs_cnic(bp);
1479 void bnx2x_free_skbs(struct bnx2x *bp)
1481 bnx2x_free_tx_skbs(bp);
1482 bnx2x_free_rx_skbs(bp);
1485 void bnx2x_update_max_mf_config(struct bnx2x *bp, u32 value)
1487 /* load old values */
1488 u32 mf_cfg = bp->mf_config[BP_VN(bp)];
1490 if (value != bnx2x_extract_max_cfg(bp, mf_cfg)) {
1491 /* leave all but MAX value */
1492 mf_cfg &= ~FUNC_MF_CFG_MAX_BW_MASK;
1494 /* set new MAX value */
1495 mf_cfg |= (value << FUNC_MF_CFG_MAX_BW_SHIFT)
1496 & FUNC_MF_CFG_MAX_BW_MASK;
1498 bnx2x_fw_command(bp, DRV_MSG_CODE_SET_MF_BW, mf_cfg);
1503 * bnx2x_free_msix_irqs - free previously requested MSI-X IRQ vectors
1505 * @bp: driver handle
1506 * @nvecs: number of vectors to be released
1508 static void bnx2x_free_msix_irqs(struct bnx2x *bp, int nvecs)
1512 if (nvecs == offset)
1515 /* VFs don't have a default SB */
1517 free_irq(bp->msix_table[offset].vector, bp->dev);
1518 DP(NETIF_MSG_IFDOWN, "released sp irq (%d)\n",
1519 bp->msix_table[offset].vector);
1523 if (CNIC_SUPPORT(bp)) {
1524 if (nvecs == offset)
1529 for_each_eth_queue(bp, i) {
1530 if (nvecs == offset)
1532 DP(NETIF_MSG_IFDOWN, "about to release fp #%d->%d irq\n",
1533 i, bp->msix_table[offset].vector);
1535 free_irq(bp->msix_table[offset++].vector, &bp->fp[i]);
1539 void bnx2x_free_irq(struct bnx2x *bp)
1541 if (bp->flags & USING_MSIX_FLAG &&
1542 !(bp->flags & USING_SINGLE_MSIX_FLAG)) {
1543 int nvecs = BNX2X_NUM_ETH_QUEUES(bp) + CNIC_SUPPORT(bp);
1545 /* vfs don't have a default status block */
1549 bnx2x_free_msix_irqs(bp, nvecs);
1551 free_irq(bp->dev->irq, bp->dev);
1555 int bnx2x_enable_msix(struct bnx2x *bp)
1557 int msix_vec = 0, i, rc;
1559 /* VFs don't have a default status block */
1561 bp->msix_table[msix_vec].entry = msix_vec;
1562 BNX2X_DEV_INFO("msix_table[0].entry = %d (slowpath)\n",
1563 bp->msix_table[0].entry);
1567 /* Cnic requires an msix vector for itself */
1568 if (CNIC_SUPPORT(bp)) {
1569 bp->msix_table[msix_vec].entry = msix_vec;
1570 BNX2X_DEV_INFO("msix_table[%d].entry = %d (CNIC)\n",
1571 msix_vec, bp->msix_table[msix_vec].entry);
1575 /* We need separate vectors for ETH queues only (not FCoE) */
1576 for_each_eth_queue(bp, i) {
1577 bp->msix_table[msix_vec].entry = msix_vec;
1578 BNX2X_DEV_INFO("msix_table[%d].entry = %d (fastpath #%u)\n",
1579 msix_vec, msix_vec, i);
1583 DP(BNX2X_MSG_SP, "about to request enable msix with %d vectors\n",
1586 rc = pci_enable_msix(bp->pdev, &bp->msix_table[0], msix_vec);
1589 * reconfigure number of tx/rx queues according to available
1592 if (rc >= BNX2X_MIN_MSIX_VEC_CNT(bp)) {
1593 /* how less vectors we will have? */
1594 int diff = msix_vec - rc;
1596 BNX2X_DEV_INFO("Trying to use less MSI-X vectors: %d\n", rc);
1598 rc = pci_enable_msix(bp->pdev, &bp->msix_table[0], rc);
1601 BNX2X_DEV_INFO("MSI-X is not attainable rc %d\n", rc);
1605 * decrease number of queues by number of unallocated entries
1607 bp->num_ethernet_queues -= diff;
1608 bp->num_queues = bp->num_ethernet_queues + bp->num_cnic_queues;
1610 BNX2X_DEV_INFO("New queue configuration set: %d\n",
1612 } else if (rc > 0) {
1613 /* Get by with single vector */
1614 rc = pci_enable_msix(bp->pdev, &bp->msix_table[0], 1);
1616 BNX2X_DEV_INFO("Single MSI-X is not attainable rc %d\n",
1621 BNX2X_DEV_INFO("Using single MSI-X vector\n");
1622 bp->flags |= USING_SINGLE_MSIX_FLAG;
1624 BNX2X_DEV_INFO("set number of queues to 1\n");
1625 bp->num_ethernet_queues = 1;
1626 bp->num_queues = bp->num_ethernet_queues + bp->num_cnic_queues;
1627 } else if (rc < 0) {
1628 BNX2X_DEV_INFO("MSI-X is not attainable rc %d\n", rc);
1632 bp->flags |= USING_MSIX_FLAG;
1637 /* fall to INTx if not enough memory */
1639 bp->flags |= DISABLE_MSI_FLAG;
1644 static int bnx2x_req_msix_irqs(struct bnx2x *bp)
1646 int i, rc, offset = 0;
1648 /* no default status block for vf */
1650 rc = request_irq(bp->msix_table[offset++].vector,
1651 bnx2x_msix_sp_int, 0,
1652 bp->dev->name, bp->dev);
1654 BNX2X_ERR("request sp irq failed\n");
1659 if (CNIC_SUPPORT(bp))
1662 for_each_eth_queue(bp, i) {
1663 struct bnx2x_fastpath *fp = &bp->fp[i];
1664 snprintf(fp->name, sizeof(fp->name), "%s-fp-%d",
1667 rc = request_irq(bp->msix_table[offset].vector,
1668 bnx2x_msix_fp_int, 0, fp->name, fp);
1670 BNX2X_ERR("request fp #%d irq (%d) failed rc %d\n", i,
1671 bp->msix_table[offset].vector, rc);
1672 bnx2x_free_msix_irqs(bp, offset);
1679 i = BNX2X_NUM_ETH_QUEUES(bp);
1681 offset = 1 + CNIC_SUPPORT(bp);
1682 netdev_info(bp->dev,
1683 "using MSI-X IRQs: sp %d fp[%d] %d ... fp[%d] %d\n",
1684 bp->msix_table[0].vector,
1685 0, bp->msix_table[offset].vector,
1686 i - 1, bp->msix_table[offset + i - 1].vector);
1688 offset = CNIC_SUPPORT(bp);
1689 netdev_info(bp->dev,
1690 "using MSI-X IRQs: fp[%d] %d ... fp[%d] %d\n",
1691 0, bp->msix_table[offset].vector,
1692 i - 1, bp->msix_table[offset + i - 1].vector);
1697 int bnx2x_enable_msi(struct bnx2x *bp)
1701 rc = pci_enable_msi(bp->pdev);
1703 BNX2X_DEV_INFO("MSI is not attainable\n");
1706 bp->flags |= USING_MSI_FLAG;
1711 static int bnx2x_req_irq(struct bnx2x *bp)
1713 unsigned long flags;
1716 if (bp->flags & (USING_MSI_FLAG | USING_MSIX_FLAG))
1719 flags = IRQF_SHARED;
1721 if (bp->flags & USING_MSIX_FLAG)
1722 irq = bp->msix_table[0].vector;
1724 irq = bp->pdev->irq;
1726 return request_irq(irq, bnx2x_interrupt, flags, bp->dev->name, bp->dev);
1729 int bnx2x_setup_irqs(struct bnx2x *bp)
1732 if (bp->flags & USING_MSIX_FLAG &&
1733 !(bp->flags & USING_SINGLE_MSIX_FLAG)) {
1734 rc = bnx2x_req_msix_irqs(bp);
1738 rc = bnx2x_req_irq(bp);
1740 BNX2X_ERR("IRQ request failed rc %d, aborting\n", rc);
1743 if (bp->flags & USING_MSI_FLAG) {
1744 bp->dev->irq = bp->pdev->irq;
1745 netdev_info(bp->dev, "using MSI IRQ %d\n",
1748 if (bp->flags & USING_MSIX_FLAG) {
1749 bp->dev->irq = bp->msix_table[0].vector;
1750 netdev_info(bp->dev, "using MSIX IRQ %d\n",
1758 static void bnx2x_napi_enable_cnic(struct bnx2x *bp)
1762 for_each_rx_queue_cnic(bp, i)
1763 napi_enable(&bnx2x_fp(bp, i, napi));
1766 static void bnx2x_napi_enable(struct bnx2x *bp)
1770 for_each_eth_queue(bp, i)
1771 napi_enable(&bnx2x_fp(bp, i, napi));
1774 static void bnx2x_napi_disable_cnic(struct bnx2x *bp)
1778 for_each_rx_queue_cnic(bp, i)
1779 napi_disable(&bnx2x_fp(bp, i, napi));
1782 static void bnx2x_napi_disable(struct bnx2x *bp)
1786 for_each_eth_queue(bp, i)
1787 napi_disable(&bnx2x_fp(bp, i, napi));
1790 void bnx2x_netif_start(struct bnx2x *bp)
1792 if (netif_running(bp->dev)) {
1793 bnx2x_napi_enable(bp);
1794 if (CNIC_LOADED(bp))
1795 bnx2x_napi_enable_cnic(bp);
1796 bnx2x_int_enable(bp);
1797 if (bp->state == BNX2X_STATE_OPEN)
1798 netif_tx_wake_all_queues(bp->dev);
1802 void bnx2x_netif_stop(struct bnx2x *bp, int disable_hw)
1804 bnx2x_int_disable_sync(bp, disable_hw);
1805 bnx2x_napi_disable(bp);
1806 if (CNIC_LOADED(bp))
1807 bnx2x_napi_disable_cnic(bp);
1810 u16 bnx2x_select_queue(struct net_device *dev, struct sk_buff *skb)
1812 struct bnx2x *bp = netdev_priv(dev);
1814 if (CNIC_LOADED(bp) && !NO_FCOE(bp)) {
1815 struct ethhdr *hdr = (struct ethhdr *)skb->data;
1816 u16 ether_type = ntohs(hdr->h_proto);
1818 /* Skip VLAN tag if present */
1819 if (ether_type == ETH_P_8021Q) {
1820 struct vlan_ethhdr *vhdr =
1821 (struct vlan_ethhdr *)skb->data;
1823 ether_type = ntohs(vhdr->h_vlan_encapsulated_proto);
1826 /* If ethertype is FCoE or FIP - use FCoE ring */
1827 if ((ether_type == ETH_P_FCOE) || (ether_type == ETH_P_FIP))
1828 return bnx2x_fcoe_tx(bp, txq_index);
1831 /* select a non-FCoE queue */
1832 return __netdev_pick_tx(dev, skb) % BNX2X_NUM_ETH_QUEUES(bp);
1835 void bnx2x_set_num_queues(struct bnx2x *bp)
1838 bp->num_ethernet_queues = bnx2x_calc_num_queues(bp);
1840 /* override in STORAGE SD modes */
1841 if (IS_MF_STORAGE_SD(bp) || IS_MF_FCOE_AFEX(bp))
1842 bp->num_ethernet_queues = 1;
1844 /* Add special queues */
1845 bp->num_cnic_queues = CNIC_SUPPORT(bp); /* For FCOE */
1846 bp->num_queues = bp->num_ethernet_queues + bp->num_cnic_queues;
1848 BNX2X_DEV_INFO("set number of queues to %d\n", bp->num_queues);
1852 * bnx2x_set_real_num_queues - configure netdev->real_num_[tx,rx]_queues
1854 * @bp: Driver handle
1856 * We currently support for at most 16 Tx queues for each CoS thus we will
1857 * allocate a multiple of 16 for ETH L2 rings according to the value of the
1860 * If there is an FCoE L2 queue the appropriate Tx queue will have the next
1861 * index after all ETH L2 indices.
1863 * If the actual number of Tx queues (for each CoS) is less than 16 then there
1864 * will be the holes at the end of each group of 16 ETh L2 indices (0..15,
1865 * 16..31,...) with indicies that are not coupled with any real Tx queue.
1867 * The proper configuration of skb->queue_mapping is handled by
1868 * bnx2x_select_queue() and __skb_tx_hash().
1870 * bnx2x_setup_tc() takes care of the proper TC mappings so that __skb_tx_hash()
1871 * will return a proper Tx index if TC is enabled (netdev->num_tc > 0).
1873 static int bnx2x_set_real_num_queues(struct bnx2x *bp, int include_cnic)
1877 tx = BNX2X_NUM_ETH_QUEUES(bp) * bp->max_cos;
1878 rx = BNX2X_NUM_ETH_QUEUES(bp);
1880 /* account for fcoe queue */
1881 if (include_cnic && !NO_FCOE(bp)) {
1886 rc = netif_set_real_num_tx_queues(bp->dev, tx);
1888 BNX2X_ERR("Failed to set real number of Tx queues: %d\n", rc);
1891 rc = netif_set_real_num_rx_queues(bp->dev, rx);
1893 BNX2X_ERR("Failed to set real number of Rx queues: %d\n", rc);
1897 DP(NETIF_MSG_IFUP, "Setting real num queues to (tx, rx) (%d, %d)\n",
1903 static void bnx2x_set_rx_buf_size(struct bnx2x *bp)
1907 for_each_queue(bp, i) {
1908 struct bnx2x_fastpath *fp = &bp->fp[i];
1911 /* Always use a mini-jumbo MTU for the FCoE L2 ring */
1914 * Although there are no IP frames expected to arrive to
1915 * this ring we still want to add an
1916 * IP_HEADER_ALIGNMENT_PADDING to prevent a buffer
1919 mtu = BNX2X_FCOE_MINI_JUMBO_MTU;
1922 fp->rx_buf_size = BNX2X_FW_RX_ALIGN_START +
1923 IP_HEADER_ALIGNMENT_PADDING +
1926 BNX2X_FW_RX_ALIGN_END;
1927 /* Note : rx_buf_size doesnt take into account NET_SKB_PAD */
1928 if (fp->rx_buf_size + NET_SKB_PAD <= PAGE_SIZE)
1929 fp->rx_frag_size = fp->rx_buf_size + NET_SKB_PAD;
1931 fp->rx_frag_size = 0;
1935 static int bnx2x_init_rss_pf(struct bnx2x *bp)
1938 u8 num_eth_queues = BNX2X_NUM_ETH_QUEUES(bp);
1940 /* Prepare the initial contents fo the indirection table if RSS is
1943 for (i = 0; i < sizeof(bp->rss_conf_obj.ind_table); i++)
1944 bp->rss_conf_obj.ind_table[i] =
1946 ethtool_rxfh_indir_default(i, num_eth_queues);
1949 * For 57710 and 57711 SEARCHER configuration (rss_keys) is
1950 * per-port, so if explicit configuration is needed , do it only
1953 * For 57712 and newer on the other hand it's a per-function
1956 return bnx2x_config_rss_eth(bp, bp->port.pmf || !CHIP_IS_E1x(bp));
1959 int bnx2x_config_rss_pf(struct bnx2x *bp, struct bnx2x_rss_config_obj *rss_obj,
1962 struct bnx2x_config_rss_params params = {NULL};
1964 /* Although RSS is meaningless when there is a single HW queue we
1965 * still need it enabled in order to have HW Rx hash generated.
1967 * if (!is_eth_multi(bp))
1968 * bp->multi_mode = ETH_RSS_MODE_DISABLED;
1971 params.rss_obj = rss_obj;
1973 __set_bit(RAMROD_COMP_WAIT, ¶ms.ramrod_flags);
1975 __set_bit(BNX2X_RSS_MODE_REGULAR, ¶ms.rss_flags);
1977 /* RSS configuration */
1978 __set_bit(BNX2X_RSS_IPV4, ¶ms.rss_flags);
1979 __set_bit(BNX2X_RSS_IPV4_TCP, ¶ms.rss_flags);
1980 __set_bit(BNX2X_RSS_IPV6, ¶ms.rss_flags);
1981 __set_bit(BNX2X_RSS_IPV6_TCP, ¶ms.rss_flags);
1982 if (rss_obj->udp_rss_v4)
1983 __set_bit(BNX2X_RSS_IPV4_UDP, ¶ms.rss_flags);
1984 if (rss_obj->udp_rss_v6)
1985 __set_bit(BNX2X_RSS_IPV6_UDP, ¶ms.rss_flags);
1988 params.rss_result_mask = MULTI_MASK;
1990 memcpy(params.ind_table, rss_obj->ind_table, sizeof(params.ind_table));
1994 prandom_bytes(params.rss_key, sizeof(params.rss_key));
1995 __set_bit(BNX2X_RSS_SET_SRCH, ¶ms.rss_flags);
1998 return bnx2x_config_rss(bp, ¶ms);
2001 static int bnx2x_init_hw(struct bnx2x *bp, u32 load_code)
2003 struct bnx2x_func_state_params func_params = {NULL};
2005 /* Prepare parameters for function state transitions */
2006 __set_bit(RAMROD_COMP_WAIT, &func_params.ramrod_flags);
2008 func_params.f_obj = &bp->func_obj;
2009 func_params.cmd = BNX2X_F_CMD_HW_INIT;
2011 func_params.params.hw_init.load_phase = load_code;
2013 return bnx2x_func_state_change(bp, &func_params);
2017 * Cleans the object that have internal lists without sending
2018 * ramrods. Should be run when interrutps are disabled.
2020 void bnx2x_squeeze_objects(struct bnx2x *bp)
2023 unsigned long ramrod_flags = 0, vlan_mac_flags = 0;
2024 struct bnx2x_mcast_ramrod_params rparam = {NULL};
2025 struct bnx2x_vlan_mac_obj *mac_obj = &bp->sp_objs->mac_obj;
2027 /***************** Cleanup MACs' object first *************************/
2029 /* Wait for completion of requested */
2030 __set_bit(RAMROD_COMP_WAIT, &ramrod_flags);
2031 /* Perform a dry cleanup */
2032 __set_bit(RAMROD_DRV_CLR_ONLY, &ramrod_flags);
2034 /* Clean ETH primary MAC */
2035 __set_bit(BNX2X_ETH_MAC, &vlan_mac_flags);
2036 rc = mac_obj->delete_all(bp, &bp->sp_objs->mac_obj, &vlan_mac_flags,
2039 BNX2X_ERR("Failed to clean ETH MACs: %d\n", rc);
2041 /* Cleanup UC list */
2043 __set_bit(BNX2X_UC_LIST_MAC, &vlan_mac_flags);
2044 rc = mac_obj->delete_all(bp, mac_obj, &vlan_mac_flags,
2047 BNX2X_ERR("Failed to clean UC list MACs: %d\n", rc);
2049 /***************** Now clean mcast object *****************************/
2050 rparam.mcast_obj = &bp->mcast_obj;
2051 __set_bit(RAMROD_DRV_CLR_ONLY, &rparam.ramrod_flags);
2053 /* Add a DEL command... */
2054 rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_DEL);
2056 BNX2X_ERR("Failed to add a new DEL command to a multi-cast object: %d\n",
2059 /* ...and wait until all pending commands are cleared */
2060 rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_CONT);
2063 BNX2X_ERR("Failed to clean multi-cast object: %d\n",
2068 rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_CONT);
2072 #ifndef BNX2X_STOP_ON_ERROR
2073 #define LOAD_ERROR_EXIT(bp, label) \
2075 (bp)->state = BNX2X_STATE_ERROR; \
2079 #define LOAD_ERROR_EXIT_CNIC(bp, label) \
2081 bp->cnic_loaded = false; \
2084 #else /*BNX2X_STOP_ON_ERROR*/
2085 #define LOAD_ERROR_EXIT(bp, label) \
2087 (bp)->state = BNX2X_STATE_ERROR; \
2091 #define LOAD_ERROR_EXIT_CNIC(bp, label) \
2093 bp->cnic_loaded = false; \
2097 #endif /*BNX2X_STOP_ON_ERROR*/
2099 static void bnx2x_free_fw_stats_mem(struct bnx2x *bp)
2101 BNX2X_PCI_FREE(bp->fw_stats, bp->fw_stats_mapping,
2102 bp->fw_stats_data_sz + bp->fw_stats_req_sz);
2106 static int bnx2x_alloc_fw_stats_mem(struct bnx2x *bp)
2108 int num_groups, vf_headroom = 0;
2109 int is_fcoe_stats = NO_FCOE(bp) ? 0 : 1;
2111 /* number of queues for statistics is number of eth queues + FCoE */
2112 u8 num_queue_stats = BNX2X_NUM_ETH_QUEUES(bp) + is_fcoe_stats;
2114 /* Total number of FW statistics requests =
2115 * 1 for port stats + 1 for PF stats + potential 2 for FCoE (fcoe proper
2116 * and fcoe l2 queue) stats + num of queues (which includes another 1
2117 * for fcoe l2 queue if applicable)
2119 bp->fw_stats_num = 2 + is_fcoe_stats + num_queue_stats;
2121 /* vf stats appear in the request list, but their data is allocated by
2122 * the VFs themselves. We don't include them in the bp->fw_stats_num as
2123 * it is used to determine where to place the vf stats queries in the
2127 vf_headroom = bnx2x_vf_headroom(bp);
2129 /* Request is built from stats_query_header and an array of
2130 * stats_query_cmd_group each of which contains
2131 * STATS_QUERY_CMD_COUNT rules. The real number or requests is
2132 * configured in the stats_query_header.
2135 (((bp->fw_stats_num + vf_headroom) / STATS_QUERY_CMD_COUNT) +
2136 (((bp->fw_stats_num + vf_headroom) % STATS_QUERY_CMD_COUNT) ?
2139 DP(BNX2X_MSG_SP, "stats fw_stats_num %d, vf headroom %d, num_groups %d\n",
2140 bp->fw_stats_num, vf_headroom, num_groups);
2141 bp->fw_stats_req_sz = sizeof(struct stats_query_header) +
2142 num_groups * sizeof(struct stats_query_cmd_group);
2144 /* Data for statistics requests + stats_counter
2145 * stats_counter holds per-STORM counters that are incremented
2146 * when STORM has finished with the current request.
2147 * memory for FCoE offloaded statistics are counted anyway,
2148 * even if they will not be sent.
2149 * VF stats are not accounted for here as the data of VF stats is stored
2150 * in memory allocated by the VF, not here.
2152 bp->fw_stats_data_sz = sizeof(struct per_port_stats) +
2153 sizeof(struct per_pf_stats) +
2154 sizeof(struct fcoe_statistics_params) +
2155 sizeof(struct per_queue_stats) * num_queue_stats +
2156 sizeof(struct stats_counter);
2158 BNX2X_PCI_ALLOC(bp->fw_stats, &bp->fw_stats_mapping,
2159 bp->fw_stats_data_sz + bp->fw_stats_req_sz);
2162 bp->fw_stats_req = (struct bnx2x_fw_stats_req *)bp->fw_stats;
2163 bp->fw_stats_req_mapping = bp->fw_stats_mapping;
2164 bp->fw_stats_data = (struct bnx2x_fw_stats_data *)
2165 ((u8 *)bp->fw_stats + bp->fw_stats_req_sz);
2166 bp->fw_stats_data_mapping = bp->fw_stats_mapping +
2167 bp->fw_stats_req_sz;
2169 DP(BNX2X_MSG_SP, "statistics request base address set to %x %x",
2170 U64_HI(bp->fw_stats_req_mapping),
2171 U64_LO(bp->fw_stats_req_mapping));
2172 DP(BNX2X_MSG_SP, "statistics data base address set to %x %x",
2173 U64_HI(bp->fw_stats_data_mapping),
2174 U64_LO(bp->fw_stats_data_mapping));
2178 bnx2x_free_fw_stats_mem(bp);
2179 BNX2X_ERR("Can't allocate FW stats memory\n");
2183 /* send load request to mcp and analyze response */
2184 static int bnx2x_nic_load_request(struct bnx2x *bp, u32 *load_code)
2190 (SHMEM_RD(bp, func_mb[BP_FW_MB_IDX(bp)].drv_mb_header) &
2191 DRV_MSG_SEQ_NUMBER_MASK);
2192 BNX2X_DEV_INFO("fw_seq 0x%08x\n", bp->fw_seq);
2194 /* Get current FW pulse sequence */
2195 bp->fw_drv_pulse_wr_seq =
2196 (SHMEM_RD(bp, func_mb[BP_FW_MB_IDX(bp)].drv_pulse_mb) &
2197 DRV_PULSE_SEQ_MASK);
2198 BNX2X_DEV_INFO("drv_pulse 0x%x\n", bp->fw_drv_pulse_wr_seq);
2200 param = DRV_MSG_CODE_LOAD_REQ_WITH_LFA;
2202 if (IS_MF_SD(bp) && bnx2x_port_after_undi(bp))
2203 param |= DRV_MSG_CODE_LOAD_REQ_FORCE_LFA;
2206 (*load_code) = bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_REQ, param);
2208 /* if mcp fails to respond we must abort */
2209 if (!(*load_code)) {
2210 BNX2X_ERR("MCP response failure, aborting\n");
2214 /* If mcp refused (e.g. other port is in diagnostic mode) we
2217 if ((*load_code) == FW_MSG_CODE_DRV_LOAD_REFUSED) {
2218 BNX2X_ERR("MCP refused load request, aborting\n");
2224 /* check whether another PF has already loaded FW to chip. In
2225 * virtualized environments a pf from another VM may have already
2226 * initialized the device including loading FW
2228 int bnx2x_nic_load_analyze_req(struct bnx2x *bp, u32 load_code)
2230 /* is another pf loaded on this engine? */
2231 if (load_code != FW_MSG_CODE_DRV_LOAD_COMMON_CHIP &&
2232 load_code != FW_MSG_CODE_DRV_LOAD_COMMON) {
2233 /* build my FW version dword */
2234 u32 my_fw = (BCM_5710_FW_MAJOR_VERSION) +
2235 (BCM_5710_FW_MINOR_VERSION << 8) +
2236 (BCM_5710_FW_REVISION_VERSION << 16) +
2237 (BCM_5710_FW_ENGINEERING_VERSION << 24);
2239 /* read loaded FW from chip */
2240 u32 loaded_fw = REG_RD(bp, XSEM_REG_PRAM);
2242 DP(BNX2X_MSG_SP, "loaded fw %x, my fw %x\n",
2245 /* abort nic load if version mismatch */
2246 if (my_fw != loaded_fw) {
2247 BNX2X_ERR("bnx2x with FW %x was already loaded which mismatches my %x FW. aborting\n",
2255 /* returns the "mcp load_code" according to global load_count array */
2256 static int bnx2x_nic_load_no_mcp(struct bnx2x *bp, int port)
2258 int path = BP_PATH(bp);
2260 DP(NETIF_MSG_IFUP, "NO MCP - load counts[%d] %d, %d, %d\n",
2261 path, load_count[path][0], load_count[path][1],
2262 load_count[path][2]);
2263 load_count[path][0]++;
2264 load_count[path][1 + port]++;
2265 DP(NETIF_MSG_IFUP, "NO MCP - new load counts[%d] %d, %d, %d\n",
2266 path, load_count[path][0], load_count[path][1],
2267 load_count[path][2]);
2268 if (load_count[path][0] == 1)
2269 return FW_MSG_CODE_DRV_LOAD_COMMON;
2270 else if (load_count[path][1 + port] == 1)
2271 return FW_MSG_CODE_DRV_LOAD_PORT;
2273 return FW_MSG_CODE_DRV_LOAD_FUNCTION;
2276 /* mark PMF if applicable */
2277 static void bnx2x_nic_load_pmf(struct bnx2x *bp, u32 load_code)
2279 if ((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
2280 (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP) ||
2281 (load_code == FW_MSG_CODE_DRV_LOAD_PORT)) {
2283 /* We need the barrier to ensure the ordering between the
2284 * writing to bp->port.pmf here and reading it from the
2285 * bnx2x_periodic_task().
2292 DP(NETIF_MSG_LINK, "pmf %d\n", bp->port.pmf);
2295 static void bnx2x_nic_load_afex_dcc(struct bnx2x *bp, int load_code)
2297 if (((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
2298 (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP)) &&
2299 (bp->common.shmem2_base)) {
2300 if (SHMEM2_HAS(bp, dcc_support))
2301 SHMEM2_WR(bp, dcc_support,
2302 (SHMEM_DCC_SUPPORT_DISABLE_ENABLE_PF_TLV |
2303 SHMEM_DCC_SUPPORT_BANDWIDTH_ALLOCATION_TLV));
2304 if (SHMEM2_HAS(bp, afex_driver_support))
2305 SHMEM2_WR(bp, afex_driver_support,
2306 SHMEM_AFEX_SUPPORTED_VERSION_ONE);
2309 /* Set AFEX default VLAN tag to an invalid value */
2310 bp->afex_def_vlan_tag = -1;
2314 * bnx2x_bz_fp - zero content of the fastpath structure.
2316 * @bp: driver handle
2317 * @index: fastpath index to be zeroed
2319 * Makes sure the contents of the bp->fp[index].napi is kept
2322 static void bnx2x_bz_fp(struct bnx2x *bp, int index)
2324 struct bnx2x_fastpath *fp = &bp->fp[index];
2327 struct napi_struct orig_napi = fp->napi;
2328 struct bnx2x_agg_info *orig_tpa_info = fp->tpa_info;
2329 /* bzero bnx2x_fastpath contents */
2331 memset(fp->tpa_info, 0, ETH_MAX_AGGREGATION_QUEUES_E1H_E2 *
2332 sizeof(struct bnx2x_agg_info));
2333 memset(fp, 0, sizeof(*fp));
2335 /* Restore the NAPI object as it has been already initialized */
2336 fp->napi = orig_napi;
2337 fp->tpa_info = orig_tpa_info;
2341 fp->max_cos = bp->max_cos;
2343 /* Special queues support only one CoS */
2346 /* Init txdata pointers */
2348 fp->txdata_ptr[0] = &bp->bnx2x_txq[FCOE_TXQ_IDX(bp)];
2350 for_each_cos_in_tx_queue(fp, cos)
2351 fp->txdata_ptr[cos] = &bp->bnx2x_txq[cos *
2352 BNX2X_NUM_ETH_QUEUES(bp) + index];
2355 * set the tpa flag for each queue. The tpa flag determines the queue
2356 * minimal size so it must be set prior to queue memory allocation
2358 fp->disable_tpa = !(bp->flags & TPA_ENABLE_FLAG ||
2359 (bp->flags & GRO_ENABLE_FLAG &&
2360 bnx2x_mtu_allows_gro(bp->dev->mtu)));
2361 if (bp->flags & TPA_ENABLE_FLAG)
2362 fp->mode = TPA_MODE_LRO;
2363 else if (bp->flags & GRO_ENABLE_FLAG)
2364 fp->mode = TPA_MODE_GRO;
2366 /* We don't want TPA on an FCoE L2 ring */
2368 fp->disable_tpa = 1;
2371 int bnx2x_load_cnic(struct bnx2x *bp)
2373 int i, rc, port = BP_PORT(bp);
2375 DP(NETIF_MSG_IFUP, "Starting CNIC-related load\n");
2377 mutex_init(&bp->cnic_mutex);
2380 rc = bnx2x_alloc_mem_cnic(bp);
2382 BNX2X_ERR("Unable to allocate bp memory for cnic\n");
2383 LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic0);
2387 rc = bnx2x_alloc_fp_mem_cnic(bp);
2389 BNX2X_ERR("Unable to allocate memory for cnic fps\n");
2390 LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic0);
2393 /* Update the number of queues with the cnic queues */
2394 rc = bnx2x_set_real_num_queues(bp, 1);
2396 BNX2X_ERR("Unable to set real_num_queues including cnic\n");
2397 LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic0);
2400 /* Add all CNIC NAPI objects */
2401 bnx2x_add_all_napi_cnic(bp);
2402 DP(NETIF_MSG_IFUP, "cnic napi added\n");
2403 bnx2x_napi_enable_cnic(bp);
2405 rc = bnx2x_init_hw_func_cnic(bp);
2407 LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic1);
2409 bnx2x_nic_init_cnic(bp);
2412 /* Enable Timer scan */
2413 REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + port*4, 1);
2415 /* setup cnic queues */
2416 for_each_cnic_queue(bp, i) {
2417 rc = bnx2x_setup_queue(bp, &bp->fp[i], 0);
2419 BNX2X_ERR("Queue setup failed\n");
2420 LOAD_ERROR_EXIT(bp, load_error_cnic2);
2425 /* Initialize Rx filter. */
2426 netif_addr_lock_bh(bp->dev);
2427 bnx2x_set_rx_mode(bp->dev);
2428 netif_addr_unlock_bh(bp->dev);
2430 /* re-read iscsi info */
2431 bnx2x_get_iscsi_info(bp);
2432 bnx2x_setup_cnic_irq_info(bp);
2433 bnx2x_setup_cnic_info(bp);
2434 bp->cnic_loaded = true;
2435 if (bp->state == BNX2X_STATE_OPEN)
2436 bnx2x_cnic_notify(bp, CNIC_CTL_START_CMD);
2439 DP(NETIF_MSG_IFUP, "Ending successfully CNIC-related load\n");
2443 #ifndef BNX2X_STOP_ON_ERROR
2445 /* Disable Timer scan */
2446 REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + port*4, 0);
2449 bnx2x_napi_disable_cnic(bp);
2450 /* Update the number of queues without the cnic queues */
2451 rc = bnx2x_set_real_num_queues(bp, 0);
2453 BNX2X_ERR("Unable to set real_num_queues not including cnic\n");
2455 BNX2X_ERR("CNIC-related load failed\n");
2456 bnx2x_free_fp_mem_cnic(bp);
2457 bnx2x_free_mem_cnic(bp);
2459 #endif /* ! BNX2X_STOP_ON_ERROR */
2462 /* must be called with rtnl_lock */
2463 int bnx2x_nic_load(struct bnx2x *bp, int load_mode)
2465 int port = BP_PORT(bp);
2466 int i, rc = 0, load_code = 0;
2468 DP(NETIF_MSG_IFUP, "Starting NIC load\n");
2470 "CNIC is %s\n", CNIC_ENABLED(bp) ? "enabled" : "disabled");
2472 #ifdef BNX2X_STOP_ON_ERROR
2473 if (unlikely(bp->panic)) {
2474 BNX2X_ERR("Can't load NIC when there is panic\n");
2479 bp->state = BNX2X_STATE_OPENING_WAIT4_LOAD;
2481 memset(&bp->last_reported_link, 0, sizeof(bp->last_reported_link));
2482 __set_bit(BNX2X_LINK_REPORT_LINK_DOWN,
2483 &bp->last_reported_link.link_report_flags);
2486 /* must be called before memory allocation and HW init */
2487 bnx2x_ilt_set_info(bp);
2490 * Zero fastpath structures preserving invariants like napi, which are
2491 * allocated only once, fp index, max_cos, bp pointer.
2492 * Also set fp->disable_tpa and txdata_ptr.
2494 DP(NETIF_MSG_IFUP, "num queues: %d", bp->num_queues);
2495 for_each_queue(bp, i)
2497 memset(bp->bnx2x_txq, 0, (BNX2X_MAX_RSS_COUNT(bp) * BNX2X_MULTI_TX_COS +
2498 bp->num_cnic_queues) *
2499 sizeof(struct bnx2x_fp_txdata));
2501 bp->fcoe_init = false;
2503 /* Set the receive queues buffer size */
2504 bnx2x_set_rx_buf_size(bp);
2507 rc = bnx2x_alloc_mem(bp);
2509 BNX2X_ERR("Unable to allocate bp memory\n");
2514 /* Allocated memory for FW statistics */
2515 if (bnx2x_alloc_fw_stats_mem(bp))
2516 LOAD_ERROR_EXIT(bp, load_error0);
2518 /* need to be done after alloc mem, since it's self adjusting to amount
2519 * of memory available for RSS queues
2521 rc = bnx2x_alloc_fp_mem(bp);
2523 BNX2X_ERR("Unable to allocate memory for fps\n");
2524 LOAD_ERROR_EXIT(bp, load_error0);
2527 /* request pf to initialize status blocks */
2529 rc = bnx2x_vfpf_init(bp);
2531 LOAD_ERROR_EXIT(bp, load_error0);
2534 /* As long as bnx2x_alloc_mem() may possibly update
2535 * bp->num_queues, bnx2x_set_real_num_queues() should always
2536 * come after it. At this stage cnic queues are not counted.
2538 rc = bnx2x_set_real_num_queues(bp, 0);
2540 BNX2X_ERR("Unable to set real_num_queues\n");
2541 LOAD_ERROR_EXIT(bp, load_error0);
2544 /* configure multi cos mappings in kernel.
2545 * this configuration may be overriden by a multi class queue discipline
2546 * or by a dcbx negotiation result.
2548 bnx2x_setup_tc(bp->dev, bp->max_cos);
2550 /* Add all NAPI objects */
2551 bnx2x_add_all_napi(bp);
2552 DP(NETIF_MSG_IFUP, "napi added\n");
2553 bnx2x_napi_enable(bp);
2556 /* set pf load just before approaching the MCP */
2557 bnx2x_set_pf_load(bp);
2559 /* if mcp exists send load request and analyze response */
2560 if (!BP_NOMCP(bp)) {
2561 /* attempt to load pf */
2562 rc = bnx2x_nic_load_request(bp, &load_code);
2564 LOAD_ERROR_EXIT(bp, load_error1);
2566 /* what did mcp say? */
2567 rc = bnx2x_nic_load_analyze_req(bp, load_code);
2569 bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2570 LOAD_ERROR_EXIT(bp, load_error2);
2573 load_code = bnx2x_nic_load_no_mcp(bp, port);
2576 /* mark pmf if applicable */
2577 bnx2x_nic_load_pmf(bp, load_code);
2579 /* Init Function state controlling object */
2580 bnx2x__init_func_obj(bp);
2583 rc = bnx2x_init_hw(bp, load_code);
2585 BNX2X_ERR("HW init failed, aborting\n");
2586 bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2587 LOAD_ERROR_EXIT(bp, load_error2);
2591 bnx2x_pre_irq_nic_init(bp);
2593 /* Connect to IRQs */
2594 rc = bnx2x_setup_irqs(bp);
2596 BNX2X_ERR("setup irqs failed\n");
2598 bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2599 LOAD_ERROR_EXIT(bp, load_error2);
2602 /* Init per-function objects */
2604 /* Setup NIC internals and enable interrupts */
2605 bnx2x_post_irq_nic_init(bp, load_code);
2607 bnx2x_init_bp_objs(bp);
2608 bnx2x_iov_nic_init(bp);
2610 /* Set AFEX default VLAN tag to an invalid value */
2611 bp->afex_def_vlan_tag = -1;
2612 bnx2x_nic_load_afex_dcc(bp, load_code);
2613 bp->state = BNX2X_STATE_OPENING_WAIT4_PORT;
2614 rc = bnx2x_func_start(bp);
2616 BNX2X_ERR("Function start failed!\n");
2617 bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2619 LOAD_ERROR_EXIT(bp, load_error3);
2622 /* Send LOAD_DONE command to MCP */
2623 if (!BP_NOMCP(bp)) {
2624 load_code = bnx2x_fw_command(bp,
2625 DRV_MSG_CODE_LOAD_DONE, 0);
2627 BNX2X_ERR("MCP response failure, aborting\n");
2629 LOAD_ERROR_EXIT(bp, load_error3);
2633 /* initialize FW coalescing state machines in RAM */
2634 bnx2x_update_coalesce(bp);
2636 /* setup the leading queue */
2637 rc = bnx2x_setup_leading(bp);
2639 BNX2X_ERR("Setup leading failed!\n");
2640 LOAD_ERROR_EXIT(bp, load_error3);
2643 /* set up the rest of the queues */
2644 for_each_nondefault_eth_queue(bp, i) {
2645 rc = bnx2x_setup_queue(bp, &bp->fp[i], 0);
2647 BNX2X_ERR("Queue setup failed\n");
2648 LOAD_ERROR_EXIT(bp, load_error3);
2653 rc = bnx2x_init_rss_pf(bp);
2655 BNX2X_ERR("PF RSS init failed\n");
2656 LOAD_ERROR_EXIT(bp, load_error3);
2660 for_each_eth_queue(bp, i) {
2661 rc = bnx2x_vfpf_setup_q(bp, i);
2663 BNX2X_ERR("Queue setup failed\n");
2664 LOAD_ERROR_EXIT(bp, load_error3);
2669 /* Now when Clients are configured we are ready to work */
2670 bp->state = BNX2X_STATE_OPEN;
2672 /* Configure a ucast MAC */
2674 rc = bnx2x_set_eth_mac(bp, true);
2676 rc = bnx2x_vfpf_config_mac(bp, bp->dev->dev_addr, bp->fp->index,
2679 BNX2X_ERR("Setting Ethernet MAC failed\n");
2680 LOAD_ERROR_EXIT(bp, load_error3);
2683 if (IS_PF(bp) && bp->pending_max) {
2684 bnx2x_update_max_mf_config(bp, bp->pending_max);
2685 bp->pending_max = 0;
2689 rc = bnx2x_initial_phy_init(bp, load_mode);
2691 LOAD_ERROR_EXIT(bp, load_error3);
2693 bp->link_params.feature_config_flags &= ~FEATURE_CONFIG_BOOT_FROM_SAN;
2695 /* Start fast path */
2697 /* Initialize Rx filter. */
2698 netif_addr_lock_bh(bp->dev);
2699 bnx2x_set_rx_mode(bp->dev);
2700 netif_addr_unlock_bh(bp->dev);
2703 switch (load_mode) {
2705 /* Tx queue should be only reenabled */
2706 netif_tx_wake_all_queues(bp->dev);
2710 netif_tx_start_all_queues(bp->dev);
2711 smp_mb__after_clear_bit();
2715 case LOAD_LOOPBACK_EXT:
2716 bp->state = BNX2X_STATE_DIAG;
2724 bnx2x_update_drv_flags(bp, 1 << DRV_FLAGS_PORT_MASK, 0);
2726 bnx2x__link_status_update(bp);
2728 /* start the timer */
2729 mod_timer(&bp->timer, jiffies + bp->current_interval);
2731 if (CNIC_ENABLED(bp))
2732 bnx2x_load_cnic(bp);
2734 if (IS_PF(bp) && SHMEM2_HAS(bp, drv_capabilities_flag)) {
2735 /* mark driver is loaded in shmem2 */
2737 val = SHMEM2_RD(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)]);
2738 SHMEM2_WR(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)],
2739 val | DRV_FLAGS_CAPABILITIES_LOADED_SUPPORTED |
2740 DRV_FLAGS_CAPABILITIES_LOADED_L2);
2743 /* Wait for all pending SP commands to complete */
2744 if (IS_PF(bp) && !bnx2x_wait_sp_comp(bp, ~0x0UL)) {
2745 BNX2X_ERR("Timeout waiting for SP elements to complete\n");
2746 bnx2x_nic_unload(bp, UNLOAD_CLOSE, false);
2750 /* If PMF - send ADMIN DCBX msg to MFW to initiate DCBX FSM */
2751 if (bp->port.pmf && (bp->state != BNX2X_STATE_DIAG))
2752 bnx2x_dcbx_init(bp, false);
2754 DP(NETIF_MSG_IFUP, "Ending successfully NIC load\n");
2758 #ifndef BNX2X_STOP_ON_ERROR
2761 bnx2x_int_disable_sync(bp, 1);
2763 /* Clean queueable objects */
2764 bnx2x_squeeze_objects(bp);
2767 /* Free SKBs, SGEs, TPA pool and driver internals */
2768 bnx2x_free_skbs(bp);
2769 for_each_rx_queue(bp, i)
2770 bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
2775 if (IS_PF(bp) && !BP_NOMCP(bp)) {
2776 bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_REQ_WOL_MCP, 0);
2777 bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_DONE, 0);
2782 bnx2x_napi_disable(bp);
2783 bnx2x_del_all_napi(bp);
2785 /* clear pf_load status, as it was already set */
2787 bnx2x_clear_pf_load(bp);
2789 bnx2x_free_fp_mem(bp);
2790 bnx2x_free_fw_stats_mem(bp);
2794 #endif /* ! BNX2X_STOP_ON_ERROR */
2797 int bnx2x_drain_tx_queues(struct bnx2x *bp)
2801 /* Wait until tx fastpath tasks complete */
2802 for_each_tx_queue(bp, i) {
2803 struct bnx2x_fastpath *fp = &bp->fp[i];
2805 for_each_cos_in_tx_queue(fp, cos)
2806 rc = bnx2x_clean_tx_queue(bp, fp->txdata_ptr[cos]);
2813 /* must be called with rtnl_lock */
2814 int bnx2x_nic_unload(struct bnx2x *bp, int unload_mode, bool keep_link)
2817 bool global = false;
2819 DP(NETIF_MSG_IFUP, "Starting NIC unload\n");
2821 /* mark driver is unloaded in shmem2 */
2822 if (IS_PF(bp) && SHMEM2_HAS(bp, drv_capabilities_flag)) {
2824 val = SHMEM2_RD(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)]);
2825 SHMEM2_WR(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)],
2826 val & ~DRV_FLAGS_CAPABILITIES_LOADED_L2);
2829 if (IS_PF(bp) && bp->recovery_state != BNX2X_RECOVERY_DONE &&
2830 (bp->state == BNX2X_STATE_CLOSED ||
2831 bp->state == BNX2X_STATE_ERROR)) {
2832 /* We can get here if the driver has been unloaded
2833 * during parity error recovery and is either waiting for a
2834 * leader to complete or for other functions to unload and
2835 * then ifdown has been issued. In this case we want to
2836 * unload and let other functions to complete a recovery
2839 bp->recovery_state = BNX2X_RECOVERY_DONE;
2841 bnx2x_release_leader_lock(bp);
2844 DP(NETIF_MSG_IFDOWN, "Releasing a leadership...\n");
2845 BNX2X_ERR("Can't unload in closed or error state\n");
2849 /* Nothing to do during unload if previous bnx2x_nic_load()
2850 * have not completed succesfully - all resourses are released.
2852 * we can get here only after unsuccessful ndo_* callback, during which
2853 * dev->IFF_UP flag is still on.
2855 if (bp->state == BNX2X_STATE_CLOSED || bp->state == BNX2X_STATE_ERROR)
2858 /* It's important to set the bp->state to the value different from
2859 * BNX2X_STATE_OPEN and only then stop the Tx. Otherwise bnx2x_tx_int()
2860 * may restart the Tx from the NAPI context (see bnx2x_tx_int()).
2862 bp->state = BNX2X_STATE_CLOSING_WAIT4_HALT;
2865 if (CNIC_LOADED(bp))
2866 bnx2x_cnic_notify(bp, CNIC_CTL_STOP_CMD);
2869 bnx2x_tx_disable(bp);
2870 netdev_reset_tc(bp->dev);
2872 bp->rx_mode = BNX2X_RX_MODE_NONE;
2874 del_timer_sync(&bp->timer);
2877 /* Set ALWAYS_ALIVE bit in shmem */
2878 bp->fw_drv_pulse_wr_seq |= DRV_PULSE_ALWAYS_ALIVE;
2879 bnx2x_drv_pulse(bp);
2880 bnx2x_stats_handle(bp, STATS_EVENT_STOP);
2881 bnx2x_save_statistics(bp);
2884 /* wait till consumers catch up with producers in all queues */
2885 bnx2x_drain_tx_queues(bp);
2887 /* if VF indicate to PF this function is going down (PF will delete sp
2888 * elements and clear initializations
2891 bnx2x_vfpf_close_vf(bp);
2892 else if (unload_mode != UNLOAD_RECOVERY)
2893 /* if this is a normal/close unload need to clean up chip*/
2894 bnx2x_chip_cleanup(bp, unload_mode, keep_link);
2896 /* Send the UNLOAD_REQUEST to the MCP */
2897 bnx2x_send_unload_req(bp, unload_mode);
2900 * Prevent transactions to host from the functions on the
2901 * engine that doesn't reset global blocks in case of global
2902 * attention once gloabl blocks are reset and gates are opened
2903 * (the engine which leader will perform the recovery
2906 if (!CHIP_IS_E1x(bp))
2907 bnx2x_pf_disable(bp);
2909 /* Disable HW interrupts, NAPI */
2910 bnx2x_netif_stop(bp, 1);
2911 /* Delete all NAPI objects */
2912 bnx2x_del_all_napi(bp);
2913 if (CNIC_LOADED(bp))
2914 bnx2x_del_all_napi_cnic(bp);
2918 /* Report UNLOAD_DONE to MCP */
2919 bnx2x_send_unload_done(bp, false);
2923 * At this stage no more interrupts will arrive so we may safly clean
2924 * the queueable objects here in case they failed to get cleaned so far.
2927 bnx2x_squeeze_objects(bp);
2929 /* There should be no more pending SP commands at this stage */
2934 /* Free SKBs, SGEs, TPA pool and driver internals */
2935 bnx2x_free_skbs(bp);
2936 if (CNIC_LOADED(bp))
2937 bnx2x_free_skbs_cnic(bp);
2938 for_each_rx_queue(bp, i)
2939 bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
2941 bnx2x_free_fp_mem(bp);
2942 if (CNIC_LOADED(bp))
2943 bnx2x_free_fp_mem_cnic(bp);
2946 if (CNIC_LOADED(bp))
2947 bnx2x_free_mem_cnic(bp);
2950 bp->state = BNX2X_STATE_CLOSED;
2951 bp->cnic_loaded = false;
2953 /* Check if there are pending parity attentions. If there are - set
2954 * RECOVERY_IN_PROGRESS.
2956 if (IS_PF(bp) && bnx2x_chk_parity_attn(bp, &global, false)) {
2957 bnx2x_set_reset_in_progress(bp);
2959 /* Set RESET_IS_GLOBAL if needed */
2961 bnx2x_set_reset_global(bp);
2965 /* The last driver must disable a "close the gate" if there is no
2966 * parity attention or "process kill" pending.
2969 !bnx2x_clear_pf_load(bp) &&
2970 bnx2x_reset_is_done(bp, BP_PATH(bp)))
2971 bnx2x_disable_close_the_gate(bp);
2973 DP(NETIF_MSG_IFUP, "Ending NIC unload\n");
2978 int bnx2x_set_power_state(struct bnx2x *bp, pci_power_t state)
2982 /* If there is no power capability, silently succeed */
2984 BNX2X_DEV_INFO("No power capability. Breaking.\n");
2988 pci_read_config_word(bp->pdev, bp->pm_cap + PCI_PM_CTRL, &pmcsr);
2992 pci_write_config_word(bp->pdev, bp->pm_cap + PCI_PM_CTRL,
2993 ((pmcsr & ~PCI_PM_CTRL_STATE_MASK) |
2994 PCI_PM_CTRL_PME_STATUS));
2996 if (pmcsr & PCI_PM_CTRL_STATE_MASK)
2997 /* delay required during transition out of D3hot */
3002 /* If there are other clients above don't
3003 shut down the power */
3004 if (atomic_read(&bp->pdev->enable_cnt) != 1)
3006 /* Don't shut down the power for emulation and FPGA */
3007 if (CHIP_REV_IS_SLOW(bp))
3010 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
3014 pmcsr |= PCI_PM_CTRL_PME_ENABLE;
3016 pci_write_config_word(bp->pdev, bp->pm_cap + PCI_PM_CTRL,
3019 /* No more memory access after this point until
3020 * device is brought back to D0.
3025 dev_err(&bp->pdev->dev, "Can't support state = %d\n", state);
3032 * net_device service functions
3034 int bnx2x_poll(struct napi_struct *napi, int budget)
3038 struct bnx2x_fastpath *fp = container_of(napi, struct bnx2x_fastpath,
3040 struct bnx2x *bp = fp->bp;
3043 #ifdef BNX2X_STOP_ON_ERROR
3044 if (unlikely(bp->panic)) {
3045 napi_complete(napi);
3050 for_each_cos_in_tx_queue(fp, cos)
3051 if (bnx2x_tx_queue_has_work(fp->txdata_ptr[cos]))
3052 bnx2x_tx_int(bp, fp->txdata_ptr[cos]);
3054 if (bnx2x_has_rx_work(fp)) {
3055 work_done += bnx2x_rx_int(fp, budget - work_done);
3057 /* must not complete if we consumed full budget */
3058 if (work_done >= budget)
3062 /* Fall out from the NAPI loop if needed */
3063 if (!(bnx2x_has_rx_work(fp) || bnx2x_has_tx_work(fp))) {
3065 /* No need to update SB for FCoE L2 ring as long as
3066 * it's connected to the default SB and the SB
3067 * has been updated when NAPI was scheduled.
3069 if (IS_FCOE_FP(fp)) {
3070 napi_complete(napi);
3073 bnx2x_update_fpsb_idx(fp);
3074 /* bnx2x_has_rx_work() reads the status block,
3075 * thus we need to ensure that status block indices
3076 * have been actually read (bnx2x_update_fpsb_idx)
3077 * prior to this check (bnx2x_has_rx_work) so that
3078 * we won't write the "newer" value of the status block
3079 * to IGU (if there was a DMA right after
3080 * bnx2x_has_rx_work and if there is no rmb, the memory
3081 * reading (bnx2x_update_fpsb_idx) may be postponed
3082 * to right before bnx2x_ack_sb). In this case there
3083 * will never be another interrupt until there is
3084 * another update of the status block, while there
3085 * is still unhandled work.
3089 if (!(bnx2x_has_rx_work(fp) || bnx2x_has_tx_work(fp))) {
3090 napi_complete(napi);
3091 /* Re-enable interrupts */
3092 DP(NETIF_MSG_RX_STATUS,
3093 "Update index to %d\n", fp->fp_hc_idx);
3094 bnx2x_ack_sb(bp, fp->igu_sb_id, USTORM_ID,
3095 le16_to_cpu(fp->fp_hc_idx),
3105 /* we split the first BD into headers and data BDs
3106 * to ease the pain of our fellow microcode engineers
3107 * we use one mapping for both BDs
3109 static u16 bnx2x_tx_split(struct bnx2x *bp,
3110 struct bnx2x_fp_txdata *txdata,
3111 struct sw_tx_bd *tx_buf,
3112 struct eth_tx_start_bd **tx_bd, u16 hlen,
3115 struct eth_tx_start_bd *h_tx_bd = *tx_bd;
3116 struct eth_tx_bd *d_tx_bd;
3118 int old_len = le16_to_cpu(h_tx_bd->nbytes);
3120 /* first fix first BD */
3121 h_tx_bd->nbytes = cpu_to_le16(hlen);
3123 DP(NETIF_MSG_TX_QUEUED, "TSO split header size is %d (%x:%x)\n",
3124 h_tx_bd->nbytes, h_tx_bd->addr_hi, h_tx_bd->addr_lo);
3126 /* now get a new data BD
3127 * (after the pbd) and fill it */
3128 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3129 d_tx_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
3131 mapping = HILO_U64(le32_to_cpu(h_tx_bd->addr_hi),
3132 le32_to_cpu(h_tx_bd->addr_lo)) + hlen;
3134 d_tx_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
3135 d_tx_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
3136 d_tx_bd->nbytes = cpu_to_le16(old_len - hlen);
3138 /* this marks the BD as one that has no individual mapping */
3139 tx_buf->flags |= BNX2X_TSO_SPLIT_BD;
3141 DP(NETIF_MSG_TX_QUEUED,
3142 "TSO split data size is %d (%x:%x)\n",
3143 d_tx_bd->nbytes, d_tx_bd->addr_hi, d_tx_bd->addr_lo);
3146 *tx_bd = (struct eth_tx_start_bd *)d_tx_bd;
3151 #define bswab32(b32) ((__force __le32) swab32((__force __u32) (b32)))
3152 #define bswab16(b16) ((__force __le16) swab16((__force __u16) (b16)))
3153 static __le16 bnx2x_csum_fix(unsigned char *t_header, u16 csum, s8 fix)
3155 __sum16 tsum = (__force __sum16) csum;
3158 tsum = ~csum_fold(csum_sub((__force __wsum) csum,
3159 csum_partial(t_header - fix, fix, 0)));
3162 tsum = ~csum_fold(csum_add((__force __wsum) csum,
3163 csum_partial(t_header, -fix, 0)));
3165 return bswab16(tsum);
3168 static u32 bnx2x_xmit_type(struct bnx2x *bp, struct sk_buff *skb)
3174 if (skb->ip_summed != CHECKSUM_PARTIAL)
3177 protocol = vlan_get_protocol(skb);
3178 if (protocol == htons(ETH_P_IPV6)) {
3180 prot = ipv6_hdr(skb)->nexthdr;
3183 prot = ip_hdr(skb)->protocol;
3186 if (!CHIP_IS_E1x(bp) && skb->encapsulation) {
3187 if (inner_ip_hdr(skb)->version == 6) {
3188 rc |= XMIT_CSUM_ENC_V6;
3189 if (inner_ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3190 rc |= XMIT_CSUM_TCP;
3192 rc |= XMIT_CSUM_ENC_V4;
3193 if (inner_ip_hdr(skb)->protocol == IPPROTO_TCP)
3194 rc |= XMIT_CSUM_TCP;
3197 if (prot == IPPROTO_TCP)
3198 rc |= XMIT_CSUM_TCP;
3200 if (skb_is_gso_v6(skb)) {
3201 rc |= (XMIT_GSO_V6 | XMIT_CSUM_TCP | XMIT_CSUM_V6);
3202 if (rc & XMIT_CSUM_ENC)
3203 rc |= XMIT_GSO_ENC_V6;
3204 } else if (skb_is_gso(skb)) {
3205 rc |= (XMIT_GSO_V4 | XMIT_CSUM_V4 | XMIT_CSUM_TCP);
3206 if (rc & XMIT_CSUM_ENC)
3207 rc |= XMIT_GSO_ENC_V4;
3213 #if (MAX_SKB_FRAGS >= MAX_FETCH_BD - 3)
3214 /* check if packet requires linearization (packet is too fragmented)
3215 no need to check fragmentation if page size > 8K (there will be no
3216 violation to FW restrictions) */
3217 static int bnx2x_pkt_req_lin(struct bnx2x *bp, struct sk_buff *skb,
3222 int first_bd_sz = 0;
3224 /* 3 = 1 (for linear data BD) + 2 (for PBD and last BD) */
3225 if (skb_shinfo(skb)->nr_frags >= (MAX_FETCH_BD - 3)) {
3227 if (xmit_type & XMIT_GSO) {
3228 unsigned short lso_mss = skb_shinfo(skb)->gso_size;
3229 /* Check if LSO packet needs to be copied:
3230 3 = 1 (for headers BD) + 2 (for PBD and last BD) */
3231 int wnd_size = MAX_FETCH_BD - 3;
3232 /* Number of windows to check */
3233 int num_wnds = skb_shinfo(skb)->nr_frags - wnd_size;
3238 /* Headers length */
3239 hlen = (int)(skb_transport_header(skb) - skb->data) +
3242 /* Amount of data (w/o headers) on linear part of SKB*/
3243 first_bd_sz = skb_headlen(skb) - hlen;
3245 wnd_sum = first_bd_sz;
3247 /* Calculate the first sum - it's special */
3248 for (frag_idx = 0; frag_idx < wnd_size - 1; frag_idx++)
3250 skb_frag_size(&skb_shinfo(skb)->frags[frag_idx]);
3252 /* If there was data on linear skb data - check it */
3253 if (first_bd_sz > 0) {
3254 if (unlikely(wnd_sum < lso_mss)) {
3259 wnd_sum -= first_bd_sz;
3262 /* Others are easier: run through the frag list and
3263 check all windows */
3264 for (wnd_idx = 0; wnd_idx <= num_wnds; wnd_idx++) {
3266 skb_frag_size(&skb_shinfo(skb)->frags[wnd_idx + wnd_size - 1]);
3268 if (unlikely(wnd_sum < lso_mss)) {
3273 skb_frag_size(&skb_shinfo(skb)->frags[wnd_idx]);
3276 /* in non-LSO too fragmented packet should always
3283 if (unlikely(to_copy))
3284 DP(NETIF_MSG_TX_QUEUED,
3285 "Linearization IS REQUIRED for %s packet. num_frags %d hlen %d first_bd_sz %d\n",
3286 (xmit_type & XMIT_GSO) ? "LSO" : "non-LSO",
3287 skb_shinfo(skb)->nr_frags, hlen, first_bd_sz);
3293 static void bnx2x_set_pbd_gso_e2(struct sk_buff *skb, u32 *parsing_data,
3296 struct ipv6hdr *ipv6;
3298 *parsing_data |= (skb_shinfo(skb)->gso_size <<
3299 ETH_TX_PARSE_BD_E2_LSO_MSS_SHIFT) &
3300 ETH_TX_PARSE_BD_E2_LSO_MSS;
3302 if (xmit_type & XMIT_GSO_ENC_V6)
3303 ipv6 = inner_ipv6_hdr(skb);
3304 else if (xmit_type & XMIT_GSO_V6)
3305 ipv6 = ipv6_hdr(skb);
3309 if (ipv6 && ipv6->nexthdr == NEXTHDR_IPV6)
3310 *parsing_data |= ETH_TX_PARSE_BD_E2_IPV6_WITH_EXT_HDR;
3314 * bnx2x_set_pbd_gso - update PBD in GSO case.
3318 * @xmit_type: xmit flags
3320 static void bnx2x_set_pbd_gso(struct sk_buff *skb,
3321 struct eth_tx_parse_bd_e1x *pbd,
3322 struct eth_tx_start_bd *tx_start_bd,
3325 pbd->lso_mss = cpu_to_le16(skb_shinfo(skb)->gso_size);
3326 pbd->tcp_send_seq = bswab32(tcp_hdr(skb)->seq);
3327 pbd->tcp_flags = pbd_tcp_flags(tcp_hdr(skb));
3329 if (xmit_type & XMIT_GSO_V4) {
3330 pbd->ip_id = bswab16(ip_hdr(skb)->id);
3331 pbd->tcp_pseudo_csum =
3332 bswab16(~csum_tcpudp_magic(ip_hdr(skb)->saddr,
3334 0, IPPROTO_TCP, 0));
3336 /* GSO on 57710/57711 needs FW to calculate IP checksum */
3337 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_IP_CSUM;
3339 pbd->tcp_pseudo_csum =
3340 bswab16(~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3341 &ipv6_hdr(skb)->daddr,
3342 0, IPPROTO_TCP, 0));
3346 cpu_to_le16(ETH_TX_PARSE_BD_E1X_PSEUDO_CS_WITHOUT_LEN);
3350 * bnx2x_set_pbd_csum_enc - update PBD with checksum and return header length
3352 * @bp: driver handle
3354 * @parsing_data: data to be updated
3355 * @xmit_type: xmit flags
3357 * 57712/578xx related, when skb has encapsulation
3359 static u8 bnx2x_set_pbd_csum_enc(struct bnx2x *bp, struct sk_buff *skb,
3360 u32 *parsing_data, u32 xmit_type)
3363 ((((u8 *)skb_inner_transport_header(skb) - skb->data) >> 1) <<
3364 ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W_SHIFT) &
3365 ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W;
3367 if (xmit_type & XMIT_CSUM_TCP) {
3368 *parsing_data |= ((inner_tcp_hdrlen(skb) / 4) <<
3369 ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW_SHIFT) &
3370 ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW;
3372 return skb_inner_transport_header(skb) +
3373 inner_tcp_hdrlen(skb) - skb->data;
3376 /* We support checksum offload for TCP and UDP only.
3377 * No need to pass the UDP header length - it's a constant.
3379 return skb_inner_transport_header(skb) +
3380 sizeof(struct udphdr) - skb->data;
3384 * bnx2x_set_pbd_csum_e2 - update PBD with checksum and return header length
3386 * @bp: driver handle
3388 * @parsing_data: data to be updated
3389 * @xmit_type: xmit flags
3391 * 57712/578xx related
3393 static u8 bnx2x_set_pbd_csum_e2(struct bnx2x *bp, struct sk_buff *skb,
3394 u32 *parsing_data, u32 xmit_type)
3397 ((((u8 *)skb_transport_header(skb) - skb->data) >> 1) <<
3398 ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W_SHIFT) &
3399 ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W;
3401 if (xmit_type & XMIT_CSUM_TCP) {
3402 *parsing_data |= ((tcp_hdrlen(skb) / 4) <<
3403 ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW_SHIFT) &
3404 ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW;
3406 return skb_transport_header(skb) + tcp_hdrlen(skb) - skb->data;
3408 /* We support checksum offload for TCP and UDP only.
3409 * No need to pass the UDP header length - it's a constant.
3411 return skb_transport_header(skb) + sizeof(struct udphdr) - skb->data;
3414 /* set FW indication according to inner or outer protocols if tunneled */
3415 static void bnx2x_set_sbd_csum(struct bnx2x *bp, struct sk_buff *skb,
3416 struct eth_tx_start_bd *tx_start_bd,
3419 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_L4_CSUM;
3421 if (xmit_type & (XMIT_CSUM_ENC_V6 | XMIT_CSUM_V6))
3422 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_IPV6;
3424 if (!(xmit_type & XMIT_CSUM_TCP))
3425 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_IS_UDP;
3429 * bnx2x_set_pbd_csum - update PBD with checksum and return header length
3431 * @bp: driver handle
3433 * @pbd: parse BD to be updated
3434 * @xmit_type: xmit flags
3436 static u8 bnx2x_set_pbd_csum(struct bnx2x *bp, struct sk_buff *skb,
3437 struct eth_tx_parse_bd_e1x *pbd,
3440 u8 hlen = (skb_network_header(skb) - skb->data) >> 1;
3442 /* for now NS flag is not used in Linux */
3445 ((skb->protocol == cpu_to_be16(ETH_P_8021Q)) <<
3446 ETH_TX_PARSE_BD_E1X_LLC_SNAP_EN_SHIFT));
3448 pbd->ip_hlen_w = (skb_transport_header(skb) -
3449 skb_network_header(skb)) >> 1;
3451 hlen += pbd->ip_hlen_w;
3453 /* We support checksum offload for TCP and UDP only */
3454 if (xmit_type & XMIT_CSUM_TCP)
3455 hlen += tcp_hdrlen(skb) / 2;
3457 hlen += sizeof(struct udphdr) / 2;
3459 pbd->total_hlen_w = cpu_to_le16(hlen);
3462 if (xmit_type & XMIT_CSUM_TCP) {
3463 pbd->tcp_pseudo_csum = bswab16(tcp_hdr(skb)->check);
3466 s8 fix = SKB_CS_OFF(skb); /* signed! */
3468 DP(NETIF_MSG_TX_QUEUED,
3469 "hlen %d fix %d csum before fix %x\n",
3470 le16_to_cpu(pbd->total_hlen_w), fix, SKB_CS(skb));
3472 /* HW bug: fixup the CSUM */
3473 pbd->tcp_pseudo_csum =
3474 bnx2x_csum_fix(skb_transport_header(skb),
3477 DP(NETIF_MSG_TX_QUEUED, "csum after fix %x\n",
3478 pbd->tcp_pseudo_csum);
3484 static void bnx2x_update_pbds_gso_enc(struct sk_buff *skb,
3485 struct eth_tx_parse_bd_e2 *pbd_e2,
3486 struct eth_tx_parse_2nd_bd *pbd2,
3491 u8 outerip_off, outerip_len = 0;
3492 /* from outer IP to transport */
3493 hlen_w = (skb_inner_transport_header(skb) -
3494 skb_network_header(skb)) >> 1;
3497 if (xmit_type & XMIT_CSUM_TCP)
3498 hlen_w += inner_tcp_hdrlen(skb) >> 1;
3500 hlen_w += sizeof(struct udphdr) >> 1;
3502 pbd2->fw_ip_hdr_to_payload_w = hlen_w;
3504 if (xmit_type & XMIT_CSUM_ENC_V4) {
3505 struct iphdr *iph = ip_hdr(skb);
3506 pbd2->fw_ip_csum_wo_len_flags_frag =
3507 bswab16(csum_fold((~iph->check) -
3508 iph->tot_len - iph->frag_off));
3510 pbd2->fw_ip_hdr_to_payload_w =
3511 hlen_w - ((sizeof(struct ipv6hdr)) >> 1);
3514 pbd2->tcp_send_seq = bswab32(inner_tcp_hdr(skb)->seq);
3516 pbd2->tcp_flags = pbd_tcp_flags(inner_tcp_hdr(skb));
3518 if (xmit_type & XMIT_GSO_V4) {
3519 pbd2->hw_ip_id = bswab16(inner_ip_hdr(skb)->id);
3521 pbd_e2->data.tunnel_data.pseudo_csum =
3522 bswab16(~csum_tcpudp_magic(
3523 inner_ip_hdr(skb)->saddr,
3524 inner_ip_hdr(skb)->daddr,
3525 0, IPPROTO_TCP, 0));
3527 outerip_len = ip_hdr(skb)->ihl << 1;
3529 pbd_e2->data.tunnel_data.pseudo_csum =
3530 bswab16(~csum_ipv6_magic(
3531 &inner_ipv6_hdr(skb)->saddr,
3532 &inner_ipv6_hdr(skb)->daddr,
3533 0, IPPROTO_TCP, 0));
3536 outerip_off = (skb_network_header(skb) - skb->data) >> 1;
3540 (!!(xmit_type & XMIT_CSUM_V6) <<
3541 ETH_TX_PARSE_2ND_BD_IP_HDR_TYPE_OUTER_SHIFT) |
3543 ETH_TX_PARSE_2ND_BD_IP_HDR_LEN_OUTER_W_SHIFT) |
3544 ((skb->protocol == cpu_to_be16(ETH_P_8021Q)) <<
3545 ETH_TX_PARSE_2ND_BD_LLC_SNAP_EN_SHIFT);
3547 if (ip_hdr(skb)->protocol == IPPROTO_UDP) {
3548 SET_FLAG(*global_data, ETH_TX_PARSE_2ND_BD_TUNNEL_UDP_EXIST, 1);
3549 pbd2->tunnel_udp_hdr_start_w = skb_transport_offset(skb) >> 1;
3553 /* called with netif_tx_lock
3554 * bnx2x_tx_int() runs without netif_tx_lock unless it needs to call
3555 * netif_wake_queue()
3557 netdev_tx_t bnx2x_start_xmit(struct sk_buff *skb, struct net_device *dev)
3559 struct bnx2x *bp = netdev_priv(dev);
3561 struct netdev_queue *txq;
3562 struct bnx2x_fp_txdata *txdata;
3563 struct sw_tx_bd *tx_buf;
3564 struct eth_tx_start_bd *tx_start_bd, *first_bd;
3565 struct eth_tx_bd *tx_data_bd, *total_pkt_bd = NULL;
3566 struct eth_tx_parse_bd_e1x *pbd_e1x = NULL;
3567 struct eth_tx_parse_bd_e2 *pbd_e2 = NULL;
3568 struct eth_tx_parse_2nd_bd *pbd2 = NULL;
3569 u32 pbd_e2_parsing_data = 0;
3570 u16 pkt_prod, bd_prod;
3573 u32 xmit_type = bnx2x_xmit_type(bp, skb);
3576 __le16 pkt_size = 0;
3578 u8 mac_type = UNICAST_ADDRESS;
3580 #ifdef BNX2X_STOP_ON_ERROR
3581 if (unlikely(bp->panic))
3582 return NETDEV_TX_BUSY;
3585 txq_index = skb_get_queue_mapping(skb);
3586 txq = netdev_get_tx_queue(dev, txq_index);
3588 BUG_ON(txq_index >= MAX_ETH_TXQ_IDX(bp) + (CNIC_LOADED(bp) ? 1 : 0));
3590 txdata = &bp->bnx2x_txq[txq_index];
3592 /* enable this debug print to view the transmission queue being used
3593 DP(NETIF_MSG_TX_QUEUED, "indices: txq %d, fp %d, txdata %d\n",
3594 txq_index, fp_index, txdata_index); */
3596 /* enable this debug print to view the tranmission details
3597 DP(NETIF_MSG_TX_QUEUED,
3598 "transmitting packet cid %d fp index %d txdata_index %d tx_data ptr %p fp pointer %p\n",
3599 txdata->cid, fp_index, txdata_index, txdata, fp); */
3601 if (unlikely(bnx2x_tx_avail(bp, txdata) <
3602 skb_shinfo(skb)->nr_frags +
3604 NEXT_CNT_PER_TX_PKT(MAX_BDS_PER_TX_PKT))) {
3605 /* Handle special storage cases separately */
3606 if (txdata->tx_ring_size == 0) {
3607 struct bnx2x_eth_q_stats *q_stats =
3608 bnx2x_fp_qstats(bp, txdata->parent_fp);
3609 q_stats->driver_filtered_tx_pkt++;
3611 return NETDEV_TX_OK;
3613 bnx2x_fp_qstats(bp, txdata->parent_fp)->driver_xoff++;
3614 netif_tx_stop_queue(txq);
3615 BNX2X_ERR("BUG! Tx ring full when queue awake!\n");
3617 return NETDEV_TX_BUSY;
3620 DP(NETIF_MSG_TX_QUEUED,
3621 "queue[%d]: SKB: summed %x protocol %x protocol(%x,%x) gso type %x xmit_type %x len %d\n",
3622 txq_index, skb->ip_summed, skb->protocol, ipv6_hdr(skb)->nexthdr,
3623 ip_hdr(skb)->protocol, skb_shinfo(skb)->gso_type, xmit_type,
3626 eth = (struct ethhdr *)skb->data;
3628 /* set flag according to packet type (UNICAST_ADDRESS is default)*/
3629 if (unlikely(is_multicast_ether_addr(eth->h_dest))) {
3630 if (is_broadcast_ether_addr(eth->h_dest))
3631 mac_type = BROADCAST_ADDRESS;
3633 mac_type = MULTICAST_ADDRESS;
3636 #if (MAX_SKB_FRAGS >= MAX_FETCH_BD - BDS_PER_TX_PKT)
3637 /* First, check if we need to linearize the skb (due to FW
3638 restrictions). No need to check fragmentation if page size > 8K
3639 (there will be no violation to FW restrictions) */
3640 if (bnx2x_pkt_req_lin(bp, skb, xmit_type)) {
3641 /* Statistics of linearization */
3643 if (skb_linearize(skb) != 0) {
3644 DP(NETIF_MSG_TX_QUEUED,
3645 "SKB linearization failed - silently dropping this SKB\n");
3646 dev_kfree_skb_any(skb);
3647 return NETDEV_TX_OK;
3651 /* Map skb linear data for DMA */
3652 mapping = dma_map_single(&bp->pdev->dev, skb->data,
3653 skb_headlen(skb), DMA_TO_DEVICE);
3654 if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
3655 DP(NETIF_MSG_TX_QUEUED,
3656 "SKB mapping failed - silently dropping this SKB\n");
3657 dev_kfree_skb_any(skb);
3658 return NETDEV_TX_OK;
3661 Please read carefully. First we use one BD which we mark as start,
3662 then we have a parsing info BD (used for TSO or xsum),
3663 and only then we have the rest of the TSO BDs.
3664 (don't forget to mark the last one as last,
3665 and to unmap only AFTER you write to the BD ...)
3666 And above all, all pdb sizes are in words - NOT DWORDS!
3669 /* get current pkt produced now - advance it just before sending packet
3670 * since mapping of pages may fail and cause packet to be dropped
3672 pkt_prod = txdata->tx_pkt_prod;
3673 bd_prod = TX_BD(txdata->tx_bd_prod);
3675 /* get a tx_buf and first BD
3676 * tx_start_bd may be changed during SPLIT,
3677 * but first_bd will always stay first
3679 tx_buf = &txdata->tx_buf_ring[TX_BD(pkt_prod)];
3680 tx_start_bd = &txdata->tx_desc_ring[bd_prod].start_bd;
3681 first_bd = tx_start_bd;
3683 tx_start_bd->bd_flags.as_bitfield = ETH_TX_BD_FLAGS_START_BD;
3685 /* header nbd: indirectly zero other flags! */
3686 tx_start_bd->general_data = 1 << ETH_TX_START_BD_HDR_NBDS_SHIFT;
3688 /* remember the first BD of the packet */
3689 tx_buf->first_bd = txdata->tx_bd_prod;
3693 DP(NETIF_MSG_TX_QUEUED,
3694 "sending pkt %u @%p next_idx %u bd %u @%p\n",
3695 pkt_prod, tx_buf, txdata->tx_pkt_prod, bd_prod, tx_start_bd);
3697 if (vlan_tx_tag_present(skb)) {
3698 tx_start_bd->vlan_or_ethertype =
3699 cpu_to_le16(vlan_tx_tag_get(skb));
3700 tx_start_bd->bd_flags.as_bitfield |=
3701 (X_ETH_OUTBAND_VLAN << ETH_TX_BD_FLAGS_VLAN_MODE_SHIFT);
3703 /* when transmitting in a vf, start bd must hold the ethertype
3704 * for fw to enforce it
3707 tx_start_bd->vlan_or_ethertype =
3708 cpu_to_le16(ntohs(eth->h_proto));
3710 /* used by FW for packet accounting */
3711 tx_start_bd->vlan_or_ethertype = cpu_to_le16(pkt_prod);
3714 nbd = 2; /* start_bd + pbd + frags (updated when pages are mapped) */
3716 /* turn on parsing and get a BD */
3717 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3719 if (xmit_type & XMIT_CSUM)
3720 bnx2x_set_sbd_csum(bp, skb, tx_start_bd, xmit_type);
3722 if (!CHIP_IS_E1x(bp)) {
3723 pbd_e2 = &txdata->tx_desc_ring[bd_prod].parse_bd_e2;
3724 memset(pbd_e2, 0, sizeof(struct eth_tx_parse_bd_e2));
3726 if (xmit_type & XMIT_CSUM_ENC) {
3727 u16 global_data = 0;
3729 /* Set PBD in enc checksum offload case */
3730 hlen = bnx2x_set_pbd_csum_enc(bp, skb,
3731 &pbd_e2_parsing_data,
3734 /* turn on 2nd parsing and get a BD */
3735 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3737 pbd2 = &txdata->tx_desc_ring[bd_prod].parse_2nd_bd;
3739 memset(pbd2, 0, sizeof(*pbd2));
3741 pbd_e2->data.tunnel_data.ip_hdr_start_inner_w =
3742 (skb_inner_network_header(skb) -
3745 if (xmit_type & XMIT_GSO_ENC)
3746 bnx2x_update_pbds_gso_enc(skb, pbd_e2, pbd2,
3750 pbd2->global_data = cpu_to_le16(global_data);
3752 /* add addition parse BD indication to start BD */
3753 SET_FLAG(tx_start_bd->general_data,
3754 ETH_TX_START_BD_PARSE_NBDS, 1);
3755 /* set encapsulation flag in start BD */
3756 SET_FLAG(tx_start_bd->general_data,
3757 ETH_TX_START_BD_TUNNEL_EXIST, 1);
3759 } else if (xmit_type & XMIT_CSUM) {
3760 /* Set PBD in checksum offload case w/o encapsulation */
3761 hlen = bnx2x_set_pbd_csum_e2(bp, skb,
3762 &pbd_e2_parsing_data,
3766 /* Add the macs to the parsing BD this is a vf */
3768 /* override GRE parameters in BD */
3769 bnx2x_set_fw_mac_addr(&pbd_e2->data.mac_addr.src_hi,
3770 &pbd_e2->data.mac_addr.src_mid,
3771 &pbd_e2->data.mac_addr.src_lo,
3774 bnx2x_set_fw_mac_addr(&pbd_e2->data.mac_addr.dst_hi,
3775 &pbd_e2->data.mac_addr.dst_mid,
3776 &pbd_e2->data.mac_addr.dst_lo,
3780 SET_FLAG(pbd_e2_parsing_data,
3781 ETH_TX_PARSE_BD_E2_ETH_ADDR_TYPE, mac_type);
3783 u16 global_data = 0;
3784 pbd_e1x = &txdata->tx_desc_ring[bd_prod].parse_bd_e1x;
3785 memset(pbd_e1x, 0, sizeof(struct eth_tx_parse_bd_e1x));
3786 /* Set PBD in checksum offload case */
3787 if (xmit_type & XMIT_CSUM)
3788 hlen = bnx2x_set_pbd_csum(bp, skb, pbd_e1x, xmit_type);
3790 SET_FLAG(global_data,
3791 ETH_TX_PARSE_BD_E1X_ETH_ADDR_TYPE, mac_type);
3792 pbd_e1x->global_data |= cpu_to_le16(global_data);
3795 /* Setup the data pointer of the first BD of the packet */
3796 tx_start_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
3797 tx_start_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
3798 tx_start_bd->nbytes = cpu_to_le16(skb_headlen(skb));
3799 pkt_size = tx_start_bd->nbytes;
3801 DP(NETIF_MSG_TX_QUEUED,
3802 "first bd @%p addr (%x:%x) nbytes %d flags %x vlan %x\n",
3803 tx_start_bd, tx_start_bd->addr_hi, tx_start_bd->addr_lo,
3804 le16_to_cpu(tx_start_bd->nbytes),
3805 tx_start_bd->bd_flags.as_bitfield,
3806 le16_to_cpu(tx_start_bd->vlan_or_ethertype));
3808 if (xmit_type & XMIT_GSO) {
3810 DP(NETIF_MSG_TX_QUEUED,
3811 "TSO packet len %d hlen %d total len %d tso size %d\n",
3812 skb->len, hlen, skb_headlen(skb),
3813 skb_shinfo(skb)->gso_size);
3815 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_SW_LSO;
3817 if (unlikely(skb_headlen(skb) > hlen)) {
3819 bd_prod = bnx2x_tx_split(bp, txdata, tx_buf,
3823 if (!CHIP_IS_E1x(bp))
3824 bnx2x_set_pbd_gso_e2(skb, &pbd_e2_parsing_data,
3827 bnx2x_set_pbd_gso(skb, pbd_e1x, tx_start_bd,
3831 /* Set the PBD's parsing_data field if not zero
3832 * (for the chips newer than 57711).
3834 if (pbd_e2_parsing_data)
3835 pbd_e2->parsing_data = cpu_to_le32(pbd_e2_parsing_data);
3837 tx_data_bd = (struct eth_tx_bd *)tx_start_bd;
3839 /* Handle fragmented skb */
3840 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3841 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3843 mapping = skb_frag_dma_map(&bp->pdev->dev, frag, 0,
3844 skb_frag_size(frag), DMA_TO_DEVICE);
3845 if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
3846 unsigned int pkts_compl = 0, bytes_compl = 0;
3848 DP(NETIF_MSG_TX_QUEUED,
3849 "Unable to map page - dropping packet...\n");
3851 /* we need unmap all buffers already mapped
3853 * first_bd->nbd need to be properly updated
3854 * before call to bnx2x_free_tx_pkt
3856 first_bd->nbd = cpu_to_le16(nbd);
3857 bnx2x_free_tx_pkt(bp, txdata,
3858 TX_BD(txdata->tx_pkt_prod),
3859 &pkts_compl, &bytes_compl);
3860 return NETDEV_TX_OK;
3863 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3864 tx_data_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
3865 if (total_pkt_bd == NULL)
3866 total_pkt_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
3868 tx_data_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
3869 tx_data_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
3870 tx_data_bd->nbytes = cpu_to_le16(skb_frag_size(frag));
3871 le16_add_cpu(&pkt_size, skb_frag_size(frag));
3874 DP(NETIF_MSG_TX_QUEUED,
3875 "frag %d bd @%p addr (%x:%x) nbytes %d\n",
3876 i, tx_data_bd, tx_data_bd->addr_hi, tx_data_bd->addr_lo,
3877 le16_to_cpu(tx_data_bd->nbytes));
3880 DP(NETIF_MSG_TX_QUEUED, "last bd @%p\n", tx_data_bd);
3882 /* update with actual num BDs */
3883 first_bd->nbd = cpu_to_le16(nbd);
3885 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3887 /* now send a tx doorbell, counting the next BD
3888 * if the packet contains or ends with it
3890 if (TX_BD_POFF(bd_prod) < nbd)
3893 /* total_pkt_bytes should be set on the first data BD if
3894 * it's not an LSO packet and there is more than one
3895 * data BD. In this case pkt_size is limited by an MTU value.
3896 * However we prefer to set it for an LSO packet (while we don't
3897 * have to) in order to save some CPU cycles in a none-LSO
3898 * case, when we much more care about them.
3900 if (total_pkt_bd != NULL)
3901 total_pkt_bd->total_pkt_bytes = pkt_size;
3904 DP(NETIF_MSG_TX_QUEUED,
3905 "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",
3906 pbd_e1x, pbd_e1x->global_data, pbd_e1x->ip_hlen_w,
3907 pbd_e1x->ip_id, pbd_e1x->lso_mss, pbd_e1x->tcp_flags,
3908 pbd_e1x->tcp_pseudo_csum, pbd_e1x->tcp_send_seq,
3909 le16_to_cpu(pbd_e1x->total_hlen_w));
3911 DP(NETIF_MSG_TX_QUEUED,
3912 "PBD (E2) @%p dst %x %x %x src %x %x %x parsing_data %x\n",
3914 pbd_e2->data.mac_addr.dst_hi,
3915 pbd_e2->data.mac_addr.dst_mid,
3916 pbd_e2->data.mac_addr.dst_lo,
3917 pbd_e2->data.mac_addr.src_hi,
3918 pbd_e2->data.mac_addr.src_mid,
3919 pbd_e2->data.mac_addr.src_lo,
3920 pbd_e2->parsing_data);
3921 DP(NETIF_MSG_TX_QUEUED, "doorbell: nbd %d bd %u\n", nbd, bd_prod);
3923 netdev_tx_sent_queue(txq, skb->len);
3925 skb_tx_timestamp(skb);
3927 txdata->tx_pkt_prod++;
3929 * Make sure that the BD data is updated before updating the producer
3930 * since FW might read the BD right after the producer is updated.
3931 * This is only applicable for weak-ordered memory model archs such
3932 * as IA-64. The following barrier is also mandatory since FW will
3933 * assumes packets must have BDs.
3937 txdata->tx_db.data.prod += nbd;
3940 DOORBELL(bp, txdata->cid, txdata->tx_db.raw);
3944 txdata->tx_bd_prod += nbd;
3946 if (unlikely(bnx2x_tx_avail(bp, txdata) < MAX_DESC_PER_TX_PKT)) {
3947 netif_tx_stop_queue(txq);
3949 /* paired memory barrier is in bnx2x_tx_int(), we have to keep
3950 * ordering of set_bit() in netif_tx_stop_queue() and read of
3954 bnx2x_fp_qstats(bp, txdata->parent_fp)->driver_xoff++;
3955 if (bnx2x_tx_avail(bp, txdata) >= MAX_DESC_PER_TX_PKT)
3956 netif_tx_wake_queue(txq);
3960 return NETDEV_TX_OK;
3964 * bnx2x_setup_tc - routine to configure net_device for multi tc
3966 * @netdev: net device to configure
3967 * @tc: number of traffic classes to enable
3969 * callback connected to the ndo_setup_tc function pointer
3971 int bnx2x_setup_tc(struct net_device *dev, u8 num_tc)
3973 int cos, prio, count, offset;
3974 struct bnx2x *bp = netdev_priv(dev);
3976 /* setup tc must be called under rtnl lock */
3979 /* no traffic classes requested. aborting */
3981 netdev_reset_tc(dev);
3985 /* requested to support too many traffic classes */
3986 if (num_tc > bp->max_cos) {
3987 BNX2X_ERR("support for too many traffic classes requested: %d. max supported is %d\n",
3988 num_tc, bp->max_cos);
3992 /* declare amount of supported traffic classes */
3993 if (netdev_set_num_tc(dev, num_tc)) {
3994 BNX2X_ERR("failed to declare %d traffic classes\n", num_tc);
3998 /* configure priority to traffic class mapping */
3999 for (prio = 0; prio < BNX2X_MAX_PRIORITY; prio++) {
4000 netdev_set_prio_tc_map(dev, prio, bp->prio_to_cos[prio]);
4001 DP(BNX2X_MSG_SP | NETIF_MSG_IFUP,
4002 "mapping priority %d to tc %d\n",
4003 prio, bp->prio_to_cos[prio]);
4007 /* Use this configuration to diffrentiate tc0 from other COSes
4008 This can be used for ets or pfc, and save the effort of setting
4009 up a multio class queue disc or negotiating DCBX with a switch
4010 netdev_set_prio_tc_map(dev, 0, 0);
4011 DP(BNX2X_MSG_SP, "mapping priority %d to tc %d\n", 0, 0);
4012 for (prio = 1; prio < 16; prio++) {
4013 netdev_set_prio_tc_map(dev, prio, 1);
4014 DP(BNX2X_MSG_SP, "mapping priority %d to tc %d\n", prio, 1);
4017 /* configure traffic class to transmission queue mapping */
4018 for (cos = 0; cos < bp->max_cos; cos++) {
4019 count = BNX2X_NUM_ETH_QUEUES(bp);
4020 offset = cos * BNX2X_NUM_NON_CNIC_QUEUES(bp);
4021 netdev_set_tc_queue(dev, cos, count, offset);
4022 DP(BNX2X_MSG_SP | NETIF_MSG_IFUP,
4023 "mapping tc %d to offset %d count %d\n",
4024 cos, offset, count);
4030 /* called with rtnl_lock */
4031 int bnx2x_change_mac_addr(struct net_device *dev, void *p)
4033 struct sockaddr *addr = p;
4034 struct bnx2x *bp = netdev_priv(dev);
4037 if (!bnx2x_is_valid_ether_addr(bp, addr->sa_data)) {
4038 BNX2X_ERR("Requested MAC address is not valid\n");
4042 if ((IS_MF_STORAGE_SD(bp) || IS_MF_FCOE_AFEX(bp)) &&
4043 !is_zero_ether_addr(addr->sa_data)) {
4044 BNX2X_ERR("Can't configure non-zero address on iSCSI or FCoE functions in MF-SD mode\n");
4048 if (netif_running(dev)) {
4049 rc = bnx2x_set_eth_mac(bp, false);
4054 memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
4056 if (netif_running(dev))
4057 rc = bnx2x_set_eth_mac(bp, true);
4062 static void bnx2x_free_fp_mem_at(struct bnx2x *bp, int fp_index)
4064 union host_hc_status_block *sb = &bnx2x_fp(bp, fp_index, status_blk);
4065 struct bnx2x_fastpath *fp = &bp->fp[fp_index];
4070 if (IS_FCOE_IDX(fp_index)) {
4071 memset(sb, 0, sizeof(union host_hc_status_block));
4072 fp->status_blk_mapping = 0;
4075 if (!CHIP_IS_E1x(bp))
4076 BNX2X_PCI_FREE(sb->e2_sb,
4077 bnx2x_fp(bp, fp_index,
4078 status_blk_mapping),
4079 sizeof(struct host_hc_status_block_e2));
4081 BNX2X_PCI_FREE(sb->e1x_sb,
4082 bnx2x_fp(bp, fp_index,
4083 status_blk_mapping),
4084 sizeof(struct host_hc_status_block_e1x));
4088 if (!skip_rx_queue(bp, fp_index)) {
4089 bnx2x_free_rx_bds(fp);
4091 /* fastpath rx rings: rx_buf rx_desc rx_comp */
4092 BNX2X_FREE(bnx2x_fp(bp, fp_index, rx_buf_ring));
4093 BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_desc_ring),
4094 bnx2x_fp(bp, fp_index, rx_desc_mapping),
4095 sizeof(struct eth_rx_bd) * NUM_RX_BD);
4097 BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_comp_ring),
4098 bnx2x_fp(bp, fp_index, rx_comp_mapping),
4099 sizeof(struct eth_fast_path_rx_cqe) *
4103 BNX2X_FREE(bnx2x_fp(bp, fp_index, rx_page_ring));
4104 BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_sge_ring),
4105 bnx2x_fp(bp, fp_index, rx_sge_mapping),
4106 BCM_PAGE_SIZE * NUM_RX_SGE_PAGES);
4110 if (!skip_tx_queue(bp, fp_index)) {
4111 /* fastpath tx rings: tx_buf tx_desc */
4112 for_each_cos_in_tx_queue(fp, cos) {
4113 struct bnx2x_fp_txdata *txdata = fp->txdata_ptr[cos];
4115 DP(NETIF_MSG_IFDOWN,
4116 "freeing tx memory of fp %d cos %d cid %d\n",
4117 fp_index, cos, txdata->cid);
4119 BNX2X_FREE(txdata->tx_buf_ring);
4120 BNX2X_PCI_FREE(txdata->tx_desc_ring,
4121 txdata->tx_desc_mapping,
4122 sizeof(union eth_tx_bd_types) * NUM_TX_BD);
4125 /* end of fastpath */
4128 void bnx2x_free_fp_mem_cnic(struct bnx2x *bp)
4131 for_each_cnic_queue(bp, i)
4132 bnx2x_free_fp_mem_at(bp, i);
4135 void bnx2x_free_fp_mem(struct bnx2x *bp)
4138 for_each_eth_queue(bp, i)
4139 bnx2x_free_fp_mem_at(bp, i);
4142 static void set_sb_shortcuts(struct bnx2x *bp, int index)
4144 union host_hc_status_block status_blk = bnx2x_fp(bp, index, status_blk);
4145 if (!CHIP_IS_E1x(bp)) {
4146 bnx2x_fp(bp, index, sb_index_values) =
4147 (__le16 *)status_blk.e2_sb->sb.index_values;
4148 bnx2x_fp(bp, index, sb_running_index) =
4149 (__le16 *)status_blk.e2_sb->sb.running_index;
4151 bnx2x_fp(bp, index, sb_index_values) =
4152 (__le16 *)status_blk.e1x_sb->sb.index_values;
4153 bnx2x_fp(bp, index, sb_running_index) =
4154 (__le16 *)status_blk.e1x_sb->sb.running_index;
4158 /* Returns the number of actually allocated BDs */
4159 static int bnx2x_alloc_rx_bds(struct bnx2x_fastpath *fp,
4162 struct bnx2x *bp = fp->bp;
4163 u16 ring_prod, cqe_ring_prod;
4164 int i, failure_cnt = 0;
4166 fp->rx_comp_cons = 0;
4167 cqe_ring_prod = ring_prod = 0;
4169 /* This routine is called only during fo init so
4170 * fp->eth_q_stats.rx_skb_alloc_failed = 0
4172 for (i = 0; i < rx_ring_size; i++) {
4173 if (bnx2x_alloc_rx_data(bp, fp, ring_prod) < 0) {
4177 ring_prod = NEXT_RX_IDX(ring_prod);
4178 cqe_ring_prod = NEXT_RCQ_IDX(cqe_ring_prod);
4179 WARN_ON(ring_prod <= (i - failure_cnt));
4183 BNX2X_ERR("was only able to allocate %d rx skbs on queue[%d]\n",
4184 i - failure_cnt, fp->index);
4186 fp->rx_bd_prod = ring_prod;
4187 /* Limit the CQE producer by the CQE ring size */
4188 fp->rx_comp_prod = min_t(u16, NUM_RCQ_RINGS*RCQ_DESC_CNT,
4190 fp->rx_pkt = fp->rx_calls = 0;
4192 bnx2x_fp_stats(bp, fp)->eth_q_stats.rx_skb_alloc_failed += failure_cnt;
4194 return i - failure_cnt;
4197 static void bnx2x_set_next_page_rx_cq(struct bnx2x_fastpath *fp)
4201 for (i = 1; i <= NUM_RCQ_RINGS; i++) {
4202 struct eth_rx_cqe_next_page *nextpg;
4204 nextpg = (struct eth_rx_cqe_next_page *)
4205 &fp->rx_comp_ring[RCQ_DESC_CNT * i - 1];
4207 cpu_to_le32(U64_HI(fp->rx_comp_mapping +
4208 BCM_PAGE_SIZE*(i % NUM_RCQ_RINGS)));
4210 cpu_to_le32(U64_LO(fp->rx_comp_mapping +
4211 BCM_PAGE_SIZE*(i % NUM_RCQ_RINGS)));
4215 static int bnx2x_alloc_fp_mem_at(struct bnx2x *bp, int index)
4217 union host_hc_status_block *sb;
4218 struct bnx2x_fastpath *fp = &bp->fp[index];
4221 int rx_ring_size = 0;
4223 if (!bp->rx_ring_size &&
4224 (IS_MF_STORAGE_SD(bp) || IS_MF_FCOE_AFEX(bp))) {
4225 rx_ring_size = MIN_RX_SIZE_NONTPA;
4226 bp->rx_ring_size = rx_ring_size;
4227 } else if (!bp->rx_ring_size) {
4228 rx_ring_size = MAX_RX_AVAIL/BNX2X_NUM_RX_QUEUES(bp);
4230 if (CHIP_IS_E3(bp)) {
4231 u32 cfg = SHMEM_RD(bp,
4232 dev_info.port_hw_config[BP_PORT(bp)].
4235 /* Decrease ring size for 1G functions */
4236 if ((cfg & PORT_HW_CFG_NET_SERDES_IF_MASK) ==
4237 PORT_HW_CFG_NET_SERDES_IF_SGMII)
4241 /* allocate at least number of buffers required by FW */
4242 rx_ring_size = max_t(int, bp->disable_tpa ? MIN_RX_SIZE_NONTPA :
4243 MIN_RX_SIZE_TPA, rx_ring_size);
4245 bp->rx_ring_size = rx_ring_size;
4246 } else /* if rx_ring_size specified - use it */
4247 rx_ring_size = bp->rx_ring_size;
4249 DP(BNX2X_MSG_SP, "calculated rx_ring_size %d\n", rx_ring_size);
4252 sb = &bnx2x_fp(bp, index, status_blk);
4254 if (!IS_FCOE_IDX(index)) {
4256 if (!CHIP_IS_E1x(bp))
4257 BNX2X_PCI_ALLOC(sb->e2_sb,
4258 &bnx2x_fp(bp, index, status_blk_mapping),
4259 sizeof(struct host_hc_status_block_e2));
4261 BNX2X_PCI_ALLOC(sb->e1x_sb,
4262 &bnx2x_fp(bp, index, status_blk_mapping),
4263 sizeof(struct host_hc_status_block_e1x));
4266 /* FCoE Queue uses Default SB and doesn't ACK the SB, thus no need to
4267 * set shortcuts for it.
4269 if (!IS_FCOE_IDX(index))
4270 set_sb_shortcuts(bp, index);
4273 if (!skip_tx_queue(bp, index)) {
4274 /* fastpath tx rings: tx_buf tx_desc */
4275 for_each_cos_in_tx_queue(fp, cos) {
4276 struct bnx2x_fp_txdata *txdata = fp->txdata_ptr[cos];
4279 "allocating tx memory of fp %d cos %d\n",
4282 BNX2X_ALLOC(txdata->tx_buf_ring,
4283 sizeof(struct sw_tx_bd) * NUM_TX_BD);
4284 BNX2X_PCI_ALLOC(txdata->tx_desc_ring,
4285 &txdata->tx_desc_mapping,
4286 sizeof(union eth_tx_bd_types) * NUM_TX_BD);
4291 if (!skip_rx_queue(bp, index)) {
4292 /* fastpath rx rings: rx_buf rx_desc rx_comp */
4293 BNX2X_ALLOC(bnx2x_fp(bp, index, rx_buf_ring),
4294 sizeof(struct sw_rx_bd) * NUM_RX_BD);
4295 BNX2X_PCI_ALLOC(bnx2x_fp(bp, index, rx_desc_ring),
4296 &bnx2x_fp(bp, index, rx_desc_mapping),
4297 sizeof(struct eth_rx_bd) * NUM_RX_BD);
4299 BNX2X_PCI_ALLOC(bnx2x_fp(bp, index, rx_comp_ring),
4300 &bnx2x_fp(bp, index, rx_comp_mapping),
4301 sizeof(struct eth_fast_path_rx_cqe) *
4305 BNX2X_ALLOC(bnx2x_fp(bp, index, rx_page_ring),
4306 sizeof(struct sw_rx_page) * NUM_RX_SGE);
4307 BNX2X_PCI_ALLOC(bnx2x_fp(bp, index, rx_sge_ring),
4308 &bnx2x_fp(bp, index, rx_sge_mapping),
4309 BCM_PAGE_SIZE * NUM_RX_SGE_PAGES);
4311 bnx2x_set_next_page_rx_bd(fp);
4314 bnx2x_set_next_page_rx_cq(fp);
4317 ring_size = bnx2x_alloc_rx_bds(fp, rx_ring_size);
4318 if (ring_size < rx_ring_size)
4324 /* handles low memory cases */
4326 BNX2X_ERR("Unable to allocate full memory for queue %d (size %d)\n",
4328 /* FW will drop all packets if queue is not big enough,
4329 * In these cases we disable the queue
4330 * Min size is different for OOO, TPA and non-TPA queues
4332 if (ring_size < (fp->disable_tpa ?
4333 MIN_RX_SIZE_NONTPA : MIN_RX_SIZE_TPA)) {
4334 /* release memory allocated for this queue */
4335 bnx2x_free_fp_mem_at(bp, index);
4341 int bnx2x_alloc_fp_mem_cnic(struct bnx2x *bp)
4345 if (bnx2x_alloc_fp_mem_at(bp, FCOE_IDX(bp)))
4346 /* we will fail load process instead of mark
4354 int bnx2x_alloc_fp_mem(struct bnx2x *bp)
4358 /* 1. Allocate FP for leading - fatal if error
4359 * 2. Allocate RSS - fix number of queues if error
4363 if (bnx2x_alloc_fp_mem_at(bp, 0))
4367 for_each_nondefault_eth_queue(bp, i)
4368 if (bnx2x_alloc_fp_mem_at(bp, i))
4371 /* handle memory failures */
4372 if (i != BNX2X_NUM_ETH_QUEUES(bp)) {
4373 int delta = BNX2X_NUM_ETH_QUEUES(bp) - i;
4376 bnx2x_shrink_eth_fp(bp, delta);
4377 if (CNIC_SUPPORT(bp))
4378 /* move non eth FPs next to last eth FP
4379 * must be done in that order
4380 * FCOE_IDX < FWD_IDX < OOO_IDX
4383 /* move FCoE fp even NO_FCOE_FLAG is on */
4384 bnx2x_move_fp(bp, FCOE_IDX(bp), FCOE_IDX(bp) - delta);
4385 bp->num_ethernet_queues -= delta;
4386 bp->num_queues = bp->num_ethernet_queues +
4387 bp->num_cnic_queues;
4388 BNX2X_ERR("Adjusted num of queues from %d to %d\n",
4389 bp->num_queues + delta, bp->num_queues);
4395 void bnx2x_free_mem_bp(struct bnx2x *bp)
4399 for (i = 0; i < bp->fp_array_size; i++)
4400 kfree(bp->fp[i].tpa_info);
4403 kfree(bp->fp_stats);
4404 kfree(bp->bnx2x_txq);
4405 kfree(bp->msix_table);
4409 int bnx2x_alloc_mem_bp(struct bnx2x *bp)
4411 struct bnx2x_fastpath *fp;
4412 struct msix_entry *tbl;
4413 struct bnx2x_ilt *ilt;
4414 int msix_table_size = 0;
4415 int fp_array_size, txq_array_size;
4419 * The biggest MSI-X table we might need is as a maximum number of fast
4420 * path IGU SBs plus default SB (for PF only).
4422 msix_table_size = bp->igu_sb_cnt;
4425 BNX2X_DEV_INFO("msix_table_size %d\n", msix_table_size);
4427 /* fp array: RSS plus CNIC related L2 queues */
4428 fp_array_size = BNX2X_MAX_RSS_COUNT(bp) + CNIC_SUPPORT(bp);
4429 bp->fp_array_size = fp_array_size;
4430 BNX2X_DEV_INFO("fp_array_size %d\n", bp->fp_array_size);
4432 fp = kcalloc(bp->fp_array_size, sizeof(*fp), GFP_KERNEL);
4435 for (i = 0; i < bp->fp_array_size; i++) {
4437 kcalloc(ETH_MAX_AGGREGATION_QUEUES_E1H_E2,
4438 sizeof(struct bnx2x_agg_info), GFP_KERNEL);
4439 if (!(fp[i].tpa_info))
4445 /* allocate sp objs */
4446 bp->sp_objs = kcalloc(bp->fp_array_size, sizeof(struct bnx2x_sp_objs),
4451 /* allocate fp_stats */
4452 bp->fp_stats = kcalloc(bp->fp_array_size, sizeof(struct bnx2x_fp_stats),
4457 /* Allocate memory for the transmission queues array */
4459 BNX2X_MAX_RSS_COUNT(bp) * BNX2X_MULTI_TX_COS + CNIC_SUPPORT(bp);
4460 BNX2X_DEV_INFO("txq_array_size %d", txq_array_size);
4462 bp->bnx2x_txq = kcalloc(txq_array_size, sizeof(struct bnx2x_fp_txdata),
4468 tbl = kcalloc(msix_table_size, sizeof(*tbl), GFP_KERNEL);
4471 bp->msix_table = tbl;
4474 ilt = kzalloc(sizeof(*ilt), GFP_KERNEL);
4481 bnx2x_free_mem_bp(bp);
4486 int bnx2x_reload_if_running(struct net_device *dev)
4488 struct bnx2x *bp = netdev_priv(dev);
4490 if (unlikely(!netif_running(dev)))
4493 bnx2x_nic_unload(bp, UNLOAD_NORMAL, true);
4494 return bnx2x_nic_load(bp, LOAD_NORMAL);
4497 int bnx2x_get_cur_phy_idx(struct bnx2x *bp)
4499 u32 sel_phy_idx = 0;
4500 if (bp->link_params.num_phys <= 1)
4503 if (bp->link_vars.link_up) {
4504 sel_phy_idx = EXT_PHY1;
4505 /* In case link is SERDES, check if the EXT_PHY2 is the one */
4506 if ((bp->link_vars.link_status & LINK_STATUS_SERDES_LINK) &&
4507 (bp->link_params.phy[EXT_PHY2].supported & SUPPORTED_FIBRE))
4508 sel_phy_idx = EXT_PHY2;
4511 switch (bnx2x_phy_selection(&bp->link_params)) {
4512 case PORT_HW_CFG_PHY_SELECTION_HARDWARE_DEFAULT:
4513 case PORT_HW_CFG_PHY_SELECTION_FIRST_PHY:
4514 case PORT_HW_CFG_PHY_SELECTION_FIRST_PHY_PRIORITY:
4515 sel_phy_idx = EXT_PHY1;
4517 case PORT_HW_CFG_PHY_SELECTION_SECOND_PHY:
4518 case PORT_HW_CFG_PHY_SELECTION_SECOND_PHY_PRIORITY:
4519 sel_phy_idx = EXT_PHY2;
4527 int bnx2x_get_link_cfg_idx(struct bnx2x *bp)
4529 u32 sel_phy_idx = bnx2x_get_cur_phy_idx(bp);
4531 * The selected activated PHY is always after swapping (in case PHY
4532 * swapping is enabled). So when swapping is enabled, we need to reverse
4536 if (bp->link_params.multi_phy_config &
4537 PORT_HW_CFG_PHY_SWAPPED_ENABLED) {
4538 if (sel_phy_idx == EXT_PHY1)
4539 sel_phy_idx = EXT_PHY2;
4540 else if (sel_phy_idx == EXT_PHY2)
4541 sel_phy_idx = EXT_PHY1;
4543 return LINK_CONFIG_IDX(sel_phy_idx);
4546 #ifdef NETDEV_FCOE_WWNN
4547 int bnx2x_fcoe_get_wwn(struct net_device *dev, u64 *wwn, int type)
4549 struct bnx2x *bp = netdev_priv(dev);
4550 struct cnic_eth_dev *cp = &bp->cnic_eth_dev;
4553 case NETDEV_FCOE_WWNN:
4554 *wwn = HILO_U64(cp->fcoe_wwn_node_name_hi,
4555 cp->fcoe_wwn_node_name_lo);
4557 case NETDEV_FCOE_WWPN:
4558 *wwn = HILO_U64(cp->fcoe_wwn_port_name_hi,
4559 cp->fcoe_wwn_port_name_lo);
4562 BNX2X_ERR("Wrong WWN type requested - %d\n", type);
4570 /* called with rtnl_lock */
4571 int bnx2x_change_mtu(struct net_device *dev, int new_mtu)
4573 struct bnx2x *bp = netdev_priv(dev);
4575 if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
4576 BNX2X_ERR("Can't perform change MTU during parity recovery\n");
4580 if ((new_mtu > ETH_MAX_JUMBO_PACKET_SIZE) ||
4581 ((new_mtu + ETH_HLEN) < ETH_MIN_PACKET_SIZE)) {
4582 BNX2X_ERR("Can't support requested MTU size\n");
4586 /* This does not race with packet allocation
4587 * because the actual alloc size is
4588 * only updated as part of load
4592 return bnx2x_reload_if_running(dev);
4595 netdev_features_t bnx2x_fix_features(struct net_device *dev,
4596 netdev_features_t features)
4598 struct bnx2x *bp = netdev_priv(dev);
4600 /* TPA requires Rx CSUM offloading */
4601 if (!(features & NETIF_F_RXCSUM) || bp->disable_tpa) {
4602 features &= ~NETIF_F_LRO;
4603 features &= ~NETIF_F_GRO;
4609 int bnx2x_set_features(struct net_device *dev, netdev_features_t features)
4611 struct bnx2x *bp = netdev_priv(dev);
4612 u32 flags = bp->flags;
4614 bool bnx2x_reload = false;
4616 if (features & NETIF_F_LRO)
4617 flags |= TPA_ENABLE_FLAG;
4619 flags &= ~TPA_ENABLE_FLAG;
4621 if (features & NETIF_F_GRO)
4622 flags |= GRO_ENABLE_FLAG;
4624 flags &= ~GRO_ENABLE_FLAG;
4626 if (features & NETIF_F_LOOPBACK) {
4627 if (bp->link_params.loopback_mode != LOOPBACK_BMAC) {
4628 bp->link_params.loopback_mode = LOOPBACK_BMAC;
4629 bnx2x_reload = true;
4632 if (bp->link_params.loopback_mode != LOOPBACK_NONE) {
4633 bp->link_params.loopback_mode = LOOPBACK_NONE;
4634 bnx2x_reload = true;
4638 changes = flags ^ bp->flags;
4640 /* if GRO is changed while LRO is enabled, dont force a reload */
4641 if ((changes & GRO_ENABLE_FLAG) && (flags & TPA_ENABLE_FLAG))
4642 changes &= ~GRO_ENABLE_FLAG;
4645 bnx2x_reload = true;
4650 if (bp->recovery_state == BNX2X_RECOVERY_DONE)
4651 return bnx2x_reload_if_running(dev);
4652 /* else: bnx2x_nic_load() will be called at end of recovery */
4658 void bnx2x_tx_timeout(struct net_device *dev)
4660 struct bnx2x *bp = netdev_priv(dev);
4662 #ifdef BNX2X_STOP_ON_ERROR
4667 smp_mb__before_clear_bit();
4668 set_bit(BNX2X_SP_RTNL_TX_TIMEOUT, &bp->sp_rtnl_state);
4669 smp_mb__after_clear_bit();
4671 /* This allows the netif to be shutdown gracefully before resetting */
4672 schedule_delayed_work(&bp->sp_rtnl_task, 0);
4675 int bnx2x_suspend(struct pci_dev *pdev, pm_message_t state)
4677 struct net_device *dev = pci_get_drvdata(pdev);
4681 dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
4684 bp = netdev_priv(dev);
4688 pci_save_state(pdev);
4690 if (!netif_running(dev)) {
4695 netif_device_detach(dev);
4697 bnx2x_nic_unload(bp, UNLOAD_CLOSE, false);
4699 bnx2x_set_power_state(bp, pci_choose_state(pdev, state));
4706 int bnx2x_resume(struct pci_dev *pdev)
4708 struct net_device *dev = pci_get_drvdata(pdev);
4713 dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
4716 bp = netdev_priv(dev);
4718 if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
4719 BNX2X_ERR("Handling parity error recovery. Try again later\n");
4725 pci_restore_state(pdev);
4727 if (!netif_running(dev)) {
4732 bnx2x_set_power_state(bp, PCI_D0);
4733 netif_device_attach(dev);
4735 rc = bnx2x_nic_load(bp, LOAD_OPEN);
4743 void bnx2x_set_ctx_validation(struct bnx2x *bp, struct eth_context *cxt,
4746 /* ustorm cxt validation */
4747 cxt->ustorm_ag_context.cdu_usage =
4748 CDU_RSRVD_VALUE_TYPE_A(HW_CID(bp, cid),
4749 CDU_REGION_NUMBER_UCM_AG, ETH_CONNECTION_TYPE);
4750 /* xcontext validation */
4751 cxt->xstorm_ag_context.cdu_reserved =
4752 CDU_RSRVD_VALUE_TYPE_A(HW_CID(bp, cid),
4753 CDU_REGION_NUMBER_XCM_AG, ETH_CONNECTION_TYPE);
4756 static void storm_memset_hc_timeout(struct bnx2x *bp, u8 port,
4757 u8 fw_sb_id, u8 sb_index,
4761 u32 addr = BAR_CSTRORM_INTMEM +
4762 CSTORM_STATUS_BLOCK_DATA_TIMEOUT_OFFSET(fw_sb_id, sb_index);
4763 REG_WR8(bp, addr, ticks);
4765 "port %x fw_sb_id %d sb_index %d ticks %d\n",
4766 port, fw_sb_id, sb_index, ticks);
4769 static void storm_memset_hc_disable(struct bnx2x *bp, u8 port,
4770 u16 fw_sb_id, u8 sb_index,
4773 u32 enable_flag = disable ? 0 : (1 << HC_INDEX_DATA_HC_ENABLED_SHIFT);
4774 u32 addr = BAR_CSTRORM_INTMEM +
4775 CSTORM_STATUS_BLOCK_DATA_FLAGS_OFFSET(fw_sb_id, sb_index);
4776 u8 flags = REG_RD8(bp, addr);
4778 flags &= ~HC_INDEX_DATA_HC_ENABLED;
4779 flags |= enable_flag;
4780 REG_WR8(bp, addr, flags);
4782 "port %x fw_sb_id %d sb_index %d disable %d\n",
4783 port, fw_sb_id, sb_index, disable);
4786 void bnx2x_update_coalesce_sb_index(struct bnx2x *bp, u8 fw_sb_id,
4787 u8 sb_index, u8 disable, u16 usec)
4789 int port = BP_PORT(bp);
4790 u8 ticks = usec / BNX2X_BTR;
4792 storm_memset_hc_timeout(bp, port, fw_sb_id, sb_index, ticks);
4794 disable = disable ? 1 : (usec ? 0 : 1);
4795 storm_memset_hc_disable(bp, port, fw_sb_id, sb_index, disable);