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 __skb_tx_hash(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)
2188 (SHMEM_RD(bp, func_mb[BP_FW_MB_IDX(bp)].drv_mb_header) &
2189 DRV_MSG_SEQ_NUMBER_MASK);
2190 BNX2X_DEV_INFO("fw_seq 0x%08x\n", bp->fw_seq);
2192 /* Get current FW pulse sequence */
2193 bp->fw_drv_pulse_wr_seq =
2194 (SHMEM_RD(bp, func_mb[BP_FW_MB_IDX(bp)].drv_pulse_mb) &
2195 DRV_PULSE_SEQ_MASK);
2196 BNX2X_DEV_INFO("drv_pulse 0x%x\n", bp->fw_drv_pulse_wr_seq);
2199 (*load_code) = bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_REQ,
2200 DRV_MSG_CODE_LOAD_REQ_WITH_LFA);
2202 /* if mcp fails to respond we must abort */
2203 if (!(*load_code)) {
2204 BNX2X_ERR("MCP response failure, aborting\n");
2208 /* If mcp refused (e.g. other port is in diagnostic mode) we
2211 if ((*load_code) == FW_MSG_CODE_DRV_LOAD_REFUSED) {
2212 BNX2X_ERR("MCP refused load request, aborting\n");
2218 /* check whether another PF has already loaded FW to chip. In
2219 * virtualized environments a pf from another VM may have already
2220 * initialized the device including loading FW
2222 int bnx2x_nic_load_analyze_req(struct bnx2x *bp, u32 load_code)
2224 /* is another pf loaded on this engine? */
2225 if (load_code != FW_MSG_CODE_DRV_LOAD_COMMON_CHIP &&
2226 load_code != FW_MSG_CODE_DRV_LOAD_COMMON) {
2227 /* build my FW version dword */
2228 u32 my_fw = (BCM_5710_FW_MAJOR_VERSION) +
2229 (BCM_5710_FW_MINOR_VERSION << 8) +
2230 (BCM_5710_FW_REVISION_VERSION << 16) +
2231 (BCM_5710_FW_ENGINEERING_VERSION << 24);
2233 /* read loaded FW from chip */
2234 u32 loaded_fw = REG_RD(bp, XSEM_REG_PRAM);
2236 DP(BNX2X_MSG_SP, "loaded fw %x, my fw %x\n",
2239 /* abort nic load if version mismatch */
2240 if (my_fw != loaded_fw) {
2241 BNX2X_ERR("bnx2x with FW %x was already loaded which mismatches my %x FW. aborting\n",
2249 /* returns the "mcp load_code" according to global load_count array */
2250 static int bnx2x_nic_load_no_mcp(struct bnx2x *bp, int port)
2252 int path = BP_PATH(bp);
2254 DP(NETIF_MSG_IFUP, "NO MCP - load counts[%d] %d, %d, %d\n",
2255 path, load_count[path][0], load_count[path][1],
2256 load_count[path][2]);
2257 load_count[path][0]++;
2258 load_count[path][1 + port]++;
2259 DP(NETIF_MSG_IFUP, "NO MCP - new load counts[%d] %d, %d, %d\n",
2260 path, load_count[path][0], load_count[path][1],
2261 load_count[path][2]);
2262 if (load_count[path][0] == 1)
2263 return FW_MSG_CODE_DRV_LOAD_COMMON;
2264 else if (load_count[path][1 + port] == 1)
2265 return FW_MSG_CODE_DRV_LOAD_PORT;
2267 return FW_MSG_CODE_DRV_LOAD_FUNCTION;
2270 /* mark PMF if applicable */
2271 static void bnx2x_nic_load_pmf(struct bnx2x *bp, u32 load_code)
2273 if ((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
2274 (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP) ||
2275 (load_code == FW_MSG_CODE_DRV_LOAD_PORT)) {
2277 /* We need the barrier to ensure the ordering between the
2278 * writing to bp->port.pmf here and reading it from the
2279 * bnx2x_periodic_task().
2286 DP(NETIF_MSG_LINK, "pmf %d\n", bp->port.pmf);
2289 static void bnx2x_nic_load_afex_dcc(struct bnx2x *bp, int load_code)
2291 if (((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
2292 (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP)) &&
2293 (bp->common.shmem2_base)) {
2294 if (SHMEM2_HAS(bp, dcc_support))
2295 SHMEM2_WR(bp, dcc_support,
2296 (SHMEM_DCC_SUPPORT_DISABLE_ENABLE_PF_TLV |
2297 SHMEM_DCC_SUPPORT_BANDWIDTH_ALLOCATION_TLV));
2298 if (SHMEM2_HAS(bp, afex_driver_support))
2299 SHMEM2_WR(bp, afex_driver_support,
2300 SHMEM_AFEX_SUPPORTED_VERSION_ONE);
2303 /* Set AFEX default VLAN tag to an invalid value */
2304 bp->afex_def_vlan_tag = -1;
2308 * bnx2x_bz_fp - zero content of the fastpath structure.
2310 * @bp: driver handle
2311 * @index: fastpath index to be zeroed
2313 * Makes sure the contents of the bp->fp[index].napi is kept
2316 static void bnx2x_bz_fp(struct bnx2x *bp, int index)
2318 struct bnx2x_fastpath *fp = &bp->fp[index];
2321 struct napi_struct orig_napi = fp->napi;
2322 struct bnx2x_agg_info *orig_tpa_info = fp->tpa_info;
2323 /* bzero bnx2x_fastpath contents */
2325 memset(fp->tpa_info, 0, ETH_MAX_AGGREGATION_QUEUES_E1H_E2 *
2326 sizeof(struct bnx2x_agg_info));
2327 memset(fp, 0, sizeof(*fp));
2329 /* Restore the NAPI object as it has been already initialized */
2330 fp->napi = orig_napi;
2331 fp->tpa_info = orig_tpa_info;
2335 fp->max_cos = bp->max_cos;
2337 /* Special queues support only one CoS */
2340 /* Init txdata pointers */
2342 fp->txdata_ptr[0] = &bp->bnx2x_txq[FCOE_TXQ_IDX(bp)];
2344 for_each_cos_in_tx_queue(fp, cos)
2345 fp->txdata_ptr[cos] = &bp->bnx2x_txq[cos *
2346 BNX2X_NUM_ETH_QUEUES(bp) + index];
2349 * set the tpa flag for each queue. The tpa flag determines the queue
2350 * minimal size so it must be set prior to queue memory allocation
2352 fp->disable_tpa = !(bp->flags & TPA_ENABLE_FLAG ||
2353 (bp->flags & GRO_ENABLE_FLAG &&
2354 bnx2x_mtu_allows_gro(bp->dev->mtu)));
2355 if (bp->flags & TPA_ENABLE_FLAG)
2356 fp->mode = TPA_MODE_LRO;
2357 else if (bp->flags & GRO_ENABLE_FLAG)
2358 fp->mode = TPA_MODE_GRO;
2360 /* We don't want TPA on an FCoE L2 ring */
2362 fp->disable_tpa = 1;
2365 int bnx2x_load_cnic(struct bnx2x *bp)
2367 int i, rc, port = BP_PORT(bp);
2369 DP(NETIF_MSG_IFUP, "Starting CNIC-related load\n");
2371 mutex_init(&bp->cnic_mutex);
2374 rc = bnx2x_alloc_mem_cnic(bp);
2376 BNX2X_ERR("Unable to allocate bp memory for cnic\n");
2377 LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic0);
2381 rc = bnx2x_alloc_fp_mem_cnic(bp);
2383 BNX2X_ERR("Unable to allocate memory for cnic fps\n");
2384 LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic0);
2387 /* Update the number of queues with the cnic queues */
2388 rc = bnx2x_set_real_num_queues(bp, 1);
2390 BNX2X_ERR("Unable to set real_num_queues including cnic\n");
2391 LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic0);
2394 /* Add all CNIC NAPI objects */
2395 bnx2x_add_all_napi_cnic(bp);
2396 DP(NETIF_MSG_IFUP, "cnic napi added\n");
2397 bnx2x_napi_enable_cnic(bp);
2399 rc = bnx2x_init_hw_func_cnic(bp);
2401 LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic1);
2403 bnx2x_nic_init_cnic(bp);
2406 /* Enable Timer scan */
2407 REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + port*4, 1);
2409 /* setup cnic queues */
2410 for_each_cnic_queue(bp, i) {
2411 rc = bnx2x_setup_queue(bp, &bp->fp[i], 0);
2413 BNX2X_ERR("Queue setup failed\n");
2414 LOAD_ERROR_EXIT(bp, load_error_cnic2);
2419 /* Initialize Rx filter. */
2420 netif_addr_lock_bh(bp->dev);
2421 bnx2x_set_rx_mode(bp->dev);
2422 netif_addr_unlock_bh(bp->dev);
2424 /* re-read iscsi info */
2425 bnx2x_get_iscsi_info(bp);
2426 bnx2x_setup_cnic_irq_info(bp);
2427 bnx2x_setup_cnic_info(bp);
2428 bp->cnic_loaded = true;
2429 if (bp->state == BNX2X_STATE_OPEN)
2430 bnx2x_cnic_notify(bp, CNIC_CTL_START_CMD);
2433 DP(NETIF_MSG_IFUP, "Ending successfully CNIC-related load\n");
2437 #ifndef BNX2X_STOP_ON_ERROR
2439 /* Disable Timer scan */
2440 REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + port*4, 0);
2443 bnx2x_napi_disable_cnic(bp);
2444 /* Update the number of queues without the cnic queues */
2445 rc = bnx2x_set_real_num_queues(bp, 0);
2447 BNX2X_ERR("Unable to set real_num_queues not including cnic\n");
2449 BNX2X_ERR("CNIC-related load failed\n");
2450 bnx2x_free_fp_mem_cnic(bp);
2451 bnx2x_free_mem_cnic(bp);
2453 #endif /* ! BNX2X_STOP_ON_ERROR */
2456 /* must be called with rtnl_lock */
2457 int bnx2x_nic_load(struct bnx2x *bp, int load_mode)
2459 int port = BP_PORT(bp);
2460 int i, rc = 0, load_code = 0;
2462 DP(NETIF_MSG_IFUP, "Starting NIC load\n");
2464 "CNIC is %s\n", CNIC_ENABLED(bp) ? "enabled" : "disabled");
2466 #ifdef BNX2X_STOP_ON_ERROR
2467 if (unlikely(bp->panic)) {
2468 BNX2X_ERR("Can't load NIC when there is panic\n");
2473 bp->state = BNX2X_STATE_OPENING_WAIT4_LOAD;
2475 memset(&bp->last_reported_link, 0, sizeof(bp->last_reported_link));
2476 __set_bit(BNX2X_LINK_REPORT_LINK_DOWN,
2477 &bp->last_reported_link.link_report_flags);
2480 /* must be called before memory allocation and HW init */
2481 bnx2x_ilt_set_info(bp);
2484 * Zero fastpath structures preserving invariants like napi, which are
2485 * allocated only once, fp index, max_cos, bp pointer.
2486 * Also set fp->disable_tpa and txdata_ptr.
2488 DP(NETIF_MSG_IFUP, "num queues: %d", bp->num_queues);
2489 for_each_queue(bp, i)
2491 memset(bp->bnx2x_txq, 0, (BNX2X_MAX_RSS_COUNT(bp) * BNX2X_MULTI_TX_COS +
2492 bp->num_cnic_queues) *
2493 sizeof(struct bnx2x_fp_txdata));
2495 bp->fcoe_init = false;
2497 /* Set the receive queues buffer size */
2498 bnx2x_set_rx_buf_size(bp);
2501 rc = bnx2x_alloc_mem(bp);
2503 BNX2X_ERR("Unable to allocate bp memory\n");
2508 /* Allocated memory for FW statistics */
2509 if (bnx2x_alloc_fw_stats_mem(bp))
2510 LOAD_ERROR_EXIT(bp, load_error0);
2512 /* need to be done after alloc mem, since it's self adjusting to amount
2513 * of memory available for RSS queues
2515 rc = bnx2x_alloc_fp_mem(bp);
2517 BNX2X_ERR("Unable to allocate memory for fps\n");
2518 LOAD_ERROR_EXIT(bp, load_error0);
2521 /* request pf to initialize status blocks */
2523 rc = bnx2x_vfpf_init(bp);
2525 LOAD_ERROR_EXIT(bp, load_error0);
2528 /* As long as bnx2x_alloc_mem() may possibly update
2529 * bp->num_queues, bnx2x_set_real_num_queues() should always
2530 * come after it. At this stage cnic queues are not counted.
2532 rc = bnx2x_set_real_num_queues(bp, 0);
2534 BNX2X_ERR("Unable to set real_num_queues\n");
2535 LOAD_ERROR_EXIT(bp, load_error0);
2538 /* configure multi cos mappings in kernel.
2539 * this configuration may be overriden by a multi class queue discipline
2540 * or by a dcbx negotiation result.
2542 bnx2x_setup_tc(bp->dev, bp->max_cos);
2544 /* Add all NAPI objects */
2545 bnx2x_add_all_napi(bp);
2546 DP(NETIF_MSG_IFUP, "napi added\n");
2547 bnx2x_napi_enable(bp);
2550 /* set pf load just before approaching the MCP */
2551 bnx2x_set_pf_load(bp);
2553 /* if mcp exists send load request and analyze response */
2554 if (!BP_NOMCP(bp)) {
2555 /* attempt to load pf */
2556 rc = bnx2x_nic_load_request(bp, &load_code);
2558 LOAD_ERROR_EXIT(bp, load_error1);
2560 /* what did mcp say? */
2561 rc = bnx2x_nic_load_analyze_req(bp, load_code);
2563 bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2564 LOAD_ERROR_EXIT(bp, load_error2);
2567 load_code = bnx2x_nic_load_no_mcp(bp, port);
2570 /* mark pmf if applicable */
2571 bnx2x_nic_load_pmf(bp, load_code);
2573 /* Init Function state controlling object */
2574 bnx2x__init_func_obj(bp);
2577 rc = bnx2x_init_hw(bp, load_code);
2579 BNX2X_ERR("HW init failed, aborting\n");
2580 bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2581 LOAD_ERROR_EXIT(bp, load_error2);
2585 bnx2x_pre_irq_nic_init(bp);
2587 /* Connect to IRQs */
2588 rc = bnx2x_setup_irqs(bp);
2590 BNX2X_ERR("setup irqs failed\n");
2592 bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2593 LOAD_ERROR_EXIT(bp, load_error2);
2596 /* Init per-function objects */
2598 /* Setup NIC internals and enable interrupts */
2599 bnx2x_post_irq_nic_init(bp, load_code);
2601 bnx2x_init_bp_objs(bp);
2602 bnx2x_iov_nic_init(bp);
2604 /* Set AFEX default VLAN tag to an invalid value */
2605 bp->afex_def_vlan_tag = -1;
2606 bnx2x_nic_load_afex_dcc(bp, load_code);
2607 bp->state = BNX2X_STATE_OPENING_WAIT4_PORT;
2608 rc = bnx2x_func_start(bp);
2610 BNX2X_ERR("Function start failed!\n");
2611 bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2613 LOAD_ERROR_EXIT(bp, load_error3);
2616 /* Send LOAD_DONE command to MCP */
2617 if (!BP_NOMCP(bp)) {
2618 load_code = bnx2x_fw_command(bp,
2619 DRV_MSG_CODE_LOAD_DONE, 0);
2621 BNX2X_ERR("MCP response failure, aborting\n");
2623 LOAD_ERROR_EXIT(bp, load_error3);
2627 /* initialize FW coalescing state machines in RAM */
2628 bnx2x_update_coalesce(bp);
2630 /* setup the leading queue */
2631 rc = bnx2x_setup_leading(bp);
2633 BNX2X_ERR("Setup leading failed!\n");
2634 LOAD_ERROR_EXIT(bp, load_error3);
2637 /* set up the rest of the queues */
2638 for_each_nondefault_eth_queue(bp, i) {
2639 rc = bnx2x_setup_queue(bp, &bp->fp[i], 0);
2641 BNX2X_ERR("Queue setup failed\n");
2642 LOAD_ERROR_EXIT(bp, load_error3);
2647 rc = bnx2x_init_rss_pf(bp);
2649 BNX2X_ERR("PF RSS init failed\n");
2650 LOAD_ERROR_EXIT(bp, load_error3);
2654 for_each_eth_queue(bp, i) {
2655 rc = bnx2x_vfpf_setup_q(bp, i);
2657 BNX2X_ERR("Queue setup failed\n");
2658 LOAD_ERROR_EXIT(bp, load_error3);
2663 /* Now when Clients are configured we are ready to work */
2664 bp->state = BNX2X_STATE_OPEN;
2666 /* Configure a ucast MAC */
2668 rc = bnx2x_set_eth_mac(bp, true);
2670 rc = bnx2x_vfpf_config_mac(bp, bp->dev->dev_addr, bp->fp->index,
2673 BNX2X_ERR("Setting Ethernet MAC failed\n");
2674 LOAD_ERROR_EXIT(bp, load_error3);
2677 if (IS_PF(bp) && bp->pending_max) {
2678 bnx2x_update_max_mf_config(bp, bp->pending_max);
2679 bp->pending_max = 0;
2683 rc = bnx2x_initial_phy_init(bp, load_mode);
2685 LOAD_ERROR_EXIT(bp, load_error3);
2687 bp->link_params.feature_config_flags &= ~FEATURE_CONFIG_BOOT_FROM_SAN;
2689 /* Start fast path */
2691 /* Initialize Rx filter. */
2692 netif_addr_lock_bh(bp->dev);
2693 bnx2x_set_rx_mode(bp->dev);
2694 netif_addr_unlock_bh(bp->dev);
2697 switch (load_mode) {
2699 /* Tx queue should be only reenabled */
2700 netif_tx_wake_all_queues(bp->dev);
2704 netif_tx_start_all_queues(bp->dev);
2705 smp_mb__after_clear_bit();
2709 case LOAD_LOOPBACK_EXT:
2710 bp->state = BNX2X_STATE_DIAG;
2718 bnx2x_update_drv_flags(bp, 1 << DRV_FLAGS_PORT_MASK, 0);
2720 bnx2x__link_status_update(bp);
2722 /* start the timer */
2723 mod_timer(&bp->timer, jiffies + bp->current_interval);
2725 if (CNIC_ENABLED(bp))
2726 bnx2x_load_cnic(bp);
2728 if (IS_PF(bp) && SHMEM2_HAS(bp, drv_capabilities_flag)) {
2729 /* mark driver is loaded in shmem2 */
2731 val = SHMEM2_RD(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)]);
2732 SHMEM2_WR(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)],
2733 val | DRV_FLAGS_CAPABILITIES_LOADED_SUPPORTED |
2734 DRV_FLAGS_CAPABILITIES_LOADED_L2);
2737 /* Wait for all pending SP commands to complete */
2738 if (IS_PF(bp) && !bnx2x_wait_sp_comp(bp, ~0x0UL)) {
2739 BNX2X_ERR("Timeout waiting for SP elements to complete\n");
2740 bnx2x_nic_unload(bp, UNLOAD_CLOSE, false);
2744 /* If PMF - send ADMIN DCBX msg to MFW to initiate DCBX FSM */
2745 if (bp->port.pmf && (bp->state != BNX2X_STATE_DIAG))
2746 bnx2x_dcbx_init(bp, false);
2748 DP(NETIF_MSG_IFUP, "Ending successfully NIC load\n");
2752 #ifndef BNX2X_STOP_ON_ERROR
2755 bnx2x_int_disable_sync(bp, 1);
2757 /* Clean queueable objects */
2758 bnx2x_squeeze_objects(bp);
2761 /* Free SKBs, SGEs, TPA pool and driver internals */
2762 bnx2x_free_skbs(bp);
2763 for_each_rx_queue(bp, i)
2764 bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
2769 if (IS_PF(bp) && !BP_NOMCP(bp)) {
2770 bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_REQ_WOL_MCP, 0);
2771 bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_DONE, 0);
2776 bnx2x_napi_disable(bp);
2777 bnx2x_del_all_napi(bp);
2779 /* clear pf_load status, as it was already set */
2781 bnx2x_clear_pf_load(bp);
2783 bnx2x_free_fp_mem(bp);
2784 bnx2x_free_fw_stats_mem(bp);
2788 #endif /* ! BNX2X_STOP_ON_ERROR */
2791 int bnx2x_drain_tx_queues(struct bnx2x *bp)
2795 /* Wait until tx fastpath tasks complete */
2796 for_each_tx_queue(bp, i) {
2797 struct bnx2x_fastpath *fp = &bp->fp[i];
2799 for_each_cos_in_tx_queue(fp, cos)
2800 rc = bnx2x_clean_tx_queue(bp, fp->txdata_ptr[cos]);
2807 /* must be called with rtnl_lock */
2808 int bnx2x_nic_unload(struct bnx2x *bp, int unload_mode, bool keep_link)
2811 bool global = false;
2813 DP(NETIF_MSG_IFUP, "Starting NIC unload\n");
2815 /* mark driver is unloaded in shmem2 */
2816 if (IS_PF(bp) && SHMEM2_HAS(bp, drv_capabilities_flag)) {
2818 val = SHMEM2_RD(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)]);
2819 SHMEM2_WR(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)],
2820 val & ~DRV_FLAGS_CAPABILITIES_LOADED_L2);
2823 if (IS_PF(bp) && bp->recovery_state != BNX2X_RECOVERY_DONE &&
2824 (bp->state == BNX2X_STATE_CLOSED ||
2825 bp->state == BNX2X_STATE_ERROR)) {
2826 /* We can get here if the driver has been unloaded
2827 * during parity error recovery and is either waiting for a
2828 * leader to complete or for other functions to unload and
2829 * then ifdown has been issued. In this case we want to
2830 * unload and let other functions to complete a recovery
2833 bp->recovery_state = BNX2X_RECOVERY_DONE;
2835 bnx2x_release_leader_lock(bp);
2838 DP(NETIF_MSG_IFDOWN, "Releasing a leadership...\n");
2839 BNX2X_ERR("Can't unload in closed or error state\n");
2843 /* Nothing to do during unload if previous bnx2x_nic_load()
2844 * have not completed succesfully - all resourses are released.
2846 * we can get here only after unsuccessful ndo_* callback, during which
2847 * dev->IFF_UP flag is still on.
2849 if (bp->state == BNX2X_STATE_CLOSED || bp->state == BNX2X_STATE_ERROR)
2852 /* It's important to set the bp->state to the value different from
2853 * BNX2X_STATE_OPEN and only then stop the Tx. Otherwise bnx2x_tx_int()
2854 * may restart the Tx from the NAPI context (see bnx2x_tx_int()).
2856 bp->state = BNX2X_STATE_CLOSING_WAIT4_HALT;
2859 if (CNIC_LOADED(bp))
2860 bnx2x_cnic_notify(bp, CNIC_CTL_STOP_CMD);
2863 bnx2x_tx_disable(bp);
2864 netdev_reset_tc(bp->dev);
2866 bp->rx_mode = BNX2X_RX_MODE_NONE;
2868 del_timer_sync(&bp->timer);
2871 /* Set ALWAYS_ALIVE bit in shmem */
2872 bp->fw_drv_pulse_wr_seq |= DRV_PULSE_ALWAYS_ALIVE;
2873 bnx2x_drv_pulse(bp);
2874 bnx2x_stats_handle(bp, STATS_EVENT_STOP);
2875 bnx2x_save_statistics(bp);
2878 /* wait till consumers catch up with producers in all queues */
2879 bnx2x_drain_tx_queues(bp);
2881 /* if VF indicate to PF this function is going down (PF will delete sp
2882 * elements and clear initializations
2885 bnx2x_vfpf_close_vf(bp);
2886 else if (unload_mode != UNLOAD_RECOVERY)
2887 /* if this is a normal/close unload need to clean up chip*/
2888 bnx2x_chip_cleanup(bp, unload_mode, keep_link);
2890 /* Send the UNLOAD_REQUEST to the MCP */
2891 bnx2x_send_unload_req(bp, unload_mode);
2894 * Prevent transactions to host from the functions on the
2895 * engine that doesn't reset global blocks in case of global
2896 * attention once gloabl blocks are reset and gates are opened
2897 * (the engine which leader will perform the recovery
2900 if (!CHIP_IS_E1x(bp))
2901 bnx2x_pf_disable(bp);
2903 /* Disable HW interrupts, NAPI */
2904 bnx2x_netif_stop(bp, 1);
2905 /* Delete all NAPI objects */
2906 bnx2x_del_all_napi(bp);
2907 if (CNIC_LOADED(bp))
2908 bnx2x_del_all_napi_cnic(bp);
2912 /* Report UNLOAD_DONE to MCP */
2913 bnx2x_send_unload_done(bp, false);
2917 * At this stage no more interrupts will arrive so we may safly clean
2918 * the queueable objects here in case they failed to get cleaned so far.
2921 bnx2x_squeeze_objects(bp);
2923 /* There should be no more pending SP commands at this stage */
2928 /* Free SKBs, SGEs, TPA pool and driver internals */
2929 bnx2x_free_skbs(bp);
2930 if (CNIC_LOADED(bp))
2931 bnx2x_free_skbs_cnic(bp);
2932 for_each_rx_queue(bp, i)
2933 bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
2935 bnx2x_free_fp_mem(bp);
2936 if (CNIC_LOADED(bp))
2937 bnx2x_free_fp_mem_cnic(bp);
2940 if (CNIC_LOADED(bp))
2941 bnx2x_free_mem_cnic(bp);
2944 bp->state = BNX2X_STATE_CLOSED;
2945 bp->cnic_loaded = false;
2947 /* Check if there are pending parity attentions. If there are - set
2948 * RECOVERY_IN_PROGRESS.
2950 if (IS_PF(bp) && bnx2x_chk_parity_attn(bp, &global, false)) {
2951 bnx2x_set_reset_in_progress(bp);
2953 /* Set RESET_IS_GLOBAL if needed */
2955 bnx2x_set_reset_global(bp);
2959 /* The last driver must disable a "close the gate" if there is no
2960 * parity attention or "process kill" pending.
2963 !bnx2x_clear_pf_load(bp) &&
2964 bnx2x_reset_is_done(bp, BP_PATH(bp)))
2965 bnx2x_disable_close_the_gate(bp);
2967 DP(NETIF_MSG_IFUP, "Ending NIC unload\n");
2972 int bnx2x_set_power_state(struct bnx2x *bp, pci_power_t state)
2976 /* If there is no power capability, silently succeed */
2978 BNX2X_DEV_INFO("No power capability. Breaking.\n");
2982 pci_read_config_word(bp->pdev, bp->pm_cap + PCI_PM_CTRL, &pmcsr);
2986 pci_write_config_word(bp->pdev, bp->pm_cap + PCI_PM_CTRL,
2987 ((pmcsr & ~PCI_PM_CTRL_STATE_MASK) |
2988 PCI_PM_CTRL_PME_STATUS));
2990 if (pmcsr & PCI_PM_CTRL_STATE_MASK)
2991 /* delay required during transition out of D3hot */
2996 /* If there are other clients above don't
2997 shut down the power */
2998 if (atomic_read(&bp->pdev->enable_cnt) != 1)
3000 /* Don't shut down the power for emulation and FPGA */
3001 if (CHIP_REV_IS_SLOW(bp))
3004 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
3008 pmcsr |= PCI_PM_CTRL_PME_ENABLE;
3010 pci_write_config_word(bp->pdev, bp->pm_cap + PCI_PM_CTRL,
3013 /* No more memory access after this point until
3014 * device is brought back to D0.
3019 dev_err(&bp->pdev->dev, "Can't support state = %d\n", state);
3026 * net_device service functions
3028 int bnx2x_poll(struct napi_struct *napi, int budget)
3032 struct bnx2x_fastpath *fp = container_of(napi, struct bnx2x_fastpath,
3034 struct bnx2x *bp = fp->bp;
3037 #ifdef BNX2X_STOP_ON_ERROR
3038 if (unlikely(bp->panic)) {
3039 napi_complete(napi);
3044 for_each_cos_in_tx_queue(fp, cos)
3045 if (bnx2x_tx_queue_has_work(fp->txdata_ptr[cos]))
3046 bnx2x_tx_int(bp, fp->txdata_ptr[cos]);
3048 if (bnx2x_has_rx_work(fp)) {
3049 work_done += bnx2x_rx_int(fp, budget - work_done);
3051 /* must not complete if we consumed full budget */
3052 if (work_done >= budget)
3056 /* Fall out from the NAPI loop if needed */
3057 if (!(bnx2x_has_rx_work(fp) || bnx2x_has_tx_work(fp))) {
3059 /* No need to update SB for FCoE L2 ring as long as
3060 * it's connected to the default SB and the SB
3061 * has been updated when NAPI was scheduled.
3063 if (IS_FCOE_FP(fp)) {
3064 napi_complete(napi);
3067 bnx2x_update_fpsb_idx(fp);
3068 /* bnx2x_has_rx_work() reads the status block,
3069 * thus we need to ensure that status block indices
3070 * have been actually read (bnx2x_update_fpsb_idx)
3071 * prior to this check (bnx2x_has_rx_work) so that
3072 * we won't write the "newer" value of the status block
3073 * to IGU (if there was a DMA right after
3074 * bnx2x_has_rx_work and if there is no rmb, the memory
3075 * reading (bnx2x_update_fpsb_idx) may be postponed
3076 * to right before bnx2x_ack_sb). In this case there
3077 * will never be another interrupt until there is
3078 * another update of the status block, while there
3079 * is still unhandled work.
3083 if (!(bnx2x_has_rx_work(fp) || bnx2x_has_tx_work(fp))) {
3084 napi_complete(napi);
3085 /* Re-enable interrupts */
3086 DP(NETIF_MSG_RX_STATUS,
3087 "Update index to %d\n", fp->fp_hc_idx);
3088 bnx2x_ack_sb(bp, fp->igu_sb_id, USTORM_ID,
3089 le16_to_cpu(fp->fp_hc_idx),
3099 /* we split the first BD into headers and data BDs
3100 * to ease the pain of our fellow microcode engineers
3101 * we use one mapping for both BDs
3103 static u16 bnx2x_tx_split(struct bnx2x *bp,
3104 struct bnx2x_fp_txdata *txdata,
3105 struct sw_tx_bd *tx_buf,
3106 struct eth_tx_start_bd **tx_bd, u16 hlen,
3109 struct eth_tx_start_bd *h_tx_bd = *tx_bd;
3110 struct eth_tx_bd *d_tx_bd;
3112 int old_len = le16_to_cpu(h_tx_bd->nbytes);
3114 /* first fix first BD */
3115 h_tx_bd->nbytes = cpu_to_le16(hlen);
3117 DP(NETIF_MSG_TX_QUEUED, "TSO split header size is %d (%x:%x)\n",
3118 h_tx_bd->nbytes, h_tx_bd->addr_hi, h_tx_bd->addr_lo);
3120 /* now get a new data BD
3121 * (after the pbd) and fill it */
3122 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3123 d_tx_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
3125 mapping = HILO_U64(le32_to_cpu(h_tx_bd->addr_hi),
3126 le32_to_cpu(h_tx_bd->addr_lo)) + hlen;
3128 d_tx_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
3129 d_tx_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
3130 d_tx_bd->nbytes = cpu_to_le16(old_len - hlen);
3132 /* this marks the BD as one that has no individual mapping */
3133 tx_buf->flags |= BNX2X_TSO_SPLIT_BD;
3135 DP(NETIF_MSG_TX_QUEUED,
3136 "TSO split data size is %d (%x:%x)\n",
3137 d_tx_bd->nbytes, d_tx_bd->addr_hi, d_tx_bd->addr_lo);
3140 *tx_bd = (struct eth_tx_start_bd *)d_tx_bd;
3145 #define bswab32(b32) ((__force __le32) swab32((__force __u32) (b32)))
3146 #define bswab16(b16) ((__force __le16) swab16((__force __u16) (b16)))
3147 static __le16 bnx2x_csum_fix(unsigned char *t_header, u16 csum, s8 fix)
3149 __sum16 tsum = (__force __sum16) csum;
3152 tsum = ~csum_fold(csum_sub((__force __wsum) csum,
3153 csum_partial(t_header - fix, fix, 0)));
3156 tsum = ~csum_fold(csum_add((__force __wsum) csum,
3157 csum_partial(t_header, -fix, 0)));
3159 return bswab16(tsum);
3162 static u32 bnx2x_xmit_type(struct bnx2x *bp, struct sk_buff *skb)
3168 if (skb->ip_summed != CHECKSUM_PARTIAL)
3171 protocol = vlan_get_protocol(skb);
3172 if (protocol == htons(ETH_P_IPV6)) {
3174 prot = ipv6_hdr(skb)->nexthdr;
3177 prot = ip_hdr(skb)->protocol;
3180 if (!CHIP_IS_E1x(bp) && skb->encapsulation) {
3181 if (inner_ip_hdr(skb)->version == 6) {
3182 rc |= XMIT_CSUM_ENC_V6;
3183 if (inner_ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3184 rc |= XMIT_CSUM_TCP;
3186 rc |= XMIT_CSUM_ENC_V4;
3187 if (inner_ip_hdr(skb)->protocol == IPPROTO_TCP)
3188 rc |= XMIT_CSUM_TCP;
3191 if (prot == IPPROTO_TCP)
3192 rc |= XMIT_CSUM_TCP;
3194 if (skb_is_gso_v6(skb)) {
3195 rc |= (XMIT_GSO_V6 | XMIT_CSUM_TCP | XMIT_CSUM_V6);
3196 if (rc & XMIT_CSUM_ENC)
3197 rc |= XMIT_GSO_ENC_V6;
3198 } else if (skb_is_gso(skb)) {
3199 rc |= (XMIT_GSO_V4 | XMIT_CSUM_V4 | XMIT_CSUM_TCP);
3200 if (rc & XMIT_CSUM_ENC)
3201 rc |= XMIT_GSO_ENC_V4;
3207 #if (MAX_SKB_FRAGS >= MAX_FETCH_BD - 3)
3208 /* check if packet requires linearization (packet is too fragmented)
3209 no need to check fragmentation if page size > 8K (there will be no
3210 violation to FW restrictions) */
3211 static int bnx2x_pkt_req_lin(struct bnx2x *bp, struct sk_buff *skb,
3216 int first_bd_sz = 0;
3218 /* 3 = 1 (for linear data BD) + 2 (for PBD and last BD) */
3219 if (skb_shinfo(skb)->nr_frags >= (MAX_FETCH_BD - 3)) {
3221 if (xmit_type & XMIT_GSO) {
3222 unsigned short lso_mss = skb_shinfo(skb)->gso_size;
3223 /* Check if LSO packet needs to be copied:
3224 3 = 1 (for headers BD) + 2 (for PBD and last BD) */
3225 int wnd_size = MAX_FETCH_BD - 3;
3226 /* Number of windows to check */
3227 int num_wnds = skb_shinfo(skb)->nr_frags - wnd_size;
3232 /* Headers length */
3233 hlen = (int)(skb_transport_header(skb) - skb->data) +
3236 /* Amount of data (w/o headers) on linear part of SKB*/
3237 first_bd_sz = skb_headlen(skb) - hlen;
3239 wnd_sum = first_bd_sz;
3241 /* Calculate the first sum - it's special */
3242 for (frag_idx = 0; frag_idx < wnd_size - 1; frag_idx++)
3244 skb_frag_size(&skb_shinfo(skb)->frags[frag_idx]);
3246 /* If there was data on linear skb data - check it */
3247 if (first_bd_sz > 0) {
3248 if (unlikely(wnd_sum < lso_mss)) {
3253 wnd_sum -= first_bd_sz;
3256 /* Others are easier: run through the frag list and
3257 check all windows */
3258 for (wnd_idx = 0; wnd_idx <= num_wnds; wnd_idx++) {
3260 skb_frag_size(&skb_shinfo(skb)->frags[wnd_idx + wnd_size - 1]);
3262 if (unlikely(wnd_sum < lso_mss)) {
3267 skb_frag_size(&skb_shinfo(skb)->frags[wnd_idx]);
3270 /* in non-LSO too fragmented packet should always
3277 if (unlikely(to_copy))
3278 DP(NETIF_MSG_TX_QUEUED,
3279 "Linearization IS REQUIRED for %s packet. num_frags %d hlen %d first_bd_sz %d\n",
3280 (xmit_type & XMIT_GSO) ? "LSO" : "non-LSO",
3281 skb_shinfo(skb)->nr_frags, hlen, first_bd_sz);
3287 static void bnx2x_set_pbd_gso_e2(struct sk_buff *skb, u32 *parsing_data,
3290 struct ipv6hdr *ipv6;
3292 *parsing_data |= (skb_shinfo(skb)->gso_size <<
3293 ETH_TX_PARSE_BD_E2_LSO_MSS_SHIFT) &
3294 ETH_TX_PARSE_BD_E2_LSO_MSS;
3296 if (xmit_type & XMIT_GSO_ENC_V6)
3297 ipv6 = inner_ipv6_hdr(skb);
3298 else if (xmit_type & XMIT_GSO_V6)
3299 ipv6 = ipv6_hdr(skb);
3303 if (ipv6 && ipv6->nexthdr == NEXTHDR_IPV6)
3304 *parsing_data |= ETH_TX_PARSE_BD_E2_IPV6_WITH_EXT_HDR;
3308 * bnx2x_set_pbd_gso - update PBD in GSO case.
3312 * @xmit_type: xmit flags
3314 static void bnx2x_set_pbd_gso(struct sk_buff *skb,
3315 struct eth_tx_parse_bd_e1x *pbd,
3316 struct eth_tx_start_bd *tx_start_bd,
3319 pbd->lso_mss = cpu_to_le16(skb_shinfo(skb)->gso_size);
3320 pbd->tcp_send_seq = bswab32(tcp_hdr(skb)->seq);
3321 pbd->tcp_flags = pbd_tcp_flags(tcp_hdr(skb));
3323 if (xmit_type & XMIT_GSO_V4) {
3324 pbd->ip_id = bswab16(ip_hdr(skb)->id);
3325 pbd->tcp_pseudo_csum =
3326 bswab16(~csum_tcpudp_magic(ip_hdr(skb)->saddr,
3328 0, IPPROTO_TCP, 0));
3330 /* GSO on 57710/57711 needs FW to calculate IP checksum */
3331 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_IP_CSUM;
3333 pbd->tcp_pseudo_csum =
3334 bswab16(~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3335 &ipv6_hdr(skb)->daddr,
3336 0, IPPROTO_TCP, 0));
3340 cpu_to_le16(ETH_TX_PARSE_BD_E1X_PSEUDO_CS_WITHOUT_LEN);
3344 * bnx2x_set_pbd_csum_enc - update PBD with checksum and return header length
3346 * @bp: driver handle
3348 * @parsing_data: data to be updated
3349 * @xmit_type: xmit flags
3351 * 57712/578xx related, when skb has encapsulation
3353 static u8 bnx2x_set_pbd_csum_enc(struct bnx2x *bp, struct sk_buff *skb,
3354 u32 *parsing_data, u32 xmit_type)
3357 ((((u8 *)skb_inner_transport_header(skb) - skb->data) >> 1) <<
3358 ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W_SHIFT) &
3359 ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W;
3361 if (xmit_type & XMIT_CSUM_TCP) {
3362 *parsing_data |= ((inner_tcp_hdrlen(skb) / 4) <<
3363 ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW_SHIFT) &
3364 ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW;
3366 return skb_inner_transport_header(skb) +
3367 inner_tcp_hdrlen(skb) - skb->data;
3370 /* We support checksum offload for TCP and UDP only.
3371 * No need to pass the UDP header length - it's a constant.
3373 return skb_inner_transport_header(skb) +
3374 sizeof(struct udphdr) - skb->data;
3378 * bnx2x_set_pbd_csum_e2 - update PBD with checksum and return header length
3380 * @bp: driver handle
3382 * @parsing_data: data to be updated
3383 * @xmit_type: xmit flags
3385 * 57712/578xx related
3387 static u8 bnx2x_set_pbd_csum_e2(struct bnx2x *bp, struct sk_buff *skb,
3388 u32 *parsing_data, u32 xmit_type)
3391 ((((u8 *)skb_transport_header(skb) - skb->data) >> 1) <<
3392 ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W_SHIFT) &
3393 ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W;
3395 if (xmit_type & XMIT_CSUM_TCP) {
3396 *parsing_data |= ((tcp_hdrlen(skb) / 4) <<
3397 ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW_SHIFT) &
3398 ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW;
3400 return skb_transport_header(skb) + tcp_hdrlen(skb) - skb->data;
3402 /* We support checksum offload for TCP and UDP only.
3403 * No need to pass the UDP header length - it's a constant.
3405 return skb_transport_header(skb) + sizeof(struct udphdr) - skb->data;
3408 /* set FW indication according to inner or outer protocols if tunneled */
3409 static void bnx2x_set_sbd_csum(struct bnx2x *bp, struct sk_buff *skb,
3410 struct eth_tx_start_bd *tx_start_bd,
3413 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_L4_CSUM;
3415 if (xmit_type & (XMIT_CSUM_ENC_V6 | XMIT_CSUM_V6))
3416 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_IPV6;
3418 if (!(xmit_type & XMIT_CSUM_TCP))
3419 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_IS_UDP;
3423 * bnx2x_set_pbd_csum - update PBD with checksum and return header length
3425 * @bp: driver handle
3427 * @pbd: parse BD to be updated
3428 * @xmit_type: xmit flags
3430 static u8 bnx2x_set_pbd_csum(struct bnx2x *bp, struct sk_buff *skb,
3431 struct eth_tx_parse_bd_e1x *pbd,
3434 u8 hlen = (skb_network_header(skb) - skb->data) >> 1;
3436 /* for now NS flag is not used in Linux */
3439 ((skb->protocol == cpu_to_be16(ETH_P_8021Q)) <<
3440 ETH_TX_PARSE_BD_E1X_LLC_SNAP_EN_SHIFT));
3442 pbd->ip_hlen_w = (skb_transport_header(skb) -
3443 skb_network_header(skb)) >> 1;
3445 hlen += pbd->ip_hlen_w;
3447 /* We support checksum offload for TCP and UDP only */
3448 if (xmit_type & XMIT_CSUM_TCP)
3449 hlen += tcp_hdrlen(skb) / 2;
3451 hlen += sizeof(struct udphdr) / 2;
3453 pbd->total_hlen_w = cpu_to_le16(hlen);
3456 if (xmit_type & XMIT_CSUM_TCP) {
3457 pbd->tcp_pseudo_csum = bswab16(tcp_hdr(skb)->check);
3460 s8 fix = SKB_CS_OFF(skb); /* signed! */
3462 DP(NETIF_MSG_TX_QUEUED,
3463 "hlen %d fix %d csum before fix %x\n",
3464 le16_to_cpu(pbd->total_hlen_w), fix, SKB_CS(skb));
3466 /* HW bug: fixup the CSUM */
3467 pbd->tcp_pseudo_csum =
3468 bnx2x_csum_fix(skb_transport_header(skb),
3471 DP(NETIF_MSG_TX_QUEUED, "csum after fix %x\n",
3472 pbd->tcp_pseudo_csum);
3478 static void bnx2x_update_pbds_gso_enc(struct sk_buff *skb,
3479 struct eth_tx_parse_bd_e2 *pbd_e2,
3480 struct eth_tx_parse_2nd_bd *pbd2,
3485 u8 outerip_off, outerip_len = 0;
3486 /* from outer IP to transport */
3487 hlen_w = (skb_inner_transport_header(skb) -
3488 skb_network_header(skb)) >> 1;
3491 if (xmit_type & XMIT_CSUM_TCP)
3492 hlen_w += inner_tcp_hdrlen(skb) >> 1;
3494 hlen_w += sizeof(struct udphdr) >> 1;
3496 pbd2->fw_ip_hdr_to_payload_w = hlen_w;
3498 if (xmit_type & XMIT_CSUM_ENC_V4) {
3499 struct iphdr *iph = ip_hdr(skb);
3500 pbd2->fw_ip_csum_wo_len_flags_frag =
3501 bswab16(csum_fold((~iph->check) -
3502 iph->tot_len - iph->frag_off));
3504 pbd2->fw_ip_hdr_to_payload_w =
3505 hlen_w - ((sizeof(struct ipv6hdr)) >> 1);
3508 pbd2->tcp_send_seq = bswab32(inner_tcp_hdr(skb)->seq);
3510 pbd2->tcp_flags = pbd_tcp_flags(inner_tcp_hdr(skb));
3512 if (xmit_type & XMIT_GSO_V4) {
3513 pbd2->hw_ip_id = bswab16(inner_ip_hdr(skb)->id);
3515 pbd_e2->data.tunnel_data.pseudo_csum =
3516 bswab16(~csum_tcpudp_magic(
3517 inner_ip_hdr(skb)->saddr,
3518 inner_ip_hdr(skb)->daddr,
3519 0, IPPROTO_TCP, 0));
3521 outerip_len = ip_hdr(skb)->ihl << 1;
3523 pbd_e2->data.tunnel_data.pseudo_csum =
3524 bswab16(~csum_ipv6_magic(
3525 &inner_ipv6_hdr(skb)->saddr,
3526 &inner_ipv6_hdr(skb)->daddr,
3527 0, IPPROTO_TCP, 0));
3530 outerip_off = (skb_network_header(skb) - skb->data) >> 1;
3534 (!!(xmit_type & XMIT_CSUM_V6) <<
3535 ETH_TX_PARSE_2ND_BD_IP_HDR_TYPE_OUTER_SHIFT) |
3537 ETH_TX_PARSE_2ND_BD_IP_HDR_LEN_OUTER_W_SHIFT) |
3538 ((skb->protocol == cpu_to_be16(ETH_P_8021Q)) <<
3539 ETH_TX_PARSE_2ND_BD_LLC_SNAP_EN_SHIFT);
3541 if (ip_hdr(skb)->protocol == IPPROTO_UDP) {
3542 SET_FLAG(*global_data, ETH_TX_PARSE_2ND_BD_TUNNEL_UDP_EXIST, 1);
3543 pbd2->tunnel_udp_hdr_start_w = skb_transport_offset(skb) >> 1;
3547 /* called with netif_tx_lock
3548 * bnx2x_tx_int() runs without netif_tx_lock unless it needs to call
3549 * netif_wake_queue()
3551 netdev_tx_t bnx2x_start_xmit(struct sk_buff *skb, struct net_device *dev)
3553 struct bnx2x *bp = netdev_priv(dev);
3555 struct netdev_queue *txq;
3556 struct bnx2x_fp_txdata *txdata;
3557 struct sw_tx_bd *tx_buf;
3558 struct eth_tx_start_bd *tx_start_bd, *first_bd;
3559 struct eth_tx_bd *tx_data_bd, *total_pkt_bd = NULL;
3560 struct eth_tx_parse_bd_e1x *pbd_e1x = NULL;
3561 struct eth_tx_parse_bd_e2 *pbd_e2 = NULL;
3562 struct eth_tx_parse_2nd_bd *pbd2 = NULL;
3563 u32 pbd_e2_parsing_data = 0;
3564 u16 pkt_prod, bd_prod;
3567 u32 xmit_type = bnx2x_xmit_type(bp, skb);
3570 __le16 pkt_size = 0;
3572 u8 mac_type = UNICAST_ADDRESS;
3574 #ifdef BNX2X_STOP_ON_ERROR
3575 if (unlikely(bp->panic))
3576 return NETDEV_TX_BUSY;
3579 txq_index = skb_get_queue_mapping(skb);
3580 txq = netdev_get_tx_queue(dev, txq_index);
3582 BUG_ON(txq_index >= MAX_ETH_TXQ_IDX(bp) + (CNIC_LOADED(bp) ? 1 : 0));
3584 txdata = &bp->bnx2x_txq[txq_index];
3586 /* enable this debug print to view the transmission queue being used
3587 DP(NETIF_MSG_TX_QUEUED, "indices: txq %d, fp %d, txdata %d\n",
3588 txq_index, fp_index, txdata_index); */
3590 /* enable this debug print to view the tranmission details
3591 DP(NETIF_MSG_TX_QUEUED,
3592 "transmitting packet cid %d fp index %d txdata_index %d tx_data ptr %p fp pointer %p\n",
3593 txdata->cid, fp_index, txdata_index, txdata, fp); */
3595 if (unlikely(bnx2x_tx_avail(bp, txdata) <
3596 skb_shinfo(skb)->nr_frags +
3598 NEXT_CNT_PER_TX_PKT(MAX_BDS_PER_TX_PKT))) {
3599 /* Handle special storage cases separately */
3600 if (txdata->tx_ring_size == 0) {
3601 struct bnx2x_eth_q_stats *q_stats =
3602 bnx2x_fp_qstats(bp, txdata->parent_fp);
3603 q_stats->driver_filtered_tx_pkt++;
3605 return NETDEV_TX_OK;
3607 bnx2x_fp_qstats(bp, txdata->parent_fp)->driver_xoff++;
3608 netif_tx_stop_queue(txq);
3609 BNX2X_ERR("BUG! Tx ring full when queue awake!\n");
3611 return NETDEV_TX_BUSY;
3614 DP(NETIF_MSG_TX_QUEUED,
3615 "queue[%d]: SKB: summed %x protocol %x protocol(%x,%x) gso type %x xmit_type %x len %d\n",
3616 txq_index, skb->ip_summed, skb->protocol, ipv6_hdr(skb)->nexthdr,
3617 ip_hdr(skb)->protocol, skb_shinfo(skb)->gso_type, xmit_type,
3620 eth = (struct ethhdr *)skb->data;
3622 /* set flag according to packet type (UNICAST_ADDRESS is default)*/
3623 if (unlikely(is_multicast_ether_addr(eth->h_dest))) {
3624 if (is_broadcast_ether_addr(eth->h_dest))
3625 mac_type = BROADCAST_ADDRESS;
3627 mac_type = MULTICAST_ADDRESS;
3630 #if (MAX_SKB_FRAGS >= MAX_FETCH_BD - BDS_PER_TX_PKT)
3631 /* First, check if we need to linearize the skb (due to FW
3632 restrictions). No need to check fragmentation if page size > 8K
3633 (there will be no violation to FW restrictions) */
3634 if (bnx2x_pkt_req_lin(bp, skb, xmit_type)) {
3635 /* Statistics of linearization */
3637 if (skb_linearize(skb) != 0) {
3638 DP(NETIF_MSG_TX_QUEUED,
3639 "SKB linearization failed - silently dropping this SKB\n");
3640 dev_kfree_skb_any(skb);
3641 return NETDEV_TX_OK;
3645 /* Map skb linear data for DMA */
3646 mapping = dma_map_single(&bp->pdev->dev, skb->data,
3647 skb_headlen(skb), DMA_TO_DEVICE);
3648 if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
3649 DP(NETIF_MSG_TX_QUEUED,
3650 "SKB mapping failed - silently dropping this SKB\n");
3651 dev_kfree_skb_any(skb);
3652 return NETDEV_TX_OK;
3655 Please read carefully. First we use one BD which we mark as start,
3656 then we have a parsing info BD (used for TSO or xsum),
3657 and only then we have the rest of the TSO BDs.
3658 (don't forget to mark the last one as last,
3659 and to unmap only AFTER you write to the BD ...)
3660 And above all, all pdb sizes are in words - NOT DWORDS!
3663 /* get current pkt produced now - advance it just before sending packet
3664 * since mapping of pages may fail and cause packet to be dropped
3666 pkt_prod = txdata->tx_pkt_prod;
3667 bd_prod = TX_BD(txdata->tx_bd_prod);
3669 /* get a tx_buf and first BD
3670 * tx_start_bd may be changed during SPLIT,
3671 * but first_bd will always stay first
3673 tx_buf = &txdata->tx_buf_ring[TX_BD(pkt_prod)];
3674 tx_start_bd = &txdata->tx_desc_ring[bd_prod].start_bd;
3675 first_bd = tx_start_bd;
3677 tx_start_bd->bd_flags.as_bitfield = ETH_TX_BD_FLAGS_START_BD;
3679 /* header nbd: indirectly zero other flags! */
3680 tx_start_bd->general_data = 1 << ETH_TX_START_BD_HDR_NBDS_SHIFT;
3682 /* remember the first BD of the packet */
3683 tx_buf->first_bd = txdata->tx_bd_prod;
3687 DP(NETIF_MSG_TX_QUEUED,
3688 "sending pkt %u @%p next_idx %u bd %u @%p\n",
3689 pkt_prod, tx_buf, txdata->tx_pkt_prod, bd_prod, tx_start_bd);
3691 if (vlan_tx_tag_present(skb)) {
3692 tx_start_bd->vlan_or_ethertype =
3693 cpu_to_le16(vlan_tx_tag_get(skb));
3694 tx_start_bd->bd_flags.as_bitfield |=
3695 (X_ETH_OUTBAND_VLAN << ETH_TX_BD_FLAGS_VLAN_MODE_SHIFT);
3697 /* when transmitting in a vf, start bd must hold the ethertype
3698 * for fw to enforce it
3701 tx_start_bd->vlan_or_ethertype =
3702 cpu_to_le16(ntohs(eth->h_proto));
3704 /* used by FW for packet accounting */
3705 tx_start_bd->vlan_or_ethertype = cpu_to_le16(pkt_prod);
3708 nbd = 2; /* start_bd + pbd + frags (updated when pages are mapped) */
3710 /* turn on parsing and get a BD */
3711 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3713 if (xmit_type & XMIT_CSUM)
3714 bnx2x_set_sbd_csum(bp, skb, tx_start_bd, xmit_type);
3716 if (!CHIP_IS_E1x(bp)) {
3717 pbd_e2 = &txdata->tx_desc_ring[bd_prod].parse_bd_e2;
3718 memset(pbd_e2, 0, sizeof(struct eth_tx_parse_bd_e2));
3720 if (xmit_type & XMIT_CSUM_ENC) {
3721 u16 global_data = 0;
3723 /* Set PBD in enc checksum offload case */
3724 hlen = bnx2x_set_pbd_csum_enc(bp, skb,
3725 &pbd_e2_parsing_data,
3728 /* turn on 2nd parsing and get a BD */
3729 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3731 pbd2 = &txdata->tx_desc_ring[bd_prod].parse_2nd_bd;
3733 memset(pbd2, 0, sizeof(*pbd2));
3735 pbd_e2->data.tunnel_data.ip_hdr_start_inner_w =
3736 (skb_inner_network_header(skb) -
3739 if (xmit_type & XMIT_GSO_ENC)
3740 bnx2x_update_pbds_gso_enc(skb, pbd_e2, pbd2,
3744 pbd2->global_data = cpu_to_le16(global_data);
3746 /* add addition parse BD indication to start BD */
3747 SET_FLAG(tx_start_bd->general_data,
3748 ETH_TX_START_BD_PARSE_NBDS, 1);
3749 /* set encapsulation flag in start BD */
3750 SET_FLAG(tx_start_bd->general_data,
3751 ETH_TX_START_BD_TUNNEL_EXIST, 1);
3753 } else if (xmit_type & XMIT_CSUM) {
3754 /* Set PBD in checksum offload case w/o encapsulation */
3755 hlen = bnx2x_set_pbd_csum_e2(bp, skb,
3756 &pbd_e2_parsing_data,
3760 /* Add the macs to the parsing BD this is a vf */
3762 /* override GRE parameters in BD */
3763 bnx2x_set_fw_mac_addr(&pbd_e2->data.mac_addr.src_hi,
3764 &pbd_e2->data.mac_addr.src_mid,
3765 &pbd_e2->data.mac_addr.src_lo,
3768 bnx2x_set_fw_mac_addr(&pbd_e2->data.mac_addr.dst_hi,
3769 &pbd_e2->data.mac_addr.dst_mid,
3770 &pbd_e2->data.mac_addr.dst_lo,
3774 SET_FLAG(pbd_e2_parsing_data,
3775 ETH_TX_PARSE_BD_E2_ETH_ADDR_TYPE, mac_type);
3777 u16 global_data = 0;
3778 pbd_e1x = &txdata->tx_desc_ring[bd_prod].parse_bd_e1x;
3779 memset(pbd_e1x, 0, sizeof(struct eth_tx_parse_bd_e1x));
3780 /* Set PBD in checksum offload case */
3781 if (xmit_type & XMIT_CSUM)
3782 hlen = bnx2x_set_pbd_csum(bp, skb, pbd_e1x, xmit_type);
3784 SET_FLAG(global_data,
3785 ETH_TX_PARSE_BD_E1X_ETH_ADDR_TYPE, mac_type);
3786 pbd_e1x->global_data |= cpu_to_le16(global_data);
3789 /* Setup the data pointer of the first BD of the packet */
3790 tx_start_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
3791 tx_start_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
3792 tx_start_bd->nbytes = cpu_to_le16(skb_headlen(skb));
3793 pkt_size = tx_start_bd->nbytes;
3795 DP(NETIF_MSG_TX_QUEUED,
3796 "first bd @%p addr (%x:%x) nbytes %d flags %x vlan %x\n",
3797 tx_start_bd, tx_start_bd->addr_hi, tx_start_bd->addr_lo,
3798 le16_to_cpu(tx_start_bd->nbytes),
3799 tx_start_bd->bd_flags.as_bitfield,
3800 le16_to_cpu(tx_start_bd->vlan_or_ethertype));
3802 if (xmit_type & XMIT_GSO) {
3804 DP(NETIF_MSG_TX_QUEUED,
3805 "TSO packet len %d hlen %d total len %d tso size %d\n",
3806 skb->len, hlen, skb_headlen(skb),
3807 skb_shinfo(skb)->gso_size);
3809 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_SW_LSO;
3811 if (unlikely(skb_headlen(skb) > hlen)) {
3813 bd_prod = bnx2x_tx_split(bp, txdata, tx_buf,
3817 if (!CHIP_IS_E1x(bp))
3818 bnx2x_set_pbd_gso_e2(skb, &pbd_e2_parsing_data,
3821 bnx2x_set_pbd_gso(skb, pbd_e1x, tx_start_bd,
3825 /* Set the PBD's parsing_data field if not zero
3826 * (for the chips newer than 57711).
3828 if (pbd_e2_parsing_data)
3829 pbd_e2->parsing_data = cpu_to_le32(pbd_e2_parsing_data);
3831 tx_data_bd = (struct eth_tx_bd *)tx_start_bd;
3833 /* Handle fragmented skb */
3834 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3835 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3837 mapping = skb_frag_dma_map(&bp->pdev->dev, frag, 0,
3838 skb_frag_size(frag), DMA_TO_DEVICE);
3839 if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
3840 unsigned int pkts_compl = 0, bytes_compl = 0;
3842 DP(NETIF_MSG_TX_QUEUED,
3843 "Unable to map page - dropping packet...\n");
3845 /* we need unmap all buffers already mapped
3847 * first_bd->nbd need to be properly updated
3848 * before call to bnx2x_free_tx_pkt
3850 first_bd->nbd = cpu_to_le16(nbd);
3851 bnx2x_free_tx_pkt(bp, txdata,
3852 TX_BD(txdata->tx_pkt_prod),
3853 &pkts_compl, &bytes_compl);
3854 return NETDEV_TX_OK;
3857 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3858 tx_data_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
3859 if (total_pkt_bd == NULL)
3860 total_pkt_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
3862 tx_data_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
3863 tx_data_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
3864 tx_data_bd->nbytes = cpu_to_le16(skb_frag_size(frag));
3865 le16_add_cpu(&pkt_size, skb_frag_size(frag));
3868 DP(NETIF_MSG_TX_QUEUED,
3869 "frag %d bd @%p addr (%x:%x) nbytes %d\n",
3870 i, tx_data_bd, tx_data_bd->addr_hi, tx_data_bd->addr_lo,
3871 le16_to_cpu(tx_data_bd->nbytes));
3874 DP(NETIF_MSG_TX_QUEUED, "last bd @%p\n", tx_data_bd);
3876 /* update with actual num BDs */
3877 first_bd->nbd = cpu_to_le16(nbd);
3879 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3881 /* now send a tx doorbell, counting the next BD
3882 * if the packet contains or ends with it
3884 if (TX_BD_POFF(bd_prod) < nbd)
3887 /* total_pkt_bytes should be set on the first data BD if
3888 * it's not an LSO packet and there is more than one
3889 * data BD. In this case pkt_size is limited by an MTU value.
3890 * However we prefer to set it for an LSO packet (while we don't
3891 * have to) in order to save some CPU cycles in a none-LSO
3892 * case, when we much more care about them.
3894 if (total_pkt_bd != NULL)
3895 total_pkt_bd->total_pkt_bytes = pkt_size;
3898 DP(NETIF_MSG_TX_QUEUED,
3899 "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",
3900 pbd_e1x, pbd_e1x->global_data, pbd_e1x->ip_hlen_w,
3901 pbd_e1x->ip_id, pbd_e1x->lso_mss, pbd_e1x->tcp_flags,
3902 pbd_e1x->tcp_pseudo_csum, pbd_e1x->tcp_send_seq,
3903 le16_to_cpu(pbd_e1x->total_hlen_w));
3905 DP(NETIF_MSG_TX_QUEUED,
3906 "PBD (E2) @%p dst %x %x %x src %x %x %x parsing_data %x\n",
3908 pbd_e2->data.mac_addr.dst_hi,
3909 pbd_e2->data.mac_addr.dst_mid,
3910 pbd_e2->data.mac_addr.dst_lo,
3911 pbd_e2->data.mac_addr.src_hi,
3912 pbd_e2->data.mac_addr.src_mid,
3913 pbd_e2->data.mac_addr.src_lo,
3914 pbd_e2->parsing_data);
3915 DP(NETIF_MSG_TX_QUEUED, "doorbell: nbd %d bd %u\n", nbd, bd_prod);
3917 netdev_tx_sent_queue(txq, skb->len);
3919 skb_tx_timestamp(skb);
3921 txdata->tx_pkt_prod++;
3923 * Make sure that the BD data is updated before updating the producer
3924 * since FW might read the BD right after the producer is updated.
3925 * This is only applicable for weak-ordered memory model archs such
3926 * as IA-64. The following barrier is also mandatory since FW will
3927 * assumes packets must have BDs.
3931 txdata->tx_db.data.prod += nbd;
3934 DOORBELL(bp, txdata->cid, txdata->tx_db.raw);
3938 txdata->tx_bd_prod += nbd;
3940 if (unlikely(bnx2x_tx_avail(bp, txdata) < MAX_DESC_PER_TX_PKT)) {
3941 netif_tx_stop_queue(txq);
3943 /* paired memory barrier is in bnx2x_tx_int(), we have to keep
3944 * ordering of set_bit() in netif_tx_stop_queue() and read of
3948 bnx2x_fp_qstats(bp, txdata->parent_fp)->driver_xoff++;
3949 if (bnx2x_tx_avail(bp, txdata) >= MAX_DESC_PER_TX_PKT)
3950 netif_tx_wake_queue(txq);
3954 return NETDEV_TX_OK;
3958 * bnx2x_setup_tc - routine to configure net_device for multi tc
3960 * @netdev: net device to configure
3961 * @tc: number of traffic classes to enable
3963 * callback connected to the ndo_setup_tc function pointer
3965 int bnx2x_setup_tc(struct net_device *dev, u8 num_tc)
3967 int cos, prio, count, offset;
3968 struct bnx2x *bp = netdev_priv(dev);
3970 /* setup tc must be called under rtnl lock */
3973 /* no traffic classes requested. aborting */
3975 netdev_reset_tc(dev);
3979 /* requested to support too many traffic classes */
3980 if (num_tc > bp->max_cos) {
3981 BNX2X_ERR("support for too many traffic classes requested: %d. max supported is %d\n",
3982 num_tc, bp->max_cos);
3986 /* declare amount of supported traffic classes */
3987 if (netdev_set_num_tc(dev, num_tc)) {
3988 BNX2X_ERR("failed to declare %d traffic classes\n", num_tc);
3992 /* configure priority to traffic class mapping */
3993 for (prio = 0; prio < BNX2X_MAX_PRIORITY; prio++) {
3994 netdev_set_prio_tc_map(dev, prio, bp->prio_to_cos[prio]);
3995 DP(BNX2X_MSG_SP | NETIF_MSG_IFUP,
3996 "mapping priority %d to tc %d\n",
3997 prio, bp->prio_to_cos[prio]);
4001 /* Use this configuration to diffrentiate tc0 from other COSes
4002 This can be used for ets or pfc, and save the effort of setting
4003 up a multio class queue disc or negotiating DCBX with a switch
4004 netdev_set_prio_tc_map(dev, 0, 0);
4005 DP(BNX2X_MSG_SP, "mapping priority %d to tc %d\n", 0, 0);
4006 for (prio = 1; prio < 16; prio++) {
4007 netdev_set_prio_tc_map(dev, prio, 1);
4008 DP(BNX2X_MSG_SP, "mapping priority %d to tc %d\n", prio, 1);
4011 /* configure traffic class to transmission queue mapping */
4012 for (cos = 0; cos < bp->max_cos; cos++) {
4013 count = BNX2X_NUM_ETH_QUEUES(bp);
4014 offset = cos * BNX2X_NUM_NON_CNIC_QUEUES(bp);
4015 netdev_set_tc_queue(dev, cos, count, offset);
4016 DP(BNX2X_MSG_SP | NETIF_MSG_IFUP,
4017 "mapping tc %d to offset %d count %d\n",
4018 cos, offset, count);
4024 /* called with rtnl_lock */
4025 int bnx2x_change_mac_addr(struct net_device *dev, void *p)
4027 struct sockaddr *addr = p;
4028 struct bnx2x *bp = netdev_priv(dev);
4031 if (!bnx2x_is_valid_ether_addr(bp, addr->sa_data)) {
4032 BNX2X_ERR("Requested MAC address is not valid\n");
4036 if ((IS_MF_STORAGE_SD(bp) || IS_MF_FCOE_AFEX(bp)) &&
4037 !is_zero_ether_addr(addr->sa_data)) {
4038 BNX2X_ERR("Can't configure non-zero address on iSCSI or FCoE functions in MF-SD mode\n");
4042 if (netif_running(dev)) {
4043 rc = bnx2x_set_eth_mac(bp, false);
4048 memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
4050 if (netif_running(dev))
4051 rc = bnx2x_set_eth_mac(bp, true);
4056 static void bnx2x_free_fp_mem_at(struct bnx2x *bp, int fp_index)
4058 union host_hc_status_block *sb = &bnx2x_fp(bp, fp_index, status_blk);
4059 struct bnx2x_fastpath *fp = &bp->fp[fp_index];
4064 if (IS_FCOE_IDX(fp_index)) {
4065 memset(sb, 0, sizeof(union host_hc_status_block));
4066 fp->status_blk_mapping = 0;
4069 if (!CHIP_IS_E1x(bp))
4070 BNX2X_PCI_FREE(sb->e2_sb,
4071 bnx2x_fp(bp, fp_index,
4072 status_blk_mapping),
4073 sizeof(struct host_hc_status_block_e2));
4075 BNX2X_PCI_FREE(sb->e1x_sb,
4076 bnx2x_fp(bp, fp_index,
4077 status_blk_mapping),
4078 sizeof(struct host_hc_status_block_e1x));
4082 if (!skip_rx_queue(bp, fp_index)) {
4083 bnx2x_free_rx_bds(fp);
4085 /* fastpath rx rings: rx_buf rx_desc rx_comp */
4086 BNX2X_FREE(bnx2x_fp(bp, fp_index, rx_buf_ring));
4087 BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_desc_ring),
4088 bnx2x_fp(bp, fp_index, rx_desc_mapping),
4089 sizeof(struct eth_rx_bd) * NUM_RX_BD);
4091 BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_comp_ring),
4092 bnx2x_fp(bp, fp_index, rx_comp_mapping),
4093 sizeof(struct eth_fast_path_rx_cqe) *
4097 BNX2X_FREE(bnx2x_fp(bp, fp_index, rx_page_ring));
4098 BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_sge_ring),
4099 bnx2x_fp(bp, fp_index, rx_sge_mapping),
4100 BCM_PAGE_SIZE * NUM_RX_SGE_PAGES);
4104 if (!skip_tx_queue(bp, fp_index)) {
4105 /* fastpath tx rings: tx_buf tx_desc */
4106 for_each_cos_in_tx_queue(fp, cos) {
4107 struct bnx2x_fp_txdata *txdata = fp->txdata_ptr[cos];
4109 DP(NETIF_MSG_IFDOWN,
4110 "freeing tx memory of fp %d cos %d cid %d\n",
4111 fp_index, cos, txdata->cid);
4113 BNX2X_FREE(txdata->tx_buf_ring);
4114 BNX2X_PCI_FREE(txdata->tx_desc_ring,
4115 txdata->tx_desc_mapping,
4116 sizeof(union eth_tx_bd_types) * NUM_TX_BD);
4119 /* end of fastpath */
4122 void bnx2x_free_fp_mem_cnic(struct bnx2x *bp)
4125 for_each_cnic_queue(bp, i)
4126 bnx2x_free_fp_mem_at(bp, i);
4129 void bnx2x_free_fp_mem(struct bnx2x *bp)
4132 for_each_eth_queue(bp, i)
4133 bnx2x_free_fp_mem_at(bp, i);
4136 static void set_sb_shortcuts(struct bnx2x *bp, int index)
4138 union host_hc_status_block status_blk = bnx2x_fp(bp, index, status_blk);
4139 if (!CHIP_IS_E1x(bp)) {
4140 bnx2x_fp(bp, index, sb_index_values) =
4141 (__le16 *)status_blk.e2_sb->sb.index_values;
4142 bnx2x_fp(bp, index, sb_running_index) =
4143 (__le16 *)status_blk.e2_sb->sb.running_index;
4145 bnx2x_fp(bp, index, sb_index_values) =
4146 (__le16 *)status_blk.e1x_sb->sb.index_values;
4147 bnx2x_fp(bp, index, sb_running_index) =
4148 (__le16 *)status_blk.e1x_sb->sb.running_index;
4152 /* Returns the number of actually allocated BDs */
4153 static int bnx2x_alloc_rx_bds(struct bnx2x_fastpath *fp,
4156 struct bnx2x *bp = fp->bp;
4157 u16 ring_prod, cqe_ring_prod;
4158 int i, failure_cnt = 0;
4160 fp->rx_comp_cons = 0;
4161 cqe_ring_prod = ring_prod = 0;
4163 /* This routine is called only during fo init so
4164 * fp->eth_q_stats.rx_skb_alloc_failed = 0
4166 for (i = 0; i < rx_ring_size; i++) {
4167 if (bnx2x_alloc_rx_data(bp, fp, ring_prod) < 0) {
4171 ring_prod = NEXT_RX_IDX(ring_prod);
4172 cqe_ring_prod = NEXT_RCQ_IDX(cqe_ring_prod);
4173 WARN_ON(ring_prod <= (i - failure_cnt));
4177 BNX2X_ERR("was only able to allocate %d rx skbs on queue[%d]\n",
4178 i - failure_cnt, fp->index);
4180 fp->rx_bd_prod = ring_prod;
4181 /* Limit the CQE producer by the CQE ring size */
4182 fp->rx_comp_prod = min_t(u16, NUM_RCQ_RINGS*RCQ_DESC_CNT,
4184 fp->rx_pkt = fp->rx_calls = 0;
4186 bnx2x_fp_stats(bp, fp)->eth_q_stats.rx_skb_alloc_failed += failure_cnt;
4188 return i - failure_cnt;
4191 static void bnx2x_set_next_page_rx_cq(struct bnx2x_fastpath *fp)
4195 for (i = 1; i <= NUM_RCQ_RINGS; i++) {
4196 struct eth_rx_cqe_next_page *nextpg;
4198 nextpg = (struct eth_rx_cqe_next_page *)
4199 &fp->rx_comp_ring[RCQ_DESC_CNT * i - 1];
4201 cpu_to_le32(U64_HI(fp->rx_comp_mapping +
4202 BCM_PAGE_SIZE*(i % NUM_RCQ_RINGS)));
4204 cpu_to_le32(U64_LO(fp->rx_comp_mapping +
4205 BCM_PAGE_SIZE*(i % NUM_RCQ_RINGS)));
4209 static int bnx2x_alloc_fp_mem_at(struct bnx2x *bp, int index)
4211 union host_hc_status_block *sb;
4212 struct bnx2x_fastpath *fp = &bp->fp[index];
4215 int rx_ring_size = 0;
4217 if (!bp->rx_ring_size &&
4218 (IS_MF_STORAGE_SD(bp) || IS_MF_FCOE_AFEX(bp))) {
4219 rx_ring_size = MIN_RX_SIZE_NONTPA;
4220 bp->rx_ring_size = rx_ring_size;
4221 } else if (!bp->rx_ring_size) {
4222 rx_ring_size = MAX_RX_AVAIL/BNX2X_NUM_RX_QUEUES(bp);
4224 if (CHIP_IS_E3(bp)) {
4225 u32 cfg = SHMEM_RD(bp,
4226 dev_info.port_hw_config[BP_PORT(bp)].
4229 /* Decrease ring size for 1G functions */
4230 if ((cfg & PORT_HW_CFG_NET_SERDES_IF_MASK) ==
4231 PORT_HW_CFG_NET_SERDES_IF_SGMII)
4235 /* allocate at least number of buffers required by FW */
4236 rx_ring_size = max_t(int, bp->disable_tpa ? MIN_RX_SIZE_NONTPA :
4237 MIN_RX_SIZE_TPA, rx_ring_size);
4239 bp->rx_ring_size = rx_ring_size;
4240 } else /* if rx_ring_size specified - use it */
4241 rx_ring_size = bp->rx_ring_size;
4243 DP(BNX2X_MSG_SP, "calculated rx_ring_size %d\n", rx_ring_size);
4246 sb = &bnx2x_fp(bp, index, status_blk);
4248 if (!IS_FCOE_IDX(index)) {
4250 if (!CHIP_IS_E1x(bp))
4251 BNX2X_PCI_ALLOC(sb->e2_sb,
4252 &bnx2x_fp(bp, index, status_blk_mapping),
4253 sizeof(struct host_hc_status_block_e2));
4255 BNX2X_PCI_ALLOC(sb->e1x_sb,
4256 &bnx2x_fp(bp, index, status_blk_mapping),
4257 sizeof(struct host_hc_status_block_e1x));
4260 /* FCoE Queue uses Default SB and doesn't ACK the SB, thus no need to
4261 * set shortcuts for it.
4263 if (!IS_FCOE_IDX(index))
4264 set_sb_shortcuts(bp, index);
4267 if (!skip_tx_queue(bp, index)) {
4268 /* fastpath tx rings: tx_buf tx_desc */
4269 for_each_cos_in_tx_queue(fp, cos) {
4270 struct bnx2x_fp_txdata *txdata = fp->txdata_ptr[cos];
4273 "allocating tx memory of fp %d cos %d\n",
4276 BNX2X_ALLOC(txdata->tx_buf_ring,
4277 sizeof(struct sw_tx_bd) * NUM_TX_BD);
4278 BNX2X_PCI_ALLOC(txdata->tx_desc_ring,
4279 &txdata->tx_desc_mapping,
4280 sizeof(union eth_tx_bd_types) * NUM_TX_BD);
4285 if (!skip_rx_queue(bp, index)) {
4286 /* fastpath rx rings: rx_buf rx_desc rx_comp */
4287 BNX2X_ALLOC(bnx2x_fp(bp, index, rx_buf_ring),
4288 sizeof(struct sw_rx_bd) * NUM_RX_BD);
4289 BNX2X_PCI_ALLOC(bnx2x_fp(bp, index, rx_desc_ring),
4290 &bnx2x_fp(bp, index, rx_desc_mapping),
4291 sizeof(struct eth_rx_bd) * NUM_RX_BD);
4293 BNX2X_PCI_ALLOC(bnx2x_fp(bp, index, rx_comp_ring),
4294 &bnx2x_fp(bp, index, rx_comp_mapping),
4295 sizeof(struct eth_fast_path_rx_cqe) *
4299 BNX2X_ALLOC(bnx2x_fp(bp, index, rx_page_ring),
4300 sizeof(struct sw_rx_page) * NUM_RX_SGE);
4301 BNX2X_PCI_ALLOC(bnx2x_fp(bp, index, rx_sge_ring),
4302 &bnx2x_fp(bp, index, rx_sge_mapping),
4303 BCM_PAGE_SIZE * NUM_RX_SGE_PAGES);
4305 bnx2x_set_next_page_rx_bd(fp);
4308 bnx2x_set_next_page_rx_cq(fp);
4311 ring_size = bnx2x_alloc_rx_bds(fp, rx_ring_size);
4312 if (ring_size < rx_ring_size)
4318 /* handles low memory cases */
4320 BNX2X_ERR("Unable to allocate full memory for queue %d (size %d)\n",
4322 /* FW will drop all packets if queue is not big enough,
4323 * In these cases we disable the queue
4324 * Min size is different for OOO, TPA and non-TPA queues
4326 if (ring_size < (fp->disable_tpa ?
4327 MIN_RX_SIZE_NONTPA : MIN_RX_SIZE_TPA)) {
4328 /* release memory allocated for this queue */
4329 bnx2x_free_fp_mem_at(bp, index);
4335 int bnx2x_alloc_fp_mem_cnic(struct bnx2x *bp)
4339 if (bnx2x_alloc_fp_mem_at(bp, FCOE_IDX(bp)))
4340 /* we will fail load process instead of mark
4348 int bnx2x_alloc_fp_mem(struct bnx2x *bp)
4352 /* 1. Allocate FP for leading - fatal if error
4353 * 2. Allocate RSS - fix number of queues if error
4357 if (bnx2x_alloc_fp_mem_at(bp, 0))
4361 for_each_nondefault_eth_queue(bp, i)
4362 if (bnx2x_alloc_fp_mem_at(bp, i))
4365 /* handle memory failures */
4366 if (i != BNX2X_NUM_ETH_QUEUES(bp)) {
4367 int delta = BNX2X_NUM_ETH_QUEUES(bp) - i;
4370 bnx2x_shrink_eth_fp(bp, delta);
4371 if (CNIC_SUPPORT(bp))
4372 /* move non eth FPs next to last eth FP
4373 * must be done in that order
4374 * FCOE_IDX < FWD_IDX < OOO_IDX
4377 /* move FCoE fp even NO_FCOE_FLAG is on */
4378 bnx2x_move_fp(bp, FCOE_IDX(bp), FCOE_IDX(bp) - delta);
4379 bp->num_ethernet_queues -= delta;
4380 bp->num_queues = bp->num_ethernet_queues +
4381 bp->num_cnic_queues;
4382 BNX2X_ERR("Adjusted num of queues from %d to %d\n",
4383 bp->num_queues + delta, bp->num_queues);
4389 void bnx2x_free_mem_bp(struct bnx2x *bp)
4393 for (i = 0; i < bp->fp_array_size; i++)
4394 kfree(bp->fp[i].tpa_info);
4397 kfree(bp->fp_stats);
4398 kfree(bp->bnx2x_txq);
4399 kfree(bp->msix_table);
4403 int bnx2x_alloc_mem_bp(struct bnx2x *bp)
4405 struct bnx2x_fastpath *fp;
4406 struct msix_entry *tbl;
4407 struct bnx2x_ilt *ilt;
4408 int msix_table_size = 0;
4409 int fp_array_size, txq_array_size;
4413 * The biggest MSI-X table we might need is as a maximum number of fast
4414 * path IGU SBs plus default SB (for PF only).
4416 msix_table_size = bp->igu_sb_cnt;
4419 BNX2X_DEV_INFO("msix_table_size %d\n", msix_table_size);
4421 /* fp array: RSS plus CNIC related L2 queues */
4422 fp_array_size = BNX2X_MAX_RSS_COUNT(bp) + CNIC_SUPPORT(bp);
4423 bp->fp_array_size = fp_array_size;
4424 BNX2X_DEV_INFO("fp_array_size %d\n", bp->fp_array_size);
4426 fp = kcalloc(bp->fp_array_size, sizeof(*fp), GFP_KERNEL);
4429 for (i = 0; i < bp->fp_array_size; i++) {
4431 kcalloc(ETH_MAX_AGGREGATION_QUEUES_E1H_E2,
4432 sizeof(struct bnx2x_agg_info), GFP_KERNEL);
4433 if (!(fp[i].tpa_info))
4439 /* allocate sp objs */
4440 bp->sp_objs = kcalloc(bp->fp_array_size, sizeof(struct bnx2x_sp_objs),
4445 /* allocate fp_stats */
4446 bp->fp_stats = kcalloc(bp->fp_array_size, sizeof(struct bnx2x_fp_stats),
4451 /* Allocate memory for the transmission queues array */
4453 BNX2X_MAX_RSS_COUNT(bp) * BNX2X_MULTI_TX_COS + CNIC_SUPPORT(bp);
4454 BNX2X_DEV_INFO("txq_array_size %d", txq_array_size);
4456 bp->bnx2x_txq = kcalloc(txq_array_size, sizeof(struct bnx2x_fp_txdata),
4462 tbl = kcalloc(msix_table_size, sizeof(*tbl), GFP_KERNEL);
4465 bp->msix_table = tbl;
4468 ilt = kzalloc(sizeof(*ilt), GFP_KERNEL);
4475 bnx2x_free_mem_bp(bp);
4480 int bnx2x_reload_if_running(struct net_device *dev)
4482 struct bnx2x *bp = netdev_priv(dev);
4484 if (unlikely(!netif_running(dev)))
4487 bnx2x_nic_unload(bp, UNLOAD_NORMAL, true);
4488 return bnx2x_nic_load(bp, LOAD_NORMAL);
4491 int bnx2x_get_cur_phy_idx(struct bnx2x *bp)
4493 u32 sel_phy_idx = 0;
4494 if (bp->link_params.num_phys <= 1)
4497 if (bp->link_vars.link_up) {
4498 sel_phy_idx = EXT_PHY1;
4499 /* In case link is SERDES, check if the EXT_PHY2 is the one */
4500 if ((bp->link_vars.link_status & LINK_STATUS_SERDES_LINK) &&
4501 (bp->link_params.phy[EXT_PHY2].supported & SUPPORTED_FIBRE))
4502 sel_phy_idx = EXT_PHY2;
4505 switch (bnx2x_phy_selection(&bp->link_params)) {
4506 case PORT_HW_CFG_PHY_SELECTION_HARDWARE_DEFAULT:
4507 case PORT_HW_CFG_PHY_SELECTION_FIRST_PHY:
4508 case PORT_HW_CFG_PHY_SELECTION_FIRST_PHY_PRIORITY:
4509 sel_phy_idx = EXT_PHY1;
4511 case PORT_HW_CFG_PHY_SELECTION_SECOND_PHY:
4512 case PORT_HW_CFG_PHY_SELECTION_SECOND_PHY_PRIORITY:
4513 sel_phy_idx = EXT_PHY2;
4521 int bnx2x_get_link_cfg_idx(struct bnx2x *bp)
4523 u32 sel_phy_idx = bnx2x_get_cur_phy_idx(bp);
4525 * The selected activated PHY is always after swapping (in case PHY
4526 * swapping is enabled). So when swapping is enabled, we need to reverse
4530 if (bp->link_params.multi_phy_config &
4531 PORT_HW_CFG_PHY_SWAPPED_ENABLED) {
4532 if (sel_phy_idx == EXT_PHY1)
4533 sel_phy_idx = EXT_PHY2;
4534 else if (sel_phy_idx == EXT_PHY2)
4535 sel_phy_idx = EXT_PHY1;
4537 return LINK_CONFIG_IDX(sel_phy_idx);
4540 #ifdef NETDEV_FCOE_WWNN
4541 int bnx2x_fcoe_get_wwn(struct net_device *dev, u64 *wwn, int type)
4543 struct bnx2x *bp = netdev_priv(dev);
4544 struct cnic_eth_dev *cp = &bp->cnic_eth_dev;
4547 case NETDEV_FCOE_WWNN:
4548 *wwn = HILO_U64(cp->fcoe_wwn_node_name_hi,
4549 cp->fcoe_wwn_node_name_lo);
4551 case NETDEV_FCOE_WWPN:
4552 *wwn = HILO_U64(cp->fcoe_wwn_port_name_hi,
4553 cp->fcoe_wwn_port_name_lo);
4556 BNX2X_ERR("Wrong WWN type requested - %d\n", type);
4564 /* called with rtnl_lock */
4565 int bnx2x_change_mtu(struct net_device *dev, int new_mtu)
4567 struct bnx2x *bp = netdev_priv(dev);
4569 if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
4570 BNX2X_ERR("Can't perform change MTU during parity recovery\n");
4574 if ((new_mtu > ETH_MAX_JUMBO_PACKET_SIZE) ||
4575 ((new_mtu + ETH_HLEN) < ETH_MIN_PACKET_SIZE)) {
4576 BNX2X_ERR("Can't support requested MTU size\n");
4580 /* This does not race with packet allocation
4581 * because the actual alloc size is
4582 * only updated as part of load
4586 return bnx2x_reload_if_running(dev);
4589 netdev_features_t bnx2x_fix_features(struct net_device *dev,
4590 netdev_features_t features)
4592 struct bnx2x *bp = netdev_priv(dev);
4594 /* TPA requires Rx CSUM offloading */
4595 if (!(features & NETIF_F_RXCSUM) || bp->disable_tpa) {
4596 features &= ~NETIF_F_LRO;
4597 features &= ~NETIF_F_GRO;
4603 int bnx2x_set_features(struct net_device *dev, netdev_features_t features)
4605 struct bnx2x *bp = netdev_priv(dev);
4606 u32 flags = bp->flags;
4607 bool bnx2x_reload = false;
4609 if (features & NETIF_F_LRO)
4610 flags |= TPA_ENABLE_FLAG;
4612 flags &= ~TPA_ENABLE_FLAG;
4614 if (features & NETIF_F_GRO)
4615 flags |= GRO_ENABLE_FLAG;
4617 flags &= ~GRO_ENABLE_FLAG;
4619 if (features & NETIF_F_LOOPBACK) {
4620 if (bp->link_params.loopback_mode != LOOPBACK_BMAC) {
4621 bp->link_params.loopback_mode = LOOPBACK_BMAC;
4622 bnx2x_reload = true;
4625 if (bp->link_params.loopback_mode != LOOPBACK_NONE) {
4626 bp->link_params.loopback_mode = LOOPBACK_NONE;
4627 bnx2x_reload = true;
4631 if (flags ^ bp->flags) {
4633 bnx2x_reload = true;
4637 if (bp->recovery_state == BNX2X_RECOVERY_DONE)
4638 return bnx2x_reload_if_running(dev);
4639 /* else: bnx2x_nic_load() will be called at end of recovery */
4645 void bnx2x_tx_timeout(struct net_device *dev)
4647 struct bnx2x *bp = netdev_priv(dev);
4649 #ifdef BNX2X_STOP_ON_ERROR
4654 smp_mb__before_clear_bit();
4655 set_bit(BNX2X_SP_RTNL_TX_TIMEOUT, &bp->sp_rtnl_state);
4656 smp_mb__after_clear_bit();
4658 /* This allows the netif to be shutdown gracefully before resetting */
4659 schedule_delayed_work(&bp->sp_rtnl_task, 0);
4662 int bnx2x_suspend(struct pci_dev *pdev, pm_message_t state)
4664 struct net_device *dev = pci_get_drvdata(pdev);
4668 dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
4671 bp = netdev_priv(dev);
4675 pci_save_state(pdev);
4677 if (!netif_running(dev)) {
4682 netif_device_detach(dev);
4684 bnx2x_nic_unload(bp, UNLOAD_CLOSE, false);
4686 bnx2x_set_power_state(bp, pci_choose_state(pdev, state));
4693 int bnx2x_resume(struct pci_dev *pdev)
4695 struct net_device *dev = pci_get_drvdata(pdev);
4700 dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
4703 bp = netdev_priv(dev);
4705 if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
4706 BNX2X_ERR("Handling parity error recovery. Try again later\n");
4712 pci_restore_state(pdev);
4714 if (!netif_running(dev)) {
4719 bnx2x_set_power_state(bp, PCI_D0);
4720 netif_device_attach(dev);
4722 rc = bnx2x_nic_load(bp, LOAD_OPEN);
4730 void bnx2x_set_ctx_validation(struct bnx2x *bp, struct eth_context *cxt,
4733 /* ustorm cxt validation */
4734 cxt->ustorm_ag_context.cdu_usage =
4735 CDU_RSRVD_VALUE_TYPE_A(HW_CID(bp, cid),
4736 CDU_REGION_NUMBER_UCM_AG, ETH_CONNECTION_TYPE);
4737 /* xcontext validation */
4738 cxt->xstorm_ag_context.cdu_reserved =
4739 CDU_RSRVD_VALUE_TYPE_A(HW_CID(bp, cid),
4740 CDU_REGION_NUMBER_XCM_AG, ETH_CONNECTION_TYPE);
4743 static void storm_memset_hc_timeout(struct bnx2x *bp, u8 port,
4744 u8 fw_sb_id, u8 sb_index,
4748 u32 addr = BAR_CSTRORM_INTMEM +
4749 CSTORM_STATUS_BLOCK_DATA_TIMEOUT_OFFSET(fw_sb_id, sb_index);
4750 REG_WR8(bp, addr, ticks);
4752 "port %x fw_sb_id %d sb_index %d ticks %d\n",
4753 port, fw_sb_id, sb_index, ticks);
4756 static void storm_memset_hc_disable(struct bnx2x *bp, u8 port,
4757 u16 fw_sb_id, u8 sb_index,
4760 u32 enable_flag = disable ? 0 : (1 << HC_INDEX_DATA_HC_ENABLED_SHIFT);
4761 u32 addr = BAR_CSTRORM_INTMEM +
4762 CSTORM_STATUS_BLOCK_DATA_FLAGS_OFFSET(fw_sb_id, sb_index);
4763 u8 flags = REG_RD8(bp, addr);
4765 flags &= ~HC_INDEX_DATA_HC_ENABLED;
4766 flags |= enable_flag;
4767 REG_WR8(bp, addr, flags);
4769 "port %x fw_sb_id %d sb_index %d disable %d\n",
4770 port, fw_sb_id, sb_index, disable);
4773 void bnx2x_update_coalesce_sb_index(struct bnx2x *bp, u8 fw_sb_id,
4774 u8 sb_index, u8 disable, u16 usec)
4776 int port = BP_PORT(bp);
4777 u8 ticks = usec / BNX2X_BTR;
4779 storm_memset_hc_timeout(bp, port, fw_sb_id, sb_index, ticks);
4781 disable = disable ? 1 : (usec ? 0 : 1);
4782 storm_memset_hc_disable(bp, port, fw_sb_id, sb_index, disable);