1 /* bnx2x_cmn.c: Broadcom Everest network driver.
3 * Copyright (c) 2007-2010 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 #include <linux/etherdevice.h>
19 #include <linux/if_vlan.h>
22 #include <net/ip6_checksum.h>
23 #include <linux/firmware.h>
24 #include "bnx2x_cmn.h"
26 #include "bnx2x_init.h"
28 static int bnx2x_setup_irqs(struct bnx2x *bp);
30 /* free skb in the packet ring at pos idx
31 * return idx of last bd freed
33 static u16 bnx2x_free_tx_pkt(struct bnx2x *bp, struct bnx2x_fastpath *fp,
36 struct sw_tx_bd *tx_buf = &fp->tx_buf_ring[idx];
37 struct eth_tx_start_bd *tx_start_bd;
38 struct eth_tx_bd *tx_data_bd;
39 struct sk_buff *skb = tx_buf->skb;
40 u16 bd_idx = TX_BD(tx_buf->first_bd), new_cons;
43 /* prefetch skb end pointer to speedup dev_kfree_skb() */
46 DP(BNX2X_MSG_OFF, "pkt_idx %d buff @(%p)->skb %p\n",
50 DP(BNX2X_MSG_OFF, "free bd_idx %d\n", bd_idx);
51 tx_start_bd = &fp->tx_desc_ring[bd_idx].start_bd;
52 dma_unmap_single(&bp->pdev->dev, BD_UNMAP_ADDR(tx_start_bd),
53 BD_UNMAP_LEN(tx_start_bd), DMA_TO_DEVICE);
55 nbd = le16_to_cpu(tx_start_bd->nbd) - 1;
56 #ifdef BNX2X_STOP_ON_ERROR
57 if ((nbd - 1) > (MAX_SKB_FRAGS + 2)) {
58 BNX2X_ERR("BAD nbd!\n");
62 new_cons = nbd + tx_buf->first_bd;
65 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
67 /* Skip a parse bd... */
69 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
71 /* ...and the TSO split header bd since they have no mapping */
72 if (tx_buf->flags & BNX2X_TSO_SPLIT_BD) {
74 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
80 DP(BNX2X_MSG_OFF, "free frag bd_idx %d\n", bd_idx);
81 tx_data_bd = &fp->tx_desc_ring[bd_idx].reg_bd;
82 dma_unmap_page(&bp->pdev->dev, BD_UNMAP_ADDR(tx_data_bd),
83 BD_UNMAP_LEN(tx_data_bd), DMA_TO_DEVICE);
85 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
97 int bnx2x_tx_int(struct bnx2x_fastpath *fp)
99 struct bnx2x *bp = fp->bp;
100 struct netdev_queue *txq;
101 u16 hw_cons, sw_cons, bd_cons = fp->tx_bd_cons;
103 #ifdef BNX2X_STOP_ON_ERROR
104 if (unlikely(bp->panic))
108 txq = netdev_get_tx_queue(bp->dev, fp->index);
109 hw_cons = le16_to_cpu(*fp->tx_cons_sb);
110 sw_cons = fp->tx_pkt_cons;
112 while (sw_cons != hw_cons) {
115 pkt_cons = TX_BD(sw_cons);
117 DP(NETIF_MSG_TX_DONE, "queue[%d]: hw_cons %u sw_cons %u "
119 fp->index, hw_cons, sw_cons, pkt_cons);
121 bd_cons = bnx2x_free_tx_pkt(bp, fp, pkt_cons);
125 fp->tx_pkt_cons = sw_cons;
126 fp->tx_bd_cons = bd_cons;
128 /* Need to make the tx_bd_cons update visible to start_xmit()
129 * before checking for netif_tx_queue_stopped(). Without the
130 * memory barrier, there is a small possibility that
131 * start_xmit() will miss it and cause the queue to be stopped
136 if (unlikely(netif_tx_queue_stopped(txq))) {
137 /* Taking tx_lock() is needed to prevent reenabling the queue
138 * while it's empty. This could have happen if rx_action() gets
139 * suspended in bnx2x_tx_int() after the condition before
140 * netif_tx_wake_queue(), while tx_action (bnx2x_start_xmit()):
142 * stops the queue->sees fresh tx_bd_cons->releases the queue->
143 * sends some packets consuming the whole queue again->
147 __netif_tx_lock(txq, smp_processor_id());
149 if ((netif_tx_queue_stopped(txq)) &&
150 (bp->state == BNX2X_STATE_OPEN) &&
151 (bnx2x_tx_avail(fp) >= MAX_SKB_FRAGS + 3))
152 netif_tx_wake_queue(txq);
154 __netif_tx_unlock(txq);
159 static inline void bnx2x_update_last_max_sge(struct bnx2x_fastpath *fp,
162 u16 last_max = fp->last_max_sge;
164 if (SUB_S16(idx, last_max) > 0)
165 fp->last_max_sge = idx;
168 static void bnx2x_update_sge_prod(struct bnx2x_fastpath *fp,
169 struct eth_fast_path_rx_cqe *fp_cqe)
171 struct bnx2x *bp = fp->bp;
172 u16 sge_len = SGE_PAGE_ALIGN(le16_to_cpu(fp_cqe->pkt_len) -
173 le16_to_cpu(fp_cqe->len_on_bd)) >>
175 u16 last_max, last_elem, first_elem;
182 /* First mark all used pages */
183 for (i = 0; i < sge_len; i++)
184 SGE_MASK_CLEAR_BIT(fp,
185 RX_SGE(le16_to_cpu(fp_cqe->sgl_or_raw_data.sgl[i])));
187 DP(NETIF_MSG_RX_STATUS, "fp_cqe->sgl[%d] = %d\n",
188 sge_len - 1, le16_to_cpu(fp_cqe->sgl_or_raw_data.sgl[sge_len - 1]));
190 /* Here we assume that the last SGE index is the biggest */
191 prefetch((void *)(fp->sge_mask));
192 bnx2x_update_last_max_sge(fp,
193 le16_to_cpu(fp_cqe->sgl_or_raw_data.sgl[sge_len - 1]));
195 last_max = RX_SGE(fp->last_max_sge);
196 last_elem = last_max >> RX_SGE_MASK_ELEM_SHIFT;
197 first_elem = RX_SGE(fp->rx_sge_prod) >> RX_SGE_MASK_ELEM_SHIFT;
199 /* If ring is not full */
200 if (last_elem + 1 != first_elem)
203 /* Now update the prod */
204 for (i = first_elem; i != last_elem; i = NEXT_SGE_MASK_ELEM(i)) {
205 if (likely(fp->sge_mask[i]))
208 fp->sge_mask[i] = RX_SGE_MASK_ELEM_ONE_MASK;
209 delta += RX_SGE_MASK_ELEM_SZ;
213 fp->rx_sge_prod += delta;
214 /* clear page-end entries */
215 bnx2x_clear_sge_mask_next_elems(fp);
218 DP(NETIF_MSG_RX_STATUS,
219 "fp->last_max_sge = %d fp->rx_sge_prod = %d\n",
220 fp->last_max_sge, fp->rx_sge_prod);
223 static void bnx2x_tpa_start(struct bnx2x_fastpath *fp, u16 queue,
224 struct sk_buff *skb, u16 cons, u16 prod)
226 struct bnx2x *bp = fp->bp;
227 struct sw_rx_bd *cons_rx_buf = &fp->rx_buf_ring[cons];
228 struct sw_rx_bd *prod_rx_buf = &fp->rx_buf_ring[prod];
229 struct eth_rx_bd *prod_bd = &fp->rx_desc_ring[prod];
232 /* move empty skb from pool to prod and map it */
233 prod_rx_buf->skb = fp->tpa_pool[queue].skb;
234 mapping = dma_map_single(&bp->pdev->dev, fp->tpa_pool[queue].skb->data,
235 bp->rx_buf_size, DMA_FROM_DEVICE);
236 dma_unmap_addr_set(prod_rx_buf, mapping, mapping);
238 /* move partial skb from cons to pool (don't unmap yet) */
239 fp->tpa_pool[queue] = *cons_rx_buf;
241 /* mark bin state as start - print error if current state != stop */
242 if (fp->tpa_state[queue] != BNX2X_TPA_STOP)
243 BNX2X_ERR("start of bin not in stop [%d]\n", queue);
245 fp->tpa_state[queue] = BNX2X_TPA_START;
247 /* point prod_bd to new skb */
248 prod_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
249 prod_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
251 #ifdef BNX2X_STOP_ON_ERROR
252 fp->tpa_queue_used |= (1 << queue);
253 #ifdef _ASM_GENERIC_INT_L64_H
254 DP(NETIF_MSG_RX_STATUS, "fp->tpa_queue_used = 0x%lx\n",
256 DP(NETIF_MSG_RX_STATUS, "fp->tpa_queue_used = 0x%llx\n",
262 /* Timestamp option length allowed for TPA aggregation:
264 * nop nop kind length echo val
266 #define TPA_TSTAMP_OPT_LEN 12
268 * Calculate the approximate value of the MSS for this
269 * aggregation using the first packet of it.
272 * @param parsing_flags Parsing flags from the START CQE
273 * @param len_on_bd Total length of the first packet for the
276 static inline u16 bnx2x_set_lro_mss(struct bnx2x *bp, u16 parsing_flags,
279 /* TPA arrgregation won't have an IP options and TCP options
280 * other than timestamp.
282 u16 hdrs_len = ETH_HLEN + sizeof(struct iphdr) + sizeof(struct tcphdr);
285 /* Check if there was a TCP timestamp, if there is it's will
286 * always be 12 bytes length: nop nop kind length echo val.
288 * Otherwise FW would close the aggregation.
290 if (parsing_flags & PARSING_FLAGS_TIME_STAMP_EXIST_FLAG)
291 hdrs_len += TPA_TSTAMP_OPT_LEN;
293 return len_on_bd - hdrs_len;
296 static int bnx2x_fill_frag_skb(struct bnx2x *bp, struct bnx2x_fastpath *fp,
298 struct eth_fast_path_rx_cqe *fp_cqe,
299 u16 cqe_idx, u16 parsing_flags)
301 struct sw_rx_page *rx_pg, old_rx_pg;
302 u16 len_on_bd = le16_to_cpu(fp_cqe->len_on_bd);
303 u32 i, frag_len, frag_size, pages;
307 frag_size = le16_to_cpu(fp_cqe->pkt_len) - len_on_bd;
308 pages = SGE_PAGE_ALIGN(frag_size) >> SGE_PAGE_SHIFT;
310 /* This is needed in order to enable forwarding support */
312 skb_shinfo(skb)->gso_size = bnx2x_set_lro_mss(bp, parsing_flags,
315 #ifdef BNX2X_STOP_ON_ERROR
316 if (pages > min_t(u32, 8, MAX_SKB_FRAGS)*SGE_PAGE_SIZE*PAGES_PER_SGE) {
317 BNX2X_ERR("SGL length is too long: %d. CQE index is %d\n",
319 BNX2X_ERR("fp_cqe->pkt_len = %d fp_cqe->len_on_bd = %d\n",
320 fp_cqe->pkt_len, len_on_bd);
326 /* Run through the SGL and compose the fragmented skb */
327 for (i = 0, j = 0; i < pages; i += PAGES_PER_SGE, j++) {
329 RX_SGE(le16_to_cpu(fp_cqe->sgl_or_raw_data.sgl[j]));
331 /* FW gives the indices of the SGE as if the ring is an array
332 (meaning that "next" element will consume 2 indices) */
333 frag_len = min(frag_size, (u32)(SGE_PAGE_SIZE*PAGES_PER_SGE));
334 rx_pg = &fp->rx_page_ring[sge_idx];
337 /* If we fail to allocate a substitute page, we simply stop
338 where we are and drop the whole packet */
339 err = bnx2x_alloc_rx_sge(bp, fp, sge_idx);
341 fp->eth_q_stats.rx_skb_alloc_failed++;
345 /* Unmap the page as we r going to pass it to the stack */
346 dma_unmap_page(&bp->pdev->dev,
347 dma_unmap_addr(&old_rx_pg, mapping),
348 SGE_PAGE_SIZE*PAGES_PER_SGE, DMA_FROM_DEVICE);
350 /* Add one frag and update the appropriate fields in the skb */
351 skb_fill_page_desc(skb, j, old_rx_pg.page, 0, frag_len);
353 skb->data_len += frag_len;
354 skb->truesize += frag_len;
355 skb->len += frag_len;
357 frag_size -= frag_len;
363 static void bnx2x_tpa_stop(struct bnx2x *bp, struct bnx2x_fastpath *fp,
364 u16 queue, int pad, int len, union eth_rx_cqe *cqe,
367 struct sw_rx_bd *rx_buf = &fp->tpa_pool[queue];
368 struct sk_buff *skb = rx_buf->skb;
370 struct sk_buff *new_skb = netdev_alloc_skb(bp->dev, bp->rx_buf_size);
372 /* Unmap skb in the pool anyway, as we are going to change
373 pool entry status to BNX2X_TPA_STOP even if new skb allocation
375 dma_unmap_single(&bp->pdev->dev, dma_unmap_addr(rx_buf, mapping),
376 bp->rx_buf_size, DMA_FROM_DEVICE);
378 if (likely(new_skb)) {
379 /* fix ip xsum and give it to the stack */
380 /* (no need to map the new skb) */
382 le16_to_cpu(cqe->fast_path_cqe.pars_flags.flags);
385 prefetch(((char *)(skb)) + L1_CACHE_BYTES);
387 #ifdef BNX2X_STOP_ON_ERROR
388 if (pad + len > bp->rx_buf_size) {
389 BNX2X_ERR("skb_put is about to fail... "
390 "pad %d len %d rx_buf_size %d\n",
391 pad, len, bp->rx_buf_size);
397 skb_reserve(skb, pad);
400 skb->protocol = eth_type_trans(skb, bp->dev);
401 skb->ip_summed = CHECKSUM_UNNECESSARY;
406 iph = (struct iphdr *)skb->data;
408 iph->check = ip_fast_csum((u8 *)iph, iph->ihl);
411 if (!bnx2x_fill_frag_skb(bp, fp, skb,
412 &cqe->fast_path_cqe, cqe_idx,
414 if (parsing_flags & PARSING_FLAGS_VLAN)
415 __vlan_hwaccel_put_tag(skb,
416 le16_to_cpu(cqe->fast_path_cqe.
418 napi_gro_receive(&fp->napi, skb);
420 DP(NETIF_MSG_RX_STATUS, "Failed to allocate new pages"
421 " - dropping packet!\n");
426 /* put new skb in bin */
427 fp->tpa_pool[queue].skb = new_skb;
430 /* else drop the packet and keep the buffer in the bin */
431 DP(NETIF_MSG_RX_STATUS,
432 "Failed to allocate new skb - dropping packet!\n");
433 fp->eth_q_stats.rx_skb_alloc_failed++;
436 fp->tpa_state[queue] = BNX2X_TPA_STOP;
439 /* Set Toeplitz hash value in the skb using the value from the
440 * CQE (calculated by HW).
442 static inline void bnx2x_set_skb_rxhash(struct bnx2x *bp, union eth_rx_cqe *cqe,
445 /* Set Toeplitz hash from CQE */
446 if ((bp->dev->features & NETIF_F_RXHASH) &&
447 (cqe->fast_path_cqe.status_flags &
448 ETH_FAST_PATH_RX_CQE_RSS_HASH_FLG))
450 le32_to_cpu(cqe->fast_path_cqe.rss_hash_result);
453 int bnx2x_rx_int(struct bnx2x_fastpath *fp, int budget)
455 struct bnx2x *bp = fp->bp;
456 u16 bd_cons, bd_prod, bd_prod_fw, comp_ring_cons;
457 u16 hw_comp_cons, sw_comp_cons, sw_comp_prod;
460 #ifdef BNX2X_STOP_ON_ERROR
461 if (unlikely(bp->panic))
465 /* CQ "next element" is of the size of the regular element,
466 that's why it's ok here */
467 hw_comp_cons = le16_to_cpu(*fp->rx_cons_sb);
468 if ((hw_comp_cons & MAX_RCQ_DESC_CNT) == MAX_RCQ_DESC_CNT)
471 bd_cons = fp->rx_bd_cons;
472 bd_prod = fp->rx_bd_prod;
473 bd_prod_fw = bd_prod;
474 sw_comp_cons = fp->rx_comp_cons;
475 sw_comp_prod = fp->rx_comp_prod;
477 /* Memory barrier necessary as speculative reads of the rx
478 * buffer can be ahead of the index in the status block
482 DP(NETIF_MSG_RX_STATUS,
483 "queue[%d]: hw_comp_cons %u sw_comp_cons %u\n",
484 fp->index, hw_comp_cons, sw_comp_cons);
486 while (sw_comp_cons != hw_comp_cons) {
487 struct sw_rx_bd *rx_buf = NULL;
489 union eth_rx_cqe *cqe;
493 comp_ring_cons = RCQ_BD(sw_comp_cons);
494 bd_prod = RX_BD(bd_prod);
495 bd_cons = RX_BD(bd_cons);
497 /* Prefetch the page containing the BD descriptor
498 at producer's index. It will be needed when new skb is
500 prefetch((void *)(PAGE_ALIGN((unsigned long)
501 (&fp->rx_desc_ring[bd_prod])) -
504 cqe = &fp->rx_comp_ring[comp_ring_cons];
505 cqe_fp_flags = cqe->fast_path_cqe.type_error_flags;
507 DP(NETIF_MSG_RX_STATUS, "CQE type %x err %x status %x"
508 " queue %x vlan %x len %u\n", CQE_TYPE(cqe_fp_flags),
509 cqe_fp_flags, cqe->fast_path_cqe.status_flags,
510 le32_to_cpu(cqe->fast_path_cqe.rss_hash_result),
511 le16_to_cpu(cqe->fast_path_cqe.vlan_tag),
512 le16_to_cpu(cqe->fast_path_cqe.pkt_len));
514 /* is this a slowpath msg? */
515 if (unlikely(CQE_TYPE(cqe_fp_flags))) {
516 bnx2x_sp_event(fp, cqe);
519 /* this is an rx packet */
521 rx_buf = &fp->rx_buf_ring[bd_cons];
524 len = le16_to_cpu(cqe->fast_path_cqe.pkt_len);
525 pad = cqe->fast_path_cqe.placement_offset;
527 /* - If CQE is marked both TPA_START and TPA_END it is
529 * - FP CQE will always have either TPA_START or/and
530 * TPA_STOP flags set.
532 if ((!fp->disable_tpa) &&
533 (TPA_TYPE(cqe_fp_flags) !=
534 (TPA_TYPE_START | TPA_TYPE_END))) {
535 u16 queue = cqe->fast_path_cqe.queue_index;
537 if (TPA_TYPE(cqe_fp_flags) == TPA_TYPE_START) {
538 DP(NETIF_MSG_RX_STATUS,
539 "calling tpa_start on queue %d\n",
542 bnx2x_tpa_start(fp, queue, skb,
545 /* Set Toeplitz hash for an LRO skb */
546 bnx2x_set_skb_rxhash(bp, cqe, skb);
549 } else { /* TPA_STOP */
550 DP(NETIF_MSG_RX_STATUS,
551 "calling tpa_stop on queue %d\n",
554 if (!BNX2X_RX_SUM_FIX(cqe))
555 BNX2X_ERR("STOP on none TCP "
558 /* This is a size of the linear data
560 len = le16_to_cpu(cqe->fast_path_cqe.
562 bnx2x_tpa_stop(bp, fp, queue, pad,
563 len, cqe, comp_ring_cons);
564 #ifdef BNX2X_STOP_ON_ERROR
569 bnx2x_update_sge_prod(fp,
570 &cqe->fast_path_cqe);
575 dma_sync_single_for_device(&bp->pdev->dev,
576 dma_unmap_addr(rx_buf, mapping),
577 pad + RX_COPY_THRESH,
579 prefetch(((char *)(skb)) + L1_CACHE_BYTES);
581 /* is this an error packet? */
582 if (unlikely(cqe_fp_flags & ETH_RX_ERROR_FALGS)) {
584 "ERROR flags %x rx packet %u\n",
585 cqe_fp_flags, sw_comp_cons);
586 fp->eth_q_stats.rx_err_discard_pkt++;
590 /* Since we don't have a jumbo ring
591 * copy small packets if mtu > 1500
593 if ((bp->dev->mtu > ETH_MAX_PACKET_SIZE) &&
594 (len <= RX_COPY_THRESH)) {
595 struct sk_buff *new_skb;
597 new_skb = netdev_alloc_skb(bp->dev,
599 if (new_skb == NULL) {
601 "ERROR packet dropped "
602 "because of alloc failure\n");
603 fp->eth_q_stats.rx_skb_alloc_failed++;
608 skb_copy_from_linear_data_offset(skb, pad,
609 new_skb->data + pad, len);
610 skb_reserve(new_skb, pad);
611 skb_put(new_skb, len);
613 bnx2x_reuse_rx_skb(fp, bd_cons, bd_prod);
618 if (likely(bnx2x_alloc_rx_skb(bp, fp, bd_prod) == 0)) {
619 dma_unmap_single(&bp->pdev->dev,
620 dma_unmap_addr(rx_buf, mapping),
623 skb_reserve(skb, pad);
628 "ERROR packet dropped because "
629 "of alloc failure\n");
630 fp->eth_q_stats.rx_skb_alloc_failed++;
632 bnx2x_reuse_rx_skb(fp, bd_cons, bd_prod);
636 skb->protocol = eth_type_trans(skb, bp->dev);
638 /* Set Toeplitz hash for a none-LRO skb */
639 bnx2x_set_skb_rxhash(bp, cqe, skb);
641 skb_checksum_none_assert(skb);
644 if (likely(BNX2X_RX_CSUM_OK(cqe)))
645 skb->ip_summed = CHECKSUM_UNNECESSARY;
647 fp->eth_q_stats.hw_csum_err++;
651 skb_record_rx_queue(skb, fp->index);
653 if (le16_to_cpu(cqe->fast_path_cqe.pars_flags.flags) &
655 __vlan_hwaccel_put_tag(skb,
656 le16_to_cpu(cqe->fast_path_cqe.vlan_tag));
657 napi_gro_receive(&fp->napi, skb);
663 bd_cons = NEXT_RX_IDX(bd_cons);
664 bd_prod = NEXT_RX_IDX(bd_prod);
665 bd_prod_fw = NEXT_RX_IDX(bd_prod_fw);
668 sw_comp_prod = NEXT_RCQ_IDX(sw_comp_prod);
669 sw_comp_cons = NEXT_RCQ_IDX(sw_comp_cons);
671 if (rx_pkt == budget)
675 fp->rx_bd_cons = bd_cons;
676 fp->rx_bd_prod = bd_prod_fw;
677 fp->rx_comp_cons = sw_comp_cons;
678 fp->rx_comp_prod = sw_comp_prod;
680 /* Update producers */
681 bnx2x_update_rx_prod(bp, fp, bd_prod_fw, sw_comp_prod,
684 fp->rx_pkt += rx_pkt;
690 static irqreturn_t bnx2x_msix_fp_int(int irq, void *fp_cookie)
692 struct bnx2x_fastpath *fp = fp_cookie;
693 struct bnx2x *bp = fp->bp;
695 /* Return here if interrupt is disabled */
696 if (unlikely(atomic_read(&bp->intr_sem) != 0)) {
697 DP(NETIF_MSG_INTR, "called but intr_sem not 0, returning\n");
701 DP(BNX2X_MSG_FP, "got an MSI-X interrupt on IDX:SB "
702 "[fp %d fw_sd %d igusb %d]\n",
703 fp->index, fp->fw_sb_id, fp->igu_sb_id);
704 bnx2x_ack_sb(bp, fp->igu_sb_id, USTORM_ID, 0, IGU_INT_DISABLE, 0);
706 #ifdef BNX2X_STOP_ON_ERROR
707 if (unlikely(bp->panic))
711 /* Handle Rx and Tx according to MSI-X vector */
712 prefetch(fp->rx_cons_sb);
713 prefetch(fp->tx_cons_sb);
714 prefetch(&fp->sb_running_index[SM_RX_ID]);
715 napi_schedule(&bnx2x_fp(bp, fp->index, napi));
720 /* HW Lock for shared dual port PHYs */
721 void bnx2x_acquire_phy_lock(struct bnx2x *bp)
723 mutex_lock(&bp->port.phy_mutex);
725 if (bp->port.need_hw_lock)
726 bnx2x_acquire_hw_lock(bp, HW_LOCK_RESOURCE_MDIO);
729 void bnx2x_release_phy_lock(struct bnx2x *bp)
731 if (bp->port.need_hw_lock)
732 bnx2x_release_hw_lock(bp, HW_LOCK_RESOURCE_MDIO);
734 mutex_unlock(&bp->port.phy_mutex);
737 /* calculates MF speed according to current linespeed and MF configuration */
738 u16 bnx2x_get_mf_speed(struct bnx2x *bp)
740 u16 line_speed = bp->link_vars.line_speed;
742 u16 maxCfg = bnx2x_extract_max_cfg(bp,
743 bp->mf_config[BP_VN(bp)]);
745 /* Calculate the current MAX line speed limit for the MF
749 line_speed = (line_speed * maxCfg) / 100;
751 u16 vn_max_rate = maxCfg * 100;
753 if (vn_max_rate < line_speed)
754 line_speed = vn_max_rate;
761 void bnx2x_link_report(struct bnx2x *bp)
763 if (bp->flags & MF_FUNC_DIS) {
764 netif_carrier_off(bp->dev);
765 netdev_err(bp->dev, "NIC Link is Down\n");
769 if (bp->link_vars.link_up) {
772 if (bp->state == BNX2X_STATE_OPEN)
773 netif_carrier_on(bp->dev);
774 netdev_info(bp->dev, "NIC Link is Up, ");
776 line_speed = bnx2x_get_mf_speed(bp);
778 pr_cont("%d Mbps ", line_speed);
780 if (bp->link_vars.duplex == DUPLEX_FULL)
781 pr_cont("full duplex");
783 pr_cont("half duplex");
785 if (bp->link_vars.flow_ctrl != BNX2X_FLOW_CTRL_NONE) {
786 if (bp->link_vars.flow_ctrl & BNX2X_FLOW_CTRL_RX) {
787 pr_cont(", receive ");
788 if (bp->link_vars.flow_ctrl &
790 pr_cont("& transmit ");
792 pr_cont(", transmit ");
794 pr_cont("flow control ON");
798 } else { /* link_down */
799 netif_carrier_off(bp->dev);
800 netdev_err(bp->dev, "NIC Link is Down\n");
804 /* Returns the number of actually allocated BDs */
805 static inline int bnx2x_alloc_rx_bds(struct bnx2x_fastpath *fp,
808 struct bnx2x *bp = fp->bp;
809 u16 ring_prod, cqe_ring_prod;
812 fp->rx_comp_cons = 0;
813 cqe_ring_prod = ring_prod = 0;
814 for (i = 0; i < rx_ring_size; i++) {
815 if (bnx2x_alloc_rx_skb(bp, fp, ring_prod) < 0) {
816 BNX2X_ERR("was only able to allocate "
817 "%d rx skbs on queue[%d]\n", i, fp->index);
818 fp->eth_q_stats.rx_skb_alloc_failed++;
821 ring_prod = NEXT_RX_IDX(ring_prod);
822 cqe_ring_prod = NEXT_RCQ_IDX(cqe_ring_prod);
823 WARN_ON(ring_prod <= i);
826 fp->rx_bd_prod = ring_prod;
827 /* Limit the CQE producer by the CQE ring size */
828 fp->rx_comp_prod = min_t(u16, NUM_RCQ_RINGS*RCQ_DESC_CNT,
830 fp->rx_pkt = fp->rx_calls = 0;
835 static inline void bnx2x_alloc_rx_bd_ring(struct bnx2x_fastpath *fp)
837 struct bnx2x *bp = fp->bp;
838 int rx_ring_size = bp->rx_ring_size ? bp->rx_ring_size :
839 MAX_RX_AVAIL/bp->num_queues;
841 rx_ring_size = max_t(int, MIN_RX_AVAIL, rx_ring_size);
843 bnx2x_alloc_rx_bds(fp, rx_ring_size);
846 * this will generate an interrupt (to the TSTORM)
847 * must only be done after chip is initialized
849 bnx2x_update_rx_prod(bp, fp, fp->rx_bd_prod, fp->rx_comp_prod,
853 void bnx2x_init_rx_rings(struct bnx2x *bp)
855 int func = BP_FUNC(bp);
856 int max_agg_queues = CHIP_IS_E1(bp) ? ETH_MAX_AGGREGATION_QUEUES_E1 :
857 ETH_MAX_AGGREGATION_QUEUES_E1H;
861 bp->rx_buf_size = bp->dev->mtu + ETH_OVREHEAD + BNX2X_RX_ALIGN +
862 IP_HEADER_ALIGNMENT_PADDING;
865 "mtu %d rx_buf_size %d\n", bp->dev->mtu, bp->rx_buf_size);
867 for_each_rx_queue(bp, j) {
868 struct bnx2x_fastpath *fp = &bp->fp[j];
870 if (!fp->disable_tpa) {
871 for (i = 0; i < max_agg_queues; i++) {
872 fp->tpa_pool[i].skb =
873 netdev_alloc_skb(bp->dev, bp->rx_buf_size);
874 if (!fp->tpa_pool[i].skb) {
875 BNX2X_ERR("Failed to allocate TPA "
876 "skb pool for queue[%d] - "
877 "disabling TPA on this "
879 bnx2x_free_tpa_pool(bp, fp, i);
883 dma_unmap_addr_set((struct sw_rx_bd *)
884 &bp->fp->tpa_pool[i],
886 fp->tpa_state[i] = BNX2X_TPA_STOP;
889 /* "next page" elements initialization */
890 bnx2x_set_next_page_sgl(fp);
892 /* set SGEs bit mask */
893 bnx2x_init_sge_ring_bit_mask(fp);
895 /* Allocate SGEs and initialize the ring elements */
896 for (i = 0, ring_prod = 0;
897 i < MAX_RX_SGE_CNT*NUM_RX_SGE_PAGES; i++) {
899 if (bnx2x_alloc_rx_sge(bp, fp, ring_prod) < 0) {
900 BNX2X_ERR("was only able to allocate "
902 BNX2X_ERR("disabling TPA for"
904 /* Cleanup already allocated elements */
905 bnx2x_free_rx_sge_range(bp,
907 bnx2x_free_tpa_pool(bp,
913 ring_prod = NEXT_SGE_IDX(ring_prod);
916 fp->rx_sge_prod = ring_prod;
920 for_each_rx_queue(bp, j) {
921 struct bnx2x_fastpath *fp = &bp->fp[j];
925 bnx2x_set_next_page_rx_bd(fp);
928 bnx2x_set_next_page_rx_cq(fp);
930 /* Allocate BDs and initialize BD ring */
931 bnx2x_alloc_rx_bd_ring(fp);
936 if (!CHIP_IS_E2(bp)) {
937 REG_WR(bp, BAR_USTRORM_INTMEM +
938 USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func),
939 U64_LO(fp->rx_comp_mapping));
940 REG_WR(bp, BAR_USTRORM_INTMEM +
941 USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func) + 4,
942 U64_HI(fp->rx_comp_mapping));
947 static void bnx2x_free_tx_skbs(struct bnx2x *bp)
951 for_each_tx_queue(bp, i) {
952 struct bnx2x_fastpath *fp = &bp->fp[i];
954 u16 bd_cons = fp->tx_bd_cons;
955 u16 sw_prod = fp->tx_pkt_prod;
956 u16 sw_cons = fp->tx_pkt_cons;
958 while (sw_cons != sw_prod) {
959 bd_cons = bnx2x_free_tx_pkt(bp, fp, TX_BD(sw_cons));
965 static void bnx2x_free_rx_skbs(struct bnx2x *bp)
969 for_each_rx_queue(bp, j) {
970 struct bnx2x_fastpath *fp = &bp->fp[j];
972 for (i = 0; i < NUM_RX_BD; i++) {
973 struct sw_rx_bd *rx_buf = &fp->rx_buf_ring[i];
974 struct sk_buff *skb = rx_buf->skb;
979 dma_unmap_single(&bp->pdev->dev,
980 dma_unmap_addr(rx_buf, mapping),
981 bp->rx_buf_size, DMA_FROM_DEVICE);
986 if (!fp->disable_tpa)
987 bnx2x_free_tpa_pool(bp, fp, CHIP_IS_E1(bp) ?
988 ETH_MAX_AGGREGATION_QUEUES_E1 :
989 ETH_MAX_AGGREGATION_QUEUES_E1H);
993 void bnx2x_free_skbs(struct bnx2x *bp)
995 bnx2x_free_tx_skbs(bp);
996 bnx2x_free_rx_skbs(bp);
999 static void bnx2x_free_msix_irqs(struct bnx2x *bp)
1003 free_irq(bp->msix_table[0].vector, bp->dev);
1004 DP(NETIF_MSG_IFDOWN, "released sp irq (%d)\n",
1005 bp->msix_table[0].vector);
1010 for_each_eth_queue(bp, i) {
1011 DP(NETIF_MSG_IFDOWN, "about to release fp #%d->%d irq "
1012 "state %x\n", i, bp->msix_table[i + offset].vector,
1013 bnx2x_fp(bp, i, state));
1015 free_irq(bp->msix_table[i + offset].vector, &bp->fp[i]);
1019 void bnx2x_free_irq(struct bnx2x *bp)
1021 if (bp->flags & USING_MSIX_FLAG)
1022 bnx2x_free_msix_irqs(bp);
1023 else if (bp->flags & USING_MSI_FLAG)
1024 free_irq(bp->pdev->irq, bp->dev);
1026 free_irq(bp->pdev->irq, bp->dev);
1029 int bnx2x_enable_msix(struct bnx2x *bp)
1031 int msix_vec = 0, i, rc, req_cnt;
1033 bp->msix_table[msix_vec].entry = msix_vec;
1034 DP(NETIF_MSG_IFUP, "msix_table[0].entry = %d (slowpath)\n",
1035 bp->msix_table[0].entry);
1039 bp->msix_table[msix_vec].entry = msix_vec;
1040 DP(NETIF_MSG_IFUP, "msix_table[%d].entry = %d (CNIC)\n",
1041 bp->msix_table[msix_vec].entry, bp->msix_table[msix_vec].entry);
1044 for_each_eth_queue(bp, i) {
1045 bp->msix_table[msix_vec].entry = msix_vec;
1046 DP(NETIF_MSG_IFUP, "msix_table[%d].entry = %d "
1047 "(fastpath #%u)\n", msix_vec, msix_vec, i);
1051 req_cnt = BNX2X_NUM_ETH_QUEUES(bp) + CNIC_CONTEXT_USE + 1;
1053 rc = pci_enable_msix(bp->pdev, &bp->msix_table[0], req_cnt);
1056 * reconfigure number of tx/rx queues according to available
1059 if (rc >= BNX2X_MIN_MSIX_VEC_CNT) {
1060 /* how less vectors we will have? */
1061 int diff = req_cnt - rc;
1064 "Trying to use less MSI-X vectors: %d\n", rc);
1066 rc = pci_enable_msix(bp->pdev, &bp->msix_table[0], rc);
1070 "MSI-X is not attainable rc %d\n", rc);
1074 * decrease number of queues by number of unallocated entries
1076 bp->num_queues -= diff;
1078 DP(NETIF_MSG_IFUP, "New queue configuration set: %d\n",
1081 /* fall to INTx if not enough memory */
1083 bp->flags |= DISABLE_MSI_FLAG;
1084 DP(NETIF_MSG_IFUP, "MSI-X is not attainable rc %d\n", rc);
1088 bp->flags |= USING_MSIX_FLAG;
1093 static int bnx2x_req_msix_irqs(struct bnx2x *bp)
1095 int i, rc, offset = 1;
1097 rc = request_irq(bp->msix_table[0].vector, bnx2x_msix_sp_int, 0,
1098 bp->dev->name, bp->dev);
1100 BNX2X_ERR("request sp irq failed\n");
1107 for_each_eth_queue(bp, i) {
1108 struct bnx2x_fastpath *fp = &bp->fp[i];
1109 snprintf(fp->name, sizeof(fp->name), "%s-fp-%d",
1112 rc = request_irq(bp->msix_table[offset].vector,
1113 bnx2x_msix_fp_int, 0, fp->name, fp);
1115 BNX2X_ERR("request fp #%d irq failed rc %d\n", i, rc);
1116 bnx2x_free_msix_irqs(bp);
1121 fp->state = BNX2X_FP_STATE_IRQ;
1124 i = BNX2X_NUM_ETH_QUEUES(bp);
1125 offset = 1 + CNIC_CONTEXT_USE;
1126 netdev_info(bp->dev, "using MSI-X IRQs: sp %d fp[%d] %d"
1128 bp->msix_table[0].vector,
1129 0, bp->msix_table[offset].vector,
1130 i - 1, bp->msix_table[offset + i - 1].vector);
1135 int bnx2x_enable_msi(struct bnx2x *bp)
1139 rc = pci_enable_msi(bp->pdev);
1141 DP(NETIF_MSG_IFUP, "MSI is not attainable\n");
1144 bp->flags |= USING_MSI_FLAG;
1149 static int bnx2x_req_irq(struct bnx2x *bp)
1151 unsigned long flags;
1154 if (bp->flags & USING_MSI_FLAG)
1157 flags = IRQF_SHARED;
1159 rc = request_irq(bp->pdev->irq, bnx2x_interrupt, flags,
1160 bp->dev->name, bp->dev);
1162 bnx2x_fp(bp, 0, state) = BNX2X_FP_STATE_IRQ;
1167 static void bnx2x_napi_enable(struct bnx2x *bp)
1171 for_each_napi_queue(bp, i)
1172 napi_enable(&bnx2x_fp(bp, i, napi));
1175 static void bnx2x_napi_disable(struct bnx2x *bp)
1179 for_each_napi_queue(bp, i)
1180 napi_disable(&bnx2x_fp(bp, i, napi));
1183 void bnx2x_netif_start(struct bnx2x *bp)
1187 intr_sem = atomic_dec_and_test(&bp->intr_sem);
1188 smp_wmb(); /* Ensure that bp->intr_sem update is SMP-safe */
1191 if (netif_running(bp->dev)) {
1192 bnx2x_napi_enable(bp);
1193 bnx2x_int_enable(bp);
1194 if (bp->state == BNX2X_STATE_OPEN)
1195 netif_tx_wake_all_queues(bp->dev);
1200 void bnx2x_netif_stop(struct bnx2x *bp, int disable_hw)
1202 bnx2x_int_disable_sync(bp, disable_hw);
1203 bnx2x_napi_disable(bp);
1204 netif_tx_disable(bp->dev);
1207 u16 bnx2x_select_queue(struct net_device *dev, struct sk_buff *skb)
1210 struct bnx2x *bp = netdev_priv(dev);
1212 return skb_tx_hash(dev, skb);
1214 struct ethhdr *hdr = (struct ethhdr *)skb->data;
1215 u16 ether_type = ntohs(hdr->h_proto);
1217 /* Skip VLAN tag if present */
1218 if (ether_type == ETH_P_8021Q) {
1219 struct vlan_ethhdr *vhdr =
1220 (struct vlan_ethhdr *)skb->data;
1222 ether_type = ntohs(vhdr->h_vlan_encapsulated_proto);
1225 /* If ethertype is FCoE or FIP - use FCoE ring */
1226 if ((ether_type == ETH_P_FCOE) || (ether_type == ETH_P_FIP))
1227 return bnx2x_fcoe(bp, index);
1230 /* Select a none-FCoE queue: if FCoE is enabled, exclude FCoE L2 ring
1232 return __skb_tx_hash(dev, skb,
1233 dev->real_num_tx_queues - FCOE_CONTEXT_USE);
1236 void bnx2x_set_num_queues(struct bnx2x *bp)
1238 switch (bp->multi_mode) {
1239 case ETH_RSS_MODE_DISABLED:
1242 case ETH_RSS_MODE_REGULAR:
1243 bp->num_queues = bnx2x_calc_num_queues(bp);
1251 /* Add special queues */
1252 bp->num_queues += NONE_ETH_CONTEXT_USE;
1256 static inline void bnx2x_set_fcoe_eth_macs(struct bnx2x *bp)
1260 bnx2x_set_fip_eth_mac_addr(bp, 1);
1261 bnx2x_set_all_enode_macs(bp, 1);
1262 bp->flags |= FCOE_MACS_SET;
1267 static void bnx2x_release_firmware(struct bnx2x *bp)
1269 kfree(bp->init_ops_offsets);
1270 kfree(bp->init_ops);
1271 kfree(bp->init_data);
1272 release_firmware(bp->firmware);
1275 static inline int bnx2x_set_real_num_queues(struct bnx2x *bp)
1277 int rc, num = bp->num_queues;
1281 num -= FCOE_CONTEXT_USE;
1284 netif_set_real_num_tx_queues(bp->dev, num);
1285 rc = netif_set_real_num_rx_queues(bp->dev, num);
1289 /* must be called with rtnl_lock */
1290 int bnx2x_nic_load(struct bnx2x *bp, int load_mode)
1295 /* Set init arrays */
1296 rc = bnx2x_init_firmware(bp);
1298 BNX2X_ERR("Error loading firmware\n");
1302 #ifdef BNX2X_STOP_ON_ERROR
1303 if (unlikely(bp->panic))
1307 bp->state = BNX2X_STATE_OPENING_WAIT4_LOAD;
1309 /* must be called before memory allocation and HW init */
1310 bnx2x_ilt_set_info(bp);
1312 if (bnx2x_alloc_mem(bp))
1315 rc = bnx2x_set_real_num_queues(bp);
1317 BNX2X_ERR("Unable to set real_num_queues\n");
1321 for_each_queue(bp, i)
1322 bnx2x_fp(bp, i, disable_tpa) =
1323 ((bp->flags & TPA_ENABLE_FLAG) == 0);
1326 /* We don't want TPA on FCoE L2 ring */
1327 bnx2x_fcoe(bp, disable_tpa) = 1;
1329 bnx2x_napi_enable(bp);
1331 /* Send LOAD_REQUEST command to MCP
1332 Returns the type of LOAD command:
1333 if it is the first port to be initialized
1334 common blocks should be initialized, otherwise - not
1336 if (!BP_NOMCP(bp)) {
1337 load_code = bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_REQ, 0);
1339 BNX2X_ERR("MCP response failure, aborting\n");
1343 if (load_code == FW_MSG_CODE_DRV_LOAD_REFUSED) {
1344 rc = -EBUSY; /* other port in diagnostic mode */
1349 int path = BP_PATH(bp);
1350 int port = BP_PORT(bp);
1352 DP(NETIF_MSG_IFUP, "NO MCP - load counts[%d] %d, %d, %d\n",
1353 path, load_count[path][0], load_count[path][1],
1354 load_count[path][2]);
1355 load_count[path][0]++;
1356 load_count[path][1 + port]++;
1357 DP(NETIF_MSG_IFUP, "NO MCP - new load counts[%d] %d, %d, %d\n",
1358 path, load_count[path][0], load_count[path][1],
1359 load_count[path][2]);
1360 if (load_count[path][0] == 1)
1361 load_code = FW_MSG_CODE_DRV_LOAD_COMMON;
1362 else if (load_count[path][1 + port] == 1)
1363 load_code = FW_MSG_CODE_DRV_LOAD_PORT;
1365 load_code = FW_MSG_CODE_DRV_LOAD_FUNCTION;
1368 if ((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
1369 (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP) ||
1370 (load_code == FW_MSG_CODE_DRV_LOAD_PORT))
1374 DP(NETIF_MSG_LINK, "pmf %d\n", bp->port.pmf);
1377 rc = bnx2x_init_hw(bp, load_code);
1379 BNX2X_ERR("HW init failed, aborting\n");
1380 bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
1384 /* Connect to IRQs */
1385 rc = bnx2x_setup_irqs(bp);
1387 bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
1391 /* Setup NIC internals and enable interrupts */
1392 bnx2x_nic_init(bp, load_code);
1394 if (((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
1395 (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP)) &&
1396 (bp->common.shmem2_base))
1397 SHMEM2_WR(bp, dcc_support,
1398 (SHMEM_DCC_SUPPORT_DISABLE_ENABLE_PF_TLV |
1399 SHMEM_DCC_SUPPORT_BANDWIDTH_ALLOCATION_TLV));
1401 /* Send LOAD_DONE command to MCP */
1402 if (!BP_NOMCP(bp)) {
1403 load_code = bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
1405 BNX2X_ERR("MCP response failure, aborting\n");
1411 bnx2x_dcbx_init(bp);
1413 bp->state = BNX2X_STATE_OPENING_WAIT4_PORT;
1415 rc = bnx2x_func_start(bp);
1417 BNX2X_ERR("Function start failed!\n");
1418 #ifndef BNX2X_STOP_ON_ERROR
1426 rc = bnx2x_setup_client(bp, &bp->fp[0], 1 /* Leading */);
1428 BNX2X_ERR("Setup leading failed!\n");
1429 #ifndef BNX2X_STOP_ON_ERROR
1437 if (!CHIP_IS_E1(bp) &&
1438 (bp->mf_config[BP_VN(bp)] & FUNC_MF_CFG_FUNC_DISABLED)) {
1439 DP(NETIF_MSG_IFUP, "mf_cfg function disabled\n");
1440 bp->flags |= MF_FUNC_DIS;
1444 /* Enable Timer scan */
1445 REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + BP_PORT(bp)*4, 1);
1448 for_each_nondefault_queue(bp, i) {
1449 rc = bnx2x_setup_client(bp, &bp->fp[i], 0);
1458 /* Now when Clients are configured we are ready to work */
1459 bp->state = BNX2X_STATE_OPEN;
1462 bnx2x_set_fcoe_eth_macs(bp);
1465 bnx2x_set_eth_mac(bp, 1);
1468 bnx2x_initial_phy_init(bp, load_mode);
1470 /* Start fast path */
1471 switch (load_mode) {
1473 /* Tx queue should be only reenabled */
1474 netif_tx_wake_all_queues(bp->dev);
1475 /* Initialize the receive filter. */
1476 bnx2x_set_rx_mode(bp->dev);
1480 netif_tx_start_all_queues(bp->dev);
1481 smp_mb__after_clear_bit();
1482 /* Initialize the receive filter. */
1483 bnx2x_set_rx_mode(bp->dev);
1487 /* Initialize the receive filter. */
1488 bnx2x_set_rx_mode(bp->dev);
1489 bp->state = BNX2X_STATE_DIAG;
1497 bnx2x__link_status_update(bp);
1499 /* start the timer */
1500 mod_timer(&bp->timer, jiffies + bp->current_interval);
1503 bnx2x_setup_cnic_irq_info(bp);
1504 if (bp->state == BNX2X_STATE_OPEN)
1505 bnx2x_cnic_notify(bp, CNIC_CTL_START_CMD);
1507 bnx2x_inc_load_cnt(bp);
1509 bnx2x_release_firmware(bp);
1515 /* Disable Timer scan */
1516 REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + BP_PORT(bp)*4, 0);
1519 bnx2x_int_disable_sync(bp, 1);
1521 /* Free SKBs, SGEs, TPA pool and driver internals */
1522 bnx2x_free_skbs(bp);
1523 for_each_rx_queue(bp, i)
1524 bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
1529 if (!BP_NOMCP(bp)) {
1530 bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_REQ_WOL_MCP, 0);
1531 bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_DONE, 0);
1536 bnx2x_napi_disable(bp);
1540 bnx2x_release_firmware(bp);
1545 /* must be called with rtnl_lock */
1546 int bnx2x_nic_unload(struct bnx2x *bp, int unload_mode)
1550 if (bp->state == BNX2X_STATE_CLOSED) {
1551 /* Interface has been removed - nothing to recover */
1552 bp->recovery_state = BNX2X_RECOVERY_DONE;
1554 bnx2x_release_hw_lock(bp, HW_LOCK_RESOURCE_RESERVED_08);
1561 bnx2x_cnic_notify(bp, CNIC_CTL_STOP_CMD);
1563 bp->state = BNX2X_STATE_CLOSING_WAIT4_HALT;
1565 /* Set "drop all" */
1566 bp->rx_mode = BNX2X_RX_MODE_NONE;
1567 bnx2x_set_storm_rx_mode(bp);
1570 bnx2x_tx_disable(bp);
1572 del_timer_sync(&bp->timer);
1574 SHMEM_WR(bp, func_mb[BP_FW_MB_IDX(bp)].drv_pulse_mb,
1575 (DRV_PULSE_ALWAYS_ALIVE | bp->fw_drv_pulse_wr_seq));
1577 bnx2x_stats_handle(bp, STATS_EVENT_STOP);
1579 /* Cleanup the chip if needed */
1580 if (unload_mode != UNLOAD_RECOVERY)
1581 bnx2x_chip_cleanup(bp, unload_mode);
1583 /* Disable HW interrupts, NAPI and Tx */
1584 bnx2x_netif_stop(bp, 1);
1592 /* Free SKBs, SGEs, TPA pool and driver internals */
1593 bnx2x_free_skbs(bp);
1594 for_each_rx_queue(bp, i)
1595 bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
1599 bp->state = BNX2X_STATE_CLOSED;
1601 /* The last driver must disable a "close the gate" if there is no
1602 * parity attention or "process kill" pending.
1604 if ((!bnx2x_dec_load_cnt(bp)) && (!bnx2x_chk_parity_attn(bp)) &&
1605 bnx2x_reset_is_done(bp))
1606 bnx2x_disable_close_the_gate(bp);
1608 /* Reset MCP mail box sequence if there is on going recovery */
1609 if (unload_mode == UNLOAD_RECOVERY)
1615 int bnx2x_set_power_state(struct bnx2x *bp, pci_power_t state)
1619 /* If there is no power capability, silently succeed */
1621 DP(NETIF_MSG_HW, "No power capability. Breaking.\n");
1625 pci_read_config_word(bp->pdev, bp->pm_cap + PCI_PM_CTRL, &pmcsr);
1629 pci_write_config_word(bp->pdev, bp->pm_cap + PCI_PM_CTRL,
1630 ((pmcsr & ~PCI_PM_CTRL_STATE_MASK) |
1631 PCI_PM_CTRL_PME_STATUS));
1633 if (pmcsr & PCI_PM_CTRL_STATE_MASK)
1634 /* delay required during transition out of D3hot */
1639 /* If there are other clients above don't
1640 shut down the power */
1641 if (atomic_read(&bp->pdev->enable_cnt) != 1)
1643 /* Don't shut down the power for emulation and FPGA */
1644 if (CHIP_REV_IS_SLOW(bp))
1647 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
1651 pmcsr |= PCI_PM_CTRL_PME_ENABLE;
1653 pci_write_config_word(bp->pdev, bp->pm_cap + PCI_PM_CTRL,
1656 /* No more memory access after this point until
1657 * device is brought back to D0.
1668 * net_device service functions
1670 int bnx2x_poll(struct napi_struct *napi, int budget)
1673 struct bnx2x_fastpath *fp = container_of(napi, struct bnx2x_fastpath,
1675 struct bnx2x *bp = fp->bp;
1678 #ifdef BNX2X_STOP_ON_ERROR
1679 if (unlikely(bp->panic)) {
1680 napi_complete(napi);
1685 if (bnx2x_has_tx_work(fp))
1688 if (bnx2x_has_rx_work(fp)) {
1689 work_done += bnx2x_rx_int(fp, budget - work_done);
1691 /* must not complete if we consumed full budget */
1692 if (work_done >= budget)
1696 /* Fall out from the NAPI loop if needed */
1697 if (!(bnx2x_has_rx_work(fp) || bnx2x_has_tx_work(fp))) {
1699 /* No need to update SB for FCoE L2 ring as long as
1700 * it's connected to the default SB and the SB
1701 * has been updated when NAPI was scheduled.
1703 if (IS_FCOE_FP(fp)) {
1704 napi_complete(napi);
1709 bnx2x_update_fpsb_idx(fp);
1710 /* bnx2x_has_rx_work() reads the status block,
1711 * thus we need to ensure that status block indices
1712 * have been actually read (bnx2x_update_fpsb_idx)
1713 * prior to this check (bnx2x_has_rx_work) so that
1714 * we won't write the "newer" value of the status block
1715 * to IGU (if there was a DMA right after
1716 * bnx2x_has_rx_work and if there is no rmb, the memory
1717 * reading (bnx2x_update_fpsb_idx) may be postponed
1718 * to right before bnx2x_ack_sb). In this case there
1719 * will never be another interrupt until there is
1720 * another update of the status block, while there
1721 * is still unhandled work.
1725 if (!(bnx2x_has_rx_work(fp) || bnx2x_has_tx_work(fp))) {
1726 napi_complete(napi);
1727 /* Re-enable interrupts */
1729 "Update index to %d\n", fp->fp_hc_idx);
1730 bnx2x_ack_sb(bp, fp->igu_sb_id, USTORM_ID,
1731 le16_to_cpu(fp->fp_hc_idx),
1741 /* we split the first BD into headers and data BDs
1742 * to ease the pain of our fellow microcode engineers
1743 * we use one mapping for both BDs
1744 * So far this has only been observed to happen
1745 * in Other Operating Systems(TM)
1747 static noinline u16 bnx2x_tx_split(struct bnx2x *bp,
1748 struct bnx2x_fastpath *fp,
1749 struct sw_tx_bd *tx_buf,
1750 struct eth_tx_start_bd **tx_bd, u16 hlen,
1751 u16 bd_prod, int nbd)
1753 struct eth_tx_start_bd *h_tx_bd = *tx_bd;
1754 struct eth_tx_bd *d_tx_bd;
1756 int old_len = le16_to_cpu(h_tx_bd->nbytes);
1758 /* first fix first BD */
1759 h_tx_bd->nbd = cpu_to_le16(nbd);
1760 h_tx_bd->nbytes = cpu_to_le16(hlen);
1762 DP(NETIF_MSG_TX_QUEUED, "TSO split header size is %d "
1763 "(%x:%x) nbd %d\n", h_tx_bd->nbytes, h_tx_bd->addr_hi,
1764 h_tx_bd->addr_lo, h_tx_bd->nbd);
1766 /* now get a new data BD
1767 * (after the pbd) and fill it */
1768 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
1769 d_tx_bd = &fp->tx_desc_ring[bd_prod].reg_bd;
1771 mapping = HILO_U64(le32_to_cpu(h_tx_bd->addr_hi),
1772 le32_to_cpu(h_tx_bd->addr_lo)) + hlen;
1774 d_tx_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
1775 d_tx_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
1776 d_tx_bd->nbytes = cpu_to_le16(old_len - hlen);
1778 /* this marks the BD as one that has no individual mapping */
1779 tx_buf->flags |= BNX2X_TSO_SPLIT_BD;
1781 DP(NETIF_MSG_TX_QUEUED,
1782 "TSO split data size is %d (%x:%x)\n",
1783 d_tx_bd->nbytes, d_tx_bd->addr_hi, d_tx_bd->addr_lo);
1786 *tx_bd = (struct eth_tx_start_bd *)d_tx_bd;
1791 static inline u16 bnx2x_csum_fix(unsigned char *t_header, u16 csum, s8 fix)
1794 csum = (u16) ~csum_fold(csum_sub(csum,
1795 csum_partial(t_header - fix, fix, 0)));
1798 csum = (u16) ~csum_fold(csum_add(csum,
1799 csum_partial(t_header, -fix, 0)));
1801 return swab16(csum);
1804 static inline u32 bnx2x_xmit_type(struct bnx2x *bp, struct sk_buff *skb)
1808 if (skb->ip_summed != CHECKSUM_PARTIAL)
1812 if (vlan_get_protocol(skb) == htons(ETH_P_IPV6)) {
1814 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
1815 rc |= XMIT_CSUM_TCP;
1819 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
1820 rc |= XMIT_CSUM_TCP;
1824 if (skb_is_gso_v6(skb))
1825 rc |= XMIT_GSO_V6 | XMIT_CSUM_TCP | XMIT_CSUM_V6;
1826 else if (skb_is_gso(skb))
1827 rc |= XMIT_GSO_V4 | XMIT_CSUM_V4 | XMIT_CSUM_TCP;
1832 #if (MAX_SKB_FRAGS >= MAX_FETCH_BD - 3)
1833 /* check if packet requires linearization (packet is too fragmented)
1834 no need to check fragmentation if page size > 8K (there will be no
1835 violation to FW restrictions) */
1836 static int bnx2x_pkt_req_lin(struct bnx2x *bp, struct sk_buff *skb,
1841 int first_bd_sz = 0;
1843 /* 3 = 1 (for linear data BD) + 2 (for PBD and last BD) */
1844 if (skb_shinfo(skb)->nr_frags >= (MAX_FETCH_BD - 3)) {
1846 if (xmit_type & XMIT_GSO) {
1847 unsigned short lso_mss = skb_shinfo(skb)->gso_size;
1848 /* Check if LSO packet needs to be copied:
1849 3 = 1 (for headers BD) + 2 (for PBD and last BD) */
1850 int wnd_size = MAX_FETCH_BD - 3;
1851 /* Number of windows to check */
1852 int num_wnds = skb_shinfo(skb)->nr_frags - wnd_size;
1857 /* Headers length */
1858 hlen = (int)(skb_transport_header(skb) - skb->data) +
1861 /* Amount of data (w/o headers) on linear part of SKB*/
1862 first_bd_sz = skb_headlen(skb) - hlen;
1864 wnd_sum = first_bd_sz;
1866 /* Calculate the first sum - it's special */
1867 for (frag_idx = 0; frag_idx < wnd_size - 1; frag_idx++)
1869 skb_shinfo(skb)->frags[frag_idx].size;
1871 /* If there was data on linear skb data - check it */
1872 if (first_bd_sz > 0) {
1873 if (unlikely(wnd_sum < lso_mss)) {
1878 wnd_sum -= first_bd_sz;
1881 /* Others are easier: run through the frag list and
1882 check all windows */
1883 for (wnd_idx = 0; wnd_idx <= num_wnds; wnd_idx++) {
1885 skb_shinfo(skb)->frags[wnd_idx + wnd_size - 1].size;
1887 if (unlikely(wnd_sum < lso_mss)) {
1892 skb_shinfo(skb)->frags[wnd_idx].size;
1895 /* in non-LSO too fragmented packet should always
1902 if (unlikely(to_copy))
1903 DP(NETIF_MSG_TX_QUEUED,
1904 "Linearization IS REQUIRED for %s packet. "
1905 "num_frags %d hlen %d first_bd_sz %d\n",
1906 (xmit_type & XMIT_GSO) ? "LSO" : "non-LSO",
1907 skb_shinfo(skb)->nr_frags, hlen, first_bd_sz);
1913 static inline void bnx2x_set_pbd_gso_e2(struct sk_buff *skb, u32 *parsing_data,
1916 *parsing_data |= (skb_shinfo(skb)->gso_size <<
1917 ETH_TX_PARSE_BD_E2_LSO_MSS_SHIFT) &
1918 ETH_TX_PARSE_BD_E2_LSO_MSS;
1919 if ((xmit_type & XMIT_GSO_V6) &&
1920 (ipv6_hdr(skb)->nexthdr == NEXTHDR_IPV6))
1921 *parsing_data |= ETH_TX_PARSE_BD_E2_IPV6_WITH_EXT_HDR;
1925 * Update PBD in GSO case.
1928 * @param tx_start_bd
1932 static inline void bnx2x_set_pbd_gso(struct sk_buff *skb,
1933 struct eth_tx_parse_bd_e1x *pbd,
1936 pbd->lso_mss = cpu_to_le16(skb_shinfo(skb)->gso_size);
1937 pbd->tcp_send_seq = swab32(tcp_hdr(skb)->seq);
1938 pbd->tcp_flags = pbd_tcp_flags(skb);
1940 if (xmit_type & XMIT_GSO_V4) {
1941 pbd->ip_id = swab16(ip_hdr(skb)->id);
1942 pbd->tcp_pseudo_csum =
1943 swab16(~csum_tcpudp_magic(ip_hdr(skb)->saddr,
1945 0, IPPROTO_TCP, 0));
1948 pbd->tcp_pseudo_csum =
1949 swab16(~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
1950 &ipv6_hdr(skb)->daddr,
1951 0, IPPROTO_TCP, 0));
1953 pbd->global_data |= ETH_TX_PARSE_BD_E1X_PSEUDO_CS_WITHOUT_LEN;
1959 * @param tx_start_bd
1963 * @return header len
1965 static inline u8 bnx2x_set_pbd_csum_e2(struct bnx2x *bp, struct sk_buff *skb,
1966 u32 *parsing_data, u32 xmit_type)
1968 *parsing_data |= ((tcp_hdrlen(skb)/4) <<
1969 ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW_SHIFT) &
1970 ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW;
1972 *parsing_data |= ((((u8 *)tcp_hdr(skb) - skb->data) / 2) <<
1973 ETH_TX_PARSE_BD_E2_TCP_HDR_START_OFFSET_W_SHIFT) &
1974 ETH_TX_PARSE_BD_E2_TCP_HDR_START_OFFSET_W;
1976 return skb_transport_header(skb) + tcp_hdrlen(skb) - skb->data;
1982 * @param tx_start_bd
1986 * @return Header length
1988 static inline u8 bnx2x_set_pbd_csum(struct bnx2x *bp, struct sk_buff *skb,
1989 struct eth_tx_parse_bd_e1x *pbd,
1992 u8 hlen = (skb_network_header(skb) - skb->data) / 2;
1994 /* for now NS flag is not used in Linux */
1996 (hlen | ((skb->protocol == cpu_to_be16(ETH_P_8021Q)) <<
1997 ETH_TX_PARSE_BD_E1X_LLC_SNAP_EN_SHIFT));
1999 pbd->ip_hlen_w = (skb_transport_header(skb) -
2000 skb_network_header(skb)) / 2;
2002 hlen += pbd->ip_hlen_w + tcp_hdrlen(skb) / 2;
2004 pbd->total_hlen_w = cpu_to_le16(hlen);
2007 if (xmit_type & XMIT_CSUM_TCP) {
2008 pbd->tcp_pseudo_csum = swab16(tcp_hdr(skb)->check);
2011 s8 fix = SKB_CS_OFF(skb); /* signed! */
2013 DP(NETIF_MSG_TX_QUEUED,
2014 "hlen %d fix %d csum before fix %x\n",
2015 le16_to_cpu(pbd->total_hlen_w), fix, SKB_CS(skb));
2017 /* HW bug: fixup the CSUM */
2018 pbd->tcp_pseudo_csum =
2019 bnx2x_csum_fix(skb_transport_header(skb),
2022 DP(NETIF_MSG_TX_QUEUED, "csum after fix %x\n",
2023 pbd->tcp_pseudo_csum);
2029 /* called with netif_tx_lock
2030 * bnx2x_tx_int() runs without netif_tx_lock unless it needs to call
2031 * netif_wake_queue()
2033 netdev_tx_t bnx2x_start_xmit(struct sk_buff *skb, struct net_device *dev)
2035 struct bnx2x *bp = netdev_priv(dev);
2036 struct bnx2x_fastpath *fp;
2037 struct netdev_queue *txq;
2038 struct sw_tx_bd *tx_buf;
2039 struct eth_tx_start_bd *tx_start_bd;
2040 struct eth_tx_bd *tx_data_bd, *total_pkt_bd = NULL;
2041 struct eth_tx_parse_bd_e1x *pbd_e1x = NULL;
2042 struct eth_tx_parse_bd_e2 *pbd_e2 = NULL;
2043 u32 pbd_e2_parsing_data = 0;
2044 u16 pkt_prod, bd_prod;
2047 u32 xmit_type = bnx2x_xmit_type(bp, skb);
2050 __le16 pkt_size = 0;
2052 u8 mac_type = UNICAST_ADDRESS;
2054 #ifdef BNX2X_STOP_ON_ERROR
2055 if (unlikely(bp->panic))
2056 return NETDEV_TX_BUSY;
2059 fp_index = skb_get_queue_mapping(skb);
2060 txq = netdev_get_tx_queue(dev, fp_index);
2062 fp = &bp->fp[fp_index];
2064 if (unlikely(bnx2x_tx_avail(fp) < (skb_shinfo(skb)->nr_frags + 3))) {
2065 fp->eth_q_stats.driver_xoff++;
2066 netif_tx_stop_queue(txq);
2067 BNX2X_ERR("BUG! Tx ring full when queue awake!\n");
2068 return NETDEV_TX_BUSY;
2071 DP(NETIF_MSG_TX_QUEUED, "queue[%d]: SKB: summed %x protocol %x "
2072 "protocol(%x,%x) gso type %x xmit_type %x\n",
2073 fp_index, skb->ip_summed, skb->protocol, ipv6_hdr(skb)->nexthdr,
2074 ip_hdr(skb)->protocol, skb_shinfo(skb)->gso_type, xmit_type);
2076 eth = (struct ethhdr *)skb->data;
2078 /* set flag according to packet type (UNICAST_ADDRESS is default)*/
2079 if (unlikely(is_multicast_ether_addr(eth->h_dest))) {
2080 if (is_broadcast_ether_addr(eth->h_dest))
2081 mac_type = BROADCAST_ADDRESS;
2083 mac_type = MULTICAST_ADDRESS;
2086 #if (MAX_SKB_FRAGS >= MAX_FETCH_BD - 3)
2087 /* First, check if we need to linearize the skb (due to FW
2088 restrictions). No need to check fragmentation if page size > 8K
2089 (there will be no violation to FW restrictions) */
2090 if (bnx2x_pkt_req_lin(bp, skb, xmit_type)) {
2091 /* Statistics of linearization */
2093 if (skb_linearize(skb) != 0) {
2094 DP(NETIF_MSG_TX_QUEUED, "SKB linearization failed - "
2095 "silently dropping this SKB\n");
2096 dev_kfree_skb_any(skb);
2097 return NETDEV_TX_OK;
2103 Please read carefully. First we use one BD which we mark as start,
2104 then we have a parsing info BD (used for TSO or xsum),
2105 and only then we have the rest of the TSO BDs.
2106 (don't forget to mark the last one as last,
2107 and to unmap only AFTER you write to the BD ...)
2108 And above all, all pdb sizes are in words - NOT DWORDS!
2111 pkt_prod = fp->tx_pkt_prod++;
2112 bd_prod = TX_BD(fp->tx_bd_prod);
2114 /* get a tx_buf and first BD */
2115 tx_buf = &fp->tx_buf_ring[TX_BD(pkt_prod)];
2116 tx_start_bd = &fp->tx_desc_ring[bd_prod].start_bd;
2118 tx_start_bd->bd_flags.as_bitfield = ETH_TX_BD_FLAGS_START_BD;
2119 SET_FLAG(tx_start_bd->general_data, ETH_TX_START_BD_ETH_ADDR_TYPE,
2123 SET_FLAG(tx_start_bd->general_data, ETH_TX_START_BD_HDR_NBDS, 1);
2125 /* remember the first BD of the packet */
2126 tx_buf->first_bd = fp->tx_bd_prod;
2130 DP(NETIF_MSG_TX_QUEUED,
2131 "sending pkt %u @%p next_idx %u bd %u @%p\n",
2132 pkt_prod, tx_buf, fp->tx_pkt_prod, bd_prod, tx_start_bd);
2134 if (vlan_tx_tag_present(skb)) {
2135 tx_start_bd->vlan_or_ethertype =
2136 cpu_to_le16(vlan_tx_tag_get(skb));
2137 tx_start_bd->bd_flags.as_bitfield |=
2138 (X_ETH_OUTBAND_VLAN << ETH_TX_BD_FLAGS_VLAN_MODE_SHIFT);
2140 tx_start_bd->vlan_or_ethertype = cpu_to_le16(pkt_prod);
2142 /* turn on parsing and get a BD */
2143 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
2145 if (xmit_type & XMIT_CSUM) {
2146 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_L4_CSUM;
2148 if (xmit_type & XMIT_CSUM_V4)
2149 tx_start_bd->bd_flags.as_bitfield |=
2150 ETH_TX_BD_FLAGS_IP_CSUM;
2152 tx_start_bd->bd_flags.as_bitfield |=
2153 ETH_TX_BD_FLAGS_IPV6;
2155 if (!(xmit_type & XMIT_CSUM_TCP))
2156 tx_start_bd->bd_flags.as_bitfield |=
2157 ETH_TX_BD_FLAGS_IS_UDP;
2160 if (CHIP_IS_E2(bp)) {
2161 pbd_e2 = &fp->tx_desc_ring[bd_prod].parse_bd_e2;
2162 memset(pbd_e2, 0, sizeof(struct eth_tx_parse_bd_e2));
2163 /* Set PBD in checksum offload case */
2164 if (xmit_type & XMIT_CSUM)
2165 hlen = bnx2x_set_pbd_csum_e2(bp, skb,
2166 &pbd_e2_parsing_data,
2169 pbd_e1x = &fp->tx_desc_ring[bd_prod].parse_bd_e1x;
2170 memset(pbd_e1x, 0, sizeof(struct eth_tx_parse_bd_e1x));
2171 /* Set PBD in checksum offload case */
2172 if (xmit_type & XMIT_CSUM)
2173 hlen = bnx2x_set_pbd_csum(bp, skb, pbd_e1x, xmit_type);
2177 /* Map skb linear data for DMA */
2178 mapping = dma_map_single(&bp->pdev->dev, skb->data,
2179 skb_headlen(skb), DMA_TO_DEVICE);
2181 /* Setup the data pointer of the first BD of the packet */
2182 tx_start_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
2183 tx_start_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
2184 nbd = skb_shinfo(skb)->nr_frags + 2; /* start_bd + pbd + frags */
2185 tx_start_bd->nbd = cpu_to_le16(nbd);
2186 tx_start_bd->nbytes = cpu_to_le16(skb_headlen(skb));
2187 pkt_size = tx_start_bd->nbytes;
2189 DP(NETIF_MSG_TX_QUEUED, "first bd @%p addr (%x:%x) nbd %d"
2190 " nbytes %d flags %x vlan %x\n",
2191 tx_start_bd, tx_start_bd->addr_hi, tx_start_bd->addr_lo,
2192 le16_to_cpu(tx_start_bd->nbd), le16_to_cpu(tx_start_bd->nbytes),
2193 tx_start_bd->bd_flags.as_bitfield,
2194 le16_to_cpu(tx_start_bd->vlan_or_ethertype));
2196 if (xmit_type & XMIT_GSO) {
2198 DP(NETIF_MSG_TX_QUEUED,
2199 "TSO packet len %d hlen %d total len %d tso size %d\n",
2200 skb->len, hlen, skb_headlen(skb),
2201 skb_shinfo(skb)->gso_size);
2203 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_SW_LSO;
2205 if (unlikely(skb_headlen(skb) > hlen))
2206 bd_prod = bnx2x_tx_split(bp, fp, tx_buf, &tx_start_bd,
2207 hlen, bd_prod, ++nbd);
2209 bnx2x_set_pbd_gso_e2(skb, &pbd_e2_parsing_data,
2212 bnx2x_set_pbd_gso(skb, pbd_e1x, xmit_type);
2215 /* Set the PBD's parsing_data field if not zero
2216 * (for the chips newer than 57711).
2218 if (pbd_e2_parsing_data)
2219 pbd_e2->parsing_data = cpu_to_le32(pbd_e2_parsing_data);
2221 tx_data_bd = (struct eth_tx_bd *)tx_start_bd;
2223 /* Handle fragmented skb */
2224 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2225 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2227 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
2228 tx_data_bd = &fp->tx_desc_ring[bd_prod].reg_bd;
2229 if (total_pkt_bd == NULL)
2230 total_pkt_bd = &fp->tx_desc_ring[bd_prod].reg_bd;
2232 mapping = dma_map_page(&bp->pdev->dev, frag->page,
2234 frag->size, DMA_TO_DEVICE);
2236 tx_data_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
2237 tx_data_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
2238 tx_data_bd->nbytes = cpu_to_le16(frag->size);
2239 le16_add_cpu(&pkt_size, frag->size);
2241 DP(NETIF_MSG_TX_QUEUED,
2242 "frag %d bd @%p addr (%x:%x) nbytes %d\n",
2243 i, tx_data_bd, tx_data_bd->addr_hi, tx_data_bd->addr_lo,
2244 le16_to_cpu(tx_data_bd->nbytes));
2247 DP(NETIF_MSG_TX_QUEUED, "last bd @%p\n", tx_data_bd);
2249 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
2251 /* now send a tx doorbell, counting the next BD
2252 * if the packet contains or ends with it
2254 if (TX_BD_POFF(bd_prod) < nbd)
2257 if (total_pkt_bd != NULL)
2258 total_pkt_bd->total_pkt_bytes = pkt_size;
2261 DP(NETIF_MSG_TX_QUEUED,
2262 "PBD (E1X) @%p ip_data %x ip_hlen %u ip_id %u lso_mss %u"
2263 " tcp_flags %x xsum %x seq %u hlen %u\n",
2264 pbd_e1x, pbd_e1x->global_data, pbd_e1x->ip_hlen_w,
2265 pbd_e1x->ip_id, pbd_e1x->lso_mss, pbd_e1x->tcp_flags,
2266 pbd_e1x->tcp_pseudo_csum, pbd_e1x->tcp_send_seq,
2267 le16_to_cpu(pbd_e1x->total_hlen_w));
2269 DP(NETIF_MSG_TX_QUEUED,
2270 "PBD (E2) @%p dst %x %x %x src %x %x %x parsing_data %x\n",
2271 pbd_e2, pbd_e2->dst_mac_addr_hi, pbd_e2->dst_mac_addr_mid,
2272 pbd_e2->dst_mac_addr_lo, pbd_e2->src_mac_addr_hi,
2273 pbd_e2->src_mac_addr_mid, pbd_e2->src_mac_addr_lo,
2274 pbd_e2->parsing_data);
2275 DP(NETIF_MSG_TX_QUEUED, "doorbell: nbd %d bd %u\n", nbd, bd_prod);
2278 * Make sure that the BD data is updated before updating the producer
2279 * since FW might read the BD right after the producer is updated.
2280 * This is only applicable for weak-ordered memory model archs such
2281 * as IA-64. The following barrier is also mandatory since FW will
2282 * assumes packets must have BDs.
2286 fp->tx_db.data.prod += nbd;
2289 DOORBELL(bp, fp->cid, fp->tx_db.raw);
2293 fp->tx_bd_prod += nbd;
2295 if (unlikely(bnx2x_tx_avail(fp) < MAX_SKB_FRAGS + 3)) {
2296 netif_tx_stop_queue(txq);
2298 /* paired memory barrier is in bnx2x_tx_int(), we have to keep
2299 * ordering of set_bit() in netif_tx_stop_queue() and read of
2303 fp->eth_q_stats.driver_xoff++;
2304 if (bnx2x_tx_avail(fp) >= MAX_SKB_FRAGS + 3)
2305 netif_tx_wake_queue(txq);
2309 return NETDEV_TX_OK;
2312 /* called with rtnl_lock */
2313 int bnx2x_change_mac_addr(struct net_device *dev, void *p)
2315 struct sockaddr *addr = p;
2316 struct bnx2x *bp = netdev_priv(dev);
2318 if (!is_valid_ether_addr((u8 *)(addr->sa_data)))
2321 memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
2322 if (netif_running(dev))
2323 bnx2x_set_eth_mac(bp, 1);
2329 static int bnx2x_setup_irqs(struct bnx2x *bp)
2332 if (bp->flags & USING_MSIX_FLAG) {
2333 rc = bnx2x_req_msix_irqs(bp);
2338 rc = bnx2x_req_irq(bp);
2340 BNX2X_ERR("IRQ request failed rc %d, aborting\n", rc);
2343 if (bp->flags & USING_MSI_FLAG) {
2344 bp->dev->irq = bp->pdev->irq;
2345 netdev_info(bp->dev, "using MSI IRQ %d\n",
2353 void bnx2x_free_mem_bp(struct bnx2x *bp)
2356 kfree(bp->msix_table);
2360 int __devinit bnx2x_alloc_mem_bp(struct bnx2x *bp)
2362 struct bnx2x_fastpath *fp;
2363 struct msix_entry *tbl;
2364 struct bnx2x_ilt *ilt;
2367 fp = kzalloc(L2_FP_COUNT(bp->l2_cid_count)*sizeof(*fp), GFP_KERNEL);
2373 tbl = kzalloc((FP_SB_COUNT(bp->l2_cid_count) + 1) * sizeof(*tbl),
2377 bp->msix_table = tbl;
2380 ilt = kzalloc(sizeof(*ilt), GFP_KERNEL);
2387 bnx2x_free_mem_bp(bp);
2392 /* called with rtnl_lock */
2393 int bnx2x_change_mtu(struct net_device *dev, int new_mtu)
2395 struct bnx2x *bp = netdev_priv(dev);
2398 if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
2399 printk(KERN_ERR "Handling parity error recovery. Try again later\n");
2403 if ((new_mtu > ETH_MAX_JUMBO_PACKET_SIZE) ||
2404 ((new_mtu + ETH_HLEN) < ETH_MIN_PACKET_SIZE))
2407 /* This does not race with packet allocation
2408 * because the actual alloc size is
2409 * only updated as part of load
2413 if (netif_running(dev)) {
2414 bnx2x_nic_unload(bp, UNLOAD_NORMAL);
2415 rc = bnx2x_nic_load(bp, LOAD_NORMAL);
2421 void bnx2x_tx_timeout(struct net_device *dev)
2423 struct bnx2x *bp = netdev_priv(dev);
2425 #ifdef BNX2X_STOP_ON_ERROR
2429 /* This allows the netif to be shutdown gracefully before resetting */
2430 schedule_delayed_work(&bp->reset_task, 0);
2433 int bnx2x_suspend(struct pci_dev *pdev, pm_message_t state)
2435 struct net_device *dev = pci_get_drvdata(pdev);
2439 dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
2442 bp = netdev_priv(dev);
2446 pci_save_state(pdev);
2448 if (!netif_running(dev)) {
2453 netif_device_detach(dev);
2455 bnx2x_nic_unload(bp, UNLOAD_CLOSE);
2457 bnx2x_set_power_state(bp, pci_choose_state(pdev, state));
2464 int bnx2x_resume(struct pci_dev *pdev)
2466 struct net_device *dev = pci_get_drvdata(pdev);
2471 dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
2474 bp = netdev_priv(dev);
2476 if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
2477 printk(KERN_ERR "Handling parity error recovery. Try again later\n");
2483 pci_restore_state(pdev);
2485 if (!netif_running(dev)) {
2490 bnx2x_set_power_state(bp, PCI_D0);
2491 netif_device_attach(dev);
2493 /* Since the chip was reset, clear the FW sequence number */
2495 rc = bnx2x_nic_load(bp, LOAD_OPEN);