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>
21 #include <net/ip6_checksum.h>
22 #include <linux/firmware.h>
23 #include "bnx2x_cmn.h"
26 #include <linux/if_vlan.h>
29 #include "bnx2x_init.h"
32 /* free skb in the packet ring at pos idx
33 * return idx of last bd freed
35 static u16 bnx2x_free_tx_pkt(struct bnx2x *bp, struct bnx2x_fastpath *fp,
38 struct sw_tx_bd *tx_buf = &fp->tx_buf_ring[idx];
39 struct eth_tx_start_bd *tx_start_bd;
40 struct eth_tx_bd *tx_data_bd;
41 struct sk_buff *skb = tx_buf->skb;
42 u16 bd_idx = TX_BD(tx_buf->first_bd), new_cons;
45 /* prefetch skb end pointer to speedup dev_kfree_skb() */
48 DP(BNX2X_MSG_OFF, "pkt_idx %d buff @(%p)->skb %p\n",
52 DP(BNX2X_MSG_OFF, "free bd_idx %d\n", bd_idx);
53 tx_start_bd = &fp->tx_desc_ring[bd_idx].start_bd;
54 dma_unmap_single(&bp->pdev->dev, BD_UNMAP_ADDR(tx_start_bd),
55 BD_UNMAP_LEN(tx_start_bd), DMA_TO_DEVICE);
57 nbd = le16_to_cpu(tx_start_bd->nbd) - 1;
58 #ifdef BNX2X_STOP_ON_ERROR
59 if ((nbd - 1) > (MAX_SKB_FRAGS + 2)) {
60 BNX2X_ERR("BAD nbd!\n");
64 new_cons = nbd + tx_buf->first_bd;
67 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
69 /* Skip a parse bd... */
71 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
73 /* ...and the TSO split header bd since they have no mapping */
74 if (tx_buf->flags & BNX2X_TSO_SPLIT_BD) {
76 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
82 DP(BNX2X_MSG_OFF, "free frag bd_idx %d\n", bd_idx);
83 tx_data_bd = &fp->tx_desc_ring[bd_idx].reg_bd;
84 dma_unmap_page(&bp->pdev->dev, BD_UNMAP_ADDR(tx_data_bd),
85 BD_UNMAP_LEN(tx_data_bd), DMA_TO_DEVICE);
87 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
99 int bnx2x_tx_int(struct bnx2x_fastpath *fp)
101 struct bnx2x *bp = fp->bp;
102 struct netdev_queue *txq;
103 u16 hw_cons, sw_cons, bd_cons = fp->tx_bd_cons;
105 #ifdef BNX2X_STOP_ON_ERROR
106 if (unlikely(bp->panic))
110 txq = netdev_get_tx_queue(bp->dev, fp->index);
111 hw_cons = le16_to_cpu(*fp->tx_cons_sb);
112 sw_cons = fp->tx_pkt_cons;
114 while (sw_cons != hw_cons) {
117 pkt_cons = TX_BD(sw_cons);
119 DP(NETIF_MSG_TX_DONE, "queue[%d]: hw_cons %u sw_cons %u "
121 fp->index, hw_cons, sw_cons, pkt_cons);
123 bd_cons = bnx2x_free_tx_pkt(bp, fp, pkt_cons);
127 fp->tx_pkt_cons = sw_cons;
128 fp->tx_bd_cons = bd_cons;
130 /* Need to make the tx_bd_cons update visible to start_xmit()
131 * before checking for netif_tx_queue_stopped(). Without the
132 * memory barrier, there is a small possibility that
133 * start_xmit() will miss it and cause the queue to be stopped
138 if (unlikely(netif_tx_queue_stopped(txq))) {
139 /* Taking tx_lock() is needed to prevent reenabling the queue
140 * while it's empty. This could have happen if rx_action() gets
141 * suspended in bnx2x_tx_int() after the condition before
142 * netif_tx_wake_queue(), while tx_action (bnx2x_start_xmit()):
144 * stops the queue->sees fresh tx_bd_cons->releases the queue->
145 * sends some packets consuming the whole queue again->
149 __netif_tx_lock(txq, smp_processor_id());
151 if ((netif_tx_queue_stopped(txq)) &&
152 (bp->state == BNX2X_STATE_OPEN) &&
153 (bnx2x_tx_avail(fp) >= MAX_SKB_FRAGS + 3))
154 netif_tx_wake_queue(txq);
156 __netif_tx_unlock(txq);
161 static inline void bnx2x_update_last_max_sge(struct bnx2x_fastpath *fp,
164 u16 last_max = fp->last_max_sge;
166 if (SUB_S16(idx, last_max) > 0)
167 fp->last_max_sge = idx;
170 static void bnx2x_update_sge_prod(struct bnx2x_fastpath *fp,
171 struct eth_fast_path_rx_cqe *fp_cqe)
173 struct bnx2x *bp = fp->bp;
174 u16 sge_len = SGE_PAGE_ALIGN(le16_to_cpu(fp_cqe->pkt_len) -
175 le16_to_cpu(fp_cqe->len_on_bd)) >>
177 u16 last_max, last_elem, first_elem;
184 /* First mark all used pages */
185 for (i = 0; i < sge_len; i++)
186 SGE_MASK_CLEAR_BIT(fp,
187 RX_SGE(le16_to_cpu(fp_cqe->sgl_or_raw_data.sgl[i])));
189 DP(NETIF_MSG_RX_STATUS, "fp_cqe->sgl[%d] = %d\n",
190 sge_len - 1, le16_to_cpu(fp_cqe->sgl_or_raw_data.sgl[sge_len - 1]));
192 /* Here we assume that the last SGE index is the biggest */
193 prefetch((void *)(fp->sge_mask));
194 bnx2x_update_last_max_sge(fp,
195 le16_to_cpu(fp_cqe->sgl_or_raw_data.sgl[sge_len - 1]));
197 last_max = RX_SGE(fp->last_max_sge);
198 last_elem = last_max >> RX_SGE_MASK_ELEM_SHIFT;
199 first_elem = RX_SGE(fp->rx_sge_prod) >> RX_SGE_MASK_ELEM_SHIFT;
201 /* If ring is not full */
202 if (last_elem + 1 != first_elem)
205 /* Now update the prod */
206 for (i = first_elem; i != last_elem; i = NEXT_SGE_MASK_ELEM(i)) {
207 if (likely(fp->sge_mask[i]))
210 fp->sge_mask[i] = RX_SGE_MASK_ELEM_ONE_MASK;
211 delta += RX_SGE_MASK_ELEM_SZ;
215 fp->rx_sge_prod += delta;
216 /* clear page-end entries */
217 bnx2x_clear_sge_mask_next_elems(fp);
220 DP(NETIF_MSG_RX_STATUS,
221 "fp->last_max_sge = %d fp->rx_sge_prod = %d\n",
222 fp->last_max_sge, fp->rx_sge_prod);
225 static void bnx2x_tpa_start(struct bnx2x_fastpath *fp, u16 queue,
226 struct sk_buff *skb, u16 cons, u16 prod)
228 struct bnx2x *bp = fp->bp;
229 struct sw_rx_bd *cons_rx_buf = &fp->rx_buf_ring[cons];
230 struct sw_rx_bd *prod_rx_buf = &fp->rx_buf_ring[prod];
231 struct eth_rx_bd *prod_bd = &fp->rx_desc_ring[prod];
234 /* move empty skb from pool to prod and map it */
235 prod_rx_buf->skb = fp->tpa_pool[queue].skb;
236 mapping = dma_map_single(&bp->pdev->dev, fp->tpa_pool[queue].skb->data,
237 bp->rx_buf_size, DMA_FROM_DEVICE);
238 dma_unmap_addr_set(prod_rx_buf, mapping, mapping);
240 /* move partial skb from cons to pool (don't unmap yet) */
241 fp->tpa_pool[queue] = *cons_rx_buf;
243 /* mark bin state as start - print error if current state != stop */
244 if (fp->tpa_state[queue] != BNX2X_TPA_STOP)
245 BNX2X_ERR("start of bin not in stop [%d]\n", queue);
247 fp->tpa_state[queue] = BNX2X_TPA_START;
249 /* point prod_bd to new skb */
250 prod_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
251 prod_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
253 #ifdef BNX2X_STOP_ON_ERROR
254 fp->tpa_queue_used |= (1 << queue);
255 #ifdef _ASM_GENERIC_INT_L64_H
256 DP(NETIF_MSG_RX_STATUS, "fp->tpa_queue_used = 0x%lx\n",
258 DP(NETIF_MSG_RX_STATUS, "fp->tpa_queue_used = 0x%llx\n",
264 static int bnx2x_fill_frag_skb(struct bnx2x *bp, struct bnx2x_fastpath *fp,
266 struct eth_fast_path_rx_cqe *fp_cqe,
269 struct sw_rx_page *rx_pg, old_rx_pg;
270 u16 len_on_bd = le16_to_cpu(fp_cqe->len_on_bd);
271 u32 i, frag_len, frag_size, pages;
275 frag_size = le16_to_cpu(fp_cqe->pkt_len) - len_on_bd;
276 pages = SGE_PAGE_ALIGN(frag_size) >> SGE_PAGE_SHIFT;
278 /* This is needed in order to enable forwarding support */
280 skb_shinfo(skb)->gso_size = min((u32)SGE_PAGE_SIZE,
281 max(frag_size, (u32)len_on_bd));
283 #ifdef BNX2X_STOP_ON_ERROR
284 if (pages > min_t(u32, 8, MAX_SKB_FRAGS)*SGE_PAGE_SIZE*PAGES_PER_SGE) {
285 BNX2X_ERR("SGL length is too long: %d. CQE index is %d\n",
287 BNX2X_ERR("fp_cqe->pkt_len = %d fp_cqe->len_on_bd = %d\n",
288 fp_cqe->pkt_len, len_on_bd);
294 /* Run through the SGL and compose the fragmented skb */
295 for (i = 0, j = 0; i < pages; i += PAGES_PER_SGE, j++) {
297 RX_SGE(le16_to_cpu(fp_cqe->sgl_or_raw_data.sgl[j]));
299 /* FW gives the indices of the SGE as if the ring is an array
300 (meaning that "next" element will consume 2 indices) */
301 frag_len = min(frag_size, (u32)(SGE_PAGE_SIZE*PAGES_PER_SGE));
302 rx_pg = &fp->rx_page_ring[sge_idx];
305 /* If we fail to allocate a substitute page, we simply stop
306 where we are and drop the whole packet */
307 err = bnx2x_alloc_rx_sge(bp, fp, sge_idx);
309 fp->eth_q_stats.rx_skb_alloc_failed++;
313 /* Unmap the page as we r going to pass it to the stack */
314 dma_unmap_page(&bp->pdev->dev,
315 dma_unmap_addr(&old_rx_pg, mapping),
316 SGE_PAGE_SIZE*PAGES_PER_SGE, DMA_FROM_DEVICE);
318 /* Add one frag and update the appropriate fields in the skb */
319 skb_fill_page_desc(skb, j, old_rx_pg.page, 0, frag_len);
321 skb->data_len += frag_len;
322 skb->truesize += frag_len;
323 skb->len += frag_len;
325 frag_size -= frag_len;
331 static void bnx2x_tpa_stop(struct bnx2x *bp, struct bnx2x_fastpath *fp,
332 u16 queue, int pad, int len, union eth_rx_cqe *cqe,
335 struct sw_rx_bd *rx_buf = &fp->tpa_pool[queue];
336 struct sk_buff *skb = rx_buf->skb;
338 struct sk_buff *new_skb = netdev_alloc_skb(bp->dev, bp->rx_buf_size);
340 /* Unmap skb in the pool anyway, as we are going to change
341 pool entry status to BNX2X_TPA_STOP even if new skb allocation
343 dma_unmap_single(&bp->pdev->dev, dma_unmap_addr(rx_buf, mapping),
344 bp->rx_buf_size, DMA_FROM_DEVICE);
346 if (likely(new_skb)) {
347 /* fix ip xsum and give it to the stack */
348 /* (no need to map the new skb) */
351 (le16_to_cpu(cqe->fast_path_cqe.pars_flags.flags) &
353 int is_not_hwaccel_vlan_cqe =
354 (is_vlan_cqe && (!(bp->flags & HW_VLAN_RX_FLAG)));
358 prefetch(((char *)(skb)) + L1_CACHE_BYTES);
360 #ifdef BNX2X_STOP_ON_ERROR
361 if (pad + len > bp->rx_buf_size) {
362 BNX2X_ERR("skb_put is about to fail... "
363 "pad %d len %d rx_buf_size %d\n",
364 pad, len, bp->rx_buf_size);
370 skb_reserve(skb, pad);
373 skb->protocol = eth_type_trans(skb, bp->dev);
374 skb->ip_summed = CHECKSUM_UNNECESSARY;
379 iph = (struct iphdr *)skb->data;
381 /* If there is no Rx VLAN offloading -
382 take VLAN tag into an account */
383 if (unlikely(is_not_hwaccel_vlan_cqe))
384 iph = (struct iphdr *)((u8 *)iph + VLAN_HLEN);
387 iph->check = ip_fast_csum((u8 *)iph, iph->ihl);
390 if (!bnx2x_fill_frag_skb(bp, fp, skb,
391 &cqe->fast_path_cqe, cqe_idx)) {
393 if ((bp->vlgrp != NULL) &&
394 (le16_to_cpu(cqe->fast_path_cqe.
395 pars_flags.flags) & PARSING_FLAGS_VLAN))
396 vlan_gro_receive(&fp->napi, bp->vlgrp,
397 le16_to_cpu(cqe->fast_path_cqe.
401 napi_gro_receive(&fp->napi, skb);
403 DP(NETIF_MSG_RX_STATUS, "Failed to allocate new pages"
404 " - dropping packet!\n");
409 /* put new skb in bin */
410 fp->tpa_pool[queue].skb = new_skb;
413 /* else drop the packet and keep the buffer in the bin */
414 DP(NETIF_MSG_RX_STATUS,
415 "Failed to allocate new skb - dropping packet!\n");
416 fp->eth_q_stats.rx_skb_alloc_failed++;
419 fp->tpa_state[queue] = BNX2X_TPA_STOP;
422 /* Set Toeplitz hash value in the skb using the value from the
423 * CQE (calculated by HW).
425 static inline void bnx2x_set_skb_rxhash(struct bnx2x *bp, union eth_rx_cqe *cqe,
428 /* Set Toeplitz hash from CQE */
429 if ((bp->dev->features & NETIF_F_RXHASH) &&
430 (cqe->fast_path_cqe.status_flags &
431 ETH_FAST_PATH_RX_CQE_RSS_HASH_FLG))
433 le32_to_cpu(cqe->fast_path_cqe.rss_hash_result);
436 int bnx2x_rx_int(struct bnx2x_fastpath *fp, int budget)
438 struct bnx2x *bp = fp->bp;
439 u16 bd_cons, bd_prod, bd_prod_fw, comp_ring_cons;
440 u16 hw_comp_cons, sw_comp_cons, sw_comp_prod;
443 #ifdef BNX2X_STOP_ON_ERROR
444 if (unlikely(bp->panic))
448 /* CQ "next element" is of the size of the regular element,
449 that's why it's ok here */
450 hw_comp_cons = le16_to_cpu(*fp->rx_cons_sb);
451 if ((hw_comp_cons & MAX_RCQ_DESC_CNT) == MAX_RCQ_DESC_CNT)
454 bd_cons = fp->rx_bd_cons;
455 bd_prod = fp->rx_bd_prod;
456 bd_prod_fw = bd_prod;
457 sw_comp_cons = fp->rx_comp_cons;
458 sw_comp_prod = fp->rx_comp_prod;
460 /* Memory barrier necessary as speculative reads of the rx
461 * buffer can be ahead of the index in the status block
465 DP(NETIF_MSG_RX_STATUS,
466 "queue[%d]: hw_comp_cons %u sw_comp_cons %u\n",
467 fp->index, hw_comp_cons, sw_comp_cons);
469 while (sw_comp_cons != hw_comp_cons) {
470 struct sw_rx_bd *rx_buf = NULL;
472 union eth_rx_cqe *cqe;
476 comp_ring_cons = RCQ_BD(sw_comp_cons);
477 bd_prod = RX_BD(bd_prod);
478 bd_cons = RX_BD(bd_cons);
480 /* Prefetch the page containing the BD descriptor
481 at producer's index. It will be needed when new skb is
483 prefetch((void *)(PAGE_ALIGN((unsigned long)
484 (&fp->rx_desc_ring[bd_prod])) -
487 cqe = &fp->rx_comp_ring[comp_ring_cons];
488 cqe_fp_flags = cqe->fast_path_cqe.type_error_flags;
490 DP(NETIF_MSG_RX_STATUS, "CQE type %x err %x status %x"
491 " queue %x vlan %x len %u\n", CQE_TYPE(cqe_fp_flags),
492 cqe_fp_flags, cqe->fast_path_cqe.status_flags,
493 le32_to_cpu(cqe->fast_path_cqe.rss_hash_result),
494 le16_to_cpu(cqe->fast_path_cqe.vlan_tag),
495 le16_to_cpu(cqe->fast_path_cqe.pkt_len));
497 /* is this a slowpath msg? */
498 if (unlikely(CQE_TYPE(cqe_fp_flags))) {
499 bnx2x_sp_event(fp, cqe);
502 /* this is an rx packet */
504 rx_buf = &fp->rx_buf_ring[bd_cons];
507 len = le16_to_cpu(cqe->fast_path_cqe.pkt_len);
508 pad = cqe->fast_path_cqe.placement_offset;
510 /* - If CQE is marked both TPA_START and TPA_END it is
512 * - FP CQE will always have either TPA_START or/and
513 * TPA_STOP flags set.
515 if ((!fp->disable_tpa) &&
516 (TPA_TYPE(cqe_fp_flags) !=
517 (TPA_TYPE_START | TPA_TYPE_END))) {
518 u16 queue = cqe->fast_path_cqe.queue_index;
520 if (TPA_TYPE(cqe_fp_flags) == TPA_TYPE_START) {
521 DP(NETIF_MSG_RX_STATUS,
522 "calling tpa_start on queue %d\n",
525 bnx2x_tpa_start(fp, queue, skb,
528 /* Set Toeplitz hash for an LRO skb */
529 bnx2x_set_skb_rxhash(bp, cqe, skb);
532 } else { /* TPA_STOP */
533 DP(NETIF_MSG_RX_STATUS,
534 "calling tpa_stop on queue %d\n",
537 if (!BNX2X_RX_SUM_FIX(cqe))
538 BNX2X_ERR("STOP on none TCP "
541 /* This is a size of the linear data
543 len = le16_to_cpu(cqe->fast_path_cqe.
545 bnx2x_tpa_stop(bp, fp, queue, pad,
546 len, cqe, comp_ring_cons);
547 #ifdef BNX2X_STOP_ON_ERROR
552 bnx2x_update_sge_prod(fp,
553 &cqe->fast_path_cqe);
558 dma_sync_single_for_device(&bp->pdev->dev,
559 dma_unmap_addr(rx_buf, mapping),
560 pad + RX_COPY_THRESH,
562 prefetch(((char *)(skb)) + L1_CACHE_BYTES);
564 /* is this an error packet? */
565 if (unlikely(cqe_fp_flags & ETH_RX_ERROR_FALGS)) {
567 "ERROR flags %x rx packet %u\n",
568 cqe_fp_flags, sw_comp_cons);
569 fp->eth_q_stats.rx_err_discard_pkt++;
573 /* Since we don't have a jumbo ring
574 * copy small packets if mtu > 1500
576 if ((bp->dev->mtu > ETH_MAX_PACKET_SIZE) &&
577 (len <= RX_COPY_THRESH)) {
578 struct sk_buff *new_skb;
580 new_skb = netdev_alloc_skb(bp->dev,
582 if (new_skb == NULL) {
584 "ERROR packet dropped "
585 "because of alloc failure\n");
586 fp->eth_q_stats.rx_skb_alloc_failed++;
591 skb_copy_from_linear_data_offset(skb, pad,
592 new_skb->data + pad, len);
593 skb_reserve(new_skb, pad);
594 skb_put(new_skb, len);
596 bnx2x_reuse_rx_skb(fp, bd_cons, bd_prod);
601 if (likely(bnx2x_alloc_rx_skb(bp, fp, bd_prod) == 0)) {
602 dma_unmap_single(&bp->pdev->dev,
603 dma_unmap_addr(rx_buf, mapping),
606 skb_reserve(skb, pad);
611 "ERROR packet dropped because "
612 "of alloc failure\n");
613 fp->eth_q_stats.rx_skb_alloc_failed++;
615 bnx2x_reuse_rx_skb(fp, bd_cons, bd_prod);
619 skb->protocol = eth_type_trans(skb, bp->dev);
621 /* Set Toeplitz hash for a none-LRO skb */
622 bnx2x_set_skb_rxhash(bp, cqe, skb);
624 skb_checksum_none_assert(skb);
627 if (likely(BNX2X_RX_CSUM_OK(cqe)))
628 skb->ip_summed = CHECKSUM_UNNECESSARY;
630 fp->eth_q_stats.hw_csum_err++;
634 skb_record_rx_queue(skb, fp->index);
637 if ((bp->vlgrp != NULL) && (bp->flags & HW_VLAN_RX_FLAG) &&
638 (le16_to_cpu(cqe->fast_path_cqe.pars_flags.flags) &
640 vlan_gro_receive(&fp->napi, bp->vlgrp,
641 le16_to_cpu(cqe->fast_path_cqe.vlan_tag), skb);
644 napi_gro_receive(&fp->napi, skb);
650 bd_cons = NEXT_RX_IDX(bd_cons);
651 bd_prod = NEXT_RX_IDX(bd_prod);
652 bd_prod_fw = NEXT_RX_IDX(bd_prod_fw);
655 sw_comp_prod = NEXT_RCQ_IDX(sw_comp_prod);
656 sw_comp_cons = NEXT_RCQ_IDX(sw_comp_cons);
658 if (rx_pkt == budget)
662 fp->rx_bd_cons = bd_cons;
663 fp->rx_bd_prod = bd_prod_fw;
664 fp->rx_comp_cons = sw_comp_cons;
665 fp->rx_comp_prod = sw_comp_prod;
667 /* Update producers */
668 bnx2x_update_rx_prod(bp, fp, bd_prod_fw, sw_comp_prod,
671 fp->rx_pkt += rx_pkt;
677 static irqreturn_t bnx2x_msix_fp_int(int irq, void *fp_cookie)
679 struct bnx2x_fastpath *fp = fp_cookie;
680 struct bnx2x *bp = fp->bp;
682 /* Return here if interrupt is disabled */
683 if (unlikely(atomic_read(&bp->intr_sem) != 0)) {
684 DP(NETIF_MSG_INTR, "called but intr_sem not 0, returning\n");
688 DP(BNX2X_MSG_FP, "got an MSI-X interrupt on IDX:SB "
689 "[fp %d fw_sd %d igusb %d]\n",
690 fp->index, fp->fw_sb_id, fp->igu_sb_id);
691 bnx2x_ack_sb(bp, fp->igu_sb_id, USTORM_ID, 0, IGU_INT_DISABLE, 0);
693 #ifdef BNX2X_STOP_ON_ERROR
694 if (unlikely(bp->panic))
698 /* Handle Rx and Tx according to MSI-X vector */
699 prefetch(fp->rx_cons_sb);
700 prefetch(fp->tx_cons_sb);
701 prefetch(&fp->sb_running_index[SM_RX_ID]);
702 napi_schedule(&bnx2x_fp(bp, fp->index, napi));
707 /* HW Lock for shared dual port PHYs */
708 void bnx2x_acquire_phy_lock(struct bnx2x *bp)
710 mutex_lock(&bp->port.phy_mutex);
712 if (bp->port.need_hw_lock)
713 bnx2x_acquire_hw_lock(bp, HW_LOCK_RESOURCE_MDIO);
716 void bnx2x_release_phy_lock(struct bnx2x *bp)
718 if (bp->port.need_hw_lock)
719 bnx2x_release_hw_lock(bp, HW_LOCK_RESOURCE_MDIO);
721 mutex_unlock(&bp->port.phy_mutex);
724 void bnx2x_link_report(struct bnx2x *bp)
726 if (bp->flags & MF_FUNC_DIS) {
727 netif_carrier_off(bp->dev);
728 netdev_err(bp->dev, "NIC Link is Down\n");
732 if (bp->link_vars.link_up) {
735 if (bp->state == BNX2X_STATE_OPEN)
736 netif_carrier_on(bp->dev);
737 netdev_info(bp->dev, "NIC Link is Up, ");
739 line_speed = bp->link_vars.line_speed;
744 ((bp->mf_config[BP_VN(bp)] &
745 FUNC_MF_CFG_MAX_BW_MASK) >>
746 FUNC_MF_CFG_MAX_BW_SHIFT) * 100;
747 if (vn_max_rate < line_speed)
748 line_speed = vn_max_rate;
750 pr_cont("%d Mbps ", line_speed);
752 if (bp->link_vars.duplex == DUPLEX_FULL)
753 pr_cont("full duplex");
755 pr_cont("half duplex");
757 if (bp->link_vars.flow_ctrl != BNX2X_FLOW_CTRL_NONE) {
758 if (bp->link_vars.flow_ctrl & BNX2X_FLOW_CTRL_RX) {
759 pr_cont(", receive ");
760 if (bp->link_vars.flow_ctrl &
762 pr_cont("& transmit ");
764 pr_cont(", transmit ");
766 pr_cont("flow control ON");
770 } else { /* link_down */
771 netif_carrier_off(bp->dev);
772 netdev_err(bp->dev, "NIC Link is Down\n");
776 /* Returns the number of actually allocated BDs */
777 static inline int bnx2x_alloc_rx_bds(struct bnx2x_fastpath *fp,
780 struct bnx2x *bp = fp->bp;
781 u16 ring_prod, cqe_ring_prod;
784 fp->rx_comp_cons = 0;
785 cqe_ring_prod = ring_prod = 0;
786 for (i = 0; i < rx_ring_size; i++) {
787 if (bnx2x_alloc_rx_skb(bp, fp, ring_prod) < 0) {
788 BNX2X_ERR("was only able to allocate "
789 "%d rx skbs on queue[%d]\n", i, fp->index);
790 fp->eth_q_stats.rx_skb_alloc_failed++;
793 ring_prod = NEXT_RX_IDX(ring_prod);
794 cqe_ring_prod = NEXT_RCQ_IDX(cqe_ring_prod);
795 WARN_ON(ring_prod <= i);
798 fp->rx_bd_prod = ring_prod;
799 /* Limit the CQE producer by the CQE ring size */
800 fp->rx_comp_prod = min_t(u16, NUM_RCQ_RINGS*RCQ_DESC_CNT,
802 fp->rx_pkt = fp->rx_calls = 0;
807 static inline void bnx2x_alloc_rx_bd_ring(struct bnx2x_fastpath *fp)
809 struct bnx2x *bp = fp->bp;
810 int rx_ring_size = bp->rx_ring_size ? bp->rx_ring_size :
811 MAX_RX_AVAIL/bp->num_queues;
813 rx_ring_size = max_t(int, MIN_RX_AVAIL, rx_ring_size);
815 bnx2x_alloc_rx_bds(fp, rx_ring_size);
818 * this will generate an interrupt (to the TSTORM)
819 * must only be done after chip is initialized
821 bnx2x_update_rx_prod(bp, fp, fp->rx_bd_prod, fp->rx_comp_prod,
825 void bnx2x_init_rx_rings(struct bnx2x *bp)
827 int func = BP_FUNC(bp);
828 int max_agg_queues = CHIP_IS_E1(bp) ? ETH_MAX_AGGREGATION_QUEUES_E1 :
829 ETH_MAX_AGGREGATION_QUEUES_E1H;
833 bp->rx_buf_size = bp->dev->mtu + ETH_OVREHEAD + BNX2X_RX_ALIGN +
834 BNX2X_FW_IP_HDR_ALIGN_PAD;
837 "mtu %d rx_buf_size %d\n", bp->dev->mtu, bp->rx_buf_size);
839 for_each_queue(bp, j) {
840 struct bnx2x_fastpath *fp = &bp->fp[j];
842 if (!fp->disable_tpa) {
843 for (i = 0; i < max_agg_queues; i++) {
844 fp->tpa_pool[i].skb =
845 netdev_alloc_skb(bp->dev, bp->rx_buf_size);
846 if (!fp->tpa_pool[i].skb) {
847 BNX2X_ERR("Failed to allocate TPA "
848 "skb pool for queue[%d] - "
849 "disabling TPA on this "
851 bnx2x_free_tpa_pool(bp, fp, i);
855 dma_unmap_addr_set((struct sw_rx_bd *)
856 &bp->fp->tpa_pool[i],
858 fp->tpa_state[i] = BNX2X_TPA_STOP;
861 /* "next page" elements initialization */
862 bnx2x_set_next_page_sgl(fp);
864 /* set SGEs bit mask */
865 bnx2x_init_sge_ring_bit_mask(fp);
867 /* Allocate SGEs and initialize the ring elements */
868 for (i = 0, ring_prod = 0;
869 i < MAX_RX_SGE_CNT*NUM_RX_SGE_PAGES; i++) {
871 if (bnx2x_alloc_rx_sge(bp, fp, ring_prod) < 0) {
872 BNX2X_ERR("was only able to allocate "
874 BNX2X_ERR("disabling TPA for"
876 /* Cleanup already allocated elements */
877 bnx2x_free_rx_sge_range(bp,
879 bnx2x_free_tpa_pool(bp,
885 ring_prod = NEXT_SGE_IDX(ring_prod);
888 fp->rx_sge_prod = ring_prod;
892 for_each_queue(bp, j) {
893 struct bnx2x_fastpath *fp = &bp->fp[j];
897 bnx2x_set_next_page_rx_bd(fp);
900 bnx2x_set_next_page_rx_cq(fp);
902 /* Allocate BDs and initialize BD ring */
903 bnx2x_alloc_rx_bd_ring(fp);
908 if (!CHIP_IS_E2(bp)) {
909 REG_WR(bp, BAR_USTRORM_INTMEM +
910 USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func),
911 U64_LO(fp->rx_comp_mapping));
912 REG_WR(bp, BAR_USTRORM_INTMEM +
913 USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func) + 4,
914 U64_HI(fp->rx_comp_mapping));
919 static void bnx2x_free_tx_skbs(struct bnx2x *bp)
923 for_each_queue(bp, i) {
924 struct bnx2x_fastpath *fp = &bp->fp[i];
926 u16 bd_cons = fp->tx_bd_cons;
927 u16 sw_prod = fp->tx_pkt_prod;
928 u16 sw_cons = fp->tx_pkt_cons;
930 while (sw_cons != sw_prod) {
931 bd_cons = bnx2x_free_tx_pkt(bp, fp, TX_BD(sw_cons));
937 static void bnx2x_free_rx_skbs(struct bnx2x *bp)
941 for_each_queue(bp, j) {
942 struct bnx2x_fastpath *fp = &bp->fp[j];
944 for (i = 0; i < NUM_RX_BD; i++) {
945 struct sw_rx_bd *rx_buf = &fp->rx_buf_ring[i];
946 struct sk_buff *skb = rx_buf->skb;
951 dma_unmap_single(&bp->pdev->dev,
952 dma_unmap_addr(rx_buf, mapping),
953 bp->rx_buf_size, DMA_FROM_DEVICE);
958 if (!fp->disable_tpa)
959 bnx2x_free_tpa_pool(bp, fp, CHIP_IS_E1(bp) ?
960 ETH_MAX_AGGREGATION_QUEUES_E1 :
961 ETH_MAX_AGGREGATION_QUEUES_E1H);
965 void bnx2x_free_skbs(struct bnx2x *bp)
967 bnx2x_free_tx_skbs(bp);
968 bnx2x_free_rx_skbs(bp);
971 static void bnx2x_free_msix_irqs(struct bnx2x *bp)
975 free_irq(bp->msix_table[0].vector, bp->dev);
976 DP(NETIF_MSG_IFDOWN, "released sp irq (%d)\n",
977 bp->msix_table[0].vector);
982 for_each_queue(bp, i) {
983 DP(NETIF_MSG_IFDOWN, "about to release fp #%d->%d irq "
984 "state %x\n", i, bp->msix_table[i + offset].vector,
985 bnx2x_fp(bp, i, state));
987 free_irq(bp->msix_table[i + offset].vector, &bp->fp[i]);
991 void bnx2x_free_irq(struct bnx2x *bp)
993 if (bp->flags & USING_MSIX_FLAG)
994 bnx2x_free_msix_irqs(bp);
995 else if (bp->flags & USING_MSI_FLAG)
996 free_irq(bp->pdev->irq, bp->dev);
998 free_irq(bp->pdev->irq, bp->dev);
1001 int bnx2x_enable_msix(struct bnx2x *bp)
1003 int msix_vec = 0, i, rc, req_cnt;
1005 bp->msix_table[msix_vec].entry = msix_vec;
1006 DP(NETIF_MSG_IFUP, "msix_table[0].entry = %d (slowpath)\n",
1007 bp->msix_table[0].entry);
1011 bp->msix_table[msix_vec].entry = msix_vec;
1012 DP(NETIF_MSG_IFUP, "msix_table[%d].entry = %d (CNIC)\n",
1013 bp->msix_table[msix_vec].entry, bp->msix_table[msix_vec].entry);
1016 for_each_queue(bp, i) {
1017 bp->msix_table[msix_vec].entry = msix_vec;
1018 DP(NETIF_MSG_IFUP, "msix_table[%d].entry = %d "
1019 "(fastpath #%u)\n", msix_vec, msix_vec, i);
1023 req_cnt = BNX2X_NUM_QUEUES(bp) + CNIC_CONTEXT_USE + 1;
1025 rc = pci_enable_msix(bp->pdev, &bp->msix_table[0], req_cnt);
1028 * reconfigure number of tx/rx queues according to available
1031 if (rc >= BNX2X_MIN_MSIX_VEC_CNT) {
1032 /* how less vectors we will have? */
1033 int diff = req_cnt - rc;
1036 "Trying to use less MSI-X vectors: %d\n", rc);
1038 rc = pci_enable_msix(bp->pdev, &bp->msix_table[0], rc);
1042 "MSI-X is not attainable rc %d\n", rc);
1046 * decrease number of queues by number of unallocated entries
1048 bp->num_queues -= diff;
1050 DP(NETIF_MSG_IFUP, "New queue configuration set: %d\n",
1053 /* fall to INTx if not enough memory */
1055 bp->flags |= DISABLE_MSI_FLAG;
1056 DP(NETIF_MSG_IFUP, "MSI-X is not attainable rc %d\n", rc);
1060 bp->flags |= USING_MSIX_FLAG;
1065 static int bnx2x_req_msix_irqs(struct bnx2x *bp)
1067 int i, rc, offset = 1;
1069 rc = request_irq(bp->msix_table[0].vector, bnx2x_msix_sp_int, 0,
1070 bp->dev->name, bp->dev);
1072 BNX2X_ERR("request sp irq failed\n");
1079 for_each_queue(bp, i) {
1080 struct bnx2x_fastpath *fp = &bp->fp[i];
1081 snprintf(fp->name, sizeof(fp->name), "%s-fp-%d",
1084 rc = request_irq(bp->msix_table[offset].vector,
1085 bnx2x_msix_fp_int, 0, fp->name, fp);
1087 BNX2X_ERR("request fp #%d irq failed rc %d\n", i, rc);
1088 bnx2x_free_msix_irqs(bp);
1093 fp->state = BNX2X_FP_STATE_IRQ;
1096 i = BNX2X_NUM_QUEUES(bp);
1097 offset = 1 + CNIC_CONTEXT_USE;
1098 netdev_info(bp->dev, "using MSI-X IRQs: sp %d fp[%d] %d"
1100 bp->msix_table[0].vector,
1101 0, bp->msix_table[offset].vector,
1102 i - 1, bp->msix_table[offset + i - 1].vector);
1107 int bnx2x_enable_msi(struct bnx2x *bp)
1111 rc = pci_enable_msi(bp->pdev);
1113 DP(NETIF_MSG_IFUP, "MSI is not attainable\n");
1116 bp->flags |= USING_MSI_FLAG;
1121 static int bnx2x_req_irq(struct bnx2x *bp)
1123 unsigned long flags;
1126 if (bp->flags & USING_MSI_FLAG)
1129 flags = IRQF_SHARED;
1131 rc = request_irq(bp->pdev->irq, bnx2x_interrupt, flags,
1132 bp->dev->name, bp->dev);
1134 bnx2x_fp(bp, 0, state) = BNX2X_FP_STATE_IRQ;
1139 static void bnx2x_napi_enable(struct bnx2x *bp)
1143 for_each_queue(bp, i)
1144 napi_enable(&bnx2x_fp(bp, i, napi));
1147 static void bnx2x_napi_disable(struct bnx2x *bp)
1151 for_each_queue(bp, i)
1152 napi_disable(&bnx2x_fp(bp, i, napi));
1155 void bnx2x_netif_start(struct bnx2x *bp)
1159 intr_sem = atomic_dec_and_test(&bp->intr_sem);
1160 smp_wmb(); /* Ensure that bp->intr_sem update is SMP-safe */
1163 if (netif_running(bp->dev)) {
1164 bnx2x_napi_enable(bp);
1165 bnx2x_int_enable(bp);
1166 if (bp->state == BNX2X_STATE_OPEN)
1167 netif_tx_wake_all_queues(bp->dev);
1172 void bnx2x_netif_stop(struct bnx2x *bp, int disable_hw)
1174 bnx2x_int_disable_sync(bp, disable_hw);
1175 bnx2x_napi_disable(bp);
1176 netif_tx_disable(bp->dev);
1179 void bnx2x_set_num_queues(struct bnx2x *bp)
1181 switch (bp->multi_mode) {
1182 case ETH_RSS_MODE_DISABLED:
1185 case ETH_RSS_MODE_REGULAR:
1186 bp->num_queues = bnx2x_calc_num_queues(bp);
1195 static void bnx2x_release_firmware(struct bnx2x *bp)
1197 kfree(bp->init_ops_offsets);
1198 kfree(bp->init_ops);
1199 kfree(bp->init_data);
1200 release_firmware(bp->firmware);
1203 /* must be called with rtnl_lock */
1204 int bnx2x_nic_load(struct bnx2x *bp, int load_mode)
1209 /* Set init arrays */
1210 rc = bnx2x_init_firmware(bp);
1212 BNX2X_ERR("Error loading firmware\n");
1216 #ifdef BNX2X_STOP_ON_ERROR
1217 if (unlikely(bp->panic))
1221 bp->state = BNX2X_STATE_OPENING_WAIT4_LOAD;
1223 /* must be called before memory allocation and HW init */
1224 bnx2x_ilt_set_info(bp);
1226 if (bnx2x_alloc_mem(bp))
1229 netif_set_real_num_tx_queues(bp->dev, bp->num_queues);
1230 rc = netif_set_real_num_rx_queues(bp->dev, bp->num_queues);
1232 BNX2X_ERR("Unable to update real_num_rx_queues\n");
1236 for_each_queue(bp, i)
1237 bnx2x_fp(bp, i, disable_tpa) =
1238 ((bp->flags & TPA_ENABLE_FLAG) == 0);
1240 bnx2x_napi_enable(bp);
1242 /* Send LOAD_REQUEST command to MCP
1243 Returns the type of LOAD command:
1244 if it is the first port to be initialized
1245 common blocks should be initialized, otherwise - not
1247 if (!BP_NOMCP(bp)) {
1248 load_code = bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_REQ, 0);
1250 BNX2X_ERR("MCP response failure, aborting\n");
1254 if (load_code == FW_MSG_CODE_DRV_LOAD_REFUSED) {
1255 rc = -EBUSY; /* other port in diagnostic mode */
1260 int path = BP_PATH(bp);
1261 int port = BP_PORT(bp);
1263 DP(NETIF_MSG_IFUP, "NO MCP - load counts[%d] %d, %d, %d\n",
1264 path, load_count[path][0], load_count[path][1],
1265 load_count[path][2]);
1266 load_count[path][0]++;
1267 load_count[path][1 + port]++;
1268 DP(NETIF_MSG_IFUP, "NO MCP - new load counts[%d] %d, %d, %d\n",
1269 path, load_count[path][0], load_count[path][1],
1270 load_count[path][2]);
1271 if (load_count[path][0] == 1)
1272 load_code = FW_MSG_CODE_DRV_LOAD_COMMON;
1273 else if (load_count[path][1 + port] == 1)
1274 load_code = FW_MSG_CODE_DRV_LOAD_PORT;
1276 load_code = FW_MSG_CODE_DRV_LOAD_FUNCTION;
1279 if ((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
1280 (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP) ||
1281 (load_code == FW_MSG_CODE_DRV_LOAD_PORT))
1285 DP(NETIF_MSG_LINK, "pmf %d\n", bp->port.pmf);
1288 rc = bnx2x_init_hw(bp, load_code);
1290 BNX2X_ERR("HW init failed, aborting\n");
1291 bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
1292 bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_REQ_WOL_MCP, 0);
1293 bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_DONE, 0);
1297 /* Connect to IRQs */
1298 rc = bnx2x_setup_irqs(bp);
1300 bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
1304 /* Setup NIC internals and enable interrupts */
1305 bnx2x_nic_init(bp, load_code);
1307 if (((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
1308 (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP)) &&
1309 (bp->common.shmem2_base))
1310 SHMEM2_WR(bp, dcc_support,
1311 (SHMEM_DCC_SUPPORT_DISABLE_ENABLE_PF_TLV |
1312 SHMEM_DCC_SUPPORT_BANDWIDTH_ALLOCATION_TLV));
1314 /* Send LOAD_DONE command to MCP */
1315 if (!BP_NOMCP(bp)) {
1316 load_code = bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
1318 BNX2X_ERR("MCP response failure, aborting\n");
1324 bp->state = BNX2X_STATE_OPENING_WAIT4_PORT;
1326 rc = bnx2x_func_start(bp);
1328 BNX2X_ERR("Function start failed!\n");
1329 #ifndef BNX2X_STOP_ON_ERROR
1337 rc = bnx2x_setup_client(bp, &bp->fp[0], 1 /* Leading */);
1339 BNX2X_ERR("Setup leading failed!\n");
1340 #ifndef BNX2X_STOP_ON_ERROR
1348 if (!CHIP_IS_E1(bp) &&
1349 (bp->mf_config[BP_VN(bp)] & FUNC_MF_CFG_FUNC_DISABLED)) {
1350 DP(NETIF_MSG_IFUP, "mf_cfg function disabled\n");
1351 bp->flags |= MF_FUNC_DIS;
1355 /* Enable Timer scan */
1356 REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + BP_PORT(bp)*4, 1);
1359 for_each_nondefault_queue(bp, i) {
1360 rc = bnx2x_setup_client(bp, &bp->fp[i], 0);
1369 /* Now when Clients are configured we are ready to work */
1370 bp->state = BNX2X_STATE_OPEN;
1372 bnx2x_set_eth_mac(bp, 1);
1375 bnx2x_initial_phy_init(bp, load_mode);
1377 /* Start fast path */
1378 switch (load_mode) {
1380 /* Tx queue should be only reenabled */
1381 netif_tx_wake_all_queues(bp->dev);
1382 /* Initialize the receive filter. */
1383 bnx2x_set_rx_mode(bp->dev);
1387 netif_tx_start_all_queues(bp->dev);
1388 smp_mb__after_clear_bit();
1389 /* Initialize the receive filter. */
1390 bnx2x_set_rx_mode(bp->dev);
1394 /* Initialize the receive filter. */
1395 bnx2x_set_rx_mode(bp->dev);
1396 bp->state = BNX2X_STATE_DIAG;
1404 bnx2x__link_status_update(bp);
1406 /* start the timer */
1407 mod_timer(&bp->timer, jiffies + bp->current_interval);
1410 bnx2x_setup_cnic_irq_info(bp);
1411 if (bp->state == BNX2X_STATE_OPEN)
1412 bnx2x_cnic_notify(bp, CNIC_CTL_START_CMD);
1414 bnx2x_inc_load_cnt(bp);
1416 bnx2x_release_firmware(bp);
1422 /* Disable Timer scan */
1423 REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + BP_PORT(bp)*4, 0);
1426 bnx2x_int_disable_sync(bp, 1);
1428 /* Free SKBs, SGEs, TPA pool and driver internals */
1429 bnx2x_free_skbs(bp);
1430 for_each_queue(bp, i)
1431 bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
1436 if (!BP_NOMCP(bp)) {
1437 bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_REQ_WOL_MCP, 0);
1438 bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_DONE, 0);
1443 bnx2x_napi_disable(bp);
1447 bnx2x_release_firmware(bp);
1452 /* must be called with rtnl_lock */
1453 int bnx2x_nic_unload(struct bnx2x *bp, int unload_mode)
1457 if (bp->state == BNX2X_STATE_CLOSED) {
1458 /* Interface has been removed - nothing to recover */
1459 bp->recovery_state = BNX2X_RECOVERY_DONE;
1461 bnx2x_release_hw_lock(bp, HW_LOCK_RESOURCE_RESERVED_08);
1468 bnx2x_cnic_notify(bp, CNIC_CTL_STOP_CMD);
1470 bp->state = BNX2X_STATE_CLOSING_WAIT4_HALT;
1472 /* Set "drop all" */
1473 bp->rx_mode = BNX2X_RX_MODE_NONE;
1474 bnx2x_set_storm_rx_mode(bp);
1477 bnx2x_tx_disable(bp);
1479 del_timer_sync(&bp->timer);
1481 SHMEM_WR(bp, func_mb[BP_FW_MB_IDX(bp)].drv_pulse_mb,
1482 (DRV_PULSE_ALWAYS_ALIVE | bp->fw_drv_pulse_wr_seq));
1484 bnx2x_stats_handle(bp, STATS_EVENT_STOP);
1486 /* Cleanup the chip if needed */
1487 if (unload_mode != UNLOAD_RECOVERY)
1488 bnx2x_chip_cleanup(bp, unload_mode);
1490 /* Disable HW interrupts, NAPI and Tx */
1491 bnx2x_netif_stop(bp, 1);
1499 /* Free SKBs, SGEs, TPA pool and driver internals */
1500 bnx2x_free_skbs(bp);
1501 for_each_queue(bp, i)
1502 bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
1506 bp->state = BNX2X_STATE_CLOSED;
1508 /* The last driver must disable a "close the gate" if there is no
1509 * parity attention or "process kill" pending.
1511 if ((!bnx2x_dec_load_cnt(bp)) && (!bnx2x_chk_parity_attn(bp)) &&
1512 bnx2x_reset_is_done(bp))
1513 bnx2x_disable_close_the_gate(bp);
1515 /* Reset MCP mail box sequence if there is on going recovery */
1516 if (unload_mode == UNLOAD_RECOVERY)
1522 int bnx2x_set_power_state(struct bnx2x *bp, pci_power_t state)
1526 pci_read_config_word(bp->pdev, bp->pm_cap + PCI_PM_CTRL, &pmcsr);
1530 pci_write_config_word(bp->pdev, bp->pm_cap + PCI_PM_CTRL,
1531 ((pmcsr & ~PCI_PM_CTRL_STATE_MASK) |
1532 PCI_PM_CTRL_PME_STATUS));
1534 if (pmcsr & PCI_PM_CTRL_STATE_MASK)
1535 /* delay required during transition out of D3hot */
1540 /* If there are other clients above don't
1541 shut down the power */
1542 if (atomic_read(&bp->pdev->enable_cnt) != 1)
1544 /* Don't shut down the power for emulation and FPGA */
1545 if (CHIP_REV_IS_SLOW(bp))
1548 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
1552 pmcsr |= PCI_PM_CTRL_PME_ENABLE;
1554 pci_write_config_word(bp->pdev, bp->pm_cap + PCI_PM_CTRL,
1557 /* No more memory access after this point until
1558 * device is brought back to D0.
1569 * net_device service functions
1571 int bnx2x_poll(struct napi_struct *napi, int budget)
1574 struct bnx2x_fastpath *fp = container_of(napi, struct bnx2x_fastpath,
1576 struct bnx2x *bp = fp->bp;
1579 #ifdef BNX2X_STOP_ON_ERROR
1580 if (unlikely(bp->panic)) {
1581 napi_complete(napi);
1586 if (bnx2x_has_tx_work(fp))
1589 if (bnx2x_has_rx_work(fp)) {
1590 work_done += bnx2x_rx_int(fp, budget - work_done);
1592 /* must not complete if we consumed full budget */
1593 if (work_done >= budget)
1597 /* Fall out from the NAPI loop if needed */
1598 if (!(bnx2x_has_rx_work(fp) || bnx2x_has_tx_work(fp))) {
1599 bnx2x_update_fpsb_idx(fp);
1600 /* bnx2x_has_rx_work() reads the status block,
1601 * thus we need to ensure that status block indices
1602 * have been actually read (bnx2x_update_fpsb_idx)
1603 * prior to this check (bnx2x_has_rx_work) so that
1604 * we won't write the "newer" value of the status block
1605 * to IGU (if there was a DMA right after
1606 * bnx2x_has_rx_work and if there is no rmb, the memory
1607 * reading (bnx2x_update_fpsb_idx) may be postponed
1608 * to right before bnx2x_ack_sb). In this case there
1609 * will never be another interrupt until there is
1610 * another update of the status block, while there
1611 * is still unhandled work.
1615 if (!(bnx2x_has_rx_work(fp) || bnx2x_has_tx_work(fp))) {
1616 napi_complete(napi);
1617 /* Re-enable interrupts */
1619 "Update index to %d\n", fp->fp_hc_idx);
1620 bnx2x_ack_sb(bp, fp->igu_sb_id, USTORM_ID,
1621 le16_to_cpu(fp->fp_hc_idx),
1631 /* we split the first BD into headers and data BDs
1632 * to ease the pain of our fellow microcode engineers
1633 * we use one mapping for both BDs
1634 * So far this has only been observed to happen
1635 * in Other Operating Systems(TM)
1637 static noinline u16 bnx2x_tx_split(struct bnx2x *bp,
1638 struct bnx2x_fastpath *fp,
1639 struct sw_tx_bd *tx_buf,
1640 struct eth_tx_start_bd **tx_bd, u16 hlen,
1641 u16 bd_prod, int nbd)
1643 struct eth_tx_start_bd *h_tx_bd = *tx_bd;
1644 struct eth_tx_bd *d_tx_bd;
1646 int old_len = le16_to_cpu(h_tx_bd->nbytes);
1648 /* first fix first BD */
1649 h_tx_bd->nbd = cpu_to_le16(nbd);
1650 h_tx_bd->nbytes = cpu_to_le16(hlen);
1652 DP(NETIF_MSG_TX_QUEUED, "TSO split header size is %d "
1653 "(%x:%x) nbd %d\n", h_tx_bd->nbytes, h_tx_bd->addr_hi,
1654 h_tx_bd->addr_lo, h_tx_bd->nbd);
1656 /* now get a new data BD
1657 * (after the pbd) and fill it */
1658 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
1659 d_tx_bd = &fp->tx_desc_ring[bd_prod].reg_bd;
1661 mapping = HILO_U64(le32_to_cpu(h_tx_bd->addr_hi),
1662 le32_to_cpu(h_tx_bd->addr_lo)) + hlen;
1664 d_tx_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
1665 d_tx_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
1666 d_tx_bd->nbytes = cpu_to_le16(old_len - hlen);
1668 /* this marks the BD as one that has no individual mapping */
1669 tx_buf->flags |= BNX2X_TSO_SPLIT_BD;
1671 DP(NETIF_MSG_TX_QUEUED,
1672 "TSO split data size is %d (%x:%x)\n",
1673 d_tx_bd->nbytes, d_tx_bd->addr_hi, d_tx_bd->addr_lo);
1676 *tx_bd = (struct eth_tx_start_bd *)d_tx_bd;
1681 static inline u16 bnx2x_csum_fix(unsigned char *t_header, u16 csum, s8 fix)
1684 csum = (u16) ~csum_fold(csum_sub(csum,
1685 csum_partial(t_header - fix, fix, 0)));
1688 csum = (u16) ~csum_fold(csum_add(csum,
1689 csum_partial(t_header, -fix, 0)));
1691 return swab16(csum);
1694 static inline u32 bnx2x_xmit_type(struct bnx2x *bp, struct sk_buff *skb)
1698 if (skb->ip_summed != CHECKSUM_PARTIAL)
1702 if (skb->protocol == htons(ETH_P_IPV6)) {
1704 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
1705 rc |= XMIT_CSUM_TCP;
1709 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
1710 rc |= XMIT_CSUM_TCP;
1714 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
1715 rc |= (XMIT_GSO_V4 | XMIT_CSUM_V4 | XMIT_CSUM_TCP);
1717 else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
1718 rc |= (XMIT_GSO_V6 | XMIT_CSUM_TCP | XMIT_CSUM_V6);
1723 #if (MAX_SKB_FRAGS >= MAX_FETCH_BD - 3)
1724 /* check if packet requires linearization (packet is too fragmented)
1725 no need to check fragmentation if page size > 8K (there will be no
1726 violation to FW restrictions) */
1727 static int bnx2x_pkt_req_lin(struct bnx2x *bp, struct sk_buff *skb,
1732 int first_bd_sz = 0;
1734 /* 3 = 1 (for linear data BD) + 2 (for PBD and last BD) */
1735 if (skb_shinfo(skb)->nr_frags >= (MAX_FETCH_BD - 3)) {
1737 if (xmit_type & XMIT_GSO) {
1738 unsigned short lso_mss = skb_shinfo(skb)->gso_size;
1739 /* Check if LSO packet needs to be copied:
1740 3 = 1 (for headers BD) + 2 (for PBD and last BD) */
1741 int wnd_size = MAX_FETCH_BD - 3;
1742 /* Number of windows to check */
1743 int num_wnds = skb_shinfo(skb)->nr_frags - wnd_size;
1748 /* Headers length */
1749 hlen = (int)(skb_transport_header(skb) - skb->data) +
1752 /* Amount of data (w/o headers) on linear part of SKB*/
1753 first_bd_sz = skb_headlen(skb) - hlen;
1755 wnd_sum = first_bd_sz;
1757 /* Calculate the first sum - it's special */
1758 for (frag_idx = 0; frag_idx < wnd_size - 1; frag_idx++)
1760 skb_shinfo(skb)->frags[frag_idx].size;
1762 /* If there was data on linear skb data - check it */
1763 if (first_bd_sz > 0) {
1764 if (unlikely(wnd_sum < lso_mss)) {
1769 wnd_sum -= first_bd_sz;
1772 /* Others are easier: run through the frag list and
1773 check all windows */
1774 for (wnd_idx = 0; wnd_idx <= num_wnds; wnd_idx++) {
1776 skb_shinfo(skb)->frags[wnd_idx + wnd_size - 1].size;
1778 if (unlikely(wnd_sum < lso_mss)) {
1783 skb_shinfo(skb)->frags[wnd_idx].size;
1786 /* in non-LSO too fragmented packet should always
1793 if (unlikely(to_copy))
1794 DP(NETIF_MSG_TX_QUEUED,
1795 "Linearization IS REQUIRED for %s packet. "
1796 "num_frags %d hlen %d first_bd_sz %d\n",
1797 (xmit_type & XMIT_GSO) ? "LSO" : "non-LSO",
1798 skb_shinfo(skb)->nr_frags, hlen, first_bd_sz);
1804 static inline void bnx2x_set_pbd_gso_e2(struct sk_buff *skb,
1805 struct eth_tx_parse_bd_e2 *pbd,
1808 pbd->parsing_data |= cpu_to_le16(skb_shinfo(skb)->gso_size) <<
1809 ETH_TX_PARSE_BD_E2_LSO_MSS_SHIFT;
1810 if ((xmit_type & XMIT_GSO_V6) &&
1811 (ipv6_hdr(skb)->nexthdr == NEXTHDR_IPV6))
1812 pbd->parsing_data |= ETH_TX_PARSE_BD_E2_IPV6_WITH_EXT_HDR;
1816 * Update PBD in GSO case.
1819 * @param tx_start_bd
1823 static inline void bnx2x_set_pbd_gso(struct sk_buff *skb,
1824 struct eth_tx_parse_bd_e1x *pbd,
1827 pbd->lso_mss = cpu_to_le16(skb_shinfo(skb)->gso_size);
1828 pbd->tcp_send_seq = swab32(tcp_hdr(skb)->seq);
1829 pbd->tcp_flags = pbd_tcp_flags(skb);
1831 if (xmit_type & XMIT_GSO_V4) {
1832 pbd->ip_id = swab16(ip_hdr(skb)->id);
1833 pbd->tcp_pseudo_csum =
1834 swab16(~csum_tcpudp_magic(ip_hdr(skb)->saddr,
1836 0, IPPROTO_TCP, 0));
1839 pbd->tcp_pseudo_csum =
1840 swab16(~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
1841 &ipv6_hdr(skb)->daddr,
1842 0, IPPROTO_TCP, 0));
1844 pbd->global_data |= ETH_TX_PARSE_BD_E1X_PSEUDO_CS_WITHOUT_LEN;
1850 * @param tx_start_bd
1854 * @return header len
1856 static inline u8 bnx2x_set_pbd_csum_e2(struct bnx2x *bp, struct sk_buff *skb,
1857 struct eth_tx_parse_bd_e2 *pbd,
1860 pbd->parsing_data |= cpu_to_le16(tcp_hdrlen(skb)/4) <<
1861 ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW_SHIFT;
1863 pbd->parsing_data |= cpu_to_le16(((unsigned char *)tcp_hdr(skb) -
1865 ETH_TX_PARSE_BD_E2_TCP_HDR_START_OFFSET_W_SHIFT;
1867 return skb_transport_header(skb) + tcp_hdrlen(skb) - skb->data;
1873 * @param tx_start_bd
1877 * @return Header length
1879 static inline u8 bnx2x_set_pbd_csum(struct bnx2x *bp, struct sk_buff *skb,
1880 struct eth_tx_parse_bd_e1x *pbd,
1883 u8 hlen = (skb_network_header(skb) - skb->data) / 2;
1885 /* for now NS flag is not used in Linux */
1887 (hlen | ((skb->protocol == cpu_to_be16(ETH_P_8021Q)) <<
1888 ETH_TX_PARSE_BD_E1X_LLC_SNAP_EN_SHIFT));
1890 pbd->ip_hlen_w = (skb_transport_header(skb) -
1891 skb_network_header(skb)) / 2;
1893 hlen += pbd->ip_hlen_w + tcp_hdrlen(skb) / 2;
1895 pbd->total_hlen_w = cpu_to_le16(hlen);
1898 if (xmit_type & XMIT_CSUM_TCP) {
1899 pbd->tcp_pseudo_csum = swab16(tcp_hdr(skb)->check);
1902 s8 fix = SKB_CS_OFF(skb); /* signed! */
1904 DP(NETIF_MSG_TX_QUEUED,
1905 "hlen %d fix %d csum before fix %x\n",
1906 le16_to_cpu(pbd->total_hlen_w), fix, SKB_CS(skb));
1908 /* HW bug: fixup the CSUM */
1909 pbd->tcp_pseudo_csum =
1910 bnx2x_csum_fix(skb_transport_header(skb),
1913 DP(NETIF_MSG_TX_QUEUED, "csum after fix %x\n",
1914 pbd->tcp_pseudo_csum);
1920 /* called with netif_tx_lock
1921 * bnx2x_tx_int() runs without netif_tx_lock unless it needs to call
1922 * netif_wake_queue()
1924 netdev_tx_t bnx2x_start_xmit(struct sk_buff *skb, struct net_device *dev)
1926 struct bnx2x *bp = netdev_priv(dev);
1927 struct bnx2x_fastpath *fp;
1928 struct netdev_queue *txq;
1929 struct sw_tx_bd *tx_buf;
1930 struct eth_tx_start_bd *tx_start_bd;
1931 struct eth_tx_bd *tx_data_bd, *total_pkt_bd = NULL;
1932 struct eth_tx_parse_bd_e1x *pbd_e1x = NULL;
1933 struct eth_tx_parse_bd_e2 *pbd_e2 = NULL;
1934 u16 pkt_prod, bd_prod;
1937 u32 xmit_type = bnx2x_xmit_type(bp, skb);
1940 __le16 pkt_size = 0;
1942 u8 mac_type = UNICAST_ADDRESS;
1944 #ifdef BNX2X_STOP_ON_ERROR
1945 if (unlikely(bp->panic))
1946 return NETDEV_TX_BUSY;
1949 fp_index = skb_get_queue_mapping(skb);
1950 txq = netdev_get_tx_queue(dev, fp_index);
1952 fp = &bp->fp[fp_index];
1954 if (unlikely(bnx2x_tx_avail(fp) < (skb_shinfo(skb)->nr_frags + 3))) {
1955 fp->eth_q_stats.driver_xoff++;
1956 netif_tx_stop_queue(txq);
1957 BNX2X_ERR("BUG! Tx ring full when queue awake!\n");
1958 return NETDEV_TX_BUSY;
1961 DP(NETIF_MSG_TX_QUEUED, "queue[%d]: SKB: summed %x protocol %x "
1962 "protocol(%x,%x) gso type %x xmit_type %x\n",
1963 fp_index, skb->ip_summed, skb->protocol, ipv6_hdr(skb)->nexthdr,
1964 ip_hdr(skb)->protocol, skb_shinfo(skb)->gso_type, xmit_type);
1966 eth = (struct ethhdr *)skb->data;
1968 /* set flag according to packet type (UNICAST_ADDRESS is default)*/
1969 if (unlikely(is_multicast_ether_addr(eth->h_dest))) {
1970 if (is_broadcast_ether_addr(eth->h_dest))
1971 mac_type = BROADCAST_ADDRESS;
1973 mac_type = MULTICAST_ADDRESS;
1976 #if (MAX_SKB_FRAGS >= MAX_FETCH_BD - 3)
1977 /* First, check if we need to linearize the skb (due to FW
1978 restrictions). No need to check fragmentation if page size > 8K
1979 (there will be no violation to FW restrictions) */
1980 if (bnx2x_pkt_req_lin(bp, skb, xmit_type)) {
1981 /* Statistics of linearization */
1983 if (skb_linearize(skb) != 0) {
1984 DP(NETIF_MSG_TX_QUEUED, "SKB linearization failed - "
1985 "silently dropping this SKB\n");
1986 dev_kfree_skb_any(skb);
1987 return NETDEV_TX_OK;
1993 Please read carefully. First we use one BD which we mark as start,
1994 then we have a parsing info BD (used for TSO or xsum),
1995 and only then we have the rest of the TSO BDs.
1996 (don't forget to mark the last one as last,
1997 and to unmap only AFTER you write to the BD ...)
1998 And above all, all pdb sizes are in words - NOT DWORDS!
2001 pkt_prod = fp->tx_pkt_prod++;
2002 bd_prod = TX_BD(fp->tx_bd_prod);
2004 /* get a tx_buf and first BD */
2005 tx_buf = &fp->tx_buf_ring[TX_BD(pkt_prod)];
2006 tx_start_bd = &fp->tx_desc_ring[bd_prod].start_bd;
2008 tx_start_bd->bd_flags.as_bitfield = ETH_TX_BD_FLAGS_START_BD;
2009 SET_FLAG(tx_start_bd->general_data, ETH_TX_START_BD_ETH_ADDR_TYPE,
2013 SET_FLAG(tx_start_bd->general_data, ETH_TX_START_BD_HDR_NBDS, 1);
2015 /* remember the first BD of the packet */
2016 tx_buf->first_bd = fp->tx_bd_prod;
2020 DP(NETIF_MSG_TX_QUEUED,
2021 "sending pkt %u @%p next_idx %u bd %u @%p\n",
2022 pkt_prod, tx_buf, fp->tx_pkt_prod, bd_prod, tx_start_bd);
2025 if ((bp->vlgrp != NULL) && vlan_tx_tag_present(skb) &&
2026 (bp->flags & HW_VLAN_TX_FLAG)) {
2027 tx_start_bd->vlan_or_ethertype =
2028 cpu_to_le16(vlan_tx_tag_get(skb));
2029 tx_start_bd->bd_flags.as_bitfield |=
2030 (X_ETH_OUTBAND_VLAN << ETH_TX_BD_FLAGS_VLAN_MODE_SHIFT);
2033 tx_start_bd->vlan_or_ethertype = cpu_to_le16(pkt_prod);
2035 /* turn on parsing and get a BD */
2036 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
2038 if (xmit_type & XMIT_CSUM) {
2039 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_L4_CSUM;
2041 if (xmit_type & XMIT_CSUM_V4)
2042 tx_start_bd->bd_flags.as_bitfield |=
2043 ETH_TX_BD_FLAGS_IP_CSUM;
2045 tx_start_bd->bd_flags.as_bitfield |=
2046 ETH_TX_BD_FLAGS_IPV6;
2048 if (!(xmit_type & XMIT_CSUM_TCP))
2049 tx_start_bd->bd_flags.as_bitfield |=
2050 ETH_TX_BD_FLAGS_IS_UDP;
2053 if (CHIP_IS_E2(bp)) {
2054 pbd_e2 = &fp->tx_desc_ring[bd_prod].parse_bd_e2;
2055 memset(pbd_e2, 0, sizeof(struct eth_tx_parse_bd_e2));
2056 /* Set PBD in checksum offload case */
2057 if (xmit_type & XMIT_CSUM)
2058 hlen = bnx2x_set_pbd_csum_e2(bp,
2059 skb, pbd_e2, xmit_type);
2061 pbd_e1x = &fp->tx_desc_ring[bd_prod].parse_bd_e1x;
2062 memset(pbd_e1x, 0, sizeof(struct eth_tx_parse_bd_e1x));
2063 /* Set PBD in checksum offload case */
2064 if (xmit_type & XMIT_CSUM)
2065 hlen = bnx2x_set_pbd_csum(bp, skb, pbd_e1x, xmit_type);
2069 /* Map skb linear data for DMA */
2070 mapping = dma_map_single(&bp->pdev->dev, skb->data,
2071 skb_headlen(skb), DMA_TO_DEVICE);
2073 /* Setup the data pointer of the first BD of the packet */
2074 tx_start_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
2075 tx_start_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
2076 nbd = skb_shinfo(skb)->nr_frags + 2; /* start_bd + pbd + frags */
2077 tx_start_bd->nbd = cpu_to_le16(nbd);
2078 tx_start_bd->nbytes = cpu_to_le16(skb_headlen(skb));
2079 pkt_size = tx_start_bd->nbytes;
2081 DP(NETIF_MSG_TX_QUEUED, "first bd @%p addr (%x:%x) nbd %d"
2082 " nbytes %d flags %x vlan %x\n",
2083 tx_start_bd, tx_start_bd->addr_hi, tx_start_bd->addr_lo,
2084 le16_to_cpu(tx_start_bd->nbd), le16_to_cpu(tx_start_bd->nbytes),
2085 tx_start_bd->bd_flags.as_bitfield,
2086 le16_to_cpu(tx_start_bd->vlan_or_ethertype));
2088 if (xmit_type & XMIT_GSO) {
2090 DP(NETIF_MSG_TX_QUEUED,
2091 "TSO packet len %d hlen %d total len %d tso size %d\n",
2092 skb->len, hlen, skb_headlen(skb),
2093 skb_shinfo(skb)->gso_size);
2095 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_SW_LSO;
2097 if (unlikely(skb_headlen(skb) > hlen))
2098 bd_prod = bnx2x_tx_split(bp, fp, tx_buf, &tx_start_bd,
2099 hlen, bd_prod, ++nbd);
2101 bnx2x_set_pbd_gso_e2(skb, pbd_e2, xmit_type);
2103 bnx2x_set_pbd_gso(skb, pbd_e1x, xmit_type);
2105 tx_data_bd = (struct eth_tx_bd *)tx_start_bd;
2107 /* Handle fragmented skb */
2108 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2109 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2111 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
2112 tx_data_bd = &fp->tx_desc_ring[bd_prod].reg_bd;
2113 if (total_pkt_bd == NULL)
2114 total_pkt_bd = &fp->tx_desc_ring[bd_prod].reg_bd;
2116 mapping = dma_map_page(&bp->pdev->dev, frag->page,
2118 frag->size, DMA_TO_DEVICE);
2120 tx_data_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
2121 tx_data_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
2122 tx_data_bd->nbytes = cpu_to_le16(frag->size);
2123 le16_add_cpu(&pkt_size, frag->size);
2125 DP(NETIF_MSG_TX_QUEUED,
2126 "frag %d bd @%p addr (%x:%x) nbytes %d\n",
2127 i, tx_data_bd, tx_data_bd->addr_hi, tx_data_bd->addr_lo,
2128 le16_to_cpu(tx_data_bd->nbytes));
2131 DP(NETIF_MSG_TX_QUEUED, "last bd @%p\n", tx_data_bd);
2133 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
2135 /* now send a tx doorbell, counting the next BD
2136 * if the packet contains or ends with it
2138 if (TX_BD_POFF(bd_prod) < nbd)
2141 if (total_pkt_bd != NULL)
2142 total_pkt_bd->total_pkt_bytes = pkt_size;
2145 DP(NETIF_MSG_TX_QUEUED,
2146 "PBD (E1X) @%p ip_data %x ip_hlen %u ip_id %u lso_mss %u"
2147 " tcp_flags %x xsum %x seq %u hlen %u\n",
2148 pbd_e1x, pbd_e1x->global_data, pbd_e1x->ip_hlen_w,
2149 pbd_e1x->ip_id, pbd_e1x->lso_mss, pbd_e1x->tcp_flags,
2150 pbd_e1x->tcp_pseudo_csum, pbd_e1x->tcp_send_seq,
2151 le16_to_cpu(pbd_e1x->total_hlen_w));
2153 DP(NETIF_MSG_TX_QUEUED,
2154 "PBD (E2) @%p dst %x %x %x src %x %x %x parsing_data %x\n",
2155 pbd_e2, pbd_e2->dst_mac_addr_hi, pbd_e2->dst_mac_addr_mid,
2156 pbd_e2->dst_mac_addr_lo, pbd_e2->src_mac_addr_hi,
2157 pbd_e2->src_mac_addr_mid, pbd_e2->src_mac_addr_lo,
2158 pbd_e2->parsing_data);
2159 DP(NETIF_MSG_TX_QUEUED, "doorbell: nbd %d bd %u\n", nbd, bd_prod);
2162 * Make sure that the BD data is updated before updating the producer
2163 * since FW might read the BD right after the producer is updated.
2164 * This is only applicable for weak-ordered memory model archs such
2165 * as IA-64. The following barrier is also mandatory since FW will
2166 * assumes packets must have BDs.
2170 fp->tx_db.data.prod += nbd;
2173 DOORBELL(bp, fp->cid, fp->tx_db.raw);
2177 fp->tx_bd_prod += nbd;
2179 if (unlikely(bnx2x_tx_avail(fp) < MAX_SKB_FRAGS + 3)) {
2180 netif_tx_stop_queue(txq);
2182 /* paired memory barrier is in bnx2x_tx_int(), we have to keep
2183 * ordering of set_bit() in netif_tx_stop_queue() and read of
2187 fp->eth_q_stats.driver_xoff++;
2188 if (bnx2x_tx_avail(fp) >= MAX_SKB_FRAGS + 3)
2189 netif_tx_wake_queue(txq);
2193 return NETDEV_TX_OK;
2196 /* called with rtnl_lock */
2197 int bnx2x_change_mac_addr(struct net_device *dev, void *p)
2199 struct sockaddr *addr = p;
2200 struct bnx2x *bp = netdev_priv(dev);
2202 if (!is_valid_ether_addr((u8 *)(addr->sa_data)))
2205 memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
2206 if (netif_running(dev))
2207 bnx2x_set_eth_mac(bp, 1);
2213 int bnx2x_setup_irqs(struct bnx2x *bp)
2216 if (bp->flags & USING_MSIX_FLAG) {
2217 rc = bnx2x_req_msix_irqs(bp);
2222 rc = bnx2x_req_irq(bp);
2224 BNX2X_ERR("IRQ request failed rc %d, aborting\n", rc);
2227 if (bp->flags & USING_MSI_FLAG) {
2228 bp->dev->irq = bp->pdev->irq;
2229 netdev_info(bp->dev, "using MSI IRQ %d\n",
2237 void bnx2x_free_mem_bp(struct bnx2x *bp)
2240 kfree(bp->msix_table);
2244 int __devinit bnx2x_alloc_mem_bp(struct bnx2x *bp)
2246 struct bnx2x_fastpath *fp;
2247 struct msix_entry *tbl;
2248 struct bnx2x_ilt *ilt;
2251 fp = kzalloc(L2_FP_COUNT(bp->l2_cid_count)*sizeof(*fp), GFP_KERNEL);
2257 tbl = kzalloc((bp->l2_cid_count + 1) * sizeof(*tbl),
2261 bp->msix_table = tbl;
2264 ilt = kzalloc(sizeof(*ilt), GFP_KERNEL);
2271 bnx2x_free_mem_bp(bp);
2276 /* called with rtnl_lock */
2277 int bnx2x_change_mtu(struct net_device *dev, int new_mtu)
2279 struct bnx2x *bp = netdev_priv(dev);
2282 if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
2283 printk(KERN_ERR "Handling parity error recovery. Try again later\n");
2287 if ((new_mtu > ETH_MAX_JUMBO_PACKET_SIZE) ||
2288 ((new_mtu + ETH_HLEN) < ETH_MIN_PACKET_SIZE))
2291 /* This does not race with packet allocation
2292 * because the actual alloc size is
2293 * only updated as part of load
2297 if (netif_running(dev)) {
2298 bnx2x_nic_unload(bp, UNLOAD_NORMAL);
2299 rc = bnx2x_nic_load(bp, LOAD_NORMAL);
2305 void bnx2x_tx_timeout(struct net_device *dev)
2307 struct bnx2x *bp = netdev_priv(dev);
2309 #ifdef BNX2X_STOP_ON_ERROR
2313 /* This allows the netif to be shutdown gracefully before resetting */
2314 schedule_delayed_work(&bp->reset_task, 0);
2318 /* called with rtnl_lock */
2319 void bnx2x_vlan_rx_register(struct net_device *dev,
2320 struct vlan_group *vlgrp)
2322 struct bnx2x *bp = netdev_priv(dev);
2329 int bnx2x_suspend(struct pci_dev *pdev, pm_message_t state)
2331 struct net_device *dev = pci_get_drvdata(pdev);
2335 dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
2338 bp = netdev_priv(dev);
2342 pci_save_state(pdev);
2344 if (!netif_running(dev)) {
2349 netif_device_detach(dev);
2351 bnx2x_nic_unload(bp, UNLOAD_CLOSE);
2353 bnx2x_set_power_state(bp, pci_choose_state(pdev, state));
2360 int bnx2x_resume(struct pci_dev *pdev)
2362 struct net_device *dev = pci_get_drvdata(pdev);
2367 dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
2370 bp = netdev_priv(dev);
2372 if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
2373 printk(KERN_ERR "Handling parity error recovery. Try again later\n");
2379 pci_restore_state(pdev);
2381 if (!netif_running(dev)) {
2386 bnx2x_set_power_state(bp, PCI_D0);
2387 netif_device_attach(dev);
2389 /* Since the chip was reset, clear the FW sequence number */
2391 rc = bnx2x_nic_load(bp, LOAD_OPEN);