proc: use seq_puts()/seq_putc() where possible
[linux-2.6/next.git] / drivers / net / bnx2x / bnx2x_cmn.c
blob710ce5d04c530056a1682dc7992bf6ba414b0b34
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>
20 #include <linux/ip.h>
21 #include <net/ipv6.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,
34 u16 idx)
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;
41 int nbd;
43 /* prefetch skb end pointer to speedup dev_kfree_skb() */
44 prefetch(&skb->end);
46 DP(BNX2X_MSG_OFF, "pkt_idx %d buff @(%p)->skb %p\n",
47 idx, tx_buf, skb);
49 /* unmap first bd */
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");
59 bnx2x_panic();
61 #endif
62 new_cons = nbd + tx_buf->first_bd;
64 /* Get the next bd */
65 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
67 /* Skip a parse bd... */
68 --nbd;
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) {
73 --nbd;
74 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
77 /* now free frags */
78 while (nbd > 0) {
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);
84 if (--nbd)
85 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
88 /* release skb */
89 WARN_ON(!skb);
90 dev_kfree_skb(skb);
91 tx_buf->first_bd = 0;
92 tx_buf->skb = NULL;
94 return new_cons;
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))
105 return -1;
106 #endif
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) {
113 u16 pkt_cons;
115 pkt_cons = TX_BD(sw_cons);
117 DP(NETIF_MSG_TX_DONE, "queue[%d]: hw_cons %u sw_cons %u "
118 " pkt_cons %u\n",
119 fp->index, hw_cons, sw_cons, pkt_cons);
121 bd_cons = bnx2x_free_tx_pkt(bp, fp, pkt_cons);
122 sw_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
132 * forever.
134 smp_mb();
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->
144 * stops the queue
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);
156 return 0;
159 static inline void bnx2x_update_last_max_sge(struct bnx2x_fastpath *fp,
160 u16 idx)
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)) >>
174 SGE_PAGE_SHIFT;
175 u16 last_max, last_elem, first_elem;
176 u16 delta = 0;
177 u16 i;
179 if (!sge_len)
180 return;
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)
201 last_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]))
206 break;
208 fp->sge_mask[i] = RX_SGE_MASK_ELEM_ONE_MASK;
209 delta += RX_SGE_MASK_ELEM_SZ;
212 if (delta > 0) {
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];
230 dma_addr_t mapping;
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",
255 #else
256 DP(NETIF_MSG_RX_STATUS, "fp->tpa_queue_used = 0x%llx\n",
257 #endif
258 fp->tpa_queue_used);
259 #endif
262 static int bnx2x_fill_frag_skb(struct bnx2x *bp, struct bnx2x_fastpath *fp,
263 struct sk_buff *skb,
264 struct eth_fast_path_rx_cqe *fp_cqe,
265 u16 cqe_idx)
267 struct sw_rx_page *rx_pg, old_rx_pg;
268 u16 len_on_bd = le16_to_cpu(fp_cqe->len_on_bd);
269 u32 i, frag_len, frag_size, pages;
270 int err;
271 int j;
273 frag_size = le16_to_cpu(fp_cqe->pkt_len) - len_on_bd;
274 pages = SGE_PAGE_ALIGN(frag_size) >> SGE_PAGE_SHIFT;
276 /* This is needed in order to enable forwarding support */
277 if (frag_size)
278 skb_shinfo(skb)->gso_size = min((u32)SGE_PAGE_SIZE,
279 max(frag_size, (u32)len_on_bd));
281 #ifdef BNX2X_STOP_ON_ERROR
282 if (pages > min_t(u32, 8, MAX_SKB_FRAGS)*SGE_PAGE_SIZE*PAGES_PER_SGE) {
283 BNX2X_ERR("SGL length is too long: %d. CQE index is %d\n",
284 pages, cqe_idx);
285 BNX2X_ERR("fp_cqe->pkt_len = %d fp_cqe->len_on_bd = %d\n",
286 fp_cqe->pkt_len, len_on_bd);
287 bnx2x_panic();
288 return -EINVAL;
290 #endif
292 /* Run through the SGL and compose the fragmented skb */
293 for (i = 0, j = 0; i < pages; i += PAGES_PER_SGE, j++) {
294 u16 sge_idx =
295 RX_SGE(le16_to_cpu(fp_cqe->sgl_or_raw_data.sgl[j]));
297 /* FW gives the indices of the SGE as if the ring is an array
298 (meaning that "next" element will consume 2 indices) */
299 frag_len = min(frag_size, (u32)(SGE_PAGE_SIZE*PAGES_PER_SGE));
300 rx_pg = &fp->rx_page_ring[sge_idx];
301 old_rx_pg = *rx_pg;
303 /* If we fail to allocate a substitute page, we simply stop
304 where we are and drop the whole packet */
305 err = bnx2x_alloc_rx_sge(bp, fp, sge_idx);
306 if (unlikely(err)) {
307 fp->eth_q_stats.rx_skb_alloc_failed++;
308 return err;
311 /* Unmap the page as we r going to pass it to the stack */
312 dma_unmap_page(&bp->pdev->dev,
313 dma_unmap_addr(&old_rx_pg, mapping),
314 SGE_PAGE_SIZE*PAGES_PER_SGE, DMA_FROM_DEVICE);
316 /* Add one frag and update the appropriate fields in the skb */
317 skb_fill_page_desc(skb, j, old_rx_pg.page, 0, frag_len);
319 skb->data_len += frag_len;
320 skb->truesize += frag_len;
321 skb->len += frag_len;
323 frag_size -= frag_len;
326 return 0;
329 static void bnx2x_tpa_stop(struct bnx2x *bp, struct bnx2x_fastpath *fp,
330 u16 queue, int pad, int len, union eth_rx_cqe *cqe,
331 u16 cqe_idx)
333 struct sw_rx_bd *rx_buf = &fp->tpa_pool[queue];
334 struct sk_buff *skb = rx_buf->skb;
335 /* alloc new skb */
336 struct sk_buff *new_skb = netdev_alloc_skb(bp->dev, bp->rx_buf_size);
338 /* Unmap skb in the pool anyway, as we are going to change
339 pool entry status to BNX2X_TPA_STOP even if new skb allocation
340 fails. */
341 dma_unmap_single(&bp->pdev->dev, dma_unmap_addr(rx_buf, mapping),
342 bp->rx_buf_size, DMA_FROM_DEVICE);
344 if (likely(new_skb)) {
345 /* fix ip xsum and give it to the stack */
346 /* (no need to map the new skb) */
348 prefetch(skb);
349 prefetch(((char *)(skb)) + L1_CACHE_BYTES);
351 #ifdef BNX2X_STOP_ON_ERROR
352 if (pad + len > bp->rx_buf_size) {
353 BNX2X_ERR("skb_put is about to fail... "
354 "pad %d len %d rx_buf_size %d\n",
355 pad, len, bp->rx_buf_size);
356 bnx2x_panic();
357 return;
359 #endif
361 skb_reserve(skb, pad);
362 skb_put(skb, len);
364 skb->protocol = eth_type_trans(skb, bp->dev);
365 skb->ip_summed = CHECKSUM_UNNECESSARY;
368 struct iphdr *iph;
370 iph = (struct iphdr *)skb->data;
371 iph->check = 0;
372 iph->check = ip_fast_csum((u8 *)iph, iph->ihl);
375 if (!bnx2x_fill_frag_skb(bp, fp, skb,
376 &cqe->fast_path_cqe, cqe_idx)) {
377 if ((le16_to_cpu(cqe->fast_path_cqe.
378 pars_flags.flags) & PARSING_FLAGS_VLAN))
379 __vlan_hwaccel_put_tag(skb,
380 le16_to_cpu(cqe->fast_path_cqe.
381 vlan_tag));
382 napi_gro_receive(&fp->napi, skb);
383 } else {
384 DP(NETIF_MSG_RX_STATUS, "Failed to allocate new pages"
385 " - dropping packet!\n");
386 dev_kfree_skb(skb);
390 /* put new skb in bin */
391 fp->tpa_pool[queue].skb = new_skb;
393 } else {
394 /* else drop the packet and keep the buffer in the bin */
395 DP(NETIF_MSG_RX_STATUS,
396 "Failed to allocate new skb - dropping packet!\n");
397 fp->eth_q_stats.rx_skb_alloc_failed++;
400 fp->tpa_state[queue] = BNX2X_TPA_STOP;
403 /* Set Toeplitz hash value in the skb using the value from the
404 * CQE (calculated by HW).
406 static inline void bnx2x_set_skb_rxhash(struct bnx2x *bp, union eth_rx_cqe *cqe,
407 struct sk_buff *skb)
409 /* Set Toeplitz hash from CQE */
410 if ((bp->dev->features & NETIF_F_RXHASH) &&
411 (cqe->fast_path_cqe.status_flags &
412 ETH_FAST_PATH_RX_CQE_RSS_HASH_FLG))
413 skb->rxhash =
414 le32_to_cpu(cqe->fast_path_cqe.rss_hash_result);
417 int bnx2x_rx_int(struct bnx2x_fastpath *fp, int budget)
419 struct bnx2x *bp = fp->bp;
420 u16 bd_cons, bd_prod, bd_prod_fw, comp_ring_cons;
421 u16 hw_comp_cons, sw_comp_cons, sw_comp_prod;
422 int rx_pkt = 0;
424 #ifdef BNX2X_STOP_ON_ERROR
425 if (unlikely(bp->panic))
426 return 0;
427 #endif
429 /* CQ "next element" is of the size of the regular element,
430 that's why it's ok here */
431 hw_comp_cons = le16_to_cpu(*fp->rx_cons_sb);
432 if ((hw_comp_cons & MAX_RCQ_DESC_CNT) == MAX_RCQ_DESC_CNT)
433 hw_comp_cons++;
435 bd_cons = fp->rx_bd_cons;
436 bd_prod = fp->rx_bd_prod;
437 bd_prod_fw = bd_prod;
438 sw_comp_cons = fp->rx_comp_cons;
439 sw_comp_prod = fp->rx_comp_prod;
441 /* Memory barrier necessary as speculative reads of the rx
442 * buffer can be ahead of the index in the status block
444 rmb();
446 DP(NETIF_MSG_RX_STATUS,
447 "queue[%d]: hw_comp_cons %u sw_comp_cons %u\n",
448 fp->index, hw_comp_cons, sw_comp_cons);
450 while (sw_comp_cons != hw_comp_cons) {
451 struct sw_rx_bd *rx_buf = NULL;
452 struct sk_buff *skb;
453 union eth_rx_cqe *cqe;
454 u8 cqe_fp_flags;
455 u16 len, pad;
457 comp_ring_cons = RCQ_BD(sw_comp_cons);
458 bd_prod = RX_BD(bd_prod);
459 bd_cons = RX_BD(bd_cons);
461 /* Prefetch the page containing the BD descriptor
462 at producer's index. It will be needed when new skb is
463 allocated */
464 prefetch((void *)(PAGE_ALIGN((unsigned long)
465 (&fp->rx_desc_ring[bd_prod])) -
466 PAGE_SIZE + 1));
468 cqe = &fp->rx_comp_ring[comp_ring_cons];
469 cqe_fp_flags = cqe->fast_path_cqe.type_error_flags;
471 DP(NETIF_MSG_RX_STATUS, "CQE type %x err %x status %x"
472 " queue %x vlan %x len %u\n", CQE_TYPE(cqe_fp_flags),
473 cqe_fp_flags, cqe->fast_path_cqe.status_flags,
474 le32_to_cpu(cqe->fast_path_cqe.rss_hash_result),
475 le16_to_cpu(cqe->fast_path_cqe.vlan_tag),
476 le16_to_cpu(cqe->fast_path_cqe.pkt_len));
478 /* is this a slowpath msg? */
479 if (unlikely(CQE_TYPE(cqe_fp_flags))) {
480 bnx2x_sp_event(fp, cqe);
481 goto next_cqe;
483 /* this is an rx packet */
484 } else {
485 rx_buf = &fp->rx_buf_ring[bd_cons];
486 skb = rx_buf->skb;
487 prefetch(skb);
488 len = le16_to_cpu(cqe->fast_path_cqe.pkt_len);
489 pad = cqe->fast_path_cqe.placement_offset;
491 /* - If CQE is marked both TPA_START and TPA_END it is
492 * a non-TPA CQE.
493 * - FP CQE will always have either TPA_START or/and
494 * TPA_STOP flags set.
496 if ((!fp->disable_tpa) &&
497 (TPA_TYPE(cqe_fp_flags) !=
498 (TPA_TYPE_START | TPA_TYPE_END))) {
499 u16 queue = cqe->fast_path_cqe.queue_index;
501 if (TPA_TYPE(cqe_fp_flags) == TPA_TYPE_START) {
502 DP(NETIF_MSG_RX_STATUS,
503 "calling tpa_start on queue %d\n",
504 queue);
506 bnx2x_tpa_start(fp, queue, skb,
507 bd_cons, bd_prod);
509 /* Set Toeplitz hash for an LRO skb */
510 bnx2x_set_skb_rxhash(bp, cqe, skb);
512 goto next_rx;
513 } else { /* TPA_STOP */
514 DP(NETIF_MSG_RX_STATUS,
515 "calling tpa_stop on queue %d\n",
516 queue);
518 if (!BNX2X_RX_SUM_FIX(cqe))
519 BNX2X_ERR("STOP on none TCP "
520 "data\n");
522 /* This is a size of the linear data
523 on this skb */
524 len = le16_to_cpu(cqe->fast_path_cqe.
525 len_on_bd);
526 bnx2x_tpa_stop(bp, fp, queue, pad,
527 len, cqe, comp_ring_cons);
528 #ifdef BNX2X_STOP_ON_ERROR
529 if (bp->panic)
530 return 0;
531 #endif
533 bnx2x_update_sge_prod(fp,
534 &cqe->fast_path_cqe);
535 goto next_cqe;
539 dma_sync_single_for_device(&bp->pdev->dev,
540 dma_unmap_addr(rx_buf, mapping),
541 pad + RX_COPY_THRESH,
542 DMA_FROM_DEVICE);
543 prefetch(((char *)(skb)) + L1_CACHE_BYTES);
545 /* is this an error packet? */
546 if (unlikely(cqe_fp_flags & ETH_RX_ERROR_FALGS)) {
547 DP(NETIF_MSG_RX_ERR,
548 "ERROR flags %x rx packet %u\n",
549 cqe_fp_flags, sw_comp_cons);
550 fp->eth_q_stats.rx_err_discard_pkt++;
551 goto reuse_rx;
554 /* Since we don't have a jumbo ring
555 * copy small packets if mtu > 1500
557 if ((bp->dev->mtu > ETH_MAX_PACKET_SIZE) &&
558 (len <= RX_COPY_THRESH)) {
559 struct sk_buff *new_skb;
561 new_skb = netdev_alloc_skb(bp->dev,
562 len + pad);
563 if (new_skb == NULL) {
564 DP(NETIF_MSG_RX_ERR,
565 "ERROR packet dropped "
566 "because of alloc failure\n");
567 fp->eth_q_stats.rx_skb_alloc_failed++;
568 goto reuse_rx;
571 /* aligned copy */
572 skb_copy_from_linear_data_offset(skb, pad,
573 new_skb->data + pad, len);
574 skb_reserve(new_skb, pad);
575 skb_put(new_skb, len);
577 bnx2x_reuse_rx_skb(fp, bd_cons, bd_prod);
579 skb = new_skb;
581 } else
582 if (likely(bnx2x_alloc_rx_skb(bp, fp, bd_prod) == 0)) {
583 dma_unmap_single(&bp->pdev->dev,
584 dma_unmap_addr(rx_buf, mapping),
585 bp->rx_buf_size,
586 DMA_FROM_DEVICE);
587 skb_reserve(skb, pad);
588 skb_put(skb, len);
590 } else {
591 DP(NETIF_MSG_RX_ERR,
592 "ERROR packet dropped because "
593 "of alloc failure\n");
594 fp->eth_q_stats.rx_skb_alloc_failed++;
595 reuse_rx:
596 bnx2x_reuse_rx_skb(fp, bd_cons, bd_prod);
597 goto next_rx;
600 skb->protocol = eth_type_trans(skb, bp->dev);
602 /* Set Toeplitz hash for a none-LRO skb */
603 bnx2x_set_skb_rxhash(bp, cqe, skb);
605 skb_checksum_none_assert(skb);
607 if (bp->rx_csum) {
608 if (likely(BNX2X_RX_CSUM_OK(cqe)))
609 skb->ip_summed = CHECKSUM_UNNECESSARY;
610 else
611 fp->eth_q_stats.hw_csum_err++;
615 skb_record_rx_queue(skb, fp->index);
617 if (le16_to_cpu(cqe->fast_path_cqe.pars_flags.flags) &
618 PARSING_FLAGS_VLAN)
619 __vlan_hwaccel_put_tag(skb,
620 le16_to_cpu(cqe->fast_path_cqe.vlan_tag));
621 napi_gro_receive(&fp->napi, skb);
624 next_rx:
625 rx_buf->skb = NULL;
627 bd_cons = NEXT_RX_IDX(bd_cons);
628 bd_prod = NEXT_RX_IDX(bd_prod);
629 bd_prod_fw = NEXT_RX_IDX(bd_prod_fw);
630 rx_pkt++;
631 next_cqe:
632 sw_comp_prod = NEXT_RCQ_IDX(sw_comp_prod);
633 sw_comp_cons = NEXT_RCQ_IDX(sw_comp_cons);
635 if (rx_pkt == budget)
636 break;
637 } /* while */
639 fp->rx_bd_cons = bd_cons;
640 fp->rx_bd_prod = bd_prod_fw;
641 fp->rx_comp_cons = sw_comp_cons;
642 fp->rx_comp_prod = sw_comp_prod;
644 /* Update producers */
645 bnx2x_update_rx_prod(bp, fp, bd_prod_fw, sw_comp_prod,
646 fp->rx_sge_prod);
648 fp->rx_pkt += rx_pkt;
649 fp->rx_calls++;
651 return rx_pkt;
654 static irqreturn_t bnx2x_msix_fp_int(int irq, void *fp_cookie)
656 struct bnx2x_fastpath *fp = fp_cookie;
657 struct bnx2x *bp = fp->bp;
659 /* Return here if interrupt is disabled */
660 if (unlikely(atomic_read(&bp->intr_sem) != 0)) {
661 DP(NETIF_MSG_INTR, "called but intr_sem not 0, returning\n");
662 return IRQ_HANDLED;
665 DP(BNX2X_MSG_FP, "got an MSI-X interrupt on IDX:SB "
666 "[fp %d fw_sd %d igusb %d]\n",
667 fp->index, fp->fw_sb_id, fp->igu_sb_id);
668 bnx2x_ack_sb(bp, fp->igu_sb_id, USTORM_ID, 0, IGU_INT_DISABLE, 0);
670 #ifdef BNX2X_STOP_ON_ERROR
671 if (unlikely(bp->panic))
672 return IRQ_HANDLED;
673 #endif
675 /* Handle Rx and Tx according to MSI-X vector */
676 prefetch(fp->rx_cons_sb);
677 prefetch(fp->tx_cons_sb);
678 prefetch(&fp->sb_running_index[SM_RX_ID]);
679 napi_schedule(&bnx2x_fp(bp, fp->index, napi));
681 return IRQ_HANDLED;
684 /* HW Lock for shared dual port PHYs */
685 void bnx2x_acquire_phy_lock(struct bnx2x *bp)
687 mutex_lock(&bp->port.phy_mutex);
689 if (bp->port.need_hw_lock)
690 bnx2x_acquire_hw_lock(bp, HW_LOCK_RESOURCE_MDIO);
693 void bnx2x_release_phy_lock(struct bnx2x *bp)
695 if (bp->port.need_hw_lock)
696 bnx2x_release_hw_lock(bp, HW_LOCK_RESOURCE_MDIO);
698 mutex_unlock(&bp->port.phy_mutex);
701 /* calculates MF speed according to current linespeed and MF configuration */
702 u16 bnx2x_get_mf_speed(struct bnx2x *bp)
704 u16 line_speed = bp->link_vars.line_speed;
705 if (IS_MF(bp)) {
706 u16 maxCfg = (bp->mf_config[BP_VN(bp)] &
707 FUNC_MF_CFG_MAX_BW_MASK) >>
708 FUNC_MF_CFG_MAX_BW_SHIFT;
709 /* Calculate the current MAX line speed limit for the DCC
710 * capable devices
712 if (IS_MF_SD(bp)) {
713 u16 vn_max_rate = maxCfg * 100;
715 if (vn_max_rate < line_speed)
716 line_speed = vn_max_rate;
717 } else /* IS_MF_SI(bp)) */
718 line_speed = (line_speed * maxCfg) / 100;
721 return line_speed;
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");
729 return;
732 if (bp->link_vars.link_up) {
733 u16 line_speed;
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 = bnx2x_get_mf_speed(bp);
741 pr_cont("%d Mbps ", line_speed);
743 if (bp->link_vars.duplex == DUPLEX_FULL)
744 pr_cont("full duplex");
745 else
746 pr_cont("half duplex");
748 if (bp->link_vars.flow_ctrl != BNX2X_FLOW_CTRL_NONE) {
749 if (bp->link_vars.flow_ctrl & BNX2X_FLOW_CTRL_RX) {
750 pr_cont(", receive ");
751 if (bp->link_vars.flow_ctrl &
752 BNX2X_FLOW_CTRL_TX)
753 pr_cont("& transmit ");
754 } else {
755 pr_cont(", transmit ");
757 pr_cont("flow control ON");
759 pr_cont("\n");
761 } else { /* link_down */
762 netif_carrier_off(bp->dev);
763 netdev_err(bp->dev, "NIC Link is Down\n");
767 /* Returns the number of actually allocated BDs */
768 static inline int bnx2x_alloc_rx_bds(struct bnx2x_fastpath *fp,
769 int rx_ring_size)
771 struct bnx2x *bp = fp->bp;
772 u16 ring_prod, cqe_ring_prod;
773 int i;
775 fp->rx_comp_cons = 0;
776 cqe_ring_prod = ring_prod = 0;
777 for (i = 0; i < rx_ring_size; i++) {
778 if (bnx2x_alloc_rx_skb(bp, fp, ring_prod) < 0) {
779 BNX2X_ERR("was only able to allocate "
780 "%d rx skbs on queue[%d]\n", i, fp->index);
781 fp->eth_q_stats.rx_skb_alloc_failed++;
782 break;
784 ring_prod = NEXT_RX_IDX(ring_prod);
785 cqe_ring_prod = NEXT_RCQ_IDX(cqe_ring_prod);
786 WARN_ON(ring_prod <= i);
789 fp->rx_bd_prod = ring_prod;
790 /* Limit the CQE producer by the CQE ring size */
791 fp->rx_comp_prod = min_t(u16, NUM_RCQ_RINGS*RCQ_DESC_CNT,
792 cqe_ring_prod);
793 fp->rx_pkt = fp->rx_calls = 0;
795 return i;
798 static inline void bnx2x_alloc_rx_bd_ring(struct bnx2x_fastpath *fp)
800 struct bnx2x *bp = fp->bp;
801 int rx_ring_size = bp->rx_ring_size ? bp->rx_ring_size :
802 MAX_RX_AVAIL/bp->num_queues;
804 rx_ring_size = max_t(int, MIN_RX_AVAIL, rx_ring_size);
806 bnx2x_alloc_rx_bds(fp, rx_ring_size);
808 /* Warning!
809 * this will generate an interrupt (to the TSTORM)
810 * must only be done after chip is initialized
812 bnx2x_update_rx_prod(bp, fp, fp->rx_bd_prod, fp->rx_comp_prod,
813 fp->rx_sge_prod);
816 void bnx2x_init_rx_rings(struct bnx2x *bp)
818 int func = BP_FUNC(bp);
819 int max_agg_queues = CHIP_IS_E1(bp) ? ETH_MAX_AGGREGATION_QUEUES_E1 :
820 ETH_MAX_AGGREGATION_QUEUES_E1H;
821 u16 ring_prod;
822 int i, j;
824 bp->rx_buf_size = bp->dev->mtu + ETH_OVREHEAD + BNX2X_RX_ALIGN +
825 IP_HEADER_ALIGNMENT_PADDING;
827 DP(NETIF_MSG_IFUP,
828 "mtu %d rx_buf_size %d\n", bp->dev->mtu, bp->rx_buf_size);
830 for_each_rx_queue(bp, j) {
831 struct bnx2x_fastpath *fp = &bp->fp[j];
833 if (!fp->disable_tpa) {
834 for (i = 0; i < max_agg_queues; i++) {
835 fp->tpa_pool[i].skb =
836 netdev_alloc_skb(bp->dev, bp->rx_buf_size);
837 if (!fp->tpa_pool[i].skb) {
838 BNX2X_ERR("Failed to allocate TPA "
839 "skb pool for queue[%d] - "
840 "disabling TPA on this "
841 "queue!\n", j);
842 bnx2x_free_tpa_pool(bp, fp, i);
843 fp->disable_tpa = 1;
844 break;
846 dma_unmap_addr_set((struct sw_rx_bd *)
847 &bp->fp->tpa_pool[i],
848 mapping, 0);
849 fp->tpa_state[i] = BNX2X_TPA_STOP;
852 /* "next page" elements initialization */
853 bnx2x_set_next_page_sgl(fp);
855 /* set SGEs bit mask */
856 bnx2x_init_sge_ring_bit_mask(fp);
858 /* Allocate SGEs and initialize the ring elements */
859 for (i = 0, ring_prod = 0;
860 i < MAX_RX_SGE_CNT*NUM_RX_SGE_PAGES; i++) {
862 if (bnx2x_alloc_rx_sge(bp, fp, ring_prod) < 0) {
863 BNX2X_ERR("was only able to allocate "
864 "%d rx sges\n", i);
865 BNX2X_ERR("disabling TPA for"
866 " queue[%d]\n", j);
867 /* Cleanup already allocated elements */
868 bnx2x_free_rx_sge_range(bp,
869 fp, ring_prod);
870 bnx2x_free_tpa_pool(bp,
871 fp, max_agg_queues);
872 fp->disable_tpa = 1;
873 ring_prod = 0;
874 break;
876 ring_prod = NEXT_SGE_IDX(ring_prod);
879 fp->rx_sge_prod = ring_prod;
883 for_each_rx_queue(bp, j) {
884 struct bnx2x_fastpath *fp = &bp->fp[j];
886 fp->rx_bd_cons = 0;
888 bnx2x_set_next_page_rx_bd(fp);
890 /* CQ ring */
891 bnx2x_set_next_page_rx_cq(fp);
893 /* Allocate BDs and initialize BD ring */
894 bnx2x_alloc_rx_bd_ring(fp);
896 if (j != 0)
897 continue;
899 if (!CHIP_IS_E2(bp)) {
900 REG_WR(bp, BAR_USTRORM_INTMEM +
901 USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func),
902 U64_LO(fp->rx_comp_mapping));
903 REG_WR(bp, BAR_USTRORM_INTMEM +
904 USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func) + 4,
905 U64_HI(fp->rx_comp_mapping));
910 static void bnx2x_free_tx_skbs(struct bnx2x *bp)
912 int i;
914 for_each_tx_queue(bp, i) {
915 struct bnx2x_fastpath *fp = &bp->fp[i];
917 u16 bd_cons = fp->tx_bd_cons;
918 u16 sw_prod = fp->tx_pkt_prod;
919 u16 sw_cons = fp->tx_pkt_cons;
921 while (sw_cons != sw_prod) {
922 bd_cons = bnx2x_free_tx_pkt(bp, fp, TX_BD(sw_cons));
923 sw_cons++;
928 static void bnx2x_free_rx_skbs(struct bnx2x *bp)
930 int i, j;
932 for_each_rx_queue(bp, j) {
933 struct bnx2x_fastpath *fp = &bp->fp[j];
935 for (i = 0; i < NUM_RX_BD; i++) {
936 struct sw_rx_bd *rx_buf = &fp->rx_buf_ring[i];
937 struct sk_buff *skb = rx_buf->skb;
939 if (skb == NULL)
940 continue;
942 dma_unmap_single(&bp->pdev->dev,
943 dma_unmap_addr(rx_buf, mapping),
944 bp->rx_buf_size, DMA_FROM_DEVICE);
946 rx_buf->skb = NULL;
947 dev_kfree_skb(skb);
949 if (!fp->disable_tpa)
950 bnx2x_free_tpa_pool(bp, fp, CHIP_IS_E1(bp) ?
951 ETH_MAX_AGGREGATION_QUEUES_E1 :
952 ETH_MAX_AGGREGATION_QUEUES_E1H);
956 void bnx2x_free_skbs(struct bnx2x *bp)
958 bnx2x_free_tx_skbs(bp);
959 bnx2x_free_rx_skbs(bp);
962 static void bnx2x_free_msix_irqs(struct bnx2x *bp)
964 int i, offset = 1;
966 free_irq(bp->msix_table[0].vector, bp->dev);
967 DP(NETIF_MSG_IFDOWN, "released sp irq (%d)\n",
968 bp->msix_table[0].vector);
970 #ifdef BCM_CNIC
971 offset++;
972 #endif
973 for_each_eth_queue(bp, i) {
974 DP(NETIF_MSG_IFDOWN, "about to release fp #%d->%d irq "
975 "state %x\n", i, bp->msix_table[i + offset].vector,
976 bnx2x_fp(bp, i, state));
978 free_irq(bp->msix_table[i + offset].vector, &bp->fp[i]);
982 void bnx2x_free_irq(struct bnx2x *bp)
984 if (bp->flags & USING_MSIX_FLAG)
985 bnx2x_free_msix_irqs(bp);
986 else if (bp->flags & USING_MSI_FLAG)
987 free_irq(bp->pdev->irq, bp->dev);
988 else
989 free_irq(bp->pdev->irq, bp->dev);
992 int bnx2x_enable_msix(struct bnx2x *bp)
994 int msix_vec = 0, i, rc, req_cnt;
996 bp->msix_table[msix_vec].entry = msix_vec;
997 DP(NETIF_MSG_IFUP, "msix_table[0].entry = %d (slowpath)\n",
998 bp->msix_table[0].entry);
999 msix_vec++;
1001 #ifdef BCM_CNIC
1002 bp->msix_table[msix_vec].entry = msix_vec;
1003 DP(NETIF_MSG_IFUP, "msix_table[%d].entry = %d (CNIC)\n",
1004 bp->msix_table[msix_vec].entry, bp->msix_table[msix_vec].entry);
1005 msix_vec++;
1006 #endif
1007 for_each_eth_queue(bp, i) {
1008 bp->msix_table[msix_vec].entry = msix_vec;
1009 DP(NETIF_MSG_IFUP, "msix_table[%d].entry = %d "
1010 "(fastpath #%u)\n", msix_vec, msix_vec, i);
1011 msix_vec++;
1014 req_cnt = BNX2X_NUM_ETH_QUEUES(bp) + CNIC_CONTEXT_USE + 1;
1016 rc = pci_enable_msix(bp->pdev, &bp->msix_table[0], req_cnt);
1019 * reconfigure number of tx/rx queues according to available
1020 * MSI-X vectors
1022 if (rc >= BNX2X_MIN_MSIX_VEC_CNT) {
1023 /* how less vectors we will have? */
1024 int diff = req_cnt - rc;
1026 DP(NETIF_MSG_IFUP,
1027 "Trying to use less MSI-X vectors: %d\n", rc);
1029 rc = pci_enable_msix(bp->pdev, &bp->msix_table[0], rc);
1031 if (rc) {
1032 DP(NETIF_MSG_IFUP,
1033 "MSI-X is not attainable rc %d\n", rc);
1034 return rc;
1037 * decrease number of queues by number of unallocated entries
1039 bp->num_queues -= diff;
1041 DP(NETIF_MSG_IFUP, "New queue configuration set: %d\n",
1042 bp->num_queues);
1043 } else if (rc) {
1044 /* fall to INTx if not enough memory */
1045 if (rc == -ENOMEM)
1046 bp->flags |= DISABLE_MSI_FLAG;
1047 DP(NETIF_MSG_IFUP, "MSI-X is not attainable rc %d\n", rc);
1048 return rc;
1051 bp->flags |= USING_MSIX_FLAG;
1053 return 0;
1056 static int bnx2x_req_msix_irqs(struct bnx2x *bp)
1058 int i, rc, offset = 1;
1060 rc = request_irq(bp->msix_table[0].vector, bnx2x_msix_sp_int, 0,
1061 bp->dev->name, bp->dev);
1062 if (rc) {
1063 BNX2X_ERR("request sp irq failed\n");
1064 return -EBUSY;
1067 #ifdef BCM_CNIC
1068 offset++;
1069 #endif
1070 for_each_eth_queue(bp, i) {
1071 struct bnx2x_fastpath *fp = &bp->fp[i];
1072 snprintf(fp->name, sizeof(fp->name), "%s-fp-%d",
1073 bp->dev->name, i);
1075 rc = request_irq(bp->msix_table[offset].vector,
1076 bnx2x_msix_fp_int, 0, fp->name, fp);
1077 if (rc) {
1078 BNX2X_ERR("request fp #%d irq failed rc %d\n", i, rc);
1079 bnx2x_free_msix_irqs(bp);
1080 return -EBUSY;
1083 offset++;
1084 fp->state = BNX2X_FP_STATE_IRQ;
1087 i = BNX2X_NUM_ETH_QUEUES(bp);
1088 offset = 1 + CNIC_CONTEXT_USE;
1089 netdev_info(bp->dev, "using MSI-X IRQs: sp %d fp[%d] %d"
1090 " ... fp[%d] %d\n",
1091 bp->msix_table[0].vector,
1092 0, bp->msix_table[offset].vector,
1093 i - 1, bp->msix_table[offset + i - 1].vector);
1095 return 0;
1098 int bnx2x_enable_msi(struct bnx2x *bp)
1100 int rc;
1102 rc = pci_enable_msi(bp->pdev);
1103 if (rc) {
1104 DP(NETIF_MSG_IFUP, "MSI is not attainable\n");
1105 return -1;
1107 bp->flags |= USING_MSI_FLAG;
1109 return 0;
1112 static int bnx2x_req_irq(struct bnx2x *bp)
1114 unsigned long flags;
1115 int rc;
1117 if (bp->flags & USING_MSI_FLAG)
1118 flags = 0;
1119 else
1120 flags = IRQF_SHARED;
1122 rc = request_irq(bp->pdev->irq, bnx2x_interrupt, flags,
1123 bp->dev->name, bp->dev);
1124 if (!rc)
1125 bnx2x_fp(bp, 0, state) = BNX2X_FP_STATE_IRQ;
1127 return rc;
1130 static void bnx2x_napi_enable(struct bnx2x *bp)
1132 int i;
1134 for_each_napi_queue(bp, i)
1135 napi_enable(&bnx2x_fp(bp, i, napi));
1138 static void bnx2x_napi_disable(struct bnx2x *bp)
1140 int i;
1142 for_each_napi_queue(bp, i)
1143 napi_disable(&bnx2x_fp(bp, i, napi));
1146 void bnx2x_netif_start(struct bnx2x *bp)
1148 int intr_sem;
1150 intr_sem = atomic_dec_and_test(&bp->intr_sem);
1151 smp_wmb(); /* Ensure that bp->intr_sem update is SMP-safe */
1153 if (intr_sem) {
1154 if (netif_running(bp->dev)) {
1155 bnx2x_napi_enable(bp);
1156 bnx2x_int_enable(bp);
1157 if (bp->state == BNX2X_STATE_OPEN)
1158 netif_tx_wake_all_queues(bp->dev);
1163 void bnx2x_netif_stop(struct bnx2x *bp, int disable_hw)
1165 bnx2x_int_disable_sync(bp, disable_hw);
1166 bnx2x_napi_disable(bp);
1167 netif_tx_disable(bp->dev);
1170 u16 bnx2x_select_queue(struct net_device *dev, struct sk_buff *skb)
1172 #ifdef BCM_CNIC
1173 struct bnx2x *bp = netdev_priv(dev);
1174 if (NO_FCOE(bp))
1175 return skb_tx_hash(dev, skb);
1176 else {
1177 struct ethhdr *hdr = (struct ethhdr *)skb->data;
1178 u16 ether_type = ntohs(hdr->h_proto);
1180 /* Skip VLAN tag if present */
1181 if (ether_type == ETH_P_8021Q) {
1182 struct vlan_ethhdr *vhdr =
1183 (struct vlan_ethhdr *)skb->data;
1185 ether_type = ntohs(vhdr->h_vlan_encapsulated_proto);
1188 /* If ethertype is FCoE or FIP - use FCoE ring */
1189 if ((ether_type == ETH_P_FCOE) || (ether_type == ETH_P_FIP))
1190 return bnx2x_fcoe(bp, index);
1192 #endif
1193 /* Select a none-FCoE queue: if FCoE is enabled, exclude FCoE L2 ring
1195 return __skb_tx_hash(dev, skb,
1196 dev->real_num_tx_queues - FCOE_CONTEXT_USE);
1199 void bnx2x_set_num_queues(struct bnx2x *bp)
1201 switch (bp->multi_mode) {
1202 case ETH_RSS_MODE_DISABLED:
1203 bp->num_queues = 1;
1204 break;
1205 case ETH_RSS_MODE_REGULAR:
1206 bp->num_queues = bnx2x_calc_num_queues(bp);
1207 break;
1209 default:
1210 bp->num_queues = 1;
1211 break;
1214 /* Add special queues */
1215 bp->num_queues += NONE_ETH_CONTEXT_USE;
1218 #ifdef BCM_CNIC
1219 static inline void bnx2x_set_fcoe_eth_macs(struct bnx2x *bp)
1221 if (!NO_FCOE(bp)) {
1222 if (!IS_MF_SD(bp))
1223 bnx2x_set_fip_eth_mac_addr(bp, 1);
1224 bnx2x_set_all_enode_macs(bp, 1);
1225 bp->flags |= FCOE_MACS_SET;
1228 #endif
1230 static void bnx2x_release_firmware(struct bnx2x *bp)
1232 kfree(bp->init_ops_offsets);
1233 kfree(bp->init_ops);
1234 kfree(bp->init_data);
1235 release_firmware(bp->firmware);
1238 static inline int bnx2x_set_real_num_queues(struct bnx2x *bp)
1240 int rc, num = bp->num_queues;
1242 #ifdef BCM_CNIC
1243 if (NO_FCOE(bp))
1244 num -= FCOE_CONTEXT_USE;
1246 #endif
1247 netif_set_real_num_tx_queues(bp->dev, num);
1248 rc = netif_set_real_num_rx_queues(bp->dev, num);
1249 return rc;
1252 /* must be called with rtnl_lock */
1253 int bnx2x_nic_load(struct bnx2x *bp, int load_mode)
1255 u32 load_code;
1256 int i, rc;
1258 /* Set init arrays */
1259 rc = bnx2x_init_firmware(bp);
1260 if (rc) {
1261 BNX2X_ERR("Error loading firmware\n");
1262 return rc;
1265 #ifdef BNX2X_STOP_ON_ERROR
1266 if (unlikely(bp->panic))
1267 return -EPERM;
1268 #endif
1270 bp->state = BNX2X_STATE_OPENING_WAIT4_LOAD;
1272 /* must be called before memory allocation and HW init */
1273 bnx2x_ilt_set_info(bp);
1275 if (bnx2x_alloc_mem(bp))
1276 return -ENOMEM;
1278 rc = bnx2x_set_real_num_queues(bp);
1279 if (rc) {
1280 BNX2X_ERR("Unable to set real_num_queues\n");
1281 goto load_error0;
1284 for_each_queue(bp, i)
1285 bnx2x_fp(bp, i, disable_tpa) =
1286 ((bp->flags & TPA_ENABLE_FLAG) == 0);
1288 #ifdef BCM_CNIC
1289 /* We don't want TPA on FCoE L2 ring */
1290 bnx2x_fcoe(bp, disable_tpa) = 1;
1291 #endif
1292 bnx2x_napi_enable(bp);
1294 /* Send LOAD_REQUEST command to MCP
1295 Returns the type of LOAD command:
1296 if it is the first port to be initialized
1297 common blocks should be initialized, otherwise - not
1299 if (!BP_NOMCP(bp)) {
1300 load_code = bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_REQ, 0);
1301 if (!load_code) {
1302 BNX2X_ERR("MCP response failure, aborting\n");
1303 rc = -EBUSY;
1304 goto load_error1;
1306 if (load_code == FW_MSG_CODE_DRV_LOAD_REFUSED) {
1307 rc = -EBUSY; /* other port in diagnostic mode */
1308 goto load_error1;
1311 } else {
1312 int path = BP_PATH(bp);
1313 int port = BP_PORT(bp);
1315 DP(NETIF_MSG_IFUP, "NO MCP - load counts[%d] %d, %d, %d\n",
1316 path, load_count[path][0], load_count[path][1],
1317 load_count[path][2]);
1318 load_count[path][0]++;
1319 load_count[path][1 + port]++;
1320 DP(NETIF_MSG_IFUP, "NO MCP - new load counts[%d] %d, %d, %d\n",
1321 path, load_count[path][0], load_count[path][1],
1322 load_count[path][2]);
1323 if (load_count[path][0] == 1)
1324 load_code = FW_MSG_CODE_DRV_LOAD_COMMON;
1325 else if (load_count[path][1 + port] == 1)
1326 load_code = FW_MSG_CODE_DRV_LOAD_PORT;
1327 else
1328 load_code = FW_MSG_CODE_DRV_LOAD_FUNCTION;
1331 if ((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
1332 (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP) ||
1333 (load_code == FW_MSG_CODE_DRV_LOAD_PORT))
1334 bp->port.pmf = 1;
1335 else
1336 bp->port.pmf = 0;
1337 DP(NETIF_MSG_LINK, "pmf %d\n", bp->port.pmf);
1339 /* Initialize HW */
1340 rc = bnx2x_init_hw(bp, load_code);
1341 if (rc) {
1342 BNX2X_ERR("HW init failed, aborting\n");
1343 bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
1344 goto load_error2;
1347 /* Connect to IRQs */
1348 rc = bnx2x_setup_irqs(bp);
1349 if (rc) {
1350 bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
1351 goto load_error2;
1354 /* Setup NIC internals and enable interrupts */
1355 bnx2x_nic_init(bp, load_code);
1357 if (((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
1358 (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP)) &&
1359 (bp->common.shmem2_base))
1360 SHMEM2_WR(bp, dcc_support,
1361 (SHMEM_DCC_SUPPORT_DISABLE_ENABLE_PF_TLV |
1362 SHMEM_DCC_SUPPORT_BANDWIDTH_ALLOCATION_TLV));
1364 /* Send LOAD_DONE command to MCP */
1365 if (!BP_NOMCP(bp)) {
1366 load_code = bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
1367 if (!load_code) {
1368 BNX2X_ERR("MCP response failure, aborting\n");
1369 rc = -EBUSY;
1370 goto load_error3;
1374 bnx2x_dcbx_init(bp);
1376 bp->state = BNX2X_STATE_OPENING_WAIT4_PORT;
1378 rc = bnx2x_func_start(bp);
1379 if (rc) {
1380 BNX2X_ERR("Function start failed!\n");
1381 #ifndef BNX2X_STOP_ON_ERROR
1382 goto load_error3;
1383 #else
1384 bp->panic = 1;
1385 return -EBUSY;
1386 #endif
1389 rc = bnx2x_setup_client(bp, &bp->fp[0], 1 /* Leading */);
1390 if (rc) {
1391 BNX2X_ERR("Setup leading failed!\n");
1392 #ifndef BNX2X_STOP_ON_ERROR
1393 goto load_error3;
1394 #else
1395 bp->panic = 1;
1396 return -EBUSY;
1397 #endif
1400 if (!CHIP_IS_E1(bp) &&
1401 (bp->mf_config[BP_VN(bp)] & FUNC_MF_CFG_FUNC_DISABLED)) {
1402 DP(NETIF_MSG_IFUP, "mf_cfg function disabled\n");
1403 bp->flags |= MF_FUNC_DIS;
1406 #ifdef BCM_CNIC
1407 /* Enable Timer scan */
1408 REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + BP_PORT(bp)*4, 1);
1409 #endif
1411 for_each_nondefault_queue(bp, i) {
1412 rc = bnx2x_setup_client(bp, &bp->fp[i], 0);
1413 if (rc)
1414 #ifdef BCM_CNIC
1415 goto load_error4;
1416 #else
1417 goto load_error3;
1418 #endif
1421 /* Now when Clients are configured we are ready to work */
1422 bp->state = BNX2X_STATE_OPEN;
1424 #ifdef BCM_CNIC
1425 bnx2x_set_fcoe_eth_macs(bp);
1426 #endif
1428 bnx2x_set_eth_mac(bp, 1);
1430 if (bp->port.pmf)
1431 bnx2x_initial_phy_init(bp, load_mode);
1433 /* Start fast path */
1434 switch (load_mode) {
1435 case LOAD_NORMAL:
1436 /* Tx queue should be only reenabled */
1437 netif_tx_wake_all_queues(bp->dev);
1438 /* Initialize the receive filter. */
1439 bnx2x_set_rx_mode(bp->dev);
1440 break;
1442 case LOAD_OPEN:
1443 netif_tx_start_all_queues(bp->dev);
1444 smp_mb__after_clear_bit();
1445 /* Initialize the receive filter. */
1446 bnx2x_set_rx_mode(bp->dev);
1447 break;
1449 case LOAD_DIAG:
1450 /* Initialize the receive filter. */
1451 bnx2x_set_rx_mode(bp->dev);
1452 bp->state = BNX2X_STATE_DIAG;
1453 break;
1455 default:
1456 break;
1459 if (!bp->port.pmf)
1460 bnx2x__link_status_update(bp);
1462 /* start the timer */
1463 mod_timer(&bp->timer, jiffies + bp->current_interval);
1465 #ifdef BCM_CNIC
1466 bnx2x_setup_cnic_irq_info(bp);
1467 if (bp->state == BNX2X_STATE_OPEN)
1468 bnx2x_cnic_notify(bp, CNIC_CTL_START_CMD);
1469 #endif
1470 bnx2x_inc_load_cnt(bp);
1472 bnx2x_release_firmware(bp);
1474 return 0;
1476 #ifdef BCM_CNIC
1477 load_error4:
1478 /* Disable Timer scan */
1479 REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + BP_PORT(bp)*4, 0);
1480 #endif
1481 load_error3:
1482 bnx2x_int_disable_sync(bp, 1);
1484 /* Free SKBs, SGEs, TPA pool and driver internals */
1485 bnx2x_free_skbs(bp);
1486 for_each_rx_queue(bp, i)
1487 bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
1489 /* Release IRQs */
1490 bnx2x_free_irq(bp);
1491 load_error2:
1492 if (!BP_NOMCP(bp)) {
1493 bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_REQ_WOL_MCP, 0);
1494 bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_DONE, 0);
1497 bp->port.pmf = 0;
1498 load_error1:
1499 bnx2x_napi_disable(bp);
1500 load_error0:
1501 bnx2x_free_mem(bp);
1503 bnx2x_release_firmware(bp);
1505 return rc;
1508 /* must be called with rtnl_lock */
1509 int bnx2x_nic_unload(struct bnx2x *bp, int unload_mode)
1511 int i;
1513 if (bp->state == BNX2X_STATE_CLOSED) {
1514 /* Interface has been removed - nothing to recover */
1515 bp->recovery_state = BNX2X_RECOVERY_DONE;
1516 bp->is_leader = 0;
1517 bnx2x_release_hw_lock(bp, HW_LOCK_RESOURCE_RESERVED_08);
1518 smp_wmb();
1520 return -EINVAL;
1523 #ifdef BCM_CNIC
1524 bnx2x_cnic_notify(bp, CNIC_CTL_STOP_CMD);
1525 #endif
1526 bp->state = BNX2X_STATE_CLOSING_WAIT4_HALT;
1528 /* Set "drop all" */
1529 bp->rx_mode = BNX2X_RX_MODE_NONE;
1530 bnx2x_set_storm_rx_mode(bp);
1532 /* Stop Tx */
1533 bnx2x_tx_disable(bp);
1535 del_timer_sync(&bp->timer);
1537 SHMEM_WR(bp, func_mb[BP_FW_MB_IDX(bp)].drv_pulse_mb,
1538 (DRV_PULSE_ALWAYS_ALIVE | bp->fw_drv_pulse_wr_seq));
1540 bnx2x_stats_handle(bp, STATS_EVENT_STOP);
1542 /* Cleanup the chip if needed */
1543 if (unload_mode != UNLOAD_RECOVERY)
1544 bnx2x_chip_cleanup(bp, unload_mode);
1545 else {
1546 /* Disable HW interrupts, NAPI and Tx */
1547 bnx2x_netif_stop(bp, 1);
1549 /* Release IRQs */
1550 bnx2x_free_irq(bp);
1553 bp->port.pmf = 0;
1555 /* Free SKBs, SGEs, TPA pool and driver internals */
1556 bnx2x_free_skbs(bp);
1557 for_each_rx_queue(bp, i)
1558 bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
1560 bnx2x_free_mem(bp);
1562 bp->state = BNX2X_STATE_CLOSED;
1564 /* The last driver must disable a "close the gate" if there is no
1565 * parity attention or "process kill" pending.
1567 if ((!bnx2x_dec_load_cnt(bp)) && (!bnx2x_chk_parity_attn(bp)) &&
1568 bnx2x_reset_is_done(bp))
1569 bnx2x_disable_close_the_gate(bp);
1571 /* Reset MCP mail box sequence if there is on going recovery */
1572 if (unload_mode == UNLOAD_RECOVERY)
1573 bp->fw_seq = 0;
1575 return 0;
1578 int bnx2x_set_power_state(struct bnx2x *bp, pci_power_t state)
1580 u16 pmcsr;
1582 /* If there is no power capability, silently succeed */
1583 if (!bp->pm_cap) {
1584 DP(NETIF_MSG_HW, "No power capability. Breaking.\n");
1585 return 0;
1588 pci_read_config_word(bp->pdev, bp->pm_cap + PCI_PM_CTRL, &pmcsr);
1590 switch (state) {
1591 case PCI_D0:
1592 pci_write_config_word(bp->pdev, bp->pm_cap + PCI_PM_CTRL,
1593 ((pmcsr & ~PCI_PM_CTRL_STATE_MASK) |
1594 PCI_PM_CTRL_PME_STATUS));
1596 if (pmcsr & PCI_PM_CTRL_STATE_MASK)
1597 /* delay required during transition out of D3hot */
1598 msleep(20);
1599 break;
1601 case PCI_D3hot:
1602 /* If there are other clients above don't
1603 shut down the power */
1604 if (atomic_read(&bp->pdev->enable_cnt) != 1)
1605 return 0;
1606 /* Don't shut down the power for emulation and FPGA */
1607 if (CHIP_REV_IS_SLOW(bp))
1608 return 0;
1610 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
1611 pmcsr |= 3;
1613 if (bp->wol)
1614 pmcsr |= PCI_PM_CTRL_PME_ENABLE;
1616 pci_write_config_word(bp->pdev, bp->pm_cap + PCI_PM_CTRL,
1617 pmcsr);
1619 /* No more memory access after this point until
1620 * device is brought back to D0.
1622 break;
1624 default:
1625 return -EINVAL;
1627 return 0;
1631 * net_device service functions
1633 int bnx2x_poll(struct napi_struct *napi, int budget)
1635 int work_done = 0;
1636 struct bnx2x_fastpath *fp = container_of(napi, struct bnx2x_fastpath,
1637 napi);
1638 struct bnx2x *bp = fp->bp;
1640 while (1) {
1641 #ifdef BNX2X_STOP_ON_ERROR
1642 if (unlikely(bp->panic)) {
1643 napi_complete(napi);
1644 return 0;
1646 #endif
1648 if (bnx2x_has_tx_work(fp))
1649 bnx2x_tx_int(fp);
1651 if (bnx2x_has_rx_work(fp)) {
1652 work_done += bnx2x_rx_int(fp, budget - work_done);
1654 /* must not complete if we consumed full budget */
1655 if (work_done >= budget)
1656 break;
1659 /* Fall out from the NAPI loop if needed */
1660 if (!(bnx2x_has_rx_work(fp) || bnx2x_has_tx_work(fp))) {
1661 #ifdef BCM_CNIC
1662 /* No need to update SB for FCoE L2 ring as long as
1663 * it's connected to the default SB and the SB
1664 * has been updated when NAPI was scheduled.
1666 if (IS_FCOE_FP(fp)) {
1667 napi_complete(napi);
1668 break;
1670 #endif
1672 bnx2x_update_fpsb_idx(fp);
1673 /* bnx2x_has_rx_work() reads the status block,
1674 * thus we need to ensure that status block indices
1675 * have been actually read (bnx2x_update_fpsb_idx)
1676 * prior to this check (bnx2x_has_rx_work) so that
1677 * we won't write the "newer" value of the status block
1678 * to IGU (if there was a DMA right after
1679 * bnx2x_has_rx_work and if there is no rmb, the memory
1680 * reading (bnx2x_update_fpsb_idx) may be postponed
1681 * to right before bnx2x_ack_sb). In this case there
1682 * will never be another interrupt until there is
1683 * another update of the status block, while there
1684 * is still unhandled work.
1686 rmb();
1688 if (!(bnx2x_has_rx_work(fp) || bnx2x_has_tx_work(fp))) {
1689 napi_complete(napi);
1690 /* Re-enable interrupts */
1691 DP(NETIF_MSG_HW,
1692 "Update index to %d\n", fp->fp_hc_idx);
1693 bnx2x_ack_sb(bp, fp->igu_sb_id, USTORM_ID,
1694 le16_to_cpu(fp->fp_hc_idx),
1695 IGU_INT_ENABLE, 1);
1696 break;
1701 return work_done;
1704 /* we split the first BD into headers and data BDs
1705 * to ease the pain of our fellow microcode engineers
1706 * we use one mapping for both BDs
1707 * So far this has only been observed to happen
1708 * in Other Operating Systems(TM)
1710 static noinline u16 bnx2x_tx_split(struct bnx2x *bp,
1711 struct bnx2x_fastpath *fp,
1712 struct sw_tx_bd *tx_buf,
1713 struct eth_tx_start_bd **tx_bd, u16 hlen,
1714 u16 bd_prod, int nbd)
1716 struct eth_tx_start_bd *h_tx_bd = *tx_bd;
1717 struct eth_tx_bd *d_tx_bd;
1718 dma_addr_t mapping;
1719 int old_len = le16_to_cpu(h_tx_bd->nbytes);
1721 /* first fix first BD */
1722 h_tx_bd->nbd = cpu_to_le16(nbd);
1723 h_tx_bd->nbytes = cpu_to_le16(hlen);
1725 DP(NETIF_MSG_TX_QUEUED, "TSO split header size is %d "
1726 "(%x:%x) nbd %d\n", h_tx_bd->nbytes, h_tx_bd->addr_hi,
1727 h_tx_bd->addr_lo, h_tx_bd->nbd);
1729 /* now get a new data BD
1730 * (after the pbd) and fill it */
1731 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
1732 d_tx_bd = &fp->tx_desc_ring[bd_prod].reg_bd;
1734 mapping = HILO_U64(le32_to_cpu(h_tx_bd->addr_hi),
1735 le32_to_cpu(h_tx_bd->addr_lo)) + hlen;
1737 d_tx_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
1738 d_tx_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
1739 d_tx_bd->nbytes = cpu_to_le16(old_len - hlen);
1741 /* this marks the BD as one that has no individual mapping */
1742 tx_buf->flags |= BNX2X_TSO_SPLIT_BD;
1744 DP(NETIF_MSG_TX_QUEUED,
1745 "TSO split data size is %d (%x:%x)\n",
1746 d_tx_bd->nbytes, d_tx_bd->addr_hi, d_tx_bd->addr_lo);
1748 /* update tx_bd */
1749 *tx_bd = (struct eth_tx_start_bd *)d_tx_bd;
1751 return bd_prod;
1754 static inline u16 bnx2x_csum_fix(unsigned char *t_header, u16 csum, s8 fix)
1756 if (fix > 0)
1757 csum = (u16) ~csum_fold(csum_sub(csum,
1758 csum_partial(t_header - fix, fix, 0)));
1760 else if (fix < 0)
1761 csum = (u16) ~csum_fold(csum_add(csum,
1762 csum_partial(t_header, -fix, 0)));
1764 return swab16(csum);
1767 static inline u32 bnx2x_xmit_type(struct bnx2x *bp, struct sk_buff *skb)
1769 u32 rc;
1771 if (skb->ip_summed != CHECKSUM_PARTIAL)
1772 rc = XMIT_PLAIN;
1774 else {
1775 if (vlan_get_protocol(skb) == htons(ETH_P_IPV6)) {
1776 rc = XMIT_CSUM_V6;
1777 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
1778 rc |= XMIT_CSUM_TCP;
1780 } else {
1781 rc = XMIT_CSUM_V4;
1782 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
1783 rc |= XMIT_CSUM_TCP;
1787 if (skb_is_gso_v6(skb))
1788 rc |= XMIT_GSO_V6 | XMIT_CSUM_TCP | XMIT_CSUM_V6;
1789 else if (skb_is_gso(skb))
1790 rc |= XMIT_GSO_V4 | XMIT_CSUM_V4 | XMIT_CSUM_TCP;
1792 return rc;
1795 #if (MAX_SKB_FRAGS >= MAX_FETCH_BD - 3)
1796 /* check if packet requires linearization (packet is too fragmented)
1797 no need to check fragmentation if page size > 8K (there will be no
1798 violation to FW restrictions) */
1799 static int bnx2x_pkt_req_lin(struct bnx2x *bp, struct sk_buff *skb,
1800 u32 xmit_type)
1802 int to_copy = 0;
1803 int hlen = 0;
1804 int first_bd_sz = 0;
1806 /* 3 = 1 (for linear data BD) + 2 (for PBD and last BD) */
1807 if (skb_shinfo(skb)->nr_frags >= (MAX_FETCH_BD - 3)) {
1809 if (xmit_type & XMIT_GSO) {
1810 unsigned short lso_mss = skb_shinfo(skb)->gso_size;
1811 /* Check if LSO packet needs to be copied:
1812 3 = 1 (for headers BD) + 2 (for PBD and last BD) */
1813 int wnd_size = MAX_FETCH_BD - 3;
1814 /* Number of windows to check */
1815 int num_wnds = skb_shinfo(skb)->nr_frags - wnd_size;
1816 int wnd_idx = 0;
1817 int frag_idx = 0;
1818 u32 wnd_sum = 0;
1820 /* Headers length */
1821 hlen = (int)(skb_transport_header(skb) - skb->data) +
1822 tcp_hdrlen(skb);
1824 /* Amount of data (w/o headers) on linear part of SKB*/
1825 first_bd_sz = skb_headlen(skb) - hlen;
1827 wnd_sum = first_bd_sz;
1829 /* Calculate the first sum - it's special */
1830 for (frag_idx = 0; frag_idx < wnd_size - 1; frag_idx++)
1831 wnd_sum +=
1832 skb_shinfo(skb)->frags[frag_idx].size;
1834 /* If there was data on linear skb data - check it */
1835 if (first_bd_sz > 0) {
1836 if (unlikely(wnd_sum < lso_mss)) {
1837 to_copy = 1;
1838 goto exit_lbl;
1841 wnd_sum -= first_bd_sz;
1844 /* Others are easier: run through the frag list and
1845 check all windows */
1846 for (wnd_idx = 0; wnd_idx <= num_wnds; wnd_idx++) {
1847 wnd_sum +=
1848 skb_shinfo(skb)->frags[wnd_idx + wnd_size - 1].size;
1850 if (unlikely(wnd_sum < lso_mss)) {
1851 to_copy = 1;
1852 break;
1854 wnd_sum -=
1855 skb_shinfo(skb)->frags[wnd_idx].size;
1857 } else {
1858 /* in non-LSO too fragmented packet should always
1859 be linearized */
1860 to_copy = 1;
1864 exit_lbl:
1865 if (unlikely(to_copy))
1866 DP(NETIF_MSG_TX_QUEUED,
1867 "Linearization IS REQUIRED for %s packet. "
1868 "num_frags %d hlen %d first_bd_sz %d\n",
1869 (xmit_type & XMIT_GSO) ? "LSO" : "non-LSO",
1870 skb_shinfo(skb)->nr_frags, hlen, first_bd_sz);
1872 return to_copy;
1874 #endif
1876 static inline void bnx2x_set_pbd_gso_e2(struct sk_buff *skb, u32 *parsing_data,
1877 u32 xmit_type)
1879 *parsing_data |= (skb_shinfo(skb)->gso_size <<
1880 ETH_TX_PARSE_BD_E2_LSO_MSS_SHIFT) &
1881 ETH_TX_PARSE_BD_E2_LSO_MSS;
1882 if ((xmit_type & XMIT_GSO_V6) &&
1883 (ipv6_hdr(skb)->nexthdr == NEXTHDR_IPV6))
1884 *parsing_data |= ETH_TX_PARSE_BD_E2_IPV6_WITH_EXT_HDR;
1888 * Update PBD in GSO case.
1890 * @param skb
1891 * @param tx_start_bd
1892 * @param pbd
1893 * @param xmit_type
1895 static inline void bnx2x_set_pbd_gso(struct sk_buff *skb,
1896 struct eth_tx_parse_bd_e1x *pbd,
1897 u32 xmit_type)
1899 pbd->lso_mss = cpu_to_le16(skb_shinfo(skb)->gso_size);
1900 pbd->tcp_send_seq = swab32(tcp_hdr(skb)->seq);
1901 pbd->tcp_flags = pbd_tcp_flags(skb);
1903 if (xmit_type & XMIT_GSO_V4) {
1904 pbd->ip_id = swab16(ip_hdr(skb)->id);
1905 pbd->tcp_pseudo_csum =
1906 swab16(~csum_tcpudp_magic(ip_hdr(skb)->saddr,
1907 ip_hdr(skb)->daddr,
1908 0, IPPROTO_TCP, 0));
1910 } else
1911 pbd->tcp_pseudo_csum =
1912 swab16(~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
1913 &ipv6_hdr(skb)->daddr,
1914 0, IPPROTO_TCP, 0));
1916 pbd->global_data |= ETH_TX_PARSE_BD_E1X_PSEUDO_CS_WITHOUT_LEN;
1921 * @param skb
1922 * @param tx_start_bd
1923 * @param pbd_e2
1924 * @param xmit_type
1926 * @return header len
1928 static inline u8 bnx2x_set_pbd_csum_e2(struct bnx2x *bp, struct sk_buff *skb,
1929 u32 *parsing_data, u32 xmit_type)
1931 *parsing_data |= ((tcp_hdrlen(skb)/4) <<
1932 ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW_SHIFT) &
1933 ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW;
1935 *parsing_data |= ((((u8 *)tcp_hdr(skb) - skb->data) / 2) <<
1936 ETH_TX_PARSE_BD_E2_TCP_HDR_START_OFFSET_W_SHIFT) &
1937 ETH_TX_PARSE_BD_E2_TCP_HDR_START_OFFSET_W;
1939 return skb_transport_header(skb) + tcp_hdrlen(skb) - skb->data;
1944 * @param skb
1945 * @param tx_start_bd
1946 * @param pbd
1947 * @param xmit_type
1949 * @return Header length
1951 static inline u8 bnx2x_set_pbd_csum(struct bnx2x *bp, struct sk_buff *skb,
1952 struct eth_tx_parse_bd_e1x *pbd,
1953 u32 xmit_type)
1955 u8 hlen = (skb_network_header(skb) - skb->data) / 2;
1957 /* for now NS flag is not used in Linux */
1958 pbd->global_data =
1959 (hlen | ((skb->protocol == cpu_to_be16(ETH_P_8021Q)) <<
1960 ETH_TX_PARSE_BD_E1X_LLC_SNAP_EN_SHIFT));
1962 pbd->ip_hlen_w = (skb_transport_header(skb) -
1963 skb_network_header(skb)) / 2;
1965 hlen += pbd->ip_hlen_w + tcp_hdrlen(skb) / 2;
1967 pbd->total_hlen_w = cpu_to_le16(hlen);
1968 hlen = hlen*2;
1970 if (xmit_type & XMIT_CSUM_TCP) {
1971 pbd->tcp_pseudo_csum = swab16(tcp_hdr(skb)->check);
1973 } else {
1974 s8 fix = SKB_CS_OFF(skb); /* signed! */
1976 DP(NETIF_MSG_TX_QUEUED,
1977 "hlen %d fix %d csum before fix %x\n",
1978 le16_to_cpu(pbd->total_hlen_w), fix, SKB_CS(skb));
1980 /* HW bug: fixup the CSUM */
1981 pbd->tcp_pseudo_csum =
1982 bnx2x_csum_fix(skb_transport_header(skb),
1983 SKB_CS(skb), fix);
1985 DP(NETIF_MSG_TX_QUEUED, "csum after fix %x\n",
1986 pbd->tcp_pseudo_csum);
1989 return hlen;
1992 /* called with netif_tx_lock
1993 * bnx2x_tx_int() runs without netif_tx_lock unless it needs to call
1994 * netif_wake_queue()
1996 netdev_tx_t bnx2x_start_xmit(struct sk_buff *skb, struct net_device *dev)
1998 struct bnx2x *bp = netdev_priv(dev);
1999 struct bnx2x_fastpath *fp;
2000 struct netdev_queue *txq;
2001 struct sw_tx_bd *tx_buf;
2002 struct eth_tx_start_bd *tx_start_bd;
2003 struct eth_tx_bd *tx_data_bd, *total_pkt_bd = NULL;
2004 struct eth_tx_parse_bd_e1x *pbd_e1x = NULL;
2005 struct eth_tx_parse_bd_e2 *pbd_e2 = NULL;
2006 u32 pbd_e2_parsing_data = 0;
2007 u16 pkt_prod, bd_prod;
2008 int nbd, fp_index;
2009 dma_addr_t mapping;
2010 u32 xmit_type = bnx2x_xmit_type(bp, skb);
2011 int i;
2012 u8 hlen = 0;
2013 __le16 pkt_size = 0;
2014 struct ethhdr *eth;
2015 u8 mac_type = UNICAST_ADDRESS;
2017 #ifdef BNX2X_STOP_ON_ERROR
2018 if (unlikely(bp->panic))
2019 return NETDEV_TX_BUSY;
2020 #endif
2022 fp_index = skb_get_queue_mapping(skb);
2023 txq = netdev_get_tx_queue(dev, fp_index);
2025 fp = &bp->fp[fp_index];
2027 if (unlikely(bnx2x_tx_avail(fp) < (skb_shinfo(skb)->nr_frags + 3))) {
2028 fp->eth_q_stats.driver_xoff++;
2029 netif_tx_stop_queue(txq);
2030 BNX2X_ERR("BUG! Tx ring full when queue awake!\n");
2031 return NETDEV_TX_BUSY;
2034 DP(NETIF_MSG_TX_QUEUED, "queue[%d]: SKB: summed %x protocol %x "
2035 "protocol(%x,%x) gso type %x xmit_type %x\n",
2036 fp_index, skb->ip_summed, skb->protocol, ipv6_hdr(skb)->nexthdr,
2037 ip_hdr(skb)->protocol, skb_shinfo(skb)->gso_type, xmit_type);
2039 eth = (struct ethhdr *)skb->data;
2041 /* set flag according to packet type (UNICAST_ADDRESS is default)*/
2042 if (unlikely(is_multicast_ether_addr(eth->h_dest))) {
2043 if (is_broadcast_ether_addr(eth->h_dest))
2044 mac_type = BROADCAST_ADDRESS;
2045 else
2046 mac_type = MULTICAST_ADDRESS;
2049 #if (MAX_SKB_FRAGS >= MAX_FETCH_BD - 3)
2050 /* First, check if we need to linearize the skb (due to FW
2051 restrictions). No need to check fragmentation if page size > 8K
2052 (there will be no violation to FW restrictions) */
2053 if (bnx2x_pkt_req_lin(bp, skb, xmit_type)) {
2054 /* Statistics of linearization */
2055 bp->lin_cnt++;
2056 if (skb_linearize(skb) != 0) {
2057 DP(NETIF_MSG_TX_QUEUED, "SKB linearization failed - "
2058 "silently dropping this SKB\n");
2059 dev_kfree_skb_any(skb);
2060 return NETDEV_TX_OK;
2063 #endif
2066 Please read carefully. First we use one BD which we mark as start,
2067 then we have a parsing info BD (used for TSO or xsum),
2068 and only then we have the rest of the TSO BDs.
2069 (don't forget to mark the last one as last,
2070 and to unmap only AFTER you write to the BD ...)
2071 And above all, all pdb sizes are in words - NOT DWORDS!
2074 pkt_prod = fp->tx_pkt_prod++;
2075 bd_prod = TX_BD(fp->tx_bd_prod);
2077 /* get a tx_buf and first BD */
2078 tx_buf = &fp->tx_buf_ring[TX_BD(pkt_prod)];
2079 tx_start_bd = &fp->tx_desc_ring[bd_prod].start_bd;
2081 tx_start_bd->bd_flags.as_bitfield = ETH_TX_BD_FLAGS_START_BD;
2082 SET_FLAG(tx_start_bd->general_data, ETH_TX_START_BD_ETH_ADDR_TYPE,
2083 mac_type);
2085 /* header nbd */
2086 SET_FLAG(tx_start_bd->general_data, ETH_TX_START_BD_HDR_NBDS, 1);
2088 /* remember the first BD of the packet */
2089 tx_buf->first_bd = fp->tx_bd_prod;
2090 tx_buf->skb = skb;
2091 tx_buf->flags = 0;
2093 DP(NETIF_MSG_TX_QUEUED,
2094 "sending pkt %u @%p next_idx %u bd %u @%p\n",
2095 pkt_prod, tx_buf, fp->tx_pkt_prod, bd_prod, tx_start_bd);
2097 if (vlan_tx_tag_present(skb)) {
2098 tx_start_bd->vlan_or_ethertype =
2099 cpu_to_le16(vlan_tx_tag_get(skb));
2100 tx_start_bd->bd_flags.as_bitfield |=
2101 (X_ETH_OUTBAND_VLAN << ETH_TX_BD_FLAGS_VLAN_MODE_SHIFT);
2102 } else
2103 tx_start_bd->vlan_or_ethertype = cpu_to_le16(pkt_prod);
2105 /* turn on parsing and get a BD */
2106 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
2108 if (xmit_type & XMIT_CSUM) {
2109 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_L4_CSUM;
2111 if (xmit_type & XMIT_CSUM_V4)
2112 tx_start_bd->bd_flags.as_bitfield |=
2113 ETH_TX_BD_FLAGS_IP_CSUM;
2114 else
2115 tx_start_bd->bd_flags.as_bitfield |=
2116 ETH_TX_BD_FLAGS_IPV6;
2118 if (!(xmit_type & XMIT_CSUM_TCP))
2119 tx_start_bd->bd_flags.as_bitfield |=
2120 ETH_TX_BD_FLAGS_IS_UDP;
2123 if (CHIP_IS_E2(bp)) {
2124 pbd_e2 = &fp->tx_desc_ring[bd_prod].parse_bd_e2;
2125 memset(pbd_e2, 0, sizeof(struct eth_tx_parse_bd_e2));
2126 /* Set PBD in checksum offload case */
2127 if (xmit_type & XMIT_CSUM)
2128 hlen = bnx2x_set_pbd_csum_e2(bp, skb,
2129 &pbd_e2_parsing_data,
2130 xmit_type);
2131 } else {
2132 pbd_e1x = &fp->tx_desc_ring[bd_prod].parse_bd_e1x;
2133 memset(pbd_e1x, 0, sizeof(struct eth_tx_parse_bd_e1x));
2134 /* Set PBD in checksum offload case */
2135 if (xmit_type & XMIT_CSUM)
2136 hlen = bnx2x_set_pbd_csum(bp, skb, pbd_e1x, xmit_type);
2140 /* Map skb linear data for DMA */
2141 mapping = dma_map_single(&bp->pdev->dev, skb->data,
2142 skb_headlen(skb), DMA_TO_DEVICE);
2144 /* Setup the data pointer of the first BD of the packet */
2145 tx_start_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
2146 tx_start_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
2147 nbd = skb_shinfo(skb)->nr_frags + 2; /* start_bd + pbd + frags */
2148 tx_start_bd->nbd = cpu_to_le16(nbd);
2149 tx_start_bd->nbytes = cpu_to_le16(skb_headlen(skb));
2150 pkt_size = tx_start_bd->nbytes;
2152 DP(NETIF_MSG_TX_QUEUED, "first bd @%p addr (%x:%x) nbd %d"
2153 " nbytes %d flags %x vlan %x\n",
2154 tx_start_bd, tx_start_bd->addr_hi, tx_start_bd->addr_lo,
2155 le16_to_cpu(tx_start_bd->nbd), le16_to_cpu(tx_start_bd->nbytes),
2156 tx_start_bd->bd_flags.as_bitfield,
2157 le16_to_cpu(tx_start_bd->vlan_or_ethertype));
2159 if (xmit_type & XMIT_GSO) {
2161 DP(NETIF_MSG_TX_QUEUED,
2162 "TSO packet len %d hlen %d total len %d tso size %d\n",
2163 skb->len, hlen, skb_headlen(skb),
2164 skb_shinfo(skb)->gso_size);
2166 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_SW_LSO;
2168 if (unlikely(skb_headlen(skb) > hlen))
2169 bd_prod = bnx2x_tx_split(bp, fp, tx_buf, &tx_start_bd,
2170 hlen, bd_prod, ++nbd);
2171 if (CHIP_IS_E2(bp))
2172 bnx2x_set_pbd_gso_e2(skb, &pbd_e2_parsing_data,
2173 xmit_type);
2174 else
2175 bnx2x_set_pbd_gso(skb, pbd_e1x, xmit_type);
2178 /* Set the PBD's parsing_data field if not zero
2179 * (for the chips newer than 57711).
2181 if (pbd_e2_parsing_data)
2182 pbd_e2->parsing_data = cpu_to_le32(pbd_e2_parsing_data);
2184 tx_data_bd = (struct eth_tx_bd *)tx_start_bd;
2186 /* Handle fragmented skb */
2187 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2188 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2190 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
2191 tx_data_bd = &fp->tx_desc_ring[bd_prod].reg_bd;
2192 if (total_pkt_bd == NULL)
2193 total_pkt_bd = &fp->tx_desc_ring[bd_prod].reg_bd;
2195 mapping = dma_map_page(&bp->pdev->dev, frag->page,
2196 frag->page_offset,
2197 frag->size, DMA_TO_DEVICE);
2199 tx_data_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
2200 tx_data_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
2201 tx_data_bd->nbytes = cpu_to_le16(frag->size);
2202 le16_add_cpu(&pkt_size, frag->size);
2204 DP(NETIF_MSG_TX_QUEUED,
2205 "frag %d bd @%p addr (%x:%x) nbytes %d\n",
2206 i, tx_data_bd, tx_data_bd->addr_hi, tx_data_bd->addr_lo,
2207 le16_to_cpu(tx_data_bd->nbytes));
2210 DP(NETIF_MSG_TX_QUEUED, "last bd @%p\n", tx_data_bd);
2212 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
2214 /* now send a tx doorbell, counting the next BD
2215 * if the packet contains or ends with it
2217 if (TX_BD_POFF(bd_prod) < nbd)
2218 nbd++;
2220 if (total_pkt_bd != NULL)
2221 total_pkt_bd->total_pkt_bytes = pkt_size;
2223 if (pbd_e1x)
2224 DP(NETIF_MSG_TX_QUEUED,
2225 "PBD (E1X) @%p ip_data %x ip_hlen %u ip_id %u lso_mss %u"
2226 " tcp_flags %x xsum %x seq %u hlen %u\n",
2227 pbd_e1x, pbd_e1x->global_data, pbd_e1x->ip_hlen_w,
2228 pbd_e1x->ip_id, pbd_e1x->lso_mss, pbd_e1x->tcp_flags,
2229 pbd_e1x->tcp_pseudo_csum, pbd_e1x->tcp_send_seq,
2230 le16_to_cpu(pbd_e1x->total_hlen_w));
2231 if (pbd_e2)
2232 DP(NETIF_MSG_TX_QUEUED,
2233 "PBD (E2) @%p dst %x %x %x src %x %x %x parsing_data %x\n",
2234 pbd_e2, pbd_e2->dst_mac_addr_hi, pbd_e2->dst_mac_addr_mid,
2235 pbd_e2->dst_mac_addr_lo, pbd_e2->src_mac_addr_hi,
2236 pbd_e2->src_mac_addr_mid, pbd_e2->src_mac_addr_lo,
2237 pbd_e2->parsing_data);
2238 DP(NETIF_MSG_TX_QUEUED, "doorbell: nbd %d bd %u\n", nbd, bd_prod);
2241 * Make sure that the BD data is updated before updating the producer
2242 * since FW might read the BD right after the producer is updated.
2243 * This is only applicable for weak-ordered memory model archs such
2244 * as IA-64. The following barrier is also mandatory since FW will
2245 * assumes packets must have BDs.
2247 wmb();
2249 fp->tx_db.data.prod += nbd;
2250 barrier();
2252 DOORBELL(bp, fp->cid, fp->tx_db.raw);
2254 mmiowb();
2256 fp->tx_bd_prod += nbd;
2258 if (unlikely(bnx2x_tx_avail(fp) < MAX_SKB_FRAGS + 3)) {
2259 netif_tx_stop_queue(txq);
2261 /* paired memory barrier is in bnx2x_tx_int(), we have to keep
2262 * ordering of set_bit() in netif_tx_stop_queue() and read of
2263 * fp->bd_tx_cons */
2264 smp_mb();
2266 fp->eth_q_stats.driver_xoff++;
2267 if (bnx2x_tx_avail(fp) >= MAX_SKB_FRAGS + 3)
2268 netif_tx_wake_queue(txq);
2270 fp->tx_pkt++;
2272 return NETDEV_TX_OK;
2275 /* called with rtnl_lock */
2276 int bnx2x_change_mac_addr(struct net_device *dev, void *p)
2278 struct sockaddr *addr = p;
2279 struct bnx2x *bp = netdev_priv(dev);
2281 if (!is_valid_ether_addr((u8 *)(addr->sa_data)))
2282 return -EINVAL;
2284 memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
2285 if (netif_running(dev))
2286 bnx2x_set_eth_mac(bp, 1);
2288 return 0;
2292 static int bnx2x_setup_irqs(struct bnx2x *bp)
2294 int rc = 0;
2295 if (bp->flags & USING_MSIX_FLAG) {
2296 rc = bnx2x_req_msix_irqs(bp);
2297 if (rc)
2298 return rc;
2299 } else {
2300 bnx2x_ack_int(bp);
2301 rc = bnx2x_req_irq(bp);
2302 if (rc) {
2303 BNX2X_ERR("IRQ request failed rc %d, aborting\n", rc);
2304 return rc;
2306 if (bp->flags & USING_MSI_FLAG) {
2307 bp->dev->irq = bp->pdev->irq;
2308 netdev_info(bp->dev, "using MSI IRQ %d\n",
2309 bp->pdev->irq);
2313 return 0;
2316 void bnx2x_free_mem_bp(struct bnx2x *bp)
2318 kfree(bp->fp);
2319 kfree(bp->msix_table);
2320 kfree(bp->ilt);
2323 int __devinit bnx2x_alloc_mem_bp(struct bnx2x *bp)
2325 struct bnx2x_fastpath *fp;
2326 struct msix_entry *tbl;
2327 struct bnx2x_ilt *ilt;
2329 /* fp array */
2330 fp = kzalloc(L2_FP_COUNT(bp->l2_cid_count)*sizeof(*fp), GFP_KERNEL);
2331 if (!fp)
2332 goto alloc_err;
2333 bp->fp = fp;
2335 /* msix table */
2336 tbl = kzalloc((FP_SB_COUNT(bp->l2_cid_count) + 1) * sizeof(*tbl),
2337 GFP_KERNEL);
2338 if (!tbl)
2339 goto alloc_err;
2340 bp->msix_table = tbl;
2342 /* ilt */
2343 ilt = kzalloc(sizeof(*ilt), GFP_KERNEL);
2344 if (!ilt)
2345 goto alloc_err;
2346 bp->ilt = ilt;
2348 return 0;
2349 alloc_err:
2350 bnx2x_free_mem_bp(bp);
2351 return -ENOMEM;
2355 /* called with rtnl_lock */
2356 int bnx2x_change_mtu(struct net_device *dev, int new_mtu)
2358 struct bnx2x *bp = netdev_priv(dev);
2359 int rc = 0;
2361 if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
2362 printk(KERN_ERR "Handling parity error recovery. Try again later\n");
2363 return -EAGAIN;
2366 if ((new_mtu > ETH_MAX_JUMBO_PACKET_SIZE) ||
2367 ((new_mtu + ETH_HLEN) < ETH_MIN_PACKET_SIZE))
2368 return -EINVAL;
2370 /* This does not race with packet allocation
2371 * because the actual alloc size is
2372 * only updated as part of load
2374 dev->mtu = new_mtu;
2376 if (netif_running(dev)) {
2377 bnx2x_nic_unload(bp, UNLOAD_NORMAL);
2378 rc = bnx2x_nic_load(bp, LOAD_NORMAL);
2381 return rc;
2384 void bnx2x_tx_timeout(struct net_device *dev)
2386 struct bnx2x *bp = netdev_priv(dev);
2388 #ifdef BNX2X_STOP_ON_ERROR
2389 if (!bp->panic)
2390 bnx2x_panic();
2391 #endif
2392 /* This allows the netif to be shutdown gracefully before resetting */
2393 schedule_delayed_work(&bp->reset_task, 0);
2396 int bnx2x_suspend(struct pci_dev *pdev, pm_message_t state)
2398 struct net_device *dev = pci_get_drvdata(pdev);
2399 struct bnx2x *bp;
2401 if (!dev) {
2402 dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
2403 return -ENODEV;
2405 bp = netdev_priv(dev);
2407 rtnl_lock();
2409 pci_save_state(pdev);
2411 if (!netif_running(dev)) {
2412 rtnl_unlock();
2413 return 0;
2416 netif_device_detach(dev);
2418 bnx2x_nic_unload(bp, UNLOAD_CLOSE);
2420 bnx2x_set_power_state(bp, pci_choose_state(pdev, state));
2422 rtnl_unlock();
2424 return 0;
2427 int bnx2x_resume(struct pci_dev *pdev)
2429 struct net_device *dev = pci_get_drvdata(pdev);
2430 struct bnx2x *bp;
2431 int rc;
2433 if (!dev) {
2434 dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
2435 return -ENODEV;
2437 bp = netdev_priv(dev);
2439 if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
2440 printk(KERN_ERR "Handling parity error recovery. Try again later\n");
2441 return -EAGAIN;
2444 rtnl_lock();
2446 pci_restore_state(pdev);
2448 if (!netif_running(dev)) {
2449 rtnl_unlock();
2450 return 0;
2453 bnx2x_set_power_state(bp, PCI_D0);
2454 netif_device_attach(dev);
2456 /* Since the chip was reset, clear the FW sequence number */
2457 bp->fw_seq = 0;
2458 rc = bnx2x_nic_load(bp, LOAD_OPEN);
2460 rtnl_unlock();
2462 return rc;