bnx2x: put start bd csum in separate function
[linux-2.6.git] / drivers / net / bnx2x / bnx2x_cmn.c
blob3a2a50c36b499dc1fd5a324607c2965d9cfa411d
1 /* bnx2x_cmn.c: Broadcom Everest network driver.
3 * Copyright (c) 2007-2011 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/interrupt.h>
21 #include <linux/ip.h>
22 #include <net/ipv6.h>
23 #include <net/ip6_checksum.h>
24 #include <linux/firmware.h>
25 #include <linux/prefetch.h>
26 #include "bnx2x_cmn.h"
28 #include "bnx2x_init.h"
30 static int bnx2x_setup_irqs(struct bnx2x *bp);
32 /**
33 * bnx2x_bz_fp - zero content of the fastpath structure.
35 * @bp: driver handle
36 * @index: fastpath index to be zeroed
38 * Makes sure the contents of the bp->fp[index].napi is kept
39 * intact.
41 static inline void bnx2x_bz_fp(struct bnx2x *bp, int index)
43 struct bnx2x_fastpath *fp = &bp->fp[index];
44 struct napi_struct orig_napi = fp->napi;
45 /* bzero bnx2x_fastpath contents */
46 memset(fp, 0, sizeof(*fp));
48 /* Restore the NAPI object as it has been already initialized */
49 fp->napi = orig_napi;
52 /**
53 * bnx2x_move_fp - move content of the fastpath structure.
55 * @bp: driver handle
56 * @from: source FP index
57 * @to: destination FP index
59 * Makes sure the contents of the bp->fp[to].napi is kept
60 * intact.
62 static inline void bnx2x_move_fp(struct bnx2x *bp, int from, int to)
64 struct bnx2x_fastpath *from_fp = &bp->fp[from];
65 struct bnx2x_fastpath *to_fp = &bp->fp[to];
66 struct napi_struct orig_napi = to_fp->napi;
67 /* Move bnx2x_fastpath contents */
68 memcpy(to_fp, from_fp, sizeof(*to_fp));
69 to_fp->index = to;
71 /* Restore the NAPI object as it has been already initialized */
72 to_fp->napi = orig_napi;
75 /* free skb in the packet ring at pos idx
76 * return idx of last bd freed
78 static u16 bnx2x_free_tx_pkt(struct bnx2x *bp, struct bnx2x_fastpath *fp,
79 u16 idx)
81 struct sw_tx_bd *tx_buf = &fp->tx_buf_ring[idx];
82 struct eth_tx_start_bd *tx_start_bd;
83 struct eth_tx_bd *tx_data_bd;
84 struct sk_buff *skb = tx_buf->skb;
85 u16 bd_idx = TX_BD(tx_buf->first_bd), new_cons;
86 int nbd;
88 /* prefetch skb end pointer to speedup dev_kfree_skb() */
89 prefetch(&skb->end);
91 DP(BNX2X_MSG_OFF, "pkt_idx %d buff @(%p)->skb %p\n",
92 idx, tx_buf, skb);
94 /* unmap first bd */
95 DP(BNX2X_MSG_OFF, "free bd_idx %d\n", bd_idx);
96 tx_start_bd = &fp->tx_desc_ring[bd_idx].start_bd;
97 dma_unmap_single(&bp->pdev->dev, BD_UNMAP_ADDR(tx_start_bd),
98 BD_UNMAP_LEN(tx_start_bd), DMA_TO_DEVICE);
100 nbd = le16_to_cpu(tx_start_bd->nbd) - 1;
101 #ifdef BNX2X_STOP_ON_ERROR
102 if ((nbd - 1) > (MAX_SKB_FRAGS + 2)) {
103 BNX2X_ERR("BAD nbd!\n");
104 bnx2x_panic();
106 #endif
107 new_cons = nbd + tx_buf->first_bd;
109 /* Get the next bd */
110 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
112 /* Skip a parse bd... */
113 --nbd;
114 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
116 /* ...and the TSO split header bd since they have no mapping */
117 if (tx_buf->flags & BNX2X_TSO_SPLIT_BD) {
118 --nbd;
119 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
122 /* now free frags */
123 while (nbd > 0) {
125 DP(BNX2X_MSG_OFF, "free frag bd_idx %d\n", bd_idx);
126 tx_data_bd = &fp->tx_desc_ring[bd_idx].reg_bd;
127 dma_unmap_page(&bp->pdev->dev, BD_UNMAP_ADDR(tx_data_bd),
128 BD_UNMAP_LEN(tx_data_bd), DMA_TO_DEVICE);
129 if (--nbd)
130 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
133 /* release skb */
134 WARN_ON(!skb);
135 dev_kfree_skb_any(skb);
136 tx_buf->first_bd = 0;
137 tx_buf->skb = NULL;
139 return new_cons;
142 int bnx2x_tx_int(struct bnx2x_fastpath *fp)
144 struct bnx2x *bp = fp->bp;
145 struct netdev_queue *txq;
146 u16 hw_cons, sw_cons, bd_cons = fp->tx_bd_cons;
148 #ifdef BNX2X_STOP_ON_ERROR
149 if (unlikely(bp->panic))
150 return -1;
151 #endif
153 txq = netdev_get_tx_queue(bp->dev, fp->index);
154 hw_cons = le16_to_cpu(*fp->tx_cons_sb);
155 sw_cons = fp->tx_pkt_cons;
157 while (sw_cons != hw_cons) {
158 u16 pkt_cons;
160 pkt_cons = TX_BD(sw_cons);
162 DP(NETIF_MSG_TX_DONE, "queue[%d]: hw_cons %u sw_cons %u "
163 " pkt_cons %u\n",
164 fp->index, hw_cons, sw_cons, pkt_cons);
166 bd_cons = bnx2x_free_tx_pkt(bp, fp, pkt_cons);
167 sw_cons++;
170 fp->tx_pkt_cons = sw_cons;
171 fp->tx_bd_cons = bd_cons;
173 /* Need to make the tx_bd_cons update visible to start_xmit()
174 * before checking for netif_tx_queue_stopped(). Without the
175 * memory barrier, there is a small possibility that
176 * start_xmit() will miss it and cause the queue to be stopped
177 * forever.
179 smp_mb();
181 if (unlikely(netif_tx_queue_stopped(txq))) {
182 /* Taking tx_lock() is needed to prevent reenabling the queue
183 * while it's empty. This could have happen if rx_action() gets
184 * suspended in bnx2x_tx_int() after the condition before
185 * netif_tx_wake_queue(), while tx_action (bnx2x_start_xmit()):
187 * stops the queue->sees fresh tx_bd_cons->releases the queue->
188 * sends some packets consuming the whole queue again->
189 * stops the queue
192 __netif_tx_lock(txq, smp_processor_id());
194 if ((netif_tx_queue_stopped(txq)) &&
195 (bp->state == BNX2X_STATE_OPEN) &&
196 (bnx2x_tx_avail(fp) >= MAX_SKB_FRAGS + 3))
197 netif_tx_wake_queue(txq);
199 __netif_tx_unlock(txq);
201 return 0;
204 static inline void bnx2x_update_last_max_sge(struct bnx2x_fastpath *fp,
205 u16 idx)
207 u16 last_max = fp->last_max_sge;
209 if (SUB_S16(idx, last_max) > 0)
210 fp->last_max_sge = idx;
213 static void bnx2x_update_sge_prod(struct bnx2x_fastpath *fp,
214 struct eth_fast_path_rx_cqe *fp_cqe)
216 struct bnx2x *bp = fp->bp;
217 u16 sge_len = SGE_PAGE_ALIGN(le16_to_cpu(fp_cqe->pkt_len) -
218 le16_to_cpu(fp_cqe->len_on_bd)) >>
219 SGE_PAGE_SHIFT;
220 u16 last_max, last_elem, first_elem;
221 u16 delta = 0;
222 u16 i;
224 if (!sge_len)
225 return;
227 /* First mark all used pages */
228 for (i = 0; i < sge_len; i++)
229 SGE_MASK_CLEAR_BIT(fp,
230 RX_SGE(le16_to_cpu(fp_cqe->sgl_or_raw_data.sgl[i])));
232 DP(NETIF_MSG_RX_STATUS, "fp_cqe->sgl[%d] = %d\n",
233 sge_len - 1, le16_to_cpu(fp_cqe->sgl_or_raw_data.sgl[sge_len - 1]));
235 /* Here we assume that the last SGE index is the biggest */
236 prefetch((void *)(fp->sge_mask));
237 bnx2x_update_last_max_sge(fp,
238 le16_to_cpu(fp_cqe->sgl_or_raw_data.sgl[sge_len - 1]));
240 last_max = RX_SGE(fp->last_max_sge);
241 last_elem = last_max >> RX_SGE_MASK_ELEM_SHIFT;
242 first_elem = RX_SGE(fp->rx_sge_prod) >> RX_SGE_MASK_ELEM_SHIFT;
244 /* If ring is not full */
245 if (last_elem + 1 != first_elem)
246 last_elem++;
248 /* Now update the prod */
249 for (i = first_elem; i != last_elem; i = NEXT_SGE_MASK_ELEM(i)) {
250 if (likely(fp->sge_mask[i]))
251 break;
253 fp->sge_mask[i] = RX_SGE_MASK_ELEM_ONE_MASK;
254 delta += RX_SGE_MASK_ELEM_SZ;
257 if (delta > 0) {
258 fp->rx_sge_prod += delta;
259 /* clear page-end entries */
260 bnx2x_clear_sge_mask_next_elems(fp);
263 DP(NETIF_MSG_RX_STATUS,
264 "fp->last_max_sge = %d fp->rx_sge_prod = %d\n",
265 fp->last_max_sge, fp->rx_sge_prod);
268 static void bnx2x_tpa_start(struct bnx2x_fastpath *fp, u16 queue,
269 struct sk_buff *skb, u16 cons, u16 prod)
271 struct bnx2x *bp = fp->bp;
272 struct sw_rx_bd *cons_rx_buf = &fp->rx_buf_ring[cons];
273 struct sw_rx_bd *prod_rx_buf = &fp->rx_buf_ring[prod];
274 struct eth_rx_bd *prod_bd = &fp->rx_desc_ring[prod];
275 dma_addr_t mapping;
277 /* move empty skb from pool to prod and map it */
278 prod_rx_buf->skb = fp->tpa_pool[queue].skb;
279 mapping = dma_map_single(&bp->pdev->dev, fp->tpa_pool[queue].skb->data,
280 fp->rx_buf_size, DMA_FROM_DEVICE);
281 dma_unmap_addr_set(prod_rx_buf, mapping, mapping);
283 /* move partial skb from cons to pool (don't unmap yet) */
284 fp->tpa_pool[queue] = *cons_rx_buf;
286 /* mark bin state as start - print error if current state != stop */
287 if (fp->tpa_state[queue] != BNX2X_TPA_STOP)
288 BNX2X_ERR("start of bin not in stop [%d]\n", queue);
290 fp->tpa_state[queue] = BNX2X_TPA_START;
292 /* point prod_bd to new skb */
293 prod_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
294 prod_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
296 #ifdef BNX2X_STOP_ON_ERROR
297 fp->tpa_queue_used |= (1 << queue);
298 #ifdef _ASM_GENERIC_INT_L64_H
299 DP(NETIF_MSG_RX_STATUS, "fp->tpa_queue_used = 0x%lx\n",
300 #else
301 DP(NETIF_MSG_RX_STATUS, "fp->tpa_queue_used = 0x%llx\n",
302 #endif
303 fp->tpa_queue_used);
304 #endif
307 /* Timestamp option length allowed for TPA aggregation:
309 * nop nop kind length echo val
311 #define TPA_TSTAMP_OPT_LEN 12
313 * bnx2x_set_lro_mss - calculate the approximate value of the MSS
315 * @bp: driver handle
316 * @parsing_flags: parsing flags from the START CQE
317 * @len_on_bd: total length of the first packet for the
318 * aggregation.
320 * Approximate value of the MSS for this aggregation calculated using
321 * the first packet of it.
323 static inline u16 bnx2x_set_lro_mss(struct bnx2x *bp, u16 parsing_flags,
324 u16 len_on_bd)
326 /* TPA arrgregation won't have an IP options and TCP options
327 * other than timestamp.
329 u16 hdrs_len = ETH_HLEN + sizeof(struct iphdr) + sizeof(struct tcphdr);
332 /* Check if there was a TCP timestamp, if there is it's will
333 * always be 12 bytes length: nop nop kind length echo val.
335 * Otherwise FW would close the aggregation.
337 if (parsing_flags & PARSING_FLAGS_TIME_STAMP_EXIST_FLAG)
338 hdrs_len += TPA_TSTAMP_OPT_LEN;
340 return len_on_bd - hdrs_len;
343 static int bnx2x_fill_frag_skb(struct bnx2x *bp, struct bnx2x_fastpath *fp,
344 struct sk_buff *skb,
345 struct eth_fast_path_rx_cqe *fp_cqe,
346 u16 cqe_idx, u16 parsing_flags)
348 struct sw_rx_page *rx_pg, old_rx_pg;
349 u16 len_on_bd = le16_to_cpu(fp_cqe->len_on_bd);
350 u32 i, frag_len, frag_size, pages;
351 int err;
352 int j;
354 frag_size = le16_to_cpu(fp_cqe->pkt_len) - len_on_bd;
355 pages = SGE_PAGE_ALIGN(frag_size) >> SGE_PAGE_SHIFT;
357 /* This is needed in order to enable forwarding support */
358 if (frag_size)
359 skb_shinfo(skb)->gso_size = bnx2x_set_lro_mss(bp, parsing_flags,
360 len_on_bd);
362 #ifdef BNX2X_STOP_ON_ERROR
363 if (pages > min_t(u32, 8, MAX_SKB_FRAGS)*SGE_PAGE_SIZE*PAGES_PER_SGE) {
364 BNX2X_ERR("SGL length is too long: %d. CQE index is %d\n",
365 pages, cqe_idx);
366 BNX2X_ERR("fp_cqe->pkt_len = %d fp_cqe->len_on_bd = %d\n",
367 fp_cqe->pkt_len, len_on_bd);
368 bnx2x_panic();
369 return -EINVAL;
371 #endif
373 /* Run through the SGL and compose the fragmented skb */
374 for (i = 0, j = 0; i < pages; i += PAGES_PER_SGE, j++) {
375 u16 sge_idx =
376 RX_SGE(le16_to_cpu(fp_cqe->sgl_or_raw_data.sgl[j]));
378 /* FW gives the indices of the SGE as if the ring is an array
379 (meaning that "next" element will consume 2 indices) */
380 frag_len = min(frag_size, (u32)(SGE_PAGE_SIZE*PAGES_PER_SGE));
381 rx_pg = &fp->rx_page_ring[sge_idx];
382 old_rx_pg = *rx_pg;
384 /* If we fail to allocate a substitute page, we simply stop
385 where we are and drop the whole packet */
386 err = bnx2x_alloc_rx_sge(bp, fp, sge_idx);
387 if (unlikely(err)) {
388 fp->eth_q_stats.rx_skb_alloc_failed++;
389 return err;
392 /* Unmap the page as we r going to pass it to the stack */
393 dma_unmap_page(&bp->pdev->dev,
394 dma_unmap_addr(&old_rx_pg, mapping),
395 SGE_PAGE_SIZE*PAGES_PER_SGE, DMA_FROM_DEVICE);
397 /* Add one frag and update the appropriate fields in the skb */
398 skb_fill_page_desc(skb, j, old_rx_pg.page, 0, frag_len);
400 skb->data_len += frag_len;
401 skb->truesize += frag_len;
402 skb->len += frag_len;
404 frag_size -= frag_len;
407 return 0;
410 static void bnx2x_tpa_stop(struct bnx2x *bp, struct bnx2x_fastpath *fp,
411 u16 queue, int pad, int len, union eth_rx_cqe *cqe,
412 u16 cqe_idx)
414 struct sw_rx_bd *rx_buf = &fp->tpa_pool[queue];
415 struct sk_buff *skb = rx_buf->skb;
416 /* alloc new skb */
417 struct sk_buff *new_skb = netdev_alloc_skb(bp->dev, fp->rx_buf_size);
419 /* Unmap skb in the pool anyway, as we are going to change
420 pool entry status to BNX2X_TPA_STOP even if new skb allocation
421 fails. */
422 dma_unmap_single(&bp->pdev->dev, dma_unmap_addr(rx_buf, mapping),
423 fp->rx_buf_size, DMA_FROM_DEVICE);
425 if (likely(new_skb)) {
426 /* fix ip xsum and give it to the stack */
427 /* (no need to map the new skb) */
428 u16 parsing_flags =
429 le16_to_cpu(cqe->fast_path_cqe.pars_flags.flags);
431 prefetch(skb);
432 prefetch(((char *)(skb)) + L1_CACHE_BYTES);
434 #ifdef BNX2X_STOP_ON_ERROR
435 if (pad + len > fp->rx_buf_size) {
436 BNX2X_ERR("skb_put is about to fail... "
437 "pad %d len %d rx_buf_size %d\n",
438 pad, len, fp->rx_buf_size);
439 bnx2x_panic();
440 return;
442 #endif
444 skb_reserve(skb, pad);
445 skb_put(skb, len);
447 skb->protocol = eth_type_trans(skb, bp->dev);
448 skb->ip_summed = CHECKSUM_UNNECESSARY;
451 struct iphdr *iph;
453 iph = (struct iphdr *)skb->data;
454 iph->check = 0;
455 iph->check = ip_fast_csum((u8 *)iph, iph->ihl);
458 if (!bnx2x_fill_frag_skb(bp, fp, skb,
459 &cqe->fast_path_cqe, cqe_idx,
460 parsing_flags)) {
461 if (parsing_flags & PARSING_FLAGS_VLAN)
462 __vlan_hwaccel_put_tag(skb,
463 le16_to_cpu(cqe->fast_path_cqe.
464 vlan_tag));
465 napi_gro_receive(&fp->napi, skb);
466 } else {
467 DP(NETIF_MSG_RX_STATUS, "Failed to allocate new pages"
468 " - dropping packet!\n");
469 dev_kfree_skb_any(skb);
473 /* put new skb in bin */
474 fp->tpa_pool[queue].skb = new_skb;
476 } else {
477 /* else drop the packet and keep the buffer in the bin */
478 DP(NETIF_MSG_RX_STATUS,
479 "Failed to allocate new skb - dropping packet!\n");
480 fp->eth_q_stats.rx_skb_alloc_failed++;
483 fp->tpa_state[queue] = BNX2X_TPA_STOP;
486 /* Set Toeplitz hash value in the skb using the value from the
487 * CQE (calculated by HW).
489 static inline void bnx2x_set_skb_rxhash(struct bnx2x *bp, union eth_rx_cqe *cqe,
490 struct sk_buff *skb)
492 /* Set Toeplitz hash from CQE */
493 if ((bp->dev->features & NETIF_F_RXHASH) &&
494 (cqe->fast_path_cqe.status_flags &
495 ETH_FAST_PATH_RX_CQE_RSS_HASH_FLG))
496 skb->rxhash =
497 le32_to_cpu(cqe->fast_path_cqe.rss_hash_result);
500 int bnx2x_rx_int(struct bnx2x_fastpath *fp, int budget)
502 struct bnx2x *bp = fp->bp;
503 u16 bd_cons, bd_prod, bd_prod_fw, comp_ring_cons;
504 u16 hw_comp_cons, sw_comp_cons, sw_comp_prod;
505 int rx_pkt = 0;
507 #ifdef BNX2X_STOP_ON_ERROR
508 if (unlikely(bp->panic))
509 return 0;
510 #endif
512 /* CQ "next element" is of the size of the regular element,
513 that's why it's ok here */
514 hw_comp_cons = le16_to_cpu(*fp->rx_cons_sb);
515 if ((hw_comp_cons & MAX_RCQ_DESC_CNT) == MAX_RCQ_DESC_CNT)
516 hw_comp_cons++;
518 bd_cons = fp->rx_bd_cons;
519 bd_prod = fp->rx_bd_prod;
520 bd_prod_fw = bd_prod;
521 sw_comp_cons = fp->rx_comp_cons;
522 sw_comp_prod = fp->rx_comp_prod;
524 /* Memory barrier necessary as speculative reads of the rx
525 * buffer can be ahead of the index in the status block
527 rmb();
529 DP(NETIF_MSG_RX_STATUS,
530 "queue[%d]: hw_comp_cons %u sw_comp_cons %u\n",
531 fp->index, hw_comp_cons, sw_comp_cons);
533 while (sw_comp_cons != hw_comp_cons) {
534 struct sw_rx_bd *rx_buf = NULL;
535 struct sk_buff *skb;
536 union eth_rx_cqe *cqe;
537 u8 cqe_fp_flags;
538 u16 len, pad;
540 comp_ring_cons = RCQ_BD(sw_comp_cons);
541 bd_prod = RX_BD(bd_prod);
542 bd_cons = RX_BD(bd_cons);
544 /* Prefetch the page containing the BD descriptor
545 at producer's index. It will be needed when new skb is
546 allocated */
547 prefetch((void *)(PAGE_ALIGN((unsigned long)
548 (&fp->rx_desc_ring[bd_prod])) -
549 PAGE_SIZE + 1));
551 cqe = &fp->rx_comp_ring[comp_ring_cons];
552 cqe_fp_flags = cqe->fast_path_cqe.type_error_flags;
554 DP(NETIF_MSG_RX_STATUS, "CQE type %x err %x status %x"
555 " queue %x vlan %x len %u\n", CQE_TYPE(cqe_fp_flags),
556 cqe_fp_flags, cqe->fast_path_cqe.status_flags,
557 le32_to_cpu(cqe->fast_path_cqe.rss_hash_result),
558 le16_to_cpu(cqe->fast_path_cqe.vlan_tag),
559 le16_to_cpu(cqe->fast_path_cqe.pkt_len));
561 /* is this a slowpath msg? */
562 if (unlikely(CQE_TYPE(cqe_fp_flags))) {
563 bnx2x_sp_event(fp, cqe);
564 goto next_cqe;
566 /* this is an rx packet */
567 } else {
568 rx_buf = &fp->rx_buf_ring[bd_cons];
569 skb = rx_buf->skb;
570 prefetch(skb);
571 len = le16_to_cpu(cqe->fast_path_cqe.pkt_len);
572 pad = cqe->fast_path_cqe.placement_offset;
574 /* - If CQE is marked both TPA_START and TPA_END it is
575 * a non-TPA CQE.
576 * - FP CQE will always have either TPA_START or/and
577 * TPA_STOP flags set.
579 if ((!fp->disable_tpa) &&
580 (TPA_TYPE(cqe_fp_flags) !=
581 (TPA_TYPE_START | TPA_TYPE_END))) {
582 u16 queue = cqe->fast_path_cqe.queue_index;
584 if (TPA_TYPE(cqe_fp_flags) == TPA_TYPE_START) {
585 DP(NETIF_MSG_RX_STATUS,
586 "calling tpa_start on queue %d\n",
587 queue);
589 bnx2x_tpa_start(fp, queue, skb,
590 bd_cons, bd_prod);
592 /* Set Toeplitz hash for an LRO skb */
593 bnx2x_set_skb_rxhash(bp, cqe, skb);
595 goto next_rx;
596 } else { /* TPA_STOP */
597 DP(NETIF_MSG_RX_STATUS,
598 "calling tpa_stop on queue %d\n",
599 queue);
601 if (!BNX2X_RX_SUM_FIX(cqe))
602 BNX2X_ERR("STOP on none TCP "
603 "data\n");
605 /* This is a size of the linear data
606 on this skb */
607 len = le16_to_cpu(cqe->fast_path_cqe.
608 len_on_bd);
609 bnx2x_tpa_stop(bp, fp, queue, pad,
610 len, cqe, comp_ring_cons);
611 #ifdef BNX2X_STOP_ON_ERROR
612 if (bp->panic)
613 return 0;
614 #endif
616 bnx2x_update_sge_prod(fp,
617 &cqe->fast_path_cqe);
618 goto next_cqe;
622 dma_sync_single_for_device(&bp->pdev->dev,
623 dma_unmap_addr(rx_buf, mapping),
624 pad + RX_COPY_THRESH,
625 DMA_FROM_DEVICE);
626 prefetch(((char *)(skb)) + L1_CACHE_BYTES);
628 /* is this an error packet? */
629 if (unlikely(cqe_fp_flags & ETH_RX_ERROR_FALGS)) {
630 DP(NETIF_MSG_RX_ERR,
631 "ERROR flags %x rx packet %u\n",
632 cqe_fp_flags, sw_comp_cons);
633 fp->eth_q_stats.rx_err_discard_pkt++;
634 goto reuse_rx;
637 /* Since we don't have a jumbo ring
638 * copy small packets if mtu > 1500
640 if ((bp->dev->mtu > ETH_MAX_PACKET_SIZE) &&
641 (len <= RX_COPY_THRESH)) {
642 struct sk_buff *new_skb;
644 new_skb = netdev_alloc_skb(bp->dev,
645 len + pad);
646 if (new_skb == NULL) {
647 DP(NETIF_MSG_RX_ERR,
648 "ERROR packet dropped "
649 "because of alloc failure\n");
650 fp->eth_q_stats.rx_skb_alloc_failed++;
651 goto reuse_rx;
654 /* aligned copy */
655 skb_copy_from_linear_data_offset(skb, pad,
656 new_skb->data + pad, len);
657 skb_reserve(new_skb, pad);
658 skb_put(new_skb, len);
660 bnx2x_reuse_rx_skb(fp, bd_cons, bd_prod);
662 skb = new_skb;
664 } else
665 if (likely(bnx2x_alloc_rx_skb(bp, fp, bd_prod) == 0)) {
666 dma_unmap_single(&bp->pdev->dev,
667 dma_unmap_addr(rx_buf, mapping),
668 fp->rx_buf_size,
669 DMA_FROM_DEVICE);
670 skb_reserve(skb, pad);
671 skb_put(skb, len);
673 } else {
674 DP(NETIF_MSG_RX_ERR,
675 "ERROR packet dropped because "
676 "of alloc failure\n");
677 fp->eth_q_stats.rx_skb_alloc_failed++;
678 reuse_rx:
679 bnx2x_reuse_rx_skb(fp, bd_cons, bd_prod);
680 goto next_rx;
683 skb->protocol = eth_type_trans(skb, bp->dev);
685 /* Set Toeplitz hash for a none-LRO skb */
686 bnx2x_set_skb_rxhash(bp, cqe, skb);
688 skb_checksum_none_assert(skb);
690 if (bp->dev->features & NETIF_F_RXCSUM) {
691 if (likely(BNX2X_RX_CSUM_OK(cqe)))
692 skb->ip_summed = CHECKSUM_UNNECESSARY;
693 else
694 fp->eth_q_stats.hw_csum_err++;
698 skb_record_rx_queue(skb, fp->index);
700 if (le16_to_cpu(cqe->fast_path_cqe.pars_flags.flags) &
701 PARSING_FLAGS_VLAN)
702 __vlan_hwaccel_put_tag(skb,
703 le16_to_cpu(cqe->fast_path_cqe.vlan_tag));
704 napi_gro_receive(&fp->napi, skb);
707 next_rx:
708 rx_buf->skb = NULL;
710 bd_cons = NEXT_RX_IDX(bd_cons);
711 bd_prod = NEXT_RX_IDX(bd_prod);
712 bd_prod_fw = NEXT_RX_IDX(bd_prod_fw);
713 rx_pkt++;
714 next_cqe:
715 sw_comp_prod = NEXT_RCQ_IDX(sw_comp_prod);
716 sw_comp_cons = NEXT_RCQ_IDX(sw_comp_cons);
718 if (rx_pkt == budget)
719 break;
720 } /* while */
722 fp->rx_bd_cons = bd_cons;
723 fp->rx_bd_prod = bd_prod_fw;
724 fp->rx_comp_cons = sw_comp_cons;
725 fp->rx_comp_prod = sw_comp_prod;
727 /* Update producers */
728 bnx2x_update_rx_prod(bp, fp, bd_prod_fw, sw_comp_prod,
729 fp->rx_sge_prod);
731 fp->rx_pkt += rx_pkt;
732 fp->rx_calls++;
734 return rx_pkt;
737 static irqreturn_t bnx2x_msix_fp_int(int irq, void *fp_cookie)
739 struct bnx2x_fastpath *fp = fp_cookie;
740 struct bnx2x *bp = fp->bp;
742 DP(BNX2X_MSG_FP, "got an MSI-X interrupt on IDX:SB "
743 "[fp %d fw_sd %d igusb %d]\n",
744 fp->index, fp->fw_sb_id, fp->igu_sb_id);
745 bnx2x_ack_sb(bp, fp->igu_sb_id, USTORM_ID, 0, IGU_INT_DISABLE, 0);
747 #ifdef BNX2X_STOP_ON_ERROR
748 if (unlikely(bp->panic))
749 return IRQ_HANDLED;
750 #endif
752 /* Handle Rx and Tx according to MSI-X vector */
753 prefetch(fp->rx_cons_sb);
754 prefetch(fp->tx_cons_sb);
755 prefetch(&fp->sb_running_index[SM_RX_ID]);
756 napi_schedule(&bnx2x_fp(bp, fp->index, napi));
758 return IRQ_HANDLED;
761 /* HW Lock for shared dual port PHYs */
762 void bnx2x_acquire_phy_lock(struct bnx2x *bp)
764 mutex_lock(&bp->port.phy_mutex);
766 if (bp->port.need_hw_lock)
767 bnx2x_acquire_hw_lock(bp, HW_LOCK_RESOURCE_MDIO);
770 void bnx2x_release_phy_lock(struct bnx2x *bp)
772 if (bp->port.need_hw_lock)
773 bnx2x_release_hw_lock(bp, HW_LOCK_RESOURCE_MDIO);
775 mutex_unlock(&bp->port.phy_mutex);
778 /* calculates MF speed according to current linespeed and MF configuration */
779 u16 bnx2x_get_mf_speed(struct bnx2x *bp)
781 u16 line_speed = bp->link_vars.line_speed;
782 if (IS_MF(bp)) {
783 u16 maxCfg = bnx2x_extract_max_cfg(bp,
784 bp->mf_config[BP_VN(bp)]);
786 /* Calculate the current MAX line speed limit for the MF
787 * devices
789 if (IS_MF_SI(bp))
790 line_speed = (line_speed * maxCfg) / 100;
791 else { /* SD mode */
792 u16 vn_max_rate = maxCfg * 100;
794 if (vn_max_rate < line_speed)
795 line_speed = vn_max_rate;
799 return line_speed;
803 * bnx2x_fill_report_data - fill link report data to report
805 * @bp: driver handle
806 * @data: link state to update
808 * It uses a none-atomic bit operations because is called under the mutex.
810 static inline void bnx2x_fill_report_data(struct bnx2x *bp,
811 struct bnx2x_link_report_data *data)
813 u16 line_speed = bnx2x_get_mf_speed(bp);
815 memset(data, 0, sizeof(*data));
817 /* Fill the report data: efective line speed */
818 data->line_speed = line_speed;
820 /* Link is down */
821 if (!bp->link_vars.link_up || (bp->flags & MF_FUNC_DIS))
822 __set_bit(BNX2X_LINK_REPORT_LINK_DOWN,
823 &data->link_report_flags);
825 /* Full DUPLEX */
826 if (bp->link_vars.duplex == DUPLEX_FULL)
827 __set_bit(BNX2X_LINK_REPORT_FD, &data->link_report_flags);
829 /* Rx Flow Control is ON */
830 if (bp->link_vars.flow_ctrl & BNX2X_FLOW_CTRL_RX)
831 __set_bit(BNX2X_LINK_REPORT_RX_FC_ON, &data->link_report_flags);
833 /* Tx Flow Control is ON */
834 if (bp->link_vars.flow_ctrl & BNX2X_FLOW_CTRL_TX)
835 __set_bit(BNX2X_LINK_REPORT_TX_FC_ON, &data->link_report_flags);
839 * bnx2x_link_report - report link status to OS.
841 * @bp: driver handle
843 * Calls the __bnx2x_link_report() under the same locking scheme
844 * as a link/PHY state managing code to ensure a consistent link
845 * reporting.
848 void bnx2x_link_report(struct bnx2x *bp)
850 bnx2x_acquire_phy_lock(bp);
851 __bnx2x_link_report(bp);
852 bnx2x_release_phy_lock(bp);
856 * __bnx2x_link_report - report link status to OS.
858 * @bp: driver handle
860 * None atomic inmlementation.
861 * Should be called under the phy_lock.
863 void __bnx2x_link_report(struct bnx2x *bp)
865 struct bnx2x_link_report_data cur_data;
867 /* reread mf_cfg */
868 if (!CHIP_IS_E1(bp))
869 bnx2x_read_mf_cfg(bp);
871 /* Read the current link report info */
872 bnx2x_fill_report_data(bp, &cur_data);
874 /* Don't report link down or exactly the same link status twice */
875 if (!memcmp(&cur_data, &bp->last_reported_link, sizeof(cur_data)) ||
876 (test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
877 &bp->last_reported_link.link_report_flags) &&
878 test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
879 &cur_data.link_report_flags)))
880 return;
882 bp->link_cnt++;
884 /* We are going to report a new link parameters now -
885 * remember the current data for the next time.
887 memcpy(&bp->last_reported_link, &cur_data, sizeof(cur_data));
889 if (test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
890 &cur_data.link_report_flags)) {
891 netif_carrier_off(bp->dev);
892 netdev_err(bp->dev, "NIC Link is Down\n");
893 return;
894 } else {
895 netif_carrier_on(bp->dev);
896 netdev_info(bp->dev, "NIC Link is Up, ");
897 pr_cont("%d Mbps ", cur_data.line_speed);
899 if (test_and_clear_bit(BNX2X_LINK_REPORT_FD,
900 &cur_data.link_report_flags))
901 pr_cont("full duplex");
902 else
903 pr_cont("half duplex");
905 /* Handle the FC at the end so that only these flags would be
906 * possibly set. This way we may easily check if there is no FC
907 * enabled.
909 if (cur_data.link_report_flags) {
910 if (test_bit(BNX2X_LINK_REPORT_RX_FC_ON,
911 &cur_data.link_report_flags)) {
912 pr_cont(", receive ");
913 if (test_bit(BNX2X_LINK_REPORT_TX_FC_ON,
914 &cur_data.link_report_flags))
915 pr_cont("& transmit ");
916 } else {
917 pr_cont(", transmit ");
919 pr_cont("flow control ON");
921 pr_cont("\n");
925 void bnx2x_init_rx_rings(struct bnx2x *bp)
927 int func = BP_FUNC(bp);
928 int max_agg_queues = CHIP_IS_E1(bp) ? ETH_MAX_AGGREGATION_QUEUES_E1 :
929 ETH_MAX_AGGREGATION_QUEUES_E1H;
930 u16 ring_prod;
931 int i, j;
933 /* Allocate TPA resources */
934 for_each_rx_queue(bp, j) {
935 struct bnx2x_fastpath *fp = &bp->fp[j];
937 DP(NETIF_MSG_IFUP,
938 "mtu %d rx_buf_size %d\n", bp->dev->mtu, fp->rx_buf_size);
940 if (!fp->disable_tpa) {
941 /* Fill the per-aggregation pool */
942 for (i = 0; i < max_agg_queues; i++) {
943 fp->tpa_pool[i].skb =
944 netdev_alloc_skb(bp->dev, fp->rx_buf_size);
945 if (!fp->tpa_pool[i].skb) {
946 BNX2X_ERR("Failed to allocate TPA "
947 "skb pool for queue[%d] - "
948 "disabling TPA on this "
949 "queue!\n", j);
950 bnx2x_free_tpa_pool(bp, fp, i);
951 fp->disable_tpa = 1;
952 break;
954 dma_unmap_addr_set((struct sw_rx_bd *)
955 &bp->fp->tpa_pool[i],
956 mapping, 0);
957 fp->tpa_state[i] = BNX2X_TPA_STOP;
960 /* "next page" elements initialization */
961 bnx2x_set_next_page_sgl(fp);
963 /* set SGEs bit mask */
964 bnx2x_init_sge_ring_bit_mask(fp);
966 /* Allocate SGEs and initialize the ring elements */
967 for (i = 0, ring_prod = 0;
968 i < MAX_RX_SGE_CNT*NUM_RX_SGE_PAGES; i++) {
970 if (bnx2x_alloc_rx_sge(bp, fp, ring_prod) < 0) {
971 BNX2X_ERR("was only able to allocate "
972 "%d rx sges\n", i);
973 BNX2X_ERR("disabling TPA for"
974 " queue[%d]\n", j);
975 /* Cleanup already allocated elements */
976 bnx2x_free_rx_sge_range(bp,
977 fp, ring_prod);
978 bnx2x_free_tpa_pool(bp,
979 fp, max_agg_queues);
980 fp->disable_tpa = 1;
981 ring_prod = 0;
982 break;
984 ring_prod = NEXT_SGE_IDX(ring_prod);
987 fp->rx_sge_prod = ring_prod;
991 for_each_rx_queue(bp, j) {
992 struct bnx2x_fastpath *fp = &bp->fp[j];
994 fp->rx_bd_cons = 0;
996 /* Activate BD ring */
997 /* Warning!
998 * this will generate an interrupt (to the TSTORM)
999 * must only be done after chip is initialized
1001 bnx2x_update_rx_prod(bp, fp, fp->rx_bd_prod, fp->rx_comp_prod,
1002 fp->rx_sge_prod);
1004 if (j != 0)
1005 continue;
1007 if (!CHIP_IS_E2(bp)) {
1008 REG_WR(bp, BAR_USTRORM_INTMEM +
1009 USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func),
1010 U64_LO(fp->rx_comp_mapping));
1011 REG_WR(bp, BAR_USTRORM_INTMEM +
1012 USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func) + 4,
1013 U64_HI(fp->rx_comp_mapping));
1018 static void bnx2x_free_tx_skbs(struct bnx2x *bp)
1020 int i;
1022 for_each_tx_queue(bp, i) {
1023 struct bnx2x_fastpath *fp = &bp->fp[i];
1025 u16 bd_cons = fp->tx_bd_cons;
1026 u16 sw_prod = fp->tx_pkt_prod;
1027 u16 sw_cons = fp->tx_pkt_cons;
1029 while (sw_cons != sw_prod) {
1030 bd_cons = bnx2x_free_tx_pkt(bp, fp, TX_BD(sw_cons));
1031 sw_cons++;
1036 static void bnx2x_free_rx_bds(struct bnx2x_fastpath *fp)
1038 struct bnx2x *bp = fp->bp;
1039 int i;
1041 /* ring wasn't allocated */
1042 if (fp->rx_buf_ring == NULL)
1043 return;
1045 for (i = 0; i < NUM_RX_BD; i++) {
1046 struct sw_rx_bd *rx_buf = &fp->rx_buf_ring[i];
1047 struct sk_buff *skb = rx_buf->skb;
1049 if (skb == NULL)
1050 continue;
1052 dma_unmap_single(&bp->pdev->dev,
1053 dma_unmap_addr(rx_buf, mapping),
1054 fp->rx_buf_size, DMA_FROM_DEVICE);
1056 rx_buf->skb = NULL;
1057 dev_kfree_skb(skb);
1061 static void bnx2x_free_rx_skbs(struct bnx2x *bp)
1063 int j;
1065 for_each_rx_queue(bp, j) {
1066 struct bnx2x_fastpath *fp = &bp->fp[j];
1068 bnx2x_free_rx_bds(fp);
1070 if (!fp->disable_tpa)
1071 bnx2x_free_tpa_pool(bp, fp, CHIP_IS_E1(bp) ?
1072 ETH_MAX_AGGREGATION_QUEUES_E1 :
1073 ETH_MAX_AGGREGATION_QUEUES_E1H);
1077 void bnx2x_free_skbs(struct bnx2x *bp)
1079 bnx2x_free_tx_skbs(bp);
1080 bnx2x_free_rx_skbs(bp);
1083 void bnx2x_update_max_mf_config(struct bnx2x *bp, u32 value)
1085 /* load old values */
1086 u32 mf_cfg = bp->mf_config[BP_VN(bp)];
1088 if (value != bnx2x_extract_max_cfg(bp, mf_cfg)) {
1089 /* leave all but MAX value */
1090 mf_cfg &= ~FUNC_MF_CFG_MAX_BW_MASK;
1092 /* set new MAX value */
1093 mf_cfg |= (value << FUNC_MF_CFG_MAX_BW_SHIFT)
1094 & FUNC_MF_CFG_MAX_BW_MASK;
1096 bnx2x_fw_command(bp, DRV_MSG_CODE_SET_MF_BW, mf_cfg);
1100 static void bnx2x_free_msix_irqs(struct bnx2x *bp)
1102 int i, offset = 1;
1104 free_irq(bp->msix_table[0].vector, bp->dev);
1105 DP(NETIF_MSG_IFDOWN, "released sp irq (%d)\n",
1106 bp->msix_table[0].vector);
1108 #ifdef BCM_CNIC
1109 offset++;
1110 #endif
1111 for_each_eth_queue(bp, i) {
1112 DP(NETIF_MSG_IFDOWN, "about to release fp #%d->%d irq "
1113 "state %x\n", i, bp->msix_table[i + offset].vector,
1114 bnx2x_fp(bp, i, state));
1116 free_irq(bp->msix_table[i + offset].vector, &bp->fp[i]);
1120 void bnx2x_free_irq(struct bnx2x *bp)
1122 if (bp->flags & USING_MSIX_FLAG)
1123 bnx2x_free_msix_irqs(bp);
1124 else if (bp->flags & USING_MSI_FLAG)
1125 free_irq(bp->pdev->irq, bp->dev);
1126 else
1127 free_irq(bp->pdev->irq, bp->dev);
1130 int bnx2x_enable_msix(struct bnx2x *bp)
1132 int msix_vec = 0, i, rc, req_cnt;
1134 bp->msix_table[msix_vec].entry = msix_vec;
1135 DP(NETIF_MSG_IFUP, "msix_table[0].entry = %d (slowpath)\n",
1136 bp->msix_table[0].entry);
1137 msix_vec++;
1139 #ifdef BCM_CNIC
1140 bp->msix_table[msix_vec].entry = msix_vec;
1141 DP(NETIF_MSG_IFUP, "msix_table[%d].entry = %d (CNIC)\n",
1142 bp->msix_table[msix_vec].entry, bp->msix_table[msix_vec].entry);
1143 msix_vec++;
1144 #endif
1145 for_each_eth_queue(bp, i) {
1146 bp->msix_table[msix_vec].entry = msix_vec;
1147 DP(NETIF_MSG_IFUP, "msix_table[%d].entry = %d "
1148 "(fastpath #%u)\n", msix_vec, msix_vec, i);
1149 msix_vec++;
1152 req_cnt = BNX2X_NUM_ETH_QUEUES(bp) + CNIC_CONTEXT_USE + 1;
1154 rc = pci_enable_msix(bp->pdev, &bp->msix_table[0], req_cnt);
1157 * reconfigure number of tx/rx queues according to available
1158 * MSI-X vectors
1160 if (rc >= BNX2X_MIN_MSIX_VEC_CNT) {
1161 /* how less vectors we will have? */
1162 int diff = req_cnt - rc;
1164 DP(NETIF_MSG_IFUP,
1165 "Trying to use less MSI-X vectors: %d\n", rc);
1167 rc = pci_enable_msix(bp->pdev, &bp->msix_table[0], rc);
1169 if (rc) {
1170 DP(NETIF_MSG_IFUP,
1171 "MSI-X is not attainable rc %d\n", rc);
1172 return rc;
1175 * decrease number of queues by number of unallocated entries
1177 bp->num_queues -= diff;
1179 DP(NETIF_MSG_IFUP, "New queue configuration set: %d\n",
1180 bp->num_queues);
1181 } else if (rc) {
1182 /* fall to INTx if not enough memory */
1183 if (rc == -ENOMEM)
1184 bp->flags |= DISABLE_MSI_FLAG;
1185 DP(NETIF_MSG_IFUP, "MSI-X is not attainable rc %d\n", rc);
1186 return rc;
1189 bp->flags |= USING_MSIX_FLAG;
1191 return 0;
1194 static int bnx2x_req_msix_irqs(struct bnx2x *bp)
1196 int i, rc, offset = 1;
1198 rc = request_irq(bp->msix_table[0].vector, bnx2x_msix_sp_int, 0,
1199 bp->dev->name, bp->dev);
1200 if (rc) {
1201 BNX2X_ERR("request sp irq failed\n");
1202 return -EBUSY;
1205 #ifdef BCM_CNIC
1206 offset++;
1207 #endif
1208 for_each_eth_queue(bp, i) {
1209 struct bnx2x_fastpath *fp = &bp->fp[i];
1210 snprintf(fp->name, sizeof(fp->name), "%s-fp-%d",
1211 bp->dev->name, i);
1213 rc = request_irq(bp->msix_table[offset].vector,
1214 bnx2x_msix_fp_int, 0, fp->name, fp);
1215 if (rc) {
1216 BNX2X_ERR("request fp #%d irq failed rc %d\n", i, rc);
1217 bnx2x_free_msix_irqs(bp);
1218 return -EBUSY;
1221 offset++;
1222 fp->state = BNX2X_FP_STATE_IRQ;
1225 i = BNX2X_NUM_ETH_QUEUES(bp);
1226 offset = 1 + CNIC_CONTEXT_USE;
1227 netdev_info(bp->dev, "using MSI-X IRQs: sp %d fp[%d] %d"
1228 " ... fp[%d] %d\n",
1229 bp->msix_table[0].vector,
1230 0, bp->msix_table[offset].vector,
1231 i - 1, bp->msix_table[offset + i - 1].vector);
1233 return 0;
1236 int bnx2x_enable_msi(struct bnx2x *bp)
1238 int rc;
1240 rc = pci_enable_msi(bp->pdev);
1241 if (rc) {
1242 DP(NETIF_MSG_IFUP, "MSI is not attainable\n");
1243 return -1;
1245 bp->flags |= USING_MSI_FLAG;
1247 return 0;
1250 static int bnx2x_req_irq(struct bnx2x *bp)
1252 unsigned long flags;
1253 int rc;
1255 if (bp->flags & USING_MSI_FLAG)
1256 flags = 0;
1257 else
1258 flags = IRQF_SHARED;
1260 rc = request_irq(bp->pdev->irq, bnx2x_interrupt, flags,
1261 bp->dev->name, bp->dev);
1262 if (!rc)
1263 bnx2x_fp(bp, 0, state) = BNX2X_FP_STATE_IRQ;
1265 return rc;
1268 static void bnx2x_napi_enable(struct bnx2x *bp)
1270 int i;
1272 for_each_napi_queue(bp, i)
1273 napi_enable(&bnx2x_fp(bp, i, napi));
1276 static void bnx2x_napi_disable(struct bnx2x *bp)
1278 int i;
1280 for_each_napi_queue(bp, i)
1281 napi_disable(&bnx2x_fp(bp, i, napi));
1284 void bnx2x_netif_start(struct bnx2x *bp)
1286 if (netif_running(bp->dev)) {
1287 bnx2x_napi_enable(bp);
1288 bnx2x_int_enable(bp);
1289 if (bp->state == BNX2X_STATE_OPEN)
1290 netif_tx_wake_all_queues(bp->dev);
1294 void bnx2x_netif_stop(struct bnx2x *bp, int disable_hw)
1296 bnx2x_int_disable_sync(bp, disable_hw);
1297 bnx2x_napi_disable(bp);
1298 netif_tx_disable(bp->dev);
1301 u16 bnx2x_select_queue(struct net_device *dev, struct sk_buff *skb)
1303 #ifdef BCM_CNIC
1304 struct bnx2x *bp = netdev_priv(dev);
1305 if (NO_FCOE(bp))
1306 return skb_tx_hash(dev, skb);
1307 else {
1308 struct ethhdr *hdr = (struct ethhdr *)skb->data;
1309 u16 ether_type = ntohs(hdr->h_proto);
1311 /* Skip VLAN tag if present */
1312 if (ether_type == ETH_P_8021Q) {
1313 struct vlan_ethhdr *vhdr =
1314 (struct vlan_ethhdr *)skb->data;
1316 ether_type = ntohs(vhdr->h_vlan_encapsulated_proto);
1319 /* If ethertype is FCoE or FIP - use FCoE ring */
1320 if ((ether_type == ETH_P_FCOE) || (ether_type == ETH_P_FIP))
1321 return bnx2x_fcoe(bp, index);
1323 #endif
1324 /* Select a none-FCoE queue: if FCoE is enabled, exclude FCoE L2 ring
1326 return __skb_tx_hash(dev, skb,
1327 dev->real_num_tx_queues - FCOE_CONTEXT_USE);
1330 void bnx2x_set_num_queues(struct bnx2x *bp)
1332 switch (bp->multi_mode) {
1333 case ETH_RSS_MODE_DISABLED:
1334 bp->num_queues = 1;
1335 break;
1336 case ETH_RSS_MODE_REGULAR:
1337 bp->num_queues = bnx2x_calc_num_queues(bp);
1338 break;
1340 default:
1341 bp->num_queues = 1;
1342 break;
1345 /* Add special queues */
1346 bp->num_queues += NONE_ETH_CONTEXT_USE;
1349 #ifdef BCM_CNIC
1350 static inline void bnx2x_set_fcoe_eth_macs(struct bnx2x *bp)
1352 if (!NO_FCOE(bp)) {
1353 if (!IS_MF_SD(bp))
1354 bnx2x_set_fip_eth_mac_addr(bp, 1);
1355 bnx2x_set_all_enode_macs(bp, 1);
1356 bp->flags |= FCOE_MACS_SET;
1359 #endif
1361 static void bnx2x_release_firmware(struct bnx2x *bp)
1363 kfree(bp->init_ops_offsets);
1364 kfree(bp->init_ops);
1365 kfree(bp->init_data);
1366 release_firmware(bp->firmware);
1369 static inline int bnx2x_set_real_num_queues(struct bnx2x *bp)
1371 int rc, num = bp->num_queues;
1373 #ifdef BCM_CNIC
1374 if (NO_FCOE(bp))
1375 num -= FCOE_CONTEXT_USE;
1377 #endif
1378 netif_set_real_num_tx_queues(bp->dev, num);
1379 rc = netif_set_real_num_rx_queues(bp->dev, num);
1380 return rc;
1383 static inline void bnx2x_set_rx_buf_size(struct bnx2x *bp)
1385 int i;
1387 for_each_queue(bp, i) {
1388 struct bnx2x_fastpath *fp = &bp->fp[i];
1390 /* Always use a mini-jumbo MTU for the FCoE L2 ring */
1391 if (IS_FCOE_IDX(i))
1393 * Although there are no IP frames expected to arrive to
1394 * this ring we still want to add an
1395 * IP_HEADER_ALIGNMENT_PADDING to prevent a buffer
1396 * overrun attack.
1398 fp->rx_buf_size =
1399 BNX2X_FCOE_MINI_JUMBO_MTU + ETH_OVREHEAD +
1400 BNX2X_RX_ALIGN + IP_HEADER_ALIGNMENT_PADDING;
1401 else
1402 fp->rx_buf_size =
1403 bp->dev->mtu + ETH_OVREHEAD + BNX2X_RX_ALIGN +
1404 IP_HEADER_ALIGNMENT_PADDING;
1408 /* must be called with rtnl_lock */
1409 int bnx2x_nic_load(struct bnx2x *bp, int load_mode)
1411 u32 load_code;
1412 int i, rc;
1414 /* Set init arrays */
1415 rc = bnx2x_init_firmware(bp);
1416 if (rc) {
1417 BNX2X_ERR("Error loading firmware\n");
1418 return rc;
1421 #ifdef BNX2X_STOP_ON_ERROR
1422 if (unlikely(bp->panic))
1423 return -EPERM;
1424 #endif
1426 bp->state = BNX2X_STATE_OPENING_WAIT4_LOAD;
1428 /* Set the initial link reported state to link down */
1429 bnx2x_acquire_phy_lock(bp);
1430 memset(&bp->last_reported_link, 0, sizeof(bp->last_reported_link));
1431 __set_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1432 &bp->last_reported_link.link_report_flags);
1433 bnx2x_release_phy_lock(bp);
1435 /* must be called before memory allocation and HW init */
1436 bnx2x_ilt_set_info(bp);
1438 /* zero fastpath structures preserving invariants like napi which are
1439 * allocated only once
1441 for_each_queue(bp, i)
1442 bnx2x_bz_fp(bp, i);
1444 /* Set the receive queues buffer size */
1445 bnx2x_set_rx_buf_size(bp);
1447 for_each_queue(bp, i)
1448 bnx2x_fp(bp, i, disable_tpa) =
1449 ((bp->flags & TPA_ENABLE_FLAG) == 0);
1451 #ifdef BCM_CNIC
1452 /* We don't want TPA on FCoE L2 ring */
1453 bnx2x_fcoe(bp, disable_tpa) = 1;
1454 #endif
1456 if (bnx2x_alloc_mem(bp))
1457 return -ENOMEM;
1459 /* As long as bnx2x_alloc_mem() may possibly update
1460 * bp->num_queues, bnx2x_set_real_num_queues() should always
1461 * come after it.
1463 rc = bnx2x_set_real_num_queues(bp);
1464 if (rc) {
1465 BNX2X_ERR("Unable to set real_num_queues\n");
1466 goto load_error0;
1469 bnx2x_napi_enable(bp);
1471 /* Send LOAD_REQUEST command to MCP
1472 Returns the type of LOAD command:
1473 if it is the first port to be initialized
1474 common blocks should be initialized, otherwise - not
1476 if (!BP_NOMCP(bp)) {
1477 load_code = bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_REQ, 0);
1478 if (!load_code) {
1479 BNX2X_ERR("MCP response failure, aborting\n");
1480 rc = -EBUSY;
1481 goto load_error1;
1483 if (load_code == FW_MSG_CODE_DRV_LOAD_REFUSED) {
1484 rc = -EBUSY; /* other port in diagnostic mode */
1485 goto load_error1;
1488 } else {
1489 int path = BP_PATH(bp);
1490 int port = BP_PORT(bp);
1492 DP(NETIF_MSG_IFUP, "NO MCP - load counts[%d] %d, %d, %d\n",
1493 path, load_count[path][0], load_count[path][1],
1494 load_count[path][2]);
1495 load_count[path][0]++;
1496 load_count[path][1 + port]++;
1497 DP(NETIF_MSG_IFUP, "NO MCP - new load counts[%d] %d, %d, %d\n",
1498 path, load_count[path][0], load_count[path][1],
1499 load_count[path][2]);
1500 if (load_count[path][0] == 1)
1501 load_code = FW_MSG_CODE_DRV_LOAD_COMMON;
1502 else if (load_count[path][1 + port] == 1)
1503 load_code = FW_MSG_CODE_DRV_LOAD_PORT;
1504 else
1505 load_code = FW_MSG_CODE_DRV_LOAD_FUNCTION;
1508 if ((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
1509 (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP) ||
1510 (load_code == FW_MSG_CODE_DRV_LOAD_PORT))
1511 bp->port.pmf = 1;
1512 else
1513 bp->port.pmf = 0;
1514 DP(NETIF_MSG_LINK, "pmf %d\n", bp->port.pmf);
1516 /* Initialize HW */
1517 rc = bnx2x_init_hw(bp, load_code);
1518 if (rc) {
1519 BNX2X_ERR("HW init failed, aborting\n");
1520 bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
1521 goto load_error2;
1524 /* Connect to IRQs */
1525 rc = bnx2x_setup_irqs(bp);
1526 if (rc) {
1527 bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
1528 goto load_error2;
1531 /* Setup NIC internals and enable interrupts */
1532 bnx2x_nic_init(bp, load_code);
1534 if (((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
1535 (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP)) &&
1536 (bp->common.shmem2_base))
1537 SHMEM2_WR(bp, dcc_support,
1538 (SHMEM_DCC_SUPPORT_DISABLE_ENABLE_PF_TLV |
1539 SHMEM_DCC_SUPPORT_BANDWIDTH_ALLOCATION_TLV));
1541 /* Send LOAD_DONE command to MCP */
1542 if (!BP_NOMCP(bp)) {
1543 load_code = bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
1544 if (!load_code) {
1545 BNX2X_ERR("MCP response failure, aborting\n");
1546 rc = -EBUSY;
1547 goto load_error3;
1551 bnx2x_dcbx_init(bp);
1553 bp->state = BNX2X_STATE_OPENING_WAIT4_PORT;
1555 rc = bnx2x_func_start(bp);
1556 if (rc) {
1557 BNX2X_ERR("Function start failed!\n");
1558 #ifndef BNX2X_STOP_ON_ERROR
1559 goto load_error3;
1560 #else
1561 bp->panic = 1;
1562 return -EBUSY;
1563 #endif
1566 rc = bnx2x_setup_client(bp, &bp->fp[0], 1 /* Leading */);
1567 if (rc) {
1568 BNX2X_ERR("Setup leading failed!\n");
1569 #ifndef BNX2X_STOP_ON_ERROR
1570 goto load_error3;
1571 #else
1572 bp->panic = 1;
1573 return -EBUSY;
1574 #endif
1577 if (!CHIP_IS_E1(bp) &&
1578 (bp->mf_config[BP_VN(bp)] & FUNC_MF_CFG_FUNC_DISABLED)) {
1579 DP(NETIF_MSG_IFUP, "mf_cfg function disabled\n");
1580 bp->flags |= MF_FUNC_DIS;
1583 #ifdef BCM_CNIC
1584 /* Enable Timer scan */
1585 REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + BP_PORT(bp)*4, 1);
1586 #endif
1588 for_each_nondefault_queue(bp, i) {
1589 rc = bnx2x_setup_client(bp, &bp->fp[i], 0);
1590 if (rc)
1591 #ifdef BCM_CNIC
1592 goto load_error4;
1593 #else
1594 goto load_error3;
1595 #endif
1598 /* Now when Clients are configured we are ready to work */
1599 bp->state = BNX2X_STATE_OPEN;
1601 #ifdef BCM_CNIC
1602 bnx2x_set_fcoe_eth_macs(bp);
1603 #endif
1605 bnx2x_set_eth_mac(bp, 1);
1607 /* Clear MC configuration */
1608 if (CHIP_IS_E1(bp))
1609 bnx2x_invalidate_e1_mc_list(bp);
1610 else
1611 bnx2x_invalidate_e1h_mc_list(bp);
1613 /* Clear UC lists configuration */
1614 bnx2x_invalidate_uc_list(bp);
1616 if (bp->pending_max) {
1617 bnx2x_update_max_mf_config(bp, bp->pending_max);
1618 bp->pending_max = 0;
1621 if (bp->port.pmf)
1622 bnx2x_initial_phy_init(bp, load_mode);
1624 /* Initialize Rx filtering */
1625 bnx2x_set_rx_mode(bp->dev);
1627 /* Start fast path */
1628 switch (load_mode) {
1629 case LOAD_NORMAL:
1630 /* Tx queue should be only reenabled */
1631 netif_tx_wake_all_queues(bp->dev);
1632 /* Initialize the receive filter. */
1633 break;
1635 case LOAD_OPEN:
1636 netif_tx_start_all_queues(bp->dev);
1637 smp_mb__after_clear_bit();
1638 break;
1640 case LOAD_DIAG:
1641 bp->state = BNX2X_STATE_DIAG;
1642 break;
1644 default:
1645 break;
1648 if (!bp->port.pmf)
1649 bnx2x__link_status_update(bp);
1651 /* start the timer */
1652 mod_timer(&bp->timer, jiffies + bp->current_interval);
1654 #ifdef BCM_CNIC
1655 bnx2x_setup_cnic_irq_info(bp);
1656 if (bp->state == BNX2X_STATE_OPEN)
1657 bnx2x_cnic_notify(bp, CNIC_CTL_START_CMD);
1658 #endif
1659 bnx2x_inc_load_cnt(bp);
1661 bnx2x_release_firmware(bp);
1663 return 0;
1665 #ifdef BCM_CNIC
1666 load_error4:
1667 /* Disable Timer scan */
1668 REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + BP_PORT(bp)*4, 0);
1669 #endif
1670 load_error3:
1671 bnx2x_int_disable_sync(bp, 1);
1673 /* Free SKBs, SGEs, TPA pool and driver internals */
1674 bnx2x_free_skbs(bp);
1675 for_each_rx_queue(bp, i)
1676 bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
1678 /* Release IRQs */
1679 bnx2x_free_irq(bp);
1680 load_error2:
1681 if (!BP_NOMCP(bp)) {
1682 bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_REQ_WOL_MCP, 0);
1683 bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_DONE, 0);
1686 bp->port.pmf = 0;
1687 load_error1:
1688 bnx2x_napi_disable(bp);
1689 load_error0:
1690 bnx2x_free_mem(bp);
1692 bnx2x_release_firmware(bp);
1694 return rc;
1697 /* must be called with rtnl_lock */
1698 int bnx2x_nic_unload(struct bnx2x *bp, int unload_mode)
1700 int i;
1702 if (bp->state == BNX2X_STATE_CLOSED) {
1703 /* Interface has been removed - nothing to recover */
1704 bp->recovery_state = BNX2X_RECOVERY_DONE;
1705 bp->is_leader = 0;
1706 bnx2x_release_hw_lock(bp, HW_LOCK_RESOURCE_RESERVED_08);
1707 smp_wmb();
1709 return -EINVAL;
1712 #ifdef BCM_CNIC
1713 bnx2x_cnic_notify(bp, CNIC_CTL_STOP_CMD);
1714 #endif
1715 bp->state = BNX2X_STATE_CLOSING_WAIT4_HALT;
1717 /* Set "drop all" */
1718 bp->rx_mode = BNX2X_RX_MODE_NONE;
1719 bnx2x_set_storm_rx_mode(bp);
1721 /* Stop Tx */
1722 bnx2x_tx_disable(bp);
1724 del_timer_sync(&bp->timer);
1726 SHMEM_WR(bp, func_mb[BP_FW_MB_IDX(bp)].drv_pulse_mb,
1727 (DRV_PULSE_ALWAYS_ALIVE | bp->fw_drv_pulse_wr_seq));
1729 bnx2x_stats_handle(bp, STATS_EVENT_STOP);
1731 /* Cleanup the chip if needed */
1732 if (unload_mode != UNLOAD_RECOVERY)
1733 bnx2x_chip_cleanup(bp, unload_mode);
1734 else {
1735 /* Disable HW interrupts, NAPI and Tx */
1736 bnx2x_netif_stop(bp, 1);
1738 /* Release IRQs */
1739 bnx2x_free_irq(bp);
1742 bp->port.pmf = 0;
1744 /* Free SKBs, SGEs, TPA pool and driver internals */
1745 bnx2x_free_skbs(bp);
1746 for_each_rx_queue(bp, i)
1747 bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
1749 bnx2x_free_mem(bp);
1751 bp->state = BNX2X_STATE_CLOSED;
1753 /* The last driver must disable a "close the gate" if there is no
1754 * parity attention or "process kill" pending.
1756 if ((!bnx2x_dec_load_cnt(bp)) && (!bnx2x_chk_parity_attn(bp)) &&
1757 bnx2x_reset_is_done(bp))
1758 bnx2x_disable_close_the_gate(bp);
1760 /* Reset MCP mail box sequence if there is on going recovery */
1761 if (unload_mode == UNLOAD_RECOVERY)
1762 bp->fw_seq = 0;
1764 return 0;
1767 int bnx2x_set_power_state(struct bnx2x *bp, pci_power_t state)
1769 u16 pmcsr;
1771 /* If there is no power capability, silently succeed */
1772 if (!bp->pm_cap) {
1773 DP(NETIF_MSG_HW, "No power capability. Breaking.\n");
1774 return 0;
1777 pci_read_config_word(bp->pdev, bp->pm_cap + PCI_PM_CTRL, &pmcsr);
1779 switch (state) {
1780 case PCI_D0:
1781 pci_write_config_word(bp->pdev, bp->pm_cap + PCI_PM_CTRL,
1782 ((pmcsr & ~PCI_PM_CTRL_STATE_MASK) |
1783 PCI_PM_CTRL_PME_STATUS));
1785 if (pmcsr & PCI_PM_CTRL_STATE_MASK)
1786 /* delay required during transition out of D3hot */
1787 msleep(20);
1788 break;
1790 case PCI_D3hot:
1791 /* If there are other clients above don't
1792 shut down the power */
1793 if (atomic_read(&bp->pdev->enable_cnt) != 1)
1794 return 0;
1795 /* Don't shut down the power for emulation and FPGA */
1796 if (CHIP_REV_IS_SLOW(bp))
1797 return 0;
1799 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
1800 pmcsr |= 3;
1802 if (bp->wol)
1803 pmcsr |= PCI_PM_CTRL_PME_ENABLE;
1805 pci_write_config_word(bp->pdev, bp->pm_cap + PCI_PM_CTRL,
1806 pmcsr);
1808 /* No more memory access after this point until
1809 * device is brought back to D0.
1811 break;
1813 default:
1814 return -EINVAL;
1816 return 0;
1820 * net_device service functions
1822 int bnx2x_poll(struct napi_struct *napi, int budget)
1824 int work_done = 0;
1825 struct bnx2x_fastpath *fp = container_of(napi, struct bnx2x_fastpath,
1826 napi);
1827 struct bnx2x *bp = fp->bp;
1829 while (1) {
1830 #ifdef BNX2X_STOP_ON_ERROR
1831 if (unlikely(bp->panic)) {
1832 napi_complete(napi);
1833 return 0;
1835 #endif
1837 if (bnx2x_has_tx_work(fp))
1838 bnx2x_tx_int(fp);
1840 if (bnx2x_has_rx_work(fp)) {
1841 work_done += bnx2x_rx_int(fp, budget - work_done);
1843 /* must not complete if we consumed full budget */
1844 if (work_done >= budget)
1845 break;
1848 /* Fall out from the NAPI loop if needed */
1849 if (!(bnx2x_has_rx_work(fp) || bnx2x_has_tx_work(fp))) {
1850 #ifdef BCM_CNIC
1851 /* No need to update SB for FCoE L2 ring as long as
1852 * it's connected to the default SB and the SB
1853 * has been updated when NAPI was scheduled.
1855 if (IS_FCOE_FP(fp)) {
1856 napi_complete(napi);
1857 break;
1859 #endif
1861 bnx2x_update_fpsb_idx(fp);
1862 /* bnx2x_has_rx_work() reads the status block,
1863 * thus we need to ensure that status block indices
1864 * have been actually read (bnx2x_update_fpsb_idx)
1865 * prior to this check (bnx2x_has_rx_work) so that
1866 * we won't write the "newer" value of the status block
1867 * to IGU (if there was a DMA right after
1868 * bnx2x_has_rx_work and if there is no rmb, the memory
1869 * reading (bnx2x_update_fpsb_idx) may be postponed
1870 * to right before bnx2x_ack_sb). In this case there
1871 * will never be another interrupt until there is
1872 * another update of the status block, while there
1873 * is still unhandled work.
1875 rmb();
1877 if (!(bnx2x_has_rx_work(fp) || bnx2x_has_tx_work(fp))) {
1878 napi_complete(napi);
1879 /* Re-enable interrupts */
1880 DP(NETIF_MSG_HW,
1881 "Update index to %d\n", fp->fp_hc_idx);
1882 bnx2x_ack_sb(bp, fp->igu_sb_id, USTORM_ID,
1883 le16_to_cpu(fp->fp_hc_idx),
1884 IGU_INT_ENABLE, 1);
1885 break;
1890 return work_done;
1893 /* we split the first BD into headers and data BDs
1894 * to ease the pain of our fellow microcode engineers
1895 * we use one mapping for both BDs
1896 * So far this has only been observed to happen
1897 * in Other Operating Systems(TM)
1899 static noinline u16 bnx2x_tx_split(struct bnx2x *bp,
1900 struct bnx2x_fastpath *fp,
1901 struct sw_tx_bd *tx_buf,
1902 struct eth_tx_start_bd **tx_bd, u16 hlen,
1903 u16 bd_prod, int nbd)
1905 struct eth_tx_start_bd *h_tx_bd = *tx_bd;
1906 struct eth_tx_bd *d_tx_bd;
1907 dma_addr_t mapping;
1908 int old_len = le16_to_cpu(h_tx_bd->nbytes);
1910 /* first fix first BD */
1911 h_tx_bd->nbd = cpu_to_le16(nbd);
1912 h_tx_bd->nbytes = cpu_to_le16(hlen);
1914 DP(NETIF_MSG_TX_QUEUED, "TSO split header size is %d "
1915 "(%x:%x) nbd %d\n", h_tx_bd->nbytes, h_tx_bd->addr_hi,
1916 h_tx_bd->addr_lo, h_tx_bd->nbd);
1918 /* now get a new data BD
1919 * (after the pbd) and fill it */
1920 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
1921 d_tx_bd = &fp->tx_desc_ring[bd_prod].reg_bd;
1923 mapping = HILO_U64(le32_to_cpu(h_tx_bd->addr_hi),
1924 le32_to_cpu(h_tx_bd->addr_lo)) + hlen;
1926 d_tx_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
1927 d_tx_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
1928 d_tx_bd->nbytes = cpu_to_le16(old_len - hlen);
1930 /* this marks the BD as one that has no individual mapping */
1931 tx_buf->flags |= BNX2X_TSO_SPLIT_BD;
1933 DP(NETIF_MSG_TX_QUEUED,
1934 "TSO split data size is %d (%x:%x)\n",
1935 d_tx_bd->nbytes, d_tx_bd->addr_hi, d_tx_bd->addr_lo);
1937 /* update tx_bd */
1938 *tx_bd = (struct eth_tx_start_bd *)d_tx_bd;
1940 return bd_prod;
1943 static inline u16 bnx2x_csum_fix(unsigned char *t_header, u16 csum, s8 fix)
1945 if (fix > 0)
1946 csum = (u16) ~csum_fold(csum_sub(csum,
1947 csum_partial(t_header - fix, fix, 0)));
1949 else if (fix < 0)
1950 csum = (u16) ~csum_fold(csum_add(csum,
1951 csum_partial(t_header, -fix, 0)));
1953 return swab16(csum);
1956 static inline u32 bnx2x_xmit_type(struct bnx2x *bp, struct sk_buff *skb)
1958 u32 rc;
1960 if (skb->ip_summed != CHECKSUM_PARTIAL)
1961 rc = XMIT_PLAIN;
1963 else {
1964 if (vlan_get_protocol(skb) == htons(ETH_P_IPV6)) {
1965 rc = XMIT_CSUM_V6;
1966 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
1967 rc |= XMIT_CSUM_TCP;
1969 } else {
1970 rc = XMIT_CSUM_V4;
1971 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
1972 rc |= XMIT_CSUM_TCP;
1976 if (skb_is_gso_v6(skb))
1977 rc |= XMIT_GSO_V6 | XMIT_CSUM_TCP | XMIT_CSUM_V6;
1978 else if (skb_is_gso(skb))
1979 rc |= XMIT_GSO_V4 | XMIT_CSUM_V4 | XMIT_CSUM_TCP;
1981 return rc;
1984 #if (MAX_SKB_FRAGS >= MAX_FETCH_BD - 3)
1985 /* check if packet requires linearization (packet is too fragmented)
1986 no need to check fragmentation if page size > 8K (there will be no
1987 violation to FW restrictions) */
1988 static int bnx2x_pkt_req_lin(struct bnx2x *bp, struct sk_buff *skb,
1989 u32 xmit_type)
1991 int to_copy = 0;
1992 int hlen = 0;
1993 int first_bd_sz = 0;
1995 /* 3 = 1 (for linear data BD) + 2 (for PBD and last BD) */
1996 if (skb_shinfo(skb)->nr_frags >= (MAX_FETCH_BD - 3)) {
1998 if (xmit_type & XMIT_GSO) {
1999 unsigned short lso_mss = skb_shinfo(skb)->gso_size;
2000 /* Check if LSO packet needs to be copied:
2001 3 = 1 (for headers BD) + 2 (for PBD and last BD) */
2002 int wnd_size = MAX_FETCH_BD - 3;
2003 /* Number of windows to check */
2004 int num_wnds = skb_shinfo(skb)->nr_frags - wnd_size;
2005 int wnd_idx = 0;
2006 int frag_idx = 0;
2007 u32 wnd_sum = 0;
2009 /* Headers length */
2010 hlen = (int)(skb_transport_header(skb) - skb->data) +
2011 tcp_hdrlen(skb);
2013 /* Amount of data (w/o headers) on linear part of SKB*/
2014 first_bd_sz = skb_headlen(skb) - hlen;
2016 wnd_sum = first_bd_sz;
2018 /* Calculate the first sum - it's special */
2019 for (frag_idx = 0; frag_idx < wnd_size - 1; frag_idx++)
2020 wnd_sum +=
2021 skb_shinfo(skb)->frags[frag_idx].size;
2023 /* If there was data on linear skb data - check it */
2024 if (first_bd_sz > 0) {
2025 if (unlikely(wnd_sum < lso_mss)) {
2026 to_copy = 1;
2027 goto exit_lbl;
2030 wnd_sum -= first_bd_sz;
2033 /* Others are easier: run through the frag list and
2034 check all windows */
2035 for (wnd_idx = 0; wnd_idx <= num_wnds; wnd_idx++) {
2036 wnd_sum +=
2037 skb_shinfo(skb)->frags[wnd_idx + wnd_size - 1].size;
2039 if (unlikely(wnd_sum < lso_mss)) {
2040 to_copy = 1;
2041 break;
2043 wnd_sum -=
2044 skb_shinfo(skb)->frags[wnd_idx].size;
2046 } else {
2047 /* in non-LSO too fragmented packet should always
2048 be linearized */
2049 to_copy = 1;
2053 exit_lbl:
2054 if (unlikely(to_copy))
2055 DP(NETIF_MSG_TX_QUEUED,
2056 "Linearization IS REQUIRED for %s packet. "
2057 "num_frags %d hlen %d first_bd_sz %d\n",
2058 (xmit_type & XMIT_GSO) ? "LSO" : "non-LSO",
2059 skb_shinfo(skb)->nr_frags, hlen, first_bd_sz);
2061 return to_copy;
2063 #endif
2065 static inline void bnx2x_set_pbd_gso_e2(struct sk_buff *skb, u32 *parsing_data,
2066 u32 xmit_type)
2068 *parsing_data |= (skb_shinfo(skb)->gso_size <<
2069 ETH_TX_PARSE_BD_E2_LSO_MSS_SHIFT) &
2070 ETH_TX_PARSE_BD_E2_LSO_MSS;
2071 if ((xmit_type & XMIT_GSO_V6) &&
2072 (ipv6_hdr(skb)->nexthdr == NEXTHDR_IPV6))
2073 *parsing_data |= ETH_TX_PARSE_BD_E2_IPV6_WITH_EXT_HDR;
2077 * bnx2x_set_pbd_gso - update PBD in GSO case.
2079 * @skb: packet skb
2080 * @pbd: parse BD
2081 * @xmit_type: xmit flags
2083 static inline void bnx2x_set_pbd_gso(struct sk_buff *skb,
2084 struct eth_tx_parse_bd_e1x *pbd,
2085 u32 xmit_type)
2087 pbd->lso_mss = cpu_to_le16(skb_shinfo(skb)->gso_size);
2088 pbd->tcp_send_seq = swab32(tcp_hdr(skb)->seq);
2089 pbd->tcp_flags = pbd_tcp_flags(skb);
2091 if (xmit_type & XMIT_GSO_V4) {
2092 pbd->ip_id = swab16(ip_hdr(skb)->id);
2093 pbd->tcp_pseudo_csum =
2094 swab16(~csum_tcpudp_magic(ip_hdr(skb)->saddr,
2095 ip_hdr(skb)->daddr,
2096 0, IPPROTO_TCP, 0));
2098 } else
2099 pbd->tcp_pseudo_csum =
2100 swab16(~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2101 &ipv6_hdr(skb)->daddr,
2102 0, IPPROTO_TCP, 0));
2104 pbd->global_data |= ETH_TX_PARSE_BD_E1X_PSEUDO_CS_WITHOUT_LEN;
2108 * bnx2x_set_pbd_csum_e2 - update PBD with checksum and return header length
2110 * @bp: driver handle
2111 * @skb: packet skb
2112 * @parsing_data: data to be updated
2113 * @xmit_type: xmit flags
2115 * 57712 related
2117 static inline u8 bnx2x_set_pbd_csum_e2(struct bnx2x *bp, struct sk_buff *skb,
2118 u32 *parsing_data, u32 xmit_type)
2120 *parsing_data |=
2121 ((((u8 *)skb_transport_header(skb) - skb->data) >> 1) <<
2122 ETH_TX_PARSE_BD_E2_TCP_HDR_START_OFFSET_W_SHIFT) &
2123 ETH_TX_PARSE_BD_E2_TCP_HDR_START_OFFSET_W;
2125 if (xmit_type & XMIT_CSUM_TCP) {
2126 *parsing_data |= ((tcp_hdrlen(skb) / 4) <<
2127 ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW_SHIFT) &
2128 ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW;
2130 return skb_transport_header(skb) + tcp_hdrlen(skb) - skb->data;
2131 } else
2132 /* We support checksum offload for TCP and UDP only.
2133 * No need to pass the UDP header length - it's a constant.
2135 return skb_transport_header(skb) +
2136 sizeof(struct udphdr) - skb->data;
2139 static inline void bnx2x_set_sbd_csum(struct bnx2x *bp, struct sk_buff *skb,
2140 struct eth_tx_start_bd *tx_start_bd, u32 xmit_type)
2143 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_L4_CSUM;
2145 if (xmit_type & XMIT_CSUM_V4)
2146 tx_start_bd->bd_flags.as_bitfield |=
2147 ETH_TX_BD_FLAGS_IP_CSUM;
2148 else
2149 tx_start_bd->bd_flags.as_bitfield |=
2150 ETH_TX_BD_FLAGS_IPV6;
2152 if (!(xmit_type & XMIT_CSUM_TCP))
2153 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_IS_UDP;
2158 * bnx2x_set_pbd_csum - update PBD with checksum and return header length
2160 * @bp: driver handle
2161 * @skb: packet skb
2162 * @pbd: parse BD to be updated
2163 * @xmit_type: xmit flags
2165 static inline u8 bnx2x_set_pbd_csum(struct bnx2x *bp, struct sk_buff *skb,
2166 struct eth_tx_parse_bd_e1x *pbd,
2167 u32 xmit_type)
2169 u8 hlen = (skb_network_header(skb) - skb->data) >> 1;
2171 /* for now NS flag is not used in Linux */
2172 pbd->global_data =
2173 (hlen | ((skb->protocol == cpu_to_be16(ETH_P_8021Q)) <<
2174 ETH_TX_PARSE_BD_E1X_LLC_SNAP_EN_SHIFT));
2176 pbd->ip_hlen_w = (skb_transport_header(skb) -
2177 skb_network_header(skb)) >> 1;
2179 hlen += pbd->ip_hlen_w;
2181 /* We support checksum offload for TCP and UDP only */
2182 if (xmit_type & XMIT_CSUM_TCP)
2183 hlen += tcp_hdrlen(skb) / 2;
2184 else
2185 hlen += sizeof(struct udphdr) / 2;
2187 pbd->total_hlen_w = cpu_to_le16(hlen);
2188 hlen = hlen*2;
2190 if (xmit_type & XMIT_CSUM_TCP) {
2191 pbd->tcp_pseudo_csum = swab16(tcp_hdr(skb)->check);
2193 } else {
2194 s8 fix = SKB_CS_OFF(skb); /* signed! */
2196 DP(NETIF_MSG_TX_QUEUED,
2197 "hlen %d fix %d csum before fix %x\n",
2198 le16_to_cpu(pbd->total_hlen_w), fix, SKB_CS(skb));
2200 /* HW bug: fixup the CSUM */
2201 pbd->tcp_pseudo_csum =
2202 bnx2x_csum_fix(skb_transport_header(skb),
2203 SKB_CS(skb), fix);
2205 DP(NETIF_MSG_TX_QUEUED, "csum after fix %x\n",
2206 pbd->tcp_pseudo_csum);
2209 return hlen;
2212 /* called with netif_tx_lock
2213 * bnx2x_tx_int() runs without netif_tx_lock unless it needs to call
2214 * netif_wake_queue()
2216 netdev_tx_t bnx2x_start_xmit(struct sk_buff *skb, struct net_device *dev)
2218 struct bnx2x *bp = netdev_priv(dev);
2219 struct bnx2x_fastpath *fp;
2220 struct netdev_queue *txq;
2221 struct sw_tx_bd *tx_buf;
2222 struct eth_tx_start_bd *tx_start_bd;
2223 struct eth_tx_bd *tx_data_bd, *total_pkt_bd = NULL;
2224 struct eth_tx_parse_bd_e1x *pbd_e1x = NULL;
2225 struct eth_tx_parse_bd_e2 *pbd_e2 = NULL;
2226 u32 pbd_e2_parsing_data = 0;
2227 u16 pkt_prod, bd_prod;
2228 int nbd, fp_index;
2229 dma_addr_t mapping;
2230 u32 xmit_type = bnx2x_xmit_type(bp, skb);
2231 int i;
2232 u8 hlen = 0;
2233 __le16 pkt_size = 0;
2234 struct ethhdr *eth;
2235 u8 mac_type = UNICAST_ADDRESS;
2237 #ifdef BNX2X_STOP_ON_ERROR
2238 if (unlikely(bp->panic))
2239 return NETDEV_TX_BUSY;
2240 #endif
2242 fp_index = skb_get_queue_mapping(skb);
2243 txq = netdev_get_tx_queue(dev, fp_index);
2245 fp = &bp->fp[fp_index];
2247 if (unlikely(bnx2x_tx_avail(fp) < (skb_shinfo(skb)->nr_frags + 3))) {
2248 fp->eth_q_stats.driver_xoff++;
2249 netif_tx_stop_queue(txq);
2250 BNX2X_ERR("BUG! Tx ring full when queue awake!\n");
2251 return NETDEV_TX_BUSY;
2254 DP(NETIF_MSG_TX_QUEUED, "queue[%d]: SKB: summed %x protocol %x "
2255 "protocol(%x,%x) gso type %x xmit_type %x\n",
2256 fp_index, skb->ip_summed, skb->protocol, ipv6_hdr(skb)->nexthdr,
2257 ip_hdr(skb)->protocol, skb_shinfo(skb)->gso_type, xmit_type);
2259 eth = (struct ethhdr *)skb->data;
2261 /* set flag according to packet type (UNICAST_ADDRESS is default)*/
2262 if (unlikely(is_multicast_ether_addr(eth->h_dest))) {
2263 if (is_broadcast_ether_addr(eth->h_dest))
2264 mac_type = BROADCAST_ADDRESS;
2265 else
2266 mac_type = MULTICAST_ADDRESS;
2269 #if (MAX_SKB_FRAGS >= MAX_FETCH_BD - 3)
2270 /* First, check if we need to linearize the skb (due to FW
2271 restrictions). No need to check fragmentation if page size > 8K
2272 (there will be no violation to FW restrictions) */
2273 if (bnx2x_pkt_req_lin(bp, skb, xmit_type)) {
2274 /* Statistics of linearization */
2275 bp->lin_cnt++;
2276 if (skb_linearize(skb) != 0) {
2277 DP(NETIF_MSG_TX_QUEUED, "SKB linearization failed - "
2278 "silently dropping this SKB\n");
2279 dev_kfree_skb_any(skb);
2280 return NETDEV_TX_OK;
2283 #endif
2286 Please read carefully. First we use one BD which we mark as start,
2287 then we have a parsing info BD (used for TSO or xsum),
2288 and only then we have the rest of the TSO BDs.
2289 (don't forget to mark the last one as last,
2290 and to unmap only AFTER you write to the BD ...)
2291 And above all, all pdb sizes are in words - NOT DWORDS!
2294 pkt_prod = fp->tx_pkt_prod++;
2295 bd_prod = TX_BD(fp->tx_bd_prod);
2297 /* get a tx_buf and first BD */
2298 tx_buf = &fp->tx_buf_ring[TX_BD(pkt_prod)];
2299 tx_start_bd = &fp->tx_desc_ring[bd_prod].start_bd;
2301 tx_start_bd->bd_flags.as_bitfield = ETH_TX_BD_FLAGS_START_BD;
2302 SET_FLAG(tx_start_bd->general_data, ETH_TX_START_BD_ETH_ADDR_TYPE,
2303 mac_type);
2305 /* header nbd */
2306 SET_FLAG(tx_start_bd->general_data, ETH_TX_START_BD_HDR_NBDS, 1);
2308 /* remember the first BD of the packet */
2309 tx_buf->first_bd = fp->tx_bd_prod;
2310 tx_buf->skb = skb;
2311 tx_buf->flags = 0;
2313 DP(NETIF_MSG_TX_QUEUED,
2314 "sending pkt %u @%p next_idx %u bd %u @%p\n",
2315 pkt_prod, tx_buf, fp->tx_pkt_prod, bd_prod, tx_start_bd);
2317 if (vlan_tx_tag_present(skb)) {
2318 tx_start_bd->vlan_or_ethertype =
2319 cpu_to_le16(vlan_tx_tag_get(skb));
2320 tx_start_bd->bd_flags.as_bitfield |=
2321 (X_ETH_OUTBAND_VLAN << ETH_TX_BD_FLAGS_VLAN_MODE_SHIFT);
2322 } else
2323 tx_start_bd->vlan_or_ethertype = cpu_to_le16(pkt_prod);
2325 /* turn on parsing and get a BD */
2326 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
2328 if (xmit_type & XMIT_CSUM)
2329 bnx2x_set_sbd_csum(bp, skb, tx_start_bd, xmit_type);
2331 if (CHIP_IS_E2(bp)) {
2332 pbd_e2 = &fp->tx_desc_ring[bd_prod].parse_bd_e2;
2333 memset(pbd_e2, 0, sizeof(struct eth_tx_parse_bd_e2));
2334 /* Set PBD in checksum offload case */
2335 if (xmit_type & XMIT_CSUM)
2336 hlen = bnx2x_set_pbd_csum_e2(bp, skb,
2337 &pbd_e2_parsing_data,
2338 xmit_type);
2339 } else {
2340 pbd_e1x = &fp->tx_desc_ring[bd_prod].parse_bd_e1x;
2341 memset(pbd_e1x, 0, sizeof(struct eth_tx_parse_bd_e1x));
2342 /* Set PBD in checksum offload case */
2343 if (xmit_type & XMIT_CSUM)
2344 hlen = bnx2x_set_pbd_csum(bp, skb, pbd_e1x, xmit_type);
2348 /* Map skb linear data for DMA */
2349 mapping = dma_map_single(&bp->pdev->dev, skb->data,
2350 skb_headlen(skb), DMA_TO_DEVICE);
2352 /* Setup the data pointer of the first BD of the packet */
2353 tx_start_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
2354 tx_start_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
2355 nbd = skb_shinfo(skb)->nr_frags + 2; /* start_bd + pbd + frags */
2356 tx_start_bd->nbd = cpu_to_le16(nbd);
2357 tx_start_bd->nbytes = cpu_to_le16(skb_headlen(skb));
2358 pkt_size = tx_start_bd->nbytes;
2360 DP(NETIF_MSG_TX_QUEUED, "first bd @%p addr (%x:%x) nbd %d"
2361 " nbytes %d flags %x vlan %x\n",
2362 tx_start_bd, tx_start_bd->addr_hi, tx_start_bd->addr_lo,
2363 le16_to_cpu(tx_start_bd->nbd), le16_to_cpu(tx_start_bd->nbytes),
2364 tx_start_bd->bd_flags.as_bitfield,
2365 le16_to_cpu(tx_start_bd->vlan_or_ethertype));
2367 if (xmit_type & XMIT_GSO) {
2369 DP(NETIF_MSG_TX_QUEUED,
2370 "TSO packet len %d hlen %d total len %d tso size %d\n",
2371 skb->len, hlen, skb_headlen(skb),
2372 skb_shinfo(skb)->gso_size);
2374 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_SW_LSO;
2376 if (unlikely(skb_headlen(skb) > hlen))
2377 bd_prod = bnx2x_tx_split(bp, fp, tx_buf, &tx_start_bd,
2378 hlen, bd_prod, ++nbd);
2379 if (CHIP_IS_E2(bp))
2380 bnx2x_set_pbd_gso_e2(skb, &pbd_e2_parsing_data,
2381 xmit_type);
2382 else
2383 bnx2x_set_pbd_gso(skb, pbd_e1x, xmit_type);
2386 /* Set the PBD's parsing_data field if not zero
2387 * (for the chips newer than 57711).
2389 if (pbd_e2_parsing_data)
2390 pbd_e2->parsing_data = cpu_to_le32(pbd_e2_parsing_data);
2392 tx_data_bd = (struct eth_tx_bd *)tx_start_bd;
2394 /* Handle fragmented skb */
2395 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2396 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2398 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
2399 tx_data_bd = &fp->tx_desc_ring[bd_prod].reg_bd;
2400 if (total_pkt_bd == NULL)
2401 total_pkt_bd = &fp->tx_desc_ring[bd_prod].reg_bd;
2403 mapping = dma_map_page(&bp->pdev->dev, frag->page,
2404 frag->page_offset,
2405 frag->size, DMA_TO_DEVICE);
2407 tx_data_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
2408 tx_data_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
2409 tx_data_bd->nbytes = cpu_to_le16(frag->size);
2410 le16_add_cpu(&pkt_size, frag->size);
2412 DP(NETIF_MSG_TX_QUEUED,
2413 "frag %d bd @%p addr (%x:%x) nbytes %d\n",
2414 i, tx_data_bd, tx_data_bd->addr_hi, tx_data_bd->addr_lo,
2415 le16_to_cpu(tx_data_bd->nbytes));
2418 DP(NETIF_MSG_TX_QUEUED, "last bd @%p\n", tx_data_bd);
2420 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
2422 /* now send a tx doorbell, counting the next BD
2423 * if the packet contains or ends with it
2425 if (TX_BD_POFF(bd_prod) < nbd)
2426 nbd++;
2428 if (total_pkt_bd != NULL)
2429 total_pkt_bd->total_pkt_bytes = pkt_size;
2431 if (pbd_e1x)
2432 DP(NETIF_MSG_TX_QUEUED,
2433 "PBD (E1X) @%p ip_data %x ip_hlen %u ip_id %u lso_mss %u"
2434 " tcp_flags %x xsum %x seq %u hlen %u\n",
2435 pbd_e1x, pbd_e1x->global_data, pbd_e1x->ip_hlen_w,
2436 pbd_e1x->ip_id, pbd_e1x->lso_mss, pbd_e1x->tcp_flags,
2437 pbd_e1x->tcp_pseudo_csum, pbd_e1x->tcp_send_seq,
2438 le16_to_cpu(pbd_e1x->total_hlen_w));
2439 if (pbd_e2)
2440 DP(NETIF_MSG_TX_QUEUED,
2441 "PBD (E2) @%p dst %x %x %x src %x %x %x parsing_data %x\n",
2442 pbd_e2, pbd_e2->dst_mac_addr_hi, pbd_e2->dst_mac_addr_mid,
2443 pbd_e2->dst_mac_addr_lo, pbd_e2->src_mac_addr_hi,
2444 pbd_e2->src_mac_addr_mid, pbd_e2->src_mac_addr_lo,
2445 pbd_e2->parsing_data);
2446 DP(NETIF_MSG_TX_QUEUED, "doorbell: nbd %d bd %u\n", nbd, bd_prod);
2449 * Make sure that the BD data is updated before updating the producer
2450 * since FW might read the BD right after the producer is updated.
2451 * This is only applicable for weak-ordered memory model archs such
2452 * as IA-64. The following barrier is also mandatory since FW will
2453 * assumes packets must have BDs.
2455 wmb();
2457 fp->tx_db.data.prod += nbd;
2458 barrier();
2460 DOORBELL(bp, fp->cid, fp->tx_db.raw);
2462 mmiowb();
2464 fp->tx_bd_prod += nbd;
2466 if (unlikely(bnx2x_tx_avail(fp) < MAX_SKB_FRAGS + 3)) {
2467 netif_tx_stop_queue(txq);
2469 /* paired memory barrier is in bnx2x_tx_int(), we have to keep
2470 * ordering of set_bit() in netif_tx_stop_queue() and read of
2471 * fp->bd_tx_cons */
2472 smp_mb();
2474 fp->eth_q_stats.driver_xoff++;
2475 if (bnx2x_tx_avail(fp) >= MAX_SKB_FRAGS + 3)
2476 netif_tx_wake_queue(txq);
2478 fp->tx_pkt++;
2480 return NETDEV_TX_OK;
2483 /* called with rtnl_lock */
2484 int bnx2x_change_mac_addr(struct net_device *dev, void *p)
2486 struct sockaddr *addr = p;
2487 struct bnx2x *bp = netdev_priv(dev);
2489 if (!is_valid_ether_addr((u8 *)(addr->sa_data)))
2490 return -EINVAL;
2492 memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
2493 if (netif_running(dev))
2494 bnx2x_set_eth_mac(bp, 1);
2496 return 0;
2499 static void bnx2x_free_fp_mem_at(struct bnx2x *bp, int fp_index)
2501 union host_hc_status_block *sb = &bnx2x_fp(bp, fp_index, status_blk);
2502 struct bnx2x_fastpath *fp = &bp->fp[fp_index];
2504 /* Common */
2505 #ifdef BCM_CNIC
2506 if (IS_FCOE_IDX(fp_index)) {
2507 memset(sb, 0, sizeof(union host_hc_status_block));
2508 fp->status_blk_mapping = 0;
2510 } else {
2511 #endif
2512 /* status blocks */
2513 if (CHIP_IS_E2(bp))
2514 BNX2X_PCI_FREE(sb->e2_sb,
2515 bnx2x_fp(bp, fp_index,
2516 status_blk_mapping),
2517 sizeof(struct host_hc_status_block_e2));
2518 else
2519 BNX2X_PCI_FREE(sb->e1x_sb,
2520 bnx2x_fp(bp, fp_index,
2521 status_blk_mapping),
2522 sizeof(struct host_hc_status_block_e1x));
2523 #ifdef BCM_CNIC
2525 #endif
2526 /* Rx */
2527 if (!skip_rx_queue(bp, fp_index)) {
2528 bnx2x_free_rx_bds(fp);
2530 /* fastpath rx rings: rx_buf rx_desc rx_comp */
2531 BNX2X_FREE(bnx2x_fp(bp, fp_index, rx_buf_ring));
2532 BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_desc_ring),
2533 bnx2x_fp(bp, fp_index, rx_desc_mapping),
2534 sizeof(struct eth_rx_bd) * NUM_RX_BD);
2536 BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_comp_ring),
2537 bnx2x_fp(bp, fp_index, rx_comp_mapping),
2538 sizeof(struct eth_fast_path_rx_cqe) *
2539 NUM_RCQ_BD);
2541 /* SGE ring */
2542 BNX2X_FREE(bnx2x_fp(bp, fp_index, rx_page_ring));
2543 BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_sge_ring),
2544 bnx2x_fp(bp, fp_index, rx_sge_mapping),
2545 BCM_PAGE_SIZE * NUM_RX_SGE_PAGES);
2548 /* Tx */
2549 if (!skip_tx_queue(bp, fp_index)) {
2550 /* fastpath tx rings: tx_buf tx_desc */
2551 BNX2X_FREE(bnx2x_fp(bp, fp_index, tx_buf_ring));
2552 BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, tx_desc_ring),
2553 bnx2x_fp(bp, fp_index, tx_desc_mapping),
2554 sizeof(union eth_tx_bd_types) * NUM_TX_BD);
2556 /* end of fastpath */
2559 void bnx2x_free_fp_mem(struct bnx2x *bp)
2561 int i;
2562 for_each_queue(bp, i)
2563 bnx2x_free_fp_mem_at(bp, i);
2566 static inline void set_sb_shortcuts(struct bnx2x *bp, int index)
2568 union host_hc_status_block status_blk = bnx2x_fp(bp, index, status_blk);
2569 if (CHIP_IS_E2(bp)) {
2570 bnx2x_fp(bp, index, sb_index_values) =
2571 (__le16 *)status_blk.e2_sb->sb.index_values;
2572 bnx2x_fp(bp, index, sb_running_index) =
2573 (__le16 *)status_blk.e2_sb->sb.running_index;
2574 } else {
2575 bnx2x_fp(bp, index, sb_index_values) =
2576 (__le16 *)status_blk.e1x_sb->sb.index_values;
2577 bnx2x_fp(bp, index, sb_running_index) =
2578 (__le16 *)status_blk.e1x_sb->sb.running_index;
2582 static int bnx2x_alloc_fp_mem_at(struct bnx2x *bp, int index)
2584 union host_hc_status_block *sb;
2585 struct bnx2x_fastpath *fp = &bp->fp[index];
2586 int ring_size = 0;
2588 /* if rx_ring_size specified - use it */
2589 int rx_ring_size = bp->rx_ring_size ? bp->rx_ring_size :
2590 MAX_RX_AVAIL/bp->num_queues;
2592 /* allocate at least number of buffers required by FW */
2593 rx_ring_size = max_t(int, fp->disable_tpa ? MIN_RX_SIZE_NONTPA :
2594 MIN_RX_SIZE_TPA,
2595 rx_ring_size);
2597 bnx2x_fp(bp, index, bp) = bp;
2598 bnx2x_fp(bp, index, index) = index;
2600 /* Common */
2601 sb = &bnx2x_fp(bp, index, status_blk);
2602 #ifdef BCM_CNIC
2603 if (!IS_FCOE_IDX(index)) {
2604 #endif
2605 /* status blocks */
2606 if (CHIP_IS_E2(bp))
2607 BNX2X_PCI_ALLOC(sb->e2_sb,
2608 &bnx2x_fp(bp, index, status_blk_mapping),
2609 sizeof(struct host_hc_status_block_e2));
2610 else
2611 BNX2X_PCI_ALLOC(sb->e1x_sb,
2612 &bnx2x_fp(bp, index, status_blk_mapping),
2613 sizeof(struct host_hc_status_block_e1x));
2614 #ifdef BCM_CNIC
2616 #endif
2618 /* FCoE Queue uses Default SB and doesn't ACK the SB, thus no need to
2619 * set shortcuts for it.
2621 if (!IS_FCOE_IDX(index))
2622 set_sb_shortcuts(bp, index);
2624 /* Tx */
2625 if (!skip_tx_queue(bp, index)) {
2626 /* fastpath tx rings: tx_buf tx_desc */
2627 BNX2X_ALLOC(bnx2x_fp(bp, index, tx_buf_ring),
2628 sizeof(struct sw_tx_bd) * NUM_TX_BD);
2629 BNX2X_PCI_ALLOC(bnx2x_fp(bp, index, tx_desc_ring),
2630 &bnx2x_fp(bp, index, tx_desc_mapping),
2631 sizeof(union eth_tx_bd_types) * NUM_TX_BD);
2634 /* Rx */
2635 if (!skip_rx_queue(bp, index)) {
2636 /* fastpath rx rings: rx_buf rx_desc rx_comp */
2637 BNX2X_ALLOC(bnx2x_fp(bp, index, rx_buf_ring),
2638 sizeof(struct sw_rx_bd) * NUM_RX_BD);
2639 BNX2X_PCI_ALLOC(bnx2x_fp(bp, index, rx_desc_ring),
2640 &bnx2x_fp(bp, index, rx_desc_mapping),
2641 sizeof(struct eth_rx_bd) * NUM_RX_BD);
2643 BNX2X_PCI_ALLOC(bnx2x_fp(bp, index, rx_comp_ring),
2644 &bnx2x_fp(bp, index, rx_comp_mapping),
2645 sizeof(struct eth_fast_path_rx_cqe) *
2646 NUM_RCQ_BD);
2648 /* SGE ring */
2649 BNX2X_ALLOC(bnx2x_fp(bp, index, rx_page_ring),
2650 sizeof(struct sw_rx_page) * NUM_RX_SGE);
2651 BNX2X_PCI_ALLOC(bnx2x_fp(bp, index, rx_sge_ring),
2652 &bnx2x_fp(bp, index, rx_sge_mapping),
2653 BCM_PAGE_SIZE * NUM_RX_SGE_PAGES);
2654 /* RX BD ring */
2655 bnx2x_set_next_page_rx_bd(fp);
2657 /* CQ ring */
2658 bnx2x_set_next_page_rx_cq(fp);
2660 /* BDs */
2661 ring_size = bnx2x_alloc_rx_bds(fp, rx_ring_size);
2662 if (ring_size < rx_ring_size)
2663 goto alloc_mem_err;
2666 return 0;
2668 /* handles low memory cases */
2669 alloc_mem_err:
2670 BNX2X_ERR("Unable to allocate full memory for queue %d (size %d)\n",
2671 index, ring_size);
2672 /* FW will drop all packets if queue is not big enough,
2673 * In these cases we disable the queue
2674 * Min size diferent for TPA and non-TPA queues
2676 if (ring_size < (fp->disable_tpa ?
2677 MIN_RX_SIZE_NONTPA : MIN_RX_SIZE_TPA)) {
2678 /* release memory allocated for this queue */
2679 bnx2x_free_fp_mem_at(bp, index);
2680 return -ENOMEM;
2682 return 0;
2685 int bnx2x_alloc_fp_mem(struct bnx2x *bp)
2687 int i;
2690 * 1. Allocate FP for leading - fatal if error
2691 * 2. {CNIC} Allocate FCoE FP - fatal if error
2692 * 3. Allocate RSS - fix number of queues if error
2695 /* leading */
2696 if (bnx2x_alloc_fp_mem_at(bp, 0))
2697 return -ENOMEM;
2698 #ifdef BCM_CNIC
2699 if (!NO_FCOE(bp))
2700 /* FCoE */
2701 if (bnx2x_alloc_fp_mem_at(bp, FCOE_IDX))
2702 /* we will fail load process instead of mark
2703 * NO_FCOE_FLAG
2705 return -ENOMEM;
2706 #endif
2707 /* RSS */
2708 for_each_nondefault_eth_queue(bp, i)
2709 if (bnx2x_alloc_fp_mem_at(bp, i))
2710 break;
2712 /* handle memory failures */
2713 if (i != BNX2X_NUM_ETH_QUEUES(bp)) {
2714 int delta = BNX2X_NUM_ETH_QUEUES(bp) - i;
2716 WARN_ON(delta < 0);
2717 #ifdef BCM_CNIC
2719 * move non eth FPs next to last eth FP
2720 * must be done in that order
2721 * FCOE_IDX < FWD_IDX < OOO_IDX
2724 /* move FCoE fp */
2725 bnx2x_move_fp(bp, FCOE_IDX, FCOE_IDX - delta);
2726 #endif
2727 bp->num_queues -= delta;
2728 BNX2X_ERR("Adjusted num of queues from %d to %d\n",
2729 bp->num_queues + delta, bp->num_queues);
2732 return 0;
2735 static int bnx2x_setup_irqs(struct bnx2x *bp)
2737 int rc = 0;
2738 if (bp->flags & USING_MSIX_FLAG) {
2739 rc = bnx2x_req_msix_irqs(bp);
2740 if (rc)
2741 return rc;
2742 } else {
2743 bnx2x_ack_int(bp);
2744 rc = bnx2x_req_irq(bp);
2745 if (rc) {
2746 BNX2X_ERR("IRQ request failed rc %d, aborting\n", rc);
2747 return rc;
2749 if (bp->flags & USING_MSI_FLAG) {
2750 bp->dev->irq = bp->pdev->irq;
2751 netdev_info(bp->dev, "using MSI IRQ %d\n",
2752 bp->pdev->irq);
2756 return 0;
2759 void bnx2x_free_mem_bp(struct bnx2x *bp)
2761 kfree(bp->fp);
2762 kfree(bp->msix_table);
2763 kfree(bp->ilt);
2766 int __devinit bnx2x_alloc_mem_bp(struct bnx2x *bp)
2768 struct bnx2x_fastpath *fp;
2769 struct msix_entry *tbl;
2770 struct bnx2x_ilt *ilt;
2772 /* fp array */
2773 fp = kzalloc(L2_FP_COUNT(bp->l2_cid_count)*sizeof(*fp), GFP_KERNEL);
2774 if (!fp)
2775 goto alloc_err;
2776 bp->fp = fp;
2778 /* msix table */
2779 tbl = kzalloc((FP_SB_COUNT(bp->l2_cid_count) + 1) * sizeof(*tbl),
2780 GFP_KERNEL);
2781 if (!tbl)
2782 goto alloc_err;
2783 bp->msix_table = tbl;
2785 /* ilt */
2786 ilt = kzalloc(sizeof(*ilt), GFP_KERNEL);
2787 if (!ilt)
2788 goto alloc_err;
2789 bp->ilt = ilt;
2791 return 0;
2792 alloc_err:
2793 bnx2x_free_mem_bp(bp);
2794 return -ENOMEM;
2798 static int bnx2x_reload_if_running(struct net_device *dev)
2800 struct bnx2x *bp = netdev_priv(dev);
2802 if (unlikely(!netif_running(dev)))
2803 return 0;
2805 bnx2x_nic_unload(bp, UNLOAD_NORMAL);
2806 return bnx2x_nic_load(bp, LOAD_NORMAL);
2809 int bnx2x_get_cur_phy_idx(struct bnx2x *bp)
2811 u32 sel_phy_idx = 0;
2812 if (bp->link_params.num_phys <= 1)
2813 return INT_PHY;
2815 if (bp->link_vars.link_up) {
2816 sel_phy_idx = EXT_PHY1;
2817 /* In case link is SERDES, check if the EXT_PHY2 is the one */
2818 if ((bp->link_vars.link_status & LINK_STATUS_SERDES_LINK) &&
2819 (bp->link_params.phy[EXT_PHY2].supported & SUPPORTED_FIBRE))
2820 sel_phy_idx = EXT_PHY2;
2821 } else {
2823 switch (bnx2x_phy_selection(&bp->link_params)) {
2824 case PORT_HW_CFG_PHY_SELECTION_HARDWARE_DEFAULT:
2825 case PORT_HW_CFG_PHY_SELECTION_FIRST_PHY:
2826 case PORT_HW_CFG_PHY_SELECTION_FIRST_PHY_PRIORITY:
2827 sel_phy_idx = EXT_PHY1;
2828 break;
2829 case PORT_HW_CFG_PHY_SELECTION_SECOND_PHY:
2830 case PORT_HW_CFG_PHY_SELECTION_SECOND_PHY_PRIORITY:
2831 sel_phy_idx = EXT_PHY2;
2832 break;
2836 return sel_phy_idx;
2839 int bnx2x_get_link_cfg_idx(struct bnx2x *bp)
2841 u32 sel_phy_idx = bnx2x_get_cur_phy_idx(bp);
2843 * The selected actived PHY is always after swapping (in case PHY
2844 * swapping is enabled). So when swapping is enabled, we need to reverse
2845 * the configuration
2848 if (bp->link_params.multi_phy_config &
2849 PORT_HW_CFG_PHY_SWAPPED_ENABLED) {
2850 if (sel_phy_idx == EXT_PHY1)
2851 sel_phy_idx = EXT_PHY2;
2852 else if (sel_phy_idx == EXT_PHY2)
2853 sel_phy_idx = EXT_PHY1;
2855 return LINK_CONFIG_IDX(sel_phy_idx);
2858 /* called with rtnl_lock */
2859 int bnx2x_change_mtu(struct net_device *dev, int new_mtu)
2861 struct bnx2x *bp = netdev_priv(dev);
2863 if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
2864 printk(KERN_ERR "Handling parity error recovery. Try again later\n");
2865 return -EAGAIN;
2868 if ((new_mtu > ETH_MAX_JUMBO_PACKET_SIZE) ||
2869 ((new_mtu + ETH_HLEN) < ETH_MIN_PACKET_SIZE))
2870 return -EINVAL;
2872 /* This does not race with packet allocation
2873 * because the actual alloc size is
2874 * only updated as part of load
2876 dev->mtu = new_mtu;
2878 return bnx2x_reload_if_running(dev);
2881 u32 bnx2x_fix_features(struct net_device *dev, u32 features)
2883 struct bnx2x *bp = netdev_priv(dev);
2885 /* TPA requires Rx CSUM offloading */
2886 if (!(features & NETIF_F_RXCSUM) || bp->disable_tpa)
2887 features &= ~NETIF_F_LRO;
2889 return features;
2892 int bnx2x_set_features(struct net_device *dev, u32 features)
2894 struct bnx2x *bp = netdev_priv(dev);
2895 u32 flags = bp->flags;
2896 bool bnx2x_reload = false;
2898 if (features & NETIF_F_LRO)
2899 flags |= TPA_ENABLE_FLAG;
2900 else
2901 flags &= ~TPA_ENABLE_FLAG;
2903 if (features & NETIF_F_LOOPBACK) {
2904 if (bp->link_params.loopback_mode != LOOPBACK_BMAC) {
2905 bp->link_params.loopback_mode = LOOPBACK_BMAC;
2906 bnx2x_reload = true;
2908 } else {
2909 if (bp->link_params.loopback_mode != LOOPBACK_NONE) {
2910 bp->link_params.loopback_mode = LOOPBACK_NONE;
2911 bnx2x_reload = true;
2915 if (flags ^ bp->flags) {
2916 bp->flags = flags;
2917 bnx2x_reload = true;
2920 if (bnx2x_reload) {
2921 if (bp->recovery_state == BNX2X_RECOVERY_DONE)
2922 return bnx2x_reload_if_running(dev);
2923 /* else: bnx2x_nic_load() will be called at end of recovery */
2926 return 0;
2929 void bnx2x_tx_timeout(struct net_device *dev)
2931 struct bnx2x *bp = netdev_priv(dev);
2933 #ifdef BNX2X_STOP_ON_ERROR
2934 if (!bp->panic)
2935 bnx2x_panic();
2936 #endif
2937 /* This allows the netif to be shutdown gracefully before resetting */
2938 schedule_delayed_work(&bp->reset_task, 0);
2941 int bnx2x_suspend(struct pci_dev *pdev, pm_message_t state)
2943 struct net_device *dev = pci_get_drvdata(pdev);
2944 struct bnx2x *bp;
2946 if (!dev) {
2947 dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
2948 return -ENODEV;
2950 bp = netdev_priv(dev);
2952 rtnl_lock();
2954 pci_save_state(pdev);
2956 if (!netif_running(dev)) {
2957 rtnl_unlock();
2958 return 0;
2961 netif_device_detach(dev);
2963 bnx2x_nic_unload(bp, UNLOAD_CLOSE);
2965 bnx2x_set_power_state(bp, pci_choose_state(pdev, state));
2967 rtnl_unlock();
2969 return 0;
2972 int bnx2x_resume(struct pci_dev *pdev)
2974 struct net_device *dev = pci_get_drvdata(pdev);
2975 struct bnx2x *bp;
2976 int rc;
2978 if (!dev) {
2979 dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
2980 return -ENODEV;
2982 bp = netdev_priv(dev);
2984 if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
2985 printk(KERN_ERR "Handling parity error recovery. Try again later\n");
2986 return -EAGAIN;
2989 rtnl_lock();
2991 pci_restore_state(pdev);
2993 if (!netif_running(dev)) {
2994 rtnl_unlock();
2995 return 0;
2998 bnx2x_set_power_state(bp, PCI_D0);
2999 netif_device_attach(dev);
3001 /* Since the chip was reset, clear the FW sequence number */
3002 bp->fw_seq = 0;
3003 rc = bnx2x_nic_load(bp, LOAD_OPEN);
3005 rtnl_unlock();
3007 return rc;