269d6d1e0d2f2f887ef8de858b7c86032db8e125
[deliverable/linux.git] / drivers / net / bnx2x / bnx2x_cmn.c
1 /* bnx2x_cmn.c: Broadcom Everest network driver.
2 *
3 * Copyright (c) 2007-2011 Broadcom Corporation
4 *
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.
8 *
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
15 *
16 */
17
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"
25
26 #include "bnx2x_init.h"
27
28 static int bnx2x_setup_irqs(struct bnx2x *bp);
29
30 /**
31 * bnx2x_bz_fp - zero content of the fastpath structure.
32 *
33 * @bp: driver handle
34 * @index: fastpath index to be zeroed
35 *
36 * Makes sure the contents of the bp->fp[index].napi is kept
37 * intact.
38 */
39 static inline void bnx2x_bz_fp(struct bnx2x *bp, int index)
40 {
41 struct bnx2x_fastpath *fp = &bp->fp[index];
42 struct napi_struct orig_napi = fp->napi;
43 /* bzero bnx2x_fastpath contents */
44 memset(fp, 0, sizeof(*fp));
45
46 /* Restore the NAPI object as it has been already initialized */
47 fp->napi = orig_napi;
48 }
49
50 /**
51 * bnx2x_move_fp - move content of the fastpath structure.
52 *
53 * @bp: driver handle
54 * @from: source FP index
55 * @to: destination FP index
56 *
57 * Makes sure the contents of the bp->fp[to].napi is kept
58 * intact.
59 */
60 static inline void bnx2x_move_fp(struct bnx2x *bp, int from, int to)
61 {
62 struct bnx2x_fastpath *from_fp = &bp->fp[from];
63 struct bnx2x_fastpath *to_fp = &bp->fp[to];
64 struct napi_struct orig_napi = to_fp->napi;
65 /* Move bnx2x_fastpath contents */
66 memcpy(to_fp, from_fp, sizeof(*to_fp));
67 to_fp->index = to;
68
69 /* Restore the NAPI object as it has been already initialized */
70 to_fp->napi = orig_napi;
71 }
72
73 /* free skb in the packet ring at pos idx
74 * return idx of last bd freed
75 */
76 static u16 bnx2x_free_tx_pkt(struct bnx2x *bp, struct bnx2x_fastpath *fp,
77 u16 idx)
78 {
79 struct sw_tx_bd *tx_buf = &fp->tx_buf_ring[idx];
80 struct eth_tx_start_bd *tx_start_bd;
81 struct eth_tx_bd *tx_data_bd;
82 struct sk_buff *skb = tx_buf->skb;
83 u16 bd_idx = TX_BD(tx_buf->first_bd), new_cons;
84 int nbd;
85
86 /* prefetch skb end pointer to speedup dev_kfree_skb() */
87 prefetch(&skb->end);
88
89 DP(BNX2X_MSG_OFF, "pkt_idx %d buff @(%p)->skb %p\n",
90 idx, tx_buf, skb);
91
92 /* unmap first bd */
93 DP(BNX2X_MSG_OFF, "free bd_idx %d\n", bd_idx);
94 tx_start_bd = &fp->tx_desc_ring[bd_idx].start_bd;
95 dma_unmap_single(&bp->pdev->dev, BD_UNMAP_ADDR(tx_start_bd),
96 BD_UNMAP_LEN(tx_start_bd), DMA_TO_DEVICE);
97
98 nbd = le16_to_cpu(tx_start_bd->nbd) - 1;
99 #ifdef BNX2X_STOP_ON_ERROR
100 if ((nbd - 1) > (MAX_SKB_FRAGS + 2)) {
101 BNX2X_ERR("BAD nbd!\n");
102 bnx2x_panic();
103 }
104 #endif
105 new_cons = nbd + tx_buf->first_bd;
106
107 /* Get the next bd */
108 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
109
110 /* Skip a parse bd... */
111 --nbd;
112 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
113
114 /* ...and the TSO split header bd since they have no mapping */
115 if (tx_buf->flags & BNX2X_TSO_SPLIT_BD) {
116 --nbd;
117 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
118 }
119
120 /* now free frags */
121 while (nbd > 0) {
122
123 DP(BNX2X_MSG_OFF, "free frag bd_idx %d\n", bd_idx);
124 tx_data_bd = &fp->tx_desc_ring[bd_idx].reg_bd;
125 dma_unmap_page(&bp->pdev->dev, BD_UNMAP_ADDR(tx_data_bd),
126 BD_UNMAP_LEN(tx_data_bd), DMA_TO_DEVICE);
127 if (--nbd)
128 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
129 }
130
131 /* release skb */
132 WARN_ON(!skb);
133 dev_kfree_skb(skb);
134 tx_buf->first_bd = 0;
135 tx_buf->skb = NULL;
136
137 return new_cons;
138 }
139
140 int bnx2x_tx_int(struct bnx2x_fastpath *fp)
141 {
142 struct bnx2x *bp = fp->bp;
143 struct netdev_queue *txq;
144 u16 hw_cons, sw_cons, bd_cons = fp->tx_bd_cons;
145
146 #ifdef BNX2X_STOP_ON_ERROR
147 if (unlikely(bp->panic))
148 return -1;
149 #endif
150
151 txq = netdev_get_tx_queue(bp->dev, fp->index);
152 hw_cons = le16_to_cpu(*fp->tx_cons_sb);
153 sw_cons = fp->tx_pkt_cons;
154
155 while (sw_cons != hw_cons) {
156 u16 pkt_cons;
157
158 pkt_cons = TX_BD(sw_cons);
159
160 DP(NETIF_MSG_TX_DONE, "queue[%d]: hw_cons %u sw_cons %u "
161 " pkt_cons %u\n",
162 fp->index, hw_cons, sw_cons, pkt_cons);
163
164 bd_cons = bnx2x_free_tx_pkt(bp, fp, pkt_cons);
165 sw_cons++;
166 }
167
168 fp->tx_pkt_cons = sw_cons;
169 fp->tx_bd_cons = bd_cons;
170
171 /* Need to make the tx_bd_cons update visible to start_xmit()
172 * before checking for netif_tx_queue_stopped(). Without the
173 * memory barrier, there is a small possibility that
174 * start_xmit() will miss it and cause the queue to be stopped
175 * forever.
176 */
177 smp_mb();
178
179 if (unlikely(netif_tx_queue_stopped(txq))) {
180 /* Taking tx_lock() is needed to prevent reenabling the queue
181 * while it's empty. This could have happen if rx_action() gets
182 * suspended in bnx2x_tx_int() after the condition before
183 * netif_tx_wake_queue(), while tx_action (bnx2x_start_xmit()):
184 *
185 * stops the queue->sees fresh tx_bd_cons->releases the queue->
186 * sends some packets consuming the whole queue again->
187 * stops the queue
188 */
189
190 __netif_tx_lock(txq, smp_processor_id());
191
192 if ((netif_tx_queue_stopped(txq)) &&
193 (bp->state == BNX2X_STATE_OPEN) &&
194 (bnx2x_tx_avail(fp) >= MAX_SKB_FRAGS + 3))
195 netif_tx_wake_queue(txq);
196
197 __netif_tx_unlock(txq);
198 }
199 return 0;
200 }
201
202 static inline void bnx2x_update_last_max_sge(struct bnx2x_fastpath *fp,
203 u16 idx)
204 {
205 u16 last_max = fp->last_max_sge;
206
207 if (SUB_S16(idx, last_max) > 0)
208 fp->last_max_sge = idx;
209 }
210
211 static void bnx2x_update_sge_prod(struct bnx2x_fastpath *fp,
212 struct eth_fast_path_rx_cqe *fp_cqe)
213 {
214 struct bnx2x *bp = fp->bp;
215 u16 sge_len = SGE_PAGE_ALIGN(le16_to_cpu(fp_cqe->pkt_len) -
216 le16_to_cpu(fp_cqe->len_on_bd)) >>
217 SGE_PAGE_SHIFT;
218 u16 last_max, last_elem, first_elem;
219 u16 delta = 0;
220 u16 i;
221
222 if (!sge_len)
223 return;
224
225 /* First mark all used pages */
226 for (i = 0; i < sge_len; i++)
227 SGE_MASK_CLEAR_BIT(fp,
228 RX_SGE(le16_to_cpu(fp_cqe->sgl_or_raw_data.sgl[i])));
229
230 DP(NETIF_MSG_RX_STATUS, "fp_cqe->sgl[%d] = %d\n",
231 sge_len - 1, le16_to_cpu(fp_cqe->sgl_or_raw_data.sgl[sge_len - 1]));
232
233 /* Here we assume that the last SGE index is the biggest */
234 prefetch((void *)(fp->sge_mask));
235 bnx2x_update_last_max_sge(fp,
236 le16_to_cpu(fp_cqe->sgl_or_raw_data.sgl[sge_len - 1]));
237
238 last_max = RX_SGE(fp->last_max_sge);
239 last_elem = last_max >> RX_SGE_MASK_ELEM_SHIFT;
240 first_elem = RX_SGE(fp->rx_sge_prod) >> RX_SGE_MASK_ELEM_SHIFT;
241
242 /* If ring is not full */
243 if (last_elem + 1 != first_elem)
244 last_elem++;
245
246 /* Now update the prod */
247 for (i = first_elem; i != last_elem; i = NEXT_SGE_MASK_ELEM(i)) {
248 if (likely(fp->sge_mask[i]))
249 break;
250
251 fp->sge_mask[i] = RX_SGE_MASK_ELEM_ONE_MASK;
252 delta += RX_SGE_MASK_ELEM_SZ;
253 }
254
255 if (delta > 0) {
256 fp->rx_sge_prod += delta;
257 /* clear page-end entries */
258 bnx2x_clear_sge_mask_next_elems(fp);
259 }
260
261 DP(NETIF_MSG_RX_STATUS,
262 "fp->last_max_sge = %d fp->rx_sge_prod = %d\n",
263 fp->last_max_sge, fp->rx_sge_prod);
264 }
265
266 static void bnx2x_tpa_start(struct bnx2x_fastpath *fp, u16 queue,
267 struct sk_buff *skb, u16 cons, u16 prod)
268 {
269 struct bnx2x *bp = fp->bp;
270 struct sw_rx_bd *cons_rx_buf = &fp->rx_buf_ring[cons];
271 struct sw_rx_bd *prod_rx_buf = &fp->rx_buf_ring[prod];
272 struct eth_rx_bd *prod_bd = &fp->rx_desc_ring[prod];
273 dma_addr_t mapping;
274
275 /* move empty skb from pool to prod and map it */
276 prod_rx_buf->skb = fp->tpa_pool[queue].skb;
277 mapping = dma_map_single(&bp->pdev->dev, fp->tpa_pool[queue].skb->data,
278 fp->rx_buf_size, DMA_FROM_DEVICE);
279 dma_unmap_addr_set(prod_rx_buf, mapping, mapping);
280
281 /* move partial skb from cons to pool (don't unmap yet) */
282 fp->tpa_pool[queue] = *cons_rx_buf;
283
284 /* mark bin state as start - print error if current state != stop */
285 if (fp->tpa_state[queue] != BNX2X_TPA_STOP)
286 BNX2X_ERR("start of bin not in stop [%d]\n", queue);
287
288 fp->tpa_state[queue] = BNX2X_TPA_START;
289
290 /* point prod_bd to new skb */
291 prod_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
292 prod_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
293
294 #ifdef BNX2X_STOP_ON_ERROR
295 fp->tpa_queue_used |= (1 << queue);
296 #ifdef _ASM_GENERIC_INT_L64_H
297 DP(NETIF_MSG_RX_STATUS, "fp->tpa_queue_used = 0x%lx\n",
298 #else
299 DP(NETIF_MSG_RX_STATUS, "fp->tpa_queue_used = 0x%llx\n",
300 #endif
301 fp->tpa_queue_used);
302 #endif
303 }
304
305 /* Timestamp option length allowed for TPA aggregation:
306 *
307 * nop nop kind length echo val
308 */
309 #define TPA_TSTAMP_OPT_LEN 12
310 /**
311 * Calculate the approximate value of the MSS for this
312 * aggregation using the first packet of it.
313 *
314 * @param bp
315 * @param parsing_flags Parsing flags from the START CQE
316 * @param len_on_bd Total length of the first packet for the
317 * aggregation.
318 */
319 static inline u16 bnx2x_set_lro_mss(struct bnx2x *bp, u16 parsing_flags,
320 u16 len_on_bd)
321 {
322 /* TPA arrgregation won't have an IP options and TCP options
323 * other than timestamp.
324 */
325 u16 hdrs_len = ETH_HLEN + sizeof(struct iphdr) + sizeof(struct tcphdr);
326
327
328 /* Check if there was a TCP timestamp, if there is it's will
329 * always be 12 bytes length: nop nop kind length echo val.
330 *
331 * Otherwise FW would close the aggregation.
332 */
333 if (parsing_flags & PARSING_FLAGS_TIME_STAMP_EXIST_FLAG)
334 hdrs_len += TPA_TSTAMP_OPT_LEN;
335
336 return len_on_bd - hdrs_len;
337 }
338
339 static int bnx2x_fill_frag_skb(struct bnx2x *bp, struct bnx2x_fastpath *fp,
340 struct sk_buff *skb,
341 struct eth_fast_path_rx_cqe *fp_cqe,
342 u16 cqe_idx, u16 parsing_flags)
343 {
344 struct sw_rx_page *rx_pg, old_rx_pg;
345 u16 len_on_bd = le16_to_cpu(fp_cqe->len_on_bd);
346 u32 i, frag_len, frag_size, pages;
347 int err;
348 int j;
349
350 frag_size = le16_to_cpu(fp_cqe->pkt_len) - len_on_bd;
351 pages = SGE_PAGE_ALIGN(frag_size) >> SGE_PAGE_SHIFT;
352
353 /* This is needed in order to enable forwarding support */
354 if (frag_size)
355 skb_shinfo(skb)->gso_size = bnx2x_set_lro_mss(bp, parsing_flags,
356 len_on_bd);
357
358 #ifdef BNX2X_STOP_ON_ERROR
359 if (pages > min_t(u32, 8, MAX_SKB_FRAGS)*SGE_PAGE_SIZE*PAGES_PER_SGE) {
360 BNX2X_ERR("SGL length is too long: %d. CQE index is %d\n",
361 pages, cqe_idx);
362 BNX2X_ERR("fp_cqe->pkt_len = %d fp_cqe->len_on_bd = %d\n",
363 fp_cqe->pkt_len, len_on_bd);
364 bnx2x_panic();
365 return -EINVAL;
366 }
367 #endif
368
369 /* Run through the SGL and compose the fragmented skb */
370 for (i = 0, j = 0; i < pages; i += PAGES_PER_SGE, j++) {
371 u16 sge_idx =
372 RX_SGE(le16_to_cpu(fp_cqe->sgl_or_raw_data.sgl[j]));
373
374 /* FW gives the indices of the SGE as if the ring is an array
375 (meaning that "next" element will consume 2 indices) */
376 frag_len = min(frag_size, (u32)(SGE_PAGE_SIZE*PAGES_PER_SGE));
377 rx_pg = &fp->rx_page_ring[sge_idx];
378 old_rx_pg = *rx_pg;
379
380 /* If we fail to allocate a substitute page, we simply stop
381 where we are and drop the whole packet */
382 err = bnx2x_alloc_rx_sge(bp, fp, sge_idx);
383 if (unlikely(err)) {
384 fp->eth_q_stats.rx_skb_alloc_failed++;
385 return err;
386 }
387
388 /* Unmap the page as we r going to pass it to the stack */
389 dma_unmap_page(&bp->pdev->dev,
390 dma_unmap_addr(&old_rx_pg, mapping),
391 SGE_PAGE_SIZE*PAGES_PER_SGE, DMA_FROM_DEVICE);
392
393 /* Add one frag and update the appropriate fields in the skb */
394 skb_fill_page_desc(skb, j, old_rx_pg.page, 0, frag_len);
395
396 skb->data_len += frag_len;
397 skb->truesize += frag_len;
398 skb->len += frag_len;
399
400 frag_size -= frag_len;
401 }
402
403 return 0;
404 }
405
406 static void bnx2x_tpa_stop(struct bnx2x *bp, struct bnx2x_fastpath *fp,
407 u16 queue, int pad, int len, union eth_rx_cqe *cqe,
408 u16 cqe_idx)
409 {
410 struct sw_rx_bd *rx_buf = &fp->tpa_pool[queue];
411 struct sk_buff *skb = rx_buf->skb;
412 /* alloc new skb */
413 struct sk_buff *new_skb = netdev_alloc_skb(bp->dev, fp->rx_buf_size);
414
415 /* Unmap skb in the pool anyway, as we are going to change
416 pool entry status to BNX2X_TPA_STOP even if new skb allocation
417 fails. */
418 dma_unmap_single(&bp->pdev->dev, dma_unmap_addr(rx_buf, mapping),
419 fp->rx_buf_size, DMA_FROM_DEVICE);
420
421 if (likely(new_skb)) {
422 /* fix ip xsum and give it to the stack */
423 /* (no need to map the new skb) */
424 u16 parsing_flags =
425 le16_to_cpu(cqe->fast_path_cqe.pars_flags.flags);
426
427 prefetch(skb);
428 prefetch(((char *)(skb)) + L1_CACHE_BYTES);
429
430 #ifdef BNX2X_STOP_ON_ERROR
431 if (pad + len > fp->rx_buf_size) {
432 BNX2X_ERR("skb_put is about to fail... "
433 "pad %d len %d rx_buf_size %d\n",
434 pad, len, fp->rx_buf_size);
435 bnx2x_panic();
436 return;
437 }
438 #endif
439
440 skb_reserve(skb, pad);
441 skb_put(skb, len);
442
443 skb->protocol = eth_type_trans(skb, bp->dev);
444 skb->ip_summed = CHECKSUM_UNNECESSARY;
445
446 {
447 struct iphdr *iph;
448
449 iph = (struct iphdr *)skb->data;
450 iph->check = 0;
451 iph->check = ip_fast_csum((u8 *)iph, iph->ihl);
452 }
453
454 if (!bnx2x_fill_frag_skb(bp, fp, skb,
455 &cqe->fast_path_cqe, cqe_idx,
456 parsing_flags)) {
457 if (parsing_flags & PARSING_FLAGS_VLAN)
458 __vlan_hwaccel_put_tag(skb,
459 le16_to_cpu(cqe->fast_path_cqe.
460 vlan_tag));
461 napi_gro_receive(&fp->napi, skb);
462 } else {
463 DP(NETIF_MSG_RX_STATUS, "Failed to allocate new pages"
464 " - dropping packet!\n");
465 dev_kfree_skb(skb);
466 }
467
468
469 /* put new skb in bin */
470 fp->tpa_pool[queue].skb = new_skb;
471
472 } else {
473 /* else drop the packet and keep the buffer in the bin */
474 DP(NETIF_MSG_RX_STATUS,
475 "Failed to allocate new skb - dropping packet!\n");
476 fp->eth_q_stats.rx_skb_alloc_failed++;
477 }
478
479 fp->tpa_state[queue] = BNX2X_TPA_STOP;
480 }
481
482 /* Set Toeplitz hash value in the skb using the value from the
483 * CQE (calculated by HW).
484 */
485 static inline void bnx2x_set_skb_rxhash(struct bnx2x *bp, union eth_rx_cqe *cqe,
486 struct sk_buff *skb)
487 {
488 /* Set Toeplitz hash from CQE */
489 if ((bp->dev->features & NETIF_F_RXHASH) &&
490 (cqe->fast_path_cqe.status_flags &
491 ETH_FAST_PATH_RX_CQE_RSS_HASH_FLG))
492 skb->rxhash =
493 le32_to_cpu(cqe->fast_path_cqe.rss_hash_result);
494 }
495
496 int bnx2x_rx_int(struct bnx2x_fastpath *fp, int budget)
497 {
498 struct bnx2x *bp = fp->bp;
499 u16 bd_cons, bd_prod, bd_prod_fw, comp_ring_cons;
500 u16 hw_comp_cons, sw_comp_cons, sw_comp_prod;
501 int rx_pkt = 0;
502
503 #ifdef BNX2X_STOP_ON_ERROR
504 if (unlikely(bp->panic))
505 return 0;
506 #endif
507
508 /* CQ "next element" is of the size of the regular element,
509 that's why it's ok here */
510 hw_comp_cons = le16_to_cpu(*fp->rx_cons_sb);
511 if ((hw_comp_cons & MAX_RCQ_DESC_CNT) == MAX_RCQ_DESC_CNT)
512 hw_comp_cons++;
513
514 bd_cons = fp->rx_bd_cons;
515 bd_prod = fp->rx_bd_prod;
516 bd_prod_fw = bd_prod;
517 sw_comp_cons = fp->rx_comp_cons;
518 sw_comp_prod = fp->rx_comp_prod;
519
520 /* Memory barrier necessary as speculative reads of the rx
521 * buffer can be ahead of the index in the status block
522 */
523 rmb();
524
525 DP(NETIF_MSG_RX_STATUS,
526 "queue[%d]: hw_comp_cons %u sw_comp_cons %u\n",
527 fp->index, hw_comp_cons, sw_comp_cons);
528
529 while (sw_comp_cons != hw_comp_cons) {
530 struct sw_rx_bd *rx_buf = NULL;
531 struct sk_buff *skb;
532 union eth_rx_cqe *cqe;
533 u8 cqe_fp_flags;
534 u16 len, pad;
535
536 comp_ring_cons = RCQ_BD(sw_comp_cons);
537 bd_prod = RX_BD(bd_prod);
538 bd_cons = RX_BD(bd_cons);
539
540 /* Prefetch the page containing the BD descriptor
541 at producer's index. It will be needed when new skb is
542 allocated */
543 prefetch((void *)(PAGE_ALIGN((unsigned long)
544 (&fp->rx_desc_ring[bd_prod])) -
545 PAGE_SIZE + 1));
546
547 cqe = &fp->rx_comp_ring[comp_ring_cons];
548 cqe_fp_flags = cqe->fast_path_cqe.type_error_flags;
549
550 DP(NETIF_MSG_RX_STATUS, "CQE type %x err %x status %x"
551 " queue %x vlan %x len %u\n", CQE_TYPE(cqe_fp_flags),
552 cqe_fp_flags, cqe->fast_path_cqe.status_flags,
553 le32_to_cpu(cqe->fast_path_cqe.rss_hash_result),
554 le16_to_cpu(cqe->fast_path_cqe.vlan_tag),
555 le16_to_cpu(cqe->fast_path_cqe.pkt_len));
556
557 /* is this a slowpath msg? */
558 if (unlikely(CQE_TYPE(cqe_fp_flags))) {
559 bnx2x_sp_event(fp, cqe);
560 goto next_cqe;
561
562 /* this is an rx packet */
563 } else {
564 rx_buf = &fp->rx_buf_ring[bd_cons];
565 skb = rx_buf->skb;
566 prefetch(skb);
567 len = le16_to_cpu(cqe->fast_path_cqe.pkt_len);
568 pad = cqe->fast_path_cqe.placement_offset;
569
570 /* - If CQE is marked both TPA_START and TPA_END it is
571 * a non-TPA CQE.
572 * - FP CQE will always have either TPA_START or/and
573 * TPA_STOP flags set.
574 */
575 if ((!fp->disable_tpa) &&
576 (TPA_TYPE(cqe_fp_flags) !=
577 (TPA_TYPE_START | TPA_TYPE_END))) {
578 u16 queue = cqe->fast_path_cqe.queue_index;
579
580 if (TPA_TYPE(cqe_fp_flags) == TPA_TYPE_START) {
581 DP(NETIF_MSG_RX_STATUS,
582 "calling tpa_start on queue %d\n",
583 queue);
584
585 bnx2x_tpa_start(fp, queue, skb,
586 bd_cons, bd_prod);
587
588 /* Set Toeplitz hash for an LRO skb */
589 bnx2x_set_skb_rxhash(bp, cqe, skb);
590
591 goto next_rx;
592 } else { /* TPA_STOP */
593 DP(NETIF_MSG_RX_STATUS,
594 "calling tpa_stop on queue %d\n",
595 queue);
596
597 if (!BNX2X_RX_SUM_FIX(cqe))
598 BNX2X_ERR("STOP on none TCP "
599 "data\n");
600
601 /* This is a size of the linear data
602 on this skb */
603 len = le16_to_cpu(cqe->fast_path_cqe.
604 len_on_bd);
605 bnx2x_tpa_stop(bp, fp, queue, pad,
606 len, cqe, comp_ring_cons);
607 #ifdef BNX2X_STOP_ON_ERROR
608 if (bp->panic)
609 return 0;
610 #endif
611
612 bnx2x_update_sge_prod(fp,
613 &cqe->fast_path_cqe);
614 goto next_cqe;
615 }
616 }
617
618 dma_sync_single_for_device(&bp->pdev->dev,
619 dma_unmap_addr(rx_buf, mapping),
620 pad + RX_COPY_THRESH,
621 DMA_FROM_DEVICE);
622 prefetch(((char *)(skb)) + L1_CACHE_BYTES);
623
624 /* is this an error packet? */
625 if (unlikely(cqe_fp_flags & ETH_RX_ERROR_FALGS)) {
626 DP(NETIF_MSG_RX_ERR,
627 "ERROR flags %x rx packet %u\n",
628 cqe_fp_flags, sw_comp_cons);
629 fp->eth_q_stats.rx_err_discard_pkt++;
630 goto reuse_rx;
631 }
632
633 /* Since we don't have a jumbo ring
634 * copy small packets if mtu > 1500
635 */
636 if ((bp->dev->mtu > ETH_MAX_PACKET_SIZE) &&
637 (len <= RX_COPY_THRESH)) {
638 struct sk_buff *new_skb;
639
640 new_skb = netdev_alloc_skb(bp->dev,
641 len + pad);
642 if (new_skb == NULL) {
643 DP(NETIF_MSG_RX_ERR,
644 "ERROR packet dropped "
645 "because of alloc failure\n");
646 fp->eth_q_stats.rx_skb_alloc_failed++;
647 goto reuse_rx;
648 }
649
650 /* aligned copy */
651 skb_copy_from_linear_data_offset(skb, pad,
652 new_skb->data + pad, len);
653 skb_reserve(new_skb, pad);
654 skb_put(new_skb, len);
655
656 bnx2x_reuse_rx_skb(fp, bd_cons, bd_prod);
657
658 skb = new_skb;
659
660 } else
661 if (likely(bnx2x_alloc_rx_skb(bp, fp, bd_prod) == 0)) {
662 dma_unmap_single(&bp->pdev->dev,
663 dma_unmap_addr(rx_buf, mapping),
664 fp->rx_buf_size,
665 DMA_FROM_DEVICE);
666 skb_reserve(skb, pad);
667 skb_put(skb, len);
668
669 } else {
670 DP(NETIF_MSG_RX_ERR,
671 "ERROR packet dropped because "
672 "of alloc failure\n");
673 fp->eth_q_stats.rx_skb_alloc_failed++;
674 reuse_rx:
675 bnx2x_reuse_rx_skb(fp, bd_cons, bd_prod);
676 goto next_rx;
677 }
678
679 skb->protocol = eth_type_trans(skb, bp->dev);
680
681 /* Set Toeplitz hash for a none-LRO skb */
682 bnx2x_set_skb_rxhash(bp, cqe, skb);
683
684 skb_checksum_none_assert(skb);
685
686 if (bp->dev->features & NETIF_F_RXCSUM) {
687 if (likely(BNX2X_RX_CSUM_OK(cqe)))
688 skb->ip_summed = CHECKSUM_UNNECESSARY;
689 else
690 fp->eth_q_stats.hw_csum_err++;
691 }
692 }
693
694 skb_record_rx_queue(skb, fp->index);
695
696 if (le16_to_cpu(cqe->fast_path_cqe.pars_flags.flags) &
697 PARSING_FLAGS_VLAN)
698 __vlan_hwaccel_put_tag(skb,
699 le16_to_cpu(cqe->fast_path_cqe.vlan_tag));
700 napi_gro_receive(&fp->napi, skb);
701
702
703 next_rx:
704 rx_buf->skb = NULL;
705
706 bd_cons = NEXT_RX_IDX(bd_cons);
707 bd_prod = NEXT_RX_IDX(bd_prod);
708 bd_prod_fw = NEXT_RX_IDX(bd_prod_fw);
709 rx_pkt++;
710 next_cqe:
711 sw_comp_prod = NEXT_RCQ_IDX(sw_comp_prod);
712 sw_comp_cons = NEXT_RCQ_IDX(sw_comp_cons);
713
714 if (rx_pkt == budget)
715 break;
716 } /* while */
717
718 fp->rx_bd_cons = bd_cons;
719 fp->rx_bd_prod = bd_prod_fw;
720 fp->rx_comp_cons = sw_comp_cons;
721 fp->rx_comp_prod = sw_comp_prod;
722
723 /* Update producers */
724 bnx2x_update_rx_prod(bp, fp, bd_prod_fw, sw_comp_prod,
725 fp->rx_sge_prod);
726
727 fp->rx_pkt += rx_pkt;
728 fp->rx_calls++;
729
730 return rx_pkt;
731 }
732
733 static irqreturn_t bnx2x_msix_fp_int(int irq, void *fp_cookie)
734 {
735 struct bnx2x_fastpath *fp = fp_cookie;
736 struct bnx2x *bp = fp->bp;
737
738 /* Return here if interrupt is disabled */
739 if (unlikely(atomic_read(&bp->intr_sem) != 0)) {
740 DP(NETIF_MSG_INTR, "called but intr_sem not 0, returning\n");
741 return IRQ_HANDLED;
742 }
743
744 DP(BNX2X_MSG_FP, "got an MSI-X interrupt on IDX:SB "
745 "[fp %d fw_sd %d igusb %d]\n",
746 fp->index, fp->fw_sb_id, fp->igu_sb_id);
747 bnx2x_ack_sb(bp, fp->igu_sb_id, USTORM_ID, 0, IGU_INT_DISABLE, 0);
748
749 #ifdef BNX2X_STOP_ON_ERROR
750 if (unlikely(bp->panic))
751 return IRQ_HANDLED;
752 #endif
753
754 /* Handle Rx and Tx according to MSI-X vector */
755 prefetch(fp->rx_cons_sb);
756 prefetch(fp->tx_cons_sb);
757 prefetch(&fp->sb_running_index[SM_RX_ID]);
758 napi_schedule(&bnx2x_fp(bp, fp->index, napi));
759
760 return IRQ_HANDLED;
761 }
762
763 /* HW Lock for shared dual port PHYs */
764 void bnx2x_acquire_phy_lock(struct bnx2x *bp)
765 {
766 mutex_lock(&bp->port.phy_mutex);
767
768 if (bp->port.need_hw_lock)
769 bnx2x_acquire_hw_lock(bp, HW_LOCK_RESOURCE_MDIO);
770 }
771
772 void bnx2x_release_phy_lock(struct bnx2x *bp)
773 {
774 if (bp->port.need_hw_lock)
775 bnx2x_release_hw_lock(bp, HW_LOCK_RESOURCE_MDIO);
776
777 mutex_unlock(&bp->port.phy_mutex);
778 }
779
780 /* calculates MF speed according to current linespeed and MF configuration */
781 u16 bnx2x_get_mf_speed(struct bnx2x *bp)
782 {
783 u16 line_speed = bp->link_vars.line_speed;
784 if (IS_MF(bp)) {
785 u16 maxCfg = bnx2x_extract_max_cfg(bp,
786 bp->mf_config[BP_VN(bp)]);
787
788 /* Calculate the current MAX line speed limit for the MF
789 * devices
790 */
791 if (IS_MF_SI(bp))
792 line_speed = (line_speed * maxCfg) / 100;
793 else { /* SD mode */
794 u16 vn_max_rate = maxCfg * 100;
795
796 if (vn_max_rate < line_speed)
797 line_speed = vn_max_rate;
798 }
799 }
800
801 return line_speed;
802 }
803
804 /**
805 * bnx2x_fill_report_data - fill link report data to report
806 *
807 * @bp: driver handle
808 * @data: link state to update
809 *
810 * It uses a none-atomic bit operations because is called under the mutex.
811 */
812 static inline void bnx2x_fill_report_data(struct bnx2x *bp,
813 struct bnx2x_link_report_data *data)
814 {
815 u16 line_speed = bnx2x_get_mf_speed(bp);
816
817 memset(data, 0, sizeof(*data));
818
819 /* Fill the report data: efective line speed */
820 data->line_speed = line_speed;
821
822 /* Link is down */
823 if (!bp->link_vars.link_up || (bp->flags & MF_FUNC_DIS))
824 __set_bit(BNX2X_LINK_REPORT_LINK_DOWN,
825 &data->link_report_flags);
826
827 /* Full DUPLEX */
828 if (bp->link_vars.duplex == DUPLEX_FULL)
829 __set_bit(BNX2X_LINK_REPORT_FD, &data->link_report_flags);
830
831 /* Rx Flow Control is ON */
832 if (bp->link_vars.flow_ctrl & BNX2X_FLOW_CTRL_RX)
833 __set_bit(BNX2X_LINK_REPORT_RX_FC_ON, &data->link_report_flags);
834
835 /* Tx Flow Control is ON */
836 if (bp->link_vars.flow_ctrl & BNX2X_FLOW_CTRL_TX)
837 __set_bit(BNX2X_LINK_REPORT_TX_FC_ON, &data->link_report_flags);
838 }
839
840 /**
841 * bnx2x_link_report - report link status to OS.
842 *
843 * @bp: driver handle
844 *
845 * Calls the __bnx2x_link_report() under the same locking scheme
846 * as a link/PHY state managing code to ensure a consistent link
847 * reporting.
848 */
849
850 void bnx2x_link_report(struct bnx2x *bp)
851 {
852 bnx2x_acquire_phy_lock(bp);
853 __bnx2x_link_report(bp);
854 bnx2x_release_phy_lock(bp);
855 }
856
857 /**
858 * __bnx2x_link_report - report link status to OS.
859 *
860 * @bp: driver handle
861 *
862 * None atomic inmlementation.
863 * Should be called under the phy_lock.
864 */
865 void __bnx2x_link_report(struct bnx2x *bp)
866 {
867 struct bnx2x_link_report_data cur_data;
868
869 /* reread mf_cfg */
870 if (!CHIP_IS_E1(bp))
871 bnx2x_read_mf_cfg(bp);
872
873 /* Read the current link report info */
874 bnx2x_fill_report_data(bp, &cur_data);
875
876 /* Don't report link down or exactly the same link status twice */
877 if (!memcmp(&cur_data, &bp->last_reported_link, sizeof(cur_data)) ||
878 (test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
879 &bp->last_reported_link.link_report_flags) &&
880 test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
881 &cur_data.link_report_flags)))
882 return;
883
884 bp->link_cnt++;
885
886 /* We are going to report a new link parameters now -
887 * remember the current data for the next time.
888 */
889 memcpy(&bp->last_reported_link, &cur_data, sizeof(cur_data));
890
891 if (test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
892 &cur_data.link_report_flags)) {
893 netif_carrier_off(bp->dev);
894 netdev_err(bp->dev, "NIC Link is Down\n");
895 return;
896 } else {
897 netif_carrier_on(bp->dev);
898 netdev_info(bp->dev, "NIC Link is Up, ");
899 pr_cont("%d Mbps ", cur_data.line_speed);
900
901 if (test_and_clear_bit(BNX2X_LINK_REPORT_FD,
902 &cur_data.link_report_flags))
903 pr_cont("full duplex");
904 else
905 pr_cont("half duplex");
906
907 /* Handle the FC at the end so that only these flags would be
908 * possibly set. This way we may easily check if there is no FC
909 * enabled.
910 */
911 if (cur_data.link_report_flags) {
912 if (test_bit(BNX2X_LINK_REPORT_RX_FC_ON,
913 &cur_data.link_report_flags)) {
914 pr_cont(", receive ");
915 if (test_bit(BNX2X_LINK_REPORT_TX_FC_ON,
916 &cur_data.link_report_flags))
917 pr_cont("& transmit ");
918 } else {
919 pr_cont(", transmit ");
920 }
921 pr_cont("flow control ON");
922 }
923 pr_cont("\n");
924 }
925 }
926
927 void bnx2x_init_rx_rings(struct bnx2x *bp)
928 {
929 int func = BP_FUNC(bp);
930 int max_agg_queues = CHIP_IS_E1(bp) ? ETH_MAX_AGGREGATION_QUEUES_E1 :
931 ETH_MAX_AGGREGATION_QUEUES_E1H;
932 u16 ring_prod;
933 int i, j;
934
935 /* Allocate TPA resources */
936 for_each_rx_queue(bp, j) {
937 struct bnx2x_fastpath *fp = &bp->fp[j];
938
939 DP(NETIF_MSG_IFUP,
940 "mtu %d rx_buf_size %d\n", bp->dev->mtu, fp->rx_buf_size);
941
942 if (!fp->disable_tpa) {
943 /* Fill the per-aggregation pool */
944 for (i = 0; i < max_agg_queues; i++) {
945 fp->tpa_pool[i].skb =
946 netdev_alloc_skb(bp->dev, fp->rx_buf_size);
947 if (!fp->tpa_pool[i].skb) {
948 BNX2X_ERR("Failed to allocate TPA "
949 "skb pool for queue[%d] - "
950 "disabling TPA on this "
951 "queue!\n", j);
952 bnx2x_free_tpa_pool(bp, fp, i);
953 fp->disable_tpa = 1;
954 break;
955 }
956 dma_unmap_addr_set((struct sw_rx_bd *)
957 &bp->fp->tpa_pool[i],
958 mapping, 0);
959 fp->tpa_state[i] = BNX2X_TPA_STOP;
960 }
961
962 /* "next page" elements initialization */
963 bnx2x_set_next_page_sgl(fp);
964
965 /* set SGEs bit mask */
966 bnx2x_init_sge_ring_bit_mask(fp);
967
968 /* Allocate SGEs and initialize the ring elements */
969 for (i = 0, ring_prod = 0;
970 i < MAX_RX_SGE_CNT*NUM_RX_SGE_PAGES; i++) {
971
972 if (bnx2x_alloc_rx_sge(bp, fp, ring_prod) < 0) {
973 BNX2X_ERR("was only able to allocate "
974 "%d rx sges\n", i);
975 BNX2X_ERR("disabling TPA for"
976 " queue[%d]\n", j);
977 /* Cleanup already allocated elements */
978 bnx2x_free_rx_sge_range(bp,
979 fp, ring_prod);
980 bnx2x_free_tpa_pool(bp,
981 fp, max_agg_queues);
982 fp->disable_tpa = 1;
983 ring_prod = 0;
984 break;
985 }
986 ring_prod = NEXT_SGE_IDX(ring_prod);
987 }
988
989 fp->rx_sge_prod = ring_prod;
990 }
991 }
992
993 for_each_rx_queue(bp, j) {
994 struct bnx2x_fastpath *fp = &bp->fp[j];
995
996 fp->rx_bd_cons = 0;
997
998 /* Activate BD ring */
999 /* Warning!
1000 * this will generate an interrupt (to the TSTORM)
1001 * must only be done after chip is initialized
1002 */
1003 bnx2x_update_rx_prod(bp, fp, fp->rx_bd_prod, fp->rx_comp_prod,
1004 fp->rx_sge_prod);
1005
1006 if (j != 0)
1007 continue;
1008
1009 if (!CHIP_IS_E2(bp)) {
1010 REG_WR(bp, BAR_USTRORM_INTMEM +
1011 USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func),
1012 U64_LO(fp->rx_comp_mapping));
1013 REG_WR(bp, BAR_USTRORM_INTMEM +
1014 USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func) + 4,
1015 U64_HI(fp->rx_comp_mapping));
1016 }
1017 }
1018 }
1019
1020 static void bnx2x_free_tx_skbs(struct bnx2x *bp)
1021 {
1022 int i;
1023
1024 for_each_tx_queue(bp, i) {
1025 struct bnx2x_fastpath *fp = &bp->fp[i];
1026
1027 u16 bd_cons = fp->tx_bd_cons;
1028 u16 sw_prod = fp->tx_pkt_prod;
1029 u16 sw_cons = fp->tx_pkt_cons;
1030
1031 while (sw_cons != sw_prod) {
1032 bd_cons = bnx2x_free_tx_pkt(bp, fp, TX_BD(sw_cons));
1033 sw_cons++;
1034 }
1035 }
1036 }
1037
1038 static void bnx2x_free_rx_bds(struct bnx2x_fastpath *fp)
1039 {
1040 struct bnx2x *bp = fp->bp;
1041 int i;
1042
1043 /* ring wasn't allocated */
1044 if (fp->rx_buf_ring == NULL)
1045 return;
1046
1047 for (i = 0; i < NUM_RX_BD; i++) {
1048 struct sw_rx_bd *rx_buf = &fp->rx_buf_ring[i];
1049 struct sk_buff *skb = rx_buf->skb;
1050
1051 if (skb == NULL)
1052 continue;
1053
1054 dma_unmap_single(&bp->pdev->dev,
1055 dma_unmap_addr(rx_buf, mapping),
1056 fp->rx_buf_size, DMA_FROM_DEVICE);
1057
1058 rx_buf->skb = NULL;
1059 dev_kfree_skb(skb);
1060 }
1061 }
1062
1063 static void bnx2x_free_rx_skbs(struct bnx2x *bp)
1064 {
1065 int j;
1066
1067 for_each_rx_queue(bp, j) {
1068 struct bnx2x_fastpath *fp = &bp->fp[j];
1069
1070 bnx2x_free_rx_bds(fp);
1071
1072 if (!fp->disable_tpa)
1073 bnx2x_free_tpa_pool(bp, fp, CHIP_IS_E1(bp) ?
1074 ETH_MAX_AGGREGATION_QUEUES_E1 :
1075 ETH_MAX_AGGREGATION_QUEUES_E1H);
1076 }
1077 }
1078
1079 void bnx2x_free_skbs(struct bnx2x *bp)
1080 {
1081 bnx2x_free_tx_skbs(bp);
1082 bnx2x_free_rx_skbs(bp);
1083 }
1084
1085 void bnx2x_update_max_mf_config(struct bnx2x *bp, u32 value)
1086 {
1087 /* load old values */
1088 u32 mf_cfg = bp->mf_config[BP_VN(bp)];
1089
1090 if (value != bnx2x_extract_max_cfg(bp, mf_cfg)) {
1091 /* leave all but MAX value */
1092 mf_cfg &= ~FUNC_MF_CFG_MAX_BW_MASK;
1093
1094 /* set new MAX value */
1095 mf_cfg |= (value << FUNC_MF_CFG_MAX_BW_SHIFT)
1096 & FUNC_MF_CFG_MAX_BW_MASK;
1097
1098 bnx2x_fw_command(bp, DRV_MSG_CODE_SET_MF_BW, mf_cfg);
1099 }
1100 }
1101
1102 static void bnx2x_free_msix_irqs(struct bnx2x *bp)
1103 {
1104 int i, offset = 1;
1105
1106 free_irq(bp->msix_table[0].vector, bp->dev);
1107 DP(NETIF_MSG_IFDOWN, "released sp irq (%d)\n",
1108 bp->msix_table[0].vector);
1109
1110 #ifdef BCM_CNIC
1111 offset++;
1112 #endif
1113 for_each_eth_queue(bp, i) {
1114 DP(NETIF_MSG_IFDOWN, "about to release fp #%d->%d irq "
1115 "state %x\n", i, bp->msix_table[i + offset].vector,
1116 bnx2x_fp(bp, i, state));
1117
1118 free_irq(bp->msix_table[i + offset].vector, &bp->fp[i]);
1119 }
1120 }
1121
1122 void bnx2x_free_irq(struct bnx2x *bp)
1123 {
1124 if (bp->flags & USING_MSIX_FLAG)
1125 bnx2x_free_msix_irqs(bp);
1126 else if (bp->flags & USING_MSI_FLAG)
1127 free_irq(bp->pdev->irq, bp->dev);
1128 else
1129 free_irq(bp->pdev->irq, bp->dev);
1130 }
1131
1132 int bnx2x_enable_msix(struct bnx2x *bp)
1133 {
1134 int msix_vec = 0, i, rc, req_cnt;
1135
1136 bp->msix_table[msix_vec].entry = msix_vec;
1137 DP(NETIF_MSG_IFUP, "msix_table[0].entry = %d (slowpath)\n",
1138 bp->msix_table[0].entry);
1139 msix_vec++;
1140
1141 #ifdef BCM_CNIC
1142 bp->msix_table[msix_vec].entry = msix_vec;
1143 DP(NETIF_MSG_IFUP, "msix_table[%d].entry = %d (CNIC)\n",
1144 bp->msix_table[msix_vec].entry, bp->msix_table[msix_vec].entry);
1145 msix_vec++;
1146 #endif
1147 for_each_eth_queue(bp, i) {
1148 bp->msix_table[msix_vec].entry = msix_vec;
1149 DP(NETIF_MSG_IFUP, "msix_table[%d].entry = %d "
1150 "(fastpath #%u)\n", msix_vec, msix_vec, i);
1151 msix_vec++;
1152 }
1153
1154 req_cnt = BNX2X_NUM_ETH_QUEUES(bp) + CNIC_CONTEXT_USE + 1;
1155
1156 rc = pci_enable_msix(bp->pdev, &bp->msix_table[0], req_cnt);
1157
1158 /*
1159 * reconfigure number of tx/rx queues according to available
1160 * MSI-X vectors
1161 */
1162 if (rc >= BNX2X_MIN_MSIX_VEC_CNT) {
1163 /* how less vectors we will have? */
1164 int diff = req_cnt - rc;
1165
1166 DP(NETIF_MSG_IFUP,
1167 "Trying to use less MSI-X vectors: %d\n", rc);
1168
1169 rc = pci_enable_msix(bp->pdev, &bp->msix_table[0], rc);
1170
1171 if (rc) {
1172 DP(NETIF_MSG_IFUP,
1173 "MSI-X is not attainable rc %d\n", rc);
1174 return rc;
1175 }
1176 /*
1177 * decrease number of queues by number of unallocated entries
1178 */
1179 bp->num_queues -= diff;
1180
1181 DP(NETIF_MSG_IFUP, "New queue configuration set: %d\n",
1182 bp->num_queues);
1183 } else if (rc) {
1184 /* fall to INTx if not enough memory */
1185 if (rc == -ENOMEM)
1186 bp->flags |= DISABLE_MSI_FLAG;
1187 DP(NETIF_MSG_IFUP, "MSI-X is not attainable rc %d\n", rc);
1188 return rc;
1189 }
1190
1191 bp->flags |= USING_MSIX_FLAG;
1192
1193 return 0;
1194 }
1195
1196 static int bnx2x_req_msix_irqs(struct bnx2x *bp)
1197 {
1198 int i, rc, offset = 1;
1199
1200 rc = request_irq(bp->msix_table[0].vector, bnx2x_msix_sp_int, 0,
1201 bp->dev->name, bp->dev);
1202 if (rc) {
1203 BNX2X_ERR("request sp irq failed\n");
1204 return -EBUSY;
1205 }
1206
1207 #ifdef BCM_CNIC
1208 offset++;
1209 #endif
1210 for_each_eth_queue(bp, i) {
1211 struct bnx2x_fastpath *fp = &bp->fp[i];
1212 snprintf(fp->name, sizeof(fp->name), "%s-fp-%d",
1213 bp->dev->name, i);
1214
1215 rc = request_irq(bp->msix_table[offset].vector,
1216 bnx2x_msix_fp_int, 0, fp->name, fp);
1217 if (rc) {
1218 BNX2X_ERR("request fp #%d irq failed rc %d\n", i, rc);
1219 bnx2x_free_msix_irqs(bp);
1220 return -EBUSY;
1221 }
1222
1223 offset++;
1224 fp->state = BNX2X_FP_STATE_IRQ;
1225 }
1226
1227 i = BNX2X_NUM_ETH_QUEUES(bp);
1228 offset = 1 + CNIC_CONTEXT_USE;
1229 netdev_info(bp->dev, "using MSI-X IRQs: sp %d fp[%d] %d"
1230 " ... fp[%d] %d\n",
1231 bp->msix_table[0].vector,
1232 0, bp->msix_table[offset].vector,
1233 i - 1, bp->msix_table[offset + i - 1].vector);
1234
1235 return 0;
1236 }
1237
1238 int bnx2x_enable_msi(struct bnx2x *bp)
1239 {
1240 int rc;
1241
1242 rc = pci_enable_msi(bp->pdev);
1243 if (rc) {
1244 DP(NETIF_MSG_IFUP, "MSI is not attainable\n");
1245 return -1;
1246 }
1247 bp->flags |= USING_MSI_FLAG;
1248
1249 return 0;
1250 }
1251
1252 static int bnx2x_req_irq(struct bnx2x *bp)
1253 {
1254 unsigned long flags;
1255 int rc;
1256
1257 if (bp->flags & USING_MSI_FLAG)
1258 flags = 0;
1259 else
1260 flags = IRQF_SHARED;
1261
1262 rc = request_irq(bp->pdev->irq, bnx2x_interrupt, flags,
1263 bp->dev->name, bp->dev);
1264 if (!rc)
1265 bnx2x_fp(bp, 0, state) = BNX2X_FP_STATE_IRQ;
1266
1267 return rc;
1268 }
1269
1270 static void bnx2x_napi_enable(struct bnx2x *bp)
1271 {
1272 int i;
1273
1274 for_each_napi_queue(bp, i)
1275 napi_enable(&bnx2x_fp(bp, i, napi));
1276 }
1277
1278 static void bnx2x_napi_disable(struct bnx2x *bp)
1279 {
1280 int i;
1281
1282 for_each_napi_queue(bp, i)
1283 napi_disable(&bnx2x_fp(bp, i, napi));
1284 }
1285
1286 void bnx2x_netif_start(struct bnx2x *bp)
1287 {
1288 int intr_sem;
1289
1290 intr_sem = atomic_dec_and_test(&bp->intr_sem);
1291 smp_wmb(); /* Ensure that bp->intr_sem update is SMP-safe */
1292
1293 if (intr_sem) {
1294 if (netif_running(bp->dev)) {
1295 bnx2x_napi_enable(bp);
1296 bnx2x_int_enable(bp);
1297 if (bp->state == BNX2X_STATE_OPEN)
1298 netif_tx_wake_all_queues(bp->dev);
1299 }
1300 }
1301 }
1302
1303 void bnx2x_netif_stop(struct bnx2x *bp, int disable_hw)
1304 {
1305 bnx2x_int_disable_sync(bp, disable_hw);
1306 bnx2x_napi_disable(bp);
1307 netif_tx_disable(bp->dev);
1308 }
1309
1310 u16 bnx2x_select_queue(struct net_device *dev, struct sk_buff *skb)
1311 {
1312 #ifdef BCM_CNIC
1313 struct bnx2x *bp = netdev_priv(dev);
1314 if (NO_FCOE(bp))
1315 return skb_tx_hash(dev, skb);
1316 else {
1317 struct ethhdr *hdr = (struct ethhdr *)skb->data;
1318 u16 ether_type = ntohs(hdr->h_proto);
1319
1320 /* Skip VLAN tag if present */
1321 if (ether_type == ETH_P_8021Q) {
1322 struct vlan_ethhdr *vhdr =
1323 (struct vlan_ethhdr *)skb->data;
1324
1325 ether_type = ntohs(vhdr->h_vlan_encapsulated_proto);
1326 }
1327
1328 /* If ethertype is FCoE or FIP - use FCoE ring */
1329 if ((ether_type == ETH_P_FCOE) || (ether_type == ETH_P_FIP))
1330 return bnx2x_fcoe(bp, index);
1331 }
1332 #endif
1333 /* Select a none-FCoE queue: if FCoE is enabled, exclude FCoE L2 ring
1334 */
1335 return __skb_tx_hash(dev, skb,
1336 dev->real_num_tx_queues - FCOE_CONTEXT_USE);
1337 }
1338
1339 void bnx2x_set_num_queues(struct bnx2x *bp)
1340 {
1341 switch (bp->multi_mode) {
1342 case ETH_RSS_MODE_DISABLED:
1343 bp->num_queues = 1;
1344 break;
1345 case ETH_RSS_MODE_REGULAR:
1346 bp->num_queues = bnx2x_calc_num_queues(bp);
1347 break;
1348
1349 default:
1350 bp->num_queues = 1;
1351 break;
1352 }
1353
1354 /* Add special queues */
1355 bp->num_queues += NONE_ETH_CONTEXT_USE;
1356 }
1357
1358 #ifdef BCM_CNIC
1359 static inline void bnx2x_set_fcoe_eth_macs(struct bnx2x *bp)
1360 {
1361 if (!NO_FCOE(bp)) {
1362 if (!IS_MF_SD(bp))
1363 bnx2x_set_fip_eth_mac_addr(bp, 1);
1364 bnx2x_set_all_enode_macs(bp, 1);
1365 bp->flags |= FCOE_MACS_SET;
1366 }
1367 }
1368 #endif
1369
1370 static void bnx2x_release_firmware(struct bnx2x *bp)
1371 {
1372 kfree(bp->init_ops_offsets);
1373 kfree(bp->init_ops);
1374 kfree(bp->init_data);
1375 release_firmware(bp->firmware);
1376 }
1377
1378 static inline int bnx2x_set_real_num_queues(struct bnx2x *bp)
1379 {
1380 int rc, num = bp->num_queues;
1381
1382 #ifdef BCM_CNIC
1383 if (NO_FCOE(bp))
1384 num -= FCOE_CONTEXT_USE;
1385
1386 #endif
1387 netif_set_real_num_tx_queues(bp->dev, num);
1388 rc = netif_set_real_num_rx_queues(bp->dev, num);
1389 return rc;
1390 }
1391
1392 static inline void bnx2x_set_rx_buf_size(struct bnx2x *bp)
1393 {
1394 int i;
1395
1396 for_each_queue(bp, i) {
1397 struct bnx2x_fastpath *fp = &bp->fp[i];
1398
1399 /* Always use a mini-jumbo MTU for the FCoE L2 ring */
1400 if (IS_FCOE_IDX(i))
1401 /*
1402 * Although there are no IP frames expected to arrive to
1403 * this ring we still want to add an
1404 * IP_HEADER_ALIGNMENT_PADDING to prevent a buffer
1405 * overrun attack.
1406 */
1407 fp->rx_buf_size =
1408 BNX2X_FCOE_MINI_JUMBO_MTU + ETH_OVREHEAD +
1409 BNX2X_RX_ALIGN + IP_HEADER_ALIGNMENT_PADDING;
1410 else
1411 fp->rx_buf_size =
1412 bp->dev->mtu + ETH_OVREHEAD + BNX2X_RX_ALIGN +
1413 IP_HEADER_ALIGNMENT_PADDING;
1414 }
1415 }
1416
1417 /* must be called with rtnl_lock */
1418 int bnx2x_nic_load(struct bnx2x *bp, int load_mode)
1419 {
1420 u32 load_code;
1421 int i, rc;
1422
1423 /* Set init arrays */
1424 rc = bnx2x_init_firmware(bp);
1425 if (rc) {
1426 BNX2X_ERR("Error loading firmware\n");
1427 return rc;
1428 }
1429
1430 #ifdef BNX2X_STOP_ON_ERROR
1431 if (unlikely(bp->panic))
1432 return -EPERM;
1433 #endif
1434
1435 bp->state = BNX2X_STATE_OPENING_WAIT4_LOAD;
1436
1437 /* Set the initial link reported state to link down */
1438 bnx2x_acquire_phy_lock(bp);
1439 memset(&bp->last_reported_link, 0, sizeof(bp->last_reported_link));
1440 __set_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1441 &bp->last_reported_link.link_report_flags);
1442 bnx2x_release_phy_lock(bp);
1443
1444 /* must be called before memory allocation and HW init */
1445 bnx2x_ilt_set_info(bp);
1446
1447 /* zero fastpath structures preserving invariants like napi which are
1448 * allocated only once
1449 */
1450 for_each_queue(bp, i)
1451 bnx2x_bz_fp(bp, i);
1452
1453 /* Set the receive queues buffer size */
1454 bnx2x_set_rx_buf_size(bp);
1455
1456 for_each_queue(bp, i)
1457 bnx2x_fp(bp, i, disable_tpa) =
1458 ((bp->flags & TPA_ENABLE_FLAG) == 0);
1459
1460 #ifdef BCM_CNIC
1461 /* We don't want TPA on FCoE L2 ring */
1462 bnx2x_fcoe(bp, disable_tpa) = 1;
1463 #endif
1464
1465 if (bnx2x_alloc_mem(bp))
1466 return -ENOMEM;
1467
1468 /* As long as bnx2x_alloc_mem() may possibly update
1469 * bp->num_queues, bnx2x_set_real_num_queues() should always
1470 * come after it.
1471 */
1472 rc = bnx2x_set_real_num_queues(bp);
1473 if (rc) {
1474 BNX2X_ERR("Unable to set real_num_queues\n");
1475 goto load_error0;
1476 }
1477
1478 bnx2x_napi_enable(bp);
1479
1480 /* Send LOAD_REQUEST command to MCP
1481 Returns the type of LOAD command:
1482 if it is the first port to be initialized
1483 common blocks should be initialized, otherwise - not
1484 */
1485 if (!BP_NOMCP(bp)) {
1486 load_code = bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_REQ, 0);
1487 if (!load_code) {
1488 BNX2X_ERR("MCP response failure, aborting\n");
1489 rc = -EBUSY;
1490 goto load_error1;
1491 }
1492 if (load_code == FW_MSG_CODE_DRV_LOAD_REFUSED) {
1493 rc = -EBUSY; /* other port in diagnostic mode */
1494 goto load_error1;
1495 }
1496
1497 } else {
1498 int path = BP_PATH(bp);
1499 int port = BP_PORT(bp);
1500
1501 DP(NETIF_MSG_IFUP, "NO MCP - load counts[%d] %d, %d, %d\n",
1502 path, load_count[path][0], load_count[path][1],
1503 load_count[path][2]);
1504 load_count[path][0]++;
1505 load_count[path][1 + port]++;
1506 DP(NETIF_MSG_IFUP, "NO MCP - new load counts[%d] %d, %d, %d\n",
1507 path, load_count[path][0], load_count[path][1],
1508 load_count[path][2]);
1509 if (load_count[path][0] == 1)
1510 load_code = FW_MSG_CODE_DRV_LOAD_COMMON;
1511 else if (load_count[path][1 + port] == 1)
1512 load_code = FW_MSG_CODE_DRV_LOAD_PORT;
1513 else
1514 load_code = FW_MSG_CODE_DRV_LOAD_FUNCTION;
1515 }
1516
1517 if ((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
1518 (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP) ||
1519 (load_code == FW_MSG_CODE_DRV_LOAD_PORT))
1520 bp->port.pmf = 1;
1521 else
1522 bp->port.pmf = 0;
1523 DP(NETIF_MSG_LINK, "pmf %d\n", bp->port.pmf);
1524
1525 /* Initialize HW */
1526 rc = bnx2x_init_hw(bp, load_code);
1527 if (rc) {
1528 BNX2X_ERR("HW init failed, aborting\n");
1529 bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
1530 goto load_error2;
1531 }
1532
1533 /* Connect to IRQs */
1534 rc = bnx2x_setup_irqs(bp);
1535 if (rc) {
1536 bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
1537 goto load_error2;
1538 }
1539
1540 /* Setup NIC internals and enable interrupts */
1541 bnx2x_nic_init(bp, load_code);
1542
1543 if (((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
1544 (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP)) &&
1545 (bp->common.shmem2_base))
1546 SHMEM2_WR(bp, dcc_support,
1547 (SHMEM_DCC_SUPPORT_DISABLE_ENABLE_PF_TLV |
1548 SHMEM_DCC_SUPPORT_BANDWIDTH_ALLOCATION_TLV));
1549
1550 /* Send LOAD_DONE command to MCP */
1551 if (!BP_NOMCP(bp)) {
1552 load_code = bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
1553 if (!load_code) {
1554 BNX2X_ERR("MCP response failure, aborting\n");
1555 rc = -EBUSY;
1556 goto load_error3;
1557 }
1558 }
1559
1560 bnx2x_dcbx_init(bp);
1561
1562 bp->state = BNX2X_STATE_OPENING_WAIT4_PORT;
1563
1564 rc = bnx2x_func_start(bp);
1565 if (rc) {
1566 BNX2X_ERR("Function start failed!\n");
1567 #ifndef BNX2X_STOP_ON_ERROR
1568 goto load_error3;
1569 #else
1570 bp->panic = 1;
1571 return -EBUSY;
1572 #endif
1573 }
1574
1575 rc = bnx2x_setup_client(bp, &bp->fp[0], 1 /* Leading */);
1576 if (rc) {
1577 BNX2X_ERR("Setup leading failed!\n");
1578 #ifndef BNX2X_STOP_ON_ERROR
1579 goto load_error3;
1580 #else
1581 bp->panic = 1;
1582 return -EBUSY;
1583 #endif
1584 }
1585
1586 if (!CHIP_IS_E1(bp) &&
1587 (bp->mf_config[BP_VN(bp)] & FUNC_MF_CFG_FUNC_DISABLED)) {
1588 DP(NETIF_MSG_IFUP, "mf_cfg function disabled\n");
1589 bp->flags |= MF_FUNC_DIS;
1590 }
1591
1592 #ifdef BCM_CNIC
1593 /* Enable Timer scan */
1594 REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + BP_PORT(bp)*4, 1);
1595 #endif
1596
1597 for_each_nondefault_queue(bp, i) {
1598 rc = bnx2x_setup_client(bp, &bp->fp[i], 0);
1599 if (rc)
1600 #ifdef BCM_CNIC
1601 goto load_error4;
1602 #else
1603 goto load_error3;
1604 #endif
1605 }
1606
1607 /* Now when Clients are configured we are ready to work */
1608 bp->state = BNX2X_STATE_OPEN;
1609
1610 #ifdef BCM_CNIC
1611 bnx2x_set_fcoe_eth_macs(bp);
1612 #endif
1613
1614 bnx2x_set_eth_mac(bp, 1);
1615
1616 /* Clear MC configuration */
1617 if (CHIP_IS_E1(bp))
1618 bnx2x_invalidate_e1_mc_list(bp);
1619 else
1620 bnx2x_invalidate_e1h_mc_list(bp);
1621
1622 /* Clear UC lists configuration */
1623 bnx2x_invalidate_uc_list(bp);
1624
1625 if (bp->pending_max) {
1626 bnx2x_update_max_mf_config(bp, bp->pending_max);
1627 bp->pending_max = 0;
1628 }
1629
1630 if (bp->port.pmf)
1631 bnx2x_initial_phy_init(bp, load_mode);
1632
1633 /* Initialize Rx filtering */
1634 bnx2x_set_rx_mode(bp->dev);
1635
1636 /* Start fast path */
1637 switch (load_mode) {
1638 case LOAD_NORMAL:
1639 /* Tx queue should be only reenabled */
1640 netif_tx_wake_all_queues(bp->dev);
1641 /* Initialize the receive filter. */
1642 break;
1643
1644 case LOAD_OPEN:
1645 netif_tx_start_all_queues(bp->dev);
1646 smp_mb__after_clear_bit();
1647 break;
1648
1649 case LOAD_DIAG:
1650 bp->state = BNX2X_STATE_DIAG;
1651 break;
1652
1653 default:
1654 break;
1655 }
1656
1657 if (!bp->port.pmf)
1658 bnx2x__link_status_update(bp);
1659
1660 /* start the timer */
1661 mod_timer(&bp->timer, jiffies + bp->current_interval);
1662
1663 #ifdef BCM_CNIC
1664 bnx2x_setup_cnic_irq_info(bp);
1665 if (bp->state == BNX2X_STATE_OPEN)
1666 bnx2x_cnic_notify(bp, CNIC_CTL_START_CMD);
1667 #endif
1668 bnx2x_inc_load_cnt(bp);
1669
1670 bnx2x_release_firmware(bp);
1671
1672 return 0;
1673
1674 #ifdef BCM_CNIC
1675 load_error4:
1676 /* Disable Timer scan */
1677 REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + BP_PORT(bp)*4, 0);
1678 #endif
1679 load_error3:
1680 bnx2x_int_disable_sync(bp, 1);
1681
1682 /* Free SKBs, SGEs, TPA pool and driver internals */
1683 bnx2x_free_skbs(bp);
1684 for_each_rx_queue(bp, i)
1685 bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
1686
1687 /* Release IRQs */
1688 bnx2x_free_irq(bp);
1689 load_error2:
1690 if (!BP_NOMCP(bp)) {
1691 bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_REQ_WOL_MCP, 0);
1692 bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_DONE, 0);
1693 }
1694
1695 bp->port.pmf = 0;
1696 load_error1:
1697 bnx2x_napi_disable(bp);
1698 load_error0:
1699 bnx2x_free_mem(bp);
1700
1701 bnx2x_release_firmware(bp);
1702
1703 return rc;
1704 }
1705
1706 /* must be called with rtnl_lock */
1707 int bnx2x_nic_unload(struct bnx2x *bp, int unload_mode)
1708 {
1709 int i;
1710
1711 if (bp->state == BNX2X_STATE_CLOSED) {
1712 /* Interface has been removed - nothing to recover */
1713 bp->recovery_state = BNX2X_RECOVERY_DONE;
1714 bp->is_leader = 0;
1715 bnx2x_release_hw_lock(bp, HW_LOCK_RESOURCE_RESERVED_08);
1716 smp_wmb();
1717
1718 return -EINVAL;
1719 }
1720
1721 #ifdef BCM_CNIC
1722 bnx2x_cnic_notify(bp, CNIC_CTL_STOP_CMD);
1723 #endif
1724 bp->state = BNX2X_STATE_CLOSING_WAIT4_HALT;
1725
1726 /* Set "drop all" */
1727 bp->rx_mode = BNX2X_RX_MODE_NONE;
1728 bnx2x_set_storm_rx_mode(bp);
1729
1730 /* Stop Tx */
1731 bnx2x_tx_disable(bp);
1732
1733 del_timer_sync(&bp->timer);
1734
1735 SHMEM_WR(bp, func_mb[BP_FW_MB_IDX(bp)].drv_pulse_mb,
1736 (DRV_PULSE_ALWAYS_ALIVE | bp->fw_drv_pulse_wr_seq));
1737
1738 bnx2x_stats_handle(bp, STATS_EVENT_STOP);
1739
1740 /* Cleanup the chip if needed */
1741 if (unload_mode != UNLOAD_RECOVERY)
1742 bnx2x_chip_cleanup(bp, unload_mode);
1743 else {
1744 /* Disable HW interrupts, NAPI and Tx */
1745 bnx2x_netif_stop(bp, 1);
1746
1747 /* Release IRQs */
1748 bnx2x_free_irq(bp);
1749 }
1750
1751 bp->port.pmf = 0;
1752
1753 /* Free SKBs, SGEs, TPA pool and driver internals */
1754 bnx2x_free_skbs(bp);
1755 for_each_rx_queue(bp, i)
1756 bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
1757
1758 bnx2x_free_mem(bp);
1759
1760 bp->state = BNX2X_STATE_CLOSED;
1761
1762 /* The last driver must disable a "close the gate" if there is no
1763 * parity attention or "process kill" pending.
1764 */
1765 if ((!bnx2x_dec_load_cnt(bp)) && (!bnx2x_chk_parity_attn(bp)) &&
1766 bnx2x_reset_is_done(bp))
1767 bnx2x_disable_close_the_gate(bp);
1768
1769 /* Reset MCP mail box sequence if there is on going recovery */
1770 if (unload_mode == UNLOAD_RECOVERY)
1771 bp->fw_seq = 0;
1772
1773 return 0;
1774 }
1775
1776 int bnx2x_set_power_state(struct bnx2x *bp, pci_power_t state)
1777 {
1778 u16 pmcsr;
1779
1780 /* If there is no power capability, silently succeed */
1781 if (!bp->pm_cap) {
1782 DP(NETIF_MSG_HW, "No power capability. Breaking.\n");
1783 return 0;
1784 }
1785
1786 pci_read_config_word(bp->pdev, bp->pm_cap + PCI_PM_CTRL, &pmcsr);
1787
1788 switch (state) {
1789 case PCI_D0:
1790 pci_write_config_word(bp->pdev, bp->pm_cap + PCI_PM_CTRL,
1791 ((pmcsr & ~PCI_PM_CTRL_STATE_MASK) |
1792 PCI_PM_CTRL_PME_STATUS));
1793
1794 if (pmcsr & PCI_PM_CTRL_STATE_MASK)
1795 /* delay required during transition out of D3hot */
1796 msleep(20);
1797 break;
1798
1799 case PCI_D3hot:
1800 /* If there are other clients above don't
1801 shut down the power */
1802 if (atomic_read(&bp->pdev->enable_cnt) != 1)
1803 return 0;
1804 /* Don't shut down the power for emulation and FPGA */
1805 if (CHIP_REV_IS_SLOW(bp))
1806 return 0;
1807
1808 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
1809 pmcsr |= 3;
1810
1811 if (bp->wol)
1812 pmcsr |= PCI_PM_CTRL_PME_ENABLE;
1813
1814 pci_write_config_word(bp->pdev, bp->pm_cap + PCI_PM_CTRL,
1815 pmcsr);
1816
1817 /* No more memory access after this point until
1818 * device is brought back to D0.
1819 */
1820 break;
1821
1822 default:
1823 return -EINVAL;
1824 }
1825 return 0;
1826 }
1827
1828 /*
1829 * net_device service functions
1830 */
1831 int bnx2x_poll(struct napi_struct *napi, int budget)
1832 {
1833 int work_done = 0;
1834 struct bnx2x_fastpath *fp = container_of(napi, struct bnx2x_fastpath,
1835 napi);
1836 struct bnx2x *bp = fp->bp;
1837
1838 while (1) {
1839 #ifdef BNX2X_STOP_ON_ERROR
1840 if (unlikely(bp->panic)) {
1841 napi_complete(napi);
1842 return 0;
1843 }
1844 #endif
1845
1846 if (bnx2x_has_tx_work(fp))
1847 bnx2x_tx_int(fp);
1848
1849 if (bnx2x_has_rx_work(fp)) {
1850 work_done += bnx2x_rx_int(fp, budget - work_done);
1851
1852 /* must not complete if we consumed full budget */
1853 if (work_done >= budget)
1854 break;
1855 }
1856
1857 /* Fall out from the NAPI loop if needed */
1858 if (!(bnx2x_has_rx_work(fp) || bnx2x_has_tx_work(fp))) {
1859 #ifdef BCM_CNIC
1860 /* No need to update SB for FCoE L2 ring as long as
1861 * it's connected to the default SB and the SB
1862 * has been updated when NAPI was scheduled.
1863 */
1864 if (IS_FCOE_FP(fp)) {
1865 napi_complete(napi);
1866 break;
1867 }
1868 #endif
1869
1870 bnx2x_update_fpsb_idx(fp);
1871 /* bnx2x_has_rx_work() reads the status block,
1872 * thus we need to ensure that status block indices
1873 * have been actually read (bnx2x_update_fpsb_idx)
1874 * prior to this check (bnx2x_has_rx_work) so that
1875 * we won't write the "newer" value of the status block
1876 * to IGU (if there was a DMA right after
1877 * bnx2x_has_rx_work and if there is no rmb, the memory
1878 * reading (bnx2x_update_fpsb_idx) may be postponed
1879 * to right before bnx2x_ack_sb). In this case there
1880 * will never be another interrupt until there is
1881 * another update of the status block, while there
1882 * is still unhandled work.
1883 */
1884 rmb();
1885
1886 if (!(bnx2x_has_rx_work(fp) || bnx2x_has_tx_work(fp))) {
1887 napi_complete(napi);
1888 /* Re-enable interrupts */
1889 DP(NETIF_MSG_HW,
1890 "Update index to %d\n", fp->fp_hc_idx);
1891 bnx2x_ack_sb(bp, fp->igu_sb_id, USTORM_ID,
1892 le16_to_cpu(fp->fp_hc_idx),
1893 IGU_INT_ENABLE, 1);
1894 break;
1895 }
1896 }
1897 }
1898
1899 return work_done;
1900 }
1901
1902 /* we split the first BD into headers and data BDs
1903 * to ease the pain of our fellow microcode engineers
1904 * we use one mapping for both BDs
1905 * So far this has only been observed to happen
1906 * in Other Operating Systems(TM)
1907 */
1908 static noinline u16 bnx2x_tx_split(struct bnx2x *bp,
1909 struct bnx2x_fastpath *fp,
1910 struct sw_tx_bd *tx_buf,
1911 struct eth_tx_start_bd **tx_bd, u16 hlen,
1912 u16 bd_prod, int nbd)
1913 {
1914 struct eth_tx_start_bd *h_tx_bd = *tx_bd;
1915 struct eth_tx_bd *d_tx_bd;
1916 dma_addr_t mapping;
1917 int old_len = le16_to_cpu(h_tx_bd->nbytes);
1918
1919 /* first fix first BD */
1920 h_tx_bd->nbd = cpu_to_le16(nbd);
1921 h_tx_bd->nbytes = cpu_to_le16(hlen);
1922
1923 DP(NETIF_MSG_TX_QUEUED, "TSO split header size is %d "
1924 "(%x:%x) nbd %d\n", h_tx_bd->nbytes, h_tx_bd->addr_hi,
1925 h_tx_bd->addr_lo, h_tx_bd->nbd);
1926
1927 /* now get a new data BD
1928 * (after the pbd) and fill it */
1929 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
1930 d_tx_bd = &fp->tx_desc_ring[bd_prod].reg_bd;
1931
1932 mapping = HILO_U64(le32_to_cpu(h_tx_bd->addr_hi),
1933 le32_to_cpu(h_tx_bd->addr_lo)) + hlen;
1934
1935 d_tx_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
1936 d_tx_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
1937 d_tx_bd->nbytes = cpu_to_le16(old_len - hlen);
1938
1939 /* this marks the BD as one that has no individual mapping */
1940 tx_buf->flags |= BNX2X_TSO_SPLIT_BD;
1941
1942 DP(NETIF_MSG_TX_QUEUED,
1943 "TSO split data size is %d (%x:%x)\n",
1944 d_tx_bd->nbytes, d_tx_bd->addr_hi, d_tx_bd->addr_lo);
1945
1946 /* update tx_bd */
1947 *tx_bd = (struct eth_tx_start_bd *)d_tx_bd;
1948
1949 return bd_prod;
1950 }
1951
1952 static inline u16 bnx2x_csum_fix(unsigned char *t_header, u16 csum, s8 fix)
1953 {
1954 if (fix > 0)
1955 csum = (u16) ~csum_fold(csum_sub(csum,
1956 csum_partial(t_header - fix, fix, 0)));
1957
1958 else if (fix < 0)
1959 csum = (u16) ~csum_fold(csum_add(csum,
1960 csum_partial(t_header, -fix, 0)));
1961
1962 return swab16(csum);
1963 }
1964
1965 static inline u32 bnx2x_xmit_type(struct bnx2x *bp, struct sk_buff *skb)
1966 {
1967 u32 rc;
1968
1969 if (skb->ip_summed != CHECKSUM_PARTIAL)
1970 rc = XMIT_PLAIN;
1971
1972 else {
1973 if (vlan_get_protocol(skb) == htons(ETH_P_IPV6)) {
1974 rc = XMIT_CSUM_V6;
1975 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
1976 rc |= XMIT_CSUM_TCP;
1977
1978 } else {
1979 rc = XMIT_CSUM_V4;
1980 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
1981 rc |= XMIT_CSUM_TCP;
1982 }
1983 }
1984
1985 if (skb_is_gso_v6(skb))
1986 rc |= XMIT_GSO_V6 | XMIT_CSUM_TCP | XMIT_CSUM_V6;
1987 else if (skb_is_gso(skb))
1988 rc |= XMIT_GSO_V4 | XMIT_CSUM_V4 | XMIT_CSUM_TCP;
1989
1990 return rc;
1991 }
1992
1993 #if (MAX_SKB_FRAGS >= MAX_FETCH_BD - 3)
1994 /* check if packet requires linearization (packet is too fragmented)
1995 no need to check fragmentation if page size > 8K (there will be no
1996 violation to FW restrictions) */
1997 static int bnx2x_pkt_req_lin(struct bnx2x *bp, struct sk_buff *skb,
1998 u32 xmit_type)
1999 {
2000 int to_copy = 0;
2001 int hlen = 0;
2002 int first_bd_sz = 0;
2003
2004 /* 3 = 1 (for linear data BD) + 2 (for PBD and last BD) */
2005 if (skb_shinfo(skb)->nr_frags >= (MAX_FETCH_BD - 3)) {
2006
2007 if (xmit_type & XMIT_GSO) {
2008 unsigned short lso_mss = skb_shinfo(skb)->gso_size;
2009 /* Check if LSO packet needs to be copied:
2010 3 = 1 (for headers BD) + 2 (for PBD and last BD) */
2011 int wnd_size = MAX_FETCH_BD - 3;
2012 /* Number of windows to check */
2013 int num_wnds = skb_shinfo(skb)->nr_frags - wnd_size;
2014 int wnd_idx = 0;
2015 int frag_idx = 0;
2016 u32 wnd_sum = 0;
2017
2018 /* Headers length */
2019 hlen = (int)(skb_transport_header(skb) - skb->data) +
2020 tcp_hdrlen(skb);
2021
2022 /* Amount of data (w/o headers) on linear part of SKB*/
2023 first_bd_sz = skb_headlen(skb) - hlen;
2024
2025 wnd_sum = first_bd_sz;
2026
2027 /* Calculate the first sum - it's special */
2028 for (frag_idx = 0; frag_idx < wnd_size - 1; frag_idx++)
2029 wnd_sum +=
2030 skb_shinfo(skb)->frags[frag_idx].size;
2031
2032 /* If there was data on linear skb data - check it */
2033 if (first_bd_sz > 0) {
2034 if (unlikely(wnd_sum < lso_mss)) {
2035 to_copy = 1;
2036 goto exit_lbl;
2037 }
2038
2039 wnd_sum -= first_bd_sz;
2040 }
2041
2042 /* Others are easier: run through the frag list and
2043 check all windows */
2044 for (wnd_idx = 0; wnd_idx <= num_wnds; wnd_idx++) {
2045 wnd_sum +=
2046 skb_shinfo(skb)->frags[wnd_idx + wnd_size - 1].size;
2047
2048 if (unlikely(wnd_sum < lso_mss)) {
2049 to_copy = 1;
2050 break;
2051 }
2052 wnd_sum -=
2053 skb_shinfo(skb)->frags[wnd_idx].size;
2054 }
2055 } else {
2056 /* in non-LSO too fragmented packet should always
2057 be linearized */
2058 to_copy = 1;
2059 }
2060 }
2061
2062 exit_lbl:
2063 if (unlikely(to_copy))
2064 DP(NETIF_MSG_TX_QUEUED,
2065 "Linearization IS REQUIRED for %s packet. "
2066 "num_frags %d hlen %d first_bd_sz %d\n",
2067 (xmit_type & XMIT_GSO) ? "LSO" : "non-LSO",
2068 skb_shinfo(skb)->nr_frags, hlen, first_bd_sz);
2069
2070 return to_copy;
2071 }
2072 #endif
2073
2074 static inline void bnx2x_set_pbd_gso_e2(struct sk_buff *skb, u32 *parsing_data,
2075 u32 xmit_type)
2076 {
2077 *parsing_data |= (skb_shinfo(skb)->gso_size <<
2078 ETH_TX_PARSE_BD_E2_LSO_MSS_SHIFT) &
2079 ETH_TX_PARSE_BD_E2_LSO_MSS;
2080 if ((xmit_type & XMIT_GSO_V6) &&
2081 (ipv6_hdr(skb)->nexthdr == NEXTHDR_IPV6))
2082 *parsing_data |= ETH_TX_PARSE_BD_E2_IPV6_WITH_EXT_HDR;
2083 }
2084
2085 /**
2086 * Update PBD in GSO case.
2087 *
2088 * @param skb
2089 * @param tx_start_bd
2090 * @param pbd
2091 * @param xmit_type
2092 */
2093 static inline void bnx2x_set_pbd_gso(struct sk_buff *skb,
2094 struct eth_tx_parse_bd_e1x *pbd,
2095 u32 xmit_type)
2096 {
2097 pbd->lso_mss = cpu_to_le16(skb_shinfo(skb)->gso_size);
2098 pbd->tcp_send_seq = swab32(tcp_hdr(skb)->seq);
2099 pbd->tcp_flags = pbd_tcp_flags(skb);
2100
2101 if (xmit_type & XMIT_GSO_V4) {
2102 pbd->ip_id = swab16(ip_hdr(skb)->id);
2103 pbd->tcp_pseudo_csum =
2104 swab16(~csum_tcpudp_magic(ip_hdr(skb)->saddr,
2105 ip_hdr(skb)->daddr,
2106 0, IPPROTO_TCP, 0));
2107
2108 } else
2109 pbd->tcp_pseudo_csum =
2110 swab16(~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2111 &ipv6_hdr(skb)->daddr,
2112 0, IPPROTO_TCP, 0));
2113
2114 pbd->global_data |= ETH_TX_PARSE_BD_E1X_PSEUDO_CS_WITHOUT_LEN;
2115 }
2116
2117 /**
2118 *
2119 * @param skb
2120 * @param tx_start_bd
2121 * @param pbd_e2
2122 * @param xmit_type
2123 *
2124 * @return header len
2125 */
2126 static inline u8 bnx2x_set_pbd_csum_e2(struct bnx2x *bp, struct sk_buff *skb,
2127 u32 *parsing_data, u32 xmit_type)
2128 {
2129 *parsing_data |=
2130 ((((u8 *)skb_transport_header(skb) - skb->data) >> 1) <<
2131 ETH_TX_PARSE_BD_E2_TCP_HDR_START_OFFSET_W_SHIFT) &
2132 ETH_TX_PARSE_BD_E2_TCP_HDR_START_OFFSET_W;
2133
2134 if (xmit_type & XMIT_CSUM_TCP) {
2135 *parsing_data |= ((tcp_hdrlen(skb) / 4) <<
2136 ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW_SHIFT) &
2137 ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW;
2138
2139 return skb_transport_header(skb) + tcp_hdrlen(skb) - skb->data;
2140 } else
2141 /* We support checksum offload for TCP and UDP only.
2142 * No need to pass the UDP header length - it's a constant.
2143 */
2144 return skb_transport_header(skb) +
2145 sizeof(struct udphdr) - skb->data;
2146 }
2147
2148 /**
2149 *
2150 * @param skb
2151 * @param tx_start_bd
2152 * @param pbd
2153 * @param xmit_type
2154 *
2155 * @return Header length
2156 */
2157 static inline u8 bnx2x_set_pbd_csum(struct bnx2x *bp, struct sk_buff *skb,
2158 struct eth_tx_parse_bd_e1x *pbd,
2159 u32 xmit_type)
2160 {
2161 u8 hlen = (skb_network_header(skb) - skb->data) >> 1;
2162
2163 /* for now NS flag is not used in Linux */
2164 pbd->global_data =
2165 (hlen | ((skb->protocol == cpu_to_be16(ETH_P_8021Q)) <<
2166 ETH_TX_PARSE_BD_E1X_LLC_SNAP_EN_SHIFT));
2167
2168 pbd->ip_hlen_w = (skb_transport_header(skb) -
2169 skb_network_header(skb)) >> 1;
2170
2171 hlen += pbd->ip_hlen_w;
2172
2173 /* We support checksum offload for TCP and UDP only */
2174 if (xmit_type & XMIT_CSUM_TCP)
2175 hlen += tcp_hdrlen(skb) / 2;
2176 else
2177 hlen += sizeof(struct udphdr) / 2;
2178
2179 pbd->total_hlen_w = cpu_to_le16(hlen);
2180 hlen = hlen*2;
2181
2182 if (xmit_type & XMIT_CSUM_TCP) {
2183 pbd->tcp_pseudo_csum = swab16(tcp_hdr(skb)->check);
2184
2185 } else {
2186 s8 fix = SKB_CS_OFF(skb); /* signed! */
2187
2188 DP(NETIF_MSG_TX_QUEUED,
2189 "hlen %d fix %d csum before fix %x\n",
2190 le16_to_cpu(pbd->total_hlen_w), fix, SKB_CS(skb));
2191
2192 /* HW bug: fixup the CSUM */
2193 pbd->tcp_pseudo_csum =
2194 bnx2x_csum_fix(skb_transport_header(skb),
2195 SKB_CS(skb), fix);
2196
2197 DP(NETIF_MSG_TX_QUEUED, "csum after fix %x\n",
2198 pbd->tcp_pseudo_csum);
2199 }
2200
2201 return hlen;
2202 }
2203
2204 /* called with netif_tx_lock
2205 * bnx2x_tx_int() runs without netif_tx_lock unless it needs to call
2206 * netif_wake_queue()
2207 */
2208 netdev_tx_t bnx2x_start_xmit(struct sk_buff *skb, struct net_device *dev)
2209 {
2210 struct bnx2x *bp = netdev_priv(dev);
2211 struct bnx2x_fastpath *fp;
2212 struct netdev_queue *txq;
2213 struct sw_tx_bd *tx_buf;
2214 struct eth_tx_start_bd *tx_start_bd;
2215 struct eth_tx_bd *tx_data_bd, *total_pkt_bd = NULL;
2216 struct eth_tx_parse_bd_e1x *pbd_e1x = NULL;
2217 struct eth_tx_parse_bd_e2 *pbd_e2 = NULL;
2218 u32 pbd_e2_parsing_data = 0;
2219 u16 pkt_prod, bd_prod;
2220 int nbd, fp_index;
2221 dma_addr_t mapping;
2222 u32 xmit_type = bnx2x_xmit_type(bp, skb);
2223 int i;
2224 u8 hlen = 0;
2225 __le16 pkt_size = 0;
2226 struct ethhdr *eth;
2227 u8 mac_type = UNICAST_ADDRESS;
2228
2229 #ifdef BNX2X_STOP_ON_ERROR
2230 if (unlikely(bp->panic))
2231 return NETDEV_TX_BUSY;
2232 #endif
2233
2234 fp_index = skb_get_queue_mapping(skb);
2235 txq = netdev_get_tx_queue(dev, fp_index);
2236
2237 fp = &bp->fp[fp_index];
2238
2239 if (unlikely(bnx2x_tx_avail(fp) < (skb_shinfo(skb)->nr_frags + 3))) {
2240 fp->eth_q_stats.driver_xoff++;
2241 netif_tx_stop_queue(txq);
2242 BNX2X_ERR("BUG! Tx ring full when queue awake!\n");
2243 return NETDEV_TX_BUSY;
2244 }
2245
2246 DP(NETIF_MSG_TX_QUEUED, "queue[%d]: SKB: summed %x protocol %x "
2247 "protocol(%x,%x) gso type %x xmit_type %x\n",
2248 fp_index, skb->ip_summed, skb->protocol, ipv6_hdr(skb)->nexthdr,
2249 ip_hdr(skb)->protocol, skb_shinfo(skb)->gso_type, xmit_type);
2250
2251 eth = (struct ethhdr *)skb->data;
2252
2253 /* set flag according to packet type (UNICAST_ADDRESS is default)*/
2254 if (unlikely(is_multicast_ether_addr(eth->h_dest))) {
2255 if (is_broadcast_ether_addr(eth->h_dest))
2256 mac_type = BROADCAST_ADDRESS;
2257 else
2258 mac_type = MULTICAST_ADDRESS;
2259 }
2260
2261 #if (MAX_SKB_FRAGS >= MAX_FETCH_BD - 3)
2262 /* First, check if we need to linearize the skb (due to FW
2263 restrictions). No need to check fragmentation if page size > 8K
2264 (there will be no violation to FW restrictions) */
2265 if (bnx2x_pkt_req_lin(bp, skb, xmit_type)) {
2266 /* Statistics of linearization */
2267 bp->lin_cnt++;
2268 if (skb_linearize(skb) != 0) {
2269 DP(NETIF_MSG_TX_QUEUED, "SKB linearization failed - "
2270 "silently dropping this SKB\n");
2271 dev_kfree_skb_any(skb);
2272 return NETDEV_TX_OK;
2273 }
2274 }
2275 #endif
2276
2277 /*
2278 Please read carefully. First we use one BD which we mark as start,
2279 then we have a parsing info BD (used for TSO or xsum),
2280 and only then we have the rest of the TSO BDs.
2281 (don't forget to mark the last one as last,
2282 and to unmap only AFTER you write to the BD ...)
2283 And above all, all pdb sizes are in words - NOT DWORDS!
2284 */
2285
2286 pkt_prod = fp->tx_pkt_prod++;
2287 bd_prod = TX_BD(fp->tx_bd_prod);
2288
2289 /* get a tx_buf and first BD */
2290 tx_buf = &fp->tx_buf_ring[TX_BD(pkt_prod)];
2291 tx_start_bd = &fp->tx_desc_ring[bd_prod].start_bd;
2292
2293 tx_start_bd->bd_flags.as_bitfield = ETH_TX_BD_FLAGS_START_BD;
2294 SET_FLAG(tx_start_bd->general_data, ETH_TX_START_BD_ETH_ADDR_TYPE,
2295 mac_type);
2296
2297 /* header nbd */
2298 SET_FLAG(tx_start_bd->general_data, ETH_TX_START_BD_HDR_NBDS, 1);
2299
2300 /* remember the first BD of the packet */
2301 tx_buf->first_bd = fp->tx_bd_prod;
2302 tx_buf->skb = skb;
2303 tx_buf->flags = 0;
2304
2305 DP(NETIF_MSG_TX_QUEUED,
2306 "sending pkt %u @%p next_idx %u bd %u @%p\n",
2307 pkt_prod, tx_buf, fp->tx_pkt_prod, bd_prod, tx_start_bd);
2308
2309 if (vlan_tx_tag_present(skb)) {
2310 tx_start_bd->vlan_or_ethertype =
2311 cpu_to_le16(vlan_tx_tag_get(skb));
2312 tx_start_bd->bd_flags.as_bitfield |=
2313 (X_ETH_OUTBAND_VLAN << ETH_TX_BD_FLAGS_VLAN_MODE_SHIFT);
2314 } else
2315 tx_start_bd->vlan_or_ethertype = cpu_to_le16(pkt_prod);
2316
2317 /* turn on parsing and get a BD */
2318 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
2319
2320 if (xmit_type & XMIT_CSUM) {
2321 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_L4_CSUM;
2322
2323 if (xmit_type & XMIT_CSUM_V4)
2324 tx_start_bd->bd_flags.as_bitfield |=
2325 ETH_TX_BD_FLAGS_IP_CSUM;
2326 else
2327 tx_start_bd->bd_flags.as_bitfield |=
2328 ETH_TX_BD_FLAGS_IPV6;
2329
2330 if (!(xmit_type & XMIT_CSUM_TCP))
2331 tx_start_bd->bd_flags.as_bitfield |=
2332 ETH_TX_BD_FLAGS_IS_UDP;
2333 }
2334
2335 if (CHIP_IS_E2(bp)) {
2336 pbd_e2 = &fp->tx_desc_ring[bd_prod].parse_bd_e2;
2337 memset(pbd_e2, 0, sizeof(struct eth_tx_parse_bd_e2));
2338 /* Set PBD in checksum offload case */
2339 if (xmit_type & XMIT_CSUM)
2340 hlen = bnx2x_set_pbd_csum_e2(bp, skb,
2341 &pbd_e2_parsing_data,
2342 xmit_type);
2343 } else {
2344 pbd_e1x = &fp->tx_desc_ring[bd_prod].parse_bd_e1x;
2345 memset(pbd_e1x, 0, sizeof(struct eth_tx_parse_bd_e1x));
2346 /* Set PBD in checksum offload case */
2347 if (xmit_type & XMIT_CSUM)
2348 hlen = bnx2x_set_pbd_csum(bp, skb, pbd_e1x, xmit_type);
2349
2350 }
2351
2352 /* Map skb linear data for DMA */
2353 mapping = dma_map_single(&bp->pdev->dev, skb->data,
2354 skb_headlen(skb), DMA_TO_DEVICE);
2355
2356 /* Setup the data pointer of the first BD of the packet */
2357 tx_start_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
2358 tx_start_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
2359 nbd = skb_shinfo(skb)->nr_frags + 2; /* start_bd + pbd + frags */
2360 tx_start_bd->nbd = cpu_to_le16(nbd);
2361 tx_start_bd->nbytes = cpu_to_le16(skb_headlen(skb));
2362 pkt_size = tx_start_bd->nbytes;
2363
2364 DP(NETIF_MSG_TX_QUEUED, "first bd @%p addr (%x:%x) nbd %d"
2365 " nbytes %d flags %x vlan %x\n",
2366 tx_start_bd, tx_start_bd->addr_hi, tx_start_bd->addr_lo,
2367 le16_to_cpu(tx_start_bd->nbd), le16_to_cpu(tx_start_bd->nbytes),
2368 tx_start_bd->bd_flags.as_bitfield,
2369 le16_to_cpu(tx_start_bd->vlan_or_ethertype));
2370
2371 if (xmit_type & XMIT_GSO) {
2372
2373 DP(NETIF_MSG_TX_QUEUED,
2374 "TSO packet len %d hlen %d total len %d tso size %d\n",
2375 skb->len, hlen, skb_headlen(skb),
2376 skb_shinfo(skb)->gso_size);
2377
2378 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_SW_LSO;
2379
2380 if (unlikely(skb_headlen(skb) > hlen))
2381 bd_prod = bnx2x_tx_split(bp, fp, tx_buf, &tx_start_bd,
2382 hlen, bd_prod, ++nbd);
2383 if (CHIP_IS_E2(bp))
2384 bnx2x_set_pbd_gso_e2(skb, &pbd_e2_parsing_data,
2385 xmit_type);
2386 else
2387 bnx2x_set_pbd_gso(skb, pbd_e1x, xmit_type);
2388 }
2389
2390 /* Set the PBD's parsing_data field if not zero
2391 * (for the chips newer than 57711).
2392 */
2393 if (pbd_e2_parsing_data)
2394 pbd_e2->parsing_data = cpu_to_le32(pbd_e2_parsing_data);
2395
2396 tx_data_bd = (struct eth_tx_bd *)tx_start_bd;
2397
2398 /* Handle fragmented skb */
2399 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2400 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2401
2402 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
2403 tx_data_bd = &fp->tx_desc_ring[bd_prod].reg_bd;
2404 if (total_pkt_bd == NULL)
2405 total_pkt_bd = &fp->tx_desc_ring[bd_prod].reg_bd;
2406
2407 mapping = dma_map_page(&bp->pdev->dev, frag->page,
2408 frag->page_offset,
2409 frag->size, DMA_TO_DEVICE);
2410
2411 tx_data_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
2412 tx_data_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
2413 tx_data_bd->nbytes = cpu_to_le16(frag->size);
2414 le16_add_cpu(&pkt_size, frag->size);
2415
2416 DP(NETIF_MSG_TX_QUEUED,
2417 "frag %d bd @%p addr (%x:%x) nbytes %d\n",
2418 i, tx_data_bd, tx_data_bd->addr_hi, tx_data_bd->addr_lo,
2419 le16_to_cpu(tx_data_bd->nbytes));
2420 }
2421
2422 DP(NETIF_MSG_TX_QUEUED, "last bd @%p\n", tx_data_bd);
2423
2424 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
2425
2426 /* now send a tx doorbell, counting the next BD
2427 * if the packet contains or ends with it
2428 */
2429 if (TX_BD_POFF(bd_prod) < nbd)
2430 nbd++;
2431
2432 if (total_pkt_bd != NULL)
2433 total_pkt_bd->total_pkt_bytes = pkt_size;
2434
2435 if (pbd_e1x)
2436 DP(NETIF_MSG_TX_QUEUED,
2437 "PBD (E1X) @%p ip_data %x ip_hlen %u ip_id %u lso_mss %u"
2438 " tcp_flags %x xsum %x seq %u hlen %u\n",
2439 pbd_e1x, pbd_e1x->global_data, pbd_e1x->ip_hlen_w,
2440 pbd_e1x->ip_id, pbd_e1x->lso_mss, pbd_e1x->tcp_flags,
2441 pbd_e1x->tcp_pseudo_csum, pbd_e1x->tcp_send_seq,
2442 le16_to_cpu(pbd_e1x->total_hlen_w));
2443 if (pbd_e2)
2444 DP(NETIF_MSG_TX_QUEUED,
2445 "PBD (E2) @%p dst %x %x %x src %x %x %x parsing_data %x\n",
2446 pbd_e2, pbd_e2->dst_mac_addr_hi, pbd_e2->dst_mac_addr_mid,
2447 pbd_e2->dst_mac_addr_lo, pbd_e2->src_mac_addr_hi,
2448 pbd_e2->src_mac_addr_mid, pbd_e2->src_mac_addr_lo,
2449 pbd_e2->parsing_data);
2450 DP(NETIF_MSG_TX_QUEUED, "doorbell: nbd %d bd %u\n", nbd, bd_prod);
2451
2452 /*
2453 * Make sure that the BD data is updated before updating the producer
2454 * since FW might read the BD right after the producer is updated.
2455 * This is only applicable for weak-ordered memory model archs such
2456 * as IA-64. The following barrier is also mandatory since FW will
2457 * assumes packets must have BDs.
2458 */
2459 wmb();
2460
2461 fp->tx_db.data.prod += nbd;
2462 barrier();
2463
2464 DOORBELL(bp, fp->cid, fp->tx_db.raw);
2465
2466 mmiowb();
2467
2468 fp->tx_bd_prod += nbd;
2469
2470 if (unlikely(bnx2x_tx_avail(fp) < MAX_SKB_FRAGS + 3)) {
2471 netif_tx_stop_queue(txq);
2472
2473 /* paired memory barrier is in bnx2x_tx_int(), we have to keep
2474 * ordering of set_bit() in netif_tx_stop_queue() and read of
2475 * fp->bd_tx_cons */
2476 smp_mb();
2477
2478 fp->eth_q_stats.driver_xoff++;
2479 if (bnx2x_tx_avail(fp) >= MAX_SKB_FRAGS + 3)
2480 netif_tx_wake_queue(txq);
2481 }
2482 fp->tx_pkt++;
2483
2484 return NETDEV_TX_OK;
2485 }
2486
2487 /* called with rtnl_lock */
2488 int bnx2x_change_mac_addr(struct net_device *dev, void *p)
2489 {
2490 struct sockaddr *addr = p;
2491 struct bnx2x *bp = netdev_priv(dev);
2492
2493 if (!is_valid_ether_addr((u8 *)(addr->sa_data)))
2494 return -EINVAL;
2495
2496 memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
2497 if (netif_running(dev))
2498 bnx2x_set_eth_mac(bp, 1);
2499
2500 return 0;
2501 }
2502
2503 static void bnx2x_free_fp_mem_at(struct bnx2x *bp, int fp_index)
2504 {
2505 union host_hc_status_block *sb = &bnx2x_fp(bp, fp_index, status_blk);
2506 struct bnx2x_fastpath *fp = &bp->fp[fp_index];
2507
2508 /* Common */
2509 #ifdef BCM_CNIC
2510 if (IS_FCOE_IDX(fp_index)) {
2511 memset(sb, 0, sizeof(union host_hc_status_block));
2512 fp->status_blk_mapping = 0;
2513
2514 } else {
2515 #endif
2516 /* status blocks */
2517 if (CHIP_IS_E2(bp))
2518 BNX2X_PCI_FREE(sb->e2_sb,
2519 bnx2x_fp(bp, fp_index,
2520 status_blk_mapping),
2521 sizeof(struct host_hc_status_block_e2));
2522 else
2523 BNX2X_PCI_FREE(sb->e1x_sb,
2524 bnx2x_fp(bp, fp_index,
2525 status_blk_mapping),
2526 sizeof(struct host_hc_status_block_e1x));
2527 #ifdef BCM_CNIC
2528 }
2529 #endif
2530 /* Rx */
2531 if (!skip_rx_queue(bp, fp_index)) {
2532 bnx2x_free_rx_bds(fp);
2533
2534 /* fastpath rx rings: rx_buf rx_desc rx_comp */
2535 BNX2X_FREE(bnx2x_fp(bp, fp_index, rx_buf_ring));
2536 BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_desc_ring),
2537 bnx2x_fp(bp, fp_index, rx_desc_mapping),
2538 sizeof(struct eth_rx_bd) * NUM_RX_BD);
2539
2540 BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_comp_ring),
2541 bnx2x_fp(bp, fp_index, rx_comp_mapping),
2542 sizeof(struct eth_fast_path_rx_cqe) *
2543 NUM_RCQ_BD);
2544
2545 /* SGE ring */
2546 BNX2X_FREE(bnx2x_fp(bp, fp_index, rx_page_ring));
2547 BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_sge_ring),
2548 bnx2x_fp(bp, fp_index, rx_sge_mapping),
2549 BCM_PAGE_SIZE * NUM_RX_SGE_PAGES);
2550 }
2551
2552 /* Tx */
2553 if (!skip_tx_queue(bp, fp_index)) {
2554 /* fastpath tx rings: tx_buf tx_desc */
2555 BNX2X_FREE(bnx2x_fp(bp, fp_index, tx_buf_ring));
2556 BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, tx_desc_ring),
2557 bnx2x_fp(bp, fp_index, tx_desc_mapping),
2558 sizeof(union eth_tx_bd_types) * NUM_TX_BD);
2559 }
2560 /* end of fastpath */
2561 }
2562
2563 void bnx2x_free_fp_mem(struct bnx2x *bp)
2564 {
2565 int i;
2566 for_each_queue(bp, i)
2567 bnx2x_free_fp_mem_at(bp, i);
2568 }
2569
2570 static inline void set_sb_shortcuts(struct bnx2x *bp, int index)
2571 {
2572 union host_hc_status_block status_blk = bnx2x_fp(bp, index, status_blk);
2573 if (CHIP_IS_E2(bp)) {
2574 bnx2x_fp(bp, index, sb_index_values) =
2575 (__le16 *)status_blk.e2_sb->sb.index_values;
2576 bnx2x_fp(bp, index, sb_running_index) =
2577 (__le16 *)status_blk.e2_sb->sb.running_index;
2578 } else {
2579 bnx2x_fp(bp, index, sb_index_values) =
2580 (__le16 *)status_blk.e1x_sb->sb.index_values;
2581 bnx2x_fp(bp, index, sb_running_index) =
2582 (__le16 *)status_blk.e1x_sb->sb.running_index;
2583 }
2584 }
2585
2586 static int bnx2x_alloc_fp_mem_at(struct bnx2x *bp, int index)
2587 {
2588 union host_hc_status_block *sb;
2589 struct bnx2x_fastpath *fp = &bp->fp[index];
2590 int ring_size = 0;
2591
2592 /* if rx_ring_size specified - use it */
2593 int rx_ring_size = bp->rx_ring_size ? bp->rx_ring_size :
2594 MAX_RX_AVAIL/bp->num_queues;
2595
2596 /* allocate at least number of buffers required by FW */
2597 rx_ring_size = max_t(int, fp->disable_tpa ? MIN_RX_SIZE_NONTPA :
2598 MIN_RX_SIZE_TPA,
2599 rx_ring_size);
2600
2601 bnx2x_fp(bp, index, bp) = bp;
2602 bnx2x_fp(bp, index, index) = index;
2603
2604 /* Common */
2605 sb = &bnx2x_fp(bp, index, status_blk);
2606 #ifdef BCM_CNIC
2607 if (!IS_FCOE_IDX(index)) {
2608 #endif
2609 /* status blocks */
2610 if (CHIP_IS_E2(bp))
2611 BNX2X_PCI_ALLOC(sb->e2_sb,
2612 &bnx2x_fp(bp, index, status_blk_mapping),
2613 sizeof(struct host_hc_status_block_e2));
2614 else
2615 BNX2X_PCI_ALLOC(sb->e1x_sb,
2616 &bnx2x_fp(bp, index, status_blk_mapping),
2617 sizeof(struct host_hc_status_block_e1x));
2618 #ifdef BCM_CNIC
2619 }
2620 #endif
2621 set_sb_shortcuts(bp, index);
2622
2623 /* Tx */
2624 if (!skip_tx_queue(bp, index)) {
2625 /* fastpath tx rings: tx_buf tx_desc */
2626 BNX2X_ALLOC(bnx2x_fp(bp, index, tx_buf_ring),
2627 sizeof(struct sw_tx_bd) * NUM_TX_BD);
2628 BNX2X_PCI_ALLOC(bnx2x_fp(bp, index, tx_desc_ring),
2629 &bnx2x_fp(bp, index, tx_desc_mapping),
2630 sizeof(union eth_tx_bd_types) * NUM_TX_BD);
2631 }
2632
2633 /* Rx */
2634 if (!skip_rx_queue(bp, index)) {
2635 /* fastpath rx rings: rx_buf rx_desc rx_comp */
2636 BNX2X_ALLOC(bnx2x_fp(bp, index, rx_buf_ring),
2637 sizeof(struct sw_rx_bd) * NUM_RX_BD);
2638 BNX2X_PCI_ALLOC(bnx2x_fp(bp, index, rx_desc_ring),
2639 &bnx2x_fp(bp, index, rx_desc_mapping),
2640 sizeof(struct eth_rx_bd) * NUM_RX_BD);
2641
2642 BNX2X_PCI_ALLOC(bnx2x_fp(bp, index, rx_comp_ring),
2643 &bnx2x_fp(bp, index, rx_comp_mapping),
2644 sizeof(struct eth_fast_path_rx_cqe) *
2645 NUM_RCQ_BD);
2646
2647 /* SGE ring */
2648 BNX2X_ALLOC(bnx2x_fp(bp, index, rx_page_ring),
2649 sizeof(struct sw_rx_page) * NUM_RX_SGE);
2650 BNX2X_PCI_ALLOC(bnx2x_fp(bp, index, rx_sge_ring),
2651 &bnx2x_fp(bp, index, rx_sge_mapping),
2652 BCM_PAGE_SIZE * NUM_RX_SGE_PAGES);
2653 /* RX BD ring */
2654 bnx2x_set_next_page_rx_bd(fp);
2655
2656 /* CQ ring */
2657 bnx2x_set_next_page_rx_cq(fp);
2658
2659 /* BDs */
2660 ring_size = bnx2x_alloc_rx_bds(fp, rx_ring_size);
2661 if (ring_size < rx_ring_size)
2662 goto alloc_mem_err;
2663 }
2664
2665 return 0;
2666
2667 /* handles low memory cases */
2668 alloc_mem_err:
2669 BNX2X_ERR("Unable to allocate full memory for queue %d (size %d)\n",
2670 index, ring_size);
2671 /* FW will drop all packets if queue is not big enough,
2672 * In these cases we disable the queue
2673 * Min size diferent for TPA and non-TPA queues
2674 */
2675 if (ring_size < (fp->disable_tpa ?
2676 MIN_RX_SIZE_TPA : MIN_RX_SIZE_NONTPA)) {
2677 /* release memory allocated for this queue */
2678 bnx2x_free_fp_mem_at(bp, index);
2679 return -ENOMEM;
2680 }
2681 return 0;
2682 }
2683
2684 int bnx2x_alloc_fp_mem(struct bnx2x *bp)
2685 {
2686 int i;
2687
2688 /**
2689 * 1. Allocate FP for leading - fatal if error
2690 * 2. {CNIC} Allocate FCoE FP - fatal if error
2691 * 3. Allocate RSS - fix number of queues if error
2692 */
2693
2694 /* leading */
2695 if (bnx2x_alloc_fp_mem_at(bp, 0))
2696 return -ENOMEM;
2697 #ifdef BCM_CNIC
2698 /* FCoE */
2699 if (bnx2x_alloc_fp_mem_at(bp, FCOE_IDX))
2700 return -ENOMEM;
2701 #endif
2702 /* RSS */
2703 for_each_nondefault_eth_queue(bp, i)
2704 if (bnx2x_alloc_fp_mem_at(bp, i))
2705 break;
2706
2707 /* handle memory failures */
2708 if (i != BNX2X_NUM_ETH_QUEUES(bp)) {
2709 int delta = BNX2X_NUM_ETH_QUEUES(bp) - i;
2710
2711 WARN_ON(delta < 0);
2712 #ifdef BCM_CNIC
2713 /**
2714 * move non eth FPs next to last eth FP
2715 * must be done in that order
2716 * FCOE_IDX < FWD_IDX < OOO_IDX
2717 */
2718
2719 /* move FCoE fp */
2720 bnx2x_move_fp(bp, FCOE_IDX, FCOE_IDX - delta);
2721 #endif
2722 bp->num_queues -= delta;
2723 BNX2X_ERR("Adjusted num of queues from %d to %d\n",
2724 bp->num_queues + delta, bp->num_queues);
2725 }
2726
2727 return 0;
2728 }
2729
2730 static int bnx2x_setup_irqs(struct bnx2x *bp)
2731 {
2732 int rc = 0;
2733 if (bp->flags & USING_MSIX_FLAG) {
2734 rc = bnx2x_req_msix_irqs(bp);
2735 if (rc)
2736 return rc;
2737 } else {
2738 bnx2x_ack_int(bp);
2739 rc = bnx2x_req_irq(bp);
2740 if (rc) {
2741 BNX2X_ERR("IRQ request failed rc %d, aborting\n", rc);
2742 return rc;
2743 }
2744 if (bp->flags & USING_MSI_FLAG) {
2745 bp->dev->irq = bp->pdev->irq;
2746 netdev_info(bp->dev, "using MSI IRQ %d\n",
2747 bp->pdev->irq);
2748 }
2749 }
2750
2751 return 0;
2752 }
2753
2754 void bnx2x_free_mem_bp(struct bnx2x *bp)
2755 {
2756 kfree(bp->fp);
2757 kfree(bp->msix_table);
2758 kfree(bp->ilt);
2759 }
2760
2761 int __devinit bnx2x_alloc_mem_bp(struct bnx2x *bp)
2762 {
2763 struct bnx2x_fastpath *fp;
2764 struct msix_entry *tbl;
2765 struct bnx2x_ilt *ilt;
2766
2767 /* fp array */
2768 fp = kzalloc(L2_FP_COUNT(bp->l2_cid_count)*sizeof(*fp), GFP_KERNEL);
2769 if (!fp)
2770 goto alloc_err;
2771 bp->fp = fp;
2772
2773 /* msix table */
2774 tbl = kzalloc((FP_SB_COUNT(bp->l2_cid_count) + 1) * sizeof(*tbl),
2775 GFP_KERNEL);
2776 if (!tbl)
2777 goto alloc_err;
2778 bp->msix_table = tbl;
2779
2780 /* ilt */
2781 ilt = kzalloc(sizeof(*ilt), GFP_KERNEL);
2782 if (!ilt)
2783 goto alloc_err;
2784 bp->ilt = ilt;
2785
2786 return 0;
2787 alloc_err:
2788 bnx2x_free_mem_bp(bp);
2789 return -ENOMEM;
2790
2791 }
2792
2793 static int bnx2x_reload_if_running(struct net_device *dev)
2794 {
2795 struct bnx2x *bp = netdev_priv(dev);
2796
2797 if (unlikely(!netif_running(dev)))
2798 return 0;
2799
2800 bnx2x_nic_unload(bp, UNLOAD_NORMAL);
2801 return bnx2x_nic_load(bp, LOAD_NORMAL);
2802 }
2803
2804 /* called with rtnl_lock */
2805 int bnx2x_change_mtu(struct net_device *dev, int new_mtu)
2806 {
2807 struct bnx2x *bp = netdev_priv(dev);
2808
2809 if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
2810 printk(KERN_ERR "Handling parity error recovery. Try again later\n");
2811 return -EAGAIN;
2812 }
2813
2814 if ((new_mtu > ETH_MAX_JUMBO_PACKET_SIZE) ||
2815 ((new_mtu + ETH_HLEN) < ETH_MIN_PACKET_SIZE))
2816 return -EINVAL;
2817
2818 /* This does not race with packet allocation
2819 * because the actual alloc size is
2820 * only updated as part of load
2821 */
2822 dev->mtu = new_mtu;
2823
2824 return bnx2x_reload_if_running(dev);
2825 }
2826
2827 u32 bnx2x_fix_features(struct net_device *dev, u32 features)
2828 {
2829 struct bnx2x *bp = netdev_priv(dev);
2830
2831 /* TPA requires Rx CSUM offloading */
2832 if (!(features & NETIF_F_RXCSUM) || bp->disable_tpa)
2833 features &= ~NETIF_F_LRO;
2834
2835 return features;
2836 }
2837
2838 int bnx2x_set_features(struct net_device *dev, u32 features)
2839 {
2840 struct bnx2x *bp = netdev_priv(dev);
2841 u32 flags = bp->flags;
2842
2843 if (features & NETIF_F_LRO)
2844 flags |= TPA_ENABLE_FLAG;
2845 else
2846 flags &= ~TPA_ENABLE_FLAG;
2847
2848 if (flags ^ bp->flags) {
2849 bp->flags = flags;
2850
2851 if (bp->recovery_state == BNX2X_RECOVERY_DONE)
2852 return bnx2x_reload_if_running(dev);
2853 /* else: bnx2x_nic_load() will be called at end of recovery */
2854 }
2855
2856 return 0;
2857 }
2858
2859 void bnx2x_tx_timeout(struct net_device *dev)
2860 {
2861 struct bnx2x *bp = netdev_priv(dev);
2862
2863 #ifdef BNX2X_STOP_ON_ERROR
2864 if (!bp->panic)
2865 bnx2x_panic();
2866 #endif
2867 /* This allows the netif to be shutdown gracefully before resetting */
2868 schedule_delayed_work(&bp->reset_task, 0);
2869 }
2870
2871 int bnx2x_suspend(struct pci_dev *pdev, pm_message_t state)
2872 {
2873 struct net_device *dev = pci_get_drvdata(pdev);
2874 struct bnx2x *bp;
2875
2876 if (!dev) {
2877 dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
2878 return -ENODEV;
2879 }
2880 bp = netdev_priv(dev);
2881
2882 rtnl_lock();
2883
2884 pci_save_state(pdev);
2885
2886 if (!netif_running(dev)) {
2887 rtnl_unlock();
2888 return 0;
2889 }
2890
2891 netif_device_detach(dev);
2892
2893 bnx2x_nic_unload(bp, UNLOAD_CLOSE);
2894
2895 bnx2x_set_power_state(bp, pci_choose_state(pdev, state));
2896
2897 rtnl_unlock();
2898
2899 return 0;
2900 }
2901
2902 int bnx2x_resume(struct pci_dev *pdev)
2903 {
2904 struct net_device *dev = pci_get_drvdata(pdev);
2905 struct bnx2x *bp;
2906 int rc;
2907
2908 if (!dev) {
2909 dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
2910 return -ENODEV;
2911 }
2912 bp = netdev_priv(dev);
2913
2914 if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
2915 printk(KERN_ERR "Handling parity error recovery. Try again later\n");
2916 return -EAGAIN;
2917 }
2918
2919 rtnl_lock();
2920
2921 pci_restore_state(pdev);
2922
2923 if (!netif_running(dev)) {
2924 rtnl_unlock();
2925 return 0;
2926 }
2927
2928 bnx2x_set_power_state(bp, PCI_D0);
2929 netif_device_attach(dev);
2930
2931 /* Since the chip was reset, clear the FW sequence number */
2932 bp->fw_seq = 0;
2933 rc = bnx2x_nic_load(bp, LOAD_OPEN);
2934
2935 rtnl_unlock();
2936
2937 return rc;
2938 }
This page took 0.151011 seconds and 4 git commands to generate.