Merge branch 'linus' into x86/urgent
[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 <linux/prefetch.h>
25 #include "bnx2x_cmn.h"
26
27 #include "bnx2x_init.h"
28
29 static int bnx2x_setup_irqs(struct bnx2x *bp);
30
31 /**
32 * bnx2x_bz_fp - zero content of the fastpath structure.
33 *
34 * @bp: driver handle
35 * @index: fastpath index to be zeroed
36 *
37 * Makes sure the contents of the bp->fp[index].napi is kept
38 * intact.
39 */
40 static inline void bnx2x_bz_fp(struct bnx2x *bp, int index)
41 {
42 struct bnx2x_fastpath *fp = &bp->fp[index];
43 struct napi_struct orig_napi = fp->napi;
44 /* bzero bnx2x_fastpath contents */
45 memset(fp, 0, sizeof(*fp));
46
47 /* Restore the NAPI object as it has been already initialized */
48 fp->napi = orig_napi;
49 }
50
51 /**
52 * bnx2x_move_fp - move content of the fastpath structure.
53 *
54 * @bp: driver handle
55 * @from: source FP index
56 * @to: destination FP index
57 *
58 * Makes sure the contents of the bp->fp[to].napi is kept
59 * intact.
60 */
61 static inline void bnx2x_move_fp(struct bnx2x *bp, int from, int to)
62 {
63 struct bnx2x_fastpath *from_fp = &bp->fp[from];
64 struct bnx2x_fastpath *to_fp = &bp->fp[to];
65 struct napi_struct orig_napi = to_fp->napi;
66 /* Move bnx2x_fastpath contents */
67 memcpy(to_fp, from_fp, sizeof(*to_fp));
68 to_fp->index = to;
69
70 /* Restore the NAPI object as it has been already initialized */
71 to_fp->napi = orig_napi;
72 }
73
74 /* free skb in the packet ring at pos idx
75 * return idx of last bd freed
76 */
77 static u16 bnx2x_free_tx_pkt(struct bnx2x *bp, struct bnx2x_fastpath *fp,
78 u16 idx)
79 {
80 struct sw_tx_bd *tx_buf = &fp->tx_buf_ring[idx];
81 struct eth_tx_start_bd *tx_start_bd;
82 struct eth_tx_bd *tx_data_bd;
83 struct sk_buff *skb = tx_buf->skb;
84 u16 bd_idx = TX_BD(tx_buf->first_bd), new_cons;
85 int nbd;
86
87 /* prefetch skb end pointer to speedup dev_kfree_skb() */
88 prefetch(&skb->end);
89
90 DP(BNX2X_MSG_OFF, "pkt_idx %d buff @(%p)->skb %p\n",
91 idx, tx_buf, skb);
92
93 /* unmap first bd */
94 DP(BNX2X_MSG_OFF, "free bd_idx %d\n", bd_idx);
95 tx_start_bd = &fp->tx_desc_ring[bd_idx].start_bd;
96 dma_unmap_single(&bp->pdev->dev, BD_UNMAP_ADDR(tx_start_bd),
97 BD_UNMAP_LEN(tx_start_bd), DMA_TO_DEVICE);
98
99 nbd = le16_to_cpu(tx_start_bd->nbd) - 1;
100 #ifdef BNX2X_STOP_ON_ERROR
101 if ((nbd - 1) > (MAX_SKB_FRAGS + 2)) {
102 BNX2X_ERR("BAD nbd!\n");
103 bnx2x_panic();
104 }
105 #endif
106 new_cons = nbd + tx_buf->first_bd;
107
108 /* Get the next bd */
109 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
110
111 /* Skip a parse bd... */
112 --nbd;
113 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
114
115 /* ...and the TSO split header bd since they have no mapping */
116 if (tx_buf->flags & BNX2X_TSO_SPLIT_BD) {
117 --nbd;
118 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
119 }
120
121 /* now free frags */
122 while (nbd > 0) {
123
124 DP(BNX2X_MSG_OFF, "free frag bd_idx %d\n", bd_idx);
125 tx_data_bd = &fp->tx_desc_ring[bd_idx].reg_bd;
126 dma_unmap_page(&bp->pdev->dev, BD_UNMAP_ADDR(tx_data_bd),
127 BD_UNMAP_LEN(tx_data_bd), DMA_TO_DEVICE);
128 if (--nbd)
129 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
130 }
131
132 /* release skb */
133 WARN_ON(!skb);
134 dev_kfree_skb_any(skb);
135 tx_buf->first_bd = 0;
136 tx_buf->skb = NULL;
137
138 return new_cons;
139 }
140
141 int bnx2x_tx_int(struct bnx2x_fastpath *fp)
142 {
143 struct bnx2x *bp = fp->bp;
144 struct netdev_queue *txq;
145 u16 hw_cons, sw_cons, bd_cons = fp->tx_bd_cons;
146
147 #ifdef BNX2X_STOP_ON_ERROR
148 if (unlikely(bp->panic))
149 return -1;
150 #endif
151
152 txq = netdev_get_tx_queue(bp->dev, fp->index);
153 hw_cons = le16_to_cpu(*fp->tx_cons_sb);
154 sw_cons = fp->tx_pkt_cons;
155
156 while (sw_cons != hw_cons) {
157 u16 pkt_cons;
158
159 pkt_cons = TX_BD(sw_cons);
160
161 DP(NETIF_MSG_TX_DONE, "queue[%d]: hw_cons %u sw_cons %u "
162 " pkt_cons %u\n",
163 fp->index, hw_cons, sw_cons, pkt_cons);
164
165 bd_cons = bnx2x_free_tx_pkt(bp, fp, pkt_cons);
166 sw_cons++;
167 }
168
169 fp->tx_pkt_cons = sw_cons;
170 fp->tx_bd_cons = bd_cons;
171
172 /* Need to make the tx_bd_cons update visible to start_xmit()
173 * before checking for netif_tx_queue_stopped(). Without the
174 * memory barrier, there is a small possibility that
175 * start_xmit() will miss it and cause the queue to be stopped
176 * forever.
177 */
178 smp_mb();
179
180 if (unlikely(netif_tx_queue_stopped(txq))) {
181 /* Taking tx_lock() is needed to prevent reenabling the queue
182 * while it's empty. This could have happen if rx_action() gets
183 * suspended in bnx2x_tx_int() after the condition before
184 * netif_tx_wake_queue(), while tx_action (bnx2x_start_xmit()):
185 *
186 * stops the queue->sees fresh tx_bd_cons->releases the queue->
187 * sends some packets consuming the whole queue again->
188 * stops the queue
189 */
190
191 __netif_tx_lock(txq, smp_processor_id());
192
193 if ((netif_tx_queue_stopped(txq)) &&
194 (bp->state == BNX2X_STATE_OPEN) &&
195 (bnx2x_tx_avail(fp) >= MAX_SKB_FRAGS + 3))
196 netif_tx_wake_queue(txq);
197
198 __netif_tx_unlock(txq);
199 }
200 return 0;
201 }
202
203 static inline void bnx2x_update_last_max_sge(struct bnx2x_fastpath *fp,
204 u16 idx)
205 {
206 u16 last_max = fp->last_max_sge;
207
208 if (SUB_S16(idx, last_max) > 0)
209 fp->last_max_sge = idx;
210 }
211
212 static void bnx2x_update_sge_prod(struct bnx2x_fastpath *fp,
213 struct eth_fast_path_rx_cqe *fp_cqe)
214 {
215 struct bnx2x *bp = fp->bp;
216 u16 sge_len = SGE_PAGE_ALIGN(le16_to_cpu(fp_cqe->pkt_len) -
217 le16_to_cpu(fp_cqe->len_on_bd)) >>
218 SGE_PAGE_SHIFT;
219 u16 last_max, last_elem, first_elem;
220 u16 delta = 0;
221 u16 i;
222
223 if (!sge_len)
224 return;
225
226 /* First mark all used pages */
227 for (i = 0; i < sge_len; i++)
228 SGE_MASK_CLEAR_BIT(fp,
229 RX_SGE(le16_to_cpu(fp_cqe->sgl_or_raw_data.sgl[i])));
230
231 DP(NETIF_MSG_RX_STATUS, "fp_cqe->sgl[%d] = %d\n",
232 sge_len - 1, le16_to_cpu(fp_cqe->sgl_or_raw_data.sgl[sge_len - 1]));
233
234 /* Here we assume that the last SGE index is the biggest */
235 prefetch((void *)(fp->sge_mask));
236 bnx2x_update_last_max_sge(fp,
237 le16_to_cpu(fp_cqe->sgl_or_raw_data.sgl[sge_len - 1]));
238
239 last_max = RX_SGE(fp->last_max_sge);
240 last_elem = last_max >> RX_SGE_MASK_ELEM_SHIFT;
241 first_elem = RX_SGE(fp->rx_sge_prod) >> RX_SGE_MASK_ELEM_SHIFT;
242
243 /* If ring is not full */
244 if (last_elem + 1 != first_elem)
245 last_elem++;
246
247 /* Now update the prod */
248 for (i = first_elem; i != last_elem; i = NEXT_SGE_MASK_ELEM(i)) {
249 if (likely(fp->sge_mask[i]))
250 break;
251
252 fp->sge_mask[i] = RX_SGE_MASK_ELEM_ONE_MASK;
253 delta += RX_SGE_MASK_ELEM_SZ;
254 }
255
256 if (delta > 0) {
257 fp->rx_sge_prod += delta;
258 /* clear page-end entries */
259 bnx2x_clear_sge_mask_next_elems(fp);
260 }
261
262 DP(NETIF_MSG_RX_STATUS,
263 "fp->last_max_sge = %d fp->rx_sge_prod = %d\n",
264 fp->last_max_sge, fp->rx_sge_prod);
265 }
266
267 static void bnx2x_tpa_start(struct bnx2x_fastpath *fp, u16 queue,
268 struct sk_buff *skb, u16 cons, u16 prod)
269 {
270 struct bnx2x *bp = fp->bp;
271 struct sw_rx_bd *cons_rx_buf = &fp->rx_buf_ring[cons];
272 struct sw_rx_bd *prod_rx_buf = &fp->rx_buf_ring[prod];
273 struct eth_rx_bd *prod_bd = &fp->rx_desc_ring[prod];
274 dma_addr_t mapping;
275
276 /* move empty skb from pool to prod and map it */
277 prod_rx_buf->skb = fp->tpa_pool[queue].skb;
278 mapping = dma_map_single(&bp->pdev->dev, fp->tpa_pool[queue].skb->data,
279 fp->rx_buf_size, DMA_FROM_DEVICE);
280 dma_unmap_addr_set(prod_rx_buf, mapping, mapping);
281
282 /* move partial skb from cons to pool (don't unmap yet) */
283 fp->tpa_pool[queue] = *cons_rx_buf;
284
285 /* mark bin state as start - print error if current state != stop */
286 if (fp->tpa_state[queue] != BNX2X_TPA_STOP)
287 BNX2X_ERR("start of bin not in stop [%d]\n", queue);
288
289 fp->tpa_state[queue] = BNX2X_TPA_START;
290
291 /* point prod_bd to new skb */
292 prod_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
293 prod_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
294
295 #ifdef BNX2X_STOP_ON_ERROR
296 fp->tpa_queue_used |= (1 << queue);
297 #ifdef _ASM_GENERIC_INT_L64_H
298 DP(NETIF_MSG_RX_STATUS, "fp->tpa_queue_used = 0x%lx\n",
299 #else
300 DP(NETIF_MSG_RX_STATUS, "fp->tpa_queue_used = 0x%llx\n",
301 #endif
302 fp->tpa_queue_used);
303 #endif
304 }
305
306 /* Timestamp option length allowed for TPA aggregation:
307 *
308 * nop nop kind length echo val
309 */
310 #define TPA_TSTAMP_OPT_LEN 12
311 /**
312 * bnx2x_set_lro_mss - calculate the approximate value of the MSS
313 *
314 * @bp: driver handle
315 * @parsing_flags: parsing flags from the START CQE
316 * @len_on_bd: total length of the first packet for the
317 * aggregation.
318 *
319 * Approximate value of the MSS for this aggregation calculated using
320 * the first packet of it.
321 */
322 static inline u16 bnx2x_set_lro_mss(struct bnx2x *bp, u16 parsing_flags,
323 u16 len_on_bd)
324 {
325 /* TPA arrgregation won't have an IP options and TCP options
326 * other than timestamp.
327 */
328 u16 hdrs_len = ETH_HLEN + sizeof(struct iphdr) + sizeof(struct tcphdr);
329
330
331 /* Check if there was a TCP timestamp, if there is it's will
332 * always be 12 bytes length: nop nop kind length echo val.
333 *
334 * Otherwise FW would close the aggregation.
335 */
336 if (parsing_flags & PARSING_FLAGS_TIME_STAMP_EXIST_FLAG)
337 hdrs_len += TPA_TSTAMP_OPT_LEN;
338
339 return len_on_bd - hdrs_len;
340 }
341
342 static int bnx2x_fill_frag_skb(struct bnx2x *bp, struct bnx2x_fastpath *fp,
343 struct sk_buff *skb,
344 struct eth_fast_path_rx_cqe *fp_cqe,
345 u16 cqe_idx, u16 parsing_flags)
346 {
347 struct sw_rx_page *rx_pg, old_rx_pg;
348 u16 len_on_bd = le16_to_cpu(fp_cqe->len_on_bd);
349 u32 i, frag_len, frag_size, pages;
350 int err;
351 int j;
352
353 frag_size = le16_to_cpu(fp_cqe->pkt_len) - len_on_bd;
354 pages = SGE_PAGE_ALIGN(frag_size) >> SGE_PAGE_SHIFT;
355
356 /* This is needed in order to enable forwarding support */
357 if (frag_size)
358 skb_shinfo(skb)->gso_size = bnx2x_set_lro_mss(bp, parsing_flags,
359 len_on_bd);
360
361 #ifdef BNX2X_STOP_ON_ERROR
362 if (pages > min_t(u32, 8, MAX_SKB_FRAGS)*SGE_PAGE_SIZE*PAGES_PER_SGE) {
363 BNX2X_ERR("SGL length is too long: %d. CQE index is %d\n",
364 pages, cqe_idx);
365 BNX2X_ERR("fp_cqe->pkt_len = %d fp_cqe->len_on_bd = %d\n",
366 fp_cqe->pkt_len, len_on_bd);
367 bnx2x_panic();
368 return -EINVAL;
369 }
370 #endif
371
372 /* Run through the SGL and compose the fragmented skb */
373 for (i = 0, j = 0; i < pages; i += PAGES_PER_SGE, j++) {
374 u16 sge_idx =
375 RX_SGE(le16_to_cpu(fp_cqe->sgl_or_raw_data.sgl[j]));
376
377 /* FW gives the indices of the SGE as if the ring is an array
378 (meaning that "next" element will consume 2 indices) */
379 frag_len = min(frag_size, (u32)(SGE_PAGE_SIZE*PAGES_PER_SGE));
380 rx_pg = &fp->rx_page_ring[sge_idx];
381 old_rx_pg = *rx_pg;
382
383 /* If we fail to allocate a substitute page, we simply stop
384 where we are and drop the whole packet */
385 err = bnx2x_alloc_rx_sge(bp, fp, sge_idx);
386 if (unlikely(err)) {
387 fp->eth_q_stats.rx_skb_alloc_failed++;
388 return err;
389 }
390
391 /* Unmap the page as we r going to pass it to the stack */
392 dma_unmap_page(&bp->pdev->dev,
393 dma_unmap_addr(&old_rx_pg, mapping),
394 SGE_PAGE_SIZE*PAGES_PER_SGE, DMA_FROM_DEVICE);
395
396 /* Add one frag and update the appropriate fields in the skb */
397 skb_fill_page_desc(skb, j, old_rx_pg.page, 0, frag_len);
398
399 skb->data_len += frag_len;
400 skb->truesize += frag_len;
401 skb->len += frag_len;
402
403 frag_size -= frag_len;
404 }
405
406 return 0;
407 }
408
409 static void bnx2x_tpa_stop(struct bnx2x *bp, struct bnx2x_fastpath *fp,
410 u16 queue, int pad, int len, union eth_rx_cqe *cqe,
411 u16 cqe_idx)
412 {
413 struct sw_rx_bd *rx_buf = &fp->tpa_pool[queue];
414 struct sk_buff *skb = rx_buf->skb;
415 /* alloc new skb */
416 struct sk_buff *new_skb = netdev_alloc_skb(bp->dev, fp->rx_buf_size);
417
418 /* Unmap skb in the pool anyway, as we are going to change
419 pool entry status to BNX2X_TPA_STOP even if new skb allocation
420 fails. */
421 dma_unmap_single(&bp->pdev->dev, dma_unmap_addr(rx_buf, mapping),
422 fp->rx_buf_size, DMA_FROM_DEVICE);
423
424 if (likely(new_skb)) {
425 /* fix ip xsum and give it to the stack */
426 /* (no need to map the new skb) */
427 u16 parsing_flags =
428 le16_to_cpu(cqe->fast_path_cqe.pars_flags.flags);
429
430 prefetch(skb);
431 prefetch(((char *)(skb)) + L1_CACHE_BYTES);
432
433 #ifdef BNX2X_STOP_ON_ERROR
434 if (pad + len > fp->rx_buf_size) {
435 BNX2X_ERR("skb_put is about to fail... "
436 "pad %d len %d rx_buf_size %d\n",
437 pad, len, fp->rx_buf_size);
438 bnx2x_panic();
439 return;
440 }
441 #endif
442
443 skb_reserve(skb, pad);
444 skb_put(skb, len);
445
446 skb->protocol = eth_type_trans(skb, bp->dev);
447 skb->ip_summed = CHECKSUM_UNNECESSARY;
448
449 {
450 struct iphdr *iph;
451
452 iph = (struct iphdr *)skb->data;
453 iph->check = 0;
454 iph->check = ip_fast_csum((u8 *)iph, iph->ihl);
455 }
456
457 if (!bnx2x_fill_frag_skb(bp, fp, skb,
458 &cqe->fast_path_cqe, cqe_idx,
459 parsing_flags)) {
460 if (parsing_flags & PARSING_FLAGS_VLAN)
461 __vlan_hwaccel_put_tag(skb,
462 le16_to_cpu(cqe->fast_path_cqe.
463 vlan_tag));
464 napi_gro_receive(&fp->napi, skb);
465 } else {
466 DP(NETIF_MSG_RX_STATUS, "Failed to allocate new pages"
467 " - dropping packet!\n");
468 dev_kfree_skb_any(skb);
469 }
470
471
472 /* put new skb in bin */
473 fp->tpa_pool[queue].skb = new_skb;
474
475 } else {
476 /* else drop the packet and keep the buffer in the bin */
477 DP(NETIF_MSG_RX_STATUS,
478 "Failed to allocate new skb - dropping packet!\n");
479 fp->eth_q_stats.rx_skb_alloc_failed++;
480 }
481
482 fp->tpa_state[queue] = BNX2X_TPA_STOP;
483 }
484
485 /* Set Toeplitz hash value in the skb using the value from the
486 * CQE (calculated by HW).
487 */
488 static inline void bnx2x_set_skb_rxhash(struct bnx2x *bp, union eth_rx_cqe *cqe,
489 struct sk_buff *skb)
490 {
491 /* Set Toeplitz hash from CQE */
492 if ((bp->dev->features & NETIF_F_RXHASH) &&
493 (cqe->fast_path_cqe.status_flags &
494 ETH_FAST_PATH_RX_CQE_RSS_HASH_FLG))
495 skb->rxhash =
496 le32_to_cpu(cqe->fast_path_cqe.rss_hash_result);
497 }
498
499 int bnx2x_rx_int(struct bnx2x_fastpath *fp, int budget)
500 {
501 struct bnx2x *bp = fp->bp;
502 u16 bd_cons, bd_prod, bd_prod_fw, comp_ring_cons;
503 u16 hw_comp_cons, sw_comp_cons, sw_comp_prod;
504 int rx_pkt = 0;
505
506 #ifdef BNX2X_STOP_ON_ERROR
507 if (unlikely(bp->panic))
508 return 0;
509 #endif
510
511 /* CQ "next element" is of the size of the regular element,
512 that's why it's ok here */
513 hw_comp_cons = le16_to_cpu(*fp->rx_cons_sb);
514 if ((hw_comp_cons & MAX_RCQ_DESC_CNT) == MAX_RCQ_DESC_CNT)
515 hw_comp_cons++;
516
517 bd_cons = fp->rx_bd_cons;
518 bd_prod = fp->rx_bd_prod;
519 bd_prod_fw = bd_prod;
520 sw_comp_cons = fp->rx_comp_cons;
521 sw_comp_prod = fp->rx_comp_prod;
522
523 /* Memory barrier necessary as speculative reads of the rx
524 * buffer can be ahead of the index in the status block
525 */
526 rmb();
527
528 DP(NETIF_MSG_RX_STATUS,
529 "queue[%d]: hw_comp_cons %u sw_comp_cons %u\n",
530 fp->index, hw_comp_cons, sw_comp_cons);
531
532 while (sw_comp_cons != hw_comp_cons) {
533 struct sw_rx_bd *rx_buf = NULL;
534 struct sk_buff *skb;
535 union eth_rx_cqe *cqe;
536 u8 cqe_fp_flags;
537 u16 len, pad;
538
539 comp_ring_cons = RCQ_BD(sw_comp_cons);
540 bd_prod = RX_BD(bd_prod);
541 bd_cons = RX_BD(bd_cons);
542
543 /* Prefetch the page containing the BD descriptor
544 at producer's index. It will be needed when new skb is
545 allocated */
546 prefetch((void *)(PAGE_ALIGN((unsigned long)
547 (&fp->rx_desc_ring[bd_prod])) -
548 PAGE_SIZE + 1));
549
550 cqe = &fp->rx_comp_ring[comp_ring_cons];
551 cqe_fp_flags = cqe->fast_path_cqe.type_error_flags;
552
553 DP(NETIF_MSG_RX_STATUS, "CQE type %x err %x status %x"
554 " queue %x vlan %x len %u\n", CQE_TYPE(cqe_fp_flags),
555 cqe_fp_flags, cqe->fast_path_cqe.status_flags,
556 le32_to_cpu(cqe->fast_path_cqe.rss_hash_result),
557 le16_to_cpu(cqe->fast_path_cqe.vlan_tag),
558 le16_to_cpu(cqe->fast_path_cqe.pkt_len));
559
560 /* is this a slowpath msg? */
561 if (unlikely(CQE_TYPE(cqe_fp_flags))) {
562 bnx2x_sp_event(fp, cqe);
563 goto next_cqe;
564
565 /* this is an rx packet */
566 } else {
567 rx_buf = &fp->rx_buf_ring[bd_cons];
568 skb = rx_buf->skb;
569 prefetch(skb);
570 len = le16_to_cpu(cqe->fast_path_cqe.pkt_len);
571 pad = cqe->fast_path_cqe.placement_offset;
572
573 /* - If CQE is marked both TPA_START and TPA_END it is
574 * a non-TPA CQE.
575 * - FP CQE will always have either TPA_START or/and
576 * TPA_STOP flags set.
577 */
578 if ((!fp->disable_tpa) &&
579 (TPA_TYPE(cqe_fp_flags) !=
580 (TPA_TYPE_START | TPA_TYPE_END))) {
581 u16 queue = cqe->fast_path_cqe.queue_index;
582
583 if (TPA_TYPE(cqe_fp_flags) == TPA_TYPE_START) {
584 DP(NETIF_MSG_RX_STATUS,
585 "calling tpa_start on queue %d\n",
586 queue);
587
588 bnx2x_tpa_start(fp, queue, skb,
589 bd_cons, bd_prod);
590
591 /* Set Toeplitz hash for an LRO skb */
592 bnx2x_set_skb_rxhash(bp, cqe, skb);
593
594 goto next_rx;
595 } else { /* TPA_STOP */
596 DP(NETIF_MSG_RX_STATUS,
597 "calling tpa_stop on queue %d\n",
598 queue);
599
600 if (!BNX2X_RX_SUM_FIX(cqe))
601 BNX2X_ERR("STOP on none TCP "
602 "data\n");
603
604 /* This is a size of the linear data
605 on this skb */
606 len = le16_to_cpu(cqe->fast_path_cqe.
607 len_on_bd);
608 bnx2x_tpa_stop(bp, fp, queue, pad,
609 len, cqe, comp_ring_cons);
610 #ifdef BNX2X_STOP_ON_ERROR
611 if (bp->panic)
612 return 0;
613 #endif
614
615 bnx2x_update_sge_prod(fp,
616 &cqe->fast_path_cqe);
617 goto next_cqe;
618 }
619 }
620
621 dma_sync_single_for_device(&bp->pdev->dev,
622 dma_unmap_addr(rx_buf, mapping),
623 pad + RX_COPY_THRESH,
624 DMA_FROM_DEVICE);
625 prefetch(((char *)(skb)) + L1_CACHE_BYTES);
626
627 /* is this an error packet? */
628 if (unlikely(cqe_fp_flags & ETH_RX_ERROR_FALGS)) {
629 DP(NETIF_MSG_RX_ERR,
630 "ERROR flags %x rx packet %u\n",
631 cqe_fp_flags, sw_comp_cons);
632 fp->eth_q_stats.rx_err_discard_pkt++;
633 goto reuse_rx;
634 }
635
636 /* Since we don't have a jumbo ring
637 * copy small packets if mtu > 1500
638 */
639 if ((bp->dev->mtu > ETH_MAX_PACKET_SIZE) &&
640 (len <= RX_COPY_THRESH)) {
641 struct sk_buff *new_skb;
642
643 new_skb = netdev_alloc_skb(bp->dev,
644 len + pad);
645 if (new_skb == NULL) {
646 DP(NETIF_MSG_RX_ERR,
647 "ERROR packet dropped "
648 "because of alloc failure\n");
649 fp->eth_q_stats.rx_skb_alloc_failed++;
650 goto reuse_rx;
651 }
652
653 /* aligned copy */
654 skb_copy_from_linear_data_offset(skb, pad,
655 new_skb->data + pad, len);
656 skb_reserve(new_skb, pad);
657 skb_put(new_skb, len);
658
659 bnx2x_reuse_rx_skb(fp, bd_cons, bd_prod);
660
661 skb = new_skb;
662
663 } else
664 if (likely(bnx2x_alloc_rx_skb(bp, fp, bd_prod) == 0)) {
665 dma_unmap_single(&bp->pdev->dev,
666 dma_unmap_addr(rx_buf, mapping),
667 fp->rx_buf_size,
668 DMA_FROM_DEVICE);
669 skb_reserve(skb, pad);
670 skb_put(skb, len);
671
672 } else {
673 DP(NETIF_MSG_RX_ERR,
674 "ERROR packet dropped because "
675 "of alloc failure\n");
676 fp->eth_q_stats.rx_skb_alloc_failed++;
677 reuse_rx:
678 bnx2x_reuse_rx_skb(fp, bd_cons, bd_prod);
679 goto next_rx;
680 }
681
682 skb->protocol = eth_type_trans(skb, bp->dev);
683
684 /* Set Toeplitz hash for a none-LRO skb */
685 bnx2x_set_skb_rxhash(bp, cqe, skb);
686
687 skb_checksum_none_assert(skb);
688
689 if (bp->dev->features & NETIF_F_RXCSUM) {
690 if (likely(BNX2X_RX_CSUM_OK(cqe)))
691 skb->ip_summed = CHECKSUM_UNNECESSARY;
692 else
693 fp->eth_q_stats.hw_csum_err++;
694 }
695 }
696
697 skb_record_rx_queue(skb, fp->index);
698
699 if (le16_to_cpu(cqe->fast_path_cqe.pars_flags.flags) &
700 PARSING_FLAGS_VLAN)
701 __vlan_hwaccel_put_tag(skb,
702 le16_to_cpu(cqe->fast_path_cqe.vlan_tag));
703 napi_gro_receive(&fp->napi, skb);
704
705
706 next_rx:
707 rx_buf->skb = NULL;
708
709 bd_cons = NEXT_RX_IDX(bd_cons);
710 bd_prod = NEXT_RX_IDX(bd_prod);
711 bd_prod_fw = NEXT_RX_IDX(bd_prod_fw);
712 rx_pkt++;
713 next_cqe:
714 sw_comp_prod = NEXT_RCQ_IDX(sw_comp_prod);
715 sw_comp_cons = NEXT_RCQ_IDX(sw_comp_cons);
716
717 if (rx_pkt == budget)
718 break;
719 } /* while */
720
721 fp->rx_bd_cons = bd_cons;
722 fp->rx_bd_prod = bd_prod_fw;
723 fp->rx_comp_cons = sw_comp_cons;
724 fp->rx_comp_prod = sw_comp_prod;
725
726 /* Update producers */
727 bnx2x_update_rx_prod(bp, fp, bd_prod_fw, sw_comp_prod,
728 fp->rx_sge_prod);
729
730 fp->rx_pkt += rx_pkt;
731 fp->rx_calls++;
732
733 return rx_pkt;
734 }
735
736 static irqreturn_t bnx2x_msix_fp_int(int irq, void *fp_cookie)
737 {
738 struct bnx2x_fastpath *fp = fp_cookie;
739 struct bnx2x *bp = fp->bp;
740
741 /* Return here if interrupt is disabled */
742 if (unlikely(atomic_read(&bp->intr_sem) != 0)) {
743 DP(NETIF_MSG_INTR, "called but intr_sem not 0, returning\n");
744 return IRQ_HANDLED;
745 }
746
747 DP(BNX2X_MSG_FP, "got an MSI-X interrupt on IDX:SB "
748 "[fp %d fw_sd %d igusb %d]\n",
749 fp->index, fp->fw_sb_id, fp->igu_sb_id);
750 bnx2x_ack_sb(bp, fp->igu_sb_id, USTORM_ID, 0, IGU_INT_DISABLE, 0);
751
752 #ifdef BNX2X_STOP_ON_ERROR
753 if (unlikely(bp->panic))
754 return IRQ_HANDLED;
755 #endif
756
757 /* Handle Rx and Tx according to MSI-X vector */
758 prefetch(fp->rx_cons_sb);
759 prefetch(fp->tx_cons_sb);
760 prefetch(&fp->sb_running_index[SM_RX_ID]);
761 napi_schedule(&bnx2x_fp(bp, fp->index, napi));
762
763 return IRQ_HANDLED;
764 }
765
766 /* HW Lock for shared dual port PHYs */
767 void bnx2x_acquire_phy_lock(struct bnx2x *bp)
768 {
769 mutex_lock(&bp->port.phy_mutex);
770
771 if (bp->port.need_hw_lock)
772 bnx2x_acquire_hw_lock(bp, HW_LOCK_RESOURCE_MDIO);
773 }
774
775 void bnx2x_release_phy_lock(struct bnx2x *bp)
776 {
777 if (bp->port.need_hw_lock)
778 bnx2x_release_hw_lock(bp, HW_LOCK_RESOURCE_MDIO);
779
780 mutex_unlock(&bp->port.phy_mutex);
781 }
782
783 /* calculates MF speed according to current linespeed and MF configuration */
784 u16 bnx2x_get_mf_speed(struct bnx2x *bp)
785 {
786 u16 line_speed = bp->link_vars.line_speed;
787 if (IS_MF(bp)) {
788 u16 maxCfg = bnx2x_extract_max_cfg(bp,
789 bp->mf_config[BP_VN(bp)]);
790
791 /* Calculate the current MAX line speed limit for the MF
792 * devices
793 */
794 if (IS_MF_SI(bp))
795 line_speed = (line_speed * maxCfg) / 100;
796 else { /* SD mode */
797 u16 vn_max_rate = maxCfg * 100;
798
799 if (vn_max_rate < line_speed)
800 line_speed = vn_max_rate;
801 }
802 }
803
804 return line_speed;
805 }
806
807 /**
808 * bnx2x_fill_report_data - fill link report data to report
809 *
810 * @bp: driver handle
811 * @data: link state to update
812 *
813 * It uses a none-atomic bit operations because is called under the mutex.
814 */
815 static inline void bnx2x_fill_report_data(struct bnx2x *bp,
816 struct bnx2x_link_report_data *data)
817 {
818 u16 line_speed = bnx2x_get_mf_speed(bp);
819
820 memset(data, 0, sizeof(*data));
821
822 /* Fill the report data: efective line speed */
823 data->line_speed = line_speed;
824
825 /* Link is down */
826 if (!bp->link_vars.link_up || (bp->flags & MF_FUNC_DIS))
827 __set_bit(BNX2X_LINK_REPORT_LINK_DOWN,
828 &data->link_report_flags);
829
830 /* Full DUPLEX */
831 if (bp->link_vars.duplex == DUPLEX_FULL)
832 __set_bit(BNX2X_LINK_REPORT_FD, &data->link_report_flags);
833
834 /* Rx Flow Control is ON */
835 if (bp->link_vars.flow_ctrl & BNX2X_FLOW_CTRL_RX)
836 __set_bit(BNX2X_LINK_REPORT_RX_FC_ON, &data->link_report_flags);
837
838 /* Tx Flow Control is ON */
839 if (bp->link_vars.flow_ctrl & BNX2X_FLOW_CTRL_TX)
840 __set_bit(BNX2X_LINK_REPORT_TX_FC_ON, &data->link_report_flags);
841 }
842
843 /**
844 * bnx2x_link_report - report link status to OS.
845 *
846 * @bp: driver handle
847 *
848 * Calls the __bnx2x_link_report() under the same locking scheme
849 * as a link/PHY state managing code to ensure a consistent link
850 * reporting.
851 */
852
853 void bnx2x_link_report(struct bnx2x *bp)
854 {
855 bnx2x_acquire_phy_lock(bp);
856 __bnx2x_link_report(bp);
857 bnx2x_release_phy_lock(bp);
858 }
859
860 /**
861 * __bnx2x_link_report - report link status to OS.
862 *
863 * @bp: driver handle
864 *
865 * None atomic inmlementation.
866 * Should be called under the phy_lock.
867 */
868 void __bnx2x_link_report(struct bnx2x *bp)
869 {
870 struct bnx2x_link_report_data cur_data;
871
872 /* reread mf_cfg */
873 if (!CHIP_IS_E1(bp))
874 bnx2x_read_mf_cfg(bp);
875
876 /* Read the current link report info */
877 bnx2x_fill_report_data(bp, &cur_data);
878
879 /* Don't report link down or exactly the same link status twice */
880 if (!memcmp(&cur_data, &bp->last_reported_link, sizeof(cur_data)) ||
881 (test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
882 &bp->last_reported_link.link_report_flags) &&
883 test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
884 &cur_data.link_report_flags)))
885 return;
886
887 bp->link_cnt++;
888
889 /* We are going to report a new link parameters now -
890 * remember the current data for the next time.
891 */
892 memcpy(&bp->last_reported_link, &cur_data, sizeof(cur_data));
893
894 if (test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
895 &cur_data.link_report_flags)) {
896 netif_carrier_off(bp->dev);
897 netdev_err(bp->dev, "NIC Link is Down\n");
898 return;
899 } else {
900 netif_carrier_on(bp->dev);
901 netdev_info(bp->dev, "NIC Link is Up, ");
902 pr_cont("%d Mbps ", cur_data.line_speed);
903
904 if (test_and_clear_bit(BNX2X_LINK_REPORT_FD,
905 &cur_data.link_report_flags))
906 pr_cont("full duplex");
907 else
908 pr_cont("half duplex");
909
910 /* Handle the FC at the end so that only these flags would be
911 * possibly set. This way we may easily check if there is no FC
912 * enabled.
913 */
914 if (cur_data.link_report_flags) {
915 if (test_bit(BNX2X_LINK_REPORT_RX_FC_ON,
916 &cur_data.link_report_flags)) {
917 pr_cont(", receive ");
918 if (test_bit(BNX2X_LINK_REPORT_TX_FC_ON,
919 &cur_data.link_report_flags))
920 pr_cont("& transmit ");
921 } else {
922 pr_cont(", transmit ");
923 }
924 pr_cont("flow control ON");
925 }
926 pr_cont("\n");
927 }
928 }
929
930 void bnx2x_init_rx_rings(struct bnx2x *bp)
931 {
932 int func = BP_FUNC(bp);
933 int max_agg_queues = CHIP_IS_E1(bp) ? ETH_MAX_AGGREGATION_QUEUES_E1 :
934 ETH_MAX_AGGREGATION_QUEUES_E1H;
935 u16 ring_prod;
936 int i, j;
937
938 /* Allocate TPA resources */
939 for_each_rx_queue(bp, j) {
940 struct bnx2x_fastpath *fp = &bp->fp[j];
941
942 DP(NETIF_MSG_IFUP,
943 "mtu %d rx_buf_size %d\n", bp->dev->mtu, fp->rx_buf_size);
944
945 if (!fp->disable_tpa) {
946 /* Fill the per-aggregation pool */
947 for (i = 0; i < max_agg_queues; i++) {
948 fp->tpa_pool[i].skb =
949 netdev_alloc_skb(bp->dev, fp->rx_buf_size);
950 if (!fp->tpa_pool[i].skb) {
951 BNX2X_ERR("Failed to allocate TPA "
952 "skb pool for queue[%d] - "
953 "disabling TPA on this "
954 "queue!\n", j);
955 bnx2x_free_tpa_pool(bp, fp, i);
956 fp->disable_tpa = 1;
957 break;
958 }
959 dma_unmap_addr_set((struct sw_rx_bd *)
960 &bp->fp->tpa_pool[i],
961 mapping, 0);
962 fp->tpa_state[i] = BNX2X_TPA_STOP;
963 }
964
965 /* "next page" elements initialization */
966 bnx2x_set_next_page_sgl(fp);
967
968 /* set SGEs bit mask */
969 bnx2x_init_sge_ring_bit_mask(fp);
970
971 /* Allocate SGEs and initialize the ring elements */
972 for (i = 0, ring_prod = 0;
973 i < MAX_RX_SGE_CNT*NUM_RX_SGE_PAGES; i++) {
974
975 if (bnx2x_alloc_rx_sge(bp, fp, ring_prod) < 0) {
976 BNX2X_ERR("was only able to allocate "
977 "%d rx sges\n", i);
978 BNX2X_ERR("disabling TPA for"
979 " queue[%d]\n", j);
980 /* Cleanup already allocated elements */
981 bnx2x_free_rx_sge_range(bp,
982 fp, ring_prod);
983 bnx2x_free_tpa_pool(bp,
984 fp, max_agg_queues);
985 fp->disable_tpa = 1;
986 ring_prod = 0;
987 break;
988 }
989 ring_prod = NEXT_SGE_IDX(ring_prod);
990 }
991
992 fp->rx_sge_prod = ring_prod;
993 }
994 }
995
996 for_each_rx_queue(bp, j) {
997 struct bnx2x_fastpath *fp = &bp->fp[j];
998
999 fp->rx_bd_cons = 0;
1000
1001 /* Activate BD ring */
1002 /* Warning!
1003 * this will generate an interrupt (to the TSTORM)
1004 * must only be done after chip is initialized
1005 */
1006 bnx2x_update_rx_prod(bp, fp, fp->rx_bd_prod, fp->rx_comp_prod,
1007 fp->rx_sge_prod);
1008
1009 if (j != 0)
1010 continue;
1011
1012 if (!CHIP_IS_E2(bp)) {
1013 REG_WR(bp, BAR_USTRORM_INTMEM +
1014 USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func),
1015 U64_LO(fp->rx_comp_mapping));
1016 REG_WR(bp, BAR_USTRORM_INTMEM +
1017 USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func) + 4,
1018 U64_HI(fp->rx_comp_mapping));
1019 }
1020 }
1021 }
1022
1023 static void bnx2x_free_tx_skbs(struct bnx2x *bp)
1024 {
1025 int i;
1026
1027 for_each_tx_queue(bp, i) {
1028 struct bnx2x_fastpath *fp = &bp->fp[i];
1029
1030 u16 bd_cons = fp->tx_bd_cons;
1031 u16 sw_prod = fp->tx_pkt_prod;
1032 u16 sw_cons = fp->tx_pkt_cons;
1033
1034 while (sw_cons != sw_prod) {
1035 bd_cons = bnx2x_free_tx_pkt(bp, fp, TX_BD(sw_cons));
1036 sw_cons++;
1037 }
1038 }
1039 }
1040
1041 static void bnx2x_free_rx_bds(struct bnx2x_fastpath *fp)
1042 {
1043 struct bnx2x *bp = fp->bp;
1044 int i;
1045
1046 /* ring wasn't allocated */
1047 if (fp->rx_buf_ring == NULL)
1048 return;
1049
1050 for (i = 0; i < NUM_RX_BD; i++) {
1051 struct sw_rx_bd *rx_buf = &fp->rx_buf_ring[i];
1052 struct sk_buff *skb = rx_buf->skb;
1053
1054 if (skb == NULL)
1055 continue;
1056
1057 dma_unmap_single(&bp->pdev->dev,
1058 dma_unmap_addr(rx_buf, mapping),
1059 fp->rx_buf_size, DMA_FROM_DEVICE);
1060
1061 rx_buf->skb = NULL;
1062 dev_kfree_skb(skb);
1063 }
1064 }
1065
1066 static void bnx2x_free_rx_skbs(struct bnx2x *bp)
1067 {
1068 int j;
1069
1070 for_each_rx_queue(bp, j) {
1071 struct bnx2x_fastpath *fp = &bp->fp[j];
1072
1073 bnx2x_free_rx_bds(fp);
1074
1075 if (!fp->disable_tpa)
1076 bnx2x_free_tpa_pool(bp, fp, CHIP_IS_E1(bp) ?
1077 ETH_MAX_AGGREGATION_QUEUES_E1 :
1078 ETH_MAX_AGGREGATION_QUEUES_E1H);
1079 }
1080 }
1081
1082 void bnx2x_free_skbs(struct bnx2x *bp)
1083 {
1084 bnx2x_free_tx_skbs(bp);
1085 bnx2x_free_rx_skbs(bp);
1086 }
1087
1088 void bnx2x_update_max_mf_config(struct bnx2x *bp, u32 value)
1089 {
1090 /* load old values */
1091 u32 mf_cfg = bp->mf_config[BP_VN(bp)];
1092
1093 if (value != bnx2x_extract_max_cfg(bp, mf_cfg)) {
1094 /* leave all but MAX value */
1095 mf_cfg &= ~FUNC_MF_CFG_MAX_BW_MASK;
1096
1097 /* set new MAX value */
1098 mf_cfg |= (value << FUNC_MF_CFG_MAX_BW_SHIFT)
1099 & FUNC_MF_CFG_MAX_BW_MASK;
1100
1101 bnx2x_fw_command(bp, DRV_MSG_CODE_SET_MF_BW, mf_cfg);
1102 }
1103 }
1104
1105 static void bnx2x_free_msix_irqs(struct bnx2x *bp)
1106 {
1107 int i, offset = 1;
1108
1109 free_irq(bp->msix_table[0].vector, bp->dev);
1110 DP(NETIF_MSG_IFDOWN, "released sp irq (%d)\n",
1111 bp->msix_table[0].vector);
1112
1113 #ifdef BCM_CNIC
1114 offset++;
1115 #endif
1116 for_each_eth_queue(bp, i) {
1117 DP(NETIF_MSG_IFDOWN, "about to release fp #%d->%d irq "
1118 "state %x\n", i, bp->msix_table[i + offset].vector,
1119 bnx2x_fp(bp, i, state));
1120
1121 free_irq(bp->msix_table[i + offset].vector, &bp->fp[i]);
1122 }
1123 }
1124
1125 void bnx2x_free_irq(struct bnx2x *bp)
1126 {
1127 if (bp->flags & USING_MSIX_FLAG)
1128 bnx2x_free_msix_irqs(bp);
1129 else if (bp->flags & USING_MSI_FLAG)
1130 free_irq(bp->pdev->irq, bp->dev);
1131 else
1132 free_irq(bp->pdev->irq, bp->dev);
1133 }
1134
1135 int bnx2x_enable_msix(struct bnx2x *bp)
1136 {
1137 int msix_vec = 0, i, rc, req_cnt;
1138
1139 bp->msix_table[msix_vec].entry = msix_vec;
1140 DP(NETIF_MSG_IFUP, "msix_table[0].entry = %d (slowpath)\n",
1141 bp->msix_table[0].entry);
1142 msix_vec++;
1143
1144 #ifdef BCM_CNIC
1145 bp->msix_table[msix_vec].entry = msix_vec;
1146 DP(NETIF_MSG_IFUP, "msix_table[%d].entry = %d (CNIC)\n",
1147 bp->msix_table[msix_vec].entry, bp->msix_table[msix_vec].entry);
1148 msix_vec++;
1149 #endif
1150 for_each_eth_queue(bp, i) {
1151 bp->msix_table[msix_vec].entry = msix_vec;
1152 DP(NETIF_MSG_IFUP, "msix_table[%d].entry = %d "
1153 "(fastpath #%u)\n", msix_vec, msix_vec, i);
1154 msix_vec++;
1155 }
1156
1157 req_cnt = BNX2X_NUM_ETH_QUEUES(bp) + CNIC_CONTEXT_USE + 1;
1158
1159 rc = pci_enable_msix(bp->pdev, &bp->msix_table[0], req_cnt);
1160
1161 /*
1162 * reconfigure number of tx/rx queues according to available
1163 * MSI-X vectors
1164 */
1165 if (rc >= BNX2X_MIN_MSIX_VEC_CNT) {
1166 /* how less vectors we will have? */
1167 int diff = req_cnt - rc;
1168
1169 DP(NETIF_MSG_IFUP,
1170 "Trying to use less MSI-X vectors: %d\n", rc);
1171
1172 rc = pci_enable_msix(bp->pdev, &bp->msix_table[0], rc);
1173
1174 if (rc) {
1175 DP(NETIF_MSG_IFUP,
1176 "MSI-X is not attainable rc %d\n", rc);
1177 return rc;
1178 }
1179 /*
1180 * decrease number of queues by number of unallocated entries
1181 */
1182 bp->num_queues -= diff;
1183
1184 DP(NETIF_MSG_IFUP, "New queue configuration set: %d\n",
1185 bp->num_queues);
1186 } else if (rc) {
1187 /* fall to INTx if not enough memory */
1188 if (rc == -ENOMEM)
1189 bp->flags |= DISABLE_MSI_FLAG;
1190 DP(NETIF_MSG_IFUP, "MSI-X is not attainable rc %d\n", rc);
1191 return rc;
1192 }
1193
1194 bp->flags |= USING_MSIX_FLAG;
1195
1196 return 0;
1197 }
1198
1199 static int bnx2x_req_msix_irqs(struct bnx2x *bp)
1200 {
1201 int i, rc, offset = 1;
1202
1203 rc = request_irq(bp->msix_table[0].vector, bnx2x_msix_sp_int, 0,
1204 bp->dev->name, bp->dev);
1205 if (rc) {
1206 BNX2X_ERR("request sp irq failed\n");
1207 return -EBUSY;
1208 }
1209
1210 #ifdef BCM_CNIC
1211 offset++;
1212 #endif
1213 for_each_eth_queue(bp, i) {
1214 struct bnx2x_fastpath *fp = &bp->fp[i];
1215 snprintf(fp->name, sizeof(fp->name), "%s-fp-%d",
1216 bp->dev->name, i);
1217
1218 rc = request_irq(bp->msix_table[offset].vector,
1219 bnx2x_msix_fp_int, 0, fp->name, fp);
1220 if (rc) {
1221 BNX2X_ERR("request fp #%d irq failed rc %d\n", i, rc);
1222 bnx2x_free_msix_irqs(bp);
1223 return -EBUSY;
1224 }
1225
1226 offset++;
1227 fp->state = BNX2X_FP_STATE_IRQ;
1228 }
1229
1230 i = BNX2X_NUM_ETH_QUEUES(bp);
1231 offset = 1 + CNIC_CONTEXT_USE;
1232 netdev_info(bp->dev, "using MSI-X IRQs: sp %d fp[%d] %d"
1233 " ... fp[%d] %d\n",
1234 bp->msix_table[0].vector,
1235 0, bp->msix_table[offset].vector,
1236 i - 1, bp->msix_table[offset + i - 1].vector);
1237
1238 return 0;
1239 }
1240
1241 int bnx2x_enable_msi(struct bnx2x *bp)
1242 {
1243 int rc;
1244
1245 rc = pci_enable_msi(bp->pdev);
1246 if (rc) {
1247 DP(NETIF_MSG_IFUP, "MSI is not attainable\n");
1248 return -1;
1249 }
1250 bp->flags |= USING_MSI_FLAG;
1251
1252 return 0;
1253 }
1254
1255 static int bnx2x_req_irq(struct bnx2x *bp)
1256 {
1257 unsigned long flags;
1258 int rc;
1259
1260 if (bp->flags & USING_MSI_FLAG)
1261 flags = 0;
1262 else
1263 flags = IRQF_SHARED;
1264
1265 rc = request_irq(bp->pdev->irq, bnx2x_interrupt, flags,
1266 bp->dev->name, bp->dev);
1267 if (!rc)
1268 bnx2x_fp(bp, 0, state) = BNX2X_FP_STATE_IRQ;
1269
1270 return rc;
1271 }
1272
1273 static void bnx2x_napi_enable(struct bnx2x *bp)
1274 {
1275 int i;
1276
1277 for_each_napi_queue(bp, i)
1278 napi_enable(&bnx2x_fp(bp, i, napi));
1279 }
1280
1281 static void bnx2x_napi_disable(struct bnx2x *bp)
1282 {
1283 int i;
1284
1285 for_each_napi_queue(bp, i)
1286 napi_disable(&bnx2x_fp(bp, i, napi));
1287 }
1288
1289 void bnx2x_netif_start(struct bnx2x *bp)
1290 {
1291 int intr_sem;
1292
1293 intr_sem = atomic_dec_and_test(&bp->intr_sem);
1294 smp_wmb(); /* Ensure that bp->intr_sem update is SMP-safe */
1295
1296 if (intr_sem) {
1297 if (netif_running(bp->dev)) {
1298 bnx2x_napi_enable(bp);
1299 bnx2x_int_enable(bp);
1300 if (bp->state == BNX2X_STATE_OPEN)
1301 netif_tx_wake_all_queues(bp->dev);
1302 }
1303 }
1304 }
1305
1306 void bnx2x_netif_stop(struct bnx2x *bp, int disable_hw)
1307 {
1308 bnx2x_int_disable_sync(bp, disable_hw);
1309 bnx2x_napi_disable(bp);
1310 netif_tx_disable(bp->dev);
1311 }
1312
1313 u16 bnx2x_select_queue(struct net_device *dev, struct sk_buff *skb)
1314 {
1315 #ifdef BCM_CNIC
1316 struct bnx2x *bp = netdev_priv(dev);
1317 if (NO_FCOE(bp))
1318 return skb_tx_hash(dev, skb);
1319 else {
1320 struct ethhdr *hdr = (struct ethhdr *)skb->data;
1321 u16 ether_type = ntohs(hdr->h_proto);
1322
1323 /* Skip VLAN tag if present */
1324 if (ether_type == ETH_P_8021Q) {
1325 struct vlan_ethhdr *vhdr =
1326 (struct vlan_ethhdr *)skb->data;
1327
1328 ether_type = ntohs(vhdr->h_vlan_encapsulated_proto);
1329 }
1330
1331 /* If ethertype is FCoE or FIP - use FCoE ring */
1332 if ((ether_type == ETH_P_FCOE) || (ether_type == ETH_P_FIP))
1333 return bnx2x_fcoe(bp, index);
1334 }
1335 #endif
1336 /* Select a none-FCoE queue: if FCoE is enabled, exclude FCoE L2 ring
1337 */
1338 return __skb_tx_hash(dev, skb,
1339 dev->real_num_tx_queues - FCOE_CONTEXT_USE);
1340 }
1341
1342 void bnx2x_set_num_queues(struct bnx2x *bp)
1343 {
1344 switch (bp->multi_mode) {
1345 case ETH_RSS_MODE_DISABLED:
1346 bp->num_queues = 1;
1347 break;
1348 case ETH_RSS_MODE_REGULAR:
1349 bp->num_queues = bnx2x_calc_num_queues(bp);
1350 break;
1351
1352 default:
1353 bp->num_queues = 1;
1354 break;
1355 }
1356
1357 /* Add special queues */
1358 bp->num_queues += NONE_ETH_CONTEXT_USE;
1359 }
1360
1361 #ifdef BCM_CNIC
1362 static inline void bnx2x_set_fcoe_eth_macs(struct bnx2x *bp)
1363 {
1364 if (!NO_FCOE(bp)) {
1365 if (!IS_MF_SD(bp))
1366 bnx2x_set_fip_eth_mac_addr(bp, 1);
1367 bnx2x_set_all_enode_macs(bp, 1);
1368 bp->flags |= FCOE_MACS_SET;
1369 }
1370 }
1371 #endif
1372
1373 static void bnx2x_release_firmware(struct bnx2x *bp)
1374 {
1375 kfree(bp->init_ops_offsets);
1376 kfree(bp->init_ops);
1377 kfree(bp->init_data);
1378 release_firmware(bp->firmware);
1379 }
1380
1381 static inline int bnx2x_set_real_num_queues(struct bnx2x *bp)
1382 {
1383 int rc, num = bp->num_queues;
1384
1385 #ifdef BCM_CNIC
1386 if (NO_FCOE(bp))
1387 num -= FCOE_CONTEXT_USE;
1388
1389 #endif
1390 netif_set_real_num_tx_queues(bp->dev, num);
1391 rc = netif_set_real_num_rx_queues(bp->dev, num);
1392 return rc;
1393 }
1394
1395 static inline void bnx2x_set_rx_buf_size(struct bnx2x *bp)
1396 {
1397 int i;
1398
1399 for_each_queue(bp, i) {
1400 struct bnx2x_fastpath *fp = &bp->fp[i];
1401
1402 /* Always use a mini-jumbo MTU for the FCoE L2 ring */
1403 if (IS_FCOE_IDX(i))
1404 /*
1405 * Although there are no IP frames expected to arrive to
1406 * this ring we still want to add an
1407 * IP_HEADER_ALIGNMENT_PADDING to prevent a buffer
1408 * overrun attack.
1409 */
1410 fp->rx_buf_size =
1411 BNX2X_FCOE_MINI_JUMBO_MTU + ETH_OVREHEAD +
1412 BNX2X_RX_ALIGN + IP_HEADER_ALIGNMENT_PADDING;
1413 else
1414 fp->rx_buf_size =
1415 bp->dev->mtu + ETH_OVREHEAD + BNX2X_RX_ALIGN +
1416 IP_HEADER_ALIGNMENT_PADDING;
1417 }
1418 }
1419
1420 /* must be called with rtnl_lock */
1421 int bnx2x_nic_load(struct bnx2x *bp, int load_mode)
1422 {
1423 u32 load_code;
1424 int i, rc;
1425
1426 /* Set init arrays */
1427 rc = bnx2x_init_firmware(bp);
1428 if (rc) {
1429 BNX2X_ERR("Error loading firmware\n");
1430 return rc;
1431 }
1432
1433 #ifdef BNX2X_STOP_ON_ERROR
1434 if (unlikely(bp->panic))
1435 return -EPERM;
1436 #endif
1437
1438 bp->state = BNX2X_STATE_OPENING_WAIT4_LOAD;
1439
1440 /* Set the initial link reported state to link down */
1441 bnx2x_acquire_phy_lock(bp);
1442 memset(&bp->last_reported_link, 0, sizeof(bp->last_reported_link));
1443 __set_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1444 &bp->last_reported_link.link_report_flags);
1445 bnx2x_release_phy_lock(bp);
1446
1447 /* must be called before memory allocation and HW init */
1448 bnx2x_ilt_set_info(bp);
1449
1450 /* zero fastpath structures preserving invariants like napi which are
1451 * allocated only once
1452 */
1453 for_each_queue(bp, i)
1454 bnx2x_bz_fp(bp, i);
1455
1456 /* Set the receive queues buffer size */
1457 bnx2x_set_rx_buf_size(bp);
1458
1459 for_each_queue(bp, i)
1460 bnx2x_fp(bp, i, disable_tpa) =
1461 ((bp->flags & TPA_ENABLE_FLAG) == 0);
1462
1463 #ifdef BCM_CNIC
1464 /* We don't want TPA on FCoE L2 ring */
1465 bnx2x_fcoe(bp, disable_tpa) = 1;
1466 #endif
1467
1468 if (bnx2x_alloc_mem(bp))
1469 return -ENOMEM;
1470
1471 /* As long as bnx2x_alloc_mem() may possibly update
1472 * bp->num_queues, bnx2x_set_real_num_queues() should always
1473 * come after it.
1474 */
1475 rc = bnx2x_set_real_num_queues(bp);
1476 if (rc) {
1477 BNX2X_ERR("Unable to set real_num_queues\n");
1478 goto load_error0;
1479 }
1480
1481 bnx2x_napi_enable(bp);
1482
1483 /* Send LOAD_REQUEST command to MCP
1484 Returns the type of LOAD command:
1485 if it is the first port to be initialized
1486 common blocks should be initialized, otherwise - not
1487 */
1488 if (!BP_NOMCP(bp)) {
1489 load_code = bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_REQ, 0);
1490 if (!load_code) {
1491 BNX2X_ERR("MCP response failure, aborting\n");
1492 rc = -EBUSY;
1493 goto load_error1;
1494 }
1495 if (load_code == FW_MSG_CODE_DRV_LOAD_REFUSED) {
1496 rc = -EBUSY; /* other port in diagnostic mode */
1497 goto load_error1;
1498 }
1499
1500 } else {
1501 int path = BP_PATH(bp);
1502 int port = BP_PORT(bp);
1503
1504 DP(NETIF_MSG_IFUP, "NO MCP - load counts[%d] %d, %d, %d\n",
1505 path, load_count[path][0], load_count[path][1],
1506 load_count[path][2]);
1507 load_count[path][0]++;
1508 load_count[path][1 + port]++;
1509 DP(NETIF_MSG_IFUP, "NO MCP - new load counts[%d] %d, %d, %d\n",
1510 path, load_count[path][0], load_count[path][1],
1511 load_count[path][2]);
1512 if (load_count[path][0] == 1)
1513 load_code = FW_MSG_CODE_DRV_LOAD_COMMON;
1514 else if (load_count[path][1 + port] == 1)
1515 load_code = FW_MSG_CODE_DRV_LOAD_PORT;
1516 else
1517 load_code = FW_MSG_CODE_DRV_LOAD_FUNCTION;
1518 }
1519
1520 if ((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
1521 (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP) ||
1522 (load_code == FW_MSG_CODE_DRV_LOAD_PORT))
1523 bp->port.pmf = 1;
1524 else
1525 bp->port.pmf = 0;
1526 DP(NETIF_MSG_LINK, "pmf %d\n", bp->port.pmf);
1527
1528 /* Initialize HW */
1529 rc = bnx2x_init_hw(bp, load_code);
1530 if (rc) {
1531 BNX2X_ERR("HW init failed, aborting\n");
1532 bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
1533 goto load_error2;
1534 }
1535
1536 /* Connect to IRQs */
1537 rc = bnx2x_setup_irqs(bp);
1538 if (rc) {
1539 bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
1540 goto load_error2;
1541 }
1542
1543 /* Setup NIC internals and enable interrupts */
1544 bnx2x_nic_init(bp, load_code);
1545
1546 if (((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
1547 (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP)) &&
1548 (bp->common.shmem2_base))
1549 SHMEM2_WR(bp, dcc_support,
1550 (SHMEM_DCC_SUPPORT_DISABLE_ENABLE_PF_TLV |
1551 SHMEM_DCC_SUPPORT_BANDWIDTH_ALLOCATION_TLV));
1552
1553 /* Send LOAD_DONE command to MCP */
1554 if (!BP_NOMCP(bp)) {
1555 load_code = bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
1556 if (!load_code) {
1557 BNX2X_ERR("MCP response failure, aborting\n");
1558 rc = -EBUSY;
1559 goto load_error3;
1560 }
1561 }
1562
1563 bnx2x_dcbx_init(bp);
1564
1565 bp->state = BNX2X_STATE_OPENING_WAIT4_PORT;
1566
1567 rc = bnx2x_func_start(bp);
1568 if (rc) {
1569 BNX2X_ERR("Function start failed!\n");
1570 #ifndef BNX2X_STOP_ON_ERROR
1571 goto load_error3;
1572 #else
1573 bp->panic = 1;
1574 return -EBUSY;
1575 #endif
1576 }
1577
1578 rc = bnx2x_setup_client(bp, &bp->fp[0], 1 /* Leading */);
1579 if (rc) {
1580 BNX2X_ERR("Setup leading failed!\n");
1581 #ifndef BNX2X_STOP_ON_ERROR
1582 goto load_error3;
1583 #else
1584 bp->panic = 1;
1585 return -EBUSY;
1586 #endif
1587 }
1588
1589 if (!CHIP_IS_E1(bp) &&
1590 (bp->mf_config[BP_VN(bp)] & FUNC_MF_CFG_FUNC_DISABLED)) {
1591 DP(NETIF_MSG_IFUP, "mf_cfg function disabled\n");
1592 bp->flags |= MF_FUNC_DIS;
1593 }
1594
1595 #ifdef BCM_CNIC
1596 /* Enable Timer scan */
1597 REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + BP_PORT(bp)*4, 1);
1598 #endif
1599
1600 for_each_nondefault_queue(bp, i) {
1601 rc = bnx2x_setup_client(bp, &bp->fp[i], 0);
1602 if (rc)
1603 #ifdef BCM_CNIC
1604 goto load_error4;
1605 #else
1606 goto load_error3;
1607 #endif
1608 }
1609
1610 /* Now when Clients are configured we are ready to work */
1611 bp->state = BNX2X_STATE_OPEN;
1612
1613 #ifdef BCM_CNIC
1614 bnx2x_set_fcoe_eth_macs(bp);
1615 #endif
1616
1617 bnx2x_set_eth_mac(bp, 1);
1618
1619 /* Clear MC configuration */
1620 if (CHIP_IS_E1(bp))
1621 bnx2x_invalidate_e1_mc_list(bp);
1622 else
1623 bnx2x_invalidate_e1h_mc_list(bp);
1624
1625 /* Clear UC lists configuration */
1626 bnx2x_invalidate_uc_list(bp);
1627
1628 if (bp->pending_max) {
1629 bnx2x_update_max_mf_config(bp, bp->pending_max);
1630 bp->pending_max = 0;
1631 }
1632
1633 if (bp->port.pmf)
1634 bnx2x_initial_phy_init(bp, load_mode);
1635
1636 /* Initialize Rx filtering */
1637 bnx2x_set_rx_mode(bp->dev);
1638
1639 /* Start fast path */
1640 switch (load_mode) {
1641 case LOAD_NORMAL:
1642 /* Tx queue should be only reenabled */
1643 netif_tx_wake_all_queues(bp->dev);
1644 /* Initialize the receive filter. */
1645 break;
1646
1647 case LOAD_OPEN:
1648 netif_tx_start_all_queues(bp->dev);
1649 smp_mb__after_clear_bit();
1650 break;
1651
1652 case LOAD_DIAG:
1653 bp->state = BNX2X_STATE_DIAG;
1654 break;
1655
1656 default:
1657 break;
1658 }
1659
1660 if (!bp->port.pmf)
1661 bnx2x__link_status_update(bp);
1662
1663 /* start the timer */
1664 mod_timer(&bp->timer, jiffies + bp->current_interval);
1665
1666 #ifdef BCM_CNIC
1667 bnx2x_setup_cnic_irq_info(bp);
1668 if (bp->state == BNX2X_STATE_OPEN)
1669 bnx2x_cnic_notify(bp, CNIC_CTL_START_CMD);
1670 #endif
1671 bnx2x_inc_load_cnt(bp);
1672
1673 bnx2x_release_firmware(bp);
1674
1675 return 0;
1676
1677 #ifdef BCM_CNIC
1678 load_error4:
1679 /* Disable Timer scan */
1680 REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + BP_PORT(bp)*4, 0);
1681 #endif
1682 load_error3:
1683 bnx2x_int_disable_sync(bp, 1);
1684
1685 /* Free SKBs, SGEs, TPA pool and driver internals */
1686 bnx2x_free_skbs(bp);
1687 for_each_rx_queue(bp, i)
1688 bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
1689
1690 /* Release IRQs */
1691 bnx2x_free_irq(bp);
1692 load_error2:
1693 if (!BP_NOMCP(bp)) {
1694 bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_REQ_WOL_MCP, 0);
1695 bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_DONE, 0);
1696 }
1697
1698 bp->port.pmf = 0;
1699 load_error1:
1700 bnx2x_napi_disable(bp);
1701 load_error0:
1702 bnx2x_free_mem(bp);
1703
1704 bnx2x_release_firmware(bp);
1705
1706 return rc;
1707 }
1708
1709 /* must be called with rtnl_lock */
1710 int bnx2x_nic_unload(struct bnx2x *bp, int unload_mode)
1711 {
1712 int i;
1713
1714 if (bp->state == BNX2X_STATE_CLOSED) {
1715 /* Interface has been removed - nothing to recover */
1716 bp->recovery_state = BNX2X_RECOVERY_DONE;
1717 bp->is_leader = 0;
1718 bnx2x_release_hw_lock(bp, HW_LOCK_RESOURCE_RESERVED_08);
1719 smp_wmb();
1720
1721 return -EINVAL;
1722 }
1723
1724 #ifdef BCM_CNIC
1725 bnx2x_cnic_notify(bp, CNIC_CTL_STOP_CMD);
1726 #endif
1727 bp->state = BNX2X_STATE_CLOSING_WAIT4_HALT;
1728
1729 /* Set "drop all" */
1730 bp->rx_mode = BNX2X_RX_MODE_NONE;
1731 bnx2x_set_storm_rx_mode(bp);
1732
1733 /* Stop Tx */
1734 bnx2x_tx_disable(bp);
1735
1736 del_timer_sync(&bp->timer);
1737
1738 SHMEM_WR(bp, func_mb[BP_FW_MB_IDX(bp)].drv_pulse_mb,
1739 (DRV_PULSE_ALWAYS_ALIVE | bp->fw_drv_pulse_wr_seq));
1740
1741 bnx2x_stats_handle(bp, STATS_EVENT_STOP);
1742
1743 /* Cleanup the chip if needed */
1744 if (unload_mode != UNLOAD_RECOVERY)
1745 bnx2x_chip_cleanup(bp, unload_mode);
1746 else {
1747 /* Disable HW interrupts, NAPI and Tx */
1748 bnx2x_netif_stop(bp, 1);
1749
1750 /* Release IRQs */
1751 bnx2x_free_irq(bp);
1752 }
1753
1754 bp->port.pmf = 0;
1755
1756 /* Free SKBs, SGEs, TPA pool and driver internals */
1757 bnx2x_free_skbs(bp);
1758 for_each_rx_queue(bp, i)
1759 bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
1760
1761 bnx2x_free_mem(bp);
1762
1763 bp->state = BNX2X_STATE_CLOSED;
1764
1765 /* The last driver must disable a "close the gate" if there is no
1766 * parity attention or "process kill" pending.
1767 */
1768 if ((!bnx2x_dec_load_cnt(bp)) && (!bnx2x_chk_parity_attn(bp)) &&
1769 bnx2x_reset_is_done(bp))
1770 bnx2x_disable_close_the_gate(bp);
1771
1772 /* Reset MCP mail box sequence if there is on going recovery */
1773 if (unload_mode == UNLOAD_RECOVERY)
1774 bp->fw_seq = 0;
1775
1776 return 0;
1777 }
1778
1779 int bnx2x_set_power_state(struct bnx2x *bp, pci_power_t state)
1780 {
1781 u16 pmcsr;
1782
1783 /* If there is no power capability, silently succeed */
1784 if (!bp->pm_cap) {
1785 DP(NETIF_MSG_HW, "No power capability. Breaking.\n");
1786 return 0;
1787 }
1788
1789 pci_read_config_word(bp->pdev, bp->pm_cap + PCI_PM_CTRL, &pmcsr);
1790
1791 switch (state) {
1792 case PCI_D0:
1793 pci_write_config_word(bp->pdev, bp->pm_cap + PCI_PM_CTRL,
1794 ((pmcsr & ~PCI_PM_CTRL_STATE_MASK) |
1795 PCI_PM_CTRL_PME_STATUS));
1796
1797 if (pmcsr & PCI_PM_CTRL_STATE_MASK)
1798 /* delay required during transition out of D3hot */
1799 msleep(20);
1800 break;
1801
1802 case PCI_D3hot:
1803 /* If there are other clients above don't
1804 shut down the power */
1805 if (atomic_read(&bp->pdev->enable_cnt) != 1)
1806 return 0;
1807 /* Don't shut down the power for emulation and FPGA */
1808 if (CHIP_REV_IS_SLOW(bp))
1809 return 0;
1810
1811 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
1812 pmcsr |= 3;
1813
1814 if (bp->wol)
1815 pmcsr |= PCI_PM_CTRL_PME_ENABLE;
1816
1817 pci_write_config_word(bp->pdev, bp->pm_cap + PCI_PM_CTRL,
1818 pmcsr);
1819
1820 /* No more memory access after this point until
1821 * device is brought back to D0.
1822 */
1823 break;
1824
1825 default:
1826 return -EINVAL;
1827 }
1828 return 0;
1829 }
1830
1831 /*
1832 * net_device service functions
1833 */
1834 int bnx2x_poll(struct napi_struct *napi, int budget)
1835 {
1836 int work_done = 0;
1837 struct bnx2x_fastpath *fp = container_of(napi, struct bnx2x_fastpath,
1838 napi);
1839 struct bnx2x *bp = fp->bp;
1840
1841 while (1) {
1842 #ifdef BNX2X_STOP_ON_ERROR
1843 if (unlikely(bp->panic)) {
1844 napi_complete(napi);
1845 return 0;
1846 }
1847 #endif
1848
1849 if (bnx2x_has_tx_work(fp))
1850 bnx2x_tx_int(fp);
1851
1852 if (bnx2x_has_rx_work(fp)) {
1853 work_done += bnx2x_rx_int(fp, budget - work_done);
1854
1855 /* must not complete if we consumed full budget */
1856 if (work_done >= budget)
1857 break;
1858 }
1859
1860 /* Fall out from the NAPI loop if needed */
1861 if (!(bnx2x_has_rx_work(fp) || bnx2x_has_tx_work(fp))) {
1862 #ifdef BCM_CNIC
1863 /* No need to update SB for FCoE L2 ring as long as
1864 * it's connected to the default SB and the SB
1865 * has been updated when NAPI was scheduled.
1866 */
1867 if (IS_FCOE_FP(fp)) {
1868 napi_complete(napi);
1869 break;
1870 }
1871 #endif
1872
1873 bnx2x_update_fpsb_idx(fp);
1874 /* bnx2x_has_rx_work() reads the status block,
1875 * thus we need to ensure that status block indices
1876 * have been actually read (bnx2x_update_fpsb_idx)
1877 * prior to this check (bnx2x_has_rx_work) so that
1878 * we won't write the "newer" value of the status block
1879 * to IGU (if there was a DMA right after
1880 * bnx2x_has_rx_work and if there is no rmb, the memory
1881 * reading (bnx2x_update_fpsb_idx) may be postponed
1882 * to right before bnx2x_ack_sb). In this case there
1883 * will never be another interrupt until there is
1884 * another update of the status block, while there
1885 * is still unhandled work.
1886 */
1887 rmb();
1888
1889 if (!(bnx2x_has_rx_work(fp) || bnx2x_has_tx_work(fp))) {
1890 napi_complete(napi);
1891 /* Re-enable interrupts */
1892 DP(NETIF_MSG_HW,
1893 "Update index to %d\n", fp->fp_hc_idx);
1894 bnx2x_ack_sb(bp, fp->igu_sb_id, USTORM_ID,
1895 le16_to_cpu(fp->fp_hc_idx),
1896 IGU_INT_ENABLE, 1);
1897 break;
1898 }
1899 }
1900 }
1901
1902 return work_done;
1903 }
1904
1905 /* we split the first BD into headers and data BDs
1906 * to ease the pain of our fellow microcode engineers
1907 * we use one mapping for both BDs
1908 * So far this has only been observed to happen
1909 * in Other Operating Systems(TM)
1910 */
1911 static noinline u16 bnx2x_tx_split(struct bnx2x *bp,
1912 struct bnx2x_fastpath *fp,
1913 struct sw_tx_bd *tx_buf,
1914 struct eth_tx_start_bd **tx_bd, u16 hlen,
1915 u16 bd_prod, int nbd)
1916 {
1917 struct eth_tx_start_bd *h_tx_bd = *tx_bd;
1918 struct eth_tx_bd *d_tx_bd;
1919 dma_addr_t mapping;
1920 int old_len = le16_to_cpu(h_tx_bd->nbytes);
1921
1922 /* first fix first BD */
1923 h_tx_bd->nbd = cpu_to_le16(nbd);
1924 h_tx_bd->nbytes = cpu_to_le16(hlen);
1925
1926 DP(NETIF_MSG_TX_QUEUED, "TSO split header size is %d "
1927 "(%x:%x) nbd %d\n", h_tx_bd->nbytes, h_tx_bd->addr_hi,
1928 h_tx_bd->addr_lo, h_tx_bd->nbd);
1929
1930 /* now get a new data BD
1931 * (after the pbd) and fill it */
1932 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
1933 d_tx_bd = &fp->tx_desc_ring[bd_prod].reg_bd;
1934
1935 mapping = HILO_U64(le32_to_cpu(h_tx_bd->addr_hi),
1936 le32_to_cpu(h_tx_bd->addr_lo)) + hlen;
1937
1938 d_tx_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
1939 d_tx_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
1940 d_tx_bd->nbytes = cpu_to_le16(old_len - hlen);
1941
1942 /* this marks the BD as one that has no individual mapping */
1943 tx_buf->flags |= BNX2X_TSO_SPLIT_BD;
1944
1945 DP(NETIF_MSG_TX_QUEUED,
1946 "TSO split data size is %d (%x:%x)\n",
1947 d_tx_bd->nbytes, d_tx_bd->addr_hi, d_tx_bd->addr_lo);
1948
1949 /* update tx_bd */
1950 *tx_bd = (struct eth_tx_start_bd *)d_tx_bd;
1951
1952 return bd_prod;
1953 }
1954
1955 static inline u16 bnx2x_csum_fix(unsigned char *t_header, u16 csum, s8 fix)
1956 {
1957 if (fix > 0)
1958 csum = (u16) ~csum_fold(csum_sub(csum,
1959 csum_partial(t_header - fix, fix, 0)));
1960
1961 else if (fix < 0)
1962 csum = (u16) ~csum_fold(csum_add(csum,
1963 csum_partial(t_header, -fix, 0)));
1964
1965 return swab16(csum);
1966 }
1967
1968 static inline u32 bnx2x_xmit_type(struct bnx2x *bp, struct sk_buff *skb)
1969 {
1970 u32 rc;
1971
1972 if (skb->ip_summed != CHECKSUM_PARTIAL)
1973 rc = XMIT_PLAIN;
1974
1975 else {
1976 if (vlan_get_protocol(skb) == htons(ETH_P_IPV6)) {
1977 rc = XMIT_CSUM_V6;
1978 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
1979 rc |= XMIT_CSUM_TCP;
1980
1981 } else {
1982 rc = XMIT_CSUM_V4;
1983 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
1984 rc |= XMIT_CSUM_TCP;
1985 }
1986 }
1987
1988 if (skb_is_gso_v6(skb))
1989 rc |= XMIT_GSO_V6 | XMIT_CSUM_TCP | XMIT_CSUM_V6;
1990 else if (skb_is_gso(skb))
1991 rc |= XMIT_GSO_V4 | XMIT_CSUM_V4 | XMIT_CSUM_TCP;
1992
1993 return rc;
1994 }
1995
1996 #if (MAX_SKB_FRAGS >= MAX_FETCH_BD - 3)
1997 /* check if packet requires linearization (packet is too fragmented)
1998 no need to check fragmentation if page size > 8K (there will be no
1999 violation to FW restrictions) */
2000 static int bnx2x_pkt_req_lin(struct bnx2x *bp, struct sk_buff *skb,
2001 u32 xmit_type)
2002 {
2003 int to_copy = 0;
2004 int hlen = 0;
2005 int first_bd_sz = 0;
2006
2007 /* 3 = 1 (for linear data BD) + 2 (for PBD and last BD) */
2008 if (skb_shinfo(skb)->nr_frags >= (MAX_FETCH_BD - 3)) {
2009
2010 if (xmit_type & XMIT_GSO) {
2011 unsigned short lso_mss = skb_shinfo(skb)->gso_size;
2012 /* Check if LSO packet needs to be copied:
2013 3 = 1 (for headers BD) + 2 (for PBD and last BD) */
2014 int wnd_size = MAX_FETCH_BD - 3;
2015 /* Number of windows to check */
2016 int num_wnds = skb_shinfo(skb)->nr_frags - wnd_size;
2017 int wnd_idx = 0;
2018 int frag_idx = 0;
2019 u32 wnd_sum = 0;
2020
2021 /* Headers length */
2022 hlen = (int)(skb_transport_header(skb) - skb->data) +
2023 tcp_hdrlen(skb);
2024
2025 /* Amount of data (w/o headers) on linear part of SKB*/
2026 first_bd_sz = skb_headlen(skb) - hlen;
2027
2028 wnd_sum = first_bd_sz;
2029
2030 /* Calculate the first sum - it's special */
2031 for (frag_idx = 0; frag_idx < wnd_size - 1; frag_idx++)
2032 wnd_sum +=
2033 skb_shinfo(skb)->frags[frag_idx].size;
2034
2035 /* If there was data on linear skb data - check it */
2036 if (first_bd_sz > 0) {
2037 if (unlikely(wnd_sum < lso_mss)) {
2038 to_copy = 1;
2039 goto exit_lbl;
2040 }
2041
2042 wnd_sum -= first_bd_sz;
2043 }
2044
2045 /* Others are easier: run through the frag list and
2046 check all windows */
2047 for (wnd_idx = 0; wnd_idx <= num_wnds; wnd_idx++) {
2048 wnd_sum +=
2049 skb_shinfo(skb)->frags[wnd_idx + wnd_size - 1].size;
2050
2051 if (unlikely(wnd_sum < lso_mss)) {
2052 to_copy = 1;
2053 break;
2054 }
2055 wnd_sum -=
2056 skb_shinfo(skb)->frags[wnd_idx].size;
2057 }
2058 } else {
2059 /* in non-LSO too fragmented packet should always
2060 be linearized */
2061 to_copy = 1;
2062 }
2063 }
2064
2065 exit_lbl:
2066 if (unlikely(to_copy))
2067 DP(NETIF_MSG_TX_QUEUED,
2068 "Linearization IS REQUIRED for %s packet. "
2069 "num_frags %d hlen %d first_bd_sz %d\n",
2070 (xmit_type & XMIT_GSO) ? "LSO" : "non-LSO",
2071 skb_shinfo(skb)->nr_frags, hlen, first_bd_sz);
2072
2073 return to_copy;
2074 }
2075 #endif
2076
2077 static inline void bnx2x_set_pbd_gso_e2(struct sk_buff *skb, u32 *parsing_data,
2078 u32 xmit_type)
2079 {
2080 *parsing_data |= (skb_shinfo(skb)->gso_size <<
2081 ETH_TX_PARSE_BD_E2_LSO_MSS_SHIFT) &
2082 ETH_TX_PARSE_BD_E2_LSO_MSS;
2083 if ((xmit_type & XMIT_GSO_V6) &&
2084 (ipv6_hdr(skb)->nexthdr == NEXTHDR_IPV6))
2085 *parsing_data |= ETH_TX_PARSE_BD_E2_IPV6_WITH_EXT_HDR;
2086 }
2087
2088 /**
2089 * bnx2x_set_pbd_gso - update PBD in GSO case.
2090 *
2091 * @skb: packet skb
2092 * @pbd: parse BD
2093 * @xmit_type: xmit flags
2094 */
2095 static inline void bnx2x_set_pbd_gso(struct sk_buff *skb,
2096 struct eth_tx_parse_bd_e1x *pbd,
2097 u32 xmit_type)
2098 {
2099 pbd->lso_mss = cpu_to_le16(skb_shinfo(skb)->gso_size);
2100 pbd->tcp_send_seq = swab32(tcp_hdr(skb)->seq);
2101 pbd->tcp_flags = pbd_tcp_flags(skb);
2102
2103 if (xmit_type & XMIT_GSO_V4) {
2104 pbd->ip_id = swab16(ip_hdr(skb)->id);
2105 pbd->tcp_pseudo_csum =
2106 swab16(~csum_tcpudp_magic(ip_hdr(skb)->saddr,
2107 ip_hdr(skb)->daddr,
2108 0, IPPROTO_TCP, 0));
2109
2110 } else
2111 pbd->tcp_pseudo_csum =
2112 swab16(~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2113 &ipv6_hdr(skb)->daddr,
2114 0, IPPROTO_TCP, 0));
2115
2116 pbd->global_data |= ETH_TX_PARSE_BD_E1X_PSEUDO_CS_WITHOUT_LEN;
2117 }
2118
2119 /**
2120 * bnx2x_set_pbd_csum_e2 - update PBD with checksum and return header length
2121 *
2122 * @bp: driver handle
2123 * @skb: packet skb
2124 * @parsing_data: data to be updated
2125 * @xmit_type: xmit flags
2126 *
2127 * 57712 related
2128 */
2129 static inline u8 bnx2x_set_pbd_csum_e2(struct bnx2x *bp, struct sk_buff *skb,
2130 u32 *parsing_data, u32 xmit_type)
2131 {
2132 *parsing_data |=
2133 ((((u8 *)skb_transport_header(skb) - skb->data) >> 1) <<
2134 ETH_TX_PARSE_BD_E2_TCP_HDR_START_OFFSET_W_SHIFT) &
2135 ETH_TX_PARSE_BD_E2_TCP_HDR_START_OFFSET_W;
2136
2137 if (xmit_type & XMIT_CSUM_TCP) {
2138 *parsing_data |= ((tcp_hdrlen(skb) / 4) <<
2139 ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW_SHIFT) &
2140 ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW;
2141
2142 return skb_transport_header(skb) + tcp_hdrlen(skb) - skb->data;
2143 } else
2144 /* We support checksum offload for TCP and UDP only.
2145 * No need to pass the UDP header length - it's a constant.
2146 */
2147 return skb_transport_header(skb) +
2148 sizeof(struct udphdr) - skb->data;
2149 }
2150
2151 /**
2152 * bnx2x_set_pbd_csum - update PBD with checksum and return header length
2153 *
2154 * @bp: driver handle
2155 * @skb: packet skb
2156 * @pbd: parse BD to be updated
2157 * @xmit_type: xmit flags
2158 */
2159 static inline u8 bnx2x_set_pbd_csum(struct bnx2x *bp, struct sk_buff *skb,
2160 struct eth_tx_parse_bd_e1x *pbd,
2161 u32 xmit_type)
2162 {
2163 u8 hlen = (skb_network_header(skb) - skb->data) >> 1;
2164
2165 /* for now NS flag is not used in Linux */
2166 pbd->global_data =
2167 (hlen | ((skb->protocol == cpu_to_be16(ETH_P_8021Q)) <<
2168 ETH_TX_PARSE_BD_E1X_LLC_SNAP_EN_SHIFT));
2169
2170 pbd->ip_hlen_w = (skb_transport_header(skb) -
2171 skb_network_header(skb)) >> 1;
2172
2173 hlen += pbd->ip_hlen_w;
2174
2175 /* We support checksum offload for TCP and UDP only */
2176 if (xmit_type & XMIT_CSUM_TCP)
2177 hlen += tcp_hdrlen(skb) / 2;
2178 else
2179 hlen += sizeof(struct udphdr) / 2;
2180
2181 pbd->total_hlen_w = cpu_to_le16(hlen);
2182 hlen = hlen*2;
2183
2184 if (xmit_type & XMIT_CSUM_TCP) {
2185 pbd->tcp_pseudo_csum = swab16(tcp_hdr(skb)->check);
2186
2187 } else {
2188 s8 fix = SKB_CS_OFF(skb); /* signed! */
2189
2190 DP(NETIF_MSG_TX_QUEUED,
2191 "hlen %d fix %d csum before fix %x\n",
2192 le16_to_cpu(pbd->total_hlen_w), fix, SKB_CS(skb));
2193
2194 /* HW bug: fixup the CSUM */
2195 pbd->tcp_pseudo_csum =
2196 bnx2x_csum_fix(skb_transport_header(skb),
2197 SKB_CS(skb), fix);
2198
2199 DP(NETIF_MSG_TX_QUEUED, "csum after fix %x\n",
2200 pbd->tcp_pseudo_csum);
2201 }
2202
2203 return hlen;
2204 }
2205
2206 /* called with netif_tx_lock
2207 * bnx2x_tx_int() runs without netif_tx_lock unless it needs to call
2208 * netif_wake_queue()
2209 */
2210 netdev_tx_t bnx2x_start_xmit(struct sk_buff *skb, struct net_device *dev)
2211 {
2212 struct bnx2x *bp = netdev_priv(dev);
2213 struct bnx2x_fastpath *fp;
2214 struct netdev_queue *txq;
2215 struct sw_tx_bd *tx_buf;
2216 struct eth_tx_start_bd *tx_start_bd;
2217 struct eth_tx_bd *tx_data_bd, *total_pkt_bd = NULL;
2218 struct eth_tx_parse_bd_e1x *pbd_e1x = NULL;
2219 struct eth_tx_parse_bd_e2 *pbd_e2 = NULL;
2220 u32 pbd_e2_parsing_data = 0;
2221 u16 pkt_prod, bd_prod;
2222 int nbd, fp_index;
2223 dma_addr_t mapping;
2224 u32 xmit_type = bnx2x_xmit_type(bp, skb);
2225 int i;
2226 u8 hlen = 0;
2227 __le16 pkt_size = 0;
2228 struct ethhdr *eth;
2229 u8 mac_type = UNICAST_ADDRESS;
2230
2231 #ifdef BNX2X_STOP_ON_ERROR
2232 if (unlikely(bp->panic))
2233 return NETDEV_TX_BUSY;
2234 #endif
2235
2236 fp_index = skb_get_queue_mapping(skb);
2237 txq = netdev_get_tx_queue(dev, fp_index);
2238
2239 fp = &bp->fp[fp_index];
2240
2241 if (unlikely(bnx2x_tx_avail(fp) < (skb_shinfo(skb)->nr_frags + 3))) {
2242 fp->eth_q_stats.driver_xoff++;
2243 netif_tx_stop_queue(txq);
2244 BNX2X_ERR("BUG! Tx ring full when queue awake!\n");
2245 return NETDEV_TX_BUSY;
2246 }
2247
2248 DP(NETIF_MSG_TX_QUEUED, "queue[%d]: SKB: summed %x protocol %x "
2249 "protocol(%x,%x) gso type %x xmit_type %x\n",
2250 fp_index, skb->ip_summed, skb->protocol, ipv6_hdr(skb)->nexthdr,
2251 ip_hdr(skb)->protocol, skb_shinfo(skb)->gso_type, xmit_type);
2252
2253 eth = (struct ethhdr *)skb->data;
2254
2255 /* set flag according to packet type (UNICAST_ADDRESS is default)*/
2256 if (unlikely(is_multicast_ether_addr(eth->h_dest))) {
2257 if (is_broadcast_ether_addr(eth->h_dest))
2258 mac_type = BROADCAST_ADDRESS;
2259 else
2260 mac_type = MULTICAST_ADDRESS;
2261 }
2262
2263 #if (MAX_SKB_FRAGS >= MAX_FETCH_BD - 3)
2264 /* First, check if we need to linearize the skb (due to FW
2265 restrictions). No need to check fragmentation if page size > 8K
2266 (there will be no violation to FW restrictions) */
2267 if (bnx2x_pkt_req_lin(bp, skb, xmit_type)) {
2268 /* Statistics of linearization */
2269 bp->lin_cnt++;
2270 if (skb_linearize(skb) != 0) {
2271 DP(NETIF_MSG_TX_QUEUED, "SKB linearization failed - "
2272 "silently dropping this SKB\n");
2273 dev_kfree_skb_any(skb);
2274 return NETDEV_TX_OK;
2275 }
2276 }
2277 #endif
2278
2279 /*
2280 Please read carefully. First we use one BD which we mark as start,
2281 then we have a parsing info BD (used for TSO or xsum),
2282 and only then we have the rest of the TSO BDs.
2283 (don't forget to mark the last one as last,
2284 and to unmap only AFTER you write to the BD ...)
2285 And above all, all pdb sizes are in words - NOT DWORDS!
2286 */
2287
2288 pkt_prod = fp->tx_pkt_prod++;
2289 bd_prod = TX_BD(fp->tx_bd_prod);
2290
2291 /* get a tx_buf and first BD */
2292 tx_buf = &fp->tx_buf_ring[TX_BD(pkt_prod)];
2293 tx_start_bd = &fp->tx_desc_ring[bd_prod].start_bd;
2294
2295 tx_start_bd->bd_flags.as_bitfield = ETH_TX_BD_FLAGS_START_BD;
2296 SET_FLAG(tx_start_bd->general_data, ETH_TX_START_BD_ETH_ADDR_TYPE,
2297 mac_type);
2298
2299 /* header nbd */
2300 SET_FLAG(tx_start_bd->general_data, ETH_TX_START_BD_HDR_NBDS, 1);
2301
2302 /* remember the first BD of the packet */
2303 tx_buf->first_bd = fp->tx_bd_prod;
2304 tx_buf->skb = skb;
2305 tx_buf->flags = 0;
2306
2307 DP(NETIF_MSG_TX_QUEUED,
2308 "sending pkt %u @%p next_idx %u bd %u @%p\n",
2309 pkt_prod, tx_buf, fp->tx_pkt_prod, bd_prod, tx_start_bd);
2310
2311 if (vlan_tx_tag_present(skb)) {
2312 tx_start_bd->vlan_or_ethertype =
2313 cpu_to_le16(vlan_tx_tag_get(skb));
2314 tx_start_bd->bd_flags.as_bitfield |=
2315 (X_ETH_OUTBAND_VLAN << ETH_TX_BD_FLAGS_VLAN_MODE_SHIFT);
2316 } else
2317 tx_start_bd->vlan_or_ethertype = cpu_to_le16(pkt_prod);
2318
2319 /* turn on parsing and get a BD */
2320 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
2321
2322 if (xmit_type & XMIT_CSUM) {
2323 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_L4_CSUM;
2324
2325 if (xmit_type & XMIT_CSUM_V4)
2326 tx_start_bd->bd_flags.as_bitfield |=
2327 ETH_TX_BD_FLAGS_IP_CSUM;
2328 else
2329 tx_start_bd->bd_flags.as_bitfield |=
2330 ETH_TX_BD_FLAGS_IPV6;
2331
2332 if (!(xmit_type & XMIT_CSUM_TCP))
2333 tx_start_bd->bd_flags.as_bitfield |=
2334 ETH_TX_BD_FLAGS_IS_UDP;
2335 }
2336
2337 if (CHIP_IS_E2(bp)) {
2338 pbd_e2 = &fp->tx_desc_ring[bd_prod].parse_bd_e2;
2339 memset(pbd_e2, 0, sizeof(struct eth_tx_parse_bd_e2));
2340 /* Set PBD in checksum offload case */
2341 if (xmit_type & XMIT_CSUM)
2342 hlen = bnx2x_set_pbd_csum_e2(bp, skb,
2343 &pbd_e2_parsing_data,
2344 xmit_type);
2345 } else {
2346 pbd_e1x = &fp->tx_desc_ring[bd_prod].parse_bd_e1x;
2347 memset(pbd_e1x, 0, sizeof(struct eth_tx_parse_bd_e1x));
2348 /* Set PBD in checksum offload case */
2349 if (xmit_type & XMIT_CSUM)
2350 hlen = bnx2x_set_pbd_csum(bp, skb, pbd_e1x, xmit_type);
2351
2352 }
2353
2354 /* Map skb linear data for DMA */
2355 mapping = dma_map_single(&bp->pdev->dev, skb->data,
2356 skb_headlen(skb), DMA_TO_DEVICE);
2357
2358 /* Setup the data pointer of the first BD of the packet */
2359 tx_start_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
2360 tx_start_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
2361 nbd = skb_shinfo(skb)->nr_frags + 2; /* start_bd + pbd + frags */
2362 tx_start_bd->nbd = cpu_to_le16(nbd);
2363 tx_start_bd->nbytes = cpu_to_le16(skb_headlen(skb));
2364 pkt_size = tx_start_bd->nbytes;
2365
2366 DP(NETIF_MSG_TX_QUEUED, "first bd @%p addr (%x:%x) nbd %d"
2367 " nbytes %d flags %x vlan %x\n",
2368 tx_start_bd, tx_start_bd->addr_hi, tx_start_bd->addr_lo,
2369 le16_to_cpu(tx_start_bd->nbd), le16_to_cpu(tx_start_bd->nbytes),
2370 tx_start_bd->bd_flags.as_bitfield,
2371 le16_to_cpu(tx_start_bd->vlan_or_ethertype));
2372
2373 if (xmit_type & XMIT_GSO) {
2374
2375 DP(NETIF_MSG_TX_QUEUED,
2376 "TSO packet len %d hlen %d total len %d tso size %d\n",
2377 skb->len, hlen, skb_headlen(skb),
2378 skb_shinfo(skb)->gso_size);
2379
2380 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_SW_LSO;
2381
2382 if (unlikely(skb_headlen(skb) > hlen))
2383 bd_prod = bnx2x_tx_split(bp, fp, tx_buf, &tx_start_bd,
2384 hlen, bd_prod, ++nbd);
2385 if (CHIP_IS_E2(bp))
2386 bnx2x_set_pbd_gso_e2(skb, &pbd_e2_parsing_data,
2387 xmit_type);
2388 else
2389 bnx2x_set_pbd_gso(skb, pbd_e1x, xmit_type);
2390 }
2391
2392 /* Set the PBD's parsing_data field if not zero
2393 * (for the chips newer than 57711).
2394 */
2395 if (pbd_e2_parsing_data)
2396 pbd_e2->parsing_data = cpu_to_le32(pbd_e2_parsing_data);
2397
2398 tx_data_bd = (struct eth_tx_bd *)tx_start_bd;
2399
2400 /* Handle fragmented skb */
2401 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2402 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2403
2404 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
2405 tx_data_bd = &fp->tx_desc_ring[bd_prod].reg_bd;
2406 if (total_pkt_bd == NULL)
2407 total_pkt_bd = &fp->tx_desc_ring[bd_prod].reg_bd;
2408
2409 mapping = dma_map_page(&bp->pdev->dev, frag->page,
2410 frag->page_offset,
2411 frag->size, DMA_TO_DEVICE);
2412
2413 tx_data_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
2414 tx_data_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
2415 tx_data_bd->nbytes = cpu_to_le16(frag->size);
2416 le16_add_cpu(&pkt_size, frag->size);
2417
2418 DP(NETIF_MSG_TX_QUEUED,
2419 "frag %d bd @%p addr (%x:%x) nbytes %d\n",
2420 i, tx_data_bd, tx_data_bd->addr_hi, tx_data_bd->addr_lo,
2421 le16_to_cpu(tx_data_bd->nbytes));
2422 }
2423
2424 DP(NETIF_MSG_TX_QUEUED, "last bd @%p\n", tx_data_bd);
2425
2426 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
2427
2428 /* now send a tx doorbell, counting the next BD
2429 * if the packet contains or ends with it
2430 */
2431 if (TX_BD_POFF(bd_prod) < nbd)
2432 nbd++;
2433
2434 if (total_pkt_bd != NULL)
2435 total_pkt_bd->total_pkt_bytes = pkt_size;
2436
2437 if (pbd_e1x)
2438 DP(NETIF_MSG_TX_QUEUED,
2439 "PBD (E1X) @%p ip_data %x ip_hlen %u ip_id %u lso_mss %u"
2440 " tcp_flags %x xsum %x seq %u hlen %u\n",
2441 pbd_e1x, pbd_e1x->global_data, pbd_e1x->ip_hlen_w,
2442 pbd_e1x->ip_id, pbd_e1x->lso_mss, pbd_e1x->tcp_flags,
2443 pbd_e1x->tcp_pseudo_csum, pbd_e1x->tcp_send_seq,
2444 le16_to_cpu(pbd_e1x->total_hlen_w));
2445 if (pbd_e2)
2446 DP(NETIF_MSG_TX_QUEUED,
2447 "PBD (E2) @%p dst %x %x %x src %x %x %x parsing_data %x\n",
2448 pbd_e2, pbd_e2->dst_mac_addr_hi, pbd_e2->dst_mac_addr_mid,
2449 pbd_e2->dst_mac_addr_lo, pbd_e2->src_mac_addr_hi,
2450 pbd_e2->src_mac_addr_mid, pbd_e2->src_mac_addr_lo,
2451 pbd_e2->parsing_data);
2452 DP(NETIF_MSG_TX_QUEUED, "doorbell: nbd %d bd %u\n", nbd, bd_prod);
2453
2454 /*
2455 * Make sure that the BD data is updated before updating the producer
2456 * since FW might read the BD right after the producer is updated.
2457 * This is only applicable for weak-ordered memory model archs such
2458 * as IA-64. The following barrier is also mandatory since FW will
2459 * assumes packets must have BDs.
2460 */
2461 wmb();
2462
2463 fp->tx_db.data.prod += nbd;
2464 barrier();
2465
2466 DOORBELL(bp, fp->cid, fp->tx_db.raw);
2467
2468 mmiowb();
2469
2470 fp->tx_bd_prod += nbd;
2471
2472 if (unlikely(bnx2x_tx_avail(fp) < MAX_SKB_FRAGS + 3)) {
2473 netif_tx_stop_queue(txq);
2474
2475 /* paired memory barrier is in bnx2x_tx_int(), we have to keep
2476 * ordering of set_bit() in netif_tx_stop_queue() and read of
2477 * fp->bd_tx_cons */
2478 smp_mb();
2479
2480 fp->eth_q_stats.driver_xoff++;
2481 if (bnx2x_tx_avail(fp) >= MAX_SKB_FRAGS + 3)
2482 netif_tx_wake_queue(txq);
2483 }
2484 fp->tx_pkt++;
2485
2486 return NETDEV_TX_OK;
2487 }
2488
2489 /* called with rtnl_lock */
2490 int bnx2x_change_mac_addr(struct net_device *dev, void *p)
2491 {
2492 struct sockaddr *addr = p;
2493 struct bnx2x *bp = netdev_priv(dev);
2494
2495 if (!is_valid_ether_addr((u8 *)(addr->sa_data)))
2496 return -EINVAL;
2497
2498 memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
2499 if (netif_running(dev))
2500 bnx2x_set_eth_mac(bp, 1);
2501
2502 return 0;
2503 }
2504
2505 static void bnx2x_free_fp_mem_at(struct bnx2x *bp, int fp_index)
2506 {
2507 union host_hc_status_block *sb = &bnx2x_fp(bp, fp_index, status_blk);
2508 struct bnx2x_fastpath *fp = &bp->fp[fp_index];
2509
2510 /* Common */
2511 #ifdef BCM_CNIC
2512 if (IS_FCOE_IDX(fp_index)) {
2513 memset(sb, 0, sizeof(union host_hc_status_block));
2514 fp->status_blk_mapping = 0;
2515
2516 } else {
2517 #endif
2518 /* status blocks */
2519 if (CHIP_IS_E2(bp))
2520 BNX2X_PCI_FREE(sb->e2_sb,
2521 bnx2x_fp(bp, fp_index,
2522 status_blk_mapping),
2523 sizeof(struct host_hc_status_block_e2));
2524 else
2525 BNX2X_PCI_FREE(sb->e1x_sb,
2526 bnx2x_fp(bp, fp_index,
2527 status_blk_mapping),
2528 sizeof(struct host_hc_status_block_e1x));
2529 #ifdef BCM_CNIC
2530 }
2531 #endif
2532 /* Rx */
2533 if (!skip_rx_queue(bp, fp_index)) {
2534 bnx2x_free_rx_bds(fp);
2535
2536 /* fastpath rx rings: rx_buf rx_desc rx_comp */
2537 BNX2X_FREE(bnx2x_fp(bp, fp_index, rx_buf_ring));
2538 BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_desc_ring),
2539 bnx2x_fp(bp, fp_index, rx_desc_mapping),
2540 sizeof(struct eth_rx_bd) * NUM_RX_BD);
2541
2542 BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_comp_ring),
2543 bnx2x_fp(bp, fp_index, rx_comp_mapping),
2544 sizeof(struct eth_fast_path_rx_cqe) *
2545 NUM_RCQ_BD);
2546
2547 /* SGE ring */
2548 BNX2X_FREE(bnx2x_fp(bp, fp_index, rx_page_ring));
2549 BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_sge_ring),
2550 bnx2x_fp(bp, fp_index, rx_sge_mapping),
2551 BCM_PAGE_SIZE * NUM_RX_SGE_PAGES);
2552 }
2553
2554 /* Tx */
2555 if (!skip_tx_queue(bp, fp_index)) {
2556 /* fastpath tx rings: tx_buf tx_desc */
2557 BNX2X_FREE(bnx2x_fp(bp, fp_index, tx_buf_ring));
2558 BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, tx_desc_ring),
2559 bnx2x_fp(bp, fp_index, tx_desc_mapping),
2560 sizeof(union eth_tx_bd_types) * NUM_TX_BD);
2561 }
2562 /* end of fastpath */
2563 }
2564
2565 void bnx2x_free_fp_mem(struct bnx2x *bp)
2566 {
2567 int i;
2568 for_each_queue(bp, i)
2569 bnx2x_free_fp_mem_at(bp, i);
2570 }
2571
2572 static inline void set_sb_shortcuts(struct bnx2x *bp, int index)
2573 {
2574 union host_hc_status_block status_blk = bnx2x_fp(bp, index, status_blk);
2575 if (CHIP_IS_E2(bp)) {
2576 bnx2x_fp(bp, index, sb_index_values) =
2577 (__le16 *)status_blk.e2_sb->sb.index_values;
2578 bnx2x_fp(bp, index, sb_running_index) =
2579 (__le16 *)status_blk.e2_sb->sb.running_index;
2580 } else {
2581 bnx2x_fp(bp, index, sb_index_values) =
2582 (__le16 *)status_blk.e1x_sb->sb.index_values;
2583 bnx2x_fp(bp, index, sb_running_index) =
2584 (__le16 *)status_blk.e1x_sb->sb.running_index;
2585 }
2586 }
2587
2588 static int bnx2x_alloc_fp_mem_at(struct bnx2x *bp, int index)
2589 {
2590 union host_hc_status_block *sb;
2591 struct bnx2x_fastpath *fp = &bp->fp[index];
2592 int ring_size = 0;
2593
2594 /* if rx_ring_size specified - use it */
2595 int rx_ring_size = bp->rx_ring_size ? bp->rx_ring_size :
2596 MAX_RX_AVAIL/bp->num_queues;
2597
2598 /* allocate at least number of buffers required by FW */
2599 rx_ring_size = max_t(int, fp->disable_tpa ? MIN_RX_SIZE_NONTPA :
2600 MIN_RX_SIZE_TPA,
2601 rx_ring_size);
2602
2603 bnx2x_fp(bp, index, bp) = bp;
2604 bnx2x_fp(bp, index, index) = index;
2605
2606 /* Common */
2607 sb = &bnx2x_fp(bp, index, status_blk);
2608 #ifdef BCM_CNIC
2609 if (!IS_FCOE_IDX(index)) {
2610 #endif
2611 /* status blocks */
2612 if (CHIP_IS_E2(bp))
2613 BNX2X_PCI_ALLOC(sb->e2_sb,
2614 &bnx2x_fp(bp, index, status_blk_mapping),
2615 sizeof(struct host_hc_status_block_e2));
2616 else
2617 BNX2X_PCI_ALLOC(sb->e1x_sb,
2618 &bnx2x_fp(bp, index, status_blk_mapping),
2619 sizeof(struct host_hc_status_block_e1x));
2620 #ifdef BCM_CNIC
2621 }
2622 #endif
2623 set_sb_shortcuts(bp, index);
2624
2625 /* Tx */
2626 if (!skip_tx_queue(bp, index)) {
2627 /* fastpath tx rings: tx_buf tx_desc */
2628 BNX2X_ALLOC(bnx2x_fp(bp, index, tx_buf_ring),
2629 sizeof(struct sw_tx_bd) * NUM_TX_BD);
2630 BNX2X_PCI_ALLOC(bnx2x_fp(bp, index, tx_desc_ring),
2631 &bnx2x_fp(bp, index, tx_desc_mapping),
2632 sizeof(union eth_tx_bd_types) * NUM_TX_BD);
2633 }
2634
2635 /* Rx */
2636 if (!skip_rx_queue(bp, index)) {
2637 /* fastpath rx rings: rx_buf rx_desc rx_comp */
2638 BNX2X_ALLOC(bnx2x_fp(bp, index, rx_buf_ring),
2639 sizeof(struct sw_rx_bd) * NUM_RX_BD);
2640 BNX2X_PCI_ALLOC(bnx2x_fp(bp, index, rx_desc_ring),
2641 &bnx2x_fp(bp, index, rx_desc_mapping),
2642 sizeof(struct eth_rx_bd) * NUM_RX_BD);
2643
2644 BNX2X_PCI_ALLOC(bnx2x_fp(bp, index, rx_comp_ring),
2645 &bnx2x_fp(bp, index, rx_comp_mapping),
2646 sizeof(struct eth_fast_path_rx_cqe) *
2647 NUM_RCQ_BD);
2648
2649 /* SGE ring */
2650 BNX2X_ALLOC(bnx2x_fp(bp, index, rx_page_ring),
2651 sizeof(struct sw_rx_page) * NUM_RX_SGE);
2652 BNX2X_PCI_ALLOC(bnx2x_fp(bp, index, rx_sge_ring),
2653 &bnx2x_fp(bp, index, rx_sge_mapping),
2654 BCM_PAGE_SIZE * NUM_RX_SGE_PAGES);
2655 /* RX BD ring */
2656 bnx2x_set_next_page_rx_bd(fp);
2657
2658 /* CQ ring */
2659 bnx2x_set_next_page_rx_cq(fp);
2660
2661 /* BDs */
2662 ring_size = bnx2x_alloc_rx_bds(fp, rx_ring_size);
2663 if (ring_size < rx_ring_size)
2664 goto alloc_mem_err;
2665 }
2666
2667 return 0;
2668
2669 /* handles low memory cases */
2670 alloc_mem_err:
2671 BNX2X_ERR("Unable to allocate full memory for queue %d (size %d)\n",
2672 index, ring_size);
2673 /* FW will drop all packets if queue is not big enough,
2674 * In these cases we disable the queue
2675 * Min size diferent for TPA and non-TPA queues
2676 */
2677 if (ring_size < (fp->disable_tpa ?
2678 MIN_RX_SIZE_NONTPA : MIN_RX_SIZE_TPA)) {
2679 /* release memory allocated for this queue */
2680 bnx2x_free_fp_mem_at(bp, index);
2681 return -ENOMEM;
2682 }
2683 return 0;
2684 }
2685
2686 int bnx2x_alloc_fp_mem(struct bnx2x *bp)
2687 {
2688 int i;
2689
2690 /**
2691 * 1. Allocate FP for leading - fatal if error
2692 * 2. {CNIC} Allocate FCoE FP - fatal if error
2693 * 3. Allocate RSS - fix number of queues if error
2694 */
2695
2696 /* leading */
2697 if (bnx2x_alloc_fp_mem_at(bp, 0))
2698 return -ENOMEM;
2699 #ifdef BCM_CNIC
2700 /* FCoE */
2701 if (bnx2x_alloc_fp_mem_at(bp, FCOE_IDX))
2702 return -ENOMEM;
2703 #endif
2704 /* RSS */
2705 for_each_nondefault_eth_queue(bp, i)
2706 if (bnx2x_alloc_fp_mem_at(bp, i))
2707 break;
2708
2709 /* handle memory failures */
2710 if (i != BNX2X_NUM_ETH_QUEUES(bp)) {
2711 int delta = BNX2X_NUM_ETH_QUEUES(bp) - i;
2712
2713 WARN_ON(delta < 0);
2714 #ifdef BCM_CNIC
2715 /**
2716 * move non eth FPs next to last eth FP
2717 * must be done in that order
2718 * FCOE_IDX < FWD_IDX < OOO_IDX
2719 */
2720
2721 /* move FCoE fp */
2722 bnx2x_move_fp(bp, FCOE_IDX, FCOE_IDX - delta);
2723 #endif
2724 bp->num_queues -= delta;
2725 BNX2X_ERR("Adjusted num of queues from %d to %d\n",
2726 bp->num_queues + delta, bp->num_queues);
2727 }
2728
2729 return 0;
2730 }
2731
2732 static int bnx2x_setup_irqs(struct bnx2x *bp)
2733 {
2734 int rc = 0;
2735 if (bp->flags & USING_MSIX_FLAG) {
2736 rc = bnx2x_req_msix_irqs(bp);
2737 if (rc)
2738 return rc;
2739 } else {
2740 bnx2x_ack_int(bp);
2741 rc = bnx2x_req_irq(bp);
2742 if (rc) {
2743 BNX2X_ERR("IRQ request failed rc %d, aborting\n", rc);
2744 return rc;
2745 }
2746 if (bp->flags & USING_MSI_FLAG) {
2747 bp->dev->irq = bp->pdev->irq;
2748 netdev_info(bp->dev, "using MSI IRQ %d\n",
2749 bp->pdev->irq);
2750 }
2751 }
2752
2753 return 0;
2754 }
2755
2756 void bnx2x_free_mem_bp(struct bnx2x *bp)
2757 {
2758 kfree(bp->fp);
2759 kfree(bp->msix_table);
2760 kfree(bp->ilt);
2761 }
2762
2763 int __devinit bnx2x_alloc_mem_bp(struct bnx2x *bp)
2764 {
2765 struct bnx2x_fastpath *fp;
2766 struct msix_entry *tbl;
2767 struct bnx2x_ilt *ilt;
2768
2769 /* fp array */
2770 fp = kzalloc(L2_FP_COUNT(bp->l2_cid_count)*sizeof(*fp), GFP_KERNEL);
2771 if (!fp)
2772 goto alloc_err;
2773 bp->fp = fp;
2774
2775 /* msix table */
2776 tbl = kzalloc((FP_SB_COUNT(bp->l2_cid_count) + 1) * sizeof(*tbl),
2777 GFP_KERNEL);
2778 if (!tbl)
2779 goto alloc_err;
2780 bp->msix_table = tbl;
2781
2782 /* ilt */
2783 ilt = kzalloc(sizeof(*ilt), GFP_KERNEL);
2784 if (!ilt)
2785 goto alloc_err;
2786 bp->ilt = ilt;
2787
2788 return 0;
2789 alloc_err:
2790 bnx2x_free_mem_bp(bp);
2791 return -ENOMEM;
2792
2793 }
2794
2795 static int bnx2x_reload_if_running(struct net_device *dev)
2796 {
2797 struct bnx2x *bp = netdev_priv(dev);
2798
2799 if (unlikely(!netif_running(dev)))
2800 return 0;
2801
2802 bnx2x_nic_unload(bp, UNLOAD_NORMAL);
2803 return bnx2x_nic_load(bp, LOAD_NORMAL);
2804 }
2805
2806 /* called with rtnl_lock */
2807 int bnx2x_change_mtu(struct net_device *dev, int new_mtu)
2808 {
2809 struct bnx2x *bp = netdev_priv(dev);
2810
2811 if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
2812 printk(KERN_ERR "Handling parity error recovery. Try again later\n");
2813 return -EAGAIN;
2814 }
2815
2816 if ((new_mtu > ETH_MAX_JUMBO_PACKET_SIZE) ||
2817 ((new_mtu + ETH_HLEN) < ETH_MIN_PACKET_SIZE))
2818 return -EINVAL;
2819
2820 /* This does not race with packet allocation
2821 * because the actual alloc size is
2822 * only updated as part of load
2823 */
2824 dev->mtu = new_mtu;
2825
2826 return bnx2x_reload_if_running(dev);
2827 }
2828
2829 u32 bnx2x_fix_features(struct net_device *dev, u32 features)
2830 {
2831 struct bnx2x *bp = netdev_priv(dev);
2832
2833 /* TPA requires Rx CSUM offloading */
2834 if (!(features & NETIF_F_RXCSUM) || bp->disable_tpa)
2835 features &= ~NETIF_F_LRO;
2836
2837 return features;
2838 }
2839
2840 int bnx2x_set_features(struct net_device *dev, u32 features)
2841 {
2842 struct bnx2x *bp = netdev_priv(dev);
2843 u32 flags = bp->flags;
2844 bool bnx2x_reload = false;
2845
2846 if (features & NETIF_F_LRO)
2847 flags |= TPA_ENABLE_FLAG;
2848 else
2849 flags &= ~TPA_ENABLE_FLAG;
2850
2851 if (features & NETIF_F_LOOPBACK) {
2852 if (bp->link_params.loopback_mode != LOOPBACK_BMAC) {
2853 bp->link_params.loopback_mode = LOOPBACK_BMAC;
2854 bnx2x_reload = true;
2855 }
2856 } else {
2857 if (bp->link_params.loopback_mode != LOOPBACK_NONE) {
2858 bp->link_params.loopback_mode = LOOPBACK_NONE;
2859 bnx2x_reload = true;
2860 }
2861 }
2862
2863 if (flags ^ bp->flags) {
2864 bp->flags = flags;
2865 bnx2x_reload = true;
2866 }
2867
2868 if (bnx2x_reload) {
2869 if (bp->recovery_state == BNX2X_RECOVERY_DONE)
2870 return bnx2x_reload_if_running(dev);
2871 /* else: bnx2x_nic_load() will be called at end of recovery */
2872 }
2873
2874 return 0;
2875 }
2876
2877 void bnx2x_tx_timeout(struct net_device *dev)
2878 {
2879 struct bnx2x *bp = netdev_priv(dev);
2880
2881 #ifdef BNX2X_STOP_ON_ERROR
2882 if (!bp->panic)
2883 bnx2x_panic();
2884 #endif
2885 /* This allows the netif to be shutdown gracefully before resetting */
2886 schedule_delayed_work(&bp->reset_task, 0);
2887 }
2888
2889 int bnx2x_suspend(struct pci_dev *pdev, pm_message_t state)
2890 {
2891 struct net_device *dev = pci_get_drvdata(pdev);
2892 struct bnx2x *bp;
2893
2894 if (!dev) {
2895 dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
2896 return -ENODEV;
2897 }
2898 bp = netdev_priv(dev);
2899
2900 rtnl_lock();
2901
2902 pci_save_state(pdev);
2903
2904 if (!netif_running(dev)) {
2905 rtnl_unlock();
2906 return 0;
2907 }
2908
2909 netif_device_detach(dev);
2910
2911 bnx2x_nic_unload(bp, UNLOAD_CLOSE);
2912
2913 bnx2x_set_power_state(bp, pci_choose_state(pdev, state));
2914
2915 rtnl_unlock();
2916
2917 return 0;
2918 }
2919
2920 int bnx2x_resume(struct pci_dev *pdev)
2921 {
2922 struct net_device *dev = pci_get_drvdata(pdev);
2923 struct bnx2x *bp;
2924 int rc;
2925
2926 if (!dev) {
2927 dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
2928 return -ENODEV;
2929 }
2930 bp = netdev_priv(dev);
2931
2932 if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
2933 printk(KERN_ERR "Handling parity error recovery. Try again later\n");
2934 return -EAGAIN;
2935 }
2936
2937 rtnl_lock();
2938
2939 pci_restore_state(pdev);
2940
2941 if (!netif_running(dev)) {
2942 rtnl_unlock();
2943 return 0;
2944 }
2945
2946 bnx2x_set_power_state(bp, PCI_D0);
2947 netif_device_attach(dev);
2948
2949 /* Since the chip was reset, clear the FW sequence number */
2950 bp->fw_seq = 0;
2951 rc = bnx2x_nic_load(bp, LOAD_OPEN);
2952
2953 rtnl_unlock();
2954
2955 return rc;
2956 }
This page took 0.090736 seconds and 6 git commands to generate.