Loading...
Note: File does not exist in v6.13.7.
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/interrupt.h>
21#include <linux/ip.h>
22#include <net/ipv6.h>
23#include <net/ip6_checksum.h>
24#include <linux/firmware.h>
25#include <linux/prefetch.h>
26#include "bnx2x_cmn.h"
27#include "bnx2x_init.h"
28#include "bnx2x_sp.h"
29
30
31
32/**
33 * bnx2x_bz_fp - zero content of the fastpath structure.
34 *
35 * @bp: driver handle
36 * @index: fastpath index to be zeroed
37 *
38 * Makes sure the contents of the bp->fp[index].napi is kept
39 * intact.
40 */
41static inline void bnx2x_bz_fp(struct bnx2x *bp, int index)
42{
43 struct bnx2x_fastpath *fp = &bp->fp[index];
44 struct napi_struct orig_napi = fp->napi;
45 /* bzero bnx2x_fastpath contents */
46 memset(fp, 0, sizeof(*fp));
47
48 /* Restore the NAPI object as it has been already initialized */
49 fp->napi = orig_napi;
50
51 fp->bp = bp;
52 fp->index = index;
53 if (IS_ETH_FP(fp))
54 fp->max_cos = bp->max_cos;
55 else
56 /* Special queues support only one CoS */
57 fp->max_cos = 1;
58
59 /*
60 * set the tpa flag for each queue. The tpa flag determines the queue
61 * minimal size so it must be set prior to queue memory allocation
62 */
63 fp->disable_tpa = ((bp->flags & TPA_ENABLE_FLAG) == 0);
64
65#ifdef BCM_CNIC
66 /* We don't want TPA on an FCoE L2 ring */
67 if (IS_FCOE_FP(fp))
68 fp->disable_tpa = 1;
69#endif
70}
71
72/**
73 * bnx2x_move_fp - move content of the fastpath structure.
74 *
75 * @bp: driver handle
76 * @from: source FP index
77 * @to: destination FP index
78 *
79 * Makes sure the contents of the bp->fp[to].napi is kept
80 * intact.
81 */
82static inline void bnx2x_move_fp(struct bnx2x *bp, int from, int to)
83{
84 struct bnx2x_fastpath *from_fp = &bp->fp[from];
85 struct bnx2x_fastpath *to_fp = &bp->fp[to];
86 struct napi_struct orig_napi = to_fp->napi;
87 /* Move bnx2x_fastpath contents */
88 memcpy(to_fp, from_fp, sizeof(*to_fp));
89 to_fp->index = to;
90
91 /* Restore the NAPI object as it has been already initialized */
92 to_fp->napi = orig_napi;
93}
94
95int load_count[2][3] = { {0} }; /* per-path: 0-common, 1-port0, 2-port1 */
96
97/* free skb in the packet ring at pos idx
98 * return idx of last bd freed
99 */
100static u16 bnx2x_free_tx_pkt(struct bnx2x *bp, struct bnx2x_fp_txdata *txdata,
101 u16 idx)
102{
103 struct sw_tx_bd *tx_buf = &txdata->tx_buf_ring[idx];
104 struct eth_tx_start_bd *tx_start_bd;
105 struct eth_tx_bd *tx_data_bd;
106 struct sk_buff *skb = tx_buf->skb;
107 u16 bd_idx = TX_BD(tx_buf->first_bd), new_cons;
108 int nbd;
109
110 /* prefetch skb end pointer to speedup dev_kfree_skb() */
111 prefetch(&skb->end);
112
113 DP(BNX2X_MSG_FP, "fp[%d]: pkt_idx %d buff @(%p)->skb %p\n",
114 txdata->txq_index, idx, tx_buf, skb);
115
116 /* unmap first bd */
117 DP(BNX2X_MSG_OFF, "free bd_idx %d\n", bd_idx);
118 tx_start_bd = &txdata->tx_desc_ring[bd_idx].start_bd;
119 dma_unmap_single(&bp->pdev->dev, BD_UNMAP_ADDR(tx_start_bd),
120 BD_UNMAP_LEN(tx_start_bd), DMA_TO_DEVICE);
121
122
123 nbd = le16_to_cpu(tx_start_bd->nbd) - 1;
124#ifdef BNX2X_STOP_ON_ERROR
125 if ((nbd - 1) > (MAX_SKB_FRAGS + 2)) {
126 BNX2X_ERR("BAD nbd!\n");
127 bnx2x_panic();
128 }
129#endif
130 new_cons = nbd + tx_buf->first_bd;
131
132 /* Get the next bd */
133 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
134
135 /* Skip a parse bd... */
136 --nbd;
137 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
138
139 /* ...and the TSO split header bd since they have no mapping */
140 if (tx_buf->flags & BNX2X_TSO_SPLIT_BD) {
141 --nbd;
142 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
143 }
144
145 /* now free frags */
146 while (nbd > 0) {
147
148 DP(BNX2X_MSG_OFF, "free frag bd_idx %d\n", bd_idx);
149 tx_data_bd = &txdata->tx_desc_ring[bd_idx].reg_bd;
150 dma_unmap_page(&bp->pdev->dev, BD_UNMAP_ADDR(tx_data_bd),
151 BD_UNMAP_LEN(tx_data_bd), DMA_TO_DEVICE);
152 if (--nbd)
153 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
154 }
155
156 /* release skb */
157 WARN_ON(!skb);
158 dev_kfree_skb_any(skb);
159 tx_buf->first_bd = 0;
160 tx_buf->skb = NULL;
161
162 return new_cons;
163}
164
165int bnx2x_tx_int(struct bnx2x *bp, struct bnx2x_fp_txdata *txdata)
166{
167 struct netdev_queue *txq;
168 u16 hw_cons, sw_cons, bd_cons = txdata->tx_bd_cons;
169
170#ifdef BNX2X_STOP_ON_ERROR
171 if (unlikely(bp->panic))
172 return -1;
173#endif
174
175 txq = netdev_get_tx_queue(bp->dev, txdata->txq_index);
176 hw_cons = le16_to_cpu(*txdata->tx_cons_sb);
177 sw_cons = txdata->tx_pkt_cons;
178
179 while (sw_cons != hw_cons) {
180 u16 pkt_cons;
181
182 pkt_cons = TX_BD(sw_cons);
183
184 DP(NETIF_MSG_TX_DONE, "queue[%d]: hw_cons %u sw_cons %u "
185 " pkt_cons %u\n",
186 txdata->txq_index, hw_cons, sw_cons, pkt_cons);
187
188 bd_cons = bnx2x_free_tx_pkt(bp, txdata, pkt_cons);
189 sw_cons++;
190 }
191
192 txdata->tx_pkt_cons = sw_cons;
193 txdata->tx_bd_cons = bd_cons;
194
195 /* Need to make the tx_bd_cons update visible to start_xmit()
196 * before checking for netif_tx_queue_stopped(). Without the
197 * memory barrier, there is a small possibility that
198 * start_xmit() will miss it and cause the queue to be stopped
199 * forever.
200 * On the other hand we need an rmb() here to ensure the proper
201 * ordering of bit testing in the following
202 * netif_tx_queue_stopped(txq) call.
203 */
204 smp_mb();
205
206 if (unlikely(netif_tx_queue_stopped(txq))) {
207 /* Taking tx_lock() is needed to prevent reenabling the queue
208 * while it's empty. This could have happen if rx_action() gets
209 * suspended in bnx2x_tx_int() after the condition before
210 * netif_tx_wake_queue(), while tx_action (bnx2x_start_xmit()):
211 *
212 * stops the queue->sees fresh tx_bd_cons->releases the queue->
213 * sends some packets consuming the whole queue again->
214 * stops the queue
215 */
216
217 __netif_tx_lock(txq, smp_processor_id());
218
219 if ((netif_tx_queue_stopped(txq)) &&
220 (bp->state == BNX2X_STATE_OPEN) &&
221 (bnx2x_tx_avail(bp, txdata) >= MAX_SKB_FRAGS + 3))
222 netif_tx_wake_queue(txq);
223
224 __netif_tx_unlock(txq);
225 }
226 return 0;
227}
228
229static inline void bnx2x_update_last_max_sge(struct bnx2x_fastpath *fp,
230 u16 idx)
231{
232 u16 last_max = fp->last_max_sge;
233
234 if (SUB_S16(idx, last_max) > 0)
235 fp->last_max_sge = idx;
236}
237
238static void bnx2x_update_sge_prod(struct bnx2x_fastpath *fp,
239 struct eth_fast_path_rx_cqe *fp_cqe)
240{
241 struct bnx2x *bp = fp->bp;
242 u16 sge_len = SGE_PAGE_ALIGN(le16_to_cpu(fp_cqe->pkt_len) -
243 le16_to_cpu(fp_cqe->len_on_bd)) >>
244 SGE_PAGE_SHIFT;
245 u16 last_max, last_elem, first_elem;
246 u16 delta = 0;
247 u16 i;
248
249 if (!sge_len)
250 return;
251
252 /* First mark all used pages */
253 for (i = 0; i < sge_len; i++)
254 BIT_VEC64_CLEAR_BIT(fp->sge_mask,
255 RX_SGE(le16_to_cpu(fp_cqe->sgl_or_raw_data.sgl[i])));
256
257 DP(NETIF_MSG_RX_STATUS, "fp_cqe->sgl[%d] = %d\n",
258 sge_len - 1, le16_to_cpu(fp_cqe->sgl_or_raw_data.sgl[sge_len - 1]));
259
260 /* Here we assume that the last SGE index is the biggest */
261 prefetch((void *)(fp->sge_mask));
262 bnx2x_update_last_max_sge(fp,
263 le16_to_cpu(fp_cqe->sgl_or_raw_data.sgl[sge_len - 1]));
264
265 last_max = RX_SGE(fp->last_max_sge);
266 last_elem = last_max >> BIT_VEC64_ELEM_SHIFT;
267 first_elem = RX_SGE(fp->rx_sge_prod) >> BIT_VEC64_ELEM_SHIFT;
268
269 /* If ring is not full */
270 if (last_elem + 1 != first_elem)
271 last_elem++;
272
273 /* Now update the prod */
274 for (i = first_elem; i != last_elem; i = NEXT_SGE_MASK_ELEM(i)) {
275 if (likely(fp->sge_mask[i]))
276 break;
277
278 fp->sge_mask[i] = BIT_VEC64_ELEM_ONE_MASK;
279 delta += BIT_VEC64_ELEM_SZ;
280 }
281
282 if (delta > 0) {
283 fp->rx_sge_prod += delta;
284 /* clear page-end entries */
285 bnx2x_clear_sge_mask_next_elems(fp);
286 }
287
288 DP(NETIF_MSG_RX_STATUS,
289 "fp->last_max_sge = %d fp->rx_sge_prod = %d\n",
290 fp->last_max_sge, fp->rx_sge_prod);
291}
292
293static void bnx2x_tpa_start(struct bnx2x_fastpath *fp, u16 queue,
294 struct sk_buff *skb, u16 cons, u16 prod,
295 struct eth_fast_path_rx_cqe *cqe)
296{
297 struct bnx2x *bp = fp->bp;
298 struct sw_rx_bd *cons_rx_buf = &fp->rx_buf_ring[cons];
299 struct sw_rx_bd *prod_rx_buf = &fp->rx_buf_ring[prod];
300 struct eth_rx_bd *prod_bd = &fp->rx_desc_ring[prod];
301 dma_addr_t mapping;
302 struct bnx2x_agg_info *tpa_info = &fp->tpa_info[queue];
303 struct sw_rx_bd *first_buf = &tpa_info->first_buf;
304
305 /* print error if current state != stop */
306 if (tpa_info->tpa_state != BNX2X_TPA_STOP)
307 BNX2X_ERR("start of bin not in stop [%d]\n", queue);
308
309 /* Try to map an empty skb from the aggregation info */
310 mapping = dma_map_single(&bp->pdev->dev,
311 first_buf->skb->data,
312 fp->rx_buf_size, DMA_FROM_DEVICE);
313 /*
314 * ...if it fails - move the skb from the consumer to the producer
315 * and set the current aggregation state as ERROR to drop it
316 * when TPA_STOP arrives.
317 */
318
319 if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
320 /* Move the BD from the consumer to the producer */
321 bnx2x_reuse_rx_skb(fp, cons, prod);
322 tpa_info->tpa_state = BNX2X_TPA_ERROR;
323 return;
324 }
325
326 /* move empty skb from pool to prod */
327 prod_rx_buf->skb = first_buf->skb;
328 dma_unmap_addr_set(prod_rx_buf, mapping, mapping);
329 /* point prod_bd to new skb */
330 prod_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
331 prod_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
332
333 /* move partial skb from cons to pool (don't unmap yet) */
334 *first_buf = *cons_rx_buf;
335
336 /* mark bin state as START */
337 tpa_info->parsing_flags =
338 le16_to_cpu(cqe->pars_flags.flags);
339 tpa_info->vlan_tag = le16_to_cpu(cqe->vlan_tag);
340 tpa_info->tpa_state = BNX2X_TPA_START;
341 tpa_info->len_on_bd = le16_to_cpu(cqe->len_on_bd);
342 tpa_info->placement_offset = cqe->placement_offset;
343
344#ifdef BNX2X_STOP_ON_ERROR
345 fp->tpa_queue_used |= (1 << queue);
346#ifdef _ASM_GENERIC_INT_L64_H
347 DP(NETIF_MSG_RX_STATUS, "fp->tpa_queue_used = 0x%lx\n",
348#else
349 DP(NETIF_MSG_RX_STATUS, "fp->tpa_queue_used = 0x%llx\n",
350#endif
351 fp->tpa_queue_used);
352#endif
353}
354
355/* Timestamp option length allowed for TPA aggregation:
356 *
357 * nop nop kind length echo val
358 */
359#define TPA_TSTAMP_OPT_LEN 12
360/**
361 * bnx2x_set_lro_mss - calculate the approximate value of the MSS
362 *
363 * @bp: driver handle
364 * @parsing_flags: parsing flags from the START CQE
365 * @len_on_bd: total length of the first packet for the
366 * aggregation.
367 *
368 * Approximate value of the MSS for this aggregation calculated using
369 * the first packet of it.
370 */
371static inline u16 bnx2x_set_lro_mss(struct bnx2x *bp, u16 parsing_flags,
372 u16 len_on_bd)
373{
374 /*
375 * TPA arrgregation won't have either IP options or TCP options
376 * other than timestamp or IPv6 extension headers.
377 */
378 u16 hdrs_len = ETH_HLEN + sizeof(struct tcphdr);
379
380 if (GET_FLAG(parsing_flags, PARSING_FLAGS_OVER_ETHERNET_PROTOCOL) ==
381 PRS_FLAG_OVERETH_IPV6)
382 hdrs_len += sizeof(struct ipv6hdr);
383 else /* IPv4 */
384 hdrs_len += sizeof(struct iphdr);
385
386
387 /* Check if there was a TCP timestamp, if there is it's will
388 * always be 12 bytes length: nop nop kind length echo val.
389 *
390 * Otherwise FW would close the aggregation.
391 */
392 if (parsing_flags & PARSING_FLAGS_TIME_STAMP_EXIST_FLAG)
393 hdrs_len += TPA_TSTAMP_OPT_LEN;
394
395 return len_on_bd - hdrs_len;
396}
397
398static int bnx2x_fill_frag_skb(struct bnx2x *bp, struct bnx2x_fastpath *fp,
399 u16 queue, struct sk_buff *skb,
400 struct eth_end_agg_rx_cqe *cqe,
401 u16 cqe_idx)
402{
403 struct sw_rx_page *rx_pg, old_rx_pg;
404 u32 i, frag_len, frag_size, pages;
405 int err;
406 int j;
407 struct bnx2x_agg_info *tpa_info = &fp->tpa_info[queue];
408 u16 len_on_bd = tpa_info->len_on_bd;
409
410 frag_size = le16_to_cpu(cqe->pkt_len) - len_on_bd;
411 pages = SGE_PAGE_ALIGN(frag_size) >> SGE_PAGE_SHIFT;
412
413 /* This is needed in order to enable forwarding support */
414 if (frag_size)
415 skb_shinfo(skb)->gso_size = bnx2x_set_lro_mss(bp,
416 tpa_info->parsing_flags, len_on_bd);
417
418#ifdef BNX2X_STOP_ON_ERROR
419 if (pages > min_t(u32, 8, MAX_SKB_FRAGS)*SGE_PAGE_SIZE*PAGES_PER_SGE) {
420 BNX2X_ERR("SGL length is too long: %d. CQE index is %d\n",
421 pages, cqe_idx);
422 BNX2X_ERR("cqe->pkt_len = %d\n", cqe->pkt_len);
423 bnx2x_panic();
424 return -EINVAL;
425 }
426#endif
427
428 /* Run through the SGL and compose the fragmented skb */
429 for (i = 0, j = 0; i < pages; i += PAGES_PER_SGE, j++) {
430 u16 sge_idx = RX_SGE(le16_to_cpu(cqe->sgl_or_raw_data.sgl[j]));
431
432 /* FW gives the indices of the SGE as if the ring is an array
433 (meaning that "next" element will consume 2 indices) */
434 frag_len = min(frag_size, (u32)(SGE_PAGE_SIZE*PAGES_PER_SGE));
435 rx_pg = &fp->rx_page_ring[sge_idx];
436 old_rx_pg = *rx_pg;
437
438 /* If we fail to allocate a substitute page, we simply stop
439 where we are and drop the whole packet */
440 err = bnx2x_alloc_rx_sge(bp, fp, sge_idx);
441 if (unlikely(err)) {
442 fp->eth_q_stats.rx_skb_alloc_failed++;
443 return err;
444 }
445
446 /* Unmap the page as we r going to pass it to the stack */
447 dma_unmap_page(&bp->pdev->dev,
448 dma_unmap_addr(&old_rx_pg, mapping),
449 SGE_PAGE_SIZE*PAGES_PER_SGE, DMA_FROM_DEVICE);
450
451 /* Add one frag and update the appropriate fields in the skb */
452 skb_fill_page_desc(skb, j, old_rx_pg.page, 0, frag_len);
453
454 skb->data_len += frag_len;
455 skb->truesize += frag_len;
456 skb->len += frag_len;
457
458 frag_size -= frag_len;
459 }
460
461 return 0;
462}
463
464static void bnx2x_tpa_stop(struct bnx2x *bp, struct bnx2x_fastpath *fp,
465 u16 queue, struct eth_end_agg_rx_cqe *cqe,
466 u16 cqe_idx)
467{
468 struct bnx2x_agg_info *tpa_info = &fp->tpa_info[queue];
469 struct sw_rx_bd *rx_buf = &tpa_info->first_buf;
470 u8 pad = tpa_info->placement_offset;
471 u16 len = tpa_info->len_on_bd;
472 struct sk_buff *skb = rx_buf->skb;
473 /* alloc new skb */
474 struct sk_buff *new_skb;
475 u8 old_tpa_state = tpa_info->tpa_state;
476
477 tpa_info->tpa_state = BNX2X_TPA_STOP;
478
479 /* If we there was an error during the handling of the TPA_START -
480 * drop this aggregation.
481 */
482 if (old_tpa_state == BNX2X_TPA_ERROR)
483 goto drop;
484
485 /* Try to allocate the new skb */
486 new_skb = netdev_alloc_skb(bp->dev, fp->rx_buf_size);
487
488 /* Unmap skb in the pool anyway, as we are going to change
489 pool entry status to BNX2X_TPA_STOP even if new skb allocation
490 fails. */
491 dma_unmap_single(&bp->pdev->dev, dma_unmap_addr(rx_buf, mapping),
492 fp->rx_buf_size, DMA_FROM_DEVICE);
493
494 if (likely(new_skb)) {
495 prefetch(skb);
496 prefetch(((char *)(skb)) + L1_CACHE_BYTES);
497
498#ifdef BNX2X_STOP_ON_ERROR
499 if (pad + len > fp->rx_buf_size) {
500 BNX2X_ERR("skb_put is about to fail... "
501 "pad %d len %d rx_buf_size %d\n",
502 pad, len, fp->rx_buf_size);
503 bnx2x_panic();
504 return;
505 }
506#endif
507
508 skb_reserve(skb, pad);
509 skb_put(skb, len);
510
511 skb->protocol = eth_type_trans(skb, bp->dev);
512 skb->ip_summed = CHECKSUM_UNNECESSARY;
513
514 if (!bnx2x_fill_frag_skb(bp, fp, queue, skb, cqe, cqe_idx)) {
515 if (tpa_info->parsing_flags & PARSING_FLAGS_VLAN)
516 __vlan_hwaccel_put_tag(skb, tpa_info->vlan_tag);
517 napi_gro_receive(&fp->napi, skb);
518 } else {
519 DP(NETIF_MSG_RX_STATUS, "Failed to allocate new pages"
520 " - dropping packet!\n");
521 dev_kfree_skb_any(skb);
522 }
523
524
525 /* put new skb in bin */
526 rx_buf->skb = new_skb;
527
528 return;
529 }
530
531drop:
532 /* drop the packet and keep the buffer in the bin */
533 DP(NETIF_MSG_RX_STATUS,
534 "Failed to allocate or map a new skb - dropping packet!\n");
535 fp->eth_q_stats.rx_skb_alloc_failed++;
536}
537
538/* Set Toeplitz hash value in the skb using the value from the
539 * CQE (calculated by HW).
540 */
541static inline void bnx2x_set_skb_rxhash(struct bnx2x *bp, union eth_rx_cqe *cqe,
542 struct sk_buff *skb)
543{
544 /* Set Toeplitz hash from CQE */
545 if ((bp->dev->features & NETIF_F_RXHASH) &&
546 (cqe->fast_path_cqe.status_flags &
547 ETH_FAST_PATH_RX_CQE_RSS_HASH_FLG))
548 skb->rxhash =
549 le32_to_cpu(cqe->fast_path_cqe.rss_hash_result);
550}
551
552int bnx2x_rx_int(struct bnx2x_fastpath *fp, int budget)
553{
554 struct bnx2x *bp = fp->bp;
555 u16 bd_cons, bd_prod, bd_prod_fw, comp_ring_cons;
556 u16 hw_comp_cons, sw_comp_cons, sw_comp_prod;
557 int rx_pkt = 0;
558
559#ifdef BNX2X_STOP_ON_ERROR
560 if (unlikely(bp->panic))
561 return 0;
562#endif
563
564 /* CQ "next element" is of the size of the regular element,
565 that's why it's ok here */
566 hw_comp_cons = le16_to_cpu(*fp->rx_cons_sb);
567 if ((hw_comp_cons & MAX_RCQ_DESC_CNT) == MAX_RCQ_DESC_CNT)
568 hw_comp_cons++;
569
570 bd_cons = fp->rx_bd_cons;
571 bd_prod = fp->rx_bd_prod;
572 bd_prod_fw = bd_prod;
573 sw_comp_cons = fp->rx_comp_cons;
574 sw_comp_prod = fp->rx_comp_prod;
575
576 /* Memory barrier necessary as speculative reads of the rx
577 * buffer can be ahead of the index in the status block
578 */
579 rmb();
580
581 DP(NETIF_MSG_RX_STATUS,
582 "queue[%d]: hw_comp_cons %u sw_comp_cons %u\n",
583 fp->index, hw_comp_cons, sw_comp_cons);
584
585 while (sw_comp_cons != hw_comp_cons) {
586 struct sw_rx_bd *rx_buf = NULL;
587 struct sk_buff *skb;
588 union eth_rx_cqe *cqe;
589 struct eth_fast_path_rx_cqe *cqe_fp;
590 u8 cqe_fp_flags;
591 enum eth_rx_cqe_type cqe_fp_type;
592 u16 len, pad;
593
594#ifdef BNX2X_STOP_ON_ERROR
595 if (unlikely(bp->panic))
596 return 0;
597#endif
598
599 comp_ring_cons = RCQ_BD(sw_comp_cons);
600 bd_prod = RX_BD(bd_prod);
601 bd_cons = RX_BD(bd_cons);
602
603 /* Prefetch the page containing the BD descriptor
604 at producer's index. It will be needed when new skb is
605 allocated */
606 prefetch((void *)(PAGE_ALIGN((unsigned long)
607 (&fp->rx_desc_ring[bd_prod])) -
608 PAGE_SIZE + 1));
609
610 cqe = &fp->rx_comp_ring[comp_ring_cons];
611 cqe_fp = &cqe->fast_path_cqe;
612 cqe_fp_flags = cqe_fp->type_error_flags;
613 cqe_fp_type = cqe_fp_flags & ETH_FAST_PATH_RX_CQE_TYPE;
614
615 DP(NETIF_MSG_RX_STATUS, "CQE type %x err %x status %x"
616 " queue %x vlan %x len %u\n", CQE_TYPE(cqe_fp_flags),
617 cqe_fp_flags, cqe_fp->status_flags,
618 le32_to_cpu(cqe_fp->rss_hash_result),
619 le16_to_cpu(cqe_fp->vlan_tag), le16_to_cpu(cqe_fp->pkt_len));
620
621 /* is this a slowpath msg? */
622 if (unlikely(CQE_TYPE_SLOW(cqe_fp_type))) {
623 bnx2x_sp_event(fp, cqe);
624 goto next_cqe;
625
626 /* this is an rx packet */
627 } else {
628 rx_buf = &fp->rx_buf_ring[bd_cons];
629 skb = rx_buf->skb;
630 prefetch(skb);
631
632 if (!CQE_TYPE_FAST(cqe_fp_type)) {
633#ifdef BNX2X_STOP_ON_ERROR
634 /* sanity check */
635 if (fp->disable_tpa &&
636 (CQE_TYPE_START(cqe_fp_type) ||
637 CQE_TYPE_STOP(cqe_fp_type)))
638 BNX2X_ERR("START/STOP packet while "
639 "disable_tpa type %x\n",
640 CQE_TYPE(cqe_fp_type));
641#endif
642
643 if (CQE_TYPE_START(cqe_fp_type)) {
644 u16 queue = cqe_fp->queue_index;
645 DP(NETIF_MSG_RX_STATUS,
646 "calling tpa_start on queue %d\n",
647 queue);
648
649 bnx2x_tpa_start(fp, queue, skb,
650 bd_cons, bd_prod,
651 cqe_fp);
652
653 /* Set Toeplitz hash for LRO skb */
654 bnx2x_set_skb_rxhash(bp, cqe, skb);
655
656 goto next_rx;
657
658 } else {
659 u16 queue =
660 cqe->end_agg_cqe.queue_index;
661 DP(NETIF_MSG_RX_STATUS,
662 "calling tpa_stop on queue %d\n",
663 queue);
664
665 bnx2x_tpa_stop(bp, fp, queue,
666 &cqe->end_agg_cqe,
667 comp_ring_cons);
668#ifdef BNX2X_STOP_ON_ERROR
669 if (bp->panic)
670 return 0;
671#endif
672
673 bnx2x_update_sge_prod(fp, cqe_fp);
674 goto next_cqe;
675 }
676 }
677 /* non TPA */
678 len = le16_to_cpu(cqe_fp->pkt_len);
679 pad = cqe_fp->placement_offset;
680 dma_sync_single_for_cpu(&bp->pdev->dev,
681 dma_unmap_addr(rx_buf, mapping),
682 pad + RX_COPY_THRESH,
683 DMA_FROM_DEVICE);
684 prefetch(((char *)(skb)) + L1_CACHE_BYTES);
685
686 /* is this an error packet? */
687 if (unlikely(cqe_fp_flags & ETH_RX_ERROR_FALGS)) {
688 DP(NETIF_MSG_RX_ERR,
689 "ERROR flags %x rx packet %u\n",
690 cqe_fp_flags, sw_comp_cons);
691 fp->eth_q_stats.rx_err_discard_pkt++;
692 goto reuse_rx;
693 }
694
695 /* Since we don't have a jumbo ring
696 * copy small packets if mtu > 1500
697 */
698 if ((bp->dev->mtu > ETH_MAX_PACKET_SIZE) &&
699 (len <= RX_COPY_THRESH)) {
700 struct sk_buff *new_skb;
701
702 new_skb = netdev_alloc_skb(bp->dev, len + pad);
703 if (new_skb == NULL) {
704 DP(NETIF_MSG_RX_ERR,
705 "ERROR packet dropped "
706 "because of alloc failure\n");
707 fp->eth_q_stats.rx_skb_alloc_failed++;
708 goto reuse_rx;
709 }
710
711 /* aligned copy */
712 skb_copy_from_linear_data_offset(skb, pad,
713 new_skb->data + pad, len);
714 skb_reserve(new_skb, pad);
715 skb_put(new_skb, len);
716
717 bnx2x_reuse_rx_skb(fp, bd_cons, bd_prod);
718
719 skb = new_skb;
720
721 } else
722 if (likely(bnx2x_alloc_rx_skb(bp, fp, bd_prod) == 0)) {
723 dma_unmap_single(&bp->pdev->dev,
724 dma_unmap_addr(rx_buf, mapping),
725 fp->rx_buf_size,
726 DMA_FROM_DEVICE);
727 skb_reserve(skb, pad);
728 skb_put(skb, len);
729
730 } else {
731 DP(NETIF_MSG_RX_ERR,
732 "ERROR packet dropped because "
733 "of alloc failure\n");
734 fp->eth_q_stats.rx_skb_alloc_failed++;
735reuse_rx:
736 bnx2x_reuse_rx_skb(fp, bd_cons, bd_prod);
737 goto next_rx;
738 }
739
740 skb->protocol = eth_type_trans(skb, bp->dev);
741
742 /* Set Toeplitz hash for a none-LRO skb */
743 bnx2x_set_skb_rxhash(bp, cqe, skb);
744
745 skb_checksum_none_assert(skb);
746
747 if (bp->dev->features & NETIF_F_RXCSUM) {
748
749 if (likely(BNX2X_RX_CSUM_OK(cqe)))
750 skb->ip_summed = CHECKSUM_UNNECESSARY;
751 else
752 fp->eth_q_stats.hw_csum_err++;
753 }
754 }
755
756 skb_record_rx_queue(skb, fp->index);
757
758 if (le16_to_cpu(cqe_fp->pars_flags.flags) &
759 PARSING_FLAGS_VLAN)
760 __vlan_hwaccel_put_tag(skb,
761 le16_to_cpu(cqe_fp->vlan_tag));
762 napi_gro_receive(&fp->napi, skb);
763
764
765next_rx:
766 rx_buf->skb = NULL;
767
768 bd_cons = NEXT_RX_IDX(bd_cons);
769 bd_prod = NEXT_RX_IDX(bd_prod);
770 bd_prod_fw = NEXT_RX_IDX(bd_prod_fw);
771 rx_pkt++;
772next_cqe:
773 sw_comp_prod = NEXT_RCQ_IDX(sw_comp_prod);
774 sw_comp_cons = NEXT_RCQ_IDX(sw_comp_cons);
775
776 if (rx_pkt == budget)
777 break;
778 } /* while */
779
780 fp->rx_bd_cons = bd_cons;
781 fp->rx_bd_prod = bd_prod_fw;
782 fp->rx_comp_cons = sw_comp_cons;
783 fp->rx_comp_prod = sw_comp_prod;
784
785 /* Update producers */
786 bnx2x_update_rx_prod(bp, fp, bd_prod_fw, sw_comp_prod,
787 fp->rx_sge_prod);
788
789 fp->rx_pkt += rx_pkt;
790 fp->rx_calls++;
791
792 return rx_pkt;
793}
794
795static irqreturn_t bnx2x_msix_fp_int(int irq, void *fp_cookie)
796{
797 struct bnx2x_fastpath *fp = fp_cookie;
798 struct bnx2x *bp = fp->bp;
799 u8 cos;
800
801 DP(BNX2X_MSG_FP, "got an MSI-X interrupt on IDX:SB "
802 "[fp %d fw_sd %d igusb %d]\n",
803 fp->index, fp->fw_sb_id, fp->igu_sb_id);
804 bnx2x_ack_sb(bp, fp->igu_sb_id, USTORM_ID, 0, IGU_INT_DISABLE, 0);
805
806#ifdef BNX2X_STOP_ON_ERROR
807 if (unlikely(bp->panic))
808 return IRQ_HANDLED;
809#endif
810
811 /* Handle Rx and Tx according to MSI-X vector */
812 prefetch(fp->rx_cons_sb);
813
814 for_each_cos_in_tx_queue(fp, cos)
815 prefetch(fp->txdata[cos].tx_cons_sb);
816
817 prefetch(&fp->sb_running_index[SM_RX_ID]);
818 napi_schedule(&bnx2x_fp(bp, fp->index, napi));
819
820 return IRQ_HANDLED;
821}
822
823/* HW Lock for shared dual port PHYs */
824void bnx2x_acquire_phy_lock(struct bnx2x *bp)
825{
826 mutex_lock(&bp->port.phy_mutex);
827
828 if (bp->port.need_hw_lock)
829 bnx2x_acquire_hw_lock(bp, HW_LOCK_RESOURCE_MDIO);
830}
831
832void bnx2x_release_phy_lock(struct bnx2x *bp)
833{
834 if (bp->port.need_hw_lock)
835 bnx2x_release_hw_lock(bp, HW_LOCK_RESOURCE_MDIO);
836
837 mutex_unlock(&bp->port.phy_mutex);
838}
839
840/* calculates MF speed according to current linespeed and MF configuration */
841u16 bnx2x_get_mf_speed(struct bnx2x *bp)
842{
843 u16 line_speed = bp->link_vars.line_speed;
844 if (IS_MF(bp)) {
845 u16 maxCfg = bnx2x_extract_max_cfg(bp,
846 bp->mf_config[BP_VN(bp)]);
847
848 /* Calculate the current MAX line speed limit for the MF
849 * devices
850 */
851 if (IS_MF_SI(bp))
852 line_speed = (line_speed * maxCfg) / 100;
853 else { /* SD mode */
854 u16 vn_max_rate = maxCfg * 100;
855
856 if (vn_max_rate < line_speed)
857 line_speed = vn_max_rate;
858 }
859 }
860
861 return line_speed;
862}
863
864/**
865 * bnx2x_fill_report_data - fill link report data to report
866 *
867 * @bp: driver handle
868 * @data: link state to update
869 *
870 * It uses a none-atomic bit operations because is called under the mutex.
871 */
872static inline void bnx2x_fill_report_data(struct bnx2x *bp,
873 struct bnx2x_link_report_data *data)
874{
875 u16 line_speed = bnx2x_get_mf_speed(bp);
876
877 memset(data, 0, sizeof(*data));
878
879 /* Fill the report data: efective line speed */
880 data->line_speed = line_speed;
881
882 /* Link is down */
883 if (!bp->link_vars.link_up || (bp->flags & MF_FUNC_DIS))
884 __set_bit(BNX2X_LINK_REPORT_LINK_DOWN,
885 &data->link_report_flags);
886
887 /* Full DUPLEX */
888 if (bp->link_vars.duplex == DUPLEX_FULL)
889 __set_bit(BNX2X_LINK_REPORT_FD, &data->link_report_flags);
890
891 /* Rx Flow Control is ON */
892 if (bp->link_vars.flow_ctrl & BNX2X_FLOW_CTRL_RX)
893 __set_bit(BNX2X_LINK_REPORT_RX_FC_ON, &data->link_report_flags);
894
895 /* Tx Flow Control is ON */
896 if (bp->link_vars.flow_ctrl & BNX2X_FLOW_CTRL_TX)
897 __set_bit(BNX2X_LINK_REPORT_TX_FC_ON, &data->link_report_flags);
898}
899
900/**
901 * bnx2x_link_report - report link status to OS.
902 *
903 * @bp: driver handle
904 *
905 * Calls the __bnx2x_link_report() under the same locking scheme
906 * as a link/PHY state managing code to ensure a consistent link
907 * reporting.
908 */
909
910void bnx2x_link_report(struct bnx2x *bp)
911{
912 bnx2x_acquire_phy_lock(bp);
913 __bnx2x_link_report(bp);
914 bnx2x_release_phy_lock(bp);
915}
916
917/**
918 * __bnx2x_link_report - report link status to OS.
919 *
920 * @bp: driver handle
921 *
922 * None atomic inmlementation.
923 * Should be called under the phy_lock.
924 */
925void __bnx2x_link_report(struct bnx2x *bp)
926{
927 struct bnx2x_link_report_data cur_data;
928
929 /* reread mf_cfg */
930 if (!CHIP_IS_E1(bp))
931 bnx2x_read_mf_cfg(bp);
932
933 /* Read the current link report info */
934 bnx2x_fill_report_data(bp, &cur_data);
935
936 /* Don't report link down or exactly the same link status twice */
937 if (!memcmp(&cur_data, &bp->last_reported_link, sizeof(cur_data)) ||
938 (test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
939 &bp->last_reported_link.link_report_flags) &&
940 test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
941 &cur_data.link_report_flags)))
942 return;
943
944 bp->link_cnt++;
945
946 /* We are going to report a new link parameters now -
947 * remember the current data for the next time.
948 */
949 memcpy(&bp->last_reported_link, &cur_data, sizeof(cur_data));
950
951 if (test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
952 &cur_data.link_report_flags)) {
953 netif_carrier_off(bp->dev);
954 netdev_err(bp->dev, "NIC Link is Down\n");
955 return;
956 } else {
957 netif_carrier_on(bp->dev);
958 netdev_info(bp->dev, "NIC Link is Up, ");
959 pr_cont("%d Mbps ", cur_data.line_speed);
960
961 if (test_and_clear_bit(BNX2X_LINK_REPORT_FD,
962 &cur_data.link_report_flags))
963 pr_cont("full duplex");
964 else
965 pr_cont("half duplex");
966
967 /* Handle the FC at the end so that only these flags would be
968 * possibly set. This way we may easily check if there is no FC
969 * enabled.
970 */
971 if (cur_data.link_report_flags) {
972 if (test_bit(BNX2X_LINK_REPORT_RX_FC_ON,
973 &cur_data.link_report_flags)) {
974 pr_cont(", receive ");
975 if (test_bit(BNX2X_LINK_REPORT_TX_FC_ON,
976 &cur_data.link_report_flags))
977 pr_cont("& transmit ");
978 } else {
979 pr_cont(", transmit ");
980 }
981 pr_cont("flow control ON");
982 }
983 pr_cont("\n");
984 }
985}
986
987void bnx2x_init_rx_rings(struct bnx2x *bp)
988{
989 int func = BP_FUNC(bp);
990 u16 ring_prod;
991 int i, j;
992
993 /* Allocate TPA resources */
994 for_each_rx_queue(bp, j) {
995 struct bnx2x_fastpath *fp = &bp->fp[j];
996
997 DP(NETIF_MSG_IFUP,
998 "mtu %d rx_buf_size %d\n", bp->dev->mtu, fp->rx_buf_size);
999
1000 if (!fp->disable_tpa) {
1001 /* Fill the per-aggregtion pool */
1002 for (i = 0; i < MAX_AGG_QS(bp); i++) {
1003 struct bnx2x_agg_info *tpa_info =
1004 &fp->tpa_info[i];
1005 struct sw_rx_bd *first_buf =
1006 &tpa_info->first_buf;
1007
1008 first_buf->skb = netdev_alloc_skb(bp->dev,
1009 fp->rx_buf_size);
1010 if (!first_buf->skb) {
1011 BNX2X_ERR("Failed to allocate TPA "
1012 "skb pool for queue[%d] - "
1013 "disabling TPA on this "
1014 "queue!\n", j);
1015 bnx2x_free_tpa_pool(bp, fp, i);
1016 fp->disable_tpa = 1;
1017 break;
1018 }
1019 dma_unmap_addr_set(first_buf, mapping, 0);
1020 tpa_info->tpa_state = BNX2X_TPA_STOP;
1021 }
1022
1023 /* "next page" elements initialization */
1024 bnx2x_set_next_page_sgl(fp);
1025
1026 /* set SGEs bit mask */
1027 bnx2x_init_sge_ring_bit_mask(fp);
1028
1029 /* Allocate SGEs and initialize the ring elements */
1030 for (i = 0, ring_prod = 0;
1031 i < MAX_RX_SGE_CNT*NUM_RX_SGE_PAGES; i++) {
1032
1033 if (bnx2x_alloc_rx_sge(bp, fp, ring_prod) < 0) {
1034 BNX2X_ERR("was only able to allocate "
1035 "%d rx sges\n", i);
1036 BNX2X_ERR("disabling TPA for "
1037 "queue[%d]\n", j);
1038 /* Cleanup already allocated elements */
1039 bnx2x_free_rx_sge_range(bp, fp,
1040 ring_prod);
1041 bnx2x_free_tpa_pool(bp, fp,
1042 MAX_AGG_QS(bp));
1043 fp->disable_tpa = 1;
1044 ring_prod = 0;
1045 break;
1046 }
1047 ring_prod = NEXT_SGE_IDX(ring_prod);
1048 }
1049
1050 fp->rx_sge_prod = ring_prod;
1051 }
1052 }
1053
1054 for_each_rx_queue(bp, j) {
1055 struct bnx2x_fastpath *fp = &bp->fp[j];
1056
1057 fp->rx_bd_cons = 0;
1058
1059 /* Activate BD ring */
1060 /* Warning!
1061 * this will generate an interrupt (to the TSTORM)
1062 * must only be done after chip is initialized
1063 */
1064 bnx2x_update_rx_prod(bp, fp, fp->rx_bd_prod, fp->rx_comp_prod,
1065 fp->rx_sge_prod);
1066
1067 if (j != 0)
1068 continue;
1069
1070 if (CHIP_IS_E1(bp)) {
1071 REG_WR(bp, BAR_USTRORM_INTMEM +
1072 USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func),
1073 U64_LO(fp->rx_comp_mapping));
1074 REG_WR(bp, BAR_USTRORM_INTMEM +
1075 USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func) + 4,
1076 U64_HI(fp->rx_comp_mapping));
1077 }
1078 }
1079}
1080
1081static void bnx2x_free_tx_skbs(struct bnx2x *bp)
1082{
1083 int i;
1084 u8 cos;
1085
1086 for_each_tx_queue(bp, i) {
1087 struct bnx2x_fastpath *fp = &bp->fp[i];
1088 for_each_cos_in_tx_queue(fp, cos) {
1089 struct bnx2x_fp_txdata *txdata = &fp->txdata[cos];
1090
1091 u16 bd_cons = txdata->tx_bd_cons;
1092 u16 sw_prod = txdata->tx_pkt_prod;
1093 u16 sw_cons = txdata->tx_pkt_cons;
1094
1095 while (sw_cons != sw_prod) {
1096 bd_cons = bnx2x_free_tx_pkt(bp, txdata,
1097 TX_BD(sw_cons));
1098 sw_cons++;
1099 }
1100 }
1101 }
1102}
1103
1104static void bnx2x_free_rx_bds(struct bnx2x_fastpath *fp)
1105{
1106 struct bnx2x *bp = fp->bp;
1107 int i;
1108
1109 /* ring wasn't allocated */
1110 if (fp->rx_buf_ring == NULL)
1111 return;
1112
1113 for (i = 0; i < NUM_RX_BD; i++) {
1114 struct sw_rx_bd *rx_buf = &fp->rx_buf_ring[i];
1115 struct sk_buff *skb = rx_buf->skb;
1116
1117 if (skb == NULL)
1118 continue;
1119 dma_unmap_single(&bp->pdev->dev,
1120 dma_unmap_addr(rx_buf, mapping),
1121 fp->rx_buf_size, DMA_FROM_DEVICE);
1122
1123 rx_buf->skb = NULL;
1124 dev_kfree_skb(skb);
1125 }
1126}
1127
1128static void bnx2x_free_rx_skbs(struct bnx2x *bp)
1129{
1130 int j;
1131
1132 for_each_rx_queue(bp, j) {
1133 struct bnx2x_fastpath *fp = &bp->fp[j];
1134
1135 bnx2x_free_rx_bds(fp);
1136
1137 if (!fp->disable_tpa)
1138 bnx2x_free_tpa_pool(bp, fp, MAX_AGG_QS(bp));
1139 }
1140}
1141
1142void bnx2x_free_skbs(struct bnx2x *bp)
1143{
1144 bnx2x_free_tx_skbs(bp);
1145 bnx2x_free_rx_skbs(bp);
1146}
1147
1148void bnx2x_update_max_mf_config(struct bnx2x *bp, u32 value)
1149{
1150 /* load old values */
1151 u32 mf_cfg = bp->mf_config[BP_VN(bp)];
1152
1153 if (value != bnx2x_extract_max_cfg(bp, mf_cfg)) {
1154 /* leave all but MAX value */
1155 mf_cfg &= ~FUNC_MF_CFG_MAX_BW_MASK;
1156
1157 /* set new MAX value */
1158 mf_cfg |= (value << FUNC_MF_CFG_MAX_BW_SHIFT)
1159 & FUNC_MF_CFG_MAX_BW_MASK;
1160
1161 bnx2x_fw_command(bp, DRV_MSG_CODE_SET_MF_BW, mf_cfg);
1162 }
1163}
1164
1165/**
1166 * bnx2x_free_msix_irqs - free previously requested MSI-X IRQ vectors
1167 *
1168 * @bp: driver handle
1169 * @nvecs: number of vectors to be released
1170 */
1171static void bnx2x_free_msix_irqs(struct bnx2x *bp, int nvecs)
1172{
1173 int i, offset = 0;
1174
1175 if (nvecs == offset)
1176 return;
1177 free_irq(bp->msix_table[offset].vector, bp->dev);
1178 DP(NETIF_MSG_IFDOWN, "released sp irq (%d)\n",
1179 bp->msix_table[offset].vector);
1180 offset++;
1181#ifdef BCM_CNIC
1182 if (nvecs == offset)
1183 return;
1184 offset++;
1185#endif
1186
1187 for_each_eth_queue(bp, i) {
1188 if (nvecs == offset)
1189 return;
1190 DP(NETIF_MSG_IFDOWN, "about to release fp #%d->%d "
1191 "irq\n", i, bp->msix_table[offset].vector);
1192
1193 free_irq(bp->msix_table[offset++].vector, &bp->fp[i]);
1194 }
1195}
1196
1197void bnx2x_free_irq(struct bnx2x *bp)
1198{
1199 if (bp->flags & USING_MSIX_FLAG)
1200 bnx2x_free_msix_irqs(bp, BNX2X_NUM_ETH_QUEUES(bp) +
1201 CNIC_PRESENT + 1);
1202 else if (bp->flags & USING_MSI_FLAG)
1203 free_irq(bp->pdev->irq, bp->dev);
1204 else
1205 free_irq(bp->pdev->irq, bp->dev);
1206}
1207
1208int bnx2x_enable_msix(struct bnx2x *bp)
1209{
1210 int msix_vec = 0, i, rc, req_cnt;
1211
1212 bp->msix_table[msix_vec].entry = msix_vec;
1213 DP(NETIF_MSG_IFUP, "msix_table[0].entry = %d (slowpath)\n",
1214 bp->msix_table[0].entry);
1215 msix_vec++;
1216
1217#ifdef BCM_CNIC
1218 bp->msix_table[msix_vec].entry = msix_vec;
1219 DP(NETIF_MSG_IFUP, "msix_table[%d].entry = %d (CNIC)\n",
1220 bp->msix_table[msix_vec].entry, bp->msix_table[msix_vec].entry);
1221 msix_vec++;
1222#endif
1223 /* We need separate vectors for ETH queues only (not FCoE) */
1224 for_each_eth_queue(bp, i) {
1225 bp->msix_table[msix_vec].entry = msix_vec;
1226 DP(NETIF_MSG_IFUP, "msix_table[%d].entry = %d "
1227 "(fastpath #%u)\n", msix_vec, msix_vec, i);
1228 msix_vec++;
1229 }
1230
1231 req_cnt = BNX2X_NUM_ETH_QUEUES(bp) + CNIC_PRESENT + 1;
1232
1233 rc = pci_enable_msix(bp->pdev, &bp->msix_table[0], req_cnt);
1234
1235 /*
1236 * reconfigure number of tx/rx queues according to available
1237 * MSI-X vectors
1238 */
1239 if (rc >= BNX2X_MIN_MSIX_VEC_CNT) {
1240 /* how less vectors we will have? */
1241 int diff = req_cnt - rc;
1242
1243 DP(NETIF_MSG_IFUP,
1244 "Trying to use less MSI-X vectors: %d\n", rc);
1245
1246 rc = pci_enable_msix(bp->pdev, &bp->msix_table[0], rc);
1247
1248 if (rc) {
1249 DP(NETIF_MSG_IFUP,
1250 "MSI-X is not attainable rc %d\n", rc);
1251 return rc;
1252 }
1253 /*
1254 * decrease number of queues by number of unallocated entries
1255 */
1256 bp->num_queues -= diff;
1257
1258 DP(NETIF_MSG_IFUP, "New queue configuration set: %d\n",
1259 bp->num_queues);
1260 } else if (rc) {
1261 /* fall to INTx if not enough memory */
1262 if (rc == -ENOMEM)
1263 bp->flags |= DISABLE_MSI_FLAG;
1264 DP(NETIF_MSG_IFUP, "MSI-X is not attainable rc %d\n", rc);
1265 return rc;
1266 }
1267
1268 bp->flags |= USING_MSIX_FLAG;
1269
1270 return 0;
1271}
1272
1273static int bnx2x_req_msix_irqs(struct bnx2x *bp)
1274{
1275 int i, rc, offset = 0;
1276
1277 rc = request_irq(bp->msix_table[offset++].vector,
1278 bnx2x_msix_sp_int, 0,
1279 bp->dev->name, bp->dev);
1280 if (rc) {
1281 BNX2X_ERR("request sp irq failed\n");
1282 return -EBUSY;
1283 }
1284
1285#ifdef BCM_CNIC
1286 offset++;
1287#endif
1288 for_each_eth_queue(bp, i) {
1289 struct bnx2x_fastpath *fp = &bp->fp[i];
1290 snprintf(fp->name, sizeof(fp->name), "%s-fp-%d",
1291 bp->dev->name, i);
1292
1293 rc = request_irq(bp->msix_table[offset].vector,
1294 bnx2x_msix_fp_int, 0, fp->name, fp);
1295 if (rc) {
1296 BNX2X_ERR("request fp #%d irq (%d) failed rc %d\n", i,
1297 bp->msix_table[offset].vector, rc);
1298 bnx2x_free_msix_irqs(bp, offset);
1299 return -EBUSY;
1300 }
1301
1302 offset++;
1303 }
1304
1305 i = BNX2X_NUM_ETH_QUEUES(bp);
1306 offset = 1 + CNIC_PRESENT;
1307 netdev_info(bp->dev, "using MSI-X IRQs: sp %d fp[%d] %d"
1308 " ... fp[%d] %d\n",
1309 bp->msix_table[0].vector,
1310 0, bp->msix_table[offset].vector,
1311 i - 1, bp->msix_table[offset + i - 1].vector);
1312
1313 return 0;
1314}
1315
1316int bnx2x_enable_msi(struct bnx2x *bp)
1317{
1318 int rc;
1319
1320 rc = pci_enable_msi(bp->pdev);
1321 if (rc) {
1322 DP(NETIF_MSG_IFUP, "MSI is not attainable\n");
1323 return -1;
1324 }
1325 bp->flags |= USING_MSI_FLAG;
1326
1327 return 0;
1328}
1329
1330static int bnx2x_req_irq(struct bnx2x *bp)
1331{
1332 unsigned long flags;
1333 int rc;
1334
1335 if (bp->flags & USING_MSI_FLAG)
1336 flags = 0;
1337 else
1338 flags = IRQF_SHARED;
1339
1340 rc = request_irq(bp->pdev->irq, bnx2x_interrupt, flags,
1341 bp->dev->name, bp->dev);
1342 return rc;
1343}
1344
1345static inline int bnx2x_setup_irqs(struct bnx2x *bp)
1346{
1347 int rc = 0;
1348 if (bp->flags & USING_MSIX_FLAG) {
1349 rc = bnx2x_req_msix_irqs(bp);
1350 if (rc)
1351 return rc;
1352 } else {
1353 bnx2x_ack_int(bp);
1354 rc = bnx2x_req_irq(bp);
1355 if (rc) {
1356 BNX2X_ERR("IRQ request failed rc %d, aborting\n", rc);
1357 return rc;
1358 }
1359 if (bp->flags & USING_MSI_FLAG) {
1360 bp->dev->irq = bp->pdev->irq;
1361 netdev_info(bp->dev, "using MSI IRQ %d\n",
1362 bp->pdev->irq);
1363 }
1364 }
1365
1366 return 0;
1367}
1368
1369static inline void bnx2x_napi_enable(struct bnx2x *bp)
1370{
1371 int i;
1372
1373 for_each_rx_queue(bp, i)
1374 napi_enable(&bnx2x_fp(bp, i, napi));
1375}
1376
1377static inline void bnx2x_napi_disable(struct bnx2x *bp)
1378{
1379 int i;
1380
1381 for_each_rx_queue(bp, i)
1382 napi_disable(&bnx2x_fp(bp, i, napi));
1383}
1384
1385void bnx2x_netif_start(struct bnx2x *bp)
1386{
1387 if (netif_running(bp->dev)) {
1388 bnx2x_napi_enable(bp);
1389 bnx2x_int_enable(bp);
1390 if (bp->state == BNX2X_STATE_OPEN)
1391 netif_tx_wake_all_queues(bp->dev);
1392 }
1393}
1394
1395void bnx2x_netif_stop(struct bnx2x *bp, int disable_hw)
1396{
1397 bnx2x_int_disable_sync(bp, disable_hw);
1398 bnx2x_napi_disable(bp);
1399}
1400
1401u16 bnx2x_select_queue(struct net_device *dev, struct sk_buff *skb)
1402{
1403 struct bnx2x *bp = netdev_priv(dev);
1404
1405#ifdef BCM_CNIC
1406 if (!NO_FCOE(bp)) {
1407 struct ethhdr *hdr = (struct ethhdr *)skb->data;
1408 u16 ether_type = ntohs(hdr->h_proto);
1409
1410 /* Skip VLAN tag if present */
1411 if (ether_type == ETH_P_8021Q) {
1412 struct vlan_ethhdr *vhdr =
1413 (struct vlan_ethhdr *)skb->data;
1414
1415 ether_type = ntohs(vhdr->h_vlan_encapsulated_proto);
1416 }
1417
1418 /* If ethertype is FCoE or FIP - use FCoE ring */
1419 if ((ether_type == ETH_P_FCOE) || (ether_type == ETH_P_FIP))
1420 return bnx2x_fcoe_tx(bp, txq_index);
1421 }
1422#endif
1423 /* select a non-FCoE queue */
1424 return __skb_tx_hash(dev, skb, BNX2X_NUM_ETH_QUEUES(bp));
1425}
1426
1427void bnx2x_set_num_queues(struct bnx2x *bp)
1428{
1429 switch (bp->multi_mode) {
1430 case ETH_RSS_MODE_DISABLED:
1431 bp->num_queues = 1;
1432 break;
1433 case ETH_RSS_MODE_REGULAR:
1434 bp->num_queues = bnx2x_calc_num_queues(bp);
1435 break;
1436
1437 default:
1438 bp->num_queues = 1;
1439 break;
1440 }
1441
1442 /* Add special queues */
1443 bp->num_queues += NON_ETH_CONTEXT_USE;
1444}
1445
1446/**
1447 * bnx2x_set_real_num_queues - configure netdev->real_num_[tx,rx]_queues
1448 *
1449 * @bp: Driver handle
1450 *
1451 * We currently support for at most 16 Tx queues for each CoS thus we will
1452 * allocate a multiple of 16 for ETH L2 rings according to the value of the
1453 * bp->max_cos.
1454 *
1455 * If there is an FCoE L2 queue the appropriate Tx queue will have the next
1456 * index after all ETH L2 indices.
1457 *
1458 * If the actual number of Tx queues (for each CoS) is less than 16 then there
1459 * will be the holes at the end of each group of 16 ETh L2 indices (0..15,
1460 * 16..31,...) with indicies that are not coupled with any real Tx queue.
1461 *
1462 * The proper configuration of skb->queue_mapping is handled by
1463 * bnx2x_select_queue() and __skb_tx_hash().
1464 *
1465 * bnx2x_setup_tc() takes care of the proper TC mappings so that __skb_tx_hash()
1466 * will return a proper Tx index if TC is enabled (netdev->num_tc > 0).
1467 */
1468static inline int bnx2x_set_real_num_queues(struct bnx2x *bp)
1469{
1470 int rc, tx, rx;
1471
1472 tx = MAX_TXQS_PER_COS * bp->max_cos;
1473 rx = BNX2X_NUM_ETH_QUEUES(bp);
1474
1475/* account for fcoe queue */
1476#ifdef BCM_CNIC
1477 if (!NO_FCOE(bp)) {
1478 rx += FCOE_PRESENT;
1479 tx += FCOE_PRESENT;
1480 }
1481#endif
1482
1483 rc = netif_set_real_num_tx_queues(bp->dev, tx);
1484 if (rc) {
1485 BNX2X_ERR("Failed to set real number of Tx queues: %d\n", rc);
1486 return rc;
1487 }
1488 rc = netif_set_real_num_rx_queues(bp->dev, rx);
1489 if (rc) {
1490 BNX2X_ERR("Failed to set real number of Rx queues: %d\n", rc);
1491 return rc;
1492 }
1493
1494 DP(NETIF_MSG_DRV, "Setting real num queues to (tx, rx) (%d, %d)\n",
1495 tx, rx);
1496
1497 return rc;
1498}
1499
1500static inline void bnx2x_set_rx_buf_size(struct bnx2x *bp)
1501{
1502 int i;
1503
1504 for_each_queue(bp, i) {
1505 struct bnx2x_fastpath *fp = &bp->fp[i];
1506
1507 /* Always use a mini-jumbo MTU for the FCoE L2 ring */
1508 if (IS_FCOE_IDX(i))
1509 /*
1510 * Although there are no IP frames expected to arrive to
1511 * this ring we still want to add an
1512 * IP_HEADER_ALIGNMENT_PADDING to prevent a buffer
1513 * overrun attack.
1514 */
1515 fp->rx_buf_size =
1516 BNX2X_FCOE_MINI_JUMBO_MTU + ETH_OVREHEAD +
1517 BNX2X_FW_RX_ALIGN + IP_HEADER_ALIGNMENT_PADDING;
1518 else
1519 fp->rx_buf_size =
1520 bp->dev->mtu + ETH_OVREHEAD +
1521 BNX2X_FW_RX_ALIGN + IP_HEADER_ALIGNMENT_PADDING;
1522 }
1523}
1524
1525static inline int bnx2x_init_rss_pf(struct bnx2x *bp)
1526{
1527 int i;
1528 u8 ind_table[T_ETH_INDIRECTION_TABLE_SIZE] = {0};
1529 u8 num_eth_queues = BNX2X_NUM_ETH_QUEUES(bp);
1530
1531 /*
1532 * Prepare the inital contents fo the indirection table if RSS is
1533 * enabled
1534 */
1535 if (bp->multi_mode != ETH_RSS_MODE_DISABLED) {
1536 for (i = 0; i < sizeof(ind_table); i++)
1537 ind_table[i] =
1538 bp->fp->cl_id + (i % num_eth_queues);
1539 }
1540
1541 /*
1542 * For 57710 and 57711 SEARCHER configuration (rss_keys) is
1543 * per-port, so if explicit configuration is needed , do it only
1544 * for a PMF.
1545 *
1546 * For 57712 and newer on the other hand it's a per-function
1547 * configuration.
1548 */
1549 return bnx2x_config_rss_pf(bp, ind_table,
1550 bp->port.pmf || !CHIP_IS_E1x(bp));
1551}
1552
1553int bnx2x_config_rss_pf(struct bnx2x *bp, u8 *ind_table, bool config_hash)
1554{
1555 struct bnx2x_config_rss_params params = {0};
1556 int i;
1557
1558 /* Although RSS is meaningless when there is a single HW queue we
1559 * still need it enabled in order to have HW Rx hash generated.
1560 *
1561 * if (!is_eth_multi(bp))
1562 * bp->multi_mode = ETH_RSS_MODE_DISABLED;
1563 */
1564
1565 params.rss_obj = &bp->rss_conf_obj;
1566
1567 __set_bit(RAMROD_COMP_WAIT, ¶ms.ramrod_flags);
1568
1569 /* RSS mode */
1570 switch (bp->multi_mode) {
1571 case ETH_RSS_MODE_DISABLED:
1572 __set_bit(BNX2X_RSS_MODE_DISABLED, ¶ms.rss_flags);
1573 break;
1574 case ETH_RSS_MODE_REGULAR:
1575 __set_bit(BNX2X_RSS_MODE_REGULAR, ¶ms.rss_flags);
1576 break;
1577 case ETH_RSS_MODE_VLAN_PRI:
1578 __set_bit(BNX2X_RSS_MODE_VLAN_PRI, ¶ms.rss_flags);
1579 break;
1580 case ETH_RSS_MODE_E1HOV_PRI:
1581 __set_bit(BNX2X_RSS_MODE_E1HOV_PRI, ¶ms.rss_flags);
1582 break;
1583 case ETH_RSS_MODE_IP_DSCP:
1584 __set_bit(BNX2X_RSS_MODE_IP_DSCP, ¶ms.rss_flags);
1585 break;
1586 default:
1587 BNX2X_ERR("Unknown multi_mode: %d\n", bp->multi_mode);
1588 return -EINVAL;
1589 }
1590
1591 /* If RSS is enabled */
1592 if (bp->multi_mode != ETH_RSS_MODE_DISABLED) {
1593 /* RSS configuration */
1594 __set_bit(BNX2X_RSS_IPV4, ¶ms.rss_flags);
1595 __set_bit(BNX2X_RSS_IPV4_TCP, ¶ms.rss_flags);
1596 __set_bit(BNX2X_RSS_IPV6, ¶ms.rss_flags);
1597 __set_bit(BNX2X_RSS_IPV6_TCP, ¶ms.rss_flags);
1598
1599 /* Hash bits */
1600 params.rss_result_mask = MULTI_MASK;
1601
1602 memcpy(params.ind_table, ind_table, sizeof(params.ind_table));
1603
1604 if (config_hash) {
1605 /* RSS keys */
1606 for (i = 0; i < sizeof(params.rss_key) / 4; i++)
1607 params.rss_key[i] = random32();
1608
1609 __set_bit(BNX2X_RSS_SET_SRCH, ¶ms.rss_flags);
1610 }
1611 }
1612
1613 return bnx2x_config_rss(bp, ¶ms);
1614}
1615
1616static inline int bnx2x_init_hw(struct bnx2x *bp, u32 load_code)
1617{
1618 struct bnx2x_func_state_params func_params = {0};
1619
1620 /* Prepare parameters for function state transitions */
1621 __set_bit(RAMROD_COMP_WAIT, &func_params.ramrod_flags);
1622
1623 func_params.f_obj = &bp->func_obj;
1624 func_params.cmd = BNX2X_F_CMD_HW_INIT;
1625
1626 func_params.params.hw_init.load_phase = load_code;
1627
1628 return bnx2x_func_state_change(bp, &func_params);
1629}
1630
1631/*
1632 * Cleans the object that have internal lists without sending
1633 * ramrods. Should be run when interrutps are disabled.
1634 */
1635static void bnx2x_squeeze_objects(struct bnx2x *bp)
1636{
1637 int rc;
1638 unsigned long ramrod_flags = 0, vlan_mac_flags = 0;
1639 struct bnx2x_mcast_ramrod_params rparam = {0};
1640 struct bnx2x_vlan_mac_obj *mac_obj = &bp->fp->mac_obj;
1641
1642 /***************** Cleanup MACs' object first *************************/
1643
1644 /* Wait for completion of requested */
1645 __set_bit(RAMROD_COMP_WAIT, &ramrod_flags);
1646 /* Perform a dry cleanup */
1647 __set_bit(RAMROD_DRV_CLR_ONLY, &ramrod_flags);
1648
1649 /* Clean ETH primary MAC */
1650 __set_bit(BNX2X_ETH_MAC, &vlan_mac_flags);
1651 rc = mac_obj->delete_all(bp, &bp->fp->mac_obj, &vlan_mac_flags,
1652 &ramrod_flags);
1653 if (rc != 0)
1654 BNX2X_ERR("Failed to clean ETH MACs: %d\n", rc);
1655
1656 /* Cleanup UC list */
1657 vlan_mac_flags = 0;
1658 __set_bit(BNX2X_UC_LIST_MAC, &vlan_mac_flags);
1659 rc = mac_obj->delete_all(bp, mac_obj, &vlan_mac_flags,
1660 &ramrod_flags);
1661 if (rc != 0)
1662 BNX2X_ERR("Failed to clean UC list MACs: %d\n", rc);
1663
1664 /***************** Now clean mcast object *****************************/
1665 rparam.mcast_obj = &bp->mcast_obj;
1666 __set_bit(RAMROD_DRV_CLR_ONLY, &rparam.ramrod_flags);
1667
1668 /* Add a DEL command... */
1669 rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_DEL);
1670 if (rc < 0)
1671 BNX2X_ERR("Failed to add a new DEL command to a multi-cast "
1672 "object: %d\n", rc);
1673
1674 /* ...and wait until all pending commands are cleared */
1675 rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_CONT);
1676 while (rc != 0) {
1677 if (rc < 0) {
1678 BNX2X_ERR("Failed to clean multi-cast object: %d\n",
1679 rc);
1680 return;
1681 }
1682
1683 rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_CONT);
1684 }
1685}
1686
1687#ifndef BNX2X_STOP_ON_ERROR
1688#define LOAD_ERROR_EXIT(bp, label) \
1689 do { \
1690 (bp)->state = BNX2X_STATE_ERROR; \
1691 goto label; \
1692 } while (0)
1693#else
1694#define LOAD_ERROR_EXIT(bp, label) \
1695 do { \
1696 (bp)->state = BNX2X_STATE_ERROR; \
1697 (bp)->panic = 1; \
1698 return -EBUSY; \
1699 } while (0)
1700#endif
1701
1702/* must be called with rtnl_lock */
1703int bnx2x_nic_load(struct bnx2x *bp, int load_mode)
1704{
1705 int port = BP_PORT(bp);
1706 u32 load_code;
1707 int i, rc;
1708
1709#ifdef BNX2X_STOP_ON_ERROR
1710 if (unlikely(bp->panic))
1711 return -EPERM;
1712#endif
1713
1714 bp->state = BNX2X_STATE_OPENING_WAIT4_LOAD;
1715
1716 /* Set the initial link reported state to link down */
1717 bnx2x_acquire_phy_lock(bp);
1718 memset(&bp->last_reported_link, 0, sizeof(bp->last_reported_link));
1719 __set_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1720 &bp->last_reported_link.link_report_flags);
1721 bnx2x_release_phy_lock(bp);
1722
1723 /* must be called before memory allocation and HW init */
1724 bnx2x_ilt_set_info(bp);
1725
1726 /*
1727 * Zero fastpath structures preserving invariants like napi, which are
1728 * allocated only once, fp index, max_cos, bp pointer.
1729 * Also set fp->disable_tpa.
1730 */
1731 for_each_queue(bp, i)
1732 bnx2x_bz_fp(bp, i);
1733
1734
1735 /* Set the receive queues buffer size */
1736 bnx2x_set_rx_buf_size(bp);
1737
1738 if (bnx2x_alloc_mem(bp))
1739 return -ENOMEM;
1740
1741 /* As long as bnx2x_alloc_mem() may possibly update
1742 * bp->num_queues, bnx2x_set_real_num_queues() should always
1743 * come after it.
1744 */
1745 rc = bnx2x_set_real_num_queues(bp);
1746 if (rc) {
1747 BNX2X_ERR("Unable to set real_num_queues\n");
1748 LOAD_ERROR_EXIT(bp, load_error0);
1749 }
1750
1751 /* configure multi cos mappings in kernel.
1752 * this configuration may be overriden by a multi class queue discipline
1753 * or by a dcbx negotiation result.
1754 */
1755 bnx2x_setup_tc(bp->dev, bp->max_cos);
1756
1757 bnx2x_napi_enable(bp);
1758
1759 /* Send LOAD_REQUEST command to MCP
1760 * Returns the type of LOAD command:
1761 * if it is the first port to be initialized
1762 * common blocks should be initialized, otherwise - not
1763 */
1764 if (!BP_NOMCP(bp)) {
1765 load_code = bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_REQ, 0);
1766 if (!load_code) {
1767 BNX2X_ERR("MCP response failure, aborting\n");
1768 rc = -EBUSY;
1769 LOAD_ERROR_EXIT(bp, load_error1);
1770 }
1771 if (load_code == FW_MSG_CODE_DRV_LOAD_REFUSED) {
1772 rc = -EBUSY; /* other port in diagnostic mode */
1773 LOAD_ERROR_EXIT(bp, load_error1);
1774 }
1775
1776 } else {
1777 int path = BP_PATH(bp);
1778
1779 DP(NETIF_MSG_IFUP, "NO MCP - load counts[%d] %d, %d, %d\n",
1780 path, load_count[path][0], load_count[path][1],
1781 load_count[path][2]);
1782 load_count[path][0]++;
1783 load_count[path][1 + port]++;
1784 DP(NETIF_MSG_IFUP, "NO MCP - new load counts[%d] %d, %d, %d\n",
1785 path, load_count[path][0], load_count[path][1],
1786 load_count[path][2]);
1787 if (load_count[path][0] == 1)
1788 load_code = FW_MSG_CODE_DRV_LOAD_COMMON;
1789 else if (load_count[path][1 + port] == 1)
1790 load_code = FW_MSG_CODE_DRV_LOAD_PORT;
1791 else
1792 load_code = FW_MSG_CODE_DRV_LOAD_FUNCTION;
1793 }
1794
1795 if ((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
1796 (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP) ||
1797 (load_code == FW_MSG_CODE_DRV_LOAD_PORT)) {
1798 bp->port.pmf = 1;
1799 /*
1800 * We need the barrier to ensure the ordering between the
1801 * writing to bp->port.pmf here and reading it from the
1802 * bnx2x_periodic_task().
1803 */
1804 smp_mb();
1805 queue_delayed_work(bnx2x_wq, &bp->period_task, 0);
1806 } else
1807 bp->port.pmf = 0;
1808
1809 DP(NETIF_MSG_LINK, "pmf %d\n", bp->port.pmf);
1810
1811 /* Init Function state controlling object */
1812 bnx2x__init_func_obj(bp);
1813
1814 /* Initialize HW */
1815 rc = bnx2x_init_hw(bp, load_code);
1816 if (rc) {
1817 BNX2X_ERR("HW init failed, aborting\n");
1818 bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
1819 LOAD_ERROR_EXIT(bp, load_error2);
1820 }
1821
1822 /* Connect to IRQs */
1823 rc = bnx2x_setup_irqs(bp);
1824 if (rc) {
1825 bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
1826 LOAD_ERROR_EXIT(bp, load_error2);
1827 }
1828
1829 /* Setup NIC internals and enable interrupts */
1830 bnx2x_nic_init(bp, load_code);
1831
1832 /* Init per-function objects */
1833 bnx2x_init_bp_objs(bp);
1834
1835 if (((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
1836 (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP)) &&
1837 (bp->common.shmem2_base)) {
1838 if (SHMEM2_HAS(bp, dcc_support))
1839 SHMEM2_WR(bp, dcc_support,
1840 (SHMEM_DCC_SUPPORT_DISABLE_ENABLE_PF_TLV |
1841 SHMEM_DCC_SUPPORT_BANDWIDTH_ALLOCATION_TLV));
1842 }
1843
1844 bp->state = BNX2X_STATE_OPENING_WAIT4_PORT;
1845 rc = bnx2x_func_start(bp);
1846 if (rc) {
1847 BNX2X_ERR("Function start failed!\n");
1848 bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
1849 LOAD_ERROR_EXIT(bp, load_error3);
1850 }
1851
1852 /* Send LOAD_DONE command to MCP */
1853 if (!BP_NOMCP(bp)) {
1854 load_code = bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
1855 if (!load_code) {
1856 BNX2X_ERR("MCP response failure, aborting\n");
1857 rc = -EBUSY;
1858 LOAD_ERROR_EXIT(bp, load_error3);
1859 }
1860 }
1861
1862 rc = bnx2x_setup_leading(bp);
1863 if (rc) {
1864 BNX2X_ERR("Setup leading failed!\n");
1865 LOAD_ERROR_EXIT(bp, load_error3);
1866 }
1867
1868#ifdef BCM_CNIC
1869 /* Enable Timer scan */
1870 REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + port*4, 1);
1871#endif
1872
1873 for_each_nondefault_queue(bp, i) {
1874 rc = bnx2x_setup_queue(bp, &bp->fp[i], 0);
1875 if (rc)
1876 LOAD_ERROR_EXIT(bp, load_error4);
1877 }
1878
1879 rc = bnx2x_init_rss_pf(bp);
1880 if (rc)
1881 LOAD_ERROR_EXIT(bp, load_error4);
1882
1883 /* Now when Clients are configured we are ready to work */
1884 bp->state = BNX2X_STATE_OPEN;
1885
1886 /* Configure a ucast MAC */
1887 rc = bnx2x_set_eth_mac(bp, true);
1888 if (rc)
1889 LOAD_ERROR_EXIT(bp, load_error4);
1890
1891 if (bp->pending_max) {
1892 bnx2x_update_max_mf_config(bp, bp->pending_max);
1893 bp->pending_max = 0;
1894 }
1895
1896 if (bp->port.pmf)
1897 bnx2x_initial_phy_init(bp, load_mode);
1898
1899 /* Start fast path */
1900
1901 /* Initialize Rx filter. */
1902 netif_addr_lock_bh(bp->dev);
1903 bnx2x_set_rx_mode(bp->dev);
1904 netif_addr_unlock_bh(bp->dev);
1905
1906 /* Start the Tx */
1907 switch (load_mode) {
1908 case LOAD_NORMAL:
1909 /* Tx queue should be only reenabled */
1910 netif_tx_wake_all_queues(bp->dev);
1911 break;
1912
1913 case LOAD_OPEN:
1914 netif_tx_start_all_queues(bp->dev);
1915 smp_mb__after_clear_bit();
1916 break;
1917
1918 case LOAD_DIAG:
1919 bp->state = BNX2X_STATE_DIAG;
1920 break;
1921
1922 default:
1923 break;
1924 }
1925
1926 if (!bp->port.pmf)
1927 bnx2x__link_status_update(bp);
1928
1929 /* start the timer */
1930 mod_timer(&bp->timer, jiffies + bp->current_interval);
1931
1932#ifdef BCM_CNIC
1933 bnx2x_setup_cnic_irq_info(bp);
1934 if (bp->state == BNX2X_STATE_OPEN)
1935 bnx2x_cnic_notify(bp, CNIC_CTL_START_CMD);
1936#endif
1937 bnx2x_inc_load_cnt(bp);
1938
1939 /* Wait for all pending SP commands to complete */
1940 if (!bnx2x_wait_sp_comp(bp, ~0x0UL)) {
1941 BNX2X_ERR("Timeout waiting for SP elements to complete\n");
1942 bnx2x_nic_unload(bp, UNLOAD_CLOSE);
1943 return -EBUSY;
1944 }
1945
1946 bnx2x_dcbx_init(bp);
1947 return 0;
1948
1949#ifndef BNX2X_STOP_ON_ERROR
1950load_error4:
1951#ifdef BCM_CNIC
1952 /* Disable Timer scan */
1953 REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + port*4, 0);
1954#endif
1955load_error3:
1956 bnx2x_int_disable_sync(bp, 1);
1957
1958 /* Clean queueable objects */
1959 bnx2x_squeeze_objects(bp);
1960
1961 /* Free SKBs, SGEs, TPA pool and driver internals */
1962 bnx2x_free_skbs(bp);
1963 for_each_rx_queue(bp, i)
1964 bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
1965
1966 /* Release IRQs */
1967 bnx2x_free_irq(bp);
1968load_error2:
1969 if (!BP_NOMCP(bp)) {
1970 bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_REQ_WOL_MCP, 0);
1971 bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_DONE, 0);
1972 }
1973
1974 bp->port.pmf = 0;
1975load_error1:
1976 bnx2x_napi_disable(bp);
1977load_error0:
1978 bnx2x_free_mem(bp);
1979
1980 return rc;
1981#endif /* ! BNX2X_STOP_ON_ERROR */
1982}
1983
1984/* must be called with rtnl_lock */
1985int bnx2x_nic_unload(struct bnx2x *bp, int unload_mode)
1986{
1987 int i;
1988 bool global = false;
1989
1990 if ((bp->state == BNX2X_STATE_CLOSED) ||
1991 (bp->state == BNX2X_STATE_ERROR)) {
1992 /* We can get here if the driver has been unloaded
1993 * during parity error recovery and is either waiting for a
1994 * leader to complete or for other functions to unload and
1995 * then ifdown has been issued. In this case we want to
1996 * unload and let other functions to complete a recovery
1997 * process.
1998 */
1999 bp->recovery_state = BNX2X_RECOVERY_DONE;
2000 bp->is_leader = 0;
2001 bnx2x_release_leader_lock(bp);
2002 smp_mb();
2003
2004 DP(NETIF_MSG_HW, "Releasing a leadership...\n");
2005
2006 return -EINVAL;
2007 }
2008
2009 /*
2010 * It's important to set the bp->state to the value different from
2011 * BNX2X_STATE_OPEN and only then stop the Tx. Otherwise bnx2x_tx_int()
2012 * may restart the Tx from the NAPI context (see bnx2x_tx_int()).
2013 */
2014 bp->state = BNX2X_STATE_CLOSING_WAIT4_HALT;
2015 smp_mb();
2016
2017 /* Stop Tx */
2018 bnx2x_tx_disable(bp);
2019
2020#ifdef BCM_CNIC
2021 bnx2x_cnic_notify(bp, CNIC_CTL_STOP_CMD);
2022#endif
2023
2024 bp->rx_mode = BNX2X_RX_MODE_NONE;
2025
2026 del_timer_sync(&bp->timer);
2027
2028 /* Set ALWAYS_ALIVE bit in shmem */
2029 bp->fw_drv_pulse_wr_seq |= DRV_PULSE_ALWAYS_ALIVE;
2030
2031 bnx2x_drv_pulse(bp);
2032
2033 bnx2x_stats_handle(bp, STATS_EVENT_STOP);
2034
2035 /* Cleanup the chip if needed */
2036 if (unload_mode != UNLOAD_RECOVERY)
2037 bnx2x_chip_cleanup(bp, unload_mode);
2038 else {
2039 /* Send the UNLOAD_REQUEST to the MCP */
2040 bnx2x_send_unload_req(bp, unload_mode);
2041
2042 /*
2043 * Prevent transactions to host from the functions on the
2044 * engine that doesn't reset global blocks in case of global
2045 * attention once gloabl blocks are reset and gates are opened
2046 * (the engine which leader will perform the recovery
2047 * last).
2048 */
2049 if (!CHIP_IS_E1x(bp))
2050 bnx2x_pf_disable(bp);
2051
2052 /* Disable HW interrupts, NAPI */
2053 bnx2x_netif_stop(bp, 1);
2054
2055 /* Release IRQs */
2056 bnx2x_free_irq(bp);
2057
2058 /* Report UNLOAD_DONE to MCP */
2059 bnx2x_send_unload_done(bp);
2060 }
2061
2062 /*
2063 * At this stage no more interrupts will arrive so we may safly clean
2064 * the queueable objects here in case they failed to get cleaned so far.
2065 */
2066 bnx2x_squeeze_objects(bp);
2067
2068 /* There should be no more pending SP commands at this stage */
2069 bp->sp_state = 0;
2070
2071 bp->port.pmf = 0;
2072
2073 /* Free SKBs, SGEs, TPA pool and driver internals */
2074 bnx2x_free_skbs(bp);
2075 for_each_rx_queue(bp, i)
2076 bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
2077
2078 bnx2x_free_mem(bp);
2079
2080 bp->state = BNX2X_STATE_CLOSED;
2081
2082 /* Check if there are pending parity attentions. If there are - set
2083 * RECOVERY_IN_PROGRESS.
2084 */
2085 if (bnx2x_chk_parity_attn(bp, &global, false)) {
2086 bnx2x_set_reset_in_progress(bp);
2087
2088 /* Set RESET_IS_GLOBAL if needed */
2089 if (global)
2090 bnx2x_set_reset_global(bp);
2091 }
2092
2093
2094 /* The last driver must disable a "close the gate" if there is no
2095 * parity attention or "process kill" pending.
2096 */
2097 if (!bnx2x_dec_load_cnt(bp) && bnx2x_reset_is_done(bp, BP_PATH(bp)))
2098 bnx2x_disable_close_the_gate(bp);
2099
2100 return 0;
2101}
2102
2103int bnx2x_set_power_state(struct bnx2x *bp, pci_power_t state)
2104{
2105 u16 pmcsr;
2106
2107 /* If there is no power capability, silently succeed */
2108 if (!bp->pm_cap) {
2109 DP(NETIF_MSG_HW, "No power capability. Breaking.\n");
2110 return 0;
2111 }
2112
2113 pci_read_config_word(bp->pdev, bp->pm_cap + PCI_PM_CTRL, &pmcsr);
2114
2115 switch (state) {
2116 case PCI_D0:
2117 pci_write_config_word(bp->pdev, bp->pm_cap + PCI_PM_CTRL,
2118 ((pmcsr & ~PCI_PM_CTRL_STATE_MASK) |
2119 PCI_PM_CTRL_PME_STATUS));
2120
2121 if (pmcsr & PCI_PM_CTRL_STATE_MASK)
2122 /* delay required during transition out of D3hot */
2123 msleep(20);
2124 break;
2125
2126 case PCI_D3hot:
2127 /* If there are other clients above don't
2128 shut down the power */
2129 if (atomic_read(&bp->pdev->enable_cnt) != 1)
2130 return 0;
2131 /* Don't shut down the power for emulation and FPGA */
2132 if (CHIP_REV_IS_SLOW(bp))
2133 return 0;
2134
2135 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
2136 pmcsr |= 3;
2137
2138 if (bp->wol)
2139 pmcsr |= PCI_PM_CTRL_PME_ENABLE;
2140
2141 pci_write_config_word(bp->pdev, bp->pm_cap + PCI_PM_CTRL,
2142 pmcsr);
2143
2144 /* No more memory access after this point until
2145 * device is brought back to D0.
2146 */
2147 break;
2148
2149 default:
2150 return -EINVAL;
2151 }
2152 return 0;
2153}
2154
2155/*
2156 * net_device service functions
2157 */
2158int bnx2x_poll(struct napi_struct *napi, int budget)
2159{
2160 int work_done = 0;
2161 u8 cos;
2162 struct bnx2x_fastpath *fp = container_of(napi, struct bnx2x_fastpath,
2163 napi);
2164 struct bnx2x *bp = fp->bp;
2165
2166 while (1) {
2167#ifdef BNX2X_STOP_ON_ERROR
2168 if (unlikely(bp->panic)) {
2169 napi_complete(napi);
2170 return 0;
2171 }
2172#endif
2173
2174 for_each_cos_in_tx_queue(fp, cos)
2175 if (bnx2x_tx_queue_has_work(&fp->txdata[cos]))
2176 bnx2x_tx_int(bp, &fp->txdata[cos]);
2177
2178
2179 if (bnx2x_has_rx_work(fp)) {
2180 work_done += bnx2x_rx_int(fp, budget - work_done);
2181
2182 /* must not complete if we consumed full budget */
2183 if (work_done >= budget)
2184 break;
2185 }
2186
2187 /* Fall out from the NAPI loop if needed */
2188 if (!(bnx2x_has_rx_work(fp) || bnx2x_has_tx_work(fp))) {
2189#ifdef BCM_CNIC
2190 /* No need to update SB for FCoE L2 ring as long as
2191 * it's connected to the default SB and the SB
2192 * has been updated when NAPI was scheduled.
2193 */
2194 if (IS_FCOE_FP(fp)) {
2195 napi_complete(napi);
2196 break;
2197 }
2198#endif
2199
2200 bnx2x_update_fpsb_idx(fp);
2201 /* bnx2x_has_rx_work() reads the status block,
2202 * thus we need to ensure that status block indices
2203 * have been actually read (bnx2x_update_fpsb_idx)
2204 * prior to this check (bnx2x_has_rx_work) so that
2205 * we won't write the "newer" value of the status block
2206 * to IGU (if there was a DMA right after
2207 * bnx2x_has_rx_work and if there is no rmb, the memory
2208 * reading (bnx2x_update_fpsb_idx) may be postponed
2209 * to right before bnx2x_ack_sb). In this case there
2210 * will never be another interrupt until there is
2211 * another update of the status block, while there
2212 * is still unhandled work.
2213 */
2214 rmb();
2215
2216 if (!(bnx2x_has_rx_work(fp) || bnx2x_has_tx_work(fp))) {
2217 napi_complete(napi);
2218 /* Re-enable interrupts */
2219 DP(NETIF_MSG_HW,
2220 "Update index to %d\n", fp->fp_hc_idx);
2221 bnx2x_ack_sb(bp, fp->igu_sb_id, USTORM_ID,
2222 le16_to_cpu(fp->fp_hc_idx),
2223 IGU_INT_ENABLE, 1);
2224 break;
2225 }
2226 }
2227 }
2228
2229 return work_done;
2230}
2231
2232/* we split the first BD into headers and data BDs
2233 * to ease the pain of our fellow microcode engineers
2234 * we use one mapping for both BDs
2235 * So far this has only been observed to happen
2236 * in Other Operating Systems(TM)
2237 */
2238static noinline u16 bnx2x_tx_split(struct bnx2x *bp,
2239 struct bnx2x_fp_txdata *txdata,
2240 struct sw_tx_bd *tx_buf,
2241 struct eth_tx_start_bd **tx_bd, u16 hlen,
2242 u16 bd_prod, int nbd)
2243{
2244 struct eth_tx_start_bd *h_tx_bd = *tx_bd;
2245 struct eth_tx_bd *d_tx_bd;
2246 dma_addr_t mapping;
2247 int old_len = le16_to_cpu(h_tx_bd->nbytes);
2248
2249 /* first fix first BD */
2250 h_tx_bd->nbd = cpu_to_le16(nbd);
2251 h_tx_bd->nbytes = cpu_to_le16(hlen);
2252
2253 DP(NETIF_MSG_TX_QUEUED, "TSO split header size is %d "
2254 "(%x:%x) nbd %d\n", h_tx_bd->nbytes, h_tx_bd->addr_hi,
2255 h_tx_bd->addr_lo, h_tx_bd->nbd);
2256
2257 /* now get a new data BD
2258 * (after the pbd) and fill it */
2259 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
2260 d_tx_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
2261
2262 mapping = HILO_U64(le32_to_cpu(h_tx_bd->addr_hi),
2263 le32_to_cpu(h_tx_bd->addr_lo)) + hlen;
2264
2265 d_tx_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
2266 d_tx_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
2267 d_tx_bd->nbytes = cpu_to_le16(old_len - hlen);
2268
2269 /* this marks the BD as one that has no individual mapping */
2270 tx_buf->flags |= BNX2X_TSO_SPLIT_BD;
2271
2272 DP(NETIF_MSG_TX_QUEUED,
2273 "TSO split data size is %d (%x:%x)\n",
2274 d_tx_bd->nbytes, d_tx_bd->addr_hi, d_tx_bd->addr_lo);
2275
2276 /* update tx_bd */
2277 *tx_bd = (struct eth_tx_start_bd *)d_tx_bd;
2278
2279 return bd_prod;
2280}
2281
2282static inline u16 bnx2x_csum_fix(unsigned char *t_header, u16 csum, s8 fix)
2283{
2284 if (fix > 0)
2285 csum = (u16) ~csum_fold(csum_sub(csum,
2286 csum_partial(t_header - fix, fix, 0)));
2287
2288 else if (fix < 0)
2289 csum = (u16) ~csum_fold(csum_add(csum,
2290 csum_partial(t_header, -fix, 0)));
2291
2292 return swab16(csum);
2293}
2294
2295static inline u32 bnx2x_xmit_type(struct bnx2x *bp, struct sk_buff *skb)
2296{
2297 u32 rc;
2298
2299 if (skb->ip_summed != CHECKSUM_PARTIAL)
2300 rc = XMIT_PLAIN;
2301
2302 else {
2303 if (vlan_get_protocol(skb) == htons(ETH_P_IPV6)) {
2304 rc = XMIT_CSUM_V6;
2305 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2306 rc |= XMIT_CSUM_TCP;
2307
2308 } else {
2309 rc = XMIT_CSUM_V4;
2310 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2311 rc |= XMIT_CSUM_TCP;
2312 }
2313 }
2314
2315 if (skb_is_gso_v6(skb))
2316 rc |= XMIT_GSO_V6 | XMIT_CSUM_TCP | XMIT_CSUM_V6;
2317 else if (skb_is_gso(skb))
2318 rc |= XMIT_GSO_V4 | XMIT_CSUM_V4 | XMIT_CSUM_TCP;
2319
2320 return rc;
2321}
2322
2323#if (MAX_SKB_FRAGS >= MAX_FETCH_BD - 3)
2324/* check if packet requires linearization (packet is too fragmented)
2325 no need to check fragmentation if page size > 8K (there will be no
2326 violation to FW restrictions) */
2327static int bnx2x_pkt_req_lin(struct bnx2x *bp, struct sk_buff *skb,
2328 u32 xmit_type)
2329{
2330 int to_copy = 0;
2331 int hlen = 0;
2332 int first_bd_sz = 0;
2333
2334 /* 3 = 1 (for linear data BD) + 2 (for PBD and last BD) */
2335 if (skb_shinfo(skb)->nr_frags >= (MAX_FETCH_BD - 3)) {
2336
2337 if (xmit_type & XMIT_GSO) {
2338 unsigned short lso_mss = skb_shinfo(skb)->gso_size;
2339 /* Check if LSO packet needs to be copied:
2340 3 = 1 (for headers BD) + 2 (for PBD and last BD) */
2341 int wnd_size = MAX_FETCH_BD - 3;
2342 /* Number of windows to check */
2343 int num_wnds = skb_shinfo(skb)->nr_frags - wnd_size;
2344 int wnd_idx = 0;
2345 int frag_idx = 0;
2346 u32 wnd_sum = 0;
2347
2348 /* Headers length */
2349 hlen = (int)(skb_transport_header(skb) - skb->data) +
2350 tcp_hdrlen(skb);
2351
2352 /* Amount of data (w/o headers) on linear part of SKB*/
2353 first_bd_sz = skb_headlen(skb) - hlen;
2354
2355 wnd_sum = first_bd_sz;
2356
2357 /* Calculate the first sum - it's special */
2358 for (frag_idx = 0; frag_idx < wnd_size - 1; frag_idx++)
2359 wnd_sum +=
2360 skb_shinfo(skb)->frags[frag_idx].size;
2361
2362 /* If there was data on linear skb data - check it */
2363 if (first_bd_sz > 0) {
2364 if (unlikely(wnd_sum < lso_mss)) {
2365 to_copy = 1;
2366 goto exit_lbl;
2367 }
2368
2369 wnd_sum -= first_bd_sz;
2370 }
2371
2372 /* Others are easier: run through the frag list and
2373 check all windows */
2374 for (wnd_idx = 0; wnd_idx <= num_wnds; wnd_idx++) {
2375 wnd_sum +=
2376 skb_shinfo(skb)->frags[wnd_idx + wnd_size - 1].size;
2377
2378 if (unlikely(wnd_sum < lso_mss)) {
2379 to_copy = 1;
2380 break;
2381 }
2382 wnd_sum -=
2383 skb_shinfo(skb)->frags[wnd_idx].size;
2384 }
2385 } else {
2386 /* in non-LSO too fragmented packet should always
2387 be linearized */
2388 to_copy = 1;
2389 }
2390 }
2391
2392exit_lbl:
2393 if (unlikely(to_copy))
2394 DP(NETIF_MSG_TX_QUEUED,
2395 "Linearization IS REQUIRED for %s packet. "
2396 "num_frags %d hlen %d first_bd_sz %d\n",
2397 (xmit_type & XMIT_GSO) ? "LSO" : "non-LSO",
2398 skb_shinfo(skb)->nr_frags, hlen, first_bd_sz);
2399
2400 return to_copy;
2401}
2402#endif
2403
2404static inline void bnx2x_set_pbd_gso_e2(struct sk_buff *skb, u32 *parsing_data,
2405 u32 xmit_type)
2406{
2407 *parsing_data |= (skb_shinfo(skb)->gso_size <<
2408 ETH_TX_PARSE_BD_E2_LSO_MSS_SHIFT) &
2409 ETH_TX_PARSE_BD_E2_LSO_MSS;
2410 if ((xmit_type & XMIT_GSO_V6) &&
2411 (ipv6_hdr(skb)->nexthdr == NEXTHDR_IPV6))
2412 *parsing_data |= ETH_TX_PARSE_BD_E2_IPV6_WITH_EXT_HDR;
2413}
2414
2415/**
2416 * bnx2x_set_pbd_gso - update PBD in GSO case.
2417 *
2418 * @skb: packet skb
2419 * @pbd: parse BD
2420 * @xmit_type: xmit flags
2421 */
2422static inline void bnx2x_set_pbd_gso(struct sk_buff *skb,
2423 struct eth_tx_parse_bd_e1x *pbd,
2424 u32 xmit_type)
2425{
2426 pbd->lso_mss = cpu_to_le16(skb_shinfo(skb)->gso_size);
2427 pbd->tcp_send_seq = swab32(tcp_hdr(skb)->seq);
2428 pbd->tcp_flags = pbd_tcp_flags(skb);
2429
2430 if (xmit_type & XMIT_GSO_V4) {
2431 pbd->ip_id = swab16(ip_hdr(skb)->id);
2432 pbd->tcp_pseudo_csum =
2433 swab16(~csum_tcpudp_magic(ip_hdr(skb)->saddr,
2434 ip_hdr(skb)->daddr,
2435 0, IPPROTO_TCP, 0));
2436
2437 } else
2438 pbd->tcp_pseudo_csum =
2439 swab16(~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2440 &ipv6_hdr(skb)->daddr,
2441 0, IPPROTO_TCP, 0));
2442
2443 pbd->global_data |= ETH_TX_PARSE_BD_E1X_PSEUDO_CS_WITHOUT_LEN;
2444}
2445
2446/**
2447 * bnx2x_set_pbd_csum_e2 - update PBD with checksum and return header length
2448 *
2449 * @bp: driver handle
2450 * @skb: packet skb
2451 * @parsing_data: data to be updated
2452 * @xmit_type: xmit flags
2453 *
2454 * 57712 related
2455 */
2456static inline u8 bnx2x_set_pbd_csum_e2(struct bnx2x *bp, struct sk_buff *skb,
2457 u32 *parsing_data, u32 xmit_type)
2458{
2459 *parsing_data |=
2460 ((((u8 *)skb_transport_header(skb) - skb->data) >> 1) <<
2461 ETH_TX_PARSE_BD_E2_TCP_HDR_START_OFFSET_W_SHIFT) &
2462 ETH_TX_PARSE_BD_E2_TCP_HDR_START_OFFSET_W;
2463
2464 if (xmit_type & XMIT_CSUM_TCP) {
2465 *parsing_data |= ((tcp_hdrlen(skb) / 4) <<
2466 ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW_SHIFT) &
2467 ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW;
2468
2469 return skb_transport_header(skb) + tcp_hdrlen(skb) - skb->data;
2470 } else
2471 /* We support checksum offload for TCP and UDP only.
2472 * No need to pass the UDP header length - it's a constant.
2473 */
2474 return skb_transport_header(skb) +
2475 sizeof(struct udphdr) - skb->data;
2476}
2477
2478static inline void bnx2x_set_sbd_csum(struct bnx2x *bp, struct sk_buff *skb,
2479 struct eth_tx_start_bd *tx_start_bd, u32 xmit_type)
2480{
2481 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_L4_CSUM;
2482
2483 if (xmit_type & XMIT_CSUM_V4)
2484 tx_start_bd->bd_flags.as_bitfield |=
2485 ETH_TX_BD_FLAGS_IP_CSUM;
2486 else
2487 tx_start_bd->bd_flags.as_bitfield |=
2488 ETH_TX_BD_FLAGS_IPV6;
2489
2490 if (!(xmit_type & XMIT_CSUM_TCP))
2491 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_IS_UDP;
2492}
2493
2494/**
2495 * bnx2x_set_pbd_csum - update PBD with checksum and return header length
2496 *
2497 * @bp: driver handle
2498 * @skb: packet skb
2499 * @pbd: parse BD to be updated
2500 * @xmit_type: xmit flags
2501 */
2502static inline u8 bnx2x_set_pbd_csum(struct bnx2x *bp, struct sk_buff *skb,
2503 struct eth_tx_parse_bd_e1x *pbd,
2504 u32 xmit_type)
2505{
2506 u8 hlen = (skb_network_header(skb) - skb->data) >> 1;
2507
2508 /* for now NS flag is not used in Linux */
2509 pbd->global_data =
2510 (hlen | ((skb->protocol == cpu_to_be16(ETH_P_8021Q)) <<
2511 ETH_TX_PARSE_BD_E1X_LLC_SNAP_EN_SHIFT));
2512
2513 pbd->ip_hlen_w = (skb_transport_header(skb) -
2514 skb_network_header(skb)) >> 1;
2515
2516 hlen += pbd->ip_hlen_w;
2517
2518 /* We support checksum offload for TCP and UDP only */
2519 if (xmit_type & XMIT_CSUM_TCP)
2520 hlen += tcp_hdrlen(skb) / 2;
2521 else
2522 hlen += sizeof(struct udphdr) / 2;
2523
2524 pbd->total_hlen_w = cpu_to_le16(hlen);
2525 hlen = hlen*2;
2526
2527 if (xmit_type & XMIT_CSUM_TCP) {
2528 pbd->tcp_pseudo_csum = swab16(tcp_hdr(skb)->check);
2529
2530 } else {
2531 s8 fix = SKB_CS_OFF(skb); /* signed! */
2532
2533 DP(NETIF_MSG_TX_QUEUED,
2534 "hlen %d fix %d csum before fix %x\n",
2535 le16_to_cpu(pbd->total_hlen_w), fix, SKB_CS(skb));
2536
2537 /* HW bug: fixup the CSUM */
2538 pbd->tcp_pseudo_csum =
2539 bnx2x_csum_fix(skb_transport_header(skb),
2540 SKB_CS(skb), fix);
2541
2542 DP(NETIF_MSG_TX_QUEUED, "csum after fix %x\n",
2543 pbd->tcp_pseudo_csum);
2544 }
2545
2546 return hlen;
2547}
2548
2549/* called with netif_tx_lock
2550 * bnx2x_tx_int() runs without netif_tx_lock unless it needs to call
2551 * netif_wake_queue()
2552 */
2553netdev_tx_t bnx2x_start_xmit(struct sk_buff *skb, struct net_device *dev)
2554{
2555 struct bnx2x *bp = netdev_priv(dev);
2556
2557 struct bnx2x_fastpath *fp;
2558 struct netdev_queue *txq;
2559 struct bnx2x_fp_txdata *txdata;
2560 struct sw_tx_bd *tx_buf;
2561 struct eth_tx_start_bd *tx_start_bd, *first_bd;
2562 struct eth_tx_bd *tx_data_bd, *total_pkt_bd = NULL;
2563 struct eth_tx_parse_bd_e1x *pbd_e1x = NULL;
2564 struct eth_tx_parse_bd_e2 *pbd_e2 = NULL;
2565 u32 pbd_e2_parsing_data = 0;
2566 u16 pkt_prod, bd_prod;
2567 int nbd, txq_index, fp_index, txdata_index;
2568 dma_addr_t mapping;
2569 u32 xmit_type = bnx2x_xmit_type(bp, skb);
2570 int i;
2571 u8 hlen = 0;
2572 __le16 pkt_size = 0;
2573 struct ethhdr *eth;
2574 u8 mac_type = UNICAST_ADDRESS;
2575
2576#ifdef BNX2X_STOP_ON_ERROR
2577 if (unlikely(bp->panic))
2578 return NETDEV_TX_BUSY;
2579#endif
2580
2581 txq_index = skb_get_queue_mapping(skb);
2582 txq = netdev_get_tx_queue(dev, txq_index);
2583
2584 BUG_ON(txq_index >= MAX_ETH_TXQ_IDX(bp) + FCOE_PRESENT);
2585
2586 /* decode the fastpath index and the cos index from the txq */
2587 fp_index = TXQ_TO_FP(txq_index);
2588 txdata_index = TXQ_TO_COS(txq_index);
2589
2590#ifdef BCM_CNIC
2591 /*
2592 * Override the above for the FCoE queue:
2593 * - FCoE fp entry is right after the ETH entries.
2594 * - FCoE L2 queue uses bp->txdata[0] only.
2595 */
2596 if (unlikely(!NO_FCOE(bp) && (txq_index ==
2597 bnx2x_fcoe_tx(bp, txq_index)))) {
2598 fp_index = FCOE_IDX;
2599 txdata_index = 0;
2600 }
2601#endif
2602
2603 /* enable this debug print to view the transmission queue being used
2604 DP(BNX2X_MSG_FP, "indices: txq %d, fp %d, txdata %d",
2605 txq_index, fp_index, txdata_index); */
2606
2607 /* locate the fastpath and the txdata */
2608 fp = &bp->fp[fp_index];
2609 txdata = &fp->txdata[txdata_index];
2610
2611 /* enable this debug print to view the tranmission details
2612 DP(BNX2X_MSG_FP,"transmitting packet cid %d fp index %d txdata_index %d"
2613 " tx_data ptr %p fp pointer %p",
2614 txdata->cid, fp_index, txdata_index, txdata, fp); */
2615
2616 if (unlikely(bnx2x_tx_avail(bp, txdata) <
2617 (skb_shinfo(skb)->nr_frags + 3))) {
2618 fp->eth_q_stats.driver_xoff++;
2619 netif_tx_stop_queue(txq);
2620 BNX2X_ERR("BUG! Tx ring full when queue awake!\n");
2621 return NETDEV_TX_BUSY;
2622 }
2623
2624 DP(NETIF_MSG_TX_QUEUED, "queue[%d]: SKB: summed %x protocol %x "
2625 "protocol(%x,%x) gso type %x xmit_type %x\n",
2626 txq_index, skb->ip_summed, skb->protocol, ipv6_hdr(skb)->nexthdr,
2627 ip_hdr(skb)->protocol, skb_shinfo(skb)->gso_type, xmit_type);
2628
2629 eth = (struct ethhdr *)skb->data;
2630
2631 /* set flag according to packet type (UNICAST_ADDRESS is default)*/
2632 if (unlikely(is_multicast_ether_addr(eth->h_dest))) {
2633 if (is_broadcast_ether_addr(eth->h_dest))
2634 mac_type = BROADCAST_ADDRESS;
2635 else
2636 mac_type = MULTICAST_ADDRESS;
2637 }
2638
2639#if (MAX_SKB_FRAGS >= MAX_FETCH_BD - 3)
2640 /* First, check if we need to linearize the skb (due to FW
2641 restrictions). No need to check fragmentation if page size > 8K
2642 (there will be no violation to FW restrictions) */
2643 if (bnx2x_pkt_req_lin(bp, skb, xmit_type)) {
2644 /* Statistics of linearization */
2645 bp->lin_cnt++;
2646 if (skb_linearize(skb) != 0) {
2647 DP(NETIF_MSG_TX_QUEUED, "SKB linearization failed - "
2648 "silently dropping this SKB\n");
2649 dev_kfree_skb_any(skb);
2650 return NETDEV_TX_OK;
2651 }
2652 }
2653#endif
2654 /* Map skb linear data for DMA */
2655 mapping = dma_map_single(&bp->pdev->dev, skb->data,
2656 skb_headlen(skb), DMA_TO_DEVICE);
2657 if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
2658 DP(NETIF_MSG_TX_QUEUED, "SKB mapping failed - "
2659 "silently dropping this SKB\n");
2660 dev_kfree_skb_any(skb);
2661 return NETDEV_TX_OK;
2662 }
2663 /*
2664 Please read carefully. First we use one BD which we mark as start,
2665 then we have a parsing info BD (used for TSO or xsum),
2666 and only then we have the rest of the TSO BDs.
2667 (don't forget to mark the last one as last,
2668 and to unmap only AFTER you write to the BD ...)
2669 And above all, all pdb sizes are in words - NOT DWORDS!
2670 */
2671
2672 /* get current pkt produced now - advance it just before sending packet
2673 * since mapping of pages may fail and cause packet to be dropped
2674 */
2675 pkt_prod = txdata->tx_pkt_prod;
2676 bd_prod = TX_BD(txdata->tx_bd_prod);
2677
2678 /* get a tx_buf and first BD
2679 * tx_start_bd may be changed during SPLIT,
2680 * but first_bd will always stay first
2681 */
2682 tx_buf = &txdata->tx_buf_ring[TX_BD(pkt_prod)];
2683 tx_start_bd = &txdata->tx_desc_ring[bd_prod].start_bd;
2684 first_bd = tx_start_bd;
2685
2686 tx_start_bd->bd_flags.as_bitfield = ETH_TX_BD_FLAGS_START_BD;
2687 SET_FLAG(tx_start_bd->general_data, ETH_TX_START_BD_ETH_ADDR_TYPE,
2688 mac_type);
2689
2690 /* header nbd */
2691 SET_FLAG(tx_start_bd->general_data, ETH_TX_START_BD_HDR_NBDS, 1);
2692
2693 /* remember the first BD of the packet */
2694 tx_buf->first_bd = txdata->tx_bd_prod;
2695 tx_buf->skb = skb;
2696 tx_buf->flags = 0;
2697
2698 DP(NETIF_MSG_TX_QUEUED,
2699 "sending pkt %u @%p next_idx %u bd %u @%p\n",
2700 pkt_prod, tx_buf, txdata->tx_pkt_prod, bd_prod, tx_start_bd);
2701
2702 if (vlan_tx_tag_present(skb)) {
2703 tx_start_bd->vlan_or_ethertype =
2704 cpu_to_le16(vlan_tx_tag_get(skb));
2705 tx_start_bd->bd_flags.as_bitfield |=
2706 (X_ETH_OUTBAND_VLAN << ETH_TX_BD_FLAGS_VLAN_MODE_SHIFT);
2707 } else
2708 tx_start_bd->vlan_or_ethertype = cpu_to_le16(pkt_prod);
2709
2710 /* turn on parsing and get a BD */
2711 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
2712
2713 if (xmit_type & XMIT_CSUM)
2714 bnx2x_set_sbd_csum(bp, skb, tx_start_bd, xmit_type);
2715
2716 if (!CHIP_IS_E1x(bp)) {
2717 pbd_e2 = &txdata->tx_desc_ring[bd_prod].parse_bd_e2;
2718 memset(pbd_e2, 0, sizeof(struct eth_tx_parse_bd_e2));
2719 /* Set PBD in checksum offload case */
2720 if (xmit_type & XMIT_CSUM)
2721 hlen = bnx2x_set_pbd_csum_e2(bp, skb,
2722 &pbd_e2_parsing_data,
2723 xmit_type);
2724 if (IS_MF_SI(bp)) {
2725 /*
2726 * fill in the MAC addresses in the PBD - for local
2727 * switching
2728 */
2729 bnx2x_set_fw_mac_addr(&pbd_e2->src_mac_addr_hi,
2730 &pbd_e2->src_mac_addr_mid,
2731 &pbd_e2->src_mac_addr_lo,
2732 eth->h_source);
2733 bnx2x_set_fw_mac_addr(&pbd_e2->dst_mac_addr_hi,
2734 &pbd_e2->dst_mac_addr_mid,
2735 &pbd_e2->dst_mac_addr_lo,
2736 eth->h_dest);
2737 }
2738 } else {
2739 pbd_e1x = &txdata->tx_desc_ring[bd_prod].parse_bd_e1x;
2740 memset(pbd_e1x, 0, sizeof(struct eth_tx_parse_bd_e1x));
2741 /* Set PBD in checksum offload case */
2742 if (xmit_type & XMIT_CSUM)
2743 hlen = bnx2x_set_pbd_csum(bp, skb, pbd_e1x, xmit_type);
2744
2745 }
2746
2747 /* Setup the data pointer of the first BD of the packet */
2748 tx_start_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
2749 tx_start_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
2750 nbd = 2; /* start_bd + pbd + frags (updated when pages are mapped) */
2751 tx_start_bd->nbytes = cpu_to_le16(skb_headlen(skb));
2752 pkt_size = tx_start_bd->nbytes;
2753
2754 DP(NETIF_MSG_TX_QUEUED, "first bd @%p addr (%x:%x) nbd %d"
2755 " nbytes %d flags %x vlan %x\n",
2756 tx_start_bd, tx_start_bd->addr_hi, tx_start_bd->addr_lo,
2757 le16_to_cpu(tx_start_bd->nbd), le16_to_cpu(tx_start_bd->nbytes),
2758 tx_start_bd->bd_flags.as_bitfield,
2759 le16_to_cpu(tx_start_bd->vlan_or_ethertype));
2760
2761 if (xmit_type & XMIT_GSO) {
2762
2763 DP(NETIF_MSG_TX_QUEUED,
2764 "TSO packet len %d hlen %d total len %d tso size %d\n",
2765 skb->len, hlen, skb_headlen(skb),
2766 skb_shinfo(skb)->gso_size);
2767
2768 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_SW_LSO;
2769
2770 if (unlikely(skb_headlen(skb) > hlen))
2771 bd_prod = bnx2x_tx_split(bp, txdata, tx_buf,
2772 &tx_start_bd, hlen,
2773 bd_prod, ++nbd);
2774 if (!CHIP_IS_E1x(bp))
2775 bnx2x_set_pbd_gso_e2(skb, &pbd_e2_parsing_data,
2776 xmit_type);
2777 else
2778 bnx2x_set_pbd_gso(skb, pbd_e1x, xmit_type);
2779 }
2780
2781 /* Set the PBD's parsing_data field if not zero
2782 * (for the chips newer than 57711).
2783 */
2784 if (pbd_e2_parsing_data)
2785 pbd_e2->parsing_data = cpu_to_le32(pbd_e2_parsing_data);
2786
2787 tx_data_bd = (struct eth_tx_bd *)tx_start_bd;
2788
2789 /* Handle fragmented skb */
2790 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2791 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2792
2793 mapping = dma_map_page(&bp->pdev->dev, frag->page,
2794 frag->page_offset, frag->size,
2795 DMA_TO_DEVICE);
2796 if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
2797
2798 DP(NETIF_MSG_TX_QUEUED, "Unable to map page - "
2799 "dropping packet...\n");
2800
2801 /* we need unmap all buffers already mapped
2802 * for this SKB;
2803 * first_bd->nbd need to be properly updated
2804 * before call to bnx2x_free_tx_pkt
2805 */
2806 first_bd->nbd = cpu_to_le16(nbd);
2807 bnx2x_free_tx_pkt(bp, txdata,
2808 TX_BD(txdata->tx_pkt_prod));
2809 return NETDEV_TX_OK;
2810 }
2811
2812 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
2813 tx_data_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
2814 if (total_pkt_bd == NULL)
2815 total_pkt_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
2816
2817 tx_data_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
2818 tx_data_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
2819 tx_data_bd->nbytes = cpu_to_le16(frag->size);
2820 le16_add_cpu(&pkt_size, frag->size);
2821 nbd++;
2822
2823 DP(NETIF_MSG_TX_QUEUED,
2824 "frag %d bd @%p addr (%x:%x) nbytes %d\n",
2825 i, tx_data_bd, tx_data_bd->addr_hi, tx_data_bd->addr_lo,
2826 le16_to_cpu(tx_data_bd->nbytes));
2827 }
2828
2829 DP(NETIF_MSG_TX_QUEUED, "last bd @%p\n", tx_data_bd);
2830
2831 /* update with actual num BDs */
2832 first_bd->nbd = cpu_to_le16(nbd);
2833
2834 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
2835
2836 /* now send a tx doorbell, counting the next BD
2837 * if the packet contains or ends with it
2838 */
2839 if (TX_BD_POFF(bd_prod) < nbd)
2840 nbd++;
2841
2842 /* total_pkt_bytes should be set on the first data BD if
2843 * it's not an LSO packet and there is more than one
2844 * data BD. In this case pkt_size is limited by an MTU value.
2845 * However we prefer to set it for an LSO packet (while we don't
2846 * have to) in order to save some CPU cycles in a none-LSO
2847 * case, when we much more care about them.
2848 */
2849 if (total_pkt_bd != NULL)
2850 total_pkt_bd->total_pkt_bytes = pkt_size;
2851
2852 if (pbd_e1x)
2853 DP(NETIF_MSG_TX_QUEUED,
2854 "PBD (E1X) @%p ip_data %x ip_hlen %u ip_id %u lso_mss %u"
2855 " tcp_flags %x xsum %x seq %u hlen %u\n",
2856 pbd_e1x, pbd_e1x->global_data, pbd_e1x->ip_hlen_w,
2857 pbd_e1x->ip_id, pbd_e1x->lso_mss, pbd_e1x->tcp_flags,
2858 pbd_e1x->tcp_pseudo_csum, pbd_e1x->tcp_send_seq,
2859 le16_to_cpu(pbd_e1x->total_hlen_w));
2860 if (pbd_e2)
2861 DP(NETIF_MSG_TX_QUEUED,
2862 "PBD (E2) @%p dst %x %x %x src %x %x %x parsing_data %x\n",
2863 pbd_e2, pbd_e2->dst_mac_addr_hi, pbd_e2->dst_mac_addr_mid,
2864 pbd_e2->dst_mac_addr_lo, pbd_e2->src_mac_addr_hi,
2865 pbd_e2->src_mac_addr_mid, pbd_e2->src_mac_addr_lo,
2866 pbd_e2->parsing_data);
2867 DP(NETIF_MSG_TX_QUEUED, "doorbell: nbd %d bd %u\n", nbd, bd_prod);
2868
2869 txdata->tx_pkt_prod++;
2870 /*
2871 * Make sure that the BD data is updated before updating the producer
2872 * since FW might read the BD right after the producer is updated.
2873 * This is only applicable for weak-ordered memory model archs such
2874 * as IA-64. The following barrier is also mandatory since FW will
2875 * assumes packets must have BDs.
2876 */
2877 wmb();
2878
2879 txdata->tx_db.data.prod += nbd;
2880 barrier();
2881
2882 DOORBELL(bp, txdata->cid, txdata->tx_db.raw);
2883
2884 mmiowb();
2885
2886 txdata->tx_bd_prod += nbd;
2887
2888 if (unlikely(bnx2x_tx_avail(bp, txdata) < MAX_SKB_FRAGS + 3)) {
2889 netif_tx_stop_queue(txq);
2890
2891 /* paired memory barrier is in bnx2x_tx_int(), we have to keep
2892 * ordering of set_bit() in netif_tx_stop_queue() and read of
2893 * fp->bd_tx_cons */
2894 smp_mb();
2895
2896 fp->eth_q_stats.driver_xoff++;
2897 if (bnx2x_tx_avail(bp, txdata) >= MAX_SKB_FRAGS + 3)
2898 netif_tx_wake_queue(txq);
2899 }
2900 txdata->tx_pkt++;
2901
2902 return NETDEV_TX_OK;
2903}
2904
2905/**
2906 * bnx2x_setup_tc - routine to configure net_device for multi tc
2907 *
2908 * @netdev: net device to configure
2909 * @tc: number of traffic classes to enable
2910 *
2911 * callback connected to the ndo_setup_tc function pointer
2912 */
2913int bnx2x_setup_tc(struct net_device *dev, u8 num_tc)
2914{
2915 int cos, prio, count, offset;
2916 struct bnx2x *bp = netdev_priv(dev);
2917
2918 /* setup tc must be called under rtnl lock */
2919 ASSERT_RTNL();
2920
2921 /* no traffic classes requested. aborting */
2922 if (!num_tc) {
2923 netdev_reset_tc(dev);
2924 return 0;
2925 }
2926
2927 /* requested to support too many traffic classes */
2928 if (num_tc > bp->max_cos) {
2929 DP(NETIF_MSG_TX_ERR, "support for too many traffic classes"
2930 " requested: %d. max supported is %d",
2931 num_tc, bp->max_cos);
2932 return -EINVAL;
2933 }
2934
2935 /* declare amount of supported traffic classes */
2936 if (netdev_set_num_tc(dev, num_tc)) {
2937 DP(NETIF_MSG_TX_ERR, "failed to declare %d traffic classes",
2938 num_tc);
2939 return -EINVAL;
2940 }
2941
2942 /* configure priority to traffic class mapping */
2943 for (prio = 0; prio < BNX2X_MAX_PRIORITY; prio++) {
2944 netdev_set_prio_tc_map(dev, prio, bp->prio_to_cos[prio]);
2945 DP(BNX2X_MSG_SP, "mapping priority %d to tc %d",
2946 prio, bp->prio_to_cos[prio]);
2947 }
2948
2949
2950 /* Use this configuration to diffrentiate tc0 from other COSes
2951 This can be used for ets or pfc, and save the effort of setting
2952 up a multio class queue disc or negotiating DCBX with a switch
2953 netdev_set_prio_tc_map(dev, 0, 0);
2954 DP(BNX2X_MSG_SP, "mapping priority %d to tc %d", 0, 0);
2955 for (prio = 1; prio < 16; prio++) {
2956 netdev_set_prio_tc_map(dev, prio, 1);
2957 DP(BNX2X_MSG_SP, "mapping priority %d to tc %d", prio, 1);
2958 } */
2959
2960 /* configure traffic class to transmission queue mapping */
2961 for (cos = 0; cos < bp->max_cos; cos++) {
2962 count = BNX2X_NUM_ETH_QUEUES(bp);
2963 offset = cos * MAX_TXQS_PER_COS;
2964 netdev_set_tc_queue(dev, cos, count, offset);
2965 DP(BNX2X_MSG_SP, "mapping tc %d to offset %d count %d",
2966 cos, offset, count);
2967 }
2968
2969 return 0;
2970}
2971
2972/* called with rtnl_lock */
2973int bnx2x_change_mac_addr(struct net_device *dev, void *p)
2974{
2975 struct sockaddr *addr = p;
2976 struct bnx2x *bp = netdev_priv(dev);
2977 int rc = 0;
2978
2979 if (!is_valid_ether_addr((u8 *)(addr->sa_data)))
2980 return -EINVAL;
2981
2982 if (netif_running(dev)) {
2983 rc = bnx2x_set_eth_mac(bp, false);
2984 if (rc)
2985 return rc;
2986 }
2987
2988 memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
2989
2990 if (netif_running(dev))
2991 rc = bnx2x_set_eth_mac(bp, true);
2992
2993 return rc;
2994}
2995
2996static void bnx2x_free_fp_mem_at(struct bnx2x *bp, int fp_index)
2997{
2998 union host_hc_status_block *sb = &bnx2x_fp(bp, fp_index, status_blk);
2999 struct bnx2x_fastpath *fp = &bp->fp[fp_index];
3000 u8 cos;
3001
3002 /* Common */
3003#ifdef BCM_CNIC
3004 if (IS_FCOE_IDX(fp_index)) {
3005 memset(sb, 0, sizeof(union host_hc_status_block));
3006 fp->status_blk_mapping = 0;
3007
3008 } else {
3009#endif
3010 /* status blocks */
3011 if (!CHIP_IS_E1x(bp))
3012 BNX2X_PCI_FREE(sb->e2_sb,
3013 bnx2x_fp(bp, fp_index,
3014 status_blk_mapping),
3015 sizeof(struct host_hc_status_block_e2));
3016 else
3017 BNX2X_PCI_FREE(sb->e1x_sb,
3018 bnx2x_fp(bp, fp_index,
3019 status_blk_mapping),
3020 sizeof(struct host_hc_status_block_e1x));
3021#ifdef BCM_CNIC
3022 }
3023#endif
3024 /* Rx */
3025 if (!skip_rx_queue(bp, fp_index)) {
3026 bnx2x_free_rx_bds(fp);
3027
3028 /* fastpath rx rings: rx_buf rx_desc rx_comp */
3029 BNX2X_FREE(bnx2x_fp(bp, fp_index, rx_buf_ring));
3030 BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_desc_ring),
3031 bnx2x_fp(bp, fp_index, rx_desc_mapping),
3032 sizeof(struct eth_rx_bd) * NUM_RX_BD);
3033
3034 BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_comp_ring),
3035 bnx2x_fp(bp, fp_index, rx_comp_mapping),
3036 sizeof(struct eth_fast_path_rx_cqe) *
3037 NUM_RCQ_BD);
3038
3039 /* SGE ring */
3040 BNX2X_FREE(bnx2x_fp(bp, fp_index, rx_page_ring));
3041 BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_sge_ring),
3042 bnx2x_fp(bp, fp_index, rx_sge_mapping),
3043 BCM_PAGE_SIZE * NUM_RX_SGE_PAGES);
3044 }
3045
3046 /* Tx */
3047 if (!skip_tx_queue(bp, fp_index)) {
3048 /* fastpath tx rings: tx_buf tx_desc */
3049 for_each_cos_in_tx_queue(fp, cos) {
3050 struct bnx2x_fp_txdata *txdata = &fp->txdata[cos];
3051
3052 DP(BNX2X_MSG_SP,
3053 "freeing tx memory of fp %d cos %d cid %d",
3054 fp_index, cos, txdata->cid);
3055
3056 BNX2X_FREE(txdata->tx_buf_ring);
3057 BNX2X_PCI_FREE(txdata->tx_desc_ring,
3058 txdata->tx_desc_mapping,
3059 sizeof(union eth_tx_bd_types) * NUM_TX_BD);
3060 }
3061 }
3062 /* end of fastpath */
3063}
3064
3065void bnx2x_free_fp_mem(struct bnx2x *bp)
3066{
3067 int i;
3068 for_each_queue(bp, i)
3069 bnx2x_free_fp_mem_at(bp, i);
3070}
3071
3072static inline void set_sb_shortcuts(struct bnx2x *bp, int index)
3073{
3074 union host_hc_status_block status_blk = bnx2x_fp(bp, index, status_blk);
3075 if (!CHIP_IS_E1x(bp)) {
3076 bnx2x_fp(bp, index, sb_index_values) =
3077 (__le16 *)status_blk.e2_sb->sb.index_values;
3078 bnx2x_fp(bp, index, sb_running_index) =
3079 (__le16 *)status_blk.e2_sb->sb.running_index;
3080 } else {
3081 bnx2x_fp(bp, index, sb_index_values) =
3082 (__le16 *)status_blk.e1x_sb->sb.index_values;
3083 bnx2x_fp(bp, index, sb_running_index) =
3084 (__le16 *)status_blk.e1x_sb->sb.running_index;
3085 }
3086}
3087
3088static int bnx2x_alloc_fp_mem_at(struct bnx2x *bp, int index)
3089{
3090 union host_hc_status_block *sb;
3091 struct bnx2x_fastpath *fp = &bp->fp[index];
3092 int ring_size = 0;
3093 u8 cos;
3094 int rx_ring_size = 0;
3095
3096 /* if rx_ring_size specified - use it */
3097 if (!bp->rx_ring_size) {
3098
3099 rx_ring_size = MAX_RX_AVAIL/BNX2X_NUM_RX_QUEUES(bp);
3100
3101 /* allocate at least number of buffers required by FW */
3102 rx_ring_size = max_t(int, bp->disable_tpa ? MIN_RX_SIZE_NONTPA :
3103 MIN_RX_SIZE_TPA, rx_ring_size);
3104
3105 bp->rx_ring_size = rx_ring_size;
3106 } else
3107 rx_ring_size = bp->rx_ring_size;
3108
3109 /* Common */
3110 sb = &bnx2x_fp(bp, index, status_blk);
3111#ifdef BCM_CNIC
3112 if (!IS_FCOE_IDX(index)) {
3113#endif
3114 /* status blocks */
3115 if (!CHIP_IS_E1x(bp))
3116 BNX2X_PCI_ALLOC(sb->e2_sb,
3117 &bnx2x_fp(bp, index, status_blk_mapping),
3118 sizeof(struct host_hc_status_block_e2));
3119 else
3120 BNX2X_PCI_ALLOC(sb->e1x_sb,
3121 &bnx2x_fp(bp, index, status_blk_mapping),
3122 sizeof(struct host_hc_status_block_e1x));
3123#ifdef BCM_CNIC
3124 }
3125#endif
3126
3127 /* FCoE Queue uses Default SB and doesn't ACK the SB, thus no need to
3128 * set shortcuts for it.
3129 */
3130 if (!IS_FCOE_IDX(index))
3131 set_sb_shortcuts(bp, index);
3132
3133 /* Tx */
3134 if (!skip_tx_queue(bp, index)) {
3135 /* fastpath tx rings: tx_buf tx_desc */
3136 for_each_cos_in_tx_queue(fp, cos) {
3137 struct bnx2x_fp_txdata *txdata = &fp->txdata[cos];
3138
3139 DP(BNX2X_MSG_SP, "allocating tx memory of "
3140 "fp %d cos %d",
3141 index, cos);
3142
3143 BNX2X_ALLOC(txdata->tx_buf_ring,
3144 sizeof(struct sw_tx_bd) * NUM_TX_BD);
3145 BNX2X_PCI_ALLOC(txdata->tx_desc_ring,
3146 &txdata->tx_desc_mapping,
3147 sizeof(union eth_tx_bd_types) * NUM_TX_BD);
3148 }
3149 }
3150
3151 /* Rx */
3152 if (!skip_rx_queue(bp, index)) {
3153 /* fastpath rx rings: rx_buf rx_desc rx_comp */
3154 BNX2X_ALLOC(bnx2x_fp(bp, index, rx_buf_ring),
3155 sizeof(struct sw_rx_bd) * NUM_RX_BD);
3156 BNX2X_PCI_ALLOC(bnx2x_fp(bp, index, rx_desc_ring),
3157 &bnx2x_fp(bp, index, rx_desc_mapping),
3158 sizeof(struct eth_rx_bd) * NUM_RX_BD);
3159
3160 BNX2X_PCI_ALLOC(bnx2x_fp(bp, index, rx_comp_ring),
3161 &bnx2x_fp(bp, index, rx_comp_mapping),
3162 sizeof(struct eth_fast_path_rx_cqe) *
3163 NUM_RCQ_BD);
3164
3165 /* SGE ring */
3166 BNX2X_ALLOC(bnx2x_fp(bp, index, rx_page_ring),
3167 sizeof(struct sw_rx_page) * NUM_RX_SGE);
3168 BNX2X_PCI_ALLOC(bnx2x_fp(bp, index, rx_sge_ring),
3169 &bnx2x_fp(bp, index, rx_sge_mapping),
3170 BCM_PAGE_SIZE * NUM_RX_SGE_PAGES);
3171 /* RX BD ring */
3172 bnx2x_set_next_page_rx_bd(fp);
3173
3174 /* CQ ring */
3175 bnx2x_set_next_page_rx_cq(fp);
3176
3177 /* BDs */
3178 ring_size = bnx2x_alloc_rx_bds(fp, rx_ring_size);
3179 if (ring_size < rx_ring_size)
3180 goto alloc_mem_err;
3181 }
3182
3183 return 0;
3184
3185/* handles low memory cases */
3186alloc_mem_err:
3187 BNX2X_ERR("Unable to allocate full memory for queue %d (size %d)\n",
3188 index, ring_size);
3189 /* FW will drop all packets if queue is not big enough,
3190 * In these cases we disable the queue
3191 * Min size is different for OOO, TPA and non-TPA queues
3192 */
3193 if (ring_size < (fp->disable_tpa ?
3194 MIN_RX_SIZE_NONTPA : MIN_RX_SIZE_TPA)) {
3195 /* release memory allocated for this queue */
3196 bnx2x_free_fp_mem_at(bp, index);
3197 return -ENOMEM;
3198 }
3199 return 0;
3200}
3201
3202int bnx2x_alloc_fp_mem(struct bnx2x *bp)
3203{
3204 int i;
3205
3206 /**
3207 * 1. Allocate FP for leading - fatal if error
3208 * 2. {CNIC} Allocate FCoE FP - fatal if error
3209 * 3. {CNIC} Allocate OOO + FWD - disable OOO if error
3210 * 4. Allocate RSS - fix number of queues if error
3211 */
3212
3213 /* leading */
3214 if (bnx2x_alloc_fp_mem_at(bp, 0))
3215 return -ENOMEM;
3216
3217#ifdef BCM_CNIC
3218 if (!NO_FCOE(bp))
3219 /* FCoE */
3220 if (bnx2x_alloc_fp_mem_at(bp, FCOE_IDX))
3221 /* we will fail load process instead of mark
3222 * NO_FCOE_FLAG
3223 */
3224 return -ENOMEM;
3225#endif
3226
3227 /* RSS */
3228 for_each_nondefault_eth_queue(bp, i)
3229 if (bnx2x_alloc_fp_mem_at(bp, i))
3230 break;
3231
3232 /* handle memory failures */
3233 if (i != BNX2X_NUM_ETH_QUEUES(bp)) {
3234 int delta = BNX2X_NUM_ETH_QUEUES(bp) - i;
3235
3236 WARN_ON(delta < 0);
3237#ifdef BCM_CNIC
3238 /**
3239 * move non eth FPs next to last eth FP
3240 * must be done in that order
3241 * FCOE_IDX < FWD_IDX < OOO_IDX
3242 */
3243
3244 /* move FCoE fp even NO_FCOE_FLAG is on */
3245 bnx2x_move_fp(bp, FCOE_IDX, FCOE_IDX - delta);
3246#endif
3247 bp->num_queues -= delta;
3248 BNX2X_ERR("Adjusted num of queues from %d to %d\n",
3249 bp->num_queues + delta, bp->num_queues);
3250 }
3251
3252 return 0;
3253}
3254
3255void bnx2x_free_mem_bp(struct bnx2x *bp)
3256{
3257 kfree(bp->fp);
3258 kfree(bp->msix_table);
3259 kfree(bp->ilt);
3260}
3261
3262int __devinit bnx2x_alloc_mem_bp(struct bnx2x *bp)
3263{
3264 struct bnx2x_fastpath *fp;
3265 struct msix_entry *tbl;
3266 struct bnx2x_ilt *ilt;
3267 int msix_table_size = 0;
3268
3269 /*
3270 * The biggest MSI-X table we might need is as a maximum number of fast
3271 * path IGU SBs plus default SB (for PF).
3272 */
3273 msix_table_size = bp->igu_sb_cnt + 1;
3274
3275 /* fp array: RSS plus CNIC related L2 queues */
3276 fp = kzalloc((BNX2X_MAX_RSS_COUNT(bp) + NON_ETH_CONTEXT_USE) *
3277 sizeof(*fp), GFP_KERNEL);
3278 if (!fp)
3279 goto alloc_err;
3280 bp->fp = fp;
3281
3282 /* msix table */
3283 tbl = kzalloc(msix_table_size * sizeof(*tbl), GFP_KERNEL);
3284 if (!tbl)
3285 goto alloc_err;
3286 bp->msix_table = tbl;
3287
3288 /* ilt */
3289 ilt = kzalloc(sizeof(*ilt), GFP_KERNEL);
3290 if (!ilt)
3291 goto alloc_err;
3292 bp->ilt = ilt;
3293
3294 return 0;
3295alloc_err:
3296 bnx2x_free_mem_bp(bp);
3297 return -ENOMEM;
3298
3299}
3300
3301int bnx2x_reload_if_running(struct net_device *dev)
3302{
3303 struct bnx2x *bp = netdev_priv(dev);
3304
3305 if (unlikely(!netif_running(dev)))
3306 return 0;
3307
3308 bnx2x_nic_unload(bp, UNLOAD_NORMAL);
3309 return bnx2x_nic_load(bp, LOAD_NORMAL);
3310}
3311
3312int bnx2x_get_cur_phy_idx(struct bnx2x *bp)
3313{
3314 u32 sel_phy_idx = 0;
3315 if (bp->link_params.num_phys <= 1)
3316 return INT_PHY;
3317
3318 if (bp->link_vars.link_up) {
3319 sel_phy_idx = EXT_PHY1;
3320 /* In case link is SERDES, check if the EXT_PHY2 is the one */
3321 if ((bp->link_vars.link_status & LINK_STATUS_SERDES_LINK) &&
3322 (bp->link_params.phy[EXT_PHY2].supported & SUPPORTED_FIBRE))
3323 sel_phy_idx = EXT_PHY2;
3324 } else {
3325
3326 switch (bnx2x_phy_selection(&bp->link_params)) {
3327 case PORT_HW_CFG_PHY_SELECTION_HARDWARE_DEFAULT:
3328 case PORT_HW_CFG_PHY_SELECTION_FIRST_PHY:
3329 case PORT_HW_CFG_PHY_SELECTION_FIRST_PHY_PRIORITY:
3330 sel_phy_idx = EXT_PHY1;
3331 break;
3332 case PORT_HW_CFG_PHY_SELECTION_SECOND_PHY:
3333 case PORT_HW_CFG_PHY_SELECTION_SECOND_PHY_PRIORITY:
3334 sel_phy_idx = EXT_PHY2;
3335 break;
3336 }
3337 }
3338
3339 return sel_phy_idx;
3340
3341}
3342int bnx2x_get_link_cfg_idx(struct bnx2x *bp)
3343{
3344 u32 sel_phy_idx = bnx2x_get_cur_phy_idx(bp);
3345 /*
3346 * The selected actived PHY is always after swapping (in case PHY
3347 * swapping is enabled). So when swapping is enabled, we need to reverse
3348 * the configuration
3349 */
3350
3351 if (bp->link_params.multi_phy_config &
3352 PORT_HW_CFG_PHY_SWAPPED_ENABLED) {
3353 if (sel_phy_idx == EXT_PHY1)
3354 sel_phy_idx = EXT_PHY2;
3355 else if (sel_phy_idx == EXT_PHY2)
3356 sel_phy_idx = EXT_PHY1;
3357 }
3358 return LINK_CONFIG_IDX(sel_phy_idx);
3359}
3360
3361#if defined(NETDEV_FCOE_WWNN) && defined(BCM_CNIC)
3362int bnx2x_fcoe_get_wwn(struct net_device *dev, u64 *wwn, int type)
3363{
3364 struct bnx2x *bp = netdev_priv(dev);
3365 struct cnic_eth_dev *cp = &bp->cnic_eth_dev;
3366
3367 switch (type) {
3368 case NETDEV_FCOE_WWNN:
3369 *wwn = HILO_U64(cp->fcoe_wwn_node_name_hi,
3370 cp->fcoe_wwn_node_name_lo);
3371 break;
3372 case NETDEV_FCOE_WWPN:
3373 *wwn = HILO_U64(cp->fcoe_wwn_port_name_hi,
3374 cp->fcoe_wwn_port_name_lo);
3375 break;
3376 default:
3377 return -EINVAL;
3378 }
3379
3380 return 0;
3381}
3382#endif
3383
3384/* called with rtnl_lock */
3385int bnx2x_change_mtu(struct net_device *dev, int new_mtu)
3386{
3387 struct bnx2x *bp = netdev_priv(dev);
3388
3389 if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
3390 printk(KERN_ERR "Handling parity error recovery. Try again later\n");
3391 return -EAGAIN;
3392 }
3393
3394 if ((new_mtu > ETH_MAX_JUMBO_PACKET_SIZE) ||
3395 ((new_mtu + ETH_HLEN) < ETH_MIN_PACKET_SIZE))
3396 return -EINVAL;
3397
3398 /* This does not race with packet allocation
3399 * because the actual alloc size is
3400 * only updated as part of load
3401 */
3402 dev->mtu = new_mtu;
3403
3404 return bnx2x_reload_if_running(dev);
3405}
3406
3407u32 bnx2x_fix_features(struct net_device *dev, u32 features)
3408{
3409 struct bnx2x *bp = netdev_priv(dev);
3410
3411 /* TPA requires Rx CSUM offloading */
3412 if (!(features & NETIF_F_RXCSUM) || bp->disable_tpa)
3413 features &= ~NETIF_F_LRO;
3414
3415 return features;
3416}
3417
3418int bnx2x_set_features(struct net_device *dev, u32 features)
3419{
3420 struct bnx2x *bp = netdev_priv(dev);
3421 u32 flags = bp->flags;
3422 bool bnx2x_reload = false;
3423
3424 if (features & NETIF_F_LRO)
3425 flags |= TPA_ENABLE_FLAG;
3426 else
3427 flags &= ~TPA_ENABLE_FLAG;
3428
3429 if (features & NETIF_F_LOOPBACK) {
3430 if (bp->link_params.loopback_mode != LOOPBACK_BMAC) {
3431 bp->link_params.loopback_mode = LOOPBACK_BMAC;
3432 bnx2x_reload = true;
3433 }
3434 } else {
3435 if (bp->link_params.loopback_mode != LOOPBACK_NONE) {
3436 bp->link_params.loopback_mode = LOOPBACK_NONE;
3437 bnx2x_reload = true;
3438 }
3439 }
3440
3441 if (flags ^ bp->flags) {
3442 bp->flags = flags;
3443 bnx2x_reload = true;
3444 }
3445
3446 if (bnx2x_reload) {
3447 if (bp->recovery_state == BNX2X_RECOVERY_DONE)
3448 return bnx2x_reload_if_running(dev);
3449 /* else: bnx2x_nic_load() will be called at end of recovery */
3450 }
3451
3452 return 0;
3453}
3454
3455void bnx2x_tx_timeout(struct net_device *dev)
3456{
3457 struct bnx2x *bp = netdev_priv(dev);
3458
3459#ifdef BNX2X_STOP_ON_ERROR
3460 if (!bp->panic)
3461 bnx2x_panic();
3462#endif
3463
3464 smp_mb__before_clear_bit();
3465 set_bit(BNX2X_SP_RTNL_TX_TIMEOUT, &bp->sp_rtnl_state);
3466 smp_mb__after_clear_bit();
3467
3468 /* This allows the netif to be shutdown gracefully before resetting */
3469 schedule_delayed_work(&bp->sp_rtnl_task, 0);
3470}
3471
3472int bnx2x_suspend(struct pci_dev *pdev, pm_message_t state)
3473{
3474 struct net_device *dev = pci_get_drvdata(pdev);
3475 struct bnx2x *bp;
3476
3477 if (!dev) {
3478 dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
3479 return -ENODEV;
3480 }
3481 bp = netdev_priv(dev);
3482
3483 rtnl_lock();
3484
3485 pci_save_state(pdev);
3486
3487 if (!netif_running(dev)) {
3488 rtnl_unlock();
3489 return 0;
3490 }
3491
3492 netif_device_detach(dev);
3493
3494 bnx2x_nic_unload(bp, UNLOAD_CLOSE);
3495
3496 bnx2x_set_power_state(bp, pci_choose_state(pdev, state));
3497
3498 rtnl_unlock();
3499
3500 return 0;
3501}
3502
3503int bnx2x_resume(struct pci_dev *pdev)
3504{
3505 struct net_device *dev = pci_get_drvdata(pdev);
3506 struct bnx2x *bp;
3507 int rc;
3508
3509 if (!dev) {
3510 dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
3511 return -ENODEV;
3512 }
3513 bp = netdev_priv(dev);
3514
3515 if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
3516 printk(KERN_ERR "Handling parity error recovery. Try again later\n");
3517 return -EAGAIN;
3518 }
3519
3520 rtnl_lock();
3521
3522 pci_restore_state(pdev);
3523
3524 if (!netif_running(dev)) {
3525 rtnl_unlock();
3526 return 0;
3527 }
3528
3529 bnx2x_set_power_state(bp, PCI_D0);
3530 netif_device_attach(dev);
3531
3532 /* Since the chip was reset, clear the FW sequence number */
3533 bp->fw_seq = 0;
3534 rc = bnx2x_nic_load(bp, LOAD_OPEN);
3535
3536 rtnl_unlock();
3537
3538 return rc;
3539}
3540
3541
3542void bnx2x_set_ctx_validation(struct bnx2x *bp, struct eth_context *cxt,
3543 u32 cid)
3544{
3545 /* ustorm cxt validation */
3546 cxt->ustorm_ag_context.cdu_usage =
3547 CDU_RSRVD_VALUE_TYPE_A(HW_CID(bp, cid),
3548 CDU_REGION_NUMBER_UCM_AG, ETH_CONNECTION_TYPE);
3549 /* xcontext validation */
3550 cxt->xstorm_ag_context.cdu_reserved =
3551 CDU_RSRVD_VALUE_TYPE_A(HW_CID(bp, cid),
3552 CDU_REGION_NUMBER_XCM_AG, ETH_CONNECTION_TYPE);
3553}
3554
3555static inline void storm_memset_hc_timeout(struct bnx2x *bp, u8 port,
3556 u8 fw_sb_id, u8 sb_index,
3557 u8 ticks)
3558{
3559
3560 u32 addr = BAR_CSTRORM_INTMEM +
3561 CSTORM_STATUS_BLOCK_DATA_TIMEOUT_OFFSET(fw_sb_id, sb_index);
3562 REG_WR8(bp, addr, ticks);
3563 DP(NETIF_MSG_HW, "port %x fw_sb_id %d sb_index %d ticks %d\n",
3564 port, fw_sb_id, sb_index, ticks);
3565}
3566
3567static inline void storm_memset_hc_disable(struct bnx2x *bp, u8 port,
3568 u16 fw_sb_id, u8 sb_index,
3569 u8 disable)
3570{
3571 u32 enable_flag = disable ? 0 : (1 << HC_INDEX_DATA_HC_ENABLED_SHIFT);
3572 u32 addr = BAR_CSTRORM_INTMEM +
3573 CSTORM_STATUS_BLOCK_DATA_FLAGS_OFFSET(fw_sb_id, sb_index);
3574 u16 flags = REG_RD16(bp, addr);
3575 /* clear and set */
3576 flags &= ~HC_INDEX_DATA_HC_ENABLED;
3577 flags |= enable_flag;
3578 REG_WR16(bp, addr, flags);
3579 DP(NETIF_MSG_HW, "port %x fw_sb_id %d sb_index %d disable %d\n",
3580 port, fw_sb_id, sb_index, disable);
3581}
3582
3583void bnx2x_update_coalesce_sb_index(struct bnx2x *bp, u8 fw_sb_id,
3584 u8 sb_index, u8 disable, u16 usec)
3585{
3586 int port = BP_PORT(bp);
3587 u8 ticks = usec / BNX2X_BTR;
3588
3589 storm_memset_hc_timeout(bp, port, fw_sb_id, sb_index, ticks);
3590
3591 disable = disable ? 1 : (usec ? 0 : 1);
3592 storm_memset_hc_disable(bp, port, fw_sb_id, sb_index, disable);
3593}