bnx2x: Fix fastpath structures when memory allocation fails
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / drivers / net / ethernet / broadcom / bnx2x / bnx2x_cmn.c
blobf771ddfba646b0a773e89f3365b8add151087836
1 /* bnx2x_cmn.c: Broadcom Everest network driver.
3 * Copyright (c) 2007-2012 Broadcom Corporation
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation.
9 * Maintained by: Eilon Greenstein <eilong@broadcom.com>
10 * Written by: Eliezer Tamir
11 * Based on code from Michael Chan's bnx2 driver
12 * UDP CSUM errata workaround by Arik Gendelman
13 * Slowpath and fastpath rework by Vladislav Zolotarov
14 * Statistics and Link management by Yitchak Gertner
18 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
20 #include <linux/etherdevice.h>
21 #include <linux/if_vlan.h>
22 #include <linux/interrupt.h>
23 #include <linux/ip.h>
24 #include <net/ipv6.h>
25 #include <net/ip6_checksum.h>
26 #include <linux/prefetch.h>
27 #include "bnx2x_cmn.h"
28 #include "bnx2x_init.h"
29 #include "bnx2x_sp.h"
33 /**
34 * bnx2x_move_fp - move content of the fastpath structure.
36 * @bp: driver handle
37 * @from: source FP index
38 * @to: destination FP index
40 * Makes sure the contents of the bp->fp[to].napi is kept
41 * intact. This is done by first copying the napi struct from
42 * the target to the source, and then mem copying the entire
43 * source onto the target. Update txdata pointers and related
44 * content.
46 static inline void bnx2x_move_fp(struct bnx2x *bp, int from, int to)
48 struct bnx2x_fastpath *from_fp = &bp->fp[from];
49 struct bnx2x_fastpath *to_fp = &bp->fp[to];
50 struct bnx2x_sp_objs *from_sp_objs = &bp->sp_objs[from];
51 struct bnx2x_sp_objs *to_sp_objs = &bp->sp_objs[to];
52 struct bnx2x_fp_stats *from_fp_stats = &bp->fp_stats[from];
53 struct bnx2x_fp_stats *to_fp_stats = &bp->fp_stats[to];
54 int old_max_eth_txqs, new_max_eth_txqs;
55 int old_txdata_index = 0, new_txdata_index = 0;
57 /* Copy the NAPI object as it has been already initialized */
58 from_fp->napi = to_fp->napi;
60 /* Move bnx2x_fastpath contents */
61 memcpy(to_fp, from_fp, sizeof(*to_fp));
62 to_fp->index = to;
64 /* move sp_objs contents as well, as their indices match fp ones */
65 memcpy(to_sp_objs, from_sp_objs, sizeof(*to_sp_objs));
67 /* move fp_stats contents as well, as their indices match fp ones */
68 memcpy(to_fp_stats, from_fp_stats, sizeof(*to_fp_stats));
70 /* Update txdata pointers in fp and move txdata content accordingly:
71 * Each fp consumes 'max_cos' txdata structures, so the index should be
72 * decremented by max_cos x delta.
75 old_max_eth_txqs = BNX2X_NUM_ETH_QUEUES(bp) * (bp)->max_cos;
76 new_max_eth_txqs = (BNX2X_NUM_ETH_QUEUES(bp) - from + to) *
77 (bp)->max_cos;
78 if (from == FCOE_IDX(bp)) {
79 old_txdata_index = old_max_eth_txqs + FCOE_TXQ_IDX_OFFSET;
80 new_txdata_index = new_max_eth_txqs + FCOE_TXQ_IDX_OFFSET;
83 memcpy(&bp->bnx2x_txq[new_txdata_index],
84 &bp->bnx2x_txq[old_txdata_index],
85 sizeof(struct bnx2x_fp_txdata));
86 to_fp->txdata_ptr[0] = &bp->bnx2x_txq[new_txdata_index];
89 /**
90 * bnx2x_shrink_eth_fp - guarantees fastpath structures stay intact
92 * @bp: driver handle
93 * @delta: number of eth queues which were not allocated
95 static void bnx2x_shrink_eth_fp(struct bnx2x *bp, int delta)
97 int i, cos, old_eth_num = BNX2X_NUM_ETH_QUEUES(bp);
99 /* Queue pointer cannot be re-set on an fp-basis, as moving pointer
100 * backward along the array could cause memory to be overriden
102 for (cos = 1; cos < bp->max_cos; cos++) {
103 for (i = 0; i < old_eth_num - delta; i++) {
104 struct bnx2x_fastpath *fp = &bp->fp[i];
105 int new_idx = cos * (old_eth_num - delta) + i;
107 memcpy(&bp->bnx2x_txq[new_idx], fp->txdata_ptr[cos],
108 sizeof(struct bnx2x_fp_txdata));
109 fp->txdata_ptr[cos] = &bp->bnx2x_txq[new_idx];
114 int load_count[2][3] = { {0} }; /* per-path: 0-common, 1-port0, 2-port1 */
116 /* free skb in the packet ring at pos idx
117 * return idx of last bd freed
119 static u16 bnx2x_free_tx_pkt(struct bnx2x *bp, struct bnx2x_fp_txdata *txdata,
120 u16 idx, unsigned int *pkts_compl,
121 unsigned int *bytes_compl)
123 struct sw_tx_bd *tx_buf = &txdata->tx_buf_ring[idx];
124 struct eth_tx_start_bd *tx_start_bd;
125 struct eth_tx_bd *tx_data_bd;
126 struct sk_buff *skb = tx_buf->skb;
127 u16 bd_idx = TX_BD(tx_buf->first_bd), new_cons;
128 int nbd;
130 /* prefetch skb end pointer to speedup dev_kfree_skb() */
131 prefetch(&skb->end);
133 DP(NETIF_MSG_TX_DONE, "fp[%d]: pkt_idx %d buff @(%p)->skb %p\n",
134 txdata->txq_index, idx, tx_buf, skb);
136 /* unmap first bd */
137 tx_start_bd = &txdata->tx_desc_ring[bd_idx].start_bd;
138 dma_unmap_single(&bp->pdev->dev, BD_UNMAP_ADDR(tx_start_bd),
139 BD_UNMAP_LEN(tx_start_bd), DMA_TO_DEVICE);
142 nbd = le16_to_cpu(tx_start_bd->nbd) - 1;
143 #ifdef BNX2X_STOP_ON_ERROR
144 if ((nbd - 1) > (MAX_SKB_FRAGS + 2)) {
145 BNX2X_ERR("BAD nbd!\n");
146 bnx2x_panic();
148 #endif
149 new_cons = nbd + tx_buf->first_bd;
151 /* Get the next bd */
152 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
154 /* Skip a parse bd... */
155 --nbd;
156 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
158 /* ...and the TSO split header bd since they have no mapping */
159 if (tx_buf->flags & BNX2X_TSO_SPLIT_BD) {
160 --nbd;
161 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
164 /* now free frags */
165 while (nbd > 0) {
167 tx_data_bd = &txdata->tx_desc_ring[bd_idx].reg_bd;
168 dma_unmap_page(&bp->pdev->dev, BD_UNMAP_ADDR(tx_data_bd),
169 BD_UNMAP_LEN(tx_data_bd), DMA_TO_DEVICE);
170 if (--nbd)
171 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
174 /* release skb */
175 WARN_ON(!skb);
176 if (likely(skb)) {
177 (*pkts_compl)++;
178 (*bytes_compl) += skb->len;
181 dev_kfree_skb_any(skb);
182 tx_buf->first_bd = 0;
183 tx_buf->skb = NULL;
185 return new_cons;
188 int bnx2x_tx_int(struct bnx2x *bp, struct bnx2x_fp_txdata *txdata)
190 struct netdev_queue *txq;
191 u16 hw_cons, sw_cons, bd_cons = txdata->tx_bd_cons;
192 unsigned int pkts_compl = 0, bytes_compl = 0;
194 #ifdef BNX2X_STOP_ON_ERROR
195 if (unlikely(bp->panic))
196 return -1;
197 #endif
199 txq = netdev_get_tx_queue(bp->dev, txdata->txq_index);
200 hw_cons = le16_to_cpu(*txdata->tx_cons_sb);
201 sw_cons = txdata->tx_pkt_cons;
203 while (sw_cons != hw_cons) {
204 u16 pkt_cons;
206 pkt_cons = TX_BD(sw_cons);
208 DP(NETIF_MSG_TX_DONE,
209 "queue[%d]: hw_cons %u sw_cons %u pkt_cons %u\n",
210 txdata->txq_index, hw_cons, sw_cons, pkt_cons);
212 bd_cons = bnx2x_free_tx_pkt(bp, txdata, pkt_cons,
213 &pkts_compl, &bytes_compl);
215 sw_cons++;
218 netdev_tx_completed_queue(txq, pkts_compl, bytes_compl);
220 txdata->tx_pkt_cons = sw_cons;
221 txdata->tx_bd_cons = bd_cons;
223 /* Need to make the tx_bd_cons update visible to start_xmit()
224 * before checking for netif_tx_queue_stopped(). Without the
225 * memory barrier, there is a small possibility that
226 * start_xmit() will miss it and cause the queue to be stopped
227 * forever.
228 * On the other hand we need an rmb() here to ensure the proper
229 * ordering of bit testing in the following
230 * netif_tx_queue_stopped(txq) call.
232 smp_mb();
234 if (unlikely(netif_tx_queue_stopped(txq))) {
235 /* Taking tx_lock() is needed to prevent reenabling the queue
236 * while it's empty. This could have happen if rx_action() gets
237 * suspended in bnx2x_tx_int() after the condition before
238 * netif_tx_wake_queue(), while tx_action (bnx2x_start_xmit()):
240 * stops the queue->sees fresh tx_bd_cons->releases the queue->
241 * sends some packets consuming the whole queue again->
242 * stops the queue
245 __netif_tx_lock(txq, smp_processor_id());
247 if ((netif_tx_queue_stopped(txq)) &&
248 (bp->state == BNX2X_STATE_OPEN) &&
249 (bnx2x_tx_avail(bp, txdata) >= MAX_DESC_PER_TX_PKT))
250 netif_tx_wake_queue(txq);
252 __netif_tx_unlock(txq);
254 return 0;
257 static inline void bnx2x_update_last_max_sge(struct bnx2x_fastpath *fp,
258 u16 idx)
260 u16 last_max = fp->last_max_sge;
262 if (SUB_S16(idx, last_max) > 0)
263 fp->last_max_sge = idx;
266 static inline void bnx2x_update_sge_prod(struct bnx2x_fastpath *fp,
267 u16 sge_len,
268 struct eth_end_agg_rx_cqe *cqe)
270 struct bnx2x *bp = fp->bp;
271 u16 last_max, last_elem, first_elem;
272 u16 delta = 0;
273 u16 i;
275 if (!sge_len)
276 return;
278 /* First mark all used pages */
279 for (i = 0; i < sge_len; i++)
280 BIT_VEC64_CLEAR_BIT(fp->sge_mask,
281 RX_SGE(le16_to_cpu(cqe->sgl_or_raw_data.sgl[i])));
283 DP(NETIF_MSG_RX_STATUS, "fp_cqe->sgl[%d] = %d\n",
284 sge_len - 1, le16_to_cpu(cqe->sgl_or_raw_data.sgl[sge_len - 1]));
286 /* Here we assume that the last SGE index is the biggest */
287 prefetch((void *)(fp->sge_mask));
288 bnx2x_update_last_max_sge(fp,
289 le16_to_cpu(cqe->sgl_or_raw_data.sgl[sge_len - 1]));
291 last_max = RX_SGE(fp->last_max_sge);
292 last_elem = last_max >> BIT_VEC64_ELEM_SHIFT;
293 first_elem = RX_SGE(fp->rx_sge_prod) >> BIT_VEC64_ELEM_SHIFT;
295 /* If ring is not full */
296 if (last_elem + 1 != first_elem)
297 last_elem++;
299 /* Now update the prod */
300 for (i = first_elem; i != last_elem; i = NEXT_SGE_MASK_ELEM(i)) {
301 if (likely(fp->sge_mask[i]))
302 break;
304 fp->sge_mask[i] = BIT_VEC64_ELEM_ONE_MASK;
305 delta += BIT_VEC64_ELEM_SZ;
308 if (delta > 0) {
309 fp->rx_sge_prod += delta;
310 /* clear page-end entries */
311 bnx2x_clear_sge_mask_next_elems(fp);
314 DP(NETIF_MSG_RX_STATUS,
315 "fp->last_max_sge = %d fp->rx_sge_prod = %d\n",
316 fp->last_max_sge, fp->rx_sge_prod);
319 /* Set Toeplitz hash value in the skb using the value from the
320 * CQE (calculated by HW).
322 static u32 bnx2x_get_rxhash(const struct bnx2x *bp,
323 const struct eth_fast_path_rx_cqe *cqe,
324 bool *l4_rxhash)
326 /* Set Toeplitz hash from CQE */
327 if ((bp->dev->features & NETIF_F_RXHASH) &&
328 (cqe->status_flags & ETH_FAST_PATH_RX_CQE_RSS_HASH_FLG)) {
329 enum eth_rss_hash_type htype;
331 htype = cqe->status_flags & ETH_FAST_PATH_RX_CQE_RSS_HASH_TYPE;
332 *l4_rxhash = (htype == TCP_IPV4_HASH_TYPE) ||
333 (htype == TCP_IPV6_HASH_TYPE);
334 return le32_to_cpu(cqe->rss_hash_result);
336 *l4_rxhash = false;
337 return 0;
340 static void bnx2x_tpa_start(struct bnx2x_fastpath *fp, u16 queue,
341 u16 cons, u16 prod,
342 struct eth_fast_path_rx_cqe *cqe)
344 struct bnx2x *bp = fp->bp;
345 struct sw_rx_bd *cons_rx_buf = &fp->rx_buf_ring[cons];
346 struct sw_rx_bd *prod_rx_buf = &fp->rx_buf_ring[prod];
347 struct eth_rx_bd *prod_bd = &fp->rx_desc_ring[prod];
348 dma_addr_t mapping;
349 struct bnx2x_agg_info *tpa_info = &fp->tpa_info[queue];
350 struct sw_rx_bd *first_buf = &tpa_info->first_buf;
352 /* print error if current state != stop */
353 if (tpa_info->tpa_state != BNX2X_TPA_STOP)
354 BNX2X_ERR("start of bin not in stop [%d]\n", queue);
356 /* Try to map an empty data buffer from the aggregation info */
357 mapping = dma_map_single(&bp->pdev->dev,
358 first_buf->data + NET_SKB_PAD,
359 fp->rx_buf_size, DMA_FROM_DEVICE);
361 * ...if it fails - move the skb from the consumer to the producer
362 * and set the current aggregation state as ERROR to drop it
363 * when TPA_STOP arrives.
366 if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
367 /* Move the BD from the consumer to the producer */
368 bnx2x_reuse_rx_data(fp, cons, prod);
369 tpa_info->tpa_state = BNX2X_TPA_ERROR;
370 return;
373 /* move empty data from pool to prod */
374 prod_rx_buf->data = first_buf->data;
375 dma_unmap_addr_set(prod_rx_buf, mapping, mapping);
376 /* point prod_bd to new data */
377 prod_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
378 prod_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
380 /* move partial skb from cons to pool (don't unmap yet) */
381 *first_buf = *cons_rx_buf;
383 /* mark bin state as START */
384 tpa_info->parsing_flags =
385 le16_to_cpu(cqe->pars_flags.flags);
386 tpa_info->vlan_tag = le16_to_cpu(cqe->vlan_tag);
387 tpa_info->tpa_state = BNX2X_TPA_START;
388 tpa_info->len_on_bd = le16_to_cpu(cqe->len_on_bd);
389 tpa_info->placement_offset = cqe->placement_offset;
390 tpa_info->rxhash = bnx2x_get_rxhash(bp, cqe, &tpa_info->l4_rxhash);
391 if (fp->mode == TPA_MODE_GRO) {
392 u16 gro_size = le16_to_cpu(cqe->pkt_len_or_gro_seg_len);
393 tpa_info->full_page =
394 SGE_PAGE_SIZE * PAGES_PER_SGE / gro_size * gro_size;
395 tpa_info->gro_size = gro_size;
398 #ifdef BNX2X_STOP_ON_ERROR
399 fp->tpa_queue_used |= (1 << queue);
400 #ifdef _ASM_GENERIC_INT_L64_H
401 DP(NETIF_MSG_RX_STATUS, "fp->tpa_queue_used = 0x%lx\n",
402 #else
403 DP(NETIF_MSG_RX_STATUS, "fp->tpa_queue_used = 0x%llx\n",
404 #endif
405 fp->tpa_queue_used);
406 #endif
409 /* Timestamp option length allowed for TPA aggregation:
411 * nop nop kind length echo val
413 #define TPA_TSTAMP_OPT_LEN 12
415 * bnx2x_set_lro_mss - calculate the approximate value of the MSS
417 * @bp: driver handle
418 * @parsing_flags: parsing flags from the START CQE
419 * @len_on_bd: total length of the first packet for the
420 * aggregation.
422 * Approximate value of the MSS for this aggregation calculated using
423 * the first packet of it.
425 static u16 bnx2x_set_lro_mss(struct bnx2x *bp, u16 parsing_flags,
426 u16 len_on_bd)
429 * TPA arrgregation won't have either IP options or TCP options
430 * other than timestamp or IPv6 extension headers.
432 u16 hdrs_len = ETH_HLEN + sizeof(struct tcphdr);
434 if (GET_FLAG(parsing_flags, PARSING_FLAGS_OVER_ETHERNET_PROTOCOL) ==
435 PRS_FLAG_OVERETH_IPV6)
436 hdrs_len += sizeof(struct ipv6hdr);
437 else /* IPv4 */
438 hdrs_len += sizeof(struct iphdr);
441 /* Check if there was a TCP timestamp, if there is it's will
442 * always be 12 bytes length: nop nop kind length echo val.
444 * Otherwise FW would close the aggregation.
446 if (parsing_flags & PARSING_FLAGS_TIME_STAMP_EXIST_FLAG)
447 hdrs_len += TPA_TSTAMP_OPT_LEN;
449 return len_on_bd - hdrs_len;
452 static int bnx2x_alloc_rx_sge(struct bnx2x *bp,
453 struct bnx2x_fastpath *fp, u16 index)
455 struct page *page = alloc_pages(GFP_ATOMIC, PAGES_PER_SGE_SHIFT);
456 struct sw_rx_page *sw_buf = &fp->rx_page_ring[index];
457 struct eth_rx_sge *sge = &fp->rx_sge_ring[index];
458 dma_addr_t mapping;
460 if (unlikely(page == NULL)) {
461 BNX2X_ERR("Can't alloc sge\n");
462 return -ENOMEM;
465 mapping = dma_map_page(&bp->pdev->dev, page, 0,
466 SGE_PAGE_SIZE*PAGES_PER_SGE, DMA_FROM_DEVICE);
467 if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
468 __free_pages(page, PAGES_PER_SGE_SHIFT);
469 BNX2X_ERR("Can't map sge\n");
470 return -ENOMEM;
473 sw_buf->page = page;
474 dma_unmap_addr_set(sw_buf, mapping, mapping);
476 sge->addr_hi = cpu_to_le32(U64_HI(mapping));
477 sge->addr_lo = cpu_to_le32(U64_LO(mapping));
479 return 0;
482 static int bnx2x_fill_frag_skb(struct bnx2x *bp, struct bnx2x_fastpath *fp,
483 struct bnx2x_agg_info *tpa_info,
484 u16 pages,
485 struct sk_buff *skb,
486 struct eth_end_agg_rx_cqe *cqe,
487 u16 cqe_idx)
489 struct sw_rx_page *rx_pg, old_rx_pg;
490 u32 i, frag_len, frag_size;
491 int err, j, frag_id = 0;
492 u16 len_on_bd = tpa_info->len_on_bd;
493 u16 full_page = 0, gro_size = 0;
495 frag_size = le16_to_cpu(cqe->pkt_len) - len_on_bd;
497 if (fp->mode == TPA_MODE_GRO) {
498 gro_size = tpa_info->gro_size;
499 full_page = tpa_info->full_page;
502 /* This is needed in order to enable forwarding support */
503 if (frag_size) {
504 skb_shinfo(skb)->gso_size = bnx2x_set_lro_mss(bp,
505 tpa_info->parsing_flags, len_on_bd);
507 /* set for GRO */
508 if (fp->mode == TPA_MODE_GRO)
509 skb_shinfo(skb)->gso_type =
510 (GET_FLAG(tpa_info->parsing_flags,
511 PARSING_FLAGS_OVER_ETHERNET_PROTOCOL) ==
512 PRS_FLAG_OVERETH_IPV6) ?
513 SKB_GSO_TCPV6 : SKB_GSO_TCPV4;
517 #ifdef BNX2X_STOP_ON_ERROR
518 if (pages > min_t(u32, 8, MAX_SKB_FRAGS)*SGE_PAGE_SIZE*PAGES_PER_SGE) {
519 BNX2X_ERR("SGL length is too long: %d. CQE index is %d\n",
520 pages, cqe_idx);
521 BNX2X_ERR("cqe->pkt_len = %d\n", cqe->pkt_len);
522 bnx2x_panic();
523 return -EINVAL;
525 #endif
527 /* Run through the SGL and compose the fragmented skb */
528 for (i = 0, j = 0; i < pages; i += PAGES_PER_SGE, j++) {
529 u16 sge_idx = RX_SGE(le16_to_cpu(cqe->sgl_or_raw_data.sgl[j]));
531 /* FW gives the indices of the SGE as if the ring is an array
532 (meaning that "next" element will consume 2 indices) */
533 if (fp->mode == TPA_MODE_GRO)
534 frag_len = min_t(u32, frag_size, (u32)full_page);
535 else /* LRO */
536 frag_len = min_t(u32, frag_size,
537 (u32)(SGE_PAGE_SIZE * PAGES_PER_SGE));
539 rx_pg = &fp->rx_page_ring[sge_idx];
540 old_rx_pg = *rx_pg;
542 /* If we fail to allocate a substitute page, we simply stop
543 where we are and drop the whole packet */
544 err = bnx2x_alloc_rx_sge(bp, fp, sge_idx);
545 if (unlikely(err)) {
546 bnx2x_fp_qstats(bp, fp)->rx_skb_alloc_failed++;
547 return err;
550 /* Unmap the page as we r going to pass it to the stack */
551 dma_unmap_page(&bp->pdev->dev,
552 dma_unmap_addr(&old_rx_pg, mapping),
553 SGE_PAGE_SIZE*PAGES_PER_SGE, DMA_FROM_DEVICE);
554 /* Add one frag and update the appropriate fields in the skb */
555 if (fp->mode == TPA_MODE_LRO)
556 skb_fill_page_desc(skb, j, old_rx_pg.page, 0, frag_len);
557 else { /* GRO */
558 int rem;
559 int offset = 0;
560 for (rem = frag_len; rem > 0; rem -= gro_size) {
561 int len = rem > gro_size ? gro_size : rem;
562 skb_fill_page_desc(skb, frag_id++,
563 old_rx_pg.page, offset, len);
564 if (offset)
565 get_page(old_rx_pg.page);
566 offset += len;
570 skb->data_len += frag_len;
571 skb->truesize += SGE_PAGE_SIZE * PAGES_PER_SGE;
572 skb->len += frag_len;
574 frag_size -= frag_len;
577 return 0;
580 static void bnx2x_frag_free(const struct bnx2x_fastpath *fp, void *data)
582 if (fp->rx_frag_size)
583 put_page(virt_to_head_page(data));
584 else
585 kfree(data);
588 static void *bnx2x_frag_alloc(const struct bnx2x_fastpath *fp)
590 if (fp->rx_frag_size)
591 return netdev_alloc_frag(fp->rx_frag_size);
593 return kmalloc(fp->rx_buf_size + NET_SKB_PAD, GFP_ATOMIC);
597 static void bnx2x_tpa_stop(struct bnx2x *bp, struct bnx2x_fastpath *fp,
598 struct bnx2x_agg_info *tpa_info,
599 u16 pages,
600 struct eth_end_agg_rx_cqe *cqe,
601 u16 cqe_idx)
603 struct sw_rx_bd *rx_buf = &tpa_info->first_buf;
604 u8 pad = tpa_info->placement_offset;
605 u16 len = tpa_info->len_on_bd;
606 struct sk_buff *skb = NULL;
607 u8 *new_data, *data = rx_buf->data;
608 u8 old_tpa_state = tpa_info->tpa_state;
610 tpa_info->tpa_state = BNX2X_TPA_STOP;
612 /* If we there was an error during the handling of the TPA_START -
613 * drop this aggregation.
615 if (old_tpa_state == BNX2X_TPA_ERROR)
616 goto drop;
618 /* Try to allocate the new data */
619 new_data = bnx2x_frag_alloc(fp);
620 /* Unmap skb in the pool anyway, as we are going to change
621 pool entry status to BNX2X_TPA_STOP even if new skb allocation
622 fails. */
623 dma_unmap_single(&bp->pdev->dev, dma_unmap_addr(rx_buf, mapping),
624 fp->rx_buf_size, DMA_FROM_DEVICE);
625 if (likely(new_data))
626 skb = build_skb(data, fp->rx_frag_size);
628 if (likely(skb)) {
629 #ifdef BNX2X_STOP_ON_ERROR
630 if (pad + len > fp->rx_buf_size) {
631 BNX2X_ERR("skb_put is about to fail... pad %d len %d rx_buf_size %d\n",
632 pad, len, fp->rx_buf_size);
633 bnx2x_panic();
634 return;
636 #endif
638 skb_reserve(skb, pad + NET_SKB_PAD);
639 skb_put(skb, len);
640 skb->rxhash = tpa_info->rxhash;
641 skb->l4_rxhash = tpa_info->l4_rxhash;
643 skb->protocol = eth_type_trans(skb, bp->dev);
644 skb->ip_summed = CHECKSUM_UNNECESSARY;
646 if (!bnx2x_fill_frag_skb(bp, fp, tpa_info, pages,
647 skb, cqe, cqe_idx)) {
648 if (tpa_info->parsing_flags & PARSING_FLAGS_VLAN)
649 __vlan_hwaccel_put_tag(skb, tpa_info->vlan_tag);
650 napi_gro_receive(&fp->napi, skb);
651 } else {
652 DP(NETIF_MSG_RX_STATUS,
653 "Failed to allocate new pages - dropping packet!\n");
654 dev_kfree_skb_any(skb);
658 /* put new data in bin */
659 rx_buf->data = new_data;
661 return;
663 bnx2x_frag_free(fp, new_data);
664 drop:
665 /* drop the packet and keep the buffer in the bin */
666 DP(NETIF_MSG_RX_STATUS,
667 "Failed to allocate or map a new skb - dropping packet!\n");
668 bnx2x_fp_stats(bp, fp)->eth_q_stats.rx_skb_alloc_failed++;
671 static int bnx2x_alloc_rx_data(struct bnx2x *bp,
672 struct bnx2x_fastpath *fp, u16 index)
674 u8 *data;
675 struct sw_rx_bd *rx_buf = &fp->rx_buf_ring[index];
676 struct eth_rx_bd *rx_bd = &fp->rx_desc_ring[index];
677 dma_addr_t mapping;
679 data = bnx2x_frag_alloc(fp);
680 if (unlikely(data == NULL))
681 return -ENOMEM;
683 mapping = dma_map_single(&bp->pdev->dev, data + NET_SKB_PAD,
684 fp->rx_buf_size,
685 DMA_FROM_DEVICE);
686 if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
687 bnx2x_frag_free(fp, data);
688 BNX2X_ERR("Can't map rx data\n");
689 return -ENOMEM;
692 rx_buf->data = data;
693 dma_unmap_addr_set(rx_buf, mapping, mapping);
695 rx_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
696 rx_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
698 return 0;
701 static
702 void bnx2x_csum_validate(struct sk_buff *skb, union eth_rx_cqe *cqe,
703 struct bnx2x_fastpath *fp,
704 struct bnx2x_eth_q_stats *qstats)
706 /* Do nothing if no L4 csum validation was done.
707 * We do not check whether IP csum was validated. For IPv4 we assume
708 * that if the card got as far as validating the L4 csum, it also
709 * validated the IP csum. IPv6 has no IP csum.
711 if (cqe->fast_path_cqe.status_flags &
712 ETH_FAST_PATH_RX_CQE_L4_XSUM_NO_VALIDATION_FLG)
713 return;
715 /* If L4 validation was done, check if an error was found. */
717 if (cqe->fast_path_cqe.type_error_flags &
718 (ETH_FAST_PATH_RX_CQE_IP_BAD_XSUM_FLG |
719 ETH_FAST_PATH_RX_CQE_L4_BAD_XSUM_FLG))
720 qstats->hw_csum_err++;
721 else
722 skb->ip_summed = CHECKSUM_UNNECESSARY;
725 int bnx2x_rx_int(struct bnx2x_fastpath *fp, int budget)
727 struct bnx2x *bp = fp->bp;
728 u16 bd_cons, bd_prod, bd_prod_fw, comp_ring_cons;
729 u16 hw_comp_cons, sw_comp_cons, sw_comp_prod;
730 int rx_pkt = 0;
732 #ifdef BNX2X_STOP_ON_ERROR
733 if (unlikely(bp->panic))
734 return 0;
735 #endif
737 /* CQ "next element" is of the size of the regular element,
738 that's why it's ok here */
739 hw_comp_cons = le16_to_cpu(*fp->rx_cons_sb);
740 if ((hw_comp_cons & MAX_RCQ_DESC_CNT) == MAX_RCQ_DESC_CNT)
741 hw_comp_cons++;
743 bd_cons = fp->rx_bd_cons;
744 bd_prod = fp->rx_bd_prod;
745 bd_prod_fw = bd_prod;
746 sw_comp_cons = fp->rx_comp_cons;
747 sw_comp_prod = fp->rx_comp_prod;
749 /* Memory barrier necessary as speculative reads of the rx
750 * buffer can be ahead of the index in the status block
752 rmb();
754 DP(NETIF_MSG_RX_STATUS,
755 "queue[%d]: hw_comp_cons %u sw_comp_cons %u\n",
756 fp->index, hw_comp_cons, sw_comp_cons);
758 while (sw_comp_cons != hw_comp_cons) {
759 struct sw_rx_bd *rx_buf = NULL;
760 struct sk_buff *skb;
761 union eth_rx_cqe *cqe;
762 struct eth_fast_path_rx_cqe *cqe_fp;
763 u8 cqe_fp_flags;
764 enum eth_rx_cqe_type cqe_fp_type;
765 u16 len, pad, queue;
766 u8 *data;
767 bool l4_rxhash;
769 #ifdef BNX2X_STOP_ON_ERROR
770 if (unlikely(bp->panic))
771 return 0;
772 #endif
774 comp_ring_cons = RCQ_BD(sw_comp_cons);
775 bd_prod = RX_BD(bd_prod);
776 bd_cons = RX_BD(bd_cons);
778 cqe = &fp->rx_comp_ring[comp_ring_cons];
779 cqe_fp = &cqe->fast_path_cqe;
780 cqe_fp_flags = cqe_fp->type_error_flags;
781 cqe_fp_type = cqe_fp_flags & ETH_FAST_PATH_RX_CQE_TYPE;
783 DP(NETIF_MSG_RX_STATUS,
784 "CQE type %x err %x status %x queue %x vlan %x len %u\n",
785 CQE_TYPE(cqe_fp_flags),
786 cqe_fp_flags, cqe_fp->status_flags,
787 le32_to_cpu(cqe_fp->rss_hash_result),
788 le16_to_cpu(cqe_fp->vlan_tag),
789 le16_to_cpu(cqe_fp->pkt_len_or_gro_seg_len));
791 /* is this a slowpath msg? */
792 if (unlikely(CQE_TYPE_SLOW(cqe_fp_type))) {
793 bnx2x_sp_event(fp, cqe);
794 goto next_cqe;
797 rx_buf = &fp->rx_buf_ring[bd_cons];
798 data = rx_buf->data;
800 if (!CQE_TYPE_FAST(cqe_fp_type)) {
801 struct bnx2x_agg_info *tpa_info;
802 u16 frag_size, pages;
803 #ifdef BNX2X_STOP_ON_ERROR
804 /* sanity check */
805 if (fp->disable_tpa &&
806 (CQE_TYPE_START(cqe_fp_type) ||
807 CQE_TYPE_STOP(cqe_fp_type)))
808 BNX2X_ERR("START/STOP packet while disable_tpa type %x\n",
809 CQE_TYPE(cqe_fp_type));
810 #endif
812 if (CQE_TYPE_START(cqe_fp_type)) {
813 u16 queue = cqe_fp->queue_index;
814 DP(NETIF_MSG_RX_STATUS,
815 "calling tpa_start on queue %d\n",
816 queue);
818 bnx2x_tpa_start(fp, queue,
819 bd_cons, bd_prod,
820 cqe_fp);
822 goto next_rx;
825 queue = cqe->end_agg_cqe.queue_index;
826 tpa_info = &fp->tpa_info[queue];
827 DP(NETIF_MSG_RX_STATUS,
828 "calling tpa_stop on queue %d\n",
829 queue);
831 frag_size = le16_to_cpu(cqe->end_agg_cqe.pkt_len) -
832 tpa_info->len_on_bd;
834 if (fp->mode == TPA_MODE_GRO)
835 pages = (frag_size + tpa_info->full_page - 1) /
836 tpa_info->full_page;
837 else
838 pages = SGE_PAGE_ALIGN(frag_size) >>
839 SGE_PAGE_SHIFT;
841 bnx2x_tpa_stop(bp, fp, tpa_info, pages,
842 &cqe->end_agg_cqe, comp_ring_cons);
843 #ifdef BNX2X_STOP_ON_ERROR
844 if (bp->panic)
845 return 0;
846 #endif
848 bnx2x_update_sge_prod(fp, pages, &cqe->end_agg_cqe);
849 goto next_cqe;
851 /* non TPA */
852 len = le16_to_cpu(cqe_fp->pkt_len_or_gro_seg_len);
853 pad = cqe_fp->placement_offset;
854 dma_sync_single_for_cpu(&bp->pdev->dev,
855 dma_unmap_addr(rx_buf, mapping),
856 pad + RX_COPY_THRESH,
857 DMA_FROM_DEVICE);
858 pad += NET_SKB_PAD;
859 prefetch(data + pad); /* speedup eth_type_trans() */
860 /* is this an error packet? */
861 if (unlikely(cqe_fp_flags & ETH_RX_ERROR_FALGS)) {
862 DP(NETIF_MSG_RX_ERR | NETIF_MSG_RX_STATUS,
863 "ERROR flags %x rx packet %u\n",
864 cqe_fp_flags, sw_comp_cons);
865 bnx2x_fp_qstats(bp, fp)->rx_err_discard_pkt++;
866 goto reuse_rx;
869 /* Since we don't have a jumbo ring
870 * copy small packets if mtu > 1500
872 if ((bp->dev->mtu > ETH_MAX_PACKET_SIZE) &&
873 (len <= RX_COPY_THRESH)) {
874 skb = netdev_alloc_skb_ip_align(bp->dev, len);
875 if (skb == NULL) {
876 DP(NETIF_MSG_RX_ERR | NETIF_MSG_RX_STATUS,
877 "ERROR packet dropped because of alloc failure\n");
878 bnx2x_fp_qstats(bp, fp)->rx_skb_alloc_failed++;
879 goto reuse_rx;
881 memcpy(skb->data, data + pad, len);
882 bnx2x_reuse_rx_data(fp, bd_cons, bd_prod);
883 } else {
884 if (likely(bnx2x_alloc_rx_data(bp, fp, bd_prod) == 0)) {
885 dma_unmap_single(&bp->pdev->dev,
886 dma_unmap_addr(rx_buf, mapping),
887 fp->rx_buf_size,
888 DMA_FROM_DEVICE);
889 skb = build_skb(data, fp->rx_frag_size);
890 if (unlikely(!skb)) {
891 bnx2x_frag_free(fp, data);
892 bnx2x_fp_qstats(bp, fp)->
893 rx_skb_alloc_failed++;
894 goto next_rx;
896 skb_reserve(skb, pad);
897 } else {
898 DP(NETIF_MSG_RX_ERR | NETIF_MSG_RX_STATUS,
899 "ERROR packet dropped because of alloc failure\n");
900 bnx2x_fp_qstats(bp, fp)->rx_skb_alloc_failed++;
901 reuse_rx:
902 bnx2x_reuse_rx_data(fp, bd_cons, bd_prod);
903 goto next_rx;
907 skb_put(skb, len);
908 skb->protocol = eth_type_trans(skb, bp->dev);
910 /* Set Toeplitz hash for a none-LRO skb */
911 skb->rxhash = bnx2x_get_rxhash(bp, cqe_fp, &l4_rxhash);
912 skb->l4_rxhash = l4_rxhash;
914 skb_checksum_none_assert(skb);
916 if (bp->dev->features & NETIF_F_RXCSUM)
917 bnx2x_csum_validate(skb, cqe, fp,
918 bnx2x_fp_qstats(bp, fp));
920 skb_record_rx_queue(skb, fp->rx_queue);
922 if (le16_to_cpu(cqe_fp->pars_flags.flags) &
923 PARSING_FLAGS_VLAN)
924 __vlan_hwaccel_put_tag(skb,
925 le16_to_cpu(cqe_fp->vlan_tag));
926 napi_gro_receive(&fp->napi, skb);
929 next_rx:
930 rx_buf->data = NULL;
932 bd_cons = NEXT_RX_IDX(bd_cons);
933 bd_prod = NEXT_RX_IDX(bd_prod);
934 bd_prod_fw = NEXT_RX_IDX(bd_prod_fw);
935 rx_pkt++;
936 next_cqe:
937 sw_comp_prod = NEXT_RCQ_IDX(sw_comp_prod);
938 sw_comp_cons = NEXT_RCQ_IDX(sw_comp_cons);
940 if (rx_pkt == budget)
941 break;
942 } /* while */
944 fp->rx_bd_cons = bd_cons;
945 fp->rx_bd_prod = bd_prod_fw;
946 fp->rx_comp_cons = sw_comp_cons;
947 fp->rx_comp_prod = sw_comp_prod;
949 /* Update producers */
950 bnx2x_update_rx_prod(bp, fp, bd_prod_fw, sw_comp_prod,
951 fp->rx_sge_prod);
953 fp->rx_pkt += rx_pkt;
954 fp->rx_calls++;
956 return rx_pkt;
959 static irqreturn_t bnx2x_msix_fp_int(int irq, void *fp_cookie)
961 struct bnx2x_fastpath *fp = fp_cookie;
962 struct bnx2x *bp = fp->bp;
963 u8 cos;
965 DP(NETIF_MSG_INTR,
966 "got an MSI-X interrupt on IDX:SB [fp %d fw_sd %d igusb %d]\n",
967 fp->index, fp->fw_sb_id, fp->igu_sb_id);
968 bnx2x_ack_sb(bp, fp->igu_sb_id, USTORM_ID, 0, IGU_INT_DISABLE, 0);
970 #ifdef BNX2X_STOP_ON_ERROR
971 if (unlikely(bp->panic))
972 return IRQ_HANDLED;
973 #endif
975 /* Handle Rx and Tx according to MSI-X vector */
976 prefetch(fp->rx_cons_sb);
978 for_each_cos_in_tx_queue(fp, cos)
979 prefetch(fp->txdata_ptr[cos]->tx_cons_sb);
981 prefetch(&fp->sb_running_index[SM_RX_ID]);
982 napi_schedule(&bnx2x_fp(bp, fp->index, napi));
984 return IRQ_HANDLED;
987 /* HW Lock for shared dual port PHYs */
988 void bnx2x_acquire_phy_lock(struct bnx2x *bp)
990 mutex_lock(&bp->port.phy_mutex);
992 bnx2x_acquire_hw_lock(bp, HW_LOCK_RESOURCE_MDIO);
995 void bnx2x_release_phy_lock(struct bnx2x *bp)
997 bnx2x_release_hw_lock(bp, HW_LOCK_RESOURCE_MDIO);
999 mutex_unlock(&bp->port.phy_mutex);
1002 /* calculates MF speed according to current linespeed and MF configuration */
1003 u16 bnx2x_get_mf_speed(struct bnx2x *bp)
1005 u16 line_speed = bp->link_vars.line_speed;
1006 if (IS_MF(bp)) {
1007 u16 maxCfg = bnx2x_extract_max_cfg(bp,
1008 bp->mf_config[BP_VN(bp)]);
1010 /* Calculate the current MAX line speed limit for the MF
1011 * devices
1013 if (IS_MF_SI(bp))
1014 line_speed = (line_speed * maxCfg) / 100;
1015 else { /* SD mode */
1016 u16 vn_max_rate = maxCfg * 100;
1018 if (vn_max_rate < line_speed)
1019 line_speed = vn_max_rate;
1023 return line_speed;
1027 * bnx2x_fill_report_data - fill link report data to report
1029 * @bp: driver handle
1030 * @data: link state to update
1032 * It uses a none-atomic bit operations because is called under the mutex.
1034 static void bnx2x_fill_report_data(struct bnx2x *bp,
1035 struct bnx2x_link_report_data *data)
1037 u16 line_speed = bnx2x_get_mf_speed(bp);
1039 memset(data, 0, sizeof(*data));
1041 /* Fill the report data: efective line speed */
1042 data->line_speed = line_speed;
1044 /* Link is down */
1045 if (!bp->link_vars.link_up || (bp->flags & MF_FUNC_DIS))
1046 __set_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1047 &data->link_report_flags);
1049 /* Full DUPLEX */
1050 if (bp->link_vars.duplex == DUPLEX_FULL)
1051 __set_bit(BNX2X_LINK_REPORT_FD, &data->link_report_flags);
1053 /* Rx Flow Control is ON */
1054 if (bp->link_vars.flow_ctrl & BNX2X_FLOW_CTRL_RX)
1055 __set_bit(BNX2X_LINK_REPORT_RX_FC_ON, &data->link_report_flags);
1057 /* Tx Flow Control is ON */
1058 if (bp->link_vars.flow_ctrl & BNX2X_FLOW_CTRL_TX)
1059 __set_bit(BNX2X_LINK_REPORT_TX_FC_ON, &data->link_report_flags);
1063 * bnx2x_link_report - report link status to OS.
1065 * @bp: driver handle
1067 * Calls the __bnx2x_link_report() under the same locking scheme
1068 * as a link/PHY state managing code to ensure a consistent link
1069 * reporting.
1072 void bnx2x_link_report(struct bnx2x *bp)
1074 bnx2x_acquire_phy_lock(bp);
1075 __bnx2x_link_report(bp);
1076 bnx2x_release_phy_lock(bp);
1080 * __bnx2x_link_report - report link status to OS.
1082 * @bp: driver handle
1084 * None atomic inmlementation.
1085 * Should be called under the phy_lock.
1087 void __bnx2x_link_report(struct bnx2x *bp)
1089 struct bnx2x_link_report_data cur_data;
1091 /* reread mf_cfg */
1092 if (!CHIP_IS_E1(bp))
1093 bnx2x_read_mf_cfg(bp);
1095 /* Read the current link report info */
1096 bnx2x_fill_report_data(bp, &cur_data);
1098 /* Don't report link down or exactly the same link status twice */
1099 if (!memcmp(&cur_data, &bp->last_reported_link, sizeof(cur_data)) ||
1100 (test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1101 &bp->last_reported_link.link_report_flags) &&
1102 test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1103 &cur_data.link_report_flags)))
1104 return;
1106 bp->link_cnt++;
1108 /* We are going to report a new link parameters now -
1109 * remember the current data for the next time.
1111 memcpy(&bp->last_reported_link, &cur_data, sizeof(cur_data));
1113 if (test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1114 &cur_data.link_report_flags)) {
1115 netif_carrier_off(bp->dev);
1116 netdev_err(bp->dev, "NIC Link is Down\n");
1117 return;
1118 } else {
1119 const char *duplex;
1120 const char *flow;
1122 netif_carrier_on(bp->dev);
1124 if (test_and_clear_bit(BNX2X_LINK_REPORT_FD,
1125 &cur_data.link_report_flags))
1126 duplex = "full";
1127 else
1128 duplex = "half";
1130 /* Handle the FC at the end so that only these flags would be
1131 * possibly set. This way we may easily check if there is no FC
1132 * enabled.
1134 if (cur_data.link_report_flags) {
1135 if (test_bit(BNX2X_LINK_REPORT_RX_FC_ON,
1136 &cur_data.link_report_flags)) {
1137 if (test_bit(BNX2X_LINK_REPORT_TX_FC_ON,
1138 &cur_data.link_report_flags))
1139 flow = "ON - receive & transmit";
1140 else
1141 flow = "ON - receive";
1142 } else {
1143 flow = "ON - transmit";
1145 } else {
1146 flow = "none";
1148 netdev_info(bp->dev, "NIC Link is Up, %d Mbps %s duplex, Flow control: %s\n",
1149 cur_data.line_speed, duplex, flow);
1153 static void bnx2x_set_next_page_sgl(struct bnx2x_fastpath *fp)
1155 int i;
1157 for (i = 1; i <= NUM_RX_SGE_PAGES; i++) {
1158 struct eth_rx_sge *sge;
1160 sge = &fp->rx_sge_ring[RX_SGE_CNT * i - 2];
1161 sge->addr_hi =
1162 cpu_to_le32(U64_HI(fp->rx_sge_mapping +
1163 BCM_PAGE_SIZE*(i % NUM_RX_SGE_PAGES)));
1165 sge->addr_lo =
1166 cpu_to_le32(U64_LO(fp->rx_sge_mapping +
1167 BCM_PAGE_SIZE*(i % NUM_RX_SGE_PAGES)));
1171 static void bnx2x_free_tpa_pool(struct bnx2x *bp,
1172 struct bnx2x_fastpath *fp, int last)
1174 int i;
1176 for (i = 0; i < last; i++) {
1177 struct bnx2x_agg_info *tpa_info = &fp->tpa_info[i];
1178 struct sw_rx_bd *first_buf = &tpa_info->first_buf;
1179 u8 *data = first_buf->data;
1181 if (data == NULL) {
1182 DP(NETIF_MSG_IFDOWN, "tpa bin %d empty on free\n", i);
1183 continue;
1185 if (tpa_info->tpa_state == BNX2X_TPA_START)
1186 dma_unmap_single(&bp->pdev->dev,
1187 dma_unmap_addr(first_buf, mapping),
1188 fp->rx_buf_size, DMA_FROM_DEVICE);
1189 bnx2x_frag_free(fp, data);
1190 first_buf->data = NULL;
1194 void bnx2x_init_rx_rings_cnic(struct bnx2x *bp)
1196 int j;
1198 for_each_rx_queue_cnic(bp, j) {
1199 struct bnx2x_fastpath *fp = &bp->fp[j];
1201 fp->rx_bd_cons = 0;
1203 /* Activate BD ring */
1204 /* Warning!
1205 * this will generate an interrupt (to the TSTORM)
1206 * must only be done after chip is initialized
1208 bnx2x_update_rx_prod(bp, fp, fp->rx_bd_prod, fp->rx_comp_prod,
1209 fp->rx_sge_prod);
1213 void bnx2x_init_rx_rings(struct bnx2x *bp)
1215 int func = BP_FUNC(bp);
1216 u16 ring_prod;
1217 int i, j;
1219 /* Allocate TPA resources */
1220 for_each_eth_queue(bp, j) {
1221 struct bnx2x_fastpath *fp = &bp->fp[j];
1223 DP(NETIF_MSG_IFUP,
1224 "mtu %d rx_buf_size %d\n", bp->dev->mtu, fp->rx_buf_size);
1226 if (!fp->disable_tpa) {
1227 /* Fill the per-aggregtion pool */
1228 for (i = 0; i < MAX_AGG_QS(bp); i++) {
1229 struct bnx2x_agg_info *tpa_info =
1230 &fp->tpa_info[i];
1231 struct sw_rx_bd *first_buf =
1232 &tpa_info->first_buf;
1234 first_buf->data = bnx2x_frag_alloc(fp);
1235 if (!first_buf->data) {
1236 BNX2X_ERR("Failed to allocate TPA skb pool for queue[%d] - disabling TPA on this queue!\n",
1238 bnx2x_free_tpa_pool(bp, fp, i);
1239 fp->disable_tpa = 1;
1240 break;
1242 dma_unmap_addr_set(first_buf, mapping, 0);
1243 tpa_info->tpa_state = BNX2X_TPA_STOP;
1246 /* "next page" elements initialization */
1247 bnx2x_set_next_page_sgl(fp);
1249 /* set SGEs bit mask */
1250 bnx2x_init_sge_ring_bit_mask(fp);
1252 /* Allocate SGEs and initialize the ring elements */
1253 for (i = 0, ring_prod = 0;
1254 i < MAX_RX_SGE_CNT*NUM_RX_SGE_PAGES; i++) {
1256 if (bnx2x_alloc_rx_sge(bp, fp, ring_prod) < 0) {
1257 BNX2X_ERR("was only able to allocate %d rx sges\n",
1259 BNX2X_ERR("disabling TPA for queue[%d]\n",
1261 /* Cleanup already allocated elements */
1262 bnx2x_free_rx_sge_range(bp, fp,
1263 ring_prod);
1264 bnx2x_free_tpa_pool(bp, fp,
1265 MAX_AGG_QS(bp));
1266 fp->disable_tpa = 1;
1267 ring_prod = 0;
1268 break;
1270 ring_prod = NEXT_SGE_IDX(ring_prod);
1273 fp->rx_sge_prod = ring_prod;
1277 for_each_eth_queue(bp, j) {
1278 struct bnx2x_fastpath *fp = &bp->fp[j];
1280 fp->rx_bd_cons = 0;
1282 /* Activate BD ring */
1283 /* Warning!
1284 * this will generate an interrupt (to the TSTORM)
1285 * must only be done after chip is initialized
1287 bnx2x_update_rx_prod(bp, fp, fp->rx_bd_prod, fp->rx_comp_prod,
1288 fp->rx_sge_prod);
1290 if (j != 0)
1291 continue;
1293 if (CHIP_IS_E1(bp)) {
1294 REG_WR(bp, BAR_USTRORM_INTMEM +
1295 USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func),
1296 U64_LO(fp->rx_comp_mapping));
1297 REG_WR(bp, BAR_USTRORM_INTMEM +
1298 USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func) + 4,
1299 U64_HI(fp->rx_comp_mapping));
1304 static void bnx2x_free_tx_skbs_queue(struct bnx2x_fastpath *fp)
1306 u8 cos;
1307 struct bnx2x *bp = fp->bp;
1309 for_each_cos_in_tx_queue(fp, cos) {
1310 struct bnx2x_fp_txdata *txdata = fp->txdata_ptr[cos];
1311 unsigned pkts_compl = 0, bytes_compl = 0;
1313 u16 sw_prod = txdata->tx_pkt_prod;
1314 u16 sw_cons = txdata->tx_pkt_cons;
1316 while (sw_cons != sw_prod) {
1317 bnx2x_free_tx_pkt(bp, txdata, TX_BD(sw_cons),
1318 &pkts_compl, &bytes_compl);
1319 sw_cons++;
1322 netdev_tx_reset_queue(
1323 netdev_get_tx_queue(bp->dev,
1324 txdata->txq_index));
1328 static void bnx2x_free_tx_skbs_cnic(struct bnx2x *bp)
1330 int i;
1332 for_each_tx_queue_cnic(bp, i) {
1333 bnx2x_free_tx_skbs_queue(&bp->fp[i]);
1337 static void bnx2x_free_tx_skbs(struct bnx2x *bp)
1339 int i;
1341 for_each_eth_queue(bp, i) {
1342 bnx2x_free_tx_skbs_queue(&bp->fp[i]);
1346 static void bnx2x_free_rx_bds(struct bnx2x_fastpath *fp)
1348 struct bnx2x *bp = fp->bp;
1349 int i;
1351 /* ring wasn't allocated */
1352 if (fp->rx_buf_ring == NULL)
1353 return;
1355 for (i = 0; i < NUM_RX_BD; i++) {
1356 struct sw_rx_bd *rx_buf = &fp->rx_buf_ring[i];
1357 u8 *data = rx_buf->data;
1359 if (data == NULL)
1360 continue;
1361 dma_unmap_single(&bp->pdev->dev,
1362 dma_unmap_addr(rx_buf, mapping),
1363 fp->rx_buf_size, DMA_FROM_DEVICE);
1365 rx_buf->data = NULL;
1366 bnx2x_frag_free(fp, data);
1370 static void bnx2x_free_rx_skbs_cnic(struct bnx2x *bp)
1372 int j;
1374 for_each_rx_queue_cnic(bp, j) {
1375 bnx2x_free_rx_bds(&bp->fp[j]);
1379 static void bnx2x_free_rx_skbs(struct bnx2x *bp)
1381 int j;
1383 for_each_eth_queue(bp, j) {
1384 struct bnx2x_fastpath *fp = &bp->fp[j];
1386 bnx2x_free_rx_bds(fp);
1388 if (!fp->disable_tpa)
1389 bnx2x_free_tpa_pool(bp, fp, MAX_AGG_QS(bp));
1393 void bnx2x_free_skbs_cnic(struct bnx2x *bp)
1395 bnx2x_free_tx_skbs_cnic(bp);
1396 bnx2x_free_rx_skbs_cnic(bp);
1399 void bnx2x_free_skbs(struct bnx2x *bp)
1401 bnx2x_free_tx_skbs(bp);
1402 bnx2x_free_rx_skbs(bp);
1405 void bnx2x_update_max_mf_config(struct bnx2x *bp, u32 value)
1407 /* load old values */
1408 u32 mf_cfg = bp->mf_config[BP_VN(bp)];
1410 if (value != bnx2x_extract_max_cfg(bp, mf_cfg)) {
1411 /* leave all but MAX value */
1412 mf_cfg &= ~FUNC_MF_CFG_MAX_BW_MASK;
1414 /* set new MAX value */
1415 mf_cfg |= (value << FUNC_MF_CFG_MAX_BW_SHIFT)
1416 & FUNC_MF_CFG_MAX_BW_MASK;
1418 bnx2x_fw_command(bp, DRV_MSG_CODE_SET_MF_BW, mf_cfg);
1423 * bnx2x_free_msix_irqs - free previously requested MSI-X IRQ vectors
1425 * @bp: driver handle
1426 * @nvecs: number of vectors to be released
1428 static void bnx2x_free_msix_irqs(struct bnx2x *bp, int nvecs)
1430 int i, offset = 0;
1432 if (nvecs == offset)
1433 return;
1434 free_irq(bp->msix_table[offset].vector, bp->dev);
1435 DP(NETIF_MSG_IFDOWN, "released sp irq (%d)\n",
1436 bp->msix_table[offset].vector);
1437 offset++;
1439 if (CNIC_SUPPORT(bp)) {
1440 if (nvecs == offset)
1441 return;
1442 offset++;
1445 for_each_eth_queue(bp, i) {
1446 if (nvecs == offset)
1447 return;
1448 DP(NETIF_MSG_IFDOWN, "about to release fp #%d->%d irq\n",
1449 i, bp->msix_table[offset].vector);
1451 free_irq(bp->msix_table[offset++].vector, &bp->fp[i]);
1455 void bnx2x_free_irq(struct bnx2x *bp)
1457 if (bp->flags & USING_MSIX_FLAG &&
1458 !(bp->flags & USING_SINGLE_MSIX_FLAG))
1459 bnx2x_free_msix_irqs(bp, BNX2X_NUM_ETH_QUEUES(bp) +
1460 CNIC_SUPPORT(bp) + 1);
1461 else
1462 free_irq(bp->dev->irq, bp->dev);
1465 int bnx2x_enable_msix(struct bnx2x *bp)
1467 int msix_vec = 0, i, rc, req_cnt;
1469 bp->msix_table[msix_vec].entry = msix_vec;
1470 BNX2X_DEV_INFO("msix_table[0].entry = %d (slowpath)\n",
1471 bp->msix_table[0].entry);
1472 msix_vec++;
1474 /* Cnic requires an msix vector for itself */
1475 if (CNIC_SUPPORT(bp)) {
1476 bp->msix_table[msix_vec].entry = msix_vec;
1477 BNX2X_DEV_INFO("msix_table[%d].entry = %d (CNIC)\n",
1478 msix_vec, bp->msix_table[msix_vec].entry);
1479 msix_vec++;
1482 /* We need separate vectors for ETH queues only (not FCoE) */
1483 for_each_eth_queue(bp, i) {
1484 bp->msix_table[msix_vec].entry = msix_vec;
1485 BNX2X_DEV_INFO("msix_table[%d].entry = %d (fastpath #%u)\n",
1486 msix_vec, msix_vec, i);
1487 msix_vec++;
1490 req_cnt = BNX2X_NUM_ETH_QUEUES(bp) + CNIC_SUPPORT(bp) + 1;
1492 rc = pci_enable_msix(bp->pdev, &bp->msix_table[0], req_cnt);
1495 * reconfigure number of tx/rx queues according to available
1496 * MSI-X vectors
1498 if (rc >= BNX2X_MIN_MSIX_VEC_CNT(bp)) {
1499 /* how less vectors we will have? */
1500 int diff = req_cnt - rc;
1502 BNX2X_DEV_INFO("Trying to use less MSI-X vectors: %d\n", rc);
1504 rc = pci_enable_msix(bp->pdev, &bp->msix_table[0], rc);
1506 if (rc) {
1507 BNX2X_DEV_INFO("MSI-X is not attainable rc %d\n", rc);
1508 goto no_msix;
1511 * decrease number of queues by number of unallocated entries
1513 bp->num_ethernet_queues -= diff;
1514 bp->num_queues = bp->num_ethernet_queues + bp->num_cnic_queues;
1516 BNX2X_DEV_INFO("New queue configuration set: %d\n",
1517 bp->num_queues);
1518 } else if (rc > 0) {
1519 /* Get by with single vector */
1520 rc = pci_enable_msix(bp->pdev, &bp->msix_table[0], 1);
1521 if (rc) {
1522 BNX2X_DEV_INFO("Single MSI-X is not attainable rc %d\n",
1523 rc);
1524 goto no_msix;
1527 BNX2X_DEV_INFO("Using single MSI-X vector\n");
1528 bp->flags |= USING_SINGLE_MSIX_FLAG;
1530 BNX2X_DEV_INFO("set number of queues to 1\n");
1531 bp->num_ethernet_queues = 1;
1532 bp->num_queues = bp->num_ethernet_queues + bp->num_cnic_queues;
1533 } else if (rc < 0) {
1534 BNX2X_DEV_INFO("MSI-X is not attainable rc %d\n", rc);
1535 goto no_msix;
1538 bp->flags |= USING_MSIX_FLAG;
1540 return 0;
1542 no_msix:
1543 /* fall to INTx if not enough memory */
1544 if (rc == -ENOMEM)
1545 bp->flags |= DISABLE_MSI_FLAG;
1547 return rc;
1550 static int bnx2x_req_msix_irqs(struct bnx2x *bp)
1552 int i, rc, offset = 0;
1554 rc = request_irq(bp->msix_table[offset++].vector,
1555 bnx2x_msix_sp_int, 0,
1556 bp->dev->name, bp->dev);
1557 if (rc) {
1558 BNX2X_ERR("request sp irq failed\n");
1559 return -EBUSY;
1562 if (CNIC_SUPPORT(bp))
1563 offset++;
1565 for_each_eth_queue(bp, i) {
1566 struct bnx2x_fastpath *fp = &bp->fp[i];
1567 snprintf(fp->name, sizeof(fp->name), "%s-fp-%d",
1568 bp->dev->name, i);
1570 rc = request_irq(bp->msix_table[offset].vector,
1571 bnx2x_msix_fp_int, 0, fp->name, fp);
1572 if (rc) {
1573 BNX2X_ERR("request fp #%d irq (%d) failed rc %d\n", i,
1574 bp->msix_table[offset].vector, rc);
1575 bnx2x_free_msix_irqs(bp, offset);
1576 return -EBUSY;
1579 offset++;
1582 i = BNX2X_NUM_ETH_QUEUES(bp);
1583 offset = 1 + CNIC_SUPPORT(bp);
1584 netdev_info(bp->dev, "using MSI-X IRQs: sp %d fp[%d] %d ... fp[%d] %d\n",
1585 bp->msix_table[0].vector,
1586 0, bp->msix_table[offset].vector,
1587 i - 1, bp->msix_table[offset + i - 1].vector);
1589 return 0;
1592 int bnx2x_enable_msi(struct bnx2x *bp)
1594 int rc;
1596 rc = pci_enable_msi(bp->pdev);
1597 if (rc) {
1598 BNX2X_DEV_INFO("MSI is not attainable\n");
1599 return -1;
1601 bp->flags |= USING_MSI_FLAG;
1603 return 0;
1606 static int bnx2x_req_irq(struct bnx2x *bp)
1608 unsigned long flags;
1609 unsigned int irq;
1611 if (bp->flags & (USING_MSI_FLAG | USING_MSIX_FLAG))
1612 flags = 0;
1613 else
1614 flags = IRQF_SHARED;
1616 if (bp->flags & USING_MSIX_FLAG)
1617 irq = bp->msix_table[0].vector;
1618 else
1619 irq = bp->pdev->irq;
1621 return request_irq(irq, bnx2x_interrupt, flags, bp->dev->name, bp->dev);
1624 static int bnx2x_setup_irqs(struct bnx2x *bp)
1626 int rc = 0;
1627 if (bp->flags & USING_MSIX_FLAG &&
1628 !(bp->flags & USING_SINGLE_MSIX_FLAG)) {
1629 rc = bnx2x_req_msix_irqs(bp);
1630 if (rc)
1631 return rc;
1632 } else {
1633 bnx2x_ack_int(bp);
1634 rc = bnx2x_req_irq(bp);
1635 if (rc) {
1636 BNX2X_ERR("IRQ request failed rc %d, aborting\n", rc);
1637 return rc;
1639 if (bp->flags & USING_MSI_FLAG) {
1640 bp->dev->irq = bp->pdev->irq;
1641 netdev_info(bp->dev, "using MSI IRQ %d\n",
1642 bp->dev->irq);
1644 if (bp->flags & USING_MSIX_FLAG) {
1645 bp->dev->irq = bp->msix_table[0].vector;
1646 netdev_info(bp->dev, "using MSIX IRQ %d\n",
1647 bp->dev->irq);
1651 return 0;
1654 static void bnx2x_napi_enable_cnic(struct bnx2x *bp)
1656 int i;
1658 for_each_rx_queue_cnic(bp, i)
1659 napi_enable(&bnx2x_fp(bp, i, napi));
1662 static void bnx2x_napi_enable(struct bnx2x *bp)
1664 int i;
1666 for_each_eth_queue(bp, i)
1667 napi_enable(&bnx2x_fp(bp, i, napi));
1670 static void bnx2x_napi_disable_cnic(struct bnx2x *bp)
1672 int i;
1674 for_each_rx_queue_cnic(bp, i)
1675 napi_disable(&bnx2x_fp(bp, i, napi));
1678 static void bnx2x_napi_disable(struct bnx2x *bp)
1680 int i;
1682 for_each_eth_queue(bp, i)
1683 napi_disable(&bnx2x_fp(bp, i, napi));
1686 void bnx2x_netif_start(struct bnx2x *bp)
1688 if (netif_running(bp->dev)) {
1689 bnx2x_napi_enable(bp);
1690 if (CNIC_LOADED(bp))
1691 bnx2x_napi_enable_cnic(bp);
1692 bnx2x_int_enable(bp);
1693 if (bp->state == BNX2X_STATE_OPEN)
1694 netif_tx_wake_all_queues(bp->dev);
1698 void bnx2x_netif_stop(struct bnx2x *bp, int disable_hw)
1700 bnx2x_int_disable_sync(bp, disable_hw);
1701 bnx2x_napi_disable(bp);
1702 if (CNIC_LOADED(bp))
1703 bnx2x_napi_disable_cnic(bp);
1706 u16 bnx2x_select_queue(struct net_device *dev, struct sk_buff *skb)
1708 struct bnx2x *bp = netdev_priv(dev);
1710 if (CNIC_LOADED(bp) && !NO_FCOE(bp)) {
1711 struct ethhdr *hdr = (struct ethhdr *)skb->data;
1712 u16 ether_type = ntohs(hdr->h_proto);
1714 /* Skip VLAN tag if present */
1715 if (ether_type == ETH_P_8021Q) {
1716 struct vlan_ethhdr *vhdr =
1717 (struct vlan_ethhdr *)skb->data;
1719 ether_type = ntohs(vhdr->h_vlan_encapsulated_proto);
1722 /* If ethertype is FCoE or FIP - use FCoE ring */
1723 if ((ether_type == ETH_P_FCOE) || (ether_type == ETH_P_FIP))
1724 return bnx2x_fcoe_tx(bp, txq_index);
1727 /* select a non-FCoE queue */
1728 return __skb_tx_hash(dev, skb, BNX2X_NUM_ETH_QUEUES(bp));
1732 void bnx2x_set_num_queues(struct bnx2x *bp)
1734 /* RSS queues */
1735 bp->num_ethernet_queues = bnx2x_calc_num_queues(bp);
1737 /* override in STORAGE SD modes */
1738 if (IS_MF_STORAGE_SD(bp) || IS_MF_FCOE_AFEX(bp))
1739 bp->num_ethernet_queues = 1;
1741 /* Add special queues */
1742 bp->num_cnic_queues = CNIC_SUPPORT(bp); /* For FCOE */
1743 bp->num_queues = bp->num_ethernet_queues + bp->num_cnic_queues;
1745 BNX2X_DEV_INFO("set number of queues to %d\n", bp->num_queues);
1749 * bnx2x_set_real_num_queues - configure netdev->real_num_[tx,rx]_queues
1751 * @bp: Driver handle
1753 * We currently support for at most 16 Tx queues for each CoS thus we will
1754 * allocate a multiple of 16 for ETH L2 rings according to the value of the
1755 * bp->max_cos.
1757 * If there is an FCoE L2 queue the appropriate Tx queue will have the next
1758 * index after all ETH L2 indices.
1760 * If the actual number of Tx queues (for each CoS) is less than 16 then there
1761 * will be the holes at the end of each group of 16 ETh L2 indices (0..15,
1762 * 16..31,...) with indicies that are not coupled with any real Tx queue.
1764 * The proper configuration of skb->queue_mapping is handled by
1765 * bnx2x_select_queue() and __skb_tx_hash().
1767 * bnx2x_setup_tc() takes care of the proper TC mappings so that __skb_tx_hash()
1768 * will return a proper Tx index if TC is enabled (netdev->num_tc > 0).
1770 static int bnx2x_set_real_num_queues(struct bnx2x *bp, int include_cnic)
1772 int rc, tx, rx;
1774 tx = BNX2X_NUM_ETH_QUEUES(bp) * bp->max_cos;
1775 rx = BNX2X_NUM_ETH_QUEUES(bp);
1777 /* account for fcoe queue */
1778 if (include_cnic && !NO_FCOE(bp)) {
1779 rx++;
1780 tx++;
1783 rc = netif_set_real_num_tx_queues(bp->dev, tx);
1784 if (rc) {
1785 BNX2X_ERR("Failed to set real number of Tx queues: %d\n", rc);
1786 return rc;
1788 rc = netif_set_real_num_rx_queues(bp->dev, rx);
1789 if (rc) {
1790 BNX2X_ERR("Failed to set real number of Rx queues: %d\n", rc);
1791 return rc;
1794 DP(NETIF_MSG_IFUP, "Setting real num queues to (tx, rx) (%d, %d)\n",
1795 tx, rx);
1797 return rc;
1800 static void bnx2x_set_rx_buf_size(struct bnx2x *bp)
1802 int i;
1804 for_each_queue(bp, i) {
1805 struct bnx2x_fastpath *fp = &bp->fp[i];
1806 u32 mtu;
1808 /* Always use a mini-jumbo MTU for the FCoE L2 ring */
1809 if (IS_FCOE_IDX(i))
1811 * Although there are no IP frames expected to arrive to
1812 * this ring we still want to add an
1813 * IP_HEADER_ALIGNMENT_PADDING to prevent a buffer
1814 * overrun attack.
1816 mtu = BNX2X_FCOE_MINI_JUMBO_MTU;
1817 else
1818 mtu = bp->dev->mtu;
1819 fp->rx_buf_size = BNX2X_FW_RX_ALIGN_START +
1820 IP_HEADER_ALIGNMENT_PADDING +
1821 ETH_OVREHEAD +
1822 mtu +
1823 BNX2X_FW_RX_ALIGN_END;
1824 /* Note : rx_buf_size doesnt take into account NET_SKB_PAD */
1825 if (fp->rx_buf_size + NET_SKB_PAD <= PAGE_SIZE)
1826 fp->rx_frag_size = fp->rx_buf_size + NET_SKB_PAD;
1827 else
1828 fp->rx_frag_size = 0;
1832 static int bnx2x_init_rss_pf(struct bnx2x *bp)
1834 int i;
1835 u8 num_eth_queues = BNX2X_NUM_ETH_QUEUES(bp);
1837 /* Prepare the initial contents fo the indirection table if RSS is
1838 * enabled
1840 for (i = 0; i < sizeof(bp->rss_conf_obj.ind_table); i++)
1841 bp->rss_conf_obj.ind_table[i] =
1842 bp->fp->cl_id +
1843 ethtool_rxfh_indir_default(i, num_eth_queues);
1846 * For 57710 and 57711 SEARCHER configuration (rss_keys) is
1847 * per-port, so if explicit configuration is needed , do it only
1848 * for a PMF.
1850 * For 57712 and newer on the other hand it's a per-function
1851 * configuration.
1853 return bnx2x_config_rss_eth(bp, bp->port.pmf || !CHIP_IS_E1x(bp));
1856 int bnx2x_config_rss_pf(struct bnx2x *bp, struct bnx2x_rss_config_obj *rss_obj,
1857 bool config_hash)
1859 struct bnx2x_config_rss_params params = {NULL};
1861 /* Although RSS is meaningless when there is a single HW queue we
1862 * still need it enabled in order to have HW Rx hash generated.
1864 * if (!is_eth_multi(bp))
1865 * bp->multi_mode = ETH_RSS_MODE_DISABLED;
1868 params.rss_obj = rss_obj;
1870 __set_bit(RAMROD_COMP_WAIT, &params.ramrod_flags);
1872 __set_bit(BNX2X_RSS_MODE_REGULAR, &params.rss_flags);
1874 /* RSS configuration */
1875 __set_bit(BNX2X_RSS_IPV4, &params.rss_flags);
1876 __set_bit(BNX2X_RSS_IPV4_TCP, &params.rss_flags);
1877 __set_bit(BNX2X_RSS_IPV6, &params.rss_flags);
1878 __set_bit(BNX2X_RSS_IPV6_TCP, &params.rss_flags);
1879 if (rss_obj->udp_rss_v4)
1880 __set_bit(BNX2X_RSS_IPV4_UDP, &params.rss_flags);
1881 if (rss_obj->udp_rss_v6)
1882 __set_bit(BNX2X_RSS_IPV6_UDP, &params.rss_flags);
1884 /* Hash bits */
1885 params.rss_result_mask = MULTI_MASK;
1887 memcpy(params.ind_table, rss_obj->ind_table, sizeof(params.ind_table));
1889 if (config_hash) {
1890 /* RSS keys */
1891 prandom_bytes(params.rss_key, sizeof(params.rss_key));
1892 __set_bit(BNX2X_RSS_SET_SRCH, &params.rss_flags);
1895 return bnx2x_config_rss(bp, &params);
1898 static int bnx2x_init_hw(struct bnx2x *bp, u32 load_code)
1900 struct bnx2x_func_state_params func_params = {NULL};
1902 /* Prepare parameters for function state transitions */
1903 __set_bit(RAMROD_COMP_WAIT, &func_params.ramrod_flags);
1905 func_params.f_obj = &bp->func_obj;
1906 func_params.cmd = BNX2X_F_CMD_HW_INIT;
1908 func_params.params.hw_init.load_phase = load_code;
1910 return bnx2x_func_state_change(bp, &func_params);
1914 * Cleans the object that have internal lists without sending
1915 * ramrods. Should be run when interrutps are disabled.
1917 static void bnx2x_squeeze_objects(struct bnx2x *bp)
1919 int rc;
1920 unsigned long ramrod_flags = 0, vlan_mac_flags = 0;
1921 struct bnx2x_mcast_ramrod_params rparam = {NULL};
1922 struct bnx2x_vlan_mac_obj *mac_obj = &bp->sp_objs->mac_obj;
1924 /***************** Cleanup MACs' object first *************************/
1926 /* Wait for completion of requested */
1927 __set_bit(RAMROD_COMP_WAIT, &ramrod_flags);
1928 /* Perform a dry cleanup */
1929 __set_bit(RAMROD_DRV_CLR_ONLY, &ramrod_flags);
1931 /* Clean ETH primary MAC */
1932 __set_bit(BNX2X_ETH_MAC, &vlan_mac_flags);
1933 rc = mac_obj->delete_all(bp, &bp->sp_objs->mac_obj, &vlan_mac_flags,
1934 &ramrod_flags);
1935 if (rc != 0)
1936 BNX2X_ERR("Failed to clean ETH MACs: %d\n", rc);
1938 /* Cleanup UC list */
1939 vlan_mac_flags = 0;
1940 __set_bit(BNX2X_UC_LIST_MAC, &vlan_mac_flags);
1941 rc = mac_obj->delete_all(bp, mac_obj, &vlan_mac_flags,
1942 &ramrod_flags);
1943 if (rc != 0)
1944 BNX2X_ERR("Failed to clean UC list MACs: %d\n", rc);
1946 /***************** Now clean mcast object *****************************/
1947 rparam.mcast_obj = &bp->mcast_obj;
1948 __set_bit(RAMROD_DRV_CLR_ONLY, &rparam.ramrod_flags);
1950 /* Add a DEL command... */
1951 rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_DEL);
1952 if (rc < 0)
1953 BNX2X_ERR("Failed to add a new DEL command to a multi-cast object: %d\n",
1954 rc);
1956 /* ...and wait until all pending commands are cleared */
1957 rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_CONT);
1958 while (rc != 0) {
1959 if (rc < 0) {
1960 BNX2X_ERR("Failed to clean multi-cast object: %d\n",
1961 rc);
1962 return;
1965 rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_CONT);
1969 #ifndef BNX2X_STOP_ON_ERROR
1970 #define LOAD_ERROR_EXIT(bp, label) \
1971 do { \
1972 (bp)->state = BNX2X_STATE_ERROR; \
1973 goto label; \
1974 } while (0)
1976 #define LOAD_ERROR_EXIT_CNIC(bp, label) \
1977 do { \
1978 bp->cnic_loaded = false; \
1979 goto label; \
1980 } while (0)
1981 #else /*BNX2X_STOP_ON_ERROR*/
1982 #define LOAD_ERROR_EXIT(bp, label) \
1983 do { \
1984 (bp)->state = BNX2X_STATE_ERROR; \
1985 (bp)->panic = 1; \
1986 return -EBUSY; \
1987 } while (0)
1988 #define LOAD_ERROR_EXIT_CNIC(bp, label) \
1989 do { \
1990 bp->cnic_loaded = false; \
1991 (bp)->panic = 1; \
1992 return -EBUSY; \
1993 } while (0)
1994 #endif /*BNX2X_STOP_ON_ERROR*/
1996 bool bnx2x_test_firmware_version(struct bnx2x *bp, bool is_err)
1998 /* build FW version dword */
1999 u32 my_fw = (BCM_5710_FW_MAJOR_VERSION) +
2000 (BCM_5710_FW_MINOR_VERSION << 8) +
2001 (BCM_5710_FW_REVISION_VERSION << 16) +
2002 (BCM_5710_FW_ENGINEERING_VERSION << 24);
2004 /* read loaded FW from chip */
2005 u32 loaded_fw = REG_RD(bp, XSEM_REG_PRAM);
2007 DP(NETIF_MSG_IFUP, "loaded fw %x, my fw %x\n", loaded_fw, my_fw);
2009 if (loaded_fw != my_fw) {
2010 if (is_err)
2011 BNX2X_ERR("bnx2x with FW %x was already loaded, which mismatches my %x FW. aborting\n",
2012 loaded_fw, my_fw);
2013 return false;
2016 return true;
2020 * bnx2x_bz_fp - zero content of the fastpath structure.
2022 * @bp: driver handle
2023 * @index: fastpath index to be zeroed
2025 * Makes sure the contents of the bp->fp[index].napi is kept
2026 * intact.
2028 static void bnx2x_bz_fp(struct bnx2x *bp, int index)
2030 struct bnx2x_fastpath *fp = &bp->fp[index];
2031 struct bnx2x_fp_stats *fp_stats = &bp->fp_stats[index];
2033 int cos;
2034 struct napi_struct orig_napi = fp->napi;
2035 struct bnx2x_agg_info *orig_tpa_info = fp->tpa_info;
2036 /* bzero bnx2x_fastpath contents */
2037 if (bp->stats_init) {
2038 memset(fp->tpa_info, 0, sizeof(*fp->tpa_info));
2039 memset(fp, 0, sizeof(*fp));
2040 } else {
2041 /* Keep Queue statistics */
2042 struct bnx2x_eth_q_stats *tmp_eth_q_stats;
2043 struct bnx2x_eth_q_stats_old *tmp_eth_q_stats_old;
2045 tmp_eth_q_stats = kzalloc(sizeof(struct bnx2x_eth_q_stats),
2046 GFP_KERNEL);
2047 if (tmp_eth_q_stats)
2048 memcpy(tmp_eth_q_stats, &fp_stats->eth_q_stats,
2049 sizeof(struct bnx2x_eth_q_stats));
2051 tmp_eth_q_stats_old =
2052 kzalloc(sizeof(struct bnx2x_eth_q_stats_old),
2053 GFP_KERNEL);
2054 if (tmp_eth_q_stats_old)
2055 memcpy(tmp_eth_q_stats_old, &fp_stats->eth_q_stats_old,
2056 sizeof(struct bnx2x_eth_q_stats_old));
2058 memset(fp->tpa_info, 0, sizeof(*fp->tpa_info));
2059 memset(fp, 0, sizeof(*fp));
2061 if (tmp_eth_q_stats) {
2062 memcpy(&fp_stats->eth_q_stats, tmp_eth_q_stats,
2063 sizeof(struct bnx2x_eth_q_stats));
2064 kfree(tmp_eth_q_stats);
2067 if (tmp_eth_q_stats_old) {
2068 memcpy(&fp_stats->eth_q_stats_old, tmp_eth_q_stats_old,
2069 sizeof(struct bnx2x_eth_q_stats_old));
2070 kfree(tmp_eth_q_stats_old);
2075 /* Restore the NAPI object as it has been already initialized */
2076 fp->napi = orig_napi;
2077 fp->tpa_info = orig_tpa_info;
2078 fp->bp = bp;
2079 fp->index = index;
2080 if (IS_ETH_FP(fp))
2081 fp->max_cos = bp->max_cos;
2082 else
2083 /* Special queues support only one CoS */
2084 fp->max_cos = 1;
2086 /* Init txdata pointers */
2087 if (IS_FCOE_FP(fp))
2088 fp->txdata_ptr[0] = &bp->bnx2x_txq[FCOE_TXQ_IDX(bp)];
2089 if (IS_ETH_FP(fp))
2090 for_each_cos_in_tx_queue(fp, cos)
2091 fp->txdata_ptr[cos] = &bp->bnx2x_txq[cos *
2092 BNX2X_NUM_ETH_QUEUES(bp) + index];
2095 * set the tpa flag for each queue. The tpa flag determines the queue
2096 * minimal size so it must be set prior to queue memory allocation
2098 fp->disable_tpa = !(bp->flags & TPA_ENABLE_FLAG ||
2099 (bp->flags & GRO_ENABLE_FLAG &&
2100 bnx2x_mtu_allows_gro(bp->dev->mtu)));
2101 if (bp->flags & TPA_ENABLE_FLAG)
2102 fp->mode = TPA_MODE_LRO;
2103 else if (bp->flags & GRO_ENABLE_FLAG)
2104 fp->mode = TPA_MODE_GRO;
2106 /* We don't want TPA on an FCoE L2 ring */
2107 if (IS_FCOE_FP(fp))
2108 fp->disable_tpa = 1;
2111 int bnx2x_load_cnic(struct bnx2x *bp)
2113 int i, rc, port = BP_PORT(bp);
2115 DP(NETIF_MSG_IFUP, "Starting CNIC-related load\n");
2117 mutex_init(&bp->cnic_mutex);
2119 rc = bnx2x_alloc_mem_cnic(bp);
2120 if (rc) {
2121 BNX2X_ERR("Unable to allocate bp memory for cnic\n");
2122 LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic0);
2125 rc = bnx2x_alloc_fp_mem_cnic(bp);
2126 if (rc) {
2127 BNX2X_ERR("Unable to allocate memory for cnic fps\n");
2128 LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic0);
2131 /* Update the number of queues with the cnic queues */
2132 rc = bnx2x_set_real_num_queues(bp, 1);
2133 if (rc) {
2134 BNX2X_ERR("Unable to set real_num_queues including cnic\n");
2135 LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic0);
2138 /* Add all CNIC NAPI objects */
2139 bnx2x_add_all_napi_cnic(bp);
2140 DP(NETIF_MSG_IFUP, "cnic napi added\n");
2141 bnx2x_napi_enable_cnic(bp);
2143 rc = bnx2x_init_hw_func_cnic(bp);
2144 if (rc)
2145 LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic1);
2147 bnx2x_nic_init_cnic(bp);
2149 /* Enable Timer scan */
2150 REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + port*4, 1);
2152 for_each_cnic_queue(bp, i) {
2153 rc = bnx2x_setup_queue(bp, &bp->fp[i], 0);
2154 if (rc) {
2155 BNX2X_ERR("Queue setup failed\n");
2156 LOAD_ERROR_EXIT(bp, load_error_cnic2);
2160 /* Initialize Rx filter. */
2161 netif_addr_lock_bh(bp->dev);
2162 bnx2x_set_rx_mode(bp->dev);
2163 netif_addr_unlock_bh(bp->dev);
2165 /* re-read iscsi info */
2166 bnx2x_get_iscsi_info(bp);
2167 bnx2x_setup_cnic_irq_info(bp);
2168 bnx2x_setup_cnic_info(bp);
2169 bp->cnic_loaded = true;
2170 if (bp->state == BNX2X_STATE_OPEN)
2171 bnx2x_cnic_notify(bp, CNIC_CTL_START_CMD);
2174 DP(NETIF_MSG_IFUP, "Ending successfully CNIC-related load\n");
2176 return 0;
2178 #ifndef BNX2X_STOP_ON_ERROR
2179 load_error_cnic2:
2180 /* Disable Timer scan */
2181 REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + port*4, 0);
2183 load_error_cnic1:
2184 bnx2x_napi_disable_cnic(bp);
2185 /* Update the number of queues without the cnic queues */
2186 rc = bnx2x_set_real_num_queues(bp, 0);
2187 if (rc)
2188 BNX2X_ERR("Unable to set real_num_queues not including cnic\n");
2189 load_error_cnic0:
2190 BNX2X_ERR("CNIC-related load failed\n");
2191 bnx2x_free_fp_mem_cnic(bp);
2192 bnx2x_free_mem_cnic(bp);
2193 return rc;
2194 #endif /* ! BNX2X_STOP_ON_ERROR */
2198 /* must be called with rtnl_lock */
2199 int bnx2x_nic_load(struct bnx2x *bp, int load_mode)
2201 int port = BP_PORT(bp);
2202 u32 load_code;
2203 int i, rc;
2205 DP(NETIF_MSG_IFUP, "Starting NIC load\n");
2206 DP(NETIF_MSG_IFUP,
2207 "CNIC is %s\n", CNIC_ENABLED(bp) ? "enabled" : "disabled");
2209 #ifdef BNX2X_STOP_ON_ERROR
2210 if (unlikely(bp->panic)) {
2211 BNX2X_ERR("Can't load NIC when there is panic\n");
2212 return -EPERM;
2214 #endif
2216 bp->state = BNX2X_STATE_OPENING_WAIT4_LOAD;
2218 /* Set the initial link reported state to link down */
2219 bnx2x_acquire_phy_lock(bp);
2220 memset(&bp->last_reported_link, 0, sizeof(bp->last_reported_link));
2221 __set_bit(BNX2X_LINK_REPORT_LINK_DOWN,
2222 &bp->last_reported_link.link_report_flags);
2223 bnx2x_release_phy_lock(bp);
2225 /* must be called before memory allocation and HW init */
2226 bnx2x_ilt_set_info(bp);
2229 * Zero fastpath structures preserving invariants like napi, which are
2230 * allocated only once, fp index, max_cos, bp pointer.
2231 * Also set fp->disable_tpa and txdata_ptr.
2233 DP(NETIF_MSG_IFUP, "num queues: %d", bp->num_queues);
2234 for_each_queue(bp, i)
2235 bnx2x_bz_fp(bp, i);
2236 memset(bp->bnx2x_txq, 0, (BNX2X_MAX_RSS_COUNT(bp) * BNX2X_MULTI_TX_COS +
2237 bp->num_cnic_queues) *
2238 sizeof(struct bnx2x_fp_txdata));
2240 bp->fcoe_init = false;
2242 /* Set the receive queues buffer size */
2243 bnx2x_set_rx_buf_size(bp);
2245 if (bnx2x_alloc_mem(bp))
2246 return -ENOMEM;
2248 /* As long as bnx2x_alloc_mem() may possibly update
2249 * bp->num_queues, bnx2x_set_real_num_queues() should always
2250 * come after it. At this stage cnic queues are not counted.
2252 rc = bnx2x_set_real_num_queues(bp, 0);
2253 if (rc) {
2254 BNX2X_ERR("Unable to set real_num_queues\n");
2255 LOAD_ERROR_EXIT(bp, load_error0);
2258 /* configure multi cos mappings in kernel.
2259 * this configuration may be overriden by a multi class queue discipline
2260 * or by a dcbx negotiation result.
2262 bnx2x_setup_tc(bp->dev, bp->max_cos);
2264 /* Add all NAPI objects */
2265 bnx2x_add_all_napi(bp);
2266 DP(NETIF_MSG_IFUP, "napi added\n");
2267 bnx2x_napi_enable(bp);
2269 /* set pf load just before approaching the MCP */
2270 bnx2x_set_pf_load(bp);
2272 /* Send LOAD_REQUEST command to MCP
2273 * Returns the type of LOAD command:
2274 * if it is the first port to be initialized
2275 * common blocks should be initialized, otherwise - not
2277 if (!BP_NOMCP(bp)) {
2278 /* init fw_seq */
2279 bp->fw_seq =
2280 (SHMEM_RD(bp, func_mb[BP_FW_MB_IDX(bp)].drv_mb_header) &
2281 DRV_MSG_SEQ_NUMBER_MASK);
2282 BNX2X_DEV_INFO("fw_seq 0x%08x\n", bp->fw_seq);
2284 /* Get current FW pulse sequence */
2285 bp->fw_drv_pulse_wr_seq =
2286 (SHMEM_RD(bp, func_mb[BP_FW_MB_IDX(bp)].drv_pulse_mb) &
2287 DRV_PULSE_SEQ_MASK);
2288 BNX2X_DEV_INFO("drv_pulse 0x%x\n", bp->fw_drv_pulse_wr_seq);
2290 load_code = bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_REQ,
2291 DRV_MSG_CODE_LOAD_REQ_WITH_LFA);
2292 if (!load_code) {
2293 BNX2X_ERR("MCP response failure, aborting\n");
2294 rc = -EBUSY;
2295 LOAD_ERROR_EXIT(bp, load_error1);
2297 if (load_code == FW_MSG_CODE_DRV_LOAD_REFUSED) {
2298 BNX2X_ERR("Driver load refused\n");
2299 rc = -EBUSY; /* other port in diagnostic mode */
2300 LOAD_ERROR_EXIT(bp, load_error1);
2302 if (load_code != FW_MSG_CODE_DRV_LOAD_COMMON_CHIP &&
2303 load_code != FW_MSG_CODE_DRV_LOAD_COMMON) {
2304 /* abort nic load if version mismatch */
2305 if (!bnx2x_test_firmware_version(bp, true)) {
2306 rc = -EBUSY;
2307 LOAD_ERROR_EXIT(bp, load_error2);
2311 } else {
2312 int path = BP_PATH(bp);
2314 DP(NETIF_MSG_IFUP, "NO MCP - load counts[%d] %d, %d, %d\n",
2315 path, load_count[path][0], load_count[path][1],
2316 load_count[path][2]);
2317 load_count[path][0]++;
2318 load_count[path][1 + port]++;
2319 DP(NETIF_MSG_IFUP, "NO MCP - new load counts[%d] %d, %d, %d\n",
2320 path, load_count[path][0], load_count[path][1],
2321 load_count[path][2]);
2322 if (load_count[path][0] == 1)
2323 load_code = FW_MSG_CODE_DRV_LOAD_COMMON;
2324 else if (load_count[path][1 + port] == 1)
2325 load_code = FW_MSG_CODE_DRV_LOAD_PORT;
2326 else
2327 load_code = FW_MSG_CODE_DRV_LOAD_FUNCTION;
2330 if ((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
2331 (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP) ||
2332 (load_code == FW_MSG_CODE_DRV_LOAD_PORT)) {
2333 bp->port.pmf = 1;
2335 * We need the barrier to ensure the ordering between the
2336 * writing to bp->port.pmf here and reading it from the
2337 * bnx2x_periodic_task().
2339 smp_mb();
2340 } else
2341 bp->port.pmf = 0;
2343 DP(NETIF_MSG_IFUP, "pmf %d\n", bp->port.pmf);
2345 /* Init Function state controlling object */
2346 bnx2x__init_func_obj(bp);
2348 /* Initialize HW */
2349 rc = bnx2x_init_hw(bp, load_code);
2350 if (rc) {
2351 BNX2X_ERR("HW init failed, aborting\n");
2352 bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2353 LOAD_ERROR_EXIT(bp, load_error2);
2356 /* Connect to IRQs */
2357 rc = bnx2x_setup_irqs(bp);
2358 if (rc) {
2359 BNX2X_ERR("IRQs setup failed\n");
2360 bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2361 LOAD_ERROR_EXIT(bp, load_error2);
2364 /* Setup NIC internals and enable interrupts */
2365 bnx2x_nic_init(bp, load_code);
2367 /* Init per-function objects */
2368 bnx2x_init_bp_objs(bp);
2370 if (((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
2371 (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP)) &&
2372 (bp->common.shmem2_base)) {
2373 if (SHMEM2_HAS(bp, dcc_support))
2374 SHMEM2_WR(bp, dcc_support,
2375 (SHMEM_DCC_SUPPORT_DISABLE_ENABLE_PF_TLV |
2376 SHMEM_DCC_SUPPORT_BANDWIDTH_ALLOCATION_TLV));
2377 if (SHMEM2_HAS(bp, afex_driver_support))
2378 SHMEM2_WR(bp, afex_driver_support,
2379 SHMEM_AFEX_SUPPORTED_VERSION_ONE);
2382 /* Set AFEX default VLAN tag to an invalid value */
2383 bp->afex_def_vlan_tag = -1;
2385 bp->state = BNX2X_STATE_OPENING_WAIT4_PORT;
2386 rc = bnx2x_func_start(bp);
2387 if (rc) {
2388 BNX2X_ERR("Function start failed!\n");
2389 bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2390 LOAD_ERROR_EXIT(bp, load_error3);
2393 /* Send LOAD_DONE command to MCP */
2394 if (!BP_NOMCP(bp)) {
2395 load_code = bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2396 if (!load_code) {
2397 BNX2X_ERR("MCP response failure, aborting\n");
2398 rc = -EBUSY;
2399 LOAD_ERROR_EXIT(bp, load_error3);
2403 rc = bnx2x_setup_leading(bp);
2404 if (rc) {
2405 BNX2X_ERR("Setup leading failed!\n");
2406 LOAD_ERROR_EXIT(bp, load_error3);
2409 for_each_nondefault_eth_queue(bp, i) {
2410 rc = bnx2x_setup_queue(bp, &bp->fp[i], 0);
2411 if (rc) {
2412 BNX2X_ERR("Queue setup failed\n");
2413 LOAD_ERROR_EXIT(bp, load_error3);
2417 rc = bnx2x_init_rss_pf(bp);
2418 if (rc) {
2419 BNX2X_ERR("PF RSS init failed\n");
2420 LOAD_ERROR_EXIT(bp, load_error3);
2423 /* Now when Clients are configured we are ready to work */
2424 bp->state = BNX2X_STATE_OPEN;
2426 /* Configure a ucast MAC */
2427 rc = bnx2x_set_eth_mac(bp, true);
2428 if (rc) {
2429 BNX2X_ERR("Setting Ethernet MAC failed\n");
2430 LOAD_ERROR_EXIT(bp, load_error3);
2433 if (bp->pending_max) {
2434 bnx2x_update_max_mf_config(bp, bp->pending_max);
2435 bp->pending_max = 0;
2438 if (bp->port.pmf)
2439 bnx2x_initial_phy_init(bp, load_mode);
2440 bp->link_params.feature_config_flags &= ~FEATURE_CONFIG_BOOT_FROM_SAN;
2442 /* Start fast path */
2444 /* Initialize Rx filter. */
2445 netif_addr_lock_bh(bp->dev);
2446 bnx2x_set_rx_mode(bp->dev);
2447 netif_addr_unlock_bh(bp->dev);
2449 /* Start the Tx */
2450 switch (load_mode) {
2451 case LOAD_NORMAL:
2452 /* Tx queue should be only reenabled */
2453 netif_tx_wake_all_queues(bp->dev);
2454 break;
2456 case LOAD_OPEN:
2457 netif_tx_start_all_queues(bp->dev);
2458 smp_mb__after_clear_bit();
2459 break;
2461 case LOAD_DIAG:
2462 case LOAD_LOOPBACK_EXT:
2463 bp->state = BNX2X_STATE_DIAG;
2464 break;
2466 default:
2467 break;
2470 if (bp->port.pmf)
2471 bnx2x_update_drv_flags(bp, 1 << DRV_FLAGS_PORT_MASK, 0);
2472 else
2473 bnx2x__link_status_update(bp);
2475 /* start the timer */
2476 mod_timer(&bp->timer, jiffies + bp->current_interval);
2478 if (CNIC_ENABLED(bp))
2479 bnx2x_load_cnic(bp);
2481 /* mark driver is loaded in shmem2 */
2482 if (SHMEM2_HAS(bp, drv_capabilities_flag)) {
2483 u32 val;
2484 val = SHMEM2_RD(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)]);
2485 SHMEM2_WR(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)],
2486 val | DRV_FLAGS_CAPABILITIES_LOADED_SUPPORTED |
2487 DRV_FLAGS_CAPABILITIES_LOADED_L2);
2490 /* Wait for all pending SP commands to complete */
2491 if (!bnx2x_wait_sp_comp(bp, ~0x0UL)) {
2492 BNX2X_ERR("Timeout waiting for SP elements to complete\n");
2493 bnx2x_nic_unload(bp, UNLOAD_CLOSE, false);
2494 return -EBUSY;
2497 /* If PMF - send ADMIN DCBX msg to MFW to initiate DCBX FSM */
2498 if (bp->port.pmf && (bp->state != BNX2X_STATE_DIAG))
2499 bnx2x_dcbx_init(bp, false);
2501 DP(NETIF_MSG_IFUP, "Ending successfully NIC load\n");
2503 return 0;
2505 #ifndef BNX2X_STOP_ON_ERROR
2506 load_error3:
2507 bnx2x_int_disable_sync(bp, 1);
2509 /* Clean queueable objects */
2510 bnx2x_squeeze_objects(bp);
2512 /* Free SKBs, SGEs, TPA pool and driver internals */
2513 bnx2x_free_skbs(bp);
2514 for_each_rx_queue(bp, i)
2515 bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
2517 /* Release IRQs */
2518 bnx2x_free_irq(bp);
2519 load_error2:
2520 if (!BP_NOMCP(bp)) {
2521 bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_REQ_WOL_MCP, 0);
2522 bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_DONE, 0);
2525 bp->port.pmf = 0;
2526 load_error1:
2527 bnx2x_napi_disable(bp);
2528 /* clear pf_load status, as it was already set */
2529 bnx2x_clear_pf_load(bp);
2530 load_error0:
2531 bnx2x_free_mem(bp);
2533 return rc;
2534 #endif /* ! BNX2X_STOP_ON_ERROR */
2537 /* must be called with rtnl_lock */
2538 int bnx2x_nic_unload(struct bnx2x *bp, int unload_mode, bool keep_link)
2540 int i;
2541 bool global = false;
2543 DP(NETIF_MSG_IFUP, "Starting NIC unload\n");
2545 /* mark driver is unloaded in shmem2 */
2546 if (SHMEM2_HAS(bp, drv_capabilities_flag)) {
2547 u32 val;
2548 val = SHMEM2_RD(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)]);
2549 SHMEM2_WR(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)],
2550 val & ~DRV_FLAGS_CAPABILITIES_LOADED_L2);
2553 if ((bp->state == BNX2X_STATE_CLOSED) ||
2554 (bp->state == BNX2X_STATE_ERROR)) {
2555 /* We can get here if the driver has been unloaded
2556 * during parity error recovery and is either waiting for a
2557 * leader to complete or for other functions to unload and
2558 * then ifdown has been issued. In this case we want to
2559 * unload and let other functions to complete a recovery
2560 * process.
2562 bp->recovery_state = BNX2X_RECOVERY_DONE;
2563 bp->is_leader = 0;
2564 bnx2x_release_leader_lock(bp);
2565 smp_mb();
2567 DP(NETIF_MSG_IFDOWN, "Releasing a leadership...\n");
2568 BNX2X_ERR("Can't unload in closed or error state\n");
2569 return -EINVAL;
2573 * It's important to set the bp->state to the value different from
2574 * BNX2X_STATE_OPEN and only then stop the Tx. Otherwise bnx2x_tx_int()
2575 * may restart the Tx from the NAPI context (see bnx2x_tx_int()).
2577 bp->state = BNX2X_STATE_CLOSING_WAIT4_HALT;
2578 smp_mb();
2580 if (CNIC_LOADED(bp))
2581 bnx2x_cnic_notify(bp, CNIC_CTL_STOP_CMD);
2583 /* Stop Tx */
2584 bnx2x_tx_disable(bp);
2585 netdev_reset_tc(bp->dev);
2587 bp->rx_mode = BNX2X_RX_MODE_NONE;
2589 del_timer_sync(&bp->timer);
2591 /* Set ALWAYS_ALIVE bit in shmem */
2592 bp->fw_drv_pulse_wr_seq |= DRV_PULSE_ALWAYS_ALIVE;
2594 bnx2x_drv_pulse(bp);
2596 bnx2x_stats_handle(bp, STATS_EVENT_STOP);
2597 bnx2x_save_statistics(bp);
2599 /* Cleanup the chip if needed */
2600 if (unload_mode != UNLOAD_RECOVERY)
2601 bnx2x_chip_cleanup(bp, unload_mode, keep_link);
2602 else {
2603 /* Send the UNLOAD_REQUEST to the MCP */
2604 bnx2x_send_unload_req(bp, unload_mode);
2607 * Prevent transactions to host from the functions on the
2608 * engine that doesn't reset global blocks in case of global
2609 * attention once gloabl blocks are reset and gates are opened
2610 * (the engine which leader will perform the recovery
2611 * last).
2613 if (!CHIP_IS_E1x(bp))
2614 bnx2x_pf_disable(bp);
2616 /* Disable HW interrupts, NAPI */
2617 bnx2x_netif_stop(bp, 1);
2618 /* Delete all NAPI objects */
2619 bnx2x_del_all_napi(bp);
2620 if (CNIC_LOADED(bp))
2621 bnx2x_del_all_napi_cnic(bp);
2622 /* Release IRQs */
2623 bnx2x_free_irq(bp);
2625 /* Report UNLOAD_DONE to MCP */
2626 bnx2x_send_unload_done(bp, false);
2630 * At this stage no more interrupts will arrive so we may safly clean
2631 * the queueable objects here in case they failed to get cleaned so far.
2633 bnx2x_squeeze_objects(bp);
2635 /* There should be no more pending SP commands at this stage */
2636 bp->sp_state = 0;
2638 bp->port.pmf = 0;
2640 /* Free SKBs, SGEs, TPA pool and driver internals */
2641 bnx2x_free_skbs(bp);
2642 if (CNIC_LOADED(bp))
2643 bnx2x_free_skbs_cnic(bp);
2644 for_each_rx_queue(bp, i)
2645 bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
2647 if (CNIC_LOADED(bp)) {
2648 bnx2x_free_fp_mem_cnic(bp);
2649 bnx2x_free_mem_cnic(bp);
2651 bnx2x_free_mem(bp);
2653 bp->state = BNX2X_STATE_CLOSED;
2654 bp->cnic_loaded = false;
2656 /* Check if there are pending parity attentions. If there are - set
2657 * RECOVERY_IN_PROGRESS.
2659 if (bnx2x_chk_parity_attn(bp, &global, false)) {
2660 bnx2x_set_reset_in_progress(bp);
2662 /* Set RESET_IS_GLOBAL if needed */
2663 if (global)
2664 bnx2x_set_reset_global(bp);
2668 /* The last driver must disable a "close the gate" if there is no
2669 * parity attention or "process kill" pending.
2671 if (!bnx2x_clear_pf_load(bp) && bnx2x_reset_is_done(bp, BP_PATH(bp)))
2672 bnx2x_disable_close_the_gate(bp);
2674 DP(NETIF_MSG_IFUP, "Ending NIC unload\n");
2676 return 0;
2679 int bnx2x_set_power_state(struct bnx2x *bp, pci_power_t state)
2681 u16 pmcsr;
2683 /* If there is no power capability, silently succeed */
2684 if (!bp->pm_cap) {
2685 BNX2X_DEV_INFO("No power capability. Breaking.\n");
2686 return 0;
2689 pci_read_config_word(bp->pdev, bp->pm_cap + PCI_PM_CTRL, &pmcsr);
2691 switch (state) {
2692 case PCI_D0:
2693 pci_write_config_word(bp->pdev, bp->pm_cap + PCI_PM_CTRL,
2694 ((pmcsr & ~PCI_PM_CTRL_STATE_MASK) |
2695 PCI_PM_CTRL_PME_STATUS));
2697 if (pmcsr & PCI_PM_CTRL_STATE_MASK)
2698 /* delay required during transition out of D3hot */
2699 msleep(20);
2700 break;
2702 case PCI_D3hot:
2703 /* If there are other clients above don't
2704 shut down the power */
2705 if (atomic_read(&bp->pdev->enable_cnt) != 1)
2706 return 0;
2707 /* Don't shut down the power for emulation and FPGA */
2708 if (CHIP_REV_IS_SLOW(bp))
2709 return 0;
2711 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
2712 pmcsr |= 3;
2714 if (bp->wol)
2715 pmcsr |= PCI_PM_CTRL_PME_ENABLE;
2717 pci_write_config_word(bp->pdev, bp->pm_cap + PCI_PM_CTRL,
2718 pmcsr);
2720 /* No more memory access after this point until
2721 * device is brought back to D0.
2723 break;
2725 default:
2726 dev_err(&bp->pdev->dev, "Can't support state = %d\n", state);
2727 return -EINVAL;
2729 return 0;
2733 * net_device service functions
2735 int bnx2x_poll(struct napi_struct *napi, int budget)
2737 int work_done = 0;
2738 u8 cos;
2739 struct bnx2x_fastpath *fp = container_of(napi, struct bnx2x_fastpath,
2740 napi);
2741 struct bnx2x *bp = fp->bp;
2743 while (1) {
2744 #ifdef BNX2X_STOP_ON_ERROR
2745 if (unlikely(bp->panic)) {
2746 napi_complete(napi);
2747 return 0;
2749 #endif
2751 for_each_cos_in_tx_queue(fp, cos)
2752 if (bnx2x_tx_queue_has_work(fp->txdata_ptr[cos]))
2753 bnx2x_tx_int(bp, fp->txdata_ptr[cos]);
2756 if (bnx2x_has_rx_work(fp)) {
2757 work_done += bnx2x_rx_int(fp, budget - work_done);
2759 /* must not complete if we consumed full budget */
2760 if (work_done >= budget)
2761 break;
2764 /* Fall out from the NAPI loop if needed */
2765 if (!(bnx2x_has_rx_work(fp) || bnx2x_has_tx_work(fp))) {
2767 /* No need to update SB for FCoE L2 ring as long as
2768 * it's connected to the default SB and the SB
2769 * has been updated when NAPI was scheduled.
2771 if (IS_FCOE_FP(fp)) {
2772 napi_complete(napi);
2773 break;
2775 bnx2x_update_fpsb_idx(fp);
2776 /* bnx2x_has_rx_work() reads the status block,
2777 * thus we need to ensure that status block indices
2778 * have been actually read (bnx2x_update_fpsb_idx)
2779 * prior to this check (bnx2x_has_rx_work) so that
2780 * we won't write the "newer" value of the status block
2781 * to IGU (if there was a DMA right after
2782 * bnx2x_has_rx_work and if there is no rmb, the memory
2783 * reading (bnx2x_update_fpsb_idx) may be postponed
2784 * to right before bnx2x_ack_sb). In this case there
2785 * will never be another interrupt until there is
2786 * another update of the status block, while there
2787 * is still unhandled work.
2789 rmb();
2791 if (!(bnx2x_has_rx_work(fp) || bnx2x_has_tx_work(fp))) {
2792 napi_complete(napi);
2793 /* Re-enable interrupts */
2794 DP(NETIF_MSG_RX_STATUS,
2795 "Update index to %d\n", fp->fp_hc_idx);
2796 bnx2x_ack_sb(bp, fp->igu_sb_id, USTORM_ID,
2797 le16_to_cpu(fp->fp_hc_idx),
2798 IGU_INT_ENABLE, 1);
2799 break;
2804 return work_done;
2807 /* we split the first BD into headers and data BDs
2808 * to ease the pain of our fellow microcode engineers
2809 * we use one mapping for both BDs
2811 static noinline u16 bnx2x_tx_split(struct bnx2x *bp,
2812 struct bnx2x_fp_txdata *txdata,
2813 struct sw_tx_bd *tx_buf,
2814 struct eth_tx_start_bd **tx_bd, u16 hlen,
2815 u16 bd_prod, int nbd)
2817 struct eth_tx_start_bd *h_tx_bd = *tx_bd;
2818 struct eth_tx_bd *d_tx_bd;
2819 dma_addr_t mapping;
2820 int old_len = le16_to_cpu(h_tx_bd->nbytes);
2822 /* first fix first BD */
2823 h_tx_bd->nbd = cpu_to_le16(nbd);
2824 h_tx_bd->nbytes = cpu_to_le16(hlen);
2826 DP(NETIF_MSG_TX_QUEUED, "TSO split header size is %d (%x:%x) nbd %d\n",
2827 h_tx_bd->nbytes, h_tx_bd->addr_hi, h_tx_bd->addr_lo, h_tx_bd->nbd);
2829 /* now get a new data BD
2830 * (after the pbd) and fill it */
2831 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
2832 d_tx_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
2834 mapping = HILO_U64(le32_to_cpu(h_tx_bd->addr_hi),
2835 le32_to_cpu(h_tx_bd->addr_lo)) + hlen;
2837 d_tx_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
2838 d_tx_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
2839 d_tx_bd->nbytes = cpu_to_le16(old_len - hlen);
2841 /* this marks the BD as one that has no individual mapping */
2842 tx_buf->flags |= BNX2X_TSO_SPLIT_BD;
2844 DP(NETIF_MSG_TX_QUEUED,
2845 "TSO split data size is %d (%x:%x)\n",
2846 d_tx_bd->nbytes, d_tx_bd->addr_hi, d_tx_bd->addr_lo);
2848 /* update tx_bd */
2849 *tx_bd = (struct eth_tx_start_bd *)d_tx_bd;
2851 return bd_prod;
2854 static inline u16 bnx2x_csum_fix(unsigned char *t_header, u16 csum, s8 fix)
2856 if (fix > 0)
2857 csum = (u16) ~csum_fold(csum_sub(csum,
2858 csum_partial(t_header - fix, fix, 0)));
2860 else if (fix < 0)
2861 csum = (u16) ~csum_fold(csum_add(csum,
2862 csum_partial(t_header, -fix, 0)));
2864 return swab16(csum);
2867 static inline u32 bnx2x_xmit_type(struct bnx2x *bp, struct sk_buff *skb)
2869 u32 rc;
2871 if (skb->ip_summed != CHECKSUM_PARTIAL)
2872 rc = XMIT_PLAIN;
2874 else {
2875 if (vlan_get_protocol(skb) == htons(ETH_P_IPV6)) {
2876 rc = XMIT_CSUM_V6;
2877 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2878 rc |= XMIT_CSUM_TCP;
2880 } else {
2881 rc = XMIT_CSUM_V4;
2882 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2883 rc |= XMIT_CSUM_TCP;
2887 if (skb_is_gso_v6(skb))
2888 rc |= XMIT_GSO_V6 | XMIT_CSUM_TCP | XMIT_CSUM_V6;
2889 else if (skb_is_gso(skb))
2890 rc |= XMIT_GSO_V4 | XMIT_CSUM_V4 | XMIT_CSUM_TCP;
2892 return rc;
2895 #if (MAX_SKB_FRAGS >= MAX_FETCH_BD - 3)
2896 /* check if packet requires linearization (packet is too fragmented)
2897 no need to check fragmentation if page size > 8K (there will be no
2898 violation to FW restrictions) */
2899 static int bnx2x_pkt_req_lin(struct bnx2x *bp, struct sk_buff *skb,
2900 u32 xmit_type)
2902 int to_copy = 0;
2903 int hlen = 0;
2904 int first_bd_sz = 0;
2906 /* 3 = 1 (for linear data BD) + 2 (for PBD and last BD) */
2907 if (skb_shinfo(skb)->nr_frags >= (MAX_FETCH_BD - 3)) {
2909 if (xmit_type & XMIT_GSO) {
2910 unsigned short lso_mss = skb_shinfo(skb)->gso_size;
2911 /* Check if LSO packet needs to be copied:
2912 3 = 1 (for headers BD) + 2 (for PBD and last BD) */
2913 int wnd_size = MAX_FETCH_BD - 3;
2914 /* Number of windows to check */
2915 int num_wnds = skb_shinfo(skb)->nr_frags - wnd_size;
2916 int wnd_idx = 0;
2917 int frag_idx = 0;
2918 u32 wnd_sum = 0;
2920 /* Headers length */
2921 hlen = (int)(skb_transport_header(skb) - skb->data) +
2922 tcp_hdrlen(skb);
2924 /* Amount of data (w/o headers) on linear part of SKB*/
2925 first_bd_sz = skb_headlen(skb) - hlen;
2927 wnd_sum = first_bd_sz;
2929 /* Calculate the first sum - it's special */
2930 for (frag_idx = 0; frag_idx < wnd_size - 1; frag_idx++)
2931 wnd_sum +=
2932 skb_frag_size(&skb_shinfo(skb)->frags[frag_idx]);
2934 /* If there was data on linear skb data - check it */
2935 if (first_bd_sz > 0) {
2936 if (unlikely(wnd_sum < lso_mss)) {
2937 to_copy = 1;
2938 goto exit_lbl;
2941 wnd_sum -= first_bd_sz;
2944 /* Others are easier: run through the frag list and
2945 check all windows */
2946 for (wnd_idx = 0; wnd_idx <= num_wnds; wnd_idx++) {
2947 wnd_sum +=
2948 skb_frag_size(&skb_shinfo(skb)->frags[wnd_idx + wnd_size - 1]);
2950 if (unlikely(wnd_sum < lso_mss)) {
2951 to_copy = 1;
2952 break;
2954 wnd_sum -=
2955 skb_frag_size(&skb_shinfo(skb)->frags[wnd_idx]);
2957 } else {
2958 /* in non-LSO too fragmented packet should always
2959 be linearized */
2960 to_copy = 1;
2964 exit_lbl:
2965 if (unlikely(to_copy))
2966 DP(NETIF_MSG_TX_QUEUED,
2967 "Linearization IS REQUIRED for %s packet. num_frags %d hlen %d first_bd_sz %d\n",
2968 (xmit_type & XMIT_GSO) ? "LSO" : "non-LSO",
2969 skb_shinfo(skb)->nr_frags, hlen, first_bd_sz);
2971 return to_copy;
2973 #endif
2975 static inline void bnx2x_set_pbd_gso_e2(struct sk_buff *skb, u32 *parsing_data,
2976 u32 xmit_type)
2978 *parsing_data |= (skb_shinfo(skb)->gso_size <<
2979 ETH_TX_PARSE_BD_E2_LSO_MSS_SHIFT) &
2980 ETH_TX_PARSE_BD_E2_LSO_MSS;
2981 if ((xmit_type & XMIT_GSO_V6) &&
2982 (ipv6_hdr(skb)->nexthdr == NEXTHDR_IPV6))
2983 *parsing_data |= ETH_TX_PARSE_BD_E2_IPV6_WITH_EXT_HDR;
2987 * bnx2x_set_pbd_gso - update PBD in GSO case.
2989 * @skb: packet skb
2990 * @pbd: parse BD
2991 * @xmit_type: xmit flags
2993 static inline void bnx2x_set_pbd_gso(struct sk_buff *skb,
2994 struct eth_tx_parse_bd_e1x *pbd,
2995 u32 xmit_type)
2997 pbd->lso_mss = cpu_to_le16(skb_shinfo(skb)->gso_size);
2998 pbd->tcp_send_seq = swab32(tcp_hdr(skb)->seq);
2999 pbd->tcp_flags = pbd_tcp_flags(skb);
3001 if (xmit_type & XMIT_GSO_V4) {
3002 pbd->ip_id = swab16(ip_hdr(skb)->id);
3003 pbd->tcp_pseudo_csum =
3004 swab16(~csum_tcpudp_magic(ip_hdr(skb)->saddr,
3005 ip_hdr(skb)->daddr,
3006 0, IPPROTO_TCP, 0));
3008 } else
3009 pbd->tcp_pseudo_csum =
3010 swab16(~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3011 &ipv6_hdr(skb)->daddr,
3012 0, IPPROTO_TCP, 0));
3014 pbd->global_data |= ETH_TX_PARSE_BD_E1X_PSEUDO_CS_WITHOUT_LEN;
3018 * bnx2x_set_pbd_csum_e2 - update PBD with checksum and return header length
3020 * @bp: driver handle
3021 * @skb: packet skb
3022 * @parsing_data: data to be updated
3023 * @xmit_type: xmit flags
3025 * 57712 related
3027 static inline u8 bnx2x_set_pbd_csum_e2(struct bnx2x *bp, struct sk_buff *skb,
3028 u32 *parsing_data, u32 xmit_type)
3030 *parsing_data |=
3031 ((((u8 *)skb_transport_header(skb) - skb->data) >> 1) <<
3032 ETH_TX_PARSE_BD_E2_TCP_HDR_START_OFFSET_W_SHIFT) &
3033 ETH_TX_PARSE_BD_E2_TCP_HDR_START_OFFSET_W;
3035 if (xmit_type & XMIT_CSUM_TCP) {
3036 *parsing_data |= ((tcp_hdrlen(skb) / 4) <<
3037 ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW_SHIFT) &
3038 ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW;
3040 return skb_transport_header(skb) + tcp_hdrlen(skb) - skb->data;
3041 } else
3042 /* We support checksum offload for TCP and UDP only.
3043 * No need to pass the UDP header length - it's a constant.
3045 return skb_transport_header(skb) +
3046 sizeof(struct udphdr) - skb->data;
3049 static inline void bnx2x_set_sbd_csum(struct bnx2x *bp, struct sk_buff *skb,
3050 struct eth_tx_start_bd *tx_start_bd, u32 xmit_type)
3052 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_L4_CSUM;
3054 if (xmit_type & XMIT_CSUM_V4)
3055 tx_start_bd->bd_flags.as_bitfield |=
3056 ETH_TX_BD_FLAGS_IP_CSUM;
3057 else
3058 tx_start_bd->bd_flags.as_bitfield |=
3059 ETH_TX_BD_FLAGS_IPV6;
3061 if (!(xmit_type & XMIT_CSUM_TCP))
3062 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_IS_UDP;
3066 * bnx2x_set_pbd_csum - update PBD with checksum and return header length
3068 * @bp: driver handle
3069 * @skb: packet skb
3070 * @pbd: parse BD to be updated
3071 * @xmit_type: xmit flags
3073 static inline u8 bnx2x_set_pbd_csum(struct bnx2x *bp, struct sk_buff *skb,
3074 struct eth_tx_parse_bd_e1x *pbd,
3075 u32 xmit_type)
3077 u8 hlen = (skb_network_header(skb) - skb->data) >> 1;
3079 /* for now NS flag is not used in Linux */
3080 pbd->global_data =
3081 (hlen | ((skb->protocol == cpu_to_be16(ETH_P_8021Q)) <<
3082 ETH_TX_PARSE_BD_E1X_LLC_SNAP_EN_SHIFT));
3084 pbd->ip_hlen_w = (skb_transport_header(skb) -
3085 skb_network_header(skb)) >> 1;
3087 hlen += pbd->ip_hlen_w;
3089 /* We support checksum offload for TCP and UDP only */
3090 if (xmit_type & XMIT_CSUM_TCP)
3091 hlen += tcp_hdrlen(skb) / 2;
3092 else
3093 hlen += sizeof(struct udphdr) / 2;
3095 pbd->total_hlen_w = cpu_to_le16(hlen);
3096 hlen = hlen*2;
3098 if (xmit_type & XMIT_CSUM_TCP) {
3099 pbd->tcp_pseudo_csum = swab16(tcp_hdr(skb)->check);
3101 } else {
3102 s8 fix = SKB_CS_OFF(skb); /* signed! */
3104 DP(NETIF_MSG_TX_QUEUED,
3105 "hlen %d fix %d csum before fix %x\n",
3106 le16_to_cpu(pbd->total_hlen_w), fix, SKB_CS(skb));
3108 /* HW bug: fixup the CSUM */
3109 pbd->tcp_pseudo_csum =
3110 bnx2x_csum_fix(skb_transport_header(skb),
3111 SKB_CS(skb), fix);
3113 DP(NETIF_MSG_TX_QUEUED, "csum after fix %x\n",
3114 pbd->tcp_pseudo_csum);
3117 return hlen;
3120 /* called with netif_tx_lock
3121 * bnx2x_tx_int() runs without netif_tx_lock unless it needs to call
3122 * netif_wake_queue()
3124 netdev_tx_t bnx2x_start_xmit(struct sk_buff *skb, struct net_device *dev)
3126 struct bnx2x *bp = netdev_priv(dev);
3128 struct netdev_queue *txq;
3129 struct bnx2x_fp_txdata *txdata;
3130 struct sw_tx_bd *tx_buf;
3131 struct eth_tx_start_bd *tx_start_bd, *first_bd;
3132 struct eth_tx_bd *tx_data_bd, *total_pkt_bd = NULL;
3133 struct eth_tx_parse_bd_e1x *pbd_e1x = NULL;
3134 struct eth_tx_parse_bd_e2 *pbd_e2 = NULL;
3135 u32 pbd_e2_parsing_data = 0;
3136 u16 pkt_prod, bd_prod;
3137 int nbd, txq_index;
3138 dma_addr_t mapping;
3139 u32 xmit_type = bnx2x_xmit_type(bp, skb);
3140 int i;
3141 u8 hlen = 0;
3142 __le16 pkt_size = 0;
3143 struct ethhdr *eth;
3144 u8 mac_type = UNICAST_ADDRESS;
3146 #ifdef BNX2X_STOP_ON_ERROR
3147 if (unlikely(bp->panic))
3148 return NETDEV_TX_BUSY;
3149 #endif
3151 txq_index = skb_get_queue_mapping(skb);
3152 txq = netdev_get_tx_queue(dev, txq_index);
3154 BUG_ON(txq_index >= MAX_ETH_TXQ_IDX(bp) + (CNIC_LOADED(bp) ? 1 : 0));
3156 txdata = &bp->bnx2x_txq[txq_index];
3158 /* enable this debug print to view the transmission queue being used
3159 DP(NETIF_MSG_TX_QUEUED, "indices: txq %d, fp %d, txdata %d\n",
3160 txq_index, fp_index, txdata_index); */
3162 /* enable this debug print to view the tranmission details
3163 DP(NETIF_MSG_TX_QUEUED,
3164 "transmitting packet cid %d fp index %d txdata_index %d tx_data ptr %p fp pointer %p\n",
3165 txdata->cid, fp_index, txdata_index, txdata, fp); */
3167 if (unlikely(bnx2x_tx_avail(bp, txdata) <
3168 skb_shinfo(skb)->nr_frags +
3169 BDS_PER_TX_PKT +
3170 NEXT_CNT_PER_TX_PKT(MAX_BDS_PER_TX_PKT))) {
3171 /* Handle special storage cases separately */
3172 if (txdata->tx_ring_size == 0) {
3173 struct bnx2x_eth_q_stats *q_stats =
3174 bnx2x_fp_qstats(bp, txdata->parent_fp);
3175 q_stats->driver_filtered_tx_pkt++;
3176 dev_kfree_skb(skb);
3177 return NETDEV_TX_OK;
3179 bnx2x_fp_qstats(bp, txdata->parent_fp)->driver_xoff++;
3180 netif_tx_stop_queue(txq);
3181 BNX2X_ERR("BUG! Tx ring full when queue awake!\n");
3183 return NETDEV_TX_BUSY;
3186 DP(NETIF_MSG_TX_QUEUED,
3187 "queue[%d]: SKB: summed %x protocol %x protocol(%x,%x) gso type %x xmit_type %x\n",
3188 txq_index, skb->ip_summed, skb->protocol, ipv6_hdr(skb)->nexthdr,
3189 ip_hdr(skb)->protocol, skb_shinfo(skb)->gso_type, xmit_type);
3191 eth = (struct ethhdr *)skb->data;
3193 /* set flag according to packet type (UNICAST_ADDRESS is default)*/
3194 if (unlikely(is_multicast_ether_addr(eth->h_dest))) {
3195 if (is_broadcast_ether_addr(eth->h_dest))
3196 mac_type = BROADCAST_ADDRESS;
3197 else
3198 mac_type = MULTICAST_ADDRESS;
3201 #if (MAX_SKB_FRAGS >= MAX_FETCH_BD - 3)
3202 /* First, check if we need to linearize the skb (due to FW
3203 restrictions). No need to check fragmentation if page size > 8K
3204 (there will be no violation to FW restrictions) */
3205 if (bnx2x_pkt_req_lin(bp, skb, xmit_type)) {
3206 /* Statistics of linearization */
3207 bp->lin_cnt++;
3208 if (skb_linearize(skb) != 0) {
3209 DP(NETIF_MSG_TX_QUEUED,
3210 "SKB linearization failed - silently dropping this SKB\n");
3211 dev_kfree_skb_any(skb);
3212 return NETDEV_TX_OK;
3215 #endif
3216 /* Map skb linear data for DMA */
3217 mapping = dma_map_single(&bp->pdev->dev, skb->data,
3218 skb_headlen(skb), DMA_TO_DEVICE);
3219 if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
3220 DP(NETIF_MSG_TX_QUEUED,
3221 "SKB mapping failed - silently dropping this SKB\n");
3222 dev_kfree_skb_any(skb);
3223 return NETDEV_TX_OK;
3226 Please read carefully. First we use one BD which we mark as start,
3227 then we have a parsing info BD (used for TSO or xsum),
3228 and only then we have the rest of the TSO BDs.
3229 (don't forget to mark the last one as last,
3230 and to unmap only AFTER you write to the BD ...)
3231 And above all, all pdb sizes are in words - NOT DWORDS!
3234 /* get current pkt produced now - advance it just before sending packet
3235 * since mapping of pages may fail and cause packet to be dropped
3237 pkt_prod = txdata->tx_pkt_prod;
3238 bd_prod = TX_BD(txdata->tx_bd_prod);
3240 /* get a tx_buf and first BD
3241 * tx_start_bd may be changed during SPLIT,
3242 * but first_bd will always stay first
3244 tx_buf = &txdata->tx_buf_ring[TX_BD(pkt_prod)];
3245 tx_start_bd = &txdata->tx_desc_ring[bd_prod].start_bd;
3246 first_bd = tx_start_bd;
3248 tx_start_bd->bd_flags.as_bitfield = ETH_TX_BD_FLAGS_START_BD;
3249 SET_FLAG(tx_start_bd->general_data,
3250 ETH_TX_START_BD_PARSE_NBDS,
3253 /* header nbd */
3254 SET_FLAG(tx_start_bd->general_data, ETH_TX_START_BD_HDR_NBDS, 1);
3256 /* remember the first BD of the packet */
3257 tx_buf->first_bd = txdata->tx_bd_prod;
3258 tx_buf->skb = skb;
3259 tx_buf->flags = 0;
3261 DP(NETIF_MSG_TX_QUEUED,
3262 "sending pkt %u @%p next_idx %u bd %u @%p\n",
3263 pkt_prod, tx_buf, txdata->tx_pkt_prod, bd_prod, tx_start_bd);
3265 if (vlan_tx_tag_present(skb)) {
3266 tx_start_bd->vlan_or_ethertype =
3267 cpu_to_le16(vlan_tx_tag_get(skb));
3268 tx_start_bd->bd_flags.as_bitfield |=
3269 (X_ETH_OUTBAND_VLAN << ETH_TX_BD_FLAGS_VLAN_MODE_SHIFT);
3270 } else
3271 tx_start_bd->vlan_or_ethertype = cpu_to_le16(pkt_prod);
3273 /* turn on parsing and get a BD */
3274 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3276 if (xmit_type & XMIT_CSUM)
3277 bnx2x_set_sbd_csum(bp, skb, tx_start_bd, xmit_type);
3279 if (!CHIP_IS_E1x(bp)) {
3280 pbd_e2 = &txdata->tx_desc_ring[bd_prod].parse_bd_e2;
3281 memset(pbd_e2, 0, sizeof(struct eth_tx_parse_bd_e2));
3282 /* Set PBD in checksum offload case */
3283 if (xmit_type & XMIT_CSUM)
3284 hlen = bnx2x_set_pbd_csum_e2(bp, skb,
3285 &pbd_e2_parsing_data,
3286 xmit_type);
3287 if (IS_MF_SI(bp)) {
3289 * fill in the MAC addresses in the PBD - for local
3290 * switching
3292 bnx2x_set_fw_mac_addr(&pbd_e2->src_mac_addr_hi,
3293 &pbd_e2->src_mac_addr_mid,
3294 &pbd_e2->src_mac_addr_lo,
3295 eth->h_source);
3296 bnx2x_set_fw_mac_addr(&pbd_e2->dst_mac_addr_hi,
3297 &pbd_e2->dst_mac_addr_mid,
3298 &pbd_e2->dst_mac_addr_lo,
3299 eth->h_dest);
3302 SET_FLAG(pbd_e2_parsing_data,
3303 ETH_TX_PARSE_BD_E2_ETH_ADDR_TYPE, mac_type);
3304 } else {
3305 u16 global_data = 0;
3306 pbd_e1x = &txdata->tx_desc_ring[bd_prod].parse_bd_e1x;
3307 memset(pbd_e1x, 0, sizeof(struct eth_tx_parse_bd_e1x));
3308 /* Set PBD in checksum offload case */
3309 if (xmit_type & XMIT_CSUM)
3310 hlen = bnx2x_set_pbd_csum(bp, skb, pbd_e1x, xmit_type);
3312 SET_FLAG(global_data,
3313 ETH_TX_PARSE_BD_E1X_ETH_ADDR_TYPE, mac_type);
3314 pbd_e1x->global_data |= cpu_to_le16(global_data);
3317 /* Setup the data pointer of the first BD of the packet */
3318 tx_start_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
3319 tx_start_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
3320 nbd = 2; /* start_bd + pbd + frags (updated when pages are mapped) */
3321 tx_start_bd->nbytes = cpu_to_le16(skb_headlen(skb));
3322 pkt_size = tx_start_bd->nbytes;
3324 DP(NETIF_MSG_TX_QUEUED,
3325 "first bd @%p addr (%x:%x) nbd %d nbytes %d flags %x vlan %x\n",
3326 tx_start_bd, tx_start_bd->addr_hi, tx_start_bd->addr_lo,
3327 le16_to_cpu(tx_start_bd->nbd), le16_to_cpu(tx_start_bd->nbytes),
3328 tx_start_bd->bd_flags.as_bitfield,
3329 le16_to_cpu(tx_start_bd->vlan_or_ethertype));
3331 if (xmit_type & XMIT_GSO) {
3333 DP(NETIF_MSG_TX_QUEUED,
3334 "TSO packet len %d hlen %d total len %d tso size %d\n",
3335 skb->len, hlen, skb_headlen(skb),
3336 skb_shinfo(skb)->gso_size);
3338 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_SW_LSO;
3340 if (unlikely(skb_headlen(skb) > hlen))
3341 bd_prod = bnx2x_tx_split(bp, txdata, tx_buf,
3342 &tx_start_bd, hlen,
3343 bd_prod, ++nbd);
3344 if (!CHIP_IS_E1x(bp))
3345 bnx2x_set_pbd_gso_e2(skb, &pbd_e2_parsing_data,
3346 xmit_type);
3347 else
3348 bnx2x_set_pbd_gso(skb, pbd_e1x, xmit_type);
3351 /* Set the PBD's parsing_data field if not zero
3352 * (for the chips newer than 57711).
3354 if (pbd_e2_parsing_data)
3355 pbd_e2->parsing_data = cpu_to_le32(pbd_e2_parsing_data);
3357 tx_data_bd = (struct eth_tx_bd *)tx_start_bd;
3359 /* Handle fragmented skb */
3360 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3361 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3363 mapping = skb_frag_dma_map(&bp->pdev->dev, frag, 0,
3364 skb_frag_size(frag), DMA_TO_DEVICE);
3365 if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
3366 unsigned int pkts_compl = 0, bytes_compl = 0;
3368 DP(NETIF_MSG_TX_QUEUED,
3369 "Unable to map page - dropping packet...\n");
3371 /* we need unmap all buffers already mapped
3372 * for this SKB;
3373 * first_bd->nbd need to be properly updated
3374 * before call to bnx2x_free_tx_pkt
3376 first_bd->nbd = cpu_to_le16(nbd);
3377 bnx2x_free_tx_pkt(bp, txdata,
3378 TX_BD(txdata->tx_pkt_prod),
3379 &pkts_compl, &bytes_compl);
3380 return NETDEV_TX_OK;
3383 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3384 tx_data_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
3385 if (total_pkt_bd == NULL)
3386 total_pkt_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
3388 tx_data_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
3389 tx_data_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
3390 tx_data_bd->nbytes = cpu_to_le16(skb_frag_size(frag));
3391 le16_add_cpu(&pkt_size, skb_frag_size(frag));
3392 nbd++;
3394 DP(NETIF_MSG_TX_QUEUED,
3395 "frag %d bd @%p addr (%x:%x) nbytes %d\n",
3396 i, tx_data_bd, tx_data_bd->addr_hi, tx_data_bd->addr_lo,
3397 le16_to_cpu(tx_data_bd->nbytes));
3400 DP(NETIF_MSG_TX_QUEUED, "last bd @%p\n", tx_data_bd);
3402 /* update with actual num BDs */
3403 first_bd->nbd = cpu_to_le16(nbd);
3405 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3407 /* now send a tx doorbell, counting the next BD
3408 * if the packet contains or ends with it
3410 if (TX_BD_POFF(bd_prod) < nbd)
3411 nbd++;
3413 /* total_pkt_bytes should be set on the first data BD if
3414 * it's not an LSO packet and there is more than one
3415 * data BD. In this case pkt_size is limited by an MTU value.
3416 * However we prefer to set it for an LSO packet (while we don't
3417 * have to) in order to save some CPU cycles in a none-LSO
3418 * case, when we much more care about them.
3420 if (total_pkt_bd != NULL)
3421 total_pkt_bd->total_pkt_bytes = pkt_size;
3423 if (pbd_e1x)
3424 DP(NETIF_MSG_TX_QUEUED,
3425 "PBD (E1X) @%p ip_data %x ip_hlen %u ip_id %u lso_mss %u tcp_flags %x xsum %x seq %u hlen %u\n",
3426 pbd_e1x, pbd_e1x->global_data, pbd_e1x->ip_hlen_w,
3427 pbd_e1x->ip_id, pbd_e1x->lso_mss, pbd_e1x->tcp_flags,
3428 pbd_e1x->tcp_pseudo_csum, pbd_e1x->tcp_send_seq,
3429 le16_to_cpu(pbd_e1x->total_hlen_w));
3430 if (pbd_e2)
3431 DP(NETIF_MSG_TX_QUEUED,
3432 "PBD (E2) @%p dst %x %x %x src %x %x %x parsing_data %x\n",
3433 pbd_e2, pbd_e2->dst_mac_addr_hi, pbd_e2->dst_mac_addr_mid,
3434 pbd_e2->dst_mac_addr_lo, pbd_e2->src_mac_addr_hi,
3435 pbd_e2->src_mac_addr_mid, pbd_e2->src_mac_addr_lo,
3436 pbd_e2->parsing_data);
3437 DP(NETIF_MSG_TX_QUEUED, "doorbell: nbd %d bd %u\n", nbd, bd_prod);
3439 netdev_tx_sent_queue(txq, skb->len);
3441 skb_tx_timestamp(skb);
3443 txdata->tx_pkt_prod++;
3445 * Make sure that the BD data is updated before updating the producer
3446 * since FW might read the BD right after the producer is updated.
3447 * This is only applicable for weak-ordered memory model archs such
3448 * as IA-64. The following barrier is also mandatory since FW will
3449 * assumes packets must have BDs.
3451 wmb();
3453 txdata->tx_db.data.prod += nbd;
3454 barrier();
3456 DOORBELL(bp, txdata->cid, txdata->tx_db.raw);
3458 mmiowb();
3460 txdata->tx_bd_prod += nbd;
3462 if (unlikely(bnx2x_tx_avail(bp, txdata) < MAX_DESC_PER_TX_PKT)) {
3463 netif_tx_stop_queue(txq);
3465 /* paired memory barrier is in bnx2x_tx_int(), we have to keep
3466 * ordering of set_bit() in netif_tx_stop_queue() and read of
3467 * fp->bd_tx_cons */
3468 smp_mb();
3470 bnx2x_fp_qstats(bp, txdata->parent_fp)->driver_xoff++;
3471 if (bnx2x_tx_avail(bp, txdata) >= MAX_DESC_PER_TX_PKT)
3472 netif_tx_wake_queue(txq);
3474 txdata->tx_pkt++;
3476 return NETDEV_TX_OK;
3480 * bnx2x_setup_tc - routine to configure net_device for multi tc
3482 * @netdev: net device to configure
3483 * @tc: number of traffic classes to enable
3485 * callback connected to the ndo_setup_tc function pointer
3487 int bnx2x_setup_tc(struct net_device *dev, u8 num_tc)
3489 int cos, prio, count, offset;
3490 struct bnx2x *bp = netdev_priv(dev);
3492 /* setup tc must be called under rtnl lock */
3493 ASSERT_RTNL();
3495 /* no traffic classes requested. aborting */
3496 if (!num_tc) {
3497 netdev_reset_tc(dev);
3498 return 0;
3501 /* requested to support too many traffic classes */
3502 if (num_tc > bp->max_cos) {
3503 BNX2X_ERR("support for too many traffic classes requested: %d. max supported is %d\n",
3504 num_tc, bp->max_cos);
3505 return -EINVAL;
3508 /* declare amount of supported traffic classes */
3509 if (netdev_set_num_tc(dev, num_tc)) {
3510 BNX2X_ERR("failed to declare %d traffic classes\n", num_tc);
3511 return -EINVAL;
3514 /* configure priority to traffic class mapping */
3515 for (prio = 0; prio < BNX2X_MAX_PRIORITY; prio++) {
3516 netdev_set_prio_tc_map(dev, prio, bp->prio_to_cos[prio]);
3517 DP(BNX2X_MSG_SP | NETIF_MSG_IFUP,
3518 "mapping priority %d to tc %d\n",
3519 prio, bp->prio_to_cos[prio]);
3523 /* Use this configuration to diffrentiate tc0 from other COSes
3524 This can be used for ets or pfc, and save the effort of setting
3525 up a multio class queue disc or negotiating DCBX with a switch
3526 netdev_set_prio_tc_map(dev, 0, 0);
3527 DP(BNX2X_MSG_SP, "mapping priority %d to tc %d\n", 0, 0);
3528 for (prio = 1; prio < 16; prio++) {
3529 netdev_set_prio_tc_map(dev, prio, 1);
3530 DP(BNX2X_MSG_SP, "mapping priority %d to tc %d\n", prio, 1);
3531 } */
3533 /* configure traffic class to transmission queue mapping */
3534 for (cos = 0; cos < bp->max_cos; cos++) {
3535 count = BNX2X_NUM_ETH_QUEUES(bp);
3536 offset = cos * BNX2X_NUM_NON_CNIC_QUEUES(bp);
3537 netdev_set_tc_queue(dev, cos, count, offset);
3538 DP(BNX2X_MSG_SP | NETIF_MSG_IFUP,
3539 "mapping tc %d to offset %d count %d\n",
3540 cos, offset, count);
3543 return 0;
3546 /* called with rtnl_lock */
3547 int bnx2x_change_mac_addr(struct net_device *dev, void *p)
3549 struct sockaddr *addr = p;
3550 struct bnx2x *bp = netdev_priv(dev);
3551 int rc = 0;
3553 if (!bnx2x_is_valid_ether_addr(bp, addr->sa_data)) {
3554 BNX2X_ERR("Requested MAC address is not valid\n");
3555 return -EINVAL;
3558 if ((IS_MF_STORAGE_SD(bp) || IS_MF_FCOE_AFEX(bp)) &&
3559 !is_zero_ether_addr(addr->sa_data)) {
3560 BNX2X_ERR("Can't configure non-zero address on iSCSI or FCoE functions in MF-SD mode\n");
3561 return -EINVAL;
3564 if (netif_running(dev)) {
3565 rc = bnx2x_set_eth_mac(bp, false);
3566 if (rc)
3567 return rc;
3570 dev->addr_assign_type &= ~NET_ADDR_RANDOM;
3571 memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
3573 if (netif_running(dev))
3574 rc = bnx2x_set_eth_mac(bp, true);
3576 return rc;
3579 static void bnx2x_free_fp_mem_at(struct bnx2x *bp, int fp_index)
3581 union host_hc_status_block *sb = &bnx2x_fp(bp, fp_index, status_blk);
3582 struct bnx2x_fastpath *fp = &bp->fp[fp_index];
3583 u8 cos;
3585 /* Common */
3587 if (IS_FCOE_IDX(fp_index)) {
3588 memset(sb, 0, sizeof(union host_hc_status_block));
3589 fp->status_blk_mapping = 0;
3590 } else {
3591 /* status blocks */
3592 if (!CHIP_IS_E1x(bp))
3593 BNX2X_PCI_FREE(sb->e2_sb,
3594 bnx2x_fp(bp, fp_index,
3595 status_blk_mapping),
3596 sizeof(struct host_hc_status_block_e2));
3597 else
3598 BNX2X_PCI_FREE(sb->e1x_sb,
3599 bnx2x_fp(bp, fp_index,
3600 status_blk_mapping),
3601 sizeof(struct host_hc_status_block_e1x));
3604 /* Rx */
3605 if (!skip_rx_queue(bp, fp_index)) {
3606 bnx2x_free_rx_bds(fp);
3608 /* fastpath rx rings: rx_buf rx_desc rx_comp */
3609 BNX2X_FREE(bnx2x_fp(bp, fp_index, rx_buf_ring));
3610 BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_desc_ring),
3611 bnx2x_fp(bp, fp_index, rx_desc_mapping),
3612 sizeof(struct eth_rx_bd) * NUM_RX_BD);
3614 BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_comp_ring),
3615 bnx2x_fp(bp, fp_index, rx_comp_mapping),
3616 sizeof(struct eth_fast_path_rx_cqe) *
3617 NUM_RCQ_BD);
3619 /* SGE ring */
3620 BNX2X_FREE(bnx2x_fp(bp, fp_index, rx_page_ring));
3621 BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_sge_ring),
3622 bnx2x_fp(bp, fp_index, rx_sge_mapping),
3623 BCM_PAGE_SIZE * NUM_RX_SGE_PAGES);
3626 /* Tx */
3627 if (!skip_tx_queue(bp, fp_index)) {
3628 /* fastpath tx rings: tx_buf tx_desc */
3629 for_each_cos_in_tx_queue(fp, cos) {
3630 struct bnx2x_fp_txdata *txdata = fp->txdata_ptr[cos];
3632 DP(NETIF_MSG_IFDOWN,
3633 "freeing tx memory of fp %d cos %d cid %d\n",
3634 fp_index, cos, txdata->cid);
3636 BNX2X_FREE(txdata->tx_buf_ring);
3637 BNX2X_PCI_FREE(txdata->tx_desc_ring,
3638 txdata->tx_desc_mapping,
3639 sizeof(union eth_tx_bd_types) * NUM_TX_BD);
3642 /* end of fastpath */
3645 void bnx2x_free_fp_mem_cnic(struct bnx2x *bp)
3647 int i;
3648 for_each_cnic_queue(bp, i)
3649 bnx2x_free_fp_mem_at(bp, i);
3652 void bnx2x_free_fp_mem(struct bnx2x *bp)
3654 int i;
3655 for_each_eth_queue(bp, i)
3656 bnx2x_free_fp_mem_at(bp, i);
3659 static void set_sb_shortcuts(struct bnx2x *bp, int index)
3661 union host_hc_status_block status_blk = bnx2x_fp(bp, index, status_blk);
3662 if (!CHIP_IS_E1x(bp)) {
3663 bnx2x_fp(bp, index, sb_index_values) =
3664 (__le16 *)status_blk.e2_sb->sb.index_values;
3665 bnx2x_fp(bp, index, sb_running_index) =
3666 (__le16 *)status_blk.e2_sb->sb.running_index;
3667 } else {
3668 bnx2x_fp(bp, index, sb_index_values) =
3669 (__le16 *)status_blk.e1x_sb->sb.index_values;
3670 bnx2x_fp(bp, index, sb_running_index) =
3671 (__le16 *)status_blk.e1x_sb->sb.running_index;
3675 /* Returns the number of actually allocated BDs */
3676 static int bnx2x_alloc_rx_bds(struct bnx2x_fastpath *fp,
3677 int rx_ring_size)
3679 struct bnx2x *bp = fp->bp;
3680 u16 ring_prod, cqe_ring_prod;
3681 int i, failure_cnt = 0;
3683 fp->rx_comp_cons = 0;
3684 cqe_ring_prod = ring_prod = 0;
3686 /* This routine is called only during fo init so
3687 * fp->eth_q_stats.rx_skb_alloc_failed = 0
3689 for (i = 0; i < rx_ring_size; i++) {
3690 if (bnx2x_alloc_rx_data(bp, fp, ring_prod) < 0) {
3691 failure_cnt++;
3692 continue;
3694 ring_prod = NEXT_RX_IDX(ring_prod);
3695 cqe_ring_prod = NEXT_RCQ_IDX(cqe_ring_prod);
3696 WARN_ON(ring_prod <= (i - failure_cnt));
3699 if (failure_cnt)
3700 BNX2X_ERR("was only able to allocate %d rx skbs on queue[%d]\n",
3701 i - failure_cnt, fp->index);
3703 fp->rx_bd_prod = ring_prod;
3704 /* Limit the CQE producer by the CQE ring size */
3705 fp->rx_comp_prod = min_t(u16, NUM_RCQ_RINGS*RCQ_DESC_CNT,
3706 cqe_ring_prod);
3707 fp->rx_pkt = fp->rx_calls = 0;
3709 bnx2x_fp_stats(bp, fp)->eth_q_stats.rx_skb_alloc_failed += failure_cnt;
3711 return i - failure_cnt;
3714 static void bnx2x_set_next_page_rx_cq(struct bnx2x_fastpath *fp)
3716 int i;
3718 for (i = 1; i <= NUM_RCQ_RINGS; i++) {
3719 struct eth_rx_cqe_next_page *nextpg;
3721 nextpg = (struct eth_rx_cqe_next_page *)
3722 &fp->rx_comp_ring[RCQ_DESC_CNT * i - 1];
3723 nextpg->addr_hi =
3724 cpu_to_le32(U64_HI(fp->rx_comp_mapping +
3725 BCM_PAGE_SIZE*(i % NUM_RCQ_RINGS)));
3726 nextpg->addr_lo =
3727 cpu_to_le32(U64_LO(fp->rx_comp_mapping +
3728 BCM_PAGE_SIZE*(i % NUM_RCQ_RINGS)));
3732 static int bnx2x_alloc_fp_mem_at(struct bnx2x *bp, int index)
3734 union host_hc_status_block *sb;
3735 struct bnx2x_fastpath *fp = &bp->fp[index];
3736 int ring_size = 0;
3737 u8 cos;
3738 int rx_ring_size = 0;
3740 if (!bp->rx_ring_size &&
3741 (IS_MF_STORAGE_SD(bp) || IS_MF_FCOE_AFEX(bp))) {
3742 rx_ring_size = MIN_RX_SIZE_NONTPA;
3743 bp->rx_ring_size = rx_ring_size;
3744 } else if (!bp->rx_ring_size) {
3745 rx_ring_size = MAX_RX_AVAIL/BNX2X_NUM_RX_QUEUES(bp);
3747 if (CHIP_IS_E3(bp)) {
3748 u32 cfg = SHMEM_RD(bp,
3749 dev_info.port_hw_config[BP_PORT(bp)].
3750 default_cfg);
3752 /* Decrease ring size for 1G functions */
3753 if ((cfg & PORT_HW_CFG_NET_SERDES_IF_MASK) ==
3754 PORT_HW_CFG_NET_SERDES_IF_SGMII)
3755 rx_ring_size /= 10;
3758 /* allocate at least number of buffers required by FW */
3759 rx_ring_size = max_t(int, bp->disable_tpa ? MIN_RX_SIZE_NONTPA :
3760 MIN_RX_SIZE_TPA, rx_ring_size);
3762 bp->rx_ring_size = rx_ring_size;
3763 } else /* if rx_ring_size specified - use it */
3764 rx_ring_size = bp->rx_ring_size;
3766 /* Common */
3767 sb = &bnx2x_fp(bp, index, status_blk);
3769 if (!IS_FCOE_IDX(index)) {
3770 /* status blocks */
3771 if (!CHIP_IS_E1x(bp))
3772 BNX2X_PCI_ALLOC(sb->e2_sb,
3773 &bnx2x_fp(bp, index, status_blk_mapping),
3774 sizeof(struct host_hc_status_block_e2));
3775 else
3776 BNX2X_PCI_ALLOC(sb->e1x_sb,
3777 &bnx2x_fp(bp, index, status_blk_mapping),
3778 sizeof(struct host_hc_status_block_e1x));
3781 /* FCoE Queue uses Default SB and doesn't ACK the SB, thus no need to
3782 * set shortcuts for it.
3784 if (!IS_FCOE_IDX(index))
3785 set_sb_shortcuts(bp, index);
3787 /* Tx */
3788 if (!skip_tx_queue(bp, index)) {
3789 /* fastpath tx rings: tx_buf tx_desc */
3790 for_each_cos_in_tx_queue(fp, cos) {
3791 struct bnx2x_fp_txdata *txdata = fp->txdata_ptr[cos];
3793 DP(NETIF_MSG_IFUP,
3794 "allocating tx memory of fp %d cos %d\n",
3795 index, cos);
3797 BNX2X_ALLOC(txdata->tx_buf_ring,
3798 sizeof(struct sw_tx_bd) * NUM_TX_BD);
3799 BNX2X_PCI_ALLOC(txdata->tx_desc_ring,
3800 &txdata->tx_desc_mapping,
3801 sizeof(union eth_tx_bd_types) * NUM_TX_BD);
3805 /* Rx */
3806 if (!skip_rx_queue(bp, index)) {
3807 /* fastpath rx rings: rx_buf rx_desc rx_comp */
3808 BNX2X_ALLOC(bnx2x_fp(bp, index, rx_buf_ring),
3809 sizeof(struct sw_rx_bd) * NUM_RX_BD);
3810 BNX2X_PCI_ALLOC(bnx2x_fp(bp, index, rx_desc_ring),
3811 &bnx2x_fp(bp, index, rx_desc_mapping),
3812 sizeof(struct eth_rx_bd) * NUM_RX_BD);
3814 BNX2X_PCI_ALLOC(bnx2x_fp(bp, index, rx_comp_ring),
3815 &bnx2x_fp(bp, index, rx_comp_mapping),
3816 sizeof(struct eth_fast_path_rx_cqe) *
3817 NUM_RCQ_BD);
3819 /* SGE ring */
3820 BNX2X_ALLOC(bnx2x_fp(bp, index, rx_page_ring),
3821 sizeof(struct sw_rx_page) * NUM_RX_SGE);
3822 BNX2X_PCI_ALLOC(bnx2x_fp(bp, index, rx_sge_ring),
3823 &bnx2x_fp(bp, index, rx_sge_mapping),
3824 BCM_PAGE_SIZE * NUM_RX_SGE_PAGES);
3825 /* RX BD ring */
3826 bnx2x_set_next_page_rx_bd(fp);
3828 /* CQ ring */
3829 bnx2x_set_next_page_rx_cq(fp);
3831 /* BDs */
3832 ring_size = bnx2x_alloc_rx_bds(fp, rx_ring_size);
3833 if (ring_size < rx_ring_size)
3834 goto alloc_mem_err;
3837 return 0;
3839 /* handles low memory cases */
3840 alloc_mem_err:
3841 BNX2X_ERR("Unable to allocate full memory for queue %d (size %d)\n",
3842 index, ring_size);
3843 /* FW will drop all packets if queue is not big enough,
3844 * In these cases we disable the queue
3845 * Min size is different for OOO, TPA and non-TPA queues
3847 if (ring_size < (fp->disable_tpa ?
3848 MIN_RX_SIZE_NONTPA : MIN_RX_SIZE_TPA)) {
3849 /* release memory allocated for this queue */
3850 bnx2x_free_fp_mem_at(bp, index);
3851 return -ENOMEM;
3853 return 0;
3856 int bnx2x_alloc_fp_mem_cnic(struct bnx2x *bp)
3858 if (!NO_FCOE(bp))
3859 /* FCoE */
3860 if (bnx2x_alloc_fp_mem_at(bp, FCOE_IDX(bp)))
3861 /* we will fail load process instead of mark
3862 * NO_FCOE_FLAG
3864 return -ENOMEM;
3866 return 0;
3869 int bnx2x_alloc_fp_mem(struct bnx2x *bp)
3871 int i;
3873 /* 1. Allocate FP for leading - fatal if error
3874 * 2. Allocate RSS - fix number of queues if error
3877 /* leading */
3878 if (bnx2x_alloc_fp_mem_at(bp, 0))
3879 return -ENOMEM;
3881 /* RSS */
3882 for_each_nondefault_eth_queue(bp, i)
3883 if (bnx2x_alloc_fp_mem_at(bp, i))
3884 break;
3886 /* handle memory failures */
3887 if (i != BNX2X_NUM_ETH_QUEUES(bp)) {
3888 int delta = BNX2X_NUM_ETH_QUEUES(bp) - i;
3890 WARN_ON(delta < 0);
3891 bnx2x_shrink_eth_fp(bp, delta);
3892 if (CNIC_SUPPORT(bp))
3893 /* move non eth FPs next to last eth FP
3894 * must be done in that order
3895 * FCOE_IDX < FWD_IDX < OOO_IDX
3898 /* move FCoE fp even NO_FCOE_FLAG is on */
3899 bnx2x_move_fp(bp, FCOE_IDX(bp), FCOE_IDX(bp) - delta);
3900 bp->num_ethernet_queues -= delta;
3901 bp->num_queues = bp->num_ethernet_queues +
3902 bp->num_cnic_queues;
3903 BNX2X_ERR("Adjusted num of queues from %d to %d\n",
3904 bp->num_queues + delta, bp->num_queues);
3907 return 0;
3910 void bnx2x_free_mem_bp(struct bnx2x *bp)
3912 kfree(bp->fp->tpa_info);
3913 kfree(bp->fp);
3914 kfree(bp->sp_objs);
3915 kfree(bp->fp_stats);
3916 kfree(bp->bnx2x_txq);
3917 kfree(bp->msix_table);
3918 kfree(bp->ilt);
3921 int bnx2x_alloc_mem_bp(struct bnx2x *bp)
3923 struct bnx2x_fastpath *fp;
3924 struct msix_entry *tbl;
3925 struct bnx2x_ilt *ilt;
3926 int msix_table_size = 0;
3927 int fp_array_size, txq_array_size;
3928 int i;
3931 * The biggest MSI-X table we might need is as a maximum number of fast
3932 * path IGU SBs plus default SB (for PF).
3934 msix_table_size = bp->igu_sb_cnt + 1;
3936 /* fp array: RSS plus CNIC related L2 queues */
3937 fp_array_size = BNX2X_MAX_RSS_COUNT(bp) + CNIC_SUPPORT(bp);
3938 BNX2X_DEV_INFO("fp_array_size %d", fp_array_size);
3940 fp = kcalloc(fp_array_size, sizeof(*fp), GFP_KERNEL);
3941 if (!fp)
3942 goto alloc_err;
3943 for (i = 0; i < fp_array_size; i++) {
3944 fp[i].tpa_info =
3945 kcalloc(ETH_MAX_AGGREGATION_QUEUES_E1H_E2,
3946 sizeof(struct bnx2x_agg_info), GFP_KERNEL);
3947 if (!(fp[i].tpa_info))
3948 goto alloc_err;
3951 bp->fp = fp;
3953 /* allocate sp objs */
3954 bp->sp_objs = kcalloc(fp_array_size, sizeof(struct bnx2x_sp_objs),
3955 GFP_KERNEL);
3956 if (!bp->sp_objs)
3957 goto alloc_err;
3959 /* allocate fp_stats */
3960 bp->fp_stats = kcalloc(fp_array_size, sizeof(struct bnx2x_fp_stats),
3961 GFP_KERNEL);
3962 if (!bp->fp_stats)
3963 goto alloc_err;
3965 /* Allocate memory for the transmission queues array */
3966 txq_array_size =
3967 BNX2X_MAX_RSS_COUNT(bp) * BNX2X_MULTI_TX_COS + CNIC_SUPPORT(bp);
3968 BNX2X_DEV_INFO("txq_array_size %d", txq_array_size);
3970 bp->bnx2x_txq = kcalloc(txq_array_size, sizeof(struct bnx2x_fp_txdata),
3971 GFP_KERNEL);
3972 if (!bp->bnx2x_txq)
3973 goto alloc_err;
3975 /* msix table */
3976 tbl = kcalloc(msix_table_size, sizeof(*tbl), GFP_KERNEL);
3977 if (!tbl)
3978 goto alloc_err;
3979 bp->msix_table = tbl;
3981 /* ilt */
3982 ilt = kzalloc(sizeof(*ilt), GFP_KERNEL);
3983 if (!ilt)
3984 goto alloc_err;
3985 bp->ilt = ilt;
3987 return 0;
3988 alloc_err:
3989 bnx2x_free_mem_bp(bp);
3990 return -ENOMEM;
3994 int bnx2x_reload_if_running(struct net_device *dev)
3996 struct bnx2x *bp = netdev_priv(dev);
3998 if (unlikely(!netif_running(dev)))
3999 return 0;
4001 bnx2x_nic_unload(bp, UNLOAD_NORMAL, true);
4002 return bnx2x_nic_load(bp, LOAD_NORMAL);
4005 int bnx2x_get_cur_phy_idx(struct bnx2x *bp)
4007 u32 sel_phy_idx = 0;
4008 if (bp->link_params.num_phys <= 1)
4009 return INT_PHY;
4011 if (bp->link_vars.link_up) {
4012 sel_phy_idx = EXT_PHY1;
4013 /* In case link is SERDES, check if the EXT_PHY2 is the one */
4014 if ((bp->link_vars.link_status & LINK_STATUS_SERDES_LINK) &&
4015 (bp->link_params.phy[EXT_PHY2].supported & SUPPORTED_FIBRE))
4016 sel_phy_idx = EXT_PHY2;
4017 } else {
4019 switch (bnx2x_phy_selection(&bp->link_params)) {
4020 case PORT_HW_CFG_PHY_SELECTION_HARDWARE_DEFAULT:
4021 case PORT_HW_CFG_PHY_SELECTION_FIRST_PHY:
4022 case PORT_HW_CFG_PHY_SELECTION_FIRST_PHY_PRIORITY:
4023 sel_phy_idx = EXT_PHY1;
4024 break;
4025 case PORT_HW_CFG_PHY_SELECTION_SECOND_PHY:
4026 case PORT_HW_CFG_PHY_SELECTION_SECOND_PHY_PRIORITY:
4027 sel_phy_idx = EXT_PHY2;
4028 break;
4032 return sel_phy_idx;
4035 int bnx2x_get_link_cfg_idx(struct bnx2x *bp)
4037 u32 sel_phy_idx = bnx2x_get_cur_phy_idx(bp);
4039 * The selected actived PHY is always after swapping (in case PHY
4040 * swapping is enabled). So when swapping is enabled, we need to reverse
4041 * the configuration
4044 if (bp->link_params.multi_phy_config &
4045 PORT_HW_CFG_PHY_SWAPPED_ENABLED) {
4046 if (sel_phy_idx == EXT_PHY1)
4047 sel_phy_idx = EXT_PHY2;
4048 else if (sel_phy_idx == EXT_PHY2)
4049 sel_phy_idx = EXT_PHY1;
4051 return LINK_CONFIG_IDX(sel_phy_idx);
4054 #ifdef NETDEV_FCOE_WWNN
4055 int bnx2x_fcoe_get_wwn(struct net_device *dev, u64 *wwn, int type)
4057 struct bnx2x *bp = netdev_priv(dev);
4058 struct cnic_eth_dev *cp = &bp->cnic_eth_dev;
4060 switch (type) {
4061 case NETDEV_FCOE_WWNN:
4062 *wwn = HILO_U64(cp->fcoe_wwn_node_name_hi,
4063 cp->fcoe_wwn_node_name_lo);
4064 break;
4065 case NETDEV_FCOE_WWPN:
4066 *wwn = HILO_U64(cp->fcoe_wwn_port_name_hi,
4067 cp->fcoe_wwn_port_name_lo);
4068 break;
4069 default:
4070 BNX2X_ERR("Wrong WWN type requested - %d\n", type);
4071 return -EINVAL;
4074 return 0;
4076 #endif
4078 /* called with rtnl_lock */
4079 int bnx2x_change_mtu(struct net_device *dev, int new_mtu)
4081 struct bnx2x *bp = netdev_priv(dev);
4083 if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
4084 BNX2X_ERR("Can't perform change MTU during parity recovery\n");
4085 return -EAGAIN;
4088 if ((new_mtu > ETH_MAX_JUMBO_PACKET_SIZE) ||
4089 ((new_mtu + ETH_HLEN) < ETH_MIN_PACKET_SIZE)) {
4090 BNX2X_ERR("Can't support requested MTU size\n");
4091 return -EINVAL;
4094 /* This does not race with packet allocation
4095 * because the actual alloc size is
4096 * only updated as part of load
4098 dev->mtu = new_mtu;
4100 return bnx2x_reload_if_running(dev);
4103 netdev_features_t bnx2x_fix_features(struct net_device *dev,
4104 netdev_features_t features)
4106 struct bnx2x *bp = netdev_priv(dev);
4108 /* TPA requires Rx CSUM offloading */
4109 if (!(features & NETIF_F_RXCSUM) || bp->disable_tpa) {
4110 features &= ~NETIF_F_LRO;
4111 features &= ~NETIF_F_GRO;
4114 return features;
4117 int bnx2x_set_features(struct net_device *dev, netdev_features_t features)
4119 struct bnx2x *bp = netdev_priv(dev);
4120 u32 flags = bp->flags;
4121 bool bnx2x_reload = false;
4123 if (features & NETIF_F_LRO)
4124 flags |= TPA_ENABLE_FLAG;
4125 else
4126 flags &= ~TPA_ENABLE_FLAG;
4128 if (features & NETIF_F_GRO)
4129 flags |= GRO_ENABLE_FLAG;
4130 else
4131 flags &= ~GRO_ENABLE_FLAG;
4133 if (features & NETIF_F_LOOPBACK) {
4134 if (bp->link_params.loopback_mode != LOOPBACK_BMAC) {
4135 bp->link_params.loopback_mode = LOOPBACK_BMAC;
4136 bnx2x_reload = true;
4138 } else {
4139 if (bp->link_params.loopback_mode != LOOPBACK_NONE) {
4140 bp->link_params.loopback_mode = LOOPBACK_NONE;
4141 bnx2x_reload = true;
4145 if (flags ^ bp->flags) {
4146 bp->flags = flags;
4147 bnx2x_reload = true;
4150 if (bnx2x_reload) {
4151 if (bp->recovery_state == BNX2X_RECOVERY_DONE)
4152 return bnx2x_reload_if_running(dev);
4153 /* else: bnx2x_nic_load() will be called at end of recovery */
4156 return 0;
4159 void bnx2x_tx_timeout(struct net_device *dev)
4161 struct bnx2x *bp = netdev_priv(dev);
4163 #ifdef BNX2X_STOP_ON_ERROR
4164 if (!bp->panic)
4165 bnx2x_panic();
4166 #endif
4168 smp_mb__before_clear_bit();
4169 set_bit(BNX2X_SP_RTNL_TX_TIMEOUT, &bp->sp_rtnl_state);
4170 smp_mb__after_clear_bit();
4172 /* This allows the netif to be shutdown gracefully before resetting */
4173 schedule_delayed_work(&bp->sp_rtnl_task, 0);
4176 int bnx2x_suspend(struct pci_dev *pdev, pm_message_t state)
4178 struct net_device *dev = pci_get_drvdata(pdev);
4179 struct bnx2x *bp;
4181 if (!dev) {
4182 dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
4183 return -ENODEV;
4185 bp = netdev_priv(dev);
4187 rtnl_lock();
4189 pci_save_state(pdev);
4191 if (!netif_running(dev)) {
4192 rtnl_unlock();
4193 return 0;
4196 netif_device_detach(dev);
4198 bnx2x_nic_unload(bp, UNLOAD_CLOSE, false);
4200 bnx2x_set_power_state(bp, pci_choose_state(pdev, state));
4202 rtnl_unlock();
4204 return 0;
4207 int bnx2x_resume(struct pci_dev *pdev)
4209 struct net_device *dev = pci_get_drvdata(pdev);
4210 struct bnx2x *bp;
4211 int rc;
4213 if (!dev) {
4214 dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
4215 return -ENODEV;
4217 bp = netdev_priv(dev);
4219 if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
4220 BNX2X_ERR("Handling parity error recovery. Try again later\n");
4221 return -EAGAIN;
4224 rtnl_lock();
4226 pci_restore_state(pdev);
4228 if (!netif_running(dev)) {
4229 rtnl_unlock();
4230 return 0;
4233 bnx2x_set_power_state(bp, PCI_D0);
4234 netif_device_attach(dev);
4236 rc = bnx2x_nic_load(bp, LOAD_OPEN);
4238 rtnl_unlock();
4240 return rc;
4244 void bnx2x_set_ctx_validation(struct bnx2x *bp, struct eth_context *cxt,
4245 u32 cid)
4247 /* ustorm cxt validation */
4248 cxt->ustorm_ag_context.cdu_usage =
4249 CDU_RSRVD_VALUE_TYPE_A(HW_CID(bp, cid),
4250 CDU_REGION_NUMBER_UCM_AG, ETH_CONNECTION_TYPE);
4251 /* xcontext validation */
4252 cxt->xstorm_ag_context.cdu_reserved =
4253 CDU_RSRVD_VALUE_TYPE_A(HW_CID(bp, cid),
4254 CDU_REGION_NUMBER_XCM_AG, ETH_CONNECTION_TYPE);
4257 static void storm_memset_hc_timeout(struct bnx2x *bp, u8 port,
4258 u8 fw_sb_id, u8 sb_index,
4259 u8 ticks)
4262 u32 addr = BAR_CSTRORM_INTMEM +
4263 CSTORM_STATUS_BLOCK_DATA_TIMEOUT_OFFSET(fw_sb_id, sb_index);
4264 REG_WR8(bp, addr, ticks);
4265 DP(NETIF_MSG_IFUP,
4266 "port %x fw_sb_id %d sb_index %d ticks %d\n",
4267 port, fw_sb_id, sb_index, ticks);
4270 static void storm_memset_hc_disable(struct bnx2x *bp, u8 port,
4271 u16 fw_sb_id, u8 sb_index,
4272 u8 disable)
4274 u32 enable_flag = disable ? 0 : (1 << HC_INDEX_DATA_HC_ENABLED_SHIFT);
4275 u32 addr = BAR_CSTRORM_INTMEM +
4276 CSTORM_STATUS_BLOCK_DATA_FLAGS_OFFSET(fw_sb_id, sb_index);
4277 u16 flags = REG_RD16(bp, addr);
4278 /* clear and set */
4279 flags &= ~HC_INDEX_DATA_HC_ENABLED;
4280 flags |= enable_flag;
4281 REG_WR16(bp, addr, flags);
4282 DP(NETIF_MSG_IFUP,
4283 "port %x fw_sb_id %d sb_index %d disable %d\n",
4284 port, fw_sb_id, sb_index, disable);
4287 void bnx2x_update_coalesce_sb_index(struct bnx2x *bp, u8 fw_sb_id,
4288 u8 sb_index, u8 disable, u16 usec)
4290 int port = BP_PORT(bp);
4291 u8 ticks = usec / BNX2X_BTR;
4293 storm_memset_hc_timeout(bp, port, fw_sb_id, sb_index, ticks);
4295 disable = disable ? 1 : (usec ? 0 : 1);
4296 storm_memset_hc_disable(bp, port, fw_sb_id, sb_index, disable);