Linux 3.4.102
[linux/fpc-iii.git] / drivers / net / ethernet / broadcom / bnx2x / bnx2x_cmn.c
blob0f05cef534559a000453baedcff6a15742b348c1
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/firmware.h>
27 #include <linux/prefetch.h>
28 #include "bnx2x_cmn.h"
29 #include "bnx2x_init.h"
30 #include "bnx2x_sp.h"
34 /**
35 * bnx2x_move_fp - move content of the fastpath structure.
37 * @bp: driver handle
38 * @from: source FP index
39 * @to: destination FP index
41 * Makes sure the contents of the bp->fp[to].napi is kept
42 * intact. This is done by first copying the napi struct from
43 * the target to the source, and then mem copying the entire
44 * source onto the target
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];
51 /* Copy the NAPI object as it has been already initialized */
52 from_fp->napi = to_fp->napi;
54 /* Move bnx2x_fastpath contents */
55 memcpy(to_fp, from_fp, sizeof(*to_fp));
56 to_fp->index = to;
59 int load_count[2][3] = { {0} }; /* per-path: 0-common, 1-port0, 2-port1 */
61 /* free skb in the packet ring at pos idx
62 * return idx of last bd freed
64 static u16 bnx2x_free_tx_pkt(struct bnx2x *bp, struct bnx2x_fp_txdata *txdata,
65 u16 idx, unsigned int *pkts_compl,
66 unsigned int *bytes_compl)
68 struct sw_tx_bd *tx_buf = &txdata->tx_buf_ring[idx];
69 struct eth_tx_start_bd *tx_start_bd;
70 struct eth_tx_bd *tx_data_bd;
71 struct sk_buff *skb = tx_buf->skb;
72 u16 bd_idx = TX_BD(tx_buf->first_bd), new_cons;
73 int nbd;
74 u16 split_bd_len = 0;
76 /* prefetch skb end pointer to speedup dev_kfree_skb() */
77 prefetch(&skb->end);
79 DP(NETIF_MSG_TX_DONE, "fp[%d]: pkt_idx %d buff @(%p)->skb %p\n",
80 txdata->txq_index, idx, tx_buf, skb);
82 tx_start_bd = &txdata->tx_desc_ring[bd_idx].start_bd;
85 nbd = le16_to_cpu(tx_start_bd->nbd) - 1;
86 #ifdef BNX2X_STOP_ON_ERROR
87 if ((nbd - 1) > (MAX_SKB_FRAGS + 2)) {
88 BNX2X_ERR("BAD nbd!\n");
89 bnx2x_panic();
91 #endif
92 new_cons = nbd + tx_buf->first_bd;
94 /* Get the next bd */
95 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
97 /* Skip a parse bd... */
98 --nbd;
99 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
101 /* TSO headers+data bds share a common mapping. See bnx2x_tx_split() */
102 if (tx_buf->flags & BNX2X_TSO_SPLIT_BD) {
103 tx_data_bd = &txdata->tx_desc_ring[bd_idx].reg_bd;
104 split_bd_len = BD_UNMAP_LEN(tx_data_bd);
105 --nbd;
106 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
109 /* unmap first bd */
110 dma_unmap_single(&bp->pdev->dev, BD_UNMAP_ADDR(tx_start_bd),
111 BD_UNMAP_LEN(tx_start_bd) + split_bd_len,
112 DMA_TO_DEVICE);
114 /* now free frags */
115 while (nbd > 0) {
117 tx_data_bd = &txdata->tx_desc_ring[bd_idx].reg_bd;
118 dma_unmap_page(&bp->pdev->dev, BD_UNMAP_ADDR(tx_data_bd),
119 BD_UNMAP_LEN(tx_data_bd), DMA_TO_DEVICE);
120 if (--nbd)
121 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
124 /* release skb */
125 WARN_ON(!skb);
126 if (likely(skb)) {
127 (*pkts_compl)++;
128 (*bytes_compl) += skb->len;
131 dev_kfree_skb_any(skb);
132 tx_buf->first_bd = 0;
133 tx_buf->skb = NULL;
135 return new_cons;
138 int bnx2x_tx_int(struct bnx2x *bp, struct bnx2x_fp_txdata *txdata)
140 struct netdev_queue *txq;
141 u16 hw_cons, sw_cons, bd_cons = txdata->tx_bd_cons;
142 unsigned int pkts_compl = 0, bytes_compl = 0;
144 #ifdef BNX2X_STOP_ON_ERROR
145 if (unlikely(bp->panic))
146 return -1;
147 #endif
149 txq = netdev_get_tx_queue(bp->dev, txdata->txq_index);
150 hw_cons = le16_to_cpu(*txdata->tx_cons_sb);
151 sw_cons = txdata->tx_pkt_cons;
153 while (sw_cons != hw_cons) {
154 u16 pkt_cons;
156 pkt_cons = TX_BD(sw_cons);
158 DP(NETIF_MSG_TX_DONE,
159 "queue[%d]: hw_cons %u sw_cons %u pkt_cons %u\n",
160 txdata->txq_index, hw_cons, sw_cons, pkt_cons);
162 bd_cons = bnx2x_free_tx_pkt(bp, txdata, pkt_cons,
163 &pkts_compl, &bytes_compl);
165 sw_cons++;
168 netdev_tx_completed_queue(txq, pkts_compl, bytes_compl);
170 txdata->tx_pkt_cons = sw_cons;
171 txdata->tx_bd_cons = bd_cons;
173 /* Need to make the tx_bd_cons update visible to start_xmit()
174 * before checking for netif_tx_queue_stopped(). Without the
175 * memory barrier, there is a small possibility that
176 * start_xmit() will miss it and cause the queue to be stopped
177 * forever.
178 * On the other hand we need an rmb() here to ensure the proper
179 * ordering of bit testing in the following
180 * netif_tx_queue_stopped(txq) call.
182 smp_mb();
184 if (unlikely(netif_tx_queue_stopped(txq))) {
185 /* Taking tx_lock() is needed to prevent reenabling the queue
186 * while it's empty. This could have happen if rx_action() gets
187 * suspended in bnx2x_tx_int() after the condition before
188 * netif_tx_wake_queue(), while tx_action (bnx2x_start_xmit()):
190 * stops the queue->sees fresh tx_bd_cons->releases the queue->
191 * sends some packets consuming the whole queue again->
192 * stops the queue
195 __netif_tx_lock(txq, smp_processor_id());
197 if ((netif_tx_queue_stopped(txq)) &&
198 (bp->state == BNX2X_STATE_OPEN) &&
199 (bnx2x_tx_avail(bp, txdata) >= MAX_SKB_FRAGS + 4))
200 netif_tx_wake_queue(txq);
202 __netif_tx_unlock(txq);
204 return 0;
207 static inline void bnx2x_update_last_max_sge(struct bnx2x_fastpath *fp,
208 u16 idx)
210 u16 last_max = fp->last_max_sge;
212 if (SUB_S16(idx, last_max) > 0)
213 fp->last_max_sge = idx;
216 static inline void bnx2x_update_sge_prod(struct bnx2x_fastpath *fp,
217 u16 sge_len,
218 struct eth_end_agg_rx_cqe *cqe)
220 struct bnx2x *bp = fp->bp;
221 u16 last_max, last_elem, first_elem;
222 u16 delta = 0;
223 u16 i;
225 if (!sge_len)
226 return;
228 /* First mark all used pages */
229 for (i = 0; i < sge_len; i++)
230 BIT_VEC64_CLEAR_BIT(fp->sge_mask,
231 RX_SGE(le16_to_cpu(cqe->sgl_or_raw_data.sgl[i])));
233 DP(NETIF_MSG_RX_STATUS, "fp_cqe->sgl[%d] = %d\n",
234 sge_len - 1, le16_to_cpu(cqe->sgl_or_raw_data.sgl[sge_len - 1]));
236 /* Here we assume that the last SGE index is the biggest */
237 prefetch((void *)(fp->sge_mask));
238 bnx2x_update_last_max_sge(fp,
239 le16_to_cpu(cqe->sgl_or_raw_data.sgl[sge_len - 1]));
241 last_max = RX_SGE(fp->last_max_sge);
242 last_elem = last_max >> BIT_VEC64_ELEM_SHIFT;
243 first_elem = RX_SGE(fp->rx_sge_prod) >> BIT_VEC64_ELEM_SHIFT;
245 /* If ring is not full */
246 if (last_elem + 1 != first_elem)
247 last_elem++;
249 /* Now update the prod */
250 for (i = first_elem; i != last_elem; i = NEXT_SGE_MASK_ELEM(i)) {
251 if (likely(fp->sge_mask[i]))
252 break;
254 fp->sge_mask[i] = BIT_VEC64_ELEM_ONE_MASK;
255 delta += BIT_VEC64_ELEM_SZ;
258 if (delta > 0) {
259 fp->rx_sge_prod += delta;
260 /* clear page-end entries */
261 bnx2x_clear_sge_mask_next_elems(fp);
264 DP(NETIF_MSG_RX_STATUS,
265 "fp->last_max_sge = %d fp->rx_sge_prod = %d\n",
266 fp->last_max_sge, fp->rx_sge_prod);
269 /* Set Toeplitz hash value in the skb using the value from the
270 * CQE (calculated by HW).
272 static u32 bnx2x_get_rxhash(const struct bnx2x *bp,
273 const struct eth_fast_path_rx_cqe *cqe)
275 /* Set Toeplitz hash from CQE */
276 if ((bp->dev->features & NETIF_F_RXHASH) &&
277 (cqe->status_flags & ETH_FAST_PATH_RX_CQE_RSS_HASH_FLG))
278 return le32_to_cpu(cqe->rss_hash_result);
279 return 0;
282 static void bnx2x_tpa_start(struct bnx2x_fastpath *fp, u16 queue,
283 u16 cons, u16 prod,
284 struct eth_fast_path_rx_cqe *cqe)
286 struct bnx2x *bp = fp->bp;
287 struct sw_rx_bd *cons_rx_buf = &fp->rx_buf_ring[cons];
288 struct sw_rx_bd *prod_rx_buf = &fp->rx_buf_ring[prod];
289 struct eth_rx_bd *prod_bd = &fp->rx_desc_ring[prod];
290 dma_addr_t mapping;
291 struct bnx2x_agg_info *tpa_info = &fp->tpa_info[queue];
292 struct sw_rx_bd *first_buf = &tpa_info->first_buf;
294 /* print error if current state != stop */
295 if (tpa_info->tpa_state != BNX2X_TPA_STOP)
296 BNX2X_ERR("start of bin not in stop [%d]\n", queue);
298 /* Try to map an empty data buffer from the aggregation info */
299 mapping = dma_map_single(&bp->pdev->dev,
300 first_buf->data + NET_SKB_PAD,
301 fp->rx_buf_size, DMA_FROM_DEVICE);
303 * ...if it fails - move the skb from the consumer to the producer
304 * and set the current aggregation state as ERROR to drop it
305 * when TPA_STOP arrives.
308 if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
309 /* Move the BD from the consumer to the producer */
310 bnx2x_reuse_rx_data(fp, cons, prod);
311 tpa_info->tpa_state = BNX2X_TPA_ERROR;
312 return;
315 /* move empty data from pool to prod */
316 prod_rx_buf->data = first_buf->data;
317 dma_unmap_addr_set(prod_rx_buf, mapping, mapping);
318 /* point prod_bd to new data */
319 prod_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
320 prod_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
322 /* move partial skb from cons to pool (don't unmap yet) */
323 *first_buf = *cons_rx_buf;
325 /* mark bin state as START */
326 tpa_info->parsing_flags =
327 le16_to_cpu(cqe->pars_flags.flags);
328 tpa_info->vlan_tag = le16_to_cpu(cqe->vlan_tag);
329 tpa_info->tpa_state = BNX2X_TPA_START;
330 tpa_info->len_on_bd = le16_to_cpu(cqe->len_on_bd);
331 tpa_info->placement_offset = cqe->placement_offset;
332 tpa_info->rxhash = bnx2x_get_rxhash(bp, cqe);
333 if (fp->mode == TPA_MODE_GRO) {
334 u16 gro_size = le16_to_cpu(cqe->pkt_len_or_gro_seg_len);
335 tpa_info->full_page =
336 SGE_PAGE_SIZE * PAGES_PER_SGE / gro_size * gro_size;
338 * FW 7.2.16 BUG workaround:
339 * if SGE size is (exactly) multiple gro_size
340 * fw will place one less frag on SGE.
341 * the calculation is done only for potentially
342 * dangerous MTUs.
344 if (unlikely(bp->gro_check))
345 if (!(SGE_PAGE_SIZE * PAGES_PER_SGE % gro_size))
346 tpa_info->full_page -= gro_size;
347 tpa_info->gro_size = gro_size;
350 #ifdef BNX2X_STOP_ON_ERROR
351 fp->tpa_queue_used |= (1 << queue);
352 #ifdef _ASM_GENERIC_INT_L64_H
353 DP(NETIF_MSG_RX_STATUS, "fp->tpa_queue_used = 0x%lx\n",
354 #else
355 DP(NETIF_MSG_RX_STATUS, "fp->tpa_queue_used = 0x%llx\n",
356 #endif
357 fp->tpa_queue_used);
358 #endif
361 /* Timestamp option length allowed for TPA aggregation:
363 * nop nop kind length echo val
365 #define TPA_TSTAMP_OPT_LEN 12
367 * bnx2x_set_lro_mss - calculate the approximate value of the MSS
369 * @bp: driver handle
370 * @parsing_flags: parsing flags from the START CQE
371 * @len_on_bd: total length of the first packet for the
372 * aggregation.
374 * Approximate value of the MSS for this aggregation calculated using
375 * the first packet of it.
377 static inline u16 bnx2x_set_lro_mss(struct bnx2x *bp, u16 parsing_flags,
378 u16 len_on_bd)
381 * TPA arrgregation won't have either IP options or TCP options
382 * other than timestamp or IPv6 extension headers.
384 u16 hdrs_len = ETH_HLEN + sizeof(struct tcphdr);
386 if (GET_FLAG(parsing_flags, PARSING_FLAGS_OVER_ETHERNET_PROTOCOL) ==
387 PRS_FLAG_OVERETH_IPV6)
388 hdrs_len += sizeof(struct ipv6hdr);
389 else /* IPv4 */
390 hdrs_len += sizeof(struct iphdr);
393 /* Check if there was a TCP timestamp, if there is it's will
394 * always be 12 bytes length: nop nop kind length echo val.
396 * Otherwise FW would close the aggregation.
398 if (parsing_flags & PARSING_FLAGS_TIME_STAMP_EXIST_FLAG)
399 hdrs_len += TPA_TSTAMP_OPT_LEN;
401 return len_on_bd - hdrs_len;
404 static int bnx2x_fill_frag_skb(struct bnx2x *bp, struct bnx2x_fastpath *fp,
405 struct bnx2x_agg_info *tpa_info,
406 u16 pages,
407 struct sk_buff *skb,
408 struct eth_end_agg_rx_cqe *cqe,
409 u16 cqe_idx)
411 struct sw_rx_page *rx_pg, old_rx_pg;
412 u32 i, frag_len, frag_size;
413 int err, j, frag_id = 0;
414 u16 len_on_bd = tpa_info->len_on_bd;
415 u16 full_page = 0, gro_size = 0;
417 frag_size = le16_to_cpu(cqe->pkt_len) - len_on_bd;
419 if (fp->mode == TPA_MODE_GRO) {
420 gro_size = tpa_info->gro_size;
421 full_page = tpa_info->full_page;
424 /* This is needed in order to enable forwarding support */
425 if (frag_size) {
426 skb_shinfo(skb)->gso_size = bnx2x_set_lro_mss(bp,
427 tpa_info->parsing_flags, len_on_bd);
429 /* set for GRO */
430 if (fp->mode == TPA_MODE_GRO)
431 skb_shinfo(skb)->gso_type =
432 (GET_FLAG(tpa_info->parsing_flags,
433 PARSING_FLAGS_OVER_ETHERNET_PROTOCOL) ==
434 PRS_FLAG_OVERETH_IPV6) ?
435 SKB_GSO_TCPV6 : SKB_GSO_TCPV4;
439 #ifdef BNX2X_STOP_ON_ERROR
440 if (pages > min_t(u32, 8, MAX_SKB_FRAGS)*SGE_PAGE_SIZE*PAGES_PER_SGE) {
441 BNX2X_ERR("SGL length is too long: %d. CQE index is %d\n",
442 pages, cqe_idx);
443 BNX2X_ERR("cqe->pkt_len = %d\n", cqe->pkt_len);
444 bnx2x_panic();
445 return -EINVAL;
447 #endif
449 /* Run through the SGL and compose the fragmented skb */
450 for (i = 0, j = 0; i < pages; i += PAGES_PER_SGE, j++) {
451 u16 sge_idx = RX_SGE(le16_to_cpu(cqe->sgl_or_raw_data.sgl[j]));
453 /* FW gives the indices of the SGE as if the ring is an array
454 (meaning that "next" element will consume 2 indices) */
455 if (fp->mode == TPA_MODE_GRO)
456 frag_len = min_t(u32, frag_size, (u32)full_page);
457 else /* LRO */
458 frag_len = min_t(u32, frag_size,
459 (u32)(SGE_PAGE_SIZE * PAGES_PER_SGE));
461 rx_pg = &fp->rx_page_ring[sge_idx];
462 old_rx_pg = *rx_pg;
464 /* If we fail to allocate a substitute page, we simply stop
465 where we are and drop the whole packet */
466 err = bnx2x_alloc_rx_sge(bp, fp, sge_idx);
467 if (unlikely(err)) {
468 fp->eth_q_stats.rx_skb_alloc_failed++;
469 return err;
472 /* Unmap the page as we r going to pass it to the stack */
473 dma_unmap_page(&bp->pdev->dev,
474 dma_unmap_addr(&old_rx_pg, mapping),
475 SGE_PAGE_SIZE*PAGES_PER_SGE, DMA_FROM_DEVICE);
476 /* Add one frag and update the appropriate fields in the skb */
477 if (fp->mode == TPA_MODE_LRO)
478 skb_fill_page_desc(skb, j, old_rx_pg.page, 0, frag_len);
479 else { /* GRO */
480 int rem;
481 int offset = 0;
482 for (rem = frag_len; rem > 0; rem -= gro_size) {
483 int len = rem > gro_size ? gro_size : rem;
484 skb_fill_page_desc(skb, frag_id++,
485 old_rx_pg.page, offset, len);
486 if (offset)
487 get_page(old_rx_pg.page);
488 offset += len;
492 skb->data_len += frag_len;
493 skb->truesize += SGE_PAGE_SIZE * PAGES_PER_SGE;
494 skb->len += frag_len;
496 frag_size -= frag_len;
499 return 0;
502 static inline void bnx2x_tpa_stop(struct bnx2x *bp, struct bnx2x_fastpath *fp,
503 struct bnx2x_agg_info *tpa_info,
504 u16 pages,
505 struct eth_end_agg_rx_cqe *cqe,
506 u16 cqe_idx)
508 struct sw_rx_bd *rx_buf = &tpa_info->first_buf;
509 u8 pad = tpa_info->placement_offset;
510 u16 len = tpa_info->len_on_bd;
511 struct sk_buff *skb = NULL;
512 u8 *new_data, *data = rx_buf->data;
513 u8 old_tpa_state = tpa_info->tpa_state;
515 tpa_info->tpa_state = BNX2X_TPA_STOP;
517 /* If we there was an error during the handling of the TPA_START -
518 * drop this aggregation.
520 if (old_tpa_state == BNX2X_TPA_ERROR)
521 goto drop;
523 /* Try to allocate the new data */
524 new_data = kmalloc(fp->rx_buf_size + NET_SKB_PAD, GFP_ATOMIC);
526 /* Unmap skb in the pool anyway, as we are going to change
527 pool entry status to BNX2X_TPA_STOP even if new skb allocation
528 fails. */
529 dma_unmap_single(&bp->pdev->dev, dma_unmap_addr(rx_buf, mapping),
530 fp->rx_buf_size, DMA_FROM_DEVICE);
531 if (likely(new_data))
532 skb = build_skb(data);
534 if (likely(skb)) {
535 #ifdef BNX2X_STOP_ON_ERROR
536 if (pad + len > fp->rx_buf_size) {
537 BNX2X_ERR("skb_put is about to fail... pad %d len %d rx_buf_size %d\n",
538 pad, len, fp->rx_buf_size);
539 bnx2x_panic();
540 return;
542 #endif
544 skb_reserve(skb, pad + NET_SKB_PAD);
545 skb_put(skb, len);
546 skb->rxhash = tpa_info->rxhash;
548 skb->protocol = eth_type_trans(skb, bp->dev);
549 skb->ip_summed = CHECKSUM_UNNECESSARY;
551 if (!bnx2x_fill_frag_skb(bp, fp, tpa_info, pages,
552 skb, cqe, cqe_idx)) {
553 if (tpa_info->parsing_flags & PARSING_FLAGS_VLAN)
554 __vlan_hwaccel_put_tag(skb, tpa_info->vlan_tag);
555 skb_record_rx_queue(skb, fp->rx_queue);
556 napi_gro_receive(&fp->napi, skb);
557 } else {
558 DP(NETIF_MSG_RX_STATUS,
559 "Failed to allocate new pages - dropping packet!\n");
560 dev_kfree_skb_any(skb);
564 /* put new data in bin */
565 rx_buf->data = new_data;
567 return;
569 kfree(new_data);
570 drop:
571 /* drop the packet and keep the buffer in the bin */
572 DP(NETIF_MSG_RX_STATUS,
573 "Failed to allocate or map a new skb - dropping packet!\n");
574 fp->eth_q_stats.rx_skb_alloc_failed++;
577 static void bnx2x_csum_validate(struct sk_buff *skb, union eth_rx_cqe *cqe,
578 struct bnx2x_fastpath *fp)
580 /* Do nothing if no L4 csum validation was done.
581 * We do not check whether IP csum was validated. For IPv4 we assume
582 * that if the card got as far as validating the L4 csum, it also
583 * validated the IP csum. IPv6 has no IP csum.
585 if (cqe->fast_path_cqe.status_flags &
586 ETH_FAST_PATH_RX_CQE_L4_XSUM_NO_VALIDATION_FLG)
587 return;
589 /* If L4 validation was done, check if an error was found. */
591 if (cqe->fast_path_cqe.type_error_flags &
592 (ETH_FAST_PATH_RX_CQE_IP_BAD_XSUM_FLG |
593 ETH_FAST_PATH_RX_CQE_L4_BAD_XSUM_FLG))
594 fp->eth_q_stats.hw_csum_err++;
595 else
596 skb->ip_summed = CHECKSUM_UNNECESSARY;
599 int bnx2x_rx_int(struct bnx2x_fastpath *fp, int budget)
601 struct bnx2x *bp = fp->bp;
602 u16 bd_cons, bd_prod, bd_prod_fw, comp_ring_cons;
603 u16 hw_comp_cons, sw_comp_cons, sw_comp_prod;
604 int rx_pkt = 0;
606 #ifdef BNX2X_STOP_ON_ERROR
607 if (unlikely(bp->panic))
608 return 0;
609 #endif
611 /* CQ "next element" is of the size of the regular element,
612 that's why it's ok here */
613 hw_comp_cons = le16_to_cpu(*fp->rx_cons_sb);
614 if ((hw_comp_cons & MAX_RCQ_DESC_CNT) == MAX_RCQ_DESC_CNT)
615 hw_comp_cons++;
617 bd_cons = fp->rx_bd_cons;
618 bd_prod = fp->rx_bd_prod;
619 bd_prod_fw = bd_prod;
620 sw_comp_cons = fp->rx_comp_cons;
621 sw_comp_prod = fp->rx_comp_prod;
623 /* Memory barrier necessary as speculative reads of the rx
624 * buffer can be ahead of the index in the status block
626 rmb();
628 DP(NETIF_MSG_RX_STATUS,
629 "queue[%d]: hw_comp_cons %u sw_comp_cons %u\n",
630 fp->index, hw_comp_cons, sw_comp_cons);
632 while (sw_comp_cons != hw_comp_cons) {
633 struct sw_rx_bd *rx_buf = NULL;
634 struct sk_buff *skb;
635 union eth_rx_cqe *cqe;
636 struct eth_fast_path_rx_cqe *cqe_fp;
637 u8 cqe_fp_flags;
638 enum eth_rx_cqe_type cqe_fp_type;
639 u16 len, pad, queue;
640 u8 *data;
642 #ifdef BNX2X_STOP_ON_ERROR
643 if (unlikely(bp->panic))
644 return 0;
645 #endif
647 comp_ring_cons = RCQ_BD(sw_comp_cons);
648 bd_prod = RX_BD(bd_prod);
649 bd_cons = RX_BD(bd_cons);
651 cqe = &fp->rx_comp_ring[comp_ring_cons];
652 cqe_fp = &cqe->fast_path_cqe;
653 cqe_fp_flags = cqe_fp->type_error_flags;
654 cqe_fp_type = cqe_fp_flags & ETH_FAST_PATH_RX_CQE_TYPE;
656 DP(NETIF_MSG_RX_STATUS,
657 "CQE type %x err %x status %x queue %x vlan %x len %u\n",
658 CQE_TYPE(cqe_fp_flags),
659 cqe_fp_flags, cqe_fp->status_flags,
660 le32_to_cpu(cqe_fp->rss_hash_result),
661 le16_to_cpu(cqe_fp->vlan_tag),
662 le16_to_cpu(cqe_fp->pkt_len_or_gro_seg_len));
664 /* is this a slowpath msg? */
665 if (unlikely(CQE_TYPE_SLOW(cqe_fp_type))) {
666 bnx2x_sp_event(fp, cqe);
667 goto next_cqe;
670 rx_buf = &fp->rx_buf_ring[bd_cons];
671 data = rx_buf->data;
673 if (!CQE_TYPE_FAST(cqe_fp_type)) {
674 struct bnx2x_agg_info *tpa_info;
675 u16 frag_size, pages;
676 #ifdef BNX2X_STOP_ON_ERROR
677 /* sanity check */
678 if (fp->disable_tpa &&
679 (CQE_TYPE_START(cqe_fp_type) ||
680 CQE_TYPE_STOP(cqe_fp_type)))
681 BNX2X_ERR("START/STOP packet while disable_tpa type %x\n",
682 CQE_TYPE(cqe_fp_type));
683 #endif
685 if (CQE_TYPE_START(cqe_fp_type)) {
686 u16 queue = cqe_fp->queue_index;
687 DP(NETIF_MSG_RX_STATUS,
688 "calling tpa_start on queue %d\n",
689 queue);
691 bnx2x_tpa_start(fp, queue,
692 bd_cons, bd_prod,
693 cqe_fp);
695 goto next_rx;
698 queue = cqe->end_agg_cqe.queue_index;
699 tpa_info = &fp->tpa_info[queue];
700 DP(NETIF_MSG_RX_STATUS,
701 "calling tpa_stop on queue %d\n",
702 queue);
704 frag_size = le16_to_cpu(cqe->end_agg_cqe.pkt_len) -
705 tpa_info->len_on_bd;
707 if (fp->mode == TPA_MODE_GRO)
708 pages = (frag_size + tpa_info->full_page - 1) /
709 tpa_info->full_page;
710 else
711 pages = SGE_PAGE_ALIGN(frag_size) >>
712 SGE_PAGE_SHIFT;
714 bnx2x_tpa_stop(bp, fp, tpa_info, pages,
715 &cqe->end_agg_cqe, comp_ring_cons);
716 #ifdef BNX2X_STOP_ON_ERROR
717 if (bp->panic)
718 return 0;
719 #endif
721 bnx2x_update_sge_prod(fp, pages, &cqe->end_agg_cqe);
722 goto next_cqe;
724 /* non TPA */
725 len = le16_to_cpu(cqe_fp->pkt_len_or_gro_seg_len);
726 pad = cqe_fp->placement_offset;
727 dma_sync_single_for_cpu(&bp->pdev->dev,
728 dma_unmap_addr(rx_buf, mapping),
729 pad + RX_COPY_THRESH,
730 DMA_FROM_DEVICE);
731 pad += NET_SKB_PAD;
732 prefetch(data + pad); /* speedup eth_type_trans() */
733 /* is this an error packet? */
734 if (unlikely(cqe_fp_flags & ETH_RX_ERROR_FALGS)) {
735 DP(NETIF_MSG_RX_ERR | NETIF_MSG_RX_STATUS,
736 "ERROR flags %x rx packet %u\n",
737 cqe_fp_flags, sw_comp_cons);
738 fp->eth_q_stats.rx_err_discard_pkt++;
739 goto reuse_rx;
742 /* Since we don't have a jumbo ring
743 * copy small packets if mtu > 1500
745 if ((bp->dev->mtu > ETH_MAX_PACKET_SIZE) &&
746 (len <= RX_COPY_THRESH)) {
747 skb = netdev_alloc_skb_ip_align(bp->dev, len);
748 if (skb == NULL) {
749 DP(NETIF_MSG_RX_ERR | NETIF_MSG_RX_STATUS,
750 "ERROR packet dropped because of alloc failure\n");
751 fp->eth_q_stats.rx_skb_alloc_failed++;
752 goto reuse_rx;
754 memcpy(skb->data, data + pad, len);
755 bnx2x_reuse_rx_data(fp, bd_cons, bd_prod);
756 } else {
757 if (likely(bnx2x_alloc_rx_data(bp, fp, bd_prod) == 0)) {
758 dma_unmap_single(&bp->pdev->dev,
759 dma_unmap_addr(rx_buf, mapping),
760 fp->rx_buf_size,
761 DMA_FROM_DEVICE);
762 skb = build_skb(data);
763 if (unlikely(!skb)) {
764 kfree(data);
765 fp->eth_q_stats.rx_skb_alloc_failed++;
766 goto next_rx;
768 skb_reserve(skb, pad);
769 } else {
770 DP(NETIF_MSG_RX_ERR | NETIF_MSG_RX_STATUS,
771 "ERROR packet dropped because of alloc failure\n");
772 fp->eth_q_stats.rx_skb_alloc_failed++;
773 reuse_rx:
774 bnx2x_reuse_rx_data(fp, bd_cons, bd_prod);
775 goto next_rx;
779 skb_put(skb, len);
780 skb->protocol = eth_type_trans(skb, bp->dev);
782 /* Set Toeplitz hash for a none-LRO skb */
783 skb->rxhash = bnx2x_get_rxhash(bp, cqe_fp);
785 skb_checksum_none_assert(skb);
787 if (bp->dev->features & NETIF_F_RXCSUM)
788 bnx2x_csum_validate(skb, cqe, fp);
791 skb_record_rx_queue(skb, fp->rx_queue);
793 if (le16_to_cpu(cqe_fp->pars_flags.flags) &
794 PARSING_FLAGS_VLAN)
795 __vlan_hwaccel_put_tag(skb,
796 le16_to_cpu(cqe_fp->vlan_tag));
797 napi_gro_receive(&fp->napi, skb);
800 next_rx:
801 rx_buf->data = NULL;
803 bd_cons = NEXT_RX_IDX(bd_cons);
804 bd_prod = NEXT_RX_IDX(bd_prod);
805 bd_prod_fw = NEXT_RX_IDX(bd_prod_fw);
806 rx_pkt++;
807 next_cqe:
808 sw_comp_prod = NEXT_RCQ_IDX(sw_comp_prod);
809 sw_comp_cons = NEXT_RCQ_IDX(sw_comp_cons);
811 if (rx_pkt == budget)
812 break;
813 } /* while */
815 fp->rx_bd_cons = bd_cons;
816 fp->rx_bd_prod = bd_prod_fw;
817 fp->rx_comp_cons = sw_comp_cons;
818 fp->rx_comp_prod = sw_comp_prod;
820 /* Update producers */
821 bnx2x_update_rx_prod(bp, fp, bd_prod_fw, sw_comp_prod,
822 fp->rx_sge_prod);
824 fp->rx_pkt += rx_pkt;
825 fp->rx_calls++;
827 return rx_pkt;
830 static irqreturn_t bnx2x_msix_fp_int(int irq, void *fp_cookie)
832 struct bnx2x_fastpath *fp = fp_cookie;
833 struct bnx2x *bp = fp->bp;
834 u8 cos;
836 DP(NETIF_MSG_INTR,
837 "got an MSI-X interrupt on IDX:SB [fp %d fw_sd %d igusb %d]\n",
838 fp->index, fp->fw_sb_id, fp->igu_sb_id);
839 bnx2x_ack_sb(bp, fp->igu_sb_id, USTORM_ID, 0, IGU_INT_DISABLE, 0);
841 #ifdef BNX2X_STOP_ON_ERROR
842 if (unlikely(bp->panic))
843 return IRQ_HANDLED;
844 #endif
846 /* Handle Rx and Tx according to MSI-X vector */
847 prefetch(fp->rx_cons_sb);
849 for_each_cos_in_tx_queue(fp, cos)
850 prefetch(fp->txdata[cos].tx_cons_sb);
852 prefetch(&fp->sb_running_index[SM_RX_ID]);
853 napi_schedule(&bnx2x_fp(bp, fp->index, napi));
855 return IRQ_HANDLED;
858 /* HW Lock for shared dual port PHYs */
859 void bnx2x_acquire_phy_lock(struct bnx2x *bp)
861 mutex_lock(&bp->port.phy_mutex);
863 if (bp->port.need_hw_lock)
864 bnx2x_acquire_hw_lock(bp, HW_LOCK_RESOURCE_MDIO);
867 void bnx2x_release_phy_lock(struct bnx2x *bp)
869 if (bp->port.need_hw_lock)
870 bnx2x_release_hw_lock(bp, HW_LOCK_RESOURCE_MDIO);
872 mutex_unlock(&bp->port.phy_mutex);
875 /* calculates MF speed according to current linespeed and MF configuration */
876 u16 bnx2x_get_mf_speed(struct bnx2x *bp)
878 u16 line_speed = bp->link_vars.line_speed;
879 if (IS_MF(bp)) {
880 u16 maxCfg = bnx2x_extract_max_cfg(bp,
881 bp->mf_config[BP_VN(bp)]);
883 /* Calculate the current MAX line speed limit for the MF
884 * devices
886 if (IS_MF_SI(bp))
887 line_speed = (line_speed * maxCfg) / 100;
888 else { /* SD mode */
889 u16 vn_max_rate = maxCfg * 100;
891 if (vn_max_rate < line_speed)
892 line_speed = vn_max_rate;
896 return line_speed;
900 * bnx2x_fill_report_data - fill link report data to report
902 * @bp: driver handle
903 * @data: link state to update
905 * It uses a none-atomic bit operations because is called under the mutex.
907 static inline void bnx2x_fill_report_data(struct bnx2x *bp,
908 struct bnx2x_link_report_data *data)
910 u16 line_speed = bnx2x_get_mf_speed(bp);
912 memset(data, 0, sizeof(*data));
914 /* Fill the report data: efective line speed */
915 data->line_speed = line_speed;
917 /* Link is down */
918 if (!bp->link_vars.link_up || (bp->flags & MF_FUNC_DIS))
919 __set_bit(BNX2X_LINK_REPORT_LINK_DOWN,
920 &data->link_report_flags);
922 /* Full DUPLEX */
923 if (bp->link_vars.duplex == DUPLEX_FULL)
924 __set_bit(BNX2X_LINK_REPORT_FD, &data->link_report_flags);
926 /* Rx Flow Control is ON */
927 if (bp->link_vars.flow_ctrl & BNX2X_FLOW_CTRL_RX)
928 __set_bit(BNX2X_LINK_REPORT_RX_FC_ON, &data->link_report_flags);
930 /* Tx Flow Control is ON */
931 if (bp->link_vars.flow_ctrl & BNX2X_FLOW_CTRL_TX)
932 __set_bit(BNX2X_LINK_REPORT_TX_FC_ON, &data->link_report_flags);
936 * bnx2x_link_report - report link status to OS.
938 * @bp: driver handle
940 * Calls the __bnx2x_link_report() under the same locking scheme
941 * as a link/PHY state managing code to ensure a consistent link
942 * reporting.
945 void bnx2x_link_report(struct bnx2x *bp)
947 bnx2x_acquire_phy_lock(bp);
948 __bnx2x_link_report(bp);
949 bnx2x_release_phy_lock(bp);
953 * __bnx2x_link_report - report link status to OS.
955 * @bp: driver handle
957 * None atomic inmlementation.
958 * Should be called under the phy_lock.
960 void __bnx2x_link_report(struct bnx2x *bp)
962 struct bnx2x_link_report_data cur_data;
964 /* reread mf_cfg */
965 if (!CHIP_IS_E1(bp))
966 bnx2x_read_mf_cfg(bp);
968 /* Read the current link report info */
969 bnx2x_fill_report_data(bp, &cur_data);
971 /* Don't report link down or exactly the same link status twice */
972 if (!memcmp(&cur_data, &bp->last_reported_link, sizeof(cur_data)) ||
973 (test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
974 &bp->last_reported_link.link_report_flags) &&
975 test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
976 &cur_data.link_report_flags)))
977 return;
979 bp->link_cnt++;
981 /* We are going to report a new link parameters now -
982 * remember the current data for the next time.
984 memcpy(&bp->last_reported_link, &cur_data, sizeof(cur_data));
986 if (test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
987 &cur_data.link_report_flags)) {
988 netif_carrier_off(bp->dev);
989 netdev_err(bp->dev, "NIC Link is Down\n");
990 return;
991 } else {
992 const char *duplex;
993 const char *flow;
995 netif_carrier_on(bp->dev);
997 if (test_and_clear_bit(BNX2X_LINK_REPORT_FD,
998 &cur_data.link_report_flags))
999 duplex = "full";
1000 else
1001 duplex = "half";
1003 /* Handle the FC at the end so that only these flags would be
1004 * possibly set. This way we may easily check if there is no FC
1005 * enabled.
1007 if (cur_data.link_report_flags) {
1008 if (test_bit(BNX2X_LINK_REPORT_RX_FC_ON,
1009 &cur_data.link_report_flags)) {
1010 if (test_bit(BNX2X_LINK_REPORT_TX_FC_ON,
1011 &cur_data.link_report_flags))
1012 flow = "ON - receive & transmit";
1013 else
1014 flow = "ON - receive";
1015 } else {
1016 flow = "ON - transmit";
1018 } else {
1019 flow = "none";
1021 netdev_info(bp->dev, "NIC Link is Up, %d Mbps %s duplex, Flow control: %s\n",
1022 cur_data.line_speed, duplex, flow);
1026 void bnx2x_init_rx_rings(struct bnx2x *bp)
1028 int func = BP_FUNC(bp);
1029 u16 ring_prod;
1030 int i, j;
1032 /* Allocate TPA resources */
1033 for_each_rx_queue(bp, j) {
1034 struct bnx2x_fastpath *fp = &bp->fp[j];
1036 DP(NETIF_MSG_IFUP,
1037 "mtu %d rx_buf_size %d\n", bp->dev->mtu, fp->rx_buf_size);
1039 if (!fp->disable_tpa) {
1040 /* Fill the per-aggregtion pool */
1041 for (i = 0; i < MAX_AGG_QS(bp); i++) {
1042 struct bnx2x_agg_info *tpa_info =
1043 &fp->tpa_info[i];
1044 struct sw_rx_bd *first_buf =
1045 &tpa_info->first_buf;
1047 first_buf->data = kmalloc(fp->rx_buf_size + NET_SKB_PAD,
1048 GFP_ATOMIC);
1049 if (!first_buf->data) {
1050 BNX2X_ERR("Failed to allocate TPA skb pool for queue[%d] - disabling TPA on this queue!\n",
1052 bnx2x_free_tpa_pool(bp, fp, i);
1053 fp->disable_tpa = 1;
1054 break;
1056 dma_unmap_addr_set(first_buf, mapping, 0);
1057 tpa_info->tpa_state = BNX2X_TPA_STOP;
1060 /* "next page" elements initialization */
1061 bnx2x_set_next_page_sgl(fp);
1063 /* set SGEs bit mask */
1064 bnx2x_init_sge_ring_bit_mask(fp);
1066 /* Allocate SGEs and initialize the ring elements */
1067 for (i = 0, ring_prod = 0;
1068 i < MAX_RX_SGE_CNT*NUM_RX_SGE_PAGES; i++) {
1070 if (bnx2x_alloc_rx_sge(bp, fp, ring_prod) < 0) {
1071 BNX2X_ERR("was only able to allocate %d rx sges\n",
1073 BNX2X_ERR("disabling TPA for queue[%d]\n",
1075 /* Cleanup already allocated elements */
1076 bnx2x_free_rx_sge_range(bp, fp,
1077 ring_prod);
1078 bnx2x_free_tpa_pool(bp, fp,
1079 MAX_AGG_QS(bp));
1080 fp->disable_tpa = 1;
1081 ring_prod = 0;
1082 break;
1084 ring_prod = NEXT_SGE_IDX(ring_prod);
1087 fp->rx_sge_prod = ring_prod;
1091 for_each_rx_queue(bp, j) {
1092 struct bnx2x_fastpath *fp = &bp->fp[j];
1094 fp->rx_bd_cons = 0;
1096 /* Activate BD ring */
1097 /* Warning!
1098 * this will generate an interrupt (to the TSTORM)
1099 * must only be done after chip is initialized
1101 bnx2x_update_rx_prod(bp, fp, fp->rx_bd_prod, fp->rx_comp_prod,
1102 fp->rx_sge_prod);
1104 if (j != 0)
1105 continue;
1107 if (CHIP_IS_E1(bp)) {
1108 REG_WR(bp, BAR_USTRORM_INTMEM +
1109 USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func),
1110 U64_LO(fp->rx_comp_mapping));
1111 REG_WR(bp, BAR_USTRORM_INTMEM +
1112 USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func) + 4,
1113 U64_HI(fp->rx_comp_mapping));
1118 static void bnx2x_free_tx_skbs(struct bnx2x *bp)
1120 int i;
1121 u8 cos;
1123 for_each_tx_queue(bp, i) {
1124 struct bnx2x_fastpath *fp = &bp->fp[i];
1125 for_each_cos_in_tx_queue(fp, cos) {
1126 struct bnx2x_fp_txdata *txdata = &fp->txdata[cos];
1127 unsigned pkts_compl = 0, bytes_compl = 0;
1129 u16 sw_prod = txdata->tx_pkt_prod;
1130 u16 sw_cons = txdata->tx_pkt_cons;
1132 while (sw_cons != sw_prod) {
1133 bnx2x_free_tx_pkt(bp, txdata, TX_BD(sw_cons),
1134 &pkts_compl, &bytes_compl);
1135 sw_cons++;
1137 netdev_tx_reset_queue(
1138 netdev_get_tx_queue(bp->dev, txdata->txq_index));
1143 static void bnx2x_free_rx_bds(struct bnx2x_fastpath *fp)
1145 struct bnx2x *bp = fp->bp;
1146 int i;
1148 /* ring wasn't allocated */
1149 if (fp->rx_buf_ring == NULL)
1150 return;
1152 for (i = 0; i < NUM_RX_BD; i++) {
1153 struct sw_rx_bd *rx_buf = &fp->rx_buf_ring[i];
1154 u8 *data = rx_buf->data;
1156 if (data == NULL)
1157 continue;
1158 dma_unmap_single(&bp->pdev->dev,
1159 dma_unmap_addr(rx_buf, mapping),
1160 fp->rx_buf_size, DMA_FROM_DEVICE);
1162 rx_buf->data = NULL;
1163 kfree(data);
1167 static void bnx2x_free_rx_skbs(struct bnx2x *bp)
1169 int j;
1171 for_each_rx_queue(bp, j) {
1172 struct bnx2x_fastpath *fp = &bp->fp[j];
1174 bnx2x_free_rx_bds(fp);
1176 if (!fp->disable_tpa)
1177 bnx2x_free_tpa_pool(bp, fp, MAX_AGG_QS(bp));
1181 void bnx2x_free_skbs(struct bnx2x *bp)
1183 bnx2x_free_tx_skbs(bp);
1184 bnx2x_free_rx_skbs(bp);
1187 void bnx2x_update_max_mf_config(struct bnx2x *bp, u32 value)
1189 /* load old values */
1190 u32 mf_cfg = bp->mf_config[BP_VN(bp)];
1192 if (value != bnx2x_extract_max_cfg(bp, mf_cfg)) {
1193 /* leave all but MAX value */
1194 mf_cfg &= ~FUNC_MF_CFG_MAX_BW_MASK;
1196 /* set new MAX value */
1197 mf_cfg |= (value << FUNC_MF_CFG_MAX_BW_SHIFT)
1198 & FUNC_MF_CFG_MAX_BW_MASK;
1200 bnx2x_fw_command(bp, DRV_MSG_CODE_SET_MF_BW, mf_cfg);
1205 * bnx2x_free_msix_irqs - free previously requested MSI-X IRQ vectors
1207 * @bp: driver handle
1208 * @nvecs: number of vectors to be released
1210 static void bnx2x_free_msix_irqs(struct bnx2x *bp, int nvecs)
1212 int i, offset = 0;
1214 if (nvecs == offset)
1215 return;
1216 free_irq(bp->msix_table[offset].vector, bp->dev);
1217 DP(NETIF_MSG_IFDOWN, "released sp irq (%d)\n",
1218 bp->msix_table[offset].vector);
1219 offset++;
1220 #ifdef BCM_CNIC
1221 if (nvecs == offset)
1222 return;
1223 offset++;
1224 #endif
1226 for_each_eth_queue(bp, i) {
1227 if (nvecs == offset)
1228 return;
1229 DP(NETIF_MSG_IFDOWN, "about to release fp #%d->%d irq\n",
1230 i, bp->msix_table[offset].vector);
1232 free_irq(bp->msix_table[offset++].vector, &bp->fp[i]);
1236 void bnx2x_free_irq(struct bnx2x *bp)
1238 if (bp->flags & USING_MSIX_FLAG)
1239 bnx2x_free_msix_irqs(bp, BNX2X_NUM_ETH_QUEUES(bp) +
1240 CNIC_PRESENT + 1);
1241 else if (bp->flags & USING_MSI_FLAG)
1242 free_irq(bp->pdev->irq, bp->dev);
1243 else
1244 free_irq(bp->pdev->irq, bp->dev);
1247 int bnx2x_enable_msix(struct bnx2x *bp)
1249 int msix_vec = 0, i, rc, req_cnt;
1251 bp->msix_table[msix_vec].entry = msix_vec;
1252 BNX2X_DEV_INFO("msix_table[0].entry = %d (slowpath)\n",
1253 bp->msix_table[0].entry);
1254 msix_vec++;
1256 #ifdef BCM_CNIC
1257 bp->msix_table[msix_vec].entry = msix_vec;
1258 BNX2X_DEV_INFO("msix_table[%d].entry = %d (CNIC)\n",
1259 bp->msix_table[msix_vec].entry, bp->msix_table[msix_vec].entry);
1260 msix_vec++;
1261 #endif
1262 /* We need separate vectors for ETH queues only (not FCoE) */
1263 for_each_eth_queue(bp, i) {
1264 bp->msix_table[msix_vec].entry = msix_vec;
1265 BNX2X_DEV_INFO("msix_table[%d].entry = %d (fastpath #%u)\n",
1266 msix_vec, msix_vec, i);
1267 msix_vec++;
1270 req_cnt = BNX2X_NUM_ETH_QUEUES(bp) + CNIC_PRESENT + 1;
1272 rc = pci_enable_msix(bp->pdev, &bp->msix_table[0], req_cnt);
1275 * reconfigure number of tx/rx queues according to available
1276 * MSI-X vectors
1278 if (rc >= BNX2X_MIN_MSIX_VEC_CNT) {
1279 /* how less vectors we will have? */
1280 int diff = req_cnt - rc;
1282 BNX2X_DEV_INFO("Trying to use less MSI-X vectors: %d\n", rc);
1284 rc = pci_enable_msix(bp->pdev, &bp->msix_table[0], rc);
1286 if (rc) {
1287 BNX2X_DEV_INFO("MSI-X is not attainable rc %d\n", rc);
1288 return rc;
1291 * decrease number of queues by number of unallocated entries
1293 bp->num_queues -= diff;
1295 BNX2X_DEV_INFO("New queue configuration set: %d\n",
1296 bp->num_queues);
1297 } else if (rc) {
1298 /* fall to INTx if not enough memory */
1299 if (rc == -ENOMEM)
1300 bp->flags |= DISABLE_MSI_FLAG;
1301 BNX2X_DEV_INFO("MSI-X is not attainable rc %d\n", rc);
1302 return rc;
1305 bp->flags |= USING_MSIX_FLAG;
1307 return 0;
1310 static int bnx2x_req_msix_irqs(struct bnx2x *bp)
1312 int i, rc, offset = 0;
1314 rc = request_irq(bp->msix_table[offset++].vector,
1315 bnx2x_msix_sp_int, 0,
1316 bp->dev->name, bp->dev);
1317 if (rc) {
1318 BNX2X_ERR("request sp irq failed\n");
1319 return -EBUSY;
1322 #ifdef BCM_CNIC
1323 offset++;
1324 #endif
1325 for_each_eth_queue(bp, i) {
1326 struct bnx2x_fastpath *fp = &bp->fp[i];
1327 snprintf(fp->name, sizeof(fp->name), "%s-fp-%d",
1328 bp->dev->name, i);
1330 rc = request_irq(bp->msix_table[offset].vector,
1331 bnx2x_msix_fp_int, 0, fp->name, fp);
1332 if (rc) {
1333 BNX2X_ERR("request fp #%d irq (%d) failed rc %d\n", i,
1334 bp->msix_table[offset].vector, rc);
1335 bnx2x_free_msix_irqs(bp, offset);
1336 return -EBUSY;
1339 offset++;
1342 i = BNX2X_NUM_ETH_QUEUES(bp);
1343 offset = 1 + CNIC_PRESENT;
1344 netdev_info(bp->dev, "using MSI-X IRQs: sp %d fp[%d] %d ... fp[%d] %d\n",
1345 bp->msix_table[0].vector,
1346 0, bp->msix_table[offset].vector,
1347 i - 1, bp->msix_table[offset + i - 1].vector);
1349 return 0;
1352 int bnx2x_enable_msi(struct bnx2x *bp)
1354 int rc;
1356 rc = pci_enable_msi(bp->pdev);
1357 if (rc) {
1358 BNX2X_DEV_INFO("MSI is not attainable\n");
1359 return -1;
1361 bp->flags |= USING_MSI_FLAG;
1363 return 0;
1366 static int bnx2x_req_irq(struct bnx2x *bp)
1368 unsigned long flags;
1369 int rc;
1371 if (bp->flags & USING_MSI_FLAG)
1372 flags = 0;
1373 else
1374 flags = IRQF_SHARED;
1376 rc = request_irq(bp->pdev->irq, bnx2x_interrupt, flags,
1377 bp->dev->name, bp->dev);
1378 return rc;
1381 static inline int bnx2x_setup_irqs(struct bnx2x *bp)
1383 int rc = 0;
1384 if (bp->flags & USING_MSIX_FLAG) {
1385 rc = bnx2x_req_msix_irqs(bp);
1386 if (rc)
1387 return rc;
1388 } else {
1389 bnx2x_ack_int(bp);
1390 rc = bnx2x_req_irq(bp);
1391 if (rc) {
1392 BNX2X_ERR("IRQ request failed rc %d, aborting\n", rc);
1393 return rc;
1395 if (bp->flags & USING_MSI_FLAG) {
1396 bp->dev->irq = bp->pdev->irq;
1397 netdev_info(bp->dev, "using MSI IRQ %d\n",
1398 bp->pdev->irq);
1402 return 0;
1405 static inline void bnx2x_napi_enable(struct bnx2x *bp)
1407 int i;
1409 for_each_rx_queue(bp, i)
1410 napi_enable(&bnx2x_fp(bp, i, napi));
1413 static inline void bnx2x_napi_disable(struct bnx2x *bp)
1415 int i;
1417 for_each_rx_queue(bp, i)
1418 napi_disable(&bnx2x_fp(bp, i, napi));
1421 void bnx2x_netif_start(struct bnx2x *bp)
1423 if (netif_running(bp->dev)) {
1424 bnx2x_napi_enable(bp);
1425 bnx2x_int_enable(bp);
1426 if (bp->state == BNX2X_STATE_OPEN)
1427 netif_tx_wake_all_queues(bp->dev);
1431 void bnx2x_netif_stop(struct bnx2x *bp, int disable_hw)
1433 bnx2x_int_disable_sync(bp, disable_hw);
1434 bnx2x_napi_disable(bp);
1437 u16 bnx2x_select_queue(struct net_device *dev, struct sk_buff *skb)
1439 struct bnx2x *bp = netdev_priv(dev);
1441 #ifdef BCM_CNIC
1442 if (!NO_FCOE(bp)) {
1443 struct ethhdr *hdr = (struct ethhdr *)skb->data;
1444 u16 ether_type = ntohs(hdr->h_proto);
1446 /* Skip VLAN tag if present */
1447 if (ether_type == ETH_P_8021Q) {
1448 struct vlan_ethhdr *vhdr =
1449 (struct vlan_ethhdr *)skb->data;
1451 ether_type = ntohs(vhdr->h_vlan_encapsulated_proto);
1454 /* If ethertype is FCoE or FIP - use FCoE ring */
1455 if ((ether_type == ETH_P_FCOE) || (ether_type == ETH_P_FIP))
1456 return bnx2x_fcoe_tx(bp, txq_index);
1458 #endif
1459 /* select a non-FCoE queue */
1460 return __skb_tx_hash(dev, skb, BNX2X_NUM_ETH_QUEUES(bp));
1463 void bnx2x_set_num_queues(struct bnx2x *bp)
1465 switch (bp->multi_mode) {
1466 case ETH_RSS_MODE_DISABLED:
1467 bp->num_queues = 1;
1468 break;
1469 case ETH_RSS_MODE_REGULAR:
1470 bp->num_queues = bnx2x_calc_num_queues(bp);
1471 break;
1473 default:
1474 bp->num_queues = 1;
1475 break;
1478 #ifdef BCM_CNIC
1479 /* override in STORAGE SD mode */
1480 if (IS_MF_STORAGE_SD(bp))
1481 bp->num_queues = 1;
1482 #endif
1483 /* Add special queues */
1484 bp->num_queues += NON_ETH_CONTEXT_USE;
1488 * bnx2x_set_real_num_queues - configure netdev->real_num_[tx,rx]_queues
1490 * @bp: Driver handle
1492 * We currently support for at most 16 Tx queues for each CoS thus we will
1493 * allocate a multiple of 16 for ETH L2 rings according to the value of the
1494 * bp->max_cos.
1496 * If there is an FCoE L2 queue the appropriate Tx queue will have the next
1497 * index after all ETH L2 indices.
1499 * If the actual number of Tx queues (for each CoS) is less than 16 then there
1500 * will be the holes at the end of each group of 16 ETh L2 indices (0..15,
1501 * 16..31,...) with indicies that are not coupled with any real Tx queue.
1503 * The proper configuration of skb->queue_mapping is handled by
1504 * bnx2x_select_queue() and __skb_tx_hash().
1506 * bnx2x_setup_tc() takes care of the proper TC mappings so that __skb_tx_hash()
1507 * will return a proper Tx index if TC is enabled (netdev->num_tc > 0).
1509 static inline int bnx2x_set_real_num_queues(struct bnx2x *bp)
1511 int rc, tx, rx;
1513 tx = MAX_TXQS_PER_COS * bp->max_cos;
1514 rx = BNX2X_NUM_ETH_QUEUES(bp);
1516 /* account for fcoe queue */
1517 #ifdef BCM_CNIC
1518 if (!NO_FCOE(bp)) {
1519 rx += FCOE_PRESENT;
1520 tx += FCOE_PRESENT;
1522 #endif
1524 rc = netif_set_real_num_tx_queues(bp->dev, tx);
1525 if (rc) {
1526 BNX2X_ERR("Failed to set real number of Tx queues: %d\n", rc);
1527 return rc;
1529 rc = netif_set_real_num_rx_queues(bp->dev, rx);
1530 if (rc) {
1531 BNX2X_ERR("Failed to set real number of Rx queues: %d\n", rc);
1532 return rc;
1535 DP(NETIF_MSG_IFUP, "Setting real num queues to (tx, rx) (%d, %d)\n",
1536 tx, rx);
1538 return rc;
1541 static inline void bnx2x_set_rx_buf_size(struct bnx2x *bp)
1543 int i;
1545 for_each_queue(bp, i) {
1546 struct bnx2x_fastpath *fp = &bp->fp[i];
1547 u32 mtu;
1549 /* Always use a mini-jumbo MTU for the FCoE L2 ring */
1550 if (IS_FCOE_IDX(i))
1552 * Although there are no IP frames expected to arrive to
1553 * this ring we still want to add an
1554 * IP_HEADER_ALIGNMENT_PADDING to prevent a buffer
1555 * overrun attack.
1557 mtu = BNX2X_FCOE_MINI_JUMBO_MTU;
1558 else
1559 mtu = bp->dev->mtu;
1560 fp->rx_buf_size = BNX2X_FW_RX_ALIGN_START +
1561 IP_HEADER_ALIGNMENT_PADDING +
1562 ETH_OVREHEAD +
1563 mtu +
1564 BNX2X_FW_RX_ALIGN_END;
1565 /* Note : rx_buf_size doesnt take into account NET_SKB_PAD */
1569 static inline int bnx2x_init_rss_pf(struct bnx2x *bp)
1571 int i;
1572 u8 ind_table[T_ETH_INDIRECTION_TABLE_SIZE] = {0};
1573 u8 num_eth_queues = BNX2X_NUM_ETH_QUEUES(bp);
1576 * Prepare the inital contents fo the indirection table if RSS is
1577 * enabled
1579 if (bp->multi_mode != ETH_RSS_MODE_DISABLED) {
1580 for (i = 0; i < sizeof(ind_table); i++)
1581 ind_table[i] =
1582 bp->fp->cl_id +
1583 ethtool_rxfh_indir_default(i, num_eth_queues);
1587 * For 57710 and 57711 SEARCHER configuration (rss_keys) is
1588 * per-port, so if explicit configuration is needed , do it only
1589 * for a PMF.
1591 * For 57712 and newer on the other hand it's a per-function
1592 * configuration.
1594 return bnx2x_config_rss_pf(bp, ind_table,
1595 bp->port.pmf || !CHIP_IS_E1x(bp));
1598 int bnx2x_config_rss_pf(struct bnx2x *bp, u8 *ind_table, bool config_hash)
1600 struct bnx2x_config_rss_params params = {NULL};
1601 int i;
1603 /* Although RSS is meaningless when there is a single HW queue we
1604 * still need it enabled in order to have HW Rx hash generated.
1606 * if (!is_eth_multi(bp))
1607 * bp->multi_mode = ETH_RSS_MODE_DISABLED;
1610 params.rss_obj = &bp->rss_conf_obj;
1612 __set_bit(RAMROD_COMP_WAIT, &params.ramrod_flags);
1614 /* RSS mode */
1615 switch (bp->multi_mode) {
1616 case ETH_RSS_MODE_DISABLED:
1617 __set_bit(BNX2X_RSS_MODE_DISABLED, &params.rss_flags);
1618 break;
1619 case ETH_RSS_MODE_REGULAR:
1620 __set_bit(BNX2X_RSS_MODE_REGULAR, &params.rss_flags);
1621 break;
1622 case ETH_RSS_MODE_VLAN_PRI:
1623 __set_bit(BNX2X_RSS_MODE_VLAN_PRI, &params.rss_flags);
1624 break;
1625 case ETH_RSS_MODE_E1HOV_PRI:
1626 __set_bit(BNX2X_RSS_MODE_E1HOV_PRI, &params.rss_flags);
1627 break;
1628 case ETH_RSS_MODE_IP_DSCP:
1629 __set_bit(BNX2X_RSS_MODE_IP_DSCP, &params.rss_flags);
1630 break;
1631 default:
1632 BNX2X_ERR("Unknown multi_mode: %d\n", bp->multi_mode);
1633 return -EINVAL;
1636 /* If RSS is enabled */
1637 if (bp->multi_mode != ETH_RSS_MODE_DISABLED) {
1638 /* RSS configuration */
1639 __set_bit(BNX2X_RSS_IPV4, &params.rss_flags);
1640 __set_bit(BNX2X_RSS_IPV4_TCP, &params.rss_flags);
1641 __set_bit(BNX2X_RSS_IPV6, &params.rss_flags);
1642 __set_bit(BNX2X_RSS_IPV6_TCP, &params.rss_flags);
1644 /* Hash bits */
1645 params.rss_result_mask = MULTI_MASK;
1647 memcpy(params.ind_table, ind_table, sizeof(params.ind_table));
1649 if (config_hash) {
1650 /* RSS keys */
1651 for (i = 0; i < sizeof(params.rss_key) / 4; i++)
1652 params.rss_key[i] = random32();
1654 __set_bit(BNX2X_RSS_SET_SRCH, &params.rss_flags);
1658 return bnx2x_config_rss(bp, &params);
1661 static inline int bnx2x_init_hw(struct bnx2x *bp, u32 load_code)
1663 struct bnx2x_func_state_params func_params = {NULL};
1665 /* Prepare parameters for function state transitions */
1666 __set_bit(RAMROD_COMP_WAIT, &func_params.ramrod_flags);
1668 func_params.f_obj = &bp->func_obj;
1669 func_params.cmd = BNX2X_F_CMD_HW_INIT;
1671 func_params.params.hw_init.load_phase = load_code;
1673 return bnx2x_func_state_change(bp, &func_params);
1677 * Cleans the object that have internal lists without sending
1678 * ramrods. Should be run when interrutps are disabled.
1680 static void bnx2x_squeeze_objects(struct bnx2x *bp)
1682 int rc;
1683 unsigned long ramrod_flags = 0, vlan_mac_flags = 0;
1684 struct bnx2x_mcast_ramrod_params rparam = {NULL};
1685 struct bnx2x_vlan_mac_obj *mac_obj = &bp->fp->mac_obj;
1687 /***************** Cleanup MACs' object first *************************/
1689 /* Wait for completion of requested */
1690 __set_bit(RAMROD_COMP_WAIT, &ramrod_flags);
1691 /* Perform a dry cleanup */
1692 __set_bit(RAMROD_DRV_CLR_ONLY, &ramrod_flags);
1694 /* Clean ETH primary MAC */
1695 __set_bit(BNX2X_ETH_MAC, &vlan_mac_flags);
1696 rc = mac_obj->delete_all(bp, &bp->fp->mac_obj, &vlan_mac_flags,
1697 &ramrod_flags);
1698 if (rc != 0)
1699 BNX2X_ERR("Failed to clean ETH MACs: %d\n", rc);
1701 /* Cleanup UC list */
1702 vlan_mac_flags = 0;
1703 __set_bit(BNX2X_UC_LIST_MAC, &vlan_mac_flags);
1704 rc = mac_obj->delete_all(bp, mac_obj, &vlan_mac_flags,
1705 &ramrod_flags);
1706 if (rc != 0)
1707 BNX2X_ERR("Failed to clean UC list MACs: %d\n", rc);
1709 /***************** Now clean mcast object *****************************/
1710 rparam.mcast_obj = &bp->mcast_obj;
1711 __set_bit(RAMROD_DRV_CLR_ONLY, &rparam.ramrod_flags);
1713 /* Add a DEL command... */
1714 rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_DEL);
1715 if (rc < 0)
1716 BNX2X_ERR("Failed to add a new DEL command to a multi-cast object: %d\n",
1717 rc);
1719 /* ...and wait until all pending commands are cleared */
1720 rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_CONT);
1721 while (rc != 0) {
1722 if (rc < 0) {
1723 BNX2X_ERR("Failed to clean multi-cast object: %d\n",
1724 rc);
1725 return;
1728 rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_CONT);
1732 #ifndef BNX2X_STOP_ON_ERROR
1733 #define LOAD_ERROR_EXIT(bp, label) \
1734 do { \
1735 (bp)->state = BNX2X_STATE_ERROR; \
1736 goto label; \
1737 } while (0)
1738 #else
1739 #define LOAD_ERROR_EXIT(bp, label) \
1740 do { \
1741 (bp)->state = BNX2X_STATE_ERROR; \
1742 (bp)->panic = 1; \
1743 return -EBUSY; \
1744 } while (0)
1745 #endif
1747 bool bnx2x_test_firmware_version(struct bnx2x *bp, bool is_err)
1749 /* build FW version dword */
1750 u32 my_fw = (BCM_5710_FW_MAJOR_VERSION) +
1751 (BCM_5710_FW_MINOR_VERSION << 8) +
1752 (BCM_5710_FW_REVISION_VERSION << 16) +
1753 (BCM_5710_FW_ENGINEERING_VERSION << 24);
1755 /* read loaded FW from chip */
1756 u32 loaded_fw = REG_RD(bp, XSEM_REG_PRAM);
1758 DP(NETIF_MSG_IFUP, "loaded fw %x, my fw %x\n", loaded_fw, my_fw);
1760 if (loaded_fw != my_fw) {
1761 if (is_err)
1762 BNX2X_ERR("bnx2x with FW %x was already loaded, which mismatches my %x FW. aborting\n",
1763 loaded_fw, my_fw);
1764 return false;
1767 return true;
1770 /* must be called with rtnl_lock */
1771 int bnx2x_nic_load(struct bnx2x *bp, int load_mode)
1773 int port = BP_PORT(bp);
1774 u32 load_code;
1775 int i, rc;
1777 #ifdef BNX2X_STOP_ON_ERROR
1778 if (unlikely(bp->panic)) {
1779 BNX2X_ERR("Can't load NIC when there is panic\n");
1780 return -EPERM;
1782 #endif
1784 bp->state = BNX2X_STATE_OPENING_WAIT4_LOAD;
1786 /* Set the initial link reported state to link down */
1787 bnx2x_acquire_phy_lock(bp);
1788 memset(&bp->last_reported_link, 0, sizeof(bp->last_reported_link));
1789 __set_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1790 &bp->last_reported_link.link_report_flags);
1791 bnx2x_release_phy_lock(bp);
1793 /* must be called before memory allocation and HW init */
1794 bnx2x_ilt_set_info(bp);
1797 * Zero fastpath structures preserving invariants like napi, which are
1798 * allocated only once, fp index, max_cos, bp pointer.
1799 * Also set fp->disable_tpa.
1801 DP(NETIF_MSG_IFUP, "num queues: %d", bp->num_queues);
1802 for_each_queue(bp, i)
1803 bnx2x_bz_fp(bp, i);
1806 /* Set the receive queues buffer size */
1807 bnx2x_set_rx_buf_size(bp);
1809 if (bnx2x_alloc_mem(bp))
1810 return -ENOMEM;
1812 /* As long as bnx2x_alloc_mem() may possibly update
1813 * bp->num_queues, bnx2x_set_real_num_queues() should always
1814 * come after it.
1816 rc = bnx2x_set_real_num_queues(bp);
1817 if (rc) {
1818 BNX2X_ERR("Unable to set real_num_queues\n");
1819 LOAD_ERROR_EXIT(bp, load_error0);
1822 /* configure multi cos mappings in kernel.
1823 * this configuration may be overriden by a multi class queue discipline
1824 * or by a dcbx negotiation result.
1826 bnx2x_setup_tc(bp->dev, bp->max_cos);
1828 bnx2x_napi_enable(bp);
1830 /* set pf load just before approaching the MCP */
1831 bnx2x_set_pf_load(bp);
1833 /* Send LOAD_REQUEST command to MCP
1834 * Returns the type of LOAD command:
1835 * if it is the first port to be initialized
1836 * common blocks should be initialized, otherwise - not
1838 if (!BP_NOMCP(bp)) {
1839 /* init fw_seq */
1840 bp->fw_seq =
1841 (SHMEM_RD(bp, func_mb[BP_FW_MB_IDX(bp)].drv_mb_header) &
1842 DRV_MSG_SEQ_NUMBER_MASK);
1843 BNX2X_DEV_INFO("fw_seq 0x%08x\n", bp->fw_seq);
1845 /* Get current FW pulse sequence */
1846 bp->fw_drv_pulse_wr_seq =
1847 (SHMEM_RD(bp, func_mb[BP_FW_MB_IDX(bp)].drv_pulse_mb) &
1848 DRV_PULSE_SEQ_MASK);
1849 BNX2X_DEV_INFO("drv_pulse 0x%x\n", bp->fw_drv_pulse_wr_seq);
1851 load_code = bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_REQ, 0);
1852 if (!load_code) {
1853 BNX2X_ERR("MCP response failure, aborting\n");
1854 rc = -EBUSY;
1855 LOAD_ERROR_EXIT(bp, load_error1);
1857 if (load_code == FW_MSG_CODE_DRV_LOAD_REFUSED) {
1858 BNX2X_ERR("Driver load refused\n");
1859 rc = -EBUSY; /* other port in diagnostic mode */
1860 LOAD_ERROR_EXIT(bp, load_error1);
1862 if (load_code != FW_MSG_CODE_DRV_LOAD_COMMON_CHIP &&
1863 load_code != FW_MSG_CODE_DRV_LOAD_COMMON) {
1864 /* abort nic load if version mismatch */
1865 if (!bnx2x_test_firmware_version(bp, true)) {
1866 rc = -EBUSY;
1867 LOAD_ERROR_EXIT(bp, load_error2);
1871 } else {
1872 int path = BP_PATH(bp);
1874 DP(NETIF_MSG_IFUP, "NO MCP - load counts[%d] %d, %d, %d\n",
1875 path, load_count[path][0], load_count[path][1],
1876 load_count[path][2]);
1877 load_count[path][0]++;
1878 load_count[path][1 + port]++;
1879 DP(NETIF_MSG_IFUP, "NO MCP - new load counts[%d] %d, %d, %d\n",
1880 path, load_count[path][0], load_count[path][1],
1881 load_count[path][2]);
1882 if (load_count[path][0] == 1)
1883 load_code = FW_MSG_CODE_DRV_LOAD_COMMON;
1884 else if (load_count[path][1 + port] == 1)
1885 load_code = FW_MSG_CODE_DRV_LOAD_PORT;
1886 else
1887 load_code = FW_MSG_CODE_DRV_LOAD_FUNCTION;
1890 if ((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
1891 (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP) ||
1892 (load_code == FW_MSG_CODE_DRV_LOAD_PORT)) {
1893 bp->port.pmf = 1;
1895 * We need the barrier to ensure the ordering between the
1896 * writing to bp->port.pmf here and reading it from the
1897 * bnx2x_periodic_task().
1899 smp_mb();
1900 } else
1901 bp->port.pmf = 0;
1903 DP(NETIF_MSG_IFUP, "pmf %d\n", bp->port.pmf);
1905 /* Init Function state controlling object */
1906 bnx2x__init_func_obj(bp);
1908 /* Initialize HW */
1909 rc = bnx2x_init_hw(bp, load_code);
1910 if (rc) {
1911 BNX2X_ERR("HW init failed, aborting\n");
1912 bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
1913 LOAD_ERROR_EXIT(bp, load_error2);
1916 /* Connect to IRQs */
1917 rc = bnx2x_setup_irqs(bp);
1918 if (rc) {
1919 BNX2X_ERR("IRQs setup failed\n");
1920 bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
1921 LOAD_ERROR_EXIT(bp, load_error2);
1924 /* Setup NIC internals and enable interrupts */
1925 bnx2x_nic_init(bp, load_code);
1927 /* Init per-function objects */
1928 bnx2x_init_bp_objs(bp);
1930 if (((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
1931 (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP)) &&
1932 (bp->common.shmem2_base)) {
1933 if (SHMEM2_HAS(bp, dcc_support))
1934 SHMEM2_WR(bp, dcc_support,
1935 (SHMEM_DCC_SUPPORT_DISABLE_ENABLE_PF_TLV |
1936 SHMEM_DCC_SUPPORT_BANDWIDTH_ALLOCATION_TLV));
1939 bp->state = BNX2X_STATE_OPENING_WAIT4_PORT;
1940 rc = bnx2x_func_start(bp);
1941 if (rc) {
1942 BNX2X_ERR("Function start failed!\n");
1943 bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
1944 LOAD_ERROR_EXIT(bp, load_error3);
1947 /* Send LOAD_DONE command to MCP */
1948 if (!BP_NOMCP(bp)) {
1949 load_code = bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
1950 if (!load_code) {
1951 BNX2X_ERR("MCP response failure, aborting\n");
1952 rc = -EBUSY;
1953 LOAD_ERROR_EXIT(bp, load_error3);
1957 rc = bnx2x_setup_leading(bp);
1958 if (rc) {
1959 BNX2X_ERR("Setup leading failed!\n");
1960 LOAD_ERROR_EXIT(bp, load_error3);
1963 #ifdef BCM_CNIC
1964 /* Enable Timer scan */
1965 REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + port*4, 1);
1966 #endif
1968 for_each_nondefault_queue(bp, i) {
1969 rc = bnx2x_setup_queue(bp, &bp->fp[i], 0);
1970 if (rc) {
1971 BNX2X_ERR("Queue setup failed\n");
1972 LOAD_ERROR_EXIT(bp, load_error4);
1976 rc = bnx2x_init_rss_pf(bp);
1977 if (rc) {
1978 BNX2X_ERR("PF RSS init failed\n");
1979 LOAD_ERROR_EXIT(bp, load_error4);
1982 /* Now when Clients are configured we are ready to work */
1983 bp->state = BNX2X_STATE_OPEN;
1985 /* Configure a ucast MAC */
1986 rc = bnx2x_set_eth_mac(bp, true);
1987 if (rc) {
1988 BNX2X_ERR("Setting Ethernet MAC failed\n");
1989 LOAD_ERROR_EXIT(bp, load_error4);
1992 if (bp->pending_max) {
1993 bnx2x_update_max_mf_config(bp, bp->pending_max);
1994 bp->pending_max = 0;
1997 if (bp->port.pmf)
1998 bnx2x_initial_phy_init(bp, load_mode);
2000 /* Start fast path */
2002 /* Initialize Rx filter. */
2003 netif_addr_lock_bh(bp->dev);
2004 bnx2x_set_rx_mode(bp->dev);
2005 netif_addr_unlock_bh(bp->dev);
2007 /* Start the Tx */
2008 switch (load_mode) {
2009 case LOAD_NORMAL:
2010 /* Tx queue should be only reenabled */
2011 netif_tx_wake_all_queues(bp->dev);
2012 break;
2014 case LOAD_OPEN:
2015 netif_tx_start_all_queues(bp->dev);
2016 smp_mb__after_clear_bit();
2017 break;
2019 case LOAD_DIAG:
2020 bp->state = BNX2X_STATE_DIAG;
2021 break;
2023 default:
2024 break;
2027 if (bp->port.pmf)
2028 bnx2x_update_drv_flags(bp, 1 << DRV_FLAGS_DCB_CONFIGURED, 0);
2029 else
2030 bnx2x__link_status_update(bp);
2032 /* start the timer */
2033 mod_timer(&bp->timer, jiffies + bp->current_interval);
2035 #ifdef BCM_CNIC
2036 /* re-read iscsi info */
2037 bnx2x_get_iscsi_info(bp);
2038 bnx2x_setup_cnic_irq_info(bp);
2039 if (bp->state == BNX2X_STATE_OPEN)
2040 bnx2x_cnic_notify(bp, CNIC_CTL_START_CMD);
2041 #endif
2043 /* mark driver is loaded in shmem2 */
2044 if (SHMEM2_HAS(bp, drv_capabilities_flag)) {
2045 u32 val;
2046 val = SHMEM2_RD(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)]);
2047 SHMEM2_WR(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)],
2048 val | DRV_FLAGS_CAPABILITIES_LOADED_SUPPORTED |
2049 DRV_FLAGS_CAPABILITIES_LOADED_L2);
2052 /* Wait for all pending SP commands to complete */
2053 if (!bnx2x_wait_sp_comp(bp, ~0x0UL)) {
2054 BNX2X_ERR("Timeout waiting for SP elements to complete\n");
2055 bnx2x_nic_unload(bp, UNLOAD_CLOSE);
2056 return -EBUSY;
2059 bnx2x_dcbx_init(bp);
2060 return 0;
2062 #ifndef BNX2X_STOP_ON_ERROR
2063 load_error4:
2064 #ifdef BCM_CNIC
2065 /* Disable Timer scan */
2066 REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + port*4, 0);
2067 #endif
2068 load_error3:
2069 bnx2x_int_disable_sync(bp, 1);
2071 /* Clean queueable objects */
2072 bnx2x_squeeze_objects(bp);
2074 /* Free SKBs, SGEs, TPA pool and driver internals */
2075 bnx2x_free_skbs(bp);
2076 for_each_rx_queue(bp, i)
2077 bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
2079 /* Release IRQs */
2080 bnx2x_free_irq(bp);
2081 load_error2:
2082 if (!BP_NOMCP(bp)) {
2083 bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_REQ_WOL_MCP, 0);
2084 bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_DONE, 0);
2087 bp->port.pmf = 0;
2088 load_error1:
2089 bnx2x_napi_disable(bp);
2090 /* clear pf_load status, as it was already set */
2091 bnx2x_clear_pf_load(bp);
2092 load_error0:
2093 bnx2x_free_mem(bp);
2095 return rc;
2096 #endif /* ! BNX2X_STOP_ON_ERROR */
2099 /* must be called with rtnl_lock */
2100 int bnx2x_nic_unload(struct bnx2x *bp, int unload_mode)
2102 int i;
2103 bool global = false;
2105 /* mark driver is unloaded in shmem2 */
2106 if (SHMEM2_HAS(bp, drv_capabilities_flag)) {
2107 u32 val;
2108 val = SHMEM2_RD(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)]);
2109 SHMEM2_WR(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)],
2110 val & ~DRV_FLAGS_CAPABILITIES_LOADED_L2);
2113 if ((bp->state == BNX2X_STATE_CLOSED) ||
2114 (bp->state == BNX2X_STATE_ERROR)) {
2115 /* We can get here if the driver has been unloaded
2116 * during parity error recovery and is either waiting for a
2117 * leader to complete or for other functions to unload and
2118 * then ifdown has been issued. In this case we want to
2119 * unload and let other functions to complete a recovery
2120 * process.
2122 bp->recovery_state = BNX2X_RECOVERY_DONE;
2123 bp->is_leader = 0;
2124 bnx2x_release_leader_lock(bp);
2125 smp_mb();
2127 DP(NETIF_MSG_IFDOWN, "Releasing a leadership...\n");
2128 BNX2X_ERR("Can't unload in closed or error state\n");
2129 return -EINVAL;
2133 * It's important to set the bp->state to the value different from
2134 * BNX2X_STATE_OPEN and only then stop the Tx. Otherwise bnx2x_tx_int()
2135 * may restart the Tx from the NAPI context (see bnx2x_tx_int()).
2137 bp->state = BNX2X_STATE_CLOSING_WAIT4_HALT;
2138 smp_mb();
2140 /* Stop Tx */
2141 bnx2x_tx_disable(bp);
2143 #ifdef BCM_CNIC
2144 bnx2x_cnic_notify(bp, CNIC_CTL_STOP_CMD);
2145 #endif
2147 bp->rx_mode = BNX2X_RX_MODE_NONE;
2149 del_timer_sync(&bp->timer);
2151 /* Set ALWAYS_ALIVE bit in shmem */
2152 bp->fw_drv_pulse_wr_seq |= DRV_PULSE_ALWAYS_ALIVE;
2154 bnx2x_drv_pulse(bp);
2156 bnx2x_stats_handle(bp, STATS_EVENT_STOP);
2157 bnx2x_save_statistics(bp);
2159 /* Cleanup the chip if needed */
2160 if (unload_mode != UNLOAD_RECOVERY)
2161 bnx2x_chip_cleanup(bp, unload_mode);
2162 else {
2163 /* Send the UNLOAD_REQUEST to the MCP */
2164 bnx2x_send_unload_req(bp, unload_mode);
2167 * Prevent transactions to host from the functions on the
2168 * engine that doesn't reset global blocks in case of global
2169 * attention once gloabl blocks are reset and gates are opened
2170 * (the engine which leader will perform the recovery
2171 * last).
2173 if (!CHIP_IS_E1x(bp))
2174 bnx2x_pf_disable(bp);
2176 /* Disable HW interrupts, NAPI */
2177 bnx2x_netif_stop(bp, 1);
2179 /* Release IRQs */
2180 bnx2x_free_irq(bp);
2182 /* Report UNLOAD_DONE to MCP */
2183 bnx2x_send_unload_done(bp);
2187 * At this stage no more interrupts will arrive so we may safly clean
2188 * the queueable objects here in case they failed to get cleaned so far.
2190 bnx2x_squeeze_objects(bp);
2192 /* There should be no more pending SP commands at this stage */
2193 bp->sp_state = 0;
2195 bp->port.pmf = 0;
2197 /* Free SKBs, SGEs, TPA pool and driver internals */
2198 bnx2x_free_skbs(bp);
2199 for_each_rx_queue(bp, i)
2200 bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
2202 bnx2x_free_mem(bp);
2204 bp->state = BNX2X_STATE_CLOSED;
2206 /* Check if there are pending parity attentions. If there are - set
2207 * RECOVERY_IN_PROGRESS.
2209 if (bnx2x_chk_parity_attn(bp, &global, false)) {
2210 bnx2x_set_reset_in_progress(bp);
2212 /* Set RESET_IS_GLOBAL if needed */
2213 if (global)
2214 bnx2x_set_reset_global(bp);
2218 /* The last driver must disable a "close the gate" if there is no
2219 * parity attention or "process kill" pending.
2221 if (!bnx2x_clear_pf_load(bp) && bnx2x_reset_is_done(bp, BP_PATH(bp)))
2222 bnx2x_disable_close_the_gate(bp);
2224 return 0;
2227 int bnx2x_set_power_state(struct bnx2x *bp, pci_power_t state)
2229 u16 pmcsr;
2231 /* If there is no power capability, silently succeed */
2232 if (!bp->pm_cap) {
2233 BNX2X_DEV_INFO("No power capability. Breaking.\n");
2234 return 0;
2237 pci_read_config_word(bp->pdev, bp->pm_cap + PCI_PM_CTRL, &pmcsr);
2239 switch (state) {
2240 case PCI_D0:
2241 pci_write_config_word(bp->pdev, bp->pm_cap + PCI_PM_CTRL,
2242 ((pmcsr & ~PCI_PM_CTRL_STATE_MASK) |
2243 PCI_PM_CTRL_PME_STATUS));
2245 if (pmcsr & PCI_PM_CTRL_STATE_MASK)
2246 /* delay required during transition out of D3hot */
2247 msleep(20);
2248 break;
2250 case PCI_D3hot:
2251 /* If there are other clients above don't
2252 shut down the power */
2253 if (atomic_read(&bp->pdev->enable_cnt) != 1)
2254 return 0;
2255 /* Don't shut down the power for emulation and FPGA */
2256 if (CHIP_REV_IS_SLOW(bp))
2257 return 0;
2259 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
2260 pmcsr |= 3;
2262 if (bp->wol)
2263 pmcsr |= PCI_PM_CTRL_PME_ENABLE;
2265 pci_write_config_word(bp->pdev, bp->pm_cap + PCI_PM_CTRL,
2266 pmcsr);
2268 /* No more memory access after this point until
2269 * device is brought back to D0.
2271 break;
2273 default:
2274 dev_err(&bp->pdev->dev, "Can't support state = %d\n", state);
2275 return -EINVAL;
2277 return 0;
2281 * net_device service functions
2283 int bnx2x_poll(struct napi_struct *napi, int budget)
2285 int work_done = 0;
2286 u8 cos;
2287 struct bnx2x_fastpath *fp = container_of(napi, struct bnx2x_fastpath,
2288 napi);
2289 struct bnx2x *bp = fp->bp;
2291 while (1) {
2292 #ifdef BNX2X_STOP_ON_ERROR
2293 if (unlikely(bp->panic)) {
2294 napi_complete(napi);
2295 return 0;
2297 #endif
2299 for_each_cos_in_tx_queue(fp, cos)
2300 if (bnx2x_tx_queue_has_work(&fp->txdata[cos]))
2301 bnx2x_tx_int(bp, &fp->txdata[cos]);
2304 if (bnx2x_has_rx_work(fp)) {
2305 work_done += bnx2x_rx_int(fp, budget - work_done);
2307 /* must not complete if we consumed full budget */
2308 if (work_done >= budget)
2309 break;
2312 /* Fall out from the NAPI loop if needed */
2313 if (!(bnx2x_has_rx_work(fp) || bnx2x_has_tx_work(fp))) {
2314 #ifdef BCM_CNIC
2315 /* No need to update SB for FCoE L2 ring as long as
2316 * it's connected to the default SB and the SB
2317 * has been updated when NAPI was scheduled.
2319 if (IS_FCOE_FP(fp)) {
2320 napi_complete(napi);
2321 break;
2323 #endif
2325 bnx2x_update_fpsb_idx(fp);
2326 /* bnx2x_has_rx_work() reads the status block,
2327 * thus we need to ensure that status block indices
2328 * have been actually read (bnx2x_update_fpsb_idx)
2329 * prior to this check (bnx2x_has_rx_work) so that
2330 * we won't write the "newer" value of the status block
2331 * to IGU (if there was a DMA right after
2332 * bnx2x_has_rx_work and if there is no rmb, the memory
2333 * reading (bnx2x_update_fpsb_idx) may be postponed
2334 * to right before bnx2x_ack_sb). In this case there
2335 * will never be another interrupt until there is
2336 * another update of the status block, while there
2337 * is still unhandled work.
2339 rmb();
2341 if (!(bnx2x_has_rx_work(fp) || bnx2x_has_tx_work(fp))) {
2342 napi_complete(napi);
2343 /* Re-enable interrupts */
2344 DP(NETIF_MSG_RX_STATUS,
2345 "Update index to %d\n", fp->fp_hc_idx);
2346 bnx2x_ack_sb(bp, fp->igu_sb_id, USTORM_ID,
2347 le16_to_cpu(fp->fp_hc_idx),
2348 IGU_INT_ENABLE, 1);
2349 break;
2354 return work_done;
2357 /* we split the first BD into headers and data BDs
2358 * to ease the pain of our fellow microcode engineers
2359 * we use one mapping for both BDs
2361 static noinline u16 bnx2x_tx_split(struct bnx2x *bp,
2362 struct bnx2x_fp_txdata *txdata,
2363 struct sw_tx_bd *tx_buf,
2364 struct eth_tx_start_bd **tx_bd, u16 hlen,
2365 u16 bd_prod, int nbd)
2367 struct eth_tx_start_bd *h_tx_bd = *tx_bd;
2368 struct eth_tx_bd *d_tx_bd;
2369 dma_addr_t mapping;
2370 int old_len = le16_to_cpu(h_tx_bd->nbytes);
2372 /* first fix first BD */
2373 h_tx_bd->nbd = cpu_to_le16(nbd);
2374 h_tx_bd->nbytes = cpu_to_le16(hlen);
2376 DP(NETIF_MSG_TX_QUEUED, "TSO split header size is %d (%x:%x) nbd %d\n",
2377 h_tx_bd->nbytes, h_tx_bd->addr_hi, h_tx_bd->addr_lo, h_tx_bd->nbd);
2379 /* now get a new data BD
2380 * (after the pbd) and fill it */
2381 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
2382 d_tx_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
2384 mapping = HILO_U64(le32_to_cpu(h_tx_bd->addr_hi),
2385 le32_to_cpu(h_tx_bd->addr_lo)) + hlen;
2387 d_tx_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
2388 d_tx_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
2389 d_tx_bd->nbytes = cpu_to_le16(old_len - hlen);
2391 /* this marks the BD as one that has no individual mapping */
2392 tx_buf->flags |= BNX2X_TSO_SPLIT_BD;
2394 DP(NETIF_MSG_TX_QUEUED,
2395 "TSO split data size is %d (%x:%x)\n",
2396 d_tx_bd->nbytes, d_tx_bd->addr_hi, d_tx_bd->addr_lo);
2398 /* update tx_bd */
2399 *tx_bd = (struct eth_tx_start_bd *)d_tx_bd;
2401 return bd_prod;
2404 static inline u16 bnx2x_csum_fix(unsigned char *t_header, u16 csum, s8 fix)
2406 if (fix > 0)
2407 csum = (u16) ~csum_fold(csum_sub(csum,
2408 csum_partial(t_header - fix, fix, 0)));
2410 else if (fix < 0)
2411 csum = (u16) ~csum_fold(csum_add(csum,
2412 csum_partial(t_header, -fix, 0)));
2414 return swab16(csum);
2417 static inline u32 bnx2x_xmit_type(struct bnx2x *bp, struct sk_buff *skb)
2419 u32 rc;
2421 if (skb->ip_summed != CHECKSUM_PARTIAL)
2422 rc = XMIT_PLAIN;
2424 else {
2425 if (vlan_get_protocol(skb) == htons(ETH_P_IPV6)) {
2426 rc = XMIT_CSUM_V6;
2427 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2428 rc |= XMIT_CSUM_TCP;
2430 } else {
2431 rc = XMIT_CSUM_V4;
2432 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2433 rc |= XMIT_CSUM_TCP;
2437 if (skb_is_gso_v6(skb))
2438 rc |= XMIT_GSO_V6 | XMIT_CSUM_TCP | XMIT_CSUM_V6;
2439 else if (skb_is_gso(skb))
2440 rc |= XMIT_GSO_V4 | XMIT_CSUM_V4 | XMIT_CSUM_TCP;
2442 return rc;
2445 #if (MAX_SKB_FRAGS >= MAX_FETCH_BD - 3)
2446 /* check if packet requires linearization (packet is too fragmented)
2447 no need to check fragmentation if page size > 8K (there will be no
2448 violation to FW restrictions) */
2449 static int bnx2x_pkt_req_lin(struct bnx2x *bp, struct sk_buff *skb,
2450 u32 xmit_type)
2452 int to_copy = 0;
2453 int hlen = 0;
2454 int first_bd_sz = 0;
2456 /* 3 = 1 (for linear data BD) + 2 (for PBD and last BD) */
2457 if (skb_shinfo(skb)->nr_frags >= (MAX_FETCH_BD - 3)) {
2459 if (xmit_type & XMIT_GSO) {
2460 unsigned short lso_mss = skb_shinfo(skb)->gso_size;
2461 /* Check if LSO packet needs to be copied:
2462 3 = 1 (for headers BD) + 2 (for PBD and last BD) */
2463 int wnd_size = MAX_FETCH_BD - 3;
2464 /* Number of windows to check */
2465 int num_wnds = skb_shinfo(skb)->nr_frags - wnd_size;
2466 int wnd_idx = 0;
2467 int frag_idx = 0;
2468 u32 wnd_sum = 0;
2470 /* Headers length */
2471 hlen = (int)(skb_transport_header(skb) - skb->data) +
2472 tcp_hdrlen(skb);
2474 /* Amount of data (w/o headers) on linear part of SKB*/
2475 first_bd_sz = skb_headlen(skb) - hlen;
2477 wnd_sum = first_bd_sz;
2479 /* Calculate the first sum - it's special */
2480 for (frag_idx = 0; frag_idx < wnd_size - 1; frag_idx++)
2481 wnd_sum +=
2482 skb_frag_size(&skb_shinfo(skb)->frags[frag_idx]);
2484 /* If there was data on linear skb data - check it */
2485 if (first_bd_sz > 0) {
2486 if (unlikely(wnd_sum < lso_mss)) {
2487 to_copy = 1;
2488 goto exit_lbl;
2491 wnd_sum -= first_bd_sz;
2494 /* Others are easier: run through the frag list and
2495 check all windows */
2496 for (wnd_idx = 0; wnd_idx <= num_wnds; wnd_idx++) {
2497 wnd_sum +=
2498 skb_frag_size(&skb_shinfo(skb)->frags[wnd_idx + wnd_size - 1]);
2500 if (unlikely(wnd_sum < lso_mss)) {
2501 to_copy = 1;
2502 break;
2504 wnd_sum -=
2505 skb_frag_size(&skb_shinfo(skb)->frags[wnd_idx]);
2507 } else {
2508 /* in non-LSO too fragmented packet should always
2509 be linearized */
2510 to_copy = 1;
2514 exit_lbl:
2515 if (unlikely(to_copy))
2516 DP(NETIF_MSG_TX_QUEUED,
2517 "Linearization IS REQUIRED for %s packet. num_frags %d hlen %d first_bd_sz %d\n",
2518 (xmit_type & XMIT_GSO) ? "LSO" : "non-LSO",
2519 skb_shinfo(skb)->nr_frags, hlen, first_bd_sz);
2521 return to_copy;
2523 #endif
2525 static inline void bnx2x_set_pbd_gso_e2(struct sk_buff *skb, u32 *parsing_data,
2526 u32 xmit_type)
2528 *parsing_data |= (skb_shinfo(skb)->gso_size <<
2529 ETH_TX_PARSE_BD_E2_LSO_MSS_SHIFT) &
2530 ETH_TX_PARSE_BD_E2_LSO_MSS;
2531 if ((xmit_type & XMIT_GSO_V6) &&
2532 (ipv6_hdr(skb)->nexthdr == NEXTHDR_IPV6))
2533 *parsing_data |= ETH_TX_PARSE_BD_E2_IPV6_WITH_EXT_HDR;
2537 * bnx2x_set_pbd_gso - update PBD in GSO case.
2539 * @skb: packet skb
2540 * @pbd: parse BD
2541 * @xmit_type: xmit flags
2543 static inline void bnx2x_set_pbd_gso(struct sk_buff *skb,
2544 struct eth_tx_parse_bd_e1x *pbd,
2545 u32 xmit_type)
2547 pbd->lso_mss = cpu_to_le16(skb_shinfo(skb)->gso_size);
2548 pbd->tcp_send_seq = swab32(tcp_hdr(skb)->seq);
2549 pbd->tcp_flags = pbd_tcp_flags(skb);
2551 if (xmit_type & XMIT_GSO_V4) {
2552 pbd->ip_id = swab16(ip_hdr(skb)->id);
2553 pbd->tcp_pseudo_csum =
2554 swab16(~csum_tcpudp_magic(ip_hdr(skb)->saddr,
2555 ip_hdr(skb)->daddr,
2556 0, IPPROTO_TCP, 0));
2558 } else
2559 pbd->tcp_pseudo_csum =
2560 swab16(~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2561 &ipv6_hdr(skb)->daddr,
2562 0, IPPROTO_TCP, 0));
2564 pbd->global_data |= ETH_TX_PARSE_BD_E1X_PSEUDO_CS_WITHOUT_LEN;
2568 * bnx2x_set_pbd_csum_e2 - update PBD with checksum and return header length
2570 * @bp: driver handle
2571 * @skb: packet skb
2572 * @parsing_data: data to be updated
2573 * @xmit_type: xmit flags
2575 * 57712 related
2577 static inline u8 bnx2x_set_pbd_csum_e2(struct bnx2x *bp, struct sk_buff *skb,
2578 u32 *parsing_data, u32 xmit_type)
2580 *parsing_data |=
2581 ((((u8 *)skb_transport_header(skb) - skb->data) >> 1) <<
2582 ETH_TX_PARSE_BD_E2_TCP_HDR_START_OFFSET_W_SHIFT) &
2583 ETH_TX_PARSE_BD_E2_TCP_HDR_START_OFFSET_W;
2585 if (xmit_type & XMIT_CSUM_TCP) {
2586 *parsing_data |= ((tcp_hdrlen(skb) / 4) <<
2587 ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW_SHIFT) &
2588 ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW;
2590 return skb_transport_header(skb) + tcp_hdrlen(skb) - skb->data;
2591 } else
2592 /* We support checksum offload for TCP and UDP only.
2593 * No need to pass the UDP header length - it's a constant.
2595 return skb_transport_header(skb) +
2596 sizeof(struct udphdr) - skb->data;
2599 static inline void bnx2x_set_sbd_csum(struct bnx2x *bp, struct sk_buff *skb,
2600 struct eth_tx_start_bd *tx_start_bd, u32 xmit_type)
2602 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_L4_CSUM;
2604 if (xmit_type & XMIT_CSUM_V4)
2605 tx_start_bd->bd_flags.as_bitfield |=
2606 ETH_TX_BD_FLAGS_IP_CSUM;
2607 else
2608 tx_start_bd->bd_flags.as_bitfield |=
2609 ETH_TX_BD_FLAGS_IPV6;
2611 if (!(xmit_type & XMIT_CSUM_TCP))
2612 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_IS_UDP;
2616 * bnx2x_set_pbd_csum - update PBD with checksum and return header length
2618 * @bp: driver handle
2619 * @skb: packet skb
2620 * @pbd: parse BD to be updated
2621 * @xmit_type: xmit flags
2623 static inline u8 bnx2x_set_pbd_csum(struct bnx2x *bp, struct sk_buff *skb,
2624 struct eth_tx_parse_bd_e1x *pbd,
2625 u32 xmit_type)
2627 u8 hlen = (skb_network_header(skb) - skb->data) >> 1;
2629 /* for now NS flag is not used in Linux */
2630 pbd->global_data =
2631 (hlen | ((skb->protocol == cpu_to_be16(ETH_P_8021Q)) <<
2632 ETH_TX_PARSE_BD_E1X_LLC_SNAP_EN_SHIFT));
2634 pbd->ip_hlen_w = (skb_transport_header(skb) -
2635 skb_network_header(skb)) >> 1;
2637 hlen += pbd->ip_hlen_w;
2639 /* We support checksum offload for TCP and UDP only */
2640 if (xmit_type & XMIT_CSUM_TCP)
2641 hlen += tcp_hdrlen(skb) / 2;
2642 else
2643 hlen += sizeof(struct udphdr) / 2;
2645 pbd->total_hlen_w = cpu_to_le16(hlen);
2646 hlen = hlen*2;
2648 if (xmit_type & XMIT_CSUM_TCP) {
2649 pbd->tcp_pseudo_csum = swab16(tcp_hdr(skb)->check);
2651 } else {
2652 s8 fix = SKB_CS_OFF(skb); /* signed! */
2654 DP(NETIF_MSG_TX_QUEUED,
2655 "hlen %d fix %d csum before fix %x\n",
2656 le16_to_cpu(pbd->total_hlen_w), fix, SKB_CS(skb));
2658 /* HW bug: fixup the CSUM */
2659 pbd->tcp_pseudo_csum =
2660 bnx2x_csum_fix(skb_transport_header(skb),
2661 SKB_CS(skb), fix);
2663 DP(NETIF_MSG_TX_QUEUED, "csum after fix %x\n",
2664 pbd->tcp_pseudo_csum);
2667 return hlen;
2670 /* called with netif_tx_lock
2671 * bnx2x_tx_int() runs without netif_tx_lock unless it needs to call
2672 * netif_wake_queue()
2674 netdev_tx_t bnx2x_start_xmit(struct sk_buff *skb, struct net_device *dev)
2676 struct bnx2x *bp = netdev_priv(dev);
2678 struct bnx2x_fastpath *fp;
2679 struct netdev_queue *txq;
2680 struct bnx2x_fp_txdata *txdata;
2681 struct sw_tx_bd *tx_buf;
2682 struct eth_tx_start_bd *tx_start_bd, *first_bd;
2683 struct eth_tx_bd *tx_data_bd, *total_pkt_bd = NULL;
2684 struct eth_tx_parse_bd_e1x *pbd_e1x = NULL;
2685 struct eth_tx_parse_bd_e2 *pbd_e2 = NULL;
2686 u32 pbd_e2_parsing_data = 0;
2687 u16 pkt_prod, bd_prod;
2688 int nbd, txq_index, fp_index, txdata_index;
2689 dma_addr_t mapping;
2690 u32 xmit_type = bnx2x_xmit_type(bp, skb);
2691 int i;
2692 u8 hlen = 0;
2693 __le16 pkt_size = 0;
2694 struct ethhdr *eth;
2695 u8 mac_type = UNICAST_ADDRESS;
2697 #ifdef BNX2X_STOP_ON_ERROR
2698 if (unlikely(bp->panic))
2699 return NETDEV_TX_BUSY;
2700 #endif
2702 txq_index = skb_get_queue_mapping(skb);
2703 txq = netdev_get_tx_queue(dev, txq_index);
2705 BUG_ON(txq_index >= MAX_ETH_TXQ_IDX(bp) + FCOE_PRESENT);
2707 /* decode the fastpath index and the cos index from the txq */
2708 fp_index = TXQ_TO_FP(txq_index);
2709 txdata_index = TXQ_TO_COS(txq_index);
2711 #ifdef BCM_CNIC
2713 * Override the above for the FCoE queue:
2714 * - FCoE fp entry is right after the ETH entries.
2715 * - FCoE L2 queue uses bp->txdata[0] only.
2717 if (unlikely(!NO_FCOE(bp) && (txq_index ==
2718 bnx2x_fcoe_tx(bp, txq_index)))) {
2719 fp_index = FCOE_IDX;
2720 txdata_index = 0;
2722 #endif
2724 /* enable this debug print to view the transmission queue being used
2725 DP(NETIF_MSG_TX_QUEUED, "indices: txq %d, fp %d, txdata %d\n",
2726 txq_index, fp_index, txdata_index); */
2728 /* locate the fastpath and the txdata */
2729 fp = &bp->fp[fp_index];
2730 txdata = &fp->txdata[txdata_index];
2732 /* enable this debug print to view the tranmission details
2733 DP(NETIF_MSG_TX_QUEUED,
2734 "transmitting packet cid %d fp index %d txdata_index %d tx_data ptr %p fp pointer %p\n",
2735 txdata->cid, fp_index, txdata_index, txdata, fp); */
2737 if (unlikely(bnx2x_tx_avail(bp, txdata) <
2738 (skb_shinfo(skb)->nr_frags + 3))) {
2739 fp->eth_q_stats.driver_xoff++;
2740 netif_tx_stop_queue(txq);
2741 BNX2X_ERR("BUG! Tx ring full when queue awake!\n");
2742 return NETDEV_TX_BUSY;
2745 DP(NETIF_MSG_TX_QUEUED,
2746 "queue[%d]: SKB: summed %x protocol %x protocol(%x,%x) gso type %x xmit_type %x\n",
2747 txq_index, skb->ip_summed, skb->protocol, ipv6_hdr(skb)->nexthdr,
2748 ip_hdr(skb)->protocol, skb_shinfo(skb)->gso_type, xmit_type);
2750 eth = (struct ethhdr *)skb->data;
2752 /* set flag according to packet type (UNICAST_ADDRESS is default)*/
2753 if (unlikely(is_multicast_ether_addr(eth->h_dest))) {
2754 if (is_broadcast_ether_addr(eth->h_dest))
2755 mac_type = BROADCAST_ADDRESS;
2756 else
2757 mac_type = MULTICAST_ADDRESS;
2760 #if (MAX_SKB_FRAGS >= MAX_FETCH_BD - 3)
2761 /* First, check if we need to linearize the skb (due to FW
2762 restrictions). No need to check fragmentation if page size > 8K
2763 (there will be no violation to FW restrictions) */
2764 if (bnx2x_pkt_req_lin(bp, skb, xmit_type)) {
2765 /* Statistics of linearization */
2766 bp->lin_cnt++;
2767 if (skb_linearize(skb) != 0) {
2768 DP(NETIF_MSG_TX_QUEUED,
2769 "SKB linearization failed - silently dropping this SKB\n");
2770 dev_kfree_skb_any(skb);
2771 return NETDEV_TX_OK;
2774 #endif
2775 /* Map skb linear data for DMA */
2776 mapping = dma_map_single(&bp->pdev->dev, skb->data,
2777 skb_headlen(skb), DMA_TO_DEVICE);
2778 if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
2779 DP(NETIF_MSG_TX_QUEUED,
2780 "SKB mapping failed - silently dropping this SKB\n");
2781 dev_kfree_skb_any(skb);
2782 return NETDEV_TX_OK;
2785 Please read carefully. First we use one BD which we mark as start,
2786 then we have a parsing info BD (used for TSO or xsum),
2787 and only then we have the rest of the TSO BDs.
2788 (don't forget to mark the last one as last,
2789 and to unmap only AFTER you write to the BD ...)
2790 And above all, all pdb sizes are in words - NOT DWORDS!
2793 /* get current pkt produced now - advance it just before sending packet
2794 * since mapping of pages may fail and cause packet to be dropped
2796 pkt_prod = txdata->tx_pkt_prod;
2797 bd_prod = TX_BD(txdata->tx_bd_prod);
2799 /* get a tx_buf and first BD
2800 * tx_start_bd may be changed during SPLIT,
2801 * but first_bd will always stay first
2803 tx_buf = &txdata->tx_buf_ring[TX_BD(pkt_prod)];
2804 tx_start_bd = &txdata->tx_desc_ring[bd_prod].start_bd;
2805 first_bd = tx_start_bd;
2807 tx_start_bd->bd_flags.as_bitfield = ETH_TX_BD_FLAGS_START_BD;
2808 SET_FLAG(tx_start_bd->general_data, ETH_TX_START_BD_ETH_ADDR_TYPE,
2809 mac_type);
2811 /* header nbd */
2812 SET_FLAG(tx_start_bd->general_data, ETH_TX_START_BD_HDR_NBDS, 1);
2814 /* remember the first BD of the packet */
2815 tx_buf->first_bd = txdata->tx_bd_prod;
2816 tx_buf->skb = skb;
2817 tx_buf->flags = 0;
2819 DP(NETIF_MSG_TX_QUEUED,
2820 "sending pkt %u @%p next_idx %u bd %u @%p\n",
2821 pkt_prod, tx_buf, txdata->tx_pkt_prod, bd_prod, tx_start_bd);
2823 if (vlan_tx_tag_present(skb)) {
2824 tx_start_bd->vlan_or_ethertype =
2825 cpu_to_le16(vlan_tx_tag_get(skb));
2826 tx_start_bd->bd_flags.as_bitfield |=
2827 (X_ETH_OUTBAND_VLAN << ETH_TX_BD_FLAGS_VLAN_MODE_SHIFT);
2828 } else
2829 tx_start_bd->vlan_or_ethertype = cpu_to_le16(pkt_prod);
2831 /* turn on parsing and get a BD */
2832 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
2834 if (xmit_type & XMIT_CSUM)
2835 bnx2x_set_sbd_csum(bp, skb, tx_start_bd, xmit_type);
2837 if (!CHIP_IS_E1x(bp)) {
2838 pbd_e2 = &txdata->tx_desc_ring[bd_prod].parse_bd_e2;
2839 memset(pbd_e2, 0, sizeof(struct eth_tx_parse_bd_e2));
2840 /* Set PBD in checksum offload case */
2841 if (xmit_type & XMIT_CSUM)
2842 hlen = bnx2x_set_pbd_csum_e2(bp, skb,
2843 &pbd_e2_parsing_data,
2844 xmit_type);
2845 if (IS_MF_SI(bp)) {
2847 * fill in the MAC addresses in the PBD - for local
2848 * switching
2850 bnx2x_set_fw_mac_addr(&pbd_e2->src_mac_addr_hi,
2851 &pbd_e2->src_mac_addr_mid,
2852 &pbd_e2->src_mac_addr_lo,
2853 eth->h_source);
2854 bnx2x_set_fw_mac_addr(&pbd_e2->dst_mac_addr_hi,
2855 &pbd_e2->dst_mac_addr_mid,
2856 &pbd_e2->dst_mac_addr_lo,
2857 eth->h_dest);
2859 } else {
2860 pbd_e1x = &txdata->tx_desc_ring[bd_prod].parse_bd_e1x;
2861 memset(pbd_e1x, 0, sizeof(struct eth_tx_parse_bd_e1x));
2862 /* Set PBD in checksum offload case */
2863 if (xmit_type & XMIT_CSUM)
2864 hlen = bnx2x_set_pbd_csum(bp, skb, pbd_e1x, xmit_type);
2868 /* Setup the data pointer of the first BD of the packet */
2869 tx_start_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
2870 tx_start_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
2871 nbd = 2; /* start_bd + pbd + frags (updated when pages are mapped) */
2872 tx_start_bd->nbytes = cpu_to_le16(skb_headlen(skb));
2873 pkt_size = tx_start_bd->nbytes;
2875 DP(NETIF_MSG_TX_QUEUED,
2876 "first bd @%p addr (%x:%x) nbd %d nbytes %d flags %x vlan %x\n",
2877 tx_start_bd, tx_start_bd->addr_hi, tx_start_bd->addr_lo,
2878 le16_to_cpu(tx_start_bd->nbd), le16_to_cpu(tx_start_bd->nbytes),
2879 tx_start_bd->bd_flags.as_bitfield,
2880 le16_to_cpu(tx_start_bd->vlan_or_ethertype));
2882 if (xmit_type & XMIT_GSO) {
2884 DP(NETIF_MSG_TX_QUEUED,
2885 "TSO packet len %d hlen %d total len %d tso size %d\n",
2886 skb->len, hlen, skb_headlen(skb),
2887 skb_shinfo(skb)->gso_size);
2889 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_SW_LSO;
2891 if (unlikely(skb_headlen(skb) > hlen))
2892 bd_prod = bnx2x_tx_split(bp, txdata, tx_buf,
2893 &tx_start_bd, hlen,
2894 bd_prod, ++nbd);
2895 if (!CHIP_IS_E1x(bp))
2896 bnx2x_set_pbd_gso_e2(skb, &pbd_e2_parsing_data,
2897 xmit_type);
2898 else
2899 bnx2x_set_pbd_gso(skb, pbd_e1x, xmit_type);
2902 /* Set the PBD's parsing_data field if not zero
2903 * (for the chips newer than 57711).
2905 if (pbd_e2_parsing_data)
2906 pbd_e2->parsing_data = cpu_to_le32(pbd_e2_parsing_data);
2908 tx_data_bd = (struct eth_tx_bd *)tx_start_bd;
2910 /* Handle fragmented skb */
2911 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2912 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2914 mapping = skb_frag_dma_map(&bp->pdev->dev, frag, 0,
2915 skb_frag_size(frag), DMA_TO_DEVICE);
2916 if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
2917 unsigned int pkts_compl = 0, bytes_compl = 0;
2919 DP(NETIF_MSG_TX_QUEUED,
2920 "Unable to map page - dropping packet...\n");
2922 /* we need unmap all buffers already mapped
2923 * for this SKB;
2924 * first_bd->nbd need to be properly updated
2925 * before call to bnx2x_free_tx_pkt
2927 first_bd->nbd = cpu_to_le16(nbd);
2928 bnx2x_free_tx_pkt(bp, txdata,
2929 TX_BD(txdata->tx_pkt_prod),
2930 &pkts_compl, &bytes_compl);
2931 return NETDEV_TX_OK;
2934 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
2935 tx_data_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
2936 if (total_pkt_bd == NULL)
2937 total_pkt_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
2939 tx_data_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
2940 tx_data_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
2941 tx_data_bd->nbytes = cpu_to_le16(skb_frag_size(frag));
2942 le16_add_cpu(&pkt_size, skb_frag_size(frag));
2943 nbd++;
2945 DP(NETIF_MSG_TX_QUEUED,
2946 "frag %d bd @%p addr (%x:%x) nbytes %d\n",
2947 i, tx_data_bd, tx_data_bd->addr_hi, tx_data_bd->addr_lo,
2948 le16_to_cpu(tx_data_bd->nbytes));
2951 DP(NETIF_MSG_TX_QUEUED, "last bd @%p\n", tx_data_bd);
2953 /* update with actual num BDs */
2954 first_bd->nbd = cpu_to_le16(nbd);
2956 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
2958 /* now send a tx doorbell, counting the next BD
2959 * if the packet contains or ends with it
2961 if (TX_BD_POFF(bd_prod) < nbd)
2962 nbd++;
2964 /* total_pkt_bytes should be set on the first data BD if
2965 * it's not an LSO packet and there is more than one
2966 * data BD. In this case pkt_size is limited by an MTU value.
2967 * However we prefer to set it for an LSO packet (while we don't
2968 * have to) in order to save some CPU cycles in a none-LSO
2969 * case, when we much more care about them.
2971 if (total_pkt_bd != NULL)
2972 total_pkt_bd->total_pkt_bytes = pkt_size;
2974 if (pbd_e1x)
2975 DP(NETIF_MSG_TX_QUEUED,
2976 "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",
2977 pbd_e1x, pbd_e1x->global_data, pbd_e1x->ip_hlen_w,
2978 pbd_e1x->ip_id, pbd_e1x->lso_mss, pbd_e1x->tcp_flags,
2979 pbd_e1x->tcp_pseudo_csum, pbd_e1x->tcp_send_seq,
2980 le16_to_cpu(pbd_e1x->total_hlen_w));
2981 if (pbd_e2)
2982 DP(NETIF_MSG_TX_QUEUED,
2983 "PBD (E2) @%p dst %x %x %x src %x %x %x parsing_data %x\n",
2984 pbd_e2, pbd_e2->dst_mac_addr_hi, pbd_e2->dst_mac_addr_mid,
2985 pbd_e2->dst_mac_addr_lo, pbd_e2->src_mac_addr_hi,
2986 pbd_e2->src_mac_addr_mid, pbd_e2->src_mac_addr_lo,
2987 pbd_e2->parsing_data);
2988 DP(NETIF_MSG_TX_QUEUED, "doorbell: nbd %d bd %u\n", nbd, bd_prod);
2990 netdev_tx_sent_queue(txq, skb->len);
2992 txdata->tx_pkt_prod++;
2994 * Make sure that the BD data is updated before updating the producer
2995 * since FW might read the BD right after the producer is updated.
2996 * This is only applicable for weak-ordered memory model archs such
2997 * as IA-64. The following barrier is also mandatory since FW will
2998 * assumes packets must have BDs.
3000 wmb();
3002 txdata->tx_db.data.prod += nbd;
3003 barrier();
3005 DOORBELL(bp, txdata->cid, txdata->tx_db.raw);
3007 mmiowb();
3009 txdata->tx_bd_prod += nbd;
3011 if (unlikely(bnx2x_tx_avail(bp, txdata) < MAX_SKB_FRAGS + 4)) {
3012 netif_tx_stop_queue(txq);
3014 /* paired memory barrier is in bnx2x_tx_int(), we have to keep
3015 * ordering of set_bit() in netif_tx_stop_queue() and read of
3016 * fp->bd_tx_cons */
3017 smp_mb();
3019 fp->eth_q_stats.driver_xoff++;
3020 if (bnx2x_tx_avail(bp, txdata) >= MAX_SKB_FRAGS + 4)
3021 netif_tx_wake_queue(txq);
3023 txdata->tx_pkt++;
3025 return NETDEV_TX_OK;
3029 * bnx2x_setup_tc - routine to configure net_device for multi tc
3031 * @netdev: net device to configure
3032 * @tc: number of traffic classes to enable
3034 * callback connected to the ndo_setup_tc function pointer
3036 int bnx2x_setup_tc(struct net_device *dev, u8 num_tc)
3038 int cos, prio, count, offset;
3039 struct bnx2x *bp = netdev_priv(dev);
3041 /* setup tc must be called under rtnl lock */
3042 ASSERT_RTNL();
3044 /* no traffic classes requested. aborting */
3045 if (!num_tc) {
3046 netdev_reset_tc(dev);
3047 return 0;
3050 /* requested to support too many traffic classes */
3051 if (num_tc > bp->max_cos) {
3052 BNX2X_ERR("support for too many traffic classes requested: %d. max supported is %d\n",
3053 num_tc, bp->max_cos);
3054 return -EINVAL;
3057 /* declare amount of supported traffic classes */
3058 if (netdev_set_num_tc(dev, num_tc)) {
3059 BNX2X_ERR("failed to declare %d traffic classes\n", num_tc);
3060 return -EINVAL;
3063 /* configure priority to traffic class mapping */
3064 for (prio = 0; prio < BNX2X_MAX_PRIORITY; prio++) {
3065 netdev_set_prio_tc_map(dev, prio, bp->prio_to_cos[prio]);
3066 DP(BNX2X_MSG_SP | NETIF_MSG_IFUP,
3067 "mapping priority %d to tc %d\n",
3068 prio, bp->prio_to_cos[prio]);
3072 /* Use this configuration to diffrentiate tc0 from other COSes
3073 This can be used for ets or pfc, and save the effort of setting
3074 up a multio class queue disc or negotiating DCBX with a switch
3075 netdev_set_prio_tc_map(dev, 0, 0);
3076 DP(BNX2X_MSG_SP, "mapping priority %d to tc %d\n", 0, 0);
3077 for (prio = 1; prio < 16; prio++) {
3078 netdev_set_prio_tc_map(dev, prio, 1);
3079 DP(BNX2X_MSG_SP, "mapping priority %d to tc %d\n", prio, 1);
3080 } */
3082 /* configure traffic class to transmission queue mapping */
3083 for (cos = 0; cos < bp->max_cos; cos++) {
3084 count = BNX2X_NUM_ETH_QUEUES(bp);
3085 offset = cos * MAX_TXQS_PER_COS;
3086 netdev_set_tc_queue(dev, cos, count, offset);
3087 DP(BNX2X_MSG_SP | NETIF_MSG_IFUP,
3088 "mapping tc %d to offset %d count %d\n",
3089 cos, offset, count);
3092 return 0;
3095 /* called with rtnl_lock */
3096 int bnx2x_change_mac_addr(struct net_device *dev, void *p)
3098 struct sockaddr *addr = p;
3099 struct bnx2x *bp = netdev_priv(dev);
3100 int rc = 0;
3102 if (!bnx2x_is_valid_ether_addr(bp, addr->sa_data)) {
3103 BNX2X_ERR("Requested MAC address is not valid\n");
3104 return -EINVAL;
3107 #ifdef BCM_CNIC
3108 if (IS_MF_STORAGE_SD(bp) && !is_zero_ether_addr(addr->sa_data)) {
3109 BNX2X_ERR("Can't configure non-zero address on iSCSI or FCoE functions in MF-SD mode\n");
3110 return -EINVAL;
3112 #endif
3114 if (netif_running(dev)) {
3115 rc = bnx2x_set_eth_mac(bp, false);
3116 if (rc)
3117 return rc;
3120 dev->addr_assign_type &= ~NET_ADDR_RANDOM;
3121 memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
3123 if (netif_running(dev))
3124 rc = bnx2x_set_eth_mac(bp, true);
3126 return rc;
3129 static void bnx2x_free_fp_mem_at(struct bnx2x *bp, int fp_index)
3131 union host_hc_status_block *sb = &bnx2x_fp(bp, fp_index, status_blk);
3132 struct bnx2x_fastpath *fp = &bp->fp[fp_index];
3133 u8 cos;
3135 /* Common */
3136 #ifdef BCM_CNIC
3137 if (IS_FCOE_IDX(fp_index)) {
3138 memset(sb, 0, sizeof(union host_hc_status_block));
3139 fp->status_blk_mapping = 0;
3141 } else {
3142 #endif
3143 /* status blocks */
3144 if (!CHIP_IS_E1x(bp))
3145 BNX2X_PCI_FREE(sb->e2_sb,
3146 bnx2x_fp(bp, fp_index,
3147 status_blk_mapping),
3148 sizeof(struct host_hc_status_block_e2));
3149 else
3150 BNX2X_PCI_FREE(sb->e1x_sb,
3151 bnx2x_fp(bp, fp_index,
3152 status_blk_mapping),
3153 sizeof(struct host_hc_status_block_e1x));
3154 #ifdef BCM_CNIC
3156 #endif
3157 /* Rx */
3158 if (!skip_rx_queue(bp, fp_index)) {
3159 bnx2x_free_rx_bds(fp);
3161 /* fastpath rx rings: rx_buf rx_desc rx_comp */
3162 BNX2X_FREE(bnx2x_fp(bp, fp_index, rx_buf_ring));
3163 BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_desc_ring),
3164 bnx2x_fp(bp, fp_index, rx_desc_mapping),
3165 sizeof(struct eth_rx_bd) * NUM_RX_BD);
3167 BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_comp_ring),
3168 bnx2x_fp(bp, fp_index, rx_comp_mapping),
3169 sizeof(struct eth_fast_path_rx_cqe) *
3170 NUM_RCQ_BD);
3172 /* SGE ring */
3173 BNX2X_FREE(bnx2x_fp(bp, fp_index, rx_page_ring));
3174 BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_sge_ring),
3175 bnx2x_fp(bp, fp_index, rx_sge_mapping),
3176 BCM_PAGE_SIZE * NUM_RX_SGE_PAGES);
3179 /* Tx */
3180 if (!skip_tx_queue(bp, fp_index)) {
3181 /* fastpath tx rings: tx_buf tx_desc */
3182 for_each_cos_in_tx_queue(fp, cos) {
3183 struct bnx2x_fp_txdata *txdata = &fp->txdata[cos];
3185 DP(NETIF_MSG_IFDOWN,
3186 "freeing tx memory of fp %d cos %d cid %d\n",
3187 fp_index, cos, txdata->cid);
3189 BNX2X_FREE(txdata->tx_buf_ring);
3190 BNX2X_PCI_FREE(txdata->tx_desc_ring,
3191 txdata->tx_desc_mapping,
3192 sizeof(union eth_tx_bd_types) * NUM_TX_BD);
3195 /* end of fastpath */
3198 void bnx2x_free_fp_mem(struct bnx2x *bp)
3200 int i;
3201 for_each_queue(bp, i)
3202 bnx2x_free_fp_mem_at(bp, i);
3205 static inline void set_sb_shortcuts(struct bnx2x *bp, int index)
3207 union host_hc_status_block status_blk = bnx2x_fp(bp, index, status_blk);
3208 if (!CHIP_IS_E1x(bp)) {
3209 bnx2x_fp(bp, index, sb_index_values) =
3210 (__le16 *)status_blk.e2_sb->sb.index_values;
3211 bnx2x_fp(bp, index, sb_running_index) =
3212 (__le16 *)status_blk.e2_sb->sb.running_index;
3213 } else {
3214 bnx2x_fp(bp, index, sb_index_values) =
3215 (__le16 *)status_blk.e1x_sb->sb.index_values;
3216 bnx2x_fp(bp, index, sb_running_index) =
3217 (__le16 *)status_blk.e1x_sb->sb.running_index;
3221 static int bnx2x_alloc_fp_mem_at(struct bnx2x *bp, int index)
3223 union host_hc_status_block *sb;
3224 struct bnx2x_fastpath *fp = &bp->fp[index];
3225 int ring_size = 0;
3226 u8 cos;
3227 int rx_ring_size = 0;
3229 #ifdef BCM_CNIC
3230 if (!bp->rx_ring_size && IS_MF_STORAGE_SD(bp)) {
3231 rx_ring_size = MIN_RX_SIZE_NONTPA;
3232 bp->rx_ring_size = rx_ring_size;
3233 } else
3234 #endif
3235 if (!bp->rx_ring_size) {
3236 u32 cfg = SHMEM_RD(bp,
3237 dev_info.port_hw_config[BP_PORT(bp)].default_cfg);
3239 rx_ring_size = MAX_RX_AVAIL/BNX2X_NUM_RX_QUEUES(bp);
3241 /* Dercease ring size for 1G functions */
3242 if ((cfg & PORT_HW_CFG_NET_SERDES_IF_MASK) ==
3243 PORT_HW_CFG_NET_SERDES_IF_SGMII)
3244 rx_ring_size /= 10;
3246 /* allocate at least number of buffers required by FW */
3247 rx_ring_size = max_t(int, bp->disable_tpa ? MIN_RX_SIZE_NONTPA :
3248 MIN_RX_SIZE_TPA, rx_ring_size);
3250 bp->rx_ring_size = rx_ring_size;
3251 } else /* if rx_ring_size specified - use it */
3252 rx_ring_size = bp->rx_ring_size;
3254 /* Common */
3255 sb = &bnx2x_fp(bp, index, status_blk);
3256 #ifdef BCM_CNIC
3257 if (!IS_FCOE_IDX(index)) {
3258 #endif
3259 /* status blocks */
3260 if (!CHIP_IS_E1x(bp))
3261 BNX2X_PCI_ALLOC(sb->e2_sb,
3262 &bnx2x_fp(bp, index, status_blk_mapping),
3263 sizeof(struct host_hc_status_block_e2));
3264 else
3265 BNX2X_PCI_ALLOC(sb->e1x_sb,
3266 &bnx2x_fp(bp, index, status_blk_mapping),
3267 sizeof(struct host_hc_status_block_e1x));
3268 #ifdef BCM_CNIC
3270 #endif
3272 /* FCoE Queue uses Default SB and doesn't ACK the SB, thus no need to
3273 * set shortcuts for it.
3275 if (!IS_FCOE_IDX(index))
3276 set_sb_shortcuts(bp, index);
3278 /* Tx */
3279 if (!skip_tx_queue(bp, index)) {
3280 /* fastpath tx rings: tx_buf tx_desc */
3281 for_each_cos_in_tx_queue(fp, cos) {
3282 struct bnx2x_fp_txdata *txdata = &fp->txdata[cos];
3284 DP(NETIF_MSG_IFUP,
3285 "allocating tx memory of fp %d cos %d\n",
3286 index, cos);
3288 BNX2X_ALLOC(txdata->tx_buf_ring,
3289 sizeof(struct sw_tx_bd) * NUM_TX_BD);
3290 BNX2X_PCI_ALLOC(txdata->tx_desc_ring,
3291 &txdata->tx_desc_mapping,
3292 sizeof(union eth_tx_bd_types) * NUM_TX_BD);
3296 /* Rx */
3297 if (!skip_rx_queue(bp, index)) {
3298 /* fastpath rx rings: rx_buf rx_desc rx_comp */
3299 BNX2X_ALLOC(bnx2x_fp(bp, index, rx_buf_ring),
3300 sizeof(struct sw_rx_bd) * NUM_RX_BD);
3301 BNX2X_PCI_ALLOC(bnx2x_fp(bp, index, rx_desc_ring),
3302 &bnx2x_fp(bp, index, rx_desc_mapping),
3303 sizeof(struct eth_rx_bd) * NUM_RX_BD);
3305 BNX2X_PCI_ALLOC(bnx2x_fp(bp, index, rx_comp_ring),
3306 &bnx2x_fp(bp, index, rx_comp_mapping),
3307 sizeof(struct eth_fast_path_rx_cqe) *
3308 NUM_RCQ_BD);
3310 /* SGE ring */
3311 BNX2X_ALLOC(bnx2x_fp(bp, index, rx_page_ring),
3312 sizeof(struct sw_rx_page) * NUM_RX_SGE);
3313 BNX2X_PCI_ALLOC(bnx2x_fp(bp, index, rx_sge_ring),
3314 &bnx2x_fp(bp, index, rx_sge_mapping),
3315 BCM_PAGE_SIZE * NUM_RX_SGE_PAGES);
3316 /* RX BD ring */
3317 bnx2x_set_next_page_rx_bd(fp);
3319 /* CQ ring */
3320 bnx2x_set_next_page_rx_cq(fp);
3322 /* BDs */
3323 ring_size = bnx2x_alloc_rx_bds(fp, rx_ring_size);
3324 if (ring_size < rx_ring_size)
3325 goto alloc_mem_err;
3328 return 0;
3330 /* handles low memory cases */
3331 alloc_mem_err:
3332 BNX2X_ERR("Unable to allocate full memory for queue %d (size %d)\n",
3333 index, ring_size);
3334 /* FW will drop all packets if queue is not big enough,
3335 * In these cases we disable the queue
3336 * Min size is different for OOO, TPA and non-TPA queues
3338 if (ring_size < (fp->disable_tpa ?
3339 MIN_RX_SIZE_NONTPA : MIN_RX_SIZE_TPA)) {
3340 /* release memory allocated for this queue */
3341 bnx2x_free_fp_mem_at(bp, index);
3342 return -ENOMEM;
3344 return 0;
3347 int bnx2x_alloc_fp_mem(struct bnx2x *bp)
3349 int i;
3352 * 1. Allocate FP for leading - fatal if error
3353 * 2. {CNIC} Allocate FCoE FP - fatal if error
3354 * 3. {CNIC} Allocate OOO + FWD - disable OOO if error
3355 * 4. Allocate RSS - fix number of queues if error
3358 /* leading */
3359 if (bnx2x_alloc_fp_mem_at(bp, 0))
3360 return -ENOMEM;
3362 #ifdef BCM_CNIC
3363 if (!NO_FCOE(bp))
3364 /* FCoE */
3365 if (bnx2x_alloc_fp_mem_at(bp, FCOE_IDX))
3366 /* we will fail load process instead of mark
3367 * NO_FCOE_FLAG
3369 return -ENOMEM;
3370 #endif
3372 /* RSS */
3373 for_each_nondefault_eth_queue(bp, i)
3374 if (bnx2x_alloc_fp_mem_at(bp, i))
3375 break;
3377 /* handle memory failures */
3378 if (i != BNX2X_NUM_ETH_QUEUES(bp)) {
3379 int delta = BNX2X_NUM_ETH_QUEUES(bp) - i;
3381 WARN_ON(delta < 0);
3382 #ifdef BCM_CNIC
3384 * move non eth FPs next to last eth FP
3385 * must be done in that order
3386 * FCOE_IDX < FWD_IDX < OOO_IDX
3389 /* move FCoE fp even NO_FCOE_FLAG is on */
3390 bnx2x_move_fp(bp, FCOE_IDX, FCOE_IDX - delta);
3391 #endif
3392 bp->num_queues -= delta;
3393 BNX2X_ERR("Adjusted num of queues from %d to %d\n",
3394 bp->num_queues + delta, bp->num_queues);
3397 return 0;
3400 void bnx2x_free_mem_bp(struct bnx2x *bp)
3402 kfree(bp->fp);
3403 kfree(bp->msix_table);
3404 kfree(bp->ilt);
3407 int __devinit bnx2x_alloc_mem_bp(struct bnx2x *bp)
3409 struct bnx2x_fastpath *fp;
3410 struct msix_entry *tbl;
3411 struct bnx2x_ilt *ilt;
3412 int msix_table_size = 0;
3415 * The biggest MSI-X table we might need is as a maximum number of fast
3416 * path IGU SBs plus default SB (for PF).
3418 msix_table_size = bp->igu_sb_cnt + 1;
3420 /* fp array: RSS plus CNIC related L2 queues */
3421 fp = kcalloc(BNX2X_MAX_RSS_COUNT(bp) + NON_ETH_CONTEXT_USE,
3422 sizeof(*fp), GFP_KERNEL);
3423 if (!fp)
3424 goto alloc_err;
3425 bp->fp = fp;
3427 /* msix table */
3428 tbl = kcalloc(msix_table_size, sizeof(*tbl), GFP_KERNEL);
3429 if (!tbl)
3430 goto alloc_err;
3431 bp->msix_table = tbl;
3433 /* ilt */
3434 ilt = kzalloc(sizeof(*ilt), GFP_KERNEL);
3435 if (!ilt)
3436 goto alloc_err;
3437 bp->ilt = ilt;
3439 return 0;
3440 alloc_err:
3441 bnx2x_free_mem_bp(bp);
3442 return -ENOMEM;
3446 int bnx2x_reload_if_running(struct net_device *dev)
3448 struct bnx2x *bp = netdev_priv(dev);
3450 if (unlikely(!netif_running(dev)))
3451 return 0;
3453 bnx2x_nic_unload(bp, UNLOAD_NORMAL);
3454 return bnx2x_nic_load(bp, LOAD_NORMAL);
3457 int bnx2x_get_cur_phy_idx(struct bnx2x *bp)
3459 u32 sel_phy_idx = 0;
3460 if (bp->link_params.num_phys <= 1)
3461 return INT_PHY;
3463 if (bp->link_vars.link_up) {
3464 sel_phy_idx = EXT_PHY1;
3465 /* In case link is SERDES, check if the EXT_PHY2 is the one */
3466 if ((bp->link_vars.link_status & LINK_STATUS_SERDES_LINK) &&
3467 (bp->link_params.phy[EXT_PHY2].supported & SUPPORTED_FIBRE))
3468 sel_phy_idx = EXT_PHY2;
3469 } else {
3471 switch (bnx2x_phy_selection(&bp->link_params)) {
3472 case PORT_HW_CFG_PHY_SELECTION_HARDWARE_DEFAULT:
3473 case PORT_HW_CFG_PHY_SELECTION_FIRST_PHY:
3474 case PORT_HW_CFG_PHY_SELECTION_FIRST_PHY_PRIORITY:
3475 sel_phy_idx = EXT_PHY1;
3476 break;
3477 case PORT_HW_CFG_PHY_SELECTION_SECOND_PHY:
3478 case PORT_HW_CFG_PHY_SELECTION_SECOND_PHY_PRIORITY:
3479 sel_phy_idx = EXT_PHY2;
3480 break;
3484 return sel_phy_idx;
3487 int bnx2x_get_link_cfg_idx(struct bnx2x *bp)
3489 u32 sel_phy_idx = bnx2x_get_cur_phy_idx(bp);
3491 * The selected actived PHY is always after swapping (in case PHY
3492 * swapping is enabled). So when swapping is enabled, we need to reverse
3493 * the configuration
3496 if (bp->link_params.multi_phy_config &
3497 PORT_HW_CFG_PHY_SWAPPED_ENABLED) {
3498 if (sel_phy_idx == EXT_PHY1)
3499 sel_phy_idx = EXT_PHY2;
3500 else if (sel_phy_idx == EXT_PHY2)
3501 sel_phy_idx = EXT_PHY1;
3503 return LINK_CONFIG_IDX(sel_phy_idx);
3506 #if defined(NETDEV_FCOE_WWNN) && defined(BCM_CNIC)
3507 int bnx2x_fcoe_get_wwn(struct net_device *dev, u64 *wwn, int type)
3509 struct bnx2x *bp = netdev_priv(dev);
3510 struct cnic_eth_dev *cp = &bp->cnic_eth_dev;
3512 switch (type) {
3513 case NETDEV_FCOE_WWNN:
3514 *wwn = HILO_U64(cp->fcoe_wwn_node_name_hi,
3515 cp->fcoe_wwn_node_name_lo);
3516 break;
3517 case NETDEV_FCOE_WWPN:
3518 *wwn = HILO_U64(cp->fcoe_wwn_port_name_hi,
3519 cp->fcoe_wwn_port_name_lo);
3520 break;
3521 default:
3522 BNX2X_ERR("Wrong WWN type requested - %d\n", type);
3523 return -EINVAL;
3526 return 0;
3528 #endif
3530 /* called with rtnl_lock */
3531 int bnx2x_change_mtu(struct net_device *dev, int new_mtu)
3533 struct bnx2x *bp = netdev_priv(dev);
3535 if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
3536 BNX2X_ERR("Can't perform change MTU during parity recovery\n");
3537 return -EAGAIN;
3540 if ((new_mtu > ETH_MAX_JUMBO_PACKET_SIZE) ||
3541 ((new_mtu + ETH_HLEN) < ETH_MIN_PACKET_SIZE)) {
3542 BNX2X_ERR("Can't support requested MTU size\n");
3543 return -EINVAL;
3546 /* This does not race with packet allocation
3547 * because the actual alloc size is
3548 * only updated as part of load
3550 dev->mtu = new_mtu;
3552 bp->gro_check = bnx2x_need_gro_check(new_mtu);
3554 return bnx2x_reload_if_running(dev);
3557 netdev_features_t bnx2x_fix_features(struct net_device *dev,
3558 netdev_features_t features)
3560 struct bnx2x *bp = netdev_priv(dev);
3562 /* TPA requires Rx CSUM offloading */
3563 if (!(features & NETIF_F_RXCSUM) || bp->disable_tpa) {
3564 features &= ~NETIF_F_LRO;
3565 features &= ~NETIF_F_GRO;
3568 return features;
3571 int bnx2x_set_features(struct net_device *dev, netdev_features_t features)
3573 struct bnx2x *bp = netdev_priv(dev);
3574 u32 flags = bp->flags;
3575 bool bnx2x_reload = false;
3577 if (features & NETIF_F_LRO)
3578 flags |= TPA_ENABLE_FLAG;
3579 else
3580 flags &= ~TPA_ENABLE_FLAG;
3582 if (features & NETIF_F_GRO)
3583 flags |= GRO_ENABLE_FLAG;
3584 else
3585 flags &= ~GRO_ENABLE_FLAG;
3587 if (features & NETIF_F_LOOPBACK) {
3588 if (bp->link_params.loopback_mode != LOOPBACK_BMAC) {
3589 bp->link_params.loopback_mode = LOOPBACK_BMAC;
3590 bnx2x_reload = true;
3592 } else {
3593 if (bp->link_params.loopback_mode != LOOPBACK_NONE) {
3594 bp->link_params.loopback_mode = LOOPBACK_NONE;
3595 bnx2x_reload = true;
3599 if (flags ^ bp->flags) {
3600 bp->flags = flags;
3601 bnx2x_reload = true;
3604 if (bnx2x_reload) {
3605 if (bp->recovery_state == BNX2X_RECOVERY_DONE)
3606 return bnx2x_reload_if_running(dev);
3607 /* else: bnx2x_nic_load() will be called at end of recovery */
3610 return 0;
3613 void bnx2x_tx_timeout(struct net_device *dev)
3615 struct bnx2x *bp = netdev_priv(dev);
3617 #ifdef BNX2X_STOP_ON_ERROR
3618 if (!bp->panic)
3619 bnx2x_panic();
3620 #endif
3622 smp_mb__before_clear_bit();
3623 set_bit(BNX2X_SP_RTNL_TX_TIMEOUT, &bp->sp_rtnl_state);
3624 smp_mb__after_clear_bit();
3626 /* This allows the netif to be shutdown gracefully before resetting */
3627 schedule_delayed_work(&bp->sp_rtnl_task, 0);
3630 int bnx2x_suspend(struct pci_dev *pdev, pm_message_t state)
3632 struct net_device *dev = pci_get_drvdata(pdev);
3633 struct bnx2x *bp;
3635 if (!dev) {
3636 dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
3637 return -ENODEV;
3639 bp = netdev_priv(dev);
3641 rtnl_lock();
3643 pci_save_state(pdev);
3645 if (!netif_running(dev)) {
3646 rtnl_unlock();
3647 return 0;
3650 netif_device_detach(dev);
3652 bnx2x_nic_unload(bp, UNLOAD_CLOSE);
3654 bnx2x_set_power_state(bp, pci_choose_state(pdev, state));
3656 rtnl_unlock();
3658 return 0;
3661 int bnx2x_resume(struct pci_dev *pdev)
3663 struct net_device *dev = pci_get_drvdata(pdev);
3664 struct bnx2x *bp;
3665 int rc;
3667 if (!dev) {
3668 dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
3669 return -ENODEV;
3671 bp = netdev_priv(dev);
3673 if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
3674 BNX2X_ERR("Handling parity error recovery. Try again later\n");
3675 return -EAGAIN;
3678 rtnl_lock();
3680 pci_restore_state(pdev);
3682 if (!netif_running(dev)) {
3683 rtnl_unlock();
3684 return 0;
3687 bnx2x_set_power_state(bp, PCI_D0);
3688 netif_device_attach(dev);
3690 rc = bnx2x_nic_load(bp, LOAD_OPEN);
3692 rtnl_unlock();
3694 return rc;
3698 void bnx2x_set_ctx_validation(struct bnx2x *bp, struct eth_context *cxt,
3699 u32 cid)
3701 /* ustorm cxt validation */
3702 cxt->ustorm_ag_context.cdu_usage =
3703 CDU_RSRVD_VALUE_TYPE_A(HW_CID(bp, cid),
3704 CDU_REGION_NUMBER_UCM_AG, ETH_CONNECTION_TYPE);
3705 /* xcontext validation */
3706 cxt->xstorm_ag_context.cdu_reserved =
3707 CDU_RSRVD_VALUE_TYPE_A(HW_CID(bp, cid),
3708 CDU_REGION_NUMBER_XCM_AG, ETH_CONNECTION_TYPE);
3711 static inline void storm_memset_hc_timeout(struct bnx2x *bp, u8 port,
3712 u8 fw_sb_id, u8 sb_index,
3713 u8 ticks)
3716 u32 addr = BAR_CSTRORM_INTMEM +
3717 CSTORM_STATUS_BLOCK_DATA_TIMEOUT_OFFSET(fw_sb_id, sb_index);
3718 REG_WR8(bp, addr, ticks);
3719 DP(NETIF_MSG_IFUP,
3720 "port %x fw_sb_id %d sb_index %d ticks %d\n",
3721 port, fw_sb_id, sb_index, ticks);
3724 static inline void storm_memset_hc_disable(struct bnx2x *bp, u8 port,
3725 u16 fw_sb_id, u8 sb_index,
3726 u8 disable)
3728 u32 enable_flag = disable ? 0 : (1 << HC_INDEX_DATA_HC_ENABLED_SHIFT);
3729 u32 addr = BAR_CSTRORM_INTMEM +
3730 CSTORM_STATUS_BLOCK_DATA_FLAGS_OFFSET(fw_sb_id, sb_index);
3731 u16 flags = REG_RD16(bp, addr);
3732 /* clear and set */
3733 flags &= ~HC_INDEX_DATA_HC_ENABLED;
3734 flags |= enable_flag;
3735 REG_WR16(bp, addr, flags);
3736 DP(NETIF_MSG_IFUP,
3737 "port %x fw_sb_id %d sb_index %d disable %d\n",
3738 port, fw_sb_id, sb_index, disable);
3741 void bnx2x_update_coalesce_sb_index(struct bnx2x *bp, u8 fw_sb_id,
3742 u8 sb_index, u8 disable, u16 usec)
3744 int port = BP_PORT(bp);
3745 u8 ticks = usec / BNX2X_BTR;
3747 storm_memset_hc_timeout(bp, port, fw_sb_id, sb_index, ticks);
3749 disable = disable ? 1 : (usec ? 0 : 1);
3750 storm_memset_hc_disable(bp, port, fw_sb_id, sb_index, disable);