1 /* bnx2x_cmn.c: Broadcom Everest network driver.
3 * Copyright (c) 2007-2011 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 #include <linux/etherdevice.h>
19 #include <linux/if_vlan.h>
20 #include <linux/interrupt.h>
23 #include <net/ip6_checksum.h>
24 #include <linux/firmware.h>
25 #include <linux/prefetch.h>
26 #include "bnx2x_cmn.h"
27 #include "bnx2x_init.h"
33 * bnx2x_bz_fp - zero content of the fastpath structure.
36 * @index: fastpath index to be zeroed
38 * Makes sure the contents of the bp->fp[index].napi is kept
41 static inline void bnx2x_bz_fp(struct bnx2x *bp, int index)
43 struct bnx2x_fastpath *fp = &bp->fp[index];
44 struct napi_struct orig_napi = fp->napi;
45 /* bzero bnx2x_fastpath contents */
46 memset(fp, 0, sizeof(*fp));
48 /* Restore the NAPI object as it has been already initialized */
53 * bnx2x_move_fp - move content of the fastpath structure.
56 * @from: source FP index
57 * @to: destination FP index
59 * Makes sure the contents of the bp->fp[to].napi is kept
62 static inline void bnx2x_move_fp(struct bnx2x *bp, int from, int to)
64 struct bnx2x_fastpath *from_fp = &bp->fp[from];
65 struct bnx2x_fastpath *to_fp = &bp->fp[to];
66 struct napi_struct orig_napi = to_fp->napi;
67 /* Move bnx2x_fastpath contents */
68 memcpy(to_fp, from_fp, sizeof(*to_fp));
71 /* Restore the NAPI object as it has been already initialized */
72 to_fp->napi = orig_napi;
75 int load_count[2][3] = { {0} }; /* per-path: 0-common, 1-port0, 2-port1 */
77 /* free skb in the packet ring at pos idx
78 * return idx of last bd freed
80 static u16 bnx2x_free_tx_pkt(struct bnx2x *bp, struct bnx2x_fastpath *fp,
83 struct sw_tx_bd *tx_buf = &fp->tx_buf_ring[idx];
84 struct eth_tx_start_bd *tx_start_bd;
85 struct eth_tx_bd *tx_data_bd;
86 struct sk_buff *skb = tx_buf->skb;
87 u16 bd_idx = TX_BD(tx_buf->first_bd), new_cons;
90 /* prefetch skb end pointer to speedup dev_kfree_skb() */
93 DP(BNX2X_MSG_FP, "fp[%d]: pkt_idx %d buff @(%p)->skb %p\n",
94 fp->index, idx, tx_buf, skb);
97 DP(BNX2X_MSG_OFF, "free bd_idx %d\n", bd_idx);
98 tx_start_bd = &fp->tx_desc_ring[bd_idx].start_bd;
99 dma_unmap_single(&bp->pdev->dev, BD_UNMAP_ADDR(tx_start_bd),
100 BD_UNMAP_LEN(tx_start_bd), DMA_TO_DEVICE);
103 nbd = le16_to_cpu(tx_start_bd->nbd) - 1;
104 #ifdef BNX2X_STOP_ON_ERROR
105 if ((nbd - 1) > (MAX_SKB_FRAGS + 2)) {
106 BNX2X_ERR("BAD nbd!\n");
110 new_cons = nbd + tx_buf->first_bd;
112 /* Get the next bd */
113 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
115 /* Skip a parse bd... */
117 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
119 /* ...and the TSO split header bd since they have no mapping */
120 if (tx_buf->flags & BNX2X_TSO_SPLIT_BD) {
122 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
128 DP(BNX2X_MSG_OFF, "free frag bd_idx %d\n", bd_idx);
129 tx_data_bd = &fp->tx_desc_ring[bd_idx].reg_bd;
130 dma_unmap_page(&bp->pdev->dev, BD_UNMAP_ADDR(tx_data_bd),
131 BD_UNMAP_LEN(tx_data_bd), DMA_TO_DEVICE);
133 bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
138 dev_kfree_skb_any(skb);
139 tx_buf->first_bd = 0;
145 int bnx2x_tx_int(struct bnx2x_fastpath *fp)
147 struct bnx2x *bp = fp->bp;
148 struct netdev_queue *txq;
149 u16 hw_cons, sw_cons, bd_cons = fp->tx_bd_cons;
151 #ifdef BNX2X_STOP_ON_ERROR
152 if (unlikely(bp->panic))
156 txq = netdev_get_tx_queue(bp->dev, fp->index);
157 hw_cons = le16_to_cpu(*fp->tx_cons_sb);
158 sw_cons = fp->tx_pkt_cons;
160 while (sw_cons != hw_cons) {
163 pkt_cons = TX_BD(sw_cons);
165 DP(NETIF_MSG_TX_DONE, "queue[%d]: hw_cons %u sw_cons %u "
167 fp->index, hw_cons, sw_cons, pkt_cons);
169 bd_cons = bnx2x_free_tx_pkt(bp, fp, pkt_cons);
173 fp->tx_pkt_cons = sw_cons;
174 fp->tx_bd_cons = bd_cons;
176 /* Need to make the tx_bd_cons update visible to start_xmit()
177 * before checking for netif_tx_queue_stopped(). Without the
178 * memory barrier, there is a small possibility that
179 * start_xmit() will miss it and cause the queue to be stopped
181 * On the other hand we need an rmb() here to ensure the proper
182 * ordering of bit testing in the following
183 * netif_tx_queue_stopped(txq) call.
187 if (unlikely(netif_tx_queue_stopped(txq))) {
188 /* Taking tx_lock() is needed to prevent reenabling the queue
189 * while it's empty. This could have happen if rx_action() gets
190 * suspended in bnx2x_tx_int() after the condition before
191 * netif_tx_wake_queue(), while tx_action (bnx2x_start_xmit()):
193 * stops the queue->sees fresh tx_bd_cons->releases the queue->
194 * sends some packets consuming the whole queue again->
198 __netif_tx_lock(txq, smp_processor_id());
200 if ((netif_tx_queue_stopped(txq)) &&
201 (bp->state == BNX2X_STATE_OPEN) &&
202 (bnx2x_tx_avail(fp) >= MAX_SKB_FRAGS + 3))
203 netif_tx_wake_queue(txq);
205 __netif_tx_unlock(txq);
210 static inline void bnx2x_update_last_max_sge(struct bnx2x_fastpath *fp,
213 u16 last_max = fp->last_max_sge;
215 if (SUB_S16(idx, last_max) > 0)
216 fp->last_max_sge = idx;
219 static void bnx2x_update_sge_prod(struct bnx2x_fastpath *fp,
220 struct eth_fast_path_rx_cqe *fp_cqe)
222 struct bnx2x *bp = fp->bp;
223 u16 sge_len = SGE_PAGE_ALIGN(le16_to_cpu(fp_cqe->pkt_len) -
224 le16_to_cpu(fp_cqe->len_on_bd)) >>
226 u16 last_max, last_elem, first_elem;
233 /* First mark all used pages */
234 for (i = 0; i < sge_len; i++)
235 BIT_VEC64_CLEAR_BIT(fp->sge_mask,
236 RX_SGE(le16_to_cpu(fp_cqe->sgl_or_raw_data.sgl[i])));
238 DP(NETIF_MSG_RX_STATUS, "fp_cqe->sgl[%d] = %d\n",
239 sge_len - 1, le16_to_cpu(fp_cqe->sgl_or_raw_data.sgl[sge_len - 1]));
241 /* Here we assume that the last SGE index is the biggest */
242 prefetch((void *)(fp->sge_mask));
243 bnx2x_update_last_max_sge(fp,
244 le16_to_cpu(fp_cqe->sgl_or_raw_data.sgl[sge_len - 1]));
246 last_max = RX_SGE(fp->last_max_sge);
247 last_elem = last_max >> BIT_VEC64_ELEM_SHIFT;
248 first_elem = RX_SGE(fp->rx_sge_prod) >> BIT_VEC64_ELEM_SHIFT;
250 /* If ring is not full */
251 if (last_elem + 1 != first_elem)
254 /* Now update the prod */
255 for (i = first_elem; i != last_elem; i = NEXT_SGE_MASK_ELEM(i)) {
256 if (likely(fp->sge_mask[i]))
259 fp->sge_mask[i] = BIT_VEC64_ELEM_ONE_MASK;
260 delta += BIT_VEC64_ELEM_SZ;
264 fp->rx_sge_prod += delta;
265 /* clear page-end entries */
266 bnx2x_clear_sge_mask_next_elems(fp);
269 DP(NETIF_MSG_RX_STATUS,
270 "fp->last_max_sge = %d fp->rx_sge_prod = %d\n",
271 fp->last_max_sge, fp->rx_sge_prod);
274 static void bnx2x_tpa_start(struct bnx2x_fastpath *fp, u16 queue,
275 struct sk_buff *skb, u16 cons, u16 prod,
276 struct eth_fast_path_rx_cqe *cqe)
278 struct bnx2x *bp = fp->bp;
279 struct sw_rx_bd *cons_rx_buf = &fp->rx_buf_ring[cons];
280 struct sw_rx_bd *prod_rx_buf = &fp->rx_buf_ring[prod];
281 struct eth_rx_bd *prod_bd = &fp->rx_desc_ring[prod];
283 struct bnx2x_agg_info *tpa_info = &fp->tpa_info[queue];
284 struct sw_rx_bd *first_buf = &tpa_info->first_buf;
286 /* print error if current state != stop */
287 if (tpa_info->tpa_state != BNX2X_TPA_STOP)
288 BNX2X_ERR("start of bin not in stop [%d]\n", queue);
290 /* Try to map an empty skb from the aggregation info */
291 mapping = dma_map_single(&bp->pdev->dev,
292 first_buf->skb->data,
293 fp->rx_buf_size, DMA_FROM_DEVICE);
295 * ...if it fails - move the skb from the consumer to the producer
296 * and set the current aggregation state as ERROR to drop it
297 * when TPA_STOP arrives.
300 if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
301 /* Move the BD from the consumer to the producer */
302 bnx2x_reuse_rx_skb(fp, cons, prod);
303 tpa_info->tpa_state = BNX2X_TPA_ERROR;
307 /* move empty skb from pool to prod */
308 prod_rx_buf->skb = first_buf->skb;
309 dma_unmap_addr_set(prod_rx_buf, mapping, mapping);
310 /* point prod_bd to new skb */
311 prod_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
312 prod_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
314 /* move partial skb from cons to pool (don't unmap yet) */
315 *first_buf = *cons_rx_buf;
317 /* mark bin state as START */
318 tpa_info->parsing_flags =
319 le16_to_cpu(cqe->pars_flags.flags);
320 tpa_info->vlan_tag = le16_to_cpu(cqe->vlan_tag);
321 tpa_info->tpa_state = BNX2X_TPA_START;
322 tpa_info->len_on_bd = le16_to_cpu(cqe->len_on_bd);
323 tpa_info->placement_offset = cqe->placement_offset;
325 #ifdef BNX2X_STOP_ON_ERROR
326 fp->tpa_queue_used |= (1 << queue);
327 #ifdef _ASM_GENERIC_INT_L64_H
328 DP(NETIF_MSG_RX_STATUS, "fp->tpa_queue_used = 0x%lx\n",
330 DP(NETIF_MSG_RX_STATUS, "fp->tpa_queue_used = 0x%llx\n",
336 /* Timestamp option length allowed for TPA aggregation:
338 * nop nop kind length echo val
340 #define TPA_TSTAMP_OPT_LEN 12
342 * bnx2x_set_lro_mss - calculate the approximate value of the MSS
345 * @parsing_flags: parsing flags from the START CQE
346 * @len_on_bd: total length of the first packet for the
349 * Approximate value of the MSS for this aggregation calculated using
350 * the first packet of it.
352 static inline u16 bnx2x_set_lro_mss(struct bnx2x *bp, u16 parsing_flags,
356 * TPA arrgregation won't have either IP options or TCP options
357 * other than timestamp or IPv6 extension headers.
359 u16 hdrs_len = ETH_HLEN + sizeof(struct tcphdr);
361 if (GET_FLAG(parsing_flags, PARSING_FLAGS_OVER_ETHERNET_PROTOCOL) ==
362 PRS_FLAG_OVERETH_IPV6)
363 hdrs_len += sizeof(struct ipv6hdr);
365 hdrs_len += sizeof(struct iphdr);
368 /* Check if there was a TCP timestamp, if there is it's will
369 * always be 12 bytes length: nop nop kind length echo val.
371 * Otherwise FW would close the aggregation.
373 if (parsing_flags & PARSING_FLAGS_TIME_STAMP_EXIST_FLAG)
374 hdrs_len += TPA_TSTAMP_OPT_LEN;
376 return len_on_bd - hdrs_len;
379 static int bnx2x_fill_frag_skb(struct bnx2x *bp, struct bnx2x_fastpath *fp,
380 u16 queue, struct sk_buff *skb,
381 struct eth_end_agg_rx_cqe *cqe,
384 struct sw_rx_page *rx_pg, old_rx_pg;
385 u32 i, frag_len, frag_size, pages;
388 struct bnx2x_agg_info *tpa_info = &fp->tpa_info[queue];
389 u16 len_on_bd = tpa_info->len_on_bd;
391 frag_size = le16_to_cpu(cqe->pkt_len) - len_on_bd;
392 pages = SGE_PAGE_ALIGN(frag_size) >> SGE_PAGE_SHIFT;
394 /* This is needed in order to enable forwarding support */
396 skb_shinfo(skb)->gso_size = bnx2x_set_lro_mss(bp,
397 tpa_info->parsing_flags, len_on_bd);
399 #ifdef BNX2X_STOP_ON_ERROR
400 if (pages > min_t(u32, 8, MAX_SKB_FRAGS)*SGE_PAGE_SIZE*PAGES_PER_SGE) {
401 BNX2X_ERR("SGL length is too long: %d. CQE index is %d\n",
403 BNX2X_ERR("cqe->pkt_len = %d\n", cqe->pkt_len);
409 /* Run through the SGL and compose the fragmented skb */
410 for (i = 0, j = 0; i < pages; i += PAGES_PER_SGE, j++) {
411 u16 sge_idx = RX_SGE(le16_to_cpu(cqe->sgl_or_raw_data.sgl[j]));
413 /* FW gives the indices of the SGE as if the ring is an array
414 (meaning that "next" element will consume 2 indices) */
415 frag_len = min(frag_size, (u32)(SGE_PAGE_SIZE*PAGES_PER_SGE));
416 rx_pg = &fp->rx_page_ring[sge_idx];
419 /* If we fail to allocate a substitute page, we simply stop
420 where we are and drop the whole packet */
421 err = bnx2x_alloc_rx_sge(bp, fp, sge_idx);
423 fp->eth_q_stats.rx_skb_alloc_failed++;
427 /* Unmap the page as we r going to pass it to the stack */
428 dma_unmap_page(&bp->pdev->dev,
429 dma_unmap_addr(&old_rx_pg, mapping),
430 SGE_PAGE_SIZE*PAGES_PER_SGE, DMA_FROM_DEVICE);
432 /* Add one frag and update the appropriate fields in the skb */
433 skb_fill_page_desc(skb, j, old_rx_pg.page, 0, frag_len);
435 skb->data_len += frag_len;
436 skb->truesize += frag_len;
437 skb->len += frag_len;
439 frag_size -= frag_len;
445 static void bnx2x_tpa_stop(struct bnx2x *bp, struct bnx2x_fastpath *fp,
446 u16 queue, struct eth_end_agg_rx_cqe *cqe,
449 struct bnx2x_agg_info *tpa_info = &fp->tpa_info[queue];
450 struct sw_rx_bd *rx_buf = &tpa_info->first_buf;
451 u8 pad = tpa_info->placement_offset;
452 u16 len = tpa_info->len_on_bd;
453 struct sk_buff *skb = rx_buf->skb;
455 struct sk_buff *new_skb;
456 u8 old_tpa_state = tpa_info->tpa_state;
458 tpa_info->tpa_state = BNX2X_TPA_STOP;
460 /* If we there was an error during the handling of the TPA_START -
461 * drop this aggregation.
463 if (old_tpa_state == BNX2X_TPA_ERROR)
466 /* Try to allocate the new skb */
467 new_skb = netdev_alloc_skb(bp->dev, fp->rx_buf_size);
469 /* Unmap skb in the pool anyway, as we are going to change
470 pool entry status to BNX2X_TPA_STOP even if new skb allocation
472 dma_unmap_single(&bp->pdev->dev, dma_unmap_addr(rx_buf, mapping),
473 fp->rx_buf_size, DMA_FROM_DEVICE);
475 if (likely(new_skb)) {
477 prefetch(((char *)(skb)) + L1_CACHE_BYTES);
479 #ifdef BNX2X_STOP_ON_ERROR
480 if (pad + len > fp->rx_buf_size) {
481 BNX2X_ERR("skb_put is about to fail... "
482 "pad %d len %d rx_buf_size %d\n",
483 pad, len, fp->rx_buf_size);
489 skb_reserve(skb, pad);
492 skb->protocol = eth_type_trans(skb, bp->dev);
493 skb->ip_summed = CHECKSUM_UNNECESSARY;
495 if (!bnx2x_fill_frag_skb(bp, fp, queue, skb, cqe, cqe_idx)) {
496 if (tpa_info->parsing_flags & PARSING_FLAGS_VLAN)
497 __vlan_hwaccel_put_tag(skb, tpa_info->vlan_tag);
498 napi_gro_receive(&fp->napi, skb);
500 DP(NETIF_MSG_RX_STATUS, "Failed to allocate new pages"
501 " - dropping packet!\n");
502 dev_kfree_skb_any(skb);
506 /* put new skb in bin */
507 rx_buf->skb = new_skb;
513 /* drop the packet and keep the buffer in the bin */
514 DP(NETIF_MSG_RX_STATUS,
515 "Failed to allocate or map a new skb - dropping packet!\n");
516 fp->eth_q_stats.rx_skb_alloc_failed++;
519 /* Set Toeplitz hash value in the skb using the value from the
520 * CQE (calculated by HW).
522 static inline void bnx2x_set_skb_rxhash(struct bnx2x *bp, union eth_rx_cqe *cqe,
525 /* Set Toeplitz hash from CQE */
526 if ((bp->dev->features & NETIF_F_RXHASH) &&
527 (cqe->fast_path_cqe.status_flags &
528 ETH_FAST_PATH_RX_CQE_RSS_HASH_FLG))
530 le32_to_cpu(cqe->fast_path_cqe.rss_hash_result);
533 int bnx2x_rx_int(struct bnx2x_fastpath *fp, int budget)
535 struct bnx2x *bp = fp->bp;
536 u16 bd_cons, bd_prod, bd_prod_fw, comp_ring_cons;
537 u16 hw_comp_cons, sw_comp_cons, sw_comp_prod;
540 #ifdef BNX2X_STOP_ON_ERROR
541 if (unlikely(bp->panic))
545 /* CQ "next element" is of the size of the regular element,
546 that's why it's ok here */
547 hw_comp_cons = le16_to_cpu(*fp->rx_cons_sb);
548 if ((hw_comp_cons & MAX_RCQ_DESC_CNT) == MAX_RCQ_DESC_CNT)
551 bd_cons = fp->rx_bd_cons;
552 bd_prod = fp->rx_bd_prod;
553 bd_prod_fw = bd_prod;
554 sw_comp_cons = fp->rx_comp_cons;
555 sw_comp_prod = fp->rx_comp_prod;
557 /* Memory barrier necessary as speculative reads of the rx
558 * buffer can be ahead of the index in the status block
562 DP(NETIF_MSG_RX_STATUS,
563 "queue[%d]: hw_comp_cons %u sw_comp_cons %u\n",
564 fp->index, hw_comp_cons, sw_comp_cons);
566 while (sw_comp_cons != hw_comp_cons) {
567 struct sw_rx_bd *rx_buf = NULL;
569 union eth_rx_cqe *cqe;
570 struct eth_fast_path_rx_cqe *cqe_fp;
572 enum eth_rx_cqe_type cqe_fp_type;
575 #ifdef BNX2X_STOP_ON_ERROR
576 if (unlikely(bp->panic))
580 comp_ring_cons = RCQ_BD(sw_comp_cons);
581 bd_prod = RX_BD(bd_prod);
582 bd_cons = RX_BD(bd_cons);
584 /* Prefetch the page containing the BD descriptor
585 at producer's index. It will be needed when new skb is
587 prefetch((void *)(PAGE_ALIGN((unsigned long)
588 (&fp->rx_desc_ring[bd_prod])) -
591 cqe = &fp->rx_comp_ring[comp_ring_cons];
592 cqe_fp = &cqe->fast_path_cqe;
593 cqe_fp_flags = cqe_fp->type_error_flags;
594 cqe_fp_type = cqe_fp_flags & ETH_FAST_PATH_RX_CQE_TYPE;
596 DP(NETIF_MSG_RX_STATUS, "CQE type %x err %x status %x"
597 " queue %x vlan %x len %u\n", CQE_TYPE(cqe_fp_flags),
598 cqe_fp_flags, cqe_fp->status_flags,
599 le32_to_cpu(cqe_fp->rss_hash_result),
600 le16_to_cpu(cqe_fp->vlan_tag), le16_to_cpu(cqe_fp->pkt_len));
602 /* is this a slowpath msg? */
603 if (unlikely(CQE_TYPE_SLOW(cqe_fp_type))) {
604 bnx2x_sp_event(fp, cqe);
607 /* this is an rx packet */
609 rx_buf = &fp->rx_buf_ring[bd_cons];
613 if (!CQE_TYPE_FAST(cqe_fp_type)) {
614 #ifdef BNX2X_STOP_ON_ERROR
616 if (fp->disable_tpa &&
617 (CQE_TYPE_START(cqe_fp_type) ||
618 CQE_TYPE_STOP(cqe_fp_type)))
619 BNX2X_ERR("START/STOP packet while "
620 "disable_tpa type %x\n",
621 CQE_TYPE(cqe_fp_type));
624 if (CQE_TYPE_START(cqe_fp_type)) {
625 u16 queue = cqe_fp->queue_index;
626 DP(NETIF_MSG_RX_STATUS,
627 "calling tpa_start on queue %d\n",
630 bnx2x_tpa_start(fp, queue, skb,
634 /* Set Toeplitz hash for LRO skb */
635 bnx2x_set_skb_rxhash(bp, cqe, skb);
641 cqe->end_agg_cqe.queue_index;
642 DP(NETIF_MSG_RX_STATUS,
643 "calling tpa_stop on queue %d\n",
646 bnx2x_tpa_stop(bp, fp, queue,
649 #ifdef BNX2X_STOP_ON_ERROR
654 bnx2x_update_sge_prod(fp, cqe_fp);
659 len = le16_to_cpu(cqe_fp->pkt_len);
660 pad = cqe_fp->placement_offset;
661 dma_sync_single_for_device(&bp->pdev->dev,
662 dma_unmap_addr(rx_buf, mapping),
663 pad + RX_COPY_THRESH,
665 prefetch(((char *)(skb)) + L1_CACHE_BYTES);
667 /* is this an error packet? */
668 if (unlikely(cqe_fp_flags & ETH_RX_ERROR_FALGS)) {
670 "ERROR flags %x rx packet %u\n",
671 cqe_fp_flags, sw_comp_cons);
672 fp->eth_q_stats.rx_err_discard_pkt++;
676 /* Since we don't have a jumbo ring
677 * copy small packets if mtu > 1500
679 if ((bp->dev->mtu > ETH_MAX_PACKET_SIZE) &&
680 (len <= RX_COPY_THRESH)) {
681 struct sk_buff *new_skb;
683 new_skb = netdev_alloc_skb(bp->dev, len + pad);
684 if (new_skb == NULL) {
686 "ERROR packet dropped "
687 "because of alloc failure\n");
688 fp->eth_q_stats.rx_skb_alloc_failed++;
693 skb_copy_from_linear_data_offset(skb, pad,
694 new_skb->data + pad, len);
695 skb_reserve(new_skb, pad);
696 skb_put(new_skb, len);
698 bnx2x_reuse_rx_skb(fp, bd_cons, bd_prod);
703 if (likely(bnx2x_alloc_rx_skb(bp, fp, bd_prod) == 0)) {
704 dma_unmap_single(&bp->pdev->dev,
705 dma_unmap_addr(rx_buf, mapping),
708 skb_reserve(skb, pad);
713 "ERROR packet dropped because "
714 "of alloc failure\n");
715 fp->eth_q_stats.rx_skb_alloc_failed++;
717 bnx2x_reuse_rx_skb(fp, bd_cons, bd_prod);
721 skb->protocol = eth_type_trans(skb, bp->dev);
723 /* Set Toeplitz hash for a none-LRO skb */
724 bnx2x_set_skb_rxhash(bp, cqe, skb);
726 skb_checksum_none_assert(skb);
728 if (bp->dev->features & NETIF_F_RXCSUM) {
730 if (likely(BNX2X_RX_CSUM_OK(cqe)))
731 skb->ip_summed = CHECKSUM_UNNECESSARY;
733 fp->eth_q_stats.hw_csum_err++;
737 skb_record_rx_queue(skb, fp->index);
739 if (le16_to_cpu(cqe_fp->pars_flags.flags) &
741 __vlan_hwaccel_put_tag(skb,
742 le16_to_cpu(cqe_fp->vlan_tag));
743 napi_gro_receive(&fp->napi, skb);
749 bd_cons = NEXT_RX_IDX(bd_cons);
750 bd_prod = NEXT_RX_IDX(bd_prod);
751 bd_prod_fw = NEXT_RX_IDX(bd_prod_fw);
754 sw_comp_prod = NEXT_RCQ_IDX(sw_comp_prod);
755 sw_comp_cons = NEXT_RCQ_IDX(sw_comp_cons);
757 if (rx_pkt == budget)
761 fp->rx_bd_cons = bd_cons;
762 fp->rx_bd_prod = bd_prod_fw;
763 fp->rx_comp_cons = sw_comp_cons;
764 fp->rx_comp_prod = sw_comp_prod;
766 /* Update producers */
767 bnx2x_update_rx_prod(bp, fp, bd_prod_fw, sw_comp_prod,
770 fp->rx_pkt += rx_pkt;
776 static irqreturn_t bnx2x_msix_fp_int(int irq, void *fp_cookie)
778 struct bnx2x_fastpath *fp = fp_cookie;
779 struct bnx2x *bp = fp->bp;
781 DP(BNX2X_MSG_FP, "got an MSI-X interrupt on IDX:SB "
782 "[fp %d fw_sd %d igusb %d]\n",
783 fp->index, fp->fw_sb_id, fp->igu_sb_id);
784 bnx2x_ack_sb(bp, fp->igu_sb_id, USTORM_ID, 0, IGU_INT_DISABLE, 0);
786 #ifdef BNX2X_STOP_ON_ERROR
787 if (unlikely(bp->panic))
791 /* Handle Rx and Tx according to MSI-X vector */
792 prefetch(fp->rx_cons_sb);
793 prefetch(fp->tx_cons_sb);
794 prefetch(&fp->sb_running_index[SM_RX_ID]);
795 napi_schedule(&bnx2x_fp(bp, fp->index, napi));
800 /* HW Lock for shared dual port PHYs */
801 void bnx2x_acquire_phy_lock(struct bnx2x *bp)
803 mutex_lock(&bp->port.phy_mutex);
805 if (bp->port.need_hw_lock)
806 bnx2x_acquire_hw_lock(bp, HW_LOCK_RESOURCE_MDIO);
809 void bnx2x_release_phy_lock(struct bnx2x *bp)
811 if (bp->port.need_hw_lock)
812 bnx2x_release_hw_lock(bp, HW_LOCK_RESOURCE_MDIO);
814 mutex_unlock(&bp->port.phy_mutex);
817 /* calculates MF speed according to current linespeed and MF configuration */
818 u16 bnx2x_get_mf_speed(struct bnx2x *bp)
820 u16 line_speed = bp->link_vars.line_speed;
822 u16 maxCfg = bnx2x_extract_max_cfg(bp,
823 bp->mf_config[BP_VN(bp)]);
825 /* Calculate the current MAX line speed limit for the MF
829 line_speed = (line_speed * maxCfg) / 100;
831 u16 vn_max_rate = maxCfg * 100;
833 if (vn_max_rate < line_speed)
834 line_speed = vn_max_rate;
842 * bnx2x_fill_report_data - fill link report data to report
845 * @data: link state to update
847 * It uses a none-atomic bit operations because is called under the mutex.
849 static inline void bnx2x_fill_report_data(struct bnx2x *bp,
850 struct bnx2x_link_report_data *data)
852 u16 line_speed = bnx2x_get_mf_speed(bp);
854 memset(data, 0, sizeof(*data));
856 /* Fill the report data: efective line speed */
857 data->line_speed = line_speed;
860 if (!bp->link_vars.link_up || (bp->flags & MF_FUNC_DIS))
861 __set_bit(BNX2X_LINK_REPORT_LINK_DOWN,
862 &data->link_report_flags);
865 if (bp->link_vars.duplex == DUPLEX_FULL)
866 __set_bit(BNX2X_LINK_REPORT_FD, &data->link_report_flags);
868 /* Rx Flow Control is ON */
869 if (bp->link_vars.flow_ctrl & BNX2X_FLOW_CTRL_RX)
870 __set_bit(BNX2X_LINK_REPORT_RX_FC_ON, &data->link_report_flags);
872 /* Tx Flow Control is ON */
873 if (bp->link_vars.flow_ctrl & BNX2X_FLOW_CTRL_TX)
874 __set_bit(BNX2X_LINK_REPORT_TX_FC_ON, &data->link_report_flags);
878 * bnx2x_link_report - report link status to OS.
882 * Calls the __bnx2x_link_report() under the same locking scheme
883 * as a link/PHY state managing code to ensure a consistent link
887 void bnx2x_link_report(struct bnx2x *bp)
889 bnx2x_acquire_phy_lock(bp);
890 __bnx2x_link_report(bp);
891 bnx2x_release_phy_lock(bp);
895 * __bnx2x_link_report - report link status to OS.
899 * None atomic inmlementation.
900 * Should be called under the phy_lock.
902 void __bnx2x_link_report(struct bnx2x *bp)
904 struct bnx2x_link_report_data cur_data;
908 bnx2x_read_mf_cfg(bp);
910 /* Read the current link report info */
911 bnx2x_fill_report_data(bp, &cur_data);
913 /* Don't report link down or exactly the same link status twice */
914 if (!memcmp(&cur_data, &bp->last_reported_link, sizeof(cur_data)) ||
915 (test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
916 &bp->last_reported_link.link_report_flags) &&
917 test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
918 &cur_data.link_report_flags)))
923 /* We are going to report a new link parameters now -
924 * remember the current data for the next time.
926 memcpy(&bp->last_reported_link, &cur_data, sizeof(cur_data));
928 if (test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
929 &cur_data.link_report_flags)) {
930 netif_carrier_off(bp->dev);
931 netdev_err(bp->dev, "NIC Link is Down\n");
934 netif_carrier_on(bp->dev);
935 netdev_info(bp->dev, "NIC Link is Up, ");
936 pr_cont("%d Mbps ", cur_data.line_speed);
938 if (test_and_clear_bit(BNX2X_LINK_REPORT_FD,
939 &cur_data.link_report_flags))
940 pr_cont("full duplex");
942 pr_cont("half duplex");
944 /* Handle the FC at the end so that only these flags would be
945 * possibly set. This way we may easily check if there is no FC
948 if (cur_data.link_report_flags) {
949 if (test_bit(BNX2X_LINK_REPORT_RX_FC_ON,
950 &cur_data.link_report_flags)) {
951 pr_cont(", receive ");
952 if (test_bit(BNX2X_LINK_REPORT_TX_FC_ON,
953 &cur_data.link_report_flags))
954 pr_cont("& transmit ");
956 pr_cont(", transmit ");
958 pr_cont("flow control ON");
964 void bnx2x_init_rx_rings(struct bnx2x *bp)
966 int func = BP_FUNC(bp);
967 int max_agg_queues = CHIP_IS_E1(bp) ? ETH_MAX_AGGREGATION_QUEUES_E1 :
968 ETH_MAX_AGGREGATION_QUEUES_E1H_E2;
972 /* Allocate TPA resources */
973 for_each_rx_queue(bp, j) {
974 struct bnx2x_fastpath *fp = &bp->fp[j];
977 "mtu %d rx_buf_size %d\n", bp->dev->mtu, fp->rx_buf_size);
979 if (!fp->disable_tpa) {
980 /* Fill the per-aggregtion pool */
981 for (i = 0; i < max_agg_queues; i++) {
982 struct bnx2x_agg_info *tpa_info =
984 struct sw_rx_bd *first_buf =
985 &tpa_info->first_buf;
987 first_buf->skb = netdev_alloc_skb(bp->dev,
989 if (!first_buf->skb) {
990 BNX2X_ERR("Failed to allocate TPA "
991 "skb pool for queue[%d] - "
992 "disabling TPA on this "
994 bnx2x_free_tpa_pool(bp, fp, i);
998 dma_unmap_addr_set(first_buf, mapping, 0);
999 tpa_info->tpa_state = BNX2X_TPA_STOP;
1002 /* "next page" elements initialization */
1003 bnx2x_set_next_page_sgl(fp);
1005 /* set SGEs bit mask */
1006 bnx2x_init_sge_ring_bit_mask(fp);
1008 /* Allocate SGEs and initialize the ring elements */
1009 for (i = 0, ring_prod = 0;
1010 i < MAX_RX_SGE_CNT*NUM_RX_SGE_PAGES; i++) {
1012 if (bnx2x_alloc_rx_sge(bp, fp, ring_prod) < 0) {
1013 BNX2X_ERR("was only able to allocate "
1015 BNX2X_ERR("disabling TPA for "
1017 /* Cleanup already allocated elements */
1018 bnx2x_free_rx_sge_range(bp, fp,
1020 bnx2x_free_tpa_pool(bp, fp,
1022 fp->disable_tpa = 1;
1026 ring_prod = NEXT_SGE_IDX(ring_prod);
1029 fp->rx_sge_prod = ring_prod;
1033 for_each_rx_queue(bp, j) {
1034 struct bnx2x_fastpath *fp = &bp->fp[j];
1038 /* Activate BD ring */
1040 * this will generate an interrupt (to the TSTORM)
1041 * must only be done after chip is initialized
1043 bnx2x_update_rx_prod(bp, fp, fp->rx_bd_prod, fp->rx_comp_prod,
1049 if (CHIP_IS_E1(bp)) {
1050 REG_WR(bp, BAR_USTRORM_INTMEM +
1051 USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func),
1052 U64_LO(fp->rx_comp_mapping));
1053 REG_WR(bp, BAR_USTRORM_INTMEM +
1054 USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func) + 4,
1055 U64_HI(fp->rx_comp_mapping));
1060 static void bnx2x_free_tx_skbs(struct bnx2x *bp)
1064 for_each_tx_queue(bp, i) {
1065 struct bnx2x_fastpath *fp = &bp->fp[i];
1067 u16 bd_cons = fp->tx_bd_cons;
1068 u16 sw_prod = fp->tx_pkt_prod;
1069 u16 sw_cons = fp->tx_pkt_cons;
1071 while (sw_cons != sw_prod) {
1072 bd_cons = bnx2x_free_tx_pkt(bp, fp, TX_BD(sw_cons));
1078 static void bnx2x_free_rx_bds(struct bnx2x_fastpath *fp)
1080 struct bnx2x *bp = fp->bp;
1083 /* ring wasn't allocated */
1084 if (fp->rx_buf_ring == NULL)
1087 for (i = 0; i < NUM_RX_BD; i++) {
1088 struct sw_rx_bd *rx_buf = &fp->rx_buf_ring[i];
1089 struct sk_buff *skb = rx_buf->skb;
1093 dma_unmap_single(&bp->pdev->dev,
1094 dma_unmap_addr(rx_buf, mapping),
1095 fp->rx_buf_size, DMA_FROM_DEVICE);
1102 static void bnx2x_free_rx_skbs(struct bnx2x *bp)
1106 for_each_rx_queue(bp, j) {
1107 struct bnx2x_fastpath *fp = &bp->fp[j];
1109 bnx2x_free_rx_bds(fp);
1111 if (!fp->disable_tpa)
1112 bnx2x_free_tpa_pool(bp, fp, CHIP_IS_E1(bp) ?
1113 ETH_MAX_AGGREGATION_QUEUES_E1 :
1114 ETH_MAX_AGGREGATION_QUEUES_E1H_E2);
1118 void bnx2x_free_skbs(struct bnx2x *bp)
1120 bnx2x_free_tx_skbs(bp);
1121 bnx2x_free_rx_skbs(bp);
1124 void bnx2x_update_max_mf_config(struct bnx2x *bp, u32 value)
1126 /* load old values */
1127 u32 mf_cfg = bp->mf_config[BP_VN(bp)];
1129 if (value != bnx2x_extract_max_cfg(bp, mf_cfg)) {
1130 /* leave all but MAX value */
1131 mf_cfg &= ~FUNC_MF_CFG_MAX_BW_MASK;
1133 /* set new MAX value */
1134 mf_cfg |= (value << FUNC_MF_CFG_MAX_BW_SHIFT)
1135 & FUNC_MF_CFG_MAX_BW_MASK;
1137 bnx2x_fw_command(bp, DRV_MSG_CODE_SET_MF_BW, mf_cfg);
1142 * bnx2x_free_msix_irqs - free previously requested MSI-X IRQ vectors
1144 * @bp: driver handle
1145 * @nvecs: number of vectors to be released
1147 static void bnx2x_free_msix_irqs(struct bnx2x *bp, int nvecs)
1151 if (nvecs == offset)
1153 free_irq(bp->msix_table[offset].vector, bp->dev);
1154 DP(NETIF_MSG_IFDOWN, "released sp irq (%d)\n",
1155 bp->msix_table[offset].vector);
1158 if (nvecs == offset)
1163 for_each_eth_queue(bp, i) {
1164 if (nvecs == offset)
1166 DP(NETIF_MSG_IFDOWN, "about to release fp #%d->%d "
1167 "irq\n", i, bp->msix_table[offset].vector);
1169 free_irq(bp->msix_table[offset++].vector, &bp->fp[i]);
1173 void bnx2x_free_irq(struct bnx2x *bp)
1175 if (bp->flags & USING_MSIX_FLAG)
1176 bnx2x_free_msix_irqs(bp, BNX2X_NUM_ETH_QUEUES(bp) +
1177 CNIC_CONTEXT_USE + 1);
1178 else if (bp->flags & USING_MSI_FLAG)
1179 free_irq(bp->pdev->irq, bp->dev);
1181 free_irq(bp->pdev->irq, bp->dev);
1184 int bnx2x_enable_msix(struct bnx2x *bp)
1186 int msix_vec = 0, i, rc, req_cnt;
1188 bp->msix_table[msix_vec].entry = msix_vec;
1189 DP(NETIF_MSG_IFUP, "msix_table[0].entry = %d (slowpath)\n",
1190 bp->msix_table[0].entry);
1194 bp->msix_table[msix_vec].entry = msix_vec;
1195 DP(NETIF_MSG_IFUP, "msix_table[%d].entry = %d (CNIC)\n",
1196 bp->msix_table[msix_vec].entry, bp->msix_table[msix_vec].entry);
1199 for_each_eth_queue(bp, i) {
1200 bp->msix_table[msix_vec].entry = msix_vec;
1201 DP(NETIF_MSG_IFUP, "msix_table[%d].entry = %d "
1202 "(fastpath #%u)\n", msix_vec, msix_vec, i);
1206 req_cnt = BNX2X_NUM_ETH_QUEUES(bp) + CNIC_CONTEXT_USE + 1;
1208 rc = pci_enable_msix(bp->pdev, &bp->msix_table[0], req_cnt);
1211 * reconfigure number of tx/rx queues according to available
1214 if (rc >= BNX2X_MIN_MSIX_VEC_CNT) {
1215 /* how less vectors we will have? */
1216 int diff = req_cnt - rc;
1219 "Trying to use less MSI-X vectors: %d\n", rc);
1221 rc = pci_enable_msix(bp->pdev, &bp->msix_table[0], rc);
1225 "MSI-X is not attainable rc %d\n", rc);
1229 * decrease number of queues by number of unallocated entries
1231 bp->num_queues -= diff;
1233 DP(NETIF_MSG_IFUP, "New queue configuration set: %d\n",
1236 /* fall to INTx if not enough memory */
1238 bp->flags |= DISABLE_MSI_FLAG;
1239 DP(NETIF_MSG_IFUP, "MSI-X is not attainable rc %d\n", rc);
1243 bp->flags |= USING_MSIX_FLAG;
1248 static int bnx2x_req_msix_irqs(struct bnx2x *bp)
1250 int i, rc, offset = 0;
1252 rc = request_irq(bp->msix_table[offset++].vector,
1253 bnx2x_msix_sp_int, 0,
1254 bp->dev->name, bp->dev);
1256 BNX2X_ERR("request sp irq failed\n");
1263 for_each_eth_queue(bp, i) {
1264 struct bnx2x_fastpath *fp = &bp->fp[i];
1265 snprintf(fp->name, sizeof(fp->name), "%s-fp-%d",
1268 rc = request_irq(bp->msix_table[offset].vector,
1269 bnx2x_msix_fp_int, 0, fp->name, fp);
1271 BNX2X_ERR("request fp #%d irq (%d) failed rc %d\n", i,
1272 bp->msix_table[offset].vector, rc);
1273 bnx2x_free_msix_irqs(bp, offset);
1280 i = BNX2X_NUM_ETH_QUEUES(bp);
1281 offset = 1 + CNIC_CONTEXT_USE;
1282 netdev_info(bp->dev, "using MSI-X IRQs: sp %d fp[%d] %d"
1284 bp->msix_table[0].vector,
1285 0, bp->msix_table[offset].vector,
1286 i - 1, bp->msix_table[offset + i - 1].vector);
1291 int bnx2x_enable_msi(struct bnx2x *bp)
1295 rc = pci_enable_msi(bp->pdev);
1297 DP(NETIF_MSG_IFUP, "MSI is not attainable\n");
1300 bp->flags |= USING_MSI_FLAG;
1305 static int bnx2x_req_irq(struct bnx2x *bp)
1307 unsigned long flags;
1310 if (bp->flags & USING_MSI_FLAG)
1313 flags = IRQF_SHARED;
1315 rc = request_irq(bp->pdev->irq, bnx2x_interrupt, flags,
1316 bp->dev->name, bp->dev);
1320 static inline int bnx2x_setup_irqs(struct bnx2x *bp)
1323 if (bp->flags & USING_MSIX_FLAG) {
1324 rc = bnx2x_req_msix_irqs(bp);
1329 rc = bnx2x_req_irq(bp);
1331 BNX2X_ERR("IRQ request failed rc %d, aborting\n", rc);
1334 if (bp->flags & USING_MSI_FLAG) {
1335 bp->dev->irq = bp->pdev->irq;
1336 netdev_info(bp->dev, "using MSI IRQ %d\n",
1344 static inline void bnx2x_napi_enable(struct bnx2x *bp)
1348 for_each_rx_queue(bp, i)
1349 napi_enable(&bnx2x_fp(bp, i, napi));
1352 static inline void bnx2x_napi_disable(struct bnx2x *bp)
1356 for_each_rx_queue(bp, i)
1357 napi_disable(&bnx2x_fp(bp, i, napi));
1360 void bnx2x_netif_start(struct bnx2x *bp)
1362 if (netif_running(bp->dev)) {
1363 bnx2x_napi_enable(bp);
1364 bnx2x_int_enable(bp);
1365 if (bp->state == BNX2X_STATE_OPEN)
1366 netif_tx_wake_all_queues(bp->dev);
1370 void bnx2x_netif_stop(struct bnx2x *bp, int disable_hw)
1372 bnx2x_int_disable_sync(bp, disable_hw);
1373 bnx2x_napi_disable(bp);
1376 u16 bnx2x_select_queue(struct net_device *dev, struct sk_buff *skb)
1379 struct bnx2x *bp = netdev_priv(dev);
1381 return skb_tx_hash(dev, skb);
1383 struct ethhdr *hdr = (struct ethhdr *)skb->data;
1384 u16 ether_type = ntohs(hdr->h_proto);
1386 /* Skip VLAN tag if present */
1387 if (ether_type == ETH_P_8021Q) {
1388 struct vlan_ethhdr *vhdr =
1389 (struct vlan_ethhdr *)skb->data;
1391 ether_type = ntohs(vhdr->h_vlan_encapsulated_proto);
1394 /* If ethertype is FCoE or FIP - use FCoE ring */
1395 if ((ether_type == ETH_P_FCOE) || (ether_type == ETH_P_FIP))
1396 return bnx2x_fcoe(bp, index);
1399 /* Select a none-FCoE queue: if FCoE is enabled, exclude FCoE L2 ring
1401 return __skb_tx_hash(dev, skb,
1402 dev->real_num_tx_queues - FCOE_CONTEXT_USE);
1405 void bnx2x_set_num_queues(struct bnx2x *bp)
1407 switch (bp->multi_mode) {
1408 case ETH_RSS_MODE_DISABLED:
1411 case ETH_RSS_MODE_REGULAR:
1412 bp->num_queues = bnx2x_calc_num_queues(bp);
1420 /* Add special queues */
1421 bp->num_queues += NONE_ETH_CONTEXT_USE;
1424 static inline int bnx2x_set_real_num_queues(struct bnx2x *bp)
1426 int rc, num = bp->num_queues;
1430 num -= FCOE_CONTEXT_USE;
1433 netif_set_real_num_tx_queues(bp->dev, num);
1434 rc = netif_set_real_num_rx_queues(bp->dev, num);
1438 static inline void bnx2x_set_rx_buf_size(struct bnx2x *bp)
1442 for_each_queue(bp, i) {
1443 struct bnx2x_fastpath *fp = &bp->fp[i];
1445 /* Always use a mini-jumbo MTU for the FCoE L2 ring */
1448 * Although there are no IP frames expected to arrive to
1449 * this ring we still want to add an
1450 * IP_HEADER_ALIGNMENT_PADDING to prevent a buffer
1454 BNX2X_FCOE_MINI_JUMBO_MTU + ETH_OVREHEAD +
1455 BNX2X_FW_RX_ALIGN + IP_HEADER_ALIGNMENT_PADDING;
1458 bp->dev->mtu + ETH_OVREHEAD +
1459 BNX2X_FW_RX_ALIGN + IP_HEADER_ALIGNMENT_PADDING;
1463 static inline int bnx2x_init_rss_pf(struct bnx2x *bp)
1466 u8 ind_table[T_ETH_INDIRECTION_TABLE_SIZE] = {0};
1467 u8 num_eth_queues = BNX2X_NUM_ETH_QUEUES(bp);
1470 * Prepare the inital contents fo the indirection table if RSS is
1473 if (bp->multi_mode != ETH_RSS_MODE_DISABLED) {
1474 for (i = 0; i < sizeof(ind_table); i++)
1476 bp->fp->cl_id + (i % num_eth_queues);
1480 * For 57710 and 57711 SEARCHER configuration (rss_keys) is
1481 * per-port, so if explicit configuration is needed , do it only
1484 * For 57712 and newer on the other hand it's a per-function
1487 return bnx2x_config_rss_pf(bp, ind_table,
1488 bp->port.pmf || !CHIP_IS_E1x(bp));
1491 int bnx2x_config_rss_pf(struct bnx2x *bp, u8 *ind_table, bool config_hash)
1493 struct bnx2x_config_rss_params params = {0};
1496 /* Although RSS is meaningless when there is a single HW queue we
1497 * still need it enabled in order to have HW Rx hash generated.
1499 * if (!is_eth_multi(bp))
1500 * bp->multi_mode = ETH_RSS_MODE_DISABLED;
1503 params.rss_obj = &bp->rss_conf_obj;
1505 __set_bit(RAMROD_COMP_WAIT, ¶ms.ramrod_flags);
1508 switch (bp->multi_mode) {
1509 case ETH_RSS_MODE_DISABLED:
1510 __set_bit(BNX2X_RSS_MODE_DISABLED, ¶ms.rss_flags);
1512 case ETH_RSS_MODE_REGULAR:
1513 __set_bit(BNX2X_RSS_MODE_REGULAR, ¶ms.rss_flags);
1515 case ETH_RSS_MODE_VLAN_PRI:
1516 __set_bit(BNX2X_RSS_MODE_VLAN_PRI, ¶ms.rss_flags);
1518 case ETH_RSS_MODE_E1HOV_PRI:
1519 __set_bit(BNX2X_RSS_MODE_E1HOV_PRI, ¶ms.rss_flags);
1521 case ETH_RSS_MODE_IP_DSCP:
1522 __set_bit(BNX2X_RSS_MODE_IP_DSCP, ¶ms.rss_flags);
1525 BNX2X_ERR("Unknown multi_mode: %d\n", bp->multi_mode);
1529 /* If RSS is enabled */
1530 if (bp->multi_mode != ETH_RSS_MODE_DISABLED) {
1531 /* RSS configuration */
1532 __set_bit(BNX2X_RSS_IPV4, ¶ms.rss_flags);
1533 __set_bit(BNX2X_RSS_IPV4_TCP, ¶ms.rss_flags);
1534 __set_bit(BNX2X_RSS_IPV6, ¶ms.rss_flags);
1535 __set_bit(BNX2X_RSS_IPV6_TCP, ¶ms.rss_flags);
1538 params.rss_result_mask = MULTI_MASK;
1540 memcpy(params.ind_table, ind_table, sizeof(params.ind_table));
1544 for (i = 0; i < sizeof(params.rss_key) / 4; i++)
1545 params.rss_key[i] = random32();
1547 __set_bit(BNX2X_RSS_SET_SRCH, ¶ms.rss_flags);
1551 return bnx2x_config_rss(bp, ¶ms);
1554 static inline int bnx2x_init_hw(struct bnx2x *bp, u32 load_code)
1556 struct bnx2x_func_state_params func_params = {0};
1558 /* Prepare parameters for function state transitions */
1559 __set_bit(RAMROD_COMP_WAIT, &func_params.ramrod_flags);
1561 func_params.f_obj = &bp->func_obj;
1562 func_params.cmd = BNX2X_F_CMD_HW_INIT;
1564 func_params.params.hw_init.load_phase = load_code;
1566 return bnx2x_func_state_change(bp, &func_params);
1570 * Cleans the object that have internal lists without sending
1571 * ramrods. Should be run when interrutps are disabled.
1573 static void bnx2x_squeeze_objects(struct bnx2x *bp)
1576 unsigned long ramrod_flags = 0, vlan_mac_flags = 0;
1577 struct bnx2x_mcast_ramrod_params rparam = {0};
1578 struct bnx2x_vlan_mac_obj *mac_obj = &bp->fp->mac_obj;
1580 /***************** Cleanup MACs' object first *************************/
1582 /* Wait for completion of requested */
1583 __set_bit(RAMROD_COMP_WAIT, &ramrod_flags);
1584 /* Perform a dry cleanup */
1585 __set_bit(RAMROD_DRV_CLR_ONLY, &ramrod_flags);
1587 /* Clean ETH primary MAC */
1588 __set_bit(BNX2X_ETH_MAC, &vlan_mac_flags);
1589 rc = mac_obj->delete_all(bp, &bp->fp->mac_obj, &vlan_mac_flags,
1592 BNX2X_ERR("Failed to clean ETH MACs: %d\n", rc);
1594 /* Cleanup UC list */
1596 __set_bit(BNX2X_UC_LIST_MAC, &vlan_mac_flags);
1597 rc = mac_obj->delete_all(bp, mac_obj, &vlan_mac_flags,
1600 BNX2X_ERR("Failed to clean UC list MACs: %d\n", rc);
1602 /***************** Now clean mcast object *****************************/
1603 rparam.mcast_obj = &bp->mcast_obj;
1604 __set_bit(RAMROD_DRV_CLR_ONLY, &rparam.ramrod_flags);
1606 /* Add a DEL command... */
1607 rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_DEL);
1609 BNX2X_ERR("Failed to add a new DEL command to a multi-cast "
1610 "object: %d\n", rc);
1612 /* ...and wait until all pending commands are cleared */
1613 rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_CONT);
1616 BNX2X_ERR("Failed to clean multi-cast object: %d\n",
1621 rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_CONT);
1625 #ifndef BNX2X_STOP_ON_ERROR
1626 #define LOAD_ERROR_EXIT(bp, label) \
1628 (bp)->state = BNX2X_STATE_ERROR; \
1632 #define LOAD_ERROR_EXIT(bp, label) \
1634 (bp)->state = BNX2X_STATE_ERROR; \
1640 /* must be called with rtnl_lock */
1641 int bnx2x_nic_load(struct bnx2x *bp, int load_mode)
1643 int port = BP_PORT(bp);
1647 #ifdef BNX2X_STOP_ON_ERROR
1648 if (unlikely(bp->panic))
1652 bp->state = BNX2X_STATE_OPENING_WAIT4_LOAD;
1654 /* Set the initial link reported state to link down */
1655 bnx2x_acquire_phy_lock(bp);
1656 memset(&bp->last_reported_link, 0, sizeof(bp->last_reported_link));
1657 __set_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1658 &bp->last_reported_link.link_report_flags);
1659 bnx2x_release_phy_lock(bp);
1661 /* must be called before memory allocation and HW init */
1662 bnx2x_ilt_set_info(bp);
1664 /* zero fastpath structures preserving invariants like napi which are
1665 * allocated only once
1667 for_each_queue(bp, i)
1670 /* Set the receive queues buffer size */
1671 bnx2x_set_rx_buf_size(bp);
1674 * set the tpa flag for each queue. The tpa flag determines the queue
1675 * minimal size so it must be set prior to queue memory allocation
1677 for_each_queue(bp, i)
1678 bnx2x_fp(bp, i, disable_tpa) =
1679 ((bp->flags & TPA_ENABLE_FLAG) == 0);
1682 /* We don't want TPA on FCoE L2 ring */
1683 bnx2x_fcoe(bp, disable_tpa) = 1;
1686 if (bnx2x_alloc_mem(bp))
1689 /* As long as bnx2x_alloc_mem() may possibly update
1690 * bp->num_queues, bnx2x_set_real_num_queues() should always
1693 rc = bnx2x_set_real_num_queues(bp);
1695 BNX2X_ERR("Unable to set real_num_queues\n");
1696 LOAD_ERROR_EXIT(bp, load_error0);
1699 bnx2x_napi_enable(bp);
1701 /* Send LOAD_REQUEST command to MCP
1702 * Returns the type of LOAD command:
1703 * if it is the first port to be initialized
1704 * common blocks should be initialized, otherwise - not
1706 if (!BP_NOMCP(bp)) {
1707 load_code = bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_REQ, 0);
1709 BNX2X_ERR("MCP response failure, aborting\n");
1711 LOAD_ERROR_EXIT(bp, load_error1);
1713 if (load_code == FW_MSG_CODE_DRV_LOAD_REFUSED) {
1714 rc = -EBUSY; /* other port in diagnostic mode */
1715 LOAD_ERROR_EXIT(bp, load_error1);
1719 int path = BP_PATH(bp);
1721 DP(NETIF_MSG_IFUP, "NO MCP - load counts[%d] %d, %d, %d\n",
1722 path, load_count[path][0], load_count[path][1],
1723 load_count[path][2]);
1724 load_count[path][0]++;
1725 load_count[path][1 + port]++;
1726 DP(NETIF_MSG_IFUP, "NO MCP - new load counts[%d] %d, %d, %d\n",
1727 path, load_count[path][0], load_count[path][1],
1728 load_count[path][2]);
1729 if (load_count[path][0] == 1)
1730 load_code = FW_MSG_CODE_DRV_LOAD_COMMON;
1731 else if (load_count[path][1 + port] == 1)
1732 load_code = FW_MSG_CODE_DRV_LOAD_PORT;
1734 load_code = FW_MSG_CODE_DRV_LOAD_FUNCTION;
1737 if ((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
1738 (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP) ||
1739 (load_code == FW_MSG_CODE_DRV_LOAD_PORT)) {
1742 * We need the barrier to ensure the ordering between the
1743 * writing to bp->port.pmf here and reading it from the
1744 * bnx2x_periodic_task().
1747 queue_delayed_work(bnx2x_wq, &bp->period_task, 0);
1750 DP(NETIF_MSG_LINK, "pmf %d\n", bp->port.pmf);
1752 /* Init Function state controlling object */
1753 bnx2x__init_func_obj(bp);
1756 rc = bnx2x_init_hw(bp, load_code);
1758 BNX2X_ERR("HW init failed, aborting\n");
1759 bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
1760 LOAD_ERROR_EXIT(bp, load_error2);
1763 /* Connect to IRQs */
1764 rc = bnx2x_setup_irqs(bp);
1766 bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
1767 LOAD_ERROR_EXIT(bp, load_error2);
1770 /* Setup NIC internals and enable interrupts */
1771 bnx2x_nic_init(bp, load_code);
1773 /* Init per-function objects */
1774 bnx2x_init_bp_objs(bp);
1776 if (((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
1777 (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP)) &&
1778 (bp->common.shmem2_base)) {
1779 if (SHMEM2_HAS(bp, dcc_support))
1780 SHMEM2_WR(bp, dcc_support,
1781 (SHMEM_DCC_SUPPORT_DISABLE_ENABLE_PF_TLV |
1782 SHMEM_DCC_SUPPORT_BANDWIDTH_ALLOCATION_TLV));
1785 bp->state = BNX2X_STATE_OPENING_WAIT4_PORT;
1786 rc = bnx2x_func_start(bp);
1788 BNX2X_ERR("Function start failed!\n");
1789 LOAD_ERROR_EXIT(bp, load_error3);
1792 /* Send LOAD_DONE command to MCP */
1793 if (!BP_NOMCP(bp)) {
1794 load_code = bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
1796 BNX2X_ERR("MCP response failure, aborting\n");
1798 LOAD_ERROR_EXIT(bp, load_error3);
1802 rc = bnx2x_setup_leading(bp);
1804 BNX2X_ERR("Setup leading failed!\n");
1805 LOAD_ERROR_EXIT(bp, load_error3);
1809 /* Enable Timer scan */
1810 REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + port*4, 1);
1813 for_each_nondefault_queue(bp, i) {
1814 rc = bnx2x_setup_queue(bp, &bp->fp[i], 0);
1816 LOAD_ERROR_EXIT(bp, load_error4);
1819 rc = bnx2x_init_rss_pf(bp);
1821 LOAD_ERROR_EXIT(bp, load_error4);
1823 /* Now when Clients are configured we are ready to work */
1824 bp->state = BNX2X_STATE_OPEN;
1826 /* Configure a ucast MAC */
1827 rc = bnx2x_set_eth_mac(bp, true);
1829 LOAD_ERROR_EXIT(bp, load_error4);
1831 if (bp->pending_max) {
1832 bnx2x_update_max_mf_config(bp, bp->pending_max);
1833 bp->pending_max = 0;
1837 bnx2x_initial_phy_init(bp, load_mode);
1839 /* Start fast path */
1841 /* Initialize Rx filter. */
1842 netif_addr_lock_bh(bp->dev);
1843 bnx2x_set_rx_mode(bp->dev);
1844 netif_addr_unlock_bh(bp->dev);
1847 switch (load_mode) {
1849 /* Tx queue should be only reenabled */
1850 netif_tx_wake_all_queues(bp->dev);
1854 netif_tx_start_all_queues(bp->dev);
1855 smp_mb__after_clear_bit();
1859 bp->state = BNX2X_STATE_DIAG;
1867 bnx2x__link_status_update(bp);
1869 /* start the timer */
1870 mod_timer(&bp->timer, jiffies + bp->current_interval);
1873 bnx2x_setup_cnic_irq_info(bp);
1874 if (bp->state == BNX2X_STATE_OPEN)
1875 bnx2x_cnic_notify(bp, CNIC_CTL_START_CMD);
1877 bnx2x_inc_load_cnt(bp);
1879 /* Wait for all pending SP commands to complete */
1880 if (!bnx2x_wait_sp_comp(bp, ~0x0UL)) {
1881 BNX2X_ERR("Timeout waiting for SP elements to complete\n");
1882 bnx2x_nic_unload(bp, UNLOAD_CLOSE);
1886 bnx2x_dcbx_init(bp);
1889 #ifndef BNX2X_STOP_ON_ERROR
1892 /* Disable Timer scan */
1893 REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + port*4, 0);
1896 bnx2x_int_disable_sync(bp, 1);
1898 /* Clean queueable objects */
1899 bnx2x_squeeze_objects(bp);
1901 /* Free SKBs, SGEs, TPA pool and driver internals */
1902 bnx2x_free_skbs(bp);
1903 for_each_rx_queue(bp, i)
1904 bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
1909 if (!BP_NOMCP(bp)) {
1910 bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_REQ_WOL_MCP, 0);
1911 bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_DONE, 0);
1916 bnx2x_napi_disable(bp);
1921 #endif /* ! BNX2X_STOP_ON_ERROR */
1924 /* must be called with rtnl_lock */
1925 int bnx2x_nic_unload(struct bnx2x *bp, int unload_mode)
1928 bool global = false;
1930 if ((bp->state == BNX2X_STATE_CLOSED) ||
1931 (bp->state == BNX2X_STATE_ERROR)) {
1932 /* We can get here if the driver has been unloaded
1933 * during parity error recovery and is either waiting for a
1934 * leader to complete or for other functions to unload and
1935 * then ifdown has been issued. In this case we want to
1936 * unload and let other functions to complete a recovery
1939 bp->recovery_state = BNX2X_RECOVERY_DONE;
1941 bnx2x_release_leader_lock(bp);
1944 DP(NETIF_MSG_HW, "Releasing a leadership...\n");
1950 bnx2x_cnic_notify(bp, CNIC_CTL_STOP_CMD);
1952 bp->state = BNX2X_STATE_CLOSING_WAIT4_HALT;
1955 bp->rx_mode = BNX2X_RX_MODE_NONE;
1958 bnx2x_tx_disable(bp);
1960 del_timer_sync(&bp->timer);
1962 /* Set ALWAYS_ALIVE bit in shmem */
1963 bp->fw_drv_pulse_wr_seq |= DRV_PULSE_ALWAYS_ALIVE;
1965 bnx2x_drv_pulse(bp);
1967 bnx2x_stats_handle(bp, STATS_EVENT_STOP);
1969 /* Cleanup the chip if needed */
1970 if (unload_mode != UNLOAD_RECOVERY)
1971 bnx2x_chip_cleanup(bp, unload_mode);
1973 /* Send the UNLOAD_REQUEST to the MCP */
1974 bnx2x_send_unload_req(bp, unload_mode);
1977 * Prevent transactions to host from the functions on the
1978 * engine that doesn't reset global blocks in case of global
1979 * attention once gloabl blocks are reset and gates are opened
1980 * (the engine which leader will perform the recovery
1983 if (!CHIP_IS_E1x(bp))
1984 bnx2x_pf_disable(bp);
1986 /* Disable HW interrupts, NAPI */
1987 bnx2x_netif_stop(bp, 1);
1992 /* Report UNLOAD_DONE to MCP */
1993 bnx2x_send_unload_done(bp);
1997 * At this stage no more interrupts will arrive so we may safly clean
1998 * the queueable objects here in case they failed to get cleaned so far.
2000 bnx2x_squeeze_objects(bp);
2004 /* Free SKBs, SGEs, TPA pool and driver internals */
2005 bnx2x_free_skbs(bp);
2006 for_each_rx_queue(bp, i)
2007 bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
2011 bp->state = BNX2X_STATE_CLOSED;
2013 /* Check if there are pending parity attentions. If there are - set
2014 * RECOVERY_IN_PROGRESS.
2016 if (bnx2x_chk_parity_attn(bp, &global, false)) {
2017 bnx2x_set_reset_in_progress(bp);
2019 /* Set RESET_IS_GLOBAL if needed */
2021 bnx2x_set_reset_global(bp);
2025 /* The last driver must disable a "close the gate" if there is no
2026 * parity attention or "process kill" pending.
2028 if (!bnx2x_dec_load_cnt(bp) && bnx2x_reset_is_done(bp, BP_PATH(bp)))
2029 bnx2x_disable_close_the_gate(bp);
2034 int bnx2x_set_power_state(struct bnx2x *bp, pci_power_t state)
2038 /* If there is no power capability, silently succeed */
2040 DP(NETIF_MSG_HW, "No power capability. Breaking.\n");
2044 pci_read_config_word(bp->pdev, bp->pm_cap + PCI_PM_CTRL, &pmcsr);
2048 pci_write_config_word(bp->pdev, bp->pm_cap + PCI_PM_CTRL,
2049 ((pmcsr & ~PCI_PM_CTRL_STATE_MASK) |
2050 PCI_PM_CTRL_PME_STATUS));
2052 if (pmcsr & PCI_PM_CTRL_STATE_MASK)
2053 /* delay required during transition out of D3hot */
2058 /* If there are other clients above don't
2059 shut down the power */
2060 if (atomic_read(&bp->pdev->enable_cnt) != 1)
2062 /* Don't shut down the power for emulation and FPGA */
2063 if (CHIP_REV_IS_SLOW(bp))
2066 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
2070 pmcsr |= PCI_PM_CTRL_PME_ENABLE;
2072 pci_write_config_word(bp->pdev, bp->pm_cap + PCI_PM_CTRL,
2075 /* No more memory access after this point until
2076 * device is brought back to D0.
2087 * net_device service functions
2089 int bnx2x_poll(struct napi_struct *napi, int budget)
2092 struct bnx2x_fastpath *fp = container_of(napi, struct bnx2x_fastpath,
2094 struct bnx2x *bp = fp->bp;
2097 #ifdef BNX2X_STOP_ON_ERROR
2098 if (unlikely(bp->panic)) {
2099 napi_complete(napi);
2104 if (bnx2x_has_tx_work(fp))
2107 if (bnx2x_has_rx_work(fp)) {
2108 work_done += bnx2x_rx_int(fp, budget - work_done);
2110 /* must not complete if we consumed full budget */
2111 if (work_done >= budget)
2115 /* Fall out from the NAPI loop if needed */
2116 if (!(bnx2x_has_rx_work(fp) || bnx2x_has_tx_work(fp))) {
2118 /* No need to update SB for FCoE L2 ring as long as
2119 * it's connected to the default SB and the SB
2120 * has been updated when NAPI was scheduled.
2122 if (IS_FCOE_FP(fp)) {
2123 napi_complete(napi);
2128 bnx2x_update_fpsb_idx(fp);
2129 /* bnx2x_has_rx_work() reads the status block,
2130 * thus we need to ensure that status block indices
2131 * have been actually read (bnx2x_update_fpsb_idx)
2132 * prior to this check (bnx2x_has_rx_work) so that
2133 * we won't write the "newer" value of the status block
2134 * to IGU (if there was a DMA right after
2135 * bnx2x_has_rx_work and if there is no rmb, the memory
2136 * reading (bnx2x_update_fpsb_idx) may be postponed
2137 * to right before bnx2x_ack_sb). In this case there
2138 * will never be another interrupt until there is
2139 * another update of the status block, while there
2140 * is still unhandled work.
2144 if (!(bnx2x_has_rx_work(fp) || bnx2x_has_tx_work(fp))) {
2145 napi_complete(napi);
2146 /* Re-enable interrupts */
2148 "Update index to %d\n", fp->fp_hc_idx);
2149 bnx2x_ack_sb(bp, fp->igu_sb_id, USTORM_ID,
2150 le16_to_cpu(fp->fp_hc_idx),
2160 /* we split the first BD into headers and data BDs
2161 * to ease the pain of our fellow microcode engineers
2162 * we use one mapping for both BDs
2163 * So far this has only been observed to happen
2164 * in Other Operating Systems(TM)
2166 static noinline u16 bnx2x_tx_split(struct bnx2x *bp,
2167 struct bnx2x_fastpath *fp,
2168 struct sw_tx_bd *tx_buf,
2169 struct eth_tx_start_bd **tx_bd, u16 hlen,
2170 u16 bd_prod, int nbd)
2172 struct eth_tx_start_bd *h_tx_bd = *tx_bd;
2173 struct eth_tx_bd *d_tx_bd;
2175 int old_len = le16_to_cpu(h_tx_bd->nbytes);
2177 /* first fix first BD */
2178 h_tx_bd->nbd = cpu_to_le16(nbd);
2179 h_tx_bd->nbytes = cpu_to_le16(hlen);
2181 DP(NETIF_MSG_TX_QUEUED, "TSO split header size is %d "
2182 "(%x:%x) nbd %d\n", h_tx_bd->nbytes, h_tx_bd->addr_hi,
2183 h_tx_bd->addr_lo, h_tx_bd->nbd);
2185 /* now get a new data BD
2186 * (after the pbd) and fill it */
2187 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
2188 d_tx_bd = &fp->tx_desc_ring[bd_prod].reg_bd;
2190 mapping = HILO_U64(le32_to_cpu(h_tx_bd->addr_hi),
2191 le32_to_cpu(h_tx_bd->addr_lo)) + hlen;
2193 d_tx_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
2194 d_tx_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
2195 d_tx_bd->nbytes = cpu_to_le16(old_len - hlen);
2197 /* this marks the BD as one that has no individual mapping */
2198 tx_buf->flags |= BNX2X_TSO_SPLIT_BD;
2200 DP(NETIF_MSG_TX_QUEUED,
2201 "TSO split data size is %d (%x:%x)\n",
2202 d_tx_bd->nbytes, d_tx_bd->addr_hi, d_tx_bd->addr_lo);
2205 *tx_bd = (struct eth_tx_start_bd *)d_tx_bd;
2210 static inline u16 bnx2x_csum_fix(unsigned char *t_header, u16 csum, s8 fix)
2213 csum = (u16) ~csum_fold(csum_sub(csum,
2214 csum_partial(t_header - fix, fix, 0)));
2217 csum = (u16) ~csum_fold(csum_add(csum,
2218 csum_partial(t_header, -fix, 0)));
2220 return swab16(csum);
2223 static inline u32 bnx2x_xmit_type(struct bnx2x *bp, struct sk_buff *skb)
2227 if (skb->ip_summed != CHECKSUM_PARTIAL)
2231 if (vlan_get_protocol(skb) == htons(ETH_P_IPV6)) {
2233 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2234 rc |= XMIT_CSUM_TCP;
2238 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2239 rc |= XMIT_CSUM_TCP;
2243 if (skb_is_gso_v6(skb))
2244 rc |= XMIT_GSO_V6 | XMIT_CSUM_TCP | XMIT_CSUM_V6;
2245 else if (skb_is_gso(skb))
2246 rc |= XMIT_GSO_V4 | XMIT_CSUM_V4 | XMIT_CSUM_TCP;
2251 #if (MAX_SKB_FRAGS >= MAX_FETCH_BD - 3)
2252 /* check if packet requires linearization (packet is too fragmented)
2253 no need to check fragmentation if page size > 8K (there will be no
2254 violation to FW restrictions) */
2255 static int bnx2x_pkt_req_lin(struct bnx2x *bp, struct sk_buff *skb,
2260 int first_bd_sz = 0;
2262 /* 3 = 1 (for linear data BD) + 2 (for PBD and last BD) */
2263 if (skb_shinfo(skb)->nr_frags >= (MAX_FETCH_BD - 3)) {
2265 if (xmit_type & XMIT_GSO) {
2266 unsigned short lso_mss = skb_shinfo(skb)->gso_size;
2267 /* Check if LSO packet needs to be copied:
2268 3 = 1 (for headers BD) + 2 (for PBD and last BD) */
2269 int wnd_size = MAX_FETCH_BD - 3;
2270 /* Number of windows to check */
2271 int num_wnds = skb_shinfo(skb)->nr_frags - wnd_size;
2276 /* Headers length */
2277 hlen = (int)(skb_transport_header(skb) - skb->data) +
2280 /* Amount of data (w/o headers) on linear part of SKB*/
2281 first_bd_sz = skb_headlen(skb) - hlen;
2283 wnd_sum = first_bd_sz;
2285 /* Calculate the first sum - it's special */
2286 for (frag_idx = 0; frag_idx < wnd_size - 1; frag_idx++)
2288 skb_shinfo(skb)->frags[frag_idx].size;
2290 /* If there was data on linear skb data - check it */
2291 if (first_bd_sz > 0) {
2292 if (unlikely(wnd_sum < lso_mss)) {
2297 wnd_sum -= first_bd_sz;
2300 /* Others are easier: run through the frag list and
2301 check all windows */
2302 for (wnd_idx = 0; wnd_idx <= num_wnds; wnd_idx++) {
2304 skb_shinfo(skb)->frags[wnd_idx + wnd_size - 1].size;
2306 if (unlikely(wnd_sum < lso_mss)) {
2311 skb_shinfo(skb)->frags[wnd_idx].size;
2314 /* in non-LSO too fragmented packet should always
2321 if (unlikely(to_copy))
2322 DP(NETIF_MSG_TX_QUEUED,
2323 "Linearization IS REQUIRED for %s packet. "
2324 "num_frags %d hlen %d first_bd_sz %d\n",
2325 (xmit_type & XMIT_GSO) ? "LSO" : "non-LSO",
2326 skb_shinfo(skb)->nr_frags, hlen, first_bd_sz);
2332 static inline void bnx2x_set_pbd_gso_e2(struct sk_buff *skb, u32 *parsing_data,
2335 *parsing_data |= (skb_shinfo(skb)->gso_size <<
2336 ETH_TX_PARSE_BD_E2_LSO_MSS_SHIFT) &
2337 ETH_TX_PARSE_BD_E2_LSO_MSS;
2338 if ((xmit_type & XMIT_GSO_V6) &&
2339 (ipv6_hdr(skb)->nexthdr == NEXTHDR_IPV6))
2340 *parsing_data |= ETH_TX_PARSE_BD_E2_IPV6_WITH_EXT_HDR;
2344 * bnx2x_set_pbd_gso - update PBD in GSO case.
2348 * @xmit_type: xmit flags
2350 static inline void bnx2x_set_pbd_gso(struct sk_buff *skb,
2351 struct eth_tx_parse_bd_e1x *pbd,
2354 pbd->lso_mss = cpu_to_le16(skb_shinfo(skb)->gso_size);
2355 pbd->tcp_send_seq = swab32(tcp_hdr(skb)->seq);
2356 pbd->tcp_flags = pbd_tcp_flags(skb);
2358 if (xmit_type & XMIT_GSO_V4) {
2359 pbd->ip_id = swab16(ip_hdr(skb)->id);
2360 pbd->tcp_pseudo_csum =
2361 swab16(~csum_tcpudp_magic(ip_hdr(skb)->saddr,
2363 0, IPPROTO_TCP, 0));
2366 pbd->tcp_pseudo_csum =
2367 swab16(~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2368 &ipv6_hdr(skb)->daddr,
2369 0, IPPROTO_TCP, 0));
2371 pbd->global_data |= ETH_TX_PARSE_BD_E1X_PSEUDO_CS_WITHOUT_LEN;
2375 * bnx2x_set_pbd_csum_e2 - update PBD with checksum and return header length
2377 * @bp: driver handle
2379 * @parsing_data: data to be updated
2380 * @xmit_type: xmit flags
2384 static inline u8 bnx2x_set_pbd_csum_e2(struct bnx2x *bp, struct sk_buff *skb,
2385 u32 *parsing_data, u32 xmit_type)
2388 ((((u8 *)skb_transport_header(skb) - skb->data) >> 1) <<
2389 ETH_TX_PARSE_BD_E2_TCP_HDR_START_OFFSET_W_SHIFT) &
2390 ETH_TX_PARSE_BD_E2_TCP_HDR_START_OFFSET_W;
2392 if (xmit_type & XMIT_CSUM_TCP) {
2393 *parsing_data |= ((tcp_hdrlen(skb) / 4) <<
2394 ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW_SHIFT) &
2395 ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW;
2397 return skb_transport_header(skb) + tcp_hdrlen(skb) - skb->data;
2399 /* We support checksum offload for TCP and UDP only.
2400 * No need to pass the UDP header length - it's a constant.
2402 return skb_transport_header(skb) +
2403 sizeof(struct udphdr) - skb->data;
2406 static inline void bnx2x_set_sbd_csum(struct bnx2x *bp, struct sk_buff *skb,
2407 struct eth_tx_start_bd *tx_start_bd, u32 xmit_type)
2409 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_L4_CSUM;
2411 if (xmit_type & XMIT_CSUM_V4)
2412 tx_start_bd->bd_flags.as_bitfield |=
2413 ETH_TX_BD_FLAGS_IP_CSUM;
2415 tx_start_bd->bd_flags.as_bitfield |=
2416 ETH_TX_BD_FLAGS_IPV6;
2418 if (!(xmit_type & XMIT_CSUM_TCP))
2419 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_IS_UDP;
2423 * bnx2x_set_pbd_csum - update PBD with checksum and return header length
2425 * @bp: driver handle
2427 * @pbd: parse BD to be updated
2428 * @xmit_type: xmit flags
2430 static inline u8 bnx2x_set_pbd_csum(struct bnx2x *bp, struct sk_buff *skb,
2431 struct eth_tx_parse_bd_e1x *pbd,
2434 u8 hlen = (skb_network_header(skb) - skb->data) >> 1;
2436 /* for now NS flag is not used in Linux */
2438 (hlen | ((skb->protocol == cpu_to_be16(ETH_P_8021Q)) <<
2439 ETH_TX_PARSE_BD_E1X_LLC_SNAP_EN_SHIFT));
2441 pbd->ip_hlen_w = (skb_transport_header(skb) -
2442 skb_network_header(skb)) >> 1;
2444 hlen += pbd->ip_hlen_w;
2446 /* We support checksum offload for TCP and UDP only */
2447 if (xmit_type & XMIT_CSUM_TCP)
2448 hlen += tcp_hdrlen(skb) / 2;
2450 hlen += sizeof(struct udphdr) / 2;
2452 pbd->total_hlen_w = cpu_to_le16(hlen);
2455 if (xmit_type & XMIT_CSUM_TCP) {
2456 pbd->tcp_pseudo_csum = swab16(tcp_hdr(skb)->check);
2459 s8 fix = SKB_CS_OFF(skb); /* signed! */
2461 DP(NETIF_MSG_TX_QUEUED,
2462 "hlen %d fix %d csum before fix %x\n",
2463 le16_to_cpu(pbd->total_hlen_w), fix, SKB_CS(skb));
2465 /* HW bug: fixup the CSUM */
2466 pbd->tcp_pseudo_csum =
2467 bnx2x_csum_fix(skb_transport_header(skb),
2470 DP(NETIF_MSG_TX_QUEUED, "csum after fix %x\n",
2471 pbd->tcp_pseudo_csum);
2477 /* called with netif_tx_lock
2478 * bnx2x_tx_int() runs without netif_tx_lock unless it needs to call
2479 * netif_wake_queue()
2481 netdev_tx_t bnx2x_start_xmit(struct sk_buff *skb, struct net_device *dev)
2483 struct bnx2x *bp = netdev_priv(dev);
2484 struct bnx2x_fastpath *fp;
2485 struct netdev_queue *txq;
2486 struct sw_tx_bd *tx_buf;
2487 struct eth_tx_start_bd *tx_start_bd, *first_bd;
2488 struct eth_tx_bd *tx_data_bd, *total_pkt_bd = NULL;
2489 struct eth_tx_parse_bd_e1x *pbd_e1x = NULL;
2490 struct eth_tx_parse_bd_e2 *pbd_e2 = NULL;
2491 u32 pbd_e2_parsing_data = 0;
2492 u16 pkt_prod, bd_prod;
2495 u32 xmit_type = bnx2x_xmit_type(bp, skb);
2498 __le16 pkt_size = 0;
2500 u8 mac_type = UNICAST_ADDRESS;
2502 #ifdef BNX2X_STOP_ON_ERROR
2503 if (unlikely(bp->panic))
2504 return NETDEV_TX_BUSY;
2507 fp_index = skb_get_queue_mapping(skb);
2508 txq = netdev_get_tx_queue(dev, fp_index);
2510 fp = &bp->fp[fp_index];
2512 if (unlikely(bnx2x_tx_avail(fp) < (skb_shinfo(skb)->nr_frags + 3))) {
2513 fp->eth_q_stats.driver_xoff++;
2514 netif_tx_stop_queue(txq);
2515 BNX2X_ERR("BUG! Tx ring full when queue awake!\n");
2516 return NETDEV_TX_BUSY;
2519 DP(NETIF_MSG_TX_QUEUED, "queue[%d]: SKB: summed %x protocol %x "
2520 "protocol(%x,%x) gso type %x xmit_type %x\n",
2521 fp_index, skb->ip_summed, skb->protocol, ipv6_hdr(skb)->nexthdr,
2522 ip_hdr(skb)->protocol, skb_shinfo(skb)->gso_type, xmit_type);
2524 eth = (struct ethhdr *)skb->data;
2526 /* set flag according to packet type (UNICAST_ADDRESS is default)*/
2527 if (unlikely(is_multicast_ether_addr(eth->h_dest))) {
2528 if (is_broadcast_ether_addr(eth->h_dest))
2529 mac_type = BROADCAST_ADDRESS;
2531 mac_type = MULTICAST_ADDRESS;
2534 #if (MAX_SKB_FRAGS >= MAX_FETCH_BD - 3)
2535 /* First, check if we need to linearize the skb (due to FW
2536 restrictions). No need to check fragmentation if page size > 8K
2537 (there will be no violation to FW restrictions) */
2538 if (bnx2x_pkt_req_lin(bp, skb, xmit_type)) {
2539 /* Statistics of linearization */
2541 if (skb_linearize(skb) != 0) {
2542 DP(NETIF_MSG_TX_QUEUED, "SKB linearization failed - "
2543 "silently dropping this SKB\n");
2544 dev_kfree_skb_any(skb);
2545 return NETDEV_TX_OK;
2549 /* Map skb linear data for DMA */
2550 mapping = dma_map_single(&bp->pdev->dev, skb->data,
2551 skb_headlen(skb), DMA_TO_DEVICE);
2552 if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
2553 DP(NETIF_MSG_TX_QUEUED, "SKB mapping failed - "
2554 "silently dropping this SKB\n");
2555 dev_kfree_skb_any(skb);
2556 return NETDEV_TX_OK;
2559 Please read carefully. First we use one BD which we mark as start,
2560 then we have a parsing info BD (used for TSO or xsum),
2561 and only then we have the rest of the TSO BDs.
2562 (don't forget to mark the last one as last,
2563 and to unmap only AFTER you write to the BD ...)
2564 And above all, all pdb sizes are in words - NOT DWORDS!
2567 /* get current pkt produced now - advance it just before sending packet
2568 * since mapping of pages may fail and cause packet to be dropped
2570 pkt_prod = fp->tx_pkt_prod;
2571 bd_prod = TX_BD(fp->tx_bd_prod);
2573 /* get a tx_buf and first BD
2574 * tx_start_bd may be changed during SPLIT,
2575 * but first_bd will always stay first
2577 tx_buf = &fp->tx_buf_ring[TX_BD(pkt_prod)];
2578 tx_start_bd = &fp->tx_desc_ring[bd_prod].start_bd;
2579 first_bd = tx_start_bd;
2581 tx_start_bd->bd_flags.as_bitfield = ETH_TX_BD_FLAGS_START_BD;
2582 SET_FLAG(tx_start_bd->general_data, ETH_TX_START_BD_ETH_ADDR_TYPE,
2586 SET_FLAG(tx_start_bd->general_data, ETH_TX_START_BD_HDR_NBDS, 1);
2588 /* remember the first BD of the packet */
2589 tx_buf->first_bd = fp->tx_bd_prod;
2593 DP(NETIF_MSG_TX_QUEUED,
2594 "sending pkt %u @%p next_idx %u bd %u @%p\n",
2595 pkt_prod, tx_buf, fp->tx_pkt_prod, bd_prod, tx_start_bd);
2597 if (vlan_tx_tag_present(skb)) {
2598 tx_start_bd->vlan_or_ethertype =
2599 cpu_to_le16(vlan_tx_tag_get(skb));
2600 tx_start_bd->bd_flags.as_bitfield |=
2601 (X_ETH_OUTBAND_VLAN << ETH_TX_BD_FLAGS_VLAN_MODE_SHIFT);
2603 tx_start_bd->vlan_or_ethertype = cpu_to_le16(pkt_prod);
2605 /* turn on parsing and get a BD */
2606 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
2608 if (xmit_type & XMIT_CSUM)
2609 bnx2x_set_sbd_csum(bp, skb, tx_start_bd, xmit_type);
2611 if (!CHIP_IS_E1x(bp)) {
2612 pbd_e2 = &fp->tx_desc_ring[bd_prod].parse_bd_e2;
2613 memset(pbd_e2, 0, sizeof(struct eth_tx_parse_bd_e2));
2614 /* Set PBD in checksum offload case */
2615 if (xmit_type & XMIT_CSUM)
2616 hlen = bnx2x_set_pbd_csum_e2(bp, skb,
2617 &pbd_e2_parsing_data,
2621 * fill in the MAC addresses in the PBD - for local
2624 bnx2x_set_fw_mac_addr(&pbd_e2->src_mac_addr_hi,
2625 &pbd_e2->src_mac_addr_mid,
2626 &pbd_e2->src_mac_addr_lo,
2628 bnx2x_set_fw_mac_addr(&pbd_e2->dst_mac_addr_hi,
2629 &pbd_e2->dst_mac_addr_mid,
2630 &pbd_e2->dst_mac_addr_lo,
2634 pbd_e1x = &fp->tx_desc_ring[bd_prod].parse_bd_e1x;
2635 memset(pbd_e1x, 0, sizeof(struct eth_tx_parse_bd_e1x));
2636 /* Set PBD in checksum offload case */
2637 if (xmit_type & XMIT_CSUM)
2638 hlen = bnx2x_set_pbd_csum(bp, skb, pbd_e1x, xmit_type);
2642 /* Setup the data pointer of the first BD of the packet */
2643 tx_start_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
2644 tx_start_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
2645 nbd = 2; /* start_bd + pbd + frags (updated when pages are mapped) */
2646 tx_start_bd->nbytes = cpu_to_le16(skb_headlen(skb));
2647 pkt_size = tx_start_bd->nbytes;
2649 DP(NETIF_MSG_TX_QUEUED, "first bd @%p addr (%x:%x) nbd %d"
2650 " nbytes %d flags %x vlan %x\n",
2651 tx_start_bd, tx_start_bd->addr_hi, tx_start_bd->addr_lo,
2652 le16_to_cpu(tx_start_bd->nbd), le16_to_cpu(tx_start_bd->nbytes),
2653 tx_start_bd->bd_flags.as_bitfield,
2654 le16_to_cpu(tx_start_bd->vlan_or_ethertype));
2656 if (xmit_type & XMIT_GSO) {
2658 DP(NETIF_MSG_TX_QUEUED,
2659 "TSO packet len %d hlen %d total len %d tso size %d\n",
2660 skb->len, hlen, skb_headlen(skb),
2661 skb_shinfo(skb)->gso_size);
2663 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_SW_LSO;
2665 if (unlikely(skb_headlen(skb) > hlen))
2666 bd_prod = bnx2x_tx_split(bp, fp, tx_buf, &tx_start_bd,
2667 hlen, bd_prod, ++nbd);
2668 if (!CHIP_IS_E1x(bp))
2669 bnx2x_set_pbd_gso_e2(skb, &pbd_e2_parsing_data,
2672 bnx2x_set_pbd_gso(skb, pbd_e1x, xmit_type);
2675 /* Set the PBD's parsing_data field if not zero
2676 * (for the chips newer than 57711).
2678 if (pbd_e2_parsing_data)
2679 pbd_e2->parsing_data = cpu_to_le32(pbd_e2_parsing_data);
2681 tx_data_bd = (struct eth_tx_bd *)tx_start_bd;
2683 /* Handle fragmented skb */
2684 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2685 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2687 mapping = dma_map_page(&bp->pdev->dev, frag->page,
2688 frag->page_offset, frag->size,
2690 if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
2692 DP(NETIF_MSG_TX_QUEUED, "Unable to map page - "
2693 "dropping packet...\n");
2695 /* we need unmap all buffers already mapped
2697 * first_bd->nbd need to be properly updated
2698 * before call to bnx2x_free_tx_pkt
2700 first_bd->nbd = cpu_to_le16(nbd);
2701 bnx2x_free_tx_pkt(bp, fp, TX_BD(fp->tx_pkt_prod));
2702 return NETDEV_TX_OK;
2705 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
2706 tx_data_bd = &fp->tx_desc_ring[bd_prod].reg_bd;
2707 if (total_pkt_bd == NULL)
2708 total_pkt_bd = &fp->tx_desc_ring[bd_prod].reg_bd;
2710 tx_data_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
2711 tx_data_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
2712 tx_data_bd->nbytes = cpu_to_le16(frag->size);
2713 le16_add_cpu(&pkt_size, frag->size);
2716 DP(NETIF_MSG_TX_QUEUED,
2717 "frag %d bd @%p addr (%x:%x) nbytes %d\n",
2718 i, tx_data_bd, tx_data_bd->addr_hi, tx_data_bd->addr_lo,
2719 le16_to_cpu(tx_data_bd->nbytes));
2722 DP(NETIF_MSG_TX_QUEUED, "last bd @%p\n", tx_data_bd);
2724 /* update with actual num BDs */
2725 first_bd->nbd = cpu_to_le16(nbd);
2727 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
2729 /* now send a tx doorbell, counting the next BD
2730 * if the packet contains or ends with it
2732 if (TX_BD_POFF(bd_prod) < nbd)
2735 /* total_pkt_bytes should be set on the first data BD if
2736 * it's not an LSO packet and there is more than one
2737 * data BD. In this case pkt_size is limited by an MTU value.
2738 * However we prefer to set it for an LSO packet (while we don't
2739 * have to) in order to save some CPU cycles in a none-LSO
2740 * case, when we much more care about them.
2742 if (total_pkt_bd != NULL)
2743 total_pkt_bd->total_pkt_bytes = pkt_size;
2746 DP(NETIF_MSG_TX_QUEUED,
2747 "PBD (E1X) @%p ip_data %x ip_hlen %u ip_id %u lso_mss %u"
2748 " tcp_flags %x xsum %x seq %u hlen %u\n",
2749 pbd_e1x, pbd_e1x->global_data, pbd_e1x->ip_hlen_w,
2750 pbd_e1x->ip_id, pbd_e1x->lso_mss, pbd_e1x->tcp_flags,
2751 pbd_e1x->tcp_pseudo_csum, pbd_e1x->tcp_send_seq,
2752 le16_to_cpu(pbd_e1x->total_hlen_w));
2754 DP(NETIF_MSG_TX_QUEUED,
2755 "PBD (E2) @%p dst %x %x %x src %x %x %x parsing_data %x\n",
2756 pbd_e2, pbd_e2->dst_mac_addr_hi, pbd_e2->dst_mac_addr_mid,
2757 pbd_e2->dst_mac_addr_lo, pbd_e2->src_mac_addr_hi,
2758 pbd_e2->src_mac_addr_mid, pbd_e2->src_mac_addr_lo,
2759 pbd_e2->parsing_data);
2760 DP(NETIF_MSG_TX_QUEUED, "doorbell: nbd %d bd %u\n", nbd, bd_prod);
2764 * Make sure that the BD data is updated before updating the producer
2765 * since FW might read the BD right after the producer is updated.
2766 * This is only applicable for weak-ordered memory model archs such
2767 * as IA-64. The following barrier is also mandatory since FW will
2768 * assumes packets must have BDs.
2772 fp->tx_db.data.prod += nbd;
2775 DOORBELL(bp, fp->cid, fp->tx_db.raw);
2779 fp->tx_bd_prod += nbd;
2781 if (unlikely(bnx2x_tx_avail(fp) < MAX_SKB_FRAGS + 3)) {
2782 netif_tx_stop_queue(txq);
2784 /* paired memory barrier is in bnx2x_tx_int(), we have to keep
2785 * ordering of set_bit() in netif_tx_stop_queue() and read of
2789 fp->eth_q_stats.driver_xoff++;
2790 if (bnx2x_tx_avail(fp) >= MAX_SKB_FRAGS + 3)
2791 netif_tx_wake_queue(txq);
2795 return NETDEV_TX_OK;
2798 /* called with rtnl_lock */
2799 int bnx2x_change_mac_addr(struct net_device *dev, void *p)
2801 struct sockaddr *addr = p;
2802 struct bnx2x *bp = netdev_priv(dev);
2805 if (!is_valid_ether_addr((u8 *)(addr->sa_data)))
2808 if (netif_running(dev)) {
2809 rc = bnx2x_set_eth_mac(bp, false);
2814 memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
2816 if (netif_running(dev))
2817 rc = bnx2x_set_eth_mac(bp, true);
2822 static void bnx2x_free_fp_mem_at(struct bnx2x *bp, int fp_index)
2824 union host_hc_status_block *sb = &bnx2x_fp(bp, fp_index, status_blk);
2825 struct bnx2x_fastpath *fp = &bp->fp[fp_index];
2829 if (IS_FCOE_IDX(fp_index)) {
2830 memset(sb, 0, sizeof(union host_hc_status_block));
2831 fp->status_blk_mapping = 0;
2836 if (!CHIP_IS_E1x(bp))
2837 BNX2X_PCI_FREE(sb->e2_sb,
2838 bnx2x_fp(bp, fp_index,
2839 status_blk_mapping),
2840 sizeof(struct host_hc_status_block_e2));
2842 BNX2X_PCI_FREE(sb->e1x_sb,
2843 bnx2x_fp(bp, fp_index,
2844 status_blk_mapping),
2845 sizeof(struct host_hc_status_block_e1x));
2850 if (!skip_rx_queue(bp, fp_index)) {
2851 bnx2x_free_rx_bds(fp);
2853 /* fastpath rx rings: rx_buf rx_desc rx_comp */
2854 BNX2X_FREE(bnx2x_fp(bp, fp_index, rx_buf_ring));
2855 BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_desc_ring),
2856 bnx2x_fp(bp, fp_index, rx_desc_mapping),
2857 sizeof(struct eth_rx_bd) * NUM_RX_BD);
2859 BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_comp_ring),
2860 bnx2x_fp(bp, fp_index, rx_comp_mapping),
2861 sizeof(struct eth_fast_path_rx_cqe) *
2865 BNX2X_FREE(bnx2x_fp(bp, fp_index, rx_page_ring));
2866 BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_sge_ring),
2867 bnx2x_fp(bp, fp_index, rx_sge_mapping),
2868 BCM_PAGE_SIZE * NUM_RX_SGE_PAGES);
2872 if (!skip_tx_queue(bp, fp_index)) {
2873 /* fastpath tx rings: tx_buf tx_desc */
2874 BNX2X_FREE(bnx2x_fp(bp, fp_index, tx_buf_ring));
2875 BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, tx_desc_ring),
2876 bnx2x_fp(bp, fp_index, tx_desc_mapping),
2877 sizeof(union eth_tx_bd_types) * NUM_TX_BD);
2879 /* end of fastpath */
2882 void bnx2x_free_fp_mem(struct bnx2x *bp)
2885 for_each_queue(bp, i)
2886 bnx2x_free_fp_mem_at(bp, i);
2889 static inline void set_sb_shortcuts(struct bnx2x *bp, int index)
2891 union host_hc_status_block status_blk = bnx2x_fp(bp, index, status_blk);
2892 if (!CHIP_IS_E1x(bp)) {
2893 bnx2x_fp(bp, index, sb_index_values) =
2894 (__le16 *)status_blk.e2_sb->sb.index_values;
2895 bnx2x_fp(bp, index, sb_running_index) =
2896 (__le16 *)status_blk.e2_sb->sb.running_index;
2898 bnx2x_fp(bp, index, sb_index_values) =
2899 (__le16 *)status_blk.e1x_sb->sb.index_values;
2900 bnx2x_fp(bp, index, sb_running_index) =
2901 (__le16 *)status_blk.e1x_sb->sb.running_index;
2905 static int bnx2x_alloc_fp_mem_at(struct bnx2x *bp, int index)
2907 union host_hc_status_block *sb;
2908 struct bnx2x_fastpath *fp = &bp->fp[index];
2911 /* if rx_ring_size specified - use it */
2912 int rx_ring_size = bp->rx_ring_size ? bp->rx_ring_size :
2913 MAX_RX_AVAIL/bp->num_queues;
2915 /* allocate at least number of buffers required by FW */
2916 rx_ring_size = max_t(int, fp->disable_tpa ? MIN_RX_SIZE_NONTPA :
2920 bnx2x_fp(bp, index, bp) = bp;
2921 bnx2x_fp(bp, index, index) = index;
2924 sb = &bnx2x_fp(bp, index, status_blk);
2926 if (!IS_FCOE_IDX(index)) {
2929 if (!CHIP_IS_E1x(bp))
2930 BNX2X_PCI_ALLOC(sb->e2_sb,
2931 &bnx2x_fp(bp, index, status_blk_mapping),
2932 sizeof(struct host_hc_status_block_e2));
2934 BNX2X_PCI_ALLOC(sb->e1x_sb,
2935 &bnx2x_fp(bp, index, status_blk_mapping),
2936 sizeof(struct host_hc_status_block_e1x));
2941 /* FCoE Queue uses Default SB and doesn't ACK the SB, thus no need to
2942 * set shortcuts for it.
2944 if (!IS_FCOE_IDX(index))
2945 set_sb_shortcuts(bp, index);
2948 if (!skip_tx_queue(bp, index)) {
2949 /* fastpath tx rings: tx_buf tx_desc */
2950 BNX2X_ALLOC(bnx2x_fp(bp, index, tx_buf_ring),
2951 sizeof(struct sw_tx_bd) * NUM_TX_BD);
2952 BNX2X_PCI_ALLOC(bnx2x_fp(bp, index, tx_desc_ring),
2953 &bnx2x_fp(bp, index, tx_desc_mapping),
2954 sizeof(union eth_tx_bd_types) * NUM_TX_BD);
2958 if (!skip_rx_queue(bp, index)) {
2959 /* fastpath rx rings: rx_buf rx_desc rx_comp */
2960 BNX2X_ALLOC(bnx2x_fp(bp, index, rx_buf_ring),
2961 sizeof(struct sw_rx_bd) * NUM_RX_BD);
2962 BNX2X_PCI_ALLOC(bnx2x_fp(bp, index, rx_desc_ring),
2963 &bnx2x_fp(bp, index, rx_desc_mapping),
2964 sizeof(struct eth_rx_bd) * NUM_RX_BD);
2966 BNX2X_PCI_ALLOC(bnx2x_fp(bp, index, rx_comp_ring),
2967 &bnx2x_fp(bp, index, rx_comp_mapping),
2968 sizeof(struct eth_fast_path_rx_cqe) *
2972 BNX2X_ALLOC(bnx2x_fp(bp, index, rx_page_ring),
2973 sizeof(struct sw_rx_page) * NUM_RX_SGE);
2974 BNX2X_PCI_ALLOC(bnx2x_fp(bp, index, rx_sge_ring),
2975 &bnx2x_fp(bp, index, rx_sge_mapping),
2976 BCM_PAGE_SIZE * NUM_RX_SGE_PAGES);
2978 bnx2x_set_next_page_rx_bd(fp);
2981 bnx2x_set_next_page_rx_cq(fp);
2984 ring_size = bnx2x_alloc_rx_bds(fp, rx_ring_size);
2985 if (ring_size < rx_ring_size)
2991 /* handles low memory cases */
2993 BNX2X_ERR("Unable to allocate full memory for queue %d (size %d)\n",
2995 /* FW will drop all packets if queue is not big enough,
2996 * In these cases we disable the queue
2997 * Min size diferent for TPA and non-TPA queues
2999 if (ring_size < (fp->disable_tpa ?
3000 MIN_RX_SIZE_NONTPA : MIN_RX_SIZE_TPA)) {
3001 /* release memory allocated for this queue */
3002 bnx2x_free_fp_mem_at(bp, index);
3008 int bnx2x_alloc_fp_mem(struct bnx2x *bp)
3013 * 1. Allocate FP for leading - fatal if error
3014 * 2. {CNIC} Allocate FCoE FP - fatal if error
3015 * 3. Allocate RSS - fix number of queues if error
3019 if (bnx2x_alloc_fp_mem_at(bp, 0))
3024 if (bnx2x_alloc_fp_mem_at(bp, FCOE_IDX))
3025 /* we will fail load process instead of mark
3031 for_each_nondefault_eth_queue(bp, i)
3032 if (bnx2x_alloc_fp_mem_at(bp, i))
3035 /* handle memory failures */
3036 if (i != BNX2X_NUM_ETH_QUEUES(bp)) {
3037 int delta = BNX2X_NUM_ETH_QUEUES(bp) - i;
3042 * move non eth FPs next to last eth FP
3043 * must be done in that order
3044 * FCOE_IDX < FWD_IDX < OOO_IDX
3048 bnx2x_move_fp(bp, FCOE_IDX, FCOE_IDX - delta);
3050 bp->num_queues -= delta;
3051 BNX2X_ERR("Adjusted num of queues from %d to %d\n",
3052 bp->num_queues + delta, bp->num_queues);
3058 void bnx2x_free_mem_bp(struct bnx2x *bp)
3061 kfree(bp->msix_table);
3065 int __devinit bnx2x_alloc_mem_bp(struct bnx2x *bp)
3067 struct bnx2x_fastpath *fp;
3068 struct msix_entry *tbl;
3069 struct bnx2x_ilt *ilt;
3072 fp = kzalloc(L2_FP_COUNT(bp->l2_cid_count)*sizeof(*fp), GFP_KERNEL);
3078 tbl = kzalloc((FP_SB_COUNT(bp->l2_cid_count) + 1) * sizeof(*tbl),
3082 bp->msix_table = tbl;
3085 ilt = kzalloc(sizeof(*ilt), GFP_KERNEL);
3092 bnx2x_free_mem_bp(bp);
3097 int bnx2x_reload_if_running(struct net_device *dev)
3099 struct bnx2x *bp = netdev_priv(dev);
3101 if (unlikely(!netif_running(dev)))
3104 bnx2x_nic_unload(bp, UNLOAD_NORMAL);
3105 return bnx2x_nic_load(bp, LOAD_NORMAL);
3108 int bnx2x_get_cur_phy_idx(struct bnx2x *bp)
3110 u32 sel_phy_idx = 0;
3111 if (bp->link_params.num_phys <= 1)
3114 if (bp->link_vars.link_up) {
3115 sel_phy_idx = EXT_PHY1;
3116 /* In case link is SERDES, check if the EXT_PHY2 is the one */
3117 if ((bp->link_vars.link_status & LINK_STATUS_SERDES_LINK) &&
3118 (bp->link_params.phy[EXT_PHY2].supported & SUPPORTED_FIBRE))
3119 sel_phy_idx = EXT_PHY2;
3122 switch (bnx2x_phy_selection(&bp->link_params)) {
3123 case PORT_HW_CFG_PHY_SELECTION_HARDWARE_DEFAULT:
3124 case PORT_HW_CFG_PHY_SELECTION_FIRST_PHY:
3125 case PORT_HW_CFG_PHY_SELECTION_FIRST_PHY_PRIORITY:
3126 sel_phy_idx = EXT_PHY1;
3128 case PORT_HW_CFG_PHY_SELECTION_SECOND_PHY:
3129 case PORT_HW_CFG_PHY_SELECTION_SECOND_PHY_PRIORITY:
3130 sel_phy_idx = EXT_PHY2;
3138 int bnx2x_get_link_cfg_idx(struct bnx2x *bp)
3140 u32 sel_phy_idx = bnx2x_get_cur_phy_idx(bp);
3142 * The selected actived PHY is always after swapping (in case PHY
3143 * swapping is enabled). So when swapping is enabled, we need to reverse
3147 if (bp->link_params.multi_phy_config &
3148 PORT_HW_CFG_PHY_SWAPPED_ENABLED) {
3149 if (sel_phy_idx == EXT_PHY1)
3150 sel_phy_idx = EXT_PHY2;
3151 else if (sel_phy_idx == EXT_PHY2)
3152 sel_phy_idx = EXT_PHY1;
3154 return LINK_CONFIG_IDX(sel_phy_idx);
3157 /* called with rtnl_lock */
3158 int bnx2x_change_mtu(struct net_device *dev, int new_mtu)
3160 struct bnx2x *bp = netdev_priv(dev);
3162 if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
3163 printk(KERN_ERR "Handling parity error recovery. Try again later\n");
3167 if ((new_mtu > ETH_MAX_JUMBO_PACKET_SIZE) ||
3168 ((new_mtu + ETH_HLEN) < ETH_MIN_PACKET_SIZE))
3171 /* This does not race with packet allocation
3172 * because the actual alloc size is
3173 * only updated as part of load
3177 return bnx2x_reload_if_running(dev);
3180 u32 bnx2x_fix_features(struct net_device *dev, u32 features)
3182 struct bnx2x *bp = netdev_priv(dev);
3184 /* TPA requires Rx CSUM offloading */
3185 if (!(features & NETIF_F_RXCSUM) || bp->disable_tpa)
3186 features &= ~NETIF_F_LRO;
3191 int bnx2x_set_features(struct net_device *dev, u32 features)
3193 struct bnx2x *bp = netdev_priv(dev);
3194 u32 flags = bp->flags;
3195 bool bnx2x_reload = false;
3197 if (features & NETIF_F_LRO)
3198 flags |= TPA_ENABLE_FLAG;
3200 flags &= ~TPA_ENABLE_FLAG;
3202 if (features & NETIF_F_LOOPBACK) {
3203 if (bp->link_params.loopback_mode != LOOPBACK_BMAC) {
3204 bp->link_params.loopback_mode = LOOPBACK_BMAC;
3205 bnx2x_reload = true;
3208 if (bp->link_params.loopback_mode != LOOPBACK_NONE) {
3209 bp->link_params.loopback_mode = LOOPBACK_NONE;
3210 bnx2x_reload = true;
3214 if (flags ^ bp->flags) {
3216 bnx2x_reload = true;
3220 if (bp->recovery_state == BNX2X_RECOVERY_DONE)
3221 return bnx2x_reload_if_running(dev);
3222 /* else: bnx2x_nic_load() will be called at end of recovery */
3228 void bnx2x_tx_timeout(struct net_device *dev)
3230 struct bnx2x *bp = netdev_priv(dev);
3232 #ifdef BNX2X_STOP_ON_ERROR
3236 /* This allows the netif to be shutdown gracefully before resetting */
3237 schedule_delayed_work(&bp->reset_task, 0);
3240 int bnx2x_suspend(struct pci_dev *pdev, pm_message_t state)
3242 struct net_device *dev = pci_get_drvdata(pdev);
3246 dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
3249 bp = netdev_priv(dev);
3253 pci_save_state(pdev);
3255 if (!netif_running(dev)) {
3260 netif_device_detach(dev);
3262 bnx2x_nic_unload(bp, UNLOAD_CLOSE);
3264 bnx2x_set_power_state(bp, pci_choose_state(pdev, state));
3271 int bnx2x_resume(struct pci_dev *pdev)
3273 struct net_device *dev = pci_get_drvdata(pdev);
3278 dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
3281 bp = netdev_priv(dev);
3283 if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
3284 printk(KERN_ERR "Handling parity error recovery. Try again later\n");
3290 pci_restore_state(pdev);
3292 if (!netif_running(dev)) {
3297 bnx2x_set_power_state(bp, PCI_D0);
3298 netif_device_attach(dev);
3300 /* Since the chip was reset, clear the FW sequence number */
3302 rc = bnx2x_nic_load(bp, LOAD_OPEN);
3310 void bnx2x_set_ctx_validation(struct bnx2x *bp, struct eth_context *cxt,
3313 /* ustorm cxt validation */
3314 cxt->ustorm_ag_context.cdu_usage =
3315 CDU_RSRVD_VALUE_TYPE_A(HW_CID(bp, cid),
3316 CDU_REGION_NUMBER_UCM_AG, ETH_CONNECTION_TYPE);
3317 /* xcontext validation */
3318 cxt->xstorm_ag_context.cdu_reserved =
3319 CDU_RSRVD_VALUE_TYPE_A(HW_CID(bp, cid),
3320 CDU_REGION_NUMBER_XCM_AG, ETH_CONNECTION_TYPE);
3323 static inline void storm_memset_hc_timeout(struct bnx2x *bp, u8 port,
3324 u8 fw_sb_id, u8 sb_index,
3328 u32 addr = BAR_CSTRORM_INTMEM +
3329 CSTORM_STATUS_BLOCK_DATA_TIMEOUT_OFFSET(fw_sb_id, sb_index);
3330 REG_WR8(bp, addr, ticks);
3331 DP(NETIF_MSG_HW, "port %x fw_sb_id %d sb_index %d ticks %d\n",
3332 port, fw_sb_id, sb_index, ticks);
3335 static inline void storm_memset_hc_disable(struct bnx2x *bp, u8 port,
3336 u16 fw_sb_id, u8 sb_index,
3339 u32 enable_flag = disable ? 0 : (1 << HC_INDEX_DATA_HC_ENABLED_SHIFT);
3340 u32 addr = BAR_CSTRORM_INTMEM +
3341 CSTORM_STATUS_BLOCK_DATA_FLAGS_OFFSET(fw_sb_id, sb_index);
3342 u16 flags = REG_RD16(bp, addr);
3344 flags &= ~HC_INDEX_DATA_HC_ENABLED;
3345 flags |= enable_flag;
3346 REG_WR16(bp, addr, flags);
3347 DP(NETIF_MSG_HW, "port %x fw_sb_id %d sb_index %d disable %d\n",
3348 port, fw_sb_id, sb_index, disable);
3351 void bnx2x_update_coalesce_sb_index(struct bnx2x *bp, u8 fw_sb_id,
3352 u8 sb_index, u8 disable, u16 usec)
3354 int port = BP_PORT(bp);
3355 u8 ticks = usec / BNX2X_BTR;
3357 storm_memset_hc_timeout(bp, port, fw_sb_id, sb_index, ticks);
3359 disable = disable ? 1 : (usec ? 0 : 1);
3360 storm_memset_hc_disable(bp, port, fw_sb_id, sb_index, disable);