1/* bnx2x_cmn.c: QLogic Everest network driver.
2 *
3 * Copyright (c) 2007-2013 Broadcom Corporation
4 * Copyright (c) 2014 QLogic Corporation
5 * All rights reserved
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation.
10 *
11 * Maintained by: Ariel Elior <ariel.elior@qlogic.com>
12 * Written by: Eliezer Tamir
13 * Based on code from Michael Chan's bnx2 driver
14 * UDP CSUM errata workaround by Arik Gendelman
15 * Slowpath and fastpath rework by Vladislav Zolotarov
16 * Statistics and Link management by Yitchak Gertner
17 *
18 */
19
20#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
21
22#include <linux/etherdevice.h>
23#include <linux/if_vlan.h>
24#include <linux/interrupt.h>
25#include <linux/ip.h>
26#include <linux/crash_dump.h>
27#include <net/tcp.h>
28#include <net/ipv6.h>
29#include <net/ip6_checksum.h>
30#include <linux/prefetch.h>
31#include "bnx2x_cmn.h"
32#include "bnx2x_init.h"
33#include "bnx2x_sp.h"
34
35static void bnx2x_free_fp_mem_cnic(struct bnx2x *bp);
36static int bnx2x_alloc_fp_mem_cnic(struct bnx2x *bp);
37static int bnx2x_alloc_fp_mem(struct bnx2x *bp);
38static int bnx2x_poll(struct napi_struct *napi, int budget);
39
40static void bnx2x_add_all_napi_cnic(struct bnx2x *bp)
41{
42	int i;
43
44	/* Add NAPI objects */
45	for_each_rx_queue_cnic(bp, i) {
46		netif_napi_add(bp->dev, &bnx2x_fp(bp, i, napi),
47			       bnx2x_poll, NAPI_POLL_WEIGHT);
48	}
49}
50
51static void bnx2x_add_all_napi(struct bnx2x *bp)
52{
53	int i;
54
55	/* Add NAPI objects */
56	for_each_eth_queue(bp, i) {
57		netif_napi_add(bp->dev, &bnx2x_fp(bp, i, napi),
58			       bnx2x_poll, NAPI_POLL_WEIGHT);
59	}
60}
61
62static int bnx2x_calc_num_queues(struct bnx2x *bp)
63{
64	int nq = bnx2x_num_queues ? : netif_get_num_default_rss_queues();
65
66	/* Reduce memory usage in kdump environment by using only one queue */
67	if (is_kdump_kernel())
68		nq = 1;
69
70	nq = clamp(nq, 1, BNX2X_MAX_QUEUES(bp));
71	return nq;
72}
73
74/**
75 * bnx2x_move_fp - move content of the fastpath structure.
76 *
77 * @bp:		driver handle
78 * @from:	source FP index
79 * @to:		destination FP index
80 *
81 * Makes sure the contents of the bp->fp[to].napi is kept
82 * intact. This is done by first copying the napi struct from
83 * the target to the source, and then mem copying the entire
84 * source onto the target. Update txdata pointers and related
85 * content.
86 */
87static inline void bnx2x_move_fp(struct bnx2x *bp, int from, int to)
88{
89	struct bnx2x_fastpath *from_fp = &bp->fp[from];
90	struct bnx2x_fastpath *to_fp = &bp->fp[to];
91	struct bnx2x_sp_objs *from_sp_objs = &bp->sp_objs[from];
92	struct bnx2x_sp_objs *to_sp_objs = &bp->sp_objs[to];
93	struct bnx2x_fp_stats *from_fp_stats = &bp->fp_stats[from];
94	struct bnx2x_fp_stats *to_fp_stats = &bp->fp_stats[to];
95	int old_max_eth_txqs, new_max_eth_txqs;
96	int old_txdata_index = 0, new_txdata_index = 0;
97	struct bnx2x_agg_info *old_tpa_info = to_fp->tpa_info;
98
99	/* Copy the NAPI object as it has been already initialized */
100	from_fp->napi = to_fp->napi;
101
102	/* Move bnx2x_fastpath contents */
103	memcpy(to_fp, from_fp, sizeof(*to_fp));
104	to_fp->index = to;
105
106	/* Retain the tpa_info of the original `to' version as we don't want
107	 * 2 FPs to contain the same tpa_info pointer.
108	 */
109	to_fp->tpa_info = old_tpa_info;
110
111	/* move sp_objs contents as well, as their indices match fp ones */
112	memcpy(to_sp_objs, from_sp_objs, sizeof(*to_sp_objs));
113
114	/* move fp_stats contents as well, as their indices match fp ones */
115	memcpy(to_fp_stats, from_fp_stats, sizeof(*to_fp_stats));
116
117	/* Update txdata pointers in fp and move txdata content accordingly:
118	 * Each fp consumes 'max_cos' txdata structures, so the index should be
119	 * decremented by max_cos x delta.
120	 */
121
122	old_max_eth_txqs = BNX2X_NUM_ETH_QUEUES(bp) * (bp)->max_cos;
123	new_max_eth_txqs = (BNX2X_NUM_ETH_QUEUES(bp) - from + to) *
124				(bp)->max_cos;
125	if (from == FCOE_IDX(bp)) {
126		old_txdata_index = old_max_eth_txqs + FCOE_TXQ_IDX_OFFSET;
127		new_txdata_index = new_max_eth_txqs + FCOE_TXQ_IDX_OFFSET;
128	}
129
130	memcpy(&bp->bnx2x_txq[new_txdata_index],
131	       &bp->bnx2x_txq[old_txdata_index],
132	       sizeof(struct bnx2x_fp_txdata));
133	to_fp->txdata_ptr[0] = &bp->bnx2x_txq[new_txdata_index];
134}
135
136/**
137 * bnx2x_fill_fw_str - Fill buffer with FW version string.
138 *
139 * @bp:        driver handle
140 * @buf:       character buffer to fill with the fw name
141 * @buf_len:   length of the above buffer
142 *
143 */
144void bnx2x_fill_fw_str(struct bnx2x *bp, char *buf, size_t buf_len)
145{
146	if (IS_PF(bp)) {
147		u8 phy_fw_ver[PHY_FW_VER_LEN];
148
149		phy_fw_ver[0] = '\0';
150		bnx2x_get_ext_phy_fw_version(&bp->link_params,
151					     phy_fw_ver, PHY_FW_VER_LEN);
152		strlcpy(buf, bp->fw_ver, buf_len);
153		snprintf(buf + strlen(bp->fw_ver), 32 - strlen(bp->fw_ver),
154			 "bc %d.%d.%d%s%s",
155			 (bp->common.bc_ver & 0xff0000) >> 16,
156			 (bp->common.bc_ver & 0xff00) >> 8,
157			 (bp->common.bc_ver & 0xff),
158			 ((phy_fw_ver[0] != '\0') ? " phy " : ""), phy_fw_ver);
159	} else {
160		bnx2x_vf_fill_fw_str(bp, buf, buf_len);
161	}
162}
163
164/**
165 * bnx2x_shrink_eth_fp - guarantees fastpath structures stay intact
166 *
167 * @bp:	driver handle
168 * @delta:	number of eth queues which were not allocated
169 */
170static void bnx2x_shrink_eth_fp(struct bnx2x *bp, int delta)
171{
172	int i, cos, old_eth_num = BNX2X_NUM_ETH_QUEUES(bp);
173
174	/* Queue pointer cannot be re-set on an fp-basis, as moving pointer
175	 * backward along the array could cause memory to be overridden
176	 */
177	for (cos = 1; cos < bp->max_cos; cos++) {
178		for (i = 0; i < old_eth_num - delta; i++) {
179			struct bnx2x_fastpath *fp = &bp->fp[i];
180			int new_idx = cos * (old_eth_num - delta) + i;
181
182			memcpy(&bp->bnx2x_txq[new_idx], fp->txdata_ptr[cos],
183			       sizeof(struct bnx2x_fp_txdata));
184			fp->txdata_ptr[cos] = &bp->bnx2x_txq[new_idx];
185		}
186	}
187}
188
189int bnx2x_load_count[2][3] = { {0} }; /* per-path: 0-common, 1-port0, 2-port1 */
190
191/* free skb in the packet ring at pos idx
192 * return idx of last bd freed
193 */
194static u16 bnx2x_free_tx_pkt(struct bnx2x *bp, struct bnx2x_fp_txdata *txdata,
195			     u16 idx, unsigned int *pkts_compl,
196			     unsigned int *bytes_compl)
197{
198	struct sw_tx_bd *tx_buf = &txdata->tx_buf_ring[idx];
199	struct eth_tx_start_bd *tx_start_bd;
200	struct eth_tx_bd *tx_data_bd;
201	struct sk_buff *skb = tx_buf->skb;
202	u16 bd_idx = TX_BD(tx_buf->first_bd), new_cons;
203	int nbd;
204	u16 split_bd_len = 0;
205
206	/* prefetch skb end pointer to speedup dev_kfree_skb() */
207	prefetch(&skb->end);
208
209	DP(NETIF_MSG_TX_DONE, "fp[%d]: pkt_idx %d  buff @(%p)->skb %p\n",
210	   txdata->txq_index, idx, tx_buf, skb);
211
212	tx_start_bd = &txdata->tx_desc_ring[bd_idx].start_bd;
213
214	nbd = le16_to_cpu(tx_start_bd->nbd) - 1;
215#ifdef BNX2X_STOP_ON_ERROR
216	if ((nbd - 1) > (MAX_SKB_FRAGS + 2)) {
217		BNX2X_ERR("BAD nbd!\n");
218		bnx2x_panic();
219	}
220#endif
221	new_cons = nbd + tx_buf->first_bd;
222
223	/* Get the next bd */
224	bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
225
226	/* Skip a parse bd... */
227	--nbd;
228	bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
229
230	if (tx_buf->flags & BNX2X_HAS_SECOND_PBD) {
231		/* Skip second parse bd... */
232		--nbd;
233		bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
234	}
235
236	/* TSO headers+data bds share a common mapping. See bnx2x_tx_split() */
237	if (tx_buf->flags & BNX2X_TSO_SPLIT_BD) {
238		tx_data_bd = &txdata->tx_desc_ring[bd_idx].reg_bd;
239		split_bd_len = BD_UNMAP_LEN(tx_data_bd);
240		--nbd;
241		bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
242	}
243
244	/* unmap first bd */
245	dma_unmap_single(&bp->pdev->dev, BD_UNMAP_ADDR(tx_start_bd),
246			 BD_UNMAP_LEN(tx_start_bd) + split_bd_len,
247			 DMA_TO_DEVICE);
248
249	/* now free frags */
250	while (nbd > 0) {
251
252		tx_data_bd = &txdata->tx_desc_ring[bd_idx].reg_bd;
253		dma_unmap_page(&bp->pdev->dev, BD_UNMAP_ADDR(tx_data_bd),
254			       BD_UNMAP_LEN(tx_data_bd), DMA_TO_DEVICE);
255		if (--nbd)
256			bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
257	}
258
259	/* release skb */
260	WARN_ON(!skb);
261	if (likely(skb)) {
262		(*pkts_compl)++;
263		(*bytes_compl) += skb->len;
264		dev_kfree_skb_any(skb);
265	}
266
267	tx_buf->first_bd = 0;
268	tx_buf->skb = NULL;
269
270	return new_cons;
271}
272
273int bnx2x_tx_int(struct bnx2x *bp, struct bnx2x_fp_txdata *txdata)
274{
275	struct netdev_queue *txq;
276	u16 hw_cons, sw_cons, bd_cons = txdata->tx_bd_cons;
277	unsigned int pkts_compl = 0, bytes_compl = 0;
278
279#ifdef BNX2X_STOP_ON_ERROR
280	if (unlikely(bp->panic))
281		return -1;
282#endif
283
284	txq = netdev_get_tx_queue(bp->dev, txdata->txq_index);
285	hw_cons = le16_to_cpu(*txdata->tx_cons_sb);
286	sw_cons = txdata->tx_pkt_cons;
287
288	/* Ensure subsequent loads occur after hw_cons */
289	smp_rmb();
290
291	while (sw_cons != hw_cons) {
292		u16 pkt_cons;
293
294		pkt_cons = TX_BD(sw_cons);
295
296		DP(NETIF_MSG_TX_DONE,
297		   "queue[%d]: hw_cons %u  sw_cons %u  pkt_cons %u\n",
298		   txdata->txq_index, hw_cons, sw_cons, pkt_cons);
299
300		bd_cons = bnx2x_free_tx_pkt(bp, txdata, pkt_cons,
301					    &pkts_compl, &bytes_compl);
302
303		sw_cons++;
304	}
305
306	netdev_tx_completed_queue(txq, pkts_compl, bytes_compl);
307
308	txdata->tx_pkt_cons = sw_cons;
309	txdata->tx_bd_cons = bd_cons;
310
311	/* Need to make the tx_bd_cons update visible to start_xmit()
312	 * before checking for netif_tx_queue_stopped().  Without the
313	 * memory barrier, there is a small possibility that
314	 * start_xmit() will miss it and cause the queue to be stopped
315	 * forever.
316	 * On the other hand we need an rmb() here to ensure the proper
317	 * ordering of bit testing in the following
318	 * netif_tx_queue_stopped(txq) call.
319	 */
320	smp_mb();
321
322	if (unlikely(netif_tx_queue_stopped(txq))) {
323		/* Taking tx_lock() is needed to prevent re-enabling the queue
324		 * while it's empty. This could have happen if rx_action() gets
325		 * suspended in bnx2x_tx_int() after the condition before
326		 * netif_tx_wake_queue(), while tx_action (bnx2x_start_xmit()):
327		 *
328		 * stops the queue->sees fresh tx_bd_cons->releases the queue->
329		 * sends some packets consuming the whole queue again->
330		 * stops the queue
331		 */
332
333		__netif_tx_lock(txq, smp_processor_id());
334
335		if ((netif_tx_queue_stopped(txq)) &&
336		    (bp->state == BNX2X_STATE_OPEN) &&
337		    (bnx2x_tx_avail(bp, txdata) >= MAX_DESC_PER_TX_PKT))
338			netif_tx_wake_queue(txq);
339
340		__netif_tx_unlock(txq);
341	}
342	return 0;
343}
344
345static inline void bnx2x_update_last_max_sge(struct bnx2x_fastpath *fp,
346					     u16 idx)
347{
348	u16 last_max = fp->last_max_sge;
349
350	if (SUB_S16(idx, last_max) > 0)
351		fp->last_max_sge = idx;
352}
353
354static inline void bnx2x_update_sge_prod(struct bnx2x_fastpath *fp,
355					 u16 sge_len,
356					 struct eth_end_agg_rx_cqe *cqe)
357{
358	struct bnx2x *bp = fp->bp;
359	u16 last_max, last_elem, first_elem;
360	u16 delta = 0;
361	u16 i;
362
363	if (!sge_len)
364		return;
365
366	/* First mark all used pages */
367	for (i = 0; i < sge_len; i++)
368		BIT_VEC64_CLEAR_BIT(fp->sge_mask,
369			RX_SGE(le16_to_cpu(cqe->sgl_or_raw_data.sgl[i])));
370
371	DP(NETIF_MSG_RX_STATUS, "fp_cqe->sgl[%d] = %d\n",
372	   sge_len - 1, le16_to_cpu(cqe->sgl_or_raw_data.sgl[sge_len - 1]));
373
374	/* Here we assume that the last SGE index is the biggest */
375	prefetch((void *)(fp->sge_mask));
376	bnx2x_update_last_max_sge(fp,
377		le16_to_cpu(cqe->sgl_or_raw_data.sgl[sge_len - 1]));
378
379	last_max = RX_SGE(fp->last_max_sge);
380	last_elem = last_max >> BIT_VEC64_ELEM_SHIFT;
381	first_elem = RX_SGE(fp->rx_sge_prod) >> BIT_VEC64_ELEM_SHIFT;
382
383	/* If ring is not full */
384	if (last_elem + 1 != first_elem)
385		last_elem++;
386
387	/* Now update the prod */
388	for (i = first_elem; i != last_elem; i = NEXT_SGE_MASK_ELEM(i)) {
389		if (likely(fp->sge_mask[i]))
390			break;
391
392		fp->sge_mask[i] = BIT_VEC64_ELEM_ONE_MASK;
393		delta += BIT_VEC64_ELEM_SZ;
394	}
395
396	if (delta > 0) {
397		fp->rx_sge_prod += delta;
398		/* clear page-end entries */
399		bnx2x_clear_sge_mask_next_elems(fp);
400	}
401
402	DP(NETIF_MSG_RX_STATUS,
403	   "fp->last_max_sge = %d  fp->rx_sge_prod = %d\n",
404	   fp->last_max_sge, fp->rx_sge_prod);
405}
406
407/* Get Toeplitz hash value in the skb using the value from the
408 * CQE (calculated by HW).
409 */
410static u32 bnx2x_get_rxhash(const struct bnx2x *bp,
411			    const struct eth_fast_path_rx_cqe *cqe,
412			    enum pkt_hash_types *rxhash_type)
413{
414	/* Get Toeplitz hash from CQE */
415	if ((bp->dev->features & NETIF_F_RXHASH) &&
416	    (cqe->status_flags & ETH_FAST_PATH_RX_CQE_RSS_HASH_FLG)) {
417		enum eth_rss_hash_type htype;
418
419		htype = cqe->status_flags & ETH_FAST_PATH_RX_CQE_RSS_HASH_TYPE;
420		*rxhash_type = ((htype == TCP_IPV4_HASH_TYPE) ||
421				(htype == TCP_IPV6_HASH_TYPE)) ?
422			       PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3;
423
424		return le32_to_cpu(cqe->rss_hash_result);
425	}
426	*rxhash_type = PKT_HASH_TYPE_NONE;
427	return 0;
428}
429
430static void bnx2x_tpa_start(struct bnx2x_fastpath *fp, u16 queue,
431			    u16 cons, u16 prod,
432			    struct eth_fast_path_rx_cqe *cqe)
433{
434	struct bnx2x *bp = fp->bp;
435	struct sw_rx_bd *cons_rx_buf = &fp->rx_buf_ring[cons];
436	struct sw_rx_bd *prod_rx_buf = &fp->rx_buf_ring[prod];
437	struct eth_rx_bd *prod_bd = &fp->rx_desc_ring[prod];
438	dma_addr_t mapping;
439	struct bnx2x_agg_info *tpa_info = &fp->tpa_info[queue];
440	struct sw_rx_bd *first_buf = &tpa_info->first_buf;
441
442	/* print error if current state != stop */
443	if (tpa_info->tpa_state != BNX2X_TPA_STOP)
444		BNX2X_ERR("start of bin not in stop [%d]\n", queue);
445
446	/* Try to map an empty data buffer from the aggregation info  */
447	mapping = dma_map_single(&bp->pdev->dev,
448				 first_buf->data + NET_SKB_PAD,
449				 fp->rx_buf_size, DMA_FROM_DEVICE);
450	/*
451	 *  ...if it fails - move the skb from the consumer to the producer
452	 *  and set the current aggregation state as ERROR to drop it
453	 *  when TPA_STOP arrives.
454	 */
455
456	if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
457		/* Move the BD from the consumer to the producer */
458		bnx2x_reuse_rx_data(fp, cons, prod);
459		tpa_info->tpa_state = BNX2X_TPA_ERROR;
460		return;
461	}
462
463	/* move empty data from pool to prod */
464	prod_rx_buf->data = first_buf->data;
465	dma_unmap_addr_set(prod_rx_buf, mapping, mapping);
466	/* point prod_bd to new data */
467	prod_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
468	prod_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
469
470	/* move partial skb from cons to pool (don't unmap yet) */
471	*first_buf = *cons_rx_buf;
472
473	/* mark bin state as START */
474	tpa_info->parsing_flags =
475		le16_to_cpu(cqe->pars_flags.flags);
476	tpa_info->vlan_tag = le16_to_cpu(cqe->vlan_tag);
477	tpa_info->tpa_state = BNX2X_TPA_START;
478	tpa_info->len_on_bd = le16_to_cpu(cqe->len_on_bd);
479	tpa_info->placement_offset = cqe->placement_offset;
480	tpa_info->rxhash = bnx2x_get_rxhash(bp, cqe, &tpa_info->rxhash_type);
481	if (fp->mode == TPA_MODE_GRO) {
482		u16 gro_size = le16_to_cpu(cqe->pkt_len_or_gro_seg_len);
483		tpa_info->full_page = SGE_PAGES / gro_size * gro_size;
484		tpa_info->gro_size = gro_size;
485	}
486
487#ifdef BNX2X_STOP_ON_ERROR
488	fp->tpa_queue_used |= (1 << queue);
489	DP(NETIF_MSG_RX_STATUS, "fp->tpa_queue_used = 0x%llx\n",
490	   fp->tpa_queue_used);
491#endif
492}
493
494/* Timestamp option length allowed for TPA aggregation:
495 *
496 *		nop nop kind length echo val
497 */
498#define TPA_TSTAMP_OPT_LEN	12
499/**
500 * bnx2x_set_gro_params - compute GRO values
501 *
502 * @skb:		packet skb
503 * @parsing_flags:	parsing flags from the START CQE
504 * @len_on_bd:		total length of the first packet for the
505 *			aggregation.
506 * @pkt_len:		length of all segments
507 * @num_of_coalesced_segs: count of segments
508 *
509 * Approximate value of the MSS for this aggregation calculated using
510 * the first packet of it.
511 * Compute number of aggregated segments, and gso_type.
512 */
513static void bnx2x_set_gro_params(struct sk_buff *skb, u16 parsing_flags,
514				 u16 len_on_bd, unsigned int pkt_len,
515				 u16 num_of_coalesced_segs)
516{
517	/* TPA aggregation won't have either IP options or TCP options
518	 * other than timestamp or IPv6 extension headers.
519	 */
520	u16 hdrs_len = ETH_HLEN + sizeof(struct tcphdr);
521
522	if (GET_FLAG(parsing_flags, PARSING_FLAGS_OVER_ETHERNET_PROTOCOL) ==
523	    PRS_FLAG_OVERETH_IPV6) {
524		hdrs_len += sizeof(struct ipv6hdr);
525		skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
526	} else {
527		hdrs_len += sizeof(struct iphdr);
528		skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
529	}
530
531	/* Check if there was a TCP timestamp, if there is it's will
532	 * always be 12 bytes length: nop nop kind length echo val.
533	 *
534	 * Otherwise FW would close the aggregation.
535	 */
536	if (parsing_flags & PARSING_FLAGS_TIME_STAMP_EXIST_FLAG)
537		hdrs_len += TPA_TSTAMP_OPT_LEN;
538
539	skb_shinfo(skb)->gso_size = len_on_bd - hdrs_len;
540
541	/* tcp_gro_complete() will copy NAPI_GRO_CB(skb)->count
542	 * to skb_shinfo(skb)->gso_segs
543	 */
544	NAPI_GRO_CB(skb)->count = num_of_coalesced_segs;
545}
546
547static int bnx2x_alloc_rx_sge(struct bnx2x *bp, struct bnx2x_fastpath *fp,
548			      u16 index, gfp_t gfp_mask)
549{
550	struct sw_rx_page *sw_buf = &fp->rx_page_ring[index];
551	struct eth_rx_sge *sge = &fp->rx_sge_ring[index];
552	struct bnx2x_alloc_pool *pool = &fp->page_pool;
553	dma_addr_t mapping;
554
555	if (!pool->page) {
556		pool->page = alloc_pages(gfp_mask, PAGES_PER_SGE_SHIFT);
557		if (unlikely(!pool->page))
558			return -ENOMEM;
559
560		pool->offset = 0;
561	}
562
563	mapping = dma_map_page(&bp->pdev->dev, pool->page,
564			       pool->offset, SGE_PAGE_SIZE, DMA_FROM_DEVICE);
565	if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
566		BNX2X_ERR("Can't map sge\n");
567		return -ENOMEM;
568	}
569
570	sw_buf->page = pool->page;
571	sw_buf->offset = pool->offset;
572
573	dma_unmap_addr_set(sw_buf, mapping, mapping);
574
575	sge->addr_hi = cpu_to_le32(U64_HI(mapping));
576	sge->addr_lo = cpu_to_le32(U64_LO(mapping));
577
578	pool->offset += SGE_PAGE_SIZE;
579	if (PAGE_SIZE - pool->offset >= SGE_PAGE_SIZE)
580		get_page(pool->page);
581	else
582		pool->page = NULL;
583	return 0;
584}
585
586static int bnx2x_fill_frag_skb(struct bnx2x *bp, struct bnx2x_fastpath *fp,
587			       struct bnx2x_agg_info *tpa_info,
588			       u16 pages,
589			       struct sk_buff *skb,
590			       struct eth_end_agg_rx_cqe *cqe,
591			       u16 cqe_idx)
592{
593	struct sw_rx_page *rx_pg, old_rx_pg;
594	u32 i, frag_len, frag_size;
595	int err, j, frag_id = 0;
596	u16 len_on_bd = tpa_info->len_on_bd;
597	u16 full_page = 0, gro_size = 0;
598
599	frag_size = le16_to_cpu(cqe->pkt_len) - len_on_bd;
600
601	if (fp->mode == TPA_MODE_GRO) {
602		gro_size = tpa_info->gro_size;
603		full_page = tpa_info->full_page;
604	}
605
606	/* This is needed in order to enable forwarding support */
607	if (frag_size)
608		bnx2x_set_gro_params(skb, tpa_info->parsing_flags, len_on_bd,
609				     le16_to_cpu(cqe->pkt_len),
610				     le16_to_cpu(cqe->num_of_coalesced_segs));
611
612#ifdef BNX2X_STOP_ON_ERROR
613	if (pages > min_t(u32, 8, MAX_SKB_FRAGS) * SGE_PAGES) {
614		BNX2X_ERR("SGL length is too long: %d. CQE index is %d\n",
615			  pages, cqe_idx);
616		BNX2X_ERR("cqe->pkt_len = %d\n", cqe->pkt_len);
617		bnx2x_panic();
618		return -EINVAL;
619	}
620#endif
621
622	/* Run through the SGL and compose the fragmented skb */
623	for (i = 0, j = 0; i < pages; i += PAGES_PER_SGE, j++) {
624		u16 sge_idx = RX_SGE(le16_to_cpu(cqe->sgl_or_raw_data.sgl[j]));
625
626		/* FW gives the indices of the SGE as if the ring is an array
627		   (meaning that "next" element will consume 2 indices) */
628		if (fp->mode == TPA_MODE_GRO)
629			frag_len = min_t(u32, frag_size, (u32)full_page);
630		else /* LRO */
631			frag_len = min_t(u32, frag_size, (u32)SGE_PAGES);
632
633		rx_pg = &fp->rx_page_ring[sge_idx];
634		old_rx_pg = *rx_pg;
635
636		/* If we fail to allocate a substitute page, we simply stop
637		   where we are and drop the whole packet */
638		err = bnx2x_alloc_rx_sge(bp, fp, sge_idx, GFP_ATOMIC);
639		if (unlikely(err)) {
640			bnx2x_fp_qstats(bp, fp)->rx_skb_alloc_failed++;
641			return err;
642		}
643
644		dma_unmap_page(&bp->pdev->dev,
645			       dma_unmap_addr(&old_rx_pg, mapping),
646			       SGE_PAGE_SIZE, DMA_FROM_DEVICE);
647		/* Add one frag and update the appropriate fields in the skb */
648		if (fp->mode == TPA_MODE_LRO)
649			skb_fill_page_desc(skb, j, old_rx_pg.page,
650					   old_rx_pg.offset, frag_len);
651		else { /* GRO */
652			int rem;
653			int offset = 0;
654			for (rem = frag_len; rem > 0; rem -= gro_size) {
655				int len = rem > gro_size ? gro_size : rem;
656				skb_fill_page_desc(skb, frag_id++,
657						   old_rx_pg.page,
658						   old_rx_pg.offset + offset,
659						   len);
660				if (offset)
661					get_page(old_rx_pg.page);
662				offset += len;
663			}
664		}
665
666		skb->data_len += frag_len;
667		skb->truesize += SGE_PAGES;
668		skb->len += frag_len;
669
670		frag_size -= frag_len;
671	}
672
673	return 0;
674}
675
676static void bnx2x_frag_free(const struct bnx2x_fastpath *fp, void *data)
677{
678	if (fp->rx_frag_size)
679		skb_free_frag(data);
680	else
681		kfree(data);
682}
683
684static void *bnx2x_frag_alloc(const struct bnx2x_fastpath *fp, gfp_t gfp_mask)
685{
686	if (fp->rx_frag_size) {
687		/* GFP_KERNEL allocations are used only during initialization */
688		if (unlikely(gfpflags_allow_blocking(gfp_mask)))
689			return (void *)__get_free_page(gfp_mask);
690
691		return napi_alloc_frag(fp->rx_frag_size);
692	}
693
694	return kmalloc(fp->rx_buf_size + NET_SKB_PAD, gfp_mask);
695}
696
697#ifdef CONFIG_INET
698static void bnx2x_gro_ip_csum(struct bnx2x *bp, struct sk_buff *skb)
699{
700	const struct iphdr *iph = ip_hdr(skb);
701	struct tcphdr *th;
702
703	skb_set_transport_header(skb, sizeof(struct iphdr));
704	th = tcp_hdr(skb);
705
706	th->check = ~tcp_v4_check(skb->len - skb_transport_offset(skb),
707				  iph->saddr, iph->daddr, 0);
708}
709
710static void bnx2x_gro_ipv6_csum(struct bnx2x *bp, struct sk_buff *skb)
711{
712	struct ipv6hdr *iph = ipv6_hdr(skb);
713	struct tcphdr *th;
714
715	skb_set_transport_header(skb, sizeof(struct ipv6hdr));
716	th = tcp_hdr(skb);
717
718	th->check = ~tcp_v6_check(skb->len - skb_transport_offset(skb),
719				  &iph->saddr, &iph->daddr, 0);
720}
721
722static void bnx2x_gro_csum(struct bnx2x *bp, struct sk_buff *skb,
723			    void (*gro_func)(struct bnx2x*, struct sk_buff*))
724{
725	skb_reset_network_header(skb);
726	gro_func(bp, skb);
727	tcp_gro_complete(skb);
728}
729#endif
730
731static void bnx2x_gro_receive(struct bnx2x *bp, struct bnx2x_fastpath *fp,
732			       struct sk_buff *skb)
733{
734#ifdef CONFIG_INET
735	if (skb_shinfo(skb)->gso_size) {
736		switch (be16_to_cpu(skb->protocol)) {
737		case ETH_P_IP:
738			bnx2x_gro_csum(bp, skb, bnx2x_gro_ip_csum);
739			break;
740		case ETH_P_IPV6:
741			bnx2x_gro_csum(bp, skb, bnx2x_gro_ipv6_csum);
742			break;
743		default:
744			netdev_WARN_ONCE(bp->dev,
745					 "Error: FW GRO supports only IPv4/IPv6, not 0x%04x\n",
746					 be16_to_cpu(skb->protocol));
747		}
748	}
749#endif
750	skb_record_rx_queue(skb, fp->rx_queue);
751	napi_gro_receive(&fp->napi, skb);
752}
753
754static void bnx2x_tpa_stop(struct bnx2x *bp, struct bnx2x_fastpath *fp,
755			   struct bnx2x_agg_info *tpa_info,
756			   u16 pages,
757			   struct eth_end_agg_rx_cqe *cqe,
758			   u16 cqe_idx)
759{
760	struct sw_rx_bd *rx_buf = &tpa_info->first_buf;
761	u8 pad = tpa_info->placement_offset;
762	u16 len = tpa_info->len_on_bd;
763	struct sk_buff *skb = NULL;
764	u8 *new_data, *data = rx_buf->data;
765	u8 old_tpa_state = tpa_info->tpa_state;
766
767	tpa_info->tpa_state = BNX2X_TPA_STOP;
768
769	/* If we there was an error during the handling of the TPA_START -
770	 * drop this aggregation.
771	 */
772	if (old_tpa_state == BNX2X_TPA_ERROR)
773		goto drop;
774
775	/* Try to allocate the new data */
776	new_data = bnx2x_frag_alloc(fp, GFP_ATOMIC);
777	/* Unmap skb in the pool anyway, as we are going to change
778	   pool entry status to BNX2X_TPA_STOP even if new skb allocation
779	   fails. */
780	dma_unmap_single(&bp->pdev->dev, dma_unmap_addr(rx_buf, mapping),
781			 fp->rx_buf_size, DMA_FROM_DEVICE);
782	if (likely(new_data))
783		skb = build_skb(data, fp->rx_frag_size);
784
785	if (likely(skb)) {
786#ifdef BNX2X_STOP_ON_ERROR
787		if (pad + len > fp->rx_buf_size) {
788			BNX2X_ERR("skb_put is about to fail...  pad %d  len %d  rx_buf_size %d\n",
789				  pad, len, fp->rx_buf_size);
790			bnx2x_panic();
791			bnx2x_frag_free(fp, new_data);
792			return;
793		}
794#endif
795
796		skb_reserve(skb, pad + NET_SKB_PAD);
797		skb_put(skb, len);
798		skb_set_hash(skb, tpa_info->rxhash, tpa_info->rxhash_type);
799
800		skb->protocol = eth_type_trans(skb, bp->dev);
801		skb->ip_summed = CHECKSUM_UNNECESSARY;
802
803		if (!bnx2x_fill_frag_skb(bp, fp, tpa_info, pages,
804					 skb, cqe, cqe_idx)) {
805			if (tpa_info->parsing_flags & PARSING_FLAGS_VLAN)
806				__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), tpa_info->vlan_tag);
807			bnx2x_gro_receive(bp, fp, skb);
808		} else {
809			DP(NETIF_MSG_RX_STATUS,
810			   "Failed to allocate new pages - dropping packet!\n");
811			dev_kfree_skb_any(skb);
812		}
813
814		/* put new data in bin */
815		rx_buf->data = new_data;
816
817		return;
818	}
819	if (new_data)
820		bnx2x_frag_free(fp, new_data);
821drop:
822	/* drop the packet and keep the buffer in the bin */
823	DP(NETIF_MSG_RX_STATUS,
824	   "Failed to allocate or map a new skb - dropping packet!\n");
825	bnx2x_fp_stats(bp, fp)->eth_q_stats.rx_skb_alloc_failed++;
826}
827
828static int bnx2x_alloc_rx_data(struct bnx2x *bp, struct bnx2x_fastpath *fp,
829			       u16 index, gfp_t gfp_mask)
830{
831	u8 *data;
832	struct sw_rx_bd *rx_buf = &fp->rx_buf_ring[index];
833	struct eth_rx_bd *rx_bd = &fp->rx_desc_ring[index];
834	dma_addr_t mapping;
835
836	data = bnx2x_frag_alloc(fp, gfp_mask);
837	if (unlikely(data == NULL))
838		return -ENOMEM;
839
840	mapping = dma_map_single(&bp->pdev->dev, data + NET_SKB_PAD,
841				 fp->rx_buf_size,
842				 DMA_FROM_DEVICE);
843	if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
844		bnx2x_frag_free(fp, data);
845		BNX2X_ERR("Can't map rx data\n");
846		return -ENOMEM;
847	}
848
849	rx_buf->data = data;
850	dma_unmap_addr_set(rx_buf, mapping, mapping);
851
852	rx_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
853	rx_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
854
855	return 0;
856}
857
858static
859void bnx2x_csum_validate(struct sk_buff *skb, union eth_rx_cqe *cqe,
860				 struct bnx2x_fastpath *fp,
861				 struct bnx2x_eth_q_stats *qstats)
862{
863	/* Do nothing if no L4 csum validation was done.
864	 * We do not check whether IP csum was validated. For IPv4 we assume
865	 * that if the card got as far as validating the L4 csum, it also
866	 * validated the IP csum. IPv6 has no IP csum.
867	 */
868	if (cqe->fast_path_cqe.status_flags &
869	    ETH_FAST_PATH_RX_CQE_L4_XSUM_NO_VALIDATION_FLG)
870		return;
871
872	/* If L4 validation was done, check if an error was found. */
873
874	if (cqe->fast_path_cqe.type_error_flags &
875	    (ETH_FAST_PATH_RX_CQE_IP_BAD_XSUM_FLG |
876	     ETH_FAST_PATH_RX_CQE_L4_BAD_XSUM_FLG))
877		qstats->hw_csum_err++;
878	else
879		skb->ip_summed = CHECKSUM_UNNECESSARY;
880}
881
882static int bnx2x_rx_int(struct bnx2x_fastpath *fp, int budget)
883{
884	struct bnx2x *bp = fp->bp;
885	u16 bd_cons, bd_prod, bd_prod_fw, comp_ring_cons;
886	u16 sw_comp_cons, sw_comp_prod;
887	int rx_pkt = 0;
888	union eth_rx_cqe *cqe;
889	struct eth_fast_path_rx_cqe *cqe_fp;
890
891#ifdef BNX2X_STOP_ON_ERROR
892	if (unlikely(bp->panic))
893		return 0;
894#endif
895	if (budget <= 0)
896		return rx_pkt;
897
898	bd_cons = fp->rx_bd_cons;
899	bd_prod = fp->rx_bd_prod;
900	bd_prod_fw = bd_prod;
901	sw_comp_cons = fp->rx_comp_cons;
902	sw_comp_prod = fp->rx_comp_prod;
903
904	comp_ring_cons = RCQ_BD(sw_comp_cons);
905	cqe = &fp->rx_comp_ring[comp_ring_cons];
906	cqe_fp = &cqe->fast_path_cqe;
907
908	DP(NETIF_MSG_RX_STATUS,
909	   "queue[%d]: sw_comp_cons %u\n", fp->index, sw_comp_cons);
910
911	while (BNX2X_IS_CQE_COMPLETED(cqe_fp)) {
912		struct sw_rx_bd *rx_buf = NULL;
913		struct sk_buff *skb;
914		u8 cqe_fp_flags;
915		enum eth_rx_cqe_type cqe_fp_type;
916		u16 len, pad, queue;
917		u8 *data;
918		u32 rxhash;
919		enum pkt_hash_types rxhash_type;
920
921#ifdef BNX2X_STOP_ON_ERROR
922		if (unlikely(bp->panic))
923			return 0;
924#endif
925
926		bd_prod = RX_BD(bd_prod);
927		bd_cons = RX_BD(bd_cons);
928
929		/* A rmb() is required to ensure that the CQE is not read
930		 * before it is written by the adapter DMA.  PCI ordering
931		 * rules will make sure the other fields are written before
932		 * the marker at the end of struct eth_fast_path_rx_cqe
933		 * but without rmb() a weakly ordered processor can process
934		 * stale data.  Without the barrier TPA state-machine might
935		 * enter inconsistent state and kernel stack might be
936		 * provided with incorrect packet description - these lead
937		 * to various kernel crashed.
938		 */
939		rmb();
940
941		cqe_fp_flags = cqe_fp->type_error_flags;
942		cqe_fp_type = cqe_fp_flags & ETH_FAST_PATH_RX_CQE_TYPE;
943
944		DP(NETIF_MSG_RX_STATUS,
945		   "CQE type %x  err %x  status %x  queue %x  vlan %x  len %u\n",
946		   CQE_TYPE(cqe_fp_flags),
947		   cqe_fp_flags, cqe_fp->status_flags,
948		   le32_to_cpu(cqe_fp->rss_hash_result),
949		   le16_to_cpu(cqe_fp->vlan_tag),
950		   le16_to_cpu(cqe_fp->pkt_len_or_gro_seg_len));
951
952		/* is this a slowpath msg? */
953		if (unlikely(CQE_TYPE_SLOW(cqe_fp_type))) {
954			bnx2x_sp_event(fp, cqe);
955			goto next_cqe;
956		}
957
958		rx_buf = &fp->rx_buf_ring[bd_cons];
959		data = rx_buf->data;
960
961		if (!CQE_TYPE_FAST(cqe_fp_type)) {
962			struct bnx2x_agg_info *tpa_info;
963			u16 frag_size, pages;
964#ifdef BNX2X_STOP_ON_ERROR
965			/* sanity check */
966			if (fp->mode == TPA_MODE_DISABLED &&
967			    (CQE_TYPE_START(cqe_fp_type) ||
968			     CQE_TYPE_STOP(cqe_fp_type)))
969				BNX2X_ERR("START/STOP packet while TPA disabled, type %x\n",
970					  CQE_TYPE(cqe_fp_type));
971#endif
972
973			if (CQE_TYPE_START(cqe_fp_type)) {
974				u16 queue = cqe_fp->queue_index;
975				DP(NETIF_MSG_RX_STATUS,
976				   "calling tpa_start on queue %d\n",
977				   queue);
978
979				bnx2x_tpa_start(fp, queue,
980						bd_cons, bd_prod,
981						cqe_fp);
982
983				goto next_rx;
984			}
985			queue = cqe->end_agg_cqe.queue_index;
986			tpa_info = &fp->tpa_info[queue];
987			DP(NETIF_MSG_RX_STATUS,
988			   "calling tpa_stop on queue %d\n",
989			   queue);
990
991			frag_size = le16_to_cpu(cqe->end_agg_cqe.pkt_len) -
992				    tpa_info->len_on_bd;
993
994			if (fp->mode == TPA_MODE_GRO)
995				pages = (frag_size + tpa_info->full_page - 1) /
996					 tpa_info->full_page;
997			else
998				pages = SGE_PAGE_ALIGN(frag_size) >>
999					SGE_PAGE_SHIFT;
1000
1001			bnx2x_tpa_stop(bp, fp, tpa_info, pages,
1002				       &cqe->end_agg_cqe, comp_ring_cons);
1003#ifdef BNX2X_STOP_ON_ERROR
1004			if (bp->panic)
1005				return 0;
1006#endif
1007
1008			bnx2x_update_sge_prod(fp, pages, &cqe->end_agg_cqe);
1009			goto next_cqe;
1010		}
1011		/* non TPA */
1012		len = le16_to_cpu(cqe_fp->pkt_len_or_gro_seg_len);
1013		pad = cqe_fp->placement_offset;
1014		dma_sync_single_for_cpu(&bp->pdev->dev,
1015					dma_unmap_addr(rx_buf, mapping),
1016					pad + RX_COPY_THRESH,
1017					DMA_FROM_DEVICE);
1018		pad += NET_SKB_PAD;
1019		prefetch(data + pad); /* speedup eth_type_trans() */
1020		/* is this an error packet? */
1021		if (unlikely(cqe_fp_flags & ETH_RX_ERROR_FALGS)) {
1022			DP(NETIF_MSG_RX_ERR | NETIF_MSG_RX_STATUS,
1023			   "ERROR  flags %x  rx packet %u\n",
1024			   cqe_fp_flags, sw_comp_cons);
1025			bnx2x_fp_qstats(bp, fp)->rx_err_discard_pkt++;
1026			goto reuse_rx;
1027		}
1028
1029		/* Since we don't have a jumbo ring
1030		 * copy small packets if mtu > 1500
1031		 */
1032		if ((bp->dev->mtu > ETH_MAX_PACKET_SIZE) &&
1033		    (len <= RX_COPY_THRESH)) {
1034			skb = napi_alloc_skb(&fp->napi, len);
1035			if (skb == NULL) {
1036				DP(NETIF_MSG_RX_ERR | NETIF_MSG_RX_STATUS,
1037				   "ERROR  packet dropped because of alloc failure\n");
1038				bnx2x_fp_qstats(bp, fp)->rx_skb_alloc_failed++;
1039				goto reuse_rx;
1040			}
1041			memcpy(skb->data, data + pad, len);
1042			bnx2x_reuse_rx_data(fp, bd_cons, bd_prod);
1043		} else {
1044			if (likely(bnx2x_alloc_rx_data(bp, fp, bd_prod,
1045						       GFP_ATOMIC) == 0)) {
1046				dma_unmap_single(&bp->pdev->dev,
1047						 dma_unmap_addr(rx_buf, mapping),
1048						 fp->rx_buf_size,
1049						 DMA_FROM_DEVICE);
1050				skb = build_skb(data, fp->rx_frag_size);
1051				if (unlikely(!skb)) {
1052					bnx2x_frag_free(fp, data);
1053					bnx2x_fp_qstats(bp, fp)->
1054							rx_skb_alloc_failed++;
1055					goto next_rx;
1056				}
1057				skb_reserve(skb, pad);
1058			} else {
1059				DP(NETIF_MSG_RX_ERR | NETIF_MSG_RX_STATUS,
1060				   "ERROR  packet dropped because of alloc failure\n");
1061				bnx2x_fp_qstats(bp, fp)->rx_skb_alloc_failed++;
1062reuse_rx:
1063				bnx2x_reuse_rx_data(fp, bd_cons, bd_prod);
1064				goto next_rx;
1065			}
1066		}
1067
1068		skb_put(skb, len);
1069		skb->protocol = eth_type_trans(skb, bp->dev);
1070
1071		/* Set Toeplitz hash for a none-LRO skb */
1072		rxhash = bnx2x_get_rxhash(bp, cqe_fp, &rxhash_type);
1073		skb_set_hash(skb, rxhash, rxhash_type);
1074
1075		skb_checksum_none_assert(skb);
1076
1077		if (bp->dev->features & NETIF_F_RXCSUM)
1078			bnx2x_csum_validate(skb, cqe, fp,
1079					    bnx2x_fp_qstats(bp, fp));
1080
1081		skb_record_rx_queue(skb, fp->rx_queue);
1082
1083		/* Check if this packet was timestamped */
1084		if (unlikely(cqe->fast_path_cqe.type_error_flags &
1085			     (1 << ETH_FAST_PATH_RX_CQE_PTP_PKT_SHIFT)))
1086			bnx2x_set_rx_ts(bp, skb);
1087
1088		if (le16_to_cpu(cqe_fp->pars_flags.flags) &
1089		    PARSING_FLAGS_VLAN)
1090			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
1091					       le16_to_cpu(cqe_fp->vlan_tag));
1092
1093		napi_gro_receive(&fp->napi, skb);
1094next_rx:
1095		rx_buf->data = NULL;
1096
1097		bd_cons = NEXT_RX_IDX(bd_cons);
1098		bd_prod = NEXT_RX_IDX(bd_prod);
1099		bd_prod_fw = NEXT_RX_IDX(bd_prod_fw);
1100		rx_pkt++;
1101next_cqe:
1102		sw_comp_prod = NEXT_RCQ_IDX(sw_comp_prod);
1103		sw_comp_cons = NEXT_RCQ_IDX(sw_comp_cons);
1104
1105		/* mark CQE as free */
1106		BNX2X_SEED_CQE(cqe_fp);
1107
1108		if (rx_pkt == budget)
1109			break;
1110
1111		comp_ring_cons = RCQ_BD(sw_comp_cons);
1112		cqe = &fp->rx_comp_ring[comp_ring_cons];
1113		cqe_fp = &cqe->fast_path_cqe;
1114	} /* while */
1115
1116	fp->rx_bd_cons = bd_cons;
1117	fp->rx_bd_prod = bd_prod_fw;
1118	fp->rx_comp_cons = sw_comp_cons;
1119	fp->rx_comp_prod = sw_comp_prod;
1120
1121	/* Update producers */
1122	bnx2x_update_rx_prod(bp, fp, bd_prod_fw, sw_comp_prod,
1123			     fp->rx_sge_prod);
1124
1125	return rx_pkt;
1126}
1127
1128static irqreturn_t bnx2x_msix_fp_int(int irq, void *fp_cookie)
1129{
1130	struct bnx2x_fastpath *fp = fp_cookie;
1131	struct bnx2x *bp = fp->bp;
1132	u8 cos;
1133
1134	DP(NETIF_MSG_INTR,
1135	   "got an MSI-X interrupt on IDX:SB [fp %d fw_sd %d igusb %d]\n",
1136	   fp->index, fp->fw_sb_id, fp->igu_sb_id);
1137
1138	bnx2x_ack_sb(bp, fp->igu_sb_id, USTORM_ID, 0, IGU_INT_DISABLE, 0);
1139
1140#ifdef BNX2X_STOP_ON_ERROR
1141	if (unlikely(bp->panic))
1142		return IRQ_HANDLED;
1143#endif
1144
1145	/* Handle Rx and Tx according to MSI-X vector */
1146	for_each_cos_in_tx_queue(fp, cos)
1147		prefetch(fp->txdata_ptr[cos]->tx_cons_sb);
1148
1149	prefetch(&fp->sb_running_index[SM_RX_ID]);
1150	napi_schedule_irqoff(&bnx2x_fp(bp, fp->index, napi));
1151
1152	return IRQ_HANDLED;
1153}
1154
1155/* HW Lock for shared dual port PHYs */
1156void bnx2x_acquire_phy_lock(struct bnx2x *bp)
1157{
1158	mutex_lock(&bp->port.phy_mutex);
1159
1160	bnx2x_acquire_hw_lock(bp, HW_LOCK_RESOURCE_MDIO);
1161}
1162
1163void bnx2x_release_phy_lock(struct bnx2x *bp)
1164{
1165	bnx2x_release_hw_lock(bp, HW_LOCK_RESOURCE_MDIO);
1166
1167	mutex_unlock(&bp->port.phy_mutex);
1168}
1169
1170/* calculates MF speed according to current linespeed and MF configuration */
1171u16 bnx2x_get_mf_speed(struct bnx2x *bp)
1172{
1173	u16 line_speed = bp->link_vars.line_speed;
1174	if (IS_MF(bp)) {
1175		u16 maxCfg = bnx2x_extract_max_cfg(bp,
1176						   bp->mf_config[BP_VN(bp)]);
1177
1178		/* Calculate the current MAX line speed limit for the MF
1179		 * devices
1180		 */
1181		if (IS_MF_PERCENT_BW(bp))
1182			line_speed = (line_speed * maxCfg) / 100;
1183		else { /* SD mode */
1184			u16 vn_max_rate = maxCfg * 100;
1185
1186			if (vn_max_rate < line_speed)
1187				line_speed = vn_max_rate;
1188		}
1189	}
1190
1191	return line_speed;
1192}
1193
1194/**
1195 * bnx2x_fill_report_data - fill link report data to report
1196 *
1197 * @bp:		driver handle
1198 * @data:	link state to update
1199 *
1200 * It uses a none-atomic bit operations because is called under the mutex.
1201 */
1202static void bnx2x_fill_report_data(struct bnx2x *bp,
1203				   struct bnx2x_link_report_data *data)
1204{
1205	memset(data, 0, sizeof(*data));
1206
1207	if (IS_PF(bp)) {
1208		/* Fill the report data: effective line speed */
1209		data->line_speed = bnx2x_get_mf_speed(bp);
1210
1211		/* Link is down */
1212		if (!bp->link_vars.link_up || (bp->flags & MF_FUNC_DIS))
1213			__set_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1214				  &data->link_report_flags);
1215
1216		if (!BNX2X_NUM_ETH_QUEUES(bp))
1217			__set_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1218				  &data->link_report_flags);
1219
1220		/* Full DUPLEX */
1221		if (bp->link_vars.duplex == DUPLEX_FULL)
1222			__set_bit(BNX2X_LINK_REPORT_FD,
1223				  &data->link_report_flags);
1224
1225		/* Rx Flow Control is ON */
1226		if (bp->link_vars.flow_ctrl & BNX2X_FLOW_CTRL_RX)
1227			__set_bit(BNX2X_LINK_REPORT_RX_FC_ON,
1228				  &data->link_report_flags);
1229
1230		/* Tx Flow Control is ON */
1231		if (bp->link_vars.flow_ctrl & BNX2X_FLOW_CTRL_TX)
1232			__set_bit(BNX2X_LINK_REPORT_TX_FC_ON,
1233				  &data->link_report_flags);
1234	} else { /* VF */
1235		*data = bp->vf_link_vars;
1236	}
1237}
1238
1239/**
1240 * bnx2x_link_report - report link status to OS.
1241 *
1242 * @bp:		driver handle
1243 *
1244 * Calls the __bnx2x_link_report() under the same locking scheme
1245 * as a link/PHY state managing code to ensure a consistent link
1246 * reporting.
1247 */
1248
1249void bnx2x_link_report(struct bnx2x *bp)
1250{
1251	bnx2x_acquire_phy_lock(bp);
1252	__bnx2x_link_report(bp);
1253	bnx2x_release_phy_lock(bp);
1254}
1255
1256/**
1257 * __bnx2x_link_report - report link status to OS.
1258 *
1259 * @bp:		driver handle
1260 *
1261 * None atomic implementation.
1262 * Should be called under the phy_lock.
1263 */
1264void __bnx2x_link_report(struct bnx2x *bp)
1265{
1266	struct bnx2x_link_report_data cur_data;
1267
1268	if (bp->force_link_down) {
1269		bp->link_vars.link_up = 0;
1270		return;
1271	}
1272
1273	/* reread mf_cfg */
1274	if (IS_PF(bp) && !CHIP_IS_E1(bp))
1275		bnx2x_read_mf_cfg(bp);
1276
1277	/* Read the current link report info */
1278	bnx2x_fill_report_data(bp, &cur_data);
1279
1280	/* Don't report link down or exactly the same link status twice */
1281	if (!memcmp(&cur_data, &bp->last_reported_link, sizeof(cur_data)) ||
1282	    (test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1283		      &bp->last_reported_link.link_report_flags) &&
1284	     test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1285		      &cur_data.link_report_flags)))
1286		return;
1287
1288	bp->link_cnt++;
1289
1290	/* We are going to report a new link parameters now -
1291	 * remember the current data for the next time.
1292	 */
1293	memcpy(&bp->last_reported_link, &cur_data, sizeof(cur_data));
1294
1295	/* propagate status to VFs */
1296	if (IS_PF(bp))
1297		bnx2x_iov_link_update(bp);
1298
1299	if (test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1300		     &cur_data.link_report_flags)) {
1301		netif_carrier_off(bp->dev);
1302		netdev_err(bp->dev, "NIC Link is Down\n");
1303		return;
1304	} else {
1305		const char *duplex;
1306		const char *flow;
1307
1308		netif_carrier_on(bp->dev);
1309
1310		if (test_and_clear_bit(BNX2X_LINK_REPORT_FD,
1311				       &cur_data.link_report_flags))
1312			duplex = "full";
1313		else
1314			duplex = "half";
1315
1316		/* Handle the FC at the end so that only these flags would be
1317		 * possibly set. This way we may easily check if there is no FC
1318		 * enabled.
1319		 */
1320		if (cur_data.link_report_flags) {
1321			if (test_bit(BNX2X_LINK_REPORT_RX_FC_ON,
1322				     &cur_data.link_report_flags)) {
1323				if (test_bit(BNX2X_LINK_REPORT_TX_FC_ON,
1324				     &cur_data.link_report_flags))
1325					flow = "ON - receive & transmit";
1326				else
1327					flow = "ON - receive";
1328			} else {
1329				flow = "ON - transmit";
1330			}
1331		} else {
1332			flow = "none";
1333		}
1334		netdev_info(bp->dev, "NIC Link is Up, %d Mbps %s duplex, Flow control: %s\n",
1335			    cur_data.line_speed, duplex, flow);
1336	}
1337}
1338
1339static void bnx2x_set_next_page_sgl(struct bnx2x_fastpath *fp)
1340{
1341	int i;
1342
1343	for (i = 1; i <= NUM_RX_SGE_PAGES; i++) {
1344		struct eth_rx_sge *sge;
1345
1346		sge = &fp->rx_sge_ring[RX_SGE_CNT * i - 2];
1347		sge->addr_hi =
1348			cpu_to_le32(U64_HI(fp->rx_sge_mapping +
1349			BCM_PAGE_SIZE*(i % NUM_RX_SGE_PAGES)));
1350
1351		sge->addr_lo =
1352			cpu_to_le32(U64_LO(fp->rx_sge_mapping +
1353			BCM_PAGE_SIZE*(i % NUM_RX_SGE_PAGES)));
1354	}
1355}
1356
1357static void bnx2x_free_tpa_pool(struct bnx2x *bp,
1358				struct bnx2x_fastpath *fp, int last)
1359{
1360	int i;
1361
1362	for (i = 0; i < last; i++) {
1363		struct bnx2x_agg_info *tpa_info = &fp->tpa_info[i];
1364		struct sw_rx_bd *first_buf = &tpa_info->first_buf;
1365		u8 *data = first_buf->data;
1366
1367		if (data == NULL) {
1368			DP(NETIF_MSG_IFDOWN, "tpa bin %d empty on free\n", i);
1369			continue;
1370		}
1371		if (tpa_info->tpa_state == BNX2X_TPA_START)
1372			dma_unmap_single(&bp->pdev->dev,
1373					 dma_unmap_addr(first_buf, mapping),
1374					 fp->rx_buf_size, DMA_FROM_DEVICE);
1375		bnx2x_frag_free(fp, data);
1376		first_buf->data = NULL;
1377	}
1378}
1379
1380void bnx2x_init_rx_rings_cnic(struct bnx2x *bp)
1381{
1382	int j;
1383
1384	for_each_rx_queue_cnic(bp, j) {
1385		struct bnx2x_fastpath *fp = &bp->fp[j];
1386
1387		fp->rx_bd_cons = 0;
1388
1389		/* Activate BD ring */
1390		/* Warning!
1391		 * this will generate an interrupt (to the TSTORM)
1392		 * must only be done after chip is initialized
1393		 */
1394		bnx2x_update_rx_prod(bp, fp, fp->rx_bd_prod, fp->rx_comp_prod,
1395				     fp->rx_sge_prod);
1396	}
1397}
1398
1399void bnx2x_init_rx_rings(struct bnx2x *bp)
1400{
1401	int func = BP_FUNC(bp);
1402	u16 ring_prod;
1403	int i, j;
1404
1405	/* Allocate TPA resources */
1406	for_each_eth_queue(bp, j) {
1407		struct bnx2x_fastpath *fp = &bp->fp[j];
1408
1409		DP(NETIF_MSG_IFUP,
1410		   "mtu %d  rx_buf_size %d\n", bp->dev->mtu, fp->rx_buf_size);
1411
1412		if (fp->mode != TPA_MODE_DISABLED) {
1413			/* Fill the per-aggregation pool */
1414			for (i = 0; i < MAX_AGG_QS(bp); i++) {
1415				struct bnx2x_agg_info *tpa_info =
1416					&fp->tpa_info[i];
1417				struct sw_rx_bd *first_buf =
1418					&tpa_info->first_buf;
1419
1420				first_buf->data =
1421					bnx2x_frag_alloc(fp, GFP_KERNEL);
1422				if (!first_buf->data) {
1423					BNX2X_ERR("Failed to allocate TPA skb pool for queue[%d] - disabling TPA on this queue!\n",
1424						  j);
1425					bnx2x_free_tpa_pool(bp, fp, i);
1426					fp->mode = TPA_MODE_DISABLED;
1427					break;
1428				}
1429				dma_unmap_addr_set(first_buf, mapping, 0);
1430				tpa_info->tpa_state = BNX2X_TPA_STOP;
1431			}
1432
1433			/* "next page" elements initialization */
1434			bnx2x_set_next_page_sgl(fp);
1435
1436			/* set SGEs bit mask */
1437			bnx2x_init_sge_ring_bit_mask(fp);
1438
1439			/* Allocate SGEs and initialize the ring elements */
1440			for (i = 0, ring_prod = 0;
1441			     i < MAX_RX_SGE_CNT*NUM_RX_SGE_PAGES; i++) {
1442
1443				if (bnx2x_alloc_rx_sge(bp, fp, ring_prod,
1444						       GFP_KERNEL) < 0) {
1445					BNX2X_ERR("was only able to allocate %d rx sges\n",
1446						  i);
1447					BNX2X_ERR("disabling TPA for queue[%d]\n",
1448						  j);
1449					/* Cleanup already allocated elements */
1450					bnx2x_free_rx_sge_range(bp, fp,
1451								ring_prod);
1452					bnx2x_free_tpa_pool(bp, fp,
1453							    MAX_AGG_QS(bp));
1454					fp->mode = TPA_MODE_DISABLED;
1455					ring_prod = 0;
1456					break;
1457				}
1458				ring_prod = NEXT_SGE_IDX(ring_prod);
1459			}
1460
1461			fp->rx_sge_prod = ring_prod;
1462		}
1463	}
1464
1465	for_each_eth_queue(bp, j) {
1466		struct bnx2x_fastpath *fp = &bp->fp[j];
1467
1468		fp->rx_bd_cons = 0;
1469
1470		/* Activate BD ring */
1471		/* Warning!
1472		 * this will generate an interrupt (to the TSTORM)
1473		 * must only be done after chip is initialized
1474		 */
1475		bnx2x_update_rx_prod(bp, fp, fp->rx_bd_prod, fp->rx_comp_prod,
1476				     fp->rx_sge_prod);
1477
1478		if (j != 0)
1479			continue;
1480
1481		if (CHIP_IS_E1(bp)) {
1482			REG_WR(bp, BAR_USTRORM_INTMEM +
1483			       USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func),
1484			       U64_LO(fp->rx_comp_mapping));
1485			REG_WR(bp, BAR_USTRORM_INTMEM +
1486			       USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func) + 4,
1487			       U64_HI(fp->rx_comp_mapping));
1488		}
1489	}
1490}
1491
1492static void bnx2x_free_tx_skbs_queue(struct bnx2x_fastpath *fp)
1493{
1494	u8 cos;
1495	struct bnx2x *bp = fp->bp;
1496
1497	for_each_cos_in_tx_queue(fp, cos) {
1498		struct bnx2x_fp_txdata *txdata = fp->txdata_ptr[cos];
1499		unsigned pkts_compl = 0, bytes_compl = 0;
1500
1501		u16 sw_prod = txdata->tx_pkt_prod;
1502		u16 sw_cons = txdata->tx_pkt_cons;
1503
1504		while (sw_cons != sw_prod) {
1505			bnx2x_free_tx_pkt(bp, txdata, TX_BD(sw_cons),
1506					  &pkts_compl, &bytes_compl);
1507			sw_cons++;
1508		}
1509
1510		netdev_tx_reset_queue(
1511			netdev_get_tx_queue(bp->dev,
1512					    txdata->txq_index));
1513	}
1514}
1515
1516static void bnx2x_free_tx_skbs_cnic(struct bnx2x *bp)
1517{
1518	int i;
1519
1520	for_each_tx_queue_cnic(bp, i) {
1521		bnx2x_free_tx_skbs_queue(&bp->fp[i]);
1522	}
1523}
1524
1525static void bnx2x_free_tx_skbs(struct bnx2x *bp)
1526{
1527	int i;
1528
1529	for_each_eth_queue(bp, i) {
1530		bnx2x_free_tx_skbs_queue(&bp->fp[i]);
1531	}
1532}
1533
1534static void bnx2x_free_rx_bds(struct bnx2x_fastpath *fp)
1535{
1536	struct bnx2x *bp = fp->bp;
1537	int i;
1538
1539	/* ring wasn't allocated */
1540	if (fp->rx_buf_ring == NULL)
1541		return;
1542
1543	for (i = 0; i < NUM_RX_BD; i++) {
1544		struct sw_rx_bd *rx_buf = &fp->rx_buf_ring[i];
1545		u8 *data = rx_buf->data;
1546
1547		if (data == NULL)
1548			continue;
1549		dma_unmap_single(&bp->pdev->dev,
1550				 dma_unmap_addr(rx_buf, mapping),
1551				 fp->rx_buf_size, DMA_FROM_DEVICE);
1552
1553		rx_buf->data = NULL;
1554		bnx2x_frag_free(fp, data);
1555	}
1556}
1557
1558static void bnx2x_free_rx_skbs_cnic(struct bnx2x *bp)
1559{
1560	int j;
1561
1562	for_each_rx_queue_cnic(bp, j) {
1563		bnx2x_free_rx_bds(&bp->fp[j]);
1564	}
1565}
1566
1567static void bnx2x_free_rx_skbs(struct bnx2x *bp)
1568{
1569	int j;
1570
1571	for_each_eth_queue(bp, j) {
1572		struct bnx2x_fastpath *fp = &bp->fp[j];
1573
1574		bnx2x_free_rx_bds(fp);
1575
1576		if (fp->mode != TPA_MODE_DISABLED)
1577			bnx2x_free_tpa_pool(bp, fp, MAX_AGG_QS(bp));
1578	}
1579}
1580
1581static void bnx2x_free_skbs_cnic(struct bnx2x *bp)
1582{
1583	bnx2x_free_tx_skbs_cnic(bp);
1584	bnx2x_free_rx_skbs_cnic(bp);
1585}
1586
1587void bnx2x_free_skbs(struct bnx2x *bp)
1588{
1589	bnx2x_free_tx_skbs(bp);
1590	bnx2x_free_rx_skbs(bp);
1591}
1592
1593void bnx2x_update_max_mf_config(struct bnx2x *bp, u32 value)
1594{
1595	/* load old values */
1596	u32 mf_cfg = bp->mf_config[BP_VN(bp)];
1597
1598	if (value != bnx2x_extract_max_cfg(bp, mf_cfg)) {
1599		/* leave all but MAX value */
1600		mf_cfg &= ~FUNC_MF_CFG_MAX_BW_MASK;
1601
1602		/* set new MAX value */
1603		mf_cfg |= (value << FUNC_MF_CFG_MAX_BW_SHIFT)
1604				& FUNC_MF_CFG_MAX_BW_MASK;
1605
1606		bnx2x_fw_command(bp, DRV_MSG_CODE_SET_MF_BW, mf_cfg);
1607	}
1608}
1609
1610/**
1611 * bnx2x_free_msix_irqs - free previously requested MSI-X IRQ vectors
1612 *
1613 * @bp:		driver handle
1614 * @nvecs:	number of vectors to be released
1615 */
1616static void bnx2x_free_msix_irqs(struct bnx2x *bp, int nvecs)
1617{
1618	int i, offset = 0;
1619
1620	if (nvecs == offset)
1621		return;
1622
1623	/* VFs don't have a default SB */
1624	if (IS_PF(bp)) {
1625		free_irq(bp->msix_table[offset].vector, bp->dev);
1626		DP(NETIF_MSG_IFDOWN, "released sp irq (%d)\n",
1627		   bp->msix_table[offset].vector);
1628		offset++;
1629	}
1630
1631	if (CNIC_SUPPORT(bp)) {
1632		if (nvecs == offset)
1633			return;
1634		offset++;
1635	}
1636
1637	for_each_eth_queue(bp, i) {
1638		if (nvecs == offset)
1639			return;
1640		DP(NETIF_MSG_IFDOWN, "about to release fp #%d->%d irq\n",
1641		   i, bp->msix_table[offset].vector);
1642
1643		free_irq(bp->msix_table[offset++].vector, &bp->fp[i]);
1644	}
1645}
1646
1647void bnx2x_free_irq(struct bnx2x *bp)
1648{
1649	if (bp->flags & USING_MSIX_FLAG &&
1650	    !(bp->flags & USING_SINGLE_MSIX_FLAG)) {
1651		int nvecs = BNX2X_NUM_ETH_QUEUES(bp) + CNIC_SUPPORT(bp);
1652
1653		/* vfs don't have a default status block */
1654		if (IS_PF(bp))
1655			nvecs++;
1656
1657		bnx2x_free_msix_irqs(bp, nvecs);
1658	} else {
1659		free_irq(bp->dev->irq, bp->dev);
1660	}
1661}
1662
1663int bnx2x_enable_msix(struct bnx2x *bp)
1664{
1665	int msix_vec = 0, i, rc;
1666
1667	/* VFs don't have a default status block */
1668	if (IS_PF(bp)) {
1669		bp->msix_table[msix_vec].entry = msix_vec;
1670		BNX2X_DEV_INFO("msix_table[0].entry = %d (slowpath)\n",
1671			       bp->msix_table[0].entry);
1672		msix_vec++;
1673	}
1674
1675	/* Cnic requires an msix vector for itself */
1676	if (CNIC_SUPPORT(bp)) {
1677		bp->msix_table[msix_vec].entry = msix_vec;
1678		BNX2X_DEV_INFO("msix_table[%d].entry = %d (CNIC)\n",
1679			       msix_vec, bp->msix_table[msix_vec].entry);
1680		msix_vec++;
1681	}
1682
1683	/* We need separate vectors for ETH queues only (not FCoE) */
1684	for_each_eth_queue(bp, i) {
1685		bp->msix_table[msix_vec].entry = msix_vec;
1686		BNX2X_DEV_INFO("msix_table[%d].entry = %d (fastpath #%u)\n",
1687			       msix_vec, msix_vec, i);
1688		msix_vec++;
1689	}
1690
1691	DP(BNX2X_MSG_SP, "about to request enable msix with %d vectors\n",
1692	   msix_vec);
1693
1694	rc = pci_enable_msix_range(bp->pdev, &bp->msix_table[0],
1695				   BNX2X_MIN_MSIX_VEC_CNT(bp), msix_vec);
1696	/*
1697	 * reconfigure number of tx/rx queues according to available
1698	 * MSI-X vectors
1699	 */
1700	if (rc == -ENOSPC) {
1701		/* Get by with single vector */
1702		rc = pci_enable_msix_range(bp->pdev, &bp->msix_table[0], 1, 1);
1703		if (rc < 0) {
1704			BNX2X_DEV_INFO("Single MSI-X is not attainable rc %d\n",
1705				       rc);
1706			goto no_msix;
1707		}
1708
1709		BNX2X_DEV_INFO("Using single MSI-X vector\n");
1710		bp->flags |= USING_SINGLE_MSIX_FLAG;
1711
1712		BNX2X_DEV_INFO("set number of queues to 1\n");
1713		bp->num_ethernet_queues = 1;
1714		bp->num_queues = bp->num_ethernet_queues + bp->num_cnic_queues;
1715	} else if (rc < 0) {
1716		BNX2X_DEV_INFO("MSI-X is not attainable rc %d\n", rc);
1717		goto no_msix;
1718	} else if (rc < msix_vec) {
1719		/* how less vectors we will have? */
1720		int diff = msix_vec - rc;
1721
1722		BNX2X_DEV_INFO("Trying to use less MSI-X vectors: %d\n", rc);
1723
1724		/*
1725		 * decrease number of queues by number of unallocated entries
1726		 */
1727		bp->num_ethernet_queues -= diff;
1728		bp->num_queues = bp->num_ethernet_queues + bp->num_cnic_queues;
1729
1730		BNX2X_DEV_INFO("New queue configuration set: %d\n",
1731			       bp->num_queues);
1732	}
1733
1734	bp->flags |= USING_MSIX_FLAG;
1735
1736	return 0;
1737
1738no_msix:
1739	/* fall to INTx if not enough memory */
1740	if (rc == -ENOMEM)
1741		bp->flags |= DISABLE_MSI_FLAG;
1742
1743	return rc;
1744}
1745
1746static int bnx2x_req_msix_irqs(struct bnx2x *bp)
1747{
1748	int i, rc, offset = 0;
1749
1750	/* no default status block for vf */
1751	if (IS_PF(bp)) {
1752		rc = request_irq(bp->msix_table[offset++].vector,
1753				 bnx2x_msix_sp_int, 0,
1754				 bp->dev->name, bp->dev);
1755		if (rc) {
1756			BNX2X_ERR("request sp irq failed\n");
1757			return -EBUSY;
1758		}
1759	}
1760
1761	if (CNIC_SUPPORT(bp))
1762		offset++;
1763
1764	for_each_eth_queue(bp, i) {
1765		struct bnx2x_fastpath *fp = &bp->fp[i];
1766		snprintf(fp->name, sizeof(fp->name), "%s-fp-%d",
1767			 bp->dev->name, i);
1768
1769		rc = request_irq(bp->msix_table[offset].vector,
1770				 bnx2x_msix_fp_int, 0, fp->name, fp);
1771		if (rc) {
1772			BNX2X_ERR("request fp #%d irq (%d) failed  rc %d\n", i,
1773			      bp->msix_table[offset].vector, rc);
1774			bnx2x_free_msix_irqs(bp, offset);
1775			return -EBUSY;
1776		}
1777
1778		offset++;
1779	}
1780
1781	i = BNX2X_NUM_ETH_QUEUES(bp);
1782	if (IS_PF(bp)) {
1783		offset = 1 + CNIC_SUPPORT(bp);
1784		netdev_info(bp->dev,
1785			    "using MSI-X  IRQs: sp %d  fp[%d] %d ... fp[%d] %d\n",
1786			    bp->msix_table[0].vector,
1787			    0, bp->msix_table[offset].vector,
1788			    i - 1, bp->msix_table[offset + i - 1].vector);
1789	} else {
1790		offset = CNIC_SUPPORT(bp);
1791		netdev_info(bp->dev,
1792			    "using MSI-X  IRQs: fp[%d] %d ... fp[%d] %d\n",
1793			    0, bp->msix_table[offset].vector,
1794			    i - 1, bp->msix_table[offset + i - 1].vector);
1795	}
1796	return 0;
1797}
1798
1799int bnx2x_enable_msi(struct bnx2x *bp)
1800{
1801	int rc;
1802
1803	rc = pci_enable_msi(bp->pdev);
1804	if (rc) {
1805		BNX2X_DEV_INFO("MSI is not attainable\n");
1806		return -1;
1807	}
1808	bp->flags |= USING_MSI_FLAG;
1809
1810	return 0;
1811}
1812
1813static int bnx2x_req_irq(struct bnx2x *bp)
1814{
1815	unsigned long flags;
1816	unsigned int irq;
1817
1818	if (bp->flags & (USING_MSI_FLAG | USING_MSIX_FLAG))
1819		flags = 0;
1820	else
1821		flags = IRQF_SHARED;
1822
1823	if (bp->flags & USING_MSIX_FLAG)
1824		irq = bp->msix_table[0].vector;
1825	else
1826		irq = bp->pdev->irq;
1827
1828	return request_irq(irq, bnx2x_interrupt, flags, bp->dev->name, bp->dev);
1829}
1830
1831static int bnx2x_setup_irqs(struct bnx2x *bp)
1832{
1833	int rc = 0;
1834	if (bp->flags & USING_MSIX_FLAG &&
1835	    !(bp->flags & USING_SINGLE_MSIX_FLAG)) {
1836		rc = bnx2x_req_msix_irqs(bp);
1837		if (rc)
1838			return rc;
1839	} else {
1840		rc = bnx2x_req_irq(bp);
1841		if (rc) {
1842			BNX2X_ERR("IRQ request failed  rc %d, aborting\n", rc);
1843			return rc;
1844		}
1845		if (bp->flags & USING_MSI_FLAG) {
1846			bp->dev->irq = bp->pdev->irq;
1847			netdev_info(bp->dev, "using MSI IRQ %d\n",
1848				    bp->dev->irq);
1849		}
1850		if (bp->flags & USING_MSIX_FLAG) {
1851			bp->dev->irq = bp->msix_table[0].vector;
1852			netdev_info(bp->dev, "using MSIX IRQ %d\n",
1853				    bp->dev->irq);
1854		}
1855	}
1856
1857	return 0;
1858}
1859
1860static void bnx2x_napi_enable_cnic(struct bnx2x *bp)
1861{
1862	int i;
1863
1864	for_each_rx_queue_cnic(bp, i) {
1865		napi_enable(&bnx2x_fp(bp, i, napi));
1866	}
1867}
1868
1869static void bnx2x_napi_enable(struct bnx2x *bp)
1870{
1871	int i;
1872
1873	for_each_eth_queue(bp, i) {
1874		napi_enable(&bnx2x_fp(bp, i, napi));
1875	}
1876}
1877
1878static void bnx2x_napi_disable_cnic(struct bnx2x *bp)
1879{
1880	int i;
1881
1882	for_each_rx_queue_cnic(bp, i) {
1883		napi_disable(&bnx2x_fp(bp, i, napi));
1884	}
1885}
1886
1887static void bnx2x_napi_disable(struct bnx2x *bp)
1888{
1889	int i;
1890
1891	for_each_eth_queue(bp, i) {
1892		napi_disable(&bnx2x_fp(bp, i, napi));
1893	}
1894}
1895
1896void bnx2x_netif_start(struct bnx2x *bp)
1897{
1898	if (netif_running(bp->dev)) {
1899		bnx2x_napi_enable(bp);
1900		if (CNIC_LOADED(bp))
1901			bnx2x_napi_enable_cnic(bp);
1902		bnx2x_int_enable(bp);
1903		if (bp->state == BNX2X_STATE_OPEN)
1904			netif_tx_wake_all_queues(bp->dev);
1905	}
1906}
1907
1908void bnx2x_netif_stop(struct bnx2x *bp, int disable_hw)
1909{
1910	bnx2x_int_disable_sync(bp, disable_hw);
1911	bnx2x_napi_disable(bp);
1912	if (CNIC_LOADED(bp))
1913		bnx2x_napi_disable_cnic(bp);
1914}
1915
1916u16 bnx2x_select_queue(struct net_device *dev, struct sk_buff *skb,
1917		       struct net_device *sb_dev)
1918{
1919	struct bnx2x *bp = netdev_priv(dev);
1920
1921	if (CNIC_LOADED(bp) && !NO_FCOE(bp)) {
1922		struct ethhdr *hdr = (struct ethhdr *)skb->data;
1923		u16 ether_type = ntohs(hdr->h_proto);
1924
1925		/* Skip VLAN tag if present */
1926		if (ether_type == ETH_P_8021Q) {
1927			struct vlan_ethhdr *vhdr = skb_vlan_eth_hdr(skb);
1928
1929			ether_type = ntohs(vhdr->h_vlan_encapsulated_proto);
1930		}
1931
1932		/* If ethertype is FCoE or FIP - use FCoE ring */
1933		if ((ether_type == ETH_P_FCOE) || (ether_type == ETH_P_FIP))
1934			return bnx2x_fcoe_tx(bp, txq_index);
1935	}
1936
1937	/* select a non-FCoE queue */
1938	return netdev_pick_tx(dev, skb, NULL) %
1939			(BNX2X_NUM_ETH_QUEUES(bp) * bp->max_cos);
1940}
1941
1942void bnx2x_set_num_queues(struct bnx2x *bp)
1943{
1944	/* RSS queues */
1945	bp->num_ethernet_queues = bnx2x_calc_num_queues(bp);
1946
1947	/* override in STORAGE SD modes */
1948	if (IS_MF_STORAGE_ONLY(bp))
1949		bp->num_ethernet_queues = 1;
1950
1951	/* Add special queues */
1952	bp->num_cnic_queues = CNIC_SUPPORT(bp); /* For FCOE */
1953	bp->num_queues = bp->num_ethernet_queues + bp->num_cnic_queues;
1954
1955	BNX2X_DEV_INFO("set number of queues to %d\n", bp->num_queues);
1956}
1957
1958/**
1959 * bnx2x_set_real_num_queues - configure netdev->real_num_[tx,rx]_queues
1960 *
1961 * @bp:		Driver handle
1962 * @include_cnic: handle cnic case
1963 *
1964 * We currently support for at most 16 Tx queues for each CoS thus we will
1965 * allocate a multiple of 16 for ETH L2 rings according to the value of the
1966 * bp->max_cos.
1967 *
1968 * If there is an FCoE L2 queue the appropriate Tx queue will have the next
1969 * index after all ETH L2 indices.
1970 *
1971 * If the actual number of Tx queues (for each CoS) is less than 16 then there
1972 * will be the holes at the end of each group of 16 ETh L2 indices (0..15,
1973 * 16..31,...) with indices that are not coupled with any real Tx queue.
1974 *
1975 * The proper configuration of skb->queue_mapping is handled by
1976 * bnx2x_select_queue() and __skb_tx_hash().
1977 *
1978 * bnx2x_setup_tc() takes care of the proper TC mappings so that __skb_tx_hash()
1979 * will return a proper Tx index if TC is enabled (netdev->num_tc > 0).
1980 */
1981static int bnx2x_set_real_num_queues(struct bnx2x *bp, int include_cnic)
1982{
1983	int rc, tx, rx;
1984
1985	tx = BNX2X_NUM_ETH_QUEUES(bp) * bp->max_cos;
1986	rx = BNX2X_NUM_ETH_QUEUES(bp);
1987
1988/* account for fcoe queue */
1989	if (include_cnic && !NO_FCOE(bp)) {
1990		rx++;
1991		tx++;
1992	}
1993
1994	rc = netif_set_real_num_tx_queues(bp->dev, tx);
1995	if (rc) {
1996		BNX2X_ERR("Failed to set real number of Tx queues: %d\n", rc);
1997		return rc;
1998	}
1999	rc = netif_set_real_num_rx_queues(bp->dev, rx);
2000	if (rc) {
2001		BNX2X_ERR("Failed to set real number of Rx queues: %d\n", rc);
2002		return rc;
2003	}
2004
2005	DP(NETIF_MSG_IFUP, "Setting real num queues to (tx, rx) (%d, %d)\n",
2006			  tx, rx);
2007
2008	return rc;
2009}
2010
2011static void bnx2x_set_rx_buf_size(struct bnx2x *bp)
2012{
2013	int i;
2014
2015	for_each_queue(bp, i) {
2016		struct bnx2x_fastpath *fp = &bp->fp[i];
2017		u32 mtu;
2018
2019		/* Always use a mini-jumbo MTU for the FCoE L2 ring */
2020		if (IS_FCOE_IDX(i))
2021			/*
2022			 * Although there are no IP frames expected to arrive to
2023			 * this ring we still want to add an
2024			 * IP_HEADER_ALIGNMENT_PADDING to prevent a buffer
2025			 * overrun attack.
2026			 */
2027			mtu = BNX2X_FCOE_MINI_JUMBO_MTU;
2028		else
2029			mtu = bp->dev->mtu;
2030		fp->rx_buf_size = BNX2X_FW_RX_ALIGN_START +
2031				  IP_HEADER_ALIGNMENT_PADDING +
2032				  ETH_OVERHEAD +
2033				  mtu +
2034				  BNX2X_FW_RX_ALIGN_END;
2035		fp->rx_buf_size = SKB_DATA_ALIGN(fp->rx_buf_size);
2036		/* Note : rx_buf_size doesn't take into account NET_SKB_PAD */
2037		if (fp->rx_buf_size + NET_SKB_PAD <= PAGE_SIZE)
2038			fp->rx_frag_size = fp->rx_buf_size + NET_SKB_PAD;
2039		else
2040			fp->rx_frag_size = 0;
2041	}
2042}
2043
2044static int bnx2x_init_rss(struct bnx2x *bp)
2045{
2046	int i;
2047	u8 num_eth_queues = BNX2X_NUM_ETH_QUEUES(bp);
2048
2049	/* Prepare the initial contents for the indirection table if RSS is
2050	 * enabled
2051	 */
2052	for (i = 0; i < sizeof(bp->rss_conf_obj.ind_table); i++)
2053		bp->rss_conf_obj.ind_table[i] =
2054			bp->fp->cl_id +
2055			ethtool_rxfh_indir_default(i, num_eth_queues);
2056
2057	/*
2058	 * For 57710 and 57711 SEARCHER configuration (rss_keys) is
2059	 * per-port, so if explicit configuration is needed , do it only
2060	 * for a PMF.
2061	 *
2062	 * For 57712 and newer on the other hand it's a per-function
2063	 * configuration.
2064	 */
2065	return bnx2x_config_rss_eth(bp, bp->port.pmf || !CHIP_IS_E1x(bp));
2066}
2067
2068int bnx2x_rss(struct bnx2x *bp, struct bnx2x_rss_config_obj *rss_obj,
2069	      bool config_hash, bool enable)
2070{
2071	struct bnx2x_config_rss_params params = {NULL};
2072
2073	/* Although RSS is meaningless when there is a single HW queue we
2074	 * still need it enabled in order to have HW Rx hash generated.
2075	 *
2076	 * if (!is_eth_multi(bp))
2077	 *      bp->multi_mode = ETH_RSS_MODE_DISABLED;
2078	 */
2079
2080	params.rss_obj = rss_obj;
2081
2082	__set_bit(RAMROD_COMP_WAIT, &params.ramrod_flags);
2083
2084	if (enable) {
2085		__set_bit(BNX2X_RSS_MODE_REGULAR, &params.rss_flags);
2086
2087		/* RSS configuration */
2088		__set_bit(BNX2X_RSS_IPV4, &params.rss_flags);
2089		__set_bit(BNX2X_RSS_IPV4_TCP, &params.rss_flags);
2090		__set_bit(BNX2X_RSS_IPV6, &params.rss_flags);
2091		__set_bit(BNX2X_RSS_IPV6_TCP, &params.rss_flags);
2092		if (rss_obj->udp_rss_v4)
2093			__set_bit(BNX2X_RSS_IPV4_UDP, &params.rss_flags);
2094		if (rss_obj->udp_rss_v6)
2095			__set_bit(BNX2X_RSS_IPV6_UDP, &params.rss_flags);
2096
2097		if (!CHIP_IS_E1x(bp)) {
2098			/* valid only for TUNN_MODE_VXLAN tunnel mode */
2099			__set_bit(BNX2X_RSS_IPV4_VXLAN, &params.rss_flags);
2100			__set_bit(BNX2X_RSS_IPV6_VXLAN, &params.rss_flags);
2101
2102			/* valid only for TUNN_MODE_GRE tunnel mode */
2103			__set_bit(BNX2X_RSS_TUNN_INNER_HDRS, &params.rss_flags);
2104		}
2105	} else {
2106		__set_bit(BNX2X_RSS_MODE_DISABLED, &params.rss_flags);
2107	}
2108
2109	/* Hash bits */
2110	params.rss_result_mask = MULTI_MASK;
2111
2112	memcpy(params.ind_table, rss_obj->ind_table, sizeof(params.ind_table));
2113
2114	if (config_hash) {
2115		/* RSS keys */
2116		netdev_rss_key_fill(params.rss_key, T_ETH_RSS_KEY * 4);
2117		__set_bit(BNX2X_RSS_SET_SRCH, &params.rss_flags);
2118	}
2119
2120	if (IS_PF(bp))
2121		return bnx2x_config_rss(bp, &params);
2122	else
2123		return bnx2x_vfpf_config_rss(bp, &params);
2124}
2125
2126static int bnx2x_init_hw(struct bnx2x *bp, u32 load_code)
2127{
2128	struct bnx2x_func_state_params func_params = {NULL};
2129
2130	/* Prepare parameters for function state transitions */
2131	__set_bit(RAMROD_COMP_WAIT, &func_params.ramrod_flags);
2132
2133	func_params.f_obj = &bp->func_obj;
2134	func_params.cmd = BNX2X_F_CMD_HW_INIT;
2135
2136	func_params.params.hw_init.load_phase = load_code;
2137
2138	return bnx2x_func_state_change(bp, &func_params);
2139}
2140
2141/*
2142 * Cleans the object that have internal lists without sending
2143 * ramrods. Should be run when interrupts are disabled.
2144 */
2145void bnx2x_squeeze_objects(struct bnx2x *bp)
2146{
2147	int rc;
2148	unsigned long ramrod_flags = 0, vlan_mac_flags = 0;
2149	struct bnx2x_mcast_ramrod_params rparam = {NULL};
2150	struct bnx2x_vlan_mac_obj *mac_obj = &bp->sp_objs->mac_obj;
2151
2152	/***************** Cleanup MACs' object first *************************/
2153
2154	/* Wait for completion of requested */
2155	__set_bit(RAMROD_COMP_WAIT, &ramrod_flags);
2156	/* Perform a dry cleanup */
2157	__set_bit(RAMROD_DRV_CLR_ONLY, &ramrod_flags);
2158
2159	/* Clean ETH primary MAC */
2160	__set_bit(BNX2X_ETH_MAC, &vlan_mac_flags);
2161	rc = mac_obj->delete_all(bp, &bp->sp_objs->mac_obj, &vlan_mac_flags,
2162				 &ramrod_flags);
2163	if (rc != 0)
2164		BNX2X_ERR("Failed to clean ETH MACs: %d\n", rc);
2165
2166	/* Cleanup UC list */
2167	vlan_mac_flags = 0;
2168	__set_bit(BNX2X_UC_LIST_MAC, &vlan_mac_flags);
2169	rc = mac_obj->delete_all(bp, mac_obj, &vlan_mac_flags,
2170				 &ramrod_flags);
2171	if (rc != 0)
2172		BNX2X_ERR("Failed to clean UC list MACs: %d\n", rc);
2173
2174	/***************** Now clean mcast object *****************************/
2175	rparam.mcast_obj = &bp->mcast_obj;
2176	__set_bit(RAMROD_DRV_CLR_ONLY, &rparam.ramrod_flags);
2177
2178	/* Add a DEL command... - Since we're doing a driver cleanup only,
2179	 * we take a lock surrounding both the initial send and the CONTs,
2180	 * as we don't want a true completion to disrupt us in the middle.
2181	 */
2182	netif_addr_lock_bh(bp->dev);
2183	rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_DEL);
2184	if (rc < 0)
2185		BNX2X_ERR("Failed to add a new DEL command to a multi-cast object: %d\n",
2186			  rc);
2187
2188	/* ...and wait until all pending commands are cleared */
2189	rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_CONT);
2190	while (rc != 0) {
2191		if (rc < 0) {
2192			BNX2X_ERR("Failed to clean multi-cast object: %d\n",
2193				  rc);
2194			netif_addr_unlock_bh(bp->dev);
2195			return;
2196		}
2197
2198		rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_CONT);
2199	}
2200	netif_addr_unlock_bh(bp->dev);
2201}
2202
2203#ifndef BNX2X_STOP_ON_ERROR
2204#define LOAD_ERROR_EXIT(bp, label) \
2205	do { \
2206		(bp)->state = BNX2X_STATE_ERROR; \
2207		goto label; \
2208	} while (0)
2209
2210#define LOAD_ERROR_EXIT_CNIC(bp, label) \
2211	do { \
2212		bp->cnic_loaded = false; \
2213		goto label; \
2214	} while (0)
2215#else /*BNX2X_STOP_ON_ERROR*/
2216#define LOAD_ERROR_EXIT(bp, label) \
2217	do { \
2218		(bp)->state = BNX2X_STATE_ERROR; \
2219		(bp)->panic = 1; \
2220		return -EBUSY; \
2221	} while (0)
2222#define LOAD_ERROR_EXIT_CNIC(bp, label) \
2223	do { \
2224		bp->cnic_loaded = false; \
2225		(bp)->panic = 1; \
2226		return -EBUSY; \
2227	} while (0)
2228#endif /*BNX2X_STOP_ON_ERROR*/
2229
2230static void bnx2x_free_fw_stats_mem(struct bnx2x *bp)
2231{
2232	BNX2X_PCI_FREE(bp->fw_stats, bp->fw_stats_mapping,
2233		       bp->fw_stats_data_sz + bp->fw_stats_req_sz);
2234	return;
2235}
2236
2237static int bnx2x_alloc_fw_stats_mem(struct bnx2x *bp)
2238{
2239	int num_groups, vf_headroom = 0;
2240	int is_fcoe_stats = NO_FCOE(bp) ? 0 : 1;
2241
2242	/* number of queues for statistics is number of eth queues + FCoE */
2243	u8 num_queue_stats = BNX2X_NUM_ETH_QUEUES(bp) + is_fcoe_stats;
2244
2245	/* Total number of FW statistics requests =
2246	 * 1 for port stats + 1 for PF stats + potential 2 for FCoE (fcoe proper
2247	 * and fcoe l2 queue) stats + num of queues (which includes another 1
2248	 * for fcoe l2 queue if applicable)
2249	 */
2250	bp->fw_stats_num = 2 + is_fcoe_stats + num_queue_stats;
2251
2252	/* vf stats appear in the request list, but their data is allocated by
2253	 * the VFs themselves. We don't include them in the bp->fw_stats_num as
2254	 * it is used to determine where to place the vf stats queries in the
2255	 * request struct
2256	 */
2257	if (IS_SRIOV(bp))
2258		vf_headroom = bnx2x_vf_headroom(bp);
2259
2260	/* Request is built from stats_query_header and an array of
2261	 * stats_query_cmd_group each of which contains
2262	 * STATS_QUERY_CMD_COUNT rules. The real number or requests is
2263	 * configured in the stats_query_header.
2264	 */
2265	num_groups =
2266		(((bp->fw_stats_num + vf_headroom) / STATS_QUERY_CMD_COUNT) +
2267		 (((bp->fw_stats_num + vf_headroom) % STATS_QUERY_CMD_COUNT) ?
2268		 1 : 0));
2269
2270	DP(BNX2X_MSG_SP, "stats fw_stats_num %d, vf headroom %d, num_groups %d\n",
2271	   bp->fw_stats_num, vf_headroom, num_groups);
2272	bp->fw_stats_req_sz = sizeof(struct stats_query_header) +
2273		num_groups * sizeof(struct stats_query_cmd_group);
2274
2275	/* Data for statistics requests + stats_counter
2276	 * stats_counter holds per-STORM counters that are incremented
2277	 * when STORM has finished with the current request.
2278	 * memory for FCoE offloaded statistics are counted anyway,
2279	 * even if they will not be sent.
2280	 * VF stats are not accounted for here as the data of VF stats is stored
2281	 * in memory allocated by the VF, not here.
2282	 */
2283	bp->fw_stats_data_sz = sizeof(struct per_port_stats) +
2284		sizeof(struct per_pf_stats) +
2285		sizeof(struct fcoe_statistics_params) +
2286		sizeof(struct per_queue_stats) * num_queue_stats +
2287		sizeof(struct stats_counter);
2288
2289	bp->fw_stats = BNX2X_PCI_ALLOC(&bp->fw_stats_mapping,
2290				       bp->fw_stats_data_sz + bp->fw_stats_req_sz);
2291	if (!bp->fw_stats)
2292		goto alloc_mem_err;
2293
2294	/* Set shortcuts */
2295	bp->fw_stats_req = (struct bnx2x_fw_stats_req *)bp->fw_stats;
2296	bp->fw_stats_req_mapping = bp->fw_stats_mapping;
2297	bp->fw_stats_data = (struct bnx2x_fw_stats_data *)
2298		((u8 *)bp->fw_stats + bp->fw_stats_req_sz);
2299	bp->fw_stats_data_mapping = bp->fw_stats_mapping +
2300		bp->fw_stats_req_sz;
2301
2302	DP(BNX2X_MSG_SP, "statistics request base address set to %x %x\n",
2303	   U64_HI(bp->fw_stats_req_mapping),
2304	   U64_LO(bp->fw_stats_req_mapping));
2305	DP(BNX2X_MSG_SP, "statistics data base address set to %x %x\n",
2306	   U64_HI(bp->fw_stats_data_mapping),
2307	   U64_LO(bp->fw_stats_data_mapping));
2308	return 0;
2309
2310alloc_mem_err:
2311	bnx2x_free_fw_stats_mem(bp);
2312	BNX2X_ERR("Can't allocate FW stats memory\n");
2313	return -ENOMEM;
2314}
2315
2316/* send load request to mcp and analyze response */
2317static int bnx2x_nic_load_request(struct bnx2x *bp, u32 *load_code)
2318{
2319	u32 param;
2320
2321	/* init fw_seq */
2322	bp->fw_seq =
2323		(SHMEM_RD(bp, func_mb[BP_FW_MB_IDX(bp)].drv_mb_header) &
2324		 DRV_MSG_SEQ_NUMBER_MASK);
2325	BNX2X_DEV_INFO("fw_seq 0x%08x\n", bp->fw_seq);
2326
2327	/* Get current FW pulse sequence */
2328	bp->fw_drv_pulse_wr_seq =
2329		(SHMEM_RD(bp, func_mb[BP_FW_MB_IDX(bp)].drv_pulse_mb) &
2330		 DRV_PULSE_SEQ_MASK);
2331	BNX2X_DEV_INFO("drv_pulse 0x%x\n", bp->fw_drv_pulse_wr_seq);
2332
2333	param = DRV_MSG_CODE_LOAD_REQ_WITH_LFA;
2334
2335	if (IS_MF_SD(bp) && bnx2x_port_after_undi(bp))
2336		param |= DRV_MSG_CODE_LOAD_REQ_FORCE_LFA;
2337
2338	/* load request */
2339	(*load_code) = bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_REQ, param);
2340
2341	/* if mcp fails to respond we must abort */
2342	if (!(*load_code)) {
2343		BNX2X_ERR("MCP response failure, aborting\n");
2344		return -EBUSY;
2345	}
2346
2347	/* If mcp refused (e.g. other port is in diagnostic mode) we
2348	 * must abort
2349	 */
2350	if ((*load_code) == FW_MSG_CODE_DRV_LOAD_REFUSED) {
2351		BNX2X_ERR("MCP refused load request, aborting\n");
2352		return -EBUSY;
2353	}
2354	return 0;
2355}
2356
2357/* check whether another PF has already loaded FW to chip. In
2358 * virtualized environments a pf from another VM may have already
2359 * initialized the device including loading FW
2360 */
2361int bnx2x_compare_fw_ver(struct bnx2x *bp, u32 load_code, bool print_err)
2362{
2363	/* is another pf loaded on this engine? */
2364	if (load_code != FW_MSG_CODE_DRV_LOAD_COMMON_CHIP &&
2365	    load_code != FW_MSG_CODE_DRV_LOAD_COMMON) {
2366		u8 loaded_fw_major, loaded_fw_minor, loaded_fw_rev, loaded_fw_eng;
2367		u32 loaded_fw;
2368
2369		/* read loaded FW from chip */
2370		loaded_fw = REG_RD(bp, XSEM_REG_PRAM);
2371
2372		loaded_fw_major = loaded_fw & 0xff;
2373		loaded_fw_minor = (loaded_fw >> 8) & 0xff;
2374		loaded_fw_rev = (loaded_fw >> 16) & 0xff;
2375		loaded_fw_eng = (loaded_fw >> 24) & 0xff;
2376
2377		DP(BNX2X_MSG_SP, "loaded fw 0x%x major 0x%x minor 0x%x rev 0x%x eng 0x%x\n",
2378		   loaded_fw, loaded_fw_major, loaded_fw_minor, loaded_fw_rev, loaded_fw_eng);
2379
2380		/* abort nic load if version mismatch */
2381		if (loaded_fw_major != BCM_5710_FW_MAJOR_VERSION ||
2382		    loaded_fw_minor != BCM_5710_FW_MINOR_VERSION ||
2383		    loaded_fw_eng != BCM_5710_FW_ENGINEERING_VERSION ||
2384		    loaded_fw_rev < BCM_5710_FW_REVISION_VERSION_V15) {
2385			if (print_err)
2386				BNX2X_ERR("loaded FW incompatible. Aborting\n");
2387			else
2388				BNX2X_DEV_INFO("loaded FW incompatible, possibly due to MF UNDI\n");
2389
2390			return -EBUSY;
2391		}
2392	}
2393	return 0;
2394}
2395
2396/* returns the "mcp load_code" according to global load_count array */
2397static int bnx2x_nic_load_no_mcp(struct bnx2x *bp, int port)
2398{
2399	int path = BP_PATH(bp);
2400
2401	DP(NETIF_MSG_IFUP, "NO MCP - load counts[%d]      %d, %d, %d\n",
2402	   path, bnx2x_load_count[path][0], bnx2x_load_count[path][1],
2403	   bnx2x_load_count[path][2]);
2404	bnx2x_load_count[path][0]++;
2405	bnx2x_load_count[path][1 + port]++;
2406	DP(NETIF_MSG_IFUP, "NO MCP - new load counts[%d]  %d, %d, %d\n",
2407	   path, bnx2x_load_count[path][0], bnx2x_load_count[path][1],
2408	   bnx2x_load_count[path][2]);
2409	if (bnx2x_load_count[path][0] == 1)
2410		return FW_MSG_CODE_DRV_LOAD_COMMON;
2411	else if (bnx2x_load_count[path][1 + port] == 1)
2412		return FW_MSG_CODE_DRV_LOAD_PORT;
2413	else
2414		return FW_MSG_CODE_DRV_LOAD_FUNCTION;
2415}
2416
2417/* mark PMF if applicable */
2418static void bnx2x_nic_load_pmf(struct bnx2x *bp, u32 load_code)
2419{
2420	if ((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
2421	    (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP) ||
2422	    (load_code == FW_MSG_CODE_DRV_LOAD_PORT)) {
2423		bp->port.pmf = 1;
2424		/* We need the barrier to ensure the ordering between the
2425		 * writing to bp->port.pmf here and reading it from the
2426		 * bnx2x_periodic_task().
2427		 */
2428		smp_mb();
2429	} else {
2430		bp->port.pmf = 0;
2431	}
2432
2433	DP(NETIF_MSG_LINK, "pmf %d\n", bp->port.pmf);
2434}
2435
2436static void bnx2x_nic_load_afex_dcc(struct bnx2x *bp, int load_code)
2437{
2438	if (((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
2439	     (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP)) &&
2440	    (bp->common.shmem2_base)) {
2441		if (SHMEM2_HAS(bp, dcc_support))
2442			SHMEM2_WR(bp, dcc_support,
2443				  (SHMEM_DCC_SUPPORT_DISABLE_ENABLE_PF_TLV |
2444				   SHMEM_DCC_SUPPORT_BANDWIDTH_ALLOCATION_TLV));
2445		if (SHMEM2_HAS(bp, afex_driver_support))
2446			SHMEM2_WR(bp, afex_driver_support,
2447				  SHMEM_AFEX_SUPPORTED_VERSION_ONE);
2448	}
2449
2450	/* Set AFEX default VLAN tag to an invalid value */
2451	bp->afex_def_vlan_tag = -1;
2452}
2453
2454/**
2455 * bnx2x_bz_fp - zero content of the fastpath structure.
2456 *
2457 * @bp:		driver handle
2458 * @index:	fastpath index to be zeroed
2459 *
2460 * Makes sure the contents of the bp->fp[index].napi is kept
2461 * intact.
2462 */
2463static void bnx2x_bz_fp(struct bnx2x *bp, int index)
2464{
2465	struct bnx2x_fastpath *fp = &bp->fp[index];
2466	int cos;
2467	struct napi_struct orig_napi = fp->napi;
2468	struct bnx2x_agg_info *orig_tpa_info = fp->tpa_info;
2469
2470	/* bzero bnx2x_fastpath contents */
2471	if (fp->tpa_info)
2472		memset(fp->tpa_info, 0, ETH_MAX_AGGREGATION_QUEUES_E1H_E2 *
2473		       sizeof(struct bnx2x_agg_info));
2474	memset(fp, 0, sizeof(*fp));
2475
2476	/* Restore the NAPI object as it has been already initialized */
2477	fp->napi = orig_napi;
2478	fp->tpa_info = orig_tpa_info;
2479	fp->bp = bp;
2480	fp->index = index;
2481	if (IS_ETH_FP(fp))
2482		fp->max_cos = bp->max_cos;
2483	else
2484		/* Special queues support only one CoS */
2485		fp->max_cos = 1;
2486
2487	/* Init txdata pointers */
2488	if (IS_FCOE_FP(fp))
2489		fp->txdata_ptr[0] = &bp->bnx2x_txq[FCOE_TXQ_IDX(bp)];
2490	if (IS_ETH_FP(fp))
2491		for_each_cos_in_tx_queue(fp, cos)
2492			fp->txdata_ptr[cos] = &bp->bnx2x_txq[cos *
2493				BNX2X_NUM_ETH_QUEUES(bp) + index];
2494
2495	/* set the tpa flag for each queue. The tpa flag determines the queue
2496	 * minimal size so it must be set prior to queue memory allocation
2497	 */
2498	if (bp->dev->features & NETIF_F_LRO)
2499		fp->mode = TPA_MODE_LRO;
2500	else if (bp->dev->features & NETIF_F_GRO_HW)
2501		fp->mode = TPA_MODE_GRO;
2502	else
2503		fp->mode = TPA_MODE_DISABLED;
2504
2505	/* We don't want TPA if it's disabled in bp
2506	 * or if this is an FCoE L2 ring.
2507	 */
2508	if (bp->disable_tpa || IS_FCOE_FP(fp))
2509		fp->mode = TPA_MODE_DISABLED;
2510}
2511
2512void bnx2x_set_os_driver_state(struct bnx2x *bp, u32 state)
2513{
2514	u32 cur;
2515
2516	if (!IS_MF_BD(bp) || !SHMEM2_HAS(bp, os_driver_state) || IS_VF(bp))
2517		return;
2518
2519	cur = SHMEM2_RD(bp, os_driver_state[BP_FW_MB_IDX(bp)]);
2520	DP(NETIF_MSG_IFUP, "Driver state %08x-->%08x\n",
2521	   cur, state);
2522
2523	SHMEM2_WR(bp, os_driver_state[BP_FW_MB_IDX(bp)], state);
2524}
2525
2526int bnx2x_load_cnic(struct bnx2x *bp)
2527{
2528	int i, rc, port = BP_PORT(bp);
2529
2530	DP(NETIF_MSG_IFUP, "Starting CNIC-related load\n");
2531
2532	mutex_init(&bp->cnic_mutex);
2533
2534	if (IS_PF(bp)) {
2535		rc = bnx2x_alloc_mem_cnic(bp);
2536		if (rc) {
2537			BNX2X_ERR("Unable to allocate bp memory for cnic\n");
2538			LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic0);
2539		}
2540	}
2541
2542	rc = bnx2x_alloc_fp_mem_cnic(bp);
2543	if (rc) {
2544		BNX2X_ERR("Unable to allocate memory for cnic fps\n");
2545		LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic0);
2546	}
2547
2548	/* Update the number of queues with the cnic queues */
2549	rc = bnx2x_set_real_num_queues(bp, 1);
2550	if (rc) {
2551		BNX2X_ERR("Unable to set real_num_queues including cnic\n");
2552		LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic0);
2553	}
2554
2555	/* Add all CNIC NAPI objects */
2556	bnx2x_add_all_napi_cnic(bp);
2557	DP(NETIF_MSG_IFUP, "cnic napi added\n");
2558	bnx2x_napi_enable_cnic(bp);
2559
2560	rc = bnx2x_init_hw_func_cnic(bp);
2561	if (rc)
2562		LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic1);
2563
2564	bnx2x_nic_init_cnic(bp);
2565
2566	if (IS_PF(bp)) {
2567		/* Enable Timer scan */
2568		REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + port*4, 1);
2569
2570		/* setup cnic queues */
2571		for_each_cnic_queue(bp, i) {
2572			rc = bnx2x_setup_queue(bp, &bp->fp[i], 0);
2573			if (rc) {
2574				BNX2X_ERR("Queue setup failed\n");
2575				LOAD_ERROR_EXIT(bp, load_error_cnic2);
2576			}
2577		}
2578	}
2579
2580	/* Initialize Rx filter. */
2581	bnx2x_set_rx_mode_inner(bp);
2582
2583	/* re-read iscsi info */
2584	bnx2x_get_iscsi_info(bp);
2585	bnx2x_setup_cnic_irq_info(bp);
2586	bnx2x_setup_cnic_info(bp);
2587	bp->cnic_loaded = true;
2588	if (bp->state == BNX2X_STATE_OPEN)
2589		bnx2x_cnic_notify(bp, CNIC_CTL_START_CMD);
2590
2591	DP(NETIF_MSG_IFUP, "Ending successfully CNIC-related load\n");
2592
2593	return 0;
2594
2595#ifndef BNX2X_STOP_ON_ERROR
2596load_error_cnic2:
2597	/* Disable Timer scan */
2598	REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + port*4, 0);
2599
2600load_error_cnic1:
2601	bnx2x_napi_disable_cnic(bp);
2602	/* Update the number of queues without the cnic queues */
2603	if (bnx2x_set_real_num_queues(bp, 0))
2604		BNX2X_ERR("Unable to set real_num_queues not including cnic\n");
2605load_error_cnic0:
2606	BNX2X_ERR("CNIC-related load failed\n");
2607	bnx2x_free_fp_mem_cnic(bp);
2608	bnx2x_free_mem_cnic(bp);
2609	return rc;
2610#endif /* ! BNX2X_STOP_ON_ERROR */
2611}
2612
2613/* must be called with rtnl_lock */
2614int bnx2x_nic_load(struct bnx2x *bp, int load_mode)
2615{
2616	int port = BP_PORT(bp);
2617	int i, rc = 0, load_code = 0;
2618
2619	DP(NETIF_MSG_IFUP, "Starting NIC load\n");
2620	DP(NETIF_MSG_IFUP,
2621	   "CNIC is %s\n", CNIC_ENABLED(bp) ? "enabled" : "disabled");
2622
2623#ifdef BNX2X_STOP_ON_ERROR
2624	if (unlikely(bp->panic)) {
2625		BNX2X_ERR("Can't load NIC when there is panic\n");
2626		return -EPERM;
2627	}
2628#endif
2629
2630	bp->state = BNX2X_STATE_OPENING_WAIT4_LOAD;
2631
2632	/* zero the structure w/o any lock, before SP handler is initialized */
2633	memset(&bp->last_reported_link, 0, sizeof(bp->last_reported_link));
2634	__set_bit(BNX2X_LINK_REPORT_LINK_DOWN,
2635		&bp->last_reported_link.link_report_flags);
2636
2637	if (IS_PF(bp))
2638		/* must be called before memory allocation and HW init */
2639		bnx2x_ilt_set_info(bp);
2640
2641	/*
2642	 * Zero fastpath structures preserving invariants like napi, which are
2643	 * allocated only once, fp index, max_cos, bp pointer.
2644	 * Also set fp->mode and txdata_ptr.
2645	 */
2646	DP(NETIF_MSG_IFUP, "num queues: %d", bp->num_queues);
2647	for_each_queue(bp, i)
2648		bnx2x_bz_fp(bp, i);
2649	memset(bp->bnx2x_txq, 0, (BNX2X_MAX_RSS_COUNT(bp) * BNX2X_MULTI_TX_COS +
2650				  bp->num_cnic_queues) *
2651				  sizeof(struct bnx2x_fp_txdata));
2652
2653	bp->fcoe_init = false;
2654
2655	/* Set the receive queues buffer size */
2656	bnx2x_set_rx_buf_size(bp);
2657
2658	if (IS_PF(bp)) {
2659		rc = bnx2x_alloc_mem(bp);
2660		if (rc) {
2661			BNX2X_ERR("Unable to allocate bp memory\n");
2662			return rc;
2663		}
2664	}
2665
2666	/* need to be done after alloc mem, since it's self adjusting to amount
2667	 * of memory available for RSS queues
2668	 */
2669	rc = bnx2x_alloc_fp_mem(bp);
2670	if (rc) {
2671		BNX2X_ERR("Unable to allocate memory for fps\n");
2672		LOAD_ERROR_EXIT(bp, load_error0);
2673	}
2674
2675	/* Allocated memory for FW statistics  */
2676	rc = bnx2x_alloc_fw_stats_mem(bp);
2677	if (rc)
2678		LOAD_ERROR_EXIT(bp, load_error0);
2679
2680	/* request pf to initialize status blocks */
2681	if (IS_VF(bp)) {
2682		rc = bnx2x_vfpf_init(bp);
2683		if (rc)
2684			LOAD_ERROR_EXIT(bp, load_error0);
2685	}
2686
2687	/* As long as bnx2x_alloc_mem() may possibly update
2688	 * bp->num_queues, bnx2x_set_real_num_queues() should always
2689	 * come after it. At this stage cnic queues are not counted.
2690	 */
2691	rc = bnx2x_set_real_num_queues(bp, 0);
2692	if (rc) {
2693		BNX2X_ERR("Unable to set real_num_queues\n");
2694		LOAD_ERROR_EXIT(bp, load_error0);
2695	}
2696
2697	/* configure multi cos mappings in kernel.
2698	 * this configuration may be overridden by a multi class queue
2699	 * discipline or by a dcbx negotiation result.
2700	 */
2701	bnx2x_setup_tc(bp->dev, bp->max_cos);
2702
2703	/* Add all NAPI objects */
2704	bnx2x_add_all_napi(bp);
2705	DP(NETIF_MSG_IFUP, "napi added\n");
2706	bnx2x_napi_enable(bp);
2707
2708	if (IS_PF(bp)) {
2709		/* set pf load just before approaching the MCP */
2710		bnx2x_set_pf_load(bp);
2711
2712		/* if mcp exists send load request and analyze response */
2713		if (!BP_NOMCP(bp)) {
2714			/* attempt to load pf */
2715			rc = bnx2x_nic_load_request(bp, &load_code);
2716			if (rc)
2717				LOAD_ERROR_EXIT(bp, load_error1);
2718
2719			/* what did mcp say? */
2720			rc = bnx2x_compare_fw_ver(bp, load_code, true);
2721			if (rc) {
2722				bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2723				LOAD_ERROR_EXIT(bp, load_error2);
2724			}
2725		} else {
2726			load_code = bnx2x_nic_load_no_mcp(bp, port);
2727		}
2728
2729		/* mark pmf if applicable */
2730		bnx2x_nic_load_pmf(bp, load_code);
2731
2732		/* Init Function state controlling object */
2733		bnx2x__init_func_obj(bp);
2734
2735		/* Initialize HW */
2736		rc = bnx2x_init_hw(bp, load_code);
2737		if (rc) {
2738			BNX2X_ERR("HW init failed, aborting\n");
2739			bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2740			LOAD_ERROR_EXIT(bp, load_error2);
2741		}
2742	}
2743
2744	bnx2x_pre_irq_nic_init(bp);
2745
2746	/* Connect to IRQs */
2747	rc = bnx2x_setup_irqs(bp);
2748	if (rc) {
2749		BNX2X_ERR("setup irqs failed\n");
2750		if (IS_PF(bp))
2751			bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2752		LOAD_ERROR_EXIT(bp, load_error2);
2753	}
2754
2755	/* Init per-function objects */
2756	if (IS_PF(bp)) {
2757		/* Setup NIC internals and enable interrupts */
2758		bnx2x_post_irq_nic_init(bp, load_code);
2759
2760		bnx2x_init_bp_objs(bp);
2761		bnx2x_iov_nic_init(bp);
2762
2763		/* Set AFEX default VLAN tag to an invalid value */
2764		bp->afex_def_vlan_tag = -1;
2765		bnx2x_nic_load_afex_dcc(bp, load_code);
2766		bp->state = BNX2X_STATE_OPENING_WAIT4_PORT;
2767		rc = bnx2x_func_start(bp);
2768		if (rc) {
2769			BNX2X_ERR("Function start failed!\n");
2770			bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2771
2772			LOAD_ERROR_EXIT(bp, load_error3);
2773		}
2774
2775		/* Send LOAD_DONE command to MCP */
2776		if (!BP_NOMCP(bp)) {
2777			load_code = bnx2x_fw_command(bp,
2778						     DRV_MSG_CODE_LOAD_DONE, 0);
2779			if (!load_code) {
2780				BNX2X_ERR("MCP response failure, aborting\n");
2781				rc = -EBUSY;
2782				LOAD_ERROR_EXIT(bp, load_error3);
2783			}
2784		}
2785
2786		/* initialize FW coalescing state machines in RAM */
2787		bnx2x_update_coalesce(bp);
2788	}
2789
2790	/* setup the leading queue */
2791	rc = bnx2x_setup_leading(bp);
2792	if (rc) {
2793		BNX2X_ERR("Setup leading failed!\n");
2794		LOAD_ERROR_EXIT(bp, load_error3);
2795	}
2796
2797	/* set up the rest of the queues */
2798	for_each_nondefault_eth_queue(bp, i) {
2799		if (IS_PF(bp))
2800			rc = bnx2x_setup_queue(bp, &bp->fp[i], false);
2801		else /* VF */
2802			rc = bnx2x_vfpf_setup_q(bp, &bp->fp[i], false);
2803		if (rc) {
2804			BNX2X_ERR("Queue %d setup failed\n", i);
2805			LOAD_ERROR_EXIT(bp, load_error3);
2806		}
2807	}
2808
2809	/* setup rss */
2810	rc = bnx2x_init_rss(bp);
2811	if (rc) {
2812		BNX2X_ERR("PF RSS init failed\n");
2813		LOAD_ERROR_EXIT(bp, load_error3);
2814	}
2815
2816	/* Now when Clients are configured we are ready to work */
2817	bp->state = BNX2X_STATE_OPEN;
2818
2819	/* Configure a ucast MAC */
2820	if (IS_PF(bp))
2821		rc = bnx2x_set_eth_mac(bp, true);
2822	else /* vf */
2823		rc = bnx2x_vfpf_config_mac(bp, bp->dev->dev_addr, bp->fp->index,
2824					   true);
2825	if (rc) {
2826		BNX2X_ERR("Setting Ethernet MAC failed\n");
2827		LOAD_ERROR_EXIT(bp, load_error3);
2828	}
2829
2830	if (IS_PF(bp) && bp->pending_max) {
2831		bnx2x_update_max_mf_config(bp, bp->pending_max);
2832		bp->pending_max = 0;
2833	}
2834
2835	bp->force_link_down = false;
2836	if (bp->port.pmf) {
2837		rc = bnx2x_initial_phy_init(bp, load_mode);
2838		if (rc)
2839			LOAD_ERROR_EXIT(bp, load_error3);
2840	}
2841	bp->link_params.feature_config_flags &= ~FEATURE_CONFIG_BOOT_FROM_SAN;
2842
2843	/* Start fast path */
2844
2845	/* Re-configure vlan filters */
2846	rc = bnx2x_vlan_reconfigure_vid(bp);
2847	if (rc)
2848		LOAD_ERROR_EXIT(bp, load_error3);
2849
2850	/* Initialize Rx filter. */
2851	bnx2x_set_rx_mode_inner(bp);
2852
2853	if (bp->flags & PTP_SUPPORTED) {
2854		bnx2x_register_phc(bp);
2855		bnx2x_init_ptp(bp);
2856		bnx2x_configure_ptp_filters(bp);
2857	}
2858	/* Start Tx */
2859	switch (load_mode) {
2860	case LOAD_NORMAL:
2861		/* Tx queue should be only re-enabled */
2862		netif_tx_wake_all_queues(bp->dev);
2863		break;
2864
2865	case LOAD_OPEN:
2866		netif_tx_start_all_queues(bp->dev);
2867		smp_mb__after_atomic();
2868		break;
2869
2870	case LOAD_DIAG:
2871	case LOAD_LOOPBACK_EXT:
2872		bp->state = BNX2X_STATE_DIAG;
2873		break;
2874
2875	default:
2876		break;
2877	}
2878
2879	if (bp->port.pmf)
2880		bnx2x_update_drv_flags(bp, 1 << DRV_FLAGS_PORT_MASK, 0);
2881	else
2882		bnx2x__link_status_update(bp);
2883
2884	/* start the timer */
2885	mod_timer(&bp->timer, jiffies + bp->current_interval);
2886
2887	if (CNIC_ENABLED(bp))
2888		bnx2x_load_cnic(bp);
2889
2890	if (IS_PF(bp))
2891		bnx2x_schedule_sp_rtnl(bp, BNX2X_SP_RTNL_GET_DRV_VERSION, 0);
2892
2893	if (IS_PF(bp) && SHMEM2_HAS(bp, drv_capabilities_flag)) {
2894		/* mark driver is loaded in shmem2 */
2895		u32 val;
2896		val = SHMEM2_RD(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)]);
2897		val &= ~DRV_FLAGS_MTU_MASK;
2898		val |= (bp->dev->mtu << DRV_FLAGS_MTU_SHIFT);
2899		SHMEM2_WR(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)],
2900			  val | DRV_FLAGS_CAPABILITIES_LOADED_SUPPORTED |
2901			  DRV_FLAGS_CAPABILITIES_LOADED_L2);
2902	}
2903
2904	/* Wait for all pending SP commands to complete */
2905	if (IS_PF(bp) && !bnx2x_wait_sp_comp(bp, ~0x0UL)) {
2906		BNX2X_ERR("Timeout waiting for SP elements to complete\n");
2907		bnx2x_nic_unload(bp, UNLOAD_CLOSE, false);
2908		return -EBUSY;
2909	}
2910
2911	/* Update driver data for On-Chip MFW dump. */
2912	if (IS_PF(bp))
2913		bnx2x_update_mfw_dump(bp);
2914
2915	/* If PMF - send ADMIN DCBX msg to MFW to initiate DCBX FSM */
2916	if (bp->port.pmf && (bp->state != BNX2X_STATE_DIAG))
2917		bnx2x_dcbx_init(bp, false);
2918
2919	if (!IS_MF_SD_STORAGE_PERSONALITY_ONLY(bp))
2920		bnx2x_set_os_driver_state(bp, OS_DRIVER_STATE_ACTIVE);
2921
2922	DP(NETIF_MSG_IFUP, "Ending successfully NIC load\n");
2923
2924	return 0;
2925
2926#ifndef BNX2X_STOP_ON_ERROR
2927load_error3:
2928	if (IS_PF(bp)) {
2929		bnx2x_int_disable_sync(bp, 1);
2930
2931		/* Clean queueable objects */
2932		bnx2x_squeeze_objects(bp);
2933	}
2934
2935	/* Free SKBs, SGEs, TPA pool and driver internals */
2936	bnx2x_free_skbs(bp);
2937	for_each_rx_queue(bp, i)
2938		bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
2939
2940	/* Release IRQs */
2941	bnx2x_free_irq(bp);
2942load_error2:
2943	if (IS_PF(bp) && !BP_NOMCP(bp)) {
2944		bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_REQ_WOL_MCP, 0);
2945		bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_DONE, 0);
2946	}
2947
2948	bp->port.pmf = 0;
2949load_error1:
2950	bnx2x_napi_disable(bp);
2951	bnx2x_del_all_napi(bp);
2952
2953	/* clear pf_load status, as it was already set */
2954	if (IS_PF(bp))
2955		bnx2x_clear_pf_load(bp);
2956load_error0:
2957	bnx2x_free_fw_stats_mem(bp);
2958	bnx2x_free_fp_mem(bp);
2959	bnx2x_free_mem(bp);
2960
2961	return rc;
2962#endif /* ! BNX2X_STOP_ON_ERROR */
2963}
2964
2965int bnx2x_drain_tx_queues(struct bnx2x *bp)
2966{
2967	u8 rc = 0, cos, i;
2968
2969	/* Wait until tx fastpath tasks complete */
2970	for_each_tx_queue(bp, i) {
2971		struct bnx2x_fastpath *fp = &bp->fp[i];
2972
2973		for_each_cos_in_tx_queue(fp, cos)
2974			rc = bnx2x_clean_tx_queue(bp, fp->txdata_ptr[cos]);
2975		if (rc)
2976			return rc;
2977	}
2978	return 0;
2979}
2980
2981/* must be called with rtnl_lock */
2982int bnx2x_nic_unload(struct bnx2x *bp, int unload_mode, bool keep_link)
2983{
2984	int i;
2985	bool global = false;
2986
2987	DP(NETIF_MSG_IFUP, "Starting NIC unload\n");
2988
2989	if (!IS_MF_SD_STORAGE_PERSONALITY_ONLY(bp))
2990		bnx2x_set_os_driver_state(bp, OS_DRIVER_STATE_DISABLED);
2991
2992	/* mark driver is unloaded in shmem2 */
2993	if (IS_PF(bp) && SHMEM2_HAS(bp, drv_capabilities_flag)) {
2994		u32 val;
2995		val = SHMEM2_RD(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)]);
2996		SHMEM2_WR(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)],
2997			  val & ~DRV_FLAGS_CAPABILITIES_LOADED_L2);
2998	}
2999
3000	if (IS_PF(bp) && bp->recovery_state != BNX2X_RECOVERY_DONE &&
3001	    (bp->state == BNX2X_STATE_CLOSED ||
3002	     bp->state == BNX2X_STATE_ERROR)) {
3003		/* We can get here if the driver has been unloaded
3004		 * during parity error recovery and is either waiting for a
3005		 * leader to complete or for other functions to unload and
3006		 * then ifdown has been issued. In this case we want to
3007		 * unload and let other functions to complete a recovery
3008		 * process.
3009		 */
3010		bp->recovery_state = BNX2X_RECOVERY_DONE;
3011		bp->is_leader = 0;
3012		bnx2x_release_leader_lock(bp);
3013		smp_mb();
3014
3015		DP(NETIF_MSG_IFDOWN, "Releasing a leadership...\n");
3016		BNX2X_ERR("Can't unload in closed or error state\n");
3017		return -EINVAL;
3018	}
3019
3020	/* Nothing to do during unload if previous bnx2x_nic_load()
3021	 * have not completed successfully - all resources are released.
3022	 *
3023	 * we can get here only after unsuccessful ndo_* callback, during which
3024	 * dev->IFF_UP flag is still on.
3025	 */
3026	if (bp->state == BNX2X_STATE_CLOSED || bp->state == BNX2X_STATE_ERROR)
3027		return 0;
3028
3029	/* It's important to set the bp->state to the value different from
3030	 * BNX2X_STATE_OPEN and only then stop the Tx. Otherwise bnx2x_tx_int()
3031	 * may restart the Tx from the NAPI context (see bnx2x_tx_int()).
3032	 */
3033	bp->state = BNX2X_STATE_CLOSING_WAIT4_HALT;
3034	smp_mb();
3035
3036	/* indicate to VFs that the PF is going down */
3037	bnx2x_iov_channel_down(bp);
3038
3039	if (CNIC_LOADED(bp))
3040		bnx2x_cnic_notify(bp, CNIC_CTL_STOP_CMD);
3041
3042	/* Stop Tx */
3043	bnx2x_tx_disable(bp);
3044	netdev_reset_tc(bp->dev);
3045
3046	bp->rx_mode = BNX2X_RX_MODE_NONE;
3047
3048	del_timer_sync(&bp->timer);
3049
3050	if (IS_PF(bp) && !BP_NOMCP(bp)) {
3051		/* Set ALWAYS_ALIVE bit in shmem */
3052		bp->fw_drv_pulse_wr_seq |= DRV_PULSE_ALWAYS_ALIVE;
3053		bnx2x_drv_pulse(bp);
3054		bnx2x_stats_handle(bp, STATS_EVENT_STOP);
3055		bnx2x_save_statistics(bp);
3056	}
3057
3058	/* wait till consumers catch up with producers in all queues.
3059	 * If we're recovering, FW can't write to host so no reason
3060	 * to wait for the queues to complete all Tx.
3061	 */
3062	if (unload_mode != UNLOAD_RECOVERY)
3063		bnx2x_drain_tx_queues(bp);
3064
3065	/* if VF indicate to PF this function is going down (PF will delete sp
3066	 * elements and clear initializations
3067	 */
3068	if (IS_VF(bp)) {
3069		bnx2x_clear_vlan_info(bp);
3070		bnx2x_vfpf_close_vf(bp);
3071	} else if (unload_mode != UNLOAD_RECOVERY) {
3072		/* if this is a normal/close unload need to clean up chip*/
3073		bnx2x_chip_cleanup(bp, unload_mode, keep_link);
3074	} else {
3075		/* Send the UNLOAD_REQUEST to the MCP */
3076		bnx2x_send_unload_req(bp, unload_mode);
3077
3078		/* Prevent transactions to host from the functions on the
3079		 * engine that doesn't reset global blocks in case of global
3080		 * attention once global blocks are reset and gates are opened
3081		 * (the engine which leader will perform the recovery
3082		 * last).
3083		 */
3084		if (!CHIP_IS_E1x(bp))
3085			bnx2x_pf_disable(bp);
3086
3087		/* Disable HW interrupts, NAPI */
3088		bnx2x_netif_stop(bp, 1);
3089		/* Delete all NAPI objects */
3090		bnx2x_del_all_napi(bp);
3091		if (CNIC_LOADED(bp))
3092			bnx2x_del_all_napi_cnic(bp);
3093		/* Release IRQs */
3094		bnx2x_free_irq(bp);
3095
3096		/* Report UNLOAD_DONE to MCP */
3097		bnx2x_send_unload_done(bp, false);
3098	}
3099
3100	/*
3101	 * At this stage no more interrupts will arrive so we may safely clean
3102	 * the queueable objects here in case they failed to get cleaned so far.
3103	 */
3104	if (IS_PF(bp))
3105		bnx2x_squeeze_objects(bp);
3106
3107	/* There should be no more pending SP commands at this stage */
3108	bp->sp_state = 0;
3109
3110	bp->port.pmf = 0;
3111
3112	/* clear pending work in rtnl task */
3113	bp->sp_rtnl_state = 0;
3114	smp_mb();
3115
3116	/* Free SKBs, SGEs, TPA pool and driver internals */
3117	bnx2x_free_skbs(bp);
3118	if (CNIC_LOADED(bp))
3119		bnx2x_free_skbs_cnic(bp);
3120	for_each_rx_queue(bp, i)
3121		bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
3122
3123	bnx2x_free_fp_mem(bp);
3124	if (CNIC_LOADED(bp))
3125		bnx2x_free_fp_mem_cnic(bp);
3126
3127	if (IS_PF(bp)) {
3128		if (CNIC_LOADED(bp))
3129			bnx2x_free_mem_cnic(bp);
3130	}
3131	bnx2x_free_mem(bp);
3132
3133	bp->state = BNX2X_STATE_CLOSED;
3134	bp->cnic_loaded = false;
3135
3136	/* Clear driver version indication in shmem */
3137	if (IS_PF(bp) && !BP_NOMCP(bp))
3138		bnx2x_update_mng_version(bp);
3139
3140	/* Check if there are pending parity attentions. If there are - set
3141	 * RECOVERY_IN_PROGRESS.
3142	 */
3143	if (IS_PF(bp) && bnx2x_chk_parity_attn(bp, &global, false)) {
3144		bnx2x_set_reset_in_progress(bp);
3145
3146		/* Set RESET_IS_GLOBAL if needed */
3147		if (global)
3148			bnx2x_set_reset_global(bp);
3149	}
3150
3151	/* The last driver must disable a "close the gate" if there is no
3152	 * parity attention or "process kill" pending.
3153	 */
3154	if (IS_PF(bp) &&
3155	    !bnx2x_clear_pf_load(bp) &&
3156	    bnx2x_reset_is_done(bp, BP_PATH(bp)))
3157		bnx2x_disable_close_the_gate(bp);
3158
3159	DP(NETIF_MSG_IFUP, "Ending NIC unload\n");
3160
3161	return 0;
3162}
3163
3164int bnx2x_set_power_state(struct bnx2x *bp, pci_power_t state)
3165{
3166	u16 pmcsr;
3167
3168	/* If there is no power capability, silently succeed */
3169	if (!bp->pdev->pm_cap) {
3170		BNX2X_DEV_INFO("No power capability. Breaking.\n");
3171		return 0;
3172	}
3173
3174	pci_read_config_word(bp->pdev, bp->pdev->pm_cap + PCI_PM_CTRL, &pmcsr);
3175
3176	switch (state) {
3177	case PCI_D0:
3178		pci_write_config_word(bp->pdev, bp->pdev->pm_cap + PCI_PM_CTRL,
3179				      ((pmcsr & ~PCI_PM_CTRL_STATE_MASK) |
3180				       PCI_PM_CTRL_PME_STATUS));
3181
3182		if (pmcsr & PCI_PM_CTRL_STATE_MASK)
3183			/* delay required during transition out of D3hot */
3184			msleep(20);
3185		break;
3186
3187	case PCI_D3hot:
3188		/* If there are other clients above don't
3189		   shut down the power */
3190		if (atomic_read(&bp->pdev->enable_cnt) != 1)
3191			return 0;
3192		/* Don't shut down the power for emulation and FPGA */
3193		if (CHIP_REV_IS_SLOW(bp))
3194			return 0;
3195
3196		pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
3197		pmcsr |= 3;
3198
3199		if (bp->wol)
3200			pmcsr |= PCI_PM_CTRL_PME_ENABLE;
3201
3202		pci_write_config_word(bp->pdev, bp->pdev->pm_cap + PCI_PM_CTRL,
3203				      pmcsr);
3204
3205		/* No more memory access after this point until
3206		* device is brought back to D0.
3207		*/
3208		break;
3209
3210	default:
3211		dev_err(&bp->pdev->dev, "Can't support state = %d\n", state);
3212		return -EINVAL;
3213	}
3214	return 0;
3215}
3216
3217/*
3218 * net_device service functions
3219 */
3220static int bnx2x_poll(struct napi_struct *napi, int budget)
3221{
3222	struct bnx2x_fastpath *fp = container_of(napi, struct bnx2x_fastpath,
3223						 napi);
3224	struct bnx2x *bp = fp->bp;
3225	int rx_work_done;
3226	u8 cos;
3227
3228#ifdef BNX2X_STOP_ON_ERROR
3229	if (unlikely(bp->panic)) {
3230		napi_complete(napi);
3231		return 0;
3232	}
3233#endif
3234	for_each_cos_in_tx_queue(fp, cos)
3235		if (bnx2x_tx_queue_has_work(fp->txdata_ptr[cos]))
3236			bnx2x_tx_int(bp, fp->txdata_ptr[cos]);
3237
3238	rx_work_done = (bnx2x_has_rx_work(fp)) ? bnx2x_rx_int(fp, budget) : 0;
3239
3240	if (rx_work_done < budget) {
3241		/* No need to update SB for FCoE L2 ring as long as
3242		 * it's connected to the default SB and the SB
3243		 * has been updated when NAPI was scheduled.
3244		 */
3245		if (IS_FCOE_FP(fp)) {
3246			napi_complete_done(napi, rx_work_done);
3247		} else {
3248			bnx2x_update_fpsb_idx(fp);
3249			/* bnx2x_has_rx_work() reads the status block,
3250			 * thus we need to ensure that status block indices
3251			 * have been actually read (bnx2x_update_fpsb_idx)
3252			 * prior to this check (bnx2x_has_rx_work) so that
3253			 * we won't write the "newer" value of the status block
3254			 * to IGU (if there was a DMA right after
3255			 * bnx2x_has_rx_work and if there is no rmb, the memory
3256			 * reading (bnx2x_update_fpsb_idx) may be postponed
3257			 * to right before bnx2x_ack_sb). In this case there
3258			 * will never be another interrupt until there is
3259			 * another update of the status block, while there
3260			 * is still unhandled work.
3261			 */
3262			rmb();
3263
3264			if (!(bnx2x_has_rx_work(fp) || bnx2x_has_tx_work(fp))) {
3265				if (napi_complete_done(napi, rx_work_done)) {
3266					/* Re-enable interrupts */
3267					DP(NETIF_MSG_RX_STATUS,
3268					   "Update index to %d\n", fp->fp_hc_idx);
3269					bnx2x_ack_sb(bp, fp->igu_sb_id, USTORM_ID,
3270						     le16_to_cpu(fp->fp_hc_idx),
3271						     IGU_INT_ENABLE, 1);
3272				}
3273			} else {
3274				rx_work_done = budget;
3275			}
3276		}
3277	}
3278
3279	return rx_work_done;
3280}
3281
3282/* we split the first BD into headers and data BDs
3283 * to ease the pain of our fellow microcode engineers
3284 * we use one mapping for both BDs
3285 */
3286static u16 bnx2x_tx_split(struct bnx2x *bp,
3287			  struct bnx2x_fp_txdata *txdata,
3288			  struct sw_tx_bd *tx_buf,
3289			  struct eth_tx_start_bd **tx_bd, u16 hlen,
3290			  u16 bd_prod)
3291{
3292	struct eth_tx_start_bd *h_tx_bd = *tx_bd;
3293	struct eth_tx_bd *d_tx_bd;
3294	dma_addr_t mapping;
3295	int old_len = le16_to_cpu(h_tx_bd->nbytes);
3296
3297	/* first fix first BD */
3298	h_tx_bd->nbytes = cpu_to_le16(hlen);
3299
3300	DP(NETIF_MSG_TX_QUEUED,	"TSO split header size is %d (%x:%x)\n",
3301	   h_tx_bd->nbytes, h_tx_bd->addr_hi, h_tx_bd->addr_lo);
3302
3303	/* now get a new data BD
3304	 * (after the pbd) and fill it */
3305	bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3306	d_tx_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
3307
3308	mapping = HILO_U64(le32_to_cpu(h_tx_bd->addr_hi),
3309			   le32_to_cpu(h_tx_bd->addr_lo)) + hlen;
3310
3311	d_tx_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
3312	d_tx_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
3313	d_tx_bd->nbytes = cpu_to_le16(old_len - hlen);
3314
3315	/* this marks the BD as one that has no individual mapping */
3316	tx_buf->flags |= BNX2X_TSO_SPLIT_BD;
3317
3318	DP(NETIF_MSG_TX_QUEUED,
3319	   "TSO split data size is %d (%x:%x)\n",
3320	   d_tx_bd->nbytes, d_tx_bd->addr_hi, d_tx_bd->addr_lo);
3321
3322	/* update tx_bd */
3323	*tx_bd = (struct eth_tx_start_bd *)d_tx_bd;
3324
3325	return bd_prod;
3326}
3327
3328#define bswab32(b32) ((__force __le32) swab32((__force __u32) (b32)))
3329#define bswab16(b16) ((__force __le16) swab16((__force __u16) (b16)))
3330static __le16 bnx2x_csum_fix(unsigned char *t_header, u16 csum, s8 fix)
3331{
3332	__sum16 tsum = (__force __sum16) csum;
3333
3334	if (fix > 0)
3335		tsum = ~csum_fold(csum_sub((__force __wsum) csum,
3336				  csum_partial(t_header - fix, fix, 0)));
3337
3338	else if (fix < 0)
3339		tsum = ~csum_fold(csum_add((__force __wsum) csum,
3340				  csum_partial(t_header, -fix, 0)));
3341
3342	return bswab16(tsum);
3343}
3344
3345static u32 bnx2x_xmit_type(struct bnx2x *bp, struct sk_buff *skb)
3346{
3347	u32 rc;
3348	__u8 prot = 0;
3349	__be16 protocol;
3350
3351	if (skb->ip_summed != CHECKSUM_PARTIAL)
3352		return XMIT_PLAIN;
3353
3354	protocol = vlan_get_protocol(skb);
3355	if (protocol == htons(ETH_P_IPV6)) {
3356		rc = XMIT_CSUM_V6;
3357		prot = ipv6_hdr(skb)->nexthdr;
3358	} else {
3359		rc = XMIT_CSUM_V4;
3360		prot = ip_hdr(skb)->protocol;
3361	}
3362
3363	if (!CHIP_IS_E1x(bp) && skb->encapsulation) {
3364		if (inner_ip_hdr(skb)->version == 6) {
3365			rc |= XMIT_CSUM_ENC_V6;
3366			if (inner_ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3367				rc |= XMIT_CSUM_TCP;
3368		} else {
3369			rc |= XMIT_CSUM_ENC_V4;
3370			if (inner_ip_hdr(skb)->protocol == IPPROTO_TCP)
3371				rc |= XMIT_CSUM_TCP;
3372		}
3373	}
3374	if (prot == IPPROTO_TCP)
3375		rc |= XMIT_CSUM_TCP;
3376
3377	if (skb_is_gso(skb)) {
3378		if (skb_is_gso_v6(skb)) {
3379			rc |= (XMIT_GSO_V6 | XMIT_CSUM_TCP);
3380			if (rc & XMIT_CSUM_ENC)
3381				rc |= XMIT_GSO_ENC_V6;
3382		} else {
3383			rc |= (XMIT_GSO_V4 | XMIT_CSUM_TCP);
3384			if (rc & XMIT_CSUM_ENC)
3385				rc |= XMIT_GSO_ENC_V4;
3386		}
3387	}
3388
3389	return rc;
3390}
3391
3392/* VXLAN: 4 = 1 (for linear data BD) + 3 (2 for PBD and last BD) */
3393#define BNX2X_NUM_VXLAN_TSO_WIN_SUB_BDS         4
3394
3395/* Regular: 3 = 1 (for linear data BD) + 2 (for PBD and last BD) */
3396#define BNX2X_NUM_TSO_WIN_SUB_BDS               3
3397
3398#if (MAX_SKB_FRAGS >= MAX_FETCH_BD - BDS_PER_TX_PKT)
3399/* check if packet requires linearization (packet is too fragmented)
3400   no need to check fragmentation if page size > 8K (there will be no
3401   violation to FW restrictions) */
3402static int bnx2x_pkt_req_lin(struct bnx2x *bp, struct sk_buff *skb,
3403			     u32 xmit_type)
3404{
3405	int first_bd_sz = 0, num_tso_win_sub = BNX2X_NUM_TSO_WIN_SUB_BDS;
3406	int to_copy = 0, hlen = 0;
3407
3408	if (xmit_type & XMIT_GSO_ENC)
3409		num_tso_win_sub = BNX2X_NUM_VXLAN_TSO_WIN_SUB_BDS;
3410
3411	if (skb_shinfo(skb)->nr_frags >= (MAX_FETCH_BD - num_tso_win_sub)) {
3412		if (xmit_type & XMIT_GSO) {
3413			unsigned short lso_mss = skb_shinfo(skb)->gso_size;
3414			int wnd_size = MAX_FETCH_BD - num_tso_win_sub;
3415			/* Number of windows to check */
3416			int num_wnds = skb_shinfo(skb)->nr_frags - wnd_size;
3417			int wnd_idx = 0;
3418			int frag_idx = 0;
3419			u32 wnd_sum = 0;
3420
3421			/* Headers length */
3422			if (xmit_type & XMIT_GSO_ENC)
3423				hlen = (int)(skb_inner_transport_header(skb) -
3424					     skb->data) +
3425					     inner_tcp_hdrlen(skb);
3426			else
3427				hlen = (int)(skb_transport_header(skb) -
3428					     skb->data) + tcp_hdrlen(skb);
3429
3430			/* Amount of data (w/o headers) on linear part of SKB*/
3431			first_bd_sz = skb_headlen(skb) - hlen;
3432
3433			wnd_sum  = first_bd_sz;
3434
3435			/* Calculate the first sum - it's special */
3436			for (frag_idx = 0; frag_idx < wnd_size - 1; frag_idx++)
3437				wnd_sum +=
3438					skb_frag_size(&skb_shinfo(skb)->frags[frag_idx]);
3439
3440			/* If there was data on linear skb data - check it */
3441			if (first_bd_sz > 0) {
3442				if (unlikely(wnd_sum < lso_mss)) {
3443					to_copy = 1;
3444					goto exit_lbl;
3445				}
3446
3447				wnd_sum -= first_bd_sz;
3448			}
3449
3450			/* Others are easier: run through the frag list and
3451			   check all windows */
3452			for (wnd_idx = 0; wnd_idx <= num_wnds; wnd_idx++) {
3453				wnd_sum +=
3454			  skb_frag_size(&skb_shinfo(skb)->frags[wnd_idx + wnd_size - 1]);
3455
3456				if (unlikely(wnd_sum < lso_mss)) {
3457					to_copy = 1;
3458					break;
3459				}
3460				wnd_sum -=
3461					skb_frag_size(&skb_shinfo(skb)->frags[wnd_idx]);
3462			}
3463		} else {
3464			/* in non-LSO too fragmented packet should always
3465			   be linearized */
3466			to_copy = 1;
3467		}
3468	}
3469
3470exit_lbl:
3471	if (unlikely(to_copy))
3472		DP(NETIF_MSG_TX_QUEUED,
3473		   "Linearization IS REQUIRED for %s packet. num_frags %d  hlen %d  first_bd_sz %d\n",
3474		   (xmit_type & XMIT_GSO) ? "LSO" : "non-LSO",
3475		   skb_shinfo(skb)->nr_frags, hlen, first_bd_sz);
3476
3477	return to_copy;
3478}
3479#endif
3480
3481/**
3482 * bnx2x_set_pbd_gso - update PBD in GSO case.
3483 *
3484 * @skb:	packet skb
3485 * @pbd:	parse BD
3486 * @xmit_type:	xmit flags
3487 */
3488static void bnx2x_set_pbd_gso(struct sk_buff *skb,
3489			      struct eth_tx_parse_bd_e1x *pbd,
3490			      u32 xmit_type)
3491{
3492	pbd->lso_mss = cpu_to_le16(skb_shinfo(skb)->gso_size);
3493	pbd->tcp_send_seq = bswab32(tcp_hdr(skb)->seq);
3494	pbd->tcp_flags = pbd_tcp_flags(tcp_hdr(skb));
3495
3496	if (xmit_type & XMIT_GSO_V4) {
3497		pbd->ip_id = bswab16(ip_hdr(skb)->id);
3498		pbd->tcp_pseudo_csum =
3499			bswab16(~csum_tcpudp_magic(ip_hdr(skb)->saddr,
3500						   ip_hdr(skb)->daddr,
3501						   0, IPPROTO_TCP, 0));
3502	} else {
3503		pbd->tcp_pseudo_csum =
3504			bswab16(~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3505						 &ipv6_hdr(skb)->daddr,
3506						 0, IPPROTO_TCP, 0));
3507	}
3508
3509	pbd->global_data |=
3510		cpu_to_le16(ETH_TX_PARSE_BD_E1X_PSEUDO_CS_WITHOUT_LEN);
3511}
3512
3513/**
3514 * bnx2x_set_pbd_csum_enc - update PBD with checksum and return header length
3515 *
3516 * @bp:			driver handle
3517 * @skb:		packet skb
3518 * @parsing_data:	data to be updated
3519 * @xmit_type:		xmit flags
3520 *
3521 * 57712/578xx related, when skb has encapsulation
3522 */
3523static u8 bnx2x_set_pbd_csum_enc(struct bnx2x *bp, struct sk_buff *skb,
3524				 u32 *parsing_data, u32 xmit_type)
3525{
3526	*parsing_data |=
3527		((((u8 *)skb_inner_transport_header(skb) - skb->data) >> 1) <<
3528		ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W_SHIFT) &
3529		ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W;
3530
3531	if (xmit_type & XMIT_CSUM_TCP) {
3532		*parsing_data |= ((inner_tcp_hdrlen(skb) / 4) <<
3533			ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW_SHIFT) &
3534			ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW;
3535
3536		return skb_inner_transport_header(skb) +
3537			inner_tcp_hdrlen(skb) - skb->data;
3538	}
3539
3540	/* We support checksum offload for TCP and UDP only.
3541	 * No need to pass the UDP header length - it's a constant.
3542	 */
3543	return skb_inner_transport_header(skb) +
3544		sizeof(struct udphdr) - skb->data;
3545}
3546
3547/**
3548 * bnx2x_set_pbd_csum_e2 - update PBD with checksum and return header length
3549 *
3550 * @bp:			driver handle
3551 * @skb:		packet skb
3552 * @parsing_data:	data to be updated
3553 * @xmit_type:		xmit flags
3554 *
3555 * 57712/578xx related
3556 */
3557static u8 bnx2x_set_pbd_csum_e2(struct bnx2x *bp, struct sk_buff *skb,
3558				u32 *parsing_data, u32 xmit_type)
3559{
3560	*parsing_data |=
3561		((((u8 *)skb_transport_header(skb) - skb->data) >> 1) <<
3562		ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W_SHIFT) &
3563		ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W;
3564
3565	if (xmit_type & XMIT_CSUM_TCP) {
3566		*parsing_data |= ((tcp_hdrlen(skb) / 4) <<
3567			ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW_SHIFT) &
3568			ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW;
3569
3570		return skb_transport_header(skb) + tcp_hdrlen(skb) - skb->data;
3571	}
3572	/* We support checksum offload for TCP and UDP only.
3573	 * No need to pass the UDP header length - it's a constant.
3574	 */
3575	return skb_transport_header(skb) + sizeof(struct udphdr) - skb->data;
3576}
3577
3578/* set FW indication according to inner or outer protocols if tunneled */
3579static void bnx2x_set_sbd_csum(struct bnx2x *bp, struct sk_buff *skb,
3580			       struct eth_tx_start_bd *tx_start_bd,
3581			       u32 xmit_type)
3582{
3583	tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_L4_CSUM;
3584
3585	if (xmit_type & (XMIT_CSUM_ENC_V6 | XMIT_CSUM_V6))
3586		tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_IPV6;
3587
3588	if (!(xmit_type & XMIT_CSUM_TCP))
3589		tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_IS_UDP;
3590}
3591
3592/**
3593 * bnx2x_set_pbd_csum - update PBD with checksum and return header length
3594 *
3595 * @bp:		driver handle
3596 * @skb:	packet skb
3597 * @pbd:	parse BD to be updated
3598 * @xmit_type:	xmit flags
3599 */
3600static u8 bnx2x_set_pbd_csum(struct bnx2x *bp, struct sk_buff *skb,
3601			     struct eth_tx_parse_bd_e1x *pbd,
3602			     u32 xmit_type)
3603{
3604	u8 hlen = (skb_network_header(skb) - skb->data) >> 1;
3605
3606	/* for now NS flag is not used in Linux */
3607	pbd->global_data =
3608		cpu_to_le16(hlen |
3609			    ((skb->protocol == cpu_to_be16(ETH_P_8021Q)) <<
3610			     ETH_TX_PARSE_BD_E1X_LLC_SNAP_EN_SHIFT));
3611
3612	pbd->ip_hlen_w = (skb_transport_header(skb) -
3613			skb_network_header(skb)) >> 1;
3614
3615	hlen += pbd->ip_hlen_w;
3616
3617	/* We support checksum offload for TCP and UDP only */
3618	if (xmit_type & XMIT_CSUM_TCP)
3619		hlen += tcp_hdrlen(skb) / 2;
3620	else
3621		hlen += sizeof(struct udphdr) / 2;
3622
3623	pbd->total_hlen_w = cpu_to_le16(hlen);
3624	hlen = hlen*2;
3625
3626	if (xmit_type & XMIT_CSUM_TCP) {
3627		pbd->tcp_pseudo_csum = bswab16(tcp_hdr(skb)->check);
3628
3629	} else {
3630		s8 fix = SKB_CS_OFF(skb); /* signed! */
3631
3632		DP(NETIF_MSG_TX_QUEUED,
3633		   "hlen %d  fix %d  csum before fix %x\n",
3634		   le16_to_cpu(pbd->total_hlen_w), fix, SKB_CS(skb));
3635
3636		/* HW bug: fixup the CSUM */
3637		pbd->tcp_pseudo_csum =
3638			bnx2x_csum_fix(skb_transport_header(skb),
3639				       SKB_CS(skb), fix);
3640
3641		DP(NETIF_MSG_TX_QUEUED, "csum after fix %x\n",
3642		   pbd->tcp_pseudo_csum);
3643	}
3644
3645	return hlen;
3646}
3647
3648static void bnx2x_update_pbds_gso_enc(struct sk_buff *skb,
3649				      struct eth_tx_parse_bd_e2 *pbd_e2,
3650				      struct eth_tx_parse_2nd_bd *pbd2,
3651				      u16 *global_data,
3652				      u32 xmit_type)
3653{
3654	u16 hlen_w = 0;
3655	u8 outerip_off, outerip_len = 0;
3656
3657	/* from outer IP to transport */
3658	hlen_w = (skb_inner_transport_header(skb) -
3659		  skb_network_header(skb)) >> 1;
3660
3661	/* transport len */
3662	hlen_w += inner_tcp_hdrlen(skb) >> 1;
3663
3664	pbd2->fw_ip_hdr_to_payload_w = hlen_w;
3665
3666	/* outer IP header info */
3667	if (xmit_type & XMIT_CSUM_V4) {
3668		struct iphdr *iph = ip_hdr(skb);
3669		u32 csum = (__force u32)(~iph->check) -
3670			   (__force u32)iph->tot_len -
3671			   (__force u32)iph->frag_off;
3672
3673		outerip_len = iph->ihl << 1;
3674
3675		pbd2->fw_ip_csum_wo_len_flags_frag =
3676			bswab16(csum_fold((__force __wsum)csum));
3677	} else {
3678		pbd2->fw_ip_hdr_to_payload_w =
3679			hlen_w - ((sizeof(struct ipv6hdr)) >> 1);
3680		pbd_e2->data.tunnel_data.flags |=
3681			ETH_TUNNEL_DATA_IPV6_OUTER;
3682	}
3683
3684	pbd2->tcp_send_seq = bswab32(inner_tcp_hdr(skb)->seq);
3685
3686	pbd2->tcp_flags = pbd_tcp_flags(inner_tcp_hdr(skb));
3687
3688	/* inner IP header info */
3689	if (xmit_type & XMIT_CSUM_ENC_V4) {
3690		pbd2->hw_ip_id = bswab16(inner_ip_hdr(skb)->id);
3691
3692		pbd_e2->data.tunnel_data.pseudo_csum =
3693			bswab16(~csum_tcpudp_magic(
3694					inner_ip_hdr(skb)->saddr,
3695					inner_ip_hdr(skb)->daddr,
3696					0, IPPROTO_TCP, 0));
3697	} else {
3698		pbd_e2->data.tunnel_data.pseudo_csum =
3699			bswab16(~csum_ipv6_magic(
3700					&inner_ipv6_hdr(skb)->saddr,
3701					&inner_ipv6_hdr(skb)->daddr,
3702					0, IPPROTO_TCP, 0));
3703	}
3704
3705	outerip_off = (skb_network_header(skb) - skb->data) >> 1;
3706
3707	*global_data |=
3708		outerip_off |
3709		(outerip_len <<
3710			ETH_TX_PARSE_2ND_BD_IP_HDR_LEN_OUTER_W_SHIFT) |
3711		((skb->protocol == cpu_to_be16(ETH_P_8021Q)) <<
3712			ETH_TX_PARSE_2ND_BD_LLC_SNAP_EN_SHIFT);
3713
3714	if (ip_hdr(skb)->protocol == IPPROTO_UDP) {
3715		SET_FLAG(*global_data, ETH_TX_PARSE_2ND_BD_TUNNEL_UDP_EXIST, 1);
3716		pbd2->tunnel_udp_hdr_start_w = skb_transport_offset(skb) >> 1;
3717	}
3718}
3719
3720static inline void bnx2x_set_ipv6_ext_e2(struct sk_buff *skb, u32 *parsing_data,
3721					 u32 xmit_type)
3722{
3723	struct ipv6hdr *ipv6;
3724
3725	if (!(xmit_type & (XMIT_GSO_ENC_V6 | XMIT_GSO_V6)))
3726		return;
3727
3728	if (xmit_type & XMIT_GSO_ENC_V6)
3729		ipv6 = inner_ipv6_hdr(skb);
3730	else /* XMIT_GSO_V6 */
3731		ipv6 = ipv6_hdr(skb);
3732
3733	if (ipv6->nexthdr == NEXTHDR_IPV6)
3734		*parsing_data |= ETH_TX_PARSE_BD_E2_IPV6_WITH_EXT_HDR;
3735}
3736
3737/* called with netif_tx_lock
3738 * bnx2x_tx_int() runs without netif_tx_lock unless it needs to call
3739 * netif_wake_queue()
3740 */
3741netdev_tx_t bnx2x_start_xmit(struct sk_buff *skb, struct net_device *dev)
3742{
3743	struct bnx2x *bp = netdev_priv(dev);
3744
3745	struct netdev_queue *txq;
3746	struct bnx2x_fp_txdata *txdata;
3747	struct sw_tx_bd *tx_buf;
3748	struct eth_tx_start_bd *tx_start_bd, *first_bd;
3749	struct eth_tx_bd *tx_data_bd, *total_pkt_bd = NULL;
3750	struct eth_tx_parse_bd_e1x *pbd_e1x = NULL;
3751	struct eth_tx_parse_bd_e2 *pbd_e2 = NULL;
3752	struct eth_tx_parse_2nd_bd *pbd2 = NULL;
3753	u32 pbd_e2_parsing_data = 0;
3754	u16 pkt_prod, bd_prod;
3755	int nbd, txq_index;
3756	dma_addr_t mapping;
3757	u32 xmit_type = bnx2x_xmit_type(bp, skb);
3758	int i;
3759	u8 hlen = 0;
3760	__le16 pkt_size = 0;
3761	struct ethhdr *eth;
3762	u8 mac_type = UNICAST_ADDRESS;
3763
3764#ifdef BNX2X_STOP_ON_ERROR
3765	if (unlikely(bp->panic))
3766		return NETDEV_TX_BUSY;
3767#endif
3768
3769	txq_index = skb_get_queue_mapping(skb);
3770	txq = netdev_get_tx_queue(dev, txq_index);
3771
3772	BUG_ON(txq_index >= MAX_ETH_TXQ_IDX(bp) + (CNIC_LOADED(bp) ? 1 : 0));
3773
3774	txdata = &bp->bnx2x_txq[txq_index];
3775
3776	/* enable this debug print to view the transmission queue being used
3777	DP(NETIF_MSG_TX_QUEUED, "indices: txq %d, fp %d, txdata %d\n",
3778	   txq_index, fp_index, txdata_index); */
3779
3780	/* enable this debug print to view the transmission details
3781	DP(NETIF_MSG_TX_QUEUED,
3782	   "transmitting packet cid %d fp index %d txdata_index %d tx_data ptr %p fp pointer %p\n",
3783	   txdata->cid, fp_index, txdata_index, txdata, fp); */
3784
3785	if (unlikely(bnx2x_tx_avail(bp, txdata) <
3786			skb_shinfo(skb)->nr_frags +
3787			BDS_PER_TX_PKT +
3788			NEXT_CNT_PER_TX_PKT(MAX_BDS_PER_TX_PKT))) {
3789		/* Handle special storage cases separately */
3790		if (txdata->tx_ring_size == 0) {
3791			struct bnx2x_eth_q_stats *q_stats =
3792				bnx2x_fp_qstats(bp, txdata->parent_fp);
3793			q_stats->driver_filtered_tx_pkt++;
3794			dev_kfree_skb(skb);
3795			return NETDEV_TX_OK;
3796		}
3797		bnx2x_fp_qstats(bp, txdata->parent_fp)->driver_xoff++;
3798		netif_tx_stop_queue(txq);
3799		BNX2X_ERR("BUG! Tx ring full when queue awake!\n");
3800
3801		return NETDEV_TX_BUSY;
3802	}
3803
3804	DP(NETIF_MSG_TX_QUEUED,
3805	   "queue[%d]: SKB: summed %x  protocol %x protocol(%x,%x) gso type %x  xmit_type %x len %d\n",
3806	   txq_index, skb->ip_summed, skb->protocol, ipv6_hdr(skb)->nexthdr,
3807	   ip_hdr(skb)->protocol, skb_shinfo(skb)->gso_type, xmit_type,
3808	   skb->len);
3809
3810	eth = (struct ethhdr *)skb->data;
3811
3812	/* set flag according to packet type (UNICAST_ADDRESS is default)*/
3813	if (unlikely(is_multicast_ether_addr(eth->h_dest))) {
3814		if (is_broadcast_ether_addr(eth->h_dest))
3815			mac_type = BROADCAST_ADDRESS;
3816		else
3817			mac_type = MULTICAST_ADDRESS;
3818	}
3819
3820#if (MAX_SKB_FRAGS >= MAX_FETCH_BD - BDS_PER_TX_PKT)
3821	/* First, check if we need to linearize the skb (due to FW
3822	   restrictions). No need to check fragmentation if page size > 8K
3823	   (there will be no violation to FW restrictions) */
3824	if (bnx2x_pkt_req_lin(bp, skb, xmit_type)) {
3825		/* Statistics of linearization */
3826		bp->lin_cnt++;
3827		if (skb_linearize(skb) != 0) {
3828			DP(NETIF_MSG_TX_QUEUED,
3829			   "SKB linearization failed - silently dropping this SKB\n");
3830			dev_kfree_skb_any(skb);
3831			return NETDEV_TX_OK;
3832		}
3833	}
3834#endif
3835	/* Map skb linear data for DMA */
3836	mapping = dma_map_single(&bp->pdev->dev, skb->data,
3837				 skb_headlen(skb), DMA_TO_DEVICE);
3838	if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
3839		DP(NETIF_MSG_TX_QUEUED,
3840		   "SKB mapping failed - silently dropping this SKB\n");
3841		dev_kfree_skb_any(skb);
3842		return NETDEV_TX_OK;
3843	}
3844	/*
3845	Please read carefully. First we use one BD which we mark as start,
3846	then we have a parsing info BD (used for TSO or xsum),
3847	and only then we have the rest of the TSO BDs.
3848	(don't forget to mark the last one as last,
3849	and to unmap only AFTER you write to the BD ...)
3850	And above all, all pdb sizes are in words - NOT DWORDS!
3851	*/
3852
3853	/* get current pkt produced now - advance it just before sending packet
3854	 * since mapping of pages may fail and cause packet to be dropped
3855	 */
3856	pkt_prod = txdata->tx_pkt_prod;
3857	bd_prod = TX_BD(txdata->tx_bd_prod);
3858
3859	/* get a tx_buf and first BD
3860	 * tx_start_bd may be changed during SPLIT,
3861	 * but first_bd will always stay first
3862	 */
3863	tx_buf = &txdata->tx_buf_ring[TX_BD(pkt_prod)];
3864	tx_start_bd = &txdata->tx_desc_ring[bd_prod].start_bd;
3865	first_bd = tx_start_bd;
3866
3867	tx_start_bd->bd_flags.as_bitfield = ETH_TX_BD_FLAGS_START_BD;
3868
3869	if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) {
3870		if (!(bp->flags & TX_TIMESTAMPING_EN)) {
3871			bp->eth_stats.ptp_skip_tx_ts++;
3872			BNX2X_ERR("Tx timestamping was not enabled, this packet will not be timestamped\n");
3873		} else if (bp->ptp_tx_skb) {
3874			bp->eth_stats.ptp_skip_tx_ts++;
3875			netdev_err_once(bp->dev,
3876					"Device supports only a single outstanding packet to timestamp, this packet won't be timestamped\n");
3877		} else {
3878			skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
3879			/* schedule check for Tx timestamp */
3880			bp->ptp_tx_skb = skb_get(skb);
3881			bp->ptp_tx_start = jiffies;
3882			schedule_work(&bp->ptp_task);
3883		}
3884	}
3885
3886	/* header nbd: indirectly zero other flags! */
3887	tx_start_bd->general_data = 1 << ETH_TX_START_BD_HDR_NBDS_SHIFT;
3888
3889	/* remember the first BD of the packet */
3890	tx_buf->first_bd = txdata->tx_bd_prod;
3891	tx_buf->skb = skb;
3892	tx_buf->flags = 0;
3893
3894	DP(NETIF_MSG_TX_QUEUED,
3895	   "sending pkt %u @%p  next_idx %u  bd %u @%p\n",
3896	   pkt_prod, tx_buf, txdata->tx_pkt_prod, bd_prod, tx_start_bd);
3897
3898	if (skb_vlan_tag_present(skb)) {
3899		tx_start_bd->vlan_or_ethertype =
3900		    cpu_to_le16(skb_vlan_tag_get(skb));
3901		tx_start_bd->bd_flags.as_bitfield |=
3902		    (X_ETH_OUTBAND_VLAN << ETH_TX_BD_FLAGS_VLAN_MODE_SHIFT);
3903	} else {
3904		/* when transmitting in a vf, start bd must hold the ethertype
3905		 * for fw to enforce it
3906		 */
3907		u16 vlan_tci = 0;
3908#ifndef BNX2X_STOP_ON_ERROR
3909		if (IS_VF(bp)) {
3910#endif
3911			/* Still need to consider inband vlan for enforced */
3912			if (__vlan_get_tag(skb, &vlan_tci)) {
3913				tx_start_bd->vlan_or_ethertype =
3914					cpu_to_le16(ntohs(eth->h_proto));
3915			} else {
3916				tx_start_bd->bd_flags.as_bitfield |=
3917					(X_ETH_INBAND_VLAN <<
3918					 ETH_TX_BD_FLAGS_VLAN_MODE_SHIFT);
3919				tx_start_bd->vlan_or_ethertype =
3920					cpu_to_le16(vlan_tci);
3921			}
3922#ifndef BNX2X_STOP_ON_ERROR
3923		} else {
3924			/* used by FW for packet accounting */
3925			tx_start_bd->vlan_or_ethertype = cpu_to_le16(pkt_prod);
3926		}
3927#endif
3928	}
3929
3930	nbd = 2; /* start_bd + pbd + frags (updated when pages are mapped) */
3931
3932	/* turn on parsing and get a BD */
3933	bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3934
3935	if (xmit_type & XMIT_CSUM)
3936		bnx2x_set_sbd_csum(bp, skb, tx_start_bd, xmit_type);
3937
3938	if (!CHIP_IS_E1x(bp)) {
3939		pbd_e2 = &txdata->tx_desc_ring[bd_prod].parse_bd_e2;
3940		memset(pbd_e2, 0, sizeof(struct eth_tx_parse_bd_e2));
3941
3942		if (xmit_type & XMIT_CSUM_ENC) {
3943			u16 global_data = 0;
3944
3945			/* Set PBD in enc checksum offload case */
3946			hlen = bnx2x_set_pbd_csum_enc(bp, skb,
3947						      &pbd_e2_parsing_data,
3948						      xmit_type);
3949
3950			/* turn on 2nd parsing and get a BD */
3951			bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3952
3953			pbd2 = &txdata->tx_desc_ring[bd_prod].parse_2nd_bd;
3954
3955			memset(pbd2, 0, sizeof(*pbd2));
3956
3957			pbd_e2->data.tunnel_data.ip_hdr_start_inner_w =
3958				(skb_inner_network_header(skb) -
3959				 skb->data) >> 1;
3960
3961			if (xmit_type & XMIT_GSO_ENC)
3962				bnx2x_update_pbds_gso_enc(skb, pbd_e2, pbd2,
3963							  &global_data,
3964							  xmit_type);
3965
3966			pbd2->global_data = cpu_to_le16(global_data);
3967
3968			/* add addition parse BD indication to start BD */
3969			SET_FLAG(tx_start_bd->general_data,
3970				 ETH_TX_START_BD_PARSE_NBDS, 1);
3971			/* set encapsulation flag in start BD */
3972			SET_FLAG(tx_start_bd->general_data,
3973				 ETH_TX_START_BD_TUNNEL_EXIST, 1);
3974
3975			tx_buf->flags |= BNX2X_HAS_SECOND_PBD;
3976
3977			nbd++;
3978		} else if (xmit_type & XMIT_CSUM) {
3979			/* Set PBD in checksum offload case w/o encapsulation */
3980			hlen = bnx2x_set_pbd_csum_e2(bp, skb,
3981						     &pbd_e2_parsing_data,
3982						     xmit_type);
3983		}
3984
3985		bnx2x_set_ipv6_ext_e2(skb, &pbd_e2_parsing_data, xmit_type);
3986		/* Add the macs to the parsing BD if this is a vf or if
3987		 * Tx Switching is enabled.
3988		 */
3989		if (IS_VF(bp)) {
3990			/* override GRE parameters in BD */
3991			bnx2x_set_fw_mac_addr(&pbd_e2->data.mac_addr.src_hi,
3992					      &pbd_e2->data.mac_addr.src_mid,
3993					      &pbd_e2->data.mac_addr.src_lo,
3994					      eth->h_source);
3995
3996			bnx2x_set_fw_mac_addr(&pbd_e2->data.mac_addr.dst_hi,
3997					      &pbd_e2->data.mac_addr.dst_mid,
3998					      &pbd_e2->data.mac_addr.dst_lo,
3999					      eth->h_dest);
4000		} else {
4001			if (bp->flags & TX_SWITCHING)
4002				bnx2x_set_fw_mac_addr(
4003						&pbd_e2->data.mac_addr.dst_hi,
4004						&pbd_e2->data.mac_addr.dst_mid,
4005						&pbd_e2->data.mac_addr.dst_lo,
4006						eth->h_dest);
4007#ifdef BNX2X_STOP_ON_ERROR
4008			/* Enforce security is always set in Stop on Error -
4009			 * source mac should be present in the parsing BD
4010			 */
4011			bnx2x_set_fw_mac_addr(&pbd_e2->data.mac_addr.src_hi,
4012					      &pbd_e2->data.mac_addr.src_mid,
4013					      &pbd_e2->data.mac_addr.src_lo,
4014					      eth->h_source);
4015#endif
4016		}
4017
4018		SET_FLAG(pbd_e2_parsing_data,
4019			 ETH_TX_PARSE_BD_E2_ETH_ADDR_TYPE, mac_type);
4020	} else {
4021		u16 global_data = 0;
4022		pbd_e1x = &txdata->tx_desc_ring[bd_prod].parse_bd_e1x;
4023		memset(pbd_e1x, 0, sizeof(struct eth_tx_parse_bd_e1x));
4024		/* Set PBD in checksum offload case */
4025		if (xmit_type & XMIT_CSUM)
4026			hlen = bnx2x_set_pbd_csum(bp, skb, pbd_e1x, xmit_type);
4027
4028		SET_FLAG(global_data,
4029			 ETH_TX_PARSE_BD_E1X_ETH_ADDR_TYPE, mac_type);
4030		pbd_e1x->global_data |= cpu_to_le16(global_data);
4031	}
4032
4033	/* Setup the data pointer of the first BD of the packet */
4034	tx_start_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
4035	tx_start_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
4036	tx_start_bd->nbytes = cpu_to_le16(skb_headlen(skb));
4037	pkt_size = tx_start_bd->nbytes;
4038
4039	DP(NETIF_MSG_TX_QUEUED,
4040	   "first bd @%p  addr (%x:%x)  nbytes %d  flags %x  vlan %x\n",
4041	   tx_start_bd, tx_start_bd->addr_hi, tx_start_bd->addr_lo,
4042	   le16_to_cpu(tx_start_bd->nbytes),
4043	   tx_start_bd->bd_flags.as_bitfield,
4044	   le16_to_cpu(tx_start_bd->vlan_or_ethertype));
4045
4046	if (xmit_type & XMIT_GSO) {
4047
4048		DP(NETIF_MSG_TX_QUEUED,
4049		   "TSO packet len %d  hlen %d  total len %d  tso size %d\n",
4050		   skb->len, hlen, skb_headlen(skb),
4051		   skb_shinfo(skb)->gso_size);
4052
4053		tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_SW_LSO;
4054
4055		if (unlikely(skb_headlen(skb) > hlen)) {
4056			nbd++;
4057			bd_prod = bnx2x_tx_split(bp, txdata, tx_buf,
4058						 &tx_start_bd, hlen,
4059						 bd_prod);
4060		}
4061		if (!CHIP_IS_E1x(bp))
4062			pbd_e2_parsing_data |=
4063				(skb_shinfo(skb)->gso_size <<
4064				 ETH_TX_PARSE_BD_E2_LSO_MSS_SHIFT) &
4065				 ETH_TX_PARSE_BD_E2_LSO_MSS;
4066		else
4067			bnx2x_set_pbd_gso(skb, pbd_e1x, xmit_type);
4068	}
4069
4070	/* Set the PBD's parsing_data field if not zero
4071	 * (for the chips newer than 57711).
4072	 */
4073	if (pbd_e2_parsing_data)
4074		pbd_e2->parsing_data = cpu_to_le32(pbd_e2_parsing_data);
4075
4076	tx_data_bd = (struct eth_tx_bd *)tx_start_bd;
4077
4078	/* Handle fragmented skb */
4079	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
4080		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
4081
4082		mapping = skb_frag_dma_map(&bp->pdev->dev, frag, 0,
4083					   skb_frag_size(frag), DMA_TO_DEVICE);
4084		if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
4085			unsigned int pkts_compl = 0, bytes_compl = 0;
4086
4087			DP(NETIF_MSG_TX_QUEUED,
4088			   "Unable to map page - dropping packet...\n");
4089
4090			/* we need unmap all buffers already mapped
4091			 * for this SKB;
4092			 * first_bd->nbd need to be properly updated
4093			 * before call to bnx2x_free_tx_pkt
4094			 */
4095			first_bd->nbd = cpu_to_le16(nbd);
4096			bnx2x_free_tx_pkt(bp, txdata,
4097					  TX_BD(txdata->tx_pkt_prod),
4098					  &pkts_compl, &bytes_compl);
4099			return NETDEV_TX_OK;
4100		}
4101
4102		bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
4103		tx_data_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
4104		if (total_pkt_bd == NULL)
4105			total_pkt_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
4106
4107		tx_data_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
4108		tx_data_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
4109		tx_data_bd->nbytes = cpu_to_le16(skb_frag_size(frag));
4110		le16_add_cpu(&pkt_size, skb_frag_size(frag));
4111		nbd++;
4112
4113		DP(NETIF_MSG_TX_QUEUED,
4114		   "frag %d  bd @%p  addr (%x:%x)  nbytes %d\n",
4115		   i, tx_data_bd, tx_data_bd->addr_hi, tx_data_bd->addr_lo,
4116		   le16_to_cpu(tx_data_bd->nbytes));
4117	}
4118
4119	DP(NETIF_MSG_TX_QUEUED, "last bd @%p\n", tx_data_bd);
4120
4121	/* update with actual num BDs */
4122	first_bd->nbd = cpu_to_le16(nbd);
4123
4124	bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
4125
4126	/* now send a tx doorbell, counting the next BD
4127	 * if the packet contains or ends with it
4128	 */
4129	if (TX_BD_POFF(bd_prod) < nbd)
4130		nbd++;
4131
4132	/* total_pkt_bytes should be set on the first data BD if
4133	 * it's not an LSO packet and there is more than one
4134	 * data BD. In this case pkt_size is limited by an MTU value.
4135	 * However we prefer to set it for an LSO packet (while we don't
4136	 * have to) in order to save some CPU cycles in a none-LSO
4137	 * case, when we much more care about them.
4138	 */
4139	if (total_pkt_bd != NULL)
4140		total_pkt_bd->total_pkt_bytes = pkt_size;
4141
4142	if (pbd_e1x)
4143		DP(NETIF_MSG_TX_QUEUED,
4144		   "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",
4145		   pbd_e1x, pbd_e1x->global_data, pbd_e1x->ip_hlen_w,
4146		   pbd_e1x->ip_id, pbd_e1x->lso_mss, pbd_e1x->tcp_flags,
4147		   pbd_e1x->tcp_pseudo_csum, pbd_e1x->tcp_send_seq,
4148		    le16_to_cpu(pbd_e1x->total_hlen_w));
4149	if (pbd_e2)
4150		DP(NETIF_MSG_TX_QUEUED,
4151		   "PBD (E2) @%p  dst %x %x %x src %x %x %x parsing_data %x\n",
4152		   pbd_e2,
4153		   pbd_e2->data.mac_addr.dst_hi,
4154		   pbd_e2->data.mac_addr.dst_mid,
4155		   pbd_e2->data.mac_addr.dst_lo,
4156		   pbd_e2->data.mac_addr.src_hi,
4157		   pbd_e2->data.mac_addr.src_mid,
4158		   pbd_e2->data.mac_addr.src_lo,
4159		   pbd_e2->parsing_data);
4160	DP(NETIF_MSG_TX_QUEUED, "doorbell: nbd %d  bd %u\n", nbd, bd_prod);
4161
4162	netdev_tx_sent_queue(txq, skb->len);
4163
4164	skb_tx_timestamp(skb);
4165
4166	txdata->tx_pkt_prod++;
4167	/*
4168	 * Make sure that the BD data is updated before updating the producer
4169	 * since FW might read the BD right after the producer is updated.
4170	 * This is only applicable for weak-ordered memory model archs such
4171	 * as IA-64. The following barrier is also mandatory since FW will
4172	 * assumes packets must have BDs.
4173	 */
4174	wmb();
4175
4176	txdata->tx_db.data.prod += nbd;
4177	/* make sure descriptor update is observed by HW */
4178	wmb();
4179
4180	DOORBELL_RELAXED(bp, txdata->cid, txdata->tx_db.raw);
4181
4182	txdata->tx_bd_prod += nbd;
4183
4184	if (unlikely(bnx2x_tx_avail(bp, txdata) < MAX_DESC_PER_TX_PKT)) {
4185		netif_tx_stop_queue(txq);
4186
4187		/* paired memory barrier is in bnx2x_tx_int(), we have to keep
4188		 * ordering of set_bit() in netif_tx_stop_queue() and read of
4189		 * fp->bd_tx_cons */
4190		smp_mb();
4191
4192		bnx2x_fp_qstats(bp, txdata->parent_fp)->driver_xoff++;
4193		if (bnx2x_tx_avail(bp, txdata) >= MAX_DESC_PER_TX_PKT)
4194			netif_tx_wake_queue(txq);
4195	}
4196	txdata->tx_pkt++;
4197
4198	return NETDEV_TX_OK;
4199}
4200
4201void bnx2x_get_c2s_mapping(struct bnx2x *bp, u8 *c2s_map, u8 *c2s_default)
4202{
4203	int mfw_vn = BP_FW_MB_IDX(bp);
4204	u32 tmp;
4205
4206	/* If the shmem shouldn't affect configuration, reflect */
4207	if (!IS_MF_BD(bp)) {
4208		int i;
4209
4210		for (i = 0; i < BNX2X_MAX_PRIORITY; i++)
4211			c2s_map[i] = i;
4212		*c2s_default = 0;
4213
4214		return;
4215	}
4216
4217	tmp = SHMEM2_RD(bp, c2s_pcp_map_lower[mfw_vn]);
4218	tmp = (__force u32)be32_to_cpu((__force __be32)tmp);
4219	c2s_map[0] = tmp & 0xff;
4220	c2s_map[1] = (tmp >> 8) & 0xff;
4221	c2s_map[2] = (tmp >> 16) & 0xff;
4222	c2s_map[3] = (tmp >> 24) & 0xff;
4223
4224	tmp = SHMEM2_RD(bp, c2s_pcp_map_upper[mfw_vn]);
4225	tmp = (__force u32)be32_to_cpu((__force __be32)tmp);
4226	c2s_map[4] = tmp & 0xff;
4227	c2s_map[5] = (tmp >> 8) & 0xff;
4228	c2s_map[6] = (tmp >> 16) & 0xff;
4229	c2s_map[7] = (tmp >> 24) & 0xff;
4230
4231	tmp = SHMEM2_RD(bp, c2s_pcp_map_default[mfw_vn]);
4232	tmp = (__force u32)be32_to_cpu((__force __be32)tmp);
4233	*c2s_default = (tmp >> (8 * mfw_vn)) & 0xff;
4234}
4235
4236/**
4237 * bnx2x_setup_tc - routine to configure net_device for multi tc
4238 *
4239 * @dev: net device to configure
4240 * @num_tc: number of traffic classes to enable
4241 *
4242 * callback connected to the ndo_setup_tc function pointer
4243 */
4244int bnx2x_setup_tc(struct net_device *dev, u8 num_tc)
4245{
4246	struct bnx2x *bp = netdev_priv(dev);
4247	u8 c2s_map[BNX2X_MAX_PRIORITY], c2s_def;
4248	int cos, prio, count, offset;
4249
4250	/* setup tc must be called under rtnl lock */
4251	ASSERT_RTNL();
4252
4253	/* no traffic classes requested. Aborting */
4254	if (!num_tc) {
4255		netdev_reset_tc(dev);
4256		return 0;
4257	}
4258
4259	/* requested to support too many traffic classes */
4260	if (num_tc > bp->max_cos) {
4261		BNX2X_ERR("support for too many traffic classes requested: %d. Max supported is %d\n",
4262			  num_tc, bp->max_cos);
4263		return -EINVAL;
4264	}
4265
4266	/* declare amount of supported traffic classes */
4267	if (netdev_set_num_tc(dev, num_tc)) {
4268		BNX2X_ERR("failed to declare %d traffic classes\n", num_tc);
4269		return -EINVAL;
4270	}
4271
4272	bnx2x_get_c2s_mapping(bp, c2s_map, &c2s_def);
4273
4274	/* configure priority to traffic class mapping */
4275	for (prio = 0; prio < BNX2X_MAX_PRIORITY; prio++) {
4276		int outer_prio = c2s_map[prio];
4277
4278		netdev_set_prio_tc_map(dev, prio, bp->prio_to_cos[outer_prio]);
4279		DP(BNX2X_MSG_SP | NETIF_MSG_IFUP,
4280		   "mapping priority %d to tc %d\n",
4281		   outer_prio, bp->prio_to_cos[outer_prio]);
4282	}
4283
4284	/* Use this configuration to differentiate tc0 from other COSes
4285	   This can be used for ets or pfc, and save the effort of setting
4286	   up a multio class queue disc or negotiating DCBX with a switch
4287	netdev_set_prio_tc_map(dev, 0, 0);
4288	DP(BNX2X_MSG_SP, "mapping priority %d to tc %d\n", 0, 0);
4289	for (prio = 1; prio < 16; prio++) {
4290		netdev_set_prio_tc_map(dev, prio, 1);
4291		DP(BNX2X_MSG_SP, "mapping priority %d to tc %d\n", prio, 1);
4292	} */
4293
4294	/* configure traffic class to transmission queue mapping */
4295	for (cos = 0; cos < bp->max_cos; cos++) {
4296		count = BNX2X_NUM_ETH_QUEUES(bp);
4297		offset = cos * BNX2X_NUM_NON_CNIC_QUEUES(bp);
4298		netdev_set_tc_queue(dev, cos, count, offset);
4299		DP(BNX2X_MSG_SP | NETIF_MSG_IFUP,
4300		   "mapping tc %d to offset %d count %d\n",
4301		   cos, offset, count);
4302	}
4303
4304	return 0;
4305}
4306
4307int __bnx2x_setup_tc(struct net_device *dev, enum tc_setup_type type,
4308		     void *type_data)
4309{
4310	struct tc_mqprio_qopt *mqprio = type_data;
4311
4312	if (type != TC_SETUP_QDISC_MQPRIO)
4313		return -EOPNOTSUPP;
4314
4315	mqprio->hw = TC_MQPRIO_HW_OFFLOAD_TCS;
4316
4317	return bnx2x_setup_tc(dev, mqprio->num_tc);
4318}
4319
4320/* called with rtnl_lock */
4321int bnx2x_change_mac_addr(struct net_device *dev, void *p)
4322{
4323	struct sockaddr *addr = p;
4324	struct bnx2x *bp = netdev_priv(dev);
4325	int rc = 0;
4326
4327	if (!is_valid_ether_addr(addr->sa_data)) {
4328		BNX2X_ERR("Requested MAC address is not valid\n");
4329		return -EINVAL;
4330	}
4331
4332	if (IS_MF_STORAGE_ONLY(bp)) {
4333		BNX2X_ERR("Can't change address on STORAGE ONLY function\n");
4334		return -EINVAL;
4335	}
4336
4337	if (netif_running(dev))  {
4338		rc = bnx2x_set_eth_mac(bp, false);
4339		if (rc)
4340			return rc;
4341	}
4342
4343	memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
4344
4345	if (netif_running(dev))
4346		rc = bnx2x_set_eth_mac(bp, true);
4347
4348	if (IS_PF(bp) && SHMEM2_HAS(bp, curr_cfg))
4349		SHMEM2_WR(bp, curr_cfg, CURR_CFG_MET_OS);
4350
4351	return rc;
4352}
4353
4354static void bnx2x_free_fp_mem_at(struct bnx2x *bp, int fp_index)
4355{
4356	union host_hc_status_block *sb = &bnx2x_fp(bp, fp_index, status_blk);
4357	struct bnx2x_fastpath *fp = &bp->fp[fp_index];
4358	u8 cos;
4359
4360	/* Common */
4361
4362	if (IS_FCOE_IDX(fp_index)) {
4363		memset(sb, 0, sizeof(union host_hc_status_block));
4364		fp->status_blk_mapping = 0;
4365	} else {
4366		/* status blocks */
4367		if (!CHIP_IS_E1x(bp))
4368			BNX2X_PCI_FREE(sb->e2_sb,
4369				       bnx2x_fp(bp, fp_index,
4370						status_blk_mapping),
4371				       sizeof(struct host_hc_status_block_e2));
4372		else
4373			BNX2X_PCI_FREE(sb->e1x_sb,
4374				       bnx2x_fp(bp, fp_index,
4375						status_blk_mapping),
4376				       sizeof(struct host_hc_status_block_e1x));
4377	}
4378
4379	/* Rx */
4380	if (!skip_rx_queue(bp, fp_index)) {
4381		bnx2x_free_rx_bds(fp);
4382
4383		/* fastpath rx rings: rx_buf rx_desc rx_comp */
4384		BNX2X_FREE(bnx2x_fp(bp, fp_index, rx_buf_ring));
4385		BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_desc_ring),
4386			       bnx2x_fp(bp, fp_index, rx_desc_mapping),
4387			       sizeof(struct eth_rx_bd) * NUM_RX_BD);
4388
4389		BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_comp_ring),
4390			       bnx2x_fp(bp, fp_index, rx_comp_mapping),
4391			       sizeof(struct eth_fast_path_rx_cqe) *
4392			       NUM_RCQ_BD);
4393
4394		/* SGE ring */
4395		BNX2X_FREE(bnx2x_fp(bp, fp_index, rx_page_ring));
4396		BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_sge_ring),
4397			       bnx2x_fp(bp, fp_index, rx_sge_mapping),
4398			       BCM_PAGE_SIZE * NUM_RX_SGE_PAGES);
4399	}
4400
4401	/* Tx */
4402	if (!skip_tx_queue(bp, fp_index)) {
4403		/* fastpath tx rings: tx_buf tx_desc */
4404		for_each_cos_in_tx_queue(fp, cos) {
4405			struct bnx2x_fp_txdata *txdata = fp->txdata_ptr[cos];
4406
4407			DP(NETIF_MSG_IFDOWN,
4408			   "freeing tx memory of fp %d cos %d cid %d\n",
4409			   fp_index, cos, txdata->cid);
4410
4411			BNX2X_FREE(txdata->tx_buf_ring);
4412			BNX2X_PCI_FREE(txdata->tx_desc_ring,
4413				txdata->tx_desc_mapping,
4414				sizeof(union eth_tx_bd_types) * NUM_TX_BD);
4415		}
4416	}
4417	/* end of fastpath */
4418}
4419
4420static void bnx2x_free_fp_mem_cnic(struct bnx2x *bp)
4421{
4422	int i;
4423	for_each_cnic_queue(bp, i)
4424		bnx2x_free_fp_mem_at(bp, i);
4425}
4426
4427void bnx2x_free_fp_mem(struct bnx2x *bp)
4428{
4429	int i;
4430	for_each_eth_queue(bp, i)
4431		bnx2x_free_fp_mem_at(bp, i);
4432}
4433
4434static void set_sb_shortcuts(struct bnx2x *bp, int index)
4435{
4436	union host_hc_status_block status_blk = bnx2x_fp(bp, index, status_blk);
4437	if (!CHIP_IS_E1x(bp)) {
4438		bnx2x_fp(bp, index, sb_index_values) =
4439			(__le16 *)status_blk.e2_sb->sb.index_values;
4440		bnx2x_fp(bp, index, sb_running_index) =
4441			(__le16 *)status_blk.e2_sb->sb.running_index;
4442	} else {
4443		bnx2x_fp(bp, index, sb_index_values) =
4444			(__le16 *)status_blk.e1x_sb->sb.index_values;
4445		bnx2x_fp(bp, index, sb_running_index) =
4446			(__le16 *)status_blk.e1x_sb->sb.running_index;
4447	}
4448}
4449
4450/* Returns the number of actually allocated BDs */
4451static int bnx2x_alloc_rx_bds(struct bnx2x_fastpath *fp,
4452			      int rx_ring_size)
4453{
4454	struct bnx2x *bp = fp->bp;
4455	u16 ring_prod, cqe_ring_prod;
4456	int i, failure_cnt = 0;
4457
4458	fp->rx_comp_cons = 0;
4459	cqe_ring_prod = ring_prod = 0;
4460
4461	/* This routine is called only during fo init so
4462	 * fp->eth_q_stats.rx_skb_alloc_failed = 0
4463	 */
4464	for (i = 0; i < rx_ring_size; i++) {
4465		if (bnx2x_alloc_rx_data(bp, fp, ring_prod, GFP_KERNEL) < 0) {
4466			failure_cnt++;
4467			continue;
4468		}
4469		ring_prod = NEXT_RX_IDX(ring_prod);
4470		cqe_ring_prod = NEXT_RCQ_IDX(cqe_ring_prod);
4471		WARN_ON(ring_prod <= (i - failure_cnt));
4472	}
4473
4474	if (failure_cnt)
4475		BNX2X_ERR("was only able to allocate %d rx skbs on queue[%d]\n",
4476			  i - failure_cnt, fp->index);
4477
4478	fp->rx_bd_prod = ring_prod;
4479	/* Limit the CQE producer by the CQE ring size */
4480	fp->rx_comp_prod = min_t(u16, NUM_RCQ_RINGS*RCQ_DESC_CNT,
4481			       cqe_ring_prod);
4482
4483	bnx2x_fp_stats(bp, fp)->eth_q_stats.rx_skb_alloc_failed += failure_cnt;
4484
4485	return i - failure_cnt;
4486}
4487
4488static void bnx2x_set_next_page_rx_cq(struct bnx2x_fastpath *fp)
4489{
4490	int i;
4491
4492	for (i = 1; i <= NUM_RCQ_RINGS; i++) {
4493		struct eth_rx_cqe_next_page *nextpg;
4494
4495		nextpg = (struct eth_rx_cqe_next_page *)
4496			&fp->rx_comp_ring[RCQ_DESC_CNT * i - 1];
4497		nextpg->addr_hi =
4498			cpu_to_le32(U64_HI(fp->rx_comp_mapping +
4499				   BCM_PAGE_SIZE*(i % NUM_RCQ_RINGS)));
4500		nextpg->addr_lo =
4501			cpu_to_le32(U64_LO(fp->rx_comp_mapping +
4502				   BCM_PAGE_SIZE*(i % NUM_RCQ_RINGS)));
4503	}
4504}
4505
4506static int bnx2x_alloc_fp_mem_at(struct bnx2x *bp, int index)
4507{
4508	union host_hc_status_block *sb;
4509	struct bnx2x_fastpath *fp = &bp->fp[index];
4510	int ring_size = 0;
4511	u8 cos;
4512	int rx_ring_size = 0;
4513
4514	if (!bp->rx_ring_size && IS_MF_STORAGE_ONLY(bp)) {
4515		rx_ring_size = MIN_RX_SIZE_NONTPA;
4516		bp->rx_ring_size = rx_ring_size;
4517	} else if (!bp->rx_ring_size) {
4518		rx_ring_size = MAX_RX_AVAIL/BNX2X_NUM_RX_QUEUES(bp);
4519
4520		if (CHIP_IS_E3(bp)) {
4521			u32 cfg = SHMEM_RD(bp,
4522					   dev_info.port_hw_config[BP_PORT(bp)].
4523					   default_cfg);
4524
4525			/* Decrease ring size for 1G functions */
4526			if ((cfg & PORT_HW_CFG_NET_SERDES_IF_MASK) ==
4527			    PORT_HW_CFG_NET_SERDES_IF_SGMII)
4528				rx_ring_size /= 10;
4529		}
4530
4531		/* allocate at least number of buffers required by FW */
4532		rx_ring_size = max_t(int, bp->disable_tpa ? MIN_RX_SIZE_NONTPA :
4533				     MIN_RX_SIZE_TPA, rx_ring_size);
4534
4535		bp->rx_ring_size = rx_ring_size;
4536	} else /* if rx_ring_size specified - use it */
4537		rx_ring_size = bp->rx_ring_size;
4538
4539	DP(BNX2X_MSG_SP, "calculated rx_ring_size %d\n", rx_ring_size);
4540
4541	/* Common */
4542	sb = &bnx2x_fp(bp, index, status_blk);
4543
4544	if (!IS_FCOE_IDX(index)) {
4545		/* status blocks */
4546		if (!CHIP_IS_E1x(bp)) {
4547			sb->e2_sb = BNX2X_PCI_ALLOC(&bnx2x_fp(bp, index, status_blk_mapping),
4548						    sizeof(struct host_hc_status_block_e2));
4549			if (!sb->e2_sb)
4550				goto alloc_mem_err;
4551		} else {
4552			sb->e1x_sb = BNX2X_PCI_ALLOC(&bnx2x_fp(bp, index, status_blk_mapping),
4553						     sizeof(struct host_hc_status_block_e1x));
4554			if (!sb->e1x_sb)
4555				goto alloc_mem_err;
4556		}
4557	}
4558
4559	/* FCoE Queue uses Default SB and doesn't ACK the SB, thus no need to
4560	 * set shortcuts for it.
4561	 */
4562	if (!IS_FCOE_IDX(index))
4563		set_sb_shortcuts(bp, index);
4564
4565	/* Tx */
4566	if (!skip_tx_queue(bp, index)) {
4567		/* fastpath tx rings: tx_buf tx_desc */
4568		for_each_cos_in_tx_queue(fp, cos) {
4569			struct bnx2x_fp_txdata *txdata = fp->txdata_ptr[cos];
4570
4571			DP(NETIF_MSG_IFUP,
4572			   "allocating tx memory of fp %d cos %d\n",
4573			   index, cos);
4574
4575			txdata->tx_buf_ring = kcalloc(NUM_TX_BD,
4576						      sizeof(struct sw_tx_bd),
4577						      GFP_KERNEL);
4578			if (!txdata->tx_buf_ring)
4579				goto alloc_mem_err;
4580			txdata->tx_desc_ring = BNX2X_PCI_ALLOC(&txdata->tx_desc_mapping,
4581							       sizeof(union eth_tx_bd_types) * NUM_TX_BD);
4582			if (!txdata->tx_desc_ring)
4583				goto alloc_mem_err;
4584		}
4585	}
4586
4587	/* Rx */
4588	if (!skip_rx_queue(bp, index)) {
4589		/* fastpath rx rings: rx_buf rx_desc rx_comp */
4590		bnx2x_fp(bp, index, rx_buf_ring) =
4591			kcalloc(NUM_RX_BD, sizeof(struct sw_rx_bd), GFP_KERNEL);
4592		if (!bnx2x_fp(bp, index, rx_buf_ring))
4593			goto alloc_mem_err;
4594		bnx2x_fp(bp, index, rx_desc_ring) =
4595			BNX2X_PCI_ALLOC(&bnx2x_fp(bp, index, rx_desc_mapping),
4596					sizeof(struct eth_rx_bd) * NUM_RX_BD);
4597		if (!bnx2x_fp(bp, index, rx_desc_ring))
4598			goto alloc_mem_err;
4599
4600		/* Seed all CQEs by 1s */
4601		bnx2x_fp(bp, index, rx_comp_ring) =
4602			BNX2X_PCI_FALLOC(&bnx2x_fp(bp, index, rx_comp_mapping),
4603					 sizeof(struct eth_fast_path_rx_cqe) * NUM_RCQ_BD);
4604		if (!bnx2x_fp(bp, index, rx_comp_ring))
4605			goto alloc_mem_err;
4606
4607		/* SGE ring */
4608		bnx2x_fp(bp, index, rx_page_ring) =
4609			kcalloc(NUM_RX_SGE, sizeof(struct sw_rx_page),
4610				GFP_KERNEL);
4611		if (!bnx2x_fp(bp, index, rx_page_ring))
4612			goto alloc_mem_err;
4613		bnx2x_fp(bp, index, rx_sge_ring) =
4614			BNX2X_PCI_ALLOC(&bnx2x_fp(bp, index, rx_sge_mapping),
4615					BCM_PAGE_SIZE * NUM_RX_SGE_PAGES);
4616		if (!bnx2x_fp(bp, index, rx_sge_ring))
4617			goto alloc_mem_err;
4618		/* RX BD ring */
4619		bnx2x_set_next_page_rx_bd(fp);
4620
4621		/* CQ ring */
4622		bnx2x_set_next_page_rx_cq(fp);
4623
4624		/* BDs */
4625		ring_size = bnx2x_alloc_rx_bds(fp, rx_ring_size);
4626		if (ring_size < rx_ring_size)
4627			goto alloc_mem_err;
4628	}
4629
4630	return 0;
4631
4632/* handles low memory cases */
4633alloc_mem_err:
4634	BNX2X_ERR("Unable to allocate full memory for queue %d (size %d)\n",
4635						index, ring_size);
4636	/* FW will drop all packets if queue is not big enough,
4637	 * In these cases we disable the queue
4638	 * Min size is different for OOO, TPA and non-TPA queues
4639	 */
4640	if (ring_size < (fp->mode == TPA_MODE_DISABLED ?
4641				MIN_RX_SIZE_NONTPA : MIN_RX_SIZE_TPA)) {
4642			/* release memory allocated for this queue */
4643			bnx2x_free_fp_mem_at(bp, index);
4644			return -ENOMEM;
4645	}
4646	return 0;
4647}
4648
4649static int bnx2x_alloc_fp_mem_cnic(struct bnx2x *bp)
4650{
4651	if (!NO_FCOE(bp))
4652		/* FCoE */
4653		if (bnx2x_alloc_fp_mem_at(bp, FCOE_IDX(bp)))
4654			/* we will fail load process instead of mark
4655			 * NO_FCOE_FLAG
4656			 */
4657			return -ENOMEM;
4658
4659	return 0;
4660}
4661
4662static int bnx2x_alloc_fp_mem(struct bnx2x *bp)
4663{
4664	int i;
4665
4666	/* 1. Allocate FP for leading - fatal if error
4667	 * 2. Allocate RSS - fix number of queues if error
4668	 */
4669
4670	/* leading */
4671	if (bnx2x_alloc_fp_mem_at(bp, 0))
4672		return -ENOMEM;
4673
4674	/* RSS */
4675	for_each_nondefault_eth_queue(bp, i)
4676		if (bnx2x_alloc_fp_mem_at(bp, i))
4677			break;
4678
4679	/* handle memory failures */
4680	if (i != BNX2X_NUM_ETH_QUEUES(bp)) {
4681		int delta = BNX2X_NUM_ETH_QUEUES(bp) - i;
4682
4683		WARN_ON(delta < 0);
4684		bnx2x_shrink_eth_fp(bp, delta);
4685		if (CNIC_SUPPORT(bp))
4686			/* move non eth FPs next to last eth FP
4687			 * must be done in that order
4688			 * FCOE_IDX < FWD_IDX < OOO_IDX
4689			 */
4690
4691			/* move FCoE fp even NO_FCOE_FLAG is on */
4692			bnx2x_move_fp(bp, FCOE_IDX(bp), FCOE_IDX(bp) - delta);
4693		bp->num_ethernet_queues -= delta;
4694		bp->num_queues = bp->num_ethernet_queues +
4695				 bp->num_cnic_queues;
4696		BNX2X_ERR("Adjusted num of queues from %d to %d\n",
4697			  bp->num_queues + delta, bp->num_queues);
4698	}
4699
4700	return 0;
4701}
4702
4703void bnx2x_free_mem_bp(struct bnx2x *bp)
4704{
4705	int i;
4706
4707	for (i = 0; i < bp->fp_array_size; i++)
4708		kfree(bp->fp[i].tpa_info);
4709	kfree(bp->fp);
4710	kfree(bp->sp_objs);
4711	kfree(bp->fp_stats);
4712	kfree(bp->bnx2x_txq);
4713	kfree(bp->msix_table);
4714	kfree(bp->ilt);
4715}
4716
4717int bnx2x_alloc_mem_bp(struct bnx2x *bp)
4718{
4719	struct bnx2x_fastpath *fp;
4720	struct msix_entry *tbl;
4721	struct bnx2x_ilt *ilt;
4722	int msix_table_size = 0;
4723	int fp_array_size, txq_array_size;
4724	int i;
4725
4726	/*
4727	 * The biggest MSI-X table we might need is as a maximum number of fast
4728	 * path IGU SBs plus default SB (for PF only).
4729	 */
4730	msix_table_size = bp->igu_sb_cnt;
4731	if (IS_PF(bp))
4732		msix_table_size++;
4733	BNX2X_DEV_INFO("msix_table_size %d\n", msix_table_size);
4734
4735	/* fp array: RSS plus CNIC related L2 queues */
4736	fp_array_size = BNX2X_MAX_RSS_COUNT(bp) + CNIC_SUPPORT(bp);
4737	bp->fp_array_size = fp_array_size;
4738	BNX2X_DEV_INFO("fp_array_size %d\n", bp->fp_array_size);
4739
4740	fp = kcalloc(bp->fp_array_size, sizeof(*fp), GFP_KERNEL);
4741	if (!fp)
4742		goto alloc_err;
4743	for (i = 0; i < bp->fp_array_size; i++) {
4744		fp[i].tpa_info =
4745			kcalloc(ETH_MAX_AGGREGATION_QUEUES_E1H_E2,
4746				sizeof(struct bnx2x_agg_info), GFP_KERNEL);
4747		if (!(fp[i].tpa_info))
4748			goto alloc_err;
4749	}
4750
4751	bp->fp = fp;
4752
4753	/* allocate sp objs */
4754	bp->sp_objs = kcalloc(bp->fp_array_size, sizeof(struct bnx2x_sp_objs),
4755			      GFP_KERNEL);
4756	if (!bp->sp_objs)
4757		goto alloc_err;
4758
4759	/* allocate fp_stats */
4760	bp->fp_stats = kcalloc(bp->fp_array_size, sizeof(struct bnx2x_fp_stats),
4761			       GFP_KERNEL);
4762	if (!bp->fp_stats)
4763		goto alloc_err;
4764
4765	/* Allocate memory for the transmission queues array */
4766	txq_array_size =
4767		BNX2X_MAX_RSS_COUNT(bp) * BNX2X_MULTI_TX_COS + CNIC_SUPPORT(bp);
4768	BNX2X_DEV_INFO("txq_array_size %d", txq_array_size);
4769
4770	bp->bnx2x_txq = kcalloc(txq_array_size, sizeof(struct bnx2x_fp_txdata),
4771				GFP_KERNEL);
4772	if (!bp->bnx2x_txq)
4773		goto alloc_err;
4774
4775	/* msix table */
4776	tbl = kcalloc(msix_table_size, sizeof(*tbl), GFP_KERNEL);
4777	if (!tbl)
4778		goto alloc_err;
4779	bp->msix_table = tbl;
4780
4781	/* ilt */
4782	ilt = kzalloc(sizeof(*ilt), GFP_KERNEL);
4783	if (!ilt)
4784		goto alloc_err;
4785	bp->ilt = ilt;
4786
4787	return 0;
4788alloc_err:
4789	bnx2x_free_mem_bp(bp);
4790	return -ENOMEM;
4791}
4792
4793int bnx2x_reload_if_running(struct net_device *dev)
4794{
4795	struct bnx2x *bp = netdev_priv(dev);
4796
4797	if (unlikely(!netif_running(dev)))
4798		return 0;
4799
4800	bnx2x_nic_unload(bp, UNLOAD_NORMAL, true);
4801	return bnx2x_nic_load(bp, LOAD_NORMAL);
4802}
4803
4804int bnx2x_get_cur_phy_idx(struct bnx2x *bp)
4805{
4806	u32 sel_phy_idx = 0;
4807	if (bp->link_params.num_phys <= 1)
4808		return INT_PHY;
4809
4810	if (bp->link_vars.link_up) {
4811		sel_phy_idx = EXT_PHY1;
4812		/* In case link is SERDES, check if the EXT_PHY2 is the one */
4813		if ((bp->link_vars.link_status & LINK_STATUS_SERDES_LINK) &&
4814		    (bp->link_params.phy[EXT_PHY2].supported & SUPPORTED_FIBRE))
4815			sel_phy_idx = EXT_PHY2;
4816	} else {
4817
4818		switch (bnx2x_phy_selection(&bp->link_params)) {
4819		case PORT_HW_CFG_PHY_SELECTION_HARDWARE_DEFAULT:
4820		case PORT_HW_CFG_PHY_SELECTION_FIRST_PHY:
4821		case PORT_HW_CFG_PHY_SELECTION_FIRST_PHY_PRIORITY:
4822		       sel_phy_idx = EXT_PHY1;
4823		       break;
4824		case PORT_HW_CFG_PHY_SELECTION_SECOND_PHY:
4825		case PORT_HW_CFG_PHY_SELECTION_SECOND_PHY_PRIORITY:
4826		       sel_phy_idx = EXT_PHY2;
4827		       break;
4828		}
4829	}
4830
4831	return sel_phy_idx;
4832}
4833int bnx2x_get_link_cfg_idx(struct bnx2x *bp)
4834{
4835	u32 sel_phy_idx = bnx2x_get_cur_phy_idx(bp);
4836	/*
4837	 * The selected activated PHY is always after swapping (in case PHY
4838	 * swapping is enabled). So when swapping is enabled, we need to reverse
4839	 * the configuration
4840	 */
4841
4842	if (bp->link_params.multi_phy_config &
4843	    PORT_HW_CFG_PHY_SWAPPED_ENABLED) {
4844		if (sel_phy_idx == EXT_PHY1)
4845			sel_phy_idx = EXT_PHY2;
4846		else if (sel_phy_idx == EXT_PHY2)
4847			sel_phy_idx = EXT_PHY1;
4848	}
4849	return LINK_CONFIG_IDX(sel_phy_idx);
4850}
4851
4852#ifdef NETDEV_FCOE_WWNN
4853int bnx2x_fcoe_get_wwn(struct net_device *dev, u64 *wwn, int type)
4854{
4855	struct bnx2x *bp = netdev_priv(dev);
4856	struct cnic_eth_dev *cp = &bp->cnic_eth_dev;
4857
4858	switch (type) {
4859	case NETDEV_FCOE_WWNN:
4860		*wwn = HILO_U64(cp->fcoe_wwn_node_name_hi,
4861				cp->fcoe_wwn_node_name_lo);
4862		break;
4863	case NETDEV_FCOE_WWPN:
4864		*wwn = HILO_U64(cp->fcoe_wwn_port_name_hi,
4865				cp->fcoe_wwn_port_name_lo);
4866		break;
4867	default:
4868		BNX2X_ERR("Wrong WWN type requested - %d\n", type);
4869		return -EINVAL;
4870	}
4871
4872	return 0;
4873}
4874#endif
4875
4876/* called with rtnl_lock */
4877int bnx2x_change_mtu(struct net_device *dev, int new_mtu)
4878{
4879	struct bnx2x *bp = netdev_priv(dev);
4880
4881	if (pci_num_vf(bp->pdev)) {
4882		DP(BNX2X_MSG_IOV, "VFs are enabled, can not change MTU\n");
4883		return -EPERM;
4884	}
4885
4886	if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
4887		BNX2X_ERR("Can't perform change MTU during parity recovery\n");
4888		return -EAGAIN;
4889	}
4890
4891	/* This does not race with packet allocation
4892	 * because the actual alloc size is
4893	 * only updated as part of load
4894	 */
4895	dev->mtu = new_mtu;
4896
4897	if (!bnx2x_mtu_allows_gro(new_mtu))
4898		dev->features &= ~NETIF_F_GRO_HW;
4899
4900	if (IS_PF(bp) && SHMEM2_HAS(bp, curr_cfg))
4901		SHMEM2_WR(bp, curr_cfg, CURR_CFG_MET_OS);
4902
4903	return bnx2x_reload_if_running(dev);
4904}
4905
4906netdev_features_t bnx2x_fix_features(struct net_device *dev,
4907				     netdev_features_t features)
4908{
4909	struct bnx2x *bp = netdev_priv(dev);
4910
4911	if (pci_num_vf(bp->pdev)) {
4912		netdev_features_t changed = dev->features ^ features;
4913
4914		/* Revert the requested changes in features if they
4915		 * would require internal reload of PF in bnx2x_set_features().
4916		 */
4917		if (!(features & NETIF_F_RXCSUM) && !bp->disable_tpa) {
4918			features &= ~NETIF_F_RXCSUM;
4919			features |= dev->features & NETIF_F_RXCSUM;
4920		}
4921
4922		if (changed & NETIF_F_LOOPBACK) {
4923			features &= ~NETIF_F_LOOPBACK;
4924			features |= dev->features & NETIF_F_LOOPBACK;
4925		}
4926	}
4927
4928	/* TPA requires Rx CSUM offloading */
4929	if (!(features & NETIF_F_RXCSUM))
4930		features &= ~NETIF_F_LRO;
4931
4932	if (!(features & NETIF_F_GRO) || !bnx2x_mtu_allows_gro(dev->mtu))
4933		features &= ~NETIF_F_GRO_HW;
4934	if (features & NETIF_F_GRO_HW)
4935		features &= ~NETIF_F_LRO;
4936
4937	return features;
4938}
4939
4940int bnx2x_set_features(struct net_device *dev, netdev_features_t features)
4941{
4942	struct bnx2x *bp = netdev_priv(dev);
4943	netdev_features_t changes = features ^ dev->features;
4944	bool bnx2x_reload = false;
4945	int rc;
4946
4947	/* VFs or non SRIOV PFs should be able to change loopback feature */
4948	if (!pci_num_vf(bp->pdev)) {
4949		if (features & NETIF_F_LOOPBACK) {
4950			if (bp->link_params.loopback_mode != LOOPBACK_BMAC) {
4951				bp->link_params.loopback_mode = LOOPBACK_BMAC;
4952				bnx2x_reload = true;
4953			}
4954		} else {
4955			if (bp->link_params.loopback_mode != LOOPBACK_NONE) {
4956				bp->link_params.loopback_mode = LOOPBACK_NONE;
4957				bnx2x_reload = true;
4958			}
4959		}
4960	}
4961
4962	/* Don't care about GRO changes */
4963	changes &= ~NETIF_F_GRO;
4964
4965	if (changes)
4966		bnx2x_reload = true;
4967
4968	if (bnx2x_reload) {
4969		if (bp->recovery_state == BNX2X_RECOVERY_DONE) {
4970			dev->features = features;
4971			rc = bnx2x_reload_if_running(dev);
4972			return rc ? rc : 1;
4973		}
4974		/* else: bnx2x_nic_load() will be called at end of recovery */
4975	}
4976
4977	return 0;
4978}
4979
4980void bnx2x_tx_timeout(struct net_device *dev, unsigned int txqueue)
4981{
4982	struct bnx2x *bp = netdev_priv(dev);
4983
4984	/* We want the information of the dump logged,
4985	 * but calling bnx2x_panic() would kill all chances of recovery.
4986	 */
4987	if (!bp->panic)
4988#ifndef BNX2X_STOP_ON_ERROR
4989		bnx2x_panic_dump(bp, false);
4990#else
4991		bnx2x_panic();
4992#endif
4993
4994	/* This allows the netif to be shutdown gracefully before resetting */
4995	bnx2x_schedule_sp_rtnl(bp, BNX2X_SP_RTNL_TX_TIMEOUT, 0);
4996}
4997
4998static int __maybe_unused bnx2x_suspend(struct device *dev_d)
4999{
5000	struct pci_dev *pdev = to_pci_dev(dev_d);
5001	struct net_device *dev = pci_get_drvdata(pdev);
5002	struct bnx2x *bp;
5003
5004	if (!dev) {
5005		dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
5006		return -ENODEV;
5007	}
5008	bp = netdev_priv(dev);
5009
5010	rtnl_lock();
5011
5012	if (!netif_running(dev)) {
5013		rtnl_unlock();
5014		return 0;
5015	}
5016
5017	netif_device_detach(dev);
5018
5019	bnx2x_nic_unload(bp, UNLOAD_CLOSE, false);
5020
5021	rtnl_unlock();
5022
5023	return 0;
5024}
5025
5026static int __maybe_unused bnx2x_resume(struct device *dev_d)
5027{
5028	struct pci_dev *pdev = to_pci_dev(dev_d);
5029	struct net_device *dev = pci_get_drvdata(pdev);
5030	struct bnx2x *bp;
5031	int rc;
5032
5033	if (!dev) {
5034		dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
5035		return -ENODEV;
5036	}
5037	bp = netdev_priv(dev);
5038
5039	if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
5040		BNX2X_ERR("Handling parity error recovery. Try again later\n");
5041		return -EAGAIN;
5042	}
5043
5044	rtnl_lock();
5045
5046	if (!netif_running(dev)) {
5047		rtnl_unlock();
5048		return 0;
5049	}
5050
5051	netif_device_attach(dev);
5052
5053	rc = bnx2x_nic_load(bp, LOAD_OPEN);
5054
5055	rtnl_unlock();
5056
5057	return rc;
5058}
5059
5060SIMPLE_DEV_PM_OPS(bnx2x_pm_ops, bnx2x_suspend, bnx2x_resume);
5061
5062void bnx2x_set_ctx_validation(struct bnx2x *bp, struct eth_context *cxt,
5063			      u32 cid)
5064{
5065	if (!cxt) {
5066		BNX2X_ERR("bad context pointer %p\n", cxt);
5067		return;
5068	}
5069
5070	/* ustorm cxt validation */
5071	cxt->ustorm_ag_context.cdu_usage =
5072		CDU_RSRVD_VALUE_TYPE_A(HW_CID(bp, cid),
5073			CDU_REGION_NUMBER_UCM_AG, ETH_CONNECTION_TYPE);
5074	/* xcontext validation */
5075	cxt->xstorm_ag_context.cdu_reserved =
5076		CDU_RSRVD_VALUE_TYPE_A(HW_CID(bp, cid),
5077			CDU_REGION_NUMBER_XCM_AG, ETH_CONNECTION_TYPE);
5078}
5079
5080static void storm_memset_hc_timeout(struct bnx2x *bp, u8 port,
5081				    u8 fw_sb_id, u8 sb_index,
5082				    u8 ticks)
5083{
5084	u32 addr = BAR_CSTRORM_INTMEM +
5085		   CSTORM_STATUS_BLOCK_DATA_TIMEOUT_OFFSET(fw_sb_id, sb_index);
5086	REG_WR8(bp, addr, ticks);
5087	DP(NETIF_MSG_IFUP,
5088	   "port %x fw_sb_id %d sb_index %d ticks %d\n",
5089	   port, fw_sb_id, sb_index, ticks);
5090}
5091
5092static void storm_memset_hc_disable(struct bnx2x *bp, u8 port,
5093				    u16 fw_sb_id, u8 sb_index,
5094				    u8 disable)
5095{
5096	u32 enable_flag = disable ? 0 : (1 << HC_INDEX_DATA_HC_ENABLED_SHIFT);
5097	u32 addr = BAR_CSTRORM_INTMEM +
5098		   CSTORM_STATUS_BLOCK_DATA_FLAGS_OFFSET(fw_sb_id, sb_index);
5099	u8 flags = REG_RD8(bp, addr);
5100	/* clear and set */
5101	flags &= ~HC_INDEX_DATA_HC_ENABLED;
5102	flags |= enable_flag;
5103	REG_WR8(bp, addr, flags);
5104	DP(NETIF_MSG_IFUP,
5105	   "port %x fw_sb_id %d sb_index %d disable %d\n",
5106	   port, fw_sb_id, sb_index, disable);
5107}
5108
5109void bnx2x_update_coalesce_sb_index(struct bnx2x *bp, u8 fw_sb_id,
5110				    u8 sb_index, u8 disable, u16 usec)
5111{
5112	int port = BP_PORT(bp);
5113	u8 ticks = usec / BNX2X_BTR;
5114
5115	storm_memset_hc_timeout(bp, port, fw_sb_id, sb_index, ticks);
5116
5117	disable = disable ? 1 : (usec ? 0 : 1);
5118	storm_memset_hc_disable(bp, port, fw_sb_id, sb_index, disable);
5119}
5120
5121void bnx2x_schedule_sp_rtnl(struct bnx2x *bp, enum sp_rtnl_flag flag,
5122			    u32 verbose)
5123{
5124	smp_mb__before_atomic();
5125	set_bit(flag, &bp->sp_rtnl_state);
5126	smp_mb__after_atomic();
5127	DP((BNX2X_MSG_SP | verbose), "Scheduling sp_rtnl task [Flag: %d]\n",
5128	   flag);
5129	schedule_delayed_work(&bp->sp_rtnl_task, 0);
5130}
5131