xref: /linux/drivers/net/ethernet/broadcom/bnx2x/bnx2x_cmn.c (revision fbce3befd60d40639bf3c6b60f7477b2f988f92d)
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 <net/busy_poll.h>
31 #include <linux/prefetch.h>
32 #include "bnx2x_cmn.h"
33 #include "bnx2x_init.h"
34 #include "bnx2x_sp.h"
35 
36 static void bnx2x_free_fp_mem_cnic(struct bnx2x *bp);
37 static int bnx2x_alloc_fp_mem_cnic(struct bnx2x *bp);
38 static int bnx2x_alloc_fp_mem(struct bnx2x *bp);
39 static int bnx2x_poll(struct napi_struct *napi, int budget);
40 
41 static void bnx2x_add_all_napi_cnic(struct bnx2x *bp)
42 {
43 	int i;
44 
45 	/* Add NAPI objects */
46 	for_each_rx_queue_cnic(bp, i) {
47 		netif_napi_add(bp->dev, &bnx2x_fp(bp, i, napi),
48 			       bnx2x_poll, NAPI_POLL_WEIGHT);
49 	}
50 }
51 
52 static void bnx2x_add_all_napi(struct bnx2x *bp)
53 {
54 	int i;
55 
56 	/* Add NAPI objects */
57 	for_each_eth_queue(bp, i) {
58 		netif_napi_add(bp->dev, &bnx2x_fp(bp, i, napi),
59 			       bnx2x_poll, NAPI_POLL_WEIGHT);
60 	}
61 }
62 
63 static int bnx2x_calc_num_queues(struct bnx2x *bp)
64 {
65 	int nq = bnx2x_num_queues ? : netif_get_num_default_rss_queues();
66 
67 	/* Reduce memory usage in kdump environment by using only one queue */
68 	if (is_kdump_kernel())
69 		nq = 1;
70 
71 	nq = clamp(nq, 1, BNX2X_MAX_QUEUES(bp));
72 	return nq;
73 }
74 
75 /**
76  * bnx2x_move_fp - move content of the fastpath structure.
77  *
78  * @bp:		driver handle
79  * @from:	source FP index
80  * @to:		destination FP index
81  *
82  * Makes sure the contents of the bp->fp[to].napi is kept
83  * intact. This is done by first copying the napi struct from
84  * the target to the source, and then mem copying the entire
85  * source onto the target. Update txdata pointers and related
86  * content.
87  */
88 static inline void bnx2x_move_fp(struct bnx2x *bp, int from, int to)
89 {
90 	struct bnx2x_fastpath *from_fp = &bp->fp[from];
91 	struct bnx2x_fastpath *to_fp = &bp->fp[to];
92 	struct bnx2x_sp_objs *from_sp_objs = &bp->sp_objs[from];
93 	struct bnx2x_sp_objs *to_sp_objs = &bp->sp_objs[to];
94 	struct bnx2x_fp_stats *from_fp_stats = &bp->fp_stats[from];
95 	struct bnx2x_fp_stats *to_fp_stats = &bp->fp_stats[to];
96 	int old_max_eth_txqs, new_max_eth_txqs;
97 	int old_txdata_index = 0, new_txdata_index = 0;
98 	struct bnx2x_agg_info *old_tpa_info = to_fp->tpa_info;
99 
100 	/* Copy the NAPI object as it has been already initialized */
101 	from_fp->napi = to_fp->napi;
102 
103 	/* Move bnx2x_fastpath contents */
104 	memcpy(to_fp, from_fp, sizeof(*to_fp));
105 	to_fp->index = to;
106 
107 	/* Retain the tpa_info of the original `to' version as we don't want
108 	 * 2 FPs to contain the same tpa_info pointer.
109 	 */
110 	to_fp->tpa_info = old_tpa_info;
111 
112 	/* move sp_objs contents as well, as their indices match fp ones */
113 	memcpy(to_sp_objs, from_sp_objs, sizeof(*to_sp_objs));
114 
115 	/* move fp_stats contents as well, as their indices match fp ones */
116 	memcpy(to_fp_stats, from_fp_stats, sizeof(*to_fp_stats));
117 
118 	/* Update txdata pointers in fp and move txdata content accordingly:
119 	 * Each fp consumes 'max_cos' txdata structures, so the index should be
120 	 * decremented by max_cos x delta.
121 	 */
122 
123 	old_max_eth_txqs = BNX2X_NUM_ETH_QUEUES(bp) * (bp)->max_cos;
124 	new_max_eth_txqs = (BNX2X_NUM_ETH_QUEUES(bp) - from + to) *
125 				(bp)->max_cos;
126 	if (from == FCOE_IDX(bp)) {
127 		old_txdata_index = old_max_eth_txqs + FCOE_TXQ_IDX_OFFSET;
128 		new_txdata_index = new_max_eth_txqs + FCOE_TXQ_IDX_OFFSET;
129 	}
130 
131 	memcpy(&bp->bnx2x_txq[new_txdata_index],
132 	       &bp->bnx2x_txq[old_txdata_index],
133 	       sizeof(struct bnx2x_fp_txdata));
134 	to_fp->txdata_ptr[0] = &bp->bnx2x_txq[new_txdata_index];
135 }
136 
137 /**
138  * bnx2x_fill_fw_str - Fill buffer with FW version string.
139  *
140  * @bp:        driver handle
141  * @buf:       character buffer to fill with the fw name
142  * @buf_len:   length of the above buffer
143  *
144  */
145 void bnx2x_fill_fw_str(struct bnx2x *bp, char *buf, size_t buf_len)
146 {
147 	if (IS_PF(bp)) {
148 		u8 phy_fw_ver[PHY_FW_VER_LEN];
149 
150 		phy_fw_ver[0] = '\0';
151 		bnx2x_get_ext_phy_fw_version(&bp->link_params,
152 					     phy_fw_ver, PHY_FW_VER_LEN);
153 		strlcpy(buf, bp->fw_ver, buf_len);
154 		snprintf(buf + strlen(bp->fw_ver), 32 - strlen(bp->fw_ver),
155 			 "bc %d.%d.%d%s%s",
156 			 (bp->common.bc_ver & 0xff0000) >> 16,
157 			 (bp->common.bc_ver & 0xff00) >> 8,
158 			 (bp->common.bc_ver & 0xff),
159 			 ((phy_fw_ver[0] != '\0') ? " phy " : ""), phy_fw_ver);
160 	} else {
161 		bnx2x_vf_fill_fw_str(bp, buf, buf_len);
162 	}
163 }
164 
165 /**
166  * bnx2x_shrink_eth_fp - guarantees fastpath structures stay intact
167  *
168  * @bp:	driver handle
169  * @delta:	number of eth queues which were not allocated
170  */
171 static void bnx2x_shrink_eth_fp(struct bnx2x *bp, int delta)
172 {
173 	int i, cos, old_eth_num = BNX2X_NUM_ETH_QUEUES(bp);
174 
175 	/* Queue pointer cannot be re-set on an fp-basis, as moving pointer
176 	 * backward along the array could cause memory to be overridden
177 	 */
178 	for (cos = 1; cos < bp->max_cos; cos++) {
179 		for (i = 0; i < old_eth_num - delta; i++) {
180 			struct bnx2x_fastpath *fp = &bp->fp[i];
181 			int new_idx = cos * (old_eth_num - delta) + i;
182 
183 			memcpy(&bp->bnx2x_txq[new_idx], fp->txdata_ptr[cos],
184 			       sizeof(struct bnx2x_fp_txdata));
185 			fp->txdata_ptr[cos] = &bp->bnx2x_txq[new_idx];
186 		}
187 	}
188 }
189 
190 int bnx2x_load_count[2][3] = { {0} }; /* per-path: 0-common, 1-port0, 2-port1 */
191 
192 /* free skb in the packet ring at pos idx
193  * return idx of last bd freed
194  */
195 static u16 bnx2x_free_tx_pkt(struct bnx2x *bp, struct bnx2x_fp_txdata *txdata,
196 			     u16 idx, unsigned int *pkts_compl,
197 			     unsigned int *bytes_compl)
198 {
199 	struct sw_tx_bd *tx_buf = &txdata->tx_buf_ring[idx];
200 	struct eth_tx_start_bd *tx_start_bd;
201 	struct eth_tx_bd *tx_data_bd;
202 	struct sk_buff *skb = tx_buf->skb;
203 	u16 bd_idx = TX_BD(tx_buf->first_bd), new_cons;
204 	int nbd;
205 	u16 split_bd_len = 0;
206 
207 	/* prefetch skb end pointer to speedup dev_kfree_skb() */
208 	prefetch(&skb->end);
209 
210 	DP(NETIF_MSG_TX_DONE, "fp[%d]: pkt_idx %d  buff @(%p)->skb %p\n",
211 	   txdata->txq_index, idx, tx_buf, skb);
212 
213 	tx_start_bd = &txdata->tx_desc_ring[bd_idx].start_bd;
214 
215 	nbd = le16_to_cpu(tx_start_bd->nbd) - 1;
216 #ifdef BNX2X_STOP_ON_ERROR
217 	if ((nbd - 1) > (MAX_SKB_FRAGS + 2)) {
218 		BNX2X_ERR("BAD nbd!\n");
219 		bnx2x_panic();
220 	}
221 #endif
222 	new_cons = nbd + tx_buf->first_bd;
223 
224 	/* Get the next bd */
225 	bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
226 
227 	/* Skip a parse bd... */
228 	--nbd;
229 	bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
230 
231 	if (tx_buf->flags & BNX2X_HAS_SECOND_PBD) {
232 		/* Skip second parse bd... */
233 		--nbd;
234 		bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
235 	}
236 
237 	/* TSO headers+data bds share a common mapping. See bnx2x_tx_split() */
238 	if (tx_buf->flags & BNX2X_TSO_SPLIT_BD) {
239 		tx_data_bd = &txdata->tx_desc_ring[bd_idx].reg_bd;
240 		split_bd_len = BD_UNMAP_LEN(tx_data_bd);
241 		--nbd;
242 		bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
243 	}
244 
245 	/* unmap first bd */
246 	dma_unmap_single(&bp->pdev->dev, BD_UNMAP_ADDR(tx_start_bd),
247 			 BD_UNMAP_LEN(tx_start_bd) + split_bd_len,
248 			 DMA_TO_DEVICE);
249 
250 	/* now free frags */
251 	while (nbd > 0) {
252 
253 		tx_data_bd = &txdata->tx_desc_ring[bd_idx].reg_bd;
254 		dma_unmap_page(&bp->pdev->dev, BD_UNMAP_ADDR(tx_data_bd),
255 			       BD_UNMAP_LEN(tx_data_bd), DMA_TO_DEVICE);
256 		if (--nbd)
257 			bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
258 	}
259 
260 	/* release skb */
261 	WARN_ON(!skb);
262 	if (likely(skb)) {
263 		(*pkts_compl)++;
264 		(*bytes_compl) += skb->len;
265 		dev_kfree_skb_any(skb);
266 	}
267 
268 	tx_buf->first_bd = 0;
269 	tx_buf->skb = NULL;
270 
271 	return new_cons;
272 }
273 
274 int bnx2x_tx_int(struct bnx2x *bp, struct bnx2x_fp_txdata *txdata)
275 {
276 	struct netdev_queue *txq;
277 	u16 hw_cons, sw_cons, bd_cons = txdata->tx_bd_cons;
278 	unsigned int pkts_compl = 0, bytes_compl = 0;
279 
280 #ifdef BNX2X_STOP_ON_ERROR
281 	if (unlikely(bp->panic))
282 		return -1;
283 #endif
284 
285 	txq = netdev_get_tx_queue(bp->dev, txdata->txq_index);
286 	hw_cons = le16_to_cpu(*txdata->tx_cons_sb);
287 	sw_cons = txdata->tx_pkt_cons;
288 
289 	while (sw_cons != hw_cons) {
290 		u16 pkt_cons;
291 
292 		pkt_cons = TX_BD(sw_cons);
293 
294 		DP(NETIF_MSG_TX_DONE,
295 		   "queue[%d]: hw_cons %u  sw_cons %u  pkt_cons %u\n",
296 		   txdata->txq_index, hw_cons, sw_cons, pkt_cons);
297 
298 		bd_cons = bnx2x_free_tx_pkt(bp, txdata, pkt_cons,
299 					    &pkts_compl, &bytes_compl);
300 
301 		sw_cons++;
302 	}
303 
304 	netdev_tx_completed_queue(txq, pkts_compl, bytes_compl);
305 
306 	txdata->tx_pkt_cons = sw_cons;
307 	txdata->tx_bd_cons = bd_cons;
308 
309 	/* Need to make the tx_bd_cons update visible to start_xmit()
310 	 * before checking for netif_tx_queue_stopped().  Without the
311 	 * memory barrier, there is a small possibility that
312 	 * start_xmit() will miss it and cause the queue to be stopped
313 	 * forever.
314 	 * On the other hand we need an rmb() here to ensure the proper
315 	 * ordering of bit testing in the following
316 	 * netif_tx_queue_stopped(txq) call.
317 	 */
318 	smp_mb();
319 
320 	if (unlikely(netif_tx_queue_stopped(txq))) {
321 		/* Taking tx_lock() is needed to prevent re-enabling the queue
322 		 * while it's empty. This could have happen if rx_action() gets
323 		 * suspended in bnx2x_tx_int() after the condition before
324 		 * netif_tx_wake_queue(), while tx_action (bnx2x_start_xmit()):
325 		 *
326 		 * stops the queue->sees fresh tx_bd_cons->releases the queue->
327 		 * sends some packets consuming the whole queue again->
328 		 * stops the queue
329 		 */
330 
331 		__netif_tx_lock(txq, smp_processor_id());
332 
333 		if ((netif_tx_queue_stopped(txq)) &&
334 		    (bp->state == BNX2X_STATE_OPEN) &&
335 		    (bnx2x_tx_avail(bp, txdata) >= MAX_DESC_PER_TX_PKT))
336 			netif_tx_wake_queue(txq);
337 
338 		__netif_tx_unlock(txq);
339 	}
340 	return 0;
341 }
342 
343 static inline void bnx2x_update_last_max_sge(struct bnx2x_fastpath *fp,
344 					     u16 idx)
345 {
346 	u16 last_max = fp->last_max_sge;
347 
348 	if (SUB_S16(idx, last_max) > 0)
349 		fp->last_max_sge = idx;
350 }
351 
352 static inline void bnx2x_update_sge_prod(struct bnx2x_fastpath *fp,
353 					 u16 sge_len,
354 					 struct eth_end_agg_rx_cqe *cqe)
355 {
356 	struct bnx2x *bp = fp->bp;
357 	u16 last_max, last_elem, first_elem;
358 	u16 delta = 0;
359 	u16 i;
360 
361 	if (!sge_len)
362 		return;
363 
364 	/* First mark all used pages */
365 	for (i = 0; i < sge_len; i++)
366 		BIT_VEC64_CLEAR_BIT(fp->sge_mask,
367 			RX_SGE(le16_to_cpu(cqe->sgl_or_raw_data.sgl[i])));
368 
369 	DP(NETIF_MSG_RX_STATUS, "fp_cqe->sgl[%d] = %d\n",
370 	   sge_len - 1, le16_to_cpu(cqe->sgl_or_raw_data.sgl[sge_len - 1]));
371 
372 	/* Here we assume that the last SGE index is the biggest */
373 	prefetch((void *)(fp->sge_mask));
374 	bnx2x_update_last_max_sge(fp,
375 		le16_to_cpu(cqe->sgl_or_raw_data.sgl[sge_len - 1]));
376 
377 	last_max = RX_SGE(fp->last_max_sge);
378 	last_elem = last_max >> BIT_VEC64_ELEM_SHIFT;
379 	first_elem = RX_SGE(fp->rx_sge_prod) >> BIT_VEC64_ELEM_SHIFT;
380 
381 	/* If ring is not full */
382 	if (last_elem + 1 != first_elem)
383 		last_elem++;
384 
385 	/* Now update the prod */
386 	for (i = first_elem; i != last_elem; i = NEXT_SGE_MASK_ELEM(i)) {
387 		if (likely(fp->sge_mask[i]))
388 			break;
389 
390 		fp->sge_mask[i] = BIT_VEC64_ELEM_ONE_MASK;
391 		delta += BIT_VEC64_ELEM_SZ;
392 	}
393 
394 	if (delta > 0) {
395 		fp->rx_sge_prod += delta;
396 		/* clear page-end entries */
397 		bnx2x_clear_sge_mask_next_elems(fp);
398 	}
399 
400 	DP(NETIF_MSG_RX_STATUS,
401 	   "fp->last_max_sge = %d  fp->rx_sge_prod = %d\n",
402 	   fp->last_max_sge, fp->rx_sge_prod);
403 }
404 
405 /* Get Toeplitz hash value in the skb using the value from the
406  * CQE (calculated by HW).
407  */
408 static u32 bnx2x_get_rxhash(const struct bnx2x *bp,
409 			    const struct eth_fast_path_rx_cqe *cqe,
410 			    enum pkt_hash_types *rxhash_type)
411 {
412 	/* Get Toeplitz hash from CQE */
413 	if ((bp->dev->features & NETIF_F_RXHASH) &&
414 	    (cqe->status_flags & ETH_FAST_PATH_RX_CQE_RSS_HASH_FLG)) {
415 		enum eth_rss_hash_type htype;
416 
417 		htype = cqe->status_flags & ETH_FAST_PATH_RX_CQE_RSS_HASH_TYPE;
418 		*rxhash_type = ((htype == TCP_IPV4_HASH_TYPE) ||
419 				(htype == TCP_IPV6_HASH_TYPE)) ?
420 			       PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3;
421 
422 		return le32_to_cpu(cqe->rss_hash_result);
423 	}
424 	*rxhash_type = PKT_HASH_TYPE_NONE;
425 	return 0;
426 }
427 
428 static void bnx2x_tpa_start(struct bnx2x_fastpath *fp, u16 queue,
429 			    u16 cons, u16 prod,
430 			    struct eth_fast_path_rx_cqe *cqe)
431 {
432 	struct bnx2x *bp = fp->bp;
433 	struct sw_rx_bd *cons_rx_buf = &fp->rx_buf_ring[cons];
434 	struct sw_rx_bd *prod_rx_buf = &fp->rx_buf_ring[prod];
435 	struct eth_rx_bd *prod_bd = &fp->rx_desc_ring[prod];
436 	dma_addr_t mapping;
437 	struct bnx2x_agg_info *tpa_info = &fp->tpa_info[queue];
438 	struct sw_rx_bd *first_buf = &tpa_info->first_buf;
439 
440 	/* print error if current state != stop */
441 	if (tpa_info->tpa_state != BNX2X_TPA_STOP)
442 		BNX2X_ERR("start of bin not in stop [%d]\n", queue);
443 
444 	/* Try to map an empty data buffer from the aggregation info  */
445 	mapping = dma_map_single(&bp->pdev->dev,
446 				 first_buf->data + NET_SKB_PAD,
447 				 fp->rx_buf_size, DMA_FROM_DEVICE);
448 	/*
449 	 *  ...if it fails - move the skb from the consumer to the producer
450 	 *  and set the current aggregation state as ERROR to drop it
451 	 *  when TPA_STOP arrives.
452 	 */
453 
454 	if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
455 		/* Move the BD from the consumer to the producer */
456 		bnx2x_reuse_rx_data(fp, cons, prod);
457 		tpa_info->tpa_state = BNX2X_TPA_ERROR;
458 		return;
459 	}
460 
461 	/* move empty data from pool to prod */
462 	prod_rx_buf->data = first_buf->data;
463 	dma_unmap_addr_set(prod_rx_buf, mapping, mapping);
464 	/* point prod_bd to new data */
465 	prod_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
466 	prod_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
467 
468 	/* move partial skb from cons to pool (don't unmap yet) */
469 	*first_buf = *cons_rx_buf;
470 
471 	/* mark bin state as START */
472 	tpa_info->parsing_flags =
473 		le16_to_cpu(cqe->pars_flags.flags);
474 	tpa_info->vlan_tag = le16_to_cpu(cqe->vlan_tag);
475 	tpa_info->tpa_state = BNX2X_TPA_START;
476 	tpa_info->len_on_bd = le16_to_cpu(cqe->len_on_bd);
477 	tpa_info->placement_offset = cqe->placement_offset;
478 	tpa_info->rxhash = bnx2x_get_rxhash(bp, cqe, &tpa_info->rxhash_type);
479 	if (fp->mode == TPA_MODE_GRO) {
480 		u16 gro_size = le16_to_cpu(cqe->pkt_len_or_gro_seg_len);
481 		tpa_info->full_page = SGE_PAGES / gro_size * gro_size;
482 		tpa_info->gro_size = gro_size;
483 	}
484 
485 #ifdef BNX2X_STOP_ON_ERROR
486 	fp->tpa_queue_used |= (1 << queue);
487 	DP(NETIF_MSG_RX_STATUS, "fp->tpa_queue_used = 0x%llx\n",
488 	   fp->tpa_queue_used);
489 #endif
490 }
491 
492 /* Timestamp option length allowed for TPA aggregation:
493  *
494  *		nop nop kind length echo val
495  */
496 #define TPA_TSTAMP_OPT_LEN	12
497 /**
498  * bnx2x_set_gro_params - compute GRO values
499  *
500  * @skb:		packet skb
501  * @parsing_flags:	parsing flags from the START CQE
502  * @len_on_bd:		total length of the first packet for the
503  *			aggregation.
504  * @pkt_len:		length of all segments
505  *
506  * Approximate value of the MSS for this aggregation calculated using
507  * the first packet of it.
508  * Compute number of aggregated segments, and gso_type.
509  */
510 static void bnx2x_set_gro_params(struct sk_buff *skb, u16 parsing_flags,
511 				 u16 len_on_bd, unsigned int pkt_len,
512 				 u16 num_of_coalesced_segs)
513 {
514 	/* TPA aggregation won't have either IP options or TCP options
515 	 * other than timestamp or IPv6 extension headers.
516 	 */
517 	u16 hdrs_len = ETH_HLEN + sizeof(struct tcphdr);
518 
519 	if (GET_FLAG(parsing_flags, PARSING_FLAGS_OVER_ETHERNET_PROTOCOL) ==
520 	    PRS_FLAG_OVERETH_IPV6) {
521 		hdrs_len += sizeof(struct ipv6hdr);
522 		skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
523 	} else {
524 		hdrs_len += sizeof(struct iphdr);
525 		skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
526 	}
527 
528 	/* Check if there was a TCP timestamp, if there is it's will
529 	 * always be 12 bytes length: nop nop kind length echo val.
530 	 *
531 	 * Otherwise FW would close the aggregation.
532 	 */
533 	if (parsing_flags & PARSING_FLAGS_TIME_STAMP_EXIST_FLAG)
534 		hdrs_len += TPA_TSTAMP_OPT_LEN;
535 
536 	skb_shinfo(skb)->gso_size = len_on_bd - hdrs_len;
537 
538 	/* tcp_gro_complete() will copy NAPI_GRO_CB(skb)->count
539 	 * to skb_shinfo(skb)->gso_segs
540 	 */
541 	NAPI_GRO_CB(skb)->count = num_of_coalesced_segs;
542 }
543 
544 static int bnx2x_alloc_rx_sge(struct bnx2x *bp, struct bnx2x_fastpath *fp,
545 			      u16 index, gfp_t gfp_mask)
546 {
547 	struct sw_rx_page *sw_buf = &fp->rx_page_ring[index];
548 	struct eth_rx_sge *sge = &fp->rx_sge_ring[index];
549 	struct bnx2x_alloc_pool *pool = &fp->page_pool;
550 	dma_addr_t mapping;
551 
552 	if (!pool->page || (PAGE_SIZE - pool->offset) < SGE_PAGE_SIZE) {
553 
554 		/* put page reference used by the memory pool, since we
555 		 * won't be using this page as the mempool anymore.
556 		 */
557 		if (pool->page)
558 			put_page(pool->page);
559 
560 		pool->page = alloc_pages(gfp_mask, PAGES_PER_SGE_SHIFT);
561 		if (unlikely(!pool->page))
562 			return -ENOMEM;
563 
564 		pool->offset = 0;
565 	}
566 
567 	mapping = dma_map_page(&bp->pdev->dev, pool->page,
568 			       pool->offset, SGE_PAGE_SIZE, DMA_FROM_DEVICE);
569 	if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
570 		BNX2X_ERR("Can't map sge\n");
571 		return -ENOMEM;
572 	}
573 
574 	get_page(pool->page);
575 	sw_buf->page = pool->page;
576 	sw_buf->offset = pool->offset;
577 
578 	dma_unmap_addr_set(sw_buf, mapping, mapping);
579 
580 	sge->addr_hi = cpu_to_le32(U64_HI(mapping));
581 	sge->addr_lo = cpu_to_le32(U64_LO(mapping));
582 
583 	pool->offset += SGE_PAGE_SIZE;
584 
585 	return 0;
586 }
587 
588 static int bnx2x_fill_frag_skb(struct bnx2x *bp, struct bnx2x_fastpath *fp,
589 			       struct bnx2x_agg_info *tpa_info,
590 			       u16 pages,
591 			       struct sk_buff *skb,
592 			       struct eth_end_agg_rx_cqe *cqe,
593 			       u16 cqe_idx)
594 {
595 	struct sw_rx_page *rx_pg, old_rx_pg;
596 	u32 i, frag_len, frag_size;
597 	int err, j, frag_id = 0;
598 	u16 len_on_bd = tpa_info->len_on_bd;
599 	u16 full_page = 0, gro_size = 0;
600 
601 	frag_size = le16_to_cpu(cqe->pkt_len) - len_on_bd;
602 
603 	if (fp->mode == TPA_MODE_GRO) {
604 		gro_size = tpa_info->gro_size;
605 		full_page = tpa_info->full_page;
606 	}
607 
608 	/* This is needed in order to enable forwarding support */
609 	if (frag_size)
610 		bnx2x_set_gro_params(skb, tpa_info->parsing_flags, len_on_bd,
611 				     le16_to_cpu(cqe->pkt_len),
612 				     le16_to_cpu(cqe->num_of_coalesced_segs));
613 
614 #ifdef BNX2X_STOP_ON_ERROR
615 	if (pages > min_t(u32, 8, MAX_SKB_FRAGS) * SGE_PAGES) {
616 		BNX2X_ERR("SGL length is too long: %d. CQE index is %d\n",
617 			  pages, cqe_idx);
618 		BNX2X_ERR("cqe->pkt_len = %d\n", cqe->pkt_len);
619 		bnx2x_panic();
620 		return -EINVAL;
621 	}
622 #endif
623 
624 	/* Run through the SGL and compose the fragmented skb */
625 	for (i = 0, j = 0; i < pages; i += PAGES_PER_SGE, j++) {
626 		u16 sge_idx = RX_SGE(le16_to_cpu(cqe->sgl_or_raw_data.sgl[j]));
627 
628 		/* FW gives the indices of the SGE as if the ring is an array
629 		   (meaning that "next" element will consume 2 indices) */
630 		if (fp->mode == TPA_MODE_GRO)
631 			frag_len = min_t(u32, frag_size, (u32)full_page);
632 		else /* LRO */
633 			frag_len = min_t(u32, frag_size, (u32)SGE_PAGES);
634 
635 		rx_pg = &fp->rx_page_ring[sge_idx];
636 		old_rx_pg = *rx_pg;
637 
638 		/* If we fail to allocate a substitute page, we simply stop
639 		   where we are and drop the whole packet */
640 		err = bnx2x_alloc_rx_sge(bp, fp, sge_idx, GFP_ATOMIC);
641 		if (unlikely(err)) {
642 			bnx2x_fp_qstats(bp, fp)->rx_skb_alloc_failed++;
643 			return err;
644 		}
645 
646 		dma_unmap_page(&bp->pdev->dev,
647 			       dma_unmap_addr(&old_rx_pg, mapping),
648 			       SGE_PAGE_SIZE, DMA_FROM_DEVICE);
649 		/* Add one frag and update the appropriate fields in the skb */
650 		if (fp->mode == TPA_MODE_LRO)
651 			skb_fill_page_desc(skb, j, old_rx_pg.page,
652 					   old_rx_pg.offset, frag_len);
653 		else { /* GRO */
654 			int rem;
655 			int offset = 0;
656 			for (rem = frag_len; rem > 0; rem -= gro_size) {
657 				int len = rem > gro_size ? gro_size : rem;
658 				skb_fill_page_desc(skb, frag_id++,
659 						   old_rx_pg.page,
660 						   old_rx_pg.offset + offset,
661 						   len);
662 				if (offset)
663 					get_page(old_rx_pg.page);
664 				offset += len;
665 			}
666 		}
667 
668 		skb->data_len += frag_len;
669 		skb->truesize += SGE_PAGES;
670 		skb->len += frag_len;
671 
672 		frag_size -= frag_len;
673 	}
674 
675 	return 0;
676 }
677 
678 static void bnx2x_frag_free(const struct bnx2x_fastpath *fp, void *data)
679 {
680 	if (fp->rx_frag_size)
681 		skb_free_frag(data);
682 	else
683 		kfree(data);
684 }
685 
686 static void *bnx2x_frag_alloc(const struct bnx2x_fastpath *fp, gfp_t gfp_mask)
687 {
688 	if (fp->rx_frag_size) {
689 		/* GFP_KERNEL allocations are used only during initialization */
690 		if (unlikely(gfpflags_allow_blocking(gfp_mask)))
691 			return (void *)__get_free_page(gfp_mask);
692 
693 		return netdev_alloc_frag(fp->rx_frag_size);
694 	}
695 
696 	return kmalloc(fp->rx_buf_size + NET_SKB_PAD, gfp_mask);
697 }
698 
699 #ifdef CONFIG_INET
700 static void bnx2x_gro_ip_csum(struct bnx2x *bp, struct sk_buff *skb)
701 {
702 	const struct iphdr *iph = ip_hdr(skb);
703 	struct tcphdr *th;
704 
705 	skb_set_transport_header(skb, sizeof(struct iphdr));
706 	th = tcp_hdr(skb);
707 
708 	th->check = ~tcp_v4_check(skb->len - skb_transport_offset(skb),
709 				  iph->saddr, iph->daddr, 0);
710 }
711 
712 static void bnx2x_gro_ipv6_csum(struct bnx2x *bp, struct sk_buff *skb)
713 {
714 	struct ipv6hdr *iph = ipv6_hdr(skb);
715 	struct tcphdr *th;
716 
717 	skb_set_transport_header(skb, sizeof(struct ipv6hdr));
718 	th = tcp_hdr(skb);
719 
720 	th->check = ~tcp_v6_check(skb->len - skb_transport_offset(skb),
721 				  &iph->saddr, &iph->daddr, 0);
722 }
723 
724 static void bnx2x_gro_csum(struct bnx2x *bp, struct sk_buff *skb,
725 			    void (*gro_func)(struct bnx2x*, struct sk_buff*))
726 {
727 	skb_set_network_header(skb, 0);
728 	gro_func(bp, skb);
729 	tcp_gro_complete(skb);
730 }
731 #endif
732 
733 static void bnx2x_gro_receive(struct bnx2x *bp, struct bnx2x_fastpath *fp,
734 			       struct sk_buff *skb)
735 {
736 #ifdef CONFIG_INET
737 	if (skb_shinfo(skb)->gso_size) {
738 		switch (be16_to_cpu(skb->protocol)) {
739 		case ETH_P_IP:
740 			bnx2x_gro_csum(bp, skb, bnx2x_gro_ip_csum);
741 			break;
742 		case ETH_P_IPV6:
743 			bnx2x_gro_csum(bp, skb, bnx2x_gro_ipv6_csum);
744 			break;
745 		default:
746 			WARN_ONCE(1, "Error: FW GRO supports only IPv4/IPv6, not 0x%04x\n",
747 				  be16_to_cpu(skb->protocol));
748 		}
749 	}
750 #endif
751 	skb_record_rx_queue(skb, fp->rx_queue);
752 	napi_gro_receive(&fp->napi, skb);
753 }
754 
755 static void bnx2x_tpa_stop(struct bnx2x *bp, struct bnx2x_fastpath *fp,
756 			   struct bnx2x_agg_info *tpa_info,
757 			   u16 pages,
758 			   struct eth_end_agg_rx_cqe *cqe,
759 			   u16 cqe_idx)
760 {
761 	struct sw_rx_bd *rx_buf = &tpa_info->first_buf;
762 	u8 pad = tpa_info->placement_offset;
763 	u16 len = tpa_info->len_on_bd;
764 	struct sk_buff *skb = NULL;
765 	u8 *new_data, *data = rx_buf->data;
766 	u8 old_tpa_state = tpa_info->tpa_state;
767 
768 	tpa_info->tpa_state = BNX2X_TPA_STOP;
769 
770 	/* If we there was an error during the handling of the TPA_START -
771 	 * drop this aggregation.
772 	 */
773 	if (old_tpa_state == BNX2X_TPA_ERROR)
774 		goto drop;
775 
776 	/* Try to allocate the new data */
777 	new_data = bnx2x_frag_alloc(fp, GFP_ATOMIC);
778 	/* Unmap skb in the pool anyway, as we are going to change
779 	   pool entry status to BNX2X_TPA_STOP even if new skb allocation
780 	   fails. */
781 	dma_unmap_single(&bp->pdev->dev, dma_unmap_addr(rx_buf, mapping),
782 			 fp->rx_buf_size, DMA_FROM_DEVICE);
783 	if (likely(new_data))
784 		skb = build_skb(data, fp->rx_frag_size);
785 
786 	if (likely(skb)) {
787 #ifdef BNX2X_STOP_ON_ERROR
788 		if (pad + len > fp->rx_buf_size) {
789 			BNX2X_ERR("skb_put is about to fail...  pad %d  len %d  rx_buf_size %d\n",
790 				  pad, len, fp->rx_buf_size);
791 			bnx2x_panic();
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);
821 drop:
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 
828 static 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 
858 static
859 void 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 
882 static 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++;
1062 reuse_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);
1094 next_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++;
1101 next_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 
1128 static 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 */
1156 void 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 
1163 void 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 */
1171 u16 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  */
1202 static 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 
1249 void 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  */
1264 void __bnx2x_link_report(struct bnx2x *bp)
1265 {
1266 	struct bnx2x_link_report_data cur_data;
1267 
1268 	/* reread mf_cfg */
1269 	if (IS_PF(bp) && !CHIP_IS_E1(bp))
1270 		bnx2x_read_mf_cfg(bp);
1271 
1272 	/* Read the current link report info */
1273 	bnx2x_fill_report_data(bp, &cur_data);
1274 
1275 	/* Don't report link down or exactly the same link status twice */
1276 	if (!memcmp(&cur_data, &bp->last_reported_link, sizeof(cur_data)) ||
1277 	    (test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1278 		      &bp->last_reported_link.link_report_flags) &&
1279 	     test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1280 		      &cur_data.link_report_flags)))
1281 		return;
1282 
1283 	bp->link_cnt++;
1284 
1285 	/* We are going to report a new link parameters now -
1286 	 * remember the current data for the next time.
1287 	 */
1288 	memcpy(&bp->last_reported_link, &cur_data, sizeof(cur_data));
1289 
1290 	/* propagate status to VFs */
1291 	if (IS_PF(bp))
1292 		bnx2x_iov_link_update(bp);
1293 
1294 	if (test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1295 		     &cur_data.link_report_flags)) {
1296 		netif_carrier_off(bp->dev);
1297 		netdev_err(bp->dev, "NIC Link is Down\n");
1298 		return;
1299 	} else {
1300 		const char *duplex;
1301 		const char *flow;
1302 
1303 		netif_carrier_on(bp->dev);
1304 
1305 		if (test_and_clear_bit(BNX2X_LINK_REPORT_FD,
1306 				       &cur_data.link_report_flags))
1307 			duplex = "full";
1308 		else
1309 			duplex = "half";
1310 
1311 		/* Handle the FC at the end so that only these flags would be
1312 		 * possibly set. This way we may easily check if there is no FC
1313 		 * enabled.
1314 		 */
1315 		if (cur_data.link_report_flags) {
1316 			if (test_bit(BNX2X_LINK_REPORT_RX_FC_ON,
1317 				     &cur_data.link_report_flags)) {
1318 				if (test_bit(BNX2X_LINK_REPORT_TX_FC_ON,
1319 				     &cur_data.link_report_flags))
1320 					flow = "ON - receive & transmit";
1321 				else
1322 					flow = "ON - receive";
1323 			} else {
1324 				flow = "ON - transmit";
1325 			}
1326 		} else {
1327 			flow = "none";
1328 		}
1329 		netdev_info(bp->dev, "NIC Link is Up, %d Mbps %s duplex, Flow control: %s\n",
1330 			    cur_data.line_speed, duplex, flow);
1331 	}
1332 }
1333 
1334 static void bnx2x_set_next_page_sgl(struct bnx2x_fastpath *fp)
1335 {
1336 	int i;
1337 
1338 	for (i = 1; i <= NUM_RX_SGE_PAGES; i++) {
1339 		struct eth_rx_sge *sge;
1340 
1341 		sge = &fp->rx_sge_ring[RX_SGE_CNT * i - 2];
1342 		sge->addr_hi =
1343 			cpu_to_le32(U64_HI(fp->rx_sge_mapping +
1344 			BCM_PAGE_SIZE*(i % NUM_RX_SGE_PAGES)));
1345 
1346 		sge->addr_lo =
1347 			cpu_to_le32(U64_LO(fp->rx_sge_mapping +
1348 			BCM_PAGE_SIZE*(i % NUM_RX_SGE_PAGES)));
1349 	}
1350 }
1351 
1352 static void bnx2x_free_tpa_pool(struct bnx2x *bp,
1353 				struct bnx2x_fastpath *fp, int last)
1354 {
1355 	int i;
1356 
1357 	for (i = 0; i < last; i++) {
1358 		struct bnx2x_agg_info *tpa_info = &fp->tpa_info[i];
1359 		struct sw_rx_bd *first_buf = &tpa_info->first_buf;
1360 		u8 *data = first_buf->data;
1361 
1362 		if (data == NULL) {
1363 			DP(NETIF_MSG_IFDOWN, "tpa bin %d empty on free\n", i);
1364 			continue;
1365 		}
1366 		if (tpa_info->tpa_state == BNX2X_TPA_START)
1367 			dma_unmap_single(&bp->pdev->dev,
1368 					 dma_unmap_addr(first_buf, mapping),
1369 					 fp->rx_buf_size, DMA_FROM_DEVICE);
1370 		bnx2x_frag_free(fp, data);
1371 		first_buf->data = NULL;
1372 	}
1373 }
1374 
1375 void bnx2x_init_rx_rings_cnic(struct bnx2x *bp)
1376 {
1377 	int j;
1378 
1379 	for_each_rx_queue_cnic(bp, j) {
1380 		struct bnx2x_fastpath *fp = &bp->fp[j];
1381 
1382 		fp->rx_bd_cons = 0;
1383 
1384 		/* Activate BD ring */
1385 		/* Warning!
1386 		 * this will generate an interrupt (to the TSTORM)
1387 		 * must only be done after chip is initialized
1388 		 */
1389 		bnx2x_update_rx_prod(bp, fp, fp->rx_bd_prod, fp->rx_comp_prod,
1390 				     fp->rx_sge_prod);
1391 	}
1392 }
1393 
1394 void bnx2x_init_rx_rings(struct bnx2x *bp)
1395 {
1396 	int func = BP_FUNC(bp);
1397 	u16 ring_prod;
1398 	int i, j;
1399 
1400 	/* Allocate TPA resources */
1401 	for_each_eth_queue(bp, j) {
1402 		struct bnx2x_fastpath *fp = &bp->fp[j];
1403 
1404 		DP(NETIF_MSG_IFUP,
1405 		   "mtu %d  rx_buf_size %d\n", bp->dev->mtu, fp->rx_buf_size);
1406 
1407 		if (fp->mode != TPA_MODE_DISABLED) {
1408 			/* Fill the per-aggregation pool */
1409 			for (i = 0; i < MAX_AGG_QS(bp); i++) {
1410 				struct bnx2x_agg_info *tpa_info =
1411 					&fp->tpa_info[i];
1412 				struct sw_rx_bd *first_buf =
1413 					&tpa_info->first_buf;
1414 
1415 				first_buf->data =
1416 					bnx2x_frag_alloc(fp, GFP_KERNEL);
1417 				if (!first_buf->data) {
1418 					BNX2X_ERR("Failed to allocate TPA skb pool for queue[%d] - disabling TPA on this queue!\n",
1419 						  j);
1420 					bnx2x_free_tpa_pool(bp, fp, i);
1421 					fp->mode = TPA_MODE_DISABLED;
1422 					break;
1423 				}
1424 				dma_unmap_addr_set(first_buf, mapping, 0);
1425 				tpa_info->tpa_state = BNX2X_TPA_STOP;
1426 			}
1427 
1428 			/* "next page" elements initialization */
1429 			bnx2x_set_next_page_sgl(fp);
1430 
1431 			/* set SGEs bit mask */
1432 			bnx2x_init_sge_ring_bit_mask(fp);
1433 
1434 			/* Allocate SGEs and initialize the ring elements */
1435 			for (i = 0, ring_prod = 0;
1436 			     i < MAX_RX_SGE_CNT*NUM_RX_SGE_PAGES; i++) {
1437 
1438 				if (bnx2x_alloc_rx_sge(bp, fp, ring_prod,
1439 						       GFP_KERNEL) < 0) {
1440 					BNX2X_ERR("was only able to allocate %d rx sges\n",
1441 						  i);
1442 					BNX2X_ERR("disabling TPA for queue[%d]\n",
1443 						  j);
1444 					/* Cleanup already allocated elements */
1445 					bnx2x_free_rx_sge_range(bp, fp,
1446 								ring_prod);
1447 					bnx2x_free_tpa_pool(bp, fp,
1448 							    MAX_AGG_QS(bp));
1449 					fp->mode = TPA_MODE_DISABLED;
1450 					ring_prod = 0;
1451 					break;
1452 				}
1453 				ring_prod = NEXT_SGE_IDX(ring_prod);
1454 			}
1455 
1456 			fp->rx_sge_prod = ring_prod;
1457 		}
1458 	}
1459 
1460 	for_each_eth_queue(bp, j) {
1461 		struct bnx2x_fastpath *fp = &bp->fp[j];
1462 
1463 		fp->rx_bd_cons = 0;
1464 
1465 		/* Activate BD ring */
1466 		/* Warning!
1467 		 * this will generate an interrupt (to the TSTORM)
1468 		 * must only be done after chip is initialized
1469 		 */
1470 		bnx2x_update_rx_prod(bp, fp, fp->rx_bd_prod, fp->rx_comp_prod,
1471 				     fp->rx_sge_prod);
1472 
1473 		if (j != 0)
1474 			continue;
1475 
1476 		if (CHIP_IS_E1(bp)) {
1477 			REG_WR(bp, BAR_USTRORM_INTMEM +
1478 			       USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func),
1479 			       U64_LO(fp->rx_comp_mapping));
1480 			REG_WR(bp, BAR_USTRORM_INTMEM +
1481 			       USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func) + 4,
1482 			       U64_HI(fp->rx_comp_mapping));
1483 		}
1484 	}
1485 }
1486 
1487 static void bnx2x_free_tx_skbs_queue(struct bnx2x_fastpath *fp)
1488 {
1489 	u8 cos;
1490 	struct bnx2x *bp = fp->bp;
1491 
1492 	for_each_cos_in_tx_queue(fp, cos) {
1493 		struct bnx2x_fp_txdata *txdata = fp->txdata_ptr[cos];
1494 		unsigned pkts_compl = 0, bytes_compl = 0;
1495 
1496 		u16 sw_prod = txdata->tx_pkt_prod;
1497 		u16 sw_cons = txdata->tx_pkt_cons;
1498 
1499 		while (sw_cons != sw_prod) {
1500 			bnx2x_free_tx_pkt(bp, txdata, TX_BD(sw_cons),
1501 					  &pkts_compl, &bytes_compl);
1502 			sw_cons++;
1503 		}
1504 
1505 		netdev_tx_reset_queue(
1506 			netdev_get_tx_queue(bp->dev,
1507 					    txdata->txq_index));
1508 	}
1509 }
1510 
1511 static void bnx2x_free_tx_skbs_cnic(struct bnx2x *bp)
1512 {
1513 	int i;
1514 
1515 	for_each_tx_queue_cnic(bp, i) {
1516 		bnx2x_free_tx_skbs_queue(&bp->fp[i]);
1517 	}
1518 }
1519 
1520 static void bnx2x_free_tx_skbs(struct bnx2x *bp)
1521 {
1522 	int i;
1523 
1524 	for_each_eth_queue(bp, i) {
1525 		bnx2x_free_tx_skbs_queue(&bp->fp[i]);
1526 	}
1527 }
1528 
1529 static void bnx2x_free_rx_bds(struct bnx2x_fastpath *fp)
1530 {
1531 	struct bnx2x *bp = fp->bp;
1532 	int i;
1533 
1534 	/* ring wasn't allocated */
1535 	if (fp->rx_buf_ring == NULL)
1536 		return;
1537 
1538 	for (i = 0; i < NUM_RX_BD; i++) {
1539 		struct sw_rx_bd *rx_buf = &fp->rx_buf_ring[i];
1540 		u8 *data = rx_buf->data;
1541 
1542 		if (data == NULL)
1543 			continue;
1544 		dma_unmap_single(&bp->pdev->dev,
1545 				 dma_unmap_addr(rx_buf, mapping),
1546 				 fp->rx_buf_size, DMA_FROM_DEVICE);
1547 
1548 		rx_buf->data = NULL;
1549 		bnx2x_frag_free(fp, data);
1550 	}
1551 }
1552 
1553 static void bnx2x_free_rx_skbs_cnic(struct bnx2x *bp)
1554 {
1555 	int j;
1556 
1557 	for_each_rx_queue_cnic(bp, j) {
1558 		bnx2x_free_rx_bds(&bp->fp[j]);
1559 	}
1560 }
1561 
1562 static void bnx2x_free_rx_skbs(struct bnx2x *bp)
1563 {
1564 	int j;
1565 
1566 	for_each_eth_queue(bp, j) {
1567 		struct bnx2x_fastpath *fp = &bp->fp[j];
1568 
1569 		bnx2x_free_rx_bds(fp);
1570 
1571 		if (fp->mode != TPA_MODE_DISABLED)
1572 			bnx2x_free_tpa_pool(bp, fp, MAX_AGG_QS(bp));
1573 	}
1574 }
1575 
1576 static void bnx2x_free_skbs_cnic(struct bnx2x *bp)
1577 {
1578 	bnx2x_free_tx_skbs_cnic(bp);
1579 	bnx2x_free_rx_skbs_cnic(bp);
1580 }
1581 
1582 void bnx2x_free_skbs(struct bnx2x *bp)
1583 {
1584 	bnx2x_free_tx_skbs(bp);
1585 	bnx2x_free_rx_skbs(bp);
1586 }
1587 
1588 void bnx2x_update_max_mf_config(struct bnx2x *bp, u32 value)
1589 {
1590 	/* load old values */
1591 	u32 mf_cfg = bp->mf_config[BP_VN(bp)];
1592 
1593 	if (value != bnx2x_extract_max_cfg(bp, mf_cfg)) {
1594 		/* leave all but MAX value */
1595 		mf_cfg &= ~FUNC_MF_CFG_MAX_BW_MASK;
1596 
1597 		/* set new MAX value */
1598 		mf_cfg |= (value << FUNC_MF_CFG_MAX_BW_SHIFT)
1599 				& FUNC_MF_CFG_MAX_BW_MASK;
1600 
1601 		bnx2x_fw_command(bp, DRV_MSG_CODE_SET_MF_BW, mf_cfg);
1602 	}
1603 }
1604 
1605 /**
1606  * bnx2x_free_msix_irqs - free previously requested MSI-X IRQ vectors
1607  *
1608  * @bp:		driver handle
1609  * @nvecs:	number of vectors to be released
1610  */
1611 static void bnx2x_free_msix_irqs(struct bnx2x *bp, int nvecs)
1612 {
1613 	int i, offset = 0;
1614 
1615 	if (nvecs == offset)
1616 		return;
1617 
1618 	/* VFs don't have a default SB */
1619 	if (IS_PF(bp)) {
1620 		free_irq(bp->msix_table[offset].vector, bp->dev);
1621 		DP(NETIF_MSG_IFDOWN, "released sp irq (%d)\n",
1622 		   bp->msix_table[offset].vector);
1623 		offset++;
1624 	}
1625 
1626 	if (CNIC_SUPPORT(bp)) {
1627 		if (nvecs == offset)
1628 			return;
1629 		offset++;
1630 	}
1631 
1632 	for_each_eth_queue(bp, i) {
1633 		if (nvecs == offset)
1634 			return;
1635 		DP(NETIF_MSG_IFDOWN, "about to release fp #%d->%d irq\n",
1636 		   i, bp->msix_table[offset].vector);
1637 
1638 		free_irq(bp->msix_table[offset++].vector, &bp->fp[i]);
1639 	}
1640 }
1641 
1642 void bnx2x_free_irq(struct bnx2x *bp)
1643 {
1644 	if (bp->flags & USING_MSIX_FLAG &&
1645 	    !(bp->flags & USING_SINGLE_MSIX_FLAG)) {
1646 		int nvecs = BNX2X_NUM_ETH_QUEUES(bp) + CNIC_SUPPORT(bp);
1647 
1648 		/* vfs don't have a default status block */
1649 		if (IS_PF(bp))
1650 			nvecs++;
1651 
1652 		bnx2x_free_msix_irqs(bp, nvecs);
1653 	} else {
1654 		free_irq(bp->dev->irq, bp->dev);
1655 	}
1656 }
1657 
1658 int bnx2x_enable_msix(struct bnx2x *bp)
1659 {
1660 	int msix_vec = 0, i, rc;
1661 
1662 	/* VFs don't have a default status block */
1663 	if (IS_PF(bp)) {
1664 		bp->msix_table[msix_vec].entry = msix_vec;
1665 		BNX2X_DEV_INFO("msix_table[0].entry = %d (slowpath)\n",
1666 			       bp->msix_table[0].entry);
1667 		msix_vec++;
1668 	}
1669 
1670 	/* Cnic requires an msix vector for itself */
1671 	if (CNIC_SUPPORT(bp)) {
1672 		bp->msix_table[msix_vec].entry = msix_vec;
1673 		BNX2X_DEV_INFO("msix_table[%d].entry = %d (CNIC)\n",
1674 			       msix_vec, bp->msix_table[msix_vec].entry);
1675 		msix_vec++;
1676 	}
1677 
1678 	/* We need separate vectors for ETH queues only (not FCoE) */
1679 	for_each_eth_queue(bp, i) {
1680 		bp->msix_table[msix_vec].entry = msix_vec;
1681 		BNX2X_DEV_INFO("msix_table[%d].entry = %d (fastpath #%u)\n",
1682 			       msix_vec, msix_vec, i);
1683 		msix_vec++;
1684 	}
1685 
1686 	DP(BNX2X_MSG_SP, "about to request enable msix with %d vectors\n",
1687 	   msix_vec);
1688 
1689 	rc = pci_enable_msix_range(bp->pdev, &bp->msix_table[0],
1690 				   BNX2X_MIN_MSIX_VEC_CNT(bp), msix_vec);
1691 	/*
1692 	 * reconfigure number of tx/rx queues according to available
1693 	 * MSI-X vectors
1694 	 */
1695 	if (rc == -ENOSPC) {
1696 		/* Get by with single vector */
1697 		rc = pci_enable_msix_range(bp->pdev, &bp->msix_table[0], 1, 1);
1698 		if (rc < 0) {
1699 			BNX2X_DEV_INFO("Single MSI-X is not attainable rc %d\n",
1700 				       rc);
1701 			goto no_msix;
1702 		}
1703 
1704 		BNX2X_DEV_INFO("Using single MSI-X vector\n");
1705 		bp->flags |= USING_SINGLE_MSIX_FLAG;
1706 
1707 		BNX2X_DEV_INFO("set number of queues to 1\n");
1708 		bp->num_ethernet_queues = 1;
1709 		bp->num_queues = bp->num_ethernet_queues + bp->num_cnic_queues;
1710 	} else if (rc < 0) {
1711 		BNX2X_DEV_INFO("MSI-X is not attainable rc %d\n", rc);
1712 		goto no_msix;
1713 	} else if (rc < msix_vec) {
1714 		/* how less vectors we will have? */
1715 		int diff = msix_vec - rc;
1716 
1717 		BNX2X_DEV_INFO("Trying to use less MSI-X vectors: %d\n", rc);
1718 
1719 		/*
1720 		 * decrease number of queues by number of unallocated entries
1721 		 */
1722 		bp->num_ethernet_queues -= diff;
1723 		bp->num_queues = bp->num_ethernet_queues + bp->num_cnic_queues;
1724 
1725 		BNX2X_DEV_INFO("New queue configuration set: %d\n",
1726 			       bp->num_queues);
1727 	}
1728 
1729 	bp->flags |= USING_MSIX_FLAG;
1730 
1731 	return 0;
1732 
1733 no_msix:
1734 	/* fall to INTx if not enough memory */
1735 	if (rc == -ENOMEM)
1736 		bp->flags |= DISABLE_MSI_FLAG;
1737 
1738 	return rc;
1739 }
1740 
1741 static int bnx2x_req_msix_irqs(struct bnx2x *bp)
1742 {
1743 	int i, rc, offset = 0;
1744 
1745 	/* no default status block for vf */
1746 	if (IS_PF(bp)) {
1747 		rc = request_irq(bp->msix_table[offset++].vector,
1748 				 bnx2x_msix_sp_int, 0,
1749 				 bp->dev->name, bp->dev);
1750 		if (rc) {
1751 			BNX2X_ERR("request sp irq failed\n");
1752 			return -EBUSY;
1753 		}
1754 	}
1755 
1756 	if (CNIC_SUPPORT(bp))
1757 		offset++;
1758 
1759 	for_each_eth_queue(bp, i) {
1760 		struct bnx2x_fastpath *fp = &bp->fp[i];
1761 		snprintf(fp->name, sizeof(fp->name), "%s-fp-%d",
1762 			 bp->dev->name, i);
1763 
1764 		rc = request_irq(bp->msix_table[offset].vector,
1765 				 bnx2x_msix_fp_int, 0, fp->name, fp);
1766 		if (rc) {
1767 			BNX2X_ERR("request fp #%d irq (%d) failed  rc %d\n", i,
1768 			      bp->msix_table[offset].vector, rc);
1769 			bnx2x_free_msix_irqs(bp, offset);
1770 			return -EBUSY;
1771 		}
1772 
1773 		offset++;
1774 	}
1775 
1776 	i = BNX2X_NUM_ETH_QUEUES(bp);
1777 	if (IS_PF(bp)) {
1778 		offset = 1 + CNIC_SUPPORT(bp);
1779 		netdev_info(bp->dev,
1780 			    "using MSI-X  IRQs: sp %d  fp[%d] %d ... fp[%d] %d\n",
1781 			    bp->msix_table[0].vector,
1782 			    0, bp->msix_table[offset].vector,
1783 			    i - 1, bp->msix_table[offset + i - 1].vector);
1784 	} else {
1785 		offset = CNIC_SUPPORT(bp);
1786 		netdev_info(bp->dev,
1787 			    "using MSI-X  IRQs: fp[%d] %d ... fp[%d] %d\n",
1788 			    0, bp->msix_table[offset].vector,
1789 			    i - 1, bp->msix_table[offset + i - 1].vector);
1790 	}
1791 	return 0;
1792 }
1793 
1794 int bnx2x_enable_msi(struct bnx2x *bp)
1795 {
1796 	int rc;
1797 
1798 	rc = pci_enable_msi(bp->pdev);
1799 	if (rc) {
1800 		BNX2X_DEV_INFO("MSI is not attainable\n");
1801 		return -1;
1802 	}
1803 	bp->flags |= USING_MSI_FLAG;
1804 
1805 	return 0;
1806 }
1807 
1808 static int bnx2x_req_irq(struct bnx2x *bp)
1809 {
1810 	unsigned long flags;
1811 	unsigned int irq;
1812 
1813 	if (bp->flags & (USING_MSI_FLAG | USING_MSIX_FLAG))
1814 		flags = 0;
1815 	else
1816 		flags = IRQF_SHARED;
1817 
1818 	if (bp->flags & USING_MSIX_FLAG)
1819 		irq = bp->msix_table[0].vector;
1820 	else
1821 		irq = bp->pdev->irq;
1822 
1823 	return request_irq(irq, bnx2x_interrupt, flags, bp->dev->name, bp->dev);
1824 }
1825 
1826 static int bnx2x_setup_irqs(struct bnx2x *bp)
1827 {
1828 	int rc = 0;
1829 	if (bp->flags & USING_MSIX_FLAG &&
1830 	    !(bp->flags & USING_SINGLE_MSIX_FLAG)) {
1831 		rc = bnx2x_req_msix_irqs(bp);
1832 		if (rc)
1833 			return rc;
1834 	} else {
1835 		rc = bnx2x_req_irq(bp);
1836 		if (rc) {
1837 			BNX2X_ERR("IRQ request failed  rc %d, aborting\n", rc);
1838 			return rc;
1839 		}
1840 		if (bp->flags & USING_MSI_FLAG) {
1841 			bp->dev->irq = bp->pdev->irq;
1842 			netdev_info(bp->dev, "using MSI IRQ %d\n",
1843 				    bp->dev->irq);
1844 		}
1845 		if (bp->flags & USING_MSIX_FLAG) {
1846 			bp->dev->irq = bp->msix_table[0].vector;
1847 			netdev_info(bp->dev, "using MSIX IRQ %d\n",
1848 				    bp->dev->irq);
1849 		}
1850 	}
1851 
1852 	return 0;
1853 }
1854 
1855 static void bnx2x_napi_enable_cnic(struct bnx2x *bp)
1856 {
1857 	int i;
1858 
1859 	for_each_rx_queue_cnic(bp, i) {
1860 		napi_enable(&bnx2x_fp(bp, i, napi));
1861 	}
1862 }
1863 
1864 static void bnx2x_napi_enable(struct bnx2x *bp)
1865 {
1866 	int i;
1867 
1868 	for_each_eth_queue(bp, i) {
1869 		napi_enable(&bnx2x_fp(bp, i, napi));
1870 	}
1871 }
1872 
1873 static void bnx2x_napi_disable_cnic(struct bnx2x *bp)
1874 {
1875 	int i;
1876 
1877 	for_each_rx_queue_cnic(bp, i) {
1878 		napi_disable(&bnx2x_fp(bp, i, napi));
1879 	}
1880 }
1881 
1882 static void bnx2x_napi_disable(struct bnx2x *bp)
1883 {
1884 	int i;
1885 
1886 	for_each_eth_queue(bp, i) {
1887 		napi_disable(&bnx2x_fp(bp, i, napi));
1888 	}
1889 }
1890 
1891 void bnx2x_netif_start(struct bnx2x *bp)
1892 {
1893 	if (netif_running(bp->dev)) {
1894 		bnx2x_napi_enable(bp);
1895 		if (CNIC_LOADED(bp))
1896 			bnx2x_napi_enable_cnic(bp);
1897 		bnx2x_int_enable(bp);
1898 		if (bp->state == BNX2X_STATE_OPEN)
1899 			netif_tx_wake_all_queues(bp->dev);
1900 	}
1901 }
1902 
1903 void bnx2x_netif_stop(struct bnx2x *bp, int disable_hw)
1904 {
1905 	bnx2x_int_disable_sync(bp, disable_hw);
1906 	bnx2x_napi_disable(bp);
1907 	if (CNIC_LOADED(bp))
1908 		bnx2x_napi_disable_cnic(bp);
1909 }
1910 
1911 u16 bnx2x_select_queue(struct net_device *dev, struct sk_buff *skb,
1912 		       void *accel_priv, select_queue_fallback_t fallback)
1913 {
1914 	struct bnx2x *bp = netdev_priv(dev);
1915 
1916 	if (CNIC_LOADED(bp) && !NO_FCOE(bp)) {
1917 		struct ethhdr *hdr = (struct ethhdr *)skb->data;
1918 		u16 ether_type = ntohs(hdr->h_proto);
1919 
1920 		/* Skip VLAN tag if present */
1921 		if (ether_type == ETH_P_8021Q) {
1922 			struct vlan_ethhdr *vhdr =
1923 				(struct vlan_ethhdr *)skb->data;
1924 
1925 			ether_type = ntohs(vhdr->h_vlan_encapsulated_proto);
1926 		}
1927 
1928 		/* If ethertype is FCoE or FIP - use FCoE ring */
1929 		if ((ether_type == ETH_P_FCOE) || (ether_type == ETH_P_FIP))
1930 			return bnx2x_fcoe_tx(bp, txq_index);
1931 	}
1932 
1933 	/* select a non-FCoE queue */
1934 	return fallback(dev, skb) % BNX2X_NUM_ETH_QUEUES(bp);
1935 }
1936 
1937 void bnx2x_set_num_queues(struct bnx2x *bp)
1938 {
1939 	/* RSS queues */
1940 	bp->num_ethernet_queues = bnx2x_calc_num_queues(bp);
1941 
1942 	/* override in STORAGE SD modes */
1943 	if (IS_MF_STORAGE_ONLY(bp))
1944 		bp->num_ethernet_queues = 1;
1945 
1946 	/* Add special queues */
1947 	bp->num_cnic_queues = CNIC_SUPPORT(bp); /* For FCOE */
1948 	bp->num_queues = bp->num_ethernet_queues + bp->num_cnic_queues;
1949 
1950 	BNX2X_DEV_INFO("set number of queues to %d\n", bp->num_queues);
1951 }
1952 
1953 /**
1954  * bnx2x_set_real_num_queues - configure netdev->real_num_[tx,rx]_queues
1955  *
1956  * @bp:		Driver handle
1957  *
1958  * We currently support for at most 16 Tx queues for each CoS thus we will
1959  * allocate a multiple of 16 for ETH L2 rings according to the value of the
1960  * bp->max_cos.
1961  *
1962  * If there is an FCoE L2 queue the appropriate Tx queue will have the next
1963  * index after all ETH L2 indices.
1964  *
1965  * If the actual number of Tx queues (for each CoS) is less than 16 then there
1966  * will be the holes at the end of each group of 16 ETh L2 indices (0..15,
1967  * 16..31,...) with indices that are not coupled with any real Tx queue.
1968  *
1969  * The proper configuration of skb->queue_mapping is handled by
1970  * bnx2x_select_queue() and __skb_tx_hash().
1971  *
1972  * bnx2x_setup_tc() takes care of the proper TC mappings so that __skb_tx_hash()
1973  * will return a proper Tx index if TC is enabled (netdev->num_tc > 0).
1974  */
1975 static int bnx2x_set_real_num_queues(struct bnx2x *bp, int include_cnic)
1976 {
1977 	int rc, tx, rx;
1978 
1979 	tx = BNX2X_NUM_ETH_QUEUES(bp) * bp->max_cos;
1980 	rx = BNX2X_NUM_ETH_QUEUES(bp);
1981 
1982 /* account for fcoe queue */
1983 	if (include_cnic && !NO_FCOE(bp)) {
1984 		rx++;
1985 		tx++;
1986 	}
1987 
1988 	rc = netif_set_real_num_tx_queues(bp->dev, tx);
1989 	if (rc) {
1990 		BNX2X_ERR("Failed to set real number of Tx queues: %d\n", rc);
1991 		return rc;
1992 	}
1993 	rc = netif_set_real_num_rx_queues(bp->dev, rx);
1994 	if (rc) {
1995 		BNX2X_ERR("Failed to set real number of Rx queues: %d\n", rc);
1996 		return rc;
1997 	}
1998 
1999 	DP(NETIF_MSG_IFUP, "Setting real num queues to (tx, rx) (%d, %d)\n",
2000 			  tx, rx);
2001 
2002 	return rc;
2003 }
2004 
2005 static void bnx2x_set_rx_buf_size(struct bnx2x *bp)
2006 {
2007 	int i;
2008 
2009 	for_each_queue(bp, i) {
2010 		struct bnx2x_fastpath *fp = &bp->fp[i];
2011 		u32 mtu;
2012 
2013 		/* Always use a mini-jumbo MTU for the FCoE L2 ring */
2014 		if (IS_FCOE_IDX(i))
2015 			/*
2016 			 * Although there are no IP frames expected to arrive to
2017 			 * this ring we still want to add an
2018 			 * IP_HEADER_ALIGNMENT_PADDING to prevent a buffer
2019 			 * overrun attack.
2020 			 */
2021 			mtu = BNX2X_FCOE_MINI_JUMBO_MTU;
2022 		else
2023 			mtu = bp->dev->mtu;
2024 		fp->rx_buf_size = BNX2X_FW_RX_ALIGN_START +
2025 				  IP_HEADER_ALIGNMENT_PADDING +
2026 				  ETH_OVREHEAD +
2027 				  mtu +
2028 				  BNX2X_FW_RX_ALIGN_END;
2029 		/* Note : rx_buf_size doesn't take into account NET_SKB_PAD */
2030 		if (fp->rx_buf_size + NET_SKB_PAD <= PAGE_SIZE)
2031 			fp->rx_frag_size = fp->rx_buf_size + NET_SKB_PAD;
2032 		else
2033 			fp->rx_frag_size = 0;
2034 	}
2035 }
2036 
2037 static int bnx2x_init_rss(struct bnx2x *bp)
2038 {
2039 	int i;
2040 	u8 num_eth_queues = BNX2X_NUM_ETH_QUEUES(bp);
2041 
2042 	/* Prepare the initial contents for the indirection table if RSS is
2043 	 * enabled
2044 	 */
2045 	for (i = 0; i < sizeof(bp->rss_conf_obj.ind_table); i++)
2046 		bp->rss_conf_obj.ind_table[i] =
2047 			bp->fp->cl_id +
2048 			ethtool_rxfh_indir_default(i, num_eth_queues);
2049 
2050 	/*
2051 	 * For 57710 and 57711 SEARCHER configuration (rss_keys) is
2052 	 * per-port, so if explicit configuration is needed , do it only
2053 	 * for a PMF.
2054 	 *
2055 	 * For 57712 and newer on the other hand it's a per-function
2056 	 * configuration.
2057 	 */
2058 	return bnx2x_config_rss_eth(bp, bp->port.pmf || !CHIP_IS_E1x(bp));
2059 }
2060 
2061 int bnx2x_rss(struct bnx2x *bp, struct bnx2x_rss_config_obj *rss_obj,
2062 	      bool config_hash, bool enable)
2063 {
2064 	struct bnx2x_config_rss_params params = {NULL};
2065 
2066 	/* Although RSS is meaningless when there is a single HW queue we
2067 	 * still need it enabled in order to have HW Rx hash generated.
2068 	 *
2069 	 * if (!is_eth_multi(bp))
2070 	 *      bp->multi_mode = ETH_RSS_MODE_DISABLED;
2071 	 */
2072 
2073 	params.rss_obj = rss_obj;
2074 
2075 	__set_bit(RAMROD_COMP_WAIT, &params.ramrod_flags);
2076 
2077 	if (enable) {
2078 		__set_bit(BNX2X_RSS_MODE_REGULAR, &params.rss_flags);
2079 
2080 		/* RSS configuration */
2081 		__set_bit(BNX2X_RSS_IPV4, &params.rss_flags);
2082 		__set_bit(BNX2X_RSS_IPV4_TCP, &params.rss_flags);
2083 		__set_bit(BNX2X_RSS_IPV6, &params.rss_flags);
2084 		__set_bit(BNX2X_RSS_IPV6_TCP, &params.rss_flags);
2085 		if (rss_obj->udp_rss_v4)
2086 			__set_bit(BNX2X_RSS_IPV4_UDP, &params.rss_flags);
2087 		if (rss_obj->udp_rss_v6)
2088 			__set_bit(BNX2X_RSS_IPV6_UDP, &params.rss_flags);
2089 
2090 		if (!CHIP_IS_E1x(bp)) {
2091 			/* valid only for TUNN_MODE_VXLAN tunnel mode */
2092 			__set_bit(BNX2X_RSS_IPV4_VXLAN, &params.rss_flags);
2093 			__set_bit(BNX2X_RSS_IPV6_VXLAN, &params.rss_flags);
2094 
2095 			/* valid only for TUNN_MODE_GRE tunnel mode */
2096 			__set_bit(BNX2X_RSS_TUNN_INNER_HDRS, &params.rss_flags);
2097 		}
2098 	} else {
2099 		__set_bit(BNX2X_RSS_MODE_DISABLED, &params.rss_flags);
2100 	}
2101 
2102 	/* Hash bits */
2103 	params.rss_result_mask = MULTI_MASK;
2104 
2105 	memcpy(params.ind_table, rss_obj->ind_table, sizeof(params.ind_table));
2106 
2107 	if (config_hash) {
2108 		/* RSS keys */
2109 		netdev_rss_key_fill(params.rss_key, T_ETH_RSS_KEY * 4);
2110 		__set_bit(BNX2X_RSS_SET_SRCH, &params.rss_flags);
2111 	}
2112 
2113 	if (IS_PF(bp))
2114 		return bnx2x_config_rss(bp, &params);
2115 	else
2116 		return bnx2x_vfpf_config_rss(bp, &params);
2117 }
2118 
2119 static int bnx2x_init_hw(struct bnx2x *bp, u32 load_code)
2120 {
2121 	struct bnx2x_func_state_params func_params = {NULL};
2122 
2123 	/* Prepare parameters for function state transitions */
2124 	__set_bit(RAMROD_COMP_WAIT, &func_params.ramrod_flags);
2125 
2126 	func_params.f_obj = &bp->func_obj;
2127 	func_params.cmd = BNX2X_F_CMD_HW_INIT;
2128 
2129 	func_params.params.hw_init.load_phase = load_code;
2130 
2131 	return bnx2x_func_state_change(bp, &func_params);
2132 }
2133 
2134 /*
2135  * Cleans the object that have internal lists without sending
2136  * ramrods. Should be run when interrupts are disabled.
2137  */
2138 void bnx2x_squeeze_objects(struct bnx2x *bp)
2139 {
2140 	int rc;
2141 	unsigned long ramrod_flags = 0, vlan_mac_flags = 0;
2142 	struct bnx2x_mcast_ramrod_params rparam = {NULL};
2143 	struct bnx2x_vlan_mac_obj *mac_obj = &bp->sp_objs->mac_obj;
2144 
2145 	/***************** Cleanup MACs' object first *************************/
2146 
2147 	/* Wait for completion of requested */
2148 	__set_bit(RAMROD_COMP_WAIT, &ramrod_flags);
2149 	/* Perform a dry cleanup */
2150 	__set_bit(RAMROD_DRV_CLR_ONLY, &ramrod_flags);
2151 
2152 	/* Clean ETH primary MAC */
2153 	__set_bit(BNX2X_ETH_MAC, &vlan_mac_flags);
2154 	rc = mac_obj->delete_all(bp, &bp->sp_objs->mac_obj, &vlan_mac_flags,
2155 				 &ramrod_flags);
2156 	if (rc != 0)
2157 		BNX2X_ERR("Failed to clean ETH MACs: %d\n", rc);
2158 
2159 	/* Cleanup UC list */
2160 	vlan_mac_flags = 0;
2161 	__set_bit(BNX2X_UC_LIST_MAC, &vlan_mac_flags);
2162 	rc = mac_obj->delete_all(bp, mac_obj, &vlan_mac_flags,
2163 				 &ramrod_flags);
2164 	if (rc != 0)
2165 		BNX2X_ERR("Failed to clean UC list MACs: %d\n", rc);
2166 
2167 	/***************** Now clean mcast object *****************************/
2168 	rparam.mcast_obj = &bp->mcast_obj;
2169 	__set_bit(RAMROD_DRV_CLR_ONLY, &rparam.ramrod_flags);
2170 
2171 	/* Add a DEL command... - Since we're doing a driver cleanup only,
2172 	 * we take a lock surrounding both the initial send and the CONTs,
2173 	 * as we don't want a true completion to disrupt us in the middle.
2174 	 */
2175 	netif_addr_lock_bh(bp->dev);
2176 	rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_DEL);
2177 	if (rc < 0)
2178 		BNX2X_ERR("Failed to add a new DEL command to a multi-cast object: %d\n",
2179 			  rc);
2180 
2181 	/* ...and wait until all pending commands are cleared */
2182 	rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_CONT);
2183 	while (rc != 0) {
2184 		if (rc < 0) {
2185 			BNX2X_ERR("Failed to clean multi-cast object: %d\n",
2186 				  rc);
2187 			netif_addr_unlock_bh(bp->dev);
2188 			return;
2189 		}
2190 
2191 		rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_CONT);
2192 	}
2193 	netif_addr_unlock_bh(bp->dev);
2194 }
2195 
2196 #ifndef BNX2X_STOP_ON_ERROR
2197 #define LOAD_ERROR_EXIT(bp, label) \
2198 	do { \
2199 		(bp)->state = BNX2X_STATE_ERROR; \
2200 		goto label; \
2201 	} while (0)
2202 
2203 #define LOAD_ERROR_EXIT_CNIC(bp, label) \
2204 	do { \
2205 		bp->cnic_loaded = false; \
2206 		goto label; \
2207 	} while (0)
2208 #else /*BNX2X_STOP_ON_ERROR*/
2209 #define LOAD_ERROR_EXIT(bp, label) \
2210 	do { \
2211 		(bp)->state = BNX2X_STATE_ERROR; \
2212 		(bp)->panic = 1; \
2213 		return -EBUSY; \
2214 	} while (0)
2215 #define LOAD_ERROR_EXIT_CNIC(bp, label) \
2216 	do { \
2217 		bp->cnic_loaded = false; \
2218 		(bp)->panic = 1; \
2219 		return -EBUSY; \
2220 	} while (0)
2221 #endif /*BNX2X_STOP_ON_ERROR*/
2222 
2223 static void bnx2x_free_fw_stats_mem(struct bnx2x *bp)
2224 {
2225 	BNX2X_PCI_FREE(bp->fw_stats, bp->fw_stats_mapping,
2226 		       bp->fw_stats_data_sz + bp->fw_stats_req_sz);
2227 	return;
2228 }
2229 
2230 static int bnx2x_alloc_fw_stats_mem(struct bnx2x *bp)
2231 {
2232 	int num_groups, vf_headroom = 0;
2233 	int is_fcoe_stats = NO_FCOE(bp) ? 0 : 1;
2234 
2235 	/* number of queues for statistics is number of eth queues + FCoE */
2236 	u8 num_queue_stats = BNX2X_NUM_ETH_QUEUES(bp) + is_fcoe_stats;
2237 
2238 	/* Total number of FW statistics requests =
2239 	 * 1 for port stats + 1 for PF stats + potential 2 for FCoE (fcoe proper
2240 	 * and fcoe l2 queue) stats + num of queues (which includes another 1
2241 	 * for fcoe l2 queue if applicable)
2242 	 */
2243 	bp->fw_stats_num = 2 + is_fcoe_stats + num_queue_stats;
2244 
2245 	/* vf stats appear in the request list, but their data is allocated by
2246 	 * the VFs themselves. We don't include them in the bp->fw_stats_num as
2247 	 * it is used to determine where to place the vf stats queries in the
2248 	 * request struct
2249 	 */
2250 	if (IS_SRIOV(bp))
2251 		vf_headroom = bnx2x_vf_headroom(bp);
2252 
2253 	/* Request is built from stats_query_header and an array of
2254 	 * stats_query_cmd_group each of which contains
2255 	 * STATS_QUERY_CMD_COUNT rules. The real number or requests is
2256 	 * configured in the stats_query_header.
2257 	 */
2258 	num_groups =
2259 		(((bp->fw_stats_num + vf_headroom) / STATS_QUERY_CMD_COUNT) +
2260 		 (((bp->fw_stats_num + vf_headroom) % STATS_QUERY_CMD_COUNT) ?
2261 		 1 : 0));
2262 
2263 	DP(BNX2X_MSG_SP, "stats fw_stats_num %d, vf headroom %d, num_groups %d\n",
2264 	   bp->fw_stats_num, vf_headroom, num_groups);
2265 	bp->fw_stats_req_sz = sizeof(struct stats_query_header) +
2266 		num_groups * sizeof(struct stats_query_cmd_group);
2267 
2268 	/* Data for statistics requests + stats_counter
2269 	 * stats_counter holds per-STORM counters that are incremented
2270 	 * when STORM has finished with the current request.
2271 	 * memory for FCoE offloaded statistics are counted anyway,
2272 	 * even if they will not be sent.
2273 	 * VF stats are not accounted for here as the data of VF stats is stored
2274 	 * in memory allocated by the VF, not here.
2275 	 */
2276 	bp->fw_stats_data_sz = sizeof(struct per_port_stats) +
2277 		sizeof(struct per_pf_stats) +
2278 		sizeof(struct fcoe_statistics_params) +
2279 		sizeof(struct per_queue_stats) * num_queue_stats +
2280 		sizeof(struct stats_counter);
2281 
2282 	bp->fw_stats = BNX2X_PCI_ALLOC(&bp->fw_stats_mapping,
2283 				       bp->fw_stats_data_sz + bp->fw_stats_req_sz);
2284 	if (!bp->fw_stats)
2285 		goto alloc_mem_err;
2286 
2287 	/* Set shortcuts */
2288 	bp->fw_stats_req = (struct bnx2x_fw_stats_req *)bp->fw_stats;
2289 	bp->fw_stats_req_mapping = bp->fw_stats_mapping;
2290 	bp->fw_stats_data = (struct bnx2x_fw_stats_data *)
2291 		((u8 *)bp->fw_stats + bp->fw_stats_req_sz);
2292 	bp->fw_stats_data_mapping = bp->fw_stats_mapping +
2293 		bp->fw_stats_req_sz;
2294 
2295 	DP(BNX2X_MSG_SP, "statistics request base address set to %x %x\n",
2296 	   U64_HI(bp->fw_stats_req_mapping),
2297 	   U64_LO(bp->fw_stats_req_mapping));
2298 	DP(BNX2X_MSG_SP, "statistics data base address set to %x %x\n",
2299 	   U64_HI(bp->fw_stats_data_mapping),
2300 	   U64_LO(bp->fw_stats_data_mapping));
2301 	return 0;
2302 
2303 alloc_mem_err:
2304 	bnx2x_free_fw_stats_mem(bp);
2305 	BNX2X_ERR("Can't allocate FW stats memory\n");
2306 	return -ENOMEM;
2307 }
2308 
2309 /* send load request to mcp and analyze response */
2310 static int bnx2x_nic_load_request(struct bnx2x *bp, u32 *load_code)
2311 {
2312 	u32 param;
2313 
2314 	/* init fw_seq */
2315 	bp->fw_seq =
2316 		(SHMEM_RD(bp, func_mb[BP_FW_MB_IDX(bp)].drv_mb_header) &
2317 		 DRV_MSG_SEQ_NUMBER_MASK);
2318 	BNX2X_DEV_INFO("fw_seq 0x%08x\n", bp->fw_seq);
2319 
2320 	/* Get current FW pulse sequence */
2321 	bp->fw_drv_pulse_wr_seq =
2322 		(SHMEM_RD(bp, func_mb[BP_FW_MB_IDX(bp)].drv_pulse_mb) &
2323 		 DRV_PULSE_SEQ_MASK);
2324 	BNX2X_DEV_INFO("drv_pulse 0x%x\n", bp->fw_drv_pulse_wr_seq);
2325 
2326 	param = DRV_MSG_CODE_LOAD_REQ_WITH_LFA;
2327 
2328 	if (IS_MF_SD(bp) && bnx2x_port_after_undi(bp))
2329 		param |= DRV_MSG_CODE_LOAD_REQ_FORCE_LFA;
2330 
2331 	/* load request */
2332 	(*load_code) = bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_REQ, param);
2333 
2334 	/* if mcp fails to respond we must abort */
2335 	if (!(*load_code)) {
2336 		BNX2X_ERR("MCP response failure, aborting\n");
2337 		return -EBUSY;
2338 	}
2339 
2340 	/* If mcp refused (e.g. other port is in diagnostic mode) we
2341 	 * must abort
2342 	 */
2343 	if ((*load_code) == FW_MSG_CODE_DRV_LOAD_REFUSED) {
2344 		BNX2X_ERR("MCP refused load request, aborting\n");
2345 		return -EBUSY;
2346 	}
2347 	return 0;
2348 }
2349 
2350 /* check whether another PF has already loaded FW to chip. In
2351  * virtualized environments a pf from another VM may have already
2352  * initialized the device including loading FW
2353  */
2354 int bnx2x_compare_fw_ver(struct bnx2x *bp, u32 load_code, bool print_err)
2355 {
2356 	/* is another pf loaded on this engine? */
2357 	if (load_code != FW_MSG_CODE_DRV_LOAD_COMMON_CHIP &&
2358 	    load_code != FW_MSG_CODE_DRV_LOAD_COMMON) {
2359 		/* build my FW version dword */
2360 		u32 my_fw = (BCM_5710_FW_MAJOR_VERSION) +
2361 			(BCM_5710_FW_MINOR_VERSION << 8) +
2362 			(BCM_5710_FW_REVISION_VERSION << 16) +
2363 			(BCM_5710_FW_ENGINEERING_VERSION << 24);
2364 
2365 		/* read loaded FW from chip */
2366 		u32 loaded_fw = REG_RD(bp, XSEM_REG_PRAM);
2367 
2368 		DP(BNX2X_MSG_SP, "loaded fw %x, my fw %x\n",
2369 		   loaded_fw, my_fw);
2370 
2371 		/* abort nic load if version mismatch */
2372 		if (my_fw != loaded_fw) {
2373 			if (print_err)
2374 				BNX2X_ERR("bnx2x with FW %x was already loaded which mismatches my %x FW. Aborting\n",
2375 					  loaded_fw, my_fw);
2376 			else
2377 				BNX2X_DEV_INFO("bnx2x with FW %x was already loaded which mismatches my %x FW, possibly due to MF UNDI\n",
2378 					       loaded_fw, my_fw);
2379 			return -EBUSY;
2380 		}
2381 	}
2382 	return 0;
2383 }
2384 
2385 /* returns the "mcp load_code" according to global load_count array */
2386 static int bnx2x_nic_load_no_mcp(struct bnx2x *bp, int port)
2387 {
2388 	int path = BP_PATH(bp);
2389 
2390 	DP(NETIF_MSG_IFUP, "NO MCP - load counts[%d]      %d, %d, %d\n",
2391 	   path, bnx2x_load_count[path][0], bnx2x_load_count[path][1],
2392 	   bnx2x_load_count[path][2]);
2393 	bnx2x_load_count[path][0]++;
2394 	bnx2x_load_count[path][1 + port]++;
2395 	DP(NETIF_MSG_IFUP, "NO MCP - new load counts[%d]  %d, %d, %d\n",
2396 	   path, bnx2x_load_count[path][0], bnx2x_load_count[path][1],
2397 	   bnx2x_load_count[path][2]);
2398 	if (bnx2x_load_count[path][0] == 1)
2399 		return FW_MSG_CODE_DRV_LOAD_COMMON;
2400 	else if (bnx2x_load_count[path][1 + port] == 1)
2401 		return FW_MSG_CODE_DRV_LOAD_PORT;
2402 	else
2403 		return FW_MSG_CODE_DRV_LOAD_FUNCTION;
2404 }
2405 
2406 /* mark PMF if applicable */
2407 static void bnx2x_nic_load_pmf(struct bnx2x *bp, u32 load_code)
2408 {
2409 	if ((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
2410 	    (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP) ||
2411 	    (load_code == FW_MSG_CODE_DRV_LOAD_PORT)) {
2412 		bp->port.pmf = 1;
2413 		/* We need the barrier to ensure the ordering between the
2414 		 * writing to bp->port.pmf here and reading it from the
2415 		 * bnx2x_periodic_task().
2416 		 */
2417 		smp_mb();
2418 	} else {
2419 		bp->port.pmf = 0;
2420 	}
2421 
2422 	DP(NETIF_MSG_LINK, "pmf %d\n", bp->port.pmf);
2423 }
2424 
2425 static void bnx2x_nic_load_afex_dcc(struct bnx2x *bp, int load_code)
2426 {
2427 	if (((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
2428 	     (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP)) &&
2429 	    (bp->common.shmem2_base)) {
2430 		if (SHMEM2_HAS(bp, dcc_support))
2431 			SHMEM2_WR(bp, dcc_support,
2432 				  (SHMEM_DCC_SUPPORT_DISABLE_ENABLE_PF_TLV |
2433 				   SHMEM_DCC_SUPPORT_BANDWIDTH_ALLOCATION_TLV));
2434 		if (SHMEM2_HAS(bp, afex_driver_support))
2435 			SHMEM2_WR(bp, afex_driver_support,
2436 				  SHMEM_AFEX_SUPPORTED_VERSION_ONE);
2437 	}
2438 
2439 	/* Set AFEX default VLAN tag to an invalid value */
2440 	bp->afex_def_vlan_tag = -1;
2441 }
2442 
2443 /**
2444  * bnx2x_bz_fp - zero content of the fastpath structure.
2445  *
2446  * @bp:		driver handle
2447  * @index:	fastpath index to be zeroed
2448  *
2449  * Makes sure the contents of the bp->fp[index].napi is kept
2450  * intact.
2451  */
2452 static void bnx2x_bz_fp(struct bnx2x *bp, int index)
2453 {
2454 	struct bnx2x_fastpath *fp = &bp->fp[index];
2455 	int cos;
2456 	struct napi_struct orig_napi = fp->napi;
2457 	struct bnx2x_agg_info *orig_tpa_info = fp->tpa_info;
2458 
2459 	/* bzero bnx2x_fastpath contents */
2460 	if (fp->tpa_info)
2461 		memset(fp->tpa_info, 0, ETH_MAX_AGGREGATION_QUEUES_E1H_E2 *
2462 		       sizeof(struct bnx2x_agg_info));
2463 	memset(fp, 0, sizeof(*fp));
2464 
2465 	/* Restore the NAPI object as it has been already initialized */
2466 	fp->napi = orig_napi;
2467 	fp->tpa_info = orig_tpa_info;
2468 	fp->bp = bp;
2469 	fp->index = index;
2470 	if (IS_ETH_FP(fp))
2471 		fp->max_cos = bp->max_cos;
2472 	else
2473 		/* Special queues support only one CoS */
2474 		fp->max_cos = 1;
2475 
2476 	/* Init txdata pointers */
2477 	if (IS_FCOE_FP(fp))
2478 		fp->txdata_ptr[0] = &bp->bnx2x_txq[FCOE_TXQ_IDX(bp)];
2479 	if (IS_ETH_FP(fp))
2480 		for_each_cos_in_tx_queue(fp, cos)
2481 			fp->txdata_ptr[cos] = &bp->bnx2x_txq[cos *
2482 				BNX2X_NUM_ETH_QUEUES(bp) + index];
2483 
2484 	/* set the tpa flag for each queue. The tpa flag determines the queue
2485 	 * minimal size so it must be set prior to queue memory allocation
2486 	 */
2487 	if (bp->dev->features & NETIF_F_LRO)
2488 		fp->mode = TPA_MODE_LRO;
2489 	else if (bp->dev->features & NETIF_F_GRO &&
2490 		 bnx2x_mtu_allows_gro(bp->dev->mtu))
2491 		fp->mode = TPA_MODE_GRO;
2492 	else
2493 		fp->mode = TPA_MODE_DISABLED;
2494 
2495 	/* We don't want TPA if it's disabled in bp
2496 	 * or if this is an FCoE L2 ring.
2497 	 */
2498 	if (bp->disable_tpa || IS_FCOE_FP(fp))
2499 		fp->mode = TPA_MODE_DISABLED;
2500 }
2501 
2502 void bnx2x_set_os_driver_state(struct bnx2x *bp, u32 state)
2503 {
2504 	u32 cur;
2505 
2506 	if (!IS_MF_BD(bp) || !SHMEM2_HAS(bp, os_driver_state) || IS_VF(bp))
2507 		return;
2508 
2509 	cur = SHMEM2_RD(bp, os_driver_state[BP_FW_MB_IDX(bp)]);
2510 	DP(NETIF_MSG_IFUP, "Driver state %08x-->%08x\n",
2511 	   cur, state);
2512 
2513 	SHMEM2_WR(bp, os_driver_state[BP_FW_MB_IDX(bp)], state);
2514 }
2515 
2516 int bnx2x_load_cnic(struct bnx2x *bp)
2517 {
2518 	int i, rc, port = BP_PORT(bp);
2519 
2520 	DP(NETIF_MSG_IFUP, "Starting CNIC-related load\n");
2521 
2522 	mutex_init(&bp->cnic_mutex);
2523 
2524 	if (IS_PF(bp)) {
2525 		rc = bnx2x_alloc_mem_cnic(bp);
2526 		if (rc) {
2527 			BNX2X_ERR("Unable to allocate bp memory for cnic\n");
2528 			LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic0);
2529 		}
2530 	}
2531 
2532 	rc = bnx2x_alloc_fp_mem_cnic(bp);
2533 	if (rc) {
2534 		BNX2X_ERR("Unable to allocate memory for cnic fps\n");
2535 		LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic0);
2536 	}
2537 
2538 	/* Update the number of queues with the cnic queues */
2539 	rc = bnx2x_set_real_num_queues(bp, 1);
2540 	if (rc) {
2541 		BNX2X_ERR("Unable to set real_num_queues including cnic\n");
2542 		LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic0);
2543 	}
2544 
2545 	/* Add all CNIC NAPI objects */
2546 	bnx2x_add_all_napi_cnic(bp);
2547 	DP(NETIF_MSG_IFUP, "cnic napi added\n");
2548 	bnx2x_napi_enable_cnic(bp);
2549 
2550 	rc = bnx2x_init_hw_func_cnic(bp);
2551 	if (rc)
2552 		LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic1);
2553 
2554 	bnx2x_nic_init_cnic(bp);
2555 
2556 	if (IS_PF(bp)) {
2557 		/* Enable Timer scan */
2558 		REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + port*4, 1);
2559 
2560 		/* setup cnic queues */
2561 		for_each_cnic_queue(bp, i) {
2562 			rc = bnx2x_setup_queue(bp, &bp->fp[i], 0);
2563 			if (rc) {
2564 				BNX2X_ERR("Queue setup failed\n");
2565 				LOAD_ERROR_EXIT(bp, load_error_cnic2);
2566 			}
2567 		}
2568 	}
2569 
2570 	/* Initialize Rx filter. */
2571 	bnx2x_set_rx_mode_inner(bp);
2572 
2573 	/* re-read iscsi info */
2574 	bnx2x_get_iscsi_info(bp);
2575 	bnx2x_setup_cnic_irq_info(bp);
2576 	bnx2x_setup_cnic_info(bp);
2577 	bp->cnic_loaded = true;
2578 	if (bp->state == BNX2X_STATE_OPEN)
2579 		bnx2x_cnic_notify(bp, CNIC_CTL_START_CMD);
2580 
2581 	DP(NETIF_MSG_IFUP, "Ending successfully CNIC-related load\n");
2582 
2583 	return 0;
2584 
2585 #ifndef BNX2X_STOP_ON_ERROR
2586 load_error_cnic2:
2587 	/* Disable Timer scan */
2588 	REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + port*4, 0);
2589 
2590 load_error_cnic1:
2591 	bnx2x_napi_disable_cnic(bp);
2592 	/* Update the number of queues without the cnic queues */
2593 	if (bnx2x_set_real_num_queues(bp, 0))
2594 		BNX2X_ERR("Unable to set real_num_queues not including cnic\n");
2595 load_error_cnic0:
2596 	BNX2X_ERR("CNIC-related load failed\n");
2597 	bnx2x_free_fp_mem_cnic(bp);
2598 	bnx2x_free_mem_cnic(bp);
2599 	return rc;
2600 #endif /* ! BNX2X_STOP_ON_ERROR */
2601 }
2602 
2603 /* must be called with rtnl_lock */
2604 int bnx2x_nic_load(struct bnx2x *bp, int load_mode)
2605 {
2606 	int port = BP_PORT(bp);
2607 	int i, rc = 0, load_code = 0;
2608 
2609 	DP(NETIF_MSG_IFUP, "Starting NIC load\n");
2610 	DP(NETIF_MSG_IFUP,
2611 	   "CNIC is %s\n", CNIC_ENABLED(bp) ? "enabled" : "disabled");
2612 
2613 #ifdef BNX2X_STOP_ON_ERROR
2614 	if (unlikely(bp->panic)) {
2615 		BNX2X_ERR("Can't load NIC when there is panic\n");
2616 		return -EPERM;
2617 	}
2618 #endif
2619 
2620 	bp->state = BNX2X_STATE_OPENING_WAIT4_LOAD;
2621 
2622 	/* zero the structure w/o any lock, before SP handler is initialized */
2623 	memset(&bp->last_reported_link, 0, sizeof(bp->last_reported_link));
2624 	__set_bit(BNX2X_LINK_REPORT_LINK_DOWN,
2625 		&bp->last_reported_link.link_report_flags);
2626 
2627 	if (IS_PF(bp))
2628 		/* must be called before memory allocation and HW init */
2629 		bnx2x_ilt_set_info(bp);
2630 
2631 	/*
2632 	 * Zero fastpath structures preserving invariants like napi, which are
2633 	 * allocated only once, fp index, max_cos, bp pointer.
2634 	 * Also set fp->mode and txdata_ptr.
2635 	 */
2636 	DP(NETIF_MSG_IFUP, "num queues: %d", bp->num_queues);
2637 	for_each_queue(bp, i)
2638 		bnx2x_bz_fp(bp, i);
2639 	memset(bp->bnx2x_txq, 0, (BNX2X_MAX_RSS_COUNT(bp) * BNX2X_MULTI_TX_COS +
2640 				  bp->num_cnic_queues) *
2641 				  sizeof(struct bnx2x_fp_txdata));
2642 
2643 	bp->fcoe_init = false;
2644 
2645 	/* Set the receive queues buffer size */
2646 	bnx2x_set_rx_buf_size(bp);
2647 
2648 	if (IS_PF(bp)) {
2649 		rc = bnx2x_alloc_mem(bp);
2650 		if (rc) {
2651 			BNX2X_ERR("Unable to allocate bp memory\n");
2652 			return rc;
2653 		}
2654 	}
2655 
2656 	/* need to be done after alloc mem, since it's self adjusting to amount
2657 	 * of memory available for RSS queues
2658 	 */
2659 	rc = bnx2x_alloc_fp_mem(bp);
2660 	if (rc) {
2661 		BNX2X_ERR("Unable to allocate memory for fps\n");
2662 		LOAD_ERROR_EXIT(bp, load_error0);
2663 	}
2664 
2665 	/* Allocated memory for FW statistics  */
2666 	if (bnx2x_alloc_fw_stats_mem(bp))
2667 		LOAD_ERROR_EXIT(bp, load_error0);
2668 
2669 	/* request pf to initialize status blocks */
2670 	if (IS_VF(bp)) {
2671 		rc = bnx2x_vfpf_init(bp);
2672 		if (rc)
2673 			LOAD_ERROR_EXIT(bp, load_error0);
2674 	}
2675 
2676 	/* As long as bnx2x_alloc_mem() may possibly update
2677 	 * bp->num_queues, bnx2x_set_real_num_queues() should always
2678 	 * come after it. At this stage cnic queues are not counted.
2679 	 */
2680 	rc = bnx2x_set_real_num_queues(bp, 0);
2681 	if (rc) {
2682 		BNX2X_ERR("Unable to set real_num_queues\n");
2683 		LOAD_ERROR_EXIT(bp, load_error0);
2684 	}
2685 
2686 	/* configure multi cos mappings in kernel.
2687 	 * this configuration may be overridden by a multi class queue
2688 	 * discipline or by a dcbx negotiation result.
2689 	 */
2690 	bnx2x_setup_tc(bp->dev, bp->max_cos);
2691 
2692 	/* Add all NAPI objects */
2693 	bnx2x_add_all_napi(bp);
2694 	DP(NETIF_MSG_IFUP, "napi added\n");
2695 	bnx2x_napi_enable(bp);
2696 
2697 	if (IS_PF(bp)) {
2698 		/* set pf load just before approaching the MCP */
2699 		bnx2x_set_pf_load(bp);
2700 
2701 		/* if mcp exists send load request and analyze response */
2702 		if (!BP_NOMCP(bp)) {
2703 			/* attempt to load pf */
2704 			rc = bnx2x_nic_load_request(bp, &load_code);
2705 			if (rc)
2706 				LOAD_ERROR_EXIT(bp, load_error1);
2707 
2708 			/* what did mcp say? */
2709 			rc = bnx2x_compare_fw_ver(bp, load_code, true);
2710 			if (rc) {
2711 				bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2712 				LOAD_ERROR_EXIT(bp, load_error2);
2713 			}
2714 		} else {
2715 			load_code = bnx2x_nic_load_no_mcp(bp, port);
2716 		}
2717 
2718 		/* mark pmf if applicable */
2719 		bnx2x_nic_load_pmf(bp, load_code);
2720 
2721 		/* Init Function state controlling object */
2722 		bnx2x__init_func_obj(bp);
2723 
2724 		/* Initialize HW */
2725 		rc = bnx2x_init_hw(bp, load_code);
2726 		if (rc) {
2727 			BNX2X_ERR("HW init failed, aborting\n");
2728 			bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2729 			LOAD_ERROR_EXIT(bp, load_error2);
2730 		}
2731 	}
2732 
2733 	bnx2x_pre_irq_nic_init(bp);
2734 
2735 	/* Connect to IRQs */
2736 	rc = bnx2x_setup_irqs(bp);
2737 	if (rc) {
2738 		BNX2X_ERR("setup irqs failed\n");
2739 		if (IS_PF(bp))
2740 			bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2741 		LOAD_ERROR_EXIT(bp, load_error2);
2742 	}
2743 
2744 	/* Init per-function objects */
2745 	if (IS_PF(bp)) {
2746 		/* Setup NIC internals and enable interrupts */
2747 		bnx2x_post_irq_nic_init(bp, load_code);
2748 
2749 		bnx2x_init_bp_objs(bp);
2750 		bnx2x_iov_nic_init(bp);
2751 
2752 		/* Set AFEX default VLAN tag to an invalid value */
2753 		bp->afex_def_vlan_tag = -1;
2754 		bnx2x_nic_load_afex_dcc(bp, load_code);
2755 		bp->state = BNX2X_STATE_OPENING_WAIT4_PORT;
2756 		rc = bnx2x_func_start(bp);
2757 		if (rc) {
2758 			BNX2X_ERR("Function start failed!\n");
2759 			bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2760 
2761 			LOAD_ERROR_EXIT(bp, load_error3);
2762 		}
2763 
2764 		/* Send LOAD_DONE command to MCP */
2765 		if (!BP_NOMCP(bp)) {
2766 			load_code = bnx2x_fw_command(bp,
2767 						     DRV_MSG_CODE_LOAD_DONE, 0);
2768 			if (!load_code) {
2769 				BNX2X_ERR("MCP response failure, aborting\n");
2770 				rc = -EBUSY;
2771 				LOAD_ERROR_EXIT(bp, load_error3);
2772 			}
2773 		}
2774 
2775 		/* initialize FW coalescing state machines in RAM */
2776 		bnx2x_update_coalesce(bp);
2777 	}
2778 
2779 	/* setup the leading queue */
2780 	rc = bnx2x_setup_leading(bp);
2781 	if (rc) {
2782 		BNX2X_ERR("Setup leading failed!\n");
2783 		LOAD_ERROR_EXIT(bp, load_error3);
2784 	}
2785 
2786 	/* set up the rest of the queues */
2787 	for_each_nondefault_eth_queue(bp, i) {
2788 		if (IS_PF(bp))
2789 			rc = bnx2x_setup_queue(bp, &bp->fp[i], false);
2790 		else /* VF */
2791 			rc = bnx2x_vfpf_setup_q(bp, &bp->fp[i], false);
2792 		if (rc) {
2793 			BNX2X_ERR("Queue %d setup failed\n", i);
2794 			LOAD_ERROR_EXIT(bp, load_error3);
2795 		}
2796 	}
2797 
2798 	/* setup rss */
2799 	rc = bnx2x_init_rss(bp);
2800 	if (rc) {
2801 		BNX2X_ERR("PF RSS init failed\n");
2802 		LOAD_ERROR_EXIT(bp, load_error3);
2803 	}
2804 
2805 	/* Now when Clients are configured we are ready to work */
2806 	bp->state = BNX2X_STATE_OPEN;
2807 
2808 	/* Configure a ucast MAC */
2809 	if (IS_PF(bp))
2810 		rc = bnx2x_set_eth_mac(bp, true);
2811 	else /* vf */
2812 		rc = bnx2x_vfpf_config_mac(bp, bp->dev->dev_addr, bp->fp->index,
2813 					   true);
2814 	if (rc) {
2815 		BNX2X_ERR("Setting Ethernet MAC failed\n");
2816 		LOAD_ERROR_EXIT(bp, load_error3);
2817 	}
2818 
2819 	if (IS_PF(bp) && bp->pending_max) {
2820 		bnx2x_update_max_mf_config(bp, bp->pending_max);
2821 		bp->pending_max = 0;
2822 	}
2823 
2824 	if (bp->port.pmf) {
2825 		rc = bnx2x_initial_phy_init(bp, load_mode);
2826 		if (rc)
2827 			LOAD_ERROR_EXIT(bp, load_error3);
2828 	}
2829 	bp->link_params.feature_config_flags &= ~FEATURE_CONFIG_BOOT_FROM_SAN;
2830 
2831 	/* Start fast path */
2832 
2833 	/* Re-configure vlan filters */
2834 	rc = bnx2x_vlan_reconfigure_vid(bp);
2835 	if (rc)
2836 		LOAD_ERROR_EXIT(bp, load_error3);
2837 
2838 	/* Initialize Rx filter. */
2839 	bnx2x_set_rx_mode_inner(bp);
2840 
2841 	if (bp->flags & PTP_SUPPORTED) {
2842 		bnx2x_init_ptp(bp);
2843 		bnx2x_configure_ptp_filters(bp);
2844 	}
2845 	/* Start Tx */
2846 	switch (load_mode) {
2847 	case LOAD_NORMAL:
2848 		/* Tx queue should be only re-enabled */
2849 		netif_tx_wake_all_queues(bp->dev);
2850 		break;
2851 
2852 	case LOAD_OPEN:
2853 		netif_tx_start_all_queues(bp->dev);
2854 		smp_mb__after_atomic();
2855 		break;
2856 
2857 	case LOAD_DIAG:
2858 	case LOAD_LOOPBACK_EXT:
2859 		bp->state = BNX2X_STATE_DIAG;
2860 		break;
2861 
2862 	default:
2863 		break;
2864 	}
2865 
2866 	if (bp->port.pmf)
2867 		bnx2x_update_drv_flags(bp, 1 << DRV_FLAGS_PORT_MASK, 0);
2868 	else
2869 		bnx2x__link_status_update(bp);
2870 
2871 	/* start the timer */
2872 	mod_timer(&bp->timer, jiffies + bp->current_interval);
2873 
2874 	if (CNIC_ENABLED(bp))
2875 		bnx2x_load_cnic(bp);
2876 
2877 	if (IS_PF(bp))
2878 		bnx2x_schedule_sp_rtnl(bp, BNX2X_SP_RTNL_GET_DRV_VERSION, 0);
2879 
2880 	if (IS_PF(bp) && SHMEM2_HAS(bp, drv_capabilities_flag)) {
2881 		/* mark driver is loaded in shmem2 */
2882 		u32 val;
2883 		val = SHMEM2_RD(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)]);
2884 		val &= ~DRV_FLAGS_MTU_MASK;
2885 		val |= (bp->dev->mtu << DRV_FLAGS_MTU_SHIFT);
2886 		SHMEM2_WR(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)],
2887 			  val | DRV_FLAGS_CAPABILITIES_LOADED_SUPPORTED |
2888 			  DRV_FLAGS_CAPABILITIES_LOADED_L2);
2889 	}
2890 
2891 	/* Wait for all pending SP commands to complete */
2892 	if (IS_PF(bp) && !bnx2x_wait_sp_comp(bp, ~0x0UL)) {
2893 		BNX2X_ERR("Timeout waiting for SP elements to complete\n");
2894 		bnx2x_nic_unload(bp, UNLOAD_CLOSE, false);
2895 		return -EBUSY;
2896 	}
2897 
2898 	/* Update driver data for On-Chip MFW dump. */
2899 	if (IS_PF(bp))
2900 		bnx2x_update_mfw_dump(bp);
2901 
2902 	/* If PMF - send ADMIN DCBX msg to MFW to initiate DCBX FSM */
2903 	if (bp->port.pmf && (bp->state != BNX2X_STATE_DIAG))
2904 		bnx2x_dcbx_init(bp, false);
2905 
2906 	if (!IS_MF_SD_STORAGE_PERSONALITY_ONLY(bp))
2907 		bnx2x_set_os_driver_state(bp, OS_DRIVER_STATE_ACTIVE);
2908 
2909 	DP(NETIF_MSG_IFUP, "Ending successfully NIC load\n");
2910 
2911 	return 0;
2912 
2913 #ifndef BNX2X_STOP_ON_ERROR
2914 load_error3:
2915 	if (IS_PF(bp)) {
2916 		bnx2x_int_disable_sync(bp, 1);
2917 
2918 		/* Clean queueable objects */
2919 		bnx2x_squeeze_objects(bp);
2920 	}
2921 
2922 	/* Free SKBs, SGEs, TPA pool and driver internals */
2923 	bnx2x_free_skbs(bp);
2924 	for_each_rx_queue(bp, i)
2925 		bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
2926 
2927 	/* Release IRQs */
2928 	bnx2x_free_irq(bp);
2929 load_error2:
2930 	if (IS_PF(bp) && !BP_NOMCP(bp)) {
2931 		bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_REQ_WOL_MCP, 0);
2932 		bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_DONE, 0);
2933 	}
2934 
2935 	bp->port.pmf = 0;
2936 load_error1:
2937 	bnx2x_napi_disable(bp);
2938 	bnx2x_del_all_napi(bp);
2939 
2940 	/* clear pf_load status, as it was already set */
2941 	if (IS_PF(bp))
2942 		bnx2x_clear_pf_load(bp);
2943 load_error0:
2944 	bnx2x_free_fw_stats_mem(bp);
2945 	bnx2x_free_fp_mem(bp);
2946 	bnx2x_free_mem(bp);
2947 
2948 	return rc;
2949 #endif /* ! BNX2X_STOP_ON_ERROR */
2950 }
2951 
2952 int bnx2x_drain_tx_queues(struct bnx2x *bp)
2953 {
2954 	u8 rc = 0, cos, i;
2955 
2956 	/* Wait until tx fastpath tasks complete */
2957 	for_each_tx_queue(bp, i) {
2958 		struct bnx2x_fastpath *fp = &bp->fp[i];
2959 
2960 		for_each_cos_in_tx_queue(fp, cos)
2961 			rc = bnx2x_clean_tx_queue(bp, fp->txdata_ptr[cos]);
2962 		if (rc)
2963 			return rc;
2964 	}
2965 	return 0;
2966 }
2967 
2968 /* must be called with rtnl_lock */
2969 int bnx2x_nic_unload(struct bnx2x *bp, int unload_mode, bool keep_link)
2970 {
2971 	int i;
2972 	bool global = false;
2973 
2974 	DP(NETIF_MSG_IFUP, "Starting NIC unload\n");
2975 
2976 	if (!IS_MF_SD_STORAGE_PERSONALITY_ONLY(bp))
2977 		bnx2x_set_os_driver_state(bp, OS_DRIVER_STATE_DISABLED);
2978 
2979 	/* mark driver is unloaded in shmem2 */
2980 	if (IS_PF(bp) && SHMEM2_HAS(bp, drv_capabilities_flag)) {
2981 		u32 val;
2982 		val = SHMEM2_RD(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)]);
2983 		SHMEM2_WR(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)],
2984 			  val & ~DRV_FLAGS_CAPABILITIES_LOADED_L2);
2985 	}
2986 
2987 	if (IS_PF(bp) && bp->recovery_state != BNX2X_RECOVERY_DONE &&
2988 	    (bp->state == BNX2X_STATE_CLOSED ||
2989 	     bp->state == BNX2X_STATE_ERROR)) {
2990 		/* We can get here if the driver has been unloaded
2991 		 * during parity error recovery and is either waiting for a
2992 		 * leader to complete or for other functions to unload and
2993 		 * then ifdown has been issued. In this case we want to
2994 		 * unload and let other functions to complete a recovery
2995 		 * process.
2996 		 */
2997 		bp->recovery_state = BNX2X_RECOVERY_DONE;
2998 		bp->is_leader = 0;
2999 		bnx2x_release_leader_lock(bp);
3000 		smp_mb();
3001 
3002 		DP(NETIF_MSG_IFDOWN, "Releasing a leadership...\n");
3003 		BNX2X_ERR("Can't unload in closed or error state\n");
3004 		return -EINVAL;
3005 	}
3006 
3007 	/* Nothing to do during unload if previous bnx2x_nic_load()
3008 	 * have not completed successfully - all resources are released.
3009 	 *
3010 	 * we can get here only after unsuccessful ndo_* callback, during which
3011 	 * dev->IFF_UP flag is still on.
3012 	 */
3013 	if (bp->state == BNX2X_STATE_CLOSED || bp->state == BNX2X_STATE_ERROR)
3014 		return 0;
3015 
3016 	/* It's important to set the bp->state to the value different from
3017 	 * BNX2X_STATE_OPEN and only then stop the Tx. Otherwise bnx2x_tx_int()
3018 	 * may restart the Tx from the NAPI context (see bnx2x_tx_int()).
3019 	 */
3020 	bp->state = BNX2X_STATE_CLOSING_WAIT4_HALT;
3021 	smp_mb();
3022 
3023 	/* indicate to VFs that the PF is going down */
3024 	bnx2x_iov_channel_down(bp);
3025 
3026 	if (CNIC_LOADED(bp))
3027 		bnx2x_cnic_notify(bp, CNIC_CTL_STOP_CMD);
3028 
3029 	/* Stop Tx */
3030 	bnx2x_tx_disable(bp);
3031 	netdev_reset_tc(bp->dev);
3032 
3033 	bp->rx_mode = BNX2X_RX_MODE_NONE;
3034 
3035 	del_timer_sync(&bp->timer);
3036 
3037 	if (IS_PF(bp)) {
3038 		/* Set ALWAYS_ALIVE bit in shmem */
3039 		bp->fw_drv_pulse_wr_seq |= DRV_PULSE_ALWAYS_ALIVE;
3040 		bnx2x_drv_pulse(bp);
3041 		bnx2x_stats_handle(bp, STATS_EVENT_STOP);
3042 		bnx2x_save_statistics(bp);
3043 	}
3044 
3045 	/* wait till consumers catch up with producers in all queues.
3046 	 * If we're recovering, FW can't write to host so no reason
3047 	 * to wait for the queues to complete all Tx.
3048 	 */
3049 	if (unload_mode != UNLOAD_RECOVERY)
3050 		bnx2x_drain_tx_queues(bp);
3051 
3052 	/* if VF indicate to PF this function is going down (PF will delete sp
3053 	 * elements and clear initializations
3054 	 */
3055 	if (IS_VF(bp))
3056 		bnx2x_vfpf_close_vf(bp);
3057 	else if (unload_mode != UNLOAD_RECOVERY)
3058 		/* if this is a normal/close unload need to clean up chip*/
3059 		bnx2x_chip_cleanup(bp, unload_mode, keep_link);
3060 	else {
3061 		/* Send the UNLOAD_REQUEST to the MCP */
3062 		bnx2x_send_unload_req(bp, unload_mode);
3063 
3064 		/* Prevent transactions to host from the functions on the
3065 		 * engine that doesn't reset global blocks in case of global
3066 		 * attention once global blocks are reset and gates are opened
3067 		 * (the engine which leader will perform the recovery
3068 		 * last).
3069 		 */
3070 		if (!CHIP_IS_E1x(bp))
3071 			bnx2x_pf_disable(bp);
3072 
3073 		/* Disable HW interrupts, NAPI */
3074 		bnx2x_netif_stop(bp, 1);
3075 		/* Delete all NAPI objects */
3076 		bnx2x_del_all_napi(bp);
3077 		if (CNIC_LOADED(bp))
3078 			bnx2x_del_all_napi_cnic(bp);
3079 		/* Release IRQs */
3080 		bnx2x_free_irq(bp);
3081 
3082 		/* Report UNLOAD_DONE to MCP */
3083 		bnx2x_send_unload_done(bp, false);
3084 	}
3085 
3086 	/*
3087 	 * At this stage no more interrupts will arrive so we may safely clean
3088 	 * the queueable objects here in case they failed to get cleaned so far.
3089 	 */
3090 	if (IS_PF(bp))
3091 		bnx2x_squeeze_objects(bp);
3092 
3093 	/* There should be no more pending SP commands at this stage */
3094 	bp->sp_state = 0;
3095 
3096 	bp->port.pmf = 0;
3097 
3098 	/* clear pending work in rtnl task */
3099 	bp->sp_rtnl_state = 0;
3100 	smp_mb();
3101 
3102 	/* Free SKBs, SGEs, TPA pool and driver internals */
3103 	bnx2x_free_skbs(bp);
3104 	if (CNIC_LOADED(bp))
3105 		bnx2x_free_skbs_cnic(bp);
3106 	for_each_rx_queue(bp, i)
3107 		bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
3108 
3109 	bnx2x_free_fp_mem(bp);
3110 	if (CNIC_LOADED(bp))
3111 		bnx2x_free_fp_mem_cnic(bp);
3112 
3113 	if (IS_PF(bp)) {
3114 		if (CNIC_LOADED(bp))
3115 			bnx2x_free_mem_cnic(bp);
3116 	}
3117 	bnx2x_free_mem(bp);
3118 
3119 	bp->state = BNX2X_STATE_CLOSED;
3120 	bp->cnic_loaded = false;
3121 
3122 	/* Clear driver version indication in shmem */
3123 	if (IS_PF(bp))
3124 		bnx2x_update_mng_version(bp);
3125 
3126 	/* Check if there are pending parity attentions. If there are - set
3127 	 * RECOVERY_IN_PROGRESS.
3128 	 */
3129 	if (IS_PF(bp) && bnx2x_chk_parity_attn(bp, &global, false)) {
3130 		bnx2x_set_reset_in_progress(bp);
3131 
3132 		/* Set RESET_IS_GLOBAL if needed */
3133 		if (global)
3134 			bnx2x_set_reset_global(bp);
3135 	}
3136 
3137 	/* The last driver must disable a "close the gate" if there is no
3138 	 * parity attention or "process kill" pending.
3139 	 */
3140 	if (IS_PF(bp) &&
3141 	    !bnx2x_clear_pf_load(bp) &&
3142 	    bnx2x_reset_is_done(bp, BP_PATH(bp)))
3143 		bnx2x_disable_close_the_gate(bp);
3144 
3145 	DP(NETIF_MSG_IFUP, "Ending NIC unload\n");
3146 
3147 	return 0;
3148 }
3149 
3150 int bnx2x_set_power_state(struct bnx2x *bp, pci_power_t state)
3151 {
3152 	u16 pmcsr;
3153 
3154 	/* If there is no power capability, silently succeed */
3155 	if (!bp->pdev->pm_cap) {
3156 		BNX2X_DEV_INFO("No power capability. Breaking.\n");
3157 		return 0;
3158 	}
3159 
3160 	pci_read_config_word(bp->pdev, bp->pdev->pm_cap + PCI_PM_CTRL, &pmcsr);
3161 
3162 	switch (state) {
3163 	case PCI_D0:
3164 		pci_write_config_word(bp->pdev, bp->pdev->pm_cap + PCI_PM_CTRL,
3165 				      ((pmcsr & ~PCI_PM_CTRL_STATE_MASK) |
3166 				       PCI_PM_CTRL_PME_STATUS));
3167 
3168 		if (pmcsr & PCI_PM_CTRL_STATE_MASK)
3169 			/* delay required during transition out of D3hot */
3170 			msleep(20);
3171 		break;
3172 
3173 	case PCI_D3hot:
3174 		/* If there are other clients above don't
3175 		   shut down the power */
3176 		if (atomic_read(&bp->pdev->enable_cnt) != 1)
3177 			return 0;
3178 		/* Don't shut down the power for emulation and FPGA */
3179 		if (CHIP_REV_IS_SLOW(bp))
3180 			return 0;
3181 
3182 		pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
3183 		pmcsr |= 3;
3184 
3185 		if (bp->wol)
3186 			pmcsr |= PCI_PM_CTRL_PME_ENABLE;
3187 
3188 		pci_write_config_word(bp->pdev, bp->pdev->pm_cap + PCI_PM_CTRL,
3189 				      pmcsr);
3190 
3191 		/* No more memory access after this point until
3192 		* device is brought back to D0.
3193 		*/
3194 		break;
3195 
3196 	default:
3197 		dev_err(&bp->pdev->dev, "Can't support state = %d\n", state);
3198 		return -EINVAL;
3199 	}
3200 	return 0;
3201 }
3202 
3203 /*
3204  * net_device service functions
3205  */
3206 static int bnx2x_poll(struct napi_struct *napi, int budget)
3207 {
3208 	struct bnx2x_fastpath *fp = container_of(napi, struct bnx2x_fastpath,
3209 						 napi);
3210 	struct bnx2x *bp = fp->bp;
3211 	int rx_work_done;
3212 	u8 cos;
3213 
3214 #ifdef BNX2X_STOP_ON_ERROR
3215 	if (unlikely(bp->panic)) {
3216 		napi_complete(napi);
3217 		return 0;
3218 	}
3219 #endif
3220 	for_each_cos_in_tx_queue(fp, cos)
3221 		if (bnx2x_tx_queue_has_work(fp->txdata_ptr[cos]))
3222 			bnx2x_tx_int(bp, fp->txdata_ptr[cos]);
3223 
3224 	rx_work_done = (bnx2x_has_rx_work(fp)) ? bnx2x_rx_int(fp, budget) : 0;
3225 
3226 	if (rx_work_done < budget) {
3227 		/* No need to update SB for FCoE L2 ring as long as
3228 		 * it's connected to the default SB and the SB
3229 		 * has been updated when NAPI was scheduled.
3230 		 */
3231 		if (IS_FCOE_FP(fp)) {
3232 			napi_complete(napi);
3233 		} else {
3234 			bnx2x_update_fpsb_idx(fp);
3235 			/* bnx2x_has_rx_work() reads the status block,
3236 			 * thus we need to ensure that status block indices
3237 			 * have been actually read (bnx2x_update_fpsb_idx)
3238 			 * prior to this check (bnx2x_has_rx_work) so that
3239 			 * we won't write the "newer" value of the status block
3240 			 * to IGU (if there was a DMA right after
3241 			 * bnx2x_has_rx_work and if there is no rmb, the memory
3242 			 * reading (bnx2x_update_fpsb_idx) may be postponed
3243 			 * to right before bnx2x_ack_sb). In this case there
3244 			 * will never be another interrupt until there is
3245 			 * another update of the status block, while there
3246 			 * is still unhandled work.
3247 			 */
3248 			rmb();
3249 
3250 			if (!(bnx2x_has_rx_work(fp) || bnx2x_has_tx_work(fp))) {
3251 				napi_complete(napi);
3252 				/* Re-enable interrupts */
3253 				DP(NETIF_MSG_RX_STATUS,
3254 				   "Update index to %d\n", fp->fp_hc_idx);
3255 				bnx2x_ack_sb(bp, fp->igu_sb_id, USTORM_ID,
3256 					     le16_to_cpu(fp->fp_hc_idx),
3257 					     IGU_INT_ENABLE, 1);
3258 			} else {
3259 				rx_work_done = budget;
3260 			}
3261 		}
3262 	}
3263 
3264 	return rx_work_done;
3265 }
3266 
3267 /* we split the first BD into headers and data BDs
3268  * to ease the pain of our fellow microcode engineers
3269  * we use one mapping for both BDs
3270  */
3271 static u16 bnx2x_tx_split(struct bnx2x *bp,
3272 			  struct bnx2x_fp_txdata *txdata,
3273 			  struct sw_tx_bd *tx_buf,
3274 			  struct eth_tx_start_bd **tx_bd, u16 hlen,
3275 			  u16 bd_prod)
3276 {
3277 	struct eth_tx_start_bd *h_tx_bd = *tx_bd;
3278 	struct eth_tx_bd *d_tx_bd;
3279 	dma_addr_t mapping;
3280 	int old_len = le16_to_cpu(h_tx_bd->nbytes);
3281 
3282 	/* first fix first BD */
3283 	h_tx_bd->nbytes = cpu_to_le16(hlen);
3284 
3285 	DP(NETIF_MSG_TX_QUEUED,	"TSO split header size is %d (%x:%x)\n",
3286 	   h_tx_bd->nbytes, h_tx_bd->addr_hi, h_tx_bd->addr_lo);
3287 
3288 	/* now get a new data BD
3289 	 * (after the pbd) and fill it */
3290 	bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3291 	d_tx_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
3292 
3293 	mapping = HILO_U64(le32_to_cpu(h_tx_bd->addr_hi),
3294 			   le32_to_cpu(h_tx_bd->addr_lo)) + hlen;
3295 
3296 	d_tx_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
3297 	d_tx_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
3298 	d_tx_bd->nbytes = cpu_to_le16(old_len - hlen);
3299 
3300 	/* this marks the BD as one that has no individual mapping */
3301 	tx_buf->flags |= BNX2X_TSO_SPLIT_BD;
3302 
3303 	DP(NETIF_MSG_TX_QUEUED,
3304 	   "TSO split data size is %d (%x:%x)\n",
3305 	   d_tx_bd->nbytes, d_tx_bd->addr_hi, d_tx_bd->addr_lo);
3306 
3307 	/* update tx_bd */
3308 	*tx_bd = (struct eth_tx_start_bd *)d_tx_bd;
3309 
3310 	return bd_prod;
3311 }
3312 
3313 #define bswab32(b32) ((__force __le32) swab32((__force __u32) (b32)))
3314 #define bswab16(b16) ((__force __le16) swab16((__force __u16) (b16)))
3315 static __le16 bnx2x_csum_fix(unsigned char *t_header, u16 csum, s8 fix)
3316 {
3317 	__sum16 tsum = (__force __sum16) csum;
3318 
3319 	if (fix > 0)
3320 		tsum = ~csum_fold(csum_sub((__force __wsum) csum,
3321 				  csum_partial(t_header - fix, fix, 0)));
3322 
3323 	else if (fix < 0)
3324 		tsum = ~csum_fold(csum_add((__force __wsum) csum,
3325 				  csum_partial(t_header, -fix, 0)));
3326 
3327 	return bswab16(tsum);
3328 }
3329 
3330 static u32 bnx2x_xmit_type(struct bnx2x *bp, struct sk_buff *skb)
3331 {
3332 	u32 rc;
3333 	__u8 prot = 0;
3334 	__be16 protocol;
3335 
3336 	if (skb->ip_summed != CHECKSUM_PARTIAL)
3337 		return XMIT_PLAIN;
3338 
3339 	protocol = vlan_get_protocol(skb);
3340 	if (protocol == htons(ETH_P_IPV6)) {
3341 		rc = XMIT_CSUM_V6;
3342 		prot = ipv6_hdr(skb)->nexthdr;
3343 	} else {
3344 		rc = XMIT_CSUM_V4;
3345 		prot = ip_hdr(skb)->protocol;
3346 	}
3347 
3348 	if (!CHIP_IS_E1x(bp) && skb->encapsulation) {
3349 		if (inner_ip_hdr(skb)->version == 6) {
3350 			rc |= XMIT_CSUM_ENC_V6;
3351 			if (inner_ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3352 				rc |= XMIT_CSUM_TCP;
3353 		} else {
3354 			rc |= XMIT_CSUM_ENC_V4;
3355 			if (inner_ip_hdr(skb)->protocol == IPPROTO_TCP)
3356 				rc |= XMIT_CSUM_TCP;
3357 		}
3358 	}
3359 	if (prot == IPPROTO_TCP)
3360 		rc |= XMIT_CSUM_TCP;
3361 
3362 	if (skb_is_gso(skb)) {
3363 		if (skb_is_gso_v6(skb)) {
3364 			rc |= (XMIT_GSO_V6 | XMIT_CSUM_TCP);
3365 			if (rc & XMIT_CSUM_ENC)
3366 				rc |= XMIT_GSO_ENC_V6;
3367 		} else {
3368 			rc |= (XMIT_GSO_V4 | XMIT_CSUM_TCP);
3369 			if (rc & XMIT_CSUM_ENC)
3370 				rc |= XMIT_GSO_ENC_V4;
3371 		}
3372 	}
3373 
3374 	return rc;
3375 }
3376 
3377 /* VXLAN: 4 = 1 (for linear data BD) + 3 (2 for PBD and last BD) */
3378 #define BNX2X_NUM_VXLAN_TSO_WIN_SUB_BDS         4
3379 
3380 /* Regular: 3 = 1 (for linear data BD) + 2 (for PBD and last BD) */
3381 #define BNX2X_NUM_TSO_WIN_SUB_BDS               3
3382 
3383 #if (MAX_SKB_FRAGS >= MAX_FETCH_BD - BDS_PER_TX_PKT)
3384 /* check if packet requires linearization (packet is too fragmented)
3385    no need to check fragmentation if page size > 8K (there will be no
3386    violation to FW restrictions) */
3387 static int bnx2x_pkt_req_lin(struct bnx2x *bp, struct sk_buff *skb,
3388 			     u32 xmit_type)
3389 {
3390 	int first_bd_sz = 0, num_tso_win_sub = BNX2X_NUM_TSO_WIN_SUB_BDS;
3391 	int to_copy = 0, hlen = 0;
3392 
3393 	if (xmit_type & XMIT_GSO_ENC)
3394 		num_tso_win_sub = BNX2X_NUM_VXLAN_TSO_WIN_SUB_BDS;
3395 
3396 	if (skb_shinfo(skb)->nr_frags >= (MAX_FETCH_BD - num_tso_win_sub)) {
3397 		if (xmit_type & XMIT_GSO) {
3398 			unsigned short lso_mss = skb_shinfo(skb)->gso_size;
3399 			int wnd_size = MAX_FETCH_BD - num_tso_win_sub;
3400 			/* Number of windows to check */
3401 			int num_wnds = skb_shinfo(skb)->nr_frags - wnd_size;
3402 			int wnd_idx = 0;
3403 			int frag_idx = 0;
3404 			u32 wnd_sum = 0;
3405 
3406 			/* Headers length */
3407 			if (xmit_type & XMIT_GSO_ENC)
3408 				hlen = (int)(skb_inner_transport_header(skb) -
3409 					     skb->data) +
3410 					     inner_tcp_hdrlen(skb);
3411 			else
3412 				hlen = (int)(skb_transport_header(skb) -
3413 					     skb->data) + tcp_hdrlen(skb);
3414 
3415 			/* Amount of data (w/o headers) on linear part of SKB*/
3416 			first_bd_sz = skb_headlen(skb) - hlen;
3417 
3418 			wnd_sum  = first_bd_sz;
3419 
3420 			/* Calculate the first sum - it's special */
3421 			for (frag_idx = 0; frag_idx < wnd_size - 1; frag_idx++)
3422 				wnd_sum +=
3423 					skb_frag_size(&skb_shinfo(skb)->frags[frag_idx]);
3424 
3425 			/* If there was data on linear skb data - check it */
3426 			if (first_bd_sz > 0) {
3427 				if (unlikely(wnd_sum < lso_mss)) {
3428 					to_copy = 1;
3429 					goto exit_lbl;
3430 				}
3431 
3432 				wnd_sum -= first_bd_sz;
3433 			}
3434 
3435 			/* Others are easier: run through the frag list and
3436 			   check all windows */
3437 			for (wnd_idx = 0; wnd_idx <= num_wnds; wnd_idx++) {
3438 				wnd_sum +=
3439 			  skb_frag_size(&skb_shinfo(skb)->frags[wnd_idx + wnd_size - 1]);
3440 
3441 				if (unlikely(wnd_sum < lso_mss)) {
3442 					to_copy = 1;
3443 					break;
3444 				}
3445 				wnd_sum -=
3446 					skb_frag_size(&skb_shinfo(skb)->frags[wnd_idx]);
3447 			}
3448 		} else {
3449 			/* in non-LSO too fragmented packet should always
3450 			   be linearized */
3451 			to_copy = 1;
3452 		}
3453 	}
3454 
3455 exit_lbl:
3456 	if (unlikely(to_copy))
3457 		DP(NETIF_MSG_TX_QUEUED,
3458 		   "Linearization IS REQUIRED for %s packet. num_frags %d  hlen %d  first_bd_sz %d\n",
3459 		   (xmit_type & XMIT_GSO) ? "LSO" : "non-LSO",
3460 		   skb_shinfo(skb)->nr_frags, hlen, first_bd_sz);
3461 
3462 	return to_copy;
3463 }
3464 #endif
3465 
3466 /**
3467  * bnx2x_set_pbd_gso - update PBD in GSO case.
3468  *
3469  * @skb:	packet skb
3470  * @pbd:	parse BD
3471  * @xmit_type:	xmit flags
3472  */
3473 static void bnx2x_set_pbd_gso(struct sk_buff *skb,
3474 			      struct eth_tx_parse_bd_e1x *pbd,
3475 			      u32 xmit_type)
3476 {
3477 	pbd->lso_mss = cpu_to_le16(skb_shinfo(skb)->gso_size);
3478 	pbd->tcp_send_seq = bswab32(tcp_hdr(skb)->seq);
3479 	pbd->tcp_flags = pbd_tcp_flags(tcp_hdr(skb));
3480 
3481 	if (xmit_type & XMIT_GSO_V4) {
3482 		pbd->ip_id = bswab16(ip_hdr(skb)->id);
3483 		pbd->tcp_pseudo_csum =
3484 			bswab16(~csum_tcpudp_magic(ip_hdr(skb)->saddr,
3485 						   ip_hdr(skb)->daddr,
3486 						   0, IPPROTO_TCP, 0));
3487 	} else {
3488 		pbd->tcp_pseudo_csum =
3489 			bswab16(~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3490 						 &ipv6_hdr(skb)->daddr,
3491 						 0, IPPROTO_TCP, 0));
3492 	}
3493 
3494 	pbd->global_data |=
3495 		cpu_to_le16(ETH_TX_PARSE_BD_E1X_PSEUDO_CS_WITHOUT_LEN);
3496 }
3497 
3498 /**
3499  * bnx2x_set_pbd_csum_enc - update PBD with checksum and return header length
3500  *
3501  * @bp:			driver handle
3502  * @skb:		packet skb
3503  * @parsing_data:	data to be updated
3504  * @xmit_type:		xmit flags
3505  *
3506  * 57712/578xx related, when skb has encapsulation
3507  */
3508 static u8 bnx2x_set_pbd_csum_enc(struct bnx2x *bp, struct sk_buff *skb,
3509 				 u32 *parsing_data, u32 xmit_type)
3510 {
3511 	*parsing_data |=
3512 		((((u8 *)skb_inner_transport_header(skb) - skb->data) >> 1) <<
3513 		ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W_SHIFT) &
3514 		ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W;
3515 
3516 	if (xmit_type & XMIT_CSUM_TCP) {
3517 		*parsing_data |= ((inner_tcp_hdrlen(skb) / 4) <<
3518 			ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW_SHIFT) &
3519 			ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW;
3520 
3521 		return skb_inner_transport_header(skb) +
3522 			inner_tcp_hdrlen(skb) - skb->data;
3523 	}
3524 
3525 	/* We support checksum offload for TCP and UDP only.
3526 	 * No need to pass the UDP header length - it's a constant.
3527 	 */
3528 	return skb_inner_transport_header(skb) +
3529 		sizeof(struct udphdr) - skb->data;
3530 }
3531 
3532 /**
3533  * bnx2x_set_pbd_csum_e2 - update PBD with checksum and return header length
3534  *
3535  * @bp:			driver handle
3536  * @skb:		packet skb
3537  * @parsing_data:	data to be updated
3538  * @xmit_type:		xmit flags
3539  *
3540  * 57712/578xx related
3541  */
3542 static u8 bnx2x_set_pbd_csum_e2(struct bnx2x *bp, struct sk_buff *skb,
3543 				u32 *parsing_data, u32 xmit_type)
3544 {
3545 	*parsing_data |=
3546 		((((u8 *)skb_transport_header(skb) - skb->data) >> 1) <<
3547 		ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W_SHIFT) &
3548 		ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W;
3549 
3550 	if (xmit_type & XMIT_CSUM_TCP) {
3551 		*parsing_data |= ((tcp_hdrlen(skb) / 4) <<
3552 			ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW_SHIFT) &
3553 			ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW;
3554 
3555 		return skb_transport_header(skb) + tcp_hdrlen(skb) - skb->data;
3556 	}
3557 	/* We support checksum offload for TCP and UDP only.
3558 	 * No need to pass the UDP header length - it's a constant.
3559 	 */
3560 	return skb_transport_header(skb) + sizeof(struct udphdr) - skb->data;
3561 }
3562 
3563 /* set FW indication according to inner or outer protocols if tunneled */
3564 static void bnx2x_set_sbd_csum(struct bnx2x *bp, struct sk_buff *skb,
3565 			       struct eth_tx_start_bd *tx_start_bd,
3566 			       u32 xmit_type)
3567 {
3568 	tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_L4_CSUM;
3569 
3570 	if (xmit_type & (XMIT_CSUM_ENC_V6 | XMIT_CSUM_V6))
3571 		tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_IPV6;
3572 
3573 	if (!(xmit_type & XMIT_CSUM_TCP))
3574 		tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_IS_UDP;
3575 }
3576 
3577 /**
3578  * bnx2x_set_pbd_csum - update PBD with checksum and return header length
3579  *
3580  * @bp:		driver handle
3581  * @skb:	packet skb
3582  * @pbd:	parse BD to be updated
3583  * @xmit_type:	xmit flags
3584  */
3585 static u8 bnx2x_set_pbd_csum(struct bnx2x *bp, struct sk_buff *skb,
3586 			     struct eth_tx_parse_bd_e1x *pbd,
3587 			     u32 xmit_type)
3588 {
3589 	u8 hlen = (skb_network_header(skb) - skb->data) >> 1;
3590 
3591 	/* for now NS flag is not used in Linux */
3592 	pbd->global_data =
3593 		cpu_to_le16(hlen |
3594 			    ((skb->protocol == cpu_to_be16(ETH_P_8021Q)) <<
3595 			     ETH_TX_PARSE_BD_E1X_LLC_SNAP_EN_SHIFT));
3596 
3597 	pbd->ip_hlen_w = (skb_transport_header(skb) -
3598 			skb_network_header(skb)) >> 1;
3599 
3600 	hlen += pbd->ip_hlen_w;
3601 
3602 	/* We support checksum offload for TCP and UDP only */
3603 	if (xmit_type & XMIT_CSUM_TCP)
3604 		hlen += tcp_hdrlen(skb) / 2;
3605 	else
3606 		hlen += sizeof(struct udphdr) / 2;
3607 
3608 	pbd->total_hlen_w = cpu_to_le16(hlen);
3609 	hlen = hlen*2;
3610 
3611 	if (xmit_type & XMIT_CSUM_TCP) {
3612 		pbd->tcp_pseudo_csum = bswab16(tcp_hdr(skb)->check);
3613 
3614 	} else {
3615 		s8 fix = SKB_CS_OFF(skb); /* signed! */
3616 
3617 		DP(NETIF_MSG_TX_QUEUED,
3618 		   "hlen %d  fix %d  csum before fix %x\n",
3619 		   le16_to_cpu(pbd->total_hlen_w), fix, SKB_CS(skb));
3620 
3621 		/* HW bug: fixup the CSUM */
3622 		pbd->tcp_pseudo_csum =
3623 			bnx2x_csum_fix(skb_transport_header(skb),
3624 				       SKB_CS(skb), fix);
3625 
3626 		DP(NETIF_MSG_TX_QUEUED, "csum after fix %x\n",
3627 		   pbd->tcp_pseudo_csum);
3628 	}
3629 
3630 	return hlen;
3631 }
3632 
3633 static void bnx2x_update_pbds_gso_enc(struct sk_buff *skb,
3634 				      struct eth_tx_parse_bd_e2 *pbd_e2,
3635 				      struct eth_tx_parse_2nd_bd *pbd2,
3636 				      u16 *global_data,
3637 				      u32 xmit_type)
3638 {
3639 	u16 hlen_w = 0;
3640 	u8 outerip_off, outerip_len = 0;
3641 
3642 	/* from outer IP to transport */
3643 	hlen_w = (skb_inner_transport_header(skb) -
3644 		  skb_network_header(skb)) >> 1;
3645 
3646 	/* transport len */
3647 	hlen_w += inner_tcp_hdrlen(skb) >> 1;
3648 
3649 	pbd2->fw_ip_hdr_to_payload_w = hlen_w;
3650 
3651 	/* outer IP header info */
3652 	if (xmit_type & XMIT_CSUM_V4) {
3653 		struct iphdr *iph = ip_hdr(skb);
3654 		u32 csum = (__force u32)(~iph->check) -
3655 			   (__force u32)iph->tot_len -
3656 			   (__force u32)iph->frag_off;
3657 
3658 		outerip_len = iph->ihl << 1;
3659 
3660 		pbd2->fw_ip_csum_wo_len_flags_frag =
3661 			bswab16(csum_fold((__force __wsum)csum));
3662 	} else {
3663 		pbd2->fw_ip_hdr_to_payload_w =
3664 			hlen_w - ((sizeof(struct ipv6hdr)) >> 1);
3665 		pbd_e2->data.tunnel_data.flags |=
3666 			ETH_TUNNEL_DATA_IPV6_OUTER;
3667 	}
3668 
3669 	pbd2->tcp_send_seq = bswab32(inner_tcp_hdr(skb)->seq);
3670 
3671 	pbd2->tcp_flags = pbd_tcp_flags(inner_tcp_hdr(skb));
3672 
3673 	/* inner IP header info */
3674 	if (xmit_type & XMIT_CSUM_ENC_V4) {
3675 		pbd2->hw_ip_id = bswab16(inner_ip_hdr(skb)->id);
3676 
3677 		pbd_e2->data.tunnel_data.pseudo_csum =
3678 			bswab16(~csum_tcpudp_magic(
3679 					inner_ip_hdr(skb)->saddr,
3680 					inner_ip_hdr(skb)->daddr,
3681 					0, IPPROTO_TCP, 0));
3682 	} else {
3683 		pbd_e2->data.tunnel_data.pseudo_csum =
3684 			bswab16(~csum_ipv6_magic(
3685 					&inner_ipv6_hdr(skb)->saddr,
3686 					&inner_ipv6_hdr(skb)->daddr,
3687 					0, IPPROTO_TCP, 0));
3688 	}
3689 
3690 	outerip_off = (skb_network_header(skb) - skb->data) >> 1;
3691 
3692 	*global_data |=
3693 		outerip_off |
3694 		(outerip_len <<
3695 			ETH_TX_PARSE_2ND_BD_IP_HDR_LEN_OUTER_W_SHIFT) |
3696 		((skb->protocol == cpu_to_be16(ETH_P_8021Q)) <<
3697 			ETH_TX_PARSE_2ND_BD_LLC_SNAP_EN_SHIFT);
3698 
3699 	if (ip_hdr(skb)->protocol == IPPROTO_UDP) {
3700 		SET_FLAG(*global_data, ETH_TX_PARSE_2ND_BD_TUNNEL_UDP_EXIST, 1);
3701 		pbd2->tunnel_udp_hdr_start_w = skb_transport_offset(skb) >> 1;
3702 	}
3703 }
3704 
3705 static inline void bnx2x_set_ipv6_ext_e2(struct sk_buff *skb, u32 *parsing_data,
3706 					 u32 xmit_type)
3707 {
3708 	struct ipv6hdr *ipv6;
3709 
3710 	if (!(xmit_type & (XMIT_GSO_ENC_V6 | XMIT_GSO_V6)))
3711 		return;
3712 
3713 	if (xmit_type & XMIT_GSO_ENC_V6)
3714 		ipv6 = inner_ipv6_hdr(skb);
3715 	else /* XMIT_GSO_V6 */
3716 		ipv6 = ipv6_hdr(skb);
3717 
3718 	if (ipv6->nexthdr == NEXTHDR_IPV6)
3719 		*parsing_data |= ETH_TX_PARSE_BD_E2_IPV6_WITH_EXT_HDR;
3720 }
3721 
3722 /* called with netif_tx_lock
3723  * bnx2x_tx_int() runs without netif_tx_lock unless it needs to call
3724  * netif_wake_queue()
3725  */
3726 netdev_tx_t bnx2x_start_xmit(struct sk_buff *skb, struct net_device *dev)
3727 {
3728 	struct bnx2x *bp = netdev_priv(dev);
3729 
3730 	struct netdev_queue *txq;
3731 	struct bnx2x_fp_txdata *txdata;
3732 	struct sw_tx_bd *tx_buf;
3733 	struct eth_tx_start_bd *tx_start_bd, *first_bd;
3734 	struct eth_tx_bd *tx_data_bd, *total_pkt_bd = NULL;
3735 	struct eth_tx_parse_bd_e1x *pbd_e1x = NULL;
3736 	struct eth_tx_parse_bd_e2 *pbd_e2 = NULL;
3737 	struct eth_tx_parse_2nd_bd *pbd2 = NULL;
3738 	u32 pbd_e2_parsing_data = 0;
3739 	u16 pkt_prod, bd_prod;
3740 	int nbd, txq_index;
3741 	dma_addr_t mapping;
3742 	u32 xmit_type = bnx2x_xmit_type(bp, skb);
3743 	int i;
3744 	u8 hlen = 0;
3745 	__le16 pkt_size = 0;
3746 	struct ethhdr *eth;
3747 	u8 mac_type = UNICAST_ADDRESS;
3748 
3749 #ifdef BNX2X_STOP_ON_ERROR
3750 	if (unlikely(bp->panic))
3751 		return NETDEV_TX_BUSY;
3752 #endif
3753 
3754 	txq_index = skb_get_queue_mapping(skb);
3755 	txq = netdev_get_tx_queue(dev, txq_index);
3756 
3757 	BUG_ON(txq_index >= MAX_ETH_TXQ_IDX(bp) + (CNIC_LOADED(bp) ? 1 : 0));
3758 
3759 	txdata = &bp->bnx2x_txq[txq_index];
3760 
3761 	/* enable this debug print to view the transmission queue being used
3762 	DP(NETIF_MSG_TX_QUEUED, "indices: txq %d, fp %d, txdata %d\n",
3763 	   txq_index, fp_index, txdata_index); */
3764 
3765 	/* enable this debug print to view the transmission details
3766 	DP(NETIF_MSG_TX_QUEUED,
3767 	   "transmitting packet cid %d fp index %d txdata_index %d tx_data ptr %p fp pointer %p\n",
3768 	   txdata->cid, fp_index, txdata_index, txdata, fp); */
3769 
3770 	if (unlikely(bnx2x_tx_avail(bp, txdata) <
3771 			skb_shinfo(skb)->nr_frags +
3772 			BDS_PER_TX_PKT +
3773 			NEXT_CNT_PER_TX_PKT(MAX_BDS_PER_TX_PKT))) {
3774 		/* Handle special storage cases separately */
3775 		if (txdata->tx_ring_size == 0) {
3776 			struct bnx2x_eth_q_stats *q_stats =
3777 				bnx2x_fp_qstats(bp, txdata->parent_fp);
3778 			q_stats->driver_filtered_tx_pkt++;
3779 			dev_kfree_skb(skb);
3780 			return NETDEV_TX_OK;
3781 		}
3782 		bnx2x_fp_qstats(bp, txdata->parent_fp)->driver_xoff++;
3783 		netif_tx_stop_queue(txq);
3784 		BNX2X_ERR("BUG! Tx ring full when queue awake!\n");
3785 
3786 		return NETDEV_TX_BUSY;
3787 	}
3788 
3789 	DP(NETIF_MSG_TX_QUEUED,
3790 	   "queue[%d]: SKB: summed %x  protocol %x protocol(%x,%x) gso type %x  xmit_type %x len %d\n",
3791 	   txq_index, skb->ip_summed, skb->protocol, ipv6_hdr(skb)->nexthdr,
3792 	   ip_hdr(skb)->protocol, skb_shinfo(skb)->gso_type, xmit_type,
3793 	   skb->len);
3794 
3795 	eth = (struct ethhdr *)skb->data;
3796 
3797 	/* set flag according to packet type (UNICAST_ADDRESS is default)*/
3798 	if (unlikely(is_multicast_ether_addr(eth->h_dest))) {
3799 		if (is_broadcast_ether_addr(eth->h_dest))
3800 			mac_type = BROADCAST_ADDRESS;
3801 		else
3802 			mac_type = MULTICAST_ADDRESS;
3803 	}
3804 
3805 #if (MAX_SKB_FRAGS >= MAX_FETCH_BD - BDS_PER_TX_PKT)
3806 	/* First, check if we need to linearize the skb (due to FW
3807 	   restrictions). No need to check fragmentation if page size > 8K
3808 	   (there will be no violation to FW restrictions) */
3809 	if (bnx2x_pkt_req_lin(bp, skb, xmit_type)) {
3810 		/* Statistics of linearization */
3811 		bp->lin_cnt++;
3812 		if (skb_linearize(skb) != 0) {
3813 			DP(NETIF_MSG_TX_QUEUED,
3814 			   "SKB linearization failed - silently dropping this SKB\n");
3815 			dev_kfree_skb_any(skb);
3816 			return NETDEV_TX_OK;
3817 		}
3818 	}
3819 #endif
3820 	/* Map skb linear data for DMA */
3821 	mapping = dma_map_single(&bp->pdev->dev, skb->data,
3822 				 skb_headlen(skb), DMA_TO_DEVICE);
3823 	if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
3824 		DP(NETIF_MSG_TX_QUEUED,
3825 		   "SKB mapping failed - silently dropping this SKB\n");
3826 		dev_kfree_skb_any(skb);
3827 		return NETDEV_TX_OK;
3828 	}
3829 	/*
3830 	Please read carefully. First we use one BD which we mark as start,
3831 	then we have a parsing info BD (used for TSO or xsum),
3832 	and only then we have the rest of the TSO BDs.
3833 	(don't forget to mark the last one as last,
3834 	and to unmap only AFTER you write to the BD ...)
3835 	And above all, all pdb sizes are in words - NOT DWORDS!
3836 	*/
3837 
3838 	/* get current pkt produced now - advance it just before sending packet
3839 	 * since mapping of pages may fail and cause packet to be dropped
3840 	 */
3841 	pkt_prod = txdata->tx_pkt_prod;
3842 	bd_prod = TX_BD(txdata->tx_bd_prod);
3843 
3844 	/* get a tx_buf and first BD
3845 	 * tx_start_bd may be changed during SPLIT,
3846 	 * but first_bd will always stay first
3847 	 */
3848 	tx_buf = &txdata->tx_buf_ring[TX_BD(pkt_prod)];
3849 	tx_start_bd = &txdata->tx_desc_ring[bd_prod].start_bd;
3850 	first_bd = tx_start_bd;
3851 
3852 	tx_start_bd->bd_flags.as_bitfield = ETH_TX_BD_FLAGS_START_BD;
3853 
3854 	if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) {
3855 		if (!(bp->flags & TX_TIMESTAMPING_EN)) {
3856 			BNX2X_ERR("Tx timestamping was not enabled, this packet will not be timestamped\n");
3857 		} else if (bp->ptp_tx_skb) {
3858 			BNX2X_ERR("The device supports only a single outstanding packet to timestamp, this packet will not be timestamped\n");
3859 		} else {
3860 			skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
3861 			/* schedule check for Tx timestamp */
3862 			bp->ptp_tx_skb = skb_get(skb);
3863 			bp->ptp_tx_start = jiffies;
3864 			schedule_work(&bp->ptp_task);
3865 		}
3866 	}
3867 
3868 	/* header nbd: indirectly zero other flags! */
3869 	tx_start_bd->general_data = 1 << ETH_TX_START_BD_HDR_NBDS_SHIFT;
3870 
3871 	/* remember the first BD of the packet */
3872 	tx_buf->first_bd = txdata->tx_bd_prod;
3873 	tx_buf->skb = skb;
3874 	tx_buf->flags = 0;
3875 
3876 	DP(NETIF_MSG_TX_QUEUED,
3877 	   "sending pkt %u @%p  next_idx %u  bd %u @%p\n",
3878 	   pkt_prod, tx_buf, txdata->tx_pkt_prod, bd_prod, tx_start_bd);
3879 
3880 	if (skb_vlan_tag_present(skb)) {
3881 		tx_start_bd->vlan_or_ethertype =
3882 		    cpu_to_le16(skb_vlan_tag_get(skb));
3883 		tx_start_bd->bd_flags.as_bitfield |=
3884 		    (X_ETH_OUTBAND_VLAN << ETH_TX_BD_FLAGS_VLAN_MODE_SHIFT);
3885 	} else {
3886 		/* when transmitting in a vf, start bd must hold the ethertype
3887 		 * for fw to enforce it
3888 		 */
3889 #ifndef BNX2X_STOP_ON_ERROR
3890 		if (IS_VF(bp))
3891 #endif
3892 			tx_start_bd->vlan_or_ethertype =
3893 				cpu_to_le16(ntohs(eth->h_proto));
3894 #ifndef BNX2X_STOP_ON_ERROR
3895 		else
3896 			/* used by FW for packet accounting */
3897 			tx_start_bd->vlan_or_ethertype = cpu_to_le16(pkt_prod);
3898 #endif
3899 	}
3900 
3901 	nbd = 2; /* start_bd + pbd + frags (updated when pages are mapped) */
3902 
3903 	/* turn on parsing and get a BD */
3904 	bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3905 
3906 	if (xmit_type & XMIT_CSUM)
3907 		bnx2x_set_sbd_csum(bp, skb, tx_start_bd, xmit_type);
3908 
3909 	if (!CHIP_IS_E1x(bp)) {
3910 		pbd_e2 = &txdata->tx_desc_ring[bd_prod].parse_bd_e2;
3911 		memset(pbd_e2, 0, sizeof(struct eth_tx_parse_bd_e2));
3912 
3913 		if (xmit_type & XMIT_CSUM_ENC) {
3914 			u16 global_data = 0;
3915 
3916 			/* Set PBD in enc checksum offload case */
3917 			hlen = bnx2x_set_pbd_csum_enc(bp, skb,
3918 						      &pbd_e2_parsing_data,
3919 						      xmit_type);
3920 
3921 			/* turn on 2nd parsing and get a BD */
3922 			bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3923 
3924 			pbd2 = &txdata->tx_desc_ring[bd_prod].parse_2nd_bd;
3925 
3926 			memset(pbd2, 0, sizeof(*pbd2));
3927 
3928 			pbd_e2->data.tunnel_data.ip_hdr_start_inner_w =
3929 				(skb_inner_network_header(skb) -
3930 				 skb->data) >> 1;
3931 
3932 			if (xmit_type & XMIT_GSO_ENC)
3933 				bnx2x_update_pbds_gso_enc(skb, pbd_e2, pbd2,
3934 							  &global_data,
3935 							  xmit_type);
3936 
3937 			pbd2->global_data = cpu_to_le16(global_data);
3938 
3939 			/* add addition parse BD indication to start BD */
3940 			SET_FLAG(tx_start_bd->general_data,
3941 				 ETH_TX_START_BD_PARSE_NBDS, 1);
3942 			/* set encapsulation flag in start BD */
3943 			SET_FLAG(tx_start_bd->general_data,
3944 				 ETH_TX_START_BD_TUNNEL_EXIST, 1);
3945 
3946 			tx_buf->flags |= BNX2X_HAS_SECOND_PBD;
3947 
3948 			nbd++;
3949 		} else if (xmit_type & XMIT_CSUM) {
3950 			/* Set PBD in checksum offload case w/o encapsulation */
3951 			hlen = bnx2x_set_pbd_csum_e2(bp, skb,
3952 						     &pbd_e2_parsing_data,
3953 						     xmit_type);
3954 		}
3955 
3956 		bnx2x_set_ipv6_ext_e2(skb, &pbd_e2_parsing_data, xmit_type);
3957 		/* Add the macs to the parsing BD if this is a vf or if
3958 		 * Tx Switching is enabled.
3959 		 */
3960 		if (IS_VF(bp)) {
3961 			/* override GRE parameters in BD */
3962 			bnx2x_set_fw_mac_addr(&pbd_e2->data.mac_addr.src_hi,
3963 					      &pbd_e2->data.mac_addr.src_mid,
3964 					      &pbd_e2->data.mac_addr.src_lo,
3965 					      eth->h_source);
3966 
3967 			bnx2x_set_fw_mac_addr(&pbd_e2->data.mac_addr.dst_hi,
3968 					      &pbd_e2->data.mac_addr.dst_mid,
3969 					      &pbd_e2->data.mac_addr.dst_lo,
3970 					      eth->h_dest);
3971 		} else {
3972 			if (bp->flags & TX_SWITCHING)
3973 				bnx2x_set_fw_mac_addr(
3974 						&pbd_e2->data.mac_addr.dst_hi,
3975 						&pbd_e2->data.mac_addr.dst_mid,
3976 						&pbd_e2->data.mac_addr.dst_lo,
3977 						eth->h_dest);
3978 #ifdef BNX2X_STOP_ON_ERROR
3979 			/* Enforce security is always set in Stop on Error -
3980 			 * source mac should be present in the parsing BD
3981 			 */
3982 			bnx2x_set_fw_mac_addr(&pbd_e2->data.mac_addr.src_hi,
3983 					      &pbd_e2->data.mac_addr.src_mid,
3984 					      &pbd_e2->data.mac_addr.src_lo,
3985 					      eth->h_source);
3986 #endif
3987 		}
3988 
3989 		SET_FLAG(pbd_e2_parsing_data,
3990 			 ETH_TX_PARSE_BD_E2_ETH_ADDR_TYPE, mac_type);
3991 	} else {
3992 		u16 global_data = 0;
3993 		pbd_e1x = &txdata->tx_desc_ring[bd_prod].parse_bd_e1x;
3994 		memset(pbd_e1x, 0, sizeof(struct eth_tx_parse_bd_e1x));
3995 		/* Set PBD in checksum offload case */
3996 		if (xmit_type & XMIT_CSUM)
3997 			hlen = bnx2x_set_pbd_csum(bp, skb, pbd_e1x, xmit_type);
3998 
3999 		SET_FLAG(global_data,
4000 			 ETH_TX_PARSE_BD_E1X_ETH_ADDR_TYPE, mac_type);
4001 		pbd_e1x->global_data |= cpu_to_le16(global_data);
4002 	}
4003 
4004 	/* Setup the data pointer of the first BD of the packet */
4005 	tx_start_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
4006 	tx_start_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
4007 	tx_start_bd->nbytes = cpu_to_le16(skb_headlen(skb));
4008 	pkt_size = tx_start_bd->nbytes;
4009 
4010 	DP(NETIF_MSG_TX_QUEUED,
4011 	   "first bd @%p  addr (%x:%x)  nbytes %d  flags %x  vlan %x\n",
4012 	   tx_start_bd, tx_start_bd->addr_hi, tx_start_bd->addr_lo,
4013 	   le16_to_cpu(tx_start_bd->nbytes),
4014 	   tx_start_bd->bd_flags.as_bitfield,
4015 	   le16_to_cpu(tx_start_bd->vlan_or_ethertype));
4016 
4017 	if (xmit_type & XMIT_GSO) {
4018 
4019 		DP(NETIF_MSG_TX_QUEUED,
4020 		   "TSO packet len %d  hlen %d  total len %d  tso size %d\n",
4021 		   skb->len, hlen, skb_headlen(skb),
4022 		   skb_shinfo(skb)->gso_size);
4023 
4024 		tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_SW_LSO;
4025 
4026 		if (unlikely(skb_headlen(skb) > hlen)) {
4027 			nbd++;
4028 			bd_prod = bnx2x_tx_split(bp, txdata, tx_buf,
4029 						 &tx_start_bd, hlen,
4030 						 bd_prod);
4031 		}
4032 		if (!CHIP_IS_E1x(bp))
4033 			pbd_e2_parsing_data |=
4034 				(skb_shinfo(skb)->gso_size <<
4035 				 ETH_TX_PARSE_BD_E2_LSO_MSS_SHIFT) &
4036 				 ETH_TX_PARSE_BD_E2_LSO_MSS;
4037 		else
4038 			bnx2x_set_pbd_gso(skb, pbd_e1x, xmit_type);
4039 	}
4040 
4041 	/* Set the PBD's parsing_data field if not zero
4042 	 * (for the chips newer than 57711).
4043 	 */
4044 	if (pbd_e2_parsing_data)
4045 		pbd_e2->parsing_data = cpu_to_le32(pbd_e2_parsing_data);
4046 
4047 	tx_data_bd = (struct eth_tx_bd *)tx_start_bd;
4048 
4049 	/* Handle fragmented skb */
4050 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
4051 		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
4052 
4053 		mapping = skb_frag_dma_map(&bp->pdev->dev, frag, 0,
4054 					   skb_frag_size(frag), DMA_TO_DEVICE);
4055 		if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
4056 			unsigned int pkts_compl = 0, bytes_compl = 0;
4057 
4058 			DP(NETIF_MSG_TX_QUEUED,
4059 			   "Unable to map page - dropping packet...\n");
4060 
4061 			/* we need unmap all buffers already mapped
4062 			 * for this SKB;
4063 			 * first_bd->nbd need to be properly updated
4064 			 * before call to bnx2x_free_tx_pkt
4065 			 */
4066 			first_bd->nbd = cpu_to_le16(nbd);
4067 			bnx2x_free_tx_pkt(bp, txdata,
4068 					  TX_BD(txdata->tx_pkt_prod),
4069 					  &pkts_compl, &bytes_compl);
4070 			return NETDEV_TX_OK;
4071 		}
4072 
4073 		bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
4074 		tx_data_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
4075 		if (total_pkt_bd == NULL)
4076 			total_pkt_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
4077 
4078 		tx_data_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
4079 		tx_data_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
4080 		tx_data_bd->nbytes = cpu_to_le16(skb_frag_size(frag));
4081 		le16_add_cpu(&pkt_size, skb_frag_size(frag));
4082 		nbd++;
4083 
4084 		DP(NETIF_MSG_TX_QUEUED,
4085 		   "frag %d  bd @%p  addr (%x:%x)  nbytes %d\n",
4086 		   i, tx_data_bd, tx_data_bd->addr_hi, tx_data_bd->addr_lo,
4087 		   le16_to_cpu(tx_data_bd->nbytes));
4088 	}
4089 
4090 	DP(NETIF_MSG_TX_QUEUED, "last bd @%p\n", tx_data_bd);
4091 
4092 	/* update with actual num BDs */
4093 	first_bd->nbd = cpu_to_le16(nbd);
4094 
4095 	bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
4096 
4097 	/* now send a tx doorbell, counting the next BD
4098 	 * if the packet contains or ends with it
4099 	 */
4100 	if (TX_BD_POFF(bd_prod) < nbd)
4101 		nbd++;
4102 
4103 	/* total_pkt_bytes should be set on the first data BD if
4104 	 * it's not an LSO packet and there is more than one
4105 	 * data BD. In this case pkt_size is limited by an MTU value.
4106 	 * However we prefer to set it for an LSO packet (while we don't
4107 	 * have to) in order to save some CPU cycles in a none-LSO
4108 	 * case, when we much more care about them.
4109 	 */
4110 	if (total_pkt_bd != NULL)
4111 		total_pkt_bd->total_pkt_bytes = pkt_size;
4112 
4113 	if (pbd_e1x)
4114 		DP(NETIF_MSG_TX_QUEUED,
4115 		   "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",
4116 		   pbd_e1x, pbd_e1x->global_data, pbd_e1x->ip_hlen_w,
4117 		   pbd_e1x->ip_id, pbd_e1x->lso_mss, pbd_e1x->tcp_flags,
4118 		   pbd_e1x->tcp_pseudo_csum, pbd_e1x->tcp_send_seq,
4119 		    le16_to_cpu(pbd_e1x->total_hlen_w));
4120 	if (pbd_e2)
4121 		DP(NETIF_MSG_TX_QUEUED,
4122 		   "PBD (E2) @%p  dst %x %x %x src %x %x %x parsing_data %x\n",
4123 		   pbd_e2,
4124 		   pbd_e2->data.mac_addr.dst_hi,
4125 		   pbd_e2->data.mac_addr.dst_mid,
4126 		   pbd_e2->data.mac_addr.dst_lo,
4127 		   pbd_e2->data.mac_addr.src_hi,
4128 		   pbd_e2->data.mac_addr.src_mid,
4129 		   pbd_e2->data.mac_addr.src_lo,
4130 		   pbd_e2->parsing_data);
4131 	DP(NETIF_MSG_TX_QUEUED, "doorbell: nbd %d  bd %u\n", nbd, bd_prod);
4132 
4133 	netdev_tx_sent_queue(txq, skb->len);
4134 
4135 	skb_tx_timestamp(skb);
4136 
4137 	txdata->tx_pkt_prod++;
4138 	/*
4139 	 * Make sure that the BD data is updated before updating the producer
4140 	 * since FW might read the BD right after the producer is updated.
4141 	 * This is only applicable for weak-ordered memory model archs such
4142 	 * as IA-64. The following barrier is also mandatory since FW will
4143 	 * assumes packets must have BDs.
4144 	 */
4145 	wmb();
4146 
4147 	txdata->tx_db.data.prod += nbd;
4148 	barrier();
4149 
4150 	DOORBELL(bp, txdata->cid, txdata->tx_db.raw);
4151 
4152 	mmiowb();
4153 
4154 	txdata->tx_bd_prod += nbd;
4155 
4156 	if (unlikely(bnx2x_tx_avail(bp, txdata) < MAX_DESC_PER_TX_PKT)) {
4157 		netif_tx_stop_queue(txq);
4158 
4159 		/* paired memory barrier is in bnx2x_tx_int(), we have to keep
4160 		 * ordering of set_bit() in netif_tx_stop_queue() and read of
4161 		 * fp->bd_tx_cons */
4162 		smp_mb();
4163 
4164 		bnx2x_fp_qstats(bp, txdata->parent_fp)->driver_xoff++;
4165 		if (bnx2x_tx_avail(bp, txdata) >= MAX_DESC_PER_TX_PKT)
4166 			netif_tx_wake_queue(txq);
4167 	}
4168 	txdata->tx_pkt++;
4169 
4170 	return NETDEV_TX_OK;
4171 }
4172 
4173 void bnx2x_get_c2s_mapping(struct bnx2x *bp, u8 *c2s_map, u8 *c2s_default)
4174 {
4175 	int mfw_vn = BP_FW_MB_IDX(bp);
4176 	u32 tmp;
4177 
4178 	/* If the shmem shouldn't affect configuration, reflect */
4179 	if (!IS_MF_BD(bp)) {
4180 		int i;
4181 
4182 		for (i = 0; i < BNX2X_MAX_PRIORITY; i++)
4183 			c2s_map[i] = i;
4184 		*c2s_default = 0;
4185 
4186 		return;
4187 	}
4188 
4189 	tmp = SHMEM2_RD(bp, c2s_pcp_map_lower[mfw_vn]);
4190 	tmp = (__force u32)be32_to_cpu((__force __be32)tmp);
4191 	c2s_map[0] = tmp & 0xff;
4192 	c2s_map[1] = (tmp >> 8) & 0xff;
4193 	c2s_map[2] = (tmp >> 16) & 0xff;
4194 	c2s_map[3] = (tmp >> 24) & 0xff;
4195 
4196 	tmp = SHMEM2_RD(bp, c2s_pcp_map_upper[mfw_vn]);
4197 	tmp = (__force u32)be32_to_cpu((__force __be32)tmp);
4198 	c2s_map[4] = tmp & 0xff;
4199 	c2s_map[5] = (tmp >> 8) & 0xff;
4200 	c2s_map[6] = (tmp >> 16) & 0xff;
4201 	c2s_map[7] = (tmp >> 24) & 0xff;
4202 
4203 	tmp = SHMEM2_RD(bp, c2s_pcp_map_default[mfw_vn]);
4204 	tmp = (__force u32)be32_to_cpu((__force __be32)tmp);
4205 	*c2s_default = (tmp >> (8 * mfw_vn)) & 0xff;
4206 }
4207 
4208 /**
4209  * bnx2x_setup_tc - routine to configure net_device for multi tc
4210  *
4211  * @netdev: net device to configure
4212  * @tc: number of traffic classes to enable
4213  *
4214  * callback connected to the ndo_setup_tc function pointer
4215  */
4216 int bnx2x_setup_tc(struct net_device *dev, u8 num_tc)
4217 {
4218 	struct bnx2x *bp = netdev_priv(dev);
4219 	u8 c2s_map[BNX2X_MAX_PRIORITY], c2s_def;
4220 	int cos, prio, count, offset;
4221 
4222 	/* setup tc must be called under rtnl lock */
4223 	ASSERT_RTNL();
4224 
4225 	/* no traffic classes requested. Aborting */
4226 	if (!num_tc) {
4227 		netdev_reset_tc(dev);
4228 		return 0;
4229 	}
4230 
4231 	/* requested to support too many traffic classes */
4232 	if (num_tc > bp->max_cos) {
4233 		BNX2X_ERR("support for too many traffic classes requested: %d. Max supported is %d\n",
4234 			  num_tc, bp->max_cos);
4235 		return -EINVAL;
4236 	}
4237 
4238 	/* declare amount of supported traffic classes */
4239 	if (netdev_set_num_tc(dev, num_tc)) {
4240 		BNX2X_ERR("failed to declare %d traffic classes\n", num_tc);
4241 		return -EINVAL;
4242 	}
4243 
4244 	bnx2x_get_c2s_mapping(bp, c2s_map, &c2s_def);
4245 
4246 	/* configure priority to traffic class mapping */
4247 	for (prio = 0; prio < BNX2X_MAX_PRIORITY; prio++) {
4248 		int outer_prio = c2s_map[prio];
4249 
4250 		netdev_set_prio_tc_map(dev, prio, bp->prio_to_cos[outer_prio]);
4251 		DP(BNX2X_MSG_SP | NETIF_MSG_IFUP,
4252 		   "mapping priority %d to tc %d\n",
4253 		   outer_prio, bp->prio_to_cos[outer_prio]);
4254 	}
4255 
4256 	/* Use this configuration to differentiate tc0 from other COSes
4257 	   This can be used for ets or pfc, and save the effort of setting
4258 	   up a multio class queue disc or negotiating DCBX with a switch
4259 	netdev_set_prio_tc_map(dev, 0, 0);
4260 	DP(BNX2X_MSG_SP, "mapping priority %d to tc %d\n", 0, 0);
4261 	for (prio = 1; prio < 16; prio++) {
4262 		netdev_set_prio_tc_map(dev, prio, 1);
4263 		DP(BNX2X_MSG_SP, "mapping priority %d to tc %d\n", prio, 1);
4264 	} */
4265 
4266 	/* configure traffic class to transmission queue mapping */
4267 	for (cos = 0; cos < bp->max_cos; cos++) {
4268 		count = BNX2X_NUM_ETH_QUEUES(bp);
4269 		offset = cos * BNX2X_NUM_NON_CNIC_QUEUES(bp);
4270 		netdev_set_tc_queue(dev, cos, count, offset);
4271 		DP(BNX2X_MSG_SP | NETIF_MSG_IFUP,
4272 		   "mapping tc %d to offset %d count %d\n",
4273 		   cos, offset, count);
4274 	}
4275 
4276 	return 0;
4277 }
4278 
4279 int __bnx2x_setup_tc(struct net_device *dev, u32 handle, __be16 proto,
4280 		     struct tc_to_netdev *tc)
4281 {
4282 	if (tc->type != TC_SETUP_MQPRIO)
4283 		return -EINVAL;
4284 	return bnx2x_setup_tc(dev, tc->tc);
4285 }
4286 
4287 /* called with rtnl_lock */
4288 int bnx2x_change_mac_addr(struct net_device *dev, void *p)
4289 {
4290 	struct sockaddr *addr = p;
4291 	struct bnx2x *bp = netdev_priv(dev);
4292 	int rc = 0;
4293 
4294 	if (!is_valid_ether_addr(addr->sa_data)) {
4295 		BNX2X_ERR("Requested MAC address is not valid\n");
4296 		return -EINVAL;
4297 	}
4298 
4299 	if (IS_MF_STORAGE_ONLY(bp)) {
4300 		BNX2X_ERR("Can't change address on STORAGE ONLY function\n");
4301 		return -EINVAL;
4302 	}
4303 
4304 	if (netif_running(dev))  {
4305 		rc = bnx2x_set_eth_mac(bp, false);
4306 		if (rc)
4307 			return rc;
4308 	}
4309 
4310 	memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
4311 
4312 	if (netif_running(dev))
4313 		rc = bnx2x_set_eth_mac(bp, true);
4314 
4315 	if (IS_PF(bp) && SHMEM2_HAS(bp, curr_cfg))
4316 		SHMEM2_WR(bp, curr_cfg, CURR_CFG_MET_OS);
4317 
4318 	return rc;
4319 }
4320 
4321 static void bnx2x_free_fp_mem_at(struct bnx2x *bp, int fp_index)
4322 {
4323 	union host_hc_status_block *sb = &bnx2x_fp(bp, fp_index, status_blk);
4324 	struct bnx2x_fastpath *fp = &bp->fp[fp_index];
4325 	u8 cos;
4326 
4327 	/* Common */
4328 
4329 	if (IS_FCOE_IDX(fp_index)) {
4330 		memset(sb, 0, sizeof(union host_hc_status_block));
4331 		fp->status_blk_mapping = 0;
4332 	} else {
4333 		/* status blocks */
4334 		if (!CHIP_IS_E1x(bp))
4335 			BNX2X_PCI_FREE(sb->e2_sb,
4336 				       bnx2x_fp(bp, fp_index,
4337 						status_blk_mapping),
4338 				       sizeof(struct host_hc_status_block_e2));
4339 		else
4340 			BNX2X_PCI_FREE(sb->e1x_sb,
4341 				       bnx2x_fp(bp, fp_index,
4342 						status_blk_mapping),
4343 				       sizeof(struct host_hc_status_block_e1x));
4344 	}
4345 
4346 	/* Rx */
4347 	if (!skip_rx_queue(bp, fp_index)) {
4348 		bnx2x_free_rx_bds(fp);
4349 
4350 		/* fastpath rx rings: rx_buf rx_desc rx_comp */
4351 		BNX2X_FREE(bnx2x_fp(bp, fp_index, rx_buf_ring));
4352 		BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_desc_ring),
4353 			       bnx2x_fp(bp, fp_index, rx_desc_mapping),
4354 			       sizeof(struct eth_rx_bd) * NUM_RX_BD);
4355 
4356 		BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_comp_ring),
4357 			       bnx2x_fp(bp, fp_index, rx_comp_mapping),
4358 			       sizeof(struct eth_fast_path_rx_cqe) *
4359 			       NUM_RCQ_BD);
4360 
4361 		/* SGE ring */
4362 		BNX2X_FREE(bnx2x_fp(bp, fp_index, rx_page_ring));
4363 		BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_sge_ring),
4364 			       bnx2x_fp(bp, fp_index, rx_sge_mapping),
4365 			       BCM_PAGE_SIZE * NUM_RX_SGE_PAGES);
4366 	}
4367 
4368 	/* Tx */
4369 	if (!skip_tx_queue(bp, fp_index)) {
4370 		/* fastpath tx rings: tx_buf tx_desc */
4371 		for_each_cos_in_tx_queue(fp, cos) {
4372 			struct bnx2x_fp_txdata *txdata = fp->txdata_ptr[cos];
4373 
4374 			DP(NETIF_MSG_IFDOWN,
4375 			   "freeing tx memory of fp %d cos %d cid %d\n",
4376 			   fp_index, cos, txdata->cid);
4377 
4378 			BNX2X_FREE(txdata->tx_buf_ring);
4379 			BNX2X_PCI_FREE(txdata->tx_desc_ring,
4380 				txdata->tx_desc_mapping,
4381 				sizeof(union eth_tx_bd_types) * NUM_TX_BD);
4382 		}
4383 	}
4384 	/* end of fastpath */
4385 }
4386 
4387 static void bnx2x_free_fp_mem_cnic(struct bnx2x *bp)
4388 {
4389 	int i;
4390 	for_each_cnic_queue(bp, i)
4391 		bnx2x_free_fp_mem_at(bp, i);
4392 }
4393 
4394 void bnx2x_free_fp_mem(struct bnx2x *bp)
4395 {
4396 	int i;
4397 	for_each_eth_queue(bp, i)
4398 		bnx2x_free_fp_mem_at(bp, i);
4399 }
4400 
4401 static void set_sb_shortcuts(struct bnx2x *bp, int index)
4402 {
4403 	union host_hc_status_block status_blk = bnx2x_fp(bp, index, status_blk);
4404 	if (!CHIP_IS_E1x(bp)) {
4405 		bnx2x_fp(bp, index, sb_index_values) =
4406 			(__le16 *)status_blk.e2_sb->sb.index_values;
4407 		bnx2x_fp(bp, index, sb_running_index) =
4408 			(__le16 *)status_blk.e2_sb->sb.running_index;
4409 	} else {
4410 		bnx2x_fp(bp, index, sb_index_values) =
4411 			(__le16 *)status_blk.e1x_sb->sb.index_values;
4412 		bnx2x_fp(bp, index, sb_running_index) =
4413 			(__le16 *)status_blk.e1x_sb->sb.running_index;
4414 	}
4415 }
4416 
4417 /* Returns the number of actually allocated BDs */
4418 static int bnx2x_alloc_rx_bds(struct bnx2x_fastpath *fp,
4419 			      int rx_ring_size)
4420 {
4421 	struct bnx2x *bp = fp->bp;
4422 	u16 ring_prod, cqe_ring_prod;
4423 	int i, failure_cnt = 0;
4424 
4425 	fp->rx_comp_cons = 0;
4426 	cqe_ring_prod = ring_prod = 0;
4427 
4428 	/* This routine is called only during fo init so
4429 	 * fp->eth_q_stats.rx_skb_alloc_failed = 0
4430 	 */
4431 	for (i = 0; i < rx_ring_size; i++) {
4432 		if (bnx2x_alloc_rx_data(bp, fp, ring_prod, GFP_KERNEL) < 0) {
4433 			failure_cnt++;
4434 			continue;
4435 		}
4436 		ring_prod = NEXT_RX_IDX(ring_prod);
4437 		cqe_ring_prod = NEXT_RCQ_IDX(cqe_ring_prod);
4438 		WARN_ON(ring_prod <= (i - failure_cnt));
4439 	}
4440 
4441 	if (failure_cnt)
4442 		BNX2X_ERR("was only able to allocate %d rx skbs on queue[%d]\n",
4443 			  i - failure_cnt, fp->index);
4444 
4445 	fp->rx_bd_prod = ring_prod;
4446 	/* Limit the CQE producer by the CQE ring size */
4447 	fp->rx_comp_prod = min_t(u16, NUM_RCQ_RINGS*RCQ_DESC_CNT,
4448 			       cqe_ring_prod);
4449 
4450 	bnx2x_fp_stats(bp, fp)->eth_q_stats.rx_skb_alloc_failed += failure_cnt;
4451 
4452 	return i - failure_cnt;
4453 }
4454 
4455 static void bnx2x_set_next_page_rx_cq(struct bnx2x_fastpath *fp)
4456 {
4457 	int i;
4458 
4459 	for (i = 1; i <= NUM_RCQ_RINGS; i++) {
4460 		struct eth_rx_cqe_next_page *nextpg;
4461 
4462 		nextpg = (struct eth_rx_cqe_next_page *)
4463 			&fp->rx_comp_ring[RCQ_DESC_CNT * i - 1];
4464 		nextpg->addr_hi =
4465 			cpu_to_le32(U64_HI(fp->rx_comp_mapping +
4466 				   BCM_PAGE_SIZE*(i % NUM_RCQ_RINGS)));
4467 		nextpg->addr_lo =
4468 			cpu_to_le32(U64_LO(fp->rx_comp_mapping +
4469 				   BCM_PAGE_SIZE*(i % NUM_RCQ_RINGS)));
4470 	}
4471 }
4472 
4473 static int bnx2x_alloc_fp_mem_at(struct bnx2x *bp, int index)
4474 {
4475 	union host_hc_status_block *sb;
4476 	struct bnx2x_fastpath *fp = &bp->fp[index];
4477 	int ring_size = 0;
4478 	u8 cos;
4479 	int rx_ring_size = 0;
4480 
4481 	if (!bp->rx_ring_size && IS_MF_STORAGE_ONLY(bp)) {
4482 		rx_ring_size = MIN_RX_SIZE_NONTPA;
4483 		bp->rx_ring_size = rx_ring_size;
4484 	} else if (!bp->rx_ring_size) {
4485 		rx_ring_size = MAX_RX_AVAIL/BNX2X_NUM_RX_QUEUES(bp);
4486 
4487 		if (CHIP_IS_E3(bp)) {
4488 			u32 cfg = SHMEM_RD(bp,
4489 					   dev_info.port_hw_config[BP_PORT(bp)].
4490 					   default_cfg);
4491 
4492 			/* Decrease ring size for 1G functions */
4493 			if ((cfg & PORT_HW_CFG_NET_SERDES_IF_MASK) ==
4494 			    PORT_HW_CFG_NET_SERDES_IF_SGMII)
4495 				rx_ring_size /= 10;
4496 		}
4497 
4498 		/* allocate at least number of buffers required by FW */
4499 		rx_ring_size = max_t(int, bp->disable_tpa ? MIN_RX_SIZE_NONTPA :
4500 				     MIN_RX_SIZE_TPA, rx_ring_size);
4501 
4502 		bp->rx_ring_size = rx_ring_size;
4503 	} else /* if rx_ring_size specified - use it */
4504 		rx_ring_size = bp->rx_ring_size;
4505 
4506 	DP(BNX2X_MSG_SP, "calculated rx_ring_size %d\n", rx_ring_size);
4507 
4508 	/* Common */
4509 	sb = &bnx2x_fp(bp, index, status_blk);
4510 
4511 	if (!IS_FCOE_IDX(index)) {
4512 		/* status blocks */
4513 		if (!CHIP_IS_E1x(bp)) {
4514 			sb->e2_sb = BNX2X_PCI_ALLOC(&bnx2x_fp(bp, index, status_blk_mapping),
4515 						    sizeof(struct host_hc_status_block_e2));
4516 			if (!sb->e2_sb)
4517 				goto alloc_mem_err;
4518 		} else {
4519 			sb->e1x_sb = BNX2X_PCI_ALLOC(&bnx2x_fp(bp, index, status_blk_mapping),
4520 						     sizeof(struct host_hc_status_block_e1x));
4521 			if (!sb->e1x_sb)
4522 				goto alloc_mem_err;
4523 		}
4524 	}
4525 
4526 	/* FCoE Queue uses Default SB and doesn't ACK the SB, thus no need to
4527 	 * set shortcuts for it.
4528 	 */
4529 	if (!IS_FCOE_IDX(index))
4530 		set_sb_shortcuts(bp, index);
4531 
4532 	/* Tx */
4533 	if (!skip_tx_queue(bp, index)) {
4534 		/* fastpath tx rings: tx_buf tx_desc */
4535 		for_each_cos_in_tx_queue(fp, cos) {
4536 			struct bnx2x_fp_txdata *txdata = fp->txdata_ptr[cos];
4537 
4538 			DP(NETIF_MSG_IFUP,
4539 			   "allocating tx memory of fp %d cos %d\n",
4540 			   index, cos);
4541 
4542 			txdata->tx_buf_ring = kcalloc(NUM_TX_BD,
4543 						      sizeof(struct sw_tx_bd),
4544 						      GFP_KERNEL);
4545 			if (!txdata->tx_buf_ring)
4546 				goto alloc_mem_err;
4547 			txdata->tx_desc_ring = BNX2X_PCI_ALLOC(&txdata->tx_desc_mapping,
4548 							       sizeof(union eth_tx_bd_types) * NUM_TX_BD);
4549 			if (!txdata->tx_desc_ring)
4550 				goto alloc_mem_err;
4551 		}
4552 	}
4553 
4554 	/* Rx */
4555 	if (!skip_rx_queue(bp, index)) {
4556 		/* fastpath rx rings: rx_buf rx_desc rx_comp */
4557 		bnx2x_fp(bp, index, rx_buf_ring) =
4558 			kcalloc(NUM_RX_BD, sizeof(struct sw_rx_bd), GFP_KERNEL);
4559 		if (!bnx2x_fp(bp, index, rx_buf_ring))
4560 			goto alloc_mem_err;
4561 		bnx2x_fp(bp, index, rx_desc_ring) =
4562 			BNX2X_PCI_ALLOC(&bnx2x_fp(bp, index, rx_desc_mapping),
4563 					sizeof(struct eth_rx_bd) * NUM_RX_BD);
4564 		if (!bnx2x_fp(bp, index, rx_desc_ring))
4565 			goto alloc_mem_err;
4566 
4567 		/* Seed all CQEs by 1s */
4568 		bnx2x_fp(bp, index, rx_comp_ring) =
4569 			BNX2X_PCI_FALLOC(&bnx2x_fp(bp, index, rx_comp_mapping),
4570 					 sizeof(struct eth_fast_path_rx_cqe) * NUM_RCQ_BD);
4571 		if (!bnx2x_fp(bp, index, rx_comp_ring))
4572 			goto alloc_mem_err;
4573 
4574 		/* SGE ring */
4575 		bnx2x_fp(bp, index, rx_page_ring) =
4576 			kcalloc(NUM_RX_SGE, sizeof(struct sw_rx_page),
4577 				GFP_KERNEL);
4578 		if (!bnx2x_fp(bp, index, rx_page_ring))
4579 			goto alloc_mem_err;
4580 		bnx2x_fp(bp, index, rx_sge_ring) =
4581 			BNX2X_PCI_ALLOC(&bnx2x_fp(bp, index, rx_sge_mapping),
4582 					BCM_PAGE_SIZE * NUM_RX_SGE_PAGES);
4583 		if (!bnx2x_fp(bp, index, rx_sge_ring))
4584 			goto alloc_mem_err;
4585 		/* RX BD ring */
4586 		bnx2x_set_next_page_rx_bd(fp);
4587 
4588 		/* CQ ring */
4589 		bnx2x_set_next_page_rx_cq(fp);
4590 
4591 		/* BDs */
4592 		ring_size = bnx2x_alloc_rx_bds(fp, rx_ring_size);
4593 		if (ring_size < rx_ring_size)
4594 			goto alloc_mem_err;
4595 	}
4596 
4597 	return 0;
4598 
4599 /* handles low memory cases */
4600 alloc_mem_err:
4601 	BNX2X_ERR("Unable to allocate full memory for queue %d (size %d)\n",
4602 						index, ring_size);
4603 	/* FW will drop all packets if queue is not big enough,
4604 	 * In these cases we disable the queue
4605 	 * Min size is different for OOO, TPA and non-TPA queues
4606 	 */
4607 	if (ring_size < (fp->mode == TPA_MODE_DISABLED ?
4608 				MIN_RX_SIZE_NONTPA : MIN_RX_SIZE_TPA)) {
4609 			/* release memory allocated for this queue */
4610 			bnx2x_free_fp_mem_at(bp, index);
4611 			return -ENOMEM;
4612 	}
4613 	return 0;
4614 }
4615 
4616 static int bnx2x_alloc_fp_mem_cnic(struct bnx2x *bp)
4617 {
4618 	if (!NO_FCOE(bp))
4619 		/* FCoE */
4620 		if (bnx2x_alloc_fp_mem_at(bp, FCOE_IDX(bp)))
4621 			/* we will fail load process instead of mark
4622 			 * NO_FCOE_FLAG
4623 			 */
4624 			return -ENOMEM;
4625 
4626 	return 0;
4627 }
4628 
4629 static int bnx2x_alloc_fp_mem(struct bnx2x *bp)
4630 {
4631 	int i;
4632 
4633 	/* 1. Allocate FP for leading - fatal if error
4634 	 * 2. Allocate RSS - fix number of queues if error
4635 	 */
4636 
4637 	/* leading */
4638 	if (bnx2x_alloc_fp_mem_at(bp, 0))
4639 		return -ENOMEM;
4640 
4641 	/* RSS */
4642 	for_each_nondefault_eth_queue(bp, i)
4643 		if (bnx2x_alloc_fp_mem_at(bp, i))
4644 			break;
4645 
4646 	/* handle memory failures */
4647 	if (i != BNX2X_NUM_ETH_QUEUES(bp)) {
4648 		int delta = BNX2X_NUM_ETH_QUEUES(bp) - i;
4649 
4650 		WARN_ON(delta < 0);
4651 		bnx2x_shrink_eth_fp(bp, delta);
4652 		if (CNIC_SUPPORT(bp))
4653 			/* move non eth FPs next to last eth FP
4654 			 * must be done in that order
4655 			 * FCOE_IDX < FWD_IDX < OOO_IDX
4656 			 */
4657 
4658 			/* move FCoE fp even NO_FCOE_FLAG is on */
4659 			bnx2x_move_fp(bp, FCOE_IDX(bp), FCOE_IDX(bp) - delta);
4660 		bp->num_ethernet_queues -= delta;
4661 		bp->num_queues = bp->num_ethernet_queues +
4662 				 bp->num_cnic_queues;
4663 		BNX2X_ERR("Adjusted num of queues from %d to %d\n",
4664 			  bp->num_queues + delta, bp->num_queues);
4665 	}
4666 
4667 	return 0;
4668 }
4669 
4670 void bnx2x_free_mem_bp(struct bnx2x *bp)
4671 {
4672 	int i;
4673 
4674 	for (i = 0; i < bp->fp_array_size; i++)
4675 		kfree(bp->fp[i].tpa_info);
4676 	kfree(bp->fp);
4677 	kfree(bp->sp_objs);
4678 	kfree(bp->fp_stats);
4679 	kfree(bp->bnx2x_txq);
4680 	kfree(bp->msix_table);
4681 	kfree(bp->ilt);
4682 }
4683 
4684 int bnx2x_alloc_mem_bp(struct bnx2x *bp)
4685 {
4686 	struct bnx2x_fastpath *fp;
4687 	struct msix_entry *tbl;
4688 	struct bnx2x_ilt *ilt;
4689 	int msix_table_size = 0;
4690 	int fp_array_size, txq_array_size;
4691 	int i;
4692 
4693 	/*
4694 	 * The biggest MSI-X table we might need is as a maximum number of fast
4695 	 * path IGU SBs plus default SB (for PF only).
4696 	 */
4697 	msix_table_size = bp->igu_sb_cnt;
4698 	if (IS_PF(bp))
4699 		msix_table_size++;
4700 	BNX2X_DEV_INFO("msix_table_size %d\n", msix_table_size);
4701 
4702 	/* fp array: RSS plus CNIC related L2 queues */
4703 	fp_array_size = BNX2X_MAX_RSS_COUNT(bp) + CNIC_SUPPORT(bp);
4704 	bp->fp_array_size = fp_array_size;
4705 	BNX2X_DEV_INFO("fp_array_size %d\n", bp->fp_array_size);
4706 
4707 	fp = kcalloc(bp->fp_array_size, sizeof(*fp), GFP_KERNEL);
4708 	if (!fp)
4709 		goto alloc_err;
4710 	for (i = 0; i < bp->fp_array_size; i++) {
4711 		fp[i].tpa_info =
4712 			kcalloc(ETH_MAX_AGGREGATION_QUEUES_E1H_E2,
4713 				sizeof(struct bnx2x_agg_info), GFP_KERNEL);
4714 		if (!(fp[i].tpa_info))
4715 			goto alloc_err;
4716 	}
4717 
4718 	bp->fp = fp;
4719 
4720 	/* allocate sp objs */
4721 	bp->sp_objs = kcalloc(bp->fp_array_size, sizeof(struct bnx2x_sp_objs),
4722 			      GFP_KERNEL);
4723 	if (!bp->sp_objs)
4724 		goto alloc_err;
4725 
4726 	/* allocate fp_stats */
4727 	bp->fp_stats = kcalloc(bp->fp_array_size, sizeof(struct bnx2x_fp_stats),
4728 			       GFP_KERNEL);
4729 	if (!bp->fp_stats)
4730 		goto alloc_err;
4731 
4732 	/* Allocate memory for the transmission queues array */
4733 	txq_array_size =
4734 		BNX2X_MAX_RSS_COUNT(bp) * BNX2X_MULTI_TX_COS + CNIC_SUPPORT(bp);
4735 	BNX2X_DEV_INFO("txq_array_size %d", txq_array_size);
4736 
4737 	bp->bnx2x_txq = kcalloc(txq_array_size, sizeof(struct bnx2x_fp_txdata),
4738 				GFP_KERNEL);
4739 	if (!bp->bnx2x_txq)
4740 		goto alloc_err;
4741 
4742 	/* msix table */
4743 	tbl = kcalloc(msix_table_size, sizeof(*tbl), GFP_KERNEL);
4744 	if (!tbl)
4745 		goto alloc_err;
4746 	bp->msix_table = tbl;
4747 
4748 	/* ilt */
4749 	ilt = kzalloc(sizeof(*ilt), GFP_KERNEL);
4750 	if (!ilt)
4751 		goto alloc_err;
4752 	bp->ilt = ilt;
4753 
4754 	return 0;
4755 alloc_err:
4756 	bnx2x_free_mem_bp(bp);
4757 	return -ENOMEM;
4758 }
4759 
4760 int bnx2x_reload_if_running(struct net_device *dev)
4761 {
4762 	struct bnx2x *bp = netdev_priv(dev);
4763 
4764 	if (unlikely(!netif_running(dev)))
4765 		return 0;
4766 
4767 	bnx2x_nic_unload(bp, UNLOAD_NORMAL, true);
4768 	return bnx2x_nic_load(bp, LOAD_NORMAL);
4769 }
4770 
4771 int bnx2x_get_cur_phy_idx(struct bnx2x *bp)
4772 {
4773 	u32 sel_phy_idx = 0;
4774 	if (bp->link_params.num_phys <= 1)
4775 		return INT_PHY;
4776 
4777 	if (bp->link_vars.link_up) {
4778 		sel_phy_idx = EXT_PHY1;
4779 		/* In case link is SERDES, check if the EXT_PHY2 is the one */
4780 		if ((bp->link_vars.link_status & LINK_STATUS_SERDES_LINK) &&
4781 		    (bp->link_params.phy[EXT_PHY2].supported & SUPPORTED_FIBRE))
4782 			sel_phy_idx = EXT_PHY2;
4783 	} else {
4784 
4785 		switch (bnx2x_phy_selection(&bp->link_params)) {
4786 		case PORT_HW_CFG_PHY_SELECTION_HARDWARE_DEFAULT:
4787 		case PORT_HW_CFG_PHY_SELECTION_FIRST_PHY:
4788 		case PORT_HW_CFG_PHY_SELECTION_FIRST_PHY_PRIORITY:
4789 		       sel_phy_idx = EXT_PHY1;
4790 		       break;
4791 		case PORT_HW_CFG_PHY_SELECTION_SECOND_PHY:
4792 		case PORT_HW_CFG_PHY_SELECTION_SECOND_PHY_PRIORITY:
4793 		       sel_phy_idx = EXT_PHY2;
4794 		       break;
4795 		}
4796 	}
4797 
4798 	return sel_phy_idx;
4799 }
4800 int bnx2x_get_link_cfg_idx(struct bnx2x *bp)
4801 {
4802 	u32 sel_phy_idx = bnx2x_get_cur_phy_idx(bp);
4803 	/*
4804 	 * The selected activated PHY is always after swapping (in case PHY
4805 	 * swapping is enabled). So when swapping is enabled, we need to reverse
4806 	 * the configuration
4807 	 */
4808 
4809 	if (bp->link_params.multi_phy_config &
4810 	    PORT_HW_CFG_PHY_SWAPPED_ENABLED) {
4811 		if (sel_phy_idx == EXT_PHY1)
4812 			sel_phy_idx = EXT_PHY2;
4813 		else if (sel_phy_idx == EXT_PHY2)
4814 			sel_phy_idx = EXT_PHY1;
4815 	}
4816 	return LINK_CONFIG_IDX(sel_phy_idx);
4817 }
4818 
4819 #ifdef NETDEV_FCOE_WWNN
4820 int bnx2x_fcoe_get_wwn(struct net_device *dev, u64 *wwn, int type)
4821 {
4822 	struct bnx2x *bp = netdev_priv(dev);
4823 	struct cnic_eth_dev *cp = &bp->cnic_eth_dev;
4824 
4825 	switch (type) {
4826 	case NETDEV_FCOE_WWNN:
4827 		*wwn = HILO_U64(cp->fcoe_wwn_node_name_hi,
4828 				cp->fcoe_wwn_node_name_lo);
4829 		break;
4830 	case NETDEV_FCOE_WWPN:
4831 		*wwn = HILO_U64(cp->fcoe_wwn_port_name_hi,
4832 				cp->fcoe_wwn_port_name_lo);
4833 		break;
4834 	default:
4835 		BNX2X_ERR("Wrong WWN type requested - %d\n", type);
4836 		return -EINVAL;
4837 	}
4838 
4839 	return 0;
4840 }
4841 #endif
4842 
4843 /* called with rtnl_lock */
4844 int bnx2x_change_mtu(struct net_device *dev, int new_mtu)
4845 {
4846 	struct bnx2x *bp = netdev_priv(dev);
4847 
4848 	if (pci_num_vf(bp->pdev)) {
4849 		DP(BNX2X_MSG_IOV, "VFs are enabled, can not change MTU\n");
4850 		return -EPERM;
4851 	}
4852 
4853 	if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
4854 		BNX2X_ERR("Can't perform change MTU during parity recovery\n");
4855 		return -EAGAIN;
4856 	}
4857 
4858 	if ((new_mtu > ETH_MAX_JUMBO_PACKET_SIZE) ||
4859 	    ((new_mtu + ETH_HLEN) < ETH_MIN_PACKET_SIZE)) {
4860 		BNX2X_ERR("Can't support requested MTU size\n");
4861 		return -EINVAL;
4862 	}
4863 
4864 	/* This does not race with packet allocation
4865 	 * because the actual alloc size is
4866 	 * only updated as part of load
4867 	 */
4868 	dev->mtu = new_mtu;
4869 
4870 	if (IS_PF(bp) && SHMEM2_HAS(bp, curr_cfg))
4871 		SHMEM2_WR(bp, curr_cfg, CURR_CFG_MET_OS);
4872 
4873 	return bnx2x_reload_if_running(dev);
4874 }
4875 
4876 netdev_features_t bnx2x_fix_features(struct net_device *dev,
4877 				     netdev_features_t features)
4878 {
4879 	struct bnx2x *bp = netdev_priv(dev);
4880 
4881 	if (pci_num_vf(bp->pdev)) {
4882 		netdev_features_t changed = dev->features ^ features;
4883 
4884 		/* Revert the requested changes in features if they
4885 		 * would require internal reload of PF in bnx2x_set_features().
4886 		 */
4887 		if (!(features & NETIF_F_RXCSUM) && !bp->disable_tpa) {
4888 			features &= ~NETIF_F_RXCSUM;
4889 			features |= dev->features & NETIF_F_RXCSUM;
4890 		}
4891 
4892 		if (changed & NETIF_F_LOOPBACK) {
4893 			features &= ~NETIF_F_LOOPBACK;
4894 			features |= dev->features & NETIF_F_LOOPBACK;
4895 		}
4896 	}
4897 
4898 	/* TPA requires Rx CSUM offloading */
4899 	if (!(features & NETIF_F_RXCSUM)) {
4900 		features &= ~NETIF_F_LRO;
4901 		features &= ~NETIF_F_GRO;
4902 	}
4903 
4904 	return features;
4905 }
4906 
4907 int bnx2x_set_features(struct net_device *dev, netdev_features_t features)
4908 {
4909 	struct bnx2x *bp = netdev_priv(dev);
4910 	netdev_features_t changes = features ^ dev->features;
4911 	bool bnx2x_reload = false;
4912 	int rc;
4913 
4914 	/* VFs or non SRIOV PFs should be able to change loopback feature */
4915 	if (!pci_num_vf(bp->pdev)) {
4916 		if (features & NETIF_F_LOOPBACK) {
4917 			if (bp->link_params.loopback_mode != LOOPBACK_BMAC) {
4918 				bp->link_params.loopback_mode = LOOPBACK_BMAC;
4919 				bnx2x_reload = true;
4920 			}
4921 		} else {
4922 			if (bp->link_params.loopback_mode != LOOPBACK_NONE) {
4923 				bp->link_params.loopback_mode = LOOPBACK_NONE;
4924 				bnx2x_reload = true;
4925 			}
4926 		}
4927 	}
4928 
4929 	/* if GRO is changed while LRO is enabled, don't force a reload */
4930 	if ((changes & NETIF_F_GRO) && (features & NETIF_F_LRO))
4931 		changes &= ~NETIF_F_GRO;
4932 
4933 	/* if GRO is changed while HW TPA is off, don't force a reload */
4934 	if ((changes & NETIF_F_GRO) && bp->disable_tpa)
4935 		changes &= ~NETIF_F_GRO;
4936 
4937 	if (changes)
4938 		bnx2x_reload = true;
4939 
4940 	if (bnx2x_reload) {
4941 		if (bp->recovery_state == BNX2X_RECOVERY_DONE) {
4942 			dev->features = features;
4943 			rc = bnx2x_reload_if_running(dev);
4944 			return rc ? rc : 1;
4945 		}
4946 		/* else: bnx2x_nic_load() will be called at end of recovery */
4947 	}
4948 
4949 	return 0;
4950 }
4951 
4952 void bnx2x_tx_timeout(struct net_device *dev)
4953 {
4954 	struct bnx2x *bp = netdev_priv(dev);
4955 
4956 #ifdef BNX2X_STOP_ON_ERROR
4957 	if (!bp->panic)
4958 		bnx2x_panic();
4959 #endif
4960 
4961 	/* This allows the netif to be shutdown gracefully before resetting */
4962 	bnx2x_schedule_sp_rtnl(bp, BNX2X_SP_RTNL_TX_TIMEOUT, 0);
4963 }
4964 
4965 int bnx2x_suspend(struct pci_dev *pdev, pm_message_t state)
4966 {
4967 	struct net_device *dev = pci_get_drvdata(pdev);
4968 	struct bnx2x *bp;
4969 
4970 	if (!dev) {
4971 		dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
4972 		return -ENODEV;
4973 	}
4974 	bp = netdev_priv(dev);
4975 
4976 	rtnl_lock();
4977 
4978 	pci_save_state(pdev);
4979 
4980 	if (!netif_running(dev)) {
4981 		rtnl_unlock();
4982 		return 0;
4983 	}
4984 
4985 	netif_device_detach(dev);
4986 
4987 	bnx2x_nic_unload(bp, UNLOAD_CLOSE, false);
4988 
4989 	bnx2x_set_power_state(bp, pci_choose_state(pdev, state));
4990 
4991 	rtnl_unlock();
4992 
4993 	return 0;
4994 }
4995 
4996 int bnx2x_resume(struct pci_dev *pdev)
4997 {
4998 	struct net_device *dev = pci_get_drvdata(pdev);
4999 	struct bnx2x *bp;
5000 	int rc;
5001 
5002 	if (!dev) {
5003 		dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
5004 		return -ENODEV;
5005 	}
5006 	bp = netdev_priv(dev);
5007 
5008 	if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
5009 		BNX2X_ERR("Handling parity error recovery. Try again later\n");
5010 		return -EAGAIN;
5011 	}
5012 
5013 	rtnl_lock();
5014 
5015 	pci_restore_state(pdev);
5016 
5017 	if (!netif_running(dev)) {
5018 		rtnl_unlock();
5019 		return 0;
5020 	}
5021 
5022 	bnx2x_set_power_state(bp, PCI_D0);
5023 	netif_device_attach(dev);
5024 
5025 	rc = bnx2x_nic_load(bp, LOAD_OPEN);
5026 
5027 	rtnl_unlock();
5028 
5029 	return rc;
5030 }
5031 
5032 void bnx2x_set_ctx_validation(struct bnx2x *bp, struct eth_context *cxt,
5033 			      u32 cid)
5034 {
5035 	if (!cxt) {
5036 		BNX2X_ERR("bad context pointer %p\n", cxt);
5037 		return;
5038 	}
5039 
5040 	/* ustorm cxt validation */
5041 	cxt->ustorm_ag_context.cdu_usage =
5042 		CDU_RSRVD_VALUE_TYPE_A(HW_CID(bp, cid),
5043 			CDU_REGION_NUMBER_UCM_AG, ETH_CONNECTION_TYPE);
5044 	/* xcontext validation */
5045 	cxt->xstorm_ag_context.cdu_reserved =
5046 		CDU_RSRVD_VALUE_TYPE_A(HW_CID(bp, cid),
5047 			CDU_REGION_NUMBER_XCM_AG, ETH_CONNECTION_TYPE);
5048 }
5049 
5050 static void storm_memset_hc_timeout(struct bnx2x *bp, u8 port,
5051 				    u8 fw_sb_id, u8 sb_index,
5052 				    u8 ticks)
5053 {
5054 	u32 addr = BAR_CSTRORM_INTMEM +
5055 		   CSTORM_STATUS_BLOCK_DATA_TIMEOUT_OFFSET(fw_sb_id, sb_index);
5056 	REG_WR8(bp, addr, ticks);
5057 	DP(NETIF_MSG_IFUP,
5058 	   "port %x fw_sb_id %d sb_index %d ticks %d\n",
5059 	   port, fw_sb_id, sb_index, ticks);
5060 }
5061 
5062 static void storm_memset_hc_disable(struct bnx2x *bp, u8 port,
5063 				    u16 fw_sb_id, u8 sb_index,
5064 				    u8 disable)
5065 {
5066 	u32 enable_flag = disable ? 0 : (1 << HC_INDEX_DATA_HC_ENABLED_SHIFT);
5067 	u32 addr = BAR_CSTRORM_INTMEM +
5068 		   CSTORM_STATUS_BLOCK_DATA_FLAGS_OFFSET(fw_sb_id, sb_index);
5069 	u8 flags = REG_RD8(bp, addr);
5070 	/* clear and set */
5071 	flags &= ~HC_INDEX_DATA_HC_ENABLED;
5072 	flags |= enable_flag;
5073 	REG_WR8(bp, addr, flags);
5074 	DP(NETIF_MSG_IFUP,
5075 	   "port %x fw_sb_id %d sb_index %d disable %d\n",
5076 	   port, fw_sb_id, sb_index, disable);
5077 }
5078 
5079 void bnx2x_update_coalesce_sb_index(struct bnx2x *bp, u8 fw_sb_id,
5080 				    u8 sb_index, u8 disable, u16 usec)
5081 {
5082 	int port = BP_PORT(bp);
5083 	u8 ticks = usec / BNX2X_BTR;
5084 
5085 	storm_memset_hc_timeout(bp, port, fw_sb_id, sb_index, ticks);
5086 
5087 	disable = disable ? 1 : (usec ? 0 : 1);
5088 	storm_memset_hc_disable(bp, port, fw_sb_id, sb_index, disable);
5089 }
5090 
5091 void bnx2x_schedule_sp_rtnl(struct bnx2x *bp, enum sp_rtnl_flag flag,
5092 			    u32 verbose)
5093 {
5094 	smp_mb__before_atomic();
5095 	set_bit(flag, &bp->sp_rtnl_state);
5096 	smp_mb__after_atomic();
5097 	DP((BNX2X_MSG_SP | verbose), "Scheduling sp_rtnl task [Flag: %d]\n",
5098 	   flag);
5099 	schedule_delayed_work(&bp->sp_rtnl_task, 0);
5100 }
5101