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) {
553 		pool->page = alloc_pages(gfp_mask, PAGES_PER_SGE_SHIFT);
554 		if (unlikely(!pool->page))
555 			return -ENOMEM;
556 
557 		pool->offset = 0;
558 	}
559 
560 	mapping = dma_map_page(&bp->pdev->dev, pool->page,
561 			       pool->offset, SGE_PAGE_SIZE, DMA_FROM_DEVICE);
562 	if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
563 		BNX2X_ERR("Can't map sge\n");
564 		return -ENOMEM;
565 	}
566 
567 	sw_buf->page = pool->page;
568 	sw_buf->offset = pool->offset;
569 
570 	dma_unmap_addr_set(sw_buf, mapping, mapping);
571 
572 	sge->addr_hi = cpu_to_le32(U64_HI(mapping));
573 	sge->addr_lo = cpu_to_le32(U64_LO(mapping));
574 
575 	pool->offset += SGE_PAGE_SIZE;
576 	if (PAGE_SIZE - pool->offset >= SGE_PAGE_SIZE)
577 		get_page(pool->page);
578 	else
579 		pool->page = NULL;
580 	return 0;
581 }
582 
583 static int bnx2x_fill_frag_skb(struct bnx2x *bp, struct bnx2x_fastpath *fp,
584 			       struct bnx2x_agg_info *tpa_info,
585 			       u16 pages,
586 			       struct sk_buff *skb,
587 			       struct eth_end_agg_rx_cqe *cqe,
588 			       u16 cqe_idx)
589 {
590 	struct sw_rx_page *rx_pg, old_rx_pg;
591 	u32 i, frag_len, frag_size;
592 	int err, j, frag_id = 0;
593 	u16 len_on_bd = tpa_info->len_on_bd;
594 	u16 full_page = 0, gro_size = 0;
595 
596 	frag_size = le16_to_cpu(cqe->pkt_len) - len_on_bd;
597 
598 	if (fp->mode == TPA_MODE_GRO) {
599 		gro_size = tpa_info->gro_size;
600 		full_page = tpa_info->full_page;
601 	}
602 
603 	/* This is needed in order to enable forwarding support */
604 	if (frag_size)
605 		bnx2x_set_gro_params(skb, tpa_info->parsing_flags, len_on_bd,
606 				     le16_to_cpu(cqe->pkt_len),
607 				     le16_to_cpu(cqe->num_of_coalesced_segs));
608 
609 #ifdef BNX2X_STOP_ON_ERROR
610 	if (pages > min_t(u32, 8, MAX_SKB_FRAGS) * SGE_PAGES) {
611 		BNX2X_ERR("SGL length is too long: %d. CQE index is %d\n",
612 			  pages, cqe_idx);
613 		BNX2X_ERR("cqe->pkt_len = %d\n", cqe->pkt_len);
614 		bnx2x_panic();
615 		return -EINVAL;
616 	}
617 #endif
618 
619 	/* Run through the SGL and compose the fragmented skb */
620 	for (i = 0, j = 0; i < pages; i += PAGES_PER_SGE, j++) {
621 		u16 sge_idx = RX_SGE(le16_to_cpu(cqe->sgl_or_raw_data.sgl[j]));
622 
623 		/* FW gives the indices of the SGE as if the ring is an array
624 		   (meaning that "next" element will consume 2 indices) */
625 		if (fp->mode == TPA_MODE_GRO)
626 			frag_len = min_t(u32, frag_size, (u32)full_page);
627 		else /* LRO */
628 			frag_len = min_t(u32, frag_size, (u32)SGE_PAGES);
629 
630 		rx_pg = &fp->rx_page_ring[sge_idx];
631 		old_rx_pg = *rx_pg;
632 
633 		/* If we fail to allocate a substitute page, we simply stop
634 		   where we are and drop the whole packet */
635 		err = bnx2x_alloc_rx_sge(bp, fp, sge_idx, GFP_ATOMIC);
636 		if (unlikely(err)) {
637 			bnx2x_fp_qstats(bp, fp)->rx_skb_alloc_failed++;
638 			return err;
639 		}
640 
641 		dma_unmap_page(&bp->pdev->dev,
642 			       dma_unmap_addr(&old_rx_pg, mapping),
643 			       SGE_PAGE_SIZE, DMA_FROM_DEVICE);
644 		/* Add one frag and update the appropriate fields in the skb */
645 		if (fp->mode == TPA_MODE_LRO)
646 			skb_fill_page_desc(skb, j, old_rx_pg.page,
647 					   old_rx_pg.offset, frag_len);
648 		else { /* GRO */
649 			int rem;
650 			int offset = 0;
651 			for (rem = frag_len; rem > 0; rem -= gro_size) {
652 				int len = rem > gro_size ? gro_size : rem;
653 				skb_fill_page_desc(skb, frag_id++,
654 						   old_rx_pg.page,
655 						   old_rx_pg.offset + offset,
656 						   len);
657 				if (offset)
658 					get_page(old_rx_pg.page);
659 				offset += len;
660 			}
661 		}
662 
663 		skb->data_len += frag_len;
664 		skb->truesize += SGE_PAGES;
665 		skb->len += frag_len;
666 
667 		frag_size -= frag_len;
668 	}
669 
670 	return 0;
671 }
672 
673 static void bnx2x_frag_free(const struct bnx2x_fastpath *fp, void *data)
674 {
675 	if (fp->rx_frag_size)
676 		skb_free_frag(data);
677 	else
678 		kfree(data);
679 }
680 
681 static void *bnx2x_frag_alloc(const struct bnx2x_fastpath *fp, gfp_t gfp_mask)
682 {
683 	if (fp->rx_frag_size) {
684 		/* GFP_KERNEL allocations are used only during initialization */
685 		if (unlikely(gfpflags_allow_blocking(gfp_mask)))
686 			return (void *)__get_free_page(gfp_mask);
687 
688 		return netdev_alloc_frag(fp->rx_frag_size);
689 	}
690 
691 	return kmalloc(fp->rx_buf_size + NET_SKB_PAD, gfp_mask);
692 }
693 
694 #ifdef CONFIG_INET
695 static void bnx2x_gro_ip_csum(struct bnx2x *bp, struct sk_buff *skb)
696 {
697 	const struct iphdr *iph = ip_hdr(skb);
698 	struct tcphdr *th;
699 
700 	skb_set_transport_header(skb, sizeof(struct iphdr));
701 	th = tcp_hdr(skb);
702 
703 	th->check = ~tcp_v4_check(skb->len - skb_transport_offset(skb),
704 				  iph->saddr, iph->daddr, 0);
705 }
706 
707 static void bnx2x_gro_ipv6_csum(struct bnx2x *bp, struct sk_buff *skb)
708 {
709 	struct ipv6hdr *iph = ipv6_hdr(skb);
710 	struct tcphdr *th;
711 
712 	skb_set_transport_header(skb, sizeof(struct ipv6hdr));
713 	th = tcp_hdr(skb);
714 
715 	th->check = ~tcp_v6_check(skb->len - skb_transport_offset(skb),
716 				  &iph->saddr, &iph->daddr, 0);
717 }
718 
719 static void bnx2x_gro_csum(struct bnx2x *bp, struct sk_buff *skb,
720 			    void (*gro_func)(struct bnx2x*, struct sk_buff*))
721 {
722 	skb_reset_network_header(skb);
723 	gro_func(bp, skb);
724 	tcp_gro_complete(skb);
725 }
726 #endif
727 
728 static void bnx2x_gro_receive(struct bnx2x *bp, struct bnx2x_fastpath *fp,
729 			       struct sk_buff *skb)
730 {
731 #ifdef CONFIG_INET
732 	if (skb_shinfo(skb)->gso_size) {
733 		switch (be16_to_cpu(skb->protocol)) {
734 		case ETH_P_IP:
735 			bnx2x_gro_csum(bp, skb, bnx2x_gro_ip_csum);
736 			break;
737 		case ETH_P_IPV6:
738 			bnx2x_gro_csum(bp, skb, bnx2x_gro_ipv6_csum);
739 			break;
740 		default:
741 			netdev_WARN_ONCE(bp->dev,
742 					 "Error: FW GRO supports only IPv4/IPv6, not 0x%04x\n",
743 					 be16_to_cpu(skb->protocol));
744 		}
745 	}
746 #endif
747 	skb_record_rx_queue(skb, fp->rx_queue);
748 	napi_gro_receive(&fp->napi, skb);
749 }
750 
751 static void bnx2x_tpa_stop(struct bnx2x *bp, struct bnx2x_fastpath *fp,
752 			   struct bnx2x_agg_info *tpa_info,
753 			   u16 pages,
754 			   struct eth_end_agg_rx_cqe *cqe,
755 			   u16 cqe_idx)
756 {
757 	struct sw_rx_bd *rx_buf = &tpa_info->first_buf;
758 	u8 pad = tpa_info->placement_offset;
759 	u16 len = tpa_info->len_on_bd;
760 	struct sk_buff *skb = NULL;
761 	u8 *new_data, *data = rx_buf->data;
762 	u8 old_tpa_state = tpa_info->tpa_state;
763 
764 	tpa_info->tpa_state = BNX2X_TPA_STOP;
765 
766 	/* If we there was an error during the handling of the TPA_START -
767 	 * drop this aggregation.
768 	 */
769 	if (old_tpa_state == BNX2X_TPA_ERROR)
770 		goto drop;
771 
772 	/* Try to allocate the new data */
773 	new_data = bnx2x_frag_alloc(fp, GFP_ATOMIC);
774 	/* Unmap skb in the pool anyway, as we are going to change
775 	   pool entry status to BNX2X_TPA_STOP even if new skb allocation
776 	   fails. */
777 	dma_unmap_single(&bp->pdev->dev, dma_unmap_addr(rx_buf, mapping),
778 			 fp->rx_buf_size, DMA_FROM_DEVICE);
779 	if (likely(new_data))
780 		skb = build_skb(data, fp->rx_frag_size);
781 
782 	if (likely(skb)) {
783 #ifdef BNX2X_STOP_ON_ERROR
784 		if (pad + len > fp->rx_buf_size) {
785 			BNX2X_ERR("skb_put is about to fail...  pad %d  len %d  rx_buf_size %d\n",
786 				  pad, len, fp->rx_buf_size);
787 			bnx2x_panic();
788 			return;
789 		}
790 #endif
791 
792 		skb_reserve(skb, pad + NET_SKB_PAD);
793 		skb_put(skb, len);
794 		skb_set_hash(skb, tpa_info->rxhash, tpa_info->rxhash_type);
795 
796 		skb->protocol = eth_type_trans(skb, bp->dev);
797 		skb->ip_summed = CHECKSUM_UNNECESSARY;
798 
799 		if (!bnx2x_fill_frag_skb(bp, fp, tpa_info, pages,
800 					 skb, cqe, cqe_idx)) {
801 			if (tpa_info->parsing_flags & PARSING_FLAGS_VLAN)
802 				__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), tpa_info->vlan_tag);
803 			bnx2x_gro_receive(bp, fp, skb);
804 		} else {
805 			DP(NETIF_MSG_RX_STATUS,
806 			   "Failed to allocate new pages - dropping packet!\n");
807 			dev_kfree_skb_any(skb);
808 		}
809 
810 		/* put new data in bin */
811 		rx_buf->data = new_data;
812 
813 		return;
814 	}
815 	if (new_data)
816 		bnx2x_frag_free(fp, new_data);
817 drop:
818 	/* drop the packet and keep the buffer in the bin */
819 	DP(NETIF_MSG_RX_STATUS,
820 	   "Failed to allocate or map a new skb - dropping packet!\n");
821 	bnx2x_fp_stats(bp, fp)->eth_q_stats.rx_skb_alloc_failed++;
822 }
823 
824 static int bnx2x_alloc_rx_data(struct bnx2x *bp, struct bnx2x_fastpath *fp,
825 			       u16 index, gfp_t gfp_mask)
826 {
827 	u8 *data;
828 	struct sw_rx_bd *rx_buf = &fp->rx_buf_ring[index];
829 	struct eth_rx_bd *rx_bd = &fp->rx_desc_ring[index];
830 	dma_addr_t mapping;
831 
832 	data = bnx2x_frag_alloc(fp, gfp_mask);
833 	if (unlikely(data == NULL))
834 		return -ENOMEM;
835 
836 	mapping = dma_map_single(&bp->pdev->dev, data + NET_SKB_PAD,
837 				 fp->rx_buf_size,
838 				 DMA_FROM_DEVICE);
839 	if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
840 		bnx2x_frag_free(fp, data);
841 		BNX2X_ERR("Can't map rx data\n");
842 		return -ENOMEM;
843 	}
844 
845 	rx_buf->data = data;
846 	dma_unmap_addr_set(rx_buf, mapping, mapping);
847 
848 	rx_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
849 	rx_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
850 
851 	return 0;
852 }
853 
854 static
855 void bnx2x_csum_validate(struct sk_buff *skb, union eth_rx_cqe *cqe,
856 				 struct bnx2x_fastpath *fp,
857 				 struct bnx2x_eth_q_stats *qstats)
858 {
859 	/* Do nothing if no L4 csum validation was done.
860 	 * We do not check whether IP csum was validated. For IPv4 we assume
861 	 * that if the card got as far as validating the L4 csum, it also
862 	 * validated the IP csum. IPv6 has no IP csum.
863 	 */
864 	if (cqe->fast_path_cqe.status_flags &
865 	    ETH_FAST_PATH_RX_CQE_L4_XSUM_NO_VALIDATION_FLG)
866 		return;
867 
868 	/* If L4 validation was done, check if an error was found. */
869 
870 	if (cqe->fast_path_cqe.type_error_flags &
871 	    (ETH_FAST_PATH_RX_CQE_IP_BAD_XSUM_FLG |
872 	     ETH_FAST_PATH_RX_CQE_L4_BAD_XSUM_FLG))
873 		qstats->hw_csum_err++;
874 	else
875 		skb->ip_summed = CHECKSUM_UNNECESSARY;
876 }
877 
878 static int bnx2x_rx_int(struct bnx2x_fastpath *fp, int budget)
879 {
880 	struct bnx2x *bp = fp->bp;
881 	u16 bd_cons, bd_prod, bd_prod_fw, comp_ring_cons;
882 	u16 sw_comp_cons, sw_comp_prod;
883 	int rx_pkt = 0;
884 	union eth_rx_cqe *cqe;
885 	struct eth_fast_path_rx_cqe *cqe_fp;
886 
887 #ifdef BNX2X_STOP_ON_ERROR
888 	if (unlikely(bp->panic))
889 		return 0;
890 #endif
891 	if (budget <= 0)
892 		return rx_pkt;
893 
894 	bd_cons = fp->rx_bd_cons;
895 	bd_prod = fp->rx_bd_prod;
896 	bd_prod_fw = bd_prod;
897 	sw_comp_cons = fp->rx_comp_cons;
898 	sw_comp_prod = fp->rx_comp_prod;
899 
900 	comp_ring_cons = RCQ_BD(sw_comp_cons);
901 	cqe = &fp->rx_comp_ring[comp_ring_cons];
902 	cqe_fp = &cqe->fast_path_cqe;
903 
904 	DP(NETIF_MSG_RX_STATUS,
905 	   "queue[%d]: sw_comp_cons %u\n", fp->index, sw_comp_cons);
906 
907 	while (BNX2X_IS_CQE_COMPLETED(cqe_fp)) {
908 		struct sw_rx_bd *rx_buf = NULL;
909 		struct sk_buff *skb;
910 		u8 cqe_fp_flags;
911 		enum eth_rx_cqe_type cqe_fp_type;
912 		u16 len, pad, queue;
913 		u8 *data;
914 		u32 rxhash;
915 		enum pkt_hash_types rxhash_type;
916 
917 #ifdef BNX2X_STOP_ON_ERROR
918 		if (unlikely(bp->panic))
919 			return 0;
920 #endif
921 
922 		bd_prod = RX_BD(bd_prod);
923 		bd_cons = RX_BD(bd_cons);
924 
925 		/* A rmb() is required to ensure that the CQE is not read
926 		 * before it is written by the adapter DMA.  PCI ordering
927 		 * rules will make sure the other fields are written before
928 		 * the marker at the end of struct eth_fast_path_rx_cqe
929 		 * but without rmb() a weakly ordered processor can process
930 		 * stale data.  Without the barrier TPA state-machine might
931 		 * enter inconsistent state and kernel stack might be
932 		 * provided with incorrect packet description - these lead
933 		 * to various kernel crashed.
934 		 */
935 		rmb();
936 
937 		cqe_fp_flags = cqe_fp->type_error_flags;
938 		cqe_fp_type = cqe_fp_flags & ETH_FAST_PATH_RX_CQE_TYPE;
939 
940 		DP(NETIF_MSG_RX_STATUS,
941 		   "CQE type %x  err %x  status %x  queue %x  vlan %x  len %u\n",
942 		   CQE_TYPE(cqe_fp_flags),
943 		   cqe_fp_flags, cqe_fp->status_flags,
944 		   le32_to_cpu(cqe_fp->rss_hash_result),
945 		   le16_to_cpu(cqe_fp->vlan_tag),
946 		   le16_to_cpu(cqe_fp->pkt_len_or_gro_seg_len));
947 
948 		/* is this a slowpath msg? */
949 		if (unlikely(CQE_TYPE_SLOW(cqe_fp_type))) {
950 			bnx2x_sp_event(fp, cqe);
951 			goto next_cqe;
952 		}
953 
954 		rx_buf = &fp->rx_buf_ring[bd_cons];
955 		data = rx_buf->data;
956 
957 		if (!CQE_TYPE_FAST(cqe_fp_type)) {
958 			struct bnx2x_agg_info *tpa_info;
959 			u16 frag_size, pages;
960 #ifdef BNX2X_STOP_ON_ERROR
961 			/* sanity check */
962 			if (fp->mode == TPA_MODE_DISABLED &&
963 			    (CQE_TYPE_START(cqe_fp_type) ||
964 			     CQE_TYPE_STOP(cqe_fp_type)))
965 				BNX2X_ERR("START/STOP packet while TPA disabled, type %x\n",
966 					  CQE_TYPE(cqe_fp_type));
967 #endif
968 
969 			if (CQE_TYPE_START(cqe_fp_type)) {
970 				u16 queue = cqe_fp->queue_index;
971 				DP(NETIF_MSG_RX_STATUS,
972 				   "calling tpa_start on queue %d\n",
973 				   queue);
974 
975 				bnx2x_tpa_start(fp, queue,
976 						bd_cons, bd_prod,
977 						cqe_fp);
978 
979 				goto next_rx;
980 			}
981 			queue = cqe->end_agg_cqe.queue_index;
982 			tpa_info = &fp->tpa_info[queue];
983 			DP(NETIF_MSG_RX_STATUS,
984 			   "calling tpa_stop on queue %d\n",
985 			   queue);
986 
987 			frag_size = le16_to_cpu(cqe->end_agg_cqe.pkt_len) -
988 				    tpa_info->len_on_bd;
989 
990 			if (fp->mode == TPA_MODE_GRO)
991 				pages = (frag_size + tpa_info->full_page - 1) /
992 					 tpa_info->full_page;
993 			else
994 				pages = SGE_PAGE_ALIGN(frag_size) >>
995 					SGE_PAGE_SHIFT;
996 
997 			bnx2x_tpa_stop(bp, fp, tpa_info, pages,
998 				       &cqe->end_agg_cqe, comp_ring_cons);
999 #ifdef BNX2X_STOP_ON_ERROR
1000 			if (bp->panic)
1001 				return 0;
1002 #endif
1003 
1004 			bnx2x_update_sge_prod(fp, pages, &cqe->end_agg_cqe);
1005 			goto next_cqe;
1006 		}
1007 		/* non TPA */
1008 		len = le16_to_cpu(cqe_fp->pkt_len_or_gro_seg_len);
1009 		pad = cqe_fp->placement_offset;
1010 		dma_sync_single_for_cpu(&bp->pdev->dev,
1011 					dma_unmap_addr(rx_buf, mapping),
1012 					pad + RX_COPY_THRESH,
1013 					DMA_FROM_DEVICE);
1014 		pad += NET_SKB_PAD;
1015 		prefetch(data + pad); /* speedup eth_type_trans() */
1016 		/* is this an error packet? */
1017 		if (unlikely(cqe_fp_flags & ETH_RX_ERROR_FALGS)) {
1018 			DP(NETIF_MSG_RX_ERR | NETIF_MSG_RX_STATUS,
1019 			   "ERROR  flags %x  rx packet %u\n",
1020 			   cqe_fp_flags, sw_comp_cons);
1021 			bnx2x_fp_qstats(bp, fp)->rx_err_discard_pkt++;
1022 			goto reuse_rx;
1023 		}
1024 
1025 		/* Since we don't have a jumbo ring
1026 		 * copy small packets if mtu > 1500
1027 		 */
1028 		if ((bp->dev->mtu > ETH_MAX_PACKET_SIZE) &&
1029 		    (len <= RX_COPY_THRESH)) {
1030 			skb = napi_alloc_skb(&fp->napi, len);
1031 			if (skb == NULL) {
1032 				DP(NETIF_MSG_RX_ERR | NETIF_MSG_RX_STATUS,
1033 				   "ERROR  packet dropped because of alloc failure\n");
1034 				bnx2x_fp_qstats(bp, fp)->rx_skb_alloc_failed++;
1035 				goto reuse_rx;
1036 			}
1037 			memcpy(skb->data, data + pad, len);
1038 			bnx2x_reuse_rx_data(fp, bd_cons, bd_prod);
1039 		} else {
1040 			if (likely(bnx2x_alloc_rx_data(bp, fp, bd_prod,
1041 						       GFP_ATOMIC) == 0)) {
1042 				dma_unmap_single(&bp->pdev->dev,
1043 						 dma_unmap_addr(rx_buf, mapping),
1044 						 fp->rx_buf_size,
1045 						 DMA_FROM_DEVICE);
1046 				skb = build_skb(data, fp->rx_frag_size);
1047 				if (unlikely(!skb)) {
1048 					bnx2x_frag_free(fp, data);
1049 					bnx2x_fp_qstats(bp, fp)->
1050 							rx_skb_alloc_failed++;
1051 					goto next_rx;
1052 				}
1053 				skb_reserve(skb, pad);
1054 			} else {
1055 				DP(NETIF_MSG_RX_ERR | NETIF_MSG_RX_STATUS,
1056 				   "ERROR  packet dropped because of alloc failure\n");
1057 				bnx2x_fp_qstats(bp, fp)->rx_skb_alloc_failed++;
1058 reuse_rx:
1059 				bnx2x_reuse_rx_data(fp, bd_cons, bd_prod);
1060 				goto next_rx;
1061 			}
1062 		}
1063 
1064 		skb_put(skb, len);
1065 		skb->protocol = eth_type_trans(skb, bp->dev);
1066 
1067 		/* Set Toeplitz hash for a none-LRO skb */
1068 		rxhash = bnx2x_get_rxhash(bp, cqe_fp, &rxhash_type);
1069 		skb_set_hash(skb, rxhash, rxhash_type);
1070 
1071 		skb_checksum_none_assert(skb);
1072 
1073 		if (bp->dev->features & NETIF_F_RXCSUM)
1074 			bnx2x_csum_validate(skb, cqe, fp,
1075 					    bnx2x_fp_qstats(bp, fp));
1076 
1077 		skb_record_rx_queue(skb, fp->rx_queue);
1078 
1079 		/* Check if this packet was timestamped */
1080 		if (unlikely(cqe->fast_path_cqe.type_error_flags &
1081 			     (1 << ETH_FAST_PATH_RX_CQE_PTP_PKT_SHIFT)))
1082 			bnx2x_set_rx_ts(bp, skb);
1083 
1084 		if (le16_to_cpu(cqe_fp->pars_flags.flags) &
1085 		    PARSING_FLAGS_VLAN)
1086 			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
1087 					       le16_to_cpu(cqe_fp->vlan_tag));
1088 
1089 		napi_gro_receive(&fp->napi, skb);
1090 next_rx:
1091 		rx_buf->data = NULL;
1092 
1093 		bd_cons = NEXT_RX_IDX(bd_cons);
1094 		bd_prod = NEXT_RX_IDX(bd_prod);
1095 		bd_prod_fw = NEXT_RX_IDX(bd_prod_fw);
1096 		rx_pkt++;
1097 next_cqe:
1098 		sw_comp_prod = NEXT_RCQ_IDX(sw_comp_prod);
1099 		sw_comp_cons = NEXT_RCQ_IDX(sw_comp_cons);
1100 
1101 		/* mark CQE as free */
1102 		BNX2X_SEED_CQE(cqe_fp);
1103 
1104 		if (rx_pkt == budget)
1105 			break;
1106 
1107 		comp_ring_cons = RCQ_BD(sw_comp_cons);
1108 		cqe = &fp->rx_comp_ring[comp_ring_cons];
1109 		cqe_fp = &cqe->fast_path_cqe;
1110 	} /* while */
1111 
1112 	fp->rx_bd_cons = bd_cons;
1113 	fp->rx_bd_prod = bd_prod_fw;
1114 	fp->rx_comp_cons = sw_comp_cons;
1115 	fp->rx_comp_prod = sw_comp_prod;
1116 
1117 	/* Update producers */
1118 	bnx2x_update_rx_prod(bp, fp, bd_prod_fw, sw_comp_prod,
1119 			     fp->rx_sge_prod);
1120 
1121 	return rx_pkt;
1122 }
1123 
1124 static irqreturn_t bnx2x_msix_fp_int(int irq, void *fp_cookie)
1125 {
1126 	struct bnx2x_fastpath *fp = fp_cookie;
1127 	struct bnx2x *bp = fp->bp;
1128 	u8 cos;
1129 
1130 	DP(NETIF_MSG_INTR,
1131 	   "got an MSI-X interrupt on IDX:SB [fp %d fw_sd %d igusb %d]\n",
1132 	   fp->index, fp->fw_sb_id, fp->igu_sb_id);
1133 
1134 	bnx2x_ack_sb(bp, fp->igu_sb_id, USTORM_ID, 0, IGU_INT_DISABLE, 0);
1135 
1136 #ifdef BNX2X_STOP_ON_ERROR
1137 	if (unlikely(bp->panic))
1138 		return IRQ_HANDLED;
1139 #endif
1140 
1141 	/* Handle Rx and Tx according to MSI-X vector */
1142 	for_each_cos_in_tx_queue(fp, cos)
1143 		prefetch(fp->txdata_ptr[cos]->tx_cons_sb);
1144 
1145 	prefetch(&fp->sb_running_index[SM_RX_ID]);
1146 	napi_schedule_irqoff(&bnx2x_fp(bp, fp->index, napi));
1147 
1148 	return IRQ_HANDLED;
1149 }
1150 
1151 /* HW Lock for shared dual port PHYs */
1152 void bnx2x_acquire_phy_lock(struct bnx2x *bp)
1153 {
1154 	mutex_lock(&bp->port.phy_mutex);
1155 
1156 	bnx2x_acquire_hw_lock(bp, HW_LOCK_RESOURCE_MDIO);
1157 }
1158 
1159 void bnx2x_release_phy_lock(struct bnx2x *bp)
1160 {
1161 	bnx2x_release_hw_lock(bp, HW_LOCK_RESOURCE_MDIO);
1162 
1163 	mutex_unlock(&bp->port.phy_mutex);
1164 }
1165 
1166 /* calculates MF speed according to current linespeed and MF configuration */
1167 u16 bnx2x_get_mf_speed(struct bnx2x *bp)
1168 {
1169 	u16 line_speed = bp->link_vars.line_speed;
1170 	if (IS_MF(bp)) {
1171 		u16 maxCfg = bnx2x_extract_max_cfg(bp,
1172 						   bp->mf_config[BP_VN(bp)]);
1173 
1174 		/* Calculate the current MAX line speed limit for the MF
1175 		 * devices
1176 		 */
1177 		if (IS_MF_PERCENT_BW(bp))
1178 			line_speed = (line_speed * maxCfg) / 100;
1179 		else { /* SD mode */
1180 			u16 vn_max_rate = maxCfg * 100;
1181 
1182 			if (vn_max_rate < line_speed)
1183 				line_speed = vn_max_rate;
1184 		}
1185 	}
1186 
1187 	return line_speed;
1188 }
1189 
1190 /**
1191  * bnx2x_fill_report_data - fill link report data to report
1192  *
1193  * @bp:		driver handle
1194  * @data:	link state to update
1195  *
1196  * It uses a none-atomic bit operations because is called under the mutex.
1197  */
1198 static void bnx2x_fill_report_data(struct bnx2x *bp,
1199 				   struct bnx2x_link_report_data *data)
1200 {
1201 	memset(data, 0, sizeof(*data));
1202 
1203 	if (IS_PF(bp)) {
1204 		/* Fill the report data: effective line speed */
1205 		data->line_speed = bnx2x_get_mf_speed(bp);
1206 
1207 		/* Link is down */
1208 		if (!bp->link_vars.link_up || (bp->flags & MF_FUNC_DIS))
1209 			__set_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1210 				  &data->link_report_flags);
1211 
1212 		if (!BNX2X_NUM_ETH_QUEUES(bp))
1213 			__set_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1214 				  &data->link_report_flags);
1215 
1216 		/* Full DUPLEX */
1217 		if (bp->link_vars.duplex == DUPLEX_FULL)
1218 			__set_bit(BNX2X_LINK_REPORT_FD,
1219 				  &data->link_report_flags);
1220 
1221 		/* Rx Flow Control is ON */
1222 		if (bp->link_vars.flow_ctrl & BNX2X_FLOW_CTRL_RX)
1223 			__set_bit(BNX2X_LINK_REPORT_RX_FC_ON,
1224 				  &data->link_report_flags);
1225 
1226 		/* Tx Flow Control is ON */
1227 		if (bp->link_vars.flow_ctrl & BNX2X_FLOW_CTRL_TX)
1228 			__set_bit(BNX2X_LINK_REPORT_TX_FC_ON,
1229 				  &data->link_report_flags);
1230 	} else { /* VF */
1231 		*data = bp->vf_link_vars;
1232 	}
1233 }
1234 
1235 /**
1236  * bnx2x_link_report - report link status to OS.
1237  *
1238  * @bp:		driver handle
1239  *
1240  * Calls the __bnx2x_link_report() under the same locking scheme
1241  * as a link/PHY state managing code to ensure a consistent link
1242  * reporting.
1243  */
1244 
1245 void bnx2x_link_report(struct bnx2x *bp)
1246 {
1247 	bnx2x_acquire_phy_lock(bp);
1248 	__bnx2x_link_report(bp);
1249 	bnx2x_release_phy_lock(bp);
1250 }
1251 
1252 /**
1253  * __bnx2x_link_report - report link status to OS.
1254  *
1255  * @bp:		driver handle
1256  *
1257  * None atomic implementation.
1258  * Should be called under the phy_lock.
1259  */
1260 void __bnx2x_link_report(struct bnx2x *bp)
1261 {
1262 	struct bnx2x_link_report_data cur_data;
1263 
1264 	if (bp->force_link_down) {
1265 		bp->link_vars.link_up = 0;
1266 		return;
1267 	}
1268 
1269 	/* reread mf_cfg */
1270 	if (IS_PF(bp) && !CHIP_IS_E1(bp))
1271 		bnx2x_read_mf_cfg(bp);
1272 
1273 	/* Read the current link report info */
1274 	bnx2x_fill_report_data(bp, &cur_data);
1275 
1276 	/* Don't report link down or exactly the same link status twice */
1277 	if (!memcmp(&cur_data, &bp->last_reported_link, sizeof(cur_data)) ||
1278 	    (test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1279 		      &bp->last_reported_link.link_report_flags) &&
1280 	     test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1281 		      &cur_data.link_report_flags)))
1282 		return;
1283 
1284 	bp->link_cnt++;
1285 
1286 	/* We are going to report a new link parameters now -
1287 	 * remember the current data for the next time.
1288 	 */
1289 	memcpy(&bp->last_reported_link, &cur_data, sizeof(cur_data));
1290 
1291 	/* propagate status to VFs */
1292 	if (IS_PF(bp))
1293 		bnx2x_iov_link_update(bp);
1294 
1295 	if (test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1296 		     &cur_data.link_report_flags)) {
1297 		netif_carrier_off(bp->dev);
1298 		netdev_err(bp->dev, "NIC Link is Down\n");
1299 		return;
1300 	} else {
1301 		const char *duplex;
1302 		const char *flow;
1303 
1304 		netif_carrier_on(bp->dev);
1305 
1306 		if (test_and_clear_bit(BNX2X_LINK_REPORT_FD,
1307 				       &cur_data.link_report_flags))
1308 			duplex = "full";
1309 		else
1310 			duplex = "half";
1311 
1312 		/* Handle the FC at the end so that only these flags would be
1313 		 * possibly set. This way we may easily check if there is no FC
1314 		 * enabled.
1315 		 */
1316 		if (cur_data.link_report_flags) {
1317 			if (test_bit(BNX2X_LINK_REPORT_RX_FC_ON,
1318 				     &cur_data.link_report_flags)) {
1319 				if (test_bit(BNX2X_LINK_REPORT_TX_FC_ON,
1320 				     &cur_data.link_report_flags))
1321 					flow = "ON - receive & transmit";
1322 				else
1323 					flow = "ON - receive";
1324 			} else {
1325 				flow = "ON - transmit";
1326 			}
1327 		} else {
1328 			flow = "none";
1329 		}
1330 		netdev_info(bp->dev, "NIC Link is Up, %d Mbps %s duplex, Flow control: %s\n",
1331 			    cur_data.line_speed, duplex, flow);
1332 	}
1333 }
1334 
1335 static void bnx2x_set_next_page_sgl(struct bnx2x_fastpath *fp)
1336 {
1337 	int i;
1338 
1339 	for (i = 1; i <= NUM_RX_SGE_PAGES; i++) {
1340 		struct eth_rx_sge *sge;
1341 
1342 		sge = &fp->rx_sge_ring[RX_SGE_CNT * i - 2];
1343 		sge->addr_hi =
1344 			cpu_to_le32(U64_HI(fp->rx_sge_mapping +
1345 			BCM_PAGE_SIZE*(i % NUM_RX_SGE_PAGES)));
1346 
1347 		sge->addr_lo =
1348 			cpu_to_le32(U64_LO(fp->rx_sge_mapping +
1349 			BCM_PAGE_SIZE*(i % NUM_RX_SGE_PAGES)));
1350 	}
1351 }
1352 
1353 static void bnx2x_free_tpa_pool(struct bnx2x *bp,
1354 				struct bnx2x_fastpath *fp, int last)
1355 {
1356 	int i;
1357 
1358 	for (i = 0; i < last; i++) {
1359 		struct bnx2x_agg_info *tpa_info = &fp->tpa_info[i];
1360 		struct sw_rx_bd *first_buf = &tpa_info->first_buf;
1361 		u8 *data = first_buf->data;
1362 
1363 		if (data == NULL) {
1364 			DP(NETIF_MSG_IFDOWN, "tpa bin %d empty on free\n", i);
1365 			continue;
1366 		}
1367 		if (tpa_info->tpa_state == BNX2X_TPA_START)
1368 			dma_unmap_single(&bp->pdev->dev,
1369 					 dma_unmap_addr(first_buf, mapping),
1370 					 fp->rx_buf_size, DMA_FROM_DEVICE);
1371 		bnx2x_frag_free(fp, data);
1372 		first_buf->data = NULL;
1373 	}
1374 }
1375 
1376 void bnx2x_init_rx_rings_cnic(struct bnx2x *bp)
1377 {
1378 	int j;
1379 
1380 	for_each_rx_queue_cnic(bp, j) {
1381 		struct bnx2x_fastpath *fp = &bp->fp[j];
1382 
1383 		fp->rx_bd_cons = 0;
1384 
1385 		/* Activate BD ring */
1386 		/* Warning!
1387 		 * this will generate an interrupt (to the TSTORM)
1388 		 * must only be done after chip is initialized
1389 		 */
1390 		bnx2x_update_rx_prod(bp, fp, fp->rx_bd_prod, fp->rx_comp_prod,
1391 				     fp->rx_sge_prod);
1392 	}
1393 }
1394 
1395 void bnx2x_init_rx_rings(struct bnx2x *bp)
1396 {
1397 	int func = BP_FUNC(bp);
1398 	u16 ring_prod;
1399 	int i, j;
1400 
1401 	/* Allocate TPA resources */
1402 	for_each_eth_queue(bp, j) {
1403 		struct bnx2x_fastpath *fp = &bp->fp[j];
1404 
1405 		DP(NETIF_MSG_IFUP,
1406 		   "mtu %d  rx_buf_size %d\n", bp->dev->mtu, fp->rx_buf_size);
1407 
1408 		if (fp->mode != TPA_MODE_DISABLED) {
1409 			/* Fill the per-aggregation pool */
1410 			for (i = 0; i < MAX_AGG_QS(bp); i++) {
1411 				struct bnx2x_agg_info *tpa_info =
1412 					&fp->tpa_info[i];
1413 				struct sw_rx_bd *first_buf =
1414 					&tpa_info->first_buf;
1415 
1416 				first_buf->data =
1417 					bnx2x_frag_alloc(fp, GFP_KERNEL);
1418 				if (!first_buf->data) {
1419 					BNX2X_ERR("Failed to allocate TPA skb pool for queue[%d] - disabling TPA on this queue!\n",
1420 						  j);
1421 					bnx2x_free_tpa_pool(bp, fp, i);
1422 					fp->mode = TPA_MODE_DISABLED;
1423 					break;
1424 				}
1425 				dma_unmap_addr_set(first_buf, mapping, 0);
1426 				tpa_info->tpa_state = BNX2X_TPA_STOP;
1427 			}
1428 
1429 			/* "next page" elements initialization */
1430 			bnx2x_set_next_page_sgl(fp);
1431 
1432 			/* set SGEs bit mask */
1433 			bnx2x_init_sge_ring_bit_mask(fp);
1434 
1435 			/* Allocate SGEs and initialize the ring elements */
1436 			for (i = 0, ring_prod = 0;
1437 			     i < MAX_RX_SGE_CNT*NUM_RX_SGE_PAGES; i++) {
1438 
1439 				if (bnx2x_alloc_rx_sge(bp, fp, ring_prod,
1440 						       GFP_KERNEL) < 0) {
1441 					BNX2X_ERR("was only able to allocate %d rx sges\n",
1442 						  i);
1443 					BNX2X_ERR("disabling TPA for queue[%d]\n",
1444 						  j);
1445 					/* Cleanup already allocated elements */
1446 					bnx2x_free_rx_sge_range(bp, fp,
1447 								ring_prod);
1448 					bnx2x_free_tpa_pool(bp, fp,
1449 							    MAX_AGG_QS(bp));
1450 					fp->mode = TPA_MODE_DISABLED;
1451 					ring_prod = 0;
1452 					break;
1453 				}
1454 				ring_prod = NEXT_SGE_IDX(ring_prod);
1455 			}
1456 
1457 			fp->rx_sge_prod = ring_prod;
1458 		}
1459 	}
1460 
1461 	for_each_eth_queue(bp, j) {
1462 		struct bnx2x_fastpath *fp = &bp->fp[j];
1463 
1464 		fp->rx_bd_cons = 0;
1465 
1466 		/* Activate BD ring */
1467 		/* Warning!
1468 		 * this will generate an interrupt (to the TSTORM)
1469 		 * must only be done after chip is initialized
1470 		 */
1471 		bnx2x_update_rx_prod(bp, fp, fp->rx_bd_prod, fp->rx_comp_prod,
1472 				     fp->rx_sge_prod);
1473 
1474 		if (j != 0)
1475 			continue;
1476 
1477 		if (CHIP_IS_E1(bp)) {
1478 			REG_WR(bp, BAR_USTRORM_INTMEM +
1479 			       USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func),
1480 			       U64_LO(fp->rx_comp_mapping));
1481 			REG_WR(bp, BAR_USTRORM_INTMEM +
1482 			       USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func) + 4,
1483 			       U64_HI(fp->rx_comp_mapping));
1484 		}
1485 	}
1486 }
1487 
1488 static void bnx2x_free_tx_skbs_queue(struct bnx2x_fastpath *fp)
1489 {
1490 	u8 cos;
1491 	struct bnx2x *bp = fp->bp;
1492 
1493 	for_each_cos_in_tx_queue(fp, cos) {
1494 		struct bnx2x_fp_txdata *txdata = fp->txdata_ptr[cos];
1495 		unsigned pkts_compl = 0, bytes_compl = 0;
1496 
1497 		u16 sw_prod = txdata->tx_pkt_prod;
1498 		u16 sw_cons = txdata->tx_pkt_cons;
1499 
1500 		while (sw_cons != sw_prod) {
1501 			bnx2x_free_tx_pkt(bp, txdata, TX_BD(sw_cons),
1502 					  &pkts_compl, &bytes_compl);
1503 			sw_cons++;
1504 		}
1505 
1506 		netdev_tx_reset_queue(
1507 			netdev_get_tx_queue(bp->dev,
1508 					    txdata->txq_index));
1509 	}
1510 }
1511 
1512 static void bnx2x_free_tx_skbs_cnic(struct bnx2x *bp)
1513 {
1514 	int i;
1515 
1516 	for_each_tx_queue_cnic(bp, i) {
1517 		bnx2x_free_tx_skbs_queue(&bp->fp[i]);
1518 	}
1519 }
1520 
1521 static void bnx2x_free_tx_skbs(struct bnx2x *bp)
1522 {
1523 	int i;
1524 
1525 	for_each_eth_queue(bp, i) {
1526 		bnx2x_free_tx_skbs_queue(&bp->fp[i]);
1527 	}
1528 }
1529 
1530 static void bnx2x_free_rx_bds(struct bnx2x_fastpath *fp)
1531 {
1532 	struct bnx2x *bp = fp->bp;
1533 	int i;
1534 
1535 	/* ring wasn't allocated */
1536 	if (fp->rx_buf_ring == NULL)
1537 		return;
1538 
1539 	for (i = 0; i < NUM_RX_BD; i++) {
1540 		struct sw_rx_bd *rx_buf = &fp->rx_buf_ring[i];
1541 		u8 *data = rx_buf->data;
1542 
1543 		if (data == NULL)
1544 			continue;
1545 		dma_unmap_single(&bp->pdev->dev,
1546 				 dma_unmap_addr(rx_buf, mapping),
1547 				 fp->rx_buf_size, DMA_FROM_DEVICE);
1548 
1549 		rx_buf->data = NULL;
1550 		bnx2x_frag_free(fp, data);
1551 	}
1552 }
1553 
1554 static void bnx2x_free_rx_skbs_cnic(struct bnx2x *bp)
1555 {
1556 	int j;
1557 
1558 	for_each_rx_queue_cnic(bp, j) {
1559 		bnx2x_free_rx_bds(&bp->fp[j]);
1560 	}
1561 }
1562 
1563 static void bnx2x_free_rx_skbs(struct bnx2x *bp)
1564 {
1565 	int j;
1566 
1567 	for_each_eth_queue(bp, j) {
1568 		struct bnx2x_fastpath *fp = &bp->fp[j];
1569 
1570 		bnx2x_free_rx_bds(fp);
1571 
1572 		if (fp->mode != TPA_MODE_DISABLED)
1573 			bnx2x_free_tpa_pool(bp, fp, MAX_AGG_QS(bp));
1574 	}
1575 }
1576 
1577 static void bnx2x_free_skbs_cnic(struct bnx2x *bp)
1578 {
1579 	bnx2x_free_tx_skbs_cnic(bp);
1580 	bnx2x_free_rx_skbs_cnic(bp);
1581 }
1582 
1583 void bnx2x_free_skbs(struct bnx2x *bp)
1584 {
1585 	bnx2x_free_tx_skbs(bp);
1586 	bnx2x_free_rx_skbs(bp);
1587 }
1588 
1589 void bnx2x_update_max_mf_config(struct bnx2x *bp, u32 value)
1590 {
1591 	/* load old values */
1592 	u32 mf_cfg = bp->mf_config[BP_VN(bp)];
1593 
1594 	if (value != bnx2x_extract_max_cfg(bp, mf_cfg)) {
1595 		/* leave all but MAX value */
1596 		mf_cfg &= ~FUNC_MF_CFG_MAX_BW_MASK;
1597 
1598 		/* set new MAX value */
1599 		mf_cfg |= (value << FUNC_MF_CFG_MAX_BW_SHIFT)
1600 				& FUNC_MF_CFG_MAX_BW_MASK;
1601 
1602 		bnx2x_fw_command(bp, DRV_MSG_CODE_SET_MF_BW, mf_cfg);
1603 	}
1604 }
1605 
1606 /**
1607  * bnx2x_free_msix_irqs - free previously requested MSI-X IRQ vectors
1608  *
1609  * @bp:		driver handle
1610  * @nvecs:	number of vectors to be released
1611  */
1612 static void bnx2x_free_msix_irqs(struct bnx2x *bp, int nvecs)
1613 {
1614 	int i, offset = 0;
1615 
1616 	if (nvecs == offset)
1617 		return;
1618 
1619 	/* VFs don't have a default SB */
1620 	if (IS_PF(bp)) {
1621 		free_irq(bp->msix_table[offset].vector, bp->dev);
1622 		DP(NETIF_MSG_IFDOWN, "released sp irq (%d)\n",
1623 		   bp->msix_table[offset].vector);
1624 		offset++;
1625 	}
1626 
1627 	if (CNIC_SUPPORT(bp)) {
1628 		if (nvecs == offset)
1629 			return;
1630 		offset++;
1631 	}
1632 
1633 	for_each_eth_queue(bp, i) {
1634 		if (nvecs == offset)
1635 			return;
1636 		DP(NETIF_MSG_IFDOWN, "about to release fp #%d->%d irq\n",
1637 		   i, bp->msix_table[offset].vector);
1638 
1639 		free_irq(bp->msix_table[offset++].vector, &bp->fp[i]);
1640 	}
1641 }
1642 
1643 void bnx2x_free_irq(struct bnx2x *bp)
1644 {
1645 	if (bp->flags & USING_MSIX_FLAG &&
1646 	    !(bp->flags & USING_SINGLE_MSIX_FLAG)) {
1647 		int nvecs = BNX2X_NUM_ETH_QUEUES(bp) + CNIC_SUPPORT(bp);
1648 
1649 		/* vfs don't have a default status block */
1650 		if (IS_PF(bp))
1651 			nvecs++;
1652 
1653 		bnx2x_free_msix_irqs(bp, nvecs);
1654 	} else {
1655 		free_irq(bp->dev->irq, bp->dev);
1656 	}
1657 }
1658 
1659 int bnx2x_enable_msix(struct bnx2x *bp)
1660 {
1661 	int msix_vec = 0, i, rc;
1662 
1663 	/* VFs don't have a default status block */
1664 	if (IS_PF(bp)) {
1665 		bp->msix_table[msix_vec].entry = msix_vec;
1666 		BNX2X_DEV_INFO("msix_table[0].entry = %d (slowpath)\n",
1667 			       bp->msix_table[0].entry);
1668 		msix_vec++;
1669 	}
1670 
1671 	/* Cnic requires an msix vector for itself */
1672 	if (CNIC_SUPPORT(bp)) {
1673 		bp->msix_table[msix_vec].entry = msix_vec;
1674 		BNX2X_DEV_INFO("msix_table[%d].entry = %d (CNIC)\n",
1675 			       msix_vec, bp->msix_table[msix_vec].entry);
1676 		msix_vec++;
1677 	}
1678 
1679 	/* We need separate vectors for ETH queues only (not FCoE) */
1680 	for_each_eth_queue(bp, i) {
1681 		bp->msix_table[msix_vec].entry = msix_vec;
1682 		BNX2X_DEV_INFO("msix_table[%d].entry = %d (fastpath #%u)\n",
1683 			       msix_vec, msix_vec, i);
1684 		msix_vec++;
1685 	}
1686 
1687 	DP(BNX2X_MSG_SP, "about to request enable msix with %d vectors\n",
1688 	   msix_vec);
1689 
1690 	rc = pci_enable_msix_range(bp->pdev, &bp->msix_table[0],
1691 				   BNX2X_MIN_MSIX_VEC_CNT(bp), msix_vec);
1692 	/*
1693 	 * reconfigure number of tx/rx queues according to available
1694 	 * MSI-X vectors
1695 	 */
1696 	if (rc == -ENOSPC) {
1697 		/* Get by with single vector */
1698 		rc = pci_enable_msix_range(bp->pdev, &bp->msix_table[0], 1, 1);
1699 		if (rc < 0) {
1700 			BNX2X_DEV_INFO("Single MSI-X is not attainable rc %d\n",
1701 				       rc);
1702 			goto no_msix;
1703 		}
1704 
1705 		BNX2X_DEV_INFO("Using single MSI-X vector\n");
1706 		bp->flags |= USING_SINGLE_MSIX_FLAG;
1707 
1708 		BNX2X_DEV_INFO("set number of queues to 1\n");
1709 		bp->num_ethernet_queues = 1;
1710 		bp->num_queues = bp->num_ethernet_queues + bp->num_cnic_queues;
1711 	} else if (rc < 0) {
1712 		BNX2X_DEV_INFO("MSI-X is not attainable rc %d\n", rc);
1713 		goto no_msix;
1714 	} else if (rc < msix_vec) {
1715 		/* how less vectors we will have? */
1716 		int diff = msix_vec - rc;
1717 
1718 		BNX2X_DEV_INFO("Trying to use less MSI-X vectors: %d\n", rc);
1719 
1720 		/*
1721 		 * decrease number of queues by number of unallocated entries
1722 		 */
1723 		bp->num_ethernet_queues -= diff;
1724 		bp->num_queues = bp->num_ethernet_queues + bp->num_cnic_queues;
1725 
1726 		BNX2X_DEV_INFO("New queue configuration set: %d\n",
1727 			       bp->num_queues);
1728 	}
1729 
1730 	bp->flags |= USING_MSIX_FLAG;
1731 
1732 	return 0;
1733 
1734 no_msix:
1735 	/* fall to INTx if not enough memory */
1736 	if (rc == -ENOMEM)
1737 		bp->flags |= DISABLE_MSI_FLAG;
1738 
1739 	return rc;
1740 }
1741 
1742 static int bnx2x_req_msix_irqs(struct bnx2x *bp)
1743 {
1744 	int i, rc, offset = 0;
1745 
1746 	/* no default status block for vf */
1747 	if (IS_PF(bp)) {
1748 		rc = request_irq(bp->msix_table[offset++].vector,
1749 				 bnx2x_msix_sp_int, 0,
1750 				 bp->dev->name, bp->dev);
1751 		if (rc) {
1752 			BNX2X_ERR("request sp irq failed\n");
1753 			return -EBUSY;
1754 		}
1755 	}
1756 
1757 	if (CNIC_SUPPORT(bp))
1758 		offset++;
1759 
1760 	for_each_eth_queue(bp, i) {
1761 		struct bnx2x_fastpath *fp = &bp->fp[i];
1762 		snprintf(fp->name, sizeof(fp->name), "%s-fp-%d",
1763 			 bp->dev->name, i);
1764 
1765 		rc = request_irq(bp->msix_table[offset].vector,
1766 				 bnx2x_msix_fp_int, 0, fp->name, fp);
1767 		if (rc) {
1768 			BNX2X_ERR("request fp #%d irq (%d) failed  rc %d\n", i,
1769 			      bp->msix_table[offset].vector, rc);
1770 			bnx2x_free_msix_irqs(bp, offset);
1771 			return -EBUSY;
1772 		}
1773 
1774 		offset++;
1775 	}
1776 
1777 	i = BNX2X_NUM_ETH_QUEUES(bp);
1778 	if (IS_PF(bp)) {
1779 		offset = 1 + CNIC_SUPPORT(bp);
1780 		netdev_info(bp->dev,
1781 			    "using MSI-X  IRQs: sp %d  fp[%d] %d ... fp[%d] %d\n",
1782 			    bp->msix_table[0].vector,
1783 			    0, bp->msix_table[offset].vector,
1784 			    i - 1, bp->msix_table[offset + i - 1].vector);
1785 	} else {
1786 		offset = CNIC_SUPPORT(bp);
1787 		netdev_info(bp->dev,
1788 			    "using MSI-X  IRQs: fp[%d] %d ... fp[%d] %d\n",
1789 			    0, bp->msix_table[offset].vector,
1790 			    i - 1, bp->msix_table[offset + i - 1].vector);
1791 	}
1792 	return 0;
1793 }
1794 
1795 int bnx2x_enable_msi(struct bnx2x *bp)
1796 {
1797 	int rc;
1798 
1799 	rc = pci_enable_msi(bp->pdev);
1800 	if (rc) {
1801 		BNX2X_DEV_INFO("MSI is not attainable\n");
1802 		return -1;
1803 	}
1804 	bp->flags |= USING_MSI_FLAG;
1805 
1806 	return 0;
1807 }
1808 
1809 static int bnx2x_req_irq(struct bnx2x *bp)
1810 {
1811 	unsigned long flags;
1812 	unsigned int irq;
1813 
1814 	if (bp->flags & (USING_MSI_FLAG | USING_MSIX_FLAG))
1815 		flags = 0;
1816 	else
1817 		flags = IRQF_SHARED;
1818 
1819 	if (bp->flags & USING_MSIX_FLAG)
1820 		irq = bp->msix_table[0].vector;
1821 	else
1822 		irq = bp->pdev->irq;
1823 
1824 	return request_irq(irq, bnx2x_interrupt, flags, bp->dev->name, bp->dev);
1825 }
1826 
1827 static int bnx2x_setup_irqs(struct bnx2x *bp)
1828 {
1829 	int rc = 0;
1830 	if (bp->flags & USING_MSIX_FLAG &&
1831 	    !(bp->flags & USING_SINGLE_MSIX_FLAG)) {
1832 		rc = bnx2x_req_msix_irqs(bp);
1833 		if (rc)
1834 			return rc;
1835 	} else {
1836 		rc = bnx2x_req_irq(bp);
1837 		if (rc) {
1838 			BNX2X_ERR("IRQ request failed  rc %d, aborting\n", rc);
1839 			return rc;
1840 		}
1841 		if (bp->flags & USING_MSI_FLAG) {
1842 			bp->dev->irq = bp->pdev->irq;
1843 			netdev_info(bp->dev, "using MSI IRQ %d\n",
1844 				    bp->dev->irq);
1845 		}
1846 		if (bp->flags & USING_MSIX_FLAG) {
1847 			bp->dev->irq = bp->msix_table[0].vector;
1848 			netdev_info(bp->dev, "using MSIX IRQ %d\n",
1849 				    bp->dev->irq);
1850 		}
1851 	}
1852 
1853 	return 0;
1854 }
1855 
1856 static void bnx2x_napi_enable_cnic(struct bnx2x *bp)
1857 {
1858 	int i;
1859 
1860 	for_each_rx_queue_cnic(bp, i) {
1861 		napi_enable(&bnx2x_fp(bp, i, napi));
1862 	}
1863 }
1864 
1865 static void bnx2x_napi_enable(struct bnx2x *bp)
1866 {
1867 	int i;
1868 
1869 	for_each_eth_queue(bp, i) {
1870 		napi_enable(&bnx2x_fp(bp, i, napi));
1871 	}
1872 }
1873 
1874 static void bnx2x_napi_disable_cnic(struct bnx2x *bp)
1875 {
1876 	int i;
1877 
1878 	for_each_rx_queue_cnic(bp, i) {
1879 		napi_disable(&bnx2x_fp(bp, i, napi));
1880 	}
1881 }
1882 
1883 static void bnx2x_napi_disable(struct bnx2x *bp)
1884 {
1885 	int i;
1886 
1887 	for_each_eth_queue(bp, i) {
1888 		napi_disable(&bnx2x_fp(bp, i, napi));
1889 	}
1890 }
1891 
1892 void bnx2x_netif_start(struct bnx2x *bp)
1893 {
1894 	if (netif_running(bp->dev)) {
1895 		bnx2x_napi_enable(bp);
1896 		if (CNIC_LOADED(bp))
1897 			bnx2x_napi_enable_cnic(bp);
1898 		bnx2x_int_enable(bp);
1899 		if (bp->state == BNX2X_STATE_OPEN)
1900 			netif_tx_wake_all_queues(bp->dev);
1901 	}
1902 }
1903 
1904 void bnx2x_netif_stop(struct bnx2x *bp, int disable_hw)
1905 {
1906 	bnx2x_int_disable_sync(bp, disable_hw);
1907 	bnx2x_napi_disable(bp);
1908 	if (CNIC_LOADED(bp))
1909 		bnx2x_napi_disable_cnic(bp);
1910 }
1911 
1912 u16 bnx2x_select_queue(struct net_device *dev, struct sk_buff *skb,
1913 		       struct net_device *sb_dev,
1914 		       select_queue_fallback_t fallback)
1915 {
1916 	struct bnx2x *bp = netdev_priv(dev);
1917 
1918 	if (CNIC_LOADED(bp) && !NO_FCOE(bp)) {
1919 		struct ethhdr *hdr = (struct ethhdr *)skb->data;
1920 		u16 ether_type = ntohs(hdr->h_proto);
1921 
1922 		/* Skip VLAN tag if present */
1923 		if (ether_type == ETH_P_8021Q) {
1924 			struct vlan_ethhdr *vhdr =
1925 				(struct vlan_ethhdr *)skb->data;
1926 
1927 			ether_type = ntohs(vhdr->h_vlan_encapsulated_proto);
1928 		}
1929 
1930 		/* If ethertype is FCoE or FIP - use FCoE ring */
1931 		if ((ether_type == ETH_P_FCOE) || (ether_type == ETH_P_FIP))
1932 			return bnx2x_fcoe_tx(bp, txq_index);
1933 	}
1934 
1935 	/* select a non-FCoE queue */
1936 	return fallback(dev, skb, NULL) %
1937 	       (BNX2X_NUM_ETH_QUEUES(bp) * bp->max_cos);
1938 }
1939 
1940 void bnx2x_set_num_queues(struct bnx2x *bp)
1941 {
1942 	/* RSS queues */
1943 	bp->num_ethernet_queues = bnx2x_calc_num_queues(bp);
1944 
1945 	/* override in STORAGE SD modes */
1946 	if (IS_MF_STORAGE_ONLY(bp))
1947 		bp->num_ethernet_queues = 1;
1948 
1949 	/* Add special queues */
1950 	bp->num_cnic_queues = CNIC_SUPPORT(bp); /* For FCOE */
1951 	bp->num_queues = bp->num_ethernet_queues + bp->num_cnic_queues;
1952 
1953 	BNX2X_DEV_INFO("set number of queues to %d\n", bp->num_queues);
1954 }
1955 
1956 /**
1957  * bnx2x_set_real_num_queues - configure netdev->real_num_[tx,rx]_queues
1958  *
1959  * @bp:		Driver handle
1960  *
1961  * We currently support for at most 16 Tx queues for each CoS thus we will
1962  * allocate a multiple of 16 for ETH L2 rings according to the value of the
1963  * bp->max_cos.
1964  *
1965  * If there is an FCoE L2 queue the appropriate Tx queue will have the next
1966  * index after all ETH L2 indices.
1967  *
1968  * If the actual number of Tx queues (for each CoS) is less than 16 then there
1969  * will be the holes at the end of each group of 16 ETh L2 indices (0..15,
1970  * 16..31,...) with indices that are not coupled with any real Tx queue.
1971  *
1972  * The proper configuration of skb->queue_mapping is handled by
1973  * bnx2x_select_queue() and __skb_tx_hash().
1974  *
1975  * bnx2x_setup_tc() takes care of the proper TC mappings so that __skb_tx_hash()
1976  * will return a proper Tx index if TC is enabled (netdev->num_tc > 0).
1977  */
1978 static int bnx2x_set_real_num_queues(struct bnx2x *bp, int include_cnic)
1979 {
1980 	int rc, tx, rx;
1981 
1982 	tx = BNX2X_NUM_ETH_QUEUES(bp) * bp->max_cos;
1983 	rx = BNX2X_NUM_ETH_QUEUES(bp);
1984 
1985 /* account for fcoe queue */
1986 	if (include_cnic && !NO_FCOE(bp)) {
1987 		rx++;
1988 		tx++;
1989 	}
1990 
1991 	rc = netif_set_real_num_tx_queues(bp->dev, tx);
1992 	if (rc) {
1993 		BNX2X_ERR("Failed to set real number of Tx queues: %d\n", rc);
1994 		return rc;
1995 	}
1996 	rc = netif_set_real_num_rx_queues(bp->dev, rx);
1997 	if (rc) {
1998 		BNX2X_ERR("Failed to set real number of Rx queues: %d\n", rc);
1999 		return rc;
2000 	}
2001 
2002 	DP(NETIF_MSG_IFUP, "Setting real num queues to (tx, rx) (%d, %d)\n",
2003 			  tx, rx);
2004 
2005 	return rc;
2006 }
2007 
2008 static void bnx2x_set_rx_buf_size(struct bnx2x *bp)
2009 {
2010 	int i;
2011 
2012 	for_each_queue(bp, i) {
2013 		struct bnx2x_fastpath *fp = &bp->fp[i];
2014 		u32 mtu;
2015 
2016 		/* Always use a mini-jumbo MTU for the FCoE L2 ring */
2017 		if (IS_FCOE_IDX(i))
2018 			/*
2019 			 * Although there are no IP frames expected to arrive to
2020 			 * this ring we still want to add an
2021 			 * IP_HEADER_ALIGNMENT_PADDING to prevent a buffer
2022 			 * overrun attack.
2023 			 */
2024 			mtu = BNX2X_FCOE_MINI_JUMBO_MTU;
2025 		else
2026 			mtu = bp->dev->mtu;
2027 		fp->rx_buf_size = BNX2X_FW_RX_ALIGN_START +
2028 				  IP_HEADER_ALIGNMENT_PADDING +
2029 				  ETH_OVERHEAD +
2030 				  mtu +
2031 				  BNX2X_FW_RX_ALIGN_END;
2032 		fp->rx_buf_size = SKB_DATA_ALIGN(fp->rx_buf_size);
2033 		/* Note : rx_buf_size doesn't take into account NET_SKB_PAD */
2034 		if (fp->rx_buf_size + NET_SKB_PAD <= PAGE_SIZE)
2035 			fp->rx_frag_size = fp->rx_buf_size + NET_SKB_PAD;
2036 		else
2037 			fp->rx_frag_size = 0;
2038 	}
2039 }
2040 
2041 static int bnx2x_init_rss(struct bnx2x *bp)
2042 {
2043 	int i;
2044 	u8 num_eth_queues = BNX2X_NUM_ETH_QUEUES(bp);
2045 
2046 	/* Prepare the initial contents for the indirection table if RSS is
2047 	 * enabled
2048 	 */
2049 	for (i = 0; i < sizeof(bp->rss_conf_obj.ind_table); i++)
2050 		bp->rss_conf_obj.ind_table[i] =
2051 			bp->fp->cl_id +
2052 			ethtool_rxfh_indir_default(i, num_eth_queues);
2053 
2054 	/*
2055 	 * For 57710 and 57711 SEARCHER configuration (rss_keys) is
2056 	 * per-port, so if explicit configuration is needed , do it only
2057 	 * for a PMF.
2058 	 *
2059 	 * For 57712 and newer on the other hand it's a per-function
2060 	 * configuration.
2061 	 */
2062 	return bnx2x_config_rss_eth(bp, bp->port.pmf || !CHIP_IS_E1x(bp));
2063 }
2064 
2065 int bnx2x_rss(struct bnx2x *bp, struct bnx2x_rss_config_obj *rss_obj,
2066 	      bool config_hash, bool enable)
2067 {
2068 	struct bnx2x_config_rss_params params = {NULL};
2069 
2070 	/* Although RSS is meaningless when there is a single HW queue we
2071 	 * still need it enabled in order to have HW Rx hash generated.
2072 	 *
2073 	 * if (!is_eth_multi(bp))
2074 	 *      bp->multi_mode = ETH_RSS_MODE_DISABLED;
2075 	 */
2076 
2077 	params.rss_obj = rss_obj;
2078 
2079 	__set_bit(RAMROD_COMP_WAIT, &params.ramrod_flags);
2080 
2081 	if (enable) {
2082 		__set_bit(BNX2X_RSS_MODE_REGULAR, &params.rss_flags);
2083 
2084 		/* RSS configuration */
2085 		__set_bit(BNX2X_RSS_IPV4, &params.rss_flags);
2086 		__set_bit(BNX2X_RSS_IPV4_TCP, &params.rss_flags);
2087 		__set_bit(BNX2X_RSS_IPV6, &params.rss_flags);
2088 		__set_bit(BNX2X_RSS_IPV6_TCP, &params.rss_flags);
2089 		if (rss_obj->udp_rss_v4)
2090 			__set_bit(BNX2X_RSS_IPV4_UDP, &params.rss_flags);
2091 		if (rss_obj->udp_rss_v6)
2092 			__set_bit(BNX2X_RSS_IPV6_UDP, &params.rss_flags);
2093 
2094 		if (!CHIP_IS_E1x(bp)) {
2095 			/* valid only for TUNN_MODE_VXLAN tunnel mode */
2096 			__set_bit(BNX2X_RSS_IPV4_VXLAN, &params.rss_flags);
2097 			__set_bit(BNX2X_RSS_IPV6_VXLAN, &params.rss_flags);
2098 
2099 			/* valid only for TUNN_MODE_GRE tunnel mode */
2100 			__set_bit(BNX2X_RSS_TUNN_INNER_HDRS, &params.rss_flags);
2101 		}
2102 	} else {
2103 		__set_bit(BNX2X_RSS_MODE_DISABLED, &params.rss_flags);
2104 	}
2105 
2106 	/* Hash bits */
2107 	params.rss_result_mask = MULTI_MASK;
2108 
2109 	memcpy(params.ind_table, rss_obj->ind_table, sizeof(params.ind_table));
2110 
2111 	if (config_hash) {
2112 		/* RSS keys */
2113 		netdev_rss_key_fill(params.rss_key, T_ETH_RSS_KEY * 4);
2114 		__set_bit(BNX2X_RSS_SET_SRCH, &params.rss_flags);
2115 	}
2116 
2117 	if (IS_PF(bp))
2118 		return bnx2x_config_rss(bp, &params);
2119 	else
2120 		return bnx2x_vfpf_config_rss(bp, &params);
2121 }
2122 
2123 static int bnx2x_init_hw(struct bnx2x *bp, u32 load_code)
2124 {
2125 	struct bnx2x_func_state_params func_params = {NULL};
2126 
2127 	/* Prepare parameters for function state transitions */
2128 	__set_bit(RAMROD_COMP_WAIT, &func_params.ramrod_flags);
2129 
2130 	func_params.f_obj = &bp->func_obj;
2131 	func_params.cmd = BNX2X_F_CMD_HW_INIT;
2132 
2133 	func_params.params.hw_init.load_phase = load_code;
2134 
2135 	return bnx2x_func_state_change(bp, &func_params);
2136 }
2137 
2138 /*
2139  * Cleans the object that have internal lists without sending
2140  * ramrods. Should be run when interrupts are disabled.
2141  */
2142 void bnx2x_squeeze_objects(struct bnx2x *bp)
2143 {
2144 	int rc;
2145 	unsigned long ramrod_flags = 0, vlan_mac_flags = 0;
2146 	struct bnx2x_mcast_ramrod_params rparam = {NULL};
2147 	struct bnx2x_vlan_mac_obj *mac_obj = &bp->sp_objs->mac_obj;
2148 
2149 	/***************** Cleanup MACs' object first *************************/
2150 
2151 	/* Wait for completion of requested */
2152 	__set_bit(RAMROD_COMP_WAIT, &ramrod_flags);
2153 	/* Perform a dry cleanup */
2154 	__set_bit(RAMROD_DRV_CLR_ONLY, &ramrod_flags);
2155 
2156 	/* Clean ETH primary MAC */
2157 	__set_bit(BNX2X_ETH_MAC, &vlan_mac_flags);
2158 	rc = mac_obj->delete_all(bp, &bp->sp_objs->mac_obj, &vlan_mac_flags,
2159 				 &ramrod_flags);
2160 	if (rc != 0)
2161 		BNX2X_ERR("Failed to clean ETH MACs: %d\n", rc);
2162 
2163 	/* Cleanup UC list */
2164 	vlan_mac_flags = 0;
2165 	__set_bit(BNX2X_UC_LIST_MAC, &vlan_mac_flags);
2166 	rc = mac_obj->delete_all(bp, mac_obj, &vlan_mac_flags,
2167 				 &ramrod_flags);
2168 	if (rc != 0)
2169 		BNX2X_ERR("Failed to clean UC list MACs: %d\n", rc);
2170 
2171 	/***************** Now clean mcast object *****************************/
2172 	rparam.mcast_obj = &bp->mcast_obj;
2173 	__set_bit(RAMROD_DRV_CLR_ONLY, &rparam.ramrod_flags);
2174 
2175 	/* Add a DEL command... - Since we're doing a driver cleanup only,
2176 	 * we take a lock surrounding both the initial send and the CONTs,
2177 	 * as we don't want a true completion to disrupt us in the middle.
2178 	 */
2179 	netif_addr_lock_bh(bp->dev);
2180 	rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_DEL);
2181 	if (rc < 0)
2182 		BNX2X_ERR("Failed to add a new DEL command to a multi-cast object: %d\n",
2183 			  rc);
2184 
2185 	/* ...and wait until all pending commands are cleared */
2186 	rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_CONT);
2187 	while (rc != 0) {
2188 		if (rc < 0) {
2189 			BNX2X_ERR("Failed to clean multi-cast object: %d\n",
2190 				  rc);
2191 			netif_addr_unlock_bh(bp->dev);
2192 			return;
2193 		}
2194 
2195 		rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_CONT);
2196 	}
2197 	netif_addr_unlock_bh(bp->dev);
2198 }
2199 
2200 #ifndef BNX2X_STOP_ON_ERROR
2201 #define LOAD_ERROR_EXIT(bp, label) \
2202 	do { \
2203 		(bp)->state = BNX2X_STATE_ERROR; \
2204 		goto label; \
2205 	} while (0)
2206 
2207 #define LOAD_ERROR_EXIT_CNIC(bp, label) \
2208 	do { \
2209 		bp->cnic_loaded = false; \
2210 		goto label; \
2211 	} while (0)
2212 #else /*BNX2X_STOP_ON_ERROR*/
2213 #define LOAD_ERROR_EXIT(bp, label) \
2214 	do { \
2215 		(bp)->state = BNX2X_STATE_ERROR; \
2216 		(bp)->panic = 1; \
2217 		return -EBUSY; \
2218 	} while (0)
2219 #define LOAD_ERROR_EXIT_CNIC(bp, label) \
2220 	do { \
2221 		bp->cnic_loaded = false; \
2222 		(bp)->panic = 1; \
2223 		return -EBUSY; \
2224 	} while (0)
2225 #endif /*BNX2X_STOP_ON_ERROR*/
2226 
2227 static void bnx2x_free_fw_stats_mem(struct bnx2x *bp)
2228 {
2229 	BNX2X_PCI_FREE(bp->fw_stats, bp->fw_stats_mapping,
2230 		       bp->fw_stats_data_sz + bp->fw_stats_req_sz);
2231 	return;
2232 }
2233 
2234 static int bnx2x_alloc_fw_stats_mem(struct bnx2x *bp)
2235 {
2236 	int num_groups, vf_headroom = 0;
2237 	int is_fcoe_stats = NO_FCOE(bp) ? 0 : 1;
2238 
2239 	/* number of queues for statistics is number of eth queues + FCoE */
2240 	u8 num_queue_stats = BNX2X_NUM_ETH_QUEUES(bp) + is_fcoe_stats;
2241 
2242 	/* Total number of FW statistics requests =
2243 	 * 1 for port stats + 1 for PF stats + potential 2 for FCoE (fcoe proper
2244 	 * and fcoe l2 queue) stats + num of queues (which includes another 1
2245 	 * for fcoe l2 queue if applicable)
2246 	 */
2247 	bp->fw_stats_num = 2 + is_fcoe_stats + num_queue_stats;
2248 
2249 	/* vf stats appear in the request list, but their data is allocated by
2250 	 * the VFs themselves. We don't include them in the bp->fw_stats_num as
2251 	 * it is used to determine where to place the vf stats queries in the
2252 	 * request struct
2253 	 */
2254 	if (IS_SRIOV(bp))
2255 		vf_headroom = bnx2x_vf_headroom(bp);
2256 
2257 	/* Request is built from stats_query_header and an array of
2258 	 * stats_query_cmd_group each of which contains
2259 	 * STATS_QUERY_CMD_COUNT rules. The real number or requests is
2260 	 * configured in the stats_query_header.
2261 	 */
2262 	num_groups =
2263 		(((bp->fw_stats_num + vf_headroom) / STATS_QUERY_CMD_COUNT) +
2264 		 (((bp->fw_stats_num + vf_headroom) % STATS_QUERY_CMD_COUNT) ?
2265 		 1 : 0));
2266 
2267 	DP(BNX2X_MSG_SP, "stats fw_stats_num %d, vf headroom %d, num_groups %d\n",
2268 	   bp->fw_stats_num, vf_headroom, num_groups);
2269 	bp->fw_stats_req_sz = sizeof(struct stats_query_header) +
2270 		num_groups * sizeof(struct stats_query_cmd_group);
2271 
2272 	/* Data for statistics requests + stats_counter
2273 	 * stats_counter holds per-STORM counters that are incremented
2274 	 * when STORM has finished with the current request.
2275 	 * memory for FCoE offloaded statistics are counted anyway,
2276 	 * even if they will not be sent.
2277 	 * VF stats are not accounted for here as the data of VF stats is stored
2278 	 * in memory allocated by the VF, not here.
2279 	 */
2280 	bp->fw_stats_data_sz = sizeof(struct per_port_stats) +
2281 		sizeof(struct per_pf_stats) +
2282 		sizeof(struct fcoe_statistics_params) +
2283 		sizeof(struct per_queue_stats) * num_queue_stats +
2284 		sizeof(struct stats_counter);
2285 
2286 	bp->fw_stats = BNX2X_PCI_ALLOC(&bp->fw_stats_mapping,
2287 				       bp->fw_stats_data_sz + bp->fw_stats_req_sz);
2288 	if (!bp->fw_stats)
2289 		goto alloc_mem_err;
2290 
2291 	/* Set shortcuts */
2292 	bp->fw_stats_req = (struct bnx2x_fw_stats_req *)bp->fw_stats;
2293 	bp->fw_stats_req_mapping = bp->fw_stats_mapping;
2294 	bp->fw_stats_data = (struct bnx2x_fw_stats_data *)
2295 		((u8 *)bp->fw_stats + bp->fw_stats_req_sz);
2296 	bp->fw_stats_data_mapping = bp->fw_stats_mapping +
2297 		bp->fw_stats_req_sz;
2298 
2299 	DP(BNX2X_MSG_SP, "statistics request base address set to %x %x\n",
2300 	   U64_HI(bp->fw_stats_req_mapping),
2301 	   U64_LO(bp->fw_stats_req_mapping));
2302 	DP(BNX2X_MSG_SP, "statistics data base address set to %x %x\n",
2303 	   U64_HI(bp->fw_stats_data_mapping),
2304 	   U64_LO(bp->fw_stats_data_mapping));
2305 	return 0;
2306 
2307 alloc_mem_err:
2308 	bnx2x_free_fw_stats_mem(bp);
2309 	BNX2X_ERR("Can't allocate FW stats memory\n");
2310 	return -ENOMEM;
2311 }
2312 
2313 /* send load request to mcp and analyze response */
2314 static int bnx2x_nic_load_request(struct bnx2x *bp, u32 *load_code)
2315 {
2316 	u32 param;
2317 
2318 	/* init fw_seq */
2319 	bp->fw_seq =
2320 		(SHMEM_RD(bp, func_mb[BP_FW_MB_IDX(bp)].drv_mb_header) &
2321 		 DRV_MSG_SEQ_NUMBER_MASK);
2322 	BNX2X_DEV_INFO("fw_seq 0x%08x\n", bp->fw_seq);
2323 
2324 	/* Get current FW pulse sequence */
2325 	bp->fw_drv_pulse_wr_seq =
2326 		(SHMEM_RD(bp, func_mb[BP_FW_MB_IDX(bp)].drv_pulse_mb) &
2327 		 DRV_PULSE_SEQ_MASK);
2328 	BNX2X_DEV_INFO("drv_pulse 0x%x\n", bp->fw_drv_pulse_wr_seq);
2329 
2330 	param = DRV_MSG_CODE_LOAD_REQ_WITH_LFA;
2331 
2332 	if (IS_MF_SD(bp) && bnx2x_port_after_undi(bp))
2333 		param |= DRV_MSG_CODE_LOAD_REQ_FORCE_LFA;
2334 
2335 	/* load request */
2336 	(*load_code) = bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_REQ, param);
2337 
2338 	/* if mcp fails to respond we must abort */
2339 	if (!(*load_code)) {
2340 		BNX2X_ERR("MCP response failure, aborting\n");
2341 		return -EBUSY;
2342 	}
2343 
2344 	/* If mcp refused (e.g. other port is in diagnostic mode) we
2345 	 * must abort
2346 	 */
2347 	if ((*load_code) == FW_MSG_CODE_DRV_LOAD_REFUSED) {
2348 		BNX2X_ERR("MCP refused load request, aborting\n");
2349 		return -EBUSY;
2350 	}
2351 	return 0;
2352 }
2353 
2354 /* check whether another PF has already loaded FW to chip. In
2355  * virtualized environments a pf from another VM may have already
2356  * initialized the device including loading FW
2357  */
2358 int bnx2x_compare_fw_ver(struct bnx2x *bp, u32 load_code, bool print_err)
2359 {
2360 	/* is another pf loaded on this engine? */
2361 	if (load_code != FW_MSG_CODE_DRV_LOAD_COMMON_CHIP &&
2362 	    load_code != FW_MSG_CODE_DRV_LOAD_COMMON) {
2363 		/* build my FW version dword */
2364 		u32 my_fw = (BCM_5710_FW_MAJOR_VERSION) +
2365 			(BCM_5710_FW_MINOR_VERSION << 8) +
2366 			(BCM_5710_FW_REVISION_VERSION << 16) +
2367 			(BCM_5710_FW_ENGINEERING_VERSION << 24);
2368 
2369 		/* read loaded FW from chip */
2370 		u32 loaded_fw = REG_RD(bp, XSEM_REG_PRAM);
2371 
2372 		DP(BNX2X_MSG_SP, "loaded fw %x, my fw %x\n",
2373 		   loaded_fw, my_fw);
2374 
2375 		/* abort nic load if version mismatch */
2376 		if (my_fw != loaded_fw) {
2377 			if (print_err)
2378 				BNX2X_ERR("bnx2x with FW %x was already loaded which mismatches my %x FW. Aborting\n",
2379 					  loaded_fw, my_fw);
2380 			else
2381 				BNX2X_DEV_INFO("bnx2x with FW %x was already loaded which mismatches my %x FW, possibly due to MF UNDI\n",
2382 					       loaded_fw, my_fw);
2383 			return -EBUSY;
2384 		}
2385 	}
2386 	return 0;
2387 }
2388 
2389 /* returns the "mcp load_code" according to global load_count array */
2390 static int bnx2x_nic_load_no_mcp(struct bnx2x *bp, int port)
2391 {
2392 	int path = BP_PATH(bp);
2393 
2394 	DP(NETIF_MSG_IFUP, "NO MCP - load counts[%d]      %d, %d, %d\n",
2395 	   path, bnx2x_load_count[path][0], bnx2x_load_count[path][1],
2396 	   bnx2x_load_count[path][2]);
2397 	bnx2x_load_count[path][0]++;
2398 	bnx2x_load_count[path][1 + port]++;
2399 	DP(NETIF_MSG_IFUP, "NO MCP - new load counts[%d]  %d, %d, %d\n",
2400 	   path, bnx2x_load_count[path][0], bnx2x_load_count[path][1],
2401 	   bnx2x_load_count[path][2]);
2402 	if (bnx2x_load_count[path][0] == 1)
2403 		return FW_MSG_CODE_DRV_LOAD_COMMON;
2404 	else if (bnx2x_load_count[path][1 + port] == 1)
2405 		return FW_MSG_CODE_DRV_LOAD_PORT;
2406 	else
2407 		return FW_MSG_CODE_DRV_LOAD_FUNCTION;
2408 }
2409 
2410 /* mark PMF if applicable */
2411 static void bnx2x_nic_load_pmf(struct bnx2x *bp, u32 load_code)
2412 {
2413 	if ((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
2414 	    (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP) ||
2415 	    (load_code == FW_MSG_CODE_DRV_LOAD_PORT)) {
2416 		bp->port.pmf = 1;
2417 		/* We need the barrier to ensure the ordering between the
2418 		 * writing to bp->port.pmf here and reading it from the
2419 		 * bnx2x_periodic_task().
2420 		 */
2421 		smp_mb();
2422 	} else {
2423 		bp->port.pmf = 0;
2424 	}
2425 
2426 	DP(NETIF_MSG_LINK, "pmf %d\n", bp->port.pmf);
2427 }
2428 
2429 static void bnx2x_nic_load_afex_dcc(struct bnx2x *bp, int load_code)
2430 {
2431 	if (((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
2432 	     (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP)) &&
2433 	    (bp->common.shmem2_base)) {
2434 		if (SHMEM2_HAS(bp, dcc_support))
2435 			SHMEM2_WR(bp, dcc_support,
2436 				  (SHMEM_DCC_SUPPORT_DISABLE_ENABLE_PF_TLV |
2437 				   SHMEM_DCC_SUPPORT_BANDWIDTH_ALLOCATION_TLV));
2438 		if (SHMEM2_HAS(bp, afex_driver_support))
2439 			SHMEM2_WR(bp, afex_driver_support,
2440 				  SHMEM_AFEX_SUPPORTED_VERSION_ONE);
2441 	}
2442 
2443 	/* Set AFEX default VLAN tag to an invalid value */
2444 	bp->afex_def_vlan_tag = -1;
2445 }
2446 
2447 /**
2448  * bnx2x_bz_fp - zero content of the fastpath structure.
2449  *
2450  * @bp:		driver handle
2451  * @index:	fastpath index to be zeroed
2452  *
2453  * Makes sure the contents of the bp->fp[index].napi is kept
2454  * intact.
2455  */
2456 static void bnx2x_bz_fp(struct bnx2x *bp, int index)
2457 {
2458 	struct bnx2x_fastpath *fp = &bp->fp[index];
2459 	int cos;
2460 	struct napi_struct orig_napi = fp->napi;
2461 	struct bnx2x_agg_info *orig_tpa_info = fp->tpa_info;
2462 
2463 	/* bzero bnx2x_fastpath contents */
2464 	if (fp->tpa_info)
2465 		memset(fp->tpa_info, 0, ETH_MAX_AGGREGATION_QUEUES_E1H_E2 *
2466 		       sizeof(struct bnx2x_agg_info));
2467 	memset(fp, 0, sizeof(*fp));
2468 
2469 	/* Restore the NAPI object as it has been already initialized */
2470 	fp->napi = orig_napi;
2471 	fp->tpa_info = orig_tpa_info;
2472 	fp->bp = bp;
2473 	fp->index = index;
2474 	if (IS_ETH_FP(fp))
2475 		fp->max_cos = bp->max_cos;
2476 	else
2477 		/* Special queues support only one CoS */
2478 		fp->max_cos = 1;
2479 
2480 	/* Init txdata pointers */
2481 	if (IS_FCOE_FP(fp))
2482 		fp->txdata_ptr[0] = &bp->bnx2x_txq[FCOE_TXQ_IDX(bp)];
2483 	if (IS_ETH_FP(fp))
2484 		for_each_cos_in_tx_queue(fp, cos)
2485 			fp->txdata_ptr[cos] = &bp->bnx2x_txq[cos *
2486 				BNX2X_NUM_ETH_QUEUES(bp) + index];
2487 
2488 	/* set the tpa flag for each queue. The tpa flag determines the queue
2489 	 * minimal size so it must be set prior to queue memory allocation
2490 	 */
2491 	if (bp->dev->features & NETIF_F_LRO)
2492 		fp->mode = TPA_MODE_LRO;
2493 	else if (bp->dev->features & NETIF_F_GRO_HW)
2494 		fp->mode = TPA_MODE_GRO;
2495 	else
2496 		fp->mode = TPA_MODE_DISABLED;
2497 
2498 	/* We don't want TPA if it's disabled in bp
2499 	 * or if this is an FCoE L2 ring.
2500 	 */
2501 	if (bp->disable_tpa || IS_FCOE_FP(fp))
2502 		fp->mode = TPA_MODE_DISABLED;
2503 }
2504 
2505 void bnx2x_set_os_driver_state(struct bnx2x *bp, u32 state)
2506 {
2507 	u32 cur;
2508 
2509 	if (!IS_MF_BD(bp) || !SHMEM2_HAS(bp, os_driver_state) || IS_VF(bp))
2510 		return;
2511 
2512 	cur = SHMEM2_RD(bp, os_driver_state[BP_FW_MB_IDX(bp)]);
2513 	DP(NETIF_MSG_IFUP, "Driver state %08x-->%08x\n",
2514 	   cur, state);
2515 
2516 	SHMEM2_WR(bp, os_driver_state[BP_FW_MB_IDX(bp)], state);
2517 }
2518 
2519 int bnx2x_load_cnic(struct bnx2x *bp)
2520 {
2521 	int i, rc, port = BP_PORT(bp);
2522 
2523 	DP(NETIF_MSG_IFUP, "Starting CNIC-related load\n");
2524 
2525 	mutex_init(&bp->cnic_mutex);
2526 
2527 	if (IS_PF(bp)) {
2528 		rc = bnx2x_alloc_mem_cnic(bp);
2529 		if (rc) {
2530 			BNX2X_ERR("Unable to allocate bp memory for cnic\n");
2531 			LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic0);
2532 		}
2533 	}
2534 
2535 	rc = bnx2x_alloc_fp_mem_cnic(bp);
2536 	if (rc) {
2537 		BNX2X_ERR("Unable to allocate memory for cnic fps\n");
2538 		LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic0);
2539 	}
2540 
2541 	/* Update the number of queues with the cnic queues */
2542 	rc = bnx2x_set_real_num_queues(bp, 1);
2543 	if (rc) {
2544 		BNX2X_ERR("Unable to set real_num_queues including cnic\n");
2545 		LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic0);
2546 	}
2547 
2548 	/* Add all CNIC NAPI objects */
2549 	bnx2x_add_all_napi_cnic(bp);
2550 	DP(NETIF_MSG_IFUP, "cnic napi added\n");
2551 	bnx2x_napi_enable_cnic(bp);
2552 
2553 	rc = bnx2x_init_hw_func_cnic(bp);
2554 	if (rc)
2555 		LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic1);
2556 
2557 	bnx2x_nic_init_cnic(bp);
2558 
2559 	if (IS_PF(bp)) {
2560 		/* Enable Timer scan */
2561 		REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + port*4, 1);
2562 
2563 		/* setup cnic queues */
2564 		for_each_cnic_queue(bp, i) {
2565 			rc = bnx2x_setup_queue(bp, &bp->fp[i], 0);
2566 			if (rc) {
2567 				BNX2X_ERR("Queue setup failed\n");
2568 				LOAD_ERROR_EXIT(bp, load_error_cnic2);
2569 			}
2570 		}
2571 	}
2572 
2573 	/* Initialize Rx filter. */
2574 	bnx2x_set_rx_mode_inner(bp);
2575 
2576 	/* re-read iscsi info */
2577 	bnx2x_get_iscsi_info(bp);
2578 	bnx2x_setup_cnic_irq_info(bp);
2579 	bnx2x_setup_cnic_info(bp);
2580 	bp->cnic_loaded = true;
2581 	if (bp->state == BNX2X_STATE_OPEN)
2582 		bnx2x_cnic_notify(bp, CNIC_CTL_START_CMD);
2583 
2584 	DP(NETIF_MSG_IFUP, "Ending successfully CNIC-related load\n");
2585 
2586 	return 0;
2587 
2588 #ifndef BNX2X_STOP_ON_ERROR
2589 load_error_cnic2:
2590 	/* Disable Timer scan */
2591 	REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + port*4, 0);
2592 
2593 load_error_cnic1:
2594 	bnx2x_napi_disable_cnic(bp);
2595 	/* Update the number of queues without the cnic queues */
2596 	if (bnx2x_set_real_num_queues(bp, 0))
2597 		BNX2X_ERR("Unable to set real_num_queues not including cnic\n");
2598 load_error_cnic0:
2599 	BNX2X_ERR("CNIC-related load failed\n");
2600 	bnx2x_free_fp_mem_cnic(bp);
2601 	bnx2x_free_mem_cnic(bp);
2602 	return rc;
2603 #endif /* ! BNX2X_STOP_ON_ERROR */
2604 }
2605 
2606 /* must be called with rtnl_lock */
2607 int bnx2x_nic_load(struct bnx2x *bp, int load_mode)
2608 {
2609 	int port = BP_PORT(bp);
2610 	int i, rc = 0, load_code = 0;
2611 
2612 	DP(NETIF_MSG_IFUP, "Starting NIC load\n");
2613 	DP(NETIF_MSG_IFUP,
2614 	   "CNIC is %s\n", CNIC_ENABLED(bp) ? "enabled" : "disabled");
2615 
2616 #ifdef BNX2X_STOP_ON_ERROR
2617 	if (unlikely(bp->panic)) {
2618 		BNX2X_ERR("Can't load NIC when there is panic\n");
2619 		return -EPERM;
2620 	}
2621 #endif
2622 
2623 	bp->state = BNX2X_STATE_OPENING_WAIT4_LOAD;
2624 
2625 	/* zero the structure w/o any lock, before SP handler is initialized */
2626 	memset(&bp->last_reported_link, 0, sizeof(bp->last_reported_link));
2627 	__set_bit(BNX2X_LINK_REPORT_LINK_DOWN,
2628 		&bp->last_reported_link.link_report_flags);
2629 
2630 	if (IS_PF(bp))
2631 		/* must be called before memory allocation and HW init */
2632 		bnx2x_ilt_set_info(bp);
2633 
2634 	/*
2635 	 * Zero fastpath structures preserving invariants like napi, which are
2636 	 * allocated only once, fp index, max_cos, bp pointer.
2637 	 * Also set fp->mode and txdata_ptr.
2638 	 */
2639 	DP(NETIF_MSG_IFUP, "num queues: %d", bp->num_queues);
2640 	for_each_queue(bp, i)
2641 		bnx2x_bz_fp(bp, i);
2642 	memset(bp->bnx2x_txq, 0, (BNX2X_MAX_RSS_COUNT(bp) * BNX2X_MULTI_TX_COS +
2643 				  bp->num_cnic_queues) *
2644 				  sizeof(struct bnx2x_fp_txdata));
2645 
2646 	bp->fcoe_init = false;
2647 
2648 	/* Set the receive queues buffer size */
2649 	bnx2x_set_rx_buf_size(bp);
2650 
2651 	if (IS_PF(bp)) {
2652 		rc = bnx2x_alloc_mem(bp);
2653 		if (rc) {
2654 			BNX2X_ERR("Unable to allocate bp memory\n");
2655 			return rc;
2656 		}
2657 	}
2658 
2659 	/* need to be done after alloc mem, since it's self adjusting to amount
2660 	 * of memory available for RSS queues
2661 	 */
2662 	rc = bnx2x_alloc_fp_mem(bp);
2663 	if (rc) {
2664 		BNX2X_ERR("Unable to allocate memory for fps\n");
2665 		LOAD_ERROR_EXIT(bp, load_error0);
2666 	}
2667 
2668 	/* Allocated memory for FW statistics  */
2669 	if (bnx2x_alloc_fw_stats_mem(bp))
2670 		LOAD_ERROR_EXIT(bp, load_error0);
2671 
2672 	/* request pf to initialize status blocks */
2673 	if (IS_VF(bp)) {
2674 		rc = bnx2x_vfpf_init(bp);
2675 		if (rc)
2676 			LOAD_ERROR_EXIT(bp, load_error0);
2677 	}
2678 
2679 	/* As long as bnx2x_alloc_mem() may possibly update
2680 	 * bp->num_queues, bnx2x_set_real_num_queues() should always
2681 	 * come after it. At this stage cnic queues are not counted.
2682 	 */
2683 	rc = bnx2x_set_real_num_queues(bp, 0);
2684 	if (rc) {
2685 		BNX2X_ERR("Unable to set real_num_queues\n");
2686 		LOAD_ERROR_EXIT(bp, load_error0);
2687 	}
2688 
2689 	/* configure multi cos mappings in kernel.
2690 	 * this configuration may be overridden by a multi class queue
2691 	 * discipline or by a dcbx negotiation result.
2692 	 */
2693 	bnx2x_setup_tc(bp->dev, bp->max_cos);
2694 
2695 	/* Add all NAPI objects */
2696 	bnx2x_add_all_napi(bp);
2697 	DP(NETIF_MSG_IFUP, "napi added\n");
2698 	bnx2x_napi_enable(bp);
2699 
2700 	if (IS_PF(bp)) {
2701 		/* set pf load just before approaching the MCP */
2702 		bnx2x_set_pf_load(bp);
2703 
2704 		/* if mcp exists send load request and analyze response */
2705 		if (!BP_NOMCP(bp)) {
2706 			/* attempt to load pf */
2707 			rc = bnx2x_nic_load_request(bp, &load_code);
2708 			if (rc)
2709 				LOAD_ERROR_EXIT(bp, load_error1);
2710 
2711 			/* what did mcp say? */
2712 			rc = bnx2x_compare_fw_ver(bp, load_code, true);
2713 			if (rc) {
2714 				bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2715 				LOAD_ERROR_EXIT(bp, load_error2);
2716 			}
2717 		} else {
2718 			load_code = bnx2x_nic_load_no_mcp(bp, port);
2719 		}
2720 
2721 		/* mark pmf if applicable */
2722 		bnx2x_nic_load_pmf(bp, load_code);
2723 
2724 		/* Init Function state controlling object */
2725 		bnx2x__init_func_obj(bp);
2726 
2727 		/* Initialize HW */
2728 		rc = bnx2x_init_hw(bp, load_code);
2729 		if (rc) {
2730 			BNX2X_ERR("HW init failed, aborting\n");
2731 			bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2732 			LOAD_ERROR_EXIT(bp, load_error2);
2733 		}
2734 	}
2735 
2736 	bnx2x_pre_irq_nic_init(bp);
2737 
2738 	/* Connect to IRQs */
2739 	rc = bnx2x_setup_irqs(bp);
2740 	if (rc) {
2741 		BNX2X_ERR("setup irqs failed\n");
2742 		if (IS_PF(bp))
2743 			bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2744 		LOAD_ERROR_EXIT(bp, load_error2);
2745 	}
2746 
2747 	/* Init per-function objects */
2748 	if (IS_PF(bp)) {
2749 		/* Setup NIC internals and enable interrupts */
2750 		bnx2x_post_irq_nic_init(bp, load_code);
2751 
2752 		bnx2x_init_bp_objs(bp);
2753 		bnx2x_iov_nic_init(bp);
2754 
2755 		/* Set AFEX default VLAN tag to an invalid value */
2756 		bp->afex_def_vlan_tag = -1;
2757 		bnx2x_nic_load_afex_dcc(bp, load_code);
2758 		bp->state = BNX2X_STATE_OPENING_WAIT4_PORT;
2759 		rc = bnx2x_func_start(bp);
2760 		if (rc) {
2761 			BNX2X_ERR("Function start failed!\n");
2762 			bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2763 
2764 			LOAD_ERROR_EXIT(bp, load_error3);
2765 		}
2766 
2767 		/* Send LOAD_DONE command to MCP */
2768 		if (!BP_NOMCP(bp)) {
2769 			load_code = bnx2x_fw_command(bp,
2770 						     DRV_MSG_CODE_LOAD_DONE, 0);
2771 			if (!load_code) {
2772 				BNX2X_ERR("MCP response failure, aborting\n");
2773 				rc = -EBUSY;
2774 				LOAD_ERROR_EXIT(bp, load_error3);
2775 			}
2776 		}
2777 
2778 		/* initialize FW coalescing state machines in RAM */
2779 		bnx2x_update_coalesce(bp);
2780 	}
2781 
2782 	/* setup the leading queue */
2783 	rc = bnx2x_setup_leading(bp);
2784 	if (rc) {
2785 		BNX2X_ERR("Setup leading failed!\n");
2786 		LOAD_ERROR_EXIT(bp, load_error3);
2787 	}
2788 
2789 	/* set up the rest of the queues */
2790 	for_each_nondefault_eth_queue(bp, i) {
2791 		if (IS_PF(bp))
2792 			rc = bnx2x_setup_queue(bp, &bp->fp[i], false);
2793 		else /* VF */
2794 			rc = bnx2x_vfpf_setup_q(bp, &bp->fp[i], false);
2795 		if (rc) {
2796 			BNX2X_ERR("Queue %d setup failed\n", i);
2797 			LOAD_ERROR_EXIT(bp, load_error3);
2798 		}
2799 	}
2800 
2801 	/* setup rss */
2802 	rc = bnx2x_init_rss(bp);
2803 	if (rc) {
2804 		BNX2X_ERR("PF RSS init failed\n");
2805 		LOAD_ERROR_EXIT(bp, load_error3);
2806 	}
2807 
2808 	/* Now when Clients are configured we are ready to work */
2809 	bp->state = BNX2X_STATE_OPEN;
2810 
2811 	/* Configure a ucast MAC */
2812 	if (IS_PF(bp))
2813 		rc = bnx2x_set_eth_mac(bp, true);
2814 	else /* vf */
2815 		rc = bnx2x_vfpf_config_mac(bp, bp->dev->dev_addr, bp->fp->index,
2816 					   true);
2817 	if (rc) {
2818 		BNX2X_ERR("Setting Ethernet MAC failed\n");
2819 		LOAD_ERROR_EXIT(bp, load_error3);
2820 	}
2821 
2822 	if (IS_PF(bp) && bp->pending_max) {
2823 		bnx2x_update_max_mf_config(bp, bp->pending_max);
2824 		bp->pending_max = 0;
2825 	}
2826 
2827 	bp->force_link_down = false;
2828 	if (bp->port.pmf) {
2829 		rc = bnx2x_initial_phy_init(bp, load_mode);
2830 		if (rc)
2831 			LOAD_ERROR_EXIT(bp, load_error3);
2832 	}
2833 	bp->link_params.feature_config_flags &= ~FEATURE_CONFIG_BOOT_FROM_SAN;
2834 
2835 	/* Start fast path */
2836 
2837 	/* Re-configure vlan filters */
2838 	rc = bnx2x_vlan_reconfigure_vid(bp);
2839 	if (rc)
2840 		LOAD_ERROR_EXIT(bp, load_error3);
2841 
2842 	/* Initialize Rx filter. */
2843 	bnx2x_set_rx_mode_inner(bp);
2844 
2845 	if (bp->flags & PTP_SUPPORTED) {
2846 		bnx2x_init_ptp(bp);
2847 		bnx2x_configure_ptp_filters(bp);
2848 	}
2849 	/* Start Tx */
2850 	switch (load_mode) {
2851 	case LOAD_NORMAL:
2852 		/* Tx queue should be only re-enabled */
2853 		netif_tx_wake_all_queues(bp->dev);
2854 		break;
2855 
2856 	case LOAD_OPEN:
2857 		netif_tx_start_all_queues(bp->dev);
2858 		smp_mb__after_atomic();
2859 		break;
2860 
2861 	case LOAD_DIAG:
2862 	case LOAD_LOOPBACK_EXT:
2863 		bp->state = BNX2X_STATE_DIAG;
2864 		break;
2865 
2866 	default:
2867 		break;
2868 	}
2869 
2870 	if (bp->port.pmf)
2871 		bnx2x_update_drv_flags(bp, 1 << DRV_FLAGS_PORT_MASK, 0);
2872 	else
2873 		bnx2x__link_status_update(bp);
2874 
2875 	/* start the timer */
2876 	mod_timer(&bp->timer, jiffies + bp->current_interval);
2877 
2878 	if (CNIC_ENABLED(bp))
2879 		bnx2x_load_cnic(bp);
2880 
2881 	if (IS_PF(bp))
2882 		bnx2x_schedule_sp_rtnl(bp, BNX2X_SP_RTNL_GET_DRV_VERSION, 0);
2883 
2884 	if (IS_PF(bp) && SHMEM2_HAS(bp, drv_capabilities_flag)) {
2885 		/* mark driver is loaded in shmem2 */
2886 		u32 val;
2887 		val = SHMEM2_RD(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)]);
2888 		val &= ~DRV_FLAGS_MTU_MASK;
2889 		val |= (bp->dev->mtu << DRV_FLAGS_MTU_SHIFT);
2890 		SHMEM2_WR(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)],
2891 			  val | DRV_FLAGS_CAPABILITIES_LOADED_SUPPORTED |
2892 			  DRV_FLAGS_CAPABILITIES_LOADED_L2);
2893 	}
2894 
2895 	/* Wait for all pending SP commands to complete */
2896 	if (IS_PF(bp) && !bnx2x_wait_sp_comp(bp, ~0x0UL)) {
2897 		BNX2X_ERR("Timeout waiting for SP elements to complete\n");
2898 		bnx2x_nic_unload(bp, UNLOAD_CLOSE, false);
2899 		return -EBUSY;
2900 	}
2901 
2902 	/* Update driver data for On-Chip MFW dump. */
2903 	if (IS_PF(bp))
2904 		bnx2x_update_mfw_dump(bp);
2905 
2906 	/* If PMF - send ADMIN DCBX msg to MFW to initiate DCBX FSM */
2907 	if (bp->port.pmf && (bp->state != BNX2X_STATE_DIAG))
2908 		bnx2x_dcbx_init(bp, false);
2909 
2910 	if (!IS_MF_SD_STORAGE_PERSONALITY_ONLY(bp))
2911 		bnx2x_set_os_driver_state(bp, OS_DRIVER_STATE_ACTIVE);
2912 
2913 	DP(NETIF_MSG_IFUP, "Ending successfully NIC load\n");
2914 
2915 	return 0;
2916 
2917 #ifndef BNX2X_STOP_ON_ERROR
2918 load_error3:
2919 	if (IS_PF(bp)) {
2920 		bnx2x_int_disable_sync(bp, 1);
2921 
2922 		/* Clean queueable objects */
2923 		bnx2x_squeeze_objects(bp);
2924 	}
2925 
2926 	/* Free SKBs, SGEs, TPA pool and driver internals */
2927 	bnx2x_free_skbs(bp);
2928 	for_each_rx_queue(bp, i)
2929 		bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
2930 
2931 	/* Release IRQs */
2932 	bnx2x_free_irq(bp);
2933 load_error2:
2934 	if (IS_PF(bp) && !BP_NOMCP(bp)) {
2935 		bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_REQ_WOL_MCP, 0);
2936 		bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_DONE, 0);
2937 	}
2938 
2939 	bp->port.pmf = 0;
2940 load_error1:
2941 	bnx2x_napi_disable(bp);
2942 	bnx2x_del_all_napi(bp);
2943 
2944 	/* clear pf_load status, as it was already set */
2945 	if (IS_PF(bp))
2946 		bnx2x_clear_pf_load(bp);
2947 load_error0:
2948 	bnx2x_free_fw_stats_mem(bp);
2949 	bnx2x_free_fp_mem(bp);
2950 	bnx2x_free_mem(bp);
2951 
2952 	return rc;
2953 #endif /* ! BNX2X_STOP_ON_ERROR */
2954 }
2955 
2956 int bnx2x_drain_tx_queues(struct bnx2x *bp)
2957 {
2958 	u8 rc = 0, cos, i;
2959 
2960 	/* Wait until tx fastpath tasks complete */
2961 	for_each_tx_queue(bp, i) {
2962 		struct bnx2x_fastpath *fp = &bp->fp[i];
2963 
2964 		for_each_cos_in_tx_queue(fp, cos)
2965 			rc = bnx2x_clean_tx_queue(bp, fp->txdata_ptr[cos]);
2966 		if (rc)
2967 			return rc;
2968 	}
2969 	return 0;
2970 }
2971 
2972 /* must be called with rtnl_lock */
2973 int bnx2x_nic_unload(struct bnx2x *bp, int unload_mode, bool keep_link)
2974 {
2975 	int i;
2976 	bool global = false;
2977 
2978 	DP(NETIF_MSG_IFUP, "Starting NIC unload\n");
2979 
2980 	if (!IS_MF_SD_STORAGE_PERSONALITY_ONLY(bp))
2981 		bnx2x_set_os_driver_state(bp, OS_DRIVER_STATE_DISABLED);
2982 
2983 	/* mark driver is unloaded in shmem2 */
2984 	if (IS_PF(bp) && SHMEM2_HAS(bp, drv_capabilities_flag)) {
2985 		u32 val;
2986 		val = SHMEM2_RD(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)]);
2987 		SHMEM2_WR(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)],
2988 			  val & ~DRV_FLAGS_CAPABILITIES_LOADED_L2);
2989 	}
2990 
2991 	if (IS_PF(bp) && bp->recovery_state != BNX2X_RECOVERY_DONE &&
2992 	    (bp->state == BNX2X_STATE_CLOSED ||
2993 	     bp->state == BNX2X_STATE_ERROR)) {
2994 		/* We can get here if the driver has been unloaded
2995 		 * during parity error recovery and is either waiting for a
2996 		 * leader to complete or for other functions to unload and
2997 		 * then ifdown has been issued. In this case we want to
2998 		 * unload and let other functions to complete a recovery
2999 		 * process.
3000 		 */
3001 		bp->recovery_state = BNX2X_RECOVERY_DONE;
3002 		bp->is_leader = 0;
3003 		bnx2x_release_leader_lock(bp);
3004 		smp_mb();
3005 
3006 		DP(NETIF_MSG_IFDOWN, "Releasing a leadership...\n");
3007 		BNX2X_ERR("Can't unload in closed or error state\n");
3008 		return -EINVAL;
3009 	}
3010 
3011 	/* Nothing to do during unload if previous bnx2x_nic_load()
3012 	 * have not completed successfully - all resources are released.
3013 	 *
3014 	 * we can get here only after unsuccessful ndo_* callback, during which
3015 	 * dev->IFF_UP flag is still on.
3016 	 */
3017 	if (bp->state == BNX2X_STATE_CLOSED || bp->state == BNX2X_STATE_ERROR)
3018 		return 0;
3019 
3020 	/* It's important to set the bp->state to the value different from
3021 	 * BNX2X_STATE_OPEN and only then stop the Tx. Otherwise bnx2x_tx_int()
3022 	 * may restart the Tx from the NAPI context (see bnx2x_tx_int()).
3023 	 */
3024 	bp->state = BNX2X_STATE_CLOSING_WAIT4_HALT;
3025 	smp_mb();
3026 
3027 	/* indicate to VFs that the PF is going down */
3028 	bnx2x_iov_channel_down(bp);
3029 
3030 	if (CNIC_LOADED(bp))
3031 		bnx2x_cnic_notify(bp, CNIC_CTL_STOP_CMD);
3032 
3033 	/* Stop Tx */
3034 	bnx2x_tx_disable(bp);
3035 	netdev_reset_tc(bp->dev);
3036 
3037 	bp->rx_mode = BNX2X_RX_MODE_NONE;
3038 
3039 	del_timer_sync(&bp->timer);
3040 
3041 	if (IS_PF(bp) && !BP_NOMCP(bp)) {
3042 		/* Set ALWAYS_ALIVE bit in shmem */
3043 		bp->fw_drv_pulse_wr_seq |= DRV_PULSE_ALWAYS_ALIVE;
3044 		bnx2x_drv_pulse(bp);
3045 		bnx2x_stats_handle(bp, STATS_EVENT_STOP);
3046 		bnx2x_save_statistics(bp);
3047 	}
3048 
3049 	/* wait till consumers catch up with producers in all queues.
3050 	 * If we're recovering, FW can't write to host so no reason
3051 	 * to wait for the queues to complete all Tx.
3052 	 */
3053 	if (unload_mode != UNLOAD_RECOVERY)
3054 		bnx2x_drain_tx_queues(bp);
3055 
3056 	/* if VF indicate to PF this function is going down (PF will delete sp
3057 	 * elements and clear initializations
3058 	 */
3059 	if (IS_VF(bp))
3060 		bnx2x_vfpf_close_vf(bp);
3061 	else if (unload_mode != UNLOAD_RECOVERY)
3062 		/* if this is a normal/close unload need to clean up chip*/
3063 		bnx2x_chip_cleanup(bp, unload_mode, keep_link);
3064 	else {
3065 		/* Send the UNLOAD_REQUEST to the MCP */
3066 		bnx2x_send_unload_req(bp, unload_mode);
3067 
3068 		/* Prevent transactions to host from the functions on the
3069 		 * engine that doesn't reset global blocks in case of global
3070 		 * attention once global blocks are reset and gates are opened
3071 		 * (the engine which leader will perform the recovery
3072 		 * last).
3073 		 */
3074 		if (!CHIP_IS_E1x(bp))
3075 			bnx2x_pf_disable(bp);
3076 
3077 		/* Disable HW interrupts, NAPI */
3078 		bnx2x_netif_stop(bp, 1);
3079 		/* Delete all NAPI objects */
3080 		bnx2x_del_all_napi(bp);
3081 		if (CNIC_LOADED(bp))
3082 			bnx2x_del_all_napi_cnic(bp);
3083 		/* Release IRQs */
3084 		bnx2x_free_irq(bp);
3085 
3086 		/* Report UNLOAD_DONE to MCP */
3087 		bnx2x_send_unload_done(bp, false);
3088 	}
3089 
3090 	/*
3091 	 * At this stage no more interrupts will arrive so we may safely clean
3092 	 * the queueable objects here in case they failed to get cleaned so far.
3093 	 */
3094 	if (IS_PF(bp))
3095 		bnx2x_squeeze_objects(bp);
3096 
3097 	/* There should be no more pending SP commands at this stage */
3098 	bp->sp_state = 0;
3099 
3100 	bp->port.pmf = 0;
3101 
3102 	/* clear pending work in rtnl task */
3103 	bp->sp_rtnl_state = 0;
3104 	smp_mb();
3105 
3106 	/* Free SKBs, SGEs, TPA pool and driver internals */
3107 	bnx2x_free_skbs(bp);
3108 	if (CNIC_LOADED(bp))
3109 		bnx2x_free_skbs_cnic(bp);
3110 	for_each_rx_queue(bp, i)
3111 		bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
3112 
3113 	bnx2x_free_fp_mem(bp);
3114 	if (CNIC_LOADED(bp))
3115 		bnx2x_free_fp_mem_cnic(bp);
3116 
3117 	if (IS_PF(bp)) {
3118 		if (CNIC_LOADED(bp))
3119 			bnx2x_free_mem_cnic(bp);
3120 	}
3121 	bnx2x_free_mem(bp);
3122 
3123 	bp->state = BNX2X_STATE_CLOSED;
3124 	bp->cnic_loaded = false;
3125 
3126 	/* Clear driver version indication in shmem */
3127 	if (IS_PF(bp) && !BP_NOMCP(bp))
3128 		bnx2x_update_mng_version(bp);
3129 
3130 	/* Check if there are pending parity attentions. If there are - set
3131 	 * RECOVERY_IN_PROGRESS.
3132 	 */
3133 	if (IS_PF(bp) && bnx2x_chk_parity_attn(bp, &global, false)) {
3134 		bnx2x_set_reset_in_progress(bp);
3135 
3136 		/* Set RESET_IS_GLOBAL if needed */
3137 		if (global)
3138 			bnx2x_set_reset_global(bp);
3139 	}
3140 
3141 	/* The last driver must disable a "close the gate" if there is no
3142 	 * parity attention or "process kill" pending.
3143 	 */
3144 	if (IS_PF(bp) &&
3145 	    !bnx2x_clear_pf_load(bp) &&
3146 	    bnx2x_reset_is_done(bp, BP_PATH(bp)))
3147 		bnx2x_disable_close_the_gate(bp);
3148 
3149 	DP(NETIF_MSG_IFUP, "Ending NIC unload\n");
3150 
3151 	return 0;
3152 }
3153 
3154 int bnx2x_set_power_state(struct bnx2x *bp, pci_power_t state)
3155 {
3156 	u16 pmcsr;
3157 
3158 	/* If there is no power capability, silently succeed */
3159 	if (!bp->pdev->pm_cap) {
3160 		BNX2X_DEV_INFO("No power capability. Breaking.\n");
3161 		return 0;
3162 	}
3163 
3164 	pci_read_config_word(bp->pdev, bp->pdev->pm_cap + PCI_PM_CTRL, &pmcsr);
3165 
3166 	switch (state) {
3167 	case PCI_D0:
3168 		pci_write_config_word(bp->pdev, bp->pdev->pm_cap + PCI_PM_CTRL,
3169 				      ((pmcsr & ~PCI_PM_CTRL_STATE_MASK) |
3170 				       PCI_PM_CTRL_PME_STATUS));
3171 
3172 		if (pmcsr & PCI_PM_CTRL_STATE_MASK)
3173 			/* delay required during transition out of D3hot */
3174 			msleep(20);
3175 		break;
3176 
3177 	case PCI_D3hot:
3178 		/* If there are other clients above don't
3179 		   shut down the power */
3180 		if (atomic_read(&bp->pdev->enable_cnt) != 1)
3181 			return 0;
3182 		/* Don't shut down the power for emulation and FPGA */
3183 		if (CHIP_REV_IS_SLOW(bp))
3184 			return 0;
3185 
3186 		pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
3187 		pmcsr |= 3;
3188 
3189 		if (bp->wol)
3190 			pmcsr |= PCI_PM_CTRL_PME_ENABLE;
3191 
3192 		pci_write_config_word(bp->pdev, bp->pdev->pm_cap + PCI_PM_CTRL,
3193 				      pmcsr);
3194 
3195 		/* No more memory access after this point until
3196 		* device is brought back to D0.
3197 		*/
3198 		break;
3199 
3200 	default:
3201 		dev_err(&bp->pdev->dev, "Can't support state = %d\n", state);
3202 		return -EINVAL;
3203 	}
3204 	return 0;
3205 }
3206 
3207 /*
3208  * net_device service functions
3209  */
3210 static int bnx2x_poll(struct napi_struct *napi, int budget)
3211 {
3212 	struct bnx2x_fastpath *fp = container_of(napi, struct bnx2x_fastpath,
3213 						 napi);
3214 	struct bnx2x *bp = fp->bp;
3215 	int rx_work_done;
3216 	u8 cos;
3217 
3218 #ifdef BNX2X_STOP_ON_ERROR
3219 	if (unlikely(bp->panic)) {
3220 		napi_complete(napi);
3221 		return 0;
3222 	}
3223 #endif
3224 	for_each_cos_in_tx_queue(fp, cos)
3225 		if (bnx2x_tx_queue_has_work(fp->txdata_ptr[cos]))
3226 			bnx2x_tx_int(bp, fp->txdata_ptr[cos]);
3227 
3228 	rx_work_done = (bnx2x_has_rx_work(fp)) ? bnx2x_rx_int(fp, budget) : 0;
3229 
3230 	if (rx_work_done < budget) {
3231 		/* No need to update SB for FCoE L2 ring as long as
3232 		 * it's connected to the default SB and the SB
3233 		 * has been updated when NAPI was scheduled.
3234 		 */
3235 		if (IS_FCOE_FP(fp)) {
3236 			napi_complete_done(napi, rx_work_done);
3237 		} else {
3238 			bnx2x_update_fpsb_idx(fp);
3239 			/* bnx2x_has_rx_work() reads the status block,
3240 			 * thus we need to ensure that status block indices
3241 			 * have been actually read (bnx2x_update_fpsb_idx)
3242 			 * prior to this check (bnx2x_has_rx_work) so that
3243 			 * we won't write the "newer" value of the status block
3244 			 * to IGU (if there was a DMA right after
3245 			 * bnx2x_has_rx_work and if there is no rmb, the memory
3246 			 * reading (bnx2x_update_fpsb_idx) may be postponed
3247 			 * to right before bnx2x_ack_sb). In this case there
3248 			 * will never be another interrupt until there is
3249 			 * another update of the status block, while there
3250 			 * is still unhandled work.
3251 			 */
3252 			rmb();
3253 
3254 			if (!(bnx2x_has_rx_work(fp) || bnx2x_has_tx_work(fp))) {
3255 				if (napi_complete_done(napi, rx_work_done)) {
3256 					/* Re-enable interrupts */
3257 					DP(NETIF_MSG_RX_STATUS,
3258 					   "Update index to %d\n", fp->fp_hc_idx);
3259 					bnx2x_ack_sb(bp, fp->igu_sb_id, USTORM_ID,
3260 						     le16_to_cpu(fp->fp_hc_idx),
3261 						     IGU_INT_ENABLE, 1);
3262 				}
3263 			} else {
3264 				rx_work_done = budget;
3265 			}
3266 		}
3267 	}
3268 
3269 	return rx_work_done;
3270 }
3271 
3272 /* we split the first BD into headers and data BDs
3273  * to ease the pain of our fellow microcode engineers
3274  * we use one mapping for both BDs
3275  */
3276 static u16 bnx2x_tx_split(struct bnx2x *bp,
3277 			  struct bnx2x_fp_txdata *txdata,
3278 			  struct sw_tx_bd *tx_buf,
3279 			  struct eth_tx_start_bd **tx_bd, u16 hlen,
3280 			  u16 bd_prod)
3281 {
3282 	struct eth_tx_start_bd *h_tx_bd = *tx_bd;
3283 	struct eth_tx_bd *d_tx_bd;
3284 	dma_addr_t mapping;
3285 	int old_len = le16_to_cpu(h_tx_bd->nbytes);
3286 
3287 	/* first fix first BD */
3288 	h_tx_bd->nbytes = cpu_to_le16(hlen);
3289 
3290 	DP(NETIF_MSG_TX_QUEUED,	"TSO split header size is %d (%x:%x)\n",
3291 	   h_tx_bd->nbytes, h_tx_bd->addr_hi, h_tx_bd->addr_lo);
3292 
3293 	/* now get a new data BD
3294 	 * (after the pbd) and fill it */
3295 	bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3296 	d_tx_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
3297 
3298 	mapping = HILO_U64(le32_to_cpu(h_tx_bd->addr_hi),
3299 			   le32_to_cpu(h_tx_bd->addr_lo)) + hlen;
3300 
3301 	d_tx_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
3302 	d_tx_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
3303 	d_tx_bd->nbytes = cpu_to_le16(old_len - hlen);
3304 
3305 	/* this marks the BD as one that has no individual mapping */
3306 	tx_buf->flags |= BNX2X_TSO_SPLIT_BD;
3307 
3308 	DP(NETIF_MSG_TX_QUEUED,
3309 	   "TSO split data size is %d (%x:%x)\n",
3310 	   d_tx_bd->nbytes, d_tx_bd->addr_hi, d_tx_bd->addr_lo);
3311 
3312 	/* update tx_bd */
3313 	*tx_bd = (struct eth_tx_start_bd *)d_tx_bd;
3314 
3315 	return bd_prod;
3316 }
3317 
3318 #define bswab32(b32) ((__force __le32) swab32((__force __u32) (b32)))
3319 #define bswab16(b16) ((__force __le16) swab16((__force __u16) (b16)))
3320 static __le16 bnx2x_csum_fix(unsigned char *t_header, u16 csum, s8 fix)
3321 {
3322 	__sum16 tsum = (__force __sum16) csum;
3323 
3324 	if (fix > 0)
3325 		tsum = ~csum_fold(csum_sub((__force __wsum) csum,
3326 				  csum_partial(t_header - fix, fix, 0)));
3327 
3328 	else if (fix < 0)
3329 		tsum = ~csum_fold(csum_add((__force __wsum) csum,
3330 				  csum_partial(t_header, -fix, 0)));
3331 
3332 	return bswab16(tsum);
3333 }
3334 
3335 static u32 bnx2x_xmit_type(struct bnx2x *bp, struct sk_buff *skb)
3336 {
3337 	u32 rc;
3338 	__u8 prot = 0;
3339 	__be16 protocol;
3340 
3341 	if (skb->ip_summed != CHECKSUM_PARTIAL)
3342 		return XMIT_PLAIN;
3343 
3344 	protocol = vlan_get_protocol(skb);
3345 	if (protocol == htons(ETH_P_IPV6)) {
3346 		rc = XMIT_CSUM_V6;
3347 		prot = ipv6_hdr(skb)->nexthdr;
3348 	} else {
3349 		rc = XMIT_CSUM_V4;
3350 		prot = ip_hdr(skb)->protocol;
3351 	}
3352 
3353 	if (!CHIP_IS_E1x(bp) && skb->encapsulation) {
3354 		if (inner_ip_hdr(skb)->version == 6) {
3355 			rc |= XMIT_CSUM_ENC_V6;
3356 			if (inner_ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3357 				rc |= XMIT_CSUM_TCP;
3358 		} else {
3359 			rc |= XMIT_CSUM_ENC_V4;
3360 			if (inner_ip_hdr(skb)->protocol == IPPROTO_TCP)
3361 				rc |= XMIT_CSUM_TCP;
3362 		}
3363 	}
3364 	if (prot == IPPROTO_TCP)
3365 		rc |= XMIT_CSUM_TCP;
3366 
3367 	if (skb_is_gso(skb)) {
3368 		if (skb_is_gso_v6(skb)) {
3369 			rc |= (XMIT_GSO_V6 | XMIT_CSUM_TCP);
3370 			if (rc & XMIT_CSUM_ENC)
3371 				rc |= XMIT_GSO_ENC_V6;
3372 		} else {
3373 			rc |= (XMIT_GSO_V4 | XMIT_CSUM_TCP);
3374 			if (rc & XMIT_CSUM_ENC)
3375 				rc |= XMIT_GSO_ENC_V4;
3376 		}
3377 	}
3378 
3379 	return rc;
3380 }
3381 
3382 /* VXLAN: 4 = 1 (for linear data BD) + 3 (2 for PBD and last BD) */
3383 #define BNX2X_NUM_VXLAN_TSO_WIN_SUB_BDS         4
3384 
3385 /* Regular: 3 = 1 (for linear data BD) + 2 (for PBD and last BD) */
3386 #define BNX2X_NUM_TSO_WIN_SUB_BDS               3
3387 
3388 #if (MAX_SKB_FRAGS >= MAX_FETCH_BD - BDS_PER_TX_PKT)
3389 /* check if packet requires linearization (packet is too fragmented)
3390    no need to check fragmentation if page size > 8K (there will be no
3391    violation to FW restrictions) */
3392 static int bnx2x_pkt_req_lin(struct bnx2x *bp, struct sk_buff *skb,
3393 			     u32 xmit_type)
3394 {
3395 	int first_bd_sz = 0, num_tso_win_sub = BNX2X_NUM_TSO_WIN_SUB_BDS;
3396 	int to_copy = 0, hlen = 0;
3397 
3398 	if (xmit_type & XMIT_GSO_ENC)
3399 		num_tso_win_sub = BNX2X_NUM_VXLAN_TSO_WIN_SUB_BDS;
3400 
3401 	if (skb_shinfo(skb)->nr_frags >= (MAX_FETCH_BD - num_tso_win_sub)) {
3402 		if (xmit_type & XMIT_GSO) {
3403 			unsigned short lso_mss = skb_shinfo(skb)->gso_size;
3404 			int wnd_size = MAX_FETCH_BD - num_tso_win_sub;
3405 			/* Number of windows to check */
3406 			int num_wnds = skb_shinfo(skb)->nr_frags - wnd_size;
3407 			int wnd_idx = 0;
3408 			int frag_idx = 0;
3409 			u32 wnd_sum = 0;
3410 
3411 			/* Headers length */
3412 			if (xmit_type & XMIT_GSO_ENC)
3413 				hlen = (int)(skb_inner_transport_header(skb) -
3414 					     skb->data) +
3415 					     inner_tcp_hdrlen(skb);
3416 			else
3417 				hlen = (int)(skb_transport_header(skb) -
3418 					     skb->data) + tcp_hdrlen(skb);
3419 
3420 			/* Amount of data (w/o headers) on linear part of SKB*/
3421 			first_bd_sz = skb_headlen(skb) - hlen;
3422 
3423 			wnd_sum  = first_bd_sz;
3424 
3425 			/* Calculate the first sum - it's special */
3426 			for (frag_idx = 0; frag_idx < wnd_size - 1; frag_idx++)
3427 				wnd_sum +=
3428 					skb_frag_size(&skb_shinfo(skb)->frags[frag_idx]);
3429 
3430 			/* If there was data on linear skb data - check it */
3431 			if (first_bd_sz > 0) {
3432 				if (unlikely(wnd_sum < lso_mss)) {
3433 					to_copy = 1;
3434 					goto exit_lbl;
3435 				}
3436 
3437 				wnd_sum -= first_bd_sz;
3438 			}
3439 
3440 			/* Others are easier: run through the frag list and
3441 			   check all windows */
3442 			for (wnd_idx = 0; wnd_idx <= num_wnds; wnd_idx++) {
3443 				wnd_sum +=
3444 			  skb_frag_size(&skb_shinfo(skb)->frags[wnd_idx + wnd_size - 1]);
3445 
3446 				if (unlikely(wnd_sum < lso_mss)) {
3447 					to_copy = 1;
3448 					break;
3449 				}
3450 				wnd_sum -=
3451 					skb_frag_size(&skb_shinfo(skb)->frags[wnd_idx]);
3452 			}
3453 		} else {
3454 			/* in non-LSO too fragmented packet should always
3455 			   be linearized */
3456 			to_copy = 1;
3457 		}
3458 	}
3459 
3460 exit_lbl:
3461 	if (unlikely(to_copy))
3462 		DP(NETIF_MSG_TX_QUEUED,
3463 		   "Linearization IS REQUIRED for %s packet. num_frags %d  hlen %d  first_bd_sz %d\n",
3464 		   (xmit_type & XMIT_GSO) ? "LSO" : "non-LSO",
3465 		   skb_shinfo(skb)->nr_frags, hlen, first_bd_sz);
3466 
3467 	return to_copy;
3468 }
3469 #endif
3470 
3471 /**
3472  * bnx2x_set_pbd_gso - update PBD in GSO case.
3473  *
3474  * @skb:	packet skb
3475  * @pbd:	parse BD
3476  * @xmit_type:	xmit flags
3477  */
3478 static void bnx2x_set_pbd_gso(struct sk_buff *skb,
3479 			      struct eth_tx_parse_bd_e1x *pbd,
3480 			      u32 xmit_type)
3481 {
3482 	pbd->lso_mss = cpu_to_le16(skb_shinfo(skb)->gso_size);
3483 	pbd->tcp_send_seq = bswab32(tcp_hdr(skb)->seq);
3484 	pbd->tcp_flags = pbd_tcp_flags(tcp_hdr(skb));
3485 
3486 	if (xmit_type & XMIT_GSO_V4) {
3487 		pbd->ip_id = bswab16(ip_hdr(skb)->id);
3488 		pbd->tcp_pseudo_csum =
3489 			bswab16(~csum_tcpudp_magic(ip_hdr(skb)->saddr,
3490 						   ip_hdr(skb)->daddr,
3491 						   0, IPPROTO_TCP, 0));
3492 	} else {
3493 		pbd->tcp_pseudo_csum =
3494 			bswab16(~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3495 						 &ipv6_hdr(skb)->daddr,
3496 						 0, IPPROTO_TCP, 0));
3497 	}
3498 
3499 	pbd->global_data |=
3500 		cpu_to_le16(ETH_TX_PARSE_BD_E1X_PSEUDO_CS_WITHOUT_LEN);
3501 }
3502 
3503 /**
3504  * bnx2x_set_pbd_csum_enc - update PBD with checksum and return header length
3505  *
3506  * @bp:			driver handle
3507  * @skb:		packet skb
3508  * @parsing_data:	data to be updated
3509  * @xmit_type:		xmit flags
3510  *
3511  * 57712/578xx related, when skb has encapsulation
3512  */
3513 static u8 bnx2x_set_pbd_csum_enc(struct bnx2x *bp, struct sk_buff *skb,
3514 				 u32 *parsing_data, u32 xmit_type)
3515 {
3516 	*parsing_data |=
3517 		((((u8 *)skb_inner_transport_header(skb) - skb->data) >> 1) <<
3518 		ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W_SHIFT) &
3519 		ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W;
3520 
3521 	if (xmit_type & XMIT_CSUM_TCP) {
3522 		*parsing_data |= ((inner_tcp_hdrlen(skb) / 4) <<
3523 			ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW_SHIFT) &
3524 			ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW;
3525 
3526 		return skb_inner_transport_header(skb) +
3527 			inner_tcp_hdrlen(skb) - skb->data;
3528 	}
3529 
3530 	/* We support checksum offload for TCP and UDP only.
3531 	 * No need to pass the UDP header length - it's a constant.
3532 	 */
3533 	return skb_inner_transport_header(skb) +
3534 		sizeof(struct udphdr) - skb->data;
3535 }
3536 
3537 /**
3538  * bnx2x_set_pbd_csum_e2 - update PBD with checksum and return header length
3539  *
3540  * @bp:			driver handle
3541  * @skb:		packet skb
3542  * @parsing_data:	data to be updated
3543  * @xmit_type:		xmit flags
3544  *
3545  * 57712/578xx related
3546  */
3547 static u8 bnx2x_set_pbd_csum_e2(struct bnx2x *bp, struct sk_buff *skb,
3548 				u32 *parsing_data, u32 xmit_type)
3549 {
3550 	*parsing_data |=
3551 		((((u8 *)skb_transport_header(skb) - skb->data) >> 1) <<
3552 		ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W_SHIFT) &
3553 		ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W;
3554 
3555 	if (xmit_type & XMIT_CSUM_TCP) {
3556 		*parsing_data |= ((tcp_hdrlen(skb) / 4) <<
3557 			ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW_SHIFT) &
3558 			ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW;
3559 
3560 		return skb_transport_header(skb) + tcp_hdrlen(skb) - skb->data;
3561 	}
3562 	/* We support checksum offload for TCP and UDP only.
3563 	 * No need to pass the UDP header length - it's a constant.
3564 	 */
3565 	return skb_transport_header(skb) + sizeof(struct udphdr) - skb->data;
3566 }
3567 
3568 /* set FW indication according to inner or outer protocols if tunneled */
3569 static void bnx2x_set_sbd_csum(struct bnx2x *bp, struct sk_buff *skb,
3570 			       struct eth_tx_start_bd *tx_start_bd,
3571 			       u32 xmit_type)
3572 {
3573 	tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_L4_CSUM;
3574 
3575 	if (xmit_type & (XMIT_CSUM_ENC_V6 | XMIT_CSUM_V6))
3576 		tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_IPV6;
3577 
3578 	if (!(xmit_type & XMIT_CSUM_TCP))
3579 		tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_IS_UDP;
3580 }
3581 
3582 /**
3583  * bnx2x_set_pbd_csum - update PBD with checksum and return header length
3584  *
3585  * @bp:		driver handle
3586  * @skb:	packet skb
3587  * @pbd:	parse BD to be updated
3588  * @xmit_type:	xmit flags
3589  */
3590 static u8 bnx2x_set_pbd_csum(struct bnx2x *bp, struct sk_buff *skb,
3591 			     struct eth_tx_parse_bd_e1x *pbd,
3592 			     u32 xmit_type)
3593 {
3594 	u8 hlen = (skb_network_header(skb) - skb->data) >> 1;
3595 
3596 	/* for now NS flag is not used in Linux */
3597 	pbd->global_data =
3598 		cpu_to_le16(hlen |
3599 			    ((skb->protocol == cpu_to_be16(ETH_P_8021Q)) <<
3600 			     ETH_TX_PARSE_BD_E1X_LLC_SNAP_EN_SHIFT));
3601 
3602 	pbd->ip_hlen_w = (skb_transport_header(skb) -
3603 			skb_network_header(skb)) >> 1;
3604 
3605 	hlen += pbd->ip_hlen_w;
3606 
3607 	/* We support checksum offload for TCP and UDP only */
3608 	if (xmit_type & XMIT_CSUM_TCP)
3609 		hlen += tcp_hdrlen(skb) / 2;
3610 	else
3611 		hlen += sizeof(struct udphdr) / 2;
3612 
3613 	pbd->total_hlen_w = cpu_to_le16(hlen);
3614 	hlen = hlen*2;
3615 
3616 	if (xmit_type & XMIT_CSUM_TCP) {
3617 		pbd->tcp_pseudo_csum = bswab16(tcp_hdr(skb)->check);
3618 
3619 	} else {
3620 		s8 fix = SKB_CS_OFF(skb); /* signed! */
3621 
3622 		DP(NETIF_MSG_TX_QUEUED,
3623 		   "hlen %d  fix %d  csum before fix %x\n",
3624 		   le16_to_cpu(pbd->total_hlen_w), fix, SKB_CS(skb));
3625 
3626 		/* HW bug: fixup the CSUM */
3627 		pbd->tcp_pseudo_csum =
3628 			bnx2x_csum_fix(skb_transport_header(skb),
3629 				       SKB_CS(skb), fix);
3630 
3631 		DP(NETIF_MSG_TX_QUEUED, "csum after fix %x\n",
3632 		   pbd->tcp_pseudo_csum);
3633 	}
3634 
3635 	return hlen;
3636 }
3637 
3638 static void bnx2x_update_pbds_gso_enc(struct sk_buff *skb,
3639 				      struct eth_tx_parse_bd_e2 *pbd_e2,
3640 				      struct eth_tx_parse_2nd_bd *pbd2,
3641 				      u16 *global_data,
3642 				      u32 xmit_type)
3643 {
3644 	u16 hlen_w = 0;
3645 	u8 outerip_off, outerip_len = 0;
3646 
3647 	/* from outer IP to transport */
3648 	hlen_w = (skb_inner_transport_header(skb) -
3649 		  skb_network_header(skb)) >> 1;
3650 
3651 	/* transport len */
3652 	hlen_w += inner_tcp_hdrlen(skb) >> 1;
3653 
3654 	pbd2->fw_ip_hdr_to_payload_w = hlen_w;
3655 
3656 	/* outer IP header info */
3657 	if (xmit_type & XMIT_CSUM_V4) {
3658 		struct iphdr *iph = ip_hdr(skb);
3659 		u32 csum = (__force u32)(~iph->check) -
3660 			   (__force u32)iph->tot_len -
3661 			   (__force u32)iph->frag_off;
3662 
3663 		outerip_len = iph->ihl << 1;
3664 
3665 		pbd2->fw_ip_csum_wo_len_flags_frag =
3666 			bswab16(csum_fold((__force __wsum)csum));
3667 	} else {
3668 		pbd2->fw_ip_hdr_to_payload_w =
3669 			hlen_w - ((sizeof(struct ipv6hdr)) >> 1);
3670 		pbd_e2->data.tunnel_data.flags |=
3671 			ETH_TUNNEL_DATA_IPV6_OUTER;
3672 	}
3673 
3674 	pbd2->tcp_send_seq = bswab32(inner_tcp_hdr(skb)->seq);
3675 
3676 	pbd2->tcp_flags = pbd_tcp_flags(inner_tcp_hdr(skb));
3677 
3678 	/* inner IP header info */
3679 	if (xmit_type & XMIT_CSUM_ENC_V4) {
3680 		pbd2->hw_ip_id = bswab16(inner_ip_hdr(skb)->id);
3681 
3682 		pbd_e2->data.tunnel_data.pseudo_csum =
3683 			bswab16(~csum_tcpudp_magic(
3684 					inner_ip_hdr(skb)->saddr,
3685 					inner_ip_hdr(skb)->daddr,
3686 					0, IPPROTO_TCP, 0));
3687 	} else {
3688 		pbd_e2->data.tunnel_data.pseudo_csum =
3689 			bswab16(~csum_ipv6_magic(
3690 					&inner_ipv6_hdr(skb)->saddr,
3691 					&inner_ipv6_hdr(skb)->daddr,
3692 					0, IPPROTO_TCP, 0));
3693 	}
3694 
3695 	outerip_off = (skb_network_header(skb) - skb->data) >> 1;
3696 
3697 	*global_data |=
3698 		outerip_off |
3699 		(outerip_len <<
3700 			ETH_TX_PARSE_2ND_BD_IP_HDR_LEN_OUTER_W_SHIFT) |
3701 		((skb->protocol == cpu_to_be16(ETH_P_8021Q)) <<
3702 			ETH_TX_PARSE_2ND_BD_LLC_SNAP_EN_SHIFT);
3703 
3704 	if (ip_hdr(skb)->protocol == IPPROTO_UDP) {
3705 		SET_FLAG(*global_data, ETH_TX_PARSE_2ND_BD_TUNNEL_UDP_EXIST, 1);
3706 		pbd2->tunnel_udp_hdr_start_w = skb_transport_offset(skb) >> 1;
3707 	}
3708 }
3709 
3710 static inline void bnx2x_set_ipv6_ext_e2(struct sk_buff *skb, u32 *parsing_data,
3711 					 u32 xmit_type)
3712 {
3713 	struct ipv6hdr *ipv6;
3714 
3715 	if (!(xmit_type & (XMIT_GSO_ENC_V6 | XMIT_GSO_V6)))
3716 		return;
3717 
3718 	if (xmit_type & XMIT_GSO_ENC_V6)
3719 		ipv6 = inner_ipv6_hdr(skb);
3720 	else /* XMIT_GSO_V6 */
3721 		ipv6 = ipv6_hdr(skb);
3722 
3723 	if (ipv6->nexthdr == NEXTHDR_IPV6)
3724 		*parsing_data |= ETH_TX_PARSE_BD_E2_IPV6_WITH_EXT_HDR;
3725 }
3726 
3727 /* called with netif_tx_lock
3728  * bnx2x_tx_int() runs without netif_tx_lock unless it needs to call
3729  * netif_wake_queue()
3730  */
3731 netdev_tx_t bnx2x_start_xmit(struct sk_buff *skb, struct net_device *dev)
3732 {
3733 	struct bnx2x *bp = netdev_priv(dev);
3734 
3735 	struct netdev_queue *txq;
3736 	struct bnx2x_fp_txdata *txdata;
3737 	struct sw_tx_bd *tx_buf;
3738 	struct eth_tx_start_bd *tx_start_bd, *first_bd;
3739 	struct eth_tx_bd *tx_data_bd, *total_pkt_bd = NULL;
3740 	struct eth_tx_parse_bd_e1x *pbd_e1x = NULL;
3741 	struct eth_tx_parse_bd_e2 *pbd_e2 = NULL;
3742 	struct eth_tx_parse_2nd_bd *pbd2 = NULL;
3743 	u32 pbd_e2_parsing_data = 0;
3744 	u16 pkt_prod, bd_prod;
3745 	int nbd, txq_index;
3746 	dma_addr_t mapping;
3747 	u32 xmit_type = bnx2x_xmit_type(bp, skb);
3748 	int i;
3749 	u8 hlen = 0;
3750 	__le16 pkt_size = 0;
3751 	struct ethhdr *eth;
3752 	u8 mac_type = UNICAST_ADDRESS;
3753 
3754 #ifdef BNX2X_STOP_ON_ERROR
3755 	if (unlikely(bp->panic))
3756 		return NETDEV_TX_BUSY;
3757 #endif
3758 
3759 	txq_index = skb_get_queue_mapping(skb);
3760 	txq = netdev_get_tx_queue(dev, txq_index);
3761 
3762 	BUG_ON(txq_index >= MAX_ETH_TXQ_IDX(bp) + (CNIC_LOADED(bp) ? 1 : 0));
3763 
3764 	txdata = &bp->bnx2x_txq[txq_index];
3765 
3766 	/* enable this debug print to view the transmission queue being used
3767 	DP(NETIF_MSG_TX_QUEUED, "indices: txq %d, fp %d, txdata %d\n",
3768 	   txq_index, fp_index, txdata_index); */
3769 
3770 	/* enable this debug print to view the transmission details
3771 	DP(NETIF_MSG_TX_QUEUED,
3772 	   "transmitting packet cid %d fp index %d txdata_index %d tx_data ptr %p fp pointer %p\n",
3773 	   txdata->cid, fp_index, txdata_index, txdata, fp); */
3774 
3775 	if (unlikely(bnx2x_tx_avail(bp, txdata) <
3776 			skb_shinfo(skb)->nr_frags +
3777 			BDS_PER_TX_PKT +
3778 			NEXT_CNT_PER_TX_PKT(MAX_BDS_PER_TX_PKT))) {
3779 		/* Handle special storage cases separately */
3780 		if (txdata->tx_ring_size == 0) {
3781 			struct bnx2x_eth_q_stats *q_stats =
3782 				bnx2x_fp_qstats(bp, txdata->parent_fp);
3783 			q_stats->driver_filtered_tx_pkt++;
3784 			dev_kfree_skb(skb);
3785 			return NETDEV_TX_OK;
3786 		}
3787 		bnx2x_fp_qstats(bp, txdata->parent_fp)->driver_xoff++;
3788 		netif_tx_stop_queue(txq);
3789 		BNX2X_ERR("BUG! Tx ring full when queue awake!\n");
3790 
3791 		return NETDEV_TX_BUSY;
3792 	}
3793 
3794 	DP(NETIF_MSG_TX_QUEUED,
3795 	   "queue[%d]: SKB: summed %x  protocol %x protocol(%x,%x) gso type %x  xmit_type %x len %d\n",
3796 	   txq_index, skb->ip_summed, skb->protocol, ipv6_hdr(skb)->nexthdr,
3797 	   ip_hdr(skb)->protocol, skb_shinfo(skb)->gso_type, xmit_type,
3798 	   skb->len);
3799 
3800 	eth = (struct ethhdr *)skb->data;
3801 
3802 	/* set flag according to packet type (UNICAST_ADDRESS is default)*/
3803 	if (unlikely(is_multicast_ether_addr(eth->h_dest))) {
3804 		if (is_broadcast_ether_addr(eth->h_dest))
3805 			mac_type = BROADCAST_ADDRESS;
3806 		else
3807 			mac_type = MULTICAST_ADDRESS;
3808 	}
3809 
3810 #if (MAX_SKB_FRAGS >= MAX_FETCH_BD - BDS_PER_TX_PKT)
3811 	/* First, check if we need to linearize the skb (due to FW
3812 	   restrictions). No need to check fragmentation if page size > 8K
3813 	   (there will be no violation to FW restrictions) */
3814 	if (bnx2x_pkt_req_lin(bp, skb, xmit_type)) {
3815 		/* Statistics of linearization */
3816 		bp->lin_cnt++;
3817 		if (skb_linearize(skb) != 0) {
3818 			DP(NETIF_MSG_TX_QUEUED,
3819 			   "SKB linearization failed - silently dropping this SKB\n");
3820 			dev_kfree_skb_any(skb);
3821 			return NETDEV_TX_OK;
3822 		}
3823 	}
3824 #endif
3825 	/* Map skb linear data for DMA */
3826 	mapping = dma_map_single(&bp->pdev->dev, skb->data,
3827 				 skb_headlen(skb), DMA_TO_DEVICE);
3828 	if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
3829 		DP(NETIF_MSG_TX_QUEUED,
3830 		   "SKB mapping failed - silently dropping this SKB\n");
3831 		dev_kfree_skb_any(skb);
3832 		return NETDEV_TX_OK;
3833 	}
3834 	/*
3835 	Please read carefully. First we use one BD which we mark as start,
3836 	then we have a parsing info BD (used for TSO or xsum),
3837 	and only then we have the rest of the TSO BDs.
3838 	(don't forget to mark the last one as last,
3839 	and to unmap only AFTER you write to the BD ...)
3840 	And above all, all pdb sizes are in words - NOT DWORDS!
3841 	*/
3842 
3843 	/* get current pkt produced now - advance it just before sending packet
3844 	 * since mapping of pages may fail and cause packet to be dropped
3845 	 */
3846 	pkt_prod = txdata->tx_pkt_prod;
3847 	bd_prod = TX_BD(txdata->tx_bd_prod);
3848 
3849 	/* get a tx_buf and first BD
3850 	 * tx_start_bd may be changed during SPLIT,
3851 	 * but first_bd will always stay first
3852 	 */
3853 	tx_buf = &txdata->tx_buf_ring[TX_BD(pkt_prod)];
3854 	tx_start_bd = &txdata->tx_desc_ring[bd_prod].start_bd;
3855 	first_bd = tx_start_bd;
3856 
3857 	tx_start_bd->bd_flags.as_bitfield = ETH_TX_BD_FLAGS_START_BD;
3858 
3859 	if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) {
3860 		if (!(bp->flags & TX_TIMESTAMPING_EN)) {
3861 			BNX2X_ERR("Tx timestamping was not enabled, this packet will not be timestamped\n");
3862 		} else if (bp->ptp_tx_skb) {
3863 			BNX2X_ERR("The device supports only a single outstanding packet to timestamp, this packet will not be timestamped\n");
3864 		} else {
3865 			skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
3866 			/* schedule check for Tx timestamp */
3867 			bp->ptp_tx_skb = skb_get(skb);
3868 			bp->ptp_tx_start = jiffies;
3869 			schedule_work(&bp->ptp_task);
3870 		}
3871 	}
3872 
3873 	/* header nbd: indirectly zero other flags! */
3874 	tx_start_bd->general_data = 1 << ETH_TX_START_BD_HDR_NBDS_SHIFT;
3875 
3876 	/* remember the first BD of the packet */
3877 	tx_buf->first_bd = txdata->tx_bd_prod;
3878 	tx_buf->skb = skb;
3879 	tx_buf->flags = 0;
3880 
3881 	DP(NETIF_MSG_TX_QUEUED,
3882 	   "sending pkt %u @%p  next_idx %u  bd %u @%p\n",
3883 	   pkt_prod, tx_buf, txdata->tx_pkt_prod, bd_prod, tx_start_bd);
3884 
3885 	if (skb_vlan_tag_present(skb)) {
3886 		tx_start_bd->vlan_or_ethertype =
3887 		    cpu_to_le16(skb_vlan_tag_get(skb));
3888 		tx_start_bd->bd_flags.as_bitfield |=
3889 		    (X_ETH_OUTBAND_VLAN << ETH_TX_BD_FLAGS_VLAN_MODE_SHIFT);
3890 	} else {
3891 		/* when transmitting in a vf, start bd must hold the ethertype
3892 		 * for fw to enforce it
3893 		 */
3894 		u16 vlan_tci = 0;
3895 #ifndef BNX2X_STOP_ON_ERROR
3896 		if (IS_VF(bp)) {
3897 #endif
3898 			/* Still need to consider inband vlan for enforced */
3899 			if (__vlan_get_tag(skb, &vlan_tci)) {
3900 				tx_start_bd->vlan_or_ethertype =
3901 					cpu_to_le16(ntohs(eth->h_proto));
3902 			} else {
3903 				tx_start_bd->bd_flags.as_bitfield |=
3904 					(X_ETH_INBAND_VLAN <<
3905 					 ETH_TX_BD_FLAGS_VLAN_MODE_SHIFT);
3906 				tx_start_bd->vlan_or_ethertype =
3907 					cpu_to_le16(vlan_tci);
3908 			}
3909 #ifndef BNX2X_STOP_ON_ERROR
3910 		} else {
3911 			/* used by FW for packet accounting */
3912 			tx_start_bd->vlan_or_ethertype = cpu_to_le16(pkt_prod);
3913 		}
3914 #endif
3915 	}
3916 
3917 	nbd = 2; /* start_bd + pbd + frags (updated when pages are mapped) */
3918 
3919 	/* turn on parsing and get a BD */
3920 	bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3921 
3922 	if (xmit_type & XMIT_CSUM)
3923 		bnx2x_set_sbd_csum(bp, skb, tx_start_bd, xmit_type);
3924 
3925 	if (!CHIP_IS_E1x(bp)) {
3926 		pbd_e2 = &txdata->tx_desc_ring[bd_prod].parse_bd_e2;
3927 		memset(pbd_e2, 0, sizeof(struct eth_tx_parse_bd_e2));
3928 
3929 		if (xmit_type & XMIT_CSUM_ENC) {
3930 			u16 global_data = 0;
3931 
3932 			/* Set PBD in enc checksum offload case */
3933 			hlen = bnx2x_set_pbd_csum_enc(bp, skb,
3934 						      &pbd_e2_parsing_data,
3935 						      xmit_type);
3936 
3937 			/* turn on 2nd parsing and get a BD */
3938 			bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3939 
3940 			pbd2 = &txdata->tx_desc_ring[bd_prod].parse_2nd_bd;
3941 
3942 			memset(pbd2, 0, sizeof(*pbd2));
3943 
3944 			pbd_e2->data.tunnel_data.ip_hdr_start_inner_w =
3945 				(skb_inner_network_header(skb) -
3946 				 skb->data) >> 1;
3947 
3948 			if (xmit_type & XMIT_GSO_ENC)
3949 				bnx2x_update_pbds_gso_enc(skb, pbd_e2, pbd2,
3950 							  &global_data,
3951 							  xmit_type);
3952 
3953 			pbd2->global_data = cpu_to_le16(global_data);
3954 
3955 			/* add addition parse BD indication to start BD */
3956 			SET_FLAG(tx_start_bd->general_data,
3957 				 ETH_TX_START_BD_PARSE_NBDS, 1);
3958 			/* set encapsulation flag in start BD */
3959 			SET_FLAG(tx_start_bd->general_data,
3960 				 ETH_TX_START_BD_TUNNEL_EXIST, 1);
3961 
3962 			tx_buf->flags |= BNX2X_HAS_SECOND_PBD;
3963 
3964 			nbd++;
3965 		} else if (xmit_type & XMIT_CSUM) {
3966 			/* Set PBD in checksum offload case w/o encapsulation */
3967 			hlen = bnx2x_set_pbd_csum_e2(bp, skb,
3968 						     &pbd_e2_parsing_data,
3969 						     xmit_type);
3970 		}
3971 
3972 		bnx2x_set_ipv6_ext_e2(skb, &pbd_e2_parsing_data, xmit_type);
3973 		/* Add the macs to the parsing BD if this is a vf or if
3974 		 * Tx Switching is enabled.
3975 		 */
3976 		if (IS_VF(bp)) {
3977 			/* override GRE parameters in BD */
3978 			bnx2x_set_fw_mac_addr(&pbd_e2->data.mac_addr.src_hi,
3979 					      &pbd_e2->data.mac_addr.src_mid,
3980 					      &pbd_e2->data.mac_addr.src_lo,
3981 					      eth->h_source);
3982 
3983 			bnx2x_set_fw_mac_addr(&pbd_e2->data.mac_addr.dst_hi,
3984 					      &pbd_e2->data.mac_addr.dst_mid,
3985 					      &pbd_e2->data.mac_addr.dst_lo,
3986 					      eth->h_dest);
3987 		} else {
3988 			if (bp->flags & TX_SWITCHING)
3989 				bnx2x_set_fw_mac_addr(
3990 						&pbd_e2->data.mac_addr.dst_hi,
3991 						&pbd_e2->data.mac_addr.dst_mid,
3992 						&pbd_e2->data.mac_addr.dst_lo,
3993 						eth->h_dest);
3994 #ifdef BNX2X_STOP_ON_ERROR
3995 			/* Enforce security is always set in Stop on Error -
3996 			 * source mac should be present in the parsing BD
3997 			 */
3998 			bnx2x_set_fw_mac_addr(&pbd_e2->data.mac_addr.src_hi,
3999 					      &pbd_e2->data.mac_addr.src_mid,
4000 					      &pbd_e2->data.mac_addr.src_lo,
4001 					      eth->h_source);
4002 #endif
4003 		}
4004 
4005 		SET_FLAG(pbd_e2_parsing_data,
4006 			 ETH_TX_PARSE_BD_E2_ETH_ADDR_TYPE, mac_type);
4007 	} else {
4008 		u16 global_data = 0;
4009 		pbd_e1x = &txdata->tx_desc_ring[bd_prod].parse_bd_e1x;
4010 		memset(pbd_e1x, 0, sizeof(struct eth_tx_parse_bd_e1x));
4011 		/* Set PBD in checksum offload case */
4012 		if (xmit_type & XMIT_CSUM)
4013 			hlen = bnx2x_set_pbd_csum(bp, skb, pbd_e1x, xmit_type);
4014 
4015 		SET_FLAG(global_data,
4016 			 ETH_TX_PARSE_BD_E1X_ETH_ADDR_TYPE, mac_type);
4017 		pbd_e1x->global_data |= cpu_to_le16(global_data);
4018 	}
4019 
4020 	/* Setup the data pointer of the first BD of the packet */
4021 	tx_start_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
4022 	tx_start_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
4023 	tx_start_bd->nbytes = cpu_to_le16(skb_headlen(skb));
4024 	pkt_size = tx_start_bd->nbytes;
4025 
4026 	DP(NETIF_MSG_TX_QUEUED,
4027 	   "first bd @%p  addr (%x:%x)  nbytes %d  flags %x  vlan %x\n",
4028 	   tx_start_bd, tx_start_bd->addr_hi, tx_start_bd->addr_lo,
4029 	   le16_to_cpu(tx_start_bd->nbytes),
4030 	   tx_start_bd->bd_flags.as_bitfield,
4031 	   le16_to_cpu(tx_start_bd->vlan_or_ethertype));
4032 
4033 	if (xmit_type & XMIT_GSO) {
4034 
4035 		DP(NETIF_MSG_TX_QUEUED,
4036 		   "TSO packet len %d  hlen %d  total len %d  tso size %d\n",
4037 		   skb->len, hlen, skb_headlen(skb),
4038 		   skb_shinfo(skb)->gso_size);
4039 
4040 		tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_SW_LSO;
4041 
4042 		if (unlikely(skb_headlen(skb) > hlen)) {
4043 			nbd++;
4044 			bd_prod = bnx2x_tx_split(bp, txdata, tx_buf,
4045 						 &tx_start_bd, hlen,
4046 						 bd_prod);
4047 		}
4048 		if (!CHIP_IS_E1x(bp))
4049 			pbd_e2_parsing_data |=
4050 				(skb_shinfo(skb)->gso_size <<
4051 				 ETH_TX_PARSE_BD_E2_LSO_MSS_SHIFT) &
4052 				 ETH_TX_PARSE_BD_E2_LSO_MSS;
4053 		else
4054 			bnx2x_set_pbd_gso(skb, pbd_e1x, xmit_type);
4055 	}
4056 
4057 	/* Set the PBD's parsing_data field if not zero
4058 	 * (for the chips newer than 57711).
4059 	 */
4060 	if (pbd_e2_parsing_data)
4061 		pbd_e2->parsing_data = cpu_to_le32(pbd_e2_parsing_data);
4062 
4063 	tx_data_bd = (struct eth_tx_bd *)tx_start_bd;
4064 
4065 	/* Handle fragmented skb */
4066 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
4067 		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
4068 
4069 		mapping = skb_frag_dma_map(&bp->pdev->dev, frag, 0,
4070 					   skb_frag_size(frag), DMA_TO_DEVICE);
4071 		if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
4072 			unsigned int pkts_compl = 0, bytes_compl = 0;
4073 
4074 			DP(NETIF_MSG_TX_QUEUED,
4075 			   "Unable to map page - dropping packet...\n");
4076 
4077 			/* we need unmap all buffers already mapped
4078 			 * for this SKB;
4079 			 * first_bd->nbd need to be properly updated
4080 			 * before call to bnx2x_free_tx_pkt
4081 			 */
4082 			first_bd->nbd = cpu_to_le16(nbd);
4083 			bnx2x_free_tx_pkt(bp, txdata,
4084 					  TX_BD(txdata->tx_pkt_prod),
4085 					  &pkts_compl, &bytes_compl);
4086 			return NETDEV_TX_OK;
4087 		}
4088 
4089 		bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
4090 		tx_data_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
4091 		if (total_pkt_bd == NULL)
4092 			total_pkt_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
4093 
4094 		tx_data_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
4095 		tx_data_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
4096 		tx_data_bd->nbytes = cpu_to_le16(skb_frag_size(frag));
4097 		le16_add_cpu(&pkt_size, skb_frag_size(frag));
4098 		nbd++;
4099 
4100 		DP(NETIF_MSG_TX_QUEUED,
4101 		   "frag %d  bd @%p  addr (%x:%x)  nbytes %d\n",
4102 		   i, tx_data_bd, tx_data_bd->addr_hi, tx_data_bd->addr_lo,
4103 		   le16_to_cpu(tx_data_bd->nbytes));
4104 	}
4105 
4106 	DP(NETIF_MSG_TX_QUEUED, "last bd @%p\n", tx_data_bd);
4107 
4108 	/* update with actual num BDs */
4109 	first_bd->nbd = cpu_to_le16(nbd);
4110 
4111 	bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
4112 
4113 	/* now send a tx doorbell, counting the next BD
4114 	 * if the packet contains or ends with it
4115 	 */
4116 	if (TX_BD_POFF(bd_prod) < nbd)
4117 		nbd++;
4118 
4119 	/* total_pkt_bytes should be set on the first data BD if
4120 	 * it's not an LSO packet and there is more than one
4121 	 * data BD. In this case pkt_size is limited by an MTU value.
4122 	 * However we prefer to set it for an LSO packet (while we don't
4123 	 * have to) in order to save some CPU cycles in a none-LSO
4124 	 * case, when we much more care about them.
4125 	 */
4126 	if (total_pkt_bd != NULL)
4127 		total_pkt_bd->total_pkt_bytes = pkt_size;
4128 
4129 	if (pbd_e1x)
4130 		DP(NETIF_MSG_TX_QUEUED,
4131 		   "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",
4132 		   pbd_e1x, pbd_e1x->global_data, pbd_e1x->ip_hlen_w,
4133 		   pbd_e1x->ip_id, pbd_e1x->lso_mss, pbd_e1x->tcp_flags,
4134 		   pbd_e1x->tcp_pseudo_csum, pbd_e1x->tcp_send_seq,
4135 		    le16_to_cpu(pbd_e1x->total_hlen_w));
4136 	if (pbd_e2)
4137 		DP(NETIF_MSG_TX_QUEUED,
4138 		   "PBD (E2) @%p  dst %x %x %x src %x %x %x parsing_data %x\n",
4139 		   pbd_e2,
4140 		   pbd_e2->data.mac_addr.dst_hi,
4141 		   pbd_e2->data.mac_addr.dst_mid,
4142 		   pbd_e2->data.mac_addr.dst_lo,
4143 		   pbd_e2->data.mac_addr.src_hi,
4144 		   pbd_e2->data.mac_addr.src_mid,
4145 		   pbd_e2->data.mac_addr.src_lo,
4146 		   pbd_e2->parsing_data);
4147 	DP(NETIF_MSG_TX_QUEUED, "doorbell: nbd %d  bd %u\n", nbd, bd_prod);
4148 
4149 	netdev_tx_sent_queue(txq, skb->len);
4150 
4151 	skb_tx_timestamp(skb);
4152 
4153 	txdata->tx_pkt_prod++;
4154 	/*
4155 	 * Make sure that the BD data is updated before updating the producer
4156 	 * since FW might read the BD right after the producer is updated.
4157 	 * This is only applicable for weak-ordered memory model archs such
4158 	 * as IA-64. The following barrier is also mandatory since FW will
4159 	 * assumes packets must have BDs.
4160 	 */
4161 	wmb();
4162 
4163 	txdata->tx_db.data.prod += nbd;
4164 	/* make sure descriptor update is observed by HW */
4165 	wmb();
4166 
4167 	DOORBELL_RELAXED(bp, txdata->cid, txdata->tx_db.raw);
4168 
4169 	mmiowb();
4170 
4171 	txdata->tx_bd_prod += nbd;
4172 
4173 	if (unlikely(bnx2x_tx_avail(bp, txdata) < MAX_DESC_PER_TX_PKT)) {
4174 		netif_tx_stop_queue(txq);
4175 
4176 		/* paired memory barrier is in bnx2x_tx_int(), we have to keep
4177 		 * ordering of set_bit() in netif_tx_stop_queue() and read of
4178 		 * fp->bd_tx_cons */
4179 		smp_mb();
4180 
4181 		bnx2x_fp_qstats(bp, txdata->parent_fp)->driver_xoff++;
4182 		if (bnx2x_tx_avail(bp, txdata) >= MAX_DESC_PER_TX_PKT)
4183 			netif_tx_wake_queue(txq);
4184 	}
4185 	txdata->tx_pkt++;
4186 
4187 	return NETDEV_TX_OK;
4188 }
4189 
4190 void bnx2x_get_c2s_mapping(struct bnx2x *bp, u8 *c2s_map, u8 *c2s_default)
4191 {
4192 	int mfw_vn = BP_FW_MB_IDX(bp);
4193 	u32 tmp;
4194 
4195 	/* If the shmem shouldn't affect configuration, reflect */
4196 	if (!IS_MF_BD(bp)) {
4197 		int i;
4198 
4199 		for (i = 0; i < BNX2X_MAX_PRIORITY; i++)
4200 			c2s_map[i] = i;
4201 		*c2s_default = 0;
4202 
4203 		return;
4204 	}
4205 
4206 	tmp = SHMEM2_RD(bp, c2s_pcp_map_lower[mfw_vn]);
4207 	tmp = (__force u32)be32_to_cpu((__force __be32)tmp);
4208 	c2s_map[0] = tmp & 0xff;
4209 	c2s_map[1] = (tmp >> 8) & 0xff;
4210 	c2s_map[2] = (tmp >> 16) & 0xff;
4211 	c2s_map[3] = (tmp >> 24) & 0xff;
4212 
4213 	tmp = SHMEM2_RD(bp, c2s_pcp_map_upper[mfw_vn]);
4214 	tmp = (__force u32)be32_to_cpu((__force __be32)tmp);
4215 	c2s_map[4] = tmp & 0xff;
4216 	c2s_map[5] = (tmp >> 8) & 0xff;
4217 	c2s_map[6] = (tmp >> 16) & 0xff;
4218 	c2s_map[7] = (tmp >> 24) & 0xff;
4219 
4220 	tmp = SHMEM2_RD(bp, c2s_pcp_map_default[mfw_vn]);
4221 	tmp = (__force u32)be32_to_cpu((__force __be32)tmp);
4222 	*c2s_default = (tmp >> (8 * mfw_vn)) & 0xff;
4223 }
4224 
4225 /**
4226  * bnx2x_setup_tc - routine to configure net_device for multi tc
4227  *
4228  * @netdev: net device to configure
4229  * @tc: number of traffic classes to enable
4230  *
4231  * callback connected to the ndo_setup_tc function pointer
4232  */
4233 int bnx2x_setup_tc(struct net_device *dev, u8 num_tc)
4234 {
4235 	struct bnx2x *bp = netdev_priv(dev);
4236 	u8 c2s_map[BNX2X_MAX_PRIORITY], c2s_def;
4237 	int cos, prio, count, offset;
4238 
4239 	/* setup tc must be called under rtnl lock */
4240 	ASSERT_RTNL();
4241 
4242 	/* no traffic classes requested. Aborting */
4243 	if (!num_tc) {
4244 		netdev_reset_tc(dev);
4245 		return 0;
4246 	}
4247 
4248 	/* requested to support too many traffic classes */
4249 	if (num_tc > bp->max_cos) {
4250 		BNX2X_ERR("support for too many traffic classes requested: %d. Max supported is %d\n",
4251 			  num_tc, bp->max_cos);
4252 		return -EINVAL;
4253 	}
4254 
4255 	/* declare amount of supported traffic classes */
4256 	if (netdev_set_num_tc(dev, num_tc)) {
4257 		BNX2X_ERR("failed to declare %d traffic classes\n", num_tc);
4258 		return -EINVAL;
4259 	}
4260 
4261 	bnx2x_get_c2s_mapping(bp, c2s_map, &c2s_def);
4262 
4263 	/* configure priority to traffic class mapping */
4264 	for (prio = 0; prio < BNX2X_MAX_PRIORITY; prio++) {
4265 		int outer_prio = c2s_map[prio];
4266 
4267 		netdev_set_prio_tc_map(dev, prio, bp->prio_to_cos[outer_prio]);
4268 		DP(BNX2X_MSG_SP | NETIF_MSG_IFUP,
4269 		   "mapping priority %d to tc %d\n",
4270 		   outer_prio, bp->prio_to_cos[outer_prio]);
4271 	}
4272 
4273 	/* Use this configuration to differentiate tc0 from other COSes
4274 	   This can be used for ets or pfc, and save the effort of setting
4275 	   up a multio class queue disc or negotiating DCBX with a switch
4276 	netdev_set_prio_tc_map(dev, 0, 0);
4277 	DP(BNX2X_MSG_SP, "mapping priority %d to tc %d\n", 0, 0);
4278 	for (prio = 1; prio < 16; prio++) {
4279 		netdev_set_prio_tc_map(dev, prio, 1);
4280 		DP(BNX2X_MSG_SP, "mapping priority %d to tc %d\n", prio, 1);
4281 	} */
4282 
4283 	/* configure traffic class to transmission queue mapping */
4284 	for (cos = 0; cos < bp->max_cos; cos++) {
4285 		count = BNX2X_NUM_ETH_QUEUES(bp);
4286 		offset = cos * BNX2X_NUM_NON_CNIC_QUEUES(bp);
4287 		netdev_set_tc_queue(dev, cos, count, offset);
4288 		DP(BNX2X_MSG_SP | NETIF_MSG_IFUP,
4289 		   "mapping tc %d to offset %d count %d\n",
4290 		   cos, offset, count);
4291 	}
4292 
4293 	return 0;
4294 }
4295 
4296 int __bnx2x_setup_tc(struct net_device *dev, enum tc_setup_type type,
4297 		     void *type_data)
4298 {
4299 	struct tc_mqprio_qopt *mqprio = type_data;
4300 
4301 	if (type != TC_SETUP_QDISC_MQPRIO)
4302 		return -EOPNOTSUPP;
4303 
4304 	mqprio->hw = TC_MQPRIO_HW_OFFLOAD_TCS;
4305 
4306 	return bnx2x_setup_tc(dev, mqprio->num_tc);
4307 }
4308 
4309 /* called with rtnl_lock */
4310 int bnx2x_change_mac_addr(struct net_device *dev, void *p)
4311 {
4312 	struct sockaddr *addr = p;
4313 	struct bnx2x *bp = netdev_priv(dev);
4314 	int rc = 0;
4315 
4316 	if (!is_valid_ether_addr(addr->sa_data)) {
4317 		BNX2X_ERR("Requested MAC address is not valid\n");
4318 		return -EINVAL;
4319 	}
4320 
4321 	if (IS_MF_STORAGE_ONLY(bp)) {
4322 		BNX2X_ERR("Can't change address on STORAGE ONLY function\n");
4323 		return -EINVAL;
4324 	}
4325 
4326 	if (netif_running(dev))  {
4327 		rc = bnx2x_set_eth_mac(bp, false);
4328 		if (rc)
4329 			return rc;
4330 	}
4331 
4332 	memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
4333 
4334 	if (netif_running(dev))
4335 		rc = bnx2x_set_eth_mac(bp, true);
4336 
4337 	if (IS_PF(bp) && SHMEM2_HAS(bp, curr_cfg))
4338 		SHMEM2_WR(bp, curr_cfg, CURR_CFG_MET_OS);
4339 
4340 	return rc;
4341 }
4342 
4343 static void bnx2x_free_fp_mem_at(struct bnx2x *bp, int fp_index)
4344 {
4345 	union host_hc_status_block *sb = &bnx2x_fp(bp, fp_index, status_blk);
4346 	struct bnx2x_fastpath *fp = &bp->fp[fp_index];
4347 	u8 cos;
4348 
4349 	/* Common */
4350 
4351 	if (IS_FCOE_IDX(fp_index)) {
4352 		memset(sb, 0, sizeof(union host_hc_status_block));
4353 		fp->status_blk_mapping = 0;
4354 	} else {
4355 		/* status blocks */
4356 		if (!CHIP_IS_E1x(bp))
4357 			BNX2X_PCI_FREE(sb->e2_sb,
4358 				       bnx2x_fp(bp, fp_index,
4359 						status_blk_mapping),
4360 				       sizeof(struct host_hc_status_block_e2));
4361 		else
4362 			BNX2X_PCI_FREE(sb->e1x_sb,
4363 				       bnx2x_fp(bp, fp_index,
4364 						status_blk_mapping),
4365 				       sizeof(struct host_hc_status_block_e1x));
4366 	}
4367 
4368 	/* Rx */
4369 	if (!skip_rx_queue(bp, fp_index)) {
4370 		bnx2x_free_rx_bds(fp);
4371 
4372 		/* fastpath rx rings: rx_buf rx_desc rx_comp */
4373 		BNX2X_FREE(bnx2x_fp(bp, fp_index, rx_buf_ring));
4374 		BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_desc_ring),
4375 			       bnx2x_fp(bp, fp_index, rx_desc_mapping),
4376 			       sizeof(struct eth_rx_bd) * NUM_RX_BD);
4377 
4378 		BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_comp_ring),
4379 			       bnx2x_fp(bp, fp_index, rx_comp_mapping),
4380 			       sizeof(struct eth_fast_path_rx_cqe) *
4381 			       NUM_RCQ_BD);
4382 
4383 		/* SGE ring */
4384 		BNX2X_FREE(bnx2x_fp(bp, fp_index, rx_page_ring));
4385 		BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_sge_ring),
4386 			       bnx2x_fp(bp, fp_index, rx_sge_mapping),
4387 			       BCM_PAGE_SIZE * NUM_RX_SGE_PAGES);
4388 	}
4389 
4390 	/* Tx */
4391 	if (!skip_tx_queue(bp, fp_index)) {
4392 		/* fastpath tx rings: tx_buf tx_desc */
4393 		for_each_cos_in_tx_queue(fp, cos) {
4394 			struct bnx2x_fp_txdata *txdata = fp->txdata_ptr[cos];
4395 
4396 			DP(NETIF_MSG_IFDOWN,
4397 			   "freeing tx memory of fp %d cos %d cid %d\n",
4398 			   fp_index, cos, txdata->cid);
4399 
4400 			BNX2X_FREE(txdata->tx_buf_ring);
4401 			BNX2X_PCI_FREE(txdata->tx_desc_ring,
4402 				txdata->tx_desc_mapping,
4403 				sizeof(union eth_tx_bd_types) * NUM_TX_BD);
4404 		}
4405 	}
4406 	/* end of fastpath */
4407 }
4408 
4409 static void bnx2x_free_fp_mem_cnic(struct bnx2x *bp)
4410 {
4411 	int i;
4412 	for_each_cnic_queue(bp, i)
4413 		bnx2x_free_fp_mem_at(bp, i);
4414 }
4415 
4416 void bnx2x_free_fp_mem(struct bnx2x *bp)
4417 {
4418 	int i;
4419 	for_each_eth_queue(bp, i)
4420 		bnx2x_free_fp_mem_at(bp, i);
4421 }
4422 
4423 static void set_sb_shortcuts(struct bnx2x *bp, int index)
4424 {
4425 	union host_hc_status_block status_blk = bnx2x_fp(bp, index, status_blk);
4426 	if (!CHIP_IS_E1x(bp)) {
4427 		bnx2x_fp(bp, index, sb_index_values) =
4428 			(__le16 *)status_blk.e2_sb->sb.index_values;
4429 		bnx2x_fp(bp, index, sb_running_index) =
4430 			(__le16 *)status_blk.e2_sb->sb.running_index;
4431 	} else {
4432 		bnx2x_fp(bp, index, sb_index_values) =
4433 			(__le16 *)status_blk.e1x_sb->sb.index_values;
4434 		bnx2x_fp(bp, index, sb_running_index) =
4435 			(__le16 *)status_blk.e1x_sb->sb.running_index;
4436 	}
4437 }
4438 
4439 /* Returns the number of actually allocated BDs */
4440 static int bnx2x_alloc_rx_bds(struct bnx2x_fastpath *fp,
4441 			      int rx_ring_size)
4442 {
4443 	struct bnx2x *bp = fp->bp;
4444 	u16 ring_prod, cqe_ring_prod;
4445 	int i, failure_cnt = 0;
4446 
4447 	fp->rx_comp_cons = 0;
4448 	cqe_ring_prod = ring_prod = 0;
4449 
4450 	/* This routine is called only during fo init so
4451 	 * fp->eth_q_stats.rx_skb_alloc_failed = 0
4452 	 */
4453 	for (i = 0; i < rx_ring_size; i++) {
4454 		if (bnx2x_alloc_rx_data(bp, fp, ring_prod, GFP_KERNEL) < 0) {
4455 			failure_cnt++;
4456 			continue;
4457 		}
4458 		ring_prod = NEXT_RX_IDX(ring_prod);
4459 		cqe_ring_prod = NEXT_RCQ_IDX(cqe_ring_prod);
4460 		WARN_ON(ring_prod <= (i - failure_cnt));
4461 	}
4462 
4463 	if (failure_cnt)
4464 		BNX2X_ERR("was only able to allocate %d rx skbs on queue[%d]\n",
4465 			  i - failure_cnt, fp->index);
4466 
4467 	fp->rx_bd_prod = ring_prod;
4468 	/* Limit the CQE producer by the CQE ring size */
4469 	fp->rx_comp_prod = min_t(u16, NUM_RCQ_RINGS*RCQ_DESC_CNT,
4470 			       cqe_ring_prod);
4471 
4472 	bnx2x_fp_stats(bp, fp)->eth_q_stats.rx_skb_alloc_failed += failure_cnt;
4473 
4474 	return i - failure_cnt;
4475 }
4476 
4477 static void bnx2x_set_next_page_rx_cq(struct bnx2x_fastpath *fp)
4478 {
4479 	int i;
4480 
4481 	for (i = 1; i <= NUM_RCQ_RINGS; i++) {
4482 		struct eth_rx_cqe_next_page *nextpg;
4483 
4484 		nextpg = (struct eth_rx_cqe_next_page *)
4485 			&fp->rx_comp_ring[RCQ_DESC_CNT * i - 1];
4486 		nextpg->addr_hi =
4487 			cpu_to_le32(U64_HI(fp->rx_comp_mapping +
4488 				   BCM_PAGE_SIZE*(i % NUM_RCQ_RINGS)));
4489 		nextpg->addr_lo =
4490 			cpu_to_le32(U64_LO(fp->rx_comp_mapping +
4491 				   BCM_PAGE_SIZE*(i % NUM_RCQ_RINGS)));
4492 	}
4493 }
4494 
4495 static int bnx2x_alloc_fp_mem_at(struct bnx2x *bp, int index)
4496 {
4497 	union host_hc_status_block *sb;
4498 	struct bnx2x_fastpath *fp = &bp->fp[index];
4499 	int ring_size = 0;
4500 	u8 cos;
4501 	int rx_ring_size = 0;
4502 
4503 	if (!bp->rx_ring_size && IS_MF_STORAGE_ONLY(bp)) {
4504 		rx_ring_size = MIN_RX_SIZE_NONTPA;
4505 		bp->rx_ring_size = rx_ring_size;
4506 	} else if (!bp->rx_ring_size) {
4507 		rx_ring_size = MAX_RX_AVAIL/BNX2X_NUM_RX_QUEUES(bp);
4508 
4509 		if (CHIP_IS_E3(bp)) {
4510 			u32 cfg = SHMEM_RD(bp,
4511 					   dev_info.port_hw_config[BP_PORT(bp)].
4512 					   default_cfg);
4513 
4514 			/* Decrease ring size for 1G functions */
4515 			if ((cfg & PORT_HW_CFG_NET_SERDES_IF_MASK) ==
4516 			    PORT_HW_CFG_NET_SERDES_IF_SGMII)
4517 				rx_ring_size /= 10;
4518 		}
4519 
4520 		/* allocate at least number of buffers required by FW */
4521 		rx_ring_size = max_t(int, bp->disable_tpa ? MIN_RX_SIZE_NONTPA :
4522 				     MIN_RX_SIZE_TPA, rx_ring_size);
4523 
4524 		bp->rx_ring_size = rx_ring_size;
4525 	} else /* if rx_ring_size specified - use it */
4526 		rx_ring_size = bp->rx_ring_size;
4527 
4528 	DP(BNX2X_MSG_SP, "calculated rx_ring_size %d\n", rx_ring_size);
4529 
4530 	/* Common */
4531 	sb = &bnx2x_fp(bp, index, status_blk);
4532 
4533 	if (!IS_FCOE_IDX(index)) {
4534 		/* status blocks */
4535 		if (!CHIP_IS_E1x(bp)) {
4536 			sb->e2_sb = BNX2X_PCI_ALLOC(&bnx2x_fp(bp, index, status_blk_mapping),
4537 						    sizeof(struct host_hc_status_block_e2));
4538 			if (!sb->e2_sb)
4539 				goto alloc_mem_err;
4540 		} else {
4541 			sb->e1x_sb = BNX2X_PCI_ALLOC(&bnx2x_fp(bp, index, status_blk_mapping),
4542 						     sizeof(struct host_hc_status_block_e1x));
4543 			if (!sb->e1x_sb)
4544 				goto alloc_mem_err;
4545 		}
4546 	}
4547 
4548 	/* FCoE Queue uses Default SB and doesn't ACK the SB, thus no need to
4549 	 * set shortcuts for it.
4550 	 */
4551 	if (!IS_FCOE_IDX(index))
4552 		set_sb_shortcuts(bp, index);
4553 
4554 	/* Tx */
4555 	if (!skip_tx_queue(bp, index)) {
4556 		/* fastpath tx rings: tx_buf tx_desc */
4557 		for_each_cos_in_tx_queue(fp, cos) {
4558 			struct bnx2x_fp_txdata *txdata = fp->txdata_ptr[cos];
4559 
4560 			DP(NETIF_MSG_IFUP,
4561 			   "allocating tx memory of fp %d cos %d\n",
4562 			   index, cos);
4563 
4564 			txdata->tx_buf_ring = kcalloc(NUM_TX_BD,
4565 						      sizeof(struct sw_tx_bd),
4566 						      GFP_KERNEL);
4567 			if (!txdata->tx_buf_ring)
4568 				goto alloc_mem_err;
4569 			txdata->tx_desc_ring = BNX2X_PCI_ALLOC(&txdata->tx_desc_mapping,
4570 							       sizeof(union eth_tx_bd_types) * NUM_TX_BD);
4571 			if (!txdata->tx_desc_ring)
4572 				goto alloc_mem_err;
4573 		}
4574 	}
4575 
4576 	/* Rx */
4577 	if (!skip_rx_queue(bp, index)) {
4578 		/* fastpath rx rings: rx_buf rx_desc rx_comp */
4579 		bnx2x_fp(bp, index, rx_buf_ring) =
4580 			kcalloc(NUM_RX_BD, sizeof(struct sw_rx_bd), GFP_KERNEL);
4581 		if (!bnx2x_fp(bp, index, rx_buf_ring))
4582 			goto alloc_mem_err;
4583 		bnx2x_fp(bp, index, rx_desc_ring) =
4584 			BNX2X_PCI_ALLOC(&bnx2x_fp(bp, index, rx_desc_mapping),
4585 					sizeof(struct eth_rx_bd) * NUM_RX_BD);
4586 		if (!bnx2x_fp(bp, index, rx_desc_ring))
4587 			goto alloc_mem_err;
4588 
4589 		/* Seed all CQEs by 1s */
4590 		bnx2x_fp(bp, index, rx_comp_ring) =
4591 			BNX2X_PCI_FALLOC(&bnx2x_fp(bp, index, rx_comp_mapping),
4592 					 sizeof(struct eth_fast_path_rx_cqe) * NUM_RCQ_BD);
4593 		if (!bnx2x_fp(bp, index, rx_comp_ring))
4594 			goto alloc_mem_err;
4595 
4596 		/* SGE ring */
4597 		bnx2x_fp(bp, index, rx_page_ring) =
4598 			kcalloc(NUM_RX_SGE, sizeof(struct sw_rx_page),
4599 				GFP_KERNEL);
4600 		if (!bnx2x_fp(bp, index, rx_page_ring))
4601 			goto alloc_mem_err;
4602 		bnx2x_fp(bp, index, rx_sge_ring) =
4603 			BNX2X_PCI_ALLOC(&bnx2x_fp(bp, index, rx_sge_mapping),
4604 					BCM_PAGE_SIZE * NUM_RX_SGE_PAGES);
4605 		if (!bnx2x_fp(bp, index, rx_sge_ring))
4606 			goto alloc_mem_err;
4607 		/* RX BD ring */
4608 		bnx2x_set_next_page_rx_bd(fp);
4609 
4610 		/* CQ ring */
4611 		bnx2x_set_next_page_rx_cq(fp);
4612 
4613 		/* BDs */
4614 		ring_size = bnx2x_alloc_rx_bds(fp, rx_ring_size);
4615 		if (ring_size < rx_ring_size)
4616 			goto alloc_mem_err;
4617 	}
4618 
4619 	return 0;
4620 
4621 /* handles low memory cases */
4622 alloc_mem_err:
4623 	BNX2X_ERR("Unable to allocate full memory for queue %d (size %d)\n",
4624 						index, ring_size);
4625 	/* FW will drop all packets if queue is not big enough,
4626 	 * In these cases we disable the queue
4627 	 * Min size is different for OOO, TPA and non-TPA queues
4628 	 */
4629 	if (ring_size < (fp->mode == TPA_MODE_DISABLED ?
4630 				MIN_RX_SIZE_NONTPA : MIN_RX_SIZE_TPA)) {
4631 			/* release memory allocated for this queue */
4632 			bnx2x_free_fp_mem_at(bp, index);
4633 			return -ENOMEM;
4634 	}
4635 	return 0;
4636 }
4637 
4638 static int bnx2x_alloc_fp_mem_cnic(struct bnx2x *bp)
4639 {
4640 	if (!NO_FCOE(bp))
4641 		/* FCoE */
4642 		if (bnx2x_alloc_fp_mem_at(bp, FCOE_IDX(bp)))
4643 			/* we will fail load process instead of mark
4644 			 * NO_FCOE_FLAG
4645 			 */
4646 			return -ENOMEM;
4647 
4648 	return 0;
4649 }
4650 
4651 static int bnx2x_alloc_fp_mem(struct bnx2x *bp)
4652 {
4653 	int i;
4654 
4655 	/* 1. Allocate FP for leading - fatal if error
4656 	 * 2. Allocate RSS - fix number of queues if error
4657 	 */
4658 
4659 	/* leading */
4660 	if (bnx2x_alloc_fp_mem_at(bp, 0))
4661 		return -ENOMEM;
4662 
4663 	/* RSS */
4664 	for_each_nondefault_eth_queue(bp, i)
4665 		if (bnx2x_alloc_fp_mem_at(bp, i))
4666 			break;
4667 
4668 	/* handle memory failures */
4669 	if (i != BNX2X_NUM_ETH_QUEUES(bp)) {
4670 		int delta = BNX2X_NUM_ETH_QUEUES(bp) - i;
4671 
4672 		WARN_ON(delta < 0);
4673 		bnx2x_shrink_eth_fp(bp, delta);
4674 		if (CNIC_SUPPORT(bp))
4675 			/* move non eth FPs next to last eth FP
4676 			 * must be done in that order
4677 			 * FCOE_IDX < FWD_IDX < OOO_IDX
4678 			 */
4679 
4680 			/* move FCoE fp even NO_FCOE_FLAG is on */
4681 			bnx2x_move_fp(bp, FCOE_IDX(bp), FCOE_IDX(bp) - delta);
4682 		bp->num_ethernet_queues -= delta;
4683 		bp->num_queues = bp->num_ethernet_queues +
4684 				 bp->num_cnic_queues;
4685 		BNX2X_ERR("Adjusted num of queues from %d to %d\n",
4686 			  bp->num_queues + delta, bp->num_queues);
4687 	}
4688 
4689 	return 0;
4690 }
4691 
4692 void bnx2x_free_mem_bp(struct bnx2x *bp)
4693 {
4694 	int i;
4695 
4696 	for (i = 0; i < bp->fp_array_size; i++)
4697 		kfree(bp->fp[i].tpa_info);
4698 	kfree(bp->fp);
4699 	kfree(bp->sp_objs);
4700 	kfree(bp->fp_stats);
4701 	kfree(bp->bnx2x_txq);
4702 	kfree(bp->msix_table);
4703 	kfree(bp->ilt);
4704 }
4705 
4706 int bnx2x_alloc_mem_bp(struct bnx2x *bp)
4707 {
4708 	struct bnx2x_fastpath *fp;
4709 	struct msix_entry *tbl;
4710 	struct bnx2x_ilt *ilt;
4711 	int msix_table_size = 0;
4712 	int fp_array_size, txq_array_size;
4713 	int i;
4714 
4715 	/*
4716 	 * The biggest MSI-X table we might need is as a maximum number of fast
4717 	 * path IGU SBs plus default SB (for PF only).
4718 	 */
4719 	msix_table_size = bp->igu_sb_cnt;
4720 	if (IS_PF(bp))
4721 		msix_table_size++;
4722 	BNX2X_DEV_INFO("msix_table_size %d\n", msix_table_size);
4723 
4724 	/* fp array: RSS plus CNIC related L2 queues */
4725 	fp_array_size = BNX2X_MAX_RSS_COUNT(bp) + CNIC_SUPPORT(bp);
4726 	bp->fp_array_size = fp_array_size;
4727 	BNX2X_DEV_INFO("fp_array_size %d\n", bp->fp_array_size);
4728 
4729 	fp = kcalloc(bp->fp_array_size, sizeof(*fp), GFP_KERNEL);
4730 	if (!fp)
4731 		goto alloc_err;
4732 	for (i = 0; i < bp->fp_array_size; i++) {
4733 		fp[i].tpa_info =
4734 			kcalloc(ETH_MAX_AGGREGATION_QUEUES_E1H_E2,
4735 				sizeof(struct bnx2x_agg_info), GFP_KERNEL);
4736 		if (!(fp[i].tpa_info))
4737 			goto alloc_err;
4738 	}
4739 
4740 	bp->fp = fp;
4741 
4742 	/* allocate sp objs */
4743 	bp->sp_objs = kcalloc(bp->fp_array_size, sizeof(struct bnx2x_sp_objs),
4744 			      GFP_KERNEL);
4745 	if (!bp->sp_objs)
4746 		goto alloc_err;
4747 
4748 	/* allocate fp_stats */
4749 	bp->fp_stats = kcalloc(bp->fp_array_size, sizeof(struct bnx2x_fp_stats),
4750 			       GFP_KERNEL);
4751 	if (!bp->fp_stats)
4752 		goto alloc_err;
4753 
4754 	/* Allocate memory for the transmission queues array */
4755 	txq_array_size =
4756 		BNX2X_MAX_RSS_COUNT(bp) * BNX2X_MULTI_TX_COS + CNIC_SUPPORT(bp);
4757 	BNX2X_DEV_INFO("txq_array_size %d", txq_array_size);
4758 
4759 	bp->bnx2x_txq = kcalloc(txq_array_size, sizeof(struct bnx2x_fp_txdata),
4760 				GFP_KERNEL);
4761 	if (!bp->bnx2x_txq)
4762 		goto alloc_err;
4763 
4764 	/* msix table */
4765 	tbl = kcalloc(msix_table_size, sizeof(*tbl), GFP_KERNEL);
4766 	if (!tbl)
4767 		goto alloc_err;
4768 	bp->msix_table = tbl;
4769 
4770 	/* ilt */
4771 	ilt = kzalloc(sizeof(*ilt), GFP_KERNEL);
4772 	if (!ilt)
4773 		goto alloc_err;
4774 	bp->ilt = ilt;
4775 
4776 	return 0;
4777 alloc_err:
4778 	bnx2x_free_mem_bp(bp);
4779 	return -ENOMEM;
4780 }
4781 
4782 int bnx2x_reload_if_running(struct net_device *dev)
4783 {
4784 	struct bnx2x *bp = netdev_priv(dev);
4785 
4786 	if (unlikely(!netif_running(dev)))
4787 		return 0;
4788 
4789 	bnx2x_nic_unload(bp, UNLOAD_NORMAL, true);
4790 	return bnx2x_nic_load(bp, LOAD_NORMAL);
4791 }
4792 
4793 int bnx2x_get_cur_phy_idx(struct bnx2x *bp)
4794 {
4795 	u32 sel_phy_idx = 0;
4796 	if (bp->link_params.num_phys <= 1)
4797 		return INT_PHY;
4798 
4799 	if (bp->link_vars.link_up) {
4800 		sel_phy_idx = EXT_PHY1;
4801 		/* In case link is SERDES, check if the EXT_PHY2 is the one */
4802 		if ((bp->link_vars.link_status & LINK_STATUS_SERDES_LINK) &&
4803 		    (bp->link_params.phy[EXT_PHY2].supported & SUPPORTED_FIBRE))
4804 			sel_phy_idx = EXT_PHY2;
4805 	} else {
4806 
4807 		switch (bnx2x_phy_selection(&bp->link_params)) {
4808 		case PORT_HW_CFG_PHY_SELECTION_HARDWARE_DEFAULT:
4809 		case PORT_HW_CFG_PHY_SELECTION_FIRST_PHY:
4810 		case PORT_HW_CFG_PHY_SELECTION_FIRST_PHY_PRIORITY:
4811 		       sel_phy_idx = EXT_PHY1;
4812 		       break;
4813 		case PORT_HW_CFG_PHY_SELECTION_SECOND_PHY:
4814 		case PORT_HW_CFG_PHY_SELECTION_SECOND_PHY_PRIORITY:
4815 		       sel_phy_idx = EXT_PHY2;
4816 		       break;
4817 		}
4818 	}
4819 
4820 	return sel_phy_idx;
4821 }
4822 int bnx2x_get_link_cfg_idx(struct bnx2x *bp)
4823 {
4824 	u32 sel_phy_idx = bnx2x_get_cur_phy_idx(bp);
4825 	/*
4826 	 * The selected activated PHY is always after swapping (in case PHY
4827 	 * swapping is enabled). So when swapping is enabled, we need to reverse
4828 	 * the configuration
4829 	 */
4830 
4831 	if (bp->link_params.multi_phy_config &
4832 	    PORT_HW_CFG_PHY_SWAPPED_ENABLED) {
4833 		if (sel_phy_idx == EXT_PHY1)
4834 			sel_phy_idx = EXT_PHY2;
4835 		else if (sel_phy_idx == EXT_PHY2)
4836 			sel_phy_idx = EXT_PHY1;
4837 	}
4838 	return LINK_CONFIG_IDX(sel_phy_idx);
4839 }
4840 
4841 #ifdef NETDEV_FCOE_WWNN
4842 int bnx2x_fcoe_get_wwn(struct net_device *dev, u64 *wwn, int type)
4843 {
4844 	struct bnx2x *bp = netdev_priv(dev);
4845 	struct cnic_eth_dev *cp = &bp->cnic_eth_dev;
4846 
4847 	switch (type) {
4848 	case NETDEV_FCOE_WWNN:
4849 		*wwn = HILO_U64(cp->fcoe_wwn_node_name_hi,
4850 				cp->fcoe_wwn_node_name_lo);
4851 		break;
4852 	case NETDEV_FCOE_WWPN:
4853 		*wwn = HILO_U64(cp->fcoe_wwn_port_name_hi,
4854 				cp->fcoe_wwn_port_name_lo);
4855 		break;
4856 	default:
4857 		BNX2X_ERR("Wrong WWN type requested - %d\n", type);
4858 		return -EINVAL;
4859 	}
4860 
4861 	return 0;
4862 }
4863 #endif
4864 
4865 /* called with rtnl_lock */
4866 int bnx2x_change_mtu(struct net_device *dev, int new_mtu)
4867 {
4868 	struct bnx2x *bp = netdev_priv(dev);
4869 
4870 	if (pci_num_vf(bp->pdev)) {
4871 		DP(BNX2X_MSG_IOV, "VFs are enabled, can not change MTU\n");
4872 		return -EPERM;
4873 	}
4874 
4875 	if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
4876 		BNX2X_ERR("Can't perform change MTU during parity recovery\n");
4877 		return -EAGAIN;
4878 	}
4879 
4880 	/* This does not race with packet allocation
4881 	 * because the actual alloc size is
4882 	 * only updated as part of load
4883 	 */
4884 	dev->mtu = new_mtu;
4885 
4886 	if (!bnx2x_mtu_allows_gro(new_mtu))
4887 		dev->features &= ~NETIF_F_GRO_HW;
4888 
4889 	if (IS_PF(bp) && SHMEM2_HAS(bp, curr_cfg))
4890 		SHMEM2_WR(bp, curr_cfg, CURR_CFG_MET_OS);
4891 
4892 	return bnx2x_reload_if_running(dev);
4893 }
4894 
4895 netdev_features_t bnx2x_fix_features(struct net_device *dev,
4896 				     netdev_features_t features)
4897 {
4898 	struct bnx2x *bp = netdev_priv(dev);
4899 
4900 	if (pci_num_vf(bp->pdev)) {
4901 		netdev_features_t changed = dev->features ^ features;
4902 
4903 		/* Revert the requested changes in features if they
4904 		 * would require internal reload of PF in bnx2x_set_features().
4905 		 */
4906 		if (!(features & NETIF_F_RXCSUM) && !bp->disable_tpa) {
4907 			features &= ~NETIF_F_RXCSUM;
4908 			features |= dev->features & NETIF_F_RXCSUM;
4909 		}
4910 
4911 		if (changed & NETIF_F_LOOPBACK) {
4912 			features &= ~NETIF_F_LOOPBACK;
4913 			features |= dev->features & NETIF_F_LOOPBACK;
4914 		}
4915 	}
4916 
4917 	/* TPA requires Rx CSUM offloading */
4918 	if (!(features & NETIF_F_RXCSUM))
4919 		features &= ~NETIF_F_LRO;
4920 
4921 	if (!(features & NETIF_F_GRO) || !bnx2x_mtu_allows_gro(dev->mtu))
4922 		features &= ~NETIF_F_GRO_HW;
4923 	if (features & NETIF_F_GRO_HW)
4924 		features &= ~NETIF_F_LRO;
4925 
4926 	return features;
4927 }
4928 
4929 int bnx2x_set_features(struct net_device *dev, netdev_features_t features)
4930 {
4931 	struct bnx2x *bp = netdev_priv(dev);
4932 	netdev_features_t changes = features ^ dev->features;
4933 	bool bnx2x_reload = false;
4934 	int rc;
4935 
4936 	/* VFs or non SRIOV PFs should be able to change loopback feature */
4937 	if (!pci_num_vf(bp->pdev)) {
4938 		if (features & NETIF_F_LOOPBACK) {
4939 			if (bp->link_params.loopback_mode != LOOPBACK_BMAC) {
4940 				bp->link_params.loopback_mode = LOOPBACK_BMAC;
4941 				bnx2x_reload = true;
4942 			}
4943 		} else {
4944 			if (bp->link_params.loopback_mode != LOOPBACK_NONE) {
4945 				bp->link_params.loopback_mode = LOOPBACK_NONE;
4946 				bnx2x_reload = true;
4947 			}
4948 		}
4949 	}
4950 
4951 	/* Don't care about GRO changes */
4952 	changes &= ~NETIF_F_GRO;
4953 
4954 	if (changes)
4955 		bnx2x_reload = true;
4956 
4957 	if (bnx2x_reload) {
4958 		if (bp->recovery_state == BNX2X_RECOVERY_DONE) {
4959 			dev->features = features;
4960 			rc = bnx2x_reload_if_running(dev);
4961 			return rc ? rc : 1;
4962 		}
4963 		/* else: bnx2x_nic_load() will be called at end of recovery */
4964 	}
4965 
4966 	return 0;
4967 }
4968 
4969 void bnx2x_tx_timeout(struct net_device *dev)
4970 {
4971 	struct bnx2x *bp = netdev_priv(dev);
4972 
4973 	/* We want the information of the dump logged,
4974 	 * but calling bnx2x_panic() would kill all chances of recovery.
4975 	 */
4976 	if (!bp->panic)
4977 #ifndef BNX2X_STOP_ON_ERROR
4978 		bnx2x_panic_dump(bp, false);
4979 #else
4980 		bnx2x_panic();
4981 #endif
4982 
4983 	/* This allows the netif to be shutdown gracefully before resetting */
4984 	bnx2x_schedule_sp_rtnl(bp, BNX2X_SP_RTNL_TX_TIMEOUT, 0);
4985 }
4986 
4987 int bnx2x_suspend(struct pci_dev *pdev, pm_message_t state)
4988 {
4989 	struct net_device *dev = pci_get_drvdata(pdev);
4990 	struct bnx2x *bp;
4991 
4992 	if (!dev) {
4993 		dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
4994 		return -ENODEV;
4995 	}
4996 	bp = netdev_priv(dev);
4997 
4998 	rtnl_lock();
4999 
5000 	pci_save_state(pdev);
5001 
5002 	if (!netif_running(dev)) {
5003 		rtnl_unlock();
5004 		return 0;
5005 	}
5006 
5007 	netif_device_detach(dev);
5008 
5009 	bnx2x_nic_unload(bp, UNLOAD_CLOSE, false);
5010 
5011 	bnx2x_set_power_state(bp, pci_choose_state(pdev, state));
5012 
5013 	rtnl_unlock();
5014 
5015 	return 0;
5016 }
5017 
5018 int bnx2x_resume(struct pci_dev *pdev)
5019 {
5020 	struct net_device *dev = pci_get_drvdata(pdev);
5021 	struct bnx2x *bp;
5022 	int rc;
5023 
5024 	if (!dev) {
5025 		dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
5026 		return -ENODEV;
5027 	}
5028 	bp = netdev_priv(dev);
5029 
5030 	if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
5031 		BNX2X_ERR("Handling parity error recovery. Try again later\n");
5032 		return -EAGAIN;
5033 	}
5034 
5035 	rtnl_lock();
5036 
5037 	pci_restore_state(pdev);
5038 
5039 	if (!netif_running(dev)) {
5040 		rtnl_unlock();
5041 		return 0;
5042 	}
5043 
5044 	bnx2x_set_power_state(bp, PCI_D0);
5045 	netif_device_attach(dev);
5046 
5047 	rc = bnx2x_nic_load(bp, LOAD_OPEN);
5048 
5049 	rtnl_unlock();
5050 
5051 	return rc;
5052 }
5053 
5054 void bnx2x_set_ctx_validation(struct bnx2x *bp, struct eth_context *cxt,
5055 			      u32 cid)
5056 {
5057 	if (!cxt) {
5058 		BNX2X_ERR("bad context pointer %p\n", cxt);
5059 		return;
5060 	}
5061 
5062 	/* ustorm cxt validation */
5063 	cxt->ustorm_ag_context.cdu_usage =
5064 		CDU_RSRVD_VALUE_TYPE_A(HW_CID(bp, cid),
5065 			CDU_REGION_NUMBER_UCM_AG, ETH_CONNECTION_TYPE);
5066 	/* xcontext validation */
5067 	cxt->xstorm_ag_context.cdu_reserved =
5068 		CDU_RSRVD_VALUE_TYPE_A(HW_CID(bp, cid),
5069 			CDU_REGION_NUMBER_XCM_AG, ETH_CONNECTION_TYPE);
5070 }
5071 
5072 static void storm_memset_hc_timeout(struct bnx2x *bp, u8 port,
5073 				    u8 fw_sb_id, u8 sb_index,
5074 				    u8 ticks)
5075 {
5076 	u32 addr = BAR_CSTRORM_INTMEM +
5077 		   CSTORM_STATUS_BLOCK_DATA_TIMEOUT_OFFSET(fw_sb_id, sb_index);
5078 	REG_WR8(bp, addr, ticks);
5079 	DP(NETIF_MSG_IFUP,
5080 	   "port %x fw_sb_id %d sb_index %d ticks %d\n",
5081 	   port, fw_sb_id, sb_index, ticks);
5082 }
5083 
5084 static void storm_memset_hc_disable(struct bnx2x *bp, u8 port,
5085 				    u16 fw_sb_id, u8 sb_index,
5086 				    u8 disable)
5087 {
5088 	u32 enable_flag = disable ? 0 : (1 << HC_INDEX_DATA_HC_ENABLED_SHIFT);
5089 	u32 addr = BAR_CSTRORM_INTMEM +
5090 		   CSTORM_STATUS_BLOCK_DATA_FLAGS_OFFSET(fw_sb_id, sb_index);
5091 	u8 flags = REG_RD8(bp, addr);
5092 	/* clear and set */
5093 	flags &= ~HC_INDEX_DATA_HC_ENABLED;
5094 	flags |= enable_flag;
5095 	REG_WR8(bp, addr, flags);
5096 	DP(NETIF_MSG_IFUP,
5097 	   "port %x fw_sb_id %d sb_index %d disable %d\n",
5098 	   port, fw_sb_id, sb_index, disable);
5099 }
5100 
5101 void bnx2x_update_coalesce_sb_index(struct bnx2x *bp, u8 fw_sb_id,
5102 				    u8 sb_index, u8 disable, u16 usec)
5103 {
5104 	int port = BP_PORT(bp);
5105 	u8 ticks = usec / BNX2X_BTR;
5106 
5107 	storm_memset_hc_timeout(bp, port, fw_sb_id, sb_index, ticks);
5108 
5109 	disable = disable ? 1 : (usec ? 0 : 1);
5110 	storm_memset_hc_disable(bp, port, fw_sb_id, sb_index, disable);
5111 }
5112 
5113 void bnx2x_schedule_sp_rtnl(struct bnx2x *bp, enum sp_rtnl_flag flag,
5114 			    u32 verbose)
5115 {
5116 	smp_mb__before_atomic();
5117 	set_bit(flag, &bp->sp_rtnl_state);
5118 	smp_mb__after_atomic();
5119 	DP((BNX2X_MSG_SP | verbose), "Scheduling sp_rtnl task [Flag: %d]\n",
5120 	   flag);
5121 	schedule_delayed_work(&bp->sp_rtnl_task, 0);
5122 }
5123