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