1 /* bnx2x_cmn.c: Broadcom Everest network driver.
2  *
3  * Copyright (c) 2007-2012 Broadcom Corporation
4  *
5  * This program is free software; you can redistribute it and/or modify
6  * it under the terms of the GNU General Public License as published by
7  * the Free Software Foundation.
8  *
9  * Maintained by: Eilon Greenstein <eilong@broadcom.com>
10  * Written by: Eliezer Tamir
11  * Based on code from Michael Chan's bnx2 driver
12  * UDP CSUM errata workaround by Arik Gendelman
13  * Slowpath and fastpath rework by Vladislav Zolotarov
14  * Statistics and Link management by Yitchak Gertner
15  *
16  */
17 
18 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
19 
20 #include <linux/etherdevice.h>
21 #include <linux/if_vlan.h>
22 #include <linux/interrupt.h>
23 #include <linux/ip.h>
24 #include <net/ipv6.h>
25 #include <net/ip6_checksum.h>
26 #include <linux/firmware.h>
27 #include <linux/prefetch.h>
28 #include "bnx2x_cmn.h"
29 #include "bnx2x_init.h"
30 #include "bnx2x_sp.h"
31 
32 
33 
34 /**
35  * bnx2x_move_fp - move content of the fastpath structure.
36  *
37  * @bp:		driver handle
38  * @from:	source FP index
39  * @to:		destination FP index
40  *
41  * Makes sure the contents of the bp->fp[to].napi is kept
42  * intact. This is done by first copying the napi struct from
43  * the target to the source, and then mem copying the entire
44  * source onto the target
45  */
46 static inline void bnx2x_move_fp(struct bnx2x *bp, int from, int to)
47 {
48 	struct bnx2x_fastpath *from_fp = &bp->fp[from];
49 	struct bnx2x_fastpath *to_fp = &bp->fp[to];
50 
51 	/* Copy the NAPI object as it has been already initialized */
52 	from_fp->napi = to_fp->napi;
53 
54 	/* Move bnx2x_fastpath contents */
55 	memcpy(to_fp, from_fp, sizeof(*to_fp));
56 	to_fp->index = to;
57 }
58 
59 int load_count[2][3] = { {0} }; /* per-path: 0-common, 1-port0, 2-port1 */
60 
61 /* free skb in the packet ring at pos idx
62  * return idx of last bd freed
63  */
64 static u16 bnx2x_free_tx_pkt(struct bnx2x *bp, struct bnx2x_fp_txdata *txdata,
65 			     u16 idx, unsigned int *pkts_compl,
66 			     unsigned int *bytes_compl)
67 {
68 	struct sw_tx_bd *tx_buf = &txdata->tx_buf_ring[idx];
69 	struct eth_tx_start_bd *tx_start_bd;
70 	struct eth_tx_bd *tx_data_bd;
71 	struct sk_buff *skb = tx_buf->skb;
72 	u16 bd_idx = TX_BD(tx_buf->first_bd), new_cons;
73 	int nbd;
74 
75 	/* prefetch skb end pointer to speedup dev_kfree_skb() */
76 	prefetch(&skb->end);
77 
78 	DP(NETIF_MSG_TX_DONE, "fp[%d]: pkt_idx %d  buff @(%p)->skb %p\n",
79 	   txdata->txq_index, idx, tx_buf, skb);
80 
81 	/* unmap first bd */
82 	tx_start_bd = &txdata->tx_desc_ring[bd_idx].start_bd;
83 	dma_unmap_single(&bp->pdev->dev, BD_UNMAP_ADDR(tx_start_bd),
84 			 BD_UNMAP_LEN(tx_start_bd), DMA_TO_DEVICE);
85 
86 
87 	nbd = le16_to_cpu(tx_start_bd->nbd) - 1;
88 #ifdef BNX2X_STOP_ON_ERROR
89 	if ((nbd - 1) > (MAX_SKB_FRAGS + 2)) {
90 		BNX2X_ERR("BAD nbd!\n");
91 		bnx2x_panic();
92 	}
93 #endif
94 	new_cons = nbd + tx_buf->first_bd;
95 
96 	/* Get the next bd */
97 	bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
98 
99 	/* Skip a parse bd... */
100 	--nbd;
101 	bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
102 
103 	/* ...and the TSO split header bd since they have no mapping */
104 	if (tx_buf->flags & BNX2X_TSO_SPLIT_BD) {
105 		--nbd;
106 		bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
107 	}
108 
109 	/* now free frags */
110 	while (nbd > 0) {
111 
112 		tx_data_bd = &txdata->tx_desc_ring[bd_idx].reg_bd;
113 		dma_unmap_page(&bp->pdev->dev, BD_UNMAP_ADDR(tx_data_bd),
114 			       BD_UNMAP_LEN(tx_data_bd), DMA_TO_DEVICE);
115 		if (--nbd)
116 			bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
117 	}
118 
119 	/* release skb */
120 	WARN_ON(!skb);
121 	if (likely(skb)) {
122 		(*pkts_compl)++;
123 		(*bytes_compl) += skb->len;
124 	}
125 
126 	dev_kfree_skb_any(skb);
127 	tx_buf->first_bd = 0;
128 	tx_buf->skb = NULL;
129 
130 	return new_cons;
131 }
132 
133 int bnx2x_tx_int(struct bnx2x *bp, struct bnx2x_fp_txdata *txdata)
134 {
135 	struct netdev_queue *txq;
136 	u16 hw_cons, sw_cons, bd_cons = txdata->tx_bd_cons;
137 	unsigned int pkts_compl = 0, bytes_compl = 0;
138 
139 #ifdef BNX2X_STOP_ON_ERROR
140 	if (unlikely(bp->panic))
141 		return -1;
142 #endif
143 
144 	txq = netdev_get_tx_queue(bp->dev, txdata->txq_index);
145 	hw_cons = le16_to_cpu(*txdata->tx_cons_sb);
146 	sw_cons = txdata->tx_pkt_cons;
147 
148 	while (sw_cons != hw_cons) {
149 		u16 pkt_cons;
150 
151 		pkt_cons = TX_BD(sw_cons);
152 
153 		DP(NETIF_MSG_TX_DONE,
154 		   "queue[%d]: hw_cons %u  sw_cons %u  pkt_cons %u\n",
155 		   txdata->txq_index, hw_cons, sw_cons, pkt_cons);
156 
157 		bd_cons = bnx2x_free_tx_pkt(bp, txdata, pkt_cons,
158 		    &pkts_compl, &bytes_compl);
159 
160 		sw_cons++;
161 	}
162 
163 	netdev_tx_completed_queue(txq, pkts_compl, bytes_compl);
164 
165 	txdata->tx_pkt_cons = sw_cons;
166 	txdata->tx_bd_cons = bd_cons;
167 
168 	/* Need to make the tx_bd_cons update visible to start_xmit()
169 	 * before checking for netif_tx_queue_stopped().  Without the
170 	 * memory barrier, there is a small possibility that
171 	 * start_xmit() will miss it and cause the queue to be stopped
172 	 * forever.
173 	 * On the other hand we need an rmb() here to ensure the proper
174 	 * ordering of bit testing in the following
175 	 * netif_tx_queue_stopped(txq) call.
176 	 */
177 	smp_mb();
178 
179 	if (unlikely(netif_tx_queue_stopped(txq))) {
180 		/* Taking tx_lock() is needed to prevent reenabling the queue
181 		 * while it's empty. This could have happen if rx_action() gets
182 		 * suspended in bnx2x_tx_int() after the condition before
183 		 * netif_tx_wake_queue(), while tx_action (bnx2x_start_xmit()):
184 		 *
185 		 * stops the queue->sees fresh tx_bd_cons->releases the queue->
186 		 * sends some packets consuming the whole queue again->
187 		 * stops the queue
188 		 */
189 
190 		__netif_tx_lock(txq, smp_processor_id());
191 
192 		if ((netif_tx_queue_stopped(txq)) &&
193 		    (bp->state == BNX2X_STATE_OPEN) &&
194 		    (bnx2x_tx_avail(bp, txdata) >= MAX_SKB_FRAGS + 3))
195 			netif_tx_wake_queue(txq);
196 
197 		__netif_tx_unlock(txq);
198 	}
199 	return 0;
200 }
201 
202 static inline void bnx2x_update_last_max_sge(struct bnx2x_fastpath *fp,
203 					     u16 idx)
204 {
205 	u16 last_max = fp->last_max_sge;
206 
207 	if (SUB_S16(idx, last_max) > 0)
208 		fp->last_max_sge = idx;
209 }
210 
211 static inline void bnx2x_update_sge_prod(struct bnx2x_fastpath *fp,
212 					 u16 sge_len,
213 					 struct eth_end_agg_rx_cqe *cqe)
214 {
215 	struct bnx2x *bp = fp->bp;
216 	u16 last_max, last_elem, first_elem;
217 	u16 delta = 0;
218 	u16 i;
219 
220 	if (!sge_len)
221 		return;
222 
223 	/* First mark all used pages */
224 	for (i = 0; i < sge_len; i++)
225 		BIT_VEC64_CLEAR_BIT(fp->sge_mask,
226 			RX_SGE(le16_to_cpu(cqe->sgl_or_raw_data.sgl[i])));
227 
228 	DP(NETIF_MSG_RX_STATUS, "fp_cqe->sgl[%d] = %d\n",
229 	   sge_len - 1, le16_to_cpu(cqe->sgl_or_raw_data.sgl[sge_len - 1]));
230 
231 	/* Here we assume that the last SGE index is the biggest */
232 	prefetch((void *)(fp->sge_mask));
233 	bnx2x_update_last_max_sge(fp,
234 		le16_to_cpu(cqe->sgl_or_raw_data.sgl[sge_len - 1]));
235 
236 	last_max = RX_SGE(fp->last_max_sge);
237 	last_elem = last_max >> BIT_VEC64_ELEM_SHIFT;
238 	first_elem = RX_SGE(fp->rx_sge_prod) >> BIT_VEC64_ELEM_SHIFT;
239 
240 	/* If ring is not full */
241 	if (last_elem + 1 != first_elem)
242 		last_elem++;
243 
244 	/* Now update the prod */
245 	for (i = first_elem; i != last_elem; i = NEXT_SGE_MASK_ELEM(i)) {
246 		if (likely(fp->sge_mask[i]))
247 			break;
248 
249 		fp->sge_mask[i] = BIT_VEC64_ELEM_ONE_MASK;
250 		delta += BIT_VEC64_ELEM_SZ;
251 	}
252 
253 	if (delta > 0) {
254 		fp->rx_sge_prod += delta;
255 		/* clear page-end entries */
256 		bnx2x_clear_sge_mask_next_elems(fp);
257 	}
258 
259 	DP(NETIF_MSG_RX_STATUS,
260 	   "fp->last_max_sge = %d  fp->rx_sge_prod = %d\n",
261 	   fp->last_max_sge, fp->rx_sge_prod);
262 }
263 
264 /* Set Toeplitz hash value in the skb using the value from the
265  * CQE (calculated by HW).
266  */
267 static u32 bnx2x_get_rxhash(const struct bnx2x *bp,
268 			    const struct eth_fast_path_rx_cqe *cqe)
269 {
270 	/* Set Toeplitz hash from CQE */
271 	if ((bp->dev->features & NETIF_F_RXHASH) &&
272 	    (cqe->status_flags & ETH_FAST_PATH_RX_CQE_RSS_HASH_FLG))
273 		return le32_to_cpu(cqe->rss_hash_result);
274 	return 0;
275 }
276 
277 static void bnx2x_tpa_start(struct bnx2x_fastpath *fp, u16 queue,
278 			    u16 cons, u16 prod,
279 			    struct eth_fast_path_rx_cqe *cqe)
280 {
281 	struct bnx2x *bp = fp->bp;
282 	struct sw_rx_bd *cons_rx_buf = &fp->rx_buf_ring[cons];
283 	struct sw_rx_bd *prod_rx_buf = &fp->rx_buf_ring[prod];
284 	struct eth_rx_bd *prod_bd = &fp->rx_desc_ring[prod];
285 	dma_addr_t mapping;
286 	struct bnx2x_agg_info *tpa_info = &fp->tpa_info[queue];
287 	struct sw_rx_bd *first_buf = &tpa_info->first_buf;
288 
289 	/* print error if current state != stop */
290 	if (tpa_info->tpa_state != BNX2X_TPA_STOP)
291 		BNX2X_ERR("start of bin not in stop [%d]\n", queue);
292 
293 	/* Try to map an empty data buffer from the aggregation info  */
294 	mapping = dma_map_single(&bp->pdev->dev,
295 				 first_buf->data + NET_SKB_PAD,
296 				 fp->rx_buf_size, DMA_FROM_DEVICE);
297 	/*
298 	 *  ...if it fails - move the skb from the consumer to the producer
299 	 *  and set the current aggregation state as ERROR to drop it
300 	 *  when TPA_STOP arrives.
301 	 */
302 
303 	if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
304 		/* Move the BD from the consumer to the producer */
305 		bnx2x_reuse_rx_data(fp, cons, prod);
306 		tpa_info->tpa_state = BNX2X_TPA_ERROR;
307 		return;
308 	}
309 
310 	/* move empty data from pool to prod */
311 	prod_rx_buf->data = first_buf->data;
312 	dma_unmap_addr_set(prod_rx_buf, mapping, mapping);
313 	/* point prod_bd to new data */
314 	prod_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
315 	prod_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
316 
317 	/* move partial skb from cons to pool (don't unmap yet) */
318 	*first_buf = *cons_rx_buf;
319 
320 	/* mark bin state as START */
321 	tpa_info->parsing_flags =
322 		le16_to_cpu(cqe->pars_flags.flags);
323 	tpa_info->vlan_tag = le16_to_cpu(cqe->vlan_tag);
324 	tpa_info->tpa_state = BNX2X_TPA_START;
325 	tpa_info->len_on_bd = le16_to_cpu(cqe->len_on_bd);
326 	tpa_info->placement_offset = cqe->placement_offset;
327 	tpa_info->rxhash = bnx2x_get_rxhash(bp, cqe);
328 	if (fp->mode == TPA_MODE_GRO) {
329 		u16 gro_size = le16_to_cpu(cqe->pkt_len_or_gro_seg_len);
330 		tpa_info->full_page =
331 			SGE_PAGE_SIZE * PAGES_PER_SGE / gro_size * gro_size;
332 		/*
333 		 * FW 7.2.16 BUG workaround:
334 		 * if SGE size is (exactly) multiple gro_size
335 		 * fw will place one less frag on SGE.
336 		 * the calculation is done only for potentially
337 		 * dangerous MTUs.
338 		 */
339 		if (unlikely(bp->gro_check))
340 			if (!(SGE_PAGE_SIZE * PAGES_PER_SGE % gro_size))
341 				tpa_info->full_page -= gro_size;
342 		tpa_info->gro_size = gro_size;
343 	}
344 
345 #ifdef BNX2X_STOP_ON_ERROR
346 	fp->tpa_queue_used |= (1 << queue);
347 #ifdef _ASM_GENERIC_INT_L64_H
348 	DP(NETIF_MSG_RX_STATUS, "fp->tpa_queue_used = 0x%lx\n",
349 #else
350 	DP(NETIF_MSG_RX_STATUS, "fp->tpa_queue_used = 0x%llx\n",
351 #endif
352 	   fp->tpa_queue_used);
353 #endif
354 }
355 
356 /* Timestamp option length allowed for TPA aggregation:
357  *
358  *		nop nop kind length echo val
359  */
360 #define TPA_TSTAMP_OPT_LEN	12
361 /**
362  * bnx2x_set_lro_mss - calculate the approximate value of the MSS
363  *
364  * @bp:			driver handle
365  * @parsing_flags:	parsing flags from the START CQE
366  * @len_on_bd:		total length of the first packet for the
367  *			aggregation.
368  *
369  * Approximate value of the MSS for this aggregation calculated using
370  * the first packet of it.
371  */
372 static inline u16 bnx2x_set_lro_mss(struct bnx2x *bp, u16 parsing_flags,
373 				    u16 len_on_bd)
374 {
375 	/*
376 	 * TPA arrgregation won't have either IP options or TCP options
377 	 * other than timestamp or IPv6 extension headers.
378 	 */
379 	u16 hdrs_len = ETH_HLEN + sizeof(struct tcphdr);
380 
381 	if (GET_FLAG(parsing_flags, PARSING_FLAGS_OVER_ETHERNET_PROTOCOL) ==
382 	    PRS_FLAG_OVERETH_IPV6)
383 		hdrs_len += sizeof(struct ipv6hdr);
384 	else /* IPv4 */
385 		hdrs_len += sizeof(struct iphdr);
386 
387 
388 	/* Check if there was a TCP timestamp, if there is it's will
389 	 * always be 12 bytes length: nop nop kind length echo val.
390 	 *
391 	 * Otherwise FW would close the aggregation.
392 	 */
393 	if (parsing_flags & PARSING_FLAGS_TIME_STAMP_EXIST_FLAG)
394 		hdrs_len += TPA_TSTAMP_OPT_LEN;
395 
396 	return len_on_bd - hdrs_len;
397 }
398 
399 static int bnx2x_fill_frag_skb(struct bnx2x *bp, struct bnx2x_fastpath *fp,
400 			       struct bnx2x_agg_info *tpa_info,
401 			       u16 pages,
402 			       struct sk_buff *skb,
403 			       struct eth_end_agg_rx_cqe *cqe,
404 			       u16 cqe_idx)
405 {
406 	struct sw_rx_page *rx_pg, old_rx_pg;
407 	u32 i, frag_len, frag_size;
408 	int err, j, frag_id = 0;
409 	u16 len_on_bd = tpa_info->len_on_bd;
410 	u16 full_page = 0, gro_size = 0;
411 
412 	frag_size = le16_to_cpu(cqe->pkt_len) - len_on_bd;
413 
414 	if (fp->mode == TPA_MODE_GRO) {
415 		gro_size = tpa_info->gro_size;
416 		full_page = tpa_info->full_page;
417 	}
418 
419 	/* This is needed in order to enable forwarding support */
420 	if (frag_size) {
421 		skb_shinfo(skb)->gso_size = bnx2x_set_lro_mss(bp,
422 					tpa_info->parsing_flags, len_on_bd);
423 
424 		/* set for GRO */
425 		if (fp->mode == TPA_MODE_GRO)
426 			skb_shinfo(skb)->gso_type =
427 			    (GET_FLAG(tpa_info->parsing_flags,
428 				      PARSING_FLAGS_OVER_ETHERNET_PROTOCOL) ==
429 						PRS_FLAG_OVERETH_IPV6) ?
430 				SKB_GSO_TCPV6 : SKB_GSO_TCPV4;
431 	}
432 
433 
434 #ifdef BNX2X_STOP_ON_ERROR
435 	if (pages > min_t(u32, 8, MAX_SKB_FRAGS)*SGE_PAGE_SIZE*PAGES_PER_SGE) {
436 		BNX2X_ERR("SGL length is too long: %d. CQE index is %d\n",
437 			  pages, cqe_idx);
438 		BNX2X_ERR("cqe->pkt_len = %d\n", cqe->pkt_len);
439 		bnx2x_panic();
440 		return -EINVAL;
441 	}
442 #endif
443 
444 	/* Run through the SGL and compose the fragmented skb */
445 	for (i = 0, j = 0; i < pages; i += PAGES_PER_SGE, j++) {
446 		u16 sge_idx = RX_SGE(le16_to_cpu(cqe->sgl_or_raw_data.sgl[j]));
447 
448 		/* FW gives the indices of the SGE as if the ring is an array
449 		   (meaning that "next" element will consume 2 indices) */
450 		if (fp->mode == TPA_MODE_GRO)
451 			frag_len = min_t(u32, frag_size, (u32)full_page);
452 		else /* LRO */
453 			frag_len = min_t(u32, frag_size,
454 					 (u32)(SGE_PAGE_SIZE * PAGES_PER_SGE));
455 
456 		rx_pg = &fp->rx_page_ring[sge_idx];
457 		old_rx_pg = *rx_pg;
458 
459 		/* If we fail to allocate a substitute page, we simply stop
460 		   where we are and drop the whole packet */
461 		err = bnx2x_alloc_rx_sge(bp, fp, sge_idx);
462 		if (unlikely(err)) {
463 			fp->eth_q_stats.rx_skb_alloc_failed++;
464 			return err;
465 		}
466 
467 		/* Unmap the page as we r going to pass it to the stack */
468 		dma_unmap_page(&bp->pdev->dev,
469 			       dma_unmap_addr(&old_rx_pg, mapping),
470 			       SGE_PAGE_SIZE*PAGES_PER_SGE, DMA_FROM_DEVICE);
471 		/* Add one frag and update the appropriate fields in the skb */
472 		if (fp->mode == TPA_MODE_LRO)
473 			skb_fill_page_desc(skb, j, old_rx_pg.page, 0, frag_len);
474 		else { /* GRO */
475 			int rem;
476 			int offset = 0;
477 			for (rem = frag_len; rem > 0; rem -= gro_size) {
478 				int len = rem > gro_size ? gro_size : rem;
479 				skb_fill_page_desc(skb, frag_id++,
480 						   old_rx_pg.page, offset, len);
481 				if (offset)
482 					get_page(old_rx_pg.page);
483 				offset += len;
484 			}
485 		}
486 
487 		skb->data_len += frag_len;
488 		skb->truesize += SGE_PAGE_SIZE * PAGES_PER_SGE;
489 		skb->len += frag_len;
490 
491 		frag_size -= frag_len;
492 	}
493 
494 	return 0;
495 }
496 
497 static inline void bnx2x_tpa_stop(struct bnx2x *bp, struct bnx2x_fastpath *fp,
498 				  struct bnx2x_agg_info *tpa_info,
499 				  u16 pages,
500 				  struct eth_end_agg_rx_cqe *cqe,
501 				  u16 cqe_idx)
502 {
503 	struct sw_rx_bd *rx_buf = &tpa_info->first_buf;
504 	u8 pad = tpa_info->placement_offset;
505 	u16 len = tpa_info->len_on_bd;
506 	struct sk_buff *skb = NULL;
507 	u8 *new_data, *data = rx_buf->data;
508 	u8 old_tpa_state = tpa_info->tpa_state;
509 
510 	tpa_info->tpa_state = BNX2X_TPA_STOP;
511 
512 	/* If we there was an error during the handling of the TPA_START -
513 	 * drop this aggregation.
514 	 */
515 	if (old_tpa_state == BNX2X_TPA_ERROR)
516 		goto drop;
517 
518 	/* Try to allocate the new data */
519 	new_data = kmalloc(fp->rx_buf_size + NET_SKB_PAD, GFP_ATOMIC);
520 
521 	/* Unmap skb in the pool anyway, as we are going to change
522 	   pool entry status to BNX2X_TPA_STOP even if new skb allocation
523 	   fails. */
524 	dma_unmap_single(&bp->pdev->dev, dma_unmap_addr(rx_buf, mapping),
525 			 fp->rx_buf_size, DMA_FROM_DEVICE);
526 	if (likely(new_data))
527 		skb = build_skb(data);
528 
529 	if (likely(skb)) {
530 #ifdef BNX2X_STOP_ON_ERROR
531 		if (pad + len > fp->rx_buf_size) {
532 			BNX2X_ERR("skb_put is about to fail...  pad %d  len %d  rx_buf_size %d\n",
533 				  pad, len, fp->rx_buf_size);
534 			bnx2x_panic();
535 			return;
536 		}
537 #endif
538 
539 		skb_reserve(skb, pad + NET_SKB_PAD);
540 		skb_put(skb, len);
541 		skb->rxhash = tpa_info->rxhash;
542 
543 		skb->protocol = eth_type_trans(skb, bp->dev);
544 		skb->ip_summed = CHECKSUM_UNNECESSARY;
545 
546 		if (!bnx2x_fill_frag_skb(bp, fp, tpa_info, pages,
547 					 skb, cqe, cqe_idx)) {
548 			if (tpa_info->parsing_flags & PARSING_FLAGS_VLAN)
549 				__vlan_hwaccel_put_tag(skb, tpa_info->vlan_tag);
550 			napi_gro_receive(&fp->napi, skb);
551 		} else {
552 			DP(NETIF_MSG_RX_STATUS,
553 			   "Failed to allocate new pages - dropping packet!\n");
554 			dev_kfree_skb_any(skb);
555 		}
556 
557 
558 		/* put new data in bin */
559 		rx_buf->data = new_data;
560 
561 		return;
562 	}
563 	kfree(new_data);
564 drop:
565 	/* drop the packet and keep the buffer in the bin */
566 	DP(NETIF_MSG_RX_STATUS,
567 	   "Failed to allocate or map a new skb - dropping packet!\n");
568 	fp->eth_q_stats.rx_skb_alloc_failed++;
569 }
570 
571 
572 int bnx2x_rx_int(struct bnx2x_fastpath *fp, int budget)
573 {
574 	struct bnx2x *bp = fp->bp;
575 	u16 bd_cons, bd_prod, bd_prod_fw, comp_ring_cons;
576 	u16 hw_comp_cons, sw_comp_cons, sw_comp_prod;
577 	int rx_pkt = 0;
578 
579 #ifdef BNX2X_STOP_ON_ERROR
580 	if (unlikely(bp->panic))
581 		return 0;
582 #endif
583 
584 	/* CQ "next element" is of the size of the regular element,
585 	   that's why it's ok here */
586 	hw_comp_cons = le16_to_cpu(*fp->rx_cons_sb);
587 	if ((hw_comp_cons & MAX_RCQ_DESC_CNT) == MAX_RCQ_DESC_CNT)
588 		hw_comp_cons++;
589 
590 	bd_cons = fp->rx_bd_cons;
591 	bd_prod = fp->rx_bd_prod;
592 	bd_prod_fw = bd_prod;
593 	sw_comp_cons = fp->rx_comp_cons;
594 	sw_comp_prod = fp->rx_comp_prod;
595 
596 	/* Memory barrier necessary as speculative reads of the rx
597 	 * buffer can be ahead of the index in the status block
598 	 */
599 	rmb();
600 
601 	DP(NETIF_MSG_RX_STATUS,
602 	   "queue[%d]:  hw_comp_cons %u  sw_comp_cons %u\n",
603 	   fp->index, hw_comp_cons, sw_comp_cons);
604 
605 	while (sw_comp_cons != hw_comp_cons) {
606 		struct sw_rx_bd *rx_buf = NULL;
607 		struct sk_buff *skb;
608 		union eth_rx_cqe *cqe;
609 		struct eth_fast_path_rx_cqe *cqe_fp;
610 		u8 cqe_fp_flags;
611 		enum eth_rx_cqe_type cqe_fp_type;
612 		u16 len, pad, queue;
613 		u8 *data;
614 
615 #ifdef BNX2X_STOP_ON_ERROR
616 		if (unlikely(bp->panic))
617 			return 0;
618 #endif
619 
620 		comp_ring_cons = RCQ_BD(sw_comp_cons);
621 		bd_prod = RX_BD(bd_prod);
622 		bd_cons = RX_BD(bd_cons);
623 
624 		cqe = &fp->rx_comp_ring[comp_ring_cons];
625 		cqe_fp = &cqe->fast_path_cqe;
626 		cqe_fp_flags = cqe_fp->type_error_flags;
627 		cqe_fp_type = cqe_fp_flags & ETH_FAST_PATH_RX_CQE_TYPE;
628 
629 		DP(NETIF_MSG_RX_STATUS,
630 		   "CQE type %x  err %x  status %x  queue %x  vlan %x  len %u\n",
631 		   CQE_TYPE(cqe_fp_flags),
632 		   cqe_fp_flags, cqe_fp->status_flags,
633 		   le32_to_cpu(cqe_fp->rss_hash_result),
634 		   le16_to_cpu(cqe_fp->vlan_tag),
635 		   le16_to_cpu(cqe_fp->pkt_len_or_gro_seg_len));
636 
637 		/* is this a slowpath msg? */
638 		if (unlikely(CQE_TYPE_SLOW(cqe_fp_type))) {
639 			bnx2x_sp_event(fp, cqe);
640 			goto next_cqe;
641 		}
642 
643 		rx_buf = &fp->rx_buf_ring[bd_cons];
644 		data = rx_buf->data;
645 
646 		if (!CQE_TYPE_FAST(cqe_fp_type)) {
647 			struct bnx2x_agg_info *tpa_info;
648 			u16 frag_size, pages;
649 #ifdef BNX2X_STOP_ON_ERROR
650 			/* sanity check */
651 			if (fp->disable_tpa &&
652 			    (CQE_TYPE_START(cqe_fp_type) ||
653 			     CQE_TYPE_STOP(cqe_fp_type)))
654 				BNX2X_ERR("START/STOP packet while disable_tpa type %x\n",
655 					  CQE_TYPE(cqe_fp_type));
656 #endif
657 
658 			if (CQE_TYPE_START(cqe_fp_type)) {
659 				u16 queue = cqe_fp->queue_index;
660 				DP(NETIF_MSG_RX_STATUS,
661 				   "calling tpa_start on queue %d\n",
662 				   queue);
663 
664 				bnx2x_tpa_start(fp, queue,
665 						bd_cons, bd_prod,
666 						cqe_fp);
667 
668 				goto next_rx;
669 
670 			}
671 			queue = cqe->end_agg_cqe.queue_index;
672 			tpa_info = &fp->tpa_info[queue];
673 			DP(NETIF_MSG_RX_STATUS,
674 			   "calling tpa_stop on queue %d\n",
675 			   queue);
676 
677 			frag_size = le16_to_cpu(cqe->end_agg_cqe.pkt_len) -
678 				    tpa_info->len_on_bd;
679 
680 			if (fp->mode == TPA_MODE_GRO)
681 				pages = (frag_size + tpa_info->full_page - 1) /
682 					 tpa_info->full_page;
683 			else
684 				pages = SGE_PAGE_ALIGN(frag_size) >>
685 					SGE_PAGE_SHIFT;
686 
687 			bnx2x_tpa_stop(bp, fp, tpa_info, pages,
688 				       &cqe->end_agg_cqe, comp_ring_cons);
689 #ifdef BNX2X_STOP_ON_ERROR
690 			if (bp->panic)
691 				return 0;
692 #endif
693 
694 			bnx2x_update_sge_prod(fp, pages, &cqe->end_agg_cqe);
695 			goto next_cqe;
696 		}
697 		/* non TPA */
698 		len = le16_to_cpu(cqe_fp->pkt_len_or_gro_seg_len);
699 		pad = cqe_fp->placement_offset;
700 		dma_sync_single_for_cpu(&bp->pdev->dev,
701 					dma_unmap_addr(rx_buf, mapping),
702 					pad + RX_COPY_THRESH,
703 					DMA_FROM_DEVICE);
704 		pad += NET_SKB_PAD;
705 		prefetch(data + pad); /* speedup eth_type_trans() */
706 		/* is this an error packet? */
707 		if (unlikely(cqe_fp_flags & ETH_RX_ERROR_FALGS)) {
708 			DP(NETIF_MSG_RX_ERR | NETIF_MSG_RX_STATUS,
709 			   "ERROR  flags %x  rx packet %u\n",
710 			   cqe_fp_flags, sw_comp_cons);
711 			fp->eth_q_stats.rx_err_discard_pkt++;
712 			goto reuse_rx;
713 		}
714 
715 		/* Since we don't have a jumbo ring
716 		 * copy small packets if mtu > 1500
717 		 */
718 		if ((bp->dev->mtu > ETH_MAX_PACKET_SIZE) &&
719 		    (len <= RX_COPY_THRESH)) {
720 			skb = netdev_alloc_skb_ip_align(bp->dev, len);
721 			if (skb == NULL) {
722 				DP(NETIF_MSG_RX_ERR | NETIF_MSG_RX_STATUS,
723 				   "ERROR  packet dropped because of alloc failure\n");
724 				fp->eth_q_stats.rx_skb_alloc_failed++;
725 				goto reuse_rx;
726 			}
727 			memcpy(skb->data, data + pad, len);
728 			bnx2x_reuse_rx_data(fp, bd_cons, bd_prod);
729 		} else {
730 			if (likely(bnx2x_alloc_rx_data(bp, fp, bd_prod) == 0)) {
731 				dma_unmap_single(&bp->pdev->dev,
732 						 dma_unmap_addr(rx_buf, mapping),
733 						 fp->rx_buf_size,
734 						 DMA_FROM_DEVICE);
735 				skb = build_skb(data);
736 				if (unlikely(!skb)) {
737 					kfree(data);
738 					fp->eth_q_stats.rx_skb_alloc_failed++;
739 					goto next_rx;
740 				}
741 				skb_reserve(skb, pad);
742 			} else {
743 				DP(NETIF_MSG_RX_ERR | NETIF_MSG_RX_STATUS,
744 				   "ERROR  packet dropped because of alloc failure\n");
745 				fp->eth_q_stats.rx_skb_alloc_failed++;
746 reuse_rx:
747 				bnx2x_reuse_rx_data(fp, bd_cons, bd_prod);
748 				goto next_rx;
749 			}
750 		}
751 
752 		skb_put(skb, len);
753 		skb->protocol = eth_type_trans(skb, bp->dev);
754 
755 		/* Set Toeplitz hash for a none-LRO skb */
756 		skb->rxhash = bnx2x_get_rxhash(bp, cqe_fp);
757 
758 		skb_checksum_none_assert(skb);
759 
760 		if (bp->dev->features & NETIF_F_RXCSUM) {
761 
762 			if (likely(BNX2X_RX_CSUM_OK(cqe)))
763 				skb->ip_summed = CHECKSUM_UNNECESSARY;
764 			else
765 				fp->eth_q_stats.hw_csum_err++;
766 		}
767 
768 		skb_record_rx_queue(skb, fp->rx_queue);
769 
770 		if (le16_to_cpu(cqe_fp->pars_flags.flags) &
771 		    PARSING_FLAGS_VLAN)
772 			__vlan_hwaccel_put_tag(skb,
773 					       le16_to_cpu(cqe_fp->vlan_tag));
774 		napi_gro_receive(&fp->napi, skb);
775 
776 
777 next_rx:
778 		rx_buf->data = NULL;
779 
780 		bd_cons = NEXT_RX_IDX(bd_cons);
781 		bd_prod = NEXT_RX_IDX(bd_prod);
782 		bd_prod_fw = NEXT_RX_IDX(bd_prod_fw);
783 		rx_pkt++;
784 next_cqe:
785 		sw_comp_prod = NEXT_RCQ_IDX(sw_comp_prod);
786 		sw_comp_cons = NEXT_RCQ_IDX(sw_comp_cons);
787 
788 		if (rx_pkt == budget)
789 			break;
790 	} /* while */
791 
792 	fp->rx_bd_cons = bd_cons;
793 	fp->rx_bd_prod = bd_prod_fw;
794 	fp->rx_comp_cons = sw_comp_cons;
795 	fp->rx_comp_prod = sw_comp_prod;
796 
797 	/* Update producers */
798 	bnx2x_update_rx_prod(bp, fp, bd_prod_fw, sw_comp_prod,
799 			     fp->rx_sge_prod);
800 
801 	fp->rx_pkt += rx_pkt;
802 	fp->rx_calls++;
803 
804 	return rx_pkt;
805 }
806 
807 static irqreturn_t bnx2x_msix_fp_int(int irq, void *fp_cookie)
808 {
809 	struct bnx2x_fastpath *fp = fp_cookie;
810 	struct bnx2x *bp = fp->bp;
811 	u8 cos;
812 
813 	DP(NETIF_MSG_INTR,
814 	   "got an MSI-X interrupt on IDX:SB [fp %d fw_sd %d igusb %d]\n",
815 	   fp->index, fp->fw_sb_id, fp->igu_sb_id);
816 	bnx2x_ack_sb(bp, fp->igu_sb_id, USTORM_ID, 0, IGU_INT_DISABLE, 0);
817 
818 #ifdef BNX2X_STOP_ON_ERROR
819 	if (unlikely(bp->panic))
820 		return IRQ_HANDLED;
821 #endif
822 
823 	/* Handle Rx and Tx according to MSI-X vector */
824 	prefetch(fp->rx_cons_sb);
825 
826 	for_each_cos_in_tx_queue(fp, cos)
827 		prefetch(fp->txdata[cos].tx_cons_sb);
828 
829 	prefetch(&fp->sb_running_index[SM_RX_ID]);
830 	napi_schedule(&bnx2x_fp(bp, fp->index, napi));
831 
832 	return IRQ_HANDLED;
833 }
834 
835 /* HW Lock for shared dual port PHYs */
836 void bnx2x_acquire_phy_lock(struct bnx2x *bp)
837 {
838 	mutex_lock(&bp->port.phy_mutex);
839 
840 	if (bp->port.need_hw_lock)
841 		bnx2x_acquire_hw_lock(bp, HW_LOCK_RESOURCE_MDIO);
842 }
843 
844 void bnx2x_release_phy_lock(struct bnx2x *bp)
845 {
846 	if (bp->port.need_hw_lock)
847 		bnx2x_release_hw_lock(bp, HW_LOCK_RESOURCE_MDIO);
848 
849 	mutex_unlock(&bp->port.phy_mutex);
850 }
851 
852 /* calculates MF speed according to current linespeed and MF configuration */
853 u16 bnx2x_get_mf_speed(struct bnx2x *bp)
854 {
855 	u16 line_speed = bp->link_vars.line_speed;
856 	if (IS_MF(bp)) {
857 		u16 maxCfg = bnx2x_extract_max_cfg(bp,
858 						   bp->mf_config[BP_VN(bp)]);
859 
860 		/* Calculate the current MAX line speed limit for the MF
861 		 * devices
862 		 */
863 		if (IS_MF_SI(bp))
864 			line_speed = (line_speed * maxCfg) / 100;
865 		else { /* SD mode */
866 			u16 vn_max_rate = maxCfg * 100;
867 
868 			if (vn_max_rate < line_speed)
869 				line_speed = vn_max_rate;
870 		}
871 	}
872 
873 	return line_speed;
874 }
875 
876 /**
877  * bnx2x_fill_report_data - fill link report data to report
878  *
879  * @bp:		driver handle
880  * @data:	link state to update
881  *
882  * It uses a none-atomic bit operations because is called under the mutex.
883  */
884 static inline void bnx2x_fill_report_data(struct bnx2x *bp,
885 					  struct bnx2x_link_report_data *data)
886 {
887 	u16 line_speed = bnx2x_get_mf_speed(bp);
888 
889 	memset(data, 0, sizeof(*data));
890 
891 	/* Fill the report data: efective line speed */
892 	data->line_speed = line_speed;
893 
894 	/* Link is down */
895 	if (!bp->link_vars.link_up || (bp->flags & MF_FUNC_DIS))
896 		__set_bit(BNX2X_LINK_REPORT_LINK_DOWN,
897 			  &data->link_report_flags);
898 
899 	/* Full DUPLEX */
900 	if (bp->link_vars.duplex == DUPLEX_FULL)
901 		__set_bit(BNX2X_LINK_REPORT_FD, &data->link_report_flags);
902 
903 	/* Rx Flow Control is ON */
904 	if (bp->link_vars.flow_ctrl & BNX2X_FLOW_CTRL_RX)
905 		__set_bit(BNX2X_LINK_REPORT_RX_FC_ON, &data->link_report_flags);
906 
907 	/* Tx Flow Control is ON */
908 	if (bp->link_vars.flow_ctrl & BNX2X_FLOW_CTRL_TX)
909 		__set_bit(BNX2X_LINK_REPORT_TX_FC_ON, &data->link_report_flags);
910 }
911 
912 /**
913  * bnx2x_link_report - report link status to OS.
914  *
915  * @bp:		driver handle
916  *
917  * Calls the __bnx2x_link_report() under the same locking scheme
918  * as a link/PHY state managing code to ensure a consistent link
919  * reporting.
920  */
921 
922 void bnx2x_link_report(struct bnx2x *bp)
923 {
924 	bnx2x_acquire_phy_lock(bp);
925 	__bnx2x_link_report(bp);
926 	bnx2x_release_phy_lock(bp);
927 }
928 
929 /**
930  * __bnx2x_link_report - report link status to OS.
931  *
932  * @bp:		driver handle
933  *
934  * None atomic inmlementation.
935  * Should be called under the phy_lock.
936  */
937 void __bnx2x_link_report(struct bnx2x *bp)
938 {
939 	struct bnx2x_link_report_data cur_data;
940 
941 	/* reread mf_cfg */
942 	if (!CHIP_IS_E1(bp))
943 		bnx2x_read_mf_cfg(bp);
944 
945 	/* Read the current link report info */
946 	bnx2x_fill_report_data(bp, &cur_data);
947 
948 	/* Don't report link down or exactly the same link status twice */
949 	if (!memcmp(&cur_data, &bp->last_reported_link, sizeof(cur_data)) ||
950 	    (test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
951 		      &bp->last_reported_link.link_report_flags) &&
952 	     test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
953 		      &cur_data.link_report_flags)))
954 		return;
955 
956 	bp->link_cnt++;
957 
958 	/* We are going to report a new link parameters now -
959 	 * remember the current data for the next time.
960 	 */
961 	memcpy(&bp->last_reported_link, &cur_data, sizeof(cur_data));
962 
963 	if (test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
964 		     &cur_data.link_report_flags)) {
965 		netif_carrier_off(bp->dev);
966 		netdev_err(bp->dev, "NIC Link is Down\n");
967 		return;
968 	} else {
969 		const char *duplex;
970 		const char *flow;
971 
972 		netif_carrier_on(bp->dev);
973 
974 		if (test_and_clear_bit(BNX2X_LINK_REPORT_FD,
975 				       &cur_data.link_report_flags))
976 			duplex = "full";
977 		else
978 			duplex = "half";
979 
980 		/* Handle the FC at the end so that only these flags would be
981 		 * possibly set. This way we may easily check if there is no FC
982 		 * enabled.
983 		 */
984 		if (cur_data.link_report_flags) {
985 			if (test_bit(BNX2X_LINK_REPORT_RX_FC_ON,
986 				     &cur_data.link_report_flags)) {
987 				if (test_bit(BNX2X_LINK_REPORT_TX_FC_ON,
988 				     &cur_data.link_report_flags))
989 					flow = "ON - receive & transmit";
990 				else
991 					flow = "ON - receive";
992 			} else {
993 				flow = "ON - transmit";
994 			}
995 		} else {
996 			flow = "none";
997 		}
998 		netdev_info(bp->dev, "NIC Link is Up, %d Mbps %s duplex, Flow control: %s\n",
999 			    cur_data.line_speed, duplex, flow);
1000 	}
1001 }
1002 
1003 void bnx2x_init_rx_rings(struct bnx2x *bp)
1004 {
1005 	int func = BP_FUNC(bp);
1006 	u16 ring_prod;
1007 	int i, j;
1008 
1009 	/* Allocate TPA resources */
1010 	for_each_rx_queue(bp, j) {
1011 		struct bnx2x_fastpath *fp = &bp->fp[j];
1012 
1013 		DP(NETIF_MSG_IFUP,
1014 		   "mtu %d  rx_buf_size %d\n", bp->dev->mtu, fp->rx_buf_size);
1015 
1016 		if (!fp->disable_tpa) {
1017 			/* Fill the per-aggregtion pool */
1018 			for (i = 0; i < MAX_AGG_QS(bp); i++) {
1019 				struct bnx2x_agg_info *tpa_info =
1020 					&fp->tpa_info[i];
1021 				struct sw_rx_bd *first_buf =
1022 					&tpa_info->first_buf;
1023 
1024 				first_buf->data = kmalloc(fp->rx_buf_size + NET_SKB_PAD,
1025 							  GFP_ATOMIC);
1026 				if (!first_buf->data) {
1027 					BNX2X_ERR("Failed to allocate TPA skb pool for queue[%d] - disabling TPA on this queue!\n",
1028 						  j);
1029 					bnx2x_free_tpa_pool(bp, fp, i);
1030 					fp->disable_tpa = 1;
1031 					break;
1032 				}
1033 				dma_unmap_addr_set(first_buf, mapping, 0);
1034 				tpa_info->tpa_state = BNX2X_TPA_STOP;
1035 			}
1036 
1037 			/* "next page" elements initialization */
1038 			bnx2x_set_next_page_sgl(fp);
1039 
1040 			/* set SGEs bit mask */
1041 			bnx2x_init_sge_ring_bit_mask(fp);
1042 
1043 			/* Allocate SGEs and initialize the ring elements */
1044 			for (i = 0, ring_prod = 0;
1045 			     i < MAX_RX_SGE_CNT*NUM_RX_SGE_PAGES; i++) {
1046 
1047 				if (bnx2x_alloc_rx_sge(bp, fp, ring_prod) < 0) {
1048 					BNX2X_ERR("was only able to allocate %d rx sges\n",
1049 						  i);
1050 					BNX2X_ERR("disabling TPA for queue[%d]\n",
1051 						  j);
1052 					/* Cleanup already allocated elements */
1053 					bnx2x_free_rx_sge_range(bp, fp,
1054 								ring_prod);
1055 					bnx2x_free_tpa_pool(bp, fp,
1056 							    MAX_AGG_QS(bp));
1057 					fp->disable_tpa = 1;
1058 					ring_prod = 0;
1059 					break;
1060 				}
1061 				ring_prod = NEXT_SGE_IDX(ring_prod);
1062 			}
1063 
1064 			fp->rx_sge_prod = ring_prod;
1065 		}
1066 	}
1067 
1068 	for_each_rx_queue(bp, j) {
1069 		struct bnx2x_fastpath *fp = &bp->fp[j];
1070 
1071 		fp->rx_bd_cons = 0;
1072 
1073 		/* Activate BD ring */
1074 		/* Warning!
1075 		 * this will generate an interrupt (to the TSTORM)
1076 		 * must only be done after chip is initialized
1077 		 */
1078 		bnx2x_update_rx_prod(bp, fp, fp->rx_bd_prod, fp->rx_comp_prod,
1079 				     fp->rx_sge_prod);
1080 
1081 		if (j != 0)
1082 			continue;
1083 
1084 		if (CHIP_IS_E1(bp)) {
1085 			REG_WR(bp, BAR_USTRORM_INTMEM +
1086 			       USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func),
1087 			       U64_LO(fp->rx_comp_mapping));
1088 			REG_WR(bp, BAR_USTRORM_INTMEM +
1089 			       USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func) + 4,
1090 			       U64_HI(fp->rx_comp_mapping));
1091 		}
1092 	}
1093 }
1094 
1095 static void bnx2x_free_tx_skbs(struct bnx2x *bp)
1096 {
1097 	int i;
1098 	u8 cos;
1099 
1100 	for_each_tx_queue(bp, i) {
1101 		struct bnx2x_fastpath *fp = &bp->fp[i];
1102 		for_each_cos_in_tx_queue(fp, cos) {
1103 			struct bnx2x_fp_txdata *txdata = &fp->txdata[cos];
1104 			unsigned pkts_compl = 0, bytes_compl = 0;
1105 
1106 			u16 sw_prod = txdata->tx_pkt_prod;
1107 			u16 sw_cons = txdata->tx_pkt_cons;
1108 
1109 			while (sw_cons != sw_prod) {
1110 				bnx2x_free_tx_pkt(bp, txdata, TX_BD(sw_cons),
1111 				    &pkts_compl, &bytes_compl);
1112 				sw_cons++;
1113 			}
1114 			netdev_tx_reset_queue(
1115 			    netdev_get_tx_queue(bp->dev, txdata->txq_index));
1116 		}
1117 	}
1118 }
1119 
1120 static void bnx2x_free_rx_bds(struct bnx2x_fastpath *fp)
1121 {
1122 	struct bnx2x *bp = fp->bp;
1123 	int i;
1124 
1125 	/* ring wasn't allocated */
1126 	if (fp->rx_buf_ring == NULL)
1127 		return;
1128 
1129 	for (i = 0; i < NUM_RX_BD; i++) {
1130 		struct sw_rx_bd *rx_buf = &fp->rx_buf_ring[i];
1131 		u8 *data = rx_buf->data;
1132 
1133 		if (data == NULL)
1134 			continue;
1135 		dma_unmap_single(&bp->pdev->dev,
1136 				 dma_unmap_addr(rx_buf, mapping),
1137 				 fp->rx_buf_size, DMA_FROM_DEVICE);
1138 
1139 		rx_buf->data = NULL;
1140 		kfree(data);
1141 	}
1142 }
1143 
1144 static void bnx2x_free_rx_skbs(struct bnx2x *bp)
1145 {
1146 	int j;
1147 
1148 	for_each_rx_queue(bp, j) {
1149 		struct bnx2x_fastpath *fp = &bp->fp[j];
1150 
1151 		bnx2x_free_rx_bds(fp);
1152 
1153 		if (!fp->disable_tpa)
1154 			bnx2x_free_tpa_pool(bp, fp, MAX_AGG_QS(bp));
1155 	}
1156 }
1157 
1158 void bnx2x_free_skbs(struct bnx2x *bp)
1159 {
1160 	bnx2x_free_tx_skbs(bp);
1161 	bnx2x_free_rx_skbs(bp);
1162 }
1163 
1164 void bnx2x_update_max_mf_config(struct bnx2x *bp, u32 value)
1165 {
1166 	/* load old values */
1167 	u32 mf_cfg = bp->mf_config[BP_VN(bp)];
1168 
1169 	if (value != bnx2x_extract_max_cfg(bp, mf_cfg)) {
1170 		/* leave all but MAX value */
1171 		mf_cfg &= ~FUNC_MF_CFG_MAX_BW_MASK;
1172 
1173 		/* set new MAX value */
1174 		mf_cfg |= (value << FUNC_MF_CFG_MAX_BW_SHIFT)
1175 				& FUNC_MF_CFG_MAX_BW_MASK;
1176 
1177 		bnx2x_fw_command(bp, DRV_MSG_CODE_SET_MF_BW, mf_cfg);
1178 	}
1179 }
1180 
1181 /**
1182  * bnx2x_free_msix_irqs - free previously requested MSI-X IRQ vectors
1183  *
1184  * @bp:		driver handle
1185  * @nvecs:	number of vectors to be released
1186  */
1187 static void bnx2x_free_msix_irqs(struct bnx2x *bp, int nvecs)
1188 {
1189 	int i, offset = 0;
1190 
1191 	if (nvecs == offset)
1192 		return;
1193 	free_irq(bp->msix_table[offset].vector, bp->dev);
1194 	DP(NETIF_MSG_IFDOWN, "released sp irq (%d)\n",
1195 	   bp->msix_table[offset].vector);
1196 	offset++;
1197 #ifdef BCM_CNIC
1198 	if (nvecs == offset)
1199 		return;
1200 	offset++;
1201 #endif
1202 
1203 	for_each_eth_queue(bp, i) {
1204 		if (nvecs == offset)
1205 			return;
1206 		DP(NETIF_MSG_IFDOWN, "about to release fp #%d->%d irq\n",
1207 		   i, bp->msix_table[offset].vector);
1208 
1209 		free_irq(bp->msix_table[offset++].vector, &bp->fp[i]);
1210 	}
1211 }
1212 
1213 void bnx2x_free_irq(struct bnx2x *bp)
1214 {
1215 	if (bp->flags & USING_MSIX_FLAG)
1216 		bnx2x_free_msix_irqs(bp, BNX2X_NUM_ETH_QUEUES(bp) +
1217 				     CNIC_PRESENT + 1);
1218 	else if (bp->flags & USING_MSI_FLAG)
1219 		free_irq(bp->pdev->irq, bp->dev);
1220 	else
1221 		free_irq(bp->pdev->irq, bp->dev);
1222 }
1223 
1224 int bnx2x_enable_msix(struct bnx2x *bp)
1225 {
1226 	int msix_vec = 0, i, rc, req_cnt;
1227 
1228 	bp->msix_table[msix_vec].entry = msix_vec;
1229 	BNX2X_DEV_INFO("msix_table[0].entry = %d (slowpath)\n",
1230 	   bp->msix_table[0].entry);
1231 	msix_vec++;
1232 
1233 #ifdef BCM_CNIC
1234 	bp->msix_table[msix_vec].entry = msix_vec;
1235 	BNX2X_DEV_INFO("msix_table[%d].entry = %d (CNIC)\n",
1236 	   bp->msix_table[msix_vec].entry, bp->msix_table[msix_vec].entry);
1237 	msix_vec++;
1238 #endif
1239 	/* We need separate vectors for ETH queues only (not FCoE) */
1240 	for_each_eth_queue(bp, i) {
1241 		bp->msix_table[msix_vec].entry = msix_vec;
1242 		BNX2X_DEV_INFO("msix_table[%d].entry = %d (fastpath #%u)\n",
1243 			       msix_vec, msix_vec, i);
1244 		msix_vec++;
1245 	}
1246 
1247 	req_cnt = BNX2X_NUM_ETH_QUEUES(bp) + CNIC_PRESENT + 1;
1248 
1249 	rc = pci_enable_msix(bp->pdev, &bp->msix_table[0], req_cnt);
1250 
1251 	/*
1252 	 * reconfigure number of tx/rx queues according to available
1253 	 * MSI-X vectors
1254 	 */
1255 	if (rc >= BNX2X_MIN_MSIX_VEC_CNT) {
1256 		/* how less vectors we will have? */
1257 		int diff = req_cnt - rc;
1258 
1259 		BNX2X_DEV_INFO("Trying to use less MSI-X vectors: %d\n", rc);
1260 
1261 		rc = pci_enable_msix(bp->pdev, &bp->msix_table[0], rc);
1262 
1263 		if (rc) {
1264 			BNX2X_DEV_INFO("MSI-X is not attainable  rc %d\n", rc);
1265 			return rc;
1266 		}
1267 		/*
1268 		 * decrease number of queues by number of unallocated entries
1269 		 */
1270 		bp->num_queues -= diff;
1271 
1272 		BNX2X_DEV_INFO("New queue configuration set: %d\n",
1273 				  bp->num_queues);
1274 	} else if (rc) {
1275 		/* fall to INTx if not enough memory */
1276 		if (rc == -ENOMEM)
1277 			bp->flags |= DISABLE_MSI_FLAG;
1278 		BNX2X_DEV_INFO("MSI-X is not attainable  rc %d\n", rc);
1279 		return rc;
1280 	}
1281 
1282 	bp->flags |= USING_MSIX_FLAG;
1283 
1284 	return 0;
1285 }
1286 
1287 static int bnx2x_req_msix_irqs(struct bnx2x *bp)
1288 {
1289 	int i, rc, offset = 0;
1290 
1291 	rc = request_irq(bp->msix_table[offset++].vector,
1292 			 bnx2x_msix_sp_int, 0,
1293 			 bp->dev->name, bp->dev);
1294 	if (rc) {
1295 		BNX2X_ERR("request sp irq failed\n");
1296 		return -EBUSY;
1297 	}
1298 
1299 #ifdef BCM_CNIC
1300 	offset++;
1301 #endif
1302 	for_each_eth_queue(bp, i) {
1303 		struct bnx2x_fastpath *fp = &bp->fp[i];
1304 		snprintf(fp->name, sizeof(fp->name), "%s-fp-%d",
1305 			 bp->dev->name, i);
1306 
1307 		rc = request_irq(bp->msix_table[offset].vector,
1308 				 bnx2x_msix_fp_int, 0, fp->name, fp);
1309 		if (rc) {
1310 			BNX2X_ERR("request fp #%d irq (%d) failed  rc %d\n", i,
1311 			      bp->msix_table[offset].vector, rc);
1312 			bnx2x_free_msix_irqs(bp, offset);
1313 			return -EBUSY;
1314 		}
1315 
1316 		offset++;
1317 	}
1318 
1319 	i = BNX2X_NUM_ETH_QUEUES(bp);
1320 	offset = 1 + CNIC_PRESENT;
1321 	netdev_info(bp->dev, "using MSI-X  IRQs: sp %d  fp[%d] %d ... fp[%d] %d\n",
1322 	       bp->msix_table[0].vector,
1323 	       0, bp->msix_table[offset].vector,
1324 	       i - 1, bp->msix_table[offset + i - 1].vector);
1325 
1326 	return 0;
1327 }
1328 
1329 int bnx2x_enable_msi(struct bnx2x *bp)
1330 {
1331 	int rc;
1332 
1333 	rc = pci_enable_msi(bp->pdev);
1334 	if (rc) {
1335 		BNX2X_DEV_INFO("MSI is not attainable\n");
1336 		return -1;
1337 	}
1338 	bp->flags |= USING_MSI_FLAG;
1339 
1340 	return 0;
1341 }
1342 
1343 static int bnx2x_req_irq(struct bnx2x *bp)
1344 {
1345 	unsigned long flags;
1346 	int rc;
1347 
1348 	if (bp->flags & USING_MSI_FLAG)
1349 		flags = 0;
1350 	else
1351 		flags = IRQF_SHARED;
1352 
1353 	rc = request_irq(bp->pdev->irq, bnx2x_interrupt, flags,
1354 			 bp->dev->name, bp->dev);
1355 	return rc;
1356 }
1357 
1358 static inline int bnx2x_setup_irqs(struct bnx2x *bp)
1359 {
1360 	int rc = 0;
1361 	if (bp->flags & USING_MSIX_FLAG) {
1362 		rc = bnx2x_req_msix_irqs(bp);
1363 		if (rc)
1364 			return rc;
1365 	} else {
1366 		bnx2x_ack_int(bp);
1367 		rc = bnx2x_req_irq(bp);
1368 		if (rc) {
1369 			BNX2X_ERR("IRQ request failed  rc %d, aborting\n", rc);
1370 			return rc;
1371 		}
1372 		if (bp->flags & USING_MSI_FLAG) {
1373 			bp->dev->irq = bp->pdev->irq;
1374 			netdev_info(bp->dev, "using MSI  IRQ %d\n",
1375 			       bp->pdev->irq);
1376 		}
1377 	}
1378 
1379 	return 0;
1380 }
1381 
1382 static inline void bnx2x_napi_enable(struct bnx2x *bp)
1383 {
1384 	int i;
1385 
1386 	for_each_rx_queue(bp, i)
1387 		napi_enable(&bnx2x_fp(bp, i, napi));
1388 }
1389 
1390 static inline void bnx2x_napi_disable(struct bnx2x *bp)
1391 {
1392 	int i;
1393 
1394 	for_each_rx_queue(bp, i)
1395 		napi_disable(&bnx2x_fp(bp, i, napi));
1396 }
1397 
1398 void bnx2x_netif_start(struct bnx2x *bp)
1399 {
1400 	if (netif_running(bp->dev)) {
1401 		bnx2x_napi_enable(bp);
1402 		bnx2x_int_enable(bp);
1403 		if (bp->state == BNX2X_STATE_OPEN)
1404 			netif_tx_wake_all_queues(bp->dev);
1405 	}
1406 }
1407 
1408 void bnx2x_netif_stop(struct bnx2x *bp, int disable_hw)
1409 {
1410 	bnx2x_int_disable_sync(bp, disable_hw);
1411 	bnx2x_napi_disable(bp);
1412 }
1413 
1414 u16 bnx2x_select_queue(struct net_device *dev, struct sk_buff *skb)
1415 {
1416 	struct bnx2x *bp = netdev_priv(dev);
1417 
1418 #ifdef BCM_CNIC
1419 	if (!NO_FCOE(bp)) {
1420 		struct ethhdr *hdr = (struct ethhdr *)skb->data;
1421 		u16 ether_type = ntohs(hdr->h_proto);
1422 
1423 		/* Skip VLAN tag if present */
1424 		if (ether_type == ETH_P_8021Q) {
1425 			struct vlan_ethhdr *vhdr =
1426 				(struct vlan_ethhdr *)skb->data;
1427 
1428 			ether_type = ntohs(vhdr->h_vlan_encapsulated_proto);
1429 		}
1430 
1431 		/* If ethertype is FCoE or FIP - use FCoE ring */
1432 		if ((ether_type == ETH_P_FCOE) || (ether_type == ETH_P_FIP))
1433 			return bnx2x_fcoe_tx(bp, txq_index);
1434 	}
1435 #endif
1436 	/* select a non-FCoE queue */
1437 	return __skb_tx_hash(dev, skb, BNX2X_NUM_ETH_QUEUES(bp));
1438 }
1439 
1440 void bnx2x_set_num_queues(struct bnx2x *bp)
1441 {
1442 	switch (bp->multi_mode) {
1443 	case ETH_RSS_MODE_DISABLED:
1444 		bp->num_queues = 1;
1445 		break;
1446 	case ETH_RSS_MODE_REGULAR:
1447 		bp->num_queues = bnx2x_calc_num_queues(bp);
1448 		break;
1449 
1450 	default:
1451 		bp->num_queues = 1;
1452 		break;
1453 	}
1454 
1455 #ifdef BCM_CNIC
1456 	/* override in STORAGE SD mode */
1457 	if (IS_MF_STORAGE_SD(bp))
1458 		bp->num_queues = 1;
1459 #endif
1460 	/* Add special queues */
1461 	bp->num_queues += NON_ETH_CONTEXT_USE;
1462 }
1463 
1464 /**
1465  * bnx2x_set_real_num_queues - configure netdev->real_num_[tx,rx]_queues
1466  *
1467  * @bp:		Driver handle
1468  *
1469  * We currently support for at most 16 Tx queues for each CoS thus we will
1470  * allocate a multiple of 16 for ETH L2 rings according to the value of the
1471  * bp->max_cos.
1472  *
1473  * If there is an FCoE L2 queue the appropriate Tx queue will have the next
1474  * index after all ETH L2 indices.
1475  *
1476  * If the actual number of Tx queues (for each CoS) is less than 16 then there
1477  * will be the holes at the end of each group of 16 ETh L2 indices (0..15,
1478  * 16..31,...) with indicies that are not coupled with any real Tx queue.
1479  *
1480  * The proper configuration of skb->queue_mapping is handled by
1481  * bnx2x_select_queue() and __skb_tx_hash().
1482  *
1483  * bnx2x_setup_tc() takes care of the proper TC mappings so that __skb_tx_hash()
1484  * will return a proper Tx index if TC is enabled (netdev->num_tc > 0).
1485  */
1486 static inline int bnx2x_set_real_num_queues(struct bnx2x *bp)
1487 {
1488 	int rc, tx, rx;
1489 
1490 	tx = MAX_TXQS_PER_COS * bp->max_cos;
1491 	rx = BNX2X_NUM_ETH_QUEUES(bp);
1492 
1493 /* account for fcoe queue */
1494 #ifdef BCM_CNIC
1495 	if (!NO_FCOE(bp)) {
1496 		rx += FCOE_PRESENT;
1497 		tx += FCOE_PRESENT;
1498 	}
1499 #endif
1500 
1501 	rc = netif_set_real_num_tx_queues(bp->dev, tx);
1502 	if (rc) {
1503 		BNX2X_ERR("Failed to set real number of Tx queues: %d\n", rc);
1504 		return rc;
1505 	}
1506 	rc = netif_set_real_num_rx_queues(bp->dev, rx);
1507 	if (rc) {
1508 		BNX2X_ERR("Failed to set real number of Rx queues: %d\n", rc);
1509 		return rc;
1510 	}
1511 
1512 	DP(NETIF_MSG_IFUP, "Setting real num queues to (tx, rx) (%d, %d)\n",
1513 			  tx, rx);
1514 
1515 	return rc;
1516 }
1517 
1518 static inline void bnx2x_set_rx_buf_size(struct bnx2x *bp)
1519 {
1520 	int i;
1521 
1522 	for_each_queue(bp, i) {
1523 		struct bnx2x_fastpath *fp = &bp->fp[i];
1524 		u32 mtu;
1525 
1526 		/* Always use a mini-jumbo MTU for the FCoE L2 ring */
1527 		if (IS_FCOE_IDX(i))
1528 			/*
1529 			 * Although there are no IP frames expected to arrive to
1530 			 * this ring we still want to add an
1531 			 * IP_HEADER_ALIGNMENT_PADDING to prevent a buffer
1532 			 * overrun attack.
1533 			 */
1534 			mtu = BNX2X_FCOE_MINI_JUMBO_MTU;
1535 		else
1536 			mtu = bp->dev->mtu;
1537 		fp->rx_buf_size = BNX2X_FW_RX_ALIGN_START +
1538 				  IP_HEADER_ALIGNMENT_PADDING +
1539 				  ETH_OVREHEAD +
1540 				  mtu +
1541 				  BNX2X_FW_RX_ALIGN_END;
1542 		/* Note : rx_buf_size doesnt take into account NET_SKB_PAD */
1543 	}
1544 }
1545 
1546 static inline int bnx2x_init_rss_pf(struct bnx2x *bp)
1547 {
1548 	int i;
1549 	u8 ind_table[T_ETH_INDIRECTION_TABLE_SIZE] = {0};
1550 	u8 num_eth_queues = BNX2X_NUM_ETH_QUEUES(bp);
1551 
1552 	/*
1553 	 * Prepare the inital contents fo the indirection table if RSS is
1554 	 * enabled
1555 	 */
1556 	if (bp->multi_mode != ETH_RSS_MODE_DISABLED) {
1557 		for (i = 0; i < sizeof(ind_table); i++)
1558 			ind_table[i] =
1559 				bp->fp->cl_id +
1560 				ethtool_rxfh_indir_default(i, num_eth_queues);
1561 	}
1562 
1563 	/*
1564 	 * For 57710 and 57711 SEARCHER configuration (rss_keys) is
1565 	 * per-port, so if explicit configuration is needed , do it only
1566 	 * for a PMF.
1567 	 *
1568 	 * For 57712 and newer on the other hand it's a per-function
1569 	 * configuration.
1570 	 */
1571 	return bnx2x_config_rss_pf(bp, ind_table,
1572 				   bp->port.pmf || !CHIP_IS_E1x(bp));
1573 }
1574 
1575 int bnx2x_config_rss_pf(struct bnx2x *bp, u8 *ind_table, bool config_hash)
1576 {
1577 	struct bnx2x_config_rss_params params = {NULL};
1578 	int i;
1579 
1580 	/* Although RSS is meaningless when there is a single HW queue we
1581 	 * still need it enabled in order to have HW Rx hash generated.
1582 	 *
1583 	 * if (!is_eth_multi(bp))
1584 	 *      bp->multi_mode = ETH_RSS_MODE_DISABLED;
1585 	 */
1586 
1587 	params.rss_obj = &bp->rss_conf_obj;
1588 
1589 	__set_bit(RAMROD_COMP_WAIT, &params.ramrod_flags);
1590 
1591 	/* RSS mode */
1592 	switch (bp->multi_mode) {
1593 	case ETH_RSS_MODE_DISABLED:
1594 		__set_bit(BNX2X_RSS_MODE_DISABLED, &params.rss_flags);
1595 		break;
1596 	case ETH_RSS_MODE_REGULAR:
1597 		__set_bit(BNX2X_RSS_MODE_REGULAR, &params.rss_flags);
1598 		break;
1599 	case ETH_RSS_MODE_VLAN_PRI:
1600 		__set_bit(BNX2X_RSS_MODE_VLAN_PRI, &params.rss_flags);
1601 		break;
1602 	case ETH_RSS_MODE_E1HOV_PRI:
1603 		__set_bit(BNX2X_RSS_MODE_E1HOV_PRI, &params.rss_flags);
1604 		break;
1605 	case ETH_RSS_MODE_IP_DSCP:
1606 		__set_bit(BNX2X_RSS_MODE_IP_DSCP, &params.rss_flags);
1607 		break;
1608 	default:
1609 		BNX2X_ERR("Unknown multi_mode: %d\n", bp->multi_mode);
1610 		return -EINVAL;
1611 	}
1612 
1613 	/* If RSS is enabled */
1614 	if (bp->multi_mode != ETH_RSS_MODE_DISABLED) {
1615 		/* RSS configuration */
1616 		__set_bit(BNX2X_RSS_IPV4, &params.rss_flags);
1617 		__set_bit(BNX2X_RSS_IPV4_TCP, &params.rss_flags);
1618 		__set_bit(BNX2X_RSS_IPV6, &params.rss_flags);
1619 		__set_bit(BNX2X_RSS_IPV6_TCP, &params.rss_flags);
1620 
1621 		/* Hash bits */
1622 		params.rss_result_mask = MULTI_MASK;
1623 
1624 		memcpy(params.ind_table, ind_table, sizeof(params.ind_table));
1625 
1626 		if (config_hash) {
1627 			/* RSS keys */
1628 			for (i = 0; i < sizeof(params.rss_key) / 4; i++)
1629 				params.rss_key[i] = random32();
1630 
1631 			__set_bit(BNX2X_RSS_SET_SRCH, &params.rss_flags);
1632 		}
1633 	}
1634 
1635 	return bnx2x_config_rss(bp, &params);
1636 }
1637 
1638 static inline int bnx2x_init_hw(struct bnx2x *bp, u32 load_code)
1639 {
1640 	struct bnx2x_func_state_params func_params = {NULL};
1641 
1642 	/* Prepare parameters for function state transitions */
1643 	__set_bit(RAMROD_COMP_WAIT, &func_params.ramrod_flags);
1644 
1645 	func_params.f_obj = &bp->func_obj;
1646 	func_params.cmd = BNX2X_F_CMD_HW_INIT;
1647 
1648 	func_params.params.hw_init.load_phase = load_code;
1649 
1650 	return bnx2x_func_state_change(bp, &func_params);
1651 }
1652 
1653 /*
1654  * Cleans the object that have internal lists without sending
1655  * ramrods. Should be run when interrutps are disabled.
1656  */
1657 static void bnx2x_squeeze_objects(struct bnx2x *bp)
1658 {
1659 	int rc;
1660 	unsigned long ramrod_flags = 0, vlan_mac_flags = 0;
1661 	struct bnx2x_mcast_ramrod_params rparam = {NULL};
1662 	struct bnx2x_vlan_mac_obj *mac_obj = &bp->fp->mac_obj;
1663 
1664 	/***************** Cleanup MACs' object first *************************/
1665 
1666 	/* Wait for completion of requested */
1667 	__set_bit(RAMROD_COMP_WAIT, &ramrod_flags);
1668 	/* Perform a dry cleanup */
1669 	__set_bit(RAMROD_DRV_CLR_ONLY, &ramrod_flags);
1670 
1671 	/* Clean ETH primary MAC */
1672 	__set_bit(BNX2X_ETH_MAC, &vlan_mac_flags);
1673 	rc = mac_obj->delete_all(bp, &bp->fp->mac_obj, &vlan_mac_flags,
1674 				 &ramrod_flags);
1675 	if (rc != 0)
1676 		BNX2X_ERR("Failed to clean ETH MACs: %d\n", rc);
1677 
1678 	/* Cleanup UC list */
1679 	vlan_mac_flags = 0;
1680 	__set_bit(BNX2X_UC_LIST_MAC, &vlan_mac_flags);
1681 	rc = mac_obj->delete_all(bp, mac_obj, &vlan_mac_flags,
1682 				 &ramrod_flags);
1683 	if (rc != 0)
1684 		BNX2X_ERR("Failed to clean UC list MACs: %d\n", rc);
1685 
1686 	/***************** Now clean mcast object *****************************/
1687 	rparam.mcast_obj = &bp->mcast_obj;
1688 	__set_bit(RAMROD_DRV_CLR_ONLY, &rparam.ramrod_flags);
1689 
1690 	/* Add a DEL command... */
1691 	rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_DEL);
1692 	if (rc < 0)
1693 		BNX2X_ERR("Failed to add a new DEL command to a multi-cast object: %d\n",
1694 			  rc);
1695 
1696 	/* ...and wait until all pending commands are cleared */
1697 	rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_CONT);
1698 	while (rc != 0) {
1699 		if (rc < 0) {
1700 			BNX2X_ERR("Failed to clean multi-cast object: %d\n",
1701 				  rc);
1702 			return;
1703 		}
1704 
1705 		rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_CONT);
1706 	}
1707 }
1708 
1709 #ifndef BNX2X_STOP_ON_ERROR
1710 #define LOAD_ERROR_EXIT(bp, label) \
1711 	do { \
1712 		(bp)->state = BNX2X_STATE_ERROR; \
1713 		goto label; \
1714 	} while (0)
1715 #else
1716 #define LOAD_ERROR_EXIT(bp, label) \
1717 	do { \
1718 		(bp)->state = BNX2X_STATE_ERROR; \
1719 		(bp)->panic = 1; \
1720 		return -EBUSY; \
1721 	} while (0)
1722 #endif
1723 
1724 bool bnx2x_test_firmware_version(struct bnx2x *bp, bool is_err)
1725 {
1726 	/* build FW version dword */
1727 	u32 my_fw = (BCM_5710_FW_MAJOR_VERSION) +
1728 		    (BCM_5710_FW_MINOR_VERSION << 8) +
1729 		    (BCM_5710_FW_REVISION_VERSION << 16) +
1730 		    (BCM_5710_FW_ENGINEERING_VERSION << 24);
1731 
1732 	/* read loaded FW from chip */
1733 	u32 loaded_fw = REG_RD(bp, XSEM_REG_PRAM);
1734 
1735 	DP(NETIF_MSG_IFUP, "loaded fw %x, my fw %x\n", loaded_fw, my_fw);
1736 
1737 	if (loaded_fw != my_fw) {
1738 		if (is_err)
1739 			BNX2X_ERR("bnx2x with FW %x was already loaded, which mismatches my %x FW. aborting\n",
1740 				  loaded_fw, my_fw);
1741 		return false;
1742 	}
1743 
1744 	return true;
1745 }
1746 
1747 /* must be called with rtnl_lock */
1748 int bnx2x_nic_load(struct bnx2x *bp, int load_mode)
1749 {
1750 	int port = BP_PORT(bp);
1751 	u32 load_code;
1752 	int i, rc;
1753 
1754 #ifdef BNX2X_STOP_ON_ERROR
1755 	if (unlikely(bp->panic)) {
1756 		BNX2X_ERR("Can't load NIC when there is panic\n");
1757 		return -EPERM;
1758 	}
1759 #endif
1760 
1761 	bp->state = BNX2X_STATE_OPENING_WAIT4_LOAD;
1762 
1763 	/* Set the initial link reported state to link down */
1764 	bnx2x_acquire_phy_lock(bp);
1765 	memset(&bp->last_reported_link, 0, sizeof(bp->last_reported_link));
1766 	__set_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1767 		&bp->last_reported_link.link_report_flags);
1768 	bnx2x_release_phy_lock(bp);
1769 
1770 	/* must be called before memory allocation and HW init */
1771 	bnx2x_ilt_set_info(bp);
1772 
1773 	/*
1774 	 * Zero fastpath structures preserving invariants like napi, which are
1775 	 * allocated only once, fp index, max_cos, bp pointer.
1776 	 * Also set fp->disable_tpa.
1777 	 */
1778 	DP(NETIF_MSG_IFUP, "num queues: %d", bp->num_queues);
1779 	for_each_queue(bp, i)
1780 		bnx2x_bz_fp(bp, i);
1781 
1782 
1783 	/* Set the receive queues buffer size */
1784 	bnx2x_set_rx_buf_size(bp);
1785 
1786 	if (bnx2x_alloc_mem(bp))
1787 		return -ENOMEM;
1788 
1789 	/* As long as bnx2x_alloc_mem() may possibly update
1790 	 * bp->num_queues, bnx2x_set_real_num_queues() should always
1791 	 * come after it.
1792 	 */
1793 	rc = bnx2x_set_real_num_queues(bp);
1794 	if (rc) {
1795 		BNX2X_ERR("Unable to set real_num_queues\n");
1796 		LOAD_ERROR_EXIT(bp, load_error0);
1797 	}
1798 
1799 	/* configure multi cos mappings in kernel.
1800 	 * this configuration may be overriden by a multi class queue discipline
1801 	 * or by a dcbx negotiation result.
1802 	 */
1803 	bnx2x_setup_tc(bp->dev, bp->max_cos);
1804 
1805 	bnx2x_napi_enable(bp);
1806 
1807 	/* set pf load just before approaching the MCP */
1808 	bnx2x_set_pf_load(bp);
1809 
1810 	/* Send LOAD_REQUEST command to MCP
1811 	 * Returns the type of LOAD command:
1812 	 * if it is the first port to be initialized
1813 	 * common blocks should be initialized, otherwise - not
1814 	 */
1815 	if (!BP_NOMCP(bp)) {
1816 		/* init fw_seq */
1817 		bp->fw_seq =
1818 			(SHMEM_RD(bp, func_mb[BP_FW_MB_IDX(bp)].drv_mb_header) &
1819 			 DRV_MSG_SEQ_NUMBER_MASK);
1820 		BNX2X_DEV_INFO("fw_seq 0x%08x\n", bp->fw_seq);
1821 
1822 		/* Get current FW pulse sequence */
1823 		bp->fw_drv_pulse_wr_seq =
1824 			(SHMEM_RD(bp, func_mb[BP_FW_MB_IDX(bp)].drv_pulse_mb) &
1825 			 DRV_PULSE_SEQ_MASK);
1826 		BNX2X_DEV_INFO("drv_pulse 0x%x\n", bp->fw_drv_pulse_wr_seq);
1827 
1828 		load_code = bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_REQ, 0);
1829 		if (!load_code) {
1830 			BNX2X_ERR("MCP response failure, aborting\n");
1831 			rc = -EBUSY;
1832 			LOAD_ERROR_EXIT(bp, load_error1);
1833 		}
1834 		if (load_code == FW_MSG_CODE_DRV_LOAD_REFUSED) {
1835 			BNX2X_ERR("Driver load refused\n");
1836 			rc = -EBUSY; /* other port in diagnostic mode */
1837 			LOAD_ERROR_EXIT(bp, load_error1);
1838 		}
1839 		if (load_code != FW_MSG_CODE_DRV_LOAD_COMMON_CHIP &&
1840 		    load_code != FW_MSG_CODE_DRV_LOAD_COMMON) {
1841 			/* abort nic load if version mismatch */
1842 			if (!bnx2x_test_firmware_version(bp, true)) {
1843 				rc = -EBUSY;
1844 				LOAD_ERROR_EXIT(bp, load_error2);
1845 			}
1846 		}
1847 
1848 	} else {
1849 		int path = BP_PATH(bp);
1850 
1851 		DP(NETIF_MSG_IFUP, "NO MCP - load counts[%d]      %d, %d, %d\n",
1852 		   path, load_count[path][0], load_count[path][1],
1853 		   load_count[path][2]);
1854 		load_count[path][0]++;
1855 		load_count[path][1 + port]++;
1856 		DP(NETIF_MSG_IFUP, "NO MCP - new load counts[%d]  %d, %d, %d\n",
1857 		   path, load_count[path][0], load_count[path][1],
1858 		   load_count[path][2]);
1859 		if (load_count[path][0] == 1)
1860 			load_code = FW_MSG_CODE_DRV_LOAD_COMMON;
1861 		else if (load_count[path][1 + port] == 1)
1862 			load_code = FW_MSG_CODE_DRV_LOAD_PORT;
1863 		else
1864 			load_code = FW_MSG_CODE_DRV_LOAD_FUNCTION;
1865 	}
1866 
1867 	if ((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
1868 	    (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP) ||
1869 	    (load_code == FW_MSG_CODE_DRV_LOAD_PORT)) {
1870 		bp->port.pmf = 1;
1871 		/*
1872 		 * We need the barrier to ensure the ordering between the
1873 		 * writing to bp->port.pmf here and reading it from the
1874 		 * bnx2x_periodic_task().
1875 		 */
1876 		smp_mb();
1877 	} else
1878 		bp->port.pmf = 0;
1879 
1880 	DP(NETIF_MSG_IFUP, "pmf %d\n", bp->port.pmf);
1881 
1882 	/* Init Function state controlling object */
1883 	bnx2x__init_func_obj(bp);
1884 
1885 	/* Initialize HW */
1886 	rc = bnx2x_init_hw(bp, load_code);
1887 	if (rc) {
1888 		BNX2X_ERR("HW init failed, aborting\n");
1889 		bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
1890 		LOAD_ERROR_EXIT(bp, load_error2);
1891 	}
1892 
1893 	/* Connect to IRQs */
1894 	rc = bnx2x_setup_irqs(bp);
1895 	if (rc) {
1896 		BNX2X_ERR("IRQs setup failed\n");
1897 		bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
1898 		LOAD_ERROR_EXIT(bp, load_error2);
1899 	}
1900 
1901 	/* Setup NIC internals and enable interrupts */
1902 	bnx2x_nic_init(bp, load_code);
1903 
1904 	/* Init per-function objects */
1905 	bnx2x_init_bp_objs(bp);
1906 
1907 	if (((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
1908 	    (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP)) &&
1909 	    (bp->common.shmem2_base)) {
1910 		if (SHMEM2_HAS(bp, dcc_support))
1911 			SHMEM2_WR(bp, dcc_support,
1912 				  (SHMEM_DCC_SUPPORT_DISABLE_ENABLE_PF_TLV |
1913 				   SHMEM_DCC_SUPPORT_BANDWIDTH_ALLOCATION_TLV));
1914 	}
1915 
1916 	bp->state = BNX2X_STATE_OPENING_WAIT4_PORT;
1917 	rc = bnx2x_func_start(bp);
1918 	if (rc) {
1919 		BNX2X_ERR("Function start failed!\n");
1920 		bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
1921 		LOAD_ERROR_EXIT(bp, load_error3);
1922 	}
1923 
1924 	/* Send LOAD_DONE command to MCP */
1925 	if (!BP_NOMCP(bp)) {
1926 		load_code = bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
1927 		if (!load_code) {
1928 			BNX2X_ERR("MCP response failure, aborting\n");
1929 			rc = -EBUSY;
1930 			LOAD_ERROR_EXIT(bp, load_error3);
1931 		}
1932 	}
1933 
1934 	rc = bnx2x_setup_leading(bp);
1935 	if (rc) {
1936 		BNX2X_ERR("Setup leading failed!\n");
1937 		LOAD_ERROR_EXIT(bp, load_error3);
1938 	}
1939 
1940 #ifdef BCM_CNIC
1941 	/* Enable Timer scan */
1942 	REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + port*4, 1);
1943 #endif
1944 
1945 	for_each_nondefault_queue(bp, i) {
1946 		rc = bnx2x_setup_queue(bp, &bp->fp[i], 0);
1947 		if (rc) {
1948 			BNX2X_ERR("Queue setup failed\n");
1949 			LOAD_ERROR_EXIT(bp, load_error4);
1950 		}
1951 	}
1952 
1953 	rc = bnx2x_init_rss_pf(bp);
1954 	if (rc) {
1955 		BNX2X_ERR("PF RSS init failed\n");
1956 		LOAD_ERROR_EXIT(bp, load_error4);
1957 	}
1958 
1959 	/* Now when Clients are configured we are ready to work */
1960 	bp->state = BNX2X_STATE_OPEN;
1961 
1962 	/* Configure a ucast MAC */
1963 	rc = bnx2x_set_eth_mac(bp, true);
1964 	if (rc) {
1965 		BNX2X_ERR("Setting Ethernet MAC failed\n");
1966 		LOAD_ERROR_EXIT(bp, load_error4);
1967 	}
1968 
1969 	if (bp->pending_max) {
1970 		bnx2x_update_max_mf_config(bp, bp->pending_max);
1971 		bp->pending_max = 0;
1972 	}
1973 
1974 	if (bp->port.pmf)
1975 		bnx2x_initial_phy_init(bp, load_mode);
1976 
1977 	/* Start fast path */
1978 
1979 	/* Initialize Rx filter. */
1980 	netif_addr_lock_bh(bp->dev);
1981 	bnx2x_set_rx_mode(bp->dev);
1982 	netif_addr_unlock_bh(bp->dev);
1983 
1984 	/* Start the Tx */
1985 	switch (load_mode) {
1986 	case LOAD_NORMAL:
1987 		/* Tx queue should be only reenabled */
1988 		netif_tx_wake_all_queues(bp->dev);
1989 		break;
1990 
1991 	case LOAD_OPEN:
1992 		netif_tx_start_all_queues(bp->dev);
1993 		smp_mb__after_clear_bit();
1994 		break;
1995 
1996 	case LOAD_DIAG:
1997 		bp->state = BNX2X_STATE_DIAG;
1998 		break;
1999 
2000 	default:
2001 		break;
2002 	}
2003 
2004 	if (bp->port.pmf)
2005 		bnx2x_update_drv_flags(bp, 1 << DRV_FLAGS_DCB_CONFIGURED, 0);
2006 	else
2007 		bnx2x__link_status_update(bp);
2008 
2009 	/* start the timer */
2010 	mod_timer(&bp->timer, jiffies + bp->current_interval);
2011 
2012 #ifdef BCM_CNIC
2013 	/* re-read iscsi info */
2014 	bnx2x_get_iscsi_info(bp);
2015 	bnx2x_setup_cnic_irq_info(bp);
2016 	if (bp->state == BNX2X_STATE_OPEN)
2017 		bnx2x_cnic_notify(bp, CNIC_CTL_START_CMD);
2018 #endif
2019 
2020 	/* mark driver is loaded in shmem2 */
2021 	if (SHMEM2_HAS(bp, drv_capabilities_flag)) {
2022 		u32 val;
2023 		val = SHMEM2_RD(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)]);
2024 		SHMEM2_WR(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)],
2025 			  val | DRV_FLAGS_CAPABILITIES_LOADED_SUPPORTED |
2026 			  DRV_FLAGS_CAPABILITIES_LOADED_L2);
2027 	}
2028 
2029 	/* Wait for all pending SP commands to complete */
2030 	if (!bnx2x_wait_sp_comp(bp, ~0x0UL)) {
2031 		BNX2X_ERR("Timeout waiting for SP elements to complete\n");
2032 		bnx2x_nic_unload(bp, UNLOAD_CLOSE);
2033 		return -EBUSY;
2034 	}
2035 
2036 	bnx2x_dcbx_init(bp);
2037 	return 0;
2038 
2039 #ifndef BNX2X_STOP_ON_ERROR
2040 load_error4:
2041 #ifdef BCM_CNIC
2042 	/* Disable Timer scan */
2043 	REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + port*4, 0);
2044 #endif
2045 load_error3:
2046 	bnx2x_int_disable_sync(bp, 1);
2047 
2048 	/* Clean queueable objects */
2049 	bnx2x_squeeze_objects(bp);
2050 
2051 	/* Free SKBs, SGEs, TPA pool and driver internals */
2052 	bnx2x_free_skbs(bp);
2053 	for_each_rx_queue(bp, i)
2054 		bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
2055 
2056 	/* Release IRQs */
2057 	bnx2x_free_irq(bp);
2058 load_error2:
2059 	if (!BP_NOMCP(bp)) {
2060 		bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_REQ_WOL_MCP, 0);
2061 		bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_DONE, 0);
2062 	}
2063 
2064 	bp->port.pmf = 0;
2065 load_error1:
2066 	bnx2x_napi_disable(bp);
2067 	/* clear pf_load status, as it was already set */
2068 	bnx2x_clear_pf_load(bp);
2069 load_error0:
2070 	bnx2x_free_mem(bp);
2071 
2072 	return rc;
2073 #endif /* ! BNX2X_STOP_ON_ERROR */
2074 }
2075 
2076 /* must be called with rtnl_lock */
2077 int bnx2x_nic_unload(struct bnx2x *bp, int unload_mode)
2078 {
2079 	int i;
2080 	bool global = false;
2081 
2082 	/* mark driver is unloaded in shmem2 */
2083 	if (SHMEM2_HAS(bp, drv_capabilities_flag)) {
2084 		u32 val;
2085 		val = SHMEM2_RD(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)]);
2086 		SHMEM2_WR(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)],
2087 			  val & ~DRV_FLAGS_CAPABILITIES_LOADED_L2);
2088 	}
2089 
2090 	if ((bp->state == BNX2X_STATE_CLOSED) ||
2091 	    (bp->state == BNX2X_STATE_ERROR)) {
2092 		/* We can get here if the driver has been unloaded
2093 		 * during parity error recovery and is either waiting for a
2094 		 * leader to complete or for other functions to unload and
2095 		 * then ifdown has been issued. In this case we want to
2096 		 * unload and let other functions to complete a recovery
2097 		 * process.
2098 		 */
2099 		bp->recovery_state = BNX2X_RECOVERY_DONE;
2100 		bp->is_leader = 0;
2101 		bnx2x_release_leader_lock(bp);
2102 		smp_mb();
2103 
2104 		DP(NETIF_MSG_IFDOWN, "Releasing a leadership...\n");
2105 		BNX2X_ERR("Can't unload in closed or error state\n");
2106 		return -EINVAL;
2107 	}
2108 
2109 	/*
2110 	 * It's important to set the bp->state to the value different from
2111 	 * BNX2X_STATE_OPEN and only then stop the Tx. Otherwise bnx2x_tx_int()
2112 	 * may restart the Tx from the NAPI context (see bnx2x_tx_int()).
2113 	 */
2114 	bp->state = BNX2X_STATE_CLOSING_WAIT4_HALT;
2115 	smp_mb();
2116 
2117 	/* Stop Tx */
2118 	bnx2x_tx_disable(bp);
2119 
2120 #ifdef BCM_CNIC
2121 	bnx2x_cnic_notify(bp, CNIC_CTL_STOP_CMD);
2122 #endif
2123 
2124 	bp->rx_mode = BNX2X_RX_MODE_NONE;
2125 
2126 	del_timer_sync(&bp->timer);
2127 
2128 	/* Set ALWAYS_ALIVE bit in shmem */
2129 	bp->fw_drv_pulse_wr_seq |= DRV_PULSE_ALWAYS_ALIVE;
2130 
2131 	bnx2x_drv_pulse(bp);
2132 
2133 	bnx2x_stats_handle(bp, STATS_EVENT_STOP);
2134 	bnx2x_save_statistics(bp);
2135 
2136 	/* Cleanup the chip if needed */
2137 	if (unload_mode != UNLOAD_RECOVERY)
2138 		bnx2x_chip_cleanup(bp, unload_mode);
2139 	else {
2140 		/* Send the UNLOAD_REQUEST to the MCP */
2141 		bnx2x_send_unload_req(bp, unload_mode);
2142 
2143 		/*
2144 		 * Prevent transactions to host from the functions on the
2145 		 * engine that doesn't reset global blocks in case of global
2146 		 * attention once gloabl blocks are reset and gates are opened
2147 		 * (the engine which leader will perform the recovery
2148 		 * last).
2149 		 */
2150 		if (!CHIP_IS_E1x(bp))
2151 			bnx2x_pf_disable(bp);
2152 
2153 		/* Disable HW interrupts, NAPI */
2154 		bnx2x_netif_stop(bp, 1);
2155 
2156 		/* Release IRQs */
2157 		bnx2x_free_irq(bp);
2158 
2159 		/* Report UNLOAD_DONE to MCP */
2160 		bnx2x_send_unload_done(bp);
2161 	}
2162 
2163 	/*
2164 	 * At this stage no more interrupts will arrive so we may safly clean
2165 	 * the queueable objects here in case they failed to get cleaned so far.
2166 	 */
2167 	bnx2x_squeeze_objects(bp);
2168 
2169 	/* There should be no more pending SP commands at this stage */
2170 	bp->sp_state = 0;
2171 
2172 	bp->port.pmf = 0;
2173 
2174 	/* Free SKBs, SGEs, TPA pool and driver internals */
2175 	bnx2x_free_skbs(bp);
2176 	for_each_rx_queue(bp, i)
2177 		bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
2178 
2179 	bnx2x_free_mem(bp);
2180 
2181 	bp->state = BNX2X_STATE_CLOSED;
2182 
2183 	/* Check if there are pending parity attentions. If there are - set
2184 	 * RECOVERY_IN_PROGRESS.
2185 	 */
2186 	if (bnx2x_chk_parity_attn(bp, &global, false)) {
2187 		bnx2x_set_reset_in_progress(bp);
2188 
2189 		/* Set RESET_IS_GLOBAL if needed */
2190 		if (global)
2191 			bnx2x_set_reset_global(bp);
2192 	}
2193 
2194 
2195 	/* The last driver must disable a "close the gate" if there is no
2196 	 * parity attention or "process kill" pending.
2197 	 */
2198 	if (!bnx2x_clear_pf_load(bp) && bnx2x_reset_is_done(bp, BP_PATH(bp)))
2199 		bnx2x_disable_close_the_gate(bp);
2200 
2201 	return 0;
2202 }
2203 
2204 int bnx2x_set_power_state(struct bnx2x *bp, pci_power_t state)
2205 {
2206 	u16 pmcsr;
2207 
2208 	/* If there is no power capability, silently succeed */
2209 	if (!bp->pm_cap) {
2210 		BNX2X_DEV_INFO("No power capability. Breaking.\n");
2211 		return 0;
2212 	}
2213 
2214 	pci_read_config_word(bp->pdev, bp->pm_cap + PCI_PM_CTRL, &pmcsr);
2215 
2216 	switch (state) {
2217 	case PCI_D0:
2218 		pci_write_config_word(bp->pdev, bp->pm_cap + PCI_PM_CTRL,
2219 				      ((pmcsr & ~PCI_PM_CTRL_STATE_MASK) |
2220 				       PCI_PM_CTRL_PME_STATUS));
2221 
2222 		if (pmcsr & PCI_PM_CTRL_STATE_MASK)
2223 			/* delay required during transition out of D3hot */
2224 			msleep(20);
2225 		break;
2226 
2227 	case PCI_D3hot:
2228 		/* If there are other clients above don't
2229 		   shut down the power */
2230 		if (atomic_read(&bp->pdev->enable_cnt) != 1)
2231 			return 0;
2232 		/* Don't shut down the power for emulation and FPGA */
2233 		if (CHIP_REV_IS_SLOW(bp))
2234 			return 0;
2235 
2236 		pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
2237 		pmcsr |= 3;
2238 
2239 		if (bp->wol)
2240 			pmcsr |= PCI_PM_CTRL_PME_ENABLE;
2241 
2242 		pci_write_config_word(bp->pdev, bp->pm_cap + PCI_PM_CTRL,
2243 				      pmcsr);
2244 
2245 		/* No more memory access after this point until
2246 		* device is brought back to D0.
2247 		*/
2248 		break;
2249 
2250 	default:
2251 		dev_err(&bp->pdev->dev, "Can't support state = %d\n", state);
2252 		return -EINVAL;
2253 	}
2254 	return 0;
2255 }
2256 
2257 /*
2258  * net_device service functions
2259  */
2260 int bnx2x_poll(struct napi_struct *napi, int budget)
2261 {
2262 	int work_done = 0;
2263 	u8 cos;
2264 	struct bnx2x_fastpath *fp = container_of(napi, struct bnx2x_fastpath,
2265 						 napi);
2266 	struct bnx2x *bp = fp->bp;
2267 
2268 	while (1) {
2269 #ifdef BNX2X_STOP_ON_ERROR
2270 		if (unlikely(bp->panic)) {
2271 			napi_complete(napi);
2272 			return 0;
2273 		}
2274 #endif
2275 
2276 		for_each_cos_in_tx_queue(fp, cos)
2277 			if (bnx2x_tx_queue_has_work(&fp->txdata[cos]))
2278 				bnx2x_tx_int(bp, &fp->txdata[cos]);
2279 
2280 
2281 		if (bnx2x_has_rx_work(fp)) {
2282 			work_done += bnx2x_rx_int(fp, budget - work_done);
2283 
2284 			/* must not complete if we consumed full budget */
2285 			if (work_done >= budget)
2286 				break;
2287 		}
2288 
2289 		/* Fall out from the NAPI loop if needed */
2290 		if (!(bnx2x_has_rx_work(fp) || bnx2x_has_tx_work(fp))) {
2291 #ifdef BCM_CNIC
2292 			/* No need to update SB for FCoE L2 ring as long as
2293 			 * it's connected to the default SB and the SB
2294 			 * has been updated when NAPI was scheduled.
2295 			 */
2296 			if (IS_FCOE_FP(fp)) {
2297 				napi_complete(napi);
2298 				break;
2299 			}
2300 #endif
2301 
2302 			bnx2x_update_fpsb_idx(fp);
2303 			/* bnx2x_has_rx_work() reads the status block,
2304 			 * thus we need to ensure that status block indices
2305 			 * have been actually read (bnx2x_update_fpsb_idx)
2306 			 * prior to this check (bnx2x_has_rx_work) so that
2307 			 * we won't write the "newer" value of the status block
2308 			 * to IGU (if there was a DMA right after
2309 			 * bnx2x_has_rx_work and if there is no rmb, the memory
2310 			 * reading (bnx2x_update_fpsb_idx) may be postponed
2311 			 * to right before bnx2x_ack_sb). In this case there
2312 			 * will never be another interrupt until there is
2313 			 * another update of the status block, while there
2314 			 * is still unhandled work.
2315 			 */
2316 			rmb();
2317 
2318 			if (!(bnx2x_has_rx_work(fp) || bnx2x_has_tx_work(fp))) {
2319 				napi_complete(napi);
2320 				/* Re-enable interrupts */
2321 				DP(NETIF_MSG_RX_STATUS,
2322 				   "Update index to %d\n", fp->fp_hc_idx);
2323 				bnx2x_ack_sb(bp, fp->igu_sb_id, USTORM_ID,
2324 					     le16_to_cpu(fp->fp_hc_idx),
2325 					     IGU_INT_ENABLE, 1);
2326 				break;
2327 			}
2328 		}
2329 	}
2330 
2331 	return work_done;
2332 }
2333 
2334 /* we split the first BD into headers and data BDs
2335  * to ease the pain of our fellow microcode engineers
2336  * we use one mapping for both BDs
2337  * So far this has only been observed to happen
2338  * in Other Operating Systems(TM)
2339  */
2340 static noinline u16 bnx2x_tx_split(struct bnx2x *bp,
2341 				   struct bnx2x_fp_txdata *txdata,
2342 				   struct sw_tx_bd *tx_buf,
2343 				   struct eth_tx_start_bd **tx_bd, u16 hlen,
2344 				   u16 bd_prod, int nbd)
2345 {
2346 	struct eth_tx_start_bd *h_tx_bd = *tx_bd;
2347 	struct eth_tx_bd *d_tx_bd;
2348 	dma_addr_t mapping;
2349 	int old_len = le16_to_cpu(h_tx_bd->nbytes);
2350 
2351 	/* first fix first BD */
2352 	h_tx_bd->nbd = cpu_to_le16(nbd);
2353 	h_tx_bd->nbytes = cpu_to_le16(hlen);
2354 
2355 	DP(NETIF_MSG_TX_QUEUED,	"TSO split header size is %d (%x:%x) nbd %d\n",
2356 	   h_tx_bd->nbytes, h_tx_bd->addr_hi, h_tx_bd->addr_lo, h_tx_bd->nbd);
2357 
2358 	/* now get a new data BD
2359 	 * (after the pbd) and fill it */
2360 	bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
2361 	d_tx_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
2362 
2363 	mapping = HILO_U64(le32_to_cpu(h_tx_bd->addr_hi),
2364 			   le32_to_cpu(h_tx_bd->addr_lo)) + hlen;
2365 
2366 	d_tx_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
2367 	d_tx_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
2368 	d_tx_bd->nbytes = cpu_to_le16(old_len - hlen);
2369 
2370 	/* this marks the BD as one that has no individual mapping */
2371 	tx_buf->flags |= BNX2X_TSO_SPLIT_BD;
2372 
2373 	DP(NETIF_MSG_TX_QUEUED,
2374 	   "TSO split data size is %d (%x:%x)\n",
2375 	   d_tx_bd->nbytes, d_tx_bd->addr_hi, d_tx_bd->addr_lo);
2376 
2377 	/* update tx_bd */
2378 	*tx_bd = (struct eth_tx_start_bd *)d_tx_bd;
2379 
2380 	return bd_prod;
2381 }
2382 
2383 static inline u16 bnx2x_csum_fix(unsigned char *t_header, u16 csum, s8 fix)
2384 {
2385 	if (fix > 0)
2386 		csum = (u16) ~csum_fold(csum_sub(csum,
2387 				csum_partial(t_header - fix, fix, 0)));
2388 
2389 	else if (fix < 0)
2390 		csum = (u16) ~csum_fold(csum_add(csum,
2391 				csum_partial(t_header, -fix, 0)));
2392 
2393 	return swab16(csum);
2394 }
2395 
2396 static inline u32 bnx2x_xmit_type(struct bnx2x *bp, struct sk_buff *skb)
2397 {
2398 	u32 rc;
2399 
2400 	if (skb->ip_summed != CHECKSUM_PARTIAL)
2401 		rc = XMIT_PLAIN;
2402 
2403 	else {
2404 		if (vlan_get_protocol(skb) == htons(ETH_P_IPV6)) {
2405 			rc = XMIT_CSUM_V6;
2406 			if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2407 				rc |= XMIT_CSUM_TCP;
2408 
2409 		} else {
2410 			rc = XMIT_CSUM_V4;
2411 			if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2412 				rc |= XMIT_CSUM_TCP;
2413 		}
2414 	}
2415 
2416 	if (skb_is_gso_v6(skb))
2417 		rc |= XMIT_GSO_V6 | XMIT_CSUM_TCP | XMIT_CSUM_V6;
2418 	else if (skb_is_gso(skb))
2419 		rc |= XMIT_GSO_V4 | XMIT_CSUM_V4 | XMIT_CSUM_TCP;
2420 
2421 	return rc;
2422 }
2423 
2424 #if (MAX_SKB_FRAGS >= MAX_FETCH_BD - 3)
2425 /* check if packet requires linearization (packet is too fragmented)
2426    no need to check fragmentation if page size > 8K (there will be no
2427    violation to FW restrictions) */
2428 static int bnx2x_pkt_req_lin(struct bnx2x *bp, struct sk_buff *skb,
2429 			     u32 xmit_type)
2430 {
2431 	int to_copy = 0;
2432 	int hlen = 0;
2433 	int first_bd_sz = 0;
2434 
2435 	/* 3 = 1 (for linear data BD) + 2 (for PBD and last BD) */
2436 	if (skb_shinfo(skb)->nr_frags >= (MAX_FETCH_BD - 3)) {
2437 
2438 		if (xmit_type & XMIT_GSO) {
2439 			unsigned short lso_mss = skb_shinfo(skb)->gso_size;
2440 			/* Check if LSO packet needs to be copied:
2441 			   3 = 1 (for headers BD) + 2 (for PBD and last BD) */
2442 			int wnd_size = MAX_FETCH_BD - 3;
2443 			/* Number of windows to check */
2444 			int num_wnds = skb_shinfo(skb)->nr_frags - wnd_size;
2445 			int wnd_idx = 0;
2446 			int frag_idx = 0;
2447 			u32 wnd_sum = 0;
2448 
2449 			/* Headers length */
2450 			hlen = (int)(skb_transport_header(skb) - skb->data) +
2451 				tcp_hdrlen(skb);
2452 
2453 			/* Amount of data (w/o headers) on linear part of SKB*/
2454 			first_bd_sz = skb_headlen(skb) - hlen;
2455 
2456 			wnd_sum  = first_bd_sz;
2457 
2458 			/* Calculate the first sum - it's special */
2459 			for (frag_idx = 0; frag_idx < wnd_size - 1; frag_idx++)
2460 				wnd_sum +=
2461 					skb_frag_size(&skb_shinfo(skb)->frags[frag_idx]);
2462 
2463 			/* If there was data on linear skb data - check it */
2464 			if (first_bd_sz > 0) {
2465 				if (unlikely(wnd_sum < lso_mss)) {
2466 					to_copy = 1;
2467 					goto exit_lbl;
2468 				}
2469 
2470 				wnd_sum -= first_bd_sz;
2471 			}
2472 
2473 			/* Others are easier: run through the frag list and
2474 			   check all windows */
2475 			for (wnd_idx = 0; wnd_idx <= num_wnds; wnd_idx++) {
2476 				wnd_sum +=
2477 			  skb_frag_size(&skb_shinfo(skb)->frags[wnd_idx + wnd_size - 1]);
2478 
2479 				if (unlikely(wnd_sum < lso_mss)) {
2480 					to_copy = 1;
2481 					break;
2482 				}
2483 				wnd_sum -=
2484 					skb_frag_size(&skb_shinfo(skb)->frags[wnd_idx]);
2485 			}
2486 		} else {
2487 			/* in non-LSO too fragmented packet should always
2488 			   be linearized */
2489 			to_copy = 1;
2490 		}
2491 	}
2492 
2493 exit_lbl:
2494 	if (unlikely(to_copy))
2495 		DP(NETIF_MSG_TX_QUEUED,
2496 		   "Linearization IS REQUIRED for %s packet. num_frags %d  hlen %d  first_bd_sz %d\n",
2497 		   (xmit_type & XMIT_GSO) ? "LSO" : "non-LSO",
2498 		   skb_shinfo(skb)->nr_frags, hlen, first_bd_sz);
2499 
2500 	return to_copy;
2501 }
2502 #endif
2503 
2504 static inline void bnx2x_set_pbd_gso_e2(struct sk_buff *skb, u32 *parsing_data,
2505 					u32 xmit_type)
2506 {
2507 	*parsing_data |= (skb_shinfo(skb)->gso_size <<
2508 			      ETH_TX_PARSE_BD_E2_LSO_MSS_SHIFT) &
2509 			      ETH_TX_PARSE_BD_E2_LSO_MSS;
2510 	if ((xmit_type & XMIT_GSO_V6) &&
2511 	    (ipv6_hdr(skb)->nexthdr == NEXTHDR_IPV6))
2512 		*parsing_data |= ETH_TX_PARSE_BD_E2_IPV6_WITH_EXT_HDR;
2513 }
2514 
2515 /**
2516  * bnx2x_set_pbd_gso - update PBD in GSO case.
2517  *
2518  * @skb:	packet skb
2519  * @pbd:	parse BD
2520  * @xmit_type:	xmit flags
2521  */
2522 static inline void bnx2x_set_pbd_gso(struct sk_buff *skb,
2523 				     struct eth_tx_parse_bd_e1x *pbd,
2524 				     u32 xmit_type)
2525 {
2526 	pbd->lso_mss = cpu_to_le16(skb_shinfo(skb)->gso_size);
2527 	pbd->tcp_send_seq = swab32(tcp_hdr(skb)->seq);
2528 	pbd->tcp_flags = pbd_tcp_flags(skb);
2529 
2530 	if (xmit_type & XMIT_GSO_V4) {
2531 		pbd->ip_id = swab16(ip_hdr(skb)->id);
2532 		pbd->tcp_pseudo_csum =
2533 			swab16(~csum_tcpudp_magic(ip_hdr(skb)->saddr,
2534 						  ip_hdr(skb)->daddr,
2535 						  0, IPPROTO_TCP, 0));
2536 
2537 	} else
2538 		pbd->tcp_pseudo_csum =
2539 			swab16(~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2540 						&ipv6_hdr(skb)->daddr,
2541 						0, IPPROTO_TCP, 0));
2542 
2543 	pbd->global_data |= ETH_TX_PARSE_BD_E1X_PSEUDO_CS_WITHOUT_LEN;
2544 }
2545 
2546 /**
2547  * bnx2x_set_pbd_csum_e2 - update PBD with checksum and return header length
2548  *
2549  * @bp:			driver handle
2550  * @skb:		packet skb
2551  * @parsing_data:	data to be updated
2552  * @xmit_type:		xmit flags
2553  *
2554  * 57712 related
2555  */
2556 static inline  u8 bnx2x_set_pbd_csum_e2(struct bnx2x *bp, struct sk_buff *skb,
2557 	u32 *parsing_data, u32 xmit_type)
2558 {
2559 	*parsing_data |=
2560 			((((u8 *)skb_transport_header(skb) - skb->data) >> 1) <<
2561 			ETH_TX_PARSE_BD_E2_TCP_HDR_START_OFFSET_W_SHIFT) &
2562 			ETH_TX_PARSE_BD_E2_TCP_HDR_START_OFFSET_W;
2563 
2564 	if (xmit_type & XMIT_CSUM_TCP) {
2565 		*parsing_data |= ((tcp_hdrlen(skb) / 4) <<
2566 			ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW_SHIFT) &
2567 			ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW;
2568 
2569 		return skb_transport_header(skb) + tcp_hdrlen(skb) - skb->data;
2570 	} else
2571 		/* We support checksum offload for TCP and UDP only.
2572 		 * No need to pass the UDP header length - it's a constant.
2573 		 */
2574 		return skb_transport_header(skb) +
2575 				sizeof(struct udphdr) - skb->data;
2576 }
2577 
2578 static inline void bnx2x_set_sbd_csum(struct bnx2x *bp, struct sk_buff *skb,
2579 	struct eth_tx_start_bd *tx_start_bd, u32 xmit_type)
2580 {
2581 	tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_L4_CSUM;
2582 
2583 	if (xmit_type & XMIT_CSUM_V4)
2584 		tx_start_bd->bd_flags.as_bitfield |=
2585 					ETH_TX_BD_FLAGS_IP_CSUM;
2586 	else
2587 		tx_start_bd->bd_flags.as_bitfield |=
2588 					ETH_TX_BD_FLAGS_IPV6;
2589 
2590 	if (!(xmit_type & XMIT_CSUM_TCP))
2591 		tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_IS_UDP;
2592 }
2593 
2594 /**
2595  * bnx2x_set_pbd_csum - update PBD with checksum and return header length
2596  *
2597  * @bp:		driver handle
2598  * @skb:	packet skb
2599  * @pbd:	parse BD to be updated
2600  * @xmit_type:	xmit flags
2601  */
2602 static inline u8 bnx2x_set_pbd_csum(struct bnx2x *bp, struct sk_buff *skb,
2603 	struct eth_tx_parse_bd_e1x *pbd,
2604 	u32 xmit_type)
2605 {
2606 	u8 hlen = (skb_network_header(skb) - skb->data) >> 1;
2607 
2608 	/* for now NS flag is not used in Linux */
2609 	pbd->global_data =
2610 		(hlen | ((skb->protocol == cpu_to_be16(ETH_P_8021Q)) <<
2611 			 ETH_TX_PARSE_BD_E1X_LLC_SNAP_EN_SHIFT));
2612 
2613 	pbd->ip_hlen_w = (skb_transport_header(skb) -
2614 			skb_network_header(skb)) >> 1;
2615 
2616 	hlen += pbd->ip_hlen_w;
2617 
2618 	/* We support checksum offload for TCP and UDP only */
2619 	if (xmit_type & XMIT_CSUM_TCP)
2620 		hlen += tcp_hdrlen(skb) / 2;
2621 	else
2622 		hlen += sizeof(struct udphdr) / 2;
2623 
2624 	pbd->total_hlen_w = cpu_to_le16(hlen);
2625 	hlen = hlen*2;
2626 
2627 	if (xmit_type & XMIT_CSUM_TCP) {
2628 		pbd->tcp_pseudo_csum = swab16(tcp_hdr(skb)->check);
2629 
2630 	} else {
2631 		s8 fix = SKB_CS_OFF(skb); /* signed! */
2632 
2633 		DP(NETIF_MSG_TX_QUEUED,
2634 		   "hlen %d  fix %d  csum before fix %x\n",
2635 		   le16_to_cpu(pbd->total_hlen_w), fix, SKB_CS(skb));
2636 
2637 		/* HW bug: fixup the CSUM */
2638 		pbd->tcp_pseudo_csum =
2639 			bnx2x_csum_fix(skb_transport_header(skb),
2640 				       SKB_CS(skb), fix);
2641 
2642 		DP(NETIF_MSG_TX_QUEUED, "csum after fix %x\n",
2643 		   pbd->tcp_pseudo_csum);
2644 	}
2645 
2646 	return hlen;
2647 }
2648 
2649 /* called with netif_tx_lock
2650  * bnx2x_tx_int() runs without netif_tx_lock unless it needs to call
2651  * netif_wake_queue()
2652  */
2653 netdev_tx_t bnx2x_start_xmit(struct sk_buff *skb, struct net_device *dev)
2654 {
2655 	struct bnx2x *bp = netdev_priv(dev);
2656 
2657 	struct bnx2x_fastpath *fp;
2658 	struct netdev_queue *txq;
2659 	struct bnx2x_fp_txdata *txdata;
2660 	struct sw_tx_bd *tx_buf;
2661 	struct eth_tx_start_bd *tx_start_bd, *first_bd;
2662 	struct eth_tx_bd *tx_data_bd, *total_pkt_bd = NULL;
2663 	struct eth_tx_parse_bd_e1x *pbd_e1x = NULL;
2664 	struct eth_tx_parse_bd_e2 *pbd_e2 = NULL;
2665 	u32 pbd_e2_parsing_data = 0;
2666 	u16 pkt_prod, bd_prod;
2667 	int nbd, txq_index, fp_index, txdata_index;
2668 	dma_addr_t mapping;
2669 	u32 xmit_type = bnx2x_xmit_type(bp, skb);
2670 	int i;
2671 	u8 hlen = 0;
2672 	__le16 pkt_size = 0;
2673 	struct ethhdr *eth;
2674 	u8 mac_type = UNICAST_ADDRESS;
2675 
2676 #ifdef BNX2X_STOP_ON_ERROR
2677 	if (unlikely(bp->panic))
2678 		return NETDEV_TX_BUSY;
2679 #endif
2680 
2681 	txq_index = skb_get_queue_mapping(skb);
2682 	txq = netdev_get_tx_queue(dev, txq_index);
2683 
2684 	BUG_ON(txq_index >= MAX_ETH_TXQ_IDX(bp) + FCOE_PRESENT);
2685 
2686 	/* decode the fastpath index and the cos index from the txq */
2687 	fp_index = TXQ_TO_FP(txq_index);
2688 	txdata_index = TXQ_TO_COS(txq_index);
2689 
2690 #ifdef BCM_CNIC
2691 	/*
2692 	 * Override the above for the FCoE queue:
2693 	 *   - FCoE fp entry is right after the ETH entries.
2694 	 *   - FCoE L2 queue uses bp->txdata[0] only.
2695 	 */
2696 	if (unlikely(!NO_FCOE(bp) && (txq_index ==
2697 				      bnx2x_fcoe_tx(bp, txq_index)))) {
2698 		fp_index = FCOE_IDX;
2699 		txdata_index = 0;
2700 	}
2701 #endif
2702 
2703 	/* enable this debug print to view the transmission queue being used
2704 	DP(NETIF_MSG_TX_QUEUED, "indices: txq %d, fp %d, txdata %d\n",
2705 	   txq_index, fp_index, txdata_index); */
2706 
2707 	/* locate the fastpath and the txdata */
2708 	fp = &bp->fp[fp_index];
2709 	txdata = &fp->txdata[txdata_index];
2710 
2711 	/* enable this debug print to view the tranmission details
2712 	DP(NETIF_MSG_TX_QUEUED,
2713 	   "transmitting packet cid %d fp index %d txdata_index %d tx_data ptr %p fp pointer %p\n",
2714 	   txdata->cid, fp_index, txdata_index, txdata, fp); */
2715 
2716 	if (unlikely(bnx2x_tx_avail(bp, txdata) <
2717 		     (skb_shinfo(skb)->nr_frags + 3))) {
2718 		fp->eth_q_stats.driver_xoff++;
2719 		netif_tx_stop_queue(txq);
2720 		BNX2X_ERR("BUG! Tx ring full when queue awake!\n");
2721 		return NETDEV_TX_BUSY;
2722 	}
2723 
2724 	DP(NETIF_MSG_TX_QUEUED,
2725 	   "queue[%d]: SKB: summed %x  protocol %x protocol(%x,%x) gso type %x  xmit_type %x\n",
2726 	   txq_index, skb->ip_summed, skb->protocol, ipv6_hdr(skb)->nexthdr,
2727 	   ip_hdr(skb)->protocol, skb_shinfo(skb)->gso_type, xmit_type);
2728 
2729 	eth = (struct ethhdr *)skb->data;
2730 
2731 	/* set flag according to packet type (UNICAST_ADDRESS is default)*/
2732 	if (unlikely(is_multicast_ether_addr(eth->h_dest))) {
2733 		if (is_broadcast_ether_addr(eth->h_dest))
2734 			mac_type = BROADCAST_ADDRESS;
2735 		else
2736 			mac_type = MULTICAST_ADDRESS;
2737 	}
2738 
2739 #if (MAX_SKB_FRAGS >= MAX_FETCH_BD - 3)
2740 	/* First, check if we need to linearize the skb (due to FW
2741 	   restrictions). No need to check fragmentation if page size > 8K
2742 	   (there will be no violation to FW restrictions) */
2743 	if (bnx2x_pkt_req_lin(bp, skb, xmit_type)) {
2744 		/* Statistics of linearization */
2745 		bp->lin_cnt++;
2746 		if (skb_linearize(skb) != 0) {
2747 			DP(NETIF_MSG_TX_QUEUED,
2748 			   "SKB linearization failed - silently dropping this SKB\n");
2749 			dev_kfree_skb_any(skb);
2750 			return NETDEV_TX_OK;
2751 		}
2752 	}
2753 #endif
2754 	/* Map skb linear data for DMA */
2755 	mapping = dma_map_single(&bp->pdev->dev, skb->data,
2756 				 skb_headlen(skb), DMA_TO_DEVICE);
2757 	if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
2758 		DP(NETIF_MSG_TX_QUEUED,
2759 		   "SKB mapping failed - silently dropping this SKB\n");
2760 		dev_kfree_skb_any(skb);
2761 		return NETDEV_TX_OK;
2762 	}
2763 	/*
2764 	Please read carefully. First we use one BD which we mark as start,
2765 	then we have a parsing info BD (used for TSO or xsum),
2766 	and only then we have the rest of the TSO BDs.
2767 	(don't forget to mark the last one as last,
2768 	and to unmap only AFTER you write to the BD ...)
2769 	And above all, all pdb sizes are in words - NOT DWORDS!
2770 	*/
2771 
2772 	/* get current pkt produced now - advance it just before sending packet
2773 	 * since mapping of pages may fail and cause packet to be dropped
2774 	 */
2775 	pkt_prod = txdata->tx_pkt_prod;
2776 	bd_prod = TX_BD(txdata->tx_bd_prod);
2777 
2778 	/* get a tx_buf and first BD
2779 	 * tx_start_bd may be changed during SPLIT,
2780 	 * but first_bd will always stay first
2781 	 */
2782 	tx_buf = &txdata->tx_buf_ring[TX_BD(pkt_prod)];
2783 	tx_start_bd = &txdata->tx_desc_ring[bd_prod].start_bd;
2784 	first_bd = tx_start_bd;
2785 
2786 	tx_start_bd->bd_flags.as_bitfield = ETH_TX_BD_FLAGS_START_BD;
2787 	SET_FLAG(tx_start_bd->general_data, ETH_TX_START_BD_ETH_ADDR_TYPE,
2788 		 mac_type);
2789 
2790 	/* header nbd */
2791 	SET_FLAG(tx_start_bd->general_data, ETH_TX_START_BD_HDR_NBDS, 1);
2792 
2793 	/* remember the first BD of the packet */
2794 	tx_buf->first_bd = txdata->tx_bd_prod;
2795 	tx_buf->skb = skb;
2796 	tx_buf->flags = 0;
2797 
2798 	DP(NETIF_MSG_TX_QUEUED,
2799 	   "sending pkt %u @%p  next_idx %u  bd %u @%p\n",
2800 	   pkt_prod, tx_buf, txdata->tx_pkt_prod, bd_prod, tx_start_bd);
2801 
2802 	if (vlan_tx_tag_present(skb)) {
2803 		tx_start_bd->vlan_or_ethertype =
2804 		    cpu_to_le16(vlan_tx_tag_get(skb));
2805 		tx_start_bd->bd_flags.as_bitfield |=
2806 		    (X_ETH_OUTBAND_VLAN << ETH_TX_BD_FLAGS_VLAN_MODE_SHIFT);
2807 	} else
2808 		tx_start_bd->vlan_or_ethertype = cpu_to_le16(pkt_prod);
2809 
2810 	/* turn on parsing and get a BD */
2811 	bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
2812 
2813 	if (xmit_type & XMIT_CSUM)
2814 		bnx2x_set_sbd_csum(bp, skb, tx_start_bd, xmit_type);
2815 
2816 	if (!CHIP_IS_E1x(bp)) {
2817 		pbd_e2 = &txdata->tx_desc_ring[bd_prod].parse_bd_e2;
2818 		memset(pbd_e2, 0, sizeof(struct eth_tx_parse_bd_e2));
2819 		/* Set PBD in checksum offload case */
2820 		if (xmit_type & XMIT_CSUM)
2821 			hlen = bnx2x_set_pbd_csum_e2(bp, skb,
2822 						     &pbd_e2_parsing_data,
2823 						     xmit_type);
2824 		if (IS_MF_SI(bp)) {
2825 			/*
2826 			 * fill in the MAC addresses in the PBD - for local
2827 			 * switching
2828 			 */
2829 			bnx2x_set_fw_mac_addr(&pbd_e2->src_mac_addr_hi,
2830 					      &pbd_e2->src_mac_addr_mid,
2831 					      &pbd_e2->src_mac_addr_lo,
2832 					      eth->h_source);
2833 			bnx2x_set_fw_mac_addr(&pbd_e2->dst_mac_addr_hi,
2834 					      &pbd_e2->dst_mac_addr_mid,
2835 					      &pbd_e2->dst_mac_addr_lo,
2836 					      eth->h_dest);
2837 		}
2838 	} else {
2839 		pbd_e1x = &txdata->tx_desc_ring[bd_prod].parse_bd_e1x;
2840 		memset(pbd_e1x, 0, sizeof(struct eth_tx_parse_bd_e1x));
2841 		/* Set PBD in checksum offload case */
2842 		if (xmit_type & XMIT_CSUM)
2843 			hlen = bnx2x_set_pbd_csum(bp, skb, pbd_e1x, xmit_type);
2844 
2845 	}
2846 
2847 	/* Setup the data pointer of the first BD of the packet */
2848 	tx_start_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
2849 	tx_start_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
2850 	nbd = 2; /* start_bd + pbd + frags (updated when pages are mapped) */
2851 	tx_start_bd->nbytes = cpu_to_le16(skb_headlen(skb));
2852 	pkt_size = tx_start_bd->nbytes;
2853 
2854 	DP(NETIF_MSG_TX_QUEUED,
2855 	   "first bd @%p  addr (%x:%x)  nbd %d  nbytes %d  flags %x  vlan %x\n",
2856 	   tx_start_bd, tx_start_bd->addr_hi, tx_start_bd->addr_lo,
2857 	   le16_to_cpu(tx_start_bd->nbd), le16_to_cpu(tx_start_bd->nbytes),
2858 	   tx_start_bd->bd_flags.as_bitfield,
2859 	   le16_to_cpu(tx_start_bd->vlan_or_ethertype));
2860 
2861 	if (xmit_type & XMIT_GSO) {
2862 
2863 		DP(NETIF_MSG_TX_QUEUED,
2864 		   "TSO packet len %d  hlen %d  total len %d  tso size %d\n",
2865 		   skb->len, hlen, skb_headlen(skb),
2866 		   skb_shinfo(skb)->gso_size);
2867 
2868 		tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_SW_LSO;
2869 
2870 		if (unlikely(skb_headlen(skb) > hlen))
2871 			bd_prod = bnx2x_tx_split(bp, txdata, tx_buf,
2872 						 &tx_start_bd, hlen,
2873 						 bd_prod, ++nbd);
2874 		if (!CHIP_IS_E1x(bp))
2875 			bnx2x_set_pbd_gso_e2(skb, &pbd_e2_parsing_data,
2876 					     xmit_type);
2877 		else
2878 			bnx2x_set_pbd_gso(skb, pbd_e1x, xmit_type);
2879 	}
2880 
2881 	/* Set the PBD's parsing_data field if not zero
2882 	 * (for the chips newer than 57711).
2883 	 */
2884 	if (pbd_e2_parsing_data)
2885 		pbd_e2->parsing_data = cpu_to_le32(pbd_e2_parsing_data);
2886 
2887 	tx_data_bd = (struct eth_tx_bd *)tx_start_bd;
2888 
2889 	/* Handle fragmented skb */
2890 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2891 		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2892 
2893 		mapping = skb_frag_dma_map(&bp->pdev->dev, frag, 0,
2894 					   skb_frag_size(frag), DMA_TO_DEVICE);
2895 		if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
2896 			unsigned int pkts_compl = 0, bytes_compl = 0;
2897 
2898 			DP(NETIF_MSG_TX_QUEUED,
2899 			   "Unable to map page - dropping packet...\n");
2900 
2901 			/* we need unmap all buffers already mapped
2902 			 * for this SKB;
2903 			 * first_bd->nbd need to be properly updated
2904 			 * before call to bnx2x_free_tx_pkt
2905 			 */
2906 			first_bd->nbd = cpu_to_le16(nbd);
2907 			bnx2x_free_tx_pkt(bp, txdata,
2908 					  TX_BD(txdata->tx_pkt_prod),
2909 					  &pkts_compl, &bytes_compl);
2910 			return NETDEV_TX_OK;
2911 		}
2912 
2913 		bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
2914 		tx_data_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
2915 		if (total_pkt_bd == NULL)
2916 			total_pkt_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
2917 
2918 		tx_data_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
2919 		tx_data_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
2920 		tx_data_bd->nbytes = cpu_to_le16(skb_frag_size(frag));
2921 		le16_add_cpu(&pkt_size, skb_frag_size(frag));
2922 		nbd++;
2923 
2924 		DP(NETIF_MSG_TX_QUEUED,
2925 		   "frag %d  bd @%p  addr (%x:%x)  nbytes %d\n",
2926 		   i, tx_data_bd, tx_data_bd->addr_hi, tx_data_bd->addr_lo,
2927 		   le16_to_cpu(tx_data_bd->nbytes));
2928 	}
2929 
2930 	DP(NETIF_MSG_TX_QUEUED, "last bd @%p\n", tx_data_bd);
2931 
2932 	/* update with actual num BDs */
2933 	first_bd->nbd = cpu_to_le16(nbd);
2934 
2935 	bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
2936 
2937 	/* now send a tx doorbell, counting the next BD
2938 	 * if the packet contains or ends with it
2939 	 */
2940 	if (TX_BD_POFF(bd_prod) < nbd)
2941 		nbd++;
2942 
2943 	/* total_pkt_bytes should be set on the first data BD if
2944 	 * it's not an LSO packet and there is more than one
2945 	 * data BD. In this case pkt_size is limited by an MTU value.
2946 	 * However we prefer to set it for an LSO packet (while we don't
2947 	 * have to) in order to save some CPU cycles in a none-LSO
2948 	 * case, when we much more care about them.
2949 	 */
2950 	if (total_pkt_bd != NULL)
2951 		total_pkt_bd->total_pkt_bytes = pkt_size;
2952 
2953 	if (pbd_e1x)
2954 		DP(NETIF_MSG_TX_QUEUED,
2955 		   "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",
2956 		   pbd_e1x, pbd_e1x->global_data, pbd_e1x->ip_hlen_w,
2957 		   pbd_e1x->ip_id, pbd_e1x->lso_mss, pbd_e1x->tcp_flags,
2958 		   pbd_e1x->tcp_pseudo_csum, pbd_e1x->tcp_send_seq,
2959 		    le16_to_cpu(pbd_e1x->total_hlen_w));
2960 	if (pbd_e2)
2961 		DP(NETIF_MSG_TX_QUEUED,
2962 		   "PBD (E2) @%p  dst %x %x %x src %x %x %x parsing_data %x\n",
2963 		   pbd_e2, pbd_e2->dst_mac_addr_hi, pbd_e2->dst_mac_addr_mid,
2964 		   pbd_e2->dst_mac_addr_lo, pbd_e2->src_mac_addr_hi,
2965 		   pbd_e2->src_mac_addr_mid, pbd_e2->src_mac_addr_lo,
2966 		   pbd_e2->parsing_data);
2967 	DP(NETIF_MSG_TX_QUEUED, "doorbell: nbd %d  bd %u\n", nbd, bd_prod);
2968 
2969 	netdev_tx_sent_queue(txq, skb->len);
2970 
2971 	txdata->tx_pkt_prod++;
2972 	/*
2973 	 * Make sure that the BD data is updated before updating the producer
2974 	 * since FW might read the BD right after the producer is updated.
2975 	 * This is only applicable for weak-ordered memory model archs such
2976 	 * as IA-64. The following barrier is also mandatory since FW will
2977 	 * assumes packets must have BDs.
2978 	 */
2979 	wmb();
2980 
2981 	txdata->tx_db.data.prod += nbd;
2982 	barrier();
2983 
2984 	DOORBELL(bp, txdata->cid, txdata->tx_db.raw);
2985 
2986 	mmiowb();
2987 
2988 	txdata->tx_bd_prod += nbd;
2989 
2990 	if (unlikely(bnx2x_tx_avail(bp, txdata) < MAX_SKB_FRAGS + 3)) {
2991 		netif_tx_stop_queue(txq);
2992 
2993 		/* paired memory barrier is in bnx2x_tx_int(), we have to keep
2994 		 * ordering of set_bit() in netif_tx_stop_queue() and read of
2995 		 * fp->bd_tx_cons */
2996 		smp_mb();
2997 
2998 		fp->eth_q_stats.driver_xoff++;
2999 		if (bnx2x_tx_avail(bp, txdata) >= MAX_SKB_FRAGS + 3)
3000 			netif_tx_wake_queue(txq);
3001 	}
3002 	txdata->tx_pkt++;
3003 
3004 	return NETDEV_TX_OK;
3005 }
3006 
3007 /**
3008  * bnx2x_setup_tc - routine to configure net_device for multi tc
3009  *
3010  * @netdev: net device to configure
3011  * @tc: number of traffic classes to enable
3012  *
3013  * callback connected to the ndo_setup_tc function pointer
3014  */
3015 int bnx2x_setup_tc(struct net_device *dev, u8 num_tc)
3016 {
3017 	int cos, prio, count, offset;
3018 	struct bnx2x *bp = netdev_priv(dev);
3019 
3020 	/* setup tc must be called under rtnl lock */
3021 	ASSERT_RTNL();
3022 
3023 	/* no traffic classes requested. aborting */
3024 	if (!num_tc) {
3025 		netdev_reset_tc(dev);
3026 		return 0;
3027 	}
3028 
3029 	/* requested to support too many traffic classes */
3030 	if (num_tc > bp->max_cos) {
3031 		BNX2X_ERR("support for too many traffic classes requested: %d. max supported is %d\n",
3032 			  num_tc, bp->max_cos);
3033 		return -EINVAL;
3034 	}
3035 
3036 	/* declare amount of supported traffic classes */
3037 	if (netdev_set_num_tc(dev, num_tc)) {
3038 		BNX2X_ERR("failed to declare %d traffic classes\n", num_tc);
3039 		return -EINVAL;
3040 	}
3041 
3042 	/* configure priority to traffic class mapping */
3043 	for (prio = 0; prio < BNX2X_MAX_PRIORITY; prio++) {
3044 		netdev_set_prio_tc_map(dev, prio, bp->prio_to_cos[prio]);
3045 		DP(BNX2X_MSG_SP | NETIF_MSG_IFUP,
3046 		   "mapping priority %d to tc %d\n",
3047 		   prio, bp->prio_to_cos[prio]);
3048 	}
3049 
3050 
3051 	/* Use this configuration to diffrentiate tc0 from other COSes
3052 	   This can be used for ets or pfc, and save the effort of setting
3053 	   up a multio class queue disc or negotiating DCBX with a switch
3054 	netdev_set_prio_tc_map(dev, 0, 0);
3055 	DP(BNX2X_MSG_SP, "mapping priority %d to tc %d\n", 0, 0);
3056 	for (prio = 1; prio < 16; prio++) {
3057 		netdev_set_prio_tc_map(dev, prio, 1);
3058 		DP(BNX2X_MSG_SP, "mapping priority %d to tc %d\n", prio, 1);
3059 	} */
3060 
3061 	/* configure traffic class to transmission queue mapping */
3062 	for (cos = 0; cos < bp->max_cos; cos++) {
3063 		count = BNX2X_NUM_ETH_QUEUES(bp);
3064 		offset = cos * MAX_TXQS_PER_COS;
3065 		netdev_set_tc_queue(dev, cos, count, offset);
3066 		DP(BNX2X_MSG_SP | NETIF_MSG_IFUP,
3067 		   "mapping tc %d to offset %d count %d\n",
3068 		   cos, offset, count);
3069 	}
3070 
3071 	return 0;
3072 }
3073 
3074 /* called with rtnl_lock */
3075 int bnx2x_change_mac_addr(struct net_device *dev, void *p)
3076 {
3077 	struct sockaddr *addr = p;
3078 	struct bnx2x *bp = netdev_priv(dev);
3079 	int rc = 0;
3080 
3081 	if (!bnx2x_is_valid_ether_addr(bp, addr->sa_data)) {
3082 		BNX2X_ERR("Requested MAC address is not valid\n");
3083 		return -EINVAL;
3084 	}
3085 
3086 #ifdef BCM_CNIC
3087 	if (IS_MF_STORAGE_SD(bp) && !is_zero_ether_addr(addr->sa_data)) {
3088 		BNX2X_ERR("Can't configure non-zero address on iSCSI or FCoE functions in MF-SD mode\n");
3089 		return -EINVAL;
3090 	}
3091 #endif
3092 
3093 	if (netif_running(dev))  {
3094 		rc = bnx2x_set_eth_mac(bp, false);
3095 		if (rc)
3096 			return rc;
3097 	}
3098 
3099 	dev->addr_assign_type &= ~NET_ADDR_RANDOM;
3100 	memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
3101 
3102 	if (netif_running(dev))
3103 		rc = bnx2x_set_eth_mac(bp, true);
3104 
3105 	return rc;
3106 }
3107 
3108 static void bnx2x_free_fp_mem_at(struct bnx2x *bp, int fp_index)
3109 {
3110 	union host_hc_status_block *sb = &bnx2x_fp(bp, fp_index, status_blk);
3111 	struct bnx2x_fastpath *fp = &bp->fp[fp_index];
3112 	u8 cos;
3113 
3114 	/* Common */
3115 #ifdef BCM_CNIC
3116 	if (IS_FCOE_IDX(fp_index)) {
3117 		memset(sb, 0, sizeof(union host_hc_status_block));
3118 		fp->status_blk_mapping = 0;
3119 
3120 	} else {
3121 #endif
3122 		/* status blocks */
3123 		if (!CHIP_IS_E1x(bp))
3124 			BNX2X_PCI_FREE(sb->e2_sb,
3125 				       bnx2x_fp(bp, fp_index,
3126 						status_blk_mapping),
3127 				       sizeof(struct host_hc_status_block_e2));
3128 		else
3129 			BNX2X_PCI_FREE(sb->e1x_sb,
3130 				       bnx2x_fp(bp, fp_index,
3131 						status_blk_mapping),
3132 				       sizeof(struct host_hc_status_block_e1x));
3133 #ifdef BCM_CNIC
3134 	}
3135 #endif
3136 	/* Rx */
3137 	if (!skip_rx_queue(bp, fp_index)) {
3138 		bnx2x_free_rx_bds(fp);
3139 
3140 		/* fastpath rx rings: rx_buf rx_desc rx_comp */
3141 		BNX2X_FREE(bnx2x_fp(bp, fp_index, rx_buf_ring));
3142 		BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_desc_ring),
3143 			       bnx2x_fp(bp, fp_index, rx_desc_mapping),
3144 			       sizeof(struct eth_rx_bd) * NUM_RX_BD);
3145 
3146 		BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_comp_ring),
3147 			       bnx2x_fp(bp, fp_index, rx_comp_mapping),
3148 			       sizeof(struct eth_fast_path_rx_cqe) *
3149 			       NUM_RCQ_BD);
3150 
3151 		/* SGE ring */
3152 		BNX2X_FREE(bnx2x_fp(bp, fp_index, rx_page_ring));
3153 		BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_sge_ring),
3154 			       bnx2x_fp(bp, fp_index, rx_sge_mapping),
3155 			       BCM_PAGE_SIZE * NUM_RX_SGE_PAGES);
3156 	}
3157 
3158 	/* Tx */
3159 	if (!skip_tx_queue(bp, fp_index)) {
3160 		/* fastpath tx rings: tx_buf tx_desc */
3161 		for_each_cos_in_tx_queue(fp, cos) {
3162 			struct bnx2x_fp_txdata *txdata = &fp->txdata[cos];
3163 
3164 			DP(NETIF_MSG_IFDOWN,
3165 			   "freeing tx memory of fp %d cos %d cid %d\n",
3166 			   fp_index, cos, txdata->cid);
3167 
3168 			BNX2X_FREE(txdata->tx_buf_ring);
3169 			BNX2X_PCI_FREE(txdata->tx_desc_ring,
3170 				txdata->tx_desc_mapping,
3171 				sizeof(union eth_tx_bd_types) * NUM_TX_BD);
3172 		}
3173 	}
3174 	/* end of fastpath */
3175 }
3176 
3177 void bnx2x_free_fp_mem(struct bnx2x *bp)
3178 {
3179 	int i;
3180 	for_each_queue(bp, i)
3181 		bnx2x_free_fp_mem_at(bp, i);
3182 }
3183 
3184 static inline void set_sb_shortcuts(struct bnx2x *bp, int index)
3185 {
3186 	union host_hc_status_block status_blk = bnx2x_fp(bp, index, status_blk);
3187 	if (!CHIP_IS_E1x(bp)) {
3188 		bnx2x_fp(bp, index, sb_index_values) =
3189 			(__le16 *)status_blk.e2_sb->sb.index_values;
3190 		bnx2x_fp(bp, index, sb_running_index) =
3191 			(__le16 *)status_blk.e2_sb->sb.running_index;
3192 	} else {
3193 		bnx2x_fp(bp, index, sb_index_values) =
3194 			(__le16 *)status_blk.e1x_sb->sb.index_values;
3195 		bnx2x_fp(bp, index, sb_running_index) =
3196 			(__le16 *)status_blk.e1x_sb->sb.running_index;
3197 	}
3198 }
3199 
3200 static int bnx2x_alloc_fp_mem_at(struct bnx2x *bp, int index)
3201 {
3202 	union host_hc_status_block *sb;
3203 	struct bnx2x_fastpath *fp = &bp->fp[index];
3204 	int ring_size = 0;
3205 	u8 cos;
3206 	int rx_ring_size = 0;
3207 
3208 #ifdef BCM_CNIC
3209 	if (!bp->rx_ring_size && IS_MF_STORAGE_SD(bp)) {
3210 		rx_ring_size = MIN_RX_SIZE_NONTPA;
3211 		bp->rx_ring_size = rx_ring_size;
3212 	} else
3213 #endif
3214 	if (!bp->rx_ring_size) {
3215 		u32 cfg = SHMEM_RD(bp,
3216 			     dev_info.port_hw_config[BP_PORT(bp)].default_cfg);
3217 
3218 		rx_ring_size = MAX_RX_AVAIL/BNX2X_NUM_RX_QUEUES(bp);
3219 
3220 		/* Dercease ring size for 1G functions */
3221 		if ((cfg & PORT_HW_CFG_NET_SERDES_IF_MASK) ==
3222 		    PORT_HW_CFG_NET_SERDES_IF_SGMII)
3223 			rx_ring_size /= 10;
3224 
3225 		/* allocate at least number of buffers required by FW */
3226 		rx_ring_size = max_t(int, bp->disable_tpa ? MIN_RX_SIZE_NONTPA :
3227 				     MIN_RX_SIZE_TPA, rx_ring_size);
3228 
3229 		bp->rx_ring_size = rx_ring_size;
3230 	} else /* if rx_ring_size specified - use it */
3231 		rx_ring_size = bp->rx_ring_size;
3232 
3233 	/* Common */
3234 	sb = &bnx2x_fp(bp, index, status_blk);
3235 #ifdef BCM_CNIC
3236 	if (!IS_FCOE_IDX(index)) {
3237 #endif
3238 		/* status blocks */
3239 		if (!CHIP_IS_E1x(bp))
3240 			BNX2X_PCI_ALLOC(sb->e2_sb,
3241 				&bnx2x_fp(bp, index, status_blk_mapping),
3242 				sizeof(struct host_hc_status_block_e2));
3243 		else
3244 			BNX2X_PCI_ALLOC(sb->e1x_sb,
3245 				&bnx2x_fp(bp, index, status_blk_mapping),
3246 			    sizeof(struct host_hc_status_block_e1x));
3247 #ifdef BCM_CNIC
3248 	}
3249 #endif
3250 
3251 	/* FCoE Queue uses Default SB and doesn't ACK the SB, thus no need to
3252 	 * set shortcuts for it.
3253 	 */
3254 	if (!IS_FCOE_IDX(index))
3255 		set_sb_shortcuts(bp, index);
3256 
3257 	/* Tx */
3258 	if (!skip_tx_queue(bp, index)) {
3259 		/* fastpath tx rings: tx_buf tx_desc */
3260 		for_each_cos_in_tx_queue(fp, cos) {
3261 			struct bnx2x_fp_txdata *txdata = &fp->txdata[cos];
3262 
3263 			DP(NETIF_MSG_IFUP,
3264 			   "allocating tx memory of fp %d cos %d\n",
3265 			   index, cos);
3266 
3267 			BNX2X_ALLOC(txdata->tx_buf_ring,
3268 				sizeof(struct sw_tx_bd) * NUM_TX_BD);
3269 			BNX2X_PCI_ALLOC(txdata->tx_desc_ring,
3270 				&txdata->tx_desc_mapping,
3271 				sizeof(union eth_tx_bd_types) * NUM_TX_BD);
3272 		}
3273 	}
3274 
3275 	/* Rx */
3276 	if (!skip_rx_queue(bp, index)) {
3277 		/* fastpath rx rings: rx_buf rx_desc rx_comp */
3278 		BNX2X_ALLOC(bnx2x_fp(bp, index, rx_buf_ring),
3279 				sizeof(struct sw_rx_bd) * NUM_RX_BD);
3280 		BNX2X_PCI_ALLOC(bnx2x_fp(bp, index, rx_desc_ring),
3281 				&bnx2x_fp(bp, index, rx_desc_mapping),
3282 				sizeof(struct eth_rx_bd) * NUM_RX_BD);
3283 
3284 		BNX2X_PCI_ALLOC(bnx2x_fp(bp, index, rx_comp_ring),
3285 				&bnx2x_fp(bp, index, rx_comp_mapping),
3286 				sizeof(struct eth_fast_path_rx_cqe) *
3287 				NUM_RCQ_BD);
3288 
3289 		/* SGE ring */
3290 		BNX2X_ALLOC(bnx2x_fp(bp, index, rx_page_ring),
3291 				sizeof(struct sw_rx_page) * NUM_RX_SGE);
3292 		BNX2X_PCI_ALLOC(bnx2x_fp(bp, index, rx_sge_ring),
3293 				&bnx2x_fp(bp, index, rx_sge_mapping),
3294 				BCM_PAGE_SIZE * NUM_RX_SGE_PAGES);
3295 		/* RX BD ring */
3296 		bnx2x_set_next_page_rx_bd(fp);
3297 
3298 		/* CQ ring */
3299 		bnx2x_set_next_page_rx_cq(fp);
3300 
3301 		/* BDs */
3302 		ring_size = bnx2x_alloc_rx_bds(fp, rx_ring_size);
3303 		if (ring_size < rx_ring_size)
3304 			goto alloc_mem_err;
3305 	}
3306 
3307 	return 0;
3308 
3309 /* handles low memory cases */
3310 alloc_mem_err:
3311 	BNX2X_ERR("Unable to allocate full memory for queue %d (size %d)\n",
3312 						index, ring_size);
3313 	/* FW will drop all packets if queue is not big enough,
3314 	 * In these cases we disable the queue
3315 	 * Min size is different for OOO, TPA and non-TPA queues
3316 	 */
3317 	if (ring_size < (fp->disable_tpa ?
3318 				MIN_RX_SIZE_NONTPA : MIN_RX_SIZE_TPA)) {
3319 			/* release memory allocated for this queue */
3320 			bnx2x_free_fp_mem_at(bp, index);
3321 			return -ENOMEM;
3322 	}
3323 	return 0;
3324 }
3325 
3326 int bnx2x_alloc_fp_mem(struct bnx2x *bp)
3327 {
3328 	int i;
3329 
3330 	/**
3331 	 * 1. Allocate FP for leading - fatal if error
3332 	 * 2. {CNIC} Allocate FCoE FP - fatal if error
3333 	 * 3. {CNIC} Allocate OOO + FWD - disable OOO if error
3334 	 * 4. Allocate RSS - fix number of queues if error
3335 	 */
3336 
3337 	/* leading */
3338 	if (bnx2x_alloc_fp_mem_at(bp, 0))
3339 		return -ENOMEM;
3340 
3341 #ifdef BCM_CNIC
3342 	if (!NO_FCOE(bp))
3343 		/* FCoE */
3344 		if (bnx2x_alloc_fp_mem_at(bp, FCOE_IDX))
3345 			/* we will fail load process instead of mark
3346 			 * NO_FCOE_FLAG
3347 			 */
3348 			return -ENOMEM;
3349 #endif
3350 
3351 	/* RSS */
3352 	for_each_nondefault_eth_queue(bp, i)
3353 		if (bnx2x_alloc_fp_mem_at(bp, i))
3354 			break;
3355 
3356 	/* handle memory failures */
3357 	if (i != BNX2X_NUM_ETH_QUEUES(bp)) {
3358 		int delta = BNX2X_NUM_ETH_QUEUES(bp) - i;
3359 
3360 		WARN_ON(delta < 0);
3361 #ifdef BCM_CNIC
3362 		/**
3363 		 * move non eth FPs next to last eth FP
3364 		 * must be done in that order
3365 		 * FCOE_IDX < FWD_IDX < OOO_IDX
3366 		 */
3367 
3368 		/* move FCoE fp even NO_FCOE_FLAG is on */
3369 		bnx2x_move_fp(bp, FCOE_IDX, FCOE_IDX - delta);
3370 #endif
3371 		bp->num_queues -= delta;
3372 		BNX2X_ERR("Adjusted num of queues from %d to %d\n",
3373 			  bp->num_queues + delta, bp->num_queues);
3374 	}
3375 
3376 	return 0;
3377 }
3378 
3379 void bnx2x_free_mem_bp(struct bnx2x *bp)
3380 {
3381 	kfree(bp->fp);
3382 	kfree(bp->msix_table);
3383 	kfree(bp->ilt);
3384 }
3385 
3386 int __devinit bnx2x_alloc_mem_bp(struct bnx2x *bp)
3387 {
3388 	struct bnx2x_fastpath *fp;
3389 	struct msix_entry *tbl;
3390 	struct bnx2x_ilt *ilt;
3391 	int msix_table_size = 0;
3392 
3393 	/*
3394 	 * The biggest MSI-X table we might need is as a maximum number of fast
3395 	 * path IGU SBs plus default SB (for PF).
3396 	 */
3397 	msix_table_size = bp->igu_sb_cnt + 1;
3398 
3399 	/* fp array: RSS plus CNIC related L2 queues */
3400 	fp = kcalloc(BNX2X_MAX_RSS_COUNT(bp) + NON_ETH_CONTEXT_USE,
3401 		     sizeof(*fp), GFP_KERNEL);
3402 	if (!fp)
3403 		goto alloc_err;
3404 	bp->fp = fp;
3405 
3406 	/* msix table */
3407 	tbl = kcalloc(msix_table_size, sizeof(*tbl), GFP_KERNEL);
3408 	if (!tbl)
3409 		goto alloc_err;
3410 	bp->msix_table = tbl;
3411 
3412 	/* ilt */
3413 	ilt = kzalloc(sizeof(*ilt), GFP_KERNEL);
3414 	if (!ilt)
3415 		goto alloc_err;
3416 	bp->ilt = ilt;
3417 
3418 	return 0;
3419 alloc_err:
3420 	bnx2x_free_mem_bp(bp);
3421 	return -ENOMEM;
3422 
3423 }
3424 
3425 int bnx2x_reload_if_running(struct net_device *dev)
3426 {
3427 	struct bnx2x *bp = netdev_priv(dev);
3428 
3429 	if (unlikely(!netif_running(dev)))
3430 		return 0;
3431 
3432 	bnx2x_nic_unload(bp, UNLOAD_NORMAL);
3433 	return bnx2x_nic_load(bp, LOAD_NORMAL);
3434 }
3435 
3436 int bnx2x_get_cur_phy_idx(struct bnx2x *bp)
3437 {
3438 	u32 sel_phy_idx = 0;
3439 	if (bp->link_params.num_phys <= 1)
3440 		return INT_PHY;
3441 
3442 	if (bp->link_vars.link_up) {
3443 		sel_phy_idx = EXT_PHY1;
3444 		/* In case link is SERDES, check if the EXT_PHY2 is the one */
3445 		if ((bp->link_vars.link_status & LINK_STATUS_SERDES_LINK) &&
3446 		    (bp->link_params.phy[EXT_PHY2].supported & SUPPORTED_FIBRE))
3447 			sel_phy_idx = EXT_PHY2;
3448 	} else {
3449 
3450 		switch (bnx2x_phy_selection(&bp->link_params)) {
3451 		case PORT_HW_CFG_PHY_SELECTION_HARDWARE_DEFAULT:
3452 		case PORT_HW_CFG_PHY_SELECTION_FIRST_PHY:
3453 		case PORT_HW_CFG_PHY_SELECTION_FIRST_PHY_PRIORITY:
3454 		       sel_phy_idx = EXT_PHY1;
3455 		       break;
3456 		case PORT_HW_CFG_PHY_SELECTION_SECOND_PHY:
3457 		case PORT_HW_CFG_PHY_SELECTION_SECOND_PHY_PRIORITY:
3458 		       sel_phy_idx = EXT_PHY2;
3459 		       break;
3460 		}
3461 	}
3462 
3463 	return sel_phy_idx;
3464 
3465 }
3466 int bnx2x_get_link_cfg_idx(struct bnx2x *bp)
3467 {
3468 	u32 sel_phy_idx = bnx2x_get_cur_phy_idx(bp);
3469 	/*
3470 	 * The selected actived PHY is always after swapping (in case PHY
3471 	 * swapping is enabled). So when swapping is enabled, we need to reverse
3472 	 * the configuration
3473 	 */
3474 
3475 	if (bp->link_params.multi_phy_config &
3476 	    PORT_HW_CFG_PHY_SWAPPED_ENABLED) {
3477 		if (sel_phy_idx == EXT_PHY1)
3478 			sel_phy_idx = EXT_PHY2;
3479 		else if (sel_phy_idx == EXT_PHY2)
3480 			sel_phy_idx = EXT_PHY1;
3481 	}
3482 	return LINK_CONFIG_IDX(sel_phy_idx);
3483 }
3484 
3485 #if defined(NETDEV_FCOE_WWNN) && defined(BCM_CNIC)
3486 int bnx2x_fcoe_get_wwn(struct net_device *dev, u64 *wwn, int type)
3487 {
3488 	struct bnx2x *bp = netdev_priv(dev);
3489 	struct cnic_eth_dev *cp = &bp->cnic_eth_dev;
3490 
3491 	switch (type) {
3492 	case NETDEV_FCOE_WWNN:
3493 		*wwn = HILO_U64(cp->fcoe_wwn_node_name_hi,
3494 				cp->fcoe_wwn_node_name_lo);
3495 		break;
3496 	case NETDEV_FCOE_WWPN:
3497 		*wwn = HILO_U64(cp->fcoe_wwn_port_name_hi,
3498 				cp->fcoe_wwn_port_name_lo);
3499 		break;
3500 	default:
3501 		BNX2X_ERR("Wrong WWN type requested - %d\n", type);
3502 		return -EINVAL;
3503 	}
3504 
3505 	return 0;
3506 }
3507 #endif
3508 
3509 /* called with rtnl_lock */
3510 int bnx2x_change_mtu(struct net_device *dev, int new_mtu)
3511 {
3512 	struct bnx2x *bp = netdev_priv(dev);
3513 
3514 	if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
3515 		BNX2X_ERR("Can't perform change MTU during parity recovery\n");
3516 		return -EAGAIN;
3517 	}
3518 
3519 	if ((new_mtu > ETH_MAX_JUMBO_PACKET_SIZE) ||
3520 	    ((new_mtu + ETH_HLEN) < ETH_MIN_PACKET_SIZE)) {
3521 		BNX2X_ERR("Can't support requested MTU size\n");
3522 		return -EINVAL;
3523 	}
3524 
3525 	/* This does not race with packet allocation
3526 	 * because the actual alloc size is
3527 	 * only updated as part of load
3528 	 */
3529 	dev->mtu = new_mtu;
3530 
3531 	bp->gro_check = bnx2x_need_gro_check(new_mtu);
3532 
3533 	return bnx2x_reload_if_running(dev);
3534 }
3535 
3536 netdev_features_t bnx2x_fix_features(struct net_device *dev,
3537 				     netdev_features_t features)
3538 {
3539 	struct bnx2x *bp = netdev_priv(dev);
3540 
3541 	/* TPA requires Rx CSUM offloading */
3542 	if (!(features & NETIF_F_RXCSUM) || bp->disable_tpa) {
3543 		features &= ~NETIF_F_LRO;
3544 		features &= ~NETIF_F_GRO;
3545 	}
3546 
3547 	return features;
3548 }
3549 
3550 int bnx2x_set_features(struct net_device *dev, netdev_features_t features)
3551 {
3552 	struct bnx2x *bp = netdev_priv(dev);
3553 	u32 flags = bp->flags;
3554 	bool bnx2x_reload = false;
3555 
3556 	if (features & NETIF_F_LRO)
3557 		flags |= TPA_ENABLE_FLAG;
3558 	else
3559 		flags &= ~TPA_ENABLE_FLAG;
3560 
3561 	if (features & NETIF_F_GRO)
3562 		flags |= GRO_ENABLE_FLAG;
3563 	else
3564 		flags &= ~GRO_ENABLE_FLAG;
3565 
3566 	if (features & NETIF_F_LOOPBACK) {
3567 		if (bp->link_params.loopback_mode != LOOPBACK_BMAC) {
3568 			bp->link_params.loopback_mode = LOOPBACK_BMAC;
3569 			bnx2x_reload = true;
3570 		}
3571 	} else {
3572 		if (bp->link_params.loopback_mode != LOOPBACK_NONE) {
3573 			bp->link_params.loopback_mode = LOOPBACK_NONE;
3574 			bnx2x_reload = true;
3575 		}
3576 	}
3577 
3578 	if (flags ^ bp->flags) {
3579 		bp->flags = flags;
3580 		bnx2x_reload = true;
3581 	}
3582 
3583 	if (bnx2x_reload) {
3584 		if (bp->recovery_state == BNX2X_RECOVERY_DONE)
3585 			return bnx2x_reload_if_running(dev);
3586 		/* else: bnx2x_nic_load() will be called at end of recovery */
3587 	}
3588 
3589 	return 0;
3590 }
3591 
3592 void bnx2x_tx_timeout(struct net_device *dev)
3593 {
3594 	struct bnx2x *bp = netdev_priv(dev);
3595 
3596 #ifdef BNX2X_STOP_ON_ERROR
3597 	if (!bp->panic)
3598 		bnx2x_panic();
3599 #endif
3600 
3601 	smp_mb__before_clear_bit();
3602 	set_bit(BNX2X_SP_RTNL_TX_TIMEOUT, &bp->sp_rtnl_state);
3603 	smp_mb__after_clear_bit();
3604 
3605 	/* This allows the netif to be shutdown gracefully before resetting */
3606 	schedule_delayed_work(&bp->sp_rtnl_task, 0);
3607 }
3608 
3609 int bnx2x_suspend(struct pci_dev *pdev, pm_message_t state)
3610 {
3611 	struct net_device *dev = pci_get_drvdata(pdev);
3612 	struct bnx2x *bp;
3613 
3614 	if (!dev) {
3615 		dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
3616 		return -ENODEV;
3617 	}
3618 	bp = netdev_priv(dev);
3619 
3620 	rtnl_lock();
3621 
3622 	pci_save_state(pdev);
3623 
3624 	if (!netif_running(dev)) {
3625 		rtnl_unlock();
3626 		return 0;
3627 	}
3628 
3629 	netif_device_detach(dev);
3630 
3631 	bnx2x_nic_unload(bp, UNLOAD_CLOSE);
3632 
3633 	bnx2x_set_power_state(bp, pci_choose_state(pdev, state));
3634 
3635 	rtnl_unlock();
3636 
3637 	return 0;
3638 }
3639 
3640 int bnx2x_resume(struct pci_dev *pdev)
3641 {
3642 	struct net_device *dev = pci_get_drvdata(pdev);
3643 	struct bnx2x *bp;
3644 	int rc;
3645 
3646 	if (!dev) {
3647 		dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
3648 		return -ENODEV;
3649 	}
3650 	bp = netdev_priv(dev);
3651 
3652 	if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
3653 		BNX2X_ERR("Handling parity error recovery. Try again later\n");
3654 		return -EAGAIN;
3655 	}
3656 
3657 	rtnl_lock();
3658 
3659 	pci_restore_state(pdev);
3660 
3661 	if (!netif_running(dev)) {
3662 		rtnl_unlock();
3663 		return 0;
3664 	}
3665 
3666 	bnx2x_set_power_state(bp, PCI_D0);
3667 	netif_device_attach(dev);
3668 
3669 	rc = bnx2x_nic_load(bp, LOAD_OPEN);
3670 
3671 	rtnl_unlock();
3672 
3673 	return rc;
3674 }
3675 
3676 
3677 void bnx2x_set_ctx_validation(struct bnx2x *bp, struct eth_context *cxt,
3678 			      u32 cid)
3679 {
3680 	/* ustorm cxt validation */
3681 	cxt->ustorm_ag_context.cdu_usage =
3682 		CDU_RSRVD_VALUE_TYPE_A(HW_CID(bp, cid),
3683 			CDU_REGION_NUMBER_UCM_AG, ETH_CONNECTION_TYPE);
3684 	/* xcontext validation */
3685 	cxt->xstorm_ag_context.cdu_reserved =
3686 		CDU_RSRVD_VALUE_TYPE_A(HW_CID(bp, cid),
3687 			CDU_REGION_NUMBER_XCM_AG, ETH_CONNECTION_TYPE);
3688 }
3689 
3690 static inline void storm_memset_hc_timeout(struct bnx2x *bp, u8 port,
3691 					     u8 fw_sb_id, u8 sb_index,
3692 					     u8 ticks)
3693 {
3694 
3695 	u32 addr = BAR_CSTRORM_INTMEM +
3696 		   CSTORM_STATUS_BLOCK_DATA_TIMEOUT_OFFSET(fw_sb_id, sb_index);
3697 	REG_WR8(bp, addr, ticks);
3698 	DP(NETIF_MSG_IFUP,
3699 	   "port %x fw_sb_id %d sb_index %d ticks %d\n",
3700 	   port, fw_sb_id, sb_index, ticks);
3701 }
3702 
3703 static inline void storm_memset_hc_disable(struct bnx2x *bp, u8 port,
3704 					     u16 fw_sb_id, u8 sb_index,
3705 					     u8 disable)
3706 {
3707 	u32 enable_flag = disable ? 0 : (1 << HC_INDEX_DATA_HC_ENABLED_SHIFT);
3708 	u32 addr = BAR_CSTRORM_INTMEM +
3709 		   CSTORM_STATUS_BLOCK_DATA_FLAGS_OFFSET(fw_sb_id, sb_index);
3710 	u16 flags = REG_RD16(bp, addr);
3711 	/* clear and set */
3712 	flags &= ~HC_INDEX_DATA_HC_ENABLED;
3713 	flags |= enable_flag;
3714 	REG_WR16(bp, addr, flags);
3715 	DP(NETIF_MSG_IFUP,
3716 	   "port %x fw_sb_id %d sb_index %d disable %d\n",
3717 	   port, fw_sb_id, sb_index, disable);
3718 }
3719 
3720 void bnx2x_update_coalesce_sb_index(struct bnx2x *bp, u8 fw_sb_id,
3721 				    u8 sb_index, u8 disable, u16 usec)
3722 {
3723 	int port = BP_PORT(bp);
3724 	u8 ticks = usec / BNX2X_BTR;
3725 
3726 	storm_memset_hc_timeout(bp, port, fw_sb_id, sb_index, ticks);
3727 
3728 	disable = disable ? 1 : (usec ? 0 : 1);
3729 	storm_memset_hc_disable(bp, port, fw_sb_id, sb_index, disable);
3730 }
3731