1 /* bnx2x_cmn.c: Broadcom Everest network driver.
2  *
3  * Copyright (c) 2007-2013 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/tcp.h>
25 #include <net/ipv6.h>
26 #include <net/ip6_checksum.h>
27 #include <net/busy_poll.h>
28 #include <linux/prefetch.h>
29 #include "bnx2x_cmn.h"
30 #include "bnx2x_init.h"
31 #include "bnx2x_sp.h"
32 
33 /**
34  * bnx2x_move_fp - move content of the fastpath structure.
35  *
36  * @bp:		driver handle
37  * @from:	source FP index
38  * @to:		destination FP index
39  *
40  * Makes sure the contents of the bp->fp[to].napi is kept
41  * intact. This is done by first copying the napi struct from
42  * the target to the source, and then mem copying the entire
43  * source onto the target. Update txdata pointers and related
44  * content.
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 	struct bnx2x_sp_objs *from_sp_objs = &bp->sp_objs[from];
51 	struct bnx2x_sp_objs *to_sp_objs = &bp->sp_objs[to];
52 	struct bnx2x_fp_stats *from_fp_stats = &bp->fp_stats[from];
53 	struct bnx2x_fp_stats *to_fp_stats = &bp->fp_stats[to];
54 	int old_max_eth_txqs, new_max_eth_txqs;
55 	int old_txdata_index = 0, new_txdata_index = 0;
56 	struct bnx2x_agg_info *old_tpa_info = to_fp->tpa_info;
57 
58 	/* Copy the NAPI object as it has been already initialized */
59 	from_fp->napi = to_fp->napi;
60 
61 	/* Move bnx2x_fastpath contents */
62 	memcpy(to_fp, from_fp, sizeof(*to_fp));
63 	to_fp->index = to;
64 
65 	/* Retain the tpa_info of the original `to' version as we don't want
66 	 * 2 FPs to contain the same tpa_info pointer.
67 	 */
68 	to_fp->tpa_info = old_tpa_info;
69 
70 	/* move sp_objs contents as well, as their indices match fp ones */
71 	memcpy(to_sp_objs, from_sp_objs, sizeof(*to_sp_objs));
72 
73 	/* move fp_stats contents as well, as their indices match fp ones */
74 	memcpy(to_fp_stats, from_fp_stats, sizeof(*to_fp_stats));
75 
76 	/* Update txdata pointers in fp and move txdata content accordingly:
77 	 * Each fp consumes 'max_cos' txdata structures, so the index should be
78 	 * decremented by max_cos x delta.
79 	 */
80 
81 	old_max_eth_txqs = BNX2X_NUM_ETH_QUEUES(bp) * (bp)->max_cos;
82 	new_max_eth_txqs = (BNX2X_NUM_ETH_QUEUES(bp) - from + to) *
83 				(bp)->max_cos;
84 	if (from == FCOE_IDX(bp)) {
85 		old_txdata_index = old_max_eth_txqs + FCOE_TXQ_IDX_OFFSET;
86 		new_txdata_index = new_max_eth_txqs + FCOE_TXQ_IDX_OFFSET;
87 	}
88 
89 	memcpy(&bp->bnx2x_txq[new_txdata_index],
90 	       &bp->bnx2x_txq[old_txdata_index],
91 	       sizeof(struct bnx2x_fp_txdata));
92 	to_fp->txdata_ptr[0] = &bp->bnx2x_txq[new_txdata_index];
93 }
94 
95 /**
96  * bnx2x_fill_fw_str - Fill buffer with FW version string.
97  *
98  * @bp:        driver handle
99  * @buf:       character buffer to fill with the fw name
100  * @buf_len:   length of the above buffer
101  *
102  */
103 void bnx2x_fill_fw_str(struct bnx2x *bp, char *buf, size_t buf_len)
104 {
105 	if (IS_PF(bp)) {
106 		u8 phy_fw_ver[PHY_FW_VER_LEN];
107 
108 		phy_fw_ver[0] = '\0';
109 		bnx2x_get_ext_phy_fw_version(&bp->link_params,
110 					     phy_fw_ver, PHY_FW_VER_LEN);
111 		strlcpy(buf, bp->fw_ver, buf_len);
112 		snprintf(buf + strlen(bp->fw_ver), 32 - strlen(bp->fw_ver),
113 			 "bc %d.%d.%d%s%s",
114 			 (bp->common.bc_ver & 0xff0000) >> 16,
115 			 (bp->common.bc_ver & 0xff00) >> 8,
116 			 (bp->common.bc_ver & 0xff),
117 			 ((phy_fw_ver[0] != '\0') ? " phy " : ""), phy_fw_ver);
118 	} else {
119 		bnx2x_vf_fill_fw_str(bp, buf, buf_len);
120 	}
121 }
122 
123 /**
124  * bnx2x_shrink_eth_fp - guarantees fastpath structures stay intact
125  *
126  * @bp:	driver handle
127  * @delta:	number of eth queues which were not allocated
128  */
129 static void bnx2x_shrink_eth_fp(struct bnx2x *bp, int delta)
130 {
131 	int i, cos, old_eth_num = BNX2X_NUM_ETH_QUEUES(bp);
132 
133 	/* Queue pointer cannot be re-set on an fp-basis, as moving pointer
134 	 * backward along the array could cause memory to be overridden
135 	 */
136 	for (cos = 1; cos < bp->max_cos; cos++) {
137 		for (i = 0; i < old_eth_num - delta; i++) {
138 			struct bnx2x_fastpath *fp = &bp->fp[i];
139 			int new_idx = cos * (old_eth_num - delta) + i;
140 
141 			memcpy(&bp->bnx2x_txq[new_idx], fp->txdata_ptr[cos],
142 			       sizeof(struct bnx2x_fp_txdata));
143 			fp->txdata_ptr[cos] = &bp->bnx2x_txq[new_idx];
144 		}
145 	}
146 }
147 
148 int load_count[2][3] = { {0} }; /* per-path: 0-common, 1-port0, 2-port1 */
149 
150 /* free skb in the packet ring at pos idx
151  * return idx of last bd freed
152  */
153 static u16 bnx2x_free_tx_pkt(struct bnx2x *bp, struct bnx2x_fp_txdata *txdata,
154 			     u16 idx, unsigned int *pkts_compl,
155 			     unsigned int *bytes_compl)
156 {
157 	struct sw_tx_bd *tx_buf = &txdata->tx_buf_ring[idx];
158 	struct eth_tx_start_bd *tx_start_bd;
159 	struct eth_tx_bd *tx_data_bd;
160 	struct sk_buff *skb = tx_buf->skb;
161 	u16 bd_idx = TX_BD(tx_buf->first_bd), new_cons;
162 	int nbd;
163 
164 	/* prefetch skb end pointer to speedup dev_kfree_skb() */
165 	prefetch(&skb->end);
166 
167 	DP(NETIF_MSG_TX_DONE, "fp[%d]: pkt_idx %d  buff @(%p)->skb %p\n",
168 	   txdata->txq_index, idx, tx_buf, skb);
169 
170 	/* unmap first bd */
171 	tx_start_bd = &txdata->tx_desc_ring[bd_idx].start_bd;
172 	dma_unmap_single(&bp->pdev->dev, BD_UNMAP_ADDR(tx_start_bd),
173 			 BD_UNMAP_LEN(tx_start_bd), DMA_TO_DEVICE);
174 
175 	nbd = le16_to_cpu(tx_start_bd->nbd) - 1;
176 #ifdef BNX2X_STOP_ON_ERROR
177 	if ((nbd - 1) > (MAX_SKB_FRAGS + 2)) {
178 		BNX2X_ERR("BAD nbd!\n");
179 		bnx2x_panic();
180 	}
181 #endif
182 	new_cons = nbd + tx_buf->first_bd;
183 
184 	/* Get the next bd */
185 	bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
186 
187 	/* Skip a parse bd... */
188 	--nbd;
189 	bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
190 
191 	/* ...and the TSO split header bd since they have no mapping */
192 	if (tx_buf->flags & BNX2X_TSO_SPLIT_BD) {
193 		--nbd;
194 		bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
195 	}
196 
197 	/* now free frags */
198 	while (nbd > 0) {
199 
200 		tx_data_bd = &txdata->tx_desc_ring[bd_idx].reg_bd;
201 		dma_unmap_page(&bp->pdev->dev, BD_UNMAP_ADDR(tx_data_bd),
202 			       BD_UNMAP_LEN(tx_data_bd), DMA_TO_DEVICE);
203 		if (--nbd)
204 			bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
205 	}
206 
207 	/* release skb */
208 	WARN_ON(!skb);
209 	if (likely(skb)) {
210 		(*pkts_compl)++;
211 		(*bytes_compl) += skb->len;
212 	}
213 
214 	dev_kfree_skb_any(skb);
215 	tx_buf->first_bd = 0;
216 	tx_buf->skb = NULL;
217 
218 	return new_cons;
219 }
220 
221 int bnx2x_tx_int(struct bnx2x *bp, struct bnx2x_fp_txdata *txdata)
222 {
223 	struct netdev_queue *txq;
224 	u16 hw_cons, sw_cons, bd_cons = txdata->tx_bd_cons;
225 	unsigned int pkts_compl = 0, bytes_compl = 0;
226 
227 #ifdef BNX2X_STOP_ON_ERROR
228 	if (unlikely(bp->panic))
229 		return -1;
230 #endif
231 
232 	txq = netdev_get_tx_queue(bp->dev, txdata->txq_index);
233 	hw_cons = le16_to_cpu(*txdata->tx_cons_sb);
234 	sw_cons = txdata->tx_pkt_cons;
235 
236 	while (sw_cons != hw_cons) {
237 		u16 pkt_cons;
238 
239 		pkt_cons = TX_BD(sw_cons);
240 
241 		DP(NETIF_MSG_TX_DONE,
242 		   "queue[%d]: hw_cons %u  sw_cons %u  pkt_cons %u\n",
243 		   txdata->txq_index, hw_cons, sw_cons, pkt_cons);
244 
245 		bd_cons = bnx2x_free_tx_pkt(bp, txdata, pkt_cons,
246 					    &pkts_compl, &bytes_compl);
247 
248 		sw_cons++;
249 	}
250 
251 	netdev_tx_completed_queue(txq, pkts_compl, bytes_compl);
252 
253 	txdata->tx_pkt_cons = sw_cons;
254 	txdata->tx_bd_cons = bd_cons;
255 
256 	/* Need to make the tx_bd_cons update visible to start_xmit()
257 	 * before checking for netif_tx_queue_stopped().  Without the
258 	 * memory barrier, there is a small possibility that
259 	 * start_xmit() will miss it and cause the queue to be stopped
260 	 * forever.
261 	 * On the other hand we need an rmb() here to ensure the proper
262 	 * ordering of bit testing in the following
263 	 * netif_tx_queue_stopped(txq) call.
264 	 */
265 	smp_mb();
266 
267 	if (unlikely(netif_tx_queue_stopped(txq))) {
268 		/* Taking tx_lock() is needed to prevent re-enabling the queue
269 		 * while it's empty. This could have happen if rx_action() gets
270 		 * suspended in bnx2x_tx_int() after the condition before
271 		 * netif_tx_wake_queue(), while tx_action (bnx2x_start_xmit()):
272 		 *
273 		 * stops the queue->sees fresh tx_bd_cons->releases the queue->
274 		 * sends some packets consuming the whole queue again->
275 		 * stops the queue
276 		 */
277 
278 		__netif_tx_lock(txq, smp_processor_id());
279 
280 		if ((netif_tx_queue_stopped(txq)) &&
281 		    (bp->state == BNX2X_STATE_OPEN) &&
282 		    (bnx2x_tx_avail(bp, txdata) >= MAX_DESC_PER_TX_PKT))
283 			netif_tx_wake_queue(txq);
284 
285 		__netif_tx_unlock(txq);
286 	}
287 	return 0;
288 }
289 
290 static inline void bnx2x_update_last_max_sge(struct bnx2x_fastpath *fp,
291 					     u16 idx)
292 {
293 	u16 last_max = fp->last_max_sge;
294 
295 	if (SUB_S16(idx, last_max) > 0)
296 		fp->last_max_sge = idx;
297 }
298 
299 static inline void bnx2x_update_sge_prod(struct bnx2x_fastpath *fp,
300 					 u16 sge_len,
301 					 struct eth_end_agg_rx_cqe *cqe)
302 {
303 	struct bnx2x *bp = fp->bp;
304 	u16 last_max, last_elem, first_elem;
305 	u16 delta = 0;
306 	u16 i;
307 
308 	if (!sge_len)
309 		return;
310 
311 	/* First mark all used pages */
312 	for (i = 0; i < sge_len; i++)
313 		BIT_VEC64_CLEAR_BIT(fp->sge_mask,
314 			RX_SGE(le16_to_cpu(cqe->sgl_or_raw_data.sgl[i])));
315 
316 	DP(NETIF_MSG_RX_STATUS, "fp_cqe->sgl[%d] = %d\n",
317 	   sge_len - 1, le16_to_cpu(cqe->sgl_or_raw_data.sgl[sge_len - 1]));
318 
319 	/* Here we assume that the last SGE index is the biggest */
320 	prefetch((void *)(fp->sge_mask));
321 	bnx2x_update_last_max_sge(fp,
322 		le16_to_cpu(cqe->sgl_or_raw_data.sgl[sge_len - 1]));
323 
324 	last_max = RX_SGE(fp->last_max_sge);
325 	last_elem = last_max >> BIT_VEC64_ELEM_SHIFT;
326 	first_elem = RX_SGE(fp->rx_sge_prod) >> BIT_VEC64_ELEM_SHIFT;
327 
328 	/* If ring is not full */
329 	if (last_elem + 1 != first_elem)
330 		last_elem++;
331 
332 	/* Now update the prod */
333 	for (i = first_elem; i != last_elem; i = NEXT_SGE_MASK_ELEM(i)) {
334 		if (likely(fp->sge_mask[i]))
335 			break;
336 
337 		fp->sge_mask[i] = BIT_VEC64_ELEM_ONE_MASK;
338 		delta += BIT_VEC64_ELEM_SZ;
339 	}
340 
341 	if (delta > 0) {
342 		fp->rx_sge_prod += delta;
343 		/* clear page-end entries */
344 		bnx2x_clear_sge_mask_next_elems(fp);
345 	}
346 
347 	DP(NETIF_MSG_RX_STATUS,
348 	   "fp->last_max_sge = %d  fp->rx_sge_prod = %d\n",
349 	   fp->last_max_sge, fp->rx_sge_prod);
350 }
351 
352 /* Get Toeplitz hash value in the skb using the value from the
353  * CQE (calculated by HW).
354  */
355 static u32 bnx2x_get_rxhash(const struct bnx2x *bp,
356 			    const struct eth_fast_path_rx_cqe *cqe,
357 			    bool *l4_rxhash)
358 {
359 	/* Get Toeplitz hash from CQE */
360 	if ((bp->dev->features & NETIF_F_RXHASH) &&
361 	    (cqe->status_flags & ETH_FAST_PATH_RX_CQE_RSS_HASH_FLG)) {
362 		enum eth_rss_hash_type htype;
363 
364 		htype = cqe->status_flags & ETH_FAST_PATH_RX_CQE_RSS_HASH_TYPE;
365 		*l4_rxhash = (htype == TCP_IPV4_HASH_TYPE) ||
366 			     (htype == TCP_IPV6_HASH_TYPE);
367 		return le32_to_cpu(cqe->rss_hash_result);
368 	}
369 	*l4_rxhash = false;
370 	return 0;
371 }
372 
373 static void bnx2x_tpa_start(struct bnx2x_fastpath *fp, u16 queue,
374 			    u16 cons, u16 prod,
375 			    struct eth_fast_path_rx_cqe *cqe)
376 {
377 	struct bnx2x *bp = fp->bp;
378 	struct sw_rx_bd *cons_rx_buf = &fp->rx_buf_ring[cons];
379 	struct sw_rx_bd *prod_rx_buf = &fp->rx_buf_ring[prod];
380 	struct eth_rx_bd *prod_bd = &fp->rx_desc_ring[prod];
381 	dma_addr_t mapping;
382 	struct bnx2x_agg_info *tpa_info = &fp->tpa_info[queue];
383 	struct sw_rx_bd *first_buf = &tpa_info->first_buf;
384 
385 	/* print error if current state != stop */
386 	if (tpa_info->tpa_state != BNX2X_TPA_STOP)
387 		BNX2X_ERR("start of bin not in stop [%d]\n", queue);
388 
389 	/* Try to map an empty data buffer from the aggregation info  */
390 	mapping = dma_map_single(&bp->pdev->dev,
391 				 first_buf->data + NET_SKB_PAD,
392 				 fp->rx_buf_size, DMA_FROM_DEVICE);
393 	/*
394 	 *  ...if it fails - move the skb from the consumer to the producer
395 	 *  and set the current aggregation state as ERROR to drop it
396 	 *  when TPA_STOP arrives.
397 	 */
398 
399 	if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
400 		/* Move the BD from the consumer to the producer */
401 		bnx2x_reuse_rx_data(fp, cons, prod);
402 		tpa_info->tpa_state = BNX2X_TPA_ERROR;
403 		return;
404 	}
405 
406 	/* move empty data from pool to prod */
407 	prod_rx_buf->data = first_buf->data;
408 	dma_unmap_addr_set(prod_rx_buf, mapping, mapping);
409 	/* point prod_bd to new data */
410 	prod_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
411 	prod_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
412 
413 	/* move partial skb from cons to pool (don't unmap yet) */
414 	*first_buf = *cons_rx_buf;
415 
416 	/* mark bin state as START */
417 	tpa_info->parsing_flags =
418 		le16_to_cpu(cqe->pars_flags.flags);
419 	tpa_info->vlan_tag = le16_to_cpu(cqe->vlan_tag);
420 	tpa_info->tpa_state = BNX2X_TPA_START;
421 	tpa_info->len_on_bd = le16_to_cpu(cqe->len_on_bd);
422 	tpa_info->placement_offset = cqe->placement_offset;
423 	tpa_info->rxhash = bnx2x_get_rxhash(bp, cqe, &tpa_info->l4_rxhash);
424 	if (fp->mode == TPA_MODE_GRO) {
425 		u16 gro_size = le16_to_cpu(cqe->pkt_len_or_gro_seg_len);
426 		tpa_info->full_page = SGE_PAGES / gro_size * gro_size;
427 		tpa_info->gro_size = gro_size;
428 	}
429 
430 #ifdef BNX2X_STOP_ON_ERROR
431 	fp->tpa_queue_used |= (1 << queue);
432 #ifdef _ASM_GENERIC_INT_L64_H
433 	DP(NETIF_MSG_RX_STATUS, "fp->tpa_queue_used = 0x%lx\n",
434 #else
435 	DP(NETIF_MSG_RX_STATUS, "fp->tpa_queue_used = 0x%llx\n",
436 #endif
437 	   fp->tpa_queue_used);
438 #endif
439 }
440 
441 /* Timestamp option length allowed for TPA aggregation:
442  *
443  *		nop nop kind length echo val
444  */
445 #define TPA_TSTAMP_OPT_LEN	12
446 /**
447  * bnx2x_set_gro_params - compute GRO values
448  *
449  * @skb:		packet skb
450  * @parsing_flags:	parsing flags from the START CQE
451  * @len_on_bd:		total length of the first packet for the
452  *			aggregation.
453  * @pkt_len:		length of all segments
454  *
455  * Approximate value of the MSS for this aggregation calculated using
456  * the first packet of it.
457  * Compute number of aggregated segments, and gso_type.
458  */
459 static void bnx2x_set_gro_params(struct sk_buff *skb, u16 parsing_flags,
460 				 u16 len_on_bd, unsigned int pkt_len,
461 				 u16 num_of_coalesced_segs)
462 {
463 	/* TPA aggregation won't have either IP options or TCP options
464 	 * other than timestamp or IPv6 extension headers.
465 	 */
466 	u16 hdrs_len = ETH_HLEN + sizeof(struct tcphdr);
467 
468 	if (GET_FLAG(parsing_flags, PARSING_FLAGS_OVER_ETHERNET_PROTOCOL) ==
469 	    PRS_FLAG_OVERETH_IPV6) {
470 		hdrs_len += sizeof(struct ipv6hdr);
471 		skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
472 	} else {
473 		hdrs_len += sizeof(struct iphdr);
474 		skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
475 	}
476 
477 	/* Check if there was a TCP timestamp, if there is it's will
478 	 * always be 12 bytes length: nop nop kind length echo val.
479 	 *
480 	 * Otherwise FW would close the aggregation.
481 	 */
482 	if (parsing_flags & PARSING_FLAGS_TIME_STAMP_EXIST_FLAG)
483 		hdrs_len += TPA_TSTAMP_OPT_LEN;
484 
485 	skb_shinfo(skb)->gso_size = len_on_bd - hdrs_len;
486 
487 	/* tcp_gro_complete() will copy NAPI_GRO_CB(skb)->count
488 	 * to skb_shinfo(skb)->gso_segs
489 	 */
490 	NAPI_GRO_CB(skb)->count = num_of_coalesced_segs;
491 }
492 
493 static int bnx2x_alloc_rx_sge(struct bnx2x *bp, struct bnx2x_fastpath *fp,
494 			      u16 index, gfp_t gfp_mask)
495 {
496 	struct page *page = alloc_pages(gfp_mask, PAGES_PER_SGE_SHIFT);
497 	struct sw_rx_page *sw_buf = &fp->rx_page_ring[index];
498 	struct eth_rx_sge *sge = &fp->rx_sge_ring[index];
499 	dma_addr_t mapping;
500 
501 	if (unlikely(page == NULL)) {
502 		BNX2X_ERR("Can't alloc sge\n");
503 		return -ENOMEM;
504 	}
505 
506 	mapping = dma_map_page(&bp->pdev->dev, page, 0,
507 			       SGE_PAGES, DMA_FROM_DEVICE);
508 	if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
509 		__free_pages(page, PAGES_PER_SGE_SHIFT);
510 		BNX2X_ERR("Can't map sge\n");
511 		return -ENOMEM;
512 	}
513 
514 	sw_buf->page = page;
515 	dma_unmap_addr_set(sw_buf, mapping, mapping);
516 
517 	sge->addr_hi = cpu_to_le32(U64_HI(mapping));
518 	sge->addr_lo = cpu_to_le32(U64_LO(mapping));
519 
520 	return 0;
521 }
522 
523 static int bnx2x_fill_frag_skb(struct bnx2x *bp, struct bnx2x_fastpath *fp,
524 			       struct bnx2x_agg_info *tpa_info,
525 			       u16 pages,
526 			       struct sk_buff *skb,
527 			       struct eth_end_agg_rx_cqe *cqe,
528 			       u16 cqe_idx)
529 {
530 	struct sw_rx_page *rx_pg, old_rx_pg;
531 	u32 i, frag_len, frag_size;
532 	int err, j, frag_id = 0;
533 	u16 len_on_bd = tpa_info->len_on_bd;
534 	u16 full_page = 0, gro_size = 0;
535 
536 	frag_size = le16_to_cpu(cqe->pkt_len) - len_on_bd;
537 
538 	if (fp->mode == TPA_MODE_GRO) {
539 		gro_size = tpa_info->gro_size;
540 		full_page = tpa_info->full_page;
541 	}
542 
543 	/* This is needed in order to enable forwarding support */
544 	if (frag_size)
545 		bnx2x_set_gro_params(skb, tpa_info->parsing_flags, len_on_bd,
546 				     le16_to_cpu(cqe->pkt_len),
547 				     le16_to_cpu(cqe->num_of_coalesced_segs));
548 
549 #ifdef BNX2X_STOP_ON_ERROR
550 	if (pages > min_t(u32, 8, MAX_SKB_FRAGS) * SGE_PAGES) {
551 		BNX2X_ERR("SGL length is too long: %d. CQE index is %d\n",
552 			  pages, cqe_idx);
553 		BNX2X_ERR("cqe->pkt_len = %d\n", cqe->pkt_len);
554 		bnx2x_panic();
555 		return -EINVAL;
556 	}
557 #endif
558 
559 	/* Run through the SGL and compose the fragmented skb */
560 	for (i = 0, j = 0; i < pages; i += PAGES_PER_SGE, j++) {
561 		u16 sge_idx = RX_SGE(le16_to_cpu(cqe->sgl_or_raw_data.sgl[j]));
562 
563 		/* FW gives the indices of the SGE as if the ring is an array
564 		   (meaning that "next" element will consume 2 indices) */
565 		if (fp->mode == TPA_MODE_GRO)
566 			frag_len = min_t(u32, frag_size, (u32)full_page);
567 		else /* LRO */
568 			frag_len = min_t(u32, frag_size, (u32)SGE_PAGES);
569 
570 		rx_pg = &fp->rx_page_ring[sge_idx];
571 		old_rx_pg = *rx_pg;
572 
573 		/* If we fail to allocate a substitute page, we simply stop
574 		   where we are and drop the whole packet */
575 		err = bnx2x_alloc_rx_sge(bp, fp, sge_idx, GFP_ATOMIC);
576 		if (unlikely(err)) {
577 			bnx2x_fp_qstats(bp, fp)->rx_skb_alloc_failed++;
578 			return err;
579 		}
580 
581 		/* Unmap the page as we're going to pass it to the stack */
582 		dma_unmap_page(&bp->pdev->dev,
583 			       dma_unmap_addr(&old_rx_pg, mapping),
584 			       SGE_PAGES, DMA_FROM_DEVICE);
585 		/* Add one frag and update the appropriate fields in the skb */
586 		if (fp->mode == TPA_MODE_LRO)
587 			skb_fill_page_desc(skb, j, old_rx_pg.page, 0, frag_len);
588 		else { /* GRO */
589 			int rem;
590 			int offset = 0;
591 			for (rem = frag_len; rem > 0; rem -= gro_size) {
592 				int len = rem > gro_size ? gro_size : rem;
593 				skb_fill_page_desc(skb, frag_id++,
594 						   old_rx_pg.page, offset, len);
595 				if (offset)
596 					get_page(old_rx_pg.page);
597 				offset += len;
598 			}
599 		}
600 
601 		skb->data_len += frag_len;
602 		skb->truesize += SGE_PAGES;
603 		skb->len += frag_len;
604 
605 		frag_size -= frag_len;
606 	}
607 
608 	return 0;
609 }
610 
611 static void bnx2x_frag_free(const struct bnx2x_fastpath *fp, void *data)
612 {
613 	if (fp->rx_frag_size)
614 		put_page(virt_to_head_page(data));
615 	else
616 		kfree(data);
617 }
618 
619 static void *bnx2x_frag_alloc(const struct bnx2x_fastpath *fp, gfp_t gfp_mask)
620 {
621 	if (fp->rx_frag_size) {
622 		/* GFP_KERNEL allocations are used only during initialization */
623 		if (unlikely(gfp_mask & __GFP_WAIT))
624 			return (void *)__get_free_page(gfp_mask);
625 
626 		return netdev_alloc_frag(fp->rx_frag_size);
627 	}
628 
629 	return kmalloc(fp->rx_buf_size + NET_SKB_PAD, gfp_mask);
630 }
631 
632 #ifdef CONFIG_INET
633 static void bnx2x_gro_ip_csum(struct bnx2x *bp, struct sk_buff *skb)
634 {
635 	const struct iphdr *iph = ip_hdr(skb);
636 	struct tcphdr *th;
637 
638 	skb_set_transport_header(skb, sizeof(struct iphdr));
639 	th = tcp_hdr(skb);
640 
641 	th->check = ~tcp_v4_check(skb->len - skb_transport_offset(skb),
642 				  iph->saddr, iph->daddr, 0);
643 }
644 
645 static void bnx2x_gro_ipv6_csum(struct bnx2x *bp, struct sk_buff *skb)
646 {
647 	struct ipv6hdr *iph = ipv6_hdr(skb);
648 	struct tcphdr *th;
649 
650 	skb_set_transport_header(skb, sizeof(struct ipv6hdr));
651 	th = tcp_hdr(skb);
652 
653 	th->check = ~tcp_v6_check(skb->len - skb_transport_offset(skb),
654 				  &iph->saddr, &iph->daddr, 0);
655 }
656 
657 static void bnx2x_gro_csum(struct bnx2x *bp, struct sk_buff *skb,
658 			    void (*gro_func)(struct bnx2x*, struct sk_buff*))
659 {
660 	skb_set_network_header(skb, 0);
661 	gro_func(bp, skb);
662 	tcp_gro_complete(skb);
663 }
664 #endif
665 
666 static void bnx2x_gro_receive(struct bnx2x *bp, struct bnx2x_fastpath *fp,
667 			       struct sk_buff *skb)
668 {
669 #ifdef CONFIG_INET
670 	if (skb_shinfo(skb)->gso_size) {
671 		switch (be16_to_cpu(skb->protocol)) {
672 		case ETH_P_IP:
673 			bnx2x_gro_csum(bp, skb, bnx2x_gro_ip_csum);
674 			break;
675 		case ETH_P_IPV6:
676 			bnx2x_gro_csum(bp, skb, bnx2x_gro_ipv6_csum);
677 			break;
678 		default:
679 			BNX2X_ERR("Error: FW GRO supports only IPv4/IPv6, not 0x%04x\n",
680 				  be16_to_cpu(skb->protocol));
681 		}
682 	}
683 #endif
684 	napi_gro_receive(&fp->napi, skb);
685 }
686 
687 static void bnx2x_tpa_stop(struct bnx2x *bp, struct bnx2x_fastpath *fp,
688 			   struct bnx2x_agg_info *tpa_info,
689 			   u16 pages,
690 			   struct eth_end_agg_rx_cqe *cqe,
691 			   u16 cqe_idx)
692 {
693 	struct sw_rx_bd *rx_buf = &tpa_info->first_buf;
694 	u8 pad = tpa_info->placement_offset;
695 	u16 len = tpa_info->len_on_bd;
696 	struct sk_buff *skb = NULL;
697 	u8 *new_data, *data = rx_buf->data;
698 	u8 old_tpa_state = tpa_info->tpa_state;
699 
700 	tpa_info->tpa_state = BNX2X_TPA_STOP;
701 
702 	/* If we there was an error during the handling of the TPA_START -
703 	 * drop this aggregation.
704 	 */
705 	if (old_tpa_state == BNX2X_TPA_ERROR)
706 		goto drop;
707 
708 	/* Try to allocate the new data */
709 	new_data = bnx2x_frag_alloc(fp, GFP_ATOMIC);
710 	/* Unmap skb in the pool anyway, as we are going to change
711 	   pool entry status to BNX2X_TPA_STOP even if new skb allocation
712 	   fails. */
713 	dma_unmap_single(&bp->pdev->dev, dma_unmap_addr(rx_buf, mapping),
714 			 fp->rx_buf_size, DMA_FROM_DEVICE);
715 	if (likely(new_data))
716 		skb = build_skb(data, fp->rx_frag_size);
717 
718 	if (likely(skb)) {
719 #ifdef BNX2X_STOP_ON_ERROR
720 		if (pad + len > fp->rx_buf_size) {
721 			BNX2X_ERR("skb_put is about to fail...  pad %d  len %d  rx_buf_size %d\n",
722 				  pad, len, fp->rx_buf_size);
723 			bnx2x_panic();
724 			return;
725 		}
726 #endif
727 
728 		skb_reserve(skb, pad + NET_SKB_PAD);
729 		skb_put(skb, len);
730 		skb->rxhash = tpa_info->rxhash;
731 		skb->l4_rxhash = tpa_info->l4_rxhash;
732 
733 		skb->protocol = eth_type_trans(skb, bp->dev);
734 		skb->ip_summed = CHECKSUM_UNNECESSARY;
735 
736 		if (!bnx2x_fill_frag_skb(bp, fp, tpa_info, pages,
737 					 skb, cqe, cqe_idx)) {
738 			if (tpa_info->parsing_flags & PARSING_FLAGS_VLAN)
739 				__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), tpa_info->vlan_tag);
740 			bnx2x_gro_receive(bp, fp, skb);
741 		} else {
742 			DP(NETIF_MSG_RX_STATUS,
743 			   "Failed to allocate new pages - dropping packet!\n");
744 			dev_kfree_skb_any(skb);
745 		}
746 
747 		/* put new data in bin */
748 		rx_buf->data = new_data;
749 
750 		return;
751 	}
752 	bnx2x_frag_free(fp, new_data);
753 drop:
754 	/* drop the packet and keep the buffer in the bin */
755 	DP(NETIF_MSG_RX_STATUS,
756 	   "Failed to allocate or map a new skb - dropping packet!\n");
757 	bnx2x_fp_stats(bp, fp)->eth_q_stats.rx_skb_alloc_failed++;
758 }
759 
760 static int bnx2x_alloc_rx_data(struct bnx2x *bp, struct bnx2x_fastpath *fp,
761 			       u16 index, gfp_t gfp_mask)
762 {
763 	u8 *data;
764 	struct sw_rx_bd *rx_buf = &fp->rx_buf_ring[index];
765 	struct eth_rx_bd *rx_bd = &fp->rx_desc_ring[index];
766 	dma_addr_t mapping;
767 
768 	data = bnx2x_frag_alloc(fp, gfp_mask);
769 	if (unlikely(data == NULL))
770 		return -ENOMEM;
771 
772 	mapping = dma_map_single(&bp->pdev->dev, data + NET_SKB_PAD,
773 				 fp->rx_buf_size,
774 				 DMA_FROM_DEVICE);
775 	if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
776 		bnx2x_frag_free(fp, data);
777 		BNX2X_ERR("Can't map rx data\n");
778 		return -ENOMEM;
779 	}
780 
781 	rx_buf->data = data;
782 	dma_unmap_addr_set(rx_buf, mapping, mapping);
783 
784 	rx_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
785 	rx_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
786 
787 	return 0;
788 }
789 
790 static
791 void bnx2x_csum_validate(struct sk_buff *skb, union eth_rx_cqe *cqe,
792 				 struct bnx2x_fastpath *fp,
793 				 struct bnx2x_eth_q_stats *qstats)
794 {
795 	/* Do nothing if no L4 csum validation was done.
796 	 * We do not check whether IP csum was validated. For IPv4 we assume
797 	 * that if the card got as far as validating the L4 csum, it also
798 	 * validated the IP csum. IPv6 has no IP csum.
799 	 */
800 	if (cqe->fast_path_cqe.status_flags &
801 	    ETH_FAST_PATH_RX_CQE_L4_XSUM_NO_VALIDATION_FLG)
802 		return;
803 
804 	/* If L4 validation was done, check if an error was found. */
805 
806 	if (cqe->fast_path_cqe.type_error_flags &
807 	    (ETH_FAST_PATH_RX_CQE_IP_BAD_XSUM_FLG |
808 	     ETH_FAST_PATH_RX_CQE_L4_BAD_XSUM_FLG))
809 		qstats->hw_csum_err++;
810 	else
811 		skb->ip_summed = CHECKSUM_UNNECESSARY;
812 }
813 
814 int bnx2x_rx_int(struct bnx2x_fastpath *fp, int budget)
815 {
816 	struct bnx2x *bp = fp->bp;
817 	u16 bd_cons, bd_prod, bd_prod_fw, comp_ring_cons;
818 	u16 sw_comp_cons, sw_comp_prod;
819 	int rx_pkt = 0;
820 	union eth_rx_cqe *cqe;
821 	struct eth_fast_path_rx_cqe *cqe_fp;
822 
823 #ifdef BNX2X_STOP_ON_ERROR
824 	if (unlikely(bp->panic))
825 		return 0;
826 #endif
827 
828 	bd_cons = fp->rx_bd_cons;
829 	bd_prod = fp->rx_bd_prod;
830 	bd_prod_fw = bd_prod;
831 	sw_comp_cons = fp->rx_comp_cons;
832 	sw_comp_prod = fp->rx_comp_prod;
833 
834 	comp_ring_cons = RCQ_BD(sw_comp_cons);
835 	cqe = &fp->rx_comp_ring[comp_ring_cons];
836 	cqe_fp = &cqe->fast_path_cqe;
837 
838 	DP(NETIF_MSG_RX_STATUS,
839 	   "queue[%d]: sw_comp_cons %u\n", fp->index, sw_comp_cons);
840 
841 	while (BNX2X_IS_CQE_COMPLETED(cqe_fp)) {
842 		struct sw_rx_bd *rx_buf = NULL;
843 		struct sk_buff *skb;
844 		u8 cqe_fp_flags;
845 		enum eth_rx_cqe_type cqe_fp_type;
846 		u16 len, pad, queue;
847 		u8 *data;
848 		bool l4_rxhash;
849 
850 #ifdef BNX2X_STOP_ON_ERROR
851 		if (unlikely(bp->panic))
852 			return 0;
853 #endif
854 
855 		bd_prod = RX_BD(bd_prod);
856 		bd_cons = RX_BD(bd_cons);
857 
858 		cqe_fp_flags = cqe_fp->type_error_flags;
859 		cqe_fp_type = cqe_fp_flags & ETH_FAST_PATH_RX_CQE_TYPE;
860 
861 		DP(NETIF_MSG_RX_STATUS,
862 		   "CQE type %x  err %x  status %x  queue %x  vlan %x  len %u\n",
863 		   CQE_TYPE(cqe_fp_flags),
864 		   cqe_fp_flags, cqe_fp->status_flags,
865 		   le32_to_cpu(cqe_fp->rss_hash_result),
866 		   le16_to_cpu(cqe_fp->vlan_tag),
867 		   le16_to_cpu(cqe_fp->pkt_len_or_gro_seg_len));
868 
869 		/* is this a slowpath msg? */
870 		if (unlikely(CQE_TYPE_SLOW(cqe_fp_type))) {
871 			bnx2x_sp_event(fp, cqe);
872 			goto next_cqe;
873 		}
874 
875 		rx_buf = &fp->rx_buf_ring[bd_cons];
876 		data = rx_buf->data;
877 
878 		if (!CQE_TYPE_FAST(cqe_fp_type)) {
879 			struct bnx2x_agg_info *tpa_info;
880 			u16 frag_size, pages;
881 #ifdef BNX2X_STOP_ON_ERROR
882 			/* sanity check */
883 			if (fp->disable_tpa &&
884 			    (CQE_TYPE_START(cqe_fp_type) ||
885 			     CQE_TYPE_STOP(cqe_fp_type)))
886 				BNX2X_ERR("START/STOP packet while disable_tpa type %x\n",
887 					  CQE_TYPE(cqe_fp_type));
888 #endif
889 
890 			if (CQE_TYPE_START(cqe_fp_type)) {
891 				u16 queue = cqe_fp->queue_index;
892 				DP(NETIF_MSG_RX_STATUS,
893 				   "calling tpa_start on queue %d\n",
894 				   queue);
895 
896 				bnx2x_tpa_start(fp, queue,
897 						bd_cons, bd_prod,
898 						cqe_fp);
899 
900 				goto next_rx;
901 			}
902 			queue = cqe->end_agg_cqe.queue_index;
903 			tpa_info = &fp->tpa_info[queue];
904 			DP(NETIF_MSG_RX_STATUS,
905 			   "calling tpa_stop on queue %d\n",
906 			   queue);
907 
908 			frag_size = le16_to_cpu(cqe->end_agg_cqe.pkt_len) -
909 				    tpa_info->len_on_bd;
910 
911 			if (fp->mode == TPA_MODE_GRO)
912 				pages = (frag_size + tpa_info->full_page - 1) /
913 					 tpa_info->full_page;
914 			else
915 				pages = SGE_PAGE_ALIGN(frag_size) >>
916 					SGE_PAGE_SHIFT;
917 
918 			bnx2x_tpa_stop(bp, fp, tpa_info, pages,
919 				       &cqe->end_agg_cqe, comp_ring_cons);
920 #ifdef BNX2X_STOP_ON_ERROR
921 			if (bp->panic)
922 				return 0;
923 #endif
924 
925 			bnx2x_update_sge_prod(fp, pages, &cqe->end_agg_cqe);
926 			goto next_cqe;
927 		}
928 		/* non TPA */
929 		len = le16_to_cpu(cqe_fp->pkt_len_or_gro_seg_len);
930 		pad = cqe_fp->placement_offset;
931 		dma_sync_single_for_cpu(&bp->pdev->dev,
932 					dma_unmap_addr(rx_buf, mapping),
933 					pad + RX_COPY_THRESH,
934 					DMA_FROM_DEVICE);
935 		pad += NET_SKB_PAD;
936 		prefetch(data + pad); /* speedup eth_type_trans() */
937 		/* is this an error packet? */
938 		if (unlikely(cqe_fp_flags & ETH_RX_ERROR_FALGS)) {
939 			DP(NETIF_MSG_RX_ERR | NETIF_MSG_RX_STATUS,
940 			   "ERROR  flags %x  rx packet %u\n",
941 			   cqe_fp_flags, sw_comp_cons);
942 			bnx2x_fp_qstats(bp, fp)->rx_err_discard_pkt++;
943 			goto reuse_rx;
944 		}
945 
946 		/* Since we don't have a jumbo ring
947 		 * copy small packets if mtu > 1500
948 		 */
949 		if ((bp->dev->mtu > ETH_MAX_PACKET_SIZE) &&
950 		    (len <= RX_COPY_THRESH)) {
951 			skb = netdev_alloc_skb_ip_align(bp->dev, len);
952 			if (skb == NULL) {
953 				DP(NETIF_MSG_RX_ERR | NETIF_MSG_RX_STATUS,
954 				   "ERROR  packet dropped because of alloc failure\n");
955 				bnx2x_fp_qstats(bp, fp)->rx_skb_alloc_failed++;
956 				goto reuse_rx;
957 			}
958 			memcpy(skb->data, data + pad, len);
959 			bnx2x_reuse_rx_data(fp, bd_cons, bd_prod);
960 		} else {
961 			if (likely(bnx2x_alloc_rx_data(bp, fp, bd_prod,
962 						       GFP_ATOMIC) == 0)) {
963 				dma_unmap_single(&bp->pdev->dev,
964 						 dma_unmap_addr(rx_buf, mapping),
965 						 fp->rx_buf_size,
966 						 DMA_FROM_DEVICE);
967 				skb = build_skb(data, fp->rx_frag_size);
968 				if (unlikely(!skb)) {
969 					bnx2x_frag_free(fp, data);
970 					bnx2x_fp_qstats(bp, fp)->
971 							rx_skb_alloc_failed++;
972 					goto next_rx;
973 				}
974 				skb_reserve(skb, pad);
975 			} else {
976 				DP(NETIF_MSG_RX_ERR | NETIF_MSG_RX_STATUS,
977 				   "ERROR  packet dropped because of alloc failure\n");
978 				bnx2x_fp_qstats(bp, fp)->rx_skb_alloc_failed++;
979 reuse_rx:
980 				bnx2x_reuse_rx_data(fp, bd_cons, bd_prod);
981 				goto next_rx;
982 			}
983 		}
984 
985 		skb_put(skb, len);
986 		skb->protocol = eth_type_trans(skb, bp->dev);
987 
988 		/* Set Toeplitz hash for a none-LRO skb */
989 		skb->rxhash = bnx2x_get_rxhash(bp, cqe_fp, &l4_rxhash);
990 		skb->l4_rxhash = l4_rxhash;
991 
992 		skb_checksum_none_assert(skb);
993 
994 		if (bp->dev->features & NETIF_F_RXCSUM)
995 			bnx2x_csum_validate(skb, cqe, fp,
996 					    bnx2x_fp_qstats(bp, fp));
997 
998 		skb_record_rx_queue(skb, fp->rx_queue);
999 
1000 		if (le16_to_cpu(cqe_fp->pars_flags.flags) &
1001 		    PARSING_FLAGS_VLAN)
1002 			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
1003 					       le16_to_cpu(cqe_fp->vlan_tag));
1004 
1005 		skb_mark_napi_id(skb, &fp->napi);
1006 
1007 		if (bnx2x_fp_ll_polling(fp))
1008 			netif_receive_skb(skb);
1009 		else
1010 			napi_gro_receive(&fp->napi, skb);
1011 next_rx:
1012 		rx_buf->data = NULL;
1013 
1014 		bd_cons = NEXT_RX_IDX(bd_cons);
1015 		bd_prod = NEXT_RX_IDX(bd_prod);
1016 		bd_prod_fw = NEXT_RX_IDX(bd_prod_fw);
1017 		rx_pkt++;
1018 next_cqe:
1019 		sw_comp_prod = NEXT_RCQ_IDX(sw_comp_prod);
1020 		sw_comp_cons = NEXT_RCQ_IDX(sw_comp_cons);
1021 
1022 		/* mark CQE as free */
1023 		BNX2X_SEED_CQE(cqe_fp);
1024 
1025 		if (rx_pkt == budget)
1026 			break;
1027 
1028 		comp_ring_cons = RCQ_BD(sw_comp_cons);
1029 		cqe = &fp->rx_comp_ring[comp_ring_cons];
1030 		cqe_fp = &cqe->fast_path_cqe;
1031 	} /* while */
1032 
1033 	fp->rx_bd_cons = bd_cons;
1034 	fp->rx_bd_prod = bd_prod_fw;
1035 	fp->rx_comp_cons = sw_comp_cons;
1036 	fp->rx_comp_prod = sw_comp_prod;
1037 
1038 	/* Update producers */
1039 	bnx2x_update_rx_prod(bp, fp, bd_prod_fw, sw_comp_prod,
1040 			     fp->rx_sge_prod);
1041 
1042 	fp->rx_pkt += rx_pkt;
1043 	fp->rx_calls++;
1044 
1045 	return rx_pkt;
1046 }
1047 
1048 static irqreturn_t bnx2x_msix_fp_int(int irq, void *fp_cookie)
1049 {
1050 	struct bnx2x_fastpath *fp = fp_cookie;
1051 	struct bnx2x *bp = fp->bp;
1052 	u8 cos;
1053 
1054 	DP(NETIF_MSG_INTR,
1055 	   "got an MSI-X interrupt on IDX:SB [fp %d fw_sd %d igusb %d]\n",
1056 	   fp->index, fp->fw_sb_id, fp->igu_sb_id);
1057 
1058 	bnx2x_ack_sb(bp, fp->igu_sb_id, USTORM_ID, 0, IGU_INT_DISABLE, 0);
1059 
1060 #ifdef BNX2X_STOP_ON_ERROR
1061 	if (unlikely(bp->panic))
1062 		return IRQ_HANDLED;
1063 #endif
1064 
1065 	/* Handle Rx and Tx according to MSI-X vector */
1066 	for_each_cos_in_tx_queue(fp, cos)
1067 		prefetch(fp->txdata_ptr[cos]->tx_cons_sb);
1068 
1069 	prefetch(&fp->sb_running_index[SM_RX_ID]);
1070 	napi_schedule(&bnx2x_fp(bp, fp->index, napi));
1071 
1072 	return IRQ_HANDLED;
1073 }
1074 
1075 /* HW Lock for shared dual port PHYs */
1076 void bnx2x_acquire_phy_lock(struct bnx2x *bp)
1077 {
1078 	mutex_lock(&bp->port.phy_mutex);
1079 
1080 	bnx2x_acquire_hw_lock(bp, HW_LOCK_RESOURCE_MDIO);
1081 }
1082 
1083 void bnx2x_release_phy_lock(struct bnx2x *bp)
1084 {
1085 	bnx2x_release_hw_lock(bp, HW_LOCK_RESOURCE_MDIO);
1086 
1087 	mutex_unlock(&bp->port.phy_mutex);
1088 }
1089 
1090 /* calculates MF speed according to current linespeed and MF configuration */
1091 u16 bnx2x_get_mf_speed(struct bnx2x *bp)
1092 {
1093 	u16 line_speed = bp->link_vars.line_speed;
1094 	if (IS_MF(bp)) {
1095 		u16 maxCfg = bnx2x_extract_max_cfg(bp,
1096 						   bp->mf_config[BP_VN(bp)]);
1097 
1098 		/* Calculate the current MAX line speed limit for the MF
1099 		 * devices
1100 		 */
1101 		if (IS_MF_SI(bp))
1102 			line_speed = (line_speed * maxCfg) / 100;
1103 		else { /* SD mode */
1104 			u16 vn_max_rate = maxCfg * 100;
1105 
1106 			if (vn_max_rate < line_speed)
1107 				line_speed = vn_max_rate;
1108 		}
1109 	}
1110 
1111 	return line_speed;
1112 }
1113 
1114 /**
1115  * bnx2x_fill_report_data - fill link report data to report
1116  *
1117  * @bp:		driver handle
1118  * @data:	link state to update
1119  *
1120  * It uses a none-atomic bit operations because is called under the mutex.
1121  */
1122 static void bnx2x_fill_report_data(struct bnx2x *bp,
1123 				   struct bnx2x_link_report_data *data)
1124 {
1125 	u16 line_speed = bnx2x_get_mf_speed(bp);
1126 
1127 	memset(data, 0, sizeof(*data));
1128 
1129 	/* Fill the report data: effective line speed */
1130 	data->line_speed = line_speed;
1131 
1132 	/* Link is down */
1133 	if (!bp->link_vars.link_up || (bp->flags & MF_FUNC_DIS))
1134 		__set_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1135 			  &data->link_report_flags);
1136 
1137 	/* Full DUPLEX */
1138 	if (bp->link_vars.duplex == DUPLEX_FULL)
1139 		__set_bit(BNX2X_LINK_REPORT_FD, &data->link_report_flags);
1140 
1141 	/* Rx Flow Control is ON */
1142 	if (bp->link_vars.flow_ctrl & BNX2X_FLOW_CTRL_RX)
1143 		__set_bit(BNX2X_LINK_REPORT_RX_FC_ON, &data->link_report_flags);
1144 
1145 	/* Tx Flow Control is ON */
1146 	if (bp->link_vars.flow_ctrl & BNX2X_FLOW_CTRL_TX)
1147 		__set_bit(BNX2X_LINK_REPORT_TX_FC_ON, &data->link_report_flags);
1148 }
1149 
1150 /**
1151  * bnx2x_link_report - report link status to OS.
1152  *
1153  * @bp:		driver handle
1154  *
1155  * Calls the __bnx2x_link_report() under the same locking scheme
1156  * as a link/PHY state managing code to ensure a consistent link
1157  * reporting.
1158  */
1159 
1160 void bnx2x_link_report(struct bnx2x *bp)
1161 {
1162 	bnx2x_acquire_phy_lock(bp);
1163 	__bnx2x_link_report(bp);
1164 	bnx2x_release_phy_lock(bp);
1165 }
1166 
1167 /**
1168  * __bnx2x_link_report - report link status to OS.
1169  *
1170  * @bp:		driver handle
1171  *
1172  * None atomic implementation.
1173  * Should be called under the phy_lock.
1174  */
1175 void __bnx2x_link_report(struct bnx2x *bp)
1176 {
1177 	struct bnx2x_link_report_data cur_data;
1178 
1179 	/* reread mf_cfg */
1180 	if (IS_PF(bp) && !CHIP_IS_E1(bp))
1181 		bnx2x_read_mf_cfg(bp);
1182 
1183 	/* Read the current link report info */
1184 	bnx2x_fill_report_data(bp, &cur_data);
1185 
1186 	/* Don't report link down or exactly the same link status twice */
1187 	if (!memcmp(&cur_data, &bp->last_reported_link, sizeof(cur_data)) ||
1188 	    (test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1189 		      &bp->last_reported_link.link_report_flags) &&
1190 	     test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1191 		      &cur_data.link_report_flags)))
1192 		return;
1193 
1194 	bp->link_cnt++;
1195 
1196 	/* We are going to report a new link parameters now -
1197 	 * remember the current data for the next time.
1198 	 */
1199 	memcpy(&bp->last_reported_link, &cur_data, sizeof(cur_data));
1200 
1201 	if (test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1202 		     &cur_data.link_report_flags)) {
1203 		netif_carrier_off(bp->dev);
1204 		netdev_err(bp->dev, "NIC Link is Down\n");
1205 		return;
1206 	} else {
1207 		const char *duplex;
1208 		const char *flow;
1209 
1210 		netif_carrier_on(bp->dev);
1211 
1212 		if (test_and_clear_bit(BNX2X_LINK_REPORT_FD,
1213 				       &cur_data.link_report_flags))
1214 			duplex = "full";
1215 		else
1216 			duplex = "half";
1217 
1218 		/* Handle the FC at the end so that only these flags would be
1219 		 * possibly set. This way we may easily check if there is no FC
1220 		 * enabled.
1221 		 */
1222 		if (cur_data.link_report_flags) {
1223 			if (test_bit(BNX2X_LINK_REPORT_RX_FC_ON,
1224 				     &cur_data.link_report_flags)) {
1225 				if (test_bit(BNX2X_LINK_REPORT_TX_FC_ON,
1226 				     &cur_data.link_report_flags))
1227 					flow = "ON - receive & transmit";
1228 				else
1229 					flow = "ON - receive";
1230 			} else {
1231 				flow = "ON - transmit";
1232 			}
1233 		} else {
1234 			flow = "none";
1235 		}
1236 		netdev_info(bp->dev, "NIC Link is Up, %d Mbps %s duplex, Flow control: %s\n",
1237 			    cur_data.line_speed, duplex, flow);
1238 	}
1239 }
1240 
1241 static void bnx2x_set_next_page_sgl(struct bnx2x_fastpath *fp)
1242 {
1243 	int i;
1244 
1245 	for (i = 1; i <= NUM_RX_SGE_PAGES; i++) {
1246 		struct eth_rx_sge *sge;
1247 
1248 		sge = &fp->rx_sge_ring[RX_SGE_CNT * i - 2];
1249 		sge->addr_hi =
1250 			cpu_to_le32(U64_HI(fp->rx_sge_mapping +
1251 			BCM_PAGE_SIZE*(i % NUM_RX_SGE_PAGES)));
1252 
1253 		sge->addr_lo =
1254 			cpu_to_le32(U64_LO(fp->rx_sge_mapping +
1255 			BCM_PAGE_SIZE*(i % NUM_RX_SGE_PAGES)));
1256 	}
1257 }
1258 
1259 static void bnx2x_free_tpa_pool(struct bnx2x *bp,
1260 				struct bnx2x_fastpath *fp, int last)
1261 {
1262 	int i;
1263 
1264 	for (i = 0; i < last; i++) {
1265 		struct bnx2x_agg_info *tpa_info = &fp->tpa_info[i];
1266 		struct sw_rx_bd *first_buf = &tpa_info->first_buf;
1267 		u8 *data = first_buf->data;
1268 
1269 		if (data == NULL) {
1270 			DP(NETIF_MSG_IFDOWN, "tpa bin %d empty on free\n", i);
1271 			continue;
1272 		}
1273 		if (tpa_info->tpa_state == BNX2X_TPA_START)
1274 			dma_unmap_single(&bp->pdev->dev,
1275 					 dma_unmap_addr(first_buf, mapping),
1276 					 fp->rx_buf_size, DMA_FROM_DEVICE);
1277 		bnx2x_frag_free(fp, data);
1278 		first_buf->data = NULL;
1279 	}
1280 }
1281 
1282 void bnx2x_init_rx_rings_cnic(struct bnx2x *bp)
1283 {
1284 	int j;
1285 
1286 	for_each_rx_queue_cnic(bp, j) {
1287 		struct bnx2x_fastpath *fp = &bp->fp[j];
1288 
1289 		fp->rx_bd_cons = 0;
1290 
1291 		/* Activate BD ring */
1292 		/* Warning!
1293 		 * this will generate an interrupt (to the TSTORM)
1294 		 * must only be done after chip is initialized
1295 		 */
1296 		bnx2x_update_rx_prod(bp, fp, fp->rx_bd_prod, fp->rx_comp_prod,
1297 				     fp->rx_sge_prod);
1298 	}
1299 }
1300 
1301 void bnx2x_init_rx_rings(struct bnx2x *bp)
1302 {
1303 	int func = BP_FUNC(bp);
1304 	u16 ring_prod;
1305 	int i, j;
1306 
1307 	/* Allocate TPA resources */
1308 	for_each_eth_queue(bp, j) {
1309 		struct bnx2x_fastpath *fp = &bp->fp[j];
1310 
1311 		DP(NETIF_MSG_IFUP,
1312 		   "mtu %d  rx_buf_size %d\n", bp->dev->mtu, fp->rx_buf_size);
1313 
1314 		if (!fp->disable_tpa) {
1315 			/* Fill the per-aggregation pool */
1316 			for (i = 0; i < MAX_AGG_QS(bp); i++) {
1317 				struct bnx2x_agg_info *tpa_info =
1318 					&fp->tpa_info[i];
1319 				struct sw_rx_bd *first_buf =
1320 					&tpa_info->first_buf;
1321 
1322 				first_buf->data =
1323 					bnx2x_frag_alloc(fp, GFP_KERNEL);
1324 				if (!first_buf->data) {
1325 					BNX2X_ERR("Failed to allocate TPA skb pool for queue[%d] - disabling TPA on this queue!\n",
1326 						  j);
1327 					bnx2x_free_tpa_pool(bp, fp, i);
1328 					fp->disable_tpa = 1;
1329 					break;
1330 				}
1331 				dma_unmap_addr_set(first_buf, mapping, 0);
1332 				tpa_info->tpa_state = BNX2X_TPA_STOP;
1333 			}
1334 
1335 			/* "next page" elements initialization */
1336 			bnx2x_set_next_page_sgl(fp);
1337 
1338 			/* set SGEs bit mask */
1339 			bnx2x_init_sge_ring_bit_mask(fp);
1340 
1341 			/* Allocate SGEs and initialize the ring elements */
1342 			for (i = 0, ring_prod = 0;
1343 			     i < MAX_RX_SGE_CNT*NUM_RX_SGE_PAGES; i++) {
1344 
1345 				if (bnx2x_alloc_rx_sge(bp, fp, ring_prod,
1346 						       GFP_KERNEL) < 0) {
1347 					BNX2X_ERR("was only able to allocate %d rx sges\n",
1348 						  i);
1349 					BNX2X_ERR("disabling TPA for queue[%d]\n",
1350 						  j);
1351 					/* Cleanup already allocated elements */
1352 					bnx2x_free_rx_sge_range(bp, fp,
1353 								ring_prod);
1354 					bnx2x_free_tpa_pool(bp, fp,
1355 							    MAX_AGG_QS(bp));
1356 					fp->disable_tpa = 1;
1357 					ring_prod = 0;
1358 					break;
1359 				}
1360 				ring_prod = NEXT_SGE_IDX(ring_prod);
1361 			}
1362 
1363 			fp->rx_sge_prod = ring_prod;
1364 		}
1365 	}
1366 
1367 	for_each_eth_queue(bp, j) {
1368 		struct bnx2x_fastpath *fp = &bp->fp[j];
1369 
1370 		fp->rx_bd_cons = 0;
1371 
1372 		/* Activate BD ring */
1373 		/* Warning!
1374 		 * this will generate an interrupt (to the TSTORM)
1375 		 * must only be done after chip is initialized
1376 		 */
1377 		bnx2x_update_rx_prod(bp, fp, fp->rx_bd_prod, fp->rx_comp_prod,
1378 				     fp->rx_sge_prod);
1379 
1380 		if (j != 0)
1381 			continue;
1382 
1383 		if (CHIP_IS_E1(bp)) {
1384 			REG_WR(bp, BAR_USTRORM_INTMEM +
1385 			       USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func),
1386 			       U64_LO(fp->rx_comp_mapping));
1387 			REG_WR(bp, BAR_USTRORM_INTMEM +
1388 			       USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func) + 4,
1389 			       U64_HI(fp->rx_comp_mapping));
1390 		}
1391 	}
1392 }
1393 
1394 static void bnx2x_free_tx_skbs_queue(struct bnx2x_fastpath *fp)
1395 {
1396 	u8 cos;
1397 	struct bnx2x *bp = fp->bp;
1398 
1399 	for_each_cos_in_tx_queue(fp, cos) {
1400 		struct bnx2x_fp_txdata *txdata = fp->txdata_ptr[cos];
1401 		unsigned pkts_compl = 0, bytes_compl = 0;
1402 
1403 		u16 sw_prod = txdata->tx_pkt_prod;
1404 		u16 sw_cons = txdata->tx_pkt_cons;
1405 
1406 		while (sw_cons != sw_prod) {
1407 			bnx2x_free_tx_pkt(bp, txdata, TX_BD(sw_cons),
1408 					  &pkts_compl, &bytes_compl);
1409 			sw_cons++;
1410 		}
1411 
1412 		netdev_tx_reset_queue(
1413 			netdev_get_tx_queue(bp->dev,
1414 					    txdata->txq_index));
1415 	}
1416 }
1417 
1418 static void bnx2x_free_tx_skbs_cnic(struct bnx2x *bp)
1419 {
1420 	int i;
1421 
1422 	for_each_tx_queue_cnic(bp, i) {
1423 		bnx2x_free_tx_skbs_queue(&bp->fp[i]);
1424 	}
1425 }
1426 
1427 static void bnx2x_free_tx_skbs(struct bnx2x *bp)
1428 {
1429 	int i;
1430 
1431 	for_each_eth_queue(bp, i) {
1432 		bnx2x_free_tx_skbs_queue(&bp->fp[i]);
1433 	}
1434 }
1435 
1436 static void bnx2x_free_rx_bds(struct bnx2x_fastpath *fp)
1437 {
1438 	struct bnx2x *bp = fp->bp;
1439 	int i;
1440 
1441 	/* ring wasn't allocated */
1442 	if (fp->rx_buf_ring == NULL)
1443 		return;
1444 
1445 	for (i = 0; i < NUM_RX_BD; i++) {
1446 		struct sw_rx_bd *rx_buf = &fp->rx_buf_ring[i];
1447 		u8 *data = rx_buf->data;
1448 
1449 		if (data == NULL)
1450 			continue;
1451 		dma_unmap_single(&bp->pdev->dev,
1452 				 dma_unmap_addr(rx_buf, mapping),
1453 				 fp->rx_buf_size, DMA_FROM_DEVICE);
1454 
1455 		rx_buf->data = NULL;
1456 		bnx2x_frag_free(fp, data);
1457 	}
1458 }
1459 
1460 static void bnx2x_free_rx_skbs_cnic(struct bnx2x *bp)
1461 {
1462 	int j;
1463 
1464 	for_each_rx_queue_cnic(bp, j) {
1465 		bnx2x_free_rx_bds(&bp->fp[j]);
1466 	}
1467 }
1468 
1469 static void bnx2x_free_rx_skbs(struct bnx2x *bp)
1470 {
1471 	int j;
1472 
1473 	for_each_eth_queue(bp, j) {
1474 		struct bnx2x_fastpath *fp = &bp->fp[j];
1475 
1476 		bnx2x_free_rx_bds(fp);
1477 
1478 		if (!fp->disable_tpa)
1479 			bnx2x_free_tpa_pool(bp, fp, MAX_AGG_QS(bp));
1480 	}
1481 }
1482 
1483 void bnx2x_free_skbs_cnic(struct bnx2x *bp)
1484 {
1485 	bnx2x_free_tx_skbs_cnic(bp);
1486 	bnx2x_free_rx_skbs_cnic(bp);
1487 }
1488 
1489 void bnx2x_free_skbs(struct bnx2x *bp)
1490 {
1491 	bnx2x_free_tx_skbs(bp);
1492 	bnx2x_free_rx_skbs(bp);
1493 }
1494 
1495 void bnx2x_update_max_mf_config(struct bnx2x *bp, u32 value)
1496 {
1497 	/* load old values */
1498 	u32 mf_cfg = bp->mf_config[BP_VN(bp)];
1499 
1500 	if (value != bnx2x_extract_max_cfg(bp, mf_cfg)) {
1501 		/* leave all but MAX value */
1502 		mf_cfg &= ~FUNC_MF_CFG_MAX_BW_MASK;
1503 
1504 		/* set new MAX value */
1505 		mf_cfg |= (value << FUNC_MF_CFG_MAX_BW_SHIFT)
1506 				& FUNC_MF_CFG_MAX_BW_MASK;
1507 
1508 		bnx2x_fw_command(bp, DRV_MSG_CODE_SET_MF_BW, mf_cfg);
1509 	}
1510 }
1511 
1512 /**
1513  * bnx2x_free_msix_irqs - free previously requested MSI-X IRQ vectors
1514  *
1515  * @bp:		driver handle
1516  * @nvecs:	number of vectors to be released
1517  */
1518 static void bnx2x_free_msix_irqs(struct bnx2x *bp, int nvecs)
1519 {
1520 	int i, offset = 0;
1521 
1522 	if (nvecs == offset)
1523 		return;
1524 
1525 	/* VFs don't have a default SB */
1526 	if (IS_PF(bp)) {
1527 		free_irq(bp->msix_table[offset].vector, bp->dev);
1528 		DP(NETIF_MSG_IFDOWN, "released sp irq (%d)\n",
1529 		   bp->msix_table[offset].vector);
1530 		offset++;
1531 	}
1532 
1533 	if (CNIC_SUPPORT(bp)) {
1534 		if (nvecs == offset)
1535 			return;
1536 		offset++;
1537 	}
1538 
1539 	for_each_eth_queue(bp, i) {
1540 		if (nvecs == offset)
1541 			return;
1542 		DP(NETIF_MSG_IFDOWN, "about to release fp #%d->%d irq\n",
1543 		   i, bp->msix_table[offset].vector);
1544 
1545 		free_irq(bp->msix_table[offset++].vector, &bp->fp[i]);
1546 	}
1547 }
1548 
1549 void bnx2x_free_irq(struct bnx2x *bp)
1550 {
1551 	if (bp->flags & USING_MSIX_FLAG &&
1552 	    !(bp->flags & USING_SINGLE_MSIX_FLAG)) {
1553 		int nvecs = BNX2X_NUM_ETH_QUEUES(bp) + CNIC_SUPPORT(bp);
1554 
1555 		/* vfs don't have a default status block */
1556 		if (IS_PF(bp))
1557 			nvecs++;
1558 
1559 		bnx2x_free_msix_irqs(bp, nvecs);
1560 	} else {
1561 		free_irq(bp->dev->irq, bp->dev);
1562 	}
1563 }
1564 
1565 int bnx2x_enable_msix(struct bnx2x *bp)
1566 {
1567 	int msix_vec = 0, i, rc;
1568 
1569 	/* VFs don't have a default status block */
1570 	if (IS_PF(bp)) {
1571 		bp->msix_table[msix_vec].entry = msix_vec;
1572 		BNX2X_DEV_INFO("msix_table[0].entry = %d (slowpath)\n",
1573 			       bp->msix_table[0].entry);
1574 		msix_vec++;
1575 	}
1576 
1577 	/* Cnic requires an msix vector for itself */
1578 	if (CNIC_SUPPORT(bp)) {
1579 		bp->msix_table[msix_vec].entry = msix_vec;
1580 		BNX2X_DEV_INFO("msix_table[%d].entry = %d (CNIC)\n",
1581 			       msix_vec, bp->msix_table[msix_vec].entry);
1582 		msix_vec++;
1583 	}
1584 
1585 	/* We need separate vectors for ETH queues only (not FCoE) */
1586 	for_each_eth_queue(bp, i) {
1587 		bp->msix_table[msix_vec].entry = msix_vec;
1588 		BNX2X_DEV_INFO("msix_table[%d].entry = %d (fastpath #%u)\n",
1589 			       msix_vec, msix_vec, i);
1590 		msix_vec++;
1591 	}
1592 
1593 	DP(BNX2X_MSG_SP, "about to request enable msix with %d vectors\n",
1594 	   msix_vec);
1595 
1596 	rc = pci_enable_msix(bp->pdev, &bp->msix_table[0], msix_vec);
1597 
1598 	/*
1599 	 * reconfigure number of tx/rx queues according to available
1600 	 * MSI-X vectors
1601 	 */
1602 	if (rc >= BNX2X_MIN_MSIX_VEC_CNT(bp)) {
1603 		/* how less vectors we will have? */
1604 		int diff = msix_vec - rc;
1605 
1606 		BNX2X_DEV_INFO("Trying to use less MSI-X vectors: %d\n", rc);
1607 
1608 		rc = pci_enable_msix(bp->pdev, &bp->msix_table[0], rc);
1609 
1610 		if (rc) {
1611 			BNX2X_DEV_INFO("MSI-X is not attainable rc %d\n", rc);
1612 			goto no_msix;
1613 		}
1614 		/*
1615 		 * decrease number of queues by number of unallocated entries
1616 		 */
1617 		bp->num_ethernet_queues -= diff;
1618 		bp->num_queues = bp->num_ethernet_queues + bp->num_cnic_queues;
1619 
1620 		BNX2X_DEV_INFO("New queue configuration set: %d\n",
1621 			       bp->num_queues);
1622 	} else if (rc > 0) {
1623 		/* Get by with single vector */
1624 		rc = pci_enable_msix(bp->pdev, &bp->msix_table[0], 1);
1625 		if (rc) {
1626 			BNX2X_DEV_INFO("Single MSI-X is not attainable rc %d\n",
1627 				       rc);
1628 			goto no_msix;
1629 		}
1630 
1631 		BNX2X_DEV_INFO("Using single MSI-X vector\n");
1632 		bp->flags |= USING_SINGLE_MSIX_FLAG;
1633 
1634 		BNX2X_DEV_INFO("set number of queues to 1\n");
1635 		bp->num_ethernet_queues = 1;
1636 		bp->num_queues = bp->num_ethernet_queues + bp->num_cnic_queues;
1637 	} else if (rc < 0) {
1638 		BNX2X_DEV_INFO("MSI-X is not attainable  rc %d\n", rc);
1639 		goto no_msix;
1640 	}
1641 
1642 	bp->flags |= USING_MSIX_FLAG;
1643 
1644 	return 0;
1645 
1646 no_msix:
1647 	/* fall to INTx if not enough memory */
1648 	if (rc == -ENOMEM)
1649 		bp->flags |= DISABLE_MSI_FLAG;
1650 
1651 	return rc;
1652 }
1653 
1654 static int bnx2x_req_msix_irqs(struct bnx2x *bp)
1655 {
1656 	int i, rc, offset = 0;
1657 
1658 	/* no default status block for vf */
1659 	if (IS_PF(bp)) {
1660 		rc = request_irq(bp->msix_table[offset++].vector,
1661 				 bnx2x_msix_sp_int, 0,
1662 				 bp->dev->name, bp->dev);
1663 		if (rc) {
1664 			BNX2X_ERR("request sp irq failed\n");
1665 			return -EBUSY;
1666 		}
1667 	}
1668 
1669 	if (CNIC_SUPPORT(bp))
1670 		offset++;
1671 
1672 	for_each_eth_queue(bp, i) {
1673 		struct bnx2x_fastpath *fp = &bp->fp[i];
1674 		snprintf(fp->name, sizeof(fp->name), "%s-fp-%d",
1675 			 bp->dev->name, i);
1676 
1677 		rc = request_irq(bp->msix_table[offset].vector,
1678 				 bnx2x_msix_fp_int, 0, fp->name, fp);
1679 		if (rc) {
1680 			BNX2X_ERR("request fp #%d irq (%d) failed  rc %d\n", i,
1681 			      bp->msix_table[offset].vector, rc);
1682 			bnx2x_free_msix_irqs(bp, offset);
1683 			return -EBUSY;
1684 		}
1685 
1686 		offset++;
1687 	}
1688 
1689 	i = BNX2X_NUM_ETH_QUEUES(bp);
1690 	if (IS_PF(bp)) {
1691 		offset = 1 + CNIC_SUPPORT(bp);
1692 		netdev_info(bp->dev,
1693 			    "using MSI-X  IRQs: sp %d  fp[%d] %d ... fp[%d] %d\n",
1694 			    bp->msix_table[0].vector,
1695 			    0, bp->msix_table[offset].vector,
1696 			    i - 1, bp->msix_table[offset + i - 1].vector);
1697 	} else {
1698 		offset = CNIC_SUPPORT(bp);
1699 		netdev_info(bp->dev,
1700 			    "using MSI-X  IRQs: fp[%d] %d ... fp[%d] %d\n",
1701 			    0, bp->msix_table[offset].vector,
1702 			    i - 1, bp->msix_table[offset + i - 1].vector);
1703 	}
1704 	return 0;
1705 }
1706 
1707 int bnx2x_enable_msi(struct bnx2x *bp)
1708 {
1709 	int rc;
1710 
1711 	rc = pci_enable_msi(bp->pdev);
1712 	if (rc) {
1713 		BNX2X_DEV_INFO("MSI is not attainable\n");
1714 		return -1;
1715 	}
1716 	bp->flags |= USING_MSI_FLAG;
1717 
1718 	return 0;
1719 }
1720 
1721 static int bnx2x_req_irq(struct bnx2x *bp)
1722 {
1723 	unsigned long flags;
1724 	unsigned int irq;
1725 
1726 	if (bp->flags & (USING_MSI_FLAG | USING_MSIX_FLAG))
1727 		flags = 0;
1728 	else
1729 		flags = IRQF_SHARED;
1730 
1731 	if (bp->flags & USING_MSIX_FLAG)
1732 		irq = bp->msix_table[0].vector;
1733 	else
1734 		irq = bp->pdev->irq;
1735 
1736 	return request_irq(irq, bnx2x_interrupt, flags, bp->dev->name, bp->dev);
1737 }
1738 
1739 static int bnx2x_setup_irqs(struct bnx2x *bp)
1740 {
1741 	int rc = 0;
1742 	if (bp->flags & USING_MSIX_FLAG &&
1743 	    !(bp->flags & USING_SINGLE_MSIX_FLAG)) {
1744 		rc = bnx2x_req_msix_irqs(bp);
1745 		if (rc)
1746 			return rc;
1747 	} else {
1748 		rc = bnx2x_req_irq(bp);
1749 		if (rc) {
1750 			BNX2X_ERR("IRQ request failed  rc %d, aborting\n", rc);
1751 			return rc;
1752 		}
1753 		if (bp->flags & USING_MSI_FLAG) {
1754 			bp->dev->irq = bp->pdev->irq;
1755 			netdev_info(bp->dev, "using MSI IRQ %d\n",
1756 				    bp->dev->irq);
1757 		}
1758 		if (bp->flags & USING_MSIX_FLAG) {
1759 			bp->dev->irq = bp->msix_table[0].vector;
1760 			netdev_info(bp->dev, "using MSIX IRQ %d\n",
1761 				    bp->dev->irq);
1762 		}
1763 	}
1764 
1765 	return 0;
1766 }
1767 
1768 static void bnx2x_napi_enable_cnic(struct bnx2x *bp)
1769 {
1770 	int i;
1771 
1772 	for_each_rx_queue_cnic(bp, i) {
1773 		bnx2x_fp_init_lock(&bp->fp[i]);
1774 		napi_enable(&bnx2x_fp(bp, i, napi));
1775 	}
1776 }
1777 
1778 static void bnx2x_napi_enable(struct bnx2x *bp)
1779 {
1780 	int i;
1781 
1782 	for_each_eth_queue(bp, i) {
1783 		bnx2x_fp_init_lock(&bp->fp[i]);
1784 		napi_enable(&bnx2x_fp(bp, i, napi));
1785 	}
1786 }
1787 
1788 static void bnx2x_napi_disable_cnic(struct bnx2x *bp)
1789 {
1790 	int i;
1791 
1792 	local_bh_disable();
1793 	for_each_rx_queue_cnic(bp, i) {
1794 		napi_disable(&bnx2x_fp(bp, i, napi));
1795 		while (!bnx2x_fp_lock_napi(&bp->fp[i]))
1796 			mdelay(1);
1797 	}
1798 	local_bh_enable();
1799 }
1800 
1801 static void bnx2x_napi_disable(struct bnx2x *bp)
1802 {
1803 	int i;
1804 
1805 	local_bh_disable();
1806 	for_each_eth_queue(bp, i) {
1807 		napi_disable(&bnx2x_fp(bp, i, napi));
1808 		while (!bnx2x_fp_lock_napi(&bp->fp[i]))
1809 			mdelay(1);
1810 	}
1811 	local_bh_enable();
1812 }
1813 
1814 void bnx2x_netif_start(struct bnx2x *bp)
1815 {
1816 	if (netif_running(bp->dev)) {
1817 		bnx2x_napi_enable(bp);
1818 		if (CNIC_LOADED(bp))
1819 			bnx2x_napi_enable_cnic(bp);
1820 		bnx2x_int_enable(bp);
1821 		if (bp->state == BNX2X_STATE_OPEN)
1822 			netif_tx_wake_all_queues(bp->dev);
1823 	}
1824 }
1825 
1826 void bnx2x_netif_stop(struct bnx2x *bp, int disable_hw)
1827 {
1828 	bnx2x_int_disable_sync(bp, disable_hw);
1829 	bnx2x_napi_disable(bp);
1830 	if (CNIC_LOADED(bp))
1831 		bnx2x_napi_disable_cnic(bp);
1832 }
1833 
1834 u16 bnx2x_select_queue(struct net_device *dev, struct sk_buff *skb)
1835 {
1836 	struct bnx2x *bp = netdev_priv(dev);
1837 
1838 	if (CNIC_LOADED(bp) && !NO_FCOE(bp)) {
1839 		struct ethhdr *hdr = (struct ethhdr *)skb->data;
1840 		u16 ether_type = ntohs(hdr->h_proto);
1841 
1842 		/* Skip VLAN tag if present */
1843 		if (ether_type == ETH_P_8021Q) {
1844 			struct vlan_ethhdr *vhdr =
1845 				(struct vlan_ethhdr *)skb->data;
1846 
1847 			ether_type = ntohs(vhdr->h_vlan_encapsulated_proto);
1848 		}
1849 
1850 		/* If ethertype is FCoE or FIP - use FCoE ring */
1851 		if ((ether_type == ETH_P_FCOE) || (ether_type == ETH_P_FIP))
1852 			return bnx2x_fcoe_tx(bp, txq_index);
1853 	}
1854 
1855 	/* select a non-FCoE queue */
1856 	return __netdev_pick_tx(dev, skb) % BNX2X_NUM_ETH_QUEUES(bp);
1857 }
1858 
1859 void bnx2x_set_num_queues(struct bnx2x *bp)
1860 {
1861 	/* RSS queues */
1862 	bp->num_ethernet_queues = bnx2x_calc_num_queues(bp);
1863 
1864 	/* override in STORAGE SD modes */
1865 	if (IS_MF_STORAGE_SD(bp) || IS_MF_FCOE_AFEX(bp))
1866 		bp->num_ethernet_queues = 1;
1867 
1868 	/* Add special queues */
1869 	bp->num_cnic_queues = CNIC_SUPPORT(bp); /* For FCOE */
1870 	bp->num_queues = bp->num_ethernet_queues + bp->num_cnic_queues;
1871 
1872 	BNX2X_DEV_INFO("set number of queues to %d\n", bp->num_queues);
1873 }
1874 
1875 /**
1876  * bnx2x_set_real_num_queues - configure netdev->real_num_[tx,rx]_queues
1877  *
1878  * @bp:		Driver handle
1879  *
1880  * We currently support for at most 16 Tx queues for each CoS thus we will
1881  * allocate a multiple of 16 for ETH L2 rings according to the value of the
1882  * bp->max_cos.
1883  *
1884  * If there is an FCoE L2 queue the appropriate Tx queue will have the next
1885  * index after all ETH L2 indices.
1886  *
1887  * If the actual number of Tx queues (for each CoS) is less than 16 then there
1888  * will be the holes at the end of each group of 16 ETh L2 indices (0..15,
1889  * 16..31,...) with indices that are not coupled with any real Tx queue.
1890  *
1891  * The proper configuration of skb->queue_mapping is handled by
1892  * bnx2x_select_queue() and __skb_tx_hash().
1893  *
1894  * bnx2x_setup_tc() takes care of the proper TC mappings so that __skb_tx_hash()
1895  * will return a proper Tx index if TC is enabled (netdev->num_tc > 0).
1896  */
1897 static int bnx2x_set_real_num_queues(struct bnx2x *bp, int include_cnic)
1898 {
1899 	int rc, tx, rx;
1900 
1901 	tx = BNX2X_NUM_ETH_QUEUES(bp) * bp->max_cos;
1902 	rx = BNX2X_NUM_ETH_QUEUES(bp);
1903 
1904 /* account for fcoe queue */
1905 	if (include_cnic && !NO_FCOE(bp)) {
1906 		rx++;
1907 		tx++;
1908 	}
1909 
1910 	rc = netif_set_real_num_tx_queues(bp->dev, tx);
1911 	if (rc) {
1912 		BNX2X_ERR("Failed to set real number of Tx queues: %d\n", rc);
1913 		return rc;
1914 	}
1915 	rc = netif_set_real_num_rx_queues(bp->dev, rx);
1916 	if (rc) {
1917 		BNX2X_ERR("Failed to set real number of Rx queues: %d\n", rc);
1918 		return rc;
1919 	}
1920 
1921 	DP(NETIF_MSG_IFUP, "Setting real num queues to (tx, rx) (%d, %d)\n",
1922 			  tx, rx);
1923 
1924 	return rc;
1925 }
1926 
1927 static void bnx2x_set_rx_buf_size(struct bnx2x *bp)
1928 {
1929 	int i;
1930 
1931 	for_each_queue(bp, i) {
1932 		struct bnx2x_fastpath *fp = &bp->fp[i];
1933 		u32 mtu;
1934 
1935 		/* Always use a mini-jumbo MTU for the FCoE L2 ring */
1936 		if (IS_FCOE_IDX(i))
1937 			/*
1938 			 * Although there are no IP frames expected to arrive to
1939 			 * this ring we still want to add an
1940 			 * IP_HEADER_ALIGNMENT_PADDING to prevent a buffer
1941 			 * overrun attack.
1942 			 */
1943 			mtu = BNX2X_FCOE_MINI_JUMBO_MTU;
1944 		else
1945 			mtu = bp->dev->mtu;
1946 		fp->rx_buf_size = BNX2X_FW_RX_ALIGN_START +
1947 				  IP_HEADER_ALIGNMENT_PADDING +
1948 				  ETH_OVREHEAD +
1949 				  mtu +
1950 				  BNX2X_FW_RX_ALIGN_END;
1951 		/* Note : rx_buf_size doesn't take into account NET_SKB_PAD */
1952 		if (fp->rx_buf_size + NET_SKB_PAD <= PAGE_SIZE)
1953 			fp->rx_frag_size = fp->rx_buf_size + NET_SKB_PAD;
1954 		else
1955 			fp->rx_frag_size = 0;
1956 	}
1957 }
1958 
1959 static int bnx2x_init_rss(struct bnx2x *bp)
1960 {
1961 	int i;
1962 	u8 num_eth_queues = BNX2X_NUM_ETH_QUEUES(bp);
1963 
1964 	/* Prepare the initial contents for the indirection table if RSS is
1965 	 * enabled
1966 	 */
1967 	for (i = 0; i < sizeof(bp->rss_conf_obj.ind_table); i++)
1968 		bp->rss_conf_obj.ind_table[i] =
1969 			bp->fp->cl_id +
1970 			ethtool_rxfh_indir_default(i, num_eth_queues);
1971 
1972 	/*
1973 	 * For 57710 and 57711 SEARCHER configuration (rss_keys) is
1974 	 * per-port, so if explicit configuration is needed , do it only
1975 	 * for a PMF.
1976 	 *
1977 	 * For 57712 and newer on the other hand it's a per-function
1978 	 * configuration.
1979 	 */
1980 	return bnx2x_config_rss_eth(bp, bp->port.pmf || !CHIP_IS_E1x(bp));
1981 }
1982 
1983 int bnx2x_rss(struct bnx2x *bp, struct bnx2x_rss_config_obj *rss_obj,
1984 	      bool config_hash, bool enable)
1985 {
1986 	struct bnx2x_config_rss_params params = {NULL};
1987 
1988 	/* Although RSS is meaningless when there is a single HW queue we
1989 	 * still need it enabled in order to have HW Rx hash generated.
1990 	 *
1991 	 * if (!is_eth_multi(bp))
1992 	 *      bp->multi_mode = ETH_RSS_MODE_DISABLED;
1993 	 */
1994 
1995 	params.rss_obj = rss_obj;
1996 
1997 	__set_bit(RAMROD_COMP_WAIT, &params.ramrod_flags);
1998 
1999 	if (enable) {
2000 		__set_bit(BNX2X_RSS_MODE_REGULAR, &params.rss_flags);
2001 
2002 		/* RSS configuration */
2003 		__set_bit(BNX2X_RSS_IPV4, &params.rss_flags);
2004 		__set_bit(BNX2X_RSS_IPV4_TCP, &params.rss_flags);
2005 		__set_bit(BNX2X_RSS_IPV6, &params.rss_flags);
2006 		__set_bit(BNX2X_RSS_IPV6_TCP, &params.rss_flags);
2007 		if (rss_obj->udp_rss_v4)
2008 			__set_bit(BNX2X_RSS_IPV4_UDP, &params.rss_flags);
2009 		if (rss_obj->udp_rss_v6)
2010 			__set_bit(BNX2X_RSS_IPV6_UDP, &params.rss_flags);
2011 	} else {
2012 		__set_bit(BNX2X_RSS_MODE_DISABLED, &params.rss_flags);
2013 	}
2014 
2015 	/* Hash bits */
2016 	params.rss_result_mask = MULTI_MASK;
2017 
2018 	memcpy(params.ind_table, rss_obj->ind_table, sizeof(params.ind_table));
2019 
2020 	if (config_hash) {
2021 		/* RSS keys */
2022 		prandom_bytes(params.rss_key, T_ETH_RSS_KEY * 4);
2023 		__set_bit(BNX2X_RSS_SET_SRCH, &params.rss_flags);
2024 	}
2025 
2026 	if (IS_PF(bp))
2027 		return bnx2x_config_rss(bp, &params);
2028 	else
2029 		return bnx2x_vfpf_config_rss(bp, &params);
2030 }
2031 
2032 static int bnx2x_init_hw(struct bnx2x *bp, u32 load_code)
2033 {
2034 	struct bnx2x_func_state_params func_params = {NULL};
2035 
2036 	/* Prepare parameters for function state transitions */
2037 	__set_bit(RAMROD_COMP_WAIT, &func_params.ramrod_flags);
2038 
2039 	func_params.f_obj = &bp->func_obj;
2040 	func_params.cmd = BNX2X_F_CMD_HW_INIT;
2041 
2042 	func_params.params.hw_init.load_phase = load_code;
2043 
2044 	return bnx2x_func_state_change(bp, &func_params);
2045 }
2046 
2047 /*
2048  * Cleans the object that have internal lists without sending
2049  * ramrods. Should be run when interrupts are disabled.
2050  */
2051 void bnx2x_squeeze_objects(struct bnx2x *bp)
2052 {
2053 	int rc;
2054 	unsigned long ramrod_flags = 0, vlan_mac_flags = 0;
2055 	struct bnx2x_mcast_ramrod_params rparam = {NULL};
2056 	struct bnx2x_vlan_mac_obj *mac_obj = &bp->sp_objs->mac_obj;
2057 
2058 	/***************** Cleanup MACs' object first *************************/
2059 
2060 	/* Wait for completion of requested */
2061 	__set_bit(RAMROD_COMP_WAIT, &ramrod_flags);
2062 	/* Perform a dry cleanup */
2063 	__set_bit(RAMROD_DRV_CLR_ONLY, &ramrod_flags);
2064 
2065 	/* Clean ETH primary MAC */
2066 	__set_bit(BNX2X_ETH_MAC, &vlan_mac_flags);
2067 	rc = mac_obj->delete_all(bp, &bp->sp_objs->mac_obj, &vlan_mac_flags,
2068 				 &ramrod_flags);
2069 	if (rc != 0)
2070 		BNX2X_ERR("Failed to clean ETH MACs: %d\n", rc);
2071 
2072 	/* Cleanup UC list */
2073 	vlan_mac_flags = 0;
2074 	__set_bit(BNX2X_UC_LIST_MAC, &vlan_mac_flags);
2075 	rc = mac_obj->delete_all(bp, mac_obj, &vlan_mac_flags,
2076 				 &ramrod_flags);
2077 	if (rc != 0)
2078 		BNX2X_ERR("Failed to clean UC list MACs: %d\n", rc);
2079 
2080 	/***************** Now clean mcast object *****************************/
2081 	rparam.mcast_obj = &bp->mcast_obj;
2082 	__set_bit(RAMROD_DRV_CLR_ONLY, &rparam.ramrod_flags);
2083 
2084 	/* Add a DEL command... - Since we're doing a driver cleanup only,
2085 	 * we take a lock surrounding both the initial send and the CONTs,
2086 	 * as we don't want a true completion to disrupt us in the middle.
2087 	 */
2088 	netif_addr_lock_bh(bp->dev);
2089 	rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_DEL);
2090 	if (rc < 0)
2091 		BNX2X_ERR("Failed to add a new DEL command to a multi-cast object: %d\n",
2092 			  rc);
2093 
2094 	/* ...and wait until all pending commands are cleared */
2095 	rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_CONT);
2096 	while (rc != 0) {
2097 		if (rc < 0) {
2098 			BNX2X_ERR("Failed to clean multi-cast object: %d\n",
2099 				  rc);
2100 			netif_addr_unlock_bh(bp->dev);
2101 			return;
2102 		}
2103 
2104 		rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_CONT);
2105 	}
2106 	netif_addr_unlock_bh(bp->dev);
2107 }
2108 
2109 #ifndef BNX2X_STOP_ON_ERROR
2110 #define LOAD_ERROR_EXIT(bp, label) \
2111 	do { \
2112 		(bp)->state = BNX2X_STATE_ERROR; \
2113 		goto label; \
2114 	} while (0)
2115 
2116 #define LOAD_ERROR_EXIT_CNIC(bp, label) \
2117 	do { \
2118 		bp->cnic_loaded = false; \
2119 		goto label; \
2120 	} while (0)
2121 #else /*BNX2X_STOP_ON_ERROR*/
2122 #define LOAD_ERROR_EXIT(bp, label) \
2123 	do { \
2124 		(bp)->state = BNX2X_STATE_ERROR; \
2125 		(bp)->panic = 1; \
2126 		return -EBUSY; \
2127 	} while (0)
2128 #define LOAD_ERROR_EXIT_CNIC(bp, label) \
2129 	do { \
2130 		bp->cnic_loaded = false; \
2131 		(bp)->panic = 1; \
2132 		return -EBUSY; \
2133 	} while (0)
2134 #endif /*BNX2X_STOP_ON_ERROR*/
2135 
2136 static void bnx2x_free_fw_stats_mem(struct bnx2x *bp)
2137 {
2138 	BNX2X_PCI_FREE(bp->fw_stats, bp->fw_stats_mapping,
2139 		       bp->fw_stats_data_sz + bp->fw_stats_req_sz);
2140 	return;
2141 }
2142 
2143 static int bnx2x_alloc_fw_stats_mem(struct bnx2x *bp)
2144 {
2145 	int num_groups, vf_headroom = 0;
2146 	int is_fcoe_stats = NO_FCOE(bp) ? 0 : 1;
2147 
2148 	/* number of queues for statistics is number of eth queues + FCoE */
2149 	u8 num_queue_stats = BNX2X_NUM_ETH_QUEUES(bp) + is_fcoe_stats;
2150 
2151 	/* Total number of FW statistics requests =
2152 	 * 1 for port stats + 1 for PF stats + potential 2 for FCoE (fcoe proper
2153 	 * and fcoe l2 queue) stats + num of queues (which includes another 1
2154 	 * for fcoe l2 queue if applicable)
2155 	 */
2156 	bp->fw_stats_num = 2 + is_fcoe_stats + num_queue_stats;
2157 
2158 	/* vf stats appear in the request list, but their data is allocated by
2159 	 * the VFs themselves. We don't include them in the bp->fw_stats_num as
2160 	 * it is used to determine where to place the vf stats queries in the
2161 	 * request struct
2162 	 */
2163 	if (IS_SRIOV(bp))
2164 		vf_headroom = bnx2x_vf_headroom(bp);
2165 
2166 	/* Request is built from stats_query_header and an array of
2167 	 * stats_query_cmd_group each of which contains
2168 	 * STATS_QUERY_CMD_COUNT rules. The real number or requests is
2169 	 * configured in the stats_query_header.
2170 	 */
2171 	num_groups =
2172 		(((bp->fw_stats_num + vf_headroom) / STATS_QUERY_CMD_COUNT) +
2173 		 (((bp->fw_stats_num + vf_headroom) % STATS_QUERY_CMD_COUNT) ?
2174 		 1 : 0));
2175 
2176 	DP(BNX2X_MSG_SP, "stats fw_stats_num %d, vf headroom %d, num_groups %d\n",
2177 	   bp->fw_stats_num, vf_headroom, num_groups);
2178 	bp->fw_stats_req_sz = sizeof(struct stats_query_header) +
2179 		num_groups * sizeof(struct stats_query_cmd_group);
2180 
2181 	/* Data for statistics requests + stats_counter
2182 	 * stats_counter holds per-STORM counters that are incremented
2183 	 * when STORM has finished with the current request.
2184 	 * memory for FCoE offloaded statistics are counted anyway,
2185 	 * even if they will not be sent.
2186 	 * VF stats are not accounted for here as the data of VF stats is stored
2187 	 * in memory allocated by the VF, not here.
2188 	 */
2189 	bp->fw_stats_data_sz = sizeof(struct per_port_stats) +
2190 		sizeof(struct per_pf_stats) +
2191 		sizeof(struct fcoe_statistics_params) +
2192 		sizeof(struct per_queue_stats) * num_queue_stats +
2193 		sizeof(struct stats_counter);
2194 
2195 	BNX2X_PCI_ALLOC(bp->fw_stats, &bp->fw_stats_mapping,
2196 			bp->fw_stats_data_sz + bp->fw_stats_req_sz);
2197 
2198 	/* Set shortcuts */
2199 	bp->fw_stats_req = (struct bnx2x_fw_stats_req *)bp->fw_stats;
2200 	bp->fw_stats_req_mapping = bp->fw_stats_mapping;
2201 	bp->fw_stats_data = (struct bnx2x_fw_stats_data *)
2202 		((u8 *)bp->fw_stats + bp->fw_stats_req_sz);
2203 	bp->fw_stats_data_mapping = bp->fw_stats_mapping +
2204 		bp->fw_stats_req_sz;
2205 
2206 	DP(BNX2X_MSG_SP, "statistics request base address set to %x %x\n",
2207 	   U64_HI(bp->fw_stats_req_mapping),
2208 	   U64_LO(bp->fw_stats_req_mapping));
2209 	DP(BNX2X_MSG_SP, "statistics data base address set to %x %x\n",
2210 	   U64_HI(bp->fw_stats_data_mapping),
2211 	   U64_LO(bp->fw_stats_data_mapping));
2212 	return 0;
2213 
2214 alloc_mem_err:
2215 	bnx2x_free_fw_stats_mem(bp);
2216 	BNX2X_ERR("Can't allocate FW stats memory\n");
2217 	return -ENOMEM;
2218 }
2219 
2220 /* send load request to mcp and analyze response */
2221 static int bnx2x_nic_load_request(struct bnx2x *bp, u32 *load_code)
2222 {
2223 	u32 param;
2224 
2225 	/* init fw_seq */
2226 	bp->fw_seq =
2227 		(SHMEM_RD(bp, func_mb[BP_FW_MB_IDX(bp)].drv_mb_header) &
2228 		 DRV_MSG_SEQ_NUMBER_MASK);
2229 	BNX2X_DEV_INFO("fw_seq 0x%08x\n", bp->fw_seq);
2230 
2231 	/* Get current FW pulse sequence */
2232 	bp->fw_drv_pulse_wr_seq =
2233 		(SHMEM_RD(bp, func_mb[BP_FW_MB_IDX(bp)].drv_pulse_mb) &
2234 		 DRV_PULSE_SEQ_MASK);
2235 	BNX2X_DEV_INFO("drv_pulse 0x%x\n", bp->fw_drv_pulse_wr_seq);
2236 
2237 	param = DRV_MSG_CODE_LOAD_REQ_WITH_LFA;
2238 
2239 	if (IS_MF_SD(bp) && bnx2x_port_after_undi(bp))
2240 		param |= DRV_MSG_CODE_LOAD_REQ_FORCE_LFA;
2241 
2242 	/* load request */
2243 	(*load_code) = bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_REQ, param);
2244 
2245 	/* if mcp fails to respond we must abort */
2246 	if (!(*load_code)) {
2247 		BNX2X_ERR("MCP response failure, aborting\n");
2248 		return -EBUSY;
2249 	}
2250 
2251 	/* If mcp refused (e.g. other port is in diagnostic mode) we
2252 	 * must abort
2253 	 */
2254 	if ((*load_code) == FW_MSG_CODE_DRV_LOAD_REFUSED) {
2255 		BNX2X_ERR("MCP refused load request, aborting\n");
2256 		return -EBUSY;
2257 	}
2258 	return 0;
2259 }
2260 
2261 /* check whether another PF has already loaded FW to chip. In
2262  * virtualized environments a pf from another VM may have already
2263  * initialized the device including loading FW
2264  */
2265 int bnx2x_nic_load_analyze_req(struct bnx2x *bp, u32 load_code)
2266 {
2267 	/* is another pf loaded on this engine? */
2268 	if (load_code != FW_MSG_CODE_DRV_LOAD_COMMON_CHIP &&
2269 	    load_code != FW_MSG_CODE_DRV_LOAD_COMMON) {
2270 		/* build my FW version dword */
2271 		u32 my_fw = (BCM_5710_FW_MAJOR_VERSION) +
2272 			(BCM_5710_FW_MINOR_VERSION << 8) +
2273 			(BCM_5710_FW_REVISION_VERSION << 16) +
2274 			(BCM_5710_FW_ENGINEERING_VERSION << 24);
2275 
2276 		/* read loaded FW from chip */
2277 		u32 loaded_fw = REG_RD(bp, XSEM_REG_PRAM);
2278 
2279 		DP(BNX2X_MSG_SP, "loaded fw %x, my fw %x\n",
2280 		   loaded_fw, my_fw);
2281 
2282 		/* abort nic load if version mismatch */
2283 		if (my_fw != loaded_fw) {
2284 			BNX2X_ERR("bnx2x with FW %x was already loaded which mismatches my %x FW. Aborting\n",
2285 				  loaded_fw, my_fw);
2286 			return -EBUSY;
2287 		}
2288 	}
2289 	return 0;
2290 }
2291 
2292 /* returns the "mcp load_code" according to global load_count array */
2293 static int bnx2x_nic_load_no_mcp(struct bnx2x *bp, int port)
2294 {
2295 	int path = BP_PATH(bp);
2296 
2297 	DP(NETIF_MSG_IFUP, "NO MCP - load counts[%d]      %d, %d, %d\n",
2298 	   path, load_count[path][0], load_count[path][1],
2299 	   load_count[path][2]);
2300 	load_count[path][0]++;
2301 	load_count[path][1 + port]++;
2302 	DP(NETIF_MSG_IFUP, "NO MCP - new load counts[%d]  %d, %d, %d\n",
2303 	   path, load_count[path][0], load_count[path][1],
2304 	   load_count[path][2]);
2305 	if (load_count[path][0] == 1)
2306 		return FW_MSG_CODE_DRV_LOAD_COMMON;
2307 	else if (load_count[path][1 + port] == 1)
2308 		return FW_MSG_CODE_DRV_LOAD_PORT;
2309 	else
2310 		return FW_MSG_CODE_DRV_LOAD_FUNCTION;
2311 }
2312 
2313 /* mark PMF if applicable */
2314 static void bnx2x_nic_load_pmf(struct bnx2x *bp, u32 load_code)
2315 {
2316 	if ((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
2317 	    (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP) ||
2318 	    (load_code == FW_MSG_CODE_DRV_LOAD_PORT)) {
2319 		bp->port.pmf = 1;
2320 		/* We need the barrier to ensure the ordering between the
2321 		 * writing to bp->port.pmf here and reading it from the
2322 		 * bnx2x_periodic_task().
2323 		 */
2324 		smp_mb();
2325 	} else {
2326 		bp->port.pmf = 0;
2327 	}
2328 
2329 	DP(NETIF_MSG_LINK, "pmf %d\n", bp->port.pmf);
2330 }
2331 
2332 static void bnx2x_nic_load_afex_dcc(struct bnx2x *bp, int load_code)
2333 {
2334 	if (((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
2335 	     (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP)) &&
2336 	    (bp->common.shmem2_base)) {
2337 		if (SHMEM2_HAS(bp, dcc_support))
2338 			SHMEM2_WR(bp, dcc_support,
2339 				  (SHMEM_DCC_SUPPORT_DISABLE_ENABLE_PF_TLV |
2340 				   SHMEM_DCC_SUPPORT_BANDWIDTH_ALLOCATION_TLV));
2341 		if (SHMEM2_HAS(bp, afex_driver_support))
2342 			SHMEM2_WR(bp, afex_driver_support,
2343 				  SHMEM_AFEX_SUPPORTED_VERSION_ONE);
2344 	}
2345 
2346 	/* Set AFEX default VLAN tag to an invalid value */
2347 	bp->afex_def_vlan_tag = -1;
2348 }
2349 
2350 /**
2351  * bnx2x_bz_fp - zero content of the fastpath structure.
2352  *
2353  * @bp:		driver handle
2354  * @index:	fastpath index to be zeroed
2355  *
2356  * Makes sure the contents of the bp->fp[index].napi is kept
2357  * intact.
2358  */
2359 static void bnx2x_bz_fp(struct bnx2x *bp, int index)
2360 {
2361 	struct bnx2x_fastpath *fp = &bp->fp[index];
2362 	int cos;
2363 	struct napi_struct orig_napi = fp->napi;
2364 	struct bnx2x_agg_info *orig_tpa_info = fp->tpa_info;
2365 
2366 	/* bzero bnx2x_fastpath contents */
2367 	if (fp->tpa_info)
2368 		memset(fp->tpa_info, 0, ETH_MAX_AGGREGATION_QUEUES_E1H_E2 *
2369 		       sizeof(struct bnx2x_agg_info));
2370 	memset(fp, 0, sizeof(*fp));
2371 
2372 	/* Restore the NAPI object as it has been already initialized */
2373 	fp->napi = orig_napi;
2374 	fp->tpa_info = orig_tpa_info;
2375 	fp->bp = bp;
2376 	fp->index = index;
2377 	if (IS_ETH_FP(fp))
2378 		fp->max_cos = bp->max_cos;
2379 	else
2380 		/* Special queues support only one CoS */
2381 		fp->max_cos = 1;
2382 
2383 	/* Init txdata pointers */
2384 	if (IS_FCOE_FP(fp))
2385 		fp->txdata_ptr[0] = &bp->bnx2x_txq[FCOE_TXQ_IDX(bp)];
2386 	if (IS_ETH_FP(fp))
2387 		for_each_cos_in_tx_queue(fp, cos)
2388 			fp->txdata_ptr[cos] = &bp->bnx2x_txq[cos *
2389 				BNX2X_NUM_ETH_QUEUES(bp) + index];
2390 
2391 	/* set the tpa flag for each queue. The tpa flag determines the queue
2392 	 * minimal size so it must be set prior to queue memory allocation
2393 	 */
2394 	fp->disable_tpa = !(bp->flags & TPA_ENABLE_FLAG ||
2395 				  (bp->flags & GRO_ENABLE_FLAG &&
2396 				   bnx2x_mtu_allows_gro(bp->dev->mtu)));
2397 	if (bp->flags & TPA_ENABLE_FLAG)
2398 		fp->mode = TPA_MODE_LRO;
2399 	else if (bp->flags & GRO_ENABLE_FLAG)
2400 		fp->mode = TPA_MODE_GRO;
2401 
2402 	/* We don't want TPA on an FCoE L2 ring */
2403 	if (IS_FCOE_FP(fp))
2404 		fp->disable_tpa = 1;
2405 }
2406 
2407 int bnx2x_load_cnic(struct bnx2x *bp)
2408 {
2409 	int i, rc, port = BP_PORT(bp);
2410 
2411 	DP(NETIF_MSG_IFUP, "Starting CNIC-related load\n");
2412 
2413 	mutex_init(&bp->cnic_mutex);
2414 
2415 	if (IS_PF(bp)) {
2416 		rc = bnx2x_alloc_mem_cnic(bp);
2417 		if (rc) {
2418 			BNX2X_ERR("Unable to allocate bp memory for cnic\n");
2419 			LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic0);
2420 		}
2421 	}
2422 
2423 	rc = bnx2x_alloc_fp_mem_cnic(bp);
2424 	if (rc) {
2425 		BNX2X_ERR("Unable to allocate memory for cnic fps\n");
2426 		LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic0);
2427 	}
2428 
2429 	/* Update the number of queues with the cnic queues */
2430 	rc = bnx2x_set_real_num_queues(bp, 1);
2431 	if (rc) {
2432 		BNX2X_ERR("Unable to set real_num_queues including cnic\n");
2433 		LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic0);
2434 	}
2435 
2436 	/* Add all CNIC NAPI objects */
2437 	bnx2x_add_all_napi_cnic(bp);
2438 	DP(NETIF_MSG_IFUP, "cnic napi added\n");
2439 	bnx2x_napi_enable_cnic(bp);
2440 
2441 	rc = bnx2x_init_hw_func_cnic(bp);
2442 	if (rc)
2443 		LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic1);
2444 
2445 	bnx2x_nic_init_cnic(bp);
2446 
2447 	if (IS_PF(bp)) {
2448 		/* Enable Timer scan */
2449 		REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + port*4, 1);
2450 
2451 		/* setup cnic queues */
2452 		for_each_cnic_queue(bp, i) {
2453 			rc = bnx2x_setup_queue(bp, &bp->fp[i], 0);
2454 			if (rc) {
2455 				BNX2X_ERR("Queue setup failed\n");
2456 				LOAD_ERROR_EXIT(bp, load_error_cnic2);
2457 			}
2458 		}
2459 	}
2460 
2461 	/* Initialize Rx filter. */
2462 	bnx2x_set_rx_mode_inner(bp);
2463 
2464 	/* re-read iscsi info */
2465 	bnx2x_get_iscsi_info(bp);
2466 	bnx2x_setup_cnic_irq_info(bp);
2467 	bnx2x_setup_cnic_info(bp);
2468 	bp->cnic_loaded = true;
2469 	if (bp->state == BNX2X_STATE_OPEN)
2470 		bnx2x_cnic_notify(bp, CNIC_CTL_START_CMD);
2471 
2472 	DP(NETIF_MSG_IFUP, "Ending successfully CNIC-related load\n");
2473 
2474 	return 0;
2475 
2476 #ifndef BNX2X_STOP_ON_ERROR
2477 load_error_cnic2:
2478 	/* Disable Timer scan */
2479 	REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + port*4, 0);
2480 
2481 load_error_cnic1:
2482 	bnx2x_napi_disable_cnic(bp);
2483 	/* Update the number of queues without the cnic queues */
2484 	if (bnx2x_set_real_num_queues(bp, 0))
2485 		BNX2X_ERR("Unable to set real_num_queues not including cnic\n");
2486 load_error_cnic0:
2487 	BNX2X_ERR("CNIC-related load failed\n");
2488 	bnx2x_free_fp_mem_cnic(bp);
2489 	bnx2x_free_mem_cnic(bp);
2490 	return rc;
2491 #endif /* ! BNX2X_STOP_ON_ERROR */
2492 }
2493 
2494 /* must be called with rtnl_lock */
2495 int bnx2x_nic_load(struct bnx2x *bp, int load_mode)
2496 {
2497 	int port = BP_PORT(bp);
2498 	int i, rc = 0, load_code = 0;
2499 
2500 	DP(NETIF_MSG_IFUP, "Starting NIC load\n");
2501 	DP(NETIF_MSG_IFUP,
2502 	   "CNIC is %s\n", CNIC_ENABLED(bp) ? "enabled" : "disabled");
2503 
2504 #ifdef BNX2X_STOP_ON_ERROR
2505 	if (unlikely(bp->panic)) {
2506 		BNX2X_ERR("Can't load NIC when there is panic\n");
2507 		return -EPERM;
2508 	}
2509 #endif
2510 
2511 	bp->state = BNX2X_STATE_OPENING_WAIT4_LOAD;
2512 
2513 	/* zero the structure w/o any lock, before SP handler is initialized */
2514 	memset(&bp->last_reported_link, 0, sizeof(bp->last_reported_link));
2515 	__set_bit(BNX2X_LINK_REPORT_LINK_DOWN,
2516 		&bp->last_reported_link.link_report_flags);
2517 
2518 	if (IS_PF(bp))
2519 		/* must be called before memory allocation and HW init */
2520 		bnx2x_ilt_set_info(bp);
2521 
2522 	/*
2523 	 * Zero fastpath structures preserving invariants like napi, which are
2524 	 * allocated only once, fp index, max_cos, bp pointer.
2525 	 * Also set fp->disable_tpa and txdata_ptr.
2526 	 */
2527 	DP(NETIF_MSG_IFUP, "num queues: %d", bp->num_queues);
2528 	for_each_queue(bp, i)
2529 		bnx2x_bz_fp(bp, i);
2530 	memset(bp->bnx2x_txq, 0, (BNX2X_MAX_RSS_COUNT(bp) * BNX2X_MULTI_TX_COS +
2531 				  bp->num_cnic_queues) *
2532 				  sizeof(struct bnx2x_fp_txdata));
2533 
2534 	bp->fcoe_init = false;
2535 
2536 	/* Set the receive queues buffer size */
2537 	bnx2x_set_rx_buf_size(bp);
2538 
2539 	if (IS_PF(bp)) {
2540 		rc = bnx2x_alloc_mem(bp);
2541 		if (rc) {
2542 			BNX2X_ERR("Unable to allocate bp memory\n");
2543 			return rc;
2544 		}
2545 	}
2546 
2547 	/* Allocated memory for FW statistics  */
2548 	if (bnx2x_alloc_fw_stats_mem(bp))
2549 		LOAD_ERROR_EXIT(bp, load_error0);
2550 
2551 	/* need to be done after alloc mem, since it's self adjusting to amount
2552 	 * of memory available for RSS queues
2553 	 */
2554 	rc = bnx2x_alloc_fp_mem(bp);
2555 	if (rc) {
2556 		BNX2X_ERR("Unable to allocate memory for fps\n");
2557 		LOAD_ERROR_EXIT(bp, load_error0);
2558 	}
2559 
2560 	/* request pf to initialize status blocks */
2561 	if (IS_VF(bp)) {
2562 		rc = bnx2x_vfpf_init(bp);
2563 		if (rc)
2564 			LOAD_ERROR_EXIT(bp, load_error0);
2565 	}
2566 
2567 	/* As long as bnx2x_alloc_mem() may possibly update
2568 	 * bp->num_queues, bnx2x_set_real_num_queues() should always
2569 	 * come after it. At this stage cnic queues are not counted.
2570 	 */
2571 	rc = bnx2x_set_real_num_queues(bp, 0);
2572 	if (rc) {
2573 		BNX2X_ERR("Unable to set real_num_queues\n");
2574 		LOAD_ERROR_EXIT(bp, load_error0);
2575 	}
2576 
2577 	/* configure multi cos mappings in kernel.
2578 	 * this configuration may be overridden by a multi class queue
2579 	 * discipline or by a dcbx negotiation result.
2580 	 */
2581 	bnx2x_setup_tc(bp->dev, bp->max_cos);
2582 
2583 	/* Add all NAPI objects */
2584 	bnx2x_add_all_napi(bp);
2585 	DP(NETIF_MSG_IFUP, "napi added\n");
2586 	bnx2x_napi_enable(bp);
2587 
2588 	if (IS_PF(bp)) {
2589 		/* set pf load just before approaching the MCP */
2590 		bnx2x_set_pf_load(bp);
2591 
2592 		/* if mcp exists send load request and analyze response */
2593 		if (!BP_NOMCP(bp)) {
2594 			/* attempt to load pf */
2595 			rc = bnx2x_nic_load_request(bp, &load_code);
2596 			if (rc)
2597 				LOAD_ERROR_EXIT(bp, load_error1);
2598 
2599 			/* what did mcp say? */
2600 			rc = bnx2x_nic_load_analyze_req(bp, load_code);
2601 			if (rc) {
2602 				bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2603 				LOAD_ERROR_EXIT(bp, load_error2);
2604 			}
2605 		} else {
2606 			load_code = bnx2x_nic_load_no_mcp(bp, port);
2607 		}
2608 
2609 		/* mark pmf if applicable */
2610 		bnx2x_nic_load_pmf(bp, load_code);
2611 
2612 		/* Init Function state controlling object */
2613 		bnx2x__init_func_obj(bp);
2614 
2615 		/* Initialize HW */
2616 		rc = bnx2x_init_hw(bp, load_code);
2617 		if (rc) {
2618 			BNX2X_ERR("HW init failed, aborting\n");
2619 			bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2620 			LOAD_ERROR_EXIT(bp, load_error2);
2621 		}
2622 	}
2623 
2624 	bnx2x_pre_irq_nic_init(bp);
2625 
2626 	/* Connect to IRQs */
2627 	rc = bnx2x_setup_irqs(bp);
2628 	if (rc) {
2629 		BNX2X_ERR("setup irqs failed\n");
2630 		if (IS_PF(bp))
2631 			bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2632 		LOAD_ERROR_EXIT(bp, load_error2);
2633 	}
2634 
2635 	/* Init per-function objects */
2636 	if (IS_PF(bp)) {
2637 		/* Setup NIC internals and enable interrupts */
2638 		bnx2x_post_irq_nic_init(bp, load_code);
2639 
2640 		bnx2x_init_bp_objs(bp);
2641 		bnx2x_iov_nic_init(bp);
2642 
2643 		/* Set AFEX default VLAN tag to an invalid value */
2644 		bp->afex_def_vlan_tag = -1;
2645 		bnx2x_nic_load_afex_dcc(bp, load_code);
2646 		bp->state = BNX2X_STATE_OPENING_WAIT4_PORT;
2647 		rc = bnx2x_func_start(bp);
2648 		if (rc) {
2649 			BNX2X_ERR("Function start failed!\n");
2650 			bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2651 
2652 			LOAD_ERROR_EXIT(bp, load_error3);
2653 		}
2654 
2655 		/* Send LOAD_DONE command to MCP */
2656 		if (!BP_NOMCP(bp)) {
2657 			load_code = bnx2x_fw_command(bp,
2658 						     DRV_MSG_CODE_LOAD_DONE, 0);
2659 			if (!load_code) {
2660 				BNX2X_ERR("MCP response failure, aborting\n");
2661 				rc = -EBUSY;
2662 				LOAD_ERROR_EXIT(bp, load_error3);
2663 			}
2664 		}
2665 
2666 		/* initialize FW coalescing state machines in RAM */
2667 		bnx2x_update_coalesce(bp);
2668 	}
2669 
2670 	/* setup the leading queue */
2671 	rc = bnx2x_setup_leading(bp);
2672 	if (rc) {
2673 		BNX2X_ERR("Setup leading failed!\n");
2674 		LOAD_ERROR_EXIT(bp, load_error3);
2675 	}
2676 
2677 	/* set up the rest of the queues */
2678 	for_each_nondefault_eth_queue(bp, i) {
2679 		if (IS_PF(bp))
2680 			rc = bnx2x_setup_queue(bp, &bp->fp[i], false);
2681 		else /* VF */
2682 			rc = bnx2x_vfpf_setup_q(bp, &bp->fp[i], false);
2683 		if (rc) {
2684 			BNX2X_ERR("Queue %d setup failed\n", i);
2685 			LOAD_ERROR_EXIT(bp, load_error3);
2686 		}
2687 	}
2688 
2689 	/* setup rss */
2690 	rc = bnx2x_init_rss(bp);
2691 	if (rc) {
2692 		BNX2X_ERR("PF RSS init failed\n");
2693 		LOAD_ERROR_EXIT(bp, load_error3);
2694 	}
2695 
2696 	/* Now when Clients are configured we are ready to work */
2697 	bp->state = BNX2X_STATE_OPEN;
2698 
2699 	/* Configure a ucast MAC */
2700 	if (IS_PF(bp))
2701 		rc = bnx2x_set_eth_mac(bp, true);
2702 	else /* vf */
2703 		rc = bnx2x_vfpf_config_mac(bp, bp->dev->dev_addr, bp->fp->index,
2704 					   true);
2705 	if (rc) {
2706 		BNX2X_ERR("Setting Ethernet MAC failed\n");
2707 		LOAD_ERROR_EXIT(bp, load_error3);
2708 	}
2709 
2710 	if (IS_PF(bp) && bp->pending_max) {
2711 		bnx2x_update_max_mf_config(bp, bp->pending_max);
2712 		bp->pending_max = 0;
2713 	}
2714 
2715 	if (bp->port.pmf) {
2716 		rc = bnx2x_initial_phy_init(bp, load_mode);
2717 		if (rc)
2718 			LOAD_ERROR_EXIT(bp, load_error3);
2719 	}
2720 	bp->link_params.feature_config_flags &= ~FEATURE_CONFIG_BOOT_FROM_SAN;
2721 
2722 	/* Start fast path */
2723 
2724 	/* Initialize Rx filter. */
2725 	bnx2x_set_rx_mode_inner(bp);
2726 
2727 	/* Start the Tx */
2728 	switch (load_mode) {
2729 	case LOAD_NORMAL:
2730 		/* Tx queue should be only re-enabled */
2731 		netif_tx_wake_all_queues(bp->dev);
2732 		break;
2733 
2734 	case LOAD_OPEN:
2735 		netif_tx_start_all_queues(bp->dev);
2736 		smp_mb__after_clear_bit();
2737 		break;
2738 
2739 	case LOAD_DIAG:
2740 	case LOAD_LOOPBACK_EXT:
2741 		bp->state = BNX2X_STATE_DIAG;
2742 		break;
2743 
2744 	default:
2745 		break;
2746 	}
2747 
2748 	if (bp->port.pmf)
2749 		bnx2x_update_drv_flags(bp, 1 << DRV_FLAGS_PORT_MASK, 0);
2750 	else
2751 		bnx2x__link_status_update(bp);
2752 
2753 	/* start the timer */
2754 	mod_timer(&bp->timer, jiffies + bp->current_interval);
2755 
2756 	if (CNIC_ENABLED(bp))
2757 		bnx2x_load_cnic(bp);
2758 
2759 	if (IS_PF(bp) && SHMEM2_HAS(bp, drv_capabilities_flag)) {
2760 		/* mark driver is loaded in shmem2 */
2761 		u32 val;
2762 		val = SHMEM2_RD(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)]);
2763 		SHMEM2_WR(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)],
2764 			  val | DRV_FLAGS_CAPABILITIES_LOADED_SUPPORTED |
2765 			  DRV_FLAGS_CAPABILITIES_LOADED_L2);
2766 	}
2767 
2768 	/* Wait for all pending SP commands to complete */
2769 	if (IS_PF(bp) && !bnx2x_wait_sp_comp(bp, ~0x0UL)) {
2770 		BNX2X_ERR("Timeout waiting for SP elements to complete\n");
2771 		bnx2x_nic_unload(bp, UNLOAD_CLOSE, false);
2772 		return -EBUSY;
2773 	}
2774 
2775 	/* If PMF - send ADMIN DCBX msg to MFW to initiate DCBX FSM */
2776 	if (bp->port.pmf && (bp->state != BNX2X_STATE_DIAG))
2777 		bnx2x_dcbx_init(bp, false);
2778 
2779 	DP(NETIF_MSG_IFUP, "Ending successfully NIC load\n");
2780 
2781 	return 0;
2782 
2783 #ifndef BNX2X_STOP_ON_ERROR
2784 load_error3:
2785 	if (IS_PF(bp)) {
2786 		bnx2x_int_disable_sync(bp, 1);
2787 
2788 		/* Clean queueable objects */
2789 		bnx2x_squeeze_objects(bp);
2790 	}
2791 
2792 	/* Free SKBs, SGEs, TPA pool and driver internals */
2793 	bnx2x_free_skbs(bp);
2794 	for_each_rx_queue(bp, i)
2795 		bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
2796 
2797 	/* Release IRQs */
2798 	bnx2x_free_irq(bp);
2799 load_error2:
2800 	if (IS_PF(bp) && !BP_NOMCP(bp)) {
2801 		bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_REQ_WOL_MCP, 0);
2802 		bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_DONE, 0);
2803 	}
2804 
2805 	bp->port.pmf = 0;
2806 load_error1:
2807 	bnx2x_napi_disable(bp);
2808 	bnx2x_del_all_napi(bp);
2809 
2810 	/* clear pf_load status, as it was already set */
2811 	if (IS_PF(bp))
2812 		bnx2x_clear_pf_load(bp);
2813 load_error0:
2814 	bnx2x_free_fp_mem(bp);
2815 	bnx2x_free_fw_stats_mem(bp);
2816 	bnx2x_free_mem(bp);
2817 
2818 	return rc;
2819 #endif /* ! BNX2X_STOP_ON_ERROR */
2820 }
2821 
2822 int bnx2x_drain_tx_queues(struct bnx2x *bp)
2823 {
2824 	u8 rc = 0, cos, i;
2825 
2826 	/* Wait until tx fastpath tasks complete */
2827 	for_each_tx_queue(bp, i) {
2828 		struct bnx2x_fastpath *fp = &bp->fp[i];
2829 
2830 		for_each_cos_in_tx_queue(fp, cos)
2831 			rc = bnx2x_clean_tx_queue(bp, fp->txdata_ptr[cos]);
2832 		if (rc)
2833 			return rc;
2834 	}
2835 	return 0;
2836 }
2837 
2838 /* must be called with rtnl_lock */
2839 int bnx2x_nic_unload(struct bnx2x *bp, int unload_mode, bool keep_link)
2840 {
2841 	int i;
2842 	bool global = false;
2843 
2844 	DP(NETIF_MSG_IFUP, "Starting NIC unload\n");
2845 
2846 	/* mark driver is unloaded in shmem2 */
2847 	if (IS_PF(bp) && SHMEM2_HAS(bp, drv_capabilities_flag)) {
2848 		u32 val;
2849 		val = SHMEM2_RD(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)]);
2850 		SHMEM2_WR(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)],
2851 			  val & ~DRV_FLAGS_CAPABILITIES_LOADED_L2);
2852 	}
2853 
2854 	if (IS_PF(bp) && bp->recovery_state != BNX2X_RECOVERY_DONE &&
2855 	    (bp->state == BNX2X_STATE_CLOSED ||
2856 	     bp->state == BNX2X_STATE_ERROR)) {
2857 		/* We can get here if the driver has been unloaded
2858 		 * during parity error recovery and is either waiting for a
2859 		 * leader to complete or for other functions to unload and
2860 		 * then ifdown has been issued. In this case we want to
2861 		 * unload and let other functions to complete a recovery
2862 		 * process.
2863 		 */
2864 		bp->recovery_state = BNX2X_RECOVERY_DONE;
2865 		bp->is_leader = 0;
2866 		bnx2x_release_leader_lock(bp);
2867 		smp_mb();
2868 
2869 		DP(NETIF_MSG_IFDOWN, "Releasing a leadership...\n");
2870 		BNX2X_ERR("Can't unload in closed or error state\n");
2871 		return -EINVAL;
2872 	}
2873 
2874 	/* Nothing to do during unload if previous bnx2x_nic_load()
2875 	 * have not completed successfully - all resources are released.
2876 	 *
2877 	 * we can get here only after unsuccessful ndo_* callback, during which
2878 	 * dev->IFF_UP flag is still on.
2879 	 */
2880 	if (bp->state == BNX2X_STATE_CLOSED || bp->state == BNX2X_STATE_ERROR)
2881 		return 0;
2882 
2883 	/* It's important to set the bp->state to the value different from
2884 	 * BNX2X_STATE_OPEN and only then stop the Tx. Otherwise bnx2x_tx_int()
2885 	 * may restart the Tx from the NAPI context (see bnx2x_tx_int()).
2886 	 */
2887 	bp->state = BNX2X_STATE_CLOSING_WAIT4_HALT;
2888 	smp_mb();
2889 
2890 	/* indicate to VFs that the PF is going down */
2891 	bnx2x_iov_channel_down(bp);
2892 
2893 	if (CNIC_LOADED(bp))
2894 		bnx2x_cnic_notify(bp, CNIC_CTL_STOP_CMD);
2895 
2896 	/* Stop Tx */
2897 	bnx2x_tx_disable(bp);
2898 	netdev_reset_tc(bp->dev);
2899 
2900 	bp->rx_mode = BNX2X_RX_MODE_NONE;
2901 
2902 	del_timer_sync(&bp->timer);
2903 
2904 	if (IS_PF(bp)) {
2905 		/* Set ALWAYS_ALIVE bit in shmem */
2906 		bp->fw_drv_pulse_wr_seq |= DRV_PULSE_ALWAYS_ALIVE;
2907 		bnx2x_drv_pulse(bp);
2908 		bnx2x_stats_handle(bp, STATS_EVENT_STOP);
2909 		bnx2x_save_statistics(bp);
2910 	}
2911 
2912 	/* wait till consumers catch up with producers in all queues */
2913 	bnx2x_drain_tx_queues(bp);
2914 
2915 	/* if VF indicate to PF this function is going down (PF will delete sp
2916 	 * elements and clear initializations
2917 	 */
2918 	if (IS_VF(bp))
2919 		bnx2x_vfpf_close_vf(bp);
2920 	else if (unload_mode != UNLOAD_RECOVERY)
2921 		/* if this is a normal/close unload need to clean up chip*/
2922 		bnx2x_chip_cleanup(bp, unload_mode, keep_link);
2923 	else {
2924 		/* Send the UNLOAD_REQUEST to the MCP */
2925 		bnx2x_send_unload_req(bp, unload_mode);
2926 
2927 		/* Prevent transactions to host from the functions on the
2928 		 * engine that doesn't reset global blocks in case of global
2929 		 * attention once global blocks are reset and gates are opened
2930 		 * (the engine which leader will perform the recovery
2931 		 * last).
2932 		 */
2933 		if (!CHIP_IS_E1x(bp))
2934 			bnx2x_pf_disable(bp);
2935 
2936 		/* Disable HW interrupts, NAPI */
2937 		bnx2x_netif_stop(bp, 1);
2938 		/* Delete all NAPI objects */
2939 		bnx2x_del_all_napi(bp);
2940 		if (CNIC_LOADED(bp))
2941 			bnx2x_del_all_napi_cnic(bp);
2942 		/* Release IRQs */
2943 		bnx2x_free_irq(bp);
2944 
2945 		/* Report UNLOAD_DONE to MCP */
2946 		bnx2x_send_unload_done(bp, false);
2947 	}
2948 
2949 	/*
2950 	 * At this stage no more interrupts will arrive so we may safely clean
2951 	 * the queueable objects here in case they failed to get cleaned so far.
2952 	 */
2953 	if (IS_PF(bp))
2954 		bnx2x_squeeze_objects(bp);
2955 
2956 	/* There should be no more pending SP commands at this stage */
2957 	bp->sp_state = 0;
2958 
2959 	bp->port.pmf = 0;
2960 
2961 	/* Free SKBs, SGEs, TPA pool and driver internals */
2962 	bnx2x_free_skbs(bp);
2963 	if (CNIC_LOADED(bp))
2964 		bnx2x_free_skbs_cnic(bp);
2965 	for_each_rx_queue(bp, i)
2966 		bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
2967 
2968 	bnx2x_free_fp_mem(bp);
2969 	if (CNIC_LOADED(bp))
2970 		bnx2x_free_fp_mem_cnic(bp);
2971 
2972 	if (IS_PF(bp)) {
2973 		if (CNIC_LOADED(bp))
2974 			bnx2x_free_mem_cnic(bp);
2975 	}
2976 	bnx2x_free_mem(bp);
2977 
2978 	bp->state = BNX2X_STATE_CLOSED;
2979 	bp->cnic_loaded = false;
2980 
2981 	/* Check if there are pending parity attentions. If there are - set
2982 	 * RECOVERY_IN_PROGRESS.
2983 	 */
2984 	if (IS_PF(bp) && bnx2x_chk_parity_attn(bp, &global, false)) {
2985 		bnx2x_set_reset_in_progress(bp);
2986 
2987 		/* Set RESET_IS_GLOBAL if needed */
2988 		if (global)
2989 			bnx2x_set_reset_global(bp);
2990 	}
2991 
2992 	/* The last driver must disable a "close the gate" if there is no
2993 	 * parity attention or "process kill" pending.
2994 	 */
2995 	if (IS_PF(bp) &&
2996 	    !bnx2x_clear_pf_load(bp) &&
2997 	    bnx2x_reset_is_done(bp, BP_PATH(bp)))
2998 		bnx2x_disable_close_the_gate(bp);
2999 
3000 	DP(NETIF_MSG_IFUP, "Ending NIC unload\n");
3001 
3002 	return 0;
3003 }
3004 
3005 int bnx2x_set_power_state(struct bnx2x *bp, pci_power_t state)
3006 {
3007 	u16 pmcsr;
3008 
3009 	/* If there is no power capability, silently succeed */
3010 	if (!bp->pdev->pm_cap) {
3011 		BNX2X_DEV_INFO("No power capability. Breaking.\n");
3012 		return 0;
3013 	}
3014 
3015 	pci_read_config_word(bp->pdev, bp->pdev->pm_cap + PCI_PM_CTRL, &pmcsr);
3016 
3017 	switch (state) {
3018 	case PCI_D0:
3019 		pci_write_config_word(bp->pdev, bp->pdev->pm_cap + PCI_PM_CTRL,
3020 				      ((pmcsr & ~PCI_PM_CTRL_STATE_MASK) |
3021 				       PCI_PM_CTRL_PME_STATUS));
3022 
3023 		if (pmcsr & PCI_PM_CTRL_STATE_MASK)
3024 			/* delay required during transition out of D3hot */
3025 			msleep(20);
3026 		break;
3027 
3028 	case PCI_D3hot:
3029 		/* If there are other clients above don't
3030 		   shut down the power */
3031 		if (atomic_read(&bp->pdev->enable_cnt) != 1)
3032 			return 0;
3033 		/* Don't shut down the power for emulation and FPGA */
3034 		if (CHIP_REV_IS_SLOW(bp))
3035 			return 0;
3036 
3037 		pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
3038 		pmcsr |= 3;
3039 
3040 		if (bp->wol)
3041 			pmcsr |= PCI_PM_CTRL_PME_ENABLE;
3042 
3043 		pci_write_config_word(bp->pdev, bp->pdev->pm_cap + PCI_PM_CTRL,
3044 				      pmcsr);
3045 
3046 		/* No more memory access after this point until
3047 		* device is brought back to D0.
3048 		*/
3049 		break;
3050 
3051 	default:
3052 		dev_err(&bp->pdev->dev, "Can't support state = %d\n", state);
3053 		return -EINVAL;
3054 	}
3055 	return 0;
3056 }
3057 
3058 /*
3059  * net_device service functions
3060  */
3061 int bnx2x_poll(struct napi_struct *napi, int budget)
3062 {
3063 	int work_done = 0;
3064 	u8 cos;
3065 	struct bnx2x_fastpath *fp = container_of(napi, struct bnx2x_fastpath,
3066 						 napi);
3067 	struct bnx2x *bp = fp->bp;
3068 
3069 	while (1) {
3070 #ifdef BNX2X_STOP_ON_ERROR
3071 		if (unlikely(bp->panic)) {
3072 			napi_complete(napi);
3073 			return 0;
3074 		}
3075 #endif
3076 		if (!bnx2x_fp_lock_napi(fp))
3077 			return work_done;
3078 
3079 		for_each_cos_in_tx_queue(fp, cos)
3080 			if (bnx2x_tx_queue_has_work(fp->txdata_ptr[cos]))
3081 				bnx2x_tx_int(bp, fp->txdata_ptr[cos]);
3082 
3083 		if (bnx2x_has_rx_work(fp)) {
3084 			work_done += bnx2x_rx_int(fp, budget - work_done);
3085 
3086 			/* must not complete if we consumed full budget */
3087 			if (work_done >= budget) {
3088 				bnx2x_fp_unlock_napi(fp);
3089 				break;
3090 			}
3091 		}
3092 
3093 		/* Fall out from the NAPI loop if needed */
3094 		if (!bnx2x_fp_unlock_napi(fp) &&
3095 		    !(bnx2x_has_rx_work(fp) || bnx2x_has_tx_work(fp))) {
3096 
3097 			/* No need to update SB for FCoE L2 ring as long as
3098 			 * it's connected to the default SB and the SB
3099 			 * has been updated when NAPI was scheduled.
3100 			 */
3101 			if (IS_FCOE_FP(fp)) {
3102 				napi_complete(napi);
3103 				break;
3104 			}
3105 			bnx2x_update_fpsb_idx(fp);
3106 			/* bnx2x_has_rx_work() reads the status block,
3107 			 * thus we need to ensure that status block indices
3108 			 * have been actually read (bnx2x_update_fpsb_idx)
3109 			 * prior to this check (bnx2x_has_rx_work) so that
3110 			 * we won't write the "newer" value of the status block
3111 			 * to IGU (if there was a DMA right after
3112 			 * bnx2x_has_rx_work and if there is no rmb, the memory
3113 			 * reading (bnx2x_update_fpsb_idx) may be postponed
3114 			 * to right before bnx2x_ack_sb). In this case there
3115 			 * will never be another interrupt until there is
3116 			 * another update of the status block, while there
3117 			 * is still unhandled work.
3118 			 */
3119 			rmb();
3120 
3121 			if (!(bnx2x_has_rx_work(fp) || bnx2x_has_tx_work(fp))) {
3122 				napi_complete(napi);
3123 				/* Re-enable interrupts */
3124 				DP(NETIF_MSG_RX_STATUS,
3125 				   "Update index to %d\n", fp->fp_hc_idx);
3126 				bnx2x_ack_sb(bp, fp->igu_sb_id, USTORM_ID,
3127 					     le16_to_cpu(fp->fp_hc_idx),
3128 					     IGU_INT_ENABLE, 1);
3129 				break;
3130 			}
3131 		}
3132 	}
3133 
3134 	return work_done;
3135 }
3136 
3137 #ifdef CONFIG_NET_RX_BUSY_POLL
3138 /* must be called with local_bh_disable()d */
3139 int bnx2x_low_latency_recv(struct napi_struct *napi)
3140 {
3141 	struct bnx2x_fastpath *fp = container_of(napi, struct bnx2x_fastpath,
3142 						 napi);
3143 	struct bnx2x *bp = fp->bp;
3144 	int found = 0;
3145 
3146 	if ((bp->state == BNX2X_STATE_CLOSED) ||
3147 	    (bp->state == BNX2X_STATE_ERROR) ||
3148 	    (bp->flags & (TPA_ENABLE_FLAG | GRO_ENABLE_FLAG)))
3149 		return LL_FLUSH_FAILED;
3150 
3151 	if (!bnx2x_fp_lock_poll(fp))
3152 		return LL_FLUSH_BUSY;
3153 
3154 	if (bnx2x_has_rx_work(fp))
3155 		found = bnx2x_rx_int(fp, 4);
3156 
3157 	bnx2x_fp_unlock_poll(fp);
3158 
3159 	return found;
3160 }
3161 #endif
3162 
3163 /* we split the first BD into headers and data BDs
3164  * to ease the pain of our fellow microcode engineers
3165  * we use one mapping for both BDs
3166  */
3167 static u16 bnx2x_tx_split(struct bnx2x *bp,
3168 			  struct bnx2x_fp_txdata *txdata,
3169 			  struct sw_tx_bd *tx_buf,
3170 			  struct eth_tx_start_bd **tx_bd, u16 hlen,
3171 			  u16 bd_prod)
3172 {
3173 	struct eth_tx_start_bd *h_tx_bd = *tx_bd;
3174 	struct eth_tx_bd *d_tx_bd;
3175 	dma_addr_t mapping;
3176 	int old_len = le16_to_cpu(h_tx_bd->nbytes);
3177 
3178 	/* first fix first BD */
3179 	h_tx_bd->nbytes = cpu_to_le16(hlen);
3180 
3181 	DP(NETIF_MSG_TX_QUEUED,	"TSO split header size is %d (%x:%x)\n",
3182 	   h_tx_bd->nbytes, h_tx_bd->addr_hi, h_tx_bd->addr_lo);
3183 
3184 	/* now get a new data BD
3185 	 * (after the pbd) and fill it */
3186 	bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3187 	d_tx_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
3188 
3189 	mapping = HILO_U64(le32_to_cpu(h_tx_bd->addr_hi),
3190 			   le32_to_cpu(h_tx_bd->addr_lo)) + hlen;
3191 
3192 	d_tx_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
3193 	d_tx_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
3194 	d_tx_bd->nbytes = cpu_to_le16(old_len - hlen);
3195 
3196 	/* this marks the BD as one that has no individual mapping */
3197 	tx_buf->flags |= BNX2X_TSO_SPLIT_BD;
3198 
3199 	DP(NETIF_MSG_TX_QUEUED,
3200 	   "TSO split data size is %d (%x:%x)\n",
3201 	   d_tx_bd->nbytes, d_tx_bd->addr_hi, d_tx_bd->addr_lo);
3202 
3203 	/* update tx_bd */
3204 	*tx_bd = (struct eth_tx_start_bd *)d_tx_bd;
3205 
3206 	return bd_prod;
3207 }
3208 
3209 #define bswab32(b32) ((__force __le32) swab32((__force __u32) (b32)))
3210 #define bswab16(b16) ((__force __le16) swab16((__force __u16) (b16)))
3211 static __le16 bnx2x_csum_fix(unsigned char *t_header, u16 csum, s8 fix)
3212 {
3213 	__sum16 tsum = (__force __sum16) csum;
3214 
3215 	if (fix > 0)
3216 		tsum = ~csum_fold(csum_sub((__force __wsum) csum,
3217 				  csum_partial(t_header - fix, fix, 0)));
3218 
3219 	else if (fix < 0)
3220 		tsum = ~csum_fold(csum_add((__force __wsum) csum,
3221 				  csum_partial(t_header, -fix, 0)));
3222 
3223 	return bswab16(tsum);
3224 }
3225 
3226 static u32 bnx2x_xmit_type(struct bnx2x *bp, struct sk_buff *skb)
3227 {
3228 	u32 rc;
3229 	__u8 prot = 0;
3230 	__be16 protocol;
3231 
3232 	if (skb->ip_summed != CHECKSUM_PARTIAL)
3233 		return XMIT_PLAIN;
3234 
3235 	protocol = vlan_get_protocol(skb);
3236 	if (protocol == htons(ETH_P_IPV6)) {
3237 		rc = XMIT_CSUM_V6;
3238 		prot = ipv6_hdr(skb)->nexthdr;
3239 	} else {
3240 		rc = XMIT_CSUM_V4;
3241 		prot = ip_hdr(skb)->protocol;
3242 	}
3243 
3244 	if (!CHIP_IS_E1x(bp) && skb->encapsulation) {
3245 		if (inner_ip_hdr(skb)->version == 6) {
3246 			rc |= XMIT_CSUM_ENC_V6;
3247 			if (inner_ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3248 				rc |= XMIT_CSUM_TCP;
3249 		} else {
3250 			rc |= XMIT_CSUM_ENC_V4;
3251 			if (inner_ip_hdr(skb)->protocol == IPPROTO_TCP)
3252 				rc |= XMIT_CSUM_TCP;
3253 		}
3254 	}
3255 	if (prot == IPPROTO_TCP)
3256 		rc |= XMIT_CSUM_TCP;
3257 
3258 	if (skb_is_gso_v6(skb)) {
3259 		rc |= (XMIT_GSO_V6 | XMIT_CSUM_TCP);
3260 		if (rc & XMIT_CSUM_ENC)
3261 			rc |= XMIT_GSO_ENC_V6;
3262 	} else if (skb_is_gso(skb)) {
3263 		rc |= (XMIT_GSO_V4 | XMIT_CSUM_TCP);
3264 		if (rc & XMIT_CSUM_ENC)
3265 			rc |= XMIT_GSO_ENC_V4;
3266 	}
3267 
3268 	return rc;
3269 }
3270 
3271 #if (MAX_SKB_FRAGS >= MAX_FETCH_BD - 3)
3272 /* check if packet requires linearization (packet is too fragmented)
3273    no need to check fragmentation if page size > 8K (there will be no
3274    violation to FW restrictions) */
3275 static int bnx2x_pkt_req_lin(struct bnx2x *bp, struct sk_buff *skb,
3276 			     u32 xmit_type)
3277 {
3278 	int to_copy = 0;
3279 	int hlen = 0;
3280 	int first_bd_sz = 0;
3281 
3282 	/* 3 = 1 (for linear data BD) + 2 (for PBD and last BD) */
3283 	if (skb_shinfo(skb)->nr_frags >= (MAX_FETCH_BD - 3)) {
3284 
3285 		if (xmit_type & XMIT_GSO) {
3286 			unsigned short lso_mss = skb_shinfo(skb)->gso_size;
3287 			/* Check if LSO packet needs to be copied:
3288 			   3 = 1 (for headers BD) + 2 (for PBD and last BD) */
3289 			int wnd_size = MAX_FETCH_BD - 3;
3290 			/* Number of windows to check */
3291 			int num_wnds = skb_shinfo(skb)->nr_frags - wnd_size;
3292 			int wnd_idx = 0;
3293 			int frag_idx = 0;
3294 			u32 wnd_sum = 0;
3295 
3296 			/* Headers length */
3297 			hlen = (int)(skb_transport_header(skb) - skb->data) +
3298 				tcp_hdrlen(skb);
3299 
3300 			/* Amount of data (w/o headers) on linear part of SKB*/
3301 			first_bd_sz = skb_headlen(skb) - hlen;
3302 
3303 			wnd_sum  = first_bd_sz;
3304 
3305 			/* Calculate the first sum - it's special */
3306 			for (frag_idx = 0; frag_idx < wnd_size - 1; frag_idx++)
3307 				wnd_sum +=
3308 					skb_frag_size(&skb_shinfo(skb)->frags[frag_idx]);
3309 
3310 			/* If there was data on linear skb data - check it */
3311 			if (first_bd_sz > 0) {
3312 				if (unlikely(wnd_sum < lso_mss)) {
3313 					to_copy = 1;
3314 					goto exit_lbl;
3315 				}
3316 
3317 				wnd_sum -= first_bd_sz;
3318 			}
3319 
3320 			/* Others are easier: run through the frag list and
3321 			   check all windows */
3322 			for (wnd_idx = 0; wnd_idx <= num_wnds; wnd_idx++) {
3323 				wnd_sum +=
3324 			  skb_frag_size(&skb_shinfo(skb)->frags[wnd_idx + wnd_size - 1]);
3325 
3326 				if (unlikely(wnd_sum < lso_mss)) {
3327 					to_copy = 1;
3328 					break;
3329 				}
3330 				wnd_sum -=
3331 					skb_frag_size(&skb_shinfo(skb)->frags[wnd_idx]);
3332 			}
3333 		} else {
3334 			/* in non-LSO too fragmented packet should always
3335 			   be linearized */
3336 			to_copy = 1;
3337 		}
3338 	}
3339 
3340 exit_lbl:
3341 	if (unlikely(to_copy))
3342 		DP(NETIF_MSG_TX_QUEUED,
3343 		   "Linearization IS REQUIRED for %s packet. num_frags %d  hlen %d  first_bd_sz %d\n",
3344 		   (xmit_type & XMIT_GSO) ? "LSO" : "non-LSO",
3345 		   skb_shinfo(skb)->nr_frags, hlen, first_bd_sz);
3346 
3347 	return to_copy;
3348 }
3349 #endif
3350 
3351 static void bnx2x_set_pbd_gso_e2(struct sk_buff *skb, u32 *parsing_data,
3352 				 u32 xmit_type)
3353 {
3354 	struct ipv6hdr *ipv6;
3355 
3356 	*parsing_data |= (skb_shinfo(skb)->gso_size <<
3357 			      ETH_TX_PARSE_BD_E2_LSO_MSS_SHIFT) &
3358 			      ETH_TX_PARSE_BD_E2_LSO_MSS;
3359 
3360 	if (xmit_type & XMIT_GSO_ENC_V6)
3361 		ipv6 = inner_ipv6_hdr(skb);
3362 	else if (xmit_type & XMIT_GSO_V6)
3363 		ipv6 = ipv6_hdr(skb);
3364 	else
3365 		ipv6 = NULL;
3366 
3367 	if (ipv6 && ipv6->nexthdr == NEXTHDR_IPV6)
3368 		*parsing_data |= ETH_TX_PARSE_BD_E2_IPV6_WITH_EXT_HDR;
3369 }
3370 
3371 /**
3372  * bnx2x_set_pbd_gso - update PBD in GSO case.
3373  *
3374  * @skb:	packet skb
3375  * @pbd:	parse BD
3376  * @xmit_type:	xmit flags
3377  */
3378 static void bnx2x_set_pbd_gso(struct sk_buff *skb,
3379 			      struct eth_tx_parse_bd_e1x *pbd,
3380 			      struct eth_tx_start_bd *tx_start_bd,
3381 			      u32 xmit_type)
3382 {
3383 	pbd->lso_mss = cpu_to_le16(skb_shinfo(skb)->gso_size);
3384 	pbd->tcp_send_seq = bswab32(tcp_hdr(skb)->seq);
3385 	pbd->tcp_flags = pbd_tcp_flags(tcp_hdr(skb));
3386 
3387 	if (xmit_type & XMIT_GSO_V4) {
3388 		pbd->ip_id = bswab16(ip_hdr(skb)->id);
3389 		pbd->tcp_pseudo_csum =
3390 			bswab16(~csum_tcpudp_magic(ip_hdr(skb)->saddr,
3391 						   ip_hdr(skb)->daddr,
3392 						   0, IPPROTO_TCP, 0));
3393 
3394 		/* GSO on 57710/57711 needs FW to calculate IP checksum */
3395 		tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_IP_CSUM;
3396 	} else {
3397 		pbd->tcp_pseudo_csum =
3398 			bswab16(~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3399 						 &ipv6_hdr(skb)->daddr,
3400 						 0, IPPROTO_TCP, 0));
3401 	}
3402 
3403 	pbd->global_data |=
3404 		cpu_to_le16(ETH_TX_PARSE_BD_E1X_PSEUDO_CS_WITHOUT_LEN);
3405 }
3406 
3407 /**
3408  * bnx2x_set_pbd_csum_enc - update PBD with checksum and return header length
3409  *
3410  * @bp:			driver handle
3411  * @skb:		packet skb
3412  * @parsing_data:	data to be updated
3413  * @xmit_type:		xmit flags
3414  *
3415  * 57712/578xx related, when skb has encapsulation
3416  */
3417 static u8 bnx2x_set_pbd_csum_enc(struct bnx2x *bp, struct sk_buff *skb,
3418 				 u32 *parsing_data, u32 xmit_type)
3419 {
3420 	*parsing_data |=
3421 		((((u8 *)skb_inner_transport_header(skb) - skb->data) >> 1) <<
3422 		ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W_SHIFT) &
3423 		ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W;
3424 
3425 	if (xmit_type & XMIT_CSUM_TCP) {
3426 		*parsing_data |= ((inner_tcp_hdrlen(skb) / 4) <<
3427 			ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW_SHIFT) &
3428 			ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW;
3429 
3430 		return skb_inner_transport_header(skb) +
3431 			inner_tcp_hdrlen(skb) - skb->data;
3432 	}
3433 
3434 	/* We support checksum offload for TCP and UDP only.
3435 	 * No need to pass the UDP header length - it's a constant.
3436 	 */
3437 	return skb_inner_transport_header(skb) +
3438 		sizeof(struct udphdr) - skb->data;
3439 }
3440 
3441 /**
3442  * bnx2x_set_pbd_csum_e2 - update PBD with checksum and return header length
3443  *
3444  * @bp:			driver handle
3445  * @skb:		packet skb
3446  * @parsing_data:	data to be updated
3447  * @xmit_type:		xmit flags
3448  *
3449  * 57712/578xx related
3450  */
3451 static u8 bnx2x_set_pbd_csum_e2(struct bnx2x *bp, struct sk_buff *skb,
3452 				u32 *parsing_data, u32 xmit_type)
3453 {
3454 	*parsing_data |=
3455 		((((u8 *)skb_transport_header(skb) - skb->data) >> 1) <<
3456 		ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W_SHIFT) &
3457 		ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W;
3458 
3459 	if (xmit_type & XMIT_CSUM_TCP) {
3460 		*parsing_data |= ((tcp_hdrlen(skb) / 4) <<
3461 			ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW_SHIFT) &
3462 			ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW;
3463 
3464 		return skb_transport_header(skb) + tcp_hdrlen(skb) - skb->data;
3465 	}
3466 	/* We support checksum offload for TCP and UDP only.
3467 	 * No need to pass the UDP header length - it's a constant.
3468 	 */
3469 	return skb_transport_header(skb) + sizeof(struct udphdr) - skb->data;
3470 }
3471 
3472 /* set FW indication according to inner or outer protocols if tunneled */
3473 static void bnx2x_set_sbd_csum(struct bnx2x *bp, struct sk_buff *skb,
3474 			       struct eth_tx_start_bd *tx_start_bd,
3475 			       u32 xmit_type)
3476 {
3477 	tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_L4_CSUM;
3478 
3479 	if (xmit_type & (XMIT_CSUM_ENC_V6 | XMIT_CSUM_V6))
3480 		tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_IPV6;
3481 
3482 	if (!(xmit_type & XMIT_CSUM_TCP))
3483 		tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_IS_UDP;
3484 }
3485 
3486 /**
3487  * bnx2x_set_pbd_csum - update PBD with checksum and return header length
3488  *
3489  * @bp:		driver handle
3490  * @skb:	packet skb
3491  * @pbd:	parse BD to be updated
3492  * @xmit_type:	xmit flags
3493  */
3494 static u8 bnx2x_set_pbd_csum(struct bnx2x *bp, struct sk_buff *skb,
3495 			     struct eth_tx_parse_bd_e1x *pbd,
3496 			     u32 xmit_type)
3497 {
3498 	u8 hlen = (skb_network_header(skb) - skb->data) >> 1;
3499 
3500 	/* for now NS flag is not used in Linux */
3501 	pbd->global_data =
3502 		cpu_to_le16(hlen |
3503 			    ((skb->protocol == cpu_to_be16(ETH_P_8021Q)) <<
3504 			     ETH_TX_PARSE_BD_E1X_LLC_SNAP_EN_SHIFT));
3505 
3506 	pbd->ip_hlen_w = (skb_transport_header(skb) -
3507 			skb_network_header(skb)) >> 1;
3508 
3509 	hlen += pbd->ip_hlen_w;
3510 
3511 	/* We support checksum offload for TCP and UDP only */
3512 	if (xmit_type & XMIT_CSUM_TCP)
3513 		hlen += tcp_hdrlen(skb) / 2;
3514 	else
3515 		hlen += sizeof(struct udphdr) / 2;
3516 
3517 	pbd->total_hlen_w = cpu_to_le16(hlen);
3518 	hlen = hlen*2;
3519 
3520 	if (xmit_type & XMIT_CSUM_TCP) {
3521 		pbd->tcp_pseudo_csum = bswab16(tcp_hdr(skb)->check);
3522 
3523 	} else {
3524 		s8 fix = SKB_CS_OFF(skb); /* signed! */
3525 
3526 		DP(NETIF_MSG_TX_QUEUED,
3527 		   "hlen %d  fix %d  csum before fix %x\n",
3528 		   le16_to_cpu(pbd->total_hlen_w), fix, SKB_CS(skb));
3529 
3530 		/* HW bug: fixup the CSUM */
3531 		pbd->tcp_pseudo_csum =
3532 			bnx2x_csum_fix(skb_transport_header(skb),
3533 				       SKB_CS(skb), fix);
3534 
3535 		DP(NETIF_MSG_TX_QUEUED, "csum after fix %x\n",
3536 		   pbd->tcp_pseudo_csum);
3537 	}
3538 
3539 	return hlen;
3540 }
3541 
3542 static void bnx2x_update_pbds_gso_enc(struct sk_buff *skb,
3543 				      struct eth_tx_parse_bd_e2 *pbd_e2,
3544 				      struct eth_tx_parse_2nd_bd *pbd2,
3545 				      u16 *global_data,
3546 				      u32 xmit_type)
3547 {
3548 	u16 hlen_w = 0;
3549 	u8 outerip_off, outerip_len = 0;
3550 
3551 	/* from outer IP to transport */
3552 	hlen_w = (skb_inner_transport_header(skb) -
3553 		  skb_network_header(skb)) >> 1;
3554 
3555 	/* transport len */
3556 	hlen_w += inner_tcp_hdrlen(skb) >> 1;
3557 
3558 	pbd2->fw_ip_hdr_to_payload_w = hlen_w;
3559 
3560 	/* outer IP header info */
3561 	if (xmit_type & XMIT_CSUM_V4) {
3562 		struct iphdr *iph = ip_hdr(skb);
3563 		u32 csum = (__force u32)(~iph->check) -
3564 			   (__force u32)iph->tot_len -
3565 			   (__force u32)iph->frag_off;
3566 
3567 		pbd2->fw_ip_csum_wo_len_flags_frag =
3568 			bswab16(csum_fold((__force __wsum)csum));
3569 	} else {
3570 		pbd2->fw_ip_hdr_to_payload_w =
3571 			hlen_w - ((sizeof(struct ipv6hdr)) >> 1);
3572 	}
3573 
3574 	pbd2->tcp_send_seq = bswab32(inner_tcp_hdr(skb)->seq);
3575 
3576 	pbd2->tcp_flags = pbd_tcp_flags(inner_tcp_hdr(skb));
3577 
3578 	if (xmit_type & XMIT_GSO_V4) {
3579 		pbd2->hw_ip_id = bswab16(inner_ip_hdr(skb)->id);
3580 
3581 		pbd_e2->data.tunnel_data.pseudo_csum =
3582 			bswab16(~csum_tcpudp_magic(
3583 					inner_ip_hdr(skb)->saddr,
3584 					inner_ip_hdr(skb)->daddr,
3585 					0, IPPROTO_TCP, 0));
3586 
3587 		outerip_len = ip_hdr(skb)->ihl << 1;
3588 	} else {
3589 		pbd_e2->data.tunnel_data.pseudo_csum =
3590 			bswab16(~csum_ipv6_magic(
3591 					&inner_ipv6_hdr(skb)->saddr,
3592 					&inner_ipv6_hdr(skb)->daddr,
3593 					0, IPPROTO_TCP, 0));
3594 	}
3595 
3596 	outerip_off = (skb_network_header(skb) - skb->data) >> 1;
3597 
3598 	*global_data |=
3599 		outerip_off |
3600 		(!!(xmit_type & XMIT_CSUM_V6) <<
3601 			ETH_TX_PARSE_2ND_BD_IP_HDR_TYPE_OUTER_SHIFT) |
3602 		(outerip_len <<
3603 			ETH_TX_PARSE_2ND_BD_IP_HDR_LEN_OUTER_W_SHIFT) |
3604 		((skb->protocol == cpu_to_be16(ETH_P_8021Q)) <<
3605 			ETH_TX_PARSE_2ND_BD_LLC_SNAP_EN_SHIFT);
3606 
3607 	if (ip_hdr(skb)->protocol == IPPROTO_UDP) {
3608 		SET_FLAG(*global_data, ETH_TX_PARSE_2ND_BD_TUNNEL_UDP_EXIST, 1);
3609 		pbd2->tunnel_udp_hdr_start_w = skb_transport_offset(skb) >> 1;
3610 	}
3611 }
3612 
3613 /* called with netif_tx_lock
3614  * bnx2x_tx_int() runs without netif_tx_lock unless it needs to call
3615  * netif_wake_queue()
3616  */
3617 netdev_tx_t bnx2x_start_xmit(struct sk_buff *skb, struct net_device *dev)
3618 {
3619 	struct bnx2x *bp = netdev_priv(dev);
3620 
3621 	struct netdev_queue *txq;
3622 	struct bnx2x_fp_txdata *txdata;
3623 	struct sw_tx_bd *tx_buf;
3624 	struct eth_tx_start_bd *tx_start_bd, *first_bd;
3625 	struct eth_tx_bd *tx_data_bd, *total_pkt_bd = NULL;
3626 	struct eth_tx_parse_bd_e1x *pbd_e1x = NULL;
3627 	struct eth_tx_parse_bd_e2 *pbd_e2 = NULL;
3628 	struct eth_tx_parse_2nd_bd *pbd2 = NULL;
3629 	u32 pbd_e2_parsing_data = 0;
3630 	u16 pkt_prod, bd_prod;
3631 	int nbd, txq_index;
3632 	dma_addr_t mapping;
3633 	u32 xmit_type = bnx2x_xmit_type(bp, skb);
3634 	int i;
3635 	u8 hlen = 0;
3636 	__le16 pkt_size = 0;
3637 	struct ethhdr *eth;
3638 	u8 mac_type = UNICAST_ADDRESS;
3639 
3640 #ifdef BNX2X_STOP_ON_ERROR
3641 	if (unlikely(bp->panic))
3642 		return NETDEV_TX_BUSY;
3643 #endif
3644 
3645 	txq_index = skb_get_queue_mapping(skb);
3646 	txq = netdev_get_tx_queue(dev, txq_index);
3647 
3648 	BUG_ON(txq_index >= MAX_ETH_TXQ_IDX(bp) + (CNIC_LOADED(bp) ? 1 : 0));
3649 
3650 	txdata = &bp->bnx2x_txq[txq_index];
3651 
3652 	/* enable this debug print to view the transmission queue being used
3653 	DP(NETIF_MSG_TX_QUEUED, "indices: txq %d, fp %d, txdata %d\n",
3654 	   txq_index, fp_index, txdata_index); */
3655 
3656 	/* enable this debug print to view the transmission details
3657 	DP(NETIF_MSG_TX_QUEUED,
3658 	   "transmitting packet cid %d fp index %d txdata_index %d tx_data ptr %p fp pointer %p\n",
3659 	   txdata->cid, fp_index, txdata_index, txdata, fp); */
3660 
3661 	if (unlikely(bnx2x_tx_avail(bp, txdata) <
3662 			skb_shinfo(skb)->nr_frags +
3663 			BDS_PER_TX_PKT +
3664 			NEXT_CNT_PER_TX_PKT(MAX_BDS_PER_TX_PKT))) {
3665 		/* Handle special storage cases separately */
3666 		if (txdata->tx_ring_size == 0) {
3667 			struct bnx2x_eth_q_stats *q_stats =
3668 				bnx2x_fp_qstats(bp, txdata->parent_fp);
3669 			q_stats->driver_filtered_tx_pkt++;
3670 			dev_kfree_skb(skb);
3671 			return NETDEV_TX_OK;
3672 		}
3673 		bnx2x_fp_qstats(bp, txdata->parent_fp)->driver_xoff++;
3674 		netif_tx_stop_queue(txq);
3675 		BNX2X_ERR("BUG! Tx ring full when queue awake!\n");
3676 
3677 		return NETDEV_TX_BUSY;
3678 	}
3679 
3680 	DP(NETIF_MSG_TX_QUEUED,
3681 	   "queue[%d]: SKB: summed %x  protocol %x protocol(%x,%x) gso type %x  xmit_type %x len %d\n",
3682 	   txq_index, skb->ip_summed, skb->protocol, ipv6_hdr(skb)->nexthdr,
3683 	   ip_hdr(skb)->protocol, skb_shinfo(skb)->gso_type, xmit_type,
3684 	   skb->len);
3685 
3686 	eth = (struct ethhdr *)skb->data;
3687 
3688 	/* set flag according to packet type (UNICAST_ADDRESS is default)*/
3689 	if (unlikely(is_multicast_ether_addr(eth->h_dest))) {
3690 		if (is_broadcast_ether_addr(eth->h_dest))
3691 			mac_type = BROADCAST_ADDRESS;
3692 		else
3693 			mac_type = MULTICAST_ADDRESS;
3694 	}
3695 
3696 #if (MAX_SKB_FRAGS >= MAX_FETCH_BD - BDS_PER_TX_PKT)
3697 	/* First, check if we need to linearize the skb (due to FW
3698 	   restrictions). No need to check fragmentation if page size > 8K
3699 	   (there will be no violation to FW restrictions) */
3700 	if (bnx2x_pkt_req_lin(bp, skb, xmit_type)) {
3701 		/* Statistics of linearization */
3702 		bp->lin_cnt++;
3703 		if (skb_linearize(skb) != 0) {
3704 			DP(NETIF_MSG_TX_QUEUED,
3705 			   "SKB linearization failed - silently dropping this SKB\n");
3706 			dev_kfree_skb_any(skb);
3707 			return NETDEV_TX_OK;
3708 		}
3709 	}
3710 #endif
3711 	/* Map skb linear data for DMA */
3712 	mapping = dma_map_single(&bp->pdev->dev, skb->data,
3713 				 skb_headlen(skb), DMA_TO_DEVICE);
3714 	if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
3715 		DP(NETIF_MSG_TX_QUEUED,
3716 		   "SKB mapping failed - silently dropping this SKB\n");
3717 		dev_kfree_skb_any(skb);
3718 		return NETDEV_TX_OK;
3719 	}
3720 	/*
3721 	Please read carefully. First we use one BD which we mark as start,
3722 	then we have a parsing info BD (used for TSO or xsum),
3723 	and only then we have the rest of the TSO BDs.
3724 	(don't forget to mark the last one as last,
3725 	and to unmap only AFTER you write to the BD ...)
3726 	And above all, all pdb sizes are in words - NOT DWORDS!
3727 	*/
3728 
3729 	/* get current pkt produced now - advance it just before sending packet
3730 	 * since mapping of pages may fail and cause packet to be dropped
3731 	 */
3732 	pkt_prod = txdata->tx_pkt_prod;
3733 	bd_prod = TX_BD(txdata->tx_bd_prod);
3734 
3735 	/* get a tx_buf and first BD
3736 	 * tx_start_bd may be changed during SPLIT,
3737 	 * but first_bd will always stay first
3738 	 */
3739 	tx_buf = &txdata->tx_buf_ring[TX_BD(pkt_prod)];
3740 	tx_start_bd = &txdata->tx_desc_ring[bd_prod].start_bd;
3741 	first_bd = tx_start_bd;
3742 
3743 	tx_start_bd->bd_flags.as_bitfield = ETH_TX_BD_FLAGS_START_BD;
3744 
3745 	/* header nbd: indirectly zero other flags! */
3746 	tx_start_bd->general_data = 1 << ETH_TX_START_BD_HDR_NBDS_SHIFT;
3747 
3748 	/* remember the first BD of the packet */
3749 	tx_buf->first_bd = txdata->tx_bd_prod;
3750 	tx_buf->skb = skb;
3751 	tx_buf->flags = 0;
3752 
3753 	DP(NETIF_MSG_TX_QUEUED,
3754 	   "sending pkt %u @%p  next_idx %u  bd %u @%p\n",
3755 	   pkt_prod, tx_buf, txdata->tx_pkt_prod, bd_prod, tx_start_bd);
3756 
3757 	if (vlan_tx_tag_present(skb)) {
3758 		tx_start_bd->vlan_or_ethertype =
3759 		    cpu_to_le16(vlan_tx_tag_get(skb));
3760 		tx_start_bd->bd_flags.as_bitfield |=
3761 		    (X_ETH_OUTBAND_VLAN << ETH_TX_BD_FLAGS_VLAN_MODE_SHIFT);
3762 	} else {
3763 		/* when transmitting in a vf, start bd must hold the ethertype
3764 		 * for fw to enforce it
3765 		 */
3766 		if (IS_VF(bp))
3767 			tx_start_bd->vlan_or_ethertype =
3768 				cpu_to_le16(ntohs(eth->h_proto));
3769 		else
3770 			/* used by FW for packet accounting */
3771 			tx_start_bd->vlan_or_ethertype = cpu_to_le16(pkt_prod);
3772 	}
3773 
3774 	nbd = 2; /* start_bd + pbd + frags (updated when pages are mapped) */
3775 
3776 	/* turn on parsing and get a BD */
3777 	bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3778 
3779 	if (xmit_type & XMIT_CSUM)
3780 		bnx2x_set_sbd_csum(bp, skb, tx_start_bd, xmit_type);
3781 
3782 	if (!CHIP_IS_E1x(bp)) {
3783 		pbd_e2 = &txdata->tx_desc_ring[bd_prod].parse_bd_e2;
3784 		memset(pbd_e2, 0, sizeof(struct eth_tx_parse_bd_e2));
3785 
3786 		if (xmit_type & XMIT_CSUM_ENC) {
3787 			u16 global_data = 0;
3788 
3789 			/* Set PBD in enc checksum offload case */
3790 			hlen = bnx2x_set_pbd_csum_enc(bp, skb,
3791 						      &pbd_e2_parsing_data,
3792 						      xmit_type);
3793 
3794 			/* turn on 2nd parsing and get a BD */
3795 			bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3796 
3797 			pbd2 = &txdata->tx_desc_ring[bd_prod].parse_2nd_bd;
3798 
3799 			memset(pbd2, 0, sizeof(*pbd2));
3800 
3801 			pbd_e2->data.tunnel_data.ip_hdr_start_inner_w =
3802 				(skb_inner_network_header(skb) -
3803 				 skb->data) >> 1;
3804 
3805 			if (xmit_type & XMIT_GSO_ENC)
3806 				bnx2x_update_pbds_gso_enc(skb, pbd_e2, pbd2,
3807 							  &global_data,
3808 							  xmit_type);
3809 
3810 			pbd2->global_data = cpu_to_le16(global_data);
3811 
3812 			/* add addition parse BD indication to start BD */
3813 			SET_FLAG(tx_start_bd->general_data,
3814 				 ETH_TX_START_BD_PARSE_NBDS, 1);
3815 			/* set encapsulation flag in start BD */
3816 			SET_FLAG(tx_start_bd->general_data,
3817 				 ETH_TX_START_BD_TUNNEL_EXIST, 1);
3818 			nbd++;
3819 		} else if (xmit_type & XMIT_CSUM) {
3820 			/* Set PBD in checksum offload case w/o encapsulation */
3821 			hlen = bnx2x_set_pbd_csum_e2(bp, skb,
3822 						     &pbd_e2_parsing_data,
3823 						     xmit_type);
3824 		}
3825 
3826 		/* Add the macs to the parsing BD this is a vf */
3827 		if (IS_VF(bp)) {
3828 			/* override GRE parameters in BD */
3829 			bnx2x_set_fw_mac_addr(&pbd_e2->data.mac_addr.src_hi,
3830 					      &pbd_e2->data.mac_addr.src_mid,
3831 					      &pbd_e2->data.mac_addr.src_lo,
3832 					      eth->h_source);
3833 
3834 			bnx2x_set_fw_mac_addr(&pbd_e2->data.mac_addr.dst_hi,
3835 					      &pbd_e2->data.mac_addr.dst_mid,
3836 					      &pbd_e2->data.mac_addr.dst_lo,
3837 					      eth->h_dest);
3838 		}
3839 
3840 		SET_FLAG(pbd_e2_parsing_data,
3841 			 ETH_TX_PARSE_BD_E2_ETH_ADDR_TYPE, mac_type);
3842 	} else {
3843 		u16 global_data = 0;
3844 		pbd_e1x = &txdata->tx_desc_ring[bd_prod].parse_bd_e1x;
3845 		memset(pbd_e1x, 0, sizeof(struct eth_tx_parse_bd_e1x));
3846 		/* Set PBD in checksum offload case */
3847 		if (xmit_type & XMIT_CSUM)
3848 			hlen = bnx2x_set_pbd_csum(bp, skb, pbd_e1x, xmit_type);
3849 
3850 		SET_FLAG(global_data,
3851 			 ETH_TX_PARSE_BD_E1X_ETH_ADDR_TYPE, mac_type);
3852 		pbd_e1x->global_data |= cpu_to_le16(global_data);
3853 	}
3854 
3855 	/* Setup the data pointer of the first BD of the packet */
3856 	tx_start_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
3857 	tx_start_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
3858 	tx_start_bd->nbytes = cpu_to_le16(skb_headlen(skb));
3859 	pkt_size = tx_start_bd->nbytes;
3860 
3861 	DP(NETIF_MSG_TX_QUEUED,
3862 	   "first bd @%p  addr (%x:%x)  nbytes %d  flags %x  vlan %x\n",
3863 	   tx_start_bd, tx_start_bd->addr_hi, tx_start_bd->addr_lo,
3864 	   le16_to_cpu(tx_start_bd->nbytes),
3865 	   tx_start_bd->bd_flags.as_bitfield,
3866 	   le16_to_cpu(tx_start_bd->vlan_or_ethertype));
3867 
3868 	if (xmit_type & XMIT_GSO) {
3869 
3870 		DP(NETIF_MSG_TX_QUEUED,
3871 		   "TSO packet len %d  hlen %d  total len %d  tso size %d\n",
3872 		   skb->len, hlen, skb_headlen(skb),
3873 		   skb_shinfo(skb)->gso_size);
3874 
3875 		tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_SW_LSO;
3876 
3877 		if (unlikely(skb_headlen(skb) > hlen)) {
3878 			nbd++;
3879 			bd_prod = bnx2x_tx_split(bp, txdata, tx_buf,
3880 						 &tx_start_bd, hlen,
3881 						 bd_prod);
3882 		}
3883 		if (!CHIP_IS_E1x(bp))
3884 			bnx2x_set_pbd_gso_e2(skb, &pbd_e2_parsing_data,
3885 					     xmit_type);
3886 		else
3887 			bnx2x_set_pbd_gso(skb, pbd_e1x, first_bd, xmit_type);
3888 	}
3889 
3890 	/* Set the PBD's parsing_data field if not zero
3891 	 * (for the chips newer than 57711).
3892 	 */
3893 	if (pbd_e2_parsing_data)
3894 		pbd_e2->parsing_data = cpu_to_le32(pbd_e2_parsing_data);
3895 
3896 	tx_data_bd = (struct eth_tx_bd *)tx_start_bd;
3897 
3898 	/* Handle fragmented skb */
3899 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3900 		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3901 
3902 		mapping = skb_frag_dma_map(&bp->pdev->dev, frag, 0,
3903 					   skb_frag_size(frag), DMA_TO_DEVICE);
3904 		if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
3905 			unsigned int pkts_compl = 0, bytes_compl = 0;
3906 
3907 			DP(NETIF_MSG_TX_QUEUED,
3908 			   "Unable to map page - dropping packet...\n");
3909 
3910 			/* we need unmap all buffers already mapped
3911 			 * for this SKB;
3912 			 * first_bd->nbd need to be properly updated
3913 			 * before call to bnx2x_free_tx_pkt
3914 			 */
3915 			first_bd->nbd = cpu_to_le16(nbd);
3916 			bnx2x_free_tx_pkt(bp, txdata,
3917 					  TX_BD(txdata->tx_pkt_prod),
3918 					  &pkts_compl, &bytes_compl);
3919 			return NETDEV_TX_OK;
3920 		}
3921 
3922 		bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3923 		tx_data_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
3924 		if (total_pkt_bd == NULL)
3925 			total_pkt_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
3926 
3927 		tx_data_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
3928 		tx_data_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
3929 		tx_data_bd->nbytes = cpu_to_le16(skb_frag_size(frag));
3930 		le16_add_cpu(&pkt_size, skb_frag_size(frag));
3931 		nbd++;
3932 
3933 		DP(NETIF_MSG_TX_QUEUED,
3934 		   "frag %d  bd @%p  addr (%x:%x)  nbytes %d\n",
3935 		   i, tx_data_bd, tx_data_bd->addr_hi, tx_data_bd->addr_lo,
3936 		   le16_to_cpu(tx_data_bd->nbytes));
3937 	}
3938 
3939 	DP(NETIF_MSG_TX_QUEUED, "last bd @%p\n", tx_data_bd);
3940 
3941 	/* update with actual num BDs */
3942 	first_bd->nbd = cpu_to_le16(nbd);
3943 
3944 	bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3945 
3946 	/* now send a tx doorbell, counting the next BD
3947 	 * if the packet contains or ends with it
3948 	 */
3949 	if (TX_BD_POFF(bd_prod) < nbd)
3950 		nbd++;
3951 
3952 	/* total_pkt_bytes should be set on the first data BD if
3953 	 * it's not an LSO packet and there is more than one
3954 	 * data BD. In this case pkt_size is limited by an MTU value.
3955 	 * However we prefer to set it for an LSO packet (while we don't
3956 	 * have to) in order to save some CPU cycles in a none-LSO
3957 	 * case, when we much more care about them.
3958 	 */
3959 	if (total_pkt_bd != NULL)
3960 		total_pkt_bd->total_pkt_bytes = pkt_size;
3961 
3962 	if (pbd_e1x)
3963 		DP(NETIF_MSG_TX_QUEUED,
3964 		   "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",
3965 		   pbd_e1x, pbd_e1x->global_data, pbd_e1x->ip_hlen_w,
3966 		   pbd_e1x->ip_id, pbd_e1x->lso_mss, pbd_e1x->tcp_flags,
3967 		   pbd_e1x->tcp_pseudo_csum, pbd_e1x->tcp_send_seq,
3968 		    le16_to_cpu(pbd_e1x->total_hlen_w));
3969 	if (pbd_e2)
3970 		DP(NETIF_MSG_TX_QUEUED,
3971 		   "PBD (E2) @%p  dst %x %x %x src %x %x %x parsing_data %x\n",
3972 		   pbd_e2,
3973 		   pbd_e2->data.mac_addr.dst_hi,
3974 		   pbd_e2->data.mac_addr.dst_mid,
3975 		   pbd_e2->data.mac_addr.dst_lo,
3976 		   pbd_e2->data.mac_addr.src_hi,
3977 		   pbd_e2->data.mac_addr.src_mid,
3978 		   pbd_e2->data.mac_addr.src_lo,
3979 		   pbd_e2->parsing_data);
3980 	DP(NETIF_MSG_TX_QUEUED, "doorbell: nbd %d  bd %u\n", nbd, bd_prod);
3981 
3982 	netdev_tx_sent_queue(txq, skb->len);
3983 
3984 	skb_tx_timestamp(skb);
3985 
3986 	txdata->tx_pkt_prod++;
3987 	/*
3988 	 * Make sure that the BD data is updated before updating the producer
3989 	 * since FW might read the BD right after the producer is updated.
3990 	 * This is only applicable for weak-ordered memory model archs such
3991 	 * as IA-64. The following barrier is also mandatory since FW will
3992 	 * assumes packets must have BDs.
3993 	 */
3994 	wmb();
3995 
3996 	txdata->tx_db.data.prod += nbd;
3997 	barrier();
3998 
3999 	DOORBELL(bp, txdata->cid, txdata->tx_db.raw);
4000 
4001 	mmiowb();
4002 
4003 	txdata->tx_bd_prod += nbd;
4004 
4005 	if (unlikely(bnx2x_tx_avail(bp, txdata) < MAX_DESC_PER_TX_PKT)) {
4006 		netif_tx_stop_queue(txq);
4007 
4008 		/* paired memory barrier is in bnx2x_tx_int(), we have to keep
4009 		 * ordering of set_bit() in netif_tx_stop_queue() and read of
4010 		 * fp->bd_tx_cons */
4011 		smp_mb();
4012 
4013 		bnx2x_fp_qstats(bp, txdata->parent_fp)->driver_xoff++;
4014 		if (bnx2x_tx_avail(bp, txdata) >= MAX_DESC_PER_TX_PKT)
4015 			netif_tx_wake_queue(txq);
4016 	}
4017 	txdata->tx_pkt++;
4018 
4019 	return NETDEV_TX_OK;
4020 }
4021 
4022 /**
4023  * bnx2x_setup_tc - routine to configure net_device for multi tc
4024  *
4025  * @netdev: net device to configure
4026  * @tc: number of traffic classes to enable
4027  *
4028  * callback connected to the ndo_setup_tc function pointer
4029  */
4030 int bnx2x_setup_tc(struct net_device *dev, u8 num_tc)
4031 {
4032 	int cos, prio, count, offset;
4033 	struct bnx2x *bp = netdev_priv(dev);
4034 
4035 	/* setup tc must be called under rtnl lock */
4036 	ASSERT_RTNL();
4037 
4038 	/* no traffic classes requested. Aborting */
4039 	if (!num_tc) {
4040 		netdev_reset_tc(dev);
4041 		return 0;
4042 	}
4043 
4044 	/* requested to support too many traffic classes */
4045 	if (num_tc > bp->max_cos) {
4046 		BNX2X_ERR("support for too many traffic classes requested: %d. Max supported is %d\n",
4047 			  num_tc, bp->max_cos);
4048 		return -EINVAL;
4049 	}
4050 
4051 	/* declare amount of supported traffic classes */
4052 	if (netdev_set_num_tc(dev, num_tc)) {
4053 		BNX2X_ERR("failed to declare %d traffic classes\n", num_tc);
4054 		return -EINVAL;
4055 	}
4056 
4057 	/* configure priority to traffic class mapping */
4058 	for (prio = 0; prio < BNX2X_MAX_PRIORITY; prio++) {
4059 		netdev_set_prio_tc_map(dev, prio, bp->prio_to_cos[prio]);
4060 		DP(BNX2X_MSG_SP | NETIF_MSG_IFUP,
4061 		   "mapping priority %d to tc %d\n",
4062 		   prio, bp->prio_to_cos[prio]);
4063 	}
4064 
4065 	/* Use this configuration to differentiate tc0 from other COSes
4066 	   This can be used for ets or pfc, and save the effort of setting
4067 	   up a multio class queue disc or negotiating DCBX with a switch
4068 	netdev_set_prio_tc_map(dev, 0, 0);
4069 	DP(BNX2X_MSG_SP, "mapping priority %d to tc %d\n", 0, 0);
4070 	for (prio = 1; prio < 16; prio++) {
4071 		netdev_set_prio_tc_map(dev, prio, 1);
4072 		DP(BNX2X_MSG_SP, "mapping priority %d to tc %d\n", prio, 1);
4073 	} */
4074 
4075 	/* configure traffic class to transmission queue mapping */
4076 	for (cos = 0; cos < bp->max_cos; cos++) {
4077 		count = BNX2X_NUM_ETH_QUEUES(bp);
4078 		offset = cos * BNX2X_NUM_NON_CNIC_QUEUES(bp);
4079 		netdev_set_tc_queue(dev, cos, count, offset);
4080 		DP(BNX2X_MSG_SP | NETIF_MSG_IFUP,
4081 		   "mapping tc %d to offset %d count %d\n",
4082 		   cos, offset, count);
4083 	}
4084 
4085 	return 0;
4086 }
4087 
4088 /* called with rtnl_lock */
4089 int bnx2x_change_mac_addr(struct net_device *dev, void *p)
4090 {
4091 	struct sockaddr *addr = p;
4092 	struct bnx2x *bp = netdev_priv(dev);
4093 	int rc = 0;
4094 
4095 	if (!bnx2x_is_valid_ether_addr(bp, addr->sa_data)) {
4096 		BNX2X_ERR("Requested MAC address is not valid\n");
4097 		return -EINVAL;
4098 	}
4099 
4100 	if ((IS_MF_STORAGE_SD(bp) || IS_MF_FCOE_AFEX(bp)) &&
4101 	    !is_zero_ether_addr(addr->sa_data)) {
4102 		BNX2X_ERR("Can't configure non-zero address on iSCSI or FCoE functions in MF-SD mode\n");
4103 		return -EINVAL;
4104 	}
4105 
4106 	if (netif_running(dev))  {
4107 		rc = bnx2x_set_eth_mac(bp, false);
4108 		if (rc)
4109 			return rc;
4110 	}
4111 
4112 	memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
4113 
4114 	if (netif_running(dev))
4115 		rc = bnx2x_set_eth_mac(bp, true);
4116 
4117 	return rc;
4118 }
4119 
4120 static void bnx2x_free_fp_mem_at(struct bnx2x *bp, int fp_index)
4121 {
4122 	union host_hc_status_block *sb = &bnx2x_fp(bp, fp_index, status_blk);
4123 	struct bnx2x_fastpath *fp = &bp->fp[fp_index];
4124 	u8 cos;
4125 
4126 	/* Common */
4127 
4128 	if (IS_FCOE_IDX(fp_index)) {
4129 		memset(sb, 0, sizeof(union host_hc_status_block));
4130 		fp->status_blk_mapping = 0;
4131 	} else {
4132 		/* status blocks */
4133 		if (!CHIP_IS_E1x(bp))
4134 			BNX2X_PCI_FREE(sb->e2_sb,
4135 				       bnx2x_fp(bp, fp_index,
4136 						status_blk_mapping),
4137 				       sizeof(struct host_hc_status_block_e2));
4138 		else
4139 			BNX2X_PCI_FREE(sb->e1x_sb,
4140 				       bnx2x_fp(bp, fp_index,
4141 						status_blk_mapping),
4142 				       sizeof(struct host_hc_status_block_e1x));
4143 	}
4144 
4145 	/* Rx */
4146 	if (!skip_rx_queue(bp, fp_index)) {
4147 		bnx2x_free_rx_bds(fp);
4148 
4149 		/* fastpath rx rings: rx_buf rx_desc rx_comp */
4150 		BNX2X_FREE(bnx2x_fp(bp, fp_index, rx_buf_ring));
4151 		BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_desc_ring),
4152 			       bnx2x_fp(bp, fp_index, rx_desc_mapping),
4153 			       sizeof(struct eth_rx_bd) * NUM_RX_BD);
4154 
4155 		BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_comp_ring),
4156 			       bnx2x_fp(bp, fp_index, rx_comp_mapping),
4157 			       sizeof(struct eth_fast_path_rx_cqe) *
4158 			       NUM_RCQ_BD);
4159 
4160 		/* SGE ring */
4161 		BNX2X_FREE(bnx2x_fp(bp, fp_index, rx_page_ring));
4162 		BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_sge_ring),
4163 			       bnx2x_fp(bp, fp_index, rx_sge_mapping),
4164 			       BCM_PAGE_SIZE * NUM_RX_SGE_PAGES);
4165 	}
4166 
4167 	/* Tx */
4168 	if (!skip_tx_queue(bp, fp_index)) {
4169 		/* fastpath tx rings: tx_buf tx_desc */
4170 		for_each_cos_in_tx_queue(fp, cos) {
4171 			struct bnx2x_fp_txdata *txdata = fp->txdata_ptr[cos];
4172 
4173 			DP(NETIF_MSG_IFDOWN,
4174 			   "freeing tx memory of fp %d cos %d cid %d\n",
4175 			   fp_index, cos, txdata->cid);
4176 
4177 			BNX2X_FREE(txdata->tx_buf_ring);
4178 			BNX2X_PCI_FREE(txdata->tx_desc_ring,
4179 				txdata->tx_desc_mapping,
4180 				sizeof(union eth_tx_bd_types) * NUM_TX_BD);
4181 		}
4182 	}
4183 	/* end of fastpath */
4184 }
4185 
4186 void bnx2x_free_fp_mem_cnic(struct bnx2x *bp)
4187 {
4188 	int i;
4189 	for_each_cnic_queue(bp, i)
4190 		bnx2x_free_fp_mem_at(bp, i);
4191 }
4192 
4193 void bnx2x_free_fp_mem(struct bnx2x *bp)
4194 {
4195 	int i;
4196 	for_each_eth_queue(bp, i)
4197 		bnx2x_free_fp_mem_at(bp, i);
4198 }
4199 
4200 static void set_sb_shortcuts(struct bnx2x *bp, int index)
4201 {
4202 	union host_hc_status_block status_blk = bnx2x_fp(bp, index, status_blk);
4203 	if (!CHIP_IS_E1x(bp)) {
4204 		bnx2x_fp(bp, index, sb_index_values) =
4205 			(__le16 *)status_blk.e2_sb->sb.index_values;
4206 		bnx2x_fp(bp, index, sb_running_index) =
4207 			(__le16 *)status_blk.e2_sb->sb.running_index;
4208 	} else {
4209 		bnx2x_fp(bp, index, sb_index_values) =
4210 			(__le16 *)status_blk.e1x_sb->sb.index_values;
4211 		bnx2x_fp(bp, index, sb_running_index) =
4212 			(__le16 *)status_blk.e1x_sb->sb.running_index;
4213 	}
4214 }
4215 
4216 /* Returns the number of actually allocated BDs */
4217 static int bnx2x_alloc_rx_bds(struct bnx2x_fastpath *fp,
4218 			      int rx_ring_size)
4219 {
4220 	struct bnx2x *bp = fp->bp;
4221 	u16 ring_prod, cqe_ring_prod;
4222 	int i, failure_cnt = 0;
4223 
4224 	fp->rx_comp_cons = 0;
4225 	cqe_ring_prod = ring_prod = 0;
4226 
4227 	/* This routine is called only during fo init so
4228 	 * fp->eth_q_stats.rx_skb_alloc_failed = 0
4229 	 */
4230 	for (i = 0; i < rx_ring_size; i++) {
4231 		if (bnx2x_alloc_rx_data(bp, fp, ring_prod, GFP_KERNEL) < 0) {
4232 			failure_cnt++;
4233 			continue;
4234 		}
4235 		ring_prod = NEXT_RX_IDX(ring_prod);
4236 		cqe_ring_prod = NEXT_RCQ_IDX(cqe_ring_prod);
4237 		WARN_ON(ring_prod <= (i - failure_cnt));
4238 	}
4239 
4240 	if (failure_cnt)
4241 		BNX2X_ERR("was only able to allocate %d rx skbs on queue[%d]\n",
4242 			  i - failure_cnt, fp->index);
4243 
4244 	fp->rx_bd_prod = ring_prod;
4245 	/* Limit the CQE producer by the CQE ring size */
4246 	fp->rx_comp_prod = min_t(u16, NUM_RCQ_RINGS*RCQ_DESC_CNT,
4247 			       cqe_ring_prod);
4248 	fp->rx_pkt = fp->rx_calls = 0;
4249 
4250 	bnx2x_fp_stats(bp, fp)->eth_q_stats.rx_skb_alloc_failed += failure_cnt;
4251 
4252 	return i - failure_cnt;
4253 }
4254 
4255 static void bnx2x_set_next_page_rx_cq(struct bnx2x_fastpath *fp)
4256 {
4257 	int i;
4258 
4259 	for (i = 1; i <= NUM_RCQ_RINGS; i++) {
4260 		struct eth_rx_cqe_next_page *nextpg;
4261 
4262 		nextpg = (struct eth_rx_cqe_next_page *)
4263 			&fp->rx_comp_ring[RCQ_DESC_CNT * i - 1];
4264 		nextpg->addr_hi =
4265 			cpu_to_le32(U64_HI(fp->rx_comp_mapping +
4266 				   BCM_PAGE_SIZE*(i % NUM_RCQ_RINGS)));
4267 		nextpg->addr_lo =
4268 			cpu_to_le32(U64_LO(fp->rx_comp_mapping +
4269 				   BCM_PAGE_SIZE*(i % NUM_RCQ_RINGS)));
4270 	}
4271 }
4272 
4273 static int bnx2x_alloc_fp_mem_at(struct bnx2x *bp, int index)
4274 {
4275 	union host_hc_status_block *sb;
4276 	struct bnx2x_fastpath *fp = &bp->fp[index];
4277 	int ring_size = 0;
4278 	u8 cos;
4279 	int rx_ring_size = 0;
4280 
4281 	if (!bp->rx_ring_size &&
4282 	    (IS_MF_STORAGE_SD(bp) || IS_MF_FCOE_AFEX(bp))) {
4283 		rx_ring_size = MIN_RX_SIZE_NONTPA;
4284 		bp->rx_ring_size = rx_ring_size;
4285 	} else if (!bp->rx_ring_size) {
4286 		rx_ring_size = MAX_RX_AVAIL/BNX2X_NUM_RX_QUEUES(bp);
4287 
4288 		if (CHIP_IS_E3(bp)) {
4289 			u32 cfg = SHMEM_RD(bp,
4290 					   dev_info.port_hw_config[BP_PORT(bp)].
4291 					   default_cfg);
4292 
4293 			/* Decrease ring size for 1G functions */
4294 			if ((cfg & PORT_HW_CFG_NET_SERDES_IF_MASK) ==
4295 			    PORT_HW_CFG_NET_SERDES_IF_SGMII)
4296 				rx_ring_size /= 10;
4297 		}
4298 
4299 		/* allocate at least number of buffers required by FW */
4300 		rx_ring_size = max_t(int, bp->disable_tpa ? MIN_RX_SIZE_NONTPA :
4301 				     MIN_RX_SIZE_TPA, rx_ring_size);
4302 
4303 		bp->rx_ring_size = rx_ring_size;
4304 	} else /* if rx_ring_size specified - use it */
4305 		rx_ring_size = bp->rx_ring_size;
4306 
4307 	DP(BNX2X_MSG_SP, "calculated rx_ring_size %d\n", rx_ring_size);
4308 
4309 	/* Common */
4310 	sb = &bnx2x_fp(bp, index, status_blk);
4311 
4312 	if (!IS_FCOE_IDX(index)) {
4313 		/* status blocks */
4314 		if (!CHIP_IS_E1x(bp))
4315 			BNX2X_PCI_ALLOC(sb->e2_sb,
4316 				&bnx2x_fp(bp, index, status_blk_mapping),
4317 				sizeof(struct host_hc_status_block_e2));
4318 		else
4319 			BNX2X_PCI_ALLOC(sb->e1x_sb,
4320 				&bnx2x_fp(bp, index, status_blk_mapping),
4321 			    sizeof(struct host_hc_status_block_e1x));
4322 	}
4323 
4324 	/* FCoE Queue uses Default SB and doesn't ACK the SB, thus no need to
4325 	 * set shortcuts for it.
4326 	 */
4327 	if (!IS_FCOE_IDX(index))
4328 		set_sb_shortcuts(bp, index);
4329 
4330 	/* Tx */
4331 	if (!skip_tx_queue(bp, index)) {
4332 		/* fastpath tx rings: tx_buf tx_desc */
4333 		for_each_cos_in_tx_queue(fp, cos) {
4334 			struct bnx2x_fp_txdata *txdata = fp->txdata_ptr[cos];
4335 
4336 			DP(NETIF_MSG_IFUP,
4337 			   "allocating tx memory of fp %d cos %d\n",
4338 			   index, cos);
4339 
4340 			BNX2X_ALLOC(txdata->tx_buf_ring,
4341 				sizeof(struct sw_tx_bd) * NUM_TX_BD);
4342 			BNX2X_PCI_ALLOC(txdata->tx_desc_ring,
4343 				&txdata->tx_desc_mapping,
4344 				sizeof(union eth_tx_bd_types) * NUM_TX_BD);
4345 		}
4346 	}
4347 
4348 	/* Rx */
4349 	if (!skip_rx_queue(bp, index)) {
4350 		/* fastpath rx rings: rx_buf rx_desc rx_comp */
4351 		BNX2X_ALLOC(bnx2x_fp(bp, index, rx_buf_ring),
4352 				sizeof(struct sw_rx_bd) * NUM_RX_BD);
4353 		BNX2X_PCI_ALLOC(bnx2x_fp(bp, index, rx_desc_ring),
4354 				&bnx2x_fp(bp, index, rx_desc_mapping),
4355 				sizeof(struct eth_rx_bd) * NUM_RX_BD);
4356 
4357 		/* Seed all CQEs by 1s */
4358 		BNX2X_PCI_FALLOC(bnx2x_fp(bp, index, rx_comp_ring),
4359 				 &bnx2x_fp(bp, index, rx_comp_mapping),
4360 				 sizeof(struct eth_fast_path_rx_cqe) *
4361 				 NUM_RCQ_BD);
4362 
4363 		/* SGE ring */
4364 		BNX2X_ALLOC(bnx2x_fp(bp, index, rx_page_ring),
4365 				sizeof(struct sw_rx_page) * NUM_RX_SGE);
4366 		BNX2X_PCI_ALLOC(bnx2x_fp(bp, index, rx_sge_ring),
4367 				&bnx2x_fp(bp, index, rx_sge_mapping),
4368 				BCM_PAGE_SIZE * NUM_RX_SGE_PAGES);
4369 		/* RX BD ring */
4370 		bnx2x_set_next_page_rx_bd(fp);
4371 
4372 		/* CQ ring */
4373 		bnx2x_set_next_page_rx_cq(fp);
4374 
4375 		/* BDs */
4376 		ring_size = bnx2x_alloc_rx_bds(fp, rx_ring_size);
4377 		if (ring_size < rx_ring_size)
4378 			goto alloc_mem_err;
4379 	}
4380 
4381 	return 0;
4382 
4383 /* handles low memory cases */
4384 alloc_mem_err:
4385 	BNX2X_ERR("Unable to allocate full memory for queue %d (size %d)\n",
4386 						index, ring_size);
4387 	/* FW will drop all packets if queue is not big enough,
4388 	 * In these cases we disable the queue
4389 	 * Min size is different for OOO, TPA and non-TPA queues
4390 	 */
4391 	if (ring_size < (fp->disable_tpa ?
4392 				MIN_RX_SIZE_NONTPA : MIN_RX_SIZE_TPA)) {
4393 			/* release memory allocated for this queue */
4394 			bnx2x_free_fp_mem_at(bp, index);
4395 			return -ENOMEM;
4396 	}
4397 	return 0;
4398 }
4399 
4400 int bnx2x_alloc_fp_mem_cnic(struct bnx2x *bp)
4401 {
4402 	if (!NO_FCOE(bp))
4403 		/* FCoE */
4404 		if (bnx2x_alloc_fp_mem_at(bp, FCOE_IDX(bp)))
4405 			/* we will fail load process instead of mark
4406 			 * NO_FCOE_FLAG
4407 			 */
4408 			return -ENOMEM;
4409 
4410 	return 0;
4411 }
4412 
4413 int bnx2x_alloc_fp_mem(struct bnx2x *bp)
4414 {
4415 	int i;
4416 
4417 	/* 1. Allocate FP for leading - fatal if error
4418 	 * 2. Allocate RSS - fix number of queues if error
4419 	 */
4420 
4421 	/* leading */
4422 	if (bnx2x_alloc_fp_mem_at(bp, 0))
4423 		return -ENOMEM;
4424 
4425 	/* RSS */
4426 	for_each_nondefault_eth_queue(bp, i)
4427 		if (bnx2x_alloc_fp_mem_at(bp, i))
4428 			break;
4429 
4430 	/* handle memory failures */
4431 	if (i != BNX2X_NUM_ETH_QUEUES(bp)) {
4432 		int delta = BNX2X_NUM_ETH_QUEUES(bp) - i;
4433 
4434 		WARN_ON(delta < 0);
4435 		bnx2x_shrink_eth_fp(bp, delta);
4436 		if (CNIC_SUPPORT(bp))
4437 			/* move non eth FPs next to last eth FP
4438 			 * must be done in that order
4439 			 * FCOE_IDX < FWD_IDX < OOO_IDX
4440 			 */
4441 
4442 			/* move FCoE fp even NO_FCOE_FLAG is on */
4443 			bnx2x_move_fp(bp, FCOE_IDX(bp), FCOE_IDX(bp) - delta);
4444 		bp->num_ethernet_queues -= delta;
4445 		bp->num_queues = bp->num_ethernet_queues +
4446 				 bp->num_cnic_queues;
4447 		BNX2X_ERR("Adjusted num of queues from %d to %d\n",
4448 			  bp->num_queues + delta, bp->num_queues);
4449 	}
4450 
4451 	return 0;
4452 }
4453 
4454 void bnx2x_free_mem_bp(struct bnx2x *bp)
4455 {
4456 	int i;
4457 
4458 	for (i = 0; i < bp->fp_array_size; i++)
4459 		kfree(bp->fp[i].tpa_info);
4460 	kfree(bp->fp);
4461 	kfree(bp->sp_objs);
4462 	kfree(bp->fp_stats);
4463 	kfree(bp->bnx2x_txq);
4464 	kfree(bp->msix_table);
4465 	kfree(bp->ilt);
4466 }
4467 
4468 int bnx2x_alloc_mem_bp(struct bnx2x *bp)
4469 {
4470 	struct bnx2x_fastpath *fp;
4471 	struct msix_entry *tbl;
4472 	struct bnx2x_ilt *ilt;
4473 	int msix_table_size = 0;
4474 	int fp_array_size, txq_array_size;
4475 	int i;
4476 
4477 	/*
4478 	 * The biggest MSI-X table we might need is as a maximum number of fast
4479 	 * path IGU SBs plus default SB (for PF only).
4480 	 */
4481 	msix_table_size = bp->igu_sb_cnt;
4482 	if (IS_PF(bp))
4483 		msix_table_size++;
4484 	BNX2X_DEV_INFO("msix_table_size %d\n", msix_table_size);
4485 
4486 	/* fp array: RSS plus CNIC related L2 queues */
4487 	fp_array_size = BNX2X_MAX_RSS_COUNT(bp) + CNIC_SUPPORT(bp);
4488 	bp->fp_array_size = fp_array_size;
4489 	BNX2X_DEV_INFO("fp_array_size %d\n", bp->fp_array_size);
4490 
4491 	fp = kcalloc(bp->fp_array_size, sizeof(*fp), GFP_KERNEL);
4492 	if (!fp)
4493 		goto alloc_err;
4494 	for (i = 0; i < bp->fp_array_size; i++) {
4495 		fp[i].tpa_info =
4496 			kcalloc(ETH_MAX_AGGREGATION_QUEUES_E1H_E2,
4497 				sizeof(struct bnx2x_agg_info), GFP_KERNEL);
4498 		if (!(fp[i].tpa_info))
4499 			goto alloc_err;
4500 	}
4501 
4502 	bp->fp = fp;
4503 
4504 	/* allocate sp objs */
4505 	bp->sp_objs = kcalloc(bp->fp_array_size, sizeof(struct bnx2x_sp_objs),
4506 			      GFP_KERNEL);
4507 	if (!bp->sp_objs)
4508 		goto alloc_err;
4509 
4510 	/* allocate fp_stats */
4511 	bp->fp_stats = kcalloc(bp->fp_array_size, sizeof(struct bnx2x_fp_stats),
4512 			       GFP_KERNEL);
4513 	if (!bp->fp_stats)
4514 		goto alloc_err;
4515 
4516 	/* Allocate memory for the transmission queues array */
4517 	txq_array_size =
4518 		BNX2X_MAX_RSS_COUNT(bp) * BNX2X_MULTI_TX_COS + CNIC_SUPPORT(bp);
4519 	BNX2X_DEV_INFO("txq_array_size %d", txq_array_size);
4520 
4521 	bp->bnx2x_txq = kcalloc(txq_array_size, sizeof(struct bnx2x_fp_txdata),
4522 				GFP_KERNEL);
4523 	if (!bp->bnx2x_txq)
4524 		goto alloc_err;
4525 
4526 	/* msix table */
4527 	tbl = kcalloc(msix_table_size, sizeof(*tbl), GFP_KERNEL);
4528 	if (!tbl)
4529 		goto alloc_err;
4530 	bp->msix_table = tbl;
4531 
4532 	/* ilt */
4533 	ilt = kzalloc(sizeof(*ilt), GFP_KERNEL);
4534 	if (!ilt)
4535 		goto alloc_err;
4536 	bp->ilt = ilt;
4537 
4538 	return 0;
4539 alloc_err:
4540 	bnx2x_free_mem_bp(bp);
4541 	return -ENOMEM;
4542 }
4543 
4544 int bnx2x_reload_if_running(struct net_device *dev)
4545 {
4546 	struct bnx2x *bp = netdev_priv(dev);
4547 
4548 	if (unlikely(!netif_running(dev)))
4549 		return 0;
4550 
4551 	bnx2x_nic_unload(bp, UNLOAD_NORMAL, true);
4552 	return bnx2x_nic_load(bp, LOAD_NORMAL);
4553 }
4554 
4555 int bnx2x_get_cur_phy_idx(struct bnx2x *bp)
4556 {
4557 	u32 sel_phy_idx = 0;
4558 	if (bp->link_params.num_phys <= 1)
4559 		return INT_PHY;
4560 
4561 	if (bp->link_vars.link_up) {
4562 		sel_phy_idx = EXT_PHY1;
4563 		/* In case link is SERDES, check if the EXT_PHY2 is the one */
4564 		if ((bp->link_vars.link_status & LINK_STATUS_SERDES_LINK) &&
4565 		    (bp->link_params.phy[EXT_PHY2].supported & SUPPORTED_FIBRE))
4566 			sel_phy_idx = EXT_PHY2;
4567 	} else {
4568 
4569 		switch (bnx2x_phy_selection(&bp->link_params)) {
4570 		case PORT_HW_CFG_PHY_SELECTION_HARDWARE_DEFAULT:
4571 		case PORT_HW_CFG_PHY_SELECTION_FIRST_PHY:
4572 		case PORT_HW_CFG_PHY_SELECTION_FIRST_PHY_PRIORITY:
4573 		       sel_phy_idx = EXT_PHY1;
4574 		       break;
4575 		case PORT_HW_CFG_PHY_SELECTION_SECOND_PHY:
4576 		case PORT_HW_CFG_PHY_SELECTION_SECOND_PHY_PRIORITY:
4577 		       sel_phy_idx = EXT_PHY2;
4578 		       break;
4579 		}
4580 	}
4581 
4582 	return sel_phy_idx;
4583 }
4584 int bnx2x_get_link_cfg_idx(struct bnx2x *bp)
4585 {
4586 	u32 sel_phy_idx = bnx2x_get_cur_phy_idx(bp);
4587 	/*
4588 	 * The selected activated PHY is always after swapping (in case PHY
4589 	 * swapping is enabled). So when swapping is enabled, we need to reverse
4590 	 * the configuration
4591 	 */
4592 
4593 	if (bp->link_params.multi_phy_config &
4594 	    PORT_HW_CFG_PHY_SWAPPED_ENABLED) {
4595 		if (sel_phy_idx == EXT_PHY1)
4596 			sel_phy_idx = EXT_PHY2;
4597 		else if (sel_phy_idx == EXT_PHY2)
4598 			sel_phy_idx = EXT_PHY1;
4599 	}
4600 	return LINK_CONFIG_IDX(sel_phy_idx);
4601 }
4602 
4603 #ifdef NETDEV_FCOE_WWNN
4604 int bnx2x_fcoe_get_wwn(struct net_device *dev, u64 *wwn, int type)
4605 {
4606 	struct bnx2x *bp = netdev_priv(dev);
4607 	struct cnic_eth_dev *cp = &bp->cnic_eth_dev;
4608 
4609 	switch (type) {
4610 	case NETDEV_FCOE_WWNN:
4611 		*wwn = HILO_U64(cp->fcoe_wwn_node_name_hi,
4612 				cp->fcoe_wwn_node_name_lo);
4613 		break;
4614 	case NETDEV_FCOE_WWPN:
4615 		*wwn = HILO_U64(cp->fcoe_wwn_port_name_hi,
4616 				cp->fcoe_wwn_port_name_lo);
4617 		break;
4618 	default:
4619 		BNX2X_ERR("Wrong WWN type requested - %d\n", type);
4620 		return -EINVAL;
4621 	}
4622 
4623 	return 0;
4624 }
4625 #endif
4626 
4627 /* called with rtnl_lock */
4628 int bnx2x_change_mtu(struct net_device *dev, int new_mtu)
4629 {
4630 	struct bnx2x *bp = netdev_priv(dev);
4631 
4632 	if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
4633 		BNX2X_ERR("Can't perform change MTU during parity recovery\n");
4634 		return -EAGAIN;
4635 	}
4636 
4637 	if ((new_mtu > ETH_MAX_JUMBO_PACKET_SIZE) ||
4638 	    ((new_mtu + ETH_HLEN) < ETH_MIN_PACKET_SIZE)) {
4639 		BNX2X_ERR("Can't support requested MTU size\n");
4640 		return -EINVAL;
4641 	}
4642 
4643 	/* This does not race with packet allocation
4644 	 * because the actual alloc size is
4645 	 * only updated as part of load
4646 	 */
4647 	dev->mtu = new_mtu;
4648 
4649 	return bnx2x_reload_if_running(dev);
4650 }
4651 
4652 netdev_features_t bnx2x_fix_features(struct net_device *dev,
4653 				     netdev_features_t features)
4654 {
4655 	struct bnx2x *bp = netdev_priv(dev);
4656 
4657 	/* TPA requires Rx CSUM offloading */
4658 	if (!(features & NETIF_F_RXCSUM) || bp->disable_tpa) {
4659 		features &= ~NETIF_F_LRO;
4660 		features &= ~NETIF_F_GRO;
4661 	}
4662 
4663 	return features;
4664 }
4665 
4666 int bnx2x_set_features(struct net_device *dev, netdev_features_t features)
4667 {
4668 	struct bnx2x *bp = netdev_priv(dev);
4669 	u32 flags = bp->flags;
4670 	u32 changes;
4671 	bool bnx2x_reload = false;
4672 
4673 	if (features & NETIF_F_LRO)
4674 		flags |= TPA_ENABLE_FLAG;
4675 	else
4676 		flags &= ~TPA_ENABLE_FLAG;
4677 
4678 	if (features & NETIF_F_GRO)
4679 		flags |= GRO_ENABLE_FLAG;
4680 	else
4681 		flags &= ~GRO_ENABLE_FLAG;
4682 
4683 	if (features & NETIF_F_LOOPBACK) {
4684 		if (bp->link_params.loopback_mode != LOOPBACK_BMAC) {
4685 			bp->link_params.loopback_mode = LOOPBACK_BMAC;
4686 			bnx2x_reload = true;
4687 		}
4688 	} else {
4689 		if (bp->link_params.loopback_mode != LOOPBACK_NONE) {
4690 			bp->link_params.loopback_mode = LOOPBACK_NONE;
4691 			bnx2x_reload = true;
4692 		}
4693 	}
4694 
4695 	changes = flags ^ bp->flags;
4696 
4697 	/* if GRO is changed while LRO is enabled, don't force a reload */
4698 	if ((changes & GRO_ENABLE_FLAG) && (flags & TPA_ENABLE_FLAG))
4699 		changes &= ~GRO_ENABLE_FLAG;
4700 
4701 	if (changes)
4702 		bnx2x_reload = true;
4703 
4704 	bp->flags = flags;
4705 
4706 	if (bnx2x_reload) {
4707 		if (bp->recovery_state == BNX2X_RECOVERY_DONE)
4708 			return bnx2x_reload_if_running(dev);
4709 		/* else: bnx2x_nic_load() will be called at end of recovery */
4710 	}
4711 
4712 	return 0;
4713 }
4714 
4715 void bnx2x_tx_timeout(struct net_device *dev)
4716 {
4717 	struct bnx2x *bp = netdev_priv(dev);
4718 
4719 #ifdef BNX2X_STOP_ON_ERROR
4720 	if (!bp->panic)
4721 		bnx2x_panic();
4722 #endif
4723 
4724 	smp_mb__before_clear_bit();
4725 	set_bit(BNX2X_SP_RTNL_TX_TIMEOUT, &bp->sp_rtnl_state);
4726 	smp_mb__after_clear_bit();
4727 
4728 	/* This allows the netif to be shutdown gracefully before resetting */
4729 	schedule_delayed_work(&bp->sp_rtnl_task, 0);
4730 }
4731 
4732 int bnx2x_suspend(struct pci_dev *pdev, pm_message_t state)
4733 {
4734 	struct net_device *dev = pci_get_drvdata(pdev);
4735 	struct bnx2x *bp;
4736 
4737 	if (!dev) {
4738 		dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
4739 		return -ENODEV;
4740 	}
4741 	bp = netdev_priv(dev);
4742 
4743 	rtnl_lock();
4744 
4745 	pci_save_state(pdev);
4746 
4747 	if (!netif_running(dev)) {
4748 		rtnl_unlock();
4749 		return 0;
4750 	}
4751 
4752 	netif_device_detach(dev);
4753 
4754 	bnx2x_nic_unload(bp, UNLOAD_CLOSE, false);
4755 
4756 	bnx2x_set_power_state(bp, pci_choose_state(pdev, state));
4757 
4758 	rtnl_unlock();
4759 
4760 	return 0;
4761 }
4762 
4763 int bnx2x_resume(struct pci_dev *pdev)
4764 {
4765 	struct net_device *dev = pci_get_drvdata(pdev);
4766 	struct bnx2x *bp;
4767 	int rc;
4768 
4769 	if (!dev) {
4770 		dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
4771 		return -ENODEV;
4772 	}
4773 	bp = netdev_priv(dev);
4774 
4775 	if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
4776 		BNX2X_ERR("Handling parity error recovery. Try again later\n");
4777 		return -EAGAIN;
4778 	}
4779 
4780 	rtnl_lock();
4781 
4782 	pci_restore_state(pdev);
4783 
4784 	if (!netif_running(dev)) {
4785 		rtnl_unlock();
4786 		return 0;
4787 	}
4788 
4789 	bnx2x_set_power_state(bp, PCI_D0);
4790 	netif_device_attach(dev);
4791 
4792 	rc = bnx2x_nic_load(bp, LOAD_OPEN);
4793 
4794 	rtnl_unlock();
4795 
4796 	return rc;
4797 }
4798 
4799 void bnx2x_set_ctx_validation(struct bnx2x *bp, struct eth_context *cxt,
4800 			      u32 cid)
4801 {
4802 	if (!cxt) {
4803 		BNX2X_ERR("bad context pointer %p\n", cxt);
4804 		return;
4805 	}
4806 
4807 	/* ustorm cxt validation */
4808 	cxt->ustorm_ag_context.cdu_usage =
4809 		CDU_RSRVD_VALUE_TYPE_A(HW_CID(bp, cid),
4810 			CDU_REGION_NUMBER_UCM_AG, ETH_CONNECTION_TYPE);
4811 	/* xcontext validation */
4812 	cxt->xstorm_ag_context.cdu_reserved =
4813 		CDU_RSRVD_VALUE_TYPE_A(HW_CID(bp, cid),
4814 			CDU_REGION_NUMBER_XCM_AG, ETH_CONNECTION_TYPE);
4815 }
4816 
4817 static void storm_memset_hc_timeout(struct bnx2x *bp, u8 port,
4818 				    u8 fw_sb_id, u8 sb_index,
4819 				    u8 ticks)
4820 {
4821 	u32 addr = BAR_CSTRORM_INTMEM +
4822 		   CSTORM_STATUS_BLOCK_DATA_TIMEOUT_OFFSET(fw_sb_id, sb_index);
4823 	REG_WR8(bp, addr, ticks);
4824 	DP(NETIF_MSG_IFUP,
4825 	   "port %x fw_sb_id %d sb_index %d ticks %d\n",
4826 	   port, fw_sb_id, sb_index, ticks);
4827 }
4828 
4829 static void storm_memset_hc_disable(struct bnx2x *bp, u8 port,
4830 				    u16 fw_sb_id, u8 sb_index,
4831 				    u8 disable)
4832 {
4833 	u32 enable_flag = disable ? 0 : (1 << HC_INDEX_DATA_HC_ENABLED_SHIFT);
4834 	u32 addr = BAR_CSTRORM_INTMEM +
4835 		   CSTORM_STATUS_BLOCK_DATA_FLAGS_OFFSET(fw_sb_id, sb_index);
4836 	u8 flags = REG_RD8(bp, addr);
4837 	/* clear and set */
4838 	flags &= ~HC_INDEX_DATA_HC_ENABLED;
4839 	flags |= enable_flag;
4840 	REG_WR8(bp, addr, flags);
4841 	DP(NETIF_MSG_IFUP,
4842 	   "port %x fw_sb_id %d sb_index %d disable %d\n",
4843 	   port, fw_sb_id, sb_index, disable);
4844 }
4845 
4846 void bnx2x_update_coalesce_sb_index(struct bnx2x *bp, u8 fw_sb_id,
4847 				    u8 sb_index, u8 disable, u16 usec)
4848 {
4849 	int port = BP_PORT(bp);
4850 	u8 ticks = usec / BNX2X_BTR;
4851 
4852 	storm_memset_hc_timeout(bp, port, fw_sb_id, sb_index, ticks);
4853 
4854 	disable = disable ? 1 : (usec ? 0 : 1);
4855 	storm_memset_hc_disable(bp, port, fw_sb_id, sb_index, disable);
4856 }
4857