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 	skb_record_rx_queue(skb, fp->rx_queue);
685 	napi_gro_receive(&fp->napi, skb);
686 }
687 
688 static void bnx2x_tpa_stop(struct bnx2x *bp, struct bnx2x_fastpath *fp,
689 			   struct bnx2x_agg_info *tpa_info,
690 			   u16 pages,
691 			   struct eth_end_agg_rx_cqe *cqe,
692 			   u16 cqe_idx)
693 {
694 	struct sw_rx_bd *rx_buf = &tpa_info->first_buf;
695 	u8 pad = tpa_info->placement_offset;
696 	u16 len = tpa_info->len_on_bd;
697 	struct sk_buff *skb = NULL;
698 	u8 *new_data, *data = rx_buf->data;
699 	u8 old_tpa_state = tpa_info->tpa_state;
700 
701 	tpa_info->tpa_state = BNX2X_TPA_STOP;
702 
703 	/* If we there was an error during the handling of the TPA_START -
704 	 * drop this aggregation.
705 	 */
706 	if (old_tpa_state == BNX2X_TPA_ERROR)
707 		goto drop;
708 
709 	/* Try to allocate the new data */
710 	new_data = bnx2x_frag_alloc(fp, GFP_ATOMIC);
711 	/* Unmap skb in the pool anyway, as we are going to change
712 	   pool entry status to BNX2X_TPA_STOP even if new skb allocation
713 	   fails. */
714 	dma_unmap_single(&bp->pdev->dev, dma_unmap_addr(rx_buf, mapping),
715 			 fp->rx_buf_size, DMA_FROM_DEVICE);
716 	if (likely(new_data))
717 		skb = build_skb(data, fp->rx_frag_size);
718 
719 	if (likely(skb)) {
720 #ifdef BNX2X_STOP_ON_ERROR
721 		if (pad + len > fp->rx_buf_size) {
722 			BNX2X_ERR("skb_put is about to fail...  pad %d  len %d  rx_buf_size %d\n",
723 				  pad, len, fp->rx_buf_size);
724 			bnx2x_panic();
725 			return;
726 		}
727 #endif
728 
729 		skb_reserve(skb, pad + NET_SKB_PAD);
730 		skb_put(skb, len);
731 		skb->rxhash = tpa_info->rxhash;
732 		skb->l4_rxhash = tpa_info->l4_rxhash;
733 
734 		skb->protocol = eth_type_trans(skb, bp->dev);
735 		skb->ip_summed = CHECKSUM_UNNECESSARY;
736 
737 		if (!bnx2x_fill_frag_skb(bp, fp, tpa_info, pages,
738 					 skb, cqe, cqe_idx)) {
739 			if (tpa_info->parsing_flags & PARSING_FLAGS_VLAN)
740 				__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), tpa_info->vlan_tag);
741 			bnx2x_gro_receive(bp, fp, skb);
742 		} else {
743 			DP(NETIF_MSG_RX_STATUS,
744 			   "Failed to allocate new pages - dropping packet!\n");
745 			dev_kfree_skb_any(skb);
746 		}
747 
748 		/* put new data in bin */
749 		rx_buf->data = new_data;
750 
751 		return;
752 	}
753 	bnx2x_frag_free(fp, new_data);
754 drop:
755 	/* drop the packet and keep the buffer in the bin */
756 	DP(NETIF_MSG_RX_STATUS,
757 	   "Failed to allocate or map a new skb - dropping packet!\n");
758 	bnx2x_fp_stats(bp, fp)->eth_q_stats.rx_skb_alloc_failed++;
759 }
760 
761 static int bnx2x_alloc_rx_data(struct bnx2x *bp, struct bnx2x_fastpath *fp,
762 			       u16 index, gfp_t gfp_mask)
763 {
764 	u8 *data;
765 	struct sw_rx_bd *rx_buf = &fp->rx_buf_ring[index];
766 	struct eth_rx_bd *rx_bd = &fp->rx_desc_ring[index];
767 	dma_addr_t mapping;
768 
769 	data = bnx2x_frag_alloc(fp, gfp_mask);
770 	if (unlikely(data == NULL))
771 		return -ENOMEM;
772 
773 	mapping = dma_map_single(&bp->pdev->dev, data + NET_SKB_PAD,
774 				 fp->rx_buf_size,
775 				 DMA_FROM_DEVICE);
776 	if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
777 		bnx2x_frag_free(fp, data);
778 		BNX2X_ERR("Can't map rx data\n");
779 		return -ENOMEM;
780 	}
781 
782 	rx_buf->data = data;
783 	dma_unmap_addr_set(rx_buf, mapping, mapping);
784 
785 	rx_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
786 	rx_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
787 
788 	return 0;
789 }
790 
791 static
792 void bnx2x_csum_validate(struct sk_buff *skb, union eth_rx_cqe *cqe,
793 				 struct bnx2x_fastpath *fp,
794 				 struct bnx2x_eth_q_stats *qstats)
795 {
796 	/* Do nothing if no L4 csum validation was done.
797 	 * We do not check whether IP csum was validated. For IPv4 we assume
798 	 * that if the card got as far as validating the L4 csum, it also
799 	 * validated the IP csum. IPv6 has no IP csum.
800 	 */
801 	if (cqe->fast_path_cqe.status_flags &
802 	    ETH_FAST_PATH_RX_CQE_L4_XSUM_NO_VALIDATION_FLG)
803 		return;
804 
805 	/* If L4 validation was done, check if an error was found. */
806 
807 	if (cqe->fast_path_cqe.type_error_flags &
808 	    (ETH_FAST_PATH_RX_CQE_IP_BAD_XSUM_FLG |
809 	     ETH_FAST_PATH_RX_CQE_L4_BAD_XSUM_FLG))
810 		qstats->hw_csum_err++;
811 	else
812 		skb->ip_summed = CHECKSUM_UNNECESSARY;
813 }
814 
815 int bnx2x_rx_int(struct bnx2x_fastpath *fp, int budget)
816 {
817 	struct bnx2x *bp = fp->bp;
818 	u16 bd_cons, bd_prod, bd_prod_fw, comp_ring_cons;
819 	u16 sw_comp_cons, sw_comp_prod;
820 	int rx_pkt = 0;
821 	union eth_rx_cqe *cqe;
822 	struct eth_fast_path_rx_cqe *cqe_fp;
823 
824 #ifdef BNX2X_STOP_ON_ERROR
825 	if (unlikely(bp->panic))
826 		return 0;
827 #endif
828 
829 	bd_cons = fp->rx_bd_cons;
830 	bd_prod = fp->rx_bd_prod;
831 	bd_prod_fw = bd_prod;
832 	sw_comp_cons = fp->rx_comp_cons;
833 	sw_comp_prod = fp->rx_comp_prod;
834 
835 	comp_ring_cons = RCQ_BD(sw_comp_cons);
836 	cqe = &fp->rx_comp_ring[comp_ring_cons];
837 	cqe_fp = &cqe->fast_path_cqe;
838 
839 	DP(NETIF_MSG_RX_STATUS,
840 	   "queue[%d]: sw_comp_cons %u\n", fp->index, sw_comp_cons);
841 
842 	while (BNX2X_IS_CQE_COMPLETED(cqe_fp)) {
843 		struct sw_rx_bd *rx_buf = NULL;
844 		struct sk_buff *skb;
845 		u8 cqe_fp_flags;
846 		enum eth_rx_cqe_type cqe_fp_type;
847 		u16 len, pad, queue;
848 		u8 *data;
849 		bool l4_rxhash;
850 
851 #ifdef BNX2X_STOP_ON_ERROR
852 		if (unlikely(bp->panic))
853 			return 0;
854 #endif
855 
856 		bd_prod = RX_BD(bd_prod);
857 		bd_cons = RX_BD(bd_cons);
858 
859 		cqe_fp_flags = cqe_fp->type_error_flags;
860 		cqe_fp_type = cqe_fp_flags & ETH_FAST_PATH_RX_CQE_TYPE;
861 
862 		DP(NETIF_MSG_RX_STATUS,
863 		   "CQE type %x  err %x  status %x  queue %x  vlan %x  len %u\n",
864 		   CQE_TYPE(cqe_fp_flags),
865 		   cqe_fp_flags, cqe_fp->status_flags,
866 		   le32_to_cpu(cqe_fp->rss_hash_result),
867 		   le16_to_cpu(cqe_fp->vlan_tag),
868 		   le16_to_cpu(cqe_fp->pkt_len_or_gro_seg_len));
869 
870 		/* is this a slowpath msg? */
871 		if (unlikely(CQE_TYPE_SLOW(cqe_fp_type))) {
872 			bnx2x_sp_event(fp, cqe);
873 			goto next_cqe;
874 		}
875 
876 		rx_buf = &fp->rx_buf_ring[bd_cons];
877 		data = rx_buf->data;
878 
879 		if (!CQE_TYPE_FAST(cqe_fp_type)) {
880 			struct bnx2x_agg_info *tpa_info;
881 			u16 frag_size, pages;
882 #ifdef BNX2X_STOP_ON_ERROR
883 			/* sanity check */
884 			if (fp->disable_tpa &&
885 			    (CQE_TYPE_START(cqe_fp_type) ||
886 			     CQE_TYPE_STOP(cqe_fp_type)))
887 				BNX2X_ERR("START/STOP packet while disable_tpa type %x\n",
888 					  CQE_TYPE(cqe_fp_type));
889 #endif
890 
891 			if (CQE_TYPE_START(cqe_fp_type)) {
892 				u16 queue = cqe_fp->queue_index;
893 				DP(NETIF_MSG_RX_STATUS,
894 				   "calling tpa_start on queue %d\n",
895 				   queue);
896 
897 				bnx2x_tpa_start(fp, queue,
898 						bd_cons, bd_prod,
899 						cqe_fp);
900 
901 				goto next_rx;
902 			}
903 			queue = cqe->end_agg_cqe.queue_index;
904 			tpa_info = &fp->tpa_info[queue];
905 			DP(NETIF_MSG_RX_STATUS,
906 			   "calling tpa_stop on queue %d\n",
907 			   queue);
908 
909 			frag_size = le16_to_cpu(cqe->end_agg_cqe.pkt_len) -
910 				    tpa_info->len_on_bd;
911 
912 			if (fp->mode == TPA_MODE_GRO)
913 				pages = (frag_size + tpa_info->full_page - 1) /
914 					 tpa_info->full_page;
915 			else
916 				pages = SGE_PAGE_ALIGN(frag_size) >>
917 					SGE_PAGE_SHIFT;
918 
919 			bnx2x_tpa_stop(bp, fp, tpa_info, pages,
920 				       &cqe->end_agg_cqe, comp_ring_cons);
921 #ifdef BNX2X_STOP_ON_ERROR
922 			if (bp->panic)
923 				return 0;
924 #endif
925 
926 			bnx2x_update_sge_prod(fp, pages, &cqe->end_agg_cqe);
927 			goto next_cqe;
928 		}
929 		/* non TPA */
930 		len = le16_to_cpu(cqe_fp->pkt_len_or_gro_seg_len);
931 		pad = cqe_fp->placement_offset;
932 		dma_sync_single_for_cpu(&bp->pdev->dev,
933 					dma_unmap_addr(rx_buf, mapping),
934 					pad + RX_COPY_THRESH,
935 					DMA_FROM_DEVICE);
936 		pad += NET_SKB_PAD;
937 		prefetch(data + pad); /* speedup eth_type_trans() */
938 		/* is this an error packet? */
939 		if (unlikely(cqe_fp_flags & ETH_RX_ERROR_FALGS)) {
940 			DP(NETIF_MSG_RX_ERR | NETIF_MSG_RX_STATUS,
941 			   "ERROR  flags %x  rx packet %u\n",
942 			   cqe_fp_flags, sw_comp_cons);
943 			bnx2x_fp_qstats(bp, fp)->rx_err_discard_pkt++;
944 			goto reuse_rx;
945 		}
946 
947 		/* Since we don't have a jumbo ring
948 		 * copy small packets if mtu > 1500
949 		 */
950 		if ((bp->dev->mtu > ETH_MAX_PACKET_SIZE) &&
951 		    (len <= RX_COPY_THRESH)) {
952 			skb = netdev_alloc_skb_ip_align(bp->dev, len);
953 			if (skb == NULL) {
954 				DP(NETIF_MSG_RX_ERR | NETIF_MSG_RX_STATUS,
955 				   "ERROR  packet dropped because of alloc failure\n");
956 				bnx2x_fp_qstats(bp, fp)->rx_skb_alloc_failed++;
957 				goto reuse_rx;
958 			}
959 			memcpy(skb->data, data + pad, len);
960 			bnx2x_reuse_rx_data(fp, bd_cons, bd_prod);
961 		} else {
962 			if (likely(bnx2x_alloc_rx_data(bp, fp, bd_prod,
963 						       GFP_ATOMIC) == 0)) {
964 				dma_unmap_single(&bp->pdev->dev,
965 						 dma_unmap_addr(rx_buf, mapping),
966 						 fp->rx_buf_size,
967 						 DMA_FROM_DEVICE);
968 				skb = build_skb(data, fp->rx_frag_size);
969 				if (unlikely(!skb)) {
970 					bnx2x_frag_free(fp, data);
971 					bnx2x_fp_qstats(bp, fp)->
972 							rx_skb_alloc_failed++;
973 					goto next_rx;
974 				}
975 				skb_reserve(skb, pad);
976 			} else {
977 				DP(NETIF_MSG_RX_ERR | NETIF_MSG_RX_STATUS,
978 				   "ERROR  packet dropped because of alloc failure\n");
979 				bnx2x_fp_qstats(bp, fp)->rx_skb_alloc_failed++;
980 reuse_rx:
981 				bnx2x_reuse_rx_data(fp, bd_cons, bd_prod);
982 				goto next_rx;
983 			}
984 		}
985 
986 		skb_put(skb, len);
987 		skb->protocol = eth_type_trans(skb, bp->dev);
988 
989 		/* Set Toeplitz hash for a none-LRO skb */
990 		skb->rxhash = bnx2x_get_rxhash(bp, cqe_fp, &l4_rxhash);
991 		skb->l4_rxhash = l4_rxhash;
992 
993 		skb_checksum_none_assert(skb);
994 
995 		if (bp->dev->features & NETIF_F_RXCSUM)
996 			bnx2x_csum_validate(skb, cqe, fp,
997 					    bnx2x_fp_qstats(bp, fp));
998 
999 		skb_record_rx_queue(skb, fp->rx_queue);
1000 
1001 		if (le16_to_cpu(cqe_fp->pars_flags.flags) &
1002 		    PARSING_FLAGS_VLAN)
1003 			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
1004 					       le16_to_cpu(cqe_fp->vlan_tag));
1005 
1006 		skb_mark_napi_id(skb, &fp->napi);
1007 
1008 		if (bnx2x_fp_ll_polling(fp))
1009 			netif_receive_skb(skb);
1010 		else
1011 			napi_gro_receive(&fp->napi, skb);
1012 next_rx:
1013 		rx_buf->data = NULL;
1014 
1015 		bd_cons = NEXT_RX_IDX(bd_cons);
1016 		bd_prod = NEXT_RX_IDX(bd_prod);
1017 		bd_prod_fw = NEXT_RX_IDX(bd_prod_fw);
1018 		rx_pkt++;
1019 next_cqe:
1020 		sw_comp_prod = NEXT_RCQ_IDX(sw_comp_prod);
1021 		sw_comp_cons = NEXT_RCQ_IDX(sw_comp_cons);
1022 
1023 		/* mark CQE as free */
1024 		BNX2X_SEED_CQE(cqe_fp);
1025 
1026 		if (rx_pkt == budget)
1027 			break;
1028 
1029 		comp_ring_cons = RCQ_BD(sw_comp_cons);
1030 		cqe = &fp->rx_comp_ring[comp_ring_cons];
1031 		cqe_fp = &cqe->fast_path_cqe;
1032 	} /* while */
1033 
1034 	fp->rx_bd_cons = bd_cons;
1035 	fp->rx_bd_prod = bd_prod_fw;
1036 	fp->rx_comp_cons = sw_comp_cons;
1037 	fp->rx_comp_prod = sw_comp_prod;
1038 
1039 	/* Update producers */
1040 	bnx2x_update_rx_prod(bp, fp, bd_prod_fw, sw_comp_prod,
1041 			     fp->rx_sge_prod);
1042 
1043 	fp->rx_pkt += rx_pkt;
1044 	fp->rx_calls++;
1045 
1046 	return rx_pkt;
1047 }
1048 
1049 static irqreturn_t bnx2x_msix_fp_int(int irq, void *fp_cookie)
1050 {
1051 	struct bnx2x_fastpath *fp = fp_cookie;
1052 	struct bnx2x *bp = fp->bp;
1053 	u8 cos;
1054 
1055 	DP(NETIF_MSG_INTR,
1056 	   "got an MSI-X interrupt on IDX:SB [fp %d fw_sd %d igusb %d]\n",
1057 	   fp->index, fp->fw_sb_id, fp->igu_sb_id);
1058 
1059 	bnx2x_ack_sb(bp, fp->igu_sb_id, USTORM_ID, 0, IGU_INT_DISABLE, 0);
1060 
1061 #ifdef BNX2X_STOP_ON_ERROR
1062 	if (unlikely(bp->panic))
1063 		return IRQ_HANDLED;
1064 #endif
1065 
1066 	/* Handle Rx and Tx according to MSI-X vector */
1067 	for_each_cos_in_tx_queue(fp, cos)
1068 		prefetch(fp->txdata_ptr[cos]->tx_cons_sb);
1069 
1070 	prefetch(&fp->sb_running_index[SM_RX_ID]);
1071 	napi_schedule(&bnx2x_fp(bp, fp->index, napi));
1072 
1073 	return IRQ_HANDLED;
1074 }
1075 
1076 /* HW Lock for shared dual port PHYs */
1077 void bnx2x_acquire_phy_lock(struct bnx2x *bp)
1078 {
1079 	mutex_lock(&bp->port.phy_mutex);
1080 
1081 	bnx2x_acquire_hw_lock(bp, HW_LOCK_RESOURCE_MDIO);
1082 }
1083 
1084 void bnx2x_release_phy_lock(struct bnx2x *bp)
1085 {
1086 	bnx2x_release_hw_lock(bp, HW_LOCK_RESOURCE_MDIO);
1087 
1088 	mutex_unlock(&bp->port.phy_mutex);
1089 }
1090 
1091 /* calculates MF speed according to current linespeed and MF configuration */
1092 u16 bnx2x_get_mf_speed(struct bnx2x *bp)
1093 {
1094 	u16 line_speed = bp->link_vars.line_speed;
1095 	if (IS_MF(bp)) {
1096 		u16 maxCfg = bnx2x_extract_max_cfg(bp,
1097 						   bp->mf_config[BP_VN(bp)]);
1098 
1099 		/* Calculate the current MAX line speed limit for the MF
1100 		 * devices
1101 		 */
1102 		if (IS_MF_SI(bp))
1103 			line_speed = (line_speed * maxCfg) / 100;
1104 		else { /* SD mode */
1105 			u16 vn_max_rate = maxCfg * 100;
1106 
1107 			if (vn_max_rate < line_speed)
1108 				line_speed = vn_max_rate;
1109 		}
1110 	}
1111 
1112 	return line_speed;
1113 }
1114 
1115 /**
1116  * bnx2x_fill_report_data - fill link report data to report
1117  *
1118  * @bp:		driver handle
1119  * @data:	link state to update
1120  *
1121  * It uses a none-atomic bit operations because is called under the mutex.
1122  */
1123 static void bnx2x_fill_report_data(struct bnx2x *bp,
1124 				   struct bnx2x_link_report_data *data)
1125 {
1126 	u16 line_speed = bnx2x_get_mf_speed(bp);
1127 
1128 	memset(data, 0, sizeof(*data));
1129 
1130 	/* Fill the report data: effective line speed */
1131 	data->line_speed = line_speed;
1132 
1133 	/* Link is down */
1134 	if (!bp->link_vars.link_up || (bp->flags & MF_FUNC_DIS))
1135 		__set_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1136 			  &data->link_report_flags);
1137 
1138 	/* Full DUPLEX */
1139 	if (bp->link_vars.duplex == DUPLEX_FULL)
1140 		__set_bit(BNX2X_LINK_REPORT_FD, &data->link_report_flags);
1141 
1142 	/* Rx Flow Control is ON */
1143 	if (bp->link_vars.flow_ctrl & BNX2X_FLOW_CTRL_RX)
1144 		__set_bit(BNX2X_LINK_REPORT_RX_FC_ON, &data->link_report_flags);
1145 
1146 	/* Tx Flow Control is ON */
1147 	if (bp->link_vars.flow_ctrl & BNX2X_FLOW_CTRL_TX)
1148 		__set_bit(BNX2X_LINK_REPORT_TX_FC_ON, &data->link_report_flags);
1149 }
1150 
1151 /**
1152  * bnx2x_link_report - report link status to OS.
1153  *
1154  * @bp:		driver handle
1155  *
1156  * Calls the __bnx2x_link_report() under the same locking scheme
1157  * as a link/PHY state managing code to ensure a consistent link
1158  * reporting.
1159  */
1160 
1161 void bnx2x_link_report(struct bnx2x *bp)
1162 {
1163 	bnx2x_acquire_phy_lock(bp);
1164 	__bnx2x_link_report(bp);
1165 	bnx2x_release_phy_lock(bp);
1166 }
1167 
1168 /**
1169  * __bnx2x_link_report - report link status to OS.
1170  *
1171  * @bp:		driver handle
1172  *
1173  * None atomic implementation.
1174  * Should be called under the phy_lock.
1175  */
1176 void __bnx2x_link_report(struct bnx2x *bp)
1177 {
1178 	struct bnx2x_link_report_data cur_data;
1179 
1180 	/* reread mf_cfg */
1181 	if (IS_PF(bp) && !CHIP_IS_E1(bp))
1182 		bnx2x_read_mf_cfg(bp);
1183 
1184 	/* Read the current link report info */
1185 	bnx2x_fill_report_data(bp, &cur_data);
1186 
1187 	/* Don't report link down or exactly the same link status twice */
1188 	if (!memcmp(&cur_data, &bp->last_reported_link, sizeof(cur_data)) ||
1189 	    (test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1190 		      &bp->last_reported_link.link_report_flags) &&
1191 	     test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1192 		      &cur_data.link_report_flags)))
1193 		return;
1194 
1195 	bp->link_cnt++;
1196 
1197 	/* We are going to report a new link parameters now -
1198 	 * remember the current data for the next time.
1199 	 */
1200 	memcpy(&bp->last_reported_link, &cur_data, sizeof(cur_data));
1201 
1202 	if (test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1203 		     &cur_data.link_report_flags)) {
1204 		netif_carrier_off(bp->dev);
1205 		netdev_err(bp->dev, "NIC Link is Down\n");
1206 		return;
1207 	} else {
1208 		const char *duplex;
1209 		const char *flow;
1210 
1211 		netif_carrier_on(bp->dev);
1212 
1213 		if (test_and_clear_bit(BNX2X_LINK_REPORT_FD,
1214 				       &cur_data.link_report_flags))
1215 			duplex = "full";
1216 		else
1217 			duplex = "half";
1218 
1219 		/* Handle the FC at the end so that only these flags would be
1220 		 * possibly set. This way we may easily check if there is no FC
1221 		 * enabled.
1222 		 */
1223 		if (cur_data.link_report_flags) {
1224 			if (test_bit(BNX2X_LINK_REPORT_RX_FC_ON,
1225 				     &cur_data.link_report_flags)) {
1226 				if (test_bit(BNX2X_LINK_REPORT_TX_FC_ON,
1227 				     &cur_data.link_report_flags))
1228 					flow = "ON - receive & transmit";
1229 				else
1230 					flow = "ON - receive";
1231 			} else {
1232 				flow = "ON - transmit";
1233 			}
1234 		} else {
1235 			flow = "none";
1236 		}
1237 		netdev_info(bp->dev, "NIC Link is Up, %d Mbps %s duplex, Flow control: %s\n",
1238 			    cur_data.line_speed, duplex, flow);
1239 	}
1240 }
1241 
1242 static void bnx2x_set_next_page_sgl(struct bnx2x_fastpath *fp)
1243 {
1244 	int i;
1245 
1246 	for (i = 1; i <= NUM_RX_SGE_PAGES; i++) {
1247 		struct eth_rx_sge *sge;
1248 
1249 		sge = &fp->rx_sge_ring[RX_SGE_CNT * i - 2];
1250 		sge->addr_hi =
1251 			cpu_to_le32(U64_HI(fp->rx_sge_mapping +
1252 			BCM_PAGE_SIZE*(i % NUM_RX_SGE_PAGES)));
1253 
1254 		sge->addr_lo =
1255 			cpu_to_le32(U64_LO(fp->rx_sge_mapping +
1256 			BCM_PAGE_SIZE*(i % NUM_RX_SGE_PAGES)));
1257 	}
1258 }
1259 
1260 static void bnx2x_free_tpa_pool(struct bnx2x *bp,
1261 				struct bnx2x_fastpath *fp, int last)
1262 {
1263 	int i;
1264 
1265 	for (i = 0; i < last; i++) {
1266 		struct bnx2x_agg_info *tpa_info = &fp->tpa_info[i];
1267 		struct sw_rx_bd *first_buf = &tpa_info->first_buf;
1268 		u8 *data = first_buf->data;
1269 
1270 		if (data == NULL) {
1271 			DP(NETIF_MSG_IFDOWN, "tpa bin %d empty on free\n", i);
1272 			continue;
1273 		}
1274 		if (tpa_info->tpa_state == BNX2X_TPA_START)
1275 			dma_unmap_single(&bp->pdev->dev,
1276 					 dma_unmap_addr(first_buf, mapping),
1277 					 fp->rx_buf_size, DMA_FROM_DEVICE);
1278 		bnx2x_frag_free(fp, data);
1279 		first_buf->data = NULL;
1280 	}
1281 }
1282 
1283 void bnx2x_init_rx_rings_cnic(struct bnx2x *bp)
1284 {
1285 	int j;
1286 
1287 	for_each_rx_queue_cnic(bp, j) {
1288 		struct bnx2x_fastpath *fp = &bp->fp[j];
1289 
1290 		fp->rx_bd_cons = 0;
1291 
1292 		/* Activate BD ring */
1293 		/* Warning!
1294 		 * this will generate an interrupt (to the TSTORM)
1295 		 * must only be done after chip is initialized
1296 		 */
1297 		bnx2x_update_rx_prod(bp, fp, fp->rx_bd_prod, fp->rx_comp_prod,
1298 				     fp->rx_sge_prod);
1299 	}
1300 }
1301 
1302 void bnx2x_init_rx_rings(struct bnx2x *bp)
1303 {
1304 	int func = BP_FUNC(bp);
1305 	u16 ring_prod;
1306 	int i, j;
1307 
1308 	/* Allocate TPA resources */
1309 	for_each_eth_queue(bp, j) {
1310 		struct bnx2x_fastpath *fp = &bp->fp[j];
1311 
1312 		DP(NETIF_MSG_IFUP,
1313 		   "mtu %d  rx_buf_size %d\n", bp->dev->mtu, fp->rx_buf_size);
1314 
1315 		if (!fp->disable_tpa) {
1316 			/* Fill the per-aggregation pool */
1317 			for (i = 0; i < MAX_AGG_QS(bp); i++) {
1318 				struct bnx2x_agg_info *tpa_info =
1319 					&fp->tpa_info[i];
1320 				struct sw_rx_bd *first_buf =
1321 					&tpa_info->first_buf;
1322 
1323 				first_buf->data =
1324 					bnx2x_frag_alloc(fp, GFP_KERNEL);
1325 				if (!first_buf->data) {
1326 					BNX2X_ERR("Failed to allocate TPA skb pool for queue[%d] - disabling TPA on this queue!\n",
1327 						  j);
1328 					bnx2x_free_tpa_pool(bp, fp, i);
1329 					fp->disable_tpa = 1;
1330 					break;
1331 				}
1332 				dma_unmap_addr_set(first_buf, mapping, 0);
1333 				tpa_info->tpa_state = BNX2X_TPA_STOP;
1334 			}
1335 
1336 			/* "next page" elements initialization */
1337 			bnx2x_set_next_page_sgl(fp);
1338 
1339 			/* set SGEs bit mask */
1340 			bnx2x_init_sge_ring_bit_mask(fp);
1341 
1342 			/* Allocate SGEs and initialize the ring elements */
1343 			for (i = 0, ring_prod = 0;
1344 			     i < MAX_RX_SGE_CNT*NUM_RX_SGE_PAGES; i++) {
1345 
1346 				if (bnx2x_alloc_rx_sge(bp, fp, ring_prod,
1347 						       GFP_KERNEL) < 0) {
1348 					BNX2X_ERR("was only able to allocate %d rx sges\n",
1349 						  i);
1350 					BNX2X_ERR("disabling TPA for queue[%d]\n",
1351 						  j);
1352 					/* Cleanup already allocated elements */
1353 					bnx2x_free_rx_sge_range(bp, fp,
1354 								ring_prod);
1355 					bnx2x_free_tpa_pool(bp, fp,
1356 							    MAX_AGG_QS(bp));
1357 					fp->disable_tpa = 1;
1358 					ring_prod = 0;
1359 					break;
1360 				}
1361 				ring_prod = NEXT_SGE_IDX(ring_prod);
1362 			}
1363 
1364 			fp->rx_sge_prod = ring_prod;
1365 		}
1366 	}
1367 
1368 	for_each_eth_queue(bp, j) {
1369 		struct bnx2x_fastpath *fp = &bp->fp[j];
1370 
1371 		fp->rx_bd_cons = 0;
1372 
1373 		/* Activate BD ring */
1374 		/* Warning!
1375 		 * this will generate an interrupt (to the TSTORM)
1376 		 * must only be done after chip is initialized
1377 		 */
1378 		bnx2x_update_rx_prod(bp, fp, fp->rx_bd_prod, fp->rx_comp_prod,
1379 				     fp->rx_sge_prod);
1380 
1381 		if (j != 0)
1382 			continue;
1383 
1384 		if (CHIP_IS_E1(bp)) {
1385 			REG_WR(bp, BAR_USTRORM_INTMEM +
1386 			       USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func),
1387 			       U64_LO(fp->rx_comp_mapping));
1388 			REG_WR(bp, BAR_USTRORM_INTMEM +
1389 			       USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func) + 4,
1390 			       U64_HI(fp->rx_comp_mapping));
1391 		}
1392 	}
1393 }
1394 
1395 static void bnx2x_free_tx_skbs_queue(struct bnx2x_fastpath *fp)
1396 {
1397 	u8 cos;
1398 	struct bnx2x *bp = fp->bp;
1399 
1400 	for_each_cos_in_tx_queue(fp, cos) {
1401 		struct bnx2x_fp_txdata *txdata = fp->txdata_ptr[cos];
1402 		unsigned pkts_compl = 0, bytes_compl = 0;
1403 
1404 		u16 sw_prod = txdata->tx_pkt_prod;
1405 		u16 sw_cons = txdata->tx_pkt_cons;
1406 
1407 		while (sw_cons != sw_prod) {
1408 			bnx2x_free_tx_pkt(bp, txdata, TX_BD(sw_cons),
1409 					  &pkts_compl, &bytes_compl);
1410 			sw_cons++;
1411 		}
1412 
1413 		netdev_tx_reset_queue(
1414 			netdev_get_tx_queue(bp->dev,
1415 					    txdata->txq_index));
1416 	}
1417 }
1418 
1419 static void bnx2x_free_tx_skbs_cnic(struct bnx2x *bp)
1420 {
1421 	int i;
1422 
1423 	for_each_tx_queue_cnic(bp, i) {
1424 		bnx2x_free_tx_skbs_queue(&bp->fp[i]);
1425 	}
1426 }
1427 
1428 static void bnx2x_free_tx_skbs(struct bnx2x *bp)
1429 {
1430 	int i;
1431 
1432 	for_each_eth_queue(bp, i) {
1433 		bnx2x_free_tx_skbs_queue(&bp->fp[i]);
1434 	}
1435 }
1436 
1437 static void bnx2x_free_rx_bds(struct bnx2x_fastpath *fp)
1438 {
1439 	struct bnx2x *bp = fp->bp;
1440 	int i;
1441 
1442 	/* ring wasn't allocated */
1443 	if (fp->rx_buf_ring == NULL)
1444 		return;
1445 
1446 	for (i = 0; i < NUM_RX_BD; i++) {
1447 		struct sw_rx_bd *rx_buf = &fp->rx_buf_ring[i];
1448 		u8 *data = rx_buf->data;
1449 
1450 		if (data == NULL)
1451 			continue;
1452 		dma_unmap_single(&bp->pdev->dev,
1453 				 dma_unmap_addr(rx_buf, mapping),
1454 				 fp->rx_buf_size, DMA_FROM_DEVICE);
1455 
1456 		rx_buf->data = NULL;
1457 		bnx2x_frag_free(fp, data);
1458 	}
1459 }
1460 
1461 static void bnx2x_free_rx_skbs_cnic(struct bnx2x *bp)
1462 {
1463 	int j;
1464 
1465 	for_each_rx_queue_cnic(bp, j) {
1466 		bnx2x_free_rx_bds(&bp->fp[j]);
1467 	}
1468 }
1469 
1470 static void bnx2x_free_rx_skbs(struct bnx2x *bp)
1471 {
1472 	int j;
1473 
1474 	for_each_eth_queue(bp, j) {
1475 		struct bnx2x_fastpath *fp = &bp->fp[j];
1476 
1477 		bnx2x_free_rx_bds(fp);
1478 
1479 		if (!fp->disable_tpa)
1480 			bnx2x_free_tpa_pool(bp, fp, MAX_AGG_QS(bp));
1481 	}
1482 }
1483 
1484 void bnx2x_free_skbs_cnic(struct bnx2x *bp)
1485 {
1486 	bnx2x_free_tx_skbs_cnic(bp);
1487 	bnx2x_free_rx_skbs_cnic(bp);
1488 }
1489 
1490 void bnx2x_free_skbs(struct bnx2x *bp)
1491 {
1492 	bnx2x_free_tx_skbs(bp);
1493 	bnx2x_free_rx_skbs(bp);
1494 }
1495 
1496 void bnx2x_update_max_mf_config(struct bnx2x *bp, u32 value)
1497 {
1498 	/* load old values */
1499 	u32 mf_cfg = bp->mf_config[BP_VN(bp)];
1500 
1501 	if (value != bnx2x_extract_max_cfg(bp, mf_cfg)) {
1502 		/* leave all but MAX value */
1503 		mf_cfg &= ~FUNC_MF_CFG_MAX_BW_MASK;
1504 
1505 		/* set new MAX value */
1506 		mf_cfg |= (value << FUNC_MF_CFG_MAX_BW_SHIFT)
1507 				& FUNC_MF_CFG_MAX_BW_MASK;
1508 
1509 		bnx2x_fw_command(bp, DRV_MSG_CODE_SET_MF_BW, mf_cfg);
1510 	}
1511 }
1512 
1513 /**
1514  * bnx2x_free_msix_irqs - free previously requested MSI-X IRQ vectors
1515  *
1516  * @bp:		driver handle
1517  * @nvecs:	number of vectors to be released
1518  */
1519 static void bnx2x_free_msix_irqs(struct bnx2x *bp, int nvecs)
1520 {
1521 	int i, offset = 0;
1522 
1523 	if (nvecs == offset)
1524 		return;
1525 
1526 	/* VFs don't have a default SB */
1527 	if (IS_PF(bp)) {
1528 		free_irq(bp->msix_table[offset].vector, bp->dev);
1529 		DP(NETIF_MSG_IFDOWN, "released sp irq (%d)\n",
1530 		   bp->msix_table[offset].vector);
1531 		offset++;
1532 	}
1533 
1534 	if (CNIC_SUPPORT(bp)) {
1535 		if (nvecs == offset)
1536 			return;
1537 		offset++;
1538 	}
1539 
1540 	for_each_eth_queue(bp, i) {
1541 		if (nvecs == offset)
1542 			return;
1543 		DP(NETIF_MSG_IFDOWN, "about to release fp #%d->%d irq\n",
1544 		   i, bp->msix_table[offset].vector);
1545 
1546 		free_irq(bp->msix_table[offset++].vector, &bp->fp[i]);
1547 	}
1548 }
1549 
1550 void bnx2x_free_irq(struct bnx2x *bp)
1551 {
1552 	if (bp->flags & USING_MSIX_FLAG &&
1553 	    !(bp->flags & USING_SINGLE_MSIX_FLAG)) {
1554 		int nvecs = BNX2X_NUM_ETH_QUEUES(bp) + CNIC_SUPPORT(bp);
1555 
1556 		/* vfs don't have a default status block */
1557 		if (IS_PF(bp))
1558 			nvecs++;
1559 
1560 		bnx2x_free_msix_irqs(bp, nvecs);
1561 	} else {
1562 		free_irq(bp->dev->irq, bp->dev);
1563 	}
1564 }
1565 
1566 int bnx2x_enable_msix(struct bnx2x *bp)
1567 {
1568 	int msix_vec = 0, i, rc;
1569 
1570 	/* VFs don't have a default status block */
1571 	if (IS_PF(bp)) {
1572 		bp->msix_table[msix_vec].entry = msix_vec;
1573 		BNX2X_DEV_INFO("msix_table[0].entry = %d (slowpath)\n",
1574 			       bp->msix_table[0].entry);
1575 		msix_vec++;
1576 	}
1577 
1578 	/* Cnic requires an msix vector for itself */
1579 	if (CNIC_SUPPORT(bp)) {
1580 		bp->msix_table[msix_vec].entry = msix_vec;
1581 		BNX2X_DEV_INFO("msix_table[%d].entry = %d (CNIC)\n",
1582 			       msix_vec, bp->msix_table[msix_vec].entry);
1583 		msix_vec++;
1584 	}
1585 
1586 	/* We need separate vectors for ETH queues only (not FCoE) */
1587 	for_each_eth_queue(bp, i) {
1588 		bp->msix_table[msix_vec].entry = msix_vec;
1589 		BNX2X_DEV_INFO("msix_table[%d].entry = %d (fastpath #%u)\n",
1590 			       msix_vec, msix_vec, i);
1591 		msix_vec++;
1592 	}
1593 
1594 	DP(BNX2X_MSG_SP, "about to request enable msix with %d vectors\n",
1595 	   msix_vec);
1596 
1597 	rc = pci_enable_msix(bp->pdev, &bp->msix_table[0], msix_vec);
1598 
1599 	/*
1600 	 * reconfigure number of tx/rx queues according to available
1601 	 * MSI-X vectors
1602 	 */
1603 	if (rc >= BNX2X_MIN_MSIX_VEC_CNT(bp)) {
1604 		/* how less vectors we will have? */
1605 		int diff = msix_vec - rc;
1606 
1607 		BNX2X_DEV_INFO("Trying to use less MSI-X vectors: %d\n", rc);
1608 
1609 		rc = pci_enable_msix(bp->pdev, &bp->msix_table[0], rc);
1610 
1611 		if (rc) {
1612 			BNX2X_DEV_INFO("MSI-X is not attainable rc %d\n", rc);
1613 			goto no_msix;
1614 		}
1615 		/*
1616 		 * decrease number of queues by number of unallocated entries
1617 		 */
1618 		bp->num_ethernet_queues -= diff;
1619 		bp->num_queues = bp->num_ethernet_queues + bp->num_cnic_queues;
1620 
1621 		BNX2X_DEV_INFO("New queue configuration set: %d\n",
1622 			       bp->num_queues);
1623 	} else if (rc > 0) {
1624 		/* Get by with single vector */
1625 		rc = pci_enable_msix(bp->pdev, &bp->msix_table[0], 1);
1626 		if (rc) {
1627 			BNX2X_DEV_INFO("Single MSI-X is not attainable rc %d\n",
1628 				       rc);
1629 			goto no_msix;
1630 		}
1631 
1632 		BNX2X_DEV_INFO("Using single MSI-X vector\n");
1633 		bp->flags |= USING_SINGLE_MSIX_FLAG;
1634 
1635 		BNX2X_DEV_INFO("set number of queues to 1\n");
1636 		bp->num_ethernet_queues = 1;
1637 		bp->num_queues = bp->num_ethernet_queues + bp->num_cnic_queues;
1638 	} else if (rc < 0) {
1639 		BNX2X_DEV_INFO("MSI-X is not attainable  rc %d\n", rc);
1640 		goto no_msix;
1641 	}
1642 
1643 	bp->flags |= USING_MSIX_FLAG;
1644 
1645 	return 0;
1646 
1647 no_msix:
1648 	/* fall to INTx if not enough memory */
1649 	if (rc == -ENOMEM)
1650 		bp->flags |= DISABLE_MSI_FLAG;
1651 
1652 	return rc;
1653 }
1654 
1655 static int bnx2x_req_msix_irqs(struct bnx2x *bp)
1656 {
1657 	int i, rc, offset = 0;
1658 
1659 	/* no default status block for vf */
1660 	if (IS_PF(bp)) {
1661 		rc = request_irq(bp->msix_table[offset++].vector,
1662 				 bnx2x_msix_sp_int, 0,
1663 				 bp->dev->name, bp->dev);
1664 		if (rc) {
1665 			BNX2X_ERR("request sp irq failed\n");
1666 			return -EBUSY;
1667 		}
1668 	}
1669 
1670 	if (CNIC_SUPPORT(bp))
1671 		offset++;
1672 
1673 	for_each_eth_queue(bp, i) {
1674 		struct bnx2x_fastpath *fp = &bp->fp[i];
1675 		snprintf(fp->name, sizeof(fp->name), "%s-fp-%d",
1676 			 bp->dev->name, i);
1677 
1678 		rc = request_irq(bp->msix_table[offset].vector,
1679 				 bnx2x_msix_fp_int, 0, fp->name, fp);
1680 		if (rc) {
1681 			BNX2X_ERR("request fp #%d irq (%d) failed  rc %d\n", i,
1682 			      bp->msix_table[offset].vector, rc);
1683 			bnx2x_free_msix_irqs(bp, offset);
1684 			return -EBUSY;
1685 		}
1686 
1687 		offset++;
1688 	}
1689 
1690 	i = BNX2X_NUM_ETH_QUEUES(bp);
1691 	if (IS_PF(bp)) {
1692 		offset = 1 + CNIC_SUPPORT(bp);
1693 		netdev_info(bp->dev,
1694 			    "using MSI-X  IRQs: sp %d  fp[%d] %d ... fp[%d] %d\n",
1695 			    bp->msix_table[0].vector,
1696 			    0, bp->msix_table[offset].vector,
1697 			    i - 1, bp->msix_table[offset + i - 1].vector);
1698 	} else {
1699 		offset = CNIC_SUPPORT(bp);
1700 		netdev_info(bp->dev,
1701 			    "using MSI-X  IRQs: fp[%d] %d ... fp[%d] %d\n",
1702 			    0, bp->msix_table[offset].vector,
1703 			    i - 1, bp->msix_table[offset + i - 1].vector);
1704 	}
1705 	return 0;
1706 }
1707 
1708 int bnx2x_enable_msi(struct bnx2x *bp)
1709 {
1710 	int rc;
1711 
1712 	rc = pci_enable_msi(bp->pdev);
1713 	if (rc) {
1714 		BNX2X_DEV_INFO("MSI is not attainable\n");
1715 		return -1;
1716 	}
1717 	bp->flags |= USING_MSI_FLAG;
1718 
1719 	return 0;
1720 }
1721 
1722 static int bnx2x_req_irq(struct bnx2x *bp)
1723 {
1724 	unsigned long flags;
1725 	unsigned int irq;
1726 
1727 	if (bp->flags & (USING_MSI_FLAG | USING_MSIX_FLAG))
1728 		flags = 0;
1729 	else
1730 		flags = IRQF_SHARED;
1731 
1732 	if (bp->flags & USING_MSIX_FLAG)
1733 		irq = bp->msix_table[0].vector;
1734 	else
1735 		irq = bp->pdev->irq;
1736 
1737 	return request_irq(irq, bnx2x_interrupt, flags, bp->dev->name, bp->dev);
1738 }
1739 
1740 static int bnx2x_setup_irqs(struct bnx2x *bp)
1741 {
1742 	int rc = 0;
1743 	if (bp->flags & USING_MSIX_FLAG &&
1744 	    !(bp->flags & USING_SINGLE_MSIX_FLAG)) {
1745 		rc = bnx2x_req_msix_irqs(bp);
1746 		if (rc)
1747 			return rc;
1748 	} else {
1749 		rc = bnx2x_req_irq(bp);
1750 		if (rc) {
1751 			BNX2X_ERR("IRQ request failed  rc %d, aborting\n", rc);
1752 			return rc;
1753 		}
1754 		if (bp->flags & USING_MSI_FLAG) {
1755 			bp->dev->irq = bp->pdev->irq;
1756 			netdev_info(bp->dev, "using MSI IRQ %d\n",
1757 				    bp->dev->irq);
1758 		}
1759 		if (bp->flags & USING_MSIX_FLAG) {
1760 			bp->dev->irq = bp->msix_table[0].vector;
1761 			netdev_info(bp->dev, "using MSIX IRQ %d\n",
1762 				    bp->dev->irq);
1763 		}
1764 	}
1765 
1766 	return 0;
1767 }
1768 
1769 static void bnx2x_napi_enable_cnic(struct bnx2x *bp)
1770 {
1771 	int i;
1772 
1773 	for_each_rx_queue_cnic(bp, i) {
1774 		bnx2x_fp_init_lock(&bp->fp[i]);
1775 		napi_enable(&bnx2x_fp(bp, i, napi));
1776 	}
1777 }
1778 
1779 static void bnx2x_napi_enable(struct bnx2x *bp)
1780 {
1781 	int i;
1782 
1783 	for_each_eth_queue(bp, i) {
1784 		bnx2x_fp_init_lock(&bp->fp[i]);
1785 		napi_enable(&bnx2x_fp(bp, i, napi));
1786 	}
1787 }
1788 
1789 static void bnx2x_napi_disable_cnic(struct bnx2x *bp)
1790 {
1791 	int i;
1792 
1793 	local_bh_disable();
1794 	for_each_rx_queue_cnic(bp, i) {
1795 		napi_disable(&bnx2x_fp(bp, i, napi));
1796 		while (!bnx2x_fp_lock_napi(&bp->fp[i]))
1797 			mdelay(1);
1798 	}
1799 	local_bh_enable();
1800 }
1801 
1802 static void bnx2x_napi_disable(struct bnx2x *bp)
1803 {
1804 	int i;
1805 
1806 	local_bh_disable();
1807 	for_each_eth_queue(bp, i) {
1808 		napi_disable(&bnx2x_fp(bp, i, napi));
1809 		while (!bnx2x_fp_lock_napi(&bp->fp[i]))
1810 			mdelay(1);
1811 	}
1812 	local_bh_enable();
1813 }
1814 
1815 void bnx2x_netif_start(struct bnx2x *bp)
1816 {
1817 	if (netif_running(bp->dev)) {
1818 		bnx2x_napi_enable(bp);
1819 		if (CNIC_LOADED(bp))
1820 			bnx2x_napi_enable_cnic(bp);
1821 		bnx2x_int_enable(bp);
1822 		if (bp->state == BNX2X_STATE_OPEN)
1823 			netif_tx_wake_all_queues(bp->dev);
1824 	}
1825 }
1826 
1827 void bnx2x_netif_stop(struct bnx2x *bp, int disable_hw)
1828 {
1829 	bnx2x_int_disable_sync(bp, disable_hw);
1830 	bnx2x_napi_disable(bp);
1831 	if (CNIC_LOADED(bp))
1832 		bnx2x_napi_disable_cnic(bp);
1833 }
1834 
1835 u16 bnx2x_select_queue(struct net_device *dev, struct sk_buff *skb)
1836 {
1837 	struct bnx2x *bp = netdev_priv(dev);
1838 
1839 	if (CNIC_LOADED(bp) && !NO_FCOE(bp)) {
1840 		struct ethhdr *hdr = (struct ethhdr *)skb->data;
1841 		u16 ether_type = ntohs(hdr->h_proto);
1842 
1843 		/* Skip VLAN tag if present */
1844 		if (ether_type == ETH_P_8021Q) {
1845 			struct vlan_ethhdr *vhdr =
1846 				(struct vlan_ethhdr *)skb->data;
1847 
1848 			ether_type = ntohs(vhdr->h_vlan_encapsulated_proto);
1849 		}
1850 
1851 		/* If ethertype is FCoE or FIP - use FCoE ring */
1852 		if ((ether_type == ETH_P_FCOE) || (ether_type == ETH_P_FIP))
1853 			return bnx2x_fcoe_tx(bp, txq_index);
1854 	}
1855 
1856 	/* select a non-FCoE queue */
1857 	return __netdev_pick_tx(dev, skb) % BNX2X_NUM_ETH_QUEUES(bp);
1858 }
1859 
1860 void bnx2x_set_num_queues(struct bnx2x *bp)
1861 {
1862 	/* RSS queues */
1863 	bp->num_ethernet_queues = bnx2x_calc_num_queues(bp);
1864 
1865 	/* override in STORAGE SD modes */
1866 	if (IS_MF_STORAGE_SD(bp) || IS_MF_FCOE_AFEX(bp))
1867 		bp->num_ethernet_queues = 1;
1868 
1869 	/* Add special queues */
1870 	bp->num_cnic_queues = CNIC_SUPPORT(bp); /* For FCOE */
1871 	bp->num_queues = bp->num_ethernet_queues + bp->num_cnic_queues;
1872 
1873 	BNX2X_DEV_INFO("set number of queues to %d\n", bp->num_queues);
1874 }
1875 
1876 /**
1877  * bnx2x_set_real_num_queues - configure netdev->real_num_[tx,rx]_queues
1878  *
1879  * @bp:		Driver handle
1880  *
1881  * We currently support for at most 16 Tx queues for each CoS thus we will
1882  * allocate a multiple of 16 for ETH L2 rings according to the value of the
1883  * bp->max_cos.
1884  *
1885  * If there is an FCoE L2 queue the appropriate Tx queue will have the next
1886  * index after all ETH L2 indices.
1887  *
1888  * If the actual number of Tx queues (for each CoS) is less than 16 then there
1889  * will be the holes at the end of each group of 16 ETh L2 indices (0..15,
1890  * 16..31,...) with indices that are not coupled with any real Tx queue.
1891  *
1892  * The proper configuration of skb->queue_mapping is handled by
1893  * bnx2x_select_queue() and __skb_tx_hash().
1894  *
1895  * bnx2x_setup_tc() takes care of the proper TC mappings so that __skb_tx_hash()
1896  * will return a proper Tx index if TC is enabled (netdev->num_tc > 0).
1897  */
1898 static int bnx2x_set_real_num_queues(struct bnx2x *bp, int include_cnic)
1899 {
1900 	int rc, tx, rx;
1901 
1902 	tx = BNX2X_NUM_ETH_QUEUES(bp) * bp->max_cos;
1903 	rx = BNX2X_NUM_ETH_QUEUES(bp);
1904 
1905 /* account for fcoe queue */
1906 	if (include_cnic && !NO_FCOE(bp)) {
1907 		rx++;
1908 		tx++;
1909 	}
1910 
1911 	rc = netif_set_real_num_tx_queues(bp->dev, tx);
1912 	if (rc) {
1913 		BNX2X_ERR("Failed to set real number of Tx queues: %d\n", rc);
1914 		return rc;
1915 	}
1916 	rc = netif_set_real_num_rx_queues(bp->dev, rx);
1917 	if (rc) {
1918 		BNX2X_ERR("Failed to set real number of Rx queues: %d\n", rc);
1919 		return rc;
1920 	}
1921 
1922 	DP(NETIF_MSG_IFUP, "Setting real num queues to (tx, rx) (%d, %d)\n",
1923 			  tx, rx);
1924 
1925 	return rc;
1926 }
1927 
1928 static void bnx2x_set_rx_buf_size(struct bnx2x *bp)
1929 {
1930 	int i;
1931 
1932 	for_each_queue(bp, i) {
1933 		struct bnx2x_fastpath *fp = &bp->fp[i];
1934 		u32 mtu;
1935 
1936 		/* Always use a mini-jumbo MTU for the FCoE L2 ring */
1937 		if (IS_FCOE_IDX(i))
1938 			/*
1939 			 * Although there are no IP frames expected to arrive to
1940 			 * this ring we still want to add an
1941 			 * IP_HEADER_ALIGNMENT_PADDING to prevent a buffer
1942 			 * overrun attack.
1943 			 */
1944 			mtu = BNX2X_FCOE_MINI_JUMBO_MTU;
1945 		else
1946 			mtu = bp->dev->mtu;
1947 		fp->rx_buf_size = BNX2X_FW_RX_ALIGN_START +
1948 				  IP_HEADER_ALIGNMENT_PADDING +
1949 				  ETH_OVREHEAD +
1950 				  mtu +
1951 				  BNX2X_FW_RX_ALIGN_END;
1952 		/* Note : rx_buf_size doesn't take into account NET_SKB_PAD */
1953 		if (fp->rx_buf_size + NET_SKB_PAD <= PAGE_SIZE)
1954 			fp->rx_frag_size = fp->rx_buf_size + NET_SKB_PAD;
1955 		else
1956 			fp->rx_frag_size = 0;
1957 	}
1958 }
1959 
1960 static int bnx2x_init_rss(struct bnx2x *bp)
1961 {
1962 	int i;
1963 	u8 num_eth_queues = BNX2X_NUM_ETH_QUEUES(bp);
1964 
1965 	/* Prepare the initial contents for the indirection table if RSS is
1966 	 * enabled
1967 	 */
1968 	for (i = 0; i < sizeof(bp->rss_conf_obj.ind_table); i++)
1969 		bp->rss_conf_obj.ind_table[i] =
1970 			bp->fp->cl_id +
1971 			ethtool_rxfh_indir_default(i, num_eth_queues);
1972 
1973 	/*
1974 	 * For 57710 and 57711 SEARCHER configuration (rss_keys) is
1975 	 * per-port, so if explicit configuration is needed , do it only
1976 	 * for a PMF.
1977 	 *
1978 	 * For 57712 and newer on the other hand it's a per-function
1979 	 * configuration.
1980 	 */
1981 	return bnx2x_config_rss_eth(bp, bp->port.pmf || !CHIP_IS_E1x(bp));
1982 }
1983 
1984 int bnx2x_rss(struct bnx2x *bp, struct bnx2x_rss_config_obj *rss_obj,
1985 	      bool config_hash, bool enable)
1986 {
1987 	struct bnx2x_config_rss_params params = {NULL};
1988 
1989 	/* Although RSS is meaningless when there is a single HW queue we
1990 	 * still need it enabled in order to have HW Rx hash generated.
1991 	 *
1992 	 * if (!is_eth_multi(bp))
1993 	 *      bp->multi_mode = ETH_RSS_MODE_DISABLED;
1994 	 */
1995 
1996 	params.rss_obj = rss_obj;
1997 
1998 	__set_bit(RAMROD_COMP_WAIT, &params.ramrod_flags);
1999 
2000 	if (enable) {
2001 		__set_bit(BNX2X_RSS_MODE_REGULAR, &params.rss_flags);
2002 
2003 		/* RSS configuration */
2004 		__set_bit(BNX2X_RSS_IPV4, &params.rss_flags);
2005 		__set_bit(BNX2X_RSS_IPV4_TCP, &params.rss_flags);
2006 		__set_bit(BNX2X_RSS_IPV6, &params.rss_flags);
2007 		__set_bit(BNX2X_RSS_IPV6_TCP, &params.rss_flags);
2008 		if (rss_obj->udp_rss_v4)
2009 			__set_bit(BNX2X_RSS_IPV4_UDP, &params.rss_flags);
2010 		if (rss_obj->udp_rss_v6)
2011 			__set_bit(BNX2X_RSS_IPV6_UDP, &params.rss_flags);
2012 	} else {
2013 		__set_bit(BNX2X_RSS_MODE_DISABLED, &params.rss_flags);
2014 	}
2015 
2016 	/* Hash bits */
2017 	params.rss_result_mask = MULTI_MASK;
2018 
2019 	memcpy(params.ind_table, rss_obj->ind_table, sizeof(params.ind_table));
2020 
2021 	if (config_hash) {
2022 		/* RSS keys */
2023 		prandom_bytes(params.rss_key, T_ETH_RSS_KEY * 4);
2024 		__set_bit(BNX2X_RSS_SET_SRCH, &params.rss_flags);
2025 	}
2026 
2027 	if (IS_PF(bp))
2028 		return bnx2x_config_rss(bp, &params);
2029 	else
2030 		return bnx2x_vfpf_config_rss(bp, &params);
2031 }
2032 
2033 static int bnx2x_init_hw(struct bnx2x *bp, u32 load_code)
2034 {
2035 	struct bnx2x_func_state_params func_params = {NULL};
2036 
2037 	/* Prepare parameters for function state transitions */
2038 	__set_bit(RAMROD_COMP_WAIT, &func_params.ramrod_flags);
2039 
2040 	func_params.f_obj = &bp->func_obj;
2041 	func_params.cmd = BNX2X_F_CMD_HW_INIT;
2042 
2043 	func_params.params.hw_init.load_phase = load_code;
2044 
2045 	return bnx2x_func_state_change(bp, &func_params);
2046 }
2047 
2048 /*
2049  * Cleans the object that have internal lists without sending
2050  * ramrods. Should be run when interrupts are disabled.
2051  */
2052 void bnx2x_squeeze_objects(struct bnx2x *bp)
2053 {
2054 	int rc;
2055 	unsigned long ramrod_flags = 0, vlan_mac_flags = 0;
2056 	struct bnx2x_mcast_ramrod_params rparam = {NULL};
2057 	struct bnx2x_vlan_mac_obj *mac_obj = &bp->sp_objs->mac_obj;
2058 
2059 	/***************** Cleanup MACs' object first *************************/
2060 
2061 	/* Wait for completion of requested */
2062 	__set_bit(RAMROD_COMP_WAIT, &ramrod_flags);
2063 	/* Perform a dry cleanup */
2064 	__set_bit(RAMROD_DRV_CLR_ONLY, &ramrod_flags);
2065 
2066 	/* Clean ETH primary MAC */
2067 	__set_bit(BNX2X_ETH_MAC, &vlan_mac_flags);
2068 	rc = mac_obj->delete_all(bp, &bp->sp_objs->mac_obj, &vlan_mac_flags,
2069 				 &ramrod_flags);
2070 	if (rc != 0)
2071 		BNX2X_ERR("Failed to clean ETH MACs: %d\n", rc);
2072 
2073 	/* Cleanup UC list */
2074 	vlan_mac_flags = 0;
2075 	__set_bit(BNX2X_UC_LIST_MAC, &vlan_mac_flags);
2076 	rc = mac_obj->delete_all(bp, mac_obj, &vlan_mac_flags,
2077 				 &ramrod_flags);
2078 	if (rc != 0)
2079 		BNX2X_ERR("Failed to clean UC list MACs: %d\n", rc);
2080 
2081 	/***************** Now clean mcast object *****************************/
2082 	rparam.mcast_obj = &bp->mcast_obj;
2083 	__set_bit(RAMROD_DRV_CLR_ONLY, &rparam.ramrod_flags);
2084 
2085 	/* Add a DEL command... - Since we're doing a driver cleanup only,
2086 	 * we take a lock surrounding both the initial send and the CONTs,
2087 	 * as we don't want a true completion to disrupt us in the middle.
2088 	 */
2089 	netif_addr_lock_bh(bp->dev);
2090 	rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_DEL);
2091 	if (rc < 0)
2092 		BNX2X_ERR("Failed to add a new DEL command to a multi-cast object: %d\n",
2093 			  rc);
2094 
2095 	/* ...and wait until all pending commands are cleared */
2096 	rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_CONT);
2097 	while (rc != 0) {
2098 		if (rc < 0) {
2099 			BNX2X_ERR("Failed to clean multi-cast object: %d\n",
2100 				  rc);
2101 			netif_addr_unlock_bh(bp->dev);
2102 			return;
2103 		}
2104 
2105 		rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_CONT);
2106 	}
2107 	netif_addr_unlock_bh(bp->dev);
2108 }
2109 
2110 #ifndef BNX2X_STOP_ON_ERROR
2111 #define LOAD_ERROR_EXIT(bp, label) \
2112 	do { \
2113 		(bp)->state = BNX2X_STATE_ERROR; \
2114 		goto label; \
2115 	} while (0)
2116 
2117 #define LOAD_ERROR_EXIT_CNIC(bp, label) \
2118 	do { \
2119 		bp->cnic_loaded = false; \
2120 		goto label; \
2121 	} while (0)
2122 #else /*BNX2X_STOP_ON_ERROR*/
2123 #define LOAD_ERROR_EXIT(bp, label) \
2124 	do { \
2125 		(bp)->state = BNX2X_STATE_ERROR; \
2126 		(bp)->panic = 1; \
2127 		return -EBUSY; \
2128 	} while (0)
2129 #define LOAD_ERROR_EXIT_CNIC(bp, label) \
2130 	do { \
2131 		bp->cnic_loaded = false; \
2132 		(bp)->panic = 1; \
2133 		return -EBUSY; \
2134 	} while (0)
2135 #endif /*BNX2X_STOP_ON_ERROR*/
2136 
2137 static void bnx2x_free_fw_stats_mem(struct bnx2x *bp)
2138 {
2139 	BNX2X_PCI_FREE(bp->fw_stats, bp->fw_stats_mapping,
2140 		       bp->fw_stats_data_sz + bp->fw_stats_req_sz);
2141 	return;
2142 }
2143 
2144 static int bnx2x_alloc_fw_stats_mem(struct bnx2x *bp)
2145 {
2146 	int num_groups, vf_headroom = 0;
2147 	int is_fcoe_stats = NO_FCOE(bp) ? 0 : 1;
2148 
2149 	/* number of queues for statistics is number of eth queues + FCoE */
2150 	u8 num_queue_stats = BNX2X_NUM_ETH_QUEUES(bp) + is_fcoe_stats;
2151 
2152 	/* Total number of FW statistics requests =
2153 	 * 1 for port stats + 1 for PF stats + potential 2 for FCoE (fcoe proper
2154 	 * and fcoe l2 queue) stats + num of queues (which includes another 1
2155 	 * for fcoe l2 queue if applicable)
2156 	 */
2157 	bp->fw_stats_num = 2 + is_fcoe_stats + num_queue_stats;
2158 
2159 	/* vf stats appear in the request list, but their data is allocated by
2160 	 * the VFs themselves. We don't include them in the bp->fw_stats_num as
2161 	 * it is used to determine where to place the vf stats queries in the
2162 	 * request struct
2163 	 */
2164 	if (IS_SRIOV(bp))
2165 		vf_headroom = bnx2x_vf_headroom(bp);
2166 
2167 	/* Request is built from stats_query_header and an array of
2168 	 * stats_query_cmd_group each of which contains
2169 	 * STATS_QUERY_CMD_COUNT rules. The real number or requests is
2170 	 * configured in the stats_query_header.
2171 	 */
2172 	num_groups =
2173 		(((bp->fw_stats_num + vf_headroom) / STATS_QUERY_CMD_COUNT) +
2174 		 (((bp->fw_stats_num + vf_headroom) % STATS_QUERY_CMD_COUNT) ?
2175 		 1 : 0));
2176 
2177 	DP(BNX2X_MSG_SP, "stats fw_stats_num %d, vf headroom %d, num_groups %d\n",
2178 	   bp->fw_stats_num, vf_headroom, num_groups);
2179 	bp->fw_stats_req_sz = sizeof(struct stats_query_header) +
2180 		num_groups * sizeof(struct stats_query_cmd_group);
2181 
2182 	/* Data for statistics requests + stats_counter
2183 	 * stats_counter holds per-STORM counters that are incremented
2184 	 * when STORM has finished with the current request.
2185 	 * memory for FCoE offloaded statistics are counted anyway,
2186 	 * even if they will not be sent.
2187 	 * VF stats are not accounted for here as the data of VF stats is stored
2188 	 * in memory allocated by the VF, not here.
2189 	 */
2190 	bp->fw_stats_data_sz = sizeof(struct per_port_stats) +
2191 		sizeof(struct per_pf_stats) +
2192 		sizeof(struct fcoe_statistics_params) +
2193 		sizeof(struct per_queue_stats) * num_queue_stats +
2194 		sizeof(struct stats_counter);
2195 
2196 	BNX2X_PCI_ALLOC(bp->fw_stats, &bp->fw_stats_mapping,
2197 			bp->fw_stats_data_sz + bp->fw_stats_req_sz);
2198 
2199 	/* Set shortcuts */
2200 	bp->fw_stats_req = (struct bnx2x_fw_stats_req *)bp->fw_stats;
2201 	bp->fw_stats_req_mapping = bp->fw_stats_mapping;
2202 	bp->fw_stats_data = (struct bnx2x_fw_stats_data *)
2203 		((u8 *)bp->fw_stats + bp->fw_stats_req_sz);
2204 	bp->fw_stats_data_mapping = bp->fw_stats_mapping +
2205 		bp->fw_stats_req_sz;
2206 
2207 	DP(BNX2X_MSG_SP, "statistics request base address set to %x %x\n",
2208 	   U64_HI(bp->fw_stats_req_mapping),
2209 	   U64_LO(bp->fw_stats_req_mapping));
2210 	DP(BNX2X_MSG_SP, "statistics data base address set to %x %x\n",
2211 	   U64_HI(bp->fw_stats_data_mapping),
2212 	   U64_LO(bp->fw_stats_data_mapping));
2213 	return 0;
2214 
2215 alloc_mem_err:
2216 	bnx2x_free_fw_stats_mem(bp);
2217 	BNX2X_ERR("Can't allocate FW stats memory\n");
2218 	return -ENOMEM;
2219 }
2220 
2221 /* send load request to mcp and analyze response */
2222 static int bnx2x_nic_load_request(struct bnx2x *bp, u32 *load_code)
2223 {
2224 	u32 param;
2225 
2226 	/* init fw_seq */
2227 	bp->fw_seq =
2228 		(SHMEM_RD(bp, func_mb[BP_FW_MB_IDX(bp)].drv_mb_header) &
2229 		 DRV_MSG_SEQ_NUMBER_MASK);
2230 	BNX2X_DEV_INFO("fw_seq 0x%08x\n", bp->fw_seq);
2231 
2232 	/* Get current FW pulse sequence */
2233 	bp->fw_drv_pulse_wr_seq =
2234 		(SHMEM_RD(bp, func_mb[BP_FW_MB_IDX(bp)].drv_pulse_mb) &
2235 		 DRV_PULSE_SEQ_MASK);
2236 	BNX2X_DEV_INFO("drv_pulse 0x%x\n", bp->fw_drv_pulse_wr_seq);
2237 
2238 	param = DRV_MSG_CODE_LOAD_REQ_WITH_LFA;
2239 
2240 	if (IS_MF_SD(bp) && bnx2x_port_after_undi(bp))
2241 		param |= DRV_MSG_CODE_LOAD_REQ_FORCE_LFA;
2242 
2243 	/* load request */
2244 	(*load_code) = bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_REQ, param);
2245 
2246 	/* if mcp fails to respond we must abort */
2247 	if (!(*load_code)) {
2248 		BNX2X_ERR("MCP response failure, aborting\n");
2249 		return -EBUSY;
2250 	}
2251 
2252 	/* If mcp refused (e.g. other port is in diagnostic mode) we
2253 	 * must abort
2254 	 */
2255 	if ((*load_code) == FW_MSG_CODE_DRV_LOAD_REFUSED) {
2256 		BNX2X_ERR("MCP refused load request, aborting\n");
2257 		return -EBUSY;
2258 	}
2259 	return 0;
2260 }
2261 
2262 /* check whether another PF has already loaded FW to chip. In
2263  * virtualized environments a pf from another VM may have already
2264  * initialized the device including loading FW
2265  */
2266 int bnx2x_nic_load_analyze_req(struct bnx2x *bp, u32 load_code)
2267 {
2268 	/* is another pf loaded on this engine? */
2269 	if (load_code != FW_MSG_CODE_DRV_LOAD_COMMON_CHIP &&
2270 	    load_code != FW_MSG_CODE_DRV_LOAD_COMMON) {
2271 		/* build my FW version dword */
2272 		u32 my_fw = (BCM_5710_FW_MAJOR_VERSION) +
2273 			(BCM_5710_FW_MINOR_VERSION << 8) +
2274 			(BCM_5710_FW_REVISION_VERSION << 16) +
2275 			(BCM_5710_FW_ENGINEERING_VERSION << 24);
2276 
2277 		/* read loaded FW from chip */
2278 		u32 loaded_fw = REG_RD(bp, XSEM_REG_PRAM);
2279 
2280 		DP(BNX2X_MSG_SP, "loaded fw %x, my fw %x\n",
2281 		   loaded_fw, my_fw);
2282 
2283 		/* abort nic load if version mismatch */
2284 		if (my_fw != loaded_fw) {
2285 			BNX2X_ERR("bnx2x with FW %x was already loaded which mismatches my %x FW. Aborting\n",
2286 				  loaded_fw, my_fw);
2287 			return -EBUSY;
2288 		}
2289 	}
2290 	return 0;
2291 }
2292 
2293 /* returns the "mcp load_code" according to global load_count array */
2294 static int bnx2x_nic_load_no_mcp(struct bnx2x *bp, int port)
2295 {
2296 	int path = BP_PATH(bp);
2297 
2298 	DP(NETIF_MSG_IFUP, "NO MCP - load counts[%d]      %d, %d, %d\n",
2299 	   path, load_count[path][0], load_count[path][1],
2300 	   load_count[path][2]);
2301 	load_count[path][0]++;
2302 	load_count[path][1 + port]++;
2303 	DP(NETIF_MSG_IFUP, "NO MCP - new load counts[%d]  %d, %d, %d\n",
2304 	   path, load_count[path][0], load_count[path][1],
2305 	   load_count[path][2]);
2306 	if (load_count[path][0] == 1)
2307 		return FW_MSG_CODE_DRV_LOAD_COMMON;
2308 	else if (load_count[path][1 + port] == 1)
2309 		return FW_MSG_CODE_DRV_LOAD_PORT;
2310 	else
2311 		return FW_MSG_CODE_DRV_LOAD_FUNCTION;
2312 }
2313 
2314 /* mark PMF if applicable */
2315 static void bnx2x_nic_load_pmf(struct bnx2x *bp, u32 load_code)
2316 {
2317 	if ((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
2318 	    (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP) ||
2319 	    (load_code == FW_MSG_CODE_DRV_LOAD_PORT)) {
2320 		bp->port.pmf = 1;
2321 		/* We need the barrier to ensure the ordering between the
2322 		 * writing to bp->port.pmf here and reading it from the
2323 		 * bnx2x_periodic_task().
2324 		 */
2325 		smp_mb();
2326 	} else {
2327 		bp->port.pmf = 0;
2328 	}
2329 
2330 	DP(NETIF_MSG_LINK, "pmf %d\n", bp->port.pmf);
2331 }
2332 
2333 static void bnx2x_nic_load_afex_dcc(struct bnx2x *bp, int load_code)
2334 {
2335 	if (((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
2336 	     (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP)) &&
2337 	    (bp->common.shmem2_base)) {
2338 		if (SHMEM2_HAS(bp, dcc_support))
2339 			SHMEM2_WR(bp, dcc_support,
2340 				  (SHMEM_DCC_SUPPORT_DISABLE_ENABLE_PF_TLV |
2341 				   SHMEM_DCC_SUPPORT_BANDWIDTH_ALLOCATION_TLV));
2342 		if (SHMEM2_HAS(bp, afex_driver_support))
2343 			SHMEM2_WR(bp, afex_driver_support,
2344 				  SHMEM_AFEX_SUPPORTED_VERSION_ONE);
2345 	}
2346 
2347 	/* Set AFEX default VLAN tag to an invalid value */
2348 	bp->afex_def_vlan_tag = -1;
2349 }
2350 
2351 /**
2352  * bnx2x_bz_fp - zero content of the fastpath structure.
2353  *
2354  * @bp:		driver handle
2355  * @index:	fastpath index to be zeroed
2356  *
2357  * Makes sure the contents of the bp->fp[index].napi is kept
2358  * intact.
2359  */
2360 static void bnx2x_bz_fp(struct bnx2x *bp, int index)
2361 {
2362 	struct bnx2x_fastpath *fp = &bp->fp[index];
2363 	int cos;
2364 	struct napi_struct orig_napi = fp->napi;
2365 	struct bnx2x_agg_info *orig_tpa_info = fp->tpa_info;
2366 
2367 	/* bzero bnx2x_fastpath contents */
2368 	if (fp->tpa_info)
2369 		memset(fp->tpa_info, 0, ETH_MAX_AGGREGATION_QUEUES_E1H_E2 *
2370 		       sizeof(struct bnx2x_agg_info));
2371 	memset(fp, 0, sizeof(*fp));
2372 
2373 	/* Restore the NAPI object as it has been already initialized */
2374 	fp->napi = orig_napi;
2375 	fp->tpa_info = orig_tpa_info;
2376 	fp->bp = bp;
2377 	fp->index = index;
2378 	if (IS_ETH_FP(fp))
2379 		fp->max_cos = bp->max_cos;
2380 	else
2381 		/* Special queues support only one CoS */
2382 		fp->max_cos = 1;
2383 
2384 	/* Init txdata pointers */
2385 	if (IS_FCOE_FP(fp))
2386 		fp->txdata_ptr[0] = &bp->bnx2x_txq[FCOE_TXQ_IDX(bp)];
2387 	if (IS_ETH_FP(fp))
2388 		for_each_cos_in_tx_queue(fp, cos)
2389 			fp->txdata_ptr[cos] = &bp->bnx2x_txq[cos *
2390 				BNX2X_NUM_ETH_QUEUES(bp) + index];
2391 
2392 	/* set the tpa flag for each queue. The tpa flag determines the queue
2393 	 * minimal size so it must be set prior to queue memory allocation
2394 	 */
2395 	fp->disable_tpa = !(bp->flags & TPA_ENABLE_FLAG ||
2396 				  (bp->flags & GRO_ENABLE_FLAG &&
2397 				   bnx2x_mtu_allows_gro(bp->dev->mtu)));
2398 	if (bp->flags & TPA_ENABLE_FLAG)
2399 		fp->mode = TPA_MODE_LRO;
2400 	else if (bp->flags & GRO_ENABLE_FLAG)
2401 		fp->mode = TPA_MODE_GRO;
2402 
2403 	/* We don't want TPA on an FCoE L2 ring */
2404 	if (IS_FCOE_FP(fp))
2405 		fp->disable_tpa = 1;
2406 }
2407 
2408 int bnx2x_load_cnic(struct bnx2x *bp)
2409 {
2410 	int i, rc, port = BP_PORT(bp);
2411 
2412 	DP(NETIF_MSG_IFUP, "Starting CNIC-related load\n");
2413 
2414 	mutex_init(&bp->cnic_mutex);
2415 
2416 	if (IS_PF(bp)) {
2417 		rc = bnx2x_alloc_mem_cnic(bp);
2418 		if (rc) {
2419 			BNX2X_ERR("Unable to allocate bp memory for cnic\n");
2420 			LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic0);
2421 		}
2422 	}
2423 
2424 	rc = bnx2x_alloc_fp_mem_cnic(bp);
2425 	if (rc) {
2426 		BNX2X_ERR("Unable to allocate memory for cnic fps\n");
2427 		LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic0);
2428 	}
2429 
2430 	/* Update the number of queues with the cnic queues */
2431 	rc = bnx2x_set_real_num_queues(bp, 1);
2432 	if (rc) {
2433 		BNX2X_ERR("Unable to set real_num_queues including cnic\n");
2434 		LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic0);
2435 	}
2436 
2437 	/* Add all CNIC NAPI objects */
2438 	bnx2x_add_all_napi_cnic(bp);
2439 	DP(NETIF_MSG_IFUP, "cnic napi added\n");
2440 	bnx2x_napi_enable_cnic(bp);
2441 
2442 	rc = bnx2x_init_hw_func_cnic(bp);
2443 	if (rc)
2444 		LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic1);
2445 
2446 	bnx2x_nic_init_cnic(bp);
2447 
2448 	if (IS_PF(bp)) {
2449 		/* Enable Timer scan */
2450 		REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + port*4, 1);
2451 
2452 		/* setup cnic queues */
2453 		for_each_cnic_queue(bp, i) {
2454 			rc = bnx2x_setup_queue(bp, &bp->fp[i], 0);
2455 			if (rc) {
2456 				BNX2X_ERR("Queue setup failed\n");
2457 				LOAD_ERROR_EXIT(bp, load_error_cnic2);
2458 			}
2459 		}
2460 	}
2461 
2462 	/* Initialize Rx filter. */
2463 	bnx2x_set_rx_mode_inner(bp);
2464 
2465 	/* re-read iscsi info */
2466 	bnx2x_get_iscsi_info(bp);
2467 	bnx2x_setup_cnic_irq_info(bp);
2468 	bnx2x_setup_cnic_info(bp);
2469 	bp->cnic_loaded = true;
2470 	if (bp->state == BNX2X_STATE_OPEN)
2471 		bnx2x_cnic_notify(bp, CNIC_CTL_START_CMD);
2472 
2473 	DP(NETIF_MSG_IFUP, "Ending successfully CNIC-related load\n");
2474 
2475 	return 0;
2476 
2477 #ifndef BNX2X_STOP_ON_ERROR
2478 load_error_cnic2:
2479 	/* Disable Timer scan */
2480 	REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + port*4, 0);
2481 
2482 load_error_cnic1:
2483 	bnx2x_napi_disable_cnic(bp);
2484 	/* Update the number of queues without the cnic queues */
2485 	if (bnx2x_set_real_num_queues(bp, 0))
2486 		BNX2X_ERR("Unable to set real_num_queues not including cnic\n");
2487 load_error_cnic0:
2488 	BNX2X_ERR("CNIC-related load failed\n");
2489 	bnx2x_free_fp_mem_cnic(bp);
2490 	bnx2x_free_mem_cnic(bp);
2491 	return rc;
2492 #endif /* ! BNX2X_STOP_ON_ERROR */
2493 }
2494 
2495 /* must be called with rtnl_lock */
2496 int bnx2x_nic_load(struct bnx2x *bp, int load_mode)
2497 {
2498 	int port = BP_PORT(bp);
2499 	int i, rc = 0, load_code = 0;
2500 
2501 	DP(NETIF_MSG_IFUP, "Starting NIC load\n");
2502 	DP(NETIF_MSG_IFUP,
2503 	   "CNIC is %s\n", CNIC_ENABLED(bp) ? "enabled" : "disabled");
2504 
2505 #ifdef BNX2X_STOP_ON_ERROR
2506 	if (unlikely(bp->panic)) {
2507 		BNX2X_ERR("Can't load NIC when there is panic\n");
2508 		return -EPERM;
2509 	}
2510 #endif
2511 
2512 	bp->state = BNX2X_STATE_OPENING_WAIT4_LOAD;
2513 
2514 	/* zero the structure w/o any lock, before SP handler is initialized */
2515 	memset(&bp->last_reported_link, 0, sizeof(bp->last_reported_link));
2516 	__set_bit(BNX2X_LINK_REPORT_LINK_DOWN,
2517 		&bp->last_reported_link.link_report_flags);
2518 
2519 	if (IS_PF(bp))
2520 		/* must be called before memory allocation and HW init */
2521 		bnx2x_ilt_set_info(bp);
2522 
2523 	/*
2524 	 * Zero fastpath structures preserving invariants like napi, which are
2525 	 * allocated only once, fp index, max_cos, bp pointer.
2526 	 * Also set fp->disable_tpa and txdata_ptr.
2527 	 */
2528 	DP(NETIF_MSG_IFUP, "num queues: %d", bp->num_queues);
2529 	for_each_queue(bp, i)
2530 		bnx2x_bz_fp(bp, i);
2531 	memset(bp->bnx2x_txq, 0, (BNX2X_MAX_RSS_COUNT(bp) * BNX2X_MULTI_TX_COS +
2532 				  bp->num_cnic_queues) *
2533 				  sizeof(struct bnx2x_fp_txdata));
2534 
2535 	bp->fcoe_init = false;
2536 
2537 	/* Set the receive queues buffer size */
2538 	bnx2x_set_rx_buf_size(bp);
2539 
2540 	if (IS_PF(bp)) {
2541 		rc = bnx2x_alloc_mem(bp);
2542 		if (rc) {
2543 			BNX2X_ERR("Unable to allocate bp memory\n");
2544 			return rc;
2545 		}
2546 	}
2547 
2548 	/* Allocated memory for FW statistics  */
2549 	if (bnx2x_alloc_fw_stats_mem(bp))
2550 		LOAD_ERROR_EXIT(bp, load_error0);
2551 
2552 	/* need to be done after alloc mem, since it's self adjusting to amount
2553 	 * of memory available for RSS queues
2554 	 */
2555 	rc = bnx2x_alloc_fp_mem(bp);
2556 	if (rc) {
2557 		BNX2X_ERR("Unable to allocate memory for fps\n");
2558 		LOAD_ERROR_EXIT(bp, load_error0);
2559 	}
2560 
2561 	/* request pf to initialize status blocks */
2562 	if (IS_VF(bp)) {
2563 		rc = bnx2x_vfpf_init(bp);
2564 		if (rc)
2565 			LOAD_ERROR_EXIT(bp, load_error0);
2566 	}
2567 
2568 	/* As long as bnx2x_alloc_mem() may possibly update
2569 	 * bp->num_queues, bnx2x_set_real_num_queues() should always
2570 	 * come after it. At this stage cnic queues are not counted.
2571 	 */
2572 	rc = bnx2x_set_real_num_queues(bp, 0);
2573 	if (rc) {
2574 		BNX2X_ERR("Unable to set real_num_queues\n");
2575 		LOAD_ERROR_EXIT(bp, load_error0);
2576 	}
2577 
2578 	/* configure multi cos mappings in kernel.
2579 	 * this configuration may be overridden by a multi class queue
2580 	 * discipline or by a dcbx negotiation result.
2581 	 */
2582 	bnx2x_setup_tc(bp->dev, bp->max_cos);
2583 
2584 	/* Add all NAPI objects */
2585 	bnx2x_add_all_napi(bp);
2586 	DP(NETIF_MSG_IFUP, "napi added\n");
2587 	bnx2x_napi_enable(bp);
2588 
2589 	if (IS_PF(bp)) {
2590 		/* set pf load just before approaching the MCP */
2591 		bnx2x_set_pf_load(bp);
2592 
2593 		/* if mcp exists send load request and analyze response */
2594 		if (!BP_NOMCP(bp)) {
2595 			/* attempt to load pf */
2596 			rc = bnx2x_nic_load_request(bp, &load_code);
2597 			if (rc)
2598 				LOAD_ERROR_EXIT(bp, load_error1);
2599 
2600 			/* what did mcp say? */
2601 			rc = bnx2x_nic_load_analyze_req(bp, load_code);
2602 			if (rc) {
2603 				bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2604 				LOAD_ERROR_EXIT(bp, load_error2);
2605 			}
2606 		} else {
2607 			load_code = bnx2x_nic_load_no_mcp(bp, port);
2608 		}
2609 
2610 		/* mark pmf if applicable */
2611 		bnx2x_nic_load_pmf(bp, load_code);
2612 
2613 		/* Init Function state controlling object */
2614 		bnx2x__init_func_obj(bp);
2615 
2616 		/* Initialize HW */
2617 		rc = bnx2x_init_hw(bp, load_code);
2618 		if (rc) {
2619 			BNX2X_ERR("HW init failed, aborting\n");
2620 			bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2621 			LOAD_ERROR_EXIT(bp, load_error2);
2622 		}
2623 	}
2624 
2625 	bnx2x_pre_irq_nic_init(bp);
2626 
2627 	/* Connect to IRQs */
2628 	rc = bnx2x_setup_irqs(bp);
2629 	if (rc) {
2630 		BNX2X_ERR("setup irqs failed\n");
2631 		if (IS_PF(bp))
2632 			bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2633 		LOAD_ERROR_EXIT(bp, load_error2);
2634 	}
2635 
2636 	/* Init per-function objects */
2637 	if (IS_PF(bp)) {
2638 		/* Setup NIC internals and enable interrupts */
2639 		bnx2x_post_irq_nic_init(bp, load_code);
2640 
2641 		bnx2x_init_bp_objs(bp);
2642 		bnx2x_iov_nic_init(bp);
2643 
2644 		/* Set AFEX default VLAN tag to an invalid value */
2645 		bp->afex_def_vlan_tag = -1;
2646 		bnx2x_nic_load_afex_dcc(bp, load_code);
2647 		bp->state = BNX2X_STATE_OPENING_WAIT4_PORT;
2648 		rc = bnx2x_func_start(bp);
2649 		if (rc) {
2650 			BNX2X_ERR("Function start failed!\n");
2651 			bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2652 
2653 			LOAD_ERROR_EXIT(bp, load_error3);
2654 		}
2655 
2656 		/* Send LOAD_DONE command to MCP */
2657 		if (!BP_NOMCP(bp)) {
2658 			load_code = bnx2x_fw_command(bp,
2659 						     DRV_MSG_CODE_LOAD_DONE, 0);
2660 			if (!load_code) {
2661 				BNX2X_ERR("MCP response failure, aborting\n");
2662 				rc = -EBUSY;
2663 				LOAD_ERROR_EXIT(bp, load_error3);
2664 			}
2665 		}
2666 
2667 		/* initialize FW coalescing state machines in RAM */
2668 		bnx2x_update_coalesce(bp);
2669 	}
2670 
2671 	/* setup the leading queue */
2672 	rc = bnx2x_setup_leading(bp);
2673 	if (rc) {
2674 		BNX2X_ERR("Setup leading failed!\n");
2675 		LOAD_ERROR_EXIT(bp, load_error3);
2676 	}
2677 
2678 	/* set up the rest of the queues */
2679 	for_each_nondefault_eth_queue(bp, i) {
2680 		if (IS_PF(bp))
2681 			rc = bnx2x_setup_queue(bp, &bp->fp[i], false);
2682 		else /* VF */
2683 			rc = bnx2x_vfpf_setup_q(bp, &bp->fp[i], false);
2684 		if (rc) {
2685 			BNX2X_ERR("Queue %d setup failed\n", i);
2686 			LOAD_ERROR_EXIT(bp, load_error3);
2687 		}
2688 	}
2689 
2690 	/* setup rss */
2691 	rc = bnx2x_init_rss(bp);
2692 	if (rc) {
2693 		BNX2X_ERR("PF RSS init failed\n");
2694 		LOAD_ERROR_EXIT(bp, load_error3);
2695 	}
2696 
2697 	/* Now when Clients are configured we are ready to work */
2698 	bp->state = BNX2X_STATE_OPEN;
2699 
2700 	/* Configure a ucast MAC */
2701 	if (IS_PF(bp))
2702 		rc = bnx2x_set_eth_mac(bp, true);
2703 	else /* vf */
2704 		rc = bnx2x_vfpf_config_mac(bp, bp->dev->dev_addr, bp->fp->index,
2705 					   true);
2706 	if (rc) {
2707 		BNX2X_ERR("Setting Ethernet MAC failed\n");
2708 		LOAD_ERROR_EXIT(bp, load_error3);
2709 	}
2710 
2711 	if (IS_PF(bp) && bp->pending_max) {
2712 		bnx2x_update_max_mf_config(bp, bp->pending_max);
2713 		bp->pending_max = 0;
2714 	}
2715 
2716 	if (bp->port.pmf) {
2717 		rc = bnx2x_initial_phy_init(bp, load_mode);
2718 		if (rc)
2719 			LOAD_ERROR_EXIT(bp, load_error3);
2720 	}
2721 	bp->link_params.feature_config_flags &= ~FEATURE_CONFIG_BOOT_FROM_SAN;
2722 
2723 	/* Start fast path */
2724 
2725 	/* Initialize Rx filter. */
2726 	bnx2x_set_rx_mode_inner(bp);
2727 
2728 	/* Start the Tx */
2729 	switch (load_mode) {
2730 	case LOAD_NORMAL:
2731 		/* Tx queue should be only re-enabled */
2732 		netif_tx_wake_all_queues(bp->dev);
2733 		break;
2734 
2735 	case LOAD_OPEN:
2736 		netif_tx_start_all_queues(bp->dev);
2737 		smp_mb__after_clear_bit();
2738 		break;
2739 
2740 	case LOAD_DIAG:
2741 	case LOAD_LOOPBACK_EXT:
2742 		bp->state = BNX2X_STATE_DIAG;
2743 		break;
2744 
2745 	default:
2746 		break;
2747 	}
2748 
2749 	if (bp->port.pmf)
2750 		bnx2x_update_drv_flags(bp, 1 << DRV_FLAGS_PORT_MASK, 0);
2751 	else
2752 		bnx2x__link_status_update(bp);
2753 
2754 	/* start the timer */
2755 	mod_timer(&bp->timer, jiffies + bp->current_interval);
2756 
2757 	if (CNIC_ENABLED(bp))
2758 		bnx2x_load_cnic(bp);
2759 
2760 	if (IS_PF(bp) && SHMEM2_HAS(bp, drv_capabilities_flag)) {
2761 		/* mark driver is loaded in shmem2 */
2762 		u32 val;
2763 		val = SHMEM2_RD(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)]);
2764 		SHMEM2_WR(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)],
2765 			  val | DRV_FLAGS_CAPABILITIES_LOADED_SUPPORTED |
2766 			  DRV_FLAGS_CAPABILITIES_LOADED_L2);
2767 	}
2768 
2769 	/* Wait for all pending SP commands to complete */
2770 	if (IS_PF(bp) && !bnx2x_wait_sp_comp(bp, ~0x0UL)) {
2771 		BNX2X_ERR("Timeout waiting for SP elements to complete\n");
2772 		bnx2x_nic_unload(bp, UNLOAD_CLOSE, false);
2773 		return -EBUSY;
2774 	}
2775 
2776 	/* If PMF - send ADMIN DCBX msg to MFW to initiate DCBX FSM */
2777 	if (bp->port.pmf && (bp->state != BNX2X_STATE_DIAG))
2778 		bnx2x_dcbx_init(bp, false);
2779 
2780 	DP(NETIF_MSG_IFUP, "Ending successfully NIC load\n");
2781 
2782 	return 0;
2783 
2784 #ifndef BNX2X_STOP_ON_ERROR
2785 load_error3:
2786 	if (IS_PF(bp)) {
2787 		bnx2x_int_disable_sync(bp, 1);
2788 
2789 		/* Clean queueable objects */
2790 		bnx2x_squeeze_objects(bp);
2791 	}
2792 
2793 	/* Free SKBs, SGEs, TPA pool and driver internals */
2794 	bnx2x_free_skbs(bp);
2795 	for_each_rx_queue(bp, i)
2796 		bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
2797 
2798 	/* Release IRQs */
2799 	bnx2x_free_irq(bp);
2800 load_error2:
2801 	if (IS_PF(bp) && !BP_NOMCP(bp)) {
2802 		bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_REQ_WOL_MCP, 0);
2803 		bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_DONE, 0);
2804 	}
2805 
2806 	bp->port.pmf = 0;
2807 load_error1:
2808 	bnx2x_napi_disable(bp);
2809 	bnx2x_del_all_napi(bp);
2810 
2811 	/* clear pf_load status, as it was already set */
2812 	if (IS_PF(bp))
2813 		bnx2x_clear_pf_load(bp);
2814 load_error0:
2815 	bnx2x_free_fp_mem(bp);
2816 	bnx2x_free_fw_stats_mem(bp);
2817 	bnx2x_free_mem(bp);
2818 
2819 	return rc;
2820 #endif /* ! BNX2X_STOP_ON_ERROR */
2821 }
2822 
2823 int bnx2x_drain_tx_queues(struct bnx2x *bp)
2824 {
2825 	u8 rc = 0, cos, i;
2826 
2827 	/* Wait until tx fastpath tasks complete */
2828 	for_each_tx_queue(bp, i) {
2829 		struct bnx2x_fastpath *fp = &bp->fp[i];
2830 
2831 		for_each_cos_in_tx_queue(fp, cos)
2832 			rc = bnx2x_clean_tx_queue(bp, fp->txdata_ptr[cos]);
2833 		if (rc)
2834 			return rc;
2835 	}
2836 	return 0;
2837 }
2838 
2839 /* must be called with rtnl_lock */
2840 int bnx2x_nic_unload(struct bnx2x *bp, int unload_mode, bool keep_link)
2841 {
2842 	int i;
2843 	bool global = false;
2844 
2845 	DP(NETIF_MSG_IFUP, "Starting NIC unload\n");
2846 
2847 	/* mark driver is unloaded in shmem2 */
2848 	if (IS_PF(bp) && SHMEM2_HAS(bp, drv_capabilities_flag)) {
2849 		u32 val;
2850 		val = SHMEM2_RD(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)]);
2851 		SHMEM2_WR(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)],
2852 			  val & ~DRV_FLAGS_CAPABILITIES_LOADED_L2);
2853 	}
2854 
2855 	if (IS_PF(bp) && bp->recovery_state != BNX2X_RECOVERY_DONE &&
2856 	    (bp->state == BNX2X_STATE_CLOSED ||
2857 	     bp->state == BNX2X_STATE_ERROR)) {
2858 		/* We can get here if the driver has been unloaded
2859 		 * during parity error recovery and is either waiting for a
2860 		 * leader to complete or for other functions to unload and
2861 		 * then ifdown has been issued. In this case we want to
2862 		 * unload and let other functions to complete a recovery
2863 		 * process.
2864 		 */
2865 		bp->recovery_state = BNX2X_RECOVERY_DONE;
2866 		bp->is_leader = 0;
2867 		bnx2x_release_leader_lock(bp);
2868 		smp_mb();
2869 
2870 		DP(NETIF_MSG_IFDOWN, "Releasing a leadership...\n");
2871 		BNX2X_ERR("Can't unload in closed or error state\n");
2872 		return -EINVAL;
2873 	}
2874 
2875 	/* Nothing to do during unload if previous bnx2x_nic_load()
2876 	 * have not completed successfully - all resources are released.
2877 	 *
2878 	 * we can get here only after unsuccessful ndo_* callback, during which
2879 	 * dev->IFF_UP flag is still on.
2880 	 */
2881 	if (bp->state == BNX2X_STATE_CLOSED || bp->state == BNX2X_STATE_ERROR)
2882 		return 0;
2883 
2884 	/* It's important to set the bp->state to the value different from
2885 	 * BNX2X_STATE_OPEN and only then stop the Tx. Otherwise bnx2x_tx_int()
2886 	 * may restart the Tx from the NAPI context (see bnx2x_tx_int()).
2887 	 */
2888 	bp->state = BNX2X_STATE_CLOSING_WAIT4_HALT;
2889 	smp_mb();
2890 
2891 	/* indicate to VFs that the PF is going down */
2892 	bnx2x_iov_channel_down(bp);
2893 
2894 	if (CNIC_LOADED(bp))
2895 		bnx2x_cnic_notify(bp, CNIC_CTL_STOP_CMD);
2896 
2897 	/* Stop Tx */
2898 	bnx2x_tx_disable(bp);
2899 	netdev_reset_tc(bp->dev);
2900 
2901 	bp->rx_mode = BNX2X_RX_MODE_NONE;
2902 
2903 	del_timer_sync(&bp->timer);
2904 
2905 	if (IS_PF(bp)) {
2906 		/* Set ALWAYS_ALIVE bit in shmem */
2907 		bp->fw_drv_pulse_wr_seq |= DRV_PULSE_ALWAYS_ALIVE;
2908 		bnx2x_drv_pulse(bp);
2909 		bnx2x_stats_handle(bp, STATS_EVENT_STOP);
2910 		bnx2x_save_statistics(bp);
2911 	}
2912 
2913 	/* wait till consumers catch up with producers in all queues */
2914 	bnx2x_drain_tx_queues(bp);
2915 
2916 	/* if VF indicate to PF this function is going down (PF will delete sp
2917 	 * elements and clear initializations
2918 	 */
2919 	if (IS_VF(bp))
2920 		bnx2x_vfpf_close_vf(bp);
2921 	else if (unload_mode != UNLOAD_RECOVERY)
2922 		/* if this is a normal/close unload need to clean up chip*/
2923 		bnx2x_chip_cleanup(bp, unload_mode, keep_link);
2924 	else {
2925 		/* Send the UNLOAD_REQUEST to the MCP */
2926 		bnx2x_send_unload_req(bp, unload_mode);
2927 
2928 		/* Prevent transactions to host from the functions on the
2929 		 * engine that doesn't reset global blocks in case of global
2930 		 * attention once global blocks are reset and gates are opened
2931 		 * (the engine which leader will perform the recovery
2932 		 * last).
2933 		 */
2934 		if (!CHIP_IS_E1x(bp))
2935 			bnx2x_pf_disable(bp);
2936 
2937 		/* Disable HW interrupts, NAPI */
2938 		bnx2x_netif_stop(bp, 1);
2939 		/* Delete all NAPI objects */
2940 		bnx2x_del_all_napi(bp);
2941 		if (CNIC_LOADED(bp))
2942 			bnx2x_del_all_napi_cnic(bp);
2943 		/* Release IRQs */
2944 		bnx2x_free_irq(bp);
2945 
2946 		/* Report UNLOAD_DONE to MCP */
2947 		bnx2x_send_unload_done(bp, false);
2948 	}
2949 
2950 	/*
2951 	 * At this stage no more interrupts will arrive so we may safely clean
2952 	 * the queueable objects here in case they failed to get cleaned so far.
2953 	 */
2954 	if (IS_PF(bp))
2955 		bnx2x_squeeze_objects(bp);
2956 
2957 	/* There should be no more pending SP commands at this stage */
2958 	bp->sp_state = 0;
2959 
2960 	bp->port.pmf = 0;
2961 
2962 	/* Free SKBs, SGEs, TPA pool and driver internals */
2963 	bnx2x_free_skbs(bp);
2964 	if (CNIC_LOADED(bp))
2965 		bnx2x_free_skbs_cnic(bp);
2966 	for_each_rx_queue(bp, i)
2967 		bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
2968 
2969 	bnx2x_free_fp_mem(bp);
2970 	if (CNIC_LOADED(bp))
2971 		bnx2x_free_fp_mem_cnic(bp);
2972 
2973 	if (IS_PF(bp)) {
2974 		if (CNIC_LOADED(bp))
2975 			bnx2x_free_mem_cnic(bp);
2976 	}
2977 	bnx2x_free_mem(bp);
2978 
2979 	bp->state = BNX2X_STATE_CLOSED;
2980 	bp->cnic_loaded = false;
2981 
2982 	/* Check if there are pending parity attentions. If there are - set
2983 	 * RECOVERY_IN_PROGRESS.
2984 	 */
2985 	if (IS_PF(bp) && bnx2x_chk_parity_attn(bp, &global, false)) {
2986 		bnx2x_set_reset_in_progress(bp);
2987 
2988 		/* Set RESET_IS_GLOBAL if needed */
2989 		if (global)
2990 			bnx2x_set_reset_global(bp);
2991 	}
2992 
2993 	/* The last driver must disable a "close the gate" if there is no
2994 	 * parity attention or "process kill" pending.
2995 	 */
2996 	if (IS_PF(bp) &&
2997 	    !bnx2x_clear_pf_load(bp) &&
2998 	    bnx2x_reset_is_done(bp, BP_PATH(bp)))
2999 		bnx2x_disable_close_the_gate(bp);
3000 
3001 	DP(NETIF_MSG_IFUP, "Ending NIC unload\n");
3002 
3003 	return 0;
3004 }
3005 
3006 int bnx2x_set_power_state(struct bnx2x *bp, pci_power_t state)
3007 {
3008 	u16 pmcsr;
3009 
3010 	/* If there is no power capability, silently succeed */
3011 	if (!bp->pdev->pm_cap) {
3012 		BNX2X_DEV_INFO("No power capability. Breaking.\n");
3013 		return 0;
3014 	}
3015 
3016 	pci_read_config_word(bp->pdev, bp->pdev->pm_cap + PCI_PM_CTRL, &pmcsr);
3017 
3018 	switch (state) {
3019 	case PCI_D0:
3020 		pci_write_config_word(bp->pdev, bp->pdev->pm_cap + PCI_PM_CTRL,
3021 				      ((pmcsr & ~PCI_PM_CTRL_STATE_MASK) |
3022 				       PCI_PM_CTRL_PME_STATUS));
3023 
3024 		if (pmcsr & PCI_PM_CTRL_STATE_MASK)
3025 			/* delay required during transition out of D3hot */
3026 			msleep(20);
3027 		break;
3028 
3029 	case PCI_D3hot:
3030 		/* If there are other clients above don't
3031 		   shut down the power */
3032 		if (atomic_read(&bp->pdev->enable_cnt) != 1)
3033 			return 0;
3034 		/* Don't shut down the power for emulation and FPGA */
3035 		if (CHIP_REV_IS_SLOW(bp))
3036 			return 0;
3037 
3038 		pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
3039 		pmcsr |= 3;
3040 
3041 		if (bp->wol)
3042 			pmcsr |= PCI_PM_CTRL_PME_ENABLE;
3043 
3044 		pci_write_config_word(bp->pdev, bp->pdev->pm_cap + PCI_PM_CTRL,
3045 				      pmcsr);
3046 
3047 		/* No more memory access after this point until
3048 		* device is brought back to D0.
3049 		*/
3050 		break;
3051 
3052 	default:
3053 		dev_err(&bp->pdev->dev, "Can't support state = %d\n", state);
3054 		return -EINVAL;
3055 	}
3056 	return 0;
3057 }
3058 
3059 /*
3060  * net_device service functions
3061  */
3062 int bnx2x_poll(struct napi_struct *napi, int budget)
3063 {
3064 	int work_done = 0;
3065 	u8 cos;
3066 	struct bnx2x_fastpath *fp = container_of(napi, struct bnx2x_fastpath,
3067 						 napi);
3068 	struct bnx2x *bp = fp->bp;
3069 
3070 	while (1) {
3071 #ifdef BNX2X_STOP_ON_ERROR
3072 		if (unlikely(bp->panic)) {
3073 			napi_complete(napi);
3074 			return 0;
3075 		}
3076 #endif
3077 		if (!bnx2x_fp_lock_napi(fp))
3078 			return work_done;
3079 
3080 		for_each_cos_in_tx_queue(fp, cos)
3081 			if (bnx2x_tx_queue_has_work(fp->txdata_ptr[cos]))
3082 				bnx2x_tx_int(bp, fp->txdata_ptr[cos]);
3083 
3084 		if (bnx2x_has_rx_work(fp)) {
3085 			work_done += bnx2x_rx_int(fp, budget - work_done);
3086 
3087 			/* must not complete if we consumed full budget */
3088 			if (work_done >= budget) {
3089 				bnx2x_fp_unlock_napi(fp);
3090 				break;
3091 			}
3092 		}
3093 
3094 		/* Fall out from the NAPI loop if needed */
3095 		if (!bnx2x_fp_unlock_napi(fp) &&
3096 		    !(bnx2x_has_rx_work(fp) || bnx2x_has_tx_work(fp))) {
3097 
3098 			/* No need to update SB for FCoE L2 ring as long as
3099 			 * it's connected to the default SB and the SB
3100 			 * has been updated when NAPI was scheduled.
3101 			 */
3102 			if (IS_FCOE_FP(fp)) {
3103 				napi_complete(napi);
3104 				break;
3105 			}
3106 			bnx2x_update_fpsb_idx(fp);
3107 			/* bnx2x_has_rx_work() reads the status block,
3108 			 * thus we need to ensure that status block indices
3109 			 * have been actually read (bnx2x_update_fpsb_idx)
3110 			 * prior to this check (bnx2x_has_rx_work) so that
3111 			 * we won't write the "newer" value of the status block
3112 			 * to IGU (if there was a DMA right after
3113 			 * bnx2x_has_rx_work and if there is no rmb, the memory
3114 			 * reading (bnx2x_update_fpsb_idx) may be postponed
3115 			 * to right before bnx2x_ack_sb). In this case there
3116 			 * will never be another interrupt until there is
3117 			 * another update of the status block, while there
3118 			 * is still unhandled work.
3119 			 */
3120 			rmb();
3121 
3122 			if (!(bnx2x_has_rx_work(fp) || bnx2x_has_tx_work(fp))) {
3123 				napi_complete(napi);
3124 				/* Re-enable interrupts */
3125 				DP(NETIF_MSG_RX_STATUS,
3126 				   "Update index to %d\n", fp->fp_hc_idx);
3127 				bnx2x_ack_sb(bp, fp->igu_sb_id, USTORM_ID,
3128 					     le16_to_cpu(fp->fp_hc_idx),
3129 					     IGU_INT_ENABLE, 1);
3130 				break;
3131 			}
3132 		}
3133 	}
3134 
3135 	return work_done;
3136 }
3137 
3138 #ifdef CONFIG_NET_RX_BUSY_POLL
3139 /* must be called with local_bh_disable()d */
3140 int bnx2x_low_latency_recv(struct napi_struct *napi)
3141 {
3142 	struct bnx2x_fastpath *fp = container_of(napi, struct bnx2x_fastpath,
3143 						 napi);
3144 	struct bnx2x *bp = fp->bp;
3145 	int found = 0;
3146 
3147 	if ((bp->state == BNX2X_STATE_CLOSED) ||
3148 	    (bp->state == BNX2X_STATE_ERROR) ||
3149 	    (bp->flags & (TPA_ENABLE_FLAG | GRO_ENABLE_FLAG)))
3150 		return LL_FLUSH_FAILED;
3151 
3152 	if (!bnx2x_fp_lock_poll(fp))
3153 		return LL_FLUSH_BUSY;
3154 
3155 	if (bnx2x_has_rx_work(fp))
3156 		found = bnx2x_rx_int(fp, 4);
3157 
3158 	bnx2x_fp_unlock_poll(fp);
3159 
3160 	return found;
3161 }
3162 #endif
3163 
3164 /* we split the first BD into headers and data BDs
3165  * to ease the pain of our fellow microcode engineers
3166  * we use one mapping for both BDs
3167  */
3168 static u16 bnx2x_tx_split(struct bnx2x *bp,
3169 			  struct bnx2x_fp_txdata *txdata,
3170 			  struct sw_tx_bd *tx_buf,
3171 			  struct eth_tx_start_bd **tx_bd, u16 hlen,
3172 			  u16 bd_prod)
3173 {
3174 	struct eth_tx_start_bd *h_tx_bd = *tx_bd;
3175 	struct eth_tx_bd *d_tx_bd;
3176 	dma_addr_t mapping;
3177 	int old_len = le16_to_cpu(h_tx_bd->nbytes);
3178 
3179 	/* first fix first BD */
3180 	h_tx_bd->nbytes = cpu_to_le16(hlen);
3181 
3182 	DP(NETIF_MSG_TX_QUEUED,	"TSO split header size is %d (%x:%x)\n",
3183 	   h_tx_bd->nbytes, h_tx_bd->addr_hi, h_tx_bd->addr_lo);
3184 
3185 	/* now get a new data BD
3186 	 * (after the pbd) and fill it */
3187 	bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3188 	d_tx_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
3189 
3190 	mapping = HILO_U64(le32_to_cpu(h_tx_bd->addr_hi),
3191 			   le32_to_cpu(h_tx_bd->addr_lo)) + hlen;
3192 
3193 	d_tx_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
3194 	d_tx_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
3195 	d_tx_bd->nbytes = cpu_to_le16(old_len - hlen);
3196 
3197 	/* this marks the BD as one that has no individual mapping */
3198 	tx_buf->flags |= BNX2X_TSO_SPLIT_BD;
3199 
3200 	DP(NETIF_MSG_TX_QUEUED,
3201 	   "TSO split data size is %d (%x:%x)\n",
3202 	   d_tx_bd->nbytes, d_tx_bd->addr_hi, d_tx_bd->addr_lo);
3203 
3204 	/* update tx_bd */
3205 	*tx_bd = (struct eth_tx_start_bd *)d_tx_bd;
3206 
3207 	return bd_prod;
3208 }
3209 
3210 #define bswab32(b32) ((__force __le32) swab32((__force __u32) (b32)))
3211 #define bswab16(b16) ((__force __le16) swab16((__force __u16) (b16)))
3212 static __le16 bnx2x_csum_fix(unsigned char *t_header, u16 csum, s8 fix)
3213 {
3214 	__sum16 tsum = (__force __sum16) csum;
3215 
3216 	if (fix > 0)
3217 		tsum = ~csum_fold(csum_sub((__force __wsum) csum,
3218 				  csum_partial(t_header - fix, fix, 0)));
3219 
3220 	else if (fix < 0)
3221 		tsum = ~csum_fold(csum_add((__force __wsum) csum,
3222 				  csum_partial(t_header, -fix, 0)));
3223 
3224 	return bswab16(tsum);
3225 }
3226 
3227 static u32 bnx2x_xmit_type(struct bnx2x *bp, struct sk_buff *skb)
3228 {
3229 	u32 rc;
3230 	__u8 prot = 0;
3231 	__be16 protocol;
3232 
3233 	if (skb->ip_summed != CHECKSUM_PARTIAL)
3234 		return XMIT_PLAIN;
3235 
3236 	protocol = vlan_get_protocol(skb);
3237 	if (protocol == htons(ETH_P_IPV6)) {
3238 		rc = XMIT_CSUM_V6;
3239 		prot = ipv6_hdr(skb)->nexthdr;
3240 	} else {
3241 		rc = XMIT_CSUM_V4;
3242 		prot = ip_hdr(skb)->protocol;
3243 	}
3244 
3245 	if (!CHIP_IS_E1x(bp) && skb->encapsulation) {
3246 		if (inner_ip_hdr(skb)->version == 6) {
3247 			rc |= XMIT_CSUM_ENC_V6;
3248 			if (inner_ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3249 				rc |= XMIT_CSUM_TCP;
3250 		} else {
3251 			rc |= XMIT_CSUM_ENC_V4;
3252 			if (inner_ip_hdr(skb)->protocol == IPPROTO_TCP)
3253 				rc |= XMIT_CSUM_TCP;
3254 		}
3255 	}
3256 	if (prot == IPPROTO_TCP)
3257 		rc |= XMIT_CSUM_TCP;
3258 
3259 	if (skb_is_gso_v6(skb)) {
3260 		rc |= (XMIT_GSO_V6 | XMIT_CSUM_TCP);
3261 		if (rc & XMIT_CSUM_ENC)
3262 			rc |= XMIT_GSO_ENC_V6;
3263 	} else if (skb_is_gso(skb)) {
3264 		rc |= (XMIT_GSO_V4 | XMIT_CSUM_TCP);
3265 		if (rc & XMIT_CSUM_ENC)
3266 			rc |= XMIT_GSO_ENC_V4;
3267 	}
3268 
3269 	return rc;
3270 }
3271 
3272 #if (MAX_SKB_FRAGS >= MAX_FETCH_BD - 3)
3273 /* check if packet requires linearization (packet is too fragmented)
3274    no need to check fragmentation if page size > 8K (there will be no
3275    violation to FW restrictions) */
3276 static int bnx2x_pkt_req_lin(struct bnx2x *bp, struct sk_buff *skb,
3277 			     u32 xmit_type)
3278 {
3279 	int to_copy = 0;
3280 	int hlen = 0;
3281 	int first_bd_sz = 0;
3282 
3283 	/* 3 = 1 (for linear data BD) + 2 (for PBD and last BD) */
3284 	if (skb_shinfo(skb)->nr_frags >= (MAX_FETCH_BD - 3)) {
3285 
3286 		if (xmit_type & XMIT_GSO) {
3287 			unsigned short lso_mss = skb_shinfo(skb)->gso_size;
3288 			/* Check if LSO packet needs to be copied:
3289 			   3 = 1 (for headers BD) + 2 (for PBD and last BD) */
3290 			int wnd_size = MAX_FETCH_BD - 3;
3291 			/* Number of windows to check */
3292 			int num_wnds = skb_shinfo(skb)->nr_frags - wnd_size;
3293 			int wnd_idx = 0;
3294 			int frag_idx = 0;
3295 			u32 wnd_sum = 0;
3296 
3297 			/* Headers length */
3298 			hlen = (int)(skb_transport_header(skb) - skb->data) +
3299 				tcp_hdrlen(skb);
3300 
3301 			/* Amount of data (w/o headers) on linear part of SKB*/
3302 			first_bd_sz = skb_headlen(skb) - hlen;
3303 
3304 			wnd_sum  = first_bd_sz;
3305 
3306 			/* Calculate the first sum - it's special */
3307 			for (frag_idx = 0; frag_idx < wnd_size - 1; frag_idx++)
3308 				wnd_sum +=
3309 					skb_frag_size(&skb_shinfo(skb)->frags[frag_idx]);
3310 
3311 			/* If there was data on linear skb data - check it */
3312 			if (first_bd_sz > 0) {
3313 				if (unlikely(wnd_sum < lso_mss)) {
3314 					to_copy = 1;
3315 					goto exit_lbl;
3316 				}
3317 
3318 				wnd_sum -= first_bd_sz;
3319 			}
3320 
3321 			/* Others are easier: run through the frag list and
3322 			   check all windows */
3323 			for (wnd_idx = 0; wnd_idx <= num_wnds; wnd_idx++) {
3324 				wnd_sum +=
3325 			  skb_frag_size(&skb_shinfo(skb)->frags[wnd_idx + wnd_size - 1]);
3326 
3327 				if (unlikely(wnd_sum < lso_mss)) {
3328 					to_copy = 1;
3329 					break;
3330 				}
3331 				wnd_sum -=
3332 					skb_frag_size(&skb_shinfo(skb)->frags[wnd_idx]);
3333 			}
3334 		} else {
3335 			/* in non-LSO too fragmented packet should always
3336 			   be linearized */
3337 			to_copy = 1;
3338 		}
3339 	}
3340 
3341 exit_lbl:
3342 	if (unlikely(to_copy))
3343 		DP(NETIF_MSG_TX_QUEUED,
3344 		   "Linearization IS REQUIRED for %s packet. num_frags %d  hlen %d  first_bd_sz %d\n",
3345 		   (xmit_type & XMIT_GSO) ? "LSO" : "non-LSO",
3346 		   skb_shinfo(skb)->nr_frags, hlen, first_bd_sz);
3347 
3348 	return to_copy;
3349 }
3350 #endif
3351 
3352 static void bnx2x_set_pbd_gso_e2(struct sk_buff *skb, u32 *parsing_data,
3353 				 u32 xmit_type)
3354 {
3355 	struct ipv6hdr *ipv6;
3356 
3357 	*parsing_data |= (skb_shinfo(skb)->gso_size <<
3358 			      ETH_TX_PARSE_BD_E2_LSO_MSS_SHIFT) &
3359 			      ETH_TX_PARSE_BD_E2_LSO_MSS;
3360 
3361 	if (xmit_type & XMIT_GSO_ENC_V6)
3362 		ipv6 = inner_ipv6_hdr(skb);
3363 	else if (xmit_type & XMIT_GSO_V6)
3364 		ipv6 = ipv6_hdr(skb);
3365 	else
3366 		ipv6 = NULL;
3367 
3368 	if (ipv6 && ipv6->nexthdr == NEXTHDR_IPV6)
3369 		*parsing_data |= ETH_TX_PARSE_BD_E2_IPV6_WITH_EXT_HDR;
3370 }
3371 
3372 /**
3373  * bnx2x_set_pbd_gso - update PBD in GSO case.
3374  *
3375  * @skb:	packet skb
3376  * @pbd:	parse BD
3377  * @xmit_type:	xmit flags
3378  */
3379 static void bnx2x_set_pbd_gso(struct sk_buff *skb,
3380 			      struct eth_tx_parse_bd_e1x *pbd,
3381 			      struct eth_tx_start_bd *tx_start_bd,
3382 			      u32 xmit_type)
3383 {
3384 	pbd->lso_mss = cpu_to_le16(skb_shinfo(skb)->gso_size);
3385 	pbd->tcp_send_seq = bswab32(tcp_hdr(skb)->seq);
3386 	pbd->tcp_flags = pbd_tcp_flags(tcp_hdr(skb));
3387 
3388 	if (xmit_type & XMIT_GSO_V4) {
3389 		pbd->ip_id = bswab16(ip_hdr(skb)->id);
3390 		pbd->tcp_pseudo_csum =
3391 			bswab16(~csum_tcpudp_magic(ip_hdr(skb)->saddr,
3392 						   ip_hdr(skb)->daddr,
3393 						   0, IPPROTO_TCP, 0));
3394 
3395 		/* GSO on 57710/57711 needs FW to calculate IP checksum */
3396 		tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_IP_CSUM;
3397 	} else {
3398 		pbd->tcp_pseudo_csum =
3399 			bswab16(~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3400 						 &ipv6_hdr(skb)->daddr,
3401 						 0, IPPROTO_TCP, 0));
3402 	}
3403 
3404 	pbd->global_data |=
3405 		cpu_to_le16(ETH_TX_PARSE_BD_E1X_PSEUDO_CS_WITHOUT_LEN);
3406 }
3407 
3408 /**
3409  * bnx2x_set_pbd_csum_enc - update PBD with checksum and return header length
3410  *
3411  * @bp:			driver handle
3412  * @skb:		packet skb
3413  * @parsing_data:	data to be updated
3414  * @xmit_type:		xmit flags
3415  *
3416  * 57712/578xx related, when skb has encapsulation
3417  */
3418 static u8 bnx2x_set_pbd_csum_enc(struct bnx2x *bp, struct sk_buff *skb,
3419 				 u32 *parsing_data, u32 xmit_type)
3420 {
3421 	*parsing_data |=
3422 		((((u8 *)skb_inner_transport_header(skb) - skb->data) >> 1) <<
3423 		ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W_SHIFT) &
3424 		ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W;
3425 
3426 	if (xmit_type & XMIT_CSUM_TCP) {
3427 		*parsing_data |= ((inner_tcp_hdrlen(skb) / 4) <<
3428 			ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW_SHIFT) &
3429 			ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW;
3430 
3431 		return skb_inner_transport_header(skb) +
3432 			inner_tcp_hdrlen(skb) - skb->data;
3433 	}
3434 
3435 	/* We support checksum offload for TCP and UDP only.
3436 	 * No need to pass the UDP header length - it's a constant.
3437 	 */
3438 	return skb_inner_transport_header(skb) +
3439 		sizeof(struct udphdr) - skb->data;
3440 }
3441 
3442 /**
3443  * bnx2x_set_pbd_csum_e2 - update PBD with checksum and return header length
3444  *
3445  * @bp:			driver handle
3446  * @skb:		packet skb
3447  * @parsing_data:	data to be updated
3448  * @xmit_type:		xmit flags
3449  *
3450  * 57712/578xx related
3451  */
3452 static u8 bnx2x_set_pbd_csum_e2(struct bnx2x *bp, struct sk_buff *skb,
3453 				u32 *parsing_data, u32 xmit_type)
3454 {
3455 	*parsing_data |=
3456 		((((u8 *)skb_transport_header(skb) - skb->data) >> 1) <<
3457 		ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W_SHIFT) &
3458 		ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W;
3459 
3460 	if (xmit_type & XMIT_CSUM_TCP) {
3461 		*parsing_data |= ((tcp_hdrlen(skb) / 4) <<
3462 			ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW_SHIFT) &
3463 			ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW;
3464 
3465 		return skb_transport_header(skb) + tcp_hdrlen(skb) - skb->data;
3466 	}
3467 	/* We support checksum offload for TCP and UDP only.
3468 	 * No need to pass the UDP header length - it's a constant.
3469 	 */
3470 	return skb_transport_header(skb) + sizeof(struct udphdr) - skb->data;
3471 }
3472 
3473 /* set FW indication according to inner or outer protocols if tunneled */
3474 static void bnx2x_set_sbd_csum(struct bnx2x *bp, struct sk_buff *skb,
3475 			       struct eth_tx_start_bd *tx_start_bd,
3476 			       u32 xmit_type)
3477 {
3478 	tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_L4_CSUM;
3479 
3480 	if (xmit_type & (XMIT_CSUM_ENC_V6 | XMIT_CSUM_V6))
3481 		tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_IPV6;
3482 
3483 	if (!(xmit_type & XMIT_CSUM_TCP))
3484 		tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_IS_UDP;
3485 }
3486 
3487 /**
3488  * bnx2x_set_pbd_csum - update PBD with checksum and return header length
3489  *
3490  * @bp:		driver handle
3491  * @skb:	packet skb
3492  * @pbd:	parse BD to be updated
3493  * @xmit_type:	xmit flags
3494  */
3495 static u8 bnx2x_set_pbd_csum(struct bnx2x *bp, struct sk_buff *skb,
3496 			     struct eth_tx_parse_bd_e1x *pbd,
3497 			     u32 xmit_type)
3498 {
3499 	u8 hlen = (skb_network_header(skb) - skb->data) >> 1;
3500 
3501 	/* for now NS flag is not used in Linux */
3502 	pbd->global_data =
3503 		cpu_to_le16(hlen |
3504 			    ((skb->protocol == cpu_to_be16(ETH_P_8021Q)) <<
3505 			     ETH_TX_PARSE_BD_E1X_LLC_SNAP_EN_SHIFT));
3506 
3507 	pbd->ip_hlen_w = (skb_transport_header(skb) -
3508 			skb_network_header(skb)) >> 1;
3509 
3510 	hlen += pbd->ip_hlen_w;
3511 
3512 	/* We support checksum offload for TCP and UDP only */
3513 	if (xmit_type & XMIT_CSUM_TCP)
3514 		hlen += tcp_hdrlen(skb) / 2;
3515 	else
3516 		hlen += sizeof(struct udphdr) / 2;
3517 
3518 	pbd->total_hlen_w = cpu_to_le16(hlen);
3519 	hlen = hlen*2;
3520 
3521 	if (xmit_type & XMIT_CSUM_TCP) {
3522 		pbd->tcp_pseudo_csum = bswab16(tcp_hdr(skb)->check);
3523 
3524 	} else {
3525 		s8 fix = SKB_CS_OFF(skb); /* signed! */
3526 
3527 		DP(NETIF_MSG_TX_QUEUED,
3528 		   "hlen %d  fix %d  csum before fix %x\n",
3529 		   le16_to_cpu(pbd->total_hlen_w), fix, SKB_CS(skb));
3530 
3531 		/* HW bug: fixup the CSUM */
3532 		pbd->tcp_pseudo_csum =
3533 			bnx2x_csum_fix(skb_transport_header(skb),
3534 				       SKB_CS(skb), fix);
3535 
3536 		DP(NETIF_MSG_TX_QUEUED, "csum after fix %x\n",
3537 		   pbd->tcp_pseudo_csum);
3538 	}
3539 
3540 	return hlen;
3541 }
3542 
3543 static void bnx2x_update_pbds_gso_enc(struct sk_buff *skb,
3544 				      struct eth_tx_parse_bd_e2 *pbd_e2,
3545 				      struct eth_tx_parse_2nd_bd *pbd2,
3546 				      u16 *global_data,
3547 				      u32 xmit_type)
3548 {
3549 	u16 hlen_w = 0;
3550 	u8 outerip_off, outerip_len = 0;
3551 
3552 	/* from outer IP to transport */
3553 	hlen_w = (skb_inner_transport_header(skb) -
3554 		  skb_network_header(skb)) >> 1;
3555 
3556 	/* transport len */
3557 	hlen_w += inner_tcp_hdrlen(skb) >> 1;
3558 
3559 	pbd2->fw_ip_hdr_to_payload_w = hlen_w;
3560 
3561 	/* outer IP header info */
3562 	if (xmit_type & XMIT_CSUM_V4) {
3563 		struct iphdr *iph = ip_hdr(skb);
3564 		u32 csum = (__force u32)(~iph->check) -
3565 			   (__force u32)iph->tot_len -
3566 			   (__force u32)iph->frag_off;
3567 
3568 		pbd2->fw_ip_csum_wo_len_flags_frag =
3569 			bswab16(csum_fold((__force __wsum)csum));
3570 	} else {
3571 		pbd2->fw_ip_hdr_to_payload_w =
3572 			hlen_w - ((sizeof(struct ipv6hdr)) >> 1);
3573 	}
3574 
3575 	pbd2->tcp_send_seq = bswab32(inner_tcp_hdr(skb)->seq);
3576 
3577 	pbd2->tcp_flags = pbd_tcp_flags(inner_tcp_hdr(skb));
3578 
3579 	if (xmit_type & XMIT_GSO_V4) {
3580 		pbd2->hw_ip_id = bswab16(inner_ip_hdr(skb)->id);
3581 
3582 		pbd_e2->data.tunnel_data.pseudo_csum =
3583 			bswab16(~csum_tcpudp_magic(
3584 					inner_ip_hdr(skb)->saddr,
3585 					inner_ip_hdr(skb)->daddr,
3586 					0, IPPROTO_TCP, 0));
3587 
3588 		outerip_len = ip_hdr(skb)->ihl << 1;
3589 	} else {
3590 		pbd_e2->data.tunnel_data.pseudo_csum =
3591 			bswab16(~csum_ipv6_magic(
3592 					&inner_ipv6_hdr(skb)->saddr,
3593 					&inner_ipv6_hdr(skb)->daddr,
3594 					0, IPPROTO_TCP, 0));
3595 	}
3596 
3597 	outerip_off = (skb_network_header(skb) - skb->data) >> 1;
3598 
3599 	*global_data |=
3600 		outerip_off |
3601 		(!!(xmit_type & XMIT_CSUM_V6) <<
3602 			ETH_TX_PARSE_2ND_BD_IP_HDR_TYPE_OUTER_SHIFT) |
3603 		(outerip_len <<
3604 			ETH_TX_PARSE_2ND_BD_IP_HDR_LEN_OUTER_W_SHIFT) |
3605 		((skb->protocol == cpu_to_be16(ETH_P_8021Q)) <<
3606 			ETH_TX_PARSE_2ND_BD_LLC_SNAP_EN_SHIFT);
3607 
3608 	if (ip_hdr(skb)->protocol == IPPROTO_UDP) {
3609 		SET_FLAG(*global_data, ETH_TX_PARSE_2ND_BD_TUNNEL_UDP_EXIST, 1);
3610 		pbd2->tunnel_udp_hdr_start_w = skb_transport_offset(skb) >> 1;
3611 	}
3612 }
3613 
3614 /* called with netif_tx_lock
3615  * bnx2x_tx_int() runs without netif_tx_lock unless it needs to call
3616  * netif_wake_queue()
3617  */
3618 netdev_tx_t bnx2x_start_xmit(struct sk_buff *skb, struct net_device *dev)
3619 {
3620 	struct bnx2x *bp = netdev_priv(dev);
3621 
3622 	struct netdev_queue *txq;
3623 	struct bnx2x_fp_txdata *txdata;
3624 	struct sw_tx_bd *tx_buf;
3625 	struct eth_tx_start_bd *tx_start_bd, *first_bd;
3626 	struct eth_tx_bd *tx_data_bd, *total_pkt_bd = NULL;
3627 	struct eth_tx_parse_bd_e1x *pbd_e1x = NULL;
3628 	struct eth_tx_parse_bd_e2 *pbd_e2 = NULL;
3629 	struct eth_tx_parse_2nd_bd *pbd2 = NULL;
3630 	u32 pbd_e2_parsing_data = 0;
3631 	u16 pkt_prod, bd_prod;
3632 	int nbd, txq_index;
3633 	dma_addr_t mapping;
3634 	u32 xmit_type = bnx2x_xmit_type(bp, skb);
3635 	int i;
3636 	u8 hlen = 0;
3637 	__le16 pkt_size = 0;
3638 	struct ethhdr *eth;
3639 	u8 mac_type = UNICAST_ADDRESS;
3640 
3641 #ifdef BNX2X_STOP_ON_ERROR
3642 	if (unlikely(bp->panic))
3643 		return NETDEV_TX_BUSY;
3644 #endif
3645 
3646 	txq_index = skb_get_queue_mapping(skb);
3647 	txq = netdev_get_tx_queue(dev, txq_index);
3648 
3649 	BUG_ON(txq_index >= MAX_ETH_TXQ_IDX(bp) + (CNIC_LOADED(bp) ? 1 : 0));
3650 
3651 	txdata = &bp->bnx2x_txq[txq_index];
3652 
3653 	/* enable this debug print to view the transmission queue being used
3654 	DP(NETIF_MSG_TX_QUEUED, "indices: txq %d, fp %d, txdata %d\n",
3655 	   txq_index, fp_index, txdata_index); */
3656 
3657 	/* enable this debug print to view the transmission details
3658 	DP(NETIF_MSG_TX_QUEUED,
3659 	   "transmitting packet cid %d fp index %d txdata_index %d tx_data ptr %p fp pointer %p\n",
3660 	   txdata->cid, fp_index, txdata_index, txdata, fp); */
3661 
3662 	if (unlikely(bnx2x_tx_avail(bp, txdata) <
3663 			skb_shinfo(skb)->nr_frags +
3664 			BDS_PER_TX_PKT +
3665 			NEXT_CNT_PER_TX_PKT(MAX_BDS_PER_TX_PKT))) {
3666 		/* Handle special storage cases separately */
3667 		if (txdata->tx_ring_size == 0) {
3668 			struct bnx2x_eth_q_stats *q_stats =
3669 				bnx2x_fp_qstats(bp, txdata->parent_fp);
3670 			q_stats->driver_filtered_tx_pkt++;
3671 			dev_kfree_skb(skb);
3672 			return NETDEV_TX_OK;
3673 		}
3674 		bnx2x_fp_qstats(bp, txdata->parent_fp)->driver_xoff++;
3675 		netif_tx_stop_queue(txq);
3676 		BNX2X_ERR("BUG! Tx ring full when queue awake!\n");
3677 
3678 		return NETDEV_TX_BUSY;
3679 	}
3680 
3681 	DP(NETIF_MSG_TX_QUEUED,
3682 	   "queue[%d]: SKB: summed %x  protocol %x protocol(%x,%x) gso type %x  xmit_type %x len %d\n",
3683 	   txq_index, skb->ip_summed, skb->protocol, ipv6_hdr(skb)->nexthdr,
3684 	   ip_hdr(skb)->protocol, skb_shinfo(skb)->gso_type, xmit_type,
3685 	   skb->len);
3686 
3687 	eth = (struct ethhdr *)skb->data;
3688 
3689 	/* set flag according to packet type (UNICAST_ADDRESS is default)*/
3690 	if (unlikely(is_multicast_ether_addr(eth->h_dest))) {
3691 		if (is_broadcast_ether_addr(eth->h_dest))
3692 			mac_type = BROADCAST_ADDRESS;
3693 		else
3694 			mac_type = MULTICAST_ADDRESS;
3695 	}
3696 
3697 #if (MAX_SKB_FRAGS >= MAX_FETCH_BD - BDS_PER_TX_PKT)
3698 	/* First, check if we need to linearize the skb (due to FW
3699 	   restrictions). No need to check fragmentation if page size > 8K
3700 	   (there will be no violation to FW restrictions) */
3701 	if (bnx2x_pkt_req_lin(bp, skb, xmit_type)) {
3702 		/* Statistics of linearization */
3703 		bp->lin_cnt++;
3704 		if (skb_linearize(skb) != 0) {
3705 			DP(NETIF_MSG_TX_QUEUED,
3706 			   "SKB linearization failed - silently dropping this SKB\n");
3707 			dev_kfree_skb_any(skb);
3708 			return NETDEV_TX_OK;
3709 		}
3710 	}
3711 #endif
3712 	/* Map skb linear data for DMA */
3713 	mapping = dma_map_single(&bp->pdev->dev, skb->data,
3714 				 skb_headlen(skb), DMA_TO_DEVICE);
3715 	if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
3716 		DP(NETIF_MSG_TX_QUEUED,
3717 		   "SKB mapping failed - silently dropping this SKB\n");
3718 		dev_kfree_skb_any(skb);
3719 		return NETDEV_TX_OK;
3720 	}
3721 	/*
3722 	Please read carefully. First we use one BD which we mark as start,
3723 	then we have a parsing info BD (used for TSO or xsum),
3724 	and only then we have the rest of the TSO BDs.
3725 	(don't forget to mark the last one as last,
3726 	and to unmap only AFTER you write to the BD ...)
3727 	And above all, all pdb sizes are in words - NOT DWORDS!
3728 	*/
3729 
3730 	/* get current pkt produced now - advance it just before sending packet
3731 	 * since mapping of pages may fail and cause packet to be dropped
3732 	 */
3733 	pkt_prod = txdata->tx_pkt_prod;
3734 	bd_prod = TX_BD(txdata->tx_bd_prod);
3735 
3736 	/* get a tx_buf and first BD
3737 	 * tx_start_bd may be changed during SPLIT,
3738 	 * but first_bd will always stay first
3739 	 */
3740 	tx_buf = &txdata->tx_buf_ring[TX_BD(pkt_prod)];
3741 	tx_start_bd = &txdata->tx_desc_ring[bd_prod].start_bd;
3742 	first_bd = tx_start_bd;
3743 
3744 	tx_start_bd->bd_flags.as_bitfield = ETH_TX_BD_FLAGS_START_BD;
3745 
3746 	/* header nbd: indirectly zero other flags! */
3747 	tx_start_bd->general_data = 1 << ETH_TX_START_BD_HDR_NBDS_SHIFT;
3748 
3749 	/* remember the first BD of the packet */
3750 	tx_buf->first_bd = txdata->tx_bd_prod;
3751 	tx_buf->skb = skb;
3752 	tx_buf->flags = 0;
3753 
3754 	DP(NETIF_MSG_TX_QUEUED,
3755 	   "sending pkt %u @%p  next_idx %u  bd %u @%p\n",
3756 	   pkt_prod, tx_buf, txdata->tx_pkt_prod, bd_prod, tx_start_bd);
3757 
3758 	if (vlan_tx_tag_present(skb)) {
3759 		tx_start_bd->vlan_or_ethertype =
3760 		    cpu_to_le16(vlan_tx_tag_get(skb));
3761 		tx_start_bd->bd_flags.as_bitfield |=
3762 		    (X_ETH_OUTBAND_VLAN << ETH_TX_BD_FLAGS_VLAN_MODE_SHIFT);
3763 	} else {
3764 		/* when transmitting in a vf, start bd must hold the ethertype
3765 		 * for fw to enforce it
3766 		 */
3767 		if (IS_VF(bp))
3768 			tx_start_bd->vlan_or_ethertype =
3769 				cpu_to_le16(ntohs(eth->h_proto));
3770 		else
3771 			/* used by FW for packet accounting */
3772 			tx_start_bd->vlan_or_ethertype = cpu_to_le16(pkt_prod);
3773 	}
3774 
3775 	nbd = 2; /* start_bd + pbd + frags (updated when pages are mapped) */
3776 
3777 	/* turn on parsing and get a BD */
3778 	bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3779 
3780 	if (xmit_type & XMIT_CSUM)
3781 		bnx2x_set_sbd_csum(bp, skb, tx_start_bd, xmit_type);
3782 
3783 	if (!CHIP_IS_E1x(bp)) {
3784 		pbd_e2 = &txdata->tx_desc_ring[bd_prod].parse_bd_e2;
3785 		memset(pbd_e2, 0, sizeof(struct eth_tx_parse_bd_e2));
3786 
3787 		if (xmit_type & XMIT_CSUM_ENC) {
3788 			u16 global_data = 0;
3789 
3790 			/* Set PBD in enc checksum offload case */
3791 			hlen = bnx2x_set_pbd_csum_enc(bp, skb,
3792 						      &pbd_e2_parsing_data,
3793 						      xmit_type);
3794 
3795 			/* turn on 2nd parsing and get a BD */
3796 			bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3797 
3798 			pbd2 = &txdata->tx_desc_ring[bd_prod].parse_2nd_bd;
3799 
3800 			memset(pbd2, 0, sizeof(*pbd2));
3801 
3802 			pbd_e2->data.tunnel_data.ip_hdr_start_inner_w =
3803 				(skb_inner_network_header(skb) -
3804 				 skb->data) >> 1;
3805 
3806 			if (xmit_type & XMIT_GSO_ENC)
3807 				bnx2x_update_pbds_gso_enc(skb, pbd_e2, pbd2,
3808 							  &global_data,
3809 							  xmit_type);
3810 
3811 			pbd2->global_data = cpu_to_le16(global_data);
3812 
3813 			/* add addition parse BD indication to start BD */
3814 			SET_FLAG(tx_start_bd->general_data,
3815 				 ETH_TX_START_BD_PARSE_NBDS, 1);
3816 			/* set encapsulation flag in start BD */
3817 			SET_FLAG(tx_start_bd->general_data,
3818 				 ETH_TX_START_BD_TUNNEL_EXIST, 1);
3819 			nbd++;
3820 		} else if (xmit_type & XMIT_CSUM) {
3821 			/* Set PBD in checksum offload case w/o encapsulation */
3822 			hlen = bnx2x_set_pbd_csum_e2(bp, skb,
3823 						     &pbd_e2_parsing_data,
3824 						     xmit_type);
3825 		}
3826 
3827 		/* Add the macs to the parsing BD this is a vf */
3828 		if (IS_VF(bp)) {
3829 			/* override GRE parameters in BD */
3830 			bnx2x_set_fw_mac_addr(&pbd_e2->data.mac_addr.src_hi,
3831 					      &pbd_e2->data.mac_addr.src_mid,
3832 					      &pbd_e2->data.mac_addr.src_lo,
3833 					      eth->h_source);
3834 
3835 			bnx2x_set_fw_mac_addr(&pbd_e2->data.mac_addr.dst_hi,
3836 					      &pbd_e2->data.mac_addr.dst_mid,
3837 					      &pbd_e2->data.mac_addr.dst_lo,
3838 					      eth->h_dest);
3839 		}
3840 
3841 		SET_FLAG(pbd_e2_parsing_data,
3842 			 ETH_TX_PARSE_BD_E2_ETH_ADDR_TYPE, mac_type);
3843 	} else {
3844 		u16 global_data = 0;
3845 		pbd_e1x = &txdata->tx_desc_ring[bd_prod].parse_bd_e1x;
3846 		memset(pbd_e1x, 0, sizeof(struct eth_tx_parse_bd_e1x));
3847 		/* Set PBD in checksum offload case */
3848 		if (xmit_type & XMIT_CSUM)
3849 			hlen = bnx2x_set_pbd_csum(bp, skb, pbd_e1x, xmit_type);
3850 
3851 		SET_FLAG(global_data,
3852 			 ETH_TX_PARSE_BD_E1X_ETH_ADDR_TYPE, mac_type);
3853 		pbd_e1x->global_data |= cpu_to_le16(global_data);
3854 	}
3855 
3856 	/* Setup the data pointer of the first BD of the packet */
3857 	tx_start_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
3858 	tx_start_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
3859 	tx_start_bd->nbytes = cpu_to_le16(skb_headlen(skb));
3860 	pkt_size = tx_start_bd->nbytes;
3861 
3862 	DP(NETIF_MSG_TX_QUEUED,
3863 	   "first bd @%p  addr (%x:%x)  nbytes %d  flags %x  vlan %x\n",
3864 	   tx_start_bd, tx_start_bd->addr_hi, tx_start_bd->addr_lo,
3865 	   le16_to_cpu(tx_start_bd->nbytes),
3866 	   tx_start_bd->bd_flags.as_bitfield,
3867 	   le16_to_cpu(tx_start_bd->vlan_or_ethertype));
3868 
3869 	if (xmit_type & XMIT_GSO) {
3870 
3871 		DP(NETIF_MSG_TX_QUEUED,
3872 		   "TSO packet len %d  hlen %d  total len %d  tso size %d\n",
3873 		   skb->len, hlen, skb_headlen(skb),
3874 		   skb_shinfo(skb)->gso_size);
3875 
3876 		tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_SW_LSO;
3877 
3878 		if (unlikely(skb_headlen(skb) > hlen)) {
3879 			nbd++;
3880 			bd_prod = bnx2x_tx_split(bp, txdata, tx_buf,
3881 						 &tx_start_bd, hlen,
3882 						 bd_prod);
3883 		}
3884 		if (!CHIP_IS_E1x(bp))
3885 			bnx2x_set_pbd_gso_e2(skb, &pbd_e2_parsing_data,
3886 					     xmit_type);
3887 		else
3888 			bnx2x_set_pbd_gso(skb, pbd_e1x, first_bd, xmit_type);
3889 	}
3890 
3891 	/* Set the PBD's parsing_data field if not zero
3892 	 * (for the chips newer than 57711).
3893 	 */
3894 	if (pbd_e2_parsing_data)
3895 		pbd_e2->parsing_data = cpu_to_le32(pbd_e2_parsing_data);
3896 
3897 	tx_data_bd = (struct eth_tx_bd *)tx_start_bd;
3898 
3899 	/* Handle fragmented skb */
3900 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3901 		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3902 
3903 		mapping = skb_frag_dma_map(&bp->pdev->dev, frag, 0,
3904 					   skb_frag_size(frag), DMA_TO_DEVICE);
3905 		if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
3906 			unsigned int pkts_compl = 0, bytes_compl = 0;
3907 
3908 			DP(NETIF_MSG_TX_QUEUED,
3909 			   "Unable to map page - dropping packet...\n");
3910 
3911 			/* we need unmap all buffers already mapped
3912 			 * for this SKB;
3913 			 * first_bd->nbd need to be properly updated
3914 			 * before call to bnx2x_free_tx_pkt
3915 			 */
3916 			first_bd->nbd = cpu_to_le16(nbd);
3917 			bnx2x_free_tx_pkt(bp, txdata,
3918 					  TX_BD(txdata->tx_pkt_prod),
3919 					  &pkts_compl, &bytes_compl);
3920 			return NETDEV_TX_OK;
3921 		}
3922 
3923 		bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3924 		tx_data_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
3925 		if (total_pkt_bd == NULL)
3926 			total_pkt_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
3927 
3928 		tx_data_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
3929 		tx_data_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
3930 		tx_data_bd->nbytes = cpu_to_le16(skb_frag_size(frag));
3931 		le16_add_cpu(&pkt_size, skb_frag_size(frag));
3932 		nbd++;
3933 
3934 		DP(NETIF_MSG_TX_QUEUED,
3935 		   "frag %d  bd @%p  addr (%x:%x)  nbytes %d\n",
3936 		   i, tx_data_bd, tx_data_bd->addr_hi, tx_data_bd->addr_lo,
3937 		   le16_to_cpu(tx_data_bd->nbytes));
3938 	}
3939 
3940 	DP(NETIF_MSG_TX_QUEUED, "last bd @%p\n", tx_data_bd);
3941 
3942 	/* update with actual num BDs */
3943 	first_bd->nbd = cpu_to_le16(nbd);
3944 
3945 	bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3946 
3947 	/* now send a tx doorbell, counting the next BD
3948 	 * if the packet contains or ends with it
3949 	 */
3950 	if (TX_BD_POFF(bd_prod) < nbd)
3951 		nbd++;
3952 
3953 	/* total_pkt_bytes should be set on the first data BD if
3954 	 * it's not an LSO packet and there is more than one
3955 	 * data BD. In this case pkt_size is limited by an MTU value.
3956 	 * However we prefer to set it for an LSO packet (while we don't
3957 	 * have to) in order to save some CPU cycles in a none-LSO
3958 	 * case, when we much more care about them.
3959 	 */
3960 	if (total_pkt_bd != NULL)
3961 		total_pkt_bd->total_pkt_bytes = pkt_size;
3962 
3963 	if (pbd_e1x)
3964 		DP(NETIF_MSG_TX_QUEUED,
3965 		   "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",
3966 		   pbd_e1x, pbd_e1x->global_data, pbd_e1x->ip_hlen_w,
3967 		   pbd_e1x->ip_id, pbd_e1x->lso_mss, pbd_e1x->tcp_flags,
3968 		   pbd_e1x->tcp_pseudo_csum, pbd_e1x->tcp_send_seq,
3969 		    le16_to_cpu(pbd_e1x->total_hlen_w));
3970 	if (pbd_e2)
3971 		DP(NETIF_MSG_TX_QUEUED,
3972 		   "PBD (E2) @%p  dst %x %x %x src %x %x %x parsing_data %x\n",
3973 		   pbd_e2,
3974 		   pbd_e2->data.mac_addr.dst_hi,
3975 		   pbd_e2->data.mac_addr.dst_mid,
3976 		   pbd_e2->data.mac_addr.dst_lo,
3977 		   pbd_e2->data.mac_addr.src_hi,
3978 		   pbd_e2->data.mac_addr.src_mid,
3979 		   pbd_e2->data.mac_addr.src_lo,
3980 		   pbd_e2->parsing_data);
3981 	DP(NETIF_MSG_TX_QUEUED, "doorbell: nbd %d  bd %u\n", nbd, bd_prod);
3982 
3983 	netdev_tx_sent_queue(txq, skb->len);
3984 
3985 	skb_tx_timestamp(skb);
3986 
3987 	txdata->tx_pkt_prod++;
3988 	/*
3989 	 * Make sure that the BD data is updated before updating the producer
3990 	 * since FW might read the BD right after the producer is updated.
3991 	 * This is only applicable for weak-ordered memory model archs such
3992 	 * as IA-64. The following barrier is also mandatory since FW will
3993 	 * assumes packets must have BDs.
3994 	 */
3995 	wmb();
3996 
3997 	txdata->tx_db.data.prod += nbd;
3998 	barrier();
3999 
4000 	DOORBELL(bp, txdata->cid, txdata->tx_db.raw);
4001 
4002 	mmiowb();
4003 
4004 	txdata->tx_bd_prod += nbd;
4005 
4006 	if (unlikely(bnx2x_tx_avail(bp, txdata) < MAX_DESC_PER_TX_PKT)) {
4007 		netif_tx_stop_queue(txq);
4008 
4009 		/* paired memory barrier is in bnx2x_tx_int(), we have to keep
4010 		 * ordering of set_bit() in netif_tx_stop_queue() and read of
4011 		 * fp->bd_tx_cons */
4012 		smp_mb();
4013 
4014 		bnx2x_fp_qstats(bp, txdata->parent_fp)->driver_xoff++;
4015 		if (bnx2x_tx_avail(bp, txdata) >= MAX_DESC_PER_TX_PKT)
4016 			netif_tx_wake_queue(txq);
4017 	}
4018 	txdata->tx_pkt++;
4019 
4020 	return NETDEV_TX_OK;
4021 }
4022 
4023 /**
4024  * bnx2x_setup_tc - routine to configure net_device for multi tc
4025  *
4026  * @netdev: net device to configure
4027  * @tc: number of traffic classes to enable
4028  *
4029  * callback connected to the ndo_setup_tc function pointer
4030  */
4031 int bnx2x_setup_tc(struct net_device *dev, u8 num_tc)
4032 {
4033 	int cos, prio, count, offset;
4034 	struct bnx2x *bp = netdev_priv(dev);
4035 
4036 	/* setup tc must be called under rtnl lock */
4037 	ASSERT_RTNL();
4038 
4039 	/* no traffic classes requested. Aborting */
4040 	if (!num_tc) {
4041 		netdev_reset_tc(dev);
4042 		return 0;
4043 	}
4044 
4045 	/* requested to support too many traffic classes */
4046 	if (num_tc > bp->max_cos) {
4047 		BNX2X_ERR("support for too many traffic classes requested: %d. Max supported is %d\n",
4048 			  num_tc, bp->max_cos);
4049 		return -EINVAL;
4050 	}
4051 
4052 	/* declare amount of supported traffic classes */
4053 	if (netdev_set_num_tc(dev, num_tc)) {
4054 		BNX2X_ERR("failed to declare %d traffic classes\n", num_tc);
4055 		return -EINVAL;
4056 	}
4057 
4058 	/* configure priority to traffic class mapping */
4059 	for (prio = 0; prio < BNX2X_MAX_PRIORITY; prio++) {
4060 		netdev_set_prio_tc_map(dev, prio, bp->prio_to_cos[prio]);
4061 		DP(BNX2X_MSG_SP | NETIF_MSG_IFUP,
4062 		   "mapping priority %d to tc %d\n",
4063 		   prio, bp->prio_to_cos[prio]);
4064 	}
4065 
4066 	/* Use this configuration to differentiate tc0 from other COSes
4067 	   This can be used for ets or pfc, and save the effort of setting
4068 	   up a multio class queue disc or negotiating DCBX with a switch
4069 	netdev_set_prio_tc_map(dev, 0, 0);
4070 	DP(BNX2X_MSG_SP, "mapping priority %d to tc %d\n", 0, 0);
4071 	for (prio = 1; prio < 16; prio++) {
4072 		netdev_set_prio_tc_map(dev, prio, 1);
4073 		DP(BNX2X_MSG_SP, "mapping priority %d to tc %d\n", prio, 1);
4074 	} */
4075 
4076 	/* configure traffic class to transmission queue mapping */
4077 	for (cos = 0; cos < bp->max_cos; cos++) {
4078 		count = BNX2X_NUM_ETH_QUEUES(bp);
4079 		offset = cos * BNX2X_NUM_NON_CNIC_QUEUES(bp);
4080 		netdev_set_tc_queue(dev, cos, count, offset);
4081 		DP(BNX2X_MSG_SP | NETIF_MSG_IFUP,
4082 		   "mapping tc %d to offset %d count %d\n",
4083 		   cos, offset, count);
4084 	}
4085 
4086 	return 0;
4087 }
4088 
4089 /* called with rtnl_lock */
4090 int bnx2x_change_mac_addr(struct net_device *dev, void *p)
4091 {
4092 	struct sockaddr *addr = p;
4093 	struct bnx2x *bp = netdev_priv(dev);
4094 	int rc = 0;
4095 
4096 	if (!bnx2x_is_valid_ether_addr(bp, addr->sa_data)) {
4097 		BNX2X_ERR("Requested MAC address is not valid\n");
4098 		return -EINVAL;
4099 	}
4100 
4101 	if ((IS_MF_STORAGE_SD(bp) || IS_MF_FCOE_AFEX(bp)) &&
4102 	    !is_zero_ether_addr(addr->sa_data)) {
4103 		BNX2X_ERR("Can't configure non-zero address on iSCSI or FCoE functions in MF-SD mode\n");
4104 		return -EINVAL;
4105 	}
4106 
4107 	if (netif_running(dev))  {
4108 		rc = bnx2x_set_eth_mac(bp, false);
4109 		if (rc)
4110 			return rc;
4111 	}
4112 
4113 	memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
4114 
4115 	if (netif_running(dev))
4116 		rc = bnx2x_set_eth_mac(bp, true);
4117 
4118 	return rc;
4119 }
4120 
4121 static void bnx2x_free_fp_mem_at(struct bnx2x *bp, int fp_index)
4122 {
4123 	union host_hc_status_block *sb = &bnx2x_fp(bp, fp_index, status_blk);
4124 	struct bnx2x_fastpath *fp = &bp->fp[fp_index];
4125 	u8 cos;
4126 
4127 	/* Common */
4128 
4129 	if (IS_FCOE_IDX(fp_index)) {
4130 		memset(sb, 0, sizeof(union host_hc_status_block));
4131 		fp->status_blk_mapping = 0;
4132 	} else {
4133 		/* status blocks */
4134 		if (!CHIP_IS_E1x(bp))
4135 			BNX2X_PCI_FREE(sb->e2_sb,
4136 				       bnx2x_fp(bp, fp_index,
4137 						status_blk_mapping),
4138 				       sizeof(struct host_hc_status_block_e2));
4139 		else
4140 			BNX2X_PCI_FREE(sb->e1x_sb,
4141 				       bnx2x_fp(bp, fp_index,
4142 						status_blk_mapping),
4143 				       sizeof(struct host_hc_status_block_e1x));
4144 	}
4145 
4146 	/* Rx */
4147 	if (!skip_rx_queue(bp, fp_index)) {
4148 		bnx2x_free_rx_bds(fp);
4149 
4150 		/* fastpath rx rings: rx_buf rx_desc rx_comp */
4151 		BNX2X_FREE(bnx2x_fp(bp, fp_index, rx_buf_ring));
4152 		BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_desc_ring),
4153 			       bnx2x_fp(bp, fp_index, rx_desc_mapping),
4154 			       sizeof(struct eth_rx_bd) * NUM_RX_BD);
4155 
4156 		BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_comp_ring),
4157 			       bnx2x_fp(bp, fp_index, rx_comp_mapping),
4158 			       sizeof(struct eth_fast_path_rx_cqe) *
4159 			       NUM_RCQ_BD);
4160 
4161 		/* SGE ring */
4162 		BNX2X_FREE(bnx2x_fp(bp, fp_index, rx_page_ring));
4163 		BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_sge_ring),
4164 			       bnx2x_fp(bp, fp_index, rx_sge_mapping),
4165 			       BCM_PAGE_SIZE * NUM_RX_SGE_PAGES);
4166 	}
4167 
4168 	/* Tx */
4169 	if (!skip_tx_queue(bp, fp_index)) {
4170 		/* fastpath tx rings: tx_buf tx_desc */
4171 		for_each_cos_in_tx_queue(fp, cos) {
4172 			struct bnx2x_fp_txdata *txdata = fp->txdata_ptr[cos];
4173 
4174 			DP(NETIF_MSG_IFDOWN,
4175 			   "freeing tx memory of fp %d cos %d cid %d\n",
4176 			   fp_index, cos, txdata->cid);
4177 
4178 			BNX2X_FREE(txdata->tx_buf_ring);
4179 			BNX2X_PCI_FREE(txdata->tx_desc_ring,
4180 				txdata->tx_desc_mapping,
4181 				sizeof(union eth_tx_bd_types) * NUM_TX_BD);
4182 		}
4183 	}
4184 	/* end of fastpath */
4185 }
4186 
4187 void bnx2x_free_fp_mem_cnic(struct bnx2x *bp)
4188 {
4189 	int i;
4190 	for_each_cnic_queue(bp, i)
4191 		bnx2x_free_fp_mem_at(bp, i);
4192 }
4193 
4194 void bnx2x_free_fp_mem(struct bnx2x *bp)
4195 {
4196 	int i;
4197 	for_each_eth_queue(bp, i)
4198 		bnx2x_free_fp_mem_at(bp, i);
4199 }
4200 
4201 static void set_sb_shortcuts(struct bnx2x *bp, int index)
4202 {
4203 	union host_hc_status_block status_blk = bnx2x_fp(bp, index, status_blk);
4204 	if (!CHIP_IS_E1x(bp)) {
4205 		bnx2x_fp(bp, index, sb_index_values) =
4206 			(__le16 *)status_blk.e2_sb->sb.index_values;
4207 		bnx2x_fp(bp, index, sb_running_index) =
4208 			(__le16 *)status_blk.e2_sb->sb.running_index;
4209 	} else {
4210 		bnx2x_fp(bp, index, sb_index_values) =
4211 			(__le16 *)status_blk.e1x_sb->sb.index_values;
4212 		bnx2x_fp(bp, index, sb_running_index) =
4213 			(__le16 *)status_blk.e1x_sb->sb.running_index;
4214 	}
4215 }
4216 
4217 /* Returns the number of actually allocated BDs */
4218 static int bnx2x_alloc_rx_bds(struct bnx2x_fastpath *fp,
4219 			      int rx_ring_size)
4220 {
4221 	struct bnx2x *bp = fp->bp;
4222 	u16 ring_prod, cqe_ring_prod;
4223 	int i, failure_cnt = 0;
4224 
4225 	fp->rx_comp_cons = 0;
4226 	cqe_ring_prod = ring_prod = 0;
4227 
4228 	/* This routine is called only during fo init so
4229 	 * fp->eth_q_stats.rx_skb_alloc_failed = 0
4230 	 */
4231 	for (i = 0; i < rx_ring_size; i++) {
4232 		if (bnx2x_alloc_rx_data(bp, fp, ring_prod, GFP_KERNEL) < 0) {
4233 			failure_cnt++;
4234 			continue;
4235 		}
4236 		ring_prod = NEXT_RX_IDX(ring_prod);
4237 		cqe_ring_prod = NEXT_RCQ_IDX(cqe_ring_prod);
4238 		WARN_ON(ring_prod <= (i - failure_cnt));
4239 	}
4240 
4241 	if (failure_cnt)
4242 		BNX2X_ERR("was only able to allocate %d rx skbs on queue[%d]\n",
4243 			  i - failure_cnt, fp->index);
4244 
4245 	fp->rx_bd_prod = ring_prod;
4246 	/* Limit the CQE producer by the CQE ring size */
4247 	fp->rx_comp_prod = min_t(u16, NUM_RCQ_RINGS*RCQ_DESC_CNT,
4248 			       cqe_ring_prod);
4249 	fp->rx_pkt = fp->rx_calls = 0;
4250 
4251 	bnx2x_fp_stats(bp, fp)->eth_q_stats.rx_skb_alloc_failed += failure_cnt;
4252 
4253 	return i - failure_cnt;
4254 }
4255 
4256 static void bnx2x_set_next_page_rx_cq(struct bnx2x_fastpath *fp)
4257 {
4258 	int i;
4259 
4260 	for (i = 1; i <= NUM_RCQ_RINGS; i++) {
4261 		struct eth_rx_cqe_next_page *nextpg;
4262 
4263 		nextpg = (struct eth_rx_cqe_next_page *)
4264 			&fp->rx_comp_ring[RCQ_DESC_CNT * i - 1];
4265 		nextpg->addr_hi =
4266 			cpu_to_le32(U64_HI(fp->rx_comp_mapping +
4267 				   BCM_PAGE_SIZE*(i % NUM_RCQ_RINGS)));
4268 		nextpg->addr_lo =
4269 			cpu_to_le32(U64_LO(fp->rx_comp_mapping +
4270 				   BCM_PAGE_SIZE*(i % NUM_RCQ_RINGS)));
4271 	}
4272 }
4273 
4274 static int bnx2x_alloc_fp_mem_at(struct bnx2x *bp, int index)
4275 {
4276 	union host_hc_status_block *sb;
4277 	struct bnx2x_fastpath *fp = &bp->fp[index];
4278 	int ring_size = 0;
4279 	u8 cos;
4280 	int rx_ring_size = 0;
4281 
4282 	if (!bp->rx_ring_size &&
4283 	    (IS_MF_STORAGE_SD(bp) || IS_MF_FCOE_AFEX(bp))) {
4284 		rx_ring_size = MIN_RX_SIZE_NONTPA;
4285 		bp->rx_ring_size = rx_ring_size;
4286 	} else if (!bp->rx_ring_size) {
4287 		rx_ring_size = MAX_RX_AVAIL/BNX2X_NUM_RX_QUEUES(bp);
4288 
4289 		if (CHIP_IS_E3(bp)) {
4290 			u32 cfg = SHMEM_RD(bp,
4291 					   dev_info.port_hw_config[BP_PORT(bp)].
4292 					   default_cfg);
4293 
4294 			/* Decrease ring size for 1G functions */
4295 			if ((cfg & PORT_HW_CFG_NET_SERDES_IF_MASK) ==
4296 			    PORT_HW_CFG_NET_SERDES_IF_SGMII)
4297 				rx_ring_size /= 10;
4298 		}
4299 
4300 		/* allocate at least number of buffers required by FW */
4301 		rx_ring_size = max_t(int, bp->disable_tpa ? MIN_RX_SIZE_NONTPA :
4302 				     MIN_RX_SIZE_TPA, rx_ring_size);
4303 
4304 		bp->rx_ring_size = rx_ring_size;
4305 	} else /* if rx_ring_size specified - use it */
4306 		rx_ring_size = bp->rx_ring_size;
4307 
4308 	DP(BNX2X_MSG_SP, "calculated rx_ring_size %d\n", rx_ring_size);
4309 
4310 	/* Common */
4311 	sb = &bnx2x_fp(bp, index, status_blk);
4312 
4313 	if (!IS_FCOE_IDX(index)) {
4314 		/* status blocks */
4315 		if (!CHIP_IS_E1x(bp))
4316 			BNX2X_PCI_ALLOC(sb->e2_sb,
4317 				&bnx2x_fp(bp, index, status_blk_mapping),
4318 				sizeof(struct host_hc_status_block_e2));
4319 		else
4320 			BNX2X_PCI_ALLOC(sb->e1x_sb,
4321 				&bnx2x_fp(bp, index, status_blk_mapping),
4322 			    sizeof(struct host_hc_status_block_e1x));
4323 	}
4324 
4325 	/* FCoE Queue uses Default SB and doesn't ACK the SB, thus no need to
4326 	 * set shortcuts for it.
4327 	 */
4328 	if (!IS_FCOE_IDX(index))
4329 		set_sb_shortcuts(bp, index);
4330 
4331 	/* Tx */
4332 	if (!skip_tx_queue(bp, index)) {
4333 		/* fastpath tx rings: tx_buf tx_desc */
4334 		for_each_cos_in_tx_queue(fp, cos) {
4335 			struct bnx2x_fp_txdata *txdata = fp->txdata_ptr[cos];
4336 
4337 			DP(NETIF_MSG_IFUP,
4338 			   "allocating tx memory of fp %d cos %d\n",
4339 			   index, cos);
4340 
4341 			BNX2X_ALLOC(txdata->tx_buf_ring,
4342 				sizeof(struct sw_tx_bd) * NUM_TX_BD);
4343 			BNX2X_PCI_ALLOC(txdata->tx_desc_ring,
4344 				&txdata->tx_desc_mapping,
4345 				sizeof(union eth_tx_bd_types) * NUM_TX_BD);
4346 		}
4347 	}
4348 
4349 	/* Rx */
4350 	if (!skip_rx_queue(bp, index)) {
4351 		/* fastpath rx rings: rx_buf rx_desc rx_comp */
4352 		BNX2X_ALLOC(bnx2x_fp(bp, index, rx_buf_ring),
4353 				sizeof(struct sw_rx_bd) * NUM_RX_BD);
4354 		BNX2X_PCI_ALLOC(bnx2x_fp(bp, index, rx_desc_ring),
4355 				&bnx2x_fp(bp, index, rx_desc_mapping),
4356 				sizeof(struct eth_rx_bd) * NUM_RX_BD);
4357 
4358 		/* Seed all CQEs by 1s */
4359 		BNX2X_PCI_FALLOC(bnx2x_fp(bp, index, rx_comp_ring),
4360 				 &bnx2x_fp(bp, index, rx_comp_mapping),
4361 				 sizeof(struct eth_fast_path_rx_cqe) *
4362 				 NUM_RCQ_BD);
4363 
4364 		/* SGE ring */
4365 		BNX2X_ALLOC(bnx2x_fp(bp, index, rx_page_ring),
4366 				sizeof(struct sw_rx_page) * NUM_RX_SGE);
4367 		BNX2X_PCI_ALLOC(bnx2x_fp(bp, index, rx_sge_ring),
4368 				&bnx2x_fp(bp, index, rx_sge_mapping),
4369 				BCM_PAGE_SIZE * NUM_RX_SGE_PAGES);
4370 		/* RX BD ring */
4371 		bnx2x_set_next_page_rx_bd(fp);
4372 
4373 		/* CQ ring */
4374 		bnx2x_set_next_page_rx_cq(fp);
4375 
4376 		/* BDs */
4377 		ring_size = bnx2x_alloc_rx_bds(fp, rx_ring_size);
4378 		if (ring_size < rx_ring_size)
4379 			goto alloc_mem_err;
4380 	}
4381 
4382 	return 0;
4383 
4384 /* handles low memory cases */
4385 alloc_mem_err:
4386 	BNX2X_ERR("Unable to allocate full memory for queue %d (size %d)\n",
4387 						index, ring_size);
4388 	/* FW will drop all packets if queue is not big enough,
4389 	 * In these cases we disable the queue
4390 	 * Min size is different for OOO, TPA and non-TPA queues
4391 	 */
4392 	if (ring_size < (fp->disable_tpa ?
4393 				MIN_RX_SIZE_NONTPA : MIN_RX_SIZE_TPA)) {
4394 			/* release memory allocated for this queue */
4395 			bnx2x_free_fp_mem_at(bp, index);
4396 			return -ENOMEM;
4397 	}
4398 	return 0;
4399 }
4400 
4401 int bnx2x_alloc_fp_mem_cnic(struct bnx2x *bp)
4402 {
4403 	if (!NO_FCOE(bp))
4404 		/* FCoE */
4405 		if (bnx2x_alloc_fp_mem_at(bp, FCOE_IDX(bp)))
4406 			/* we will fail load process instead of mark
4407 			 * NO_FCOE_FLAG
4408 			 */
4409 			return -ENOMEM;
4410 
4411 	return 0;
4412 }
4413 
4414 int bnx2x_alloc_fp_mem(struct bnx2x *bp)
4415 {
4416 	int i;
4417 
4418 	/* 1. Allocate FP for leading - fatal if error
4419 	 * 2. Allocate RSS - fix number of queues if error
4420 	 */
4421 
4422 	/* leading */
4423 	if (bnx2x_alloc_fp_mem_at(bp, 0))
4424 		return -ENOMEM;
4425 
4426 	/* RSS */
4427 	for_each_nondefault_eth_queue(bp, i)
4428 		if (bnx2x_alloc_fp_mem_at(bp, i))
4429 			break;
4430 
4431 	/* handle memory failures */
4432 	if (i != BNX2X_NUM_ETH_QUEUES(bp)) {
4433 		int delta = BNX2X_NUM_ETH_QUEUES(bp) - i;
4434 
4435 		WARN_ON(delta < 0);
4436 		bnx2x_shrink_eth_fp(bp, delta);
4437 		if (CNIC_SUPPORT(bp))
4438 			/* move non eth FPs next to last eth FP
4439 			 * must be done in that order
4440 			 * FCOE_IDX < FWD_IDX < OOO_IDX
4441 			 */
4442 
4443 			/* move FCoE fp even NO_FCOE_FLAG is on */
4444 			bnx2x_move_fp(bp, FCOE_IDX(bp), FCOE_IDX(bp) - delta);
4445 		bp->num_ethernet_queues -= delta;
4446 		bp->num_queues = bp->num_ethernet_queues +
4447 				 bp->num_cnic_queues;
4448 		BNX2X_ERR("Adjusted num of queues from %d to %d\n",
4449 			  bp->num_queues + delta, bp->num_queues);
4450 	}
4451 
4452 	return 0;
4453 }
4454 
4455 void bnx2x_free_mem_bp(struct bnx2x *bp)
4456 {
4457 	int i;
4458 
4459 	for (i = 0; i < bp->fp_array_size; i++)
4460 		kfree(bp->fp[i].tpa_info);
4461 	kfree(bp->fp);
4462 	kfree(bp->sp_objs);
4463 	kfree(bp->fp_stats);
4464 	kfree(bp->bnx2x_txq);
4465 	kfree(bp->msix_table);
4466 	kfree(bp->ilt);
4467 }
4468 
4469 int bnx2x_alloc_mem_bp(struct bnx2x *bp)
4470 {
4471 	struct bnx2x_fastpath *fp;
4472 	struct msix_entry *tbl;
4473 	struct bnx2x_ilt *ilt;
4474 	int msix_table_size = 0;
4475 	int fp_array_size, txq_array_size;
4476 	int i;
4477 
4478 	/*
4479 	 * The biggest MSI-X table we might need is as a maximum number of fast
4480 	 * path IGU SBs plus default SB (for PF only).
4481 	 */
4482 	msix_table_size = bp->igu_sb_cnt;
4483 	if (IS_PF(bp))
4484 		msix_table_size++;
4485 	BNX2X_DEV_INFO("msix_table_size %d\n", msix_table_size);
4486 
4487 	/* fp array: RSS plus CNIC related L2 queues */
4488 	fp_array_size = BNX2X_MAX_RSS_COUNT(bp) + CNIC_SUPPORT(bp);
4489 	bp->fp_array_size = fp_array_size;
4490 	BNX2X_DEV_INFO("fp_array_size %d\n", bp->fp_array_size);
4491 
4492 	fp = kcalloc(bp->fp_array_size, sizeof(*fp), GFP_KERNEL);
4493 	if (!fp)
4494 		goto alloc_err;
4495 	for (i = 0; i < bp->fp_array_size; i++) {
4496 		fp[i].tpa_info =
4497 			kcalloc(ETH_MAX_AGGREGATION_QUEUES_E1H_E2,
4498 				sizeof(struct bnx2x_agg_info), GFP_KERNEL);
4499 		if (!(fp[i].tpa_info))
4500 			goto alloc_err;
4501 	}
4502 
4503 	bp->fp = fp;
4504 
4505 	/* allocate sp objs */
4506 	bp->sp_objs = kcalloc(bp->fp_array_size, sizeof(struct bnx2x_sp_objs),
4507 			      GFP_KERNEL);
4508 	if (!bp->sp_objs)
4509 		goto alloc_err;
4510 
4511 	/* allocate fp_stats */
4512 	bp->fp_stats = kcalloc(bp->fp_array_size, sizeof(struct bnx2x_fp_stats),
4513 			       GFP_KERNEL);
4514 	if (!bp->fp_stats)
4515 		goto alloc_err;
4516 
4517 	/* Allocate memory for the transmission queues array */
4518 	txq_array_size =
4519 		BNX2X_MAX_RSS_COUNT(bp) * BNX2X_MULTI_TX_COS + CNIC_SUPPORT(bp);
4520 	BNX2X_DEV_INFO("txq_array_size %d", txq_array_size);
4521 
4522 	bp->bnx2x_txq = kcalloc(txq_array_size, sizeof(struct bnx2x_fp_txdata),
4523 				GFP_KERNEL);
4524 	if (!bp->bnx2x_txq)
4525 		goto alloc_err;
4526 
4527 	/* msix table */
4528 	tbl = kcalloc(msix_table_size, sizeof(*tbl), GFP_KERNEL);
4529 	if (!tbl)
4530 		goto alloc_err;
4531 	bp->msix_table = tbl;
4532 
4533 	/* ilt */
4534 	ilt = kzalloc(sizeof(*ilt), GFP_KERNEL);
4535 	if (!ilt)
4536 		goto alloc_err;
4537 	bp->ilt = ilt;
4538 
4539 	return 0;
4540 alloc_err:
4541 	bnx2x_free_mem_bp(bp);
4542 	return -ENOMEM;
4543 }
4544 
4545 int bnx2x_reload_if_running(struct net_device *dev)
4546 {
4547 	struct bnx2x *bp = netdev_priv(dev);
4548 
4549 	if (unlikely(!netif_running(dev)))
4550 		return 0;
4551 
4552 	bnx2x_nic_unload(bp, UNLOAD_NORMAL, true);
4553 	return bnx2x_nic_load(bp, LOAD_NORMAL);
4554 }
4555 
4556 int bnx2x_get_cur_phy_idx(struct bnx2x *bp)
4557 {
4558 	u32 sel_phy_idx = 0;
4559 	if (bp->link_params.num_phys <= 1)
4560 		return INT_PHY;
4561 
4562 	if (bp->link_vars.link_up) {
4563 		sel_phy_idx = EXT_PHY1;
4564 		/* In case link is SERDES, check if the EXT_PHY2 is the one */
4565 		if ((bp->link_vars.link_status & LINK_STATUS_SERDES_LINK) &&
4566 		    (bp->link_params.phy[EXT_PHY2].supported & SUPPORTED_FIBRE))
4567 			sel_phy_idx = EXT_PHY2;
4568 	} else {
4569 
4570 		switch (bnx2x_phy_selection(&bp->link_params)) {
4571 		case PORT_HW_CFG_PHY_SELECTION_HARDWARE_DEFAULT:
4572 		case PORT_HW_CFG_PHY_SELECTION_FIRST_PHY:
4573 		case PORT_HW_CFG_PHY_SELECTION_FIRST_PHY_PRIORITY:
4574 		       sel_phy_idx = EXT_PHY1;
4575 		       break;
4576 		case PORT_HW_CFG_PHY_SELECTION_SECOND_PHY:
4577 		case PORT_HW_CFG_PHY_SELECTION_SECOND_PHY_PRIORITY:
4578 		       sel_phy_idx = EXT_PHY2;
4579 		       break;
4580 		}
4581 	}
4582 
4583 	return sel_phy_idx;
4584 }
4585 int bnx2x_get_link_cfg_idx(struct bnx2x *bp)
4586 {
4587 	u32 sel_phy_idx = bnx2x_get_cur_phy_idx(bp);
4588 	/*
4589 	 * The selected activated PHY is always after swapping (in case PHY
4590 	 * swapping is enabled). So when swapping is enabled, we need to reverse
4591 	 * the configuration
4592 	 */
4593 
4594 	if (bp->link_params.multi_phy_config &
4595 	    PORT_HW_CFG_PHY_SWAPPED_ENABLED) {
4596 		if (sel_phy_idx == EXT_PHY1)
4597 			sel_phy_idx = EXT_PHY2;
4598 		else if (sel_phy_idx == EXT_PHY2)
4599 			sel_phy_idx = EXT_PHY1;
4600 	}
4601 	return LINK_CONFIG_IDX(sel_phy_idx);
4602 }
4603 
4604 #ifdef NETDEV_FCOE_WWNN
4605 int bnx2x_fcoe_get_wwn(struct net_device *dev, u64 *wwn, int type)
4606 {
4607 	struct bnx2x *bp = netdev_priv(dev);
4608 	struct cnic_eth_dev *cp = &bp->cnic_eth_dev;
4609 
4610 	switch (type) {
4611 	case NETDEV_FCOE_WWNN:
4612 		*wwn = HILO_U64(cp->fcoe_wwn_node_name_hi,
4613 				cp->fcoe_wwn_node_name_lo);
4614 		break;
4615 	case NETDEV_FCOE_WWPN:
4616 		*wwn = HILO_U64(cp->fcoe_wwn_port_name_hi,
4617 				cp->fcoe_wwn_port_name_lo);
4618 		break;
4619 	default:
4620 		BNX2X_ERR("Wrong WWN type requested - %d\n", type);
4621 		return -EINVAL;
4622 	}
4623 
4624 	return 0;
4625 }
4626 #endif
4627 
4628 /* called with rtnl_lock */
4629 int bnx2x_change_mtu(struct net_device *dev, int new_mtu)
4630 {
4631 	struct bnx2x *bp = netdev_priv(dev);
4632 
4633 	if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
4634 		BNX2X_ERR("Can't perform change MTU during parity recovery\n");
4635 		return -EAGAIN;
4636 	}
4637 
4638 	if ((new_mtu > ETH_MAX_JUMBO_PACKET_SIZE) ||
4639 	    ((new_mtu + ETH_HLEN) < ETH_MIN_PACKET_SIZE)) {
4640 		BNX2X_ERR("Can't support requested MTU size\n");
4641 		return -EINVAL;
4642 	}
4643 
4644 	/* This does not race with packet allocation
4645 	 * because the actual alloc size is
4646 	 * only updated as part of load
4647 	 */
4648 	dev->mtu = new_mtu;
4649 
4650 	return bnx2x_reload_if_running(dev);
4651 }
4652 
4653 netdev_features_t bnx2x_fix_features(struct net_device *dev,
4654 				     netdev_features_t features)
4655 {
4656 	struct bnx2x *bp = netdev_priv(dev);
4657 
4658 	/* TPA requires Rx CSUM offloading */
4659 	if (!(features & NETIF_F_RXCSUM) || bp->disable_tpa) {
4660 		features &= ~NETIF_F_LRO;
4661 		features &= ~NETIF_F_GRO;
4662 	}
4663 
4664 	return features;
4665 }
4666 
4667 int bnx2x_set_features(struct net_device *dev, netdev_features_t features)
4668 {
4669 	struct bnx2x *bp = netdev_priv(dev);
4670 	u32 flags = bp->flags;
4671 	u32 changes;
4672 	bool bnx2x_reload = false;
4673 
4674 	if (features & NETIF_F_LRO)
4675 		flags |= TPA_ENABLE_FLAG;
4676 	else
4677 		flags &= ~TPA_ENABLE_FLAG;
4678 
4679 	if (features & NETIF_F_GRO)
4680 		flags |= GRO_ENABLE_FLAG;
4681 	else
4682 		flags &= ~GRO_ENABLE_FLAG;
4683 
4684 	if (features & NETIF_F_LOOPBACK) {
4685 		if (bp->link_params.loopback_mode != LOOPBACK_BMAC) {
4686 			bp->link_params.loopback_mode = LOOPBACK_BMAC;
4687 			bnx2x_reload = true;
4688 		}
4689 	} else {
4690 		if (bp->link_params.loopback_mode != LOOPBACK_NONE) {
4691 			bp->link_params.loopback_mode = LOOPBACK_NONE;
4692 			bnx2x_reload = true;
4693 		}
4694 	}
4695 
4696 	changes = flags ^ bp->flags;
4697 
4698 	/* if GRO is changed while LRO is enabled, don't force a reload */
4699 	if ((changes & GRO_ENABLE_FLAG) && (flags & TPA_ENABLE_FLAG))
4700 		changes &= ~GRO_ENABLE_FLAG;
4701 
4702 	if (changes)
4703 		bnx2x_reload = true;
4704 
4705 	bp->flags = flags;
4706 
4707 	if (bnx2x_reload) {
4708 		if (bp->recovery_state == BNX2X_RECOVERY_DONE)
4709 			return bnx2x_reload_if_running(dev);
4710 		/* else: bnx2x_nic_load() will be called at end of recovery */
4711 	}
4712 
4713 	return 0;
4714 }
4715 
4716 void bnx2x_tx_timeout(struct net_device *dev)
4717 {
4718 	struct bnx2x *bp = netdev_priv(dev);
4719 
4720 #ifdef BNX2X_STOP_ON_ERROR
4721 	if (!bp->panic)
4722 		bnx2x_panic();
4723 #endif
4724 
4725 	smp_mb__before_clear_bit();
4726 	set_bit(BNX2X_SP_RTNL_TX_TIMEOUT, &bp->sp_rtnl_state);
4727 	smp_mb__after_clear_bit();
4728 
4729 	/* This allows the netif to be shutdown gracefully before resetting */
4730 	schedule_delayed_work(&bp->sp_rtnl_task, 0);
4731 }
4732 
4733 int bnx2x_suspend(struct pci_dev *pdev, pm_message_t state)
4734 {
4735 	struct net_device *dev = pci_get_drvdata(pdev);
4736 	struct bnx2x *bp;
4737 
4738 	if (!dev) {
4739 		dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
4740 		return -ENODEV;
4741 	}
4742 	bp = netdev_priv(dev);
4743 
4744 	rtnl_lock();
4745 
4746 	pci_save_state(pdev);
4747 
4748 	if (!netif_running(dev)) {
4749 		rtnl_unlock();
4750 		return 0;
4751 	}
4752 
4753 	netif_device_detach(dev);
4754 
4755 	bnx2x_nic_unload(bp, UNLOAD_CLOSE, false);
4756 
4757 	bnx2x_set_power_state(bp, pci_choose_state(pdev, state));
4758 
4759 	rtnl_unlock();
4760 
4761 	return 0;
4762 }
4763 
4764 int bnx2x_resume(struct pci_dev *pdev)
4765 {
4766 	struct net_device *dev = pci_get_drvdata(pdev);
4767 	struct bnx2x *bp;
4768 	int rc;
4769 
4770 	if (!dev) {
4771 		dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
4772 		return -ENODEV;
4773 	}
4774 	bp = netdev_priv(dev);
4775 
4776 	if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
4777 		BNX2X_ERR("Handling parity error recovery. Try again later\n");
4778 		return -EAGAIN;
4779 	}
4780 
4781 	rtnl_lock();
4782 
4783 	pci_restore_state(pdev);
4784 
4785 	if (!netif_running(dev)) {
4786 		rtnl_unlock();
4787 		return 0;
4788 	}
4789 
4790 	bnx2x_set_power_state(bp, PCI_D0);
4791 	netif_device_attach(dev);
4792 
4793 	rc = bnx2x_nic_load(bp, LOAD_OPEN);
4794 
4795 	rtnl_unlock();
4796 
4797 	return rc;
4798 }
4799 
4800 void bnx2x_set_ctx_validation(struct bnx2x *bp, struct eth_context *cxt,
4801 			      u32 cid)
4802 {
4803 	if (!cxt) {
4804 		BNX2X_ERR("bad context pointer %p\n", cxt);
4805 		return;
4806 	}
4807 
4808 	/* ustorm cxt validation */
4809 	cxt->ustorm_ag_context.cdu_usage =
4810 		CDU_RSRVD_VALUE_TYPE_A(HW_CID(bp, cid),
4811 			CDU_REGION_NUMBER_UCM_AG, ETH_CONNECTION_TYPE);
4812 	/* xcontext validation */
4813 	cxt->xstorm_ag_context.cdu_reserved =
4814 		CDU_RSRVD_VALUE_TYPE_A(HW_CID(bp, cid),
4815 			CDU_REGION_NUMBER_XCM_AG, ETH_CONNECTION_TYPE);
4816 }
4817 
4818 static void storm_memset_hc_timeout(struct bnx2x *bp, u8 port,
4819 				    u8 fw_sb_id, u8 sb_index,
4820 				    u8 ticks)
4821 {
4822 	u32 addr = BAR_CSTRORM_INTMEM +
4823 		   CSTORM_STATUS_BLOCK_DATA_TIMEOUT_OFFSET(fw_sb_id, sb_index);
4824 	REG_WR8(bp, addr, ticks);
4825 	DP(NETIF_MSG_IFUP,
4826 	   "port %x fw_sb_id %d sb_index %d ticks %d\n",
4827 	   port, fw_sb_id, sb_index, ticks);
4828 }
4829 
4830 static void storm_memset_hc_disable(struct bnx2x *bp, u8 port,
4831 				    u16 fw_sb_id, u8 sb_index,
4832 				    u8 disable)
4833 {
4834 	u32 enable_flag = disable ? 0 : (1 << HC_INDEX_DATA_HC_ENABLED_SHIFT);
4835 	u32 addr = BAR_CSTRORM_INTMEM +
4836 		   CSTORM_STATUS_BLOCK_DATA_FLAGS_OFFSET(fw_sb_id, sb_index);
4837 	u8 flags = REG_RD8(bp, addr);
4838 	/* clear and set */
4839 	flags &= ~HC_INDEX_DATA_HC_ENABLED;
4840 	flags |= enable_flag;
4841 	REG_WR8(bp, addr, flags);
4842 	DP(NETIF_MSG_IFUP,
4843 	   "port %x fw_sb_id %d sb_index %d disable %d\n",
4844 	   port, fw_sb_id, sb_index, disable);
4845 }
4846 
4847 void bnx2x_update_coalesce_sb_index(struct bnx2x *bp, u8 fw_sb_id,
4848 				    u8 sb_index, u8 disable, u16 usec)
4849 {
4850 	int port = BP_PORT(bp);
4851 	u8 ticks = usec / BNX2X_BTR;
4852 
4853 	storm_memset_hc_timeout(bp, port, fw_sb_id, sb_index, ticks);
4854 
4855 	disable = disable ? 1 : (usec ? 0 : 1);
4856 	storm_memset_hc_disable(bp, port, fw_sb_id, sb_index, disable);
4857 }
4858