xref: /openbmc/linux/drivers/net/ethernet/broadcom/bnx2x/bnx2x_cmn.c (revision 029f7f3b8701cc7aca8bdb31f0c7edd6a479e357)
1 /* bnx2x_cmn.c: QLogic Everest network driver.
2  *
3  * Copyright (c) 2007-2013 Broadcom Corporation
4  * Copyright (c) 2014 QLogic Corporation
5  * All rights reserved
6  *
7  * This program is free software; you can redistribute it and/or modify
8  * it under the terms of the GNU General Public License as published by
9  * the Free Software Foundation.
10  *
11  * Maintained by: Ariel Elior <ariel.elior@qlogic.com>
12  * Written by: Eliezer Tamir
13  * Based on code from Michael Chan's bnx2 driver
14  * UDP CSUM errata workaround by Arik Gendelman
15  * Slowpath and fastpath rework by Vladislav Zolotarov
16  * Statistics and Link management by Yitchak Gertner
17  *
18  */
19 
20 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
21 
22 #include <linux/etherdevice.h>
23 #include <linux/if_vlan.h>
24 #include <linux/interrupt.h>
25 #include <linux/ip.h>
26 #include <linux/crash_dump.h>
27 #include <net/tcp.h>
28 #include <net/ipv6.h>
29 #include <net/ip6_checksum.h>
30 #include <net/busy_poll.h>
31 #include <linux/prefetch.h>
32 #include "bnx2x_cmn.h"
33 #include "bnx2x_init.h"
34 #include "bnx2x_sp.h"
35 
36 static void bnx2x_free_fp_mem_cnic(struct bnx2x *bp);
37 static int bnx2x_alloc_fp_mem_cnic(struct bnx2x *bp);
38 static int bnx2x_alloc_fp_mem(struct bnx2x *bp);
39 static int bnx2x_poll(struct napi_struct *napi, int budget);
40 
41 static void bnx2x_add_all_napi_cnic(struct bnx2x *bp)
42 {
43 	int i;
44 
45 	/* Add NAPI objects */
46 	for_each_rx_queue_cnic(bp, i) {
47 		netif_napi_add(bp->dev, &bnx2x_fp(bp, i, napi),
48 			       bnx2x_poll, NAPI_POLL_WEIGHT);
49 	}
50 }
51 
52 static void bnx2x_add_all_napi(struct bnx2x *bp)
53 {
54 	int i;
55 
56 	/* Add NAPI objects */
57 	for_each_eth_queue(bp, i) {
58 		netif_napi_add(bp->dev, &bnx2x_fp(bp, i, napi),
59 			       bnx2x_poll, NAPI_POLL_WEIGHT);
60 	}
61 }
62 
63 static int bnx2x_calc_num_queues(struct bnx2x *bp)
64 {
65 	int nq = bnx2x_num_queues ? : netif_get_num_default_rss_queues();
66 
67 	/* Reduce memory usage in kdump environment by using only one queue */
68 	if (is_kdump_kernel())
69 		nq = 1;
70 
71 	nq = clamp(nq, 1, BNX2X_MAX_QUEUES(bp));
72 	return nq;
73 }
74 
75 /**
76  * bnx2x_move_fp - move content of the fastpath structure.
77  *
78  * @bp:		driver handle
79  * @from:	source FP index
80  * @to:		destination FP index
81  *
82  * Makes sure the contents of the bp->fp[to].napi is kept
83  * intact. This is done by first copying the napi struct from
84  * the target to the source, and then mem copying the entire
85  * source onto the target. Update txdata pointers and related
86  * content.
87  */
88 static inline void bnx2x_move_fp(struct bnx2x *bp, int from, int to)
89 {
90 	struct bnx2x_fastpath *from_fp = &bp->fp[from];
91 	struct bnx2x_fastpath *to_fp = &bp->fp[to];
92 	struct bnx2x_sp_objs *from_sp_objs = &bp->sp_objs[from];
93 	struct bnx2x_sp_objs *to_sp_objs = &bp->sp_objs[to];
94 	struct bnx2x_fp_stats *from_fp_stats = &bp->fp_stats[from];
95 	struct bnx2x_fp_stats *to_fp_stats = &bp->fp_stats[to];
96 	int old_max_eth_txqs, new_max_eth_txqs;
97 	int old_txdata_index = 0, new_txdata_index = 0;
98 	struct bnx2x_agg_info *old_tpa_info = to_fp->tpa_info;
99 
100 	/* Copy the NAPI object as it has been already initialized */
101 	from_fp->napi = to_fp->napi;
102 
103 	/* Move bnx2x_fastpath contents */
104 	memcpy(to_fp, from_fp, sizeof(*to_fp));
105 	to_fp->index = to;
106 
107 	/* Retain the tpa_info of the original `to' version as we don't want
108 	 * 2 FPs to contain the same tpa_info pointer.
109 	 */
110 	to_fp->tpa_info = old_tpa_info;
111 
112 	/* move sp_objs contents as well, as their indices match fp ones */
113 	memcpy(to_sp_objs, from_sp_objs, sizeof(*to_sp_objs));
114 
115 	/* move fp_stats contents as well, as their indices match fp ones */
116 	memcpy(to_fp_stats, from_fp_stats, sizeof(*to_fp_stats));
117 
118 	/* Update txdata pointers in fp and move txdata content accordingly:
119 	 * Each fp consumes 'max_cos' txdata structures, so the index should be
120 	 * decremented by max_cos x delta.
121 	 */
122 
123 	old_max_eth_txqs = BNX2X_NUM_ETH_QUEUES(bp) * (bp)->max_cos;
124 	new_max_eth_txqs = (BNX2X_NUM_ETH_QUEUES(bp) - from + to) *
125 				(bp)->max_cos;
126 	if (from == FCOE_IDX(bp)) {
127 		old_txdata_index = old_max_eth_txqs + FCOE_TXQ_IDX_OFFSET;
128 		new_txdata_index = new_max_eth_txqs + FCOE_TXQ_IDX_OFFSET;
129 	}
130 
131 	memcpy(&bp->bnx2x_txq[new_txdata_index],
132 	       &bp->bnx2x_txq[old_txdata_index],
133 	       sizeof(struct bnx2x_fp_txdata));
134 	to_fp->txdata_ptr[0] = &bp->bnx2x_txq[new_txdata_index];
135 }
136 
137 /**
138  * bnx2x_fill_fw_str - Fill buffer with FW version string.
139  *
140  * @bp:        driver handle
141  * @buf:       character buffer to fill with the fw name
142  * @buf_len:   length of the above buffer
143  *
144  */
145 void bnx2x_fill_fw_str(struct bnx2x *bp, char *buf, size_t buf_len)
146 {
147 	if (IS_PF(bp)) {
148 		u8 phy_fw_ver[PHY_FW_VER_LEN];
149 
150 		phy_fw_ver[0] = '\0';
151 		bnx2x_get_ext_phy_fw_version(&bp->link_params,
152 					     phy_fw_ver, PHY_FW_VER_LEN);
153 		strlcpy(buf, bp->fw_ver, buf_len);
154 		snprintf(buf + strlen(bp->fw_ver), 32 - strlen(bp->fw_ver),
155 			 "bc %d.%d.%d%s%s",
156 			 (bp->common.bc_ver & 0xff0000) >> 16,
157 			 (bp->common.bc_ver & 0xff00) >> 8,
158 			 (bp->common.bc_ver & 0xff),
159 			 ((phy_fw_ver[0] != '\0') ? " phy " : ""), phy_fw_ver);
160 	} else {
161 		bnx2x_vf_fill_fw_str(bp, buf, buf_len);
162 	}
163 }
164 
165 /**
166  * bnx2x_shrink_eth_fp - guarantees fastpath structures stay intact
167  *
168  * @bp:	driver handle
169  * @delta:	number of eth queues which were not allocated
170  */
171 static void bnx2x_shrink_eth_fp(struct bnx2x *bp, int delta)
172 {
173 	int i, cos, old_eth_num = BNX2X_NUM_ETH_QUEUES(bp);
174 
175 	/* Queue pointer cannot be re-set on an fp-basis, as moving pointer
176 	 * backward along the array could cause memory to be overridden
177 	 */
178 	for (cos = 1; cos < bp->max_cos; cos++) {
179 		for (i = 0; i < old_eth_num - delta; i++) {
180 			struct bnx2x_fastpath *fp = &bp->fp[i];
181 			int new_idx = cos * (old_eth_num - delta) + i;
182 
183 			memcpy(&bp->bnx2x_txq[new_idx], fp->txdata_ptr[cos],
184 			       sizeof(struct bnx2x_fp_txdata));
185 			fp->txdata_ptr[cos] = &bp->bnx2x_txq[new_idx];
186 		}
187 	}
188 }
189 
190 int bnx2x_load_count[2][3] = { {0} }; /* per-path: 0-common, 1-port0, 2-port1 */
191 
192 /* free skb in the packet ring at pos idx
193  * return idx of last bd freed
194  */
195 static u16 bnx2x_free_tx_pkt(struct bnx2x *bp, struct bnx2x_fp_txdata *txdata,
196 			     u16 idx, unsigned int *pkts_compl,
197 			     unsigned int *bytes_compl)
198 {
199 	struct sw_tx_bd *tx_buf = &txdata->tx_buf_ring[idx];
200 	struct eth_tx_start_bd *tx_start_bd;
201 	struct eth_tx_bd *tx_data_bd;
202 	struct sk_buff *skb = tx_buf->skb;
203 	u16 bd_idx = TX_BD(tx_buf->first_bd), new_cons;
204 	int nbd;
205 	u16 split_bd_len = 0;
206 
207 	/* prefetch skb end pointer to speedup dev_kfree_skb() */
208 	prefetch(&skb->end);
209 
210 	DP(NETIF_MSG_TX_DONE, "fp[%d]: pkt_idx %d  buff @(%p)->skb %p\n",
211 	   txdata->txq_index, idx, tx_buf, skb);
212 
213 	tx_start_bd = &txdata->tx_desc_ring[bd_idx].start_bd;
214 
215 	nbd = le16_to_cpu(tx_start_bd->nbd) - 1;
216 #ifdef BNX2X_STOP_ON_ERROR
217 	if ((nbd - 1) > (MAX_SKB_FRAGS + 2)) {
218 		BNX2X_ERR("BAD nbd!\n");
219 		bnx2x_panic();
220 	}
221 #endif
222 	new_cons = nbd + tx_buf->first_bd;
223 
224 	/* Get the next bd */
225 	bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
226 
227 	/* Skip a parse bd... */
228 	--nbd;
229 	bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
230 
231 	if (tx_buf->flags & BNX2X_HAS_SECOND_PBD) {
232 		/* Skip second parse bd... */
233 		--nbd;
234 		bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
235 	}
236 
237 	/* TSO headers+data bds share a common mapping. See bnx2x_tx_split() */
238 	if (tx_buf->flags & BNX2X_TSO_SPLIT_BD) {
239 		tx_data_bd = &txdata->tx_desc_ring[bd_idx].reg_bd;
240 		split_bd_len = BD_UNMAP_LEN(tx_data_bd);
241 		--nbd;
242 		bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
243 	}
244 
245 	/* unmap first bd */
246 	dma_unmap_single(&bp->pdev->dev, BD_UNMAP_ADDR(tx_start_bd),
247 			 BD_UNMAP_LEN(tx_start_bd) + split_bd_len,
248 			 DMA_TO_DEVICE);
249 
250 	/* now free frags */
251 	while (nbd > 0) {
252 
253 		tx_data_bd = &txdata->tx_desc_ring[bd_idx].reg_bd;
254 		dma_unmap_page(&bp->pdev->dev, BD_UNMAP_ADDR(tx_data_bd),
255 			       BD_UNMAP_LEN(tx_data_bd), DMA_TO_DEVICE);
256 		if (--nbd)
257 			bd_idx = TX_BD(NEXT_TX_IDX(bd_idx));
258 	}
259 
260 	/* release skb */
261 	WARN_ON(!skb);
262 	if (likely(skb)) {
263 		(*pkts_compl)++;
264 		(*bytes_compl) += skb->len;
265 		dev_kfree_skb_any(skb);
266 	}
267 
268 	tx_buf->first_bd = 0;
269 	tx_buf->skb = NULL;
270 
271 	return new_cons;
272 }
273 
274 int bnx2x_tx_int(struct bnx2x *bp, struct bnx2x_fp_txdata *txdata)
275 {
276 	struct netdev_queue *txq;
277 	u16 hw_cons, sw_cons, bd_cons = txdata->tx_bd_cons;
278 	unsigned int pkts_compl = 0, bytes_compl = 0;
279 
280 #ifdef BNX2X_STOP_ON_ERROR
281 	if (unlikely(bp->panic))
282 		return -1;
283 #endif
284 
285 	txq = netdev_get_tx_queue(bp->dev, txdata->txq_index);
286 	hw_cons = le16_to_cpu(*txdata->tx_cons_sb);
287 	sw_cons = txdata->tx_pkt_cons;
288 
289 	while (sw_cons != hw_cons) {
290 		u16 pkt_cons;
291 
292 		pkt_cons = TX_BD(sw_cons);
293 
294 		DP(NETIF_MSG_TX_DONE,
295 		   "queue[%d]: hw_cons %u  sw_cons %u  pkt_cons %u\n",
296 		   txdata->txq_index, hw_cons, sw_cons, pkt_cons);
297 
298 		bd_cons = bnx2x_free_tx_pkt(bp, txdata, pkt_cons,
299 					    &pkts_compl, &bytes_compl);
300 
301 		sw_cons++;
302 	}
303 
304 	netdev_tx_completed_queue(txq, pkts_compl, bytes_compl);
305 
306 	txdata->tx_pkt_cons = sw_cons;
307 	txdata->tx_bd_cons = bd_cons;
308 
309 	/* Need to make the tx_bd_cons update visible to start_xmit()
310 	 * before checking for netif_tx_queue_stopped().  Without the
311 	 * memory barrier, there is a small possibility that
312 	 * start_xmit() will miss it and cause the queue to be stopped
313 	 * forever.
314 	 * On the other hand we need an rmb() here to ensure the proper
315 	 * ordering of bit testing in the following
316 	 * netif_tx_queue_stopped(txq) call.
317 	 */
318 	smp_mb();
319 
320 	if (unlikely(netif_tx_queue_stopped(txq))) {
321 		/* Taking tx_lock() is needed to prevent re-enabling the queue
322 		 * while it's empty. This could have happen if rx_action() gets
323 		 * suspended in bnx2x_tx_int() after the condition before
324 		 * netif_tx_wake_queue(), while tx_action (bnx2x_start_xmit()):
325 		 *
326 		 * stops the queue->sees fresh tx_bd_cons->releases the queue->
327 		 * sends some packets consuming the whole queue again->
328 		 * stops the queue
329 		 */
330 
331 		__netif_tx_lock(txq, smp_processor_id());
332 
333 		if ((netif_tx_queue_stopped(txq)) &&
334 		    (bp->state == BNX2X_STATE_OPEN) &&
335 		    (bnx2x_tx_avail(bp, txdata) >= MAX_DESC_PER_TX_PKT))
336 			netif_tx_wake_queue(txq);
337 
338 		__netif_tx_unlock(txq);
339 	}
340 	return 0;
341 }
342 
343 static inline void bnx2x_update_last_max_sge(struct bnx2x_fastpath *fp,
344 					     u16 idx)
345 {
346 	u16 last_max = fp->last_max_sge;
347 
348 	if (SUB_S16(idx, last_max) > 0)
349 		fp->last_max_sge = idx;
350 }
351 
352 static inline void bnx2x_update_sge_prod(struct bnx2x_fastpath *fp,
353 					 u16 sge_len,
354 					 struct eth_end_agg_rx_cqe *cqe)
355 {
356 	struct bnx2x *bp = fp->bp;
357 	u16 last_max, last_elem, first_elem;
358 	u16 delta = 0;
359 	u16 i;
360 
361 	if (!sge_len)
362 		return;
363 
364 	/* First mark all used pages */
365 	for (i = 0; i < sge_len; i++)
366 		BIT_VEC64_CLEAR_BIT(fp->sge_mask,
367 			RX_SGE(le16_to_cpu(cqe->sgl_or_raw_data.sgl[i])));
368 
369 	DP(NETIF_MSG_RX_STATUS, "fp_cqe->sgl[%d] = %d\n",
370 	   sge_len - 1, le16_to_cpu(cqe->sgl_or_raw_data.sgl[sge_len - 1]));
371 
372 	/* Here we assume that the last SGE index is the biggest */
373 	prefetch((void *)(fp->sge_mask));
374 	bnx2x_update_last_max_sge(fp,
375 		le16_to_cpu(cqe->sgl_or_raw_data.sgl[sge_len - 1]));
376 
377 	last_max = RX_SGE(fp->last_max_sge);
378 	last_elem = last_max >> BIT_VEC64_ELEM_SHIFT;
379 	first_elem = RX_SGE(fp->rx_sge_prod) >> BIT_VEC64_ELEM_SHIFT;
380 
381 	/* If ring is not full */
382 	if (last_elem + 1 != first_elem)
383 		last_elem++;
384 
385 	/* Now update the prod */
386 	for (i = first_elem; i != last_elem; i = NEXT_SGE_MASK_ELEM(i)) {
387 		if (likely(fp->sge_mask[i]))
388 			break;
389 
390 		fp->sge_mask[i] = BIT_VEC64_ELEM_ONE_MASK;
391 		delta += BIT_VEC64_ELEM_SZ;
392 	}
393 
394 	if (delta > 0) {
395 		fp->rx_sge_prod += delta;
396 		/* clear page-end entries */
397 		bnx2x_clear_sge_mask_next_elems(fp);
398 	}
399 
400 	DP(NETIF_MSG_RX_STATUS,
401 	   "fp->last_max_sge = %d  fp->rx_sge_prod = %d\n",
402 	   fp->last_max_sge, fp->rx_sge_prod);
403 }
404 
405 /* Get Toeplitz hash value in the skb using the value from the
406  * CQE (calculated by HW).
407  */
408 static u32 bnx2x_get_rxhash(const struct bnx2x *bp,
409 			    const struct eth_fast_path_rx_cqe *cqe,
410 			    enum pkt_hash_types *rxhash_type)
411 {
412 	/* Get Toeplitz hash from CQE */
413 	if ((bp->dev->features & NETIF_F_RXHASH) &&
414 	    (cqe->status_flags & ETH_FAST_PATH_RX_CQE_RSS_HASH_FLG)) {
415 		enum eth_rss_hash_type htype;
416 
417 		htype = cqe->status_flags & ETH_FAST_PATH_RX_CQE_RSS_HASH_TYPE;
418 		*rxhash_type = ((htype == TCP_IPV4_HASH_TYPE) ||
419 				(htype == TCP_IPV6_HASH_TYPE)) ?
420 			       PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3;
421 
422 		return le32_to_cpu(cqe->rss_hash_result);
423 	}
424 	*rxhash_type = PKT_HASH_TYPE_NONE;
425 	return 0;
426 }
427 
428 static void bnx2x_tpa_start(struct bnx2x_fastpath *fp, u16 queue,
429 			    u16 cons, u16 prod,
430 			    struct eth_fast_path_rx_cqe *cqe)
431 {
432 	struct bnx2x *bp = fp->bp;
433 	struct sw_rx_bd *cons_rx_buf = &fp->rx_buf_ring[cons];
434 	struct sw_rx_bd *prod_rx_buf = &fp->rx_buf_ring[prod];
435 	struct eth_rx_bd *prod_bd = &fp->rx_desc_ring[prod];
436 	dma_addr_t mapping;
437 	struct bnx2x_agg_info *tpa_info = &fp->tpa_info[queue];
438 	struct sw_rx_bd *first_buf = &tpa_info->first_buf;
439 
440 	/* print error if current state != stop */
441 	if (tpa_info->tpa_state != BNX2X_TPA_STOP)
442 		BNX2X_ERR("start of bin not in stop [%d]\n", queue);
443 
444 	/* Try to map an empty data buffer from the aggregation info  */
445 	mapping = dma_map_single(&bp->pdev->dev,
446 				 first_buf->data + NET_SKB_PAD,
447 				 fp->rx_buf_size, DMA_FROM_DEVICE);
448 	/*
449 	 *  ...if it fails - move the skb from the consumer to the producer
450 	 *  and set the current aggregation state as ERROR to drop it
451 	 *  when TPA_STOP arrives.
452 	 */
453 
454 	if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
455 		/* Move the BD from the consumer to the producer */
456 		bnx2x_reuse_rx_data(fp, cons, prod);
457 		tpa_info->tpa_state = BNX2X_TPA_ERROR;
458 		return;
459 	}
460 
461 	/* move empty data from pool to prod */
462 	prod_rx_buf->data = first_buf->data;
463 	dma_unmap_addr_set(prod_rx_buf, mapping, mapping);
464 	/* point prod_bd to new data */
465 	prod_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
466 	prod_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
467 
468 	/* move partial skb from cons to pool (don't unmap yet) */
469 	*first_buf = *cons_rx_buf;
470 
471 	/* mark bin state as START */
472 	tpa_info->parsing_flags =
473 		le16_to_cpu(cqe->pars_flags.flags);
474 	tpa_info->vlan_tag = le16_to_cpu(cqe->vlan_tag);
475 	tpa_info->tpa_state = BNX2X_TPA_START;
476 	tpa_info->len_on_bd = le16_to_cpu(cqe->len_on_bd);
477 	tpa_info->placement_offset = cqe->placement_offset;
478 	tpa_info->rxhash = bnx2x_get_rxhash(bp, cqe, &tpa_info->rxhash_type);
479 	if (fp->mode == TPA_MODE_GRO) {
480 		u16 gro_size = le16_to_cpu(cqe->pkt_len_or_gro_seg_len);
481 		tpa_info->full_page = SGE_PAGES / gro_size * gro_size;
482 		tpa_info->gro_size = gro_size;
483 	}
484 
485 #ifdef BNX2X_STOP_ON_ERROR
486 	fp->tpa_queue_used |= (1 << queue);
487 	DP(NETIF_MSG_RX_STATUS, "fp->tpa_queue_used = 0x%llx\n",
488 	   fp->tpa_queue_used);
489 #endif
490 }
491 
492 /* Timestamp option length allowed for TPA aggregation:
493  *
494  *		nop nop kind length echo val
495  */
496 #define TPA_TSTAMP_OPT_LEN	12
497 /**
498  * bnx2x_set_gro_params - compute GRO values
499  *
500  * @skb:		packet skb
501  * @parsing_flags:	parsing flags from the START CQE
502  * @len_on_bd:		total length of the first packet for the
503  *			aggregation.
504  * @pkt_len:		length of all segments
505  *
506  * Approximate value of the MSS for this aggregation calculated using
507  * the first packet of it.
508  * Compute number of aggregated segments, and gso_type.
509  */
510 static void bnx2x_set_gro_params(struct sk_buff *skb, u16 parsing_flags,
511 				 u16 len_on_bd, unsigned int pkt_len,
512 				 u16 num_of_coalesced_segs)
513 {
514 	/* TPA aggregation won't have either IP options or TCP options
515 	 * other than timestamp or IPv6 extension headers.
516 	 */
517 	u16 hdrs_len = ETH_HLEN + sizeof(struct tcphdr);
518 
519 	if (GET_FLAG(parsing_flags, PARSING_FLAGS_OVER_ETHERNET_PROTOCOL) ==
520 	    PRS_FLAG_OVERETH_IPV6) {
521 		hdrs_len += sizeof(struct ipv6hdr);
522 		skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
523 	} else {
524 		hdrs_len += sizeof(struct iphdr);
525 		skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
526 	}
527 
528 	/* Check if there was a TCP timestamp, if there is it's will
529 	 * always be 12 bytes length: nop nop kind length echo val.
530 	 *
531 	 * Otherwise FW would close the aggregation.
532 	 */
533 	if (parsing_flags & PARSING_FLAGS_TIME_STAMP_EXIST_FLAG)
534 		hdrs_len += TPA_TSTAMP_OPT_LEN;
535 
536 	skb_shinfo(skb)->gso_size = len_on_bd - hdrs_len;
537 
538 	/* tcp_gro_complete() will copy NAPI_GRO_CB(skb)->count
539 	 * to skb_shinfo(skb)->gso_segs
540 	 */
541 	NAPI_GRO_CB(skb)->count = num_of_coalesced_segs;
542 }
543 
544 static int bnx2x_alloc_rx_sge(struct bnx2x *bp, struct bnx2x_fastpath *fp,
545 			      u16 index, gfp_t gfp_mask)
546 {
547 	struct sw_rx_page *sw_buf = &fp->rx_page_ring[index];
548 	struct eth_rx_sge *sge = &fp->rx_sge_ring[index];
549 	struct bnx2x_alloc_pool *pool = &fp->page_pool;
550 	dma_addr_t mapping;
551 
552 	if (!pool->page || (PAGE_SIZE - pool->offset) < SGE_PAGE_SIZE) {
553 
554 		/* put page reference used by the memory pool, since we
555 		 * won't be using this page as the mempool anymore.
556 		 */
557 		if (pool->page)
558 			put_page(pool->page);
559 
560 		pool->page = alloc_pages(gfp_mask, PAGES_PER_SGE_SHIFT);
561 		if (unlikely(!pool->page)) {
562 			BNX2X_ERR("Can't alloc sge\n");
563 			return -ENOMEM;
564 		}
565 
566 		pool->offset = 0;
567 	}
568 
569 	mapping = dma_map_page(&bp->pdev->dev, pool->page,
570 			       pool->offset, SGE_PAGE_SIZE, DMA_FROM_DEVICE);
571 	if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
572 		BNX2X_ERR("Can't map sge\n");
573 		return -ENOMEM;
574 	}
575 
576 	get_page(pool->page);
577 	sw_buf->page = pool->page;
578 	sw_buf->offset = pool->offset;
579 
580 	dma_unmap_addr_set(sw_buf, mapping, mapping);
581 
582 	sge->addr_hi = cpu_to_le32(U64_HI(mapping));
583 	sge->addr_lo = cpu_to_le32(U64_LO(mapping));
584 
585 	pool->offset += SGE_PAGE_SIZE;
586 
587 	return 0;
588 }
589 
590 static int bnx2x_fill_frag_skb(struct bnx2x *bp, struct bnx2x_fastpath *fp,
591 			       struct bnx2x_agg_info *tpa_info,
592 			       u16 pages,
593 			       struct sk_buff *skb,
594 			       struct eth_end_agg_rx_cqe *cqe,
595 			       u16 cqe_idx)
596 {
597 	struct sw_rx_page *rx_pg, old_rx_pg;
598 	u32 i, frag_len, frag_size;
599 	int err, j, frag_id = 0;
600 	u16 len_on_bd = tpa_info->len_on_bd;
601 	u16 full_page = 0, gro_size = 0;
602 
603 	frag_size = le16_to_cpu(cqe->pkt_len) - len_on_bd;
604 
605 	if (fp->mode == TPA_MODE_GRO) {
606 		gro_size = tpa_info->gro_size;
607 		full_page = tpa_info->full_page;
608 	}
609 
610 	/* This is needed in order to enable forwarding support */
611 	if (frag_size)
612 		bnx2x_set_gro_params(skb, tpa_info->parsing_flags, len_on_bd,
613 				     le16_to_cpu(cqe->pkt_len),
614 				     le16_to_cpu(cqe->num_of_coalesced_segs));
615 
616 #ifdef BNX2X_STOP_ON_ERROR
617 	if (pages > min_t(u32, 8, MAX_SKB_FRAGS) * SGE_PAGES) {
618 		BNX2X_ERR("SGL length is too long: %d. CQE index is %d\n",
619 			  pages, cqe_idx);
620 		BNX2X_ERR("cqe->pkt_len = %d\n", cqe->pkt_len);
621 		bnx2x_panic();
622 		return -EINVAL;
623 	}
624 #endif
625 
626 	/* Run through the SGL and compose the fragmented skb */
627 	for (i = 0, j = 0; i < pages; i += PAGES_PER_SGE, j++) {
628 		u16 sge_idx = RX_SGE(le16_to_cpu(cqe->sgl_or_raw_data.sgl[j]));
629 
630 		/* FW gives the indices of the SGE as if the ring is an array
631 		   (meaning that "next" element will consume 2 indices) */
632 		if (fp->mode == TPA_MODE_GRO)
633 			frag_len = min_t(u32, frag_size, (u32)full_page);
634 		else /* LRO */
635 			frag_len = min_t(u32, frag_size, (u32)SGE_PAGES);
636 
637 		rx_pg = &fp->rx_page_ring[sge_idx];
638 		old_rx_pg = *rx_pg;
639 
640 		/* If we fail to allocate a substitute page, we simply stop
641 		   where we are and drop the whole packet */
642 		err = bnx2x_alloc_rx_sge(bp, fp, sge_idx, GFP_ATOMIC);
643 		if (unlikely(err)) {
644 			bnx2x_fp_qstats(bp, fp)->rx_skb_alloc_failed++;
645 			return err;
646 		}
647 
648 		dma_unmap_page(&bp->pdev->dev,
649 			       dma_unmap_addr(&old_rx_pg, mapping),
650 			       SGE_PAGE_SIZE, DMA_FROM_DEVICE);
651 		/* Add one frag and update the appropriate fields in the skb */
652 		if (fp->mode == TPA_MODE_LRO)
653 			skb_fill_page_desc(skb, j, old_rx_pg.page,
654 					   old_rx_pg.offset, frag_len);
655 		else { /* GRO */
656 			int rem;
657 			int offset = 0;
658 			for (rem = frag_len; rem > 0; rem -= gro_size) {
659 				int len = rem > gro_size ? gro_size : rem;
660 				skb_fill_page_desc(skb, frag_id++,
661 						   old_rx_pg.page,
662 						   old_rx_pg.offset + offset,
663 						   len);
664 				if (offset)
665 					get_page(old_rx_pg.page);
666 				offset += len;
667 			}
668 		}
669 
670 		skb->data_len += frag_len;
671 		skb->truesize += SGE_PAGES;
672 		skb->len += frag_len;
673 
674 		frag_size -= frag_len;
675 	}
676 
677 	return 0;
678 }
679 
680 static void bnx2x_frag_free(const struct bnx2x_fastpath *fp, void *data)
681 {
682 	if (fp->rx_frag_size)
683 		skb_free_frag(data);
684 	else
685 		kfree(data);
686 }
687 
688 static void *bnx2x_frag_alloc(const struct bnx2x_fastpath *fp, gfp_t gfp_mask)
689 {
690 	if (fp->rx_frag_size) {
691 		/* GFP_KERNEL allocations are used only during initialization */
692 		if (unlikely(gfpflags_allow_blocking(gfp_mask)))
693 			return (void *)__get_free_page(gfp_mask);
694 
695 		return netdev_alloc_frag(fp->rx_frag_size);
696 	}
697 
698 	return kmalloc(fp->rx_buf_size + NET_SKB_PAD, gfp_mask);
699 }
700 
701 #ifdef CONFIG_INET
702 static void bnx2x_gro_ip_csum(struct bnx2x *bp, struct sk_buff *skb)
703 {
704 	const struct iphdr *iph = ip_hdr(skb);
705 	struct tcphdr *th;
706 
707 	skb_set_transport_header(skb, sizeof(struct iphdr));
708 	th = tcp_hdr(skb);
709 
710 	th->check = ~tcp_v4_check(skb->len - skb_transport_offset(skb),
711 				  iph->saddr, iph->daddr, 0);
712 }
713 
714 static void bnx2x_gro_ipv6_csum(struct bnx2x *bp, struct sk_buff *skb)
715 {
716 	struct ipv6hdr *iph = ipv6_hdr(skb);
717 	struct tcphdr *th;
718 
719 	skb_set_transport_header(skb, sizeof(struct ipv6hdr));
720 	th = tcp_hdr(skb);
721 
722 	th->check = ~tcp_v6_check(skb->len - skb_transport_offset(skb),
723 				  &iph->saddr, &iph->daddr, 0);
724 }
725 
726 static void bnx2x_gro_csum(struct bnx2x *bp, struct sk_buff *skb,
727 			    void (*gro_func)(struct bnx2x*, struct sk_buff*))
728 {
729 	skb_set_network_header(skb, 0);
730 	gro_func(bp, skb);
731 	tcp_gro_complete(skb);
732 }
733 #endif
734 
735 static void bnx2x_gro_receive(struct bnx2x *bp, struct bnx2x_fastpath *fp,
736 			       struct sk_buff *skb)
737 {
738 #ifdef CONFIG_INET
739 	if (skb_shinfo(skb)->gso_size) {
740 		switch (be16_to_cpu(skb->protocol)) {
741 		case ETH_P_IP:
742 			bnx2x_gro_csum(bp, skb, bnx2x_gro_ip_csum);
743 			break;
744 		case ETH_P_IPV6:
745 			bnx2x_gro_csum(bp, skb, bnx2x_gro_ipv6_csum);
746 			break;
747 		default:
748 			BNX2X_ERR("Error: FW GRO supports only IPv4/IPv6, not 0x%04x\n",
749 				  be16_to_cpu(skb->protocol));
750 		}
751 	}
752 #endif
753 	skb_record_rx_queue(skb, fp->rx_queue);
754 	napi_gro_receive(&fp->napi, skb);
755 }
756 
757 static void bnx2x_tpa_stop(struct bnx2x *bp, struct bnx2x_fastpath *fp,
758 			   struct bnx2x_agg_info *tpa_info,
759 			   u16 pages,
760 			   struct eth_end_agg_rx_cqe *cqe,
761 			   u16 cqe_idx)
762 {
763 	struct sw_rx_bd *rx_buf = &tpa_info->first_buf;
764 	u8 pad = tpa_info->placement_offset;
765 	u16 len = tpa_info->len_on_bd;
766 	struct sk_buff *skb = NULL;
767 	u8 *new_data, *data = rx_buf->data;
768 	u8 old_tpa_state = tpa_info->tpa_state;
769 
770 	tpa_info->tpa_state = BNX2X_TPA_STOP;
771 
772 	/* If we there was an error during the handling of the TPA_START -
773 	 * drop this aggregation.
774 	 */
775 	if (old_tpa_state == BNX2X_TPA_ERROR)
776 		goto drop;
777 
778 	/* Try to allocate the new data */
779 	new_data = bnx2x_frag_alloc(fp, GFP_ATOMIC);
780 	/* Unmap skb in the pool anyway, as we are going to change
781 	   pool entry status to BNX2X_TPA_STOP even if new skb allocation
782 	   fails. */
783 	dma_unmap_single(&bp->pdev->dev, dma_unmap_addr(rx_buf, mapping),
784 			 fp->rx_buf_size, DMA_FROM_DEVICE);
785 	if (likely(new_data))
786 		skb = build_skb(data, fp->rx_frag_size);
787 
788 	if (likely(skb)) {
789 #ifdef BNX2X_STOP_ON_ERROR
790 		if (pad + len > fp->rx_buf_size) {
791 			BNX2X_ERR("skb_put is about to fail...  pad %d  len %d  rx_buf_size %d\n",
792 				  pad, len, fp->rx_buf_size);
793 			bnx2x_panic();
794 			return;
795 		}
796 #endif
797 
798 		skb_reserve(skb, pad + NET_SKB_PAD);
799 		skb_put(skb, len);
800 		skb_set_hash(skb, tpa_info->rxhash, tpa_info->rxhash_type);
801 
802 		skb->protocol = eth_type_trans(skb, bp->dev);
803 		skb->ip_summed = CHECKSUM_UNNECESSARY;
804 
805 		if (!bnx2x_fill_frag_skb(bp, fp, tpa_info, pages,
806 					 skb, cqe, cqe_idx)) {
807 			if (tpa_info->parsing_flags & PARSING_FLAGS_VLAN)
808 				__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), tpa_info->vlan_tag);
809 			bnx2x_gro_receive(bp, fp, skb);
810 		} else {
811 			DP(NETIF_MSG_RX_STATUS,
812 			   "Failed to allocate new pages - dropping packet!\n");
813 			dev_kfree_skb_any(skb);
814 		}
815 
816 		/* put new data in bin */
817 		rx_buf->data = new_data;
818 
819 		return;
820 	}
821 	if (new_data)
822 		bnx2x_frag_free(fp, new_data);
823 drop:
824 	/* drop the packet and keep the buffer in the bin */
825 	DP(NETIF_MSG_RX_STATUS,
826 	   "Failed to allocate or map a new skb - dropping packet!\n");
827 	bnx2x_fp_stats(bp, fp)->eth_q_stats.rx_skb_alloc_failed++;
828 }
829 
830 static int bnx2x_alloc_rx_data(struct bnx2x *bp, struct bnx2x_fastpath *fp,
831 			       u16 index, gfp_t gfp_mask)
832 {
833 	u8 *data;
834 	struct sw_rx_bd *rx_buf = &fp->rx_buf_ring[index];
835 	struct eth_rx_bd *rx_bd = &fp->rx_desc_ring[index];
836 	dma_addr_t mapping;
837 
838 	data = bnx2x_frag_alloc(fp, gfp_mask);
839 	if (unlikely(data == NULL))
840 		return -ENOMEM;
841 
842 	mapping = dma_map_single(&bp->pdev->dev, data + NET_SKB_PAD,
843 				 fp->rx_buf_size,
844 				 DMA_FROM_DEVICE);
845 	if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
846 		bnx2x_frag_free(fp, data);
847 		BNX2X_ERR("Can't map rx data\n");
848 		return -ENOMEM;
849 	}
850 
851 	rx_buf->data = data;
852 	dma_unmap_addr_set(rx_buf, mapping, mapping);
853 
854 	rx_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
855 	rx_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
856 
857 	return 0;
858 }
859 
860 static
861 void bnx2x_csum_validate(struct sk_buff *skb, union eth_rx_cqe *cqe,
862 				 struct bnx2x_fastpath *fp,
863 				 struct bnx2x_eth_q_stats *qstats)
864 {
865 	/* Do nothing if no L4 csum validation was done.
866 	 * We do not check whether IP csum was validated. For IPv4 we assume
867 	 * that if the card got as far as validating the L4 csum, it also
868 	 * validated the IP csum. IPv6 has no IP csum.
869 	 */
870 	if (cqe->fast_path_cqe.status_flags &
871 	    ETH_FAST_PATH_RX_CQE_L4_XSUM_NO_VALIDATION_FLG)
872 		return;
873 
874 	/* If L4 validation was done, check if an error was found. */
875 
876 	if (cqe->fast_path_cqe.type_error_flags &
877 	    (ETH_FAST_PATH_RX_CQE_IP_BAD_XSUM_FLG |
878 	     ETH_FAST_PATH_RX_CQE_L4_BAD_XSUM_FLG))
879 		qstats->hw_csum_err++;
880 	else
881 		skb->ip_summed = CHECKSUM_UNNECESSARY;
882 }
883 
884 static int bnx2x_rx_int(struct bnx2x_fastpath *fp, int budget)
885 {
886 	struct bnx2x *bp = fp->bp;
887 	u16 bd_cons, bd_prod, bd_prod_fw, comp_ring_cons;
888 	u16 sw_comp_cons, sw_comp_prod;
889 	int rx_pkt = 0;
890 	union eth_rx_cqe *cqe;
891 	struct eth_fast_path_rx_cqe *cqe_fp;
892 
893 #ifdef BNX2X_STOP_ON_ERROR
894 	if (unlikely(bp->panic))
895 		return 0;
896 #endif
897 	if (budget <= 0)
898 		return rx_pkt;
899 
900 	bd_cons = fp->rx_bd_cons;
901 	bd_prod = fp->rx_bd_prod;
902 	bd_prod_fw = bd_prod;
903 	sw_comp_cons = fp->rx_comp_cons;
904 	sw_comp_prod = fp->rx_comp_prod;
905 
906 	comp_ring_cons = RCQ_BD(sw_comp_cons);
907 	cqe = &fp->rx_comp_ring[comp_ring_cons];
908 	cqe_fp = &cqe->fast_path_cqe;
909 
910 	DP(NETIF_MSG_RX_STATUS,
911 	   "queue[%d]: sw_comp_cons %u\n", fp->index, sw_comp_cons);
912 
913 	while (BNX2X_IS_CQE_COMPLETED(cqe_fp)) {
914 		struct sw_rx_bd *rx_buf = NULL;
915 		struct sk_buff *skb;
916 		u8 cqe_fp_flags;
917 		enum eth_rx_cqe_type cqe_fp_type;
918 		u16 len, pad, queue;
919 		u8 *data;
920 		u32 rxhash;
921 		enum pkt_hash_types rxhash_type;
922 
923 #ifdef BNX2X_STOP_ON_ERROR
924 		if (unlikely(bp->panic))
925 			return 0;
926 #endif
927 
928 		bd_prod = RX_BD(bd_prod);
929 		bd_cons = RX_BD(bd_cons);
930 
931 		/* A rmb() is required to ensure that the CQE is not read
932 		 * before it is written by the adapter DMA.  PCI ordering
933 		 * rules will make sure the other fields are written before
934 		 * the marker at the end of struct eth_fast_path_rx_cqe
935 		 * but without rmb() a weakly ordered processor can process
936 		 * stale data.  Without the barrier TPA state-machine might
937 		 * enter inconsistent state and kernel stack might be
938 		 * provided with incorrect packet description - these lead
939 		 * to various kernel crashed.
940 		 */
941 		rmb();
942 
943 		cqe_fp_flags = cqe_fp->type_error_flags;
944 		cqe_fp_type = cqe_fp_flags & ETH_FAST_PATH_RX_CQE_TYPE;
945 
946 		DP(NETIF_MSG_RX_STATUS,
947 		   "CQE type %x  err %x  status %x  queue %x  vlan %x  len %u\n",
948 		   CQE_TYPE(cqe_fp_flags),
949 		   cqe_fp_flags, cqe_fp->status_flags,
950 		   le32_to_cpu(cqe_fp->rss_hash_result),
951 		   le16_to_cpu(cqe_fp->vlan_tag),
952 		   le16_to_cpu(cqe_fp->pkt_len_or_gro_seg_len));
953 
954 		/* is this a slowpath msg? */
955 		if (unlikely(CQE_TYPE_SLOW(cqe_fp_type))) {
956 			bnx2x_sp_event(fp, cqe);
957 			goto next_cqe;
958 		}
959 
960 		rx_buf = &fp->rx_buf_ring[bd_cons];
961 		data = rx_buf->data;
962 
963 		if (!CQE_TYPE_FAST(cqe_fp_type)) {
964 			struct bnx2x_agg_info *tpa_info;
965 			u16 frag_size, pages;
966 #ifdef BNX2X_STOP_ON_ERROR
967 			/* sanity check */
968 			if (fp->mode == TPA_MODE_DISABLED &&
969 			    (CQE_TYPE_START(cqe_fp_type) ||
970 			     CQE_TYPE_STOP(cqe_fp_type)))
971 				BNX2X_ERR("START/STOP packet while TPA disabled, type %x\n",
972 					  CQE_TYPE(cqe_fp_type));
973 #endif
974 
975 			if (CQE_TYPE_START(cqe_fp_type)) {
976 				u16 queue = cqe_fp->queue_index;
977 				DP(NETIF_MSG_RX_STATUS,
978 				   "calling tpa_start on queue %d\n",
979 				   queue);
980 
981 				bnx2x_tpa_start(fp, queue,
982 						bd_cons, bd_prod,
983 						cqe_fp);
984 
985 				goto next_rx;
986 			}
987 			queue = cqe->end_agg_cqe.queue_index;
988 			tpa_info = &fp->tpa_info[queue];
989 			DP(NETIF_MSG_RX_STATUS,
990 			   "calling tpa_stop on queue %d\n",
991 			   queue);
992 
993 			frag_size = le16_to_cpu(cqe->end_agg_cqe.pkt_len) -
994 				    tpa_info->len_on_bd;
995 
996 			if (fp->mode == TPA_MODE_GRO)
997 				pages = (frag_size + tpa_info->full_page - 1) /
998 					 tpa_info->full_page;
999 			else
1000 				pages = SGE_PAGE_ALIGN(frag_size) >>
1001 					SGE_PAGE_SHIFT;
1002 
1003 			bnx2x_tpa_stop(bp, fp, tpa_info, pages,
1004 				       &cqe->end_agg_cqe, comp_ring_cons);
1005 #ifdef BNX2X_STOP_ON_ERROR
1006 			if (bp->panic)
1007 				return 0;
1008 #endif
1009 
1010 			bnx2x_update_sge_prod(fp, pages, &cqe->end_agg_cqe);
1011 			goto next_cqe;
1012 		}
1013 		/* non TPA */
1014 		len = le16_to_cpu(cqe_fp->pkt_len_or_gro_seg_len);
1015 		pad = cqe_fp->placement_offset;
1016 		dma_sync_single_for_cpu(&bp->pdev->dev,
1017 					dma_unmap_addr(rx_buf, mapping),
1018 					pad + RX_COPY_THRESH,
1019 					DMA_FROM_DEVICE);
1020 		pad += NET_SKB_PAD;
1021 		prefetch(data + pad); /* speedup eth_type_trans() */
1022 		/* is this an error packet? */
1023 		if (unlikely(cqe_fp_flags & ETH_RX_ERROR_FALGS)) {
1024 			DP(NETIF_MSG_RX_ERR | NETIF_MSG_RX_STATUS,
1025 			   "ERROR  flags %x  rx packet %u\n",
1026 			   cqe_fp_flags, sw_comp_cons);
1027 			bnx2x_fp_qstats(bp, fp)->rx_err_discard_pkt++;
1028 			goto reuse_rx;
1029 		}
1030 
1031 		/* Since we don't have a jumbo ring
1032 		 * copy small packets if mtu > 1500
1033 		 */
1034 		if ((bp->dev->mtu > ETH_MAX_PACKET_SIZE) &&
1035 		    (len <= RX_COPY_THRESH)) {
1036 			skb = napi_alloc_skb(&fp->napi, len);
1037 			if (skb == NULL) {
1038 				DP(NETIF_MSG_RX_ERR | NETIF_MSG_RX_STATUS,
1039 				   "ERROR  packet dropped because of alloc failure\n");
1040 				bnx2x_fp_qstats(bp, fp)->rx_skb_alloc_failed++;
1041 				goto reuse_rx;
1042 			}
1043 			memcpy(skb->data, data + pad, len);
1044 			bnx2x_reuse_rx_data(fp, bd_cons, bd_prod);
1045 		} else {
1046 			if (likely(bnx2x_alloc_rx_data(bp, fp, bd_prod,
1047 						       GFP_ATOMIC) == 0)) {
1048 				dma_unmap_single(&bp->pdev->dev,
1049 						 dma_unmap_addr(rx_buf, mapping),
1050 						 fp->rx_buf_size,
1051 						 DMA_FROM_DEVICE);
1052 				skb = build_skb(data, fp->rx_frag_size);
1053 				if (unlikely(!skb)) {
1054 					bnx2x_frag_free(fp, data);
1055 					bnx2x_fp_qstats(bp, fp)->
1056 							rx_skb_alloc_failed++;
1057 					goto next_rx;
1058 				}
1059 				skb_reserve(skb, pad);
1060 			} else {
1061 				DP(NETIF_MSG_RX_ERR | NETIF_MSG_RX_STATUS,
1062 				   "ERROR  packet dropped because of alloc failure\n");
1063 				bnx2x_fp_qstats(bp, fp)->rx_skb_alloc_failed++;
1064 reuse_rx:
1065 				bnx2x_reuse_rx_data(fp, bd_cons, bd_prod);
1066 				goto next_rx;
1067 			}
1068 		}
1069 
1070 		skb_put(skb, len);
1071 		skb->protocol = eth_type_trans(skb, bp->dev);
1072 
1073 		/* Set Toeplitz hash for a none-LRO skb */
1074 		rxhash = bnx2x_get_rxhash(bp, cqe_fp, &rxhash_type);
1075 		skb_set_hash(skb, rxhash, rxhash_type);
1076 
1077 		skb_checksum_none_assert(skb);
1078 
1079 		if (bp->dev->features & NETIF_F_RXCSUM)
1080 			bnx2x_csum_validate(skb, cqe, fp,
1081 					    bnx2x_fp_qstats(bp, fp));
1082 
1083 		skb_record_rx_queue(skb, fp->rx_queue);
1084 
1085 		/* Check if this packet was timestamped */
1086 		if (unlikely(cqe->fast_path_cqe.type_error_flags &
1087 			     (1 << ETH_FAST_PATH_RX_CQE_PTP_PKT_SHIFT)))
1088 			bnx2x_set_rx_ts(bp, skb);
1089 
1090 		if (le16_to_cpu(cqe_fp->pars_flags.flags) &
1091 		    PARSING_FLAGS_VLAN)
1092 			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
1093 					       le16_to_cpu(cqe_fp->vlan_tag));
1094 
1095 		napi_gro_receive(&fp->napi, skb);
1096 next_rx:
1097 		rx_buf->data = NULL;
1098 
1099 		bd_cons = NEXT_RX_IDX(bd_cons);
1100 		bd_prod = NEXT_RX_IDX(bd_prod);
1101 		bd_prod_fw = NEXT_RX_IDX(bd_prod_fw);
1102 		rx_pkt++;
1103 next_cqe:
1104 		sw_comp_prod = NEXT_RCQ_IDX(sw_comp_prod);
1105 		sw_comp_cons = NEXT_RCQ_IDX(sw_comp_cons);
1106 
1107 		/* mark CQE as free */
1108 		BNX2X_SEED_CQE(cqe_fp);
1109 
1110 		if (rx_pkt == budget)
1111 			break;
1112 
1113 		comp_ring_cons = RCQ_BD(sw_comp_cons);
1114 		cqe = &fp->rx_comp_ring[comp_ring_cons];
1115 		cqe_fp = &cqe->fast_path_cqe;
1116 	} /* while */
1117 
1118 	fp->rx_bd_cons = bd_cons;
1119 	fp->rx_bd_prod = bd_prod_fw;
1120 	fp->rx_comp_cons = sw_comp_cons;
1121 	fp->rx_comp_prod = sw_comp_prod;
1122 
1123 	/* Update producers */
1124 	bnx2x_update_rx_prod(bp, fp, bd_prod_fw, sw_comp_prod,
1125 			     fp->rx_sge_prod);
1126 
1127 	fp->rx_pkt += rx_pkt;
1128 	fp->rx_calls++;
1129 
1130 	return rx_pkt;
1131 }
1132 
1133 static irqreturn_t bnx2x_msix_fp_int(int irq, void *fp_cookie)
1134 {
1135 	struct bnx2x_fastpath *fp = fp_cookie;
1136 	struct bnx2x *bp = fp->bp;
1137 	u8 cos;
1138 
1139 	DP(NETIF_MSG_INTR,
1140 	   "got an MSI-X interrupt on IDX:SB [fp %d fw_sd %d igusb %d]\n",
1141 	   fp->index, fp->fw_sb_id, fp->igu_sb_id);
1142 
1143 	bnx2x_ack_sb(bp, fp->igu_sb_id, USTORM_ID, 0, IGU_INT_DISABLE, 0);
1144 
1145 #ifdef BNX2X_STOP_ON_ERROR
1146 	if (unlikely(bp->panic))
1147 		return IRQ_HANDLED;
1148 #endif
1149 
1150 	/* Handle Rx and Tx according to MSI-X vector */
1151 	for_each_cos_in_tx_queue(fp, cos)
1152 		prefetch(fp->txdata_ptr[cos]->tx_cons_sb);
1153 
1154 	prefetch(&fp->sb_running_index[SM_RX_ID]);
1155 	napi_schedule_irqoff(&bnx2x_fp(bp, fp->index, napi));
1156 
1157 	return IRQ_HANDLED;
1158 }
1159 
1160 /* HW Lock for shared dual port PHYs */
1161 void bnx2x_acquire_phy_lock(struct bnx2x *bp)
1162 {
1163 	mutex_lock(&bp->port.phy_mutex);
1164 
1165 	bnx2x_acquire_hw_lock(bp, HW_LOCK_RESOURCE_MDIO);
1166 }
1167 
1168 void bnx2x_release_phy_lock(struct bnx2x *bp)
1169 {
1170 	bnx2x_release_hw_lock(bp, HW_LOCK_RESOURCE_MDIO);
1171 
1172 	mutex_unlock(&bp->port.phy_mutex);
1173 }
1174 
1175 /* calculates MF speed according to current linespeed and MF configuration */
1176 u16 bnx2x_get_mf_speed(struct bnx2x *bp)
1177 {
1178 	u16 line_speed = bp->link_vars.line_speed;
1179 	if (IS_MF(bp)) {
1180 		u16 maxCfg = bnx2x_extract_max_cfg(bp,
1181 						   bp->mf_config[BP_VN(bp)]);
1182 
1183 		/* Calculate the current MAX line speed limit for the MF
1184 		 * devices
1185 		 */
1186 		if (IS_MF_PERCENT_BW(bp))
1187 			line_speed = (line_speed * maxCfg) / 100;
1188 		else { /* SD mode */
1189 			u16 vn_max_rate = maxCfg * 100;
1190 
1191 			if (vn_max_rate < line_speed)
1192 				line_speed = vn_max_rate;
1193 		}
1194 	}
1195 
1196 	return line_speed;
1197 }
1198 
1199 /**
1200  * bnx2x_fill_report_data - fill link report data to report
1201  *
1202  * @bp:		driver handle
1203  * @data:	link state to update
1204  *
1205  * It uses a none-atomic bit operations because is called under the mutex.
1206  */
1207 static void bnx2x_fill_report_data(struct bnx2x *bp,
1208 				   struct bnx2x_link_report_data *data)
1209 {
1210 	memset(data, 0, sizeof(*data));
1211 
1212 	if (IS_PF(bp)) {
1213 		/* Fill the report data: effective line speed */
1214 		data->line_speed = bnx2x_get_mf_speed(bp);
1215 
1216 		/* Link is down */
1217 		if (!bp->link_vars.link_up || (bp->flags & MF_FUNC_DIS))
1218 			__set_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1219 				  &data->link_report_flags);
1220 
1221 		if (!BNX2X_NUM_ETH_QUEUES(bp))
1222 			__set_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1223 				  &data->link_report_flags);
1224 
1225 		/* Full DUPLEX */
1226 		if (bp->link_vars.duplex == DUPLEX_FULL)
1227 			__set_bit(BNX2X_LINK_REPORT_FD,
1228 				  &data->link_report_flags);
1229 
1230 		/* Rx Flow Control is ON */
1231 		if (bp->link_vars.flow_ctrl & BNX2X_FLOW_CTRL_RX)
1232 			__set_bit(BNX2X_LINK_REPORT_RX_FC_ON,
1233 				  &data->link_report_flags);
1234 
1235 		/* Tx Flow Control is ON */
1236 		if (bp->link_vars.flow_ctrl & BNX2X_FLOW_CTRL_TX)
1237 			__set_bit(BNX2X_LINK_REPORT_TX_FC_ON,
1238 				  &data->link_report_flags);
1239 	} else { /* VF */
1240 		*data = bp->vf_link_vars;
1241 	}
1242 }
1243 
1244 /**
1245  * bnx2x_link_report - report link status to OS.
1246  *
1247  * @bp:		driver handle
1248  *
1249  * Calls the __bnx2x_link_report() under the same locking scheme
1250  * as a link/PHY state managing code to ensure a consistent link
1251  * reporting.
1252  */
1253 
1254 void bnx2x_link_report(struct bnx2x *bp)
1255 {
1256 	bnx2x_acquire_phy_lock(bp);
1257 	__bnx2x_link_report(bp);
1258 	bnx2x_release_phy_lock(bp);
1259 }
1260 
1261 /**
1262  * __bnx2x_link_report - report link status to OS.
1263  *
1264  * @bp:		driver handle
1265  *
1266  * None atomic implementation.
1267  * Should be called under the phy_lock.
1268  */
1269 void __bnx2x_link_report(struct bnx2x *bp)
1270 {
1271 	struct bnx2x_link_report_data cur_data;
1272 
1273 	/* reread mf_cfg */
1274 	if (IS_PF(bp) && !CHIP_IS_E1(bp))
1275 		bnx2x_read_mf_cfg(bp);
1276 
1277 	/* Read the current link report info */
1278 	bnx2x_fill_report_data(bp, &cur_data);
1279 
1280 	/* Don't report link down or exactly the same link status twice */
1281 	if (!memcmp(&cur_data, &bp->last_reported_link, sizeof(cur_data)) ||
1282 	    (test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1283 		      &bp->last_reported_link.link_report_flags) &&
1284 	     test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1285 		      &cur_data.link_report_flags)))
1286 		return;
1287 
1288 	bp->link_cnt++;
1289 
1290 	/* We are going to report a new link parameters now -
1291 	 * remember the current data for the next time.
1292 	 */
1293 	memcpy(&bp->last_reported_link, &cur_data, sizeof(cur_data));
1294 
1295 	/* propagate status to VFs */
1296 	if (IS_PF(bp))
1297 		bnx2x_iov_link_update(bp);
1298 
1299 	if (test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1300 		     &cur_data.link_report_flags)) {
1301 		netif_carrier_off(bp->dev);
1302 		netdev_err(bp->dev, "NIC Link is Down\n");
1303 		return;
1304 	} else {
1305 		const char *duplex;
1306 		const char *flow;
1307 
1308 		netif_carrier_on(bp->dev);
1309 
1310 		if (test_and_clear_bit(BNX2X_LINK_REPORT_FD,
1311 				       &cur_data.link_report_flags))
1312 			duplex = "full";
1313 		else
1314 			duplex = "half";
1315 
1316 		/* Handle the FC at the end so that only these flags would be
1317 		 * possibly set. This way we may easily check if there is no FC
1318 		 * enabled.
1319 		 */
1320 		if (cur_data.link_report_flags) {
1321 			if (test_bit(BNX2X_LINK_REPORT_RX_FC_ON,
1322 				     &cur_data.link_report_flags)) {
1323 				if (test_bit(BNX2X_LINK_REPORT_TX_FC_ON,
1324 				     &cur_data.link_report_flags))
1325 					flow = "ON - receive & transmit";
1326 				else
1327 					flow = "ON - receive";
1328 			} else {
1329 				flow = "ON - transmit";
1330 			}
1331 		} else {
1332 			flow = "none";
1333 		}
1334 		netdev_info(bp->dev, "NIC Link is Up, %d Mbps %s duplex, Flow control: %s\n",
1335 			    cur_data.line_speed, duplex, flow);
1336 	}
1337 }
1338 
1339 static void bnx2x_set_next_page_sgl(struct bnx2x_fastpath *fp)
1340 {
1341 	int i;
1342 
1343 	for (i = 1; i <= NUM_RX_SGE_PAGES; i++) {
1344 		struct eth_rx_sge *sge;
1345 
1346 		sge = &fp->rx_sge_ring[RX_SGE_CNT * i - 2];
1347 		sge->addr_hi =
1348 			cpu_to_le32(U64_HI(fp->rx_sge_mapping +
1349 			BCM_PAGE_SIZE*(i % NUM_RX_SGE_PAGES)));
1350 
1351 		sge->addr_lo =
1352 			cpu_to_le32(U64_LO(fp->rx_sge_mapping +
1353 			BCM_PAGE_SIZE*(i % NUM_RX_SGE_PAGES)));
1354 	}
1355 }
1356 
1357 static void bnx2x_free_tpa_pool(struct bnx2x *bp,
1358 				struct bnx2x_fastpath *fp, int last)
1359 {
1360 	int i;
1361 
1362 	for (i = 0; i < last; i++) {
1363 		struct bnx2x_agg_info *tpa_info = &fp->tpa_info[i];
1364 		struct sw_rx_bd *first_buf = &tpa_info->first_buf;
1365 		u8 *data = first_buf->data;
1366 
1367 		if (data == NULL) {
1368 			DP(NETIF_MSG_IFDOWN, "tpa bin %d empty on free\n", i);
1369 			continue;
1370 		}
1371 		if (tpa_info->tpa_state == BNX2X_TPA_START)
1372 			dma_unmap_single(&bp->pdev->dev,
1373 					 dma_unmap_addr(first_buf, mapping),
1374 					 fp->rx_buf_size, DMA_FROM_DEVICE);
1375 		bnx2x_frag_free(fp, data);
1376 		first_buf->data = NULL;
1377 	}
1378 }
1379 
1380 void bnx2x_init_rx_rings_cnic(struct bnx2x *bp)
1381 {
1382 	int j;
1383 
1384 	for_each_rx_queue_cnic(bp, j) {
1385 		struct bnx2x_fastpath *fp = &bp->fp[j];
1386 
1387 		fp->rx_bd_cons = 0;
1388 
1389 		/* Activate BD ring */
1390 		/* Warning!
1391 		 * this will generate an interrupt (to the TSTORM)
1392 		 * must only be done after chip is initialized
1393 		 */
1394 		bnx2x_update_rx_prod(bp, fp, fp->rx_bd_prod, fp->rx_comp_prod,
1395 				     fp->rx_sge_prod);
1396 	}
1397 }
1398 
1399 void bnx2x_init_rx_rings(struct bnx2x *bp)
1400 {
1401 	int func = BP_FUNC(bp);
1402 	u16 ring_prod;
1403 	int i, j;
1404 
1405 	/* Allocate TPA resources */
1406 	for_each_eth_queue(bp, j) {
1407 		struct bnx2x_fastpath *fp = &bp->fp[j];
1408 
1409 		DP(NETIF_MSG_IFUP,
1410 		   "mtu %d  rx_buf_size %d\n", bp->dev->mtu, fp->rx_buf_size);
1411 
1412 		if (fp->mode != TPA_MODE_DISABLED) {
1413 			/* Fill the per-aggregation pool */
1414 			for (i = 0; i < MAX_AGG_QS(bp); i++) {
1415 				struct bnx2x_agg_info *tpa_info =
1416 					&fp->tpa_info[i];
1417 				struct sw_rx_bd *first_buf =
1418 					&tpa_info->first_buf;
1419 
1420 				first_buf->data =
1421 					bnx2x_frag_alloc(fp, GFP_KERNEL);
1422 				if (!first_buf->data) {
1423 					BNX2X_ERR("Failed to allocate TPA skb pool for queue[%d] - disabling TPA on this queue!\n",
1424 						  j);
1425 					bnx2x_free_tpa_pool(bp, fp, i);
1426 					fp->mode = TPA_MODE_DISABLED;
1427 					break;
1428 				}
1429 				dma_unmap_addr_set(first_buf, mapping, 0);
1430 				tpa_info->tpa_state = BNX2X_TPA_STOP;
1431 			}
1432 
1433 			/* "next page" elements initialization */
1434 			bnx2x_set_next_page_sgl(fp);
1435 
1436 			/* set SGEs bit mask */
1437 			bnx2x_init_sge_ring_bit_mask(fp);
1438 
1439 			/* Allocate SGEs and initialize the ring elements */
1440 			for (i = 0, ring_prod = 0;
1441 			     i < MAX_RX_SGE_CNT*NUM_RX_SGE_PAGES; i++) {
1442 
1443 				if (bnx2x_alloc_rx_sge(bp, fp, ring_prod,
1444 						       GFP_KERNEL) < 0) {
1445 					BNX2X_ERR("was only able to allocate %d rx sges\n",
1446 						  i);
1447 					BNX2X_ERR("disabling TPA for queue[%d]\n",
1448 						  j);
1449 					/* Cleanup already allocated elements */
1450 					bnx2x_free_rx_sge_range(bp, fp,
1451 								ring_prod);
1452 					bnx2x_free_tpa_pool(bp, fp,
1453 							    MAX_AGG_QS(bp));
1454 					fp->mode = TPA_MODE_DISABLED;
1455 					ring_prod = 0;
1456 					break;
1457 				}
1458 				ring_prod = NEXT_SGE_IDX(ring_prod);
1459 			}
1460 
1461 			fp->rx_sge_prod = ring_prod;
1462 		}
1463 	}
1464 
1465 	for_each_eth_queue(bp, j) {
1466 		struct bnx2x_fastpath *fp = &bp->fp[j];
1467 
1468 		fp->rx_bd_cons = 0;
1469 
1470 		/* Activate BD ring */
1471 		/* Warning!
1472 		 * this will generate an interrupt (to the TSTORM)
1473 		 * must only be done after chip is initialized
1474 		 */
1475 		bnx2x_update_rx_prod(bp, fp, fp->rx_bd_prod, fp->rx_comp_prod,
1476 				     fp->rx_sge_prod);
1477 
1478 		if (j != 0)
1479 			continue;
1480 
1481 		if (CHIP_IS_E1(bp)) {
1482 			REG_WR(bp, BAR_USTRORM_INTMEM +
1483 			       USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func),
1484 			       U64_LO(fp->rx_comp_mapping));
1485 			REG_WR(bp, BAR_USTRORM_INTMEM +
1486 			       USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func) + 4,
1487 			       U64_HI(fp->rx_comp_mapping));
1488 		}
1489 	}
1490 }
1491 
1492 static void bnx2x_free_tx_skbs_queue(struct bnx2x_fastpath *fp)
1493 {
1494 	u8 cos;
1495 	struct bnx2x *bp = fp->bp;
1496 
1497 	for_each_cos_in_tx_queue(fp, cos) {
1498 		struct bnx2x_fp_txdata *txdata = fp->txdata_ptr[cos];
1499 		unsigned pkts_compl = 0, bytes_compl = 0;
1500 
1501 		u16 sw_prod = txdata->tx_pkt_prod;
1502 		u16 sw_cons = txdata->tx_pkt_cons;
1503 
1504 		while (sw_cons != sw_prod) {
1505 			bnx2x_free_tx_pkt(bp, txdata, TX_BD(sw_cons),
1506 					  &pkts_compl, &bytes_compl);
1507 			sw_cons++;
1508 		}
1509 
1510 		netdev_tx_reset_queue(
1511 			netdev_get_tx_queue(bp->dev,
1512 					    txdata->txq_index));
1513 	}
1514 }
1515 
1516 static void bnx2x_free_tx_skbs_cnic(struct bnx2x *bp)
1517 {
1518 	int i;
1519 
1520 	for_each_tx_queue_cnic(bp, i) {
1521 		bnx2x_free_tx_skbs_queue(&bp->fp[i]);
1522 	}
1523 }
1524 
1525 static void bnx2x_free_tx_skbs(struct bnx2x *bp)
1526 {
1527 	int i;
1528 
1529 	for_each_eth_queue(bp, i) {
1530 		bnx2x_free_tx_skbs_queue(&bp->fp[i]);
1531 	}
1532 }
1533 
1534 static void bnx2x_free_rx_bds(struct bnx2x_fastpath *fp)
1535 {
1536 	struct bnx2x *bp = fp->bp;
1537 	int i;
1538 
1539 	/* ring wasn't allocated */
1540 	if (fp->rx_buf_ring == NULL)
1541 		return;
1542 
1543 	for (i = 0; i < NUM_RX_BD; i++) {
1544 		struct sw_rx_bd *rx_buf = &fp->rx_buf_ring[i];
1545 		u8 *data = rx_buf->data;
1546 
1547 		if (data == NULL)
1548 			continue;
1549 		dma_unmap_single(&bp->pdev->dev,
1550 				 dma_unmap_addr(rx_buf, mapping),
1551 				 fp->rx_buf_size, DMA_FROM_DEVICE);
1552 
1553 		rx_buf->data = NULL;
1554 		bnx2x_frag_free(fp, data);
1555 	}
1556 }
1557 
1558 static void bnx2x_free_rx_skbs_cnic(struct bnx2x *bp)
1559 {
1560 	int j;
1561 
1562 	for_each_rx_queue_cnic(bp, j) {
1563 		bnx2x_free_rx_bds(&bp->fp[j]);
1564 	}
1565 }
1566 
1567 static void bnx2x_free_rx_skbs(struct bnx2x *bp)
1568 {
1569 	int j;
1570 
1571 	for_each_eth_queue(bp, j) {
1572 		struct bnx2x_fastpath *fp = &bp->fp[j];
1573 
1574 		bnx2x_free_rx_bds(fp);
1575 
1576 		if (fp->mode != TPA_MODE_DISABLED)
1577 			bnx2x_free_tpa_pool(bp, fp, MAX_AGG_QS(bp));
1578 	}
1579 }
1580 
1581 static void bnx2x_free_skbs_cnic(struct bnx2x *bp)
1582 {
1583 	bnx2x_free_tx_skbs_cnic(bp);
1584 	bnx2x_free_rx_skbs_cnic(bp);
1585 }
1586 
1587 void bnx2x_free_skbs(struct bnx2x *bp)
1588 {
1589 	bnx2x_free_tx_skbs(bp);
1590 	bnx2x_free_rx_skbs(bp);
1591 }
1592 
1593 void bnx2x_update_max_mf_config(struct bnx2x *bp, u32 value)
1594 {
1595 	/* load old values */
1596 	u32 mf_cfg = bp->mf_config[BP_VN(bp)];
1597 
1598 	if (value != bnx2x_extract_max_cfg(bp, mf_cfg)) {
1599 		/* leave all but MAX value */
1600 		mf_cfg &= ~FUNC_MF_CFG_MAX_BW_MASK;
1601 
1602 		/* set new MAX value */
1603 		mf_cfg |= (value << FUNC_MF_CFG_MAX_BW_SHIFT)
1604 				& FUNC_MF_CFG_MAX_BW_MASK;
1605 
1606 		bnx2x_fw_command(bp, DRV_MSG_CODE_SET_MF_BW, mf_cfg);
1607 	}
1608 }
1609 
1610 /**
1611  * bnx2x_free_msix_irqs - free previously requested MSI-X IRQ vectors
1612  *
1613  * @bp:		driver handle
1614  * @nvecs:	number of vectors to be released
1615  */
1616 static void bnx2x_free_msix_irqs(struct bnx2x *bp, int nvecs)
1617 {
1618 	int i, offset = 0;
1619 
1620 	if (nvecs == offset)
1621 		return;
1622 
1623 	/* VFs don't have a default SB */
1624 	if (IS_PF(bp)) {
1625 		free_irq(bp->msix_table[offset].vector, bp->dev);
1626 		DP(NETIF_MSG_IFDOWN, "released sp irq (%d)\n",
1627 		   bp->msix_table[offset].vector);
1628 		offset++;
1629 	}
1630 
1631 	if (CNIC_SUPPORT(bp)) {
1632 		if (nvecs == offset)
1633 			return;
1634 		offset++;
1635 	}
1636 
1637 	for_each_eth_queue(bp, i) {
1638 		if (nvecs == offset)
1639 			return;
1640 		DP(NETIF_MSG_IFDOWN, "about to release fp #%d->%d irq\n",
1641 		   i, bp->msix_table[offset].vector);
1642 
1643 		free_irq(bp->msix_table[offset++].vector, &bp->fp[i]);
1644 	}
1645 }
1646 
1647 void bnx2x_free_irq(struct bnx2x *bp)
1648 {
1649 	if (bp->flags & USING_MSIX_FLAG &&
1650 	    !(bp->flags & USING_SINGLE_MSIX_FLAG)) {
1651 		int nvecs = BNX2X_NUM_ETH_QUEUES(bp) + CNIC_SUPPORT(bp);
1652 
1653 		/* vfs don't have a default status block */
1654 		if (IS_PF(bp))
1655 			nvecs++;
1656 
1657 		bnx2x_free_msix_irqs(bp, nvecs);
1658 	} else {
1659 		free_irq(bp->dev->irq, bp->dev);
1660 	}
1661 }
1662 
1663 int bnx2x_enable_msix(struct bnx2x *bp)
1664 {
1665 	int msix_vec = 0, i, rc;
1666 
1667 	/* VFs don't have a default status block */
1668 	if (IS_PF(bp)) {
1669 		bp->msix_table[msix_vec].entry = msix_vec;
1670 		BNX2X_DEV_INFO("msix_table[0].entry = %d (slowpath)\n",
1671 			       bp->msix_table[0].entry);
1672 		msix_vec++;
1673 	}
1674 
1675 	/* Cnic requires an msix vector for itself */
1676 	if (CNIC_SUPPORT(bp)) {
1677 		bp->msix_table[msix_vec].entry = msix_vec;
1678 		BNX2X_DEV_INFO("msix_table[%d].entry = %d (CNIC)\n",
1679 			       msix_vec, bp->msix_table[msix_vec].entry);
1680 		msix_vec++;
1681 	}
1682 
1683 	/* We need separate vectors for ETH queues only (not FCoE) */
1684 	for_each_eth_queue(bp, i) {
1685 		bp->msix_table[msix_vec].entry = msix_vec;
1686 		BNX2X_DEV_INFO("msix_table[%d].entry = %d (fastpath #%u)\n",
1687 			       msix_vec, msix_vec, i);
1688 		msix_vec++;
1689 	}
1690 
1691 	DP(BNX2X_MSG_SP, "about to request enable msix with %d vectors\n",
1692 	   msix_vec);
1693 
1694 	rc = pci_enable_msix_range(bp->pdev, &bp->msix_table[0],
1695 				   BNX2X_MIN_MSIX_VEC_CNT(bp), msix_vec);
1696 	/*
1697 	 * reconfigure number of tx/rx queues according to available
1698 	 * MSI-X vectors
1699 	 */
1700 	if (rc == -ENOSPC) {
1701 		/* Get by with single vector */
1702 		rc = pci_enable_msix_range(bp->pdev, &bp->msix_table[0], 1, 1);
1703 		if (rc < 0) {
1704 			BNX2X_DEV_INFO("Single MSI-X is not attainable rc %d\n",
1705 				       rc);
1706 			goto no_msix;
1707 		}
1708 
1709 		BNX2X_DEV_INFO("Using single MSI-X vector\n");
1710 		bp->flags |= USING_SINGLE_MSIX_FLAG;
1711 
1712 		BNX2X_DEV_INFO("set number of queues to 1\n");
1713 		bp->num_ethernet_queues = 1;
1714 		bp->num_queues = bp->num_ethernet_queues + bp->num_cnic_queues;
1715 	} else if (rc < 0) {
1716 		BNX2X_DEV_INFO("MSI-X is not attainable rc %d\n", rc);
1717 		goto no_msix;
1718 	} else if (rc < msix_vec) {
1719 		/* how less vectors we will have? */
1720 		int diff = msix_vec - rc;
1721 
1722 		BNX2X_DEV_INFO("Trying to use less MSI-X vectors: %d\n", rc);
1723 
1724 		/*
1725 		 * decrease number of queues by number of unallocated entries
1726 		 */
1727 		bp->num_ethernet_queues -= diff;
1728 		bp->num_queues = bp->num_ethernet_queues + bp->num_cnic_queues;
1729 
1730 		BNX2X_DEV_INFO("New queue configuration set: %d\n",
1731 			       bp->num_queues);
1732 	}
1733 
1734 	bp->flags |= USING_MSIX_FLAG;
1735 
1736 	return 0;
1737 
1738 no_msix:
1739 	/* fall to INTx if not enough memory */
1740 	if (rc == -ENOMEM)
1741 		bp->flags |= DISABLE_MSI_FLAG;
1742 
1743 	return rc;
1744 }
1745 
1746 static int bnx2x_req_msix_irqs(struct bnx2x *bp)
1747 {
1748 	int i, rc, offset = 0;
1749 
1750 	/* no default status block for vf */
1751 	if (IS_PF(bp)) {
1752 		rc = request_irq(bp->msix_table[offset++].vector,
1753 				 bnx2x_msix_sp_int, 0,
1754 				 bp->dev->name, bp->dev);
1755 		if (rc) {
1756 			BNX2X_ERR("request sp irq failed\n");
1757 			return -EBUSY;
1758 		}
1759 	}
1760 
1761 	if (CNIC_SUPPORT(bp))
1762 		offset++;
1763 
1764 	for_each_eth_queue(bp, i) {
1765 		struct bnx2x_fastpath *fp = &bp->fp[i];
1766 		snprintf(fp->name, sizeof(fp->name), "%s-fp-%d",
1767 			 bp->dev->name, i);
1768 
1769 		rc = request_irq(bp->msix_table[offset].vector,
1770 				 bnx2x_msix_fp_int, 0, fp->name, fp);
1771 		if (rc) {
1772 			BNX2X_ERR("request fp #%d irq (%d) failed  rc %d\n", i,
1773 			      bp->msix_table[offset].vector, rc);
1774 			bnx2x_free_msix_irqs(bp, offset);
1775 			return -EBUSY;
1776 		}
1777 
1778 		offset++;
1779 	}
1780 
1781 	i = BNX2X_NUM_ETH_QUEUES(bp);
1782 	if (IS_PF(bp)) {
1783 		offset = 1 + CNIC_SUPPORT(bp);
1784 		netdev_info(bp->dev,
1785 			    "using MSI-X  IRQs: sp %d  fp[%d] %d ... fp[%d] %d\n",
1786 			    bp->msix_table[0].vector,
1787 			    0, bp->msix_table[offset].vector,
1788 			    i - 1, bp->msix_table[offset + i - 1].vector);
1789 	} else {
1790 		offset = CNIC_SUPPORT(bp);
1791 		netdev_info(bp->dev,
1792 			    "using MSI-X  IRQs: fp[%d] %d ... fp[%d] %d\n",
1793 			    0, bp->msix_table[offset].vector,
1794 			    i - 1, bp->msix_table[offset + i - 1].vector);
1795 	}
1796 	return 0;
1797 }
1798 
1799 int bnx2x_enable_msi(struct bnx2x *bp)
1800 {
1801 	int rc;
1802 
1803 	rc = pci_enable_msi(bp->pdev);
1804 	if (rc) {
1805 		BNX2X_DEV_INFO("MSI is not attainable\n");
1806 		return -1;
1807 	}
1808 	bp->flags |= USING_MSI_FLAG;
1809 
1810 	return 0;
1811 }
1812 
1813 static int bnx2x_req_irq(struct bnx2x *bp)
1814 {
1815 	unsigned long flags;
1816 	unsigned int irq;
1817 
1818 	if (bp->flags & (USING_MSI_FLAG | USING_MSIX_FLAG))
1819 		flags = 0;
1820 	else
1821 		flags = IRQF_SHARED;
1822 
1823 	if (bp->flags & USING_MSIX_FLAG)
1824 		irq = bp->msix_table[0].vector;
1825 	else
1826 		irq = bp->pdev->irq;
1827 
1828 	return request_irq(irq, bnx2x_interrupt, flags, bp->dev->name, bp->dev);
1829 }
1830 
1831 static int bnx2x_setup_irqs(struct bnx2x *bp)
1832 {
1833 	int rc = 0;
1834 	if (bp->flags & USING_MSIX_FLAG &&
1835 	    !(bp->flags & USING_SINGLE_MSIX_FLAG)) {
1836 		rc = bnx2x_req_msix_irqs(bp);
1837 		if (rc)
1838 			return rc;
1839 	} else {
1840 		rc = bnx2x_req_irq(bp);
1841 		if (rc) {
1842 			BNX2X_ERR("IRQ request failed  rc %d, aborting\n", rc);
1843 			return rc;
1844 		}
1845 		if (bp->flags & USING_MSI_FLAG) {
1846 			bp->dev->irq = bp->pdev->irq;
1847 			netdev_info(bp->dev, "using MSI IRQ %d\n",
1848 				    bp->dev->irq);
1849 		}
1850 		if (bp->flags & USING_MSIX_FLAG) {
1851 			bp->dev->irq = bp->msix_table[0].vector;
1852 			netdev_info(bp->dev, "using MSIX IRQ %d\n",
1853 				    bp->dev->irq);
1854 		}
1855 	}
1856 
1857 	return 0;
1858 }
1859 
1860 static void bnx2x_napi_enable_cnic(struct bnx2x *bp)
1861 {
1862 	int i;
1863 
1864 	for_each_rx_queue_cnic(bp, i) {
1865 		napi_enable(&bnx2x_fp(bp, i, napi));
1866 	}
1867 }
1868 
1869 static void bnx2x_napi_enable(struct bnx2x *bp)
1870 {
1871 	int i;
1872 
1873 	for_each_eth_queue(bp, i) {
1874 		napi_enable(&bnx2x_fp(bp, i, napi));
1875 	}
1876 }
1877 
1878 static void bnx2x_napi_disable_cnic(struct bnx2x *bp)
1879 {
1880 	int i;
1881 
1882 	for_each_rx_queue_cnic(bp, i) {
1883 		napi_disable(&bnx2x_fp(bp, i, napi));
1884 	}
1885 }
1886 
1887 static void bnx2x_napi_disable(struct bnx2x *bp)
1888 {
1889 	int i;
1890 
1891 	for_each_eth_queue(bp, i) {
1892 		napi_disable(&bnx2x_fp(bp, i, napi));
1893 	}
1894 }
1895 
1896 void bnx2x_netif_start(struct bnx2x *bp)
1897 {
1898 	if (netif_running(bp->dev)) {
1899 		bnx2x_napi_enable(bp);
1900 		if (CNIC_LOADED(bp))
1901 			bnx2x_napi_enable_cnic(bp);
1902 		bnx2x_int_enable(bp);
1903 		if (bp->state == BNX2X_STATE_OPEN)
1904 			netif_tx_wake_all_queues(bp->dev);
1905 	}
1906 }
1907 
1908 void bnx2x_netif_stop(struct bnx2x *bp, int disable_hw)
1909 {
1910 	bnx2x_int_disable_sync(bp, disable_hw);
1911 	bnx2x_napi_disable(bp);
1912 	if (CNIC_LOADED(bp))
1913 		bnx2x_napi_disable_cnic(bp);
1914 }
1915 
1916 u16 bnx2x_select_queue(struct net_device *dev, struct sk_buff *skb,
1917 		       void *accel_priv, select_queue_fallback_t fallback)
1918 {
1919 	struct bnx2x *bp = netdev_priv(dev);
1920 
1921 	if (CNIC_LOADED(bp) && !NO_FCOE(bp)) {
1922 		struct ethhdr *hdr = (struct ethhdr *)skb->data;
1923 		u16 ether_type = ntohs(hdr->h_proto);
1924 
1925 		/* Skip VLAN tag if present */
1926 		if (ether_type == ETH_P_8021Q) {
1927 			struct vlan_ethhdr *vhdr =
1928 				(struct vlan_ethhdr *)skb->data;
1929 
1930 			ether_type = ntohs(vhdr->h_vlan_encapsulated_proto);
1931 		}
1932 
1933 		/* If ethertype is FCoE or FIP - use FCoE ring */
1934 		if ((ether_type == ETH_P_FCOE) || (ether_type == ETH_P_FIP))
1935 			return bnx2x_fcoe_tx(bp, txq_index);
1936 	}
1937 
1938 	/* select a non-FCoE queue */
1939 	return fallback(dev, skb) % BNX2X_NUM_ETH_QUEUES(bp);
1940 }
1941 
1942 void bnx2x_set_num_queues(struct bnx2x *bp)
1943 {
1944 	/* RSS queues */
1945 	bp->num_ethernet_queues = bnx2x_calc_num_queues(bp);
1946 
1947 	/* override in STORAGE SD modes */
1948 	if (IS_MF_STORAGE_ONLY(bp))
1949 		bp->num_ethernet_queues = 1;
1950 
1951 	/* Add special queues */
1952 	bp->num_cnic_queues = CNIC_SUPPORT(bp); /* For FCOE */
1953 	bp->num_queues = bp->num_ethernet_queues + bp->num_cnic_queues;
1954 
1955 	BNX2X_DEV_INFO("set number of queues to %d\n", bp->num_queues);
1956 }
1957 
1958 /**
1959  * bnx2x_set_real_num_queues - configure netdev->real_num_[tx,rx]_queues
1960  *
1961  * @bp:		Driver handle
1962  *
1963  * We currently support for at most 16 Tx queues for each CoS thus we will
1964  * allocate a multiple of 16 for ETH L2 rings according to the value of the
1965  * bp->max_cos.
1966  *
1967  * If there is an FCoE L2 queue the appropriate Tx queue will have the next
1968  * index after all ETH L2 indices.
1969  *
1970  * If the actual number of Tx queues (for each CoS) is less than 16 then there
1971  * will be the holes at the end of each group of 16 ETh L2 indices (0..15,
1972  * 16..31,...) with indices that are not coupled with any real Tx queue.
1973  *
1974  * The proper configuration of skb->queue_mapping is handled by
1975  * bnx2x_select_queue() and __skb_tx_hash().
1976  *
1977  * bnx2x_setup_tc() takes care of the proper TC mappings so that __skb_tx_hash()
1978  * will return a proper Tx index if TC is enabled (netdev->num_tc > 0).
1979  */
1980 static int bnx2x_set_real_num_queues(struct bnx2x *bp, int include_cnic)
1981 {
1982 	int rc, tx, rx;
1983 
1984 	tx = BNX2X_NUM_ETH_QUEUES(bp) * bp->max_cos;
1985 	rx = BNX2X_NUM_ETH_QUEUES(bp);
1986 
1987 /* account for fcoe queue */
1988 	if (include_cnic && !NO_FCOE(bp)) {
1989 		rx++;
1990 		tx++;
1991 	}
1992 
1993 	rc = netif_set_real_num_tx_queues(bp->dev, tx);
1994 	if (rc) {
1995 		BNX2X_ERR("Failed to set real number of Tx queues: %d\n", rc);
1996 		return rc;
1997 	}
1998 	rc = netif_set_real_num_rx_queues(bp->dev, rx);
1999 	if (rc) {
2000 		BNX2X_ERR("Failed to set real number of Rx queues: %d\n", rc);
2001 		return rc;
2002 	}
2003 
2004 	DP(NETIF_MSG_IFUP, "Setting real num queues to (tx, rx) (%d, %d)\n",
2005 			  tx, rx);
2006 
2007 	return rc;
2008 }
2009 
2010 static void bnx2x_set_rx_buf_size(struct bnx2x *bp)
2011 {
2012 	int i;
2013 
2014 	for_each_queue(bp, i) {
2015 		struct bnx2x_fastpath *fp = &bp->fp[i];
2016 		u32 mtu;
2017 
2018 		/* Always use a mini-jumbo MTU for the FCoE L2 ring */
2019 		if (IS_FCOE_IDX(i))
2020 			/*
2021 			 * Although there are no IP frames expected to arrive to
2022 			 * this ring we still want to add an
2023 			 * IP_HEADER_ALIGNMENT_PADDING to prevent a buffer
2024 			 * overrun attack.
2025 			 */
2026 			mtu = BNX2X_FCOE_MINI_JUMBO_MTU;
2027 		else
2028 			mtu = bp->dev->mtu;
2029 		fp->rx_buf_size = BNX2X_FW_RX_ALIGN_START +
2030 				  IP_HEADER_ALIGNMENT_PADDING +
2031 				  ETH_OVREHEAD +
2032 				  mtu +
2033 				  BNX2X_FW_RX_ALIGN_END;
2034 		/* Note : rx_buf_size doesn't take into account NET_SKB_PAD */
2035 		if (fp->rx_buf_size + NET_SKB_PAD <= PAGE_SIZE)
2036 			fp->rx_frag_size = fp->rx_buf_size + NET_SKB_PAD;
2037 		else
2038 			fp->rx_frag_size = 0;
2039 	}
2040 }
2041 
2042 static int bnx2x_init_rss(struct bnx2x *bp)
2043 {
2044 	int i;
2045 	u8 num_eth_queues = BNX2X_NUM_ETH_QUEUES(bp);
2046 
2047 	/* Prepare the initial contents for the indirection table if RSS is
2048 	 * enabled
2049 	 */
2050 	for (i = 0; i < sizeof(bp->rss_conf_obj.ind_table); i++)
2051 		bp->rss_conf_obj.ind_table[i] =
2052 			bp->fp->cl_id +
2053 			ethtool_rxfh_indir_default(i, num_eth_queues);
2054 
2055 	/*
2056 	 * For 57710 and 57711 SEARCHER configuration (rss_keys) is
2057 	 * per-port, so if explicit configuration is needed , do it only
2058 	 * for a PMF.
2059 	 *
2060 	 * For 57712 and newer on the other hand it's a per-function
2061 	 * configuration.
2062 	 */
2063 	return bnx2x_config_rss_eth(bp, bp->port.pmf || !CHIP_IS_E1x(bp));
2064 }
2065 
2066 int bnx2x_rss(struct bnx2x *bp, struct bnx2x_rss_config_obj *rss_obj,
2067 	      bool config_hash, bool enable)
2068 {
2069 	struct bnx2x_config_rss_params params = {NULL};
2070 
2071 	/* Although RSS is meaningless when there is a single HW queue we
2072 	 * still need it enabled in order to have HW Rx hash generated.
2073 	 *
2074 	 * if (!is_eth_multi(bp))
2075 	 *      bp->multi_mode = ETH_RSS_MODE_DISABLED;
2076 	 */
2077 
2078 	params.rss_obj = rss_obj;
2079 
2080 	__set_bit(RAMROD_COMP_WAIT, &params.ramrod_flags);
2081 
2082 	if (enable) {
2083 		__set_bit(BNX2X_RSS_MODE_REGULAR, &params.rss_flags);
2084 
2085 		/* RSS configuration */
2086 		__set_bit(BNX2X_RSS_IPV4, &params.rss_flags);
2087 		__set_bit(BNX2X_RSS_IPV4_TCP, &params.rss_flags);
2088 		__set_bit(BNX2X_RSS_IPV6, &params.rss_flags);
2089 		__set_bit(BNX2X_RSS_IPV6_TCP, &params.rss_flags);
2090 		if (rss_obj->udp_rss_v4)
2091 			__set_bit(BNX2X_RSS_IPV4_UDP, &params.rss_flags);
2092 		if (rss_obj->udp_rss_v6)
2093 			__set_bit(BNX2X_RSS_IPV6_UDP, &params.rss_flags);
2094 
2095 		if (!CHIP_IS_E1x(bp)) {
2096 			/* valid only for TUNN_MODE_VXLAN tunnel mode */
2097 			__set_bit(BNX2X_RSS_IPV4_VXLAN, &params.rss_flags);
2098 			__set_bit(BNX2X_RSS_IPV6_VXLAN, &params.rss_flags);
2099 
2100 			/* valid only for TUNN_MODE_GRE tunnel mode */
2101 			__set_bit(BNX2X_RSS_TUNN_INNER_HDRS, &params.rss_flags);
2102 		}
2103 	} else {
2104 		__set_bit(BNX2X_RSS_MODE_DISABLED, &params.rss_flags);
2105 	}
2106 
2107 	/* Hash bits */
2108 	params.rss_result_mask = MULTI_MASK;
2109 
2110 	memcpy(params.ind_table, rss_obj->ind_table, sizeof(params.ind_table));
2111 
2112 	if (config_hash) {
2113 		/* RSS keys */
2114 		netdev_rss_key_fill(params.rss_key, T_ETH_RSS_KEY * 4);
2115 		__set_bit(BNX2X_RSS_SET_SRCH, &params.rss_flags);
2116 	}
2117 
2118 	if (IS_PF(bp))
2119 		return bnx2x_config_rss(bp, &params);
2120 	else
2121 		return bnx2x_vfpf_config_rss(bp, &params);
2122 }
2123 
2124 static int bnx2x_init_hw(struct bnx2x *bp, u32 load_code)
2125 {
2126 	struct bnx2x_func_state_params func_params = {NULL};
2127 
2128 	/* Prepare parameters for function state transitions */
2129 	__set_bit(RAMROD_COMP_WAIT, &func_params.ramrod_flags);
2130 
2131 	func_params.f_obj = &bp->func_obj;
2132 	func_params.cmd = BNX2X_F_CMD_HW_INIT;
2133 
2134 	func_params.params.hw_init.load_phase = load_code;
2135 
2136 	return bnx2x_func_state_change(bp, &func_params);
2137 }
2138 
2139 /*
2140  * Cleans the object that have internal lists without sending
2141  * ramrods. Should be run when interrupts are disabled.
2142  */
2143 void bnx2x_squeeze_objects(struct bnx2x *bp)
2144 {
2145 	int rc;
2146 	unsigned long ramrod_flags = 0, vlan_mac_flags = 0;
2147 	struct bnx2x_mcast_ramrod_params rparam = {NULL};
2148 	struct bnx2x_vlan_mac_obj *mac_obj = &bp->sp_objs->mac_obj;
2149 
2150 	/***************** Cleanup MACs' object first *************************/
2151 
2152 	/* Wait for completion of requested */
2153 	__set_bit(RAMROD_COMP_WAIT, &ramrod_flags);
2154 	/* Perform a dry cleanup */
2155 	__set_bit(RAMROD_DRV_CLR_ONLY, &ramrod_flags);
2156 
2157 	/* Clean ETH primary MAC */
2158 	__set_bit(BNX2X_ETH_MAC, &vlan_mac_flags);
2159 	rc = mac_obj->delete_all(bp, &bp->sp_objs->mac_obj, &vlan_mac_flags,
2160 				 &ramrod_flags);
2161 	if (rc != 0)
2162 		BNX2X_ERR("Failed to clean ETH MACs: %d\n", rc);
2163 
2164 	/* Cleanup UC list */
2165 	vlan_mac_flags = 0;
2166 	__set_bit(BNX2X_UC_LIST_MAC, &vlan_mac_flags);
2167 	rc = mac_obj->delete_all(bp, mac_obj, &vlan_mac_flags,
2168 				 &ramrod_flags);
2169 	if (rc != 0)
2170 		BNX2X_ERR("Failed to clean UC list MACs: %d\n", rc);
2171 
2172 	/***************** Now clean mcast object *****************************/
2173 	rparam.mcast_obj = &bp->mcast_obj;
2174 	__set_bit(RAMROD_DRV_CLR_ONLY, &rparam.ramrod_flags);
2175 
2176 	/* Add a DEL command... - Since we're doing a driver cleanup only,
2177 	 * we take a lock surrounding both the initial send and the CONTs,
2178 	 * as we don't want a true completion to disrupt us in the middle.
2179 	 */
2180 	netif_addr_lock_bh(bp->dev);
2181 	rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_DEL);
2182 	if (rc < 0)
2183 		BNX2X_ERR("Failed to add a new DEL command to a multi-cast object: %d\n",
2184 			  rc);
2185 
2186 	/* ...and wait until all pending commands are cleared */
2187 	rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_CONT);
2188 	while (rc != 0) {
2189 		if (rc < 0) {
2190 			BNX2X_ERR("Failed to clean multi-cast object: %d\n",
2191 				  rc);
2192 			netif_addr_unlock_bh(bp->dev);
2193 			return;
2194 		}
2195 
2196 		rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_CONT);
2197 	}
2198 	netif_addr_unlock_bh(bp->dev);
2199 }
2200 
2201 #ifndef BNX2X_STOP_ON_ERROR
2202 #define LOAD_ERROR_EXIT(bp, label) \
2203 	do { \
2204 		(bp)->state = BNX2X_STATE_ERROR; \
2205 		goto label; \
2206 	} while (0)
2207 
2208 #define LOAD_ERROR_EXIT_CNIC(bp, label) \
2209 	do { \
2210 		bp->cnic_loaded = false; \
2211 		goto label; \
2212 	} while (0)
2213 #else /*BNX2X_STOP_ON_ERROR*/
2214 #define LOAD_ERROR_EXIT(bp, label) \
2215 	do { \
2216 		(bp)->state = BNX2X_STATE_ERROR; \
2217 		(bp)->panic = 1; \
2218 		return -EBUSY; \
2219 	} while (0)
2220 #define LOAD_ERROR_EXIT_CNIC(bp, label) \
2221 	do { \
2222 		bp->cnic_loaded = false; \
2223 		(bp)->panic = 1; \
2224 		return -EBUSY; \
2225 	} while (0)
2226 #endif /*BNX2X_STOP_ON_ERROR*/
2227 
2228 static void bnx2x_free_fw_stats_mem(struct bnx2x *bp)
2229 {
2230 	BNX2X_PCI_FREE(bp->fw_stats, bp->fw_stats_mapping,
2231 		       bp->fw_stats_data_sz + bp->fw_stats_req_sz);
2232 	return;
2233 }
2234 
2235 static int bnx2x_alloc_fw_stats_mem(struct bnx2x *bp)
2236 {
2237 	int num_groups, vf_headroom = 0;
2238 	int is_fcoe_stats = NO_FCOE(bp) ? 0 : 1;
2239 
2240 	/* number of queues for statistics is number of eth queues + FCoE */
2241 	u8 num_queue_stats = BNX2X_NUM_ETH_QUEUES(bp) + is_fcoe_stats;
2242 
2243 	/* Total number of FW statistics requests =
2244 	 * 1 for port stats + 1 for PF stats + potential 2 for FCoE (fcoe proper
2245 	 * and fcoe l2 queue) stats + num of queues (which includes another 1
2246 	 * for fcoe l2 queue if applicable)
2247 	 */
2248 	bp->fw_stats_num = 2 + is_fcoe_stats + num_queue_stats;
2249 
2250 	/* vf stats appear in the request list, but their data is allocated by
2251 	 * the VFs themselves. We don't include them in the bp->fw_stats_num as
2252 	 * it is used to determine where to place the vf stats queries in the
2253 	 * request struct
2254 	 */
2255 	if (IS_SRIOV(bp))
2256 		vf_headroom = bnx2x_vf_headroom(bp);
2257 
2258 	/* Request is built from stats_query_header and an array of
2259 	 * stats_query_cmd_group each of which contains
2260 	 * STATS_QUERY_CMD_COUNT rules. The real number or requests is
2261 	 * configured in the stats_query_header.
2262 	 */
2263 	num_groups =
2264 		(((bp->fw_stats_num + vf_headroom) / STATS_QUERY_CMD_COUNT) +
2265 		 (((bp->fw_stats_num + vf_headroom) % STATS_QUERY_CMD_COUNT) ?
2266 		 1 : 0));
2267 
2268 	DP(BNX2X_MSG_SP, "stats fw_stats_num %d, vf headroom %d, num_groups %d\n",
2269 	   bp->fw_stats_num, vf_headroom, num_groups);
2270 	bp->fw_stats_req_sz = sizeof(struct stats_query_header) +
2271 		num_groups * sizeof(struct stats_query_cmd_group);
2272 
2273 	/* Data for statistics requests + stats_counter
2274 	 * stats_counter holds per-STORM counters that are incremented
2275 	 * when STORM has finished with the current request.
2276 	 * memory for FCoE offloaded statistics are counted anyway,
2277 	 * even if they will not be sent.
2278 	 * VF stats are not accounted for here as the data of VF stats is stored
2279 	 * in memory allocated by the VF, not here.
2280 	 */
2281 	bp->fw_stats_data_sz = sizeof(struct per_port_stats) +
2282 		sizeof(struct per_pf_stats) +
2283 		sizeof(struct fcoe_statistics_params) +
2284 		sizeof(struct per_queue_stats) * num_queue_stats +
2285 		sizeof(struct stats_counter);
2286 
2287 	bp->fw_stats = BNX2X_PCI_ALLOC(&bp->fw_stats_mapping,
2288 				       bp->fw_stats_data_sz + bp->fw_stats_req_sz);
2289 	if (!bp->fw_stats)
2290 		goto alloc_mem_err;
2291 
2292 	/* Set shortcuts */
2293 	bp->fw_stats_req = (struct bnx2x_fw_stats_req *)bp->fw_stats;
2294 	bp->fw_stats_req_mapping = bp->fw_stats_mapping;
2295 	bp->fw_stats_data = (struct bnx2x_fw_stats_data *)
2296 		((u8 *)bp->fw_stats + bp->fw_stats_req_sz);
2297 	bp->fw_stats_data_mapping = bp->fw_stats_mapping +
2298 		bp->fw_stats_req_sz;
2299 
2300 	DP(BNX2X_MSG_SP, "statistics request base address set to %x %x\n",
2301 	   U64_HI(bp->fw_stats_req_mapping),
2302 	   U64_LO(bp->fw_stats_req_mapping));
2303 	DP(BNX2X_MSG_SP, "statistics data base address set to %x %x\n",
2304 	   U64_HI(bp->fw_stats_data_mapping),
2305 	   U64_LO(bp->fw_stats_data_mapping));
2306 	return 0;
2307 
2308 alloc_mem_err:
2309 	bnx2x_free_fw_stats_mem(bp);
2310 	BNX2X_ERR("Can't allocate FW stats memory\n");
2311 	return -ENOMEM;
2312 }
2313 
2314 /* send load request to mcp and analyze response */
2315 static int bnx2x_nic_load_request(struct bnx2x *bp, u32 *load_code)
2316 {
2317 	u32 param;
2318 
2319 	/* init fw_seq */
2320 	bp->fw_seq =
2321 		(SHMEM_RD(bp, func_mb[BP_FW_MB_IDX(bp)].drv_mb_header) &
2322 		 DRV_MSG_SEQ_NUMBER_MASK);
2323 	BNX2X_DEV_INFO("fw_seq 0x%08x\n", bp->fw_seq);
2324 
2325 	/* Get current FW pulse sequence */
2326 	bp->fw_drv_pulse_wr_seq =
2327 		(SHMEM_RD(bp, func_mb[BP_FW_MB_IDX(bp)].drv_pulse_mb) &
2328 		 DRV_PULSE_SEQ_MASK);
2329 	BNX2X_DEV_INFO("drv_pulse 0x%x\n", bp->fw_drv_pulse_wr_seq);
2330 
2331 	param = DRV_MSG_CODE_LOAD_REQ_WITH_LFA;
2332 
2333 	if (IS_MF_SD(bp) && bnx2x_port_after_undi(bp))
2334 		param |= DRV_MSG_CODE_LOAD_REQ_FORCE_LFA;
2335 
2336 	/* load request */
2337 	(*load_code) = bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_REQ, param);
2338 
2339 	/* if mcp fails to respond we must abort */
2340 	if (!(*load_code)) {
2341 		BNX2X_ERR("MCP response failure, aborting\n");
2342 		return -EBUSY;
2343 	}
2344 
2345 	/* If mcp refused (e.g. other port is in diagnostic mode) we
2346 	 * must abort
2347 	 */
2348 	if ((*load_code) == FW_MSG_CODE_DRV_LOAD_REFUSED) {
2349 		BNX2X_ERR("MCP refused load request, aborting\n");
2350 		return -EBUSY;
2351 	}
2352 	return 0;
2353 }
2354 
2355 /* check whether another PF has already loaded FW to chip. In
2356  * virtualized environments a pf from another VM may have already
2357  * initialized the device including loading FW
2358  */
2359 int bnx2x_compare_fw_ver(struct bnx2x *bp, u32 load_code, bool print_err)
2360 {
2361 	/* is another pf loaded on this engine? */
2362 	if (load_code != FW_MSG_CODE_DRV_LOAD_COMMON_CHIP &&
2363 	    load_code != FW_MSG_CODE_DRV_LOAD_COMMON) {
2364 		/* build my FW version dword */
2365 		u32 my_fw = (BCM_5710_FW_MAJOR_VERSION) +
2366 			(BCM_5710_FW_MINOR_VERSION << 8) +
2367 			(BCM_5710_FW_REVISION_VERSION << 16) +
2368 			(BCM_5710_FW_ENGINEERING_VERSION << 24);
2369 
2370 		/* read loaded FW from chip */
2371 		u32 loaded_fw = REG_RD(bp, XSEM_REG_PRAM);
2372 
2373 		DP(BNX2X_MSG_SP, "loaded fw %x, my fw %x\n",
2374 		   loaded_fw, my_fw);
2375 
2376 		/* abort nic load if version mismatch */
2377 		if (my_fw != loaded_fw) {
2378 			if (print_err)
2379 				BNX2X_ERR("bnx2x with FW %x was already loaded which mismatches my %x FW. Aborting\n",
2380 					  loaded_fw, my_fw);
2381 			else
2382 				BNX2X_DEV_INFO("bnx2x with FW %x was already loaded which mismatches my %x FW, possibly due to MF UNDI\n",
2383 					       loaded_fw, my_fw);
2384 			return -EBUSY;
2385 		}
2386 	}
2387 	return 0;
2388 }
2389 
2390 /* returns the "mcp load_code" according to global load_count array */
2391 static int bnx2x_nic_load_no_mcp(struct bnx2x *bp, int port)
2392 {
2393 	int path = BP_PATH(bp);
2394 
2395 	DP(NETIF_MSG_IFUP, "NO MCP - load counts[%d]      %d, %d, %d\n",
2396 	   path, bnx2x_load_count[path][0], bnx2x_load_count[path][1],
2397 	   bnx2x_load_count[path][2]);
2398 	bnx2x_load_count[path][0]++;
2399 	bnx2x_load_count[path][1 + port]++;
2400 	DP(NETIF_MSG_IFUP, "NO MCP - new load counts[%d]  %d, %d, %d\n",
2401 	   path, bnx2x_load_count[path][0], bnx2x_load_count[path][1],
2402 	   bnx2x_load_count[path][2]);
2403 	if (bnx2x_load_count[path][0] == 1)
2404 		return FW_MSG_CODE_DRV_LOAD_COMMON;
2405 	else if (bnx2x_load_count[path][1 + port] == 1)
2406 		return FW_MSG_CODE_DRV_LOAD_PORT;
2407 	else
2408 		return FW_MSG_CODE_DRV_LOAD_FUNCTION;
2409 }
2410 
2411 /* mark PMF if applicable */
2412 static void bnx2x_nic_load_pmf(struct bnx2x *bp, u32 load_code)
2413 {
2414 	if ((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
2415 	    (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP) ||
2416 	    (load_code == FW_MSG_CODE_DRV_LOAD_PORT)) {
2417 		bp->port.pmf = 1;
2418 		/* We need the barrier to ensure the ordering between the
2419 		 * writing to bp->port.pmf here and reading it from the
2420 		 * bnx2x_periodic_task().
2421 		 */
2422 		smp_mb();
2423 	} else {
2424 		bp->port.pmf = 0;
2425 	}
2426 
2427 	DP(NETIF_MSG_LINK, "pmf %d\n", bp->port.pmf);
2428 }
2429 
2430 static void bnx2x_nic_load_afex_dcc(struct bnx2x *bp, int load_code)
2431 {
2432 	if (((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
2433 	     (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP)) &&
2434 	    (bp->common.shmem2_base)) {
2435 		if (SHMEM2_HAS(bp, dcc_support))
2436 			SHMEM2_WR(bp, dcc_support,
2437 				  (SHMEM_DCC_SUPPORT_DISABLE_ENABLE_PF_TLV |
2438 				   SHMEM_DCC_SUPPORT_BANDWIDTH_ALLOCATION_TLV));
2439 		if (SHMEM2_HAS(bp, afex_driver_support))
2440 			SHMEM2_WR(bp, afex_driver_support,
2441 				  SHMEM_AFEX_SUPPORTED_VERSION_ONE);
2442 	}
2443 
2444 	/* Set AFEX default VLAN tag to an invalid value */
2445 	bp->afex_def_vlan_tag = -1;
2446 }
2447 
2448 /**
2449  * bnx2x_bz_fp - zero content of the fastpath structure.
2450  *
2451  * @bp:		driver handle
2452  * @index:	fastpath index to be zeroed
2453  *
2454  * Makes sure the contents of the bp->fp[index].napi is kept
2455  * intact.
2456  */
2457 static void bnx2x_bz_fp(struct bnx2x *bp, int index)
2458 {
2459 	struct bnx2x_fastpath *fp = &bp->fp[index];
2460 	int cos;
2461 	struct napi_struct orig_napi = fp->napi;
2462 	struct bnx2x_agg_info *orig_tpa_info = fp->tpa_info;
2463 
2464 	/* bzero bnx2x_fastpath contents */
2465 	if (fp->tpa_info)
2466 		memset(fp->tpa_info, 0, ETH_MAX_AGGREGATION_QUEUES_E1H_E2 *
2467 		       sizeof(struct bnx2x_agg_info));
2468 	memset(fp, 0, sizeof(*fp));
2469 
2470 	/* Restore the NAPI object as it has been already initialized */
2471 	fp->napi = orig_napi;
2472 	fp->tpa_info = orig_tpa_info;
2473 	fp->bp = bp;
2474 	fp->index = index;
2475 	if (IS_ETH_FP(fp))
2476 		fp->max_cos = bp->max_cos;
2477 	else
2478 		/* Special queues support only one CoS */
2479 		fp->max_cos = 1;
2480 
2481 	/* Init txdata pointers */
2482 	if (IS_FCOE_FP(fp))
2483 		fp->txdata_ptr[0] = &bp->bnx2x_txq[FCOE_TXQ_IDX(bp)];
2484 	if (IS_ETH_FP(fp))
2485 		for_each_cos_in_tx_queue(fp, cos)
2486 			fp->txdata_ptr[cos] = &bp->bnx2x_txq[cos *
2487 				BNX2X_NUM_ETH_QUEUES(bp) + index];
2488 
2489 	/* set the tpa flag for each queue. The tpa flag determines the queue
2490 	 * minimal size so it must be set prior to queue memory allocation
2491 	 */
2492 	if (bp->dev->features & NETIF_F_LRO)
2493 		fp->mode = TPA_MODE_LRO;
2494 	else if (bp->dev->features & NETIF_F_GRO &&
2495 		 bnx2x_mtu_allows_gro(bp->dev->mtu))
2496 		fp->mode = TPA_MODE_GRO;
2497 	else
2498 		fp->mode = TPA_MODE_DISABLED;
2499 
2500 	/* We don't want TPA if it's disabled in bp
2501 	 * or if this is an FCoE L2 ring.
2502 	 */
2503 	if (bp->disable_tpa || IS_FCOE_FP(fp))
2504 		fp->mode = TPA_MODE_DISABLED;
2505 }
2506 
2507 void bnx2x_set_os_driver_state(struct bnx2x *bp, u32 state)
2508 {
2509 	u32 cur;
2510 
2511 	if (!IS_MF_BD(bp) || !SHMEM2_HAS(bp, os_driver_state) || IS_VF(bp))
2512 		return;
2513 
2514 	cur = SHMEM2_RD(bp, os_driver_state[BP_FW_MB_IDX(bp)]);
2515 	DP(NETIF_MSG_IFUP, "Driver state %08x-->%08x\n",
2516 	   cur, state);
2517 
2518 	SHMEM2_WR(bp, os_driver_state[BP_FW_MB_IDX(bp)], state);
2519 }
2520 
2521 int bnx2x_load_cnic(struct bnx2x *bp)
2522 {
2523 	int i, rc, port = BP_PORT(bp);
2524 
2525 	DP(NETIF_MSG_IFUP, "Starting CNIC-related load\n");
2526 
2527 	mutex_init(&bp->cnic_mutex);
2528 
2529 	if (IS_PF(bp)) {
2530 		rc = bnx2x_alloc_mem_cnic(bp);
2531 		if (rc) {
2532 			BNX2X_ERR("Unable to allocate bp memory for cnic\n");
2533 			LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic0);
2534 		}
2535 	}
2536 
2537 	rc = bnx2x_alloc_fp_mem_cnic(bp);
2538 	if (rc) {
2539 		BNX2X_ERR("Unable to allocate memory for cnic fps\n");
2540 		LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic0);
2541 	}
2542 
2543 	/* Update the number of queues with the cnic queues */
2544 	rc = bnx2x_set_real_num_queues(bp, 1);
2545 	if (rc) {
2546 		BNX2X_ERR("Unable to set real_num_queues including cnic\n");
2547 		LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic0);
2548 	}
2549 
2550 	/* Add all CNIC NAPI objects */
2551 	bnx2x_add_all_napi_cnic(bp);
2552 	DP(NETIF_MSG_IFUP, "cnic napi added\n");
2553 	bnx2x_napi_enable_cnic(bp);
2554 
2555 	rc = bnx2x_init_hw_func_cnic(bp);
2556 	if (rc)
2557 		LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic1);
2558 
2559 	bnx2x_nic_init_cnic(bp);
2560 
2561 	if (IS_PF(bp)) {
2562 		/* Enable Timer scan */
2563 		REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + port*4, 1);
2564 
2565 		/* setup cnic queues */
2566 		for_each_cnic_queue(bp, i) {
2567 			rc = bnx2x_setup_queue(bp, &bp->fp[i], 0);
2568 			if (rc) {
2569 				BNX2X_ERR("Queue setup failed\n");
2570 				LOAD_ERROR_EXIT(bp, load_error_cnic2);
2571 			}
2572 		}
2573 	}
2574 
2575 	/* Initialize Rx filter. */
2576 	bnx2x_set_rx_mode_inner(bp);
2577 
2578 	/* re-read iscsi info */
2579 	bnx2x_get_iscsi_info(bp);
2580 	bnx2x_setup_cnic_irq_info(bp);
2581 	bnx2x_setup_cnic_info(bp);
2582 	bp->cnic_loaded = true;
2583 	if (bp->state == BNX2X_STATE_OPEN)
2584 		bnx2x_cnic_notify(bp, CNIC_CTL_START_CMD);
2585 
2586 	DP(NETIF_MSG_IFUP, "Ending successfully CNIC-related load\n");
2587 
2588 	return 0;
2589 
2590 #ifndef BNX2X_STOP_ON_ERROR
2591 load_error_cnic2:
2592 	/* Disable Timer scan */
2593 	REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + port*4, 0);
2594 
2595 load_error_cnic1:
2596 	bnx2x_napi_disable_cnic(bp);
2597 	/* Update the number of queues without the cnic queues */
2598 	if (bnx2x_set_real_num_queues(bp, 0))
2599 		BNX2X_ERR("Unable to set real_num_queues not including cnic\n");
2600 load_error_cnic0:
2601 	BNX2X_ERR("CNIC-related load failed\n");
2602 	bnx2x_free_fp_mem_cnic(bp);
2603 	bnx2x_free_mem_cnic(bp);
2604 	return rc;
2605 #endif /* ! BNX2X_STOP_ON_ERROR */
2606 }
2607 
2608 /* must be called with rtnl_lock */
2609 int bnx2x_nic_load(struct bnx2x *bp, int load_mode)
2610 {
2611 	int port = BP_PORT(bp);
2612 	int i, rc = 0, load_code = 0;
2613 
2614 	DP(NETIF_MSG_IFUP, "Starting NIC load\n");
2615 	DP(NETIF_MSG_IFUP,
2616 	   "CNIC is %s\n", CNIC_ENABLED(bp) ? "enabled" : "disabled");
2617 
2618 #ifdef BNX2X_STOP_ON_ERROR
2619 	if (unlikely(bp->panic)) {
2620 		BNX2X_ERR("Can't load NIC when there is panic\n");
2621 		return -EPERM;
2622 	}
2623 #endif
2624 
2625 	bp->state = BNX2X_STATE_OPENING_WAIT4_LOAD;
2626 
2627 	/* zero the structure w/o any lock, before SP handler is initialized */
2628 	memset(&bp->last_reported_link, 0, sizeof(bp->last_reported_link));
2629 	__set_bit(BNX2X_LINK_REPORT_LINK_DOWN,
2630 		&bp->last_reported_link.link_report_flags);
2631 
2632 	if (IS_PF(bp))
2633 		/* must be called before memory allocation and HW init */
2634 		bnx2x_ilt_set_info(bp);
2635 
2636 	/*
2637 	 * Zero fastpath structures preserving invariants like napi, which are
2638 	 * allocated only once, fp index, max_cos, bp pointer.
2639 	 * Also set fp->mode and txdata_ptr.
2640 	 */
2641 	DP(NETIF_MSG_IFUP, "num queues: %d", bp->num_queues);
2642 	for_each_queue(bp, i)
2643 		bnx2x_bz_fp(bp, i);
2644 	memset(bp->bnx2x_txq, 0, (BNX2X_MAX_RSS_COUNT(bp) * BNX2X_MULTI_TX_COS +
2645 				  bp->num_cnic_queues) *
2646 				  sizeof(struct bnx2x_fp_txdata));
2647 
2648 	bp->fcoe_init = false;
2649 
2650 	/* Set the receive queues buffer size */
2651 	bnx2x_set_rx_buf_size(bp);
2652 
2653 	if (IS_PF(bp)) {
2654 		rc = bnx2x_alloc_mem(bp);
2655 		if (rc) {
2656 			BNX2X_ERR("Unable to allocate bp memory\n");
2657 			return rc;
2658 		}
2659 	}
2660 
2661 	/* need to be done after alloc mem, since it's self adjusting to amount
2662 	 * of memory available for RSS queues
2663 	 */
2664 	rc = bnx2x_alloc_fp_mem(bp);
2665 	if (rc) {
2666 		BNX2X_ERR("Unable to allocate memory for fps\n");
2667 		LOAD_ERROR_EXIT(bp, load_error0);
2668 	}
2669 
2670 	/* Allocated memory for FW statistics  */
2671 	if (bnx2x_alloc_fw_stats_mem(bp))
2672 		LOAD_ERROR_EXIT(bp, load_error0);
2673 
2674 	/* request pf to initialize status blocks */
2675 	if (IS_VF(bp)) {
2676 		rc = bnx2x_vfpf_init(bp);
2677 		if (rc)
2678 			LOAD_ERROR_EXIT(bp, load_error0);
2679 	}
2680 
2681 	/* As long as bnx2x_alloc_mem() may possibly update
2682 	 * bp->num_queues, bnx2x_set_real_num_queues() should always
2683 	 * come after it. At this stage cnic queues are not counted.
2684 	 */
2685 	rc = bnx2x_set_real_num_queues(bp, 0);
2686 	if (rc) {
2687 		BNX2X_ERR("Unable to set real_num_queues\n");
2688 		LOAD_ERROR_EXIT(bp, load_error0);
2689 	}
2690 
2691 	/* configure multi cos mappings in kernel.
2692 	 * this configuration may be overridden by a multi class queue
2693 	 * discipline or by a dcbx negotiation result.
2694 	 */
2695 	bnx2x_setup_tc(bp->dev, bp->max_cos);
2696 
2697 	/* Add all NAPI objects */
2698 	bnx2x_add_all_napi(bp);
2699 	DP(NETIF_MSG_IFUP, "napi added\n");
2700 	bnx2x_napi_enable(bp);
2701 
2702 	if (IS_PF(bp)) {
2703 		/* set pf load just before approaching the MCP */
2704 		bnx2x_set_pf_load(bp);
2705 
2706 		/* if mcp exists send load request and analyze response */
2707 		if (!BP_NOMCP(bp)) {
2708 			/* attempt to load pf */
2709 			rc = bnx2x_nic_load_request(bp, &load_code);
2710 			if (rc)
2711 				LOAD_ERROR_EXIT(bp, load_error1);
2712 
2713 			/* what did mcp say? */
2714 			rc = bnx2x_compare_fw_ver(bp, load_code, true);
2715 			if (rc) {
2716 				bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2717 				LOAD_ERROR_EXIT(bp, load_error2);
2718 			}
2719 		} else {
2720 			load_code = bnx2x_nic_load_no_mcp(bp, port);
2721 		}
2722 
2723 		/* mark pmf if applicable */
2724 		bnx2x_nic_load_pmf(bp, load_code);
2725 
2726 		/* Init Function state controlling object */
2727 		bnx2x__init_func_obj(bp);
2728 
2729 		/* Initialize HW */
2730 		rc = bnx2x_init_hw(bp, load_code);
2731 		if (rc) {
2732 			BNX2X_ERR("HW init failed, aborting\n");
2733 			bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2734 			LOAD_ERROR_EXIT(bp, load_error2);
2735 		}
2736 	}
2737 
2738 	bnx2x_pre_irq_nic_init(bp);
2739 
2740 	/* Connect to IRQs */
2741 	rc = bnx2x_setup_irqs(bp);
2742 	if (rc) {
2743 		BNX2X_ERR("setup irqs failed\n");
2744 		if (IS_PF(bp))
2745 			bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2746 		LOAD_ERROR_EXIT(bp, load_error2);
2747 	}
2748 
2749 	/* Init per-function objects */
2750 	if (IS_PF(bp)) {
2751 		/* Setup NIC internals and enable interrupts */
2752 		bnx2x_post_irq_nic_init(bp, load_code);
2753 
2754 		bnx2x_init_bp_objs(bp);
2755 		bnx2x_iov_nic_init(bp);
2756 
2757 		/* Set AFEX default VLAN tag to an invalid value */
2758 		bp->afex_def_vlan_tag = -1;
2759 		bnx2x_nic_load_afex_dcc(bp, load_code);
2760 		bp->state = BNX2X_STATE_OPENING_WAIT4_PORT;
2761 		rc = bnx2x_func_start(bp);
2762 		if (rc) {
2763 			BNX2X_ERR("Function start failed!\n");
2764 			bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2765 
2766 			LOAD_ERROR_EXIT(bp, load_error3);
2767 		}
2768 
2769 		/* Send LOAD_DONE command to MCP */
2770 		if (!BP_NOMCP(bp)) {
2771 			load_code = bnx2x_fw_command(bp,
2772 						     DRV_MSG_CODE_LOAD_DONE, 0);
2773 			if (!load_code) {
2774 				BNX2X_ERR("MCP response failure, aborting\n");
2775 				rc = -EBUSY;
2776 				LOAD_ERROR_EXIT(bp, load_error3);
2777 			}
2778 		}
2779 
2780 		/* initialize FW coalescing state machines in RAM */
2781 		bnx2x_update_coalesce(bp);
2782 	}
2783 
2784 	/* setup the leading queue */
2785 	rc = bnx2x_setup_leading(bp);
2786 	if (rc) {
2787 		BNX2X_ERR("Setup leading failed!\n");
2788 		LOAD_ERROR_EXIT(bp, load_error3);
2789 	}
2790 
2791 	/* set up the rest of the queues */
2792 	for_each_nondefault_eth_queue(bp, i) {
2793 		if (IS_PF(bp))
2794 			rc = bnx2x_setup_queue(bp, &bp->fp[i], false);
2795 		else /* VF */
2796 			rc = bnx2x_vfpf_setup_q(bp, &bp->fp[i], false);
2797 		if (rc) {
2798 			BNX2X_ERR("Queue %d setup failed\n", i);
2799 			LOAD_ERROR_EXIT(bp, load_error3);
2800 		}
2801 	}
2802 
2803 	/* setup rss */
2804 	rc = bnx2x_init_rss(bp);
2805 	if (rc) {
2806 		BNX2X_ERR("PF RSS init failed\n");
2807 		LOAD_ERROR_EXIT(bp, load_error3);
2808 	}
2809 
2810 	/* Now when Clients are configured we are ready to work */
2811 	bp->state = BNX2X_STATE_OPEN;
2812 
2813 	/* Configure a ucast MAC */
2814 	if (IS_PF(bp))
2815 		rc = bnx2x_set_eth_mac(bp, true);
2816 	else /* vf */
2817 		rc = bnx2x_vfpf_config_mac(bp, bp->dev->dev_addr, bp->fp->index,
2818 					   true);
2819 	if (rc) {
2820 		BNX2X_ERR("Setting Ethernet MAC failed\n");
2821 		LOAD_ERROR_EXIT(bp, load_error3);
2822 	}
2823 
2824 	if (IS_PF(bp) && bp->pending_max) {
2825 		bnx2x_update_max_mf_config(bp, bp->pending_max);
2826 		bp->pending_max = 0;
2827 	}
2828 
2829 	if (bp->port.pmf) {
2830 		rc = bnx2x_initial_phy_init(bp, load_mode);
2831 		if (rc)
2832 			LOAD_ERROR_EXIT(bp, load_error3);
2833 	}
2834 	bp->link_params.feature_config_flags &= ~FEATURE_CONFIG_BOOT_FROM_SAN;
2835 
2836 	/* Start fast path */
2837 
2838 	/* Re-configure vlan filters */
2839 	rc = bnx2x_vlan_reconfigure_vid(bp);
2840 	if (rc)
2841 		LOAD_ERROR_EXIT(bp, load_error3);
2842 
2843 	/* Initialize Rx filter. */
2844 	bnx2x_set_rx_mode_inner(bp);
2845 
2846 	if (bp->flags & PTP_SUPPORTED) {
2847 		bnx2x_init_ptp(bp);
2848 		bnx2x_configure_ptp_filters(bp);
2849 	}
2850 	/* Start Tx */
2851 	switch (load_mode) {
2852 	case LOAD_NORMAL:
2853 		/* Tx queue should be only re-enabled */
2854 		netif_tx_wake_all_queues(bp->dev);
2855 		break;
2856 
2857 	case LOAD_OPEN:
2858 		netif_tx_start_all_queues(bp->dev);
2859 		smp_mb__after_atomic();
2860 		break;
2861 
2862 	case LOAD_DIAG:
2863 	case LOAD_LOOPBACK_EXT:
2864 		bp->state = BNX2X_STATE_DIAG;
2865 		break;
2866 
2867 	default:
2868 		break;
2869 	}
2870 
2871 	if (bp->port.pmf)
2872 		bnx2x_update_drv_flags(bp, 1 << DRV_FLAGS_PORT_MASK, 0);
2873 	else
2874 		bnx2x__link_status_update(bp);
2875 
2876 	/* start the timer */
2877 	mod_timer(&bp->timer, jiffies + bp->current_interval);
2878 
2879 	if (CNIC_ENABLED(bp))
2880 		bnx2x_load_cnic(bp);
2881 
2882 	if (IS_PF(bp))
2883 		bnx2x_schedule_sp_rtnl(bp, BNX2X_SP_RTNL_GET_DRV_VERSION, 0);
2884 
2885 	if (IS_PF(bp) && SHMEM2_HAS(bp, drv_capabilities_flag)) {
2886 		/* mark driver is loaded in shmem2 */
2887 		u32 val;
2888 		val = SHMEM2_RD(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)]);
2889 		val &= ~DRV_FLAGS_MTU_MASK;
2890 		val |= (bp->dev->mtu << DRV_FLAGS_MTU_SHIFT);
2891 		SHMEM2_WR(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)],
2892 			  val | DRV_FLAGS_CAPABILITIES_LOADED_SUPPORTED |
2893 			  DRV_FLAGS_CAPABILITIES_LOADED_L2);
2894 	}
2895 
2896 	/* Wait for all pending SP commands to complete */
2897 	if (IS_PF(bp) && !bnx2x_wait_sp_comp(bp, ~0x0UL)) {
2898 		BNX2X_ERR("Timeout waiting for SP elements to complete\n");
2899 		bnx2x_nic_unload(bp, UNLOAD_CLOSE, false);
2900 		return -EBUSY;
2901 	}
2902 
2903 	/* Update driver data for On-Chip MFW dump. */
2904 	if (IS_PF(bp))
2905 		bnx2x_update_mfw_dump(bp);
2906 
2907 	/* If PMF - send ADMIN DCBX msg to MFW to initiate DCBX FSM */
2908 	if (bp->port.pmf && (bp->state != BNX2X_STATE_DIAG))
2909 		bnx2x_dcbx_init(bp, false);
2910 
2911 	if (!IS_MF_SD_STORAGE_PERSONALITY_ONLY(bp))
2912 		bnx2x_set_os_driver_state(bp, OS_DRIVER_STATE_ACTIVE);
2913 
2914 	DP(NETIF_MSG_IFUP, "Ending successfully NIC load\n");
2915 
2916 	return 0;
2917 
2918 #ifndef BNX2X_STOP_ON_ERROR
2919 load_error3:
2920 	if (IS_PF(bp)) {
2921 		bnx2x_int_disable_sync(bp, 1);
2922 
2923 		/* Clean queueable objects */
2924 		bnx2x_squeeze_objects(bp);
2925 	}
2926 
2927 	/* Free SKBs, SGEs, TPA pool and driver internals */
2928 	bnx2x_free_skbs(bp);
2929 	for_each_rx_queue(bp, i)
2930 		bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
2931 
2932 	/* Release IRQs */
2933 	bnx2x_free_irq(bp);
2934 load_error2:
2935 	if (IS_PF(bp) && !BP_NOMCP(bp)) {
2936 		bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_REQ_WOL_MCP, 0);
2937 		bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_DONE, 0);
2938 	}
2939 
2940 	bp->port.pmf = 0;
2941 load_error1:
2942 	bnx2x_napi_disable(bp);
2943 	bnx2x_del_all_napi(bp);
2944 
2945 	/* clear pf_load status, as it was already set */
2946 	if (IS_PF(bp))
2947 		bnx2x_clear_pf_load(bp);
2948 load_error0:
2949 	bnx2x_free_fw_stats_mem(bp);
2950 	bnx2x_free_fp_mem(bp);
2951 	bnx2x_free_mem(bp);
2952 
2953 	return rc;
2954 #endif /* ! BNX2X_STOP_ON_ERROR */
2955 }
2956 
2957 int bnx2x_drain_tx_queues(struct bnx2x *bp)
2958 {
2959 	u8 rc = 0, cos, i;
2960 
2961 	/* Wait until tx fastpath tasks complete */
2962 	for_each_tx_queue(bp, i) {
2963 		struct bnx2x_fastpath *fp = &bp->fp[i];
2964 
2965 		for_each_cos_in_tx_queue(fp, cos)
2966 			rc = bnx2x_clean_tx_queue(bp, fp->txdata_ptr[cos]);
2967 		if (rc)
2968 			return rc;
2969 	}
2970 	return 0;
2971 }
2972 
2973 /* must be called with rtnl_lock */
2974 int bnx2x_nic_unload(struct bnx2x *bp, int unload_mode, bool keep_link)
2975 {
2976 	int i;
2977 	bool global = false;
2978 
2979 	DP(NETIF_MSG_IFUP, "Starting NIC unload\n");
2980 
2981 	if (!IS_MF_SD_STORAGE_PERSONALITY_ONLY(bp))
2982 		bnx2x_set_os_driver_state(bp, OS_DRIVER_STATE_DISABLED);
2983 
2984 	/* mark driver is unloaded in shmem2 */
2985 	if (IS_PF(bp) && SHMEM2_HAS(bp, drv_capabilities_flag)) {
2986 		u32 val;
2987 		val = SHMEM2_RD(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)]);
2988 		SHMEM2_WR(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)],
2989 			  val & ~DRV_FLAGS_CAPABILITIES_LOADED_L2);
2990 	}
2991 
2992 	if (IS_PF(bp) && bp->recovery_state != BNX2X_RECOVERY_DONE &&
2993 	    (bp->state == BNX2X_STATE_CLOSED ||
2994 	     bp->state == BNX2X_STATE_ERROR)) {
2995 		/* We can get here if the driver has been unloaded
2996 		 * during parity error recovery and is either waiting for a
2997 		 * leader to complete or for other functions to unload and
2998 		 * then ifdown has been issued. In this case we want to
2999 		 * unload and let other functions to complete a recovery
3000 		 * process.
3001 		 */
3002 		bp->recovery_state = BNX2X_RECOVERY_DONE;
3003 		bp->is_leader = 0;
3004 		bnx2x_release_leader_lock(bp);
3005 		smp_mb();
3006 
3007 		DP(NETIF_MSG_IFDOWN, "Releasing a leadership...\n");
3008 		BNX2X_ERR("Can't unload in closed or error state\n");
3009 		return -EINVAL;
3010 	}
3011 
3012 	/* Nothing to do during unload if previous bnx2x_nic_load()
3013 	 * have not completed successfully - all resources are released.
3014 	 *
3015 	 * we can get here only after unsuccessful ndo_* callback, during which
3016 	 * dev->IFF_UP flag is still on.
3017 	 */
3018 	if (bp->state == BNX2X_STATE_CLOSED || bp->state == BNX2X_STATE_ERROR)
3019 		return 0;
3020 
3021 	/* It's important to set the bp->state to the value different from
3022 	 * BNX2X_STATE_OPEN and only then stop the Tx. Otherwise bnx2x_tx_int()
3023 	 * may restart the Tx from the NAPI context (see bnx2x_tx_int()).
3024 	 */
3025 	bp->state = BNX2X_STATE_CLOSING_WAIT4_HALT;
3026 	smp_mb();
3027 
3028 	/* indicate to VFs that the PF is going down */
3029 	bnx2x_iov_channel_down(bp);
3030 
3031 	if (CNIC_LOADED(bp))
3032 		bnx2x_cnic_notify(bp, CNIC_CTL_STOP_CMD);
3033 
3034 	/* Stop Tx */
3035 	bnx2x_tx_disable(bp);
3036 	netdev_reset_tc(bp->dev);
3037 
3038 	bp->rx_mode = BNX2X_RX_MODE_NONE;
3039 
3040 	del_timer_sync(&bp->timer);
3041 
3042 	if (IS_PF(bp)) {
3043 		/* Set ALWAYS_ALIVE bit in shmem */
3044 		bp->fw_drv_pulse_wr_seq |= DRV_PULSE_ALWAYS_ALIVE;
3045 		bnx2x_drv_pulse(bp);
3046 		bnx2x_stats_handle(bp, STATS_EVENT_STOP);
3047 		bnx2x_save_statistics(bp);
3048 	}
3049 
3050 	/* wait till consumers catch up with producers in all queues */
3051 	bnx2x_drain_tx_queues(bp);
3052 
3053 	/* if VF indicate to PF this function is going down (PF will delete sp
3054 	 * elements and clear initializations
3055 	 */
3056 	if (IS_VF(bp))
3057 		bnx2x_vfpf_close_vf(bp);
3058 	else if (unload_mode != UNLOAD_RECOVERY)
3059 		/* if this is a normal/close unload need to clean up chip*/
3060 		bnx2x_chip_cleanup(bp, unload_mode, keep_link);
3061 	else {
3062 		/* Send the UNLOAD_REQUEST to the MCP */
3063 		bnx2x_send_unload_req(bp, unload_mode);
3064 
3065 		/* Prevent transactions to host from the functions on the
3066 		 * engine that doesn't reset global blocks in case of global
3067 		 * attention once global blocks are reset and gates are opened
3068 		 * (the engine which leader will perform the recovery
3069 		 * last).
3070 		 */
3071 		if (!CHIP_IS_E1x(bp))
3072 			bnx2x_pf_disable(bp);
3073 
3074 		/* Disable HW interrupts, NAPI */
3075 		bnx2x_netif_stop(bp, 1);
3076 		/* Delete all NAPI objects */
3077 		bnx2x_del_all_napi(bp);
3078 		if (CNIC_LOADED(bp))
3079 			bnx2x_del_all_napi_cnic(bp);
3080 		/* Release IRQs */
3081 		bnx2x_free_irq(bp);
3082 
3083 		/* Report UNLOAD_DONE to MCP */
3084 		bnx2x_send_unload_done(bp, false);
3085 	}
3086 
3087 	/*
3088 	 * At this stage no more interrupts will arrive so we may safely clean
3089 	 * the queueable objects here in case they failed to get cleaned so far.
3090 	 */
3091 	if (IS_PF(bp))
3092 		bnx2x_squeeze_objects(bp);
3093 
3094 	/* There should be no more pending SP commands at this stage */
3095 	bp->sp_state = 0;
3096 
3097 	bp->port.pmf = 0;
3098 
3099 	/* clear pending work in rtnl task */
3100 	bp->sp_rtnl_state = 0;
3101 	smp_mb();
3102 
3103 	/* Free SKBs, SGEs, TPA pool and driver internals */
3104 	bnx2x_free_skbs(bp);
3105 	if (CNIC_LOADED(bp))
3106 		bnx2x_free_skbs_cnic(bp);
3107 	for_each_rx_queue(bp, i)
3108 		bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
3109 
3110 	bnx2x_free_fp_mem(bp);
3111 	if (CNIC_LOADED(bp))
3112 		bnx2x_free_fp_mem_cnic(bp);
3113 
3114 	if (IS_PF(bp)) {
3115 		if (CNIC_LOADED(bp))
3116 			bnx2x_free_mem_cnic(bp);
3117 	}
3118 	bnx2x_free_mem(bp);
3119 
3120 	bp->state = BNX2X_STATE_CLOSED;
3121 	bp->cnic_loaded = false;
3122 
3123 	/* Clear driver version indication in shmem */
3124 	if (IS_PF(bp))
3125 		bnx2x_update_mng_version(bp);
3126 
3127 	/* Check if there are pending parity attentions. If there are - set
3128 	 * RECOVERY_IN_PROGRESS.
3129 	 */
3130 	if (IS_PF(bp) && bnx2x_chk_parity_attn(bp, &global, false)) {
3131 		bnx2x_set_reset_in_progress(bp);
3132 
3133 		/* Set RESET_IS_GLOBAL if needed */
3134 		if (global)
3135 			bnx2x_set_reset_global(bp);
3136 	}
3137 
3138 	/* The last driver must disable a "close the gate" if there is no
3139 	 * parity attention or "process kill" pending.
3140 	 */
3141 	if (IS_PF(bp) &&
3142 	    !bnx2x_clear_pf_load(bp) &&
3143 	    bnx2x_reset_is_done(bp, BP_PATH(bp)))
3144 		bnx2x_disable_close_the_gate(bp);
3145 
3146 	DP(NETIF_MSG_IFUP, "Ending NIC unload\n");
3147 
3148 	return 0;
3149 }
3150 
3151 int bnx2x_set_power_state(struct bnx2x *bp, pci_power_t state)
3152 {
3153 	u16 pmcsr;
3154 
3155 	/* If there is no power capability, silently succeed */
3156 	if (!bp->pdev->pm_cap) {
3157 		BNX2X_DEV_INFO("No power capability. Breaking.\n");
3158 		return 0;
3159 	}
3160 
3161 	pci_read_config_word(bp->pdev, bp->pdev->pm_cap + PCI_PM_CTRL, &pmcsr);
3162 
3163 	switch (state) {
3164 	case PCI_D0:
3165 		pci_write_config_word(bp->pdev, bp->pdev->pm_cap + PCI_PM_CTRL,
3166 				      ((pmcsr & ~PCI_PM_CTRL_STATE_MASK) |
3167 				       PCI_PM_CTRL_PME_STATUS));
3168 
3169 		if (pmcsr & PCI_PM_CTRL_STATE_MASK)
3170 			/* delay required during transition out of D3hot */
3171 			msleep(20);
3172 		break;
3173 
3174 	case PCI_D3hot:
3175 		/* If there are other clients above don't
3176 		   shut down the power */
3177 		if (atomic_read(&bp->pdev->enable_cnt) != 1)
3178 			return 0;
3179 		/* Don't shut down the power for emulation and FPGA */
3180 		if (CHIP_REV_IS_SLOW(bp))
3181 			return 0;
3182 
3183 		pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
3184 		pmcsr |= 3;
3185 
3186 		if (bp->wol)
3187 			pmcsr |= PCI_PM_CTRL_PME_ENABLE;
3188 
3189 		pci_write_config_word(bp->pdev, bp->pdev->pm_cap + PCI_PM_CTRL,
3190 				      pmcsr);
3191 
3192 		/* No more memory access after this point until
3193 		* device is brought back to D0.
3194 		*/
3195 		break;
3196 
3197 	default:
3198 		dev_err(&bp->pdev->dev, "Can't support state = %d\n", state);
3199 		return -EINVAL;
3200 	}
3201 	return 0;
3202 }
3203 
3204 /*
3205  * net_device service functions
3206  */
3207 static int bnx2x_poll(struct napi_struct *napi, int budget)
3208 {
3209 	int work_done = 0;
3210 	u8 cos;
3211 	struct bnx2x_fastpath *fp = container_of(napi, struct bnx2x_fastpath,
3212 						 napi);
3213 	struct bnx2x *bp = fp->bp;
3214 
3215 	while (1) {
3216 #ifdef BNX2X_STOP_ON_ERROR
3217 		if (unlikely(bp->panic)) {
3218 			napi_complete(napi);
3219 			return 0;
3220 		}
3221 #endif
3222 		for_each_cos_in_tx_queue(fp, cos)
3223 			if (bnx2x_tx_queue_has_work(fp->txdata_ptr[cos]))
3224 				bnx2x_tx_int(bp, fp->txdata_ptr[cos]);
3225 
3226 		if (bnx2x_has_rx_work(fp)) {
3227 			work_done += bnx2x_rx_int(fp, budget - work_done);
3228 
3229 			/* must not complete if we consumed full budget */
3230 			if (work_done >= budget)
3231 				break;
3232 		}
3233 
3234 		/* Fall out from the NAPI loop if needed */
3235 		if (!(bnx2x_has_rx_work(fp) || bnx2x_has_tx_work(fp))) {
3236 
3237 			/* No need to update SB for FCoE L2 ring as long as
3238 			 * it's connected to the default SB and the SB
3239 			 * has been updated when NAPI was scheduled.
3240 			 */
3241 			if (IS_FCOE_FP(fp)) {
3242 				napi_complete(napi);
3243 				break;
3244 			}
3245 			bnx2x_update_fpsb_idx(fp);
3246 			/* bnx2x_has_rx_work() reads the status block,
3247 			 * thus we need to ensure that status block indices
3248 			 * have been actually read (bnx2x_update_fpsb_idx)
3249 			 * prior to this check (bnx2x_has_rx_work) so that
3250 			 * we won't write the "newer" value of the status block
3251 			 * to IGU (if there was a DMA right after
3252 			 * bnx2x_has_rx_work and if there is no rmb, the memory
3253 			 * reading (bnx2x_update_fpsb_idx) may be postponed
3254 			 * to right before bnx2x_ack_sb). In this case there
3255 			 * will never be another interrupt until there is
3256 			 * another update of the status block, while there
3257 			 * is still unhandled work.
3258 			 */
3259 			rmb();
3260 
3261 			if (!(bnx2x_has_rx_work(fp) || bnx2x_has_tx_work(fp))) {
3262 				napi_complete(napi);
3263 				/* Re-enable interrupts */
3264 				DP(NETIF_MSG_RX_STATUS,
3265 				   "Update index to %d\n", fp->fp_hc_idx);
3266 				bnx2x_ack_sb(bp, fp->igu_sb_id, USTORM_ID,
3267 					     le16_to_cpu(fp->fp_hc_idx),
3268 					     IGU_INT_ENABLE, 1);
3269 				break;
3270 			}
3271 		}
3272 	}
3273 
3274 	return work_done;
3275 }
3276 
3277 /* we split the first BD into headers and data BDs
3278  * to ease the pain of our fellow microcode engineers
3279  * we use one mapping for both BDs
3280  */
3281 static u16 bnx2x_tx_split(struct bnx2x *bp,
3282 			  struct bnx2x_fp_txdata *txdata,
3283 			  struct sw_tx_bd *tx_buf,
3284 			  struct eth_tx_start_bd **tx_bd, u16 hlen,
3285 			  u16 bd_prod)
3286 {
3287 	struct eth_tx_start_bd *h_tx_bd = *tx_bd;
3288 	struct eth_tx_bd *d_tx_bd;
3289 	dma_addr_t mapping;
3290 	int old_len = le16_to_cpu(h_tx_bd->nbytes);
3291 
3292 	/* first fix first BD */
3293 	h_tx_bd->nbytes = cpu_to_le16(hlen);
3294 
3295 	DP(NETIF_MSG_TX_QUEUED,	"TSO split header size is %d (%x:%x)\n",
3296 	   h_tx_bd->nbytes, h_tx_bd->addr_hi, h_tx_bd->addr_lo);
3297 
3298 	/* now get a new data BD
3299 	 * (after the pbd) and fill it */
3300 	bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3301 	d_tx_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
3302 
3303 	mapping = HILO_U64(le32_to_cpu(h_tx_bd->addr_hi),
3304 			   le32_to_cpu(h_tx_bd->addr_lo)) + hlen;
3305 
3306 	d_tx_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
3307 	d_tx_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
3308 	d_tx_bd->nbytes = cpu_to_le16(old_len - hlen);
3309 
3310 	/* this marks the BD as one that has no individual mapping */
3311 	tx_buf->flags |= BNX2X_TSO_SPLIT_BD;
3312 
3313 	DP(NETIF_MSG_TX_QUEUED,
3314 	   "TSO split data size is %d (%x:%x)\n",
3315 	   d_tx_bd->nbytes, d_tx_bd->addr_hi, d_tx_bd->addr_lo);
3316 
3317 	/* update tx_bd */
3318 	*tx_bd = (struct eth_tx_start_bd *)d_tx_bd;
3319 
3320 	return bd_prod;
3321 }
3322 
3323 #define bswab32(b32) ((__force __le32) swab32((__force __u32) (b32)))
3324 #define bswab16(b16) ((__force __le16) swab16((__force __u16) (b16)))
3325 static __le16 bnx2x_csum_fix(unsigned char *t_header, u16 csum, s8 fix)
3326 {
3327 	__sum16 tsum = (__force __sum16) csum;
3328 
3329 	if (fix > 0)
3330 		tsum = ~csum_fold(csum_sub((__force __wsum) csum,
3331 				  csum_partial(t_header - fix, fix, 0)));
3332 
3333 	else if (fix < 0)
3334 		tsum = ~csum_fold(csum_add((__force __wsum) csum,
3335 				  csum_partial(t_header, -fix, 0)));
3336 
3337 	return bswab16(tsum);
3338 }
3339 
3340 static u32 bnx2x_xmit_type(struct bnx2x *bp, struct sk_buff *skb)
3341 {
3342 	u32 rc;
3343 	__u8 prot = 0;
3344 	__be16 protocol;
3345 
3346 	if (skb->ip_summed != CHECKSUM_PARTIAL)
3347 		return XMIT_PLAIN;
3348 
3349 	protocol = vlan_get_protocol(skb);
3350 	if (protocol == htons(ETH_P_IPV6)) {
3351 		rc = XMIT_CSUM_V6;
3352 		prot = ipv6_hdr(skb)->nexthdr;
3353 	} else {
3354 		rc = XMIT_CSUM_V4;
3355 		prot = ip_hdr(skb)->protocol;
3356 	}
3357 
3358 	if (!CHIP_IS_E1x(bp) && skb->encapsulation) {
3359 		if (inner_ip_hdr(skb)->version == 6) {
3360 			rc |= XMIT_CSUM_ENC_V6;
3361 			if (inner_ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3362 				rc |= XMIT_CSUM_TCP;
3363 		} else {
3364 			rc |= XMIT_CSUM_ENC_V4;
3365 			if (inner_ip_hdr(skb)->protocol == IPPROTO_TCP)
3366 				rc |= XMIT_CSUM_TCP;
3367 		}
3368 	}
3369 	if (prot == IPPROTO_TCP)
3370 		rc |= XMIT_CSUM_TCP;
3371 
3372 	if (skb_is_gso(skb)) {
3373 		if (skb_is_gso_v6(skb)) {
3374 			rc |= (XMIT_GSO_V6 | XMIT_CSUM_TCP);
3375 			if (rc & XMIT_CSUM_ENC)
3376 				rc |= XMIT_GSO_ENC_V6;
3377 		} else {
3378 			rc |= (XMIT_GSO_V4 | XMIT_CSUM_TCP);
3379 			if (rc & XMIT_CSUM_ENC)
3380 				rc |= XMIT_GSO_ENC_V4;
3381 		}
3382 	}
3383 
3384 	return rc;
3385 }
3386 
3387 #if (MAX_SKB_FRAGS >= MAX_FETCH_BD - 3)
3388 /* check if packet requires linearization (packet is too fragmented)
3389    no need to check fragmentation if page size > 8K (there will be no
3390    violation to FW restrictions) */
3391 static int bnx2x_pkt_req_lin(struct bnx2x *bp, struct sk_buff *skb,
3392 			     u32 xmit_type)
3393 {
3394 	int to_copy = 0;
3395 	int hlen = 0;
3396 	int first_bd_sz = 0;
3397 
3398 	/* 3 = 1 (for linear data BD) + 2 (for PBD and last BD) */
3399 	if (skb_shinfo(skb)->nr_frags >= (MAX_FETCH_BD - 3)) {
3400 
3401 		if (xmit_type & XMIT_GSO) {
3402 			unsigned short lso_mss = skb_shinfo(skb)->gso_size;
3403 			/* Check if LSO packet needs to be copied:
3404 			   3 = 1 (for headers BD) + 2 (for PBD and last BD) */
3405 			int wnd_size = MAX_FETCH_BD - 3;
3406 			/* Number of windows to check */
3407 			int num_wnds = skb_shinfo(skb)->nr_frags - wnd_size;
3408 			int wnd_idx = 0;
3409 			int frag_idx = 0;
3410 			u32 wnd_sum = 0;
3411 
3412 			/* Headers length */
3413 			if (xmit_type & XMIT_GSO_ENC)
3414 				hlen = (int)(skb_inner_transport_header(skb) -
3415 					     skb->data) +
3416 					     inner_tcp_hdrlen(skb);
3417 			else
3418 				hlen = (int)(skb_transport_header(skb) -
3419 					     skb->data) + tcp_hdrlen(skb);
3420 
3421 			/* Amount of data (w/o headers) on linear part of SKB*/
3422 			first_bd_sz = skb_headlen(skb) - hlen;
3423 
3424 			wnd_sum  = first_bd_sz;
3425 
3426 			/* Calculate the first sum - it's special */
3427 			for (frag_idx = 0; frag_idx < wnd_size - 1; frag_idx++)
3428 				wnd_sum +=
3429 					skb_frag_size(&skb_shinfo(skb)->frags[frag_idx]);
3430 
3431 			/* If there was data on linear skb data - check it */
3432 			if (first_bd_sz > 0) {
3433 				if (unlikely(wnd_sum < lso_mss)) {
3434 					to_copy = 1;
3435 					goto exit_lbl;
3436 				}
3437 
3438 				wnd_sum -= first_bd_sz;
3439 			}
3440 
3441 			/* Others are easier: run through the frag list and
3442 			   check all windows */
3443 			for (wnd_idx = 0; wnd_idx <= num_wnds; wnd_idx++) {
3444 				wnd_sum +=
3445 			  skb_frag_size(&skb_shinfo(skb)->frags[wnd_idx + wnd_size - 1]);
3446 
3447 				if (unlikely(wnd_sum < lso_mss)) {
3448 					to_copy = 1;
3449 					break;
3450 				}
3451 				wnd_sum -=
3452 					skb_frag_size(&skb_shinfo(skb)->frags[wnd_idx]);
3453 			}
3454 		} else {
3455 			/* in non-LSO too fragmented packet should always
3456 			   be linearized */
3457 			to_copy = 1;
3458 		}
3459 	}
3460 
3461 exit_lbl:
3462 	if (unlikely(to_copy))
3463 		DP(NETIF_MSG_TX_QUEUED,
3464 		   "Linearization IS REQUIRED for %s packet. num_frags %d  hlen %d  first_bd_sz %d\n",
3465 		   (xmit_type & XMIT_GSO) ? "LSO" : "non-LSO",
3466 		   skb_shinfo(skb)->nr_frags, hlen, first_bd_sz);
3467 
3468 	return to_copy;
3469 }
3470 #endif
3471 
3472 /**
3473  * bnx2x_set_pbd_gso - update PBD in GSO case.
3474  *
3475  * @skb:	packet skb
3476  * @pbd:	parse BD
3477  * @xmit_type:	xmit flags
3478  */
3479 static void bnx2x_set_pbd_gso(struct sk_buff *skb,
3480 			      struct eth_tx_parse_bd_e1x *pbd,
3481 			      u32 xmit_type)
3482 {
3483 	pbd->lso_mss = cpu_to_le16(skb_shinfo(skb)->gso_size);
3484 	pbd->tcp_send_seq = bswab32(tcp_hdr(skb)->seq);
3485 	pbd->tcp_flags = pbd_tcp_flags(tcp_hdr(skb));
3486 
3487 	if (xmit_type & XMIT_GSO_V4) {
3488 		pbd->ip_id = bswab16(ip_hdr(skb)->id);
3489 		pbd->tcp_pseudo_csum =
3490 			bswab16(~csum_tcpudp_magic(ip_hdr(skb)->saddr,
3491 						   ip_hdr(skb)->daddr,
3492 						   0, IPPROTO_TCP, 0));
3493 	} else {
3494 		pbd->tcp_pseudo_csum =
3495 			bswab16(~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3496 						 &ipv6_hdr(skb)->daddr,
3497 						 0, IPPROTO_TCP, 0));
3498 	}
3499 
3500 	pbd->global_data |=
3501 		cpu_to_le16(ETH_TX_PARSE_BD_E1X_PSEUDO_CS_WITHOUT_LEN);
3502 }
3503 
3504 /**
3505  * bnx2x_set_pbd_csum_enc - update PBD with checksum and return header length
3506  *
3507  * @bp:			driver handle
3508  * @skb:		packet skb
3509  * @parsing_data:	data to be updated
3510  * @xmit_type:		xmit flags
3511  *
3512  * 57712/578xx related, when skb has encapsulation
3513  */
3514 static u8 bnx2x_set_pbd_csum_enc(struct bnx2x *bp, struct sk_buff *skb,
3515 				 u32 *parsing_data, u32 xmit_type)
3516 {
3517 	*parsing_data |=
3518 		((((u8 *)skb_inner_transport_header(skb) - skb->data) >> 1) <<
3519 		ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W_SHIFT) &
3520 		ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W;
3521 
3522 	if (xmit_type & XMIT_CSUM_TCP) {
3523 		*parsing_data |= ((inner_tcp_hdrlen(skb) / 4) <<
3524 			ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW_SHIFT) &
3525 			ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW;
3526 
3527 		return skb_inner_transport_header(skb) +
3528 			inner_tcp_hdrlen(skb) - skb->data;
3529 	}
3530 
3531 	/* We support checksum offload for TCP and UDP only.
3532 	 * No need to pass the UDP header length - it's a constant.
3533 	 */
3534 	return skb_inner_transport_header(skb) +
3535 		sizeof(struct udphdr) - skb->data;
3536 }
3537 
3538 /**
3539  * bnx2x_set_pbd_csum_e2 - update PBD with checksum and return header length
3540  *
3541  * @bp:			driver handle
3542  * @skb:		packet skb
3543  * @parsing_data:	data to be updated
3544  * @xmit_type:		xmit flags
3545  *
3546  * 57712/578xx related
3547  */
3548 static u8 bnx2x_set_pbd_csum_e2(struct bnx2x *bp, struct sk_buff *skb,
3549 				u32 *parsing_data, u32 xmit_type)
3550 {
3551 	*parsing_data |=
3552 		((((u8 *)skb_transport_header(skb) - skb->data) >> 1) <<
3553 		ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W_SHIFT) &
3554 		ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W;
3555 
3556 	if (xmit_type & XMIT_CSUM_TCP) {
3557 		*parsing_data |= ((tcp_hdrlen(skb) / 4) <<
3558 			ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW_SHIFT) &
3559 			ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW;
3560 
3561 		return skb_transport_header(skb) + tcp_hdrlen(skb) - skb->data;
3562 	}
3563 	/* We support checksum offload for TCP and UDP only.
3564 	 * No need to pass the UDP header length - it's a constant.
3565 	 */
3566 	return skb_transport_header(skb) + sizeof(struct udphdr) - skb->data;
3567 }
3568 
3569 /* set FW indication according to inner or outer protocols if tunneled */
3570 static void bnx2x_set_sbd_csum(struct bnx2x *bp, struct sk_buff *skb,
3571 			       struct eth_tx_start_bd *tx_start_bd,
3572 			       u32 xmit_type)
3573 {
3574 	tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_L4_CSUM;
3575 
3576 	if (xmit_type & (XMIT_CSUM_ENC_V6 | XMIT_CSUM_V6))
3577 		tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_IPV6;
3578 
3579 	if (!(xmit_type & XMIT_CSUM_TCP))
3580 		tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_IS_UDP;
3581 }
3582 
3583 /**
3584  * bnx2x_set_pbd_csum - update PBD with checksum and return header length
3585  *
3586  * @bp:		driver handle
3587  * @skb:	packet skb
3588  * @pbd:	parse BD to be updated
3589  * @xmit_type:	xmit flags
3590  */
3591 static u8 bnx2x_set_pbd_csum(struct bnx2x *bp, struct sk_buff *skb,
3592 			     struct eth_tx_parse_bd_e1x *pbd,
3593 			     u32 xmit_type)
3594 {
3595 	u8 hlen = (skb_network_header(skb) - skb->data) >> 1;
3596 
3597 	/* for now NS flag is not used in Linux */
3598 	pbd->global_data =
3599 		cpu_to_le16(hlen |
3600 			    ((skb->protocol == cpu_to_be16(ETH_P_8021Q)) <<
3601 			     ETH_TX_PARSE_BD_E1X_LLC_SNAP_EN_SHIFT));
3602 
3603 	pbd->ip_hlen_w = (skb_transport_header(skb) -
3604 			skb_network_header(skb)) >> 1;
3605 
3606 	hlen += pbd->ip_hlen_w;
3607 
3608 	/* We support checksum offload for TCP and UDP only */
3609 	if (xmit_type & XMIT_CSUM_TCP)
3610 		hlen += tcp_hdrlen(skb) / 2;
3611 	else
3612 		hlen += sizeof(struct udphdr) / 2;
3613 
3614 	pbd->total_hlen_w = cpu_to_le16(hlen);
3615 	hlen = hlen*2;
3616 
3617 	if (xmit_type & XMIT_CSUM_TCP) {
3618 		pbd->tcp_pseudo_csum = bswab16(tcp_hdr(skb)->check);
3619 
3620 	} else {
3621 		s8 fix = SKB_CS_OFF(skb); /* signed! */
3622 
3623 		DP(NETIF_MSG_TX_QUEUED,
3624 		   "hlen %d  fix %d  csum before fix %x\n",
3625 		   le16_to_cpu(pbd->total_hlen_w), fix, SKB_CS(skb));
3626 
3627 		/* HW bug: fixup the CSUM */
3628 		pbd->tcp_pseudo_csum =
3629 			bnx2x_csum_fix(skb_transport_header(skb),
3630 				       SKB_CS(skb), fix);
3631 
3632 		DP(NETIF_MSG_TX_QUEUED, "csum after fix %x\n",
3633 		   pbd->tcp_pseudo_csum);
3634 	}
3635 
3636 	return hlen;
3637 }
3638 
3639 static void bnx2x_update_pbds_gso_enc(struct sk_buff *skb,
3640 				      struct eth_tx_parse_bd_e2 *pbd_e2,
3641 				      struct eth_tx_parse_2nd_bd *pbd2,
3642 				      u16 *global_data,
3643 				      u32 xmit_type)
3644 {
3645 	u16 hlen_w = 0;
3646 	u8 outerip_off, outerip_len = 0;
3647 
3648 	/* from outer IP to transport */
3649 	hlen_w = (skb_inner_transport_header(skb) -
3650 		  skb_network_header(skb)) >> 1;
3651 
3652 	/* transport len */
3653 	hlen_w += inner_tcp_hdrlen(skb) >> 1;
3654 
3655 	pbd2->fw_ip_hdr_to_payload_w = hlen_w;
3656 
3657 	/* outer IP header info */
3658 	if (xmit_type & XMIT_CSUM_V4) {
3659 		struct iphdr *iph = ip_hdr(skb);
3660 		u32 csum = (__force u32)(~iph->check) -
3661 			   (__force u32)iph->tot_len -
3662 			   (__force u32)iph->frag_off;
3663 
3664 		outerip_len = iph->ihl << 1;
3665 
3666 		pbd2->fw_ip_csum_wo_len_flags_frag =
3667 			bswab16(csum_fold((__force __wsum)csum));
3668 	} else {
3669 		pbd2->fw_ip_hdr_to_payload_w =
3670 			hlen_w - ((sizeof(struct ipv6hdr)) >> 1);
3671 		pbd_e2->data.tunnel_data.flags |=
3672 			ETH_TUNNEL_DATA_IPV6_OUTER;
3673 	}
3674 
3675 	pbd2->tcp_send_seq = bswab32(inner_tcp_hdr(skb)->seq);
3676 
3677 	pbd2->tcp_flags = pbd_tcp_flags(inner_tcp_hdr(skb));
3678 
3679 	/* inner IP header info */
3680 	if (xmit_type & XMIT_CSUM_ENC_V4) {
3681 		pbd2->hw_ip_id = bswab16(inner_ip_hdr(skb)->id);
3682 
3683 		pbd_e2->data.tunnel_data.pseudo_csum =
3684 			bswab16(~csum_tcpudp_magic(
3685 					inner_ip_hdr(skb)->saddr,
3686 					inner_ip_hdr(skb)->daddr,
3687 					0, IPPROTO_TCP, 0));
3688 	} else {
3689 		pbd_e2->data.tunnel_data.pseudo_csum =
3690 			bswab16(~csum_ipv6_magic(
3691 					&inner_ipv6_hdr(skb)->saddr,
3692 					&inner_ipv6_hdr(skb)->daddr,
3693 					0, IPPROTO_TCP, 0));
3694 	}
3695 
3696 	outerip_off = (skb_network_header(skb) - skb->data) >> 1;
3697 
3698 	*global_data |=
3699 		outerip_off |
3700 		(outerip_len <<
3701 			ETH_TX_PARSE_2ND_BD_IP_HDR_LEN_OUTER_W_SHIFT) |
3702 		((skb->protocol == cpu_to_be16(ETH_P_8021Q)) <<
3703 			ETH_TX_PARSE_2ND_BD_LLC_SNAP_EN_SHIFT);
3704 
3705 	if (ip_hdr(skb)->protocol == IPPROTO_UDP) {
3706 		SET_FLAG(*global_data, ETH_TX_PARSE_2ND_BD_TUNNEL_UDP_EXIST, 1);
3707 		pbd2->tunnel_udp_hdr_start_w = skb_transport_offset(skb) >> 1;
3708 	}
3709 }
3710 
3711 static inline void bnx2x_set_ipv6_ext_e2(struct sk_buff *skb, u32 *parsing_data,
3712 					 u32 xmit_type)
3713 {
3714 	struct ipv6hdr *ipv6;
3715 
3716 	if (!(xmit_type & (XMIT_GSO_ENC_V6 | XMIT_GSO_V6)))
3717 		return;
3718 
3719 	if (xmit_type & XMIT_GSO_ENC_V6)
3720 		ipv6 = inner_ipv6_hdr(skb);
3721 	else /* XMIT_GSO_V6 */
3722 		ipv6 = ipv6_hdr(skb);
3723 
3724 	if (ipv6->nexthdr == NEXTHDR_IPV6)
3725 		*parsing_data |= ETH_TX_PARSE_BD_E2_IPV6_WITH_EXT_HDR;
3726 }
3727 
3728 /* called with netif_tx_lock
3729  * bnx2x_tx_int() runs without netif_tx_lock unless it needs to call
3730  * netif_wake_queue()
3731  */
3732 netdev_tx_t bnx2x_start_xmit(struct sk_buff *skb, struct net_device *dev)
3733 {
3734 	struct bnx2x *bp = netdev_priv(dev);
3735 
3736 	struct netdev_queue *txq;
3737 	struct bnx2x_fp_txdata *txdata;
3738 	struct sw_tx_bd *tx_buf;
3739 	struct eth_tx_start_bd *tx_start_bd, *first_bd;
3740 	struct eth_tx_bd *tx_data_bd, *total_pkt_bd = NULL;
3741 	struct eth_tx_parse_bd_e1x *pbd_e1x = NULL;
3742 	struct eth_tx_parse_bd_e2 *pbd_e2 = NULL;
3743 	struct eth_tx_parse_2nd_bd *pbd2 = NULL;
3744 	u32 pbd_e2_parsing_data = 0;
3745 	u16 pkt_prod, bd_prod;
3746 	int nbd, txq_index;
3747 	dma_addr_t mapping;
3748 	u32 xmit_type = bnx2x_xmit_type(bp, skb);
3749 	int i;
3750 	u8 hlen = 0;
3751 	__le16 pkt_size = 0;
3752 	struct ethhdr *eth;
3753 	u8 mac_type = UNICAST_ADDRESS;
3754 
3755 #ifdef BNX2X_STOP_ON_ERROR
3756 	if (unlikely(bp->panic))
3757 		return NETDEV_TX_BUSY;
3758 #endif
3759 
3760 	txq_index = skb_get_queue_mapping(skb);
3761 	txq = netdev_get_tx_queue(dev, txq_index);
3762 
3763 	BUG_ON(txq_index >= MAX_ETH_TXQ_IDX(bp) + (CNIC_LOADED(bp) ? 1 : 0));
3764 
3765 	txdata = &bp->bnx2x_txq[txq_index];
3766 
3767 	/* enable this debug print to view the transmission queue being used
3768 	DP(NETIF_MSG_TX_QUEUED, "indices: txq %d, fp %d, txdata %d\n",
3769 	   txq_index, fp_index, txdata_index); */
3770 
3771 	/* enable this debug print to view the transmission details
3772 	DP(NETIF_MSG_TX_QUEUED,
3773 	   "transmitting packet cid %d fp index %d txdata_index %d tx_data ptr %p fp pointer %p\n",
3774 	   txdata->cid, fp_index, txdata_index, txdata, fp); */
3775 
3776 	if (unlikely(bnx2x_tx_avail(bp, txdata) <
3777 			skb_shinfo(skb)->nr_frags +
3778 			BDS_PER_TX_PKT +
3779 			NEXT_CNT_PER_TX_PKT(MAX_BDS_PER_TX_PKT))) {
3780 		/* Handle special storage cases separately */
3781 		if (txdata->tx_ring_size == 0) {
3782 			struct bnx2x_eth_q_stats *q_stats =
3783 				bnx2x_fp_qstats(bp, txdata->parent_fp);
3784 			q_stats->driver_filtered_tx_pkt++;
3785 			dev_kfree_skb(skb);
3786 			return NETDEV_TX_OK;
3787 		}
3788 		bnx2x_fp_qstats(bp, txdata->parent_fp)->driver_xoff++;
3789 		netif_tx_stop_queue(txq);
3790 		BNX2X_ERR("BUG! Tx ring full when queue awake!\n");
3791 
3792 		return NETDEV_TX_BUSY;
3793 	}
3794 
3795 	DP(NETIF_MSG_TX_QUEUED,
3796 	   "queue[%d]: SKB: summed %x  protocol %x protocol(%x,%x) gso type %x  xmit_type %x len %d\n",
3797 	   txq_index, skb->ip_summed, skb->protocol, ipv6_hdr(skb)->nexthdr,
3798 	   ip_hdr(skb)->protocol, skb_shinfo(skb)->gso_type, xmit_type,
3799 	   skb->len);
3800 
3801 	eth = (struct ethhdr *)skb->data;
3802 
3803 	/* set flag according to packet type (UNICAST_ADDRESS is default)*/
3804 	if (unlikely(is_multicast_ether_addr(eth->h_dest))) {
3805 		if (is_broadcast_ether_addr(eth->h_dest))
3806 			mac_type = BROADCAST_ADDRESS;
3807 		else
3808 			mac_type = MULTICAST_ADDRESS;
3809 	}
3810 
3811 #if (MAX_SKB_FRAGS >= MAX_FETCH_BD - BDS_PER_TX_PKT)
3812 	/* First, check if we need to linearize the skb (due to FW
3813 	   restrictions). No need to check fragmentation if page size > 8K
3814 	   (there will be no violation to FW restrictions) */
3815 	if (bnx2x_pkt_req_lin(bp, skb, xmit_type)) {
3816 		/* Statistics of linearization */
3817 		bp->lin_cnt++;
3818 		if (skb_linearize(skb) != 0) {
3819 			DP(NETIF_MSG_TX_QUEUED,
3820 			   "SKB linearization failed - silently dropping this SKB\n");
3821 			dev_kfree_skb_any(skb);
3822 			return NETDEV_TX_OK;
3823 		}
3824 	}
3825 #endif
3826 	/* Map skb linear data for DMA */
3827 	mapping = dma_map_single(&bp->pdev->dev, skb->data,
3828 				 skb_headlen(skb), DMA_TO_DEVICE);
3829 	if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
3830 		DP(NETIF_MSG_TX_QUEUED,
3831 		   "SKB mapping failed - silently dropping this SKB\n");
3832 		dev_kfree_skb_any(skb);
3833 		return NETDEV_TX_OK;
3834 	}
3835 	/*
3836 	Please read carefully. First we use one BD which we mark as start,
3837 	then we have a parsing info BD (used for TSO or xsum),
3838 	and only then we have the rest of the TSO BDs.
3839 	(don't forget to mark the last one as last,
3840 	and to unmap only AFTER you write to the BD ...)
3841 	And above all, all pdb sizes are in words - NOT DWORDS!
3842 	*/
3843 
3844 	/* get current pkt produced now - advance it just before sending packet
3845 	 * since mapping of pages may fail and cause packet to be dropped
3846 	 */
3847 	pkt_prod = txdata->tx_pkt_prod;
3848 	bd_prod = TX_BD(txdata->tx_bd_prod);
3849 
3850 	/* get a tx_buf and first BD
3851 	 * tx_start_bd may be changed during SPLIT,
3852 	 * but first_bd will always stay first
3853 	 */
3854 	tx_buf = &txdata->tx_buf_ring[TX_BD(pkt_prod)];
3855 	tx_start_bd = &txdata->tx_desc_ring[bd_prod].start_bd;
3856 	first_bd = tx_start_bd;
3857 
3858 	tx_start_bd->bd_flags.as_bitfield = ETH_TX_BD_FLAGS_START_BD;
3859 
3860 	if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) {
3861 		if (!(bp->flags & TX_TIMESTAMPING_EN)) {
3862 			BNX2X_ERR("Tx timestamping was not enabled, this packet will not be timestamped\n");
3863 		} else if (bp->ptp_tx_skb) {
3864 			BNX2X_ERR("The device supports only a single outstanding packet to timestamp, this packet will not be timestamped\n");
3865 		} else {
3866 			skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
3867 			/* schedule check for Tx timestamp */
3868 			bp->ptp_tx_skb = skb_get(skb);
3869 			bp->ptp_tx_start = jiffies;
3870 			schedule_work(&bp->ptp_task);
3871 		}
3872 	}
3873 
3874 	/* header nbd: indirectly zero other flags! */
3875 	tx_start_bd->general_data = 1 << ETH_TX_START_BD_HDR_NBDS_SHIFT;
3876 
3877 	/* remember the first BD of the packet */
3878 	tx_buf->first_bd = txdata->tx_bd_prod;
3879 	tx_buf->skb = skb;
3880 	tx_buf->flags = 0;
3881 
3882 	DP(NETIF_MSG_TX_QUEUED,
3883 	   "sending pkt %u @%p  next_idx %u  bd %u @%p\n",
3884 	   pkt_prod, tx_buf, txdata->tx_pkt_prod, bd_prod, tx_start_bd);
3885 
3886 	if (skb_vlan_tag_present(skb)) {
3887 		tx_start_bd->vlan_or_ethertype =
3888 		    cpu_to_le16(skb_vlan_tag_get(skb));
3889 		tx_start_bd->bd_flags.as_bitfield |=
3890 		    (X_ETH_OUTBAND_VLAN << ETH_TX_BD_FLAGS_VLAN_MODE_SHIFT);
3891 	} else {
3892 		/* when transmitting in a vf, start bd must hold the ethertype
3893 		 * for fw to enforce it
3894 		 */
3895 #ifndef BNX2X_STOP_ON_ERROR
3896 		if (IS_VF(bp))
3897 #endif
3898 			tx_start_bd->vlan_or_ethertype =
3899 				cpu_to_le16(ntohs(eth->h_proto));
3900 #ifndef BNX2X_STOP_ON_ERROR
3901 		else
3902 			/* used by FW for packet accounting */
3903 			tx_start_bd->vlan_or_ethertype = cpu_to_le16(pkt_prod);
3904 #endif
3905 	}
3906 
3907 	nbd = 2; /* start_bd + pbd + frags (updated when pages are mapped) */
3908 
3909 	/* turn on parsing and get a BD */
3910 	bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3911 
3912 	if (xmit_type & XMIT_CSUM)
3913 		bnx2x_set_sbd_csum(bp, skb, tx_start_bd, xmit_type);
3914 
3915 	if (!CHIP_IS_E1x(bp)) {
3916 		pbd_e2 = &txdata->tx_desc_ring[bd_prod].parse_bd_e2;
3917 		memset(pbd_e2, 0, sizeof(struct eth_tx_parse_bd_e2));
3918 
3919 		if (xmit_type & XMIT_CSUM_ENC) {
3920 			u16 global_data = 0;
3921 
3922 			/* Set PBD in enc checksum offload case */
3923 			hlen = bnx2x_set_pbd_csum_enc(bp, skb,
3924 						      &pbd_e2_parsing_data,
3925 						      xmit_type);
3926 
3927 			/* turn on 2nd parsing and get a BD */
3928 			bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3929 
3930 			pbd2 = &txdata->tx_desc_ring[bd_prod].parse_2nd_bd;
3931 
3932 			memset(pbd2, 0, sizeof(*pbd2));
3933 
3934 			pbd_e2->data.tunnel_data.ip_hdr_start_inner_w =
3935 				(skb_inner_network_header(skb) -
3936 				 skb->data) >> 1;
3937 
3938 			if (xmit_type & XMIT_GSO_ENC)
3939 				bnx2x_update_pbds_gso_enc(skb, pbd_e2, pbd2,
3940 							  &global_data,
3941 							  xmit_type);
3942 
3943 			pbd2->global_data = cpu_to_le16(global_data);
3944 
3945 			/* add addition parse BD indication to start BD */
3946 			SET_FLAG(tx_start_bd->general_data,
3947 				 ETH_TX_START_BD_PARSE_NBDS, 1);
3948 			/* set encapsulation flag in start BD */
3949 			SET_FLAG(tx_start_bd->general_data,
3950 				 ETH_TX_START_BD_TUNNEL_EXIST, 1);
3951 
3952 			tx_buf->flags |= BNX2X_HAS_SECOND_PBD;
3953 
3954 			nbd++;
3955 		} else if (xmit_type & XMIT_CSUM) {
3956 			/* Set PBD in checksum offload case w/o encapsulation */
3957 			hlen = bnx2x_set_pbd_csum_e2(bp, skb,
3958 						     &pbd_e2_parsing_data,
3959 						     xmit_type);
3960 		}
3961 
3962 		bnx2x_set_ipv6_ext_e2(skb, &pbd_e2_parsing_data, xmit_type);
3963 		/* Add the macs to the parsing BD if this is a vf or if
3964 		 * Tx Switching is enabled.
3965 		 */
3966 		if (IS_VF(bp)) {
3967 			/* override GRE parameters in BD */
3968 			bnx2x_set_fw_mac_addr(&pbd_e2->data.mac_addr.src_hi,
3969 					      &pbd_e2->data.mac_addr.src_mid,
3970 					      &pbd_e2->data.mac_addr.src_lo,
3971 					      eth->h_source);
3972 
3973 			bnx2x_set_fw_mac_addr(&pbd_e2->data.mac_addr.dst_hi,
3974 					      &pbd_e2->data.mac_addr.dst_mid,
3975 					      &pbd_e2->data.mac_addr.dst_lo,
3976 					      eth->h_dest);
3977 		} else {
3978 			if (bp->flags & TX_SWITCHING)
3979 				bnx2x_set_fw_mac_addr(
3980 						&pbd_e2->data.mac_addr.dst_hi,
3981 						&pbd_e2->data.mac_addr.dst_mid,
3982 						&pbd_e2->data.mac_addr.dst_lo,
3983 						eth->h_dest);
3984 #ifdef BNX2X_STOP_ON_ERROR
3985 			/* Enforce security is always set in Stop on Error -
3986 			 * source mac should be present in the parsing BD
3987 			 */
3988 			bnx2x_set_fw_mac_addr(&pbd_e2->data.mac_addr.src_hi,
3989 					      &pbd_e2->data.mac_addr.src_mid,
3990 					      &pbd_e2->data.mac_addr.src_lo,
3991 					      eth->h_source);
3992 #endif
3993 		}
3994 
3995 		SET_FLAG(pbd_e2_parsing_data,
3996 			 ETH_TX_PARSE_BD_E2_ETH_ADDR_TYPE, mac_type);
3997 	} else {
3998 		u16 global_data = 0;
3999 		pbd_e1x = &txdata->tx_desc_ring[bd_prod].parse_bd_e1x;
4000 		memset(pbd_e1x, 0, sizeof(struct eth_tx_parse_bd_e1x));
4001 		/* Set PBD in checksum offload case */
4002 		if (xmit_type & XMIT_CSUM)
4003 			hlen = bnx2x_set_pbd_csum(bp, skb, pbd_e1x, xmit_type);
4004 
4005 		SET_FLAG(global_data,
4006 			 ETH_TX_PARSE_BD_E1X_ETH_ADDR_TYPE, mac_type);
4007 		pbd_e1x->global_data |= cpu_to_le16(global_data);
4008 	}
4009 
4010 	/* Setup the data pointer of the first BD of the packet */
4011 	tx_start_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
4012 	tx_start_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
4013 	tx_start_bd->nbytes = cpu_to_le16(skb_headlen(skb));
4014 	pkt_size = tx_start_bd->nbytes;
4015 
4016 	DP(NETIF_MSG_TX_QUEUED,
4017 	   "first bd @%p  addr (%x:%x)  nbytes %d  flags %x  vlan %x\n",
4018 	   tx_start_bd, tx_start_bd->addr_hi, tx_start_bd->addr_lo,
4019 	   le16_to_cpu(tx_start_bd->nbytes),
4020 	   tx_start_bd->bd_flags.as_bitfield,
4021 	   le16_to_cpu(tx_start_bd->vlan_or_ethertype));
4022 
4023 	if (xmit_type & XMIT_GSO) {
4024 
4025 		DP(NETIF_MSG_TX_QUEUED,
4026 		   "TSO packet len %d  hlen %d  total len %d  tso size %d\n",
4027 		   skb->len, hlen, skb_headlen(skb),
4028 		   skb_shinfo(skb)->gso_size);
4029 
4030 		tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_SW_LSO;
4031 
4032 		if (unlikely(skb_headlen(skb) > hlen)) {
4033 			nbd++;
4034 			bd_prod = bnx2x_tx_split(bp, txdata, tx_buf,
4035 						 &tx_start_bd, hlen,
4036 						 bd_prod);
4037 		}
4038 		if (!CHIP_IS_E1x(bp))
4039 			pbd_e2_parsing_data |=
4040 				(skb_shinfo(skb)->gso_size <<
4041 				 ETH_TX_PARSE_BD_E2_LSO_MSS_SHIFT) &
4042 				 ETH_TX_PARSE_BD_E2_LSO_MSS;
4043 		else
4044 			bnx2x_set_pbd_gso(skb, pbd_e1x, xmit_type);
4045 	}
4046 
4047 	/* Set the PBD's parsing_data field if not zero
4048 	 * (for the chips newer than 57711).
4049 	 */
4050 	if (pbd_e2_parsing_data)
4051 		pbd_e2->parsing_data = cpu_to_le32(pbd_e2_parsing_data);
4052 
4053 	tx_data_bd = (struct eth_tx_bd *)tx_start_bd;
4054 
4055 	/* Handle fragmented skb */
4056 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
4057 		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
4058 
4059 		mapping = skb_frag_dma_map(&bp->pdev->dev, frag, 0,
4060 					   skb_frag_size(frag), DMA_TO_DEVICE);
4061 		if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
4062 			unsigned int pkts_compl = 0, bytes_compl = 0;
4063 
4064 			DP(NETIF_MSG_TX_QUEUED,
4065 			   "Unable to map page - dropping packet...\n");
4066 
4067 			/* we need unmap all buffers already mapped
4068 			 * for this SKB;
4069 			 * first_bd->nbd need to be properly updated
4070 			 * before call to bnx2x_free_tx_pkt
4071 			 */
4072 			first_bd->nbd = cpu_to_le16(nbd);
4073 			bnx2x_free_tx_pkt(bp, txdata,
4074 					  TX_BD(txdata->tx_pkt_prod),
4075 					  &pkts_compl, &bytes_compl);
4076 			return NETDEV_TX_OK;
4077 		}
4078 
4079 		bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
4080 		tx_data_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
4081 		if (total_pkt_bd == NULL)
4082 			total_pkt_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
4083 
4084 		tx_data_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
4085 		tx_data_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
4086 		tx_data_bd->nbytes = cpu_to_le16(skb_frag_size(frag));
4087 		le16_add_cpu(&pkt_size, skb_frag_size(frag));
4088 		nbd++;
4089 
4090 		DP(NETIF_MSG_TX_QUEUED,
4091 		   "frag %d  bd @%p  addr (%x:%x)  nbytes %d\n",
4092 		   i, tx_data_bd, tx_data_bd->addr_hi, tx_data_bd->addr_lo,
4093 		   le16_to_cpu(tx_data_bd->nbytes));
4094 	}
4095 
4096 	DP(NETIF_MSG_TX_QUEUED, "last bd @%p\n", tx_data_bd);
4097 
4098 	/* update with actual num BDs */
4099 	first_bd->nbd = cpu_to_le16(nbd);
4100 
4101 	bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
4102 
4103 	/* now send a tx doorbell, counting the next BD
4104 	 * if the packet contains or ends with it
4105 	 */
4106 	if (TX_BD_POFF(bd_prod) < nbd)
4107 		nbd++;
4108 
4109 	/* total_pkt_bytes should be set on the first data BD if
4110 	 * it's not an LSO packet and there is more than one
4111 	 * data BD. In this case pkt_size is limited by an MTU value.
4112 	 * However we prefer to set it for an LSO packet (while we don't
4113 	 * have to) in order to save some CPU cycles in a none-LSO
4114 	 * case, when we much more care about them.
4115 	 */
4116 	if (total_pkt_bd != NULL)
4117 		total_pkt_bd->total_pkt_bytes = pkt_size;
4118 
4119 	if (pbd_e1x)
4120 		DP(NETIF_MSG_TX_QUEUED,
4121 		   "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",
4122 		   pbd_e1x, pbd_e1x->global_data, pbd_e1x->ip_hlen_w,
4123 		   pbd_e1x->ip_id, pbd_e1x->lso_mss, pbd_e1x->tcp_flags,
4124 		   pbd_e1x->tcp_pseudo_csum, pbd_e1x->tcp_send_seq,
4125 		    le16_to_cpu(pbd_e1x->total_hlen_w));
4126 	if (pbd_e2)
4127 		DP(NETIF_MSG_TX_QUEUED,
4128 		   "PBD (E2) @%p  dst %x %x %x src %x %x %x parsing_data %x\n",
4129 		   pbd_e2,
4130 		   pbd_e2->data.mac_addr.dst_hi,
4131 		   pbd_e2->data.mac_addr.dst_mid,
4132 		   pbd_e2->data.mac_addr.dst_lo,
4133 		   pbd_e2->data.mac_addr.src_hi,
4134 		   pbd_e2->data.mac_addr.src_mid,
4135 		   pbd_e2->data.mac_addr.src_lo,
4136 		   pbd_e2->parsing_data);
4137 	DP(NETIF_MSG_TX_QUEUED, "doorbell: nbd %d  bd %u\n", nbd, bd_prod);
4138 
4139 	netdev_tx_sent_queue(txq, skb->len);
4140 
4141 	skb_tx_timestamp(skb);
4142 
4143 	txdata->tx_pkt_prod++;
4144 	/*
4145 	 * Make sure that the BD data is updated before updating the producer
4146 	 * since FW might read the BD right after the producer is updated.
4147 	 * This is only applicable for weak-ordered memory model archs such
4148 	 * as IA-64. The following barrier is also mandatory since FW will
4149 	 * assumes packets must have BDs.
4150 	 */
4151 	wmb();
4152 
4153 	txdata->tx_db.data.prod += nbd;
4154 	barrier();
4155 
4156 	DOORBELL(bp, txdata->cid, txdata->tx_db.raw);
4157 
4158 	mmiowb();
4159 
4160 	txdata->tx_bd_prod += nbd;
4161 
4162 	if (unlikely(bnx2x_tx_avail(bp, txdata) < MAX_DESC_PER_TX_PKT)) {
4163 		netif_tx_stop_queue(txq);
4164 
4165 		/* paired memory barrier is in bnx2x_tx_int(), we have to keep
4166 		 * ordering of set_bit() in netif_tx_stop_queue() and read of
4167 		 * fp->bd_tx_cons */
4168 		smp_mb();
4169 
4170 		bnx2x_fp_qstats(bp, txdata->parent_fp)->driver_xoff++;
4171 		if (bnx2x_tx_avail(bp, txdata) >= MAX_DESC_PER_TX_PKT)
4172 			netif_tx_wake_queue(txq);
4173 	}
4174 	txdata->tx_pkt++;
4175 
4176 	return NETDEV_TX_OK;
4177 }
4178 
4179 void bnx2x_get_c2s_mapping(struct bnx2x *bp, u8 *c2s_map, u8 *c2s_default)
4180 {
4181 	int mfw_vn = BP_FW_MB_IDX(bp);
4182 	u32 tmp;
4183 
4184 	/* If the shmem shouldn't affect configuration, reflect */
4185 	if (!IS_MF_BD(bp)) {
4186 		int i;
4187 
4188 		for (i = 0; i < BNX2X_MAX_PRIORITY; i++)
4189 			c2s_map[i] = i;
4190 		*c2s_default = 0;
4191 
4192 		return;
4193 	}
4194 
4195 	tmp = SHMEM2_RD(bp, c2s_pcp_map_lower[mfw_vn]);
4196 	tmp = (__force u32)be32_to_cpu((__force __be32)tmp);
4197 	c2s_map[0] = tmp & 0xff;
4198 	c2s_map[1] = (tmp >> 8) & 0xff;
4199 	c2s_map[2] = (tmp >> 16) & 0xff;
4200 	c2s_map[3] = (tmp >> 24) & 0xff;
4201 
4202 	tmp = SHMEM2_RD(bp, c2s_pcp_map_upper[mfw_vn]);
4203 	tmp = (__force u32)be32_to_cpu((__force __be32)tmp);
4204 	c2s_map[4] = tmp & 0xff;
4205 	c2s_map[5] = (tmp >> 8) & 0xff;
4206 	c2s_map[6] = (tmp >> 16) & 0xff;
4207 	c2s_map[7] = (tmp >> 24) & 0xff;
4208 
4209 	tmp = SHMEM2_RD(bp, c2s_pcp_map_default[mfw_vn]);
4210 	tmp = (__force u32)be32_to_cpu((__force __be32)tmp);
4211 	*c2s_default = (tmp >> (8 * mfw_vn)) & 0xff;
4212 }
4213 
4214 /**
4215  * bnx2x_setup_tc - routine to configure net_device for multi tc
4216  *
4217  * @netdev: net device to configure
4218  * @tc: number of traffic classes to enable
4219  *
4220  * callback connected to the ndo_setup_tc function pointer
4221  */
4222 int bnx2x_setup_tc(struct net_device *dev, u8 num_tc)
4223 {
4224 	struct bnx2x *bp = netdev_priv(dev);
4225 	u8 c2s_map[BNX2X_MAX_PRIORITY], c2s_def;
4226 	int cos, prio, count, offset;
4227 
4228 	/* setup tc must be called under rtnl lock */
4229 	ASSERT_RTNL();
4230 
4231 	/* no traffic classes requested. Aborting */
4232 	if (!num_tc) {
4233 		netdev_reset_tc(dev);
4234 		return 0;
4235 	}
4236 
4237 	/* requested to support too many traffic classes */
4238 	if (num_tc > bp->max_cos) {
4239 		BNX2X_ERR("support for too many traffic classes requested: %d. Max supported is %d\n",
4240 			  num_tc, bp->max_cos);
4241 		return -EINVAL;
4242 	}
4243 
4244 	/* declare amount of supported traffic classes */
4245 	if (netdev_set_num_tc(dev, num_tc)) {
4246 		BNX2X_ERR("failed to declare %d traffic classes\n", num_tc);
4247 		return -EINVAL;
4248 	}
4249 
4250 	bnx2x_get_c2s_mapping(bp, c2s_map, &c2s_def);
4251 
4252 	/* configure priority to traffic class mapping */
4253 	for (prio = 0; prio < BNX2X_MAX_PRIORITY; prio++) {
4254 		int outer_prio = c2s_map[prio];
4255 
4256 		netdev_set_prio_tc_map(dev, prio, bp->prio_to_cos[outer_prio]);
4257 		DP(BNX2X_MSG_SP | NETIF_MSG_IFUP,
4258 		   "mapping priority %d to tc %d\n",
4259 		   outer_prio, bp->prio_to_cos[outer_prio]);
4260 	}
4261 
4262 	/* Use this configuration to differentiate tc0 from other COSes
4263 	   This can be used for ets or pfc, and save the effort of setting
4264 	   up a multio class queue disc or negotiating DCBX with a switch
4265 	netdev_set_prio_tc_map(dev, 0, 0);
4266 	DP(BNX2X_MSG_SP, "mapping priority %d to tc %d\n", 0, 0);
4267 	for (prio = 1; prio < 16; prio++) {
4268 		netdev_set_prio_tc_map(dev, prio, 1);
4269 		DP(BNX2X_MSG_SP, "mapping priority %d to tc %d\n", prio, 1);
4270 	} */
4271 
4272 	/* configure traffic class to transmission queue mapping */
4273 	for (cos = 0; cos < bp->max_cos; cos++) {
4274 		count = BNX2X_NUM_ETH_QUEUES(bp);
4275 		offset = cos * BNX2X_NUM_NON_CNIC_QUEUES(bp);
4276 		netdev_set_tc_queue(dev, cos, count, offset);
4277 		DP(BNX2X_MSG_SP | NETIF_MSG_IFUP,
4278 		   "mapping tc %d to offset %d count %d\n",
4279 		   cos, offset, count);
4280 	}
4281 
4282 	return 0;
4283 }
4284 
4285 /* called with rtnl_lock */
4286 int bnx2x_change_mac_addr(struct net_device *dev, void *p)
4287 {
4288 	struct sockaddr *addr = p;
4289 	struct bnx2x *bp = netdev_priv(dev);
4290 	int rc = 0;
4291 
4292 	if (!is_valid_ether_addr(addr->sa_data)) {
4293 		BNX2X_ERR("Requested MAC address is not valid\n");
4294 		return -EINVAL;
4295 	}
4296 
4297 	if (IS_MF_STORAGE_ONLY(bp)) {
4298 		BNX2X_ERR("Can't change address on STORAGE ONLY function\n");
4299 		return -EINVAL;
4300 	}
4301 
4302 	if (netif_running(dev))  {
4303 		rc = bnx2x_set_eth_mac(bp, false);
4304 		if (rc)
4305 			return rc;
4306 	}
4307 
4308 	memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
4309 
4310 	if (netif_running(dev))
4311 		rc = bnx2x_set_eth_mac(bp, true);
4312 
4313 	if (IS_PF(bp) && SHMEM2_HAS(bp, curr_cfg))
4314 		SHMEM2_WR(bp, curr_cfg, CURR_CFG_MET_OS);
4315 
4316 	return rc;
4317 }
4318 
4319 static void bnx2x_free_fp_mem_at(struct bnx2x *bp, int fp_index)
4320 {
4321 	union host_hc_status_block *sb = &bnx2x_fp(bp, fp_index, status_blk);
4322 	struct bnx2x_fastpath *fp = &bp->fp[fp_index];
4323 	u8 cos;
4324 
4325 	/* Common */
4326 
4327 	if (IS_FCOE_IDX(fp_index)) {
4328 		memset(sb, 0, sizeof(union host_hc_status_block));
4329 		fp->status_blk_mapping = 0;
4330 	} else {
4331 		/* status blocks */
4332 		if (!CHIP_IS_E1x(bp))
4333 			BNX2X_PCI_FREE(sb->e2_sb,
4334 				       bnx2x_fp(bp, fp_index,
4335 						status_blk_mapping),
4336 				       sizeof(struct host_hc_status_block_e2));
4337 		else
4338 			BNX2X_PCI_FREE(sb->e1x_sb,
4339 				       bnx2x_fp(bp, fp_index,
4340 						status_blk_mapping),
4341 				       sizeof(struct host_hc_status_block_e1x));
4342 	}
4343 
4344 	/* Rx */
4345 	if (!skip_rx_queue(bp, fp_index)) {
4346 		bnx2x_free_rx_bds(fp);
4347 
4348 		/* fastpath rx rings: rx_buf rx_desc rx_comp */
4349 		BNX2X_FREE(bnx2x_fp(bp, fp_index, rx_buf_ring));
4350 		BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_desc_ring),
4351 			       bnx2x_fp(bp, fp_index, rx_desc_mapping),
4352 			       sizeof(struct eth_rx_bd) * NUM_RX_BD);
4353 
4354 		BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_comp_ring),
4355 			       bnx2x_fp(bp, fp_index, rx_comp_mapping),
4356 			       sizeof(struct eth_fast_path_rx_cqe) *
4357 			       NUM_RCQ_BD);
4358 
4359 		/* SGE ring */
4360 		BNX2X_FREE(bnx2x_fp(bp, fp_index, rx_page_ring));
4361 		BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_sge_ring),
4362 			       bnx2x_fp(bp, fp_index, rx_sge_mapping),
4363 			       BCM_PAGE_SIZE * NUM_RX_SGE_PAGES);
4364 	}
4365 
4366 	/* Tx */
4367 	if (!skip_tx_queue(bp, fp_index)) {
4368 		/* fastpath tx rings: tx_buf tx_desc */
4369 		for_each_cos_in_tx_queue(fp, cos) {
4370 			struct bnx2x_fp_txdata *txdata = fp->txdata_ptr[cos];
4371 
4372 			DP(NETIF_MSG_IFDOWN,
4373 			   "freeing tx memory of fp %d cos %d cid %d\n",
4374 			   fp_index, cos, txdata->cid);
4375 
4376 			BNX2X_FREE(txdata->tx_buf_ring);
4377 			BNX2X_PCI_FREE(txdata->tx_desc_ring,
4378 				txdata->tx_desc_mapping,
4379 				sizeof(union eth_tx_bd_types) * NUM_TX_BD);
4380 		}
4381 	}
4382 	/* end of fastpath */
4383 }
4384 
4385 static void bnx2x_free_fp_mem_cnic(struct bnx2x *bp)
4386 {
4387 	int i;
4388 	for_each_cnic_queue(bp, i)
4389 		bnx2x_free_fp_mem_at(bp, i);
4390 }
4391 
4392 void bnx2x_free_fp_mem(struct bnx2x *bp)
4393 {
4394 	int i;
4395 	for_each_eth_queue(bp, i)
4396 		bnx2x_free_fp_mem_at(bp, i);
4397 }
4398 
4399 static void set_sb_shortcuts(struct bnx2x *bp, int index)
4400 {
4401 	union host_hc_status_block status_blk = bnx2x_fp(bp, index, status_blk);
4402 	if (!CHIP_IS_E1x(bp)) {
4403 		bnx2x_fp(bp, index, sb_index_values) =
4404 			(__le16 *)status_blk.e2_sb->sb.index_values;
4405 		bnx2x_fp(bp, index, sb_running_index) =
4406 			(__le16 *)status_blk.e2_sb->sb.running_index;
4407 	} else {
4408 		bnx2x_fp(bp, index, sb_index_values) =
4409 			(__le16 *)status_blk.e1x_sb->sb.index_values;
4410 		bnx2x_fp(bp, index, sb_running_index) =
4411 			(__le16 *)status_blk.e1x_sb->sb.running_index;
4412 	}
4413 }
4414 
4415 /* Returns the number of actually allocated BDs */
4416 static int bnx2x_alloc_rx_bds(struct bnx2x_fastpath *fp,
4417 			      int rx_ring_size)
4418 {
4419 	struct bnx2x *bp = fp->bp;
4420 	u16 ring_prod, cqe_ring_prod;
4421 	int i, failure_cnt = 0;
4422 
4423 	fp->rx_comp_cons = 0;
4424 	cqe_ring_prod = ring_prod = 0;
4425 
4426 	/* This routine is called only during fo init so
4427 	 * fp->eth_q_stats.rx_skb_alloc_failed = 0
4428 	 */
4429 	for (i = 0; i < rx_ring_size; i++) {
4430 		if (bnx2x_alloc_rx_data(bp, fp, ring_prod, GFP_KERNEL) < 0) {
4431 			failure_cnt++;
4432 			continue;
4433 		}
4434 		ring_prod = NEXT_RX_IDX(ring_prod);
4435 		cqe_ring_prod = NEXT_RCQ_IDX(cqe_ring_prod);
4436 		WARN_ON(ring_prod <= (i - failure_cnt));
4437 	}
4438 
4439 	if (failure_cnt)
4440 		BNX2X_ERR("was only able to allocate %d rx skbs on queue[%d]\n",
4441 			  i - failure_cnt, fp->index);
4442 
4443 	fp->rx_bd_prod = ring_prod;
4444 	/* Limit the CQE producer by the CQE ring size */
4445 	fp->rx_comp_prod = min_t(u16, NUM_RCQ_RINGS*RCQ_DESC_CNT,
4446 			       cqe_ring_prod);
4447 	fp->rx_pkt = fp->rx_calls = 0;
4448 
4449 	bnx2x_fp_stats(bp, fp)->eth_q_stats.rx_skb_alloc_failed += failure_cnt;
4450 
4451 	return i - failure_cnt;
4452 }
4453 
4454 static void bnx2x_set_next_page_rx_cq(struct bnx2x_fastpath *fp)
4455 {
4456 	int i;
4457 
4458 	for (i = 1; i <= NUM_RCQ_RINGS; i++) {
4459 		struct eth_rx_cqe_next_page *nextpg;
4460 
4461 		nextpg = (struct eth_rx_cqe_next_page *)
4462 			&fp->rx_comp_ring[RCQ_DESC_CNT * i - 1];
4463 		nextpg->addr_hi =
4464 			cpu_to_le32(U64_HI(fp->rx_comp_mapping +
4465 				   BCM_PAGE_SIZE*(i % NUM_RCQ_RINGS)));
4466 		nextpg->addr_lo =
4467 			cpu_to_le32(U64_LO(fp->rx_comp_mapping +
4468 				   BCM_PAGE_SIZE*(i % NUM_RCQ_RINGS)));
4469 	}
4470 }
4471 
4472 static int bnx2x_alloc_fp_mem_at(struct bnx2x *bp, int index)
4473 {
4474 	union host_hc_status_block *sb;
4475 	struct bnx2x_fastpath *fp = &bp->fp[index];
4476 	int ring_size = 0;
4477 	u8 cos;
4478 	int rx_ring_size = 0;
4479 
4480 	if (!bp->rx_ring_size && IS_MF_STORAGE_ONLY(bp)) {
4481 		rx_ring_size = MIN_RX_SIZE_NONTPA;
4482 		bp->rx_ring_size = rx_ring_size;
4483 	} else if (!bp->rx_ring_size) {
4484 		rx_ring_size = MAX_RX_AVAIL/BNX2X_NUM_RX_QUEUES(bp);
4485 
4486 		if (CHIP_IS_E3(bp)) {
4487 			u32 cfg = SHMEM_RD(bp,
4488 					   dev_info.port_hw_config[BP_PORT(bp)].
4489 					   default_cfg);
4490 
4491 			/* Decrease ring size for 1G functions */
4492 			if ((cfg & PORT_HW_CFG_NET_SERDES_IF_MASK) ==
4493 			    PORT_HW_CFG_NET_SERDES_IF_SGMII)
4494 				rx_ring_size /= 10;
4495 		}
4496 
4497 		/* allocate at least number of buffers required by FW */
4498 		rx_ring_size = max_t(int, bp->disable_tpa ? MIN_RX_SIZE_NONTPA :
4499 				     MIN_RX_SIZE_TPA, rx_ring_size);
4500 
4501 		bp->rx_ring_size = rx_ring_size;
4502 	} else /* if rx_ring_size specified - use it */
4503 		rx_ring_size = bp->rx_ring_size;
4504 
4505 	DP(BNX2X_MSG_SP, "calculated rx_ring_size %d\n", rx_ring_size);
4506 
4507 	/* Common */
4508 	sb = &bnx2x_fp(bp, index, status_blk);
4509 
4510 	if (!IS_FCOE_IDX(index)) {
4511 		/* status blocks */
4512 		if (!CHIP_IS_E1x(bp)) {
4513 			sb->e2_sb = BNX2X_PCI_ALLOC(&bnx2x_fp(bp, index, status_blk_mapping),
4514 						    sizeof(struct host_hc_status_block_e2));
4515 			if (!sb->e2_sb)
4516 				goto alloc_mem_err;
4517 		} else {
4518 			sb->e1x_sb = BNX2X_PCI_ALLOC(&bnx2x_fp(bp, index, status_blk_mapping),
4519 						     sizeof(struct host_hc_status_block_e1x));
4520 			if (!sb->e1x_sb)
4521 				goto alloc_mem_err;
4522 		}
4523 	}
4524 
4525 	/* FCoE Queue uses Default SB and doesn't ACK the SB, thus no need to
4526 	 * set shortcuts for it.
4527 	 */
4528 	if (!IS_FCOE_IDX(index))
4529 		set_sb_shortcuts(bp, index);
4530 
4531 	/* Tx */
4532 	if (!skip_tx_queue(bp, index)) {
4533 		/* fastpath tx rings: tx_buf tx_desc */
4534 		for_each_cos_in_tx_queue(fp, cos) {
4535 			struct bnx2x_fp_txdata *txdata = fp->txdata_ptr[cos];
4536 
4537 			DP(NETIF_MSG_IFUP,
4538 			   "allocating tx memory of fp %d cos %d\n",
4539 			   index, cos);
4540 
4541 			txdata->tx_buf_ring = kcalloc(NUM_TX_BD,
4542 						      sizeof(struct sw_tx_bd),
4543 						      GFP_KERNEL);
4544 			if (!txdata->tx_buf_ring)
4545 				goto alloc_mem_err;
4546 			txdata->tx_desc_ring = BNX2X_PCI_ALLOC(&txdata->tx_desc_mapping,
4547 							       sizeof(union eth_tx_bd_types) * NUM_TX_BD);
4548 			if (!txdata->tx_desc_ring)
4549 				goto alloc_mem_err;
4550 		}
4551 	}
4552 
4553 	/* Rx */
4554 	if (!skip_rx_queue(bp, index)) {
4555 		/* fastpath rx rings: rx_buf rx_desc rx_comp */
4556 		bnx2x_fp(bp, index, rx_buf_ring) =
4557 			kcalloc(NUM_RX_BD, sizeof(struct sw_rx_bd), GFP_KERNEL);
4558 		if (!bnx2x_fp(bp, index, rx_buf_ring))
4559 			goto alloc_mem_err;
4560 		bnx2x_fp(bp, index, rx_desc_ring) =
4561 			BNX2X_PCI_ALLOC(&bnx2x_fp(bp, index, rx_desc_mapping),
4562 					sizeof(struct eth_rx_bd) * NUM_RX_BD);
4563 		if (!bnx2x_fp(bp, index, rx_desc_ring))
4564 			goto alloc_mem_err;
4565 
4566 		/* Seed all CQEs by 1s */
4567 		bnx2x_fp(bp, index, rx_comp_ring) =
4568 			BNX2X_PCI_FALLOC(&bnx2x_fp(bp, index, rx_comp_mapping),
4569 					 sizeof(struct eth_fast_path_rx_cqe) * NUM_RCQ_BD);
4570 		if (!bnx2x_fp(bp, index, rx_comp_ring))
4571 			goto alloc_mem_err;
4572 
4573 		/* SGE ring */
4574 		bnx2x_fp(bp, index, rx_page_ring) =
4575 			kcalloc(NUM_RX_SGE, sizeof(struct sw_rx_page),
4576 				GFP_KERNEL);
4577 		if (!bnx2x_fp(bp, index, rx_page_ring))
4578 			goto alloc_mem_err;
4579 		bnx2x_fp(bp, index, rx_sge_ring) =
4580 			BNX2X_PCI_ALLOC(&bnx2x_fp(bp, index, rx_sge_mapping),
4581 					BCM_PAGE_SIZE * NUM_RX_SGE_PAGES);
4582 		if (!bnx2x_fp(bp, index, rx_sge_ring))
4583 			goto alloc_mem_err;
4584 		/* RX BD ring */
4585 		bnx2x_set_next_page_rx_bd(fp);
4586 
4587 		/* CQ ring */
4588 		bnx2x_set_next_page_rx_cq(fp);
4589 
4590 		/* BDs */
4591 		ring_size = bnx2x_alloc_rx_bds(fp, rx_ring_size);
4592 		if (ring_size < rx_ring_size)
4593 			goto alloc_mem_err;
4594 	}
4595 
4596 	return 0;
4597 
4598 /* handles low memory cases */
4599 alloc_mem_err:
4600 	BNX2X_ERR("Unable to allocate full memory for queue %d (size %d)\n",
4601 						index, ring_size);
4602 	/* FW will drop all packets if queue is not big enough,
4603 	 * In these cases we disable the queue
4604 	 * Min size is different for OOO, TPA and non-TPA queues
4605 	 */
4606 	if (ring_size < (fp->mode == TPA_MODE_DISABLED ?
4607 				MIN_RX_SIZE_NONTPA : MIN_RX_SIZE_TPA)) {
4608 			/* release memory allocated for this queue */
4609 			bnx2x_free_fp_mem_at(bp, index);
4610 			return -ENOMEM;
4611 	}
4612 	return 0;
4613 }
4614 
4615 static int bnx2x_alloc_fp_mem_cnic(struct bnx2x *bp)
4616 {
4617 	if (!NO_FCOE(bp))
4618 		/* FCoE */
4619 		if (bnx2x_alloc_fp_mem_at(bp, FCOE_IDX(bp)))
4620 			/* we will fail load process instead of mark
4621 			 * NO_FCOE_FLAG
4622 			 */
4623 			return -ENOMEM;
4624 
4625 	return 0;
4626 }
4627 
4628 static int bnx2x_alloc_fp_mem(struct bnx2x *bp)
4629 {
4630 	int i;
4631 
4632 	/* 1. Allocate FP for leading - fatal if error
4633 	 * 2. Allocate RSS - fix number of queues if error
4634 	 */
4635 
4636 	/* leading */
4637 	if (bnx2x_alloc_fp_mem_at(bp, 0))
4638 		return -ENOMEM;
4639 
4640 	/* RSS */
4641 	for_each_nondefault_eth_queue(bp, i)
4642 		if (bnx2x_alloc_fp_mem_at(bp, i))
4643 			break;
4644 
4645 	/* handle memory failures */
4646 	if (i != BNX2X_NUM_ETH_QUEUES(bp)) {
4647 		int delta = BNX2X_NUM_ETH_QUEUES(bp) - i;
4648 
4649 		WARN_ON(delta < 0);
4650 		bnx2x_shrink_eth_fp(bp, delta);
4651 		if (CNIC_SUPPORT(bp))
4652 			/* move non eth FPs next to last eth FP
4653 			 * must be done in that order
4654 			 * FCOE_IDX < FWD_IDX < OOO_IDX
4655 			 */
4656 
4657 			/* move FCoE fp even NO_FCOE_FLAG is on */
4658 			bnx2x_move_fp(bp, FCOE_IDX(bp), FCOE_IDX(bp) - delta);
4659 		bp->num_ethernet_queues -= delta;
4660 		bp->num_queues = bp->num_ethernet_queues +
4661 				 bp->num_cnic_queues;
4662 		BNX2X_ERR("Adjusted num of queues from %d to %d\n",
4663 			  bp->num_queues + delta, bp->num_queues);
4664 	}
4665 
4666 	return 0;
4667 }
4668 
4669 void bnx2x_free_mem_bp(struct bnx2x *bp)
4670 {
4671 	int i;
4672 
4673 	for (i = 0; i < bp->fp_array_size; i++)
4674 		kfree(bp->fp[i].tpa_info);
4675 	kfree(bp->fp);
4676 	kfree(bp->sp_objs);
4677 	kfree(bp->fp_stats);
4678 	kfree(bp->bnx2x_txq);
4679 	kfree(bp->msix_table);
4680 	kfree(bp->ilt);
4681 }
4682 
4683 int bnx2x_alloc_mem_bp(struct bnx2x *bp)
4684 {
4685 	struct bnx2x_fastpath *fp;
4686 	struct msix_entry *tbl;
4687 	struct bnx2x_ilt *ilt;
4688 	int msix_table_size = 0;
4689 	int fp_array_size, txq_array_size;
4690 	int i;
4691 
4692 	/*
4693 	 * The biggest MSI-X table we might need is as a maximum number of fast
4694 	 * path IGU SBs plus default SB (for PF only).
4695 	 */
4696 	msix_table_size = bp->igu_sb_cnt;
4697 	if (IS_PF(bp))
4698 		msix_table_size++;
4699 	BNX2X_DEV_INFO("msix_table_size %d\n", msix_table_size);
4700 
4701 	/* fp array: RSS plus CNIC related L2 queues */
4702 	fp_array_size = BNX2X_MAX_RSS_COUNT(bp) + CNIC_SUPPORT(bp);
4703 	bp->fp_array_size = fp_array_size;
4704 	BNX2X_DEV_INFO("fp_array_size %d\n", bp->fp_array_size);
4705 
4706 	fp = kcalloc(bp->fp_array_size, sizeof(*fp), GFP_KERNEL);
4707 	if (!fp)
4708 		goto alloc_err;
4709 	for (i = 0; i < bp->fp_array_size; i++) {
4710 		fp[i].tpa_info =
4711 			kcalloc(ETH_MAX_AGGREGATION_QUEUES_E1H_E2,
4712 				sizeof(struct bnx2x_agg_info), GFP_KERNEL);
4713 		if (!(fp[i].tpa_info))
4714 			goto alloc_err;
4715 	}
4716 
4717 	bp->fp = fp;
4718 
4719 	/* allocate sp objs */
4720 	bp->sp_objs = kcalloc(bp->fp_array_size, sizeof(struct bnx2x_sp_objs),
4721 			      GFP_KERNEL);
4722 	if (!bp->sp_objs)
4723 		goto alloc_err;
4724 
4725 	/* allocate fp_stats */
4726 	bp->fp_stats = kcalloc(bp->fp_array_size, sizeof(struct bnx2x_fp_stats),
4727 			       GFP_KERNEL);
4728 	if (!bp->fp_stats)
4729 		goto alloc_err;
4730 
4731 	/* Allocate memory for the transmission queues array */
4732 	txq_array_size =
4733 		BNX2X_MAX_RSS_COUNT(bp) * BNX2X_MULTI_TX_COS + CNIC_SUPPORT(bp);
4734 	BNX2X_DEV_INFO("txq_array_size %d", txq_array_size);
4735 
4736 	bp->bnx2x_txq = kcalloc(txq_array_size, sizeof(struct bnx2x_fp_txdata),
4737 				GFP_KERNEL);
4738 	if (!bp->bnx2x_txq)
4739 		goto alloc_err;
4740 
4741 	/* msix table */
4742 	tbl = kcalloc(msix_table_size, sizeof(*tbl), GFP_KERNEL);
4743 	if (!tbl)
4744 		goto alloc_err;
4745 	bp->msix_table = tbl;
4746 
4747 	/* ilt */
4748 	ilt = kzalloc(sizeof(*ilt), GFP_KERNEL);
4749 	if (!ilt)
4750 		goto alloc_err;
4751 	bp->ilt = ilt;
4752 
4753 	return 0;
4754 alloc_err:
4755 	bnx2x_free_mem_bp(bp);
4756 	return -ENOMEM;
4757 }
4758 
4759 int bnx2x_reload_if_running(struct net_device *dev)
4760 {
4761 	struct bnx2x *bp = netdev_priv(dev);
4762 
4763 	if (unlikely(!netif_running(dev)))
4764 		return 0;
4765 
4766 	bnx2x_nic_unload(bp, UNLOAD_NORMAL, true);
4767 	return bnx2x_nic_load(bp, LOAD_NORMAL);
4768 }
4769 
4770 int bnx2x_get_cur_phy_idx(struct bnx2x *bp)
4771 {
4772 	u32 sel_phy_idx = 0;
4773 	if (bp->link_params.num_phys <= 1)
4774 		return INT_PHY;
4775 
4776 	if (bp->link_vars.link_up) {
4777 		sel_phy_idx = EXT_PHY1;
4778 		/* In case link is SERDES, check if the EXT_PHY2 is the one */
4779 		if ((bp->link_vars.link_status & LINK_STATUS_SERDES_LINK) &&
4780 		    (bp->link_params.phy[EXT_PHY2].supported & SUPPORTED_FIBRE))
4781 			sel_phy_idx = EXT_PHY2;
4782 	} else {
4783 
4784 		switch (bnx2x_phy_selection(&bp->link_params)) {
4785 		case PORT_HW_CFG_PHY_SELECTION_HARDWARE_DEFAULT:
4786 		case PORT_HW_CFG_PHY_SELECTION_FIRST_PHY:
4787 		case PORT_HW_CFG_PHY_SELECTION_FIRST_PHY_PRIORITY:
4788 		       sel_phy_idx = EXT_PHY1;
4789 		       break;
4790 		case PORT_HW_CFG_PHY_SELECTION_SECOND_PHY:
4791 		case PORT_HW_CFG_PHY_SELECTION_SECOND_PHY_PRIORITY:
4792 		       sel_phy_idx = EXT_PHY2;
4793 		       break;
4794 		}
4795 	}
4796 
4797 	return sel_phy_idx;
4798 }
4799 int bnx2x_get_link_cfg_idx(struct bnx2x *bp)
4800 {
4801 	u32 sel_phy_idx = bnx2x_get_cur_phy_idx(bp);
4802 	/*
4803 	 * The selected activated PHY is always after swapping (in case PHY
4804 	 * swapping is enabled). So when swapping is enabled, we need to reverse
4805 	 * the configuration
4806 	 */
4807 
4808 	if (bp->link_params.multi_phy_config &
4809 	    PORT_HW_CFG_PHY_SWAPPED_ENABLED) {
4810 		if (sel_phy_idx == EXT_PHY1)
4811 			sel_phy_idx = EXT_PHY2;
4812 		else if (sel_phy_idx == EXT_PHY2)
4813 			sel_phy_idx = EXT_PHY1;
4814 	}
4815 	return LINK_CONFIG_IDX(sel_phy_idx);
4816 }
4817 
4818 #ifdef NETDEV_FCOE_WWNN
4819 int bnx2x_fcoe_get_wwn(struct net_device *dev, u64 *wwn, int type)
4820 {
4821 	struct bnx2x *bp = netdev_priv(dev);
4822 	struct cnic_eth_dev *cp = &bp->cnic_eth_dev;
4823 
4824 	switch (type) {
4825 	case NETDEV_FCOE_WWNN:
4826 		*wwn = HILO_U64(cp->fcoe_wwn_node_name_hi,
4827 				cp->fcoe_wwn_node_name_lo);
4828 		break;
4829 	case NETDEV_FCOE_WWPN:
4830 		*wwn = HILO_U64(cp->fcoe_wwn_port_name_hi,
4831 				cp->fcoe_wwn_port_name_lo);
4832 		break;
4833 	default:
4834 		BNX2X_ERR("Wrong WWN type requested - %d\n", type);
4835 		return -EINVAL;
4836 	}
4837 
4838 	return 0;
4839 }
4840 #endif
4841 
4842 /* called with rtnl_lock */
4843 int bnx2x_change_mtu(struct net_device *dev, int new_mtu)
4844 {
4845 	struct bnx2x *bp = netdev_priv(dev);
4846 
4847 	if (pci_num_vf(bp->pdev)) {
4848 		DP(BNX2X_MSG_IOV, "VFs are enabled, can not change MTU\n");
4849 		return -EPERM;
4850 	}
4851 
4852 	if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
4853 		BNX2X_ERR("Can't perform change MTU during parity recovery\n");
4854 		return -EAGAIN;
4855 	}
4856 
4857 	if ((new_mtu > ETH_MAX_JUMBO_PACKET_SIZE) ||
4858 	    ((new_mtu + ETH_HLEN) < ETH_MIN_PACKET_SIZE)) {
4859 		BNX2X_ERR("Can't support requested MTU size\n");
4860 		return -EINVAL;
4861 	}
4862 
4863 	/* This does not race with packet allocation
4864 	 * because the actual alloc size is
4865 	 * only updated as part of load
4866 	 */
4867 	dev->mtu = new_mtu;
4868 
4869 	if (IS_PF(bp) && SHMEM2_HAS(bp, curr_cfg))
4870 		SHMEM2_WR(bp, curr_cfg, CURR_CFG_MET_OS);
4871 
4872 	return bnx2x_reload_if_running(dev);
4873 }
4874 
4875 netdev_features_t bnx2x_fix_features(struct net_device *dev,
4876 				     netdev_features_t features)
4877 {
4878 	struct bnx2x *bp = netdev_priv(dev);
4879 
4880 	if (pci_num_vf(bp->pdev)) {
4881 		netdev_features_t changed = dev->features ^ features;
4882 
4883 		/* Revert the requested changes in features if they
4884 		 * would require internal reload of PF in bnx2x_set_features().
4885 		 */
4886 		if (!(features & NETIF_F_RXCSUM) && !bp->disable_tpa) {
4887 			features &= ~NETIF_F_RXCSUM;
4888 			features |= dev->features & NETIF_F_RXCSUM;
4889 		}
4890 
4891 		if (changed & NETIF_F_LOOPBACK) {
4892 			features &= ~NETIF_F_LOOPBACK;
4893 			features |= dev->features & NETIF_F_LOOPBACK;
4894 		}
4895 	}
4896 
4897 	/* TPA requires Rx CSUM offloading */
4898 	if (!(features & NETIF_F_RXCSUM)) {
4899 		features &= ~NETIF_F_LRO;
4900 		features &= ~NETIF_F_GRO;
4901 	}
4902 
4903 	return features;
4904 }
4905 
4906 int bnx2x_set_features(struct net_device *dev, netdev_features_t features)
4907 {
4908 	struct bnx2x *bp = netdev_priv(dev);
4909 	netdev_features_t changes = features ^ dev->features;
4910 	bool bnx2x_reload = false;
4911 	int rc;
4912 
4913 	/* VFs or non SRIOV PFs should be able to change loopback feature */
4914 	if (!pci_num_vf(bp->pdev)) {
4915 		if (features & NETIF_F_LOOPBACK) {
4916 			if (bp->link_params.loopback_mode != LOOPBACK_BMAC) {
4917 				bp->link_params.loopback_mode = LOOPBACK_BMAC;
4918 				bnx2x_reload = true;
4919 			}
4920 		} else {
4921 			if (bp->link_params.loopback_mode != LOOPBACK_NONE) {
4922 				bp->link_params.loopback_mode = LOOPBACK_NONE;
4923 				bnx2x_reload = true;
4924 			}
4925 		}
4926 	}
4927 
4928 	/* if GRO is changed while LRO is enabled, don't force a reload */
4929 	if ((changes & NETIF_F_GRO) && (features & NETIF_F_LRO))
4930 		changes &= ~NETIF_F_GRO;
4931 
4932 	/* if GRO is changed while HW TPA is off, don't force a reload */
4933 	if ((changes & NETIF_F_GRO) && bp->disable_tpa)
4934 		changes &= ~NETIF_F_GRO;
4935 
4936 	if (changes)
4937 		bnx2x_reload = true;
4938 
4939 	if (bnx2x_reload) {
4940 		if (bp->recovery_state == BNX2X_RECOVERY_DONE) {
4941 			dev->features = features;
4942 			rc = bnx2x_reload_if_running(dev);
4943 			return rc ? rc : 1;
4944 		}
4945 		/* else: bnx2x_nic_load() will be called at end of recovery */
4946 	}
4947 
4948 	return 0;
4949 }
4950 
4951 void bnx2x_tx_timeout(struct net_device *dev)
4952 {
4953 	struct bnx2x *bp = netdev_priv(dev);
4954 
4955 #ifdef BNX2X_STOP_ON_ERROR
4956 	if (!bp->panic)
4957 		bnx2x_panic();
4958 #endif
4959 
4960 	/* This allows the netif to be shutdown gracefully before resetting */
4961 	bnx2x_schedule_sp_rtnl(bp, BNX2X_SP_RTNL_TX_TIMEOUT, 0);
4962 }
4963 
4964 int bnx2x_suspend(struct pci_dev *pdev, pm_message_t state)
4965 {
4966 	struct net_device *dev = pci_get_drvdata(pdev);
4967 	struct bnx2x *bp;
4968 
4969 	if (!dev) {
4970 		dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
4971 		return -ENODEV;
4972 	}
4973 	bp = netdev_priv(dev);
4974 
4975 	rtnl_lock();
4976 
4977 	pci_save_state(pdev);
4978 
4979 	if (!netif_running(dev)) {
4980 		rtnl_unlock();
4981 		return 0;
4982 	}
4983 
4984 	netif_device_detach(dev);
4985 
4986 	bnx2x_nic_unload(bp, UNLOAD_CLOSE, false);
4987 
4988 	bnx2x_set_power_state(bp, pci_choose_state(pdev, state));
4989 
4990 	rtnl_unlock();
4991 
4992 	return 0;
4993 }
4994 
4995 int bnx2x_resume(struct pci_dev *pdev)
4996 {
4997 	struct net_device *dev = pci_get_drvdata(pdev);
4998 	struct bnx2x *bp;
4999 	int rc;
5000 
5001 	if (!dev) {
5002 		dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
5003 		return -ENODEV;
5004 	}
5005 	bp = netdev_priv(dev);
5006 
5007 	if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
5008 		BNX2X_ERR("Handling parity error recovery. Try again later\n");
5009 		return -EAGAIN;
5010 	}
5011 
5012 	rtnl_lock();
5013 
5014 	pci_restore_state(pdev);
5015 
5016 	if (!netif_running(dev)) {
5017 		rtnl_unlock();
5018 		return 0;
5019 	}
5020 
5021 	bnx2x_set_power_state(bp, PCI_D0);
5022 	netif_device_attach(dev);
5023 
5024 	rc = bnx2x_nic_load(bp, LOAD_OPEN);
5025 
5026 	rtnl_unlock();
5027 
5028 	return rc;
5029 }
5030 
5031 void bnx2x_set_ctx_validation(struct bnx2x *bp, struct eth_context *cxt,
5032 			      u32 cid)
5033 {
5034 	if (!cxt) {
5035 		BNX2X_ERR("bad context pointer %p\n", cxt);
5036 		return;
5037 	}
5038 
5039 	/* ustorm cxt validation */
5040 	cxt->ustorm_ag_context.cdu_usage =
5041 		CDU_RSRVD_VALUE_TYPE_A(HW_CID(bp, cid),
5042 			CDU_REGION_NUMBER_UCM_AG, ETH_CONNECTION_TYPE);
5043 	/* xcontext validation */
5044 	cxt->xstorm_ag_context.cdu_reserved =
5045 		CDU_RSRVD_VALUE_TYPE_A(HW_CID(bp, cid),
5046 			CDU_REGION_NUMBER_XCM_AG, ETH_CONNECTION_TYPE);
5047 }
5048 
5049 static void storm_memset_hc_timeout(struct bnx2x *bp, u8 port,
5050 				    u8 fw_sb_id, u8 sb_index,
5051 				    u8 ticks)
5052 {
5053 	u32 addr = BAR_CSTRORM_INTMEM +
5054 		   CSTORM_STATUS_BLOCK_DATA_TIMEOUT_OFFSET(fw_sb_id, sb_index);
5055 	REG_WR8(bp, addr, ticks);
5056 	DP(NETIF_MSG_IFUP,
5057 	   "port %x fw_sb_id %d sb_index %d ticks %d\n",
5058 	   port, fw_sb_id, sb_index, ticks);
5059 }
5060 
5061 static void storm_memset_hc_disable(struct bnx2x *bp, u8 port,
5062 				    u16 fw_sb_id, u8 sb_index,
5063 				    u8 disable)
5064 {
5065 	u32 enable_flag = disable ? 0 : (1 << HC_INDEX_DATA_HC_ENABLED_SHIFT);
5066 	u32 addr = BAR_CSTRORM_INTMEM +
5067 		   CSTORM_STATUS_BLOCK_DATA_FLAGS_OFFSET(fw_sb_id, sb_index);
5068 	u8 flags = REG_RD8(bp, addr);
5069 	/* clear and set */
5070 	flags &= ~HC_INDEX_DATA_HC_ENABLED;
5071 	flags |= enable_flag;
5072 	REG_WR8(bp, addr, flags);
5073 	DP(NETIF_MSG_IFUP,
5074 	   "port %x fw_sb_id %d sb_index %d disable %d\n",
5075 	   port, fw_sb_id, sb_index, disable);
5076 }
5077 
5078 void bnx2x_update_coalesce_sb_index(struct bnx2x *bp, u8 fw_sb_id,
5079 				    u8 sb_index, u8 disable, u16 usec)
5080 {
5081 	int port = BP_PORT(bp);
5082 	u8 ticks = usec / BNX2X_BTR;
5083 
5084 	storm_memset_hc_timeout(bp, port, fw_sb_id, sb_index, ticks);
5085 
5086 	disable = disable ? 1 : (usec ? 0 : 1);
5087 	storm_memset_hc_disable(bp, port, fw_sb_id, sb_index, disable);
5088 }
5089 
5090 void bnx2x_schedule_sp_rtnl(struct bnx2x *bp, enum sp_rtnl_flag flag,
5091 			    u32 verbose)
5092 {
5093 	smp_mb__before_atomic();
5094 	set_bit(flag, &bp->sp_rtnl_state);
5095 	smp_mb__after_atomic();
5096 	DP((BNX2X_MSG_SP | verbose), "Scheduling sp_rtnl task [Flag: %d]\n",
5097 	   flag);
5098 	schedule_delayed_work(&bp->sp_rtnl_task, 0);
5099 }
5100 EXPORT_SYMBOL(bnx2x_schedule_sp_rtnl);
5101