Merge branch 'acpi-ec'
[linux-drm-fsl-dcu.git] / drivers / net / ethernet / broadcom / bnx2x / bnx2x_cmn.c
1 /* bnx2x_cmn.c: Broadcom Everest network driver.
2  *
3  * Copyright (c) 2007-2013 Broadcom Corporation
4  *
5  * This program is free software; you can redistribute it and/or modify
6  * it under the terms of the GNU General Public License as published by
7  * the Free Software Foundation.
8  *
9  * Maintained by: Ariel Elior <ariel.elior@qlogic.com>
10  * Written by: Eliezer Tamir
11  * Based on code from Michael Chan's bnx2 driver
12  * UDP CSUM errata workaround by Arik Gendelman
13  * Slowpath and fastpath rework by Vladislav Zolotarov
14  * Statistics and Link management by Yitchak Gertner
15  *
16  */
17
18 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
19
20 #include <linux/etherdevice.h>
21 #include <linux/if_vlan.h>
22 #include <linux/interrupt.h>
23 #include <linux/ip.h>
24 #include <linux/crash_dump.h>
25 #include <net/tcp.h>
26 #include <net/ipv6.h>
27 #include <net/ip6_checksum.h>
28 #include <net/busy_poll.h>
29 #include <linux/prefetch.h>
30 #include "bnx2x_cmn.h"
31 #include "bnx2x_init.h"
32 #include "bnx2x_sp.h"
33
34 static void bnx2x_free_fp_mem_cnic(struct bnx2x *bp);
35 static int bnx2x_alloc_fp_mem_cnic(struct bnx2x *bp);
36 static int bnx2x_alloc_fp_mem(struct bnx2x *bp);
37 static int bnx2x_poll(struct napi_struct *napi, int budget);
38
39 static void bnx2x_add_all_napi_cnic(struct bnx2x *bp)
40 {
41         int i;
42
43         /* Add NAPI objects */
44         for_each_rx_queue_cnic(bp, i) {
45                 netif_napi_add(bp->dev, &bnx2x_fp(bp, i, napi),
46                                bnx2x_poll, NAPI_POLL_WEIGHT);
47                 napi_hash_add(&bnx2x_fp(bp, i, napi));
48         }
49 }
50
51 static void bnx2x_add_all_napi(struct bnx2x *bp)
52 {
53         int i;
54
55         /* Add NAPI objects */
56         for_each_eth_queue(bp, i) {
57                 netif_napi_add(bp->dev, &bnx2x_fp(bp, i, napi),
58                                bnx2x_poll, NAPI_POLL_WEIGHT);
59                 napi_hash_add(&bnx2x_fp(bp, i, napi));
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         }
266
267         dev_kfree_skb_any(skb);
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 page *page = alloc_pages(gfp_mask, PAGES_PER_SGE_SHIFT);
548         struct sw_rx_page *sw_buf = &fp->rx_page_ring[index];
549         struct eth_rx_sge *sge = &fp->rx_sge_ring[index];
550         dma_addr_t mapping;
551
552         if (unlikely(page == NULL)) {
553                 BNX2X_ERR("Can't alloc sge\n");
554                 return -ENOMEM;
555         }
556
557         mapping = dma_map_page(&bp->pdev->dev, page, 0,
558                                SGE_PAGES, DMA_FROM_DEVICE);
559         if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
560                 __free_pages(page, PAGES_PER_SGE_SHIFT);
561                 BNX2X_ERR("Can't map sge\n");
562                 return -ENOMEM;
563         }
564
565         sw_buf->page = page;
566         dma_unmap_addr_set(sw_buf, mapping, mapping);
567
568         sge->addr_hi = cpu_to_le32(U64_HI(mapping));
569         sge->addr_lo = cpu_to_le32(U64_LO(mapping));
570
571         return 0;
572 }
573
574 static int bnx2x_fill_frag_skb(struct bnx2x *bp, struct bnx2x_fastpath *fp,
575                                struct bnx2x_agg_info *tpa_info,
576                                u16 pages,
577                                struct sk_buff *skb,
578                                struct eth_end_agg_rx_cqe *cqe,
579                                u16 cqe_idx)
580 {
581         struct sw_rx_page *rx_pg, old_rx_pg;
582         u32 i, frag_len, frag_size;
583         int err, j, frag_id = 0;
584         u16 len_on_bd = tpa_info->len_on_bd;
585         u16 full_page = 0, gro_size = 0;
586
587         frag_size = le16_to_cpu(cqe->pkt_len) - len_on_bd;
588
589         if (fp->mode == TPA_MODE_GRO) {
590                 gro_size = tpa_info->gro_size;
591                 full_page = tpa_info->full_page;
592         }
593
594         /* This is needed in order to enable forwarding support */
595         if (frag_size)
596                 bnx2x_set_gro_params(skb, tpa_info->parsing_flags, len_on_bd,
597                                      le16_to_cpu(cqe->pkt_len),
598                                      le16_to_cpu(cqe->num_of_coalesced_segs));
599
600 #ifdef BNX2X_STOP_ON_ERROR
601         if (pages > min_t(u32, 8, MAX_SKB_FRAGS) * SGE_PAGES) {
602                 BNX2X_ERR("SGL length is too long: %d. CQE index is %d\n",
603                           pages, cqe_idx);
604                 BNX2X_ERR("cqe->pkt_len = %d\n", cqe->pkt_len);
605                 bnx2x_panic();
606                 return -EINVAL;
607         }
608 #endif
609
610         /* Run through the SGL and compose the fragmented skb */
611         for (i = 0, j = 0; i < pages; i += PAGES_PER_SGE, j++) {
612                 u16 sge_idx = RX_SGE(le16_to_cpu(cqe->sgl_or_raw_data.sgl[j]));
613
614                 /* FW gives the indices of the SGE as if the ring is an array
615                    (meaning that "next" element will consume 2 indices) */
616                 if (fp->mode == TPA_MODE_GRO)
617                         frag_len = min_t(u32, frag_size, (u32)full_page);
618                 else /* LRO */
619                         frag_len = min_t(u32, frag_size, (u32)SGE_PAGES);
620
621                 rx_pg = &fp->rx_page_ring[sge_idx];
622                 old_rx_pg = *rx_pg;
623
624                 /* If we fail to allocate a substitute page, we simply stop
625                    where we are and drop the whole packet */
626                 err = bnx2x_alloc_rx_sge(bp, fp, sge_idx, GFP_ATOMIC);
627                 if (unlikely(err)) {
628                         bnx2x_fp_qstats(bp, fp)->rx_skb_alloc_failed++;
629                         return err;
630                 }
631
632                 /* Unmap the page as we're going to pass it to the stack */
633                 dma_unmap_page(&bp->pdev->dev,
634                                dma_unmap_addr(&old_rx_pg, mapping),
635                                SGE_PAGES, DMA_FROM_DEVICE);
636                 /* Add one frag and update the appropriate fields in the skb */
637                 if (fp->mode == TPA_MODE_LRO)
638                         skb_fill_page_desc(skb, j, old_rx_pg.page, 0, frag_len);
639                 else { /* GRO */
640                         int rem;
641                         int offset = 0;
642                         for (rem = frag_len; rem > 0; rem -= gro_size) {
643                                 int len = rem > gro_size ? gro_size : rem;
644                                 skb_fill_page_desc(skb, frag_id++,
645                                                    old_rx_pg.page, offset, len);
646                                 if (offset)
647                                         get_page(old_rx_pg.page);
648                                 offset += len;
649                         }
650                 }
651
652                 skb->data_len += frag_len;
653                 skb->truesize += SGE_PAGES;
654                 skb->len += frag_len;
655
656                 frag_size -= frag_len;
657         }
658
659         return 0;
660 }
661
662 static void bnx2x_frag_free(const struct bnx2x_fastpath *fp, void *data)
663 {
664         if (fp->rx_frag_size)
665                 put_page(virt_to_head_page(data));
666         else
667                 kfree(data);
668 }
669
670 static void *bnx2x_frag_alloc(const struct bnx2x_fastpath *fp, gfp_t gfp_mask)
671 {
672         if (fp->rx_frag_size) {
673                 /* GFP_KERNEL allocations are used only during initialization */
674                 if (unlikely(gfp_mask & __GFP_WAIT))
675                         return (void *)__get_free_page(gfp_mask);
676
677                 return netdev_alloc_frag(fp->rx_frag_size);
678         }
679
680         return kmalloc(fp->rx_buf_size + NET_SKB_PAD, gfp_mask);
681 }
682
683 #ifdef CONFIG_INET
684 static void bnx2x_gro_ip_csum(struct bnx2x *bp, struct sk_buff *skb)
685 {
686         const struct iphdr *iph = ip_hdr(skb);
687         struct tcphdr *th;
688
689         skb_set_transport_header(skb, sizeof(struct iphdr));
690         th = tcp_hdr(skb);
691
692         th->check = ~tcp_v4_check(skb->len - skb_transport_offset(skb),
693                                   iph->saddr, iph->daddr, 0);
694 }
695
696 static void bnx2x_gro_ipv6_csum(struct bnx2x *bp, struct sk_buff *skb)
697 {
698         struct ipv6hdr *iph = ipv6_hdr(skb);
699         struct tcphdr *th;
700
701         skb_set_transport_header(skb, sizeof(struct ipv6hdr));
702         th = tcp_hdr(skb);
703
704         th->check = ~tcp_v6_check(skb->len - skb_transport_offset(skb),
705                                   &iph->saddr, &iph->daddr, 0);
706 }
707
708 static void bnx2x_gro_csum(struct bnx2x *bp, struct sk_buff *skb,
709                             void (*gro_func)(struct bnx2x*, struct sk_buff*))
710 {
711         skb_set_network_header(skb, 0);
712         gro_func(bp, skb);
713         tcp_gro_complete(skb);
714 }
715 #endif
716
717 static void bnx2x_gro_receive(struct bnx2x *bp, struct bnx2x_fastpath *fp,
718                                struct sk_buff *skb)
719 {
720 #ifdef CONFIG_INET
721         if (skb_shinfo(skb)->gso_size) {
722                 switch (be16_to_cpu(skb->protocol)) {
723                 case ETH_P_IP:
724                         bnx2x_gro_csum(bp, skb, bnx2x_gro_ip_csum);
725                         break;
726                 case ETH_P_IPV6:
727                         bnx2x_gro_csum(bp, skb, bnx2x_gro_ipv6_csum);
728                         break;
729                 default:
730                         BNX2X_ERR("Error: FW GRO supports only IPv4/IPv6, not 0x%04x\n",
731                                   be16_to_cpu(skb->protocol));
732                 }
733         }
734 #endif
735         skb_record_rx_queue(skb, fp->rx_queue);
736         napi_gro_receive(&fp->napi, skb);
737 }
738
739 static void bnx2x_tpa_stop(struct bnx2x *bp, struct bnx2x_fastpath *fp,
740                            struct bnx2x_agg_info *tpa_info,
741                            u16 pages,
742                            struct eth_end_agg_rx_cqe *cqe,
743                            u16 cqe_idx)
744 {
745         struct sw_rx_bd *rx_buf = &tpa_info->first_buf;
746         u8 pad = tpa_info->placement_offset;
747         u16 len = tpa_info->len_on_bd;
748         struct sk_buff *skb = NULL;
749         u8 *new_data, *data = rx_buf->data;
750         u8 old_tpa_state = tpa_info->tpa_state;
751
752         tpa_info->tpa_state = BNX2X_TPA_STOP;
753
754         /* If we there was an error during the handling of the TPA_START -
755          * drop this aggregation.
756          */
757         if (old_tpa_state == BNX2X_TPA_ERROR)
758                 goto drop;
759
760         /* Try to allocate the new data */
761         new_data = bnx2x_frag_alloc(fp, GFP_ATOMIC);
762         /* Unmap skb in the pool anyway, as we are going to change
763            pool entry status to BNX2X_TPA_STOP even if new skb allocation
764            fails. */
765         dma_unmap_single(&bp->pdev->dev, dma_unmap_addr(rx_buf, mapping),
766                          fp->rx_buf_size, DMA_FROM_DEVICE);
767         if (likely(new_data))
768                 skb = build_skb(data, fp->rx_frag_size);
769
770         if (likely(skb)) {
771 #ifdef BNX2X_STOP_ON_ERROR
772                 if (pad + len > fp->rx_buf_size) {
773                         BNX2X_ERR("skb_put is about to fail...  pad %d  len %d  rx_buf_size %d\n",
774                                   pad, len, fp->rx_buf_size);
775                         bnx2x_panic();
776                         return;
777                 }
778 #endif
779
780                 skb_reserve(skb, pad + NET_SKB_PAD);
781                 skb_put(skb, len);
782                 skb_set_hash(skb, tpa_info->rxhash, tpa_info->rxhash_type);
783
784                 skb->protocol = eth_type_trans(skb, bp->dev);
785                 skb->ip_summed = CHECKSUM_UNNECESSARY;
786
787                 if (!bnx2x_fill_frag_skb(bp, fp, tpa_info, pages,
788                                          skb, cqe, cqe_idx)) {
789                         if (tpa_info->parsing_flags & PARSING_FLAGS_VLAN)
790                                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), tpa_info->vlan_tag);
791                         bnx2x_gro_receive(bp, fp, skb);
792                 } else {
793                         DP(NETIF_MSG_RX_STATUS,
794                            "Failed to allocate new pages - dropping packet!\n");
795                         dev_kfree_skb_any(skb);
796                 }
797
798                 /* put new data in bin */
799                 rx_buf->data = new_data;
800
801                 return;
802         }
803         if (new_data)
804                 bnx2x_frag_free(fp, new_data);
805 drop:
806         /* drop the packet and keep the buffer in the bin */
807         DP(NETIF_MSG_RX_STATUS,
808            "Failed to allocate or map a new skb - dropping packet!\n");
809         bnx2x_fp_stats(bp, fp)->eth_q_stats.rx_skb_alloc_failed++;
810 }
811
812 static int bnx2x_alloc_rx_data(struct bnx2x *bp, struct bnx2x_fastpath *fp,
813                                u16 index, gfp_t gfp_mask)
814 {
815         u8 *data;
816         struct sw_rx_bd *rx_buf = &fp->rx_buf_ring[index];
817         struct eth_rx_bd *rx_bd = &fp->rx_desc_ring[index];
818         dma_addr_t mapping;
819
820         data = bnx2x_frag_alloc(fp, gfp_mask);
821         if (unlikely(data == NULL))
822                 return -ENOMEM;
823
824         mapping = dma_map_single(&bp->pdev->dev, data + NET_SKB_PAD,
825                                  fp->rx_buf_size,
826                                  DMA_FROM_DEVICE);
827         if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
828                 bnx2x_frag_free(fp, data);
829                 BNX2X_ERR("Can't map rx data\n");
830                 return -ENOMEM;
831         }
832
833         rx_buf->data = data;
834         dma_unmap_addr_set(rx_buf, mapping, mapping);
835
836         rx_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
837         rx_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
838
839         return 0;
840 }
841
842 static
843 void bnx2x_csum_validate(struct sk_buff *skb, union eth_rx_cqe *cqe,
844                                  struct bnx2x_fastpath *fp,
845                                  struct bnx2x_eth_q_stats *qstats)
846 {
847         /* Do nothing if no L4 csum validation was done.
848          * We do not check whether IP csum was validated. For IPv4 we assume
849          * that if the card got as far as validating the L4 csum, it also
850          * validated the IP csum. IPv6 has no IP csum.
851          */
852         if (cqe->fast_path_cqe.status_flags &
853             ETH_FAST_PATH_RX_CQE_L4_XSUM_NO_VALIDATION_FLG)
854                 return;
855
856         /* If L4 validation was done, check if an error was found. */
857
858         if (cqe->fast_path_cqe.type_error_flags &
859             (ETH_FAST_PATH_RX_CQE_IP_BAD_XSUM_FLG |
860              ETH_FAST_PATH_RX_CQE_L4_BAD_XSUM_FLG))
861                 qstats->hw_csum_err++;
862         else
863                 skb->ip_summed = CHECKSUM_UNNECESSARY;
864 }
865
866 static int bnx2x_rx_int(struct bnx2x_fastpath *fp, int budget)
867 {
868         struct bnx2x *bp = fp->bp;
869         u16 bd_cons, bd_prod, bd_prod_fw, comp_ring_cons;
870         u16 sw_comp_cons, sw_comp_prod;
871         int rx_pkt = 0;
872         union eth_rx_cqe *cqe;
873         struct eth_fast_path_rx_cqe *cqe_fp;
874
875 #ifdef BNX2X_STOP_ON_ERROR
876         if (unlikely(bp->panic))
877                 return 0;
878 #endif
879         if (budget <= 0)
880                 return rx_pkt;
881
882         bd_cons = fp->rx_bd_cons;
883         bd_prod = fp->rx_bd_prod;
884         bd_prod_fw = bd_prod;
885         sw_comp_cons = fp->rx_comp_cons;
886         sw_comp_prod = fp->rx_comp_prod;
887
888         comp_ring_cons = RCQ_BD(sw_comp_cons);
889         cqe = &fp->rx_comp_ring[comp_ring_cons];
890         cqe_fp = &cqe->fast_path_cqe;
891
892         DP(NETIF_MSG_RX_STATUS,
893            "queue[%d]: sw_comp_cons %u\n", fp->index, sw_comp_cons);
894
895         while (BNX2X_IS_CQE_COMPLETED(cqe_fp)) {
896                 struct sw_rx_bd *rx_buf = NULL;
897                 struct sk_buff *skb;
898                 u8 cqe_fp_flags;
899                 enum eth_rx_cqe_type cqe_fp_type;
900                 u16 len, pad, queue;
901                 u8 *data;
902                 u32 rxhash;
903                 enum pkt_hash_types rxhash_type;
904
905 #ifdef BNX2X_STOP_ON_ERROR
906                 if (unlikely(bp->panic))
907                         return 0;
908 #endif
909
910                 bd_prod = RX_BD(bd_prod);
911                 bd_cons = RX_BD(bd_cons);
912
913                 /* A rmb() is required to ensure that the CQE is not read
914                  * before it is written by the adapter DMA.  PCI ordering
915                  * rules will make sure the other fields are written before
916                  * the marker at the end of struct eth_fast_path_rx_cqe
917                  * but without rmb() a weakly ordered processor can process
918                  * stale data.  Without the barrier TPA state-machine might
919                  * enter inconsistent state and kernel stack might be
920                  * provided with incorrect packet description - these lead
921                  * to various kernel crashed.
922                  */
923                 rmb();
924
925                 cqe_fp_flags = cqe_fp->type_error_flags;
926                 cqe_fp_type = cqe_fp_flags & ETH_FAST_PATH_RX_CQE_TYPE;
927
928                 DP(NETIF_MSG_RX_STATUS,
929                    "CQE type %x  err %x  status %x  queue %x  vlan %x  len %u\n",
930                    CQE_TYPE(cqe_fp_flags),
931                    cqe_fp_flags, cqe_fp->status_flags,
932                    le32_to_cpu(cqe_fp->rss_hash_result),
933                    le16_to_cpu(cqe_fp->vlan_tag),
934                    le16_to_cpu(cqe_fp->pkt_len_or_gro_seg_len));
935
936                 /* is this a slowpath msg? */
937                 if (unlikely(CQE_TYPE_SLOW(cqe_fp_type))) {
938                         bnx2x_sp_event(fp, cqe);
939                         goto next_cqe;
940                 }
941
942                 rx_buf = &fp->rx_buf_ring[bd_cons];
943                 data = rx_buf->data;
944
945                 if (!CQE_TYPE_FAST(cqe_fp_type)) {
946                         struct bnx2x_agg_info *tpa_info;
947                         u16 frag_size, pages;
948 #ifdef BNX2X_STOP_ON_ERROR
949                         /* sanity check */
950                         if (fp->disable_tpa &&
951                             (CQE_TYPE_START(cqe_fp_type) ||
952                              CQE_TYPE_STOP(cqe_fp_type)))
953                                 BNX2X_ERR("START/STOP packet while disable_tpa type %x\n",
954                                           CQE_TYPE(cqe_fp_type));
955 #endif
956
957                         if (CQE_TYPE_START(cqe_fp_type)) {
958                                 u16 queue = cqe_fp->queue_index;
959                                 DP(NETIF_MSG_RX_STATUS,
960                                    "calling tpa_start on queue %d\n",
961                                    queue);
962
963                                 bnx2x_tpa_start(fp, queue,
964                                                 bd_cons, bd_prod,
965                                                 cqe_fp);
966
967                                 goto next_rx;
968                         }
969                         queue = cqe->end_agg_cqe.queue_index;
970                         tpa_info = &fp->tpa_info[queue];
971                         DP(NETIF_MSG_RX_STATUS,
972                            "calling tpa_stop on queue %d\n",
973                            queue);
974
975                         frag_size = le16_to_cpu(cqe->end_agg_cqe.pkt_len) -
976                                     tpa_info->len_on_bd;
977
978                         if (fp->mode == TPA_MODE_GRO)
979                                 pages = (frag_size + tpa_info->full_page - 1) /
980                                          tpa_info->full_page;
981                         else
982                                 pages = SGE_PAGE_ALIGN(frag_size) >>
983                                         SGE_PAGE_SHIFT;
984
985                         bnx2x_tpa_stop(bp, fp, tpa_info, pages,
986                                        &cqe->end_agg_cqe, comp_ring_cons);
987 #ifdef BNX2X_STOP_ON_ERROR
988                         if (bp->panic)
989                                 return 0;
990 #endif
991
992                         bnx2x_update_sge_prod(fp, pages, &cqe->end_agg_cqe);
993                         goto next_cqe;
994                 }
995                 /* non TPA */
996                 len = le16_to_cpu(cqe_fp->pkt_len_or_gro_seg_len);
997                 pad = cqe_fp->placement_offset;
998                 dma_sync_single_for_cpu(&bp->pdev->dev,
999                                         dma_unmap_addr(rx_buf, mapping),
1000                                         pad + RX_COPY_THRESH,
1001                                         DMA_FROM_DEVICE);
1002                 pad += NET_SKB_PAD;
1003                 prefetch(data + pad); /* speedup eth_type_trans() */
1004                 /* is this an error packet? */
1005                 if (unlikely(cqe_fp_flags & ETH_RX_ERROR_FALGS)) {
1006                         DP(NETIF_MSG_RX_ERR | NETIF_MSG_RX_STATUS,
1007                            "ERROR  flags %x  rx packet %u\n",
1008                            cqe_fp_flags, sw_comp_cons);
1009                         bnx2x_fp_qstats(bp, fp)->rx_err_discard_pkt++;
1010                         goto reuse_rx;
1011                 }
1012
1013                 /* Since we don't have a jumbo ring
1014                  * copy small packets if mtu > 1500
1015                  */
1016                 if ((bp->dev->mtu > ETH_MAX_PACKET_SIZE) &&
1017                     (len <= RX_COPY_THRESH)) {
1018                         skb = napi_alloc_skb(&fp->napi, len);
1019                         if (skb == NULL) {
1020                                 DP(NETIF_MSG_RX_ERR | NETIF_MSG_RX_STATUS,
1021                                    "ERROR  packet dropped because of alloc failure\n");
1022                                 bnx2x_fp_qstats(bp, fp)->rx_skb_alloc_failed++;
1023                                 goto reuse_rx;
1024                         }
1025                         memcpy(skb->data, data + pad, len);
1026                         bnx2x_reuse_rx_data(fp, bd_cons, bd_prod);
1027                 } else {
1028                         if (likely(bnx2x_alloc_rx_data(bp, fp, bd_prod,
1029                                                        GFP_ATOMIC) == 0)) {
1030                                 dma_unmap_single(&bp->pdev->dev,
1031                                                  dma_unmap_addr(rx_buf, mapping),
1032                                                  fp->rx_buf_size,
1033                                                  DMA_FROM_DEVICE);
1034                                 skb = build_skb(data, fp->rx_frag_size);
1035                                 if (unlikely(!skb)) {
1036                                         bnx2x_frag_free(fp, data);
1037                                         bnx2x_fp_qstats(bp, fp)->
1038                                                         rx_skb_alloc_failed++;
1039                                         goto next_rx;
1040                                 }
1041                                 skb_reserve(skb, pad);
1042                         } else {
1043                                 DP(NETIF_MSG_RX_ERR | NETIF_MSG_RX_STATUS,
1044                                    "ERROR  packet dropped because of alloc failure\n");
1045                                 bnx2x_fp_qstats(bp, fp)->rx_skb_alloc_failed++;
1046 reuse_rx:
1047                                 bnx2x_reuse_rx_data(fp, bd_cons, bd_prod);
1048                                 goto next_rx;
1049                         }
1050                 }
1051
1052                 skb_put(skb, len);
1053                 skb->protocol = eth_type_trans(skb, bp->dev);
1054
1055                 /* Set Toeplitz hash for a none-LRO skb */
1056                 rxhash = bnx2x_get_rxhash(bp, cqe_fp, &rxhash_type);
1057                 skb_set_hash(skb, rxhash, rxhash_type);
1058
1059                 skb_checksum_none_assert(skb);
1060
1061                 if (bp->dev->features & NETIF_F_RXCSUM)
1062                         bnx2x_csum_validate(skb, cqe, fp,
1063                                             bnx2x_fp_qstats(bp, fp));
1064
1065                 skb_record_rx_queue(skb, fp->rx_queue);
1066
1067                 /* Check if this packet was timestamped */
1068                 if (unlikely(cqe->fast_path_cqe.type_error_flags &
1069                              (1 << ETH_FAST_PATH_RX_CQE_PTP_PKT_SHIFT)))
1070                         bnx2x_set_rx_ts(bp, skb);
1071
1072                 if (le16_to_cpu(cqe_fp->pars_flags.flags) &
1073                     PARSING_FLAGS_VLAN)
1074                         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
1075                                                le16_to_cpu(cqe_fp->vlan_tag));
1076
1077                 skb_mark_napi_id(skb, &fp->napi);
1078
1079                 if (bnx2x_fp_ll_polling(fp))
1080                         netif_receive_skb(skb);
1081                 else
1082                         napi_gro_receive(&fp->napi, skb);
1083 next_rx:
1084                 rx_buf->data = NULL;
1085
1086                 bd_cons = NEXT_RX_IDX(bd_cons);
1087                 bd_prod = NEXT_RX_IDX(bd_prod);
1088                 bd_prod_fw = NEXT_RX_IDX(bd_prod_fw);
1089                 rx_pkt++;
1090 next_cqe:
1091                 sw_comp_prod = NEXT_RCQ_IDX(sw_comp_prod);
1092                 sw_comp_cons = NEXT_RCQ_IDX(sw_comp_cons);
1093
1094                 /* mark CQE as free */
1095                 BNX2X_SEED_CQE(cqe_fp);
1096
1097                 if (rx_pkt == budget)
1098                         break;
1099
1100                 comp_ring_cons = RCQ_BD(sw_comp_cons);
1101                 cqe = &fp->rx_comp_ring[comp_ring_cons];
1102                 cqe_fp = &cqe->fast_path_cqe;
1103         } /* while */
1104
1105         fp->rx_bd_cons = bd_cons;
1106         fp->rx_bd_prod = bd_prod_fw;
1107         fp->rx_comp_cons = sw_comp_cons;
1108         fp->rx_comp_prod = sw_comp_prod;
1109
1110         /* Update producers */
1111         bnx2x_update_rx_prod(bp, fp, bd_prod_fw, sw_comp_prod,
1112                              fp->rx_sge_prod);
1113
1114         fp->rx_pkt += rx_pkt;
1115         fp->rx_calls++;
1116
1117         return rx_pkt;
1118 }
1119
1120 static irqreturn_t bnx2x_msix_fp_int(int irq, void *fp_cookie)
1121 {
1122         struct bnx2x_fastpath *fp = fp_cookie;
1123         struct bnx2x *bp = fp->bp;
1124         u8 cos;
1125
1126         DP(NETIF_MSG_INTR,
1127            "got an MSI-X interrupt on IDX:SB [fp %d fw_sd %d igusb %d]\n",
1128            fp->index, fp->fw_sb_id, fp->igu_sb_id);
1129
1130         bnx2x_ack_sb(bp, fp->igu_sb_id, USTORM_ID, 0, IGU_INT_DISABLE, 0);
1131
1132 #ifdef BNX2X_STOP_ON_ERROR
1133         if (unlikely(bp->panic))
1134                 return IRQ_HANDLED;
1135 #endif
1136
1137         /* Handle Rx and Tx according to MSI-X vector */
1138         for_each_cos_in_tx_queue(fp, cos)
1139                 prefetch(fp->txdata_ptr[cos]->tx_cons_sb);
1140
1141         prefetch(&fp->sb_running_index[SM_RX_ID]);
1142         napi_schedule_irqoff(&bnx2x_fp(bp, fp->index, napi));
1143
1144         return IRQ_HANDLED;
1145 }
1146
1147 /* HW Lock for shared dual port PHYs */
1148 void bnx2x_acquire_phy_lock(struct bnx2x *bp)
1149 {
1150         mutex_lock(&bp->port.phy_mutex);
1151
1152         bnx2x_acquire_hw_lock(bp, HW_LOCK_RESOURCE_MDIO);
1153 }
1154
1155 void bnx2x_release_phy_lock(struct bnx2x *bp)
1156 {
1157         bnx2x_release_hw_lock(bp, HW_LOCK_RESOURCE_MDIO);
1158
1159         mutex_unlock(&bp->port.phy_mutex);
1160 }
1161
1162 /* calculates MF speed according to current linespeed and MF configuration */
1163 u16 bnx2x_get_mf_speed(struct bnx2x *bp)
1164 {
1165         u16 line_speed = bp->link_vars.line_speed;
1166         if (IS_MF(bp)) {
1167                 u16 maxCfg = bnx2x_extract_max_cfg(bp,
1168                                                    bp->mf_config[BP_VN(bp)]);
1169
1170                 /* Calculate the current MAX line speed limit for the MF
1171                  * devices
1172                  */
1173                 if (IS_MF_SI(bp))
1174                         line_speed = (line_speed * maxCfg) / 100;
1175                 else { /* SD mode */
1176                         u16 vn_max_rate = maxCfg * 100;
1177
1178                         if (vn_max_rate < line_speed)
1179                                 line_speed = vn_max_rate;
1180                 }
1181         }
1182
1183         return line_speed;
1184 }
1185
1186 /**
1187  * bnx2x_fill_report_data - fill link report data to report
1188  *
1189  * @bp:         driver handle
1190  * @data:       link state to update
1191  *
1192  * It uses a none-atomic bit operations because is called under the mutex.
1193  */
1194 static void bnx2x_fill_report_data(struct bnx2x *bp,
1195                                    struct bnx2x_link_report_data *data)
1196 {
1197         memset(data, 0, sizeof(*data));
1198
1199         if (IS_PF(bp)) {
1200                 /* Fill the report data: effective line speed */
1201                 data->line_speed = bnx2x_get_mf_speed(bp);
1202
1203                 /* Link is down */
1204                 if (!bp->link_vars.link_up || (bp->flags & MF_FUNC_DIS))
1205                         __set_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1206                                   &data->link_report_flags);
1207
1208                 if (!BNX2X_NUM_ETH_QUEUES(bp))
1209                         __set_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1210                                   &data->link_report_flags);
1211
1212                 /* Full DUPLEX */
1213                 if (bp->link_vars.duplex == DUPLEX_FULL)
1214                         __set_bit(BNX2X_LINK_REPORT_FD,
1215                                   &data->link_report_flags);
1216
1217                 /* Rx Flow Control is ON */
1218                 if (bp->link_vars.flow_ctrl & BNX2X_FLOW_CTRL_RX)
1219                         __set_bit(BNX2X_LINK_REPORT_RX_FC_ON,
1220                                   &data->link_report_flags);
1221
1222                 /* Tx Flow Control is ON */
1223                 if (bp->link_vars.flow_ctrl & BNX2X_FLOW_CTRL_TX)
1224                         __set_bit(BNX2X_LINK_REPORT_TX_FC_ON,
1225                                   &data->link_report_flags);
1226         } else { /* VF */
1227                 *data = bp->vf_link_vars;
1228         }
1229 }
1230
1231 /**
1232  * bnx2x_link_report - report link status to OS.
1233  *
1234  * @bp:         driver handle
1235  *
1236  * Calls the __bnx2x_link_report() under the same locking scheme
1237  * as a link/PHY state managing code to ensure a consistent link
1238  * reporting.
1239  */
1240
1241 void bnx2x_link_report(struct bnx2x *bp)
1242 {
1243         bnx2x_acquire_phy_lock(bp);
1244         __bnx2x_link_report(bp);
1245         bnx2x_release_phy_lock(bp);
1246 }
1247
1248 /**
1249  * __bnx2x_link_report - report link status to OS.
1250  *
1251  * @bp:         driver handle
1252  *
1253  * None atomic implementation.
1254  * Should be called under the phy_lock.
1255  */
1256 void __bnx2x_link_report(struct bnx2x *bp)
1257 {
1258         struct bnx2x_link_report_data cur_data;
1259
1260         /* reread mf_cfg */
1261         if (IS_PF(bp) && !CHIP_IS_E1(bp))
1262                 bnx2x_read_mf_cfg(bp);
1263
1264         /* Read the current link report info */
1265         bnx2x_fill_report_data(bp, &cur_data);
1266
1267         /* Don't report link down or exactly the same link status twice */
1268         if (!memcmp(&cur_data, &bp->last_reported_link, sizeof(cur_data)) ||
1269             (test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1270                       &bp->last_reported_link.link_report_flags) &&
1271              test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1272                       &cur_data.link_report_flags)))
1273                 return;
1274
1275         bp->link_cnt++;
1276
1277         /* We are going to report a new link parameters now -
1278          * remember the current data for the next time.
1279          */
1280         memcpy(&bp->last_reported_link, &cur_data, sizeof(cur_data));
1281
1282         /* propagate status to VFs */
1283         if (IS_PF(bp))
1284                 bnx2x_iov_link_update(bp);
1285
1286         if (test_bit(BNX2X_LINK_REPORT_LINK_DOWN,
1287                      &cur_data.link_report_flags)) {
1288                 netif_carrier_off(bp->dev);
1289                 netdev_err(bp->dev, "NIC Link is Down\n");
1290                 return;
1291         } else {
1292                 const char *duplex;
1293                 const char *flow;
1294
1295                 netif_carrier_on(bp->dev);
1296
1297                 if (test_and_clear_bit(BNX2X_LINK_REPORT_FD,
1298                                        &cur_data.link_report_flags))
1299                         duplex = "full";
1300                 else
1301                         duplex = "half";
1302
1303                 /* Handle the FC at the end so that only these flags would be
1304                  * possibly set. This way we may easily check if there is no FC
1305                  * enabled.
1306                  */
1307                 if (cur_data.link_report_flags) {
1308                         if (test_bit(BNX2X_LINK_REPORT_RX_FC_ON,
1309                                      &cur_data.link_report_flags)) {
1310                                 if (test_bit(BNX2X_LINK_REPORT_TX_FC_ON,
1311                                      &cur_data.link_report_flags))
1312                                         flow = "ON - receive & transmit";
1313                                 else
1314                                         flow = "ON - receive";
1315                         } else {
1316                                 flow = "ON - transmit";
1317                         }
1318                 } else {
1319                         flow = "none";
1320                 }
1321                 netdev_info(bp->dev, "NIC Link is Up, %d Mbps %s duplex, Flow control: %s\n",
1322                             cur_data.line_speed, duplex, flow);
1323         }
1324 }
1325
1326 static void bnx2x_set_next_page_sgl(struct bnx2x_fastpath *fp)
1327 {
1328         int i;
1329
1330         for (i = 1; i <= NUM_RX_SGE_PAGES; i++) {
1331                 struct eth_rx_sge *sge;
1332
1333                 sge = &fp->rx_sge_ring[RX_SGE_CNT * i - 2];
1334                 sge->addr_hi =
1335                         cpu_to_le32(U64_HI(fp->rx_sge_mapping +
1336                         BCM_PAGE_SIZE*(i % NUM_RX_SGE_PAGES)));
1337
1338                 sge->addr_lo =
1339                         cpu_to_le32(U64_LO(fp->rx_sge_mapping +
1340                         BCM_PAGE_SIZE*(i % NUM_RX_SGE_PAGES)));
1341         }
1342 }
1343
1344 static void bnx2x_free_tpa_pool(struct bnx2x *bp,
1345                                 struct bnx2x_fastpath *fp, int last)
1346 {
1347         int i;
1348
1349         for (i = 0; i < last; i++) {
1350                 struct bnx2x_agg_info *tpa_info = &fp->tpa_info[i];
1351                 struct sw_rx_bd *first_buf = &tpa_info->first_buf;
1352                 u8 *data = first_buf->data;
1353
1354                 if (data == NULL) {
1355                         DP(NETIF_MSG_IFDOWN, "tpa bin %d empty on free\n", i);
1356                         continue;
1357                 }
1358                 if (tpa_info->tpa_state == BNX2X_TPA_START)
1359                         dma_unmap_single(&bp->pdev->dev,
1360                                          dma_unmap_addr(first_buf, mapping),
1361                                          fp->rx_buf_size, DMA_FROM_DEVICE);
1362                 bnx2x_frag_free(fp, data);
1363                 first_buf->data = NULL;
1364         }
1365 }
1366
1367 void bnx2x_init_rx_rings_cnic(struct bnx2x *bp)
1368 {
1369         int j;
1370
1371         for_each_rx_queue_cnic(bp, j) {
1372                 struct bnx2x_fastpath *fp = &bp->fp[j];
1373
1374                 fp->rx_bd_cons = 0;
1375
1376                 /* Activate BD ring */
1377                 /* Warning!
1378                  * this will generate an interrupt (to the TSTORM)
1379                  * must only be done after chip is initialized
1380                  */
1381                 bnx2x_update_rx_prod(bp, fp, fp->rx_bd_prod, fp->rx_comp_prod,
1382                                      fp->rx_sge_prod);
1383         }
1384 }
1385
1386 void bnx2x_init_rx_rings(struct bnx2x *bp)
1387 {
1388         int func = BP_FUNC(bp);
1389         u16 ring_prod;
1390         int i, j;
1391
1392         /* Allocate TPA resources */
1393         for_each_eth_queue(bp, j) {
1394                 struct bnx2x_fastpath *fp = &bp->fp[j];
1395
1396                 DP(NETIF_MSG_IFUP,
1397                    "mtu %d  rx_buf_size %d\n", bp->dev->mtu, fp->rx_buf_size);
1398
1399                 if (!fp->disable_tpa) {
1400                         /* Fill the per-aggregation pool */
1401                         for (i = 0; i < MAX_AGG_QS(bp); i++) {
1402                                 struct bnx2x_agg_info *tpa_info =
1403                                         &fp->tpa_info[i];
1404                                 struct sw_rx_bd *first_buf =
1405                                         &tpa_info->first_buf;
1406
1407                                 first_buf->data =
1408                                         bnx2x_frag_alloc(fp, GFP_KERNEL);
1409                                 if (!first_buf->data) {
1410                                         BNX2X_ERR("Failed to allocate TPA skb pool for queue[%d] - disabling TPA on this queue!\n",
1411                                                   j);
1412                                         bnx2x_free_tpa_pool(bp, fp, i);
1413                                         fp->disable_tpa = 1;
1414                                         break;
1415                                 }
1416                                 dma_unmap_addr_set(first_buf, mapping, 0);
1417                                 tpa_info->tpa_state = BNX2X_TPA_STOP;
1418                         }
1419
1420                         /* "next page" elements initialization */
1421                         bnx2x_set_next_page_sgl(fp);
1422
1423                         /* set SGEs bit mask */
1424                         bnx2x_init_sge_ring_bit_mask(fp);
1425
1426                         /* Allocate SGEs and initialize the ring elements */
1427                         for (i = 0, ring_prod = 0;
1428                              i < MAX_RX_SGE_CNT*NUM_RX_SGE_PAGES; i++) {
1429
1430                                 if (bnx2x_alloc_rx_sge(bp, fp, ring_prod,
1431                                                        GFP_KERNEL) < 0) {
1432                                         BNX2X_ERR("was only able to allocate %d rx sges\n",
1433                                                   i);
1434                                         BNX2X_ERR("disabling TPA for queue[%d]\n",
1435                                                   j);
1436                                         /* Cleanup already allocated elements */
1437                                         bnx2x_free_rx_sge_range(bp, fp,
1438                                                                 ring_prod);
1439                                         bnx2x_free_tpa_pool(bp, fp,
1440                                                             MAX_AGG_QS(bp));
1441                                         fp->disable_tpa = 1;
1442                                         ring_prod = 0;
1443                                         break;
1444                                 }
1445                                 ring_prod = NEXT_SGE_IDX(ring_prod);
1446                         }
1447
1448                         fp->rx_sge_prod = ring_prod;
1449                 }
1450         }
1451
1452         for_each_eth_queue(bp, j) {
1453                 struct bnx2x_fastpath *fp = &bp->fp[j];
1454
1455                 fp->rx_bd_cons = 0;
1456
1457                 /* Activate BD ring */
1458                 /* Warning!
1459                  * this will generate an interrupt (to the TSTORM)
1460                  * must only be done after chip is initialized
1461                  */
1462                 bnx2x_update_rx_prod(bp, fp, fp->rx_bd_prod, fp->rx_comp_prod,
1463                                      fp->rx_sge_prod);
1464
1465                 if (j != 0)
1466                         continue;
1467
1468                 if (CHIP_IS_E1(bp)) {
1469                         REG_WR(bp, BAR_USTRORM_INTMEM +
1470                                USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func),
1471                                U64_LO(fp->rx_comp_mapping));
1472                         REG_WR(bp, BAR_USTRORM_INTMEM +
1473                                USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func) + 4,
1474                                U64_HI(fp->rx_comp_mapping));
1475                 }
1476         }
1477 }
1478
1479 static void bnx2x_free_tx_skbs_queue(struct bnx2x_fastpath *fp)
1480 {
1481         u8 cos;
1482         struct bnx2x *bp = fp->bp;
1483
1484         for_each_cos_in_tx_queue(fp, cos) {
1485                 struct bnx2x_fp_txdata *txdata = fp->txdata_ptr[cos];
1486                 unsigned pkts_compl = 0, bytes_compl = 0;
1487
1488                 u16 sw_prod = txdata->tx_pkt_prod;
1489                 u16 sw_cons = txdata->tx_pkt_cons;
1490
1491                 while (sw_cons != sw_prod) {
1492                         bnx2x_free_tx_pkt(bp, txdata, TX_BD(sw_cons),
1493                                           &pkts_compl, &bytes_compl);
1494                         sw_cons++;
1495                 }
1496
1497                 netdev_tx_reset_queue(
1498                         netdev_get_tx_queue(bp->dev,
1499                                             txdata->txq_index));
1500         }
1501 }
1502
1503 static void bnx2x_free_tx_skbs_cnic(struct bnx2x *bp)
1504 {
1505         int i;
1506
1507         for_each_tx_queue_cnic(bp, i) {
1508                 bnx2x_free_tx_skbs_queue(&bp->fp[i]);
1509         }
1510 }
1511
1512 static void bnx2x_free_tx_skbs(struct bnx2x *bp)
1513 {
1514         int i;
1515
1516         for_each_eth_queue(bp, i) {
1517                 bnx2x_free_tx_skbs_queue(&bp->fp[i]);
1518         }
1519 }
1520
1521 static void bnx2x_free_rx_bds(struct bnx2x_fastpath *fp)
1522 {
1523         struct bnx2x *bp = fp->bp;
1524         int i;
1525
1526         /* ring wasn't allocated */
1527         if (fp->rx_buf_ring == NULL)
1528                 return;
1529
1530         for (i = 0; i < NUM_RX_BD; i++) {
1531                 struct sw_rx_bd *rx_buf = &fp->rx_buf_ring[i];
1532                 u8 *data = rx_buf->data;
1533
1534                 if (data == NULL)
1535                         continue;
1536                 dma_unmap_single(&bp->pdev->dev,
1537                                  dma_unmap_addr(rx_buf, mapping),
1538                                  fp->rx_buf_size, DMA_FROM_DEVICE);
1539
1540                 rx_buf->data = NULL;
1541                 bnx2x_frag_free(fp, data);
1542         }
1543 }
1544
1545 static void bnx2x_free_rx_skbs_cnic(struct bnx2x *bp)
1546 {
1547         int j;
1548
1549         for_each_rx_queue_cnic(bp, j) {
1550                 bnx2x_free_rx_bds(&bp->fp[j]);
1551         }
1552 }
1553
1554 static void bnx2x_free_rx_skbs(struct bnx2x *bp)
1555 {
1556         int j;
1557
1558         for_each_eth_queue(bp, j) {
1559                 struct bnx2x_fastpath *fp = &bp->fp[j];
1560
1561                 bnx2x_free_rx_bds(fp);
1562
1563                 if (!fp->disable_tpa)
1564                         bnx2x_free_tpa_pool(bp, fp, MAX_AGG_QS(bp));
1565         }
1566 }
1567
1568 static void bnx2x_free_skbs_cnic(struct bnx2x *bp)
1569 {
1570         bnx2x_free_tx_skbs_cnic(bp);
1571         bnx2x_free_rx_skbs_cnic(bp);
1572 }
1573
1574 void bnx2x_free_skbs(struct bnx2x *bp)
1575 {
1576         bnx2x_free_tx_skbs(bp);
1577         bnx2x_free_rx_skbs(bp);
1578 }
1579
1580 void bnx2x_update_max_mf_config(struct bnx2x *bp, u32 value)
1581 {
1582         /* load old values */
1583         u32 mf_cfg = bp->mf_config[BP_VN(bp)];
1584
1585         if (value != bnx2x_extract_max_cfg(bp, mf_cfg)) {
1586                 /* leave all but MAX value */
1587                 mf_cfg &= ~FUNC_MF_CFG_MAX_BW_MASK;
1588
1589                 /* set new MAX value */
1590                 mf_cfg |= (value << FUNC_MF_CFG_MAX_BW_SHIFT)
1591                                 & FUNC_MF_CFG_MAX_BW_MASK;
1592
1593                 bnx2x_fw_command(bp, DRV_MSG_CODE_SET_MF_BW, mf_cfg);
1594         }
1595 }
1596
1597 /**
1598  * bnx2x_free_msix_irqs - free previously requested MSI-X IRQ vectors
1599  *
1600  * @bp:         driver handle
1601  * @nvecs:      number of vectors to be released
1602  */
1603 static void bnx2x_free_msix_irqs(struct bnx2x *bp, int nvecs)
1604 {
1605         int i, offset = 0;
1606
1607         if (nvecs == offset)
1608                 return;
1609
1610         /* VFs don't have a default SB */
1611         if (IS_PF(bp)) {
1612                 free_irq(bp->msix_table[offset].vector, bp->dev);
1613                 DP(NETIF_MSG_IFDOWN, "released sp irq (%d)\n",
1614                    bp->msix_table[offset].vector);
1615                 offset++;
1616         }
1617
1618         if (CNIC_SUPPORT(bp)) {
1619                 if (nvecs == offset)
1620                         return;
1621                 offset++;
1622         }
1623
1624         for_each_eth_queue(bp, i) {
1625                 if (nvecs == offset)
1626                         return;
1627                 DP(NETIF_MSG_IFDOWN, "about to release fp #%d->%d irq\n",
1628                    i, bp->msix_table[offset].vector);
1629
1630                 free_irq(bp->msix_table[offset++].vector, &bp->fp[i]);
1631         }
1632 }
1633
1634 void bnx2x_free_irq(struct bnx2x *bp)
1635 {
1636         if (bp->flags & USING_MSIX_FLAG &&
1637             !(bp->flags & USING_SINGLE_MSIX_FLAG)) {
1638                 int nvecs = BNX2X_NUM_ETH_QUEUES(bp) + CNIC_SUPPORT(bp);
1639
1640                 /* vfs don't have a default status block */
1641                 if (IS_PF(bp))
1642                         nvecs++;
1643
1644                 bnx2x_free_msix_irqs(bp, nvecs);
1645         } else {
1646                 free_irq(bp->dev->irq, bp->dev);
1647         }
1648 }
1649
1650 int bnx2x_enable_msix(struct bnx2x *bp)
1651 {
1652         int msix_vec = 0, i, rc;
1653
1654         /* VFs don't have a default status block */
1655         if (IS_PF(bp)) {
1656                 bp->msix_table[msix_vec].entry = msix_vec;
1657                 BNX2X_DEV_INFO("msix_table[0].entry = %d (slowpath)\n",
1658                                bp->msix_table[0].entry);
1659                 msix_vec++;
1660         }
1661
1662         /* Cnic requires an msix vector for itself */
1663         if (CNIC_SUPPORT(bp)) {
1664                 bp->msix_table[msix_vec].entry = msix_vec;
1665                 BNX2X_DEV_INFO("msix_table[%d].entry = %d (CNIC)\n",
1666                                msix_vec, bp->msix_table[msix_vec].entry);
1667                 msix_vec++;
1668         }
1669
1670         /* We need separate vectors for ETH queues only (not FCoE) */
1671         for_each_eth_queue(bp, i) {
1672                 bp->msix_table[msix_vec].entry = msix_vec;
1673                 BNX2X_DEV_INFO("msix_table[%d].entry = %d (fastpath #%u)\n",
1674                                msix_vec, msix_vec, i);
1675                 msix_vec++;
1676         }
1677
1678         DP(BNX2X_MSG_SP, "about to request enable msix with %d vectors\n",
1679            msix_vec);
1680
1681         rc = pci_enable_msix_range(bp->pdev, &bp->msix_table[0],
1682                                    BNX2X_MIN_MSIX_VEC_CNT(bp), msix_vec);
1683         /*
1684          * reconfigure number of tx/rx queues according to available
1685          * MSI-X vectors
1686          */
1687         if (rc == -ENOSPC) {
1688                 /* Get by with single vector */
1689                 rc = pci_enable_msix_range(bp->pdev, &bp->msix_table[0], 1, 1);
1690                 if (rc < 0) {
1691                         BNX2X_DEV_INFO("Single MSI-X is not attainable rc %d\n",
1692                                        rc);
1693                         goto no_msix;
1694                 }
1695
1696                 BNX2X_DEV_INFO("Using single MSI-X vector\n");
1697                 bp->flags |= USING_SINGLE_MSIX_FLAG;
1698
1699                 BNX2X_DEV_INFO("set number of queues to 1\n");
1700                 bp->num_ethernet_queues = 1;
1701                 bp->num_queues = bp->num_ethernet_queues + bp->num_cnic_queues;
1702         } else if (rc < 0) {
1703                 BNX2X_DEV_INFO("MSI-X is not attainable rc %d\n", rc);
1704                 goto no_msix;
1705         } else if (rc < msix_vec) {
1706                 /* how less vectors we will have? */
1707                 int diff = msix_vec - rc;
1708
1709                 BNX2X_DEV_INFO("Trying to use less MSI-X vectors: %d\n", rc);
1710
1711                 /*
1712                  * decrease number of queues by number of unallocated entries
1713                  */
1714                 bp->num_ethernet_queues -= diff;
1715                 bp->num_queues = bp->num_ethernet_queues + bp->num_cnic_queues;
1716
1717                 BNX2X_DEV_INFO("New queue configuration set: %d\n",
1718                                bp->num_queues);
1719         }
1720
1721         bp->flags |= USING_MSIX_FLAG;
1722
1723         return 0;
1724
1725 no_msix:
1726         /* fall to INTx if not enough memory */
1727         if (rc == -ENOMEM)
1728                 bp->flags |= DISABLE_MSI_FLAG;
1729
1730         return rc;
1731 }
1732
1733 static int bnx2x_req_msix_irqs(struct bnx2x *bp)
1734 {
1735         int i, rc, offset = 0;
1736
1737         /* no default status block for vf */
1738         if (IS_PF(bp)) {
1739                 rc = request_irq(bp->msix_table[offset++].vector,
1740                                  bnx2x_msix_sp_int, 0,
1741                                  bp->dev->name, bp->dev);
1742                 if (rc) {
1743                         BNX2X_ERR("request sp irq failed\n");
1744                         return -EBUSY;
1745                 }
1746         }
1747
1748         if (CNIC_SUPPORT(bp))
1749                 offset++;
1750
1751         for_each_eth_queue(bp, i) {
1752                 struct bnx2x_fastpath *fp = &bp->fp[i];
1753                 snprintf(fp->name, sizeof(fp->name), "%s-fp-%d",
1754                          bp->dev->name, i);
1755
1756                 rc = request_irq(bp->msix_table[offset].vector,
1757                                  bnx2x_msix_fp_int, 0, fp->name, fp);
1758                 if (rc) {
1759                         BNX2X_ERR("request fp #%d irq (%d) failed  rc %d\n", i,
1760                               bp->msix_table[offset].vector, rc);
1761                         bnx2x_free_msix_irqs(bp, offset);
1762                         return -EBUSY;
1763                 }
1764
1765                 offset++;
1766         }
1767
1768         i = BNX2X_NUM_ETH_QUEUES(bp);
1769         if (IS_PF(bp)) {
1770                 offset = 1 + CNIC_SUPPORT(bp);
1771                 netdev_info(bp->dev,
1772                             "using MSI-X  IRQs: sp %d  fp[%d] %d ... fp[%d] %d\n",
1773                             bp->msix_table[0].vector,
1774                             0, bp->msix_table[offset].vector,
1775                             i - 1, bp->msix_table[offset + i - 1].vector);
1776         } else {
1777                 offset = CNIC_SUPPORT(bp);
1778                 netdev_info(bp->dev,
1779                             "using MSI-X  IRQs: fp[%d] %d ... fp[%d] %d\n",
1780                             0, bp->msix_table[offset].vector,
1781                             i - 1, bp->msix_table[offset + i - 1].vector);
1782         }
1783         return 0;
1784 }
1785
1786 int bnx2x_enable_msi(struct bnx2x *bp)
1787 {
1788         int rc;
1789
1790         rc = pci_enable_msi(bp->pdev);
1791         if (rc) {
1792                 BNX2X_DEV_INFO("MSI is not attainable\n");
1793                 return -1;
1794         }
1795         bp->flags |= USING_MSI_FLAG;
1796
1797         return 0;
1798 }
1799
1800 static int bnx2x_req_irq(struct bnx2x *bp)
1801 {
1802         unsigned long flags;
1803         unsigned int irq;
1804
1805         if (bp->flags & (USING_MSI_FLAG | USING_MSIX_FLAG))
1806                 flags = 0;
1807         else
1808                 flags = IRQF_SHARED;
1809
1810         if (bp->flags & USING_MSIX_FLAG)
1811                 irq = bp->msix_table[0].vector;
1812         else
1813                 irq = bp->pdev->irq;
1814
1815         return request_irq(irq, bnx2x_interrupt, flags, bp->dev->name, bp->dev);
1816 }
1817
1818 static int bnx2x_setup_irqs(struct bnx2x *bp)
1819 {
1820         int rc = 0;
1821         if (bp->flags & USING_MSIX_FLAG &&
1822             !(bp->flags & USING_SINGLE_MSIX_FLAG)) {
1823                 rc = bnx2x_req_msix_irqs(bp);
1824                 if (rc)
1825                         return rc;
1826         } else {
1827                 rc = bnx2x_req_irq(bp);
1828                 if (rc) {
1829                         BNX2X_ERR("IRQ request failed  rc %d, aborting\n", rc);
1830                         return rc;
1831                 }
1832                 if (bp->flags & USING_MSI_FLAG) {
1833                         bp->dev->irq = bp->pdev->irq;
1834                         netdev_info(bp->dev, "using MSI IRQ %d\n",
1835                                     bp->dev->irq);
1836                 }
1837                 if (bp->flags & USING_MSIX_FLAG) {
1838                         bp->dev->irq = bp->msix_table[0].vector;
1839                         netdev_info(bp->dev, "using MSIX IRQ %d\n",
1840                                     bp->dev->irq);
1841                 }
1842         }
1843
1844         return 0;
1845 }
1846
1847 static void bnx2x_napi_enable_cnic(struct bnx2x *bp)
1848 {
1849         int i;
1850
1851         for_each_rx_queue_cnic(bp, i) {
1852                 bnx2x_fp_init_lock(&bp->fp[i]);
1853                 napi_enable(&bnx2x_fp(bp, i, napi));
1854         }
1855 }
1856
1857 static void bnx2x_napi_enable(struct bnx2x *bp)
1858 {
1859         int i;
1860
1861         for_each_eth_queue(bp, i) {
1862                 bnx2x_fp_init_lock(&bp->fp[i]);
1863                 napi_enable(&bnx2x_fp(bp, i, napi));
1864         }
1865 }
1866
1867 static void bnx2x_napi_disable_cnic(struct bnx2x *bp)
1868 {
1869         int i;
1870
1871         for_each_rx_queue_cnic(bp, i) {
1872                 napi_disable(&bnx2x_fp(bp, i, napi));
1873                 while (!bnx2x_fp_ll_disable(&bp->fp[i]))
1874                         usleep_range(1000, 2000);
1875         }
1876 }
1877
1878 static void bnx2x_napi_disable(struct bnx2x *bp)
1879 {
1880         int i;
1881
1882         for_each_eth_queue(bp, i) {
1883                 napi_disable(&bnx2x_fp(bp, i, napi));
1884                 while (!bnx2x_fp_ll_disable(&bp->fp[i]))
1885                         usleep_range(1000, 2000);
1886         }
1887 }
1888
1889 void bnx2x_netif_start(struct bnx2x *bp)
1890 {
1891         if (netif_running(bp->dev)) {
1892                 bnx2x_napi_enable(bp);
1893                 if (CNIC_LOADED(bp))
1894                         bnx2x_napi_enable_cnic(bp);
1895                 bnx2x_int_enable(bp);
1896                 if (bp->state == BNX2X_STATE_OPEN)
1897                         netif_tx_wake_all_queues(bp->dev);
1898         }
1899 }
1900
1901 void bnx2x_netif_stop(struct bnx2x *bp, int disable_hw)
1902 {
1903         bnx2x_int_disable_sync(bp, disable_hw);
1904         bnx2x_napi_disable(bp);
1905         if (CNIC_LOADED(bp))
1906                 bnx2x_napi_disable_cnic(bp);
1907 }
1908
1909 u16 bnx2x_select_queue(struct net_device *dev, struct sk_buff *skb,
1910                        void *accel_priv, select_queue_fallback_t fallback)
1911 {
1912         struct bnx2x *bp = netdev_priv(dev);
1913
1914         if (CNIC_LOADED(bp) && !NO_FCOE(bp)) {
1915                 struct ethhdr *hdr = (struct ethhdr *)skb->data;
1916                 u16 ether_type = ntohs(hdr->h_proto);
1917
1918                 /* Skip VLAN tag if present */
1919                 if (ether_type == ETH_P_8021Q) {
1920                         struct vlan_ethhdr *vhdr =
1921                                 (struct vlan_ethhdr *)skb->data;
1922
1923                         ether_type = ntohs(vhdr->h_vlan_encapsulated_proto);
1924                 }
1925
1926                 /* If ethertype is FCoE or FIP - use FCoE ring */
1927                 if ((ether_type == ETH_P_FCOE) || (ether_type == ETH_P_FIP))
1928                         return bnx2x_fcoe_tx(bp, txq_index);
1929         }
1930
1931         /* select a non-FCoE queue */
1932         return fallback(dev, skb) % BNX2X_NUM_ETH_QUEUES(bp);
1933 }
1934
1935 void bnx2x_set_num_queues(struct bnx2x *bp)
1936 {
1937         /* RSS queues */
1938         bp->num_ethernet_queues = bnx2x_calc_num_queues(bp);
1939
1940         /* override in STORAGE SD modes */
1941         if (IS_MF_STORAGE_ONLY(bp))
1942                 bp->num_ethernet_queues = 1;
1943
1944         /* Add special queues */
1945         bp->num_cnic_queues = CNIC_SUPPORT(bp); /* For FCOE */
1946         bp->num_queues = bp->num_ethernet_queues + bp->num_cnic_queues;
1947
1948         BNX2X_DEV_INFO("set number of queues to %d\n", bp->num_queues);
1949 }
1950
1951 /**
1952  * bnx2x_set_real_num_queues - configure netdev->real_num_[tx,rx]_queues
1953  *
1954  * @bp:         Driver handle
1955  *
1956  * We currently support for at most 16 Tx queues for each CoS thus we will
1957  * allocate a multiple of 16 for ETH L2 rings according to the value of the
1958  * bp->max_cos.
1959  *
1960  * If there is an FCoE L2 queue the appropriate Tx queue will have the next
1961  * index after all ETH L2 indices.
1962  *
1963  * If the actual number of Tx queues (for each CoS) is less than 16 then there
1964  * will be the holes at the end of each group of 16 ETh L2 indices (0..15,
1965  * 16..31,...) with indices that are not coupled with any real Tx queue.
1966  *
1967  * The proper configuration of skb->queue_mapping is handled by
1968  * bnx2x_select_queue() and __skb_tx_hash().
1969  *
1970  * bnx2x_setup_tc() takes care of the proper TC mappings so that __skb_tx_hash()
1971  * will return a proper Tx index if TC is enabled (netdev->num_tc > 0).
1972  */
1973 static int bnx2x_set_real_num_queues(struct bnx2x *bp, int include_cnic)
1974 {
1975         int rc, tx, rx;
1976
1977         tx = BNX2X_NUM_ETH_QUEUES(bp) * bp->max_cos;
1978         rx = BNX2X_NUM_ETH_QUEUES(bp);
1979
1980 /* account for fcoe queue */
1981         if (include_cnic && !NO_FCOE(bp)) {
1982                 rx++;
1983                 tx++;
1984         }
1985
1986         rc = netif_set_real_num_tx_queues(bp->dev, tx);
1987         if (rc) {
1988                 BNX2X_ERR("Failed to set real number of Tx queues: %d\n", rc);
1989                 return rc;
1990         }
1991         rc = netif_set_real_num_rx_queues(bp->dev, rx);
1992         if (rc) {
1993                 BNX2X_ERR("Failed to set real number of Rx queues: %d\n", rc);
1994                 return rc;
1995         }
1996
1997         DP(NETIF_MSG_IFUP, "Setting real num queues to (tx, rx) (%d, %d)\n",
1998                           tx, rx);
1999
2000         return rc;
2001 }
2002
2003 static void bnx2x_set_rx_buf_size(struct bnx2x *bp)
2004 {
2005         int i;
2006
2007         for_each_queue(bp, i) {
2008                 struct bnx2x_fastpath *fp = &bp->fp[i];
2009                 u32 mtu;
2010
2011                 /* Always use a mini-jumbo MTU for the FCoE L2 ring */
2012                 if (IS_FCOE_IDX(i))
2013                         /*
2014                          * Although there are no IP frames expected to arrive to
2015                          * this ring we still want to add an
2016                          * IP_HEADER_ALIGNMENT_PADDING to prevent a buffer
2017                          * overrun attack.
2018                          */
2019                         mtu = BNX2X_FCOE_MINI_JUMBO_MTU;
2020                 else
2021                         mtu = bp->dev->mtu;
2022                 fp->rx_buf_size = BNX2X_FW_RX_ALIGN_START +
2023                                   IP_HEADER_ALIGNMENT_PADDING +
2024                                   ETH_OVREHEAD +
2025                                   mtu +
2026                                   BNX2X_FW_RX_ALIGN_END;
2027                 /* Note : rx_buf_size doesn't take into account NET_SKB_PAD */
2028                 if (fp->rx_buf_size + NET_SKB_PAD <= PAGE_SIZE)
2029                         fp->rx_frag_size = fp->rx_buf_size + NET_SKB_PAD;
2030                 else
2031                         fp->rx_frag_size = 0;
2032         }
2033 }
2034
2035 static int bnx2x_init_rss(struct bnx2x *bp)
2036 {
2037         int i;
2038         u8 num_eth_queues = BNX2X_NUM_ETH_QUEUES(bp);
2039
2040         /* Prepare the initial contents for the indirection table if RSS is
2041          * enabled
2042          */
2043         for (i = 0; i < sizeof(bp->rss_conf_obj.ind_table); i++)
2044                 bp->rss_conf_obj.ind_table[i] =
2045                         bp->fp->cl_id +
2046                         ethtool_rxfh_indir_default(i, num_eth_queues);
2047
2048         /*
2049          * For 57710 and 57711 SEARCHER configuration (rss_keys) is
2050          * per-port, so if explicit configuration is needed , do it only
2051          * for a PMF.
2052          *
2053          * For 57712 and newer on the other hand it's a per-function
2054          * configuration.
2055          */
2056         return bnx2x_config_rss_eth(bp, bp->port.pmf || !CHIP_IS_E1x(bp));
2057 }
2058
2059 int bnx2x_rss(struct bnx2x *bp, struct bnx2x_rss_config_obj *rss_obj,
2060               bool config_hash, bool enable)
2061 {
2062         struct bnx2x_config_rss_params params = {NULL};
2063
2064         /* Although RSS is meaningless when there is a single HW queue we
2065          * still need it enabled in order to have HW Rx hash generated.
2066          *
2067          * if (!is_eth_multi(bp))
2068          *      bp->multi_mode = ETH_RSS_MODE_DISABLED;
2069          */
2070
2071         params.rss_obj = rss_obj;
2072
2073         __set_bit(RAMROD_COMP_WAIT, &params.ramrod_flags);
2074
2075         if (enable) {
2076                 __set_bit(BNX2X_RSS_MODE_REGULAR, &params.rss_flags);
2077
2078                 /* RSS configuration */
2079                 __set_bit(BNX2X_RSS_IPV4, &params.rss_flags);
2080                 __set_bit(BNX2X_RSS_IPV4_TCP, &params.rss_flags);
2081                 __set_bit(BNX2X_RSS_IPV6, &params.rss_flags);
2082                 __set_bit(BNX2X_RSS_IPV6_TCP, &params.rss_flags);
2083                 if (rss_obj->udp_rss_v4)
2084                         __set_bit(BNX2X_RSS_IPV4_UDP, &params.rss_flags);
2085                 if (rss_obj->udp_rss_v6)
2086                         __set_bit(BNX2X_RSS_IPV6_UDP, &params.rss_flags);
2087
2088                 if (!CHIP_IS_E1x(bp))
2089                         /* valid only for TUNN_MODE_GRE tunnel mode */
2090                         __set_bit(BNX2X_RSS_GRE_INNER_HDRS, &params.rss_flags);
2091         } else {
2092                 __set_bit(BNX2X_RSS_MODE_DISABLED, &params.rss_flags);
2093         }
2094
2095         /* Hash bits */
2096         params.rss_result_mask = MULTI_MASK;
2097
2098         memcpy(params.ind_table, rss_obj->ind_table, sizeof(params.ind_table));
2099
2100         if (config_hash) {
2101                 /* RSS keys */
2102                 netdev_rss_key_fill(params.rss_key, T_ETH_RSS_KEY * 4);
2103                 __set_bit(BNX2X_RSS_SET_SRCH, &params.rss_flags);
2104         }
2105
2106         if (IS_PF(bp))
2107                 return bnx2x_config_rss(bp, &params);
2108         else
2109                 return bnx2x_vfpf_config_rss(bp, &params);
2110 }
2111
2112 static int bnx2x_init_hw(struct bnx2x *bp, u32 load_code)
2113 {
2114         struct bnx2x_func_state_params func_params = {NULL};
2115
2116         /* Prepare parameters for function state transitions */
2117         __set_bit(RAMROD_COMP_WAIT, &func_params.ramrod_flags);
2118
2119         func_params.f_obj = &bp->func_obj;
2120         func_params.cmd = BNX2X_F_CMD_HW_INIT;
2121
2122         func_params.params.hw_init.load_phase = load_code;
2123
2124         return bnx2x_func_state_change(bp, &func_params);
2125 }
2126
2127 /*
2128  * Cleans the object that have internal lists without sending
2129  * ramrods. Should be run when interrupts are disabled.
2130  */
2131 void bnx2x_squeeze_objects(struct bnx2x *bp)
2132 {
2133         int rc;
2134         unsigned long ramrod_flags = 0, vlan_mac_flags = 0;
2135         struct bnx2x_mcast_ramrod_params rparam = {NULL};
2136         struct bnx2x_vlan_mac_obj *mac_obj = &bp->sp_objs->mac_obj;
2137
2138         /***************** Cleanup MACs' object first *************************/
2139
2140         /* Wait for completion of requested */
2141         __set_bit(RAMROD_COMP_WAIT, &ramrod_flags);
2142         /* Perform a dry cleanup */
2143         __set_bit(RAMROD_DRV_CLR_ONLY, &ramrod_flags);
2144
2145         /* Clean ETH primary MAC */
2146         __set_bit(BNX2X_ETH_MAC, &vlan_mac_flags);
2147         rc = mac_obj->delete_all(bp, &bp->sp_objs->mac_obj, &vlan_mac_flags,
2148                                  &ramrod_flags);
2149         if (rc != 0)
2150                 BNX2X_ERR("Failed to clean ETH MACs: %d\n", rc);
2151
2152         /* Cleanup UC list */
2153         vlan_mac_flags = 0;
2154         __set_bit(BNX2X_UC_LIST_MAC, &vlan_mac_flags);
2155         rc = mac_obj->delete_all(bp, mac_obj, &vlan_mac_flags,
2156                                  &ramrod_flags);
2157         if (rc != 0)
2158                 BNX2X_ERR("Failed to clean UC list MACs: %d\n", rc);
2159
2160         /***************** Now clean mcast object *****************************/
2161         rparam.mcast_obj = &bp->mcast_obj;
2162         __set_bit(RAMROD_DRV_CLR_ONLY, &rparam.ramrod_flags);
2163
2164         /* Add a DEL command... - Since we're doing a driver cleanup only,
2165          * we take a lock surrounding both the initial send and the CONTs,
2166          * as we don't want a true completion to disrupt us in the middle.
2167          */
2168         netif_addr_lock_bh(bp->dev);
2169         rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_DEL);
2170         if (rc < 0)
2171                 BNX2X_ERR("Failed to add a new DEL command to a multi-cast object: %d\n",
2172                           rc);
2173
2174         /* ...and wait until all pending commands are cleared */
2175         rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_CONT);
2176         while (rc != 0) {
2177                 if (rc < 0) {
2178                         BNX2X_ERR("Failed to clean multi-cast object: %d\n",
2179                                   rc);
2180                         netif_addr_unlock_bh(bp->dev);
2181                         return;
2182                 }
2183
2184                 rc = bnx2x_config_mcast(bp, &rparam, BNX2X_MCAST_CMD_CONT);
2185         }
2186         netif_addr_unlock_bh(bp->dev);
2187 }
2188
2189 #ifndef BNX2X_STOP_ON_ERROR
2190 #define LOAD_ERROR_EXIT(bp, label) \
2191         do { \
2192                 (bp)->state = BNX2X_STATE_ERROR; \
2193                 goto label; \
2194         } while (0)
2195
2196 #define LOAD_ERROR_EXIT_CNIC(bp, label) \
2197         do { \
2198                 bp->cnic_loaded = false; \
2199                 goto label; \
2200         } while (0)
2201 #else /*BNX2X_STOP_ON_ERROR*/
2202 #define LOAD_ERROR_EXIT(bp, label) \
2203         do { \
2204                 (bp)->state = BNX2X_STATE_ERROR; \
2205                 (bp)->panic = 1; \
2206                 return -EBUSY; \
2207         } while (0)
2208 #define LOAD_ERROR_EXIT_CNIC(bp, label) \
2209         do { \
2210                 bp->cnic_loaded = false; \
2211                 (bp)->panic = 1; \
2212                 return -EBUSY; \
2213         } while (0)
2214 #endif /*BNX2X_STOP_ON_ERROR*/
2215
2216 static void bnx2x_free_fw_stats_mem(struct bnx2x *bp)
2217 {
2218         BNX2X_PCI_FREE(bp->fw_stats, bp->fw_stats_mapping,
2219                        bp->fw_stats_data_sz + bp->fw_stats_req_sz);
2220         return;
2221 }
2222
2223 static int bnx2x_alloc_fw_stats_mem(struct bnx2x *bp)
2224 {
2225         int num_groups, vf_headroom = 0;
2226         int is_fcoe_stats = NO_FCOE(bp) ? 0 : 1;
2227
2228         /* number of queues for statistics is number of eth queues + FCoE */
2229         u8 num_queue_stats = BNX2X_NUM_ETH_QUEUES(bp) + is_fcoe_stats;
2230
2231         /* Total number of FW statistics requests =
2232          * 1 for port stats + 1 for PF stats + potential 2 for FCoE (fcoe proper
2233          * and fcoe l2 queue) stats + num of queues (which includes another 1
2234          * for fcoe l2 queue if applicable)
2235          */
2236         bp->fw_stats_num = 2 + is_fcoe_stats + num_queue_stats;
2237
2238         /* vf stats appear in the request list, but their data is allocated by
2239          * the VFs themselves. We don't include them in the bp->fw_stats_num as
2240          * it is used to determine where to place the vf stats queries in the
2241          * request struct
2242          */
2243         if (IS_SRIOV(bp))
2244                 vf_headroom = bnx2x_vf_headroom(bp);
2245
2246         /* Request is built from stats_query_header and an array of
2247          * stats_query_cmd_group each of which contains
2248          * STATS_QUERY_CMD_COUNT rules. The real number or requests is
2249          * configured in the stats_query_header.
2250          */
2251         num_groups =
2252                 (((bp->fw_stats_num + vf_headroom) / STATS_QUERY_CMD_COUNT) +
2253                  (((bp->fw_stats_num + vf_headroom) % STATS_QUERY_CMD_COUNT) ?
2254                  1 : 0));
2255
2256         DP(BNX2X_MSG_SP, "stats fw_stats_num %d, vf headroom %d, num_groups %d\n",
2257            bp->fw_stats_num, vf_headroom, num_groups);
2258         bp->fw_stats_req_sz = sizeof(struct stats_query_header) +
2259                 num_groups * sizeof(struct stats_query_cmd_group);
2260
2261         /* Data for statistics requests + stats_counter
2262          * stats_counter holds per-STORM counters that are incremented
2263          * when STORM has finished with the current request.
2264          * memory for FCoE offloaded statistics are counted anyway,
2265          * even if they will not be sent.
2266          * VF stats are not accounted for here as the data of VF stats is stored
2267          * in memory allocated by the VF, not here.
2268          */
2269         bp->fw_stats_data_sz = sizeof(struct per_port_stats) +
2270                 sizeof(struct per_pf_stats) +
2271                 sizeof(struct fcoe_statistics_params) +
2272                 sizeof(struct per_queue_stats) * num_queue_stats +
2273                 sizeof(struct stats_counter);
2274
2275         bp->fw_stats = BNX2X_PCI_ALLOC(&bp->fw_stats_mapping,
2276                                        bp->fw_stats_data_sz + bp->fw_stats_req_sz);
2277         if (!bp->fw_stats)
2278                 goto alloc_mem_err;
2279
2280         /* Set shortcuts */
2281         bp->fw_stats_req = (struct bnx2x_fw_stats_req *)bp->fw_stats;
2282         bp->fw_stats_req_mapping = bp->fw_stats_mapping;
2283         bp->fw_stats_data = (struct bnx2x_fw_stats_data *)
2284                 ((u8 *)bp->fw_stats + bp->fw_stats_req_sz);
2285         bp->fw_stats_data_mapping = bp->fw_stats_mapping +
2286                 bp->fw_stats_req_sz;
2287
2288         DP(BNX2X_MSG_SP, "statistics request base address set to %x %x\n",
2289            U64_HI(bp->fw_stats_req_mapping),
2290            U64_LO(bp->fw_stats_req_mapping));
2291         DP(BNX2X_MSG_SP, "statistics data base address set to %x %x\n",
2292            U64_HI(bp->fw_stats_data_mapping),
2293            U64_LO(bp->fw_stats_data_mapping));
2294         return 0;
2295
2296 alloc_mem_err:
2297         bnx2x_free_fw_stats_mem(bp);
2298         BNX2X_ERR("Can't allocate FW stats memory\n");
2299         return -ENOMEM;
2300 }
2301
2302 /* send load request to mcp and analyze response */
2303 static int bnx2x_nic_load_request(struct bnx2x *bp, u32 *load_code)
2304 {
2305         u32 param;
2306
2307         /* init fw_seq */
2308         bp->fw_seq =
2309                 (SHMEM_RD(bp, func_mb[BP_FW_MB_IDX(bp)].drv_mb_header) &
2310                  DRV_MSG_SEQ_NUMBER_MASK);
2311         BNX2X_DEV_INFO("fw_seq 0x%08x\n", bp->fw_seq);
2312
2313         /* Get current FW pulse sequence */
2314         bp->fw_drv_pulse_wr_seq =
2315                 (SHMEM_RD(bp, func_mb[BP_FW_MB_IDX(bp)].drv_pulse_mb) &
2316                  DRV_PULSE_SEQ_MASK);
2317         BNX2X_DEV_INFO("drv_pulse 0x%x\n", bp->fw_drv_pulse_wr_seq);
2318
2319         param = DRV_MSG_CODE_LOAD_REQ_WITH_LFA;
2320
2321         if (IS_MF_SD(bp) && bnx2x_port_after_undi(bp))
2322                 param |= DRV_MSG_CODE_LOAD_REQ_FORCE_LFA;
2323
2324         /* load request */
2325         (*load_code) = bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_REQ, param);
2326
2327         /* if mcp fails to respond we must abort */
2328         if (!(*load_code)) {
2329                 BNX2X_ERR("MCP response failure, aborting\n");
2330                 return -EBUSY;
2331         }
2332
2333         /* If mcp refused (e.g. other port is in diagnostic mode) we
2334          * must abort
2335          */
2336         if ((*load_code) == FW_MSG_CODE_DRV_LOAD_REFUSED) {
2337                 BNX2X_ERR("MCP refused load request, aborting\n");
2338                 return -EBUSY;
2339         }
2340         return 0;
2341 }
2342
2343 /* check whether another PF has already loaded FW to chip. In
2344  * virtualized environments a pf from another VM may have already
2345  * initialized the device including loading FW
2346  */
2347 int bnx2x_compare_fw_ver(struct bnx2x *bp, u32 load_code, bool print_err)
2348 {
2349         /* is another pf loaded on this engine? */
2350         if (load_code != FW_MSG_CODE_DRV_LOAD_COMMON_CHIP &&
2351             load_code != FW_MSG_CODE_DRV_LOAD_COMMON) {
2352                 /* build my FW version dword */
2353                 u32 my_fw = (BCM_5710_FW_MAJOR_VERSION) +
2354                         (BCM_5710_FW_MINOR_VERSION << 8) +
2355                         (BCM_5710_FW_REVISION_VERSION << 16) +
2356                         (BCM_5710_FW_ENGINEERING_VERSION << 24);
2357
2358                 /* read loaded FW from chip */
2359                 u32 loaded_fw = REG_RD(bp, XSEM_REG_PRAM);
2360
2361                 DP(BNX2X_MSG_SP, "loaded fw %x, my fw %x\n",
2362                    loaded_fw, my_fw);
2363
2364                 /* abort nic load if version mismatch */
2365                 if (my_fw != loaded_fw) {
2366                         if (print_err)
2367                                 BNX2X_ERR("bnx2x with FW %x was already loaded which mismatches my %x FW. Aborting\n",
2368                                           loaded_fw, my_fw);
2369                         else
2370                                 BNX2X_DEV_INFO("bnx2x with FW %x was already loaded which mismatches my %x FW, possibly due to MF UNDI\n",
2371                                                loaded_fw, my_fw);
2372                         return -EBUSY;
2373                 }
2374         }
2375         return 0;
2376 }
2377
2378 /* returns the "mcp load_code" according to global load_count array */
2379 static int bnx2x_nic_load_no_mcp(struct bnx2x *bp, int port)
2380 {
2381         int path = BP_PATH(bp);
2382
2383         DP(NETIF_MSG_IFUP, "NO MCP - load counts[%d]      %d, %d, %d\n",
2384            path, bnx2x_load_count[path][0], bnx2x_load_count[path][1],
2385            bnx2x_load_count[path][2]);
2386         bnx2x_load_count[path][0]++;
2387         bnx2x_load_count[path][1 + port]++;
2388         DP(NETIF_MSG_IFUP, "NO MCP - new load counts[%d]  %d, %d, %d\n",
2389            path, bnx2x_load_count[path][0], bnx2x_load_count[path][1],
2390            bnx2x_load_count[path][2]);
2391         if (bnx2x_load_count[path][0] == 1)
2392                 return FW_MSG_CODE_DRV_LOAD_COMMON;
2393         else if (bnx2x_load_count[path][1 + port] == 1)
2394                 return FW_MSG_CODE_DRV_LOAD_PORT;
2395         else
2396                 return FW_MSG_CODE_DRV_LOAD_FUNCTION;
2397 }
2398
2399 /* mark PMF if applicable */
2400 static void bnx2x_nic_load_pmf(struct bnx2x *bp, u32 load_code)
2401 {
2402         if ((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
2403             (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP) ||
2404             (load_code == FW_MSG_CODE_DRV_LOAD_PORT)) {
2405                 bp->port.pmf = 1;
2406                 /* We need the barrier to ensure the ordering between the
2407                  * writing to bp->port.pmf here and reading it from the
2408                  * bnx2x_periodic_task().
2409                  */
2410                 smp_mb();
2411         } else {
2412                 bp->port.pmf = 0;
2413         }
2414
2415         DP(NETIF_MSG_LINK, "pmf %d\n", bp->port.pmf);
2416 }
2417
2418 static void bnx2x_nic_load_afex_dcc(struct bnx2x *bp, int load_code)
2419 {
2420         if (((load_code == FW_MSG_CODE_DRV_LOAD_COMMON) ||
2421              (load_code == FW_MSG_CODE_DRV_LOAD_COMMON_CHIP)) &&
2422             (bp->common.shmem2_base)) {
2423                 if (SHMEM2_HAS(bp, dcc_support))
2424                         SHMEM2_WR(bp, dcc_support,
2425                                   (SHMEM_DCC_SUPPORT_DISABLE_ENABLE_PF_TLV |
2426                                    SHMEM_DCC_SUPPORT_BANDWIDTH_ALLOCATION_TLV));
2427                 if (SHMEM2_HAS(bp, afex_driver_support))
2428                         SHMEM2_WR(bp, afex_driver_support,
2429                                   SHMEM_AFEX_SUPPORTED_VERSION_ONE);
2430         }
2431
2432         /* Set AFEX default VLAN tag to an invalid value */
2433         bp->afex_def_vlan_tag = -1;
2434 }
2435
2436 /**
2437  * bnx2x_bz_fp - zero content of the fastpath structure.
2438  *
2439  * @bp:         driver handle
2440  * @index:      fastpath index to be zeroed
2441  *
2442  * Makes sure the contents of the bp->fp[index].napi is kept
2443  * intact.
2444  */
2445 static void bnx2x_bz_fp(struct bnx2x *bp, int index)
2446 {
2447         struct bnx2x_fastpath *fp = &bp->fp[index];
2448         int cos;
2449         struct napi_struct orig_napi = fp->napi;
2450         struct bnx2x_agg_info *orig_tpa_info = fp->tpa_info;
2451
2452         /* bzero bnx2x_fastpath contents */
2453         if (fp->tpa_info)
2454                 memset(fp->tpa_info, 0, ETH_MAX_AGGREGATION_QUEUES_E1H_E2 *
2455                        sizeof(struct bnx2x_agg_info));
2456         memset(fp, 0, sizeof(*fp));
2457
2458         /* Restore the NAPI object as it has been already initialized */
2459         fp->napi = orig_napi;
2460         fp->tpa_info = orig_tpa_info;
2461         fp->bp = bp;
2462         fp->index = index;
2463         if (IS_ETH_FP(fp))
2464                 fp->max_cos = bp->max_cos;
2465         else
2466                 /* Special queues support only one CoS */
2467                 fp->max_cos = 1;
2468
2469         /* Init txdata pointers */
2470         if (IS_FCOE_FP(fp))
2471                 fp->txdata_ptr[0] = &bp->bnx2x_txq[FCOE_TXQ_IDX(bp)];
2472         if (IS_ETH_FP(fp))
2473                 for_each_cos_in_tx_queue(fp, cos)
2474                         fp->txdata_ptr[cos] = &bp->bnx2x_txq[cos *
2475                                 BNX2X_NUM_ETH_QUEUES(bp) + index];
2476
2477         /* set the tpa flag for each queue. The tpa flag determines the queue
2478          * minimal size so it must be set prior to queue memory allocation
2479          */
2480         fp->disable_tpa = !(bp->flags & TPA_ENABLE_FLAG ||
2481                                   (bp->flags & GRO_ENABLE_FLAG &&
2482                                    bnx2x_mtu_allows_gro(bp->dev->mtu)));
2483         if (bp->flags & TPA_ENABLE_FLAG)
2484                 fp->mode = TPA_MODE_LRO;
2485         else if (bp->flags & GRO_ENABLE_FLAG)
2486                 fp->mode = TPA_MODE_GRO;
2487
2488         /* We don't want TPA on an FCoE L2 ring */
2489         if (IS_FCOE_FP(fp))
2490                 fp->disable_tpa = 1;
2491 }
2492
2493 int bnx2x_load_cnic(struct bnx2x *bp)
2494 {
2495         int i, rc, port = BP_PORT(bp);
2496
2497         DP(NETIF_MSG_IFUP, "Starting CNIC-related load\n");
2498
2499         mutex_init(&bp->cnic_mutex);
2500
2501         if (IS_PF(bp)) {
2502                 rc = bnx2x_alloc_mem_cnic(bp);
2503                 if (rc) {
2504                         BNX2X_ERR("Unable to allocate bp memory for cnic\n");
2505                         LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic0);
2506                 }
2507         }
2508
2509         rc = bnx2x_alloc_fp_mem_cnic(bp);
2510         if (rc) {
2511                 BNX2X_ERR("Unable to allocate memory for cnic fps\n");
2512                 LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic0);
2513         }
2514
2515         /* Update the number of queues with the cnic queues */
2516         rc = bnx2x_set_real_num_queues(bp, 1);
2517         if (rc) {
2518                 BNX2X_ERR("Unable to set real_num_queues including cnic\n");
2519                 LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic0);
2520         }
2521
2522         /* Add all CNIC NAPI objects */
2523         bnx2x_add_all_napi_cnic(bp);
2524         DP(NETIF_MSG_IFUP, "cnic napi added\n");
2525         bnx2x_napi_enable_cnic(bp);
2526
2527         rc = bnx2x_init_hw_func_cnic(bp);
2528         if (rc)
2529                 LOAD_ERROR_EXIT_CNIC(bp, load_error_cnic1);
2530
2531         bnx2x_nic_init_cnic(bp);
2532
2533         if (IS_PF(bp)) {
2534                 /* Enable Timer scan */
2535                 REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + port*4, 1);
2536
2537                 /* setup cnic queues */
2538                 for_each_cnic_queue(bp, i) {
2539                         rc = bnx2x_setup_queue(bp, &bp->fp[i], 0);
2540                         if (rc) {
2541                                 BNX2X_ERR("Queue setup failed\n");
2542                                 LOAD_ERROR_EXIT(bp, load_error_cnic2);
2543                         }
2544                 }
2545         }
2546
2547         /* Initialize Rx filter. */
2548         bnx2x_set_rx_mode_inner(bp);
2549
2550         /* re-read iscsi info */
2551         bnx2x_get_iscsi_info(bp);
2552         bnx2x_setup_cnic_irq_info(bp);
2553         bnx2x_setup_cnic_info(bp);
2554         bp->cnic_loaded = true;
2555         if (bp->state == BNX2X_STATE_OPEN)
2556                 bnx2x_cnic_notify(bp, CNIC_CTL_START_CMD);
2557
2558         DP(NETIF_MSG_IFUP, "Ending successfully CNIC-related load\n");
2559
2560         return 0;
2561
2562 #ifndef BNX2X_STOP_ON_ERROR
2563 load_error_cnic2:
2564         /* Disable Timer scan */
2565         REG_WR(bp, TM_REG_EN_LINEAR0_TIMER + port*4, 0);
2566
2567 load_error_cnic1:
2568         bnx2x_napi_disable_cnic(bp);
2569         /* Update the number of queues without the cnic queues */
2570         if (bnx2x_set_real_num_queues(bp, 0))
2571                 BNX2X_ERR("Unable to set real_num_queues not including cnic\n");
2572 load_error_cnic0:
2573         BNX2X_ERR("CNIC-related load failed\n");
2574         bnx2x_free_fp_mem_cnic(bp);
2575         bnx2x_free_mem_cnic(bp);
2576         return rc;
2577 #endif /* ! BNX2X_STOP_ON_ERROR */
2578 }
2579
2580 /* must be called with rtnl_lock */
2581 int bnx2x_nic_load(struct bnx2x *bp, int load_mode)
2582 {
2583         int port = BP_PORT(bp);
2584         int i, rc = 0, load_code = 0;
2585
2586         DP(NETIF_MSG_IFUP, "Starting NIC load\n");
2587         DP(NETIF_MSG_IFUP,
2588            "CNIC is %s\n", CNIC_ENABLED(bp) ? "enabled" : "disabled");
2589
2590 #ifdef BNX2X_STOP_ON_ERROR
2591         if (unlikely(bp->panic)) {
2592                 BNX2X_ERR("Can't load NIC when there is panic\n");
2593                 return -EPERM;
2594         }
2595 #endif
2596
2597         bp->state = BNX2X_STATE_OPENING_WAIT4_LOAD;
2598
2599         /* zero the structure w/o any lock, before SP handler is initialized */
2600         memset(&bp->last_reported_link, 0, sizeof(bp->last_reported_link));
2601         __set_bit(BNX2X_LINK_REPORT_LINK_DOWN,
2602                 &bp->last_reported_link.link_report_flags);
2603
2604         if (IS_PF(bp))
2605                 /* must be called before memory allocation and HW init */
2606                 bnx2x_ilt_set_info(bp);
2607
2608         /*
2609          * Zero fastpath structures preserving invariants like napi, which are
2610          * allocated only once, fp index, max_cos, bp pointer.
2611          * Also set fp->disable_tpa and txdata_ptr.
2612          */
2613         DP(NETIF_MSG_IFUP, "num queues: %d", bp->num_queues);
2614         for_each_queue(bp, i)
2615                 bnx2x_bz_fp(bp, i);
2616         memset(bp->bnx2x_txq, 0, (BNX2X_MAX_RSS_COUNT(bp) * BNX2X_MULTI_TX_COS +
2617                                   bp->num_cnic_queues) *
2618                                   sizeof(struct bnx2x_fp_txdata));
2619
2620         bp->fcoe_init = false;
2621
2622         /* Set the receive queues buffer size */
2623         bnx2x_set_rx_buf_size(bp);
2624
2625         if (IS_PF(bp)) {
2626                 rc = bnx2x_alloc_mem(bp);
2627                 if (rc) {
2628                         BNX2X_ERR("Unable to allocate bp memory\n");
2629                         return rc;
2630                 }
2631         }
2632
2633         /* need to be done after alloc mem, since it's self adjusting to amount
2634          * of memory available for RSS queues
2635          */
2636         rc = bnx2x_alloc_fp_mem(bp);
2637         if (rc) {
2638                 BNX2X_ERR("Unable to allocate memory for fps\n");
2639                 LOAD_ERROR_EXIT(bp, load_error0);
2640         }
2641
2642         /* Allocated memory for FW statistics  */
2643         if (bnx2x_alloc_fw_stats_mem(bp))
2644                 LOAD_ERROR_EXIT(bp, load_error0);
2645
2646         /* request pf to initialize status blocks */
2647         if (IS_VF(bp)) {
2648                 rc = bnx2x_vfpf_init(bp);
2649                 if (rc)
2650                         LOAD_ERROR_EXIT(bp, load_error0);
2651         }
2652
2653         /* As long as bnx2x_alloc_mem() may possibly update
2654          * bp->num_queues, bnx2x_set_real_num_queues() should always
2655          * come after it. At this stage cnic queues are not counted.
2656          */
2657         rc = bnx2x_set_real_num_queues(bp, 0);
2658         if (rc) {
2659                 BNX2X_ERR("Unable to set real_num_queues\n");
2660                 LOAD_ERROR_EXIT(bp, load_error0);
2661         }
2662
2663         /* configure multi cos mappings in kernel.
2664          * this configuration may be overridden by a multi class queue
2665          * discipline or by a dcbx negotiation result.
2666          */
2667         bnx2x_setup_tc(bp->dev, bp->max_cos);
2668
2669         /* Add all NAPI objects */
2670         bnx2x_add_all_napi(bp);
2671         DP(NETIF_MSG_IFUP, "napi added\n");
2672         bnx2x_napi_enable(bp);
2673
2674         if (IS_PF(bp)) {
2675                 /* set pf load just before approaching the MCP */
2676                 bnx2x_set_pf_load(bp);
2677
2678                 /* if mcp exists send load request and analyze response */
2679                 if (!BP_NOMCP(bp)) {
2680                         /* attempt to load pf */
2681                         rc = bnx2x_nic_load_request(bp, &load_code);
2682                         if (rc)
2683                                 LOAD_ERROR_EXIT(bp, load_error1);
2684
2685                         /* what did mcp say? */
2686                         rc = bnx2x_compare_fw_ver(bp, load_code, true);
2687                         if (rc) {
2688                                 bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2689                                 LOAD_ERROR_EXIT(bp, load_error2);
2690                         }
2691                 } else {
2692                         load_code = bnx2x_nic_load_no_mcp(bp, port);
2693                 }
2694
2695                 /* mark pmf if applicable */
2696                 bnx2x_nic_load_pmf(bp, load_code);
2697
2698                 /* Init Function state controlling object */
2699                 bnx2x__init_func_obj(bp);
2700
2701                 /* Initialize HW */
2702                 rc = bnx2x_init_hw(bp, load_code);
2703                 if (rc) {
2704                         BNX2X_ERR("HW init failed, aborting\n");
2705                         bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2706                         LOAD_ERROR_EXIT(bp, load_error2);
2707                 }
2708         }
2709
2710         bnx2x_pre_irq_nic_init(bp);
2711
2712         /* Connect to IRQs */
2713         rc = bnx2x_setup_irqs(bp);
2714         if (rc) {
2715                 BNX2X_ERR("setup irqs failed\n");
2716                 if (IS_PF(bp))
2717                         bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2718                 LOAD_ERROR_EXIT(bp, load_error2);
2719         }
2720
2721         /* Init per-function objects */
2722         if (IS_PF(bp)) {
2723                 /* Setup NIC internals and enable interrupts */
2724                 bnx2x_post_irq_nic_init(bp, load_code);
2725
2726                 bnx2x_init_bp_objs(bp);
2727                 bnx2x_iov_nic_init(bp);
2728
2729                 /* Set AFEX default VLAN tag to an invalid value */
2730                 bp->afex_def_vlan_tag = -1;
2731                 bnx2x_nic_load_afex_dcc(bp, load_code);
2732                 bp->state = BNX2X_STATE_OPENING_WAIT4_PORT;
2733                 rc = bnx2x_func_start(bp);
2734                 if (rc) {
2735                         BNX2X_ERR("Function start failed!\n");
2736                         bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE, 0);
2737
2738                         LOAD_ERROR_EXIT(bp, load_error3);
2739                 }
2740
2741                 /* Send LOAD_DONE command to MCP */
2742                 if (!BP_NOMCP(bp)) {
2743                         load_code = bnx2x_fw_command(bp,
2744                                                      DRV_MSG_CODE_LOAD_DONE, 0);
2745                         if (!load_code) {
2746                                 BNX2X_ERR("MCP response failure, aborting\n");
2747                                 rc = -EBUSY;
2748                                 LOAD_ERROR_EXIT(bp, load_error3);
2749                         }
2750                 }
2751
2752                 /* initialize FW coalescing state machines in RAM */
2753                 bnx2x_update_coalesce(bp);
2754         }
2755
2756         /* setup the leading queue */
2757         rc = bnx2x_setup_leading(bp);
2758         if (rc) {
2759                 BNX2X_ERR("Setup leading failed!\n");
2760                 LOAD_ERROR_EXIT(bp, load_error3);
2761         }
2762
2763         /* set up the rest of the queues */
2764         for_each_nondefault_eth_queue(bp, i) {
2765                 if (IS_PF(bp))
2766                         rc = bnx2x_setup_queue(bp, &bp->fp[i], false);
2767                 else /* VF */
2768                         rc = bnx2x_vfpf_setup_q(bp, &bp->fp[i], false);
2769                 if (rc) {
2770                         BNX2X_ERR("Queue %d setup failed\n", i);
2771                         LOAD_ERROR_EXIT(bp, load_error3);
2772                 }
2773         }
2774
2775         /* setup rss */
2776         rc = bnx2x_init_rss(bp);
2777         if (rc) {
2778                 BNX2X_ERR("PF RSS init failed\n");
2779                 LOAD_ERROR_EXIT(bp, load_error3);
2780         }
2781
2782         /* Now when Clients are configured we are ready to work */
2783         bp->state = BNX2X_STATE_OPEN;
2784
2785         /* Configure a ucast MAC */
2786         if (IS_PF(bp))
2787                 rc = bnx2x_set_eth_mac(bp, true);
2788         else /* vf */
2789                 rc = bnx2x_vfpf_config_mac(bp, bp->dev->dev_addr, bp->fp->index,
2790                                            true);
2791         if (rc) {
2792                 BNX2X_ERR("Setting Ethernet MAC failed\n");
2793                 LOAD_ERROR_EXIT(bp, load_error3);
2794         }
2795
2796         if (IS_PF(bp) && bp->pending_max) {
2797                 bnx2x_update_max_mf_config(bp, bp->pending_max);
2798                 bp->pending_max = 0;
2799         }
2800
2801         if (bp->port.pmf) {
2802                 rc = bnx2x_initial_phy_init(bp, load_mode);
2803                 if (rc)
2804                         LOAD_ERROR_EXIT(bp, load_error3);
2805         }
2806         bp->link_params.feature_config_flags &= ~FEATURE_CONFIG_BOOT_FROM_SAN;
2807
2808         /* Start fast path */
2809
2810         /* Initialize Rx filter. */
2811         bnx2x_set_rx_mode_inner(bp);
2812
2813         if (bp->flags & PTP_SUPPORTED) {
2814                 bnx2x_init_ptp(bp);
2815                 bnx2x_configure_ptp_filters(bp);
2816         }
2817         /* Start Tx */
2818         switch (load_mode) {
2819         case LOAD_NORMAL:
2820                 /* Tx queue should be only re-enabled */
2821                 netif_tx_wake_all_queues(bp->dev);
2822                 break;
2823
2824         case LOAD_OPEN:
2825                 netif_tx_start_all_queues(bp->dev);
2826                 smp_mb__after_atomic();
2827                 break;
2828
2829         case LOAD_DIAG:
2830         case LOAD_LOOPBACK_EXT:
2831                 bp->state = BNX2X_STATE_DIAG;
2832                 break;
2833
2834         default:
2835                 break;
2836         }
2837
2838         if (bp->port.pmf)
2839                 bnx2x_update_drv_flags(bp, 1 << DRV_FLAGS_PORT_MASK, 0);
2840         else
2841                 bnx2x__link_status_update(bp);
2842
2843         /* start the timer */
2844         mod_timer(&bp->timer, jiffies + bp->current_interval);
2845
2846         if (CNIC_ENABLED(bp))
2847                 bnx2x_load_cnic(bp);
2848
2849         if (IS_PF(bp))
2850                 bnx2x_schedule_sp_rtnl(bp, BNX2X_SP_RTNL_GET_DRV_VERSION, 0);
2851
2852         if (IS_PF(bp) && SHMEM2_HAS(bp, drv_capabilities_flag)) {
2853                 /* mark driver is loaded in shmem2 */
2854                 u32 val;
2855                 val = SHMEM2_RD(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)]);
2856                 SHMEM2_WR(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)],
2857                           val | DRV_FLAGS_CAPABILITIES_LOADED_SUPPORTED |
2858                           DRV_FLAGS_CAPABILITIES_LOADED_L2);
2859         }
2860
2861         /* Wait for all pending SP commands to complete */
2862         if (IS_PF(bp) && !bnx2x_wait_sp_comp(bp, ~0x0UL)) {
2863                 BNX2X_ERR("Timeout waiting for SP elements to complete\n");
2864                 bnx2x_nic_unload(bp, UNLOAD_CLOSE, false);
2865                 return -EBUSY;
2866         }
2867
2868         /* If PMF - send ADMIN DCBX msg to MFW to initiate DCBX FSM */
2869         if (bp->port.pmf && (bp->state != BNX2X_STATE_DIAG))
2870                 bnx2x_dcbx_init(bp, false);
2871
2872         DP(NETIF_MSG_IFUP, "Ending successfully NIC load\n");
2873
2874         return 0;
2875
2876 #ifndef BNX2X_STOP_ON_ERROR
2877 load_error3:
2878         if (IS_PF(bp)) {
2879                 bnx2x_int_disable_sync(bp, 1);
2880
2881                 /* Clean queueable objects */
2882                 bnx2x_squeeze_objects(bp);
2883         }
2884
2885         /* Free SKBs, SGEs, TPA pool and driver internals */
2886         bnx2x_free_skbs(bp);
2887         for_each_rx_queue(bp, i)
2888                 bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
2889
2890         /* Release IRQs */
2891         bnx2x_free_irq(bp);
2892 load_error2:
2893         if (IS_PF(bp) && !BP_NOMCP(bp)) {
2894                 bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_REQ_WOL_MCP, 0);
2895                 bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_DONE, 0);
2896         }
2897
2898         bp->port.pmf = 0;
2899 load_error1:
2900         bnx2x_napi_disable(bp);
2901         bnx2x_del_all_napi(bp);
2902
2903         /* clear pf_load status, as it was already set */
2904         if (IS_PF(bp))
2905                 bnx2x_clear_pf_load(bp);
2906 load_error0:
2907         bnx2x_free_fw_stats_mem(bp);
2908         bnx2x_free_fp_mem(bp);
2909         bnx2x_free_mem(bp);
2910
2911         return rc;
2912 #endif /* ! BNX2X_STOP_ON_ERROR */
2913 }
2914
2915 int bnx2x_drain_tx_queues(struct bnx2x *bp)
2916 {
2917         u8 rc = 0, cos, i;
2918
2919         /* Wait until tx fastpath tasks complete */
2920         for_each_tx_queue(bp, i) {
2921                 struct bnx2x_fastpath *fp = &bp->fp[i];
2922
2923                 for_each_cos_in_tx_queue(fp, cos)
2924                         rc = bnx2x_clean_tx_queue(bp, fp->txdata_ptr[cos]);
2925                 if (rc)
2926                         return rc;
2927         }
2928         return 0;
2929 }
2930
2931 /* must be called with rtnl_lock */
2932 int bnx2x_nic_unload(struct bnx2x *bp, int unload_mode, bool keep_link)
2933 {
2934         int i;
2935         bool global = false;
2936
2937         DP(NETIF_MSG_IFUP, "Starting NIC unload\n");
2938
2939         /* mark driver is unloaded in shmem2 */
2940         if (IS_PF(bp) && SHMEM2_HAS(bp, drv_capabilities_flag)) {
2941                 u32 val;
2942                 val = SHMEM2_RD(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)]);
2943                 SHMEM2_WR(bp, drv_capabilities_flag[BP_FW_MB_IDX(bp)],
2944                           val & ~DRV_FLAGS_CAPABILITIES_LOADED_L2);
2945         }
2946
2947         if (IS_PF(bp) && bp->recovery_state != BNX2X_RECOVERY_DONE &&
2948             (bp->state == BNX2X_STATE_CLOSED ||
2949              bp->state == BNX2X_STATE_ERROR)) {
2950                 /* We can get here if the driver has been unloaded
2951                  * during parity error recovery and is either waiting for a
2952                  * leader to complete or for other functions to unload and
2953                  * then ifdown has been issued. In this case we want to
2954                  * unload and let other functions to complete a recovery
2955                  * process.
2956                  */
2957                 bp->recovery_state = BNX2X_RECOVERY_DONE;
2958                 bp->is_leader = 0;
2959                 bnx2x_release_leader_lock(bp);
2960                 smp_mb();
2961
2962                 DP(NETIF_MSG_IFDOWN, "Releasing a leadership...\n");
2963                 BNX2X_ERR("Can't unload in closed or error state\n");
2964                 return -EINVAL;
2965         }
2966
2967         /* Nothing to do during unload if previous bnx2x_nic_load()
2968          * have not completed successfully - all resources are released.
2969          *
2970          * we can get here only after unsuccessful ndo_* callback, during which
2971          * dev->IFF_UP flag is still on.
2972          */
2973         if (bp->state == BNX2X_STATE_CLOSED || bp->state == BNX2X_STATE_ERROR)
2974                 return 0;
2975
2976         /* It's important to set the bp->state to the value different from
2977          * BNX2X_STATE_OPEN and only then stop the Tx. Otherwise bnx2x_tx_int()
2978          * may restart the Tx from the NAPI context (see bnx2x_tx_int()).
2979          */
2980         bp->state = BNX2X_STATE_CLOSING_WAIT4_HALT;
2981         smp_mb();
2982
2983         /* indicate to VFs that the PF is going down */
2984         bnx2x_iov_channel_down(bp);
2985
2986         if (CNIC_LOADED(bp))
2987                 bnx2x_cnic_notify(bp, CNIC_CTL_STOP_CMD);
2988
2989         /* Stop Tx */
2990         bnx2x_tx_disable(bp);
2991         netdev_reset_tc(bp->dev);
2992
2993         bp->rx_mode = BNX2X_RX_MODE_NONE;
2994
2995         del_timer_sync(&bp->timer);
2996
2997         if (IS_PF(bp)) {
2998                 /* Set ALWAYS_ALIVE bit in shmem */
2999                 bp->fw_drv_pulse_wr_seq |= DRV_PULSE_ALWAYS_ALIVE;
3000                 bnx2x_drv_pulse(bp);
3001                 bnx2x_stats_handle(bp, STATS_EVENT_STOP);
3002                 bnx2x_save_statistics(bp);
3003         }
3004
3005         /* wait till consumers catch up with producers in all queues */
3006         bnx2x_drain_tx_queues(bp);
3007
3008         /* if VF indicate to PF this function is going down (PF will delete sp
3009          * elements and clear initializations
3010          */
3011         if (IS_VF(bp))
3012                 bnx2x_vfpf_close_vf(bp);
3013         else if (unload_mode != UNLOAD_RECOVERY)
3014                 /* if this is a normal/close unload need to clean up chip*/
3015                 bnx2x_chip_cleanup(bp, unload_mode, keep_link);
3016         else {
3017                 /* Send the UNLOAD_REQUEST to the MCP */
3018                 bnx2x_send_unload_req(bp, unload_mode);
3019
3020                 /* Prevent transactions to host from the functions on the
3021                  * engine that doesn't reset global blocks in case of global
3022                  * attention once global blocks are reset and gates are opened
3023                  * (the engine which leader will perform the recovery
3024                  * last).
3025                  */
3026                 if (!CHIP_IS_E1x(bp))
3027                         bnx2x_pf_disable(bp);
3028
3029                 /* Disable HW interrupts, NAPI */
3030                 bnx2x_netif_stop(bp, 1);
3031                 /* Delete all NAPI objects */
3032                 bnx2x_del_all_napi(bp);
3033                 if (CNIC_LOADED(bp))
3034                         bnx2x_del_all_napi_cnic(bp);
3035                 /* Release IRQs */
3036                 bnx2x_free_irq(bp);
3037
3038                 /* Report UNLOAD_DONE to MCP */
3039                 bnx2x_send_unload_done(bp, false);
3040         }
3041
3042         /*
3043          * At this stage no more interrupts will arrive so we may safely clean
3044          * the queueable objects here in case they failed to get cleaned so far.
3045          */
3046         if (IS_PF(bp))
3047                 bnx2x_squeeze_objects(bp);
3048
3049         /* There should be no more pending SP commands at this stage */
3050         bp->sp_state = 0;
3051
3052         bp->port.pmf = 0;
3053
3054         /* clear pending work in rtnl task */
3055         bp->sp_rtnl_state = 0;
3056         smp_mb();
3057
3058         /* Free SKBs, SGEs, TPA pool and driver internals */
3059         bnx2x_free_skbs(bp);
3060         if (CNIC_LOADED(bp))
3061                 bnx2x_free_skbs_cnic(bp);
3062         for_each_rx_queue(bp, i)
3063                 bnx2x_free_rx_sge_range(bp, bp->fp + i, NUM_RX_SGE);
3064
3065         bnx2x_free_fp_mem(bp);
3066         if (CNIC_LOADED(bp))
3067                 bnx2x_free_fp_mem_cnic(bp);
3068
3069         if (IS_PF(bp)) {
3070                 if (CNIC_LOADED(bp))
3071                         bnx2x_free_mem_cnic(bp);
3072         }
3073         bnx2x_free_mem(bp);
3074
3075         bp->state = BNX2X_STATE_CLOSED;
3076         bp->cnic_loaded = false;
3077
3078         /* Clear driver version indication in shmem */
3079         if (IS_PF(bp))
3080                 bnx2x_update_mng_version(bp);
3081
3082         /* Check if there are pending parity attentions. If there are - set
3083          * RECOVERY_IN_PROGRESS.
3084          */
3085         if (IS_PF(bp) && bnx2x_chk_parity_attn(bp, &global, false)) {
3086                 bnx2x_set_reset_in_progress(bp);
3087
3088                 /* Set RESET_IS_GLOBAL if needed */
3089                 if (global)
3090                         bnx2x_set_reset_global(bp);
3091         }
3092
3093         /* The last driver must disable a "close the gate" if there is no
3094          * parity attention or "process kill" pending.
3095          */
3096         if (IS_PF(bp) &&
3097             !bnx2x_clear_pf_load(bp) &&
3098             bnx2x_reset_is_done(bp, BP_PATH(bp)))
3099                 bnx2x_disable_close_the_gate(bp);
3100
3101         DP(NETIF_MSG_IFUP, "Ending NIC unload\n");
3102
3103         return 0;
3104 }
3105
3106 int bnx2x_set_power_state(struct bnx2x *bp, pci_power_t state)
3107 {
3108         u16 pmcsr;
3109
3110         /* If there is no power capability, silently succeed */
3111         if (!bp->pdev->pm_cap) {
3112                 BNX2X_DEV_INFO("No power capability. Breaking.\n");
3113                 return 0;
3114         }
3115
3116         pci_read_config_word(bp->pdev, bp->pdev->pm_cap + PCI_PM_CTRL, &pmcsr);
3117
3118         switch (state) {
3119         case PCI_D0:
3120                 pci_write_config_word(bp->pdev, bp->pdev->pm_cap + PCI_PM_CTRL,
3121                                       ((pmcsr & ~PCI_PM_CTRL_STATE_MASK) |
3122                                        PCI_PM_CTRL_PME_STATUS));
3123
3124                 if (pmcsr & PCI_PM_CTRL_STATE_MASK)
3125                         /* delay required during transition out of D3hot */
3126                         msleep(20);
3127                 break;
3128
3129         case PCI_D3hot:
3130                 /* If there are other clients above don't
3131                    shut down the power */
3132                 if (atomic_read(&bp->pdev->enable_cnt) != 1)
3133                         return 0;
3134                 /* Don't shut down the power for emulation and FPGA */
3135                 if (CHIP_REV_IS_SLOW(bp))
3136                         return 0;
3137
3138                 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
3139                 pmcsr |= 3;
3140
3141                 if (bp->wol)
3142                         pmcsr |= PCI_PM_CTRL_PME_ENABLE;
3143
3144                 pci_write_config_word(bp->pdev, bp->pdev->pm_cap + PCI_PM_CTRL,
3145                                       pmcsr);
3146
3147                 /* No more memory access after this point until
3148                 * device is brought back to D0.
3149                 */
3150                 break;
3151
3152         default:
3153                 dev_err(&bp->pdev->dev, "Can't support state = %d\n", state);
3154                 return -EINVAL;
3155         }
3156         return 0;
3157 }
3158
3159 /*
3160  * net_device service functions
3161  */
3162 static int bnx2x_poll(struct napi_struct *napi, int budget)
3163 {
3164         int work_done = 0;
3165         u8 cos;
3166         struct bnx2x_fastpath *fp = container_of(napi, struct bnx2x_fastpath,
3167                                                  napi);
3168         struct bnx2x *bp = fp->bp;
3169
3170         while (1) {
3171 #ifdef BNX2X_STOP_ON_ERROR
3172                 if (unlikely(bp->panic)) {
3173                         napi_complete(napi);
3174                         return 0;
3175                 }
3176 #endif
3177                 if (!bnx2x_fp_lock_napi(fp))
3178                         return budget;
3179
3180                 for_each_cos_in_tx_queue(fp, cos)
3181                         if (bnx2x_tx_queue_has_work(fp->txdata_ptr[cos]))
3182                                 bnx2x_tx_int(bp, fp->txdata_ptr[cos]);
3183
3184                 if (bnx2x_has_rx_work(fp)) {
3185                         work_done += bnx2x_rx_int(fp, budget - work_done);
3186
3187                         /* must not complete if we consumed full budget */
3188                         if (work_done >= budget) {
3189                                 bnx2x_fp_unlock_napi(fp);
3190                                 break;
3191                         }
3192                 }
3193
3194                 /* Fall out from the NAPI loop if needed */
3195                 if (!bnx2x_fp_unlock_napi(fp) &&
3196                     !(bnx2x_has_rx_work(fp) || bnx2x_has_tx_work(fp))) {
3197
3198                         /* No need to update SB for FCoE L2 ring as long as
3199                          * it's connected to the default SB and the SB
3200                          * has been updated when NAPI was scheduled.
3201                          */
3202                         if (IS_FCOE_FP(fp)) {
3203                                 napi_complete(napi);
3204                                 break;
3205                         }
3206                         bnx2x_update_fpsb_idx(fp);
3207                         /* bnx2x_has_rx_work() reads the status block,
3208                          * thus we need to ensure that status block indices
3209                          * have been actually read (bnx2x_update_fpsb_idx)
3210                          * prior to this check (bnx2x_has_rx_work) so that
3211                          * we won't write the "newer" value of the status block
3212                          * to IGU (if there was a DMA right after
3213                          * bnx2x_has_rx_work and if there is no rmb, the memory
3214                          * reading (bnx2x_update_fpsb_idx) may be postponed
3215                          * to right before bnx2x_ack_sb). In this case there
3216                          * will never be another interrupt until there is
3217                          * another update of the status block, while there
3218                          * is still unhandled work.
3219                          */
3220                         rmb();
3221
3222                         if (!(bnx2x_has_rx_work(fp) || bnx2x_has_tx_work(fp))) {
3223                                 napi_complete(napi);
3224                                 /* Re-enable interrupts */
3225                                 DP(NETIF_MSG_RX_STATUS,
3226                                    "Update index to %d\n", fp->fp_hc_idx);
3227                                 bnx2x_ack_sb(bp, fp->igu_sb_id, USTORM_ID,
3228                                              le16_to_cpu(fp->fp_hc_idx),
3229                                              IGU_INT_ENABLE, 1);
3230                                 break;
3231                         }
3232                 }
3233         }
3234
3235         return work_done;
3236 }
3237
3238 #ifdef CONFIG_NET_RX_BUSY_POLL
3239 /* must be called with local_bh_disable()d */
3240 int bnx2x_low_latency_recv(struct napi_struct *napi)
3241 {
3242         struct bnx2x_fastpath *fp = container_of(napi, struct bnx2x_fastpath,
3243                                                  napi);
3244         struct bnx2x *bp = fp->bp;
3245         int found = 0;
3246
3247         if ((bp->state == BNX2X_STATE_CLOSED) ||
3248             (bp->state == BNX2X_STATE_ERROR) ||
3249             (bp->flags & (TPA_ENABLE_FLAG | GRO_ENABLE_FLAG)))
3250                 return LL_FLUSH_FAILED;
3251
3252         if (!bnx2x_fp_lock_poll(fp))
3253                 return LL_FLUSH_BUSY;
3254
3255         if (bnx2x_has_rx_work(fp))
3256                 found = bnx2x_rx_int(fp, 4);
3257
3258         bnx2x_fp_unlock_poll(fp);
3259
3260         return found;
3261 }
3262 #endif
3263
3264 /* we split the first BD into headers and data BDs
3265  * to ease the pain of our fellow microcode engineers
3266  * we use one mapping for both BDs
3267  */
3268 static u16 bnx2x_tx_split(struct bnx2x *bp,
3269                           struct bnx2x_fp_txdata *txdata,
3270                           struct sw_tx_bd *tx_buf,
3271                           struct eth_tx_start_bd **tx_bd, u16 hlen,
3272                           u16 bd_prod)
3273 {
3274         struct eth_tx_start_bd *h_tx_bd = *tx_bd;
3275         struct eth_tx_bd *d_tx_bd;
3276         dma_addr_t mapping;
3277         int old_len = le16_to_cpu(h_tx_bd->nbytes);
3278
3279         /* first fix first BD */
3280         h_tx_bd->nbytes = cpu_to_le16(hlen);
3281
3282         DP(NETIF_MSG_TX_QUEUED, "TSO split header size is %d (%x:%x)\n",
3283            h_tx_bd->nbytes, h_tx_bd->addr_hi, h_tx_bd->addr_lo);
3284
3285         /* now get a new data BD
3286          * (after the pbd) and fill it */
3287         bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3288         d_tx_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
3289
3290         mapping = HILO_U64(le32_to_cpu(h_tx_bd->addr_hi),
3291                            le32_to_cpu(h_tx_bd->addr_lo)) + hlen;
3292
3293         d_tx_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
3294         d_tx_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
3295         d_tx_bd->nbytes = cpu_to_le16(old_len - hlen);
3296
3297         /* this marks the BD as one that has no individual mapping */
3298         tx_buf->flags |= BNX2X_TSO_SPLIT_BD;
3299
3300         DP(NETIF_MSG_TX_QUEUED,
3301            "TSO split data size is %d (%x:%x)\n",
3302            d_tx_bd->nbytes, d_tx_bd->addr_hi, d_tx_bd->addr_lo);
3303
3304         /* update tx_bd */
3305         *tx_bd = (struct eth_tx_start_bd *)d_tx_bd;
3306
3307         return bd_prod;
3308 }
3309
3310 #define bswab32(b32) ((__force __le32) swab32((__force __u32) (b32)))
3311 #define bswab16(b16) ((__force __le16) swab16((__force __u16) (b16)))
3312 static __le16 bnx2x_csum_fix(unsigned char *t_header, u16 csum, s8 fix)
3313 {
3314         __sum16 tsum = (__force __sum16) csum;
3315
3316         if (fix > 0)
3317                 tsum = ~csum_fold(csum_sub((__force __wsum) csum,
3318                                   csum_partial(t_header - fix, fix, 0)));
3319
3320         else if (fix < 0)
3321                 tsum = ~csum_fold(csum_add((__force __wsum) csum,
3322                                   csum_partial(t_header, -fix, 0)));
3323
3324         return bswab16(tsum);
3325 }
3326
3327 static u32 bnx2x_xmit_type(struct bnx2x *bp, struct sk_buff *skb)
3328 {
3329         u32 rc;
3330         __u8 prot = 0;
3331         __be16 protocol;
3332
3333         if (skb->ip_summed != CHECKSUM_PARTIAL)
3334                 return XMIT_PLAIN;
3335
3336         protocol = vlan_get_protocol(skb);
3337         if (protocol == htons(ETH_P_IPV6)) {
3338                 rc = XMIT_CSUM_V6;
3339                 prot = ipv6_hdr(skb)->nexthdr;
3340         } else {
3341                 rc = XMIT_CSUM_V4;
3342                 prot = ip_hdr(skb)->protocol;
3343         }
3344
3345         if (!CHIP_IS_E1x(bp) && skb->encapsulation) {
3346                 if (inner_ip_hdr(skb)->version == 6) {
3347                         rc |= XMIT_CSUM_ENC_V6;
3348                         if (inner_ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3349                                 rc |= XMIT_CSUM_TCP;
3350                 } else {
3351                         rc |= XMIT_CSUM_ENC_V4;
3352                         if (inner_ip_hdr(skb)->protocol == IPPROTO_TCP)
3353                                 rc |= XMIT_CSUM_TCP;
3354                 }
3355         }
3356         if (prot == IPPROTO_TCP)
3357                 rc |= XMIT_CSUM_TCP;
3358
3359         if (skb_is_gso(skb)) {
3360                 if (skb_is_gso_v6(skb)) {
3361                         rc |= (XMIT_GSO_V6 | XMIT_CSUM_TCP);
3362                         if (rc & XMIT_CSUM_ENC)
3363                                 rc |= XMIT_GSO_ENC_V6;
3364                 } else {
3365                         rc |= (XMIT_GSO_V4 | XMIT_CSUM_TCP);
3366                         if (rc & XMIT_CSUM_ENC)
3367                                 rc |= XMIT_GSO_ENC_V4;
3368                 }
3369         }
3370
3371         return rc;
3372 }
3373
3374 #if (MAX_SKB_FRAGS >= MAX_FETCH_BD - 3)
3375 /* check if packet requires linearization (packet is too fragmented)
3376    no need to check fragmentation if page size > 8K (there will be no
3377    violation to FW restrictions) */
3378 static int bnx2x_pkt_req_lin(struct bnx2x *bp, struct sk_buff *skb,
3379                              u32 xmit_type)
3380 {
3381         int to_copy = 0;
3382         int hlen = 0;
3383         int first_bd_sz = 0;
3384
3385         /* 3 = 1 (for linear data BD) + 2 (for PBD and last BD) */
3386         if (skb_shinfo(skb)->nr_frags >= (MAX_FETCH_BD - 3)) {
3387
3388                 if (xmit_type & XMIT_GSO) {
3389                         unsigned short lso_mss = skb_shinfo(skb)->gso_size;
3390                         /* Check if LSO packet needs to be copied:
3391                            3 = 1 (for headers BD) + 2 (for PBD and last BD) */
3392                         int wnd_size = MAX_FETCH_BD - 3;
3393                         /* Number of windows to check */
3394                         int num_wnds = skb_shinfo(skb)->nr_frags - wnd_size;
3395                         int wnd_idx = 0;
3396                         int frag_idx = 0;
3397                         u32 wnd_sum = 0;
3398
3399                         /* Headers length */
3400                         hlen = (int)(skb_transport_header(skb) - skb->data) +
3401                                 tcp_hdrlen(skb);
3402
3403                         /* Amount of data (w/o headers) on linear part of SKB*/
3404                         first_bd_sz = skb_headlen(skb) - hlen;
3405
3406                         wnd_sum  = first_bd_sz;
3407
3408                         /* Calculate the first sum - it's special */
3409                         for (frag_idx = 0; frag_idx < wnd_size - 1; frag_idx++)
3410                                 wnd_sum +=
3411                                         skb_frag_size(&skb_shinfo(skb)->frags[frag_idx]);
3412
3413                         /* If there was data on linear skb data - check it */
3414                         if (first_bd_sz > 0) {
3415                                 if (unlikely(wnd_sum < lso_mss)) {
3416                                         to_copy = 1;
3417                                         goto exit_lbl;
3418                                 }
3419
3420                                 wnd_sum -= first_bd_sz;
3421                         }
3422
3423                         /* Others are easier: run through the frag list and
3424                            check all windows */
3425                         for (wnd_idx = 0; wnd_idx <= num_wnds; wnd_idx++) {
3426                                 wnd_sum +=
3427                           skb_frag_size(&skb_shinfo(skb)->frags[wnd_idx + wnd_size - 1]);
3428
3429                                 if (unlikely(wnd_sum < lso_mss)) {
3430                                         to_copy = 1;
3431                                         break;
3432                                 }
3433                                 wnd_sum -=
3434                                         skb_frag_size(&skb_shinfo(skb)->frags[wnd_idx]);
3435                         }
3436                 } else {
3437                         /* in non-LSO too fragmented packet should always
3438                            be linearized */
3439                         to_copy = 1;
3440                 }
3441         }
3442
3443 exit_lbl:
3444         if (unlikely(to_copy))
3445                 DP(NETIF_MSG_TX_QUEUED,
3446                    "Linearization IS REQUIRED for %s packet. num_frags %d  hlen %d  first_bd_sz %d\n",
3447                    (xmit_type & XMIT_GSO) ? "LSO" : "non-LSO",
3448                    skb_shinfo(skb)->nr_frags, hlen, first_bd_sz);
3449
3450         return to_copy;
3451 }
3452 #endif
3453
3454 /**
3455  * bnx2x_set_pbd_gso - update PBD in GSO case.
3456  *
3457  * @skb:        packet skb
3458  * @pbd:        parse BD
3459  * @xmit_type:  xmit flags
3460  */
3461 static void bnx2x_set_pbd_gso(struct sk_buff *skb,
3462                               struct eth_tx_parse_bd_e1x *pbd,
3463                               u32 xmit_type)
3464 {
3465         pbd->lso_mss = cpu_to_le16(skb_shinfo(skb)->gso_size);
3466         pbd->tcp_send_seq = bswab32(tcp_hdr(skb)->seq);
3467         pbd->tcp_flags = pbd_tcp_flags(tcp_hdr(skb));
3468
3469         if (xmit_type & XMIT_GSO_V4) {
3470                 pbd->ip_id = bswab16(ip_hdr(skb)->id);
3471                 pbd->tcp_pseudo_csum =
3472                         bswab16(~csum_tcpudp_magic(ip_hdr(skb)->saddr,
3473                                                    ip_hdr(skb)->daddr,
3474                                                    0, IPPROTO_TCP, 0));
3475         } else {
3476                 pbd->tcp_pseudo_csum =
3477                         bswab16(~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3478                                                  &ipv6_hdr(skb)->daddr,
3479                                                  0, IPPROTO_TCP, 0));
3480         }
3481
3482         pbd->global_data |=
3483                 cpu_to_le16(ETH_TX_PARSE_BD_E1X_PSEUDO_CS_WITHOUT_LEN);
3484 }
3485
3486 /**
3487  * bnx2x_set_pbd_csum_enc - update PBD with checksum and return header length
3488  *
3489  * @bp:                 driver handle
3490  * @skb:                packet skb
3491  * @parsing_data:       data to be updated
3492  * @xmit_type:          xmit flags
3493  *
3494  * 57712/578xx related, when skb has encapsulation
3495  */
3496 static u8 bnx2x_set_pbd_csum_enc(struct bnx2x *bp, struct sk_buff *skb,
3497                                  u32 *parsing_data, u32 xmit_type)
3498 {
3499         *parsing_data |=
3500                 ((((u8 *)skb_inner_transport_header(skb) - skb->data) >> 1) <<
3501                 ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W_SHIFT) &
3502                 ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W;
3503
3504         if (xmit_type & XMIT_CSUM_TCP) {
3505                 *parsing_data |= ((inner_tcp_hdrlen(skb) / 4) <<
3506                         ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW_SHIFT) &
3507                         ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW;
3508
3509                 return skb_inner_transport_header(skb) +
3510                         inner_tcp_hdrlen(skb) - skb->data;
3511         }
3512
3513         /* We support checksum offload for TCP and UDP only.
3514          * No need to pass the UDP header length - it's a constant.
3515          */
3516         return skb_inner_transport_header(skb) +
3517                 sizeof(struct udphdr) - skb->data;
3518 }
3519
3520 /**
3521  * bnx2x_set_pbd_csum_e2 - update PBD with checksum and return header length
3522  *
3523  * @bp:                 driver handle
3524  * @skb:                packet skb
3525  * @parsing_data:       data to be updated
3526  * @xmit_type:          xmit flags
3527  *
3528  * 57712/578xx related
3529  */
3530 static u8 bnx2x_set_pbd_csum_e2(struct bnx2x *bp, struct sk_buff *skb,
3531                                 u32 *parsing_data, u32 xmit_type)
3532 {
3533         *parsing_data |=
3534                 ((((u8 *)skb_transport_header(skb) - skb->data) >> 1) <<
3535                 ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W_SHIFT) &
3536                 ETH_TX_PARSE_BD_E2_L4_HDR_START_OFFSET_W;
3537
3538         if (xmit_type & XMIT_CSUM_TCP) {
3539                 *parsing_data |= ((tcp_hdrlen(skb) / 4) <<
3540                         ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW_SHIFT) &
3541                         ETH_TX_PARSE_BD_E2_TCP_HDR_LENGTH_DW;
3542
3543                 return skb_transport_header(skb) + tcp_hdrlen(skb) - skb->data;
3544         }
3545         /* We support checksum offload for TCP and UDP only.
3546          * No need to pass the UDP header length - it's a constant.
3547          */
3548         return skb_transport_header(skb) + sizeof(struct udphdr) - skb->data;
3549 }
3550
3551 /* set FW indication according to inner or outer protocols if tunneled */
3552 static void bnx2x_set_sbd_csum(struct bnx2x *bp, struct sk_buff *skb,
3553                                struct eth_tx_start_bd *tx_start_bd,
3554                                u32 xmit_type)
3555 {
3556         tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_L4_CSUM;
3557
3558         if (xmit_type & (XMIT_CSUM_ENC_V6 | XMIT_CSUM_V6))
3559                 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_IPV6;
3560
3561         if (!(xmit_type & XMIT_CSUM_TCP))
3562                 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_IS_UDP;
3563 }
3564
3565 /**
3566  * bnx2x_set_pbd_csum - update PBD with checksum and return header length
3567  *
3568  * @bp:         driver handle
3569  * @skb:        packet skb
3570  * @pbd:        parse BD to be updated
3571  * @xmit_type:  xmit flags
3572  */
3573 static u8 bnx2x_set_pbd_csum(struct bnx2x *bp, struct sk_buff *skb,
3574                              struct eth_tx_parse_bd_e1x *pbd,
3575                              u32 xmit_type)
3576 {
3577         u8 hlen = (skb_network_header(skb) - skb->data) >> 1;
3578
3579         /* for now NS flag is not used in Linux */
3580         pbd->global_data =
3581                 cpu_to_le16(hlen |
3582                             ((skb->protocol == cpu_to_be16(ETH_P_8021Q)) <<
3583                              ETH_TX_PARSE_BD_E1X_LLC_SNAP_EN_SHIFT));
3584
3585         pbd->ip_hlen_w = (skb_transport_header(skb) -
3586                         skb_network_header(skb)) >> 1;
3587
3588         hlen += pbd->ip_hlen_w;
3589
3590         /* We support checksum offload for TCP and UDP only */
3591         if (xmit_type & XMIT_CSUM_TCP)
3592                 hlen += tcp_hdrlen(skb) / 2;
3593         else
3594                 hlen += sizeof(struct udphdr) / 2;
3595
3596         pbd->total_hlen_w = cpu_to_le16(hlen);
3597         hlen = hlen*2;
3598
3599         if (xmit_type & XMIT_CSUM_TCP) {
3600                 pbd->tcp_pseudo_csum = bswab16(tcp_hdr(skb)->check);
3601
3602         } else {
3603                 s8 fix = SKB_CS_OFF(skb); /* signed! */
3604
3605                 DP(NETIF_MSG_TX_QUEUED,
3606                    "hlen %d  fix %d  csum before fix %x\n",
3607                    le16_to_cpu(pbd->total_hlen_w), fix, SKB_CS(skb));
3608
3609                 /* HW bug: fixup the CSUM */
3610                 pbd->tcp_pseudo_csum =
3611                         bnx2x_csum_fix(skb_transport_header(skb),
3612                                        SKB_CS(skb), fix);
3613
3614                 DP(NETIF_MSG_TX_QUEUED, "csum after fix %x\n",
3615                    pbd->tcp_pseudo_csum);
3616         }
3617
3618         return hlen;
3619 }
3620
3621 static void bnx2x_update_pbds_gso_enc(struct sk_buff *skb,
3622                                       struct eth_tx_parse_bd_e2 *pbd_e2,
3623                                       struct eth_tx_parse_2nd_bd *pbd2,
3624                                       u16 *global_data,
3625                                       u32 xmit_type)
3626 {
3627         u16 hlen_w = 0;
3628         u8 outerip_off, outerip_len = 0;
3629
3630         /* from outer IP to transport */
3631         hlen_w = (skb_inner_transport_header(skb) -
3632                   skb_network_header(skb)) >> 1;
3633
3634         /* transport len */
3635         hlen_w += inner_tcp_hdrlen(skb) >> 1;
3636
3637         pbd2->fw_ip_hdr_to_payload_w = hlen_w;
3638
3639         /* outer IP header info */
3640         if (xmit_type & XMIT_CSUM_V4) {
3641                 struct iphdr *iph = ip_hdr(skb);
3642                 u32 csum = (__force u32)(~iph->check) -
3643                            (__force u32)iph->tot_len -
3644                            (__force u32)iph->frag_off;
3645
3646                 outerip_len = iph->ihl << 1;
3647
3648                 pbd2->fw_ip_csum_wo_len_flags_frag =
3649                         bswab16(csum_fold((__force __wsum)csum));
3650         } else {
3651                 pbd2->fw_ip_hdr_to_payload_w =
3652                         hlen_w - ((sizeof(struct ipv6hdr)) >> 1);
3653                 pbd_e2->data.tunnel_data.flags |=
3654                         ETH_TUNNEL_DATA_IP_HDR_TYPE_OUTER;
3655         }
3656
3657         pbd2->tcp_send_seq = bswab32(inner_tcp_hdr(skb)->seq);
3658
3659         pbd2->tcp_flags = pbd_tcp_flags(inner_tcp_hdr(skb));
3660
3661         /* inner IP header info */
3662         if (xmit_type & XMIT_CSUM_ENC_V4) {
3663                 pbd2->hw_ip_id = bswab16(inner_ip_hdr(skb)->id);
3664
3665                 pbd_e2->data.tunnel_data.pseudo_csum =
3666                         bswab16(~csum_tcpudp_magic(
3667                                         inner_ip_hdr(skb)->saddr,
3668                                         inner_ip_hdr(skb)->daddr,
3669                                         0, IPPROTO_TCP, 0));
3670         } else {
3671                 pbd_e2->data.tunnel_data.pseudo_csum =
3672                         bswab16(~csum_ipv6_magic(
3673                                         &inner_ipv6_hdr(skb)->saddr,
3674                                         &inner_ipv6_hdr(skb)->daddr,
3675                                         0, IPPROTO_TCP, 0));
3676         }
3677
3678         outerip_off = (skb_network_header(skb) - skb->data) >> 1;
3679
3680         *global_data |=
3681                 outerip_off |
3682                 (outerip_len <<
3683                         ETH_TX_PARSE_2ND_BD_IP_HDR_LEN_OUTER_W_SHIFT) |
3684                 ((skb->protocol == cpu_to_be16(ETH_P_8021Q)) <<
3685                         ETH_TX_PARSE_2ND_BD_LLC_SNAP_EN_SHIFT);
3686
3687         if (ip_hdr(skb)->protocol == IPPROTO_UDP) {
3688                 SET_FLAG(*global_data, ETH_TX_PARSE_2ND_BD_TUNNEL_UDP_EXIST, 1);
3689                 pbd2->tunnel_udp_hdr_start_w = skb_transport_offset(skb) >> 1;
3690         }
3691 }
3692
3693 static inline void bnx2x_set_ipv6_ext_e2(struct sk_buff *skb, u32 *parsing_data,
3694                                          u32 xmit_type)
3695 {
3696         struct ipv6hdr *ipv6;
3697
3698         if (!(xmit_type & (XMIT_GSO_ENC_V6 | XMIT_GSO_V6)))
3699                 return;
3700
3701         if (xmit_type & XMIT_GSO_ENC_V6)
3702                 ipv6 = inner_ipv6_hdr(skb);
3703         else /* XMIT_GSO_V6 */
3704                 ipv6 = ipv6_hdr(skb);
3705
3706         if (ipv6->nexthdr == NEXTHDR_IPV6)
3707                 *parsing_data |= ETH_TX_PARSE_BD_E2_IPV6_WITH_EXT_HDR;
3708 }
3709
3710 /* called with netif_tx_lock
3711  * bnx2x_tx_int() runs without netif_tx_lock unless it needs to call
3712  * netif_wake_queue()
3713  */
3714 netdev_tx_t bnx2x_start_xmit(struct sk_buff *skb, struct net_device *dev)
3715 {
3716         struct bnx2x *bp = netdev_priv(dev);
3717
3718         struct netdev_queue *txq;
3719         struct bnx2x_fp_txdata *txdata;
3720         struct sw_tx_bd *tx_buf;
3721         struct eth_tx_start_bd *tx_start_bd, *first_bd;
3722         struct eth_tx_bd *tx_data_bd, *total_pkt_bd = NULL;
3723         struct eth_tx_parse_bd_e1x *pbd_e1x = NULL;
3724         struct eth_tx_parse_bd_e2 *pbd_e2 = NULL;
3725         struct eth_tx_parse_2nd_bd *pbd2 = NULL;
3726         u32 pbd_e2_parsing_data = 0;
3727         u16 pkt_prod, bd_prod;
3728         int nbd, txq_index;
3729         dma_addr_t mapping;
3730         u32 xmit_type = bnx2x_xmit_type(bp, skb);
3731         int i;
3732         u8 hlen = 0;
3733         __le16 pkt_size = 0;
3734         struct ethhdr *eth;
3735         u8 mac_type = UNICAST_ADDRESS;
3736
3737 #ifdef BNX2X_STOP_ON_ERROR
3738         if (unlikely(bp->panic))
3739                 return NETDEV_TX_BUSY;
3740 #endif
3741
3742         txq_index = skb_get_queue_mapping(skb);
3743         txq = netdev_get_tx_queue(dev, txq_index);
3744
3745         BUG_ON(txq_index >= MAX_ETH_TXQ_IDX(bp) + (CNIC_LOADED(bp) ? 1 : 0));
3746
3747         txdata = &bp->bnx2x_txq[txq_index];
3748
3749         /* enable this debug print to view the transmission queue being used
3750         DP(NETIF_MSG_TX_QUEUED, "indices: txq %d, fp %d, txdata %d\n",
3751            txq_index, fp_index, txdata_index); */
3752
3753         /* enable this debug print to view the transmission details
3754         DP(NETIF_MSG_TX_QUEUED,
3755            "transmitting packet cid %d fp index %d txdata_index %d tx_data ptr %p fp pointer %p\n",
3756            txdata->cid, fp_index, txdata_index, txdata, fp); */
3757
3758         if (unlikely(bnx2x_tx_avail(bp, txdata) <
3759                         skb_shinfo(skb)->nr_frags +
3760                         BDS_PER_TX_PKT +
3761                         NEXT_CNT_PER_TX_PKT(MAX_BDS_PER_TX_PKT))) {
3762                 /* Handle special storage cases separately */
3763                 if (txdata->tx_ring_size == 0) {
3764                         struct bnx2x_eth_q_stats *q_stats =
3765                                 bnx2x_fp_qstats(bp, txdata->parent_fp);
3766                         q_stats->driver_filtered_tx_pkt++;
3767                         dev_kfree_skb(skb);
3768                         return NETDEV_TX_OK;
3769                 }
3770                 bnx2x_fp_qstats(bp, txdata->parent_fp)->driver_xoff++;
3771                 netif_tx_stop_queue(txq);
3772                 BNX2X_ERR("BUG! Tx ring full when queue awake!\n");
3773
3774                 return NETDEV_TX_BUSY;
3775         }
3776
3777         DP(NETIF_MSG_TX_QUEUED,
3778            "queue[%d]: SKB: summed %x  protocol %x protocol(%x,%x) gso type %x  xmit_type %x len %d\n",
3779            txq_index, skb->ip_summed, skb->protocol, ipv6_hdr(skb)->nexthdr,
3780            ip_hdr(skb)->protocol, skb_shinfo(skb)->gso_type, xmit_type,
3781            skb->len);
3782
3783         eth = (struct ethhdr *)skb->data;
3784
3785         /* set flag according to packet type (UNICAST_ADDRESS is default)*/
3786         if (unlikely(is_multicast_ether_addr(eth->h_dest))) {
3787                 if (is_broadcast_ether_addr(eth->h_dest))
3788                         mac_type = BROADCAST_ADDRESS;
3789                 else
3790                         mac_type = MULTICAST_ADDRESS;
3791         }
3792
3793 #if (MAX_SKB_FRAGS >= MAX_FETCH_BD - BDS_PER_TX_PKT)
3794         /* First, check if we need to linearize the skb (due to FW
3795            restrictions). No need to check fragmentation if page size > 8K
3796            (there will be no violation to FW restrictions) */
3797         if (bnx2x_pkt_req_lin(bp, skb, xmit_type)) {
3798                 /* Statistics of linearization */
3799                 bp->lin_cnt++;
3800                 if (skb_linearize(skb) != 0) {
3801                         DP(NETIF_MSG_TX_QUEUED,
3802                            "SKB linearization failed - silently dropping this SKB\n");
3803                         dev_kfree_skb_any(skb);
3804                         return NETDEV_TX_OK;
3805                 }
3806         }
3807 #endif
3808         /* Map skb linear data for DMA */
3809         mapping = dma_map_single(&bp->pdev->dev, skb->data,
3810                                  skb_headlen(skb), DMA_TO_DEVICE);
3811         if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
3812                 DP(NETIF_MSG_TX_QUEUED,
3813                    "SKB mapping failed - silently dropping this SKB\n");
3814                 dev_kfree_skb_any(skb);
3815                 return NETDEV_TX_OK;
3816         }
3817         /*
3818         Please read carefully. First we use one BD which we mark as start,
3819         then we have a parsing info BD (used for TSO or xsum),
3820         and only then we have the rest of the TSO BDs.
3821         (don't forget to mark the last one as last,
3822         and to unmap only AFTER you write to the BD ...)
3823         And above all, all pdb sizes are in words - NOT DWORDS!
3824         */
3825
3826         /* get current pkt produced now - advance it just before sending packet
3827          * since mapping of pages may fail and cause packet to be dropped
3828          */
3829         pkt_prod = txdata->tx_pkt_prod;
3830         bd_prod = TX_BD(txdata->tx_bd_prod);
3831
3832         /* get a tx_buf and first BD
3833          * tx_start_bd may be changed during SPLIT,
3834          * but first_bd will always stay first
3835          */
3836         tx_buf = &txdata->tx_buf_ring[TX_BD(pkt_prod)];
3837         tx_start_bd = &txdata->tx_desc_ring[bd_prod].start_bd;
3838         first_bd = tx_start_bd;
3839
3840         tx_start_bd->bd_flags.as_bitfield = ETH_TX_BD_FLAGS_START_BD;
3841
3842         if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) {
3843                 if (!(bp->flags & TX_TIMESTAMPING_EN)) {
3844                         BNX2X_ERR("Tx timestamping was not enabled, this packet will not be timestamped\n");
3845                 } else if (bp->ptp_tx_skb) {
3846                         BNX2X_ERR("The device supports only a single outstanding packet to timestamp, this packet will not be timestamped\n");
3847                 } else {
3848                         skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
3849                         /* schedule check for Tx timestamp */
3850                         bp->ptp_tx_skb = skb_get(skb);
3851                         bp->ptp_tx_start = jiffies;
3852                         schedule_work(&bp->ptp_task);
3853                 }
3854         }
3855
3856         /* header nbd: indirectly zero other flags! */
3857         tx_start_bd->general_data = 1 << ETH_TX_START_BD_HDR_NBDS_SHIFT;
3858
3859         /* remember the first BD of the packet */
3860         tx_buf->first_bd = txdata->tx_bd_prod;
3861         tx_buf->skb = skb;
3862         tx_buf->flags = 0;
3863
3864         DP(NETIF_MSG_TX_QUEUED,
3865            "sending pkt %u @%p  next_idx %u  bd %u @%p\n",
3866            pkt_prod, tx_buf, txdata->tx_pkt_prod, bd_prod, tx_start_bd);
3867
3868         if (vlan_tx_tag_present(skb)) {
3869                 tx_start_bd->vlan_or_ethertype =
3870                     cpu_to_le16(vlan_tx_tag_get(skb));
3871                 tx_start_bd->bd_flags.as_bitfield |=
3872                     (X_ETH_OUTBAND_VLAN << ETH_TX_BD_FLAGS_VLAN_MODE_SHIFT);
3873         } else {
3874                 /* when transmitting in a vf, start bd must hold the ethertype
3875                  * for fw to enforce it
3876                  */
3877 #ifndef BNX2X_STOP_ON_ERROR
3878                 if (IS_VF(bp))
3879 #endif
3880                         tx_start_bd->vlan_or_ethertype =
3881                                 cpu_to_le16(ntohs(eth->h_proto));
3882 #ifndef BNX2X_STOP_ON_ERROR
3883                 else
3884                         /* used by FW for packet accounting */
3885                         tx_start_bd->vlan_or_ethertype = cpu_to_le16(pkt_prod);
3886 #endif
3887         }
3888
3889         nbd = 2; /* start_bd + pbd + frags (updated when pages are mapped) */
3890
3891         /* turn on parsing and get a BD */
3892         bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3893
3894         if (xmit_type & XMIT_CSUM)
3895                 bnx2x_set_sbd_csum(bp, skb, tx_start_bd, xmit_type);
3896
3897         if (!CHIP_IS_E1x(bp)) {
3898                 pbd_e2 = &txdata->tx_desc_ring[bd_prod].parse_bd_e2;
3899                 memset(pbd_e2, 0, sizeof(struct eth_tx_parse_bd_e2));
3900
3901                 if (xmit_type & XMIT_CSUM_ENC) {
3902                         u16 global_data = 0;
3903
3904                         /* Set PBD in enc checksum offload case */
3905                         hlen = bnx2x_set_pbd_csum_enc(bp, skb,
3906                                                       &pbd_e2_parsing_data,
3907                                                       xmit_type);
3908
3909                         /* turn on 2nd parsing and get a BD */
3910                         bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
3911
3912                         pbd2 = &txdata->tx_desc_ring[bd_prod].parse_2nd_bd;
3913
3914                         memset(pbd2, 0, sizeof(*pbd2));
3915
3916                         pbd_e2->data.tunnel_data.ip_hdr_start_inner_w =
3917                                 (skb_inner_network_header(skb) -
3918                                  skb->data) >> 1;
3919
3920                         if (xmit_type & XMIT_GSO_ENC)
3921                                 bnx2x_update_pbds_gso_enc(skb, pbd_e2, pbd2,
3922                                                           &global_data,
3923                                                           xmit_type);
3924
3925                         pbd2->global_data = cpu_to_le16(global_data);
3926
3927                         /* add addition parse BD indication to start BD */
3928                         SET_FLAG(tx_start_bd->general_data,
3929                                  ETH_TX_START_BD_PARSE_NBDS, 1);
3930                         /* set encapsulation flag in start BD */
3931                         SET_FLAG(tx_start_bd->general_data,
3932                                  ETH_TX_START_BD_TUNNEL_EXIST, 1);
3933
3934                         tx_buf->flags |= BNX2X_HAS_SECOND_PBD;
3935
3936                         nbd++;
3937                 } else if (xmit_type & XMIT_CSUM) {
3938                         /* Set PBD in checksum offload case w/o encapsulation */
3939                         hlen = bnx2x_set_pbd_csum_e2(bp, skb,
3940                                                      &pbd_e2_parsing_data,
3941                                                      xmit_type);
3942                 }
3943
3944                 bnx2x_set_ipv6_ext_e2(skb, &pbd_e2_parsing_data, xmit_type);
3945                 /* Add the macs to the parsing BD if this is a vf or if
3946                  * Tx Switching is enabled.
3947                  */
3948                 if (IS_VF(bp)) {
3949                         /* override GRE parameters in BD */
3950                         bnx2x_set_fw_mac_addr(&pbd_e2->data.mac_addr.src_hi,
3951                                               &pbd_e2->data.mac_addr.src_mid,
3952                                               &pbd_e2->data.mac_addr.src_lo,
3953                                               eth->h_source);
3954
3955                         bnx2x_set_fw_mac_addr(&pbd_e2->data.mac_addr.dst_hi,
3956                                               &pbd_e2->data.mac_addr.dst_mid,
3957                                               &pbd_e2->data.mac_addr.dst_lo,
3958                                               eth->h_dest);
3959                 } else {
3960                         if (bp->flags & TX_SWITCHING)
3961                                 bnx2x_set_fw_mac_addr(
3962                                                 &pbd_e2->data.mac_addr.dst_hi,
3963                                                 &pbd_e2->data.mac_addr.dst_mid,
3964                                                 &pbd_e2->data.mac_addr.dst_lo,
3965                                                 eth->h_dest);
3966 #ifdef BNX2X_STOP_ON_ERROR
3967                         /* Enforce security is always set in Stop on Error -
3968                          * source mac should be present in the parsing BD
3969                          */
3970                         bnx2x_set_fw_mac_addr(&pbd_e2->data.mac_addr.src_hi,
3971                                               &pbd_e2->data.mac_addr.src_mid,
3972                                               &pbd_e2->data.mac_addr.src_lo,
3973                                               eth->h_source);
3974 #endif
3975                 }
3976
3977                 SET_FLAG(pbd_e2_parsing_data,
3978                          ETH_TX_PARSE_BD_E2_ETH_ADDR_TYPE, mac_type);
3979         } else {
3980                 u16 global_data = 0;
3981                 pbd_e1x = &txdata->tx_desc_ring[bd_prod].parse_bd_e1x;
3982                 memset(pbd_e1x, 0, sizeof(struct eth_tx_parse_bd_e1x));
3983                 /* Set PBD in checksum offload case */
3984                 if (xmit_type & XMIT_CSUM)
3985                         hlen = bnx2x_set_pbd_csum(bp, skb, pbd_e1x, xmit_type);
3986
3987                 SET_FLAG(global_data,
3988                          ETH_TX_PARSE_BD_E1X_ETH_ADDR_TYPE, mac_type);
3989                 pbd_e1x->global_data |= cpu_to_le16(global_data);
3990         }
3991
3992         /* Setup the data pointer of the first BD of the packet */
3993         tx_start_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
3994         tx_start_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
3995         tx_start_bd->nbytes = cpu_to_le16(skb_headlen(skb));
3996         pkt_size = tx_start_bd->nbytes;
3997
3998         DP(NETIF_MSG_TX_QUEUED,
3999            "first bd @%p  addr (%x:%x)  nbytes %d  flags %x  vlan %x\n",
4000            tx_start_bd, tx_start_bd->addr_hi, tx_start_bd->addr_lo,
4001            le16_to_cpu(tx_start_bd->nbytes),
4002            tx_start_bd->bd_flags.as_bitfield,
4003            le16_to_cpu(tx_start_bd->vlan_or_ethertype));
4004
4005         if (xmit_type & XMIT_GSO) {
4006
4007                 DP(NETIF_MSG_TX_QUEUED,
4008                    "TSO packet len %d  hlen %d  total len %d  tso size %d\n",
4009                    skb->len, hlen, skb_headlen(skb),
4010                    skb_shinfo(skb)->gso_size);
4011
4012                 tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_SW_LSO;
4013
4014                 if (unlikely(skb_headlen(skb) > hlen)) {
4015                         nbd++;
4016                         bd_prod = bnx2x_tx_split(bp, txdata, tx_buf,
4017                                                  &tx_start_bd, hlen,
4018                                                  bd_prod);
4019                 }
4020                 if (!CHIP_IS_E1x(bp))
4021                         pbd_e2_parsing_data |=
4022                                 (skb_shinfo(skb)->gso_size <<
4023                                  ETH_TX_PARSE_BD_E2_LSO_MSS_SHIFT) &
4024                                  ETH_TX_PARSE_BD_E2_LSO_MSS;
4025                 else
4026                         bnx2x_set_pbd_gso(skb, pbd_e1x, xmit_type);
4027         }
4028
4029         /* Set the PBD's parsing_data field if not zero
4030          * (for the chips newer than 57711).
4031          */
4032         if (pbd_e2_parsing_data)
4033                 pbd_e2->parsing_data = cpu_to_le32(pbd_e2_parsing_data);
4034
4035         tx_data_bd = (struct eth_tx_bd *)tx_start_bd;
4036
4037         /* Handle fragmented skb */
4038         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
4039                 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
4040
4041                 mapping = skb_frag_dma_map(&bp->pdev->dev, frag, 0,
4042                                            skb_frag_size(frag), DMA_TO_DEVICE);
4043                 if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
4044                         unsigned int pkts_compl = 0, bytes_compl = 0;
4045
4046                         DP(NETIF_MSG_TX_QUEUED,
4047                            "Unable to map page - dropping packet...\n");
4048
4049                         /* we need unmap all buffers already mapped
4050                          * for this SKB;
4051                          * first_bd->nbd need to be properly updated
4052                          * before call to bnx2x_free_tx_pkt
4053                          */
4054                         first_bd->nbd = cpu_to_le16(nbd);
4055                         bnx2x_free_tx_pkt(bp, txdata,
4056                                           TX_BD(txdata->tx_pkt_prod),
4057                                           &pkts_compl, &bytes_compl);
4058                         return NETDEV_TX_OK;
4059                 }
4060
4061                 bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
4062                 tx_data_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
4063                 if (total_pkt_bd == NULL)
4064                         total_pkt_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
4065
4066                 tx_data_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
4067                 tx_data_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
4068                 tx_data_bd->nbytes = cpu_to_le16(skb_frag_size(frag));
4069                 le16_add_cpu(&pkt_size, skb_frag_size(frag));
4070                 nbd++;
4071
4072                 DP(NETIF_MSG_TX_QUEUED,
4073                    "frag %d  bd @%p  addr (%x:%x)  nbytes %d\n",
4074                    i, tx_data_bd, tx_data_bd->addr_hi, tx_data_bd->addr_lo,
4075                    le16_to_cpu(tx_data_bd->nbytes));
4076         }
4077
4078         DP(NETIF_MSG_TX_QUEUED, "last bd @%p\n", tx_data_bd);
4079
4080         /* update with actual num BDs */
4081         first_bd->nbd = cpu_to_le16(nbd);
4082
4083         bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
4084
4085         /* now send a tx doorbell, counting the next BD
4086          * if the packet contains or ends with it
4087          */
4088         if (TX_BD_POFF(bd_prod) < nbd)
4089                 nbd++;
4090
4091         /* total_pkt_bytes should be set on the first data BD if
4092          * it's not an LSO packet and there is more than one
4093          * data BD. In this case pkt_size is limited by an MTU value.
4094          * However we prefer to set it for an LSO packet (while we don't
4095          * have to) in order to save some CPU cycles in a none-LSO
4096          * case, when we much more care about them.
4097          */
4098         if (total_pkt_bd != NULL)
4099                 total_pkt_bd->total_pkt_bytes = pkt_size;
4100
4101         if (pbd_e1x)
4102                 DP(NETIF_MSG_TX_QUEUED,
4103                    "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",
4104                    pbd_e1x, pbd_e1x->global_data, pbd_e1x->ip_hlen_w,
4105                    pbd_e1x->ip_id, pbd_e1x->lso_mss, pbd_e1x->tcp_flags,
4106                    pbd_e1x->tcp_pseudo_csum, pbd_e1x->tcp_send_seq,
4107                     le16_to_cpu(pbd_e1x->total_hlen_w));
4108         if (pbd_e2)
4109                 DP(NETIF_MSG_TX_QUEUED,
4110                    "PBD (E2) @%p  dst %x %x %x src %x %x %x parsing_data %x\n",
4111                    pbd_e2,
4112                    pbd_e2->data.mac_addr.dst_hi,
4113                    pbd_e2->data.mac_addr.dst_mid,
4114                    pbd_e2->data.mac_addr.dst_lo,
4115                    pbd_e2->data.mac_addr.src_hi,
4116                    pbd_e2->data.mac_addr.src_mid,
4117                    pbd_e2->data.mac_addr.src_lo,
4118                    pbd_e2->parsing_data);
4119         DP(NETIF_MSG_TX_QUEUED, "doorbell: nbd %d  bd %u\n", nbd, bd_prod);
4120
4121         netdev_tx_sent_queue(txq, skb->len);
4122
4123         skb_tx_timestamp(skb);
4124
4125         txdata->tx_pkt_prod++;
4126         /*
4127          * Make sure that the BD data is updated before updating the producer
4128          * since FW might read the BD right after the producer is updated.
4129          * This is only applicable for weak-ordered memory model archs such
4130          * as IA-64. The following barrier is also mandatory since FW will
4131          * assumes packets must have BDs.
4132          */
4133         wmb();
4134
4135         txdata->tx_db.data.prod += nbd;
4136         barrier();
4137
4138         DOORBELL(bp, txdata->cid, txdata->tx_db.raw);
4139
4140         mmiowb();
4141
4142         txdata->tx_bd_prod += nbd;
4143
4144         if (unlikely(bnx2x_tx_avail(bp, txdata) < MAX_DESC_PER_TX_PKT)) {
4145                 netif_tx_stop_queue(txq);
4146
4147                 /* paired memory barrier is in bnx2x_tx_int(), we have to keep
4148                  * ordering of set_bit() in netif_tx_stop_queue() and read of
4149                  * fp->bd_tx_cons */
4150                 smp_mb();
4151
4152                 bnx2x_fp_qstats(bp, txdata->parent_fp)->driver_xoff++;
4153                 if (bnx2x_tx_avail(bp, txdata) >= MAX_DESC_PER_TX_PKT)
4154                         netif_tx_wake_queue(txq);
4155         }
4156         txdata->tx_pkt++;
4157
4158         return NETDEV_TX_OK;
4159 }
4160
4161 /**
4162  * bnx2x_setup_tc - routine to configure net_device for multi tc
4163  *
4164  * @netdev: net device to configure
4165  * @tc: number of traffic classes to enable
4166  *
4167  * callback connected to the ndo_setup_tc function pointer
4168  */
4169 int bnx2x_setup_tc(struct net_device *dev, u8 num_tc)
4170 {
4171         int cos, prio, count, offset;
4172         struct bnx2x *bp = netdev_priv(dev);
4173
4174         /* setup tc must be called under rtnl lock */
4175         ASSERT_RTNL();
4176
4177         /* no traffic classes requested. Aborting */
4178         if (!num_tc) {
4179                 netdev_reset_tc(dev);
4180                 return 0;
4181         }
4182
4183         /* requested to support too many traffic classes */
4184         if (num_tc > bp->max_cos) {
4185                 BNX2X_ERR("support for too many traffic classes requested: %d. Max supported is %d\n",
4186                           num_tc, bp->max_cos);
4187                 return -EINVAL;
4188         }
4189
4190         /* declare amount of supported traffic classes */
4191         if (netdev_set_num_tc(dev, num_tc)) {
4192                 BNX2X_ERR("failed to declare %d traffic classes\n", num_tc);
4193                 return -EINVAL;
4194         }
4195
4196         /* configure priority to traffic class mapping */
4197         for (prio = 0; prio < BNX2X_MAX_PRIORITY; prio++) {
4198                 netdev_set_prio_tc_map(dev, prio, bp->prio_to_cos[prio]);
4199                 DP(BNX2X_MSG_SP | NETIF_MSG_IFUP,
4200                    "mapping priority %d to tc %d\n",
4201                    prio, bp->prio_to_cos[prio]);
4202         }
4203
4204         /* Use this configuration to differentiate tc0 from other COSes
4205            This can be used for ets or pfc, and save the effort of setting
4206            up a multio class queue disc or negotiating DCBX with a switch
4207         netdev_set_prio_tc_map(dev, 0, 0);
4208         DP(BNX2X_MSG_SP, "mapping priority %d to tc %d\n", 0, 0);
4209         for (prio = 1; prio < 16; prio++) {
4210                 netdev_set_prio_tc_map(dev, prio, 1);
4211                 DP(BNX2X_MSG_SP, "mapping priority %d to tc %d\n", prio, 1);
4212         } */
4213
4214         /* configure traffic class to transmission queue mapping */
4215         for (cos = 0; cos < bp->max_cos; cos++) {
4216                 count = BNX2X_NUM_ETH_QUEUES(bp);
4217                 offset = cos * BNX2X_NUM_NON_CNIC_QUEUES(bp);
4218                 netdev_set_tc_queue(dev, cos, count, offset);
4219                 DP(BNX2X_MSG_SP | NETIF_MSG_IFUP,
4220                    "mapping tc %d to offset %d count %d\n",
4221                    cos, offset, count);
4222         }
4223
4224         return 0;
4225 }
4226
4227 /* called with rtnl_lock */
4228 int bnx2x_change_mac_addr(struct net_device *dev, void *p)
4229 {
4230         struct sockaddr *addr = p;
4231         struct bnx2x *bp = netdev_priv(dev);
4232         int rc = 0;
4233
4234         if (!is_valid_ether_addr(addr->sa_data)) {
4235                 BNX2X_ERR("Requested MAC address is not valid\n");
4236                 return -EINVAL;
4237         }
4238
4239         if (IS_MF_STORAGE_ONLY(bp)) {
4240                 BNX2X_ERR("Can't change address on STORAGE ONLY function\n");
4241                 return -EINVAL;
4242         }
4243
4244         if (netif_running(dev))  {
4245                 rc = bnx2x_set_eth_mac(bp, false);
4246                 if (rc)
4247                         return rc;
4248         }
4249
4250         memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
4251
4252         if (netif_running(dev))
4253                 rc = bnx2x_set_eth_mac(bp, true);
4254
4255         return rc;
4256 }
4257
4258 static void bnx2x_free_fp_mem_at(struct bnx2x *bp, int fp_index)
4259 {
4260         union host_hc_status_block *sb = &bnx2x_fp(bp, fp_index, status_blk);
4261         struct bnx2x_fastpath *fp = &bp->fp[fp_index];
4262         u8 cos;
4263
4264         /* Common */
4265
4266         if (IS_FCOE_IDX(fp_index)) {
4267                 memset(sb, 0, sizeof(union host_hc_status_block));
4268                 fp->status_blk_mapping = 0;
4269         } else {
4270                 /* status blocks */
4271                 if (!CHIP_IS_E1x(bp))
4272                         BNX2X_PCI_FREE(sb->e2_sb,
4273                                        bnx2x_fp(bp, fp_index,
4274                                                 status_blk_mapping),
4275                                        sizeof(struct host_hc_status_block_e2));
4276                 else
4277                         BNX2X_PCI_FREE(sb->e1x_sb,
4278                                        bnx2x_fp(bp, fp_index,
4279                                                 status_blk_mapping),
4280                                        sizeof(struct host_hc_status_block_e1x));
4281         }
4282
4283         /* Rx */
4284         if (!skip_rx_queue(bp, fp_index)) {
4285                 bnx2x_free_rx_bds(fp);
4286
4287                 /* fastpath rx rings: rx_buf rx_desc rx_comp */
4288                 BNX2X_FREE(bnx2x_fp(bp, fp_index, rx_buf_ring));
4289                 BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_desc_ring),
4290                                bnx2x_fp(bp, fp_index, rx_desc_mapping),
4291                                sizeof(struct eth_rx_bd) * NUM_RX_BD);
4292
4293                 BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_comp_ring),
4294                                bnx2x_fp(bp, fp_index, rx_comp_mapping),
4295                                sizeof(struct eth_fast_path_rx_cqe) *
4296                                NUM_RCQ_BD);
4297
4298                 /* SGE ring */
4299                 BNX2X_FREE(bnx2x_fp(bp, fp_index, rx_page_ring));
4300                 BNX2X_PCI_FREE(bnx2x_fp(bp, fp_index, rx_sge_ring),
4301                                bnx2x_fp(bp, fp_index, rx_sge_mapping),
4302                                BCM_PAGE_SIZE * NUM_RX_SGE_PAGES);
4303         }
4304
4305         /* Tx */
4306         if (!skip_tx_queue(bp, fp_index)) {
4307                 /* fastpath tx rings: tx_buf tx_desc */
4308                 for_each_cos_in_tx_queue(fp, cos) {
4309                         struct bnx2x_fp_txdata *txdata = fp->txdata_ptr[cos];
4310
4311                         DP(NETIF_MSG_IFDOWN,
4312                            "freeing tx memory of fp %d cos %d cid %d\n",
4313                            fp_index, cos, txdata->cid);
4314
4315                         BNX2X_FREE(txdata->tx_buf_ring);
4316                         BNX2X_PCI_FREE(txdata->tx_desc_ring,
4317                                 txdata->tx_desc_mapping,
4318                                 sizeof(union eth_tx_bd_types) * NUM_TX_BD);
4319                 }
4320         }
4321         /* end of fastpath */
4322 }
4323
4324 static void bnx2x_free_fp_mem_cnic(struct bnx2x *bp)
4325 {
4326         int i;
4327         for_each_cnic_queue(bp, i)
4328                 bnx2x_free_fp_mem_at(bp, i);
4329 }
4330
4331 void bnx2x_free_fp_mem(struct bnx2x *bp)
4332 {
4333         int i;
4334         for_each_eth_queue(bp, i)
4335                 bnx2x_free_fp_mem_at(bp, i);
4336 }
4337
4338 static void set_sb_shortcuts(struct bnx2x *bp, int index)
4339 {
4340         union host_hc_status_block status_blk = bnx2x_fp(bp, index, status_blk);
4341         if (!CHIP_IS_E1x(bp)) {
4342                 bnx2x_fp(bp, index, sb_index_values) =
4343                         (__le16 *)status_blk.e2_sb->sb.index_values;
4344                 bnx2x_fp(bp, index, sb_running_index) =
4345                         (__le16 *)status_blk.e2_sb->sb.running_index;
4346         } else {
4347                 bnx2x_fp(bp, index, sb_index_values) =
4348                         (__le16 *)status_blk.e1x_sb->sb.index_values;
4349                 bnx2x_fp(bp, index, sb_running_index) =
4350                         (__le16 *)status_blk.e1x_sb->sb.running_index;
4351         }
4352 }
4353
4354 /* Returns the number of actually allocated BDs */
4355 static int bnx2x_alloc_rx_bds(struct bnx2x_fastpath *fp,
4356                               int rx_ring_size)
4357 {
4358         struct bnx2x *bp = fp->bp;
4359         u16 ring_prod, cqe_ring_prod;
4360         int i, failure_cnt = 0;
4361
4362         fp->rx_comp_cons = 0;
4363         cqe_ring_prod = ring_prod = 0;
4364
4365         /* This routine is called only during fo init so
4366          * fp->eth_q_stats.rx_skb_alloc_failed = 0
4367          */
4368         for (i = 0; i < rx_ring_size; i++) {
4369                 if (bnx2x_alloc_rx_data(bp, fp, ring_prod, GFP_KERNEL) < 0) {
4370                         failure_cnt++;
4371                         continue;
4372                 }
4373                 ring_prod = NEXT_RX_IDX(ring_prod);
4374                 cqe_ring_prod = NEXT_RCQ_IDX(cqe_ring_prod);
4375                 WARN_ON(ring_prod <= (i - failure_cnt));
4376         }
4377
4378         if (failure_cnt)
4379                 BNX2X_ERR("was only able to allocate %d rx skbs on queue[%d]\n",
4380                           i - failure_cnt, fp->index);
4381
4382         fp->rx_bd_prod = ring_prod;
4383         /* Limit the CQE producer by the CQE ring size */
4384         fp->rx_comp_prod = min_t(u16, NUM_RCQ_RINGS*RCQ_DESC_CNT,
4385                                cqe_ring_prod);
4386         fp->rx_pkt = fp->rx_calls = 0;
4387
4388         bnx2x_fp_stats(bp, fp)->eth_q_stats.rx_skb_alloc_failed += failure_cnt;
4389
4390         return i - failure_cnt;
4391 }
4392
4393 static void bnx2x_set_next_page_rx_cq(struct bnx2x_fastpath *fp)
4394 {
4395         int i;
4396
4397         for (i = 1; i <= NUM_RCQ_RINGS; i++) {
4398                 struct eth_rx_cqe_next_page *nextpg;
4399
4400                 nextpg = (struct eth_rx_cqe_next_page *)
4401                         &fp->rx_comp_ring[RCQ_DESC_CNT * i - 1];
4402                 nextpg->addr_hi =
4403                         cpu_to_le32(U64_HI(fp->rx_comp_mapping +
4404                                    BCM_PAGE_SIZE*(i % NUM_RCQ_RINGS)));
4405                 nextpg->addr_lo =
4406                         cpu_to_le32(U64_LO(fp->rx_comp_mapping +
4407                                    BCM_PAGE_SIZE*(i % NUM_RCQ_RINGS)));
4408         }
4409 }
4410
4411 static int bnx2x_alloc_fp_mem_at(struct bnx2x *bp, int index)
4412 {
4413         union host_hc_status_block *sb;
4414         struct bnx2x_fastpath *fp = &bp->fp[index];
4415         int ring_size = 0;
4416         u8 cos;
4417         int rx_ring_size = 0;
4418
4419         if (!bp->rx_ring_size && IS_MF_STORAGE_ONLY(bp)) {
4420                 rx_ring_size = MIN_RX_SIZE_NONTPA;
4421                 bp->rx_ring_size = rx_ring_size;
4422         } else if (!bp->rx_ring_size) {
4423                 rx_ring_size = MAX_RX_AVAIL/BNX2X_NUM_RX_QUEUES(bp);
4424
4425                 if (CHIP_IS_E3(bp)) {
4426                         u32 cfg = SHMEM_RD(bp,
4427                                            dev_info.port_hw_config[BP_PORT(bp)].
4428                                            default_cfg);
4429
4430                         /* Decrease ring size for 1G functions */
4431                         if ((cfg & PORT_HW_CFG_NET_SERDES_IF_MASK) ==
4432                             PORT_HW_CFG_NET_SERDES_IF_SGMII)
4433                                 rx_ring_size /= 10;
4434                 }
4435
4436                 /* allocate at least number of buffers required by FW */
4437                 rx_ring_size = max_t(int, bp->disable_tpa ? MIN_RX_SIZE_NONTPA :
4438                                      MIN_RX_SIZE_TPA, rx_ring_size);
4439
4440                 bp->rx_ring_size = rx_ring_size;
4441         } else /* if rx_ring_size specified - use it */
4442                 rx_ring_size = bp->rx_ring_size;
4443
4444         DP(BNX2X_MSG_SP, "calculated rx_ring_size %d\n", rx_ring_size);
4445
4446         /* Common */
4447         sb = &bnx2x_fp(bp, index, status_blk);
4448
4449         if (!IS_FCOE_IDX(index)) {
4450                 /* status blocks */
4451                 if (!CHIP_IS_E1x(bp)) {
4452                         sb->e2_sb = BNX2X_PCI_ALLOC(&bnx2x_fp(bp, index, status_blk_mapping),
4453                                                     sizeof(struct host_hc_status_block_e2));
4454                         if (!sb->e2_sb)
4455                                 goto alloc_mem_err;
4456                 } else {
4457                         sb->e1x_sb = BNX2X_PCI_ALLOC(&bnx2x_fp(bp, index, status_blk_mapping),
4458                                                      sizeof(struct host_hc_status_block_e1x));
4459                         if (!sb->e1x_sb)
4460                                 goto alloc_mem_err;
4461                 }
4462         }
4463
4464         /* FCoE Queue uses Default SB and doesn't ACK the SB, thus no need to
4465          * set shortcuts for it.
4466          */
4467         if (!IS_FCOE_IDX(index))
4468                 set_sb_shortcuts(bp, index);
4469
4470         /* Tx */
4471         if (!skip_tx_queue(bp, index)) {
4472                 /* fastpath tx rings: tx_buf tx_desc */
4473                 for_each_cos_in_tx_queue(fp, cos) {
4474                         struct bnx2x_fp_txdata *txdata = fp->txdata_ptr[cos];
4475
4476                         DP(NETIF_MSG_IFUP,
4477                            "allocating tx memory of fp %d cos %d\n",
4478                            index, cos);
4479
4480                         txdata->tx_buf_ring = kcalloc(NUM_TX_BD,
4481                                                       sizeof(struct sw_tx_bd),
4482                                                       GFP_KERNEL);
4483                         if (!txdata->tx_buf_ring)
4484                                 goto alloc_mem_err;
4485                         txdata->tx_desc_ring = BNX2X_PCI_ALLOC(&txdata->tx_desc_mapping,
4486                                                                sizeof(union eth_tx_bd_types) * NUM_TX_BD);
4487                         if (!txdata->tx_desc_ring)
4488                                 goto alloc_mem_err;
4489                 }
4490         }
4491
4492         /* Rx */
4493         if (!skip_rx_queue(bp, index)) {
4494                 /* fastpath rx rings: rx_buf rx_desc rx_comp */
4495                 bnx2x_fp(bp, index, rx_buf_ring) =
4496                         kcalloc(NUM_RX_BD, sizeof(struct sw_rx_bd), GFP_KERNEL);
4497                 if (!bnx2x_fp(bp, index, rx_buf_ring))
4498                         goto alloc_mem_err;
4499                 bnx2x_fp(bp, index, rx_desc_ring) =
4500                         BNX2X_PCI_ALLOC(&bnx2x_fp(bp, index, rx_desc_mapping),
4501                                         sizeof(struct eth_rx_bd) * NUM_RX_BD);
4502                 if (!bnx2x_fp(bp, index, rx_desc_ring))
4503                         goto alloc_mem_err;
4504
4505                 /* Seed all CQEs by 1s */
4506                 bnx2x_fp(bp, index, rx_comp_ring) =
4507                         BNX2X_PCI_FALLOC(&bnx2x_fp(bp, index, rx_comp_mapping),
4508                                          sizeof(struct eth_fast_path_rx_cqe) * NUM_RCQ_BD);
4509                 if (!bnx2x_fp(bp, index, rx_comp_ring))
4510                         goto alloc_mem_err;
4511
4512                 /* SGE ring */
4513                 bnx2x_fp(bp, index, rx_page_ring) =
4514                         kcalloc(NUM_RX_SGE, sizeof(struct sw_rx_page),
4515                                 GFP_KERNEL);
4516                 if (!bnx2x_fp(bp, index, rx_page_ring))
4517                         goto alloc_mem_err;
4518                 bnx2x_fp(bp, index, rx_sge_ring) =
4519                         BNX2X_PCI_ALLOC(&bnx2x_fp(bp, index, rx_sge_mapping),
4520                                         BCM_PAGE_SIZE * NUM_RX_SGE_PAGES);
4521                 if (!bnx2x_fp(bp, index, rx_sge_ring))
4522                         goto alloc_mem_err;
4523                 /* RX BD ring */
4524                 bnx2x_set_next_page_rx_bd(fp);
4525
4526                 /* CQ ring */
4527                 bnx2x_set_next_page_rx_cq(fp);
4528
4529                 /* BDs */
4530                 ring_size = bnx2x_alloc_rx_bds(fp, rx_ring_size);
4531                 if (ring_size < rx_ring_size)
4532                         goto alloc_mem_err;
4533         }
4534
4535         return 0;
4536
4537 /* handles low memory cases */
4538 alloc_mem_err:
4539         BNX2X_ERR("Unable to allocate full memory for queue %d (size %d)\n",
4540                                                 index, ring_size);
4541         /* FW will drop all packets if queue is not big enough,
4542          * In these cases we disable the queue
4543          * Min size is different for OOO, TPA and non-TPA queues
4544          */
4545         if (ring_size < (fp->disable_tpa ?
4546                                 MIN_RX_SIZE_NONTPA : MIN_RX_SIZE_TPA)) {
4547                         /* release memory allocated for this queue */
4548                         bnx2x_free_fp_mem_at(bp, index);
4549                         return -ENOMEM;
4550         }
4551         return 0;
4552 }
4553
4554 static int bnx2x_alloc_fp_mem_cnic(struct bnx2x *bp)
4555 {
4556         if (!NO_FCOE(bp))
4557                 /* FCoE */
4558                 if (bnx2x_alloc_fp_mem_at(bp, FCOE_IDX(bp)))
4559                         /* we will fail load process instead of mark
4560                          * NO_FCOE_FLAG
4561                          */
4562                         return -ENOMEM;
4563
4564         return 0;
4565 }
4566
4567 static int bnx2x_alloc_fp_mem(struct bnx2x *bp)
4568 {
4569         int i;
4570
4571         /* 1. Allocate FP for leading - fatal if error
4572          * 2. Allocate RSS - fix number of queues if error
4573          */
4574
4575         /* leading */
4576         if (bnx2x_alloc_fp_mem_at(bp, 0))
4577                 return -ENOMEM;
4578
4579         /* RSS */
4580         for_each_nondefault_eth_queue(bp, i)
4581                 if (bnx2x_alloc_fp_mem_at(bp, i))
4582                         break;
4583
4584         /* handle memory failures */
4585         if (i != BNX2X_NUM_ETH_QUEUES(bp)) {
4586                 int delta = BNX2X_NUM_ETH_QUEUES(bp) - i;
4587
4588                 WARN_ON(delta < 0);
4589                 bnx2x_shrink_eth_fp(bp, delta);
4590                 if (CNIC_SUPPORT(bp))
4591                         /* move non eth FPs next to last eth FP
4592                          * must be done in that order
4593                          * FCOE_IDX < FWD_IDX < OOO_IDX
4594                          */
4595
4596                         /* move FCoE fp even NO_FCOE_FLAG is on */
4597                         bnx2x_move_fp(bp, FCOE_IDX(bp), FCOE_IDX(bp) - delta);
4598                 bp->num_ethernet_queues -= delta;
4599                 bp->num_queues = bp->num_ethernet_queues +
4600                                  bp->num_cnic_queues;
4601                 BNX2X_ERR("Adjusted num of queues from %d to %d\n",
4602                           bp->num_queues + delta, bp->num_queues);
4603         }
4604
4605         return 0;
4606 }
4607
4608 void bnx2x_free_mem_bp(struct bnx2x *bp)
4609 {
4610         int i;
4611
4612         for (i = 0; i < bp->fp_array_size; i++)
4613                 kfree(bp->fp[i].tpa_info);
4614         kfree(bp->fp);
4615         kfree(bp->sp_objs);
4616         kfree(bp->fp_stats);
4617         kfree(bp->bnx2x_txq);
4618         kfree(bp->msix_table);
4619         kfree(bp->ilt);
4620 }
4621
4622 int bnx2x_alloc_mem_bp(struct bnx2x *bp)
4623 {
4624         struct bnx2x_fastpath *fp;
4625         struct msix_entry *tbl;
4626         struct bnx2x_ilt *ilt;
4627         int msix_table_size = 0;
4628         int fp_array_size, txq_array_size;
4629         int i;
4630
4631         /*
4632          * The biggest MSI-X table we might need is as a maximum number of fast
4633          * path IGU SBs plus default SB (for PF only).
4634          */
4635         msix_table_size = bp->igu_sb_cnt;
4636         if (IS_PF(bp))
4637                 msix_table_size++;
4638         BNX2X_DEV_INFO("msix_table_size %d\n", msix_table_size);
4639
4640         /* fp array: RSS plus CNIC related L2 queues */
4641         fp_array_size = BNX2X_MAX_RSS_COUNT(bp) + CNIC_SUPPORT(bp);
4642         bp->fp_array_size = fp_array_size;
4643         BNX2X_DEV_INFO("fp_array_size %d\n", bp->fp_array_size);
4644
4645         fp = kcalloc(bp->fp_array_size, sizeof(*fp), GFP_KERNEL);
4646         if (!fp)
4647                 goto alloc_err;
4648         for (i = 0; i < bp->fp_array_size; i++) {
4649                 fp[i].tpa_info =
4650                         kcalloc(ETH_MAX_AGGREGATION_QUEUES_E1H_E2,
4651                                 sizeof(struct bnx2x_agg_info), GFP_KERNEL);
4652                 if (!(fp[i].tpa_info))
4653                         goto alloc_err;
4654         }
4655
4656         bp->fp = fp;
4657
4658         /* allocate sp objs */
4659         bp->sp_objs = kcalloc(bp->fp_array_size, sizeof(struct bnx2x_sp_objs),
4660                               GFP_KERNEL);
4661         if (!bp->sp_objs)
4662                 goto alloc_err;
4663
4664         /* allocate fp_stats */
4665         bp->fp_stats = kcalloc(bp->fp_array_size, sizeof(struct bnx2x_fp_stats),
4666                                GFP_KERNEL);
4667         if (!bp->fp_stats)
4668                 goto alloc_err;
4669
4670         /* Allocate memory for the transmission queues array */
4671         txq_array_size =
4672                 BNX2X_MAX_RSS_COUNT(bp) * BNX2X_MULTI_TX_COS + CNIC_SUPPORT(bp);
4673         BNX2X_DEV_INFO("txq_array_size %d", txq_array_size);
4674
4675         bp->bnx2x_txq = kcalloc(txq_array_size, sizeof(struct bnx2x_fp_txdata),
4676                                 GFP_KERNEL);
4677         if (!bp->bnx2x_txq)
4678                 goto alloc_err;
4679
4680         /* msix table */
4681         tbl = kcalloc(msix_table_size, sizeof(*tbl), GFP_KERNEL);
4682         if (!tbl)
4683                 goto alloc_err;
4684         bp->msix_table = tbl;
4685
4686         /* ilt */
4687         ilt = kzalloc(sizeof(*ilt), GFP_KERNEL);
4688         if (!ilt)
4689                 goto alloc_err;
4690         bp->ilt = ilt;
4691
4692         return 0;
4693 alloc_err:
4694         bnx2x_free_mem_bp(bp);
4695         return -ENOMEM;
4696 }
4697
4698 int bnx2x_reload_if_running(struct net_device *dev)
4699 {
4700         struct bnx2x *bp = netdev_priv(dev);
4701
4702         if (unlikely(!netif_running(dev)))
4703                 return 0;
4704
4705         bnx2x_nic_unload(bp, UNLOAD_NORMAL, true);
4706         return bnx2x_nic_load(bp, LOAD_NORMAL);
4707 }
4708
4709 int bnx2x_get_cur_phy_idx(struct bnx2x *bp)
4710 {
4711         u32 sel_phy_idx = 0;
4712         if (bp->link_params.num_phys <= 1)
4713                 return INT_PHY;
4714
4715         if (bp->link_vars.link_up) {
4716                 sel_phy_idx = EXT_PHY1;
4717                 /* In case link is SERDES, check if the EXT_PHY2 is the one */
4718                 if ((bp->link_vars.link_status & LINK_STATUS_SERDES_LINK) &&
4719                     (bp->link_params.phy[EXT_PHY2].supported & SUPPORTED_FIBRE))
4720                         sel_phy_idx = EXT_PHY2;
4721         } else {
4722
4723                 switch (bnx2x_phy_selection(&bp->link_params)) {
4724                 case PORT_HW_CFG_PHY_SELECTION_HARDWARE_DEFAULT:
4725                 case PORT_HW_CFG_PHY_SELECTION_FIRST_PHY:
4726                 case PORT_HW_CFG_PHY_SELECTION_FIRST_PHY_PRIORITY:
4727                        sel_phy_idx = EXT_PHY1;
4728                        break;
4729                 case PORT_HW_CFG_PHY_SELECTION_SECOND_PHY:
4730                 case PORT_HW_CFG_PHY_SELECTION_SECOND_PHY_PRIORITY:
4731                        sel_phy_idx = EXT_PHY2;
4732                        break;
4733                 }
4734         }
4735
4736         return sel_phy_idx;
4737 }
4738 int bnx2x_get_link_cfg_idx(struct bnx2x *bp)
4739 {
4740         u32 sel_phy_idx = bnx2x_get_cur_phy_idx(bp);
4741         /*
4742          * The selected activated PHY is always after swapping (in case PHY
4743          * swapping is enabled). So when swapping is enabled, we need to reverse
4744          * the configuration
4745          */
4746
4747         if (bp->link_params.multi_phy_config &
4748             PORT_HW_CFG_PHY_SWAPPED_ENABLED) {
4749                 if (sel_phy_idx == EXT_PHY1)
4750                         sel_phy_idx = EXT_PHY2;
4751                 else if (sel_phy_idx == EXT_PHY2)
4752                         sel_phy_idx = EXT_PHY1;
4753         }
4754         return LINK_CONFIG_IDX(sel_phy_idx);
4755 }
4756
4757 #ifdef NETDEV_FCOE_WWNN
4758 int bnx2x_fcoe_get_wwn(struct net_device *dev, u64 *wwn, int type)
4759 {
4760         struct bnx2x *bp = netdev_priv(dev);
4761         struct cnic_eth_dev *cp = &bp->cnic_eth_dev;
4762
4763         switch (type) {
4764         case NETDEV_FCOE_WWNN:
4765                 *wwn = HILO_U64(cp->fcoe_wwn_node_name_hi,
4766                                 cp->fcoe_wwn_node_name_lo);
4767                 break;
4768         case NETDEV_FCOE_WWPN:
4769                 *wwn = HILO_U64(cp->fcoe_wwn_port_name_hi,
4770                                 cp->fcoe_wwn_port_name_lo);
4771                 break;
4772         default:
4773                 BNX2X_ERR("Wrong WWN type requested - %d\n", type);
4774                 return -EINVAL;
4775         }
4776
4777         return 0;
4778 }
4779 #endif
4780
4781 /* called with rtnl_lock */
4782 int bnx2x_change_mtu(struct net_device *dev, int new_mtu)
4783 {
4784         struct bnx2x *bp = netdev_priv(dev);
4785
4786         if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
4787                 BNX2X_ERR("Can't perform change MTU during parity recovery\n");
4788                 return -EAGAIN;
4789         }
4790
4791         if ((new_mtu > ETH_MAX_JUMBO_PACKET_SIZE) ||
4792             ((new_mtu + ETH_HLEN) < ETH_MIN_PACKET_SIZE)) {
4793                 BNX2X_ERR("Can't support requested MTU size\n");
4794                 return -EINVAL;
4795         }
4796
4797         /* This does not race with packet allocation
4798          * because the actual alloc size is
4799          * only updated as part of load
4800          */
4801         dev->mtu = new_mtu;
4802
4803         return bnx2x_reload_if_running(dev);
4804 }
4805
4806 netdev_features_t bnx2x_fix_features(struct net_device *dev,
4807                                      netdev_features_t features)
4808 {
4809         struct bnx2x *bp = netdev_priv(dev);
4810
4811         /* TPA requires Rx CSUM offloading */
4812         if (!(features & NETIF_F_RXCSUM)) {
4813                 features &= ~NETIF_F_LRO;
4814                 features &= ~NETIF_F_GRO;
4815         }
4816
4817         /* Note: do not disable SW GRO in kernel when HW GRO is off */
4818         if (bp->disable_tpa)
4819                 features &= ~NETIF_F_LRO;
4820
4821         return features;
4822 }
4823
4824 int bnx2x_set_features(struct net_device *dev, netdev_features_t features)
4825 {
4826         struct bnx2x *bp = netdev_priv(dev);
4827         u32 flags = bp->flags;
4828         u32 changes;
4829         bool bnx2x_reload = false;
4830
4831         if (features & NETIF_F_LRO)
4832                 flags |= TPA_ENABLE_FLAG;
4833         else
4834                 flags &= ~TPA_ENABLE_FLAG;
4835
4836         if (features & NETIF_F_GRO)
4837                 flags |= GRO_ENABLE_FLAG;
4838         else
4839                 flags &= ~GRO_ENABLE_FLAG;
4840
4841         if (features & NETIF_F_LOOPBACK) {
4842                 if (bp->link_params.loopback_mode != LOOPBACK_BMAC) {
4843                         bp->link_params.loopback_mode = LOOPBACK_BMAC;
4844                         bnx2x_reload = true;
4845                 }
4846         } else {
4847                 if (bp->link_params.loopback_mode != LOOPBACK_NONE) {
4848                         bp->link_params.loopback_mode = LOOPBACK_NONE;
4849                         bnx2x_reload = true;
4850                 }
4851         }
4852
4853         changes = flags ^ bp->flags;
4854
4855         /* if GRO is changed while LRO is enabled, don't force a reload */
4856         if ((changes & GRO_ENABLE_FLAG) && (flags & TPA_ENABLE_FLAG))
4857                 changes &= ~GRO_ENABLE_FLAG;
4858
4859         /* if GRO is changed while HW TPA is off, don't force a reload */
4860         if ((changes & GRO_ENABLE_FLAG) && bp->disable_tpa)
4861                 changes &= ~GRO_ENABLE_FLAG;
4862
4863         if (changes)
4864                 bnx2x_reload = true;
4865
4866         bp->flags = flags;
4867
4868         if (bnx2x_reload) {
4869                 if (bp->recovery_state == BNX2X_RECOVERY_DONE)
4870                         return bnx2x_reload_if_running(dev);
4871                 /* else: bnx2x_nic_load() will be called at end of recovery */
4872         }
4873
4874         return 0;
4875 }
4876
4877 void bnx2x_tx_timeout(struct net_device *dev)
4878 {
4879         struct bnx2x *bp = netdev_priv(dev);
4880
4881 #ifdef BNX2X_STOP_ON_ERROR
4882         if (!bp->panic)
4883                 bnx2x_panic();
4884 #endif
4885
4886         /* This allows the netif to be shutdown gracefully before resetting */
4887         bnx2x_schedule_sp_rtnl(bp, BNX2X_SP_RTNL_TX_TIMEOUT, 0);
4888 }
4889
4890 int bnx2x_suspend(struct pci_dev *pdev, pm_message_t state)
4891 {
4892         struct net_device *dev = pci_get_drvdata(pdev);
4893         struct bnx2x *bp;
4894
4895         if (!dev) {
4896                 dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
4897                 return -ENODEV;
4898         }
4899         bp = netdev_priv(dev);
4900
4901         rtnl_lock();
4902
4903         pci_save_state(pdev);
4904
4905         if (!netif_running(dev)) {
4906                 rtnl_unlock();
4907                 return 0;
4908         }
4909
4910         netif_device_detach(dev);
4911
4912         bnx2x_nic_unload(bp, UNLOAD_CLOSE, false);
4913
4914         bnx2x_set_power_state(bp, pci_choose_state(pdev, state));
4915
4916         rtnl_unlock();
4917
4918         return 0;
4919 }
4920
4921 int bnx2x_resume(struct pci_dev *pdev)
4922 {
4923         struct net_device *dev = pci_get_drvdata(pdev);
4924         struct bnx2x *bp;
4925         int rc;
4926
4927         if (!dev) {
4928                 dev_err(&pdev->dev, "BAD net device from bnx2x_init_one\n");
4929                 return -ENODEV;
4930         }
4931         bp = netdev_priv(dev);
4932
4933         if (bp->recovery_state != BNX2X_RECOVERY_DONE) {
4934                 BNX2X_ERR("Handling parity error recovery. Try again later\n");
4935                 return -EAGAIN;
4936         }
4937
4938         rtnl_lock();
4939
4940         pci_restore_state(pdev);
4941
4942         if (!netif_running(dev)) {
4943                 rtnl_unlock();
4944                 return 0;
4945         }
4946
4947         bnx2x_set_power_state(bp, PCI_D0);
4948         netif_device_attach(dev);
4949
4950         rc = bnx2x_nic_load(bp, LOAD_OPEN);
4951
4952         rtnl_unlock();
4953
4954         return rc;
4955 }
4956
4957 void bnx2x_set_ctx_validation(struct bnx2x *bp, struct eth_context *cxt,
4958                               u32 cid)
4959 {
4960         if (!cxt) {
4961                 BNX2X_ERR("bad context pointer %p\n", cxt);
4962                 return;
4963         }
4964
4965         /* ustorm cxt validation */
4966         cxt->ustorm_ag_context.cdu_usage =
4967                 CDU_RSRVD_VALUE_TYPE_A(HW_CID(bp, cid),
4968                         CDU_REGION_NUMBER_UCM_AG, ETH_CONNECTION_TYPE);
4969         /* xcontext validation */
4970         cxt->xstorm_ag_context.cdu_reserved =
4971                 CDU_RSRVD_VALUE_TYPE_A(HW_CID(bp, cid),
4972                         CDU_REGION_NUMBER_XCM_AG, ETH_CONNECTION_TYPE);
4973 }
4974
4975 static void storm_memset_hc_timeout(struct bnx2x *bp, u8 port,
4976                                     u8 fw_sb_id, u8 sb_index,
4977                                     u8 ticks)
4978 {
4979         u32 addr = BAR_CSTRORM_INTMEM +
4980                    CSTORM_STATUS_BLOCK_DATA_TIMEOUT_OFFSET(fw_sb_id, sb_index);
4981         REG_WR8(bp, addr, ticks);
4982         DP(NETIF_MSG_IFUP,
4983            "port %x fw_sb_id %d sb_index %d ticks %d\n",
4984            port, fw_sb_id, sb_index, ticks);
4985 }
4986
4987 static void storm_memset_hc_disable(struct bnx2x *bp, u8 port,
4988                                     u16 fw_sb_id, u8 sb_index,
4989                                     u8 disable)
4990 {
4991         u32 enable_flag = disable ? 0 : (1 << HC_INDEX_DATA_HC_ENABLED_SHIFT);
4992         u32 addr = BAR_CSTRORM_INTMEM +
4993                    CSTORM_STATUS_BLOCK_DATA_FLAGS_OFFSET(fw_sb_id, sb_index);
4994         u8 flags = REG_RD8(bp, addr);
4995         /* clear and set */
4996         flags &= ~HC_INDEX_DATA_HC_ENABLED;
4997         flags |= enable_flag;
4998         REG_WR8(bp, addr, flags);
4999         DP(NETIF_MSG_IFUP,
5000            "port %x fw_sb_id %d sb_index %d disable %d\n",
5001            port, fw_sb_id, sb_index, disable);
5002 }
5003
5004 void bnx2x_update_coalesce_sb_index(struct bnx2x *bp, u8 fw_sb_id,
5005                                     u8 sb_index, u8 disable, u16 usec)
5006 {
5007         int port = BP_PORT(bp);
5008         u8 ticks = usec / BNX2X_BTR;
5009
5010         storm_memset_hc_timeout(bp, port, fw_sb_id, sb_index, ticks);
5011
5012         disable = disable ? 1 : (usec ? 0 : 1);
5013         storm_memset_hc_disable(bp, port, fw_sb_id, sb_index, disable);
5014 }
5015
5016 void bnx2x_schedule_sp_rtnl(struct bnx2x *bp, enum sp_rtnl_flag flag,
5017                             u32 verbose)
5018 {
5019         smp_mb__before_atomic();
5020         set_bit(flag, &bp->sp_rtnl_state);
5021         smp_mb__after_atomic();
5022         DP((BNX2X_MSG_SP | verbose), "Scheduling sp_rtnl task [Flag: %d]\n",
5023            flag);
5024         schedule_delayed_work(&bp->sp_rtnl_task, 0);
5025 }
5026 EXPORT_SYMBOL(bnx2x_schedule_sp_rtnl);