vmxnet3: Add locking for access to command register
[linux-2.6.git] / drivers / net / vmxnet3 / vmxnet3_drv.c
1 /*
2  * Linux driver for VMware's vmxnet3 ethernet NIC.
3  *
4  * Copyright (C) 2008-2009, VMware, Inc. All Rights Reserved.
5  *
6  * This program is free software; you can redistribute it and/or modify it
7  * under the terms of the GNU General Public License as published by the
8  * Free Software Foundation; version 2 of the License and no later version.
9  *
10  * This program is distributed in the hope that it will be useful, but
11  * WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
13  * NON INFRINGEMENT. See the GNU General Public License for more
14  * details.
15  *
16  * You should have received a copy of the GNU General Public License
17  * along with this program; if not, write to the Free Software
18  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19  *
20  * The full GNU General Public License is included in this distribution in
21  * the file called "COPYING".
22  *
23  * Maintained by: Shreyas Bhatewara <pv-drivers@vmware.com>
24  *
25  */
26
27 #include <net/ip6_checksum.h>
28
29 #include "vmxnet3_int.h"
30
31 char vmxnet3_driver_name[] = "vmxnet3";
32 #define VMXNET3_DRIVER_DESC "VMware vmxnet3 virtual NIC driver"
33
34 /*
35  * PCI Device ID Table
36  * Last entry must be all 0s
37  */
38 static DEFINE_PCI_DEVICE_TABLE(vmxnet3_pciid_table) = {
39         {PCI_VDEVICE(VMWARE, PCI_DEVICE_ID_VMWARE_VMXNET3)},
40         {0}
41 };
42
43 MODULE_DEVICE_TABLE(pci, vmxnet3_pciid_table);
44
45 static atomic_t devices_found;
46
47 #define VMXNET3_MAX_DEVICES 10
48 static int enable_mq = 1;
49 static int irq_share_mode;
50
51 static void
52 vmxnet3_write_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac);
53
54 /*
55  *    Enable/Disable the given intr
56  */
57 static void
58 vmxnet3_enable_intr(struct vmxnet3_adapter *adapter, unsigned intr_idx)
59 {
60         VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_IMR + intr_idx * 8, 0);
61 }
62
63
64 static void
65 vmxnet3_disable_intr(struct vmxnet3_adapter *adapter, unsigned intr_idx)
66 {
67         VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_IMR + intr_idx * 8, 1);
68 }
69
70
71 /*
72  *    Enable/Disable all intrs used by the device
73  */
74 static void
75 vmxnet3_enable_all_intrs(struct vmxnet3_adapter *adapter)
76 {
77         int i;
78
79         for (i = 0; i < adapter->intr.num_intrs; i++)
80                 vmxnet3_enable_intr(adapter, i);
81         adapter->shared->devRead.intrConf.intrCtrl &=
82                                         cpu_to_le32(~VMXNET3_IC_DISABLE_ALL);
83 }
84
85
86 static void
87 vmxnet3_disable_all_intrs(struct vmxnet3_adapter *adapter)
88 {
89         int i;
90
91         adapter->shared->devRead.intrConf.intrCtrl |=
92                                         cpu_to_le32(VMXNET3_IC_DISABLE_ALL);
93         for (i = 0; i < adapter->intr.num_intrs; i++)
94                 vmxnet3_disable_intr(adapter, i);
95 }
96
97
98 static void
99 vmxnet3_ack_events(struct vmxnet3_adapter *adapter, u32 events)
100 {
101         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_ECR, events);
102 }
103
104
105 static bool
106 vmxnet3_tq_stopped(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
107 {
108         return tq->stopped;
109 }
110
111
112 static void
113 vmxnet3_tq_start(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
114 {
115         tq->stopped = false;
116         netif_start_subqueue(adapter->netdev, tq - adapter->tx_queue);
117 }
118
119
120 static void
121 vmxnet3_tq_wake(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
122 {
123         tq->stopped = false;
124         netif_wake_subqueue(adapter->netdev, (tq - adapter->tx_queue));
125 }
126
127
128 static void
129 vmxnet3_tq_stop(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
130 {
131         tq->stopped = true;
132         tq->num_stop++;
133         netif_stop_subqueue(adapter->netdev, (tq - adapter->tx_queue));
134 }
135
136
137 /*
138  * Check the link state. This may start or stop the tx queue.
139  */
140 static void
141 vmxnet3_check_link(struct vmxnet3_adapter *adapter, bool affectTxQueue)
142 {
143         u32 ret;
144         int i;
145         unsigned long flags;
146
147         spin_lock_irqsave(&adapter->cmd_lock, flags);
148         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, VMXNET3_CMD_GET_LINK);
149         ret = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
150         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
151
152         adapter->link_speed = ret >> 16;
153         if (ret & 1) { /* Link is up. */
154                 printk(KERN_INFO "%s: NIC Link is Up %d Mbps\n",
155                        adapter->netdev->name, adapter->link_speed);
156                 if (!netif_carrier_ok(adapter->netdev))
157                         netif_carrier_on(adapter->netdev);
158
159                 if (affectTxQueue) {
160                         for (i = 0; i < adapter->num_tx_queues; i++)
161                                 vmxnet3_tq_start(&adapter->tx_queue[i],
162                                                  adapter);
163                 }
164         } else {
165                 printk(KERN_INFO "%s: NIC Link is Down\n",
166                        adapter->netdev->name);
167                 if (netif_carrier_ok(adapter->netdev))
168                         netif_carrier_off(adapter->netdev);
169
170                 if (affectTxQueue) {
171                         for (i = 0; i < adapter->num_tx_queues; i++)
172                                 vmxnet3_tq_stop(&adapter->tx_queue[i], adapter);
173                 }
174         }
175 }
176
177 static void
178 vmxnet3_process_events(struct vmxnet3_adapter *adapter)
179 {
180         int i;
181         u32 events = le32_to_cpu(adapter->shared->ecr);
182         if (!events)
183                 return;
184
185         vmxnet3_ack_events(adapter, events);
186
187         /* Check if link state has changed */
188         if (events & VMXNET3_ECR_LINK)
189                 vmxnet3_check_link(adapter, true);
190
191         /* Check if there is an error on xmit/recv queues */
192         if (events & (VMXNET3_ECR_TQERR | VMXNET3_ECR_RQERR)) {
193                 spin_lock(&adapter->cmd_lock);
194                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
195                                        VMXNET3_CMD_GET_QUEUE_STATUS);
196                 spin_unlock(&adapter->cmd_lock);
197
198                 for (i = 0; i < adapter->num_tx_queues; i++)
199                         if (adapter->tqd_start[i].status.stopped)
200                                 dev_err(&adapter->netdev->dev,
201                                         "%s: tq[%d] error 0x%x\n",
202                                         adapter->netdev->name, i, le32_to_cpu(
203                                         adapter->tqd_start[i].status.error));
204                 for (i = 0; i < adapter->num_rx_queues; i++)
205                         if (adapter->rqd_start[i].status.stopped)
206                                 dev_err(&adapter->netdev->dev,
207                                         "%s: rq[%d] error 0x%x\n",
208                                         adapter->netdev->name, i,
209                                         adapter->rqd_start[i].status.error);
210
211                 schedule_work(&adapter->work);
212         }
213 }
214
215 #ifdef __BIG_ENDIAN_BITFIELD
216 /*
217  * The device expects the bitfields in shared structures to be written in
218  * little endian. When CPU is big endian, the following routines are used to
219  * correctly read and write into ABI.
220  * The general technique used here is : double word bitfields are defined in
221  * opposite order for big endian architecture. Then before reading them in
222  * driver the complete double word is translated using le32_to_cpu. Similarly
223  * After the driver writes into bitfields, cpu_to_le32 is used to translate the
224  * double words into required format.
225  * In order to avoid touching bits in shared structure more than once, temporary
226  * descriptors are used. These are passed as srcDesc to following functions.
227  */
228 static void vmxnet3_RxDescToCPU(const struct Vmxnet3_RxDesc *srcDesc,
229                                 struct Vmxnet3_RxDesc *dstDesc)
230 {
231         u32 *src = (u32 *)srcDesc + 2;
232         u32 *dst = (u32 *)dstDesc + 2;
233         dstDesc->addr = le64_to_cpu(srcDesc->addr);
234         *dst = le32_to_cpu(*src);
235         dstDesc->ext1 = le32_to_cpu(srcDesc->ext1);
236 }
237
238 static void vmxnet3_TxDescToLe(const struct Vmxnet3_TxDesc *srcDesc,
239                                struct Vmxnet3_TxDesc *dstDesc)
240 {
241         int i;
242         u32 *src = (u32 *)(srcDesc + 1);
243         u32 *dst = (u32 *)(dstDesc + 1);
244
245         /* Working backwards so that the gen bit is set at the end. */
246         for (i = 2; i > 0; i--) {
247                 src--;
248                 dst--;
249                 *dst = cpu_to_le32(*src);
250         }
251 }
252
253
254 static void vmxnet3_RxCompToCPU(const struct Vmxnet3_RxCompDesc *srcDesc,
255                                 struct Vmxnet3_RxCompDesc *dstDesc)
256 {
257         int i = 0;
258         u32 *src = (u32 *)srcDesc;
259         u32 *dst = (u32 *)dstDesc;
260         for (i = 0; i < sizeof(struct Vmxnet3_RxCompDesc) / sizeof(u32); i++) {
261                 *dst = le32_to_cpu(*src);
262                 src++;
263                 dst++;
264         }
265 }
266
267
268 /* Used to read bitfield values from double words. */
269 static u32 get_bitfield32(const __le32 *bitfield, u32 pos, u32 size)
270 {
271         u32 temp = le32_to_cpu(*bitfield);
272         u32 mask = ((1 << size) - 1) << pos;
273         temp &= mask;
274         temp >>= pos;
275         return temp;
276 }
277
278
279
280 #endif  /* __BIG_ENDIAN_BITFIELD */
281
282 #ifdef __BIG_ENDIAN_BITFIELD
283
284 #   define VMXNET3_TXDESC_GET_GEN(txdesc) get_bitfield32(((const __le32 *) \
285                         txdesc) + VMXNET3_TXD_GEN_DWORD_SHIFT, \
286                         VMXNET3_TXD_GEN_SHIFT, VMXNET3_TXD_GEN_SIZE)
287 #   define VMXNET3_TXDESC_GET_EOP(txdesc) get_bitfield32(((const __le32 *) \
288                         txdesc) + VMXNET3_TXD_EOP_DWORD_SHIFT, \
289                         VMXNET3_TXD_EOP_SHIFT, VMXNET3_TXD_EOP_SIZE)
290 #   define VMXNET3_TCD_GET_GEN(tcd) get_bitfield32(((const __le32 *)tcd) + \
291                         VMXNET3_TCD_GEN_DWORD_SHIFT, VMXNET3_TCD_GEN_SHIFT, \
292                         VMXNET3_TCD_GEN_SIZE)
293 #   define VMXNET3_TCD_GET_TXIDX(tcd) get_bitfield32((const __le32 *)tcd, \
294                         VMXNET3_TCD_TXIDX_SHIFT, VMXNET3_TCD_TXIDX_SIZE)
295 #   define vmxnet3_getRxComp(dstrcd, rcd, tmp) do { \
296                         (dstrcd) = (tmp); \
297                         vmxnet3_RxCompToCPU((rcd), (tmp)); \
298                 } while (0)
299 #   define vmxnet3_getRxDesc(dstrxd, rxd, tmp) do { \
300                         (dstrxd) = (tmp); \
301                         vmxnet3_RxDescToCPU((rxd), (tmp)); \
302                 } while (0)
303
304 #else
305
306 #   define VMXNET3_TXDESC_GET_GEN(txdesc) ((txdesc)->gen)
307 #   define VMXNET3_TXDESC_GET_EOP(txdesc) ((txdesc)->eop)
308 #   define VMXNET3_TCD_GET_GEN(tcd) ((tcd)->gen)
309 #   define VMXNET3_TCD_GET_TXIDX(tcd) ((tcd)->txdIdx)
310 #   define vmxnet3_getRxComp(dstrcd, rcd, tmp) (dstrcd) = (rcd)
311 #   define vmxnet3_getRxDesc(dstrxd, rxd, tmp) (dstrxd) = (rxd)
312
313 #endif /* __BIG_ENDIAN_BITFIELD  */
314
315
316 static void
317 vmxnet3_unmap_tx_buf(struct vmxnet3_tx_buf_info *tbi,
318                      struct pci_dev *pdev)
319 {
320         if (tbi->map_type == VMXNET3_MAP_SINGLE)
321                 pci_unmap_single(pdev, tbi->dma_addr, tbi->len,
322                                  PCI_DMA_TODEVICE);
323         else if (tbi->map_type == VMXNET3_MAP_PAGE)
324                 pci_unmap_page(pdev, tbi->dma_addr, tbi->len,
325                                PCI_DMA_TODEVICE);
326         else
327                 BUG_ON(tbi->map_type != VMXNET3_MAP_NONE);
328
329         tbi->map_type = VMXNET3_MAP_NONE; /* to help debugging */
330 }
331
332
333 static int
334 vmxnet3_unmap_pkt(u32 eop_idx, struct vmxnet3_tx_queue *tq,
335                   struct pci_dev *pdev, struct vmxnet3_adapter *adapter)
336 {
337         struct sk_buff *skb;
338         int entries = 0;
339
340         /* no out of order completion */
341         BUG_ON(tq->buf_info[eop_idx].sop_idx != tq->tx_ring.next2comp);
342         BUG_ON(VMXNET3_TXDESC_GET_EOP(&(tq->tx_ring.base[eop_idx].txd)) != 1);
343
344         skb = tq->buf_info[eop_idx].skb;
345         BUG_ON(skb == NULL);
346         tq->buf_info[eop_idx].skb = NULL;
347
348         VMXNET3_INC_RING_IDX_ONLY(eop_idx, tq->tx_ring.size);
349
350         while (tq->tx_ring.next2comp != eop_idx) {
351                 vmxnet3_unmap_tx_buf(tq->buf_info + tq->tx_ring.next2comp,
352                                      pdev);
353
354                 /* update next2comp w/o tx_lock. Since we are marking more,
355                  * instead of less, tx ring entries avail, the worst case is
356                  * that the tx routine incorrectly re-queues a pkt due to
357                  * insufficient tx ring entries.
358                  */
359                 vmxnet3_cmd_ring_adv_next2comp(&tq->tx_ring);
360                 entries++;
361         }
362
363         dev_kfree_skb_any(skb);
364         return entries;
365 }
366
367
368 static int
369 vmxnet3_tq_tx_complete(struct vmxnet3_tx_queue *tq,
370                         struct vmxnet3_adapter *adapter)
371 {
372         int completed = 0;
373         union Vmxnet3_GenericDesc *gdesc;
374
375         gdesc = tq->comp_ring.base + tq->comp_ring.next2proc;
376         while (VMXNET3_TCD_GET_GEN(&gdesc->tcd) == tq->comp_ring.gen) {
377                 completed += vmxnet3_unmap_pkt(VMXNET3_TCD_GET_TXIDX(
378                                                &gdesc->tcd), tq, adapter->pdev,
379                                                adapter);
380
381                 vmxnet3_comp_ring_adv_next2proc(&tq->comp_ring);
382                 gdesc = tq->comp_ring.base + tq->comp_ring.next2proc;
383         }
384
385         if (completed) {
386                 spin_lock(&tq->tx_lock);
387                 if (unlikely(vmxnet3_tq_stopped(tq, adapter) &&
388                              vmxnet3_cmd_ring_desc_avail(&tq->tx_ring) >
389                              VMXNET3_WAKE_QUEUE_THRESHOLD(tq) &&
390                              netif_carrier_ok(adapter->netdev))) {
391                         vmxnet3_tq_wake(tq, adapter);
392                 }
393                 spin_unlock(&tq->tx_lock);
394         }
395         return completed;
396 }
397
398
399 static void
400 vmxnet3_tq_cleanup(struct vmxnet3_tx_queue *tq,
401                    struct vmxnet3_adapter *adapter)
402 {
403         int i;
404
405         while (tq->tx_ring.next2comp != tq->tx_ring.next2fill) {
406                 struct vmxnet3_tx_buf_info *tbi;
407                 union Vmxnet3_GenericDesc *gdesc;
408
409                 tbi = tq->buf_info + tq->tx_ring.next2comp;
410                 gdesc = tq->tx_ring.base + tq->tx_ring.next2comp;
411
412                 vmxnet3_unmap_tx_buf(tbi, adapter->pdev);
413                 if (tbi->skb) {
414                         dev_kfree_skb_any(tbi->skb);
415                         tbi->skb = NULL;
416                 }
417                 vmxnet3_cmd_ring_adv_next2comp(&tq->tx_ring);
418         }
419
420         /* sanity check, verify all buffers are indeed unmapped and freed */
421         for (i = 0; i < tq->tx_ring.size; i++) {
422                 BUG_ON(tq->buf_info[i].skb != NULL ||
423                        tq->buf_info[i].map_type != VMXNET3_MAP_NONE);
424         }
425
426         tq->tx_ring.gen = VMXNET3_INIT_GEN;
427         tq->tx_ring.next2fill = tq->tx_ring.next2comp = 0;
428
429         tq->comp_ring.gen = VMXNET3_INIT_GEN;
430         tq->comp_ring.next2proc = 0;
431 }
432
433
434 static void
435 vmxnet3_tq_destroy(struct vmxnet3_tx_queue *tq,
436                    struct vmxnet3_adapter *adapter)
437 {
438         if (tq->tx_ring.base) {
439                 pci_free_consistent(adapter->pdev, tq->tx_ring.size *
440                                     sizeof(struct Vmxnet3_TxDesc),
441                                     tq->tx_ring.base, tq->tx_ring.basePA);
442                 tq->tx_ring.base = NULL;
443         }
444         if (tq->data_ring.base) {
445                 pci_free_consistent(adapter->pdev, tq->data_ring.size *
446                                     sizeof(struct Vmxnet3_TxDataDesc),
447                                     tq->data_ring.base, tq->data_ring.basePA);
448                 tq->data_ring.base = NULL;
449         }
450         if (tq->comp_ring.base) {
451                 pci_free_consistent(adapter->pdev, tq->comp_ring.size *
452                                     sizeof(struct Vmxnet3_TxCompDesc),
453                                     tq->comp_ring.base, tq->comp_ring.basePA);
454                 tq->comp_ring.base = NULL;
455         }
456         kfree(tq->buf_info);
457         tq->buf_info = NULL;
458 }
459
460
461 /* Destroy all tx queues */
462 void
463 vmxnet3_tq_destroy_all(struct vmxnet3_adapter *adapter)
464 {
465         int i;
466
467         for (i = 0; i < adapter->num_tx_queues; i++)
468                 vmxnet3_tq_destroy(&adapter->tx_queue[i], adapter);
469 }
470
471
472 static void
473 vmxnet3_tq_init(struct vmxnet3_tx_queue *tq,
474                 struct vmxnet3_adapter *adapter)
475 {
476         int i;
477
478         /* reset the tx ring contents to 0 and reset the tx ring states */
479         memset(tq->tx_ring.base, 0, tq->tx_ring.size *
480                sizeof(struct Vmxnet3_TxDesc));
481         tq->tx_ring.next2fill = tq->tx_ring.next2comp = 0;
482         tq->tx_ring.gen = VMXNET3_INIT_GEN;
483
484         memset(tq->data_ring.base, 0, tq->data_ring.size *
485                sizeof(struct Vmxnet3_TxDataDesc));
486
487         /* reset the tx comp ring contents to 0 and reset comp ring states */
488         memset(tq->comp_ring.base, 0, tq->comp_ring.size *
489                sizeof(struct Vmxnet3_TxCompDesc));
490         tq->comp_ring.next2proc = 0;
491         tq->comp_ring.gen = VMXNET3_INIT_GEN;
492
493         /* reset the bookkeeping data */
494         memset(tq->buf_info, 0, sizeof(tq->buf_info[0]) * tq->tx_ring.size);
495         for (i = 0; i < tq->tx_ring.size; i++)
496                 tq->buf_info[i].map_type = VMXNET3_MAP_NONE;
497
498         /* stats are not reset */
499 }
500
501
502 static int
503 vmxnet3_tq_create(struct vmxnet3_tx_queue *tq,
504                   struct vmxnet3_adapter *adapter)
505 {
506         BUG_ON(tq->tx_ring.base || tq->data_ring.base ||
507                tq->comp_ring.base || tq->buf_info);
508
509         tq->tx_ring.base = pci_alloc_consistent(adapter->pdev, tq->tx_ring.size
510                            * sizeof(struct Vmxnet3_TxDesc),
511                            &tq->tx_ring.basePA);
512         if (!tq->tx_ring.base) {
513                 printk(KERN_ERR "%s: failed to allocate tx ring\n",
514                        adapter->netdev->name);
515                 goto err;
516         }
517
518         tq->data_ring.base = pci_alloc_consistent(adapter->pdev,
519                              tq->data_ring.size *
520                              sizeof(struct Vmxnet3_TxDataDesc),
521                              &tq->data_ring.basePA);
522         if (!tq->data_ring.base) {
523                 printk(KERN_ERR "%s: failed to allocate data ring\n",
524                        adapter->netdev->name);
525                 goto err;
526         }
527
528         tq->comp_ring.base = pci_alloc_consistent(adapter->pdev,
529                              tq->comp_ring.size *
530                              sizeof(struct Vmxnet3_TxCompDesc),
531                              &tq->comp_ring.basePA);
532         if (!tq->comp_ring.base) {
533                 printk(KERN_ERR "%s: failed to allocate tx comp ring\n",
534                        adapter->netdev->name);
535                 goto err;
536         }
537
538         tq->buf_info = kcalloc(tq->tx_ring.size, sizeof(tq->buf_info[0]),
539                                GFP_KERNEL);
540         if (!tq->buf_info) {
541                 printk(KERN_ERR "%s: failed to allocate tx bufinfo\n",
542                        adapter->netdev->name);
543                 goto err;
544         }
545
546         return 0;
547
548 err:
549         vmxnet3_tq_destroy(tq, adapter);
550         return -ENOMEM;
551 }
552
553 static void
554 vmxnet3_tq_cleanup_all(struct vmxnet3_adapter *adapter)
555 {
556         int i;
557
558         for (i = 0; i < adapter->num_tx_queues; i++)
559                 vmxnet3_tq_cleanup(&adapter->tx_queue[i], adapter);
560 }
561
562 /*
563  *    starting from ring->next2fill, allocate rx buffers for the given ring
564  *    of the rx queue and update the rx desc. stop after @num_to_alloc buffers
565  *    are allocated or allocation fails
566  */
567
568 static int
569 vmxnet3_rq_alloc_rx_buf(struct vmxnet3_rx_queue *rq, u32 ring_idx,
570                         int num_to_alloc, struct vmxnet3_adapter *adapter)
571 {
572         int num_allocated = 0;
573         struct vmxnet3_rx_buf_info *rbi_base = rq->buf_info[ring_idx];
574         struct vmxnet3_cmd_ring *ring = &rq->rx_ring[ring_idx];
575         u32 val;
576
577         while (num_allocated < num_to_alloc) {
578                 struct vmxnet3_rx_buf_info *rbi;
579                 union Vmxnet3_GenericDesc *gd;
580
581                 rbi = rbi_base + ring->next2fill;
582                 gd = ring->base + ring->next2fill;
583
584                 if (rbi->buf_type == VMXNET3_RX_BUF_SKB) {
585                         if (rbi->skb == NULL) {
586                                 rbi->skb = dev_alloc_skb(rbi->len +
587                                                          NET_IP_ALIGN);
588                                 if (unlikely(rbi->skb == NULL)) {
589                                         rq->stats.rx_buf_alloc_failure++;
590                                         break;
591                                 }
592                                 rbi->skb->dev = adapter->netdev;
593
594                                 skb_reserve(rbi->skb, NET_IP_ALIGN);
595                                 rbi->dma_addr = pci_map_single(adapter->pdev,
596                                                 rbi->skb->data, rbi->len,
597                                                 PCI_DMA_FROMDEVICE);
598                         } else {
599                                 /* rx buffer skipped by the device */
600                         }
601                         val = VMXNET3_RXD_BTYPE_HEAD << VMXNET3_RXD_BTYPE_SHIFT;
602                 } else {
603                         BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_PAGE ||
604                                rbi->len  != PAGE_SIZE);
605
606                         if (rbi->page == NULL) {
607                                 rbi->page = alloc_page(GFP_ATOMIC);
608                                 if (unlikely(rbi->page == NULL)) {
609                                         rq->stats.rx_buf_alloc_failure++;
610                                         break;
611                                 }
612                                 rbi->dma_addr = pci_map_page(adapter->pdev,
613                                                 rbi->page, 0, PAGE_SIZE,
614                                                 PCI_DMA_FROMDEVICE);
615                         } else {
616                                 /* rx buffers skipped by the device */
617                         }
618                         val = VMXNET3_RXD_BTYPE_BODY << VMXNET3_RXD_BTYPE_SHIFT;
619                 }
620
621                 BUG_ON(rbi->dma_addr == 0);
622                 gd->rxd.addr = cpu_to_le64(rbi->dma_addr);
623                 gd->dword[2] = cpu_to_le32((ring->gen << VMXNET3_RXD_GEN_SHIFT)
624                                            | val | rbi->len);
625
626                 num_allocated++;
627                 vmxnet3_cmd_ring_adv_next2fill(ring);
628         }
629         rq->uncommitted[ring_idx] += num_allocated;
630
631         dev_dbg(&adapter->netdev->dev,
632                 "alloc_rx_buf: %d allocated, next2fill %u, next2comp "
633                 "%u, uncommited %u\n", num_allocated, ring->next2fill,
634                 ring->next2comp, rq->uncommitted[ring_idx]);
635
636         /* so that the device can distinguish a full ring and an empty ring */
637         BUG_ON(num_allocated != 0 && ring->next2fill == ring->next2comp);
638
639         return num_allocated;
640 }
641
642
643 static void
644 vmxnet3_append_frag(struct sk_buff *skb, struct Vmxnet3_RxCompDesc *rcd,
645                     struct vmxnet3_rx_buf_info *rbi)
646 {
647         struct skb_frag_struct *frag = skb_shinfo(skb)->frags +
648                 skb_shinfo(skb)->nr_frags;
649
650         BUG_ON(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS);
651
652         frag->page = rbi->page;
653         frag->page_offset = 0;
654         frag->size = rcd->len;
655         skb->data_len += frag->size;
656         skb_shinfo(skb)->nr_frags++;
657 }
658
659
660 static void
661 vmxnet3_map_pkt(struct sk_buff *skb, struct vmxnet3_tx_ctx *ctx,
662                 struct vmxnet3_tx_queue *tq, struct pci_dev *pdev,
663                 struct vmxnet3_adapter *adapter)
664 {
665         u32 dw2, len;
666         unsigned long buf_offset;
667         int i;
668         union Vmxnet3_GenericDesc *gdesc;
669         struct vmxnet3_tx_buf_info *tbi = NULL;
670
671         BUG_ON(ctx->copy_size > skb_headlen(skb));
672
673         /* use the previous gen bit for the SOP desc */
674         dw2 = (tq->tx_ring.gen ^ 0x1) << VMXNET3_TXD_GEN_SHIFT;
675
676         ctx->sop_txd = tq->tx_ring.base + tq->tx_ring.next2fill;
677         gdesc = ctx->sop_txd; /* both loops below can be skipped */
678
679         /* no need to map the buffer if headers are copied */
680         if (ctx->copy_size) {
681                 ctx->sop_txd->txd.addr = cpu_to_le64(tq->data_ring.basePA +
682                                         tq->tx_ring.next2fill *
683                                         sizeof(struct Vmxnet3_TxDataDesc));
684                 ctx->sop_txd->dword[2] = cpu_to_le32(dw2 | ctx->copy_size);
685                 ctx->sop_txd->dword[3] = 0;
686
687                 tbi = tq->buf_info + tq->tx_ring.next2fill;
688                 tbi->map_type = VMXNET3_MAP_NONE;
689
690                 dev_dbg(&adapter->netdev->dev,
691                         "txd[%u]: 0x%Lx 0x%x 0x%x\n",
692                         tq->tx_ring.next2fill,
693                         le64_to_cpu(ctx->sop_txd->txd.addr),
694                         ctx->sop_txd->dword[2], ctx->sop_txd->dword[3]);
695                 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
696
697                 /* use the right gen for non-SOP desc */
698                 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
699         }
700
701         /* linear part can use multiple tx desc if it's big */
702         len = skb_headlen(skb) - ctx->copy_size;
703         buf_offset = ctx->copy_size;
704         while (len) {
705                 u32 buf_size;
706
707                 if (len < VMXNET3_MAX_TX_BUF_SIZE) {
708                         buf_size = len;
709                         dw2 |= len;
710                 } else {
711                         buf_size = VMXNET3_MAX_TX_BUF_SIZE;
712                         /* spec says that for TxDesc.len, 0 == 2^14 */
713                 }
714
715                 tbi = tq->buf_info + tq->tx_ring.next2fill;
716                 tbi->map_type = VMXNET3_MAP_SINGLE;
717                 tbi->dma_addr = pci_map_single(adapter->pdev,
718                                 skb->data + buf_offset, buf_size,
719                                 PCI_DMA_TODEVICE);
720
721                 tbi->len = buf_size;
722
723                 gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
724                 BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
725
726                 gdesc->txd.addr = cpu_to_le64(tbi->dma_addr);
727                 gdesc->dword[2] = cpu_to_le32(dw2);
728                 gdesc->dword[3] = 0;
729
730                 dev_dbg(&adapter->netdev->dev,
731                         "txd[%u]: 0x%Lx 0x%x 0x%x\n",
732                         tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr),
733                         le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]);
734                 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
735                 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
736
737                 len -= buf_size;
738                 buf_offset += buf_size;
739         }
740
741         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
742                 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
743
744                 tbi = tq->buf_info + tq->tx_ring.next2fill;
745                 tbi->map_type = VMXNET3_MAP_PAGE;
746                 tbi->dma_addr = pci_map_page(adapter->pdev, frag->page,
747                                              frag->page_offset, frag->size,
748                                              PCI_DMA_TODEVICE);
749
750                 tbi->len = frag->size;
751
752                 gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
753                 BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
754
755                 gdesc->txd.addr = cpu_to_le64(tbi->dma_addr);
756                 gdesc->dword[2] = cpu_to_le32(dw2 | frag->size);
757                 gdesc->dword[3] = 0;
758
759                 dev_dbg(&adapter->netdev->dev,
760                         "txd[%u]: 0x%llu %u %u\n",
761                         tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr),
762                         le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]);
763                 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
764                 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
765         }
766
767         ctx->eop_txd = gdesc;
768
769         /* set the last buf_info for the pkt */
770         tbi->skb = skb;
771         tbi->sop_idx = ctx->sop_txd - tq->tx_ring.base;
772 }
773
774
775 /* Init all tx queues */
776 static void
777 vmxnet3_tq_init_all(struct vmxnet3_adapter *adapter)
778 {
779         int i;
780
781         for (i = 0; i < adapter->num_tx_queues; i++)
782                 vmxnet3_tq_init(&adapter->tx_queue[i], adapter);
783 }
784
785
786 /*
787  *    parse and copy relevant protocol headers:
788  *      For a tso pkt, relevant headers are L2/3/4 including options
789  *      For a pkt requesting csum offloading, they are L2/3 and may include L4
790  *      if it's a TCP/UDP pkt
791  *
792  * Returns:
793  *    -1:  error happens during parsing
794  *     0:  protocol headers parsed, but too big to be copied
795  *     1:  protocol headers parsed and copied
796  *
797  * Other effects:
798  *    1. related *ctx fields are updated.
799  *    2. ctx->copy_size is # of bytes copied
800  *    3. the portion copied is guaranteed to be in the linear part
801  *
802  */
803 static int
804 vmxnet3_parse_and_copy_hdr(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
805                            struct vmxnet3_tx_ctx *ctx,
806                            struct vmxnet3_adapter *adapter)
807 {
808         struct Vmxnet3_TxDataDesc *tdd;
809
810         if (ctx->mss) { /* TSO */
811                 ctx->eth_ip_hdr_size = skb_transport_offset(skb);
812                 ctx->l4_hdr_size = ((struct tcphdr *)
813                                    skb_transport_header(skb))->doff * 4;
814                 ctx->copy_size = ctx->eth_ip_hdr_size + ctx->l4_hdr_size;
815         } else {
816                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
817                         ctx->eth_ip_hdr_size = skb_checksum_start_offset(skb);
818
819                         if (ctx->ipv4) {
820                                 struct iphdr *iph = (struct iphdr *)
821                                                     skb_network_header(skb);
822                                 if (iph->protocol == IPPROTO_TCP)
823                                         ctx->l4_hdr_size = ((struct tcphdr *)
824                                            skb_transport_header(skb))->doff * 4;
825                                 else if (iph->protocol == IPPROTO_UDP)
826                                         /*
827                                          * Use tcp header size so that bytes to
828                                          * be copied are more than required by
829                                          * the device.
830                                          */
831                                         ctx->l4_hdr_size =
832                                                         sizeof(struct tcphdr);
833                                 else
834                                         ctx->l4_hdr_size = 0;
835                         } else {
836                                 /* for simplicity, don't copy L4 headers */
837                                 ctx->l4_hdr_size = 0;
838                         }
839                         ctx->copy_size = ctx->eth_ip_hdr_size +
840                                          ctx->l4_hdr_size;
841                 } else {
842                         ctx->eth_ip_hdr_size = 0;
843                         ctx->l4_hdr_size = 0;
844                         /* copy as much as allowed */
845                         ctx->copy_size = min((unsigned int)VMXNET3_HDR_COPY_SIZE
846                                              , skb_headlen(skb));
847                 }
848
849                 /* make sure headers are accessible directly */
850                 if (unlikely(!pskb_may_pull(skb, ctx->copy_size)))
851                         goto err;
852         }
853
854         if (unlikely(ctx->copy_size > VMXNET3_HDR_COPY_SIZE)) {
855                 tq->stats.oversized_hdr++;
856                 ctx->copy_size = 0;
857                 return 0;
858         }
859
860         tdd = tq->data_ring.base + tq->tx_ring.next2fill;
861
862         memcpy(tdd->data, skb->data, ctx->copy_size);
863         dev_dbg(&adapter->netdev->dev,
864                 "copy %u bytes to dataRing[%u]\n",
865                 ctx->copy_size, tq->tx_ring.next2fill);
866         return 1;
867
868 err:
869         return -1;
870 }
871
872
873 static void
874 vmxnet3_prepare_tso(struct sk_buff *skb,
875                     struct vmxnet3_tx_ctx *ctx)
876 {
877         struct tcphdr *tcph = (struct tcphdr *)skb_transport_header(skb);
878         if (ctx->ipv4) {
879                 struct iphdr *iph = (struct iphdr *)skb_network_header(skb);
880                 iph->check = 0;
881                 tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, 0,
882                                                  IPPROTO_TCP, 0);
883         } else {
884                 struct ipv6hdr *iph = (struct ipv6hdr *)skb_network_header(skb);
885                 tcph->check = ~csum_ipv6_magic(&iph->saddr, &iph->daddr, 0,
886                                                IPPROTO_TCP, 0);
887         }
888 }
889
890
891 /*
892  * Transmits a pkt thru a given tq
893  * Returns:
894  *    NETDEV_TX_OK:      descriptors are setup successfully
895  *    NETDEV_TX_OK:      error occured, the pkt is dropped
896  *    NETDEV_TX_BUSY:    tx ring is full, queue is stopped
897  *
898  * Side-effects:
899  *    1. tx ring may be changed
900  *    2. tq stats may be updated accordingly
901  *    3. shared->txNumDeferred may be updated
902  */
903
904 static int
905 vmxnet3_tq_xmit(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
906                 struct vmxnet3_adapter *adapter, struct net_device *netdev)
907 {
908         int ret;
909         u32 count;
910         unsigned long flags;
911         struct vmxnet3_tx_ctx ctx;
912         union Vmxnet3_GenericDesc *gdesc;
913 #ifdef __BIG_ENDIAN_BITFIELD
914         /* Use temporary descriptor to avoid touching bits multiple times */
915         union Vmxnet3_GenericDesc tempTxDesc;
916 #endif
917
918         /* conservatively estimate # of descriptors to use */
919         count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) +
920                 skb_shinfo(skb)->nr_frags + 1;
921
922         ctx.ipv4 = (skb->protocol == cpu_to_be16(ETH_P_IP));
923
924         ctx.mss = skb_shinfo(skb)->gso_size;
925         if (ctx.mss) {
926                 if (skb_header_cloned(skb)) {
927                         if (unlikely(pskb_expand_head(skb, 0, 0,
928                                                       GFP_ATOMIC) != 0)) {
929                                 tq->stats.drop_tso++;
930                                 goto drop_pkt;
931                         }
932                         tq->stats.copy_skb_header++;
933                 }
934                 vmxnet3_prepare_tso(skb, &ctx);
935         } else {
936                 if (unlikely(count > VMXNET3_MAX_TXD_PER_PKT)) {
937
938                         /* non-tso pkts must not use more than
939                          * VMXNET3_MAX_TXD_PER_PKT entries
940                          */
941                         if (skb_linearize(skb) != 0) {
942                                 tq->stats.drop_too_many_frags++;
943                                 goto drop_pkt;
944                         }
945                         tq->stats.linearized++;
946
947                         /* recalculate the # of descriptors to use */
948                         count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) + 1;
949                 }
950         }
951
952         spin_lock_irqsave(&tq->tx_lock, flags);
953
954         if (count > vmxnet3_cmd_ring_desc_avail(&tq->tx_ring)) {
955                 tq->stats.tx_ring_full++;
956                 dev_dbg(&adapter->netdev->dev,
957                         "tx queue stopped on %s, next2comp %u"
958                         " next2fill %u\n", adapter->netdev->name,
959                         tq->tx_ring.next2comp, tq->tx_ring.next2fill);
960
961                 vmxnet3_tq_stop(tq, adapter);
962                 spin_unlock_irqrestore(&tq->tx_lock, flags);
963                 return NETDEV_TX_BUSY;
964         }
965
966
967         ret = vmxnet3_parse_and_copy_hdr(skb, tq, &ctx, adapter);
968         if (ret >= 0) {
969                 BUG_ON(ret <= 0 && ctx.copy_size != 0);
970                 /* hdrs parsed, check against other limits */
971                 if (ctx.mss) {
972                         if (unlikely(ctx.eth_ip_hdr_size + ctx.l4_hdr_size >
973                                      VMXNET3_MAX_TX_BUF_SIZE)) {
974                                 goto hdr_too_big;
975                         }
976                 } else {
977                         if (skb->ip_summed == CHECKSUM_PARTIAL) {
978                                 if (unlikely(ctx.eth_ip_hdr_size +
979                                              skb->csum_offset >
980                                              VMXNET3_MAX_CSUM_OFFSET)) {
981                                         goto hdr_too_big;
982                                 }
983                         }
984                 }
985         } else {
986                 tq->stats.drop_hdr_inspect_err++;
987                 goto unlock_drop_pkt;
988         }
989
990         /* fill tx descs related to addr & len */
991         vmxnet3_map_pkt(skb, &ctx, tq, adapter->pdev, adapter);
992
993         /* setup the EOP desc */
994         ctx.eop_txd->dword[3] = cpu_to_le32(VMXNET3_TXD_CQ | VMXNET3_TXD_EOP);
995
996         /* setup the SOP desc */
997 #ifdef __BIG_ENDIAN_BITFIELD
998         gdesc = &tempTxDesc;
999         gdesc->dword[2] = ctx.sop_txd->dword[2];
1000         gdesc->dword[3] = ctx.sop_txd->dword[3];
1001 #else
1002         gdesc = ctx.sop_txd;
1003 #endif
1004         if (ctx.mss) {
1005                 gdesc->txd.hlen = ctx.eth_ip_hdr_size + ctx.l4_hdr_size;
1006                 gdesc->txd.om = VMXNET3_OM_TSO;
1007                 gdesc->txd.msscof = ctx.mss;
1008                 le32_add_cpu(&tq->shared->txNumDeferred, (skb->len -
1009                              gdesc->txd.hlen + ctx.mss - 1) / ctx.mss);
1010         } else {
1011                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1012                         gdesc->txd.hlen = ctx.eth_ip_hdr_size;
1013                         gdesc->txd.om = VMXNET3_OM_CSUM;
1014                         gdesc->txd.msscof = ctx.eth_ip_hdr_size +
1015                                             skb->csum_offset;
1016                 } else {
1017                         gdesc->txd.om = 0;
1018                         gdesc->txd.msscof = 0;
1019                 }
1020                 le32_add_cpu(&tq->shared->txNumDeferred, 1);
1021         }
1022
1023         if (vlan_tx_tag_present(skb)) {
1024                 gdesc->txd.ti = 1;
1025                 gdesc->txd.tci = vlan_tx_tag_get(skb);
1026         }
1027
1028         /* finally flips the GEN bit of the SOP desc. */
1029         gdesc->dword[2] = cpu_to_le32(le32_to_cpu(gdesc->dword[2]) ^
1030                                                   VMXNET3_TXD_GEN);
1031 #ifdef __BIG_ENDIAN_BITFIELD
1032         /* Finished updating in bitfields of Tx Desc, so write them in original
1033          * place.
1034          */
1035         vmxnet3_TxDescToLe((struct Vmxnet3_TxDesc *)gdesc,
1036                            (struct Vmxnet3_TxDesc *)ctx.sop_txd);
1037         gdesc = ctx.sop_txd;
1038 #endif
1039         dev_dbg(&adapter->netdev->dev,
1040                 "txd[%u]: SOP 0x%Lx 0x%x 0x%x\n",
1041                 (u32)((union Vmxnet3_GenericDesc *)ctx.sop_txd -
1042                 tq->tx_ring.base), le64_to_cpu(gdesc->txd.addr),
1043                 le32_to_cpu(gdesc->dword[2]), le32_to_cpu(gdesc->dword[3]));
1044
1045         spin_unlock_irqrestore(&tq->tx_lock, flags);
1046
1047         if (le32_to_cpu(tq->shared->txNumDeferred) >=
1048                                         le32_to_cpu(tq->shared->txThreshold)) {
1049                 tq->shared->txNumDeferred = 0;
1050                 VMXNET3_WRITE_BAR0_REG(adapter,
1051                                        VMXNET3_REG_TXPROD + tq->qid * 8,
1052                                        tq->tx_ring.next2fill);
1053         }
1054
1055         return NETDEV_TX_OK;
1056
1057 hdr_too_big:
1058         tq->stats.drop_oversized_hdr++;
1059 unlock_drop_pkt:
1060         spin_unlock_irqrestore(&tq->tx_lock, flags);
1061 drop_pkt:
1062         tq->stats.drop_total++;
1063         dev_kfree_skb(skb);
1064         return NETDEV_TX_OK;
1065 }
1066
1067
1068 static netdev_tx_t
1069 vmxnet3_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
1070 {
1071         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1072
1073                 BUG_ON(skb->queue_mapping > adapter->num_tx_queues);
1074                 return vmxnet3_tq_xmit(skb,
1075                                        &adapter->tx_queue[skb->queue_mapping],
1076                                        adapter, netdev);
1077 }
1078
1079
1080 static void
1081 vmxnet3_rx_csum(struct vmxnet3_adapter *adapter,
1082                 struct sk_buff *skb,
1083                 union Vmxnet3_GenericDesc *gdesc)
1084 {
1085         if (!gdesc->rcd.cnc && adapter->rxcsum) {
1086                 /* typical case: TCP/UDP over IP and both csums are correct */
1087                 if ((le32_to_cpu(gdesc->dword[3]) & VMXNET3_RCD_CSUM_OK) ==
1088                                                         VMXNET3_RCD_CSUM_OK) {
1089                         skb->ip_summed = CHECKSUM_UNNECESSARY;
1090                         BUG_ON(!(gdesc->rcd.tcp || gdesc->rcd.udp));
1091                         BUG_ON(!(gdesc->rcd.v4  || gdesc->rcd.v6));
1092                         BUG_ON(gdesc->rcd.frg);
1093                 } else {
1094                         if (gdesc->rcd.csum) {
1095                                 skb->csum = htons(gdesc->rcd.csum);
1096                                 skb->ip_summed = CHECKSUM_PARTIAL;
1097                         } else {
1098                                 skb_checksum_none_assert(skb);
1099                         }
1100                 }
1101         } else {
1102                 skb_checksum_none_assert(skb);
1103         }
1104 }
1105
1106
1107 static void
1108 vmxnet3_rx_error(struct vmxnet3_rx_queue *rq, struct Vmxnet3_RxCompDesc *rcd,
1109                  struct vmxnet3_rx_ctx *ctx,  struct vmxnet3_adapter *adapter)
1110 {
1111         rq->stats.drop_err++;
1112         if (!rcd->fcs)
1113                 rq->stats.drop_fcs++;
1114
1115         rq->stats.drop_total++;
1116
1117         /*
1118          * We do not unmap and chain the rx buffer to the skb.
1119          * We basically pretend this buffer is not used and will be recycled
1120          * by vmxnet3_rq_alloc_rx_buf()
1121          */
1122
1123         /*
1124          * ctx->skb may be NULL if this is the first and the only one
1125          * desc for the pkt
1126          */
1127         if (ctx->skb)
1128                 dev_kfree_skb_irq(ctx->skb);
1129
1130         ctx->skb = NULL;
1131 }
1132
1133
1134 static int
1135 vmxnet3_rq_rx_complete(struct vmxnet3_rx_queue *rq,
1136                        struct vmxnet3_adapter *adapter, int quota)
1137 {
1138         static const u32 rxprod_reg[2] = {
1139                 VMXNET3_REG_RXPROD, VMXNET3_REG_RXPROD2
1140         };
1141         u32 num_rxd = 0;
1142         struct Vmxnet3_RxCompDesc *rcd;
1143         struct vmxnet3_rx_ctx *ctx = &rq->rx_ctx;
1144 #ifdef __BIG_ENDIAN_BITFIELD
1145         struct Vmxnet3_RxDesc rxCmdDesc;
1146         struct Vmxnet3_RxCompDesc rxComp;
1147 #endif
1148         vmxnet3_getRxComp(rcd, &rq->comp_ring.base[rq->comp_ring.next2proc].rcd,
1149                           &rxComp);
1150         while (rcd->gen == rq->comp_ring.gen) {
1151                 struct vmxnet3_rx_buf_info *rbi;
1152                 struct sk_buff *skb;
1153                 int num_to_alloc;
1154                 struct Vmxnet3_RxDesc *rxd;
1155                 u32 idx, ring_idx;
1156
1157                 if (num_rxd >= quota) {
1158                         /* we may stop even before we see the EOP desc of
1159                          * the current pkt
1160                          */
1161                         break;
1162                 }
1163                 num_rxd++;
1164                 BUG_ON(rcd->rqID != rq->qid && rcd->rqID != rq->qid2);
1165                 idx = rcd->rxdIdx;
1166                 ring_idx = rcd->rqID < adapter->num_rx_queues ? 0 : 1;
1167                 vmxnet3_getRxDesc(rxd, &rq->rx_ring[ring_idx].base[idx].rxd,
1168                                   &rxCmdDesc);
1169                 rbi = rq->buf_info[ring_idx] + idx;
1170
1171                 BUG_ON(rxd->addr != rbi->dma_addr ||
1172                        rxd->len != rbi->len);
1173
1174                 if (unlikely(rcd->eop && rcd->err)) {
1175                         vmxnet3_rx_error(rq, rcd, ctx, adapter);
1176                         goto rcd_done;
1177                 }
1178
1179                 if (rcd->sop) { /* first buf of the pkt */
1180                         BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_HEAD ||
1181                                rcd->rqID != rq->qid);
1182
1183                         BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_SKB);
1184                         BUG_ON(ctx->skb != NULL || rbi->skb == NULL);
1185
1186                         if (unlikely(rcd->len == 0)) {
1187                                 /* Pretend the rx buffer is skipped. */
1188                                 BUG_ON(!(rcd->sop && rcd->eop));
1189                                 dev_dbg(&adapter->netdev->dev,
1190                                         "rxRing[%u][%u] 0 length\n",
1191                                         ring_idx, idx);
1192                                 goto rcd_done;
1193                         }
1194
1195                         ctx->skb = rbi->skb;
1196                         rbi->skb = NULL;
1197
1198                         pci_unmap_single(adapter->pdev, rbi->dma_addr, rbi->len,
1199                                          PCI_DMA_FROMDEVICE);
1200
1201                         skb_put(ctx->skb, rcd->len);
1202                 } else {
1203                         BUG_ON(ctx->skb == NULL);
1204                         /* non SOP buffer must be type 1 in most cases */
1205                         if (rbi->buf_type == VMXNET3_RX_BUF_PAGE) {
1206                                 BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_BODY);
1207
1208                                 if (rcd->len) {
1209                                         pci_unmap_page(adapter->pdev,
1210                                                        rbi->dma_addr, rbi->len,
1211                                                        PCI_DMA_FROMDEVICE);
1212
1213                                         vmxnet3_append_frag(ctx->skb, rcd, rbi);
1214                                         rbi->page = NULL;
1215                                 }
1216                         } else {
1217                                 /*
1218                                  * The only time a non-SOP buffer is type 0 is
1219                                  * when it's EOP and error flag is raised, which
1220                                  * has already been handled.
1221                                  */
1222                                 BUG_ON(true);
1223                         }
1224                 }
1225
1226                 skb = ctx->skb;
1227                 if (rcd->eop) {
1228                         skb->len += skb->data_len;
1229                         skb->truesize += skb->data_len;
1230
1231                         vmxnet3_rx_csum(adapter, skb,
1232                                         (union Vmxnet3_GenericDesc *)rcd);
1233                         skb->protocol = eth_type_trans(skb, adapter->netdev);
1234
1235                         if (unlikely(adapter->vlan_grp && rcd->ts)) {
1236                                 vlan_hwaccel_receive_skb(skb,
1237                                                 adapter->vlan_grp, rcd->tci);
1238                         } else {
1239                                 netif_receive_skb(skb);
1240                         }
1241
1242                         ctx->skb = NULL;
1243                 }
1244
1245 rcd_done:
1246                 /* device may skip some rx descs */
1247                 rq->rx_ring[ring_idx].next2comp = idx;
1248                 VMXNET3_INC_RING_IDX_ONLY(rq->rx_ring[ring_idx].next2comp,
1249                                           rq->rx_ring[ring_idx].size);
1250
1251                 /* refill rx buffers frequently to avoid starving the h/w */
1252                 num_to_alloc = vmxnet3_cmd_ring_desc_avail(rq->rx_ring +
1253                                                            ring_idx);
1254                 if (unlikely(num_to_alloc > VMXNET3_RX_ALLOC_THRESHOLD(rq,
1255                                                         ring_idx, adapter))) {
1256                         vmxnet3_rq_alloc_rx_buf(rq, ring_idx, num_to_alloc,
1257                                                 adapter);
1258
1259                         /* if needed, update the register */
1260                         if (unlikely(rq->shared->updateRxProd)) {
1261                                 VMXNET3_WRITE_BAR0_REG(adapter,
1262                                         rxprod_reg[ring_idx] + rq->qid * 8,
1263                                         rq->rx_ring[ring_idx].next2fill);
1264                                 rq->uncommitted[ring_idx] = 0;
1265                         }
1266                 }
1267
1268                 vmxnet3_comp_ring_adv_next2proc(&rq->comp_ring);
1269                 vmxnet3_getRxComp(rcd,
1270                      &rq->comp_ring.base[rq->comp_ring.next2proc].rcd, &rxComp);
1271         }
1272
1273         return num_rxd;
1274 }
1275
1276
1277 static void
1278 vmxnet3_rq_cleanup(struct vmxnet3_rx_queue *rq,
1279                    struct vmxnet3_adapter *adapter)
1280 {
1281         u32 i, ring_idx;
1282         struct Vmxnet3_RxDesc *rxd;
1283
1284         for (ring_idx = 0; ring_idx < 2; ring_idx++) {
1285                 for (i = 0; i < rq->rx_ring[ring_idx].size; i++) {
1286 #ifdef __BIG_ENDIAN_BITFIELD
1287                         struct Vmxnet3_RxDesc rxDesc;
1288 #endif
1289                         vmxnet3_getRxDesc(rxd,
1290                                 &rq->rx_ring[ring_idx].base[i].rxd, &rxDesc);
1291
1292                         if (rxd->btype == VMXNET3_RXD_BTYPE_HEAD &&
1293                                         rq->buf_info[ring_idx][i].skb) {
1294                                 pci_unmap_single(adapter->pdev, rxd->addr,
1295                                                  rxd->len, PCI_DMA_FROMDEVICE);
1296                                 dev_kfree_skb(rq->buf_info[ring_idx][i].skb);
1297                                 rq->buf_info[ring_idx][i].skb = NULL;
1298                         } else if (rxd->btype == VMXNET3_RXD_BTYPE_BODY &&
1299                                         rq->buf_info[ring_idx][i].page) {
1300                                 pci_unmap_page(adapter->pdev, rxd->addr,
1301                                                rxd->len, PCI_DMA_FROMDEVICE);
1302                                 put_page(rq->buf_info[ring_idx][i].page);
1303                                 rq->buf_info[ring_idx][i].page = NULL;
1304                         }
1305                 }
1306
1307                 rq->rx_ring[ring_idx].gen = VMXNET3_INIT_GEN;
1308                 rq->rx_ring[ring_idx].next2fill =
1309                                         rq->rx_ring[ring_idx].next2comp = 0;
1310                 rq->uncommitted[ring_idx] = 0;
1311         }
1312
1313         rq->comp_ring.gen = VMXNET3_INIT_GEN;
1314         rq->comp_ring.next2proc = 0;
1315 }
1316
1317
1318 static void
1319 vmxnet3_rq_cleanup_all(struct vmxnet3_adapter *adapter)
1320 {
1321         int i;
1322
1323         for (i = 0; i < adapter->num_rx_queues; i++)
1324                 vmxnet3_rq_cleanup(&adapter->rx_queue[i], adapter);
1325 }
1326
1327
1328 void vmxnet3_rq_destroy(struct vmxnet3_rx_queue *rq,
1329                         struct vmxnet3_adapter *adapter)
1330 {
1331         int i;
1332         int j;
1333
1334         /* all rx buffers must have already been freed */
1335         for (i = 0; i < 2; i++) {
1336                 if (rq->buf_info[i]) {
1337                         for (j = 0; j < rq->rx_ring[i].size; j++)
1338                                 BUG_ON(rq->buf_info[i][j].page != NULL);
1339                 }
1340         }
1341
1342
1343         kfree(rq->buf_info[0]);
1344
1345         for (i = 0; i < 2; i++) {
1346                 if (rq->rx_ring[i].base) {
1347                         pci_free_consistent(adapter->pdev, rq->rx_ring[i].size
1348                                             * sizeof(struct Vmxnet3_RxDesc),
1349                                             rq->rx_ring[i].base,
1350                                             rq->rx_ring[i].basePA);
1351                         rq->rx_ring[i].base = NULL;
1352                 }
1353                 rq->buf_info[i] = NULL;
1354         }
1355
1356         if (rq->comp_ring.base) {
1357                 pci_free_consistent(adapter->pdev, rq->comp_ring.size *
1358                                     sizeof(struct Vmxnet3_RxCompDesc),
1359                                     rq->comp_ring.base, rq->comp_ring.basePA);
1360                 rq->comp_ring.base = NULL;
1361         }
1362 }
1363
1364
1365 static int
1366 vmxnet3_rq_init(struct vmxnet3_rx_queue *rq,
1367                 struct vmxnet3_adapter  *adapter)
1368 {
1369         int i;
1370
1371         /* initialize buf_info */
1372         for (i = 0; i < rq->rx_ring[0].size; i++) {
1373
1374                 /* 1st buf for a pkt is skbuff */
1375                 if (i % adapter->rx_buf_per_pkt == 0) {
1376                         rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_SKB;
1377                         rq->buf_info[0][i].len = adapter->skb_buf_size;
1378                 } else { /* subsequent bufs for a pkt is frag */
1379                         rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_PAGE;
1380                         rq->buf_info[0][i].len = PAGE_SIZE;
1381                 }
1382         }
1383         for (i = 0; i < rq->rx_ring[1].size; i++) {
1384                 rq->buf_info[1][i].buf_type = VMXNET3_RX_BUF_PAGE;
1385                 rq->buf_info[1][i].len = PAGE_SIZE;
1386         }
1387
1388         /* reset internal state and allocate buffers for both rings */
1389         for (i = 0; i < 2; i++) {
1390                 rq->rx_ring[i].next2fill = rq->rx_ring[i].next2comp = 0;
1391                 rq->uncommitted[i] = 0;
1392
1393                 memset(rq->rx_ring[i].base, 0, rq->rx_ring[i].size *
1394                        sizeof(struct Vmxnet3_RxDesc));
1395                 rq->rx_ring[i].gen = VMXNET3_INIT_GEN;
1396         }
1397         if (vmxnet3_rq_alloc_rx_buf(rq, 0, rq->rx_ring[0].size - 1,
1398                                     adapter) == 0) {
1399                 /* at least has 1 rx buffer for the 1st ring */
1400                 return -ENOMEM;
1401         }
1402         vmxnet3_rq_alloc_rx_buf(rq, 1, rq->rx_ring[1].size - 1, adapter);
1403
1404         /* reset the comp ring */
1405         rq->comp_ring.next2proc = 0;
1406         memset(rq->comp_ring.base, 0, rq->comp_ring.size *
1407                sizeof(struct Vmxnet3_RxCompDesc));
1408         rq->comp_ring.gen = VMXNET3_INIT_GEN;
1409
1410         /* reset rxctx */
1411         rq->rx_ctx.skb = NULL;
1412
1413         /* stats are not reset */
1414         return 0;
1415 }
1416
1417
1418 static int
1419 vmxnet3_rq_init_all(struct vmxnet3_adapter *adapter)
1420 {
1421         int i, err = 0;
1422
1423         for (i = 0; i < adapter->num_rx_queues; i++) {
1424                 err = vmxnet3_rq_init(&adapter->rx_queue[i], adapter);
1425                 if (unlikely(err)) {
1426                         dev_err(&adapter->netdev->dev, "%s: failed to "
1427                                 "initialize rx queue%i\n",
1428                                 adapter->netdev->name, i);
1429                         break;
1430                 }
1431         }
1432         return err;
1433
1434 }
1435
1436
1437 static int
1438 vmxnet3_rq_create(struct vmxnet3_rx_queue *rq, struct vmxnet3_adapter *adapter)
1439 {
1440         int i;
1441         size_t sz;
1442         struct vmxnet3_rx_buf_info *bi;
1443
1444         for (i = 0; i < 2; i++) {
1445
1446                 sz = rq->rx_ring[i].size * sizeof(struct Vmxnet3_RxDesc);
1447                 rq->rx_ring[i].base = pci_alloc_consistent(adapter->pdev, sz,
1448                                                         &rq->rx_ring[i].basePA);
1449                 if (!rq->rx_ring[i].base) {
1450                         printk(KERN_ERR "%s: failed to allocate rx ring %d\n",
1451                                adapter->netdev->name, i);
1452                         goto err;
1453                 }
1454         }
1455
1456         sz = rq->comp_ring.size * sizeof(struct Vmxnet3_RxCompDesc);
1457         rq->comp_ring.base = pci_alloc_consistent(adapter->pdev, sz,
1458                                                   &rq->comp_ring.basePA);
1459         if (!rq->comp_ring.base) {
1460                 printk(KERN_ERR "%s: failed to allocate rx comp ring\n",
1461                        adapter->netdev->name);
1462                 goto err;
1463         }
1464
1465         sz = sizeof(struct vmxnet3_rx_buf_info) * (rq->rx_ring[0].size +
1466                                                    rq->rx_ring[1].size);
1467         bi = kzalloc(sz, GFP_KERNEL);
1468         if (!bi) {
1469                 printk(KERN_ERR "%s: failed to allocate rx bufinfo\n",
1470                        adapter->netdev->name);
1471                 goto err;
1472         }
1473         rq->buf_info[0] = bi;
1474         rq->buf_info[1] = bi + rq->rx_ring[0].size;
1475
1476         return 0;
1477
1478 err:
1479         vmxnet3_rq_destroy(rq, adapter);
1480         return -ENOMEM;
1481 }
1482
1483
1484 static int
1485 vmxnet3_rq_create_all(struct vmxnet3_adapter *adapter)
1486 {
1487         int i, err = 0;
1488
1489         for (i = 0; i < adapter->num_rx_queues; i++) {
1490                 err = vmxnet3_rq_create(&adapter->rx_queue[i], adapter);
1491                 if (unlikely(err)) {
1492                         dev_err(&adapter->netdev->dev,
1493                                 "%s: failed to create rx queue%i\n",
1494                                 adapter->netdev->name, i);
1495                         goto err_out;
1496                 }
1497         }
1498         return err;
1499 err_out:
1500         vmxnet3_rq_destroy_all(adapter);
1501         return err;
1502
1503 }
1504
1505 /* Multiple queue aware polling function for tx and rx */
1506
1507 static int
1508 vmxnet3_do_poll(struct vmxnet3_adapter *adapter, int budget)
1509 {
1510         int rcd_done = 0, i;
1511         if (unlikely(adapter->shared->ecr))
1512                 vmxnet3_process_events(adapter);
1513         for (i = 0; i < adapter->num_tx_queues; i++)
1514                 vmxnet3_tq_tx_complete(&adapter->tx_queue[i], adapter);
1515
1516         for (i = 0; i < adapter->num_rx_queues; i++)
1517                 rcd_done += vmxnet3_rq_rx_complete(&adapter->rx_queue[i],
1518                                                    adapter, budget);
1519         return rcd_done;
1520 }
1521
1522
1523 static int
1524 vmxnet3_poll(struct napi_struct *napi, int budget)
1525 {
1526         struct vmxnet3_rx_queue *rx_queue = container_of(napi,
1527                                           struct vmxnet3_rx_queue, napi);
1528         int rxd_done;
1529
1530         rxd_done = vmxnet3_do_poll(rx_queue->adapter, budget);
1531
1532         if (rxd_done < budget) {
1533                 napi_complete(napi);
1534                 vmxnet3_enable_all_intrs(rx_queue->adapter);
1535         }
1536         return rxd_done;
1537 }
1538
1539 /*
1540  * NAPI polling function for MSI-X mode with multiple Rx queues
1541  * Returns the # of the NAPI credit consumed (# of rx descriptors processed)
1542  */
1543
1544 static int
1545 vmxnet3_poll_rx_only(struct napi_struct *napi, int budget)
1546 {
1547         struct vmxnet3_rx_queue *rq = container_of(napi,
1548                                                 struct vmxnet3_rx_queue, napi);
1549         struct vmxnet3_adapter *adapter = rq->adapter;
1550         int rxd_done;
1551
1552         /* When sharing interrupt with corresponding tx queue, process
1553          * tx completions in that queue as well
1554          */
1555         if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE) {
1556                 struct vmxnet3_tx_queue *tq =
1557                                 &adapter->tx_queue[rq - adapter->rx_queue];
1558                 vmxnet3_tq_tx_complete(tq, adapter);
1559         }
1560
1561         rxd_done = vmxnet3_rq_rx_complete(rq, adapter, budget);
1562
1563         if (rxd_done < budget) {
1564                 napi_complete(napi);
1565                 vmxnet3_enable_intr(adapter, rq->comp_ring.intr_idx);
1566         }
1567         return rxd_done;
1568 }
1569
1570
1571 #ifdef CONFIG_PCI_MSI
1572
1573 /*
1574  * Handle completion interrupts on tx queues
1575  * Returns whether or not the intr is handled
1576  */
1577
1578 static irqreturn_t
1579 vmxnet3_msix_tx(int irq, void *data)
1580 {
1581         struct vmxnet3_tx_queue *tq = data;
1582         struct vmxnet3_adapter *adapter = tq->adapter;
1583
1584         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1585                 vmxnet3_disable_intr(adapter, tq->comp_ring.intr_idx);
1586
1587         /* Handle the case where only one irq is allocate for all tx queues */
1588         if (adapter->share_intr == VMXNET3_INTR_TXSHARE) {
1589                 int i;
1590                 for (i = 0; i < adapter->num_tx_queues; i++) {
1591                         struct vmxnet3_tx_queue *txq = &adapter->tx_queue[i];
1592                         vmxnet3_tq_tx_complete(txq, adapter);
1593                 }
1594         } else {
1595                 vmxnet3_tq_tx_complete(tq, adapter);
1596         }
1597         vmxnet3_enable_intr(adapter, tq->comp_ring.intr_idx);
1598
1599         return IRQ_HANDLED;
1600 }
1601
1602
1603 /*
1604  * Handle completion interrupts on rx queues. Returns whether or not the
1605  * intr is handled
1606  */
1607
1608 static irqreturn_t
1609 vmxnet3_msix_rx(int irq, void *data)
1610 {
1611         struct vmxnet3_rx_queue *rq = data;
1612         struct vmxnet3_adapter *adapter = rq->adapter;
1613
1614         /* disable intr if needed */
1615         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1616                 vmxnet3_disable_intr(adapter, rq->comp_ring.intr_idx);
1617         napi_schedule(&rq->napi);
1618
1619         return IRQ_HANDLED;
1620 }
1621
1622 /*
1623  *----------------------------------------------------------------------------
1624  *
1625  * vmxnet3_msix_event --
1626  *
1627  *    vmxnet3 msix event intr handler
1628  *
1629  * Result:
1630  *    whether or not the intr is handled
1631  *
1632  *----------------------------------------------------------------------------
1633  */
1634
1635 static irqreturn_t
1636 vmxnet3_msix_event(int irq, void *data)
1637 {
1638         struct net_device *dev = data;
1639         struct vmxnet3_adapter *adapter = netdev_priv(dev);
1640
1641         /* disable intr if needed */
1642         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1643                 vmxnet3_disable_intr(adapter, adapter->intr.event_intr_idx);
1644
1645         if (adapter->shared->ecr)
1646                 vmxnet3_process_events(adapter);
1647
1648         vmxnet3_enable_intr(adapter, adapter->intr.event_intr_idx);
1649
1650         return IRQ_HANDLED;
1651 }
1652
1653 #endif /* CONFIG_PCI_MSI  */
1654
1655
1656 /* Interrupt handler for vmxnet3  */
1657 static irqreturn_t
1658 vmxnet3_intr(int irq, void *dev_id)
1659 {
1660         struct net_device *dev = dev_id;
1661         struct vmxnet3_adapter *adapter = netdev_priv(dev);
1662
1663         if (adapter->intr.type == VMXNET3_IT_INTX) {
1664                 u32 icr = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_ICR);
1665                 if (unlikely(icr == 0))
1666                         /* not ours */
1667                         return IRQ_NONE;
1668         }
1669
1670
1671         /* disable intr if needed */
1672         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1673                 vmxnet3_disable_all_intrs(adapter);
1674
1675         napi_schedule(&adapter->rx_queue[0].napi);
1676
1677         return IRQ_HANDLED;
1678 }
1679
1680 #ifdef CONFIG_NET_POLL_CONTROLLER
1681
1682 /* netpoll callback. */
1683 static void
1684 vmxnet3_netpoll(struct net_device *netdev)
1685 {
1686         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1687
1688         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1689                 vmxnet3_disable_all_intrs(adapter);
1690
1691         vmxnet3_do_poll(adapter, adapter->rx_queue[0].rx_ring[0].size);
1692         vmxnet3_enable_all_intrs(adapter);
1693
1694 }
1695 #endif  /* CONFIG_NET_POLL_CONTROLLER */
1696
1697 static int
1698 vmxnet3_request_irqs(struct vmxnet3_adapter *adapter)
1699 {
1700         struct vmxnet3_intr *intr = &adapter->intr;
1701         int err = 0, i;
1702         int vector = 0;
1703
1704 #ifdef CONFIG_PCI_MSI
1705         if (adapter->intr.type == VMXNET3_IT_MSIX) {
1706                 for (i = 0; i < adapter->num_tx_queues; i++) {
1707                         if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE) {
1708                                 sprintf(adapter->tx_queue[i].name, "%s-tx-%d",
1709                                         adapter->netdev->name, vector);
1710                                 err = request_irq(
1711                                               intr->msix_entries[vector].vector,
1712                                               vmxnet3_msix_tx, 0,
1713                                               adapter->tx_queue[i].name,
1714                                               &adapter->tx_queue[i]);
1715                         } else {
1716                                 sprintf(adapter->tx_queue[i].name, "%s-rxtx-%d",
1717                                         adapter->netdev->name, vector);
1718                         }
1719                         if (err) {
1720                                 dev_err(&adapter->netdev->dev,
1721                                         "Failed to request irq for MSIX, %s, "
1722                                         "error %d\n",
1723                                         adapter->tx_queue[i].name, err);
1724                                 return err;
1725                         }
1726
1727                         /* Handle the case where only 1 MSIx was allocated for
1728                          * all tx queues */
1729                         if (adapter->share_intr == VMXNET3_INTR_TXSHARE) {
1730                                 for (; i < adapter->num_tx_queues; i++)
1731                                         adapter->tx_queue[i].comp_ring.intr_idx
1732                                                                 = vector;
1733                                 vector++;
1734                                 break;
1735                         } else {
1736                                 adapter->tx_queue[i].comp_ring.intr_idx
1737                                                                 = vector++;
1738                         }
1739                 }
1740                 if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE)
1741                         vector = 0;
1742
1743                 for (i = 0; i < adapter->num_rx_queues; i++) {
1744                         if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE)
1745                                 sprintf(adapter->rx_queue[i].name, "%s-rx-%d",
1746                                         adapter->netdev->name, vector);
1747                         else
1748                                 sprintf(adapter->rx_queue[i].name, "%s-rxtx-%d",
1749                                         adapter->netdev->name, vector);
1750                         err = request_irq(intr->msix_entries[vector].vector,
1751                                           vmxnet3_msix_rx, 0,
1752                                           adapter->rx_queue[i].name,
1753                                           &(adapter->rx_queue[i]));
1754                         if (err) {
1755                                 printk(KERN_ERR "Failed to request irq for MSIX"
1756                                        ", %s, error %d\n",
1757                                        adapter->rx_queue[i].name, err);
1758                                 return err;
1759                         }
1760
1761                         adapter->rx_queue[i].comp_ring.intr_idx = vector++;
1762                 }
1763
1764                 sprintf(intr->event_msi_vector_name, "%s-event-%d",
1765                         adapter->netdev->name, vector);
1766                 err = request_irq(intr->msix_entries[vector].vector,
1767                                   vmxnet3_msix_event, 0,
1768                                   intr->event_msi_vector_name, adapter->netdev);
1769                 intr->event_intr_idx = vector;
1770
1771         } else if (intr->type == VMXNET3_IT_MSI) {
1772                 adapter->num_rx_queues = 1;
1773                 err = request_irq(adapter->pdev->irq, vmxnet3_intr, 0,
1774                                   adapter->netdev->name, adapter->netdev);
1775         } else {
1776 #endif
1777                 adapter->num_rx_queues = 1;
1778                 err = request_irq(adapter->pdev->irq, vmxnet3_intr,
1779                                   IRQF_SHARED, adapter->netdev->name,
1780                                   adapter->netdev);
1781 #ifdef CONFIG_PCI_MSI
1782         }
1783 #endif
1784         intr->num_intrs = vector + 1;
1785         if (err) {
1786                 printk(KERN_ERR "Failed to request irq %s (intr type:%d), error"
1787                        ":%d\n", adapter->netdev->name, intr->type, err);
1788         } else {
1789                 /* Number of rx queues will not change after this */
1790                 for (i = 0; i < adapter->num_rx_queues; i++) {
1791                         struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
1792                         rq->qid = i;
1793                         rq->qid2 = i + adapter->num_rx_queues;
1794                 }
1795
1796
1797
1798                 /* init our intr settings */
1799                 for (i = 0; i < intr->num_intrs; i++)
1800                         intr->mod_levels[i] = UPT1_IML_ADAPTIVE;
1801                 if (adapter->intr.type != VMXNET3_IT_MSIX) {
1802                         adapter->intr.event_intr_idx = 0;
1803                         for (i = 0; i < adapter->num_tx_queues; i++)
1804                                 adapter->tx_queue[i].comp_ring.intr_idx = 0;
1805                         adapter->rx_queue[0].comp_ring.intr_idx = 0;
1806                 }
1807
1808                 printk(KERN_INFO "%s: intr type %u, mode %u, %u vectors "
1809                        "allocated\n", adapter->netdev->name, intr->type,
1810                        intr->mask_mode, intr->num_intrs);
1811         }
1812
1813         return err;
1814 }
1815
1816
1817 static void
1818 vmxnet3_free_irqs(struct vmxnet3_adapter *adapter)
1819 {
1820         struct vmxnet3_intr *intr = &adapter->intr;
1821         BUG_ON(intr->type == VMXNET3_IT_AUTO || intr->num_intrs <= 0);
1822
1823         switch (intr->type) {
1824 #ifdef CONFIG_PCI_MSI
1825         case VMXNET3_IT_MSIX:
1826         {
1827                 int i, vector = 0;
1828
1829                 if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE) {
1830                         for (i = 0; i < adapter->num_tx_queues; i++) {
1831                                 free_irq(intr->msix_entries[vector++].vector,
1832                                          &(adapter->tx_queue[i]));
1833                                 if (adapter->share_intr == VMXNET3_INTR_TXSHARE)
1834                                         break;
1835                         }
1836                 }
1837
1838                 for (i = 0; i < adapter->num_rx_queues; i++) {
1839                         free_irq(intr->msix_entries[vector++].vector,
1840                                  &(adapter->rx_queue[i]));
1841                 }
1842
1843                 free_irq(intr->msix_entries[vector].vector,
1844                          adapter->netdev);
1845                 BUG_ON(vector >= intr->num_intrs);
1846                 break;
1847         }
1848 #endif
1849         case VMXNET3_IT_MSI:
1850                 free_irq(adapter->pdev->irq, adapter->netdev);
1851                 break;
1852         case VMXNET3_IT_INTX:
1853                 free_irq(adapter->pdev->irq, adapter->netdev);
1854                 break;
1855         default:
1856                 BUG_ON(true);
1857         }
1858 }
1859
1860 static void
1861 vmxnet3_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp)
1862 {
1863         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1864         struct Vmxnet3_DriverShared *shared = adapter->shared;
1865         u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1866         unsigned long flags;
1867
1868         if (grp) {
1869                 /* add vlan rx stripping. */
1870                 if (adapter->netdev->features & NETIF_F_HW_VLAN_RX) {
1871                         int i;
1872                         adapter->vlan_grp = grp;
1873
1874                         /*
1875                          *  Clear entire vfTable; then enable untagged pkts.
1876                          *  Note: setting one entry in vfTable to non-zero turns
1877                          *  on VLAN rx filtering.
1878                          */
1879                         for (i = 0; i < VMXNET3_VFT_SIZE; i++)
1880                                 vfTable[i] = 0;
1881
1882                         VMXNET3_SET_VFTABLE_ENTRY(vfTable, 0);
1883                         spin_lock_irqsave(&adapter->cmd_lock, flags);
1884                         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1885                                                VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1886                         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
1887                 } else {
1888                         printk(KERN_ERR "%s: vlan_rx_register when device has "
1889                                "no NETIF_F_HW_VLAN_RX\n", netdev->name);
1890                 }
1891         } else {
1892                 /* remove vlan rx stripping. */
1893                 struct Vmxnet3_DSDevRead *devRead = &shared->devRead;
1894                 adapter->vlan_grp = NULL;
1895
1896                 if (devRead->misc.uptFeatures & UPT1_F_RXVLAN) {
1897                         int i;
1898
1899                         for (i = 0; i < VMXNET3_VFT_SIZE; i++) {
1900                                 /* clear entire vfTable; this also disables
1901                                  * VLAN rx filtering
1902                                  */
1903                                 vfTable[i] = 0;
1904                         }
1905                         spin_lock_irqsave(&adapter->cmd_lock, flags);
1906                         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1907                                                VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1908                         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
1909                 }
1910         }
1911 }
1912
1913
1914 static void
1915 vmxnet3_restore_vlan(struct vmxnet3_adapter *adapter)
1916 {
1917         if (adapter->vlan_grp) {
1918                 u16 vid;
1919                 u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1920                 bool activeVlan = false;
1921
1922                 for (vid = 0; vid < VLAN_N_VID; vid++) {
1923                         if (vlan_group_get_device(adapter->vlan_grp, vid)) {
1924                                 VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid);
1925                                 activeVlan = true;
1926                         }
1927                 }
1928                 if (activeVlan) {
1929                         /* continue to allow untagged pkts */
1930                         VMXNET3_SET_VFTABLE_ENTRY(vfTable, 0);
1931                 }
1932         }
1933 }
1934
1935
1936 static void
1937 vmxnet3_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1938 {
1939         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1940         u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1941         unsigned long flags;
1942
1943         VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid);
1944         spin_lock_irqsave(&adapter->cmd_lock, flags);
1945         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1946                                VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1947         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
1948 }
1949
1950
1951 static void
1952 vmxnet3_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1953 {
1954         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1955         u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1956         unsigned long flags;
1957
1958         VMXNET3_CLEAR_VFTABLE_ENTRY(vfTable, vid);
1959         spin_lock_irqsave(&adapter->cmd_lock, flags);
1960         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1961                                VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1962         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
1963 }
1964
1965
1966 static u8 *
1967 vmxnet3_copy_mc(struct net_device *netdev)
1968 {
1969         u8 *buf = NULL;
1970         u32 sz = netdev_mc_count(netdev) * ETH_ALEN;
1971
1972         /* struct Vmxnet3_RxFilterConf.mfTableLen is u16. */
1973         if (sz <= 0xffff) {
1974                 /* We may be called with BH disabled */
1975                 buf = kmalloc(sz, GFP_ATOMIC);
1976                 if (buf) {
1977                         struct netdev_hw_addr *ha;
1978                         int i = 0;
1979
1980                         netdev_for_each_mc_addr(ha, netdev)
1981                                 memcpy(buf + i++ * ETH_ALEN, ha->addr,
1982                                        ETH_ALEN);
1983                 }
1984         }
1985         return buf;
1986 }
1987
1988
1989 static void
1990 vmxnet3_set_mc(struct net_device *netdev)
1991 {
1992         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1993         unsigned long flags;
1994         struct Vmxnet3_RxFilterConf *rxConf =
1995                                         &adapter->shared->devRead.rxFilterConf;
1996         u8 *new_table = NULL;
1997         u32 new_mode = VMXNET3_RXM_UCAST;
1998
1999         if (netdev->flags & IFF_PROMISC)
2000                 new_mode |= VMXNET3_RXM_PROMISC;
2001
2002         if (netdev->flags & IFF_BROADCAST)
2003                 new_mode |= VMXNET3_RXM_BCAST;
2004
2005         if (netdev->flags & IFF_ALLMULTI)
2006                 new_mode |= VMXNET3_RXM_ALL_MULTI;
2007         else
2008                 if (!netdev_mc_empty(netdev)) {
2009                         new_table = vmxnet3_copy_mc(netdev);
2010                         if (new_table) {
2011                                 new_mode |= VMXNET3_RXM_MCAST;
2012                                 rxConf->mfTableLen = cpu_to_le16(
2013                                         netdev_mc_count(netdev) * ETH_ALEN);
2014                                 rxConf->mfTablePA = cpu_to_le64(virt_to_phys(
2015                                                     new_table));
2016                         } else {
2017                                 printk(KERN_INFO "%s: failed to copy mcast list"
2018                                        ", setting ALL_MULTI\n", netdev->name);
2019                                 new_mode |= VMXNET3_RXM_ALL_MULTI;
2020                         }
2021                 }
2022
2023
2024         if (!(new_mode & VMXNET3_RXM_MCAST)) {
2025                 rxConf->mfTableLen = 0;
2026                 rxConf->mfTablePA = 0;
2027         }
2028
2029         spin_lock_irqsave(&adapter->cmd_lock, flags);
2030         if (new_mode != rxConf->rxMode) {
2031                 rxConf->rxMode = cpu_to_le32(new_mode);
2032                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2033                                        VMXNET3_CMD_UPDATE_RX_MODE);
2034         }
2035
2036         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2037                                VMXNET3_CMD_UPDATE_MAC_FILTERS);
2038         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2039
2040         kfree(new_table);
2041 }
2042
2043 void
2044 vmxnet3_rq_destroy_all(struct vmxnet3_adapter *adapter)
2045 {
2046         int i;
2047
2048         for (i = 0; i < adapter->num_rx_queues; i++)
2049                 vmxnet3_rq_destroy(&adapter->rx_queue[i], adapter);
2050 }
2051
2052
2053 /*
2054  *   Set up driver_shared based on settings in adapter.
2055  */
2056
2057 static void
2058 vmxnet3_setup_driver_shared(struct vmxnet3_adapter *adapter)
2059 {
2060         struct Vmxnet3_DriverShared *shared = adapter->shared;
2061         struct Vmxnet3_DSDevRead *devRead = &shared->devRead;
2062         struct Vmxnet3_TxQueueConf *tqc;
2063         struct Vmxnet3_RxQueueConf *rqc;
2064         int i;
2065
2066         memset(shared, 0, sizeof(*shared));
2067
2068         /* driver settings */
2069         shared->magic = cpu_to_le32(VMXNET3_REV1_MAGIC);
2070         devRead->misc.driverInfo.version = cpu_to_le32(
2071                                                 VMXNET3_DRIVER_VERSION_NUM);
2072         devRead->misc.driverInfo.gos.gosBits = (sizeof(void *) == 4 ?
2073                                 VMXNET3_GOS_BITS_32 : VMXNET3_GOS_BITS_64);
2074         devRead->misc.driverInfo.gos.gosType = VMXNET3_GOS_TYPE_LINUX;
2075         *((u32 *)&devRead->misc.driverInfo.gos) = cpu_to_le32(
2076                                 *((u32 *)&devRead->misc.driverInfo.gos));
2077         devRead->misc.driverInfo.vmxnet3RevSpt = cpu_to_le32(1);
2078         devRead->misc.driverInfo.uptVerSpt = cpu_to_le32(1);
2079
2080         devRead->misc.ddPA = cpu_to_le64(virt_to_phys(adapter));
2081         devRead->misc.ddLen = cpu_to_le32(sizeof(struct vmxnet3_adapter));
2082
2083         /* set up feature flags */
2084         if (adapter->rxcsum)
2085                 devRead->misc.uptFeatures |= UPT1_F_RXCSUM;
2086
2087         if (adapter->lro) {
2088                 devRead->misc.uptFeatures |= UPT1_F_LRO;
2089                 devRead->misc.maxNumRxSG = cpu_to_le16(1 + MAX_SKB_FRAGS);
2090         }
2091         if (adapter->netdev->features & NETIF_F_HW_VLAN_RX)
2092                 devRead->misc.uptFeatures |= UPT1_F_RXVLAN;
2093
2094         devRead->misc.mtu = cpu_to_le32(adapter->netdev->mtu);
2095         devRead->misc.queueDescPA = cpu_to_le64(adapter->queue_desc_pa);
2096         devRead->misc.queueDescLen = cpu_to_le32(
2097                 adapter->num_tx_queues * sizeof(struct Vmxnet3_TxQueueDesc) +
2098                 adapter->num_rx_queues * sizeof(struct Vmxnet3_RxQueueDesc));
2099
2100         /* tx queue settings */
2101         devRead->misc.numTxQueues =  adapter->num_tx_queues;
2102         for (i = 0; i < adapter->num_tx_queues; i++) {
2103                 struct vmxnet3_tx_queue *tq = &adapter->tx_queue[i];
2104                 BUG_ON(adapter->tx_queue[i].tx_ring.base == NULL);
2105                 tqc = &adapter->tqd_start[i].conf;
2106                 tqc->txRingBasePA   = cpu_to_le64(tq->tx_ring.basePA);
2107                 tqc->dataRingBasePA = cpu_to_le64(tq->data_ring.basePA);
2108                 tqc->compRingBasePA = cpu_to_le64(tq->comp_ring.basePA);
2109                 tqc->ddPA           = cpu_to_le64(virt_to_phys(tq->buf_info));
2110                 tqc->txRingSize     = cpu_to_le32(tq->tx_ring.size);
2111                 tqc->dataRingSize   = cpu_to_le32(tq->data_ring.size);
2112                 tqc->compRingSize   = cpu_to_le32(tq->comp_ring.size);
2113                 tqc->ddLen          = cpu_to_le32(
2114                                         sizeof(struct vmxnet3_tx_buf_info) *
2115                                         tqc->txRingSize);
2116                 tqc->intrIdx        = tq->comp_ring.intr_idx;
2117         }
2118
2119         /* rx queue settings */
2120         devRead->misc.numRxQueues = adapter->num_rx_queues;
2121         for (i = 0; i < adapter->num_rx_queues; i++) {
2122                 struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
2123                 rqc = &adapter->rqd_start[i].conf;
2124                 rqc->rxRingBasePA[0] = cpu_to_le64(rq->rx_ring[0].basePA);
2125                 rqc->rxRingBasePA[1] = cpu_to_le64(rq->rx_ring[1].basePA);
2126                 rqc->compRingBasePA  = cpu_to_le64(rq->comp_ring.basePA);
2127                 rqc->ddPA            = cpu_to_le64(virt_to_phys(
2128                                                         rq->buf_info));
2129                 rqc->rxRingSize[0]   = cpu_to_le32(rq->rx_ring[0].size);
2130                 rqc->rxRingSize[1]   = cpu_to_le32(rq->rx_ring[1].size);
2131                 rqc->compRingSize    = cpu_to_le32(rq->comp_ring.size);
2132                 rqc->ddLen           = cpu_to_le32(
2133                                         sizeof(struct vmxnet3_rx_buf_info) *
2134                                         (rqc->rxRingSize[0] +
2135                                          rqc->rxRingSize[1]));
2136                 rqc->intrIdx         = rq->comp_ring.intr_idx;
2137         }
2138
2139 #ifdef VMXNET3_RSS
2140         memset(adapter->rss_conf, 0, sizeof(*adapter->rss_conf));
2141
2142         if (adapter->rss) {
2143                 struct UPT1_RSSConf *rssConf = adapter->rss_conf;
2144                 devRead->misc.uptFeatures |= UPT1_F_RSS;
2145                 devRead->misc.numRxQueues = adapter->num_rx_queues;
2146                 rssConf->hashType = UPT1_RSS_HASH_TYPE_TCP_IPV4 |
2147                                     UPT1_RSS_HASH_TYPE_IPV4 |
2148                                     UPT1_RSS_HASH_TYPE_TCP_IPV6 |
2149                                     UPT1_RSS_HASH_TYPE_IPV6;
2150                 rssConf->hashFunc = UPT1_RSS_HASH_FUNC_TOEPLITZ;
2151                 rssConf->hashKeySize = UPT1_RSS_MAX_KEY_SIZE;
2152                 rssConf->indTableSize = VMXNET3_RSS_IND_TABLE_SIZE;
2153                 get_random_bytes(&rssConf->hashKey[0], rssConf->hashKeySize);
2154                 for (i = 0; i < rssConf->indTableSize; i++)
2155                         rssConf->indTable[i] = i % adapter->num_rx_queues;
2156
2157                 devRead->rssConfDesc.confVer = 1;
2158                 devRead->rssConfDesc.confLen = sizeof(*rssConf);
2159                 devRead->rssConfDesc.confPA  = virt_to_phys(rssConf);
2160         }
2161
2162 #endif /* VMXNET3_RSS */
2163
2164         /* intr settings */
2165         devRead->intrConf.autoMask = adapter->intr.mask_mode ==
2166                                      VMXNET3_IMM_AUTO;
2167         devRead->intrConf.numIntrs = adapter->intr.num_intrs;
2168         for (i = 0; i < adapter->intr.num_intrs; i++)
2169                 devRead->intrConf.modLevels[i] = adapter->intr.mod_levels[i];
2170
2171         devRead->intrConf.eventIntrIdx = adapter->intr.event_intr_idx;
2172         devRead->intrConf.intrCtrl |= cpu_to_le32(VMXNET3_IC_DISABLE_ALL);
2173
2174         /* rx filter settings */
2175         devRead->rxFilterConf.rxMode = 0;
2176         vmxnet3_restore_vlan(adapter);
2177         vmxnet3_write_mac_addr(adapter, adapter->netdev->dev_addr);
2178
2179         /* the rest are already zeroed */
2180 }
2181
2182
2183 int
2184 vmxnet3_activate_dev(struct vmxnet3_adapter *adapter)
2185 {
2186         int err, i;
2187         u32 ret;
2188         unsigned long flags;
2189
2190         dev_dbg(&adapter->netdev->dev, "%s: skb_buf_size %d, rx_buf_per_pkt %d,"
2191                 " ring sizes %u %u %u\n", adapter->netdev->name,
2192                 adapter->skb_buf_size, adapter->rx_buf_per_pkt,
2193                 adapter->tx_queue[0].tx_ring.size,
2194                 adapter->rx_queue[0].rx_ring[0].size,
2195                 adapter->rx_queue[0].rx_ring[1].size);
2196
2197         vmxnet3_tq_init_all(adapter);
2198         err = vmxnet3_rq_init_all(adapter);
2199         if (err) {
2200                 printk(KERN_ERR "Failed to init rx queue for %s: error %d\n",
2201                        adapter->netdev->name, err);
2202                 goto rq_err;
2203         }
2204
2205         err = vmxnet3_request_irqs(adapter);
2206         if (err) {
2207                 printk(KERN_ERR "Failed to setup irq for %s: error %d\n",
2208                        adapter->netdev->name, err);
2209                 goto irq_err;
2210         }
2211
2212         vmxnet3_setup_driver_shared(adapter);
2213
2214         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, VMXNET3_GET_ADDR_LO(
2215                                adapter->shared_pa));
2216         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, VMXNET3_GET_ADDR_HI(
2217                                adapter->shared_pa));
2218         spin_lock_irqsave(&adapter->cmd_lock, flags);
2219         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2220                                VMXNET3_CMD_ACTIVATE_DEV);
2221         ret = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
2222         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2223
2224         if (ret != 0) {
2225                 printk(KERN_ERR "Failed to activate dev %s: error %u\n",
2226                        adapter->netdev->name, ret);
2227                 err = -EINVAL;
2228                 goto activate_err;
2229         }
2230
2231         for (i = 0; i < adapter->num_rx_queues; i++) {
2232                 VMXNET3_WRITE_BAR0_REG(adapter,
2233                                 VMXNET3_REG_RXPROD + i * VMXNET3_REG_ALIGN,
2234                                 adapter->rx_queue[i].rx_ring[0].next2fill);
2235                 VMXNET3_WRITE_BAR0_REG(adapter, (VMXNET3_REG_RXPROD2 +
2236                                 (i * VMXNET3_REG_ALIGN)),
2237                                 adapter->rx_queue[i].rx_ring[1].next2fill);
2238         }
2239
2240         /* Apply the rx filter settins last. */
2241         vmxnet3_set_mc(adapter->netdev);
2242
2243         /*
2244          * Check link state when first activating device. It will start the
2245          * tx queue if the link is up.
2246          */
2247         vmxnet3_check_link(adapter, true);
2248         for (i = 0; i < adapter->num_rx_queues; i++)
2249                 napi_enable(&adapter->rx_queue[i].napi);
2250         vmxnet3_enable_all_intrs(adapter);
2251         clear_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
2252         return 0;
2253
2254 activate_err:
2255         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, 0);
2256         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, 0);
2257         vmxnet3_free_irqs(adapter);
2258 irq_err:
2259 rq_err:
2260         /* free up buffers we allocated */
2261         vmxnet3_rq_cleanup_all(adapter);
2262         return err;
2263 }
2264
2265
2266 void
2267 vmxnet3_reset_dev(struct vmxnet3_adapter *adapter)
2268 {
2269         unsigned long flags;
2270         spin_lock_irqsave(&adapter->cmd_lock, flags);
2271         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, VMXNET3_CMD_RESET_DEV);
2272         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2273 }
2274
2275
2276 int
2277 vmxnet3_quiesce_dev(struct vmxnet3_adapter *adapter)
2278 {
2279         int i;
2280         unsigned long flags;
2281         if (test_and_set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state))
2282                 return 0;
2283
2284
2285         spin_lock_irqsave(&adapter->cmd_lock, flags);
2286         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2287                                VMXNET3_CMD_QUIESCE_DEV);
2288         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2289         vmxnet3_disable_all_intrs(adapter);
2290
2291         for (i = 0; i < adapter->num_rx_queues; i++)
2292                 napi_disable(&adapter->rx_queue[i].napi);
2293         netif_tx_disable(adapter->netdev);
2294         adapter->link_speed = 0;
2295         netif_carrier_off(adapter->netdev);
2296
2297         vmxnet3_tq_cleanup_all(adapter);
2298         vmxnet3_rq_cleanup_all(adapter);
2299         vmxnet3_free_irqs(adapter);
2300         return 0;
2301 }
2302
2303
2304 static void
2305 vmxnet3_write_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac)
2306 {
2307         u32 tmp;
2308
2309         tmp = *(u32 *)mac;
2310         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACL, tmp);
2311
2312         tmp = (mac[5] << 8) | mac[4];
2313         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACH, tmp);
2314 }
2315
2316
2317 static int
2318 vmxnet3_set_mac_addr(struct net_device *netdev, void *p)
2319 {
2320         struct sockaddr *addr = p;
2321         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2322
2323         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2324         vmxnet3_write_mac_addr(adapter, addr->sa_data);
2325
2326         return 0;
2327 }
2328
2329
2330 /* ==================== initialization and cleanup routines ============ */
2331
2332 static int
2333 vmxnet3_alloc_pci_resources(struct vmxnet3_adapter *adapter, bool *dma64)
2334 {
2335         int err;
2336         unsigned long mmio_start, mmio_len;
2337         struct pci_dev *pdev = adapter->pdev;
2338
2339         err = pci_enable_device(pdev);
2340         if (err) {
2341                 printk(KERN_ERR "Failed to enable adapter %s: error %d\n",
2342                        pci_name(pdev), err);
2343                 return err;
2344         }
2345
2346         if (pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) == 0) {
2347                 if (pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64)) != 0) {
2348                         printk(KERN_ERR "pci_set_consistent_dma_mask failed "
2349                                "for adapter %s\n", pci_name(pdev));
2350                         err = -EIO;
2351                         goto err_set_mask;
2352                 }
2353                 *dma64 = true;
2354         } else {
2355                 if (pci_set_dma_mask(pdev, DMA_BIT_MASK(32)) != 0) {
2356                         printk(KERN_ERR "pci_set_dma_mask failed for adapter "
2357                                "%s\n",  pci_name(pdev));
2358                         err = -EIO;
2359                         goto err_set_mask;
2360                 }
2361                 *dma64 = false;
2362         }
2363
2364         err = pci_request_selected_regions(pdev, (1 << 2) - 1,
2365                                            vmxnet3_driver_name);
2366         if (err) {
2367                 printk(KERN_ERR "Failed to request region for adapter %s: "
2368                        "error %d\n", pci_name(pdev), err);
2369                 goto err_set_mask;
2370         }
2371
2372         pci_set_master(pdev);
2373
2374         mmio_start = pci_resource_start(pdev, 0);
2375         mmio_len = pci_resource_len(pdev, 0);
2376         adapter->hw_addr0 = ioremap(mmio_start, mmio_len);
2377         if (!adapter->hw_addr0) {
2378                 printk(KERN_ERR "Failed to map bar0 for adapter %s\n",
2379                        pci_name(pdev));
2380                 err = -EIO;
2381                 goto err_ioremap;
2382         }
2383
2384         mmio_start = pci_resource_start(pdev, 1);
2385         mmio_len = pci_resource_len(pdev, 1);
2386         adapter->hw_addr1 = ioremap(mmio_start, mmio_len);
2387         if (!adapter->hw_addr1) {
2388                 printk(KERN_ERR "Failed to map bar1 for adapter %s\n",
2389                        pci_name(pdev));
2390                 err = -EIO;
2391                 goto err_bar1;
2392         }
2393         return 0;
2394
2395 err_bar1:
2396         iounmap(adapter->hw_addr0);
2397 err_ioremap:
2398         pci_release_selected_regions(pdev, (1 << 2) - 1);
2399 err_set_mask:
2400         pci_disable_device(pdev);
2401         return err;
2402 }
2403
2404
2405 static void
2406 vmxnet3_free_pci_resources(struct vmxnet3_adapter *adapter)
2407 {
2408         BUG_ON(!adapter->pdev);
2409
2410         iounmap(adapter->hw_addr0);
2411         iounmap(adapter->hw_addr1);
2412         pci_release_selected_regions(adapter->pdev, (1 << 2) - 1);
2413         pci_disable_device(adapter->pdev);
2414 }
2415
2416
2417 static void
2418 vmxnet3_adjust_rx_ring_size(struct vmxnet3_adapter *adapter)
2419 {
2420         size_t sz, i, ring0_size, ring1_size, comp_size;
2421         struct vmxnet3_rx_queue *rq = &adapter->rx_queue[0];
2422
2423
2424         if (adapter->netdev->mtu <= VMXNET3_MAX_SKB_BUF_SIZE -
2425                                     VMXNET3_MAX_ETH_HDR_SIZE) {
2426                 adapter->skb_buf_size = adapter->netdev->mtu +
2427                                         VMXNET3_MAX_ETH_HDR_SIZE;
2428                 if (adapter->skb_buf_size < VMXNET3_MIN_T0_BUF_SIZE)
2429                         adapter->skb_buf_size = VMXNET3_MIN_T0_BUF_SIZE;
2430
2431                 adapter->rx_buf_per_pkt = 1;
2432         } else {
2433                 adapter->skb_buf_size = VMXNET3_MAX_SKB_BUF_SIZE;
2434                 sz = adapter->netdev->mtu - VMXNET3_MAX_SKB_BUF_SIZE +
2435                                             VMXNET3_MAX_ETH_HDR_SIZE;
2436                 adapter->rx_buf_per_pkt = 1 + (sz + PAGE_SIZE - 1) / PAGE_SIZE;
2437         }
2438
2439         /*
2440          * for simplicity, force the ring0 size to be a multiple of
2441          * rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN
2442          */
2443         sz = adapter->rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN;
2444         ring0_size = adapter->rx_queue[0].rx_ring[0].size;
2445         ring0_size = (ring0_size + sz - 1) / sz * sz;
2446         ring0_size = min_t(u32, ring0_size, VMXNET3_RX_RING_MAX_SIZE /
2447                            sz * sz);
2448         ring1_size = adapter->rx_queue[0].rx_ring[1].size;
2449         comp_size = ring0_size + ring1_size;
2450
2451         for (i = 0; i < adapter->num_rx_queues; i++) {
2452                 rq = &adapter->rx_queue[i];
2453                 rq->rx_ring[0].size = ring0_size;
2454                 rq->rx_ring[1].size = ring1_size;
2455                 rq->comp_ring.size = comp_size;
2456         }
2457 }
2458
2459
2460 int
2461 vmxnet3_create_queues(struct vmxnet3_adapter *adapter, u32 tx_ring_size,
2462                       u32 rx_ring_size, u32 rx_ring2_size)
2463 {
2464         int err = 0, i;
2465
2466         for (i = 0; i < adapter->num_tx_queues; i++) {
2467                 struct vmxnet3_tx_queue *tq = &adapter->tx_queue[i];
2468                 tq->tx_ring.size   = tx_ring_size;
2469                 tq->data_ring.size = tx_ring_size;
2470                 tq->comp_ring.size = tx_ring_size;
2471                 tq->shared = &adapter->tqd_start[i].ctrl;
2472                 tq->stopped = true;
2473                 tq->adapter = adapter;
2474                 tq->qid = i;
2475                 err = vmxnet3_tq_create(tq, adapter);
2476                 /*
2477                  * Too late to change num_tx_queues. We cannot do away with
2478                  * lesser number of queues than what we asked for
2479                  */
2480                 if (err)
2481                         goto queue_err;
2482         }
2483
2484         adapter->rx_queue[0].rx_ring[0].size = rx_ring_size;
2485         adapter->rx_queue[0].rx_ring[1].size = rx_ring2_size;
2486         vmxnet3_adjust_rx_ring_size(adapter);
2487         for (i = 0; i < adapter->num_rx_queues; i++) {
2488                 struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
2489                 /* qid and qid2 for rx queues will be assigned later when num
2490                  * of rx queues is finalized after allocating intrs */
2491                 rq->shared = &adapter->rqd_start[i].ctrl;
2492                 rq->adapter = adapter;
2493                 err = vmxnet3_rq_create(rq, adapter);
2494                 if (err) {
2495                         if (i == 0) {
2496                                 printk(KERN_ERR "Could not allocate any rx"
2497                                        "queues. Aborting.\n");
2498                                 goto queue_err;
2499                         } else {
2500                                 printk(KERN_INFO "Number of rx queues changed "
2501                                        "to : %d.\n", i);
2502                                 adapter->num_rx_queues = i;
2503                                 err = 0;
2504                                 break;
2505                         }
2506                 }
2507         }
2508         return err;
2509 queue_err:
2510         vmxnet3_tq_destroy_all(adapter);
2511         return err;
2512 }
2513
2514 static int
2515 vmxnet3_open(struct net_device *netdev)
2516 {
2517         struct vmxnet3_adapter *adapter;
2518         int err, i;
2519
2520         adapter = netdev_priv(netdev);
2521
2522         for (i = 0; i < adapter->num_tx_queues; i++)
2523                 spin_lock_init(&adapter->tx_queue[i].tx_lock);
2524
2525         err = vmxnet3_create_queues(adapter, VMXNET3_DEF_TX_RING_SIZE,
2526                                     VMXNET3_DEF_RX_RING_SIZE,
2527                                     VMXNET3_DEF_RX_RING_SIZE);
2528         if (err)
2529                 goto queue_err;
2530
2531         err = vmxnet3_activate_dev(adapter);
2532         if (err)
2533                 goto activate_err;
2534
2535         return 0;
2536
2537 activate_err:
2538         vmxnet3_rq_destroy_all(adapter);
2539         vmxnet3_tq_destroy_all(adapter);
2540 queue_err:
2541         return err;
2542 }
2543
2544
2545 static int
2546 vmxnet3_close(struct net_device *netdev)
2547 {
2548         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2549
2550         /*
2551          * Reset_work may be in the middle of resetting the device, wait for its
2552          * completion.
2553          */
2554         while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2555                 msleep(1);
2556
2557         vmxnet3_quiesce_dev(adapter);
2558
2559         vmxnet3_rq_destroy_all(adapter);
2560         vmxnet3_tq_destroy_all(adapter);
2561
2562         clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2563
2564
2565         return 0;
2566 }
2567
2568
2569 void
2570 vmxnet3_force_close(struct vmxnet3_adapter *adapter)
2571 {
2572         int i;
2573
2574         /*
2575          * we must clear VMXNET3_STATE_BIT_RESETTING, otherwise
2576          * vmxnet3_close() will deadlock.
2577          */
2578         BUG_ON(test_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state));
2579
2580         /* we need to enable NAPI, otherwise dev_close will deadlock */
2581         for (i = 0; i < adapter->num_rx_queues; i++)
2582                 napi_enable(&adapter->rx_queue[i].napi);
2583         dev_close(adapter->netdev);
2584 }
2585
2586
2587 static int
2588 vmxnet3_change_mtu(struct net_device *netdev, int new_mtu)
2589 {
2590         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2591         int err = 0;
2592
2593         if (new_mtu < VMXNET3_MIN_MTU || new_mtu > VMXNET3_MAX_MTU)
2594                 return -EINVAL;
2595
2596         if (new_mtu > 1500 && !adapter->jumbo_frame)
2597                 return -EINVAL;
2598
2599         netdev->mtu = new_mtu;
2600
2601         /*
2602          * Reset_work may be in the middle of resetting the device, wait for its
2603          * completion.
2604          */
2605         while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2606                 msleep(1);
2607
2608         if (netif_running(netdev)) {
2609                 vmxnet3_quiesce_dev(adapter);
2610                 vmxnet3_reset_dev(adapter);
2611
2612                 /* we need to re-create the rx queue based on the new mtu */
2613                 vmxnet3_rq_destroy_all(adapter);
2614                 vmxnet3_adjust_rx_ring_size(adapter);
2615                 err = vmxnet3_rq_create_all(adapter);
2616                 if (err) {
2617                         printk(KERN_ERR "%s: failed to re-create rx queues,"
2618                                 " error %d. Closing it.\n", netdev->name, err);
2619                         goto out;
2620                 }
2621
2622                 err = vmxnet3_activate_dev(adapter);
2623                 if (err) {
2624                         printk(KERN_ERR "%s: failed to re-activate, error %d. "
2625                                 "Closing it\n", netdev->name, err);
2626                         goto out;
2627                 }
2628         }
2629
2630 out:
2631         clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2632         if (err)
2633                 vmxnet3_force_close(adapter);
2634
2635         return err;
2636 }
2637
2638
2639 static void
2640 vmxnet3_declare_features(struct vmxnet3_adapter *adapter, bool dma64)
2641 {
2642         struct net_device *netdev = adapter->netdev;
2643
2644         netdev->features = NETIF_F_SG |
2645                 NETIF_F_HW_CSUM |
2646                 NETIF_F_HW_VLAN_TX |
2647                 NETIF_F_HW_VLAN_RX |
2648                 NETIF_F_HW_VLAN_FILTER |
2649                 NETIF_F_TSO |
2650                 NETIF_F_TSO6 |
2651                 NETIF_F_LRO;
2652
2653         printk(KERN_INFO "features: sg csum vlan jf tso tsoIPv6 lro");
2654
2655         adapter->rxcsum = true;
2656         adapter->jumbo_frame = true;
2657         adapter->lro = true;
2658
2659         if (dma64) {
2660                 netdev->features |= NETIF_F_HIGHDMA;
2661                 printk(" highDMA");
2662         }
2663
2664         netdev->vlan_features = netdev->features;
2665         printk("\n");
2666 }
2667
2668
2669 static void
2670 vmxnet3_read_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac)
2671 {
2672         u32 tmp;
2673
2674         tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACL);
2675         *(u32 *)mac = tmp;
2676
2677         tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACH);
2678         mac[4] = tmp & 0xff;
2679         mac[5] = (tmp >> 8) & 0xff;
2680 }
2681
2682 #ifdef CONFIG_PCI_MSI
2683
2684 /*
2685  * Enable MSIx vectors.
2686  * Returns :
2687  *      0 on successful enabling of required vectors,
2688  *      VMXNET3_LINUX_MIN_MSIX_VECT when only minumum number of vectors required
2689  *       could be enabled.
2690  *      number of vectors which can be enabled otherwise (this number is smaller
2691  *       than VMXNET3_LINUX_MIN_MSIX_VECT)
2692  */
2693
2694 static int
2695 vmxnet3_acquire_msix_vectors(struct vmxnet3_adapter *adapter,
2696                              int vectors)
2697 {
2698         int err = 0, vector_threshold;
2699         vector_threshold = VMXNET3_LINUX_MIN_MSIX_VECT;
2700
2701         while (vectors >= vector_threshold) {
2702                 err = pci_enable_msix(adapter->pdev, adapter->intr.msix_entries,
2703                                       vectors);
2704                 if (!err) {
2705                         adapter->intr.num_intrs = vectors;
2706                         return 0;
2707                 } else if (err < 0) {
2708                         printk(KERN_ERR "Failed to enable MSI-X for %s, error"
2709                                " %d\n", adapter->netdev->name, err);
2710                         vectors = 0;
2711                 } else if (err < vector_threshold) {
2712                         break;
2713                 } else {
2714                         /* If fails to enable required number of MSI-x vectors
2715                          * try enabling 3 of them. One each for rx, tx and event
2716                          */
2717                         vectors = vector_threshold;
2718                         printk(KERN_ERR "Failed to enable %d MSI-X for %s, try"
2719                                " %d instead\n", vectors, adapter->netdev->name,
2720                                vector_threshold);
2721                 }
2722         }
2723
2724         printk(KERN_INFO "Number of MSI-X interrupts which can be allocatedi"
2725                " are lower than min threshold required.\n");
2726         return err;
2727 }
2728
2729
2730 #endif /* CONFIG_PCI_MSI */
2731
2732 static void
2733 vmxnet3_alloc_intr_resources(struct vmxnet3_adapter *adapter)
2734 {
2735         u32 cfg;
2736
2737         /* intr settings */
2738         spin_lock(&adapter->cmd_lock);
2739         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2740                                VMXNET3_CMD_GET_CONF_INTR);
2741         cfg = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
2742         spin_unlock(&adapter->cmd_lock);
2743         adapter->intr.type = cfg & 0x3;
2744         adapter->intr.mask_mode = (cfg >> 2) & 0x3;
2745
2746         if (adapter->intr.type == VMXNET3_IT_AUTO) {
2747                 adapter->intr.type = VMXNET3_IT_MSIX;
2748         }
2749
2750 #ifdef CONFIG_PCI_MSI
2751         if (adapter->intr.type == VMXNET3_IT_MSIX) {
2752                 int vector, err = 0;
2753
2754                 adapter->intr.num_intrs = (adapter->share_intr ==
2755                                            VMXNET3_INTR_TXSHARE) ? 1 :
2756                                            adapter->num_tx_queues;
2757                 adapter->intr.num_intrs += (adapter->share_intr ==
2758                                            VMXNET3_INTR_BUDDYSHARE) ? 0 :
2759                                            adapter->num_rx_queues;
2760                 adapter->intr.num_intrs += 1;           /* for link event */
2761
2762                 adapter->intr.num_intrs = (adapter->intr.num_intrs >
2763                                            VMXNET3_LINUX_MIN_MSIX_VECT
2764                                            ? adapter->intr.num_intrs :
2765                                            VMXNET3_LINUX_MIN_MSIX_VECT);
2766
2767                 for (vector = 0; vector < adapter->intr.num_intrs; vector++)
2768                         adapter->intr.msix_entries[vector].entry = vector;
2769
2770                 err = vmxnet3_acquire_msix_vectors(adapter,
2771                                                    adapter->intr.num_intrs);
2772                 /* If we cannot allocate one MSIx vector per queue
2773                  * then limit the number of rx queues to 1
2774                  */
2775                 if (err == VMXNET3_LINUX_MIN_MSIX_VECT) {
2776                         if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE
2777                             || adapter->num_rx_queues != 2) {
2778                                 adapter->share_intr = VMXNET3_INTR_TXSHARE;
2779                                 printk(KERN_ERR "Number of rx queues : 1\n");
2780                                 adapter->num_rx_queues = 1;
2781                                 adapter->intr.num_intrs =
2782                                                 VMXNET3_LINUX_MIN_MSIX_VECT;
2783                         }
2784                         return;
2785                 }
2786                 if (!err)
2787                         return;
2788
2789                 /* If we cannot allocate MSIx vectors use only one rx queue */
2790                 printk(KERN_INFO "Failed to enable MSI-X for %s, error %d."
2791                        "#rx queues : 1, try MSI\n", adapter->netdev->name, err);
2792
2793                 adapter->intr.type = VMXNET3_IT_MSI;
2794         }
2795
2796         if (adapter->intr.type == VMXNET3_IT_MSI) {
2797                 int err;
2798                 err = pci_enable_msi(adapter->pdev);
2799                 if (!err) {
2800                         adapter->num_rx_queues = 1;
2801                         adapter->intr.num_intrs = 1;
2802                         return;
2803                 }
2804         }
2805 #endif /* CONFIG_PCI_MSI */
2806
2807         adapter->num_rx_queues = 1;
2808         printk(KERN_INFO "Using INTx interrupt, #Rx queues: 1.\n");
2809         adapter->intr.type = VMXNET3_IT_INTX;
2810
2811         /* INT-X related setting */
2812         adapter->intr.num_intrs = 1;
2813 }
2814
2815
2816 static void
2817 vmxnet3_free_intr_resources(struct vmxnet3_adapter *adapter)
2818 {
2819         if (adapter->intr.type == VMXNET3_IT_MSIX)
2820                 pci_disable_msix(adapter->pdev);
2821         else if (adapter->intr.type == VMXNET3_IT_MSI)
2822                 pci_disable_msi(adapter->pdev);
2823         else
2824                 BUG_ON(adapter->intr.type != VMXNET3_IT_INTX);
2825 }
2826
2827
2828 static void
2829 vmxnet3_tx_timeout(struct net_device *netdev)
2830 {
2831         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2832         adapter->tx_timeout_count++;
2833
2834         printk(KERN_ERR "%s: tx hang\n", adapter->netdev->name);
2835         schedule_work(&adapter->work);
2836         netif_wake_queue(adapter->netdev);
2837 }
2838
2839
2840 static void
2841 vmxnet3_reset_work(struct work_struct *data)
2842 {
2843         struct vmxnet3_adapter *adapter;
2844
2845         adapter = container_of(data, struct vmxnet3_adapter, work);
2846
2847         /* if another thread is resetting the device, no need to proceed */
2848         if (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2849                 return;
2850
2851         /* if the device is closed, we must leave it alone */
2852         rtnl_lock();
2853         if (netif_running(adapter->netdev)) {
2854                 printk(KERN_INFO "%s: resetting\n", adapter->netdev->name);
2855                 vmxnet3_quiesce_dev(adapter);
2856                 vmxnet3_reset_dev(adapter);
2857                 vmxnet3_activate_dev(adapter);
2858         } else {
2859                 printk(KERN_INFO "%s: already closed\n", adapter->netdev->name);
2860         }
2861         rtnl_unlock();
2862
2863         clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2864 }
2865
2866
2867 static int __devinit
2868 vmxnet3_probe_device(struct pci_dev *pdev,
2869                      const struct pci_device_id *id)
2870 {
2871         static const struct net_device_ops vmxnet3_netdev_ops = {
2872                 .ndo_open = vmxnet3_open,
2873                 .ndo_stop = vmxnet3_close,
2874                 .ndo_start_xmit = vmxnet3_xmit_frame,
2875                 .ndo_set_mac_address = vmxnet3_set_mac_addr,
2876                 .ndo_change_mtu = vmxnet3_change_mtu,
2877                 .ndo_get_stats = vmxnet3_get_stats,
2878                 .ndo_tx_timeout = vmxnet3_tx_timeout,
2879                 .ndo_set_multicast_list = vmxnet3_set_mc,
2880                 .ndo_vlan_rx_register = vmxnet3_vlan_rx_register,
2881                 .ndo_vlan_rx_add_vid = vmxnet3_vlan_rx_add_vid,
2882                 .ndo_vlan_rx_kill_vid = vmxnet3_vlan_rx_kill_vid,
2883 #ifdef CONFIG_NET_POLL_CONTROLLER
2884                 .ndo_poll_controller = vmxnet3_netpoll,
2885 #endif
2886         };
2887         int err;
2888         bool dma64 = false; /* stupid gcc */
2889         u32 ver;
2890         struct net_device *netdev;
2891         struct vmxnet3_adapter *adapter;
2892         u8 mac[ETH_ALEN];
2893         int size;
2894         int num_tx_queues;
2895         int num_rx_queues;
2896
2897 #ifdef VMXNET3_RSS
2898         if (enable_mq)
2899                 num_rx_queues = min(VMXNET3_DEVICE_MAX_RX_QUEUES,
2900                                     (int)num_online_cpus());
2901         else
2902 #endif
2903                 num_rx_queues = 1;
2904
2905         if (enable_mq)
2906                 num_tx_queues = min(VMXNET3_DEVICE_MAX_TX_QUEUES,
2907                                     (int)num_online_cpus());
2908         else
2909                 num_tx_queues = 1;
2910
2911         netdev = alloc_etherdev_mq(sizeof(struct vmxnet3_adapter),
2912                                    max(num_tx_queues, num_rx_queues));
2913         printk(KERN_INFO "# of Tx queues : %d, # of Rx queues : %d\n",
2914                num_tx_queues, num_rx_queues);
2915
2916         if (!netdev) {
2917                 printk(KERN_ERR "Failed to alloc ethernet device for adapter "
2918                         "%s\n", pci_name(pdev));
2919                 return -ENOMEM;
2920         }
2921
2922         pci_set_drvdata(pdev, netdev);
2923         adapter = netdev_priv(netdev);
2924         adapter->netdev = netdev;
2925         adapter->pdev = pdev;
2926
2927         spin_lock_init(&adapter->cmd_lock);
2928         adapter->shared = pci_alloc_consistent(adapter->pdev,
2929                           sizeof(struct Vmxnet3_DriverShared),
2930                           &adapter->shared_pa);
2931         if (!adapter->shared) {
2932                 printk(KERN_ERR "Failed to allocate memory for %s\n",
2933                         pci_name(pdev));
2934                 err = -ENOMEM;
2935                 goto err_alloc_shared;
2936         }
2937
2938         adapter->num_rx_queues = num_rx_queues;
2939         adapter->num_tx_queues = num_tx_queues;
2940
2941         size = sizeof(struct Vmxnet3_TxQueueDesc) * adapter->num_tx_queues;
2942         size += sizeof(struct Vmxnet3_RxQueueDesc) * adapter->num_rx_queues;
2943         adapter->tqd_start = pci_alloc_consistent(adapter->pdev, size,
2944                              &adapter->queue_desc_pa);
2945
2946         if (!adapter->tqd_start) {
2947                 printk(KERN_ERR "Failed to allocate memory for %s\n",
2948                         pci_name(pdev));
2949                 err = -ENOMEM;
2950                 goto err_alloc_queue_desc;
2951         }
2952         adapter->rqd_start = (struct Vmxnet3_RxQueueDesc *)(adapter->tqd_start +
2953                                                         adapter->num_tx_queues);
2954
2955         adapter->pm_conf = kmalloc(sizeof(struct Vmxnet3_PMConf), GFP_KERNEL);
2956         if (adapter->pm_conf == NULL) {
2957                 printk(KERN_ERR "Failed to allocate memory for %s\n",
2958                         pci_name(pdev));
2959                 err = -ENOMEM;
2960                 goto err_alloc_pm;
2961         }
2962
2963 #ifdef VMXNET3_RSS
2964
2965         adapter->rss_conf = kmalloc(sizeof(struct UPT1_RSSConf), GFP_KERNEL);
2966         if (adapter->rss_conf == NULL) {
2967                 printk(KERN_ERR "Failed to allocate memory for %s\n",
2968                        pci_name(pdev));
2969                 err = -ENOMEM;
2970                 goto err_alloc_rss;
2971         }
2972 #endif /* VMXNET3_RSS */
2973
2974         err = vmxnet3_alloc_pci_resources(adapter, &dma64);
2975         if (err < 0)
2976                 goto err_alloc_pci;
2977
2978         ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_VRRS);
2979         if (ver & 1) {
2980                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_VRRS, 1);
2981         } else {
2982                 printk(KERN_ERR "Incompatible h/w version (0x%x) for adapter"
2983                        " %s\n", ver, pci_name(pdev));
2984                 err = -EBUSY;
2985                 goto err_ver;
2986         }
2987
2988         ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_UVRS);
2989         if (ver & 1) {
2990                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_UVRS, 1);
2991         } else {
2992                 printk(KERN_ERR "Incompatible upt version (0x%x) for "
2993                        "adapter %s\n", ver, pci_name(pdev));
2994                 err = -EBUSY;
2995                 goto err_ver;
2996         }
2997
2998         vmxnet3_declare_features(adapter, dma64);
2999
3000         adapter->dev_number = atomic_read(&devices_found);
3001
3002          adapter->share_intr = irq_share_mode;
3003         if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE &&
3004             adapter->num_tx_queues != adapter->num_rx_queues)
3005                 adapter->share_intr = VMXNET3_INTR_DONTSHARE;
3006
3007         vmxnet3_alloc_intr_resources(adapter);
3008
3009 #ifdef VMXNET3_RSS
3010         if (adapter->num_rx_queues > 1 &&
3011             adapter->intr.type == VMXNET3_IT_MSIX) {
3012                 adapter->rss = true;
3013                 printk(KERN_INFO "RSS is enabled.\n");
3014         } else {
3015                 adapter->rss = false;
3016         }
3017 #endif
3018
3019         vmxnet3_read_mac_addr(adapter, mac);
3020         memcpy(netdev->dev_addr,  mac, netdev->addr_len);
3021
3022         netdev->netdev_ops = &vmxnet3_netdev_ops;
3023         vmxnet3_set_ethtool_ops(netdev);
3024         netdev->watchdog_timeo = 5 * HZ;
3025
3026         INIT_WORK(&adapter->work, vmxnet3_reset_work);
3027
3028         if (adapter->intr.type == VMXNET3_IT_MSIX) {
3029                 int i;
3030                 for (i = 0; i < adapter->num_rx_queues; i++) {
3031                         netif_napi_add(adapter->netdev,
3032                                        &adapter->rx_queue[i].napi,
3033                                        vmxnet3_poll_rx_only, 64);
3034                 }
3035         } else {
3036                 netif_napi_add(adapter->netdev, &adapter->rx_queue[0].napi,
3037                                vmxnet3_poll, 64);
3038         }
3039
3040         netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
3041         netif_set_real_num_rx_queues(adapter->netdev, adapter->num_rx_queues);
3042
3043         SET_NETDEV_DEV(netdev, &pdev->dev);
3044         err = register_netdev(netdev);
3045
3046         if (err) {
3047                 printk(KERN_ERR "Failed to register adapter %s\n",
3048                         pci_name(pdev));
3049                 goto err_register;
3050         }
3051
3052         set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
3053         vmxnet3_check_link(adapter, false);
3054         atomic_inc(&devices_found);
3055         return 0;
3056
3057 err_register:
3058         vmxnet3_free_intr_resources(adapter);
3059 err_ver:
3060         vmxnet3_free_pci_resources(adapter);
3061 err_alloc_pci:
3062 #ifdef VMXNET3_RSS
3063         kfree(adapter->rss_conf);
3064 err_alloc_rss:
3065 #endif
3066         kfree(adapter->pm_conf);
3067 err_alloc_pm:
3068         pci_free_consistent(adapter->pdev, size, adapter->tqd_start,
3069                             adapter->queue_desc_pa);
3070 err_alloc_queue_desc:
3071         pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_DriverShared),
3072                             adapter->shared, adapter->shared_pa);
3073 err_alloc_shared:
3074         pci_set_drvdata(pdev, NULL);
3075         free_netdev(netdev);
3076         return err;
3077 }
3078
3079
3080 static void __devexit
3081 vmxnet3_remove_device(struct pci_dev *pdev)
3082 {
3083         struct net_device *netdev = pci_get_drvdata(pdev);
3084         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3085         int size = 0;
3086         int num_rx_queues;
3087
3088 #ifdef VMXNET3_RSS
3089         if (enable_mq)
3090                 num_rx_queues = min(VMXNET3_DEVICE_MAX_RX_QUEUES,
3091                                     (int)num_online_cpus());
3092         else
3093 #endif
3094                 num_rx_queues = 1;
3095
3096         cancel_work_sync(&adapter->work);
3097
3098         unregister_netdev(netdev);
3099
3100         vmxnet3_free_intr_resources(adapter);
3101         vmxnet3_free_pci_resources(adapter);
3102 #ifdef VMXNET3_RSS
3103         kfree(adapter->rss_conf);
3104 #endif
3105         kfree(adapter->pm_conf);
3106
3107         size = sizeof(struct Vmxnet3_TxQueueDesc) * adapter->num_tx_queues;
3108         size += sizeof(struct Vmxnet3_RxQueueDesc) * num_rx_queues;
3109         pci_free_consistent(adapter->pdev, size, adapter->tqd_start,
3110                             adapter->queue_desc_pa);
3111         pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_DriverShared),
3112                             adapter->shared, adapter->shared_pa);
3113         free_netdev(netdev);
3114 }
3115
3116
3117 #ifdef CONFIG_PM
3118
3119 static int
3120 vmxnet3_suspend(struct device *device)
3121 {
3122         struct pci_dev *pdev = to_pci_dev(device);
3123         struct net_device *netdev = pci_get_drvdata(pdev);
3124         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3125         struct Vmxnet3_PMConf *pmConf;
3126         struct ethhdr *ehdr;
3127         struct arphdr *ahdr;
3128         u8 *arpreq;
3129         struct in_device *in_dev;
3130         struct in_ifaddr *ifa;
3131         unsigned long flags;
3132         int i = 0;
3133
3134         if (!netif_running(netdev))
3135                 return 0;
3136
3137         for (i = 0; i < adapter->num_rx_queues; i++)
3138                 napi_disable(&adapter->rx_queue[i].napi);
3139
3140         vmxnet3_disable_all_intrs(adapter);
3141         vmxnet3_free_irqs(adapter);
3142         vmxnet3_free_intr_resources(adapter);
3143
3144         netif_device_detach(netdev);
3145         netif_tx_stop_all_queues(netdev);
3146
3147         /* Create wake-up filters. */
3148         pmConf = adapter->pm_conf;
3149         memset(pmConf, 0, sizeof(*pmConf));
3150
3151         if (adapter->wol & WAKE_UCAST) {
3152                 pmConf->filters[i].patternSize = ETH_ALEN;
3153                 pmConf->filters[i].maskSize = 1;
3154                 memcpy(pmConf->filters[i].pattern, netdev->dev_addr, ETH_ALEN);
3155                 pmConf->filters[i].mask[0] = 0x3F; /* LSB ETH_ALEN bits */
3156
3157                 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_FILTER;
3158                 i++;
3159         }
3160
3161         if (adapter->wol & WAKE_ARP) {
3162                 in_dev = in_dev_get(netdev);
3163                 if (!in_dev)
3164                         goto skip_arp;
3165
3166                 ifa = (struct in_ifaddr *)in_dev->ifa_list;
3167                 if (!ifa)
3168                         goto skip_arp;
3169
3170                 pmConf->filters[i].patternSize = ETH_HLEN + /* Ethernet header*/
3171                         sizeof(struct arphdr) +         /* ARP header */
3172                         2 * ETH_ALEN +          /* 2 Ethernet addresses*/
3173                         2 * sizeof(u32);        /*2 IPv4 addresses */
3174                 pmConf->filters[i].maskSize =
3175                         (pmConf->filters[i].patternSize - 1) / 8 + 1;
3176
3177                 /* ETH_P_ARP in Ethernet header. */
3178                 ehdr = (struct ethhdr *)pmConf->filters[i].pattern;
3179                 ehdr->h_proto = htons(ETH_P_ARP);
3180
3181                 /* ARPOP_REQUEST in ARP header. */
3182                 ahdr = (struct arphdr *)&pmConf->filters[i].pattern[ETH_HLEN];
3183                 ahdr->ar_op = htons(ARPOP_REQUEST);
3184                 arpreq = (u8 *)(ahdr + 1);
3185
3186                 /* The Unicast IPv4 address in 'tip' field. */
3187                 arpreq += 2 * ETH_ALEN + sizeof(u32);
3188                 *(u32 *)arpreq = ifa->ifa_address;
3189
3190                 /* The mask for the relevant bits. */
3191                 pmConf->filters[i].mask[0] = 0x00;
3192                 pmConf->filters[i].mask[1] = 0x30; /* ETH_P_ARP */
3193                 pmConf->filters[i].mask[2] = 0x30; /* ARPOP_REQUEST */
3194                 pmConf->filters[i].mask[3] = 0x00;
3195                 pmConf->filters[i].mask[4] = 0xC0; /* IPv4 TIP */
3196                 pmConf->filters[i].mask[5] = 0x03; /* IPv4 TIP */
3197                 in_dev_put(in_dev);
3198
3199                 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_FILTER;
3200                 i++;
3201         }
3202
3203 skip_arp:
3204         if (adapter->wol & WAKE_MAGIC)
3205                 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_MAGIC;
3206
3207         pmConf->numFilters = i;
3208
3209         adapter->shared->devRead.pmConfDesc.confVer = cpu_to_le32(1);
3210         adapter->shared->devRead.pmConfDesc.confLen = cpu_to_le32(sizeof(
3211                                                                   *pmConf));
3212         adapter->shared->devRead.pmConfDesc.confPA = cpu_to_le64(virt_to_phys(
3213                                                                  pmConf));
3214
3215         spin_lock_irqsave(&adapter->cmd_lock, flags);
3216         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3217                                VMXNET3_CMD_UPDATE_PMCFG);
3218         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3219
3220         pci_save_state(pdev);
3221         pci_enable_wake(pdev, pci_choose_state(pdev, PMSG_SUSPEND),
3222                         adapter->wol);
3223         pci_disable_device(pdev);
3224         pci_set_power_state(pdev, pci_choose_state(pdev, PMSG_SUSPEND));
3225
3226         return 0;
3227 }
3228
3229
3230 static int
3231 vmxnet3_resume(struct device *device)
3232 {
3233         int err, i = 0;
3234         unsigned long flags;
3235         struct pci_dev *pdev = to_pci_dev(device);
3236         struct net_device *netdev = pci_get_drvdata(pdev);
3237         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3238         struct Vmxnet3_PMConf *pmConf;
3239
3240         if (!netif_running(netdev))
3241                 return 0;
3242
3243         /* Destroy wake-up filters. */
3244         pmConf = adapter->pm_conf;
3245         memset(pmConf, 0, sizeof(*pmConf));
3246
3247         adapter->shared->devRead.pmConfDesc.confVer = cpu_to_le32(1);
3248         adapter->shared->devRead.pmConfDesc.confLen = cpu_to_le32(sizeof(
3249                                                                   *pmConf));
3250         adapter->shared->devRead.pmConfDesc.confPA = cpu_to_le64(virt_to_phys(
3251                                                                  pmConf));
3252
3253         netif_device_attach(netdev);
3254         pci_set_power_state(pdev, PCI_D0);
3255         pci_restore_state(pdev);
3256         err = pci_enable_device_mem(pdev);
3257         if (err != 0)
3258                 return err;
3259
3260         pci_enable_wake(pdev, PCI_D0, 0);
3261
3262         spin_lock_irqsave(&adapter->cmd_lock, flags);
3263         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3264                                VMXNET3_CMD_UPDATE_PMCFG);
3265         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3266         vmxnet3_alloc_intr_resources(adapter);
3267         vmxnet3_request_irqs(adapter);
3268         for (i = 0; i < adapter->num_rx_queues; i++)
3269                 napi_enable(&adapter->rx_queue[i].napi);
3270         vmxnet3_enable_all_intrs(adapter);
3271
3272         return 0;
3273 }
3274
3275 static const struct dev_pm_ops vmxnet3_pm_ops = {
3276         .suspend = vmxnet3_suspend,
3277         .resume = vmxnet3_resume,
3278 };
3279 #endif
3280
3281 static struct pci_driver vmxnet3_driver = {
3282         .name           = vmxnet3_driver_name,
3283         .id_table       = vmxnet3_pciid_table,
3284         .probe          = vmxnet3_probe_device,
3285         .remove         = __devexit_p(vmxnet3_remove_device),
3286 #ifdef CONFIG_PM
3287         .driver.pm      = &vmxnet3_pm_ops,
3288 #endif
3289 };
3290
3291
3292 static int __init
3293 vmxnet3_init_module(void)
3294 {
3295         printk(KERN_INFO "%s - version %s\n", VMXNET3_DRIVER_DESC,
3296                 VMXNET3_DRIVER_VERSION_REPORT);
3297         return pci_register_driver(&vmxnet3_driver);
3298 }
3299
3300 module_init(vmxnet3_init_module);
3301
3302
3303 static void
3304 vmxnet3_exit_module(void)
3305 {
3306         pci_unregister_driver(&vmxnet3_driver);
3307 }
3308
3309 module_exit(vmxnet3_exit_module);
3310
3311 MODULE_AUTHOR("VMware, Inc.");
3312 MODULE_DESCRIPTION(VMXNET3_DRIVER_DESC);
3313 MODULE_LICENSE("GPL v2");
3314 MODULE_VERSION(VMXNET3_DRIVER_VERSION_STRING);