Merge branch 'core-iommu-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
[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         unsigned long flags;
182         u32 events = le32_to_cpu(adapter->shared->ecr);
183         if (!events)
184                 return;
185
186         vmxnet3_ack_events(adapter, events);
187
188         /* Check if link state has changed */
189         if (events & VMXNET3_ECR_LINK)
190                 vmxnet3_check_link(adapter, true);
191
192         /* Check if there is an error on xmit/recv queues */
193         if (events & (VMXNET3_ECR_TQERR | VMXNET3_ECR_RQERR)) {
194                 spin_lock_irqsave(&adapter->cmd_lock, flags);
195                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
196                                        VMXNET3_CMD_GET_QUEUE_STATUS);
197                 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
198
199                 for (i = 0; i < adapter->num_tx_queues; i++)
200                         if (adapter->tqd_start[i].status.stopped)
201                                 dev_err(&adapter->netdev->dev,
202                                         "%s: tq[%d] error 0x%x\n",
203                                         adapter->netdev->name, i, le32_to_cpu(
204                                         adapter->tqd_start[i].status.error));
205                 for (i = 0; i < adapter->num_rx_queues; i++)
206                         if (adapter->rqd_start[i].status.stopped)
207                                 dev_err(&adapter->netdev->dev,
208                                         "%s: rq[%d] error 0x%x\n",
209                                         adapter->netdev->name, i,
210                                         adapter->rqd_start[i].status.error);
211
212                 schedule_work(&adapter->work);
213         }
214 }
215
216 #ifdef __BIG_ENDIAN_BITFIELD
217 /*
218  * The device expects the bitfields in shared structures to be written in
219  * little endian. When CPU is big endian, the following routines are used to
220  * correctly read and write into ABI.
221  * The general technique used here is : double word bitfields are defined in
222  * opposite order for big endian architecture. Then before reading them in
223  * driver the complete double word is translated using le32_to_cpu. Similarly
224  * After the driver writes into bitfields, cpu_to_le32 is used to translate the
225  * double words into required format.
226  * In order to avoid touching bits in shared structure more than once, temporary
227  * descriptors are used. These are passed as srcDesc to following functions.
228  */
229 static void vmxnet3_RxDescToCPU(const struct Vmxnet3_RxDesc *srcDesc,
230                                 struct Vmxnet3_RxDesc *dstDesc)
231 {
232         u32 *src = (u32 *)srcDesc + 2;
233         u32 *dst = (u32 *)dstDesc + 2;
234         dstDesc->addr = le64_to_cpu(srcDesc->addr);
235         *dst = le32_to_cpu(*src);
236         dstDesc->ext1 = le32_to_cpu(srcDesc->ext1);
237 }
238
239 static void vmxnet3_TxDescToLe(const struct Vmxnet3_TxDesc *srcDesc,
240                                struct Vmxnet3_TxDesc *dstDesc)
241 {
242         int i;
243         u32 *src = (u32 *)(srcDesc + 1);
244         u32 *dst = (u32 *)(dstDesc + 1);
245
246         /* Working backwards so that the gen bit is set at the end. */
247         for (i = 2; i > 0; i--) {
248                 src--;
249                 dst--;
250                 *dst = cpu_to_le32(*src);
251         }
252 }
253
254
255 static void vmxnet3_RxCompToCPU(const struct Vmxnet3_RxCompDesc *srcDesc,
256                                 struct Vmxnet3_RxCompDesc *dstDesc)
257 {
258         int i = 0;
259         u32 *src = (u32 *)srcDesc;
260         u32 *dst = (u32 *)dstDesc;
261         for (i = 0; i < sizeof(struct Vmxnet3_RxCompDesc) / sizeof(u32); i++) {
262                 *dst = le32_to_cpu(*src);
263                 src++;
264                 dst++;
265         }
266 }
267
268
269 /* Used to read bitfield values from double words. */
270 static u32 get_bitfield32(const __le32 *bitfield, u32 pos, u32 size)
271 {
272         u32 temp = le32_to_cpu(*bitfield);
273         u32 mask = ((1 << size) - 1) << pos;
274         temp &= mask;
275         temp >>= pos;
276         return temp;
277 }
278
279
280
281 #endif  /* __BIG_ENDIAN_BITFIELD */
282
283 #ifdef __BIG_ENDIAN_BITFIELD
284
285 #   define VMXNET3_TXDESC_GET_GEN(txdesc) get_bitfield32(((const __le32 *) \
286                         txdesc) + VMXNET3_TXD_GEN_DWORD_SHIFT, \
287                         VMXNET3_TXD_GEN_SHIFT, VMXNET3_TXD_GEN_SIZE)
288 #   define VMXNET3_TXDESC_GET_EOP(txdesc) get_bitfield32(((const __le32 *) \
289                         txdesc) + VMXNET3_TXD_EOP_DWORD_SHIFT, \
290                         VMXNET3_TXD_EOP_SHIFT, VMXNET3_TXD_EOP_SIZE)
291 #   define VMXNET3_TCD_GET_GEN(tcd) get_bitfield32(((const __le32 *)tcd) + \
292                         VMXNET3_TCD_GEN_DWORD_SHIFT, VMXNET3_TCD_GEN_SHIFT, \
293                         VMXNET3_TCD_GEN_SIZE)
294 #   define VMXNET3_TCD_GET_TXIDX(tcd) get_bitfield32((const __le32 *)tcd, \
295                         VMXNET3_TCD_TXIDX_SHIFT, VMXNET3_TCD_TXIDX_SIZE)
296 #   define vmxnet3_getRxComp(dstrcd, rcd, tmp) do { \
297                         (dstrcd) = (tmp); \
298                         vmxnet3_RxCompToCPU((rcd), (tmp)); \
299                 } while (0)
300 #   define vmxnet3_getRxDesc(dstrxd, rxd, tmp) do { \
301                         (dstrxd) = (tmp); \
302                         vmxnet3_RxDescToCPU((rxd), (tmp)); \
303                 } while (0)
304
305 #else
306
307 #   define VMXNET3_TXDESC_GET_GEN(txdesc) ((txdesc)->gen)
308 #   define VMXNET3_TXDESC_GET_EOP(txdesc) ((txdesc)->eop)
309 #   define VMXNET3_TCD_GET_GEN(tcd) ((tcd)->gen)
310 #   define VMXNET3_TCD_GET_TXIDX(tcd) ((tcd)->txdIdx)
311 #   define vmxnet3_getRxComp(dstrcd, rcd, tmp) (dstrcd) = (rcd)
312 #   define vmxnet3_getRxDesc(dstrxd, rxd, tmp) (dstrxd) = (rxd)
313
314 #endif /* __BIG_ENDIAN_BITFIELD  */
315
316
317 static void
318 vmxnet3_unmap_tx_buf(struct vmxnet3_tx_buf_info *tbi,
319                      struct pci_dev *pdev)
320 {
321         if (tbi->map_type == VMXNET3_MAP_SINGLE)
322                 pci_unmap_single(pdev, tbi->dma_addr, tbi->len,
323                                  PCI_DMA_TODEVICE);
324         else if (tbi->map_type == VMXNET3_MAP_PAGE)
325                 pci_unmap_page(pdev, tbi->dma_addr, tbi->len,
326                                PCI_DMA_TODEVICE);
327         else
328                 BUG_ON(tbi->map_type != VMXNET3_MAP_NONE);
329
330         tbi->map_type = VMXNET3_MAP_NONE; /* to help debugging */
331 }
332
333
334 static int
335 vmxnet3_unmap_pkt(u32 eop_idx, struct vmxnet3_tx_queue *tq,
336                   struct pci_dev *pdev, struct vmxnet3_adapter *adapter)
337 {
338         struct sk_buff *skb;
339         int entries = 0;
340
341         /* no out of order completion */
342         BUG_ON(tq->buf_info[eop_idx].sop_idx != tq->tx_ring.next2comp);
343         BUG_ON(VMXNET3_TXDESC_GET_EOP(&(tq->tx_ring.base[eop_idx].txd)) != 1);
344
345         skb = tq->buf_info[eop_idx].skb;
346         BUG_ON(skb == NULL);
347         tq->buf_info[eop_idx].skb = NULL;
348
349         VMXNET3_INC_RING_IDX_ONLY(eop_idx, tq->tx_ring.size);
350
351         while (tq->tx_ring.next2comp != eop_idx) {
352                 vmxnet3_unmap_tx_buf(tq->buf_info + tq->tx_ring.next2comp,
353                                      pdev);
354
355                 /* update next2comp w/o tx_lock. Since we are marking more,
356                  * instead of less, tx ring entries avail, the worst case is
357                  * that the tx routine incorrectly re-queues a pkt due to
358                  * insufficient tx ring entries.
359                  */
360                 vmxnet3_cmd_ring_adv_next2comp(&tq->tx_ring);
361                 entries++;
362         }
363
364         dev_kfree_skb_any(skb);
365         return entries;
366 }
367
368
369 static int
370 vmxnet3_tq_tx_complete(struct vmxnet3_tx_queue *tq,
371                         struct vmxnet3_adapter *adapter)
372 {
373         int completed = 0;
374         union Vmxnet3_GenericDesc *gdesc;
375
376         gdesc = tq->comp_ring.base + tq->comp_ring.next2proc;
377         while (VMXNET3_TCD_GET_GEN(&gdesc->tcd) == tq->comp_ring.gen) {
378                 completed += vmxnet3_unmap_pkt(VMXNET3_TCD_GET_TXIDX(
379                                                &gdesc->tcd), tq, adapter->pdev,
380                                                adapter);
381
382                 vmxnet3_comp_ring_adv_next2proc(&tq->comp_ring);
383                 gdesc = tq->comp_ring.base + tq->comp_ring.next2proc;
384         }
385
386         if (completed) {
387                 spin_lock(&tq->tx_lock);
388                 if (unlikely(vmxnet3_tq_stopped(tq, adapter) &&
389                              vmxnet3_cmd_ring_desc_avail(&tq->tx_ring) >
390                              VMXNET3_WAKE_QUEUE_THRESHOLD(tq) &&
391                              netif_carrier_ok(adapter->netdev))) {
392                         vmxnet3_tq_wake(tq, adapter);
393                 }
394                 spin_unlock(&tq->tx_lock);
395         }
396         return completed;
397 }
398
399
400 static void
401 vmxnet3_tq_cleanup(struct vmxnet3_tx_queue *tq,
402                    struct vmxnet3_adapter *adapter)
403 {
404         int i;
405
406         while (tq->tx_ring.next2comp != tq->tx_ring.next2fill) {
407                 struct vmxnet3_tx_buf_info *tbi;
408                 union Vmxnet3_GenericDesc *gdesc;
409
410                 tbi = tq->buf_info + tq->tx_ring.next2comp;
411                 gdesc = tq->tx_ring.base + tq->tx_ring.next2comp;
412
413                 vmxnet3_unmap_tx_buf(tbi, adapter->pdev);
414                 if (tbi->skb) {
415                         dev_kfree_skb_any(tbi->skb);
416                         tbi->skb = NULL;
417                 }
418                 vmxnet3_cmd_ring_adv_next2comp(&tq->tx_ring);
419         }
420
421         /* sanity check, verify all buffers are indeed unmapped and freed */
422         for (i = 0; i < tq->tx_ring.size; i++) {
423                 BUG_ON(tq->buf_info[i].skb != NULL ||
424                        tq->buf_info[i].map_type != VMXNET3_MAP_NONE);
425         }
426
427         tq->tx_ring.gen = VMXNET3_INIT_GEN;
428         tq->tx_ring.next2fill = tq->tx_ring.next2comp = 0;
429
430         tq->comp_ring.gen = VMXNET3_INIT_GEN;
431         tq->comp_ring.next2proc = 0;
432 }
433
434
435 static void
436 vmxnet3_tq_destroy(struct vmxnet3_tx_queue *tq,
437                    struct vmxnet3_adapter *adapter)
438 {
439         if (tq->tx_ring.base) {
440                 pci_free_consistent(adapter->pdev, tq->tx_ring.size *
441                                     sizeof(struct Vmxnet3_TxDesc),
442                                     tq->tx_ring.base, tq->tx_ring.basePA);
443                 tq->tx_ring.base = NULL;
444         }
445         if (tq->data_ring.base) {
446                 pci_free_consistent(adapter->pdev, tq->data_ring.size *
447                                     sizeof(struct Vmxnet3_TxDataDesc),
448                                     tq->data_ring.base, tq->data_ring.basePA);
449                 tq->data_ring.base = NULL;
450         }
451         if (tq->comp_ring.base) {
452                 pci_free_consistent(adapter->pdev, tq->comp_ring.size *
453                                     sizeof(struct Vmxnet3_TxCompDesc),
454                                     tq->comp_ring.base, tq->comp_ring.basePA);
455                 tq->comp_ring.base = NULL;
456         }
457         kfree(tq->buf_info);
458         tq->buf_info = NULL;
459 }
460
461
462 /* Destroy all tx queues */
463 void
464 vmxnet3_tq_destroy_all(struct vmxnet3_adapter *adapter)
465 {
466         int i;
467
468         for (i = 0; i < adapter->num_tx_queues; i++)
469                 vmxnet3_tq_destroy(&adapter->tx_queue[i], adapter);
470 }
471
472
473 static void
474 vmxnet3_tq_init(struct vmxnet3_tx_queue *tq,
475                 struct vmxnet3_adapter *adapter)
476 {
477         int i;
478
479         /* reset the tx ring contents to 0 and reset the tx ring states */
480         memset(tq->tx_ring.base, 0, tq->tx_ring.size *
481                sizeof(struct Vmxnet3_TxDesc));
482         tq->tx_ring.next2fill = tq->tx_ring.next2comp = 0;
483         tq->tx_ring.gen = VMXNET3_INIT_GEN;
484
485         memset(tq->data_ring.base, 0, tq->data_ring.size *
486                sizeof(struct Vmxnet3_TxDataDesc));
487
488         /* reset the tx comp ring contents to 0 and reset comp ring states */
489         memset(tq->comp_ring.base, 0, tq->comp_ring.size *
490                sizeof(struct Vmxnet3_TxCompDesc));
491         tq->comp_ring.next2proc = 0;
492         tq->comp_ring.gen = VMXNET3_INIT_GEN;
493
494         /* reset the bookkeeping data */
495         memset(tq->buf_info, 0, sizeof(tq->buf_info[0]) * tq->tx_ring.size);
496         for (i = 0; i < tq->tx_ring.size; i++)
497                 tq->buf_info[i].map_type = VMXNET3_MAP_NONE;
498
499         /* stats are not reset */
500 }
501
502
503 static int
504 vmxnet3_tq_create(struct vmxnet3_tx_queue *tq,
505                   struct vmxnet3_adapter *adapter)
506 {
507         BUG_ON(tq->tx_ring.base || tq->data_ring.base ||
508                tq->comp_ring.base || tq->buf_info);
509
510         tq->tx_ring.base = pci_alloc_consistent(adapter->pdev, tq->tx_ring.size
511                            * sizeof(struct Vmxnet3_TxDesc),
512                            &tq->tx_ring.basePA);
513         if (!tq->tx_ring.base) {
514                 printk(KERN_ERR "%s: failed to allocate tx ring\n",
515                        adapter->netdev->name);
516                 goto err;
517         }
518
519         tq->data_ring.base = pci_alloc_consistent(adapter->pdev,
520                              tq->data_ring.size *
521                              sizeof(struct Vmxnet3_TxDataDesc),
522                              &tq->data_ring.basePA);
523         if (!tq->data_ring.base) {
524                 printk(KERN_ERR "%s: failed to allocate data ring\n",
525                        adapter->netdev->name);
526                 goto err;
527         }
528
529         tq->comp_ring.base = pci_alloc_consistent(adapter->pdev,
530                              tq->comp_ring.size *
531                              sizeof(struct Vmxnet3_TxCompDesc),
532                              &tq->comp_ring.basePA);
533         if (!tq->comp_ring.base) {
534                 printk(KERN_ERR "%s: failed to allocate tx comp ring\n",
535                        adapter->netdev->name);
536                 goto err;
537         }
538
539         tq->buf_info = kcalloc(tq->tx_ring.size, sizeof(tq->buf_info[0]),
540                                GFP_KERNEL);
541         if (!tq->buf_info) {
542                 printk(KERN_ERR "%s: failed to allocate tx bufinfo\n",
543                        adapter->netdev->name);
544                 goto err;
545         }
546
547         return 0;
548
549 err:
550         vmxnet3_tq_destroy(tq, adapter);
551         return -ENOMEM;
552 }
553
554 static void
555 vmxnet3_tq_cleanup_all(struct vmxnet3_adapter *adapter)
556 {
557         int i;
558
559         for (i = 0; i < adapter->num_tx_queues; i++)
560                 vmxnet3_tq_cleanup(&adapter->tx_queue[i], adapter);
561 }
562
563 /*
564  *    starting from ring->next2fill, allocate rx buffers for the given ring
565  *    of the rx queue and update the rx desc. stop after @num_to_alloc buffers
566  *    are allocated or allocation fails
567  */
568
569 static int
570 vmxnet3_rq_alloc_rx_buf(struct vmxnet3_rx_queue *rq, u32 ring_idx,
571                         int num_to_alloc, struct vmxnet3_adapter *adapter)
572 {
573         int num_allocated = 0;
574         struct vmxnet3_rx_buf_info *rbi_base = rq->buf_info[ring_idx];
575         struct vmxnet3_cmd_ring *ring = &rq->rx_ring[ring_idx];
576         u32 val;
577
578         while (num_allocated < num_to_alloc) {
579                 struct vmxnet3_rx_buf_info *rbi;
580                 union Vmxnet3_GenericDesc *gd;
581
582                 rbi = rbi_base + ring->next2fill;
583                 gd = ring->base + ring->next2fill;
584
585                 if (rbi->buf_type == VMXNET3_RX_BUF_SKB) {
586                         if (rbi->skb == NULL) {
587                                 rbi->skb = dev_alloc_skb(rbi->len +
588                                                          NET_IP_ALIGN);
589                                 if (unlikely(rbi->skb == NULL)) {
590                                         rq->stats.rx_buf_alloc_failure++;
591                                         break;
592                                 }
593                                 rbi->skb->dev = adapter->netdev;
594
595                                 skb_reserve(rbi->skb, NET_IP_ALIGN);
596                                 rbi->dma_addr = pci_map_single(adapter->pdev,
597                                                 rbi->skb->data, rbi->len,
598                                                 PCI_DMA_FROMDEVICE);
599                         } else {
600                                 /* rx buffer skipped by the device */
601                         }
602                         val = VMXNET3_RXD_BTYPE_HEAD << VMXNET3_RXD_BTYPE_SHIFT;
603                 } else {
604                         BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_PAGE ||
605                                rbi->len  != PAGE_SIZE);
606
607                         if (rbi->page == NULL) {
608                                 rbi->page = alloc_page(GFP_ATOMIC);
609                                 if (unlikely(rbi->page == NULL)) {
610                                         rq->stats.rx_buf_alloc_failure++;
611                                         break;
612                                 }
613                                 rbi->dma_addr = pci_map_page(adapter->pdev,
614                                                 rbi->page, 0, PAGE_SIZE,
615                                                 PCI_DMA_FROMDEVICE);
616                         } else {
617                                 /* rx buffers skipped by the device */
618                         }
619                         val = VMXNET3_RXD_BTYPE_BODY << VMXNET3_RXD_BTYPE_SHIFT;
620                 }
621
622                 BUG_ON(rbi->dma_addr == 0);
623                 gd->rxd.addr = cpu_to_le64(rbi->dma_addr);
624                 gd->dword[2] = cpu_to_le32((ring->gen << VMXNET3_RXD_GEN_SHIFT)
625                                            | val | rbi->len);
626
627                 num_allocated++;
628                 vmxnet3_cmd_ring_adv_next2fill(ring);
629         }
630         rq->uncommitted[ring_idx] += num_allocated;
631
632         dev_dbg(&adapter->netdev->dev,
633                 "alloc_rx_buf: %d allocated, next2fill %u, next2comp "
634                 "%u, uncommited %u\n", num_allocated, ring->next2fill,
635                 ring->next2comp, rq->uncommitted[ring_idx]);
636
637         /* so that the device can distinguish a full ring and an empty ring */
638         BUG_ON(num_allocated != 0 && ring->next2fill == ring->next2comp);
639
640         return num_allocated;
641 }
642
643
644 static void
645 vmxnet3_append_frag(struct sk_buff *skb, struct Vmxnet3_RxCompDesc *rcd,
646                     struct vmxnet3_rx_buf_info *rbi)
647 {
648         struct skb_frag_struct *frag = skb_shinfo(skb)->frags +
649                 skb_shinfo(skb)->nr_frags;
650
651         BUG_ON(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS);
652
653         frag->page = rbi->page;
654         frag->page_offset = 0;
655         frag->size = rcd->len;
656         skb->data_len += frag->size;
657         skb_shinfo(skb)->nr_frags++;
658 }
659
660
661 static void
662 vmxnet3_map_pkt(struct sk_buff *skb, struct vmxnet3_tx_ctx *ctx,
663                 struct vmxnet3_tx_queue *tq, struct pci_dev *pdev,
664                 struct vmxnet3_adapter *adapter)
665 {
666         u32 dw2, len;
667         unsigned long buf_offset;
668         int i;
669         union Vmxnet3_GenericDesc *gdesc;
670         struct vmxnet3_tx_buf_info *tbi = NULL;
671
672         BUG_ON(ctx->copy_size > skb_headlen(skb));
673
674         /* use the previous gen bit for the SOP desc */
675         dw2 = (tq->tx_ring.gen ^ 0x1) << VMXNET3_TXD_GEN_SHIFT;
676
677         ctx->sop_txd = tq->tx_ring.base + tq->tx_ring.next2fill;
678         gdesc = ctx->sop_txd; /* both loops below can be skipped */
679
680         /* no need to map the buffer if headers are copied */
681         if (ctx->copy_size) {
682                 ctx->sop_txd->txd.addr = cpu_to_le64(tq->data_ring.basePA +
683                                         tq->tx_ring.next2fill *
684                                         sizeof(struct Vmxnet3_TxDataDesc));
685                 ctx->sop_txd->dword[2] = cpu_to_le32(dw2 | ctx->copy_size);
686                 ctx->sop_txd->dword[3] = 0;
687
688                 tbi = tq->buf_info + tq->tx_ring.next2fill;
689                 tbi->map_type = VMXNET3_MAP_NONE;
690
691                 dev_dbg(&adapter->netdev->dev,
692                         "txd[%u]: 0x%Lx 0x%x 0x%x\n",
693                         tq->tx_ring.next2fill,
694                         le64_to_cpu(ctx->sop_txd->txd.addr),
695                         ctx->sop_txd->dword[2], ctx->sop_txd->dword[3]);
696                 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
697
698                 /* use the right gen for non-SOP desc */
699                 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
700         }
701
702         /* linear part can use multiple tx desc if it's big */
703         len = skb_headlen(skb) - ctx->copy_size;
704         buf_offset = ctx->copy_size;
705         while (len) {
706                 u32 buf_size;
707
708                 if (len < VMXNET3_MAX_TX_BUF_SIZE) {
709                         buf_size = len;
710                         dw2 |= len;
711                 } else {
712                         buf_size = VMXNET3_MAX_TX_BUF_SIZE;
713                         /* spec says that for TxDesc.len, 0 == 2^14 */
714                 }
715
716                 tbi = tq->buf_info + tq->tx_ring.next2fill;
717                 tbi->map_type = VMXNET3_MAP_SINGLE;
718                 tbi->dma_addr = pci_map_single(adapter->pdev,
719                                 skb->data + buf_offset, buf_size,
720                                 PCI_DMA_TODEVICE);
721
722                 tbi->len = buf_size;
723
724                 gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
725                 BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
726
727                 gdesc->txd.addr = cpu_to_le64(tbi->dma_addr);
728                 gdesc->dword[2] = cpu_to_le32(dw2);
729                 gdesc->dword[3] = 0;
730
731                 dev_dbg(&adapter->netdev->dev,
732                         "txd[%u]: 0x%Lx 0x%x 0x%x\n",
733                         tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr),
734                         le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]);
735                 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
736                 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
737
738                 len -= buf_size;
739                 buf_offset += buf_size;
740         }
741
742         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
743                 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
744
745                 tbi = tq->buf_info + tq->tx_ring.next2fill;
746                 tbi->map_type = VMXNET3_MAP_PAGE;
747                 tbi->dma_addr = pci_map_page(adapter->pdev, frag->page,
748                                              frag->page_offset, frag->size,
749                                              PCI_DMA_TODEVICE);
750
751                 tbi->len = frag->size;
752
753                 gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
754                 BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
755
756                 gdesc->txd.addr = cpu_to_le64(tbi->dma_addr);
757                 gdesc->dword[2] = cpu_to_le32(dw2 | frag->size);
758                 gdesc->dword[3] = 0;
759
760                 dev_dbg(&adapter->netdev->dev,
761                         "txd[%u]: 0x%llu %u %u\n",
762                         tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr),
763                         le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]);
764                 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
765                 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
766         }
767
768         ctx->eop_txd = gdesc;
769
770         /* set the last buf_info for the pkt */
771         tbi->skb = skb;
772         tbi->sop_idx = ctx->sop_txd - tq->tx_ring.base;
773 }
774
775
776 /* Init all tx queues */
777 static void
778 vmxnet3_tq_init_all(struct vmxnet3_adapter *adapter)
779 {
780         int i;
781
782         for (i = 0; i < adapter->num_tx_queues; i++)
783                 vmxnet3_tq_init(&adapter->tx_queue[i], adapter);
784 }
785
786
787 /*
788  *    parse and copy relevant protocol headers:
789  *      For a tso pkt, relevant headers are L2/3/4 including options
790  *      For a pkt requesting csum offloading, they are L2/3 and may include L4
791  *      if it's a TCP/UDP pkt
792  *
793  * Returns:
794  *    -1:  error happens during parsing
795  *     0:  protocol headers parsed, but too big to be copied
796  *     1:  protocol headers parsed and copied
797  *
798  * Other effects:
799  *    1. related *ctx fields are updated.
800  *    2. ctx->copy_size is # of bytes copied
801  *    3. the portion copied is guaranteed to be in the linear part
802  *
803  */
804 static int
805 vmxnet3_parse_and_copy_hdr(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
806                            struct vmxnet3_tx_ctx *ctx,
807                            struct vmxnet3_adapter *adapter)
808 {
809         struct Vmxnet3_TxDataDesc *tdd;
810
811         if (ctx->mss) { /* TSO */
812                 ctx->eth_ip_hdr_size = skb_transport_offset(skb);
813                 ctx->l4_hdr_size = ((struct tcphdr *)
814                                    skb_transport_header(skb))->doff * 4;
815                 ctx->copy_size = ctx->eth_ip_hdr_size + ctx->l4_hdr_size;
816         } else {
817                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
818                         ctx->eth_ip_hdr_size = skb_checksum_start_offset(skb);
819
820                         if (ctx->ipv4) {
821                                 struct iphdr *iph = (struct iphdr *)
822                                                     skb_network_header(skb);
823                                 if (iph->protocol == IPPROTO_TCP)
824                                         ctx->l4_hdr_size = ((struct tcphdr *)
825                                            skb_transport_header(skb))->doff * 4;
826                                 else if (iph->protocol == IPPROTO_UDP)
827                                         /*
828                                          * Use tcp header size so that bytes to
829                                          * be copied are more than required by
830                                          * the device.
831                                          */
832                                         ctx->l4_hdr_size =
833                                                         sizeof(struct tcphdr);
834                                 else
835                                         ctx->l4_hdr_size = 0;
836                         } else {
837                                 /* for simplicity, don't copy L4 headers */
838                                 ctx->l4_hdr_size = 0;
839                         }
840                         ctx->copy_size = ctx->eth_ip_hdr_size +
841                                          ctx->l4_hdr_size;
842                 } else {
843                         ctx->eth_ip_hdr_size = 0;
844                         ctx->l4_hdr_size = 0;
845                         /* copy as much as allowed */
846                         ctx->copy_size = min((unsigned int)VMXNET3_HDR_COPY_SIZE
847                                              , skb_headlen(skb));
848                 }
849
850                 /* make sure headers are accessible directly */
851                 if (unlikely(!pskb_may_pull(skb, ctx->copy_size)))
852                         goto err;
853         }
854
855         if (unlikely(ctx->copy_size > VMXNET3_HDR_COPY_SIZE)) {
856                 tq->stats.oversized_hdr++;
857                 ctx->copy_size = 0;
858                 return 0;
859         }
860
861         tdd = tq->data_ring.base + tq->tx_ring.next2fill;
862
863         memcpy(tdd->data, skb->data, ctx->copy_size);
864         dev_dbg(&adapter->netdev->dev,
865                 "copy %u bytes to dataRing[%u]\n",
866                 ctx->copy_size, tq->tx_ring.next2fill);
867         return 1;
868
869 err:
870         return -1;
871 }
872
873
874 static void
875 vmxnet3_prepare_tso(struct sk_buff *skb,
876                     struct vmxnet3_tx_ctx *ctx)
877 {
878         struct tcphdr *tcph = (struct tcphdr *)skb_transport_header(skb);
879         if (ctx->ipv4) {
880                 struct iphdr *iph = (struct iphdr *)skb_network_header(skb);
881                 iph->check = 0;
882                 tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, 0,
883                                                  IPPROTO_TCP, 0);
884         } else {
885                 struct ipv6hdr *iph = (struct ipv6hdr *)skb_network_header(skb);
886                 tcph->check = ~csum_ipv6_magic(&iph->saddr, &iph->daddr, 0,
887                                                IPPROTO_TCP, 0);
888         }
889 }
890
891
892 /*
893  * Transmits a pkt thru a given tq
894  * Returns:
895  *    NETDEV_TX_OK:      descriptors are setup successfully
896  *    NETDEV_TX_OK:      error occurred, the pkt is dropped
897  *    NETDEV_TX_BUSY:    tx ring is full, queue is stopped
898  *
899  * Side-effects:
900  *    1. tx ring may be changed
901  *    2. tq stats may be updated accordingly
902  *    3. shared->txNumDeferred may be updated
903  */
904
905 static int
906 vmxnet3_tq_xmit(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
907                 struct vmxnet3_adapter *adapter, struct net_device *netdev)
908 {
909         int ret;
910         u32 count;
911         unsigned long flags;
912         struct vmxnet3_tx_ctx ctx;
913         union Vmxnet3_GenericDesc *gdesc;
914 #ifdef __BIG_ENDIAN_BITFIELD
915         /* Use temporary descriptor to avoid touching bits multiple times */
916         union Vmxnet3_GenericDesc tempTxDesc;
917 #endif
918
919         /* conservatively estimate # of descriptors to use */
920         count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) +
921                 skb_shinfo(skb)->nr_frags + 1;
922
923         ctx.ipv4 = (skb->protocol == cpu_to_be16(ETH_P_IP));
924
925         ctx.mss = skb_shinfo(skb)->gso_size;
926         if (ctx.mss) {
927                 if (skb_header_cloned(skb)) {
928                         if (unlikely(pskb_expand_head(skb, 0, 0,
929                                                       GFP_ATOMIC) != 0)) {
930                                 tq->stats.drop_tso++;
931                                 goto drop_pkt;
932                         }
933                         tq->stats.copy_skb_header++;
934                 }
935                 vmxnet3_prepare_tso(skb, &ctx);
936         } else {
937                 if (unlikely(count > VMXNET3_MAX_TXD_PER_PKT)) {
938
939                         /* non-tso pkts must not use more than
940                          * VMXNET3_MAX_TXD_PER_PKT entries
941                          */
942                         if (skb_linearize(skb) != 0) {
943                                 tq->stats.drop_too_many_frags++;
944                                 goto drop_pkt;
945                         }
946                         tq->stats.linearized++;
947
948                         /* recalculate the # of descriptors to use */
949                         count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) + 1;
950                 }
951         }
952
953         spin_lock_irqsave(&tq->tx_lock, flags);
954
955         if (count > vmxnet3_cmd_ring_desc_avail(&tq->tx_ring)) {
956                 tq->stats.tx_ring_full++;
957                 dev_dbg(&adapter->netdev->dev,
958                         "tx queue stopped on %s, next2comp %u"
959                         " next2fill %u\n", adapter->netdev->name,
960                         tq->tx_ring.next2comp, tq->tx_ring.next2fill);
961
962                 vmxnet3_tq_stop(tq, adapter);
963                 spin_unlock_irqrestore(&tq->tx_lock, flags);
964                 return NETDEV_TX_BUSY;
965         }
966
967
968         ret = vmxnet3_parse_and_copy_hdr(skb, tq, &ctx, adapter);
969         if (ret >= 0) {
970                 BUG_ON(ret <= 0 && ctx.copy_size != 0);
971                 /* hdrs parsed, check against other limits */
972                 if (ctx.mss) {
973                         if (unlikely(ctx.eth_ip_hdr_size + ctx.l4_hdr_size >
974                                      VMXNET3_MAX_TX_BUF_SIZE)) {
975                                 goto hdr_too_big;
976                         }
977                 } else {
978                         if (skb->ip_summed == CHECKSUM_PARTIAL) {
979                                 if (unlikely(ctx.eth_ip_hdr_size +
980                                              skb->csum_offset >
981                                              VMXNET3_MAX_CSUM_OFFSET)) {
982                                         goto hdr_too_big;
983                                 }
984                         }
985                 }
986         } else {
987                 tq->stats.drop_hdr_inspect_err++;
988                 goto unlock_drop_pkt;
989         }
990
991         /* fill tx descs related to addr & len */
992         vmxnet3_map_pkt(skb, &ctx, tq, adapter->pdev, adapter);
993
994         /* setup the EOP desc */
995         ctx.eop_txd->dword[3] = cpu_to_le32(VMXNET3_TXD_CQ | VMXNET3_TXD_EOP);
996
997         /* setup the SOP desc */
998 #ifdef __BIG_ENDIAN_BITFIELD
999         gdesc = &tempTxDesc;
1000         gdesc->dword[2] = ctx.sop_txd->dword[2];
1001         gdesc->dword[3] = ctx.sop_txd->dword[3];
1002 #else
1003         gdesc = ctx.sop_txd;
1004 #endif
1005         if (ctx.mss) {
1006                 gdesc->txd.hlen = ctx.eth_ip_hdr_size + ctx.l4_hdr_size;
1007                 gdesc->txd.om = VMXNET3_OM_TSO;
1008                 gdesc->txd.msscof = ctx.mss;
1009                 le32_add_cpu(&tq->shared->txNumDeferred, (skb->len -
1010                              gdesc->txd.hlen + ctx.mss - 1) / ctx.mss);
1011         } else {
1012                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1013                         gdesc->txd.hlen = ctx.eth_ip_hdr_size;
1014                         gdesc->txd.om = VMXNET3_OM_CSUM;
1015                         gdesc->txd.msscof = ctx.eth_ip_hdr_size +
1016                                             skb->csum_offset;
1017                 } else {
1018                         gdesc->txd.om = 0;
1019                         gdesc->txd.msscof = 0;
1020                 }
1021                 le32_add_cpu(&tq->shared->txNumDeferred, 1);
1022         }
1023
1024         if (vlan_tx_tag_present(skb)) {
1025                 gdesc->txd.ti = 1;
1026                 gdesc->txd.tci = vlan_tx_tag_get(skb);
1027         }
1028
1029         /* finally flips the GEN bit of the SOP desc. */
1030         gdesc->dword[2] = cpu_to_le32(le32_to_cpu(gdesc->dword[2]) ^
1031                                                   VMXNET3_TXD_GEN);
1032 #ifdef __BIG_ENDIAN_BITFIELD
1033         /* Finished updating in bitfields of Tx Desc, so write them in original
1034          * place.
1035          */
1036         vmxnet3_TxDescToLe((struct Vmxnet3_TxDesc *)gdesc,
1037                            (struct Vmxnet3_TxDesc *)ctx.sop_txd);
1038         gdesc = ctx.sop_txd;
1039 #endif
1040         dev_dbg(&adapter->netdev->dev,
1041                 "txd[%u]: SOP 0x%Lx 0x%x 0x%x\n",
1042                 (u32)((union Vmxnet3_GenericDesc *)ctx.sop_txd -
1043                 tq->tx_ring.base), le64_to_cpu(gdesc->txd.addr),
1044                 le32_to_cpu(gdesc->dword[2]), le32_to_cpu(gdesc->dword[3]));
1045
1046         spin_unlock_irqrestore(&tq->tx_lock, flags);
1047
1048         if (le32_to_cpu(tq->shared->txNumDeferred) >=
1049                                         le32_to_cpu(tq->shared->txThreshold)) {
1050                 tq->shared->txNumDeferred = 0;
1051                 VMXNET3_WRITE_BAR0_REG(adapter,
1052                                        VMXNET3_REG_TXPROD + tq->qid * 8,
1053                                        tq->tx_ring.next2fill);
1054         }
1055
1056         return NETDEV_TX_OK;
1057
1058 hdr_too_big:
1059         tq->stats.drop_oversized_hdr++;
1060 unlock_drop_pkt:
1061         spin_unlock_irqrestore(&tq->tx_lock, flags);
1062 drop_pkt:
1063         tq->stats.drop_total++;
1064         dev_kfree_skb(skb);
1065         return NETDEV_TX_OK;
1066 }
1067
1068
1069 static netdev_tx_t
1070 vmxnet3_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
1071 {
1072         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1073
1074                 BUG_ON(skb->queue_mapping > adapter->num_tx_queues);
1075                 return vmxnet3_tq_xmit(skb,
1076                                        &adapter->tx_queue[skb->queue_mapping],
1077                                        adapter, netdev);
1078 }
1079
1080
1081 static void
1082 vmxnet3_rx_csum(struct vmxnet3_adapter *adapter,
1083                 struct sk_buff *skb,
1084                 union Vmxnet3_GenericDesc *gdesc)
1085 {
1086         if (!gdesc->rcd.cnc && adapter->rxcsum) {
1087                 /* typical case: TCP/UDP over IP and both csums are correct */
1088                 if ((le32_to_cpu(gdesc->dword[3]) & VMXNET3_RCD_CSUM_OK) ==
1089                                                         VMXNET3_RCD_CSUM_OK) {
1090                         skb->ip_summed = CHECKSUM_UNNECESSARY;
1091                         BUG_ON(!(gdesc->rcd.tcp || gdesc->rcd.udp));
1092                         BUG_ON(!(gdesc->rcd.v4  || gdesc->rcd.v6));
1093                         BUG_ON(gdesc->rcd.frg);
1094                 } else {
1095                         if (gdesc->rcd.csum) {
1096                                 skb->csum = htons(gdesc->rcd.csum);
1097                                 skb->ip_summed = CHECKSUM_PARTIAL;
1098                         } else {
1099                                 skb_checksum_none_assert(skb);
1100                         }
1101                 }
1102         } else {
1103                 skb_checksum_none_assert(skb);
1104         }
1105 }
1106
1107
1108 static void
1109 vmxnet3_rx_error(struct vmxnet3_rx_queue *rq, struct Vmxnet3_RxCompDesc *rcd,
1110                  struct vmxnet3_rx_ctx *ctx,  struct vmxnet3_adapter *adapter)
1111 {
1112         rq->stats.drop_err++;
1113         if (!rcd->fcs)
1114                 rq->stats.drop_fcs++;
1115
1116         rq->stats.drop_total++;
1117
1118         /*
1119          * We do not unmap and chain the rx buffer to the skb.
1120          * We basically pretend this buffer is not used and will be recycled
1121          * by vmxnet3_rq_alloc_rx_buf()
1122          */
1123
1124         /*
1125          * ctx->skb may be NULL if this is the first and the only one
1126          * desc for the pkt
1127          */
1128         if (ctx->skb)
1129                 dev_kfree_skb_irq(ctx->skb);
1130
1131         ctx->skb = NULL;
1132 }
1133
1134
1135 static int
1136 vmxnet3_rq_rx_complete(struct vmxnet3_rx_queue *rq,
1137                        struct vmxnet3_adapter *adapter, int quota)
1138 {
1139         static const u32 rxprod_reg[2] = {
1140                 VMXNET3_REG_RXPROD, VMXNET3_REG_RXPROD2
1141         };
1142         u32 num_rxd = 0;
1143         struct Vmxnet3_RxCompDesc *rcd;
1144         struct vmxnet3_rx_ctx *ctx = &rq->rx_ctx;
1145 #ifdef __BIG_ENDIAN_BITFIELD
1146         struct Vmxnet3_RxDesc rxCmdDesc;
1147         struct Vmxnet3_RxCompDesc rxComp;
1148 #endif
1149         vmxnet3_getRxComp(rcd, &rq->comp_ring.base[rq->comp_ring.next2proc].rcd,
1150                           &rxComp);
1151         while (rcd->gen == rq->comp_ring.gen) {
1152                 struct vmxnet3_rx_buf_info *rbi;
1153                 struct sk_buff *skb;
1154                 int num_to_alloc;
1155                 struct Vmxnet3_RxDesc *rxd;
1156                 u32 idx, ring_idx;
1157
1158                 if (num_rxd >= quota) {
1159                         /* we may stop even before we see the EOP desc of
1160                          * the current pkt
1161                          */
1162                         break;
1163                 }
1164                 num_rxd++;
1165                 BUG_ON(rcd->rqID != rq->qid && rcd->rqID != rq->qid2);
1166                 idx = rcd->rxdIdx;
1167                 ring_idx = rcd->rqID < adapter->num_rx_queues ? 0 : 1;
1168                 vmxnet3_getRxDesc(rxd, &rq->rx_ring[ring_idx].base[idx].rxd,
1169                                   &rxCmdDesc);
1170                 rbi = rq->buf_info[ring_idx] + idx;
1171
1172                 BUG_ON(rxd->addr != rbi->dma_addr ||
1173                        rxd->len != rbi->len);
1174
1175                 if (unlikely(rcd->eop && rcd->err)) {
1176                         vmxnet3_rx_error(rq, rcd, ctx, adapter);
1177                         goto rcd_done;
1178                 }
1179
1180                 if (rcd->sop) { /* first buf of the pkt */
1181                         BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_HEAD ||
1182                                rcd->rqID != rq->qid);
1183
1184                         BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_SKB);
1185                         BUG_ON(ctx->skb != NULL || rbi->skb == NULL);
1186
1187                         if (unlikely(rcd->len == 0)) {
1188                                 /* Pretend the rx buffer is skipped. */
1189                                 BUG_ON(!(rcd->sop && rcd->eop));
1190                                 dev_dbg(&adapter->netdev->dev,
1191                                         "rxRing[%u][%u] 0 length\n",
1192                                         ring_idx, idx);
1193                                 goto rcd_done;
1194                         }
1195
1196                         ctx->skb = rbi->skb;
1197                         rbi->skb = NULL;
1198
1199                         pci_unmap_single(adapter->pdev, rbi->dma_addr, rbi->len,
1200                                          PCI_DMA_FROMDEVICE);
1201
1202                         skb_put(ctx->skb, rcd->len);
1203                 } else {
1204                         BUG_ON(ctx->skb == NULL);
1205                         /* non SOP buffer must be type 1 in most cases */
1206                         if (rbi->buf_type == VMXNET3_RX_BUF_PAGE) {
1207                                 BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_BODY);
1208
1209                                 if (rcd->len) {
1210                                         pci_unmap_page(adapter->pdev,
1211                                                        rbi->dma_addr, rbi->len,
1212                                                        PCI_DMA_FROMDEVICE);
1213
1214                                         vmxnet3_append_frag(ctx->skb, rcd, rbi);
1215                                         rbi->page = NULL;
1216                                 }
1217                         } else {
1218                                 /*
1219                                  * The only time a non-SOP buffer is type 0 is
1220                                  * when it's EOP and error flag is raised, which
1221                                  * has already been handled.
1222                                  */
1223                                 BUG_ON(true);
1224                         }
1225                 }
1226
1227                 skb = ctx->skb;
1228                 if (rcd->eop) {
1229                         skb->len += skb->data_len;
1230                         skb->truesize += skb->data_len;
1231
1232                         vmxnet3_rx_csum(adapter, skb,
1233                                         (union Vmxnet3_GenericDesc *)rcd);
1234                         skb->protocol = eth_type_trans(skb, adapter->netdev);
1235
1236                         if (unlikely(adapter->vlan_grp && rcd->ts)) {
1237                                 vlan_hwaccel_receive_skb(skb,
1238                                                 adapter->vlan_grp, rcd->tci);
1239                         } else {
1240                                 netif_receive_skb(skb);
1241                         }
1242
1243                         ctx->skb = NULL;
1244                 }
1245
1246 rcd_done:
1247                 /* device may skip some rx descs */
1248                 rq->rx_ring[ring_idx].next2comp = idx;
1249                 VMXNET3_INC_RING_IDX_ONLY(rq->rx_ring[ring_idx].next2comp,
1250                                           rq->rx_ring[ring_idx].size);
1251
1252                 /* refill rx buffers frequently to avoid starving the h/w */
1253                 num_to_alloc = vmxnet3_cmd_ring_desc_avail(rq->rx_ring +
1254                                                            ring_idx);
1255                 if (unlikely(num_to_alloc > VMXNET3_RX_ALLOC_THRESHOLD(rq,
1256                                                         ring_idx, adapter))) {
1257                         vmxnet3_rq_alloc_rx_buf(rq, ring_idx, num_to_alloc,
1258                                                 adapter);
1259
1260                         /* if needed, update the register */
1261                         if (unlikely(rq->shared->updateRxProd)) {
1262                                 VMXNET3_WRITE_BAR0_REG(adapter,
1263                                         rxprod_reg[ring_idx] + rq->qid * 8,
1264                                         rq->rx_ring[ring_idx].next2fill);
1265                                 rq->uncommitted[ring_idx] = 0;
1266                         }
1267                 }
1268
1269                 vmxnet3_comp_ring_adv_next2proc(&rq->comp_ring);
1270                 vmxnet3_getRxComp(rcd,
1271                      &rq->comp_ring.base[rq->comp_ring.next2proc].rcd, &rxComp);
1272         }
1273
1274         return num_rxd;
1275 }
1276
1277
1278 static void
1279 vmxnet3_rq_cleanup(struct vmxnet3_rx_queue *rq,
1280                    struct vmxnet3_adapter *adapter)
1281 {
1282         u32 i, ring_idx;
1283         struct Vmxnet3_RxDesc *rxd;
1284
1285         for (ring_idx = 0; ring_idx < 2; ring_idx++) {
1286                 for (i = 0; i < rq->rx_ring[ring_idx].size; i++) {
1287 #ifdef __BIG_ENDIAN_BITFIELD
1288                         struct Vmxnet3_RxDesc rxDesc;
1289 #endif
1290                         vmxnet3_getRxDesc(rxd,
1291                                 &rq->rx_ring[ring_idx].base[i].rxd, &rxDesc);
1292
1293                         if (rxd->btype == VMXNET3_RXD_BTYPE_HEAD &&
1294                                         rq->buf_info[ring_idx][i].skb) {
1295                                 pci_unmap_single(adapter->pdev, rxd->addr,
1296                                                  rxd->len, PCI_DMA_FROMDEVICE);
1297                                 dev_kfree_skb(rq->buf_info[ring_idx][i].skb);
1298                                 rq->buf_info[ring_idx][i].skb = NULL;
1299                         } else if (rxd->btype == VMXNET3_RXD_BTYPE_BODY &&
1300                                         rq->buf_info[ring_idx][i].page) {
1301                                 pci_unmap_page(adapter->pdev, rxd->addr,
1302                                                rxd->len, PCI_DMA_FROMDEVICE);
1303                                 put_page(rq->buf_info[ring_idx][i].page);
1304                                 rq->buf_info[ring_idx][i].page = NULL;
1305                         }
1306                 }
1307
1308                 rq->rx_ring[ring_idx].gen = VMXNET3_INIT_GEN;
1309                 rq->rx_ring[ring_idx].next2fill =
1310                                         rq->rx_ring[ring_idx].next2comp = 0;
1311                 rq->uncommitted[ring_idx] = 0;
1312         }
1313
1314         rq->comp_ring.gen = VMXNET3_INIT_GEN;
1315         rq->comp_ring.next2proc = 0;
1316 }
1317
1318
1319 static void
1320 vmxnet3_rq_cleanup_all(struct vmxnet3_adapter *adapter)
1321 {
1322         int i;
1323
1324         for (i = 0; i < adapter->num_rx_queues; i++)
1325                 vmxnet3_rq_cleanup(&adapter->rx_queue[i], adapter);
1326 }
1327
1328
1329 void vmxnet3_rq_destroy(struct vmxnet3_rx_queue *rq,
1330                         struct vmxnet3_adapter *adapter)
1331 {
1332         int i;
1333         int j;
1334
1335         /* all rx buffers must have already been freed */
1336         for (i = 0; i < 2; i++) {
1337                 if (rq->buf_info[i]) {
1338                         for (j = 0; j < rq->rx_ring[i].size; j++)
1339                                 BUG_ON(rq->buf_info[i][j].page != NULL);
1340                 }
1341         }
1342
1343
1344         kfree(rq->buf_info[0]);
1345
1346         for (i = 0; i < 2; i++) {
1347                 if (rq->rx_ring[i].base) {
1348                         pci_free_consistent(adapter->pdev, rq->rx_ring[i].size
1349                                             * sizeof(struct Vmxnet3_RxDesc),
1350                                             rq->rx_ring[i].base,
1351                                             rq->rx_ring[i].basePA);
1352                         rq->rx_ring[i].base = NULL;
1353                 }
1354                 rq->buf_info[i] = NULL;
1355         }
1356
1357         if (rq->comp_ring.base) {
1358                 pci_free_consistent(adapter->pdev, rq->comp_ring.size *
1359                                     sizeof(struct Vmxnet3_RxCompDesc),
1360                                     rq->comp_ring.base, rq->comp_ring.basePA);
1361                 rq->comp_ring.base = NULL;
1362         }
1363 }
1364
1365
1366 static int
1367 vmxnet3_rq_init(struct vmxnet3_rx_queue *rq,
1368                 struct vmxnet3_adapter  *adapter)
1369 {
1370         int i;
1371
1372         /* initialize buf_info */
1373         for (i = 0; i < rq->rx_ring[0].size; i++) {
1374
1375                 /* 1st buf for a pkt is skbuff */
1376                 if (i % adapter->rx_buf_per_pkt == 0) {
1377                         rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_SKB;
1378                         rq->buf_info[0][i].len = adapter->skb_buf_size;
1379                 } else { /* subsequent bufs for a pkt is frag */
1380                         rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_PAGE;
1381                         rq->buf_info[0][i].len = PAGE_SIZE;
1382                 }
1383         }
1384         for (i = 0; i < rq->rx_ring[1].size; i++) {
1385                 rq->buf_info[1][i].buf_type = VMXNET3_RX_BUF_PAGE;
1386                 rq->buf_info[1][i].len = PAGE_SIZE;
1387         }
1388
1389         /* reset internal state and allocate buffers for both rings */
1390         for (i = 0; i < 2; i++) {
1391                 rq->rx_ring[i].next2fill = rq->rx_ring[i].next2comp = 0;
1392                 rq->uncommitted[i] = 0;
1393
1394                 memset(rq->rx_ring[i].base, 0, rq->rx_ring[i].size *
1395                        sizeof(struct Vmxnet3_RxDesc));
1396                 rq->rx_ring[i].gen = VMXNET3_INIT_GEN;
1397         }
1398         if (vmxnet3_rq_alloc_rx_buf(rq, 0, rq->rx_ring[0].size - 1,
1399                                     adapter) == 0) {
1400                 /* at least has 1 rx buffer for the 1st ring */
1401                 return -ENOMEM;
1402         }
1403         vmxnet3_rq_alloc_rx_buf(rq, 1, rq->rx_ring[1].size - 1, adapter);
1404
1405         /* reset the comp ring */
1406         rq->comp_ring.next2proc = 0;
1407         memset(rq->comp_ring.base, 0, rq->comp_ring.size *
1408                sizeof(struct Vmxnet3_RxCompDesc));
1409         rq->comp_ring.gen = VMXNET3_INIT_GEN;
1410
1411         /* reset rxctx */
1412         rq->rx_ctx.skb = NULL;
1413
1414         /* stats are not reset */
1415         return 0;
1416 }
1417
1418
1419 static int
1420 vmxnet3_rq_init_all(struct vmxnet3_adapter *adapter)
1421 {
1422         int i, err = 0;
1423
1424         for (i = 0; i < adapter->num_rx_queues; i++) {
1425                 err = vmxnet3_rq_init(&adapter->rx_queue[i], adapter);
1426                 if (unlikely(err)) {
1427                         dev_err(&adapter->netdev->dev, "%s: failed to "
1428                                 "initialize rx queue%i\n",
1429                                 adapter->netdev->name, i);
1430                         break;
1431                 }
1432         }
1433         return err;
1434
1435 }
1436
1437
1438 static int
1439 vmxnet3_rq_create(struct vmxnet3_rx_queue *rq, struct vmxnet3_adapter *adapter)
1440 {
1441         int i;
1442         size_t sz;
1443         struct vmxnet3_rx_buf_info *bi;
1444
1445         for (i = 0; i < 2; i++) {
1446
1447                 sz = rq->rx_ring[i].size * sizeof(struct Vmxnet3_RxDesc);
1448                 rq->rx_ring[i].base = pci_alloc_consistent(adapter->pdev, sz,
1449                                                         &rq->rx_ring[i].basePA);
1450                 if (!rq->rx_ring[i].base) {
1451                         printk(KERN_ERR "%s: failed to allocate rx ring %d\n",
1452                                adapter->netdev->name, i);
1453                         goto err;
1454                 }
1455         }
1456
1457         sz = rq->comp_ring.size * sizeof(struct Vmxnet3_RxCompDesc);
1458         rq->comp_ring.base = pci_alloc_consistent(adapter->pdev, sz,
1459                                                   &rq->comp_ring.basePA);
1460         if (!rq->comp_ring.base) {
1461                 printk(KERN_ERR "%s: failed to allocate rx comp ring\n",
1462                        adapter->netdev->name);
1463                 goto err;
1464         }
1465
1466         sz = sizeof(struct vmxnet3_rx_buf_info) * (rq->rx_ring[0].size +
1467                                                    rq->rx_ring[1].size);
1468         bi = kzalloc(sz, GFP_KERNEL);
1469         if (!bi) {
1470                 printk(KERN_ERR "%s: failed to allocate rx bufinfo\n",
1471                        adapter->netdev->name);
1472                 goto err;
1473         }
1474         rq->buf_info[0] = bi;
1475         rq->buf_info[1] = bi + rq->rx_ring[0].size;
1476
1477         return 0;
1478
1479 err:
1480         vmxnet3_rq_destroy(rq, adapter);
1481         return -ENOMEM;
1482 }
1483
1484
1485 static int
1486 vmxnet3_rq_create_all(struct vmxnet3_adapter *adapter)
1487 {
1488         int i, err = 0;
1489
1490         for (i = 0; i < adapter->num_rx_queues; i++) {
1491                 err = vmxnet3_rq_create(&adapter->rx_queue[i], adapter);
1492                 if (unlikely(err)) {
1493                         dev_err(&adapter->netdev->dev,
1494                                 "%s: failed to create rx queue%i\n",
1495                                 adapter->netdev->name, i);
1496                         goto err_out;
1497                 }
1498         }
1499         return err;
1500 err_out:
1501         vmxnet3_rq_destroy_all(adapter);
1502         return err;
1503
1504 }
1505
1506 /* Multiple queue aware polling function for tx and rx */
1507
1508 static int
1509 vmxnet3_do_poll(struct vmxnet3_adapter *adapter, int budget)
1510 {
1511         int rcd_done = 0, i;
1512         if (unlikely(adapter->shared->ecr))
1513                 vmxnet3_process_events(adapter);
1514         for (i = 0; i < adapter->num_tx_queues; i++)
1515                 vmxnet3_tq_tx_complete(&adapter->tx_queue[i], adapter);
1516
1517         for (i = 0; i < adapter->num_rx_queues; i++)
1518                 rcd_done += vmxnet3_rq_rx_complete(&adapter->rx_queue[i],
1519                                                    adapter, budget);
1520         return rcd_done;
1521 }
1522
1523
1524 static int
1525 vmxnet3_poll(struct napi_struct *napi, int budget)
1526 {
1527         struct vmxnet3_rx_queue *rx_queue = container_of(napi,
1528                                           struct vmxnet3_rx_queue, napi);
1529         int rxd_done;
1530
1531         rxd_done = vmxnet3_do_poll(rx_queue->adapter, budget);
1532
1533         if (rxd_done < budget) {
1534                 napi_complete(napi);
1535                 vmxnet3_enable_all_intrs(rx_queue->adapter);
1536         }
1537         return rxd_done;
1538 }
1539
1540 /*
1541  * NAPI polling function for MSI-X mode with multiple Rx queues
1542  * Returns the # of the NAPI credit consumed (# of rx descriptors processed)
1543  */
1544
1545 static int
1546 vmxnet3_poll_rx_only(struct napi_struct *napi, int budget)
1547 {
1548         struct vmxnet3_rx_queue *rq = container_of(napi,
1549                                                 struct vmxnet3_rx_queue, napi);
1550         struct vmxnet3_adapter *adapter = rq->adapter;
1551         int rxd_done;
1552
1553         /* When sharing interrupt with corresponding tx queue, process
1554          * tx completions in that queue as well
1555          */
1556         if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE) {
1557                 struct vmxnet3_tx_queue *tq =
1558                                 &adapter->tx_queue[rq - adapter->rx_queue];
1559                 vmxnet3_tq_tx_complete(tq, adapter);
1560         }
1561
1562         rxd_done = vmxnet3_rq_rx_complete(rq, adapter, budget);
1563
1564         if (rxd_done < budget) {
1565                 napi_complete(napi);
1566                 vmxnet3_enable_intr(adapter, rq->comp_ring.intr_idx);
1567         }
1568         return rxd_done;
1569 }
1570
1571
1572 #ifdef CONFIG_PCI_MSI
1573
1574 /*
1575  * Handle completion interrupts on tx queues
1576  * Returns whether or not the intr is handled
1577  */
1578
1579 static irqreturn_t
1580 vmxnet3_msix_tx(int irq, void *data)
1581 {
1582         struct vmxnet3_tx_queue *tq = data;
1583         struct vmxnet3_adapter *adapter = tq->adapter;
1584
1585         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1586                 vmxnet3_disable_intr(adapter, tq->comp_ring.intr_idx);
1587
1588         /* Handle the case where only one irq is allocate for all tx queues */
1589         if (adapter->share_intr == VMXNET3_INTR_TXSHARE) {
1590                 int i;
1591                 for (i = 0; i < adapter->num_tx_queues; i++) {
1592                         struct vmxnet3_tx_queue *txq = &adapter->tx_queue[i];
1593                         vmxnet3_tq_tx_complete(txq, adapter);
1594                 }
1595         } else {
1596                 vmxnet3_tq_tx_complete(tq, adapter);
1597         }
1598         vmxnet3_enable_intr(adapter, tq->comp_ring.intr_idx);
1599
1600         return IRQ_HANDLED;
1601 }
1602
1603
1604 /*
1605  * Handle completion interrupts on rx queues. Returns whether or not the
1606  * intr is handled
1607  */
1608
1609 static irqreturn_t
1610 vmxnet3_msix_rx(int irq, void *data)
1611 {
1612         struct vmxnet3_rx_queue *rq = data;
1613         struct vmxnet3_adapter *adapter = rq->adapter;
1614
1615         /* disable intr if needed */
1616         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1617                 vmxnet3_disable_intr(adapter, rq->comp_ring.intr_idx);
1618         napi_schedule(&rq->napi);
1619
1620         return IRQ_HANDLED;
1621 }
1622
1623 /*
1624  *----------------------------------------------------------------------------
1625  *
1626  * vmxnet3_msix_event --
1627  *
1628  *    vmxnet3 msix event intr handler
1629  *
1630  * Result:
1631  *    whether or not the intr is handled
1632  *
1633  *----------------------------------------------------------------------------
1634  */
1635
1636 static irqreturn_t
1637 vmxnet3_msix_event(int irq, void *data)
1638 {
1639         struct net_device *dev = data;
1640         struct vmxnet3_adapter *adapter = netdev_priv(dev);
1641
1642         /* disable intr if needed */
1643         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1644                 vmxnet3_disable_intr(adapter, adapter->intr.event_intr_idx);
1645
1646         if (adapter->shared->ecr)
1647                 vmxnet3_process_events(adapter);
1648
1649         vmxnet3_enable_intr(adapter, adapter->intr.event_intr_idx);
1650
1651         return IRQ_HANDLED;
1652 }
1653
1654 #endif /* CONFIG_PCI_MSI  */
1655
1656
1657 /* Interrupt handler for vmxnet3  */
1658 static irqreturn_t
1659 vmxnet3_intr(int irq, void *dev_id)
1660 {
1661         struct net_device *dev = dev_id;
1662         struct vmxnet3_adapter *adapter = netdev_priv(dev);
1663
1664         if (adapter->intr.type == VMXNET3_IT_INTX) {
1665                 u32 icr = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_ICR);
1666                 if (unlikely(icr == 0))
1667                         /* not ours */
1668                         return IRQ_NONE;
1669         }
1670
1671
1672         /* disable intr if needed */
1673         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1674                 vmxnet3_disable_all_intrs(adapter);
1675
1676         napi_schedule(&adapter->rx_queue[0].napi);
1677
1678         return IRQ_HANDLED;
1679 }
1680
1681 #ifdef CONFIG_NET_POLL_CONTROLLER
1682
1683 /* netpoll callback. */
1684 static void
1685 vmxnet3_netpoll(struct net_device *netdev)
1686 {
1687         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1688
1689         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1690                 vmxnet3_disable_all_intrs(adapter);
1691
1692         vmxnet3_do_poll(adapter, adapter->rx_queue[0].rx_ring[0].size);
1693         vmxnet3_enable_all_intrs(adapter);
1694
1695 }
1696 #endif  /* CONFIG_NET_POLL_CONTROLLER */
1697
1698 static int
1699 vmxnet3_request_irqs(struct vmxnet3_adapter *adapter)
1700 {
1701         struct vmxnet3_intr *intr = &adapter->intr;
1702         int err = 0, i;
1703         int vector = 0;
1704
1705 #ifdef CONFIG_PCI_MSI
1706         if (adapter->intr.type == VMXNET3_IT_MSIX) {
1707                 for (i = 0; i < adapter->num_tx_queues; i++) {
1708                         if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE) {
1709                                 sprintf(adapter->tx_queue[i].name, "%s-tx-%d",
1710                                         adapter->netdev->name, vector);
1711                                 err = request_irq(
1712                                               intr->msix_entries[vector].vector,
1713                                               vmxnet3_msix_tx, 0,
1714                                               adapter->tx_queue[i].name,
1715                                               &adapter->tx_queue[i]);
1716                         } else {
1717                                 sprintf(adapter->tx_queue[i].name, "%s-rxtx-%d",
1718                                         adapter->netdev->name, vector);
1719                         }
1720                         if (err) {
1721                                 dev_err(&adapter->netdev->dev,
1722                                         "Failed to request irq for MSIX, %s, "
1723                                         "error %d\n",
1724                                         adapter->tx_queue[i].name, err);
1725                                 return err;
1726                         }
1727
1728                         /* Handle the case where only 1 MSIx was allocated for
1729                          * all tx queues */
1730                         if (adapter->share_intr == VMXNET3_INTR_TXSHARE) {
1731                                 for (; i < adapter->num_tx_queues; i++)
1732                                         adapter->tx_queue[i].comp_ring.intr_idx
1733                                                                 = vector;
1734                                 vector++;
1735                                 break;
1736                         } else {
1737                                 adapter->tx_queue[i].comp_ring.intr_idx
1738                                                                 = vector++;
1739                         }
1740                 }
1741                 if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE)
1742                         vector = 0;
1743
1744                 for (i = 0; i < adapter->num_rx_queues; i++) {
1745                         if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE)
1746                                 sprintf(adapter->rx_queue[i].name, "%s-rx-%d",
1747                                         adapter->netdev->name, vector);
1748                         else
1749                                 sprintf(adapter->rx_queue[i].name, "%s-rxtx-%d",
1750                                         adapter->netdev->name, vector);
1751                         err = request_irq(intr->msix_entries[vector].vector,
1752                                           vmxnet3_msix_rx, 0,
1753                                           adapter->rx_queue[i].name,
1754                                           &(adapter->rx_queue[i]));
1755                         if (err) {
1756                                 printk(KERN_ERR "Failed to request irq for MSIX"
1757                                        ", %s, error %d\n",
1758                                        adapter->rx_queue[i].name, err);
1759                                 return err;
1760                         }
1761
1762                         adapter->rx_queue[i].comp_ring.intr_idx = vector++;
1763                 }
1764
1765                 sprintf(intr->event_msi_vector_name, "%s-event-%d",
1766                         adapter->netdev->name, vector);
1767                 err = request_irq(intr->msix_entries[vector].vector,
1768                                   vmxnet3_msix_event, 0,
1769                                   intr->event_msi_vector_name, adapter->netdev);
1770                 intr->event_intr_idx = vector;
1771
1772         } else if (intr->type == VMXNET3_IT_MSI) {
1773                 adapter->num_rx_queues = 1;
1774                 err = request_irq(adapter->pdev->irq, vmxnet3_intr, 0,
1775                                   adapter->netdev->name, adapter->netdev);
1776         } else {
1777 #endif
1778                 adapter->num_rx_queues = 1;
1779                 err = request_irq(adapter->pdev->irq, vmxnet3_intr,
1780                                   IRQF_SHARED, adapter->netdev->name,
1781                                   adapter->netdev);
1782 #ifdef CONFIG_PCI_MSI
1783         }
1784 #endif
1785         intr->num_intrs = vector + 1;
1786         if (err) {
1787                 printk(KERN_ERR "Failed to request irq %s (intr type:%d), error"
1788                        ":%d\n", adapter->netdev->name, intr->type, err);
1789         } else {
1790                 /* Number of rx queues will not change after this */
1791                 for (i = 0; i < adapter->num_rx_queues; i++) {
1792                         struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
1793                         rq->qid = i;
1794                         rq->qid2 = i + adapter->num_rx_queues;
1795                 }
1796
1797
1798
1799                 /* init our intr settings */
1800                 for (i = 0; i < intr->num_intrs; i++)
1801                         intr->mod_levels[i] = UPT1_IML_ADAPTIVE;
1802                 if (adapter->intr.type != VMXNET3_IT_MSIX) {
1803                         adapter->intr.event_intr_idx = 0;
1804                         for (i = 0; i < adapter->num_tx_queues; i++)
1805                                 adapter->tx_queue[i].comp_ring.intr_idx = 0;
1806                         adapter->rx_queue[0].comp_ring.intr_idx = 0;
1807                 }
1808
1809                 printk(KERN_INFO "%s: intr type %u, mode %u, %u vectors "
1810                        "allocated\n", adapter->netdev->name, intr->type,
1811                        intr->mask_mode, intr->num_intrs);
1812         }
1813
1814         return err;
1815 }
1816
1817
1818 static void
1819 vmxnet3_free_irqs(struct vmxnet3_adapter *adapter)
1820 {
1821         struct vmxnet3_intr *intr = &adapter->intr;
1822         BUG_ON(intr->type == VMXNET3_IT_AUTO || intr->num_intrs <= 0);
1823
1824         switch (intr->type) {
1825 #ifdef CONFIG_PCI_MSI
1826         case VMXNET3_IT_MSIX:
1827         {
1828                 int i, vector = 0;
1829
1830                 if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE) {
1831                         for (i = 0; i < adapter->num_tx_queues; i++) {
1832                                 free_irq(intr->msix_entries[vector++].vector,
1833                                          &(adapter->tx_queue[i]));
1834                                 if (adapter->share_intr == VMXNET3_INTR_TXSHARE)
1835                                         break;
1836                         }
1837                 }
1838
1839                 for (i = 0; i < adapter->num_rx_queues; i++) {
1840                         free_irq(intr->msix_entries[vector++].vector,
1841                                  &(adapter->rx_queue[i]));
1842                 }
1843
1844                 free_irq(intr->msix_entries[vector].vector,
1845                          adapter->netdev);
1846                 BUG_ON(vector >= intr->num_intrs);
1847                 break;
1848         }
1849 #endif
1850         case VMXNET3_IT_MSI:
1851                 free_irq(adapter->pdev->irq, adapter->netdev);
1852                 break;
1853         case VMXNET3_IT_INTX:
1854                 free_irq(adapter->pdev->irq, adapter->netdev);
1855                 break;
1856         default:
1857                 BUG_ON(true);
1858         }
1859 }
1860
1861 static void
1862 vmxnet3_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp)
1863 {
1864         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1865         struct Vmxnet3_DriverShared *shared = adapter->shared;
1866         u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1867         unsigned long flags;
1868
1869         if (grp) {
1870                 /* add vlan rx stripping. */
1871                 if (adapter->netdev->features & NETIF_F_HW_VLAN_RX) {
1872                         int i;
1873                         adapter->vlan_grp = grp;
1874
1875                         /*
1876                          *  Clear entire vfTable; then enable untagged pkts.
1877                          *  Note: setting one entry in vfTable to non-zero turns
1878                          *  on VLAN rx filtering.
1879                          */
1880                         for (i = 0; i < VMXNET3_VFT_SIZE; i++)
1881                                 vfTable[i] = 0;
1882
1883                         VMXNET3_SET_VFTABLE_ENTRY(vfTable, 0);
1884                         spin_lock_irqsave(&adapter->cmd_lock, flags);
1885                         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1886                                                VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1887                         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
1888                 } else {
1889                         printk(KERN_ERR "%s: vlan_rx_register when device has "
1890                                "no NETIF_F_HW_VLAN_RX\n", netdev->name);
1891                 }
1892         } else {
1893                 /* remove vlan rx stripping. */
1894                 struct Vmxnet3_DSDevRead *devRead = &shared->devRead;
1895                 adapter->vlan_grp = NULL;
1896
1897                 if (devRead->misc.uptFeatures & UPT1_F_RXVLAN) {
1898                         int i;
1899
1900                         for (i = 0; i < VMXNET3_VFT_SIZE; i++) {
1901                                 /* clear entire vfTable; this also disables
1902                                  * VLAN rx filtering
1903                                  */
1904                                 vfTable[i] = 0;
1905                         }
1906                         spin_lock_irqsave(&adapter->cmd_lock, flags);
1907                         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1908                                                VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1909                         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
1910                 }
1911         }
1912 }
1913
1914
1915 static void
1916 vmxnet3_restore_vlan(struct vmxnet3_adapter *adapter)
1917 {
1918         if (adapter->vlan_grp) {
1919                 u16 vid;
1920                 u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1921                 bool activeVlan = false;
1922
1923                 for (vid = 0; vid < VLAN_N_VID; vid++) {
1924                         if (vlan_group_get_device(adapter->vlan_grp, vid)) {
1925                                 VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid);
1926                                 activeVlan = true;
1927                         }
1928                 }
1929                 if (activeVlan) {
1930                         /* continue to allow untagged pkts */
1931                         VMXNET3_SET_VFTABLE_ENTRY(vfTable, 0);
1932                 }
1933         }
1934 }
1935
1936
1937 static void
1938 vmxnet3_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1939 {
1940         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1941         u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1942         unsigned long flags;
1943
1944         VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid);
1945         spin_lock_irqsave(&adapter->cmd_lock, flags);
1946         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1947                                VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1948         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
1949 }
1950
1951
1952 static void
1953 vmxnet3_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1954 {
1955         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1956         u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1957         unsigned long flags;
1958
1959         VMXNET3_CLEAR_VFTABLE_ENTRY(vfTable, vid);
1960         spin_lock_irqsave(&adapter->cmd_lock, flags);
1961         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1962                                VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1963         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
1964 }
1965
1966
1967 static u8 *
1968 vmxnet3_copy_mc(struct net_device *netdev)
1969 {
1970         u8 *buf = NULL;
1971         u32 sz = netdev_mc_count(netdev) * ETH_ALEN;
1972
1973         /* struct Vmxnet3_RxFilterConf.mfTableLen is u16. */
1974         if (sz <= 0xffff) {
1975                 /* We may be called with BH disabled */
1976                 buf = kmalloc(sz, GFP_ATOMIC);
1977                 if (buf) {
1978                         struct netdev_hw_addr *ha;
1979                         int i = 0;
1980
1981                         netdev_for_each_mc_addr(ha, netdev)
1982                                 memcpy(buf + i++ * ETH_ALEN, ha->addr,
1983                                        ETH_ALEN);
1984                 }
1985         }
1986         return buf;
1987 }
1988
1989
1990 static void
1991 vmxnet3_set_mc(struct net_device *netdev)
1992 {
1993         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1994         unsigned long flags;
1995         struct Vmxnet3_RxFilterConf *rxConf =
1996                                         &adapter->shared->devRead.rxFilterConf;
1997         u8 *new_table = NULL;
1998         u32 new_mode = VMXNET3_RXM_UCAST;
1999
2000         if (netdev->flags & IFF_PROMISC)
2001                 new_mode |= VMXNET3_RXM_PROMISC;
2002
2003         if (netdev->flags & IFF_BROADCAST)
2004                 new_mode |= VMXNET3_RXM_BCAST;
2005
2006         if (netdev->flags & IFF_ALLMULTI)
2007                 new_mode |= VMXNET3_RXM_ALL_MULTI;
2008         else
2009                 if (!netdev_mc_empty(netdev)) {
2010                         new_table = vmxnet3_copy_mc(netdev);
2011                         if (new_table) {
2012                                 new_mode |= VMXNET3_RXM_MCAST;
2013                                 rxConf->mfTableLen = cpu_to_le16(
2014                                         netdev_mc_count(netdev) * ETH_ALEN);
2015                                 rxConf->mfTablePA = cpu_to_le64(virt_to_phys(
2016                                                     new_table));
2017                         } else {
2018                                 printk(KERN_INFO "%s: failed to copy mcast list"
2019                                        ", setting ALL_MULTI\n", netdev->name);
2020                                 new_mode |= VMXNET3_RXM_ALL_MULTI;
2021                         }
2022                 }
2023
2024
2025         if (!(new_mode & VMXNET3_RXM_MCAST)) {
2026                 rxConf->mfTableLen = 0;
2027                 rxConf->mfTablePA = 0;
2028         }
2029
2030         spin_lock_irqsave(&adapter->cmd_lock, flags);
2031         if (new_mode != rxConf->rxMode) {
2032                 rxConf->rxMode = cpu_to_le32(new_mode);
2033                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2034                                        VMXNET3_CMD_UPDATE_RX_MODE);
2035         }
2036
2037         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2038                                VMXNET3_CMD_UPDATE_MAC_FILTERS);
2039         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2040
2041         kfree(new_table);
2042 }
2043
2044 void
2045 vmxnet3_rq_destroy_all(struct vmxnet3_adapter *adapter)
2046 {
2047         int i;
2048
2049         for (i = 0; i < adapter->num_rx_queues; i++)
2050                 vmxnet3_rq_destroy(&adapter->rx_queue[i], adapter);
2051 }
2052
2053
2054 /*
2055  *   Set up driver_shared based on settings in adapter.
2056  */
2057
2058 static void
2059 vmxnet3_setup_driver_shared(struct vmxnet3_adapter *adapter)
2060 {
2061         struct Vmxnet3_DriverShared *shared = adapter->shared;
2062         struct Vmxnet3_DSDevRead *devRead = &shared->devRead;
2063         struct Vmxnet3_TxQueueConf *tqc;
2064         struct Vmxnet3_RxQueueConf *rqc;
2065         int i;
2066
2067         memset(shared, 0, sizeof(*shared));
2068
2069         /* driver settings */
2070         shared->magic = cpu_to_le32(VMXNET3_REV1_MAGIC);
2071         devRead->misc.driverInfo.version = cpu_to_le32(
2072                                                 VMXNET3_DRIVER_VERSION_NUM);
2073         devRead->misc.driverInfo.gos.gosBits = (sizeof(void *) == 4 ?
2074                                 VMXNET3_GOS_BITS_32 : VMXNET3_GOS_BITS_64);
2075         devRead->misc.driverInfo.gos.gosType = VMXNET3_GOS_TYPE_LINUX;
2076         *((u32 *)&devRead->misc.driverInfo.gos) = cpu_to_le32(
2077                                 *((u32 *)&devRead->misc.driverInfo.gos));
2078         devRead->misc.driverInfo.vmxnet3RevSpt = cpu_to_le32(1);
2079         devRead->misc.driverInfo.uptVerSpt = cpu_to_le32(1);
2080
2081         devRead->misc.ddPA = cpu_to_le64(virt_to_phys(adapter));
2082         devRead->misc.ddLen = cpu_to_le32(sizeof(struct vmxnet3_adapter));
2083
2084         /* set up feature flags */
2085         if (adapter->rxcsum)
2086                 devRead->misc.uptFeatures |= UPT1_F_RXCSUM;
2087
2088         if (adapter->lro) {
2089                 devRead->misc.uptFeatures |= UPT1_F_LRO;
2090                 devRead->misc.maxNumRxSG = cpu_to_le16(1 + MAX_SKB_FRAGS);
2091         }
2092         if (adapter->netdev->features & NETIF_F_HW_VLAN_RX)
2093                 devRead->misc.uptFeatures |= UPT1_F_RXVLAN;
2094
2095         devRead->misc.mtu = cpu_to_le32(adapter->netdev->mtu);
2096         devRead->misc.queueDescPA = cpu_to_le64(adapter->queue_desc_pa);
2097         devRead->misc.queueDescLen = cpu_to_le32(
2098                 adapter->num_tx_queues * sizeof(struct Vmxnet3_TxQueueDesc) +
2099                 adapter->num_rx_queues * sizeof(struct Vmxnet3_RxQueueDesc));
2100
2101         /* tx queue settings */
2102         devRead->misc.numTxQueues =  adapter->num_tx_queues;
2103         for (i = 0; i < adapter->num_tx_queues; i++) {
2104                 struct vmxnet3_tx_queue *tq = &adapter->tx_queue[i];
2105                 BUG_ON(adapter->tx_queue[i].tx_ring.base == NULL);
2106                 tqc = &adapter->tqd_start[i].conf;
2107                 tqc->txRingBasePA   = cpu_to_le64(tq->tx_ring.basePA);
2108                 tqc->dataRingBasePA = cpu_to_le64(tq->data_ring.basePA);
2109                 tqc->compRingBasePA = cpu_to_le64(tq->comp_ring.basePA);
2110                 tqc->ddPA           = cpu_to_le64(virt_to_phys(tq->buf_info));
2111                 tqc->txRingSize     = cpu_to_le32(tq->tx_ring.size);
2112                 tqc->dataRingSize   = cpu_to_le32(tq->data_ring.size);
2113                 tqc->compRingSize   = cpu_to_le32(tq->comp_ring.size);
2114                 tqc->ddLen          = cpu_to_le32(
2115                                         sizeof(struct vmxnet3_tx_buf_info) *
2116                                         tqc->txRingSize);
2117                 tqc->intrIdx        = tq->comp_ring.intr_idx;
2118         }
2119
2120         /* rx queue settings */
2121         devRead->misc.numRxQueues = adapter->num_rx_queues;
2122         for (i = 0; i < adapter->num_rx_queues; i++) {
2123                 struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
2124                 rqc = &adapter->rqd_start[i].conf;
2125                 rqc->rxRingBasePA[0] = cpu_to_le64(rq->rx_ring[0].basePA);
2126                 rqc->rxRingBasePA[1] = cpu_to_le64(rq->rx_ring[1].basePA);
2127                 rqc->compRingBasePA  = cpu_to_le64(rq->comp_ring.basePA);
2128                 rqc->ddPA            = cpu_to_le64(virt_to_phys(
2129                                                         rq->buf_info));
2130                 rqc->rxRingSize[0]   = cpu_to_le32(rq->rx_ring[0].size);
2131                 rqc->rxRingSize[1]   = cpu_to_le32(rq->rx_ring[1].size);
2132                 rqc->compRingSize    = cpu_to_le32(rq->comp_ring.size);
2133                 rqc->ddLen           = cpu_to_le32(
2134                                         sizeof(struct vmxnet3_rx_buf_info) *
2135                                         (rqc->rxRingSize[0] +
2136                                          rqc->rxRingSize[1]));
2137                 rqc->intrIdx         = rq->comp_ring.intr_idx;
2138         }
2139
2140 #ifdef VMXNET3_RSS
2141         memset(adapter->rss_conf, 0, sizeof(*adapter->rss_conf));
2142
2143         if (adapter->rss) {
2144                 struct UPT1_RSSConf *rssConf = adapter->rss_conf;
2145                 devRead->misc.uptFeatures |= UPT1_F_RSS;
2146                 devRead->misc.numRxQueues = adapter->num_rx_queues;
2147                 rssConf->hashType = UPT1_RSS_HASH_TYPE_TCP_IPV4 |
2148                                     UPT1_RSS_HASH_TYPE_IPV4 |
2149                                     UPT1_RSS_HASH_TYPE_TCP_IPV6 |
2150                                     UPT1_RSS_HASH_TYPE_IPV6;
2151                 rssConf->hashFunc = UPT1_RSS_HASH_FUNC_TOEPLITZ;
2152                 rssConf->hashKeySize = UPT1_RSS_MAX_KEY_SIZE;
2153                 rssConf->indTableSize = VMXNET3_RSS_IND_TABLE_SIZE;
2154                 get_random_bytes(&rssConf->hashKey[0], rssConf->hashKeySize);
2155                 for (i = 0; i < rssConf->indTableSize; i++)
2156                         rssConf->indTable[i] = i % adapter->num_rx_queues;
2157
2158                 devRead->rssConfDesc.confVer = 1;
2159                 devRead->rssConfDesc.confLen = sizeof(*rssConf);
2160                 devRead->rssConfDesc.confPA  = virt_to_phys(rssConf);
2161         }
2162
2163 #endif /* VMXNET3_RSS */
2164
2165         /* intr settings */
2166         devRead->intrConf.autoMask = adapter->intr.mask_mode ==
2167                                      VMXNET3_IMM_AUTO;
2168         devRead->intrConf.numIntrs = adapter->intr.num_intrs;
2169         for (i = 0; i < adapter->intr.num_intrs; i++)
2170                 devRead->intrConf.modLevels[i] = adapter->intr.mod_levels[i];
2171
2172         devRead->intrConf.eventIntrIdx = adapter->intr.event_intr_idx;
2173         devRead->intrConf.intrCtrl |= cpu_to_le32(VMXNET3_IC_DISABLE_ALL);
2174
2175         /* rx filter settings */
2176         devRead->rxFilterConf.rxMode = 0;
2177         vmxnet3_restore_vlan(adapter);
2178         vmxnet3_write_mac_addr(adapter, adapter->netdev->dev_addr);
2179
2180         /* the rest are already zeroed */
2181 }
2182
2183
2184 int
2185 vmxnet3_activate_dev(struct vmxnet3_adapter *adapter)
2186 {
2187         int err, i;
2188         u32 ret;
2189         unsigned long flags;
2190
2191         dev_dbg(&adapter->netdev->dev, "%s: skb_buf_size %d, rx_buf_per_pkt %d,"
2192                 " ring sizes %u %u %u\n", adapter->netdev->name,
2193                 adapter->skb_buf_size, adapter->rx_buf_per_pkt,
2194                 adapter->tx_queue[0].tx_ring.size,
2195                 adapter->rx_queue[0].rx_ring[0].size,
2196                 adapter->rx_queue[0].rx_ring[1].size);
2197
2198         vmxnet3_tq_init_all(adapter);
2199         err = vmxnet3_rq_init_all(adapter);
2200         if (err) {
2201                 printk(KERN_ERR "Failed to init rx queue for %s: error %d\n",
2202                        adapter->netdev->name, err);
2203                 goto rq_err;
2204         }
2205
2206         err = vmxnet3_request_irqs(adapter);
2207         if (err) {
2208                 printk(KERN_ERR "Failed to setup irq for %s: error %d\n",
2209                        adapter->netdev->name, err);
2210                 goto irq_err;
2211         }
2212
2213         vmxnet3_setup_driver_shared(adapter);
2214
2215         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, VMXNET3_GET_ADDR_LO(
2216                                adapter->shared_pa));
2217         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, VMXNET3_GET_ADDR_HI(
2218                                adapter->shared_pa));
2219         spin_lock_irqsave(&adapter->cmd_lock, flags);
2220         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2221                                VMXNET3_CMD_ACTIVATE_DEV);
2222         ret = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
2223         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2224
2225         if (ret != 0) {
2226                 printk(KERN_ERR "Failed to activate dev %s: error %u\n",
2227                        adapter->netdev->name, ret);
2228                 err = -EINVAL;
2229                 goto activate_err;
2230         }
2231
2232         for (i = 0; i < adapter->num_rx_queues; i++) {
2233                 VMXNET3_WRITE_BAR0_REG(adapter,
2234                                 VMXNET3_REG_RXPROD + i * VMXNET3_REG_ALIGN,
2235                                 adapter->rx_queue[i].rx_ring[0].next2fill);
2236                 VMXNET3_WRITE_BAR0_REG(adapter, (VMXNET3_REG_RXPROD2 +
2237                                 (i * VMXNET3_REG_ALIGN)),
2238                                 adapter->rx_queue[i].rx_ring[1].next2fill);
2239         }
2240
2241         /* Apply the rx filter settins last. */
2242         vmxnet3_set_mc(adapter->netdev);
2243
2244         /*
2245          * Check link state when first activating device. It will start the
2246          * tx queue if the link is up.
2247          */
2248         vmxnet3_check_link(adapter, true);
2249         for (i = 0; i < adapter->num_rx_queues; i++)
2250                 napi_enable(&adapter->rx_queue[i].napi);
2251         vmxnet3_enable_all_intrs(adapter);
2252         clear_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
2253         return 0;
2254
2255 activate_err:
2256         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, 0);
2257         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, 0);
2258         vmxnet3_free_irqs(adapter);
2259 irq_err:
2260 rq_err:
2261         /* free up buffers we allocated */
2262         vmxnet3_rq_cleanup_all(adapter);
2263         return err;
2264 }
2265
2266
2267 void
2268 vmxnet3_reset_dev(struct vmxnet3_adapter *adapter)
2269 {
2270         unsigned long flags;
2271         spin_lock_irqsave(&adapter->cmd_lock, flags);
2272         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, VMXNET3_CMD_RESET_DEV);
2273         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2274 }
2275
2276
2277 int
2278 vmxnet3_quiesce_dev(struct vmxnet3_adapter *adapter)
2279 {
2280         int i;
2281         unsigned long flags;
2282         if (test_and_set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state))
2283                 return 0;
2284
2285
2286         spin_lock_irqsave(&adapter->cmd_lock, flags);
2287         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2288                                VMXNET3_CMD_QUIESCE_DEV);
2289         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2290         vmxnet3_disable_all_intrs(adapter);
2291
2292         for (i = 0; i < adapter->num_rx_queues; i++)
2293                 napi_disable(&adapter->rx_queue[i].napi);
2294         netif_tx_disable(adapter->netdev);
2295         adapter->link_speed = 0;
2296         netif_carrier_off(adapter->netdev);
2297
2298         vmxnet3_tq_cleanup_all(adapter);
2299         vmxnet3_rq_cleanup_all(adapter);
2300         vmxnet3_free_irqs(adapter);
2301         return 0;
2302 }
2303
2304
2305 static void
2306 vmxnet3_write_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac)
2307 {
2308         u32 tmp;
2309
2310         tmp = *(u32 *)mac;
2311         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACL, tmp);
2312
2313         tmp = (mac[5] << 8) | mac[4];
2314         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACH, tmp);
2315 }
2316
2317
2318 static int
2319 vmxnet3_set_mac_addr(struct net_device *netdev, void *p)
2320 {
2321         struct sockaddr *addr = p;
2322         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2323
2324         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2325         vmxnet3_write_mac_addr(adapter, addr->sa_data);
2326
2327         return 0;
2328 }
2329
2330
2331 /* ==================== initialization and cleanup routines ============ */
2332
2333 static int
2334 vmxnet3_alloc_pci_resources(struct vmxnet3_adapter *adapter, bool *dma64)
2335 {
2336         int err;
2337         unsigned long mmio_start, mmio_len;
2338         struct pci_dev *pdev = adapter->pdev;
2339
2340         err = pci_enable_device(pdev);
2341         if (err) {
2342                 printk(KERN_ERR "Failed to enable adapter %s: error %d\n",
2343                        pci_name(pdev), err);
2344                 return err;
2345         }
2346
2347         if (pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) == 0) {
2348                 if (pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64)) != 0) {
2349                         printk(KERN_ERR "pci_set_consistent_dma_mask failed "
2350                                "for adapter %s\n", pci_name(pdev));
2351                         err = -EIO;
2352                         goto err_set_mask;
2353                 }
2354                 *dma64 = true;
2355         } else {
2356                 if (pci_set_dma_mask(pdev, DMA_BIT_MASK(32)) != 0) {
2357                         printk(KERN_ERR "pci_set_dma_mask failed for adapter "
2358                                "%s\n",  pci_name(pdev));
2359                         err = -EIO;
2360                         goto err_set_mask;
2361                 }
2362                 *dma64 = false;
2363         }
2364
2365         err = pci_request_selected_regions(pdev, (1 << 2) - 1,
2366                                            vmxnet3_driver_name);
2367         if (err) {
2368                 printk(KERN_ERR "Failed to request region for adapter %s: "
2369                        "error %d\n", pci_name(pdev), err);
2370                 goto err_set_mask;
2371         }
2372
2373         pci_set_master(pdev);
2374
2375         mmio_start = pci_resource_start(pdev, 0);
2376         mmio_len = pci_resource_len(pdev, 0);
2377         adapter->hw_addr0 = ioremap(mmio_start, mmio_len);
2378         if (!adapter->hw_addr0) {
2379                 printk(KERN_ERR "Failed to map bar0 for adapter %s\n",
2380                        pci_name(pdev));
2381                 err = -EIO;
2382                 goto err_ioremap;
2383         }
2384
2385         mmio_start = pci_resource_start(pdev, 1);
2386         mmio_len = pci_resource_len(pdev, 1);
2387         adapter->hw_addr1 = ioremap(mmio_start, mmio_len);
2388         if (!adapter->hw_addr1) {
2389                 printk(KERN_ERR "Failed to map bar1 for adapter %s\n",
2390                        pci_name(pdev));
2391                 err = -EIO;
2392                 goto err_bar1;
2393         }
2394         return 0;
2395
2396 err_bar1:
2397         iounmap(adapter->hw_addr0);
2398 err_ioremap:
2399         pci_release_selected_regions(pdev, (1 << 2) - 1);
2400 err_set_mask:
2401         pci_disable_device(pdev);
2402         return err;
2403 }
2404
2405
2406 static void
2407 vmxnet3_free_pci_resources(struct vmxnet3_adapter *adapter)
2408 {
2409         BUG_ON(!adapter->pdev);
2410
2411         iounmap(adapter->hw_addr0);
2412         iounmap(adapter->hw_addr1);
2413         pci_release_selected_regions(adapter->pdev, (1 << 2) - 1);
2414         pci_disable_device(adapter->pdev);
2415 }
2416
2417
2418 static void
2419 vmxnet3_adjust_rx_ring_size(struct vmxnet3_adapter *adapter)
2420 {
2421         size_t sz, i, ring0_size, ring1_size, comp_size;
2422         struct vmxnet3_rx_queue *rq = &adapter->rx_queue[0];
2423
2424
2425         if (adapter->netdev->mtu <= VMXNET3_MAX_SKB_BUF_SIZE -
2426                                     VMXNET3_MAX_ETH_HDR_SIZE) {
2427                 adapter->skb_buf_size = adapter->netdev->mtu +
2428                                         VMXNET3_MAX_ETH_HDR_SIZE;
2429                 if (adapter->skb_buf_size < VMXNET3_MIN_T0_BUF_SIZE)
2430                         adapter->skb_buf_size = VMXNET3_MIN_T0_BUF_SIZE;
2431
2432                 adapter->rx_buf_per_pkt = 1;
2433         } else {
2434                 adapter->skb_buf_size = VMXNET3_MAX_SKB_BUF_SIZE;
2435                 sz = adapter->netdev->mtu - VMXNET3_MAX_SKB_BUF_SIZE +
2436                                             VMXNET3_MAX_ETH_HDR_SIZE;
2437                 adapter->rx_buf_per_pkt = 1 + (sz + PAGE_SIZE - 1) / PAGE_SIZE;
2438         }
2439
2440         /*
2441          * for simplicity, force the ring0 size to be a multiple of
2442          * rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN
2443          */
2444         sz = adapter->rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN;
2445         ring0_size = adapter->rx_queue[0].rx_ring[0].size;
2446         ring0_size = (ring0_size + sz - 1) / sz * sz;
2447         ring0_size = min_t(u32, ring0_size, VMXNET3_RX_RING_MAX_SIZE /
2448                            sz * sz);
2449         ring1_size = adapter->rx_queue[0].rx_ring[1].size;
2450         comp_size = ring0_size + ring1_size;
2451
2452         for (i = 0; i < adapter->num_rx_queues; i++) {
2453                 rq = &adapter->rx_queue[i];
2454                 rq->rx_ring[0].size = ring0_size;
2455                 rq->rx_ring[1].size = ring1_size;
2456                 rq->comp_ring.size = comp_size;
2457         }
2458 }
2459
2460
2461 int
2462 vmxnet3_create_queues(struct vmxnet3_adapter *adapter, u32 tx_ring_size,
2463                       u32 rx_ring_size, u32 rx_ring2_size)
2464 {
2465         int err = 0, i;
2466
2467         for (i = 0; i < adapter->num_tx_queues; i++) {
2468                 struct vmxnet3_tx_queue *tq = &adapter->tx_queue[i];
2469                 tq->tx_ring.size   = tx_ring_size;
2470                 tq->data_ring.size = tx_ring_size;
2471                 tq->comp_ring.size = tx_ring_size;
2472                 tq->shared = &adapter->tqd_start[i].ctrl;
2473                 tq->stopped = true;
2474                 tq->adapter = adapter;
2475                 tq->qid = i;
2476                 err = vmxnet3_tq_create(tq, adapter);
2477                 /*
2478                  * Too late to change num_tx_queues. We cannot do away with
2479                  * lesser number of queues than what we asked for
2480                  */
2481                 if (err)
2482                         goto queue_err;
2483         }
2484
2485         adapter->rx_queue[0].rx_ring[0].size = rx_ring_size;
2486         adapter->rx_queue[0].rx_ring[1].size = rx_ring2_size;
2487         vmxnet3_adjust_rx_ring_size(adapter);
2488         for (i = 0; i < adapter->num_rx_queues; i++) {
2489                 struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
2490                 /* qid and qid2 for rx queues will be assigned later when num
2491                  * of rx queues is finalized after allocating intrs */
2492                 rq->shared = &adapter->rqd_start[i].ctrl;
2493                 rq->adapter = adapter;
2494                 err = vmxnet3_rq_create(rq, adapter);
2495                 if (err) {
2496                         if (i == 0) {
2497                                 printk(KERN_ERR "Could not allocate any rx"
2498                                        "queues. Aborting.\n");
2499                                 goto queue_err;
2500                         } else {
2501                                 printk(KERN_INFO "Number of rx queues changed "
2502                                        "to : %d.\n", i);
2503                                 adapter->num_rx_queues = i;
2504                                 err = 0;
2505                                 break;
2506                         }
2507                 }
2508         }
2509         return err;
2510 queue_err:
2511         vmxnet3_tq_destroy_all(adapter);
2512         return err;
2513 }
2514
2515 static int
2516 vmxnet3_open(struct net_device *netdev)
2517 {
2518         struct vmxnet3_adapter *adapter;
2519         int err, i;
2520
2521         adapter = netdev_priv(netdev);
2522
2523         for (i = 0; i < adapter->num_tx_queues; i++)
2524                 spin_lock_init(&adapter->tx_queue[i].tx_lock);
2525
2526         err = vmxnet3_create_queues(adapter, VMXNET3_DEF_TX_RING_SIZE,
2527                                     VMXNET3_DEF_RX_RING_SIZE,
2528                                     VMXNET3_DEF_RX_RING_SIZE);
2529         if (err)
2530                 goto queue_err;
2531
2532         err = vmxnet3_activate_dev(adapter);
2533         if (err)
2534                 goto activate_err;
2535
2536         return 0;
2537
2538 activate_err:
2539         vmxnet3_rq_destroy_all(adapter);
2540         vmxnet3_tq_destroy_all(adapter);
2541 queue_err:
2542         return err;
2543 }
2544
2545
2546 static int
2547 vmxnet3_close(struct net_device *netdev)
2548 {
2549         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2550
2551         /*
2552          * Reset_work may be in the middle of resetting the device, wait for its
2553          * completion.
2554          */
2555         while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2556                 msleep(1);
2557
2558         vmxnet3_quiesce_dev(adapter);
2559
2560         vmxnet3_rq_destroy_all(adapter);
2561         vmxnet3_tq_destroy_all(adapter);
2562
2563         clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2564
2565
2566         return 0;
2567 }
2568
2569
2570 void
2571 vmxnet3_force_close(struct vmxnet3_adapter *adapter)
2572 {
2573         int i;
2574
2575         /*
2576          * we must clear VMXNET3_STATE_BIT_RESETTING, otherwise
2577          * vmxnet3_close() will deadlock.
2578          */
2579         BUG_ON(test_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state));
2580
2581         /* we need to enable NAPI, otherwise dev_close will deadlock */
2582         for (i = 0; i < adapter->num_rx_queues; i++)
2583                 napi_enable(&adapter->rx_queue[i].napi);
2584         dev_close(adapter->netdev);
2585 }
2586
2587
2588 static int
2589 vmxnet3_change_mtu(struct net_device *netdev, int new_mtu)
2590 {
2591         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2592         int err = 0;
2593
2594         if (new_mtu < VMXNET3_MIN_MTU || new_mtu > VMXNET3_MAX_MTU)
2595                 return -EINVAL;
2596
2597         if (new_mtu > 1500 && !adapter->jumbo_frame)
2598                 return -EINVAL;
2599
2600         netdev->mtu = new_mtu;
2601
2602         /*
2603          * Reset_work may be in the middle of resetting the device, wait for its
2604          * completion.
2605          */
2606         while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2607                 msleep(1);
2608
2609         if (netif_running(netdev)) {
2610                 vmxnet3_quiesce_dev(adapter);
2611                 vmxnet3_reset_dev(adapter);
2612
2613                 /* we need to re-create the rx queue based on the new mtu */
2614                 vmxnet3_rq_destroy_all(adapter);
2615                 vmxnet3_adjust_rx_ring_size(adapter);
2616                 err = vmxnet3_rq_create_all(adapter);
2617                 if (err) {
2618                         printk(KERN_ERR "%s: failed to re-create rx queues,"
2619                                 " error %d. Closing it.\n", netdev->name, err);
2620                         goto out;
2621                 }
2622
2623                 err = vmxnet3_activate_dev(adapter);
2624                 if (err) {
2625                         printk(KERN_ERR "%s: failed to re-activate, error %d. "
2626                                 "Closing it\n", netdev->name, err);
2627                         goto out;
2628                 }
2629         }
2630
2631 out:
2632         clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2633         if (err)
2634                 vmxnet3_force_close(adapter);
2635
2636         return err;
2637 }
2638
2639
2640 static void
2641 vmxnet3_declare_features(struct vmxnet3_adapter *adapter, bool dma64)
2642 {
2643         struct net_device *netdev = adapter->netdev;
2644
2645         netdev->features = NETIF_F_SG |
2646                 NETIF_F_HW_CSUM |
2647                 NETIF_F_HW_VLAN_TX |
2648                 NETIF_F_HW_VLAN_RX |
2649                 NETIF_F_HW_VLAN_FILTER |
2650                 NETIF_F_TSO |
2651                 NETIF_F_TSO6 |
2652                 NETIF_F_LRO;
2653
2654         printk(KERN_INFO "features: sg csum vlan jf tso tsoIPv6 lro");
2655
2656         adapter->rxcsum = true;
2657         adapter->jumbo_frame = true;
2658         adapter->lro = true;
2659
2660         if (dma64) {
2661                 netdev->features |= NETIF_F_HIGHDMA;
2662                 printk(" highDMA");
2663         }
2664
2665         netdev->vlan_features = netdev->features;
2666         printk("\n");
2667 }
2668
2669
2670 static void
2671 vmxnet3_read_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac)
2672 {
2673         u32 tmp;
2674
2675         tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACL);
2676         *(u32 *)mac = tmp;
2677
2678         tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACH);
2679         mac[4] = tmp & 0xff;
2680         mac[5] = (tmp >> 8) & 0xff;
2681 }
2682
2683 #ifdef CONFIG_PCI_MSI
2684
2685 /*
2686  * Enable MSIx vectors.
2687  * Returns :
2688  *      0 on successful enabling of required vectors,
2689  *      VMXNET3_LINUX_MIN_MSIX_VECT when only minimum number of vectors required
2690  *       could be enabled.
2691  *      number of vectors which can be enabled otherwise (this number is smaller
2692  *       than VMXNET3_LINUX_MIN_MSIX_VECT)
2693  */
2694
2695 static int
2696 vmxnet3_acquire_msix_vectors(struct vmxnet3_adapter *adapter,
2697                              int vectors)
2698 {
2699         int err = 0, vector_threshold;
2700         vector_threshold = VMXNET3_LINUX_MIN_MSIX_VECT;
2701
2702         while (vectors >= vector_threshold) {
2703                 err = pci_enable_msix(adapter->pdev, adapter->intr.msix_entries,
2704                                       vectors);
2705                 if (!err) {
2706                         adapter->intr.num_intrs = vectors;
2707                         return 0;
2708                 } else if (err < 0) {
2709                         printk(KERN_ERR "Failed to enable MSI-X for %s, error"
2710                                " %d\n", adapter->netdev->name, err);
2711                         vectors = 0;
2712                 } else if (err < vector_threshold) {
2713                         break;
2714                 } else {
2715                         /* If fails to enable required number of MSI-x vectors
2716                          * try enabling minimum number of vectors required.
2717                          */
2718                         vectors = vector_threshold;
2719                         printk(KERN_ERR "Failed to enable %d MSI-X for %s, try"
2720                                " %d instead\n", vectors, adapter->netdev->name,
2721                                vector_threshold);
2722                 }
2723         }
2724
2725         printk(KERN_INFO "Number of MSI-X interrupts which can be allocatedi"
2726                " are lower than min threshold required.\n");
2727         return err;
2728 }
2729
2730
2731 #endif /* CONFIG_PCI_MSI */
2732
2733 static void
2734 vmxnet3_alloc_intr_resources(struct vmxnet3_adapter *adapter)
2735 {
2736         u32 cfg;
2737         unsigned long flags;
2738
2739         /* intr settings */
2740         spin_lock_irqsave(&adapter->cmd_lock, flags);
2741         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2742                                VMXNET3_CMD_GET_CONF_INTR);
2743         cfg = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
2744         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2745         adapter->intr.type = cfg & 0x3;
2746         adapter->intr.mask_mode = (cfg >> 2) & 0x3;
2747
2748         if (adapter->intr.type == VMXNET3_IT_AUTO) {
2749                 adapter->intr.type = VMXNET3_IT_MSIX;
2750         }
2751
2752 #ifdef CONFIG_PCI_MSI
2753         if (adapter->intr.type == VMXNET3_IT_MSIX) {
2754                 int vector, err = 0;
2755
2756                 adapter->intr.num_intrs = (adapter->share_intr ==
2757                                            VMXNET3_INTR_TXSHARE) ? 1 :
2758                                            adapter->num_tx_queues;
2759                 adapter->intr.num_intrs += (adapter->share_intr ==
2760                                            VMXNET3_INTR_BUDDYSHARE) ? 0 :
2761                                            adapter->num_rx_queues;
2762                 adapter->intr.num_intrs += 1;           /* for link event */
2763
2764                 adapter->intr.num_intrs = (adapter->intr.num_intrs >
2765                                            VMXNET3_LINUX_MIN_MSIX_VECT
2766                                            ? adapter->intr.num_intrs :
2767                                            VMXNET3_LINUX_MIN_MSIX_VECT);
2768
2769                 for (vector = 0; vector < adapter->intr.num_intrs; vector++)
2770                         adapter->intr.msix_entries[vector].entry = vector;
2771
2772                 err = vmxnet3_acquire_msix_vectors(adapter,
2773                                                    adapter->intr.num_intrs);
2774                 /* If we cannot allocate one MSIx vector per queue
2775                  * then limit the number of rx queues to 1
2776                  */
2777                 if (err == VMXNET3_LINUX_MIN_MSIX_VECT) {
2778                         if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE
2779                             || adapter->num_rx_queues != 1) {
2780                                 adapter->share_intr = VMXNET3_INTR_TXSHARE;
2781                                 printk(KERN_ERR "Number of rx queues : 1\n");
2782                                 adapter->num_rx_queues = 1;
2783                                 adapter->intr.num_intrs =
2784                                                 VMXNET3_LINUX_MIN_MSIX_VECT;
2785                         }
2786                         return;
2787                 }
2788                 if (!err)
2789                         return;
2790
2791                 /* If we cannot allocate MSIx vectors use only one rx queue */
2792                 printk(KERN_INFO "Failed to enable MSI-X for %s, error %d."
2793                        "#rx queues : 1, try MSI\n", adapter->netdev->name, err);
2794
2795                 adapter->intr.type = VMXNET3_IT_MSI;
2796         }
2797
2798         if (adapter->intr.type == VMXNET3_IT_MSI) {
2799                 int err;
2800                 err = pci_enable_msi(adapter->pdev);
2801                 if (!err) {
2802                         adapter->num_rx_queues = 1;
2803                         adapter->intr.num_intrs = 1;
2804                         return;
2805                 }
2806         }
2807 #endif /* CONFIG_PCI_MSI */
2808
2809         adapter->num_rx_queues = 1;
2810         printk(KERN_INFO "Using INTx interrupt, #Rx queues: 1.\n");
2811         adapter->intr.type = VMXNET3_IT_INTX;
2812
2813         /* INT-X related setting */
2814         adapter->intr.num_intrs = 1;
2815 }
2816
2817
2818 static void
2819 vmxnet3_free_intr_resources(struct vmxnet3_adapter *adapter)
2820 {
2821         if (adapter->intr.type == VMXNET3_IT_MSIX)
2822                 pci_disable_msix(adapter->pdev);
2823         else if (adapter->intr.type == VMXNET3_IT_MSI)
2824                 pci_disable_msi(adapter->pdev);
2825         else
2826                 BUG_ON(adapter->intr.type != VMXNET3_IT_INTX);
2827 }
2828
2829
2830 static void
2831 vmxnet3_tx_timeout(struct net_device *netdev)
2832 {
2833         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2834         adapter->tx_timeout_count++;
2835
2836         printk(KERN_ERR "%s: tx hang\n", adapter->netdev->name);
2837         schedule_work(&adapter->work);
2838         netif_wake_queue(adapter->netdev);
2839 }
2840
2841
2842 static void
2843 vmxnet3_reset_work(struct work_struct *data)
2844 {
2845         struct vmxnet3_adapter *adapter;
2846
2847         adapter = container_of(data, struct vmxnet3_adapter, work);
2848
2849         /* if another thread is resetting the device, no need to proceed */
2850         if (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2851                 return;
2852
2853         /* if the device is closed, we must leave it alone */
2854         rtnl_lock();
2855         if (netif_running(adapter->netdev)) {
2856                 printk(KERN_INFO "%s: resetting\n", adapter->netdev->name);
2857                 vmxnet3_quiesce_dev(adapter);
2858                 vmxnet3_reset_dev(adapter);
2859                 vmxnet3_activate_dev(adapter);
2860         } else {
2861                 printk(KERN_INFO "%s: already closed\n", adapter->netdev->name);
2862         }
2863         rtnl_unlock();
2864
2865         clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2866 }
2867
2868
2869 static int __devinit
2870 vmxnet3_probe_device(struct pci_dev *pdev,
2871                      const struct pci_device_id *id)
2872 {
2873         static const struct net_device_ops vmxnet3_netdev_ops = {
2874                 .ndo_open = vmxnet3_open,
2875                 .ndo_stop = vmxnet3_close,
2876                 .ndo_start_xmit = vmxnet3_xmit_frame,
2877                 .ndo_set_mac_address = vmxnet3_set_mac_addr,
2878                 .ndo_change_mtu = vmxnet3_change_mtu,
2879                 .ndo_get_stats = vmxnet3_get_stats,
2880                 .ndo_tx_timeout = vmxnet3_tx_timeout,
2881                 .ndo_set_multicast_list = vmxnet3_set_mc,
2882                 .ndo_vlan_rx_register = vmxnet3_vlan_rx_register,
2883                 .ndo_vlan_rx_add_vid = vmxnet3_vlan_rx_add_vid,
2884                 .ndo_vlan_rx_kill_vid = vmxnet3_vlan_rx_kill_vid,
2885 #ifdef CONFIG_NET_POLL_CONTROLLER
2886                 .ndo_poll_controller = vmxnet3_netpoll,
2887 #endif
2888         };
2889         int err;
2890         bool dma64 = false; /* stupid gcc */
2891         u32 ver;
2892         struct net_device *netdev;
2893         struct vmxnet3_adapter *adapter;
2894         u8 mac[ETH_ALEN];
2895         int size;
2896         int num_tx_queues;
2897         int num_rx_queues;
2898
2899 #ifdef VMXNET3_RSS
2900         if (enable_mq)
2901                 num_rx_queues = min(VMXNET3_DEVICE_MAX_RX_QUEUES,
2902                                     (int)num_online_cpus());
2903         else
2904 #endif
2905                 num_rx_queues = 1;
2906
2907         if (enable_mq)
2908                 num_tx_queues = min(VMXNET3_DEVICE_MAX_TX_QUEUES,
2909                                     (int)num_online_cpus());
2910         else
2911                 num_tx_queues = 1;
2912
2913         netdev = alloc_etherdev_mq(sizeof(struct vmxnet3_adapter),
2914                                    max(num_tx_queues, num_rx_queues));
2915         printk(KERN_INFO "# of Tx queues : %d, # of Rx queues : %d\n",
2916                num_tx_queues, num_rx_queues);
2917
2918         if (!netdev) {
2919                 printk(KERN_ERR "Failed to alloc ethernet device for adapter "
2920                         "%s\n", pci_name(pdev));
2921                 return -ENOMEM;
2922         }
2923
2924         pci_set_drvdata(pdev, netdev);
2925         adapter = netdev_priv(netdev);
2926         adapter->netdev = netdev;
2927         adapter->pdev = pdev;
2928
2929         spin_lock_init(&adapter->cmd_lock);
2930         adapter->shared = pci_alloc_consistent(adapter->pdev,
2931                           sizeof(struct Vmxnet3_DriverShared),
2932                           &adapter->shared_pa);
2933         if (!adapter->shared) {
2934                 printk(KERN_ERR "Failed to allocate memory for %s\n",
2935                         pci_name(pdev));
2936                 err = -ENOMEM;
2937                 goto err_alloc_shared;
2938         }
2939
2940         adapter->num_rx_queues = num_rx_queues;
2941         adapter->num_tx_queues = num_tx_queues;
2942
2943         size = sizeof(struct Vmxnet3_TxQueueDesc) * adapter->num_tx_queues;
2944         size += sizeof(struct Vmxnet3_RxQueueDesc) * adapter->num_rx_queues;
2945         adapter->tqd_start = pci_alloc_consistent(adapter->pdev, size,
2946                              &adapter->queue_desc_pa);
2947
2948         if (!adapter->tqd_start) {
2949                 printk(KERN_ERR "Failed to allocate memory for %s\n",
2950                         pci_name(pdev));
2951                 err = -ENOMEM;
2952                 goto err_alloc_queue_desc;
2953         }
2954         adapter->rqd_start = (struct Vmxnet3_RxQueueDesc *)(adapter->tqd_start +
2955                                                         adapter->num_tx_queues);
2956
2957         adapter->pm_conf = kmalloc(sizeof(struct Vmxnet3_PMConf), GFP_KERNEL);
2958         if (adapter->pm_conf == NULL) {
2959                 printk(KERN_ERR "Failed to allocate memory for %s\n",
2960                         pci_name(pdev));
2961                 err = -ENOMEM;
2962                 goto err_alloc_pm;
2963         }
2964
2965 #ifdef VMXNET3_RSS
2966
2967         adapter->rss_conf = kmalloc(sizeof(struct UPT1_RSSConf), GFP_KERNEL);
2968         if (adapter->rss_conf == NULL) {
2969                 printk(KERN_ERR "Failed to allocate memory for %s\n",
2970                        pci_name(pdev));
2971                 err = -ENOMEM;
2972                 goto err_alloc_rss;
2973         }
2974 #endif /* VMXNET3_RSS */
2975
2976         err = vmxnet3_alloc_pci_resources(adapter, &dma64);
2977         if (err < 0)
2978                 goto err_alloc_pci;
2979
2980         ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_VRRS);
2981         if (ver & 1) {
2982                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_VRRS, 1);
2983         } else {
2984                 printk(KERN_ERR "Incompatible h/w version (0x%x) for adapter"
2985                        " %s\n", ver, pci_name(pdev));
2986                 err = -EBUSY;
2987                 goto err_ver;
2988         }
2989
2990         ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_UVRS);
2991         if (ver & 1) {
2992                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_UVRS, 1);
2993         } else {
2994                 printk(KERN_ERR "Incompatible upt version (0x%x) for "
2995                        "adapter %s\n", ver, pci_name(pdev));
2996                 err = -EBUSY;
2997                 goto err_ver;
2998         }
2999
3000         vmxnet3_declare_features(adapter, dma64);
3001
3002         adapter->dev_number = atomic_read(&devices_found);
3003
3004          adapter->share_intr = irq_share_mode;
3005         if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE &&
3006             adapter->num_tx_queues != adapter->num_rx_queues)
3007                 adapter->share_intr = VMXNET3_INTR_DONTSHARE;
3008
3009         vmxnet3_alloc_intr_resources(adapter);
3010
3011 #ifdef VMXNET3_RSS
3012         if (adapter->num_rx_queues > 1 &&
3013             adapter->intr.type == VMXNET3_IT_MSIX) {
3014                 adapter->rss = true;
3015                 printk(KERN_INFO "RSS is enabled.\n");
3016         } else {
3017                 adapter->rss = false;
3018         }
3019 #endif
3020
3021         vmxnet3_read_mac_addr(adapter, mac);
3022         memcpy(netdev->dev_addr,  mac, netdev->addr_len);
3023
3024         netdev->netdev_ops = &vmxnet3_netdev_ops;
3025         vmxnet3_set_ethtool_ops(netdev);
3026         netdev->watchdog_timeo = 5 * HZ;
3027
3028         INIT_WORK(&adapter->work, vmxnet3_reset_work);
3029
3030         if (adapter->intr.type == VMXNET3_IT_MSIX) {
3031                 int i;
3032                 for (i = 0; i < adapter->num_rx_queues; i++) {
3033                         netif_napi_add(adapter->netdev,
3034                                        &adapter->rx_queue[i].napi,
3035                                        vmxnet3_poll_rx_only, 64);
3036                 }
3037         } else {
3038                 netif_napi_add(adapter->netdev, &adapter->rx_queue[0].napi,
3039                                vmxnet3_poll, 64);
3040         }
3041
3042         netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
3043         netif_set_real_num_rx_queues(adapter->netdev, adapter->num_rx_queues);
3044
3045         SET_NETDEV_DEV(netdev, &pdev->dev);
3046         err = register_netdev(netdev);
3047
3048         if (err) {
3049                 printk(KERN_ERR "Failed to register adapter %s\n",
3050                         pci_name(pdev));
3051                 goto err_register;
3052         }
3053
3054         set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
3055         vmxnet3_check_link(adapter, false);
3056         atomic_inc(&devices_found);
3057         return 0;
3058
3059 err_register:
3060         vmxnet3_free_intr_resources(adapter);
3061 err_ver:
3062         vmxnet3_free_pci_resources(adapter);
3063 err_alloc_pci:
3064 #ifdef VMXNET3_RSS
3065         kfree(adapter->rss_conf);
3066 err_alloc_rss:
3067 #endif
3068         kfree(adapter->pm_conf);
3069 err_alloc_pm:
3070         pci_free_consistent(adapter->pdev, size, adapter->tqd_start,
3071                             adapter->queue_desc_pa);
3072 err_alloc_queue_desc:
3073         pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_DriverShared),
3074                             adapter->shared, adapter->shared_pa);
3075 err_alloc_shared:
3076         pci_set_drvdata(pdev, NULL);
3077         free_netdev(netdev);
3078         return err;
3079 }
3080
3081
3082 static void __devexit
3083 vmxnet3_remove_device(struct pci_dev *pdev)
3084 {
3085         struct net_device *netdev = pci_get_drvdata(pdev);
3086         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3087         int size = 0;
3088         int num_rx_queues;
3089
3090 #ifdef VMXNET3_RSS
3091         if (enable_mq)
3092                 num_rx_queues = min(VMXNET3_DEVICE_MAX_RX_QUEUES,
3093                                     (int)num_online_cpus());
3094         else
3095 #endif
3096                 num_rx_queues = 1;
3097
3098         cancel_work_sync(&adapter->work);
3099
3100         unregister_netdev(netdev);
3101
3102         vmxnet3_free_intr_resources(adapter);
3103         vmxnet3_free_pci_resources(adapter);
3104 #ifdef VMXNET3_RSS
3105         kfree(adapter->rss_conf);
3106 #endif
3107         kfree(adapter->pm_conf);
3108
3109         size = sizeof(struct Vmxnet3_TxQueueDesc) * adapter->num_tx_queues;
3110         size += sizeof(struct Vmxnet3_RxQueueDesc) * num_rx_queues;
3111         pci_free_consistent(adapter->pdev, size, adapter->tqd_start,
3112                             adapter->queue_desc_pa);
3113         pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_DriverShared),
3114                             adapter->shared, adapter->shared_pa);
3115         free_netdev(netdev);
3116 }
3117
3118
3119 #ifdef CONFIG_PM
3120
3121 static int
3122 vmxnet3_suspend(struct device *device)
3123 {
3124         struct pci_dev *pdev = to_pci_dev(device);
3125         struct net_device *netdev = pci_get_drvdata(pdev);
3126         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3127         struct Vmxnet3_PMConf *pmConf;
3128         struct ethhdr *ehdr;
3129         struct arphdr *ahdr;
3130         u8 *arpreq;
3131         struct in_device *in_dev;
3132         struct in_ifaddr *ifa;
3133         unsigned long flags;
3134         int i = 0;
3135
3136         if (!netif_running(netdev))
3137                 return 0;
3138
3139         for (i = 0; i < adapter->num_rx_queues; i++)
3140                 napi_disable(&adapter->rx_queue[i].napi);
3141
3142         vmxnet3_disable_all_intrs(adapter);
3143         vmxnet3_free_irqs(adapter);
3144         vmxnet3_free_intr_resources(adapter);
3145
3146         netif_device_detach(netdev);
3147         netif_tx_stop_all_queues(netdev);
3148
3149         /* Create wake-up filters. */
3150         pmConf = adapter->pm_conf;
3151         memset(pmConf, 0, sizeof(*pmConf));
3152
3153         if (adapter->wol & WAKE_UCAST) {
3154                 pmConf->filters[i].patternSize = ETH_ALEN;
3155                 pmConf->filters[i].maskSize = 1;
3156                 memcpy(pmConf->filters[i].pattern, netdev->dev_addr, ETH_ALEN);
3157                 pmConf->filters[i].mask[0] = 0x3F; /* LSB ETH_ALEN bits */
3158
3159                 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_FILTER;
3160                 i++;
3161         }
3162
3163         if (adapter->wol & WAKE_ARP) {
3164                 in_dev = in_dev_get(netdev);
3165                 if (!in_dev)
3166                         goto skip_arp;
3167
3168                 ifa = (struct in_ifaddr *)in_dev->ifa_list;
3169                 if (!ifa)
3170                         goto skip_arp;
3171
3172                 pmConf->filters[i].patternSize = ETH_HLEN + /* Ethernet header*/
3173                         sizeof(struct arphdr) +         /* ARP header */
3174                         2 * ETH_ALEN +          /* 2 Ethernet addresses*/
3175                         2 * sizeof(u32);        /*2 IPv4 addresses */
3176                 pmConf->filters[i].maskSize =
3177                         (pmConf->filters[i].patternSize - 1) / 8 + 1;
3178
3179                 /* ETH_P_ARP in Ethernet header. */
3180                 ehdr = (struct ethhdr *)pmConf->filters[i].pattern;
3181                 ehdr->h_proto = htons(ETH_P_ARP);
3182
3183                 /* ARPOP_REQUEST in ARP header. */
3184                 ahdr = (struct arphdr *)&pmConf->filters[i].pattern[ETH_HLEN];
3185                 ahdr->ar_op = htons(ARPOP_REQUEST);
3186                 arpreq = (u8 *)(ahdr + 1);
3187
3188                 /* The Unicast IPv4 address in 'tip' field. */
3189                 arpreq += 2 * ETH_ALEN + sizeof(u32);
3190                 *(u32 *)arpreq = ifa->ifa_address;
3191
3192                 /* The mask for the relevant bits. */
3193                 pmConf->filters[i].mask[0] = 0x00;
3194                 pmConf->filters[i].mask[1] = 0x30; /* ETH_P_ARP */
3195                 pmConf->filters[i].mask[2] = 0x30; /* ARPOP_REQUEST */
3196                 pmConf->filters[i].mask[3] = 0x00;
3197                 pmConf->filters[i].mask[4] = 0xC0; /* IPv4 TIP */
3198                 pmConf->filters[i].mask[5] = 0x03; /* IPv4 TIP */
3199                 in_dev_put(in_dev);
3200
3201                 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_FILTER;
3202                 i++;
3203         }
3204
3205 skip_arp:
3206         if (adapter->wol & WAKE_MAGIC)
3207                 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_MAGIC;
3208
3209         pmConf->numFilters = i;
3210
3211         adapter->shared->devRead.pmConfDesc.confVer = cpu_to_le32(1);
3212         adapter->shared->devRead.pmConfDesc.confLen = cpu_to_le32(sizeof(
3213                                                                   *pmConf));
3214         adapter->shared->devRead.pmConfDesc.confPA = cpu_to_le64(virt_to_phys(
3215                                                                  pmConf));
3216
3217         spin_lock_irqsave(&adapter->cmd_lock, flags);
3218         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3219                                VMXNET3_CMD_UPDATE_PMCFG);
3220         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3221
3222         pci_save_state(pdev);
3223         pci_enable_wake(pdev, pci_choose_state(pdev, PMSG_SUSPEND),
3224                         adapter->wol);
3225         pci_disable_device(pdev);
3226         pci_set_power_state(pdev, pci_choose_state(pdev, PMSG_SUSPEND));
3227
3228         return 0;
3229 }
3230
3231
3232 static int
3233 vmxnet3_resume(struct device *device)
3234 {
3235         int err, i = 0;
3236         unsigned long flags;
3237         struct pci_dev *pdev = to_pci_dev(device);
3238         struct net_device *netdev = pci_get_drvdata(pdev);
3239         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3240         struct Vmxnet3_PMConf *pmConf;
3241
3242         if (!netif_running(netdev))
3243                 return 0;
3244
3245         /* Destroy wake-up filters. */
3246         pmConf = adapter->pm_conf;
3247         memset(pmConf, 0, sizeof(*pmConf));
3248
3249         adapter->shared->devRead.pmConfDesc.confVer = cpu_to_le32(1);
3250         adapter->shared->devRead.pmConfDesc.confLen = cpu_to_le32(sizeof(
3251                                                                   *pmConf));
3252         adapter->shared->devRead.pmConfDesc.confPA = cpu_to_le64(virt_to_phys(
3253                                                                  pmConf));
3254
3255         netif_device_attach(netdev);
3256         pci_set_power_state(pdev, PCI_D0);
3257         pci_restore_state(pdev);
3258         err = pci_enable_device_mem(pdev);
3259         if (err != 0)
3260                 return err;
3261
3262         pci_enable_wake(pdev, PCI_D0, 0);
3263
3264         spin_lock_irqsave(&adapter->cmd_lock, flags);
3265         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3266                                VMXNET3_CMD_UPDATE_PMCFG);
3267         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3268         vmxnet3_alloc_intr_resources(adapter);
3269         vmxnet3_request_irqs(adapter);
3270         for (i = 0; i < adapter->num_rx_queues; i++)
3271                 napi_enable(&adapter->rx_queue[i].napi);
3272         vmxnet3_enable_all_intrs(adapter);
3273
3274         return 0;
3275 }
3276
3277 static const struct dev_pm_ops vmxnet3_pm_ops = {
3278         .suspend = vmxnet3_suspend,
3279         .resume = vmxnet3_resume,
3280 };
3281 #endif
3282
3283 static struct pci_driver vmxnet3_driver = {
3284         .name           = vmxnet3_driver_name,
3285         .id_table       = vmxnet3_pciid_table,
3286         .probe          = vmxnet3_probe_device,
3287         .remove         = __devexit_p(vmxnet3_remove_device),
3288 #ifdef CONFIG_PM
3289         .driver.pm      = &vmxnet3_pm_ops,
3290 #endif
3291 };
3292
3293
3294 static int __init
3295 vmxnet3_init_module(void)
3296 {
3297         printk(KERN_INFO "%s - version %s\n", VMXNET3_DRIVER_DESC,
3298                 VMXNET3_DRIVER_VERSION_REPORT);
3299         return pci_register_driver(&vmxnet3_driver);
3300 }
3301
3302 module_init(vmxnet3_init_module);
3303
3304
3305 static void
3306 vmxnet3_exit_module(void)
3307 {
3308         pci_unregister_driver(&vmxnet3_driver);
3309 }
3310
3311 module_exit(vmxnet3_exit_module);
3312
3313 MODULE_AUTHOR("VMware, Inc.");
3314 MODULE_DESCRIPTION(VMXNET3_DRIVER_DESC);
3315 MODULE_LICENSE("GPL v2");
3316 MODULE_VERSION(VMXNET3_DRIVER_VERSION_STRING);