vmxnet3: Preserve the MAC address configured by ifconfig
[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
146         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, VMXNET3_CMD_GET_LINK);
147         ret = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
148         adapter->link_speed = ret >> 16;
149         if (ret & 1) { /* Link is up. */
150                 printk(KERN_INFO "%s: NIC Link is Up %d Mbps\n",
151                        adapter->netdev->name, adapter->link_speed);
152                 if (!netif_carrier_ok(adapter->netdev))
153                         netif_carrier_on(adapter->netdev);
154
155                 if (affectTxQueue) {
156                         for (i = 0; i < adapter->num_tx_queues; i++)
157                                 vmxnet3_tq_start(&adapter->tx_queue[i],
158                                                  adapter);
159                 }
160         } else {
161                 printk(KERN_INFO "%s: NIC Link is Down\n",
162                        adapter->netdev->name);
163                 if (netif_carrier_ok(adapter->netdev))
164                         netif_carrier_off(adapter->netdev);
165
166                 if (affectTxQueue) {
167                         for (i = 0; i < adapter->num_tx_queues; i++)
168                                 vmxnet3_tq_stop(&adapter->tx_queue[i], adapter);
169                 }
170         }
171 }
172
173 static void
174 vmxnet3_process_events(struct vmxnet3_adapter *adapter)
175 {
176         int i;
177         u32 events = le32_to_cpu(adapter->shared->ecr);
178         if (!events)
179                 return;
180
181         vmxnet3_ack_events(adapter, events);
182
183         /* Check if link state has changed */
184         if (events & VMXNET3_ECR_LINK)
185                 vmxnet3_check_link(adapter, true);
186
187         /* Check if there is an error on xmit/recv queues */
188         if (events & (VMXNET3_ECR_TQERR | VMXNET3_ECR_RQERR)) {
189                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
190                                        VMXNET3_CMD_GET_QUEUE_STATUS);
191
192                 for (i = 0; i < adapter->num_tx_queues; i++)
193                         if (adapter->tqd_start[i].status.stopped)
194                                 dev_err(&adapter->netdev->dev,
195                                         "%s: tq[%d] error 0x%x\n",
196                                         adapter->netdev->name, i, le32_to_cpu(
197                                         adapter->tqd_start[i].status.error));
198                 for (i = 0; i < adapter->num_rx_queues; i++)
199                         if (adapter->rqd_start[i].status.stopped)
200                                 dev_err(&adapter->netdev->dev,
201                                         "%s: rq[%d] error 0x%x\n",
202                                         adapter->netdev->name, i,
203                                         adapter->rqd_start[i].status.error);
204
205                 schedule_work(&adapter->work);
206         }
207 }
208
209 #ifdef __BIG_ENDIAN_BITFIELD
210 /*
211  * The device expects the bitfields in shared structures to be written in
212  * little endian. When CPU is big endian, the following routines are used to
213  * correctly read and write into ABI.
214  * The general technique used here is : double word bitfields are defined in
215  * opposite order for big endian architecture. Then before reading them in
216  * driver the complete double word is translated using le32_to_cpu. Similarly
217  * After the driver writes into bitfields, cpu_to_le32 is used to translate the
218  * double words into required format.
219  * In order to avoid touching bits in shared structure more than once, temporary
220  * descriptors are used. These are passed as srcDesc to following functions.
221  */
222 static void vmxnet3_RxDescToCPU(const struct Vmxnet3_RxDesc *srcDesc,
223                                 struct Vmxnet3_RxDesc *dstDesc)
224 {
225         u32 *src = (u32 *)srcDesc + 2;
226         u32 *dst = (u32 *)dstDesc + 2;
227         dstDesc->addr = le64_to_cpu(srcDesc->addr);
228         *dst = le32_to_cpu(*src);
229         dstDesc->ext1 = le32_to_cpu(srcDesc->ext1);
230 }
231
232 static void vmxnet3_TxDescToLe(const struct Vmxnet3_TxDesc *srcDesc,
233                                struct Vmxnet3_TxDesc *dstDesc)
234 {
235         int i;
236         u32 *src = (u32 *)(srcDesc + 1);
237         u32 *dst = (u32 *)(dstDesc + 1);
238
239         /* Working backwards so that the gen bit is set at the end. */
240         for (i = 2; i > 0; i--) {
241                 src--;
242                 dst--;
243                 *dst = cpu_to_le32(*src);
244         }
245 }
246
247
248 static void vmxnet3_RxCompToCPU(const struct Vmxnet3_RxCompDesc *srcDesc,
249                                 struct Vmxnet3_RxCompDesc *dstDesc)
250 {
251         int i = 0;
252         u32 *src = (u32 *)srcDesc;
253         u32 *dst = (u32 *)dstDesc;
254         for (i = 0; i < sizeof(struct Vmxnet3_RxCompDesc) / sizeof(u32); i++) {
255                 *dst = le32_to_cpu(*src);
256                 src++;
257                 dst++;
258         }
259 }
260
261
262 /* Used to read bitfield values from double words. */
263 static u32 get_bitfield32(const __le32 *bitfield, u32 pos, u32 size)
264 {
265         u32 temp = le32_to_cpu(*bitfield);
266         u32 mask = ((1 << size) - 1) << pos;
267         temp &= mask;
268         temp >>= pos;
269         return temp;
270 }
271
272
273
274 #endif  /* __BIG_ENDIAN_BITFIELD */
275
276 #ifdef __BIG_ENDIAN_BITFIELD
277
278 #   define VMXNET3_TXDESC_GET_GEN(txdesc) get_bitfield32(((const __le32 *) \
279                         txdesc) + VMXNET3_TXD_GEN_DWORD_SHIFT, \
280                         VMXNET3_TXD_GEN_SHIFT, VMXNET3_TXD_GEN_SIZE)
281 #   define VMXNET3_TXDESC_GET_EOP(txdesc) get_bitfield32(((const __le32 *) \
282                         txdesc) + VMXNET3_TXD_EOP_DWORD_SHIFT, \
283                         VMXNET3_TXD_EOP_SHIFT, VMXNET3_TXD_EOP_SIZE)
284 #   define VMXNET3_TCD_GET_GEN(tcd) get_bitfield32(((const __le32 *)tcd) + \
285                         VMXNET3_TCD_GEN_DWORD_SHIFT, VMXNET3_TCD_GEN_SHIFT, \
286                         VMXNET3_TCD_GEN_SIZE)
287 #   define VMXNET3_TCD_GET_TXIDX(tcd) get_bitfield32((const __le32 *)tcd, \
288                         VMXNET3_TCD_TXIDX_SHIFT, VMXNET3_TCD_TXIDX_SIZE)
289 #   define vmxnet3_getRxComp(dstrcd, rcd, tmp) do { \
290                         (dstrcd) = (tmp); \
291                         vmxnet3_RxCompToCPU((rcd), (tmp)); \
292                 } while (0)
293 #   define vmxnet3_getRxDesc(dstrxd, rxd, tmp) do { \
294                         (dstrxd) = (tmp); \
295                         vmxnet3_RxDescToCPU((rxd), (tmp)); \
296                 } while (0)
297
298 #else
299
300 #   define VMXNET3_TXDESC_GET_GEN(txdesc) ((txdesc)->gen)
301 #   define VMXNET3_TXDESC_GET_EOP(txdesc) ((txdesc)->eop)
302 #   define VMXNET3_TCD_GET_GEN(tcd) ((tcd)->gen)
303 #   define VMXNET3_TCD_GET_TXIDX(tcd) ((tcd)->txdIdx)
304 #   define vmxnet3_getRxComp(dstrcd, rcd, tmp) (dstrcd) = (rcd)
305 #   define vmxnet3_getRxDesc(dstrxd, rxd, tmp) (dstrxd) = (rxd)
306
307 #endif /* __BIG_ENDIAN_BITFIELD  */
308
309
310 static void
311 vmxnet3_unmap_tx_buf(struct vmxnet3_tx_buf_info *tbi,
312                      struct pci_dev *pdev)
313 {
314         if (tbi->map_type == VMXNET3_MAP_SINGLE)
315                 pci_unmap_single(pdev, tbi->dma_addr, tbi->len,
316                                  PCI_DMA_TODEVICE);
317         else if (tbi->map_type == VMXNET3_MAP_PAGE)
318                 pci_unmap_page(pdev, tbi->dma_addr, tbi->len,
319                                PCI_DMA_TODEVICE);
320         else
321                 BUG_ON(tbi->map_type != VMXNET3_MAP_NONE);
322
323         tbi->map_type = VMXNET3_MAP_NONE; /* to help debugging */
324 }
325
326
327 static int
328 vmxnet3_unmap_pkt(u32 eop_idx, struct vmxnet3_tx_queue *tq,
329                   struct pci_dev *pdev, struct vmxnet3_adapter *adapter)
330 {
331         struct sk_buff *skb;
332         int entries = 0;
333
334         /* no out of order completion */
335         BUG_ON(tq->buf_info[eop_idx].sop_idx != tq->tx_ring.next2comp);
336         BUG_ON(VMXNET3_TXDESC_GET_EOP(&(tq->tx_ring.base[eop_idx].txd)) != 1);
337
338         skb = tq->buf_info[eop_idx].skb;
339         BUG_ON(skb == NULL);
340         tq->buf_info[eop_idx].skb = NULL;
341
342         VMXNET3_INC_RING_IDX_ONLY(eop_idx, tq->tx_ring.size);
343
344         while (tq->tx_ring.next2comp != eop_idx) {
345                 vmxnet3_unmap_tx_buf(tq->buf_info + tq->tx_ring.next2comp,
346                                      pdev);
347
348                 /* update next2comp w/o tx_lock. Since we are marking more,
349                  * instead of less, tx ring entries avail, the worst case is
350                  * that the tx routine incorrectly re-queues a pkt due to
351                  * insufficient tx ring entries.
352                  */
353                 vmxnet3_cmd_ring_adv_next2comp(&tq->tx_ring);
354                 entries++;
355         }
356
357         dev_kfree_skb_any(skb);
358         return entries;
359 }
360
361
362 static int
363 vmxnet3_tq_tx_complete(struct vmxnet3_tx_queue *tq,
364                         struct vmxnet3_adapter *adapter)
365 {
366         int completed = 0;
367         union Vmxnet3_GenericDesc *gdesc;
368
369         gdesc = tq->comp_ring.base + tq->comp_ring.next2proc;
370         while (VMXNET3_TCD_GET_GEN(&gdesc->tcd) == tq->comp_ring.gen) {
371                 completed += vmxnet3_unmap_pkt(VMXNET3_TCD_GET_TXIDX(
372                                                &gdesc->tcd), tq, adapter->pdev,
373                                                adapter);
374
375                 vmxnet3_comp_ring_adv_next2proc(&tq->comp_ring);
376                 gdesc = tq->comp_ring.base + tq->comp_ring.next2proc;
377         }
378
379         if (completed) {
380                 spin_lock(&tq->tx_lock);
381                 if (unlikely(vmxnet3_tq_stopped(tq, adapter) &&
382                              vmxnet3_cmd_ring_desc_avail(&tq->tx_ring) >
383                              VMXNET3_WAKE_QUEUE_THRESHOLD(tq) &&
384                              netif_carrier_ok(adapter->netdev))) {
385                         vmxnet3_tq_wake(tq, adapter);
386                 }
387                 spin_unlock(&tq->tx_lock);
388         }
389         return completed;
390 }
391
392
393 static void
394 vmxnet3_tq_cleanup(struct vmxnet3_tx_queue *tq,
395                    struct vmxnet3_adapter *adapter)
396 {
397         int i;
398
399         while (tq->tx_ring.next2comp != tq->tx_ring.next2fill) {
400                 struct vmxnet3_tx_buf_info *tbi;
401                 union Vmxnet3_GenericDesc *gdesc;
402
403                 tbi = tq->buf_info + tq->tx_ring.next2comp;
404                 gdesc = tq->tx_ring.base + tq->tx_ring.next2comp;
405
406                 vmxnet3_unmap_tx_buf(tbi, adapter->pdev);
407                 if (tbi->skb) {
408                         dev_kfree_skb_any(tbi->skb);
409                         tbi->skb = NULL;
410                 }
411                 vmxnet3_cmd_ring_adv_next2comp(&tq->tx_ring);
412         }
413
414         /* sanity check, verify all buffers are indeed unmapped and freed */
415         for (i = 0; i < tq->tx_ring.size; i++) {
416                 BUG_ON(tq->buf_info[i].skb != NULL ||
417                        tq->buf_info[i].map_type != VMXNET3_MAP_NONE);
418         }
419
420         tq->tx_ring.gen = VMXNET3_INIT_GEN;
421         tq->tx_ring.next2fill = tq->tx_ring.next2comp = 0;
422
423         tq->comp_ring.gen = VMXNET3_INIT_GEN;
424         tq->comp_ring.next2proc = 0;
425 }
426
427
428 static void
429 vmxnet3_tq_destroy(struct vmxnet3_tx_queue *tq,
430                    struct vmxnet3_adapter *adapter)
431 {
432         if (tq->tx_ring.base) {
433                 pci_free_consistent(adapter->pdev, tq->tx_ring.size *
434                                     sizeof(struct Vmxnet3_TxDesc),
435                                     tq->tx_ring.base, tq->tx_ring.basePA);
436                 tq->tx_ring.base = NULL;
437         }
438         if (tq->data_ring.base) {
439                 pci_free_consistent(adapter->pdev, tq->data_ring.size *
440                                     sizeof(struct Vmxnet3_TxDataDesc),
441                                     tq->data_ring.base, tq->data_ring.basePA);
442                 tq->data_ring.base = NULL;
443         }
444         if (tq->comp_ring.base) {
445                 pci_free_consistent(adapter->pdev, tq->comp_ring.size *
446                                     sizeof(struct Vmxnet3_TxCompDesc),
447                                     tq->comp_ring.base, tq->comp_ring.basePA);
448                 tq->comp_ring.base = NULL;
449         }
450         kfree(tq->buf_info);
451         tq->buf_info = NULL;
452 }
453
454
455 /* Destroy all tx queues */
456 void
457 vmxnet3_tq_destroy_all(struct vmxnet3_adapter *adapter)
458 {
459         int i;
460
461         for (i = 0; i < adapter->num_tx_queues; i++)
462                 vmxnet3_tq_destroy(&adapter->tx_queue[i], adapter);
463 }
464
465
466 static void
467 vmxnet3_tq_init(struct vmxnet3_tx_queue *tq,
468                 struct vmxnet3_adapter *adapter)
469 {
470         int i;
471
472         /* reset the tx ring contents to 0 and reset the tx ring states */
473         memset(tq->tx_ring.base, 0, tq->tx_ring.size *
474                sizeof(struct Vmxnet3_TxDesc));
475         tq->tx_ring.next2fill = tq->tx_ring.next2comp = 0;
476         tq->tx_ring.gen = VMXNET3_INIT_GEN;
477
478         memset(tq->data_ring.base, 0, tq->data_ring.size *
479                sizeof(struct Vmxnet3_TxDataDesc));
480
481         /* reset the tx comp ring contents to 0 and reset comp ring states */
482         memset(tq->comp_ring.base, 0, tq->comp_ring.size *
483                sizeof(struct Vmxnet3_TxCompDesc));
484         tq->comp_ring.next2proc = 0;
485         tq->comp_ring.gen = VMXNET3_INIT_GEN;
486
487         /* reset the bookkeeping data */
488         memset(tq->buf_info, 0, sizeof(tq->buf_info[0]) * tq->tx_ring.size);
489         for (i = 0; i < tq->tx_ring.size; i++)
490                 tq->buf_info[i].map_type = VMXNET3_MAP_NONE;
491
492         /* stats are not reset */
493 }
494
495
496 static int
497 vmxnet3_tq_create(struct vmxnet3_tx_queue *tq,
498                   struct vmxnet3_adapter *adapter)
499 {
500         BUG_ON(tq->tx_ring.base || tq->data_ring.base ||
501                tq->comp_ring.base || tq->buf_info);
502
503         tq->tx_ring.base = pci_alloc_consistent(adapter->pdev, tq->tx_ring.size
504                            * sizeof(struct Vmxnet3_TxDesc),
505                            &tq->tx_ring.basePA);
506         if (!tq->tx_ring.base) {
507                 printk(KERN_ERR "%s: failed to allocate tx ring\n",
508                        adapter->netdev->name);
509                 goto err;
510         }
511
512         tq->data_ring.base = pci_alloc_consistent(adapter->pdev,
513                              tq->data_ring.size *
514                              sizeof(struct Vmxnet3_TxDataDesc),
515                              &tq->data_ring.basePA);
516         if (!tq->data_ring.base) {
517                 printk(KERN_ERR "%s: failed to allocate data ring\n",
518                        adapter->netdev->name);
519                 goto err;
520         }
521
522         tq->comp_ring.base = pci_alloc_consistent(adapter->pdev,
523                              tq->comp_ring.size *
524                              sizeof(struct Vmxnet3_TxCompDesc),
525                              &tq->comp_ring.basePA);
526         if (!tq->comp_ring.base) {
527                 printk(KERN_ERR "%s: failed to allocate tx comp ring\n",
528                        adapter->netdev->name);
529                 goto err;
530         }
531
532         tq->buf_info = kcalloc(tq->tx_ring.size, sizeof(tq->buf_info[0]),
533                                GFP_KERNEL);
534         if (!tq->buf_info) {
535                 printk(KERN_ERR "%s: failed to allocate tx bufinfo\n",
536                        adapter->netdev->name);
537                 goto err;
538         }
539
540         return 0;
541
542 err:
543         vmxnet3_tq_destroy(tq, adapter);
544         return -ENOMEM;
545 }
546
547 static void
548 vmxnet3_tq_cleanup_all(struct vmxnet3_adapter *adapter)
549 {
550         int i;
551
552         for (i = 0; i < adapter->num_tx_queues; i++)
553                 vmxnet3_tq_cleanup(&adapter->tx_queue[i], adapter);
554 }
555
556 /*
557  *    starting from ring->next2fill, allocate rx buffers for the given ring
558  *    of the rx queue and update the rx desc. stop after @num_to_alloc buffers
559  *    are allocated or allocation fails
560  */
561
562 static int
563 vmxnet3_rq_alloc_rx_buf(struct vmxnet3_rx_queue *rq, u32 ring_idx,
564                         int num_to_alloc, struct vmxnet3_adapter *adapter)
565 {
566         int num_allocated = 0;
567         struct vmxnet3_rx_buf_info *rbi_base = rq->buf_info[ring_idx];
568         struct vmxnet3_cmd_ring *ring = &rq->rx_ring[ring_idx];
569         u32 val;
570
571         while (num_allocated < num_to_alloc) {
572                 struct vmxnet3_rx_buf_info *rbi;
573                 union Vmxnet3_GenericDesc *gd;
574
575                 rbi = rbi_base + ring->next2fill;
576                 gd = ring->base + ring->next2fill;
577
578                 if (rbi->buf_type == VMXNET3_RX_BUF_SKB) {
579                         if (rbi->skb == NULL) {
580                                 rbi->skb = dev_alloc_skb(rbi->len +
581                                                          NET_IP_ALIGN);
582                                 if (unlikely(rbi->skb == NULL)) {
583                                         rq->stats.rx_buf_alloc_failure++;
584                                         break;
585                                 }
586                                 rbi->skb->dev = adapter->netdev;
587
588                                 skb_reserve(rbi->skb, NET_IP_ALIGN);
589                                 rbi->dma_addr = pci_map_single(adapter->pdev,
590                                                 rbi->skb->data, rbi->len,
591                                                 PCI_DMA_FROMDEVICE);
592                         } else {
593                                 /* rx buffer skipped by the device */
594                         }
595                         val = VMXNET3_RXD_BTYPE_HEAD << VMXNET3_RXD_BTYPE_SHIFT;
596                 } else {
597                         BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_PAGE ||
598                                rbi->len  != PAGE_SIZE);
599
600                         if (rbi->page == NULL) {
601                                 rbi->page = alloc_page(GFP_ATOMIC);
602                                 if (unlikely(rbi->page == NULL)) {
603                                         rq->stats.rx_buf_alloc_failure++;
604                                         break;
605                                 }
606                                 rbi->dma_addr = pci_map_page(adapter->pdev,
607                                                 rbi->page, 0, PAGE_SIZE,
608                                                 PCI_DMA_FROMDEVICE);
609                         } else {
610                                 /* rx buffers skipped by the device */
611                         }
612                         val = VMXNET3_RXD_BTYPE_BODY << VMXNET3_RXD_BTYPE_SHIFT;
613                 }
614
615                 BUG_ON(rbi->dma_addr == 0);
616                 gd->rxd.addr = cpu_to_le64(rbi->dma_addr);
617                 gd->dword[2] = cpu_to_le32((ring->gen << VMXNET3_RXD_GEN_SHIFT)
618                                            | val | rbi->len);
619
620                 num_allocated++;
621                 vmxnet3_cmd_ring_adv_next2fill(ring);
622         }
623         rq->uncommitted[ring_idx] += num_allocated;
624
625         dev_dbg(&adapter->netdev->dev,
626                 "alloc_rx_buf: %d allocated, next2fill %u, next2comp "
627                 "%u, uncommited %u\n", num_allocated, ring->next2fill,
628                 ring->next2comp, rq->uncommitted[ring_idx]);
629
630         /* so that the device can distinguish a full ring and an empty ring */
631         BUG_ON(num_allocated != 0 && ring->next2fill == ring->next2comp);
632
633         return num_allocated;
634 }
635
636
637 static void
638 vmxnet3_append_frag(struct sk_buff *skb, struct Vmxnet3_RxCompDesc *rcd,
639                     struct vmxnet3_rx_buf_info *rbi)
640 {
641         struct skb_frag_struct *frag = skb_shinfo(skb)->frags +
642                 skb_shinfo(skb)->nr_frags;
643
644         BUG_ON(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS);
645
646         frag->page = rbi->page;
647         frag->page_offset = 0;
648         frag->size = rcd->len;
649         skb->data_len += frag->size;
650         skb_shinfo(skb)->nr_frags++;
651 }
652
653
654 static void
655 vmxnet3_map_pkt(struct sk_buff *skb, struct vmxnet3_tx_ctx *ctx,
656                 struct vmxnet3_tx_queue *tq, struct pci_dev *pdev,
657                 struct vmxnet3_adapter *adapter)
658 {
659         u32 dw2, len;
660         unsigned long buf_offset;
661         int i;
662         union Vmxnet3_GenericDesc *gdesc;
663         struct vmxnet3_tx_buf_info *tbi = NULL;
664
665         BUG_ON(ctx->copy_size > skb_headlen(skb));
666
667         /* use the previous gen bit for the SOP desc */
668         dw2 = (tq->tx_ring.gen ^ 0x1) << VMXNET3_TXD_GEN_SHIFT;
669
670         ctx->sop_txd = tq->tx_ring.base + tq->tx_ring.next2fill;
671         gdesc = ctx->sop_txd; /* both loops below can be skipped */
672
673         /* no need to map the buffer if headers are copied */
674         if (ctx->copy_size) {
675                 ctx->sop_txd->txd.addr = cpu_to_le64(tq->data_ring.basePA +
676                                         tq->tx_ring.next2fill *
677                                         sizeof(struct Vmxnet3_TxDataDesc));
678                 ctx->sop_txd->dword[2] = cpu_to_le32(dw2 | ctx->copy_size);
679                 ctx->sop_txd->dword[3] = 0;
680
681                 tbi = tq->buf_info + tq->tx_ring.next2fill;
682                 tbi->map_type = VMXNET3_MAP_NONE;
683
684                 dev_dbg(&adapter->netdev->dev,
685                         "txd[%u]: 0x%Lx 0x%x 0x%x\n",
686                         tq->tx_ring.next2fill,
687                         le64_to_cpu(ctx->sop_txd->txd.addr),
688                         ctx->sop_txd->dword[2], ctx->sop_txd->dword[3]);
689                 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
690
691                 /* use the right gen for non-SOP desc */
692                 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
693         }
694
695         /* linear part can use multiple tx desc if it's big */
696         len = skb_headlen(skb) - ctx->copy_size;
697         buf_offset = ctx->copy_size;
698         while (len) {
699                 u32 buf_size;
700
701                 if (len < VMXNET3_MAX_TX_BUF_SIZE) {
702                         buf_size = len;
703                         dw2 |= len;
704                 } else {
705                         buf_size = VMXNET3_MAX_TX_BUF_SIZE;
706                         /* spec says that for TxDesc.len, 0 == 2^14 */
707                 }
708
709                 tbi = tq->buf_info + tq->tx_ring.next2fill;
710                 tbi->map_type = VMXNET3_MAP_SINGLE;
711                 tbi->dma_addr = pci_map_single(adapter->pdev,
712                                 skb->data + buf_offset, buf_size,
713                                 PCI_DMA_TODEVICE);
714
715                 tbi->len = buf_size;
716
717                 gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
718                 BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
719
720                 gdesc->txd.addr = cpu_to_le64(tbi->dma_addr);
721                 gdesc->dword[2] = cpu_to_le32(dw2);
722                 gdesc->dword[3] = 0;
723
724                 dev_dbg(&adapter->netdev->dev,
725                         "txd[%u]: 0x%Lx 0x%x 0x%x\n",
726                         tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr),
727                         le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]);
728                 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
729                 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
730
731                 len -= buf_size;
732                 buf_offset += buf_size;
733         }
734
735         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
736                 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
737
738                 tbi = tq->buf_info + tq->tx_ring.next2fill;
739                 tbi->map_type = VMXNET3_MAP_PAGE;
740                 tbi->dma_addr = pci_map_page(adapter->pdev, frag->page,
741                                              frag->page_offset, frag->size,
742                                              PCI_DMA_TODEVICE);
743
744                 tbi->len = frag->size;
745
746                 gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
747                 BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
748
749                 gdesc->txd.addr = cpu_to_le64(tbi->dma_addr);
750                 gdesc->dword[2] = cpu_to_le32(dw2 | frag->size);
751                 gdesc->dword[3] = 0;
752
753                 dev_dbg(&adapter->netdev->dev,
754                         "txd[%u]: 0x%llu %u %u\n",
755                         tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr),
756                         le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]);
757                 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
758                 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
759         }
760
761         ctx->eop_txd = gdesc;
762
763         /* set the last buf_info for the pkt */
764         tbi->skb = skb;
765         tbi->sop_idx = ctx->sop_txd - tq->tx_ring.base;
766 }
767
768
769 /* Init all tx queues */
770 static void
771 vmxnet3_tq_init_all(struct vmxnet3_adapter *adapter)
772 {
773         int i;
774
775         for (i = 0; i < adapter->num_tx_queues; i++)
776                 vmxnet3_tq_init(&adapter->tx_queue[i], adapter);
777 }
778
779
780 /*
781  *    parse and copy relevant protocol headers:
782  *      For a tso pkt, relevant headers are L2/3/4 including options
783  *      For a pkt requesting csum offloading, they are L2/3 and may include L4
784  *      if it's a TCP/UDP pkt
785  *
786  * Returns:
787  *    -1:  error happens during parsing
788  *     0:  protocol headers parsed, but too big to be copied
789  *     1:  protocol headers parsed and copied
790  *
791  * Other effects:
792  *    1. related *ctx fields are updated.
793  *    2. ctx->copy_size is # of bytes copied
794  *    3. the portion copied is guaranteed to be in the linear part
795  *
796  */
797 static int
798 vmxnet3_parse_and_copy_hdr(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
799                            struct vmxnet3_tx_ctx *ctx,
800                            struct vmxnet3_adapter *adapter)
801 {
802         struct Vmxnet3_TxDataDesc *tdd;
803
804         if (ctx->mss) { /* TSO */
805                 ctx->eth_ip_hdr_size = skb_transport_offset(skb);
806                 ctx->l4_hdr_size = ((struct tcphdr *)
807                                    skb_transport_header(skb))->doff * 4;
808                 ctx->copy_size = ctx->eth_ip_hdr_size + ctx->l4_hdr_size;
809         } else {
810                 unsigned int pull_size;
811
812                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
813                         ctx->eth_ip_hdr_size = skb_checksum_start_offset(skb);
814
815                         if (ctx->ipv4) {
816                                 struct iphdr *iph = (struct iphdr *)
817                                                     skb_network_header(skb);
818                                 if (iph->protocol == IPPROTO_TCP) {
819                                         pull_size = ctx->eth_ip_hdr_size +
820                                                     sizeof(struct tcphdr);
821
822                                         if (unlikely(!pskb_may_pull(skb,
823                                                                 pull_size))) {
824                                                 goto err;
825                                         }
826                                         ctx->l4_hdr_size = ((struct tcphdr *)
827                                            skb_transport_header(skb))->doff * 4;
828                                 } else if (iph->protocol == IPPROTO_UDP) {
829                                         ctx->l4_hdr_size =
830                                                         sizeof(struct udphdr);
831                                 } else {
832                                         ctx->l4_hdr_size = 0;
833                                 }
834                         } else {
835                                 /* for simplicity, don't copy L4 headers */
836                                 ctx->l4_hdr_size = 0;
837                         }
838                         ctx->copy_size = ctx->eth_ip_hdr_size +
839                                          ctx->l4_hdr_size;
840                 } else {
841                         ctx->eth_ip_hdr_size = 0;
842                         ctx->l4_hdr_size = 0;
843                         /* copy as much as allowed */
844                         ctx->copy_size = min((unsigned int)VMXNET3_HDR_COPY_SIZE
845                                              , skb_headlen(skb));
846                 }
847
848                 /* make sure headers are accessible directly */
849                 if (unlikely(!pskb_may_pull(skb, ctx->copy_size)))
850                         goto err;
851         }
852
853         if (unlikely(ctx->copy_size > VMXNET3_HDR_COPY_SIZE)) {
854                 tq->stats.oversized_hdr++;
855                 ctx->copy_size = 0;
856                 return 0;
857         }
858
859         tdd = tq->data_ring.base + tq->tx_ring.next2fill;
860
861         memcpy(tdd->data, skb->data, ctx->copy_size);
862         dev_dbg(&adapter->netdev->dev,
863                 "copy %u bytes to dataRing[%u]\n",
864                 ctx->copy_size, tq->tx_ring.next2fill);
865         return 1;
866
867 err:
868         return -1;
869 }
870
871
872 static void
873 vmxnet3_prepare_tso(struct sk_buff *skb,
874                     struct vmxnet3_tx_ctx *ctx)
875 {
876         struct tcphdr *tcph = (struct tcphdr *)skb_transport_header(skb);
877         if (ctx->ipv4) {
878                 struct iphdr *iph = (struct iphdr *)skb_network_header(skb);
879                 iph->check = 0;
880                 tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, 0,
881                                                  IPPROTO_TCP, 0);
882         } else {
883                 struct ipv6hdr *iph = (struct ipv6hdr *)skb_network_header(skb);
884                 tcph->check = ~csum_ipv6_magic(&iph->saddr, &iph->daddr, 0,
885                                                IPPROTO_TCP, 0);
886         }
887 }
888
889
890 /*
891  * Transmits a pkt thru a given tq
892  * Returns:
893  *    NETDEV_TX_OK:      descriptors are setup successfully
894  *    NETDEV_TX_OK:      error occured, the pkt is dropped
895  *    NETDEV_TX_BUSY:    tx ring is full, queue is stopped
896  *
897  * Side-effects:
898  *    1. tx ring may be changed
899  *    2. tq stats may be updated accordingly
900  *    3. shared->txNumDeferred may be updated
901  */
902
903 static int
904 vmxnet3_tq_xmit(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
905                 struct vmxnet3_adapter *adapter, struct net_device *netdev)
906 {
907         int ret;
908         u32 count;
909         unsigned long flags;
910         struct vmxnet3_tx_ctx ctx;
911         union Vmxnet3_GenericDesc *gdesc;
912 #ifdef __BIG_ENDIAN_BITFIELD
913         /* Use temporary descriptor to avoid touching bits multiple times */
914         union Vmxnet3_GenericDesc tempTxDesc;
915 #endif
916
917         /* conservatively estimate # of descriptors to use */
918         count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) +
919                 skb_shinfo(skb)->nr_frags + 1;
920
921         ctx.ipv4 = (skb->protocol == cpu_to_be16(ETH_P_IP));
922
923         ctx.mss = skb_shinfo(skb)->gso_size;
924         if (ctx.mss) {
925                 if (skb_header_cloned(skb)) {
926                         if (unlikely(pskb_expand_head(skb, 0, 0,
927                                                       GFP_ATOMIC) != 0)) {
928                                 tq->stats.drop_tso++;
929                                 goto drop_pkt;
930                         }
931                         tq->stats.copy_skb_header++;
932                 }
933                 vmxnet3_prepare_tso(skb, &ctx);
934         } else {
935                 if (unlikely(count > VMXNET3_MAX_TXD_PER_PKT)) {
936
937                         /* non-tso pkts must not use more than
938                          * VMXNET3_MAX_TXD_PER_PKT entries
939                          */
940                         if (skb_linearize(skb) != 0) {
941                                 tq->stats.drop_too_many_frags++;
942                                 goto drop_pkt;
943                         }
944                         tq->stats.linearized++;
945
946                         /* recalculate the # of descriptors to use */
947                         count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) + 1;
948                 }
949         }
950
951         spin_lock_irqsave(&tq->tx_lock, flags);
952
953         if (count > vmxnet3_cmd_ring_desc_avail(&tq->tx_ring)) {
954                 tq->stats.tx_ring_full++;
955                 dev_dbg(&adapter->netdev->dev,
956                         "tx queue stopped on %s, next2comp %u"
957                         " next2fill %u\n", adapter->netdev->name,
958                         tq->tx_ring.next2comp, tq->tx_ring.next2fill);
959
960                 vmxnet3_tq_stop(tq, adapter);
961                 spin_unlock_irqrestore(&tq->tx_lock, flags);
962                 return NETDEV_TX_BUSY;
963         }
964
965
966         ret = vmxnet3_parse_and_copy_hdr(skb, tq, &ctx, adapter);
967         if (ret >= 0) {
968                 BUG_ON(ret <= 0 && ctx.copy_size != 0);
969                 /* hdrs parsed, check against other limits */
970                 if (ctx.mss) {
971                         if (unlikely(ctx.eth_ip_hdr_size + ctx.l4_hdr_size >
972                                      VMXNET3_MAX_TX_BUF_SIZE)) {
973                                 goto hdr_too_big;
974                         }
975                 } else {
976                         if (skb->ip_summed == CHECKSUM_PARTIAL) {
977                                 if (unlikely(ctx.eth_ip_hdr_size +
978                                              skb->csum_offset >
979                                              VMXNET3_MAX_CSUM_OFFSET)) {
980                                         goto hdr_too_big;
981                                 }
982                         }
983                 }
984         } else {
985                 tq->stats.drop_hdr_inspect_err++;
986                 goto unlock_drop_pkt;
987         }
988
989         /* fill tx descs related to addr & len */
990         vmxnet3_map_pkt(skb, &ctx, tq, adapter->pdev, adapter);
991
992         /* setup the EOP desc */
993         ctx.eop_txd->dword[3] = cpu_to_le32(VMXNET3_TXD_CQ | VMXNET3_TXD_EOP);
994
995         /* setup the SOP desc */
996 #ifdef __BIG_ENDIAN_BITFIELD
997         gdesc = &tempTxDesc;
998         gdesc->dword[2] = ctx.sop_txd->dword[2];
999         gdesc->dword[3] = ctx.sop_txd->dword[3];
1000 #else
1001         gdesc = ctx.sop_txd;
1002 #endif
1003         if (ctx.mss) {
1004                 gdesc->txd.hlen = ctx.eth_ip_hdr_size + ctx.l4_hdr_size;
1005                 gdesc->txd.om = VMXNET3_OM_TSO;
1006                 gdesc->txd.msscof = ctx.mss;
1007                 le32_add_cpu(&tq->shared->txNumDeferred, (skb->len -
1008                              gdesc->txd.hlen + ctx.mss - 1) / ctx.mss);
1009         } else {
1010                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1011                         gdesc->txd.hlen = ctx.eth_ip_hdr_size;
1012                         gdesc->txd.om = VMXNET3_OM_CSUM;
1013                         gdesc->txd.msscof = ctx.eth_ip_hdr_size +
1014                                             skb->csum_offset;
1015                 } else {
1016                         gdesc->txd.om = 0;
1017                         gdesc->txd.msscof = 0;
1018                 }
1019                 le32_add_cpu(&tq->shared->txNumDeferred, 1);
1020         }
1021
1022         if (vlan_tx_tag_present(skb)) {
1023                 gdesc->txd.ti = 1;
1024                 gdesc->txd.tci = vlan_tx_tag_get(skb);
1025         }
1026
1027         /* finally flips the GEN bit of the SOP desc. */
1028         gdesc->dword[2] = cpu_to_le32(le32_to_cpu(gdesc->dword[2]) ^
1029                                                   VMXNET3_TXD_GEN);
1030 #ifdef __BIG_ENDIAN_BITFIELD
1031         /* Finished updating in bitfields of Tx Desc, so write them in original
1032          * place.
1033          */
1034         vmxnet3_TxDescToLe((struct Vmxnet3_TxDesc *)gdesc,
1035                            (struct Vmxnet3_TxDesc *)ctx.sop_txd);
1036         gdesc = ctx.sop_txd;
1037 #endif
1038         dev_dbg(&adapter->netdev->dev,
1039                 "txd[%u]: SOP 0x%Lx 0x%x 0x%x\n",
1040                 (u32)((union Vmxnet3_GenericDesc *)ctx.sop_txd -
1041                 tq->tx_ring.base), le64_to_cpu(gdesc->txd.addr),
1042                 le32_to_cpu(gdesc->dword[2]), le32_to_cpu(gdesc->dword[3]));
1043
1044         spin_unlock_irqrestore(&tq->tx_lock, flags);
1045
1046         if (le32_to_cpu(tq->shared->txNumDeferred) >=
1047                                         le32_to_cpu(tq->shared->txThreshold)) {
1048                 tq->shared->txNumDeferred = 0;
1049                 VMXNET3_WRITE_BAR0_REG(adapter,
1050                                        VMXNET3_REG_TXPROD + tq->qid * 8,
1051                                        tq->tx_ring.next2fill);
1052         }
1053
1054         return NETDEV_TX_OK;
1055
1056 hdr_too_big:
1057         tq->stats.drop_oversized_hdr++;
1058 unlock_drop_pkt:
1059         spin_unlock_irqrestore(&tq->tx_lock, flags);
1060 drop_pkt:
1061         tq->stats.drop_total++;
1062         dev_kfree_skb(skb);
1063         return NETDEV_TX_OK;
1064 }
1065
1066
1067 static netdev_tx_t
1068 vmxnet3_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
1069 {
1070         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1071
1072                 BUG_ON(skb->queue_mapping > adapter->num_tx_queues);
1073                 return vmxnet3_tq_xmit(skb,
1074                                        &adapter->tx_queue[skb->queue_mapping],
1075                                        adapter, netdev);
1076 }
1077
1078
1079 static void
1080 vmxnet3_rx_csum(struct vmxnet3_adapter *adapter,
1081                 struct sk_buff *skb,
1082                 union Vmxnet3_GenericDesc *gdesc)
1083 {
1084         if (!gdesc->rcd.cnc && adapter->rxcsum) {
1085                 /* typical case: TCP/UDP over IP and both csums are correct */
1086                 if ((le32_to_cpu(gdesc->dword[3]) & VMXNET3_RCD_CSUM_OK) ==
1087                                                         VMXNET3_RCD_CSUM_OK) {
1088                         skb->ip_summed = CHECKSUM_UNNECESSARY;
1089                         BUG_ON(!(gdesc->rcd.tcp || gdesc->rcd.udp));
1090                         BUG_ON(!(gdesc->rcd.v4  || gdesc->rcd.v6));
1091                         BUG_ON(gdesc->rcd.frg);
1092                 } else {
1093                         if (gdesc->rcd.csum) {
1094                                 skb->csum = htons(gdesc->rcd.csum);
1095                                 skb->ip_summed = CHECKSUM_PARTIAL;
1096                         } else {
1097                                 skb_checksum_none_assert(skb);
1098                         }
1099                 }
1100         } else {
1101                 skb_checksum_none_assert(skb);
1102         }
1103 }
1104
1105
1106 static void
1107 vmxnet3_rx_error(struct vmxnet3_rx_queue *rq, struct Vmxnet3_RxCompDesc *rcd,
1108                  struct vmxnet3_rx_ctx *ctx,  struct vmxnet3_adapter *adapter)
1109 {
1110         rq->stats.drop_err++;
1111         if (!rcd->fcs)
1112                 rq->stats.drop_fcs++;
1113
1114         rq->stats.drop_total++;
1115
1116         /*
1117          * We do not unmap and chain the rx buffer to the skb.
1118          * We basically pretend this buffer is not used and will be recycled
1119          * by vmxnet3_rq_alloc_rx_buf()
1120          */
1121
1122         /*
1123          * ctx->skb may be NULL if this is the first and the only one
1124          * desc for the pkt
1125          */
1126         if (ctx->skb)
1127                 dev_kfree_skb_irq(ctx->skb);
1128
1129         ctx->skb = NULL;
1130 }
1131
1132
1133 static int
1134 vmxnet3_rq_rx_complete(struct vmxnet3_rx_queue *rq,
1135                        struct vmxnet3_adapter *adapter, int quota)
1136 {
1137         static const u32 rxprod_reg[2] = {
1138                 VMXNET3_REG_RXPROD, VMXNET3_REG_RXPROD2
1139         };
1140         u32 num_rxd = 0;
1141         struct Vmxnet3_RxCompDesc *rcd;
1142         struct vmxnet3_rx_ctx *ctx = &rq->rx_ctx;
1143 #ifdef __BIG_ENDIAN_BITFIELD
1144         struct Vmxnet3_RxDesc rxCmdDesc;
1145         struct Vmxnet3_RxCompDesc rxComp;
1146 #endif
1147         vmxnet3_getRxComp(rcd, &rq->comp_ring.base[rq->comp_ring.next2proc].rcd,
1148                           &rxComp);
1149         while (rcd->gen == rq->comp_ring.gen) {
1150                 struct vmxnet3_rx_buf_info *rbi;
1151                 struct sk_buff *skb;
1152                 int num_to_alloc;
1153                 struct Vmxnet3_RxDesc *rxd;
1154                 u32 idx, ring_idx;
1155
1156                 if (num_rxd >= quota) {
1157                         /* we may stop even before we see the EOP desc of
1158                          * the current pkt
1159                          */
1160                         break;
1161                 }
1162                 num_rxd++;
1163                 BUG_ON(rcd->rqID != rq->qid && rcd->rqID != rq->qid2);
1164                 idx = rcd->rxdIdx;
1165                 ring_idx = rcd->rqID < adapter->num_rx_queues ? 0 : 1;
1166                 vmxnet3_getRxDesc(rxd, &rq->rx_ring[ring_idx].base[idx].rxd,
1167                                   &rxCmdDesc);
1168                 rbi = rq->buf_info[ring_idx] + idx;
1169
1170                 BUG_ON(rxd->addr != rbi->dma_addr ||
1171                        rxd->len != rbi->len);
1172
1173                 if (unlikely(rcd->eop && rcd->err)) {
1174                         vmxnet3_rx_error(rq, rcd, ctx, adapter);
1175                         goto rcd_done;
1176                 }
1177
1178                 if (rcd->sop) { /* first buf of the pkt */
1179                         BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_HEAD ||
1180                                rcd->rqID != rq->qid);
1181
1182                         BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_SKB);
1183                         BUG_ON(ctx->skb != NULL || rbi->skb == NULL);
1184
1185                         if (unlikely(rcd->len == 0)) {
1186                                 /* Pretend the rx buffer is skipped. */
1187                                 BUG_ON(!(rcd->sop && rcd->eop));
1188                                 dev_dbg(&adapter->netdev->dev,
1189                                         "rxRing[%u][%u] 0 length\n",
1190                                         ring_idx, idx);
1191                                 goto rcd_done;
1192                         }
1193
1194                         ctx->skb = rbi->skb;
1195                         rbi->skb = NULL;
1196
1197                         pci_unmap_single(adapter->pdev, rbi->dma_addr, rbi->len,
1198                                          PCI_DMA_FROMDEVICE);
1199
1200                         skb_put(ctx->skb, rcd->len);
1201                 } else {
1202                         BUG_ON(ctx->skb == NULL);
1203                         /* non SOP buffer must be type 1 in most cases */
1204                         if (rbi->buf_type == VMXNET3_RX_BUF_PAGE) {
1205                                 BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_BODY);
1206
1207                                 if (rcd->len) {
1208                                         pci_unmap_page(adapter->pdev,
1209                                                        rbi->dma_addr, rbi->len,
1210                                                        PCI_DMA_FROMDEVICE);
1211
1212                                         vmxnet3_append_frag(ctx->skb, rcd, rbi);
1213                                         rbi->page = NULL;
1214                                 }
1215                         } else {
1216                                 /*
1217                                  * The only time a non-SOP buffer is type 0 is
1218                                  * when it's EOP and error flag is raised, which
1219                                  * has already been handled.
1220                                  */
1221                                 BUG_ON(true);
1222                         }
1223                 }
1224
1225                 skb = ctx->skb;
1226                 if (rcd->eop) {
1227                         skb->len += skb->data_len;
1228                         skb->truesize += skb->data_len;
1229
1230                         vmxnet3_rx_csum(adapter, skb,
1231                                         (union Vmxnet3_GenericDesc *)rcd);
1232                         skb->protocol = eth_type_trans(skb, adapter->netdev);
1233
1234                         if (unlikely(adapter->vlan_grp && rcd->ts)) {
1235                                 vlan_hwaccel_receive_skb(skb,
1236                                                 adapter->vlan_grp, rcd->tci);
1237                         } else {
1238                                 netif_receive_skb(skb);
1239                         }
1240
1241                         ctx->skb = NULL;
1242                 }
1243
1244 rcd_done:
1245                 /* device may skip some rx descs */
1246                 rq->rx_ring[ring_idx].next2comp = idx;
1247                 VMXNET3_INC_RING_IDX_ONLY(rq->rx_ring[ring_idx].next2comp,
1248                                           rq->rx_ring[ring_idx].size);
1249
1250                 /* refill rx buffers frequently to avoid starving the h/w */
1251                 num_to_alloc = vmxnet3_cmd_ring_desc_avail(rq->rx_ring +
1252                                                            ring_idx);
1253                 if (unlikely(num_to_alloc > VMXNET3_RX_ALLOC_THRESHOLD(rq,
1254                                                         ring_idx, adapter))) {
1255                         vmxnet3_rq_alloc_rx_buf(rq, ring_idx, num_to_alloc,
1256                                                 adapter);
1257
1258                         /* if needed, update the register */
1259                         if (unlikely(rq->shared->updateRxProd)) {
1260                                 VMXNET3_WRITE_BAR0_REG(adapter,
1261                                         rxprod_reg[ring_idx] + rq->qid * 8,
1262                                         rq->rx_ring[ring_idx].next2fill);
1263                                 rq->uncommitted[ring_idx] = 0;
1264                         }
1265                 }
1266
1267                 vmxnet3_comp_ring_adv_next2proc(&rq->comp_ring);
1268                 vmxnet3_getRxComp(rcd,
1269                      &rq->comp_ring.base[rq->comp_ring.next2proc].rcd, &rxComp);
1270         }
1271
1272         return num_rxd;
1273 }
1274
1275
1276 static void
1277 vmxnet3_rq_cleanup(struct vmxnet3_rx_queue *rq,
1278                    struct vmxnet3_adapter *adapter)
1279 {
1280         u32 i, ring_idx;
1281         struct Vmxnet3_RxDesc *rxd;
1282
1283         for (ring_idx = 0; ring_idx < 2; ring_idx++) {
1284                 for (i = 0; i < rq->rx_ring[ring_idx].size; i++) {
1285 #ifdef __BIG_ENDIAN_BITFIELD
1286                         struct Vmxnet3_RxDesc rxDesc;
1287 #endif
1288                         vmxnet3_getRxDesc(rxd,
1289                                 &rq->rx_ring[ring_idx].base[i].rxd, &rxDesc);
1290
1291                         if (rxd->btype == VMXNET3_RXD_BTYPE_HEAD &&
1292                                         rq->buf_info[ring_idx][i].skb) {
1293                                 pci_unmap_single(adapter->pdev, rxd->addr,
1294                                                  rxd->len, PCI_DMA_FROMDEVICE);
1295                                 dev_kfree_skb(rq->buf_info[ring_idx][i].skb);
1296                                 rq->buf_info[ring_idx][i].skb = NULL;
1297                         } else if (rxd->btype == VMXNET3_RXD_BTYPE_BODY &&
1298                                         rq->buf_info[ring_idx][i].page) {
1299                                 pci_unmap_page(adapter->pdev, rxd->addr,
1300                                                rxd->len, PCI_DMA_FROMDEVICE);
1301                                 put_page(rq->buf_info[ring_idx][i].page);
1302                                 rq->buf_info[ring_idx][i].page = NULL;
1303                         }
1304                 }
1305
1306                 rq->rx_ring[ring_idx].gen = VMXNET3_INIT_GEN;
1307                 rq->rx_ring[ring_idx].next2fill =
1308                                         rq->rx_ring[ring_idx].next2comp = 0;
1309                 rq->uncommitted[ring_idx] = 0;
1310         }
1311
1312         rq->comp_ring.gen = VMXNET3_INIT_GEN;
1313         rq->comp_ring.next2proc = 0;
1314 }
1315
1316
1317 static void
1318 vmxnet3_rq_cleanup_all(struct vmxnet3_adapter *adapter)
1319 {
1320         int i;
1321
1322         for (i = 0; i < adapter->num_rx_queues; i++)
1323                 vmxnet3_rq_cleanup(&adapter->rx_queue[i], adapter);
1324 }
1325
1326
1327 void vmxnet3_rq_destroy(struct vmxnet3_rx_queue *rq,
1328                         struct vmxnet3_adapter *adapter)
1329 {
1330         int i;
1331         int j;
1332
1333         /* all rx buffers must have already been freed */
1334         for (i = 0; i < 2; i++) {
1335                 if (rq->buf_info[i]) {
1336                         for (j = 0; j < rq->rx_ring[i].size; j++)
1337                                 BUG_ON(rq->buf_info[i][j].page != NULL);
1338                 }
1339         }
1340
1341
1342         kfree(rq->buf_info[0]);
1343
1344         for (i = 0; i < 2; i++) {
1345                 if (rq->rx_ring[i].base) {
1346                         pci_free_consistent(adapter->pdev, rq->rx_ring[i].size
1347                                             * sizeof(struct Vmxnet3_RxDesc),
1348                                             rq->rx_ring[i].base,
1349                                             rq->rx_ring[i].basePA);
1350                         rq->rx_ring[i].base = NULL;
1351                 }
1352                 rq->buf_info[i] = NULL;
1353         }
1354
1355         if (rq->comp_ring.base) {
1356                 pci_free_consistent(adapter->pdev, rq->comp_ring.size *
1357                                     sizeof(struct Vmxnet3_RxCompDesc),
1358                                     rq->comp_ring.base, rq->comp_ring.basePA);
1359                 rq->comp_ring.base = NULL;
1360         }
1361 }
1362
1363
1364 static int
1365 vmxnet3_rq_init(struct vmxnet3_rx_queue *rq,
1366                 struct vmxnet3_adapter  *adapter)
1367 {
1368         int i;
1369
1370         /* initialize buf_info */
1371         for (i = 0; i < rq->rx_ring[0].size; i++) {
1372
1373                 /* 1st buf for a pkt is skbuff */
1374                 if (i % adapter->rx_buf_per_pkt == 0) {
1375                         rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_SKB;
1376                         rq->buf_info[0][i].len = adapter->skb_buf_size;
1377                 } else { /* subsequent bufs for a pkt is frag */
1378                         rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_PAGE;
1379                         rq->buf_info[0][i].len = PAGE_SIZE;
1380                 }
1381         }
1382         for (i = 0; i < rq->rx_ring[1].size; i++) {
1383                 rq->buf_info[1][i].buf_type = VMXNET3_RX_BUF_PAGE;
1384                 rq->buf_info[1][i].len = PAGE_SIZE;
1385         }
1386
1387         /* reset internal state and allocate buffers for both rings */
1388         for (i = 0; i < 2; i++) {
1389                 rq->rx_ring[i].next2fill = rq->rx_ring[i].next2comp = 0;
1390                 rq->uncommitted[i] = 0;
1391
1392                 memset(rq->rx_ring[i].base, 0, rq->rx_ring[i].size *
1393                        sizeof(struct Vmxnet3_RxDesc));
1394                 rq->rx_ring[i].gen = VMXNET3_INIT_GEN;
1395         }
1396         if (vmxnet3_rq_alloc_rx_buf(rq, 0, rq->rx_ring[0].size - 1,
1397                                     adapter) == 0) {
1398                 /* at least has 1 rx buffer for the 1st ring */
1399                 return -ENOMEM;
1400         }
1401         vmxnet3_rq_alloc_rx_buf(rq, 1, rq->rx_ring[1].size - 1, adapter);
1402
1403         /* reset the comp ring */
1404         rq->comp_ring.next2proc = 0;
1405         memset(rq->comp_ring.base, 0, rq->comp_ring.size *
1406                sizeof(struct Vmxnet3_RxCompDesc));
1407         rq->comp_ring.gen = VMXNET3_INIT_GEN;
1408
1409         /* reset rxctx */
1410         rq->rx_ctx.skb = NULL;
1411
1412         /* stats are not reset */
1413         return 0;
1414 }
1415
1416
1417 static int
1418 vmxnet3_rq_init_all(struct vmxnet3_adapter *adapter)
1419 {
1420         int i, err = 0;
1421
1422         for (i = 0; i < adapter->num_rx_queues; i++) {
1423                 err = vmxnet3_rq_init(&adapter->rx_queue[i], adapter);
1424                 if (unlikely(err)) {
1425                         dev_err(&adapter->netdev->dev, "%s: failed to "
1426                                 "initialize rx queue%i\n",
1427                                 adapter->netdev->name, i);
1428                         break;
1429                 }
1430         }
1431         return err;
1432
1433 }
1434
1435
1436 static int
1437 vmxnet3_rq_create(struct vmxnet3_rx_queue *rq, struct vmxnet3_adapter *adapter)
1438 {
1439         int i;
1440         size_t sz;
1441         struct vmxnet3_rx_buf_info *bi;
1442
1443         for (i = 0; i < 2; i++) {
1444
1445                 sz = rq->rx_ring[i].size * sizeof(struct Vmxnet3_RxDesc);
1446                 rq->rx_ring[i].base = pci_alloc_consistent(adapter->pdev, sz,
1447                                                         &rq->rx_ring[i].basePA);
1448                 if (!rq->rx_ring[i].base) {
1449                         printk(KERN_ERR "%s: failed to allocate rx ring %d\n",
1450                                adapter->netdev->name, i);
1451                         goto err;
1452                 }
1453         }
1454
1455         sz = rq->comp_ring.size * sizeof(struct Vmxnet3_RxCompDesc);
1456         rq->comp_ring.base = pci_alloc_consistent(adapter->pdev, sz,
1457                                                   &rq->comp_ring.basePA);
1458         if (!rq->comp_ring.base) {
1459                 printk(KERN_ERR "%s: failed to allocate rx comp ring\n",
1460                        adapter->netdev->name);
1461                 goto err;
1462         }
1463
1464         sz = sizeof(struct vmxnet3_rx_buf_info) * (rq->rx_ring[0].size +
1465                                                    rq->rx_ring[1].size);
1466         bi = kzalloc(sz, GFP_KERNEL);
1467         if (!bi) {
1468                 printk(KERN_ERR "%s: failed to allocate rx bufinfo\n",
1469                        adapter->netdev->name);
1470                 goto err;
1471         }
1472         rq->buf_info[0] = bi;
1473         rq->buf_info[1] = bi + rq->rx_ring[0].size;
1474
1475         return 0;
1476
1477 err:
1478         vmxnet3_rq_destroy(rq, adapter);
1479         return -ENOMEM;
1480 }
1481
1482
1483 static int
1484 vmxnet3_rq_create_all(struct vmxnet3_adapter *adapter)
1485 {
1486         int i, err = 0;
1487
1488         for (i = 0; i < adapter->num_rx_queues; i++) {
1489                 err = vmxnet3_rq_create(&adapter->rx_queue[i], adapter);
1490                 if (unlikely(err)) {
1491                         dev_err(&adapter->netdev->dev,
1492                                 "%s: failed to create rx queue%i\n",
1493                                 adapter->netdev->name, i);
1494                         goto err_out;
1495                 }
1496         }
1497         return err;
1498 err_out:
1499         vmxnet3_rq_destroy_all(adapter);
1500         return err;
1501
1502 }
1503
1504 /* Multiple queue aware polling function for tx and rx */
1505
1506 static int
1507 vmxnet3_do_poll(struct vmxnet3_adapter *adapter, int budget)
1508 {
1509         int rcd_done = 0, i;
1510         if (unlikely(adapter->shared->ecr))
1511                 vmxnet3_process_events(adapter);
1512         for (i = 0; i < adapter->num_tx_queues; i++)
1513                 vmxnet3_tq_tx_complete(&adapter->tx_queue[i], adapter);
1514
1515         for (i = 0; i < adapter->num_rx_queues; i++)
1516                 rcd_done += vmxnet3_rq_rx_complete(&adapter->rx_queue[i],
1517                                                    adapter, budget);
1518         return rcd_done;
1519 }
1520
1521
1522 static int
1523 vmxnet3_poll(struct napi_struct *napi, int budget)
1524 {
1525         struct vmxnet3_rx_queue *rx_queue = container_of(napi,
1526                                           struct vmxnet3_rx_queue, napi);
1527         int rxd_done;
1528
1529         rxd_done = vmxnet3_do_poll(rx_queue->adapter, budget);
1530
1531         if (rxd_done < budget) {
1532                 napi_complete(napi);
1533                 vmxnet3_enable_all_intrs(rx_queue->adapter);
1534         }
1535         return rxd_done;
1536 }
1537
1538 /*
1539  * NAPI polling function for MSI-X mode with multiple Rx queues
1540  * Returns the # of the NAPI credit consumed (# of rx descriptors processed)
1541  */
1542
1543 static int
1544 vmxnet3_poll_rx_only(struct napi_struct *napi, int budget)
1545 {
1546         struct vmxnet3_rx_queue *rq = container_of(napi,
1547                                                 struct vmxnet3_rx_queue, napi);
1548         struct vmxnet3_adapter *adapter = rq->adapter;
1549         int rxd_done;
1550
1551         /* When sharing interrupt with corresponding tx queue, process
1552          * tx completions in that queue as well
1553          */
1554         if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE) {
1555                 struct vmxnet3_tx_queue *tq =
1556                                 &adapter->tx_queue[rq - adapter->rx_queue];
1557                 vmxnet3_tq_tx_complete(tq, adapter);
1558         }
1559
1560         rxd_done = vmxnet3_rq_rx_complete(rq, adapter, budget);
1561
1562         if (rxd_done < budget) {
1563                 napi_complete(napi);
1564                 vmxnet3_enable_intr(adapter, rq->comp_ring.intr_idx);
1565         }
1566         return rxd_done;
1567 }
1568
1569
1570 #ifdef CONFIG_PCI_MSI
1571
1572 /*
1573  * Handle completion interrupts on tx queues
1574  * Returns whether or not the intr is handled
1575  */
1576
1577 static irqreturn_t
1578 vmxnet3_msix_tx(int irq, void *data)
1579 {
1580         struct vmxnet3_tx_queue *tq = data;
1581         struct vmxnet3_adapter *adapter = tq->adapter;
1582
1583         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1584                 vmxnet3_disable_intr(adapter, tq->comp_ring.intr_idx);
1585
1586         /* Handle the case where only one irq is allocate for all tx queues */
1587         if (adapter->share_intr == VMXNET3_INTR_TXSHARE) {
1588                 int i;
1589                 for (i = 0; i < adapter->num_tx_queues; i++) {
1590                         struct vmxnet3_tx_queue *txq = &adapter->tx_queue[i];
1591                         vmxnet3_tq_tx_complete(txq, adapter);
1592                 }
1593         } else {
1594                 vmxnet3_tq_tx_complete(tq, adapter);
1595         }
1596         vmxnet3_enable_intr(adapter, tq->comp_ring.intr_idx);
1597
1598         return IRQ_HANDLED;
1599 }
1600
1601
1602 /*
1603  * Handle completion interrupts on rx queues. Returns whether or not the
1604  * intr is handled
1605  */
1606
1607 static irqreturn_t
1608 vmxnet3_msix_rx(int irq, void *data)
1609 {
1610         struct vmxnet3_rx_queue *rq = data;
1611         struct vmxnet3_adapter *adapter = rq->adapter;
1612
1613         /* disable intr if needed */
1614         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1615                 vmxnet3_disable_intr(adapter, rq->comp_ring.intr_idx);
1616         napi_schedule(&rq->napi);
1617
1618         return IRQ_HANDLED;
1619 }
1620
1621 /*
1622  *----------------------------------------------------------------------------
1623  *
1624  * vmxnet3_msix_event --
1625  *
1626  *    vmxnet3 msix event intr handler
1627  *
1628  * Result:
1629  *    whether or not the intr is handled
1630  *
1631  *----------------------------------------------------------------------------
1632  */
1633
1634 static irqreturn_t
1635 vmxnet3_msix_event(int irq, void *data)
1636 {
1637         struct net_device *dev = data;
1638         struct vmxnet3_adapter *adapter = netdev_priv(dev);
1639
1640         /* disable intr if needed */
1641         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1642                 vmxnet3_disable_intr(adapter, adapter->intr.event_intr_idx);
1643
1644         if (adapter->shared->ecr)
1645                 vmxnet3_process_events(adapter);
1646
1647         vmxnet3_enable_intr(adapter, adapter->intr.event_intr_idx);
1648
1649         return IRQ_HANDLED;
1650 }
1651
1652 #endif /* CONFIG_PCI_MSI  */
1653
1654
1655 /* Interrupt handler for vmxnet3  */
1656 static irqreturn_t
1657 vmxnet3_intr(int irq, void *dev_id)
1658 {
1659         struct net_device *dev = dev_id;
1660         struct vmxnet3_adapter *adapter = netdev_priv(dev);
1661
1662         if (adapter->intr.type == VMXNET3_IT_INTX) {
1663                 u32 icr = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_ICR);
1664                 if (unlikely(icr == 0))
1665                         /* not ours */
1666                         return IRQ_NONE;
1667         }
1668
1669
1670         /* disable intr if needed */
1671         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1672                 vmxnet3_disable_all_intrs(adapter);
1673
1674         napi_schedule(&adapter->rx_queue[0].napi);
1675
1676         return IRQ_HANDLED;
1677 }
1678
1679 #ifdef CONFIG_NET_POLL_CONTROLLER
1680
1681 /* netpoll callback. */
1682 static void
1683 vmxnet3_netpoll(struct net_device *netdev)
1684 {
1685         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1686
1687         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1688                 vmxnet3_disable_all_intrs(adapter);
1689
1690         vmxnet3_do_poll(adapter, adapter->rx_queue[0].rx_ring[0].size);
1691         vmxnet3_enable_all_intrs(adapter);
1692
1693 }
1694 #endif  /* CONFIG_NET_POLL_CONTROLLER */
1695
1696 static int
1697 vmxnet3_request_irqs(struct vmxnet3_adapter *adapter)
1698 {
1699         struct vmxnet3_intr *intr = &adapter->intr;
1700         int err = 0, i;
1701         int vector = 0;
1702
1703 #ifdef CONFIG_PCI_MSI
1704         if (adapter->intr.type == VMXNET3_IT_MSIX) {
1705                 for (i = 0; i < adapter->num_tx_queues; i++) {
1706                         if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE) {
1707                                 sprintf(adapter->tx_queue[i].name, "%s-tx-%d",
1708                                         adapter->netdev->name, vector);
1709                                 err = request_irq(
1710                                               intr->msix_entries[vector].vector,
1711                                               vmxnet3_msix_tx, 0,
1712                                               adapter->tx_queue[i].name,
1713                                               &adapter->tx_queue[i]);
1714                         } else {
1715                                 sprintf(adapter->tx_queue[i].name, "%s-rxtx-%d",
1716                                         adapter->netdev->name, vector);
1717                         }
1718                         if (err) {
1719                                 dev_err(&adapter->netdev->dev,
1720                                         "Failed to request irq for MSIX, %s, "
1721                                         "error %d\n",
1722                                         adapter->tx_queue[i].name, err);
1723                                 return err;
1724                         }
1725
1726                         /* Handle the case where only 1 MSIx was allocated for
1727                          * all tx queues */
1728                         if (adapter->share_intr == VMXNET3_INTR_TXSHARE) {
1729                                 for (; i < adapter->num_tx_queues; i++)
1730                                         adapter->tx_queue[i].comp_ring.intr_idx
1731                                                                 = vector;
1732                                 vector++;
1733                                 break;
1734                         } else {
1735                                 adapter->tx_queue[i].comp_ring.intr_idx
1736                                                                 = vector++;
1737                         }
1738                 }
1739                 if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE)
1740                         vector = 0;
1741
1742                 for (i = 0; i < adapter->num_rx_queues; i++) {
1743                         if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE)
1744                                 sprintf(adapter->rx_queue[i].name, "%s-rx-%d",
1745                                         adapter->netdev->name, vector);
1746                         else
1747                                 sprintf(adapter->rx_queue[i].name, "%s-rxtx-%d",
1748                                         adapter->netdev->name, vector);
1749                         err = request_irq(intr->msix_entries[vector].vector,
1750                                           vmxnet3_msix_rx, 0,
1751                                           adapter->rx_queue[i].name,
1752                                           &(adapter->rx_queue[i]));
1753                         if (err) {
1754                                 printk(KERN_ERR "Failed to request irq for MSIX"
1755                                        ", %s, error %d\n",
1756                                        adapter->rx_queue[i].name, err);
1757                                 return err;
1758                         }
1759
1760                         adapter->rx_queue[i].comp_ring.intr_idx = vector++;
1761                 }
1762
1763                 sprintf(intr->event_msi_vector_name, "%s-event-%d",
1764                         adapter->netdev->name, vector);
1765                 err = request_irq(intr->msix_entries[vector].vector,
1766                                   vmxnet3_msix_event, 0,
1767                                   intr->event_msi_vector_name, adapter->netdev);
1768                 intr->event_intr_idx = vector;
1769
1770         } else if (intr->type == VMXNET3_IT_MSI) {
1771                 adapter->num_rx_queues = 1;
1772                 err = request_irq(adapter->pdev->irq, vmxnet3_intr, 0,
1773                                   adapter->netdev->name, adapter->netdev);
1774         } else {
1775 #endif
1776                 adapter->num_rx_queues = 1;
1777                 err = request_irq(adapter->pdev->irq, vmxnet3_intr,
1778                                   IRQF_SHARED, adapter->netdev->name,
1779                                   adapter->netdev);
1780 #ifdef CONFIG_PCI_MSI
1781         }
1782 #endif
1783         intr->num_intrs = vector + 1;
1784         if (err) {
1785                 printk(KERN_ERR "Failed to request irq %s (intr type:%d), error"
1786                        ":%d\n", adapter->netdev->name, intr->type, err);
1787         } else {
1788                 /* Number of rx queues will not change after this */
1789                 for (i = 0; i < adapter->num_rx_queues; i++) {
1790                         struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
1791                         rq->qid = i;
1792                         rq->qid2 = i + adapter->num_rx_queues;
1793                 }
1794
1795
1796
1797                 /* init our intr settings */
1798                 for (i = 0; i < intr->num_intrs; i++)
1799                         intr->mod_levels[i] = UPT1_IML_ADAPTIVE;
1800                 if (adapter->intr.type != VMXNET3_IT_MSIX) {
1801                         adapter->intr.event_intr_idx = 0;
1802                         for (i = 0; i < adapter->num_tx_queues; i++)
1803                                 adapter->tx_queue[i].comp_ring.intr_idx = 0;
1804                         adapter->rx_queue[0].comp_ring.intr_idx = 0;
1805                 }
1806
1807                 printk(KERN_INFO "%s: intr type %u, mode %u, %u vectors "
1808                        "allocated\n", adapter->netdev->name, intr->type,
1809                        intr->mask_mode, intr->num_intrs);
1810         }
1811
1812         return err;
1813 }
1814
1815
1816 static void
1817 vmxnet3_free_irqs(struct vmxnet3_adapter *adapter)
1818 {
1819         struct vmxnet3_intr *intr = &adapter->intr;
1820         BUG_ON(intr->type == VMXNET3_IT_AUTO || intr->num_intrs <= 0);
1821
1822         switch (intr->type) {
1823 #ifdef CONFIG_PCI_MSI
1824         case VMXNET3_IT_MSIX:
1825         {
1826                 int i, vector = 0;
1827
1828                 if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE) {
1829                         for (i = 0; i < adapter->num_tx_queues; i++) {
1830                                 free_irq(intr->msix_entries[vector++].vector,
1831                                          &(adapter->tx_queue[i]));
1832                                 if (adapter->share_intr == VMXNET3_INTR_TXSHARE)
1833                                         break;
1834                         }
1835                 }
1836
1837                 for (i = 0; i < adapter->num_rx_queues; i++) {
1838                         free_irq(intr->msix_entries[vector++].vector,
1839                                  &(adapter->rx_queue[i]));
1840                 }
1841
1842                 free_irq(intr->msix_entries[vector].vector,
1843                          adapter->netdev);
1844                 BUG_ON(vector >= intr->num_intrs);
1845                 break;
1846         }
1847 #endif
1848         case VMXNET3_IT_MSI:
1849                 free_irq(adapter->pdev->irq, adapter->netdev);
1850                 break;
1851         case VMXNET3_IT_INTX:
1852                 free_irq(adapter->pdev->irq, adapter->netdev);
1853                 break;
1854         default:
1855                 BUG_ON(true);
1856         }
1857 }
1858
1859 static void
1860 vmxnet3_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp)
1861 {
1862         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1863         struct Vmxnet3_DriverShared *shared = adapter->shared;
1864         u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1865
1866         if (grp) {
1867                 /* add vlan rx stripping. */
1868                 if (adapter->netdev->features & NETIF_F_HW_VLAN_RX) {
1869                         int i;
1870                         struct Vmxnet3_DSDevRead *devRead = &shared->devRead;
1871                         adapter->vlan_grp = grp;
1872
1873                         /* update FEATURES to device */
1874                         devRead->misc.uptFeatures |= UPT1_F_RXVLAN;
1875                         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1876                                                VMXNET3_CMD_UPDATE_FEATURE);
1877                         /*
1878                          *  Clear entire vfTable; then enable untagged pkts.
1879                          *  Note: setting one entry in vfTable to non-zero turns
1880                          *  on VLAN rx filtering.
1881                          */
1882                         for (i = 0; i < VMXNET3_VFT_SIZE; i++)
1883                                 vfTable[i] = 0;
1884
1885                         VMXNET3_SET_VFTABLE_ENTRY(vfTable, 0);
1886                         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1887                                                VMXNET3_CMD_UPDATE_VLAN_FILTERS);
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                         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1907                                                VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1908
1909                         /* update FEATURES to device */
1910                         devRead->misc.uptFeatures &= ~UPT1_F_RXVLAN;
1911                         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1912                                                VMXNET3_CMD_UPDATE_FEATURE);
1913                 }
1914         }
1915 }
1916
1917
1918 static void
1919 vmxnet3_restore_vlan(struct vmxnet3_adapter *adapter)
1920 {
1921         if (adapter->vlan_grp) {
1922                 u16 vid;
1923                 u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1924                 bool activeVlan = false;
1925
1926                 for (vid = 0; vid < VLAN_N_VID; vid++) {
1927                         if (vlan_group_get_device(adapter->vlan_grp, vid)) {
1928                                 VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid);
1929                                 activeVlan = true;
1930                         }
1931                 }
1932                 if (activeVlan) {
1933                         /* continue to allow untagged pkts */
1934                         VMXNET3_SET_VFTABLE_ENTRY(vfTable, 0);
1935                 }
1936         }
1937 }
1938
1939
1940 static void
1941 vmxnet3_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1942 {
1943         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1944         u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1945
1946         VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid);
1947         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1948                                VMXNET3_CMD_UPDATE_VLAN_FILTERS);
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
1958         VMXNET3_CLEAR_VFTABLE_ENTRY(vfTable, vid);
1959         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1960                                VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1961 }
1962
1963
1964 static u8 *
1965 vmxnet3_copy_mc(struct net_device *netdev)
1966 {
1967         u8 *buf = NULL;
1968         u32 sz = netdev_mc_count(netdev) * ETH_ALEN;
1969
1970         /* struct Vmxnet3_RxFilterConf.mfTableLen is u16. */
1971         if (sz <= 0xffff) {
1972                 /* We may be called with BH disabled */
1973                 buf = kmalloc(sz, GFP_ATOMIC);
1974                 if (buf) {
1975                         struct netdev_hw_addr *ha;
1976                         int i = 0;
1977
1978                         netdev_for_each_mc_addr(ha, netdev)
1979                                 memcpy(buf + i++ * ETH_ALEN, ha->addr,
1980                                        ETH_ALEN);
1981                 }
1982         }
1983         return buf;
1984 }
1985
1986
1987 static void
1988 vmxnet3_set_mc(struct net_device *netdev)
1989 {
1990         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1991         struct Vmxnet3_RxFilterConf *rxConf =
1992                                         &adapter->shared->devRead.rxFilterConf;
1993         u8 *new_table = NULL;
1994         u32 new_mode = VMXNET3_RXM_UCAST;
1995
1996         if (netdev->flags & IFF_PROMISC)
1997                 new_mode |= VMXNET3_RXM_PROMISC;
1998
1999         if (netdev->flags & IFF_BROADCAST)
2000                 new_mode |= VMXNET3_RXM_BCAST;
2001
2002         if (netdev->flags & IFF_ALLMULTI)
2003                 new_mode |= VMXNET3_RXM_ALL_MULTI;
2004         else
2005                 if (!netdev_mc_empty(netdev)) {
2006                         new_table = vmxnet3_copy_mc(netdev);
2007                         if (new_table) {
2008                                 new_mode |= VMXNET3_RXM_MCAST;
2009                                 rxConf->mfTableLen = cpu_to_le16(
2010                                         netdev_mc_count(netdev) * ETH_ALEN);
2011                                 rxConf->mfTablePA = cpu_to_le64(virt_to_phys(
2012                                                     new_table));
2013                         } else {
2014                                 printk(KERN_INFO "%s: failed to copy mcast list"
2015                                        ", setting ALL_MULTI\n", netdev->name);
2016                                 new_mode |= VMXNET3_RXM_ALL_MULTI;
2017                         }
2018                 }
2019
2020
2021         if (!(new_mode & VMXNET3_RXM_MCAST)) {
2022                 rxConf->mfTableLen = 0;
2023                 rxConf->mfTablePA = 0;
2024         }
2025
2026         if (new_mode != rxConf->rxMode) {
2027                 rxConf->rxMode = cpu_to_le32(new_mode);
2028                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2029                                        VMXNET3_CMD_UPDATE_RX_MODE);
2030         }
2031
2032         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2033                                VMXNET3_CMD_UPDATE_MAC_FILTERS);
2034
2035         kfree(new_table);
2036 }
2037
2038 void
2039 vmxnet3_rq_destroy_all(struct vmxnet3_adapter *adapter)
2040 {
2041         int i;
2042
2043         for (i = 0; i < adapter->num_rx_queues; i++)
2044                 vmxnet3_rq_destroy(&adapter->rx_queue[i], adapter);
2045 }
2046
2047
2048 /*
2049  *   Set up driver_shared based on settings in adapter.
2050  */
2051
2052 static void
2053 vmxnet3_setup_driver_shared(struct vmxnet3_adapter *adapter)
2054 {
2055         struct Vmxnet3_DriverShared *shared = adapter->shared;
2056         struct Vmxnet3_DSDevRead *devRead = &shared->devRead;
2057         struct Vmxnet3_TxQueueConf *tqc;
2058         struct Vmxnet3_RxQueueConf *rqc;
2059         int i;
2060
2061         memset(shared, 0, sizeof(*shared));
2062
2063         /* driver settings */
2064         shared->magic = cpu_to_le32(VMXNET3_REV1_MAGIC);
2065         devRead->misc.driverInfo.version = cpu_to_le32(
2066                                                 VMXNET3_DRIVER_VERSION_NUM);
2067         devRead->misc.driverInfo.gos.gosBits = (sizeof(void *) == 4 ?
2068                                 VMXNET3_GOS_BITS_32 : VMXNET3_GOS_BITS_64);
2069         devRead->misc.driverInfo.gos.gosType = VMXNET3_GOS_TYPE_LINUX;
2070         *((u32 *)&devRead->misc.driverInfo.gos) = cpu_to_le32(
2071                                 *((u32 *)&devRead->misc.driverInfo.gos));
2072         devRead->misc.driverInfo.vmxnet3RevSpt = cpu_to_le32(1);
2073         devRead->misc.driverInfo.uptVerSpt = cpu_to_le32(1);
2074
2075         devRead->misc.ddPA = cpu_to_le64(virt_to_phys(adapter));
2076         devRead->misc.ddLen = cpu_to_le32(sizeof(struct vmxnet3_adapter));
2077
2078         /* set up feature flags */
2079         if (adapter->rxcsum)
2080                 devRead->misc.uptFeatures |= UPT1_F_RXCSUM;
2081
2082         if (adapter->lro) {
2083                 devRead->misc.uptFeatures |= UPT1_F_LRO;
2084                 devRead->misc.maxNumRxSG = cpu_to_le16(1 + MAX_SKB_FRAGS);
2085         }
2086         if ((adapter->netdev->features & NETIF_F_HW_VLAN_RX) &&
2087             adapter->vlan_grp) {
2088                 devRead->misc.uptFeatures |= UPT1_F_RXVLAN;
2089         }
2090
2091         devRead->misc.mtu = cpu_to_le32(adapter->netdev->mtu);
2092         devRead->misc.queueDescPA = cpu_to_le64(adapter->queue_desc_pa);
2093         devRead->misc.queueDescLen = cpu_to_le32(
2094                 adapter->num_tx_queues * sizeof(struct Vmxnet3_TxQueueDesc) +
2095                 adapter->num_rx_queues * sizeof(struct Vmxnet3_RxQueueDesc));
2096
2097         /* tx queue settings */
2098         devRead->misc.numTxQueues =  adapter->num_tx_queues;
2099         for (i = 0; i < adapter->num_tx_queues; i++) {
2100                 struct vmxnet3_tx_queue *tq = &adapter->tx_queue[i];
2101                 BUG_ON(adapter->tx_queue[i].tx_ring.base == NULL);
2102                 tqc = &adapter->tqd_start[i].conf;
2103                 tqc->txRingBasePA   = cpu_to_le64(tq->tx_ring.basePA);
2104                 tqc->dataRingBasePA = cpu_to_le64(tq->data_ring.basePA);
2105                 tqc->compRingBasePA = cpu_to_le64(tq->comp_ring.basePA);
2106                 tqc->ddPA           = cpu_to_le64(virt_to_phys(tq->buf_info));
2107                 tqc->txRingSize     = cpu_to_le32(tq->tx_ring.size);
2108                 tqc->dataRingSize   = cpu_to_le32(tq->data_ring.size);
2109                 tqc->compRingSize   = cpu_to_le32(tq->comp_ring.size);
2110                 tqc->ddLen          = cpu_to_le32(
2111                                         sizeof(struct vmxnet3_tx_buf_info) *
2112                                         tqc->txRingSize);
2113                 tqc->intrIdx        = tq->comp_ring.intr_idx;
2114         }
2115
2116         /* rx queue settings */
2117         devRead->misc.numRxQueues = adapter->num_rx_queues;
2118         for (i = 0; i < adapter->num_rx_queues; i++) {
2119                 struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
2120                 rqc = &adapter->rqd_start[i].conf;
2121                 rqc->rxRingBasePA[0] = cpu_to_le64(rq->rx_ring[0].basePA);
2122                 rqc->rxRingBasePA[1] = cpu_to_le64(rq->rx_ring[1].basePA);
2123                 rqc->compRingBasePA  = cpu_to_le64(rq->comp_ring.basePA);
2124                 rqc->ddPA            = cpu_to_le64(virt_to_phys(
2125                                                         rq->buf_info));
2126                 rqc->rxRingSize[0]   = cpu_to_le32(rq->rx_ring[0].size);
2127                 rqc->rxRingSize[1]   = cpu_to_le32(rq->rx_ring[1].size);
2128                 rqc->compRingSize    = cpu_to_le32(rq->comp_ring.size);
2129                 rqc->ddLen           = cpu_to_le32(
2130                                         sizeof(struct vmxnet3_rx_buf_info) *
2131                                         (rqc->rxRingSize[0] +
2132                                          rqc->rxRingSize[1]));
2133                 rqc->intrIdx         = rq->comp_ring.intr_idx;
2134         }
2135
2136 #ifdef VMXNET3_RSS
2137         memset(adapter->rss_conf, 0, sizeof(*adapter->rss_conf));
2138
2139         if (adapter->rss) {
2140                 struct UPT1_RSSConf *rssConf = adapter->rss_conf;
2141                 devRead->misc.uptFeatures |= UPT1_F_RSS;
2142                 devRead->misc.numRxQueues = adapter->num_rx_queues;
2143                 rssConf->hashType = UPT1_RSS_HASH_TYPE_TCP_IPV4 |
2144                                     UPT1_RSS_HASH_TYPE_IPV4 |
2145                                     UPT1_RSS_HASH_TYPE_TCP_IPV6 |
2146                                     UPT1_RSS_HASH_TYPE_IPV6;
2147                 rssConf->hashFunc = UPT1_RSS_HASH_FUNC_TOEPLITZ;
2148                 rssConf->hashKeySize = UPT1_RSS_MAX_KEY_SIZE;
2149                 rssConf->indTableSize = VMXNET3_RSS_IND_TABLE_SIZE;
2150                 get_random_bytes(&rssConf->hashKey[0], rssConf->hashKeySize);
2151                 for (i = 0; i < rssConf->indTableSize; i++)
2152                         rssConf->indTable[i] = i % adapter->num_rx_queues;
2153
2154                 devRead->rssConfDesc.confVer = 1;
2155                 devRead->rssConfDesc.confLen = sizeof(*rssConf);
2156                 devRead->rssConfDesc.confPA  = virt_to_phys(rssConf);
2157         }
2158
2159 #endif /* VMXNET3_RSS */
2160
2161         /* intr settings */
2162         devRead->intrConf.autoMask = adapter->intr.mask_mode ==
2163                                      VMXNET3_IMM_AUTO;
2164         devRead->intrConf.numIntrs = adapter->intr.num_intrs;
2165         for (i = 0; i < adapter->intr.num_intrs; i++)
2166                 devRead->intrConf.modLevels[i] = adapter->intr.mod_levels[i];
2167
2168         devRead->intrConf.eventIntrIdx = adapter->intr.event_intr_idx;
2169         devRead->intrConf.intrCtrl |= cpu_to_le32(VMXNET3_IC_DISABLE_ALL);
2170
2171         /* rx filter settings */
2172         devRead->rxFilterConf.rxMode = 0;
2173         vmxnet3_restore_vlan(adapter);
2174         vmxnet3_write_mac_addr(adapter, adapter->netdev->dev_addr);
2175
2176         /* the rest are already zeroed */
2177 }
2178
2179
2180 int
2181 vmxnet3_activate_dev(struct vmxnet3_adapter *adapter)
2182 {
2183         int err, i;
2184         u32 ret;
2185
2186         dev_dbg(&adapter->netdev->dev, "%s: skb_buf_size %d, rx_buf_per_pkt %d,"
2187                 " ring sizes %u %u %u\n", adapter->netdev->name,
2188                 adapter->skb_buf_size, adapter->rx_buf_per_pkt,
2189                 adapter->tx_queue[0].tx_ring.size,
2190                 adapter->rx_queue[0].rx_ring[0].size,
2191                 adapter->rx_queue[0].rx_ring[1].size);
2192
2193         vmxnet3_tq_init_all(adapter);
2194         err = vmxnet3_rq_init_all(adapter);
2195         if (err) {
2196                 printk(KERN_ERR "Failed to init rx queue for %s: error %d\n",
2197                        adapter->netdev->name, err);
2198                 goto rq_err;
2199         }
2200
2201         err = vmxnet3_request_irqs(adapter);
2202         if (err) {
2203                 printk(KERN_ERR "Failed to setup irq for %s: error %d\n",
2204                        adapter->netdev->name, err);
2205                 goto irq_err;
2206         }
2207
2208         vmxnet3_setup_driver_shared(adapter);
2209
2210         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, VMXNET3_GET_ADDR_LO(
2211                                adapter->shared_pa));
2212         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, VMXNET3_GET_ADDR_HI(
2213                                adapter->shared_pa));
2214         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2215                                VMXNET3_CMD_ACTIVATE_DEV);
2216         ret = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
2217
2218         if (ret != 0) {
2219                 printk(KERN_ERR "Failed to activate dev %s: error %u\n",
2220                        adapter->netdev->name, ret);
2221                 err = -EINVAL;
2222                 goto activate_err;
2223         }
2224
2225         for (i = 0; i < adapter->num_rx_queues; i++) {
2226                 VMXNET3_WRITE_BAR0_REG(adapter,
2227                                 VMXNET3_REG_RXPROD + i * VMXNET3_REG_ALIGN,
2228                                 adapter->rx_queue[i].rx_ring[0].next2fill);
2229                 VMXNET3_WRITE_BAR0_REG(adapter, (VMXNET3_REG_RXPROD2 +
2230                                 (i * VMXNET3_REG_ALIGN)),
2231                                 adapter->rx_queue[i].rx_ring[1].next2fill);
2232         }
2233
2234         /* Apply the rx filter settins last. */
2235         vmxnet3_set_mc(adapter->netdev);
2236
2237         /*
2238          * Check link state when first activating device. It will start the
2239          * tx queue if the link is up.
2240          */
2241         vmxnet3_check_link(adapter, true);
2242         for (i = 0; i < adapter->num_rx_queues; i++)
2243                 napi_enable(&adapter->rx_queue[i].napi);
2244         vmxnet3_enable_all_intrs(adapter);
2245         clear_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
2246         return 0;
2247
2248 activate_err:
2249         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, 0);
2250         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, 0);
2251         vmxnet3_free_irqs(adapter);
2252 irq_err:
2253 rq_err:
2254         /* free up buffers we allocated */
2255         vmxnet3_rq_cleanup_all(adapter);
2256         return err;
2257 }
2258
2259
2260 void
2261 vmxnet3_reset_dev(struct vmxnet3_adapter *adapter)
2262 {
2263         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, VMXNET3_CMD_RESET_DEV);
2264 }
2265
2266
2267 int
2268 vmxnet3_quiesce_dev(struct vmxnet3_adapter *adapter)
2269 {
2270         int i;
2271         if (test_and_set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state))
2272                 return 0;
2273
2274
2275         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2276                                VMXNET3_CMD_QUIESCE_DEV);
2277         vmxnet3_disable_all_intrs(adapter);
2278
2279         for (i = 0; i < adapter->num_rx_queues; i++)
2280                 napi_disable(&adapter->rx_queue[i].napi);
2281         netif_tx_disable(adapter->netdev);
2282         adapter->link_speed = 0;
2283         netif_carrier_off(adapter->netdev);
2284
2285         vmxnet3_tq_cleanup_all(adapter);
2286         vmxnet3_rq_cleanup_all(adapter);
2287         vmxnet3_free_irqs(adapter);
2288         return 0;
2289 }
2290
2291
2292 static void
2293 vmxnet3_write_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac)
2294 {
2295         u32 tmp;
2296
2297         tmp = *(u32 *)mac;
2298         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACL, tmp);
2299
2300         tmp = (mac[5] << 8) | mac[4];
2301         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACH, tmp);
2302 }
2303
2304
2305 static int
2306 vmxnet3_set_mac_addr(struct net_device *netdev, void *p)
2307 {
2308         struct sockaddr *addr = p;
2309         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2310
2311         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2312         vmxnet3_write_mac_addr(adapter, addr->sa_data);
2313
2314         return 0;
2315 }
2316
2317
2318 /* ==================== initialization and cleanup routines ============ */
2319
2320 static int
2321 vmxnet3_alloc_pci_resources(struct vmxnet3_adapter *adapter, bool *dma64)
2322 {
2323         int err;
2324         unsigned long mmio_start, mmio_len;
2325         struct pci_dev *pdev = adapter->pdev;
2326
2327         err = pci_enable_device(pdev);
2328         if (err) {
2329                 printk(KERN_ERR "Failed to enable adapter %s: error %d\n",
2330                        pci_name(pdev), err);
2331                 return err;
2332         }
2333
2334         if (pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) == 0) {
2335                 if (pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64)) != 0) {
2336                         printk(KERN_ERR "pci_set_consistent_dma_mask failed "
2337                                "for adapter %s\n", pci_name(pdev));
2338                         err = -EIO;
2339                         goto err_set_mask;
2340                 }
2341                 *dma64 = true;
2342         } else {
2343                 if (pci_set_dma_mask(pdev, DMA_BIT_MASK(32)) != 0) {
2344                         printk(KERN_ERR "pci_set_dma_mask failed for adapter "
2345                                "%s\n",  pci_name(pdev));
2346                         err = -EIO;
2347                         goto err_set_mask;
2348                 }
2349                 *dma64 = false;
2350         }
2351
2352         err = pci_request_selected_regions(pdev, (1 << 2) - 1,
2353                                            vmxnet3_driver_name);
2354         if (err) {
2355                 printk(KERN_ERR "Failed to request region for adapter %s: "
2356                        "error %d\n", pci_name(pdev), err);
2357                 goto err_set_mask;
2358         }
2359
2360         pci_set_master(pdev);
2361
2362         mmio_start = pci_resource_start(pdev, 0);
2363         mmio_len = pci_resource_len(pdev, 0);
2364         adapter->hw_addr0 = ioremap(mmio_start, mmio_len);
2365         if (!adapter->hw_addr0) {
2366                 printk(KERN_ERR "Failed to map bar0 for adapter %s\n",
2367                        pci_name(pdev));
2368                 err = -EIO;
2369                 goto err_ioremap;
2370         }
2371
2372         mmio_start = pci_resource_start(pdev, 1);
2373         mmio_len = pci_resource_len(pdev, 1);
2374         adapter->hw_addr1 = ioremap(mmio_start, mmio_len);
2375         if (!adapter->hw_addr1) {
2376                 printk(KERN_ERR "Failed to map bar1 for adapter %s\n",
2377                        pci_name(pdev));
2378                 err = -EIO;
2379                 goto err_bar1;
2380         }
2381         return 0;
2382
2383 err_bar1:
2384         iounmap(adapter->hw_addr0);
2385 err_ioremap:
2386         pci_release_selected_regions(pdev, (1 << 2) - 1);
2387 err_set_mask:
2388         pci_disable_device(pdev);
2389         return err;
2390 }
2391
2392
2393 static void
2394 vmxnet3_free_pci_resources(struct vmxnet3_adapter *adapter)
2395 {
2396         BUG_ON(!adapter->pdev);
2397
2398         iounmap(adapter->hw_addr0);
2399         iounmap(adapter->hw_addr1);
2400         pci_release_selected_regions(adapter->pdev, (1 << 2) - 1);
2401         pci_disable_device(adapter->pdev);
2402 }
2403
2404
2405 static void
2406 vmxnet3_adjust_rx_ring_size(struct vmxnet3_adapter *adapter)
2407 {
2408         size_t sz, i, ring0_size, ring1_size, comp_size;
2409         struct vmxnet3_rx_queue *rq = &adapter->rx_queue[0];
2410
2411
2412         if (adapter->netdev->mtu <= VMXNET3_MAX_SKB_BUF_SIZE -
2413                                     VMXNET3_MAX_ETH_HDR_SIZE) {
2414                 adapter->skb_buf_size = adapter->netdev->mtu +
2415                                         VMXNET3_MAX_ETH_HDR_SIZE;
2416                 if (adapter->skb_buf_size < VMXNET3_MIN_T0_BUF_SIZE)
2417                         adapter->skb_buf_size = VMXNET3_MIN_T0_BUF_SIZE;
2418
2419                 adapter->rx_buf_per_pkt = 1;
2420         } else {
2421                 adapter->skb_buf_size = VMXNET3_MAX_SKB_BUF_SIZE;
2422                 sz = adapter->netdev->mtu - VMXNET3_MAX_SKB_BUF_SIZE +
2423                                             VMXNET3_MAX_ETH_HDR_SIZE;
2424                 adapter->rx_buf_per_pkt = 1 + (sz + PAGE_SIZE - 1) / PAGE_SIZE;
2425         }
2426
2427         /*
2428          * for simplicity, force the ring0 size to be a multiple of
2429          * rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN
2430          */
2431         sz = adapter->rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN;
2432         ring0_size = adapter->rx_queue[0].rx_ring[0].size;
2433         ring0_size = (ring0_size + sz - 1) / sz * sz;
2434         ring0_size = min_t(u32, ring0_size, VMXNET3_RX_RING_MAX_SIZE /
2435                            sz * sz);
2436         ring1_size = adapter->rx_queue[0].rx_ring[1].size;
2437         comp_size = ring0_size + ring1_size;
2438
2439         for (i = 0; i < adapter->num_rx_queues; i++) {
2440                 rq = &adapter->rx_queue[i];
2441                 rq->rx_ring[0].size = ring0_size;
2442                 rq->rx_ring[1].size = ring1_size;
2443                 rq->comp_ring.size = comp_size;
2444         }
2445 }
2446
2447
2448 int
2449 vmxnet3_create_queues(struct vmxnet3_adapter *adapter, u32 tx_ring_size,
2450                       u32 rx_ring_size, u32 rx_ring2_size)
2451 {
2452         int err = 0, i;
2453
2454         for (i = 0; i < adapter->num_tx_queues; i++) {
2455                 struct vmxnet3_tx_queue *tq = &adapter->tx_queue[i];
2456                 tq->tx_ring.size   = tx_ring_size;
2457                 tq->data_ring.size = tx_ring_size;
2458                 tq->comp_ring.size = tx_ring_size;
2459                 tq->shared = &adapter->tqd_start[i].ctrl;
2460                 tq->stopped = true;
2461                 tq->adapter = adapter;
2462                 tq->qid = i;
2463                 err = vmxnet3_tq_create(tq, adapter);
2464                 /*
2465                  * Too late to change num_tx_queues. We cannot do away with
2466                  * lesser number of queues than what we asked for
2467                  */
2468                 if (err)
2469                         goto queue_err;
2470         }
2471
2472         adapter->rx_queue[0].rx_ring[0].size = rx_ring_size;
2473         adapter->rx_queue[0].rx_ring[1].size = rx_ring2_size;
2474         vmxnet3_adjust_rx_ring_size(adapter);
2475         for (i = 0; i < adapter->num_rx_queues; i++) {
2476                 struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
2477                 /* qid and qid2 for rx queues will be assigned later when num
2478                  * of rx queues is finalized after allocating intrs */
2479                 rq->shared = &adapter->rqd_start[i].ctrl;
2480                 rq->adapter = adapter;
2481                 err = vmxnet3_rq_create(rq, adapter);
2482                 if (err) {
2483                         if (i == 0) {
2484                                 printk(KERN_ERR "Could not allocate any rx"
2485                                        "queues. Aborting.\n");
2486                                 goto queue_err;
2487                         } else {
2488                                 printk(KERN_INFO "Number of rx queues changed "
2489                                        "to : %d.\n", i);
2490                                 adapter->num_rx_queues = i;
2491                                 err = 0;
2492                                 break;
2493                         }
2494                 }
2495         }
2496         return err;
2497 queue_err:
2498         vmxnet3_tq_destroy_all(adapter);
2499         return err;
2500 }
2501
2502 static int
2503 vmxnet3_open(struct net_device *netdev)
2504 {
2505         struct vmxnet3_adapter *adapter;
2506         int err, i;
2507
2508         adapter = netdev_priv(netdev);
2509
2510         for (i = 0; i < adapter->num_tx_queues; i++)
2511                 spin_lock_init(&adapter->tx_queue[i].tx_lock);
2512
2513         err = vmxnet3_create_queues(adapter, VMXNET3_DEF_TX_RING_SIZE,
2514                                     VMXNET3_DEF_RX_RING_SIZE,
2515                                     VMXNET3_DEF_RX_RING_SIZE);
2516         if (err)
2517                 goto queue_err;
2518
2519         err = vmxnet3_activate_dev(adapter);
2520         if (err)
2521                 goto activate_err;
2522
2523         return 0;
2524
2525 activate_err:
2526         vmxnet3_rq_destroy_all(adapter);
2527         vmxnet3_tq_destroy_all(adapter);
2528 queue_err:
2529         return err;
2530 }
2531
2532
2533 static int
2534 vmxnet3_close(struct net_device *netdev)
2535 {
2536         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2537
2538         /*
2539          * Reset_work may be in the middle of resetting the device, wait for its
2540          * completion.
2541          */
2542         while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2543                 msleep(1);
2544
2545         vmxnet3_quiesce_dev(adapter);
2546
2547         vmxnet3_rq_destroy_all(adapter);
2548         vmxnet3_tq_destroy_all(adapter);
2549
2550         clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2551
2552
2553         return 0;
2554 }
2555
2556
2557 void
2558 vmxnet3_force_close(struct vmxnet3_adapter *adapter)
2559 {
2560         int i;
2561
2562         /*
2563          * we must clear VMXNET3_STATE_BIT_RESETTING, otherwise
2564          * vmxnet3_close() will deadlock.
2565          */
2566         BUG_ON(test_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state));
2567
2568         /* we need to enable NAPI, otherwise dev_close will deadlock */
2569         for (i = 0; i < adapter->num_rx_queues; i++)
2570                 napi_enable(&adapter->rx_queue[i].napi);
2571         dev_close(adapter->netdev);
2572 }
2573
2574
2575 static int
2576 vmxnet3_change_mtu(struct net_device *netdev, int new_mtu)
2577 {
2578         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2579         int err = 0;
2580
2581         if (new_mtu < VMXNET3_MIN_MTU || new_mtu > VMXNET3_MAX_MTU)
2582                 return -EINVAL;
2583
2584         if (new_mtu > 1500 && !adapter->jumbo_frame)
2585                 return -EINVAL;
2586
2587         netdev->mtu = new_mtu;
2588
2589         /*
2590          * Reset_work may be in the middle of resetting the device, wait for its
2591          * completion.
2592          */
2593         while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2594                 msleep(1);
2595
2596         if (netif_running(netdev)) {
2597                 vmxnet3_quiesce_dev(adapter);
2598                 vmxnet3_reset_dev(adapter);
2599
2600                 /* we need to re-create the rx queue based on the new mtu */
2601                 vmxnet3_rq_destroy_all(adapter);
2602                 vmxnet3_adjust_rx_ring_size(adapter);
2603                 err = vmxnet3_rq_create_all(adapter);
2604                 if (err) {
2605                         printk(KERN_ERR "%s: failed to re-create rx queues,"
2606                                 " error %d. Closing it.\n", netdev->name, err);
2607                         goto out;
2608                 }
2609
2610                 err = vmxnet3_activate_dev(adapter);
2611                 if (err) {
2612                         printk(KERN_ERR "%s: failed to re-activate, error %d. "
2613                                 "Closing it\n", netdev->name, err);
2614                         goto out;
2615                 }
2616         }
2617
2618 out:
2619         clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2620         if (err)
2621                 vmxnet3_force_close(adapter);
2622
2623         return err;
2624 }
2625
2626
2627 static void
2628 vmxnet3_declare_features(struct vmxnet3_adapter *adapter, bool dma64)
2629 {
2630         struct net_device *netdev = adapter->netdev;
2631
2632         netdev->features = NETIF_F_SG |
2633                 NETIF_F_HW_CSUM |
2634                 NETIF_F_HW_VLAN_TX |
2635                 NETIF_F_HW_VLAN_RX |
2636                 NETIF_F_HW_VLAN_FILTER |
2637                 NETIF_F_TSO |
2638                 NETIF_F_TSO6 |
2639                 NETIF_F_LRO;
2640
2641         printk(KERN_INFO "features: sg csum vlan jf tso tsoIPv6 lro");
2642
2643         adapter->rxcsum = true;
2644         adapter->jumbo_frame = true;
2645         adapter->lro = true;
2646
2647         if (dma64) {
2648                 netdev->features |= NETIF_F_HIGHDMA;
2649                 printk(" highDMA");
2650         }
2651
2652         netdev->vlan_features = netdev->features;
2653         printk("\n");
2654 }
2655
2656
2657 static void
2658 vmxnet3_read_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac)
2659 {
2660         u32 tmp;
2661
2662         tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACL);
2663         *(u32 *)mac = tmp;
2664
2665         tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACH);
2666         mac[4] = tmp & 0xff;
2667         mac[5] = (tmp >> 8) & 0xff;
2668 }
2669
2670 #ifdef CONFIG_PCI_MSI
2671
2672 /*
2673  * Enable MSIx vectors.
2674  * Returns :
2675  *      0 on successful enabling of required vectors,
2676  *      VMXNET3_LINUX_MIN_MSIX_VECT when only minumum number of vectors required
2677  *       could be enabled.
2678  *      number of vectors which can be enabled otherwise (this number is smaller
2679  *       than VMXNET3_LINUX_MIN_MSIX_VECT)
2680  */
2681
2682 static int
2683 vmxnet3_acquire_msix_vectors(struct vmxnet3_adapter *adapter,
2684                              int vectors)
2685 {
2686         int err = 0, vector_threshold;
2687         vector_threshold = VMXNET3_LINUX_MIN_MSIX_VECT;
2688
2689         while (vectors >= vector_threshold) {
2690                 err = pci_enable_msix(adapter->pdev, adapter->intr.msix_entries,
2691                                       vectors);
2692                 if (!err) {
2693                         adapter->intr.num_intrs = vectors;
2694                         return 0;
2695                 } else if (err < 0) {
2696                         printk(KERN_ERR "Failed to enable MSI-X for %s, error"
2697                                " %d\n", adapter->netdev->name, err);
2698                         vectors = 0;
2699                 } else if (err < vector_threshold) {
2700                         break;
2701                 } else {
2702                         /* If fails to enable required number of MSI-x vectors
2703                          * try enabling 3 of them. One each for rx, tx and event
2704                          */
2705                         vectors = vector_threshold;
2706                         printk(KERN_ERR "Failed to enable %d MSI-X for %s, try"
2707                                " %d instead\n", vectors, adapter->netdev->name,
2708                                vector_threshold);
2709                 }
2710         }
2711
2712         printk(KERN_INFO "Number of MSI-X interrupts which can be allocatedi"
2713                " are lower than min threshold required.\n");
2714         return err;
2715 }
2716
2717
2718 #endif /* CONFIG_PCI_MSI */
2719
2720 static void
2721 vmxnet3_alloc_intr_resources(struct vmxnet3_adapter *adapter)
2722 {
2723         u32 cfg;
2724
2725         /* intr settings */
2726         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2727                                VMXNET3_CMD_GET_CONF_INTR);
2728         cfg = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
2729         adapter->intr.type = cfg & 0x3;
2730         adapter->intr.mask_mode = (cfg >> 2) & 0x3;
2731
2732         if (adapter->intr.type == VMXNET3_IT_AUTO) {
2733                 adapter->intr.type = VMXNET3_IT_MSIX;
2734         }
2735
2736 #ifdef CONFIG_PCI_MSI
2737         if (adapter->intr.type == VMXNET3_IT_MSIX) {
2738                 int vector, err = 0;
2739
2740                 adapter->intr.num_intrs = (adapter->share_intr ==
2741                                            VMXNET3_INTR_TXSHARE) ? 1 :
2742                                            adapter->num_tx_queues;
2743                 adapter->intr.num_intrs += (adapter->share_intr ==
2744                                            VMXNET3_INTR_BUDDYSHARE) ? 0 :
2745                                            adapter->num_rx_queues;
2746                 adapter->intr.num_intrs += 1;           /* for link event */
2747
2748                 adapter->intr.num_intrs = (adapter->intr.num_intrs >
2749                                            VMXNET3_LINUX_MIN_MSIX_VECT
2750                                            ? adapter->intr.num_intrs :
2751                                            VMXNET3_LINUX_MIN_MSIX_VECT);
2752
2753                 for (vector = 0; vector < adapter->intr.num_intrs; vector++)
2754                         adapter->intr.msix_entries[vector].entry = vector;
2755
2756                 err = vmxnet3_acquire_msix_vectors(adapter,
2757                                                    adapter->intr.num_intrs);
2758                 /* If we cannot allocate one MSIx vector per queue
2759                  * then limit the number of rx queues to 1
2760                  */
2761                 if (err == VMXNET3_LINUX_MIN_MSIX_VECT) {
2762                         if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE
2763                             || adapter->num_rx_queues != 2) {
2764                                 adapter->share_intr = VMXNET3_INTR_TXSHARE;
2765                                 printk(KERN_ERR "Number of rx queues : 1\n");
2766                                 adapter->num_rx_queues = 1;
2767                                 adapter->intr.num_intrs =
2768                                                 VMXNET3_LINUX_MIN_MSIX_VECT;
2769                         }
2770                         return;
2771                 }
2772                 if (!err)
2773                         return;
2774
2775                 /* If we cannot allocate MSIx vectors use only one rx queue */
2776                 printk(KERN_INFO "Failed to enable MSI-X for %s, error %d."
2777                        "#rx queues : 1, try MSI\n", adapter->netdev->name, err);
2778
2779                 adapter->intr.type = VMXNET3_IT_MSI;
2780         }
2781
2782         if (adapter->intr.type == VMXNET3_IT_MSI) {
2783                 int err;
2784                 err = pci_enable_msi(adapter->pdev);
2785                 if (!err) {
2786                         adapter->num_rx_queues = 1;
2787                         adapter->intr.num_intrs = 1;
2788                         return;
2789                 }
2790         }
2791 #endif /* CONFIG_PCI_MSI */
2792
2793         adapter->num_rx_queues = 1;
2794         printk(KERN_INFO "Using INTx interrupt, #Rx queues: 1.\n");
2795         adapter->intr.type = VMXNET3_IT_INTX;
2796
2797         /* INT-X related setting */
2798         adapter->intr.num_intrs = 1;
2799 }
2800
2801
2802 static void
2803 vmxnet3_free_intr_resources(struct vmxnet3_adapter *adapter)
2804 {
2805         if (adapter->intr.type == VMXNET3_IT_MSIX)
2806                 pci_disable_msix(adapter->pdev);
2807         else if (adapter->intr.type == VMXNET3_IT_MSI)
2808                 pci_disable_msi(adapter->pdev);
2809         else
2810                 BUG_ON(adapter->intr.type != VMXNET3_IT_INTX);
2811 }
2812
2813
2814 static void
2815 vmxnet3_tx_timeout(struct net_device *netdev)
2816 {
2817         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2818         adapter->tx_timeout_count++;
2819
2820         printk(KERN_ERR "%s: tx hang\n", adapter->netdev->name);
2821         schedule_work(&adapter->work);
2822         netif_wake_queue(adapter->netdev);
2823 }
2824
2825
2826 static void
2827 vmxnet3_reset_work(struct work_struct *data)
2828 {
2829         struct vmxnet3_adapter *adapter;
2830
2831         adapter = container_of(data, struct vmxnet3_adapter, work);
2832
2833         /* if another thread is resetting the device, no need to proceed */
2834         if (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2835                 return;
2836
2837         /* if the device is closed, we must leave it alone */
2838         rtnl_lock();
2839         if (netif_running(adapter->netdev)) {
2840                 printk(KERN_INFO "%s: resetting\n", adapter->netdev->name);
2841                 vmxnet3_quiesce_dev(adapter);
2842                 vmxnet3_reset_dev(adapter);
2843                 vmxnet3_activate_dev(adapter);
2844         } else {
2845                 printk(KERN_INFO "%s: already closed\n", adapter->netdev->name);
2846         }
2847         rtnl_unlock();
2848
2849         clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2850 }
2851
2852
2853 static int __devinit
2854 vmxnet3_probe_device(struct pci_dev *pdev,
2855                      const struct pci_device_id *id)
2856 {
2857         static const struct net_device_ops vmxnet3_netdev_ops = {
2858                 .ndo_open = vmxnet3_open,
2859                 .ndo_stop = vmxnet3_close,
2860                 .ndo_start_xmit = vmxnet3_xmit_frame,
2861                 .ndo_set_mac_address = vmxnet3_set_mac_addr,
2862                 .ndo_change_mtu = vmxnet3_change_mtu,
2863                 .ndo_get_stats = vmxnet3_get_stats,
2864                 .ndo_tx_timeout = vmxnet3_tx_timeout,
2865                 .ndo_set_multicast_list = vmxnet3_set_mc,
2866                 .ndo_vlan_rx_register = vmxnet3_vlan_rx_register,
2867                 .ndo_vlan_rx_add_vid = vmxnet3_vlan_rx_add_vid,
2868                 .ndo_vlan_rx_kill_vid = vmxnet3_vlan_rx_kill_vid,
2869 #ifdef CONFIG_NET_POLL_CONTROLLER
2870                 .ndo_poll_controller = vmxnet3_netpoll,
2871 #endif
2872         };
2873         int err;
2874         bool dma64 = false; /* stupid gcc */
2875         u32 ver;
2876         struct net_device *netdev;
2877         struct vmxnet3_adapter *adapter;
2878         u8 mac[ETH_ALEN];
2879         int size;
2880         int num_tx_queues;
2881         int num_rx_queues;
2882
2883 #ifdef VMXNET3_RSS
2884         if (enable_mq)
2885                 num_rx_queues = min(VMXNET3_DEVICE_MAX_RX_QUEUES,
2886                                     (int)num_online_cpus());
2887         else
2888 #endif
2889                 num_rx_queues = 1;
2890
2891         if (enable_mq)
2892                 num_tx_queues = min(VMXNET3_DEVICE_MAX_TX_QUEUES,
2893                                     (int)num_online_cpus());
2894         else
2895                 num_tx_queues = 1;
2896
2897         netdev = alloc_etherdev_mq(sizeof(struct vmxnet3_adapter),
2898                                    max(num_tx_queues, num_rx_queues));
2899         printk(KERN_INFO "# of Tx queues : %d, # of Rx queues : %d\n",
2900                num_tx_queues, num_rx_queues);
2901
2902         if (!netdev) {
2903                 printk(KERN_ERR "Failed to alloc ethernet device for adapter "
2904                         "%s\n", pci_name(pdev));
2905                 return -ENOMEM;
2906         }
2907
2908         pci_set_drvdata(pdev, netdev);
2909         adapter = netdev_priv(netdev);
2910         adapter->netdev = netdev;
2911         adapter->pdev = pdev;
2912
2913         adapter->shared = pci_alloc_consistent(adapter->pdev,
2914                           sizeof(struct Vmxnet3_DriverShared),
2915                           &adapter->shared_pa);
2916         if (!adapter->shared) {
2917                 printk(KERN_ERR "Failed to allocate memory for %s\n",
2918                         pci_name(pdev));
2919                 err = -ENOMEM;
2920                 goto err_alloc_shared;
2921         }
2922
2923         adapter->num_rx_queues = num_rx_queues;
2924         adapter->num_tx_queues = num_tx_queues;
2925
2926         size = sizeof(struct Vmxnet3_TxQueueDesc) * adapter->num_tx_queues;
2927         size += sizeof(struct Vmxnet3_RxQueueDesc) * adapter->num_rx_queues;
2928         adapter->tqd_start = pci_alloc_consistent(adapter->pdev, size,
2929                              &adapter->queue_desc_pa);
2930
2931         if (!adapter->tqd_start) {
2932                 printk(KERN_ERR "Failed to allocate memory for %s\n",
2933                         pci_name(pdev));
2934                 err = -ENOMEM;
2935                 goto err_alloc_queue_desc;
2936         }
2937         adapter->rqd_start = (struct Vmxnet3_RxQueueDesc *)(adapter->tqd_start +
2938                                                         adapter->num_tx_queues);
2939
2940         adapter->pm_conf = kmalloc(sizeof(struct Vmxnet3_PMConf), GFP_KERNEL);
2941         if (adapter->pm_conf == NULL) {
2942                 printk(KERN_ERR "Failed to allocate memory for %s\n",
2943                         pci_name(pdev));
2944                 err = -ENOMEM;
2945                 goto err_alloc_pm;
2946         }
2947
2948 #ifdef VMXNET3_RSS
2949
2950         adapter->rss_conf = kmalloc(sizeof(struct UPT1_RSSConf), GFP_KERNEL);
2951         if (adapter->rss_conf == NULL) {
2952                 printk(KERN_ERR "Failed to allocate memory for %s\n",
2953                        pci_name(pdev));
2954                 err = -ENOMEM;
2955                 goto err_alloc_rss;
2956         }
2957 #endif /* VMXNET3_RSS */
2958
2959         err = vmxnet3_alloc_pci_resources(adapter, &dma64);
2960         if (err < 0)
2961                 goto err_alloc_pci;
2962
2963         ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_VRRS);
2964         if (ver & 1) {
2965                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_VRRS, 1);
2966         } else {
2967                 printk(KERN_ERR "Incompatible h/w version (0x%x) for adapter"
2968                        " %s\n", ver, pci_name(pdev));
2969                 err = -EBUSY;
2970                 goto err_ver;
2971         }
2972
2973         ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_UVRS);
2974         if (ver & 1) {
2975                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_UVRS, 1);
2976         } else {
2977                 printk(KERN_ERR "Incompatible upt version (0x%x) for "
2978                        "adapter %s\n", ver, pci_name(pdev));
2979                 err = -EBUSY;
2980                 goto err_ver;
2981         }
2982
2983         vmxnet3_declare_features(adapter, dma64);
2984
2985         adapter->dev_number = atomic_read(&devices_found);
2986
2987          adapter->share_intr = irq_share_mode;
2988         if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE &&
2989             adapter->num_tx_queues != adapter->num_rx_queues)
2990                 adapter->share_intr = VMXNET3_INTR_DONTSHARE;
2991
2992         vmxnet3_alloc_intr_resources(adapter);
2993
2994 #ifdef VMXNET3_RSS
2995         if (adapter->num_rx_queues > 1 &&
2996             adapter->intr.type == VMXNET3_IT_MSIX) {
2997                 adapter->rss = true;
2998                 printk(KERN_INFO "RSS is enabled.\n");
2999         } else {
3000                 adapter->rss = false;
3001         }
3002 #endif
3003
3004         vmxnet3_read_mac_addr(adapter, mac);
3005         memcpy(netdev->dev_addr,  mac, netdev->addr_len);
3006
3007         netdev->netdev_ops = &vmxnet3_netdev_ops;
3008         vmxnet3_set_ethtool_ops(netdev);
3009         netdev->watchdog_timeo = 5 * HZ;
3010
3011         INIT_WORK(&adapter->work, vmxnet3_reset_work);
3012
3013         if (adapter->intr.type == VMXNET3_IT_MSIX) {
3014                 int i;
3015                 for (i = 0; i < adapter->num_rx_queues; i++) {
3016                         netif_napi_add(adapter->netdev,
3017                                        &adapter->rx_queue[i].napi,
3018                                        vmxnet3_poll_rx_only, 64);
3019                 }
3020         } else {
3021                 netif_napi_add(adapter->netdev, &adapter->rx_queue[0].napi,
3022                                vmxnet3_poll, 64);
3023         }
3024
3025         netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
3026         netif_set_real_num_rx_queues(adapter->netdev, adapter->num_rx_queues);
3027
3028         SET_NETDEV_DEV(netdev, &pdev->dev);
3029         err = register_netdev(netdev);
3030
3031         if (err) {
3032                 printk(KERN_ERR "Failed to register adapter %s\n",
3033                         pci_name(pdev));
3034                 goto err_register;
3035         }
3036
3037         set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
3038         vmxnet3_check_link(adapter, false);
3039         atomic_inc(&devices_found);
3040         return 0;
3041
3042 err_register:
3043         vmxnet3_free_intr_resources(adapter);
3044 err_ver:
3045         vmxnet3_free_pci_resources(adapter);
3046 err_alloc_pci:
3047 #ifdef VMXNET3_RSS
3048         kfree(adapter->rss_conf);
3049 err_alloc_rss:
3050 #endif
3051         kfree(adapter->pm_conf);
3052 err_alloc_pm:
3053         pci_free_consistent(adapter->pdev, size, adapter->tqd_start,
3054                             adapter->queue_desc_pa);
3055 err_alloc_queue_desc:
3056         pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_DriverShared),
3057                             adapter->shared, adapter->shared_pa);
3058 err_alloc_shared:
3059         pci_set_drvdata(pdev, NULL);
3060         free_netdev(netdev);
3061         return err;
3062 }
3063
3064
3065 static void __devexit
3066 vmxnet3_remove_device(struct pci_dev *pdev)
3067 {
3068         struct net_device *netdev = pci_get_drvdata(pdev);
3069         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3070         int size = 0;
3071         int num_rx_queues;
3072
3073 #ifdef VMXNET3_RSS
3074         if (enable_mq)
3075                 num_rx_queues = min(VMXNET3_DEVICE_MAX_RX_QUEUES,
3076                                     (int)num_online_cpus());
3077         else
3078 #endif
3079                 num_rx_queues = 1;
3080
3081         cancel_work_sync(&adapter->work);
3082
3083         unregister_netdev(netdev);
3084
3085         vmxnet3_free_intr_resources(adapter);
3086         vmxnet3_free_pci_resources(adapter);
3087 #ifdef VMXNET3_RSS
3088         kfree(adapter->rss_conf);
3089 #endif
3090         kfree(adapter->pm_conf);
3091
3092         size = sizeof(struct Vmxnet3_TxQueueDesc) * adapter->num_tx_queues;
3093         size += sizeof(struct Vmxnet3_RxQueueDesc) * num_rx_queues;
3094         pci_free_consistent(adapter->pdev, size, adapter->tqd_start,
3095                             adapter->queue_desc_pa);
3096         pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_DriverShared),
3097                             adapter->shared, adapter->shared_pa);
3098         free_netdev(netdev);
3099 }
3100
3101
3102 #ifdef CONFIG_PM
3103
3104 static int
3105 vmxnet3_suspend(struct device *device)
3106 {
3107         struct pci_dev *pdev = to_pci_dev(device);
3108         struct net_device *netdev = pci_get_drvdata(pdev);
3109         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3110         struct Vmxnet3_PMConf *pmConf;
3111         struct ethhdr *ehdr;
3112         struct arphdr *ahdr;
3113         u8 *arpreq;
3114         struct in_device *in_dev;
3115         struct in_ifaddr *ifa;
3116         int i = 0;
3117
3118         if (!netif_running(netdev))
3119                 return 0;
3120
3121         vmxnet3_disable_all_intrs(adapter);
3122         vmxnet3_free_irqs(adapter);
3123         vmxnet3_free_intr_resources(adapter);
3124
3125         netif_device_detach(netdev);
3126         netif_tx_stop_all_queues(netdev);
3127
3128         /* Create wake-up filters. */
3129         pmConf = adapter->pm_conf;
3130         memset(pmConf, 0, sizeof(*pmConf));
3131
3132         if (adapter->wol & WAKE_UCAST) {
3133                 pmConf->filters[i].patternSize = ETH_ALEN;
3134                 pmConf->filters[i].maskSize = 1;
3135                 memcpy(pmConf->filters[i].pattern, netdev->dev_addr, ETH_ALEN);
3136                 pmConf->filters[i].mask[0] = 0x3F; /* LSB ETH_ALEN bits */
3137
3138                 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_FILTER;
3139                 i++;
3140         }
3141
3142         if (adapter->wol & WAKE_ARP) {
3143                 in_dev = in_dev_get(netdev);
3144                 if (!in_dev)
3145                         goto skip_arp;
3146
3147                 ifa = (struct in_ifaddr *)in_dev->ifa_list;
3148                 if (!ifa)
3149                         goto skip_arp;
3150
3151                 pmConf->filters[i].patternSize = ETH_HLEN + /* Ethernet header*/
3152                         sizeof(struct arphdr) +         /* ARP header */
3153                         2 * ETH_ALEN +          /* 2 Ethernet addresses*/
3154                         2 * sizeof(u32);        /*2 IPv4 addresses */
3155                 pmConf->filters[i].maskSize =
3156                         (pmConf->filters[i].patternSize - 1) / 8 + 1;
3157
3158                 /* ETH_P_ARP in Ethernet header. */
3159                 ehdr = (struct ethhdr *)pmConf->filters[i].pattern;
3160                 ehdr->h_proto = htons(ETH_P_ARP);
3161
3162                 /* ARPOP_REQUEST in ARP header. */
3163                 ahdr = (struct arphdr *)&pmConf->filters[i].pattern[ETH_HLEN];
3164                 ahdr->ar_op = htons(ARPOP_REQUEST);
3165                 arpreq = (u8 *)(ahdr + 1);
3166
3167                 /* The Unicast IPv4 address in 'tip' field. */
3168                 arpreq += 2 * ETH_ALEN + sizeof(u32);
3169                 *(u32 *)arpreq = ifa->ifa_address;
3170
3171                 /* The mask for the relevant bits. */
3172                 pmConf->filters[i].mask[0] = 0x00;
3173                 pmConf->filters[i].mask[1] = 0x30; /* ETH_P_ARP */
3174                 pmConf->filters[i].mask[2] = 0x30; /* ARPOP_REQUEST */
3175                 pmConf->filters[i].mask[3] = 0x00;
3176                 pmConf->filters[i].mask[4] = 0xC0; /* IPv4 TIP */
3177                 pmConf->filters[i].mask[5] = 0x03; /* IPv4 TIP */
3178                 in_dev_put(in_dev);
3179
3180                 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_FILTER;
3181                 i++;
3182         }
3183
3184 skip_arp:
3185         if (adapter->wol & WAKE_MAGIC)
3186                 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_MAGIC;
3187
3188         pmConf->numFilters = i;
3189
3190         adapter->shared->devRead.pmConfDesc.confVer = cpu_to_le32(1);
3191         adapter->shared->devRead.pmConfDesc.confLen = cpu_to_le32(sizeof(
3192                                                                   *pmConf));
3193         adapter->shared->devRead.pmConfDesc.confPA = cpu_to_le64(virt_to_phys(
3194                                                                  pmConf));
3195
3196         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3197                                VMXNET3_CMD_UPDATE_PMCFG);
3198
3199         pci_save_state(pdev);
3200         pci_enable_wake(pdev, pci_choose_state(pdev, PMSG_SUSPEND),
3201                         adapter->wol);
3202         pci_disable_device(pdev);
3203         pci_set_power_state(pdev, pci_choose_state(pdev, PMSG_SUSPEND));
3204
3205         return 0;
3206 }
3207
3208
3209 static int
3210 vmxnet3_resume(struct device *device)
3211 {
3212         int err;
3213         struct pci_dev *pdev = to_pci_dev(device);
3214         struct net_device *netdev = pci_get_drvdata(pdev);
3215         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3216         struct Vmxnet3_PMConf *pmConf;
3217
3218         if (!netif_running(netdev))
3219                 return 0;
3220
3221         /* Destroy wake-up filters. */
3222         pmConf = adapter->pm_conf;
3223         memset(pmConf, 0, sizeof(*pmConf));
3224
3225         adapter->shared->devRead.pmConfDesc.confVer = cpu_to_le32(1);
3226         adapter->shared->devRead.pmConfDesc.confLen = cpu_to_le32(sizeof(
3227                                                                   *pmConf));
3228         adapter->shared->devRead.pmConfDesc.confPA = cpu_to_le64(virt_to_phys(
3229                                                                  pmConf));
3230
3231         netif_device_attach(netdev);
3232         pci_set_power_state(pdev, PCI_D0);
3233         pci_restore_state(pdev);
3234         err = pci_enable_device_mem(pdev);
3235         if (err != 0)
3236                 return err;
3237
3238         pci_enable_wake(pdev, PCI_D0, 0);
3239
3240         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3241                                VMXNET3_CMD_UPDATE_PMCFG);
3242         vmxnet3_alloc_intr_resources(adapter);
3243         vmxnet3_request_irqs(adapter);
3244         vmxnet3_enable_all_intrs(adapter);
3245
3246         return 0;
3247 }
3248
3249 static const struct dev_pm_ops vmxnet3_pm_ops = {
3250         .suspend = vmxnet3_suspend,
3251         .resume = vmxnet3_resume,
3252 };
3253 #endif
3254
3255 static struct pci_driver vmxnet3_driver = {
3256         .name           = vmxnet3_driver_name,
3257         .id_table       = vmxnet3_pciid_table,
3258         .probe          = vmxnet3_probe_device,
3259         .remove         = __devexit_p(vmxnet3_remove_device),
3260 #ifdef CONFIG_PM
3261         .driver.pm      = &vmxnet3_pm_ops,
3262 #endif
3263 };
3264
3265
3266 static int __init
3267 vmxnet3_init_module(void)
3268 {
3269         printk(KERN_INFO "%s - version %s\n", VMXNET3_DRIVER_DESC,
3270                 VMXNET3_DRIVER_VERSION_REPORT);
3271         return pci_register_driver(&vmxnet3_driver);
3272 }
3273
3274 module_init(vmxnet3_init_module);
3275
3276
3277 static void
3278 vmxnet3_exit_module(void)
3279 {
3280         pci_unregister_driver(&vmxnet3_driver);
3281 }
3282
3283 module_exit(vmxnet3_exit_module);
3284
3285 MODULE_AUTHOR("VMware, Inc.");
3286 MODULE_DESCRIPTION(VMXNET3_DRIVER_DESC);
3287 MODULE_LICENSE("GPL v2");
3288 MODULE_VERSION(VMXNET3_DRIVER_VERSION_STRING);