?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
vmxnet3: remove unused variable
[linux-2.6/libata-dev.git] / drivers / net / vmxnet3 / vmxnet3_drv.c
blob33097ec1c514c821d9da1c88da3c965c0508dafe
1 /*
2 * Linux driver for VMware's vmxnet3 ethernet NIC.
3 *
4 * Copyright (C) 2008-2009, VMware, Inc. All Rights Reserved.
5 *
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License as published by the
8 * Free Software Foundation; version 2 of the License and no later version.
9 *
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
13 * NON INFRINGEMENT. See the GNU General Public License for more
14 * details.
15 *
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * The full GNU General Public License is included in this distribution in
21 * the file called "COPYING".
22 *
23 * Maintained by: Shreyas Bhatewara <pv-drivers@vmware.com>
24 *
25 */
26
27 #include <net/ip6_checksum.h>
28
29 #include "vmxnet3_int.h"
30
31 char vmxnet3_driver_name[] = "vmxnet3";
32 #define VMXNET3_DRIVER_DESC "VMware vmxnet3 virtual NIC driver"
33
34 /*
35 * PCI Device ID Table
36 * Last entry must be all 0s
37 */
38 static DEFINE_PCI_DEVICE_TABLE(vmxnet3_pciid_table) = {
39 {PCI_VDEVICE(VMWARE, PCI_DEVICE_ID_VMWARE_VMXNET3)},
40 {0}
41 };
42
43 MODULE_DEVICE_TABLE(pci, vmxnet3_pciid_table);
44
45 static atomic_t devices_found;
46
47 #define VMXNET3_MAX_DEVICES 10
48 static int enable_mq = 1;
49 static int irq_share_mode;
50
51 static void
52 vmxnet3_write_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac);
53
54 /*
55 * Enable/Disable the given intr
56 */
57 static void
58 vmxnet3_enable_intr(struct vmxnet3_adapter *adapter, unsigned intr_idx)
59 {
60 VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_IMR + intr_idx * 8, 0);
61 }
62
63
64 static void
65 vmxnet3_disable_intr(struct vmxnet3_adapter *adapter, unsigned intr_idx)
66 {
67 VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_IMR + intr_idx * 8, 1);
68 }
69
70
71 /*
72 * Enable/Disable all intrs used by the device
73 */
74 static void
75 vmxnet3_enable_all_intrs(struct vmxnet3_adapter *adapter)
76 {
77 int i;
78
79 for (i = 0; i < adapter->intr.num_intrs; i++)
80 vmxnet3_enable_intr(adapter, i);
81 adapter->shared->devRead.intrConf.intrCtrl &=
82 cpu_to_le32(~VMXNET3_IC_DISABLE_ALL);
83 }
84
85
86 static void
87 vmxnet3_disable_all_intrs(struct vmxnet3_adapter *adapter)
88 {
89 int i;
90
91 adapter->shared->devRead.intrConf.intrCtrl |=
92 cpu_to_le32(VMXNET3_IC_DISABLE_ALL);
93 for (i = 0; i < adapter->intr.num_intrs; i++)
94 vmxnet3_disable_intr(adapter, i);
95 }
96
97
98 static void
99 vmxnet3_ack_events(struct vmxnet3_adapter *adapter, u32 events)
100 {
101 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_ECR, events);
102 }
103
104
105 static bool
106 vmxnet3_tq_stopped(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
107 {
108 return tq->stopped;
109 }
110
111
112 static void
113 vmxnet3_tq_start(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
114 {
115 tq->stopped = false;
116 netif_start_subqueue(adapter->netdev, tq - adapter->tx_queue);
117 }
118
119
120 static void
121 vmxnet3_tq_wake(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
122 {
123 tq->stopped = false;
124 netif_wake_subqueue(adapter->netdev, (tq - adapter->tx_queue));
125 }
126
127
128 static void
129 vmxnet3_tq_stop(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
130 {
131 tq->stopped = true;
132 tq->num_stop++;
133 netif_stop_subqueue(adapter->netdev, (tq - adapter->tx_queue));
134 }
135
136
137 /*
138 * Check the link state. This may start or stop the tx queue.
139 */
140 static void
141 vmxnet3_check_link(struct vmxnet3_adapter *adapter, bool affectTxQueue)
142 {
143 u32 ret;
144 int i;
145 unsigned long flags;
146
147 spin_lock_irqsave(&adapter->cmd_lock, flags);
148 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, VMXNET3_CMD_GET_LINK);
149 ret = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
150 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
151
152 adapter->link_speed = ret >> 16;
153 if (ret & 1) { /* Link is up. */
154 printk(KERN_INFO "%s: NIC Link is Up %d Mbps\n",
155 adapter->netdev->name, adapter->link_speed);
156 if (!netif_carrier_ok(adapter->netdev))
157 netif_carrier_on(adapter->netdev);
158
159 if (affectTxQueue) {
160 for (i = 0; i < adapter->num_tx_queues; i++)
161 vmxnet3_tq_start(&adapter->tx_queue[i],
162 adapter);
163 }
164 } else {
165 printk(KERN_INFO "%s: NIC Link is Down\n",
166 adapter->netdev->name);
167 if (netif_carrier_ok(adapter->netdev))
168 netif_carrier_off(adapter->netdev);
169
170 if (affectTxQueue) {
171 for (i = 0; i < adapter->num_tx_queues; i++)
172 vmxnet3_tq_stop(&adapter->tx_queue[i], adapter);
173 }
174 }
175 }
176
177 static void
178 vmxnet3_process_events(struct vmxnet3_adapter *adapter)
179 {
180 int i;
181 unsigned long flags;
182 u32 events = le32_to_cpu(adapter->shared->ecr);
183 if (!events)
184 return;
185
186 vmxnet3_ack_events(adapter, events);
187
188 /* Check if link state has changed */
189 if (events & VMXNET3_ECR_LINK)
190 vmxnet3_check_link(adapter, true);
191
192 /* Check if there is an error on xmit/recv queues */
193 if (events & (VMXNET3_ECR_TQERR | VMXNET3_ECR_RQERR)) {
194 spin_lock_irqsave(&adapter->cmd_lock, flags);
195 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
196 VMXNET3_CMD_GET_QUEUE_STATUS);
197 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
198
199 for (i = 0; i < adapter->num_tx_queues; i++)
200 if (adapter->tqd_start[i].status.stopped)
201 dev_err(&adapter->netdev->dev,
202 "%s: tq[%d] error 0x%x\n",
203 adapter->netdev->name, i, le32_to_cpu(
204 adapter->tqd_start[i].status.error));
205 for (i = 0; i < adapter->num_rx_queues; i++)
206 if (adapter->rqd_start[i].status.stopped)
207 dev_err(&adapter->netdev->dev,
208 "%s: rq[%d] error 0x%x\n",
209 adapter->netdev->name, i,
210 adapter->rqd_start[i].status.error);
211
212 schedule_work(&adapter->work);
213 }
214 }
215
216 #ifdef __BIG_ENDIAN_BITFIELD
217 /*
218 * The device expects the bitfields in shared structures to be written in
219 * little endian. When CPU is big endian, the following routines are used to
220 * correctly read and write into ABI.
221 * The general technique used here is : double word bitfields are defined in
222 * opposite order for big endian architecture. Then before reading them in
223 * driver the complete double word is translated using le32_to_cpu. Similarly
224 * After the driver writes into bitfields, cpu_to_le32 is used to translate the
225 * double words into required format.
226 * In order to avoid touching bits in shared structure more than once, temporary
227 * descriptors are used. These are passed as srcDesc to following functions.
228 */
229 static void vmxnet3_RxDescToCPU(const struct Vmxnet3_RxDesc *srcDesc,
230 struct Vmxnet3_RxDesc *dstDesc)
231 {
232 u32 *src = (u32 *)srcDesc + 2;
233 u32 *dst = (u32 *)dstDesc + 2;
234 dstDesc->addr = le64_to_cpu(srcDesc->addr);
235 *dst = le32_to_cpu(*src);
236 dstDesc->ext1 = le32_to_cpu(srcDesc->ext1);
237 }
238
239 static void vmxnet3_TxDescToLe(const struct Vmxnet3_TxDesc *srcDesc,
240 struct Vmxnet3_TxDesc *dstDesc)
241 {
242 int i;
243 u32 *src = (u32 *)(srcDesc + 1);
244 u32 *dst = (u32 *)(dstDesc + 1);
245
246 /* Working backwards so that the gen bit is set at the end. */
247 for (i = 2; i > 0; i--) {
248 src--;
249 dst--;
250 *dst = cpu_to_le32(*src);
251 }
252 }
253
254
255 static void vmxnet3_RxCompToCPU(const struct Vmxnet3_RxCompDesc *srcDesc,
256 struct Vmxnet3_RxCompDesc *dstDesc)
257 {
258 int i = 0;
259 u32 *src = (u32 *)srcDesc;
260 u32 *dst = (u32 *)dstDesc;
261 for (i = 0; i < sizeof(struct Vmxnet3_RxCompDesc) / sizeof(u32); i++) {
262 *dst = le32_to_cpu(*src);
263 src++;
264 dst++;
265 }
266 }
267
268
269 /* Used to read bitfield values from double words. */
270 static u32 get_bitfield32(const __le32 *bitfield, u32 pos, u32 size)
271 {
272 u32 temp = le32_to_cpu(*bitfield);
273 u32 mask = ((1 << size) - 1) << pos;
274 temp &= mask;
275 temp >>= pos;
276 return temp;
277 }
278
279
280
281 #endif /* __BIG_ENDIAN_BITFIELD */
282
283 #ifdef __BIG_ENDIAN_BITFIELD
284
285 # define VMXNET3_TXDESC_GET_GEN(txdesc) get_bitfield32(((const __le32 *) \
286 txdesc) + VMXNET3_TXD_GEN_DWORD_SHIFT, \
287 VMXNET3_TXD_GEN_SHIFT, VMXNET3_TXD_GEN_SIZE)
288 # define VMXNET3_TXDESC_GET_EOP(txdesc) get_bitfield32(((const __le32 *) \
289 txdesc) + VMXNET3_TXD_EOP_DWORD_SHIFT, \
290 VMXNET3_TXD_EOP_SHIFT, VMXNET3_TXD_EOP_SIZE)
291 # define VMXNET3_TCD_GET_GEN(tcd) get_bitfield32(((const __le32 *)tcd) + \
292 VMXNET3_TCD_GEN_DWORD_SHIFT, VMXNET3_TCD_GEN_SHIFT, \
293 VMXNET3_TCD_GEN_SIZE)
294 # define VMXNET3_TCD_GET_TXIDX(tcd) get_bitfield32((const __le32 *)tcd, \
295 VMXNET3_TCD_TXIDX_SHIFT, VMXNET3_TCD_TXIDX_SIZE)
296 # define vmxnet3_getRxComp(dstrcd, rcd, tmp) do { \
297 (dstrcd) = (tmp); \
298 vmxnet3_RxCompToCPU((rcd), (tmp)); \
299 } while (0)
300 # define vmxnet3_getRxDesc(dstrxd, rxd, tmp) do { \
301 (dstrxd) = (tmp); \
302 vmxnet3_RxDescToCPU((rxd), (tmp)); \
303 } while (0)
304
305 #else
306
307 # define VMXNET3_TXDESC_GET_GEN(txdesc) ((txdesc)->gen)
308 # define VMXNET3_TXDESC_GET_EOP(txdesc) ((txdesc)->eop)
309 # define VMXNET3_TCD_GET_GEN(tcd) ((tcd)->gen)
310 # define VMXNET3_TCD_GET_TXIDX(tcd) ((tcd)->txdIdx)
311 # define vmxnet3_getRxComp(dstrcd, rcd, tmp) (dstrcd) = (rcd)
312 # define vmxnet3_getRxDesc(dstrxd, rxd, tmp) (dstrxd) = (rxd)
313
314 #endif /* __BIG_ENDIAN_BITFIELD */
315
316
317 static void
318 vmxnet3_unmap_tx_buf(struct vmxnet3_tx_buf_info *tbi,
319 struct pci_dev *pdev)
320 {
321 if (tbi->map_type == VMXNET3_MAP_SINGLE)
322 pci_unmap_single(pdev, tbi->dma_addr, tbi->len,
323 PCI_DMA_TODEVICE);
324 else if (tbi->map_type == VMXNET3_MAP_PAGE)
325 pci_unmap_page(pdev, tbi->dma_addr, tbi->len,
326 PCI_DMA_TODEVICE);
327 else
328 BUG_ON(tbi->map_type != VMXNET3_MAP_NONE);
329
330 tbi->map_type = VMXNET3_MAP_NONE; /* to help debugging */
331 }
332
333
334 static int
335 vmxnet3_unmap_pkt(u32 eop_idx, struct vmxnet3_tx_queue *tq,
336 struct pci_dev *pdev, struct vmxnet3_adapter *adapter)
337 {
338 struct sk_buff *skb;
339 int entries = 0;
340
341 /* no out of order completion */
342 BUG_ON(tq->buf_info[eop_idx].sop_idx != tq->tx_ring.next2comp);
343 BUG_ON(VMXNET3_TXDESC_GET_EOP(&(tq->tx_ring.base[eop_idx].txd)) != 1);
344
345 skb = tq->buf_info[eop_idx].skb;
346 BUG_ON(skb == NULL);
347 tq->buf_info[eop_idx].skb = NULL;
348
349 VMXNET3_INC_RING_IDX_ONLY(eop_idx, tq->tx_ring.size);
350
351 while (tq->tx_ring.next2comp != eop_idx) {
352 vmxnet3_unmap_tx_buf(tq->buf_info + tq->tx_ring.next2comp,
353 pdev);
354
355 /* update next2comp w/o tx_lock. Since we are marking more,
356 * instead of less, tx ring entries avail, the worst case is
357 * that the tx routine incorrectly re-queues a pkt due to
358 * insufficient tx ring entries.
359 */
360 vmxnet3_cmd_ring_adv_next2comp(&tq->tx_ring);
361 entries++;
362 }
363
364 dev_kfree_skb_any(skb);
365 return entries;
366 }
367
368
369 static int
370 vmxnet3_tq_tx_complete(struct vmxnet3_tx_queue *tq,
371 struct vmxnet3_adapter *adapter)
372 {
373 int completed = 0;
374 union Vmxnet3_GenericDesc *gdesc;
375
376 gdesc = tq->comp_ring.base + tq->comp_ring.next2proc;
377 while (VMXNET3_TCD_GET_GEN(&gdesc->tcd) == tq->comp_ring.gen) {
378 completed += vmxnet3_unmap_pkt(VMXNET3_TCD_GET_TXIDX(
379 &gdesc->tcd), tq, adapter->pdev,
380 adapter);
381
382 vmxnet3_comp_ring_adv_next2proc(&tq->comp_ring);
383 gdesc = tq->comp_ring.base + tq->comp_ring.next2proc;
384 }
385
386 if (completed) {
387 spin_lock(&tq->tx_lock);
388 if (unlikely(vmxnet3_tq_stopped(tq, adapter) &&
389 vmxnet3_cmd_ring_desc_avail(&tq->tx_ring) >
390 VMXNET3_WAKE_QUEUE_THRESHOLD(tq) &&
391 netif_carrier_ok(adapter->netdev))) {
392 vmxnet3_tq_wake(tq, adapter);
393 }
394 spin_unlock(&tq->tx_lock);
395 }
396 return completed;
397 }
398
399
400 static void
401 vmxnet3_tq_cleanup(struct vmxnet3_tx_queue *tq,
402 struct vmxnet3_adapter *adapter)
403 {
404 int i;
405
406 while (tq->tx_ring.next2comp != tq->tx_ring.next2fill) {
407 struct vmxnet3_tx_buf_info *tbi;
408
409 tbi = tq->buf_info + tq->tx_ring.next2comp;
410
411 vmxnet3_unmap_tx_buf(tbi, adapter->pdev);
412 if (tbi->skb) {
413 dev_kfree_skb_any(tbi->skb);
414 tbi->skb = NULL;
415 }
416 vmxnet3_cmd_ring_adv_next2comp(&tq->tx_ring);
417 }
418
419 /* sanity check, verify all buffers are indeed unmapped and freed */
420 for (i = 0; i < tq->tx_ring.size; i++) {
421 BUG_ON(tq->buf_info[i].skb != NULL ||
422 tq->buf_info[i].map_type != VMXNET3_MAP_NONE);
423 }
424
425 tq->tx_ring.gen = VMXNET3_INIT_GEN;
426 tq->tx_ring.next2fill = tq->tx_ring.next2comp = 0;
427
428 tq->comp_ring.gen = VMXNET3_INIT_GEN;
429 tq->comp_ring.next2proc = 0;
430 }
431
432
433 static void
434 vmxnet3_tq_destroy(struct vmxnet3_tx_queue *tq,
435 struct vmxnet3_adapter *adapter)
436 {
437 if (tq->tx_ring.base) {
438 pci_free_consistent(adapter->pdev, tq->tx_ring.size *
439 sizeof(struct Vmxnet3_TxDesc),
440 tq->tx_ring.base, tq->tx_ring.basePA);
441 tq->tx_ring.base = NULL;
442 }
443 if (tq->data_ring.base) {
444 pci_free_consistent(adapter->pdev, tq->data_ring.size *
445 sizeof(struct Vmxnet3_TxDataDesc),
446 tq->data_ring.base, tq->data_ring.basePA);
447 tq->data_ring.base = NULL;
448 }
449 if (tq->comp_ring.base) {
450 pci_free_consistent(adapter->pdev, tq->comp_ring.size *
451 sizeof(struct Vmxnet3_TxCompDesc),
452 tq->comp_ring.base, tq->comp_ring.basePA);
453 tq->comp_ring.base = NULL;
454 }
455 kfree(tq->buf_info);
456 tq->buf_info = NULL;
457 }
458
459
460 /* Destroy all tx queues */
461 void
462 vmxnet3_tq_destroy_all(struct vmxnet3_adapter *adapter)
463 {
464 int i;
465
466 for (i = 0; i < adapter->num_tx_queues; i++)
467 vmxnet3_tq_destroy(&adapter->tx_queue[i], adapter);
468 }
469
470
471 static void
472 vmxnet3_tq_init(struct vmxnet3_tx_queue *tq,
473 struct vmxnet3_adapter *adapter)
474 {
475 int i;
476
477 /* reset the tx ring contents to 0 and reset the tx ring states */
478 memset(tq->tx_ring.base, 0, tq->tx_ring.size *
479 sizeof(struct Vmxnet3_TxDesc));
480 tq->tx_ring.next2fill = tq->tx_ring.next2comp = 0;
481 tq->tx_ring.gen = VMXNET3_INIT_GEN;
482
483 memset(tq->data_ring.base, 0, tq->data_ring.size *
484 sizeof(struct Vmxnet3_TxDataDesc));
485
486 /* reset the tx comp ring contents to 0 and reset comp ring states */
487 memset(tq->comp_ring.base, 0, tq->comp_ring.size *
488 sizeof(struct Vmxnet3_TxCompDesc));
489 tq->comp_ring.next2proc = 0;
490 tq->comp_ring.gen = VMXNET3_INIT_GEN;
491
492 /* reset the bookkeeping data */
493 memset(tq->buf_info, 0, sizeof(tq->buf_info[0]) * tq->tx_ring.size);
494 for (i = 0; i < tq->tx_ring.size; i++)
495 tq->buf_info[i].map_type = VMXNET3_MAP_NONE;
496
497 /* stats are not reset */
498 }
499
500
501 static int
502 vmxnet3_tq_create(struct vmxnet3_tx_queue *tq,
503 struct vmxnet3_adapter *adapter)
504 {
505 BUG_ON(tq->tx_ring.base || tq->data_ring.base ||
506 tq->comp_ring.base || tq->buf_info);
507
508 tq->tx_ring.base = pci_alloc_consistent(adapter->pdev, tq->tx_ring.size
509 * sizeof(struct Vmxnet3_TxDesc),
510 &tq->tx_ring.basePA);
511 if (!tq->tx_ring.base) {
512 printk(KERN_ERR "%s: failed to allocate tx ring\n",
513 adapter->netdev->name);
514 goto err;
515 }
516
517 tq->data_ring.base = pci_alloc_consistent(adapter->pdev,
518 tq->data_ring.size *
519 sizeof(struct Vmxnet3_TxDataDesc),
520 &tq->data_ring.basePA);
521 if (!tq->data_ring.base) {
522 printk(KERN_ERR "%s: failed to allocate data ring\n",
523 adapter->netdev->name);
524 goto err;
525 }
526
527 tq->comp_ring.base = pci_alloc_consistent(adapter->pdev,
528 tq->comp_ring.size *
529 sizeof(struct Vmxnet3_TxCompDesc),
530 &tq->comp_ring.basePA);
531 if (!tq->comp_ring.base) {
532 printk(KERN_ERR "%s: failed to allocate tx comp ring\n",
533 adapter->netdev->name);
534 goto err;
535 }
536
537 tq->buf_info = kcalloc(tq->tx_ring.size, sizeof(tq->buf_info[0]),
538 GFP_KERNEL);
539 if (!tq->buf_info) {
540 printk(KERN_ERR "%s: failed to allocate tx bufinfo\n",
541 adapter->netdev->name);
542 goto err;
543 }
544
545 return 0;
546
547 err:
548 vmxnet3_tq_destroy(tq, adapter);
549 return -ENOMEM;
550 }
551
552 static void
553 vmxnet3_tq_cleanup_all(struct vmxnet3_adapter *adapter)
554 {
555 int i;
556
557 for (i = 0; i < adapter->num_tx_queues; i++)
558 vmxnet3_tq_cleanup(&adapter->tx_queue[i], adapter);
559 }
560
561 /*
562 * starting from ring->next2fill, allocate rx buffers for the given ring
563 * of the rx queue and update the rx desc. stop after @num_to_alloc buffers
564 * are allocated or allocation fails
565 */
566
567 static int
568 vmxnet3_rq_alloc_rx_buf(struct vmxnet3_rx_queue *rq, u32 ring_idx,
569 int num_to_alloc, struct vmxnet3_adapter *adapter)
570 {
571 int num_allocated = 0;
572 struct vmxnet3_rx_buf_info *rbi_base = rq->buf_info[ring_idx];
573 struct vmxnet3_cmd_ring *ring = &rq->rx_ring[ring_idx];
574 u32 val;
575
576 while (num_allocated < num_to_alloc) {
577 struct vmxnet3_rx_buf_info *rbi;
578 union Vmxnet3_GenericDesc *gd;
579
580 rbi = rbi_base + ring->next2fill;
581 gd = ring->base + ring->next2fill;
582
583 if (rbi->buf_type == VMXNET3_RX_BUF_SKB) {
584 if (rbi->skb == NULL) {
585 rbi->skb = dev_alloc_skb(rbi->len +
586 NET_IP_ALIGN);
587 if (unlikely(rbi->skb == NULL)) {
588 rq->stats.rx_buf_alloc_failure++;
589 break;
590 }
591 rbi->skb->dev = adapter->netdev;
592
593 skb_reserve(rbi->skb, NET_IP_ALIGN);
594 rbi->dma_addr = pci_map_single(adapter->pdev,
595 rbi->skb->data, rbi->len,
596 PCI_DMA_FROMDEVICE);
597 } else {
598 /* rx buffer skipped by the device */
599 }
600 val = VMXNET3_RXD_BTYPE_HEAD << VMXNET3_RXD_BTYPE_SHIFT;
601 } else {
602 BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_PAGE ||
603 rbi->len != PAGE_SIZE);
604
605 if (rbi->page == NULL) {
606 rbi->page = alloc_page(GFP_ATOMIC);
607 if (unlikely(rbi->page == NULL)) {
608 rq->stats.rx_buf_alloc_failure++;
609 break;
610 }
611 rbi->dma_addr = pci_map_page(adapter->pdev,
612 rbi->page, 0, PAGE_SIZE,
613 PCI_DMA_FROMDEVICE);
614 } else {
615 /* rx buffers skipped by the device */
616 }
617 val = VMXNET3_RXD_BTYPE_BODY << VMXNET3_RXD_BTYPE_SHIFT;
618 }
619
620 BUG_ON(rbi->dma_addr == 0);
621 gd->rxd.addr = cpu_to_le64(rbi->dma_addr);
622 gd->dword[2] = cpu_to_le32((ring->gen << VMXNET3_RXD_GEN_SHIFT)
623 | val | rbi->len);
624
625 num_allocated++;
626 vmxnet3_cmd_ring_adv_next2fill(ring);
627 }
628 rq->uncommitted[ring_idx] += num_allocated;
629
630 dev_dbg(&adapter->netdev->dev,
631 "alloc_rx_buf: %d allocated, next2fill %u, next2comp "
632 "%u, uncommited %u\n", num_allocated, ring->next2fill,
633 ring->next2comp, rq->uncommitted[ring_idx]);
634
635 /* so that the device can distinguish a full ring and an empty ring */
636 BUG_ON(num_allocated != 0 && ring->next2fill == ring->next2comp);
637
638 return num_allocated;
639 }
640
641
642 static void
643 vmxnet3_append_frag(struct sk_buff *skb, struct Vmxnet3_RxCompDesc *rcd,
644 struct vmxnet3_rx_buf_info *rbi)
645 {
646 struct skb_frag_struct *frag = skb_shinfo(skb)->frags +
647 skb_shinfo(skb)->nr_frags;
648
649 BUG_ON(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS);
650
651 frag->page = rbi->page;
652 frag->page_offset = 0;
653 frag->size = rcd->len;
654 skb->data_len += frag->size;
655 skb_shinfo(skb)->nr_frags++;
656 }
657
658
659 static void
660 vmxnet3_map_pkt(struct sk_buff *skb, struct vmxnet3_tx_ctx *ctx,
661 struct vmxnet3_tx_queue *tq, struct pci_dev *pdev,
662 struct vmxnet3_adapter *adapter)
663 {
664 u32 dw2, len;
665 unsigned long buf_offset;
666 int i;
667 union Vmxnet3_GenericDesc *gdesc;
668 struct vmxnet3_tx_buf_info *tbi = NULL;
669
670 BUG_ON(ctx->copy_size > skb_headlen(skb));
671
672 /* use the previous gen bit for the SOP desc */
673 dw2 = (tq->tx_ring.gen ^ 0x1) << VMXNET3_TXD_GEN_SHIFT;
674
675 ctx->sop_txd = tq->tx_ring.base + tq->tx_ring.next2fill;
676 gdesc = ctx->sop_txd; /* both loops below can be skipped */
677
678 /* no need to map the buffer if headers are copied */
679 if (ctx->copy_size) {
680 ctx->sop_txd->txd.addr = cpu_to_le64(tq->data_ring.basePA +
681 tq->tx_ring.next2fill *
682 sizeof(struct Vmxnet3_TxDataDesc));
683 ctx->sop_txd->dword[2] = cpu_to_le32(dw2 | ctx->copy_size);
684 ctx->sop_txd->dword[3] = 0;
685
686 tbi = tq->buf_info + tq->tx_ring.next2fill;
687 tbi->map_type = VMXNET3_MAP_NONE;
688
689 dev_dbg(&adapter->netdev->dev,
690 "txd[%u]: 0x%Lx 0x%x 0x%x\n",
691 tq->tx_ring.next2fill,
692 le64_to_cpu(ctx->sop_txd->txd.addr),
693 ctx->sop_txd->dword[2], ctx->sop_txd->dword[3]);
694 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
695
696 /* use the right gen for non-SOP desc */
697 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
698 }
699
700 /* linear part can use multiple tx desc if it's big */
701 len = skb_headlen(skb) - ctx->copy_size;
702 buf_offset = ctx->copy_size;
703 while (len) {
704 u32 buf_size;
705
706 if (len < VMXNET3_MAX_TX_BUF_SIZE) {
707 buf_size = len;
708 dw2 |= len;
709 } else {
710 buf_size = VMXNET3_MAX_TX_BUF_SIZE;
711 /* spec says that for TxDesc.len, 0 == 2^14 */
712 }
713
714 tbi = tq->buf_info + tq->tx_ring.next2fill;
715 tbi->map_type = VMXNET3_MAP_SINGLE;
716 tbi->dma_addr = pci_map_single(adapter->pdev,
717 skb->data + buf_offset, buf_size,
718 PCI_DMA_TODEVICE);
719
720 tbi->len = buf_size;
721
722 gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
723 BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
724
725 gdesc->txd.addr = cpu_to_le64(tbi->dma_addr);
726 gdesc->dword[2] = cpu_to_le32(dw2);
727 gdesc->dword[3] = 0;
728
729 dev_dbg(&adapter->netdev->dev,
730 "txd[%u]: 0x%Lx 0x%x 0x%x\n",
731 tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr),
732 le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]);
733 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
734 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
735
736 len -= buf_size;
737 buf_offset += buf_size;
738 }
739
740 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
741 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
742
743 tbi = tq->buf_info + tq->tx_ring.next2fill;
744 tbi->map_type = VMXNET3_MAP_PAGE;
745 tbi->dma_addr = pci_map_page(adapter->pdev, frag->page,
746 frag->page_offset, frag->size,
747 PCI_DMA_TODEVICE);
748
749 tbi->len = frag->size;
750
751 gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
752 BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
753
754 gdesc->txd.addr = cpu_to_le64(tbi->dma_addr);
755 gdesc->dword[2] = cpu_to_le32(dw2 | frag->size);
756 gdesc->dword[3] = 0;
757
758 dev_dbg(&adapter->netdev->dev,
759 "txd[%u]: 0x%llu %u %u\n",
760 tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr),
761 le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]);
762 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
763 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
764 }
765
766 ctx->eop_txd = gdesc;
767
768 /* set the last buf_info for the pkt */
769 tbi->skb = skb;
770 tbi->sop_idx = ctx->sop_txd - tq->tx_ring.base;
771 }
772
773
774 /* Init all tx queues */
775 static void
776 vmxnet3_tq_init_all(struct vmxnet3_adapter *adapter)
777 {
778 int i;
779
780 for (i = 0; i < adapter->num_tx_queues; i++)
781 vmxnet3_tq_init(&adapter->tx_queue[i], adapter);
782 }
783
784
785 /*
786 * parse and copy relevant protocol headers:
787 * For a tso pkt, relevant headers are L2/3/4 including options
788 * For a pkt requesting csum offloading, they are L2/3 and may include L4
789 * if it's a TCP/UDP pkt
790 *
791 * Returns:
792 * -1: error happens during parsing
793 * 0: protocol headers parsed, but too big to be copied
794 * 1: protocol headers parsed and copied
795 *
796 * Other effects:
797 * 1. related *ctx fields are updated.
798 * 2. ctx->copy_size is # of bytes copied
799 * 3. the portion copied is guaranteed to be in the linear part
800 *
801 */
802 static int
803 vmxnet3_parse_and_copy_hdr(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
804 struct vmxnet3_tx_ctx *ctx,
805 struct vmxnet3_adapter *adapter)
806 {
807 struct Vmxnet3_TxDataDesc *tdd;
808
809 if (ctx->mss) { /* TSO */
810 ctx->eth_ip_hdr_size = skb_transport_offset(skb);
811 ctx->l4_hdr_size = ((struct tcphdr *)
812 skb_transport_header(skb))->doff * 4;
813 ctx->copy_size = ctx->eth_ip_hdr_size + ctx->l4_hdr_size;
814 } else {
815 if (skb->ip_summed == CHECKSUM_PARTIAL) {
816 ctx->eth_ip_hdr_size = skb_checksum_start_offset(skb);
817
818 if (ctx->ipv4) {
819 struct iphdr *iph = (struct iphdr *)
820 skb_network_header(skb);
821 if (iph->protocol == IPPROTO_TCP)
822 ctx->l4_hdr_size = ((struct tcphdr *)
823 skb_transport_header(skb))->doff * 4;
824 else if (iph->protocol == IPPROTO_UDP)
825 /*
826 * Use tcp header size so that bytes to
827 * be copied are more than required by
828 * the device.
829 */
830 ctx->l4_hdr_size =
831 sizeof(struct tcphdr);
832 else
833 ctx->l4_hdr_size = 0;
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 occurred, 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->netdev->features & NETIF_F_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 unsigned long flags;
1866
1867 if (grp) {
1868 /* add vlan rx stripping. */
1869 if (adapter->netdev->features & NETIF_F_HW_VLAN_RX) {
1870 int i;
1871 adapter->vlan_grp = grp;
1872
1873 /*
1874 * Clear entire vfTable; then enable untagged pkts.
1875 * Note: setting one entry in vfTable to non-zero turns
1876 * on VLAN rx filtering.
1877 */
1878 for (i = 0; i < VMXNET3_VFT_SIZE; i++)
1879 vfTable[i] = 0;
1880
1881 VMXNET3_SET_VFTABLE_ENTRY(vfTable, 0);
1882 spin_lock_irqsave(&adapter->cmd_lock, flags);
1883 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1884 VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1885 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
1886 } else {
1887 printk(KERN_ERR "%s: vlan_rx_register when device has "
1888 "no NETIF_F_HW_VLAN_RX\n", netdev->name);
1889 }
1890 } else {
1891 /* remove vlan rx stripping. */
1892 struct Vmxnet3_DSDevRead *devRead = &shared->devRead;
1893 adapter->vlan_grp = NULL;
1894
1895 if (devRead->misc.uptFeatures & UPT1_F_RXVLAN) {
1896 int i;
1897
1898 for (i = 0; i < VMXNET3_VFT_SIZE; i++) {
1899 /* clear entire vfTable; this also disables
1900 * VLAN rx filtering
1901 */
1902 vfTable[i] = 0;
1903 }
1904 spin_lock_irqsave(&adapter->cmd_lock, flags);
1905 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1906 VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1907 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
1908 }
1909 }
1910 }
1911
1912
1913 static void
1914 vmxnet3_restore_vlan(struct vmxnet3_adapter *adapter)
1915 {
1916 if (adapter->vlan_grp) {
1917 u16 vid;
1918 u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1919 bool activeVlan = false;
1920
1921 for (vid = 0; vid < VLAN_N_VID; vid++) {
1922 if (vlan_group_get_device(adapter->vlan_grp, vid)) {
1923 VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid);
1924 activeVlan = true;
1925 }
1926 }
1927 if (activeVlan) {
1928 /* continue to allow untagged pkts */
1929 VMXNET3_SET_VFTABLE_ENTRY(vfTable, 0);
1930 }
1931 }
1932 }
1933
1934
1935 static void
1936 vmxnet3_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1937 {
1938 struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1939 u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1940 unsigned long flags;
1941
1942 VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid);
1943 spin_lock_irqsave(&adapter->cmd_lock, flags);
1944 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1945 VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1946 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
1947 }
1948
1949
1950 static void
1951 vmxnet3_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1952 {
1953 struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1954 u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1955 unsigned long flags;
1956
1957 VMXNET3_CLEAR_VFTABLE_ENTRY(vfTable, vid);
1958 spin_lock_irqsave(&adapter->cmd_lock, flags);
1959 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1960 VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1961 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
1962 }
1963
1964
1965 static u8 *
1966 vmxnet3_copy_mc(struct net_device *netdev)
1967 {
1968 u8 *buf = NULL;
1969 u32 sz = netdev_mc_count(netdev) * ETH_ALEN;
1970
1971 /* struct Vmxnet3_RxFilterConf.mfTableLen is u16. */
1972 if (sz <= 0xffff) {
1973 /* We may be called with BH disabled */
1974 buf = kmalloc(sz, GFP_ATOMIC);
1975 if (buf) {
1976 struct netdev_hw_addr *ha;
1977 int i = 0;
1978
1979 netdev_for_each_mc_addr(ha, netdev)
1980 memcpy(buf + i++ * ETH_ALEN, ha->addr,
1981 ETH_ALEN);
1982 }
1983 }
1984 return buf;
1985 }
1986
1987
1988 static void
1989 vmxnet3_set_mc(struct net_device *netdev)
1990 {
1991 struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1992 unsigned long flags;
1993 struct Vmxnet3_RxFilterConf *rxConf =
1994 &adapter->shared->devRead.rxFilterConf;
1995 u8 *new_table = NULL;
1996 u32 new_mode = VMXNET3_RXM_UCAST;
1997
1998 if (netdev->flags & IFF_PROMISC)
1999 new_mode |= VMXNET3_RXM_PROMISC;
2000
2001 if (netdev->flags & IFF_BROADCAST)
2002 new_mode |= VMXNET3_RXM_BCAST;
2003
2004 if (netdev->flags & IFF_ALLMULTI)
2005 new_mode |= VMXNET3_RXM_ALL_MULTI;
2006 else
2007 if (!netdev_mc_empty(netdev)) {
2008 new_table = vmxnet3_copy_mc(netdev);
2009 if (new_table) {
2010 new_mode |= VMXNET3_RXM_MCAST;
2011 rxConf->mfTableLen = cpu_to_le16(
2012 netdev_mc_count(netdev) * ETH_ALEN);
2013 rxConf->mfTablePA = cpu_to_le64(virt_to_phys(
2014 new_table));
2015 } else {
2016 printk(KERN_INFO "%s: failed to copy mcast list"
2017 ", setting ALL_MULTI\n", netdev->name);
2018 new_mode |= VMXNET3_RXM_ALL_MULTI;
2019 }
2020 }
2021
2022
2023 if (!(new_mode & VMXNET3_RXM_MCAST)) {
2024 rxConf->mfTableLen = 0;
2025 rxConf->mfTablePA = 0;
2026 }
2027
2028 spin_lock_irqsave(&adapter->cmd_lock, flags);
2029 if (new_mode != rxConf->rxMode) {
2030 rxConf->rxMode = cpu_to_le32(new_mode);
2031 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2032 VMXNET3_CMD_UPDATE_RX_MODE);
2033 }
2034
2035 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2036 VMXNET3_CMD_UPDATE_MAC_FILTERS);
2037 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2038
2039 kfree(new_table);
2040 }
2041
2042 void
2043 vmxnet3_rq_destroy_all(struct vmxnet3_adapter *adapter)
2044 {
2045 int i;
2046
2047 for (i = 0; i < adapter->num_rx_queues; i++)
2048 vmxnet3_rq_destroy(&adapter->rx_queue[i], adapter);
2049 }
2050
2051
2052 /*
2053 * Set up driver_shared based on settings in adapter.
2054 */
2055
2056 static void
2057 vmxnet3_setup_driver_shared(struct vmxnet3_adapter *adapter)
2058 {
2059 struct Vmxnet3_DriverShared *shared = adapter->shared;
2060 struct Vmxnet3_DSDevRead *devRead = &shared->devRead;
2061 struct Vmxnet3_TxQueueConf *tqc;
2062 struct Vmxnet3_RxQueueConf *rqc;
2063 int i;
2064
2065 memset(shared, 0, sizeof(*shared));
2066
2067 /* driver settings */
2068 shared->magic = cpu_to_le32(VMXNET3_REV1_MAGIC);
2069 devRead->misc.driverInfo.version = cpu_to_le32(
2070 VMXNET3_DRIVER_VERSION_NUM);
2071 devRead->misc.driverInfo.gos.gosBits = (sizeof(void *) == 4 ?
2072 VMXNET3_GOS_BITS_32 : VMXNET3_GOS_BITS_64);
2073 devRead->misc.driverInfo.gos.gosType = VMXNET3_GOS_TYPE_LINUX;
2074 *((u32 *)&devRead->misc.driverInfo.gos) = cpu_to_le32(
2075 *((u32 *)&devRead->misc.driverInfo.gos));
2076 devRead->misc.driverInfo.vmxnet3RevSpt = cpu_to_le32(1);
2077 devRead->misc.driverInfo.uptVerSpt = cpu_to_le32(1);
2078
2079 devRead->misc.ddPA = cpu_to_le64(virt_to_phys(adapter));
2080 devRead->misc.ddLen = cpu_to_le32(sizeof(struct vmxnet3_adapter));
2081
2082 /* set up feature flags */
2083 if (adapter->netdev->features & NETIF_F_RXCSUM)
2084 devRead->misc.uptFeatures |= UPT1_F_RXCSUM;
2085
2086 if (adapter->netdev->features & NETIF_F_LRO) {
2087 devRead->misc.uptFeatures |= UPT1_F_LRO;
2088 devRead->misc.maxNumRxSG = cpu_to_le16(1 + MAX_SKB_FRAGS);
2089 }
2090 if (adapter->netdev->features & NETIF_F_HW_VLAN_RX)
2091 devRead->misc.uptFeatures |= UPT1_F_RXVLAN;
2092
2093 devRead->misc.mtu = cpu_to_le32(adapter->netdev->mtu);
2094 devRead->misc.queueDescPA = cpu_to_le64(adapter->queue_desc_pa);
2095 devRead->misc.queueDescLen = cpu_to_le32(
2096 adapter->num_tx_queues * sizeof(struct Vmxnet3_TxQueueDesc) +
2097 adapter->num_rx_queues * sizeof(struct Vmxnet3_RxQueueDesc));
2098
2099 /* tx queue settings */
2100 devRead->misc.numTxQueues = adapter->num_tx_queues;
2101 for (i = 0; i < adapter->num_tx_queues; i++) {
2102 struct vmxnet3_tx_queue *tq = &adapter->tx_queue[i];
2103 BUG_ON(adapter->tx_queue[i].tx_ring.base == NULL);
2104 tqc = &adapter->tqd_start[i].conf;
2105 tqc->txRingBasePA = cpu_to_le64(tq->tx_ring.basePA);
2106 tqc->dataRingBasePA = cpu_to_le64(tq->data_ring.basePA);
2107 tqc->compRingBasePA = cpu_to_le64(tq->comp_ring.basePA);
2108 tqc->ddPA = cpu_to_le64(virt_to_phys(tq->buf_info));
2109 tqc->txRingSize = cpu_to_le32(tq->tx_ring.size);
2110 tqc->dataRingSize = cpu_to_le32(tq->data_ring.size);
2111 tqc->compRingSize = cpu_to_le32(tq->comp_ring.size);
2112 tqc->ddLen = cpu_to_le32(
2113 sizeof(struct vmxnet3_tx_buf_info) *
2114 tqc->txRingSize);
2115 tqc->intrIdx = tq->comp_ring.intr_idx;
2116 }
2117
2118 /* rx queue settings */
2119 devRead->misc.numRxQueues = adapter->num_rx_queues;
2120 for (i = 0; i < adapter->num_rx_queues; i++) {
2121 struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
2122 rqc = &adapter->rqd_start[i].conf;
2123 rqc->rxRingBasePA[0] = cpu_to_le64(rq->rx_ring[0].basePA);
2124 rqc->rxRingBasePA[1] = cpu_to_le64(rq->rx_ring[1].basePA);
2125 rqc->compRingBasePA = cpu_to_le64(rq->comp_ring.basePA);
2126 rqc->ddPA = cpu_to_le64(virt_to_phys(
2127 rq->buf_info));
2128 rqc->rxRingSize[0] = cpu_to_le32(rq->rx_ring[0].size);
2129 rqc->rxRingSize[1] = cpu_to_le32(rq->rx_ring[1].size);
2130 rqc->compRingSize = cpu_to_le32(rq->comp_ring.size);
2131 rqc->ddLen = cpu_to_le32(
2132 sizeof(struct vmxnet3_rx_buf_info) *
2133 (rqc->rxRingSize[0] +
2134 rqc->rxRingSize[1]));
2135 rqc->intrIdx = rq->comp_ring.intr_idx;
2136 }
2137
2138 #ifdef VMXNET3_RSS
2139 memset(adapter->rss_conf, 0, sizeof(*adapter->rss_conf));
2140
2141 if (adapter->rss) {
2142 struct UPT1_RSSConf *rssConf = adapter->rss_conf;
2143 devRead->misc.uptFeatures |= UPT1_F_RSS;
2144 devRead->misc.numRxQueues = adapter->num_rx_queues;
2145 rssConf->hashType = UPT1_RSS_HASH_TYPE_TCP_IPV4 |
2146 UPT1_RSS_HASH_TYPE_IPV4 |
2147 UPT1_RSS_HASH_TYPE_TCP_IPV6 |
2148 UPT1_RSS_HASH_TYPE_IPV6;
2149 rssConf->hashFunc = UPT1_RSS_HASH_FUNC_TOEPLITZ;
2150 rssConf->hashKeySize = UPT1_RSS_MAX_KEY_SIZE;
2151 rssConf->indTableSize = VMXNET3_RSS_IND_TABLE_SIZE;
2152 get_random_bytes(&rssConf->hashKey[0], rssConf->hashKeySize);
2153 for (i = 0; i < rssConf->indTableSize; i++)
2154 rssConf->indTable[i] = i % adapter->num_rx_queues;
2155
2156 devRead->rssConfDesc.confVer = 1;
2157 devRead->rssConfDesc.confLen = sizeof(*rssConf);
2158 devRead->rssConfDesc.confPA = virt_to_phys(rssConf);
2159 }
2160
2161 #endif /* VMXNET3_RSS */
2162
2163 /* intr settings */
2164 devRead->intrConf.autoMask = adapter->intr.mask_mode ==
2165 VMXNET3_IMM_AUTO;
2166 devRead->intrConf.numIntrs = adapter->intr.num_intrs;
2167 for (i = 0; i < adapter->intr.num_intrs; i++)
2168 devRead->intrConf.modLevels[i] = adapter->intr.mod_levels[i];
2169
2170 devRead->intrConf.eventIntrIdx = adapter->intr.event_intr_idx;
2171 devRead->intrConf.intrCtrl |= cpu_to_le32(VMXNET3_IC_DISABLE_ALL);
2172
2173 /* rx filter settings */
2174 devRead->rxFilterConf.rxMode = 0;
2175 vmxnet3_restore_vlan(adapter);
2176 vmxnet3_write_mac_addr(adapter, adapter->netdev->dev_addr);
2177
2178 /* the rest are already zeroed */
2179 }
2180
2181
2182 int
2183 vmxnet3_activate_dev(struct vmxnet3_adapter *adapter)
2184 {
2185 int err, i;
2186 u32 ret;
2187 unsigned long flags;
2188
2189 dev_dbg(&adapter->netdev->dev, "%s: skb_buf_size %d, rx_buf_per_pkt %d,"
2190 " ring sizes %u %u %u\n", adapter->netdev->name,
2191 adapter->skb_buf_size, adapter->rx_buf_per_pkt,
2192 adapter->tx_queue[0].tx_ring.size,
2193 adapter->rx_queue[0].rx_ring[0].size,
2194 adapter->rx_queue[0].rx_ring[1].size);
2195
2196 vmxnet3_tq_init_all(adapter);
2197 err = vmxnet3_rq_init_all(adapter);
2198 if (err) {
2199 printk(KERN_ERR "Failed to init rx queue for %s: error %d\n",
2200 adapter->netdev->name, err);
2201 goto rq_err;
2202 }
2203
2204 err = vmxnet3_request_irqs(adapter);
2205 if (err) {
2206 printk(KERN_ERR "Failed to setup irq for %s: error %d\n",
2207 adapter->netdev->name, err);
2208 goto irq_err;
2209 }
2210
2211 vmxnet3_setup_driver_shared(adapter);
2212
2213 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, VMXNET3_GET_ADDR_LO(
2214 adapter->shared_pa));
2215 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, VMXNET3_GET_ADDR_HI(
2216 adapter->shared_pa));
2217 spin_lock_irqsave(&adapter->cmd_lock, flags);
2218 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2219 VMXNET3_CMD_ACTIVATE_DEV);
2220 ret = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
2221 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2222
2223 if (ret != 0) {
2224 printk(KERN_ERR "Failed to activate dev %s: error %u\n",
2225 adapter->netdev->name, ret);
2226 err = -EINVAL;
2227 goto activate_err;
2228 }
2229
2230 for (i = 0; i < adapter->num_rx_queues; i++) {
2231 VMXNET3_WRITE_BAR0_REG(adapter,
2232 VMXNET3_REG_RXPROD + i * VMXNET3_REG_ALIGN,
2233 adapter->rx_queue[i].rx_ring[0].next2fill);
2234 VMXNET3_WRITE_BAR0_REG(adapter, (VMXNET3_REG_RXPROD2 +
2235 (i * VMXNET3_REG_ALIGN)),
2236 adapter->rx_queue[i].rx_ring[1].next2fill);
2237 }
2238
2239 /* Apply the rx filter settins last. */
2240 vmxnet3_set_mc(adapter->netdev);
2241
2242 /*
2243 * Check link state when first activating device. It will start the
2244 * tx queue if the link is up.
2245 */
2246 vmxnet3_check_link(adapter, true);
2247 for (i = 0; i < adapter->num_rx_queues; i++)
2248 napi_enable(&adapter->rx_queue[i].napi);
2249 vmxnet3_enable_all_intrs(adapter);
2250 clear_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
2251 return 0;
2252
2253 activate_err:
2254 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, 0);
2255 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, 0);
2256 vmxnet3_free_irqs(adapter);
2257 irq_err:
2258 rq_err:
2259 /* free up buffers we allocated */
2260 vmxnet3_rq_cleanup_all(adapter);
2261 return err;
2262 }
2263
2264
2265 void
2266 vmxnet3_reset_dev(struct vmxnet3_adapter *adapter)
2267 {
2268 unsigned long flags;
2269 spin_lock_irqsave(&adapter->cmd_lock, flags);
2270 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, VMXNET3_CMD_RESET_DEV);
2271 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2272 }
2273
2274
2275 int
2276 vmxnet3_quiesce_dev(struct vmxnet3_adapter *adapter)
2277 {
2278 int i;
2279 unsigned long flags;
2280 if (test_and_set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state))
2281 return 0;
2282
2283
2284 spin_lock_irqsave(&adapter->cmd_lock, flags);
2285 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2286 VMXNET3_CMD_QUIESCE_DEV);
2287 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2288 vmxnet3_disable_all_intrs(adapter);
2289
2290 for (i = 0; i < adapter->num_rx_queues; i++)
2291 napi_disable(&adapter->rx_queue[i].napi);
2292 netif_tx_disable(adapter->netdev);
2293 adapter->link_speed = 0;
2294 netif_carrier_off(adapter->netdev);
2295
2296 vmxnet3_tq_cleanup_all(adapter);
2297 vmxnet3_rq_cleanup_all(adapter);
2298 vmxnet3_free_irqs(adapter);
2299 return 0;
2300 }
2301
2302
2303 static void
2304 vmxnet3_write_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac)
2305 {
2306 u32 tmp;
2307
2308 tmp = *(u32 *)mac;
2309 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACL, tmp);
2310
2311 tmp = (mac[5] << 8) | mac[4];
2312 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACH, tmp);
2313 }
2314
2315
2316 static int
2317 vmxnet3_set_mac_addr(struct net_device *netdev, void *p)
2318 {
2319 struct sockaddr *addr = p;
2320 struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2321
2322 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2323 vmxnet3_write_mac_addr(adapter, addr->sa_data);
2324
2325 return 0;
2326 }
2327
2328
2329 /* ==================== initialization and cleanup routines ============ */
2330
2331 static int
2332 vmxnet3_alloc_pci_resources(struct vmxnet3_adapter *adapter, bool *dma64)
2333 {
2334 int err;
2335 unsigned long mmio_start, mmio_len;
2336 struct pci_dev *pdev = adapter->pdev;
2337
2338 err = pci_enable_device(pdev);
2339 if (err) {
2340 printk(KERN_ERR "Failed to enable adapter %s: error %d\n",
2341 pci_name(pdev), err);
2342 return err;
2343 }
2344
2345 if (pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) == 0) {
2346 if (pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64)) != 0) {
2347 printk(KERN_ERR "pci_set_consistent_dma_mask failed "
2348 "for adapter %s\n", pci_name(pdev));
2349 err = -EIO;
2350 goto err_set_mask;
2351 }
2352 *dma64 = true;
2353 } else {
2354 if (pci_set_dma_mask(pdev, DMA_BIT_MASK(32)) != 0) {
2355 printk(KERN_ERR "pci_set_dma_mask failed for adapter "
2356 "%s\n", pci_name(pdev));
2357 err = -EIO;
2358 goto err_set_mask;
2359 }
2360 *dma64 = false;
2361 }
2362
2363 err = pci_request_selected_regions(pdev, (1 << 2) - 1,
2364 vmxnet3_driver_name);
2365 if (err) {
2366 printk(KERN_ERR "Failed to request region for adapter %s: "
2367 "error %d\n", pci_name(pdev), err);
2368 goto err_set_mask;
2369 }
2370
2371 pci_set_master(pdev);
2372
2373 mmio_start = pci_resource_start(pdev, 0);
2374 mmio_len = pci_resource_len(pdev, 0);
2375 adapter->hw_addr0 = ioremap(mmio_start, mmio_len);
2376 if (!adapter->hw_addr0) {
2377 printk(KERN_ERR "Failed to map bar0 for adapter %s\n",
2378 pci_name(pdev));
2379 err = -EIO;
2380 goto err_ioremap;
2381 }
2382
2383 mmio_start = pci_resource_start(pdev, 1);
2384 mmio_len = pci_resource_len(pdev, 1);
2385 adapter->hw_addr1 = ioremap(mmio_start, mmio_len);
2386 if (!adapter->hw_addr1) {
2387 printk(KERN_ERR "Failed to map bar1 for adapter %s\n",
2388 pci_name(pdev));
2389 err = -EIO;
2390 goto err_bar1;
2391 }
2392 return 0;
2393
2394 err_bar1:
2395 iounmap(adapter->hw_addr0);
2396 err_ioremap:
2397 pci_release_selected_regions(pdev, (1 << 2) - 1);
2398 err_set_mask:
2399 pci_disable_device(pdev);
2400 return err;
2401 }
2402
2403
2404 static void
2405 vmxnet3_free_pci_resources(struct vmxnet3_adapter *adapter)
2406 {
2407 BUG_ON(!adapter->pdev);
2408
2409 iounmap(adapter->hw_addr0);
2410 iounmap(adapter->hw_addr1);
2411 pci_release_selected_regions(adapter->pdev, (1 << 2) - 1);
2412 pci_disable_device(adapter->pdev);
2413 }
2414
2415
2416 static void
2417 vmxnet3_adjust_rx_ring_size(struct vmxnet3_adapter *adapter)
2418 {
2419 size_t sz, i, ring0_size, ring1_size, comp_size;
2420 struct vmxnet3_rx_queue *rq = &adapter->rx_queue[0];
2421
2422
2423 if (adapter->netdev->mtu <= VMXNET3_MAX_SKB_BUF_SIZE -
2424 VMXNET3_MAX_ETH_HDR_SIZE) {
2425 adapter->skb_buf_size = adapter->netdev->mtu +
2426 VMXNET3_MAX_ETH_HDR_SIZE;
2427 if (adapter->skb_buf_size < VMXNET3_MIN_T0_BUF_SIZE)
2428 adapter->skb_buf_size = VMXNET3_MIN_T0_BUF_SIZE;
2429
2430 adapter->rx_buf_per_pkt = 1;
2431 } else {
2432 adapter->skb_buf_size = VMXNET3_MAX_SKB_BUF_SIZE;
2433 sz = adapter->netdev->mtu - VMXNET3_MAX_SKB_BUF_SIZE +
2434 VMXNET3_MAX_ETH_HDR_SIZE;
2435 adapter->rx_buf_per_pkt = 1 + (sz + PAGE_SIZE - 1) / PAGE_SIZE;
2436 }
2437
2438 /*
2439 * for simplicity, force the ring0 size to be a multiple of
2440 * rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN
2441 */
2442 sz = adapter->rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN;
2443 ring0_size = adapter->rx_queue[0].rx_ring[0].size;
2444 ring0_size = (ring0_size + sz - 1) / sz * sz;
2445 ring0_size = min_t(u32, ring0_size, VMXNET3_RX_RING_MAX_SIZE /
2446 sz * sz);
2447 ring1_size = adapter->rx_queue[0].rx_ring[1].size;
2448 comp_size = ring0_size + ring1_size;
2449
2450 for (i = 0; i < adapter->num_rx_queues; i++) {
2451 rq = &adapter->rx_queue[i];
2452 rq->rx_ring[0].size = ring0_size;
2453 rq->rx_ring[1].size = ring1_size;
2454 rq->comp_ring.size = comp_size;
2455 }
2456 }
2457
2458
2459 int
2460 vmxnet3_create_queues(struct vmxnet3_adapter *adapter, u32 tx_ring_size,
2461 u32 rx_ring_size, u32 rx_ring2_size)
2462 {
2463 int err = 0, i;
2464
2465 for (i = 0; i < adapter->num_tx_queues; i++) {
2466 struct vmxnet3_tx_queue *tq = &adapter->tx_queue[i];
2467 tq->tx_ring.size = tx_ring_size;
2468 tq->data_ring.size = tx_ring_size;
2469 tq->comp_ring.size = tx_ring_size;
2470 tq->shared = &adapter->tqd_start[i].ctrl;
2471 tq->stopped = true;
2472 tq->adapter = adapter;
2473 tq->qid = i;
2474 err = vmxnet3_tq_create(tq, adapter);
2475 /*
2476 * Too late to change num_tx_queues. We cannot do away with
2477 * lesser number of queues than what we asked for
2478 */
2479 if (err)
2480 goto queue_err;
2481 }
2482
2483 adapter->rx_queue[0].rx_ring[0].size = rx_ring_size;
2484 adapter->rx_queue[0].rx_ring[1].size = rx_ring2_size;
2485 vmxnet3_adjust_rx_ring_size(adapter);
2486 for (i = 0; i < adapter->num_rx_queues; i++) {
2487 struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
2488 /* qid and qid2 for rx queues will be assigned later when num
2489 * of rx queues is finalized after allocating intrs */
2490 rq->shared = &adapter->rqd_start[i].ctrl;
2491 rq->adapter = adapter;
2492 err = vmxnet3_rq_create(rq, adapter);
2493 if (err) {
2494 if (i == 0) {
2495 printk(KERN_ERR "Could not allocate any rx"
2496 "queues. Aborting.\n");
2497 goto queue_err;
2498 } else {
2499 printk(KERN_INFO "Number of rx queues changed "
2500 "to : %d.\n", i);
2501 adapter->num_rx_queues = i;
2502 err = 0;
2503 break;
2504 }
2505 }
2506 }
2507 return err;
2508 queue_err:
2509 vmxnet3_tq_destroy_all(adapter);
2510 return err;
2511 }
2512
2513 static int
2514 vmxnet3_open(struct net_device *netdev)
2515 {
2516 struct vmxnet3_adapter *adapter;
2517 int err, i;
2518
2519 adapter = netdev_priv(netdev);
2520
2521 for (i = 0; i < adapter->num_tx_queues; i++)
2522 spin_lock_init(&adapter->tx_queue[i].tx_lock);
2523
2524 err = vmxnet3_create_queues(adapter, VMXNET3_DEF_TX_RING_SIZE,
2525 VMXNET3_DEF_RX_RING_SIZE,
2526 VMXNET3_DEF_RX_RING_SIZE);
2527 if (err)
2528 goto queue_err;
2529
2530 err = vmxnet3_activate_dev(adapter);
2531 if (err)
2532 goto activate_err;
2533
2534 return 0;
2535
2536 activate_err:
2537 vmxnet3_rq_destroy_all(adapter);
2538 vmxnet3_tq_destroy_all(adapter);
2539 queue_err:
2540 return err;
2541 }
2542
2543
2544 static int
2545 vmxnet3_close(struct net_device *netdev)
2546 {
2547 struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2548
2549 /*
2550 * Reset_work may be in the middle of resetting the device, wait for its
2551 * completion.
2552 */
2553 while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2554 msleep(1);
2555
2556 vmxnet3_quiesce_dev(adapter);
2557
2558 vmxnet3_rq_destroy_all(adapter);
2559 vmxnet3_tq_destroy_all(adapter);
2560
2561 clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2562
2563
2564 return 0;
2565 }
2566
2567
2568 void
2569 vmxnet3_force_close(struct vmxnet3_adapter *adapter)
2570 {
2571 int i;
2572
2573 /*
2574 * we must clear VMXNET3_STATE_BIT_RESETTING, otherwise
2575 * vmxnet3_close() will deadlock.
2576 */
2577 BUG_ON(test_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state));
2578
2579 /* we need to enable NAPI, otherwise dev_close will deadlock */
2580 for (i = 0; i < adapter->num_rx_queues; i++)
2581 napi_enable(&adapter->rx_queue[i].napi);
2582 dev_close(adapter->netdev);
2583 }
2584
2585
2586 static int
2587 vmxnet3_change_mtu(struct net_device *netdev, int new_mtu)
2588 {
2589 struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2590 int err = 0;
2591
2592 if (new_mtu < VMXNET3_MIN_MTU || new_mtu > VMXNET3_MAX_MTU)
2593 return -EINVAL;
2594
2595 netdev->mtu = new_mtu;
2596
2597 /*
2598 * Reset_work may be in the middle of resetting the device, wait for its
2599 * completion.
2600 */
2601 while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2602 msleep(1);
2603
2604 if (netif_running(netdev)) {
2605 vmxnet3_quiesce_dev(adapter);
2606 vmxnet3_reset_dev(adapter);
2607
2608 /* we need to re-create the rx queue based on the new mtu */
2609 vmxnet3_rq_destroy_all(adapter);
2610 vmxnet3_adjust_rx_ring_size(adapter);
2611 err = vmxnet3_rq_create_all(adapter);
2612 if (err) {
2613 printk(KERN_ERR "%s: failed to re-create rx queues,"
2614 " error %d. Closing it.\n", netdev->name, err);
2615 goto out;
2616 }
2617
2618 err = vmxnet3_activate_dev(adapter);
2619 if (err) {
2620 printk(KERN_ERR "%s: failed to re-activate, error %d. "
2621 "Closing it\n", netdev->name, err);
2622 goto out;
2623 }
2624 }
2625
2626 out:
2627 clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2628 if (err)
2629 vmxnet3_force_close(adapter);
2630
2631 return err;
2632 }
2633
2634
2635 static void
2636 vmxnet3_declare_features(struct vmxnet3_adapter *adapter, bool dma64)
2637 {
2638 struct net_device *netdev = adapter->netdev;
2639
2640 netdev->hw_features = NETIF_F_SG | NETIF_F_RXCSUM |
2641 NETIF_F_HW_CSUM | NETIF_F_HW_VLAN_TX |
2642 NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_LRO;
2643 if (dma64)
2644 netdev->features |= NETIF_F_HIGHDMA;
2645 netdev->vlan_features = netdev->hw_features & ~NETIF_F_HW_VLAN_TX;
2646 netdev->features = netdev->hw_features |
2647 NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_FILTER;
2648
2649 netdev_info(adapter->netdev,
2650 "features: sg csum vlan jf tso tsoIPv6 lro%s\n",
2651 dma64 ? " highDMA" : "");
2652 }
2653
2654
2655 static void
2656 vmxnet3_read_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac)
2657 {
2658 u32 tmp;
2659
2660 tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACL);
2661 *(u32 *)mac = tmp;
2662
2663 tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACH);
2664 mac[4] = tmp & 0xff;
2665 mac[5] = (tmp >> 8) & 0xff;
2666 }
2667
2668 #ifdef CONFIG_PCI_MSI
2669
2670 /*
2671 * Enable MSIx vectors.
2672 * Returns :
2673 * 0 on successful enabling of required vectors,
2674 * VMXNET3_LINUX_MIN_MSIX_VECT when only minimum number of vectors required
2675 * could be enabled.
2676 * number of vectors which can be enabled otherwise (this number is smaller
2677 * than VMXNET3_LINUX_MIN_MSIX_VECT)
2678 */
2679
2680 static int
2681 vmxnet3_acquire_msix_vectors(struct vmxnet3_adapter *adapter,
2682 int vectors)
2683 {
2684 int err = 0, vector_threshold;
2685 vector_threshold = VMXNET3_LINUX_MIN_MSIX_VECT;
2686
2687 while (vectors >= vector_threshold) {
2688 err = pci_enable_msix(adapter->pdev, adapter->intr.msix_entries,
2689 vectors);
2690 if (!err) {
2691 adapter->intr.num_intrs = vectors;
2692 return 0;
2693 } else if (err < 0) {
2694 printk(KERN_ERR "Failed to enable MSI-X for %s, error"
2695 " %d\n", adapter->netdev->name, err);
2696 vectors = 0;
2697 } else if (err < vector_threshold) {
2698 break;
2699 } else {
2700 /* If fails to enable required number of MSI-x vectors
2701 * try enabling minimum number of vectors required.
2702 */
2703 vectors = vector_threshold;
2704 printk(KERN_ERR "Failed to enable %d MSI-X for %s, try"
2705 " %d instead\n", vectors, adapter->netdev->name,
2706 vector_threshold);
2707 }
2708 }
2709
2710 printk(KERN_INFO "Number of MSI-X interrupts which can be allocatedi"
2711 " are lower than min threshold required.\n");
2712 return err;
2713 }
2714
2715
2716 #endif /* CONFIG_PCI_MSI */
2717
2718 static void
2719 vmxnet3_alloc_intr_resources(struct vmxnet3_adapter *adapter)
2720 {
2721 u32 cfg;
2722 unsigned long flags;
2723
2724 /* intr settings */
2725 spin_lock_irqsave(&adapter->cmd_lock, flags);
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 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2730 adapter->intr.type = cfg & 0x3;
2731 adapter->intr.mask_mode = (cfg >> 2) & 0x3;
2732
2733 if (adapter->intr.type == VMXNET3_IT_AUTO) {
2734 adapter->intr.type = VMXNET3_IT_MSIX;
2735 }
2736
2737 #ifdef CONFIG_PCI_MSI
2738 if (adapter->intr.type == VMXNET3_IT_MSIX) {
2739 int vector, err = 0;
2740
2741 adapter->intr.num_intrs = (adapter->share_intr ==
2742 VMXNET3_INTR_TXSHARE) ? 1 :
2743 adapter->num_tx_queues;
2744 adapter->intr.num_intrs += (adapter->share_intr ==
2745 VMXNET3_INTR_BUDDYSHARE) ? 0 :
2746 adapter->num_rx_queues;
2747 adapter->intr.num_intrs += 1; /* for link event */
2748
2749 adapter->intr.num_intrs = (adapter->intr.num_intrs >
2750 VMXNET3_LINUX_MIN_MSIX_VECT
2751 ? adapter->intr.num_intrs :
2752 VMXNET3_LINUX_MIN_MSIX_VECT);
2753
2754 for (vector = 0; vector < adapter->intr.num_intrs; vector++)
2755 adapter->intr.msix_entries[vector].entry = vector;
2756
2757 err = vmxnet3_acquire_msix_vectors(adapter,
2758 adapter->intr.num_intrs);
2759 /* If we cannot allocate one MSIx vector per queue
2760 * then limit the number of rx queues to 1
2761 */
2762 if (err == VMXNET3_LINUX_MIN_MSIX_VECT) {
2763 if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE
2764 || adapter->num_rx_queues != 1) {
2765 adapter->share_intr = VMXNET3_INTR_TXSHARE;
2766 printk(KERN_ERR "Number of rx queues : 1\n");
2767 adapter->num_rx_queues = 1;
2768 adapter->intr.num_intrs =
2769 VMXNET3_LINUX_MIN_MSIX_VECT;
2770 }
2771 return;
2772 }
2773 if (!err)
2774 return;
2775
2776 /* If we cannot allocate MSIx vectors use only one rx queue */
2777 printk(KERN_INFO "Failed to enable MSI-X for %s, error %d."
2778 "#rx queues : 1, try MSI\n", adapter->netdev->name, err);
2779
2780 adapter->intr.type = VMXNET3_IT_MSI;
2781 }
2782
2783 if (adapter->intr.type == VMXNET3_IT_MSI) {
2784 int err;
2785 err = pci_enable_msi(adapter->pdev);
2786 if (!err) {
2787 adapter->num_rx_queues = 1;
2788 adapter->intr.num_intrs = 1;
2789 return;
2790 }
2791 }
2792 #endif /* CONFIG_PCI_MSI */
2793
2794 adapter->num_rx_queues = 1;
2795 printk(KERN_INFO "Using INTx interrupt, #Rx queues: 1.\n");
2796 adapter->intr.type = VMXNET3_IT_INTX;
2797
2798 /* INT-X related setting */
2799 adapter->intr.num_intrs = 1;
2800 }
2801
2802
2803 static void
2804 vmxnet3_free_intr_resources(struct vmxnet3_adapter *adapter)
2805 {
2806 if (adapter->intr.type == VMXNET3_IT_MSIX)
2807 pci_disable_msix(adapter->pdev);
2808 else if (adapter->intr.type == VMXNET3_IT_MSI)
2809 pci_disable_msi(adapter->pdev);
2810 else
2811 BUG_ON(adapter->intr.type != VMXNET3_IT_INTX);
2812 }
2813
2814
2815 static void
2816 vmxnet3_tx_timeout(struct net_device *netdev)
2817 {
2818 struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2819 adapter->tx_timeout_count++;
2820
2821 printk(KERN_ERR "%s: tx hang\n", adapter->netdev->name);
2822 schedule_work(&adapter->work);
2823 netif_wake_queue(adapter->netdev);
2824 }
2825
2826
2827 static void
2828 vmxnet3_reset_work(struct work_struct *data)
2829 {
2830 struct vmxnet3_adapter *adapter;
2831
2832 adapter = container_of(data, struct vmxnet3_adapter, work);
2833
2834 /* if another thread is resetting the device, no need to proceed */
2835 if (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2836 return;
2837
2838 /* if the device is closed, we must leave it alone */
2839 rtnl_lock();
2840 if (netif_running(adapter->netdev)) {
2841 printk(KERN_INFO "%s: resetting\n", adapter->netdev->name);
2842 vmxnet3_quiesce_dev(adapter);
2843 vmxnet3_reset_dev(adapter);
2844 vmxnet3_activate_dev(adapter);
2845 } else {
2846 printk(KERN_INFO "%s: already closed\n", adapter->netdev->name);
2847 }
2848 rtnl_unlock();
2849
2850 clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2851 }
2852
2853
2854 static int __devinit
2855 vmxnet3_probe_device(struct pci_dev *pdev,
2856 const struct pci_device_id *id)
2857 {
2858 static const struct net_device_ops vmxnet3_netdev_ops = {
2859 .ndo_open = vmxnet3_open,
2860 .ndo_stop = vmxnet3_close,
2861 .ndo_start_xmit = vmxnet3_xmit_frame,
2862 .ndo_set_mac_address = vmxnet3_set_mac_addr,
2863 .ndo_change_mtu = vmxnet3_change_mtu,
2864 .ndo_set_features = vmxnet3_set_features,
2865 .ndo_get_stats64 = vmxnet3_get_stats64,
2866 .ndo_tx_timeout = vmxnet3_tx_timeout,
2867 .ndo_set_multicast_list = vmxnet3_set_mc,
2868 .ndo_vlan_rx_register = vmxnet3_vlan_rx_register,
2869 .ndo_vlan_rx_add_vid = vmxnet3_vlan_rx_add_vid,
2870 .ndo_vlan_rx_kill_vid = vmxnet3_vlan_rx_kill_vid,
2871 #ifdef CONFIG_NET_POLL_CONTROLLER
2872 .ndo_poll_controller = vmxnet3_netpoll,
2873 #endif
2874 };
2875 int err;
2876 bool dma64 = false; /* stupid gcc */
2877 u32 ver;
2878 struct net_device *netdev;
2879 struct vmxnet3_adapter *adapter;
2880 u8 mac[ETH_ALEN];
2881 int size;
2882 int num_tx_queues;
2883 int num_rx_queues;
2884
2885 if (!pci_msi_enabled())
2886 enable_mq = 0;
2887
2888 #ifdef VMXNET3_RSS
2889 if (enable_mq)
2890 num_rx_queues = min(VMXNET3_DEVICE_MAX_RX_QUEUES,
2891 (int)num_online_cpus());
2892 else
2893 #endif
2894 num_rx_queues = 1;
2895
2896 if (enable_mq)
2897 num_tx_queues = min(VMXNET3_DEVICE_MAX_TX_QUEUES,
2898 (int)num_online_cpus());
2899 else
2900 num_tx_queues = 1;
2901
2902 netdev = alloc_etherdev_mq(sizeof(struct vmxnet3_adapter),
2903 max(num_tx_queues, num_rx_queues));
2904 printk(KERN_INFO "# of Tx queues : %d, # of Rx queues : %d\n",
2905 num_tx_queues, num_rx_queues);
2906
2907 if (!netdev) {
2908 printk(KERN_ERR "Failed to alloc ethernet device for adapter "
2909 "%s\n", pci_name(pdev));
2910 return -ENOMEM;
2911 }
2912
2913 pci_set_drvdata(pdev, netdev);
2914 adapter = netdev_priv(netdev);
2915 adapter->netdev = netdev;
2916 adapter->pdev = pdev;
2917
2918 spin_lock_init(&adapter->cmd_lock);
2919 adapter->shared = pci_alloc_consistent(adapter->pdev,
2920 sizeof(struct Vmxnet3_DriverShared),
2921 &adapter->shared_pa);
2922 if (!adapter->shared) {
2923 printk(KERN_ERR "Failed to allocate memory for %s\n",
2924 pci_name(pdev));
2925 err = -ENOMEM;
2926 goto err_alloc_shared;
2927 }
2928
2929 adapter->num_rx_queues = num_rx_queues;
2930 adapter->num_tx_queues = num_tx_queues;
2931
2932 size = sizeof(struct Vmxnet3_TxQueueDesc) * adapter->num_tx_queues;
2933 size += sizeof(struct Vmxnet3_RxQueueDesc) * adapter->num_rx_queues;
2934 adapter->tqd_start = pci_alloc_consistent(adapter->pdev, size,
2935 &adapter->queue_desc_pa);
2936
2937 if (!adapter->tqd_start) {
2938 printk(KERN_ERR "Failed to allocate memory for %s\n",
2939 pci_name(pdev));
2940 err = -ENOMEM;
2941 goto err_alloc_queue_desc;
2942 }
2943 adapter->rqd_start = (struct Vmxnet3_RxQueueDesc *)(adapter->tqd_start +
2944 adapter->num_tx_queues);
2945
2946 adapter->pm_conf = kmalloc(sizeof(struct Vmxnet3_PMConf), GFP_KERNEL);
2947 if (adapter->pm_conf == NULL) {
2948 printk(KERN_ERR "Failed to allocate memory for %s\n",
2949 pci_name(pdev));
2950 err = -ENOMEM;
2951 goto err_alloc_pm;
2952 }
2953
2954 #ifdef VMXNET3_RSS
2955
2956 adapter->rss_conf = kmalloc(sizeof(struct UPT1_RSSConf), GFP_KERNEL);
2957 if (adapter->rss_conf == NULL) {
2958 printk(KERN_ERR "Failed to allocate memory for %s\n",
2959 pci_name(pdev));
2960 err = -ENOMEM;
2961 goto err_alloc_rss;
2962 }
2963 #endif /* VMXNET3_RSS */
2964
2965 err = vmxnet3_alloc_pci_resources(adapter, &dma64);
2966 if (err < 0)
2967 goto err_alloc_pci;
2968
2969 ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_VRRS);
2970 if (ver & 1) {
2971 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_VRRS, 1);
2972 } else {
2973 printk(KERN_ERR "Incompatible h/w version (0x%x) for adapter"
2974 " %s\n", ver, pci_name(pdev));
2975 err = -EBUSY;
2976 goto err_ver;
2977 }
2978
2979 ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_UVRS);
2980 if (ver & 1) {
2981 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_UVRS, 1);
2982 } else {
2983 printk(KERN_ERR "Incompatible upt version (0x%x) for "
2984 "adapter %s\n", ver, pci_name(pdev));
2985 err = -EBUSY;
2986 goto err_ver;
2987 }
2988
2989 vmxnet3_declare_features(adapter, dma64);
2990
2991 adapter->dev_number = atomic_read(&devices_found);
2992
2993 adapter->share_intr = irq_share_mode;
2994 if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE &&
2995 adapter->num_tx_queues != adapter->num_rx_queues)
2996 adapter->share_intr = VMXNET3_INTR_DONTSHARE;
2997
2998 vmxnet3_alloc_intr_resources(adapter);
2999
3000 #ifdef VMXNET3_RSS
3001 if (adapter->num_rx_queues > 1 &&
3002 adapter->intr.type == VMXNET3_IT_MSIX) {
3003 adapter->rss = true;
3004 printk(KERN_INFO "RSS is enabled.\n");
3005 } else {
3006 adapter->rss = false;
3007 }
3008 #endif
3009
3010 vmxnet3_read_mac_addr(adapter, mac);
3011 memcpy(netdev->dev_addr, mac, netdev->addr_len);
3012
3013 netdev->netdev_ops = &vmxnet3_netdev_ops;
3014 vmxnet3_set_ethtool_ops(netdev);
3015 netdev->watchdog_timeo = 5 * HZ;
3016
3017 INIT_WORK(&adapter->work, vmxnet3_reset_work);
3018
3019 if (adapter->intr.type == VMXNET3_IT_MSIX) {
3020 int i;
3021 for (i = 0; i < adapter->num_rx_queues; i++) {
3022 netif_napi_add(adapter->netdev,
3023 &adapter->rx_queue[i].napi,
3024 vmxnet3_poll_rx_only, 64);
3025 }
3026 } else {
3027 netif_napi_add(adapter->netdev, &adapter->rx_queue[0].napi,
3028 vmxnet3_poll, 64);
3029 }
3030
3031 netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
3032 netif_set_real_num_rx_queues(adapter->netdev, adapter->num_rx_queues);
3033
3034 SET_NETDEV_DEV(netdev, &pdev->dev);
3035 err = register_netdev(netdev);
3036
3037 if (err) {
3038 printk(KERN_ERR "Failed to register adapter %s\n",
3039 pci_name(pdev));
3040 goto err_register;
3041 }
3042
3043 set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
3044 vmxnet3_check_link(adapter, false);
3045 atomic_inc(&devices_found);
3046 return 0;
3047
3048 err_register:
3049 vmxnet3_free_intr_resources(adapter);
3050 err_ver:
3051 vmxnet3_free_pci_resources(adapter);
3052 err_alloc_pci:
3053 #ifdef VMXNET3_RSS
3054 kfree(adapter->rss_conf);
3055 err_alloc_rss:
3056 #endif
3057 kfree(adapter->pm_conf);
3058 err_alloc_pm:
3059 pci_free_consistent(adapter->pdev, size, adapter->tqd_start,
3060 adapter->queue_desc_pa);
3061 err_alloc_queue_desc:
3062 pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_DriverShared),
3063 adapter->shared, adapter->shared_pa);
3064 err_alloc_shared:
3065 pci_set_drvdata(pdev, NULL);
3066 free_netdev(netdev);
3067 return err;
3068 }
3069
3070
3071 static void __devexit
3072 vmxnet3_remove_device(struct pci_dev *pdev)
3073 {
3074 struct net_device *netdev = pci_get_drvdata(pdev);
3075 struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3076 int size = 0;
3077 int num_rx_queues;
3078
3079 #ifdef VMXNET3_RSS
3080 if (enable_mq)
3081 num_rx_queues = min(VMXNET3_DEVICE_MAX_RX_QUEUES,
3082 (int)num_online_cpus());
3083 else
3084 #endif
3085 num_rx_queues = 1;
3086
3087 cancel_work_sync(&adapter->work);
3088
3089 unregister_netdev(netdev);
3090
3091 vmxnet3_free_intr_resources(adapter);
3092 vmxnet3_free_pci_resources(adapter);
3093 #ifdef VMXNET3_RSS
3094 kfree(adapter->rss_conf);
3095 #endif
3096 kfree(adapter->pm_conf);
3097
3098 size = sizeof(struct Vmxnet3_TxQueueDesc) * adapter->num_tx_queues;
3099 size += sizeof(struct Vmxnet3_RxQueueDesc) * num_rx_queues;
3100 pci_free_consistent(adapter->pdev, size, adapter->tqd_start,
3101 adapter->queue_desc_pa);
3102 pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_DriverShared),
3103 adapter->shared, adapter->shared_pa);
3104 free_netdev(netdev);
3105 }
3106
3107
3108 #ifdef CONFIG_PM
3109
3110 static int
3111 vmxnet3_suspend(struct device *device)
3112 {
3113 struct pci_dev *pdev = to_pci_dev(device);
3114 struct net_device *netdev = pci_get_drvdata(pdev);
3115 struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3116 struct Vmxnet3_PMConf *pmConf;
3117 struct ethhdr *ehdr;
3118 struct arphdr *ahdr;
3119 u8 *arpreq;
3120 struct in_device *in_dev;
3121 struct in_ifaddr *ifa;
3122 unsigned long flags;
3123 int i = 0;
3124
3125 if (!netif_running(netdev))
3126 return 0;
3127
3128 for (i = 0; i < adapter->num_rx_queues; i++)
3129 napi_disable(&adapter->rx_queue[i].napi);
3130
3131 vmxnet3_disable_all_intrs(adapter);
3132 vmxnet3_free_irqs(adapter);
3133 vmxnet3_free_intr_resources(adapter);
3134
3135 netif_device_detach(netdev);
3136 netif_tx_stop_all_queues(netdev);
3137
3138 /* Create wake-up filters. */
3139 pmConf = adapter->pm_conf;
3140 memset(pmConf, 0, sizeof(*pmConf));
3141
3142 if (adapter->wol & WAKE_UCAST) {
3143 pmConf->filters[i].patternSize = ETH_ALEN;
3144 pmConf->filters[i].maskSize = 1;
3145 memcpy(pmConf->filters[i].pattern, netdev->dev_addr, ETH_ALEN);
3146 pmConf->filters[i].mask[0] = 0x3F; /* LSB ETH_ALEN bits */
3147
3148 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_FILTER;
3149 i++;
3150 }
3151
3152 if (adapter->wol & WAKE_ARP) {
3153 in_dev = in_dev_get(netdev);
3154 if (!in_dev)
3155 goto skip_arp;
3156
3157 ifa = (struct in_ifaddr *)in_dev->ifa_list;
3158 if (!ifa)
3159 goto skip_arp;
3160
3161 pmConf->filters[i].patternSize = ETH_HLEN + /* Ethernet header*/
3162 sizeof(struct arphdr) + /* ARP header */
3163 2 * ETH_ALEN + /* 2 Ethernet addresses*/
3164 2 * sizeof(u32); /*2 IPv4 addresses */
3165 pmConf->filters[i].maskSize =
3166 (pmConf->filters[i].patternSize - 1) / 8 + 1;
3167
3168 /* ETH_P_ARP in Ethernet header. */
3169 ehdr = (struct ethhdr *)pmConf->filters[i].pattern;
3170 ehdr->h_proto = htons(ETH_P_ARP);
3171
3172 /* ARPOP_REQUEST in ARP header. */
3173 ahdr = (struct arphdr *)&pmConf->filters[i].pattern[ETH_HLEN];
3174 ahdr->ar_op = htons(ARPOP_REQUEST);
3175 arpreq = (u8 *)(ahdr + 1);
3176
3177 /* The Unicast IPv4 address in 'tip' field. */
3178 arpreq += 2 * ETH_ALEN + sizeof(u32);
3179 *(u32 *)arpreq = ifa->ifa_address;
3180
3181 /* The mask for the relevant bits. */
3182 pmConf->filters[i].mask[0] = 0x00;
3183 pmConf->filters[i].mask[1] = 0x30; /* ETH_P_ARP */
3184 pmConf->filters[i].mask[2] = 0x30; /* ARPOP_REQUEST */
3185 pmConf->filters[i].mask[3] = 0x00;
3186 pmConf->filters[i].mask[4] = 0xC0; /* IPv4 TIP */
3187 pmConf->filters[i].mask[5] = 0x03; /* IPv4 TIP */
3188 in_dev_put(in_dev);
3189
3190 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_FILTER;
3191 i++;
3192 }
3193
3194 skip_arp:
3195 if (adapter->wol & WAKE_MAGIC)
3196 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_MAGIC;
3197
3198 pmConf->numFilters = i;
3199
3200 adapter->shared->devRead.pmConfDesc.confVer = cpu_to_le32(1);
3201 adapter->shared->devRead.pmConfDesc.confLen = cpu_to_le32(sizeof(
3202 *pmConf));
3203 adapter->shared->devRead.pmConfDesc.confPA = cpu_to_le64(virt_to_phys(
3204 pmConf));
3205
3206 spin_lock_irqsave(&adapter->cmd_lock, flags);
3207 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3208 VMXNET3_CMD_UPDATE_PMCFG);
3209 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3210
3211 pci_save_state(pdev);
3212 pci_enable_wake(pdev, pci_choose_state(pdev, PMSG_SUSPEND),
3213 adapter->wol);
3214 pci_disable_device(pdev);
3215 pci_set_power_state(pdev, pci_choose_state(pdev, PMSG_SUSPEND));
3216
3217 return 0;
3218 }
3219
3220
3221 static int
3222 vmxnet3_resume(struct device *device)
3223 {
3224 int err, i = 0;
3225 unsigned long flags;
3226 struct pci_dev *pdev = to_pci_dev(device);
3227 struct net_device *netdev = pci_get_drvdata(pdev);
3228 struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3229 struct Vmxnet3_PMConf *pmConf;
3230
3231 if (!netif_running(netdev))
3232 return 0;
3233
3234 /* Destroy wake-up filters. */
3235 pmConf = adapter->pm_conf;
3236 memset(pmConf, 0, sizeof(*pmConf));
3237
3238 adapter->shared->devRead.pmConfDesc.confVer = cpu_to_le32(1);
3239 adapter->shared->devRead.pmConfDesc.confLen = cpu_to_le32(sizeof(
3240 *pmConf));
3241 adapter->shared->devRead.pmConfDesc.confPA = cpu_to_le64(virt_to_phys(
3242 pmConf));
3243
3244 netif_device_attach(netdev);
3245 pci_set_power_state(pdev, PCI_D0);
3246 pci_restore_state(pdev);
3247 err = pci_enable_device_mem(pdev);
3248 if (err != 0)
3249 return err;
3250
3251 pci_enable_wake(pdev, PCI_D0, 0);
3252
3253 spin_lock_irqsave(&adapter->cmd_lock, flags);
3254 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3255 VMXNET3_CMD_UPDATE_PMCFG);
3256 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3257 vmxnet3_alloc_intr_resources(adapter);
3258 vmxnet3_request_irqs(adapter);
3259 for (i = 0; i < adapter->num_rx_queues; i++)
3260 napi_enable(&adapter->rx_queue[i].napi);
3261 vmxnet3_enable_all_intrs(adapter);
3262
3263 return 0;
3264 }
3265
3266 static const struct dev_pm_ops vmxnet3_pm_ops = {
3267 .suspend = vmxnet3_suspend,
3268 .resume = vmxnet3_resume,
3269 };
3270 #endif
3271
3272 static struct pci_driver vmxnet3_driver = {
3273 .name = vmxnet3_driver_name,
3274 .id_table = vmxnet3_pciid_table,
3275 .probe = vmxnet3_probe_device,
3276 .remove = __devexit_p(vmxnet3_remove_device),
3277 #ifdef CONFIG_PM
3278 .driver.pm = &vmxnet3_pm_ops,
3279 #endif
3280 };
3281
3282
3283 static int __init
3284 vmxnet3_init_module(void)
3285 {
3286 printk(KERN_INFO "%s - version %s\n", VMXNET3_DRIVER_DESC,
3287 VMXNET3_DRIVER_VERSION_REPORT);
3288 return pci_register_driver(&vmxnet3_driver);
3289 }
3290
3291 module_init(vmxnet3_init_module);
3292
3293
3294 static void
3295 vmxnet3_exit_module(void)
3296 {
3297 pci_unregister_driver(&vmxnet3_driver);
3298 }
3299
3300 module_exit(vmxnet3_exit_module);
3301
3302 MODULE_AUTHOR("VMware, Inc.");
3303 MODULE_DESCRIPTION(VMXNET3_DRIVER_DESC);
3304 MODULE_LICENSE("GPL v2");
3305 MODULE_VERSION(VMXNET3_DRIVER_VERSION_STRING);
Linux 2.6 libata-dev (mirror)