?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
mm/hmm.c: remove superfluous RCU protection around radix tree lookup
[linux/fpc-iii.git] / drivers / net / vmxnet3 / vmxnet3_drv.c
blobe04937f44f33313eb3eebf3bc1ea598686965950
1 /*
2 * Linux driver for VMware's vmxnet3 ethernet NIC.
3 *
4 * Copyright (C) 2008-2016, 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: pv-drivers@vmware.com
24 *
25 */
26
27 #include <linux/module.h>
28 #include <net/ip6_checksum.h>
29
30 #include "vmxnet3_int.h"
31
32 char vmxnet3_driver_name[] = "vmxnet3";
33 #define VMXNET3_DRIVER_DESC "VMware vmxnet3 virtual NIC driver"
34
35 /*
36 * PCI Device ID Table
37 * Last entry must be all 0s
38 */
39 static const struct pci_device_id vmxnet3_pciid_table[] = {
40 {PCI_VDEVICE(VMWARE, PCI_DEVICE_ID_VMWARE_VMXNET3)},
41 {0}
42 };
43
44 MODULE_DEVICE_TABLE(pci, vmxnet3_pciid_table);
45
46 static int enable_mq = 1;
47
48 static void
49 vmxnet3_write_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac);
50
51 /*
52 * Enable/Disable the given intr
53 */
54 static void
55 vmxnet3_enable_intr(struct vmxnet3_adapter *adapter, unsigned intr_idx)
56 {
57 VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_IMR + intr_idx * 8, 0);
58 }
59
60
61 static void
62 vmxnet3_disable_intr(struct vmxnet3_adapter *adapter, unsigned intr_idx)
63 {
64 VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_IMR + intr_idx * 8, 1);
65 }
66
67
68 /*
69 * Enable/Disable all intrs used by the device
70 */
71 static void
72 vmxnet3_enable_all_intrs(struct vmxnet3_adapter *adapter)
73 {
74 int i;
75
76 for (i = 0; i < adapter->intr.num_intrs; i++)
77 vmxnet3_enable_intr(adapter, i);
78 adapter->shared->devRead.intrConf.intrCtrl &=
79 cpu_to_le32(~VMXNET3_IC_DISABLE_ALL);
80 }
81
82
83 static void
84 vmxnet3_disable_all_intrs(struct vmxnet3_adapter *adapter)
85 {
86 int i;
87
88 adapter->shared->devRead.intrConf.intrCtrl |=
89 cpu_to_le32(VMXNET3_IC_DISABLE_ALL);
90 for (i = 0; i < adapter->intr.num_intrs; i++)
91 vmxnet3_disable_intr(adapter, i);
92 }
93
94
95 static void
96 vmxnet3_ack_events(struct vmxnet3_adapter *adapter, u32 events)
97 {
98 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_ECR, events);
99 }
100
101
102 static bool
103 vmxnet3_tq_stopped(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
104 {
105 return tq->stopped;
106 }
107
108
109 static void
110 vmxnet3_tq_start(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
111 {
112 tq->stopped = false;
113 netif_start_subqueue(adapter->netdev, tq - adapter->tx_queue);
114 }
115
116
117 static void
118 vmxnet3_tq_wake(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
119 {
120 tq->stopped = false;
121 netif_wake_subqueue(adapter->netdev, (tq - adapter->tx_queue));
122 }
123
124
125 static void
126 vmxnet3_tq_stop(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
127 {
128 tq->stopped = true;
129 tq->num_stop++;
130 netif_stop_subqueue(adapter->netdev, (tq - adapter->tx_queue));
131 }
132
133
134 /*
135 * Check the link state. This may start or stop the tx queue.
136 */
137 static void
138 vmxnet3_check_link(struct vmxnet3_adapter *adapter, bool affectTxQueue)
139 {
140 u32 ret;
141 int i;
142 unsigned long flags;
143
144 spin_lock_irqsave(&adapter->cmd_lock, flags);
145 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, VMXNET3_CMD_GET_LINK);
146 ret = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
147 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
148
149 adapter->link_speed = ret >> 16;
150 if (ret & 1) { /* Link is up. */
151 netdev_info(adapter->netdev, "NIC Link is Up %d Mbps\n",
152 adapter->link_speed);
153 netif_carrier_on(adapter->netdev);
154
155 if (affectTxQueue) {
156 for (i = 0; i < adapter->num_tx_queues; i++)
157 vmxnet3_tq_start(&adapter->tx_queue[i],
158 adapter);
159 }
160 } else {
161 netdev_info(adapter->netdev, "NIC Link is Down\n");
162 netif_carrier_off(adapter->netdev);
163
164 if (affectTxQueue) {
165 for (i = 0; i < adapter->num_tx_queues; i++)
166 vmxnet3_tq_stop(&adapter->tx_queue[i], adapter);
167 }
168 }
169 }
170
171 static void
172 vmxnet3_process_events(struct vmxnet3_adapter *adapter)
173 {
174 int i;
175 unsigned long flags;
176 u32 events = le32_to_cpu(adapter->shared->ecr);
177 if (!events)
178 return;
179
180 vmxnet3_ack_events(adapter, events);
181
182 /* Check if link state has changed */
183 if (events & VMXNET3_ECR_LINK)
184 vmxnet3_check_link(adapter, true);
185
186 /* Check if there is an error on xmit/recv queues */
187 if (events & (VMXNET3_ECR_TQERR | VMXNET3_ECR_RQERR)) {
188 spin_lock_irqsave(&adapter->cmd_lock, flags);
189 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
190 VMXNET3_CMD_GET_QUEUE_STATUS);
191 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
192
193 for (i = 0; i < adapter->num_tx_queues; i++)
194 if (adapter->tqd_start[i].status.stopped)
195 dev_err(&adapter->netdev->dev,
196 "%s: tq[%d] error 0x%x\n",
197 adapter->netdev->name, i, le32_to_cpu(
198 adapter->tqd_start[i].status.error));
199 for (i = 0; i < adapter->num_rx_queues; i++)
200 if (adapter->rqd_start[i].status.stopped)
201 dev_err(&adapter->netdev->dev,
202 "%s: rq[%d] error 0x%x\n",
203 adapter->netdev->name, i,
204 adapter->rqd_start[i].status.error);
205
206 schedule_work(&adapter->work);
207 }
208 }
209
210 #ifdef __BIG_ENDIAN_BITFIELD
211 /*
212 * The device expects the bitfields in shared structures to be written in
213 * little endian. When CPU is big endian, the following routines are used to
214 * correctly read and write into ABI.
215 * The general technique used here is : double word bitfields are defined in
216 * opposite order for big endian architecture. Then before reading them in
217 * driver the complete double word is translated using le32_to_cpu. Similarly
218 * After the driver writes into bitfields, cpu_to_le32 is used to translate the
219 * double words into required format.
220 * In order to avoid touching bits in shared structure more than once, temporary
221 * descriptors are used. These are passed as srcDesc to following functions.
222 */
223 static void vmxnet3_RxDescToCPU(const struct Vmxnet3_RxDesc *srcDesc,
224 struct Vmxnet3_RxDesc *dstDesc)
225 {
226 u32 *src = (u32 *)srcDesc + 2;
227 u32 *dst = (u32 *)dstDesc + 2;
228 dstDesc->addr = le64_to_cpu(srcDesc->addr);
229 *dst = le32_to_cpu(*src);
230 dstDesc->ext1 = le32_to_cpu(srcDesc->ext1);
231 }
232
233 static void vmxnet3_TxDescToLe(const struct Vmxnet3_TxDesc *srcDesc,
234 struct Vmxnet3_TxDesc *dstDesc)
235 {
236 int i;
237 u32 *src = (u32 *)(srcDesc + 1);
238 u32 *dst = (u32 *)(dstDesc + 1);
239
240 /* Working backwards so that the gen bit is set at the end. */
241 for (i = 2; i > 0; i--) {
242 src--;
243 dst--;
244 *dst = cpu_to_le32(*src);
245 }
246 }
247
248
249 static void vmxnet3_RxCompToCPU(const struct Vmxnet3_RxCompDesc *srcDesc,
250 struct Vmxnet3_RxCompDesc *dstDesc)
251 {
252 int i = 0;
253 u32 *src = (u32 *)srcDesc;
254 u32 *dst = (u32 *)dstDesc;
255 for (i = 0; i < sizeof(struct Vmxnet3_RxCompDesc) / sizeof(u32); i++) {
256 *dst = le32_to_cpu(*src);
257 src++;
258 dst++;
259 }
260 }
261
262
263 /* Used to read bitfield values from double words. */
264 static u32 get_bitfield32(const __le32 *bitfield, u32 pos, u32 size)
265 {
266 u32 temp = le32_to_cpu(*bitfield);
267 u32 mask = ((1 << size) - 1) << pos;
268 temp &= mask;
269 temp >>= pos;
270 return temp;
271 }
272
273
274
275 #endif /* __BIG_ENDIAN_BITFIELD */
276
277 #ifdef __BIG_ENDIAN_BITFIELD
278
279 # define VMXNET3_TXDESC_GET_GEN(txdesc) get_bitfield32(((const __le32 *) \
280 txdesc) + VMXNET3_TXD_GEN_DWORD_SHIFT, \
281 VMXNET3_TXD_GEN_SHIFT, VMXNET3_TXD_GEN_SIZE)
282 # define VMXNET3_TXDESC_GET_EOP(txdesc) get_bitfield32(((const __le32 *) \
283 txdesc) + VMXNET3_TXD_EOP_DWORD_SHIFT, \
284 VMXNET3_TXD_EOP_SHIFT, VMXNET3_TXD_EOP_SIZE)
285 # define VMXNET3_TCD_GET_GEN(tcd) get_bitfield32(((const __le32 *)tcd) + \
286 VMXNET3_TCD_GEN_DWORD_SHIFT, VMXNET3_TCD_GEN_SHIFT, \
287 VMXNET3_TCD_GEN_SIZE)
288 # define VMXNET3_TCD_GET_TXIDX(tcd) get_bitfield32((const __le32 *)tcd, \
289 VMXNET3_TCD_TXIDX_SHIFT, VMXNET3_TCD_TXIDX_SIZE)
290 # define vmxnet3_getRxComp(dstrcd, rcd, tmp) do { \
291 (dstrcd) = (tmp); \
292 vmxnet3_RxCompToCPU((rcd), (tmp)); \
293 } while (0)
294 # define vmxnet3_getRxDesc(dstrxd, rxd, tmp) do { \
295 (dstrxd) = (tmp); \
296 vmxnet3_RxDescToCPU((rxd), (tmp)); \
297 } while (0)
298
299 #else
300
301 # define VMXNET3_TXDESC_GET_GEN(txdesc) ((txdesc)->gen)
302 # define VMXNET3_TXDESC_GET_EOP(txdesc) ((txdesc)->eop)
303 # define VMXNET3_TCD_GET_GEN(tcd) ((tcd)->gen)
304 # define VMXNET3_TCD_GET_TXIDX(tcd) ((tcd)->txdIdx)
305 # define vmxnet3_getRxComp(dstrcd, rcd, tmp) (dstrcd) = (rcd)
306 # define vmxnet3_getRxDesc(dstrxd, rxd, tmp) (dstrxd) = (rxd)
307
308 #endif /* __BIG_ENDIAN_BITFIELD */
309
310
311 static void
312 vmxnet3_unmap_tx_buf(struct vmxnet3_tx_buf_info *tbi,
313 struct pci_dev *pdev)
314 {
315 if (tbi->map_type == VMXNET3_MAP_SINGLE)
316 dma_unmap_single(&pdev->dev, tbi->dma_addr, tbi->len,
317 PCI_DMA_TODEVICE);
318 else if (tbi->map_type == VMXNET3_MAP_PAGE)
319 dma_unmap_page(&pdev->dev, tbi->dma_addr, tbi->len,
320 PCI_DMA_TODEVICE);
321 else
322 BUG_ON(tbi->map_type != VMXNET3_MAP_NONE);
323
324 tbi->map_type = VMXNET3_MAP_NONE; /* to help debugging */
325 }
326
327
328 static int
329 vmxnet3_unmap_pkt(u32 eop_idx, struct vmxnet3_tx_queue *tq,
330 struct pci_dev *pdev, struct vmxnet3_adapter *adapter)
331 {
332 struct sk_buff *skb;
333 int entries = 0;
334
335 /* no out of order completion */
336 BUG_ON(tq->buf_info[eop_idx].sop_idx != tq->tx_ring.next2comp);
337 BUG_ON(VMXNET3_TXDESC_GET_EOP(&(tq->tx_ring.base[eop_idx].txd)) != 1);
338
339 skb = tq->buf_info[eop_idx].skb;
340 BUG_ON(skb == NULL);
341 tq->buf_info[eop_idx].skb = NULL;
342
343 VMXNET3_INC_RING_IDX_ONLY(eop_idx, tq->tx_ring.size);
344
345 while (tq->tx_ring.next2comp != eop_idx) {
346 vmxnet3_unmap_tx_buf(tq->buf_info + tq->tx_ring.next2comp,
347 pdev);
348
349 /* update next2comp w/o tx_lock. Since we are marking more,
350 * instead of less, tx ring entries avail, the worst case is
351 * that the tx routine incorrectly re-queues a pkt due to
352 * insufficient tx ring entries.
353 */
354 vmxnet3_cmd_ring_adv_next2comp(&tq->tx_ring);
355 entries++;
356 }
357
358 dev_kfree_skb_any(skb);
359 return entries;
360 }
361
362
363 static int
364 vmxnet3_tq_tx_complete(struct vmxnet3_tx_queue *tq,
365 struct vmxnet3_adapter *adapter)
366 {
367 int completed = 0;
368 union Vmxnet3_GenericDesc *gdesc;
369
370 gdesc = tq->comp_ring.base + tq->comp_ring.next2proc;
371 while (VMXNET3_TCD_GET_GEN(&gdesc->tcd) == tq->comp_ring.gen) {
372 completed += vmxnet3_unmap_pkt(VMXNET3_TCD_GET_TXIDX(
373 &gdesc->tcd), tq, adapter->pdev,
374 adapter);
375
376 vmxnet3_comp_ring_adv_next2proc(&tq->comp_ring);
377 gdesc = tq->comp_ring.base + tq->comp_ring.next2proc;
378 }
379
380 if (completed) {
381 spin_lock(&tq->tx_lock);
382 if (unlikely(vmxnet3_tq_stopped(tq, adapter) &&
383 vmxnet3_cmd_ring_desc_avail(&tq->tx_ring) >
384 VMXNET3_WAKE_QUEUE_THRESHOLD(tq) &&
385 netif_carrier_ok(adapter->netdev))) {
386 vmxnet3_tq_wake(tq, adapter);
387 }
388 spin_unlock(&tq->tx_lock);
389 }
390 return completed;
391 }
392
393
394 static void
395 vmxnet3_tq_cleanup(struct vmxnet3_tx_queue *tq,
396 struct vmxnet3_adapter *adapter)
397 {
398 int i;
399
400 while (tq->tx_ring.next2comp != tq->tx_ring.next2fill) {
401 struct vmxnet3_tx_buf_info *tbi;
402
403 tbi = tq->buf_info + tq->tx_ring.next2comp;
404
405 vmxnet3_unmap_tx_buf(tbi, adapter->pdev);
406 if (tbi->skb) {
407 dev_kfree_skb_any(tbi->skb);
408 tbi->skb = NULL;
409 }
410 vmxnet3_cmd_ring_adv_next2comp(&tq->tx_ring);
411 }
412
413 /* sanity check, verify all buffers are indeed unmapped and freed */
414 for (i = 0; i < tq->tx_ring.size; i++) {
415 BUG_ON(tq->buf_info[i].skb != NULL ||
416 tq->buf_info[i].map_type != VMXNET3_MAP_NONE);
417 }
418
419 tq->tx_ring.gen = VMXNET3_INIT_GEN;
420 tq->tx_ring.next2fill = tq->tx_ring.next2comp = 0;
421
422 tq->comp_ring.gen = VMXNET3_INIT_GEN;
423 tq->comp_ring.next2proc = 0;
424 }
425
426
427 static void
428 vmxnet3_tq_destroy(struct vmxnet3_tx_queue *tq,
429 struct vmxnet3_adapter *adapter)
430 {
431 if (tq->tx_ring.base) {
432 dma_free_coherent(&adapter->pdev->dev, tq->tx_ring.size *
433 sizeof(struct Vmxnet3_TxDesc),
434 tq->tx_ring.base, tq->tx_ring.basePA);
435 tq->tx_ring.base = NULL;
436 }
437 if (tq->data_ring.base) {
438 dma_free_coherent(&adapter->pdev->dev,
439 tq->data_ring.size * tq->txdata_desc_size,
440 tq->data_ring.base, tq->data_ring.basePA);
441 tq->data_ring.base = NULL;
442 }
443 if (tq->comp_ring.base) {
444 dma_free_coherent(&adapter->pdev->dev, tq->comp_ring.size *
445 sizeof(struct Vmxnet3_TxCompDesc),
446 tq->comp_ring.base, tq->comp_ring.basePA);
447 tq->comp_ring.base = NULL;
448 }
449 if (tq->buf_info) {
450 dma_free_coherent(&adapter->pdev->dev,
451 tq->tx_ring.size * sizeof(tq->buf_info[0]),
452 tq->buf_info, tq->buf_info_pa);
453 tq->buf_info = NULL;
454 }
455 }
456
457
458 /* Destroy all tx queues */
459 void
460 vmxnet3_tq_destroy_all(struct vmxnet3_adapter *adapter)
461 {
462 int i;
463
464 for (i = 0; i < adapter->num_tx_queues; i++)
465 vmxnet3_tq_destroy(&adapter->tx_queue[i], adapter);
466 }
467
468
469 static void
470 vmxnet3_tq_init(struct vmxnet3_tx_queue *tq,
471 struct vmxnet3_adapter *adapter)
472 {
473 int i;
474
475 /* reset the tx ring contents to 0 and reset the tx ring states */
476 memset(tq->tx_ring.base, 0, tq->tx_ring.size *
477 sizeof(struct Vmxnet3_TxDesc));
478 tq->tx_ring.next2fill = tq->tx_ring.next2comp = 0;
479 tq->tx_ring.gen = VMXNET3_INIT_GEN;
480
481 memset(tq->data_ring.base, 0,
482 tq->data_ring.size * tq->txdata_desc_size);
483
484 /* reset the tx comp ring contents to 0 and reset comp ring states */
485 memset(tq->comp_ring.base, 0, tq->comp_ring.size *
486 sizeof(struct Vmxnet3_TxCompDesc));
487 tq->comp_ring.next2proc = 0;
488 tq->comp_ring.gen = VMXNET3_INIT_GEN;
489
490 /* reset the bookkeeping data */
491 memset(tq->buf_info, 0, sizeof(tq->buf_info[0]) * tq->tx_ring.size);
492 for (i = 0; i < tq->tx_ring.size; i++)
493 tq->buf_info[i].map_type = VMXNET3_MAP_NONE;
494
495 /* stats are not reset */
496 }
497
498
499 static int
500 vmxnet3_tq_create(struct vmxnet3_tx_queue *tq,
501 struct vmxnet3_adapter *adapter)
502 {
503 size_t sz;
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 = dma_alloc_coherent(&adapter->pdev->dev,
509 tq->tx_ring.size * sizeof(struct Vmxnet3_TxDesc),
510 &tq->tx_ring.basePA, GFP_KERNEL);
511 if (!tq->tx_ring.base) {
512 netdev_err(adapter->netdev, "failed to allocate tx ring\n");
513 goto err;
514 }
515
516 tq->data_ring.base = dma_alloc_coherent(&adapter->pdev->dev,
517 tq->data_ring.size * tq->txdata_desc_size,
518 &tq->data_ring.basePA, GFP_KERNEL);
519 if (!tq->data_ring.base) {
520 netdev_err(adapter->netdev, "failed to allocate tx data ring\n");
521 goto err;
522 }
523
524 tq->comp_ring.base = dma_alloc_coherent(&adapter->pdev->dev,
525 tq->comp_ring.size * sizeof(struct Vmxnet3_TxCompDesc),
526 &tq->comp_ring.basePA, GFP_KERNEL);
527 if (!tq->comp_ring.base) {
528 netdev_err(adapter->netdev, "failed to allocate tx comp ring\n");
529 goto err;
530 }
531
532 sz = tq->tx_ring.size * sizeof(tq->buf_info[0]);
533 tq->buf_info = dma_zalloc_coherent(&adapter->pdev->dev, sz,
534 &tq->buf_info_pa, GFP_KERNEL);
535 if (!tq->buf_info)
536 goto err;
537
538 return 0;
539
540 err:
541 vmxnet3_tq_destroy(tq, adapter);
542 return -ENOMEM;
543 }
544
545 static void
546 vmxnet3_tq_cleanup_all(struct vmxnet3_adapter *adapter)
547 {
548 int i;
549
550 for (i = 0; i < adapter->num_tx_queues; i++)
551 vmxnet3_tq_cleanup(&adapter->tx_queue[i], adapter);
552 }
553
554 /*
555 * starting from ring->next2fill, allocate rx buffers for the given ring
556 * of the rx queue and update the rx desc. stop after @num_to_alloc buffers
557 * are allocated or allocation fails
558 */
559
560 static int
561 vmxnet3_rq_alloc_rx_buf(struct vmxnet3_rx_queue *rq, u32 ring_idx,
562 int num_to_alloc, struct vmxnet3_adapter *adapter)
563 {
564 int num_allocated = 0;
565 struct vmxnet3_rx_buf_info *rbi_base = rq->buf_info[ring_idx];
566 struct vmxnet3_cmd_ring *ring = &rq->rx_ring[ring_idx];
567 u32 val;
568
569 while (num_allocated <= num_to_alloc) {
570 struct vmxnet3_rx_buf_info *rbi;
571 union Vmxnet3_GenericDesc *gd;
572
573 rbi = rbi_base + ring->next2fill;
574 gd = ring->base + ring->next2fill;
575
576 if (rbi->buf_type == VMXNET3_RX_BUF_SKB) {
577 if (rbi->skb == NULL) {
578 rbi->skb = __netdev_alloc_skb_ip_align(adapter->netdev,
579 rbi->len,
580 GFP_KERNEL);
581 if (unlikely(rbi->skb == NULL)) {
582 rq->stats.rx_buf_alloc_failure++;
583 break;
584 }
585
586 rbi->dma_addr = dma_map_single(
587 &adapter->pdev->dev,
588 rbi->skb->data, rbi->len,
589 PCI_DMA_FROMDEVICE);
590 if (dma_mapping_error(&adapter->pdev->dev,
591 rbi->dma_addr)) {
592 dev_kfree_skb_any(rbi->skb);
593 rq->stats.rx_buf_alloc_failure++;
594 break;
595 }
596 } else {
597 /* rx buffer skipped by the device */
598 }
599 val = VMXNET3_RXD_BTYPE_HEAD << VMXNET3_RXD_BTYPE_SHIFT;
600 } else {
601 BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_PAGE ||
602 rbi->len != PAGE_SIZE);
603
604 if (rbi->page == NULL) {
605 rbi->page = alloc_page(GFP_ATOMIC);
606 if (unlikely(rbi->page == NULL)) {
607 rq->stats.rx_buf_alloc_failure++;
608 break;
609 }
610 rbi->dma_addr = dma_map_page(
611 &adapter->pdev->dev,
612 rbi->page, 0, PAGE_SIZE,
613 PCI_DMA_FROMDEVICE);
614 if (dma_mapping_error(&adapter->pdev->dev,
615 rbi->dma_addr)) {
616 put_page(rbi->page);
617 rq->stats.rx_buf_alloc_failure++;
618 break;
619 }
620 } else {
621 /* rx buffers skipped by the device */
622 }
623 val = VMXNET3_RXD_BTYPE_BODY << VMXNET3_RXD_BTYPE_SHIFT;
624 }
625
626 gd->rxd.addr = cpu_to_le64(rbi->dma_addr);
627 gd->dword[2] = cpu_to_le32((!ring->gen << VMXNET3_RXD_GEN_SHIFT)
628 | val | rbi->len);
629
630 /* Fill the last buffer but dont mark it ready, or else the
631 * device will think that the queue is full */
632 if (num_allocated == num_to_alloc)
633 break;
634
635 gd->dword[2] |= cpu_to_le32(ring->gen << VMXNET3_RXD_GEN_SHIFT);
636 num_allocated++;
637 vmxnet3_cmd_ring_adv_next2fill(ring);
638 }
639
640 netdev_dbg(adapter->netdev,
641 "alloc_rx_buf: %d allocated, next2fill %u, next2comp %u\n",
642 num_allocated, ring->next2fill, ring->next2comp);
643
644 /* so that the device can distinguish a full ring and an empty ring */
645 BUG_ON(num_allocated != 0 && ring->next2fill == ring->next2comp);
646
647 return num_allocated;
648 }
649
650
651 static void
652 vmxnet3_append_frag(struct sk_buff *skb, struct Vmxnet3_RxCompDesc *rcd,
653 struct vmxnet3_rx_buf_info *rbi)
654 {
655 struct skb_frag_struct *frag = skb_shinfo(skb)->frags +
656 skb_shinfo(skb)->nr_frags;
657
658 BUG_ON(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS);
659
660 __skb_frag_set_page(frag, rbi->page);
661 frag->page_offset = 0;
662 skb_frag_size_set(frag, rcd->len);
663 skb->data_len += rcd->len;
664 skb->truesize += PAGE_SIZE;
665 skb_shinfo(skb)->nr_frags++;
666 }
667
668
669 static int
670 vmxnet3_map_pkt(struct sk_buff *skb, struct vmxnet3_tx_ctx *ctx,
671 struct vmxnet3_tx_queue *tq, struct pci_dev *pdev,
672 struct vmxnet3_adapter *adapter)
673 {
674 u32 dw2, len;
675 unsigned long buf_offset;
676 int i;
677 union Vmxnet3_GenericDesc *gdesc;
678 struct vmxnet3_tx_buf_info *tbi = NULL;
679
680 BUG_ON(ctx->copy_size > skb_headlen(skb));
681
682 /* use the previous gen bit for the SOP desc */
683 dw2 = (tq->tx_ring.gen ^ 0x1) << VMXNET3_TXD_GEN_SHIFT;
684
685 ctx->sop_txd = tq->tx_ring.base + tq->tx_ring.next2fill;
686 gdesc = ctx->sop_txd; /* both loops below can be skipped */
687
688 /* no need to map the buffer if headers are copied */
689 if (ctx->copy_size) {
690 ctx->sop_txd->txd.addr = cpu_to_le64(tq->data_ring.basePA +
691 tq->tx_ring.next2fill *
692 tq->txdata_desc_size);
693 ctx->sop_txd->dword[2] = cpu_to_le32(dw2 | ctx->copy_size);
694 ctx->sop_txd->dword[3] = 0;
695
696 tbi = tq->buf_info + tq->tx_ring.next2fill;
697 tbi->map_type = VMXNET3_MAP_NONE;
698
699 netdev_dbg(adapter->netdev,
700 "txd[%u]: 0x%Lx 0x%x 0x%x\n",
701 tq->tx_ring.next2fill,
702 le64_to_cpu(ctx->sop_txd->txd.addr),
703 ctx->sop_txd->dword[2], ctx->sop_txd->dword[3]);
704 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
705
706 /* use the right gen for non-SOP desc */
707 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
708 }
709
710 /* linear part can use multiple tx desc if it's big */
711 len = skb_headlen(skb) - ctx->copy_size;
712 buf_offset = ctx->copy_size;
713 while (len) {
714 u32 buf_size;
715
716 if (len < VMXNET3_MAX_TX_BUF_SIZE) {
717 buf_size = len;
718 dw2 |= len;
719 } else {
720 buf_size = VMXNET3_MAX_TX_BUF_SIZE;
721 /* spec says that for TxDesc.len, 0 == 2^14 */
722 }
723
724 tbi = tq->buf_info + tq->tx_ring.next2fill;
725 tbi->map_type = VMXNET3_MAP_SINGLE;
726 tbi->dma_addr = dma_map_single(&adapter->pdev->dev,
727 skb->data + buf_offset, buf_size,
728 PCI_DMA_TODEVICE);
729 if (dma_mapping_error(&adapter->pdev->dev, tbi->dma_addr))
730 return -EFAULT;
731
732 tbi->len = buf_size;
733
734 gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
735 BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
736
737 gdesc->txd.addr = cpu_to_le64(tbi->dma_addr);
738 gdesc->dword[2] = cpu_to_le32(dw2);
739 gdesc->dword[3] = 0;
740
741 netdev_dbg(adapter->netdev,
742 "txd[%u]: 0x%Lx 0x%x 0x%x\n",
743 tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr),
744 le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]);
745 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
746 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
747
748 len -= buf_size;
749 buf_offset += buf_size;
750 }
751
752 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
753 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
754 u32 buf_size;
755
756 buf_offset = 0;
757 len = skb_frag_size(frag);
758 while (len) {
759 tbi = tq->buf_info + tq->tx_ring.next2fill;
760 if (len < VMXNET3_MAX_TX_BUF_SIZE) {
761 buf_size = len;
762 dw2 |= len;
763 } else {
764 buf_size = VMXNET3_MAX_TX_BUF_SIZE;
765 /* spec says that for TxDesc.len, 0 == 2^14 */
766 }
767 tbi->map_type = VMXNET3_MAP_PAGE;
768 tbi->dma_addr = skb_frag_dma_map(&adapter->pdev->dev, frag,
769 buf_offset, buf_size,
770 DMA_TO_DEVICE);
771 if (dma_mapping_error(&adapter->pdev->dev, tbi->dma_addr))
772 return -EFAULT;
773
774 tbi->len = buf_size;
775
776 gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
777 BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
778
779 gdesc->txd.addr = cpu_to_le64(tbi->dma_addr);
780 gdesc->dword[2] = cpu_to_le32(dw2);
781 gdesc->dword[3] = 0;
782
783 netdev_dbg(adapter->netdev,
784 "txd[%u]: 0x%llx %u %u\n",
785 tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr),
786 le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]);
787 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
788 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
789
790 len -= buf_size;
791 buf_offset += buf_size;
792 }
793 }
794
795 ctx->eop_txd = gdesc;
796
797 /* set the last buf_info for the pkt */
798 tbi->skb = skb;
799 tbi->sop_idx = ctx->sop_txd - tq->tx_ring.base;
800
801 return 0;
802 }
803
804
805 /* Init all tx queues */
806 static void
807 vmxnet3_tq_init_all(struct vmxnet3_adapter *adapter)
808 {
809 int i;
810
811 for (i = 0; i < adapter->num_tx_queues; i++)
812 vmxnet3_tq_init(&adapter->tx_queue[i], adapter);
813 }
814
815
816 /*
817 * parse relevant protocol headers:
818 * For a tso pkt, relevant headers are L2/3/4 including options
819 * For a pkt requesting csum offloading, they are L2/3 and may include L4
820 * if it's a TCP/UDP pkt
821 *
822 * Returns:
823 * -1: error happens during parsing
824 * 0: protocol headers parsed, but too big to be copied
825 * 1: protocol headers parsed and copied
826 *
827 * Other effects:
828 * 1. related *ctx fields are updated.
829 * 2. ctx->copy_size is # of bytes copied
830 * 3. the portion to be copied is guaranteed to be in the linear part
831 *
832 */
833 static int
834 vmxnet3_parse_hdr(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
835 struct vmxnet3_tx_ctx *ctx,
836 struct vmxnet3_adapter *adapter)
837 {
838 u8 protocol = 0;
839
840 if (ctx->mss) { /* TSO */
841 ctx->eth_ip_hdr_size = skb_transport_offset(skb);
842 ctx->l4_hdr_size = tcp_hdrlen(skb);
843 ctx->copy_size = ctx->eth_ip_hdr_size + ctx->l4_hdr_size;
844 } else {
845 if (skb->ip_summed == CHECKSUM_PARTIAL) {
846 ctx->eth_ip_hdr_size = skb_checksum_start_offset(skb);
847
848 if (ctx->ipv4) {
849 const struct iphdr *iph = ip_hdr(skb);
850
851 protocol = iph->protocol;
852 } else if (ctx->ipv6) {
853 const struct ipv6hdr *ipv6h = ipv6_hdr(skb);
854
855 protocol = ipv6h->nexthdr;
856 }
857
858 switch (protocol) {
859 case IPPROTO_TCP:
860 ctx->l4_hdr_size = tcp_hdrlen(skb);
861 break;
862 case IPPROTO_UDP:
863 ctx->l4_hdr_size = sizeof(struct udphdr);
864 break;
865 default:
866 ctx->l4_hdr_size = 0;
867 break;
868 }
869
870 ctx->copy_size = min(ctx->eth_ip_hdr_size +
871 ctx->l4_hdr_size, skb->len);
872 } else {
873 ctx->eth_ip_hdr_size = 0;
874 ctx->l4_hdr_size = 0;
875 /* copy as much as allowed */
876 ctx->copy_size = min_t(unsigned int,
877 tq->txdata_desc_size,
878 skb_headlen(skb));
879 }
880
881 if (skb->len <= VMXNET3_HDR_COPY_SIZE)
882 ctx->copy_size = skb->len;
883
884 /* make sure headers are accessible directly */
885 if (unlikely(!pskb_may_pull(skb, ctx->copy_size)))
886 goto err;
887 }
888
889 if (unlikely(ctx->copy_size > tq->txdata_desc_size)) {
890 tq->stats.oversized_hdr++;
891 ctx->copy_size = 0;
892 return 0;
893 }
894
895 return 1;
896 err:
897 return -1;
898 }
899
900 /*
901 * copy relevant protocol headers to the transmit ring:
902 * For a tso pkt, relevant headers are L2/3/4 including options
903 * For a pkt requesting csum offloading, they are L2/3 and may include L4
904 * if it's a TCP/UDP pkt
905 *
906 *
907 * Note that this requires that vmxnet3_parse_hdr be called first to set the
908 * appropriate bits in ctx first
909 */
910 static void
911 vmxnet3_copy_hdr(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
912 struct vmxnet3_tx_ctx *ctx,
913 struct vmxnet3_adapter *adapter)
914 {
915 struct Vmxnet3_TxDataDesc *tdd;
916
917 tdd = (struct Vmxnet3_TxDataDesc *)((u8 *)tq->data_ring.base +
918 tq->tx_ring.next2fill *
919 tq->txdata_desc_size);
920
921 memcpy(tdd->data, skb->data, ctx->copy_size);
922 netdev_dbg(adapter->netdev,
923 "copy %u bytes to dataRing[%u]\n",
924 ctx->copy_size, tq->tx_ring.next2fill);
925 }
926
927
928 static void
929 vmxnet3_prepare_tso(struct sk_buff *skb,
930 struct vmxnet3_tx_ctx *ctx)
931 {
932 struct tcphdr *tcph = tcp_hdr(skb);
933
934 if (ctx->ipv4) {
935 struct iphdr *iph = ip_hdr(skb);
936
937 iph->check = 0;
938 tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, 0,
939 IPPROTO_TCP, 0);
940 } else if (ctx->ipv6) {
941 struct ipv6hdr *iph = ipv6_hdr(skb);
942
943 tcph->check = ~csum_ipv6_magic(&iph->saddr, &iph->daddr, 0,
944 IPPROTO_TCP, 0);
945 }
946 }
947
948 static int txd_estimate(const struct sk_buff *skb)
949 {
950 int count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) + 1;
951 int i;
952
953 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
954 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
955
956 count += VMXNET3_TXD_NEEDED(skb_frag_size(frag));
957 }
958 return count;
959 }
960
961 /*
962 * Transmits a pkt thru a given tq
963 * Returns:
964 * NETDEV_TX_OK: descriptors are setup successfully
965 * NETDEV_TX_OK: error occurred, the pkt is dropped
966 * NETDEV_TX_BUSY: tx ring is full, queue is stopped
967 *
968 * Side-effects:
969 * 1. tx ring may be changed
970 * 2. tq stats may be updated accordingly
971 * 3. shared->txNumDeferred may be updated
972 */
973
974 static int
975 vmxnet3_tq_xmit(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
976 struct vmxnet3_adapter *adapter, struct net_device *netdev)
977 {
978 int ret;
979 u32 count;
980 int num_pkts;
981 int tx_num_deferred;
982 unsigned long flags;
983 struct vmxnet3_tx_ctx ctx;
984 union Vmxnet3_GenericDesc *gdesc;
985 #ifdef __BIG_ENDIAN_BITFIELD
986 /* Use temporary descriptor to avoid touching bits multiple times */
987 union Vmxnet3_GenericDesc tempTxDesc;
988 #endif
989
990 count = txd_estimate(skb);
991
992 ctx.ipv4 = (vlan_get_protocol(skb) == cpu_to_be16(ETH_P_IP));
993 ctx.ipv6 = (vlan_get_protocol(skb) == cpu_to_be16(ETH_P_IPV6));
994
995 ctx.mss = skb_shinfo(skb)->gso_size;
996 if (ctx.mss) {
997 if (skb_header_cloned(skb)) {
998 if (unlikely(pskb_expand_head(skb, 0, 0,
999 GFP_ATOMIC) != 0)) {
1000 tq->stats.drop_tso++;
1001 goto drop_pkt;
1002 }
1003 tq->stats.copy_skb_header++;
1004 }
1005 vmxnet3_prepare_tso(skb, &ctx);
1006 } else {
1007 if (unlikely(count > VMXNET3_MAX_TXD_PER_PKT)) {
1008
1009 /* non-tso pkts must not use more than
1010 * VMXNET3_MAX_TXD_PER_PKT entries
1011 */
1012 if (skb_linearize(skb) != 0) {
1013 tq->stats.drop_too_many_frags++;
1014 goto drop_pkt;
1015 }
1016 tq->stats.linearized++;
1017
1018 /* recalculate the # of descriptors to use */
1019 count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) + 1;
1020 }
1021 }
1022
1023 ret = vmxnet3_parse_hdr(skb, tq, &ctx, adapter);
1024 if (ret >= 0) {
1025 BUG_ON(ret <= 0 && ctx.copy_size != 0);
1026 /* hdrs parsed, check against other limits */
1027 if (ctx.mss) {
1028 if (unlikely(ctx.eth_ip_hdr_size + ctx.l4_hdr_size >
1029 VMXNET3_MAX_TX_BUF_SIZE)) {
1030 tq->stats.drop_oversized_hdr++;
1031 goto drop_pkt;
1032 }
1033 } else {
1034 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1035 if (unlikely(ctx.eth_ip_hdr_size +
1036 skb->csum_offset >
1037 VMXNET3_MAX_CSUM_OFFSET)) {
1038 tq->stats.drop_oversized_hdr++;
1039 goto drop_pkt;
1040 }
1041 }
1042 }
1043 } else {
1044 tq->stats.drop_hdr_inspect_err++;
1045 goto drop_pkt;
1046 }
1047
1048 spin_lock_irqsave(&tq->tx_lock, flags);
1049
1050 if (count > vmxnet3_cmd_ring_desc_avail(&tq->tx_ring)) {
1051 tq->stats.tx_ring_full++;
1052 netdev_dbg(adapter->netdev,
1053 "tx queue stopped on %s, next2comp %u"
1054 " next2fill %u\n", adapter->netdev->name,
1055 tq->tx_ring.next2comp, tq->tx_ring.next2fill);
1056
1057 vmxnet3_tq_stop(tq, adapter);
1058 spin_unlock_irqrestore(&tq->tx_lock, flags);
1059 return NETDEV_TX_BUSY;
1060 }
1061
1062
1063 vmxnet3_copy_hdr(skb, tq, &ctx, adapter);
1064
1065 /* fill tx descs related to addr & len */
1066 if (vmxnet3_map_pkt(skb, &ctx, tq, adapter->pdev, adapter))
1067 goto unlock_drop_pkt;
1068
1069 /* setup the EOP desc */
1070 ctx.eop_txd->dword[3] = cpu_to_le32(VMXNET3_TXD_CQ | VMXNET3_TXD_EOP);
1071
1072 /* setup the SOP desc */
1073 #ifdef __BIG_ENDIAN_BITFIELD
1074 gdesc = &tempTxDesc;
1075 gdesc->dword[2] = ctx.sop_txd->dword[2];
1076 gdesc->dword[3] = ctx.sop_txd->dword[3];
1077 #else
1078 gdesc = ctx.sop_txd;
1079 #endif
1080 tx_num_deferred = le32_to_cpu(tq->shared->txNumDeferred);
1081 if (ctx.mss) {
1082 gdesc->txd.hlen = ctx.eth_ip_hdr_size + ctx.l4_hdr_size;
1083 gdesc->txd.om = VMXNET3_OM_TSO;
1084 gdesc->txd.msscof = ctx.mss;
1085 num_pkts = (skb->len - gdesc->txd.hlen + ctx.mss - 1) / ctx.mss;
1086 } else {
1087 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1088 gdesc->txd.hlen = ctx.eth_ip_hdr_size;
1089 gdesc->txd.om = VMXNET3_OM_CSUM;
1090 gdesc->txd.msscof = ctx.eth_ip_hdr_size +
1091 skb->csum_offset;
1092 } else {
1093 gdesc->txd.om = 0;
1094 gdesc->txd.msscof = 0;
1095 }
1096 num_pkts = 1;
1097 }
1098 le32_add_cpu(&tq->shared->txNumDeferred, num_pkts);
1099 tx_num_deferred += num_pkts;
1100
1101 if (skb_vlan_tag_present(skb)) {
1102 gdesc->txd.ti = 1;
1103 gdesc->txd.tci = skb_vlan_tag_get(skb);
1104 }
1105
1106 /* finally flips the GEN bit of the SOP desc. */
1107 gdesc->dword[2] = cpu_to_le32(le32_to_cpu(gdesc->dword[2]) ^
1108 VMXNET3_TXD_GEN);
1109 #ifdef __BIG_ENDIAN_BITFIELD
1110 /* Finished updating in bitfields of Tx Desc, so write them in original
1111 * place.
1112 */
1113 vmxnet3_TxDescToLe((struct Vmxnet3_TxDesc *)gdesc,
1114 (struct Vmxnet3_TxDesc *)ctx.sop_txd);
1115 gdesc = ctx.sop_txd;
1116 #endif
1117 netdev_dbg(adapter->netdev,
1118 "txd[%u]: SOP 0x%Lx 0x%x 0x%x\n",
1119 (u32)(ctx.sop_txd -
1120 tq->tx_ring.base), le64_to_cpu(gdesc->txd.addr),
1121 le32_to_cpu(gdesc->dword[2]), le32_to_cpu(gdesc->dword[3]));
1122
1123 spin_unlock_irqrestore(&tq->tx_lock, flags);
1124
1125 if (tx_num_deferred >= le32_to_cpu(tq->shared->txThreshold)) {
1126 tq->shared->txNumDeferred = 0;
1127 VMXNET3_WRITE_BAR0_REG(adapter,
1128 VMXNET3_REG_TXPROD + tq->qid * 8,
1129 tq->tx_ring.next2fill);
1130 }
1131
1132 return NETDEV_TX_OK;
1133
1134 unlock_drop_pkt:
1135 spin_unlock_irqrestore(&tq->tx_lock, flags);
1136 drop_pkt:
1137 tq->stats.drop_total++;
1138 dev_kfree_skb_any(skb);
1139 return NETDEV_TX_OK;
1140 }
1141
1142
1143 static netdev_tx_t
1144 vmxnet3_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
1145 {
1146 struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1147
1148 BUG_ON(skb->queue_mapping > adapter->num_tx_queues);
1149 return vmxnet3_tq_xmit(skb,
1150 &adapter->tx_queue[skb->queue_mapping],
1151 adapter, netdev);
1152 }
1153
1154
1155 static void
1156 vmxnet3_rx_csum(struct vmxnet3_adapter *adapter,
1157 struct sk_buff *skb,
1158 union Vmxnet3_GenericDesc *gdesc)
1159 {
1160 if (!gdesc->rcd.cnc && adapter->netdev->features & NETIF_F_RXCSUM) {
1161 if (gdesc->rcd.v4 &&
1162 (le32_to_cpu(gdesc->dword[3]) &
1163 VMXNET3_RCD_CSUM_OK) == VMXNET3_RCD_CSUM_OK) {
1164 skb->ip_summed = CHECKSUM_UNNECESSARY;
1165 BUG_ON(!(gdesc->rcd.tcp || gdesc->rcd.udp));
1166 BUG_ON(gdesc->rcd.frg);
1167 } else if (gdesc->rcd.v6 && (le32_to_cpu(gdesc->dword[3]) &
1168 (1 << VMXNET3_RCD_TUC_SHIFT))) {
1169 skb->ip_summed = CHECKSUM_UNNECESSARY;
1170 BUG_ON(!(gdesc->rcd.tcp || gdesc->rcd.udp));
1171 BUG_ON(gdesc->rcd.frg);
1172 } else {
1173 if (gdesc->rcd.csum) {
1174 skb->csum = htons(gdesc->rcd.csum);
1175 skb->ip_summed = CHECKSUM_PARTIAL;
1176 } else {
1177 skb_checksum_none_assert(skb);
1178 }
1179 }
1180 } else {
1181 skb_checksum_none_assert(skb);
1182 }
1183 }
1184
1185
1186 static void
1187 vmxnet3_rx_error(struct vmxnet3_rx_queue *rq, struct Vmxnet3_RxCompDesc *rcd,
1188 struct vmxnet3_rx_ctx *ctx, struct vmxnet3_adapter *adapter)
1189 {
1190 rq->stats.drop_err++;
1191 if (!rcd->fcs)
1192 rq->stats.drop_fcs++;
1193
1194 rq->stats.drop_total++;
1195
1196 /*
1197 * We do not unmap and chain the rx buffer to the skb.
1198 * We basically pretend this buffer is not used and will be recycled
1199 * by vmxnet3_rq_alloc_rx_buf()
1200 */
1201
1202 /*
1203 * ctx->skb may be NULL if this is the first and the only one
1204 * desc for the pkt
1205 */
1206 if (ctx->skb)
1207 dev_kfree_skb_irq(ctx->skb);
1208
1209 ctx->skb = NULL;
1210 }
1211
1212
1213 static u32
1214 vmxnet3_get_hdr_len(struct vmxnet3_adapter *adapter, struct sk_buff *skb,
1215 union Vmxnet3_GenericDesc *gdesc)
1216 {
1217 u32 hlen, maplen;
1218 union {
1219 void *ptr;
1220 struct ethhdr *eth;
1221 struct iphdr *ipv4;
1222 struct ipv6hdr *ipv6;
1223 struct tcphdr *tcp;
1224 } hdr;
1225 BUG_ON(gdesc->rcd.tcp == 0);
1226
1227 maplen = skb_headlen(skb);
1228 if (unlikely(sizeof(struct iphdr) + sizeof(struct tcphdr) > maplen))
1229 return 0;
1230
1231 hdr.eth = eth_hdr(skb);
1232 if (gdesc->rcd.v4) {
1233 BUG_ON(hdr.eth->h_proto != htons(ETH_P_IP));
1234 hdr.ptr += sizeof(struct ethhdr);
1235 BUG_ON(hdr.ipv4->protocol != IPPROTO_TCP);
1236 hlen = hdr.ipv4->ihl << 2;
1237 hdr.ptr += hdr.ipv4->ihl << 2;
1238 } else if (gdesc->rcd.v6) {
1239 BUG_ON(hdr.eth->h_proto != htons(ETH_P_IPV6));
1240 hdr.ptr += sizeof(struct ethhdr);
1241 /* Use an estimated value, since we also need to handle
1242 * TSO case.
1243 */
1244 if (hdr.ipv6->nexthdr != IPPROTO_TCP)
1245 return sizeof(struct ipv6hdr) + sizeof(struct tcphdr);
1246 hlen = sizeof(struct ipv6hdr);
1247 hdr.ptr += sizeof(struct ipv6hdr);
1248 } else {
1249 /* Non-IP pkt, dont estimate header length */
1250 return 0;
1251 }
1252
1253 if (hlen + sizeof(struct tcphdr) > maplen)
1254 return 0;
1255
1256 return (hlen + (hdr.tcp->doff << 2));
1257 }
1258
1259 static int
1260 vmxnet3_rq_rx_complete(struct vmxnet3_rx_queue *rq,
1261 struct vmxnet3_adapter *adapter, int quota)
1262 {
1263 static const u32 rxprod_reg[2] = {
1264 VMXNET3_REG_RXPROD, VMXNET3_REG_RXPROD2
1265 };
1266 u32 num_pkts = 0;
1267 bool skip_page_frags = false;
1268 struct Vmxnet3_RxCompDesc *rcd;
1269 struct vmxnet3_rx_ctx *ctx = &rq->rx_ctx;
1270 u16 segCnt = 0, mss = 0;
1271 #ifdef __BIG_ENDIAN_BITFIELD
1272 struct Vmxnet3_RxDesc rxCmdDesc;
1273 struct Vmxnet3_RxCompDesc rxComp;
1274 #endif
1275 vmxnet3_getRxComp(rcd, &rq->comp_ring.base[rq->comp_ring.next2proc].rcd,
1276 &rxComp);
1277 while (rcd->gen == rq->comp_ring.gen) {
1278 struct vmxnet3_rx_buf_info *rbi;
1279 struct sk_buff *skb, *new_skb = NULL;
1280 struct page *new_page = NULL;
1281 dma_addr_t new_dma_addr;
1282 int num_to_alloc;
1283 struct Vmxnet3_RxDesc *rxd;
1284 u32 idx, ring_idx;
1285 struct vmxnet3_cmd_ring *ring = NULL;
1286 if (num_pkts >= quota) {
1287 /* we may stop even before we see the EOP desc of
1288 * the current pkt
1289 */
1290 break;
1291 }
1292 BUG_ON(rcd->rqID != rq->qid && rcd->rqID != rq->qid2 &&
1293 rcd->rqID != rq->dataRingQid);
1294 idx = rcd->rxdIdx;
1295 ring_idx = VMXNET3_GET_RING_IDX(adapter, rcd->rqID);
1296 ring = rq->rx_ring + ring_idx;
1297 vmxnet3_getRxDesc(rxd, &rq->rx_ring[ring_idx].base[idx].rxd,
1298 &rxCmdDesc);
1299 rbi = rq->buf_info[ring_idx] + idx;
1300
1301 BUG_ON(rxd->addr != rbi->dma_addr ||
1302 rxd->len != rbi->len);
1303
1304 if (unlikely(rcd->eop && rcd->err)) {
1305 vmxnet3_rx_error(rq, rcd, ctx, adapter);
1306 goto rcd_done;
1307 }
1308
1309 if (rcd->sop) { /* first buf of the pkt */
1310 bool rxDataRingUsed;
1311 u16 len;
1312
1313 BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_HEAD ||
1314 (rcd->rqID != rq->qid &&
1315 rcd->rqID != rq->dataRingQid));
1316
1317 BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_SKB);
1318 BUG_ON(ctx->skb != NULL || rbi->skb == NULL);
1319
1320 if (unlikely(rcd->len == 0)) {
1321 /* Pretend the rx buffer is skipped. */
1322 BUG_ON(!(rcd->sop && rcd->eop));
1323 netdev_dbg(adapter->netdev,
1324 "rxRing[%u][%u] 0 length\n",
1325 ring_idx, idx);
1326 goto rcd_done;
1327 }
1328
1329 skip_page_frags = false;
1330 ctx->skb = rbi->skb;
1331
1332 rxDataRingUsed =
1333 VMXNET3_RX_DATA_RING(adapter, rcd->rqID);
1334 len = rxDataRingUsed ? rcd->len : rbi->len;
1335 new_skb = netdev_alloc_skb_ip_align(adapter->netdev,
1336 len);
1337 if (new_skb == NULL) {
1338 /* Skb allocation failed, do not handover this
1339 * skb to stack. Reuse it. Drop the existing pkt
1340 */
1341 rq->stats.rx_buf_alloc_failure++;
1342 ctx->skb = NULL;
1343 rq->stats.drop_total++;
1344 skip_page_frags = true;
1345 goto rcd_done;
1346 }
1347
1348 if (rxDataRingUsed) {
1349 size_t sz;
1350
1351 BUG_ON(rcd->len > rq->data_ring.desc_size);
1352
1353 ctx->skb = new_skb;
1354 sz = rcd->rxdIdx * rq->data_ring.desc_size;
1355 memcpy(new_skb->data,
1356 &rq->data_ring.base[sz], rcd->len);
1357 } else {
1358 ctx->skb = rbi->skb;
1359
1360 new_dma_addr =
1361 dma_map_single(&adapter->pdev->dev,
1362 new_skb->data, rbi->len,
1363 PCI_DMA_FROMDEVICE);
1364 if (dma_mapping_error(&adapter->pdev->dev,
1365 new_dma_addr)) {
1366 dev_kfree_skb(new_skb);
1367 /* Skb allocation failed, do not
1368 * handover this skb to stack. Reuse
1369 * it. Drop the existing pkt.
1370 */
1371 rq->stats.rx_buf_alloc_failure++;
1372 ctx->skb = NULL;
1373 rq->stats.drop_total++;
1374 skip_page_frags = true;
1375 goto rcd_done;
1376 }
1377
1378 dma_unmap_single(&adapter->pdev->dev,
1379 rbi->dma_addr,
1380 rbi->len,
1381 PCI_DMA_FROMDEVICE);
1382
1383 /* Immediate refill */
1384 rbi->skb = new_skb;
1385 rbi->dma_addr = new_dma_addr;
1386 rxd->addr = cpu_to_le64(rbi->dma_addr);
1387 rxd->len = rbi->len;
1388 }
1389
1390 #ifdef VMXNET3_RSS
1391 if (rcd->rssType != VMXNET3_RCD_RSS_TYPE_NONE &&
1392 (adapter->netdev->features & NETIF_F_RXHASH))
1393 skb_set_hash(ctx->skb,
1394 le32_to_cpu(rcd->rssHash),
1395 PKT_HASH_TYPE_L3);
1396 #endif
1397 skb_put(ctx->skb, rcd->len);
1398
1399 if (VMXNET3_VERSION_GE_2(adapter) &&
1400 rcd->type == VMXNET3_CDTYPE_RXCOMP_LRO) {
1401 struct Vmxnet3_RxCompDescExt *rcdlro;
1402 rcdlro = (struct Vmxnet3_RxCompDescExt *)rcd;
1403
1404 segCnt = rcdlro->segCnt;
1405 WARN_ON_ONCE(segCnt == 0);
1406 mss = rcdlro->mss;
1407 if (unlikely(segCnt <= 1))
1408 segCnt = 0;
1409 } else {
1410 segCnt = 0;
1411 }
1412 } else {
1413 BUG_ON(ctx->skb == NULL && !skip_page_frags);
1414
1415 /* non SOP buffer must be type 1 in most cases */
1416 BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_PAGE);
1417 BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_BODY);
1418
1419 /* If an sop buffer was dropped, skip all
1420 * following non-sop fragments. They will be reused.
1421 */
1422 if (skip_page_frags)
1423 goto rcd_done;
1424
1425 if (rcd->len) {
1426 new_page = alloc_page(GFP_ATOMIC);
1427 /* Replacement page frag could not be allocated.
1428 * Reuse this page. Drop the pkt and free the
1429 * skb which contained this page as a frag. Skip
1430 * processing all the following non-sop frags.
1431 */
1432 if (unlikely(!new_page)) {
1433 rq->stats.rx_buf_alloc_failure++;
1434 dev_kfree_skb(ctx->skb);
1435 ctx->skb = NULL;
1436 skip_page_frags = true;
1437 goto rcd_done;
1438 }
1439 new_dma_addr = dma_map_page(&adapter->pdev->dev,
1440 new_page,
1441 0, PAGE_SIZE,
1442 PCI_DMA_FROMDEVICE);
1443 if (dma_mapping_error(&adapter->pdev->dev,
1444 new_dma_addr)) {
1445 put_page(new_page);
1446 rq->stats.rx_buf_alloc_failure++;
1447 dev_kfree_skb(ctx->skb);
1448 ctx->skb = NULL;
1449 skip_page_frags = true;
1450 goto rcd_done;
1451 }
1452
1453 dma_unmap_page(&adapter->pdev->dev,
1454 rbi->dma_addr, rbi->len,
1455 PCI_DMA_FROMDEVICE);
1456
1457 vmxnet3_append_frag(ctx->skb, rcd, rbi);
1458
1459 /* Immediate refill */
1460 rbi->page = new_page;
1461 rbi->dma_addr = new_dma_addr;
1462 rxd->addr = cpu_to_le64(rbi->dma_addr);
1463 rxd->len = rbi->len;
1464 }
1465 }
1466
1467
1468 skb = ctx->skb;
1469 if (rcd->eop) {
1470 u32 mtu = adapter->netdev->mtu;
1471 skb->len += skb->data_len;
1472
1473 vmxnet3_rx_csum(adapter, skb,
1474 (union Vmxnet3_GenericDesc *)rcd);
1475 skb->protocol = eth_type_trans(skb, adapter->netdev);
1476 if (!rcd->tcp ||
1477 !(adapter->netdev->features & NETIF_F_LRO))
1478 goto not_lro;
1479
1480 if (segCnt != 0 && mss != 0) {
1481 skb_shinfo(skb)->gso_type = rcd->v4 ?
1482 SKB_GSO_TCPV4 : SKB_GSO_TCPV6;
1483 skb_shinfo(skb)->gso_size = mss;
1484 skb_shinfo(skb)->gso_segs = segCnt;
1485 } else if (segCnt != 0 || skb->len > mtu) {
1486 u32 hlen;
1487
1488 hlen = vmxnet3_get_hdr_len(adapter, skb,
1489 (union Vmxnet3_GenericDesc *)rcd);
1490 if (hlen == 0)
1491 goto not_lro;
1492
1493 skb_shinfo(skb)->gso_type =
1494 rcd->v4 ? SKB_GSO_TCPV4 : SKB_GSO_TCPV6;
1495 if (segCnt != 0) {
1496 skb_shinfo(skb)->gso_segs = segCnt;
1497 skb_shinfo(skb)->gso_size =
1498 DIV_ROUND_UP(skb->len -
1499 hlen, segCnt);
1500 } else {
1501 skb_shinfo(skb)->gso_size = mtu - hlen;
1502 }
1503 }
1504 not_lro:
1505 if (unlikely(rcd->ts))
1506 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), rcd->tci);
1507
1508 if (adapter->netdev->features & NETIF_F_LRO)
1509 netif_receive_skb(skb);
1510 else
1511 napi_gro_receive(&rq->napi, skb);
1512
1513 ctx->skb = NULL;
1514 num_pkts++;
1515 }
1516
1517 rcd_done:
1518 /* device may have skipped some rx descs */
1519 ring->next2comp = idx;
1520 num_to_alloc = vmxnet3_cmd_ring_desc_avail(ring);
1521 ring = rq->rx_ring + ring_idx;
1522 while (num_to_alloc) {
1523 vmxnet3_getRxDesc(rxd, &ring->base[ring->next2fill].rxd,
1524 &rxCmdDesc);
1525 BUG_ON(!rxd->addr);
1526
1527 /* Recv desc is ready to be used by the device */
1528 rxd->gen = ring->gen;
1529 vmxnet3_cmd_ring_adv_next2fill(ring);
1530 num_to_alloc--;
1531 }
1532
1533 /* if needed, update the register */
1534 if (unlikely(rq->shared->updateRxProd)) {
1535 VMXNET3_WRITE_BAR0_REG(adapter,
1536 rxprod_reg[ring_idx] + rq->qid * 8,
1537 ring->next2fill);
1538 }
1539
1540 vmxnet3_comp_ring_adv_next2proc(&rq->comp_ring);
1541 vmxnet3_getRxComp(rcd,
1542 &rq->comp_ring.base[rq->comp_ring.next2proc].rcd, &rxComp);
1543 }
1544
1545 return num_pkts;
1546 }
1547
1548
1549 static void
1550 vmxnet3_rq_cleanup(struct vmxnet3_rx_queue *rq,
1551 struct vmxnet3_adapter *adapter)
1552 {
1553 u32 i, ring_idx;
1554 struct Vmxnet3_RxDesc *rxd;
1555
1556 for (ring_idx = 0; ring_idx < 2; ring_idx++) {
1557 for (i = 0; i < rq->rx_ring[ring_idx].size; i++) {
1558 #ifdef __BIG_ENDIAN_BITFIELD
1559 struct Vmxnet3_RxDesc rxDesc;
1560 #endif
1561 vmxnet3_getRxDesc(rxd,
1562 &rq->rx_ring[ring_idx].base[i].rxd, &rxDesc);
1563
1564 if (rxd->btype == VMXNET3_RXD_BTYPE_HEAD &&
1565 rq->buf_info[ring_idx][i].skb) {
1566 dma_unmap_single(&adapter->pdev->dev, rxd->addr,
1567 rxd->len, PCI_DMA_FROMDEVICE);
1568 dev_kfree_skb(rq->buf_info[ring_idx][i].skb);
1569 rq->buf_info[ring_idx][i].skb = NULL;
1570 } else if (rxd->btype == VMXNET3_RXD_BTYPE_BODY &&
1571 rq->buf_info[ring_idx][i].page) {
1572 dma_unmap_page(&adapter->pdev->dev, rxd->addr,
1573 rxd->len, PCI_DMA_FROMDEVICE);
1574 put_page(rq->buf_info[ring_idx][i].page);
1575 rq->buf_info[ring_idx][i].page = NULL;
1576 }
1577 }
1578
1579 rq->rx_ring[ring_idx].gen = VMXNET3_INIT_GEN;
1580 rq->rx_ring[ring_idx].next2fill =
1581 rq->rx_ring[ring_idx].next2comp = 0;
1582 }
1583
1584 rq->comp_ring.gen = VMXNET3_INIT_GEN;
1585 rq->comp_ring.next2proc = 0;
1586 }
1587
1588
1589 static void
1590 vmxnet3_rq_cleanup_all(struct vmxnet3_adapter *adapter)
1591 {
1592 int i;
1593
1594 for (i = 0; i < adapter->num_rx_queues; i++)
1595 vmxnet3_rq_cleanup(&adapter->rx_queue[i], adapter);
1596 }
1597
1598
1599 static void vmxnet3_rq_destroy(struct vmxnet3_rx_queue *rq,
1600 struct vmxnet3_adapter *adapter)
1601 {
1602 int i;
1603 int j;
1604
1605 /* all rx buffers must have already been freed */
1606 for (i = 0; i < 2; i++) {
1607 if (rq->buf_info[i]) {
1608 for (j = 0; j < rq->rx_ring[i].size; j++)
1609 BUG_ON(rq->buf_info[i][j].page != NULL);
1610 }
1611 }
1612
1613
1614 for (i = 0; i < 2; i++) {
1615 if (rq->rx_ring[i].base) {
1616 dma_free_coherent(&adapter->pdev->dev,
1617 rq->rx_ring[i].size
1618 * sizeof(struct Vmxnet3_RxDesc),
1619 rq->rx_ring[i].base,
1620 rq->rx_ring[i].basePA);
1621 rq->rx_ring[i].base = NULL;
1622 }
1623 }
1624
1625 if (rq->data_ring.base) {
1626 dma_free_coherent(&adapter->pdev->dev,
1627 rq->rx_ring[0].size * rq->data_ring.desc_size,
1628 rq->data_ring.base, rq->data_ring.basePA);
1629 rq->data_ring.base = NULL;
1630 }
1631
1632 if (rq->comp_ring.base) {
1633 dma_free_coherent(&adapter->pdev->dev, rq->comp_ring.size
1634 * sizeof(struct Vmxnet3_RxCompDesc),
1635 rq->comp_ring.base, rq->comp_ring.basePA);
1636 rq->comp_ring.base = NULL;
1637 }
1638
1639 if (rq->buf_info[0]) {
1640 size_t sz = sizeof(struct vmxnet3_rx_buf_info) *
1641 (rq->rx_ring[0].size + rq->rx_ring[1].size);
1642 dma_free_coherent(&adapter->pdev->dev, sz, rq->buf_info[0],
1643 rq->buf_info_pa);
1644 rq->buf_info[0] = rq->buf_info[1] = NULL;
1645 }
1646 }
1647
1648 static void
1649 vmxnet3_rq_destroy_all_rxdataring(struct vmxnet3_adapter *adapter)
1650 {
1651 int i;
1652
1653 for (i = 0; i < adapter->num_rx_queues; i++) {
1654 struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
1655
1656 if (rq->data_ring.base) {
1657 dma_free_coherent(&adapter->pdev->dev,
1658 (rq->rx_ring[0].size *
1659 rq->data_ring.desc_size),
1660 rq->data_ring.base,
1661 rq->data_ring.basePA);
1662 rq->data_ring.base = NULL;
1663 rq->data_ring.desc_size = 0;
1664 }
1665 }
1666 }
1667
1668 static int
1669 vmxnet3_rq_init(struct vmxnet3_rx_queue *rq,
1670 struct vmxnet3_adapter *adapter)
1671 {
1672 int i;
1673
1674 /* initialize buf_info */
1675 for (i = 0; i < rq->rx_ring[0].size; i++) {
1676
1677 /* 1st buf for a pkt is skbuff */
1678 if (i % adapter->rx_buf_per_pkt == 0) {
1679 rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_SKB;
1680 rq->buf_info[0][i].len = adapter->skb_buf_size;
1681 } else { /* subsequent bufs for a pkt is frag */
1682 rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_PAGE;
1683 rq->buf_info[0][i].len = PAGE_SIZE;
1684 }
1685 }
1686 for (i = 0; i < rq->rx_ring[1].size; i++) {
1687 rq->buf_info[1][i].buf_type = VMXNET3_RX_BUF_PAGE;
1688 rq->buf_info[1][i].len = PAGE_SIZE;
1689 }
1690
1691 /* reset internal state and allocate buffers for both rings */
1692 for (i = 0; i < 2; i++) {
1693 rq->rx_ring[i].next2fill = rq->rx_ring[i].next2comp = 0;
1694
1695 memset(rq->rx_ring[i].base, 0, rq->rx_ring[i].size *
1696 sizeof(struct Vmxnet3_RxDesc));
1697 rq->rx_ring[i].gen = VMXNET3_INIT_GEN;
1698 }
1699 if (vmxnet3_rq_alloc_rx_buf(rq, 0, rq->rx_ring[0].size - 1,
1700 adapter) == 0) {
1701 /* at least has 1 rx buffer for the 1st ring */
1702 return -ENOMEM;
1703 }
1704 vmxnet3_rq_alloc_rx_buf(rq, 1, rq->rx_ring[1].size - 1, adapter);
1705
1706 /* reset the comp ring */
1707 rq->comp_ring.next2proc = 0;
1708 memset(rq->comp_ring.base, 0, rq->comp_ring.size *
1709 sizeof(struct Vmxnet3_RxCompDesc));
1710 rq->comp_ring.gen = VMXNET3_INIT_GEN;
1711
1712 /* reset rxctx */
1713 rq->rx_ctx.skb = NULL;
1714
1715 /* stats are not reset */
1716 return 0;
1717 }
1718
1719
1720 static int
1721 vmxnet3_rq_init_all(struct vmxnet3_adapter *adapter)
1722 {
1723 int i, err = 0;
1724
1725 for (i = 0; i < adapter->num_rx_queues; i++) {
1726 err = vmxnet3_rq_init(&adapter->rx_queue[i], adapter);
1727 if (unlikely(err)) {
1728 dev_err(&adapter->netdev->dev, "%s: failed to "
1729 "initialize rx queue%i\n",
1730 adapter->netdev->name, i);
1731 break;
1732 }
1733 }
1734 return err;
1735
1736 }
1737
1738
1739 static int
1740 vmxnet3_rq_create(struct vmxnet3_rx_queue *rq, struct vmxnet3_adapter *adapter)
1741 {
1742 int i;
1743 size_t sz;
1744 struct vmxnet3_rx_buf_info *bi;
1745
1746 for (i = 0; i < 2; i++) {
1747
1748 sz = rq->rx_ring[i].size * sizeof(struct Vmxnet3_RxDesc);
1749 rq->rx_ring[i].base = dma_alloc_coherent(
1750 &adapter->pdev->dev, sz,
1751 &rq->rx_ring[i].basePA,
1752 GFP_KERNEL);
1753 if (!rq->rx_ring[i].base) {
1754 netdev_err(adapter->netdev,
1755 "failed to allocate rx ring %d\n", i);
1756 goto err;
1757 }
1758 }
1759
1760 if ((adapter->rxdataring_enabled) && (rq->data_ring.desc_size != 0)) {
1761 sz = rq->rx_ring[0].size * rq->data_ring.desc_size;
1762 rq->data_ring.base =
1763 dma_alloc_coherent(&adapter->pdev->dev, sz,
1764 &rq->data_ring.basePA,
1765 GFP_KERNEL);
1766 if (!rq->data_ring.base) {
1767 netdev_err(adapter->netdev,
1768 "rx data ring will be disabled\n");
1769 adapter->rxdataring_enabled = false;
1770 }
1771 } else {
1772 rq->data_ring.base = NULL;
1773 rq->data_ring.desc_size = 0;
1774 }
1775
1776 sz = rq->comp_ring.size * sizeof(struct Vmxnet3_RxCompDesc);
1777 rq->comp_ring.base = dma_alloc_coherent(&adapter->pdev->dev, sz,
1778 &rq->comp_ring.basePA,
1779 GFP_KERNEL);
1780 if (!rq->comp_ring.base) {
1781 netdev_err(adapter->netdev, "failed to allocate rx comp ring\n");
1782 goto err;
1783 }
1784
1785 sz = sizeof(struct vmxnet3_rx_buf_info) * (rq->rx_ring[0].size +
1786 rq->rx_ring[1].size);
1787 bi = dma_zalloc_coherent(&adapter->pdev->dev, sz, &rq->buf_info_pa,
1788 GFP_KERNEL);
1789 if (!bi)
1790 goto err;
1791
1792 rq->buf_info[0] = bi;
1793 rq->buf_info[1] = bi + rq->rx_ring[0].size;
1794
1795 return 0;
1796
1797 err:
1798 vmxnet3_rq_destroy(rq, adapter);
1799 return -ENOMEM;
1800 }
1801
1802
1803 static int
1804 vmxnet3_rq_create_all(struct vmxnet3_adapter *adapter)
1805 {
1806 int i, err = 0;
1807
1808 adapter->rxdataring_enabled = VMXNET3_VERSION_GE_3(adapter);
1809
1810 for (i = 0; i < adapter->num_rx_queues; i++) {
1811 err = vmxnet3_rq_create(&adapter->rx_queue[i], adapter);
1812 if (unlikely(err)) {
1813 dev_err(&adapter->netdev->dev,
1814 "%s: failed to create rx queue%i\n",
1815 adapter->netdev->name, i);
1816 goto err_out;
1817 }
1818 }
1819
1820 if (!adapter->rxdataring_enabled)
1821 vmxnet3_rq_destroy_all_rxdataring(adapter);
1822
1823 return err;
1824 err_out:
1825 vmxnet3_rq_destroy_all(adapter);
1826 return err;
1827
1828 }
1829
1830 /* Multiple queue aware polling function for tx and rx */
1831
1832 static int
1833 vmxnet3_do_poll(struct vmxnet3_adapter *adapter, int budget)
1834 {
1835 int rcd_done = 0, i;
1836 if (unlikely(adapter->shared->ecr))
1837 vmxnet3_process_events(adapter);
1838 for (i = 0; i < adapter->num_tx_queues; i++)
1839 vmxnet3_tq_tx_complete(&adapter->tx_queue[i], adapter);
1840
1841 for (i = 0; i < adapter->num_rx_queues; i++)
1842 rcd_done += vmxnet3_rq_rx_complete(&adapter->rx_queue[i],
1843 adapter, budget);
1844 return rcd_done;
1845 }
1846
1847
1848 static int
1849 vmxnet3_poll(struct napi_struct *napi, int budget)
1850 {
1851 struct vmxnet3_rx_queue *rx_queue = container_of(napi,
1852 struct vmxnet3_rx_queue, napi);
1853 int rxd_done;
1854
1855 rxd_done = vmxnet3_do_poll(rx_queue->adapter, budget);
1856
1857 if (rxd_done < budget) {
1858 napi_complete_done(napi, rxd_done);
1859 vmxnet3_enable_all_intrs(rx_queue->adapter);
1860 }
1861 return rxd_done;
1862 }
1863
1864 /*
1865 * NAPI polling function for MSI-X mode with multiple Rx queues
1866 * Returns the # of the NAPI credit consumed (# of rx descriptors processed)
1867 */
1868
1869 static int
1870 vmxnet3_poll_rx_only(struct napi_struct *napi, int budget)
1871 {
1872 struct vmxnet3_rx_queue *rq = container_of(napi,
1873 struct vmxnet3_rx_queue, napi);
1874 struct vmxnet3_adapter *adapter = rq->adapter;
1875 int rxd_done;
1876
1877 /* When sharing interrupt with corresponding tx queue, process
1878 * tx completions in that queue as well
1879 */
1880 if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE) {
1881 struct vmxnet3_tx_queue *tq =
1882 &adapter->tx_queue[rq - adapter->rx_queue];
1883 vmxnet3_tq_tx_complete(tq, adapter);
1884 }
1885
1886 rxd_done = vmxnet3_rq_rx_complete(rq, adapter, budget);
1887
1888 if (rxd_done < budget) {
1889 napi_complete_done(napi, rxd_done);
1890 vmxnet3_enable_intr(adapter, rq->comp_ring.intr_idx);
1891 }
1892 return rxd_done;
1893 }
1894
1895
1896 #ifdef CONFIG_PCI_MSI
1897
1898 /*
1899 * Handle completion interrupts on tx queues
1900 * Returns whether or not the intr is handled
1901 */
1902
1903 static irqreturn_t
1904 vmxnet3_msix_tx(int irq, void *data)
1905 {
1906 struct vmxnet3_tx_queue *tq = data;
1907 struct vmxnet3_adapter *adapter = tq->adapter;
1908
1909 if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1910 vmxnet3_disable_intr(adapter, tq->comp_ring.intr_idx);
1911
1912 /* Handle the case where only one irq is allocate for all tx queues */
1913 if (adapter->share_intr == VMXNET3_INTR_TXSHARE) {
1914 int i;
1915 for (i = 0; i < adapter->num_tx_queues; i++) {
1916 struct vmxnet3_tx_queue *txq = &adapter->tx_queue[i];
1917 vmxnet3_tq_tx_complete(txq, adapter);
1918 }
1919 } else {
1920 vmxnet3_tq_tx_complete(tq, adapter);
1921 }
1922 vmxnet3_enable_intr(adapter, tq->comp_ring.intr_idx);
1923
1924 return IRQ_HANDLED;
1925 }
1926
1927
1928 /*
1929 * Handle completion interrupts on rx queues. Returns whether or not the
1930 * intr is handled
1931 */
1932
1933 static irqreturn_t
1934 vmxnet3_msix_rx(int irq, void *data)
1935 {
1936 struct vmxnet3_rx_queue *rq = data;
1937 struct vmxnet3_adapter *adapter = rq->adapter;
1938
1939 /* disable intr if needed */
1940 if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1941 vmxnet3_disable_intr(adapter, rq->comp_ring.intr_idx);
1942 napi_schedule(&rq->napi);
1943
1944 return IRQ_HANDLED;
1945 }
1946
1947 /*
1948 *----------------------------------------------------------------------------
1949 *
1950 * vmxnet3_msix_event --
1951 *
1952 * vmxnet3 msix event intr handler
1953 *
1954 * Result:
1955 * whether or not the intr is handled
1956 *
1957 *----------------------------------------------------------------------------
1958 */
1959
1960 static irqreturn_t
1961 vmxnet3_msix_event(int irq, void *data)
1962 {
1963 struct net_device *dev = data;
1964 struct vmxnet3_adapter *adapter = netdev_priv(dev);
1965
1966 /* disable intr if needed */
1967 if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1968 vmxnet3_disable_intr(adapter, adapter->intr.event_intr_idx);
1969
1970 if (adapter->shared->ecr)
1971 vmxnet3_process_events(adapter);
1972
1973 vmxnet3_enable_intr(adapter, adapter->intr.event_intr_idx);
1974
1975 return IRQ_HANDLED;
1976 }
1977
1978 #endif /* CONFIG_PCI_MSI */
1979
1980
1981 /* Interrupt handler for vmxnet3 */
1982 static irqreturn_t
1983 vmxnet3_intr(int irq, void *dev_id)
1984 {
1985 struct net_device *dev = dev_id;
1986 struct vmxnet3_adapter *adapter = netdev_priv(dev);
1987
1988 if (adapter->intr.type == VMXNET3_IT_INTX) {
1989 u32 icr = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_ICR);
1990 if (unlikely(icr == 0))
1991 /* not ours */
1992 return IRQ_NONE;
1993 }
1994
1995
1996 /* disable intr if needed */
1997 if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1998 vmxnet3_disable_all_intrs(adapter);
1999
2000 napi_schedule(&adapter->rx_queue[0].napi);
2001
2002 return IRQ_HANDLED;
2003 }
2004
2005 #ifdef CONFIG_NET_POLL_CONTROLLER
2006
2007 /* netpoll callback. */
2008 static void
2009 vmxnet3_netpoll(struct net_device *netdev)
2010 {
2011 struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2012
2013 switch (adapter->intr.type) {
2014 #ifdef CONFIG_PCI_MSI
2015 case VMXNET3_IT_MSIX: {
2016 int i;
2017 for (i = 0; i < adapter->num_rx_queues; i++)
2018 vmxnet3_msix_rx(0, &adapter->rx_queue[i]);
2019 break;
2020 }
2021 #endif
2022 case VMXNET3_IT_MSI:
2023 default:
2024 vmxnet3_intr(0, adapter->netdev);
2025 break;
2026 }
2027
2028 }
2029 #endif /* CONFIG_NET_POLL_CONTROLLER */
2030
2031 static int
2032 vmxnet3_request_irqs(struct vmxnet3_adapter *adapter)
2033 {
2034 struct vmxnet3_intr *intr = &adapter->intr;
2035 int err = 0, i;
2036 int vector = 0;
2037
2038 #ifdef CONFIG_PCI_MSI
2039 if (adapter->intr.type == VMXNET3_IT_MSIX) {
2040 for (i = 0; i < adapter->num_tx_queues; i++) {
2041 if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE) {
2042 sprintf(adapter->tx_queue[i].name, "%s-tx-%d",
2043 adapter->netdev->name, vector);
2044 err = request_irq(
2045 intr->msix_entries[vector].vector,
2046 vmxnet3_msix_tx, 0,
2047 adapter->tx_queue[i].name,
2048 &adapter->tx_queue[i]);
2049 } else {
2050 sprintf(adapter->tx_queue[i].name, "%s-rxtx-%d",
2051 adapter->netdev->name, vector);
2052 }
2053 if (err) {
2054 dev_err(&adapter->netdev->dev,
2055 "Failed to request irq for MSIX, %s, "
2056 "error %d\n",
2057 adapter->tx_queue[i].name, err);
2058 return err;
2059 }
2060
2061 /* Handle the case where only 1 MSIx was allocated for
2062 * all tx queues */
2063 if (adapter->share_intr == VMXNET3_INTR_TXSHARE) {
2064 for (; i < adapter->num_tx_queues; i++)
2065 adapter->tx_queue[i].comp_ring.intr_idx
2066 = vector;
2067 vector++;
2068 break;
2069 } else {
2070 adapter->tx_queue[i].comp_ring.intr_idx
2071 = vector++;
2072 }
2073 }
2074 if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE)
2075 vector = 0;
2076
2077 for (i = 0; i < adapter->num_rx_queues; i++) {
2078 if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE)
2079 sprintf(adapter->rx_queue[i].name, "%s-rx-%d",
2080 adapter->netdev->name, vector);
2081 else
2082 sprintf(adapter->rx_queue[i].name, "%s-rxtx-%d",
2083 adapter->netdev->name, vector);
2084 err = request_irq(intr->msix_entries[vector].vector,
2085 vmxnet3_msix_rx, 0,
2086 adapter->rx_queue[i].name,
2087 &(adapter->rx_queue[i]));
2088 if (err) {
2089 netdev_err(adapter->netdev,
2090 "Failed to request irq for MSIX, "
2091 "%s, error %d\n",
2092 adapter->rx_queue[i].name, err);
2093 return err;
2094 }
2095
2096 adapter->rx_queue[i].comp_ring.intr_idx = vector++;
2097 }
2098
2099 sprintf(intr->event_msi_vector_name, "%s-event-%d",
2100 adapter->netdev->name, vector);
2101 err = request_irq(intr->msix_entries[vector].vector,
2102 vmxnet3_msix_event, 0,
2103 intr->event_msi_vector_name, adapter->netdev);
2104 intr->event_intr_idx = vector;
2105
2106 } else if (intr->type == VMXNET3_IT_MSI) {
2107 adapter->num_rx_queues = 1;
2108 err = request_irq(adapter->pdev->irq, vmxnet3_intr, 0,
2109 adapter->netdev->name, adapter->netdev);
2110 } else {
2111 #endif
2112 adapter->num_rx_queues = 1;
2113 err = request_irq(adapter->pdev->irq, vmxnet3_intr,
2114 IRQF_SHARED, adapter->netdev->name,
2115 adapter->netdev);
2116 #ifdef CONFIG_PCI_MSI
2117 }
2118 #endif
2119 intr->num_intrs = vector + 1;
2120 if (err) {
2121 netdev_err(adapter->netdev,
2122 "Failed to request irq (intr type:%d), error %d\n",
2123 intr->type, err);
2124 } else {
2125 /* Number of rx queues will not change after this */
2126 for (i = 0; i < adapter->num_rx_queues; i++) {
2127 struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
2128 rq->qid = i;
2129 rq->qid2 = i + adapter->num_rx_queues;
2130 rq->dataRingQid = i + 2 * adapter->num_rx_queues;
2131 }
2132
2133 /* init our intr settings */
2134 for (i = 0; i < intr->num_intrs; i++)
2135 intr->mod_levels[i] = UPT1_IML_ADAPTIVE;
2136 if (adapter->intr.type != VMXNET3_IT_MSIX) {
2137 adapter->intr.event_intr_idx = 0;
2138 for (i = 0; i < adapter->num_tx_queues; i++)
2139 adapter->tx_queue[i].comp_ring.intr_idx = 0;
2140 adapter->rx_queue[0].comp_ring.intr_idx = 0;
2141 }
2142
2143 netdev_info(adapter->netdev,
2144 "intr type %u, mode %u, %u vectors allocated\n",
2145 intr->type, intr->mask_mode, intr->num_intrs);
2146 }
2147
2148 return err;
2149 }
2150
2151
2152 static void
2153 vmxnet3_free_irqs(struct vmxnet3_adapter *adapter)
2154 {
2155 struct vmxnet3_intr *intr = &adapter->intr;
2156 BUG_ON(intr->type == VMXNET3_IT_AUTO || intr->num_intrs <= 0);
2157
2158 switch (intr->type) {
2159 #ifdef CONFIG_PCI_MSI
2160 case VMXNET3_IT_MSIX:
2161 {
2162 int i, vector = 0;
2163
2164 if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE) {
2165 for (i = 0; i < adapter->num_tx_queues; i++) {
2166 free_irq(intr->msix_entries[vector++].vector,
2167 &(adapter->tx_queue[i]));
2168 if (adapter->share_intr == VMXNET3_INTR_TXSHARE)
2169 break;
2170 }
2171 }
2172
2173 for (i = 0; i < adapter->num_rx_queues; i++) {
2174 free_irq(intr->msix_entries[vector++].vector,
2175 &(adapter->rx_queue[i]));
2176 }
2177
2178 free_irq(intr->msix_entries[vector].vector,
2179 adapter->netdev);
2180 BUG_ON(vector >= intr->num_intrs);
2181 break;
2182 }
2183 #endif
2184 case VMXNET3_IT_MSI:
2185 free_irq(adapter->pdev->irq, adapter->netdev);
2186 break;
2187 case VMXNET3_IT_INTX:
2188 free_irq(adapter->pdev->irq, adapter->netdev);
2189 break;
2190 default:
2191 BUG();
2192 }
2193 }
2194
2195
2196 static void
2197 vmxnet3_restore_vlan(struct vmxnet3_adapter *adapter)
2198 {
2199 u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
2200 u16 vid;
2201
2202 /* allow untagged pkts */
2203 VMXNET3_SET_VFTABLE_ENTRY(vfTable, 0);
2204
2205 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
2206 VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid);
2207 }
2208
2209
2210 static int
2211 vmxnet3_vlan_rx_add_vid(struct net_device *netdev, __be16 proto, u16 vid)
2212 {
2213 struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2214
2215 if (!(netdev->flags & IFF_PROMISC)) {
2216 u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
2217 unsigned long flags;
2218
2219 VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid);
2220 spin_lock_irqsave(&adapter->cmd_lock, flags);
2221 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2222 VMXNET3_CMD_UPDATE_VLAN_FILTERS);
2223 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2224 }
2225
2226 set_bit(vid, adapter->active_vlans);
2227
2228 return 0;
2229 }
2230
2231
2232 static int
2233 vmxnet3_vlan_rx_kill_vid(struct net_device *netdev, __be16 proto, u16 vid)
2234 {
2235 struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2236
2237 if (!(netdev->flags & IFF_PROMISC)) {
2238 u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
2239 unsigned long flags;
2240
2241 VMXNET3_CLEAR_VFTABLE_ENTRY(vfTable, vid);
2242 spin_lock_irqsave(&adapter->cmd_lock, flags);
2243 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2244 VMXNET3_CMD_UPDATE_VLAN_FILTERS);
2245 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2246 }
2247
2248 clear_bit(vid, adapter->active_vlans);
2249
2250 return 0;
2251 }
2252
2253
2254 static u8 *
2255 vmxnet3_copy_mc(struct net_device *netdev)
2256 {
2257 u8 *buf = NULL;
2258 u32 sz = netdev_mc_count(netdev) * ETH_ALEN;
2259
2260 /* struct Vmxnet3_RxFilterConf.mfTableLen is u16. */
2261 if (sz <= 0xffff) {
2262 /* We may be called with BH disabled */
2263 buf = kmalloc(sz, GFP_ATOMIC);
2264 if (buf) {
2265 struct netdev_hw_addr *ha;
2266 int i = 0;
2267
2268 netdev_for_each_mc_addr(ha, netdev)
2269 memcpy(buf + i++ * ETH_ALEN, ha->addr,
2270 ETH_ALEN);
2271 }
2272 }
2273 return buf;
2274 }
2275
2276
2277 static void
2278 vmxnet3_set_mc(struct net_device *netdev)
2279 {
2280 struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2281 unsigned long flags;
2282 struct Vmxnet3_RxFilterConf *rxConf =
2283 &adapter->shared->devRead.rxFilterConf;
2284 u8 *new_table = NULL;
2285 dma_addr_t new_table_pa = 0;
2286 bool new_table_pa_valid = false;
2287 u32 new_mode = VMXNET3_RXM_UCAST;
2288
2289 if (netdev->flags & IFF_PROMISC) {
2290 u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
2291 memset(vfTable, 0, VMXNET3_VFT_SIZE * sizeof(*vfTable));
2292
2293 new_mode |= VMXNET3_RXM_PROMISC;
2294 } else {
2295 vmxnet3_restore_vlan(adapter);
2296 }
2297
2298 if (netdev->flags & IFF_BROADCAST)
2299 new_mode |= VMXNET3_RXM_BCAST;
2300
2301 if (netdev->flags & IFF_ALLMULTI)
2302 new_mode |= VMXNET3_RXM_ALL_MULTI;
2303 else
2304 if (!netdev_mc_empty(netdev)) {
2305 new_table = vmxnet3_copy_mc(netdev);
2306 if (new_table) {
2307 size_t sz = netdev_mc_count(netdev) * ETH_ALEN;
2308
2309 rxConf->mfTableLen = cpu_to_le16(sz);
2310 new_table_pa = dma_map_single(
2311 &adapter->pdev->dev,
2312 new_table,
2313 sz,
2314 PCI_DMA_TODEVICE);
2315 if (!dma_mapping_error(&adapter->pdev->dev,
2316 new_table_pa)) {
2317 new_mode |= VMXNET3_RXM_MCAST;
2318 new_table_pa_valid = true;
2319 rxConf->mfTablePA = cpu_to_le64(
2320 new_table_pa);
2321 }
2322 }
2323 if (!new_table_pa_valid) {
2324 netdev_info(netdev,
2325 "failed to copy mcast list, setting ALL_MULTI\n");
2326 new_mode |= VMXNET3_RXM_ALL_MULTI;
2327 }
2328 }
2329
2330 if (!(new_mode & VMXNET3_RXM_MCAST)) {
2331 rxConf->mfTableLen = 0;
2332 rxConf->mfTablePA = 0;
2333 }
2334
2335 spin_lock_irqsave(&adapter->cmd_lock, flags);
2336 if (new_mode != rxConf->rxMode) {
2337 rxConf->rxMode = cpu_to_le32(new_mode);
2338 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2339 VMXNET3_CMD_UPDATE_RX_MODE);
2340 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2341 VMXNET3_CMD_UPDATE_VLAN_FILTERS);
2342 }
2343
2344 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2345 VMXNET3_CMD_UPDATE_MAC_FILTERS);
2346 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2347
2348 if (new_table_pa_valid)
2349 dma_unmap_single(&adapter->pdev->dev, new_table_pa,
2350 rxConf->mfTableLen, PCI_DMA_TODEVICE);
2351 kfree(new_table);
2352 }
2353
2354 void
2355 vmxnet3_rq_destroy_all(struct vmxnet3_adapter *adapter)
2356 {
2357 int i;
2358
2359 for (i = 0; i < adapter->num_rx_queues; i++)
2360 vmxnet3_rq_destroy(&adapter->rx_queue[i], adapter);
2361 }
2362
2363
2364 /*
2365 * Set up driver_shared based on settings in adapter.
2366 */
2367
2368 static void
2369 vmxnet3_setup_driver_shared(struct vmxnet3_adapter *adapter)
2370 {
2371 struct Vmxnet3_DriverShared *shared = adapter->shared;
2372 struct Vmxnet3_DSDevRead *devRead = &shared->devRead;
2373 struct Vmxnet3_TxQueueConf *tqc;
2374 struct Vmxnet3_RxQueueConf *rqc;
2375 int i;
2376
2377 memset(shared, 0, sizeof(*shared));
2378
2379 /* driver settings */
2380 shared->magic = cpu_to_le32(VMXNET3_REV1_MAGIC);
2381 devRead->misc.driverInfo.version = cpu_to_le32(
2382 VMXNET3_DRIVER_VERSION_NUM);
2383 devRead->misc.driverInfo.gos.gosBits = (sizeof(void *) == 4 ?
2384 VMXNET3_GOS_BITS_32 : VMXNET3_GOS_BITS_64);
2385 devRead->misc.driverInfo.gos.gosType = VMXNET3_GOS_TYPE_LINUX;
2386 *((u32 *)&devRead->misc.driverInfo.gos) = cpu_to_le32(
2387 *((u32 *)&devRead->misc.driverInfo.gos));
2388 devRead->misc.driverInfo.vmxnet3RevSpt = cpu_to_le32(1);
2389 devRead->misc.driverInfo.uptVerSpt = cpu_to_le32(1);
2390
2391 devRead->misc.ddPA = cpu_to_le64(adapter->adapter_pa);
2392 devRead->misc.ddLen = cpu_to_le32(sizeof(struct vmxnet3_adapter));
2393
2394 /* set up feature flags */
2395 if (adapter->netdev->features & NETIF_F_RXCSUM)
2396 devRead->misc.uptFeatures |= UPT1_F_RXCSUM;
2397
2398 if (adapter->netdev->features & NETIF_F_LRO) {
2399 devRead->misc.uptFeatures |= UPT1_F_LRO;
2400 devRead->misc.maxNumRxSG = cpu_to_le16(1 + MAX_SKB_FRAGS);
2401 }
2402 if (adapter->netdev->features & NETIF_F_HW_VLAN_CTAG_RX)
2403 devRead->misc.uptFeatures |= UPT1_F_RXVLAN;
2404
2405 devRead->misc.mtu = cpu_to_le32(adapter->netdev->mtu);
2406 devRead->misc.queueDescPA = cpu_to_le64(adapter->queue_desc_pa);
2407 devRead->misc.queueDescLen = cpu_to_le32(
2408 adapter->num_tx_queues * sizeof(struct Vmxnet3_TxQueueDesc) +
2409 adapter->num_rx_queues * sizeof(struct Vmxnet3_RxQueueDesc));
2410
2411 /* tx queue settings */
2412 devRead->misc.numTxQueues = adapter->num_tx_queues;
2413 for (i = 0; i < adapter->num_tx_queues; i++) {
2414 struct vmxnet3_tx_queue *tq = &adapter->tx_queue[i];
2415 BUG_ON(adapter->tx_queue[i].tx_ring.base == NULL);
2416 tqc = &adapter->tqd_start[i].conf;
2417 tqc->txRingBasePA = cpu_to_le64(tq->tx_ring.basePA);
2418 tqc->dataRingBasePA = cpu_to_le64(tq->data_ring.basePA);
2419 tqc->compRingBasePA = cpu_to_le64(tq->comp_ring.basePA);
2420 tqc->ddPA = cpu_to_le64(tq->buf_info_pa);
2421 tqc->txRingSize = cpu_to_le32(tq->tx_ring.size);
2422 tqc->dataRingSize = cpu_to_le32(tq->data_ring.size);
2423 tqc->txDataRingDescSize = cpu_to_le32(tq->txdata_desc_size);
2424 tqc->compRingSize = cpu_to_le32(tq->comp_ring.size);
2425 tqc->ddLen = cpu_to_le32(
2426 sizeof(struct vmxnet3_tx_buf_info) *
2427 tqc->txRingSize);
2428 tqc->intrIdx = tq->comp_ring.intr_idx;
2429 }
2430
2431 /* rx queue settings */
2432 devRead->misc.numRxQueues = adapter->num_rx_queues;
2433 for (i = 0; i < adapter->num_rx_queues; i++) {
2434 struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
2435 rqc = &adapter->rqd_start[i].conf;
2436 rqc->rxRingBasePA[0] = cpu_to_le64(rq->rx_ring[0].basePA);
2437 rqc->rxRingBasePA[1] = cpu_to_le64(rq->rx_ring[1].basePA);
2438 rqc->compRingBasePA = cpu_to_le64(rq->comp_ring.basePA);
2439 rqc->ddPA = cpu_to_le64(rq->buf_info_pa);
2440 rqc->rxRingSize[0] = cpu_to_le32(rq->rx_ring[0].size);
2441 rqc->rxRingSize[1] = cpu_to_le32(rq->rx_ring[1].size);
2442 rqc->compRingSize = cpu_to_le32(rq->comp_ring.size);
2443 rqc->ddLen = cpu_to_le32(
2444 sizeof(struct vmxnet3_rx_buf_info) *
2445 (rqc->rxRingSize[0] +
2446 rqc->rxRingSize[1]));
2447 rqc->intrIdx = rq->comp_ring.intr_idx;
2448 if (VMXNET3_VERSION_GE_3(adapter)) {
2449 rqc->rxDataRingBasePA =
2450 cpu_to_le64(rq->data_ring.basePA);
2451 rqc->rxDataRingDescSize =
2452 cpu_to_le16(rq->data_ring.desc_size);
2453 }
2454 }
2455
2456 #ifdef VMXNET3_RSS
2457 memset(adapter->rss_conf, 0, sizeof(*adapter->rss_conf));
2458
2459 if (adapter->rss) {
2460 struct UPT1_RSSConf *rssConf = adapter->rss_conf;
2461
2462 devRead->misc.uptFeatures |= UPT1_F_RSS;
2463 devRead->misc.numRxQueues = adapter->num_rx_queues;
2464 rssConf->hashType = UPT1_RSS_HASH_TYPE_TCP_IPV4 |
2465 UPT1_RSS_HASH_TYPE_IPV4 |
2466 UPT1_RSS_HASH_TYPE_TCP_IPV6 |
2467 UPT1_RSS_HASH_TYPE_IPV6;
2468 rssConf->hashFunc = UPT1_RSS_HASH_FUNC_TOEPLITZ;
2469 rssConf->hashKeySize = UPT1_RSS_MAX_KEY_SIZE;
2470 rssConf->indTableSize = VMXNET3_RSS_IND_TABLE_SIZE;
2471 netdev_rss_key_fill(rssConf->hashKey, sizeof(rssConf->hashKey));
2472
2473 for (i = 0; i < rssConf->indTableSize; i++)
2474 rssConf->indTable[i] = ethtool_rxfh_indir_default(
2475 i, adapter->num_rx_queues);
2476
2477 devRead->rssConfDesc.confVer = 1;
2478 devRead->rssConfDesc.confLen = cpu_to_le32(sizeof(*rssConf));
2479 devRead->rssConfDesc.confPA =
2480 cpu_to_le64(adapter->rss_conf_pa);
2481 }
2482
2483 #endif /* VMXNET3_RSS */
2484
2485 /* intr settings */
2486 devRead->intrConf.autoMask = adapter->intr.mask_mode ==
2487 VMXNET3_IMM_AUTO;
2488 devRead->intrConf.numIntrs = adapter->intr.num_intrs;
2489 for (i = 0; i < adapter->intr.num_intrs; i++)
2490 devRead->intrConf.modLevels[i] = adapter->intr.mod_levels[i];
2491
2492 devRead->intrConf.eventIntrIdx = adapter->intr.event_intr_idx;
2493 devRead->intrConf.intrCtrl |= cpu_to_le32(VMXNET3_IC_DISABLE_ALL);
2494
2495 /* rx filter settings */
2496 devRead->rxFilterConf.rxMode = 0;
2497 vmxnet3_restore_vlan(adapter);
2498 vmxnet3_write_mac_addr(adapter, adapter->netdev->dev_addr);
2499
2500 /* the rest are already zeroed */
2501 }
2502
2503 static void
2504 vmxnet3_init_coalesce(struct vmxnet3_adapter *adapter)
2505 {
2506 struct Vmxnet3_DriverShared *shared = adapter->shared;
2507 union Vmxnet3_CmdInfo *cmdInfo = &shared->cu.cmdInfo;
2508 unsigned long flags;
2509
2510 if (!VMXNET3_VERSION_GE_3(adapter))
2511 return;
2512
2513 spin_lock_irqsave(&adapter->cmd_lock, flags);
2514 cmdInfo->varConf.confVer = 1;
2515 cmdInfo->varConf.confLen =
2516 cpu_to_le32(sizeof(*adapter->coal_conf));
2517 cmdInfo->varConf.confPA = cpu_to_le64(adapter->coal_conf_pa);
2518
2519 if (adapter->default_coal_mode) {
2520 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2521 VMXNET3_CMD_GET_COALESCE);
2522 } else {
2523 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2524 VMXNET3_CMD_SET_COALESCE);
2525 }
2526
2527 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2528 }
2529
2530 int
2531 vmxnet3_activate_dev(struct vmxnet3_adapter *adapter)
2532 {
2533 int err, i;
2534 u32 ret;
2535 unsigned long flags;
2536
2537 netdev_dbg(adapter->netdev, "%s: skb_buf_size %d, rx_buf_per_pkt %d,"
2538 " ring sizes %u %u %u\n", adapter->netdev->name,
2539 adapter->skb_buf_size, adapter->rx_buf_per_pkt,
2540 adapter->tx_queue[0].tx_ring.size,
2541 adapter->rx_queue[0].rx_ring[0].size,
2542 adapter->rx_queue[0].rx_ring[1].size);
2543
2544 vmxnet3_tq_init_all(adapter);
2545 err = vmxnet3_rq_init_all(adapter);
2546 if (err) {
2547 netdev_err(adapter->netdev,
2548 "Failed to init rx queue error %d\n", err);
2549 goto rq_err;
2550 }
2551
2552 err = vmxnet3_request_irqs(adapter);
2553 if (err) {
2554 netdev_err(adapter->netdev,
2555 "Failed to setup irq for error %d\n", err);
2556 goto irq_err;
2557 }
2558
2559 vmxnet3_setup_driver_shared(adapter);
2560
2561 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, VMXNET3_GET_ADDR_LO(
2562 adapter->shared_pa));
2563 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, VMXNET3_GET_ADDR_HI(
2564 adapter->shared_pa));
2565 spin_lock_irqsave(&adapter->cmd_lock, flags);
2566 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2567 VMXNET3_CMD_ACTIVATE_DEV);
2568 ret = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
2569 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2570
2571 if (ret != 0) {
2572 netdev_err(adapter->netdev,
2573 "Failed to activate dev: error %u\n", ret);
2574 err = -EINVAL;
2575 goto activate_err;
2576 }
2577
2578 vmxnet3_init_coalesce(adapter);
2579
2580 for (i = 0; i < adapter->num_rx_queues; i++) {
2581 VMXNET3_WRITE_BAR0_REG(adapter,
2582 VMXNET3_REG_RXPROD + i * VMXNET3_REG_ALIGN,
2583 adapter->rx_queue[i].rx_ring[0].next2fill);
2584 VMXNET3_WRITE_BAR0_REG(adapter, (VMXNET3_REG_RXPROD2 +
2585 (i * VMXNET3_REG_ALIGN)),
2586 adapter->rx_queue[i].rx_ring[1].next2fill);
2587 }
2588
2589 /* Apply the rx filter settins last. */
2590 vmxnet3_set_mc(adapter->netdev);
2591
2592 /*
2593 * Check link state when first activating device. It will start the
2594 * tx queue if the link is up.
2595 */
2596 vmxnet3_check_link(adapter, true);
2597 for (i = 0; i < adapter->num_rx_queues; i++)
2598 napi_enable(&adapter->rx_queue[i].napi);
2599 vmxnet3_enable_all_intrs(adapter);
2600 clear_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
2601 return 0;
2602
2603 activate_err:
2604 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, 0);
2605 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, 0);
2606 vmxnet3_free_irqs(adapter);
2607 irq_err:
2608 rq_err:
2609 /* free up buffers we allocated */
2610 vmxnet3_rq_cleanup_all(adapter);
2611 return err;
2612 }
2613
2614
2615 void
2616 vmxnet3_reset_dev(struct vmxnet3_adapter *adapter)
2617 {
2618 unsigned long flags;
2619 spin_lock_irqsave(&adapter->cmd_lock, flags);
2620 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, VMXNET3_CMD_RESET_DEV);
2621 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2622 }
2623
2624
2625 int
2626 vmxnet3_quiesce_dev(struct vmxnet3_adapter *adapter)
2627 {
2628 int i;
2629 unsigned long flags;
2630 if (test_and_set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state))
2631 return 0;
2632
2633
2634 spin_lock_irqsave(&adapter->cmd_lock, flags);
2635 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2636 VMXNET3_CMD_QUIESCE_DEV);
2637 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2638 vmxnet3_disable_all_intrs(adapter);
2639
2640 for (i = 0; i < adapter->num_rx_queues; i++)
2641 napi_disable(&adapter->rx_queue[i].napi);
2642 netif_tx_disable(adapter->netdev);
2643 adapter->link_speed = 0;
2644 netif_carrier_off(adapter->netdev);
2645
2646 vmxnet3_tq_cleanup_all(adapter);
2647 vmxnet3_rq_cleanup_all(adapter);
2648 vmxnet3_free_irqs(adapter);
2649 return 0;
2650 }
2651
2652
2653 static void
2654 vmxnet3_write_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac)
2655 {
2656 u32 tmp;
2657
2658 tmp = *(u32 *)mac;
2659 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACL, tmp);
2660
2661 tmp = (mac[5] << 8) | mac[4];
2662 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACH, tmp);
2663 }
2664
2665
2666 static int
2667 vmxnet3_set_mac_addr(struct net_device *netdev, void *p)
2668 {
2669 struct sockaddr *addr = p;
2670 struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2671
2672 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2673 vmxnet3_write_mac_addr(adapter, addr->sa_data);
2674
2675 return 0;
2676 }
2677
2678
2679 /* ==================== initialization and cleanup routines ============ */
2680
2681 static int
2682 vmxnet3_alloc_pci_resources(struct vmxnet3_adapter *adapter, bool *dma64)
2683 {
2684 int err;
2685 unsigned long mmio_start, mmio_len;
2686 struct pci_dev *pdev = adapter->pdev;
2687
2688 err = pci_enable_device(pdev);
2689 if (err) {
2690 dev_err(&pdev->dev, "Failed to enable adapter: error %d\n", err);
2691 return err;
2692 }
2693
2694 if (pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) == 0) {
2695 if (pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64)) != 0) {
2696 dev_err(&pdev->dev,
2697 "pci_set_consistent_dma_mask failed\n");
2698 err = -EIO;
2699 goto err_set_mask;
2700 }
2701 *dma64 = true;
2702 } else {
2703 if (pci_set_dma_mask(pdev, DMA_BIT_MASK(32)) != 0) {
2704 dev_err(&pdev->dev,
2705 "pci_set_dma_mask failed\n");
2706 err = -EIO;
2707 goto err_set_mask;
2708 }
2709 *dma64 = false;
2710 }
2711
2712 err = pci_request_selected_regions(pdev, (1 << 2) - 1,
2713 vmxnet3_driver_name);
2714 if (err) {
2715 dev_err(&pdev->dev,
2716 "Failed to request region for adapter: error %d\n", err);
2717 goto err_set_mask;
2718 }
2719
2720 pci_set_master(pdev);
2721
2722 mmio_start = pci_resource_start(pdev, 0);
2723 mmio_len = pci_resource_len(pdev, 0);
2724 adapter->hw_addr0 = ioremap(mmio_start, mmio_len);
2725 if (!adapter->hw_addr0) {
2726 dev_err(&pdev->dev, "Failed to map bar0\n");
2727 err = -EIO;
2728 goto err_ioremap;
2729 }
2730
2731 mmio_start = pci_resource_start(pdev, 1);
2732 mmio_len = pci_resource_len(pdev, 1);
2733 adapter->hw_addr1 = ioremap(mmio_start, mmio_len);
2734 if (!adapter->hw_addr1) {
2735 dev_err(&pdev->dev, "Failed to map bar1\n");
2736 err = -EIO;
2737 goto err_bar1;
2738 }
2739 return 0;
2740
2741 err_bar1:
2742 iounmap(adapter->hw_addr0);
2743 err_ioremap:
2744 pci_release_selected_regions(pdev, (1 << 2) - 1);
2745 err_set_mask:
2746 pci_disable_device(pdev);
2747 return err;
2748 }
2749
2750
2751 static void
2752 vmxnet3_free_pci_resources(struct vmxnet3_adapter *adapter)
2753 {
2754 BUG_ON(!adapter->pdev);
2755
2756 iounmap(adapter->hw_addr0);
2757 iounmap(adapter->hw_addr1);
2758 pci_release_selected_regions(adapter->pdev, (1 << 2) - 1);
2759 pci_disable_device(adapter->pdev);
2760 }
2761
2762
2763 static void
2764 vmxnet3_adjust_rx_ring_size(struct vmxnet3_adapter *adapter)
2765 {
2766 size_t sz, i, ring0_size, ring1_size, comp_size;
2767 if (adapter->netdev->mtu <= VMXNET3_MAX_SKB_BUF_SIZE -
2768 VMXNET3_MAX_ETH_HDR_SIZE) {
2769 adapter->skb_buf_size = adapter->netdev->mtu +
2770 VMXNET3_MAX_ETH_HDR_SIZE;
2771 if (adapter->skb_buf_size < VMXNET3_MIN_T0_BUF_SIZE)
2772 adapter->skb_buf_size = VMXNET3_MIN_T0_BUF_SIZE;
2773
2774 adapter->rx_buf_per_pkt = 1;
2775 } else {
2776 adapter->skb_buf_size = VMXNET3_MAX_SKB_BUF_SIZE;
2777 sz = adapter->netdev->mtu - VMXNET3_MAX_SKB_BUF_SIZE +
2778 VMXNET3_MAX_ETH_HDR_SIZE;
2779 adapter->rx_buf_per_pkt = 1 + (sz + PAGE_SIZE - 1) / PAGE_SIZE;
2780 }
2781
2782 /*
2783 * for simplicity, force the ring0 size to be a multiple of
2784 * rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN
2785 */
2786 sz = adapter->rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN;
2787 ring0_size = adapter->rx_queue[0].rx_ring[0].size;
2788 ring0_size = (ring0_size + sz - 1) / sz * sz;
2789 ring0_size = min_t(u32, ring0_size, VMXNET3_RX_RING_MAX_SIZE /
2790 sz * sz);
2791 ring1_size = adapter->rx_queue[0].rx_ring[1].size;
2792 ring1_size = (ring1_size + sz - 1) / sz * sz;
2793 ring1_size = min_t(u32, ring1_size, VMXNET3_RX_RING2_MAX_SIZE /
2794 sz * sz);
2795 comp_size = ring0_size + ring1_size;
2796
2797 for (i = 0; i < adapter->num_rx_queues; i++) {
2798 struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
2799
2800 rq->rx_ring[0].size = ring0_size;
2801 rq->rx_ring[1].size = ring1_size;
2802 rq->comp_ring.size = comp_size;
2803 }
2804 }
2805
2806
2807 int
2808 vmxnet3_create_queues(struct vmxnet3_adapter *adapter, u32 tx_ring_size,
2809 u32 rx_ring_size, u32 rx_ring2_size,
2810 u16 txdata_desc_size, u16 rxdata_desc_size)
2811 {
2812 int err = 0, i;
2813
2814 for (i = 0; i < adapter->num_tx_queues; i++) {
2815 struct vmxnet3_tx_queue *tq = &adapter->tx_queue[i];
2816 tq->tx_ring.size = tx_ring_size;
2817 tq->data_ring.size = tx_ring_size;
2818 tq->comp_ring.size = tx_ring_size;
2819 tq->txdata_desc_size = txdata_desc_size;
2820 tq->shared = &adapter->tqd_start[i].ctrl;
2821 tq->stopped = true;
2822 tq->adapter = adapter;
2823 tq->qid = i;
2824 err = vmxnet3_tq_create(tq, adapter);
2825 /*
2826 * Too late to change num_tx_queues. We cannot do away with
2827 * lesser number of queues than what we asked for
2828 */
2829 if (err)
2830 goto queue_err;
2831 }
2832
2833 adapter->rx_queue[0].rx_ring[0].size = rx_ring_size;
2834 adapter->rx_queue[0].rx_ring[1].size = rx_ring2_size;
2835 vmxnet3_adjust_rx_ring_size(adapter);
2836
2837 adapter->rxdataring_enabled = VMXNET3_VERSION_GE_3(adapter);
2838 for (i = 0; i < adapter->num_rx_queues; i++) {
2839 struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
2840 /* qid and qid2 for rx queues will be assigned later when num
2841 * of rx queues is finalized after allocating intrs */
2842 rq->shared = &adapter->rqd_start[i].ctrl;
2843 rq->adapter = adapter;
2844 rq->data_ring.desc_size = rxdata_desc_size;
2845 err = vmxnet3_rq_create(rq, adapter);
2846 if (err) {
2847 if (i == 0) {
2848 netdev_err(adapter->netdev,
2849 "Could not allocate any rx queues. "
2850 "Aborting.\n");
2851 goto queue_err;
2852 } else {
2853 netdev_info(adapter->netdev,
2854 "Number of rx queues changed "
2855 "to : %d.\n", i);
2856 adapter->num_rx_queues = i;
2857 err = 0;
2858 break;
2859 }
2860 }
2861 }
2862
2863 if (!adapter->rxdataring_enabled)
2864 vmxnet3_rq_destroy_all_rxdataring(adapter);
2865
2866 return err;
2867 queue_err:
2868 vmxnet3_tq_destroy_all(adapter);
2869 return err;
2870 }
2871
2872 static int
2873 vmxnet3_open(struct net_device *netdev)
2874 {
2875 struct vmxnet3_adapter *adapter;
2876 int err, i;
2877
2878 adapter = netdev_priv(netdev);
2879
2880 for (i = 0; i < adapter->num_tx_queues; i++)
2881 spin_lock_init(&adapter->tx_queue[i].tx_lock);
2882
2883 if (VMXNET3_VERSION_GE_3(adapter)) {
2884 unsigned long flags;
2885 u16 txdata_desc_size;
2886
2887 spin_lock_irqsave(&adapter->cmd_lock, flags);
2888 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2889 VMXNET3_CMD_GET_TXDATA_DESC_SIZE);
2890 txdata_desc_size = VMXNET3_READ_BAR1_REG(adapter,
2891 VMXNET3_REG_CMD);
2892 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2893
2894 if ((txdata_desc_size < VMXNET3_TXDATA_DESC_MIN_SIZE) ||
2895 (txdata_desc_size > VMXNET3_TXDATA_DESC_MAX_SIZE) ||
2896 (txdata_desc_size & VMXNET3_TXDATA_DESC_SIZE_MASK)) {
2897 adapter->txdata_desc_size =
2898 sizeof(struct Vmxnet3_TxDataDesc);
2899 } else {
2900 adapter->txdata_desc_size = txdata_desc_size;
2901 }
2902 } else {
2903 adapter->txdata_desc_size = sizeof(struct Vmxnet3_TxDataDesc);
2904 }
2905
2906 err = vmxnet3_create_queues(adapter,
2907 adapter->tx_ring_size,
2908 adapter->rx_ring_size,
2909 adapter->rx_ring2_size,
2910 adapter->txdata_desc_size,
2911 adapter->rxdata_desc_size);
2912 if (err)
2913 goto queue_err;
2914
2915 err = vmxnet3_activate_dev(adapter);
2916 if (err)
2917 goto activate_err;
2918
2919 return 0;
2920
2921 activate_err:
2922 vmxnet3_rq_destroy_all(adapter);
2923 vmxnet3_tq_destroy_all(adapter);
2924 queue_err:
2925 return err;
2926 }
2927
2928
2929 static int
2930 vmxnet3_close(struct net_device *netdev)
2931 {
2932 struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2933
2934 /*
2935 * Reset_work may be in the middle of resetting the device, wait for its
2936 * completion.
2937 */
2938 while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2939 msleep(1);
2940
2941 vmxnet3_quiesce_dev(adapter);
2942
2943 vmxnet3_rq_destroy_all(adapter);
2944 vmxnet3_tq_destroy_all(adapter);
2945
2946 clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2947
2948
2949 return 0;
2950 }
2951
2952
2953 void
2954 vmxnet3_force_close(struct vmxnet3_adapter *adapter)
2955 {
2956 int i;
2957
2958 /*
2959 * we must clear VMXNET3_STATE_BIT_RESETTING, otherwise
2960 * vmxnet3_close() will deadlock.
2961 */
2962 BUG_ON(test_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state));
2963
2964 /* we need to enable NAPI, otherwise dev_close will deadlock */
2965 for (i = 0; i < adapter->num_rx_queues; i++)
2966 napi_enable(&adapter->rx_queue[i].napi);
2967 /*
2968 * Need to clear the quiesce bit to ensure that vmxnet3_close
2969 * can quiesce the device properly
2970 */
2971 clear_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
2972 dev_close(adapter->netdev);
2973 }
2974
2975
2976 static int
2977 vmxnet3_change_mtu(struct net_device *netdev, int new_mtu)
2978 {
2979 struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2980 int err = 0;
2981
2982 netdev->mtu = new_mtu;
2983
2984 /*
2985 * Reset_work may be in the middle of resetting the device, wait for its
2986 * completion.
2987 */
2988 while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2989 msleep(1);
2990
2991 if (netif_running(netdev)) {
2992 vmxnet3_quiesce_dev(adapter);
2993 vmxnet3_reset_dev(adapter);
2994
2995 /* we need to re-create the rx queue based on the new mtu */
2996 vmxnet3_rq_destroy_all(adapter);
2997 vmxnet3_adjust_rx_ring_size(adapter);
2998 err = vmxnet3_rq_create_all(adapter);
2999 if (err) {
3000 netdev_err(netdev,
3001 "failed to re-create rx queues, "
3002 " error %d. Closing it.\n", err);
3003 goto out;
3004 }
3005
3006 err = vmxnet3_activate_dev(adapter);
3007 if (err) {
3008 netdev_err(netdev,
3009 "failed to re-activate, error %d. "
3010 "Closing it\n", err);
3011 goto out;
3012 }
3013 }
3014
3015 out:
3016 clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
3017 if (err)
3018 vmxnet3_force_close(adapter);
3019
3020 return err;
3021 }
3022
3023
3024 static void
3025 vmxnet3_declare_features(struct vmxnet3_adapter *adapter, bool dma64)
3026 {
3027 struct net_device *netdev = adapter->netdev;
3028
3029 netdev->hw_features = NETIF_F_SG | NETIF_F_RXCSUM |
3030 NETIF_F_HW_CSUM | NETIF_F_HW_VLAN_CTAG_TX |
3031 NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_TSO | NETIF_F_TSO6 |
3032 NETIF_F_LRO;
3033 if (dma64)
3034 netdev->hw_features |= NETIF_F_HIGHDMA;
3035 netdev->vlan_features = netdev->hw_features &
3036 ~(NETIF_F_HW_VLAN_CTAG_TX |
3037 NETIF_F_HW_VLAN_CTAG_RX);
3038 netdev->features = netdev->hw_features | NETIF_F_HW_VLAN_CTAG_FILTER;
3039 }
3040
3041
3042 static void
3043 vmxnet3_read_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac)
3044 {
3045 u32 tmp;
3046
3047 tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACL);
3048 *(u32 *)mac = tmp;
3049
3050 tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACH);
3051 mac[4] = tmp & 0xff;
3052 mac[5] = (tmp >> 8) & 0xff;
3053 }
3054
3055 #ifdef CONFIG_PCI_MSI
3056
3057 /*
3058 * Enable MSIx vectors.
3059 * Returns :
3060 * VMXNET3_LINUX_MIN_MSIX_VECT when only minimum number of vectors required
3061 * were enabled.
3062 * number of vectors which were enabled otherwise (this number is greater
3063 * than VMXNET3_LINUX_MIN_MSIX_VECT)
3064 */
3065
3066 static int
3067 vmxnet3_acquire_msix_vectors(struct vmxnet3_adapter *adapter, int nvec)
3068 {
3069 int ret = pci_enable_msix_range(adapter->pdev,
3070 adapter->intr.msix_entries, nvec, nvec);
3071
3072 if (ret == -ENOSPC && nvec > VMXNET3_LINUX_MIN_MSIX_VECT) {
3073 dev_err(&adapter->netdev->dev,
3074 "Failed to enable %d MSI-X, trying %d\n",
3075 nvec, VMXNET3_LINUX_MIN_MSIX_VECT);
3076
3077 ret = pci_enable_msix_range(adapter->pdev,
3078 adapter->intr.msix_entries,
3079 VMXNET3_LINUX_MIN_MSIX_VECT,
3080 VMXNET3_LINUX_MIN_MSIX_VECT);
3081 }
3082
3083 if (ret < 0) {
3084 dev_err(&adapter->netdev->dev,
3085 "Failed to enable MSI-X, error: %d\n", ret);
3086 }
3087
3088 return ret;
3089 }
3090
3091
3092 #endif /* CONFIG_PCI_MSI */
3093
3094 static void
3095 vmxnet3_alloc_intr_resources(struct vmxnet3_adapter *adapter)
3096 {
3097 u32 cfg;
3098 unsigned long flags;
3099
3100 /* intr settings */
3101 spin_lock_irqsave(&adapter->cmd_lock, flags);
3102 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3103 VMXNET3_CMD_GET_CONF_INTR);
3104 cfg = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
3105 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3106 adapter->intr.type = cfg & 0x3;
3107 adapter->intr.mask_mode = (cfg >> 2) & 0x3;
3108
3109 if (adapter->intr.type == VMXNET3_IT_AUTO) {
3110 adapter->intr.type = VMXNET3_IT_MSIX;
3111 }
3112
3113 #ifdef CONFIG_PCI_MSI
3114 if (adapter->intr.type == VMXNET3_IT_MSIX) {
3115 int i, nvec;
3116
3117 nvec = adapter->share_intr == VMXNET3_INTR_TXSHARE ?
3118 1 : adapter->num_tx_queues;
3119 nvec += adapter->share_intr == VMXNET3_INTR_BUDDYSHARE ?
3120 0 : adapter->num_rx_queues;
3121 nvec += 1; /* for link event */
3122 nvec = nvec > VMXNET3_LINUX_MIN_MSIX_VECT ?
3123 nvec : VMXNET3_LINUX_MIN_MSIX_VECT;
3124
3125 for (i = 0; i < nvec; i++)
3126 adapter->intr.msix_entries[i].entry = i;
3127
3128 nvec = vmxnet3_acquire_msix_vectors(adapter, nvec);
3129 if (nvec < 0)
3130 goto msix_err;
3131
3132 /* If we cannot allocate one MSIx vector per queue
3133 * then limit the number of rx queues to 1
3134 */
3135 if (nvec == VMXNET3_LINUX_MIN_MSIX_VECT) {
3136 if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE
3137 || adapter->num_rx_queues != 1) {
3138 adapter->share_intr = VMXNET3_INTR_TXSHARE;
3139 netdev_err(adapter->netdev,
3140 "Number of rx queues : 1\n");
3141 adapter->num_rx_queues = 1;
3142 }
3143 }
3144
3145 adapter->intr.num_intrs = nvec;
3146 return;
3147
3148 msix_err:
3149 /* If we cannot allocate MSIx vectors use only one rx queue */
3150 dev_info(&adapter->pdev->dev,
3151 "Failed to enable MSI-X, error %d. "
3152 "Limiting #rx queues to 1, try MSI.\n", nvec);
3153
3154 adapter->intr.type = VMXNET3_IT_MSI;
3155 }
3156
3157 if (adapter->intr.type == VMXNET3_IT_MSI) {
3158 if (!pci_enable_msi(adapter->pdev)) {
3159 adapter->num_rx_queues = 1;
3160 adapter->intr.num_intrs = 1;
3161 return;
3162 }
3163 }
3164 #endif /* CONFIG_PCI_MSI */
3165
3166 adapter->num_rx_queues = 1;
3167 dev_info(&adapter->netdev->dev,
3168 "Using INTx interrupt, #Rx queues: 1.\n");
3169 adapter->intr.type = VMXNET3_IT_INTX;
3170
3171 /* INT-X related setting */
3172 adapter->intr.num_intrs = 1;
3173 }
3174
3175
3176 static void
3177 vmxnet3_free_intr_resources(struct vmxnet3_adapter *adapter)
3178 {
3179 if (adapter->intr.type == VMXNET3_IT_MSIX)
3180 pci_disable_msix(adapter->pdev);
3181 else if (adapter->intr.type == VMXNET3_IT_MSI)
3182 pci_disable_msi(adapter->pdev);
3183 else
3184 BUG_ON(adapter->intr.type != VMXNET3_IT_INTX);
3185 }
3186
3187
3188 static void
3189 vmxnet3_tx_timeout(struct net_device *netdev)
3190 {
3191 struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3192 adapter->tx_timeout_count++;
3193
3194 netdev_err(adapter->netdev, "tx hang\n");
3195 schedule_work(&adapter->work);
3196 }
3197
3198
3199 static void
3200 vmxnet3_reset_work(struct work_struct *data)
3201 {
3202 struct vmxnet3_adapter *adapter;
3203
3204 adapter = container_of(data, struct vmxnet3_adapter, work);
3205
3206 /* if another thread is resetting the device, no need to proceed */
3207 if (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
3208 return;
3209
3210 /* if the device is closed, we must leave it alone */
3211 rtnl_lock();
3212 if (netif_running(adapter->netdev)) {
3213 netdev_notice(adapter->netdev, "resetting\n");
3214 vmxnet3_quiesce_dev(adapter);
3215 vmxnet3_reset_dev(adapter);
3216 vmxnet3_activate_dev(adapter);
3217 } else {
3218 netdev_info(adapter->netdev, "already closed\n");
3219 }
3220 rtnl_unlock();
3221
3222 netif_wake_queue(adapter->netdev);
3223 clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
3224 }
3225
3226
3227 static int
3228 vmxnet3_probe_device(struct pci_dev *pdev,
3229 const struct pci_device_id *id)
3230 {
3231 static const struct net_device_ops vmxnet3_netdev_ops = {
3232 .ndo_open = vmxnet3_open,
3233 .ndo_stop = vmxnet3_close,
3234 .ndo_start_xmit = vmxnet3_xmit_frame,
3235 .ndo_set_mac_address = vmxnet3_set_mac_addr,
3236 .ndo_change_mtu = vmxnet3_change_mtu,
3237 .ndo_set_features = vmxnet3_set_features,
3238 .ndo_get_stats64 = vmxnet3_get_stats64,
3239 .ndo_tx_timeout = vmxnet3_tx_timeout,
3240 .ndo_set_rx_mode = vmxnet3_set_mc,
3241 .ndo_vlan_rx_add_vid = vmxnet3_vlan_rx_add_vid,
3242 .ndo_vlan_rx_kill_vid = vmxnet3_vlan_rx_kill_vid,
3243 #ifdef CONFIG_NET_POLL_CONTROLLER
3244 .ndo_poll_controller = vmxnet3_netpoll,
3245 #endif
3246 };
3247 int err;
3248 bool dma64 = false; /* stupid gcc */
3249 u32 ver;
3250 struct net_device *netdev;
3251 struct vmxnet3_adapter *adapter;
3252 u8 mac[ETH_ALEN];
3253 int size;
3254 int num_tx_queues;
3255 int num_rx_queues;
3256
3257 if (!pci_msi_enabled())
3258 enable_mq = 0;
3259
3260 #ifdef VMXNET3_RSS
3261 if (enable_mq)
3262 num_rx_queues = min(VMXNET3_DEVICE_MAX_RX_QUEUES,
3263 (int)num_online_cpus());
3264 else
3265 #endif
3266 num_rx_queues = 1;
3267 num_rx_queues = rounddown_pow_of_two(num_rx_queues);
3268
3269 if (enable_mq)
3270 num_tx_queues = min(VMXNET3_DEVICE_MAX_TX_QUEUES,
3271 (int)num_online_cpus());
3272 else
3273 num_tx_queues = 1;
3274
3275 num_tx_queues = rounddown_pow_of_two(num_tx_queues);
3276 netdev = alloc_etherdev_mq(sizeof(struct vmxnet3_adapter),
3277 max(num_tx_queues, num_rx_queues));
3278 dev_info(&pdev->dev,
3279 "# of Tx queues : %d, # of Rx queues : %d\n",
3280 num_tx_queues, num_rx_queues);
3281
3282 if (!netdev)
3283 return -ENOMEM;
3284
3285 pci_set_drvdata(pdev, netdev);
3286 adapter = netdev_priv(netdev);
3287 adapter->netdev = netdev;
3288 adapter->pdev = pdev;
3289
3290 adapter->tx_ring_size = VMXNET3_DEF_TX_RING_SIZE;
3291 adapter->rx_ring_size = VMXNET3_DEF_RX_RING_SIZE;
3292 adapter->rx_ring2_size = VMXNET3_DEF_RX_RING2_SIZE;
3293
3294 spin_lock_init(&adapter->cmd_lock);
3295 adapter->adapter_pa = dma_map_single(&adapter->pdev->dev, adapter,
3296 sizeof(struct vmxnet3_adapter),
3297 PCI_DMA_TODEVICE);
3298 if (dma_mapping_error(&adapter->pdev->dev, adapter->adapter_pa)) {
3299 dev_err(&pdev->dev, "Failed to map dma\n");
3300 err = -EFAULT;
3301 goto err_dma_map;
3302 }
3303 adapter->shared = dma_alloc_coherent(
3304 &adapter->pdev->dev,
3305 sizeof(struct Vmxnet3_DriverShared),
3306 &adapter->shared_pa, GFP_KERNEL);
3307 if (!adapter->shared) {
3308 dev_err(&pdev->dev, "Failed to allocate memory\n");
3309 err = -ENOMEM;
3310 goto err_alloc_shared;
3311 }
3312
3313 adapter->num_rx_queues = num_rx_queues;
3314 adapter->num_tx_queues = num_tx_queues;
3315 adapter->rx_buf_per_pkt = 1;
3316
3317 size = sizeof(struct Vmxnet3_TxQueueDesc) * adapter->num_tx_queues;
3318 size += sizeof(struct Vmxnet3_RxQueueDesc) * adapter->num_rx_queues;
3319 adapter->tqd_start = dma_alloc_coherent(&adapter->pdev->dev, size,
3320 &adapter->queue_desc_pa,
3321 GFP_KERNEL);
3322
3323 if (!adapter->tqd_start) {
3324 dev_err(&pdev->dev, "Failed to allocate memory\n");
3325 err = -ENOMEM;
3326 goto err_alloc_queue_desc;
3327 }
3328 adapter->rqd_start = (struct Vmxnet3_RxQueueDesc *)(adapter->tqd_start +
3329 adapter->num_tx_queues);
3330
3331 adapter->pm_conf = dma_alloc_coherent(&adapter->pdev->dev,
3332 sizeof(struct Vmxnet3_PMConf),
3333 &adapter->pm_conf_pa,
3334 GFP_KERNEL);
3335 if (adapter->pm_conf == NULL) {
3336 err = -ENOMEM;
3337 goto err_alloc_pm;
3338 }
3339
3340 #ifdef VMXNET3_RSS
3341
3342 adapter->rss_conf = dma_alloc_coherent(&adapter->pdev->dev,
3343 sizeof(struct UPT1_RSSConf),
3344 &adapter->rss_conf_pa,
3345 GFP_KERNEL);
3346 if (adapter->rss_conf == NULL) {
3347 err = -ENOMEM;
3348 goto err_alloc_rss;
3349 }
3350 #endif /* VMXNET3_RSS */
3351
3352 err = vmxnet3_alloc_pci_resources(adapter, &dma64);
3353 if (err < 0)
3354 goto err_alloc_pci;
3355
3356 ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_VRRS);
3357 if (ver & (1 << VMXNET3_REV_3)) {
3358 VMXNET3_WRITE_BAR1_REG(adapter,
3359 VMXNET3_REG_VRRS,
3360 1 << VMXNET3_REV_3);
3361 adapter->version = VMXNET3_REV_3 + 1;
3362 } else if (ver & (1 << VMXNET3_REV_2)) {
3363 VMXNET3_WRITE_BAR1_REG(adapter,
3364 VMXNET3_REG_VRRS,
3365 1 << VMXNET3_REV_2);
3366 adapter->version = VMXNET3_REV_2 + 1;
3367 } else if (ver & (1 << VMXNET3_REV_1)) {
3368 VMXNET3_WRITE_BAR1_REG(adapter,
3369 VMXNET3_REG_VRRS,
3370 1 << VMXNET3_REV_1);
3371 adapter->version = VMXNET3_REV_1 + 1;
3372 } else {
3373 dev_err(&pdev->dev,
3374 "Incompatible h/w version (0x%x) for adapter\n", ver);
3375 err = -EBUSY;
3376 goto err_ver;
3377 }
3378 dev_dbg(&pdev->dev, "Using device version %d\n", adapter->version);
3379
3380 ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_UVRS);
3381 if (ver & 1) {
3382 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_UVRS, 1);
3383 } else {
3384 dev_err(&pdev->dev,
3385 "Incompatible upt version (0x%x) for adapter\n", ver);
3386 err = -EBUSY;
3387 goto err_ver;
3388 }
3389
3390 if (VMXNET3_VERSION_GE_3(adapter)) {
3391 adapter->coal_conf =
3392 dma_alloc_coherent(&adapter->pdev->dev,
3393 sizeof(struct Vmxnet3_CoalesceScheme)
3394 ,
3395 &adapter->coal_conf_pa,
3396 GFP_KERNEL);
3397 if (!adapter->coal_conf) {
3398 err = -ENOMEM;
3399 goto err_ver;
3400 }
3401 memset(adapter->coal_conf, 0, sizeof(*adapter->coal_conf));
3402 adapter->coal_conf->coalMode = VMXNET3_COALESCE_DISABLED;
3403 adapter->default_coal_mode = true;
3404 }
3405
3406 SET_NETDEV_DEV(netdev, &pdev->dev);
3407 vmxnet3_declare_features(adapter, dma64);
3408
3409 adapter->rxdata_desc_size = VMXNET3_VERSION_GE_3(adapter) ?
3410 VMXNET3_DEF_RXDATA_DESC_SIZE : 0;
3411
3412 if (adapter->num_tx_queues == adapter->num_rx_queues)
3413 adapter->share_intr = VMXNET3_INTR_BUDDYSHARE;
3414 else
3415 adapter->share_intr = VMXNET3_INTR_DONTSHARE;
3416
3417 vmxnet3_alloc_intr_resources(adapter);
3418
3419 #ifdef VMXNET3_RSS
3420 if (adapter->num_rx_queues > 1 &&
3421 adapter->intr.type == VMXNET3_IT_MSIX) {
3422 adapter->rss = true;
3423 netdev->hw_features |= NETIF_F_RXHASH;
3424 netdev->features |= NETIF_F_RXHASH;
3425 dev_dbg(&pdev->dev, "RSS is enabled.\n");
3426 } else {
3427 adapter->rss = false;
3428 }
3429 #endif
3430
3431 vmxnet3_read_mac_addr(adapter, mac);
3432 memcpy(netdev->dev_addr, mac, netdev->addr_len);
3433
3434 netdev->netdev_ops = &vmxnet3_netdev_ops;
3435 vmxnet3_set_ethtool_ops(netdev);
3436 netdev->watchdog_timeo = 5 * HZ;
3437
3438 /* MTU range: 60 - 9000 */
3439 netdev->min_mtu = VMXNET3_MIN_MTU;
3440 netdev->max_mtu = VMXNET3_MAX_MTU;
3441
3442 INIT_WORK(&adapter->work, vmxnet3_reset_work);
3443 set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
3444
3445 if (adapter->intr.type == VMXNET3_IT_MSIX) {
3446 int i;
3447 for (i = 0; i < adapter->num_rx_queues; i++) {
3448 netif_napi_add(adapter->netdev,
3449 &adapter->rx_queue[i].napi,
3450 vmxnet3_poll_rx_only, 64);
3451 }
3452 } else {
3453 netif_napi_add(adapter->netdev, &adapter->rx_queue[0].napi,
3454 vmxnet3_poll, 64);
3455 }
3456
3457 netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
3458 netif_set_real_num_rx_queues(adapter->netdev, adapter->num_rx_queues);
3459
3460 netif_carrier_off(netdev);
3461 err = register_netdev(netdev);
3462
3463 if (err) {
3464 dev_err(&pdev->dev, "Failed to register adapter\n");
3465 goto err_register;
3466 }
3467
3468 vmxnet3_check_link(adapter, false);
3469 return 0;
3470
3471 err_register:
3472 if (VMXNET3_VERSION_GE_3(adapter)) {
3473 dma_free_coherent(&adapter->pdev->dev,
3474 sizeof(struct Vmxnet3_CoalesceScheme),
3475 adapter->coal_conf, adapter->coal_conf_pa);
3476 }
3477 vmxnet3_free_intr_resources(adapter);
3478 err_ver:
3479 vmxnet3_free_pci_resources(adapter);
3480 err_alloc_pci:
3481 #ifdef VMXNET3_RSS
3482 dma_free_coherent(&adapter->pdev->dev, sizeof(struct UPT1_RSSConf),
3483 adapter->rss_conf, adapter->rss_conf_pa);
3484 err_alloc_rss:
3485 #endif
3486 dma_free_coherent(&adapter->pdev->dev, sizeof(struct Vmxnet3_PMConf),
3487 adapter->pm_conf, adapter->pm_conf_pa);
3488 err_alloc_pm:
3489 dma_free_coherent(&adapter->pdev->dev, size, adapter->tqd_start,
3490 adapter->queue_desc_pa);
3491 err_alloc_queue_desc:
3492 dma_free_coherent(&adapter->pdev->dev,
3493 sizeof(struct Vmxnet3_DriverShared),
3494 adapter->shared, adapter->shared_pa);
3495 err_alloc_shared:
3496 dma_unmap_single(&adapter->pdev->dev, adapter->adapter_pa,
3497 sizeof(struct vmxnet3_adapter), PCI_DMA_TODEVICE);
3498 err_dma_map:
3499 free_netdev(netdev);
3500 return err;
3501 }
3502
3503
3504 static void
3505 vmxnet3_remove_device(struct pci_dev *pdev)
3506 {
3507 struct net_device *netdev = pci_get_drvdata(pdev);
3508 struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3509 int size = 0;
3510 int num_rx_queues;
3511
3512 #ifdef VMXNET3_RSS
3513 if (enable_mq)
3514 num_rx_queues = min(VMXNET3_DEVICE_MAX_RX_QUEUES,
3515 (int)num_online_cpus());
3516 else
3517 #endif
3518 num_rx_queues = 1;
3519 num_rx_queues = rounddown_pow_of_two(num_rx_queues);
3520
3521 cancel_work_sync(&adapter->work);
3522
3523 unregister_netdev(netdev);
3524
3525 vmxnet3_free_intr_resources(adapter);
3526 vmxnet3_free_pci_resources(adapter);
3527 if (VMXNET3_VERSION_GE_3(adapter)) {
3528 dma_free_coherent(&adapter->pdev->dev,
3529 sizeof(struct Vmxnet3_CoalesceScheme),
3530 adapter->coal_conf, adapter->coal_conf_pa);
3531 }
3532 #ifdef VMXNET3_RSS
3533 dma_free_coherent(&adapter->pdev->dev, sizeof(struct UPT1_RSSConf),
3534 adapter->rss_conf, adapter->rss_conf_pa);
3535 #endif
3536 dma_free_coherent(&adapter->pdev->dev, sizeof(struct Vmxnet3_PMConf),
3537 adapter->pm_conf, adapter->pm_conf_pa);
3538
3539 size = sizeof(struct Vmxnet3_TxQueueDesc) * adapter->num_tx_queues;
3540 size += sizeof(struct Vmxnet3_RxQueueDesc) * num_rx_queues;
3541 dma_free_coherent(&adapter->pdev->dev, size, adapter->tqd_start,
3542 adapter->queue_desc_pa);
3543 dma_free_coherent(&adapter->pdev->dev,
3544 sizeof(struct Vmxnet3_DriverShared),
3545 adapter->shared, adapter->shared_pa);
3546 dma_unmap_single(&adapter->pdev->dev, adapter->adapter_pa,
3547 sizeof(struct vmxnet3_adapter), PCI_DMA_TODEVICE);
3548 free_netdev(netdev);
3549 }
3550
3551 static void vmxnet3_shutdown_device(struct pci_dev *pdev)
3552 {
3553 struct net_device *netdev = pci_get_drvdata(pdev);
3554 struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3555 unsigned long flags;
3556
3557 /* Reset_work may be in the middle of resetting the device, wait for its
3558 * completion.
3559 */
3560 while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
3561 msleep(1);
3562
3563 if (test_and_set_bit(VMXNET3_STATE_BIT_QUIESCED,
3564 &adapter->state)) {
3565 clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
3566 return;
3567 }
3568 spin_lock_irqsave(&adapter->cmd_lock, flags);
3569 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3570 VMXNET3_CMD_QUIESCE_DEV);
3571 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3572 vmxnet3_disable_all_intrs(adapter);
3573
3574 clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
3575 }
3576
3577
3578 #ifdef CONFIG_PM
3579
3580 static int
3581 vmxnet3_suspend(struct device *device)
3582 {
3583 struct pci_dev *pdev = to_pci_dev(device);
3584 struct net_device *netdev = pci_get_drvdata(pdev);
3585 struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3586 struct Vmxnet3_PMConf *pmConf;
3587 struct ethhdr *ehdr;
3588 struct arphdr *ahdr;
3589 u8 *arpreq;
3590 struct in_device *in_dev;
3591 struct in_ifaddr *ifa;
3592 unsigned long flags;
3593 int i = 0;
3594
3595 if (!netif_running(netdev))
3596 return 0;
3597
3598 for (i = 0; i < adapter->num_rx_queues; i++)
3599 napi_disable(&adapter->rx_queue[i].napi);
3600
3601 vmxnet3_disable_all_intrs(adapter);
3602 vmxnet3_free_irqs(adapter);
3603 vmxnet3_free_intr_resources(adapter);
3604
3605 netif_device_detach(netdev);
3606 netif_tx_stop_all_queues(netdev);
3607
3608 /* Create wake-up filters. */
3609 pmConf = adapter->pm_conf;
3610 memset(pmConf, 0, sizeof(*pmConf));
3611
3612 if (adapter->wol & WAKE_UCAST) {
3613 pmConf->filters[i].patternSize = ETH_ALEN;
3614 pmConf->filters[i].maskSize = 1;
3615 memcpy(pmConf->filters[i].pattern, netdev->dev_addr, ETH_ALEN);
3616 pmConf->filters[i].mask[0] = 0x3F; /* LSB ETH_ALEN bits */
3617
3618 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_FILTER;
3619 i++;
3620 }
3621
3622 if (adapter->wol & WAKE_ARP) {
3623 in_dev = in_dev_get(netdev);
3624 if (!in_dev)
3625 goto skip_arp;
3626
3627 ifa = (struct in_ifaddr *)in_dev->ifa_list;
3628 if (!ifa)
3629 goto skip_arp;
3630
3631 pmConf->filters[i].patternSize = ETH_HLEN + /* Ethernet header*/
3632 sizeof(struct arphdr) + /* ARP header */
3633 2 * ETH_ALEN + /* 2 Ethernet addresses*/
3634 2 * sizeof(u32); /*2 IPv4 addresses */
3635 pmConf->filters[i].maskSize =
3636 (pmConf->filters[i].patternSize - 1) / 8 + 1;
3637
3638 /* ETH_P_ARP in Ethernet header. */
3639 ehdr = (struct ethhdr *)pmConf->filters[i].pattern;
3640 ehdr->h_proto = htons(ETH_P_ARP);
3641
3642 /* ARPOP_REQUEST in ARP header. */
3643 ahdr = (struct arphdr *)&pmConf->filters[i].pattern[ETH_HLEN];
3644 ahdr->ar_op = htons(ARPOP_REQUEST);
3645 arpreq = (u8 *)(ahdr + 1);
3646
3647 /* The Unicast IPv4 address in 'tip' field. */
3648 arpreq += 2 * ETH_ALEN + sizeof(u32);
3649 *(u32 *)arpreq = ifa->ifa_address;
3650
3651 /* The mask for the relevant bits. */
3652 pmConf->filters[i].mask[0] = 0x00;
3653 pmConf->filters[i].mask[1] = 0x30; /* ETH_P_ARP */
3654 pmConf->filters[i].mask[2] = 0x30; /* ARPOP_REQUEST */
3655 pmConf->filters[i].mask[3] = 0x00;
3656 pmConf->filters[i].mask[4] = 0xC0; /* IPv4 TIP */
3657 pmConf->filters[i].mask[5] = 0x03; /* IPv4 TIP */
3658 in_dev_put(in_dev);
3659
3660 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_FILTER;
3661 i++;
3662 }
3663
3664 skip_arp:
3665 if (adapter->wol & WAKE_MAGIC)
3666 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_MAGIC;
3667
3668 pmConf->numFilters = i;
3669
3670 adapter->shared->devRead.pmConfDesc.confVer = cpu_to_le32(1);
3671 adapter->shared->devRead.pmConfDesc.confLen = cpu_to_le32(sizeof(
3672 *pmConf));
3673 adapter->shared->devRead.pmConfDesc.confPA =
3674 cpu_to_le64(adapter->pm_conf_pa);
3675
3676 spin_lock_irqsave(&adapter->cmd_lock, flags);
3677 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3678 VMXNET3_CMD_UPDATE_PMCFG);
3679 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3680
3681 pci_save_state(pdev);
3682 pci_enable_wake(pdev, pci_choose_state(pdev, PMSG_SUSPEND),
3683 adapter->wol);
3684 pci_disable_device(pdev);
3685 pci_set_power_state(pdev, pci_choose_state(pdev, PMSG_SUSPEND));
3686
3687 return 0;
3688 }
3689
3690
3691 static int
3692 vmxnet3_resume(struct device *device)
3693 {
3694 int err;
3695 unsigned long flags;
3696 struct pci_dev *pdev = to_pci_dev(device);
3697 struct net_device *netdev = pci_get_drvdata(pdev);
3698 struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3699
3700 if (!netif_running(netdev))
3701 return 0;
3702
3703 pci_set_power_state(pdev, PCI_D0);
3704 pci_restore_state(pdev);
3705 err = pci_enable_device_mem(pdev);
3706 if (err != 0)
3707 return err;
3708
3709 pci_enable_wake(pdev, PCI_D0, 0);
3710
3711 vmxnet3_alloc_intr_resources(adapter);
3712
3713 /* During hibernate and suspend, device has to be reinitialized as the
3714 * device state need not be preserved.
3715 */
3716
3717 /* Need not check adapter state as other reset tasks cannot run during
3718 * device resume.
3719 */
3720 spin_lock_irqsave(&adapter->cmd_lock, flags);
3721 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3722 VMXNET3_CMD_QUIESCE_DEV);
3723 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3724 vmxnet3_tq_cleanup_all(adapter);
3725 vmxnet3_rq_cleanup_all(adapter);
3726
3727 vmxnet3_reset_dev(adapter);
3728 err = vmxnet3_activate_dev(adapter);
3729 if (err != 0) {
3730 netdev_err(netdev,
3731 "failed to re-activate on resume, error: %d", err);
3732 vmxnet3_force_close(adapter);
3733 return err;
3734 }
3735 netif_device_attach(netdev);
3736
3737 return 0;
3738 }
3739
3740 static const struct dev_pm_ops vmxnet3_pm_ops = {
3741 .suspend = vmxnet3_suspend,
3742 .resume = vmxnet3_resume,
3743 .freeze = vmxnet3_suspend,
3744 .restore = vmxnet3_resume,
3745 };
3746 #endif
3747
3748 static struct pci_driver vmxnet3_driver = {
3749 .name = vmxnet3_driver_name,
3750 .id_table = vmxnet3_pciid_table,
3751 .probe = vmxnet3_probe_device,
3752 .remove = vmxnet3_remove_device,
3753 .shutdown = vmxnet3_shutdown_device,
3754 #ifdef CONFIG_PM
3755 .driver.pm = &vmxnet3_pm_ops,
3756 #endif
3757 };
3758
3759
3760 static int __init
3761 vmxnet3_init_module(void)
3762 {
3763 pr_info("%s - version %s\n", VMXNET3_DRIVER_DESC,
3764 VMXNET3_DRIVER_VERSION_REPORT);
3765 return pci_register_driver(&vmxnet3_driver);
3766 }
3767
3768 module_init(vmxnet3_init_module);
3769
3770
3771 static void
3772 vmxnet3_exit_module(void)
3773 {
3774 pci_unregister_driver(&vmxnet3_driver);
3775 }
3776
3777 module_exit(vmxnet3_exit_module);
3778
3779 MODULE_AUTHOR("VMware, Inc.");
3780 MODULE_DESCRIPTION(VMXNET3_DRIVER_DESC);
3781 MODULE_LICENSE("GPL v2");
3782 MODULE_VERSION(VMXNET3_DRIVER_VERSION_STRING);
Linux with added FPC-III support