?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
WIP FPC-III support
[linux/fpc-iii.git] / drivers / net / vmxnet3 / vmxnet3_drv.c
blob336504b7531d9099eac648abf483511c696e4061
1 /*
2 * Linux driver for VMware's vmxnet3 ethernet NIC.
3 *
4 * Copyright (C) 2008-2020, 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 /* Prevent any &gdesc->tcd field from being (speculatively)
373 * read before (&gdesc->tcd)->gen is read.
374 */
375 dma_rmb();
376
377 completed += vmxnet3_unmap_pkt(VMXNET3_TCD_GET_TXIDX(
378 &gdesc->tcd), tq, adapter->pdev,
379 adapter);
380
381 vmxnet3_comp_ring_adv_next2proc(&tq->comp_ring);
382 gdesc = tq->comp_ring.base + tq->comp_ring.next2proc;
383 }
384
385 if (completed) {
386 spin_lock(&tq->tx_lock);
387 if (unlikely(vmxnet3_tq_stopped(tq, adapter) &&
388 vmxnet3_cmd_ring_desc_avail(&tq->tx_ring) >
389 VMXNET3_WAKE_QUEUE_THRESHOLD(tq) &&
390 netif_carrier_ok(adapter->netdev))) {
391 vmxnet3_tq_wake(tq, adapter);
392 }
393 spin_unlock(&tq->tx_lock);
394 }
395 return completed;
396 }
397
398
399 static void
400 vmxnet3_tq_cleanup(struct vmxnet3_tx_queue *tq,
401 struct vmxnet3_adapter *adapter)
402 {
403 int i;
404
405 while (tq->tx_ring.next2comp != tq->tx_ring.next2fill) {
406 struct vmxnet3_tx_buf_info *tbi;
407
408 tbi = tq->buf_info + tq->tx_ring.next2comp;
409
410 vmxnet3_unmap_tx_buf(tbi, adapter->pdev);
411 if (tbi->skb) {
412 dev_kfree_skb_any(tbi->skb);
413 tbi->skb = NULL;
414 }
415 vmxnet3_cmd_ring_adv_next2comp(&tq->tx_ring);
416 }
417
418 /* sanity check, verify all buffers are indeed unmapped and freed */
419 for (i = 0; i < tq->tx_ring.size; i++) {
420 BUG_ON(tq->buf_info[i].skb != NULL ||
421 tq->buf_info[i].map_type != VMXNET3_MAP_NONE);
422 }
423
424 tq->tx_ring.gen = VMXNET3_INIT_GEN;
425 tq->tx_ring.next2fill = tq->tx_ring.next2comp = 0;
426
427 tq->comp_ring.gen = VMXNET3_INIT_GEN;
428 tq->comp_ring.next2proc = 0;
429 }
430
431
432 static void
433 vmxnet3_tq_destroy(struct vmxnet3_tx_queue *tq,
434 struct vmxnet3_adapter *adapter)
435 {
436 if (tq->tx_ring.base) {
437 dma_free_coherent(&adapter->pdev->dev, tq->tx_ring.size *
438 sizeof(struct Vmxnet3_TxDesc),
439 tq->tx_ring.base, tq->tx_ring.basePA);
440 tq->tx_ring.base = NULL;
441 }
442 if (tq->data_ring.base) {
443 dma_free_coherent(&adapter->pdev->dev,
444 tq->data_ring.size * tq->txdata_desc_size,
445 tq->data_ring.base, tq->data_ring.basePA);
446 tq->data_ring.base = NULL;
447 }
448 if (tq->comp_ring.base) {
449 dma_free_coherent(&adapter->pdev->dev, tq->comp_ring.size *
450 sizeof(struct Vmxnet3_TxCompDesc),
451 tq->comp_ring.base, tq->comp_ring.basePA);
452 tq->comp_ring.base = NULL;
453 }
454 if (tq->buf_info) {
455 dma_free_coherent(&adapter->pdev->dev,
456 tq->tx_ring.size * sizeof(tq->buf_info[0]),
457 tq->buf_info, tq->buf_info_pa);
458 tq->buf_info = NULL;
459 }
460 }
461
462
463 /* Destroy all tx queues */
464 void
465 vmxnet3_tq_destroy_all(struct vmxnet3_adapter *adapter)
466 {
467 int i;
468
469 for (i = 0; i < adapter->num_tx_queues; i++)
470 vmxnet3_tq_destroy(&adapter->tx_queue[i], adapter);
471 }
472
473
474 static void
475 vmxnet3_tq_init(struct vmxnet3_tx_queue *tq,
476 struct vmxnet3_adapter *adapter)
477 {
478 int i;
479
480 /* reset the tx ring contents to 0 and reset the tx ring states */
481 memset(tq->tx_ring.base, 0, tq->tx_ring.size *
482 sizeof(struct Vmxnet3_TxDesc));
483 tq->tx_ring.next2fill = tq->tx_ring.next2comp = 0;
484 tq->tx_ring.gen = VMXNET3_INIT_GEN;
485
486 memset(tq->data_ring.base, 0,
487 tq->data_ring.size * tq->txdata_desc_size);
488
489 /* reset the tx comp ring contents to 0 and reset comp ring states */
490 memset(tq->comp_ring.base, 0, tq->comp_ring.size *
491 sizeof(struct Vmxnet3_TxCompDesc));
492 tq->comp_ring.next2proc = 0;
493 tq->comp_ring.gen = VMXNET3_INIT_GEN;
494
495 /* reset the bookkeeping data */
496 memset(tq->buf_info, 0, sizeof(tq->buf_info[0]) * tq->tx_ring.size);
497 for (i = 0; i < tq->tx_ring.size; i++)
498 tq->buf_info[i].map_type = VMXNET3_MAP_NONE;
499
500 /* stats are not reset */
501 }
502
503
504 static int
505 vmxnet3_tq_create(struct vmxnet3_tx_queue *tq,
506 struct vmxnet3_adapter *adapter)
507 {
508 size_t sz;
509
510 BUG_ON(tq->tx_ring.base || tq->data_ring.base ||
511 tq->comp_ring.base || tq->buf_info);
512
513 tq->tx_ring.base = dma_alloc_coherent(&adapter->pdev->dev,
514 tq->tx_ring.size * sizeof(struct Vmxnet3_TxDesc),
515 &tq->tx_ring.basePA, GFP_KERNEL);
516 if (!tq->tx_ring.base) {
517 netdev_err(adapter->netdev, "failed to allocate tx ring\n");
518 goto err;
519 }
520
521 tq->data_ring.base = dma_alloc_coherent(&adapter->pdev->dev,
522 tq->data_ring.size * tq->txdata_desc_size,
523 &tq->data_ring.basePA, GFP_KERNEL);
524 if (!tq->data_ring.base) {
525 netdev_err(adapter->netdev, "failed to allocate tx data ring\n");
526 goto err;
527 }
528
529 tq->comp_ring.base = dma_alloc_coherent(&adapter->pdev->dev,
530 tq->comp_ring.size * sizeof(struct Vmxnet3_TxCompDesc),
531 &tq->comp_ring.basePA, GFP_KERNEL);
532 if (!tq->comp_ring.base) {
533 netdev_err(adapter->netdev, "failed to allocate tx comp ring\n");
534 goto err;
535 }
536
537 sz = tq->tx_ring.size * sizeof(tq->buf_info[0]);
538 tq->buf_info = dma_alloc_coherent(&adapter->pdev->dev, sz,
539 &tq->buf_info_pa, GFP_KERNEL);
540 if (!tq->buf_info)
541 goto err;
542
543 return 0;
544
545 err:
546 vmxnet3_tq_destroy(tq, adapter);
547 return -ENOMEM;
548 }
549
550 static void
551 vmxnet3_tq_cleanup_all(struct vmxnet3_adapter *adapter)
552 {
553 int i;
554
555 for (i = 0; i < adapter->num_tx_queues; i++)
556 vmxnet3_tq_cleanup(&adapter->tx_queue[i], adapter);
557 }
558
559 /*
560 * starting from ring->next2fill, allocate rx buffers for the given ring
561 * of the rx queue and update the rx desc. stop after @num_to_alloc buffers
562 * are allocated or allocation fails
563 */
564
565 static int
566 vmxnet3_rq_alloc_rx_buf(struct vmxnet3_rx_queue *rq, u32 ring_idx,
567 int num_to_alloc, struct vmxnet3_adapter *adapter)
568 {
569 int num_allocated = 0;
570 struct vmxnet3_rx_buf_info *rbi_base = rq->buf_info[ring_idx];
571 struct vmxnet3_cmd_ring *ring = &rq->rx_ring[ring_idx];
572 u32 val;
573
574 while (num_allocated <= num_to_alloc) {
575 struct vmxnet3_rx_buf_info *rbi;
576 union Vmxnet3_GenericDesc *gd;
577
578 rbi = rbi_base + ring->next2fill;
579 gd = ring->base + ring->next2fill;
580
581 if (rbi->buf_type == VMXNET3_RX_BUF_SKB) {
582 if (rbi->skb == NULL) {
583 rbi->skb = __netdev_alloc_skb_ip_align(adapter->netdev,
584 rbi->len,
585 GFP_KERNEL);
586 if (unlikely(rbi->skb == NULL)) {
587 rq->stats.rx_buf_alloc_failure++;
588 break;
589 }
590
591 rbi->dma_addr = dma_map_single(
592 &adapter->pdev->dev,
593 rbi->skb->data, rbi->len,
594 PCI_DMA_FROMDEVICE);
595 if (dma_mapping_error(&adapter->pdev->dev,
596 rbi->dma_addr)) {
597 dev_kfree_skb_any(rbi->skb);
598 rq->stats.rx_buf_alloc_failure++;
599 break;
600 }
601 } else {
602 /* rx buffer skipped by the device */
603 }
604 val = VMXNET3_RXD_BTYPE_HEAD << VMXNET3_RXD_BTYPE_SHIFT;
605 } else {
606 BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_PAGE ||
607 rbi->len != PAGE_SIZE);
608
609 if (rbi->page == NULL) {
610 rbi->page = alloc_page(GFP_ATOMIC);
611 if (unlikely(rbi->page == NULL)) {
612 rq->stats.rx_buf_alloc_failure++;
613 break;
614 }
615 rbi->dma_addr = dma_map_page(
616 &adapter->pdev->dev,
617 rbi->page, 0, PAGE_SIZE,
618 PCI_DMA_FROMDEVICE);
619 if (dma_mapping_error(&adapter->pdev->dev,
620 rbi->dma_addr)) {
621 put_page(rbi->page);
622 rq->stats.rx_buf_alloc_failure++;
623 break;
624 }
625 } else {
626 /* rx buffers skipped by the device */
627 }
628 val = VMXNET3_RXD_BTYPE_BODY << VMXNET3_RXD_BTYPE_SHIFT;
629 }
630
631 gd->rxd.addr = cpu_to_le64(rbi->dma_addr);
632 gd->dword[2] = cpu_to_le32((!ring->gen << VMXNET3_RXD_GEN_SHIFT)
633 | val | rbi->len);
634
635 /* Fill the last buffer but dont mark it ready, or else the
636 * device will think that the queue is full */
637 if (num_allocated == num_to_alloc)
638 break;
639
640 gd->dword[2] |= cpu_to_le32(ring->gen << VMXNET3_RXD_GEN_SHIFT);
641 num_allocated++;
642 vmxnet3_cmd_ring_adv_next2fill(ring);
643 }
644
645 netdev_dbg(adapter->netdev,
646 "alloc_rx_buf: %d allocated, next2fill %u, next2comp %u\n",
647 num_allocated, ring->next2fill, ring->next2comp);
648
649 /* so that the device can distinguish a full ring and an empty ring */
650 BUG_ON(num_allocated != 0 && ring->next2fill == ring->next2comp);
651
652 return num_allocated;
653 }
654
655
656 static void
657 vmxnet3_append_frag(struct sk_buff *skb, struct Vmxnet3_RxCompDesc *rcd,
658 struct vmxnet3_rx_buf_info *rbi)
659 {
660 skb_frag_t *frag = skb_shinfo(skb)->frags + skb_shinfo(skb)->nr_frags;
661
662 BUG_ON(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS);
663
664 __skb_frag_set_page(frag, rbi->page);
665 skb_frag_off_set(frag, 0);
666 skb_frag_size_set(frag, rcd->len);
667 skb->data_len += rcd->len;
668 skb->truesize += PAGE_SIZE;
669 skb_shinfo(skb)->nr_frags++;
670 }
671
672
673 static int
674 vmxnet3_map_pkt(struct sk_buff *skb, struct vmxnet3_tx_ctx *ctx,
675 struct vmxnet3_tx_queue *tq, struct pci_dev *pdev,
676 struct vmxnet3_adapter *adapter)
677 {
678 u32 dw2, len;
679 unsigned long buf_offset;
680 int i;
681 union Vmxnet3_GenericDesc *gdesc;
682 struct vmxnet3_tx_buf_info *tbi = NULL;
683
684 BUG_ON(ctx->copy_size > skb_headlen(skb));
685
686 /* use the previous gen bit for the SOP desc */
687 dw2 = (tq->tx_ring.gen ^ 0x1) << VMXNET3_TXD_GEN_SHIFT;
688
689 ctx->sop_txd = tq->tx_ring.base + tq->tx_ring.next2fill;
690 gdesc = ctx->sop_txd; /* both loops below can be skipped */
691
692 /* no need to map the buffer if headers are copied */
693 if (ctx->copy_size) {
694 ctx->sop_txd->txd.addr = cpu_to_le64(tq->data_ring.basePA +
695 tq->tx_ring.next2fill *
696 tq->txdata_desc_size);
697 ctx->sop_txd->dword[2] = cpu_to_le32(dw2 | ctx->copy_size);
698 ctx->sop_txd->dword[3] = 0;
699
700 tbi = tq->buf_info + tq->tx_ring.next2fill;
701 tbi->map_type = VMXNET3_MAP_NONE;
702
703 netdev_dbg(adapter->netdev,
704 "txd[%u]: 0x%Lx 0x%x 0x%x\n",
705 tq->tx_ring.next2fill,
706 le64_to_cpu(ctx->sop_txd->txd.addr),
707 ctx->sop_txd->dword[2], ctx->sop_txd->dword[3]);
708 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
709
710 /* use the right gen for non-SOP desc */
711 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
712 }
713
714 /* linear part can use multiple tx desc if it's big */
715 len = skb_headlen(skb) - ctx->copy_size;
716 buf_offset = ctx->copy_size;
717 while (len) {
718 u32 buf_size;
719
720 if (len < VMXNET3_MAX_TX_BUF_SIZE) {
721 buf_size = len;
722 dw2 |= len;
723 } else {
724 buf_size = VMXNET3_MAX_TX_BUF_SIZE;
725 /* spec says that for TxDesc.len, 0 == 2^14 */
726 }
727
728 tbi = tq->buf_info + tq->tx_ring.next2fill;
729 tbi->map_type = VMXNET3_MAP_SINGLE;
730 tbi->dma_addr = dma_map_single(&adapter->pdev->dev,
731 skb->data + buf_offset, buf_size,
732 PCI_DMA_TODEVICE);
733 if (dma_mapping_error(&adapter->pdev->dev, tbi->dma_addr))
734 return -EFAULT;
735
736 tbi->len = buf_size;
737
738 gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
739 BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
740
741 gdesc->txd.addr = cpu_to_le64(tbi->dma_addr);
742 gdesc->dword[2] = cpu_to_le32(dw2);
743 gdesc->dword[3] = 0;
744
745 netdev_dbg(adapter->netdev,
746 "txd[%u]: 0x%Lx 0x%x 0x%x\n",
747 tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr),
748 le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]);
749 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
750 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
751
752 len -= buf_size;
753 buf_offset += buf_size;
754 }
755
756 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
757 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
758 u32 buf_size;
759
760 buf_offset = 0;
761 len = skb_frag_size(frag);
762 while (len) {
763 tbi = tq->buf_info + tq->tx_ring.next2fill;
764 if (len < VMXNET3_MAX_TX_BUF_SIZE) {
765 buf_size = len;
766 dw2 |= len;
767 } else {
768 buf_size = VMXNET3_MAX_TX_BUF_SIZE;
769 /* spec says that for TxDesc.len, 0 == 2^14 */
770 }
771 tbi->map_type = VMXNET3_MAP_PAGE;
772 tbi->dma_addr = skb_frag_dma_map(&adapter->pdev->dev, frag,
773 buf_offset, buf_size,
774 DMA_TO_DEVICE);
775 if (dma_mapping_error(&adapter->pdev->dev, tbi->dma_addr))
776 return -EFAULT;
777
778 tbi->len = buf_size;
779
780 gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
781 BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
782
783 gdesc->txd.addr = cpu_to_le64(tbi->dma_addr);
784 gdesc->dword[2] = cpu_to_le32(dw2);
785 gdesc->dword[3] = 0;
786
787 netdev_dbg(adapter->netdev,
788 "txd[%u]: 0x%llx %u %u\n",
789 tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr),
790 le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]);
791 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
792 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
793
794 len -= buf_size;
795 buf_offset += buf_size;
796 }
797 }
798
799 ctx->eop_txd = gdesc;
800
801 /* set the last buf_info for the pkt */
802 tbi->skb = skb;
803 tbi->sop_idx = ctx->sop_txd - tq->tx_ring.base;
804
805 return 0;
806 }
807
808
809 /* Init all tx queues */
810 static void
811 vmxnet3_tq_init_all(struct vmxnet3_adapter *adapter)
812 {
813 int i;
814
815 for (i = 0; i < adapter->num_tx_queues; i++)
816 vmxnet3_tq_init(&adapter->tx_queue[i], adapter);
817 }
818
819
820 /*
821 * parse relevant protocol headers:
822 * For a tso pkt, relevant headers are L2/3/4 including options
823 * For a pkt requesting csum offloading, they are L2/3 and may include L4
824 * if it's a TCP/UDP pkt
825 *
826 * Returns:
827 * -1: error happens during parsing
828 * 0: protocol headers parsed, but too big to be copied
829 * 1: protocol headers parsed and copied
830 *
831 * Other effects:
832 * 1. related *ctx fields are updated.
833 * 2. ctx->copy_size is # of bytes copied
834 * 3. the portion to be copied is guaranteed to be in the linear part
835 *
836 */
837 static int
838 vmxnet3_parse_hdr(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
839 struct vmxnet3_tx_ctx *ctx,
840 struct vmxnet3_adapter *adapter)
841 {
842 u8 protocol = 0;
843
844 if (ctx->mss) { /* TSO */
845 if (VMXNET3_VERSION_GE_4(adapter) && skb->encapsulation) {
846 ctx->l4_offset = skb_inner_transport_offset(skb);
847 ctx->l4_hdr_size = inner_tcp_hdrlen(skb);
848 ctx->copy_size = ctx->l4_offset + ctx->l4_hdr_size;
849 } else {
850 ctx->l4_offset = skb_transport_offset(skb);
851 ctx->l4_hdr_size = tcp_hdrlen(skb);
852 ctx->copy_size = ctx->l4_offset + ctx->l4_hdr_size;
853 }
854 } else {
855 if (skb->ip_summed == CHECKSUM_PARTIAL) {
856 /* For encap packets, skb_checksum_start_offset refers
857 * to inner L4 offset. Thus, below works for encap as
858 * well as non-encap case
859 */
860 ctx->l4_offset = skb_checksum_start_offset(skb);
861
862 if (VMXNET3_VERSION_GE_4(adapter) &&
863 skb->encapsulation) {
864 struct iphdr *iph = inner_ip_hdr(skb);
865
866 if (iph->version == 4) {
867 protocol = iph->protocol;
868 } else {
869 const struct ipv6hdr *ipv6h;
870
871 ipv6h = inner_ipv6_hdr(skb);
872 protocol = ipv6h->nexthdr;
873 }
874 } else {
875 if (ctx->ipv4) {
876 const struct iphdr *iph = ip_hdr(skb);
877
878 protocol = iph->protocol;
879 } else if (ctx->ipv6) {
880 const struct ipv6hdr *ipv6h;
881
882 ipv6h = ipv6_hdr(skb);
883 protocol = ipv6h->nexthdr;
884 }
885 }
886
887 switch (protocol) {
888 case IPPROTO_TCP:
889 ctx->l4_hdr_size = skb->encapsulation ? inner_tcp_hdrlen(skb) :
890 tcp_hdrlen(skb);
891 break;
892 case IPPROTO_UDP:
893 ctx->l4_hdr_size = sizeof(struct udphdr);
894 break;
895 default:
896 ctx->l4_hdr_size = 0;
897 break;
898 }
899
900 ctx->copy_size = min(ctx->l4_offset +
901 ctx->l4_hdr_size, skb->len);
902 } else {
903 ctx->l4_offset = 0;
904 ctx->l4_hdr_size = 0;
905 /* copy as much as allowed */
906 ctx->copy_size = min_t(unsigned int,
907 tq->txdata_desc_size,
908 skb_headlen(skb));
909 }
910
911 if (skb->len <= VMXNET3_HDR_COPY_SIZE)
912 ctx->copy_size = skb->len;
913
914 /* make sure headers are accessible directly */
915 if (unlikely(!pskb_may_pull(skb, ctx->copy_size)))
916 goto err;
917 }
918
919 if (unlikely(ctx->copy_size > tq->txdata_desc_size)) {
920 tq->stats.oversized_hdr++;
921 ctx->copy_size = 0;
922 return 0;
923 }
924
925 return 1;
926 err:
927 return -1;
928 }
929
930 /*
931 * copy relevant protocol headers to the transmit ring:
932 * For a tso pkt, relevant headers are L2/3/4 including options
933 * For a pkt requesting csum offloading, they are L2/3 and may include L4
934 * if it's a TCP/UDP pkt
935 *
936 *
937 * Note that this requires that vmxnet3_parse_hdr be called first to set the
938 * appropriate bits in ctx first
939 */
940 static void
941 vmxnet3_copy_hdr(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
942 struct vmxnet3_tx_ctx *ctx,
943 struct vmxnet3_adapter *adapter)
944 {
945 struct Vmxnet3_TxDataDesc *tdd;
946
947 tdd = (struct Vmxnet3_TxDataDesc *)((u8 *)tq->data_ring.base +
948 tq->tx_ring.next2fill *
949 tq->txdata_desc_size);
950
951 memcpy(tdd->data, skb->data, ctx->copy_size);
952 netdev_dbg(adapter->netdev,
953 "copy %u bytes to dataRing[%u]\n",
954 ctx->copy_size, tq->tx_ring.next2fill);
955 }
956
957
958 static void
959 vmxnet3_prepare_inner_tso(struct sk_buff *skb,
960 struct vmxnet3_tx_ctx *ctx)
961 {
962 struct tcphdr *tcph = inner_tcp_hdr(skb);
963 struct iphdr *iph = inner_ip_hdr(skb);
964
965 if (iph->version == 4) {
966 iph->check = 0;
967 tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, 0,
968 IPPROTO_TCP, 0);
969 } else {
970 struct ipv6hdr *iph = inner_ipv6_hdr(skb);
971
972 tcph->check = ~csum_ipv6_magic(&iph->saddr, &iph->daddr, 0,
973 IPPROTO_TCP, 0);
974 }
975 }
976
977 static void
978 vmxnet3_prepare_tso(struct sk_buff *skb,
979 struct vmxnet3_tx_ctx *ctx)
980 {
981 struct tcphdr *tcph = tcp_hdr(skb);
982
983 if (ctx->ipv4) {
984 struct iphdr *iph = ip_hdr(skb);
985
986 iph->check = 0;
987 tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, 0,
988 IPPROTO_TCP, 0);
989 } else if (ctx->ipv6) {
990 tcp_v6_gso_csum_prep(skb);
991 }
992 }
993
994 static int txd_estimate(const struct sk_buff *skb)
995 {
996 int count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) + 1;
997 int i;
998
999 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1000 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1001
1002 count += VMXNET3_TXD_NEEDED(skb_frag_size(frag));
1003 }
1004 return count;
1005 }
1006
1007 /*
1008 * Transmits a pkt thru a given tq
1009 * Returns:
1010 * NETDEV_TX_OK: descriptors are setup successfully
1011 * NETDEV_TX_OK: error occurred, the pkt is dropped
1012 * NETDEV_TX_BUSY: tx ring is full, queue is stopped
1013 *
1014 * Side-effects:
1015 * 1. tx ring may be changed
1016 * 2. tq stats may be updated accordingly
1017 * 3. shared->txNumDeferred may be updated
1018 */
1019
1020 static int
1021 vmxnet3_tq_xmit(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
1022 struct vmxnet3_adapter *adapter, struct net_device *netdev)
1023 {
1024 int ret;
1025 u32 count;
1026 int num_pkts;
1027 int tx_num_deferred;
1028 unsigned long flags;
1029 struct vmxnet3_tx_ctx ctx;
1030 union Vmxnet3_GenericDesc *gdesc;
1031 #ifdef __BIG_ENDIAN_BITFIELD
1032 /* Use temporary descriptor to avoid touching bits multiple times */
1033 union Vmxnet3_GenericDesc tempTxDesc;
1034 #endif
1035
1036 count = txd_estimate(skb);
1037
1038 ctx.ipv4 = (vlan_get_protocol(skb) == cpu_to_be16(ETH_P_IP));
1039 ctx.ipv6 = (vlan_get_protocol(skb) == cpu_to_be16(ETH_P_IPV6));
1040
1041 ctx.mss = skb_shinfo(skb)->gso_size;
1042 if (ctx.mss) {
1043 if (skb_header_cloned(skb)) {
1044 if (unlikely(pskb_expand_head(skb, 0, 0,
1045 GFP_ATOMIC) != 0)) {
1046 tq->stats.drop_tso++;
1047 goto drop_pkt;
1048 }
1049 tq->stats.copy_skb_header++;
1050 }
1051 if (skb->encapsulation) {
1052 vmxnet3_prepare_inner_tso(skb, &ctx);
1053 } else {
1054 vmxnet3_prepare_tso(skb, &ctx);
1055 }
1056 } else {
1057 if (unlikely(count > VMXNET3_MAX_TXD_PER_PKT)) {
1058
1059 /* non-tso pkts must not use more than
1060 * VMXNET3_MAX_TXD_PER_PKT entries
1061 */
1062 if (skb_linearize(skb) != 0) {
1063 tq->stats.drop_too_many_frags++;
1064 goto drop_pkt;
1065 }
1066 tq->stats.linearized++;
1067
1068 /* recalculate the # of descriptors to use */
1069 count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) + 1;
1070 }
1071 }
1072
1073 ret = vmxnet3_parse_hdr(skb, tq, &ctx, adapter);
1074 if (ret >= 0) {
1075 BUG_ON(ret <= 0 && ctx.copy_size != 0);
1076 /* hdrs parsed, check against other limits */
1077 if (ctx.mss) {
1078 if (unlikely(ctx.l4_offset + ctx.l4_hdr_size >
1079 VMXNET3_MAX_TX_BUF_SIZE)) {
1080 tq->stats.drop_oversized_hdr++;
1081 goto drop_pkt;
1082 }
1083 } else {
1084 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1085 if (unlikely(ctx.l4_offset +
1086 skb->csum_offset >
1087 VMXNET3_MAX_CSUM_OFFSET)) {
1088 tq->stats.drop_oversized_hdr++;
1089 goto drop_pkt;
1090 }
1091 }
1092 }
1093 } else {
1094 tq->stats.drop_hdr_inspect_err++;
1095 goto drop_pkt;
1096 }
1097
1098 spin_lock_irqsave(&tq->tx_lock, flags);
1099
1100 if (count > vmxnet3_cmd_ring_desc_avail(&tq->tx_ring)) {
1101 tq->stats.tx_ring_full++;
1102 netdev_dbg(adapter->netdev,
1103 "tx queue stopped on %s, next2comp %u"
1104 " next2fill %u\n", adapter->netdev->name,
1105 tq->tx_ring.next2comp, tq->tx_ring.next2fill);
1106
1107 vmxnet3_tq_stop(tq, adapter);
1108 spin_unlock_irqrestore(&tq->tx_lock, flags);
1109 return NETDEV_TX_BUSY;
1110 }
1111
1112
1113 vmxnet3_copy_hdr(skb, tq, &ctx, adapter);
1114
1115 /* fill tx descs related to addr & len */
1116 if (vmxnet3_map_pkt(skb, &ctx, tq, adapter->pdev, adapter))
1117 goto unlock_drop_pkt;
1118
1119 /* setup the EOP desc */
1120 ctx.eop_txd->dword[3] = cpu_to_le32(VMXNET3_TXD_CQ | VMXNET3_TXD_EOP);
1121
1122 /* setup the SOP desc */
1123 #ifdef __BIG_ENDIAN_BITFIELD
1124 gdesc = &tempTxDesc;
1125 gdesc->dword[2] = ctx.sop_txd->dword[2];
1126 gdesc->dword[3] = ctx.sop_txd->dword[3];
1127 #else
1128 gdesc = ctx.sop_txd;
1129 #endif
1130 tx_num_deferred = le32_to_cpu(tq->shared->txNumDeferred);
1131 if (ctx.mss) {
1132 if (VMXNET3_VERSION_GE_4(adapter) && skb->encapsulation) {
1133 gdesc->txd.hlen = ctx.l4_offset + ctx.l4_hdr_size;
1134 gdesc->txd.om = VMXNET3_OM_ENCAP;
1135 gdesc->txd.msscof = ctx.mss;
1136
1137 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL_CSUM)
1138 gdesc->txd.oco = 1;
1139 } else {
1140 gdesc->txd.hlen = ctx.l4_offset + ctx.l4_hdr_size;
1141 gdesc->txd.om = VMXNET3_OM_TSO;
1142 gdesc->txd.msscof = ctx.mss;
1143 }
1144 num_pkts = (skb->len - gdesc->txd.hlen + ctx.mss - 1) / ctx.mss;
1145 } else {
1146 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1147 if (VMXNET3_VERSION_GE_4(adapter) &&
1148 skb->encapsulation) {
1149 gdesc->txd.hlen = ctx.l4_offset +
1150 ctx.l4_hdr_size;
1151 gdesc->txd.om = VMXNET3_OM_ENCAP;
1152 gdesc->txd.msscof = 0; /* Reserved */
1153 } else {
1154 gdesc->txd.hlen = ctx.l4_offset;
1155 gdesc->txd.om = VMXNET3_OM_CSUM;
1156 gdesc->txd.msscof = ctx.l4_offset +
1157 skb->csum_offset;
1158 }
1159 } else {
1160 gdesc->txd.om = 0;
1161 gdesc->txd.msscof = 0;
1162 }
1163 num_pkts = 1;
1164 }
1165 le32_add_cpu(&tq->shared->txNumDeferred, num_pkts);
1166 tx_num_deferred += num_pkts;
1167
1168 if (skb_vlan_tag_present(skb)) {
1169 gdesc->txd.ti = 1;
1170 gdesc->txd.tci = skb_vlan_tag_get(skb);
1171 }
1172
1173 /* Ensure that the write to (&gdesc->txd)->gen will be observed after
1174 * all other writes to &gdesc->txd.
1175 */
1176 dma_wmb();
1177
1178 /* finally flips the GEN bit of the SOP desc. */
1179 gdesc->dword[2] = cpu_to_le32(le32_to_cpu(gdesc->dword[2]) ^
1180 VMXNET3_TXD_GEN);
1181 #ifdef __BIG_ENDIAN_BITFIELD
1182 /* Finished updating in bitfields of Tx Desc, so write them in original
1183 * place.
1184 */
1185 vmxnet3_TxDescToLe((struct Vmxnet3_TxDesc *)gdesc,
1186 (struct Vmxnet3_TxDesc *)ctx.sop_txd);
1187 gdesc = ctx.sop_txd;
1188 #endif
1189 netdev_dbg(adapter->netdev,
1190 "txd[%u]: SOP 0x%Lx 0x%x 0x%x\n",
1191 (u32)(ctx.sop_txd -
1192 tq->tx_ring.base), le64_to_cpu(gdesc->txd.addr),
1193 le32_to_cpu(gdesc->dword[2]), le32_to_cpu(gdesc->dword[3]));
1194
1195 spin_unlock_irqrestore(&tq->tx_lock, flags);
1196
1197 if (tx_num_deferred >= le32_to_cpu(tq->shared->txThreshold)) {
1198 tq->shared->txNumDeferred = 0;
1199 VMXNET3_WRITE_BAR0_REG(adapter,
1200 VMXNET3_REG_TXPROD + tq->qid * 8,
1201 tq->tx_ring.next2fill);
1202 }
1203
1204 return NETDEV_TX_OK;
1205
1206 unlock_drop_pkt:
1207 spin_unlock_irqrestore(&tq->tx_lock, flags);
1208 drop_pkt:
1209 tq->stats.drop_total++;
1210 dev_kfree_skb_any(skb);
1211 return NETDEV_TX_OK;
1212 }
1213
1214
1215 static netdev_tx_t
1216 vmxnet3_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
1217 {
1218 struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1219
1220 BUG_ON(skb->queue_mapping > adapter->num_tx_queues);
1221 return vmxnet3_tq_xmit(skb,
1222 &adapter->tx_queue[skb->queue_mapping],
1223 adapter, netdev);
1224 }
1225
1226
1227 static void
1228 vmxnet3_rx_csum(struct vmxnet3_adapter *adapter,
1229 struct sk_buff *skb,
1230 union Vmxnet3_GenericDesc *gdesc)
1231 {
1232 if (!gdesc->rcd.cnc && adapter->netdev->features & NETIF_F_RXCSUM) {
1233 if (gdesc->rcd.v4 &&
1234 (le32_to_cpu(gdesc->dword[3]) &
1235 VMXNET3_RCD_CSUM_OK) == VMXNET3_RCD_CSUM_OK) {
1236 skb->ip_summed = CHECKSUM_UNNECESSARY;
1237 WARN_ON_ONCE(!(gdesc->rcd.tcp || gdesc->rcd.udp) &&
1238 !(le32_to_cpu(gdesc->dword[0]) &
1239 (1UL << VMXNET3_RCD_HDR_INNER_SHIFT)));
1240 WARN_ON_ONCE(gdesc->rcd.frg &&
1241 !(le32_to_cpu(gdesc->dword[0]) &
1242 (1UL << VMXNET3_RCD_HDR_INNER_SHIFT)));
1243 } else if (gdesc->rcd.v6 && (le32_to_cpu(gdesc->dword[3]) &
1244 (1 << VMXNET3_RCD_TUC_SHIFT))) {
1245 skb->ip_summed = CHECKSUM_UNNECESSARY;
1246 WARN_ON_ONCE(!(gdesc->rcd.tcp || gdesc->rcd.udp) &&
1247 !(le32_to_cpu(gdesc->dword[0]) &
1248 (1UL << VMXNET3_RCD_HDR_INNER_SHIFT)));
1249 WARN_ON_ONCE(gdesc->rcd.frg &&
1250 !(le32_to_cpu(gdesc->dword[0]) &
1251 (1UL << VMXNET3_RCD_HDR_INNER_SHIFT)));
1252 } else {
1253 if (gdesc->rcd.csum) {
1254 skb->csum = htons(gdesc->rcd.csum);
1255 skb->ip_summed = CHECKSUM_PARTIAL;
1256 } else {
1257 skb_checksum_none_assert(skb);
1258 }
1259 }
1260 } else {
1261 skb_checksum_none_assert(skb);
1262 }
1263 }
1264
1265
1266 static void
1267 vmxnet3_rx_error(struct vmxnet3_rx_queue *rq, struct Vmxnet3_RxCompDesc *rcd,
1268 struct vmxnet3_rx_ctx *ctx, struct vmxnet3_adapter *adapter)
1269 {
1270 rq->stats.drop_err++;
1271 if (!rcd->fcs)
1272 rq->stats.drop_fcs++;
1273
1274 rq->stats.drop_total++;
1275
1276 /*
1277 * We do not unmap and chain the rx buffer to the skb.
1278 * We basically pretend this buffer is not used and will be recycled
1279 * by vmxnet3_rq_alloc_rx_buf()
1280 */
1281
1282 /*
1283 * ctx->skb may be NULL if this is the first and the only one
1284 * desc for the pkt
1285 */
1286 if (ctx->skb)
1287 dev_kfree_skb_irq(ctx->skb);
1288
1289 ctx->skb = NULL;
1290 }
1291
1292
1293 static u32
1294 vmxnet3_get_hdr_len(struct vmxnet3_adapter *adapter, struct sk_buff *skb,
1295 union Vmxnet3_GenericDesc *gdesc)
1296 {
1297 u32 hlen, maplen;
1298 union {
1299 void *ptr;
1300 struct ethhdr *eth;
1301 struct vlan_ethhdr *veth;
1302 struct iphdr *ipv4;
1303 struct ipv6hdr *ipv6;
1304 struct tcphdr *tcp;
1305 } hdr;
1306 BUG_ON(gdesc->rcd.tcp == 0);
1307
1308 maplen = skb_headlen(skb);
1309 if (unlikely(sizeof(struct iphdr) + sizeof(struct tcphdr) > maplen))
1310 return 0;
1311
1312 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
1313 skb->protocol == cpu_to_be16(ETH_P_8021AD))
1314 hlen = sizeof(struct vlan_ethhdr);
1315 else
1316 hlen = sizeof(struct ethhdr);
1317
1318 hdr.eth = eth_hdr(skb);
1319 if (gdesc->rcd.v4) {
1320 BUG_ON(hdr.eth->h_proto != htons(ETH_P_IP) &&
1321 hdr.veth->h_vlan_encapsulated_proto != htons(ETH_P_IP));
1322 hdr.ptr += hlen;
1323 BUG_ON(hdr.ipv4->protocol != IPPROTO_TCP);
1324 hlen = hdr.ipv4->ihl << 2;
1325 hdr.ptr += hdr.ipv4->ihl << 2;
1326 } else if (gdesc->rcd.v6) {
1327 BUG_ON(hdr.eth->h_proto != htons(ETH_P_IPV6) &&
1328 hdr.veth->h_vlan_encapsulated_proto != htons(ETH_P_IPV6));
1329 hdr.ptr += hlen;
1330 /* Use an estimated value, since we also need to handle
1331 * TSO case.
1332 */
1333 if (hdr.ipv6->nexthdr != IPPROTO_TCP)
1334 return sizeof(struct ipv6hdr) + sizeof(struct tcphdr);
1335 hlen = sizeof(struct ipv6hdr);
1336 hdr.ptr += sizeof(struct ipv6hdr);
1337 } else {
1338 /* Non-IP pkt, dont estimate header length */
1339 return 0;
1340 }
1341
1342 if (hlen + sizeof(struct tcphdr) > maplen)
1343 return 0;
1344
1345 return (hlen + (hdr.tcp->doff << 2));
1346 }
1347
1348 static int
1349 vmxnet3_rq_rx_complete(struct vmxnet3_rx_queue *rq,
1350 struct vmxnet3_adapter *adapter, int quota)
1351 {
1352 static const u32 rxprod_reg[2] = {
1353 VMXNET3_REG_RXPROD, VMXNET3_REG_RXPROD2
1354 };
1355 u32 num_pkts = 0;
1356 bool skip_page_frags = false;
1357 struct Vmxnet3_RxCompDesc *rcd;
1358 struct vmxnet3_rx_ctx *ctx = &rq->rx_ctx;
1359 u16 segCnt = 0, mss = 0;
1360 #ifdef __BIG_ENDIAN_BITFIELD
1361 struct Vmxnet3_RxDesc rxCmdDesc;
1362 struct Vmxnet3_RxCompDesc rxComp;
1363 #endif
1364 vmxnet3_getRxComp(rcd, &rq->comp_ring.base[rq->comp_ring.next2proc].rcd,
1365 &rxComp);
1366 while (rcd->gen == rq->comp_ring.gen) {
1367 struct vmxnet3_rx_buf_info *rbi;
1368 struct sk_buff *skb, *new_skb = NULL;
1369 struct page *new_page = NULL;
1370 dma_addr_t new_dma_addr;
1371 int num_to_alloc;
1372 struct Vmxnet3_RxDesc *rxd;
1373 u32 idx, ring_idx;
1374 struct vmxnet3_cmd_ring *ring = NULL;
1375 if (num_pkts >= quota) {
1376 /* we may stop even before we see the EOP desc of
1377 * the current pkt
1378 */
1379 break;
1380 }
1381
1382 /* Prevent any rcd field from being (speculatively) read before
1383 * rcd->gen is read.
1384 */
1385 dma_rmb();
1386
1387 BUG_ON(rcd->rqID != rq->qid && rcd->rqID != rq->qid2 &&
1388 rcd->rqID != rq->dataRingQid);
1389 idx = rcd->rxdIdx;
1390 ring_idx = VMXNET3_GET_RING_IDX(adapter, rcd->rqID);
1391 ring = rq->rx_ring + ring_idx;
1392 vmxnet3_getRxDesc(rxd, &rq->rx_ring[ring_idx].base[idx].rxd,
1393 &rxCmdDesc);
1394 rbi = rq->buf_info[ring_idx] + idx;
1395
1396 BUG_ON(rxd->addr != rbi->dma_addr ||
1397 rxd->len != rbi->len);
1398
1399 if (unlikely(rcd->eop && rcd->err)) {
1400 vmxnet3_rx_error(rq, rcd, ctx, adapter);
1401 goto rcd_done;
1402 }
1403
1404 if (rcd->sop) { /* first buf of the pkt */
1405 bool rxDataRingUsed;
1406 u16 len;
1407
1408 BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_HEAD ||
1409 (rcd->rqID != rq->qid &&
1410 rcd->rqID != rq->dataRingQid));
1411
1412 BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_SKB);
1413 BUG_ON(ctx->skb != NULL || rbi->skb == NULL);
1414
1415 if (unlikely(rcd->len == 0)) {
1416 /* Pretend the rx buffer is skipped. */
1417 BUG_ON(!(rcd->sop && rcd->eop));
1418 netdev_dbg(adapter->netdev,
1419 "rxRing[%u][%u] 0 length\n",
1420 ring_idx, idx);
1421 goto rcd_done;
1422 }
1423
1424 skip_page_frags = false;
1425 ctx->skb = rbi->skb;
1426
1427 rxDataRingUsed =
1428 VMXNET3_RX_DATA_RING(adapter, rcd->rqID);
1429 len = rxDataRingUsed ? rcd->len : rbi->len;
1430 new_skb = netdev_alloc_skb_ip_align(adapter->netdev,
1431 len);
1432 if (new_skb == NULL) {
1433 /* Skb allocation failed, do not handover this
1434 * skb to stack. Reuse it. Drop the existing pkt
1435 */
1436 rq->stats.rx_buf_alloc_failure++;
1437 ctx->skb = NULL;
1438 rq->stats.drop_total++;
1439 skip_page_frags = true;
1440 goto rcd_done;
1441 }
1442
1443 if (rxDataRingUsed) {
1444 size_t sz;
1445
1446 BUG_ON(rcd->len > rq->data_ring.desc_size);
1447
1448 ctx->skb = new_skb;
1449 sz = rcd->rxdIdx * rq->data_ring.desc_size;
1450 memcpy(new_skb->data,
1451 &rq->data_ring.base[sz], rcd->len);
1452 } else {
1453 ctx->skb = rbi->skb;
1454
1455 new_dma_addr =
1456 dma_map_single(&adapter->pdev->dev,
1457 new_skb->data, rbi->len,
1458 PCI_DMA_FROMDEVICE);
1459 if (dma_mapping_error(&adapter->pdev->dev,
1460 new_dma_addr)) {
1461 dev_kfree_skb(new_skb);
1462 /* Skb allocation failed, do not
1463 * handover this skb to stack. Reuse
1464 * it. Drop the existing pkt.
1465 */
1466 rq->stats.rx_buf_alloc_failure++;
1467 ctx->skb = NULL;
1468 rq->stats.drop_total++;
1469 skip_page_frags = true;
1470 goto rcd_done;
1471 }
1472
1473 dma_unmap_single(&adapter->pdev->dev,
1474 rbi->dma_addr,
1475 rbi->len,
1476 PCI_DMA_FROMDEVICE);
1477
1478 /* Immediate refill */
1479 rbi->skb = new_skb;
1480 rbi->dma_addr = new_dma_addr;
1481 rxd->addr = cpu_to_le64(rbi->dma_addr);
1482 rxd->len = rbi->len;
1483 }
1484
1485 #ifdef VMXNET3_RSS
1486 if (rcd->rssType != VMXNET3_RCD_RSS_TYPE_NONE &&
1487 (adapter->netdev->features & NETIF_F_RXHASH))
1488 skb_set_hash(ctx->skb,
1489 le32_to_cpu(rcd->rssHash),
1490 PKT_HASH_TYPE_L3);
1491 #endif
1492 skb_put(ctx->skb, rcd->len);
1493
1494 if (VMXNET3_VERSION_GE_2(adapter) &&
1495 rcd->type == VMXNET3_CDTYPE_RXCOMP_LRO) {
1496 struct Vmxnet3_RxCompDescExt *rcdlro;
1497 rcdlro = (struct Vmxnet3_RxCompDescExt *)rcd;
1498
1499 segCnt = rcdlro->segCnt;
1500 WARN_ON_ONCE(segCnt == 0);
1501 mss = rcdlro->mss;
1502 if (unlikely(segCnt <= 1))
1503 segCnt = 0;
1504 } else {
1505 segCnt = 0;
1506 }
1507 } else {
1508 BUG_ON(ctx->skb == NULL && !skip_page_frags);
1509
1510 /* non SOP buffer must be type 1 in most cases */
1511 BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_PAGE);
1512 BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_BODY);
1513
1514 /* If an sop buffer was dropped, skip all
1515 * following non-sop fragments. They will be reused.
1516 */
1517 if (skip_page_frags)
1518 goto rcd_done;
1519
1520 if (rcd->len) {
1521 new_page = alloc_page(GFP_ATOMIC);
1522 /* Replacement page frag could not be allocated.
1523 * Reuse this page. Drop the pkt and free the
1524 * skb which contained this page as a frag. Skip
1525 * processing all the following non-sop frags.
1526 */
1527 if (unlikely(!new_page)) {
1528 rq->stats.rx_buf_alloc_failure++;
1529 dev_kfree_skb(ctx->skb);
1530 ctx->skb = NULL;
1531 skip_page_frags = true;
1532 goto rcd_done;
1533 }
1534 new_dma_addr = dma_map_page(&adapter->pdev->dev,
1535 new_page,
1536 0, PAGE_SIZE,
1537 PCI_DMA_FROMDEVICE);
1538 if (dma_mapping_error(&adapter->pdev->dev,
1539 new_dma_addr)) {
1540 put_page(new_page);
1541 rq->stats.rx_buf_alloc_failure++;
1542 dev_kfree_skb(ctx->skb);
1543 ctx->skb = NULL;
1544 skip_page_frags = true;
1545 goto rcd_done;
1546 }
1547
1548 dma_unmap_page(&adapter->pdev->dev,
1549 rbi->dma_addr, rbi->len,
1550 PCI_DMA_FROMDEVICE);
1551
1552 vmxnet3_append_frag(ctx->skb, rcd, rbi);
1553
1554 /* Immediate refill */
1555 rbi->page = new_page;
1556 rbi->dma_addr = new_dma_addr;
1557 rxd->addr = cpu_to_le64(rbi->dma_addr);
1558 rxd->len = rbi->len;
1559 }
1560 }
1561
1562
1563 skb = ctx->skb;
1564 if (rcd->eop) {
1565 u32 mtu = adapter->netdev->mtu;
1566 skb->len += skb->data_len;
1567
1568 vmxnet3_rx_csum(adapter, skb,
1569 (union Vmxnet3_GenericDesc *)rcd);
1570 skb->protocol = eth_type_trans(skb, adapter->netdev);
1571 if (!rcd->tcp ||
1572 !(adapter->netdev->features & NETIF_F_LRO))
1573 goto not_lro;
1574
1575 if (segCnt != 0 && mss != 0) {
1576 skb_shinfo(skb)->gso_type = rcd->v4 ?
1577 SKB_GSO_TCPV4 : SKB_GSO_TCPV6;
1578 skb_shinfo(skb)->gso_size = mss;
1579 skb_shinfo(skb)->gso_segs = segCnt;
1580 } else if (segCnt != 0 || skb->len > mtu) {
1581 u32 hlen;
1582
1583 hlen = vmxnet3_get_hdr_len(adapter, skb,
1584 (union Vmxnet3_GenericDesc *)rcd);
1585 if (hlen == 0)
1586 goto not_lro;
1587
1588 skb_shinfo(skb)->gso_type =
1589 rcd->v4 ? SKB_GSO_TCPV4 : SKB_GSO_TCPV6;
1590 if (segCnt != 0) {
1591 skb_shinfo(skb)->gso_segs = segCnt;
1592 skb_shinfo(skb)->gso_size =
1593 DIV_ROUND_UP(skb->len -
1594 hlen, segCnt);
1595 } else {
1596 skb_shinfo(skb)->gso_size = mtu - hlen;
1597 }
1598 }
1599 not_lro:
1600 if (unlikely(rcd->ts))
1601 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), rcd->tci);
1602
1603 if (adapter->netdev->features & NETIF_F_LRO)
1604 netif_receive_skb(skb);
1605 else
1606 napi_gro_receive(&rq->napi, skb);
1607
1608 ctx->skb = NULL;
1609 num_pkts++;
1610 }
1611
1612 rcd_done:
1613 /* device may have skipped some rx descs */
1614 ring->next2comp = idx;
1615 num_to_alloc = vmxnet3_cmd_ring_desc_avail(ring);
1616 ring = rq->rx_ring + ring_idx;
1617
1618 /* Ensure that the writes to rxd->gen bits will be observed
1619 * after all other writes to rxd objects.
1620 */
1621 dma_wmb();
1622
1623 while (num_to_alloc) {
1624 vmxnet3_getRxDesc(rxd, &ring->base[ring->next2fill].rxd,
1625 &rxCmdDesc);
1626 BUG_ON(!rxd->addr);
1627
1628 /* Recv desc is ready to be used by the device */
1629 rxd->gen = ring->gen;
1630 vmxnet3_cmd_ring_adv_next2fill(ring);
1631 num_to_alloc--;
1632 }
1633
1634 /* if needed, update the register */
1635 if (unlikely(rq->shared->updateRxProd)) {
1636 VMXNET3_WRITE_BAR0_REG(adapter,
1637 rxprod_reg[ring_idx] + rq->qid * 8,
1638 ring->next2fill);
1639 }
1640
1641 vmxnet3_comp_ring_adv_next2proc(&rq->comp_ring);
1642 vmxnet3_getRxComp(rcd,
1643 &rq->comp_ring.base[rq->comp_ring.next2proc].rcd, &rxComp);
1644 }
1645
1646 return num_pkts;
1647 }
1648
1649
1650 static void
1651 vmxnet3_rq_cleanup(struct vmxnet3_rx_queue *rq,
1652 struct vmxnet3_adapter *adapter)
1653 {
1654 u32 i, ring_idx;
1655 struct Vmxnet3_RxDesc *rxd;
1656
1657 for (ring_idx = 0; ring_idx < 2; ring_idx++) {
1658 for (i = 0; i < rq->rx_ring[ring_idx].size; i++) {
1659 #ifdef __BIG_ENDIAN_BITFIELD
1660 struct Vmxnet3_RxDesc rxDesc;
1661 #endif
1662 vmxnet3_getRxDesc(rxd,
1663 &rq->rx_ring[ring_idx].base[i].rxd, &rxDesc);
1664
1665 if (rxd->btype == VMXNET3_RXD_BTYPE_HEAD &&
1666 rq->buf_info[ring_idx][i].skb) {
1667 dma_unmap_single(&adapter->pdev->dev, rxd->addr,
1668 rxd->len, PCI_DMA_FROMDEVICE);
1669 dev_kfree_skb(rq->buf_info[ring_idx][i].skb);
1670 rq->buf_info[ring_idx][i].skb = NULL;
1671 } else if (rxd->btype == VMXNET3_RXD_BTYPE_BODY &&
1672 rq->buf_info[ring_idx][i].page) {
1673 dma_unmap_page(&adapter->pdev->dev, rxd->addr,
1674 rxd->len, PCI_DMA_FROMDEVICE);
1675 put_page(rq->buf_info[ring_idx][i].page);
1676 rq->buf_info[ring_idx][i].page = NULL;
1677 }
1678 }
1679
1680 rq->rx_ring[ring_idx].gen = VMXNET3_INIT_GEN;
1681 rq->rx_ring[ring_idx].next2fill =
1682 rq->rx_ring[ring_idx].next2comp = 0;
1683 }
1684
1685 rq->comp_ring.gen = VMXNET3_INIT_GEN;
1686 rq->comp_ring.next2proc = 0;
1687 }
1688
1689
1690 static void
1691 vmxnet3_rq_cleanup_all(struct vmxnet3_adapter *adapter)
1692 {
1693 int i;
1694
1695 for (i = 0; i < adapter->num_rx_queues; i++)
1696 vmxnet3_rq_cleanup(&adapter->rx_queue[i], adapter);
1697 }
1698
1699
1700 static void vmxnet3_rq_destroy(struct vmxnet3_rx_queue *rq,
1701 struct vmxnet3_adapter *adapter)
1702 {
1703 int i;
1704 int j;
1705
1706 /* all rx buffers must have already been freed */
1707 for (i = 0; i < 2; i++) {
1708 if (rq->buf_info[i]) {
1709 for (j = 0; j < rq->rx_ring[i].size; j++)
1710 BUG_ON(rq->buf_info[i][j].page != NULL);
1711 }
1712 }
1713
1714
1715 for (i = 0; i < 2; i++) {
1716 if (rq->rx_ring[i].base) {
1717 dma_free_coherent(&adapter->pdev->dev,
1718 rq->rx_ring[i].size
1719 * sizeof(struct Vmxnet3_RxDesc),
1720 rq->rx_ring[i].base,
1721 rq->rx_ring[i].basePA);
1722 rq->rx_ring[i].base = NULL;
1723 }
1724 }
1725
1726 if (rq->data_ring.base) {
1727 dma_free_coherent(&adapter->pdev->dev,
1728 rq->rx_ring[0].size * rq->data_ring.desc_size,
1729 rq->data_ring.base, rq->data_ring.basePA);
1730 rq->data_ring.base = NULL;
1731 }
1732
1733 if (rq->comp_ring.base) {
1734 dma_free_coherent(&adapter->pdev->dev, rq->comp_ring.size
1735 * sizeof(struct Vmxnet3_RxCompDesc),
1736 rq->comp_ring.base, rq->comp_ring.basePA);
1737 rq->comp_ring.base = NULL;
1738 }
1739
1740 if (rq->buf_info[0]) {
1741 size_t sz = sizeof(struct vmxnet3_rx_buf_info) *
1742 (rq->rx_ring[0].size + rq->rx_ring[1].size);
1743 dma_free_coherent(&adapter->pdev->dev, sz, rq->buf_info[0],
1744 rq->buf_info_pa);
1745 rq->buf_info[0] = rq->buf_info[1] = NULL;
1746 }
1747 }
1748
1749 static void
1750 vmxnet3_rq_destroy_all_rxdataring(struct vmxnet3_adapter *adapter)
1751 {
1752 int i;
1753
1754 for (i = 0; i < adapter->num_rx_queues; i++) {
1755 struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
1756
1757 if (rq->data_ring.base) {
1758 dma_free_coherent(&adapter->pdev->dev,
1759 (rq->rx_ring[0].size *
1760 rq->data_ring.desc_size),
1761 rq->data_ring.base,
1762 rq->data_ring.basePA);
1763 rq->data_ring.base = NULL;
1764 rq->data_ring.desc_size = 0;
1765 }
1766 }
1767 }
1768
1769 static int
1770 vmxnet3_rq_init(struct vmxnet3_rx_queue *rq,
1771 struct vmxnet3_adapter *adapter)
1772 {
1773 int i;
1774
1775 /* initialize buf_info */
1776 for (i = 0; i < rq->rx_ring[0].size; i++) {
1777
1778 /* 1st buf for a pkt is skbuff */
1779 if (i % adapter->rx_buf_per_pkt == 0) {
1780 rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_SKB;
1781 rq->buf_info[0][i].len = adapter->skb_buf_size;
1782 } else { /* subsequent bufs for a pkt is frag */
1783 rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_PAGE;
1784 rq->buf_info[0][i].len = PAGE_SIZE;
1785 }
1786 }
1787 for (i = 0; i < rq->rx_ring[1].size; i++) {
1788 rq->buf_info[1][i].buf_type = VMXNET3_RX_BUF_PAGE;
1789 rq->buf_info[1][i].len = PAGE_SIZE;
1790 }
1791
1792 /* reset internal state and allocate buffers for both rings */
1793 for (i = 0; i < 2; i++) {
1794 rq->rx_ring[i].next2fill = rq->rx_ring[i].next2comp = 0;
1795
1796 memset(rq->rx_ring[i].base, 0, rq->rx_ring[i].size *
1797 sizeof(struct Vmxnet3_RxDesc));
1798 rq->rx_ring[i].gen = VMXNET3_INIT_GEN;
1799 }
1800 if (vmxnet3_rq_alloc_rx_buf(rq, 0, rq->rx_ring[0].size - 1,
1801 adapter) == 0) {
1802 /* at least has 1 rx buffer for the 1st ring */
1803 return -ENOMEM;
1804 }
1805 vmxnet3_rq_alloc_rx_buf(rq, 1, rq->rx_ring[1].size - 1, adapter);
1806
1807 /* reset the comp ring */
1808 rq->comp_ring.next2proc = 0;
1809 memset(rq->comp_ring.base, 0, rq->comp_ring.size *
1810 sizeof(struct Vmxnet3_RxCompDesc));
1811 rq->comp_ring.gen = VMXNET3_INIT_GEN;
1812
1813 /* reset rxctx */
1814 rq->rx_ctx.skb = NULL;
1815
1816 /* stats are not reset */
1817 return 0;
1818 }
1819
1820
1821 static int
1822 vmxnet3_rq_init_all(struct vmxnet3_adapter *adapter)
1823 {
1824 int i, err = 0;
1825
1826 for (i = 0; i < adapter->num_rx_queues; i++) {
1827 err = vmxnet3_rq_init(&adapter->rx_queue[i], adapter);
1828 if (unlikely(err)) {
1829 dev_err(&adapter->netdev->dev, "%s: failed to "
1830 "initialize rx queue%i\n",
1831 adapter->netdev->name, i);
1832 break;
1833 }
1834 }
1835 return err;
1836
1837 }
1838
1839
1840 static int
1841 vmxnet3_rq_create(struct vmxnet3_rx_queue *rq, struct vmxnet3_adapter *adapter)
1842 {
1843 int i;
1844 size_t sz;
1845 struct vmxnet3_rx_buf_info *bi;
1846
1847 for (i = 0; i < 2; i++) {
1848
1849 sz = rq->rx_ring[i].size * sizeof(struct Vmxnet3_RxDesc);
1850 rq->rx_ring[i].base = dma_alloc_coherent(
1851 &adapter->pdev->dev, sz,
1852 &rq->rx_ring[i].basePA,
1853 GFP_KERNEL);
1854 if (!rq->rx_ring[i].base) {
1855 netdev_err(adapter->netdev,
1856 "failed to allocate rx ring %d\n", i);
1857 goto err;
1858 }
1859 }
1860
1861 if ((adapter->rxdataring_enabled) && (rq->data_ring.desc_size != 0)) {
1862 sz = rq->rx_ring[0].size * rq->data_ring.desc_size;
1863 rq->data_ring.base =
1864 dma_alloc_coherent(&adapter->pdev->dev, sz,
1865 &rq->data_ring.basePA,
1866 GFP_KERNEL);
1867 if (!rq->data_ring.base) {
1868 netdev_err(adapter->netdev,
1869 "rx data ring will be disabled\n");
1870 adapter->rxdataring_enabled = false;
1871 }
1872 } else {
1873 rq->data_ring.base = NULL;
1874 rq->data_ring.desc_size = 0;
1875 }
1876
1877 sz = rq->comp_ring.size * sizeof(struct Vmxnet3_RxCompDesc);
1878 rq->comp_ring.base = dma_alloc_coherent(&adapter->pdev->dev, sz,
1879 &rq->comp_ring.basePA,
1880 GFP_KERNEL);
1881 if (!rq->comp_ring.base) {
1882 netdev_err(adapter->netdev, "failed to allocate rx comp ring\n");
1883 goto err;
1884 }
1885
1886 sz = sizeof(struct vmxnet3_rx_buf_info) * (rq->rx_ring[0].size +
1887 rq->rx_ring[1].size);
1888 bi = dma_alloc_coherent(&adapter->pdev->dev, sz, &rq->buf_info_pa,
1889 GFP_KERNEL);
1890 if (!bi)
1891 goto err;
1892
1893 rq->buf_info[0] = bi;
1894 rq->buf_info[1] = bi + rq->rx_ring[0].size;
1895
1896 return 0;
1897
1898 err:
1899 vmxnet3_rq_destroy(rq, adapter);
1900 return -ENOMEM;
1901 }
1902
1903
1904 static int
1905 vmxnet3_rq_create_all(struct vmxnet3_adapter *adapter)
1906 {
1907 int i, err = 0;
1908
1909 adapter->rxdataring_enabled = VMXNET3_VERSION_GE_3(adapter);
1910
1911 for (i = 0; i < adapter->num_rx_queues; i++) {
1912 err = vmxnet3_rq_create(&adapter->rx_queue[i], adapter);
1913 if (unlikely(err)) {
1914 dev_err(&adapter->netdev->dev,
1915 "%s: failed to create rx queue%i\n",
1916 adapter->netdev->name, i);
1917 goto err_out;
1918 }
1919 }
1920
1921 if (!adapter->rxdataring_enabled)
1922 vmxnet3_rq_destroy_all_rxdataring(adapter);
1923
1924 return err;
1925 err_out:
1926 vmxnet3_rq_destroy_all(adapter);
1927 return err;
1928
1929 }
1930
1931 /* Multiple queue aware polling function for tx and rx */
1932
1933 static int
1934 vmxnet3_do_poll(struct vmxnet3_adapter *adapter, int budget)
1935 {
1936 int rcd_done = 0, i;
1937 if (unlikely(adapter->shared->ecr))
1938 vmxnet3_process_events(adapter);
1939 for (i = 0; i < adapter->num_tx_queues; i++)
1940 vmxnet3_tq_tx_complete(&adapter->tx_queue[i], adapter);
1941
1942 for (i = 0; i < adapter->num_rx_queues; i++)
1943 rcd_done += vmxnet3_rq_rx_complete(&adapter->rx_queue[i],
1944 adapter, budget);
1945 return rcd_done;
1946 }
1947
1948
1949 static int
1950 vmxnet3_poll(struct napi_struct *napi, int budget)
1951 {
1952 struct vmxnet3_rx_queue *rx_queue = container_of(napi,
1953 struct vmxnet3_rx_queue, napi);
1954 int rxd_done;
1955
1956 rxd_done = vmxnet3_do_poll(rx_queue->adapter, budget);
1957
1958 if (rxd_done < budget) {
1959 napi_complete_done(napi, rxd_done);
1960 vmxnet3_enable_all_intrs(rx_queue->adapter);
1961 }
1962 return rxd_done;
1963 }
1964
1965 /*
1966 * NAPI polling function for MSI-X mode with multiple Rx queues
1967 * Returns the # of the NAPI credit consumed (# of rx descriptors processed)
1968 */
1969
1970 static int
1971 vmxnet3_poll_rx_only(struct napi_struct *napi, int budget)
1972 {
1973 struct vmxnet3_rx_queue *rq = container_of(napi,
1974 struct vmxnet3_rx_queue, napi);
1975 struct vmxnet3_adapter *adapter = rq->adapter;
1976 int rxd_done;
1977
1978 /* When sharing interrupt with corresponding tx queue, process
1979 * tx completions in that queue as well
1980 */
1981 if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE) {
1982 struct vmxnet3_tx_queue *tq =
1983 &adapter->tx_queue[rq - adapter->rx_queue];
1984 vmxnet3_tq_tx_complete(tq, adapter);
1985 }
1986
1987 rxd_done = vmxnet3_rq_rx_complete(rq, adapter, budget);
1988
1989 if (rxd_done < budget) {
1990 napi_complete_done(napi, rxd_done);
1991 vmxnet3_enable_intr(adapter, rq->comp_ring.intr_idx);
1992 }
1993 return rxd_done;
1994 }
1995
1996
1997 #ifdef CONFIG_PCI_MSI
1998
1999 /*
2000 * Handle completion interrupts on tx queues
2001 * Returns whether or not the intr is handled
2002 */
2003
2004 static irqreturn_t
2005 vmxnet3_msix_tx(int irq, void *data)
2006 {
2007 struct vmxnet3_tx_queue *tq = data;
2008 struct vmxnet3_adapter *adapter = tq->adapter;
2009
2010 if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
2011 vmxnet3_disable_intr(adapter, tq->comp_ring.intr_idx);
2012
2013 /* Handle the case where only one irq is allocate for all tx queues */
2014 if (adapter->share_intr == VMXNET3_INTR_TXSHARE) {
2015 int i;
2016 for (i = 0; i < adapter->num_tx_queues; i++) {
2017 struct vmxnet3_tx_queue *txq = &adapter->tx_queue[i];
2018 vmxnet3_tq_tx_complete(txq, adapter);
2019 }
2020 } else {
2021 vmxnet3_tq_tx_complete(tq, adapter);
2022 }
2023 vmxnet3_enable_intr(adapter, tq->comp_ring.intr_idx);
2024
2025 return IRQ_HANDLED;
2026 }
2027
2028
2029 /*
2030 * Handle completion interrupts on rx queues. Returns whether or not the
2031 * intr is handled
2032 */
2033
2034 static irqreturn_t
2035 vmxnet3_msix_rx(int irq, void *data)
2036 {
2037 struct vmxnet3_rx_queue *rq = data;
2038 struct vmxnet3_adapter *adapter = rq->adapter;
2039
2040 /* disable intr if needed */
2041 if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
2042 vmxnet3_disable_intr(adapter, rq->comp_ring.intr_idx);
2043 napi_schedule(&rq->napi);
2044
2045 return IRQ_HANDLED;
2046 }
2047
2048 /*
2049 *----------------------------------------------------------------------------
2050 *
2051 * vmxnet3_msix_event --
2052 *
2053 * vmxnet3 msix event intr handler
2054 *
2055 * Result:
2056 * whether or not the intr is handled
2057 *
2058 *----------------------------------------------------------------------------
2059 */
2060
2061 static irqreturn_t
2062 vmxnet3_msix_event(int irq, void *data)
2063 {
2064 struct net_device *dev = data;
2065 struct vmxnet3_adapter *adapter = netdev_priv(dev);
2066
2067 /* disable intr if needed */
2068 if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
2069 vmxnet3_disable_intr(adapter, adapter->intr.event_intr_idx);
2070
2071 if (adapter->shared->ecr)
2072 vmxnet3_process_events(adapter);
2073
2074 vmxnet3_enable_intr(adapter, adapter->intr.event_intr_idx);
2075
2076 return IRQ_HANDLED;
2077 }
2078
2079 #endif /* CONFIG_PCI_MSI */
2080
2081
2082 /* Interrupt handler for vmxnet3 */
2083 static irqreturn_t
2084 vmxnet3_intr(int irq, void *dev_id)
2085 {
2086 struct net_device *dev = dev_id;
2087 struct vmxnet3_adapter *adapter = netdev_priv(dev);
2088
2089 if (adapter->intr.type == VMXNET3_IT_INTX) {
2090 u32 icr = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_ICR);
2091 if (unlikely(icr == 0))
2092 /* not ours */
2093 return IRQ_NONE;
2094 }
2095
2096
2097 /* disable intr if needed */
2098 if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
2099 vmxnet3_disable_all_intrs(adapter);
2100
2101 napi_schedule(&adapter->rx_queue[0].napi);
2102
2103 return IRQ_HANDLED;
2104 }
2105
2106 #ifdef CONFIG_NET_POLL_CONTROLLER
2107
2108 /* netpoll callback. */
2109 static void
2110 vmxnet3_netpoll(struct net_device *netdev)
2111 {
2112 struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2113
2114 switch (adapter->intr.type) {
2115 #ifdef CONFIG_PCI_MSI
2116 case VMXNET3_IT_MSIX: {
2117 int i;
2118 for (i = 0; i < adapter->num_rx_queues; i++)
2119 vmxnet3_msix_rx(0, &adapter->rx_queue[i]);
2120 break;
2121 }
2122 #endif
2123 case VMXNET3_IT_MSI:
2124 default:
2125 vmxnet3_intr(0, adapter->netdev);
2126 break;
2127 }
2128
2129 }
2130 #endif /* CONFIG_NET_POLL_CONTROLLER */
2131
2132 static int
2133 vmxnet3_request_irqs(struct vmxnet3_adapter *adapter)
2134 {
2135 struct vmxnet3_intr *intr = &adapter->intr;
2136 int err = 0, i;
2137 int vector = 0;
2138
2139 #ifdef CONFIG_PCI_MSI
2140 if (adapter->intr.type == VMXNET3_IT_MSIX) {
2141 for (i = 0; i < adapter->num_tx_queues; i++) {
2142 if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE) {
2143 sprintf(adapter->tx_queue[i].name, "%s-tx-%d",
2144 adapter->netdev->name, vector);
2145 err = request_irq(
2146 intr->msix_entries[vector].vector,
2147 vmxnet3_msix_tx, 0,
2148 adapter->tx_queue[i].name,
2149 &adapter->tx_queue[i]);
2150 } else {
2151 sprintf(adapter->tx_queue[i].name, "%s-rxtx-%d",
2152 adapter->netdev->name, vector);
2153 }
2154 if (err) {
2155 dev_err(&adapter->netdev->dev,
2156 "Failed to request irq for MSIX, %s, "
2157 "error %d\n",
2158 adapter->tx_queue[i].name, err);
2159 return err;
2160 }
2161
2162 /* Handle the case where only 1 MSIx was allocated for
2163 * all tx queues */
2164 if (adapter->share_intr == VMXNET3_INTR_TXSHARE) {
2165 for (; i < adapter->num_tx_queues; i++)
2166 adapter->tx_queue[i].comp_ring.intr_idx
2167 = vector;
2168 vector++;
2169 break;
2170 } else {
2171 adapter->tx_queue[i].comp_ring.intr_idx
2172 = vector++;
2173 }
2174 }
2175 if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE)
2176 vector = 0;
2177
2178 for (i = 0; i < adapter->num_rx_queues; i++) {
2179 if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE)
2180 sprintf(adapter->rx_queue[i].name, "%s-rx-%d",
2181 adapter->netdev->name, vector);
2182 else
2183 sprintf(adapter->rx_queue[i].name, "%s-rxtx-%d",
2184 adapter->netdev->name, vector);
2185 err = request_irq(intr->msix_entries[vector].vector,
2186 vmxnet3_msix_rx, 0,
2187 adapter->rx_queue[i].name,
2188 &(adapter->rx_queue[i]));
2189 if (err) {
2190 netdev_err(adapter->netdev,
2191 "Failed to request irq for MSIX, "
2192 "%s, error %d\n",
2193 adapter->rx_queue[i].name, err);
2194 return err;
2195 }
2196
2197 adapter->rx_queue[i].comp_ring.intr_idx = vector++;
2198 }
2199
2200 sprintf(intr->event_msi_vector_name, "%s-event-%d",
2201 adapter->netdev->name, vector);
2202 err = request_irq(intr->msix_entries[vector].vector,
2203 vmxnet3_msix_event, 0,
2204 intr->event_msi_vector_name, adapter->netdev);
2205 intr->event_intr_idx = vector;
2206
2207 } else if (intr->type == VMXNET3_IT_MSI) {
2208 adapter->num_rx_queues = 1;
2209 err = request_irq(adapter->pdev->irq, vmxnet3_intr, 0,
2210 adapter->netdev->name, adapter->netdev);
2211 } else {
2212 #endif
2213 adapter->num_rx_queues = 1;
2214 err = request_irq(adapter->pdev->irq, vmxnet3_intr,
2215 IRQF_SHARED, adapter->netdev->name,
2216 adapter->netdev);
2217 #ifdef CONFIG_PCI_MSI
2218 }
2219 #endif
2220 intr->num_intrs = vector + 1;
2221 if (err) {
2222 netdev_err(adapter->netdev,
2223 "Failed to request irq (intr type:%d), error %d\n",
2224 intr->type, err);
2225 } else {
2226 /* Number of rx queues will not change after this */
2227 for (i = 0; i < adapter->num_rx_queues; i++) {
2228 struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
2229 rq->qid = i;
2230 rq->qid2 = i + adapter->num_rx_queues;
2231 rq->dataRingQid = i + 2 * adapter->num_rx_queues;
2232 }
2233
2234 /* init our intr settings */
2235 for (i = 0; i < intr->num_intrs; i++)
2236 intr->mod_levels[i] = UPT1_IML_ADAPTIVE;
2237 if (adapter->intr.type != VMXNET3_IT_MSIX) {
2238 adapter->intr.event_intr_idx = 0;
2239 for (i = 0; i < adapter->num_tx_queues; i++)
2240 adapter->tx_queue[i].comp_ring.intr_idx = 0;
2241 adapter->rx_queue[0].comp_ring.intr_idx = 0;
2242 }
2243
2244 netdev_info(adapter->netdev,
2245 "intr type %u, mode %u, %u vectors allocated\n",
2246 intr->type, intr->mask_mode, intr->num_intrs);
2247 }
2248
2249 return err;
2250 }
2251
2252
2253 static void
2254 vmxnet3_free_irqs(struct vmxnet3_adapter *adapter)
2255 {
2256 struct vmxnet3_intr *intr = &adapter->intr;
2257 BUG_ON(intr->type == VMXNET3_IT_AUTO || intr->num_intrs <= 0);
2258
2259 switch (intr->type) {
2260 #ifdef CONFIG_PCI_MSI
2261 case VMXNET3_IT_MSIX:
2262 {
2263 int i, vector = 0;
2264
2265 if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE) {
2266 for (i = 0; i < adapter->num_tx_queues; i++) {
2267 free_irq(intr->msix_entries[vector++].vector,
2268 &(adapter->tx_queue[i]));
2269 if (adapter->share_intr == VMXNET3_INTR_TXSHARE)
2270 break;
2271 }
2272 }
2273
2274 for (i = 0; i < adapter->num_rx_queues; i++) {
2275 free_irq(intr->msix_entries[vector++].vector,
2276 &(adapter->rx_queue[i]));
2277 }
2278
2279 free_irq(intr->msix_entries[vector].vector,
2280 adapter->netdev);
2281 BUG_ON(vector >= intr->num_intrs);
2282 break;
2283 }
2284 #endif
2285 case VMXNET3_IT_MSI:
2286 free_irq(adapter->pdev->irq, adapter->netdev);
2287 break;
2288 case VMXNET3_IT_INTX:
2289 free_irq(adapter->pdev->irq, adapter->netdev);
2290 break;
2291 default:
2292 BUG();
2293 }
2294 }
2295
2296
2297 static void
2298 vmxnet3_restore_vlan(struct vmxnet3_adapter *adapter)
2299 {
2300 u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
2301 u16 vid;
2302
2303 /* allow untagged pkts */
2304 VMXNET3_SET_VFTABLE_ENTRY(vfTable, 0);
2305
2306 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
2307 VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid);
2308 }
2309
2310
2311 static int
2312 vmxnet3_vlan_rx_add_vid(struct net_device *netdev, __be16 proto, u16 vid)
2313 {
2314 struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2315
2316 if (!(netdev->flags & IFF_PROMISC)) {
2317 u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
2318 unsigned long flags;
2319
2320 VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid);
2321 spin_lock_irqsave(&adapter->cmd_lock, flags);
2322 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2323 VMXNET3_CMD_UPDATE_VLAN_FILTERS);
2324 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2325 }
2326
2327 set_bit(vid, adapter->active_vlans);
2328
2329 return 0;
2330 }
2331
2332
2333 static int
2334 vmxnet3_vlan_rx_kill_vid(struct net_device *netdev, __be16 proto, u16 vid)
2335 {
2336 struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2337
2338 if (!(netdev->flags & IFF_PROMISC)) {
2339 u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
2340 unsigned long flags;
2341
2342 VMXNET3_CLEAR_VFTABLE_ENTRY(vfTable, vid);
2343 spin_lock_irqsave(&adapter->cmd_lock, flags);
2344 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2345 VMXNET3_CMD_UPDATE_VLAN_FILTERS);
2346 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2347 }
2348
2349 clear_bit(vid, adapter->active_vlans);
2350
2351 return 0;
2352 }
2353
2354
2355 static u8 *
2356 vmxnet3_copy_mc(struct net_device *netdev)
2357 {
2358 u8 *buf = NULL;
2359 u32 sz = netdev_mc_count(netdev) * ETH_ALEN;
2360
2361 /* struct Vmxnet3_RxFilterConf.mfTableLen is u16. */
2362 if (sz <= 0xffff) {
2363 /* We may be called with BH disabled */
2364 buf = kmalloc(sz, GFP_ATOMIC);
2365 if (buf) {
2366 struct netdev_hw_addr *ha;
2367 int i = 0;
2368
2369 netdev_for_each_mc_addr(ha, netdev)
2370 memcpy(buf + i++ * ETH_ALEN, ha->addr,
2371 ETH_ALEN);
2372 }
2373 }
2374 return buf;
2375 }
2376
2377
2378 static void
2379 vmxnet3_set_mc(struct net_device *netdev)
2380 {
2381 struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2382 unsigned long flags;
2383 struct Vmxnet3_RxFilterConf *rxConf =
2384 &adapter->shared->devRead.rxFilterConf;
2385 u8 *new_table = NULL;
2386 dma_addr_t new_table_pa = 0;
2387 bool new_table_pa_valid = false;
2388 u32 new_mode = VMXNET3_RXM_UCAST;
2389
2390 if (netdev->flags & IFF_PROMISC) {
2391 u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
2392 memset(vfTable, 0, VMXNET3_VFT_SIZE * sizeof(*vfTable));
2393
2394 new_mode |= VMXNET3_RXM_PROMISC;
2395 } else {
2396 vmxnet3_restore_vlan(adapter);
2397 }
2398
2399 if (netdev->flags & IFF_BROADCAST)
2400 new_mode |= VMXNET3_RXM_BCAST;
2401
2402 if (netdev->flags & IFF_ALLMULTI)
2403 new_mode |= VMXNET3_RXM_ALL_MULTI;
2404 else
2405 if (!netdev_mc_empty(netdev)) {
2406 new_table = vmxnet3_copy_mc(netdev);
2407 if (new_table) {
2408 size_t sz = netdev_mc_count(netdev) * ETH_ALEN;
2409
2410 rxConf->mfTableLen = cpu_to_le16(sz);
2411 new_table_pa = dma_map_single(
2412 &adapter->pdev->dev,
2413 new_table,
2414 sz,
2415 PCI_DMA_TODEVICE);
2416 if (!dma_mapping_error(&adapter->pdev->dev,
2417 new_table_pa)) {
2418 new_mode |= VMXNET3_RXM_MCAST;
2419 new_table_pa_valid = true;
2420 rxConf->mfTablePA = cpu_to_le64(
2421 new_table_pa);
2422 }
2423 }
2424 if (!new_table_pa_valid) {
2425 netdev_info(netdev,
2426 "failed to copy mcast list, setting ALL_MULTI\n");
2427 new_mode |= VMXNET3_RXM_ALL_MULTI;
2428 }
2429 }
2430
2431 if (!(new_mode & VMXNET3_RXM_MCAST)) {
2432 rxConf->mfTableLen = 0;
2433 rxConf->mfTablePA = 0;
2434 }
2435
2436 spin_lock_irqsave(&adapter->cmd_lock, flags);
2437 if (new_mode != rxConf->rxMode) {
2438 rxConf->rxMode = cpu_to_le32(new_mode);
2439 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2440 VMXNET3_CMD_UPDATE_RX_MODE);
2441 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2442 VMXNET3_CMD_UPDATE_VLAN_FILTERS);
2443 }
2444
2445 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2446 VMXNET3_CMD_UPDATE_MAC_FILTERS);
2447 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2448
2449 if (new_table_pa_valid)
2450 dma_unmap_single(&adapter->pdev->dev, new_table_pa,
2451 rxConf->mfTableLen, PCI_DMA_TODEVICE);
2452 kfree(new_table);
2453 }
2454
2455 void
2456 vmxnet3_rq_destroy_all(struct vmxnet3_adapter *adapter)
2457 {
2458 int i;
2459
2460 for (i = 0; i < adapter->num_rx_queues; i++)
2461 vmxnet3_rq_destroy(&adapter->rx_queue[i], adapter);
2462 }
2463
2464
2465 /*
2466 * Set up driver_shared based on settings in adapter.
2467 */
2468
2469 static void
2470 vmxnet3_setup_driver_shared(struct vmxnet3_adapter *adapter)
2471 {
2472 struct Vmxnet3_DriverShared *shared = adapter->shared;
2473 struct Vmxnet3_DSDevRead *devRead = &shared->devRead;
2474 struct Vmxnet3_TxQueueConf *tqc;
2475 struct Vmxnet3_RxQueueConf *rqc;
2476 int i;
2477
2478 memset(shared, 0, sizeof(*shared));
2479
2480 /* driver settings */
2481 shared->magic = cpu_to_le32(VMXNET3_REV1_MAGIC);
2482 devRead->misc.driverInfo.version = cpu_to_le32(
2483 VMXNET3_DRIVER_VERSION_NUM);
2484 devRead->misc.driverInfo.gos.gosBits = (sizeof(void *) == 4 ?
2485 VMXNET3_GOS_BITS_32 : VMXNET3_GOS_BITS_64);
2486 devRead->misc.driverInfo.gos.gosType = VMXNET3_GOS_TYPE_LINUX;
2487 *((u32 *)&devRead->misc.driverInfo.gos) = cpu_to_le32(
2488 *((u32 *)&devRead->misc.driverInfo.gos));
2489 devRead->misc.driverInfo.vmxnet3RevSpt = cpu_to_le32(1);
2490 devRead->misc.driverInfo.uptVerSpt = cpu_to_le32(1);
2491
2492 devRead->misc.ddPA = cpu_to_le64(adapter->adapter_pa);
2493 devRead->misc.ddLen = cpu_to_le32(sizeof(struct vmxnet3_adapter));
2494
2495 /* set up feature flags */
2496 if (adapter->netdev->features & NETIF_F_RXCSUM)
2497 devRead->misc.uptFeatures |= UPT1_F_RXCSUM;
2498
2499 if (adapter->netdev->features & NETIF_F_LRO) {
2500 devRead->misc.uptFeatures |= UPT1_F_LRO;
2501 devRead->misc.maxNumRxSG = cpu_to_le16(1 + MAX_SKB_FRAGS);
2502 }
2503 if (adapter->netdev->features & NETIF_F_HW_VLAN_CTAG_RX)
2504 devRead->misc.uptFeatures |= UPT1_F_RXVLAN;
2505
2506 if (adapter->netdev->features & (NETIF_F_GSO_UDP_TUNNEL |
2507 NETIF_F_GSO_UDP_TUNNEL_CSUM))
2508 devRead->misc.uptFeatures |= UPT1_F_RXINNEROFLD;
2509
2510 devRead->misc.mtu = cpu_to_le32(adapter->netdev->mtu);
2511 devRead->misc.queueDescPA = cpu_to_le64(adapter->queue_desc_pa);
2512 devRead->misc.queueDescLen = cpu_to_le32(
2513 adapter->num_tx_queues * sizeof(struct Vmxnet3_TxQueueDesc) +
2514 adapter->num_rx_queues * sizeof(struct Vmxnet3_RxQueueDesc));
2515
2516 /* tx queue settings */
2517 devRead->misc.numTxQueues = adapter->num_tx_queues;
2518 for (i = 0; i < adapter->num_tx_queues; i++) {
2519 struct vmxnet3_tx_queue *tq = &adapter->tx_queue[i];
2520 BUG_ON(adapter->tx_queue[i].tx_ring.base == NULL);
2521 tqc = &adapter->tqd_start[i].conf;
2522 tqc->txRingBasePA = cpu_to_le64(tq->tx_ring.basePA);
2523 tqc->dataRingBasePA = cpu_to_le64(tq->data_ring.basePA);
2524 tqc->compRingBasePA = cpu_to_le64(tq->comp_ring.basePA);
2525 tqc->ddPA = cpu_to_le64(tq->buf_info_pa);
2526 tqc->txRingSize = cpu_to_le32(tq->tx_ring.size);
2527 tqc->dataRingSize = cpu_to_le32(tq->data_ring.size);
2528 tqc->txDataRingDescSize = cpu_to_le32(tq->txdata_desc_size);
2529 tqc->compRingSize = cpu_to_le32(tq->comp_ring.size);
2530 tqc->ddLen = cpu_to_le32(
2531 sizeof(struct vmxnet3_tx_buf_info) *
2532 tqc->txRingSize);
2533 tqc->intrIdx = tq->comp_ring.intr_idx;
2534 }
2535
2536 /* rx queue settings */
2537 devRead->misc.numRxQueues = adapter->num_rx_queues;
2538 for (i = 0; i < adapter->num_rx_queues; i++) {
2539 struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
2540 rqc = &adapter->rqd_start[i].conf;
2541 rqc->rxRingBasePA[0] = cpu_to_le64(rq->rx_ring[0].basePA);
2542 rqc->rxRingBasePA[1] = cpu_to_le64(rq->rx_ring[1].basePA);
2543 rqc->compRingBasePA = cpu_to_le64(rq->comp_ring.basePA);
2544 rqc->ddPA = cpu_to_le64(rq->buf_info_pa);
2545 rqc->rxRingSize[0] = cpu_to_le32(rq->rx_ring[0].size);
2546 rqc->rxRingSize[1] = cpu_to_le32(rq->rx_ring[1].size);
2547 rqc->compRingSize = cpu_to_le32(rq->comp_ring.size);
2548 rqc->ddLen = cpu_to_le32(
2549 sizeof(struct vmxnet3_rx_buf_info) *
2550 (rqc->rxRingSize[0] +
2551 rqc->rxRingSize[1]));
2552 rqc->intrIdx = rq->comp_ring.intr_idx;
2553 if (VMXNET3_VERSION_GE_3(adapter)) {
2554 rqc->rxDataRingBasePA =
2555 cpu_to_le64(rq->data_ring.basePA);
2556 rqc->rxDataRingDescSize =
2557 cpu_to_le16(rq->data_ring.desc_size);
2558 }
2559 }
2560
2561 #ifdef VMXNET3_RSS
2562 memset(adapter->rss_conf, 0, sizeof(*adapter->rss_conf));
2563
2564 if (adapter->rss) {
2565 struct UPT1_RSSConf *rssConf = adapter->rss_conf;
2566
2567 devRead->misc.uptFeatures |= UPT1_F_RSS;
2568 devRead->misc.numRxQueues = adapter->num_rx_queues;
2569 rssConf->hashType = UPT1_RSS_HASH_TYPE_TCP_IPV4 |
2570 UPT1_RSS_HASH_TYPE_IPV4 |
2571 UPT1_RSS_HASH_TYPE_TCP_IPV6 |
2572 UPT1_RSS_HASH_TYPE_IPV6;
2573 rssConf->hashFunc = UPT1_RSS_HASH_FUNC_TOEPLITZ;
2574 rssConf->hashKeySize = UPT1_RSS_MAX_KEY_SIZE;
2575 rssConf->indTableSize = VMXNET3_RSS_IND_TABLE_SIZE;
2576 netdev_rss_key_fill(rssConf->hashKey, sizeof(rssConf->hashKey));
2577
2578 for (i = 0; i < rssConf->indTableSize; i++)
2579 rssConf->indTable[i] = ethtool_rxfh_indir_default(
2580 i, adapter->num_rx_queues);
2581
2582 devRead->rssConfDesc.confVer = 1;
2583 devRead->rssConfDesc.confLen = cpu_to_le32(sizeof(*rssConf));
2584 devRead->rssConfDesc.confPA =
2585 cpu_to_le64(adapter->rss_conf_pa);
2586 }
2587
2588 #endif /* VMXNET3_RSS */
2589
2590 /* intr settings */
2591 devRead->intrConf.autoMask = adapter->intr.mask_mode ==
2592 VMXNET3_IMM_AUTO;
2593 devRead->intrConf.numIntrs = adapter->intr.num_intrs;
2594 for (i = 0; i < adapter->intr.num_intrs; i++)
2595 devRead->intrConf.modLevels[i] = adapter->intr.mod_levels[i];
2596
2597 devRead->intrConf.eventIntrIdx = adapter->intr.event_intr_idx;
2598 devRead->intrConf.intrCtrl |= cpu_to_le32(VMXNET3_IC_DISABLE_ALL);
2599
2600 /* rx filter settings */
2601 devRead->rxFilterConf.rxMode = 0;
2602 vmxnet3_restore_vlan(adapter);
2603 vmxnet3_write_mac_addr(adapter, adapter->netdev->dev_addr);
2604
2605 /* the rest are already zeroed */
2606 }
2607
2608 static void
2609 vmxnet3_init_coalesce(struct vmxnet3_adapter *adapter)
2610 {
2611 struct Vmxnet3_DriverShared *shared = adapter->shared;
2612 union Vmxnet3_CmdInfo *cmdInfo = &shared->cu.cmdInfo;
2613 unsigned long flags;
2614
2615 if (!VMXNET3_VERSION_GE_3(adapter))
2616 return;
2617
2618 spin_lock_irqsave(&adapter->cmd_lock, flags);
2619 cmdInfo->varConf.confVer = 1;
2620 cmdInfo->varConf.confLen =
2621 cpu_to_le32(sizeof(*adapter->coal_conf));
2622 cmdInfo->varConf.confPA = cpu_to_le64(adapter->coal_conf_pa);
2623
2624 if (adapter->default_coal_mode) {
2625 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2626 VMXNET3_CMD_GET_COALESCE);
2627 } else {
2628 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2629 VMXNET3_CMD_SET_COALESCE);
2630 }
2631
2632 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2633 }
2634
2635 static void
2636 vmxnet3_init_rssfields(struct vmxnet3_adapter *adapter)
2637 {
2638 struct Vmxnet3_DriverShared *shared = adapter->shared;
2639 union Vmxnet3_CmdInfo *cmdInfo = &shared->cu.cmdInfo;
2640 unsigned long flags;
2641
2642 if (!VMXNET3_VERSION_GE_4(adapter))
2643 return;
2644
2645 spin_lock_irqsave(&adapter->cmd_lock, flags);
2646
2647 if (adapter->default_rss_fields) {
2648 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2649 VMXNET3_CMD_GET_RSS_FIELDS);
2650 adapter->rss_fields =
2651 VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
2652 } else {
2653 cmdInfo->setRssFields = adapter->rss_fields;
2654 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2655 VMXNET3_CMD_SET_RSS_FIELDS);
2656 /* Not all requested RSS may get applied, so get and
2657 * cache what was actually applied.
2658 */
2659 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2660 VMXNET3_CMD_GET_RSS_FIELDS);
2661 adapter->rss_fields =
2662 VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
2663 }
2664
2665 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2666 }
2667
2668 int
2669 vmxnet3_activate_dev(struct vmxnet3_adapter *adapter)
2670 {
2671 int err, i;
2672 u32 ret;
2673 unsigned long flags;
2674
2675 netdev_dbg(adapter->netdev, "%s: skb_buf_size %d, rx_buf_per_pkt %d,"
2676 " ring sizes %u %u %u\n", adapter->netdev->name,
2677 adapter->skb_buf_size, adapter->rx_buf_per_pkt,
2678 adapter->tx_queue[0].tx_ring.size,
2679 adapter->rx_queue[0].rx_ring[0].size,
2680 adapter->rx_queue[0].rx_ring[1].size);
2681
2682 vmxnet3_tq_init_all(adapter);
2683 err = vmxnet3_rq_init_all(adapter);
2684 if (err) {
2685 netdev_err(adapter->netdev,
2686 "Failed to init rx queue error %d\n", err);
2687 goto rq_err;
2688 }
2689
2690 err = vmxnet3_request_irqs(adapter);
2691 if (err) {
2692 netdev_err(adapter->netdev,
2693 "Failed to setup irq for error %d\n", err);
2694 goto irq_err;
2695 }
2696
2697 vmxnet3_setup_driver_shared(adapter);
2698
2699 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, VMXNET3_GET_ADDR_LO(
2700 adapter->shared_pa));
2701 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, VMXNET3_GET_ADDR_HI(
2702 adapter->shared_pa));
2703 spin_lock_irqsave(&adapter->cmd_lock, flags);
2704 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2705 VMXNET3_CMD_ACTIVATE_DEV);
2706 ret = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
2707 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2708
2709 if (ret != 0) {
2710 netdev_err(adapter->netdev,
2711 "Failed to activate dev: error %u\n", ret);
2712 err = -EINVAL;
2713 goto activate_err;
2714 }
2715
2716 vmxnet3_init_coalesce(adapter);
2717 vmxnet3_init_rssfields(adapter);
2718
2719 for (i = 0; i < adapter->num_rx_queues; i++) {
2720 VMXNET3_WRITE_BAR0_REG(adapter,
2721 VMXNET3_REG_RXPROD + i * VMXNET3_REG_ALIGN,
2722 adapter->rx_queue[i].rx_ring[0].next2fill);
2723 VMXNET3_WRITE_BAR0_REG(adapter, (VMXNET3_REG_RXPROD2 +
2724 (i * VMXNET3_REG_ALIGN)),
2725 adapter->rx_queue[i].rx_ring[1].next2fill);
2726 }
2727
2728 /* Apply the rx filter settins last. */
2729 vmxnet3_set_mc(adapter->netdev);
2730
2731 /*
2732 * Check link state when first activating device. It will start the
2733 * tx queue if the link is up.
2734 */
2735 vmxnet3_check_link(adapter, true);
2736 for (i = 0; i < adapter->num_rx_queues; i++)
2737 napi_enable(&adapter->rx_queue[i].napi);
2738 vmxnet3_enable_all_intrs(adapter);
2739 clear_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
2740 return 0;
2741
2742 activate_err:
2743 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, 0);
2744 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, 0);
2745 vmxnet3_free_irqs(adapter);
2746 irq_err:
2747 rq_err:
2748 /* free up buffers we allocated */
2749 vmxnet3_rq_cleanup_all(adapter);
2750 return err;
2751 }
2752
2753
2754 void
2755 vmxnet3_reset_dev(struct vmxnet3_adapter *adapter)
2756 {
2757 unsigned long flags;
2758 spin_lock_irqsave(&adapter->cmd_lock, flags);
2759 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, VMXNET3_CMD_RESET_DEV);
2760 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2761 }
2762
2763
2764 int
2765 vmxnet3_quiesce_dev(struct vmxnet3_adapter *adapter)
2766 {
2767 int i;
2768 unsigned long flags;
2769 if (test_and_set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state))
2770 return 0;
2771
2772
2773 spin_lock_irqsave(&adapter->cmd_lock, flags);
2774 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2775 VMXNET3_CMD_QUIESCE_DEV);
2776 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2777 vmxnet3_disable_all_intrs(adapter);
2778
2779 for (i = 0; i < adapter->num_rx_queues; i++)
2780 napi_disable(&adapter->rx_queue[i].napi);
2781 netif_tx_disable(adapter->netdev);
2782 adapter->link_speed = 0;
2783 netif_carrier_off(adapter->netdev);
2784
2785 vmxnet3_tq_cleanup_all(adapter);
2786 vmxnet3_rq_cleanup_all(adapter);
2787 vmxnet3_free_irqs(adapter);
2788 return 0;
2789 }
2790
2791
2792 static void
2793 vmxnet3_write_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac)
2794 {
2795 u32 tmp;
2796
2797 tmp = *(u32 *)mac;
2798 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACL, tmp);
2799
2800 tmp = (mac[5] << 8) | mac[4];
2801 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACH, tmp);
2802 }
2803
2804
2805 static int
2806 vmxnet3_set_mac_addr(struct net_device *netdev, void *p)
2807 {
2808 struct sockaddr *addr = p;
2809 struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2810
2811 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2812 vmxnet3_write_mac_addr(adapter, addr->sa_data);
2813
2814 return 0;
2815 }
2816
2817
2818 /* ==================== initialization and cleanup routines ============ */
2819
2820 static int
2821 vmxnet3_alloc_pci_resources(struct vmxnet3_adapter *adapter)
2822 {
2823 int err;
2824 unsigned long mmio_start, mmio_len;
2825 struct pci_dev *pdev = adapter->pdev;
2826
2827 err = pci_enable_device(pdev);
2828 if (err) {
2829 dev_err(&pdev->dev, "Failed to enable adapter: error %d\n", err);
2830 return err;
2831 }
2832
2833 err = pci_request_selected_regions(pdev, (1 << 2) - 1,
2834 vmxnet3_driver_name);
2835 if (err) {
2836 dev_err(&pdev->dev,
2837 "Failed to request region for adapter: error %d\n", err);
2838 goto err_enable_device;
2839 }
2840
2841 pci_set_master(pdev);
2842
2843 mmio_start = pci_resource_start(pdev, 0);
2844 mmio_len = pci_resource_len(pdev, 0);
2845 adapter->hw_addr0 = ioremap(mmio_start, mmio_len);
2846 if (!adapter->hw_addr0) {
2847 dev_err(&pdev->dev, "Failed to map bar0\n");
2848 err = -EIO;
2849 goto err_ioremap;
2850 }
2851
2852 mmio_start = pci_resource_start(pdev, 1);
2853 mmio_len = pci_resource_len(pdev, 1);
2854 adapter->hw_addr1 = ioremap(mmio_start, mmio_len);
2855 if (!adapter->hw_addr1) {
2856 dev_err(&pdev->dev, "Failed to map bar1\n");
2857 err = -EIO;
2858 goto err_bar1;
2859 }
2860 return 0;
2861
2862 err_bar1:
2863 iounmap(adapter->hw_addr0);
2864 err_ioremap:
2865 pci_release_selected_regions(pdev, (1 << 2) - 1);
2866 err_enable_device:
2867 pci_disable_device(pdev);
2868 return err;
2869 }
2870
2871
2872 static void
2873 vmxnet3_free_pci_resources(struct vmxnet3_adapter *adapter)
2874 {
2875 BUG_ON(!adapter->pdev);
2876
2877 iounmap(adapter->hw_addr0);
2878 iounmap(adapter->hw_addr1);
2879 pci_release_selected_regions(adapter->pdev, (1 << 2) - 1);
2880 pci_disable_device(adapter->pdev);
2881 }
2882
2883
2884 static void
2885 vmxnet3_adjust_rx_ring_size(struct vmxnet3_adapter *adapter)
2886 {
2887 size_t sz, i, ring0_size, ring1_size, comp_size;
2888 if (adapter->netdev->mtu <= VMXNET3_MAX_SKB_BUF_SIZE -
2889 VMXNET3_MAX_ETH_HDR_SIZE) {
2890 adapter->skb_buf_size = adapter->netdev->mtu +
2891 VMXNET3_MAX_ETH_HDR_SIZE;
2892 if (adapter->skb_buf_size < VMXNET3_MIN_T0_BUF_SIZE)
2893 adapter->skb_buf_size = VMXNET3_MIN_T0_BUF_SIZE;
2894
2895 adapter->rx_buf_per_pkt = 1;
2896 } else {
2897 adapter->skb_buf_size = VMXNET3_MAX_SKB_BUF_SIZE;
2898 sz = adapter->netdev->mtu - VMXNET3_MAX_SKB_BUF_SIZE +
2899 VMXNET3_MAX_ETH_HDR_SIZE;
2900 adapter->rx_buf_per_pkt = 1 + (sz + PAGE_SIZE - 1) / PAGE_SIZE;
2901 }
2902
2903 /*
2904 * for simplicity, force the ring0 size to be a multiple of
2905 * rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN
2906 */
2907 sz = adapter->rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN;
2908 ring0_size = adapter->rx_queue[0].rx_ring[0].size;
2909 ring0_size = (ring0_size + sz - 1) / sz * sz;
2910 ring0_size = min_t(u32, ring0_size, VMXNET3_RX_RING_MAX_SIZE /
2911 sz * sz);
2912 ring1_size = adapter->rx_queue[0].rx_ring[1].size;
2913 ring1_size = (ring1_size + sz - 1) / sz * sz;
2914 ring1_size = min_t(u32, ring1_size, VMXNET3_RX_RING2_MAX_SIZE /
2915 sz * sz);
2916 comp_size = ring0_size + ring1_size;
2917
2918 for (i = 0; i < adapter->num_rx_queues; i++) {
2919 struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
2920
2921 rq->rx_ring[0].size = ring0_size;
2922 rq->rx_ring[1].size = ring1_size;
2923 rq->comp_ring.size = comp_size;
2924 }
2925 }
2926
2927
2928 int
2929 vmxnet3_create_queues(struct vmxnet3_adapter *adapter, u32 tx_ring_size,
2930 u32 rx_ring_size, u32 rx_ring2_size,
2931 u16 txdata_desc_size, u16 rxdata_desc_size)
2932 {
2933 int err = 0, i;
2934
2935 for (i = 0; i < adapter->num_tx_queues; i++) {
2936 struct vmxnet3_tx_queue *tq = &adapter->tx_queue[i];
2937 tq->tx_ring.size = tx_ring_size;
2938 tq->data_ring.size = tx_ring_size;
2939 tq->comp_ring.size = tx_ring_size;
2940 tq->txdata_desc_size = txdata_desc_size;
2941 tq->shared = &adapter->tqd_start[i].ctrl;
2942 tq->stopped = true;
2943 tq->adapter = adapter;
2944 tq->qid = i;
2945 err = vmxnet3_tq_create(tq, adapter);
2946 /*
2947 * Too late to change num_tx_queues. We cannot do away with
2948 * lesser number of queues than what we asked for
2949 */
2950 if (err)
2951 goto queue_err;
2952 }
2953
2954 adapter->rx_queue[0].rx_ring[0].size = rx_ring_size;
2955 adapter->rx_queue[0].rx_ring[1].size = rx_ring2_size;
2956 vmxnet3_adjust_rx_ring_size(adapter);
2957
2958 adapter->rxdataring_enabled = VMXNET3_VERSION_GE_3(adapter);
2959 for (i = 0; i < adapter->num_rx_queues; i++) {
2960 struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
2961 /* qid and qid2 for rx queues will be assigned later when num
2962 * of rx queues is finalized after allocating intrs */
2963 rq->shared = &adapter->rqd_start[i].ctrl;
2964 rq->adapter = adapter;
2965 rq->data_ring.desc_size = rxdata_desc_size;
2966 err = vmxnet3_rq_create(rq, adapter);
2967 if (err) {
2968 if (i == 0) {
2969 netdev_err(adapter->netdev,
2970 "Could not allocate any rx queues. "
2971 "Aborting.\n");
2972 goto queue_err;
2973 } else {
2974 netdev_info(adapter->netdev,
2975 "Number of rx queues changed "
2976 "to : %d.\n", i);
2977 adapter->num_rx_queues = i;
2978 err = 0;
2979 break;
2980 }
2981 }
2982 }
2983
2984 if (!adapter->rxdataring_enabled)
2985 vmxnet3_rq_destroy_all_rxdataring(adapter);
2986
2987 return err;
2988 queue_err:
2989 vmxnet3_tq_destroy_all(adapter);
2990 return err;
2991 }
2992
2993 static int
2994 vmxnet3_open(struct net_device *netdev)
2995 {
2996 struct vmxnet3_adapter *adapter;
2997 int err, i;
2998
2999 adapter = netdev_priv(netdev);
3000
3001 for (i = 0; i < adapter->num_tx_queues; i++)
3002 spin_lock_init(&adapter->tx_queue[i].tx_lock);
3003
3004 if (VMXNET3_VERSION_GE_3(adapter)) {
3005 unsigned long flags;
3006 u16 txdata_desc_size;
3007
3008 spin_lock_irqsave(&adapter->cmd_lock, flags);
3009 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3010 VMXNET3_CMD_GET_TXDATA_DESC_SIZE);
3011 txdata_desc_size = VMXNET3_READ_BAR1_REG(adapter,
3012 VMXNET3_REG_CMD);
3013 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3014
3015 if ((txdata_desc_size < VMXNET3_TXDATA_DESC_MIN_SIZE) ||
3016 (txdata_desc_size > VMXNET3_TXDATA_DESC_MAX_SIZE) ||
3017 (txdata_desc_size & VMXNET3_TXDATA_DESC_SIZE_MASK)) {
3018 adapter->txdata_desc_size =
3019 sizeof(struct Vmxnet3_TxDataDesc);
3020 } else {
3021 adapter->txdata_desc_size = txdata_desc_size;
3022 }
3023 } else {
3024 adapter->txdata_desc_size = sizeof(struct Vmxnet3_TxDataDesc);
3025 }
3026
3027 err = vmxnet3_create_queues(adapter,
3028 adapter->tx_ring_size,
3029 adapter->rx_ring_size,
3030 adapter->rx_ring2_size,
3031 adapter->txdata_desc_size,
3032 adapter->rxdata_desc_size);
3033 if (err)
3034 goto queue_err;
3035
3036 err = vmxnet3_activate_dev(adapter);
3037 if (err)
3038 goto activate_err;
3039
3040 return 0;
3041
3042 activate_err:
3043 vmxnet3_rq_destroy_all(adapter);
3044 vmxnet3_tq_destroy_all(adapter);
3045 queue_err:
3046 return err;
3047 }
3048
3049
3050 static int
3051 vmxnet3_close(struct net_device *netdev)
3052 {
3053 struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3054
3055 /*
3056 * Reset_work may be in the middle of resetting the device, wait for its
3057 * completion.
3058 */
3059 while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
3060 usleep_range(1000, 2000);
3061
3062 vmxnet3_quiesce_dev(adapter);
3063
3064 vmxnet3_rq_destroy_all(adapter);
3065 vmxnet3_tq_destroy_all(adapter);
3066
3067 clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
3068
3069
3070 return 0;
3071 }
3072
3073
3074 void
3075 vmxnet3_force_close(struct vmxnet3_adapter *adapter)
3076 {
3077 int i;
3078
3079 /*
3080 * we must clear VMXNET3_STATE_BIT_RESETTING, otherwise
3081 * vmxnet3_close() will deadlock.
3082 */
3083 BUG_ON(test_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state));
3084
3085 /* we need to enable NAPI, otherwise dev_close will deadlock */
3086 for (i = 0; i < adapter->num_rx_queues; i++)
3087 napi_enable(&adapter->rx_queue[i].napi);
3088 /*
3089 * Need to clear the quiesce bit to ensure that vmxnet3_close
3090 * can quiesce the device properly
3091 */
3092 clear_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
3093 dev_close(adapter->netdev);
3094 }
3095
3096
3097 static int
3098 vmxnet3_change_mtu(struct net_device *netdev, int new_mtu)
3099 {
3100 struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3101 int err = 0;
3102
3103 netdev->mtu = new_mtu;
3104
3105 /*
3106 * Reset_work may be in the middle of resetting the device, wait for its
3107 * completion.
3108 */
3109 while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
3110 usleep_range(1000, 2000);
3111
3112 if (netif_running(netdev)) {
3113 vmxnet3_quiesce_dev(adapter);
3114 vmxnet3_reset_dev(adapter);
3115
3116 /* we need to re-create the rx queue based on the new mtu */
3117 vmxnet3_rq_destroy_all(adapter);
3118 vmxnet3_adjust_rx_ring_size(adapter);
3119 err = vmxnet3_rq_create_all(adapter);
3120 if (err) {
3121 netdev_err(netdev,
3122 "failed to re-create rx queues, "
3123 " error %d. Closing it.\n", err);
3124 goto out;
3125 }
3126
3127 err = vmxnet3_activate_dev(adapter);
3128 if (err) {
3129 netdev_err(netdev,
3130 "failed to re-activate, error %d. "
3131 "Closing it\n", err);
3132 goto out;
3133 }
3134 }
3135
3136 out:
3137 clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
3138 if (err)
3139 vmxnet3_force_close(adapter);
3140
3141 return err;
3142 }
3143
3144
3145 static void
3146 vmxnet3_declare_features(struct vmxnet3_adapter *adapter, bool dma64)
3147 {
3148 struct net_device *netdev = adapter->netdev;
3149
3150 netdev->hw_features = NETIF_F_SG | NETIF_F_RXCSUM |
3151 NETIF_F_HW_CSUM | NETIF_F_HW_VLAN_CTAG_TX |
3152 NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_TSO | NETIF_F_TSO6 |
3153 NETIF_F_LRO;
3154
3155 if (VMXNET3_VERSION_GE_4(adapter)) {
3156 netdev->hw_features |= NETIF_F_GSO_UDP_TUNNEL |
3157 NETIF_F_GSO_UDP_TUNNEL_CSUM;
3158
3159 netdev->hw_enc_features = NETIF_F_SG | NETIF_F_RXCSUM |
3160 NETIF_F_HW_CSUM | NETIF_F_HW_VLAN_CTAG_TX |
3161 NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_TSO | NETIF_F_TSO6 |
3162 NETIF_F_LRO | NETIF_F_GSO_UDP_TUNNEL |
3163 NETIF_F_GSO_UDP_TUNNEL_CSUM;
3164 }
3165
3166 if (dma64)
3167 netdev->hw_features |= NETIF_F_HIGHDMA;
3168 netdev->vlan_features = netdev->hw_features &
3169 ~(NETIF_F_HW_VLAN_CTAG_TX |
3170 NETIF_F_HW_VLAN_CTAG_RX);
3171 netdev->features = netdev->hw_features | NETIF_F_HW_VLAN_CTAG_FILTER;
3172 }
3173
3174
3175 static void
3176 vmxnet3_read_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac)
3177 {
3178 u32 tmp;
3179
3180 tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACL);
3181 *(u32 *)mac = tmp;
3182
3183 tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACH);
3184 mac[4] = tmp & 0xff;
3185 mac[5] = (tmp >> 8) & 0xff;
3186 }
3187
3188 #ifdef CONFIG_PCI_MSI
3189
3190 /*
3191 * Enable MSIx vectors.
3192 * Returns :
3193 * VMXNET3_LINUX_MIN_MSIX_VECT when only minimum number of vectors required
3194 * were enabled.
3195 * number of vectors which were enabled otherwise (this number is greater
3196 * than VMXNET3_LINUX_MIN_MSIX_VECT)
3197 */
3198
3199 static int
3200 vmxnet3_acquire_msix_vectors(struct vmxnet3_adapter *adapter, int nvec)
3201 {
3202 int ret = pci_enable_msix_range(adapter->pdev,
3203 adapter->intr.msix_entries, nvec, nvec);
3204
3205 if (ret == -ENOSPC && nvec > VMXNET3_LINUX_MIN_MSIX_VECT) {
3206 dev_err(&adapter->netdev->dev,
3207 "Failed to enable %d MSI-X, trying %d\n",
3208 nvec, VMXNET3_LINUX_MIN_MSIX_VECT);
3209
3210 ret = pci_enable_msix_range(adapter->pdev,
3211 adapter->intr.msix_entries,
3212 VMXNET3_LINUX_MIN_MSIX_VECT,
3213 VMXNET3_LINUX_MIN_MSIX_VECT);
3214 }
3215
3216 if (ret < 0) {
3217 dev_err(&adapter->netdev->dev,
3218 "Failed to enable MSI-X, error: %d\n", ret);
3219 }
3220
3221 return ret;
3222 }
3223
3224
3225 #endif /* CONFIG_PCI_MSI */
3226
3227 static void
3228 vmxnet3_alloc_intr_resources(struct vmxnet3_adapter *adapter)
3229 {
3230 u32 cfg;
3231 unsigned long flags;
3232
3233 /* intr settings */
3234 spin_lock_irqsave(&adapter->cmd_lock, flags);
3235 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3236 VMXNET3_CMD_GET_CONF_INTR);
3237 cfg = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
3238 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3239 adapter->intr.type = cfg & 0x3;
3240 adapter->intr.mask_mode = (cfg >> 2) & 0x3;
3241
3242 if (adapter->intr.type == VMXNET3_IT_AUTO) {
3243 adapter->intr.type = VMXNET3_IT_MSIX;
3244 }
3245
3246 #ifdef CONFIG_PCI_MSI
3247 if (adapter->intr.type == VMXNET3_IT_MSIX) {
3248 int i, nvec;
3249
3250 nvec = adapter->share_intr == VMXNET3_INTR_TXSHARE ?
3251 1 : adapter->num_tx_queues;
3252 nvec += adapter->share_intr == VMXNET3_INTR_BUDDYSHARE ?
3253 0 : adapter->num_rx_queues;
3254 nvec += 1; /* for link event */
3255 nvec = nvec > VMXNET3_LINUX_MIN_MSIX_VECT ?
3256 nvec : VMXNET3_LINUX_MIN_MSIX_VECT;
3257
3258 for (i = 0; i < nvec; i++)
3259 adapter->intr.msix_entries[i].entry = i;
3260
3261 nvec = vmxnet3_acquire_msix_vectors(adapter, nvec);
3262 if (nvec < 0)
3263 goto msix_err;
3264
3265 /* If we cannot allocate one MSIx vector per queue
3266 * then limit the number of rx queues to 1
3267 */
3268 if (nvec == VMXNET3_LINUX_MIN_MSIX_VECT) {
3269 if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE
3270 || adapter->num_rx_queues != 1) {
3271 adapter->share_intr = VMXNET3_INTR_TXSHARE;
3272 netdev_err(adapter->netdev,
3273 "Number of rx queues : 1\n");
3274 adapter->num_rx_queues = 1;
3275 }
3276 }
3277
3278 adapter->intr.num_intrs = nvec;
3279 return;
3280
3281 msix_err:
3282 /* If we cannot allocate MSIx vectors use only one rx queue */
3283 dev_info(&adapter->pdev->dev,
3284 "Failed to enable MSI-X, error %d. "
3285 "Limiting #rx queues to 1, try MSI.\n", nvec);
3286
3287 adapter->intr.type = VMXNET3_IT_MSI;
3288 }
3289
3290 if (adapter->intr.type == VMXNET3_IT_MSI) {
3291 if (!pci_enable_msi(adapter->pdev)) {
3292 adapter->num_rx_queues = 1;
3293 adapter->intr.num_intrs = 1;
3294 return;
3295 }
3296 }
3297 #endif /* CONFIG_PCI_MSI */
3298
3299 adapter->num_rx_queues = 1;
3300 dev_info(&adapter->netdev->dev,
3301 "Using INTx interrupt, #Rx queues: 1.\n");
3302 adapter->intr.type = VMXNET3_IT_INTX;
3303
3304 /* INT-X related setting */
3305 adapter->intr.num_intrs = 1;
3306 }
3307
3308
3309 static void
3310 vmxnet3_free_intr_resources(struct vmxnet3_adapter *adapter)
3311 {
3312 if (adapter->intr.type == VMXNET3_IT_MSIX)
3313 pci_disable_msix(adapter->pdev);
3314 else if (adapter->intr.type == VMXNET3_IT_MSI)
3315 pci_disable_msi(adapter->pdev);
3316 else
3317 BUG_ON(adapter->intr.type != VMXNET3_IT_INTX);
3318 }
3319
3320
3321 static void
3322 vmxnet3_tx_timeout(struct net_device *netdev, unsigned int txqueue)
3323 {
3324 struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3325 adapter->tx_timeout_count++;
3326
3327 netdev_err(adapter->netdev, "tx hang\n");
3328 schedule_work(&adapter->work);
3329 }
3330
3331
3332 static void
3333 vmxnet3_reset_work(struct work_struct *data)
3334 {
3335 struct vmxnet3_adapter *adapter;
3336
3337 adapter = container_of(data, struct vmxnet3_adapter, work);
3338
3339 /* if another thread is resetting the device, no need to proceed */
3340 if (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
3341 return;
3342
3343 /* if the device is closed, we must leave it alone */
3344 rtnl_lock();
3345 if (netif_running(adapter->netdev)) {
3346 netdev_notice(adapter->netdev, "resetting\n");
3347 vmxnet3_quiesce_dev(adapter);
3348 vmxnet3_reset_dev(adapter);
3349 vmxnet3_activate_dev(adapter);
3350 } else {
3351 netdev_info(adapter->netdev, "already closed\n");
3352 }
3353 rtnl_unlock();
3354
3355 netif_wake_queue(adapter->netdev);
3356 clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
3357 }
3358
3359
3360 static int
3361 vmxnet3_probe_device(struct pci_dev *pdev,
3362 const struct pci_device_id *id)
3363 {
3364 static const struct net_device_ops vmxnet3_netdev_ops = {
3365 .ndo_open = vmxnet3_open,
3366 .ndo_stop = vmxnet3_close,
3367 .ndo_start_xmit = vmxnet3_xmit_frame,
3368 .ndo_set_mac_address = vmxnet3_set_mac_addr,
3369 .ndo_change_mtu = vmxnet3_change_mtu,
3370 .ndo_fix_features = vmxnet3_fix_features,
3371 .ndo_set_features = vmxnet3_set_features,
3372 .ndo_features_check = vmxnet3_features_check,
3373 .ndo_get_stats64 = vmxnet3_get_stats64,
3374 .ndo_tx_timeout = vmxnet3_tx_timeout,
3375 .ndo_set_rx_mode = vmxnet3_set_mc,
3376 .ndo_vlan_rx_add_vid = vmxnet3_vlan_rx_add_vid,
3377 .ndo_vlan_rx_kill_vid = vmxnet3_vlan_rx_kill_vid,
3378 #ifdef CONFIG_NET_POLL_CONTROLLER
3379 .ndo_poll_controller = vmxnet3_netpoll,
3380 #endif
3381 };
3382 int err;
3383 bool dma64;
3384 u32 ver;
3385 struct net_device *netdev;
3386 struct vmxnet3_adapter *adapter;
3387 u8 mac[ETH_ALEN];
3388 int size;
3389 int num_tx_queues;
3390 int num_rx_queues;
3391
3392 if (!pci_msi_enabled())
3393 enable_mq = 0;
3394
3395 #ifdef VMXNET3_RSS
3396 if (enable_mq)
3397 num_rx_queues = min(VMXNET3_DEVICE_MAX_RX_QUEUES,
3398 (int)num_online_cpus());
3399 else
3400 #endif
3401 num_rx_queues = 1;
3402 num_rx_queues = rounddown_pow_of_two(num_rx_queues);
3403
3404 if (enable_mq)
3405 num_tx_queues = min(VMXNET3_DEVICE_MAX_TX_QUEUES,
3406 (int)num_online_cpus());
3407 else
3408 num_tx_queues = 1;
3409
3410 num_tx_queues = rounddown_pow_of_two(num_tx_queues);
3411 netdev = alloc_etherdev_mq(sizeof(struct vmxnet3_adapter),
3412 max(num_tx_queues, num_rx_queues));
3413 dev_info(&pdev->dev,
3414 "# of Tx queues : %d, # of Rx queues : %d\n",
3415 num_tx_queues, num_rx_queues);
3416
3417 if (!netdev)
3418 return -ENOMEM;
3419
3420 pci_set_drvdata(pdev, netdev);
3421 adapter = netdev_priv(netdev);
3422 adapter->netdev = netdev;
3423 adapter->pdev = pdev;
3424
3425 adapter->tx_ring_size = VMXNET3_DEF_TX_RING_SIZE;
3426 adapter->rx_ring_size = VMXNET3_DEF_RX_RING_SIZE;
3427 adapter->rx_ring2_size = VMXNET3_DEF_RX_RING2_SIZE;
3428
3429 if (pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) == 0) {
3430 if (pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64)) != 0) {
3431 dev_err(&pdev->dev,
3432 "pci_set_consistent_dma_mask failed\n");
3433 err = -EIO;
3434 goto err_set_mask;
3435 }
3436 dma64 = true;
3437 } else {
3438 if (pci_set_dma_mask(pdev, DMA_BIT_MASK(32)) != 0) {
3439 dev_err(&pdev->dev,
3440 "pci_set_dma_mask failed\n");
3441 err = -EIO;
3442 goto err_set_mask;
3443 }
3444 dma64 = false;
3445 }
3446
3447 spin_lock_init(&adapter->cmd_lock);
3448 adapter->adapter_pa = dma_map_single(&adapter->pdev->dev, adapter,
3449 sizeof(struct vmxnet3_adapter),
3450 PCI_DMA_TODEVICE);
3451 if (dma_mapping_error(&adapter->pdev->dev, adapter->adapter_pa)) {
3452 dev_err(&pdev->dev, "Failed to map dma\n");
3453 err = -EFAULT;
3454 goto err_set_mask;
3455 }
3456 adapter->shared = dma_alloc_coherent(
3457 &adapter->pdev->dev,
3458 sizeof(struct Vmxnet3_DriverShared),
3459 &adapter->shared_pa, GFP_KERNEL);
3460 if (!adapter->shared) {
3461 dev_err(&pdev->dev, "Failed to allocate memory\n");
3462 err = -ENOMEM;
3463 goto err_alloc_shared;
3464 }
3465
3466 adapter->num_rx_queues = num_rx_queues;
3467 adapter->num_tx_queues = num_tx_queues;
3468 adapter->rx_buf_per_pkt = 1;
3469
3470 size = sizeof(struct Vmxnet3_TxQueueDesc) * adapter->num_tx_queues;
3471 size += sizeof(struct Vmxnet3_RxQueueDesc) * adapter->num_rx_queues;
3472 adapter->tqd_start = dma_alloc_coherent(&adapter->pdev->dev, size,
3473 &adapter->queue_desc_pa,
3474 GFP_KERNEL);
3475
3476 if (!adapter->tqd_start) {
3477 dev_err(&pdev->dev, "Failed to allocate memory\n");
3478 err = -ENOMEM;
3479 goto err_alloc_queue_desc;
3480 }
3481 adapter->rqd_start = (struct Vmxnet3_RxQueueDesc *)(adapter->tqd_start +
3482 adapter->num_tx_queues);
3483
3484 adapter->pm_conf = dma_alloc_coherent(&adapter->pdev->dev,
3485 sizeof(struct Vmxnet3_PMConf),
3486 &adapter->pm_conf_pa,
3487 GFP_KERNEL);
3488 if (adapter->pm_conf == NULL) {
3489 err = -ENOMEM;
3490 goto err_alloc_pm;
3491 }
3492
3493 #ifdef VMXNET3_RSS
3494
3495 adapter->rss_conf = dma_alloc_coherent(&adapter->pdev->dev,
3496 sizeof(struct UPT1_RSSConf),
3497 &adapter->rss_conf_pa,
3498 GFP_KERNEL);
3499 if (adapter->rss_conf == NULL) {
3500 err = -ENOMEM;
3501 goto err_alloc_rss;
3502 }
3503 #endif /* VMXNET3_RSS */
3504
3505 err = vmxnet3_alloc_pci_resources(adapter);
3506 if (err < 0)
3507 goto err_alloc_pci;
3508
3509 ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_VRRS);
3510 if (ver & (1 << VMXNET3_REV_4)) {
3511 VMXNET3_WRITE_BAR1_REG(adapter,
3512 VMXNET3_REG_VRRS,
3513 1 << VMXNET3_REV_4);
3514 adapter->version = VMXNET3_REV_4 + 1;
3515 } else if (ver & (1 << VMXNET3_REV_3)) {
3516 VMXNET3_WRITE_BAR1_REG(adapter,
3517 VMXNET3_REG_VRRS,
3518 1 << VMXNET3_REV_3);
3519 adapter->version = VMXNET3_REV_3 + 1;
3520 } else if (ver & (1 << VMXNET3_REV_2)) {
3521 VMXNET3_WRITE_BAR1_REG(adapter,
3522 VMXNET3_REG_VRRS,
3523 1 << VMXNET3_REV_2);
3524 adapter->version = VMXNET3_REV_2 + 1;
3525 } else if (ver & (1 << VMXNET3_REV_1)) {
3526 VMXNET3_WRITE_BAR1_REG(adapter,
3527 VMXNET3_REG_VRRS,
3528 1 << VMXNET3_REV_1);
3529 adapter->version = VMXNET3_REV_1 + 1;
3530 } else {
3531 dev_err(&pdev->dev,
3532 "Incompatible h/w version (0x%x) for adapter\n", ver);
3533 err = -EBUSY;
3534 goto err_ver;
3535 }
3536 dev_dbg(&pdev->dev, "Using device version %d\n", adapter->version);
3537
3538 ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_UVRS);
3539 if (ver & 1) {
3540 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_UVRS, 1);
3541 } else {
3542 dev_err(&pdev->dev,
3543 "Incompatible upt version (0x%x) for adapter\n", ver);
3544 err = -EBUSY;
3545 goto err_ver;
3546 }
3547
3548 if (VMXNET3_VERSION_GE_3(adapter)) {
3549 adapter->coal_conf =
3550 dma_alloc_coherent(&adapter->pdev->dev,
3551 sizeof(struct Vmxnet3_CoalesceScheme)
3552 ,
3553 &adapter->coal_conf_pa,
3554 GFP_KERNEL);
3555 if (!adapter->coal_conf) {
3556 err = -ENOMEM;
3557 goto err_ver;
3558 }
3559 adapter->coal_conf->coalMode = VMXNET3_COALESCE_DISABLED;
3560 adapter->default_coal_mode = true;
3561 }
3562
3563 if (VMXNET3_VERSION_GE_4(adapter)) {
3564 adapter->default_rss_fields = true;
3565 adapter->rss_fields = VMXNET3_RSS_FIELDS_DEFAULT;
3566 }
3567
3568 SET_NETDEV_DEV(netdev, &pdev->dev);
3569 vmxnet3_declare_features(adapter, dma64);
3570
3571 adapter->rxdata_desc_size = VMXNET3_VERSION_GE_3(adapter) ?
3572 VMXNET3_DEF_RXDATA_DESC_SIZE : 0;
3573
3574 if (adapter->num_tx_queues == adapter->num_rx_queues)
3575 adapter->share_intr = VMXNET3_INTR_BUDDYSHARE;
3576 else
3577 adapter->share_intr = VMXNET3_INTR_DONTSHARE;
3578
3579 vmxnet3_alloc_intr_resources(adapter);
3580
3581 #ifdef VMXNET3_RSS
3582 if (adapter->num_rx_queues > 1 &&
3583 adapter->intr.type == VMXNET3_IT_MSIX) {
3584 adapter->rss = true;
3585 netdev->hw_features |= NETIF_F_RXHASH;
3586 netdev->features |= NETIF_F_RXHASH;
3587 dev_dbg(&pdev->dev, "RSS is enabled.\n");
3588 } else {
3589 adapter->rss = false;
3590 }
3591 #endif
3592
3593 vmxnet3_read_mac_addr(adapter, mac);
3594 memcpy(netdev->dev_addr, mac, netdev->addr_len);
3595
3596 netdev->netdev_ops = &vmxnet3_netdev_ops;
3597 vmxnet3_set_ethtool_ops(netdev);
3598 netdev->watchdog_timeo = 5 * HZ;
3599
3600 /* MTU range: 60 - 9000 */
3601 netdev->min_mtu = VMXNET3_MIN_MTU;
3602 netdev->max_mtu = VMXNET3_MAX_MTU;
3603
3604 INIT_WORK(&adapter->work, vmxnet3_reset_work);
3605 set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
3606
3607 if (adapter->intr.type == VMXNET3_IT_MSIX) {
3608 int i;
3609 for (i = 0; i < adapter->num_rx_queues; i++) {
3610 netif_napi_add(adapter->netdev,
3611 &adapter->rx_queue[i].napi,
3612 vmxnet3_poll_rx_only, 64);
3613 }
3614 } else {
3615 netif_napi_add(adapter->netdev, &adapter->rx_queue[0].napi,
3616 vmxnet3_poll, 64);
3617 }
3618
3619 netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
3620 netif_set_real_num_rx_queues(adapter->netdev, adapter->num_rx_queues);
3621
3622 netif_carrier_off(netdev);
3623 err = register_netdev(netdev);
3624
3625 if (err) {
3626 dev_err(&pdev->dev, "Failed to register adapter\n");
3627 goto err_register;
3628 }
3629
3630 vmxnet3_check_link(adapter, false);
3631 return 0;
3632
3633 err_register:
3634 if (VMXNET3_VERSION_GE_3(adapter)) {
3635 dma_free_coherent(&adapter->pdev->dev,
3636 sizeof(struct Vmxnet3_CoalesceScheme),
3637 adapter->coal_conf, adapter->coal_conf_pa);
3638 }
3639 vmxnet3_free_intr_resources(adapter);
3640 err_ver:
3641 vmxnet3_free_pci_resources(adapter);
3642 err_alloc_pci:
3643 #ifdef VMXNET3_RSS
3644 dma_free_coherent(&adapter->pdev->dev, sizeof(struct UPT1_RSSConf),
3645 adapter->rss_conf, adapter->rss_conf_pa);
3646 err_alloc_rss:
3647 #endif
3648 dma_free_coherent(&adapter->pdev->dev, sizeof(struct Vmxnet3_PMConf),
3649 adapter->pm_conf, adapter->pm_conf_pa);
3650 err_alloc_pm:
3651 dma_free_coherent(&adapter->pdev->dev, size, adapter->tqd_start,
3652 adapter->queue_desc_pa);
3653 err_alloc_queue_desc:
3654 dma_free_coherent(&adapter->pdev->dev,
3655 sizeof(struct Vmxnet3_DriverShared),
3656 adapter->shared, adapter->shared_pa);
3657 err_alloc_shared:
3658 dma_unmap_single(&adapter->pdev->dev, adapter->adapter_pa,
3659 sizeof(struct vmxnet3_adapter), PCI_DMA_TODEVICE);
3660 err_set_mask:
3661 free_netdev(netdev);
3662 return err;
3663 }
3664
3665
3666 static void
3667 vmxnet3_remove_device(struct pci_dev *pdev)
3668 {
3669 struct net_device *netdev = pci_get_drvdata(pdev);
3670 struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3671 int size = 0;
3672 int num_rx_queues;
3673
3674 #ifdef VMXNET3_RSS
3675 if (enable_mq)
3676 num_rx_queues = min(VMXNET3_DEVICE_MAX_RX_QUEUES,
3677 (int)num_online_cpus());
3678 else
3679 #endif
3680 num_rx_queues = 1;
3681 num_rx_queues = rounddown_pow_of_two(num_rx_queues);
3682
3683 cancel_work_sync(&adapter->work);
3684
3685 unregister_netdev(netdev);
3686
3687 vmxnet3_free_intr_resources(adapter);
3688 vmxnet3_free_pci_resources(adapter);
3689 if (VMXNET3_VERSION_GE_3(adapter)) {
3690 dma_free_coherent(&adapter->pdev->dev,
3691 sizeof(struct Vmxnet3_CoalesceScheme),
3692 adapter->coal_conf, adapter->coal_conf_pa);
3693 }
3694 #ifdef VMXNET3_RSS
3695 dma_free_coherent(&adapter->pdev->dev, sizeof(struct UPT1_RSSConf),
3696 adapter->rss_conf, adapter->rss_conf_pa);
3697 #endif
3698 dma_free_coherent(&adapter->pdev->dev, sizeof(struct Vmxnet3_PMConf),
3699 adapter->pm_conf, adapter->pm_conf_pa);
3700
3701 size = sizeof(struct Vmxnet3_TxQueueDesc) * adapter->num_tx_queues;
3702 size += sizeof(struct Vmxnet3_RxQueueDesc) * num_rx_queues;
3703 dma_free_coherent(&adapter->pdev->dev, size, adapter->tqd_start,
3704 adapter->queue_desc_pa);
3705 dma_free_coherent(&adapter->pdev->dev,
3706 sizeof(struct Vmxnet3_DriverShared),
3707 adapter->shared, adapter->shared_pa);
3708 dma_unmap_single(&adapter->pdev->dev, adapter->adapter_pa,
3709 sizeof(struct vmxnet3_adapter), PCI_DMA_TODEVICE);
3710 free_netdev(netdev);
3711 }
3712
3713 static void vmxnet3_shutdown_device(struct pci_dev *pdev)
3714 {
3715 struct net_device *netdev = pci_get_drvdata(pdev);
3716 struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3717 unsigned long flags;
3718
3719 /* Reset_work may be in the middle of resetting the device, wait for its
3720 * completion.
3721 */
3722 while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
3723 usleep_range(1000, 2000);
3724
3725 if (test_and_set_bit(VMXNET3_STATE_BIT_QUIESCED,
3726 &adapter->state)) {
3727 clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
3728 return;
3729 }
3730 spin_lock_irqsave(&adapter->cmd_lock, flags);
3731 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3732 VMXNET3_CMD_QUIESCE_DEV);
3733 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3734 vmxnet3_disable_all_intrs(adapter);
3735
3736 clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
3737 }
3738
3739
3740 #ifdef CONFIG_PM
3741
3742 static int
3743 vmxnet3_suspend(struct device *device)
3744 {
3745 struct pci_dev *pdev = to_pci_dev(device);
3746 struct net_device *netdev = pci_get_drvdata(pdev);
3747 struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3748 struct Vmxnet3_PMConf *pmConf;
3749 struct ethhdr *ehdr;
3750 struct arphdr *ahdr;
3751 u8 *arpreq;
3752 struct in_device *in_dev;
3753 struct in_ifaddr *ifa;
3754 unsigned long flags;
3755 int i = 0;
3756
3757 if (!netif_running(netdev))
3758 return 0;
3759
3760 for (i = 0; i < adapter->num_rx_queues; i++)
3761 napi_disable(&adapter->rx_queue[i].napi);
3762
3763 vmxnet3_disable_all_intrs(adapter);
3764 vmxnet3_free_irqs(adapter);
3765 vmxnet3_free_intr_resources(adapter);
3766
3767 netif_device_detach(netdev);
3768 netif_tx_stop_all_queues(netdev);
3769
3770 /* Create wake-up filters. */
3771 pmConf = adapter->pm_conf;
3772 memset(pmConf, 0, sizeof(*pmConf));
3773
3774 if (adapter->wol & WAKE_UCAST) {
3775 pmConf->filters[i].patternSize = ETH_ALEN;
3776 pmConf->filters[i].maskSize = 1;
3777 memcpy(pmConf->filters[i].pattern, netdev->dev_addr, ETH_ALEN);
3778 pmConf->filters[i].mask[0] = 0x3F; /* LSB ETH_ALEN bits */
3779
3780 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_FILTER;
3781 i++;
3782 }
3783
3784 if (adapter->wol & WAKE_ARP) {
3785 rcu_read_lock();
3786
3787 in_dev = __in_dev_get_rcu(netdev);
3788 if (!in_dev) {
3789 rcu_read_unlock();
3790 goto skip_arp;
3791 }
3792
3793 ifa = rcu_dereference(in_dev->ifa_list);
3794 if (!ifa) {
3795 rcu_read_unlock();
3796 goto skip_arp;
3797 }
3798
3799 pmConf->filters[i].patternSize = ETH_HLEN + /* Ethernet header*/
3800 sizeof(struct arphdr) + /* ARP header */
3801 2 * ETH_ALEN + /* 2 Ethernet addresses*/
3802 2 * sizeof(u32); /*2 IPv4 addresses */
3803 pmConf->filters[i].maskSize =
3804 (pmConf->filters[i].patternSize - 1) / 8 + 1;
3805
3806 /* ETH_P_ARP in Ethernet header. */
3807 ehdr = (struct ethhdr *)pmConf->filters[i].pattern;
3808 ehdr->h_proto = htons(ETH_P_ARP);
3809
3810 /* ARPOP_REQUEST in ARP header. */
3811 ahdr = (struct arphdr *)&pmConf->filters[i].pattern[ETH_HLEN];
3812 ahdr->ar_op = htons(ARPOP_REQUEST);
3813 arpreq = (u8 *)(ahdr + 1);
3814
3815 /* The Unicast IPv4 address in 'tip' field. */
3816 arpreq += 2 * ETH_ALEN + sizeof(u32);
3817 *(__be32 *)arpreq = ifa->ifa_address;
3818
3819 rcu_read_unlock();
3820
3821 /* The mask for the relevant bits. */
3822 pmConf->filters[i].mask[0] = 0x00;
3823 pmConf->filters[i].mask[1] = 0x30; /* ETH_P_ARP */
3824 pmConf->filters[i].mask[2] = 0x30; /* ARPOP_REQUEST */
3825 pmConf->filters[i].mask[3] = 0x00;
3826 pmConf->filters[i].mask[4] = 0xC0; /* IPv4 TIP */
3827 pmConf->filters[i].mask[5] = 0x03; /* IPv4 TIP */
3828
3829 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_FILTER;
3830 i++;
3831 }
3832
3833 skip_arp:
3834 if (adapter->wol & WAKE_MAGIC)
3835 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_MAGIC;
3836
3837 pmConf->numFilters = i;
3838
3839 adapter->shared->devRead.pmConfDesc.confVer = cpu_to_le32(1);
3840 adapter->shared->devRead.pmConfDesc.confLen = cpu_to_le32(sizeof(
3841 *pmConf));
3842 adapter->shared->devRead.pmConfDesc.confPA =
3843 cpu_to_le64(adapter->pm_conf_pa);
3844
3845 spin_lock_irqsave(&adapter->cmd_lock, flags);
3846 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3847 VMXNET3_CMD_UPDATE_PMCFG);
3848 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3849
3850 pci_save_state(pdev);
3851 pci_enable_wake(pdev, pci_choose_state(pdev, PMSG_SUSPEND),
3852 adapter->wol);
3853 pci_disable_device(pdev);
3854 pci_set_power_state(pdev, pci_choose_state(pdev, PMSG_SUSPEND));
3855
3856 return 0;
3857 }
3858
3859
3860 static int
3861 vmxnet3_resume(struct device *device)
3862 {
3863 int err;
3864 unsigned long flags;
3865 struct pci_dev *pdev = to_pci_dev(device);
3866 struct net_device *netdev = pci_get_drvdata(pdev);
3867 struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3868
3869 if (!netif_running(netdev))
3870 return 0;
3871
3872 pci_set_power_state(pdev, PCI_D0);
3873 pci_restore_state(pdev);
3874 err = pci_enable_device_mem(pdev);
3875 if (err != 0)
3876 return err;
3877
3878 pci_enable_wake(pdev, PCI_D0, 0);
3879
3880 vmxnet3_alloc_intr_resources(adapter);
3881
3882 /* During hibernate and suspend, device has to be reinitialized as the
3883 * device state need not be preserved.
3884 */
3885
3886 /* Need not check adapter state as other reset tasks cannot run during
3887 * device resume.
3888 */
3889 spin_lock_irqsave(&adapter->cmd_lock, flags);
3890 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3891 VMXNET3_CMD_QUIESCE_DEV);
3892 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3893 vmxnet3_tq_cleanup_all(adapter);
3894 vmxnet3_rq_cleanup_all(adapter);
3895
3896 vmxnet3_reset_dev(adapter);
3897 err = vmxnet3_activate_dev(adapter);
3898 if (err != 0) {
3899 netdev_err(netdev,
3900 "failed to re-activate on resume, error: %d", err);
3901 vmxnet3_force_close(adapter);
3902 return err;
3903 }
3904 netif_device_attach(netdev);
3905
3906 return 0;
3907 }
3908
3909 static const struct dev_pm_ops vmxnet3_pm_ops = {
3910 .suspend = vmxnet3_suspend,
3911 .resume = vmxnet3_resume,
3912 .freeze = vmxnet3_suspend,
3913 .restore = vmxnet3_resume,
3914 };
3915 #endif
3916
3917 static struct pci_driver vmxnet3_driver = {
3918 .name = vmxnet3_driver_name,
3919 .id_table = vmxnet3_pciid_table,
3920 .probe = vmxnet3_probe_device,
3921 .remove = vmxnet3_remove_device,
3922 .shutdown = vmxnet3_shutdown_device,
3923 #ifdef CONFIG_PM
3924 .driver.pm = &vmxnet3_pm_ops,
3925 #endif
3926 };
3927
3928
3929 static int __init
3930 vmxnet3_init_module(void)
3931 {
3932 pr_info("%s - version %s\n", VMXNET3_DRIVER_DESC,
3933 VMXNET3_DRIVER_VERSION_REPORT);
3934 return pci_register_driver(&vmxnet3_driver);
3935 }
3936
3937 module_init(vmxnet3_init_module);
3938
3939
3940 static void
3941 vmxnet3_exit_module(void)
3942 {
3943 pci_unregister_driver(&vmxnet3_driver);
3944 }
3945
3946 module_exit(vmxnet3_exit_module);
3947
3948 MODULE_AUTHOR("VMware, Inc.");
3949 MODULE_DESCRIPTION(VMXNET3_DRIVER_DESC);
3950 MODULE_LICENSE("GPL v2");
3951 MODULE_VERSION(VMXNET3_DRIVER_VERSION_STRING);
Linux with added FPC-III support