1 /*
2 * Linux driver for VMware's vmxnet3 ethernet NIC.
3 *
4 * Copyright (C) 2008-2009, VMware, Inc. All Rights Reserved.
5 *
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License as published by the
8 * Free Software Foundation; version 2 of the License and no later version.
9 *
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
13 * NON INFRINGEMENT. See the GNU General Public License for more
14 * details.
15 *
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * The full GNU General Public License is included in this distribution in
21 * the file called "COPYING".
22 *
23 * Maintained by: Shreyas Bhatewara <pv-drivers@vmware.com>
24 *
25 */
26
27 #include <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 DEFINE_PCI_DEVICE_TABLE(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 atomic_t devices_found;
47
48 #define VMXNET3_MAX_DEVICES 10
49 static int enable_mq = 1;
50 static int irq_share_mode;
51
52 static void
53 vmxnet3_write_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac);
54
55 /*
56 * Enable/Disable the given intr
57 */
58 static void
vmxnet3_enable_intr(struct vmxnet3_adapter * adapter,unsigned intr_idx)59 vmxnet3_enable_intr(struct vmxnet3_adapter *adapter, unsigned intr_idx)
60 {
61 VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_IMR + intr_idx * 8, 0);
62 }
63
64
65 static void
vmxnet3_disable_intr(struct vmxnet3_adapter * adapter,unsigned intr_idx)66 vmxnet3_disable_intr(struct vmxnet3_adapter *adapter, unsigned intr_idx)
67 {
68 VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_IMR + intr_idx * 8, 1);
69 }
70
71
72 /*
73 * Enable/Disable all intrs used by the device
74 */
75 static void
vmxnet3_enable_all_intrs(struct vmxnet3_adapter * adapter)76 vmxnet3_enable_all_intrs(struct vmxnet3_adapter *adapter)
77 {
78 int i;
79
80 for (i = 0; i < adapter->intr.num_intrs; i++)
81 vmxnet3_enable_intr(adapter, i);
82 adapter->shared->devRead.intrConf.intrCtrl &=
83 cpu_to_le32(~VMXNET3_IC_DISABLE_ALL);
84 }
85
86
87 static void
vmxnet3_disable_all_intrs(struct vmxnet3_adapter * adapter)88 vmxnet3_disable_all_intrs(struct vmxnet3_adapter *adapter)
89 {
90 int i;
91
92 adapter->shared->devRead.intrConf.intrCtrl |=
93 cpu_to_le32(VMXNET3_IC_DISABLE_ALL);
94 for (i = 0; i < adapter->intr.num_intrs; i++)
95 vmxnet3_disable_intr(adapter, i);
96 }
97
98
99 static void
vmxnet3_ack_events(struct vmxnet3_adapter * adapter,u32 events)100 vmxnet3_ack_events(struct vmxnet3_adapter *adapter, u32 events)
101 {
102 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_ECR, events);
103 }
104
105
106 static bool
vmxnet3_tq_stopped(struct vmxnet3_tx_queue * tq,struct vmxnet3_adapter * adapter)107 vmxnet3_tq_stopped(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
108 {
109 return tq->stopped;
110 }
111
112
113 static void
vmxnet3_tq_start(struct vmxnet3_tx_queue * tq,struct vmxnet3_adapter * adapter)114 vmxnet3_tq_start(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
115 {
116 tq->stopped = false;
117 netif_start_subqueue(adapter->netdev, tq - adapter->tx_queue);
118 }
119
120
121 static void
vmxnet3_tq_wake(struct vmxnet3_tx_queue * tq,struct vmxnet3_adapter * adapter)122 vmxnet3_tq_wake(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
123 {
124 tq->stopped = false;
125 netif_wake_subqueue(adapter->netdev, (tq - adapter->tx_queue));
126 }
127
128
129 static void
vmxnet3_tq_stop(struct vmxnet3_tx_queue * tq,struct vmxnet3_adapter * adapter)130 vmxnet3_tq_stop(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
131 {
132 tq->stopped = true;
133 tq->num_stop++;
134 netif_stop_subqueue(adapter->netdev, (tq - adapter->tx_queue));
135 }
136
137
138 /*
139 * Check the link state. This may start or stop the tx queue.
140 */
141 static void
vmxnet3_check_link(struct vmxnet3_adapter * adapter,bool affectTxQueue)142 vmxnet3_check_link(struct vmxnet3_adapter *adapter, bool affectTxQueue)
143 {
144 u32 ret;
145 int i;
146 unsigned long flags;
147
148 spin_lock_irqsave(&adapter->cmd_lock, flags);
149 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, VMXNET3_CMD_GET_LINK);
150 ret = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
151 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
152
153 adapter->link_speed = ret >> 16;
154 if (ret & 1) { /* Link is up. */
155 printk(KERN_INFO "%s: NIC Link is Up %d Mbps\n",
156 adapter->netdev->name, adapter->link_speed);
157 if (!netif_carrier_ok(adapter->netdev))
158 netif_carrier_on(adapter->netdev);
159
160 if (affectTxQueue) {
161 for (i = 0; i < adapter->num_tx_queues; i++)
162 vmxnet3_tq_start(&adapter->tx_queue[i],
163 adapter);
164 }
165 } else {
166 printk(KERN_INFO "%s: NIC Link is Down\n",
167 adapter->netdev->name);
168 if (netif_carrier_ok(adapter->netdev))
169 netif_carrier_off(adapter->netdev);
170
171 if (affectTxQueue) {
172 for (i = 0; i < adapter->num_tx_queues; i++)
173 vmxnet3_tq_stop(&adapter->tx_queue[i], adapter);
174 }
175 }
176 }
177
178 static void
vmxnet3_process_events(struct vmxnet3_adapter * adapter)179 vmxnet3_process_events(struct vmxnet3_adapter *adapter)
180 {
181 int i;
182 unsigned long flags;
183 u32 events = le32_to_cpu(adapter->shared->ecr);
184 if (!events)
185 return;
186
187 vmxnet3_ack_events(adapter, events);
188
189 /* Check if link state has changed */
190 if (events & VMXNET3_ECR_LINK)
191 vmxnet3_check_link(adapter, true);
192
193 /* Check if there is an error on xmit/recv queues */
194 if (events & (VMXNET3_ECR_TQERR | VMXNET3_ECR_RQERR)) {
195 spin_lock_irqsave(&adapter->cmd_lock, flags);
196 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
197 VMXNET3_CMD_GET_QUEUE_STATUS);
198 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
199
200 for (i = 0; i < adapter->num_tx_queues; i++)
201 if (adapter->tqd_start[i].status.stopped)
202 dev_err(&adapter->netdev->dev,
203 "%s: tq[%d] error 0x%x\n",
204 adapter->netdev->name, i, le32_to_cpu(
205 adapter->tqd_start[i].status.error));
206 for (i = 0; i < adapter->num_rx_queues; i++)
207 if (adapter->rqd_start[i].status.stopped)
208 dev_err(&adapter->netdev->dev,
209 "%s: rq[%d] error 0x%x\n",
210 adapter->netdev->name, i,
211 adapter->rqd_start[i].status.error);
212
213 schedule_work(&adapter->work);
214 }
215 }
216
217 #ifdef __BIG_ENDIAN_BITFIELD
218 /*
219 * The device expects the bitfields in shared structures to be written in
220 * little endian. When CPU is big endian, the following routines are used to
221 * correctly read and write into ABI.
222 * The general technique used here is : double word bitfields are defined in
223 * opposite order for big endian architecture. Then before reading them in
224 * driver the complete double word is translated using le32_to_cpu. Similarly
225 * After the driver writes into bitfields, cpu_to_le32 is used to translate the
226 * double words into required format.
227 * In order to avoid touching bits in shared structure more than once, temporary
228 * descriptors are used. These are passed as srcDesc to following functions.
229 */
vmxnet3_RxDescToCPU(const struct Vmxnet3_RxDesc * srcDesc,struct Vmxnet3_RxDesc * dstDesc)230 static void vmxnet3_RxDescToCPU(const struct Vmxnet3_RxDesc *srcDesc,
231 struct Vmxnet3_RxDesc *dstDesc)
232 {
233 u32 *src = (u32 *)srcDesc + 2;
234 u32 *dst = (u32 *)dstDesc + 2;
235 dstDesc->addr = le64_to_cpu(srcDesc->addr);
236 *dst = le32_to_cpu(*src);
237 dstDesc->ext1 = le32_to_cpu(srcDesc->ext1);
238 }
239
vmxnet3_TxDescToLe(const struct Vmxnet3_TxDesc * srcDesc,struct Vmxnet3_TxDesc * dstDesc)240 static void vmxnet3_TxDescToLe(const struct Vmxnet3_TxDesc *srcDesc,
241 struct Vmxnet3_TxDesc *dstDesc)
242 {
243 int i;
244 u32 *src = (u32 *)(srcDesc + 1);
245 u32 *dst = (u32 *)(dstDesc + 1);
246
247 /* Working backwards so that the gen bit is set at the end. */
248 for (i = 2; i > 0; i--) {
249 src--;
250 dst--;
251 *dst = cpu_to_le32(*src);
252 }
253 }
254
255
vmxnet3_RxCompToCPU(const struct Vmxnet3_RxCompDesc * srcDesc,struct Vmxnet3_RxCompDesc * dstDesc)256 static void vmxnet3_RxCompToCPU(const struct Vmxnet3_RxCompDesc *srcDesc,
257 struct Vmxnet3_RxCompDesc *dstDesc)
258 {
259 int i = 0;
260 u32 *src = (u32 *)srcDesc;
261 u32 *dst = (u32 *)dstDesc;
262 for (i = 0; i < sizeof(struct Vmxnet3_RxCompDesc) / sizeof(u32); i++) {
263 *dst = le32_to_cpu(*src);
264 src++;
265 dst++;
266 }
267 }
268
269
270 /* Used to read bitfield values from double words. */
get_bitfield32(const __le32 * bitfield,u32 pos,u32 size)271 static u32 get_bitfield32(const __le32 *bitfield, u32 pos, u32 size)
272 {
273 u32 temp = le32_to_cpu(*bitfield);
274 u32 mask = ((1 << size) - 1) << pos;
275 temp &= mask;
276 temp >>= pos;
277 return temp;
278 }
279
280
281
282 #endif /* __BIG_ENDIAN_BITFIELD */
283
284 #ifdef __BIG_ENDIAN_BITFIELD
285
286 # define VMXNET3_TXDESC_GET_GEN(txdesc) get_bitfield32(((const __le32 *) \
287 txdesc) + VMXNET3_TXD_GEN_DWORD_SHIFT, \
288 VMXNET3_TXD_GEN_SHIFT, VMXNET3_TXD_GEN_SIZE)
289 # define VMXNET3_TXDESC_GET_EOP(txdesc) get_bitfield32(((const __le32 *) \
290 txdesc) + VMXNET3_TXD_EOP_DWORD_SHIFT, \
291 VMXNET3_TXD_EOP_SHIFT, VMXNET3_TXD_EOP_SIZE)
292 # define VMXNET3_TCD_GET_GEN(tcd) get_bitfield32(((const __le32 *)tcd) + \
293 VMXNET3_TCD_GEN_DWORD_SHIFT, VMXNET3_TCD_GEN_SHIFT, \
294 VMXNET3_TCD_GEN_SIZE)
295 # define VMXNET3_TCD_GET_TXIDX(tcd) get_bitfield32((const __le32 *)tcd, \
296 VMXNET3_TCD_TXIDX_SHIFT, VMXNET3_TCD_TXIDX_SIZE)
297 # define vmxnet3_getRxComp(dstrcd, rcd, tmp) do { \
298 (dstrcd) = (tmp); \
299 vmxnet3_RxCompToCPU((rcd), (tmp)); \
300 } while (0)
301 # define vmxnet3_getRxDesc(dstrxd, rxd, tmp) do { \
302 (dstrxd) = (tmp); \
303 vmxnet3_RxDescToCPU((rxd), (tmp)); \
304 } while (0)
305
306 #else
307
308 # define VMXNET3_TXDESC_GET_GEN(txdesc) ((txdesc)->gen)
309 # define VMXNET3_TXDESC_GET_EOP(txdesc) ((txdesc)->eop)
310 # define VMXNET3_TCD_GET_GEN(tcd) ((tcd)->gen)
311 # define VMXNET3_TCD_GET_TXIDX(tcd) ((tcd)->txdIdx)
312 # define vmxnet3_getRxComp(dstrcd, rcd, tmp) (dstrcd) = (rcd)
313 # define vmxnet3_getRxDesc(dstrxd, rxd, tmp) (dstrxd) = (rxd)
314
315 #endif /* __BIG_ENDIAN_BITFIELD */
316
317
318 static void
vmxnet3_unmap_tx_buf(struct vmxnet3_tx_buf_info * tbi,struct pci_dev * pdev)319 vmxnet3_unmap_tx_buf(struct vmxnet3_tx_buf_info *tbi,
320 struct pci_dev *pdev)
321 {
322 if (tbi->map_type == VMXNET3_MAP_SINGLE)
323 pci_unmap_single(pdev, tbi->dma_addr, tbi->len,
324 PCI_DMA_TODEVICE);
325 else if (tbi->map_type == VMXNET3_MAP_PAGE)
326 pci_unmap_page(pdev, tbi->dma_addr, tbi->len,
327 PCI_DMA_TODEVICE);
328 else
329 BUG_ON(tbi->map_type != VMXNET3_MAP_NONE);
330
331 tbi->map_type = VMXNET3_MAP_NONE; /* to help debugging */
332 }
333
334
335 static int
vmxnet3_unmap_pkt(u32 eop_idx,struct vmxnet3_tx_queue * tq,struct pci_dev * pdev,struct vmxnet3_adapter * adapter)336 vmxnet3_unmap_pkt(u32 eop_idx, struct vmxnet3_tx_queue *tq,
337 struct pci_dev *pdev, struct vmxnet3_adapter *adapter)
338 {
339 struct sk_buff *skb;
340 int entries = 0;
341
342 /* no out of order completion */
343 BUG_ON(tq->buf_info[eop_idx].sop_idx != tq->tx_ring.next2comp);
344 BUG_ON(VMXNET3_TXDESC_GET_EOP(&(tq->tx_ring.base[eop_idx].txd)) != 1);
345
346 skb = tq->buf_info[eop_idx].skb;
347 BUG_ON(skb == NULL);
348 tq->buf_info[eop_idx].skb = NULL;
349
350 VMXNET3_INC_RING_IDX_ONLY(eop_idx, tq->tx_ring.size);
351
352 while (tq->tx_ring.next2comp != eop_idx) {
353 vmxnet3_unmap_tx_buf(tq->buf_info + tq->tx_ring.next2comp,
354 pdev);
355
356 /* update next2comp w/o tx_lock. Since we are marking more,
357 * instead of less, tx ring entries avail, the worst case is
358 * that the tx routine incorrectly re-queues a pkt due to
359 * insufficient tx ring entries.
360 */
361 vmxnet3_cmd_ring_adv_next2comp(&tq->tx_ring);
362 entries++;
363 }
364
365 dev_kfree_skb_any(skb);
366 return entries;
367 }
368
369
370 static int
vmxnet3_tq_tx_complete(struct vmxnet3_tx_queue * tq,struct vmxnet3_adapter * adapter)371 vmxnet3_tq_tx_complete(struct vmxnet3_tx_queue *tq,
372 struct vmxnet3_adapter *adapter)
373 {
374 int completed = 0;
375 union Vmxnet3_GenericDesc *gdesc;
376
377 gdesc = tq->comp_ring.base + tq->comp_ring.next2proc;
378 while (VMXNET3_TCD_GET_GEN(&gdesc->tcd) == tq->comp_ring.gen) {
379 completed += vmxnet3_unmap_pkt(VMXNET3_TCD_GET_TXIDX(
380 &gdesc->tcd), tq, adapter->pdev,
381 adapter);
382
383 vmxnet3_comp_ring_adv_next2proc(&tq->comp_ring);
384 gdesc = tq->comp_ring.base + tq->comp_ring.next2proc;
385 }
386
387 if (completed) {
388 spin_lock(&tq->tx_lock);
389 if (unlikely(vmxnet3_tq_stopped(tq, adapter) &&
390 vmxnet3_cmd_ring_desc_avail(&tq->tx_ring) >
391 VMXNET3_WAKE_QUEUE_THRESHOLD(tq) &&
392 netif_carrier_ok(adapter->netdev))) {
393 vmxnet3_tq_wake(tq, adapter);
394 }
395 spin_unlock(&tq->tx_lock);
396 }
397 return completed;
398 }
399
400
401 static void
vmxnet3_tq_cleanup(struct vmxnet3_tx_queue * tq,struct vmxnet3_adapter * adapter)402 vmxnet3_tq_cleanup(struct vmxnet3_tx_queue *tq,
403 struct vmxnet3_adapter *adapter)
404 {
405 int i;
406
407 while (tq->tx_ring.next2comp != tq->tx_ring.next2fill) {
408 struct vmxnet3_tx_buf_info *tbi;
409
410 tbi = tq->buf_info + tq->tx_ring.next2comp;
411
412 vmxnet3_unmap_tx_buf(tbi, adapter->pdev);
413 if (tbi->skb) {
414 dev_kfree_skb_any(tbi->skb);
415 tbi->skb = NULL;
416 }
417 vmxnet3_cmd_ring_adv_next2comp(&tq->tx_ring);
418 }
419
420 /* sanity check, verify all buffers are indeed unmapped and freed */
421 for (i = 0; i < tq->tx_ring.size; i++) {
422 BUG_ON(tq->buf_info[i].skb != NULL ||
423 tq->buf_info[i].map_type != VMXNET3_MAP_NONE);
424 }
425
426 tq->tx_ring.gen = VMXNET3_INIT_GEN;
427 tq->tx_ring.next2fill = tq->tx_ring.next2comp = 0;
428
429 tq->comp_ring.gen = VMXNET3_INIT_GEN;
430 tq->comp_ring.next2proc = 0;
431 }
432
433
434 static void
vmxnet3_tq_destroy(struct vmxnet3_tx_queue * tq,struct vmxnet3_adapter * adapter)435 vmxnet3_tq_destroy(struct vmxnet3_tx_queue *tq,
436 struct vmxnet3_adapter *adapter)
437 {
438 if (tq->tx_ring.base) {
439 pci_free_consistent(adapter->pdev, tq->tx_ring.size *
440 sizeof(struct Vmxnet3_TxDesc),
441 tq->tx_ring.base, tq->tx_ring.basePA);
442 tq->tx_ring.base = NULL;
443 }
444 if (tq->data_ring.base) {
445 pci_free_consistent(adapter->pdev, tq->data_ring.size *
446 sizeof(struct Vmxnet3_TxDataDesc),
447 tq->data_ring.base, tq->data_ring.basePA);
448 tq->data_ring.base = NULL;
449 }
450 if (tq->comp_ring.base) {
451 pci_free_consistent(adapter->pdev, tq->comp_ring.size *
452 sizeof(struct Vmxnet3_TxCompDesc),
453 tq->comp_ring.base, tq->comp_ring.basePA);
454 tq->comp_ring.base = NULL;
455 }
456 kfree(tq->buf_info);
457 tq->buf_info = NULL;
458 }
459
460
461 /* Destroy all tx queues */
462 void
vmxnet3_tq_destroy_all(struct vmxnet3_adapter * adapter)463 vmxnet3_tq_destroy_all(struct vmxnet3_adapter *adapter)
464 {
465 int i;
466
467 for (i = 0; i < adapter->num_tx_queues; i++)
468 vmxnet3_tq_destroy(&adapter->tx_queue[i], adapter);
469 }
470
471
472 static void
vmxnet3_tq_init(struct vmxnet3_tx_queue * tq,struct vmxnet3_adapter * adapter)473 vmxnet3_tq_init(struct vmxnet3_tx_queue *tq,
474 struct vmxnet3_adapter *adapter)
475 {
476 int i;
477
478 /* reset the tx ring contents to 0 and reset the tx ring states */
479 memset(tq->tx_ring.base, 0, tq->tx_ring.size *
480 sizeof(struct Vmxnet3_TxDesc));
481 tq->tx_ring.next2fill = tq->tx_ring.next2comp = 0;
482 tq->tx_ring.gen = VMXNET3_INIT_GEN;
483
484 memset(tq->data_ring.base, 0, tq->data_ring.size *
485 sizeof(struct Vmxnet3_TxDataDesc));
486
487 /* reset the tx comp ring contents to 0 and reset comp ring states */
488 memset(tq->comp_ring.base, 0, tq->comp_ring.size *
489 sizeof(struct Vmxnet3_TxCompDesc));
490 tq->comp_ring.next2proc = 0;
491 tq->comp_ring.gen = VMXNET3_INIT_GEN;
492
493 /* reset the bookkeeping data */
494 memset(tq->buf_info, 0, sizeof(tq->buf_info[0]) * tq->tx_ring.size);
495 for (i = 0; i < tq->tx_ring.size; i++)
496 tq->buf_info[i].map_type = VMXNET3_MAP_NONE;
497
498 /* stats are not reset */
499 }
500
501
502 static int
vmxnet3_tq_create(struct vmxnet3_tx_queue * tq,struct vmxnet3_adapter * adapter)503 vmxnet3_tq_create(struct vmxnet3_tx_queue *tq,
504 struct vmxnet3_adapter *adapter)
505 {
506 BUG_ON(tq->tx_ring.base || tq->data_ring.base ||
507 tq->comp_ring.base || tq->buf_info);
508
509 tq->tx_ring.base = pci_alloc_consistent(adapter->pdev, tq->tx_ring.size
510 * sizeof(struct Vmxnet3_TxDesc),
511 &tq->tx_ring.basePA);
512 if (!tq->tx_ring.base) {
513 printk(KERN_ERR "%s: failed to allocate tx ring\n",
514 adapter->netdev->name);
515 goto err;
516 }
517
518 tq->data_ring.base = pci_alloc_consistent(adapter->pdev,
519 tq->data_ring.size *
520 sizeof(struct Vmxnet3_TxDataDesc),
521 &tq->data_ring.basePA);
522 if (!tq->data_ring.base) {
523 printk(KERN_ERR "%s: failed to allocate data ring\n",
524 adapter->netdev->name);
525 goto err;
526 }
527
528 tq->comp_ring.base = pci_alloc_consistent(adapter->pdev,
529 tq->comp_ring.size *
530 sizeof(struct Vmxnet3_TxCompDesc),
531 &tq->comp_ring.basePA);
532 if (!tq->comp_ring.base) {
533 printk(KERN_ERR "%s: failed to allocate tx comp ring\n",
534 adapter->netdev->name);
535 goto err;
536 }
537
538 tq->buf_info = kcalloc(tq->tx_ring.size, sizeof(tq->buf_info[0]),
539 GFP_KERNEL);
540 if (!tq->buf_info) {
541 printk(KERN_ERR "%s: failed to allocate tx bufinfo\n",
542 adapter->netdev->name);
543 goto err;
544 }
545
546 return 0;
547
548 err:
549 vmxnet3_tq_destroy(tq, adapter);
550 return -ENOMEM;
551 }
552
553 static void
vmxnet3_tq_cleanup_all(struct vmxnet3_adapter * adapter)554 vmxnet3_tq_cleanup_all(struct vmxnet3_adapter *adapter)
555 {
556 int i;
557
558 for (i = 0; i < adapter->num_tx_queues; i++)
559 vmxnet3_tq_cleanup(&adapter->tx_queue[i], adapter);
560 }
561
562 /*
563 * starting from ring->next2fill, allocate rx buffers for the given ring
564 * of the rx queue and update the rx desc. stop after @num_to_alloc buffers
565 * are allocated or allocation fails
566 */
567
568 static int
vmxnet3_rq_alloc_rx_buf(struct vmxnet3_rx_queue * rq,u32 ring_idx,int num_to_alloc,struct vmxnet3_adapter * adapter)569 vmxnet3_rq_alloc_rx_buf(struct vmxnet3_rx_queue *rq, u32 ring_idx,
570 int num_to_alloc, struct vmxnet3_adapter *adapter)
571 {
572 int num_allocated = 0;
573 struct vmxnet3_rx_buf_info *rbi_base = rq->buf_info[ring_idx];
574 struct vmxnet3_cmd_ring *ring = &rq->rx_ring[ring_idx];
575 u32 val;
576
577 while (num_allocated <= num_to_alloc) {
578 struct vmxnet3_rx_buf_info *rbi;
579 union Vmxnet3_GenericDesc *gd;
580
581 rbi = rbi_base + ring->next2fill;
582 gd = ring->base + ring->next2fill;
583
584 if (rbi->buf_type == VMXNET3_RX_BUF_SKB) {
585 if (rbi->skb == NULL) {
586 rbi->skb = dev_alloc_skb(rbi->len +
587 NET_IP_ALIGN);
588 if (unlikely(rbi->skb == NULL)) {
589 rq->stats.rx_buf_alloc_failure++;
590 break;
591 }
592 rbi->skb->dev = adapter->netdev;
593
594 skb_reserve(rbi->skb, NET_IP_ALIGN);
595 rbi->dma_addr = pci_map_single(adapter->pdev,
596 rbi->skb->data, rbi->len,
597 PCI_DMA_FROMDEVICE);
598 } else {
599 /* rx buffer skipped by the device */
600 }
601 val = VMXNET3_RXD_BTYPE_HEAD << VMXNET3_RXD_BTYPE_SHIFT;
602 } else {
603 BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_PAGE ||
604 rbi->len != PAGE_SIZE);
605
606 if (rbi->page == NULL) {
607 rbi->page = alloc_page(GFP_ATOMIC);
608 if (unlikely(rbi->page == NULL)) {
609 rq->stats.rx_buf_alloc_failure++;
610 break;
611 }
612 rbi->dma_addr = pci_map_page(adapter->pdev,
613 rbi->page, 0, PAGE_SIZE,
614 PCI_DMA_FROMDEVICE);
615 } else {
616 /* rx buffers skipped by the device */
617 }
618 val = VMXNET3_RXD_BTYPE_BODY << VMXNET3_RXD_BTYPE_SHIFT;
619 }
620
621 BUG_ON(rbi->dma_addr == 0);
622 gd->rxd.addr = cpu_to_le64(rbi->dma_addr);
623 gd->dword[2] = cpu_to_le32((!ring->gen << VMXNET3_RXD_GEN_SHIFT)
624 | val | rbi->len);
625
626 /* Fill the last buffer but dont mark it ready, or else the
627 * device will think that the queue is full */
628 if (num_allocated == num_to_alloc)
629 break;
630
631 gd->dword[2] |= cpu_to_le32(ring->gen << VMXNET3_RXD_GEN_SHIFT);
632 num_allocated++;
633 vmxnet3_cmd_ring_adv_next2fill(ring);
634 }
635 rq->uncommitted[ring_idx] += num_allocated;
636
637 dev_dbg(&adapter->netdev->dev,
638 "alloc_rx_buf: %d allocated, next2fill %u, next2comp "
639 "%u, uncommited %u\n", num_allocated, ring->next2fill,
640 ring->next2comp, rq->uncommitted[ring_idx]);
641
642 /* so that the device can distinguish a full ring and an empty ring */
643 BUG_ON(num_allocated != 0 && ring->next2fill == ring->next2comp);
644
645 return num_allocated;
646 }
647
648
649 static void
vmxnet3_append_frag(struct sk_buff * skb,struct Vmxnet3_RxCompDesc * rcd,struct vmxnet3_rx_buf_info * rbi)650 vmxnet3_append_frag(struct sk_buff *skb, struct Vmxnet3_RxCompDesc *rcd,
651 struct vmxnet3_rx_buf_info *rbi)
652 {
653 struct skb_frag_struct *frag = skb_shinfo(skb)->frags +
654 skb_shinfo(skb)->nr_frags;
655
656 BUG_ON(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS);
657
658 __skb_frag_set_page(frag, rbi->page);
659 frag->page_offset = 0;
660 skb_frag_size_set(frag, rcd->len);
661 skb->data_len += rcd->len;
662 skb->truesize += PAGE_SIZE;
663 skb_shinfo(skb)->nr_frags++;
664 }
665
666
667 static void
vmxnet3_map_pkt(struct sk_buff * skb,struct vmxnet3_tx_ctx * ctx,struct vmxnet3_tx_queue * tq,struct pci_dev * pdev,struct vmxnet3_adapter * adapter)668 vmxnet3_map_pkt(struct sk_buff *skb, struct vmxnet3_tx_ctx *ctx,
669 struct vmxnet3_tx_queue *tq, struct pci_dev *pdev,
670 struct vmxnet3_adapter *adapter)
671 {
672 u32 dw2, len;
673 unsigned long buf_offset;
674 int i;
675 union Vmxnet3_GenericDesc *gdesc;
676 struct vmxnet3_tx_buf_info *tbi = NULL;
677
678 BUG_ON(ctx->copy_size > skb_headlen(skb));
679
680 /* use the previous gen bit for the SOP desc */
681 dw2 = (tq->tx_ring.gen ^ 0x1) << VMXNET3_TXD_GEN_SHIFT;
682
683 ctx->sop_txd = tq->tx_ring.base + tq->tx_ring.next2fill;
684 gdesc = ctx->sop_txd; /* both loops below can be skipped */
685
686 /* no need to map the buffer if headers are copied */
687 if (ctx->copy_size) {
688 ctx->sop_txd->txd.addr = cpu_to_le64(tq->data_ring.basePA +
689 tq->tx_ring.next2fill *
690 sizeof(struct Vmxnet3_TxDataDesc));
691 ctx->sop_txd->dword[2] = cpu_to_le32(dw2 | ctx->copy_size);
692 ctx->sop_txd->dword[3] = 0;
693
694 tbi = tq->buf_info + tq->tx_ring.next2fill;
695 tbi->map_type = VMXNET3_MAP_NONE;
696
697 dev_dbg(&adapter->netdev->dev,
698 "txd[%u]: 0x%Lx 0x%x 0x%x\n",
699 tq->tx_ring.next2fill,
700 le64_to_cpu(ctx->sop_txd->txd.addr),
701 ctx->sop_txd->dword[2], ctx->sop_txd->dword[3]);
702 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
703
704 /* use the right gen for non-SOP desc */
705 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
706 }
707
708 /* linear part can use multiple tx desc if it's big */
709 len = skb_headlen(skb) - ctx->copy_size;
710 buf_offset = ctx->copy_size;
711 while (len) {
712 u32 buf_size;
713
714 if (len < VMXNET3_MAX_TX_BUF_SIZE) {
715 buf_size = len;
716 dw2 |= len;
717 } else {
718 buf_size = VMXNET3_MAX_TX_BUF_SIZE;
719 /* spec says that for TxDesc.len, 0 == 2^14 */
720 }
721
722 tbi = tq->buf_info + tq->tx_ring.next2fill;
723 tbi->map_type = VMXNET3_MAP_SINGLE;
724 tbi->dma_addr = pci_map_single(adapter->pdev,
725 skb->data + buf_offset, buf_size,
726 PCI_DMA_TODEVICE);
727
728 tbi->len = buf_size;
729
730 gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
731 BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
732
733 gdesc->txd.addr = cpu_to_le64(tbi->dma_addr);
734 gdesc->dword[2] = cpu_to_le32(dw2);
735 gdesc->dword[3] = 0;
736
737 dev_dbg(&adapter->netdev->dev,
738 "txd[%u]: 0x%Lx 0x%x 0x%x\n",
739 tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr),
740 le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]);
741 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
742 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
743
744 len -= buf_size;
745 buf_offset += buf_size;
746 }
747
748 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
749 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
750
751 tbi = tq->buf_info + tq->tx_ring.next2fill;
752 tbi->map_type = VMXNET3_MAP_PAGE;
753 tbi->dma_addr = skb_frag_dma_map(&adapter->pdev->dev, frag,
754 0, skb_frag_size(frag),
755 DMA_TO_DEVICE);
756
757 tbi->len = skb_frag_size(frag);
758
759 gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
760 BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
761
762 gdesc->txd.addr = cpu_to_le64(tbi->dma_addr);
763 gdesc->dword[2] = cpu_to_le32(dw2 | skb_frag_size(frag));
764 gdesc->dword[3] = 0;
765
766 dev_dbg(&adapter->netdev->dev,
767 "txd[%u]: 0x%llu %u %u\n",
768 tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr),
769 le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]);
770 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
771 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
772 }
773
774 ctx->eop_txd = gdesc;
775
776 /* set the last buf_info for the pkt */
777 tbi->skb = skb;
778 tbi->sop_idx = ctx->sop_txd - tq->tx_ring.base;
779 }
780
781
782 /* Init all tx queues */
783 static void
vmxnet3_tq_init_all(struct vmxnet3_adapter * adapter)784 vmxnet3_tq_init_all(struct vmxnet3_adapter *adapter)
785 {
786 int i;
787
788 for (i = 0; i < adapter->num_tx_queues; i++)
789 vmxnet3_tq_init(&adapter->tx_queue[i], adapter);
790 }
791
792
793 /*
794 * parse and copy relevant protocol headers:
795 * For a tso pkt, relevant headers are L2/3/4 including options
796 * For a pkt requesting csum offloading, they are L2/3 and may include L4
797 * if it's a TCP/UDP pkt
798 *
799 * Returns:
800 * -1: error happens during parsing
801 * 0: protocol headers parsed, but too big to be copied
802 * 1: protocol headers parsed and copied
803 *
804 * Other effects:
805 * 1. related *ctx fields are updated.
806 * 2. ctx->copy_size is # of bytes copied
807 * 3. the portion copied is guaranteed to be in the linear part
808 *
809 */
810 static int
vmxnet3_parse_and_copy_hdr(struct sk_buff * skb,struct vmxnet3_tx_queue * tq,struct vmxnet3_tx_ctx * ctx,struct vmxnet3_adapter * adapter)811 vmxnet3_parse_and_copy_hdr(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
812 struct vmxnet3_tx_ctx *ctx,
813 struct vmxnet3_adapter *adapter)
814 {
815 struct Vmxnet3_TxDataDesc *tdd;
816
817 if (ctx->mss) { /* TSO */
818 ctx->eth_ip_hdr_size = skb_transport_offset(skb);
819 ctx->l4_hdr_size = ((struct tcphdr *)
820 skb_transport_header(skb))->doff * 4;
821 ctx->copy_size = ctx->eth_ip_hdr_size + ctx->l4_hdr_size;
822 } else {
823 if (skb->ip_summed == CHECKSUM_PARTIAL) {
824 ctx->eth_ip_hdr_size = skb_checksum_start_offset(skb);
825
826 if (ctx->ipv4) {
827 struct iphdr *iph = (struct iphdr *)
828 skb_network_header(skb);
829 if (iph->protocol == IPPROTO_TCP)
830 ctx->l4_hdr_size = ((struct tcphdr *)
831 skb_transport_header(skb))->doff * 4;
832 else if (iph->protocol == IPPROTO_UDP)
833 ctx->l4_hdr_size =
834 sizeof(struct udphdr);
835 else
836 ctx->l4_hdr_size = 0;
837 } else {
838 /* for simplicity, don't copy L4 headers */
839 ctx->l4_hdr_size = 0;
840 }
841 ctx->copy_size = min(ctx->eth_ip_hdr_size +
842 ctx->l4_hdr_size, skb->len);
843 } else {
844 ctx->eth_ip_hdr_size = 0;
845 ctx->l4_hdr_size = 0;
846 /* copy as much as allowed */
847 ctx->copy_size = min((unsigned int)VMXNET3_HDR_COPY_SIZE
848 , skb_headlen(skb));
849 }
850
851 /* make sure headers are accessible directly */
852 if (unlikely(!pskb_may_pull(skb, ctx->copy_size)))
853 goto err;
854 }
855
856 if (unlikely(ctx->copy_size > VMXNET3_HDR_COPY_SIZE)) {
857 tq->stats.oversized_hdr++;
858 ctx->copy_size = 0;
859 return 0;
860 }
861
862 tdd = tq->data_ring.base + tq->tx_ring.next2fill;
863
864 memcpy(tdd->data, skb->data, ctx->copy_size);
865 dev_dbg(&adapter->netdev->dev,
866 "copy %u bytes to dataRing[%u]\n",
867 ctx->copy_size, tq->tx_ring.next2fill);
868 return 1;
869
870 err:
871 return -1;
872 }
873
874
875 static void
vmxnet3_prepare_tso(struct sk_buff * skb,struct vmxnet3_tx_ctx * ctx)876 vmxnet3_prepare_tso(struct sk_buff *skb,
877 struct vmxnet3_tx_ctx *ctx)
878 {
879 struct tcphdr *tcph = (struct tcphdr *)skb_transport_header(skb);
880 if (ctx->ipv4) {
881 struct iphdr *iph = (struct iphdr *)skb_network_header(skb);
882 iph->check = 0;
883 tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, 0,
884 IPPROTO_TCP, 0);
885 } else {
886 struct ipv6hdr *iph = (struct ipv6hdr *)skb_network_header(skb);
887 tcph->check = ~csum_ipv6_magic(&iph->saddr, &iph->daddr, 0,
888 IPPROTO_TCP, 0);
889 }
890 }
891
892
893 /*
894 * Transmits a pkt thru a given tq
895 * Returns:
896 * NETDEV_TX_OK: descriptors are setup successfully
897 * NETDEV_TX_OK: error occurred, the pkt is dropped
898 * NETDEV_TX_BUSY: tx ring is full, queue is stopped
899 *
900 * Side-effects:
901 * 1. tx ring may be changed
902 * 2. tq stats may be updated accordingly
903 * 3. shared->txNumDeferred may be updated
904 */
905
906 static int
vmxnet3_tq_xmit(struct sk_buff * skb,struct vmxnet3_tx_queue * tq,struct vmxnet3_adapter * adapter,struct net_device * netdev)907 vmxnet3_tq_xmit(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
908 struct vmxnet3_adapter *adapter, struct net_device *netdev)
909 {
910 int ret;
911 u32 count;
912 unsigned long flags;
913 struct vmxnet3_tx_ctx ctx;
914 union Vmxnet3_GenericDesc *gdesc;
915 #ifdef __BIG_ENDIAN_BITFIELD
916 /* Use temporary descriptor to avoid touching bits multiple times */
917 union Vmxnet3_GenericDesc tempTxDesc;
918 #endif
919
920 /* conservatively estimate # of descriptors to use */
921 count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) +
922 skb_shinfo(skb)->nr_frags + 1;
923
924 ctx.ipv4 = (vlan_get_protocol(skb) == cpu_to_be16(ETH_P_IP));
925
926 ctx.mss = skb_shinfo(skb)->gso_size;
927 if (ctx.mss) {
928 if (skb_header_cloned(skb)) {
929 if (unlikely(pskb_expand_head(skb, 0, 0,
930 GFP_ATOMIC) != 0)) {
931 tq->stats.drop_tso++;
932 goto drop_pkt;
933 }
934 tq->stats.copy_skb_header++;
935 }
936 vmxnet3_prepare_tso(skb, &ctx);
937 } else {
938 if (unlikely(count > VMXNET3_MAX_TXD_PER_PKT)) {
939
940 /* non-tso pkts must not use more than
941 * VMXNET3_MAX_TXD_PER_PKT entries
942 */
943 if (skb_linearize(skb) != 0) {
944 tq->stats.drop_too_many_frags++;
945 goto drop_pkt;
946 }
947 tq->stats.linearized++;
948
949 /* recalculate the # of descriptors to use */
950 count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) + 1;
951 }
952 }
953
954 spin_lock_irqsave(&tq->tx_lock, flags);
955
956 if (count > vmxnet3_cmd_ring_desc_avail(&tq->tx_ring)) {
957 tq->stats.tx_ring_full++;
958 dev_dbg(&adapter->netdev->dev,
959 "tx queue stopped on %s, next2comp %u"
960 " next2fill %u\n", adapter->netdev->name,
961 tq->tx_ring.next2comp, tq->tx_ring.next2fill);
962
963 vmxnet3_tq_stop(tq, adapter);
964 spin_unlock_irqrestore(&tq->tx_lock, flags);
965 return NETDEV_TX_BUSY;
966 }
967
968
969 ret = vmxnet3_parse_and_copy_hdr(skb, tq, &ctx, adapter);
970 if (ret >= 0) {
971 BUG_ON(ret <= 0 && ctx.copy_size != 0);
972 /* hdrs parsed, check against other limits */
973 if (ctx.mss) {
974 if (unlikely(ctx.eth_ip_hdr_size + ctx.l4_hdr_size >
975 VMXNET3_MAX_TX_BUF_SIZE)) {
976 goto hdr_too_big;
977 }
978 } else {
979 if (skb->ip_summed == CHECKSUM_PARTIAL) {
980 if (unlikely(ctx.eth_ip_hdr_size +
981 skb->csum_offset >
982 VMXNET3_MAX_CSUM_OFFSET)) {
983 goto hdr_too_big;
984 }
985 }
986 }
987 } else {
988 tq->stats.drop_hdr_inspect_err++;
989 goto unlock_drop_pkt;
990 }
991
992 /* fill tx descs related to addr & len */
993 vmxnet3_map_pkt(skb, &ctx, tq, adapter->pdev, adapter);
994
995 /* setup the EOP desc */
996 ctx.eop_txd->dword[3] = cpu_to_le32(VMXNET3_TXD_CQ | VMXNET3_TXD_EOP);
997
998 /* setup the SOP desc */
999 #ifdef __BIG_ENDIAN_BITFIELD
1000 gdesc = &tempTxDesc;
1001 gdesc->dword[2] = ctx.sop_txd->dword[2];
1002 gdesc->dword[3] = ctx.sop_txd->dword[3];
1003 #else
1004 gdesc = ctx.sop_txd;
1005 #endif
1006 if (ctx.mss) {
1007 gdesc->txd.hlen = ctx.eth_ip_hdr_size + ctx.l4_hdr_size;
1008 gdesc->txd.om = VMXNET3_OM_TSO;
1009 gdesc->txd.msscof = ctx.mss;
1010 le32_add_cpu(&tq->shared->txNumDeferred, (skb->len -
1011 gdesc->txd.hlen + ctx.mss - 1) / ctx.mss);
1012 } else {
1013 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1014 gdesc->txd.hlen = ctx.eth_ip_hdr_size;
1015 gdesc->txd.om = VMXNET3_OM_CSUM;
1016 gdesc->txd.msscof = ctx.eth_ip_hdr_size +
1017 skb->csum_offset;
1018 } else {
1019 gdesc->txd.om = 0;
1020 gdesc->txd.msscof = 0;
1021 }
1022 le32_add_cpu(&tq->shared->txNumDeferred, 1);
1023 }
1024
1025 if (vlan_tx_tag_present(skb)) {
1026 gdesc->txd.ti = 1;
1027 gdesc->txd.tci = vlan_tx_tag_get(skb);
1028 }
1029
1030 /* finally flips the GEN bit of the SOP desc. */
1031 gdesc->dword[2] = cpu_to_le32(le32_to_cpu(gdesc->dword[2]) ^
1032 VMXNET3_TXD_GEN);
1033 #ifdef __BIG_ENDIAN_BITFIELD
1034 /* Finished updating in bitfields of Tx Desc, so write them in original
1035 * place.
1036 */
1037 vmxnet3_TxDescToLe((struct Vmxnet3_TxDesc *)gdesc,
1038 (struct Vmxnet3_TxDesc *)ctx.sop_txd);
1039 gdesc = ctx.sop_txd;
1040 #endif
1041 dev_dbg(&adapter->netdev->dev,
1042 "txd[%u]: SOP 0x%Lx 0x%x 0x%x\n",
1043 (u32)((union Vmxnet3_GenericDesc *)ctx.sop_txd -
1044 tq->tx_ring.base), le64_to_cpu(gdesc->txd.addr),
1045 le32_to_cpu(gdesc->dword[2]), le32_to_cpu(gdesc->dword[3]));
1046
1047 spin_unlock_irqrestore(&tq->tx_lock, flags);
1048
1049 if (le32_to_cpu(tq->shared->txNumDeferred) >=
1050 le32_to_cpu(tq->shared->txThreshold)) {
1051 tq->shared->txNumDeferred = 0;
1052 VMXNET3_WRITE_BAR0_REG(adapter,
1053 VMXNET3_REG_TXPROD + tq->qid * 8,
1054 tq->tx_ring.next2fill);
1055 }
1056
1057 return NETDEV_TX_OK;
1058
1059 hdr_too_big:
1060 tq->stats.drop_oversized_hdr++;
1061 unlock_drop_pkt:
1062 spin_unlock_irqrestore(&tq->tx_lock, flags);
1063 drop_pkt:
1064 tq->stats.drop_total++;
1065 dev_kfree_skb(skb);
1066 return NETDEV_TX_OK;
1067 }
1068
1069
1070 static netdev_tx_t
vmxnet3_xmit_frame(struct sk_buff * skb,struct net_device * netdev)1071 vmxnet3_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
1072 {
1073 struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1074
1075 BUG_ON(skb->queue_mapping > adapter->num_tx_queues);
1076 return vmxnet3_tq_xmit(skb,
1077 &adapter->tx_queue[skb->queue_mapping],
1078 adapter, netdev);
1079 }
1080
1081
1082 static void
vmxnet3_rx_csum(struct vmxnet3_adapter * adapter,struct sk_buff * skb,union Vmxnet3_GenericDesc * gdesc)1083 vmxnet3_rx_csum(struct vmxnet3_adapter *adapter,
1084 struct sk_buff *skb,
1085 union Vmxnet3_GenericDesc *gdesc)
1086 {
1087 if (!gdesc->rcd.cnc && adapter->netdev->features & NETIF_F_RXCSUM) {
1088 /* typical case: TCP/UDP over IP and both csums are correct */
1089 if ((le32_to_cpu(gdesc->dword[3]) & VMXNET3_RCD_CSUM_OK) ==
1090 VMXNET3_RCD_CSUM_OK) {
1091 skb->ip_summed = CHECKSUM_UNNECESSARY;
1092 BUG_ON(!(gdesc->rcd.tcp || gdesc->rcd.udp));
1093 BUG_ON(!(gdesc->rcd.v4 || gdesc->rcd.v6));
1094 BUG_ON(gdesc->rcd.frg);
1095 } else {
1096 if (gdesc->rcd.csum) {
1097 skb->csum = htons(gdesc->rcd.csum);
1098 skb->ip_summed = CHECKSUM_PARTIAL;
1099 } else {
1100 skb_checksum_none_assert(skb);
1101 }
1102 }
1103 } else {
1104 skb_checksum_none_assert(skb);
1105 }
1106 }
1107
1108
1109 static void
vmxnet3_rx_error(struct vmxnet3_rx_queue * rq,struct Vmxnet3_RxCompDesc * rcd,struct vmxnet3_rx_ctx * ctx,struct vmxnet3_adapter * adapter)1110 vmxnet3_rx_error(struct vmxnet3_rx_queue *rq, struct Vmxnet3_RxCompDesc *rcd,
1111 struct vmxnet3_rx_ctx *ctx, struct vmxnet3_adapter *adapter)
1112 {
1113 rq->stats.drop_err++;
1114 if (!rcd->fcs)
1115 rq->stats.drop_fcs++;
1116
1117 rq->stats.drop_total++;
1118
1119 /*
1120 * We do not unmap and chain the rx buffer to the skb.
1121 * We basically pretend this buffer is not used and will be recycled
1122 * by vmxnet3_rq_alloc_rx_buf()
1123 */
1124
1125 /*
1126 * ctx->skb may be NULL if this is the first and the only one
1127 * desc for the pkt
1128 */
1129 if (ctx->skb)
1130 dev_kfree_skb_irq(ctx->skb);
1131
1132 ctx->skb = NULL;
1133 }
1134
1135
1136 static int
vmxnet3_rq_rx_complete(struct vmxnet3_rx_queue * rq,struct vmxnet3_adapter * adapter,int quota)1137 vmxnet3_rq_rx_complete(struct vmxnet3_rx_queue *rq,
1138 struct vmxnet3_adapter *adapter, int quota)
1139 {
1140 static const u32 rxprod_reg[2] = {
1141 VMXNET3_REG_RXPROD, VMXNET3_REG_RXPROD2
1142 };
1143 u32 num_rxd = 0;
1144 bool skip_page_frags = false;
1145 struct Vmxnet3_RxCompDesc *rcd;
1146 struct vmxnet3_rx_ctx *ctx = &rq->rx_ctx;
1147 #ifdef __BIG_ENDIAN_BITFIELD
1148 struct Vmxnet3_RxDesc rxCmdDesc;
1149 struct Vmxnet3_RxCompDesc rxComp;
1150 #endif
1151 vmxnet3_getRxComp(rcd, &rq->comp_ring.base[rq->comp_ring.next2proc].rcd,
1152 &rxComp);
1153 while (rcd->gen == rq->comp_ring.gen) {
1154 struct vmxnet3_rx_buf_info *rbi;
1155 struct sk_buff *skb, *new_skb = NULL;
1156 struct page *new_page = NULL;
1157 int num_to_alloc;
1158 struct Vmxnet3_RxDesc *rxd;
1159 u32 idx, ring_idx;
1160 struct vmxnet3_cmd_ring *ring = NULL;
1161 if (num_rxd >= quota) {
1162 /* we may stop even before we see the EOP desc of
1163 * the current pkt
1164 */
1165 break;
1166 }
1167 num_rxd++;
1168 BUG_ON(rcd->rqID != rq->qid && rcd->rqID != rq->qid2);
1169 idx = rcd->rxdIdx;
1170 ring_idx = rcd->rqID < adapter->num_rx_queues ? 0 : 1;
1171 ring = rq->rx_ring + ring_idx;
1172 vmxnet3_getRxDesc(rxd, &rq->rx_ring[ring_idx].base[idx].rxd,
1173 &rxCmdDesc);
1174 rbi = rq->buf_info[ring_idx] + idx;
1175
1176 BUG_ON(rxd->addr != rbi->dma_addr ||
1177 rxd->len != rbi->len);
1178
1179 if (unlikely(rcd->eop && rcd->err)) {
1180 vmxnet3_rx_error(rq, rcd, ctx, adapter);
1181 goto rcd_done;
1182 }
1183
1184 if (rcd->sop) { /* first buf of the pkt */
1185 BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_HEAD ||
1186 rcd->rqID != rq->qid);
1187
1188 BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_SKB);
1189 BUG_ON(ctx->skb != NULL || rbi->skb == NULL);
1190
1191 if (unlikely(rcd->len == 0)) {
1192 /* Pretend the rx buffer is skipped. */
1193 BUG_ON(!(rcd->sop && rcd->eop));
1194 dev_dbg(&adapter->netdev->dev,
1195 "rxRing[%u][%u] 0 length\n",
1196 ring_idx, idx);
1197 goto rcd_done;
1198 }
1199
1200 skip_page_frags = false;
1201 ctx->skb = rbi->skb;
1202 new_skb = dev_alloc_skb(rbi->len + NET_IP_ALIGN);
1203 if (new_skb == NULL) {
1204 /* Skb allocation failed, do not handover this
1205 * skb to stack. Reuse it. Drop the existing pkt
1206 */
1207 rq->stats.rx_buf_alloc_failure++;
1208 ctx->skb = NULL;
1209 rq->stats.drop_total++;
1210 skip_page_frags = true;
1211 goto rcd_done;
1212 }
1213
1214 pci_unmap_single(adapter->pdev, rbi->dma_addr, rbi->len,
1215 PCI_DMA_FROMDEVICE);
1216
1217 skb_put(ctx->skb, rcd->len);
1218
1219 /* Immediate refill */
1220 new_skb->dev = adapter->netdev;
1221 skb_reserve(new_skb, NET_IP_ALIGN);
1222 rbi->skb = new_skb;
1223 rbi->dma_addr = pci_map_single(adapter->pdev,
1224 rbi->skb->data, rbi->len,
1225 PCI_DMA_FROMDEVICE);
1226 rxd->addr = cpu_to_le64(rbi->dma_addr);
1227 rxd->len = rbi->len;
1228
1229 } else {
1230 BUG_ON(ctx->skb == NULL && !skip_page_frags);
1231
1232 /* non SOP buffer must be type 1 in most cases */
1233 BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_PAGE);
1234 BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_BODY);
1235
1236 /* If an sop buffer was dropped, skip all
1237 * following non-sop fragments. They will be reused.
1238 */
1239 if (skip_page_frags)
1240 goto rcd_done;
1241
1242 new_page = alloc_page(GFP_ATOMIC);
1243 if (unlikely(new_page == NULL)) {
1244 /* Replacement page frag could not be allocated.
1245 * Reuse this page. Drop the pkt and free the
1246 * skb which contained this page as a frag. Skip
1247 * processing all the following non-sop frags.
1248 */
1249 rq->stats.rx_buf_alloc_failure++;
1250 dev_kfree_skb(ctx->skb);
1251 ctx->skb = NULL;
1252 skip_page_frags = true;
1253 goto rcd_done;
1254 }
1255
1256 if (rcd->len) {
1257 pci_unmap_page(adapter->pdev,
1258 rbi->dma_addr, rbi->len,
1259 PCI_DMA_FROMDEVICE);
1260
1261 vmxnet3_append_frag(ctx->skb, rcd, rbi);
1262 }
1263
1264 /* Immediate refill */
1265 rbi->page = new_page;
1266 rbi->dma_addr = pci_map_page(adapter->pdev, rbi->page,
1267 0, PAGE_SIZE,
1268 PCI_DMA_FROMDEVICE);
1269 rxd->addr = cpu_to_le64(rbi->dma_addr);
1270 rxd->len = rbi->len;
1271 }
1272
1273
1274 skb = ctx->skb;
1275 if (rcd->eop) {
1276 skb->len += skb->data_len;
1277
1278 vmxnet3_rx_csum(adapter, skb,
1279 (union Vmxnet3_GenericDesc *)rcd);
1280 skb->protocol = eth_type_trans(skb, adapter->netdev);
1281
1282 if (unlikely(rcd->ts))
1283 __vlan_hwaccel_put_tag(skb, rcd->tci);
1284
1285 if (adapter->netdev->features & NETIF_F_LRO)
1286 netif_receive_skb(skb);
1287 else
1288 napi_gro_receive(&rq->napi, skb);
1289
1290 ctx->skb = NULL;
1291 }
1292
1293 rcd_done:
1294 /* device may have skipped some rx descs */
1295 ring->next2comp = idx;
1296 num_to_alloc = vmxnet3_cmd_ring_desc_avail(ring);
1297 ring = rq->rx_ring + ring_idx;
1298 while (num_to_alloc) {
1299 vmxnet3_getRxDesc(rxd, &ring->base[ring->next2fill].rxd,
1300 &rxCmdDesc);
1301 BUG_ON(!rxd->addr);
1302
1303 /* Recv desc is ready to be used by the device */
1304 rxd->gen = ring->gen;
1305 vmxnet3_cmd_ring_adv_next2fill(ring);
1306 num_to_alloc--;
1307 }
1308
1309 /* if needed, update the register */
1310 if (unlikely(rq->shared->updateRxProd)) {
1311 VMXNET3_WRITE_BAR0_REG(adapter,
1312 rxprod_reg[ring_idx] + rq->qid * 8,
1313 ring->next2fill);
1314 rq->uncommitted[ring_idx] = 0;
1315 }
1316
1317 vmxnet3_comp_ring_adv_next2proc(&rq->comp_ring);
1318 vmxnet3_getRxComp(rcd,
1319 &rq->comp_ring.base[rq->comp_ring.next2proc].rcd, &rxComp);
1320 }
1321
1322 return num_rxd;
1323 }
1324
1325
1326 static void
vmxnet3_rq_cleanup(struct vmxnet3_rx_queue * rq,struct vmxnet3_adapter * adapter)1327 vmxnet3_rq_cleanup(struct vmxnet3_rx_queue *rq,
1328 struct vmxnet3_adapter *adapter)
1329 {
1330 u32 i, ring_idx;
1331 struct Vmxnet3_RxDesc *rxd;
1332
1333 for (ring_idx = 0; ring_idx < 2; ring_idx++) {
1334 for (i = 0; i < rq->rx_ring[ring_idx].size; i++) {
1335 #ifdef __BIG_ENDIAN_BITFIELD
1336 struct Vmxnet3_RxDesc rxDesc;
1337 #endif
1338 vmxnet3_getRxDesc(rxd,
1339 &rq->rx_ring[ring_idx].base[i].rxd, &rxDesc);
1340
1341 if (rxd->btype == VMXNET3_RXD_BTYPE_HEAD &&
1342 rq->buf_info[ring_idx][i].skb) {
1343 pci_unmap_single(adapter->pdev, rxd->addr,
1344 rxd->len, PCI_DMA_FROMDEVICE);
1345 dev_kfree_skb(rq->buf_info[ring_idx][i].skb);
1346 rq->buf_info[ring_idx][i].skb = NULL;
1347 } else if (rxd->btype == VMXNET3_RXD_BTYPE_BODY &&
1348 rq->buf_info[ring_idx][i].page) {
1349 pci_unmap_page(adapter->pdev, rxd->addr,
1350 rxd->len, PCI_DMA_FROMDEVICE);
1351 put_page(rq->buf_info[ring_idx][i].page);
1352 rq->buf_info[ring_idx][i].page = NULL;
1353 }
1354 }
1355
1356 rq->rx_ring[ring_idx].gen = VMXNET3_INIT_GEN;
1357 rq->rx_ring[ring_idx].next2fill =
1358 rq->rx_ring[ring_idx].next2comp = 0;
1359 rq->uncommitted[ring_idx] = 0;
1360 }
1361
1362 rq->comp_ring.gen = VMXNET3_INIT_GEN;
1363 rq->comp_ring.next2proc = 0;
1364 }
1365
1366
1367 static void
vmxnet3_rq_cleanup_all(struct vmxnet3_adapter * adapter)1368 vmxnet3_rq_cleanup_all(struct vmxnet3_adapter *adapter)
1369 {
1370 int i;
1371
1372 for (i = 0; i < adapter->num_rx_queues; i++)
1373 vmxnet3_rq_cleanup(&adapter->rx_queue[i], adapter);
1374 }
1375
1376
vmxnet3_rq_destroy(struct vmxnet3_rx_queue * rq,struct vmxnet3_adapter * adapter)1377 void vmxnet3_rq_destroy(struct vmxnet3_rx_queue *rq,
1378 struct vmxnet3_adapter *adapter)
1379 {
1380 int i;
1381 int j;
1382
1383 /* all rx buffers must have already been freed */
1384 for (i = 0; i < 2; i++) {
1385 if (rq->buf_info[i]) {
1386 for (j = 0; j < rq->rx_ring[i].size; j++)
1387 BUG_ON(rq->buf_info[i][j].page != NULL);
1388 }
1389 }
1390
1391
1392 kfree(rq->buf_info[0]);
1393
1394 for (i = 0; i < 2; i++) {
1395 if (rq->rx_ring[i].base) {
1396 pci_free_consistent(adapter->pdev, rq->rx_ring[i].size
1397 * sizeof(struct Vmxnet3_RxDesc),
1398 rq->rx_ring[i].base,
1399 rq->rx_ring[i].basePA);
1400 rq->rx_ring[i].base = NULL;
1401 }
1402 rq->buf_info[i] = NULL;
1403 }
1404
1405 if (rq->comp_ring.base) {
1406 pci_free_consistent(adapter->pdev, rq->comp_ring.size *
1407 sizeof(struct Vmxnet3_RxCompDesc),
1408 rq->comp_ring.base, rq->comp_ring.basePA);
1409 rq->comp_ring.base = NULL;
1410 }
1411 }
1412
1413
1414 static int
vmxnet3_rq_init(struct vmxnet3_rx_queue * rq,struct vmxnet3_adapter * adapter)1415 vmxnet3_rq_init(struct vmxnet3_rx_queue *rq,
1416 struct vmxnet3_adapter *adapter)
1417 {
1418 int i;
1419
1420 /* initialize buf_info */
1421 for (i = 0; i < rq->rx_ring[0].size; i++) {
1422
1423 /* 1st buf for a pkt is skbuff */
1424 if (i % adapter->rx_buf_per_pkt == 0) {
1425 rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_SKB;
1426 rq->buf_info[0][i].len = adapter->skb_buf_size;
1427 } else { /* subsequent bufs for a pkt is frag */
1428 rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_PAGE;
1429 rq->buf_info[0][i].len = PAGE_SIZE;
1430 }
1431 }
1432 for (i = 0; i < rq->rx_ring[1].size; i++) {
1433 rq->buf_info[1][i].buf_type = VMXNET3_RX_BUF_PAGE;
1434 rq->buf_info[1][i].len = PAGE_SIZE;
1435 }
1436
1437 /* reset internal state and allocate buffers for both rings */
1438 for (i = 0; i < 2; i++) {
1439 rq->rx_ring[i].next2fill = rq->rx_ring[i].next2comp = 0;
1440 rq->uncommitted[i] = 0;
1441
1442 memset(rq->rx_ring[i].base, 0, rq->rx_ring[i].size *
1443 sizeof(struct Vmxnet3_RxDesc));
1444 rq->rx_ring[i].gen = VMXNET3_INIT_GEN;
1445 }
1446 if (vmxnet3_rq_alloc_rx_buf(rq, 0, rq->rx_ring[0].size - 1,
1447 adapter) == 0) {
1448 /* at least has 1 rx buffer for the 1st ring */
1449 return -ENOMEM;
1450 }
1451 vmxnet3_rq_alloc_rx_buf(rq, 1, rq->rx_ring[1].size - 1, adapter);
1452
1453 /* reset the comp ring */
1454 rq->comp_ring.next2proc = 0;
1455 memset(rq->comp_ring.base, 0, rq->comp_ring.size *
1456 sizeof(struct Vmxnet3_RxCompDesc));
1457 rq->comp_ring.gen = VMXNET3_INIT_GEN;
1458
1459 /* reset rxctx */
1460 rq->rx_ctx.skb = NULL;
1461
1462 /* stats are not reset */
1463 return 0;
1464 }
1465
1466
1467 static int
vmxnet3_rq_init_all(struct vmxnet3_adapter * adapter)1468 vmxnet3_rq_init_all(struct vmxnet3_adapter *adapter)
1469 {
1470 int i, err = 0;
1471
1472 for (i = 0; i < adapter->num_rx_queues; i++) {
1473 err = vmxnet3_rq_init(&adapter->rx_queue[i], adapter);
1474 if (unlikely(err)) {
1475 dev_err(&adapter->netdev->dev, "%s: failed to "
1476 "initialize rx queue%i\n",
1477 adapter->netdev->name, i);
1478 break;
1479 }
1480 }
1481 return err;
1482
1483 }
1484
1485
1486 static int
vmxnet3_rq_create(struct vmxnet3_rx_queue * rq,struct vmxnet3_adapter * adapter)1487 vmxnet3_rq_create(struct vmxnet3_rx_queue *rq, struct vmxnet3_adapter *adapter)
1488 {
1489 int i;
1490 size_t sz;
1491 struct vmxnet3_rx_buf_info *bi;
1492
1493 for (i = 0; i < 2; i++) {
1494
1495 sz = rq->rx_ring[i].size * sizeof(struct Vmxnet3_RxDesc);
1496 rq->rx_ring[i].base = pci_alloc_consistent(adapter->pdev, sz,
1497 &rq->rx_ring[i].basePA);
1498 if (!rq->rx_ring[i].base) {
1499 printk(KERN_ERR "%s: failed to allocate rx ring %d\n",
1500 adapter->netdev->name, i);
1501 goto err;
1502 }
1503 }
1504
1505 sz = rq->comp_ring.size * sizeof(struct Vmxnet3_RxCompDesc);
1506 rq->comp_ring.base = pci_alloc_consistent(adapter->pdev, sz,
1507 &rq->comp_ring.basePA);
1508 if (!rq->comp_ring.base) {
1509 printk(KERN_ERR "%s: failed to allocate rx comp ring\n",
1510 adapter->netdev->name);
1511 goto err;
1512 }
1513
1514 sz = sizeof(struct vmxnet3_rx_buf_info) * (rq->rx_ring[0].size +
1515 rq->rx_ring[1].size);
1516 bi = kzalloc(sz, GFP_KERNEL);
1517 if (!bi) {
1518 printk(KERN_ERR "%s: failed to allocate rx bufinfo\n",
1519 adapter->netdev->name);
1520 goto err;
1521 }
1522 rq->buf_info[0] = bi;
1523 rq->buf_info[1] = bi + rq->rx_ring[0].size;
1524
1525 return 0;
1526
1527 err:
1528 vmxnet3_rq_destroy(rq, adapter);
1529 return -ENOMEM;
1530 }
1531
1532
1533 static int
vmxnet3_rq_create_all(struct vmxnet3_adapter * adapter)1534 vmxnet3_rq_create_all(struct vmxnet3_adapter *adapter)
1535 {
1536 int i, err = 0;
1537
1538 for (i = 0; i < adapter->num_rx_queues; i++) {
1539 err = vmxnet3_rq_create(&adapter->rx_queue[i], adapter);
1540 if (unlikely(err)) {
1541 dev_err(&adapter->netdev->dev,
1542 "%s: failed to create rx queue%i\n",
1543 adapter->netdev->name, i);
1544 goto err_out;
1545 }
1546 }
1547 return err;
1548 err_out:
1549 vmxnet3_rq_destroy_all(adapter);
1550 return err;
1551
1552 }
1553
1554 /* Multiple queue aware polling function for tx and rx */
1555
1556 static int
vmxnet3_do_poll(struct vmxnet3_adapter * adapter,int budget)1557 vmxnet3_do_poll(struct vmxnet3_adapter *adapter, int budget)
1558 {
1559 int rcd_done = 0, i;
1560 if (unlikely(adapter->shared->ecr))
1561 vmxnet3_process_events(adapter);
1562 for (i = 0; i < adapter->num_tx_queues; i++)
1563 vmxnet3_tq_tx_complete(&adapter->tx_queue[i], adapter);
1564
1565 for (i = 0; i < adapter->num_rx_queues; i++)
1566 rcd_done += vmxnet3_rq_rx_complete(&adapter->rx_queue[i],
1567 adapter, budget);
1568 return rcd_done;
1569 }
1570
1571
1572 static int
vmxnet3_poll(struct napi_struct * napi,int budget)1573 vmxnet3_poll(struct napi_struct *napi, int budget)
1574 {
1575 struct vmxnet3_rx_queue *rx_queue = container_of(napi,
1576 struct vmxnet3_rx_queue, napi);
1577 int rxd_done;
1578
1579 rxd_done = vmxnet3_do_poll(rx_queue->adapter, budget);
1580
1581 if (rxd_done < budget) {
1582 napi_complete(napi);
1583 vmxnet3_enable_all_intrs(rx_queue->adapter);
1584 }
1585 return rxd_done;
1586 }
1587
1588 /*
1589 * NAPI polling function for MSI-X mode with multiple Rx queues
1590 * Returns the # of the NAPI credit consumed (# of rx descriptors processed)
1591 */
1592
1593 static int
vmxnet3_poll_rx_only(struct napi_struct * napi,int budget)1594 vmxnet3_poll_rx_only(struct napi_struct *napi, int budget)
1595 {
1596 struct vmxnet3_rx_queue *rq = container_of(napi,
1597 struct vmxnet3_rx_queue, napi);
1598 struct vmxnet3_adapter *adapter = rq->adapter;
1599 int rxd_done;
1600
1601 /* When sharing interrupt with corresponding tx queue, process
1602 * tx completions in that queue as well
1603 */
1604 if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE) {
1605 struct vmxnet3_tx_queue *tq =
1606 &adapter->tx_queue[rq - adapter->rx_queue];
1607 vmxnet3_tq_tx_complete(tq, adapter);
1608 }
1609
1610 rxd_done = vmxnet3_rq_rx_complete(rq, adapter, budget);
1611
1612 if (rxd_done < budget) {
1613 napi_complete(napi);
1614 vmxnet3_enable_intr(adapter, rq->comp_ring.intr_idx);
1615 }
1616 return rxd_done;
1617 }
1618
1619
1620 #ifdef CONFIG_PCI_MSI
1621
1622 /*
1623 * Handle completion interrupts on tx queues
1624 * Returns whether or not the intr is handled
1625 */
1626
1627 static irqreturn_t
vmxnet3_msix_tx(int irq,void * data)1628 vmxnet3_msix_tx(int irq, void *data)
1629 {
1630 struct vmxnet3_tx_queue *tq = data;
1631 struct vmxnet3_adapter *adapter = tq->adapter;
1632
1633 if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1634 vmxnet3_disable_intr(adapter, tq->comp_ring.intr_idx);
1635
1636 /* Handle the case where only one irq is allocate for all tx queues */
1637 if (adapter->share_intr == VMXNET3_INTR_TXSHARE) {
1638 int i;
1639 for (i = 0; i < adapter->num_tx_queues; i++) {
1640 struct vmxnet3_tx_queue *txq = &adapter->tx_queue[i];
1641 vmxnet3_tq_tx_complete(txq, adapter);
1642 }
1643 } else {
1644 vmxnet3_tq_tx_complete(tq, adapter);
1645 }
1646 vmxnet3_enable_intr(adapter, tq->comp_ring.intr_idx);
1647
1648 return IRQ_HANDLED;
1649 }
1650
1651
1652 /*
1653 * Handle completion interrupts on rx queues. Returns whether or not the
1654 * intr is handled
1655 */
1656
1657 static irqreturn_t
vmxnet3_msix_rx(int irq,void * data)1658 vmxnet3_msix_rx(int irq, void *data)
1659 {
1660 struct vmxnet3_rx_queue *rq = data;
1661 struct vmxnet3_adapter *adapter = rq->adapter;
1662
1663 /* disable intr if needed */
1664 if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1665 vmxnet3_disable_intr(adapter, rq->comp_ring.intr_idx);
1666 napi_schedule(&rq->napi);
1667
1668 return IRQ_HANDLED;
1669 }
1670
1671 /*
1672 *----------------------------------------------------------------------------
1673 *
1674 * vmxnet3_msix_event --
1675 *
1676 * vmxnet3 msix event intr handler
1677 *
1678 * Result:
1679 * whether or not the intr is handled
1680 *
1681 *----------------------------------------------------------------------------
1682 */
1683
1684 static irqreturn_t
vmxnet3_msix_event(int irq,void * data)1685 vmxnet3_msix_event(int irq, void *data)
1686 {
1687 struct net_device *dev = data;
1688 struct vmxnet3_adapter *adapter = netdev_priv(dev);
1689
1690 /* disable intr if needed */
1691 if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1692 vmxnet3_disable_intr(adapter, adapter->intr.event_intr_idx);
1693
1694 if (adapter->shared->ecr)
1695 vmxnet3_process_events(adapter);
1696
1697 vmxnet3_enable_intr(adapter, adapter->intr.event_intr_idx);
1698
1699 return IRQ_HANDLED;
1700 }
1701
1702 #endif /* CONFIG_PCI_MSI */
1703
1704
1705 /* Interrupt handler for vmxnet3 */
1706 static irqreturn_t
vmxnet3_intr(int irq,void * dev_id)1707 vmxnet3_intr(int irq, void *dev_id)
1708 {
1709 struct net_device *dev = dev_id;
1710 struct vmxnet3_adapter *adapter = netdev_priv(dev);
1711
1712 if (adapter->intr.type == VMXNET3_IT_INTX) {
1713 u32 icr = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_ICR);
1714 if (unlikely(icr == 0))
1715 /* not ours */
1716 return IRQ_NONE;
1717 }
1718
1719
1720 /* disable intr if needed */
1721 if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1722 vmxnet3_disable_all_intrs(adapter);
1723
1724 napi_schedule(&adapter->rx_queue[0].napi);
1725
1726 return IRQ_HANDLED;
1727 }
1728
1729 #ifdef CONFIG_NET_POLL_CONTROLLER
1730
1731 /* netpoll callback. */
1732 static void
vmxnet3_netpoll(struct net_device * netdev)1733 vmxnet3_netpoll(struct net_device *netdev)
1734 {
1735 struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1736
1737 if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1738 vmxnet3_disable_all_intrs(adapter);
1739
1740 vmxnet3_do_poll(adapter, adapter->rx_queue[0].rx_ring[0].size);
1741 vmxnet3_enable_all_intrs(adapter);
1742
1743 }
1744 #endif /* CONFIG_NET_POLL_CONTROLLER */
1745
1746 static int
vmxnet3_request_irqs(struct vmxnet3_adapter * adapter)1747 vmxnet3_request_irqs(struct vmxnet3_adapter *adapter)
1748 {
1749 struct vmxnet3_intr *intr = &adapter->intr;
1750 int err = 0, i;
1751 int vector = 0;
1752
1753 #ifdef CONFIG_PCI_MSI
1754 if (adapter->intr.type == VMXNET3_IT_MSIX) {
1755 for (i = 0; i < adapter->num_tx_queues; i++) {
1756 if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE) {
1757 sprintf(adapter->tx_queue[i].name, "%s-tx-%d",
1758 adapter->netdev->name, vector);
1759 err = request_irq(
1760 intr->msix_entries[vector].vector,
1761 vmxnet3_msix_tx, 0,
1762 adapter->tx_queue[i].name,
1763 &adapter->tx_queue[i]);
1764 } else {
1765 sprintf(adapter->tx_queue[i].name, "%s-rxtx-%d",
1766 adapter->netdev->name, vector);
1767 }
1768 if (err) {
1769 dev_err(&adapter->netdev->dev,
1770 "Failed to request irq for MSIX, %s, "
1771 "error %d\n",
1772 adapter->tx_queue[i].name, err);
1773 return err;
1774 }
1775
1776 /* Handle the case where only 1 MSIx was allocated for
1777 * all tx queues */
1778 if (adapter->share_intr == VMXNET3_INTR_TXSHARE) {
1779 for (; i < adapter->num_tx_queues; i++)
1780 adapter->tx_queue[i].comp_ring.intr_idx
1781 = vector;
1782 vector++;
1783 break;
1784 } else {
1785 adapter->tx_queue[i].comp_ring.intr_idx
1786 = vector++;
1787 }
1788 }
1789 if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE)
1790 vector = 0;
1791
1792 for (i = 0; i < adapter->num_rx_queues; i++) {
1793 if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE)
1794 sprintf(adapter->rx_queue[i].name, "%s-rx-%d",
1795 adapter->netdev->name, vector);
1796 else
1797 sprintf(adapter->rx_queue[i].name, "%s-rxtx-%d",
1798 adapter->netdev->name, vector);
1799 err = request_irq(intr->msix_entries[vector].vector,
1800 vmxnet3_msix_rx, 0,
1801 adapter->rx_queue[i].name,
1802 &(adapter->rx_queue[i]));
1803 if (err) {
1804 printk(KERN_ERR "Failed to request irq for MSIX"
1805 ", %s, error %d\n",
1806 adapter->rx_queue[i].name, err);
1807 return err;
1808 }
1809
1810 adapter->rx_queue[i].comp_ring.intr_idx = vector++;
1811 }
1812
1813 sprintf(intr->event_msi_vector_name, "%s-event-%d",
1814 adapter->netdev->name, vector);
1815 err = request_irq(intr->msix_entries[vector].vector,
1816 vmxnet3_msix_event, 0,
1817 intr->event_msi_vector_name, adapter->netdev);
1818 intr->event_intr_idx = vector;
1819
1820 } else if (intr->type == VMXNET3_IT_MSI) {
1821 adapter->num_rx_queues = 1;
1822 err = request_irq(adapter->pdev->irq, vmxnet3_intr, 0,
1823 adapter->netdev->name, adapter->netdev);
1824 } else {
1825 #endif
1826 adapter->num_rx_queues = 1;
1827 err = request_irq(adapter->pdev->irq, vmxnet3_intr,
1828 IRQF_SHARED, adapter->netdev->name,
1829 adapter->netdev);
1830 #ifdef CONFIG_PCI_MSI
1831 }
1832 #endif
1833 intr->num_intrs = vector + 1;
1834 if (err) {
1835 printk(KERN_ERR "Failed to request irq %s (intr type:%d), error"
1836 ":%d\n", adapter->netdev->name, intr->type, err);
1837 } else {
1838 /* Number of rx queues will not change after this */
1839 for (i = 0; i < adapter->num_rx_queues; i++) {
1840 struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
1841 rq->qid = i;
1842 rq->qid2 = i + adapter->num_rx_queues;
1843 }
1844
1845
1846
1847 /* init our intr settings */
1848 for (i = 0; i < intr->num_intrs; i++)
1849 intr->mod_levels[i] = UPT1_IML_ADAPTIVE;
1850 if (adapter->intr.type != VMXNET3_IT_MSIX) {
1851 adapter->intr.event_intr_idx = 0;
1852 for (i = 0; i < adapter->num_tx_queues; i++)
1853 adapter->tx_queue[i].comp_ring.intr_idx = 0;
1854 adapter->rx_queue[0].comp_ring.intr_idx = 0;
1855 }
1856
1857 printk(KERN_INFO "%s: intr type %u, mode %u, %u vectors "
1858 "allocated\n", adapter->netdev->name, intr->type,
1859 intr->mask_mode, intr->num_intrs);
1860 }
1861
1862 return err;
1863 }
1864
1865
1866 static void
vmxnet3_free_irqs(struct vmxnet3_adapter * adapter)1867 vmxnet3_free_irqs(struct vmxnet3_adapter *adapter)
1868 {
1869 struct vmxnet3_intr *intr = &adapter->intr;
1870 BUG_ON(intr->type == VMXNET3_IT_AUTO || intr->num_intrs <= 0);
1871
1872 switch (intr->type) {
1873 #ifdef CONFIG_PCI_MSI
1874 case VMXNET3_IT_MSIX:
1875 {
1876 int i, vector = 0;
1877
1878 if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE) {
1879 for (i = 0; i < adapter->num_tx_queues; i++) {
1880 free_irq(intr->msix_entries[vector++].vector,
1881 &(adapter->tx_queue[i]));
1882 if (adapter->share_intr == VMXNET3_INTR_TXSHARE)
1883 break;
1884 }
1885 }
1886
1887 for (i = 0; i < adapter->num_rx_queues; i++) {
1888 free_irq(intr->msix_entries[vector++].vector,
1889 &(adapter->rx_queue[i]));
1890 }
1891
1892 free_irq(intr->msix_entries[vector].vector,
1893 adapter->netdev);
1894 BUG_ON(vector >= intr->num_intrs);
1895 break;
1896 }
1897 #endif
1898 case VMXNET3_IT_MSI:
1899 free_irq(adapter->pdev->irq, adapter->netdev);
1900 break;
1901 case VMXNET3_IT_INTX:
1902 free_irq(adapter->pdev->irq, adapter->netdev);
1903 break;
1904 default:
1905 BUG_ON(true);
1906 }
1907 }
1908
1909
1910 static void
vmxnet3_restore_vlan(struct vmxnet3_adapter * adapter)1911 vmxnet3_restore_vlan(struct vmxnet3_adapter *adapter)
1912 {
1913 u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1914 u16 vid;
1915
1916 /* allow untagged pkts */
1917 VMXNET3_SET_VFTABLE_ENTRY(vfTable, 0);
1918
1919 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
1920 VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid);
1921 }
1922
1923
1924 static int
vmxnet3_vlan_rx_add_vid(struct net_device * netdev,u16 vid)1925 vmxnet3_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1926 {
1927 struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1928
1929 if (!(netdev->flags & IFF_PROMISC)) {
1930 u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1931 unsigned long flags;
1932
1933 VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid);
1934 spin_lock_irqsave(&adapter->cmd_lock, flags);
1935 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1936 VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1937 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
1938 }
1939
1940 set_bit(vid, adapter->active_vlans);
1941
1942 return 0;
1943 }
1944
1945
1946 static int
vmxnet3_vlan_rx_kill_vid(struct net_device * netdev,u16 vid)1947 vmxnet3_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1948 {
1949 struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1950
1951 if (!(netdev->flags & IFF_PROMISC)) {
1952 u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1953 unsigned long flags;
1954
1955 VMXNET3_CLEAR_VFTABLE_ENTRY(vfTable, vid);
1956 spin_lock_irqsave(&adapter->cmd_lock, flags);
1957 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1958 VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1959 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
1960 }
1961
1962 clear_bit(vid, adapter->active_vlans);
1963
1964 return 0;
1965 }
1966
1967
1968 static u8 *
vmxnet3_copy_mc(struct net_device * netdev)1969 vmxnet3_copy_mc(struct net_device *netdev)
1970 {
1971 u8 *buf = NULL;
1972 u32 sz = netdev_mc_count(netdev) * ETH_ALEN;
1973
1974 /* struct Vmxnet3_RxFilterConf.mfTableLen is u16. */
1975 if (sz <= 0xffff) {
1976 /* We may be called with BH disabled */
1977 buf = kmalloc(sz, GFP_ATOMIC);
1978 if (buf) {
1979 struct netdev_hw_addr *ha;
1980 int i = 0;
1981
1982 netdev_for_each_mc_addr(ha, netdev)
1983 memcpy(buf + i++ * ETH_ALEN, ha->addr,
1984 ETH_ALEN);
1985 }
1986 }
1987 return buf;
1988 }
1989
1990
1991 static void
vmxnet3_set_mc(struct net_device * netdev)1992 vmxnet3_set_mc(struct net_device *netdev)
1993 {
1994 struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1995 unsigned long flags;
1996 struct Vmxnet3_RxFilterConf *rxConf =
1997 &adapter->shared->devRead.rxFilterConf;
1998 u8 *new_table = NULL;
1999 u32 new_mode = VMXNET3_RXM_UCAST;
2000
2001 if (netdev->flags & IFF_PROMISC) {
2002 u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
2003 memset(vfTable, 0, VMXNET3_VFT_SIZE * sizeof(*vfTable));
2004
2005 new_mode |= VMXNET3_RXM_PROMISC;
2006 } else {
2007 vmxnet3_restore_vlan(adapter);
2008 }
2009
2010 if (netdev->flags & IFF_BROADCAST)
2011 new_mode |= VMXNET3_RXM_BCAST;
2012
2013 if (netdev->flags & IFF_ALLMULTI)
2014 new_mode |= VMXNET3_RXM_ALL_MULTI;
2015 else
2016 if (!netdev_mc_empty(netdev)) {
2017 new_table = vmxnet3_copy_mc(netdev);
2018 if (new_table) {
2019 new_mode |= VMXNET3_RXM_MCAST;
2020 rxConf->mfTableLen = cpu_to_le16(
2021 netdev_mc_count(netdev) * ETH_ALEN);
2022 rxConf->mfTablePA = cpu_to_le64(virt_to_phys(
2023 new_table));
2024 } else {
2025 printk(KERN_INFO "%s: failed to copy mcast list"
2026 ", setting ALL_MULTI\n", netdev->name);
2027 new_mode |= VMXNET3_RXM_ALL_MULTI;
2028 }
2029 }
2030
2031
2032 if (!(new_mode & VMXNET3_RXM_MCAST)) {
2033 rxConf->mfTableLen = 0;
2034 rxConf->mfTablePA = 0;
2035 }
2036
2037 spin_lock_irqsave(&adapter->cmd_lock, flags);
2038 if (new_mode != rxConf->rxMode) {
2039 rxConf->rxMode = cpu_to_le32(new_mode);
2040 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2041 VMXNET3_CMD_UPDATE_RX_MODE);
2042 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2043 VMXNET3_CMD_UPDATE_VLAN_FILTERS);
2044 }
2045
2046 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2047 VMXNET3_CMD_UPDATE_MAC_FILTERS);
2048 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2049
2050 kfree(new_table);
2051 }
2052
2053 void
vmxnet3_rq_destroy_all(struct vmxnet3_adapter * adapter)2054 vmxnet3_rq_destroy_all(struct vmxnet3_adapter *adapter)
2055 {
2056 int i;
2057
2058 for (i = 0; i < adapter->num_rx_queues; i++)
2059 vmxnet3_rq_destroy(&adapter->rx_queue[i], adapter);
2060 }
2061
2062
2063 /*
2064 * Set up driver_shared based on settings in adapter.
2065 */
2066
2067 static void
vmxnet3_setup_driver_shared(struct vmxnet3_adapter * adapter)2068 vmxnet3_setup_driver_shared(struct vmxnet3_adapter *adapter)
2069 {
2070 struct Vmxnet3_DriverShared *shared = adapter->shared;
2071 struct Vmxnet3_DSDevRead *devRead = &shared->devRead;
2072 struct Vmxnet3_TxQueueConf *tqc;
2073 struct Vmxnet3_RxQueueConf *rqc;
2074 int i;
2075
2076 memset(shared, 0, sizeof(*shared));
2077
2078 /* driver settings */
2079 shared->magic = cpu_to_le32(VMXNET3_REV1_MAGIC);
2080 devRead->misc.driverInfo.version = cpu_to_le32(
2081 VMXNET3_DRIVER_VERSION_NUM);
2082 devRead->misc.driverInfo.gos.gosBits = (sizeof(void *) == 4 ?
2083 VMXNET3_GOS_BITS_32 : VMXNET3_GOS_BITS_64);
2084 devRead->misc.driverInfo.gos.gosType = VMXNET3_GOS_TYPE_LINUX;
2085 *((u32 *)&devRead->misc.driverInfo.gos) = cpu_to_le32(
2086 *((u32 *)&devRead->misc.driverInfo.gos));
2087 devRead->misc.driverInfo.vmxnet3RevSpt = cpu_to_le32(1);
2088 devRead->misc.driverInfo.uptVerSpt = cpu_to_le32(1);
2089
2090 devRead->misc.ddPA = cpu_to_le64(virt_to_phys(adapter));
2091 devRead->misc.ddLen = cpu_to_le32(sizeof(struct vmxnet3_adapter));
2092
2093 /* set up feature flags */
2094 if (adapter->netdev->features & NETIF_F_RXCSUM)
2095 devRead->misc.uptFeatures |= UPT1_F_RXCSUM;
2096
2097 if (adapter->netdev->features & NETIF_F_LRO) {
2098 devRead->misc.uptFeatures |= UPT1_F_LRO;
2099 devRead->misc.maxNumRxSG = cpu_to_le16(1 + MAX_SKB_FRAGS);
2100 }
2101 if (adapter->netdev->features & NETIF_F_HW_VLAN_RX)
2102 devRead->misc.uptFeatures |= UPT1_F_RXVLAN;
2103
2104 devRead->misc.mtu = cpu_to_le32(adapter->netdev->mtu);
2105 devRead->misc.queueDescPA = cpu_to_le64(adapter->queue_desc_pa);
2106 devRead->misc.queueDescLen = cpu_to_le32(
2107 adapter->num_tx_queues * sizeof(struct Vmxnet3_TxQueueDesc) +
2108 adapter->num_rx_queues * sizeof(struct Vmxnet3_RxQueueDesc));
2109
2110 /* tx queue settings */
2111 devRead->misc.numTxQueues = adapter->num_tx_queues;
2112 for (i = 0; i < adapter->num_tx_queues; i++) {
2113 struct vmxnet3_tx_queue *tq = &adapter->tx_queue[i];
2114 BUG_ON(adapter->tx_queue[i].tx_ring.base == NULL);
2115 tqc = &adapter->tqd_start[i].conf;
2116 tqc->txRingBasePA = cpu_to_le64(tq->tx_ring.basePA);
2117 tqc->dataRingBasePA = cpu_to_le64(tq->data_ring.basePA);
2118 tqc->compRingBasePA = cpu_to_le64(tq->comp_ring.basePA);
2119 tqc->ddPA = cpu_to_le64(virt_to_phys(tq->buf_info));
2120 tqc->txRingSize = cpu_to_le32(tq->tx_ring.size);
2121 tqc->dataRingSize = cpu_to_le32(tq->data_ring.size);
2122 tqc->compRingSize = cpu_to_le32(tq->comp_ring.size);
2123 tqc->ddLen = cpu_to_le32(
2124 sizeof(struct vmxnet3_tx_buf_info) *
2125 tqc->txRingSize);
2126 tqc->intrIdx = tq->comp_ring.intr_idx;
2127 }
2128
2129 /* rx queue settings */
2130 devRead->misc.numRxQueues = adapter->num_rx_queues;
2131 for (i = 0; i < adapter->num_rx_queues; i++) {
2132 struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
2133 rqc = &adapter->rqd_start[i].conf;
2134 rqc->rxRingBasePA[0] = cpu_to_le64(rq->rx_ring[0].basePA);
2135 rqc->rxRingBasePA[1] = cpu_to_le64(rq->rx_ring[1].basePA);
2136 rqc->compRingBasePA = cpu_to_le64(rq->comp_ring.basePA);
2137 rqc->ddPA = cpu_to_le64(virt_to_phys(
2138 rq->buf_info));
2139 rqc->rxRingSize[0] = cpu_to_le32(rq->rx_ring[0].size);
2140 rqc->rxRingSize[1] = cpu_to_le32(rq->rx_ring[1].size);
2141 rqc->compRingSize = cpu_to_le32(rq->comp_ring.size);
2142 rqc->ddLen = cpu_to_le32(
2143 sizeof(struct vmxnet3_rx_buf_info) *
2144 (rqc->rxRingSize[0] +
2145 rqc->rxRingSize[1]));
2146 rqc->intrIdx = rq->comp_ring.intr_idx;
2147 }
2148
2149 #ifdef VMXNET3_RSS
2150 memset(adapter->rss_conf, 0, sizeof(*adapter->rss_conf));
2151
2152 if (adapter->rss) {
2153 struct UPT1_RSSConf *rssConf = adapter->rss_conf;
2154 devRead->misc.uptFeatures |= UPT1_F_RSS;
2155 devRead->misc.numRxQueues = adapter->num_rx_queues;
2156 rssConf->hashType = UPT1_RSS_HASH_TYPE_TCP_IPV4 |
2157 UPT1_RSS_HASH_TYPE_IPV4 |
2158 UPT1_RSS_HASH_TYPE_TCP_IPV6 |
2159 UPT1_RSS_HASH_TYPE_IPV6;
2160 rssConf->hashFunc = UPT1_RSS_HASH_FUNC_TOEPLITZ;
2161 rssConf->hashKeySize = UPT1_RSS_MAX_KEY_SIZE;
2162 rssConf->indTableSize = VMXNET3_RSS_IND_TABLE_SIZE;
2163 get_random_bytes(&rssConf->hashKey[0], rssConf->hashKeySize);
2164 for (i = 0; i < rssConf->indTableSize; i++)
2165 rssConf->indTable[i] = ethtool_rxfh_indir_default(
2166 i, adapter->num_rx_queues);
2167
2168 devRead->rssConfDesc.confVer = 1;
2169 devRead->rssConfDesc.confLen = sizeof(*rssConf);
2170 devRead->rssConfDesc.confPA = virt_to_phys(rssConf);
2171 }
2172
2173 #endif /* VMXNET3_RSS */
2174
2175 /* intr settings */
2176 devRead->intrConf.autoMask = adapter->intr.mask_mode ==
2177 VMXNET3_IMM_AUTO;
2178 devRead->intrConf.numIntrs = adapter->intr.num_intrs;
2179 for (i = 0; i < adapter->intr.num_intrs; i++)
2180 devRead->intrConf.modLevels[i] = adapter->intr.mod_levels[i];
2181
2182 devRead->intrConf.eventIntrIdx = adapter->intr.event_intr_idx;
2183 devRead->intrConf.intrCtrl |= cpu_to_le32(VMXNET3_IC_DISABLE_ALL);
2184
2185 /* rx filter settings */
2186 devRead->rxFilterConf.rxMode = 0;
2187 vmxnet3_restore_vlan(adapter);
2188 vmxnet3_write_mac_addr(adapter, adapter->netdev->dev_addr);
2189
2190 /* the rest are already zeroed */
2191 }
2192
2193
2194 int
vmxnet3_activate_dev(struct vmxnet3_adapter * adapter)2195 vmxnet3_activate_dev(struct vmxnet3_adapter *adapter)
2196 {
2197 int err, i;
2198 u32 ret;
2199 unsigned long flags;
2200
2201 dev_dbg(&adapter->netdev->dev, "%s: skb_buf_size %d, rx_buf_per_pkt %d,"
2202 " ring sizes %u %u %u\n", adapter->netdev->name,
2203 adapter->skb_buf_size, adapter->rx_buf_per_pkt,
2204 adapter->tx_queue[0].tx_ring.size,
2205 adapter->rx_queue[0].rx_ring[0].size,
2206 adapter->rx_queue[0].rx_ring[1].size);
2207
2208 vmxnet3_tq_init_all(adapter);
2209 err = vmxnet3_rq_init_all(adapter);
2210 if (err) {
2211 printk(KERN_ERR "Failed to init rx queue for %s: error %d\n",
2212 adapter->netdev->name, err);
2213 goto rq_err;
2214 }
2215
2216 err = vmxnet3_request_irqs(adapter);
2217 if (err) {
2218 printk(KERN_ERR "Failed to setup irq for %s: error %d\n",
2219 adapter->netdev->name, err);
2220 goto irq_err;
2221 }
2222
2223 vmxnet3_setup_driver_shared(adapter);
2224
2225 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, VMXNET3_GET_ADDR_LO(
2226 adapter->shared_pa));
2227 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, VMXNET3_GET_ADDR_HI(
2228 adapter->shared_pa));
2229 spin_lock_irqsave(&adapter->cmd_lock, flags);
2230 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2231 VMXNET3_CMD_ACTIVATE_DEV);
2232 ret = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
2233 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2234
2235 if (ret != 0) {
2236 printk(KERN_ERR "Failed to activate dev %s: error %u\n",
2237 adapter->netdev->name, ret);
2238 err = -EINVAL;
2239 goto activate_err;
2240 }
2241
2242 for (i = 0; i < adapter->num_rx_queues; i++) {
2243 VMXNET3_WRITE_BAR0_REG(adapter,
2244 VMXNET3_REG_RXPROD + i * VMXNET3_REG_ALIGN,
2245 adapter->rx_queue[i].rx_ring[0].next2fill);
2246 VMXNET3_WRITE_BAR0_REG(adapter, (VMXNET3_REG_RXPROD2 +
2247 (i * VMXNET3_REG_ALIGN)),
2248 adapter->rx_queue[i].rx_ring[1].next2fill);
2249 }
2250
2251 /* Apply the rx filter settins last. */
2252 vmxnet3_set_mc(adapter->netdev);
2253
2254 /*
2255 * Check link state when first activating device. It will start the
2256 * tx queue if the link is up.
2257 */
2258 vmxnet3_check_link(adapter, true);
2259 for (i = 0; i < adapter->num_rx_queues; i++)
2260 napi_enable(&adapter->rx_queue[i].napi);
2261 vmxnet3_enable_all_intrs(adapter);
2262 clear_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
2263 return 0;
2264
2265 activate_err:
2266 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, 0);
2267 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, 0);
2268 vmxnet3_free_irqs(adapter);
2269 irq_err:
2270 rq_err:
2271 /* free up buffers we allocated */
2272 vmxnet3_rq_cleanup_all(adapter);
2273 return err;
2274 }
2275
2276
2277 void
vmxnet3_reset_dev(struct vmxnet3_adapter * adapter)2278 vmxnet3_reset_dev(struct vmxnet3_adapter *adapter)
2279 {
2280 unsigned long flags;
2281 spin_lock_irqsave(&adapter->cmd_lock, flags);
2282 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, VMXNET3_CMD_RESET_DEV);
2283 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2284 }
2285
2286
2287 int
vmxnet3_quiesce_dev(struct vmxnet3_adapter * adapter)2288 vmxnet3_quiesce_dev(struct vmxnet3_adapter *adapter)
2289 {
2290 int i;
2291 unsigned long flags;
2292 if (test_and_set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state))
2293 return 0;
2294
2295
2296 spin_lock_irqsave(&adapter->cmd_lock, flags);
2297 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2298 VMXNET3_CMD_QUIESCE_DEV);
2299 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2300 vmxnet3_disable_all_intrs(adapter);
2301
2302 for (i = 0; i < adapter->num_rx_queues; i++)
2303 napi_disable(&adapter->rx_queue[i].napi);
2304 netif_tx_disable(adapter->netdev);
2305 adapter->link_speed = 0;
2306 netif_carrier_off(adapter->netdev);
2307
2308 vmxnet3_tq_cleanup_all(adapter);
2309 vmxnet3_rq_cleanup_all(adapter);
2310 vmxnet3_free_irqs(adapter);
2311 return 0;
2312 }
2313
2314
2315 static void
vmxnet3_write_mac_addr(struct vmxnet3_adapter * adapter,u8 * mac)2316 vmxnet3_write_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac)
2317 {
2318 u32 tmp;
2319
2320 tmp = *(u32 *)mac;
2321 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACL, tmp);
2322
2323 tmp = (mac[5] << 8) | mac[4];
2324 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACH, tmp);
2325 }
2326
2327
2328 static int
vmxnet3_set_mac_addr(struct net_device * netdev,void * p)2329 vmxnet3_set_mac_addr(struct net_device *netdev, void *p)
2330 {
2331 struct sockaddr *addr = p;
2332 struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2333
2334 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2335 vmxnet3_write_mac_addr(adapter, addr->sa_data);
2336
2337 return 0;
2338 }
2339
2340
2341 /* ==================== initialization and cleanup routines ============ */
2342
2343 static int
vmxnet3_alloc_pci_resources(struct vmxnet3_adapter * adapter,bool * dma64)2344 vmxnet3_alloc_pci_resources(struct vmxnet3_adapter *adapter, bool *dma64)
2345 {
2346 int err;
2347 unsigned long mmio_start, mmio_len;
2348 struct pci_dev *pdev = adapter->pdev;
2349
2350 err = pci_enable_device(pdev);
2351 if (err) {
2352 printk(KERN_ERR "Failed to enable adapter %s: error %d\n",
2353 pci_name(pdev), err);
2354 return err;
2355 }
2356
2357 if (pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) == 0) {
2358 if (pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64)) != 0) {
2359 printk(KERN_ERR "pci_set_consistent_dma_mask failed "
2360 "for adapter %s\n", pci_name(pdev));
2361 err = -EIO;
2362 goto err_set_mask;
2363 }
2364 *dma64 = true;
2365 } else {
2366 if (pci_set_dma_mask(pdev, DMA_BIT_MASK(32)) != 0) {
2367 printk(KERN_ERR "pci_set_dma_mask failed for adapter "
2368 "%s\n", pci_name(pdev));
2369 err = -EIO;
2370 goto err_set_mask;
2371 }
2372 *dma64 = false;
2373 }
2374
2375 err = pci_request_selected_regions(pdev, (1 << 2) - 1,
2376 vmxnet3_driver_name);
2377 if (err) {
2378 printk(KERN_ERR "Failed to request region for adapter %s: "
2379 "error %d\n", pci_name(pdev), err);
2380 goto err_set_mask;
2381 }
2382
2383 pci_set_master(pdev);
2384
2385 mmio_start = pci_resource_start(pdev, 0);
2386 mmio_len = pci_resource_len(pdev, 0);
2387 adapter->hw_addr0 = ioremap(mmio_start, mmio_len);
2388 if (!adapter->hw_addr0) {
2389 printk(KERN_ERR "Failed to map bar0 for adapter %s\n",
2390 pci_name(pdev));
2391 err = -EIO;
2392 goto err_ioremap;
2393 }
2394
2395 mmio_start = pci_resource_start(pdev, 1);
2396 mmio_len = pci_resource_len(pdev, 1);
2397 adapter->hw_addr1 = ioremap(mmio_start, mmio_len);
2398 if (!adapter->hw_addr1) {
2399 printk(KERN_ERR "Failed to map bar1 for adapter %s\n",
2400 pci_name(pdev));
2401 err = -EIO;
2402 goto err_bar1;
2403 }
2404 return 0;
2405
2406 err_bar1:
2407 iounmap(adapter->hw_addr0);
2408 err_ioremap:
2409 pci_release_selected_regions(pdev, (1 << 2) - 1);
2410 err_set_mask:
2411 pci_disable_device(pdev);
2412 return err;
2413 }
2414
2415
2416 static void
vmxnet3_free_pci_resources(struct vmxnet3_adapter * adapter)2417 vmxnet3_free_pci_resources(struct vmxnet3_adapter *adapter)
2418 {
2419 BUG_ON(!adapter->pdev);
2420
2421 iounmap(adapter->hw_addr0);
2422 iounmap(adapter->hw_addr1);
2423 pci_release_selected_regions(adapter->pdev, (1 << 2) - 1);
2424 pci_disable_device(adapter->pdev);
2425 }
2426
2427
2428 static void
vmxnet3_adjust_rx_ring_size(struct vmxnet3_adapter * adapter)2429 vmxnet3_adjust_rx_ring_size(struct vmxnet3_adapter *adapter)
2430 {
2431 size_t sz, i, ring0_size, ring1_size, comp_size;
2432 struct vmxnet3_rx_queue *rq = &adapter->rx_queue[0];
2433
2434
2435 if (adapter->netdev->mtu <= VMXNET3_MAX_SKB_BUF_SIZE -
2436 VMXNET3_MAX_ETH_HDR_SIZE) {
2437 adapter->skb_buf_size = adapter->netdev->mtu +
2438 VMXNET3_MAX_ETH_HDR_SIZE;
2439 if (adapter->skb_buf_size < VMXNET3_MIN_T0_BUF_SIZE)
2440 adapter->skb_buf_size = VMXNET3_MIN_T0_BUF_SIZE;
2441
2442 adapter->rx_buf_per_pkt = 1;
2443 } else {
2444 adapter->skb_buf_size = VMXNET3_MAX_SKB_BUF_SIZE;
2445 sz = adapter->netdev->mtu - VMXNET3_MAX_SKB_BUF_SIZE +
2446 VMXNET3_MAX_ETH_HDR_SIZE;
2447 adapter->rx_buf_per_pkt = 1 + (sz + PAGE_SIZE - 1) / PAGE_SIZE;
2448 }
2449
2450 /*
2451 * for simplicity, force the ring0 size to be a multiple of
2452 * rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN
2453 */
2454 sz = adapter->rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN;
2455 ring0_size = adapter->rx_queue[0].rx_ring[0].size;
2456 ring0_size = (ring0_size + sz - 1) / sz * sz;
2457 ring0_size = min_t(u32, ring0_size, VMXNET3_RX_RING_MAX_SIZE /
2458 sz * sz);
2459 ring1_size = adapter->rx_queue[0].rx_ring[1].size;
2460 comp_size = ring0_size + ring1_size;
2461
2462 for (i = 0; i < adapter->num_rx_queues; i++) {
2463 rq = &adapter->rx_queue[i];
2464 rq->rx_ring[0].size = ring0_size;
2465 rq->rx_ring[1].size = ring1_size;
2466 rq->comp_ring.size = comp_size;
2467 }
2468 }
2469
2470
2471 int
vmxnet3_create_queues(struct vmxnet3_adapter * adapter,u32 tx_ring_size,u32 rx_ring_size,u32 rx_ring2_size)2472 vmxnet3_create_queues(struct vmxnet3_adapter *adapter, u32 tx_ring_size,
2473 u32 rx_ring_size, u32 rx_ring2_size)
2474 {
2475 int err = 0, i;
2476
2477 for (i = 0; i < adapter->num_tx_queues; i++) {
2478 struct vmxnet3_tx_queue *tq = &adapter->tx_queue[i];
2479 tq->tx_ring.size = tx_ring_size;
2480 tq->data_ring.size = tx_ring_size;
2481 tq->comp_ring.size = tx_ring_size;
2482 tq->shared = &adapter->tqd_start[i].ctrl;
2483 tq->stopped = true;
2484 tq->adapter = adapter;
2485 tq->qid = i;
2486 err = vmxnet3_tq_create(tq, adapter);
2487 /*
2488 * Too late to change num_tx_queues. We cannot do away with
2489 * lesser number of queues than what we asked for
2490 */
2491 if (err)
2492 goto queue_err;
2493 }
2494
2495 adapter->rx_queue[0].rx_ring[0].size = rx_ring_size;
2496 adapter->rx_queue[0].rx_ring[1].size = rx_ring2_size;
2497 vmxnet3_adjust_rx_ring_size(adapter);
2498 for (i = 0; i < adapter->num_rx_queues; i++) {
2499 struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
2500 /* qid and qid2 for rx queues will be assigned later when num
2501 * of rx queues is finalized after allocating intrs */
2502 rq->shared = &adapter->rqd_start[i].ctrl;
2503 rq->adapter = adapter;
2504 err = vmxnet3_rq_create(rq, adapter);
2505 if (err) {
2506 if (i == 0) {
2507 printk(KERN_ERR "Could not allocate any rx"
2508 "queues. Aborting.\n");
2509 goto queue_err;
2510 } else {
2511 printk(KERN_INFO "Number of rx queues changed "
2512 "to : %d.\n", i);
2513 adapter->num_rx_queues = i;
2514 err = 0;
2515 break;
2516 }
2517 }
2518 }
2519 return err;
2520 queue_err:
2521 vmxnet3_tq_destroy_all(adapter);
2522 return err;
2523 }
2524
2525 static int
vmxnet3_open(struct net_device * netdev)2526 vmxnet3_open(struct net_device *netdev)
2527 {
2528 struct vmxnet3_adapter *adapter;
2529 int err, i;
2530
2531 adapter = netdev_priv(netdev);
2532
2533 for (i = 0; i < adapter->num_tx_queues; i++)
2534 spin_lock_init(&adapter->tx_queue[i].tx_lock);
2535
2536 err = vmxnet3_create_queues(adapter, VMXNET3_DEF_TX_RING_SIZE,
2537 VMXNET3_DEF_RX_RING_SIZE,
2538 VMXNET3_DEF_RX_RING_SIZE);
2539 if (err)
2540 goto queue_err;
2541
2542 err = vmxnet3_activate_dev(adapter);
2543 if (err)
2544 goto activate_err;
2545
2546 return 0;
2547
2548 activate_err:
2549 vmxnet3_rq_destroy_all(adapter);
2550 vmxnet3_tq_destroy_all(adapter);
2551 queue_err:
2552 return err;
2553 }
2554
2555
2556 static int
vmxnet3_close(struct net_device * netdev)2557 vmxnet3_close(struct net_device *netdev)
2558 {
2559 struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2560
2561 /*
2562 * Reset_work may be in the middle of resetting the device, wait for its
2563 * completion.
2564 */
2565 while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2566 msleep(1);
2567
2568 vmxnet3_quiesce_dev(adapter);
2569
2570 vmxnet3_rq_destroy_all(adapter);
2571 vmxnet3_tq_destroy_all(adapter);
2572
2573 clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2574
2575
2576 return 0;
2577 }
2578
2579
2580 void
vmxnet3_force_close(struct vmxnet3_adapter * adapter)2581 vmxnet3_force_close(struct vmxnet3_adapter *adapter)
2582 {
2583 int i;
2584
2585 /*
2586 * we must clear VMXNET3_STATE_BIT_RESETTING, otherwise
2587 * vmxnet3_close() will deadlock.
2588 */
2589 BUG_ON(test_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state));
2590
2591 /* we need to enable NAPI, otherwise dev_close will deadlock */
2592 for (i = 0; i < adapter->num_rx_queues; i++)
2593 napi_enable(&adapter->rx_queue[i].napi);
2594 dev_close(adapter->netdev);
2595 }
2596
2597
2598 static int
vmxnet3_change_mtu(struct net_device * netdev,int new_mtu)2599 vmxnet3_change_mtu(struct net_device *netdev, int new_mtu)
2600 {
2601 struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2602 int err = 0;
2603
2604 if (new_mtu < VMXNET3_MIN_MTU || new_mtu > VMXNET3_MAX_MTU)
2605 return -EINVAL;
2606
2607 netdev->mtu = new_mtu;
2608
2609 /*
2610 * Reset_work may be in the middle of resetting the device, wait for its
2611 * completion.
2612 */
2613 while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2614 msleep(1);
2615
2616 if (netif_running(netdev)) {
2617 vmxnet3_quiesce_dev(adapter);
2618 vmxnet3_reset_dev(adapter);
2619
2620 /* we need to re-create the rx queue based on the new mtu */
2621 vmxnet3_rq_destroy_all(adapter);
2622 vmxnet3_adjust_rx_ring_size(adapter);
2623 err = vmxnet3_rq_create_all(adapter);
2624 if (err) {
2625 printk(KERN_ERR "%s: failed to re-create rx queues,"
2626 " error %d. Closing it.\n", netdev->name, err);
2627 goto out;
2628 }
2629
2630 err = vmxnet3_activate_dev(adapter);
2631 if (err) {
2632 printk(KERN_ERR "%s: failed to re-activate, error %d. "
2633 "Closing it\n", netdev->name, err);
2634 goto out;
2635 }
2636 }
2637
2638 out:
2639 clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2640 if (err)
2641 vmxnet3_force_close(adapter);
2642
2643 return err;
2644 }
2645
2646
2647 static void
vmxnet3_declare_features(struct vmxnet3_adapter * adapter,bool dma64)2648 vmxnet3_declare_features(struct vmxnet3_adapter *adapter, bool dma64)
2649 {
2650 struct net_device *netdev = adapter->netdev;
2651
2652 netdev->hw_features = NETIF_F_SG | NETIF_F_RXCSUM |
2653 NETIF_F_HW_CSUM | NETIF_F_HW_VLAN_TX |
2654 NETIF_F_HW_VLAN_RX | NETIF_F_TSO | NETIF_F_TSO6 |
2655 NETIF_F_LRO;
2656 if (dma64)
2657 netdev->hw_features |= NETIF_F_HIGHDMA;
2658 netdev->vlan_features = netdev->hw_features &
2659 ~(NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX);
2660 netdev->features = netdev->hw_features | NETIF_F_HW_VLAN_FILTER;
2661
2662 netdev_info(adapter->netdev,
2663 "features: sg csum vlan jf tso tsoIPv6 lro%s\n",
2664 dma64 ? " highDMA" : "");
2665 }
2666
2667
2668 static void
vmxnet3_read_mac_addr(struct vmxnet3_adapter * adapter,u8 * mac)2669 vmxnet3_read_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac)
2670 {
2671 u32 tmp;
2672
2673 tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACL);
2674 *(u32 *)mac = tmp;
2675
2676 tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACH);
2677 mac[4] = tmp & 0xff;
2678 mac[5] = (tmp >> 8) & 0xff;
2679 }
2680
2681 #ifdef CONFIG_PCI_MSI
2682
2683 /*
2684 * Enable MSIx vectors.
2685 * Returns :
2686 * 0 on successful enabling of required vectors,
2687 * VMXNET3_LINUX_MIN_MSIX_VECT when only minimum number of vectors required
2688 * could be enabled.
2689 * number of vectors which can be enabled otherwise (this number is smaller
2690 * than VMXNET3_LINUX_MIN_MSIX_VECT)
2691 */
2692
2693 static int
vmxnet3_acquire_msix_vectors(struct vmxnet3_adapter * adapter,int vectors)2694 vmxnet3_acquire_msix_vectors(struct vmxnet3_adapter *adapter,
2695 int vectors)
2696 {
2697 int err = 0, vector_threshold;
2698 vector_threshold = VMXNET3_LINUX_MIN_MSIX_VECT;
2699
2700 while (vectors >= vector_threshold) {
2701 err = pci_enable_msix(adapter->pdev, adapter->intr.msix_entries,
2702 vectors);
2703 if (!err) {
2704 adapter->intr.num_intrs = vectors;
2705 return 0;
2706 } else if (err < 0) {
2707 printk(KERN_ERR "Failed to enable MSI-X for %s, error"
2708 " %d\n", adapter->netdev->name, err);
2709 vectors = 0;
2710 } else if (err < vector_threshold) {
2711 break;
2712 } else {
2713 /* If fails to enable required number of MSI-x vectors
2714 * try enabling minimum number of vectors required.
2715 */
2716 vectors = vector_threshold;
2717 printk(KERN_ERR "Failed to enable %d MSI-X for %s, try"
2718 " %d instead\n", vectors, adapter->netdev->name,
2719 vector_threshold);
2720 }
2721 }
2722
2723 printk(KERN_INFO "Number of MSI-X interrupts which can be allocatedi"
2724 " are lower than min threshold required.\n");
2725 return err;
2726 }
2727
2728
2729 #endif /* CONFIG_PCI_MSI */
2730
2731 static void
vmxnet3_alloc_intr_resources(struct vmxnet3_adapter * adapter)2732 vmxnet3_alloc_intr_resources(struct vmxnet3_adapter *adapter)
2733 {
2734 u32 cfg;
2735 unsigned long flags;
2736
2737 /* intr settings */
2738 spin_lock_irqsave(&adapter->cmd_lock, flags);
2739 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2740 VMXNET3_CMD_GET_CONF_INTR);
2741 cfg = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
2742 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2743 adapter->intr.type = cfg & 0x3;
2744 adapter->intr.mask_mode = (cfg >> 2) & 0x3;
2745
2746 if (adapter->intr.type == VMXNET3_IT_AUTO) {
2747 adapter->intr.type = VMXNET3_IT_MSIX;
2748 }
2749
2750 #ifdef CONFIG_PCI_MSI
2751 if (adapter->intr.type == VMXNET3_IT_MSIX) {
2752 int vector, err = 0;
2753
2754 adapter->intr.num_intrs = (adapter->share_intr ==
2755 VMXNET3_INTR_TXSHARE) ? 1 :
2756 adapter->num_tx_queues;
2757 adapter->intr.num_intrs += (adapter->share_intr ==
2758 VMXNET3_INTR_BUDDYSHARE) ? 0 :
2759 adapter->num_rx_queues;
2760 adapter->intr.num_intrs += 1; /* for link event */
2761
2762 adapter->intr.num_intrs = (adapter->intr.num_intrs >
2763 VMXNET3_LINUX_MIN_MSIX_VECT
2764 ? adapter->intr.num_intrs :
2765 VMXNET3_LINUX_MIN_MSIX_VECT);
2766
2767 for (vector = 0; vector < adapter->intr.num_intrs; vector++)
2768 adapter->intr.msix_entries[vector].entry = vector;
2769
2770 err = vmxnet3_acquire_msix_vectors(adapter,
2771 adapter->intr.num_intrs);
2772 /* If we cannot allocate one MSIx vector per queue
2773 * then limit the number of rx queues to 1
2774 */
2775 if (err == VMXNET3_LINUX_MIN_MSIX_VECT) {
2776 if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE
2777 || adapter->num_rx_queues != 1) {
2778 adapter->share_intr = VMXNET3_INTR_TXSHARE;
2779 printk(KERN_ERR "Number of rx queues : 1\n");
2780 adapter->num_rx_queues = 1;
2781 adapter->intr.num_intrs =
2782 VMXNET3_LINUX_MIN_MSIX_VECT;
2783 }
2784 return;
2785 }
2786 if (!err)
2787 return;
2788
2789 /* If we cannot allocate MSIx vectors use only one rx queue */
2790 printk(KERN_INFO "Failed to enable MSI-X for %s, error %d."
2791 "#rx queues : 1, try MSI\n", adapter->netdev->name, err);
2792
2793 adapter->intr.type = VMXNET3_IT_MSI;
2794 }
2795
2796 if (adapter->intr.type == VMXNET3_IT_MSI) {
2797 int err;
2798 err = pci_enable_msi(adapter->pdev);
2799 if (!err) {
2800 adapter->num_rx_queues = 1;
2801 adapter->intr.num_intrs = 1;
2802 return;
2803 }
2804 }
2805 #endif /* CONFIG_PCI_MSI */
2806
2807 adapter->num_rx_queues = 1;
2808 printk(KERN_INFO "Using INTx interrupt, #Rx queues: 1.\n");
2809 adapter->intr.type = VMXNET3_IT_INTX;
2810
2811 /* INT-X related setting */
2812 adapter->intr.num_intrs = 1;
2813 }
2814
2815
2816 static void
vmxnet3_free_intr_resources(struct vmxnet3_adapter * adapter)2817 vmxnet3_free_intr_resources(struct vmxnet3_adapter *adapter)
2818 {
2819 if (adapter->intr.type == VMXNET3_IT_MSIX)
2820 pci_disable_msix(adapter->pdev);
2821 else if (adapter->intr.type == VMXNET3_IT_MSI)
2822 pci_disable_msi(adapter->pdev);
2823 else
2824 BUG_ON(adapter->intr.type != VMXNET3_IT_INTX);
2825 }
2826
2827
2828 static void
vmxnet3_tx_timeout(struct net_device * netdev)2829 vmxnet3_tx_timeout(struct net_device *netdev)
2830 {
2831 struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2832 adapter->tx_timeout_count++;
2833
2834 printk(KERN_ERR "%s: tx hang\n", adapter->netdev->name);
2835 schedule_work(&adapter->work);
2836 netif_wake_queue(adapter->netdev);
2837 }
2838
2839
2840 static void
vmxnet3_reset_work(struct work_struct * data)2841 vmxnet3_reset_work(struct work_struct *data)
2842 {
2843 struct vmxnet3_adapter *adapter;
2844
2845 adapter = container_of(data, struct vmxnet3_adapter, work);
2846
2847 /* if another thread is resetting the device, no need to proceed */
2848 if (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2849 return;
2850
2851 /* if the device is closed, we must leave it alone */
2852 rtnl_lock();
2853 if (netif_running(adapter->netdev)) {
2854 printk(KERN_INFO "%s: resetting\n", adapter->netdev->name);
2855 vmxnet3_quiesce_dev(adapter);
2856 vmxnet3_reset_dev(adapter);
2857 vmxnet3_activate_dev(adapter);
2858 } else {
2859 printk(KERN_INFO "%s: already closed\n", adapter->netdev->name);
2860 }
2861 rtnl_unlock();
2862
2863 clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2864 }
2865
2866
2867 static int __devinit
vmxnet3_probe_device(struct pci_dev * pdev,const struct pci_device_id * id)2868 vmxnet3_probe_device(struct pci_dev *pdev,
2869 const struct pci_device_id *id)
2870 {
2871 static const struct net_device_ops vmxnet3_netdev_ops = {
2872 .ndo_open = vmxnet3_open,
2873 .ndo_stop = vmxnet3_close,
2874 .ndo_start_xmit = vmxnet3_xmit_frame,
2875 .ndo_set_mac_address = vmxnet3_set_mac_addr,
2876 .ndo_change_mtu = vmxnet3_change_mtu,
2877 .ndo_set_features = vmxnet3_set_features,
2878 .ndo_get_stats64 = vmxnet3_get_stats64,
2879 .ndo_tx_timeout = vmxnet3_tx_timeout,
2880 .ndo_set_rx_mode = vmxnet3_set_mc,
2881 .ndo_vlan_rx_add_vid = vmxnet3_vlan_rx_add_vid,
2882 .ndo_vlan_rx_kill_vid = vmxnet3_vlan_rx_kill_vid,
2883 #ifdef CONFIG_NET_POLL_CONTROLLER
2884 .ndo_poll_controller = vmxnet3_netpoll,
2885 #endif
2886 };
2887 int err;
2888 bool dma64 = false; /* stupid gcc */
2889 u32 ver;
2890 struct net_device *netdev;
2891 struct vmxnet3_adapter *adapter;
2892 u8 mac[ETH_ALEN];
2893 int size;
2894 int num_tx_queues;
2895 int num_rx_queues;
2896
2897 if (!pci_msi_enabled())
2898 enable_mq = 0;
2899
2900 #ifdef VMXNET3_RSS
2901 if (enable_mq)
2902 num_rx_queues = min(VMXNET3_DEVICE_MAX_RX_QUEUES,
2903 (int)num_online_cpus());
2904 else
2905 #endif
2906 num_rx_queues = 1;
2907 num_rx_queues = rounddown_pow_of_two(num_rx_queues);
2908
2909 if (enable_mq)
2910 num_tx_queues = min(VMXNET3_DEVICE_MAX_TX_QUEUES,
2911 (int)num_online_cpus());
2912 else
2913 num_tx_queues = 1;
2914
2915 num_tx_queues = rounddown_pow_of_two(num_tx_queues);
2916 netdev = alloc_etherdev_mq(sizeof(struct vmxnet3_adapter),
2917 max(num_tx_queues, num_rx_queues));
2918 printk(KERN_INFO "# of Tx queues : %d, # of Rx queues : %d\n",
2919 num_tx_queues, num_rx_queues);
2920
2921 if (!netdev) {
2922 printk(KERN_ERR "Failed to alloc ethernet device for adapter "
2923 "%s\n", pci_name(pdev));
2924 return -ENOMEM;
2925 }
2926
2927 pci_set_drvdata(pdev, netdev);
2928 adapter = netdev_priv(netdev);
2929 adapter->netdev = netdev;
2930 adapter->pdev = pdev;
2931
2932 spin_lock_init(&adapter->cmd_lock);
2933 adapter->shared = pci_alloc_consistent(adapter->pdev,
2934 sizeof(struct Vmxnet3_DriverShared),
2935 &adapter->shared_pa);
2936 if (!adapter->shared) {
2937 printk(KERN_ERR "Failed to allocate memory for %s\n",
2938 pci_name(pdev));
2939 err = -ENOMEM;
2940 goto err_alloc_shared;
2941 }
2942
2943 adapter->num_rx_queues = num_rx_queues;
2944 adapter->num_tx_queues = num_tx_queues;
2945
2946 size = sizeof(struct Vmxnet3_TxQueueDesc) * adapter->num_tx_queues;
2947 size += sizeof(struct Vmxnet3_RxQueueDesc) * adapter->num_rx_queues;
2948 adapter->tqd_start = pci_alloc_consistent(adapter->pdev, size,
2949 &adapter->queue_desc_pa);
2950
2951 if (!adapter->tqd_start) {
2952 printk(KERN_ERR "Failed to allocate memory for %s\n",
2953 pci_name(pdev));
2954 err = -ENOMEM;
2955 goto err_alloc_queue_desc;
2956 }
2957 adapter->rqd_start = (struct Vmxnet3_RxQueueDesc *)(adapter->tqd_start +
2958 adapter->num_tx_queues);
2959
2960 adapter->pm_conf = kmalloc(sizeof(struct Vmxnet3_PMConf), GFP_KERNEL);
2961 if (adapter->pm_conf == NULL) {
2962 printk(KERN_ERR "Failed to allocate memory for %s\n",
2963 pci_name(pdev));
2964 err = -ENOMEM;
2965 goto err_alloc_pm;
2966 }
2967
2968 #ifdef VMXNET3_RSS
2969
2970 adapter->rss_conf = kmalloc(sizeof(struct UPT1_RSSConf), GFP_KERNEL);
2971 if (adapter->rss_conf == NULL) {
2972 printk(KERN_ERR "Failed to allocate memory for %s\n",
2973 pci_name(pdev));
2974 err = -ENOMEM;
2975 goto err_alloc_rss;
2976 }
2977 #endif /* VMXNET3_RSS */
2978
2979 err = vmxnet3_alloc_pci_resources(adapter, &dma64);
2980 if (err < 0)
2981 goto err_alloc_pci;
2982
2983 ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_VRRS);
2984 if (ver & 1) {
2985 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_VRRS, 1);
2986 } else {
2987 printk(KERN_ERR "Incompatible h/w version (0x%x) for adapter"
2988 " %s\n", ver, pci_name(pdev));
2989 err = -EBUSY;
2990 goto err_ver;
2991 }
2992
2993 ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_UVRS);
2994 if (ver & 1) {
2995 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_UVRS, 1);
2996 } else {
2997 printk(KERN_ERR "Incompatible upt version (0x%x) for "
2998 "adapter %s\n", ver, pci_name(pdev));
2999 err = -EBUSY;
3000 goto err_ver;
3001 }
3002
3003 SET_NETDEV_DEV(netdev, &pdev->dev);
3004 vmxnet3_declare_features(adapter, dma64);
3005
3006 adapter->dev_number = atomic_read(&devices_found);
3007
3008 adapter->share_intr = irq_share_mode;
3009 if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE &&
3010 adapter->num_tx_queues != adapter->num_rx_queues)
3011 adapter->share_intr = VMXNET3_INTR_DONTSHARE;
3012
3013 vmxnet3_alloc_intr_resources(adapter);
3014
3015 #ifdef VMXNET3_RSS
3016 if (adapter->num_rx_queues > 1 &&
3017 adapter->intr.type == VMXNET3_IT_MSIX) {
3018 adapter->rss = true;
3019 printk(KERN_INFO "RSS is enabled.\n");
3020 } else {
3021 adapter->rss = false;
3022 }
3023 #endif
3024
3025 vmxnet3_read_mac_addr(adapter, mac);
3026 memcpy(netdev->dev_addr, mac, netdev->addr_len);
3027
3028 netdev->netdev_ops = &vmxnet3_netdev_ops;
3029 vmxnet3_set_ethtool_ops(netdev);
3030 netdev->watchdog_timeo = 5 * HZ;
3031
3032 INIT_WORK(&adapter->work, vmxnet3_reset_work);
3033
3034 if (adapter->intr.type == VMXNET3_IT_MSIX) {
3035 int i;
3036 for (i = 0; i < adapter->num_rx_queues; i++) {
3037 netif_napi_add(adapter->netdev,
3038 &adapter->rx_queue[i].napi,
3039 vmxnet3_poll_rx_only, 64);
3040 }
3041 } else {
3042 netif_napi_add(adapter->netdev, &adapter->rx_queue[0].napi,
3043 vmxnet3_poll, 64);
3044 }
3045
3046 netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
3047 netif_set_real_num_rx_queues(adapter->netdev, adapter->num_rx_queues);
3048
3049 err = register_netdev(netdev);
3050
3051 if (err) {
3052 printk(KERN_ERR "Failed to register adapter %s\n",
3053 pci_name(pdev));
3054 goto err_register;
3055 }
3056
3057 set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
3058 vmxnet3_check_link(adapter, false);
3059 atomic_inc(&devices_found);
3060 return 0;
3061
3062 err_register:
3063 vmxnet3_free_intr_resources(adapter);
3064 err_ver:
3065 vmxnet3_free_pci_resources(adapter);
3066 err_alloc_pci:
3067 #ifdef VMXNET3_RSS
3068 kfree(adapter->rss_conf);
3069 err_alloc_rss:
3070 #endif
3071 kfree(adapter->pm_conf);
3072 err_alloc_pm:
3073 pci_free_consistent(adapter->pdev, size, adapter->tqd_start,
3074 adapter->queue_desc_pa);
3075 err_alloc_queue_desc:
3076 pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_DriverShared),
3077 adapter->shared, adapter->shared_pa);
3078 err_alloc_shared:
3079 pci_set_drvdata(pdev, NULL);
3080 free_netdev(netdev);
3081 return err;
3082 }
3083
3084
3085 static void __devexit
vmxnet3_remove_device(struct pci_dev * pdev)3086 vmxnet3_remove_device(struct pci_dev *pdev)
3087 {
3088 struct net_device *netdev = pci_get_drvdata(pdev);
3089 struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3090 int size = 0;
3091 int num_rx_queues;
3092
3093 #ifdef VMXNET3_RSS
3094 if (enable_mq)
3095 num_rx_queues = min(VMXNET3_DEVICE_MAX_RX_QUEUES,
3096 (int)num_online_cpus());
3097 else
3098 #endif
3099 num_rx_queues = 1;
3100 num_rx_queues = rounddown_pow_of_two(num_rx_queues);
3101
3102 cancel_work_sync(&adapter->work);
3103
3104 unregister_netdev(netdev);
3105
3106 vmxnet3_free_intr_resources(adapter);
3107 vmxnet3_free_pci_resources(adapter);
3108 #ifdef VMXNET3_RSS
3109 kfree(adapter->rss_conf);
3110 #endif
3111 kfree(adapter->pm_conf);
3112
3113 size = sizeof(struct Vmxnet3_TxQueueDesc) * adapter->num_tx_queues;
3114 size += sizeof(struct Vmxnet3_RxQueueDesc) * num_rx_queues;
3115 pci_free_consistent(adapter->pdev, size, adapter->tqd_start,
3116 adapter->queue_desc_pa);
3117 pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_DriverShared),
3118 adapter->shared, adapter->shared_pa);
3119 free_netdev(netdev);
3120 }
3121
3122
3123 #ifdef CONFIG_PM
3124
3125 static int
vmxnet3_suspend(struct device * device)3126 vmxnet3_suspend(struct device *device)
3127 {
3128 struct pci_dev *pdev = to_pci_dev(device);
3129 struct net_device *netdev = pci_get_drvdata(pdev);
3130 struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3131 struct Vmxnet3_PMConf *pmConf;
3132 struct ethhdr *ehdr;
3133 struct arphdr *ahdr;
3134 u8 *arpreq;
3135 struct in_device *in_dev;
3136 struct in_ifaddr *ifa;
3137 unsigned long flags;
3138 int i = 0;
3139
3140 if (!netif_running(netdev))
3141 return 0;
3142
3143 for (i = 0; i < adapter->num_rx_queues; i++)
3144 napi_disable(&adapter->rx_queue[i].napi);
3145
3146 vmxnet3_disable_all_intrs(adapter);
3147 vmxnet3_free_irqs(adapter);
3148 vmxnet3_free_intr_resources(adapter);
3149
3150 netif_device_detach(netdev);
3151 netif_tx_stop_all_queues(netdev);
3152
3153 /* Create wake-up filters. */
3154 pmConf = adapter->pm_conf;
3155 memset(pmConf, 0, sizeof(*pmConf));
3156
3157 if (adapter->wol & WAKE_UCAST) {
3158 pmConf->filters[i].patternSize = ETH_ALEN;
3159 pmConf->filters[i].maskSize = 1;
3160 memcpy(pmConf->filters[i].pattern, netdev->dev_addr, ETH_ALEN);
3161 pmConf->filters[i].mask[0] = 0x3F; /* LSB ETH_ALEN bits */
3162
3163 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_FILTER;
3164 i++;
3165 }
3166
3167 if (adapter->wol & WAKE_ARP) {
3168 in_dev = in_dev_get(netdev);
3169 if (!in_dev)
3170 goto skip_arp;
3171
3172 ifa = (struct in_ifaddr *)in_dev->ifa_list;
3173 if (!ifa)
3174 goto skip_arp;
3175
3176 pmConf->filters[i].patternSize = ETH_HLEN + /* Ethernet header*/
3177 sizeof(struct arphdr) + /* ARP header */
3178 2 * ETH_ALEN + /* 2 Ethernet addresses*/
3179 2 * sizeof(u32); /*2 IPv4 addresses */
3180 pmConf->filters[i].maskSize =
3181 (pmConf->filters[i].patternSize - 1) / 8 + 1;
3182
3183 /* ETH_P_ARP in Ethernet header. */
3184 ehdr = (struct ethhdr *)pmConf->filters[i].pattern;
3185 ehdr->h_proto = htons(ETH_P_ARP);
3186
3187 /* ARPOP_REQUEST in ARP header. */
3188 ahdr = (struct arphdr *)&pmConf->filters[i].pattern[ETH_HLEN];
3189 ahdr->ar_op = htons(ARPOP_REQUEST);
3190 arpreq = (u8 *)(ahdr + 1);
3191
3192 /* The Unicast IPv4 address in 'tip' field. */
3193 arpreq += 2 * ETH_ALEN + sizeof(u32);
3194 *(u32 *)arpreq = ifa->ifa_address;
3195
3196 /* The mask for the relevant bits. */
3197 pmConf->filters[i].mask[0] = 0x00;
3198 pmConf->filters[i].mask[1] = 0x30; /* ETH_P_ARP */
3199 pmConf->filters[i].mask[2] = 0x30; /* ARPOP_REQUEST */
3200 pmConf->filters[i].mask[3] = 0x00;
3201 pmConf->filters[i].mask[4] = 0xC0; /* IPv4 TIP */
3202 pmConf->filters[i].mask[5] = 0x03; /* IPv4 TIP */
3203 in_dev_put(in_dev);
3204
3205 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_FILTER;
3206 i++;
3207 }
3208
3209 skip_arp:
3210 if (adapter->wol & WAKE_MAGIC)
3211 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_MAGIC;
3212
3213 pmConf->numFilters = i;
3214
3215 adapter->shared->devRead.pmConfDesc.confVer = cpu_to_le32(1);
3216 adapter->shared->devRead.pmConfDesc.confLen = cpu_to_le32(sizeof(
3217 *pmConf));
3218 adapter->shared->devRead.pmConfDesc.confPA = cpu_to_le64(virt_to_phys(
3219 pmConf));
3220
3221 spin_lock_irqsave(&adapter->cmd_lock, flags);
3222 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3223 VMXNET3_CMD_UPDATE_PMCFG);
3224 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3225
3226 pci_save_state(pdev);
3227 pci_enable_wake(pdev, pci_choose_state(pdev, PMSG_SUSPEND),
3228 adapter->wol);
3229 pci_disable_device(pdev);
3230 pci_set_power_state(pdev, pci_choose_state(pdev, PMSG_SUSPEND));
3231
3232 return 0;
3233 }
3234
3235
3236 static int
vmxnet3_resume(struct device * device)3237 vmxnet3_resume(struct device *device)
3238 {
3239 int err, i = 0;
3240 unsigned long flags;
3241 struct pci_dev *pdev = to_pci_dev(device);
3242 struct net_device *netdev = pci_get_drvdata(pdev);
3243 struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3244 struct Vmxnet3_PMConf *pmConf;
3245
3246 if (!netif_running(netdev))
3247 return 0;
3248
3249 /* Destroy wake-up filters. */
3250 pmConf = adapter->pm_conf;
3251 memset(pmConf, 0, sizeof(*pmConf));
3252
3253 adapter->shared->devRead.pmConfDesc.confVer = cpu_to_le32(1);
3254 adapter->shared->devRead.pmConfDesc.confLen = cpu_to_le32(sizeof(
3255 *pmConf));
3256 adapter->shared->devRead.pmConfDesc.confPA = cpu_to_le64(virt_to_phys(
3257 pmConf));
3258
3259 netif_device_attach(netdev);
3260 pci_set_power_state(pdev, PCI_D0);
3261 pci_restore_state(pdev);
3262 err = pci_enable_device_mem(pdev);
3263 if (err != 0)
3264 return err;
3265
3266 pci_enable_wake(pdev, PCI_D0, 0);
3267
3268 spin_lock_irqsave(&adapter->cmd_lock, flags);
3269 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3270 VMXNET3_CMD_UPDATE_PMCFG);
3271 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3272 vmxnet3_alloc_intr_resources(adapter);
3273 vmxnet3_request_irqs(adapter);
3274 for (i = 0; i < adapter->num_rx_queues; i++)
3275 napi_enable(&adapter->rx_queue[i].napi);
3276 vmxnet3_enable_all_intrs(adapter);
3277
3278 return 0;
3279 }
3280
3281 static const struct dev_pm_ops vmxnet3_pm_ops = {
3282 .suspend = vmxnet3_suspend,
3283 .resume = vmxnet3_resume,
3284 };
3285 #endif
3286
3287 static struct pci_driver vmxnet3_driver = {
3288 .name = vmxnet3_driver_name,
3289 .id_table = vmxnet3_pciid_table,
3290 .probe = vmxnet3_probe_device,
3291 .remove = __devexit_p(vmxnet3_remove_device),
3292 #ifdef CONFIG_PM
3293 .driver.pm = &vmxnet3_pm_ops,
3294 #endif
3295 };
3296
3297
3298 static int __init
vmxnet3_init_module(void)3299 vmxnet3_init_module(void)
3300 {
3301 printk(KERN_INFO "%s - version %s\n", VMXNET3_DRIVER_DESC,
3302 VMXNET3_DRIVER_VERSION_REPORT);
3303 return pci_register_driver(&vmxnet3_driver);
3304 }
3305
3306 module_init(vmxnet3_init_module);
3307
3308
3309 static void
vmxnet3_exit_module(void)3310 vmxnet3_exit_module(void)
3311 {
3312 pci_unregister_driver(&vmxnet3_driver);
3313 }
3314
3315 module_exit(vmxnet3_exit_module);
3316
3317 MODULE_AUTHOR("VMware, Inc.");
3318 MODULE_DESCRIPTION(VMXNET3_DRIVER_DESC);
3319 MODULE_LICENSE("GPL v2");
3320 MODULE_VERSION(VMXNET3_DRIVER_VERSION_STRING);
3321