1 /*
2  * Linux driver for VMware's vmxnet3 ethernet NIC.
3  *
4  * Copyright (C) 2008-2021, VMware, Inc. All Rights Reserved.
5  *
6  * This program is free software; you can redistribute it and/or modify it
7  * under the terms of the GNU General Public License as published by the
8  * Free Software Foundation; version 2 of the License and no later version.
9  *
10  * This program is distributed in the hope that it will be useful, but
11  * WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
13  * NON INFRINGEMENT. See the GNU General Public License for more
14  * details.
15  *
16  * You should have received a copy of the GNU General Public License
17  * along with this program; if not, write to the Free Software
18  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19  *
20  * The full GNU General Public License is included in this distribution in
21  * the file called "COPYING".
22  *
23  * Maintained by: pv-drivers@vmware.com
24  *
25  */
26 
27 #include <linux/module.h>
28 #include <net/ip6_checksum.h>
29 
30 #include "vmxnet3_int.h"
31 
32 char vmxnet3_driver_name[] = "vmxnet3";
33 #define VMXNET3_DRIVER_DESC "VMware vmxnet3 virtual NIC driver"
34 
35 /*
36  * PCI Device ID Table
37  * Last entry must be all 0s
38  */
39 static const struct pci_device_id vmxnet3_pciid_table[] = {
40 	{PCI_VDEVICE(VMWARE, PCI_DEVICE_ID_VMWARE_VMXNET3)},
41 	{0}
42 };
43 
44 MODULE_DEVICE_TABLE(pci, vmxnet3_pciid_table);
45 
46 static int enable_mq = 1;
47 
48 static void
49 vmxnet3_write_mac_addr(struct vmxnet3_adapter *adapter, const u8 *mac);
50 
51 /*
52  *    Enable/Disable the given intr
53  */
54 static void
55 vmxnet3_enable_intr(struct vmxnet3_adapter *adapter, unsigned intr_idx)
56 {
57 	VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_IMR + intr_idx * 8, 0);
58 }
59 
60 
61 static void
62 vmxnet3_disable_intr(struct vmxnet3_adapter *adapter, unsigned intr_idx)
63 {
64 	VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_IMR + intr_idx * 8, 1);
65 }
66 
67 
68 /*
69  *    Enable/Disable all intrs used by the device
70  */
71 static void
72 vmxnet3_enable_all_intrs(struct vmxnet3_adapter *adapter)
73 {
74 	int i;
75 
76 	for (i = 0; i < adapter->intr.num_intrs; i++)
77 		vmxnet3_enable_intr(adapter, i);
78 	adapter->shared->devRead.intrConf.intrCtrl &=
79 					cpu_to_le32(~VMXNET3_IC_DISABLE_ALL);
80 }
81 
82 
83 static void
84 vmxnet3_disable_all_intrs(struct vmxnet3_adapter *adapter)
85 {
86 	int i;
87 
88 	adapter->shared->devRead.intrConf.intrCtrl |=
89 					cpu_to_le32(VMXNET3_IC_DISABLE_ALL);
90 	for (i = 0; i < adapter->intr.num_intrs; i++)
91 		vmxnet3_disable_intr(adapter, i);
92 }
93 
94 
95 static void
96 vmxnet3_ack_events(struct vmxnet3_adapter *adapter, u32 events)
97 {
98 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_ECR, events);
99 }
100 
101 
102 static bool
103 vmxnet3_tq_stopped(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
104 {
105 	return tq->stopped;
106 }
107 
108 
109 static void
110 vmxnet3_tq_start(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
111 {
112 	tq->stopped = false;
113 	netif_start_subqueue(adapter->netdev, tq - adapter->tx_queue);
114 }
115 
116 
117 static void
118 vmxnet3_tq_wake(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
119 {
120 	tq->stopped = false;
121 	netif_wake_subqueue(adapter->netdev, (tq - adapter->tx_queue));
122 }
123 
124 
125 static void
126 vmxnet3_tq_stop(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
127 {
128 	tq->stopped = true;
129 	tq->num_stop++;
130 	netif_stop_subqueue(adapter->netdev, (tq - adapter->tx_queue));
131 }
132 
133 
134 /*
135  * Check the link state. This may start or stop the tx queue.
136  */
137 static void
138 vmxnet3_check_link(struct vmxnet3_adapter *adapter, bool affectTxQueue)
139 {
140 	u32 ret;
141 	int i;
142 	unsigned long flags;
143 
144 	spin_lock_irqsave(&adapter->cmd_lock, flags);
145 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, VMXNET3_CMD_GET_LINK);
146 	ret = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
147 	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
148 
149 	adapter->link_speed = ret >> 16;
150 	if (ret & 1) { /* Link is up. */
151 		netdev_info(adapter->netdev, "NIC Link is Up %d Mbps\n",
152 			    adapter->link_speed);
153 		netif_carrier_on(adapter->netdev);
154 
155 		if (affectTxQueue) {
156 			for (i = 0; i < adapter->num_tx_queues; i++)
157 				vmxnet3_tq_start(&adapter->tx_queue[i],
158 						 adapter);
159 		}
160 	} else {
161 		netdev_info(adapter->netdev, "NIC Link is Down\n");
162 		netif_carrier_off(adapter->netdev);
163 
164 		if (affectTxQueue) {
165 			for (i = 0; i < adapter->num_tx_queues; i++)
166 				vmxnet3_tq_stop(&adapter->tx_queue[i], adapter);
167 		}
168 	}
169 }
170 
171 static void
172 vmxnet3_process_events(struct vmxnet3_adapter *adapter)
173 {
174 	int i;
175 	unsigned long flags;
176 	u32 events = le32_to_cpu(adapter->shared->ecr);
177 	if (!events)
178 		return;
179 
180 	vmxnet3_ack_events(adapter, events);
181 
182 	/* Check if link state has changed */
183 	if (events & VMXNET3_ECR_LINK)
184 		vmxnet3_check_link(adapter, true);
185 
186 	/* Check if there is an error on xmit/recv queues */
187 	if (events & (VMXNET3_ECR_TQERR | VMXNET3_ECR_RQERR)) {
188 		spin_lock_irqsave(&adapter->cmd_lock, flags);
189 		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
190 				       VMXNET3_CMD_GET_QUEUE_STATUS);
191 		spin_unlock_irqrestore(&adapter->cmd_lock, flags);
192 
193 		for (i = 0; i < adapter->num_tx_queues; i++)
194 			if (adapter->tqd_start[i].status.stopped)
195 				dev_err(&adapter->netdev->dev,
196 					"%s: tq[%d] error 0x%x\n",
197 					adapter->netdev->name, i, le32_to_cpu(
198 					adapter->tqd_start[i].status.error));
199 		for (i = 0; i < adapter->num_rx_queues; i++)
200 			if (adapter->rqd_start[i].status.stopped)
201 				dev_err(&adapter->netdev->dev,
202 					"%s: rq[%d] error 0x%x\n",
203 					adapter->netdev->name, i,
204 					adapter->rqd_start[i].status.error);
205 
206 		schedule_work(&adapter->work);
207 	}
208 }
209 
210 #ifdef __BIG_ENDIAN_BITFIELD
211 /*
212  * The device expects the bitfields in shared structures to be written in
213  * little endian. When CPU is big endian, the following routines are used to
214  * correctly read and write into ABI.
215  * The general technique used here is : double word bitfields are defined in
216  * opposite order for big endian architecture. Then before reading them in
217  * driver the complete double word is translated using le32_to_cpu. Similarly
218  * After the driver writes into bitfields, cpu_to_le32 is used to translate the
219  * double words into required format.
220  * In order to avoid touching bits in shared structure more than once, temporary
221  * descriptors are used. These are passed as srcDesc to following functions.
222  */
223 static void vmxnet3_RxDescToCPU(const struct Vmxnet3_RxDesc *srcDesc,
224 				struct Vmxnet3_RxDesc *dstDesc)
225 {
226 	u32 *src = (u32 *)srcDesc + 2;
227 	u32 *dst = (u32 *)dstDesc + 2;
228 	dstDesc->addr = le64_to_cpu(srcDesc->addr);
229 	*dst = le32_to_cpu(*src);
230 	dstDesc->ext1 = le32_to_cpu(srcDesc->ext1);
231 }
232 
233 static void vmxnet3_TxDescToLe(const struct Vmxnet3_TxDesc *srcDesc,
234 			       struct Vmxnet3_TxDesc *dstDesc)
235 {
236 	int i;
237 	u32 *src = (u32 *)(srcDesc + 1);
238 	u32 *dst = (u32 *)(dstDesc + 1);
239 
240 	/* Working backwards so that the gen bit is set at the end. */
241 	for (i = 2; i > 0; i--) {
242 		src--;
243 		dst--;
244 		*dst = cpu_to_le32(*src);
245 	}
246 }
247 
248 
249 static void vmxnet3_RxCompToCPU(const struct Vmxnet3_RxCompDesc *srcDesc,
250 				struct Vmxnet3_RxCompDesc *dstDesc)
251 {
252 	int i = 0;
253 	u32 *src = (u32 *)srcDesc;
254 	u32 *dst = (u32 *)dstDesc;
255 	for (i = 0; i < sizeof(struct Vmxnet3_RxCompDesc) / sizeof(u32); i++) {
256 		*dst = le32_to_cpu(*src);
257 		src++;
258 		dst++;
259 	}
260 }
261 
262 
263 /* Used to read bitfield values from double words. */
264 static u32 get_bitfield32(const __le32 *bitfield, u32 pos, u32 size)
265 {
266 	u32 temp = le32_to_cpu(*bitfield);
267 	u32 mask = ((1 << size) - 1) << pos;
268 	temp &= mask;
269 	temp >>= pos;
270 	return temp;
271 }
272 
273 
274 
275 #endif  /* __BIG_ENDIAN_BITFIELD */
276 
277 #ifdef __BIG_ENDIAN_BITFIELD
278 
279 #   define VMXNET3_TXDESC_GET_GEN(txdesc) get_bitfield32(((const __le32 *) \
280 			txdesc) + VMXNET3_TXD_GEN_DWORD_SHIFT, \
281 			VMXNET3_TXD_GEN_SHIFT, VMXNET3_TXD_GEN_SIZE)
282 #   define VMXNET3_TXDESC_GET_EOP(txdesc) get_bitfield32(((const __le32 *) \
283 			txdesc) + VMXNET3_TXD_EOP_DWORD_SHIFT, \
284 			VMXNET3_TXD_EOP_SHIFT, VMXNET3_TXD_EOP_SIZE)
285 #   define VMXNET3_TCD_GET_GEN(tcd) get_bitfield32(((const __le32 *)tcd) + \
286 			VMXNET3_TCD_GEN_DWORD_SHIFT, VMXNET3_TCD_GEN_SHIFT, \
287 			VMXNET3_TCD_GEN_SIZE)
288 #   define VMXNET3_TCD_GET_TXIDX(tcd) get_bitfield32((const __le32 *)tcd, \
289 			VMXNET3_TCD_TXIDX_SHIFT, VMXNET3_TCD_TXIDX_SIZE)
290 #   define vmxnet3_getRxComp(dstrcd, rcd, tmp) do { \
291 			(dstrcd) = (tmp); \
292 			vmxnet3_RxCompToCPU((rcd), (tmp)); \
293 		} while (0)
294 #   define vmxnet3_getRxDesc(dstrxd, rxd, tmp) do { \
295 			(dstrxd) = (tmp); \
296 			vmxnet3_RxDescToCPU((rxd), (tmp)); \
297 		} while (0)
298 
299 #else
300 
301 #   define VMXNET3_TXDESC_GET_GEN(txdesc) ((txdesc)->gen)
302 #   define VMXNET3_TXDESC_GET_EOP(txdesc) ((txdesc)->eop)
303 #   define VMXNET3_TCD_GET_GEN(tcd) ((tcd)->gen)
304 #   define VMXNET3_TCD_GET_TXIDX(tcd) ((tcd)->txdIdx)
305 #   define vmxnet3_getRxComp(dstrcd, rcd, tmp) (dstrcd) = (rcd)
306 #   define vmxnet3_getRxDesc(dstrxd, rxd, tmp) (dstrxd) = (rxd)
307 
308 #endif /* __BIG_ENDIAN_BITFIELD  */
309 
310 
311 static void
312 vmxnet3_unmap_tx_buf(struct vmxnet3_tx_buf_info *tbi,
313 		     struct pci_dev *pdev)
314 {
315 	if (tbi->map_type == VMXNET3_MAP_SINGLE)
316 		dma_unmap_single(&pdev->dev, tbi->dma_addr, tbi->len,
317 				 DMA_TO_DEVICE);
318 	else if (tbi->map_type == VMXNET3_MAP_PAGE)
319 		dma_unmap_page(&pdev->dev, tbi->dma_addr, tbi->len,
320 			       DMA_TO_DEVICE);
321 	else
322 		BUG_ON(tbi->map_type != VMXNET3_MAP_NONE);
323 
324 	tbi->map_type = VMXNET3_MAP_NONE; /* to help debugging */
325 }
326 
327 
328 static int
329 vmxnet3_unmap_pkt(u32 eop_idx, struct vmxnet3_tx_queue *tq,
330 		  struct pci_dev *pdev,	struct vmxnet3_adapter *adapter)
331 {
332 	struct sk_buff *skb;
333 	int entries = 0;
334 
335 	/* no out of order completion */
336 	BUG_ON(tq->buf_info[eop_idx].sop_idx != tq->tx_ring.next2comp);
337 	BUG_ON(VMXNET3_TXDESC_GET_EOP(&(tq->tx_ring.base[eop_idx].txd)) != 1);
338 
339 	skb = tq->buf_info[eop_idx].skb;
340 	BUG_ON(skb == NULL);
341 	tq->buf_info[eop_idx].skb = NULL;
342 
343 	VMXNET3_INC_RING_IDX_ONLY(eop_idx, tq->tx_ring.size);
344 
345 	while (tq->tx_ring.next2comp != eop_idx) {
346 		vmxnet3_unmap_tx_buf(tq->buf_info + tq->tx_ring.next2comp,
347 				     pdev);
348 
349 		/* update next2comp w/o tx_lock. Since we are marking more,
350 		 * instead of less, tx ring entries avail, the worst case is
351 		 * that the tx routine incorrectly re-queues a pkt due to
352 		 * insufficient tx ring entries.
353 		 */
354 		vmxnet3_cmd_ring_adv_next2comp(&tq->tx_ring);
355 		entries++;
356 	}
357 
358 	dev_kfree_skb_any(skb);
359 	return entries;
360 }
361 
362 
363 static int
364 vmxnet3_tq_tx_complete(struct vmxnet3_tx_queue *tq,
365 			struct vmxnet3_adapter *adapter)
366 {
367 	int completed = 0;
368 	union Vmxnet3_GenericDesc *gdesc;
369 
370 	gdesc = tq->comp_ring.base + tq->comp_ring.next2proc;
371 	while (VMXNET3_TCD_GET_GEN(&gdesc->tcd) == tq->comp_ring.gen) {
372 		/* Prevent any &gdesc->tcd field from being (speculatively)
373 		 * read before (&gdesc->tcd)->gen is read.
374 		 */
375 		dma_rmb();
376 
377 		completed += vmxnet3_unmap_pkt(VMXNET3_TCD_GET_TXIDX(
378 					       &gdesc->tcd), tq, adapter->pdev,
379 					       adapter);
380 
381 		vmxnet3_comp_ring_adv_next2proc(&tq->comp_ring);
382 		gdesc = tq->comp_ring.base + tq->comp_ring.next2proc;
383 	}
384 
385 	if (completed) {
386 		spin_lock(&tq->tx_lock);
387 		if (unlikely(vmxnet3_tq_stopped(tq, adapter) &&
388 			     vmxnet3_cmd_ring_desc_avail(&tq->tx_ring) >
389 			     VMXNET3_WAKE_QUEUE_THRESHOLD(tq) &&
390 			     netif_carrier_ok(adapter->netdev))) {
391 			vmxnet3_tq_wake(tq, adapter);
392 		}
393 		spin_unlock(&tq->tx_lock);
394 	}
395 	return completed;
396 }
397 
398 
399 static void
400 vmxnet3_tq_cleanup(struct vmxnet3_tx_queue *tq,
401 		   struct vmxnet3_adapter *adapter)
402 {
403 	int i;
404 
405 	while (tq->tx_ring.next2comp != tq->tx_ring.next2fill) {
406 		struct vmxnet3_tx_buf_info *tbi;
407 
408 		tbi = tq->buf_info + tq->tx_ring.next2comp;
409 
410 		vmxnet3_unmap_tx_buf(tbi, adapter->pdev);
411 		if (tbi->skb) {
412 			dev_kfree_skb_any(tbi->skb);
413 			tbi->skb = NULL;
414 		}
415 		vmxnet3_cmd_ring_adv_next2comp(&tq->tx_ring);
416 	}
417 
418 	/* sanity check, verify all buffers are indeed unmapped and freed */
419 	for (i = 0; i < tq->tx_ring.size; i++) {
420 		BUG_ON(tq->buf_info[i].skb != NULL ||
421 		       tq->buf_info[i].map_type != VMXNET3_MAP_NONE);
422 	}
423 
424 	tq->tx_ring.gen = VMXNET3_INIT_GEN;
425 	tq->tx_ring.next2fill = tq->tx_ring.next2comp = 0;
426 
427 	tq->comp_ring.gen = VMXNET3_INIT_GEN;
428 	tq->comp_ring.next2proc = 0;
429 }
430 
431 
432 static void
433 vmxnet3_tq_destroy(struct vmxnet3_tx_queue *tq,
434 		   struct vmxnet3_adapter *adapter)
435 {
436 	if (tq->tx_ring.base) {
437 		dma_free_coherent(&adapter->pdev->dev, tq->tx_ring.size *
438 				  sizeof(struct Vmxnet3_TxDesc),
439 				  tq->tx_ring.base, tq->tx_ring.basePA);
440 		tq->tx_ring.base = NULL;
441 	}
442 	if (tq->data_ring.base) {
443 		dma_free_coherent(&adapter->pdev->dev,
444 				  tq->data_ring.size * tq->txdata_desc_size,
445 				  tq->data_ring.base, tq->data_ring.basePA);
446 		tq->data_ring.base = NULL;
447 	}
448 	if (tq->comp_ring.base) {
449 		dma_free_coherent(&adapter->pdev->dev, tq->comp_ring.size *
450 				  sizeof(struct Vmxnet3_TxCompDesc),
451 				  tq->comp_ring.base, tq->comp_ring.basePA);
452 		tq->comp_ring.base = NULL;
453 	}
454 	kfree(tq->buf_info);
455 	tq->buf_info = NULL;
456 }
457 
458 
459 /* Destroy all tx queues */
460 void
461 vmxnet3_tq_destroy_all(struct vmxnet3_adapter *adapter)
462 {
463 	int i;
464 
465 	for (i = 0; i < adapter->num_tx_queues; i++)
466 		vmxnet3_tq_destroy(&adapter->tx_queue[i], adapter);
467 }
468 
469 
470 static void
471 vmxnet3_tq_init(struct vmxnet3_tx_queue *tq,
472 		struct vmxnet3_adapter *adapter)
473 {
474 	int i;
475 
476 	/* reset the tx ring contents to 0 and reset the tx ring states */
477 	memset(tq->tx_ring.base, 0, tq->tx_ring.size *
478 	       sizeof(struct Vmxnet3_TxDesc));
479 	tq->tx_ring.next2fill = tq->tx_ring.next2comp = 0;
480 	tq->tx_ring.gen = VMXNET3_INIT_GEN;
481 
482 	memset(tq->data_ring.base, 0,
483 	       tq->data_ring.size * tq->txdata_desc_size);
484 
485 	/* reset the tx comp ring contents to 0 and reset comp ring states */
486 	memset(tq->comp_ring.base, 0, tq->comp_ring.size *
487 	       sizeof(struct Vmxnet3_TxCompDesc));
488 	tq->comp_ring.next2proc = 0;
489 	tq->comp_ring.gen = VMXNET3_INIT_GEN;
490 
491 	/* reset the bookkeeping data */
492 	memset(tq->buf_info, 0, sizeof(tq->buf_info[0]) * tq->tx_ring.size);
493 	for (i = 0; i < tq->tx_ring.size; i++)
494 		tq->buf_info[i].map_type = VMXNET3_MAP_NONE;
495 
496 	/* stats are not reset */
497 }
498 
499 
500 static int
501 vmxnet3_tq_create(struct vmxnet3_tx_queue *tq,
502 		  struct vmxnet3_adapter *adapter)
503 {
504 	BUG_ON(tq->tx_ring.base || tq->data_ring.base ||
505 	       tq->comp_ring.base || tq->buf_info);
506 
507 	tq->tx_ring.base = dma_alloc_coherent(&adapter->pdev->dev,
508 			tq->tx_ring.size * sizeof(struct Vmxnet3_TxDesc),
509 			&tq->tx_ring.basePA, GFP_KERNEL);
510 	if (!tq->tx_ring.base) {
511 		netdev_err(adapter->netdev, "failed to allocate tx ring\n");
512 		goto err;
513 	}
514 
515 	tq->data_ring.base = dma_alloc_coherent(&adapter->pdev->dev,
516 			tq->data_ring.size * tq->txdata_desc_size,
517 			&tq->data_ring.basePA, GFP_KERNEL);
518 	if (!tq->data_ring.base) {
519 		netdev_err(adapter->netdev, "failed to allocate tx data ring\n");
520 		goto err;
521 	}
522 
523 	tq->comp_ring.base = dma_alloc_coherent(&adapter->pdev->dev,
524 			tq->comp_ring.size * sizeof(struct Vmxnet3_TxCompDesc),
525 			&tq->comp_ring.basePA, GFP_KERNEL);
526 	if (!tq->comp_ring.base) {
527 		netdev_err(adapter->netdev, "failed to allocate tx comp ring\n");
528 		goto err;
529 	}
530 
531 	tq->buf_info = kcalloc_node(tq->tx_ring.size, sizeof(tq->buf_info[0]),
532 				    GFP_KERNEL,
533 				    dev_to_node(&adapter->pdev->dev));
534 	if (!tq->buf_info)
535 		goto err;
536 
537 	return 0;
538 
539 err:
540 	vmxnet3_tq_destroy(tq, adapter);
541 	return -ENOMEM;
542 }
543 
544 static void
545 vmxnet3_tq_cleanup_all(struct vmxnet3_adapter *adapter)
546 {
547 	int i;
548 
549 	for (i = 0; i < adapter->num_tx_queues; i++)
550 		vmxnet3_tq_cleanup(&adapter->tx_queue[i], adapter);
551 }
552 
553 /*
554  *    starting from ring->next2fill, allocate rx buffers for the given ring
555  *    of the rx queue and update the rx desc. stop after @num_to_alloc buffers
556  *    are allocated or allocation fails
557  */
558 
559 static int
560 vmxnet3_rq_alloc_rx_buf(struct vmxnet3_rx_queue *rq, u32 ring_idx,
561 			int num_to_alloc, struct vmxnet3_adapter *adapter)
562 {
563 	int num_allocated = 0;
564 	struct vmxnet3_rx_buf_info *rbi_base = rq->buf_info[ring_idx];
565 	struct vmxnet3_cmd_ring *ring = &rq->rx_ring[ring_idx];
566 	u32 val;
567 
568 	while (num_allocated <= num_to_alloc) {
569 		struct vmxnet3_rx_buf_info *rbi;
570 		union Vmxnet3_GenericDesc *gd;
571 
572 		rbi = rbi_base + ring->next2fill;
573 		gd = ring->base + ring->next2fill;
574 
575 		if (rbi->buf_type == VMXNET3_RX_BUF_SKB) {
576 			if (rbi->skb == NULL) {
577 				rbi->skb = __netdev_alloc_skb_ip_align(adapter->netdev,
578 								       rbi->len,
579 								       GFP_KERNEL);
580 				if (unlikely(rbi->skb == NULL)) {
581 					rq->stats.rx_buf_alloc_failure++;
582 					break;
583 				}
584 
585 				rbi->dma_addr = dma_map_single(
586 						&adapter->pdev->dev,
587 						rbi->skb->data, rbi->len,
588 						DMA_FROM_DEVICE);
589 				if (dma_mapping_error(&adapter->pdev->dev,
590 						      rbi->dma_addr)) {
591 					dev_kfree_skb_any(rbi->skb);
592 					rq->stats.rx_buf_alloc_failure++;
593 					break;
594 				}
595 			} else {
596 				/* rx buffer skipped by the device */
597 			}
598 			val = VMXNET3_RXD_BTYPE_HEAD << VMXNET3_RXD_BTYPE_SHIFT;
599 		} else {
600 			BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_PAGE ||
601 			       rbi->len  != PAGE_SIZE);
602 
603 			if (rbi->page == NULL) {
604 				rbi->page = alloc_page(GFP_ATOMIC);
605 				if (unlikely(rbi->page == NULL)) {
606 					rq->stats.rx_buf_alloc_failure++;
607 					break;
608 				}
609 				rbi->dma_addr = dma_map_page(
610 						&adapter->pdev->dev,
611 						rbi->page, 0, PAGE_SIZE,
612 						DMA_FROM_DEVICE);
613 				if (dma_mapping_error(&adapter->pdev->dev,
614 						      rbi->dma_addr)) {
615 					put_page(rbi->page);
616 					rq->stats.rx_buf_alloc_failure++;
617 					break;
618 				}
619 			} else {
620 				/* rx buffers skipped by the device */
621 			}
622 			val = VMXNET3_RXD_BTYPE_BODY << VMXNET3_RXD_BTYPE_SHIFT;
623 		}
624 
625 		gd->rxd.addr = cpu_to_le64(rbi->dma_addr);
626 		gd->dword[2] = cpu_to_le32((!ring->gen << VMXNET3_RXD_GEN_SHIFT)
627 					   | val | rbi->len);
628 
629 		/* Fill the last buffer but dont mark it ready, or else the
630 		 * device will think that the queue is full */
631 		if (num_allocated == num_to_alloc)
632 			break;
633 
634 		gd->dword[2] |= cpu_to_le32(ring->gen << VMXNET3_RXD_GEN_SHIFT);
635 		num_allocated++;
636 		vmxnet3_cmd_ring_adv_next2fill(ring);
637 	}
638 
639 	netdev_dbg(adapter->netdev,
640 		"alloc_rx_buf: %d allocated, next2fill %u, next2comp %u\n",
641 		num_allocated, ring->next2fill, ring->next2comp);
642 
643 	/* so that the device can distinguish a full ring and an empty ring */
644 	BUG_ON(num_allocated != 0 && ring->next2fill == ring->next2comp);
645 
646 	return num_allocated;
647 }
648 
649 
650 static void
651 vmxnet3_append_frag(struct sk_buff *skb, struct Vmxnet3_RxCompDesc *rcd,
652 		    struct vmxnet3_rx_buf_info *rbi)
653 {
654 	skb_frag_t *frag = skb_shinfo(skb)->frags + 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 	skb_frag_off_set(frag, 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 int
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 					tq->txdata_desc_size);
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 		netdev_dbg(adapter->netdev,
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 = dma_map_single(&adapter->pdev->dev,
725 				skb->data + buf_offset, buf_size,
726 				DMA_TO_DEVICE);
727 		if (dma_mapping_error(&adapter->pdev->dev, tbi->dma_addr))
728 			return -EFAULT;
729 
730 		tbi->len = buf_size;
731 
732 		gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
733 		BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
734 
735 		gdesc->txd.addr = cpu_to_le64(tbi->dma_addr);
736 		gdesc->dword[2] = cpu_to_le32(dw2);
737 		gdesc->dword[3] = 0;
738 
739 		netdev_dbg(adapter->netdev,
740 			"txd[%u]: 0x%Lx 0x%x 0x%x\n",
741 			tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr),
742 			le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]);
743 		vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
744 		dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
745 
746 		len -= buf_size;
747 		buf_offset += buf_size;
748 	}
749 
750 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
751 		const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
752 		u32 buf_size;
753 
754 		buf_offset = 0;
755 		len = skb_frag_size(frag);
756 		while (len) {
757 			tbi = tq->buf_info + tq->tx_ring.next2fill;
758 			if (len < VMXNET3_MAX_TX_BUF_SIZE) {
759 				buf_size = len;
760 				dw2 |= len;
761 			} else {
762 				buf_size = VMXNET3_MAX_TX_BUF_SIZE;
763 				/* spec says that for TxDesc.len, 0 == 2^14 */
764 			}
765 			tbi->map_type = VMXNET3_MAP_PAGE;
766 			tbi->dma_addr = skb_frag_dma_map(&adapter->pdev->dev, frag,
767 							 buf_offset, buf_size,
768 							 DMA_TO_DEVICE);
769 			if (dma_mapping_error(&adapter->pdev->dev, tbi->dma_addr))
770 				return -EFAULT;
771 
772 			tbi->len = buf_size;
773 
774 			gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
775 			BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
776 
777 			gdesc->txd.addr = cpu_to_le64(tbi->dma_addr);
778 			gdesc->dword[2] = cpu_to_le32(dw2);
779 			gdesc->dword[3] = 0;
780 
781 			netdev_dbg(adapter->netdev,
782 				"txd[%u]: 0x%llx %u %u\n",
783 				tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr),
784 				le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]);
785 			vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
786 			dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
787 
788 			len -= buf_size;
789 			buf_offset += buf_size;
790 		}
791 	}
792 
793 	ctx->eop_txd = gdesc;
794 
795 	/* set the last buf_info for the pkt */
796 	tbi->skb = skb;
797 	tbi->sop_idx = ctx->sop_txd - tq->tx_ring.base;
798 
799 	return 0;
800 }
801 
802 
803 /* Init all tx queues */
804 static void
805 vmxnet3_tq_init_all(struct vmxnet3_adapter *adapter)
806 {
807 	int i;
808 
809 	for (i = 0; i < adapter->num_tx_queues; i++)
810 		vmxnet3_tq_init(&adapter->tx_queue[i], adapter);
811 }
812 
813 
814 /*
815  *    parse relevant protocol headers:
816  *      For a tso pkt, relevant headers are L2/3/4 including options
817  *      For a pkt requesting csum offloading, they are L2/3 and may include L4
818  *      if it's a TCP/UDP pkt
819  *
820  * Returns:
821  *    -1:  error happens during parsing
822  *     0:  protocol headers parsed, but too big to be copied
823  *     1:  protocol headers parsed and copied
824  *
825  * Other effects:
826  *    1. related *ctx fields are updated.
827  *    2. ctx->copy_size is # of bytes copied
828  *    3. the portion to be copied is guaranteed to be in the linear part
829  *
830  */
831 static int
832 vmxnet3_parse_hdr(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
833 		  struct vmxnet3_tx_ctx *ctx,
834 		  struct vmxnet3_adapter *adapter)
835 {
836 	u8 protocol = 0;
837 
838 	if (ctx->mss) {	/* TSO */
839 		if (VMXNET3_VERSION_GE_4(adapter) && skb->encapsulation) {
840 			ctx->l4_offset = skb_inner_transport_offset(skb);
841 			ctx->l4_hdr_size = inner_tcp_hdrlen(skb);
842 			ctx->copy_size = ctx->l4_offset + ctx->l4_hdr_size;
843 		} else {
844 			ctx->l4_offset = skb_transport_offset(skb);
845 			ctx->l4_hdr_size = tcp_hdrlen(skb);
846 			ctx->copy_size = ctx->l4_offset + ctx->l4_hdr_size;
847 		}
848 	} else {
849 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
850 			/* For encap packets, skb_checksum_start_offset refers
851 			 * to inner L4 offset. Thus, below works for encap as
852 			 * well as non-encap case
853 			 */
854 			ctx->l4_offset = skb_checksum_start_offset(skb);
855 
856 			if (VMXNET3_VERSION_GE_4(adapter) &&
857 			    skb->encapsulation) {
858 				struct iphdr *iph = inner_ip_hdr(skb);
859 
860 				if (iph->version == 4) {
861 					protocol = iph->protocol;
862 				} else {
863 					const struct ipv6hdr *ipv6h;
864 
865 					ipv6h = inner_ipv6_hdr(skb);
866 					protocol = ipv6h->nexthdr;
867 				}
868 			} else {
869 				if (ctx->ipv4) {
870 					const struct iphdr *iph = ip_hdr(skb);
871 
872 					protocol = iph->protocol;
873 				} else if (ctx->ipv6) {
874 					const struct ipv6hdr *ipv6h;
875 
876 					ipv6h = ipv6_hdr(skb);
877 					protocol = ipv6h->nexthdr;
878 				}
879 			}
880 
881 			switch (protocol) {
882 			case IPPROTO_TCP:
883 				ctx->l4_hdr_size = skb->encapsulation ? inner_tcp_hdrlen(skb) :
884 						   tcp_hdrlen(skb);
885 				break;
886 			case IPPROTO_UDP:
887 				ctx->l4_hdr_size = sizeof(struct udphdr);
888 				break;
889 			default:
890 				ctx->l4_hdr_size = 0;
891 				break;
892 			}
893 
894 			ctx->copy_size = min(ctx->l4_offset +
895 					 ctx->l4_hdr_size, skb->len);
896 		} else {
897 			ctx->l4_offset = 0;
898 			ctx->l4_hdr_size = 0;
899 			/* copy as much as allowed */
900 			ctx->copy_size = min_t(unsigned int,
901 					       tq->txdata_desc_size,
902 					       skb_headlen(skb));
903 		}
904 
905 		if (skb->len <= VMXNET3_HDR_COPY_SIZE)
906 			ctx->copy_size = skb->len;
907 
908 		/* make sure headers are accessible directly */
909 		if (unlikely(!pskb_may_pull(skb, ctx->copy_size)))
910 			goto err;
911 	}
912 
913 	if (unlikely(ctx->copy_size > tq->txdata_desc_size)) {
914 		tq->stats.oversized_hdr++;
915 		ctx->copy_size = 0;
916 		return 0;
917 	}
918 
919 	return 1;
920 err:
921 	return -1;
922 }
923 
924 /*
925  *    copy relevant protocol headers to the transmit ring:
926  *      For a tso pkt, relevant headers are L2/3/4 including options
927  *      For a pkt requesting csum offloading, they are L2/3 and may include L4
928  *      if it's a TCP/UDP pkt
929  *
930  *
931  *    Note that this requires that vmxnet3_parse_hdr be called first to set the
932  *      appropriate bits in ctx first
933  */
934 static void
935 vmxnet3_copy_hdr(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
936 		 struct vmxnet3_tx_ctx *ctx,
937 		 struct vmxnet3_adapter *adapter)
938 {
939 	struct Vmxnet3_TxDataDesc *tdd;
940 
941 	tdd = (struct Vmxnet3_TxDataDesc *)((u8 *)tq->data_ring.base +
942 					    tq->tx_ring.next2fill *
943 					    tq->txdata_desc_size);
944 
945 	memcpy(tdd->data, skb->data, ctx->copy_size);
946 	netdev_dbg(adapter->netdev,
947 		"copy %u bytes to dataRing[%u]\n",
948 		ctx->copy_size, tq->tx_ring.next2fill);
949 }
950 
951 
952 static void
953 vmxnet3_prepare_inner_tso(struct sk_buff *skb,
954 			  struct vmxnet3_tx_ctx *ctx)
955 {
956 	struct tcphdr *tcph = inner_tcp_hdr(skb);
957 	struct iphdr *iph = inner_ip_hdr(skb);
958 
959 	if (iph->version == 4) {
960 		iph->check = 0;
961 		tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, 0,
962 						 IPPROTO_TCP, 0);
963 	} else {
964 		struct ipv6hdr *iph = inner_ipv6_hdr(skb);
965 
966 		tcph->check = ~csum_ipv6_magic(&iph->saddr, &iph->daddr, 0,
967 					       IPPROTO_TCP, 0);
968 	}
969 }
970 
971 static void
972 vmxnet3_prepare_tso(struct sk_buff *skb,
973 		    struct vmxnet3_tx_ctx *ctx)
974 {
975 	struct tcphdr *tcph = tcp_hdr(skb);
976 
977 	if (ctx->ipv4) {
978 		struct iphdr *iph = ip_hdr(skb);
979 
980 		iph->check = 0;
981 		tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, 0,
982 						 IPPROTO_TCP, 0);
983 	} else if (ctx->ipv6) {
984 		tcp_v6_gso_csum_prep(skb);
985 	}
986 }
987 
988 static int txd_estimate(const struct sk_buff *skb)
989 {
990 	int count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) + 1;
991 	int i;
992 
993 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
994 		const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
995 
996 		count += VMXNET3_TXD_NEEDED(skb_frag_size(frag));
997 	}
998 	return count;
999 }
1000 
1001 /*
1002  * Transmits a pkt thru a given tq
1003  * Returns:
1004  *    NETDEV_TX_OK:      descriptors are setup successfully
1005  *    NETDEV_TX_OK:      error occurred, the pkt is dropped
1006  *    NETDEV_TX_BUSY:    tx ring is full, queue is stopped
1007  *
1008  * Side-effects:
1009  *    1. tx ring may be changed
1010  *    2. tq stats may be updated accordingly
1011  *    3. shared->txNumDeferred may be updated
1012  */
1013 
1014 static int
1015 vmxnet3_tq_xmit(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
1016 		struct vmxnet3_adapter *adapter, struct net_device *netdev)
1017 {
1018 	int ret;
1019 	u32 count;
1020 	int num_pkts;
1021 	int tx_num_deferred;
1022 	unsigned long flags;
1023 	struct vmxnet3_tx_ctx ctx;
1024 	union Vmxnet3_GenericDesc *gdesc;
1025 #ifdef __BIG_ENDIAN_BITFIELD
1026 	/* Use temporary descriptor to avoid touching bits multiple times */
1027 	union Vmxnet3_GenericDesc tempTxDesc;
1028 #endif
1029 
1030 	count = txd_estimate(skb);
1031 
1032 	ctx.ipv4 = (vlan_get_protocol(skb) == cpu_to_be16(ETH_P_IP));
1033 	ctx.ipv6 = (vlan_get_protocol(skb) == cpu_to_be16(ETH_P_IPV6));
1034 
1035 	ctx.mss = skb_shinfo(skb)->gso_size;
1036 	if (ctx.mss) {
1037 		if (skb_header_cloned(skb)) {
1038 			if (unlikely(pskb_expand_head(skb, 0, 0,
1039 						      GFP_ATOMIC) != 0)) {
1040 				tq->stats.drop_tso++;
1041 				goto drop_pkt;
1042 			}
1043 			tq->stats.copy_skb_header++;
1044 		}
1045 		if (skb->encapsulation) {
1046 			vmxnet3_prepare_inner_tso(skb, &ctx);
1047 		} else {
1048 			vmxnet3_prepare_tso(skb, &ctx);
1049 		}
1050 	} else {
1051 		if (unlikely(count > VMXNET3_MAX_TXD_PER_PKT)) {
1052 
1053 			/* non-tso pkts must not use more than
1054 			 * VMXNET3_MAX_TXD_PER_PKT entries
1055 			 */
1056 			if (skb_linearize(skb) != 0) {
1057 				tq->stats.drop_too_many_frags++;
1058 				goto drop_pkt;
1059 			}
1060 			tq->stats.linearized++;
1061 
1062 			/* recalculate the # of descriptors to use */
1063 			count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) + 1;
1064 		}
1065 	}
1066 
1067 	ret = vmxnet3_parse_hdr(skb, tq, &ctx, adapter);
1068 	if (ret >= 0) {
1069 		BUG_ON(ret <= 0 && ctx.copy_size != 0);
1070 		/* hdrs parsed, check against other limits */
1071 		if (ctx.mss) {
1072 			if (unlikely(ctx.l4_offset + ctx.l4_hdr_size >
1073 				     VMXNET3_MAX_TX_BUF_SIZE)) {
1074 				tq->stats.drop_oversized_hdr++;
1075 				goto drop_pkt;
1076 			}
1077 		} else {
1078 			if (skb->ip_summed == CHECKSUM_PARTIAL) {
1079 				if (unlikely(ctx.l4_offset +
1080 					     skb->csum_offset >
1081 					     VMXNET3_MAX_CSUM_OFFSET)) {
1082 					tq->stats.drop_oversized_hdr++;
1083 					goto drop_pkt;
1084 				}
1085 			}
1086 		}
1087 	} else {
1088 		tq->stats.drop_hdr_inspect_err++;
1089 		goto drop_pkt;
1090 	}
1091 
1092 	spin_lock_irqsave(&tq->tx_lock, flags);
1093 
1094 	if (count > vmxnet3_cmd_ring_desc_avail(&tq->tx_ring)) {
1095 		tq->stats.tx_ring_full++;
1096 		netdev_dbg(adapter->netdev,
1097 			"tx queue stopped on %s, next2comp %u"
1098 			" next2fill %u\n", adapter->netdev->name,
1099 			tq->tx_ring.next2comp, tq->tx_ring.next2fill);
1100 
1101 		vmxnet3_tq_stop(tq, adapter);
1102 		spin_unlock_irqrestore(&tq->tx_lock, flags);
1103 		return NETDEV_TX_BUSY;
1104 	}
1105 
1106 
1107 	vmxnet3_copy_hdr(skb, tq, &ctx, adapter);
1108 
1109 	/* fill tx descs related to addr & len */
1110 	if (vmxnet3_map_pkt(skb, &ctx, tq, adapter->pdev, adapter))
1111 		goto unlock_drop_pkt;
1112 
1113 	/* setup the EOP desc */
1114 	ctx.eop_txd->dword[3] = cpu_to_le32(VMXNET3_TXD_CQ | VMXNET3_TXD_EOP);
1115 
1116 	/* setup the SOP desc */
1117 #ifdef __BIG_ENDIAN_BITFIELD
1118 	gdesc = &tempTxDesc;
1119 	gdesc->dword[2] = ctx.sop_txd->dword[2];
1120 	gdesc->dword[3] = ctx.sop_txd->dword[3];
1121 #else
1122 	gdesc = ctx.sop_txd;
1123 #endif
1124 	tx_num_deferred = le32_to_cpu(tq->shared->txNumDeferred);
1125 	if (ctx.mss) {
1126 		if (VMXNET3_VERSION_GE_4(adapter) && skb->encapsulation) {
1127 			gdesc->txd.hlen = ctx.l4_offset + ctx.l4_hdr_size;
1128 			gdesc->txd.om = VMXNET3_OM_ENCAP;
1129 			gdesc->txd.msscof = ctx.mss;
1130 
1131 			if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL_CSUM)
1132 				gdesc->txd.oco = 1;
1133 		} else {
1134 			gdesc->txd.hlen = ctx.l4_offset + ctx.l4_hdr_size;
1135 			gdesc->txd.om = VMXNET3_OM_TSO;
1136 			gdesc->txd.msscof = ctx.mss;
1137 		}
1138 		num_pkts = (skb->len - gdesc->txd.hlen + ctx.mss - 1) / ctx.mss;
1139 	} else {
1140 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
1141 			if (VMXNET3_VERSION_GE_4(adapter) &&
1142 			    skb->encapsulation) {
1143 				gdesc->txd.hlen = ctx.l4_offset +
1144 						  ctx.l4_hdr_size;
1145 				gdesc->txd.om = VMXNET3_OM_ENCAP;
1146 				gdesc->txd.msscof = 0;		/* Reserved */
1147 			} else {
1148 				gdesc->txd.hlen = ctx.l4_offset;
1149 				gdesc->txd.om = VMXNET3_OM_CSUM;
1150 				gdesc->txd.msscof = ctx.l4_offset +
1151 						    skb->csum_offset;
1152 			}
1153 		} else {
1154 			gdesc->txd.om = 0;
1155 			gdesc->txd.msscof = 0;
1156 		}
1157 		num_pkts = 1;
1158 	}
1159 	le32_add_cpu(&tq->shared->txNumDeferred, num_pkts);
1160 	tx_num_deferred += num_pkts;
1161 
1162 	if (skb_vlan_tag_present(skb)) {
1163 		gdesc->txd.ti = 1;
1164 		gdesc->txd.tci = skb_vlan_tag_get(skb);
1165 	}
1166 
1167 	/* Ensure that the write to (&gdesc->txd)->gen will be observed after
1168 	 * all other writes to &gdesc->txd.
1169 	 */
1170 	dma_wmb();
1171 
1172 	/* finally flips the GEN bit of the SOP desc. */
1173 	gdesc->dword[2] = cpu_to_le32(le32_to_cpu(gdesc->dword[2]) ^
1174 						  VMXNET3_TXD_GEN);
1175 #ifdef __BIG_ENDIAN_BITFIELD
1176 	/* Finished updating in bitfields of Tx Desc, so write them in original
1177 	 * place.
1178 	 */
1179 	vmxnet3_TxDescToLe((struct Vmxnet3_TxDesc *)gdesc,
1180 			   (struct Vmxnet3_TxDesc *)ctx.sop_txd);
1181 	gdesc = ctx.sop_txd;
1182 #endif
1183 	netdev_dbg(adapter->netdev,
1184 		"txd[%u]: SOP 0x%Lx 0x%x 0x%x\n",
1185 		(u32)(ctx.sop_txd -
1186 		tq->tx_ring.base), le64_to_cpu(gdesc->txd.addr),
1187 		le32_to_cpu(gdesc->dword[2]), le32_to_cpu(gdesc->dword[3]));
1188 
1189 	spin_unlock_irqrestore(&tq->tx_lock, flags);
1190 
1191 	if (tx_num_deferred >= le32_to_cpu(tq->shared->txThreshold)) {
1192 		tq->shared->txNumDeferred = 0;
1193 		VMXNET3_WRITE_BAR0_REG(adapter,
1194 				       VMXNET3_REG_TXPROD + tq->qid * 8,
1195 				       tq->tx_ring.next2fill);
1196 	}
1197 
1198 	return NETDEV_TX_OK;
1199 
1200 unlock_drop_pkt:
1201 	spin_unlock_irqrestore(&tq->tx_lock, flags);
1202 drop_pkt:
1203 	tq->stats.drop_total++;
1204 	dev_kfree_skb_any(skb);
1205 	return NETDEV_TX_OK;
1206 }
1207 
1208 
1209 static netdev_tx_t
1210 vmxnet3_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
1211 {
1212 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1213 
1214 	BUG_ON(skb->queue_mapping > adapter->num_tx_queues);
1215 	return vmxnet3_tq_xmit(skb,
1216 			       &adapter->tx_queue[skb->queue_mapping],
1217 			       adapter, netdev);
1218 }
1219 
1220 
1221 static void
1222 vmxnet3_rx_csum(struct vmxnet3_adapter *adapter,
1223 		struct sk_buff *skb,
1224 		union Vmxnet3_GenericDesc *gdesc)
1225 {
1226 	if (!gdesc->rcd.cnc && adapter->netdev->features & NETIF_F_RXCSUM) {
1227 		if (gdesc->rcd.v4 &&
1228 		    (le32_to_cpu(gdesc->dword[3]) &
1229 		     VMXNET3_RCD_CSUM_OK) == VMXNET3_RCD_CSUM_OK) {
1230 			skb->ip_summed = CHECKSUM_UNNECESSARY;
1231 			WARN_ON_ONCE(!(gdesc->rcd.tcp || gdesc->rcd.udp) &&
1232 				     !(le32_to_cpu(gdesc->dword[0]) &
1233 				     (1UL << VMXNET3_RCD_HDR_INNER_SHIFT)));
1234 			WARN_ON_ONCE(gdesc->rcd.frg &&
1235 				     !(le32_to_cpu(gdesc->dword[0]) &
1236 				     (1UL << VMXNET3_RCD_HDR_INNER_SHIFT)));
1237 		} else if (gdesc->rcd.v6 && (le32_to_cpu(gdesc->dword[3]) &
1238 					     (1 << VMXNET3_RCD_TUC_SHIFT))) {
1239 			skb->ip_summed = CHECKSUM_UNNECESSARY;
1240 			WARN_ON_ONCE(!(gdesc->rcd.tcp || gdesc->rcd.udp) &&
1241 				     !(le32_to_cpu(gdesc->dword[0]) &
1242 				     (1UL << VMXNET3_RCD_HDR_INNER_SHIFT)));
1243 			WARN_ON_ONCE(gdesc->rcd.frg &&
1244 				     !(le32_to_cpu(gdesc->dword[0]) &
1245 				     (1UL << VMXNET3_RCD_HDR_INNER_SHIFT)));
1246 		} else {
1247 			if (gdesc->rcd.csum) {
1248 				skb->csum = htons(gdesc->rcd.csum);
1249 				skb->ip_summed = CHECKSUM_PARTIAL;
1250 			} else {
1251 				skb_checksum_none_assert(skb);
1252 			}
1253 		}
1254 	} else {
1255 		skb_checksum_none_assert(skb);
1256 	}
1257 }
1258 
1259 
1260 static void
1261 vmxnet3_rx_error(struct vmxnet3_rx_queue *rq, struct Vmxnet3_RxCompDesc *rcd,
1262 		 struct vmxnet3_rx_ctx *ctx,  struct vmxnet3_adapter *adapter)
1263 {
1264 	rq->stats.drop_err++;
1265 	if (!rcd->fcs)
1266 		rq->stats.drop_fcs++;
1267 
1268 	rq->stats.drop_total++;
1269 
1270 	/*
1271 	 * We do not unmap and chain the rx buffer to the skb.
1272 	 * We basically pretend this buffer is not used and will be recycled
1273 	 * by vmxnet3_rq_alloc_rx_buf()
1274 	 */
1275 
1276 	/*
1277 	 * ctx->skb may be NULL if this is the first and the only one
1278 	 * desc for the pkt
1279 	 */
1280 	if (ctx->skb)
1281 		dev_kfree_skb_irq(ctx->skb);
1282 
1283 	ctx->skb = NULL;
1284 }
1285 
1286 
1287 static u32
1288 vmxnet3_get_hdr_len(struct vmxnet3_adapter *adapter, struct sk_buff *skb,
1289 		    union Vmxnet3_GenericDesc *gdesc)
1290 {
1291 	u32 hlen, maplen;
1292 	union {
1293 		void *ptr;
1294 		struct ethhdr *eth;
1295 		struct vlan_ethhdr *veth;
1296 		struct iphdr *ipv4;
1297 		struct ipv6hdr *ipv6;
1298 		struct tcphdr *tcp;
1299 	} hdr;
1300 	BUG_ON(gdesc->rcd.tcp == 0);
1301 
1302 	maplen = skb_headlen(skb);
1303 	if (unlikely(sizeof(struct iphdr) + sizeof(struct tcphdr) > maplen))
1304 		return 0;
1305 
1306 	if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
1307 	    skb->protocol == cpu_to_be16(ETH_P_8021AD))
1308 		hlen = sizeof(struct vlan_ethhdr);
1309 	else
1310 		hlen = sizeof(struct ethhdr);
1311 
1312 	hdr.eth = eth_hdr(skb);
1313 	if (gdesc->rcd.v4) {
1314 		BUG_ON(hdr.eth->h_proto != htons(ETH_P_IP) &&
1315 		       hdr.veth->h_vlan_encapsulated_proto != htons(ETH_P_IP));
1316 		hdr.ptr += hlen;
1317 		BUG_ON(hdr.ipv4->protocol != IPPROTO_TCP);
1318 		hlen = hdr.ipv4->ihl << 2;
1319 		hdr.ptr += hdr.ipv4->ihl << 2;
1320 	} else if (gdesc->rcd.v6) {
1321 		BUG_ON(hdr.eth->h_proto != htons(ETH_P_IPV6) &&
1322 		       hdr.veth->h_vlan_encapsulated_proto != htons(ETH_P_IPV6));
1323 		hdr.ptr += hlen;
1324 		/* Use an estimated value, since we also need to handle
1325 		 * TSO case.
1326 		 */
1327 		if (hdr.ipv6->nexthdr != IPPROTO_TCP)
1328 			return sizeof(struct ipv6hdr) + sizeof(struct tcphdr);
1329 		hlen = sizeof(struct ipv6hdr);
1330 		hdr.ptr += sizeof(struct ipv6hdr);
1331 	} else {
1332 		/* Non-IP pkt, dont estimate header length */
1333 		return 0;
1334 	}
1335 
1336 	if (hlen + sizeof(struct tcphdr) > maplen)
1337 		return 0;
1338 
1339 	return (hlen + (hdr.tcp->doff << 2));
1340 }
1341 
1342 static int
1343 vmxnet3_rq_rx_complete(struct vmxnet3_rx_queue *rq,
1344 		       struct vmxnet3_adapter *adapter, int quota)
1345 {
1346 	static const u32 rxprod_reg[2] = {
1347 		VMXNET3_REG_RXPROD, VMXNET3_REG_RXPROD2
1348 	};
1349 	u32 num_pkts = 0;
1350 	bool skip_page_frags = false;
1351 	struct Vmxnet3_RxCompDesc *rcd;
1352 	struct vmxnet3_rx_ctx *ctx = &rq->rx_ctx;
1353 	u16 segCnt = 0, mss = 0;
1354 #ifdef __BIG_ENDIAN_BITFIELD
1355 	struct Vmxnet3_RxDesc rxCmdDesc;
1356 	struct Vmxnet3_RxCompDesc rxComp;
1357 #endif
1358 	vmxnet3_getRxComp(rcd, &rq->comp_ring.base[rq->comp_ring.next2proc].rcd,
1359 			  &rxComp);
1360 	while (rcd->gen == rq->comp_ring.gen) {
1361 		struct vmxnet3_rx_buf_info *rbi;
1362 		struct sk_buff *skb, *new_skb = NULL;
1363 		struct page *new_page = NULL;
1364 		dma_addr_t new_dma_addr;
1365 		int num_to_alloc;
1366 		struct Vmxnet3_RxDesc *rxd;
1367 		u32 idx, ring_idx;
1368 		struct vmxnet3_cmd_ring	*ring = NULL;
1369 		if (num_pkts >= quota) {
1370 			/* we may stop even before we see the EOP desc of
1371 			 * the current pkt
1372 			 */
1373 			break;
1374 		}
1375 
1376 		/* Prevent any rcd field from being (speculatively) read before
1377 		 * rcd->gen is read.
1378 		 */
1379 		dma_rmb();
1380 
1381 		BUG_ON(rcd->rqID != rq->qid && rcd->rqID != rq->qid2 &&
1382 		       rcd->rqID != rq->dataRingQid);
1383 		idx = rcd->rxdIdx;
1384 		ring_idx = VMXNET3_GET_RING_IDX(adapter, rcd->rqID);
1385 		ring = rq->rx_ring + ring_idx;
1386 		vmxnet3_getRxDesc(rxd, &rq->rx_ring[ring_idx].base[idx].rxd,
1387 				  &rxCmdDesc);
1388 		rbi = rq->buf_info[ring_idx] + idx;
1389 
1390 		BUG_ON(rxd->addr != rbi->dma_addr ||
1391 		       rxd->len != rbi->len);
1392 
1393 		if (unlikely(rcd->eop && rcd->err)) {
1394 			vmxnet3_rx_error(rq, rcd, ctx, adapter);
1395 			goto rcd_done;
1396 		}
1397 
1398 		if (rcd->sop) { /* first buf of the pkt */
1399 			bool rxDataRingUsed;
1400 			u16 len;
1401 
1402 			BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_HEAD ||
1403 			       (rcd->rqID != rq->qid &&
1404 				rcd->rqID != rq->dataRingQid));
1405 
1406 			BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_SKB);
1407 			BUG_ON(ctx->skb != NULL || rbi->skb == NULL);
1408 
1409 			if (unlikely(rcd->len == 0)) {
1410 				/* Pretend the rx buffer is skipped. */
1411 				BUG_ON(!(rcd->sop && rcd->eop));
1412 				netdev_dbg(adapter->netdev,
1413 					"rxRing[%u][%u] 0 length\n",
1414 					ring_idx, idx);
1415 				goto rcd_done;
1416 			}
1417 
1418 			skip_page_frags = false;
1419 			ctx->skb = rbi->skb;
1420 
1421 			rxDataRingUsed =
1422 				VMXNET3_RX_DATA_RING(adapter, rcd->rqID);
1423 			len = rxDataRingUsed ? rcd->len : rbi->len;
1424 			new_skb = netdev_alloc_skb_ip_align(adapter->netdev,
1425 							    len);
1426 			if (new_skb == NULL) {
1427 				/* Skb allocation failed, do not handover this
1428 				 * skb to stack. Reuse it. Drop the existing pkt
1429 				 */
1430 				rq->stats.rx_buf_alloc_failure++;
1431 				ctx->skb = NULL;
1432 				rq->stats.drop_total++;
1433 				skip_page_frags = true;
1434 				goto rcd_done;
1435 			}
1436 
1437 			if (rxDataRingUsed) {
1438 				size_t sz;
1439 
1440 				BUG_ON(rcd->len > rq->data_ring.desc_size);
1441 
1442 				ctx->skb = new_skb;
1443 				sz = rcd->rxdIdx * rq->data_ring.desc_size;
1444 				memcpy(new_skb->data,
1445 				       &rq->data_ring.base[sz], rcd->len);
1446 			} else {
1447 				ctx->skb = rbi->skb;
1448 
1449 				new_dma_addr =
1450 					dma_map_single(&adapter->pdev->dev,
1451 						       new_skb->data, rbi->len,
1452 						       DMA_FROM_DEVICE);
1453 				if (dma_mapping_error(&adapter->pdev->dev,
1454 						      new_dma_addr)) {
1455 					dev_kfree_skb(new_skb);
1456 					/* Skb allocation failed, do not
1457 					 * handover this skb to stack. Reuse
1458 					 * it. Drop the existing pkt.
1459 					 */
1460 					rq->stats.rx_buf_alloc_failure++;
1461 					ctx->skb = NULL;
1462 					rq->stats.drop_total++;
1463 					skip_page_frags = true;
1464 					goto rcd_done;
1465 				}
1466 
1467 				dma_unmap_single(&adapter->pdev->dev,
1468 						 rbi->dma_addr,
1469 						 rbi->len,
1470 						 DMA_FROM_DEVICE);
1471 
1472 				/* Immediate refill */
1473 				rbi->skb = new_skb;
1474 				rbi->dma_addr = new_dma_addr;
1475 				rxd->addr = cpu_to_le64(rbi->dma_addr);
1476 				rxd->len = rbi->len;
1477 			}
1478 
1479 #ifdef VMXNET3_RSS
1480 			if (rcd->rssType != VMXNET3_RCD_RSS_TYPE_NONE &&
1481 			    (adapter->netdev->features & NETIF_F_RXHASH)) {
1482 				enum pkt_hash_types hash_type;
1483 
1484 				switch (rcd->rssType) {
1485 				case VMXNET3_RCD_RSS_TYPE_IPV4:
1486 				case VMXNET3_RCD_RSS_TYPE_IPV6:
1487 					hash_type = PKT_HASH_TYPE_L3;
1488 					break;
1489 				case VMXNET3_RCD_RSS_TYPE_TCPIPV4:
1490 				case VMXNET3_RCD_RSS_TYPE_TCPIPV6:
1491 				case VMXNET3_RCD_RSS_TYPE_UDPIPV4:
1492 				case VMXNET3_RCD_RSS_TYPE_UDPIPV6:
1493 					hash_type = PKT_HASH_TYPE_L4;
1494 					break;
1495 				default:
1496 					hash_type = PKT_HASH_TYPE_L3;
1497 					break;
1498 				}
1499 				skb_set_hash(ctx->skb,
1500 					     le32_to_cpu(rcd->rssHash),
1501 					     hash_type);
1502 			}
1503 #endif
1504 			skb_put(ctx->skb, rcd->len);
1505 
1506 			if (VMXNET3_VERSION_GE_2(adapter) &&
1507 			    rcd->type == VMXNET3_CDTYPE_RXCOMP_LRO) {
1508 				struct Vmxnet3_RxCompDescExt *rcdlro;
1509 				rcdlro = (struct Vmxnet3_RxCompDescExt *)rcd;
1510 
1511 				segCnt = rcdlro->segCnt;
1512 				WARN_ON_ONCE(segCnt == 0);
1513 				mss = rcdlro->mss;
1514 				if (unlikely(segCnt <= 1))
1515 					segCnt = 0;
1516 			} else {
1517 				segCnt = 0;
1518 			}
1519 		} else {
1520 			BUG_ON(ctx->skb == NULL && !skip_page_frags);
1521 
1522 			/* non SOP buffer must be type 1 in most cases */
1523 			BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_PAGE);
1524 			BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_BODY);
1525 
1526 			/* If an sop buffer was dropped, skip all
1527 			 * following non-sop fragments. They will be reused.
1528 			 */
1529 			if (skip_page_frags)
1530 				goto rcd_done;
1531 
1532 			if (rcd->len) {
1533 				new_page = alloc_page(GFP_ATOMIC);
1534 				/* Replacement page frag could not be allocated.
1535 				 * Reuse this page. Drop the pkt and free the
1536 				 * skb which contained this page as a frag. Skip
1537 				 * processing all the following non-sop frags.
1538 				 */
1539 				if (unlikely(!new_page)) {
1540 					rq->stats.rx_buf_alloc_failure++;
1541 					dev_kfree_skb(ctx->skb);
1542 					ctx->skb = NULL;
1543 					skip_page_frags = true;
1544 					goto rcd_done;
1545 				}
1546 				new_dma_addr = dma_map_page(&adapter->pdev->dev,
1547 							    new_page,
1548 							    0, PAGE_SIZE,
1549 							    DMA_FROM_DEVICE);
1550 				if (dma_mapping_error(&adapter->pdev->dev,
1551 						      new_dma_addr)) {
1552 					put_page(new_page);
1553 					rq->stats.rx_buf_alloc_failure++;
1554 					dev_kfree_skb(ctx->skb);
1555 					ctx->skb = NULL;
1556 					skip_page_frags = true;
1557 					goto rcd_done;
1558 				}
1559 
1560 				dma_unmap_page(&adapter->pdev->dev,
1561 					       rbi->dma_addr, rbi->len,
1562 					       DMA_FROM_DEVICE);
1563 
1564 				vmxnet3_append_frag(ctx->skb, rcd, rbi);
1565 
1566 				/* Immediate refill */
1567 				rbi->page = new_page;
1568 				rbi->dma_addr = new_dma_addr;
1569 				rxd->addr = cpu_to_le64(rbi->dma_addr);
1570 				rxd->len = rbi->len;
1571 			}
1572 		}
1573 
1574 
1575 		skb = ctx->skb;
1576 		if (rcd->eop) {
1577 			u32 mtu = adapter->netdev->mtu;
1578 			skb->len += skb->data_len;
1579 
1580 			vmxnet3_rx_csum(adapter, skb,
1581 					(union Vmxnet3_GenericDesc *)rcd);
1582 			skb->protocol = eth_type_trans(skb, adapter->netdev);
1583 			if (!rcd->tcp ||
1584 			    !(adapter->netdev->features & NETIF_F_LRO))
1585 				goto not_lro;
1586 
1587 			if (segCnt != 0 && mss != 0) {
1588 				skb_shinfo(skb)->gso_type = rcd->v4 ?
1589 					SKB_GSO_TCPV4 : SKB_GSO_TCPV6;
1590 				skb_shinfo(skb)->gso_size = mss;
1591 				skb_shinfo(skb)->gso_segs = segCnt;
1592 			} else if (segCnt != 0 || skb->len > mtu) {
1593 				u32 hlen;
1594 
1595 				hlen = vmxnet3_get_hdr_len(adapter, skb,
1596 					(union Vmxnet3_GenericDesc *)rcd);
1597 				if (hlen == 0)
1598 					goto not_lro;
1599 
1600 				skb_shinfo(skb)->gso_type =
1601 					rcd->v4 ? SKB_GSO_TCPV4 : SKB_GSO_TCPV6;
1602 				if (segCnt != 0) {
1603 					skb_shinfo(skb)->gso_segs = segCnt;
1604 					skb_shinfo(skb)->gso_size =
1605 						DIV_ROUND_UP(skb->len -
1606 							hlen, segCnt);
1607 				} else {
1608 					skb_shinfo(skb)->gso_size = mtu - hlen;
1609 				}
1610 			}
1611 not_lro:
1612 			if (unlikely(rcd->ts))
1613 				__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), rcd->tci);
1614 
1615 			if (adapter->netdev->features & NETIF_F_LRO)
1616 				netif_receive_skb(skb);
1617 			else
1618 				napi_gro_receive(&rq->napi, skb);
1619 
1620 			ctx->skb = NULL;
1621 			num_pkts++;
1622 		}
1623 
1624 rcd_done:
1625 		/* device may have skipped some rx descs */
1626 		ring->next2comp = idx;
1627 		num_to_alloc = vmxnet3_cmd_ring_desc_avail(ring);
1628 		ring = rq->rx_ring + ring_idx;
1629 
1630 		/* Ensure that the writes to rxd->gen bits will be observed
1631 		 * after all other writes to rxd objects.
1632 		 */
1633 		dma_wmb();
1634 
1635 		while (num_to_alloc) {
1636 			vmxnet3_getRxDesc(rxd, &ring->base[ring->next2fill].rxd,
1637 					  &rxCmdDesc);
1638 			BUG_ON(!rxd->addr);
1639 
1640 			/* Recv desc is ready to be used by the device */
1641 			rxd->gen = ring->gen;
1642 			vmxnet3_cmd_ring_adv_next2fill(ring);
1643 			num_to_alloc--;
1644 		}
1645 
1646 		/* if needed, update the register */
1647 		if (unlikely(rq->shared->updateRxProd)) {
1648 			VMXNET3_WRITE_BAR0_REG(adapter,
1649 					       rxprod_reg[ring_idx] + rq->qid * 8,
1650 					       ring->next2fill);
1651 		}
1652 
1653 		vmxnet3_comp_ring_adv_next2proc(&rq->comp_ring);
1654 		vmxnet3_getRxComp(rcd,
1655 				  &rq->comp_ring.base[rq->comp_ring.next2proc].rcd, &rxComp);
1656 	}
1657 
1658 	return num_pkts;
1659 }
1660 
1661 
1662 static void
1663 vmxnet3_rq_cleanup(struct vmxnet3_rx_queue *rq,
1664 		   struct vmxnet3_adapter *adapter)
1665 {
1666 	u32 i, ring_idx;
1667 	struct Vmxnet3_RxDesc *rxd;
1668 
1669 	for (ring_idx = 0; ring_idx < 2; ring_idx++) {
1670 		for (i = 0; i < rq->rx_ring[ring_idx].size; i++) {
1671 #ifdef __BIG_ENDIAN_BITFIELD
1672 			struct Vmxnet3_RxDesc rxDesc;
1673 #endif
1674 			vmxnet3_getRxDesc(rxd,
1675 				&rq->rx_ring[ring_idx].base[i].rxd, &rxDesc);
1676 
1677 			if (rxd->btype == VMXNET3_RXD_BTYPE_HEAD &&
1678 					rq->buf_info[ring_idx][i].skb) {
1679 				dma_unmap_single(&adapter->pdev->dev, rxd->addr,
1680 						 rxd->len, DMA_FROM_DEVICE);
1681 				dev_kfree_skb(rq->buf_info[ring_idx][i].skb);
1682 				rq->buf_info[ring_idx][i].skb = NULL;
1683 			} else if (rxd->btype == VMXNET3_RXD_BTYPE_BODY &&
1684 					rq->buf_info[ring_idx][i].page) {
1685 				dma_unmap_page(&adapter->pdev->dev, rxd->addr,
1686 					       rxd->len, DMA_FROM_DEVICE);
1687 				put_page(rq->buf_info[ring_idx][i].page);
1688 				rq->buf_info[ring_idx][i].page = NULL;
1689 			}
1690 		}
1691 
1692 		rq->rx_ring[ring_idx].gen = VMXNET3_INIT_GEN;
1693 		rq->rx_ring[ring_idx].next2fill =
1694 					rq->rx_ring[ring_idx].next2comp = 0;
1695 	}
1696 
1697 	rq->comp_ring.gen = VMXNET3_INIT_GEN;
1698 	rq->comp_ring.next2proc = 0;
1699 }
1700 
1701 
1702 static void
1703 vmxnet3_rq_cleanup_all(struct vmxnet3_adapter *adapter)
1704 {
1705 	int i;
1706 
1707 	for (i = 0; i < adapter->num_rx_queues; i++)
1708 		vmxnet3_rq_cleanup(&adapter->rx_queue[i], adapter);
1709 }
1710 
1711 
1712 static void vmxnet3_rq_destroy(struct vmxnet3_rx_queue *rq,
1713 			       struct vmxnet3_adapter *adapter)
1714 {
1715 	int i;
1716 	int j;
1717 
1718 	/* all rx buffers must have already been freed */
1719 	for (i = 0; i < 2; i++) {
1720 		if (rq->buf_info[i]) {
1721 			for (j = 0; j < rq->rx_ring[i].size; j++)
1722 				BUG_ON(rq->buf_info[i][j].page != NULL);
1723 		}
1724 	}
1725 
1726 
1727 	for (i = 0; i < 2; i++) {
1728 		if (rq->rx_ring[i].base) {
1729 			dma_free_coherent(&adapter->pdev->dev,
1730 					  rq->rx_ring[i].size
1731 					  * sizeof(struct Vmxnet3_RxDesc),
1732 					  rq->rx_ring[i].base,
1733 					  rq->rx_ring[i].basePA);
1734 			rq->rx_ring[i].base = NULL;
1735 		}
1736 	}
1737 
1738 	if (rq->data_ring.base) {
1739 		dma_free_coherent(&adapter->pdev->dev,
1740 				  rq->rx_ring[0].size * rq->data_ring.desc_size,
1741 				  rq->data_ring.base, rq->data_ring.basePA);
1742 		rq->data_ring.base = NULL;
1743 	}
1744 
1745 	if (rq->comp_ring.base) {
1746 		dma_free_coherent(&adapter->pdev->dev, rq->comp_ring.size
1747 				  * sizeof(struct Vmxnet3_RxCompDesc),
1748 				  rq->comp_ring.base, rq->comp_ring.basePA);
1749 		rq->comp_ring.base = NULL;
1750 	}
1751 
1752 	kfree(rq->buf_info[0]);
1753 	rq->buf_info[0] = NULL;
1754 	rq->buf_info[1] = NULL;
1755 }
1756 
1757 static void
1758 vmxnet3_rq_destroy_all_rxdataring(struct vmxnet3_adapter *adapter)
1759 {
1760 	int i;
1761 
1762 	for (i = 0; i < adapter->num_rx_queues; i++) {
1763 		struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
1764 
1765 		if (rq->data_ring.base) {
1766 			dma_free_coherent(&adapter->pdev->dev,
1767 					  (rq->rx_ring[0].size *
1768 					  rq->data_ring.desc_size),
1769 					  rq->data_ring.base,
1770 					  rq->data_ring.basePA);
1771 			rq->data_ring.base = NULL;
1772 			rq->data_ring.desc_size = 0;
1773 		}
1774 	}
1775 }
1776 
1777 static int
1778 vmxnet3_rq_init(struct vmxnet3_rx_queue *rq,
1779 		struct vmxnet3_adapter  *adapter)
1780 {
1781 	int i;
1782 
1783 	/* initialize buf_info */
1784 	for (i = 0; i < rq->rx_ring[0].size; i++) {
1785 
1786 		/* 1st buf for a pkt is skbuff */
1787 		if (i % adapter->rx_buf_per_pkt == 0) {
1788 			rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_SKB;
1789 			rq->buf_info[0][i].len = adapter->skb_buf_size;
1790 		} else { /* subsequent bufs for a pkt is frag */
1791 			rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_PAGE;
1792 			rq->buf_info[0][i].len = PAGE_SIZE;
1793 		}
1794 	}
1795 	for (i = 0; i < rq->rx_ring[1].size; i++) {
1796 		rq->buf_info[1][i].buf_type = VMXNET3_RX_BUF_PAGE;
1797 		rq->buf_info[1][i].len = PAGE_SIZE;
1798 	}
1799 
1800 	/* reset internal state and allocate buffers for both rings */
1801 	for (i = 0; i < 2; i++) {
1802 		rq->rx_ring[i].next2fill = rq->rx_ring[i].next2comp = 0;
1803 
1804 		memset(rq->rx_ring[i].base, 0, rq->rx_ring[i].size *
1805 		       sizeof(struct Vmxnet3_RxDesc));
1806 		rq->rx_ring[i].gen = VMXNET3_INIT_GEN;
1807 	}
1808 	if (vmxnet3_rq_alloc_rx_buf(rq, 0, rq->rx_ring[0].size - 1,
1809 				    adapter) == 0) {
1810 		/* at least has 1 rx buffer for the 1st ring */
1811 		return -ENOMEM;
1812 	}
1813 	vmxnet3_rq_alloc_rx_buf(rq, 1, rq->rx_ring[1].size - 1, adapter);
1814 
1815 	/* reset the comp ring */
1816 	rq->comp_ring.next2proc = 0;
1817 	memset(rq->comp_ring.base, 0, rq->comp_ring.size *
1818 	       sizeof(struct Vmxnet3_RxCompDesc));
1819 	rq->comp_ring.gen = VMXNET3_INIT_GEN;
1820 
1821 	/* reset rxctx */
1822 	rq->rx_ctx.skb = NULL;
1823 
1824 	/* stats are not reset */
1825 	return 0;
1826 }
1827 
1828 
1829 static int
1830 vmxnet3_rq_init_all(struct vmxnet3_adapter *adapter)
1831 {
1832 	int i, err = 0;
1833 
1834 	for (i = 0; i < adapter->num_rx_queues; i++) {
1835 		err = vmxnet3_rq_init(&adapter->rx_queue[i], adapter);
1836 		if (unlikely(err)) {
1837 			dev_err(&adapter->netdev->dev, "%s: failed to "
1838 				"initialize rx queue%i\n",
1839 				adapter->netdev->name, i);
1840 			break;
1841 		}
1842 	}
1843 	return err;
1844 
1845 }
1846 
1847 
1848 static int
1849 vmxnet3_rq_create(struct vmxnet3_rx_queue *rq, struct vmxnet3_adapter *adapter)
1850 {
1851 	int i;
1852 	size_t sz;
1853 	struct vmxnet3_rx_buf_info *bi;
1854 
1855 	for (i = 0; i < 2; i++) {
1856 
1857 		sz = rq->rx_ring[i].size * sizeof(struct Vmxnet3_RxDesc);
1858 		rq->rx_ring[i].base = dma_alloc_coherent(
1859 						&adapter->pdev->dev, sz,
1860 						&rq->rx_ring[i].basePA,
1861 						GFP_KERNEL);
1862 		if (!rq->rx_ring[i].base) {
1863 			netdev_err(adapter->netdev,
1864 				   "failed to allocate rx ring %d\n", i);
1865 			goto err;
1866 		}
1867 	}
1868 
1869 	if ((adapter->rxdataring_enabled) && (rq->data_ring.desc_size != 0)) {
1870 		sz = rq->rx_ring[0].size * rq->data_ring.desc_size;
1871 		rq->data_ring.base =
1872 			dma_alloc_coherent(&adapter->pdev->dev, sz,
1873 					   &rq->data_ring.basePA,
1874 					   GFP_KERNEL);
1875 		if (!rq->data_ring.base) {
1876 			netdev_err(adapter->netdev,
1877 				   "rx data ring will be disabled\n");
1878 			adapter->rxdataring_enabled = false;
1879 		}
1880 	} else {
1881 		rq->data_ring.base = NULL;
1882 		rq->data_ring.desc_size = 0;
1883 	}
1884 
1885 	sz = rq->comp_ring.size * sizeof(struct Vmxnet3_RxCompDesc);
1886 	rq->comp_ring.base = dma_alloc_coherent(&adapter->pdev->dev, sz,
1887 						&rq->comp_ring.basePA,
1888 						GFP_KERNEL);
1889 	if (!rq->comp_ring.base) {
1890 		netdev_err(adapter->netdev, "failed to allocate rx comp ring\n");
1891 		goto err;
1892 	}
1893 
1894 	bi = kcalloc_node(rq->rx_ring[0].size + rq->rx_ring[1].size,
1895 			  sizeof(rq->buf_info[0][0]), GFP_KERNEL,
1896 			  dev_to_node(&adapter->pdev->dev));
1897 	if (!bi)
1898 		goto err;
1899 
1900 	rq->buf_info[0] = bi;
1901 	rq->buf_info[1] = bi + rq->rx_ring[0].size;
1902 
1903 	return 0;
1904 
1905 err:
1906 	vmxnet3_rq_destroy(rq, adapter);
1907 	return -ENOMEM;
1908 }
1909 
1910 
1911 static int
1912 vmxnet3_rq_create_all(struct vmxnet3_adapter *adapter)
1913 {
1914 	int i, err = 0;
1915 
1916 	adapter->rxdataring_enabled = VMXNET3_VERSION_GE_3(adapter);
1917 
1918 	for (i = 0; i < adapter->num_rx_queues; i++) {
1919 		err = vmxnet3_rq_create(&adapter->rx_queue[i], adapter);
1920 		if (unlikely(err)) {
1921 			dev_err(&adapter->netdev->dev,
1922 				"%s: failed to create rx queue%i\n",
1923 				adapter->netdev->name, i);
1924 			goto err_out;
1925 		}
1926 	}
1927 
1928 	if (!adapter->rxdataring_enabled)
1929 		vmxnet3_rq_destroy_all_rxdataring(adapter);
1930 
1931 	return err;
1932 err_out:
1933 	vmxnet3_rq_destroy_all(adapter);
1934 	return err;
1935 
1936 }
1937 
1938 /* Multiple queue aware polling function for tx and rx */
1939 
1940 static int
1941 vmxnet3_do_poll(struct vmxnet3_adapter *adapter, int budget)
1942 {
1943 	int rcd_done = 0, i;
1944 	if (unlikely(adapter->shared->ecr))
1945 		vmxnet3_process_events(adapter);
1946 	for (i = 0; i < adapter->num_tx_queues; i++)
1947 		vmxnet3_tq_tx_complete(&adapter->tx_queue[i], adapter);
1948 
1949 	for (i = 0; i < adapter->num_rx_queues; i++)
1950 		rcd_done += vmxnet3_rq_rx_complete(&adapter->rx_queue[i],
1951 						   adapter, budget);
1952 	return rcd_done;
1953 }
1954 
1955 
1956 static int
1957 vmxnet3_poll(struct napi_struct *napi, int budget)
1958 {
1959 	struct vmxnet3_rx_queue *rx_queue = container_of(napi,
1960 					  struct vmxnet3_rx_queue, napi);
1961 	int rxd_done;
1962 
1963 	rxd_done = vmxnet3_do_poll(rx_queue->adapter, budget);
1964 
1965 	if (rxd_done < budget) {
1966 		napi_complete_done(napi, rxd_done);
1967 		vmxnet3_enable_all_intrs(rx_queue->adapter);
1968 	}
1969 	return rxd_done;
1970 }
1971 
1972 /*
1973  * NAPI polling function for MSI-X mode with multiple Rx queues
1974  * Returns the # of the NAPI credit consumed (# of rx descriptors processed)
1975  */
1976 
1977 static int
1978 vmxnet3_poll_rx_only(struct napi_struct *napi, int budget)
1979 {
1980 	struct vmxnet3_rx_queue *rq = container_of(napi,
1981 						struct vmxnet3_rx_queue, napi);
1982 	struct vmxnet3_adapter *adapter = rq->adapter;
1983 	int rxd_done;
1984 
1985 	/* When sharing interrupt with corresponding tx queue, process
1986 	 * tx completions in that queue as well
1987 	 */
1988 	if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE) {
1989 		struct vmxnet3_tx_queue *tq =
1990 				&adapter->tx_queue[rq - adapter->rx_queue];
1991 		vmxnet3_tq_tx_complete(tq, adapter);
1992 	}
1993 
1994 	rxd_done = vmxnet3_rq_rx_complete(rq, adapter, budget);
1995 
1996 	if (rxd_done < budget) {
1997 		napi_complete_done(napi, rxd_done);
1998 		vmxnet3_enable_intr(adapter, rq->comp_ring.intr_idx);
1999 	}
2000 	return rxd_done;
2001 }
2002 
2003 
2004 #ifdef CONFIG_PCI_MSI
2005 
2006 /*
2007  * Handle completion interrupts on tx queues
2008  * Returns whether or not the intr is handled
2009  */
2010 
2011 static irqreturn_t
2012 vmxnet3_msix_tx(int irq, void *data)
2013 {
2014 	struct vmxnet3_tx_queue *tq = data;
2015 	struct vmxnet3_adapter *adapter = tq->adapter;
2016 
2017 	if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
2018 		vmxnet3_disable_intr(adapter, tq->comp_ring.intr_idx);
2019 
2020 	/* Handle the case where only one irq is allocate for all tx queues */
2021 	if (adapter->share_intr == VMXNET3_INTR_TXSHARE) {
2022 		int i;
2023 		for (i = 0; i < adapter->num_tx_queues; i++) {
2024 			struct vmxnet3_tx_queue *txq = &adapter->tx_queue[i];
2025 			vmxnet3_tq_tx_complete(txq, adapter);
2026 		}
2027 	} else {
2028 		vmxnet3_tq_tx_complete(tq, adapter);
2029 	}
2030 	vmxnet3_enable_intr(adapter, tq->comp_ring.intr_idx);
2031 
2032 	return IRQ_HANDLED;
2033 }
2034 
2035 
2036 /*
2037  * Handle completion interrupts on rx queues. Returns whether or not the
2038  * intr is handled
2039  */
2040 
2041 static irqreturn_t
2042 vmxnet3_msix_rx(int irq, void *data)
2043 {
2044 	struct vmxnet3_rx_queue *rq = data;
2045 	struct vmxnet3_adapter *adapter = rq->adapter;
2046 
2047 	/* disable intr if needed */
2048 	if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
2049 		vmxnet3_disable_intr(adapter, rq->comp_ring.intr_idx);
2050 	napi_schedule(&rq->napi);
2051 
2052 	return IRQ_HANDLED;
2053 }
2054 
2055 /*
2056  *----------------------------------------------------------------------------
2057  *
2058  * vmxnet3_msix_event --
2059  *
2060  *    vmxnet3 msix event intr handler
2061  *
2062  * Result:
2063  *    whether or not the intr is handled
2064  *
2065  *----------------------------------------------------------------------------
2066  */
2067 
2068 static irqreturn_t
2069 vmxnet3_msix_event(int irq, void *data)
2070 {
2071 	struct net_device *dev = data;
2072 	struct vmxnet3_adapter *adapter = netdev_priv(dev);
2073 
2074 	/* disable intr if needed */
2075 	if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
2076 		vmxnet3_disable_intr(adapter, adapter->intr.event_intr_idx);
2077 
2078 	if (adapter->shared->ecr)
2079 		vmxnet3_process_events(adapter);
2080 
2081 	vmxnet3_enable_intr(adapter, adapter->intr.event_intr_idx);
2082 
2083 	return IRQ_HANDLED;
2084 }
2085 
2086 #endif /* CONFIG_PCI_MSI  */
2087 
2088 
2089 /* Interrupt handler for vmxnet3  */
2090 static irqreturn_t
2091 vmxnet3_intr(int irq, void *dev_id)
2092 {
2093 	struct net_device *dev = dev_id;
2094 	struct vmxnet3_adapter *adapter = netdev_priv(dev);
2095 
2096 	if (adapter->intr.type == VMXNET3_IT_INTX) {
2097 		u32 icr = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_ICR);
2098 		if (unlikely(icr == 0))
2099 			/* not ours */
2100 			return IRQ_NONE;
2101 	}
2102 
2103 
2104 	/* disable intr if needed */
2105 	if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
2106 		vmxnet3_disable_all_intrs(adapter);
2107 
2108 	napi_schedule(&adapter->rx_queue[0].napi);
2109 
2110 	return IRQ_HANDLED;
2111 }
2112 
2113 #ifdef CONFIG_NET_POLL_CONTROLLER
2114 
2115 /* netpoll callback. */
2116 static void
2117 vmxnet3_netpoll(struct net_device *netdev)
2118 {
2119 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2120 
2121 	switch (adapter->intr.type) {
2122 #ifdef CONFIG_PCI_MSI
2123 	case VMXNET3_IT_MSIX: {
2124 		int i;
2125 		for (i = 0; i < adapter->num_rx_queues; i++)
2126 			vmxnet3_msix_rx(0, &adapter->rx_queue[i]);
2127 		break;
2128 	}
2129 #endif
2130 	case VMXNET3_IT_MSI:
2131 	default:
2132 		vmxnet3_intr(0, adapter->netdev);
2133 		break;
2134 	}
2135 
2136 }
2137 #endif	/* CONFIG_NET_POLL_CONTROLLER */
2138 
2139 static int
2140 vmxnet3_request_irqs(struct vmxnet3_adapter *adapter)
2141 {
2142 	struct vmxnet3_intr *intr = &adapter->intr;
2143 	int err = 0, i;
2144 	int vector = 0;
2145 
2146 #ifdef CONFIG_PCI_MSI
2147 	if (adapter->intr.type == VMXNET3_IT_MSIX) {
2148 		for (i = 0; i < adapter->num_tx_queues; i++) {
2149 			if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE) {
2150 				sprintf(adapter->tx_queue[i].name, "%s-tx-%d",
2151 					adapter->netdev->name, vector);
2152 				err = request_irq(
2153 					      intr->msix_entries[vector].vector,
2154 					      vmxnet3_msix_tx, 0,
2155 					      adapter->tx_queue[i].name,
2156 					      &adapter->tx_queue[i]);
2157 			} else {
2158 				sprintf(adapter->tx_queue[i].name, "%s-rxtx-%d",
2159 					adapter->netdev->name, vector);
2160 			}
2161 			if (err) {
2162 				dev_err(&adapter->netdev->dev,
2163 					"Failed to request irq for MSIX, %s, "
2164 					"error %d\n",
2165 					adapter->tx_queue[i].name, err);
2166 				return err;
2167 			}
2168 
2169 			/* Handle the case where only 1 MSIx was allocated for
2170 			 * all tx queues */
2171 			if (adapter->share_intr == VMXNET3_INTR_TXSHARE) {
2172 				for (; i < adapter->num_tx_queues; i++)
2173 					adapter->tx_queue[i].comp_ring.intr_idx
2174 								= vector;
2175 				vector++;
2176 				break;
2177 			} else {
2178 				adapter->tx_queue[i].comp_ring.intr_idx
2179 								= vector++;
2180 			}
2181 		}
2182 		if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE)
2183 			vector = 0;
2184 
2185 		for (i = 0; i < adapter->num_rx_queues; i++) {
2186 			if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE)
2187 				sprintf(adapter->rx_queue[i].name, "%s-rx-%d",
2188 					adapter->netdev->name, vector);
2189 			else
2190 				sprintf(adapter->rx_queue[i].name, "%s-rxtx-%d",
2191 					adapter->netdev->name, vector);
2192 			err = request_irq(intr->msix_entries[vector].vector,
2193 					  vmxnet3_msix_rx, 0,
2194 					  adapter->rx_queue[i].name,
2195 					  &(adapter->rx_queue[i]));
2196 			if (err) {
2197 				netdev_err(adapter->netdev,
2198 					   "Failed to request irq for MSIX, "
2199 					   "%s, error %d\n",
2200 					   adapter->rx_queue[i].name, err);
2201 				return err;
2202 			}
2203 
2204 			adapter->rx_queue[i].comp_ring.intr_idx = vector++;
2205 		}
2206 
2207 		sprintf(intr->event_msi_vector_name, "%s-event-%d",
2208 			adapter->netdev->name, vector);
2209 		err = request_irq(intr->msix_entries[vector].vector,
2210 				  vmxnet3_msix_event, 0,
2211 				  intr->event_msi_vector_name, adapter->netdev);
2212 		intr->event_intr_idx = vector;
2213 
2214 	} else if (intr->type == VMXNET3_IT_MSI) {
2215 		adapter->num_rx_queues = 1;
2216 		err = request_irq(adapter->pdev->irq, vmxnet3_intr, 0,
2217 				  adapter->netdev->name, adapter->netdev);
2218 	} else {
2219 #endif
2220 		adapter->num_rx_queues = 1;
2221 		err = request_irq(adapter->pdev->irq, vmxnet3_intr,
2222 				  IRQF_SHARED, adapter->netdev->name,
2223 				  adapter->netdev);
2224 #ifdef CONFIG_PCI_MSI
2225 	}
2226 #endif
2227 	intr->num_intrs = vector + 1;
2228 	if (err) {
2229 		netdev_err(adapter->netdev,
2230 			   "Failed to request irq (intr type:%d), error %d\n",
2231 			   intr->type, err);
2232 	} else {
2233 		/* Number of rx queues will not change after this */
2234 		for (i = 0; i < adapter->num_rx_queues; i++) {
2235 			struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
2236 			rq->qid = i;
2237 			rq->qid2 = i + adapter->num_rx_queues;
2238 			rq->dataRingQid = i + 2 * adapter->num_rx_queues;
2239 		}
2240 
2241 		/* init our intr settings */
2242 		for (i = 0; i < intr->num_intrs; i++)
2243 			intr->mod_levels[i] = UPT1_IML_ADAPTIVE;
2244 		if (adapter->intr.type != VMXNET3_IT_MSIX) {
2245 			adapter->intr.event_intr_idx = 0;
2246 			for (i = 0; i < adapter->num_tx_queues; i++)
2247 				adapter->tx_queue[i].comp_ring.intr_idx = 0;
2248 			adapter->rx_queue[0].comp_ring.intr_idx = 0;
2249 		}
2250 
2251 		netdev_info(adapter->netdev,
2252 			    "intr type %u, mode %u, %u vectors allocated\n",
2253 			    intr->type, intr->mask_mode, intr->num_intrs);
2254 	}
2255 
2256 	return err;
2257 }
2258 
2259 
2260 static void
2261 vmxnet3_free_irqs(struct vmxnet3_adapter *adapter)
2262 {
2263 	struct vmxnet3_intr *intr = &adapter->intr;
2264 	BUG_ON(intr->type == VMXNET3_IT_AUTO || intr->num_intrs <= 0);
2265 
2266 	switch (intr->type) {
2267 #ifdef CONFIG_PCI_MSI
2268 	case VMXNET3_IT_MSIX:
2269 	{
2270 		int i, vector = 0;
2271 
2272 		if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE) {
2273 			for (i = 0; i < adapter->num_tx_queues; i++) {
2274 				free_irq(intr->msix_entries[vector++].vector,
2275 					 &(adapter->tx_queue[i]));
2276 				if (adapter->share_intr == VMXNET3_INTR_TXSHARE)
2277 					break;
2278 			}
2279 		}
2280 
2281 		for (i = 0; i < adapter->num_rx_queues; i++) {
2282 			free_irq(intr->msix_entries[vector++].vector,
2283 				 &(adapter->rx_queue[i]));
2284 		}
2285 
2286 		free_irq(intr->msix_entries[vector].vector,
2287 			 adapter->netdev);
2288 		BUG_ON(vector >= intr->num_intrs);
2289 		break;
2290 	}
2291 #endif
2292 	case VMXNET3_IT_MSI:
2293 		free_irq(adapter->pdev->irq, adapter->netdev);
2294 		break;
2295 	case VMXNET3_IT_INTX:
2296 		free_irq(adapter->pdev->irq, adapter->netdev);
2297 		break;
2298 	default:
2299 		BUG();
2300 	}
2301 }
2302 
2303 
2304 static void
2305 vmxnet3_restore_vlan(struct vmxnet3_adapter *adapter)
2306 {
2307 	u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
2308 	u16 vid;
2309 
2310 	/* allow untagged pkts */
2311 	VMXNET3_SET_VFTABLE_ENTRY(vfTable, 0);
2312 
2313 	for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
2314 		VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid);
2315 }
2316 
2317 
2318 static int
2319 vmxnet3_vlan_rx_add_vid(struct net_device *netdev, __be16 proto, u16 vid)
2320 {
2321 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2322 
2323 	if (!(netdev->flags & IFF_PROMISC)) {
2324 		u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
2325 		unsigned long flags;
2326 
2327 		VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid);
2328 		spin_lock_irqsave(&adapter->cmd_lock, flags);
2329 		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2330 				       VMXNET3_CMD_UPDATE_VLAN_FILTERS);
2331 		spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2332 	}
2333 
2334 	set_bit(vid, adapter->active_vlans);
2335 
2336 	return 0;
2337 }
2338 
2339 
2340 static int
2341 vmxnet3_vlan_rx_kill_vid(struct net_device *netdev, __be16 proto, u16 vid)
2342 {
2343 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2344 
2345 	if (!(netdev->flags & IFF_PROMISC)) {
2346 		u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
2347 		unsigned long flags;
2348 
2349 		VMXNET3_CLEAR_VFTABLE_ENTRY(vfTable, vid);
2350 		spin_lock_irqsave(&adapter->cmd_lock, flags);
2351 		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2352 				       VMXNET3_CMD_UPDATE_VLAN_FILTERS);
2353 		spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2354 	}
2355 
2356 	clear_bit(vid, adapter->active_vlans);
2357 
2358 	return 0;
2359 }
2360 
2361 
2362 static u8 *
2363 vmxnet3_copy_mc(struct net_device *netdev)
2364 {
2365 	u8 *buf = NULL;
2366 	u32 sz = netdev_mc_count(netdev) * ETH_ALEN;
2367 
2368 	/* struct Vmxnet3_RxFilterConf.mfTableLen is u16. */
2369 	if (sz <= 0xffff) {
2370 		/* We may be called with BH disabled */
2371 		buf = kmalloc(sz, GFP_ATOMIC);
2372 		if (buf) {
2373 			struct netdev_hw_addr *ha;
2374 			int i = 0;
2375 
2376 			netdev_for_each_mc_addr(ha, netdev)
2377 				memcpy(buf + i++ * ETH_ALEN, ha->addr,
2378 				       ETH_ALEN);
2379 		}
2380 	}
2381 	return buf;
2382 }
2383 
2384 
2385 static void
2386 vmxnet3_set_mc(struct net_device *netdev)
2387 {
2388 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2389 	unsigned long flags;
2390 	struct Vmxnet3_RxFilterConf *rxConf =
2391 					&adapter->shared->devRead.rxFilterConf;
2392 	u8 *new_table = NULL;
2393 	dma_addr_t new_table_pa = 0;
2394 	bool new_table_pa_valid = false;
2395 	u32 new_mode = VMXNET3_RXM_UCAST;
2396 
2397 	if (netdev->flags & IFF_PROMISC) {
2398 		u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
2399 		memset(vfTable, 0, VMXNET3_VFT_SIZE * sizeof(*vfTable));
2400 
2401 		new_mode |= VMXNET3_RXM_PROMISC;
2402 	} else {
2403 		vmxnet3_restore_vlan(adapter);
2404 	}
2405 
2406 	if (netdev->flags & IFF_BROADCAST)
2407 		new_mode |= VMXNET3_RXM_BCAST;
2408 
2409 	if (netdev->flags & IFF_ALLMULTI)
2410 		new_mode |= VMXNET3_RXM_ALL_MULTI;
2411 	else
2412 		if (!netdev_mc_empty(netdev)) {
2413 			new_table = vmxnet3_copy_mc(netdev);
2414 			if (new_table) {
2415 				size_t sz = netdev_mc_count(netdev) * ETH_ALEN;
2416 
2417 				rxConf->mfTableLen = cpu_to_le16(sz);
2418 				new_table_pa = dma_map_single(
2419 							&adapter->pdev->dev,
2420 							new_table,
2421 							sz,
2422 							DMA_TO_DEVICE);
2423 				if (!dma_mapping_error(&adapter->pdev->dev,
2424 						       new_table_pa)) {
2425 					new_mode |= VMXNET3_RXM_MCAST;
2426 					new_table_pa_valid = true;
2427 					rxConf->mfTablePA = cpu_to_le64(
2428 								new_table_pa);
2429 				}
2430 			}
2431 			if (!new_table_pa_valid) {
2432 				netdev_info(netdev,
2433 					    "failed to copy mcast list, setting ALL_MULTI\n");
2434 				new_mode |= VMXNET3_RXM_ALL_MULTI;
2435 			}
2436 		}
2437 
2438 	if (!(new_mode & VMXNET3_RXM_MCAST)) {
2439 		rxConf->mfTableLen = 0;
2440 		rxConf->mfTablePA = 0;
2441 	}
2442 
2443 	spin_lock_irqsave(&adapter->cmd_lock, flags);
2444 	if (new_mode != rxConf->rxMode) {
2445 		rxConf->rxMode = cpu_to_le32(new_mode);
2446 		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2447 				       VMXNET3_CMD_UPDATE_RX_MODE);
2448 		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2449 				       VMXNET3_CMD_UPDATE_VLAN_FILTERS);
2450 	}
2451 
2452 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2453 			       VMXNET3_CMD_UPDATE_MAC_FILTERS);
2454 	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2455 
2456 	if (new_table_pa_valid)
2457 		dma_unmap_single(&adapter->pdev->dev, new_table_pa,
2458 				 rxConf->mfTableLen, DMA_TO_DEVICE);
2459 	kfree(new_table);
2460 }
2461 
2462 void
2463 vmxnet3_rq_destroy_all(struct vmxnet3_adapter *adapter)
2464 {
2465 	int i;
2466 
2467 	for (i = 0; i < adapter->num_rx_queues; i++)
2468 		vmxnet3_rq_destroy(&adapter->rx_queue[i], adapter);
2469 }
2470 
2471 
2472 /*
2473  *   Set up driver_shared based on settings in adapter.
2474  */
2475 
2476 static void
2477 vmxnet3_setup_driver_shared(struct vmxnet3_adapter *adapter)
2478 {
2479 	struct Vmxnet3_DriverShared *shared = adapter->shared;
2480 	struct Vmxnet3_DSDevRead *devRead = &shared->devRead;
2481 	struct Vmxnet3_DSDevReadExt *devReadExt = &shared->devReadExt;
2482 	struct Vmxnet3_TxQueueConf *tqc;
2483 	struct Vmxnet3_RxQueueConf *rqc;
2484 	int i;
2485 
2486 	memset(shared, 0, sizeof(*shared));
2487 
2488 	/* driver settings */
2489 	shared->magic = cpu_to_le32(VMXNET3_REV1_MAGIC);
2490 	devRead->misc.driverInfo.version = cpu_to_le32(
2491 						VMXNET3_DRIVER_VERSION_NUM);
2492 	devRead->misc.driverInfo.gos.gosBits = (sizeof(void *) == 4 ?
2493 				VMXNET3_GOS_BITS_32 : VMXNET3_GOS_BITS_64);
2494 	devRead->misc.driverInfo.gos.gosType = VMXNET3_GOS_TYPE_LINUX;
2495 	*((u32 *)&devRead->misc.driverInfo.gos) = cpu_to_le32(
2496 				*((u32 *)&devRead->misc.driverInfo.gos));
2497 	devRead->misc.driverInfo.vmxnet3RevSpt = cpu_to_le32(1);
2498 	devRead->misc.driverInfo.uptVerSpt = cpu_to_le32(1);
2499 
2500 	devRead->misc.ddPA = cpu_to_le64(adapter->adapter_pa);
2501 	devRead->misc.ddLen = cpu_to_le32(sizeof(struct vmxnet3_adapter));
2502 
2503 	/* set up feature flags */
2504 	if (adapter->netdev->features & NETIF_F_RXCSUM)
2505 		devRead->misc.uptFeatures |= UPT1_F_RXCSUM;
2506 
2507 	if (adapter->netdev->features & NETIF_F_LRO) {
2508 		devRead->misc.uptFeatures |= UPT1_F_LRO;
2509 		devRead->misc.maxNumRxSG = cpu_to_le16(1 + MAX_SKB_FRAGS);
2510 	}
2511 	if (adapter->netdev->features & NETIF_F_HW_VLAN_CTAG_RX)
2512 		devRead->misc.uptFeatures |= UPT1_F_RXVLAN;
2513 
2514 	if (adapter->netdev->features & (NETIF_F_GSO_UDP_TUNNEL |
2515 					 NETIF_F_GSO_UDP_TUNNEL_CSUM))
2516 		devRead->misc.uptFeatures |= UPT1_F_RXINNEROFLD;
2517 
2518 	devRead->misc.mtu = cpu_to_le32(adapter->netdev->mtu);
2519 	devRead->misc.queueDescPA = cpu_to_le64(adapter->queue_desc_pa);
2520 	devRead->misc.queueDescLen = cpu_to_le32(
2521 		adapter->num_tx_queues * sizeof(struct Vmxnet3_TxQueueDesc) +
2522 		adapter->num_rx_queues * sizeof(struct Vmxnet3_RxQueueDesc));
2523 
2524 	/* tx queue settings */
2525 	devRead->misc.numTxQueues =  adapter->num_tx_queues;
2526 	for (i = 0; i < adapter->num_tx_queues; i++) {
2527 		struct vmxnet3_tx_queue	*tq = &adapter->tx_queue[i];
2528 		BUG_ON(adapter->tx_queue[i].tx_ring.base == NULL);
2529 		tqc = &adapter->tqd_start[i].conf;
2530 		tqc->txRingBasePA   = cpu_to_le64(tq->tx_ring.basePA);
2531 		tqc->dataRingBasePA = cpu_to_le64(tq->data_ring.basePA);
2532 		tqc->compRingBasePA = cpu_to_le64(tq->comp_ring.basePA);
2533 		tqc->ddPA           = cpu_to_le64(~0ULL);
2534 		tqc->txRingSize     = cpu_to_le32(tq->tx_ring.size);
2535 		tqc->dataRingSize   = cpu_to_le32(tq->data_ring.size);
2536 		tqc->txDataRingDescSize = cpu_to_le32(tq->txdata_desc_size);
2537 		tqc->compRingSize   = cpu_to_le32(tq->comp_ring.size);
2538 		tqc->ddLen          = cpu_to_le32(0);
2539 		tqc->intrIdx        = tq->comp_ring.intr_idx;
2540 	}
2541 
2542 	/* rx queue settings */
2543 	devRead->misc.numRxQueues = adapter->num_rx_queues;
2544 	for (i = 0; i < adapter->num_rx_queues; i++) {
2545 		struct vmxnet3_rx_queue	*rq = &adapter->rx_queue[i];
2546 		rqc = &adapter->rqd_start[i].conf;
2547 		rqc->rxRingBasePA[0] = cpu_to_le64(rq->rx_ring[0].basePA);
2548 		rqc->rxRingBasePA[1] = cpu_to_le64(rq->rx_ring[1].basePA);
2549 		rqc->compRingBasePA  = cpu_to_le64(rq->comp_ring.basePA);
2550 		rqc->ddPA            = cpu_to_le64(~0ULL);
2551 		rqc->rxRingSize[0]   = cpu_to_le32(rq->rx_ring[0].size);
2552 		rqc->rxRingSize[1]   = cpu_to_le32(rq->rx_ring[1].size);
2553 		rqc->compRingSize    = cpu_to_le32(rq->comp_ring.size);
2554 		rqc->ddLen           = cpu_to_le32(0);
2555 		rqc->intrIdx         = rq->comp_ring.intr_idx;
2556 		if (VMXNET3_VERSION_GE_3(adapter)) {
2557 			rqc->rxDataRingBasePA =
2558 				cpu_to_le64(rq->data_ring.basePA);
2559 			rqc->rxDataRingDescSize =
2560 				cpu_to_le16(rq->data_ring.desc_size);
2561 		}
2562 	}
2563 
2564 #ifdef VMXNET3_RSS
2565 	memset(adapter->rss_conf, 0, sizeof(*adapter->rss_conf));
2566 
2567 	if (adapter->rss) {
2568 		struct UPT1_RSSConf *rssConf = adapter->rss_conf;
2569 
2570 		devRead->misc.uptFeatures |= UPT1_F_RSS;
2571 		devRead->misc.numRxQueues = adapter->num_rx_queues;
2572 		rssConf->hashType = UPT1_RSS_HASH_TYPE_TCP_IPV4 |
2573 				    UPT1_RSS_HASH_TYPE_IPV4 |
2574 				    UPT1_RSS_HASH_TYPE_TCP_IPV6 |
2575 				    UPT1_RSS_HASH_TYPE_IPV6;
2576 		rssConf->hashFunc = UPT1_RSS_HASH_FUNC_TOEPLITZ;
2577 		rssConf->hashKeySize = UPT1_RSS_MAX_KEY_SIZE;
2578 		rssConf->indTableSize = VMXNET3_RSS_IND_TABLE_SIZE;
2579 		netdev_rss_key_fill(rssConf->hashKey, sizeof(rssConf->hashKey));
2580 
2581 		for (i = 0; i < rssConf->indTableSize; i++)
2582 			rssConf->indTable[i] = ethtool_rxfh_indir_default(
2583 				i, adapter->num_rx_queues);
2584 
2585 		devRead->rssConfDesc.confVer = 1;
2586 		devRead->rssConfDesc.confLen = cpu_to_le32(sizeof(*rssConf));
2587 		devRead->rssConfDesc.confPA =
2588 			cpu_to_le64(adapter->rss_conf_pa);
2589 	}
2590 
2591 #endif /* VMXNET3_RSS */
2592 
2593 	/* intr settings */
2594 	if (!VMXNET3_VERSION_GE_6(adapter) ||
2595 	    !adapter->queuesExtEnabled) {
2596 		devRead->intrConf.autoMask = adapter->intr.mask_mode ==
2597 					     VMXNET3_IMM_AUTO;
2598 		devRead->intrConf.numIntrs = adapter->intr.num_intrs;
2599 		for (i = 0; i < adapter->intr.num_intrs; i++)
2600 			devRead->intrConf.modLevels[i] = adapter->intr.mod_levels[i];
2601 
2602 		devRead->intrConf.eventIntrIdx = adapter->intr.event_intr_idx;
2603 		devRead->intrConf.intrCtrl |= cpu_to_le32(VMXNET3_IC_DISABLE_ALL);
2604 	} else {
2605 		devReadExt->intrConfExt.autoMask = adapter->intr.mask_mode ==
2606 						   VMXNET3_IMM_AUTO;
2607 		devReadExt->intrConfExt.numIntrs = adapter->intr.num_intrs;
2608 		for (i = 0; i < adapter->intr.num_intrs; i++)
2609 			devReadExt->intrConfExt.modLevels[i] = adapter->intr.mod_levels[i];
2610 
2611 		devReadExt->intrConfExt.eventIntrIdx = adapter->intr.event_intr_idx;
2612 		devReadExt->intrConfExt.intrCtrl |= cpu_to_le32(VMXNET3_IC_DISABLE_ALL);
2613 	}
2614 
2615 	/* rx filter settings */
2616 	devRead->rxFilterConf.rxMode = 0;
2617 	vmxnet3_restore_vlan(adapter);
2618 	vmxnet3_write_mac_addr(adapter, adapter->netdev->dev_addr);
2619 
2620 	/* the rest are already zeroed */
2621 }
2622 
2623 static void
2624 vmxnet3_init_coalesce(struct vmxnet3_adapter *adapter)
2625 {
2626 	struct Vmxnet3_DriverShared *shared = adapter->shared;
2627 	union Vmxnet3_CmdInfo *cmdInfo = &shared->cu.cmdInfo;
2628 	unsigned long flags;
2629 
2630 	if (!VMXNET3_VERSION_GE_3(adapter))
2631 		return;
2632 
2633 	spin_lock_irqsave(&adapter->cmd_lock, flags);
2634 	cmdInfo->varConf.confVer = 1;
2635 	cmdInfo->varConf.confLen =
2636 		cpu_to_le32(sizeof(*adapter->coal_conf));
2637 	cmdInfo->varConf.confPA  = cpu_to_le64(adapter->coal_conf_pa);
2638 
2639 	if (adapter->default_coal_mode) {
2640 		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2641 				       VMXNET3_CMD_GET_COALESCE);
2642 	} else {
2643 		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2644 				       VMXNET3_CMD_SET_COALESCE);
2645 	}
2646 
2647 	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2648 }
2649 
2650 static void
2651 vmxnet3_init_rssfields(struct vmxnet3_adapter *adapter)
2652 {
2653 	struct Vmxnet3_DriverShared *shared = adapter->shared;
2654 	union Vmxnet3_CmdInfo *cmdInfo = &shared->cu.cmdInfo;
2655 	unsigned long flags;
2656 
2657 	if (!VMXNET3_VERSION_GE_4(adapter))
2658 		return;
2659 
2660 	spin_lock_irqsave(&adapter->cmd_lock, flags);
2661 
2662 	if (adapter->default_rss_fields) {
2663 		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2664 				       VMXNET3_CMD_GET_RSS_FIELDS);
2665 		adapter->rss_fields =
2666 			VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
2667 	} else {
2668 		cmdInfo->setRssFields = adapter->rss_fields;
2669 		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2670 				       VMXNET3_CMD_SET_RSS_FIELDS);
2671 		/* Not all requested RSS may get applied, so get and
2672 		 * cache what was actually applied.
2673 		 */
2674 		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2675 				       VMXNET3_CMD_GET_RSS_FIELDS);
2676 		adapter->rss_fields =
2677 			VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
2678 	}
2679 
2680 	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2681 }
2682 
2683 int
2684 vmxnet3_activate_dev(struct vmxnet3_adapter *adapter)
2685 {
2686 	int err, i;
2687 	u32 ret;
2688 	unsigned long flags;
2689 
2690 	netdev_dbg(adapter->netdev, "%s: skb_buf_size %d, rx_buf_per_pkt %d,"
2691 		" ring sizes %u %u %u\n", adapter->netdev->name,
2692 		adapter->skb_buf_size, adapter->rx_buf_per_pkt,
2693 		adapter->tx_queue[0].tx_ring.size,
2694 		adapter->rx_queue[0].rx_ring[0].size,
2695 		adapter->rx_queue[0].rx_ring[1].size);
2696 
2697 	vmxnet3_tq_init_all(adapter);
2698 	err = vmxnet3_rq_init_all(adapter);
2699 	if (err) {
2700 		netdev_err(adapter->netdev,
2701 			   "Failed to init rx queue error %d\n", err);
2702 		goto rq_err;
2703 	}
2704 
2705 	err = vmxnet3_request_irqs(adapter);
2706 	if (err) {
2707 		netdev_err(adapter->netdev,
2708 			   "Failed to setup irq for error %d\n", err);
2709 		goto irq_err;
2710 	}
2711 
2712 	vmxnet3_setup_driver_shared(adapter);
2713 
2714 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, VMXNET3_GET_ADDR_LO(
2715 			       adapter->shared_pa));
2716 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, VMXNET3_GET_ADDR_HI(
2717 			       adapter->shared_pa));
2718 	spin_lock_irqsave(&adapter->cmd_lock, flags);
2719 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2720 			       VMXNET3_CMD_ACTIVATE_DEV);
2721 	ret = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
2722 	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2723 
2724 	if (ret != 0) {
2725 		netdev_err(adapter->netdev,
2726 			   "Failed to activate dev: error %u\n", ret);
2727 		err = -EINVAL;
2728 		goto activate_err;
2729 	}
2730 
2731 	vmxnet3_init_coalesce(adapter);
2732 	vmxnet3_init_rssfields(adapter);
2733 
2734 	for (i = 0; i < adapter->num_rx_queues; i++) {
2735 		VMXNET3_WRITE_BAR0_REG(adapter,
2736 				VMXNET3_REG_RXPROD + i * VMXNET3_REG_ALIGN,
2737 				adapter->rx_queue[i].rx_ring[0].next2fill);
2738 		VMXNET3_WRITE_BAR0_REG(adapter, (VMXNET3_REG_RXPROD2 +
2739 				(i * VMXNET3_REG_ALIGN)),
2740 				adapter->rx_queue[i].rx_ring[1].next2fill);
2741 	}
2742 
2743 	/* Apply the rx filter settins last. */
2744 	vmxnet3_set_mc(adapter->netdev);
2745 
2746 	/*
2747 	 * Check link state when first activating device. It will start the
2748 	 * tx queue if the link is up.
2749 	 */
2750 	vmxnet3_check_link(adapter, true);
2751 	netif_tx_wake_all_queues(adapter->netdev);
2752 	for (i = 0; i < adapter->num_rx_queues; i++)
2753 		napi_enable(&adapter->rx_queue[i].napi);
2754 	vmxnet3_enable_all_intrs(adapter);
2755 	clear_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
2756 	return 0;
2757 
2758 activate_err:
2759 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, 0);
2760 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, 0);
2761 	vmxnet3_free_irqs(adapter);
2762 irq_err:
2763 rq_err:
2764 	/* free up buffers we allocated */
2765 	vmxnet3_rq_cleanup_all(adapter);
2766 	return err;
2767 }
2768 
2769 
2770 void
2771 vmxnet3_reset_dev(struct vmxnet3_adapter *adapter)
2772 {
2773 	unsigned long flags;
2774 	spin_lock_irqsave(&adapter->cmd_lock, flags);
2775 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, VMXNET3_CMD_RESET_DEV);
2776 	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2777 }
2778 
2779 
2780 int
2781 vmxnet3_quiesce_dev(struct vmxnet3_adapter *adapter)
2782 {
2783 	int i;
2784 	unsigned long flags;
2785 	if (test_and_set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state))
2786 		return 0;
2787 
2788 
2789 	spin_lock_irqsave(&adapter->cmd_lock, flags);
2790 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2791 			       VMXNET3_CMD_QUIESCE_DEV);
2792 	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2793 	vmxnet3_disable_all_intrs(adapter);
2794 
2795 	for (i = 0; i < adapter->num_rx_queues; i++)
2796 		napi_disable(&adapter->rx_queue[i].napi);
2797 	netif_tx_disable(adapter->netdev);
2798 	adapter->link_speed = 0;
2799 	netif_carrier_off(adapter->netdev);
2800 
2801 	vmxnet3_tq_cleanup_all(adapter);
2802 	vmxnet3_rq_cleanup_all(adapter);
2803 	vmxnet3_free_irqs(adapter);
2804 	return 0;
2805 }
2806 
2807 
2808 static void
2809 vmxnet3_write_mac_addr(struct vmxnet3_adapter *adapter, const u8 *mac)
2810 {
2811 	u32 tmp;
2812 
2813 	tmp = *(u32 *)mac;
2814 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACL, tmp);
2815 
2816 	tmp = (mac[5] << 8) | mac[4];
2817 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACH, tmp);
2818 }
2819 
2820 
2821 static int
2822 vmxnet3_set_mac_addr(struct net_device *netdev, void *p)
2823 {
2824 	struct sockaddr *addr = p;
2825 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2826 
2827 	dev_addr_set(netdev, addr->sa_data);
2828 	vmxnet3_write_mac_addr(adapter, addr->sa_data);
2829 
2830 	return 0;
2831 }
2832 
2833 
2834 /* ==================== initialization and cleanup routines ============ */
2835 
2836 static int
2837 vmxnet3_alloc_pci_resources(struct vmxnet3_adapter *adapter)
2838 {
2839 	int err;
2840 	unsigned long mmio_start, mmio_len;
2841 	struct pci_dev *pdev = adapter->pdev;
2842 
2843 	err = pci_enable_device(pdev);
2844 	if (err) {
2845 		dev_err(&pdev->dev, "Failed to enable adapter: error %d\n", err);
2846 		return err;
2847 	}
2848 
2849 	err = pci_request_selected_regions(pdev, (1 << 2) - 1,
2850 					   vmxnet3_driver_name);
2851 	if (err) {
2852 		dev_err(&pdev->dev,
2853 			"Failed to request region for adapter: error %d\n", err);
2854 		goto err_enable_device;
2855 	}
2856 
2857 	pci_set_master(pdev);
2858 
2859 	mmio_start = pci_resource_start(pdev, 0);
2860 	mmio_len = pci_resource_len(pdev, 0);
2861 	adapter->hw_addr0 = ioremap(mmio_start, mmio_len);
2862 	if (!adapter->hw_addr0) {
2863 		dev_err(&pdev->dev, "Failed to map bar0\n");
2864 		err = -EIO;
2865 		goto err_ioremap;
2866 	}
2867 
2868 	mmio_start = pci_resource_start(pdev, 1);
2869 	mmio_len = pci_resource_len(pdev, 1);
2870 	adapter->hw_addr1 = ioremap(mmio_start, mmio_len);
2871 	if (!adapter->hw_addr1) {
2872 		dev_err(&pdev->dev, "Failed to map bar1\n");
2873 		err = -EIO;
2874 		goto err_bar1;
2875 	}
2876 	return 0;
2877 
2878 err_bar1:
2879 	iounmap(adapter->hw_addr0);
2880 err_ioremap:
2881 	pci_release_selected_regions(pdev, (1 << 2) - 1);
2882 err_enable_device:
2883 	pci_disable_device(pdev);
2884 	return err;
2885 }
2886 
2887 
2888 static void
2889 vmxnet3_free_pci_resources(struct vmxnet3_adapter *adapter)
2890 {
2891 	BUG_ON(!adapter->pdev);
2892 
2893 	iounmap(adapter->hw_addr0);
2894 	iounmap(adapter->hw_addr1);
2895 	pci_release_selected_regions(adapter->pdev, (1 << 2) - 1);
2896 	pci_disable_device(adapter->pdev);
2897 }
2898 
2899 
2900 static void
2901 vmxnet3_adjust_rx_ring_size(struct vmxnet3_adapter *adapter)
2902 {
2903 	size_t sz, i, ring0_size, ring1_size, comp_size;
2904 	if (adapter->netdev->mtu <= VMXNET3_MAX_SKB_BUF_SIZE -
2905 				    VMXNET3_MAX_ETH_HDR_SIZE) {
2906 		adapter->skb_buf_size = adapter->netdev->mtu +
2907 					VMXNET3_MAX_ETH_HDR_SIZE;
2908 		if (adapter->skb_buf_size < VMXNET3_MIN_T0_BUF_SIZE)
2909 			adapter->skb_buf_size = VMXNET3_MIN_T0_BUF_SIZE;
2910 
2911 		adapter->rx_buf_per_pkt = 1;
2912 	} else {
2913 		adapter->skb_buf_size = VMXNET3_MAX_SKB_BUF_SIZE;
2914 		sz = adapter->netdev->mtu - VMXNET3_MAX_SKB_BUF_SIZE +
2915 					    VMXNET3_MAX_ETH_HDR_SIZE;
2916 		adapter->rx_buf_per_pkt = 1 + (sz + PAGE_SIZE - 1) / PAGE_SIZE;
2917 	}
2918 
2919 	/*
2920 	 * for simplicity, force the ring0 size to be a multiple of
2921 	 * rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN
2922 	 */
2923 	sz = adapter->rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN;
2924 	ring0_size = adapter->rx_queue[0].rx_ring[0].size;
2925 	ring0_size = (ring0_size + sz - 1) / sz * sz;
2926 	ring0_size = min_t(u32, ring0_size, VMXNET3_RX_RING_MAX_SIZE /
2927 			   sz * sz);
2928 	ring1_size = adapter->rx_queue[0].rx_ring[1].size;
2929 	ring1_size = (ring1_size + sz - 1) / sz * sz;
2930 	ring1_size = min_t(u32, ring1_size, VMXNET3_RX_RING2_MAX_SIZE /
2931 			   sz * sz);
2932 	comp_size = ring0_size + ring1_size;
2933 
2934 	for (i = 0; i < adapter->num_rx_queues; i++) {
2935 		struct vmxnet3_rx_queue	*rq = &adapter->rx_queue[i];
2936 
2937 		rq->rx_ring[0].size = ring0_size;
2938 		rq->rx_ring[1].size = ring1_size;
2939 		rq->comp_ring.size = comp_size;
2940 	}
2941 }
2942 
2943 
2944 int
2945 vmxnet3_create_queues(struct vmxnet3_adapter *adapter, u32 tx_ring_size,
2946 		      u32 rx_ring_size, u32 rx_ring2_size,
2947 		      u16 txdata_desc_size, u16 rxdata_desc_size)
2948 {
2949 	int err = 0, i;
2950 
2951 	for (i = 0; i < adapter->num_tx_queues; i++) {
2952 		struct vmxnet3_tx_queue	*tq = &adapter->tx_queue[i];
2953 		tq->tx_ring.size   = tx_ring_size;
2954 		tq->data_ring.size = tx_ring_size;
2955 		tq->comp_ring.size = tx_ring_size;
2956 		tq->txdata_desc_size = txdata_desc_size;
2957 		tq->shared = &adapter->tqd_start[i].ctrl;
2958 		tq->stopped = true;
2959 		tq->adapter = adapter;
2960 		tq->qid = i;
2961 		err = vmxnet3_tq_create(tq, adapter);
2962 		/*
2963 		 * Too late to change num_tx_queues. We cannot do away with
2964 		 * lesser number of queues than what we asked for
2965 		 */
2966 		if (err)
2967 			goto queue_err;
2968 	}
2969 
2970 	adapter->rx_queue[0].rx_ring[0].size = rx_ring_size;
2971 	adapter->rx_queue[0].rx_ring[1].size = rx_ring2_size;
2972 	vmxnet3_adjust_rx_ring_size(adapter);
2973 
2974 	adapter->rxdataring_enabled = VMXNET3_VERSION_GE_3(adapter);
2975 	for (i = 0; i < adapter->num_rx_queues; i++) {
2976 		struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
2977 		/* qid and qid2 for rx queues will be assigned later when num
2978 		 * of rx queues is finalized after allocating intrs */
2979 		rq->shared = &adapter->rqd_start[i].ctrl;
2980 		rq->adapter = adapter;
2981 		rq->data_ring.desc_size = rxdata_desc_size;
2982 		err = vmxnet3_rq_create(rq, adapter);
2983 		if (err) {
2984 			if (i == 0) {
2985 				netdev_err(adapter->netdev,
2986 					   "Could not allocate any rx queues. "
2987 					   "Aborting.\n");
2988 				goto queue_err;
2989 			} else {
2990 				netdev_info(adapter->netdev,
2991 					    "Number of rx queues changed "
2992 					    "to : %d.\n", i);
2993 				adapter->num_rx_queues = i;
2994 				err = 0;
2995 				break;
2996 			}
2997 		}
2998 	}
2999 
3000 	if (!adapter->rxdataring_enabled)
3001 		vmxnet3_rq_destroy_all_rxdataring(adapter);
3002 
3003 	return err;
3004 queue_err:
3005 	vmxnet3_tq_destroy_all(adapter);
3006 	return err;
3007 }
3008 
3009 static int
3010 vmxnet3_open(struct net_device *netdev)
3011 {
3012 	struct vmxnet3_adapter *adapter;
3013 	int err, i;
3014 
3015 	adapter = netdev_priv(netdev);
3016 
3017 	for (i = 0; i < adapter->num_tx_queues; i++)
3018 		spin_lock_init(&adapter->tx_queue[i].tx_lock);
3019 
3020 	if (VMXNET3_VERSION_GE_3(adapter)) {
3021 		unsigned long flags;
3022 		u16 txdata_desc_size;
3023 
3024 		spin_lock_irqsave(&adapter->cmd_lock, flags);
3025 		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3026 				       VMXNET3_CMD_GET_TXDATA_DESC_SIZE);
3027 		txdata_desc_size = VMXNET3_READ_BAR1_REG(adapter,
3028 							 VMXNET3_REG_CMD);
3029 		spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3030 
3031 		if ((txdata_desc_size < VMXNET3_TXDATA_DESC_MIN_SIZE) ||
3032 		    (txdata_desc_size > VMXNET3_TXDATA_DESC_MAX_SIZE) ||
3033 		    (txdata_desc_size & VMXNET3_TXDATA_DESC_SIZE_MASK)) {
3034 			adapter->txdata_desc_size =
3035 				sizeof(struct Vmxnet3_TxDataDesc);
3036 		} else {
3037 			adapter->txdata_desc_size = txdata_desc_size;
3038 		}
3039 	} else {
3040 		adapter->txdata_desc_size = sizeof(struct Vmxnet3_TxDataDesc);
3041 	}
3042 
3043 	err = vmxnet3_create_queues(adapter,
3044 				    adapter->tx_ring_size,
3045 				    adapter->rx_ring_size,
3046 				    adapter->rx_ring2_size,
3047 				    adapter->txdata_desc_size,
3048 				    adapter->rxdata_desc_size);
3049 	if (err)
3050 		goto queue_err;
3051 
3052 	err = vmxnet3_activate_dev(adapter);
3053 	if (err)
3054 		goto activate_err;
3055 
3056 	return 0;
3057 
3058 activate_err:
3059 	vmxnet3_rq_destroy_all(adapter);
3060 	vmxnet3_tq_destroy_all(adapter);
3061 queue_err:
3062 	return err;
3063 }
3064 
3065 
3066 static int
3067 vmxnet3_close(struct net_device *netdev)
3068 {
3069 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3070 
3071 	/*
3072 	 * Reset_work may be in the middle of resetting the device, wait for its
3073 	 * completion.
3074 	 */
3075 	while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
3076 		usleep_range(1000, 2000);
3077 
3078 	vmxnet3_quiesce_dev(adapter);
3079 
3080 	vmxnet3_rq_destroy_all(adapter);
3081 	vmxnet3_tq_destroy_all(adapter);
3082 
3083 	clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
3084 
3085 
3086 	return 0;
3087 }
3088 
3089 
3090 void
3091 vmxnet3_force_close(struct vmxnet3_adapter *adapter)
3092 {
3093 	int i;
3094 
3095 	/*
3096 	 * we must clear VMXNET3_STATE_BIT_RESETTING, otherwise
3097 	 * vmxnet3_close() will deadlock.
3098 	 */
3099 	BUG_ON(test_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state));
3100 
3101 	/* we need to enable NAPI, otherwise dev_close will deadlock */
3102 	for (i = 0; i < adapter->num_rx_queues; i++)
3103 		napi_enable(&adapter->rx_queue[i].napi);
3104 	/*
3105 	 * Need to clear the quiesce bit to ensure that vmxnet3_close
3106 	 * can quiesce the device properly
3107 	 */
3108 	clear_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
3109 	dev_close(adapter->netdev);
3110 }
3111 
3112 
3113 static int
3114 vmxnet3_change_mtu(struct net_device *netdev, int new_mtu)
3115 {
3116 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3117 	int err = 0;
3118 
3119 	netdev->mtu = new_mtu;
3120 
3121 	/*
3122 	 * Reset_work may be in the middle of resetting the device, wait for its
3123 	 * completion.
3124 	 */
3125 	while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
3126 		usleep_range(1000, 2000);
3127 
3128 	if (netif_running(netdev)) {
3129 		vmxnet3_quiesce_dev(adapter);
3130 		vmxnet3_reset_dev(adapter);
3131 
3132 		/* we need to re-create the rx queue based on the new mtu */
3133 		vmxnet3_rq_destroy_all(adapter);
3134 		vmxnet3_adjust_rx_ring_size(adapter);
3135 		err = vmxnet3_rq_create_all(adapter);
3136 		if (err) {
3137 			netdev_err(netdev,
3138 				   "failed to re-create rx queues, "
3139 				   " error %d. Closing it.\n", err);
3140 			goto out;
3141 		}
3142 
3143 		err = vmxnet3_activate_dev(adapter);
3144 		if (err) {
3145 			netdev_err(netdev,
3146 				   "failed to re-activate, error %d. "
3147 				   "Closing it\n", err);
3148 			goto out;
3149 		}
3150 	}
3151 
3152 out:
3153 	clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
3154 	if (err)
3155 		vmxnet3_force_close(adapter);
3156 
3157 	return err;
3158 }
3159 
3160 
3161 static void
3162 vmxnet3_declare_features(struct vmxnet3_adapter *adapter)
3163 {
3164 	struct net_device *netdev = adapter->netdev;
3165 
3166 	netdev->hw_features = NETIF_F_SG | NETIF_F_RXCSUM |
3167 		NETIF_F_HW_CSUM | NETIF_F_HW_VLAN_CTAG_TX |
3168 		NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_TSO | NETIF_F_TSO6 |
3169 		NETIF_F_LRO | NETIF_F_HIGHDMA;
3170 
3171 	if (VMXNET3_VERSION_GE_4(adapter)) {
3172 		netdev->hw_features |= NETIF_F_GSO_UDP_TUNNEL |
3173 				NETIF_F_GSO_UDP_TUNNEL_CSUM;
3174 
3175 		netdev->hw_enc_features = NETIF_F_SG | NETIF_F_RXCSUM |
3176 			NETIF_F_HW_CSUM | NETIF_F_HW_VLAN_CTAG_TX |
3177 			NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_TSO | NETIF_F_TSO6 |
3178 			NETIF_F_LRO | NETIF_F_GSO_UDP_TUNNEL |
3179 			NETIF_F_GSO_UDP_TUNNEL_CSUM;
3180 	}
3181 
3182 	netdev->vlan_features = netdev->hw_features &
3183 				~(NETIF_F_HW_VLAN_CTAG_TX |
3184 				  NETIF_F_HW_VLAN_CTAG_RX);
3185 	netdev->features = netdev->hw_features | NETIF_F_HW_VLAN_CTAG_FILTER;
3186 }
3187 
3188 
3189 static void
3190 vmxnet3_read_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac)
3191 {
3192 	u32 tmp;
3193 
3194 	tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACL);
3195 	*(u32 *)mac = tmp;
3196 
3197 	tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACH);
3198 	mac[4] = tmp & 0xff;
3199 	mac[5] = (tmp >> 8) & 0xff;
3200 }
3201 
3202 #ifdef CONFIG_PCI_MSI
3203 
3204 /*
3205  * Enable MSIx vectors.
3206  * Returns :
3207  *	VMXNET3_LINUX_MIN_MSIX_VECT when only minimum number of vectors required
3208  *	 were enabled.
3209  *	number of vectors which were enabled otherwise (this number is greater
3210  *	 than VMXNET3_LINUX_MIN_MSIX_VECT)
3211  */
3212 
3213 static int
3214 vmxnet3_acquire_msix_vectors(struct vmxnet3_adapter *adapter, int nvec)
3215 {
3216 	int ret = pci_enable_msix_range(adapter->pdev,
3217 					adapter->intr.msix_entries, nvec, nvec);
3218 
3219 	if (ret == -ENOSPC && nvec > VMXNET3_LINUX_MIN_MSIX_VECT) {
3220 		dev_err(&adapter->netdev->dev,
3221 			"Failed to enable %d MSI-X, trying %d\n",
3222 			nvec, VMXNET3_LINUX_MIN_MSIX_VECT);
3223 
3224 		ret = pci_enable_msix_range(adapter->pdev,
3225 					    adapter->intr.msix_entries,
3226 					    VMXNET3_LINUX_MIN_MSIX_VECT,
3227 					    VMXNET3_LINUX_MIN_MSIX_VECT);
3228 	}
3229 
3230 	if (ret < 0) {
3231 		dev_err(&adapter->netdev->dev,
3232 			"Failed to enable MSI-X, error: %d\n", ret);
3233 	}
3234 
3235 	return ret;
3236 }
3237 
3238 
3239 #endif /* CONFIG_PCI_MSI */
3240 
3241 static void
3242 vmxnet3_alloc_intr_resources(struct vmxnet3_adapter *adapter)
3243 {
3244 	u32 cfg;
3245 	unsigned long flags;
3246 
3247 	/* intr settings */
3248 	spin_lock_irqsave(&adapter->cmd_lock, flags);
3249 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3250 			       VMXNET3_CMD_GET_CONF_INTR);
3251 	cfg = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
3252 	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3253 	adapter->intr.type = cfg & 0x3;
3254 	adapter->intr.mask_mode = (cfg >> 2) & 0x3;
3255 
3256 	if (adapter->intr.type == VMXNET3_IT_AUTO) {
3257 		adapter->intr.type = VMXNET3_IT_MSIX;
3258 	}
3259 
3260 #ifdef CONFIG_PCI_MSI
3261 	if (adapter->intr.type == VMXNET3_IT_MSIX) {
3262 		int i, nvec, nvec_allocated;
3263 
3264 		nvec  = adapter->share_intr == VMXNET3_INTR_TXSHARE ?
3265 			1 : adapter->num_tx_queues;
3266 		nvec += adapter->share_intr == VMXNET3_INTR_BUDDYSHARE ?
3267 			0 : adapter->num_rx_queues;
3268 		nvec += 1;	/* for link event */
3269 		nvec = nvec > VMXNET3_LINUX_MIN_MSIX_VECT ?
3270 		       nvec : VMXNET3_LINUX_MIN_MSIX_VECT;
3271 
3272 		for (i = 0; i < nvec; i++)
3273 			adapter->intr.msix_entries[i].entry = i;
3274 
3275 		nvec_allocated = vmxnet3_acquire_msix_vectors(adapter, nvec);
3276 		if (nvec_allocated < 0)
3277 			goto msix_err;
3278 
3279 		/* If we cannot allocate one MSIx vector per queue
3280 		 * then limit the number of rx queues to 1
3281 		 */
3282 		if (nvec_allocated == VMXNET3_LINUX_MIN_MSIX_VECT &&
3283 		    nvec != VMXNET3_LINUX_MIN_MSIX_VECT) {
3284 			if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE
3285 			    || adapter->num_rx_queues != 1) {
3286 				adapter->share_intr = VMXNET3_INTR_TXSHARE;
3287 				netdev_err(adapter->netdev,
3288 					   "Number of rx queues : 1\n");
3289 				adapter->num_rx_queues = 1;
3290 			}
3291 		}
3292 
3293 		adapter->intr.num_intrs = nvec_allocated;
3294 		return;
3295 
3296 msix_err:
3297 		/* If we cannot allocate MSIx vectors use only one rx queue */
3298 		dev_info(&adapter->pdev->dev,
3299 			 "Failed to enable MSI-X, error %d. "
3300 			 "Limiting #rx queues to 1, try MSI.\n", nvec_allocated);
3301 
3302 		adapter->intr.type = VMXNET3_IT_MSI;
3303 	}
3304 
3305 	if (adapter->intr.type == VMXNET3_IT_MSI) {
3306 		if (!pci_enable_msi(adapter->pdev)) {
3307 			adapter->num_rx_queues = 1;
3308 			adapter->intr.num_intrs = 1;
3309 			return;
3310 		}
3311 	}
3312 #endif /* CONFIG_PCI_MSI */
3313 
3314 	adapter->num_rx_queues = 1;
3315 	dev_info(&adapter->netdev->dev,
3316 		 "Using INTx interrupt, #Rx queues: 1.\n");
3317 	adapter->intr.type = VMXNET3_IT_INTX;
3318 
3319 	/* INT-X related setting */
3320 	adapter->intr.num_intrs = 1;
3321 }
3322 
3323 
3324 static void
3325 vmxnet3_free_intr_resources(struct vmxnet3_adapter *adapter)
3326 {
3327 	if (adapter->intr.type == VMXNET3_IT_MSIX)
3328 		pci_disable_msix(adapter->pdev);
3329 	else if (adapter->intr.type == VMXNET3_IT_MSI)
3330 		pci_disable_msi(adapter->pdev);
3331 	else
3332 		BUG_ON(adapter->intr.type != VMXNET3_IT_INTX);
3333 }
3334 
3335 
3336 static void
3337 vmxnet3_tx_timeout(struct net_device *netdev, unsigned int txqueue)
3338 {
3339 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3340 	adapter->tx_timeout_count++;
3341 
3342 	netdev_err(adapter->netdev, "tx hang\n");
3343 	schedule_work(&adapter->work);
3344 }
3345 
3346 
3347 static void
3348 vmxnet3_reset_work(struct work_struct *data)
3349 {
3350 	struct vmxnet3_adapter *adapter;
3351 
3352 	adapter = container_of(data, struct vmxnet3_adapter, work);
3353 
3354 	/* if another thread is resetting the device, no need to proceed */
3355 	if (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
3356 		return;
3357 
3358 	/* if the device is closed, we must leave it alone */
3359 	rtnl_lock();
3360 	if (netif_running(adapter->netdev)) {
3361 		netdev_notice(adapter->netdev, "resetting\n");
3362 		vmxnet3_quiesce_dev(adapter);
3363 		vmxnet3_reset_dev(adapter);
3364 		vmxnet3_activate_dev(adapter);
3365 	} else {
3366 		netdev_info(adapter->netdev, "already closed\n");
3367 	}
3368 	rtnl_unlock();
3369 
3370 	netif_wake_queue(adapter->netdev);
3371 	clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
3372 }
3373 
3374 
3375 static int
3376 vmxnet3_probe_device(struct pci_dev *pdev,
3377 		     const struct pci_device_id *id)
3378 {
3379 	static const struct net_device_ops vmxnet3_netdev_ops = {
3380 		.ndo_open = vmxnet3_open,
3381 		.ndo_stop = vmxnet3_close,
3382 		.ndo_start_xmit = vmxnet3_xmit_frame,
3383 		.ndo_set_mac_address = vmxnet3_set_mac_addr,
3384 		.ndo_change_mtu = vmxnet3_change_mtu,
3385 		.ndo_fix_features = vmxnet3_fix_features,
3386 		.ndo_set_features = vmxnet3_set_features,
3387 		.ndo_features_check = vmxnet3_features_check,
3388 		.ndo_get_stats64 = vmxnet3_get_stats64,
3389 		.ndo_tx_timeout = vmxnet3_tx_timeout,
3390 		.ndo_set_rx_mode = vmxnet3_set_mc,
3391 		.ndo_vlan_rx_add_vid = vmxnet3_vlan_rx_add_vid,
3392 		.ndo_vlan_rx_kill_vid = vmxnet3_vlan_rx_kill_vid,
3393 #ifdef CONFIG_NET_POLL_CONTROLLER
3394 		.ndo_poll_controller = vmxnet3_netpoll,
3395 #endif
3396 	};
3397 	int err;
3398 	u32 ver;
3399 	struct net_device *netdev;
3400 	struct vmxnet3_adapter *adapter;
3401 	u8 mac[ETH_ALEN];
3402 	int size;
3403 	int num_tx_queues;
3404 	int num_rx_queues;
3405 	int queues;
3406 	unsigned long flags;
3407 
3408 	if (!pci_msi_enabled())
3409 		enable_mq = 0;
3410 
3411 #ifdef VMXNET3_RSS
3412 	if (enable_mq)
3413 		num_rx_queues = min(VMXNET3_DEVICE_MAX_RX_QUEUES,
3414 				    (int)num_online_cpus());
3415 	else
3416 #endif
3417 		num_rx_queues = 1;
3418 
3419 	if (enable_mq)
3420 		num_tx_queues = min(VMXNET3_DEVICE_MAX_TX_QUEUES,
3421 				    (int)num_online_cpus());
3422 	else
3423 		num_tx_queues = 1;
3424 
3425 	netdev = alloc_etherdev_mq(sizeof(struct vmxnet3_adapter),
3426 				   max(num_tx_queues, num_rx_queues));
3427 	if (!netdev)
3428 		return -ENOMEM;
3429 
3430 	pci_set_drvdata(pdev, netdev);
3431 	adapter = netdev_priv(netdev);
3432 	adapter->netdev = netdev;
3433 	adapter->pdev = pdev;
3434 
3435 	adapter->tx_ring_size = VMXNET3_DEF_TX_RING_SIZE;
3436 	adapter->rx_ring_size = VMXNET3_DEF_RX_RING_SIZE;
3437 	adapter->rx_ring2_size = VMXNET3_DEF_RX_RING2_SIZE;
3438 
3439 	err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
3440 	if (err) {
3441 		dev_err(&pdev->dev, "dma_set_mask failed\n");
3442 		goto err_set_mask;
3443 	}
3444 
3445 	spin_lock_init(&adapter->cmd_lock);
3446 	adapter->adapter_pa = dma_map_single(&adapter->pdev->dev, adapter,
3447 					     sizeof(struct vmxnet3_adapter),
3448 					     DMA_TO_DEVICE);
3449 	if (dma_mapping_error(&adapter->pdev->dev, adapter->adapter_pa)) {
3450 		dev_err(&pdev->dev, "Failed to map dma\n");
3451 		err = -EFAULT;
3452 		goto err_set_mask;
3453 	}
3454 	adapter->shared = dma_alloc_coherent(
3455 				&adapter->pdev->dev,
3456 				sizeof(struct Vmxnet3_DriverShared),
3457 				&adapter->shared_pa, GFP_KERNEL);
3458 	if (!adapter->shared) {
3459 		dev_err(&pdev->dev, "Failed to allocate memory\n");
3460 		err = -ENOMEM;
3461 		goto err_alloc_shared;
3462 	}
3463 
3464 	err = vmxnet3_alloc_pci_resources(adapter);
3465 	if (err < 0)
3466 		goto err_alloc_pci;
3467 
3468 	ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_VRRS);
3469 	if (ver & (1 << VMXNET3_REV_6)) {
3470 		VMXNET3_WRITE_BAR1_REG(adapter,
3471 				       VMXNET3_REG_VRRS,
3472 				       1 << VMXNET3_REV_6);
3473 		adapter->version = VMXNET3_REV_6 + 1;
3474 	} else if (ver & (1 << VMXNET3_REV_5)) {
3475 		VMXNET3_WRITE_BAR1_REG(adapter,
3476 				       VMXNET3_REG_VRRS,
3477 				       1 << VMXNET3_REV_5);
3478 		adapter->version = VMXNET3_REV_5 + 1;
3479 	} else if (ver & (1 << VMXNET3_REV_4)) {
3480 		VMXNET3_WRITE_BAR1_REG(adapter,
3481 				       VMXNET3_REG_VRRS,
3482 				       1 << VMXNET3_REV_4);
3483 		adapter->version = VMXNET3_REV_4 + 1;
3484 	} else if (ver & (1 << VMXNET3_REV_3)) {
3485 		VMXNET3_WRITE_BAR1_REG(adapter,
3486 				       VMXNET3_REG_VRRS,
3487 				       1 << VMXNET3_REV_3);
3488 		adapter->version = VMXNET3_REV_3 + 1;
3489 	} else if (ver & (1 << VMXNET3_REV_2)) {
3490 		VMXNET3_WRITE_BAR1_REG(adapter,
3491 				       VMXNET3_REG_VRRS,
3492 				       1 << VMXNET3_REV_2);
3493 		adapter->version = VMXNET3_REV_2 + 1;
3494 	} else if (ver & (1 << VMXNET3_REV_1)) {
3495 		VMXNET3_WRITE_BAR1_REG(adapter,
3496 				       VMXNET3_REG_VRRS,
3497 				       1 << VMXNET3_REV_1);
3498 		adapter->version = VMXNET3_REV_1 + 1;
3499 	} else {
3500 		dev_err(&pdev->dev,
3501 			"Incompatible h/w version (0x%x) for adapter\n", ver);
3502 		err = -EBUSY;
3503 		goto err_ver;
3504 	}
3505 	dev_dbg(&pdev->dev, "Using device version %d\n", adapter->version);
3506 
3507 	ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_UVRS);
3508 	if (ver & 1) {
3509 		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_UVRS, 1);
3510 	} else {
3511 		dev_err(&pdev->dev,
3512 			"Incompatible upt version (0x%x) for adapter\n", ver);
3513 		err = -EBUSY;
3514 		goto err_ver;
3515 	}
3516 
3517 	if (VMXNET3_VERSION_GE_6(adapter)) {
3518 		spin_lock_irqsave(&adapter->cmd_lock, flags);
3519 		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3520 				       VMXNET3_CMD_GET_MAX_QUEUES_CONF);
3521 		queues = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
3522 		spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3523 		if (queues > 0) {
3524 			adapter->num_rx_queues = min(num_rx_queues, ((queues >> 8) & 0xff));
3525 			adapter->num_tx_queues = min(num_tx_queues, (queues & 0xff));
3526 		} else {
3527 			adapter->num_rx_queues = min(num_rx_queues,
3528 						     VMXNET3_DEVICE_DEFAULT_RX_QUEUES);
3529 			adapter->num_tx_queues = min(num_tx_queues,
3530 						     VMXNET3_DEVICE_DEFAULT_TX_QUEUES);
3531 		}
3532 		if (adapter->num_rx_queues > VMXNET3_MAX_RX_QUEUES ||
3533 		    adapter->num_tx_queues > VMXNET3_MAX_TX_QUEUES) {
3534 			adapter->queuesExtEnabled = true;
3535 		} else {
3536 			adapter->queuesExtEnabled = false;
3537 		}
3538 	} else {
3539 		adapter->queuesExtEnabled = false;
3540 		num_rx_queues = rounddown_pow_of_two(num_rx_queues);
3541 		num_tx_queues = rounddown_pow_of_two(num_tx_queues);
3542 		adapter->num_rx_queues = min(num_rx_queues,
3543 					     VMXNET3_DEVICE_DEFAULT_RX_QUEUES);
3544 		adapter->num_tx_queues = min(num_tx_queues,
3545 					     VMXNET3_DEVICE_DEFAULT_TX_QUEUES);
3546 	}
3547 	dev_info(&pdev->dev,
3548 		 "# of Tx queues : %d, # of Rx queues : %d\n",
3549 		 adapter->num_tx_queues, adapter->num_rx_queues);
3550 
3551 	adapter->rx_buf_per_pkt = 1;
3552 
3553 	size = sizeof(struct Vmxnet3_TxQueueDesc) * adapter->num_tx_queues;
3554 	size += sizeof(struct Vmxnet3_RxQueueDesc) * adapter->num_rx_queues;
3555 	adapter->tqd_start = dma_alloc_coherent(&adapter->pdev->dev, size,
3556 						&adapter->queue_desc_pa,
3557 						GFP_KERNEL);
3558 
3559 	if (!adapter->tqd_start) {
3560 		dev_err(&pdev->dev, "Failed to allocate memory\n");
3561 		err = -ENOMEM;
3562 		goto err_ver;
3563 	}
3564 	adapter->rqd_start = (struct Vmxnet3_RxQueueDesc *)(adapter->tqd_start +
3565 							    adapter->num_tx_queues);
3566 
3567 	adapter->pm_conf = dma_alloc_coherent(&adapter->pdev->dev,
3568 					      sizeof(struct Vmxnet3_PMConf),
3569 					      &adapter->pm_conf_pa,
3570 					      GFP_KERNEL);
3571 	if (adapter->pm_conf == NULL) {
3572 		err = -ENOMEM;
3573 		goto err_alloc_pm;
3574 	}
3575 
3576 #ifdef VMXNET3_RSS
3577 
3578 	adapter->rss_conf = dma_alloc_coherent(&adapter->pdev->dev,
3579 					       sizeof(struct UPT1_RSSConf),
3580 					       &adapter->rss_conf_pa,
3581 					       GFP_KERNEL);
3582 	if (adapter->rss_conf == NULL) {
3583 		err = -ENOMEM;
3584 		goto err_alloc_rss;
3585 	}
3586 #endif /* VMXNET3_RSS */
3587 
3588 	if (VMXNET3_VERSION_GE_3(adapter)) {
3589 		adapter->coal_conf =
3590 			dma_alloc_coherent(&adapter->pdev->dev,
3591 					   sizeof(struct Vmxnet3_CoalesceScheme)
3592 					   ,
3593 					   &adapter->coal_conf_pa,
3594 					   GFP_KERNEL);
3595 		if (!adapter->coal_conf) {
3596 			err = -ENOMEM;
3597 			goto err_coal_conf;
3598 		}
3599 		adapter->coal_conf->coalMode = VMXNET3_COALESCE_DISABLED;
3600 		adapter->default_coal_mode = true;
3601 	}
3602 
3603 	if (VMXNET3_VERSION_GE_4(adapter)) {
3604 		adapter->default_rss_fields = true;
3605 		adapter->rss_fields = VMXNET3_RSS_FIELDS_DEFAULT;
3606 	}
3607 
3608 	SET_NETDEV_DEV(netdev, &pdev->dev);
3609 	vmxnet3_declare_features(adapter);
3610 
3611 	adapter->rxdata_desc_size = VMXNET3_VERSION_GE_3(adapter) ?
3612 		VMXNET3_DEF_RXDATA_DESC_SIZE : 0;
3613 
3614 	if (adapter->num_tx_queues == adapter->num_rx_queues)
3615 		adapter->share_intr = VMXNET3_INTR_BUDDYSHARE;
3616 	else
3617 		adapter->share_intr = VMXNET3_INTR_DONTSHARE;
3618 
3619 	vmxnet3_alloc_intr_resources(adapter);
3620 
3621 #ifdef VMXNET3_RSS
3622 	if (adapter->num_rx_queues > 1 &&
3623 	    adapter->intr.type == VMXNET3_IT_MSIX) {
3624 		adapter->rss = true;
3625 		netdev->hw_features |= NETIF_F_RXHASH;
3626 		netdev->features |= NETIF_F_RXHASH;
3627 		dev_dbg(&pdev->dev, "RSS is enabled.\n");
3628 	} else {
3629 		adapter->rss = false;
3630 	}
3631 #endif
3632 
3633 	vmxnet3_read_mac_addr(adapter, mac);
3634 	dev_addr_set(netdev, mac);
3635 
3636 	netdev->netdev_ops = &vmxnet3_netdev_ops;
3637 	vmxnet3_set_ethtool_ops(netdev);
3638 	netdev->watchdog_timeo = 5 * HZ;
3639 
3640 	/* MTU range: 60 - 9190 */
3641 	netdev->min_mtu = VMXNET3_MIN_MTU;
3642 	if (VMXNET3_VERSION_GE_6(adapter))
3643 		netdev->max_mtu = VMXNET3_V6_MAX_MTU;
3644 	else
3645 		netdev->max_mtu = VMXNET3_MAX_MTU;
3646 
3647 	INIT_WORK(&adapter->work, vmxnet3_reset_work);
3648 	set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
3649 
3650 	if (adapter->intr.type == VMXNET3_IT_MSIX) {
3651 		int i;
3652 		for (i = 0; i < adapter->num_rx_queues; i++) {
3653 			netif_napi_add(adapter->netdev,
3654 				       &adapter->rx_queue[i].napi,
3655 				       vmxnet3_poll_rx_only, 64);
3656 		}
3657 	} else {
3658 		netif_napi_add(adapter->netdev, &adapter->rx_queue[0].napi,
3659 			       vmxnet3_poll, 64);
3660 	}
3661 
3662 	netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
3663 	netif_set_real_num_rx_queues(adapter->netdev, adapter->num_rx_queues);
3664 
3665 	netif_carrier_off(netdev);
3666 	err = register_netdev(netdev);
3667 
3668 	if (err) {
3669 		dev_err(&pdev->dev, "Failed to register adapter\n");
3670 		goto err_register;
3671 	}
3672 
3673 	vmxnet3_check_link(adapter, false);
3674 	return 0;
3675 
3676 err_register:
3677 	if (VMXNET3_VERSION_GE_3(adapter)) {
3678 		dma_free_coherent(&adapter->pdev->dev,
3679 				  sizeof(struct Vmxnet3_CoalesceScheme),
3680 				  adapter->coal_conf, adapter->coal_conf_pa);
3681 	}
3682 	vmxnet3_free_intr_resources(adapter);
3683 err_coal_conf:
3684 #ifdef VMXNET3_RSS
3685 	dma_free_coherent(&adapter->pdev->dev, sizeof(struct UPT1_RSSConf),
3686 			  adapter->rss_conf, adapter->rss_conf_pa);
3687 err_alloc_rss:
3688 #endif
3689 	dma_free_coherent(&adapter->pdev->dev, sizeof(struct Vmxnet3_PMConf),
3690 			  adapter->pm_conf, adapter->pm_conf_pa);
3691 err_alloc_pm:
3692 	dma_free_coherent(&adapter->pdev->dev, size, adapter->tqd_start,
3693 			  adapter->queue_desc_pa);
3694 err_ver:
3695 	vmxnet3_free_pci_resources(adapter);
3696 err_alloc_pci:
3697 	dma_free_coherent(&adapter->pdev->dev,
3698 			  sizeof(struct Vmxnet3_DriverShared),
3699 			  adapter->shared, adapter->shared_pa);
3700 err_alloc_shared:
3701 	dma_unmap_single(&adapter->pdev->dev, adapter->adapter_pa,
3702 			 sizeof(struct vmxnet3_adapter), DMA_TO_DEVICE);
3703 err_set_mask:
3704 	free_netdev(netdev);
3705 	return err;
3706 }
3707 
3708 
3709 static void
3710 vmxnet3_remove_device(struct pci_dev *pdev)
3711 {
3712 	struct net_device *netdev = pci_get_drvdata(pdev);
3713 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3714 	int size = 0;
3715 	int num_rx_queues, rx_queues;
3716 	unsigned long flags;
3717 
3718 #ifdef VMXNET3_RSS
3719 	if (enable_mq)
3720 		num_rx_queues = min(VMXNET3_DEVICE_MAX_RX_QUEUES,
3721 				    (int)num_online_cpus());
3722 	else
3723 #endif
3724 		num_rx_queues = 1;
3725 	if (!VMXNET3_VERSION_GE_6(adapter)) {
3726 		num_rx_queues = rounddown_pow_of_two(num_rx_queues);
3727 	}
3728 	if (VMXNET3_VERSION_GE_6(adapter)) {
3729 		spin_lock_irqsave(&adapter->cmd_lock, flags);
3730 		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3731 				       VMXNET3_CMD_GET_MAX_QUEUES_CONF);
3732 		rx_queues = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
3733 		spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3734 		if (rx_queues > 0)
3735 			rx_queues = (rx_queues >> 8) & 0xff;
3736 		else
3737 			rx_queues = min(num_rx_queues, VMXNET3_DEVICE_DEFAULT_RX_QUEUES);
3738 		num_rx_queues = min(num_rx_queues, rx_queues);
3739 	} else {
3740 		num_rx_queues = min(num_rx_queues,
3741 				    VMXNET3_DEVICE_DEFAULT_RX_QUEUES);
3742 	}
3743 
3744 	cancel_work_sync(&adapter->work);
3745 
3746 	unregister_netdev(netdev);
3747 
3748 	vmxnet3_free_intr_resources(adapter);
3749 	vmxnet3_free_pci_resources(adapter);
3750 	if (VMXNET3_VERSION_GE_3(adapter)) {
3751 		dma_free_coherent(&adapter->pdev->dev,
3752 				  sizeof(struct Vmxnet3_CoalesceScheme),
3753 				  adapter->coal_conf, adapter->coal_conf_pa);
3754 	}
3755 #ifdef VMXNET3_RSS
3756 	dma_free_coherent(&adapter->pdev->dev, sizeof(struct UPT1_RSSConf),
3757 			  adapter->rss_conf, adapter->rss_conf_pa);
3758 #endif
3759 	dma_free_coherent(&adapter->pdev->dev, sizeof(struct Vmxnet3_PMConf),
3760 			  adapter->pm_conf, adapter->pm_conf_pa);
3761 
3762 	size = sizeof(struct Vmxnet3_TxQueueDesc) * adapter->num_tx_queues;
3763 	size += sizeof(struct Vmxnet3_RxQueueDesc) * num_rx_queues;
3764 	dma_free_coherent(&adapter->pdev->dev, size, adapter->tqd_start,
3765 			  adapter->queue_desc_pa);
3766 	dma_free_coherent(&adapter->pdev->dev,
3767 			  sizeof(struct Vmxnet3_DriverShared),
3768 			  adapter->shared, adapter->shared_pa);
3769 	dma_unmap_single(&adapter->pdev->dev, adapter->adapter_pa,
3770 			 sizeof(struct vmxnet3_adapter), DMA_TO_DEVICE);
3771 	free_netdev(netdev);
3772 }
3773 
3774 static void vmxnet3_shutdown_device(struct pci_dev *pdev)
3775 {
3776 	struct net_device *netdev = pci_get_drvdata(pdev);
3777 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3778 	unsigned long flags;
3779 
3780 	/* Reset_work may be in the middle of resetting the device, wait for its
3781 	 * completion.
3782 	 */
3783 	while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
3784 		usleep_range(1000, 2000);
3785 
3786 	if (test_and_set_bit(VMXNET3_STATE_BIT_QUIESCED,
3787 			     &adapter->state)) {
3788 		clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
3789 		return;
3790 	}
3791 	spin_lock_irqsave(&adapter->cmd_lock, flags);
3792 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3793 			       VMXNET3_CMD_QUIESCE_DEV);
3794 	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3795 	vmxnet3_disable_all_intrs(adapter);
3796 
3797 	clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
3798 }
3799 
3800 
3801 #ifdef CONFIG_PM
3802 
3803 static int
3804 vmxnet3_suspend(struct device *device)
3805 {
3806 	struct pci_dev *pdev = to_pci_dev(device);
3807 	struct net_device *netdev = pci_get_drvdata(pdev);
3808 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3809 	struct Vmxnet3_PMConf *pmConf;
3810 	struct ethhdr *ehdr;
3811 	struct arphdr *ahdr;
3812 	u8 *arpreq;
3813 	struct in_device *in_dev;
3814 	struct in_ifaddr *ifa;
3815 	unsigned long flags;
3816 	int i = 0;
3817 
3818 	if (!netif_running(netdev))
3819 		return 0;
3820 
3821 	for (i = 0; i < adapter->num_rx_queues; i++)
3822 		napi_disable(&adapter->rx_queue[i].napi);
3823 
3824 	vmxnet3_disable_all_intrs(adapter);
3825 	vmxnet3_free_irqs(adapter);
3826 	vmxnet3_free_intr_resources(adapter);
3827 
3828 	netif_device_detach(netdev);
3829 
3830 	/* Create wake-up filters. */
3831 	pmConf = adapter->pm_conf;
3832 	memset(pmConf, 0, sizeof(*pmConf));
3833 
3834 	if (adapter->wol & WAKE_UCAST) {
3835 		pmConf->filters[i].patternSize = ETH_ALEN;
3836 		pmConf->filters[i].maskSize = 1;
3837 		memcpy(pmConf->filters[i].pattern, netdev->dev_addr, ETH_ALEN);
3838 		pmConf->filters[i].mask[0] = 0x3F; /* LSB ETH_ALEN bits */
3839 
3840 		pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_FILTER;
3841 		i++;
3842 	}
3843 
3844 	if (adapter->wol & WAKE_ARP) {
3845 		rcu_read_lock();
3846 
3847 		in_dev = __in_dev_get_rcu(netdev);
3848 		if (!in_dev) {
3849 			rcu_read_unlock();
3850 			goto skip_arp;
3851 		}
3852 
3853 		ifa = rcu_dereference(in_dev->ifa_list);
3854 		if (!ifa) {
3855 			rcu_read_unlock();
3856 			goto skip_arp;
3857 		}
3858 
3859 		pmConf->filters[i].patternSize = ETH_HLEN + /* Ethernet header*/
3860 			sizeof(struct arphdr) +		/* ARP header */
3861 			2 * ETH_ALEN +		/* 2 Ethernet addresses*/
3862 			2 * sizeof(u32);	/*2 IPv4 addresses */
3863 		pmConf->filters[i].maskSize =
3864 			(pmConf->filters[i].patternSize - 1) / 8 + 1;
3865 
3866 		/* ETH_P_ARP in Ethernet header. */
3867 		ehdr = (struct ethhdr *)pmConf->filters[i].pattern;
3868 		ehdr->h_proto = htons(ETH_P_ARP);
3869 
3870 		/* ARPOP_REQUEST in ARP header. */
3871 		ahdr = (struct arphdr *)&pmConf->filters[i].pattern[ETH_HLEN];
3872 		ahdr->ar_op = htons(ARPOP_REQUEST);
3873 		arpreq = (u8 *)(ahdr + 1);
3874 
3875 		/* The Unicast IPv4 address in 'tip' field. */
3876 		arpreq += 2 * ETH_ALEN + sizeof(u32);
3877 		*(__be32 *)arpreq = ifa->ifa_address;
3878 
3879 		rcu_read_unlock();
3880 
3881 		/* The mask for the relevant bits. */
3882 		pmConf->filters[i].mask[0] = 0x00;
3883 		pmConf->filters[i].mask[1] = 0x30; /* ETH_P_ARP */
3884 		pmConf->filters[i].mask[2] = 0x30; /* ARPOP_REQUEST */
3885 		pmConf->filters[i].mask[3] = 0x00;
3886 		pmConf->filters[i].mask[4] = 0xC0; /* IPv4 TIP */
3887 		pmConf->filters[i].mask[5] = 0x03; /* IPv4 TIP */
3888 
3889 		pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_FILTER;
3890 		i++;
3891 	}
3892 
3893 skip_arp:
3894 	if (adapter->wol & WAKE_MAGIC)
3895 		pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_MAGIC;
3896 
3897 	pmConf->numFilters = i;
3898 
3899 	adapter->shared->devRead.pmConfDesc.confVer = cpu_to_le32(1);
3900 	adapter->shared->devRead.pmConfDesc.confLen = cpu_to_le32(sizeof(
3901 								  *pmConf));
3902 	adapter->shared->devRead.pmConfDesc.confPA =
3903 		cpu_to_le64(adapter->pm_conf_pa);
3904 
3905 	spin_lock_irqsave(&adapter->cmd_lock, flags);
3906 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3907 			       VMXNET3_CMD_UPDATE_PMCFG);
3908 	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3909 
3910 	pci_save_state(pdev);
3911 	pci_enable_wake(pdev, pci_choose_state(pdev, PMSG_SUSPEND),
3912 			adapter->wol);
3913 	pci_disable_device(pdev);
3914 	pci_set_power_state(pdev, pci_choose_state(pdev, PMSG_SUSPEND));
3915 
3916 	return 0;
3917 }
3918 
3919 
3920 static int
3921 vmxnet3_resume(struct device *device)
3922 {
3923 	int err;
3924 	unsigned long flags;
3925 	struct pci_dev *pdev = to_pci_dev(device);
3926 	struct net_device *netdev = pci_get_drvdata(pdev);
3927 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3928 
3929 	if (!netif_running(netdev))
3930 		return 0;
3931 
3932 	pci_set_power_state(pdev, PCI_D0);
3933 	pci_restore_state(pdev);
3934 	err = pci_enable_device_mem(pdev);
3935 	if (err != 0)
3936 		return err;
3937 
3938 	pci_enable_wake(pdev, PCI_D0, 0);
3939 
3940 	vmxnet3_alloc_intr_resources(adapter);
3941 
3942 	/* During hibernate and suspend, device has to be reinitialized as the
3943 	 * device state need not be preserved.
3944 	 */
3945 
3946 	/* Need not check adapter state as other reset tasks cannot run during
3947 	 * device resume.
3948 	 */
3949 	spin_lock_irqsave(&adapter->cmd_lock, flags);
3950 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3951 			       VMXNET3_CMD_QUIESCE_DEV);
3952 	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3953 	vmxnet3_tq_cleanup_all(adapter);
3954 	vmxnet3_rq_cleanup_all(adapter);
3955 
3956 	vmxnet3_reset_dev(adapter);
3957 	err = vmxnet3_activate_dev(adapter);
3958 	if (err != 0) {
3959 		netdev_err(netdev,
3960 			   "failed to re-activate on resume, error: %d", err);
3961 		vmxnet3_force_close(adapter);
3962 		return err;
3963 	}
3964 	netif_device_attach(netdev);
3965 
3966 	return 0;
3967 }
3968 
3969 static const struct dev_pm_ops vmxnet3_pm_ops = {
3970 	.suspend = vmxnet3_suspend,
3971 	.resume = vmxnet3_resume,
3972 	.freeze = vmxnet3_suspend,
3973 	.restore = vmxnet3_resume,
3974 };
3975 #endif
3976 
3977 static struct pci_driver vmxnet3_driver = {
3978 	.name		= vmxnet3_driver_name,
3979 	.id_table	= vmxnet3_pciid_table,
3980 	.probe		= vmxnet3_probe_device,
3981 	.remove		= vmxnet3_remove_device,
3982 	.shutdown	= vmxnet3_shutdown_device,
3983 #ifdef CONFIG_PM
3984 	.driver.pm	= &vmxnet3_pm_ops,
3985 #endif
3986 };
3987 
3988 
3989 static int __init
3990 vmxnet3_init_module(void)
3991 {
3992 	pr_info("%s - version %s\n", VMXNET3_DRIVER_DESC,
3993 		VMXNET3_DRIVER_VERSION_REPORT);
3994 	return pci_register_driver(&vmxnet3_driver);
3995 }
3996 
3997 module_init(vmxnet3_init_module);
3998 
3999 
4000 static void
4001 vmxnet3_exit_module(void)
4002 {
4003 	pci_unregister_driver(&vmxnet3_driver);
4004 }
4005 
4006 module_exit(vmxnet3_exit_module);
4007 
4008 MODULE_AUTHOR("VMware, Inc.");
4009 MODULE_DESCRIPTION(VMXNET3_DRIVER_DESC);
4010 MODULE_LICENSE("GPL v2");
4011 MODULE_VERSION(VMXNET3_DRIVER_VERSION_STRING);
4012