1 /*
2  * Linux driver for VMware's vmxnet3 ethernet NIC.
3  *
4  * Copyright (C) 2008-2009, VMware, Inc. All Rights Reserved.
5  *
6  * This program is free software; you can redistribute it and/or modify it
7  * under the terms of the GNU General Public License as published by the
8  * Free Software Foundation; version 2 of the License and no later version.
9  *
10  * This program is distributed in the hope that it will be useful, but
11  * WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
13  * NON INFRINGEMENT. See the GNU General Public License for more
14  * details.
15  *
16  * You should have received a copy of the GNU General Public License
17  * along with this program; if not, write to the Free Software
18  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19  *
20  * The full GNU General Public License is included in this distribution in
21  * the file called "COPYING".
22  *
23  * Maintained by: Shreyas Bhatewara <pv-drivers@vmware.com>
24  *
25  */
26 
27 #include <linux/module.h>
28 #include <net/ip6_checksum.h>
29 
30 #include "vmxnet3_int.h"
31 
32 char vmxnet3_driver_name[] = "vmxnet3";
33 #define VMXNET3_DRIVER_DESC "VMware vmxnet3 virtual NIC driver"
34 
35 /*
36  * PCI Device ID Table
37  * Last entry must be all 0s
38  */
39 static const struct pci_device_id vmxnet3_pciid_table[] = {
40 	{PCI_VDEVICE(VMWARE, PCI_DEVICE_ID_VMWARE_VMXNET3)},
41 	{0}
42 };
43 
44 MODULE_DEVICE_TABLE(pci, vmxnet3_pciid_table);
45 
46 static int enable_mq = 1;
47 
48 static void
49 vmxnet3_write_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac);
50 
51 /*
52  *    Enable/Disable the given intr
53  */
54 static void
55 vmxnet3_enable_intr(struct vmxnet3_adapter *adapter, unsigned intr_idx)
56 {
57 	VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_IMR + intr_idx * 8, 0);
58 }
59 
60 
61 static void
62 vmxnet3_disable_intr(struct vmxnet3_adapter *adapter, unsigned intr_idx)
63 {
64 	VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_IMR + intr_idx * 8, 1);
65 }
66 
67 
68 /*
69  *    Enable/Disable all intrs used by the device
70  */
71 static void
72 vmxnet3_enable_all_intrs(struct vmxnet3_adapter *adapter)
73 {
74 	int i;
75 
76 	for (i = 0; i < adapter->intr.num_intrs; i++)
77 		vmxnet3_enable_intr(adapter, i);
78 	adapter->shared->devRead.intrConf.intrCtrl &=
79 					cpu_to_le32(~VMXNET3_IC_DISABLE_ALL);
80 }
81 
82 
83 static void
84 vmxnet3_disable_all_intrs(struct vmxnet3_adapter *adapter)
85 {
86 	int i;
87 
88 	adapter->shared->devRead.intrConf.intrCtrl |=
89 					cpu_to_le32(VMXNET3_IC_DISABLE_ALL);
90 	for (i = 0; i < adapter->intr.num_intrs; i++)
91 		vmxnet3_disable_intr(adapter, i);
92 }
93 
94 
95 static void
96 vmxnet3_ack_events(struct vmxnet3_adapter *adapter, u32 events)
97 {
98 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_ECR, events);
99 }
100 
101 
102 static bool
103 vmxnet3_tq_stopped(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
104 {
105 	return tq->stopped;
106 }
107 
108 
109 static void
110 vmxnet3_tq_start(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
111 {
112 	tq->stopped = false;
113 	netif_start_subqueue(adapter->netdev, tq - adapter->tx_queue);
114 }
115 
116 
117 static void
118 vmxnet3_tq_wake(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
119 {
120 	tq->stopped = false;
121 	netif_wake_subqueue(adapter->netdev, (tq - adapter->tx_queue));
122 }
123 
124 
125 static void
126 vmxnet3_tq_stop(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
127 {
128 	tq->stopped = true;
129 	tq->num_stop++;
130 	netif_stop_subqueue(adapter->netdev, (tq - adapter->tx_queue));
131 }
132 
133 
134 /*
135  * Check the link state. This may start or stop the tx queue.
136  */
137 static void
138 vmxnet3_check_link(struct vmxnet3_adapter *adapter, bool affectTxQueue)
139 {
140 	u32 ret;
141 	int i;
142 	unsigned long flags;
143 
144 	spin_lock_irqsave(&adapter->cmd_lock, flags);
145 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, VMXNET3_CMD_GET_LINK);
146 	ret = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
147 	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
148 
149 	adapter->link_speed = ret >> 16;
150 	if (ret & 1) { /* Link is up. */
151 		netdev_info(adapter->netdev, "NIC Link is Up %d Mbps\n",
152 			    adapter->link_speed);
153 		netif_carrier_on(adapter->netdev);
154 
155 		if (affectTxQueue) {
156 			for (i = 0; i < adapter->num_tx_queues; i++)
157 				vmxnet3_tq_start(&adapter->tx_queue[i],
158 						 adapter);
159 		}
160 	} else {
161 		netdev_info(adapter->netdev, "NIC Link is Down\n");
162 		netif_carrier_off(adapter->netdev);
163 
164 		if (affectTxQueue) {
165 			for (i = 0; i < adapter->num_tx_queues; i++)
166 				vmxnet3_tq_stop(&adapter->tx_queue[i], adapter);
167 		}
168 	}
169 }
170 
171 static void
172 vmxnet3_process_events(struct vmxnet3_adapter *adapter)
173 {
174 	int i;
175 	unsigned long flags;
176 	u32 events = le32_to_cpu(adapter->shared->ecr);
177 	if (!events)
178 		return;
179 
180 	vmxnet3_ack_events(adapter, events);
181 
182 	/* Check if link state has changed */
183 	if (events & VMXNET3_ECR_LINK)
184 		vmxnet3_check_link(adapter, true);
185 
186 	/* Check if there is an error on xmit/recv queues */
187 	if (events & (VMXNET3_ECR_TQERR | VMXNET3_ECR_RQERR)) {
188 		spin_lock_irqsave(&adapter->cmd_lock, flags);
189 		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
190 				       VMXNET3_CMD_GET_QUEUE_STATUS);
191 		spin_unlock_irqrestore(&adapter->cmd_lock, flags);
192 
193 		for (i = 0; i < adapter->num_tx_queues; i++)
194 			if (adapter->tqd_start[i].status.stopped)
195 				dev_err(&adapter->netdev->dev,
196 					"%s: tq[%d] error 0x%x\n",
197 					adapter->netdev->name, i, le32_to_cpu(
198 					adapter->tqd_start[i].status.error));
199 		for (i = 0; i < adapter->num_rx_queues; i++)
200 			if (adapter->rqd_start[i].status.stopped)
201 				dev_err(&adapter->netdev->dev,
202 					"%s: rq[%d] error 0x%x\n",
203 					adapter->netdev->name, i,
204 					adapter->rqd_start[i].status.error);
205 
206 		schedule_work(&adapter->work);
207 	}
208 }
209 
210 #ifdef __BIG_ENDIAN_BITFIELD
211 /*
212  * The device expects the bitfields in shared structures to be written in
213  * little endian. When CPU is big endian, the following routines are used to
214  * correctly read and write into ABI.
215  * The general technique used here is : double word bitfields are defined in
216  * opposite order for big endian architecture. Then before reading them in
217  * driver the complete double word is translated using le32_to_cpu. Similarly
218  * After the driver writes into bitfields, cpu_to_le32 is used to translate the
219  * double words into required format.
220  * In order to avoid touching bits in shared structure more than once, temporary
221  * descriptors are used. These are passed as srcDesc to following functions.
222  */
223 static void vmxnet3_RxDescToCPU(const struct Vmxnet3_RxDesc *srcDesc,
224 				struct Vmxnet3_RxDesc *dstDesc)
225 {
226 	u32 *src = (u32 *)srcDesc + 2;
227 	u32 *dst = (u32 *)dstDesc + 2;
228 	dstDesc->addr = le64_to_cpu(srcDesc->addr);
229 	*dst = le32_to_cpu(*src);
230 	dstDesc->ext1 = le32_to_cpu(srcDesc->ext1);
231 }
232 
233 static void vmxnet3_TxDescToLe(const struct Vmxnet3_TxDesc *srcDesc,
234 			       struct Vmxnet3_TxDesc *dstDesc)
235 {
236 	int i;
237 	u32 *src = (u32 *)(srcDesc + 1);
238 	u32 *dst = (u32 *)(dstDesc + 1);
239 
240 	/* Working backwards so that the gen bit is set at the end. */
241 	for (i = 2; i > 0; i--) {
242 		src--;
243 		dst--;
244 		*dst = cpu_to_le32(*src);
245 	}
246 }
247 
248 
249 static void vmxnet3_RxCompToCPU(const struct Vmxnet3_RxCompDesc *srcDesc,
250 				struct Vmxnet3_RxCompDesc *dstDesc)
251 {
252 	int i = 0;
253 	u32 *src = (u32 *)srcDesc;
254 	u32 *dst = (u32 *)dstDesc;
255 	for (i = 0; i < sizeof(struct Vmxnet3_RxCompDesc) / sizeof(u32); i++) {
256 		*dst = le32_to_cpu(*src);
257 		src++;
258 		dst++;
259 	}
260 }
261 
262 
263 /* Used to read bitfield values from double words. */
264 static u32 get_bitfield32(const __le32 *bitfield, u32 pos, u32 size)
265 {
266 	u32 temp = le32_to_cpu(*bitfield);
267 	u32 mask = ((1 << size) - 1) << pos;
268 	temp &= mask;
269 	temp >>= pos;
270 	return temp;
271 }
272 
273 
274 
275 #endif  /* __BIG_ENDIAN_BITFIELD */
276 
277 #ifdef __BIG_ENDIAN_BITFIELD
278 
279 #   define VMXNET3_TXDESC_GET_GEN(txdesc) get_bitfield32(((const __le32 *) \
280 			txdesc) + VMXNET3_TXD_GEN_DWORD_SHIFT, \
281 			VMXNET3_TXD_GEN_SHIFT, VMXNET3_TXD_GEN_SIZE)
282 #   define VMXNET3_TXDESC_GET_EOP(txdesc) get_bitfield32(((const __le32 *) \
283 			txdesc) + VMXNET3_TXD_EOP_DWORD_SHIFT, \
284 			VMXNET3_TXD_EOP_SHIFT, VMXNET3_TXD_EOP_SIZE)
285 #   define VMXNET3_TCD_GET_GEN(tcd) get_bitfield32(((const __le32 *)tcd) + \
286 			VMXNET3_TCD_GEN_DWORD_SHIFT, VMXNET3_TCD_GEN_SHIFT, \
287 			VMXNET3_TCD_GEN_SIZE)
288 #   define VMXNET3_TCD_GET_TXIDX(tcd) get_bitfield32((const __le32 *)tcd, \
289 			VMXNET3_TCD_TXIDX_SHIFT, VMXNET3_TCD_TXIDX_SIZE)
290 #   define vmxnet3_getRxComp(dstrcd, rcd, tmp) do { \
291 			(dstrcd) = (tmp); \
292 			vmxnet3_RxCompToCPU((rcd), (tmp)); \
293 		} while (0)
294 #   define vmxnet3_getRxDesc(dstrxd, rxd, tmp) do { \
295 			(dstrxd) = (tmp); \
296 			vmxnet3_RxDescToCPU((rxd), (tmp)); \
297 		} while (0)
298 
299 #else
300 
301 #   define VMXNET3_TXDESC_GET_GEN(txdesc) ((txdesc)->gen)
302 #   define VMXNET3_TXDESC_GET_EOP(txdesc) ((txdesc)->eop)
303 #   define VMXNET3_TCD_GET_GEN(tcd) ((tcd)->gen)
304 #   define VMXNET3_TCD_GET_TXIDX(tcd) ((tcd)->txdIdx)
305 #   define vmxnet3_getRxComp(dstrcd, rcd, tmp) (dstrcd) = (rcd)
306 #   define vmxnet3_getRxDesc(dstrxd, rxd, tmp) (dstrxd) = (rxd)
307 
308 #endif /* __BIG_ENDIAN_BITFIELD  */
309 
310 
311 static void
312 vmxnet3_unmap_tx_buf(struct vmxnet3_tx_buf_info *tbi,
313 		     struct pci_dev *pdev)
314 {
315 	if (tbi->map_type == VMXNET3_MAP_SINGLE)
316 		dma_unmap_single(&pdev->dev, tbi->dma_addr, tbi->len,
317 				 PCI_DMA_TODEVICE);
318 	else if (tbi->map_type == VMXNET3_MAP_PAGE)
319 		dma_unmap_page(&pdev->dev, tbi->dma_addr, tbi->len,
320 			       PCI_DMA_TODEVICE);
321 	else
322 		BUG_ON(tbi->map_type != VMXNET3_MAP_NONE);
323 
324 	tbi->map_type = VMXNET3_MAP_NONE; /* to help debugging */
325 }
326 
327 
328 static int
329 vmxnet3_unmap_pkt(u32 eop_idx, struct vmxnet3_tx_queue *tq,
330 		  struct pci_dev *pdev,	struct vmxnet3_adapter *adapter)
331 {
332 	struct sk_buff *skb;
333 	int entries = 0;
334 
335 	/* no out of order completion */
336 	BUG_ON(tq->buf_info[eop_idx].sop_idx != tq->tx_ring.next2comp);
337 	BUG_ON(VMXNET3_TXDESC_GET_EOP(&(tq->tx_ring.base[eop_idx].txd)) != 1);
338 
339 	skb = tq->buf_info[eop_idx].skb;
340 	BUG_ON(skb == NULL);
341 	tq->buf_info[eop_idx].skb = NULL;
342 
343 	VMXNET3_INC_RING_IDX_ONLY(eop_idx, tq->tx_ring.size);
344 
345 	while (tq->tx_ring.next2comp != eop_idx) {
346 		vmxnet3_unmap_tx_buf(tq->buf_info + tq->tx_ring.next2comp,
347 				     pdev);
348 
349 		/* update next2comp w/o tx_lock. Since we are marking more,
350 		 * instead of less, tx ring entries avail, the worst case is
351 		 * that the tx routine incorrectly re-queues a pkt due to
352 		 * insufficient tx ring entries.
353 		 */
354 		vmxnet3_cmd_ring_adv_next2comp(&tq->tx_ring);
355 		entries++;
356 	}
357 
358 	dev_kfree_skb_any(skb);
359 	return entries;
360 }
361 
362 
363 static int
364 vmxnet3_tq_tx_complete(struct vmxnet3_tx_queue *tq,
365 			struct vmxnet3_adapter *adapter)
366 {
367 	int completed = 0;
368 	union Vmxnet3_GenericDesc *gdesc;
369 
370 	gdesc = tq->comp_ring.base + tq->comp_ring.next2proc;
371 	while (VMXNET3_TCD_GET_GEN(&gdesc->tcd) == tq->comp_ring.gen) {
372 		completed += vmxnet3_unmap_pkt(VMXNET3_TCD_GET_TXIDX(
373 					       &gdesc->tcd), tq, adapter->pdev,
374 					       adapter);
375 
376 		vmxnet3_comp_ring_adv_next2proc(&tq->comp_ring);
377 		gdesc = tq->comp_ring.base + tq->comp_ring.next2proc;
378 	}
379 
380 	if (completed) {
381 		spin_lock(&tq->tx_lock);
382 		if (unlikely(vmxnet3_tq_stopped(tq, adapter) &&
383 			     vmxnet3_cmd_ring_desc_avail(&tq->tx_ring) >
384 			     VMXNET3_WAKE_QUEUE_THRESHOLD(tq) &&
385 			     netif_carrier_ok(adapter->netdev))) {
386 			vmxnet3_tq_wake(tq, adapter);
387 		}
388 		spin_unlock(&tq->tx_lock);
389 	}
390 	return completed;
391 }
392 
393 
394 static void
395 vmxnet3_tq_cleanup(struct vmxnet3_tx_queue *tq,
396 		   struct vmxnet3_adapter *adapter)
397 {
398 	int i;
399 
400 	while (tq->tx_ring.next2comp != tq->tx_ring.next2fill) {
401 		struct vmxnet3_tx_buf_info *tbi;
402 
403 		tbi = tq->buf_info + tq->tx_ring.next2comp;
404 
405 		vmxnet3_unmap_tx_buf(tbi, adapter->pdev);
406 		if (tbi->skb) {
407 			dev_kfree_skb_any(tbi->skb);
408 			tbi->skb = NULL;
409 		}
410 		vmxnet3_cmd_ring_adv_next2comp(&tq->tx_ring);
411 	}
412 
413 	/* sanity check, verify all buffers are indeed unmapped and freed */
414 	for (i = 0; i < tq->tx_ring.size; i++) {
415 		BUG_ON(tq->buf_info[i].skb != NULL ||
416 		       tq->buf_info[i].map_type != VMXNET3_MAP_NONE);
417 	}
418 
419 	tq->tx_ring.gen = VMXNET3_INIT_GEN;
420 	tq->tx_ring.next2fill = tq->tx_ring.next2comp = 0;
421 
422 	tq->comp_ring.gen = VMXNET3_INIT_GEN;
423 	tq->comp_ring.next2proc = 0;
424 }
425 
426 
427 static void
428 vmxnet3_tq_destroy(struct vmxnet3_tx_queue *tq,
429 		   struct vmxnet3_adapter *adapter)
430 {
431 	if (tq->tx_ring.base) {
432 		dma_free_coherent(&adapter->pdev->dev, tq->tx_ring.size *
433 				  sizeof(struct Vmxnet3_TxDesc),
434 				  tq->tx_ring.base, tq->tx_ring.basePA);
435 		tq->tx_ring.base = NULL;
436 	}
437 	if (tq->data_ring.base) {
438 		dma_free_coherent(&adapter->pdev->dev, tq->data_ring.size *
439 				  sizeof(struct Vmxnet3_TxDataDesc),
440 				  tq->data_ring.base, tq->data_ring.basePA);
441 		tq->data_ring.base = NULL;
442 	}
443 	if (tq->comp_ring.base) {
444 		dma_free_coherent(&adapter->pdev->dev, tq->comp_ring.size *
445 				  sizeof(struct Vmxnet3_TxCompDesc),
446 				  tq->comp_ring.base, tq->comp_ring.basePA);
447 		tq->comp_ring.base = NULL;
448 	}
449 	if (tq->buf_info) {
450 		dma_free_coherent(&adapter->pdev->dev,
451 				  tq->tx_ring.size * sizeof(tq->buf_info[0]),
452 				  tq->buf_info, tq->buf_info_pa);
453 		tq->buf_info = NULL;
454 	}
455 }
456 
457 
458 /* Destroy all tx queues */
459 void
460 vmxnet3_tq_destroy_all(struct vmxnet3_adapter *adapter)
461 {
462 	int i;
463 
464 	for (i = 0; i < adapter->num_tx_queues; i++)
465 		vmxnet3_tq_destroy(&adapter->tx_queue[i], adapter);
466 }
467 
468 
469 static void
470 vmxnet3_tq_init(struct vmxnet3_tx_queue *tq,
471 		struct vmxnet3_adapter *adapter)
472 {
473 	int i;
474 
475 	/* reset the tx ring contents to 0 and reset the tx ring states */
476 	memset(tq->tx_ring.base, 0, tq->tx_ring.size *
477 	       sizeof(struct Vmxnet3_TxDesc));
478 	tq->tx_ring.next2fill = tq->tx_ring.next2comp = 0;
479 	tq->tx_ring.gen = VMXNET3_INIT_GEN;
480 
481 	memset(tq->data_ring.base, 0, tq->data_ring.size *
482 	       sizeof(struct Vmxnet3_TxDataDesc));
483 
484 	/* reset the tx comp ring contents to 0 and reset comp ring states */
485 	memset(tq->comp_ring.base, 0, tq->comp_ring.size *
486 	       sizeof(struct Vmxnet3_TxCompDesc));
487 	tq->comp_ring.next2proc = 0;
488 	tq->comp_ring.gen = VMXNET3_INIT_GEN;
489 
490 	/* reset the bookkeeping data */
491 	memset(tq->buf_info, 0, sizeof(tq->buf_info[0]) * tq->tx_ring.size);
492 	for (i = 0; i < tq->tx_ring.size; i++)
493 		tq->buf_info[i].map_type = VMXNET3_MAP_NONE;
494 
495 	/* stats are not reset */
496 }
497 
498 
499 static int
500 vmxnet3_tq_create(struct vmxnet3_tx_queue *tq,
501 		  struct vmxnet3_adapter *adapter)
502 {
503 	size_t sz;
504 
505 	BUG_ON(tq->tx_ring.base || tq->data_ring.base ||
506 	       tq->comp_ring.base || tq->buf_info);
507 
508 	tq->tx_ring.base = dma_alloc_coherent(&adapter->pdev->dev,
509 			tq->tx_ring.size * sizeof(struct Vmxnet3_TxDesc),
510 			&tq->tx_ring.basePA, GFP_KERNEL);
511 	if (!tq->tx_ring.base) {
512 		netdev_err(adapter->netdev, "failed to allocate tx ring\n");
513 		goto err;
514 	}
515 
516 	tq->data_ring.base = dma_alloc_coherent(&adapter->pdev->dev,
517 			tq->data_ring.size * sizeof(struct Vmxnet3_TxDataDesc),
518 			&tq->data_ring.basePA, GFP_KERNEL);
519 	if (!tq->data_ring.base) {
520 		netdev_err(adapter->netdev, "failed to allocate data ring\n");
521 		goto err;
522 	}
523 
524 	tq->comp_ring.base = dma_alloc_coherent(&adapter->pdev->dev,
525 			tq->comp_ring.size * sizeof(struct Vmxnet3_TxCompDesc),
526 			&tq->comp_ring.basePA, GFP_KERNEL);
527 	if (!tq->comp_ring.base) {
528 		netdev_err(adapter->netdev, "failed to allocate tx comp ring\n");
529 		goto err;
530 	}
531 
532 	sz = tq->tx_ring.size * sizeof(tq->buf_info[0]);
533 	tq->buf_info = dma_zalloc_coherent(&adapter->pdev->dev, sz,
534 					   &tq->buf_info_pa, GFP_KERNEL);
535 	if (!tq->buf_info)
536 		goto err;
537 
538 	return 0;
539 
540 err:
541 	vmxnet3_tq_destroy(tq, adapter);
542 	return -ENOMEM;
543 }
544 
545 static void
546 vmxnet3_tq_cleanup_all(struct vmxnet3_adapter *adapter)
547 {
548 	int i;
549 
550 	for (i = 0; i < adapter->num_tx_queues; i++)
551 		vmxnet3_tq_cleanup(&adapter->tx_queue[i], adapter);
552 }
553 
554 /*
555  *    starting from ring->next2fill, allocate rx buffers for the given ring
556  *    of the rx queue and update the rx desc. stop after @num_to_alloc buffers
557  *    are allocated or allocation fails
558  */
559 
560 static int
561 vmxnet3_rq_alloc_rx_buf(struct vmxnet3_rx_queue *rq, u32 ring_idx,
562 			int num_to_alloc, struct vmxnet3_adapter *adapter)
563 {
564 	int num_allocated = 0;
565 	struct vmxnet3_rx_buf_info *rbi_base = rq->buf_info[ring_idx];
566 	struct vmxnet3_cmd_ring *ring = &rq->rx_ring[ring_idx];
567 	u32 val;
568 
569 	while (num_allocated <= num_to_alloc) {
570 		struct vmxnet3_rx_buf_info *rbi;
571 		union Vmxnet3_GenericDesc *gd;
572 
573 		rbi = rbi_base + ring->next2fill;
574 		gd = ring->base + ring->next2fill;
575 
576 		if (rbi->buf_type == VMXNET3_RX_BUF_SKB) {
577 			if (rbi->skb == NULL) {
578 				rbi->skb = __netdev_alloc_skb_ip_align(adapter->netdev,
579 								       rbi->len,
580 								       GFP_KERNEL);
581 				if (unlikely(rbi->skb == NULL)) {
582 					rq->stats.rx_buf_alloc_failure++;
583 					break;
584 				}
585 
586 				rbi->dma_addr = dma_map_single(
587 						&adapter->pdev->dev,
588 						rbi->skb->data, rbi->len,
589 						PCI_DMA_FROMDEVICE);
590 				if (dma_mapping_error(&adapter->pdev->dev,
591 						      rbi->dma_addr)) {
592 					dev_kfree_skb_any(rbi->skb);
593 					rq->stats.rx_buf_alloc_failure++;
594 					break;
595 				}
596 			} else {
597 				/* rx buffer skipped by the device */
598 			}
599 			val = VMXNET3_RXD_BTYPE_HEAD << VMXNET3_RXD_BTYPE_SHIFT;
600 		} else {
601 			BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_PAGE ||
602 			       rbi->len  != PAGE_SIZE);
603 
604 			if (rbi->page == NULL) {
605 				rbi->page = alloc_page(GFP_ATOMIC);
606 				if (unlikely(rbi->page == NULL)) {
607 					rq->stats.rx_buf_alloc_failure++;
608 					break;
609 				}
610 				rbi->dma_addr = dma_map_page(
611 						&adapter->pdev->dev,
612 						rbi->page, 0, PAGE_SIZE,
613 						PCI_DMA_FROMDEVICE);
614 				if (dma_mapping_error(&adapter->pdev->dev,
615 						      rbi->dma_addr)) {
616 					put_page(rbi->page);
617 					rq->stats.rx_buf_alloc_failure++;
618 					break;
619 				}
620 			} else {
621 				/* rx buffers skipped by the device */
622 			}
623 			val = VMXNET3_RXD_BTYPE_BODY << VMXNET3_RXD_BTYPE_SHIFT;
624 		}
625 
626 		gd->rxd.addr = cpu_to_le64(rbi->dma_addr);
627 		gd->dword[2] = cpu_to_le32((!ring->gen << VMXNET3_RXD_GEN_SHIFT)
628 					   | val | rbi->len);
629 
630 		/* Fill the last buffer but dont mark it ready, or else the
631 		 * device will think that the queue is full */
632 		if (num_allocated == num_to_alloc)
633 			break;
634 
635 		gd->dword[2] |= cpu_to_le32(ring->gen << VMXNET3_RXD_GEN_SHIFT);
636 		num_allocated++;
637 		vmxnet3_cmd_ring_adv_next2fill(ring);
638 	}
639 
640 	netdev_dbg(adapter->netdev,
641 		"alloc_rx_buf: %d allocated, next2fill %u, next2comp %u\n",
642 		num_allocated, ring->next2fill, ring->next2comp);
643 
644 	/* so that the device can distinguish a full ring and an empty ring */
645 	BUG_ON(num_allocated != 0 && ring->next2fill == ring->next2comp);
646 
647 	return num_allocated;
648 }
649 
650 
651 static void
652 vmxnet3_append_frag(struct sk_buff *skb, struct Vmxnet3_RxCompDesc *rcd,
653 		    struct vmxnet3_rx_buf_info *rbi)
654 {
655 	struct skb_frag_struct *frag = skb_shinfo(skb)->frags +
656 		skb_shinfo(skb)->nr_frags;
657 
658 	BUG_ON(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS);
659 
660 	__skb_frag_set_page(frag, rbi->page);
661 	frag->page_offset = 0;
662 	skb_frag_size_set(frag, rcd->len);
663 	skb->data_len += rcd->len;
664 	skb->truesize += PAGE_SIZE;
665 	skb_shinfo(skb)->nr_frags++;
666 }
667 
668 
669 static int
670 vmxnet3_map_pkt(struct sk_buff *skb, struct vmxnet3_tx_ctx *ctx,
671 		struct vmxnet3_tx_queue *tq, struct pci_dev *pdev,
672 		struct vmxnet3_adapter *adapter)
673 {
674 	u32 dw2, len;
675 	unsigned long buf_offset;
676 	int i;
677 	union Vmxnet3_GenericDesc *gdesc;
678 	struct vmxnet3_tx_buf_info *tbi = NULL;
679 
680 	BUG_ON(ctx->copy_size > skb_headlen(skb));
681 
682 	/* use the previous gen bit for the SOP desc */
683 	dw2 = (tq->tx_ring.gen ^ 0x1) << VMXNET3_TXD_GEN_SHIFT;
684 
685 	ctx->sop_txd = tq->tx_ring.base + tq->tx_ring.next2fill;
686 	gdesc = ctx->sop_txd; /* both loops below can be skipped */
687 
688 	/* no need to map the buffer if headers are copied */
689 	if (ctx->copy_size) {
690 		ctx->sop_txd->txd.addr = cpu_to_le64(tq->data_ring.basePA +
691 					tq->tx_ring.next2fill *
692 					sizeof(struct Vmxnet3_TxDataDesc));
693 		ctx->sop_txd->dword[2] = cpu_to_le32(dw2 | ctx->copy_size);
694 		ctx->sop_txd->dword[3] = 0;
695 
696 		tbi = tq->buf_info + tq->tx_ring.next2fill;
697 		tbi->map_type = VMXNET3_MAP_NONE;
698 
699 		netdev_dbg(adapter->netdev,
700 			"txd[%u]: 0x%Lx 0x%x 0x%x\n",
701 			tq->tx_ring.next2fill,
702 			le64_to_cpu(ctx->sop_txd->txd.addr),
703 			ctx->sop_txd->dword[2], ctx->sop_txd->dword[3]);
704 		vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
705 
706 		/* use the right gen for non-SOP desc */
707 		dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
708 	}
709 
710 	/* linear part can use multiple tx desc if it's big */
711 	len = skb_headlen(skb) - ctx->copy_size;
712 	buf_offset = ctx->copy_size;
713 	while (len) {
714 		u32 buf_size;
715 
716 		if (len < VMXNET3_MAX_TX_BUF_SIZE) {
717 			buf_size = len;
718 			dw2 |= len;
719 		} else {
720 			buf_size = VMXNET3_MAX_TX_BUF_SIZE;
721 			/* spec says that for TxDesc.len, 0 == 2^14 */
722 		}
723 
724 		tbi = tq->buf_info + tq->tx_ring.next2fill;
725 		tbi->map_type = VMXNET3_MAP_SINGLE;
726 		tbi->dma_addr = dma_map_single(&adapter->pdev->dev,
727 				skb->data + buf_offset, buf_size,
728 				PCI_DMA_TODEVICE);
729 		if (dma_mapping_error(&adapter->pdev->dev, tbi->dma_addr))
730 			return -EFAULT;
731 
732 		tbi->len = buf_size;
733 
734 		gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
735 		BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
736 
737 		gdesc->txd.addr = cpu_to_le64(tbi->dma_addr);
738 		gdesc->dword[2] = cpu_to_le32(dw2);
739 		gdesc->dword[3] = 0;
740 
741 		netdev_dbg(adapter->netdev,
742 			"txd[%u]: 0x%Lx 0x%x 0x%x\n",
743 			tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr),
744 			le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]);
745 		vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
746 		dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
747 
748 		len -= buf_size;
749 		buf_offset += buf_size;
750 	}
751 
752 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
753 		const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
754 		u32 buf_size;
755 
756 		buf_offset = 0;
757 		len = skb_frag_size(frag);
758 		while (len) {
759 			tbi = tq->buf_info + tq->tx_ring.next2fill;
760 			if (len < VMXNET3_MAX_TX_BUF_SIZE) {
761 				buf_size = len;
762 				dw2 |= len;
763 			} else {
764 				buf_size = VMXNET3_MAX_TX_BUF_SIZE;
765 				/* spec says that for TxDesc.len, 0 == 2^14 */
766 			}
767 			tbi->map_type = VMXNET3_MAP_PAGE;
768 			tbi->dma_addr = skb_frag_dma_map(&adapter->pdev->dev, frag,
769 							 buf_offset, buf_size,
770 							 DMA_TO_DEVICE);
771 			if (dma_mapping_error(&adapter->pdev->dev, tbi->dma_addr))
772 				return -EFAULT;
773 
774 			tbi->len = buf_size;
775 
776 			gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
777 			BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
778 
779 			gdesc->txd.addr = cpu_to_le64(tbi->dma_addr);
780 			gdesc->dword[2] = cpu_to_le32(dw2);
781 			gdesc->dword[3] = 0;
782 
783 			netdev_dbg(adapter->netdev,
784 				"txd[%u]: 0x%llx %u %u\n",
785 				tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr),
786 				le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]);
787 			vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
788 			dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
789 
790 			len -= buf_size;
791 			buf_offset += buf_size;
792 		}
793 	}
794 
795 	ctx->eop_txd = gdesc;
796 
797 	/* set the last buf_info for the pkt */
798 	tbi->skb = skb;
799 	tbi->sop_idx = ctx->sop_txd - tq->tx_ring.base;
800 
801 	return 0;
802 }
803 
804 
805 /* Init all tx queues */
806 static void
807 vmxnet3_tq_init_all(struct vmxnet3_adapter *adapter)
808 {
809 	int i;
810 
811 	for (i = 0; i < adapter->num_tx_queues; i++)
812 		vmxnet3_tq_init(&adapter->tx_queue[i], adapter);
813 }
814 
815 
816 /*
817  *    parse relevant protocol headers:
818  *      For a tso pkt, relevant headers are L2/3/4 including options
819  *      For a pkt requesting csum offloading, they are L2/3 and may include L4
820  *      if it's a TCP/UDP pkt
821  *
822  * Returns:
823  *    -1:  error happens during parsing
824  *     0:  protocol headers parsed, but too big to be copied
825  *     1:  protocol headers parsed and copied
826  *
827  * Other effects:
828  *    1. related *ctx fields are updated.
829  *    2. ctx->copy_size is # of bytes copied
830  *    3. the portion to be copied is guaranteed to be in the linear part
831  *
832  */
833 static int
834 vmxnet3_parse_hdr(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
835 		  struct vmxnet3_tx_ctx *ctx,
836 		  struct vmxnet3_adapter *adapter)
837 {
838 	u8 protocol = 0;
839 
840 	if (ctx->mss) {	/* TSO */
841 		ctx->eth_ip_hdr_size = skb_transport_offset(skb);
842 		ctx->l4_hdr_size = tcp_hdrlen(skb);
843 		ctx->copy_size = ctx->eth_ip_hdr_size + ctx->l4_hdr_size;
844 	} else {
845 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
846 			ctx->eth_ip_hdr_size = skb_checksum_start_offset(skb);
847 
848 			if (ctx->ipv4) {
849 				const struct iphdr *iph = ip_hdr(skb);
850 
851 				protocol = iph->protocol;
852 			} else if (ctx->ipv6) {
853 				const struct ipv6hdr *ipv6h = ipv6_hdr(skb);
854 
855 				protocol = ipv6h->nexthdr;
856 			}
857 
858 			switch (protocol) {
859 			case IPPROTO_TCP:
860 				ctx->l4_hdr_size = tcp_hdrlen(skb);
861 				break;
862 			case IPPROTO_UDP:
863 				ctx->l4_hdr_size = sizeof(struct udphdr);
864 				break;
865 			default:
866 				ctx->l4_hdr_size = 0;
867 				break;
868 			}
869 
870 			ctx->copy_size = min(ctx->eth_ip_hdr_size +
871 					 ctx->l4_hdr_size, skb->len);
872 		} else {
873 			ctx->eth_ip_hdr_size = 0;
874 			ctx->l4_hdr_size = 0;
875 			/* copy as much as allowed */
876 			ctx->copy_size = min((unsigned int)VMXNET3_HDR_COPY_SIZE
877 					     , skb_headlen(skb));
878 		}
879 
880 		if (skb->len <= VMXNET3_HDR_COPY_SIZE)
881 			ctx->copy_size = skb->len;
882 
883 		/* make sure headers are accessible directly */
884 		if (unlikely(!pskb_may_pull(skb, ctx->copy_size)))
885 			goto err;
886 	}
887 
888 	if (unlikely(ctx->copy_size > VMXNET3_HDR_COPY_SIZE)) {
889 		tq->stats.oversized_hdr++;
890 		ctx->copy_size = 0;
891 		return 0;
892 	}
893 
894 	return 1;
895 err:
896 	return -1;
897 }
898 
899 /*
900  *    copy relevant protocol headers to the transmit ring:
901  *      For a tso pkt, relevant headers are L2/3/4 including options
902  *      For a pkt requesting csum offloading, they are L2/3 and may include L4
903  *      if it's a TCP/UDP pkt
904  *
905  *
906  *    Note that this requires that vmxnet3_parse_hdr be called first to set the
907  *      appropriate bits in ctx first
908  */
909 static void
910 vmxnet3_copy_hdr(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
911 		 struct vmxnet3_tx_ctx *ctx,
912 		 struct vmxnet3_adapter *adapter)
913 {
914 	struct Vmxnet3_TxDataDesc *tdd;
915 
916 	tdd = tq->data_ring.base + tq->tx_ring.next2fill;
917 
918 	memcpy(tdd->data, skb->data, ctx->copy_size);
919 	netdev_dbg(adapter->netdev,
920 		"copy %u bytes to dataRing[%u]\n",
921 		ctx->copy_size, tq->tx_ring.next2fill);
922 }
923 
924 
925 static void
926 vmxnet3_prepare_tso(struct sk_buff *skb,
927 		    struct vmxnet3_tx_ctx *ctx)
928 {
929 	struct tcphdr *tcph = tcp_hdr(skb);
930 
931 	if (ctx->ipv4) {
932 		struct iphdr *iph = ip_hdr(skb);
933 
934 		iph->check = 0;
935 		tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, 0,
936 						 IPPROTO_TCP, 0);
937 	} else if (ctx->ipv6) {
938 		struct ipv6hdr *iph = ipv6_hdr(skb);
939 
940 		tcph->check = ~csum_ipv6_magic(&iph->saddr, &iph->daddr, 0,
941 					       IPPROTO_TCP, 0);
942 	}
943 }
944 
945 static int txd_estimate(const struct sk_buff *skb)
946 {
947 	int count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) + 1;
948 	int i;
949 
950 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
951 		const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
952 
953 		count += VMXNET3_TXD_NEEDED(skb_frag_size(frag));
954 	}
955 	return count;
956 }
957 
958 /*
959  * Transmits a pkt thru a given tq
960  * Returns:
961  *    NETDEV_TX_OK:      descriptors are setup successfully
962  *    NETDEV_TX_OK:      error occurred, the pkt is dropped
963  *    NETDEV_TX_BUSY:    tx ring is full, queue is stopped
964  *
965  * Side-effects:
966  *    1. tx ring may be changed
967  *    2. tq stats may be updated accordingly
968  *    3. shared->txNumDeferred may be updated
969  */
970 
971 static int
972 vmxnet3_tq_xmit(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
973 		struct vmxnet3_adapter *adapter, struct net_device *netdev)
974 {
975 	int ret;
976 	u32 count;
977 	unsigned long flags;
978 	struct vmxnet3_tx_ctx ctx;
979 	union Vmxnet3_GenericDesc *gdesc;
980 #ifdef __BIG_ENDIAN_BITFIELD
981 	/* Use temporary descriptor to avoid touching bits multiple times */
982 	union Vmxnet3_GenericDesc tempTxDesc;
983 #endif
984 
985 	count = txd_estimate(skb);
986 
987 	ctx.ipv4 = (vlan_get_protocol(skb) == cpu_to_be16(ETH_P_IP));
988 	ctx.ipv6 = (vlan_get_protocol(skb) == cpu_to_be16(ETH_P_IPV6));
989 
990 	ctx.mss = skb_shinfo(skb)->gso_size;
991 	if (ctx.mss) {
992 		if (skb_header_cloned(skb)) {
993 			if (unlikely(pskb_expand_head(skb, 0, 0,
994 						      GFP_ATOMIC) != 0)) {
995 				tq->stats.drop_tso++;
996 				goto drop_pkt;
997 			}
998 			tq->stats.copy_skb_header++;
999 		}
1000 		vmxnet3_prepare_tso(skb, &ctx);
1001 	} else {
1002 		if (unlikely(count > VMXNET3_MAX_TXD_PER_PKT)) {
1003 
1004 			/* non-tso pkts must not use more than
1005 			 * VMXNET3_MAX_TXD_PER_PKT entries
1006 			 */
1007 			if (skb_linearize(skb) != 0) {
1008 				tq->stats.drop_too_many_frags++;
1009 				goto drop_pkt;
1010 			}
1011 			tq->stats.linearized++;
1012 
1013 			/* recalculate the # of descriptors to use */
1014 			count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) + 1;
1015 		}
1016 	}
1017 
1018 	ret = vmxnet3_parse_hdr(skb, tq, &ctx, adapter);
1019 	if (ret >= 0) {
1020 		BUG_ON(ret <= 0 && ctx.copy_size != 0);
1021 		/* hdrs parsed, check against other limits */
1022 		if (ctx.mss) {
1023 			if (unlikely(ctx.eth_ip_hdr_size + ctx.l4_hdr_size >
1024 				     VMXNET3_MAX_TX_BUF_SIZE)) {
1025 				tq->stats.drop_oversized_hdr++;
1026 				goto drop_pkt;
1027 			}
1028 		} else {
1029 			if (skb->ip_summed == CHECKSUM_PARTIAL) {
1030 				if (unlikely(ctx.eth_ip_hdr_size +
1031 					     skb->csum_offset >
1032 					     VMXNET3_MAX_CSUM_OFFSET)) {
1033 					tq->stats.drop_oversized_hdr++;
1034 					goto drop_pkt;
1035 				}
1036 			}
1037 		}
1038 	} else {
1039 		tq->stats.drop_hdr_inspect_err++;
1040 		goto drop_pkt;
1041 	}
1042 
1043 	spin_lock_irqsave(&tq->tx_lock, flags);
1044 
1045 	if (count > vmxnet3_cmd_ring_desc_avail(&tq->tx_ring)) {
1046 		tq->stats.tx_ring_full++;
1047 		netdev_dbg(adapter->netdev,
1048 			"tx queue stopped on %s, next2comp %u"
1049 			" next2fill %u\n", adapter->netdev->name,
1050 			tq->tx_ring.next2comp, tq->tx_ring.next2fill);
1051 
1052 		vmxnet3_tq_stop(tq, adapter);
1053 		spin_unlock_irqrestore(&tq->tx_lock, flags);
1054 		return NETDEV_TX_BUSY;
1055 	}
1056 
1057 
1058 	vmxnet3_copy_hdr(skb, tq, &ctx, adapter);
1059 
1060 	/* fill tx descs related to addr & len */
1061 	if (vmxnet3_map_pkt(skb, &ctx, tq, adapter->pdev, adapter))
1062 		goto unlock_drop_pkt;
1063 
1064 	/* setup the EOP desc */
1065 	ctx.eop_txd->dword[3] = cpu_to_le32(VMXNET3_TXD_CQ | VMXNET3_TXD_EOP);
1066 
1067 	/* setup the SOP desc */
1068 #ifdef __BIG_ENDIAN_BITFIELD
1069 	gdesc = &tempTxDesc;
1070 	gdesc->dword[2] = ctx.sop_txd->dword[2];
1071 	gdesc->dword[3] = ctx.sop_txd->dword[3];
1072 #else
1073 	gdesc = ctx.sop_txd;
1074 #endif
1075 	if (ctx.mss) {
1076 		gdesc->txd.hlen = ctx.eth_ip_hdr_size + ctx.l4_hdr_size;
1077 		gdesc->txd.om = VMXNET3_OM_TSO;
1078 		gdesc->txd.msscof = ctx.mss;
1079 		le32_add_cpu(&tq->shared->txNumDeferred, (skb->len -
1080 			     gdesc->txd.hlen + ctx.mss - 1) / ctx.mss);
1081 	} else {
1082 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
1083 			gdesc->txd.hlen = ctx.eth_ip_hdr_size;
1084 			gdesc->txd.om = VMXNET3_OM_CSUM;
1085 			gdesc->txd.msscof = ctx.eth_ip_hdr_size +
1086 					    skb->csum_offset;
1087 		} else {
1088 			gdesc->txd.om = 0;
1089 			gdesc->txd.msscof = 0;
1090 		}
1091 		le32_add_cpu(&tq->shared->txNumDeferred, 1);
1092 	}
1093 
1094 	if (skb_vlan_tag_present(skb)) {
1095 		gdesc->txd.ti = 1;
1096 		gdesc->txd.tci = skb_vlan_tag_get(skb);
1097 	}
1098 
1099 	/* finally flips the GEN bit of the SOP desc. */
1100 	gdesc->dword[2] = cpu_to_le32(le32_to_cpu(gdesc->dword[2]) ^
1101 						  VMXNET3_TXD_GEN);
1102 #ifdef __BIG_ENDIAN_BITFIELD
1103 	/* Finished updating in bitfields of Tx Desc, so write them in original
1104 	 * place.
1105 	 */
1106 	vmxnet3_TxDescToLe((struct Vmxnet3_TxDesc *)gdesc,
1107 			   (struct Vmxnet3_TxDesc *)ctx.sop_txd);
1108 	gdesc = ctx.sop_txd;
1109 #endif
1110 	netdev_dbg(adapter->netdev,
1111 		"txd[%u]: SOP 0x%Lx 0x%x 0x%x\n",
1112 		(u32)(ctx.sop_txd -
1113 		tq->tx_ring.base), le64_to_cpu(gdesc->txd.addr),
1114 		le32_to_cpu(gdesc->dword[2]), le32_to_cpu(gdesc->dword[3]));
1115 
1116 	spin_unlock_irqrestore(&tq->tx_lock, flags);
1117 
1118 	if (le32_to_cpu(tq->shared->txNumDeferred) >=
1119 					le32_to_cpu(tq->shared->txThreshold)) {
1120 		tq->shared->txNumDeferred = 0;
1121 		VMXNET3_WRITE_BAR0_REG(adapter,
1122 				       VMXNET3_REG_TXPROD + tq->qid * 8,
1123 				       tq->tx_ring.next2fill);
1124 	}
1125 
1126 	return NETDEV_TX_OK;
1127 
1128 unlock_drop_pkt:
1129 	spin_unlock_irqrestore(&tq->tx_lock, flags);
1130 drop_pkt:
1131 	tq->stats.drop_total++;
1132 	dev_kfree_skb_any(skb);
1133 	return NETDEV_TX_OK;
1134 }
1135 
1136 
1137 static netdev_tx_t
1138 vmxnet3_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
1139 {
1140 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1141 
1142 	BUG_ON(skb->queue_mapping > adapter->num_tx_queues);
1143 	return vmxnet3_tq_xmit(skb,
1144 			       &adapter->tx_queue[skb->queue_mapping],
1145 			       adapter, netdev);
1146 }
1147 
1148 
1149 static void
1150 vmxnet3_rx_csum(struct vmxnet3_adapter *adapter,
1151 		struct sk_buff *skb,
1152 		union Vmxnet3_GenericDesc *gdesc)
1153 {
1154 	if (!gdesc->rcd.cnc && adapter->netdev->features & NETIF_F_RXCSUM) {
1155 		/* typical case: TCP/UDP over IP and both csums are correct */
1156 		if ((le32_to_cpu(gdesc->dword[3]) & VMXNET3_RCD_CSUM_OK) ==
1157 							VMXNET3_RCD_CSUM_OK) {
1158 			skb->ip_summed = CHECKSUM_UNNECESSARY;
1159 			BUG_ON(!(gdesc->rcd.tcp || gdesc->rcd.udp));
1160 			BUG_ON(!(gdesc->rcd.v4  || gdesc->rcd.v6));
1161 			BUG_ON(gdesc->rcd.frg);
1162 		} else {
1163 			if (gdesc->rcd.csum) {
1164 				skb->csum = htons(gdesc->rcd.csum);
1165 				skb->ip_summed = CHECKSUM_PARTIAL;
1166 			} else {
1167 				skb_checksum_none_assert(skb);
1168 			}
1169 		}
1170 	} else {
1171 		skb_checksum_none_assert(skb);
1172 	}
1173 }
1174 
1175 
1176 static void
1177 vmxnet3_rx_error(struct vmxnet3_rx_queue *rq, struct Vmxnet3_RxCompDesc *rcd,
1178 		 struct vmxnet3_rx_ctx *ctx,  struct vmxnet3_adapter *adapter)
1179 {
1180 	rq->stats.drop_err++;
1181 	if (!rcd->fcs)
1182 		rq->stats.drop_fcs++;
1183 
1184 	rq->stats.drop_total++;
1185 
1186 	/*
1187 	 * We do not unmap and chain the rx buffer to the skb.
1188 	 * We basically pretend this buffer is not used and will be recycled
1189 	 * by vmxnet3_rq_alloc_rx_buf()
1190 	 */
1191 
1192 	/*
1193 	 * ctx->skb may be NULL if this is the first and the only one
1194 	 * desc for the pkt
1195 	 */
1196 	if (ctx->skb)
1197 		dev_kfree_skb_irq(ctx->skb);
1198 
1199 	ctx->skb = NULL;
1200 }
1201 
1202 
1203 static u32
1204 vmxnet3_get_hdr_len(struct vmxnet3_adapter *adapter, struct sk_buff *skb,
1205 		    union Vmxnet3_GenericDesc *gdesc)
1206 {
1207 	u32 hlen, maplen;
1208 	union {
1209 		void *ptr;
1210 		struct ethhdr *eth;
1211 		struct iphdr *ipv4;
1212 		struct ipv6hdr *ipv6;
1213 		struct tcphdr *tcp;
1214 	} hdr;
1215 	BUG_ON(gdesc->rcd.tcp == 0);
1216 
1217 	maplen = skb_headlen(skb);
1218 	if (unlikely(sizeof(struct iphdr) + sizeof(struct tcphdr) > maplen))
1219 		return 0;
1220 
1221 	hdr.eth = eth_hdr(skb);
1222 	if (gdesc->rcd.v4) {
1223 		BUG_ON(hdr.eth->h_proto != htons(ETH_P_IP));
1224 		hdr.ptr += sizeof(struct ethhdr);
1225 		BUG_ON(hdr.ipv4->protocol != IPPROTO_TCP);
1226 		hlen = hdr.ipv4->ihl << 2;
1227 		hdr.ptr += hdr.ipv4->ihl << 2;
1228 	} else if (gdesc->rcd.v6) {
1229 		BUG_ON(hdr.eth->h_proto != htons(ETH_P_IPV6));
1230 		hdr.ptr += sizeof(struct ethhdr);
1231 		/* Use an estimated value, since we also need to handle
1232 		 * TSO case.
1233 		 */
1234 		if (hdr.ipv6->nexthdr != IPPROTO_TCP)
1235 			return sizeof(struct ipv6hdr) + sizeof(struct tcphdr);
1236 		hlen = sizeof(struct ipv6hdr);
1237 		hdr.ptr += sizeof(struct ipv6hdr);
1238 	} else {
1239 		/* Non-IP pkt, dont estimate header length */
1240 		return 0;
1241 	}
1242 
1243 	if (hlen + sizeof(struct tcphdr) > maplen)
1244 		return 0;
1245 
1246 	return (hlen + (hdr.tcp->doff << 2));
1247 }
1248 
1249 static int
1250 vmxnet3_rq_rx_complete(struct vmxnet3_rx_queue *rq,
1251 		       struct vmxnet3_adapter *adapter, int quota)
1252 {
1253 	static const u32 rxprod_reg[2] = {
1254 		VMXNET3_REG_RXPROD, VMXNET3_REG_RXPROD2
1255 	};
1256 	u32 num_pkts = 0;
1257 	bool skip_page_frags = false;
1258 	struct Vmxnet3_RxCompDesc *rcd;
1259 	struct vmxnet3_rx_ctx *ctx = &rq->rx_ctx;
1260 	u16 segCnt = 0, mss = 0;
1261 #ifdef __BIG_ENDIAN_BITFIELD
1262 	struct Vmxnet3_RxDesc rxCmdDesc;
1263 	struct Vmxnet3_RxCompDesc rxComp;
1264 #endif
1265 	vmxnet3_getRxComp(rcd, &rq->comp_ring.base[rq->comp_ring.next2proc].rcd,
1266 			  &rxComp);
1267 	while (rcd->gen == rq->comp_ring.gen) {
1268 		struct vmxnet3_rx_buf_info *rbi;
1269 		struct sk_buff *skb, *new_skb = NULL;
1270 		struct page *new_page = NULL;
1271 		dma_addr_t new_dma_addr;
1272 		int num_to_alloc;
1273 		struct Vmxnet3_RxDesc *rxd;
1274 		u32 idx, ring_idx;
1275 		struct vmxnet3_cmd_ring	*ring = NULL;
1276 		if (num_pkts >= quota) {
1277 			/* we may stop even before we see the EOP desc of
1278 			 * the current pkt
1279 			 */
1280 			break;
1281 		}
1282 		BUG_ON(rcd->rqID != rq->qid && rcd->rqID != rq->qid2);
1283 		idx = rcd->rxdIdx;
1284 		ring_idx = rcd->rqID < adapter->num_rx_queues ? 0 : 1;
1285 		ring = rq->rx_ring + ring_idx;
1286 		vmxnet3_getRxDesc(rxd, &rq->rx_ring[ring_idx].base[idx].rxd,
1287 				  &rxCmdDesc);
1288 		rbi = rq->buf_info[ring_idx] + idx;
1289 
1290 		BUG_ON(rxd->addr != rbi->dma_addr ||
1291 		       rxd->len != rbi->len);
1292 
1293 		if (unlikely(rcd->eop && rcd->err)) {
1294 			vmxnet3_rx_error(rq, rcd, ctx, adapter);
1295 			goto rcd_done;
1296 		}
1297 
1298 		if (rcd->sop) { /* first buf of the pkt */
1299 			BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_HEAD ||
1300 			       rcd->rqID != rq->qid);
1301 
1302 			BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_SKB);
1303 			BUG_ON(ctx->skb != NULL || rbi->skb == NULL);
1304 
1305 			if (unlikely(rcd->len == 0)) {
1306 				/* Pretend the rx buffer is skipped. */
1307 				BUG_ON(!(rcd->sop && rcd->eop));
1308 				netdev_dbg(adapter->netdev,
1309 					"rxRing[%u][%u] 0 length\n",
1310 					ring_idx, idx);
1311 				goto rcd_done;
1312 			}
1313 
1314 			skip_page_frags = false;
1315 			ctx->skb = rbi->skb;
1316 			new_skb = netdev_alloc_skb_ip_align(adapter->netdev,
1317 							    rbi->len);
1318 			if (new_skb == NULL) {
1319 				/* Skb allocation failed, do not handover this
1320 				 * skb to stack. Reuse it. Drop the existing pkt
1321 				 */
1322 				rq->stats.rx_buf_alloc_failure++;
1323 				ctx->skb = NULL;
1324 				rq->stats.drop_total++;
1325 				skip_page_frags = true;
1326 				goto rcd_done;
1327 			}
1328 			new_dma_addr = dma_map_single(&adapter->pdev->dev,
1329 						      new_skb->data, rbi->len,
1330 						      PCI_DMA_FROMDEVICE);
1331 			if (dma_mapping_error(&adapter->pdev->dev,
1332 					      new_dma_addr)) {
1333 				dev_kfree_skb(new_skb);
1334 				/* Skb allocation failed, do not handover this
1335 				 * skb to stack. Reuse it. Drop the existing pkt
1336 				 */
1337 				rq->stats.rx_buf_alloc_failure++;
1338 				ctx->skb = NULL;
1339 				rq->stats.drop_total++;
1340 				skip_page_frags = true;
1341 				goto rcd_done;
1342 			}
1343 
1344 			dma_unmap_single(&adapter->pdev->dev, rbi->dma_addr,
1345 					 rbi->len,
1346 					 PCI_DMA_FROMDEVICE);
1347 
1348 #ifdef VMXNET3_RSS
1349 			if (rcd->rssType != VMXNET3_RCD_RSS_TYPE_NONE &&
1350 			    (adapter->netdev->features & NETIF_F_RXHASH))
1351 				skb_set_hash(ctx->skb,
1352 					     le32_to_cpu(rcd->rssHash),
1353 					     PKT_HASH_TYPE_L3);
1354 #endif
1355 			skb_put(ctx->skb, rcd->len);
1356 
1357 			/* Immediate refill */
1358 			rbi->skb = new_skb;
1359 			rbi->dma_addr = new_dma_addr;
1360 			rxd->addr = cpu_to_le64(rbi->dma_addr);
1361 			rxd->len = rbi->len;
1362 			if (adapter->version == 2 &&
1363 			    rcd->type == VMXNET3_CDTYPE_RXCOMP_LRO) {
1364 				struct Vmxnet3_RxCompDescExt *rcdlro;
1365 				rcdlro = (struct Vmxnet3_RxCompDescExt *)rcd;
1366 
1367 				segCnt = rcdlro->segCnt;
1368 				BUG_ON(segCnt <= 1);
1369 				mss = rcdlro->mss;
1370 				if (unlikely(segCnt <= 1))
1371 					segCnt = 0;
1372 			} else {
1373 				segCnt = 0;
1374 			}
1375 		} else {
1376 			BUG_ON(ctx->skb == NULL && !skip_page_frags);
1377 
1378 			/* non SOP buffer must be type 1 in most cases */
1379 			BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_PAGE);
1380 			BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_BODY);
1381 
1382 			/* If an sop buffer was dropped, skip all
1383 			 * following non-sop fragments. They will be reused.
1384 			 */
1385 			if (skip_page_frags)
1386 				goto rcd_done;
1387 
1388 			if (rcd->len) {
1389 				new_page = alloc_page(GFP_ATOMIC);
1390 				/* Replacement page frag could not be allocated.
1391 				 * Reuse this page. Drop the pkt and free the
1392 				 * skb which contained this page as a frag. Skip
1393 				 * processing all the following non-sop frags.
1394 				 */
1395 				if (unlikely(!new_page)) {
1396 					rq->stats.rx_buf_alloc_failure++;
1397 					dev_kfree_skb(ctx->skb);
1398 					ctx->skb = NULL;
1399 					skip_page_frags = true;
1400 					goto rcd_done;
1401 				}
1402 				new_dma_addr = dma_map_page(&adapter->pdev->dev,
1403 							    new_page,
1404 							    0, PAGE_SIZE,
1405 							    PCI_DMA_FROMDEVICE);
1406 				if (dma_mapping_error(&adapter->pdev->dev,
1407 						      new_dma_addr)) {
1408 					put_page(new_page);
1409 					rq->stats.rx_buf_alloc_failure++;
1410 					dev_kfree_skb(ctx->skb);
1411 					ctx->skb = NULL;
1412 					skip_page_frags = true;
1413 					goto rcd_done;
1414 				}
1415 
1416 				dma_unmap_page(&adapter->pdev->dev,
1417 					       rbi->dma_addr, rbi->len,
1418 					       PCI_DMA_FROMDEVICE);
1419 
1420 				vmxnet3_append_frag(ctx->skb, rcd, rbi);
1421 
1422 				/* Immediate refill */
1423 				rbi->page = new_page;
1424 				rbi->dma_addr = new_dma_addr;
1425 				rxd->addr = cpu_to_le64(rbi->dma_addr);
1426 				rxd->len = rbi->len;
1427 			}
1428 		}
1429 
1430 
1431 		skb = ctx->skb;
1432 		if (rcd->eop) {
1433 			u32 mtu = adapter->netdev->mtu;
1434 			skb->len += skb->data_len;
1435 
1436 			vmxnet3_rx_csum(adapter, skb,
1437 					(union Vmxnet3_GenericDesc *)rcd);
1438 			skb->protocol = eth_type_trans(skb, adapter->netdev);
1439 			if (!rcd->tcp || !adapter->lro)
1440 				goto not_lro;
1441 
1442 			if (segCnt != 0 && mss != 0) {
1443 				skb_shinfo(skb)->gso_type = rcd->v4 ?
1444 					SKB_GSO_TCPV4 : SKB_GSO_TCPV6;
1445 				skb_shinfo(skb)->gso_size = mss;
1446 				skb_shinfo(skb)->gso_segs = segCnt;
1447 			} else if (segCnt != 0 || skb->len > mtu) {
1448 				u32 hlen;
1449 
1450 				hlen = vmxnet3_get_hdr_len(adapter, skb,
1451 					(union Vmxnet3_GenericDesc *)rcd);
1452 				if (hlen == 0)
1453 					goto not_lro;
1454 
1455 				skb_shinfo(skb)->gso_type =
1456 					rcd->v4 ? SKB_GSO_TCPV4 : SKB_GSO_TCPV6;
1457 				if (segCnt != 0) {
1458 					skb_shinfo(skb)->gso_segs = segCnt;
1459 					skb_shinfo(skb)->gso_size =
1460 						DIV_ROUND_UP(skb->len -
1461 							hlen, segCnt);
1462 				} else {
1463 					skb_shinfo(skb)->gso_size = mtu - hlen;
1464 				}
1465 			}
1466 not_lro:
1467 			if (unlikely(rcd->ts))
1468 				__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), rcd->tci);
1469 
1470 			if (adapter->netdev->features & NETIF_F_LRO)
1471 				netif_receive_skb(skb);
1472 			else
1473 				napi_gro_receive(&rq->napi, skb);
1474 
1475 			ctx->skb = NULL;
1476 			num_pkts++;
1477 		}
1478 
1479 rcd_done:
1480 		/* device may have skipped some rx descs */
1481 		ring->next2comp = idx;
1482 		num_to_alloc = vmxnet3_cmd_ring_desc_avail(ring);
1483 		ring = rq->rx_ring + ring_idx;
1484 		while (num_to_alloc) {
1485 			vmxnet3_getRxDesc(rxd, &ring->base[ring->next2fill].rxd,
1486 					  &rxCmdDesc);
1487 			BUG_ON(!rxd->addr);
1488 
1489 			/* Recv desc is ready to be used by the device */
1490 			rxd->gen = ring->gen;
1491 			vmxnet3_cmd_ring_adv_next2fill(ring);
1492 			num_to_alloc--;
1493 		}
1494 
1495 		/* if needed, update the register */
1496 		if (unlikely(rq->shared->updateRxProd)) {
1497 			VMXNET3_WRITE_BAR0_REG(adapter,
1498 					       rxprod_reg[ring_idx] + rq->qid * 8,
1499 					       ring->next2fill);
1500 		}
1501 
1502 		vmxnet3_comp_ring_adv_next2proc(&rq->comp_ring);
1503 		vmxnet3_getRxComp(rcd,
1504 				  &rq->comp_ring.base[rq->comp_ring.next2proc].rcd, &rxComp);
1505 	}
1506 
1507 	return num_pkts;
1508 }
1509 
1510 
1511 static void
1512 vmxnet3_rq_cleanup(struct vmxnet3_rx_queue *rq,
1513 		   struct vmxnet3_adapter *adapter)
1514 {
1515 	u32 i, ring_idx;
1516 	struct Vmxnet3_RxDesc *rxd;
1517 
1518 	for (ring_idx = 0; ring_idx < 2; ring_idx++) {
1519 		for (i = 0; i < rq->rx_ring[ring_idx].size; i++) {
1520 #ifdef __BIG_ENDIAN_BITFIELD
1521 			struct Vmxnet3_RxDesc rxDesc;
1522 #endif
1523 			vmxnet3_getRxDesc(rxd,
1524 				&rq->rx_ring[ring_idx].base[i].rxd, &rxDesc);
1525 
1526 			if (rxd->btype == VMXNET3_RXD_BTYPE_HEAD &&
1527 					rq->buf_info[ring_idx][i].skb) {
1528 				dma_unmap_single(&adapter->pdev->dev, rxd->addr,
1529 						 rxd->len, PCI_DMA_FROMDEVICE);
1530 				dev_kfree_skb(rq->buf_info[ring_idx][i].skb);
1531 				rq->buf_info[ring_idx][i].skb = NULL;
1532 			} else if (rxd->btype == VMXNET3_RXD_BTYPE_BODY &&
1533 					rq->buf_info[ring_idx][i].page) {
1534 				dma_unmap_page(&adapter->pdev->dev, rxd->addr,
1535 					       rxd->len, PCI_DMA_FROMDEVICE);
1536 				put_page(rq->buf_info[ring_idx][i].page);
1537 				rq->buf_info[ring_idx][i].page = NULL;
1538 			}
1539 		}
1540 
1541 		rq->rx_ring[ring_idx].gen = VMXNET3_INIT_GEN;
1542 		rq->rx_ring[ring_idx].next2fill =
1543 					rq->rx_ring[ring_idx].next2comp = 0;
1544 	}
1545 
1546 	rq->comp_ring.gen = VMXNET3_INIT_GEN;
1547 	rq->comp_ring.next2proc = 0;
1548 }
1549 
1550 
1551 static void
1552 vmxnet3_rq_cleanup_all(struct vmxnet3_adapter *adapter)
1553 {
1554 	int i;
1555 
1556 	for (i = 0; i < adapter->num_rx_queues; i++)
1557 		vmxnet3_rq_cleanup(&adapter->rx_queue[i], adapter);
1558 }
1559 
1560 
1561 static void vmxnet3_rq_destroy(struct vmxnet3_rx_queue *rq,
1562 			       struct vmxnet3_adapter *adapter)
1563 {
1564 	int i;
1565 	int j;
1566 
1567 	/* all rx buffers must have already been freed */
1568 	for (i = 0; i < 2; i++) {
1569 		if (rq->buf_info[i]) {
1570 			for (j = 0; j < rq->rx_ring[i].size; j++)
1571 				BUG_ON(rq->buf_info[i][j].page != NULL);
1572 		}
1573 	}
1574 
1575 
1576 	for (i = 0; i < 2; i++) {
1577 		if (rq->rx_ring[i].base) {
1578 			dma_free_coherent(&adapter->pdev->dev,
1579 					  rq->rx_ring[i].size
1580 					  * sizeof(struct Vmxnet3_RxDesc),
1581 					  rq->rx_ring[i].base,
1582 					  rq->rx_ring[i].basePA);
1583 			rq->rx_ring[i].base = NULL;
1584 		}
1585 		rq->buf_info[i] = NULL;
1586 	}
1587 
1588 	if (rq->comp_ring.base) {
1589 		dma_free_coherent(&adapter->pdev->dev, rq->comp_ring.size
1590 				  * sizeof(struct Vmxnet3_RxCompDesc),
1591 				  rq->comp_ring.base, rq->comp_ring.basePA);
1592 		rq->comp_ring.base = NULL;
1593 	}
1594 
1595 	if (rq->buf_info[0]) {
1596 		size_t sz = sizeof(struct vmxnet3_rx_buf_info) *
1597 			(rq->rx_ring[0].size + rq->rx_ring[1].size);
1598 		dma_free_coherent(&adapter->pdev->dev, sz, rq->buf_info[0],
1599 				  rq->buf_info_pa);
1600 	}
1601 }
1602 
1603 
1604 static int
1605 vmxnet3_rq_init(struct vmxnet3_rx_queue *rq,
1606 		struct vmxnet3_adapter  *adapter)
1607 {
1608 	int i;
1609 
1610 	/* initialize buf_info */
1611 	for (i = 0; i < rq->rx_ring[0].size; i++) {
1612 
1613 		/* 1st buf for a pkt is skbuff */
1614 		if (i % adapter->rx_buf_per_pkt == 0) {
1615 			rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_SKB;
1616 			rq->buf_info[0][i].len = adapter->skb_buf_size;
1617 		} else { /* subsequent bufs for a pkt is frag */
1618 			rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_PAGE;
1619 			rq->buf_info[0][i].len = PAGE_SIZE;
1620 		}
1621 	}
1622 	for (i = 0; i < rq->rx_ring[1].size; i++) {
1623 		rq->buf_info[1][i].buf_type = VMXNET3_RX_BUF_PAGE;
1624 		rq->buf_info[1][i].len = PAGE_SIZE;
1625 	}
1626 
1627 	/* reset internal state and allocate buffers for both rings */
1628 	for (i = 0; i < 2; i++) {
1629 		rq->rx_ring[i].next2fill = rq->rx_ring[i].next2comp = 0;
1630 
1631 		memset(rq->rx_ring[i].base, 0, rq->rx_ring[i].size *
1632 		       sizeof(struct Vmxnet3_RxDesc));
1633 		rq->rx_ring[i].gen = VMXNET3_INIT_GEN;
1634 	}
1635 	if (vmxnet3_rq_alloc_rx_buf(rq, 0, rq->rx_ring[0].size - 1,
1636 				    adapter) == 0) {
1637 		/* at least has 1 rx buffer for the 1st ring */
1638 		return -ENOMEM;
1639 	}
1640 	vmxnet3_rq_alloc_rx_buf(rq, 1, rq->rx_ring[1].size - 1, adapter);
1641 
1642 	/* reset the comp ring */
1643 	rq->comp_ring.next2proc = 0;
1644 	memset(rq->comp_ring.base, 0, rq->comp_ring.size *
1645 	       sizeof(struct Vmxnet3_RxCompDesc));
1646 	rq->comp_ring.gen = VMXNET3_INIT_GEN;
1647 
1648 	/* reset rxctx */
1649 	rq->rx_ctx.skb = NULL;
1650 
1651 	/* stats are not reset */
1652 	return 0;
1653 }
1654 
1655 
1656 static int
1657 vmxnet3_rq_init_all(struct vmxnet3_adapter *adapter)
1658 {
1659 	int i, err = 0;
1660 
1661 	for (i = 0; i < adapter->num_rx_queues; i++) {
1662 		err = vmxnet3_rq_init(&adapter->rx_queue[i], adapter);
1663 		if (unlikely(err)) {
1664 			dev_err(&adapter->netdev->dev, "%s: failed to "
1665 				"initialize rx queue%i\n",
1666 				adapter->netdev->name, i);
1667 			break;
1668 		}
1669 	}
1670 	return err;
1671 
1672 }
1673 
1674 
1675 static int
1676 vmxnet3_rq_create(struct vmxnet3_rx_queue *rq, struct vmxnet3_adapter *adapter)
1677 {
1678 	int i;
1679 	size_t sz;
1680 	struct vmxnet3_rx_buf_info *bi;
1681 
1682 	for (i = 0; i < 2; i++) {
1683 
1684 		sz = rq->rx_ring[i].size * sizeof(struct Vmxnet3_RxDesc);
1685 		rq->rx_ring[i].base = dma_alloc_coherent(
1686 						&adapter->pdev->dev, sz,
1687 						&rq->rx_ring[i].basePA,
1688 						GFP_KERNEL);
1689 		if (!rq->rx_ring[i].base) {
1690 			netdev_err(adapter->netdev,
1691 				   "failed to allocate rx ring %d\n", i);
1692 			goto err;
1693 		}
1694 	}
1695 
1696 	sz = rq->comp_ring.size * sizeof(struct Vmxnet3_RxCompDesc);
1697 	rq->comp_ring.base = dma_alloc_coherent(&adapter->pdev->dev, sz,
1698 						&rq->comp_ring.basePA,
1699 						GFP_KERNEL);
1700 	if (!rq->comp_ring.base) {
1701 		netdev_err(adapter->netdev, "failed to allocate rx comp ring\n");
1702 		goto err;
1703 	}
1704 
1705 	sz = sizeof(struct vmxnet3_rx_buf_info) * (rq->rx_ring[0].size +
1706 						   rq->rx_ring[1].size);
1707 	bi = dma_zalloc_coherent(&adapter->pdev->dev, sz, &rq->buf_info_pa,
1708 				 GFP_KERNEL);
1709 	if (!bi)
1710 		goto err;
1711 
1712 	rq->buf_info[0] = bi;
1713 	rq->buf_info[1] = bi + rq->rx_ring[0].size;
1714 
1715 	return 0;
1716 
1717 err:
1718 	vmxnet3_rq_destroy(rq, adapter);
1719 	return -ENOMEM;
1720 }
1721 
1722 
1723 static int
1724 vmxnet3_rq_create_all(struct vmxnet3_adapter *adapter)
1725 {
1726 	int i, err = 0;
1727 
1728 	for (i = 0; i < adapter->num_rx_queues; i++) {
1729 		err = vmxnet3_rq_create(&adapter->rx_queue[i], adapter);
1730 		if (unlikely(err)) {
1731 			dev_err(&adapter->netdev->dev,
1732 				"%s: failed to create rx queue%i\n",
1733 				adapter->netdev->name, i);
1734 			goto err_out;
1735 		}
1736 	}
1737 	return err;
1738 err_out:
1739 	vmxnet3_rq_destroy_all(adapter);
1740 	return err;
1741 
1742 }
1743 
1744 /* Multiple queue aware polling function for tx and rx */
1745 
1746 static int
1747 vmxnet3_do_poll(struct vmxnet3_adapter *adapter, int budget)
1748 {
1749 	int rcd_done = 0, i;
1750 	if (unlikely(adapter->shared->ecr))
1751 		vmxnet3_process_events(adapter);
1752 	for (i = 0; i < adapter->num_tx_queues; i++)
1753 		vmxnet3_tq_tx_complete(&adapter->tx_queue[i], adapter);
1754 
1755 	for (i = 0; i < adapter->num_rx_queues; i++)
1756 		rcd_done += vmxnet3_rq_rx_complete(&adapter->rx_queue[i],
1757 						   adapter, budget);
1758 	return rcd_done;
1759 }
1760 
1761 
1762 static int
1763 vmxnet3_poll(struct napi_struct *napi, int budget)
1764 {
1765 	struct vmxnet3_rx_queue *rx_queue = container_of(napi,
1766 					  struct vmxnet3_rx_queue, napi);
1767 	int rxd_done;
1768 
1769 	rxd_done = vmxnet3_do_poll(rx_queue->adapter, budget);
1770 
1771 	if (rxd_done < budget) {
1772 		napi_complete(napi);
1773 		vmxnet3_enable_all_intrs(rx_queue->adapter);
1774 	}
1775 	return rxd_done;
1776 }
1777 
1778 /*
1779  * NAPI polling function for MSI-X mode with multiple Rx queues
1780  * Returns the # of the NAPI credit consumed (# of rx descriptors processed)
1781  */
1782 
1783 static int
1784 vmxnet3_poll_rx_only(struct napi_struct *napi, int budget)
1785 {
1786 	struct vmxnet3_rx_queue *rq = container_of(napi,
1787 						struct vmxnet3_rx_queue, napi);
1788 	struct vmxnet3_adapter *adapter = rq->adapter;
1789 	int rxd_done;
1790 
1791 	/* When sharing interrupt with corresponding tx queue, process
1792 	 * tx completions in that queue as well
1793 	 */
1794 	if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE) {
1795 		struct vmxnet3_tx_queue *tq =
1796 				&adapter->tx_queue[rq - adapter->rx_queue];
1797 		vmxnet3_tq_tx_complete(tq, adapter);
1798 	}
1799 
1800 	rxd_done = vmxnet3_rq_rx_complete(rq, adapter, budget);
1801 
1802 	if (rxd_done < budget) {
1803 		napi_complete(napi);
1804 		vmxnet3_enable_intr(adapter, rq->comp_ring.intr_idx);
1805 	}
1806 	return rxd_done;
1807 }
1808 
1809 
1810 #ifdef CONFIG_PCI_MSI
1811 
1812 /*
1813  * Handle completion interrupts on tx queues
1814  * Returns whether or not the intr is handled
1815  */
1816 
1817 static irqreturn_t
1818 vmxnet3_msix_tx(int irq, void *data)
1819 {
1820 	struct vmxnet3_tx_queue *tq = data;
1821 	struct vmxnet3_adapter *adapter = tq->adapter;
1822 
1823 	if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1824 		vmxnet3_disable_intr(adapter, tq->comp_ring.intr_idx);
1825 
1826 	/* Handle the case where only one irq is allocate for all tx queues */
1827 	if (adapter->share_intr == VMXNET3_INTR_TXSHARE) {
1828 		int i;
1829 		for (i = 0; i < adapter->num_tx_queues; i++) {
1830 			struct vmxnet3_tx_queue *txq = &adapter->tx_queue[i];
1831 			vmxnet3_tq_tx_complete(txq, adapter);
1832 		}
1833 	} else {
1834 		vmxnet3_tq_tx_complete(tq, adapter);
1835 	}
1836 	vmxnet3_enable_intr(adapter, tq->comp_ring.intr_idx);
1837 
1838 	return IRQ_HANDLED;
1839 }
1840 
1841 
1842 /*
1843  * Handle completion interrupts on rx queues. Returns whether or not the
1844  * intr is handled
1845  */
1846 
1847 static irqreturn_t
1848 vmxnet3_msix_rx(int irq, void *data)
1849 {
1850 	struct vmxnet3_rx_queue *rq = data;
1851 	struct vmxnet3_adapter *adapter = rq->adapter;
1852 
1853 	/* disable intr if needed */
1854 	if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1855 		vmxnet3_disable_intr(adapter, rq->comp_ring.intr_idx);
1856 	napi_schedule(&rq->napi);
1857 
1858 	return IRQ_HANDLED;
1859 }
1860 
1861 /*
1862  *----------------------------------------------------------------------------
1863  *
1864  * vmxnet3_msix_event --
1865  *
1866  *    vmxnet3 msix event intr handler
1867  *
1868  * Result:
1869  *    whether or not the intr is handled
1870  *
1871  *----------------------------------------------------------------------------
1872  */
1873 
1874 static irqreturn_t
1875 vmxnet3_msix_event(int irq, void *data)
1876 {
1877 	struct net_device *dev = data;
1878 	struct vmxnet3_adapter *adapter = netdev_priv(dev);
1879 
1880 	/* disable intr if needed */
1881 	if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1882 		vmxnet3_disable_intr(adapter, adapter->intr.event_intr_idx);
1883 
1884 	if (adapter->shared->ecr)
1885 		vmxnet3_process_events(adapter);
1886 
1887 	vmxnet3_enable_intr(adapter, adapter->intr.event_intr_idx);
1888 
1889 	return IRQ_HANDLED;
1890 }
1891 
1892 #endif /* CONFIG_PCI_MSI  */
1893 
1894 
1895 /* Interrupt handler for vmxnet3  */
1896 static irqreturn_t
1897 vmxnet3_intr(int irq, void *dev_id)
1898 {
1899 	struct net_device *dev = dev_id;
1900 	struct vmxnet3_adapter *adapter = netdev_priv(dev);
1901 
1902 	if (adapter->intr.type == VMXNET3_IT_INTX) {
1903 		u32 icr = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_ICR);
1904 		if (unlikely(icr == 0))
1905 			/* not ours */
1906 			return IRQ_NONE;
1907 	}
1908 
1909 
1910 	/* disable intr if needed */
1911 	if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1912 		vmxnet3_disable_all_intrs(adapter);
1913 
1914 	napi_schedule(&adapter->rx_queue[0].napi);
1915 
1916 	return IRQ_HANDLED;
1917 }
1918 
1919 #ifdef CONFIG_NET_POLL_CONTROLLER
1920 
1921 /* netpoll callback. */
1922 static void
1923 vmxnet3_netpoll(struct net_device *netdev)
1924 {
1925 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1926 
1927 	switch (adapter->intr.type) {
1928 #ifdef CONFIG_PCI_MSI
1929 	case VMXNET3_IT_MSIX: {
1930 		int i;
1931 		for (i = 0; i < adapter->num_rx_queues; i++)
1932 			vmxnet3_msix_rx(0, &adapter->rx_queue[i]);
1933 		break;
1934 	}
1935 #endif
1936 	case VMXNET3_IT_MSI:
1937 	default:
1938 		vmxnet3_intr(0, adapter->netdev);
1939 		break;
1940 	}
1941 
1942 }
1943 #endif	/* CONFIG_NET_POLL_CONTROLLER */
1944 
1945 static int
1946 vmxnet3_request_irqs(struct vmxnet3_adapter *adapter)
1947 {
1948 	struct vmxnet3_intr *intr = &adapter->intr;
1949 	int err = 0, i;
1950 	int vector = 0;
1951 
1952 #ifdef CONFIG_PCI_MSI
1953 	if (adapter->intr.type == VMXNET3_IT_MSIX) {
1954 		for (i = 0; i < adapter->num_tx_queues; i++) {
1955 			if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE) {
1956 				sprintf(adapter->tx_queue[i].name, "%s-tx-%d",
1957 					adapter->netdev->name, vector);
1958 				err = request_irq(
1959 					      intr->msix_entries[vector].vector,
1960 					      vmxnet3_msix_tx, 0,
1961 					      adapter->tx_queue[i].name,
1962 					      &adapter->tx_queue[i]);
1963 			} else {
1964 				sprintf(adapter->tx_queue[i].name, "%s-rxtx-%d",
1965 					adapter->netdev->name, vector);
1966 			}
1967 			if (err) {
1968 				dev_err(&adapter->netdev->dev,
1969 					"Failed to request irq for MSIX, %s, "
1970 					"error %d\n",
1971 					adapter->tx_queue[i].name, err);
1972 				return err;
1973 			}
1974 
1975 			/* Handle the case where only 1 MSIx was allocated for
1976 			 * all tx queues */
1977 			if (adapter->share_intr == VMXNET3_INTR_TXSHARE) {
1978 				for (; i < adapter->num_tx_queues; i++)
1979 					adapter->tx_queue[i].comp_ring.intr_idx
1980 								= vector;
1981 				vector++;
1982 				break;
1983 			} else {
1984 				adapter->tx_queue[i].comp_ring.intr_idx
1985 								= vector++;
1986 			}
1987 		}
1988 		if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE)
1989 			vector = 0;
1990 
1991 		for (i = 0; i < adapter->num_rx_queues; i++) {
1992 			if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE)
1993 				sprintf(adapter->rx_queue[i].name, "%s-rx-%d",
1994 					adapter->netdev->name, vector);
1995 			else
1996 				sprintf(adapter->rx_queue[i].name, "%s-rxtx-%d",
1997 					adapter->netdev->name, vector);
1998 			err = request_irq(intr->msix_entries[vector].vector,
1999 					  vmxnet3_msix_rx, 0,
2000 					  adapter->rx_queue[i].name,
2001 					  &(adapter->rx_queue[i]));
2002 			if (err) {
2003 				netdev_err(adapter->netdev,
2004 					   "Failed to request irq for MSIX, "
2005 					   "%s, error %d\n",
2006 					   adapter->rx_queue[i].name, err);
2007 				return err;
2008 			}
2009 
2010 			adapter->rx_queue[i].comp_ring.intr_idx = vector++;
2011 		}
2012 
2013 		sprintf(intr->event_msi_vector_name, "%s-event-%d",
2014 			adapter->netdev->name, vector);
2015 		err = request_irq(intr->msix_entries[vector].vector,
2016 				  vmxnet3_msix_event, 0,
2017 				  intr->event_msi_vector_name, adapter->netdev);
2018 		intr->event_intr_idx = vector;
2019 
2020 	} else if (intr->type == VMXNET3_IT_MSI) {
2021 		adapter->num_rx_queues = 1;
2022 		err = request_irq(adapter->pdev->irq, vmxnet3_intr, 0,
2023 				  adapter->netdev->name, adapter->netdev);
2024 	} else {
2025 #endif
2026 		adapter->num_rx_queues = 1;
2027 		err = request_irq(adapter->pdev->irq, vmxnet3_intr,
2028 				  IRQF_SHARED, adapter->netdev->name,
2029 				  adapter->netdev);
2030 #ifdef CONFIG_PCI_MSI
2031 	}
2032 #endif
2033 	intr->num_intrs = vector + 1;
2034 	if (err) {
2035 		netdev_err(adapter->netdev,
2036 			   "Failed to request irq (intr type:%d), error %d\n",
2037 			   intr->type, err);
2038 	} else {
2039 		/* Number of rx queues will not change after this */
2040 		for (i = 0; i < adapter->num_rx_queues; i++) {
2041 			struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
2042 			rq->qid = i;
2043 			rq->qid2 = i + adapter->num_rx_queues;
2044 		}
2045 
2046 
2047 
2048 		/* init our intr settings */
2049 		for (i = 0; i < intr->num_intrs; i++)
2050 			intr->mod_levels[i] = UPT1_IML_ADAPTIVE;
2051 		if (adapter->intr.type != VMXNET3_IT_MSIX) {
2052 			adapter->intr.event_intr_idx = 0;
2053 			for (i = 0; i < adapter->num_tx_queues; i++)
2054 				adapter->tx_queue[i].comp_ring.intr_idx = 0;
2055 			adapter->rx_queue[0].comp_ring.intr_idx = 0;
2056 		}
2057 
2058 		netdev_info(adapter->netdev,
2059 			    "intr type %u, mode %u, %u vectors allocated\n",
2060 			    intr->type, intr->mask_mode, intr->num_intrs);
2061 	}
2062 
2063 	return err;
2064 }
2065 
2066 
2067 static void
2068 vmxnet3_free_irqs(struct vmxnet3_adapter *adapter)
2069 {
2070 	struct vmxnet3_intr *intr = &adapter->intr;
2071 	BUG_ON(intr->type == VMXNET3_IT_AUTO || intr->num_intrs <= 0);
2072 
2073 	switch (intr->type) {
2074 #ifdef CONFIG_PCI_MSI
2075 	case VMXNET3_IT_MSIX:
2076 	{
2077 		int i, vector = 0;
2078 
2079 		if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE) {
2080 			for (i = 0; i < adapter->num_tx_queues; i++) {
2081 				free_irq(intr->msix_entries[vector++].vector,
2082 					 &(adapter->tx_queue[i]));
2083 				if (adapter->share_intr == VMXNET3_INTR_TXSHARE)
2084 					break;
2085 			}
2086 		}
2087 
2088 		for (i = 0; i < adapter->num_rx_queues; i++) {
2089 			free_irq(intr->msix_entries[vector++].vector,
2090 				 &(adapter->rx_queue[i]));
2091 		}
2092 
2093 		free_irq(intr->msix_entries[vector].vector,
2094 			 adapter->netdev);
2095 		BUG_ON(vector >= intr->num_intrs);
2096 		break;
2097 	}
2098 #endif
2099 	case VMXNET3_IT_MSI:
2100 		free_irq(adapter->pdev->irq, adapter->netdev);
2101 		break;
2102 	case VMXNET3_IT_INTX:
2103 		free_irq(adapter->pdev->irq, adapter->netdev);
2104 		break;
2105 	default:
2106 		BUG();
2107 	}
2108 }
2109 
2110 
2111 static void
2112 vmxnet3_restore_vlan(struct vmxnet3_adapter *adapter)
2113 {
2114 	u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
2115 	u16 vid;
2116 
2117 	/* allow untagged pkts */
2118 	VMXNET3_SET_VFTABLE_ENTRY(vfTable, 0);
2119 
2120 	for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
2121 		VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid);
2122 }
2123 
2124 
2125 static int
2126 vmxnet3_vlan_rx_add_vid(struct net_device *netdev, __be16 proto, u16 vid)
2127 {
2128 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2129 
2130 	if (!(netdev->flags & IFF_PROMISC)) {
2131 		u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
2132 		unsigned long flags;
2133 
2134 		VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid);
2135 		spin_lock_irqsave(&adapter->cmd_lock, flags);
2136 		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2137 				       VMXNET3_CMD_UPDATE_VLAN_FILTERS);
2138 		spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2139 	}
2140 
2141 	set_bit(vid, adapter->active_vlans);
2142 
2143 	return 0;
2144 }
2145 
2146 
2147 static int
2148 vmxnet3_vlan_rx_kill_vid(struct net_device *netdev, __be16 proto, u16 vid)
2149 {
2150 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2151 
2152 	if (!(netdev->flags & IFF_PROMISC)) {
2153 		u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
2154 		unsigned long flags;
2155 
2156 		VMXNET3_CLEAR_VFTABLE_ENTRY(vfTable, vid);
2157 		spin_lock_irqsave(&adapter->cmd_lock, flags);
2158 		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2159 				       VMXNET3_CMD_UPDATE_VLAN_FILTERS);
2160 		spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2161 	}
2162 
2163 	clear_bit(vid, adapter->active_vlans);
2164 
2165 	return 0;
2166 }
2167 
2168 
2169 static u8 *
2170 vmxnet3_copy_mc(struct net_device *netdev)
2171 {
2172 	u8 *buf = NULL;
2173 	u32 sz = netdev_mc_count(netdev) * ETH_ALEN;
2174 
2175 	/* struct Vmxnet3_RxFilterConf.mfTableLen is u16. */
2176 	if (sz <= 0xffff) {
2177 		/* We may be called with BH disabled */
2178 		buf = kmalloc(sz, GFP_ATOMIC);
2179 		if (buf) {
2180 			struct netdev_hw_addr *ha;
2181 			int i = 0;
2182 
2183 			netdev_for_each_mc_addr(ha, netdev)
2184 				memcpy(buf + i++ * ETH_ALEN, ha->addr,
2185 				       ETH_ALEN);
2186 		}
2187 	}
2188 	return buf;
2189 }
2190 
2191 
2192 static void
2193 vmxnet3_set_mc(struct net_device *netdev)
2194 {
2195 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2196 	unsigned long flags;
2197 	struct Vmxnet3_RxFilterConf *rxConf =
2198 					&adapter->shared->devRead.rxFilterConf;
2199 	u8 *new_table = NULL;
2200 	dma_addr_t new_table_pa = 0;
2201 	u32 new_mode = VMXNET3_RXM_UCAST;
2202 
2203 	if (netdev->flags & IFF_PROMISC) {
2204 		u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
2205 		memset(vfTable, 0, VMXNET3_VFT_SIZE * sizeof(*vfTable));
2206 
2207 		new_mode |= VMXNET3_RXM_PROMISC;
2208 	} else {
2209 		vmxnet3_restore_vlan(adapter);
2210 	}
2211 
2212 	if (netdev->flags & IFF_BROADCAST)
2213 		new_mode |= VMXNET3_RXM_BCAST;
2214 
2215 	if (netdev->flags & IFF_ALLMULTI)
2216 		new_mode |= VMXNET3_RXM_ALL_MULTI;
2217 	else
2218 		if (!netdev_mc_empty(netdev)) {
2219 			new_table = vmxnet3_copy_mc(netdev);
2220 			if (new_table) {
2221 				size_t sz = netdev_mc_count(netdev) * ETH_ALEN;
2222 
2223 				rxConf->mfTableLen = cpu_to_le16(sz);
2224 				new_table_pa = dma_map_single(
2225 							&adapter->pdev->dev,
2226 							new_table,
2227 							sz,
2228 							PCI_DMA_TODEVICE);
2229 			}
2230 
2231 			if (!dma_mapping_error(&adapter->pdev->dev,
2232 					       new_table_pa)) {
2233 				new_mode |= VMXNET3_RXM_MCAST;
2234 				rxConf->mfTablePA = cpu_to_le64(new_table_pa);
2235 			} else {
2236 				netdev_info(netdev,
2237 					    "failed to copy mcast list, setting ALL_MULTI\n");
2238 				new_mode |= VMXNET3_RXM_ALL_MULTI;
2239 			}
2240 		}
2241 
2242 	if (!(new_mode & VMXNET3_RXM_MCAST)) {
2243 		rxConf->mfTableLen = 0;
2244 		rxConf->mfTablePA = 0;
2245 	}
2246 
2247 	spin_lock_irqsave(&adapter->cmd_lock, flags);
2248 	if (new_mode != rxConf->rxMode) {
2249 		rxConf->rxMode = cpu_to_le32(new_mode);
2250 		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2251 				       VMXNET3_CMD_UPDATE_RX_MODE);
2252 		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2253 				       VMXNET3_CMD_UPDATE_VLAN_FILTERS);
2254 	}
2255 
2256 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2257 			       VMXNET3_CMD_UPDATE_MAC_FILTERS);
2258 	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2259 
2260 	if (new_table_pa)
2261 		dma_unmap_single(&adapter->pdev->dev, new_table_pa,
2262 				 rxConf->mfTableLen, PCI_DMA_TODEVICE);
2263 	kfree(new_table);
2264 }
2265 
2266 void
2267 vmxnet3_rq_destroy_all(struct vmxnet3_adapter *adapter)
2268 {
2269 	int i;
2270 
2271 	for (i = 0; i < adapter->num_rx_queues; i++)
2272 		vmxnet3_rq_destroy(&adapter->rx_queue[i], adapter);
2273 }
2274 
2275 
2276 /*
2277  *   Set up driver_shared based on settings in adapter.
2278  */
2279 
2280 static void
2281 vmxnet3_setup_driver_shared(struct vmxnet3_adapter *adapter)
2282 {
2283 	struct Vmxnet3_DriverShared *shared = adapter->shared;
2284 	struct Vmxnet3_DSDevRead *devRead = &shared->devRead;
2285 	struct Vmxnet3_TxQueueConf *tqc;
2286 	struct Vmxnet3_RxQueueConf *rqc;
2287 	int i;
2288 
2289 	memset(shared, 0, sizeof(*shared));
2290 
2291 	/* driver settings */
2292 	shared->magic = cpu_to_le32(VMXNET3_REV1_MAGIC);
2293 	devRead->misc.driverInfo.version = cpu_to_le32(
2294 						VMXNET3_DRIVER_VERSION_NUM);
2295 	devRead->misc.driverInfo.gos.gosBits = (sizeof(void *) == 4 ?
2296 				VMXNET3_GOS_BITS_32 : VMXNET3_GOS_BITS_64);
2297 	devRead->misc.driverInfo.gos.gosType = VMXNET3_GOS_TYPE_LINUX;
2298 	*((u32 *)&devRead->misc.driverInfo.gos) = cpu_to_le32(
2299 				*((u32 *)&devRead->misc.driverInfo.gos));
2300 	devRead->misc.driverInfo.vmxnet3RevSpt = cpu_to_le32(1);
2301 	devRead->misc.driverInfo.uptVerSpt = cpu_to_le32(1);
2302 
2303 	devRead->misc.ddPA = cpu_to_le64(adapter->adapter_pa);
2304 	devRead->misc.ddLen = cpu_to_le32(sizeof(struct vmxnet3_adapter));
2305 
2306 	/* set up feature flags */
2307 	if (adapter->netdev->features & NETIF_F_RXCSUM)
2308 		devRead->misc.uptFeatures |= UPT1_F_RXCSUM;
2309 
2310 	if (adapter->netdev->features & NETIF_F_LRO) {
2311 		devRead->misc.uptFeatures |= UPT1_F_LRO;
2312 		devRead->misc.maxNumRxSG = cpu_to_le16(1 + MAX_SKB_FRAGS);
2313 	}
2314 	if (adapter->netdev->features & NETIF_F_HW_VLAN_CTAG_RX)
2315 		devRead->misc.uptFeatures |= UPT1_F_RXVLAN;
2316 
2317 	devRead->misc.mtu = cpu_to_le32(adapter->netdev->mtu);
2318 	devRead->misc.queueDescPA = cpu_to_le64(adapter->queue_desc_pa);
2319 	devRead->misc.queueDescLen = cpu_to_le32(
2320 		adapter->num_tx_queues * sizeof(struct Vmxnet3_TxQueueDesc) +
2321 		adapter->num_rx_queues * sizeof(struct Vmxnet3_RxQueueDesc));
2322 
2323 	/* tx queue settings */
2324 	devRead->misc.numTxQueues =  adapter->num_tx_queues;
2325 	for (i = 0; i < adapter->num_tx_queues; i++) {
2326 		struct vmxnet3_tx_queue	*tq = &adapter->tx_queue[i];
2327 		BUG_ON(adapter->tx_queue[i].tx_ring.base == NULL);
2328 		tqc = &adapter->tqd_start[i].conf;
2329 		tqc->txRingBasePA   = cpu_to_le64(tq->tx_ring.basePA);
2330 		tqc->dataRingBasePA = cpu_to_le64(tq->data_ring.basePA);
2331 		tqc->compRingBasePA = cpu_to_le64(tq->comp_ring.basePA);
2332 		tqc->ddPA           = cpu_to_le64(tq->buf_info_pa);
2333 		tqc->txRingSize     = cpu_to_le32(tq->tx_ring.size);
2334 		tqc->dataRingSize   = cpu_to_le32(tq->data_ring.size);
2335 		tqc->compRingSize   = cpu_to_le32(tq->comp_ring.size);
2336 		tqc->ddLen          = cpu_to_le32(
2337 					sizeof(struct vmxnet3_tx_buf_info) *
2338 					tqc->txRingSize);
2339 		tqc->intrIdx        = tq->comp_ring.intr_idx;
2340 	}
2341 
2342 	/* rx queue settings */
2343 	devRead->misc.numRxQueues = adapter->num_rx_queues;
2344 	for (i = 0; i < adapter->num_rx_queues; i++) {
2345 		struct vmxnet3_rx_queue	*rq = &adapter->rx_queue[i];
2346 		rqc = &adapter->rqd_start[i].conf;
2347 		rqc->rxRingBasePA[0] = cpu_to_le64(rq->rx_ring[0].basePA);
2348 		rqc->rxRingBasePA[1] = cpu_to_le64(rq->rx_ring[1].basePA);
2349 		rqc->compRingBasePA  = cpu_to_le64(rq->comp_ring.basePA);
2350 		rqc->ddPA            = cpu_to_le64(rq->buf_info_pa);
2351 		rqc->rxRingSize[0]   = cpu_to_le32(rq->rx_ring[0].size);
2352 		rqc->rxRingSize[1]   = cpu_to_le32(rq->rx_ring[1].size);
2353 		rqc->compRingSize    = cpu_to_le32(rq->comp_ring.size);
2354 		rqc->ddLen           = cpu_to_le32(
2355 					sizeof(struct vmxnet3_rx_buf_info) *
2356 					(rqc->rxRingSize[0] +
2357 					 rqc->rxRingSize[1]));
2358 		rqc->intrIdx         = rq->comp_ring.intr_idx;
2359 	}
2360 
2361 #ifdef VMXNET3_RSS
2362 	memset(adapter->rss_conf, 0, sizeof(*adapter->rss_conf));
2363 
2364 	if (adapter->rss) {
2365 		struct UPT1_RSSConf *rssConf = adapter->rss_conf;
2366 
2367 		devRead->misc.uptFeatures |= UPT1_F_RSS;
2368 		devRead->misc.numRxQueues = adapter->num_rx_queues;
2369 		rssConf->hashType = UPT1_RSS_HASH_TYPE_TCP_IPV4 |
2370 				    UPT1_RSS_HASH_TYPE_IPV4 |
2371 				    UPT1_RSS_HASH_TYPE_TCP_IPV6 |
2372 				    UPT1_RSS_HASH_TYPE_IPV6;
2373 		rssConf->hashFunc = UPT1_RSS_HASH_FUNC_TOEPLITZ;
2374 		rssConf->hashKeySize = UPT1_RSS_MAX_KEY_SIZE;
2375 		rssConf->indTableSize = VMXNET3_RSS_IND_TABLE_SIZE;
2376 		netdev_rss_key_fill(rssConf->hashKey, sizeof(rssConf->hashKey));
2377 
2378 		for (i = 0; i < rssConf->indTableSize; i++)
2379 			rssConf->indTable[i] = ethtool_rxfh_indir_default(
2380 				i, adapter->num_rx_queues);
2381 
2382 		devRead->rssConfDesc.confVer = 1;
2383 		devRead->rssConfDesc.confLen = cpu_to_le32(sizeof(*rssConf));
2384 		devRead->rssConfDesc.confPA =
2385 			cpu_to_le64(adapter->rss_conf_pa);
2386 	}
2387 
2388 #endif /* VMXNET3_RSS */
2389 
2390 	/* intr settings */
2391 	devRead->intrConf.autoMask = adapter->intr.mask_mode ==
2392 				     VMXNET3_IMM_AUTO;
2393 	devRead->intrConf.numIntrs = adapter->intr.num_intrs;
2394 	for (i = 0; i < adapter->intr.num_intrs; i++)
2395 		devRead->intrConf.modLevels[i] = adapter->intr.mod_levels[i];
2396 
2397 	devRead->intrConf.eventIntrIdx = adapter->intr.event_intr_idx;
2398 	devRead->intrConf.intrCtrl |= cpu_to_le32(VMXNET3_IC_DISABLE_ALL);
2399 
2400 	/* rx filter settings */
2401 	devRead->rxFilterConf.rxMode = 0;
2402 	vmxnet3_restore_vlan(adapter);
2403 	vmxnet3_write_mac_addr(adapter, adapter->netdev->dev_addr);
2404 
2405 	/* the rest are already zeroed */
2406 }
2407 
2408 
2409 int
2410 vmxnet3_activate_dev(struct vmxnet3_adapter *adapter)
2411 {
2412 	int err, i;
2413 	u32 ret;
2414 	unsigned long flags;
2415 
2416 	netdev_dbg(adapter->netdev, "%s: skb_buf_size %d, rx_buf_per_pkt %d,"
2417 		" ring sizes %u %u %u\n", adapter->netdev->name,
2418 		adapter->skb_buf_size, adapter->rx_buf_per_pkt,
2419 		adapter->tx_queue[0].tx_ring.size,
2420 		adapter->rx_queue[0].rx_ring[0].size,
2421 		adapter->rx_queue[0].rx_ring[1].size);
2422 
2423 	vmxnet3_tq_init_all(adapter);
2424 	err = vmxnet3_rq_init_all(adapter);
2425 	if (err) {
2426 		netdev_err(adapter->netdev,
2427 			   "Failed to init rx queue error %d\n", err);
2428 		goto rq_err;
2429 	}
2430 
2431 	err = vmxnet3_request_irqs(adapter);
2432 	if (err) {
2433 		netdev_err(adapter->netdev,
2434 			   "Failed to setup irq for error %d\n", err);
2435 		goto irq_err;
2436 	}
2437 
2438 	vmxnet3_setup_driver_shared(adapter);
2439 
2440 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, VMXNET3_GET_ADDR_LO(
2441 			       adapter->shared_pa));
2442 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, VMXNET3_GET_ADDR_HI(
2443 			       adapter->shared_pa));
2444 	spin_lock_irqsave(&adapter->cmd_lock, flags);
2445 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2446 			       VMXNET3_CMD_ACTIVATE_DEV);
2447 	ret = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
2448 	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2449 
2450 	if (ret != 0) {
2451 		netdev_err(adapter->netdev,
2452 			   "Failed to activate dev: error %u\n", ret);
2453 		err = -EINVAL;
2454 		goto activate_err;
2455 	}
2456 
2457 	for (i = 0; i < adapter->num_rx_queues; i++) {
2458 		VMXNET3_WRITE_BAR0_REG(adapter,
2459 				VMXNET3_REG_RXPROD + i * VMXNET3_REG_ALIGN,
2460 				adapter->rx_queue[i].rx_ring[0].next2fill);
2461 		VMXNET3_WRITE_BAR0_REG(adapter, (VMXNET3_REG_RXPROD2 +
2462 				(i * VMXNET3_REG_ALIGN)),
2463 				adapter->rx_queue[i].rx_ring[1].next2fill);
2464 	}
2465 
2466 	/* Apply the rx filter settins last. */
2467 	vmxnet3_set_mc(adapter->netdev);
2468 
2469 	/*
2470 	 * Check link state when first activating device. It will start the
2471 	 * tx queue if the link is up.
2472 	 */
2473 	vmxnet3_check_link(adapter, true);
2474 	for (i = 0; i < adapter->num_rx_queues; i++)
2475 		napi_enable(&adapter->rx_queue[i].napi);
2476 	vmxnet3_enable_all_intrs(adapter);
2477 	clear_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
2478 	return 0;
2479 
2480 activate_err:
2481 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, 0);
2482 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, 0);
2483 	vmxnet3_free_irqs(adapter);
2484 irq_err:
2485 rq_err:
2486 	/* free up buffers we allocated */
2487 	vmxnet3_rq_cleanup_all(adapter);
2488 	return err;
2489 }
2490 
2491 
2492 void
2493 vmxnet3_reset_dev(struct vmxnet3_adapter *adapter)
2494 {
2495 	unsigned long flags;
2496 	spin_lock_irqsave(&adapter->cmd_lock, flags);
2497 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, VMXNET3_CMD_RESET_DEV);
2498 	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2499 }
2500 
2501 
2502 int
2503 vmxnet3_quiesce_dev(struct vmxnet3_adapter *adapter)
2504 {
2505 	int i;
2506 	unsigned long flags;
2507 	if (test_and_set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state))
2508 		return 0;
2509 
2510 
2511 	spin_lock_irqsave(&adapter->cmd_lock, flags);
2512 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2513 			       VMXNET3_CMD_QUIESCE_DEV);
2514 	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2515 	vmxnet3_disable_all_intrs(adapter);
2516 
2517 	for (i = 0; i < adapter->num_rx_queues; i++)
2518 		napi_disable(&adapter->rx_queue[i].napi);
2519 	netif_tx_disable(adapter->netdev);
2520 	adapter->link_speed = 0;
2521 	netif_carrier_off(adapter->netdev);
2522 
2523 	vmxnet3_tq_cleanup_all(adapter);
2524 	vmxnet3_rq_cleanup_all(adapter);
2525 	vmxnet3_free_irqs(adapter);
2526 	return 0;
2527 }
2528 
2529 
2530 static void
2531 vmxnet3_write_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac)
2532 {
2533 	u32 tmp;
2534 
2535 	tmp = *(u32 *)mac;
2536 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACL, tmp);
2537 
2538 	tmp = (mac[5] << 8) | mac[4];
2539 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACH, tmp);
2540 }
2541 
2542 
2543 static int
2544 vmxnet3_set_mac_addr(struct net_device *netdev, void *p)
2545 {
2546 	struct sockaddr *addr = p;
2547 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2548 
2549 	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2550 	vmxnet3_write_mac_addr(adapter, addr->sa_data);
2551 
2552 	return 0;
2553 }
2554 
2555 
2556 /* ==================== initialization and cleanup routines ============ */
2557 
2558 static int
2559 vmxnet3_alloc_pci_resources(struct vmxnet3_adapter *adapter, bool *dma64)
2560 {
2561 	int err;
2562 	unsigned long mmio_start, mmio_len;
2563 	struct pci_dev *pdev = adapter->pdev;
2564 
2565 	err = pci_enable_device(pdev);
2566 	if (err) {
2567 		dev_err(&pdev->dev, "Failed to enable adapter: error %d\n", err);
2568 		return err;
2569 	}
2570 
2571 	if (pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) == 0) {
2572 		if (pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64)) != 0) {
2573 			dev_err(&pdev->dev,
2574 				"pci_set_consistent_dma_mask failed\n");
2575 			err = -EIO;
2576 			goto err_set_mask;
2577 		}
2578 		*dma64 = true;
2579 	} else {
2580 		if (pci_set_dma_mask(pdev, DMA_BIT_MASK(32)) != 0) {
2581 			dev_err(&pdev->dev,
2582 				"pci_set_dma_mask failed\n");
2583 			err = -EIO;
2584 			goto err_set_mask;
2585 		}
2586 		*dma64 = false;
2587 	}
2588 
2589 	err = pci_request_selected_regions(pdev, (1 << 2) - 1,
2590 					   vmxnet3_driver_name);
2591 	if (err) {
2592 		dev_err(&pdev->dev,
2593 			"Failed to request region for adapter: error %d\n", err);
2594 		goto err_set_mask;
2595 	}
2596 
2597 	pci_set_master(pdev);
2598 
2599 	mmio_start = pci_resource_start(pdev, 0);
2600 	mmio_len = pci_resource_len(pdev, 0);
2601 	adapter->hw_addr0 = ioremap(mmio_start, mmio_len);
2602 	if (!adapter->hw_addr0) {
2603 		dev_err(&pdev->dev, "Failed to map bar0\n");
2604 		err = -EIO;
2605 		goto err_ioremap;
2606 	}
2607 
2608 	mmio_start = pci_resource_start(pdev, 1);
2609 	mmio_len = pci_resource_len(pdev, 1);
2610 	adapter->hw_addr1 = ioremap(mmio_start, mmio_len);
2611 	if (!adapter->hw_addr1) {
2612 		dev_err(&pdev->dev, "Failed to map bar1\n");
2613 		err = -EIO;
2614 		goto err_bar1;
2615 	}
2616 	return 0;
2617 
2618 err_bar1:
2619 	iounmap(adapter->hw_addr0);
2620 err_ioremap:
2621 	pci_release_selected_regions(pdev, (1 << 2) - 1);
2622 err_set_mask:
2623 	pci_disable_device(pdev);
2624 	return err;
2625 }
2626 
2627 
2628 static void
2629 vmxnet3_free_pci_resources(struct vmxnet3_adapter *adapter)
2630 {
2631 	BUG_ON(!adapter->pdev);
2632 
2633 	iounmap(adapter->hw_addr0);
2634 	iounmap(adapter->hw_addr1);
2635 	pci_release_selected_regions(adapter->pdev, (1 << 2) - 1);
2636 	pci_disable_device(adapter->pdev);
2637 }
2638 
2639 
2640 static void
2641 vmxnet3_adjust_rx_ring_size(struct vmxnet3_adapter *adapter)
2642 {
2643 	size_t sz, i, ring0_size, ring1_size, comp_size;
2644 	struct vmxnet3_rx_queue	*rq = &adapter->rx_queue[0];
2645 
2646 
2647 	if (adapter->netdev->mtu <= VMXNET3_MAX_SKB_BUF_SIZE -
2648 				    VMXNET3_MAX_ETH_HDR_SIZE) {
2649 		adapter->skb_buf_size = adapter->netdev->mtu +
2650 					VMXNET3_MAX_ETH_HDR_SIZE;
2651 		if (adapter->skb_buf_size < VMXNET3_MIN_T0_BUF_SIZE)
2652 			adapter->skb_buf_size = VMXNET3_MIN_T0_BUF_SIZE;
2653 
2654 		adapter->rx_buf_per_pkt = 1;
2655 	} else {
2656 		adapter->skb_buf_size = VMXNET3_MAX_SKB_BUF_SIZE;
2657 		sz = adapter->netdev->mtu - VMXNET3_MAX_SKB_BUF_SIZE +
2658 					    VMXNET3_MAX_ETH_HDR_SIZE;
2659 		adapter->rx_buf_per_pkt = 1 + (sz + PAGE_SIZE - 1) / PAGE_SIZE;
2660 	}
2661 
2662 	/*
2663 	 * for simplicity, force the ring0 size to be a multiple of
2664 	 * rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN
2665 	 */
2666 	sz = adapter->rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN;
2667 	ring0_size = adapter->rx_queue[0].rx_ring[0].size;
2668 	ring0_size = (ring0_size + sz - 1) / sz * sz;
2669 	ring0_size = min_t(u32, ring0_size, VMXNET3_RX_RING_MAX_SIZE /
2670 			   sz * sz);
2671 	ring1_size = adapter->rx_queue[0].rx_ring[1].size;
2672 	ring1_size = (ring1_size + sz - 1) / sz * sz;
2673 	ring1_size = min_t(u32, ring1_size, VMXNET3_RX_RING2_MAX_SIZE /
2674 			   sz * sz);
2675 	comp_size = ring0_size + ring1_size;
2676 
2677 	for (i = 0; i < adapter->num_rx_queues; i++) {
2678 		rq = &adapter->rx_queue[i];
2679 		rq->rx_ring[0].size = ring0_size;
2680 		rq->rx_ring[1].size = ring1_size;
2681 		rq->comp_ring.size = comp_size;
2682 	}
2683 }
2684 
2685 
2686 int
2687 vmxnet3_create_queues(struct vmxnet3_adapter *adapter, u32 tx_ring_size,
2688 		      u32 rx_ring_size, u32 rx_ring2_size)
2689 {
2690 	int err = 0, i;
2691 
2692 	for (i = 0; i < adapter->num_tx_queues; i++) {
2693 		struct vmxnet3_tx_queue	*tq = &adapter->tx_queue[i];
2694 		tq->tx_ring.size   = tx_ring_size;
2695 		tq->data_ring.size = tx_ring_size;
2696 		tq->comp_ring.size = tx_ring_size;
2697 		tq->shared = &adapter->tqd_start[i].ctrl;
2698 		tq->stopped = true;
2699 		tq->adapter = adapter;
2700 		tq->qid = i;
2701 		err = vmxnet3_tq_create(tq, adapter);
2702 		/*
2703 		 * Too late to change num_tx_queues. We cannot do away with
2704 		 * lesser number of queues than what we asked for
2705 		 */
2706 		if (err)
2707 			goto queue_err;
2708 	}
2709 
2710 	adapter->rx_queue[0].rx_ring[0].size = rx_ring_size;
2711 	adapter->rx_queue[0].rx_ring[1].size = rx_ring2_size;
2712 	vmxnet3_adjust_rx_ring_size(adapter);
2713 	for (i = 0; i < adapter->num_rx_queues; i++) {
2714 		struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
2715 		/* qid and qid2 for rx queues will be assigned later when num
2716 		 * of rx queues is finalized after allocating intrs */
2717 		rq->shared = &adapter->rqd_start[i].ctrl;
2718 		rq->adapter = adapter;
2719 		err = vmxnet3_rq_create(rq, adapter);
2720 		if (err) {
2721 			if (i == 0) {
2722 				netdev_err(adapter->netdev,
2723 					   "Could not allocate any rx queues. "
2724 					   "Aborting.\n");
2725 				goto queue_err;
2726 			} else {
2727 				netdev_info(adapter->netdev,
2728 					    "Number of rx queues changed "
2729 					    "to : %d.\n", i);
2730 				adapter->num_rx_queues = i;
2731 				err = 0;
2732 				break;
2733 			}
2734 		}
2735 	}
2736 	return err;
2737 queue_err:
2738 	vmxnet3_tq_destroy_all(adapter);
2739 	return err;
2740 }
2741 
2742 static int
2743 vmxnet3_open(struct net_device *netdev)
2744 {
2745 	struct vmxnet3_adapter *adapter;
2746 	int err, i;
2747 
2748 	adapter = netdev_priv(netdev);
2749 
2750 	for (i = 0; i < adapter->num_tx_queues; i++)
2751 		spin_lock_init(&adapter->tx_queue[i].tx_lock);
2752 
2753 	err = vmxnet3_create_queues(adapter, adapter->tx_ring_size,
2754 				    adapter->rx_ring_size,
2755 				    adapter->rx_ring2_size);
2756 	if (err)
2757 		goto queue_err;
2758 
2759 	err = vmxnet3_activate_dev(adapter);
2760 	if (err)
2761 		goto activate_err;
2762 
2763 	return 0;
2764 
2765 activate_err:
2766 	vmxnet3_rq_destroy_all(adapter);
2767 	vmxnet3_tq_destroy_all(adapter);
2768 queue_err:
2769 	return err;
2770 }
2771 
2772 
2773 static int
2774 vmxnet3_close(struct net_device *netdev)
2775 {
2776 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2777 
2778 	/*
2779 	 * Reset_work may be in the middle of resetting the device, wait for its
2780 	 * completion.
2781 	 */
2782 	while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2783 		msleep(1);
2784 
2785 	vmxnet3_quiesce_dev(adapter);
2786 
2787 	vmxnet3_rq_destroy_all(adapter);
2788 	vmxnet3_tq_destroy_all(adapter);
2789 
2790 	clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2791 
2792 
2793 	return 0;
2794 }
2795 
2796 
2797 void
2798 vmxnet3_force_close(struct vmxnet3_adapter *adapter)
2799 {
2800 	int i;
2801 
2802 	/*
2803 	 * we must clear VMXNET3_STATE_BIT_RESETTING, otherwise
2804 	 * vmxnet3_close() will deadlock.
2805 	 */
2806 	BUG_ON(test_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state));
2807 
2808 	/* we need to enable NAPI, otherwise dev_close will deadlock */
2809 	for (i = 0; i < adapter->num_rx_queues; i++)
2810 		napi_enable(&adapter->rx_queue[i].napi);
2811 	dev_close(adapter->netdev);
2812 }
2813 
2814 
2815 static int
2816 vmxnet3_change_mtu(struct net_device *netdev, int new_mtu)
2817 {
2818 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2819 	int err = 0;
2820 
2821 	if (new_mtu < VMXNET3_MIN_MTU || new_mtu > VMXNET3_MAX_MTU)
2822 		return -EINVAL;
2823 
2824 	netdev->mtu = new_mtu;
2825 
2826 	/*
2827 	 * Reset_work may be in the middle of resetting the device, wait for its
2828 	 * completion.
2829 	 */
2830 	while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2831 		msleep(1);
2832 
2833 	if (netif_running(netdev)) {
2834 		vmxnet3_quiesce_dev(adapter);
2835 		vmxnet3_reset_dev(adapter);
2836 
2837 		/* we need to re-create the rx queue based on the new mtu */
2838 		vmxnet3_rq_destroy_all(adapter);
2839 		vmxnet3_adjust_rx_ring_size(adapter);
2840 		err = vmxnet3_rq_create_all(adapter);
2841 		if (err) {
2842 			netdev_err(netdev,
2843 				   "failed to re-create rx queues, "
2844 				   " error %d. Closing it.\n", err);
2845 			goto out;
2846 		}
2847 
2848 		err = vmxnet3_activate_dev(adapter);
2849 		if (err) {
2850 			netdev_err(netdev,
2851 				   "failed to re-activate, error %d. "
2852 				   "Closing it\n", err);
2853 			goto out;
2854 		}
2855 	}
2856 
2857 out:
2858 	clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2859 	if (err)
2860 		vmxnet3_force_close(adapter);
2861 
2862 	return err;
2863 }
2864 
2865 
2866 static void
2867 vmxnet3_declare_features(struct vmxnet3_adapter *adapter, bool dma64)
2868 {
2869 	struct net_device *netdev = adapter->netdev;
2870 
2871 	netdev->hw_features = NETIF_F_SG | NETIF_F_RXCSUM |
2872 		NETIF_F_HW_CSUM | NETIF_F_HW_VLAN_CTAG_TX |
2873 		NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_TSO | NETIF_F_TSO6 |
2874 		NETIF_F_LRO;
2875 	if (dma64)
2876 		netdev->hw_features |= NETIF_F_HIGHDMA;
2877 	netdev->vlan_features = netdev->hw_features &
2878 				~(NETIF_F_HW_VLAN_CTAG_TX |
2879 				  NETIF_F_HW_VLAN_CTAG_RX);
2880 	netdev->features = netdev->hw_features | NETIF_F_HW_VLAN_CTAG_FILTER;
2881 }
2882 
2883 
2884 static void
2885 vmxnet3_read_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac)
2886 {
2887 	u32 tmp;
2888 
2889 	tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACL);
2890 	*(u32 *)mac = tmp;
2891 
2892 	tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACH);
2893 	mac[4] = tmp & 0xff;
2894 	mac[5] = (tmp >> 8) & 0xff;
2895 }
2896 
2897 #ifdef CONFIG_PCI_MSI
2898 
2899 /*
2900  * Enable MSIx vectors.
2901  * Returns :
2902  *	VMXNET3_LINUX_MIN_MSIX_VECT when only minimum number of vectors required
2903  *	 were enabled.
2904  *	number of vectors which were enabled otherwise (this number is greater
2905  *	 than VMXNET3_LINUX_MIN_MSIX_VECT)
2906  */
2907 
2908 static int
2909 vmxnet3_acquire_msix_vectors(struct vmxnet3_adapter *adapter, int nvec)
2910 {
2911 	int ret = pci_enable_msix_range(adapter->pdev,
2912 					adapter->intr.msix_entries, nvec, nvec);
2913 
2914 	if (ret == -ENOSPC && nvec > VMXNET3_LINUX_MIN_MSIX_VECT) {
2915 		dev_err(&adapter->netdev->dev,
2916 			"Failed to enable %d MSI-X, trying %d\n",
2917 			nvec, VMXNET3_LINUX_MIN_MSIX_VECT);
2918 
2919 		ret = pci_enable_msix_range(adapter->pdev,
2920 					    adapter->intr.msix_entries,
2921 					    VMXNET3_LINUX_MIN_MSIX_VECT,
2922 					    VMXNET3_LINUX_MIN_MSIX_VECT);
2923 	}
2924 
2925 	if (ret < 0) {
2926 		dev_err(&adapter->netdev->dev,
2927 			"Failed to enable MSI-X, error: %d\n", ret);
2928 	}
2929 
2930 	return ret;
2931 }
2932 
2933 
2934 #endif /* CONFIG_PCI_MSI */
2935 
2936 static void
2937 vmxnet3_alloc_intr_resources(struct vmxnet3_adapter *adapter)
2938 {
2939 	u32 cfg;
2940 	unsigned long flags;
2941 
2942 	/* intr settings */
2943 	spin_lock_irqsave(&adapter->cmd_lock, flags);
2944 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2945 			       VMXNET3_CMD_GET_CONF_INTR);
2946 	cfg = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
2947 	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2948 	adapter->intr.type = cfg & 0x3;
2949 	adapter->intr.mask_mode = (cfg >> 2) & 0x3;
2950 
2951 	if (adapter->intr.type == VMXNET3_IT_AUTO) {
2952 		adapter->intr.type = VMXNET3_IT_MSIX;
2953 	}
2954 
2955 #ifdef CONFIG_PCI_MSI
2956 	if (adapter->intr.type == VMXNET3_IT_MSIX) {
2957 		int i, nvec;
2958 
2959 		nvec  = adapter->share_intr == VMXNET3_INTR_TXSHARE ?
2960 			1 : adapter->num_tx_queues;
2961 		nvec += adapter->share_intr == VMXNET3_INTR_BUDDYSHARE ?
2962 			0 : adapter->num_rx_queues;
2963 		nvec += 1;	/* for link event */
2964 		nvec = nvec > VMXNET3_LINUX_MIN_MSIX_VECT ?
2965 		       nvec : VMXNET3_LINUX_MIN_MSIX_VECT;
2966 
2967 		for (i = 0; i < nvec; i++)
2968 			adapter->intr.msix_entries[i].entry = i;
2969 
2970 		nvec = vmxnet3_acquire_msix_vectors(adapter, nvec);
2971 		if (nvec < 0)
2972 			goto msix_err;
2973 
2974 		/* If we cannot allocate one MSIx vector per queue
2975 		 * then limit the number of rx queues to 1
2976 		 */
2977 		if (nvec == VMXNET3_LINUX_MIN_MSIX_VECT) {
2978 			if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE
2979 			    || adapter->num_rx_queues != 1) {
2980 				adapter->share_intr = VMXNET3_INTR_TXSHARE;
2981 				netdev_err(adapter->netdev,
2982 					   "Number of rx queues : 1\n");
2983 				adapter->num_rx_queues = 1;
2984 			}
2985 		}
2986 
2987 		adapter->intr.num_intrs = nvec;
2988 		return;
2989 
2990 msix_err:
2991 		/* If we cannot allocate MSIx vectors use only one rx queue */
2992 		dev_info(&adapter->pdev->dev,
2993 			 "Failed to enable MSI-X, error %d. "
2994 			 "Limiting #rx queues to 1, try MSI.\n", nvec);
2995 
2996 		adapter->intr.type = VMXNET3_IT_MSI;
2997 	}
2998 
2999 	if (adapter->intr.type == VMXNET3_IT_MSI) {
3000 		if (!pci_enable_msi(adapter->pdev)) {
3001 			adapter->num_rx_queues = 1;
3002 			adapter->intr.num_intrs = 1;
3003 			return;
3004 		}
3005 	}
3006 #endif /* CONFIG_PCI_MSI */
3007 
3008 	adapter->num_rx_queues = 1;
3009 	dev_info(&adapter->netdev->dev,
3010 		 "Using INTx interrupt, #Rx queues: 1.\n");
3011 	adapter->intr.type = VMXNET3_IT_INTX;
3012 
3013 	/* INT-X related setting */
3014 	adapter->intr.num_intrs = 1;
3015 }
3016 
3017 
3018 static void
3019 vmxnet3_free_intr_resources(struct vmxnet3_adapter *adapter)
3020 {
3021 	if (adapter->intr.type == VMXNET3_IT_MSIX)
3022 		pci_disable_msix(adapter->pdev);
3023 	else if (adapter->intr.type == VMXNET3_IT_MSI)
3024 		pci_disable_msi(adapter->pdev);
3025 	else
3026 		BUG_ON(adapter->intr.type != VMXNET3_IT_INTX);
3027 }
3028 
3029 
3030 static void
3031 vmxnet3_tx_timeout(struct net_device *netdev)
3032 {
3033 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3034 	adapter->tx_timeout_count++;
3035 
3036 	netdev_err(adapter->netdev, "tx hang\n");
3037 	schedule_work(&adapter->work);
3038 	netif_wake_queue(adapter->netdev);
3039 }
3040 
3041 
3042 static void
3043 vmxnet3_reset_work(struct work_struct *data)
3044 {
3045 	struct vmxnet3_adapter *adapter;
3046 
3047 	adapter = container_of(data, struct vmxnet3_adapter, work);
3048 
3049 	/* if another thread is resetting the device, no need to proceed */
3050 	if (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
3051 		return;
3052 
3053 	/* if the device is closed, we must leave it alone */
3054 	rtnl_lock();
3055 	if (netif_running(adapter->netdev)) {
3056 		netdev_notice(adapter->netdev, "resetting\n");
3057 		vmxnet3_quiesce_dev(adapter);
3058 		vmxnet3_reset_dev(adapter);
3059 		vmxnet3_activate_dev(adapter);
3060 	} else {
3061 		netdev_info(adapter->netdev, "already closed\n");
3062 	}
3063 	rtnl_unlock();
3064 
3065 	clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
3066 }
3067 
3068 
3069 static int
3070 vmxnet3_probe_device(struct pci_dev *pdev,
3071 		     const struct pci_device_id *id)
3072 {
3073 	static const struct net_device_ops vmxnet3_netdev_ops = {
3074 		.ndo_open = vmxnet3_open,
3075 		.ndo_stop = vmxnet3_close,
3076 		.ndo_start_xmit = vmxnet3_xmit_frame,
3077 		.ndo_set_mac_address = vmxnet3_set_mac_addr,
3078 		.ndo_change_mtu = vmxnet3_change_mtu,
3079 		.ndo_set_features = vmxnet3_set_features,
3080 		.ndo_get_stats64 = vmxnet3_get_stats64,
3081 		.ndo_tx_timeout = vmxnet3_tx_timeout,
3082 		.ndo_set_rx_mode = vmxnet3_set_mc,
3083 		.ndo_vlan_rx_add_vid = vmxnet3_vlan_rx_add_vid,
3084 		.ndo_vlan_rx_kill_vid = vmxnet3_vlan_rx_kill_vid,
3085 #ifdef CONFIG_NET_POLL_CONTROLLER
3086 		.ndo_poll_controller = vmxnet3_netpoll,
3087 #endif
3088 	};
3089 	int err;
3090 	bool dma64 = false; /* stupid gcc */
3091 	u32 ver;
3092 	struct net_device *netdev;
3093 	struct vmxnet3_adapter *adapter;
3094 	u8 mac[ETH_ALEN];
3095 	int size;
3096 	int num_tx_queues;
3097 	int num_rx_queues;
3098 
3099 	if (!pci_msi_enabled())
3100 		enable_mq = 0;
3101 
3102 #ifdef VMXNET3_RSS
3103 	if (enable_mq)
3104 		num_rx_queues = min(VMXNET3_DEVICE_MAX_RX_QUEUES,
3105 				    (int)num_online_cpus());
3106 	else
3107 #endif
3108 		num_rx_queues = 1;
3109 	num_rx_queues = rounddown_pow_of_two(num_rx_queues);
3110 
3111 	if (enable_mq)
3112 		num_tx_queues = min(VMXNET3_DEVICE_MAX_TX_QUEUES,
3113 				    (int)num_online_cpus());
3114 	else
3115 		num_tx_queues = 1;
3116 
3117 	num_tx_queues = rounddown_pow_of_two(num_tx_queues);
3118 	netdev = alloc_etherdev_mq(sizeof(struct vmxnet3_adapter),
3119 				   max(num_tx_queues, num_rx_queues));
3120 	dev_info(&pdev->dev,
3121 		 "# of Tx queues : %d, # of Rx queues : %d\n",
3122 		 num_tx_queues, num_rx_queues);
3123 
3124 	if (!netdev)
3125 		return -ENOMEM;
3126 
3127 	pci_set_drvdata(pdev, netdev);
3128 	adapter = netdev_priv(netdev);
3129 	adapter->netdev = netdev;
3130 	adapter->pdev = pdev;
3131 
3132 	adapter->tx_ring_size = VMXNET3_DEF_TX_RING_SIZE;
3133 	adapter->rx_ring_size = VMXNET3_DEF_RX_RING_SIZE;
3134 	adapter->rx_ring2_size = VMXNET3_DEF_RX_RING2_SIZE;
3135 
3136 	spin_lock_init(&adapter->cmd_lock);
3137 	adapter->adapter_pa = dma_map_single(&adapter->pdev->dev, adapter,
3138 					     sizeof(struct vmxnet3_adapter),
3139 					     PCI_DMA_TODEVICE);
3140 	if (dma_mapping_error(&adapter->pdev->dev, adapter->adapter_pa)) {
3141 		dev_err(&pdev->dev, "Failed to map dma\n");
3142 		err = -EFAULT;
3143 		goto err_dma_map;
3144 	}
3145 	adapter->shared = dma_alloc_coherent(
3146 				&adapter->pdev->dev,
3147 				sizeof(struct Vmxnet3_DriverShared),
3148 				&adapter->shared_pa, GFP_KERNEL);
3149 	if (!adapter->shared) {
3150 		dev_err(&pdev->dev, "Failed to allocate memory\n");
3151 		err = -ENOMEM;
3152 		goto err_alloc_shared;
3153 	}
3154 
3155 	adapter->num_rx_queues = num_rx_queues;
3156 	adapter->num_tx_queues = num_tx_queues;
3157 	adapter->rx_buf_per_pkt = 1;
3158 
3159 	size = sizeof(struct Vmxnet3_TxQueueDesc) * adapter->num_tx_queues;
3160 	size += sizeof(struct Vmxnet3_RxQueueDesc) * adapter->num_rx_queues;
3161 	adapter->tqd_start = dma_alloc_coherent(&adapter->pdev->dev, size,
3162 						&adapter->queue_desc_pa,
3163 						GFP_KERNEL);
3164 
3165 	if (!adapter->tqd_start) {
3166 		dev_err(&pdev->dev, "Failed to allocate memory\n");
3167 		err = -ENOMEM;
3168 		goto err_alloc_queue_desc;
3169 	}
3170 	adapter->rqd_start = (struct Vmxnet3_RxQueueDesc *)(adapter->tqd_start +
3171 							    adapter->num_tx_queues);
3172 
3173 	adapter->pm_conf = dma_alloc_coherent(&adapter->pdev->dev,
3174 					      sizeof(struct Vmxnet3_PMConf),
3175 					      &adapter->pm_conf_pa,
3176 					      GFP_KERNEL);
3177 	if (adapter->pm_conf == NULL) {
3178 		err = -ENOMEM;
3179 		goto err_alloc_pm;
3180 	}
3181 
3182 #ifdef VMXNET3_RSS
3183 
3184 	adapter->rss_conf = dma_alloc_coherent(&adapter->pdev->dev,
3185 					       sizeof(struct UPT1_RSSConf),
3186 					       &adapter->rss_conf_pa,
3187 					       GFP_KERNEL);
3188 	if (adapter->rss_conf == NULL) {
3189 		err = -ENOMEM;
3190 		goto err_alloc_rss;
3191 	}
3192 #endif /* VMXNET3_RSS */
3193 
3194 	err = vmxnet3_alloc_pci_resources(adapter, &dma64);
3195 	if (err < 0)
3196 		goto err_alloc_pci;
3197 
3198 	ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_VRRS);
3199 	if (ver & 2) {
3200 		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_VRRS, 2);
3201 		adapter->version = 2;
3202 	} else if (ver & 1) {
3203 		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_VRRS, 1);
3204 		adapter->version = 1;
3205 	} else {
3206 		dev_err(&pdev->dev,
3207 			"Incompatible h/w version (0x%x) for adapter\n", ver);
3208 		err = -EBUSY;
3209 		goto err_ver;
3210 	}
3211 	dev_dbg(&pdev->dev, "Using device version %d\n", adapter->version);
3212 
3213 	ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_UVRS);
3214 	if (ver & 1) {
3215 		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_UVRS, 1);
3216 	} else {
3217 		dev_err(&pdev->dev,
3218 			"Incompatible upt version (0x%x) for adapter\n", ver);
3219 		err = -EBUSY;
3220 		goto err_ver;
3221 	}
3222 
3223 	SET_NETDEV_DEV(netdev, &pdev->dev);
3224 	vmxnet3_declare_features(adapter, dma64);
3225 
3226 	if (adapter->num_tx_queues == adapter->num_rx_queues)
3227 		adapter->share_intr = VMXNET3_INTR_BUDDYSHARE;
3228 	else
3229 		adapter->share_intr = VMXNET3_INTR_DONTSHARE;
3230 
3231 	vmxnet3_alloc_intr_resources(adapter);
3232 
3233 #ifdef VMXNET3_RSS
3234 	if (adapter->num_rx_queues > 1 &&
3235 	    adapter->intr.type == VMXNET3_IT_MSIX) {
3236 		adapter->rss = true;
3237 		netdev->hw_features |= NETIF_F_RXHASH;
3238 		netdev->features |= NETIF_F_RXHASH;
3239 		dev_dbg(&pdev->dev, "RSS is enabled.\n");
3240 	} else {
3241 		adapter->rss = false;
3242 	}
3243 #endif
3244 
3245 	vmxnet3_read_mac_addr(adapter, mac);
3246 	memcpy(netdev->dev_addr,  mac, netdev->addr_len);
3247 
3248 	netdev->netdev_ops = &vmxnet3_netdev_ops;
3249 	vmxnet3_set_ethtool_ops(netdev);
3250 	netdev->watchdog_timeo = 5 * HZ;
3251 
3252 	INIT_WORK(&adapter->work, vmxnet3_reset_work);
3253 	set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
3254 
3255 	if (adapter->intr.type == VMXNET3_IT_MSIX) {
3256 		int i;
3257 		for (i = 0; i < adapter->num_rx_queues; i++) {
3258 			netif_napi_add(adapter->netdev,
3259 				       &adapter->rx_queue[i].napi,
3260 				       vmxnet3_poll_rx_only, 64);
3261 		}
3262 	} else {
3263 		netif_napi_add(adapter->netdev, &adapter->rx_queue[0].napi,
3264 			       vmxnet3_poll, 64);
3265 	}
3266 
3267 	netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
3268 	netif_set_real_num_rx_queues(adapter->netdev, adapter->num_rx_queues);
3269 
3270 	netif_carrier_off(netdev);
3271 	err = register_netdev(netdev);
3272 
3273 	if (err) {
3274 		dev_err(&pdev->dev, "Failed to register adapter\n");
3275 		goto err_register;
3276 	}
3277 
3278 	vmxnet3_check_link(adapter, false);
3279 	return 0;
3280 
3281 err_register:
3282 	vmxnet3_free_intr_resources(adapter);
3283 err_ver:
3284 	vmxnet3_free_pci_resources(adapter);
3285 err_alloc_pci:
3286 #ifdef VMXNET3_RSS
3287 	dma_free_coherent(&adapter->pdev->dev, sizeof(struct UPT1_RSSConf),
3288 			  adapter->rss_conf, adapter->rss_conf_pa);
3289 err_alloc_rss:
3290 #endif
3291 	dma_free_coherent(&adapter->pdev->dev, sizeof(struct Vmxnet3_PMConf),
3292 			  adapter->pm_conf, adapter->pm_conf_pa);
3293 err_alloc_pm:
3294 	dma_free_coherent(&adapter->pdev->dev, size, adapter->tqd_start,
3295 			  adapter->queue_desc_pa);
3296 err_alloc_queue_desc:
3297 	dma_free_coherent(&adapter->pdev->dev,
3298 			  sizeof(struct Vmxnet3_DriverShared),
3299 			  adapter->shared, adapter->shared_pa);
3300 err_alloc_shared:
3301 	dma_unmap_single(&adapter->pdev->dev, adapter->adapter_pa,
3302 			 sizeof(struct vmxnet3_adapter), PCI_DMA_TODEVICE);
3303 err_dma_map:
3304 	free_netdev(netdev);
3305 	return err;
3306 }
3307 
3308 
3309 static void
3310 vmxnet3_remove_device(struct pci_dev *pdev)
3311 {
3312 	struct net_device *netdev = pci_get_drvdata(pdev);
3313 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3314 	int size = 0;
3315 	int num_rx_queues;
3316 
3317 #ifdef VMXNET3_RSS
3318 	if (enable_mq)
3319 		num_rx_queues = min(VMXNET3_DEVICE_MAX_RX_QUEUES,
3320 				    (int)num_online_cpus());
3321 	else
3322 #endif
3323 		num_rx_queues = 1;
3324 	num_rx_queues = rounddown_pow_of_two(num_rx_queues);
3325 
3326 	cancel_work_sync(&adapter->work);
3327 
3328 	unregister_netdev(netdev);
3329 
3330 	vmxnet3_free_intr_resources(adapter);
3331 	vmxnet3_free_pci_resources(adapter);
3332 #ifdef VMXNET3_RSS
3333 	dma_free_coherent(&adapter->pdev->dev, sizeof(struct UPT1_RSSConf),
3334 			  adapter->rss_conf, adapter->rss_conf_pa);
3335 #endif
3336 	dma_free_coherent(&adapter->pdev->dev, sizeof(struct Vmxnet3_PMConf),
3337 			  adapter->pm_conf, adapter->pm_conf_pa);
3338 
3339 	size = sizeof(struct Vmxnet3_TxQueueDesc) * adapter->num_tx_queues;
3340 	size += sizeof(struct Vmxnet3_RxQueueDesc) * num_rx_queues;
3341 	dma_free_coherent(&adapter->pdev->dev, size, adapter->tqd_start,
3342 			  adapter->queue_desc_pa);
3343 	dma_free_coherent(&adapter->pdev->dev,
3344 			  sizeof(struct Vmxnet3_DriverShared),
3345 			  adapter->shared, adapter->shared_pa);
3346 	dma_unmap_single(&adapter->pdev->dev, adapter->adapter_pa,
3347 			 sizeof(struct vmxnet3_adapter), PCI_DMA_TODEVICE);
3348 	free_netdev(netdev);
3349 }
3350 
3351 static void vmxnet3_shutdown_device(struct pci_dev *pdev)
3352 {
3353 	struct net_device *netdev = pci_get_drvdata(pdev);
3354 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3355 	unsigned long flags;
3356 
3357 	/* Reset_work may be in the middle of resetting the device, wait for its
3358 	 * completion.
3359 	 */
3360 	while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
3361 		msleep(1);
3362 
3363 	if (test_and_set_bit(VMXNET3_STATE_BIT_QUIESCED,
3364 			     &adapter->state)) {
3365 		clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
3366 		return;
3367 	}
3368 	spin_lock_irqsave(&adapter->cmd_lock, flags);
3369 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3370 			       VMXNET3_CMD_QUIESCE_DEV);
3371 	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3372 	vmxnet3_disable_all_intrs(adapter);
3373 
3374 	clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
3375 }
3376 
3377 
3378 #ifdef CONFIG_PM
3379 
3380 static int
3381 vmxnet3_suspend(struct device *device)
3382 {
3383 	struct pci_dev *pdev = to_pci_dev(device);
3384 	struct net_device *netdev = pci_get_drvdata(pdev);
3385 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3386 	struct Vmxnet3_PMConf *pmConf;
3387 	struct ethhdr *ehdr;
3388 	struct arphdr *ahdr;
3389 	u8 *arpreq;
3390 	struct in_device *in_dev;
3391 	struct in_ifaddr *ifa;
3392 	unsigned long flags;
3393 	int i = 0;
3394 
3395 	if (!netif_running(netdev))
3396 		return 0;
3397 
3398 	for (i = 0; i < adapter->num_rx_queues; i++)
3399 		napi_disable(&adapter->rx_queue[i].napi);
3400 
3401 	vmxnet3_disable_all_intrs(adapter);
3402 	vmxnet3_free_irqs(adapter);
3403 	vmxnet3_free_intr_resources(adapter);
3404 
3405 	netif_device_detach(netdev);
3406 	netif_tx_stop_all_queues(netdev);
3407 
3408 	/* Create wake-up filters. */
3409 	pmConf = adapter->pm_conf;
3410 	memset(pmConf, 0, sizeof(*pmConf));
3411 
3412 	if (adapter->wol & WAKE_UCAST) {
3413 		pmConf->filters[i].patternSize = ETH_ALEN;
3414 		pmConf->filters[i].maskSize = 1;
3415 		memcpy(pmConf->filters[i].pattern, netdev->dev_addr, ETH_ALEN);
3416 		pmConf->filters[i].mask[0] = 0x3F; /* LSB ETH_ALEN bits */
3417 
3418 		pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_FILTER;
3419 		i++;
3420 	}
3421 
3422 	if (adapter->wol & WAKE_ARP) {
3423 		in_dev = in_dev_get(netdev);
3424 		if (!in_dev)
3425 			goto skip_arp;
3426 
3427 		ifa = (struct in_ifaddr *)in_dev->ifa_list;
3428 		if (!ifa)
3429 			goto skip_arp;
3430 
3431 		pmConf->filters[i].patternSize = ETH_HLEN + /* Ethernet header*/
3432 			sizeof(struct arphdr) +		/* ARP header */
3433 			2 * ETH_ALEN +		/* 2 Ethernet addresses*/
3434 			2 * sizeof(u32);	/*2 IPv4 addresses */
3435 		pmConf->filters[i].maskSize =
3436 			(pmConf->filters[i].patternSize - 1) / 8 + 1;
3437 
3438 		/* ETH_P_ARP in Ethernet header. */
3439 		ehdr = (struct ethhdr *)pmConf->filters[i].pattern;
3440 		ehdr->h_proto = htons(ETH_P_ARP);
3441 
3442 		/* ARPOP_REQUEST in ARP header. */
3443 		ahdr = (struct arphdr *)&pmConf->filters[i].pattern[ETH_HLEN];
3444 		ahdr->ar_op = htons(ARPOP_REQUEST);
3445 		arpreq = (u8 *)(ahdr + 1);
3446 
3447 		/* The Unicast IPv4 address in 'tip' field. */
3448 		arpreq += 2 * ETH_ALEN + sizeof(u32);
3449 		*(u32 *)arpreq = ifa->ifa_address;
3450 
3451 		/* The mask for the relevant bits. */
3452 		pmConf->filters[i].mask[0] = 0x00;
3453 		pmConf->filters[i].mask[1] = 0x30; /* ETH_P_ARP */
3454 		pmConf->filters[i].mask[2] = 0x30; /* ARPOP_REQUEST */
3455 		pmConf->filters[i].mask[3] = 0x00;
3456 		pmConf->filters[i].mask[4] = 0xC0; /* IPv4 TIP */
3457 		pmConf->filters[i].mask[5] = 0x03; /* IPv4 TIP */
3458 		in_dev_put(in_dev);
3459 
3460 		pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_FILTER;
3461 		i++;
3462 	}
3463 
3464 skip_arp:
3465 	if (adapter->wol & WAKE_MAGIC)
3466 		pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_MAGIC;
3467 
3468 	pmConf->numFilters = i;
3469 
3470 	adapter->shared->devRead.pmConfDesc.confVer = cpu_to_le32(1);
3471 	adapter->shared->devRead.pmConfDesc.confLen = cpu_to_le32(sizeof(
3472 								  *pmConf));
3473 	adapter->shared->devRead.pmConfDesc.confPA =
3474 		cpu_to_le64(adapter->pm_conf_pa);
3475 
3476 	spin_lock_irqsave(&adapter->cmd_lock, flags);
3477 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3478 			       VMXNET3_CMD_UPDATE_PMCFG);
3479 	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3480 
3481 	pci_save_state(pdev);
3482 	pci_enable_wake(pdev, pci_choose_state(pdev, PMSG_SUSPEND),
3483 			adapter->wol);
3484 	pci_disable_device(pdev);
3485 	pci_set_power_state(pdev, pci_choose_state(pdev, PMSG_SUSPEND));
3486 
3487 	return 0;
3488 }
3489 
3490 
3491 static int
3492 vmxnet3_resume(struct device *device)
3493 {
3494 	int err;
3495 	unsigned long flags;
3496 	struct pci_dev *pdev = to_pci_dev(device);
3497 	struct net_device *netdev = pci_get_drvdata(pdev);
3498 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3499 
3500 	if (!netif_running(netdev))
3501 		return 0;
3502 
3503 	pci_set_power_state(pdev, PCI_D0);
3504 	pci_restore_state(pdev);
3505 	err = pci_enable_device_mem(pdev);
3506 	if (err != 0)
3507 		return err;
3508 
3509 	pci_enable_wake(pdev, PCI_D0, 0);
3510 
3511 	vmxnet3_alloc_intr_resources(adapter);
3512 
3513 	/* During hibernate and suspend, device has to be reinitialized as the
3514 	 * device state need not be preserved.
3515 	 */
3516 
3517 	/* Need not check adapter state as other reset tasks cannot run during
3518 	 * device resume.
3519 	 */
3520 	spin_lock_irqsave(&adapter->cmd_lock, flags);
3521 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3522 			       VMXNET3_CMD_QUIESCE_DEV);
3523 	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3524 	vmxnet3_tq_cleanup_all(adapter);
3525 	vmxnet3_rq_cleanup_all(adapter);
3526 
3527 	vmxnet3_reset_dev(adapter);
3528 	err = vmxnet3_activate_dev(adapter);
3529 	if (err != 0) {
3530 		netdev_err(netdev,
3531 			   "failed to re-activate on resume, error: %d", err);
3532 		vmxnet3_force_close(adapter);
3533 		return err;
3534 	}
3535 	netif_device_attach(netdev);
3536 
3537 	return 0;
3538 }
3539 
3540 static const struct dev_pm_ops vmxnet3_pm_ops = {
3541 	.suspend = vmxnet3_suspend,
3542 	.resume = vmxnet3_resume,
3543 	.freeze = vmxnet3_suspend,
3544 	.restore = vmxnet3_resume,
3545 };
3546 #endif
3547 
3548 static struct pci_driver vmxnet3_driver = {
3549 	.name		= vmxnet3_driver_name,
3550 	.id_table	= vmxnet3_pciid_table,
3551 	.probe		= vmxnet3_probe_device,
3552 	.remove		= vmxnet3_remove_device,
3553 	.shutdown	= vmxnet3_shutdown_device,
3554 #ifdef CONFIG_PM
3555 	.driver.pm	= &vmxnet3_pm_ops,
3556 #endif
3557 };
3558 
3559 
3560 static int __init
3561 vmxnet3_init_module(void)
3562 {
3563 	pr_info("%s - version %s\n", VMXNET3_DRIVER_DESC,
3564 		VMXNET3_DRIVER_VERSION_REPORT);
3565 	return pci_register_driver(&vmxnet3_driver);
3566 }
3567 
3568 module_init(vmxnet3_init_module);
3569 
3570 
3571 static void
3572 vmxnet3_exit_module(void)
3573 {
3574 	pci_unregister_driver(&vmxnet3_driver);
3575 }
3576 
3577 module_exit(vmxnet3_exit_module);
3578 
3579 MODULE_AUTHOR("VMware, Inc.");
3580 MODULE_DESCRIPTION(VMXNET3_DRIVER_DESC);
3581 MODULE_LICENSE("GPL v2");
3582 MODULE_VERSION(VMXNET3_DRIVER_VERSION_STRING);
3583