1 /*
2  * Linux driver for VMware's vmxnet3 ethernet NIC.
3  *
4  * Copyright (C) 2008-2016, VMware, Inc. All Rights Reserved.
5  *
6  * This program is free software; you can redistribute it and/or modify it
7  * under the terms of the GNU General Public License as published by the
8  * Free Software Foundation; version 2 of the License and no later version.
9  *
10  * This program is distributed in the hope that it will be useful, but
11  * WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
13  * NON INFRINGEMENT. See the GNU General Public License for more
14  * details.
15  *
16  * You should have received a copy of the GNU General Public License
17  * along with this program; if not, write to the Free Software
18  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19  *
20  * The full GNU General Public License is included in this distribution in
21  * the file called "COPYING".
22  *
23  * Maintained by: pv-drivers@vmware.com
24  *
25  */
26 
27 #include <linux/module.h>
28 #include <net/ip6_checksum.h>
29 
30 #include "vmxnet3_int.h"
31 
32 char vmxnet3_driver_name[] = "vmxnet3";
33 #define VMXNET3_DRIVER_DESC "VMware vmxnet3 virtual NIC driver"
34 
35 /*
36  * PCI Device ID Table
37  * Last entry must be all 0s
38  */
39 static const struct pci_device_id vmxnet3_pciid_table[] = {
40 	{PCI_VDEVICE(VMWARE, PCI_DEVICE_ID_VMWARE_VMXNET3)},
41 	{0}
42 };
43 
44 MODULE_DEVICE_TABLE(pci, vmxnet3_pciid_table);
45 
46 static int enable_mq = 1;
47 
48 static void
49 vmxnet3_write_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac);
50 
51 /*
52  *    Enable/Disable the given intr
53  */
54 static void
55 vmxnet3_enable_intr(struct vmxnet3_adapter *adapter, unsigned intr_idx)
56 {
57 	VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_IMR + intr_idx * 8, 0);
58 }
59 
60 
61 static void
62 vmxnet3_disable_intr(struct vmxnet3_adapter *adapter, unsigned intr_idx)
63 {
64 	VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_IMR + intr_idx * 8, 1);
65 }
66 
67 
68 /*
69  *    Enable/Disable all intrs used by the device
70  */
71 static void
72 vmxnet3_enable_all_intrs(struct vmxnet3_adapter *adapter)
73 {
74 	int i;
75 
76 	for (i = 0; i < adapter->intr.num_intrs; i++)
77 		vmxnet3_enable_intr(adapter, i);
78 	adapter->shared->devRead.intrConf.intrCtrl &=
79 					cpu_to_le32(~VMXNET3_IC_DISABLE_ALL);
80 }
81 
82 
83 static void
84 vmxnet3_disable_all_intrs(struct vmxnet3_adapter *adapter)
85 {
86 	int i;
87 
88 	adapter->shared->devRead.intrConf.intrCtrl |=
89 					cpu_to_le32(VMXNET3_IC_DISABLE_ALL);
90 	for (i = 0; i < adapter->intr.num_intrs; i++)
91 		vmxnet3_disable_intr(adapter, i);
92 }
93 
94 
95 static void
96 vmxnet3_ack_events(struct vmxnet3_adapter *adapter, u32 events)
97 {
98 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_ECR, events);
99 }
100 
101 
102 static bool
103 vmxnet3_tq_stopped(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
104 {
105 	return tq->stopped;
106 }
107 
108 
109 static void
110 vmxnet3_tq_start(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
111 {
112 	tq->stopped = false;
113 	netif_start_subqueue(adapter->netdev, tq - adapter->tx_queue);
114 }
115 
116 
117 static void
118 vmxnet3_tq_wake(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
119 {
120 	tq->stopped = false;
121 	netif_wake_subqueue(adapter->netdev, (tq - adapter->tx_queue));
122 }
123 
124 
125 static void
126 vmxnet3_tq_stop(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
127 {
128 	tq->stopped = true;
129 	tq->num_stop++;
130 	netif_stop_subqueue(adapter->netdev, (tq - adapter->tx_queue));
131 }
132 
133 
134 /*
135  * Check the link state. This may start or stop the tx queue.
136  */
137 static void
138 vmxnet3_check_link(struct vmxnet3_adapter *adapter, bool affectTxQueue)
139 {
140 	u32 ret;
141 	int i;
142 	unsigned long flags;
143 
144 	spin_lock_irqsave(&adapter->cmd_lock, flags);
145 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, VMXNET3_CMD_GET_LINK);
146 	ret = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
147 	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
148 
149 	adapter->link_speed = ret >> 16;
150 	if (ret & 1) { /* Link is up. */
151 		netdev_info(adapter->netdev, "NIC Link is Up %d Mbps\n",
152 			    adapter->link_speed);
153 		netif_carrier_on(adapter->netdev);
154 
155 		if (affectTxQueue) {
156 			for (i = 0; i < adapter->num_tx_queues; i++)
157 				vmxnet3_tq_start(&adapter->tx_queue[i],
158 						 adapter);
159 		}
160 	} else {
161 		netdev_info(adapter->netdev, "NIC Link is Down\n");
162 		netif_carrier_off(adapter->netdev);
163 
164 		if (affectTxQueue) {
165 			for (i = 0; i < adapter->num_tx_queues; i++)
166 				vmxnet3_tq_stop(&adapter->tx_queue[i], adapter);
167 		}
168 	}
169 }
170 
171 static void
172 vmxnet3_process_events(struct vmxnet3_adapter *adapter)
173 {
174 	int i;
175 	unsigned long flags;
176 	u32 events = le32_to_cpu(adapter->shared->ecr);
177 	if (!events)
178 		return;
179 
180 	vmxnet3_ack_events(adapter, events);
181 
182 	/* Check if link state has changed */
183 	if (events & VMXNET3_ECR_LINK)
184 		vmxnet3_check_link(adapter, true);
185 
186 	/* Check if there is an error on xmit/recv queues */
187 	if (events & (VMXNET3_ECR_TQERR | VMXNET3_ECR_RQERR)) {
188 		spin_lock_irqsave(&adapter->cmd_lock, flags);
189 		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
190 				       VMXNET3_CMD_GET_QUEUE_STATUS);
191 		spin_unlock_irqrestore(&adapter->cmd_lock, flags);
192 
193 		for (i = 0; i < adapter->num_tx_queues; i++)
194 			if (adapter->tqd_start[i].status.stopped)
195 				dev_err(&adapter->netdev->dev,
196 					"%s: tq[%d] error 0x%x\n",
197 					adapter->netdev->name, i, le32_to_cpu(
198 					adapter->tqd_start[i].status.error));
199 		for (i = 0; i < adapter->num_rx_queues; i++)
200 			if (adapter->rqd_start[i].status.stopped)
201 				dev_err(&adapter->netdev->dev,
202 					"%s: rq[%d] error 0x%x\n",
203 					adapter->netdev->name, i,
204 					adapter->rqd_start[i].status.error);
205 
206 		schedule_work(&adapter->work);
207 	}
208 }
209 
210 #ifdef __BIG_ENDIAN_BITFIELD
211 /*
212  * The device expects the bitfields in shared structures to be written in
213  * little endian. When CPU is big endian, the following routines are used to
214  * correctly read and write into ABI.
215  * The general technique used here is : double word bitfields are defined in
216  * opposite order for big endian architecture. Then before reading them in
217  * driver the complete double word is translated using le32_to_cpu. Similarly
218  * After the driver writes into bitfields, cpu_to_le32 is used to translate the
219  * double words into required format.
220  * In order to avoid touching bits in shared structure more than once, temporary
221  * descriptors are used. These are passed as srcDesc to following functions.
222  */
223 static void vmxnet3_RxDescToCPU(const struct Vmxnet3_RxDesc *srcDesc,
224 				struct Vmxnet3_RxDesc *dstDesc)
225 {
226 	u32 *src = (u32 *)srcDesc + 2;
227 	u32 *dst = (u32 *)dstDesc + 2;
228 	dstDesc->addr = le64_to_cpu(srcDesc->addr);
229 	*dst = le32_to_cpu(*src);
230 	dstDesc->ext1 = le32_to_cpu(srcDesc->ext1);
231 }
232 
233 static void vmxnet3_TxDescToLe(const struct Vmxnet3_TxDesc *srcDesc,
234 			       struct Vmxnet3_TxDesc *dstDesc)
235 {
236 	int i;
237 	u32 *src = (u32 *)(srcDesc + 1);
238 	u32 *dst = (u32 *)(dstDesc + 1);
239 
240 	/* Working backwards so that the gen bit is set at the end. */
241 	for (i = 2; i > 0; i--) {
242 		src--;
243 		dst--;
244 		*dst = cpu_to_le32(*src);
245 	}
246 }
247 
248 
249 static void vmxnet3_RxCompToCPU(const struct Vmxnet3_RxCompDesc *srcDesc,
250 				struct Vmxnet3_RxCompDesc *dstDesc)
251 {
252 	int i = 0;
253 	u32 *src = (u32 *)srcDesc;
254 	u32 *dst = (u32 *)dstDesc;
255 	for (i = 0; i < sizeof(struct Vmxnet3_RxCompDesc) / sizeof(u32); i++) {
256 		*dst = le32_to_cpu(*src);
257 		src++;
258 		dst++;
259 	}
260 }
261 
262 
263 /* Used to read bitfield values from double words. */
264 static u32 get_bitfield32(const __le32 *bitfield, u32 pos, u32 size)
265 {
266 	u32 temp = le32_to_cpu(*bitfield);
267 	u32 mask = ((1 << size) - 1) << pos;
268 	temp &= mask;
269 	temp >>= pos;
270 	return temp;
271 }
272 
273 
274 
275 #endif  /* __BIG_ENDIAN_BITFIELD */
276 
277 #ifdef __BIG_ENDIAN_BITFIELD
278 
279 #   define VMXNET3_TXDESC_GET_GEN(txdesc) get_bitfield32(((const __le32 *) \
280 			txdesc) + VMXNET3_TXD_GEN_DWORD_SHIFT, \
281 			VMXNET3_TXD_GEN_SHIFT, VMXNET3_TXD_GEN_SIZE)
282 #   define VMXNET3_TXDESC_GET_EOP(txdesc) get_bitfield32(((const __le32 *) \
283 			txdesc) + VMXNET3_TXD_EOP_DWORD_SHIFT, \
284 			VMXNET3_TXD_EOP_SHIFT, VMXNET3_TXD_EOP_SIZE)
285 #   define VMXNET3_TCD_GET_GEN(tcd) get_bitfield32(((const __le32 *)tcd) + \
286 			VMXNET3_TCD_GEN_DWORD_SHIFT, VMXNET3_TCD_GEN_SHIFT, \
287 			VMXNET3_TCD_GEN_SIZE)
288 #   define VMXNET3_TCD_GET_TXIDX(tcd) get_bitfield32((const __le32 *)tcd, \
289 			VMXNET3_TCD_TXIDX_SHIFT, VMXNET3_TCD_TXIDX_SIZE)
290 #   define vmxnet3_getRxComp(dstrcd, rcd, tmp) do { \
291 			(dstrcd) = (tmp); \
292 			vmxnet3_RxCompToCPU((rcd), (tmp)); \
293 		} while (0)
294 #   define vmxnet3_getRxDesc(dstrxd, rxd, tmp) do { \
295 			(dstrxd) = (tmp); \
296 			vmxnet3_RxDescToCPU((rxd), (tmp)); \
297 		} while (0)
298 
299 #else
300 
301 #   define VMXNET3_TXDESC_GET_GEN(txdesc) ((txdesc)->gen)
302 #   define VMXNET3_TXDESC_GET_EOP(txdesc) ((txdesc)->eop)
303 #   define VMXNET3_TCD_GET_GEN(tcd) ((tcd)->gen)
304 #   define VMXNET3_TCD_GET_TXIDX(tcd) ((tcd)->txdIdx)
305 #   define vmxnet3_getRxComp(dstrcd, rcd, tmp) (dstrcd) = (rcd)
306 #   define vmxnet3_getRxDesc(dstrxd, rxd, tmp) (dstrxd) = (rxd)
307 
308 #endif /* __BIG_ENDIAN_BITFIELD  */
309 
310 
311 static void
312 vmxnet3_unmap_tx_buf(struct vmxnet3_tx_buf_info *tbi,
313 		     struct pci_dev *pdev)
314 {
315 	if (tbi->map_type == VMXNET3_MAP_SINGLE)
316 		dma_unmap_single(&pdev->dev, tbi->dma_addr, tbi->len,
317 				 PCI_DMA_TODEVICE);
318 	else if (tbi->map_type == VMXNET3_MAP_PAGE)
319 		dma_unmap_page(&pdev->dev, tbi->dma_addr, tbi->len,
320 			       PCI_DMA_TODEVICE);
321 	else
322 		BUG_ON(tbi->map_type != VMXNET3_MAP_NONE);
323 
324 	tbi->map_type = VMXNET3_MAP_NONE; /* to help debugging */
325 }
326 
327 
328 static int
329 vmxnet3_unmap_pkt(u32 eop_idx, struct vmxnet3_tx_queue *tq,
330 		  struct pci_dev *pdev,	struct vmxnet3_adapter *adapter)
331 {
332 	struct sk_buff *skb;
333 	int entries = 0;
334 
335 	/* no out of order completion */
336 	BUG_ON(tq->buf_info[eop_idx].sop_idx != tq->tx_ring.next2comp);
337 	BUG_ON(VMXNET3_TXDESC_GET_EOP(&(tq->tx_ring.base[eop_idx].txd)) != 1);
338 
339 	skb = tq->buf_info[eop_idx].skb;
340 	BUG_ON(skb == NULL);
341 	tq->buf_info[eop_idx].skb = NULL;
342 
343 	VMXNET3_INC_RING_IDX_ONLY(eop_idx, tq->tx_ring.size);
344 
345 	while (tq->tx_ring.next2comp != eop_idx) {
346 		vmxnet3_unmap_tx_buf(tq->buf_info + tq->tx_ring.next2comp,
347 				     pdev);
348 
349 		/* update next2comp w/o tx_lock. Since we are marking more,
350 		 * instead of less, tx ring entries avail, the worst case is
351 		 * that the tx routine incorrectly re-queues a pkt due to
352 		 * insufficient tx ring entries.
353 		 */
354 		vmxnet3_cmd_ring_adv_next2comp(&tq->tx_ring);
355 		entries++;
356 	}
357 
358 	dev_kfree_skb_any(skb);
359 	return entries;
360 }
361 
362 
363 static int
364 vmxnet3_tq_tx_complete(struct vmxnet3_tx_queue *tq,
365 			struct vmxnet3_adapter *adapter)
366 {
367 	int completed = 0;
368 	union Vmxnet3_GenericDesc *gdesc;
369 
370 	gdesc = tq->comp_ring.base + tq->comp_ring.next2proc;
371 	while (VMXNET3_TCD_GET_GEN(&gdesc->tcd) == tq->comp_ring.gen) {
372 		completed += vmxnet3_unmap_pkt(VMXNET3_TCD_GET_TXIDX(
373 					       &gdesc->tcd), tq, adapter->pdev,
374 					       adapter);
375 
376 		vmxnet3_comp_ring_adv_next2proc(&tq->comp_ring);
377 		gdesc = tq->comp_ring.base + tq->comp_ring.next2proc;
378 	}
379 
380 	if (completed) {
381 		spin_lock(&tq->tx_lock);
382 		if (unlikely(vmxnet3_tq_stopped(tq, adapter) &&
383 			     vmxnet3_cmd_ring_desc_avail(&tq->tx_ring) >
384 			     VMXNET3_WAKE_QUEUE_THRESHOLD(tq) &&
385 			     netif_carrier_ok(adapter->netdev))) {
386 			vmxnet3_tq_wake(tq, adapter);
387 		}
388 		spin_unlock(&tq->tx_lock);
389 	}
390 	return completed;
391 }
392 
393 
394 static void
395 vmxnet3_tq_cleanup(struct vmxnet3_tx_queue *tq,
396 		   struct vmxnet3_adapter *adapter)
397 {
398 	int i;
399 
400 	while (tq->tx_ring.next2comp != tq->tx_ring.next2fill) {
401 		struct vmxnet3_tx_buf_info *tbi;
402 
403 		tbi = tq->buf_info + tq->tx_ring.next2comp;
404 
405 		vmxnet3_unmap_tx_buf(tbi, adapter->pdev);
406 		if (tbi->skb) {
407 			dev_kfree_skb_any(tbi->skb);
408 			tbi->skb = NULL;
409 		}
410 		vmxnet3_cmd_ring_adv_next2comp(&tq->tx_ring);
411 	}
412 
413 	/* sanity check, verify all buffers are indeed unmapped and freed */
414 	for (i = 0; i < tq->tx_ring.size; i++) {
415 		BUG_ON(tq->buf_info[i].skb != NULL ||
416 		       tq->buf_info[i].map_type != VMXNET3_MAP_NONE);
417 	}
418 
419 	tq->tx_ring.gen = VMXNET3_INIT_GEN;
420 	tq->tx_ring.next2fill = tq->tx_ring.next2comp = 0;
421 
422 	tq->comp_ring.gen = VMXNET3_INIT_GEN;
423 	tq->comp_ring.next2proc = 0;
424 }
425 
426 
427 static void
428 vmxnet3_tq_destroy(struct vmxnet3_tx_queue *tq,
429 		   struct vmxnet3_adapter *adapter)
430 {
431 	if (tq->tx_ring.base) {
432 		dma_free_coherent(&adapter->pdev->dev, tq->tx_ring.size *
433 				  sizeof(struct Vmxnet3_TxDesc),
434 				  tq->tx_ring.base, tq->tx_ring.basePA);
435 		tq->tx_ring.base = NULL;
436 	}
437 	if (tq->data_ring.base) {
438 		dma_free_coherent(&adapter->pdev->dev,
439 				  tq->data_ring.size * tq->txdata_desc_size,
440 				  tq->data_ring.base, tq->data_ring.basePA);
441 		tq->data_ring.base = NULL;
442 	}
443 	if (tq->comp_ring.base) {
444 		dma_free_coherent(&adapter->pdev->dev, tq->comp_ring.size *
445 				  sizeof(struct Vmxnet3_TxCompDesc),
446 				  tq->comp_ring.base, tq->comp_ring.basePA);
447 		tq->comp_ring.base = NULL;
448 	}
449 	if (tq->buf_info) {
450 		dma_free_coherent(&adapter->pdev->dev,
451 				  tq->tx_ring.size * sizeof(tq->buf_info[0]),
452 				  tq->buf_info, tq->buf_info_pa);
453 		tq->buf_info = NULL;
454 	}
455 }
456 
457 
458 /* Destroy all tx queues */
459 void
460 vmxnet3_tq_destroy_all(struct vmxnet3_adapter *adapter)
461 {
462 	int i;
463 
464 	for (i = 0; i < adapter->num_tx_queues; i++)
465 		vmxnet3_tq_destroy(&adapter->tx_queue[i], adapter);
466 }
467 
468 
469 static void
470 vmxnet3_tq_init(struct vmxnet3_tx_queue *tq,
471 		struct vmxnet3_adapter *adapter)
472 {
473 	int i;
474 
475 	/* reset the tx ring contents to 0 and reset the tx ring states */
476 	memset(tq->tx_ring.base, 0, tq->tx_ring.size *
477 	       sizeof(struct Vmxnet3_TxDesc));
478 	tq->tx_ring.next2fill = tq->tx_ring.next2comp = 0;
479 	tq->tx_ring.gen = VMXNET3_INIT_GEN;
480 
481 	memset(tq->data_ring.base, 0,
482 	       tq->data_ring.size * tq->txdata_desc_size);
483 
484 	/* reset the tx comp ring contents to 0 and reset comp ring states */
485 	memset(tq->comp_ring.base, 0, tq->comp_ring.size *
486 	       sizeof(struct Vmxnet3_TxCompDesc));
487 	tq->comp_ring.next2proc = 0;
488 	tq->comp_ring.gen = VMXNET3_INIT_GEN;
489 
490 	/* reset the bookkeeping data */
491 	memset(tq->buf_info, 0, sizeof(tq->buf_info[0]) * tq->tx_ring.size);
492 	for (i = 0; i < tq->tx_ring.size; i++)
493 		tq->buf_info[i].map_type = VMXNET3_MAP_NONE;
494 
495 	/* stats are not reset */
496 }
497 
498 
499 static int
500 vmxnet3_tq_create(struct vmxnet3_tx_queue *tq,
501 		  struct vmxnet3_adapter *adapter)
502 {
503 	size_t sz;
504 
505 	BUG_ON(tq->tx_ring.base || tq->data_ring.base ||
506 	       tq->comp_ring.base || tq->buf_info);
507 
508 	tq->tx_ring.base = dma_alloc_coherent(&adapter->pdev->dev,
509 			tq->tx_ring.size * sizeof(struct Vmxnet3_TxDesc),
510 			&tq->tx_ring.basePA, GFP_KERNEL);
511 	if (!tq->tx_ring.base) {
512 		netdev_err(adapter->netdev, "failed to allocate tx ring\n");
513 		goto err;
514 	}
515 
516 	tq->data_ring.base = dma_alloc_coherent(&adapter->pdev->dev,
517 			tq->data_ring.size * tq->txdata_desc_size,
518 			&tq->data_ring.basePA, GFP_KERNEL);
519 	if (!tq->data_ring.base) {
520 		netdev_err(adapter->netdev, "failed to allocate tx data ring\n");
521 		goto err;
522 	}
523 
524 	tq->comp_ring.base = dma_alloc_coherent(&adapter->pdev->dev,
525 			tq->comp_ring.size * sizeof(struct Vmxnet3_TxCompDesc),
526 			&tq->comp_ring.basePA, GFP_KERNEL);
527 	if (!tq->comp_ring.base) {
528 		netdev_err(adapter->netdev, "failed to allocate tx comp ring\n");
529 		goto err;
530 	}
531 
532 	sz = tq->tx_ring.size * sizeof(tq->buf_info[0]);
533 	tq->buf_info = dma_zalloc_coherent(&adapter->pdev->dev, sz,
534 					   &tq->buf_info_pa, GFP_KERNEL);
535 	if (!tq->buf_info)
536 		goto err;
537 
538 	return 0;
539 
540 err:
541 	vmxnet3_tq_destroy(tq, adapter);
542 	return -ENOMEM;
543 }
544 
545 static void
546 vmxnet3_tq_cleanup_all(struct vmxnet3_adapter *adapter)
547 {
548 	int i;
549 
550 	for (i = 0; i < adapter->num_tx_queues; i++)
551 		vmxnet3_tq_cleanup(&adapter->tx_queue[i], adapter);
552 }
553 
554 /*
555  *    starting from ring->next2fill, allocate rx buffers for the given ring
556  *    of the rx queue and update the rx desc. stop after @num_to_alloc buffers
557  *    are allocated or allocation fails
558  */
559 
560 static int
561 vmxnet3_rq_alloc_rx_buf(struct vmxnet3_rx_queue *rq, u32 ring_idx,
562 			int num_to_alloc, struct vmxnet3_adapter *adapter)
563 {
564 	int num_allocated = 0;
565 	struct vmxnet3_rx_buf_info *rbi_base = rq->buf_info[ring_idx];
566 	struct vmxnet3_cmd_ring *ring = &rq->rx_ring[ring_idx];
567 	u32 val;
568 
569 	while (num_allocated <= num_to_alloc) {
570 		struct vmxnet3_rx_buf_info *rbi;
571 		union Vmxnet3_GenericDesc *gd;
572 
573 		rbi = rbi_base + ring->next2fill;
574 		gd = ring->base + ring->next2fill;
575 
576 		if (rbi->buf_type == VMXNET3_RX_BUF_SKB) {
577 			if (rbi->skb == NULL) {
578 				rbi->skb = __netdev_alloc_skb_ip_align(adapter->netdev,
579 								       rbi->len,
580 								       GFP_KERNEL);
581 				if (unlikely(rbi->skb == NULL)) {
582 					rq->stats.rx_buf_alloc_failure++;
583 					break;
584 				}
585 
586 				rbi->dma_addr = dma_map_single(
587 						&adapter->pdev->dev,
588 						rbi->skb->data, rbi->len,
589 						PCI_DMA_FROMDEVICE);
590 				if (dma_mapping_error(&adapter->pdev->dev,
591 						      rbi->dma_addr)) {
592 					dev_kfree_skb_any(rbi->skb);
593 					rq->stats.rx_buf_alloc_failure++;
594 					break;
595 				}
596 			} else {
597 				/* rx buffer skipped by the device */
598 			}
599 			val = VMXNET3_RXD_BTYPE_HEAD << VMXNET3_RXD_BTYPE_SHIFT;
600 		} else {
601 			BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_PAGE ||
602 			       rbi->len  != PAGE_SIZE);
603 
604 			if (rbi->page == NULL) {
605 				rbi->page = alloc_page(GFP_ATOMIC);
606 				if (unlikely(rbi->page == NULL)) {
607 					rq->stats.rx_buf_alloc_failure++;
608 					break;
609 				}
610 				rbi->dma_addr = dma_map_page(
611 						&adapter->pdev->dev,
612 						rbi->page, 0, PAGE_SIZE,
613 						PCI_DMA_FROMDEVICE);
614 				if (dma_mapping_error(&adapter->pdev->dev,
615 						      rbi->dma_addr)) {
616 					put_page(rbi->page);
617 					rq->stats.rx_buf_alloc_failure++;
618 					break;
619 				}
620 			} else {
621 				/* rx buffers skipped by the device */
622 			}
623 			val = VMXNET3_RXD_BTYPE_BODY << VMXNET3_RXD_BTYPE_SHIFT;
624 		}
625 
626 		gd->rxd.addr = cpu_to_le64(rbi->dma_addr);
627 		gd->dword[2] = cpu_to_le32((!ring->gen << VMXNET3_RXD_GEN_SHIFT)
628 					   | val | rbi->len);
629 
630 		/* Fill the last buffer but dont mark it ready, or else the
631 		 * device will think that the queue is full */
632 		if (num_allocated == num_to_alloc)
633 			break;
634 
635 		gd->dword[2] |= cpu_to_le32(ring->gen << VMXNET3_RXD_GEN_SHIFT);
636 		num_allocated++;
637 		vmxnet3_cmd_ring_adv_next2fill(ring);
638 	}
639 
640 	netdev_dbg(adapter->netdev,
641 		"alloc_rx_buf: %d allocated, next2fill %u, next2comp %u\n",
642 		num_allocated, ring->next2fill, ring->next2comp);
643 
644 	/* so that the device can distinguish a full ring and an empty ring */
645 	BUG_ON(num_allocated != 0 && ring->next2fill == ring->next2comp);
646 
647 	return num_allocated;
648 }
649 
650 
651 static void
652 vmxnet3_append_frag(struct sk_buff *skb, struct Vmxnet3_RxCompDesc *rcd,
653 		    struct vmxnet3_rx_buf_info *rbi)
654 {
655 	struct skb_frag_struct *frag = skb_shinfo(skb)->frags +
656 		skb_shinfo(skb)->nr_frags;
657 
658 	BUG_ON(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS);
659 
660 	__skb_frag_set_page(frag, rbi->page);
661 	frag->page_offset = 0;
662 	skb_frag_size_set(frag, rcd->len);
663 	skb->data_len += rcd->len;
664 	skb->truesize += PAGE_SIZE;
665 	skb_shinfo(skb)->nr_frags++;
666 }
667 
668 
669 static int
670 vmxnet3_map_pkt(struct sk_buff *skb, struct vmxnet3_tx_ctx *ctx,
671 		struct vmxnet3_tx_queue *tq, struct pci_dev *pdev,
672 		struct vmxnet3_adapter *adapter)
673 {
674 	u32 dw2, len;
675 	unsigned long buf_offset;
676 	int i;
677 	union Vmxnet3_GenericDesc *gdesc;
678 	struct vmxnet3_tx_buf_info *tbi = NULL;
679 
680 	BUG_ON(ctx->copy_size > skb_headlen(skb));
681 
682 	/* use the previous gen bit for the SOP desc */
683 	dw2 = (tq->tx_ring.gen ^ 0x1) << VMXNET3_TXD_GEN_SHIFT;
684 
685 	ctx->sop_txd = tq->tx_ring.base + tq->tx_ring.next2fill;
686 	gdesc = ctx->sop_txd; /* both loops below can be skipped */
687 
688 	/* no need to map the buffer if headers are copied */
689 	if (ctx->copy_size) {
690 		ctx->sop_txd->txd.addr = cpu_to_le64(tq->data_ring.basePA +
691 					tq->tx_ring.next2fill *
692 					tq->txdata_desc_size);
693 		ctx->sop_txd->dword[2] = cpu_to_le32(dw2 | ctx->copy_size);
694 		ctx->sop_txd->dword[3] = 0;
695 
696 		tbi = tq->buf_info + tq->tx_ring.next2fill;
697 		tbi->map_type = VMXNET3_MAP_NONE;
698 
699 		netdev_dbg(adapter->netdev,
700 			"txd[%u]: 0x%Lx 0x%x 0x%x\n",
701 			tq->tx_ring.next2fill,
702 			le64_to_cpu(ctx->sop_txd->txd.addr),
703 			ctx->sop_txd->dword[2], ctx->sop_txd->dword[3]);
704 		vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
705 
706 		/* use the right gen for non-SOP desc */
707 		dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
708 	}
709 
710 	/* linear part can use multiple tx desc if it's big */
711 	len = skb_headlen(skb) - ctx->copy_size;
712 	buf_offset = ctx->copy_size;
713 	while (len) {
714 		u32 buf_size;
715 
716 		if (len < VMXNET3_MAX_TX_BUF_SIZE) {
717 			buf_size = len;
718 			dw2 |= len;
719 		} else {
720 			buf_size = VMXNET3_MAX_TX_BUF_SIZE;
721 			/* spec says that for TxDesc.len, 0 == 2^14 */
722 		}
723 
724 		tbi = tq->buf_info + tq->tx_ring.next2fill;
725 		tbi->map_type = VMXNET3_MAP_SINGLE;
726 		tbi->dma_addr = dma_map_single(&adapter->pdev->dev,
727 				skb->data + buf_offset, buf_size,
728 				PCI_DMA_TODEVICE);
729 		if (dma_mapping_error(&adapter->pdev->dev, tbi->dma_addr))
730 			return -EFAULT;
731 
732 		tbi->len = buf_size;
733 
734 		gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
735 		BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
736 
737 		gdesc->txd.addr = cpu_to_le64(tbi->dma_addr);
738 		gdesc->dword[2] = cpu_to_le32(dw2);
739 		gdesc->dword[3] = 0;
740 
741 		netdev_dbg(adapter->netdev,
742 			"txd[%u]: 0x%Lx 0x%x 0x%x\n",
743 			tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr),
744 			le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]);
745 		vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
746 		dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
747 
748 		len -= buf_size;
749 		buf_offset += buf_size;
750 	}
751 
752 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
753 		const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
754 		u32 buf_size;
755 
756 		buf_offset = 0;
757 		len = skb_frag_size(frag);
758 		while (len) {
759 			tbi = tq->buf_info + tq->tx_ring.next2fill;
760 			if (len < VMXNET3_MAX_TX_BUF_SIZE) {
761 				buf_size = len;
762 				dw2 |= len;
763 			} else {
764 				buf_size = VMXNET3_MAX_TX_BUF_SIZE;
765 				/* spec says that for TxDesc.len, 0 == 2^14 */
766 			}
767 			tbi->map_type = VMXNET3_MAP_PAGE;
768 			tbi->dma_addr = skb_frag_dma_map(&adapter->pdev->dev, frag,
769 							 buf_offset, buf_size,
770 							 DMA_TO_DEVICE);
771 			if (dma_mapping_error(&adapter->pdev->dev, tbi->dma_addr))
772 				return -EFAULT;
773 
774 			tbi->len = buf_size;
775 
776 			gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
777 			BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
778 
779 			gdesc->txd.addr = cpu_to_le64(tbi->dma_addr);
780 			gdesc->dword[2] = cpu_to_le32(dw2);
781 			gdesc->dword[3] = 0;
782 
783 			netdev_dbg(adapter->netdev,
784 				"txd[%u]: 0x%llx %u %u\n",
785 				tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr),
786 				le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]);
787 			vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
788 			dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
789 
790 			len -= buf_size;
791 			buf_offset += buf_size;
792 		}
793 	}
794 
795 	ctx->eop_txd = gdesc;
796 
797 	/* set the last buf_info for the pkt */
798 	tbi->skb = skb;
799 	tbi->sop_idx = ctx->sop_txd - tq->tx_ring.base;
800 
801 	return 0;
802 }
803 
804 
805 /* Init all tx queues */
806 static void
807 vmxnet3_tq_init_all(struct vmxnet3_adapter *adapter)
808 {
809 	int i;
810 
811 	for (i = 0; i < adapter->num_tx_queues; i++)
812 		vmxnet3_tq_init(&adapter->tx_queue[i], adapter);
813 }
814 
815 
816 /*
817  *    parse relevant protocol headers:
818  *      For a tso pkt, relevant headers are L2/3/4 including options
819  *      For a pkt requesting csum offloading, they are L2/3 and may include L4
820  *      if it's a TCP/UDP pkt
821  *
822  * Returns:
823  *    -1:  error happens during parsing
824  *     0:  protocol headers parsed, but too big to be copied
825  *     1:  protocol headers parsed and copied
826  *
827  * Other effects:
828  *    1. related *ctx fields are updated.
829  *    2. ctx->copy_size is # of bytes copied
830  *    3. the portion to be copied is guaranteed to be in the linear part
831  *
832  */
833 static int
834 vmxnet3_parse_hdr(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
835 		  struct vmxnet3_tx_ctx *ctx,
836 		  struct vmxnet3_adapter *adapter)
837 {
838 	u8 protocol = 0;
839 
840 	if (ctx->mss) {	/* TSO */
841 		ctx->eth_ip_hdr_size = skb_transport_offset(skb);
842 		ctx->l4_hdr_size = tcp_hdrlen(skb);
843 		ctx->copy_size = ctx->eth_ip_hdr_size + ctx->l4_hdr_size;
844 	} else {
845 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
846 			ctx->eth_ip_hdr_size = skb_checksum_start_offset(skb);
847 
848 			if (ctx->ipv4) {
849 				const struct iphdr *iph = ip_hdr(skb);
850 
851 				protocol = iph->protocol;
852 			} else if (ctx->ipv6) {
853 				const struct ipv6hdr *ipv6h = ipv6_hdr(skb);
854 
855 				protocol = ipv6h->nexthdr;
856 			}
857 
858 			switch (protocol) {
859 			case IPPROTO_TCP:
860 				ctx->l4_hdr_size = tcp_hdrlen(skb);
861 				break;
862 			case IPPROTO_UDP:
863 				ctx->l4_hdr_size = sizeof(struct udphdr);
864 				break;
865 			default:
866 				ctx->l4_hdr_size = 0;
867 				break;
868 			}
869 
870 			ctx->copy_size = min(ctx->eth_ip_hdr_size +
871 					 ctx->l4_hdr_size, skb->len);
872 		} else {
873 			ctx->eth_ip_hdr_size = 0;
874 			ctx->l4_hdr_size = 0;
875 			/* copy as much as allowed */
876 			ctx->copy_size = min_t(unsigned int,
877 					       tq->txdata_desc_size,
878 					       skb_headlen(skb));
879 		}
880 
881 		if (skb->len <= VMXNET3_HDR_COPY_SIZE)
882 			ctx->copy_size = skb->len;
883 
884 		/* make sure headers are accessible directly */
885 		if (unlikely(!pskb_may_pull(skb, ctx->copy_size)))
886 			goto err;
887 	}
888 
889 	if (unlikely(ctx->copy_size > tq->txdata_desc_size)) {
890 		tq->stats.oversized_hdr++;
891 		ctx->copy_size = 0;
892 		return 0;
893 	}
894 
895 	return 1;
896 err:
897 	return -1;
898 }
899 
900 /*
901  *    copy relevant protocol headers to the transmit ring:
902  *      For a tso pkt, relevant headers are L2/3/4 including options
903  *      For a pkt requesting csum offloading, they are L2/3 and may include L4
904  *      if it's a TCP/UDP pkt
905  *
906  *
907  *    Note that this requires that vmxnet3_parse_hdr be called first to set the
908  *      appropriate bits in ctx first
909  */
910 static void
911 vmxnet3_copy_hdr(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
912 		 struct vmxnet3_tx_ctx *ctx,
913 		 struct vmxnet3_adapter *adapter)
914 {
915 	struct Vmxnet3_TxDataDesc *tdd;
916 
917 	tdd = (struct Vmxnet3_TxDataDesc *)((u8 *)tq->data_ring.base +
918 					    tq->tx_ring.next2fill *
919 					    tq->txdata_desc_size);
920 
921 	memcpy(tdd->data, skb->data, ctx->copy_size);
922 	netdev_dbg(adapter->netdev,
923 		"copy %u bytes to dataRing[%u]\n",
924 		ctx->copy_size, tq->tx_ring.next2fill);
925 }
926 
927 
928 static void
929 vmxnet3_prepare_tso(struct sk_buff *skb,
930 		    struct vmxnet3_tx_ctx *ctx)
931 {
932 	struct tcphdr *tcph = tcp_hdr(skb);
933 
934 	if (ctx->ipv4) {
935 		struct iphdr *iph = ip_hdr(skb);
936 
937 		iph->check = 0;
938 		tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, 0,
939 						 IPPROTO_TCP, 0);
940 	} else if (ctx->ipv6) {
941 		struct ipv6hdr *iph = ipv6_hdr(skb);
942 
943 		tcph->check = ~csum_ipv6_magic(&iph->saddr, &iph->daddr, 0,
944 					       IPPROTO_TCP, 0);
945 	}
946 }
947 
948 static int txd_estimate(const struct sk_buff *skb)
949 {
950 	int count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) + 1;
951 	int i;
952 
953 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
954 		const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
955 
956 		count += VMXNET3_TXD_NEEDED(skb_frag_size(frag));
957 	}
958 	return count;
959 }
960 
961 /*
962  * Transmits a pkt thru a given tq
963  * Returns:
964  *    NETDEV_TX_OK:      descriptors are setup successfully
965  *    NETDEV_TX_OK:      error occurred, the pkt is dropped
966  *    NETDEV_TX_BUSY:    tx ring is full, queue is stopped
967  *
968  * Side-effects:
969  *    1. tx ring may be changed
970  *    2. tq stats may be updated accordingly
971  *    3. shared->txNumDeferred may be updated
972  */
973 
974 static int
975 vmxnet3_tq_xmit(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
976 		struct vmxnet3_adapter *adapter, struct net_device *netdev)
977 {
978 	int ret;
979 	u32 count;
980 	unsigned long flags;
981 	struct vmxnet3_tx_ctx ctx;
982 	union Vmxnet3_GenericDesc *gdesc;
983 #ifdef __BIG_ENDIAN_BITFIELD
984 	/* Use temporary descriptor to avoid touching bits multiple times */
985 	union Vmxnet3_GenericDesc tempTxDesc;
986 #endif
987 
988 	count = txd_estimate(skb);
989 
990 	ctx.ipv4 = (vlan_get_protocol(skb) == cpu_to_be16(ETH_P_IP));
991 	ctx.ipv6 = (vlan_get_protocol(skb) == cpu_to_be16(ETH_P_IPV6));
992 
993 	ctx.mss = skb_shinfo(skb)->gso_size;
994 	if (ctx.mss) {
995 		if (skb_header_cloned(skb)) {
996 			if (unlikely(pskb_expand_head(skb, 0, 0,
997 						      GFP_ATOMIC) != 0)) {
998 				tq->stats.drop_tso++;
999 				goto drop_pkt;
1000 			}
1001 			tq->stats.copy_skb_header++;
1002 		}
1003 		vmxnet3_prepare_tso(skb, &ctx);
1004 	} else {
1005 		if (unlikely(count > VMXNET3_MAX_TXD_PER_PKT)) {
1006 
1007 			/* non-tso pkts must not use more than
1008 			 * VMXNET3_MAX_TXD_PER_PKT entries
1009 			 */
1010 			if (skb_linearize(skb) != 0) {
1011 				tq->stats.drop_too_many_frags++;
1012 				goto drop_pkt;
1013 			}
1014 			tq->stats.linearized++;
1015 
1016 			/* recalculate the # of descriptors to use */
1017 			count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) + 1;
1018 		}
1019 	}
1020 
1021 	ret = vmxnet3_parse_hdr(skb, tq, &ctx, adapter);
1022 	if (ret >= 0) {
1023 		BUG_ON(ret <= 0 && ctx.copy_size != 0);
1024 		/* hdrs parsed, check against other limits */
1025 		if (ctx.mss) {
1026 			if (unlikely(ctx.eth_ip_hdr_size + ctx.l4_hdr_size >
1027 				     VMXNET3_MAX_TX_BUF_SIZE)) {
1028 				tq->stats.drop_oversized_hdr++;
1029 				goto drop_pkt;
1030 			}
1031 		} else {
1032 			if (skb->ip_summed == CHECKSUM_PARTIAL) {
1033 				if (unlikely(ctx.eth_ip_hdr_size +
1034 					     skb->csum_offset >
1035 					     VMXNET3_MAX_CSUM_OFFSET)) {
1036 					tq->stats.drop_oversized_hdr++;
1037 					goto drop_pkt;
1038 				}
1039 			}
1040 		}
1041 	} else {
1042 		tq->stats.drop_hdr_inspect_err++;
1043 		goto drop_pkt;
1044 	}
1045 
1046 	spin_lock_irqsave(&tq->tx_lock, flags);
1047 
1048 	if (count > vmxnet3_cmd_ring_desc_avail(&tq->tx_ring)) {
1049 		tq->stats.tx_ring_full++;
1050 		netdev_dbg(adapter->netdev,
1051 			"tx queue stopped on %s, next2comp %u"
1052 			" next2fill %u\n", adapter->netdev->name,
1053 			tq->tx_ring.next2comp, tq->tx_ring.next2fill);
1054 
1055 		vmxnet3_tq_stop(tq, adapter);
1056 		spin_unlock_irqrestore(&tq->tx_lock, flags);
1057 		return NETDEV_TX_BUSY;
1058 	}
1059 
1060 
1061 	vmxnet3_copy_hdr(skb, tq, &ctx, adapter);
1062 
1063 	/* fill tx descs related to addr & len */
1064 	if (vmxnet3_map_pkt(skb, &ctx, tq, adapter->pdev, adapter))
1065 		goto unlock_drop_pkt;
1066 
1067 	/* setup the EOP desc */
1068 	ctx.eop_txd->dword[3] = cpu_to_le32(VMXNET3_TXD_CQ | VMXNET3_TXD_EOP);
1069 
1070 	/* setup the SOP desc */
1071 #ifdef __BIG_ENDIAN_BITFIELD
1072 	gdesc = &tempTxDesc;
1073 	gdesc->dword[2] = ctx.sop_txd->dword[2];
1074 	gdesc->dword[3] = ctx.sop_txd->dword[3];
1075 #else
1076 	gdesc = ctx.sop_txd;
1077 #endif
1078 	if (ctx.mss) {
1079 		gdesc->txd.hlen = ctx.eth_ip_hdr_size + ctx.l4_hdr_size;
1080 		gdesc->txd.om = VMXNET3_OM_TSO;
1081 		gdesc->txd.msscof = ctx.mss;
1082 		le32_add_cpu(&tq->shared->txNumDeferred, (skb->len -
1083 			     gdesc->txd.hlen + ctx.mss - 1) / ctx.mss);
1084 	} else {
1085 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
1086 			gdesc->txd.hlen = ctx.eth_ip_hdr_size;
1087 			gdesc->txd.om = VMXNET3_OM_CSUM;
1088 			gdesc->txd.msscof = ctx.eth_ip_hdr_size +
1089 					    skb->csum_offset;
1090 		} else {
1091 			gdesc->txd.om = 0;
1092 			gdesc->txd.msscof = 0;
1093 		}
1094 		le32_add_cpu(&tq->shared->txNumDeferred, 1);
1095 	}
1096 
1097 	if (skb_vlan_tag_present(skb)) {
1098 		gdesc->txd.ti = 1;
1099 		gdesc->txd.tci = skb_vlan_tag_get(skb);
1100 	}
1101 
1102 	/* finally flips the GEN bit of the SOP desc. */
1103 	gdesc->dword[2] = cpu_to_le32(le32_to_cpu(gdesc->dword[2]) ^
1104 						  VMXNET3_TXD_GEN);
1105 #ifdef __BIG_ENDIAN_BITFIELD
1106 	/* Finished updating in bitfields of Tx Desc, so write them in original
1107 	 * place.
1108 	 */
1109 	vmxnet3_TxDescToLe((struct Vmxnet3_TxDesc *)gdesc,
1110 			   (struct Vmxnet3_TxDesc *)ctx.sop_txd);
1111 	gdesc = ctx.sop_txd;
1112 #endif
1113 	netdev_dbg(adapter->netdev,
1114 		"txd[%u]: SOP 0x%Lx 0x%x 0x%x\n",
1115 		(u32)(ctx.sop_txd -
1116 		tq->tx_ring.base), le64_to_cpu(gdesc->txd.addr),
1117 		le32_to_cpu(gdesc->dword[2]), le32_to_cpu(gdesc->dword[3]));
1118 
1119 	spin_unlock_irqrestore(&tq->tx_lock, flags);
1120 
1121 	if (le32_to_cpu(tq->shared->txNumDeferred) >=
1122 					le32_to_cpu(tq->shared->txThreshold)) {
1123 		tq->shared->txNumDeferred = 0;
1124 		VMXNET3_WRITE_BAR0_REG(adapter,
1125 				       VMXNET3_REG_TXPROD + tq->qid * 8,
1126 				       tq->tx_ring.next2fill);
1127 	}
1128 
1129 	return NETDEV_TX_OK;
1130 
1131 unlock_drop_pkt:
1132 	spin_unlock_irqrestore(&tq->tx_lock, flags);
1133 drop_pkt:
1134 	tq->stats.drop_total++;
1135 	dev_kfree_skb_any(skb);
1136 	return NETDEV_TX_OK;
1137 }
1138 
1139 
1140 static netdev_tx_t
1141 vmxnet3_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
1142 {
1143 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1144 
1145 	BUG_ON(skb->queue_mapping > adapter->num_tx_queues);
1146 	return vmxnet3_tq_xmit(skb,
1147 			       &adapter->tx_queue[skb->queue_mapping],
1148 			       adapter, netdev);
1149 }
1150 
1151 
1152 static void
1153 vmxnet3_rx_csum(struct vmxnet3_adapter *adapter,
1154 		struct sk_buff *skb,
1155 		union Vmxnet3_GenericDesc *gdesc)
1156 {
1157 	if (!gdesc->rcd.cnc && adapter->netdev->features & NETIF_F_RXCSUM) {
1158 		if (gdesc->rcd.v4 &&
1159 		    (le32_to_cpu(gdesc->dword[3]) &
1160 		     VMXNET3_RCD_CSUM_OK) == VMXNET3_RCD_CSUM_OK) {
1161 			skb->ip_summed = CHECKSUM_UNNECESSARY;
1162 			BUG_ON(!(gdesc->rcd.tcp || gdesc->rcd.udp));
1163 			BUG_ON(gdesc->rcd.frg);
1164 		} else if (gdesc->rcd.v6 && (le32_to_cpu(gdesc->dword[3]) &
1165 					     (1 << VMXNET3_RCD_TUC_SHIFT))) {
1166 			skb->ip_summed = CHECKSUM_UNNECESSARY;
1167 			BUG_ON(!(gdesc->rcd.tcp || gdesc->rcd.udp));
1168 			BUG_ON(gdesc->rcd.frg);
1169 		} else {
1170 			if (gdesc->rcd.csum) {
1171 				skb->csum = htons(gdesc->rcd.csum);
1172 				skb->ip_summed = CHECKSUM_PARTIAL;
1173 			} else {
1174 				skb_checksum_none_assert(skb);
1175 			}
1176 		}
1177 	} else {
1178 		skb_checksum_none_assert(skb);
1179 	}
1180 }
1181 
1182 
1183 static void
1184 vmxnet3_rx_error(struct vmxnet3_rx_queue *rq, struct Vmxnet3_RxCompDesc *rcd,
1185 		 struct vmxnet3_rx_ctx *ctx,  struct vmxnet3_adapter *adapter)
1186 {
1187 	rq->stats.drop_err++;
1188 	if (!rcd->fcs)
1189 		rq->stats.drop_fcs++;
1190 
1191 	rq->stats.drop_total++;
1192 
1193 	/*
1194 	 * We do not unmap and chain the rx buffer to the skb.
1195 	 * We basically pretend this buffer is not used and will be recycled
1196 	 * by vmxnet3_rq_alloc_rx_buf()
1197 	 */
1198 
1199 	/*
1200 	 * ctx->skb may be NULL if this is the first and the only one
1201 	 * desc for the pkt
1202 	 */
1203 	if (ctx->skb)
1204 		dev_kfree_skb_irq(ctx->skb);
1205 
1206 	ctx->skb = NULL;
1207 }
1208 
1209 
1210 static u32
1211 vmxnet3_get_hdr_len(struct vmxnet3_adapter *adapter, struct sk_buff *skb,
1212 		    union Vmxnet3_GenericDesc *gdesc)
1213 {
1214 	u32 hlen, maplen;
1215 	union {
1216 		void *ptr;
1217 		struct ethhdr *eth;
1218 		struct iphdr *ipv4;
1219 		struct ipv6hdr *ipv6;
1220 		struct tcphdr *tcp;
1221 	} hdr;
1222 	BUG_ON(gdesc->rcd.tcp == 0);
1223 
1224 	maplen = skb_headlen(skb);
1225 	if (unlikely(sizeof(struct iphdr) + sizeof(struct tcphdr) > maplen))
1226 		return 0;
1227 
1228 	hdr.eth = eth_hdr(skb);
1229 	if (gdesc->rcd.v4) {
1230 		BUG_ON(hdr.eth->h_proto != htons(ETH_P_IP));
1231 		hdr.ptr += sizeof(struct ethhdr);
1232 		BUG_ON(hdr.ipv4->protocol != IPPROTO_TCP);
1233 		hlen = hdr.ipv4->ihl << 2;
1234 		hdr.ptr += hdr.ipv4->ihl << 2;
1235 	} else if (gdesc->rcd.v6) {
1236 		BUG_ON(hdr.eth->h_proto != htons(ETH_P_IPV6));
1237 		hdr.ptr += sizeof(struct ethhdr);
1238 		/* Use an estimated value, since we also need to handle
1239 		 * TSO case.
1240 		 */
1241 		if (hdr.ipv6->nexthdr != IPPROTO_TCP)
1242 			return sizeof(struct ipv6hdr) + sizeof(struct tcphdr);
1243 		hlen = sizeof(struct ipv6hdr);
1244 		hdr.ptr += sizeof(struct ipv6hdr);
1245 	} else {
1246 		/* Non-IP pkt, dont estimate header length */
1247 		return 0;
1248 	}
1249 
1250 	if (hlen + sizeof(struct tcphdr) > maplen)
1251 		return 0;
1252 
1253 	return (hlen + (hdr.tcp->doff << 2));
1254 }
1255 
1256 static int
1257 vmxnet3_rq_rx_complete(struct vmxnet3_rx_queue *rq,
1258 		       struct vmxnet3_adapter *adapter, int quota)
1259 {
1260 	static const u32 rxprod_reg[2] = {
1261 		VMXNET3_REG_RXPROD, VMXNET3_REG_RXPROD2
1262 	};
1263 	u32 num_pkts = 0;
1264 	bool skip_page_frags = false;
1265 	struct Vmxnet3_RxCompDesc *rcd;
1266 	struct vmxnet3_rx_ctx *ctx = &rq->rx_ctx;
1267 	u16 segCnt = 0, mss = 0;
1268 #ifdef __BIG_ENDIAN_BITFIELD
1269 	struct Vmxnet3_RxDesc rxCmdDesc;
1270 	struct Vmxnet3_RxCompDesc rxComp;
1271 #endif
1272 	vmxnet3_getRxComp(rcd, &rq->comp_ring.base[rq->comp_ring.next2proc].rcd,
1273 			  &rxComp);
1274 	while (rcd->gen == rq->comp_ring.gen) {
1275 		struct vmxnet3_rx_buf_info *rbi;
1276 		struct sk_buff *skb, *new_skb = NULL;
1277 		struct page *new_page = NULL;
1278 		dma_addr_t new_dma_addr;
1279 		int num_to_alloc;
1280 		struct Vmxnet3_RxDesc *rxd;
1281 		u32 idx, ring_idx;
1282 		struct vmxnet3_cmd_ring	*ring = NULL;
1283 		if (num_pkts >= quota) {
1284 			/* we may stop even before we see the EOP desc of
1285 			 * the current pkt
1286 			 */
1287 			break;
1288 		}
1289 		BUG_ON(rcd->rqID != rq->qid && rcd->rqID != rq->qid2 &&
1290 		       rcd->rqID != rq->dataRingQid);
1291 		idx = rcd->rxdIdx;
1292 		ring_idx = VMXNET3_GET_RING_IDX(adapter, rcd->rqID);
1293 		ring = rq->rx_ring + ring_idx;
1294 		vmxnet3_getRxDesc(rxd, &rq->rx_ring[ring_idx].base[idx].rxd,
1295 				  &rxCmdDesc);
1296 		rbi = rq->buf_info[ring_idx] + idx;
1297 
1298 		BUG_ON(rxd->addr != rbi->dma_addr ||
1299 		       rxd->len != rbi->len);
1300 
1301 		if (unlikely(rcd->eop && rcd->err)) {
1302 			vmxnet3_rx_error(rq, rcd, ctx, adapter);
1303 			goto rcd_done;
1304 		}
1305 
1306 		if (rcd->sop) { /* first buf of the pkt */
1307 			bool rxDataRingUsed;
1308 			u16 len;
1309 
1310 			BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_HEAD ||
1311 			       (rcd->rqID != rq->qid &&
1312 				rcd->rqID != rq->dataRingQid));
1313 
1314 			BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_SKB);
1315 			BUG_ON(ctx->skb != NULL || rbi->skb == NULL);
1316 
1317 			if (unlikely(rcd->len == 0)) {
1318 				/* Pretend the rx buffer is skipped. */
1319 				BUG_ON(!(rcd->sop && rcd->eop));
1320 				netdev_dbg(adapter->netdev,
1321 					"rxRing[%u][%u] 0 length\n",
1322 					ring_idx, idx);
1323 				goto rcd_done;
1324 			}
1325 
1326 			skip_page_frags = false;
1327 			ctx->skb = rbi->skb;
1328 
1329 			rxDataRingUsed =
1330 				VMXNET3_RX_DATA_RING(adapter, rcd->rqID);
1331 			len = rxDataRingUsed ? rcd->len : rbi->len;
1332 			new_skb = netdev_alloc_skb_ip_align(adapter->netdev,
1333 							    len);
1334 			if (new_skb == NULL) {
1335 				/* Skb allocation failed, do not handover this
1336 				 * skb to stack. Reuse it. Drop the existing pkt
1337 				 */
1338 				rq->stats.rx_buf_alloc_failure++;
1339 				ctx->skb = NULL;
1340 				rq->stats.drop_total++;
1341 				skip_page_frags = true;
1342 				goto rcd_done;
1343 			}
1344 
1345 			if (rxDataRingUsed) {
1346 				size_t sz;
1347 
1348 				BUG_ON(rcd->len > rq->data_ring.desc_size);
1349 
1350 				ctx->skb = new_skb;
1351 				sz = rcd->rxdIdx * rq->data_ring.desc_size;
1352 				memcpy(new_skb->data,
1353 				       &rq->data_ring.base[sz], rcd->len);
1354 			} else {
1355 				ctx->skb = rbi->skb;
1356 
1357 				new_dma_addr =
1358 					dma_map_single(&adapter->pdev->dev,
1359 						       new_skb->data, rbi->len,
1360 						       PCI_DMA_FROMDEVICE);
1361 				if (dma_mapping_error(&adapter->pdev->dev,
1362 						      new_dma_addr)) {
1363 					dev_kfree_skb(new_skb);
1364 					/* Skb allocation failed, do not
1365 					 * handover this skb to stack. Reuse
1366 					 * it. Drop the existing pkt.
1367 					 */
1368 					rq->stats.rx_buf_alloc_failure++;
1369 					ctx->skb = NULL;
1370 					rq->stats.drop_total++;
1371 					skip_page_frags = true;
1372 					goto rcd_done;
1373 				}
1374 
1375 				dma_unmap_single(&adapter->pdev->dev,
1376 						 rbi->dma_addr,
1377 						 rbi->len,
1378 						 PCI_DMA_FROMDEVICE);
1379 
1380 				/* Immediate refill */
1381 				rbi->skb = new_skb;
1382 				rbi->dma_addr = new_dma_addr;
1383 				rxd->addr = cpu_to_le64(rbi->dma_addr);
1384 				rxd->len = rbi->len;
1385 			}
1386 
1387 #ifdef VMXNET3_RSS
1388 			if (rcd->rssType != VMXNET3_RCD_RSS_TYPE_NONE &&
1389 			    (adapter->netdev->features & NETIF_F_RXHASH))
1390 				skb_set_hash(ctx->skb,
1391 					     le32_to_cpu(rcd->rssHash),
1392 					     PKT_HASH_TYPE_L3);
1393 #endif
1394 			skb_put(ctx->skb, rcd->len);
1395 
1396 			if (VMXNET3_VERSION_GE_2(adapter) &&
1397 			    rcd->type == VMXNET3_CDTYPE_RXCOMP_LRO) {
1398 				struct Vmxnet3_RxCompDescExt *rcdlro;
1399 				rcdlro = (struct Vmxnet3_RxCompDescExt *)rcd;
1400 
1401 				segCnt = rcdlro->segCnt;
1402 				WARN_ON_ONCE(segCnt == 0);
1403 				mss = rcdlro->mss;
1404 				if (unlikely(segCnt <= 1))
1405 					segCnt = 0;
1406 			} else {
1407 				segCnt = 0;
1408 			}
1409 		} else {
1410 			BUG_ON(ctx->skb == NULL && !skip_page_frags);
1411 
1412 			/* non SOP buffer must be type 1 in most cases */
1413 			BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_PAGE);
1414 			BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_BODY);
1415 
1416 			/* If an sop buffer was dropped, skip all
1417 			 * following non-sop fragments. They will be reused.
1418 			 */
1419 			if (skip_page_frags)
1420 				goto rcd_done;
1421 
1422 			if (rcd->len) {
1423 				new_page = alloc_page(GFP_ATOMIC);
1424 				/* Replacement page frag could not be allocated.
1425 				 * Reuse this page. Drop the pkt and free the
1426 				 * skb which contained this page as a frag. Skip
1427 				 * processing all the following non-sop frags.
1428 				 */
1429 				if (unlikely(!new_page)) {
1430 					rq->stats.rx_buf_alloc_failure++;
1431 					dev_kfree_skb(ctx->skb);
1432 					ctx->skb = NULL;
1433 					skip_page_frags = true;
1434 					goto rcd_done;
1435 				}
1436 				new_dma_addr = dma_map_page(&adapter->pdev->dev,
1437 							    new_page,
1438 							    0, PAGE_SIZE,
1439 							    PCI_DMA_FROMDEVICE);
1440 				if (dma_mapping_error(&adapter->pdev->dev,
1441 						      new_dma_addr)) {
1442 					put_page(new_page);
1443 					rq->stats.rx_buf_alloc_failure++;
1444 					dev_kfree_skb(ctx->skb);
1445 					ctx->skb = NULL;
1446 					skip_page_frags = true;
1447 					goto rcd_done;
1448 				}
1449 
1450 				dma_unmap_page(&adapter->pdev->dev,
1451 					       rbi->dma_addr, rbi->len,
1452 					       PCI_DMA_FROMDEVICE);
1453 
1454 				vmxnet3_append_frag(ctx->skb, rcd, rbi);
1455 
1456 				/* Immediate refill */
1457 				rbi->page = new_page;
1458 				rbi->dma_addr = new_dma_addr;
1459 				rxd->addr = cpu_to_le64(rbi->dma_addr);
1460 				rxd->len = rbi->len;
1461 			}
1462 		}
1463 
1464 
1465 		skb = ctx->skb;
1466 		if (rcd->eop) {
1467 			u32 mtu = adapter->netdev->mtu;
1468 			skb->len += skb->data_len;
1469 
1470 			vmxnet3_rx_csum(adapter, skb,
1471 					(union Vmxnet3_GenericDesc *)rcd);
1472 			skb->protocol = eth_type_trans(skb, adapter->netdev);
1473 			if (!rcd->tcp || !adapter->lro)
1474 				goto not_lro;
1475 
1476 			if (segCnt != 0 && mss != 0) {
1477 				skb_shinfo(skb)->gso_type = rcd->v4 ?
1478 					SKB_GSO_TCPV4 : SKB_GSO_TCPV6;
1479 				skb_shinfo(skb)->gso_size = mss;
1480 				skb_shinfo(skb)->gso_segs = segCnt;
1481 			} else if (segCnt != 0 || skb->len > mtu) {
1482 				u32 hlen;
1483 
1484 				hlen = vmxnet3_get_hdr_len(adapter, skb,
1485 					(union Vmxnet3_GenericDesc *)rcd);
1486 				if (hlen == 0)
1487 					goto not_lro;
1488 
1489 				skb_shinfo(skb)->gso_type =
1490 					rcd->v4 ? SKB_GSO_TCPV4 : SKB_GSO_TCPV6;
1491 				if (segCnt != 0) {
1492 					skb_shinfo(skb)->gso_segs = segCnt;
1493 					skb_shinfo(skb)->gso_size =
1494 						DIV_ROUND_UP(skb->len -
1495 							hlen, segCnt);
1496 				} else {
1497 					skb_shinfo(skb)->gso_size = mtu - hlen;
1498 				}
1499 			}
1500 not_lro:
1501 			if (unlikely(rcd->ts))
1502 				__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), rcd->tci);
1503 
1504 			if (adapter->netdev->features & NETIF_F_LRO)
1505 				netif_receive_skb(skb);
1506 			else
1507 				napi_gro_receive(&rq->napi, skb);
1508 
1509 			ctx->skb = NULL;
1510 			num_pkts++;
1511 		}
1512 
1513 rcd_done:
1514 		/* device may have skipped some rx descs */
1515 		ring->next2comp = idx;
1516 		num_to_alloc = vmxnet3_cmd_ring_desc_avail(ring);
1517 		ring = rq->rx_ring + ring_idx;
1518 		while (num_to_alloc) {
1519 			vmxnet3_getRxDesc(rxd, &ring->base[ring->next2fill].rxd,
1520 					  &rxCmdDesc);
1521 			BUG_ON(!rxd->addr);
1522 
1523 			/* Recv desc is ready to be used by the device */
1524 			rxd->gen = ring->gen;
1525 			vmxnet3_cmd_ring_adv_next2fill(ring);
1526 			num_to_alloc--;
1527 		}
1528 
1529 		/* if needed, update the register */
1530 		if (unlikely(rq->shared->updateRxProd)) {
1531 			VMXNET3_WRITE_BAR0_REG(adapter,
1532 					       rxprod_reg[ring_idx] + rq->qid * 8,
1533 					       ring->next2fill);
1534 		}
1535 
1536 		vmxnet3_comp_ring_adv_next2proc(&rq->comp_ring);
1537 		vmxnet3_getRxComp(rcd,
1538 				  &rq->comp_ring.base[rq->comp_ring.next2proc].rcd, &rxComp);
1539 	}
1540 
1541 	return num_pkts;
1542 }
1543 
1544 
1545 static void
1546 vmxnet3_rq_cleanup(struct vmxnet3_rx_queue *rq,
1547 		   struct vmxnet3_adapter *adapter)
1548 {
1549 	u32 i, ring_idx;
1550 	struct Vmxnet3_RxDesc *rxd;
1551 
1552 	for (ring_idx = 0; ring_idx < 2; ring_idx++) {
1553 		for (i = 0; i < rq->rx_ring[ring_idx].size; i++) {
1554 #ifdef __BIG_ENDIAN_BITFIELD
1555 			struct Vmxnet3_RxDesc rxDesc;
1556 #endif
1557 			vmxnet3_getRxDesc(rxd,
1558 				&rq->rx_ring[ring_idx].base[i].rxd, &rxDesc);
1559 
1560 			if (rxd->btype == VMXNET3_RXD_BTYPE_HEAD &&
1561 					rq->buf_info[ring_idx][i].skb) {
1562 				dma_unmap_single(&adapter->pdev->dev, rxd->addr,
1563 						 rxd->len, PCI_DMA_FROMDEVICE);
1564 				dev_kfree_skb(rq->buf_info[ring_idx][i].skb);
1565 				rq->buf_info[ring_idx][i].skb = NULL;
1566 			} else if (rxd->btype == VMXNET3_RXD_BTYPE_BODY &&
1567 					rq->buf_info[ring_idx][i].page) {
1568 				dma_unmap_page(&adapter->pdev->dev, rxd->addr,
1569 					       rxd->len, PCI_DMA_FROMDEVICE);
1570 				put_page(rq->buf_info[ring_idx][i].page);
1571 				rq->buf_info[ring_idx][i].page = NULL;
1572 			}
1573 		}
1574 
1575 		rq->rx_ring[ring_idx].gen = VMXNET3_INIT_GEN;
1576 		rq->rx_ring[ring_idx].next2fill =
1577 					rq->rx_ring[ring_idx].next2comp = 0;
1578 	}
1579 
1580 	rq->comp_ring.gen = VMXNET3_INIT_GEN;
1581 	rq->comp_ring.next2proc = 0;
1582 }
1583 
1584 
1585 static void
1586 vmxnet3_rq_cleanup_all(struct vmxnet3_adapter *adapter)
1587 {
1588 	int i;
1589 
1590 	for (i = 0; i < adapter->num_rx_queues; i++)
1591 		vmxnet3_rq_cleanup(&adapter->rx_queue[i], adapter);
1592 }
1593 
1594 
1595 static void vmxnet3_rq_destroy(struct vmxnet3_rx_queue *rq,
1596 			       struct vmxnet3_adapter *adapter)
1597 {
1598 	int i;
1599 	int j;
1600 
1601 	/* all rx buffers must have already been freed */
1602 	for (i = 0; i < 2; i++) {
1603 		if (rq->buf_info[i]) {
1604 			for (j = 0; j < rq->rx_ring[i].size; j++)
1605 				BUG_ON(rq->buf_info[i][j].page != NULL);
1606 		}
1607 	}
1608 
1609 
1610 	for (i = 0; i < 2; i++) {
1611 		if (rq->rx_ring[i].base) {
1612 			dma_free_coherent(&adapter->pdev->dev,
1613 					  rq->rx_ring[i].size
1614 					  * sizeof(struct Vmxnet3_RxDesc),
1615 					  rq->rx_ring[i].base,
1616 					  rq->rx_ring[i].basePA);
1617 			rq->rx_ring[i].base = NULL;
1618 		}
1619 	}
1620 
1621 	if (rq->data_ring.base) {
1622 		dma_free_coherent(&adapter->pdev->dev,
1623 				  rq->rx_ring[0].size * rq->data_ring.desc_size,
1624 				  rq->data_ring.base, rq->data_ring.basePA);
1625 		rq->data_ring.base = NULL;
1626 	}
1627 
1628 	if (rq->comp_ring.base) {
1629 		dma_free_coherent(&adapter->pdev->dev, rq->comp_ring.size
1630 				  * sizeof(struct Vmxnet3_RxCompDesc),
1631 				  rq->comp_ring.base, rq->comp_ring.basePA);
1632 		rq->comp_ring.base = NULL;
1633 	}
1634 
1635 	if (rq->buf_info[0]) {
1636 		size_t sz = sizeof(struct vmxnet3_rx_buf_info) *
1637 			(rq->rx_ring[0].size + rq->rx_ring[1].size);
1638 		dma_free_coherent(&adapter->pdev->dev, sz, rq->buf_info[0],
1639 				  rq->buf_info_pa);
1640 		rq->buf_info[0] = rq->buf_info[1] = NULL;
1641 	}
1642 }
1643 
1644 static void
1645 vmxnet3_rq_destroy_all_rxdataring(struct vmxnet3_adapter *adapter)
1646 {
1647 	int i;
1648 
1649 	for (i = 0; i < adapter->num_rx_queues; i++) {
1650 		struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
1651 
1652 		if (rq->data_ring.base) {
1653 			dma_free_coherent(&adapter->pdev->dev,
1654 					  (rq->rx_ring[0].size *
1655 					  rq->data_ring.desc_size),
1656 					  rq->data_ring.base,
1657 					  rq->data_ring.basePA);
1658 			rq->data_ring.base = NULL;
1659 			rq->data_ring.desc_size = 0;
1660 		}
1661 	}
1662 }
1663 
1664 static int
1665 vmxnet3_rq_init(struct vmxnet3_rx_queue *rq,
1666 		struct vmxnet3_adapter  *adapter)
1667 {
1668 	int i;
1669 
1670 	/* initialize buf_info */
1671 	for (i = 0; i < rq->rx_ring[0].size; i++) {
1672 
1673 		/* 1st buf for a pkt is skbuff */
1674 		if (i % adapter->rx_buf_per_pkt == 0) {
1675 			rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_SKB;
1676 			rq->buf_info[0][i].len = adapter->skb_buf_size;
1677 		} else { /* subsequent bufs for a pkt is frag */
1678 			rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_PAGE;
1679 			rq->buf_info[0][i].len = PAGE_SIZE;
1680 		}
1681 	}
1682 	for (i = 0; i < rq->rx_ring[1].size; i++) {
1683 		rq->buf_info[1][i].buf_type = VMXNET3_RX_BUF_PAGE;
1684 		rq->buf_info[1][i].len = PAGE_SIZE;
1685 	}
1686 
1687 	/* reset internal state and allocate buffers for both rings */
1688 	for (i = 0; i < 2; i++) {
1689 		rq->rx_ring[i].next2fill = rq->rx_ring[i].next2comp = 0;
1690 
1691 		memset(rq->rx_ring[i].base, 0, rq->rx_ring[i].size *
1692 		       sizeof(struct Vmxnet3_RxDesc));
1693 		rq->rx_ring[i].gen = VMXNET3_INIT_GEN;
1694 	}
1695 	if (vmxnet3_rq_alloc_rx_buf(rq, 0, rq->rx_ring[0].size - 1,
1696 				    adapter) == 0) {
1697 		/* at least has 1 rx buffer for the 1st ring */
1698 		return -ENOMEM;
1699 	}
1700 	vmxnet3_rq_alloc_rx_buf(rq, 1, rq->rx_ring[1].size - 1, adapter);
1701 
1702 	/* reset the comp ring */
1703 	rq->comp_ring.next2proc = 0;
1704 	memset(rq->comp_ring.base, 0, rq->comp_ring.size *
1705 	       sizeof(struct Vmxnet3_RxCompDesc));
1706 	rq->comp_ring.gen = VMXNET3_INIT_GEN;
1707 
1708 	/* reset rxctx */
1709 	rq->rx_ctx.skb = NULL;
1710 
1711 	/* stats are not reset */
1712 	return 0;
1713 }
1714 
1715 
1716 static int
1717 vmxnet3_rq_init_all(struct vmxnet3_adapter *adapter)
1718 {
1719 	int i, err = 0;
1720 
1721 	for (i = 0; i < adapter->num_rx_queues; i++) {
1722 		err = vmxnet3_rq_init(&adapter->rx_queue[i], adapter);
1723 		if (unlikely(err)) {
1724 			dev_err(&adapter->netdev->dev, "%s: failed to "
1725 				"initialize rx queue%i\n",
1726 				adapter->netdev->name, i);
1727 			break;
1728 		}
1729 	}
1730 	return err;
1731 
1732 }
1733 
1734 
1735 static int
1736 vmxnet3_rq_create(struct vmxnet3_rx_queue *rq, struct vmxnet3_adapter *adapter)
1737 {
1738 	int i;
1739 	size_t sz;
1740 	struct vmxnet3_rx_buf_info *bi;
1741 
1742 	for (i = 0; i < 2; i++) {
1743 
1744 		sz = rq->rx_ring[i].size * sizeof(struct Vmxnet3_RxDesc);
1745 		rq->rx_ring[i].base = dma_alloc_coherent(
1746 						&adapter->pdev->dev, sz,
1747 						&rq->rx_ring[i].basePA,
1748 						GFP_KERNEL);
1749 		if (!rq->rx_ring[i].base) {
1750 			netdev_err(adapter->netdev,
1751 				   "failed to allocate rx ring %d\n", i);
1752 			goto err;
1753 		}
1754 	}
1755 
1756 	if ((adapter->rxdataring_enabled) && (rq->data_ring.desc_size != 0)) {
1757 		sz = rq->rx_ring[0].size * rq->data_ring.desc_size;
1758 		rq->data_ring.base =
1759 			dma_alloc_coherent(&adapter->pdev->dev, sz,
1760 					   &rq->data_ring.basePA,
1761 					   GFP_KERNEL);
1762 		if (!rq->data_ring.base) {
1763 			netdev_err(adapter->netdev,
1764 				   "rx data ring will be disabled\n");
1765 			adapter->rxdataring_enabled = false;
1766 		}
1767 	} else {
1768 		rq->data_ring.base = NULL;
1769 		rq->data_ring.desc_size = 0;
1770 	}
1771 
1772 	sz = rq->comp_ring.size * sizeof(struct Vmxnet3_RxCompDesc);
1773 	rq->comp_ring.base = dma_alloc_coherent(&adapter->pdev->dev, sz,
1774 						&rq->comp_ring.basePA,
1775 						GFP_KERNEL);
1776 	if (!rq->comp_ring.base) {
1777 		netdev_err(adapter->netdev, "failed to allocate rx comp ring\n");
1778 		goto err;
1779 	}
1780 
1781 	sz = sizeof(struct vmxnet3_rx_buf_info) * (rq->rx_ring[0].size +
1782 						   rq->rx_ring[1].size);
1783 	bi = dma_zalloc_coherent(&adapter->pdev->dev, sz, &rq->buf_info_pa,
1784 				 GFP_KERNEL);
1785 	if (!bi)
1786 		goto err;
1787 
1788 	rq->buf_info[0] = bi;
1789 	rq->buf_info[1] = bi + rq->rx_ring[0].size;
1790 
1791 	return 0;
1792 
1793 err:
1794 	vmxnet3_rq_destroy(rq, adapter);
1795 	return -ENOMEM;
1796 }
1797 
1798 
1799 static int
1800 vmxnet3_rq_create_all(struct vmxnet3_adapter *adapter)
1801 {
1802 	int i, err = 0;
1803 
1804 	adapter->rxdataring_enabled = VMXNET3_VERSION_GE_3(adapter);
1805 
1806 	for (i = 0; i < adapter->num_rx_queues; i++) {
1807 		err = vmxnet3_rq_create(&adapter->rx_queue[i], adapter);
1808 		if (unlikely(err)) {
1809 			dev_err(&adapter->netdev->dev,
1810 				"%s: failed to create rx queue%i\n",
1811 				adapter->netdev->name, i);
1812 			goto err_out;
1813 		}
1814 	}
1815 
1816 	if (!adapter->rxdataring_enabled)
1817 		vmxnet3_rq_destroy_all_rxdataring(adapter);
1818 
1819 	return err;
1820 err_out:
1821 	vmxnet3_rq_destroy_all(adapter);
1822 	return err;
1823 
1824 }
1825 
1826 /* Multiple queue aware polling function for tx and rx */
1827 
1828 static int
1829 vmxnet3_do_poll(struct vmxnet3_adapter *adapter, int budget)
1830 {
1831 	int rcd_done = 0, i;
1832 	if (unlikely(adapter->shared->ecr))
1833 		vmxnet3_process_events(adapter);
1834 	for (i = 0; i < adapter->num_tx_queues; i++)
1835 		vmxnet3_tq_tx_complete(&adapter->tx_queue[i], adapter);
1836 
1837 	for (i = 0; i < adapter->num_rx_queues; i++)
1838 		rcd_done += vmxnet3_rq_rx_complete(&adapter->rx_queue[i],
1839 						   adapter, budget);
1840 	return rcd_done;
1841 }
1842 
1843 
1844 static int
1845 vmxnet3_poll(struct napi_struct *napi, int budget)
1846 {
1847 	struct vmxnet3_rx_queue *rx_queue = container_of(napi,
1848 					  struct vmxnet3_rx_queue, napi);
1849 	int rxd_done;
1850 
1851 	rxd_done = vmxnet3_do_poll(rx_queue->adapter, budget);
1852 
1853 	if (rxd_done < budget) {
1854 		napi_complete_done(napi, rxd_done);
1855 		vmxnet3_enable_all_intrs(rx_queue->adapter);
1856 	}
1857 	return rxd_done;
1858 }
1859 
1860 /*
1861  * NAPI polling function for MSI-X mode with multiple Rx queues
1862  * Returns the # of the NAPI credit consumed (# of rx descriptors processed)
1863  */
1864 
1865 static int
1866 vmxnet3_poll_rx_only(struct napi_struct *napi, int budget)
1867 {
1868 	struct vmxnet3_rx_queue *rq = container_of(napi,
1869 						struct vmxnet3_rx_queue, napi);
1870 	struct vmxnet3_adapter *adapter = rq->adapter;
1871 	int rxd_done;
1872 
1873 	/* When sharing interrupt with corresponding tx queue, process
1874 	 * tx completions in that queue as well
1875 	 */
1876 	if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE) {
1877 		struct vmxnet3_tx_queue *tq =
1878 				&adapter->tx_queue[rq - adapter->rx_queue];
1879 		vmxnet3_tq_tx_complete(tq, adapter);
1880 	}
1881 
1882 	rxd_done = vmxnet3_rq_rx_complete(rq, adapter, budget);
1883 
1884 	if (rxd_done < budget) {
1885 		napi_complete_done(napi, rxd_done);
1886 		vmxnet3_enable_intr(adapter, rq->comp_ring.intr_idx);
1887 	}
1888 	return rxd_done;
1889 }
1890 
1891 
1892 #ifdef CONFIG_PCI_MSI
1893 
1894 /*
1895  * Handle completion interrupts on tx queues
1896  * Returns whether or not the intr is handled
1897  */
1898 
1899 static irqreturn_t
1900 vmxnet3_msix_tx(int irq, void *data)
1901 {
1902 	struct vmxnet3_tx_queue *tq = data;
1903 	struct vmxnet3_adapter *adapter = tq->adapter;
1904 
1905 	if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1906 		vmxnet3_disable_intr(adapter, tq->comp_ring.intr_idx);
1907 
1908 	/* Handle the case where only one irq is allocate for all tx queues */
1909 	if (adapter->share_intr == VMXNET3_INTR_TXSHARE) {
1910 		int i;
1911 		for (i = 0; i < adapter->num_tx_queues; i++) {
1912 			struct vmxnet3_tx_queue *txq = &adapter->tx_queue[i];
1913 			vmxnet3_tq_tx_complete(txq, adapter);
1914 		}
1915 	} else {
1916 		vmxnet3_tq_tx_complete(tq, adapter);
1917 	}
1918 	vmxnet3_enable_intr(adapter, tq->comp_ring.intr_idx);
1919 
1920 	return IRQ_HANDLED;
1921 }
1922 
1923 
1924 /*
1925  * Handle completion interrupts on rx queues. Returns whether or not the
1926  * intr is handled
1927  */
1928 
1929 static irqreturn_t
1930 vmxnet3_msix_rx(int irq, void *data)
1931 {
1932 	struct vmxnet3_rx_queue *rq = data;
1933 	struct vmxnet3_adapter *adapter = rq->adapter;
1934 
1935 	/* disable intr if needed */
1936 	if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1937 		vmxnet3_disable_intr(adapter, rq->comp_ring.intr_idx);
1938 	napi_schedule(&rq->napi);
1939 
1940 	return IRQ_HANDLED;
1941 }
1942 
1943 /*
1944  *----------------------------------------------------------------------------
1945  *
1946  * vmxnet3_msix_event --
1947  *
1948  *    vmxnet3 msix event intr handler
1949  *
1950  * Result:
1951  *    whether or not the intr is handled
1952  *
1953  *----------------------------------------------------------------------------
1954  */
1955 
1956 static irqreturn_t
1957 vmxnet3_msix_event(int irq, void *data)
1958 {
1959 	struct net_device *dev = data;
1960 	struct vmxnet3_adapter *adapter = netdev_priv(dev);
1961 
1962 	/* disable intr if needed */
1963 	if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1964 		vmxnet3_disable_intr(adapter, adapter->intr.event_intr_idx);
1965 
1966 	if (adapter->shared->ecr)
1967 		vmxnet3_process_events(adapter);
1968 
1969 	vmxnet3_enable_intr(adapter, adapter->intr.event_intr_idx);
1970 
1971 	return IRQ_HANDLED;
1972 }
1973 
1974 #endif /* CONFIG_PCI_MSI  */
1975 
1976 
1977 /* Interrupt handler for vmxnet3  */
1978 static irqreturn_t
1979 vmxnet3_intr(int irq, void *dev_id)
1980 {
1981 	struct net_device *dev = dev_id;
1982 	struct vmxnet3_adapter *adapter = netdev_priv(dev);
1983 
1984 	if (adapter->intr.type == VMXNET3_IT_INTX) {
1985 		u32 icr = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_ICR);
1986 		if (unlikely(icr == 0))
1987 			/* not ours */
1988 			return IRQ_NONE;
1989 	}
1990 
1991 
1992 	/* disable intr if needed */
1993 	if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1994 		vmxnet3_disable_all_intrs(adapter);
1995 
1996 	napi_schedule(&adapter->rx_queue[0].napi);
1997 
1998 	return IRQ_HANDLED;
1999 }
2000 
2001 #ifdef CONFIG_NET_POLL_CONTROLLER
2002 
2003 /* netpoll callback. */
2004 static void
2005 vmxnet3_netpoll(struct net_device *netdev)
2006 {
2007 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2008 
2009 	switch (adapter->intr.type) {
2010 #ifdef CONFIG_PCI_MSI
2011 	case VMXNET3_IT_MSIX: {
2012 		int i;
2013 		for (i = 0; i < adapter->num_rx_queues; i++)
2014 			vmxnet3_msix_rx(0, &adapter->rx_queue[i]);
2015 		break;
2016 	}
2017 #endif
2018 	case VMXNET3_IT_MSI:
2019 	default:
2020 		vmxnet3_intr(0, adapter->netdev);
2021 		break;
2022 	}
2023 
2024 }
2025 #endif	/* CONFIG_NET_POLL_CONTROLLER */
2026 
2027 static int
2028 vmxnet3_request_irqs(struct vmxnet3_adapter *adapter)
2029 {
2030 	struct vmxnet3_intr *intr = &adapter->intr;
2031 	int err = 0, i;
2032 	int vector = 0;
2033 
2034 #ifdef CONFIG_PCI_MSI
2035 	if (adapter->intr.type == VMXNET3_IT_MSIX) {
2036 		for (i = 0; i < adapter->num_tx_queues; i++) {
2037 			if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE) {
2038 				sprintf(adapter->tx_queue[i].name, "%s-tx-%d",
2039 					adapter->netdev->name, vector);
2040 				err = request_irq(
2041 					      intr->msix_entries[vector].vector,
2042 					      vmxnet3_msix_tx, 0,
2043 					      adapter->tx_queue[i].name,
2044 					      &adapter->tx_queue[i]);
2045 			} else {
2046 				sprintf(adapter->tx_queue[i].name, "%s-rxtx-%d",
2047 					adapter->netdev->name, vector);
2048 			}
2049 			if (err) {
2050 				dev_err(&adapter->netdev->dev,
2051 					"Failed to request irq for MSIX, %s, "
2052 					"error %d\n",
2053 					adapter->tx_queue[i].name, err);
2054 				return err;
2055 			}
2056 
2057 			/* Handle the case where only 1 MSIx was allocated for
2058 			 * all tx queues */
2059 			if (adapter->share_intr == VMXNET3_INTR_TXSHARE) {
2060 				for (; i < adapter->num_tx_queues; i++)
2061 					adapter->tx_queue[i].comp_ring.intr_idx
2062 								= vector;
2063 				vector++;
2064 				break;
2065 			} else {
2066 				adapter->tx_queue[i].comp_ring.intr_idx
2067 								= vector++;
2068 			}
2069 		}
2070 		if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE)
2071 			vector = 0;
2072 
2073 		for (i = 0; i < adapter->num_rx_queues; i++) {
2074 			if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE)
2075 				sprintf(adapter->rx_queue[i].name, "%s-rx-%d",
2076 					adapter->netdev->name, vector);
2077 			else
2078 				sprintf(adapter->rx_queue[i].name, "%s-rxtx-%d",
2079 					adapter->netdev->name, vector);
2080 			err = request_irq(intr->msix_entries[vector].vector,
2081 					  vmxnet3_msix_rx, 0,
2082 					  adapter->rx_queue[i].name,
2083 					  &(adapter->rx_queue[i]));
2084 			if (err) {
2085 				netdev_err(adapter->netdev,
2086 					   "Failed to request irq for MSIX, "
2087 					   "%s, error %d\n",
2088 					   adapter->rx_queue[i].name, err);
2089 				return err;
2090 			}
2091 
2092 			adapter->rx_queue[i].comp_ring.intr_idx = vector++;
2093 		}
2094 
2095 		sprintf(intr->event_msi_vector_name, "%s-event-%d",
2096 			adapter->netdev->name, vector);
2097 		err = request_irq(intr->msix_entries[vector].vector,
2098 				  vmxnet3_msix_event, 0,
2099 				  intr->event_msi_vector_name, adapter->netdev);
2100 		intr->event_intr_idx = vector;
2101 
2102 	} else if (intr->type == VMXNET3_IT_MSI) {
2103 		adapter->num_rx_queues = 1;
2104 		err = request_irq(adapter->pdev->irq, vmxnet3_intr, 0,
2105 				  adapter->netdev->name, adapter->netdev);
2106 	} else {
2107 #endif
2108 		adapter->num_rx_queues = 1;
2109 		err = request_irq(adapter->pdev->irq, vmxnet3_intr,
2110 				  IRQF_SHARED, adapter->netdev->name,
2111 				  adapter->netdev);
2112 #ifdef CONFIG_PCI_MSI
2113 	}
2114 #endif
2115 	intr->num_intrs = vector + 1;
2116 	if (err) {
2117 		netdev_err(adapter->netdev,
2118 			   "Failed to request irq (intr type:%d), error %d\n",
2119 			   intr->type, err);
2120 	} else {
2121 		/* Number of rx queues will not change after this */
2122 		for (i = 0; i < adapter->num_rx_queues; i++) {
2123 			struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
2124 			rq->qid = i;
2125 			rq->qid2 = i + adapter->num_rx_queues;
2126 			rq->dataRingQid = i + 2 * adapter->num_rx_queues;
2127 		}
2128 
2129 		/* init our intr settings */
2130 		for (i = 0; i < intr->num_intrs; i++)
2131 			intr->mod_levels[i] = UPT1_IML_ADAPTIVE;
2132 		if (adapter->intr.type != VMXNET3_IT_MSIX) {
2133 			adapter->intr.event_intr_idx = 0;
2134 			for (i = 0; i < adapter->num_tx_queues; i++)
2135 				adapter->tx_queue[i].comp_ring.intr_idx = 0;
2136 			adapter->rx_queue[0].comp_ring.intr_idx = 0;
2137 		}
2138 
2139 		netdev_info(adapter->netdev,
2140 			    "intr type %u, mode %u, %u vectors allocated\n",
2141 			    intr->type, intr->mask_mode, intr->num_intrs);
2142 	}
2143 
2144 	return err;
2145 }
2146 
2147 
2148 static void
2149 vmxnet3_free_irqs(struct vmxnet3_adapter *adapter)
2150 {
2151 	struct vmxnet3_intr *intr = &adapter->intr;
2152 	BUG_ON(intr->type == VMXNET3_IT_AUTO || intr->num_intrs <= 0);
2153 
2154 	switch (intr->type) {
2155 #ifdef CONFIG_PCI_MSI
2156 	case VMXNET3_IT_MSIX:
2157 	{
2158 		int i, vector = 0;
2159 
2160 		if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE) {
2161 			for (i = 0; i < adapter->num_tx_queues; i++) {
2162 				free_irq(intr->msix_entries[vector++].vector,
2163 					 &(adapter->tx_queue[i]));
2164 				if (adapter->share_intr == VMXNET3_INTR_TXSHARE)
2165 					break;
2166 			}
2167 		}
2168 
2169 		for (i = 0; i < adapter->num_rx_queues; i++) {
2170 			free_irq(intr->msix_entries[vector++].vector,
2171 				 &(adapter->rx_queue[i]));
2172 		}
2173 
2174 		free_irq(intr->msix_entries[vector].vector,
2175 			 adapter->netdev);
2176 		BUG_ON(vector >= intr->num_intrs);
2177 		break;
2178 	}
2179 #endif
2180 	case VMXNET3_IT_MSI:
2181 		free_irq(adapter->pdev->irq, adapter->netdev);
2182 		break;
2183 	case VMXNET3_IT_INTX:
2184 		free_irq(adapter->pdev->irq, adapter->netdev);
2185 		break;
2186 	default:
2187 		BUG();
2188 	}
2189 }
2190 
2191 
2192 static void
2193 vmxnet3_restore_vlan(struct vmxnet3_adapter *adapter)
2194 {
2195 	u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
2196 	u16 vid;
2197 
2198 	/* allow untagged pkts */
2199 	VMXNET3_SET_VFTABLE_ENTRY(vfTable, 0);
2200 
2201 	for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
2202 		VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid);
2203 }
2204 
2205 
2206 static int
2207 vmxnet3_vlan_rx_add_vid(struct net_device *netdev, __be16 proto, u16 vid)
2208 {
2209 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2210 
2211 	if (!(netdev->flags & IFF_PROMISC)) {
2212 		u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
2213 		unsigned long flags;
2214 
2215 		VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid);
2216 		spin_lock_irqsave(&adapter->cmd_lock, flags);
2217 		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2218 				       VMXNET3_CMD_UPDATE_VLAN_FILTERS);
2219 		spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2220 	}
2221 
2222 	set_bit(vid, adapter->active_vlans);
2223 
2224 	return 0;
2225 }
2226 
2227 
2228 static int
2229 vmxnet3_vlan_rx_kill_vid(struct net_device *netdev, __be16 proto, u16 vid)
2230 {
2231 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2232 
2233 	if (!(netdev->flags & IFF_PROMISC)) {
2234 		u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
2235 		unsigned long flags;
2236 
2237 		VMXNET3_CLEAR_VFTABLE_ENTRY(vfTable, vid);
2238 		spin_lock_irqsave(&adapter->cmd_lock, flags);
2239 		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2240 				       VMXNET3_CMD_UPDATE_VLAN_FILTERS);
2241 		spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2242 	}
2243 
2244 	clear_bit(vid, adapter->active_vlans);
2245 
2246 	return 0;
2247 }
2248 
2249 
2250 static u8 *
2251 vmxnet3_copy_mc(struct net_device *netdev)
2252 {
2253 	u8 *buf = NULL;
2254 	u32 sz = netdev_mc_count(netdev) * ETH_ALEN;
2255 
2256 	/* struct Vmxnet3_RxFilterConf.mfTableLen is u16. */
2257 	if (sz <= 0xffff) {
2258 		/* We may be called with BH disabled */
2259 		buf = kmalloc(sz, GFP_ATOMIC);
2260 		if (buf) {
2261 			struct netdev_hw_addr *ha;
2262 			int i = 0;
2263 
2264 			netdev_for_each_mc_addr(ha, netdev)
2265 				memcpy(buf + i++ * ETH_ALEN, ha->addr,
2266 				       ETH_ALEN);
2267 		}
2268 	}
2269 	return buf;
2270 }
2271 
2272 
2273 static void
2274 vmxnet3_set_mc(struct net_device *netdev)
2275 {
2276 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2277 	unsigned long flags;
2278 	struct Vmxnet3_RxFilterConf *rxConf =
2279 					&adapter->shared->devRead.rxFilterConf;
2280 	u8 *new_table = NULL;
2281 	dma_addr_t new_table_pa = 0;
2282 	bool new_table_pa_valid = false;
2283 	u32 new_mode = VMXNET3_RXM_UCAST;
2284 
2285 	if (netdev->flags & IFF_PROMISC) {
2286 		u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
2287 		memset(vfTable, 0, VMXNET3_VFT_SIZE * sizeof(*vfTable));
2288 
2289 		new_mode |= VMXNET3_RXM_PROMISC;
2290 	} else {
2291 		vmxnet3_restore_vlan(adapter);
2292 	}
2293 
2294 	if (netdev->flags & IFF_BROADCAST)
2295 		new_mode |= VMXNET3_RXM_BCAST;
2296 
2297 	if (netdev->flags & IFF_ALLMULTI)
2298 		new_mode |= VMXNET3_RXM_ALL_MULTI;
2299 	else
2300 		if (!netdev_mc_empty(netdev)) {
2301 			new_table = vmxnet3_copy_mc(netdev);
2302 			if (new_table) {
2303 				size_t sz = netdev_mc_count(netdev) * ETH_ALEN;
2304 
2305 				rxConf->mfTableLen = cpu_to_le16(sz);
2306 				new_table_pa = dma_map_single(
2307 							&adapter->pdev->dev,
2308 							new_table,
2309 							sz,
2310 							PCI_DMA_TODEVICE);
2311 				if (!dma_mapping_error(&adapter->pdev->dev,
2312 						       new_table_pa)) {
2313 					new_mode |= VMXNET3_RXM_MCAST;
2314 					new_table_pa_valid = true;
2315 					rxConf->mfTablePA = cpu_to_le64(
2316 								new_table_pa);
2317 				}
2318 			}
2319 			if (!new_table_pa_valid) {
2320 				netdev_info(netdev,
2321 					    "failed to copy mcast list, setting ALL_MULTI\n");
2322 				new_mode |= VMXNET3_RXM_ALL_MULTI;
2323 			}
2324 		}
2325 
2326 	if (!(new_mode & VMXNET3_RXM_MCAST)) {
2327 		rxConf->mfTableLen = 0;
2328 		rxConf->mfTablePA = 0;
2329 	}
2330 
2331 	spin_lock_irqsave(&adapter->cmd_lock, flags);
2332 	if (new_mode != rxConf->rxMode) {
2333 		rxConf->rxMode = cpu_to_le32(new_mode);
2334 		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2335 				       VMXNET3_CMD_UPDATE_RX_MODE);
2336 		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2337 				       VMXNET3_CMD_UPDATE_VLAN_FILTERS);
2338 	}
2339 
2340 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2341 			       VMXNET3_CMD_UPDATE_MAC_FILTERS);
2342 	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2343 
2344 	if (new_table_pa_valid)
2345 		dma_unmap_single(&adapter->pdev->dev, new_table_pa,
2346 				 rxConf->mfTableLen, PCI_DMA_TODEVICE);
2347 	kfree(new_table);
2348 }
2349 
2350 void
2351 vmxnet3_rq_destroy_all(struct vmxnet3_adapter *adapter)
2352 {
2353 	int i;
2354 
2355 	for (i = 0; i < adapter->num_rx_queues; i++)
2356 		vmxnet3_rq_destroy(&adapter->rx_queue[i], adapter);
2357 }
2358 
2359 
2360 /*
2361  *   Set up driver_shared based on settings in adapter.
2362  */
2363 
2364 static void
2365 vmxnet3_setup_driver_shared(struct vmxnet3_adapter *adapter)
2366 {
2367 	struct Vmxnet3_DriverShared *shared = adapter->shared;
2368 	struct Vmxnet3_DSDevRead *devRead = &shared->devRead;
2369 	struct Vmxnet3_TxQueueConf *tqc;
2370 	struct Vmxnet3_RxQueueConf *rqc;
2371 	int i;
2372 
2373 	memset(shared, 0, sizeof(*shared));
2374 
2375 	/* driver settings */
2376 	shared->magic = cpu_to_le32(VMXNET3_REV1_MAGIC);
2377 	devRead->misc.driverInfo.version = cpu_to_le32(
2378 						VMXNET3_DRIVER_VERSION_NUM);
2379 	devRead->misc.driverInfo.gos.gosBits = (sizeof(void *) == 4 ?
2380 				VMXNET3_GOS_BITS_32 : VMXNET3_GOS_BITS_64);
2381 	devRead->misc.driverInfo.gos.gosType = VMXNET3_GOS_TYPE_LINUX;
2382 	*((u32 *)&devRead->misc.driverInfo.gos) = cpu_to_le32(
2383 				*((u32 *)&devRead->misc.driverInfo.gos));
2384 	devRead->misc.driverInfo.vmxnet3RevSpt = cpu_to_le32(1);
2385 	devRead->misc.driverInfo.uptVerSpt = cpu_to_le32(1);
2386 
2387 	devRead->misc.ddPA = cpu_to_le64(adapter->adapter_pa);
2388 	devRead->misc.ddLen = cpu_to_le32(sizeof(struct vmxnet3_adapter));
2389 
2390 	/* set up feature flags */
2391 	if (adapter->netdev->features & NETIF_F_RXCSUM)
2392 		devRead->misc.uptFeatures |= UPT1_F_RXCSUM;
2393 
2394 	if (adapter->netdev->features & NETIF_F_LRO) {
2395 		devRead->misc.uptFeatures |= UPT1_F_LRO;
2396 		devRead->misc.maxNumRxSG = cpu_to_le16(1 + MAX_SKB_FRAGS);
2397 	}
2398 	if (adapter->netdev->features & NETIF_F_HW_VLAN_CTAG_RX)
2399 		devRead->misc.uptFeatures |= UPT1_F_RXVLAN;
2400 
2401 	devRead->misc.mtu = cpu_to_le32(adapter->netdev->mtu);
2402 	devRead->misc.queueDescPA = cpu_to_le64(adapter->queue_desc_pa);
2403 	devRead->misc.queueDescLen = cpu_to_le32(
2404 		adapter->num_tx_queues * sizeof(struct Vmxnet3_TxQueueDesc) +
2405 		adapter->num_rx_queues * sizeof(struct Vmxnet3_RxQueueDesc));
2406 
2407 	/* tx queue settings */
2408 	devRead->misc.numTxQueues =  adapter->num_tx_queues;
2409 	for (i = 0; i < adapter->num_tx_queues; i++) {
2410 		struct vmxnet3_tx_queue	*tq = &adapter->tx_queue[i];
2411 		BUG_ON(adapter->tx_queue[i].tx_ring.base == NULL);
2412 		tqc = &adapter->tqd_start[i].conf;
2413 		tqc->txRingBasePA   = cpu_to_le64(tq->tx_ring.basePA);
2414 		tqc->dataRingBasePA = cpu_to_le64(tq->data_ring.basePA);
2415 		tqc->compRingBasePA = cpu_to_le64(tq->comp_ring.basePA);
2416 		tqc->ddPA           = cpu_to_le64(tq->buf_info_pa);
2417 		tqc->txRingSize     = cpu_to_le32(tq->tx_ring.size);
2418 		tqc->dataRingSize   = cpu_to_le32(tq->data_ring.size);
2419 		tqc->txDataRingDescSize = cpu_to_le32(tq->txdata_desc_size);
2420 		tqc->compRingSize   = cpu_to_le32(tq->comp_ring.size);
2421 		tqc->ddLen          = cpu_to_le32(
2422 					sizeof(struct vmxnet3_tx_buf_info) *
2423 					tqc->txRingSize);
2424 		tqc->intrIdx        = tq->comp_ring.intr_idx;
2425 	}
2426 
2427 	/* rx queue settings */
2428 	devRead->misc.numRxQueues = adapter->num_rx_queues;
2429 	for (i = 0; i < adapter->num_rx_queues; i++) {
2430 		struct vmxnet3_rx_queue	*rq = &adapter->rx_queue[i];
2431 		rqc = &adapter->rqd_start[i].conf;
2432 		rqc->rxRingBasePA[0] = cpu_to_le64(rq->rx_ring[0].basePA);
2433 		rqc->rxRingBasePA[1] = cpu_to_le64(rq->rx_ring[1].basePA);
2434 		rqc->compRingBasePA  = cpu_to_le64(rq->comp_ring.basePA);
2435 		rqc->ddPA            = cpu_to_le64(rq->buf_info_pa);
2436 		rqc->rxRingSize[0]   = cpu_to_le32(rq->rx_ring[0].size);
2437 		rqc->rxRingSize[1]   = cpu_to_le32(rq->rx_ring[1].size);
2438 		rqc->compRingSize    = cpu_to_le32(rq->comp_ring.size);
2439 		rqc->ddLen           = cpu_to_le32(
2440 					sizeof(struct vmxnet3_rx_buf_info) *
2441 					(rqc->rxRingSize[0] +
2442 					 rqc->rxRingSize[1]));
2443 		rqc->intrIdx         = rq->comp_ring.intr_idx;
2444 		if (VMXNET3_VERSION_GE_3(adapter)) {
2445 			rqc->rxDataRingBasePA =
2446 				cpu_to_le64(rq->data_ring.basePA);
2447 			rqc->rxDataRingDescSize =
2448 				cpu_to_le16(rq->data_ring.desc_size);
2449 		}
2450 	}
2451 
2452 #ifdef VMXNET3_RSS
2453 	memset(adapter->rss_conf, 0, sizeof(*adapter->rss_conf));
2454 
2455 	if (adapter->rss) {
2456 		struct UPT1_RSSConf *rssConf = adapter->rss_conf;
2457 
2458 		devRead->misc.uptFeatures |= UPT1_F_RSS;
2459 		devRead->misc.numRxQueues = adapter->num_rx_queues;
2460 		rssConf->hashType = UPT1_RSS_HASH_TYPE_TCP_IPV4 |
2461 				    UPT1_RSS_HASH_TYPE_IPV4 |
2462 				    UPT1_RSS_HASH_TYPE_TCP_IPV6 |
2463 				    UPT1_RSS_HASH_TYPE_IPV6;
2464 		rssConf->hashFunc = UPT1_RSS_HASH_FUNC_TOEPLITZ;
2465 		rssConf->hashKeySize = UPT1_RSS_MAX_KEY_SIZE;
2466 		rssConf->indTableSize = VMXNET3_RSS_IND_TABLE_SIZE;
2467 		netdev_rss_key_fill(rssConf->hashKey, sizeof(rssConf->hashKey));
2468 
2469 		for (i = 0; i < rssConf->indTableSize; i++)
2470 			rssConf->indTable[i] = ethtool_rxfh_indir_default(
2471 				i, adapter->num_rx_queues);
2472 
2473 		devRead->rssConfDesc.confVer = 1;
2474 		devRead->rssConfDesc.confLen = cpu_to_le32(sizeof(*rssConf));
2475 		devRead->rssConfDesc.confPA =
2476 			cpu_to_le64(adapter->rss_conf_pa);
2477 	}
2478 
2479 #endif /* VMXNET3_RSS */
2480 
2481 	/* intr settings */
2482 	devRead->intrConf.autoMask = adapter->intr.mask_mode ==
2483 				     VMXNET3_IMM_AUTO;
2484 	devRead->intrConf.numIntrs = adapter->intr.num_intrs;
2485 	for (i = 0; i < adapter->intr.num_intrs; i++)
2486 		devRead->intrConf.modLevels[i] = adapter->intr.mod_levels[i];
2487 
2488 	devRead->intrConf.eventIntrIdx = adapter->intr.event_intr_idx;
2489 	devRead->intrConf.intrCtrl |= cpu_to_le32(VMXNET3_IC_DISABLE_ALL);
2490 
2491 	/* rx filter settings */
2492 	devRead->rxFilterConf.rxMode = 0;
2493 	vmxnet3_restore_vlan(adapter);
2494 	vmxnet3_write_mac_addr(adapter, adapter->netdev->dev_addr);
2495 
2496 	/* the rest are already zeroed */
2497 }
2498 
2499 static void
2500 vmxnet3_init_coalesce(struct vmxnet3_adapter *adapter)
2501 {
2502 	struct Vmxnet3_DriverShared *shared = adapter->shared;
2503 	union Vmxnet3_CmdInfo *cmdInfo = &shared->cu.cmdInfo;
2504 	unsigned long flags;
2505 
2506 	if (!VMXNET3_VERSION_GE_3(adapter))
2507 		return;
2508 
2509 	spin_lock_irqsave(&adapter->cmd_lock, flags);
2510 	cmdInfo->varConf.confVer = 1;
2511 	cmdInfo->varConf.confLen =
2512 		cpu_to_le32(sizeof(*adapter->coal_conf));
2513 	cmdInfo->varConf.confPA  = cpu_to_le64(adapter->coal_conf_pa);
2514 
2515 	if (adapter->default_coal_mode) {
2516 		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2517 				       VMXNET3_CMD_GET_COALESCE);
2518 	} else {
2519 		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2520 				       VMXNET3_CMD_SET_COALESCE);
2521 	}
2522 
2523 	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2524 }
2525 
2526 int
2527 vmxnet3_activate_dev(struct vmxnet3_adapter *adapter)
2528 {
2529 	int err, i;
2530 	u32 ret;
2531 	unsigned long flags;
2532 
2533 	netdev_dbg(adapter->netdev, "%s: skb_buf_size %d, rx_buf_per_pkt %d,"
2534 		" ring sizes %u %u %u\n", adapter->netdev->name,
2535 		adapter->skb_buf_size, adapter->rx_buf_per_pkt,
2536 		adapter->tx_queue[0].tx_ring.size,
2537 		adapter->rx_queue[0].rx_ring[0].size,
2538 		adapter->rx_queue[0].rx_ring[1].size);
2539 
2540 	vmxnet3_tq_init_all(adapter);
2541 	err = vmxnet3_rq_init_all(adapter);
2542 	if (err) {
2543 		netdev_err(adapter->netdev,
2544 			   "Failed to init rx queue error %d\n", err);
2545 		goto rq_err;
2546 	}
2547 
2548 	err = vmxnet3_request_irqs(adapter);
2549 	if (err) {
2550 		netdev_err(adapter->netdev,
2551 			   "Failed to setup irq for error %d\n", err);
2552 		goto irq_err;
2553 	}
2554 
2555 	vmxnet3_setup_driver_shared(adapter);
2556 
2557 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, VMXNET3_GET_ADDR_LO(
2558 			       adapter->shared_pa));
2559 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, VMXNET3_GET_ADDR_HI(
2560 			       adapter->shared_pa));
2561 	spin_lock_irqsave(&adapter->cmd_lock, flags);
2562 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2563 			       VMXNET3_CMD_ACTIVATE_DEV);
2564 	ret = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
2565 	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2566 
2567 	if (ret != 0) {
2568 		netdev_err(adapter->netdev,
2569 			   "Failed to activate dev: error %u\n", ret);
2570 		err = -EINVAL;
2571 		goto activate_err;
2572 	}
2573 
2574 	vmxnet3_init_coalesce(adapter);
2575 
2576 	for (i = 0; i < adapter->num_rx_queues; i++) {
2577 		VMXNET3_WRITE_BAR0_REG(adapter,
2578 				VMXNET3_REG_RXPROD + i * VMXNET3_REG_ALIGN,
2579 				adapter->rx_queue[i].rx_ring[0].next2fill);
2580 		VMXNET3_WRITE_BAR0_REG(adapter, (VMXNET3_REG_RXPROD2 +
2581 				(i * VMXNET3_REG_ALIGN)),
2582 				adapter->rx_queue[i].rx_ring[1].next2fill);
2583 	}
2584 
2585 	/* Apply the rx filter settins last. */
2586 	vmxnet3_set_mc(adapter->netdev);
2587 
2588 	/*
2589 	 * Check link state when first activating device. It will start the
2590 	 * tx queue if the link is up.
2591 	 */
2592 	vmxnet3_check_link(adapter, true);
2593 	for (i = 0; i < adapter->num_rx_queues; i++)
2594 		napi_enable(&adapter->rx_queue[i].napi);
2595 	vmxnet3_enable_all_intrs(adapter);
2596 	clear_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
2597 	return 0;
2598 
2599 activate_err:
2600 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, 0);
2601 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, 0);
2602 	vmxnet3_free_irqs(adapter);
2603 irq_err:
2604 rq_err:
2605 	/* free up buffers we allocated */
2606 	vmxnet3_rq_cleanup_all(adapter);
2607 	return err;
2608 }
2609 
2610 
2611 void
2612 vmxnet3_reset_dev(struct vmxnet3_adapter *adapter)
2613 {
2614 	unsigned long flags;
2615 	spin_lock_irqsave(&adapter->cmd_lock, flags);
2616 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, VMXNET3_CMD_RESET_DEV);
2617 	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2618 }
2619 
2620 
2621 int
2622 vmxnet3_quiesce_dev(struct vmxnet3_adapter *adapter)
2623 {
2624 	int i;
2625 	unsigned long flags;
2626 	if (test_and_set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state))
2627 		return 0;
2628 
2629 
2630 	spin_lock_irqsave(&adapter->cmd_lock, flags);
2631 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2632 			       VMXNET3_CMD_QUIESCE_DEV);
2633 	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2634 	vmxnet3_disable_all_intrs(adapter);
2635 
2636 	for (i = 0; i < adapter->num_rx_queues; i++)
2637 		napi_disable(&adapter->rx_queue[i].napi);
2638 	netif_tx_disable(adapter->netdev);
2639 	adapter->link_speed = 0;
2640 	netif_carrier_off(adapter->netdev);
2641 
2642 	vmxnet3_tq_cleanup_all(adapter);
2643 	vmxnet3_rq_cleanup_all(adapter);
2644 	vmxnet3_free_irqs(adapter);
2645 	return 0;
2646 }
2647 
2648 
2649 static void
2650 vmxnet3_write_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac)
2651 {
2652 	u32 tmp;
2653 
2654 	tmp = *(u32 *)mac;
2655 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACL, tmp);
2656 
2657 	tmp = (mac[5] << 8) | mac[4];
2658 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACH, tmp);
2659 }
2660 
2661 
2662 static int
2663 vmxnet3_set_mac_addr(struct net_device *netdev, void *p)
2664 {
2665 	struct sockaddr *addr = p;
2666 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2667 
2668 	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2669 	vmxnet3_write_mac_addr(adapter, addr->sa_data);
2670 
2671 	return 0;
2672 }
2673 
2674 
2675 /* ==================== initialization and cleanup routines ============ */
2676 
2677 static int
2678 vmxnet3_alloc_pci_resources(struct vmxnet3_adapter *adapter, bool *dma64)
2679 {
2680 	int err;
2681 	unsigned long mmio_start, mmio_len;
2682 	struct pci_dev *pdev = adapter->pdev;
2683 
2684 	err = pci_enable_device(pdev);
2685 	if (err) {
2686 		dev_err(&pdev->dev, "Failed to enable adapter: error %d\n", err);
2687 		return err;
2688 	}
2689 
2690 	if (pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) == 0) {
2691 		if (pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64)) != 0) {
2692 			dev_err(&pdev->dev,
2693 				"pci_set_consistent_dma_mask failed\n");
2694 			err = -EIO;
2695 			goto err_set_mask;
2696 		}
2697 		*dma64 = true;
2698 	} else {
2699 		if (pci_set_dma_mask(pdev, DMA_BIT_MASK(32)) != 0) {
2700 			dev_err(&pdev->dev,
2701 				"pci_set_dma_mask failed\n");
2702 			err = -EIO;
2703 			goto err_set_mask;
2704 		}
2705 		*dma64 = false;
2706 	}
2707 
2708 	err = pci_request_selected_regions(pdev, (1 << 2) - 1,
2709 					   vmxnet3_driver_name);
2710 	if (err) {
2711 		dev_err(&pdev->dev,
2712 			"Failed to request region for adapter: error %d\n", err);
2713 		goto err_set_mask;
2714 	}
2715 
2716 	pci_set_master(pdev);
2717 
2718 	mmio_start = pci_resource_start(pdev, 0);
2719 	mmio_len = pci_resource_len(pdev, 0);
2720 	adapter->hw_addr0 = ioremap(mmio_start, mmio_len);
2721 	if (!adapter->hw_addr0) {
2722 		dev_err(&pdev->dev, "Failed to map bar0\n");
2723 		err = -EIO;
2724 		goto err_ioremap;
2725 	}
2726 
2727 	mmio_start = pci_resource_start(pdev, 1);
2728 	mmio_len = pci_resource_len(pdev, 1);
2729 	adapter->hw_addr1 = ioremap(mmio_start, mmio_len);
2730 	if (!adapter->hw_addr1) {
2731 		dev_err(&pdev->dev, "Failed to map bar1\n");
2732 		err = -EIO;
2733 		goto err_bar1;
2734 	}
2735 	return 0;
2736 
2737 err_bar1:
2738 	iounmap(adapter->hw_addr0);
2739 err_ioremap:
2740 	pci_release_selected_regions(pdev, (1 << 2) - 1);
2741 err_set_mask:
2742 	pci_disable_device(pdev);
2743 	return err;
2744 }
2745 
2746 
2747 static void
2748 vmxnet3_free_pci_resources(struct vmxnet3_adapter *adapter)
2749 {
2750 	BUG_ON(!adapter->pdev);
2751 
2752 	iounmap(adapter->hw_addr0);
2753 	iounmap(adapter->hw_addr1);
2754 	pci_release_selected_regions(adapter->pdev, (1 << 2) - 1);
2755 	pci_disable_device(adapter->pdev);
2756 }
2757 
2758 
2759 static void
2760 vmxnet3_adjust_rx_ring_size(struct vmxnet3_adapter *adapter)
2761 {
2762 	size_t sz, i, ring0_size, ring1_size, comp_size;
2763 	if (adapter->netdev->mtu <= VMXNET3_MAX_SKB_BUF_SIZE -
2764 				    VMXNET3_MAX_ETH_HDR_SIZE) {
2765 		adapter->skb_buf_size = adapter->netdev->mtu +
2766 					VMXNET3_MAX_ETH_HDR_SIZE;
2767 		if (adapter->skb_buf_size < VMXNET3_MIN_T0_BUF_SIZE)
2768 			adapter->skb_buf_size = VMXNET3_MIN_T0_BUF_SIZE;
2769 
2770 		adapter->rx_buf_per_pkt = 1;
2771 	} else {
2772 		adapter->skb_buf_size = VMXNET3_MAX_SKB_BUF_SIZE;
2773 		sz = adapter->netdev->mtu - VMXNET3_MAX_SKB_BUF_SIZE +
2774 					    VMXNET3_MAX_ETH_HDR_SIZE;
2775 		adapter->rx_buf_per_pkt = 1 + (sz + PAGE_SIZE - 1) / PAGE_SIZE;
2776 	}
2777 
2778 	/*
2779 	 * for simplicity, force the ring0 size to be a multiple of
2780 	 * rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN
2781 	 */
2782 	sz = adapter->rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN;
2783 	ring0_size = adapter->rx_queue[0].rx_ring[0].size;
2784 	ring0_size = (ring0_size + sz - 1) / sz * sz;
2785 	ring0_size = min_t(u32, ring0_size, VMXNET3_RX_RING_MAX_SIZE /
2786 			   sz * sz);
2787 	ring1_size = adapter->rx_queue[0].rx_ring[1].size;
2788 	ring1_size = (ring1_size + sz - 1) / sz * sz;
2789 	ring1_size = min_t(u32, ring1_size, VMXNET3_RX_RING2_MAX_SIZE /
2790 			   sz * sz);
2791 	comp_size = ring0_size + ring1_size;
2792 
2793 	for (i = 0; i < adapter->num_rx_queues; i++) {
2794 		struct vmxnet3_rx_queue	*rq = &adapter->rx_queue[i];
2795 
2796 		rq->rx_ring[0].size = ring0_size;
2797 		rq->rx_ring[1].size = ring1_size;
2798 		rq->comp_ring.size = comp_size;
2799 	}
2800 }
2801 
2802 
2803 int
2804 vmxnet3_create_queues(struct vmxnet3_adapter *adapter, u32 tx_ring_size,
2805 		      u32 rx_ring_size, u32 rx_ring2_size,
2806 		      u16 txdata_desc_size, u16 rxdata_desc_size)
2807 {
2808 	int err = 0, i;
2809 
2810 	for (i = 0; i < adapter->num_tx_queues; i++) {
2811 		struct vmxnet3_tx_queue	*tq = &adapter->tx_queue[i];
2812 		tq->tx_ring.size   = tx_ring_size;
2813 		tq->data_ring.size = tx_ring_size;
2814 		tq->comp_ring.size = tx_ring_size;
2815 		tq->txdata_desc_size = txdata_desc_size;
2816 		tq->shared = &adapter->tqd_start[i].ctrl;
2817 		tq->stopped = true;
2818 		tq->adapter = adapter;
2819 		tq->qid = i;
2820 		err = vmxnet3_tq_create(tq, adapter);
2821 		/*
2822 		 * Too late to change num_tx_queues. We cannot do away with
2823 		 * lesser number of queues than what we asked for
2824 		 */
2825 		if (err)
2826 			goto queue_err;
2827 	}
2828 
2829 	adapter->rx_queue[0].rx_ring[0].size = rx_ring_size;
2830 	adapter->rx_queue[0].rx_ring[1].size = rx_ring2_size;
2831 	vmxnet3_adjust_rx_ring_size(adapter);
2832 
2833 	adapter->rxdataring_enabled = VMXNET3_VERSION_GE_3(adapter);
2834 	for (i = 0; i < adapter->num_rx_queues; i++) {
2835 		struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
2836 		/* qid and qid2 for rx queues will be assigned later when num
2837 		 * of rx queues is finalized after allocating intrs */
2838 		rq->shared = &adapter->rqd_start[i].ctrl;
2839 		rq->adapter = adapter;
2840 		rq->data_ring.desc_size = rxdata_desc_size;
2841 		err = vmxnet3_rq_create(rq, adapter);
2842 		if (err) {
2843 			if (i == 0) {
2844 				netdev_err(adapter->netdev,
2845 					   "Could not allocate any rx queues. "
2846 					   "Aborting.\n");
2847 				goto queue_err;
2848 			} else {
2849 				netdev_info(adapter->netdev,
2850 					    "Number of rx queues changed "
2851 					    "to : %d.\n", i);
2852 				adapter->num_rx_queues = i;
2853 				err = 0;
2854 				break;
2855 			}
2856 		}
2857 	}
2858 
2859 	if (!adapter->rxdataring_enabled)
2860 		vmxnet3_rq_destroy_all_rxdataring(adapter);
2861 
2862 	return err;
2863 queue_err:
2864 	vmxnet3_tq_destroy_all(adapter);
2865 	return err;
2866 }
2867 
2868 static int
2869 vmxnet3_open(struct net_device *netdev)
2870 {
2871 	struct vmxnet3_adapter *adapter;
2872 	int err, i;
2873 
2874 	adapter = netdev_priv(netdev);
2875 
2876 	for (i = 0; i < adapter->num_tx_queues; i++)
2877 		spin_lock_init(&adapter->tx_queue[i].tx_lock);
2878 
2879 	if (VMXNET3_VERSION_GE_3(adapter)) {
2880 		unsigned long flags;
2881 		u16 txdata_desc_size;
2882 
2883 		spin_lock_irqsave(&adapter->cmd_lock, flags);
2884 		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2885 				       VMXNET3_CMD_GET_TXDATA_DESC_SIZE);
2886 		txdata_desc_size = VMXNET3_READ_BAR1_REG(adapter,
2887 							 VMXNET3_REG_CMD);
2888 		spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2889 
2890 		if ((txdata_desc_size < VMXNET3_TXDATA_DESC_MIN_SIZE) ||
2891 		    (txdata_desc_size > VMXNET3_TXDATA_DESC_MAX_SIZE) ||
2892 		    (txdata_desc_size & VMXNET3_TXDATA_DESC_SIZE_MASK)) {
2893 			adapter->txdata_desc_size =
2894 				sizeof(struct Vmxnet3_TxDataDesc);
2895 		} else {
2896 			adapter->txdata_desc_size = txdata_desc_size;
2897 		}
2898 	} else {
2899 		adapter->txdata_desc_size = sizeof(struct Vmxnet3_TxDataDesc);
2900 	}
2901 
2902 	err = vmxnet3_create_queues(adapter,
2903 				    adapter->tx_ring_size,
2904 				    adapter->rx_ring_size,
2905 				    adapter->rx_ring2_size,
2906 				    adapter->txdata_desc_size,
2907 				    adapter->rxdata_desc_size);
2908 	if (err)
2909 		goto queue_err;
2910 
2911 	err = vmxnet3_activate_dev(adapter);
2912 	if (err)
2913 		goto activate_err;
2914 
2915 	return 0;
2916 
2917 activate_err:
2918 	vmxnet3_rq_destroy_all(adapter);
2919 	vmxnet3_tq_destroy_all(adapter);
2920 queue_err:
2921 	return err;
2922 }
2923 
2924 
2925 static int
2926 vmxnet3_close(struct net_device *netdev)
2927 {
2928 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2929 
2930 	/*
2931 	 * Reset_work may be in the middle of resetting the device, wait for its
2932 	 * completion.
2933 	 */
2934 	while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2935 		msleep(1);
2936 
2937 	vmxnet3_quiesce_dev(adapter);
2938 
2939 	vmxnet3_rq_destroy_all(adapter);
2940 	vmxnet3_tq_destroy_all(adapter);
2941 
2942 	clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2943 
2944 
2945 	return 0;
2946 }
2947 
2948 
2949 void
2950 vmxnet3_force_close(struct vmxnet3_adapter *adapter)
2951 {
2952 	int i;
2953 
2954 	/*
2955 	 * we must clear VMXNET3_STATE_BIT_RESETTING, otherwise
2956 	 * vmxnet3_close() will deadlock.
2957 	 */
2958 	BUG_ON(test_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state));
2959 
2960 	/* we need to enable NAPI, otherwise dev_close will deadlock */
2961 	for (i = 0; i < adapter->num_rx_queues; i++)
2962 		napi_enable(&adapter->rx_queue[i].napi);
2963 	/*
2964 	 * Need to clear the quiesce bit to ensure that vmxnet3_close
2965 	 * can quiesce the device properly
2966 	 */
2967 	clear_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
2968 	dev_close(adapter->netdev);
2969 }
2970 
2971 
2972 static int
2973 vmxnet3_change_mtu(struct net_device *netdev, int new_mtu)
2974 {
2975 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2976 	int err = 0;
2977 
2978 	netdev->mtu = new_mtu;
2979 
2980 	/*
2981 	 * Reset_work may be in the middle of resetting the device, wait for its
2982 	 * completion.
2983 	 */
2984 	while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2985 		msleep(1);
2986 
2987 	if (netif_running(netdev)) {
2988 		vmxnet3_quiesce_dev(adapter);
2989 		vmxnet3_reset_dev(adapter);
2990 
2991 		/* we need to re-create the rx queue based on the new mtu */
2992 		vmxnet3_rq_destroy_all(adapter);
2993 		vmxnet3_adjust_rx_ring_size(adapter);
2994 		err = vmxnet3_rq_create_all(adapter);
2995 		if (err) {
2996 			netdev_err(netdev,
2997 				   "failed to re-create rx queues, "
2998 				   " error %d. Closing it.\n", err);
2999 			goto out;
3000 		}
3001 
3002 		err = vmxnet3_activate_dev(adapter);
3003 		if (err) {
3004 			netdev_err(netdev,
3005 				   "failed to re-activate, error %d. "
3006 				   "Closing it\n", err);
3007 			goto out;
3008 		}
3009 	}
3010 
3011 out:
3012 	clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
3013 	if (err)
3014 		vmxnet3_force_close(adapter);
3015 
3016 	return err;
3017 }
3018 
3019 
3020 static void
3021 vmxnet3_declare_features(struct vmxnet3_adapter *adapter, bool dma64)
3022 {
3023 	struct net_device *netdev = adapter->netdev;
3024 
3025 	netdev->hw_features = NETIF_F_SG | NETIF_F_RXCSUM |
3026 		NETIF_F_HW_CSUM | NETIF_F_HW_VLAN_CTAG_TX |
3027 		NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_TSO | NETIF_F_TSO6 |
3028 		NETIF_F_LRO;
3029 	if (dma64)
3030 		netdev->hw_features |= NETIF_F_HIGHDMA;
3031 	netdev->vlan_features = netdev->hw_features &
3032 				~(NETIF_F_HW_VLAN_CTAG_TX |
3033 				  NETIF_F_HW_VLAN_CTAG_RX);
3034 	netdev->features = netdev->hw_features | NETIF_F_HW_VLAN_CTAG_FILTER;
3035 }
3036 
3037 
3038 static void
3039 vmxnet3_read_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac)
3040 {
3041 	u32 tmp;
3042 
3043 	tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACL);
3044 	*(u32 *)mac = tmp;
3045 
3046 	tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACH);
3047 	mac[4] = tmp & 0xff;
3048 	mac[5] = (tmp >> 8) & 0xff;
3049 }
3050 
3051 #ifdef CONFIG_PCI_MSI
3052 
3053 /*
3054  * Enable MSIx vectors.
3055  * Returns :
3056  *	VMXNET3_LINUX_MIN_MSIX_VECT when only minimum number of vectors required
3057  *	 were enabled.
3058  *	number of vectors which were enabled otherwise (this number is greater
3059  *	 than VMXNET3_LINUX_MIN_MSIX_VECT)
3060  */
3061 
3062 static int
3063 vmxnet3_acquire_msix_vectors(struct vmxnet3_adapter *adapter, int nvec)
3064 {
3065 	int ret = pci_enable_msix_range(adapter->pdev,
3066 					adapter->intr.msix_entries, nvec, nvec);
3067 
3068 	if (ret == -ENOSPC && nvec > VMXNET3_LINUX_MIN_MSIX_VECT) {
3069 		dev_err(&adapter->netdev->dev,
3070 			"Failed to enable %d MSI-X, trying %d\n",
3071 			nvec, VMXNET3_LINUX_MIN_MSIX_VECT);
3072 
3073 		ret = pci_enable_msix_range(adapter->pdev,
3074 					    adapter->intr.msix_entries,
3075 					    VMXNET3_LINUX_MIN_MSIX_VECT,
3076 					    VMXNET3_LINUX_MIN_MSIX_VECT);
3077 	}
3078 
3079 	if (ret < 0) {
3080 		dev_err(&adapter->netdev->dev,
3081 			"Failed to enable MSI-X, error: %d\n", ret);
3082 	}
3083 
3084 	return ret;
3085 }
3086 
3087 
3088 #endif /* CONFIG_PCI_MSI */
3089 
3090 static void
3091 vmxnet3_alloc_intr_resources(struct vmxnet3_adapter *adapter)
3092 {
3093 	u32 cfg;
3094 	unsigned long flags;
3095 
3096 	/* intr settings */
3097 	spin_lock_irqsave(&adapter->cmd_lock, flags);
3098 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3099 			       VMXNET3_CMD_GET_CONF_INTR);
3100 	cfg = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
3101 	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3102 	adapter->intr.type = cfg & 0x3;
3103 	adapter->intr.mask_mode = (cfg >> 2) & 0x3;
3104 
3105 	if (adapter->intr.type == VMXNET3_IT_AUTO) {
3106 		adapter->intr.type = VMXNET3_IT_MSIX;
3107 	}
3108 
3109 #ifdef CONFIG_PCI_MSI
3110 	if (adapter->intr.type == VMXNET3_IT_MSIX) {
3111 		int i, nvec;
3112 
3113 		nvec  = adapter->share_intr == VMXNET3_INTR_TXSHARE ?
3114 			1 : adapter->num_tx_queues;
3115 		nvec += adapter->share_intr == VMXNET3_INTR_BUDDYSHARE ?
3116 			0 : adapter->num_rx_queues;
3117 		nvec += 1;	/* for link event */
3118 		nvec = nvec > VMXNET3_LINUX_MIN_MSIX_VECT ?
3119 		       nvec : VMXNET3_LINUX_MIN_MSIX_VECT;
3120 
3121 		for (i = 0; i < nvec; i++)
3122 			adapter->intr.msix_entries[i].entry = i;
3123 
3124 		nvec = vmxnet3_acquire_msix_vectors(adapter, nvec);
3125 		if (nvec < 0)
3126 			goto msix_err;
3127 
3128 		/* If we cannot allocate one MSIx vector per queue
3129 		 * then limit the number of rx queues to 1
3130 		 */
3131 		if (nvec == VMXNET3_LINUX_MIN_MSIX_VECT) {
3132 			if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE
3133 			    || adapter->num_rx_queues != 1) {
3134 				adapter->share_intr = VMXNET3_INTR_TXSHARE;
3135 				netdev_err(adapter->netdev,
3136 					   "Number of rx queues : 1\n");
3137 				adapter->num_rx_queues = 1;
3138 			}
3139 		}
3140 
3141 		adapter->intr.num_intrs = nvec;
3142 		return;
3143 
3144 msix_err:
3145 		/* If we cannot allocate MSIx vectors use only one rx queue */
3146 		dev_info(&adapter->pdev->dev,
3147 			 "Failed to enable MSI-X, error %d. "
3148 			 "Limiting #rx queues to 1, try MSI.\n", nvec);
3149 
3150 		adapter->intr.type = VMXNET3_IT_MSI;
3151 	}
3152 
3153 	if (adapter->intr.type == VMXNET3_IT_MSI) {
3154 		if (!pci_enable_msi(adapter->pdev)) {
3155 			adapter->num_rx_queues = 1;
3156 			adapter->intr.num_intrs = 1;
3157 			return;
3158 		}
3159 	}
3160 #endif /* CONFIG_PCI_MSI */
3161 
3162 	adapter->num_rx_queues = 1;
3163 	dev_info(&adapter->netdev->dev,
3164 		 "Using INTx interrupt, #Rx queues: 1.\n");
3165 	adapter->intr.type = VMXNET3_IT_INTX;
3166 
3167 	/* INT-X related setting */
3168 	adapter->intr.num_intrs = 1;
3169 }
3170 
3171 
3172 static void
3173 vmxnet3_free_intr_resources(struct vmxnet3_adapter *adapter)
3174 {
3175 	if (adapter->intr.type == VMXNET3_IT_MSIX)
3176 		pci_disable_msix(adapter->pdev);
3177 	else if (adapter->intr.type == VMXNET3_IT_MSI)
3178 		pci_disable_msi(adapter->pdev);
3179 	else
3180 		BUG_ON(adapter->intr.type != VMXNET3_IT_INTX);
3181 }
3182 
3183 
3184 static void
3185 vmxnet3_tx_timeout(struct net_device *netdev)
3186 {
3187 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3188 	adapter->tx_timeout_count++;
3189 
3190 	netdev_err(adapter->netdev, "tx hang\n");
3191 	schedule_work(&adapter->work);
3192 }
3193 
3194 
3195 static void
3196 vmxnet3_reset_work(struct work_struct *data)
3197 {
3198 	struct vmxnet3_adapter *adapter;
3199 
3200 	adapter = container_of(data, struct vmxnet3_adapter, work);
3201 
3202 	/* if another thread is resetting the device, no need to proceed */
3203 	if (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
3204 		return;
3205 
3206 	/* if the device is closed, we must leave it alone */
3207 	rtnl_lock();
3208 	if (netif_running(adapter->netdev)) {
3209 		netdev_notice(adapter->netdev, "resetting\n");
3210 		vmxnet3_quiesce_dev(adapter);
3211 		vmxnet3_reset_dev(adapter);
3212 		vmxnet3_activate_dev(adapter);
3213 	} else {
3214 		netdev_info(adapter->netdev, "already closed\n");
3215 	}
3216 	rtnl_unlock();
3217 
3218 	netif_wake_queue(adapter->netdev);
3219 	clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
3220 }
3221 
3222 
3223 static int
3224 vmxnet3_probe_device(struct pci_dev *pdev,
3225 		     const struct pci_device_id *id)
3226 {
3227 	static const struct net_device_ops vmxnet3_netdev_ops = {
3228 		.ndo_open = vmxnet3_open,
3229 		.ndo_stop = vmxnet3_close,
3230 		.ndo_start_xmit = vmxnet3_xmit_frame,
3231 		.ndo_set_mac_address = vmxnet3_set_mac_addr,
3232 		.ndo_change_mtu = vmxnet3_change_mtu,
3233 		.ndo_set_features = vmxnet3_set_features,
3234 		.ndo_get_stats64 = vmxnet3_get_stats64,
3235 		.ndo_tx_timeout = vmxnet3_tx_timeout,
3236 		.ndo_set_rx_mode = vmxnet3_set_mc,
3237 		.ndo_vlan_rx_add_vid = vmxnet3_vlan_rx_add_vid,
3238 		.ndo_vlan_rx_kill_vid = vmxnet3_vlan_rx_kill_vid,
3239 #ifdef CONFIG_NET_POLL_CONTROLLER
3240 		.ndo_poll_controller = vmxnet3_netpoll,
3241 #endif
3242 	};
3243 	int err;
3244 	bool dma64 = false; /* stupid gcc */
3245 	u32 ver;
3246 	struct net_device *netdev;
3247 	struct vmxnet3_adapter *adapter;
3248 	u8 mac[ETH_ALEN];
3249 	int size;
3250 	int num_tx_queues;
3251 	int num_rx_queues;
3252 
3253 	if (!pci_msi_enabled())
3254 		enable_mq = 0;
3255 
3256 #ifdef VMXNET3_RSS
3257 	if (enable_mq)
3258 		num_rx_queues = min(VMXNET3_DEVICE_MAX_RX_QUEUES,
3259 				    (int)num_online_cpus());
3260 	else
3261 #endif
3262 		num_rx_queues = 1;
3263 	num_rx_queues = rounddown_pow_of_two(num_rx_queues);
3264 
3265 	if (enable_mq)
3266 		num_tx_queues = min(VMXNET3_DEVICE_MAX_TX_QUEUES,
3267 				    (int)num_online_cpus());
3268 	else
3269 		num_tx_queues = 1;
3270 
3271 	num_tx_queues = rounddown_pow_of_two(num_tx_queues);
3272 	netdev = alloc_etherdev_mq(sizeof(struct vmxnet3_adapter),
3273 				   max(num_tx_queues, num_rx_queues));
3274 	dev_info(&pdev->dev,
3275 		 "# of Tx queues : %d, # of Rx queues : %d\n",
3276 		 num_tx_queues, num_rx_queues);
3277 
3278 	if (!netdev)
3279 		return -ENOMEM;
3280 
3281 	pci_set_drvdata(pdev, netdev);
3282 	adapter = netdev_priv(netdev);
3283 	adapter->netdev = netdev;
3284 	adapter->pdev = pdev;
3285 
3286 	adapter->tx_ring_size = VMXNET3_DEF_TX_RING_SIZE;
3287 	adapter->rx_ring_size = VMXNET3_DEF_RX_RING_SIZE;
3288 	adapter->rx_ring2_size = VMXNET3_DEF_RX_RING2_SIZE;
3289 
3290 	spin_lock_init(&adapter->cmd_lock);
3291 	adapter->adapter_pa = dma_map_single(&adapter->pdev->dev, adapter,
3292 					     sizeof(struct vmxnet3_adapter),
3293 					     PCI_DMA_TODEVICE);
3294 	if (dma_mapping_error(&adapter->pdev->dev, adapter->adapter_pa)) {
3295 		dev_err(&pdev->dev, "Failed to map dma\n");
3296 		err = -EFAULT;
3297 		goto err_dma_map;
3298 	}
3299 	adapter->shared = dma_alloc_coherent(
3300 				&adapter->pdev->dev,
3301 				sizeof(struct Vmxnet3_DriverShared),
3302 				&adapter->shared_pa, GFP_KERNEL);
3303 	if (!adapter->shared) {
3304 		dev_err(&pdev->dev, "Failed to allocate memory\n");
3305 		err = -ENOMEM;
3306 		goto err_alloc_shared;
3307 	}
3308 
3309 	adapter->num_rx_queues = num_rx_queues;
3310 	adapter->num_tx_queues = num_tx_queues;
3311 	adapter->rx_buf_per_pkt = 1;
3312 
3313 	size = sizeof(struct Vmxnet3_TxQueueDesc) * adapter->num_tx_queues;
3314 	size += sizeof(struct Vmxnet3_RxQueueDesc) * adapter->num_rx_queues;
3315 	adapter->tqd_start = dma_alloc_coherent(&adapter->pdev->dev, size,
3316 						&adapter->queue_desc_pa,
3317 						GFP_KERNEL);
3318 
3319 	if (!adapter->tqd_start) {
3320 		dev_err(&pdev->dev, "Failed to allocate memory\n");
3321 		err = -ENOMEM;
3322 		goto err_alloc_queue_desc;
3323 	}
3324 	adapter->rqd_start = (struct Vmxnet3_RxQueueDesc *)(adapter->tqd_start +
3325 							    adapter->num_tx_queues);
3326 
3327 	adapter->pm_conf = dma_alloc_coherent(&adapter->pdev->dev,
3328 					      sizeof(struct Vmxnet3_PMConf),
3329 					      &adapter->pm_conf_pa,
3330 					      GFP_KERNEL);
3331 	if (adapter->pm_conf == NULL) {
3332 		err = -ENOMEM;
3333 		goto err_alloc_pm;
3334 	}
3335 
3336 #ifdef VMXNET3_RSS
3337 
3338 	adapter->rss_conf = dma_alloc_coherent(&adapter->pdev->dev,
3339 					       sizeof(struct UPT1_RSSConf),
3340 					       &adapter->rss_conf_pa,
3341 					       GFP_KERNEL);
3342 	if (adapter->rss_conf == NULL) {
3343 		err = -ENOMEM;
3344 		goto err_alloc_rss;
3345 	}
3346 #endif /* VMXNET3_RSS */
3347 
3348 	err = vmxnet3_alloc_pci_resources(adapter, &dma64);
3349 	if (err < 0)
3350 		goto err_alloc_pci;
3351 
3352 	ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_VRRS);
3353 	if (ver & (1 << VMXNET3_REV_3)) {
3354 		VMXNET3_WRITE_BAR1_REG(adapter,
3355 				       VMXNET3_REG_VRRS,
3356 				       1 << VMXNET3_REV_3);
3357 		adapter->version = VMXNET3_REV_3 + 1;
3358 	} else if (ver & (1 << VMXNET3_REV_2)) {
3359 		VMXNET3_WRITE_BAR1_REG(adapter,
3360 				       VMXNET3_REG_VRRS,
3361 				       1 << VMXNET3_REV_2);
3362 		adapter->version = VMXNET3_REV_2 + 1;
3363 	} else if (ver & (1 << VMXNET3_REV_1)) {
3364 		VMXNET3_WRITE_BAR1_REG(adapter,
3365 				       VMXNET3_REG_VRRS,
3366 				       1 << VMXNET3_REV_1);
3367 		adapter->version = VMXNET3_REV_1 + 1;
3368 	} else {
3369 		dev_err(&pdev->dev,
3370 			"Incompatible h/w version (0x%x) for adapter\n", ver);
3371 		err = -EBUSY;
3372 		goto err_ver;
3373 	}
3374 	dev_dbg(&pdev->dev, "Using device version %d\n", adapter->version);
3375 
3376 	ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_UVRS);
3377 	if (ver & 1) {
3378 		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_UVRS, 1);
3379 	} else {
3380 		dev_err(&pdev->dev,
3381 			"Incompatible upt version (0x%x) for adapter\n", ver);
3382 		err = -EBUSY;
3383 		goto err_ver;
3384 	}
3385 
3386 	if (VMXNET3_VERSION_GE_3(adapter)) {
3387 		adapter->coal_conf =
3388 			dma_alloc_coherent(&adapter->pdev->dev,
3389 					   sizeof(struct Vmxnet3_CoalesceScheme)
3390 					   ,
3391 					   &adapter->coal_conf_pa,
3392 					   GFP_KERNEL);
3393 		if (!adapter->coal_conf) {
3394 			err = -ENOMEM;
3395 			goto err_ver;
3396 		}
3397 		memset(adapter->coal_conf, 0, sizeof(*adapter->coal_conf));
3398 		adapter->coal_conf->coalMode = VMXNET3_COALESCE_DISABLED;
3399 		adapter->default_coal_mode = true;
3400 	}
3401 
3402 	SET_NETDEV_DEV(netdev, &pdev->dev);
3403 	vmxnet3_declare_features(adapter, dma64);
3404 
3405 	adapter->rxdata_desc_size = VMXNET3_VERSION_GE_3(adapter) ?
3406 		VMXNET3_DEF_RXDATA_DESC_SIZE : 0;
3407 
3408 	if (adapter->num_tx_queues == adapter->num_rx_queues)
3409 		adapter->share_intr = VMXNET3_INTR_BUDDYSHARE;
3410 	else
3411 		adapter->share_intr = VMXNET3_INTR_DONTSHARE;
3412 
3413 	vmxnet3_alloc_intr_resources(adapter);
3414 
3415 #ifdef VMXNET3_RSS
3416 	if (adapter->num_rx_queues > 1 &&
3417 	    adapter->intr.type == VMXNET3_IT_MSIX) {
3418 		adapter->rss = true;
3419 		netdev->hw_features |= NETIF_F_RXHASH;
3420 		netdev->features |= NETIF_F_RXHASH;
3421 		dev_dbg(&pdev->dev, "RSS is enabled.\n");
3422 	} else {
3423 		adapter->rss = false;
3424 	}
3425 #endif
3426 
3427 	vmxnet3_read_mac_addr(adapter, mac);
3428 	memcpy(netdev->dev_addr,  mac, netdev->addr_len);
3429 
3430 	netdev->netdev_ops = &vmxnet3_netdev_ops;
3431 	vmxnet3_set_ethtool_ops(netdev);
3432 	netdev->watchdog_timeo = 5 * HZ;
3433 
3434 	/* MTU range: 60 - 9000 */
3435 	netdev->min_mtu = VMXNET3_MIN_MTU;
3436 	netdev->max_mtu = VMXNET3_MAX_MTU;
3437 
3438 	INIT_WORK(&adapter->work, vmxnet3_reset_work);
3439 	set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
3440 
3441 	if (adapter->intr.type == VMXNET3_IT_MSIX) {
3442 		int i;
3443 		for (i = 0; i < adapter->num_rx_queues; i++) {
3444 			netif_napi_add(adapter->netdev,
3445 				       &adapter->rx_queue[i].napi,
3446 				       vmxnet3_poll_rx_only, 64);
3447 		}
3448 	} else {
3449 		netif_napi_add(adapter->netdev, &adapter->rx_queue[0].napi,
3450 			       vmxnet3_poll, 64);
3451 	}
3452 
3453 	netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
3454 	netif_set_real_num_rx_queues(adapter->netdev, adapter->num_rx_queues);
3455 
3456 	netif_carrier_off(netdev);
3457 	err = register_netdev(netdev);
3458 
3459 	if (err) {
3460 		dev_err(&pdev->dev, "Failed to register adapter\n");
3461 		goto err_register;
3462 	}
3463 
3464 	vmxnet3_check_link(adapter, false);
3465 	return 0;
3466 
3467 err_register:
3468 	if (VMXNET3_VERSION_GE_3(adapter)) {
3469 		dma_free_coherent(&adapter->pdev->dev,
3470 				  sizeof(struct Vmxnet3_CoalesceScheme),
3471 				  adapter->coal_conf, adapter->coal_conf_pa);
3472 	}
3473 	vmxnet3_free_intr_resources(adapter);
3474 err_ver:
3475 	vmxnet3_free_pci_resources(adapter);
3476 err_alloc_pci:
3477 #ifdef VMXNET3_RSS
3478 	dma_free_coherent(&adapter->pdev->dev, sizeof(struct UPT1_RSSConf),
3479 			  adapter->rss_conf, adapter->rss_conf_pa);
3480 err_alloc_rss:
3481 #endif
3482 	dma_free_coherent(&adapter->pdev->dev, sizeof(struct Vmxnet3_PMConf),
3483 			  adapter->pm_conf, adapter->pm_conf_pa);
3484 err_alloc_pm:
3485 	dma_free_coherent(&adapter->pdev->dev, size, adapter->tqd_start,
3486 			  adapter->queue_desc_pa);
3487 err_alloc_queue_desc:
3488 	dma_free_coherent(&adapter->pdev->dev,
3489 			  sizeof(struct Vmxnet3_DriverShared),
3490 			  adapter->shared, adapter->shared_pa);
3491 err_alloc_shared:
3492 	dma_unmap_single(&adapter->pdev->dev, adapter->adapter_pa,
3493 			 sizeof(struct vmxnet3_adapter), PCI_DMA_TODEVICE);
3494 err_dma_map:
3495 	free_netdev(netdev);
3496 	return err;
3497 }
3498 
3499 
3500 static void
3501 vmxnet3_remove_device(struct pci_dev *pdev)
3502 {
3503 	struct net_device *netdev = pci_get_drvdata(pdev);
3504 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3505 	int size = 0;
3506 	int num_rx_queues;
3507 
3508 #ifdef VMXNET3_RSS
3509 	if (enable_mq)
3510 		num_rx_queues = min(VMXNET3_DEVICE_MAX_RX_QUEUES,
3511 				    (int)num_online_cpus());
3512 	else
3513 #endif
3514 		num_rx_queues = 1;
3515 	num_rx_queues = rounddown_pow_of_two(num_rx_queues);
3516 
3517 	cancel_work_sync(&adapter->work);
3518 
3519 	unregister_netdev(netdev);
3520 
3521 	vmxnet3_free_intr_resources(adapter);
3522 	vmxnet3_free_pci_resources(adapter);
3523 	if (VMXNET3_VERSION_GE_3(adapter)) {
3524 		dma_free_coherent(&adapter->pdev->dev,
3525 				  sizeof(struct Vmxnet3_CoalesceScheme),
3526 				  adapter->coal_conf, adapter->coal_conf_pa);
3527 	}
3528 #ifdef VMXNET3_RSS
3529 	dma_free_coherent(&adapter->pdev->dev, sizeof(struct UPT1_RSSConf),
3530 			  adapter->rss_conf, adapter->rss_conf_pa);
3531 #endif
3532 	dma_free_coherent(&adapter->pdev->dev, sizeof(struct Vmxnet3_PMConf),
3533 			  adapter->pm_conf, adapter->pm_conf_pa);
3534 
3535 	size = sizeof(struct Vmxnet3_TxQueueDesc) * adapter->num_tx_queues;
3536 	size += sizeof(struct Vmxnet3_RxQueueDesc) * num_rx_queues;
3537 	dma_free_coherent(&adapter->pdev->dev, size, adapter->tqd_start,
3538 			  adapter->queue_desc_pa);
3539 	dma_free_coherent(&adapter->pdev->dev,
3540 			  sizeof(struct Vmxnet3_DriverShared),
3541 			  adapter->shared, adapter->shared_pa);
3542 	dma_unmap_single(&adapter->pdev->dev, adapter->adapter_pa,
3543 			 sizeof(struct vmxnet3_adapter), PCI_DMA_TODEVICE);
3544 	free_netdev(netdev);
3545 }
3546 
3547 static void vmxnet3_shutdown_device(struct pci_dev *pdev)
3548 {
3549 	struct net_device *netdev = pci_get_drvdata(pdev);
3550 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3551 	unsigned long flags;
3552 
3553 	/* Reset_work may be in the middle of resetting the device, wait for its
3554 	 * completion.
3555 	 */
3556 	while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
3557 		msleep(1);
3558 
3559 	if (test_and_set_bit(VMXNET3_STATE_BIT_QUIESCED,
3560 			     &adapter->state)) {
3561 		clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
3562 		return;
3563 	}
3564 	spin_lock_irqsave(&adapter->cmd_lock, flags);
3565 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3566 			       VMXNET3_CMD_QUIESCE_DEV);
3567 	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3568 	vmxnet3_disable_all_intrs(adapter);
3569 
3570 	clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
3571 }
3572 
3573 
3574 #ifdef CONFIG_PM
3575 
3576 static int
3577 vmxnet3_suspend(struct device *device)
3578 {
3579 	struct pci_dev *pdev = to_pci_dev(device);
3580 	struct net_device *netdev = pci_get_drvdata(pdev);
3581 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3582 	struct Vmxnet3_PMConf *pmConf;
3583 	struct ethhdr *ehdr;
3584 	struct arphdr *ahdr;
3585 	u8 *arpreq;
3586 	struct in_device *in_dev;
3587 	struct in_ifaddr *ifa;
3588 	unsigned long flags;
3589 	int i = 0;
3590 
3591 	if (!netif_running(netdev))
3592 		return 0;
3593 
3594 	for (i = 0; i < adapter->num_rx_queues; i++)
3595 		napi_disable(&adapter->rx_queue[i].napi);
3596 
3597 	vmxnet3_disable_all_intrs(adapter);
3598 	vmxnet3_free_irqs(adapter);
3599 	vmxnet3_free_intr_resources(adapter);
3600 
3601 	netif_device_detach(netdev);
3602 	netif_tx_stop_all_queues(netdev);
3603 
3604 	/* Create wake-up filters. */
3605 	pmConf = adapter->pm_conf;
3606 	memset(pmConf, 0, sizeof(*pmConf));
3607 
3608 	if (adapter->wol & WAKE_UCAST) {
3609 		pmConf->filters[i].patternSize = ETH_ALEN;
3610 		pmConf->filters[i].maskSize = 1;
3611 		memcpy(pmConf->filters[i].pattern, netdev->dev_addr, ETH_ALEN);
3612 		pmConf->filters[i].mask[0] = 0x3F; /* LSB ETH_ALEN bits */
3613 
3614 		pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_FILTER;
3615 		i++;
3616 	}
3617 
3618 	if (adapter->wol & WAKE_ARP) {
3619 		in_dev = in_dev_get(netdev);
3620 		if (!in_dev)
3621 			goto skip_arp;
3622 
3623 		ifa = (struct in_ifaddr *)in_dev->ifa_list;
3624 		if (!ifa)
3625 			goto skip_arp;
3626 
3627 		pmConf->filters[i].patternSize = ETH_HLEN + /* Ethernet header*/
3628 			sizeof(struct arphdr) +		/* ARP header */
3629 			2 * ETH_ALEN +		/* 2 Ethernet addresses*/
3630 			2 * sizeof(u32);	/*2 IPv4 addresses */
3631 		pmConf->filters[i].maskSize =
3632 			(pmConf->filters[i].patternSize - 1) / 8 + 1;
3633 
3634 		/* ETH_P_ARP in Ethernet header. */
3635 		ehdr = (struct ethhdr *)pmConf->filters[i].pattern;
3636 		ehdr->h_proto = htons(ETH_P_ARP);
3637 
3638 		/* ARPOP_REQUEST in ARP header. */
3639 		ahdr = (struct arphdr *)&pmConf->filters[i].pattern[ETH_HLEN];
3640 		ahdr->ar_op = htons(ARPOP_REQUEST);
3641 		arpreq = (u8 *)(ahdr + 1);
3642 
3643 		/* The Unicast IPv4 address in 'tip' field. */
3644 		arpreq += 2 * ETH_ALEN + sizeof(u32);
3645 		*(u32 *)arpreq = ifa->ifa_address;
3646 
3647 		/* The mask for the relevant bits. */
3648 		pmConf->filters[i].mask[0] = 0x00;
3649 		pmConf->filters[i].mask[1] = 0x30; /* ETH_P_ARP */
3650 		pmConf->filters[i].mask[2] = 0x30; /* ARPOP_REQUEST */
3651 		pmConf->filters[i].mask[3] = 0x00;
3652 		pmConf->filters[i].mask[4] = 0xC0; /* IPv4 TIP */
3653 		pmConf->filters[i].mask[5] = 0x03; /* IPv4 TIP */
3654 		in_dev_put(in_dev);
3655 
3656 		pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_FILTER;
3657 		i++;
3658 	}
3659 
3660 skip_arp:
3661 	if (adapter->wol & WAKE_MAGIC)
3662 		pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_MAGIC;
3663 
3664 	pmConf->numFilters = i;
3665 
3666 	adapter->shared->devRead.pmConfDesc.confVer = cpu_to_le32(1);
3667 	adapter->shared->devRead.pmConfDesc.confLen = cpu_to_le32(sizeof(
3668 								  *pmConf));
3669 	adapter->shared->devRead.pmConfDesc.confPA =
3670 		cpu_to_le64(adapter->pm_conf_pa);
3671 
3672 	spin_lock_irqsave(&adapter->cmd_lock, flags);
3673 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3674 			       VMXNET3_CMD_UPDATE_PMCFG);
3675 	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3676 
3677 	pci_save_state(pdev);
3678 	pci_enable_wake(pdev, pci_choose_state(pdev, PMSG_SUSPEND),
3679 			adapter->wol);
3680 	pci_disable_device(pdev);
3681 	pci_set_power_state(pdev, pci_choose_state(pdev, PMSG_SUSPEND));
3682 
3683 	return 0;
3684 }
3685 
3686 
3687 static int
3688 vmxnet3_resume(struct device *device)
3689 {
3690 	int err;
3691 	unsigned long flags;
3692 	struct pci_dev *pdev = to_pci_dev(device);
3693 	struct net_device *netdev = pci_get_drvdata(pdev);
3694 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3695 
3696 	if (!netif_running(netdev))
3697 		return 0;
3698 
3699 	pci_set_power_state(pdev, PCI_D0);
3700 	pci_restore_state(pdev);
3701 	err = pci_enable_device_mem(pdev);
3702 	if (err != 0)
3703 		return err;
3704 
3705 	pci_enable_wake(pdev, PCI_D0, 0);
3706 
3707 	vmxnet3_alloc_intr_resources(adapter);
3708 
3709 	/* During hibernate and suspend, device has to be reinitialized as the
3710 	 * device state need not be preserved.
3711 	 */
3712 
3713 	/* Need not check adapter state as other reset tasks cannot run during
3714 	 * device resume.
3715 	 */
3716 	spin_lock_irqsave(&adapter->cmd_lock, flags);
3717 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3718 			       VMXNET3_CMD_QUIESCE_DEV);
3719 	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3720 	vmxnet3_tq_cleanup_all(adapter);
3721 	vmxnet3_rq_cleanup_all(adapter);
3722 
3723 	vmxnet3_reset_dev(adapter);
3724 	err = vmxnet3_activate_dev(adapter);
3725 	if (err != 0) {
3726 		netdev_err(netdev,
3727 			   "failed to re-activate on resume, error: %d", err);
3728 		vmxnet3_force_close(adapter);
3729 		return err;
3730 	}
3731 	netif_device_attach(netdev);
3732 
3733 	return 0;
3734 }
3735 
3736 static const struct dev_pm_ops vmxnet3_pm_ops = {
3737 	.suspend = vmxnet3_suspend,
3738 	.resume = vmxnet3_resume,
3739 	.freeze = vmxnet3_suspend,
3740 	.restore = vmxnet3_resume,
3741 };
3742 #endif
3743 
3744 static struct pci_driver vmxnet3_driver = {
3745 	.name		= vmxnet3_driver_name,
3746 	.id_table	= vmxnet3_pciid_table,
3747 	.probe		= vmxnet3_probe_device,
3748 	.remove		= vmxnet3_remove_device,
3749 	.shutdown	= vmxnet3_shutdown_device,
3750 #ifdef CONFIG_PM
3751 	.driver.pm	= &vmxnet3_pm_ops,
3752 #endif
3753 };
3754 
3755 
3756 static int __init
3757 vmxnet3_init_module(void)
3758 {
3759 	pr_info("%s - version %s\n", VMXNET3_DRIVER_DESC,
3760 		VMXNET3_DRIVER_VERSION_REPORT);
3761 	return pci_register_driver(&vmxnet3_driver);
3762 }
3763 
3764 module_init(vmxnet3_init_module);
3765 
3766 
3767 static void
3768 vmxnet3_exit_module(void)
3769 {
3770 	pci_unregister_driver(&vmxnet3_driver);
3771 }
3772 
3773 module_exit(vmxnet3_exit_module);
3774 
3775 MODULE_AUTHOR("VMware, Inc.");
3776 MODULE_DESCRIPTION(VMXNET3_DRIVER_DESC);
3777 MODULE_LICENSE("GPL v2");
3778 MODULE_VERSION(VMXNET3_DRIVER_VERSION_STRING);
3779