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