1 /*
2  * Linux driver for VMware's vmxnet3 ethernet NIC.
3  *
4  * Copyright (C) 2008-2009, VMware, Inc. All Rights Reserved.
5  *
6  * This program is free software; you can redistribute it and/or modify it
7  * under the terms of the GNU General Public License as published by the
8  * Free Software Foundation; version 2 of the License and no later version.
9  *
10  * This program is distributed in the hope that it will be useful, but
11  * WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
13  * NON INFRINGEMENT. See the GNU General Public License for more
14  * details.
15  *
16  * You should have received a copy of the GNU General Public License
17  * along with this program; if not, write to the Free Software
18  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19  *
20  * The full GNU General Public License is included in this distribution in
21  * the file called "COPYING".
22  *
23  * Maintained by: Shreyas Bhatewara <pv-drivers@vmware.com>
24  *
25  */
26 
27 #include <linux/module.h>
28 #include <net/ip6_checksum.h>
29 
30 #include "vmxnet3_int.h"
31 
32 char vmxnet3_driver_name[] = "vmxnet3";
33 #define VMXNET3_DRIVER_DESC "VMware vmxnet3 virtual NIC driver"
34 
35 /*
36  * PCI Device ID Table
37  * Last entry must be all 0s
38  */
39 static DEFINE_PCI_DEVICE_TABLE(vmxnet3_pciid_table) = {
40 	{PCI_VDEVICE(VMWARE, PCI_DEVICE_ID_VMWARE_VMXNET3)},
41 	{0}
42 };
43 
44 MODULE_DEVICE_TABLE(pci, vmxnet3_pciid_table);
45 
46 static atomic_t devices_found;
47 
48 #define VMXNET3_MAX_DEVICES 10
49 static int enable_mq = 1;
50 static int irq_share_mode;
51 
52 static void
53 vmxnet3_write_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac);
54 
55 /*
56  *    Enable/Disable the given intr
57  */
58 static void
59 vmxnet3_enable_intr(struct vmxnet3_adapter *adapter, unsigned intr_idx)
60 {
61 	VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_IMR + intr_idx * 8, 0);
62 }
63 
64 
65 static void
66 vmxnet3_disable_intr(struct vmxnet3_adapter *adapter, unsigned intr_idx)
67 {
68 	VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_IMR + intr_idx * 8, 1);
69 }
70 
71 
72 /*
73  *    Enable/Disable all intrs used by the device
74  */
75 static void
76 vmxnet3_enable_all_intrs(struct vmxnet3_adapter *adapter)
77 {
78 	int i;
79 
80 	for (i = 0; i < adapter->intr.num_intrs; i++)
81 		vmxnet3_enable_intr(adapter, i);
82 	adapter->shared->devRead.intrConf.intrCtrl &=
83 					cpu_to_le32(~VMXNET3_IC_DISABLE_ALL);
84 }
85 
86 
87 static void
88 vmxnet3_disable_all_intrs(struct vmxnet3_adapter *adapter)
89 {
90 	int i;
91 
92 	adapter->shared->devRead.intrConf.intrCtrl |=
93 					cpu_to_le32(VMXNET3_IC_DISABLE_ALL);
94 	for (i = 0; i < adapter->intr.num_intrs; i++)
95 		vmxnet3_disable_intr(adapter, i);
96 }
97 
98 
99 static void
100 vmxnet3_ack_events(struct vmxnet3_adapter *adapter, u32 events)
101 {
102 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_ECR, events);
103 }
104 
105 
106 static bool
107 vmxnet3_tq_stopped(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
108 {
109 	return tq->stopped;
110 }
111 
112 
113 static void
114 vmxnet3_tq_start(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
115 {
116 	tq->stopped = false;
117 	netif_start_subqueue(adapter->netdev, tq - adapter->tx_queue);
118 }
119 
120 
121 static void
122 vmxnet3_tq_wake(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
123 {
124 	tq->stopped = false;
125 	netif_wake_subqueue(adapter->netdev, (tq - adapter->tx_queue));
126 }
127 
128 
129 static void
130 vmxnet3_tq_stop(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
131 {
132 	tq->stopped = true;
133 	tq->num_stop++;
134 	netif_stop_subqueue(adapter->netdev, (tq - adapter->tx_queue));
135 }
136 
137 
138 /*
139  * Check the link state. This may start or stop the tx queue.
140  */
141 static void
142 vmxnet3_check_link(struct vmxnet3_adapter *adapter, bool affectTxQueue)
143 {
144 	u32 ret;
145 	int i;
146 	unsigned long flags;
147 
148 	spin_lock_irqsave(&adapter->cmd_lock, flags);
149 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, VMXNET3_CMD_GET_LINK);
150 	ret = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
151 	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
152 
153 	adapter->link_speed = ret >> 16;
154 	if (ret & 1) { /* Link is up. */
155 		printk(KERN_INFO "%s: NIC Link is Up %d Mbps\n",
156 		       adapter->netdev->name, adapter->link_speed);
157 		if (!netif_carrier_ok(adapter->netdev))
158 			netif_carrier_on(adapter->netdev);
159 
160 		if (affectTxQueue) {
161 			for (i = 0; i < adapter->num_tx_queues; i++)
162 				vmxnet3_tq_start(&adapter->tx_queue[i],
163 						 adapter);
164 		}
165 	} else {
166 		printk(KERN_INFO "%s: NIC Link is Down\n",
167 		       adapter->netdev->name);
168 		if (netif_carrier_ok(adapter->netdev))
169 			netif_carrier_off(adapter->netdev);
170 
171 		if (affectTxQueue) {
172 			for (i = 0; i < adapter->num_tx_queues; i++)
173 				vmxnet3_tq_stop(&adapter->tx_queue[i], adapter);
174 		}
175 	}
176 }
177 
178 static void
179 vmxnet3_process_events(struct vmxnet3_adapter *adapter)
180 {
181 	int i;
182 	unsigned long flags;
183 	u32 events = le32_to_cpu(adapter->shared->ecr);
184 	if (!events)
185 		return;
186 
187 	vmxnet3_ack_events(adapter, events);
188 
189 	/* Check if link state has changed */
190 	if (events & VMXNET3_ECR_LINK)
191 		vmxnet3_check_link(adapter, true);
192 
193 	/* Check if there is an error on xmit/recv queues */
194 	if (events & (VMXNET3_ECR_TQERR | VMXNET3_ECR_RQERR)) {
195 		spin_lock_irqsave(&adapter->cmd_lock, flags);
196 		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
197 				       VMXNET3_CMD_GET_QUEUE_STATUS);
198 		spin_unlock_irqrestore(&adapter->cmd_lock, flags);
199 
200 		for (i = 0; i < adapter->num_tx_queues; i++)
201 			if (adapter->tqd_start[i].status.stopped)
202 				dev_err(&adapter->netdev->dev,
203 					"%s: tq[%d] error 0x%x\n",
204 					adapter->netdev->name, i, le32_to_cpu(
205 					adapter->tqd_start[i].status.error));
206 		for (i = 0; i < adapter->num_rx_queues; i++)
207 			if (adapter->rqd_start[i].status.stopped)
208 				dev_err(&adapter->netdev->dev,
209 					"%s: rq[%d] error 0x%x\n",
210 					adapter->netdev->name, i,
211 					adapter->rqd_start[i].status.error);
212 
213 		schedule_work(&adapter->work);
214 	}
215 }
216 
217 #ifdef __BIG_ENDIAN_BITFIELD
218 /*
219  * The device expects the bitfields in shared structures to be written in
220  * little endian. When CPU is big endian, the following routines are used to
221  * correctly read and write into ABI.
222  * The general technique used here is : double word bitfields are defined in
223  * opposite order for big endian architecture. Then before reading them in
224  * driver the complete double word is translated using le32_to_cpu. Similarly
225  * After the driver writes into bitfields, cpu_to_le32 is used to translate the
226  * double words into required format.
227  * In order to avoid touching bits in shared structure more than once, temporary
228  * descriptors are used. These are passed as srcDesc to following functions.
229  */
230 static void vmxnet3_RxDescToCPU(const struct Vmxnet3_RxDesc *srcDesc,
231 				struct Vmxnet3_RxDesc *dstDesc)
232 {
233 	u32 *src = (u32 *)srcDesc + 2;
234 	u32 *dst = (u32 *)dstDesc + 2;
235 	dstDesc->addr = le64_to_cpu(srcDesc->addr);
236 	*dst = le32_to_cpu(*src);
237 	dstDesc->ext1 = le32_to_cpu(srcDesc->ext1);
238 }
239 
240 static void vmxnet3_TxDescToLe(const struct Vmxnet3_TxDesc *srcDesc,
241 			       struct Vmxnet3_TxDesc *dstDesc)
242 {
243 	int i;
244 	u32 *src = (u32 *)(srcDesc + 1);
245 	u32 *dst = (u32 *)(dstDesc + 1);
246 
247 	/* Working backwards so that the gen bit is set at the end. */
248 	for (i = 2; i > 0; i--) {
249 		src--;
250 		dst--;
251 		*dst = cpu_to_le32(*src);
252 	}
253 }
254 
255 
256 static void vmxnet3_RxCompToCPU(const struct Vmxnet3_RxCompDesc *srcDesc,
257 				struct Vmxnet3_RxCompDesc *dstDesc)
258 {
259 	int i = 0;
260 	u32 *src = (u32 *)srcDesc;
261 	u32 *dst = (u32 *)dstDesc;
262 	for (i = 0; i < sizeof(struct Vmxnet3_RxCompDesc) / sizeof(u32); i++) {
263 		*dst = le32_to_cpu(*src);
264 		src++;
265 		dst++;
266 	}
267 }
268 
269 
270 /* Used to read bitfield values from double words. */
271 static u32 get_bitfield32(const __le32 *bitfield, u32 pos, u32 size)
272 {
273 	u32 temp = le32_to_cpu(*bitfield);
274 	u32 mask = ((1 << size) - 1) << pos;
275 	temp &= mask;
276 	temp >>= pos;
277 	return temp;
278 }
279 
280 
281 
282 #endif  /* __BIG_ENDIAN_BITFIELD */
283 
284 #ifdef __BIG_ENDIAN_BITFIELD
285 
286 #   define VMXNET3_TXDESC_GET_GEN(txdesc) get_bitfield32(((const __le32 *) \
287 			txdesc) + VMXNET3_TXD_GEN_DWORD_SHIFT, \
288 			VMXNET3_TXD_GEN_SHIFT, VMXNET3_TXD_GEN_SIZE)
289 #   define VMXNET3_TXDESC_GET_EOP(txdesc) get_bitfield32(((const __le32 *) \
290 			txdesc) + VMXNET3_TXD_EOP_DWORD_SHIFT, \
291 			VMXNET3_TXD_EOP_SHIFT, VMXNET3_TXD_EOP_SIZE)
292 #   define VMXNET3_TCD_GET_GEN(tcd) get_bitfield32(((const __le32 *)tcd) + \
293 			VMXNET3_TCD_GEN_DWORD_SHIFT, VMXNET3_TCD_GEN_SHIFT, \
294 			VMXNET3_TCD_GEN_SIZE)
295 #   define VMXNET3_TCD_GET_TXIDX(tcd) get_bitfield32((const __le32 *)tcd, \
296 			VMXNET3_TCD_TXIDX_SHIFT, VMXNET3_TCD_TXIDX_SIZE)
297 #   define vmxnet3_getRxComp(dstrcd, rcd, tmp) do { \
298 			(dstrcd) = (tmp); \
299 			vmxnet3_RxCompToCPU((rcd), (tmp)); \
300 		} while (0)
301 #   define vmxnet3_getRxDesc(dstrxd, rxd, tmp) do { \
302 			(dstrxd) = (tmp); \
303 			vmxnet3_RxDescToCPU((rxd), (tmp)); \
304 		} while (0)
305 
306 #else
307 
308 #   define VMXNET3_TXDESC_GET_GEN(txdesc) ((txdesc)->gen)
309 #   define VMXNET3_TXDESC_GET_EOP(txdesc) ((txdesc)->eop)
310 #   define VMXNET3_TCD_GET_GEN(tcd) ((tcd)->gen)
311 #   define VMXNET3_TCD_GET_TXIDX(tcd) ((tcd)->txdIdx)
312 #   define vmxnet3_getRxComp(dstrcd, rcd, tmp) (dstrcd) = (rcd)
313 #   define vmxnet3_getRxDesc(dstrxd, rxd, tmp) (dstrxd) = (rxd)
314 
315 #endif /* __BIG_ENDIAN_BITFIELD  */
316 
317 
318 static void
319 vmxnet3_unmap_tx_buf(struct vmxnet3_tx_buf_info *tbi,
320 		     struct pci_dev *pdev)
321 {
322 	if (tbi->map_type == VMXNET3_MAP_SINGLE)
323 		pci_unmap_single(pdev, tbi->dma_addr, tbi->len,
324 				 PCI_DMA_TODEVICE);
325 	else if (tbi->map_type == VMXNET3_MAP_PAGE)
326 		pci_unmap_page(pdev, tbi->dma_addr, tbi->len,
327 			       PCI_DMA_TODEVICE);
328 	else
329 		BUG_ON(tbi->map_type != VMXNET3_MAP_NONE);
330 
331 	tbi->map_type = VMXNET3_MAP_NONE; /* to help debugging */
332 }
333 
334 
335 static int
336 vmxnet3_unmap_pkt(u32 eop_idx, struct vmxnet3_tx_queue *tq,
337 		  struct pci_dev *pdev,	struct vmxnet3_adapter *adapter)
338 {
339 	struct sk_buff *skb;
340 	int entries = 0;
341 
342 	/* no out of order completion */
343 	BUG_ON(tq->buf_info[eop_idx].sop_idx != tq->tx_ring.next2comp);
344 	BUG_ON(VMXNET3_TXDESC_GET_EOP(&(tq->tx_ring.base[eop_idx].txd)) != 1);
345 
346 	skb = tq->buf_info[eop_idx].skb;
347 	BUG_ON(skb == NULL);
348 	tq->buf_info[eop_idx].skb = NULL;
349 
350 	VMXNET3_INC_RING_IDX_ONLY(eop_idx, tq->tx_ring.size);
351 
352 	while (tq->tx_ring.next2comp != eop_idx) {
353 		vmxnet3_unmap_tx_buf(tq->buf_info + tq->tx_ring.next2comp,
354 				     pdev);
355 
356 		/* update next2comp w/o tx_lock. Since we are marking more,
357 		 * instead of less, tx ring entries avail, the worst case is
358 		 * that the tx routine incorrectly re-queues a pkt due to
359 		 * insufficient tx ring entries.
360 		 */
361 		vmxnet3_cmd_ring_adv_next2comp(&tq->tx_ring);
362 		entries++;
363 	}
364 
365 	dev_kfree_skb_any(skb);
366 	return entries;
367 }
368 
369 
370 static int
371 vmxnet3_tq_tx_complete(struct vmxnet3_tx_queue *tq,
372 			struct vmxnet3_adapter *adapter)
373 {
374 	int completed = 0;
375 	union Vmxnet3_GenericDesc *gdesc;
376 
377 	gdesc = tq->comp_ring.base + tq->comp_ring.next2proc;
378 	while (VMXNET3_TCD_GET_GEN(&gdesc->tcd) == tq->comp_ring.gen) {
379 		completed += vmxnet3_unmap_pkt(VMXNET3_TCD_GET_TXIDX(
380 					       &gdesc->tcd), tq, adapter->pdev,
381 					       adapter);
382 
383 		vmxnet3_comp_ring_adv_next2proc(&tq->comp_ring);
384 		gdesc = tq->comp_ring.base + tq->comp_ring.next2proc;
385 	}
386 
387 	if (completed) {
388 		spin_lock(&tq->tx_lock);
389 		if (unlikely(vmxnet3_tq_stopped(tq, adapter) &&
390 			     vmxnet3_cmd_ring_desc_avail(&tq->tx_ring) >
391 			     VMXNET3_WAKE_QUEUE_THRESHOLD(tq) &&
392 			     netif_carrier_ok(adapter->netdev))) {
393 			vmxnet3_tq_wake(tq, adapter);
394 		}
395 		spin_unlock(&tq->tx_lock);
396 	}
397 	return completed;
398 }
399 
400 
401 static void
402 vmxnet3_tq_cleanup(struct vmxnet3_tx_queue *tq,
403 		   struct vmxnet3_adapter *adapter)
404 {
405 	int i;
406 
407 	while (tq->tx_ring.next2comp != tq->tx_ring.next2fill) {
408 		struct vmxnet3_tx_buf_info *tbi;
409 
410 		tbi = tq->buf_info + tq->tx_ring.next2comp;
411 
412 		vmxnet3_unmap_tx_buf(tbi, adapter->pdev);
413 		if (tbi->skb) {
414 			dev_kfree_skb_any(tbi->skb);
415 			tbi->skb = NULL;
416 		}
417 		vmxnet3_cmd_ring_adv_next2comp(&tq->tx_ring);
418 	}
419 
420 	/* sanity check, verify all buffers are indeed unmapped and freed */
421 	for (i = 0; i < tq->tx_ring.size; i++) {
422 		BUG_ON(tq->buf_info[i].skb != NULL ||
423 		       tq->buf_info[i].map_type != VMXNET3_MAP_NONE);
424 	}
425 
426 	tq->tx_ring.gen = VMXNET3_INIT_GEN;
427 	tq->tx_ring.next2fill = tq->tx_ring.next2comp = 0;
428 
429 	tq->comp_ring.gen = VMXNET3_INIT_GEN;
430 	tq->comp_ring.next2proc = 0;
431 }
432 
433 
434 static void
435 vmxnet3_tq_destroy(struct vmxnet3_tx_queue *tq,
436 		   struct vmxnet3_adapter *adapter)
437 {
438 	if (tq->tx_ring.base) {
439 		pci_free_consistent(adapter->pdev, tq->tx_ring.size *
440 				    sizeof(struct Vmxnet3_TxDesc),
441 				    tq->tx_ring.base, tq->tx_ring.basePA);
442 		tq->tx_ring.base = NULL;
443 	}
444 	if (tq->data_ring.base) {
445 		pci_free_consistent(adapter->pdev, tq->data_ring.size *
446 				    sizeof(struct Vmxnet3_TxDataDesc),
447 				    tq->data_ring.base, tq->data_ring.basePA);
448 		tq->data_ring.base = NULL;
449 	}
450 	if (tq->comp_ring.base) {
451 		pci_free_consistent(adapter->pdev, tq->comp_ring.size *
452 				    sizeof(struct Vmxnet3_TxCompDesc),
453 				    tq->comp_ring.base, tq->comp_ring.basePA);
454 		tq->comp_ring.base = NULL;
455 	}
456 	kfree(tq->buf_info);
457 	tq->buf_info = NULL;
458 }
459 
460 
461 /* Destroy all tx queues */
462 void
463 vmxnet3_tq_destroy_all(struct vmxnet3_adapter *adapter)
464 {
465 	int i;
466 
467 	for (i = 0; i < adapter->num_tx_queues; i++)
468 		vmxnet3_tq_destroy(&adapter->tx_queue[i], adapter);
469 }
470 
471 
472 static void
473 vmxnet3_tq_init(struct vmxnet3_tx_queue *tq,
474 		struct vmxnet3_adapter *adapter)
475 {
476 	int i;
477 
478 	/* reset the tx ring contents to 0 and reset the tx ring states */
479 	memset(tq->tx_ring.base, 0, tq->tx_ring.size *
480 	       sizeof(struct Vmxnet3_TxDesc));
481 	tq->tx_ring.next2fill = tq->tx_ring.next2comp = 0;
482 	tq->tx_ring.gen = VMXNET3_INIT_GEN;
483 
484 	memset(tq->data_ring.base, 0, tq->data_ring.size *
485 	       sizeof(struct Vmxnet3_TxDataDesc));
486 
487 	/* reset the tx comp ring contents to 0 and reset comp ring states */
488 	memset(tq->comp_ring.base, 0, tq->comp_ring.size *
489 	       sizeof(struct Vmxnet3_TxCompDesc));
490 	tq->comp_ring.next2proc = 0;
491 	tq->comp_ring.gen = VMXNET3_INIT_GEN;
492 
493 	/* reset the bookkeeping data */
494 	memset(tq->buf_info, 0, sizeof(tq->buf_info[0]) * tq->tx_ring.size);
495 	for (i = 0; i < tq->tx_ring.size; i++)
496 		tq->buf_info[i].map_type = VMXNET3_MAP_NONE;
497 
498 	/* stats are not reset */
499 }
500 
501 
502 static int
503 vmxnet3_tq_create(struct vmxnet3_tx_queue *tq,
504 		  struct vmxnet3_adapter *adapter)
505 {
506 	BUG_ON(tq->tx_ring.base || tq->data_ring.base ||
507 	       tq->comp_ring.base || tq->buf_info);
508 
509 	tq->tx_ring.base = pci_alloc_consistent(adapter->pdev, tq->tx_ring.size
510 			   * sizeof(struct Vmxnet3_TxDesc),
511 			   &tq->tx_ring.basePA);
512 	if (!tq->tx_ring.base) {
513 		printk(KERN_ERR "%s: failed to allocate tx ring\n",
514 		       adapter->netdev->name);
515 		goto err;
516 	}
517 
518 	tq->data_ring.base = pci_alloc_consistent(adapter->pdev,
519 			     tq->data_ring.size *
520 			     sizeof(struct Vmxnet3_TxDataDesc),
521 			     &tq->data_ring.basePA);
522 	if (!tq->data_ring.base) {
523 		printk(KERN_ERR "%s: failed to allocate data ring\n",
524 		       adapter->netdev->name);
525 		goto err;
526 	}
527 
528 	tq->comp_ring.base = pci_alloc_consistent(adapter->pdev,
529 			     tq->comp_ring.size *
530 			     sizeof(struct Vmxnet3_TxCompDesc),
531 			     &tq->comp_ring.basePA);
532 	if (!tq->comp_ring.base) {
533 		printk(KERN_ERR "%s: failed to allocate tx comp ring\n",
534 		       adapter->netdev->name);
535 		goto err;
536 	}
537 
538 	tq->buf_info = kcalloc(tq->tx_ring.size, sizeof(tq->buf_info[0]),
539 			       GFP_KERNEL);
540 	if (!tq->buf_info)
541 		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 = dev_alloc_skb(rbi->len +
584 							 NET_IP_ALIGN);
585 				if (unlikely(rbi->skb == NULL)) {
586 					rq->stats.rx_buf_alloc_failure++;
587 					break;
588 				}
589 				rbi->skb->dev = adapter->netdev;
590 
591 				skb_reserve(rbi->skb, NET_IP_ALIGN);
592 				rbi->dma_addr = pci_map_single(adapter->pdev,
593 						rbi->skb->data, rbi->len,
594 						PCI_DMA_FROMDEVICE);
595 			} else {
596 				/* rx buffer skipped by the device */
597 			}
598 			val = VMXNET3_RXD_BTYPE_HEAD << VMXNET3_RXD_BTYPE_SHIFT;
599 		} else {
600 			BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_PAGE ||
601 			       rbi->len  != PAGE_SIZE);
602 
603 			if (rbi->page == NULL) {
604 				rbi->page = alloc_page(GFP_ATOMIC);
605 				if (unlikely(rbi->page == NULL)) {
606 					rq->stats.rx_buf_alloc_failure++;
607 					break;
608 				}
609 				rbi->dma_addr = pci_map_page(adapter->pdev,
610 						rbi->page, 0, PAGE_SIZE,
611 						PCI_DMA_FROMDEVICE);
612 			} else {
613 				/* rx buffers skipped by the device */
614 			}
615 			val = VMXNET3_RXD_BTYPE_BODY << VMXNET3_RXD_BTYPE_SHIFT;
616 		}
617 
618 		BUG_ON(rbi->dma_addr == 0);
619 		gd->rxd.addr = cpu_to_le64(rbi->dma_addr);
620 		gd->dword[2] = cpu_to_le32((!ring->gen << VMXNET3_RXD_GEN_SHIFT)
621 					   | val | rbi->len);
622 
623 		/* Fill the last buffer but dont mark it ready, or else the
624 		 * device will think that the queue is full */
625 		if (num_allocated == num_to_alloc)
626 			break;
627 
628 		gd->dword[2] |= cpu_to_le32(ring->gen << VMXNET3_RXD_GEN_SHIFT);
629 		num_allocated++;
630 		vmxnet3_cmd_ring_adv_next2fill(ring);
631 	}
632 	rq->uncommitted[ring_idx] += num_allocated;
633 
634 	dev_dbg(&adapter->netdev->dev,
635 		"alloc_rx_buf: %d allocated, next2fill %u, next2comp "
636 		"%u, uncommitted %u\n", num_allocated, ring->next2fill,
637 		ring->next2comp, rq->uncommitted[ring_idx]);
638 
639 	/* so that the device can distinguish a full ring and an empty ring */
640 	BUG_ON(num_allocated != 0 && ring->next2fill == ring->next2comp);
641 
642 	return num_allocated;
643 }
644 
645 
646 static void
647 vmxnet3_append_frag(struct sk_buff *skb, struct Vmxnet3_RxCompDesc *rcd,
648 		    struct vmxnet3_rx_buf_info *rbi)
649 {
650 	struct skb_frag_struct *frag = skb_shinfo(skb)->frags +
651 		skb_shinfo(skb)->nr_frags;
652 
653 	BUG_ON(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS);
654 
655 	__skb_frag_set_page(frag, rbi->page);
656 	frag->page_offset = 0;
657 	skb_frag_size_set(frag, rcd->len);
658 	skb->data_len += rcd->len;
659 	skb->truesize += PAGE_SIZE;
660 	skb_shinfo(skb)->nr_frags++;
661 }
662 
663 
664 static void
665 vmxnet3_map_pkt(struct sk_buff *skb, struct vmxnet3_tx_ctx *ctx,
666 		struct vmxnet3_tx_queue *tq, struct pci_dev *pdev,
667 		struct vmxnet3_adapter *adapter)
668 {
669 	u32 dw2, len;
670 	unsigned long buf_offset;
671 	int i;
672 	union Vmxnet3_GenericDesc *gdesc;
673 	struct vmxnet3_tx_buf_info *tbi = NULL;
674 
675 	BUG_ON(ctx->copy_size > skb_headlen(skb));
676 
677 	/* use the previous gen bit for the SOP desc */
678 	dw2 = (tq->tx_ring.gen ^ 0x1) << VMXNET3_TXD_GEN_SHIFT;
679 
680 	ctx->sop_txd = tq->tx_ring.base + tq->tx_ring.next2fill;
681 	gdesc = ctx->sop_txd; /* both loops below can be skipped */
682 
683 	/* no need to map the buffer if headers are copied */
684 	if (ctx->copy_size) {
685 		ctx->sop_txd->txd.addr = cpu_to_le64(tq->data_ring.basePA +
686 					tq->tx_ring.next2fill *
687 					sizeof(struct Vmxnet3_TxDataDesc));
688 		ctx->sop_txd->dword[2] = cpu_to_le32(dw2 | ctx->copy_size);
689 		ctx->sop_txd->dword[3] = 0;
690 
691 		tbi = tq->buf_info + tq->tx_ring.next2fill;
692 		tbi->map_type = VMXNET3_MAP_NONE;
693 
694 		dev_dbg(&adapter->netdev->dev,
695 			"txd[%u]: 0x%Lx 0x%x 0x%x\n",
696 			tq->tx_ring.next2fill,
697 			le64_to_cpu(ctx->sop_txd->txd.addr),
698 			ctx->sop_txd->dword[2], ctx->sop_txd->dword[3]);
699 		vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
700 
701 		/* use the right gen for non-SOP desc */
702 		dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
703 	}
704 
705 	/* linear part can use multiple tx desc if it's big */
706 	len = skb_headlen(skb) - ctx->copy_size;
707 	buf_offset = ctx->copy_size;
708 	while (len) {
709 		u32 buf_size;
710 
711 		if (len < VMXNET3_MAX_TX_BUF_SIZE) {
712 			buf_size = len;
713 			dw2 |= len;
714 		} else {
715 			buf_size = VMXNET3_MAX_TX_BUF_SIZE;
716 			/* spec says that for TxDesc.len, 0 == 2^14 */
717 		}
718 
719 		tbi = tq->buf_info + tq->tx_ring.next2fill;
720 		tbi->map_type = VMXNET3_MAP_SINGLE;
721 		tbi->dma_addr = pci_map_single(adapter->pdev,
722 				skb->data + buf_offset, buf_size,
723 				PCI_DMA_TODEVICE);
724 
725 		tbi->len = buf_size;
726 
727 		gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
728 		BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
729 
730 		gdesc->txd.addr = cpu_to_le64(tbi->dma_addr);
731 		gdesc->dword[2] = cpu_to_le32(dw2);
732 		gdesc->dword[3] = 0;
733 
734 		dev_dbg(&adapter->netdev->dev,
735 			"txd[%u]: 0x%Lx 0x%x 0x%x\n",
736 			tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr),
737 			le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]);
738 		vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
739 		dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
740 
741 		len -= buf_size;
742 		buf_offset += buf_size;
743 	}
744 
745 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
746 		const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
747 
748 		tbi = tq->buf_info + tq->tx_ring.next2fill;
749 		tbi->map_type = VMXNET3_MAP_PAGE;
750 		tbi->dma_addr = skb_frag_dma_map(&adapter->pdev->dev, frag,
751 						 0, skb_frag_size(frag),
752 						 DMA_TO_DEVICE);
753 
754 		tbi->len = skb_frag_size(frag);
755 
756 		gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
757 		BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
758 
759 		gdesc->txd.addr = cpu_to_le64(tbi->dma_addr);
760 		gdesc->dword[2] = cpu_to_le32(dw2 | skb_frag_size(frag));
761 		gdesc->dword[3] = 0;
762 
763 		dev_dbg(&adapter->netdev->dev,
764 			"txd[%u]: 0x%llu %u %u\n",
765 			tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr),
766 			le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]);
767 		vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
768 		dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
769 	}
770 
771 	ctx->eop_txd = gdesc;
772 
773 	/* set the last buf_info for the pkt */
774 	tbi->skb = skb;
775 	tbi->sop_idx = ctx->sop_txd - tq->tx_ring.base;
776 }
777 
778 
779 /* Init all tx queues */
780 static void
781 vmxnet3_tq_init_all(struct vmxnet3_adapter *adapter)
782 {
783 	int i;
784 
785 	for (i = 0; i < adapter->num_tx_queues; i++)
786 		vmxnet3_tq_init(&adapter->tx_queue[i], adapter);
787 }
788 
789 
790 /*
791  *    parse and copy relevant protocol headers:
792  *      For a tso pkt, relevant headers are L2/3/4 including options
793  *      For a pkt requesting csum offloading, they are L2/3 and may include L4
794  *      if it's a TCP/UDP pkt
795  *
796  * Returns:
797  *    -1:  error happens during parsing
798  *     0:  protocol headers parsed, but too big to be copied
799  *     1:  protocol headers parsed and copied
800  *
801  * Other effects:
802  *    1. related *ctx fields are updated.
803  *    2. ctx->copy_size is # of bytes copied
804  *    3. the portion copied is guaranteed to be in the linear part
805  *
806  */
807 static int
808 vmxnet3_parse_and_copy_hdr(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
809 			   struct vmxnet3_tx_ctx *ctx,
810 			   struct vmxnet3_adapter *adapter)
811 {
812 	struct Vmxnet3_TxDataDesc *tdd;
813 
814 	if (ctx->mss) {	/* TSO */
815 		ctx->eth_ip_hdr_size = skb_transport_offset(skb);
816 		ctx->l4_hdr_size = tcp_hdrlen(skb);
817 		ctx->copy_size = ctx->eth_ip_hdr_size + ctx->l4_hdr_size;
818 	} else {
819 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
820 			ctx->eth_ip_hdr_size = skb_checksum_start_offset(skb);
821 
822 			if (ctx->ipv4) {
823 				const struct iphdr *iph = ip_hdr(skb);
824 
825 				if (iph->protocol == IPPROTO_TCP)
826 					ctx->l4_hdr_size = tcp_hdrlen(skb);
827 				else if (iph->protocol == IPPROTO_UDP)
828 					ctx->l4_hdr_size = sizeof(struct udphdr);
829 				else
830 					ctx->l4_hdr_size = 0;
831 			} else {
832 				/* for simplicity, don't copy L4 headers */
833 				ctx->l4_hdr_size = 0;
834 			}
835 			ctx->copy_size = min(ctx->eth_ip_hdr_size +
836 					 ctx->l4_hdr_size, skb->len);
837 		} else {
838 			ctx->eth_ip_hdr_size = 0;
839 			ctx->l4_hdr_size = 0;
840 			/* copy as much as allowed */
841 			ctx->copy_size = min((unsigned int)VMXNET3_HDR_COPY_SIZE
842 					     , skb_headlen(skb));
843 		}
844 
845 		/* make sure headers are accessible directly */
846 		if (unlikely(!pskb_may_pull(skb, ctx->copy_size)))
847 			goto err;
848 	}
849 
850 	if (unlikely(ctx->copy_size > VMXNET3_HDR_COPY_SIZE)) {
851 		tq->stats.oversized_hdr++;
852 		ctx->copy_size = 0;
853 		return 0;
854 	}
855 
856 	tdd = tq->data_ring.base + tq->tx_ring.next2fill;
857 
858 	memcpy(tdd->data, skb->data, ctx->copy_size);
859 	dev_dbg(&adapter->netdev->dev,
860 		"copy %u bytes to dataRing[%u]\n",
861 		ctx->copy_size, tq->tx_ring.next2fill);
862 	return 1;
863 
864 err:
865 	return -1;
866 }
867 
868 
869 static void
870 vmxnet3_prepare_tso(struct sk_buff *skb,
871 		    struct vmxnet3_tx_ctx *ctx)
872 {
873 	struct tcphdr *tcph = tcp_hdr(skb);
874 
875 	if (ctx->ipv4) {
876 		struct iphdr *iph = ip_hdr(skb);
877 
878 		iph->check = 0;
879 		tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, 0,
880 						 IPPROTO_TCP, 0);
881 	} else {
882 		struct ipv6hdr *iph = ipv6_hdr(skb);
883 
884 		tcph->check = ~csum_ipv6_magic(&iph->saddr, &iph->daddr, 0,
885 					       IPPROTO_TCP, 0);
886 	}
887 }
888 
889 
890 /*
891  * Transmits a pkt thru a given tq
892  * Returns:
893  *    NETDEV_TX_OK:      descriptors are setup successfully
894  *    NETDEV_TX_OK:      error occurred, the pkt is dropped
895  *    NETDEV_TX_BUSY:    tx ring is full, queue is stopped
896  *
897  * Side-effects:
898  *    1. tx ring may be changed
899  *    2. tq stats may be updated accordingly
900  *    3. shared->txNumDeferred may be updated
901  */
902 
903 static int
904 vmxnet3_tq_xmit(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
905 		struct vmxnet3_adapter *adapter, struct net_device *netdev)
906 {
907 	int ret;
908 	u32 count;
909 	unsigned long flags;
910 	struct vmxnet3_tx_ctx ctx;
911 	union Vmxnet3_GenericDesc *gdesc;
912 #ifdef __BIG_ENDIAN_BITFIELD
913 	/* Use temporary descriptor to avoid touching bits multiple times */
914 	union Vmxnet3_GenericDesc tempTxDesc;
915 #endif
916 
917 	/* conservatively estimate # of descriptors to use */
918 	count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) +
919 		skb_shinfo(skb)->nr_frags + 1;
920 
921 	ctx.ipv4 = (vlan_get_protocol(skb) == cpu_to_be16(ETH_P_IP));
922 
923 	ctx.mss = skb_shinfo(skb)->gso_size;
924 	if (ctx.mss) {
925 		if (skb_header_cloned(skb)) {
926 			if (unlikely(pskb_expand_head(skb, 0, 0,
927 						      GFP_ATOMIC) != 0)) {
928 				tq->stats.drop_tso++;
929 				goto drop_pkt;
930 			}
931 			tq->stats.copy_skb_header++;
932 		}
933 		vmxnet3_prepare_tso(skb, &ctx);
934 	} else {
935 		if (unlikely(count > VMXNET3_MAX_TXD_PER_PKT)) {
936 
937 			/* non-tso pkts must not use more than
938 			 * VMXNET3_MAX_TXD_PER_PKT entries
939 			 */
940 			if (skb_linearize(skb) != 0) {
941 				tq->stats.drop_too_many_frags++;
942 				goto drop_pkt;
943 			}
944 			tq->stats.linearized++;
945 
946 			/* recalculate the # of descriptors to use */
947 			count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) + 1;
948 		}
949 	}
950 
951 	spin_lock_irqsave(&tq->tx_lock, flags);
952 
953 	if (count > vmxnet3_cmd_ring_desc_avail(&tq->tx_ring)) {
954 		tq->stats.tx_ring_full++;
955 		dev_dbg(&adapter->netdev->dev,
956 			"tx queue stopped on %s, next2comp %u"
957 			" next2fill %u\n", adapter->netdev->name,
958 			tq->tx_ring.next2comp, tq->tx_ring.next2fill);
959 
960 		vmxnet3_tq_stop(tq, adapter);
961 		spin_unlock_irqrestore(&tq->tx_lock, flags);
962 		return NETDEV_TX_BUSY;
963 	}
964 
965 
966 	ret = vmxnet3_parse_and_copy_hdr(skb, tq, &ctx, adapter);
967 	if (ret >= 0) {
968 		BUG_ON(ret <= 0 && ctx.copy_size != 0);
969 		/* hdrs parsed, check against other limits */
970 		if (ctx.mss) {
971 			if (unlikely(ctx.eth_ip_hdr_size + ctx.l4_hdr_size >
972 				     VMXNET3_MAX_TX_BUF_SIZE)) {
973 				goto hdr_too_big;
974 			}
975 		} else {
976 			if (skb->ip_summed == CHECKSUM_PARTIAL) {
977 				if (unlikely(ctx.eth_ip_hdr_size +
978 					     skb->csum_offset >
979 					     VMXNET3_MAX_CSUM_OFFSET)) {
980 					goto hdr_too_big;
981 				}
982 			}
983 		}
984 	} else {
985 		tq->stats.drop_hdr_inspect_err++;
986 		goto unlock_drop_pkt;
987 	}
988 
989 	/* fill tx descs related to addr & len */
990 	vmxnet3_map_pkt(skb, &ctx, tq, adapter->pdev, adapter);
991 
992 	/* setup the EOP desc */
993 	ctx.eop_txd->dword[3] = cpu_to_le32(VMXNET3_TXD_CQ | VMXNET3_TXD_EOP);
994 
995 	/* setup the SOP desc */
996 #ifdef __BIG_ENDIAN_BITFIELD
997 	gdesc = &tempTxDesc;
998 	gdesc->dword[2] = ctx.sop_txd->dword[2];
999 	gdesc->dword[3] = ctx.sop_txd->dword[3];
1000 #else
1001 	gdesc = ctx.sop_txd;
1002 #endif
1003 	if (ctx.mss) {
1004 		gdesc->txd.hlen = ctx.eth_ip_hdr_size + ctx.l4_hdr_size;
1005 		gdesc->txd.om = VMXNET3_OM_TSO;
1006 		gdesc->txd.msscof = ctx.mss;
1007 		le32_add_cpu(&tq->shared->txNumDeferred, (skb->len -
1008 			     gdesc->txd.hlen + ctx.mss - 1) / ctx.mss);
1009 	} else {
1010 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
1011 			gdesc->txd.hlen = ctx.eth_ip_hdr_size;
1012 			gdesc->txd.om = VMXNET3_OM_CSUM;
1013 			gdesc->txd.msscof = ctx.eth_ip_hdr_size +
1014 					    skb->csum_offset;
1015 		} else {
1016 			gdesc->txd.om = 0;
1017 			gdesc->txd.msscof = 0;
1018 		}
1019 		le32_add_cpu(&tq->shared->txNumDeferred, 1);
1020 	}
1021 
1022 	if (vlan_tx_tag_present(skb)) {
1023 		gdesc->txd.ti = 1;
1024 		gdesc->txd.tci = vlan_tx_tag_get(skb);
1025 	}
1026 
1027 	/* finally flips the GEN bit of the SOP desc. */
1028 	gdesc->dword[2] = cpu_to_le32(le32_to_cpu(gdesc->dword[2]) ^
1029 						  VMXNET3_TXD_GEN);
1030 #ifdef __BIG_ENDIAN_BITFIELD
1031 	/* Finished updating in bitfields of Tx Desc, so write them in original
1032 	 * place.
1033 	 */
1034 	vmxnet3_TxDescToLe((struct Vmxnet3_TxDesc *)gdesc,
1035 			   (struct Vmxnet3_TxDesc *)ctx.sop_txd);
1036 	gdesc = ctx.sop_txd;
1037 #endif
1038 	dev_dbg(&adapter->netdev->dev,
1039 		"txd[%u]: SOP 0x%Lx 0x%x 0x%x\n",
1040 		(u32)(ctx.sop_txd -
1041 		tq->tx_ring.base), le64_to_cpu(gdesc->txd.addr),
1042 		le32_to_cpu(gdesc->dword[2]), le32_to_cpu(gdesc->dword[3]));
1043 
1044 	spin_unlock_irqrestore(&tq->tx_lock, flags);
1045 
1046 	if (le32_to_cpu(tq->shared->txNumDeferred) >=
1047 					le32_to_cpu(tq->shared->txThreshold)) {
1048 		tq->shared->txNumDeferred = 0;
1049 		VMXNET3_WRITE_BAR0_REG(adapter,
1050 				       VMXNET3_REG_TXPROD + tq->qid * 8,
1051 				       tq->tx_ring.next2fill);
1052 	}
1053 
1054 	return NETDEV_TX_OK;
1055 
1056 hdr_too_big:
1057 	tq->stats.drop_oversized_hdr++;
1058 unlock_drop_pkt:
1059 	spin_unlock_irqrestore(&tq->tx_lock, flags);
1060 drop_pkt:
1061 	tq->stats.drop_total++;
1062 	dev_kfree_skb(skb);
1063 	return NETDEV_TX_OK;
1064 }
1065 
1066 
1067 static netdev_tx_t
1068 vmxnet3_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
1069 {
1070 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1071 
1072 		BUG_ON(skb->queue_mapping > adapter->num_tx_queues);
1073 		return vmxnet3_tq_xmit(skb,
1074 				       &adapter->tx_queue[skb->queue_mapping],
1075 				       adapter, netdev);
1076 }
1077 
1078 
1079 static void
1080 vmxnet3_rx_csum(struct vmxnet3_adapter *adapter,
1081 		struct sk_buff *skb,
1082 		union Vmxnet3_GenericDesc *gdesc)
1083 {
1084 	if (!gdesc->rcd.cnc && adapter->netdev->features & NETIF_F_RXCSUM) {
1085 		/* typical case: TCP/UDP over IP and both csums are correct */
1086 		if ((le32_to_cpu(gdesc->dword[3]) & VMXNET3_RCD_CSUM_OK) ==
1087 							VMXNET3_RCD_CSUM_OK) {
1088 			skb->ip_summed = CHECKSUM_UNNECESSARY;
1089 			BUG_ON(!(gdesc->rcd.tcp || gdesc->rcd.udp));
1090 			BUG_ON(!(gdesc->rcd.v4  || gdesc->rcd.v6));
1091 			BUG_ON(gdesc->rcd.frg);
1092 		} else {
1093 			if (gdesc->rcd.csum) {
1094 				skb->csum = htons(gdesc->rcd.csum);
1095 				skb->ip_summed = CHECKSUM_PARTIAL;
1096 			} else {
1097 				skb_checksum_none_assert(skb);
1098 			}
1099 		}
1100 	} else {
1101 		skb_checksum_none_assert(skb);
1102 	}
1103 }
1104 
1105 
1106 static void
1107 vmxnet3_rx_error(struct vmxnet3_rx_queue *rq, struct Vmxnet3_RxCompDesc *rcd,
1108 		 struct vmxnet3_rx_ctx *ctx,  struct vmxnet3_adapter *adapter)
1109 {
1110 	rq->stats.drop_err++;
1111 	if (!rcd->fcs)
1112 		rq->stats.drop_fcs++;
1113 
1114 	rq->stats.drop_total++;
1115 
1116 	/*
1117 	 * We do not unmap and chain the rx buffer to the skb.
1118 	 * We basically pretend this buffer is not used and will be recycled
1119 	 * by vmxnet3_rq_alloc_rx_buf()
1120 	 */
1121 
1122 	/*
1123 	 * ctx->skb may be NULL if this is the first and the only one
1124 	 * desc for the pkt
1125 	 */
1126 	if (ctx->skb)
1127 		dev_kfree_skb_irq(ctx->skb);
1128 
1129 	ctx->skb = NULL;
1130 }
1131 
1132 
1133 static int
1134 vmxnet3_rq_rx_complete(struct vmxnet3_rx_queue *rq,
1135 		       struct vmxnet3_adapter *adapter, int quota)
1136 {
1137 	static const u32 rxprod_reg[2] = {
1138 		VMXNET3_REG_RXPROD, VMXNET3_REG_RXPROD2
1139 	};
1140 	u32 num_rxd = 0;
1141 	bool skip_page_frags = false;
1142 	struct Vmxnet3_RxCompDesc *rcd;
1143 	struct vmxnet3_rx_ctx *ctx = &rq->rx_ctx;
1144 #ifdef __BIG_ENDIAN_BITFIELD
1145 	struct Vmxnet3_RxDesc rxCmdDesc;
1146 	struct Vmxnet3_RxCompDesc rxComp;
1147 #endif
1148 	vmxnet3_getRxComp(rcd, &rq->comp_ring.base[rq->comp_ring.next2proc].rcd,
1149 			  &rxComp);
1150 	while (rcd->gen == rq->comp_ring.gen) {
1151 		struct vmxnet3_rx_buf_info *rbi;
1152 		struct sk_buff *skb, *new_skb = NULL;
1153 		struct page *new_page = NULL;
1154 		int num_to_alloc;
1155 		struct Vmxnet3_RxDesc *rxd;
1156 		u32 idx, ring_idx;
1157 		struct vmxnet3_cmd_ring	*ring = NULL;
1158 		if (num_rxd >= quota) {
1159 			/* we may stop even before we see the EOP desc of
1160 			 * the current pkt
1161 			 */
1162 			break;
1163 		}
1164 		num_rxd++;
1165 		BUG_ON(rcd->rqID != rq->qid && rcd->rqID != rq->qid2);
1166 		idx = rcd->rxdIdx;
1167 		ring_idx = rcd->rqID < adapter->num_rx_queues ? 0 : 1;
1168 		ring = rq->rx_ring + ring_idx;
1169 		vmxnet3_getRxDesc(rxd, &rq->rx_ring[ring_idx].base[idx].rxd,
1170 				  &rxCmdDesc);
1171 		rbi = rq->buf_info[ring_idx] + idx;
1172 
1173 		BUG_ON(rxd->addr != rbi->dma_addr ||
1174 		       rxd->len != rbi->len);
1175 
1176 		if (unlikely(rcd->eop && rcd->err)) {
1177 			vmxnet3_rx_error(rq, rcd, ctx, adapter);
1178 			goto rcd_done;
1179 		}
1180 
1181 		if (rcd->sop) { /* first buf of the pkt */
1182 			BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_HEAD ||
1183 			       rcd->rqID != rq->qid);
1184 
1185 			BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_SKB);
1186 			BUG_ON(ctx->skb != NULL || rbi->skb == NULL);
1187 
1188 			if (unlikely(rcd->len == 0)) {
1189 				/* Pretend the rx buffer is skipped. */
1190 				BUG_ON(!(rcd->sop && rcd->eop));
1191 				dev_dbg(&adapter->netdev->dev,
1192 					"rxRing[%u][%u] 0 length\n",
1193 					ring_idx, idx);
1194 				goto rcd_done;
1195 			}
1196 
1197 			skip_page_frags = false;
1198 			ctx->skb = rbi->skb;
1199 			new_skb = dev_alloc_skb(rbi->len + NET_IP_ALIGN);
1200 			if (new_skb == NULL) {
1201 				/* Skb allocation failed, do not handover this
1202 				 * skb to stack. Reuse it. Drop the existing pkt
1203 				 */
1204 				rq->stats.rx_buf_alloc_failure++;
1205 				ctx->skb = NULL;
1206 				rq->stats.drop_total++;
1207 				skip_page_frags = true;
1208 				goto rcd_done;
1209 			}
1210 
1211 			pci_unmap_single(adapter->pdev, rbi->dma_addr, rbi->len,
1212 					 PCI_DMA_FROMDEVICE);
1213 
1214 			skb_put(ctx->skb, rcd->len);
1215 
1216 			/* Immediate refill */
1217 			new_skb->dev = adapter->netdev;
1218 			skb_reserve(new_skb, NET_IP_ALIGN);
1219 			rbi->skb = new_skb;
1220 			rbi->dma_addr = pci_map_single(adapter->pdev,
1221 					rbi->skb->data, rbi->len,
1222 					PCI_DMA_FROMDEVICE);
1223 			rxd->addr = cpu_to_le64(rbi->dma_addr);
1224 			rxd->len = rbi->len;
1225 
1226 		} else {
1227 			BUG_ON(ctx->skb == NULL && !skip_page_frags);
1228 
1229 			/* non SOP buffer must be type 1 in most cases */
1230 			BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_PAGE);
1231 			BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_BODY);
1232 
1233 			/* If an sop buffer was dropped, skip all
1234 			 * following non-sop fragments. They will be reused.
1235 			 */
1236 			if (skip_page_frags)
1237 				goto rcd_done;
1238 
1239 			new_page = alloc_page(GFP_ATOMIC);
1240 			if (unlikely(new_page == NULL)) {
1241 				/* Replacement page frag could not be allocated.
1242 				 * Reuse this page. Drop the pkt and free the
1243 				 * skb which contained this page as a frag. Skip
1244 				 * processing all the following non-sop frags.
1245 				 */
1246 				rq->stats.rx_buf_alloc_failure++;
1247 				dev_kfree_skb(ctx->skb);
1248 				ctx->skb = NULL;
1249 				skip_page_frags = true;
1250 				goto rcd_done;
1251 			}
1252 
1253 			if (rcd->len) {
1254 				pci_unmap_page(adapter->pdev,
1255 					       rbi->dma_addr, rbi->len,
1256 					       PCI_DMA_FROMDEVICE);
1257 
1258 				vmxnet3_append_frag(ctx->skb, rcd, rbi);
1259 			}
1260 
1261 			/* Immediate refill */
1262 			rbi->page = new_page;
1263 			rbi->dma_addr = pci_map_page(adapter->pdev, rbi->page,
1264 						     0, PAGE_SIZE,
1265 						     PCI_DMA_FROMDEVICE);
1266 			rxd->addr = cpu_to_le64(rbi->dma_addr);
1267 			rxd->len = rbi->len;
1268 		}
1269 
1270 
1271 		skb = ctx->skb;
1272 		if (rcd->eop) {
1273 			skb->len += skb->data_len;
1274 
1275 			vmxnet3_rx_csum(adapter, skb,
1276 					(union Vmxnet3_GenericDesc *)rcd);
1277 			skb->protocol = eth_type_trans(skb, adapter->netdev);
1278 
1279 			if (unlikely(rcd->ts))
1280 				__vlan_hwaccel_put_tag(skb, rcd->tci);
1281 
1282 			if (adapter->netdev->features & NETIF_F_LRO)
1283 				netif_receive_skb(skb);
1284 			else
1285 				napi_gro_receive(&rq->napi, skb);
1286 
1287 			ctx->skb = NULL;
1288 		}
1289 
1290 rcd_done:
1291 		/* device may have skipped some rx descs */
1292 		ring->next2comp = idx;
1293 		num_to_alloc = vmxnet3_cmd_ring_desc_avail(ring);
1294 		ring = rq->rx_ring + ring_idx;
1295 		while (num_to_alloc) {
1296 			vmxnet3_getRxDesc(rxd, &ring->base[ring->next2fill].rxd,
1297 					  &rxCmdDesc);
1298 			BUG_ON(!rxd->addr);
1299 
1300 			/* Recv desc is ready to be used by the device */
1301 			rxd->gen = ring->gen;
1302 			vmxnet3_cmd_ring_adv_next2fill(ring);
1303 			num_to_alloc--;
1304 		}
1305 
1306 		/* if needed, update the register */
1307 		if (unlikely(rq->shared->updateRxProd)) {
1308 			VMXNET3_WRITE_BAR0_REG(adapter,
1309 				rxprod_reg[ring_idx] + rq->qid * 8,
1310 				ring->next2fill);
1311 			rq->uncommitted[ring_idx] = 0;
1312 		}
1313 
1314 		vmxnet3_comp_ring_adv_next2proc(&rq->comp_ring);
1315 		vmxnet3_getRxComp(rcd,
1316 		     &rq->comp_ring.base[rq->comp_ring.next2proc].rcd, &rxComp);
1317 	}
1318 
1319 	return num_rxd;
1320 }
1321 
1322 
1323 static void
1324 vmxnet3_rq_cleanup(struct vmxnet3_rx_queue *rq,
1325 		   struct vmxnet3_adapter *adapter)
1326 {
1327 	u32 i, ring_idx;
1328 	struct Vmxnet3_RxDesc *rxd;
1329 
1330 	for (ring_idx = 0; ring_idx < 2; ring_idx++) {
1331 		for (i = 0; i < rq->rx_ring[ring_idx].size; i++) {
1332 #ifdef __BIG_ENDIAN_BITFIELD
1333 			struct Vmxnet3_RxDesc rxDesc;
1334 #endif
1335 			vmxnet3_getRxDesc(rxd,
1336 				&rq->rx_ring[ring_idx].base[i].rxd, &rxDesc);
1337 
1338 			if (rxd->btype == VMXNET3_RXD_BTYPE_HEAD &&
1339 					rq->buf_info[ring_idx][i].skb) {
1340 				pci_unmap_single(adapter->pdev, rxd->addr,
1341 						 rxd->len, PCI_DMA_FROMDEVICE);
1342 				dev_kfree_skb(rq->buf_info[ring_idx][i].skb);
1343 				rq->buf_info[ring_idx][i].skb = NULL;
1344 			} else if (rxd->btype == VMXNET3_RXD_BTYPE_BODY &&
1345 					rq->buf_info[ring_idx][i].page) {
1346 				pci_unmap_page(adapter->pdev, rxd->addr,
1347 					       rxd->len, PCI_DMA_FROMDEVICE);
1348 				put_page(rq->buf_info[ring_idx][i].page);
1349 				rq->buf_info[ring_idx][i].page = NULL;
1350 			}
1351 		}
1352 
1353 		rq->rx_ring[ring_idx].gen = VMXNET3_INIT_GEN;
1354 		rq->rx_ring[ring_idx].next2fill =
1355 					rq->rx_ring[ring_idx].next2comp = 0;
1356 		rq->uncommitted[ring_idx] = 0;
1357 	}
1358 
1359 	rq->comp_ring.gen = VMXNET3_INIT_GEN;
1360 	rq->comp_ring.next2proc = 0;
1361 }
1362 
1363 
1364 static void
1365 vmxnet3_rq_cleanup_all(struct vmxnet3_adapter *adapter)
1366 {
1367 	int i;
1368 
1369 	for (i = 0; i < adapter->num_rx_queues; i++)
1370 		vmxnet3_rq_cleanup(&adapter->rx_queue[i], adapter);
1371 }
1372 
1373 
1374 void vmxnet3_rq_destroy(struct vmxnet3_rx_queue *rq,
1375 			struct vmxnet3_adapter *adapter)
1376 {
1377 	int i;
1378 	int j;
1379 
1380 	/* all rx buffers must have already been freed */
1381 	for (i = 0; i < 2; i++) {
1382 		if (rq->buf_info[i]) {
1383 			for (j = 0; j < rq->rx_ring[i].size; j++)
1384 				BUG_ON(rq->buf_info[i][j].page != NULL);
1385 		}
1386 	}
1387 
1388 
1389 	kfree(rq->buf_info[0]);
1390 
1391 	for (i = 0; i < 2; i++) {
1392 		if (rq->rx_ring[i].base) {
1393 			pci_free_consistent(adapter->pdev, rq->rx_ring[i].size
1394 					    * sizeof(struct Vmxnet3_RxDesc),
1395 					    rq->rx_ring[i].base,
1396 					    rq->rx_ring[i].basePA);
1397 			rq->rx_ring[i].base = NULL;
1398 		}
1399 		rq->buf_info[i] = NULL;
1400 	}
1401 
1402 	if (rq->comp_ring.base) {
1403 		pci_free_consistent(adapter->pdev, rq->comp_ring.size *
1404 				    sizeof(struct Vmxnet3_RxCompDesc),
1405 				    rq->comp_ring.base, rq->comp_ring.basePA);
1406 		rq->comp_ring.base = NULL;
1407 	}
1408 }
1409 
1410 
1411 static int
1412 vmxnet3_rq_init(struct vmxnet3_rx_queue *rq,
1413 		struct vmxnet3_adapter  *adapter)
1414 {
1415 	int i;
1416 
1417 	/* initialize buf_info */
1418 	for (i = 0; i < rq->rx_ring[0].size; i++) {
1419 
1420 		/* 1st buf for a pkt is skbuff */
1421 		if (i % adapter->rx_buf_per_pkt == 0) {
1422 			rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_SKB;
1423 			rq->buf_info[0][i].len = adapter->skb_buf_size;
1424 		} else { /* subsequent bufs for a pkt is frag */
1425 			rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_PAGE;
1426 			rq->buf_info[0][i].len = PAGE_SIZE;
1427 		}
1428 	}
1429 	for (i = 0; i < rq->rx_ring[1].size; i++) {
1430 		rq->buf_info[1][i].buf_type = VMXNET3_RX_BUF_PAGE;
1431 		rq->buf_info[1][i].len = PAGE_SIZE;
1432 	}
1433 
1434 	/* reset internal state and allocate buffers for both rings */
1435 	for (i = 0; i < 2; i++) {
1436 		rq->rx_ring[i].next2fill = rq->rx_ring[i].next2comp = 0;
1437 		rq->uncommitted[i] = 0;
1438 
1439 		memset(rq->rx_ring[i].base, 0, rq->rx_ring[i].size *
1440 		       sizeof(struct Vmxnet3_RxDesc));
1441 		rq->rx_ring[i].gen = VMXNET3_INIT_GEN;
1442 	}
1443 	if (vmxnet3_rq_alloc_rx_buf(rq, 0, rq->rx_ring[0].size - 1,
1444 				    adapter) == 0) {
1445 		/* at least has 1 rx buffer for the 1st ring */
1446 		return -ENOMEM;
1447 	}
1448 	vmxnet3_rq_alloc_rx_buf(rq, 1, rq->rx_ring[1].size - 1, adapter);
1449 
1450 	/* reset the comp ring */
1451 	rq->comp_ring.next2proc = 0;
1452 	memset(rq->comp_ring.base, 0, rq->comp_ring.size *
1453 	       sizeof(struct Vmxnet3_RxCompDesc));
1454 	rq->comp_ring.gen = VMXNET3_INIT_GEN;
1455 
1456 	/* reset rxctx */
1457 	rq->rx_ctx.skb = NULL;
1458 
1459 	/* stats are not reset */
1460 	return 0;
1461 }
1462 
1463 
1464 static int
1465 vmxnet3_rq_init_all(struct vmxnet3_adapter *adapter)
1466 {
1467 	int i, err = 0;
1468 
1469 	for (i = 0; i < adapter->num_rx_queues; i++) {
1470 		err = vmxnet3_rq_init(&adapter->rx_queue[i], adapter);
1471 		if (unlikely(err)) {
1472 			dev_err(&adapter->netdev->dev, "%s: failed to "
1473 				"initialize rx queue%i\n",
1474 				adapter->netdev->name, i);
1475 			break;
1476 		}
1477 	}
1478 	return err;
1479 
1480 }
1481 
1482 
1483 static int
1484 vmxnet3_rq_create(struct vmxnet3_rx_queue *rq, struct vmxnet3_adapter *adapter)
1485 {
1486 	int i;
1487 	size_t sz;
1488 	struct vmxnet3_rx_buf_info *bi;
1489 
1490 	for (i = 0; i < 2; i++) {
1491 
1492 		sz = rq->rx_ring[i].size * sizeof(struct Vmxnet3_RxDesc);
1493 		rq->rx_ring[i].base = pci_alloc_consistent(adapter->pdev, sz,
1494 							&rq->rx_ring[i].basePA);
1495 		if (!rq->rx_ring[i].base) {
1496 			printk(KERN_ERR "%s: failed to allocate rx ring %d\n",
1497 			       adapter->netdev->name, i);
1498 			goto err;
1499 		}
1500 	}
1501 
1502 	sz = rq->comp_ring.size * sizeof(struct Vmxnet3_RxCompDesc);
1503 	rq->comp_ring.base = pci_alloc_consistent(adapter->pdev, sz,
1504 						  &rq->comp_ring.basePA);
1505 	if (!rq->comp_ring.base) {
1506 		printk(KERN_ERR "%s: failed to allocate rx comp ring\n",
1507 		       adapter->netdev->name);
1508 		goto err;
1509 	}
1510 
1511 	sz = sizeof(struct vmxnet3_rx_buf_info) * (rq->rx_ring[0].size +
1512 						   rq->rx_ring[1].size);
1513 	bi = kzalloc(sz, GFP_KERNEL);
1514 	if (!bi)
1515 		goto err;
1516 
1517 	rq->buf_info[0] = bi;
1518 	rq->buf_info[1] = bi + rq->rx_ring[0].size;
1519 
1520 	return 0;
1521 
1522 err:
1523 	vmxnet3_rq_destroy(rq, adapter);
1524 	return -ENOMEM;
1525 }
1526 
1527 
1528 static int
1529 vmxnet3_rq_create_all(struct vmxnet3_adapter *adapter)
1530 {
1531 	int i, err = 0;
1532 
1533 	for (i = 0; i < adapter->num_rx_queues; i++) {
1534 		err = vmxnet3_rq_create(&adapter->rx_queue[i], adapter);
1535 		if (unlikely(err)) {
1536 			dev_err(&adapter->netdev->dev,
1537 				"%s: failed to create rx queue%i\n",
1538 				adapter->netdev->name, i);
1539 			goto err_out;
1540 		}
1541 	}
1542 	return err;
1543 err_out:
1544 	vmxnet3_rq_destroy_all(adapter);
1545 	return err;
1546 
1547 }
1548 
1549 /* Multiple queue aware polling function for tx and rx */
1550 
1551 static int
1552 vmxnet3_do_poll(struct vmxnet3_adapter *adapter, int budget)
1553 {
1554 	int rcd_done = 0, i;
1555 	if (unlikely(adapter->shared->ecr))
1556 		vmxnet3_process_events(adapter);
1557 	for (i = 0; i < adapter->num_tx_queues; i++)
1558 		vmxnet3_tq_tx_complete(&adapter->tx_queue[i], adapter);
1559 
1560 	for (i = 0; i < adapter->num_rx_queues; i++)
1561 		rcd_done += vmxnet3_rq_rx_complete(&adapter->rx_queue[i],
1562 						   adapter, budget);
1563 	return rcd_done;
1564 }
1565 
1566 
1567 static int
1568 vmxnet3_poll(struct napi_struct *napi, int budget)
1569 {
1570 	struct vmxnet3_rx_queue *rx_queue = container_of(napi,
1571 					  struct vmxnet3_rx_queue, napi);
1572 	int rxd_done;
1573 
1574 	rxd_done = vmxnet3_do_poll(rx_queue->adapter, budget);
1575 
1576 	if (rxd_done < budget) {
1577 		napi_complete(napi);
1578 		vmxnet3_enable_all_intrs(rx_queue->adapter);
1579 	}
1580 	return rxd_done;
1581 }
1582 
1583 /*
1584  * NAPI polling function for MSI-X mode with multiple Rx queues
1585  * Returns the # of the NAPI credit consumed (# of rx descriptors processed)
1586  */
1587 
1588 static int
1589 vmxnet3_poll_rx_only(struct napi_struct *napi, int budget)
1590 {
1591 	struct vmxnet3_rx_queue *rq = container_of(napi,
1592 						struct vmxnet3_rx_queue, napi);
1593 	struct vmxnet3_adapter *adapter = rq->adapter;
1594 	int rxd_done;
1595 
1596 	/* When sharing interrupt with corresponding tx queue, process
1597 	 * tx completions in that queue as well
1598 	 */
1599 	if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE) {
1600 		struct vmxnet3_tx_queue *tq =
1601 				&adapter->tx_queue[rq - adapter->rx_queue];
1602 		vmxnet3_tq_tx_complete(tq, adapter);
1603 	}
1604 
1605 	rxd_done = vmxnet3_rq_rx_complete(rq, adapter, budget);
1606 
1607 	if (rxd_done < budget) {
1608 		napi_complete(napi);
1609 		vmxnet3_enable_intr(adapter, rq->comp_ring.intr_idx);
1610 	}
1611 	return rxd_done;
1612 }
1613 
1614 
1615 #ifdef CONFIG_PCI_MSI
1616 
1617 /*
1618  * Handle completion interrupts on tx queues
1619  * Returns whether or not the intr is handled
1620  */
1621 
1622 static irqreturn_t
1623 vmxnet3_msix_tx(int irq, void *data)
1624 {
1625 	struct vmxnet3_tx_queue *tq = data;
1626 	struct vmxnet3_adapter *adapter = tq->adapter;
1627 
1628 	if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1629 		vmxnet3_disable_intr(adapter, tq->comp_ring.intr_idx);
1630 
1631 	/* Handle the case where only one irq is allocate for all tx queues */
1632 	if (adapter->share_intr == VMXNET3_INTR_TXSHARE) {
1633 		int i;
1634 		for (i = 0; i < adapter->num_tx_queues; i++) {
1635 			struct vmxnet3_tx_queue *txq = &adapter->tx_queue[i];
1636 			vmxnet3_tq_tx_complete(txq, adapter);
1637 		}
1638 	} else {
1639 		vmxnet3_tq_tx_complete(tq, adapter);
1640 	}
1641 	vmxnet3_enable_intr(adapter, tq->comp_ring.intr_idx);
1642 
1643 	return IRQ_HANDLED;
1644 }
1645 
1646 
1647 /*
1648  * Handle completion interrupts on rx queues. Returns whether or not the
1649  * intr is handled
1650  */
1651 
1652 static irqreturn_t
1653 vmxnet3_msix_rx(int irq, void *data)
1654 {
1655 	struct vmxnet3_rx_queue *rq = data;
1656 	struct vmxnet3_adapter *adapter = rq->adapter;
1657 
1658 	/* disable intr if needed */
1659 	if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1660 		vmxnet3_disable_intr(adapter, rq->comp_ring.intr_idx);
1661 	napi_schedule(&rq->napi);
1662 
1663 	return IRQ_HANDLED;
1664 }
1665 
1666 /*
1667  *----------------------------------------------------------------------------
1668  *
1669  * vmxnet3_msix_event --
1670  *
1671  *    vmxnet3 msix event intr handler
1672  *
1673  * Result:
1674  *    whether or not the intr is handled
1675  *
1676  *----------------------------------------------------------------------------
1677  */
1678 
1679 static irqreturn_t
1680 vmxnet3_msix_event(int irq, void *data)
1681 {
1682 	struct net_device *dev = data;
1683 	struct vmxnet3_adapter *adapter = netdev_priv(dev);
1684 
1685 	/* disable intr if needed */
1686 	if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1687 		vmxnet3_disable_intr(adapter, adapter->intr.event_intr_idx);
1688 
1689 	if (adapter->shared->ecr)
1690 		vmxnet3_process_events(adapter);
1691 
1692 	vmxnet3_enable_intr(adapter, adapter->intr.event_intr_idx);
1693 
1694 	return IRQ_HANDLED;
1695 }
1696 
1697 #endif /* CONFIG_PCI_MSI  */
1698 
1699 
1700 /* Interrupt handler for vmxnet3  */
1701 static irqreturn_t
1702 vmxnet3_intr(int irq, void *dev_id)
1703 {
1704 	struct net_device *dev = dev_id;
1705 	struct vmxnet3_adapter *adapter = netdev_priv(dev);
1706 
1707 	if (adapter->intr.type == VMXNET3_IT_INTX) {
1708 		u32 icr = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_ICR);
1709 		if (unlikely(icr == 0))
1710 			/* not ours */
1711 			return IRQ_NONE;
1712 	}
1713 
1714 
1715 	/* disable intr if needed */
1716 	if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1717 		vmxnet3_disable_all_intrs(adapter);
1718 
1719 	napi_schedule(&adapter->rx_queue[0].napi);
1720 
1721 	return IRQ_HANDLED;
1722 }
1723 
1724 #ifdef CONFIG_NET_POLL_CONTROLLER
1725 
1726 /* netpoll callback. */
1727 static void
1728 vmxnet3_netpoll(struct net_device *netdev)
1729 {
1730 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1731 
1732 	if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1733 		vmxnet3_disable_all_intrs(adapter);
1734 
1735 	vmxnet3_do_poll(adapter, adapter->rx_queue[0].rx_ring[0].size);
1736 	vmxnet3_enable_all_intrs(adapter);
1737 
1738 }
1739 #endif	/* CONFIG_NET_POLL_CONTROLLER */
1740 
1741 static int
1742 vmxnet3_request_irqs(struct vmxnet3_adapter *adapter)
1743 {
1744 	struct vmxnet3_intr *intr = &adapter->intr;
1745 	int err = 0, i;
1746 	int vector = 0;
1747 
1748 #ifdef CONFIG_PCI_MSI
1749 	if (adapter->intr.type == VMXNET3_IT_MSIX) {
1750 		for (i = 0; i < adapter->num_tx_queues; i++) {
1751 			if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE) {
1752 				sprintf(adapter->tx_queue[i].name, "%s-tx-%d",
1753 					adapter->netdev->name, vector);
1754 				err = request_irq(
1755 					      intr->msix_entries[vector].vector,
1756 					      vmxnet3_msix_tx, 0,
1757 					      adapter->tx_queue[i].name,
1758 					      &adapter->tx_queue[i]);
1759 			} else {
1760 				sprintf(adapter->tx_queue[i].name, "%s-rxtx-%d",
1761 					adapter->netdev->name, vector);
1762 			}
1763 			if (err) {
1764 				dev_err(&adapter->netdev->dev,
1765 					"Failed to request irq for MSIX, %s, "
1766 					"error %d\n",
1767 					adapter->tx_queue[i].name, err);
1768 				return err;
1769 			}
1770 
1771 			/* Handle the case where only 1 MSIx was allocated for
1772 			 * all tx queues */
1773 			if (adapter->share_intr == VMXNET3_INTR_TXSHARE) {
1774 				for (; i < adapter->num_tx_queues; i++)
1775 					adapter->tx_queue[i].comp_ring.intr_idx
1776 								= vector;
1777 				vector++;
1778 				break;
1779 			} else {
1780 				adapter->tx_queue[i].comp_ring.intr_idx
1781 								= vector++;
1782 			}
1783 		}
1784 		if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE)
1785 			vector = 0;
1786 
1787 		for (i = 0; i < adapter->num_rx_queues; i++) {
1788 			if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE)
1789 				sprintf(adapter->rx_queue[i].name, "%s-rx-%d",
1790 					adapter->netdev->name, vector);
1791 			else
1792 				sprintf(adapter->rx_queue[i].name, "%s-rxtx-%d",
1793 					adapter->netdev->name, vector);
1794 			err = request_irq(intr->msix_entries[vector].vector,
1795 					  vmxnet3_msix_rx, 0,
1796 					  adapter->rx_queue[i].name,
1797 					  &(adapter->rx_queue[i]));
1798 			if (err) {
1799 				printk(KERN_ERR "Failed to request irq for MSIX"
1800 				       ", %s, error %d\n",
1801 				       adapter->rx_queue[i].name, err);
1802 				return err;
1803 			}
1804 
1805 			adapter->rx_queue[i].comp_ring.intr_idx = vector++;
1806 		}
1807 
1808 		sprintf(intr->event_msi_vector_name, "%s-event-%d",
1809 			adapter->netdev->name, vector);
1810 		err = request_irq(intr->msix_entries[vector].vector,
1811 				  vmxnet3_msix_event, 0,
1812 				  intr->event_msi_vector_name, adapter->netdev);
1813 		intr->event_intr_idx = vector;
1814 
1815 	} else if (intr->type == VMXNET3_IT_MSI) {
1816 		adapter->num_rx_queues = 1;
1817 		err = request_irq(adapter->pdev->irq, vmxnet3_intr, 0,
1818 				  adapter->netdev->name, adapter->netdev);
1819 	} else {
1820 #endif
1821 		adapter->num_rx_queues = 1;
1822 		err = request_irq(adapter->pdev->irq, vmxnet3_intr,
1823 				  IRQF_SHARED, adapter->netdev->name,
1824 				  adapter->netdev);
1825 #ifdef CONFIG_PCI_MSI
1826 	}
1827 #endif
1828 	intr->num_intrs = vector + 1;
1829 	if (err) {
1830 		printk(KERN_ERR "Failed to request irq %s (intr type:%d), error"
1831 		       ":%d\n", adapter->netdev->name, intr->type, err);
1832 	} else {
1833 		/* Number of rx queues will not change after this */
1834 		for (i = 0; i < adapter->num_rx_queues; i++) {
1835 			struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
1836 			rq->qid = i;
1837 			rq->qid2 = i + adapter->num_rx_queues;
1838 		}
1839 
1840 
1841 
1842 		/* init our intr settings */
1843 		for (i = 0; i < intr->num_intrs; i++)
1844 			intr->mod_levels[i] = UPT1_IML_ADAPTIVE;
1845 		if (adapter->intr.type != VMXNET3_IT_MSIX) {
1846 			adapter->intr.event_intr_idx = 0;
1847 			for (i = 0; i < adapter->num_tx_queues; i++)
1848 				adapter->tx_queue[i].comp_ring.intr_idx = 0;
1849 			adapter->rx_queue[0].comp_ring.intr_idx = 0;
1850 		}
1851 
1852 		printk(KERN_INFO "%s: intr type %u, mode %u, %u vectors "
1853 		       "allocated\n", adapter->netdev->name, intr->type,
1854 		       intr->mask_mode, intr->num_intrs);
1855 	}
1856 
1857 	return err;
1858 }
1859 
1860 
1861 static void
1862 vmxnet3_free_irqs(struct vmxnet3_adapter *adapter)
1863 {
1864 	struct vmxnet3_intr *intr = &adapter->intr;
1865 	BUG_ON(intr->type == VMXNET3_IT_AUTO || intr->num_intrs <= 0);
1866 
1867 	switch (intr->type) {
1868 #ifdef CONFIG_PCI_MSI
1869 	case VMXNET3_IT_MSIX:
1870 	{
1871 		int i, vector = 0;
1872 
1873 		if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE) {
1874 			for (i = 0; i < adapter->num_tx_queues; i++) {
1875 				free_irq(intr->msix_entries[vector++].vector,
1876 					 &(adapter->tx_queue[i]));
1877 				if (adapter->share_intr == VMXNET3_INTR_TXSHARE)
1878 					break;
1879 			}
1880 		}
1881 
1882 		for (i = 0; i < adapter->num_rx_queues; i++) {
1883 			free_irq(intr->msix_entries[vector++].vector,
1884 				 &(adapter->rx_queue[i]));
1885 		}
1886 
1887 		free_irq(intr->msix_entries[vector].vector,
1888 			 adapter->netdev);
1889 		BUG_ON(vector >= intr->num_intrs);
1890 		break;
1891 	}
1892 #endif
1893 	case VMXNET3_IT_MSI:
1894 		free_irq(adapter->pdev->irq, adapter->netdev);
1895 		break;
1896 	case VMXNET3_IT_INTX:
1897 		free_irq(adapter->pdev->irq, adapter->netdev);
1898 		break;
1899 	default:
1900 		BUG_ON(true);
1901 	}
1902 }
1903 
1904 
1905 static void
1906 vmxnet3_restore_vlan(struct vmxnet3_adapter *adapter)
1907 {
1908 	u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1909 	u16 vid;
1910 
1911 	/* allow untagged pkts */
1912 	VMXNET3_SET_VFTABLE_ENTRY(vfTable, 0);
1913 
1914 	for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
1915 		VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid);
1916 }
1917 
1918 
1919 static int
1920 vmxnet3_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1921 {
1922 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1923 
1924 	if (!(netdev->flags & IFF_PROMISC)) {
1925 		u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1926 		unsigned long flags;
1927 
1928 		VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid);
1929 		spin_lock_irqsave(&adapter->cmd_lock, flags);
1930 		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1931 				       VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1932 		spin_unlock_irqrestore(&adapter->cmd_lock, flags);
1933 	}
1934 
1935 	set_bit(vid, adapter->active_vlans);
1936 
1937 	return 0;
1938 }
1939 
1940 
1941 static int
1942 vmxnet3_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1943 {
1944 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1945 
1946 	if (!(netdev->flags & IFF_PROMISC)) {
1947 		u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1948 		unsigned long flags;
1949 
1950 		VMXNET3_CLEAR_VFTABLE_ENTRY(vfTable, vid);
1951 		spin_lock_irqsave(&adapter->cmd_lock, flags);
1952 		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1953 				       VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1954 		spin_unlock_irqrestore(&adapter->cmd_lock, flags);
1955 	}
1956 
1957 	clear_bit(vid, adapter->active_vlans);
1958 
1959 	return 0;
1960 }
1961 
1962 
1963 static u8 *
1964 vmxnet3_copy_mc(struct net_device *netdev)
1965 {
1966 	u8 *buf = NULL;
1967 	u32 sz = netdev_mc_count(netdev) * ETH_ALEN;
1968 
1969 	/* struct Vmxnet3_RxFilterConf.mfTableLen is u16. */
1970 	if (sz <= 0xffff) {
1971 		/* We may be called with BH disabled */
1972 		buf = kmalloc(sz, GFP_ATOMIC);
1973 		if (buf) {
1974 			struct netdev_hw_addr *ha;
1975 			int i = 0;
1976 
1977 			netdev_for_each_mc_addr(ha, netdev)
1978 				memcpy(buf + i++ * ETH_ALEN, ha->addr,
1979 				       ETH_ALEN);
1980 		}
1981 	}
1982 	return buf;
1983 }
1984 
1985 
1986 static void
1987 vmxnet3_set_mc(struct net_device *netdev)
1988 {
1989 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1990 	unsigned long flags;
1991 	struct Vmxnet3_RxFilterConf *rxConf =
1992 					&adapter->shared->devRead.rxFilterConf;
1993 	u8 *new_table = NULL;
1994 	u32 new_mode = VMXNET3_RXM_UCAST;
1995 
1996 	if (netdev->flags & IFF_PROMISC) {
1997 		u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1998 		memset(vfTable, 0, VMXNET3_VFT_SIZE * sizeof(*vfTable));
1999 
2000 		new_mode |= VMXNET3_RXM_PROMISC;
2001 	} else {
2002 		vmxnet3_restore_vlan(adapter);
2003 	}
2004 
2005 	if (netdev->flags & IFF_BROADCAST)
2006 		new_mode |= VMXNET3_RXM_BCAST;
2007 
2008 	if (netdev->flags & IFF_ALLMULTI)
2009 		new_mode |= VMXNET3_RXM_ALL_MULTI;
2010 	else
2011 		if (!netdev_mc_empty(netdev)) {
2012 			new_table = vmxnet3_copy_mc(netdev);
2013 			if (new_table) {
2014 				new_mode |= VMXNET3_RXM_MCAST;
2015 				rxConf->mfTableLen = cpu_to_le16(
2016 					netdev_mc_count(netdev) * ETH_ALEN);
2017 				rxConf->mfTablePA = cpu_to_le64(virt_to_phys(
2018 						    new_table));
2019 			} else {
2020 				printk(KERN_INFO "%s: failed to copy mcast list"
2021 				       ", setting ALL_MULTI\n", netdev->name);
2022 				new_mode |= VMXNET3_RXM_ALL_MULTI;
2023 			}
2024 		}
2025 
2026 
2027 	if (!(new_mode & VMXNET3_RXM_MCAST)) {
2028 		rxConf->mfTableLen = 0;
2029 		rxConf->mfTablePA = 0;
2030 	}
2031 
2032 	spin_lock_irqsave(&adapter->cmd_lock, flags);
2033 	if (new_mode != rxConf->rxMode) {
2034 		rxConf->rxMode = cpu_to_le32(new_mode);
2035 		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2036 				       VMXNET3_CMD_UPDATE_RX_MODE);
2037 		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2038 				       VMXNET3_CMD_UPDATE_VLAN_FILTERS);
2039 	}
2040 
2041 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2042 			       VMXNET3_CMD_UPDATE_MAC_FILTERS);
2043 	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2044 
2045 	kfree(new_table);
2046 }
2047 
2048 void
2049 vmxnet3_rq_destroy_all(struct vmxnet3_adapter *adapter)
2050 {
2051 	int i;
2052 
2053 	for (i = 0; i < adapter->num_rx_queues; i++)
2054 		vmxnet3_rq_destroy(&adapter->rx_queue[i], adapter);
2055 }
2056 
2057 
2058 /*
2059  *   Set up driver_shared based on settings in adapter.
2060  */
2061 
2062 static void
2063 vmxnet3_setup_driver_shared(struct vmxnet3_adapter *adapter)
2064 {
2065 	struct Vmxnet3_DriverShared *shared = adapter->shared;
2066 	struct Vmxnet3_DSDevRead *devRead = &shared->devRead;
2067 	struct Vmxnet3_TxQueueConf *tqc;
2068 	struct Vmxnet3_RxQueueConf *rqc;
2069 	int i;
2070 
2071 	memset(shared, 0, sizeof(*shared));
2072 
2073 	/* driver settings */
2074 	shared->magic = cpu_to_le32(VMXNET3_REV1_MAGIC);
2075 	devRead->misc.driverInfo.version = cpu_to_le32(
2076 						VMXNET3_DRIVER_VERSION_NUM);
2077 	devRead->misc.driverInfo.gos.gosBits = (sizeof(void *) == 4 ?
2078 				VMXNET3_GOS_BITS_32 : VMXNET3_GOS_BITS_64);
2079 	devRead->misc.driverInfo.gos.gosType = VMXNET3_GOS_TYPE_LINUX;
2080 	*((u32 *)&devRead->misc.driverInfo.gos) = cpu_to_le32(
2081 				*((u32 *)&devRead->misc.driverInfo.gos));
2082 	devRead->misc.driverInfo.vmxnet3RevSpt = cpu_to_le32(1);
2083 	devRead->misc.driverInfo.uptVerSpt = cpu_to_le32(1);
2084 
2085 	devRead->misc.ddPA = cpu_to_le64(virt_to_phys(adapter));
2086 	devRead->misc.ddLen = cpu_to_le32(sizeof(struct vmxnet3_adapter));
2087 
2088 	/* set up feature flags */
2089 	if (adapter->netdev->features & NETIF_F_RXCSUM)
2090 		devRead->misc.uptFeatures |= UPT1_F_RXCSUM;
2091 
2092 	if (adapter->netdev->features & NETIF_F_LRO) {
2093 		devRead->misc.uptFeatures |= UPT1_F_LRO;
2094 		devRead->misc.maxNumRxSG = cpu_to_le16(1 + MAX_SKB_FRAGS);
2095 	}
2096 	if (adapter->netdev->features & NETIF_F_HW_VLAN_RX)
2097 		devRead->misc.uptFeatures |= UPT1_F_RXVLAN;
2098 
2099 	devRead->misc.mtu = cpu_to_le32(adapter->netdev->mtu);
2100 	devRead->misc.queueDescPA = cpu_to_le64(adapter->queue_desc_pa);
2101 	devRead->misc.queueDescLen = cpu_to_le32(
2102 		adapter->num_tx_queues * sizeof(struct Vmxnet3_TxQueueDesc) +
2103 		adapter->num_rx_queues * sizeof(struct Vmxnet3_RxQueueDesc));
2104 
2105 	/* tx queue settings */
2106 	devRead->misc.numTxQueues =  adapter->num_tx_queues;
2107 	for (i = 0; i < adapter->num_tx_queues; i++) {
2108 		struct vmxnet3_tx_queue	*tq = &adapter->tx_queue[i];
2109 		BUG_ON(adapter->tx_queue[i].tx_ring.base == NULL);
2110 		tqc = &adapter->tqd_start[i].conf;
2111 		tqc->txRingBasePA   = cpu_to_le64(tq->tx_ring.basePA);
2112 		tqc->dataRingBasePA = cpu_to_le64(tq->data_ring.basePA);
2113 		tqc->compRingBasePA = cpu_to_le64(tq->comp_ring.basePA);
2114 		tqc->ddPA           = cpu_to_le64(virt_to_phys(tq->buf_info));
2115 		tqc->txRingSize     = cpu_to_le32(tq->tx_ring.size);
2116 		tqc->dataRingSize   = cpu_to_le32(tq->data_ring.size);
2117 		tqc->compRingSize   = cpu_to_le32(tq->comp_ring.size);
2118 		tqc->ddLen          = cpu_to_le32(
2119 					sizeof(struct vmxnet3_tx_buf_info) *
2120 					tqc->txRingSize);
2121 		tqc->intrIdx        = tq->comp_ring.intr_idx;
2122 	}
2123 
2124 	/* rx queue settings */
2125 	devRead->misc.numRxQueues = adapter->num_rx_queues;
2126 	for (i = 0; i < adapter->num_rx_queues; i++) {
2127 		struct vmxnet3_rx_queue	*rq = &adapter->rx_queue[i];
2128 		rqc = &adapter->rqd_start[i].conf;
2129 		rqc->rxRingBasePA[0] = cpu_to_le64(rq->rx_ring[0].basePA);
2130 		rqc->rxRingBasePA[1] = cpu_to_le64(rq->rx_ring[1].basePA);
2131 		rqc->compRingBasePA  = cpu_to_le64(rq->comp_ring.basePA);
2132 		rqc->ddPA            = cpu_to_le64(virt_to_phys(
2133 							rq->buf_info));
2134 		rqc->rxRingSize[0]   = cpu_to_le32(rq->rx_ring[0].size);
2135 		rqc->rxRingSize[1]   = cpu_to_le32(rq->rx_ring[1].size);
2136 		rqc->compRingSize    = cpu_to_le32(rq->comp_ring.size);
2137 		rqc->ddLen           = cpu_to_le32(
2138 					sizeof(struct vmxnet3_rx_buf_info) *
2139 					(rqc->rxRingSize[0] +
2140 					 rqc->rxRingSize[1]));
2141 		rqc->intrIdx         = rq->comp_ring.intr_idx;
2142 	}
2143 
2144 #ifdef VMXNET3_RSS
2145 	memset(adapter->rss_conf, 0, sizeof(*adapter->rss_conf));
2146 
2147 	if (adapter->rss) {
2148 		struct UPT1_RSSConf *rssConf = adapter->rss_conf;
2149 		devRead->misc.uptFeatures |= UPT1_F_RSS;
2150 		devRead->misc.numRxQueues = adapter->num_rx_queues;
2151 		rssConf->hashType = UPT1_RSS_HASH_TYPE_TCP_IPV4 |
2152 				    UPT1_RSS_HASH_TYPE_IPV4 |
2153 				    UPT1_RSS_HASH_TYPE_TCP_IPV6 |
2154 				    UPT1_RSS_HASH_TYPE_IPV6;
2155 		rssConf->hashFunc = UPT1_RSS_HASH_FUNC_TOEPLITZ;
2156 		rssConf->hashKeySize = UPT1_RSS_MAX_KEY_SIZE;
2157 		rssConf->indTableSize = VMXNET3_RSS_IND_TABLE_SIZE;
2158 		get_random_bytes(&rssConf->hashKey[0], rssConf->hashKeySize);
2159 		for (i = 0; i < rssConf->indTableSize; i++)
2160 			rssConf->indTable[i] = ethtool_rxfh_indir_default(
2161 				i, adapter->num_rx_queues);
2162 
2163 		devRead->rssConfDesc.confVer = 1;
2164 		devRead->rssConfDesc.confLen = sizeof(*rssConf);
2165 		devRead->rssConfDesc.confPA  = virt_to_phys(rssConf);
2166 	}
2167 
2168 #endif /* VMXNET3_RSS */
2169 
2170 	/* intr settings */
2171 	devRead->intrConf.autoMask = adapter->intr.mask_mode ==
2172 				     VMXNET3_IMM_AUTO;
2173 	devRead->intrConf.numIntrs = adapter->intr.num_intrs;
2174 	for (i = 0; i < adapter->intr.num_intrs; i++)
2175 		devRead->intrConf.modLevels[i] = adapter->intr.mod_levels[i];
2176 
2177 	devRead->intrConf.eventIntrIdx = adapter->intr.event_intr_idx;
2178 	devRead->intrConf.intrCtrl |= cpu_to_le32(VMXNET3_IC_DISABLE_ALL);
2179 
2180 	/* rx filter settings */
2181 	devRead->rxFilterConf.rxMode = 0;
2182 	vmxnet3_restore_vlan(adapter);
2183 	vmxnet3_write_mac_addr(adapter, adapter->netdev->dev_addr);
2184 
2185 	/* the rest are already zeroed */
2186 }
2187 
2188 
2189 int
2190 vmxnet3_activate_dev(struct vmxnet3_adapter *adapter)
2191 {
2192 	int err, i;
2193 	u32 ret;
2194 	unsigned long flags;
2195 
2196 	dev_dbg(&adapter->netdev->dev, "%s: skb_buf_size %d, rx_buf_per_pkt %d,"
2197 		" ring sizes %u %u %u\n", adapter->netdev->name,
2198 		adapter->skb_buf_size, adapter->rx_buf_per_pkt,
2199 		adapter->tx_queue[0].tx_ring.size,
2200 		adapter->rx_queue[0].rx_ring[0].size,
2201 		adapter->rx_queue[0].rx_ring[1].size);
2202 
2203 	vmxnet3_tq_init_all(adapter);
2204 	err = vmxnet3_rq_init_all(adapter);
2205 	if (err) {
2206 		printk(KERN_ERR "Failed to init rx queue for %s: error %d\n",
2207 		       adapter->netdev->name, err);
2208 		goto rq_err;
2209 	}
2210 
2211 	err = vmxnet3_request_irqs(adapter);
2212 	if (err) {
2213 		printk(KERN_ERR "Failed to setup irq for %s: error %d\n",
2214 		       adapter->netdev->name, err);
2215 		goto irq_err;
2216 	}
2217 
2218 	vmxnet3_setup_driver_shared(adapter);
2219 
2220 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, VMXNET3_GET_ADDR_LO(
2221 			       adapter->shared_pa));
2222 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, VMXNET3_GET_ADDR_HI(
2223 			       adapter->shared_pa));
2224 	spin_lock_irqsave(&adapter->cmd_lock, flags);
2225 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2226 			       VMXNET3_CMD_ACTIVATE_DEV);
2227 	ret = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
2228 	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2229 
2230 	if (ret != 0) {
2231 		printk(KERN_ERR "Failed to activate dev %s: error %u\n",
2232 		       adapter->netdev->name, ret);
2233 		err = -EINVAL;
2234 		goto activate_err;
2235 	}
2236 
2237 	for (i = 0; i < adapter->num_rx_queues; i++) {
2238 		VMXNET3_WRITE_BAR0_REG(adapter,
2239 				VMXNET3_REG_RXPROD + i * VMXNET3_REG_ALIGN,
2240 				adapter->rx_queue[i].rx_ring[0].next2fill);
2241 		VMXNET3_WRITE_BAR0_REG(adapter, (VMXNET3_REG_RXPROD2 +
2242 				(i * VMXNET3_REG_ALIGN)),
2243 				adapter->rx_queue[i].rx_ring[1].next2fill);
2244 	}
2245 
2246 	/* Apply the rx filter settins last. */
2247 	vmxnet3_set_mc(adapter->netdev);
2248 
2249 	/*
2250 	 * Check link state when first activating device. It will start the
2251 	 * tx queue if the link is up.
2252 	 */
2253 	vmxnet3_check_link(adapter, true);
2254 	for (i = 0; i < adapter->num_rx_queues; i++)
2255 		napi_enable(&adapter->rx_queue[i].napi);
2256 	vmxnet3_enable_all_intrs(adapter);
2257 	clear_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
2258 	return 0;
2259 
2260 activate_err:
2261 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, 0);
2262 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, 0);
2263 	vmxnet3_free_irqs(adapter);
2264 irq_err:
2265 rq_err:
2266 	/* free up buffers we allocated */
2267 	vmxnet3_rq_cleanup_all(adapter);
2268 	return err;
2269 }
2270 
2271 
2272 void
2273 vmxnet3_reset_dev(struct vmxnet3_adapter *adapter)
2274 {
2275 	unsigned long flags;
2276 	spin_lock_irqsave(&adapter->cmd_lock, flags);
2277 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, VMXNET3_CMD_RESET_DEV);
2278 	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2279 }
2280 
2281 
2282 int
2283 vmxnet3_quiesce_dev(struct vmxnet3_adapter *adapter)
2284 {
2285 	int i;
2286 	unsigned long flags;
2287 	if (test_and_set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state))
2288 		return 0;
2289 
2290 
2291 	spin_lock_irqsave(&adapter->cmd_lock, flags);
2292 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2293 			       VMXNET3_CMD_QUIESCE_DEV);
2294 	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2295 	vmxnet3_disable_all_intrs(adapter);
2296 
2297 	for (i = 0; i < adapter->num_rx_queues; i++)
2298 		napi_disable(&adapter->rx_queue[i].napi);
2299 	netif_tx_disable(adapter->netdev);
2300 	adapter->link_speed = 0;
2301 	netif_carrier_off(adapter->netdev);
2302 
2303 	vmxnet3_tq_cleanup_all(adapter);
2304 	vmxnet3_rq_cleanup_all(adapter);
2305 	vmxnet3_free_irqs(adapter);
2306 	return 0;
2307 }
2308 
2309 
2310 static void
2311 vmxnet3_write_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac)
2312 {
2313 	u32 tmp;
2314 
2315 	tmp = *(u32 *)mac;
2316 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACL, tmp);
2317 
2318 	tmp = (mac[5] << 8) | mac[4];
2319 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACH, tmp);
2320 }
2321 
2322 
2323 static int
2324 vmxnet3_set_mac_addr(struct net_device *netdev, void *p)
2325 {
2326 	struct sockaddr *addr = p;
2327 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2328 
2329 	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2330 	vmxnet3_write_mac_addr(adapter, addr->sa_data);
2331 
2332 	return 0;
2333 }
2334 
2335 
2336 /* ==================== initialization and cleanup routines ============ */
2337 
2338 static int
2339 vmxnet3_alloc_pci_resources(struct vmxnet3_adapter *adapter, bool *dma64)
2340 {
2341 	int err;
2342 	unsigned long mmio_start, mmio_len;
2343 	struct pci_dev *pdev = adapter->pdev;
2344 
2345 	err = pci_enable_device(pdev);
2346 	if (err) {
2347 		printk(KERN_ERR "Failed to enable adapter %s: error %d\n",
2348 		       pci_name(pdev), err);
2349 		return err;
2350 	}
2351 
2352 	if (pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) == 0) {
2353 		if (pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64)) != 0) {
2354 			printk(KERN_ERR "pci_set_consistent_dma_mask failed "
2355 			       "for adapter %s\n", pci_name(pdev));
2356 			err = -EIO;
2357 			goto err_set_mask;
2358 		}
2359 		*dma64 = true;
2360 	} else {
2361 		if (pci_set_dma_mask(pdev, DMA_BIT_MASK(32)) != 0) {
2362 			printk(KERN_ERR "pci_set_dma_mask failed for adapter "
2363 			       "%s\n",	pci_name(pdev));
2364 			err = -EIO;
2365 			goto err_set_mask;
2366 		}
2367 		*dma64 = false;
2368 	}
2369 
2370 	err = pci_request_selected_regions(pdev, (1 << 2) - 1,
2371 					   vmxnet3_driver_name);
2372 	if (err) {
2373 		printk(KERN_ERR "Failed to request region for adapter %s: "
2374 		       "error %d\n", pci_name(pdev), err);
2375 		goto err_set_mask;
2376 	}
2377 
2378 	pci_set_master(pdev);
2379 
2380 	mmio_start = pci_resource_start(pdev, 0);
2381 	mmio_len = pci_resource_len(pdev, 0);
2382 	adapter->hw_addr0 = ioremap(mmio_start, mmio_len);
2383 	if (!adapter->hw_addr0) {
2384 		printk(KERN_ERR "Failed to map bar0 for adapter %s\n",
2385 		       pci_name(pdev));
2386 		err = -EIO;
2387 		goto err_ioremap;
2388 	}
2389 
2390 	mmio_start = pci_resource_start(pdev, 1);
2391 	mmio_len = pci_resource_len(pdev, 1);
2392 	adapter->hw_addr1 = ioremap(mmio_start, mmio_len);
2393 	if (!adapter->hw_addr1) {
2394 		printk(KERN_ERR "Failed to map bar1 for adapter %s\n",
2395 		       pci_name(pdev));
2396 		err = -EIO;
2397 		goto err_bar1;
2398 	}
2399 	return 0;
2400 
2401 err_bar1:
2402 	iounmap(adapter->hw_addr0);
2403 err_ioremap:
2404 	pci_release_selected_regions(pdev, (1 << 2) - 1);
2405 err_set_mask:
2406 	pci_disable_device(pdev);
2407 	return err;
2408 }
2409 
2410 
2411 static void
2412 vmxnet3_free_pci_resources(struct vmxnet3_adapter *adapter)
2413 {
2414 	BUG_ON(!adapter->pdev);
2415 
2416 	iounmap(adapter->hw_addr0);
2417 	iounmap(adapter->hw_addr1);
2418 	pci_release_selected_regions(adapter->pdev, (1 << 2) - 1);
2419 	pci_disable_device(adapter->pdev);
2420 }
2421 
2422 
2423 static void
2424 vmxnet3_adjust_rx_ring_size(struct vmxnet3_adapter *adapter)
2425 {
2426 	size_t sz, i, ring0_size, ring1_size, comp_size;
2427 	struct vmxnet3_rx_queue	*rq = &adapter->rx_queue[0];
2428 
2429 
2430 	if (adapter->netdev->mtu <= VMXNET3_MAX_SKB_BUF_SIZE -
2431 				    VMXNET3_MAX_ETH_HDR_SIZE) {
2432 		adapter->skb_buf_size = adapter->netdev->mtu +
2433 					VMXNET3_MAX_ETH_HDR_SIZE;
2434 		if (adapter->skb_buf_size < VMXNET3_MIN_T0_BUF_SIZE)
2435 			adapter->skb_buf_size = VMXNET3_MIN_T0_BUF_SIZE;
2436 
2437 		adapter->rx_buf_per_pkt = 1;
2438 	} else {
2439 		adapter->skb_buf_size = VMXNET3_MAX_SKB_BUF_SIZE;
2440 		sz = adapter->netdev->mtu - VMXNET3_MAX_SKB_BUF_SIZE +
2441 					    VMXNET3_MAX_ETH_HDR_SIZE;
2442 		adapter->rx_buf_per_pkt = 1 + (sz + PAGE_SIZE - 1) / PAGE_SIZE;
2443 	}
2444 
2445 	/*
2446 	 * for simplicity, force the ring0 size to be a multiple of
2447 	 * rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN
2448 	 */
2449 	sz = adapter->rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN;
2450 	ring0_size = adapter->rx_queue[0].rx_ring[0].size;
2451 	ring0_size = (ring0_size + sz - 1) / sz * sz;
2452 	ring0_size = min_t(u32, ring0_size, VMXNET3_RX_RING_MAX_SIZE /
2453 			   sz * sz);
2454 	ring1_size = adapter->rx_queue[0].rx_ring[1].size;
2455 	comp_size = ring0_size + ring1_size;
2456 
2457 	for (i = 0; i < adapter->num_rx_queues; i++) {
2458 		rq = &adapter->rx_queue[i];
2459 		rq->rx_ring[0].size = ring0_size;
2460 		rq->rx_ring[1].size = ring1_size;
2461 		rq->comp_ring.size = comp_size;
2462 	}
2463 }
2464 
2465 
2466 int
2467 vmxnet3_create_queues(struct vmxnet3_adapter *adapter, u32 tx_ring_size,
2468 		      u32 rx_ring_size, u32 rx_ring2_size)
2469 {
2470 	int err = 0, i;
2471 
2472 	for (i = 0; i < adapter->num_tx_queues; i++) {
2473 		struct vmxnet3_tx_queue	*tq = &adapter->tx_queue[i];
2474 		tq->tx_ring.size   = tx_ring_size;
2475 		tq->data_ring.size = tx_ring_size;
2476 		tq->comp_ring.size = tx_ring_size;
2477 		tq->shared = &adapter->tqd_start[i].ctrl;
2478 		tq->stopped = true;
2479 		tq->adapter = adapter;
2480 		tq->qid = i;
2481 		err = vmxnet3_tq_create(tq, adapter);
2482 		/*
2483 		 * Too late to change num_tx_queues. We cannot do away with
2484 		 * lesser number of queues than what we asked for
2485 		 */
2486 		if (err)
2487 			goto queue_err;
2488 	}
2489 
2490 	adapter->rx_queue[0].rx_ring[0].size = rx_ring_size;
2491 	adapter->rx_queue[0].rx_ring[1].size = rx_ring2_size;
2492 	vmxnet3_adjust_rx_ring_size(adapter);
2493 	for (i = 0; i < adapter->num_rx_queues; i++) {
2494 		struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
2495 		/* qid and qid2 for rx queues will be assigned later when num
2496 		 * of rx queues is finalized after allocating intrs */
2497 		rq->shared = &adapter->rqd_start[i].ctrl;
2498 		rq->adapter = adapter;
2499 		err = vmxnet3_rq_create(rq, adapter);
2500 		if (err) {
2501 			if (i == 0) {
2502 				printk(KERN_ERR "Could not allocate any rx"
2503 				       "queues. Aborting.\n");
2504 				goto queue_err;
2505 			} else {
2506 				printk(KERN_INFO "Number of rx queues changed "
2507 				       "to : %d.\n", i);
2508 				adapter->num_rx_queues = i;
2509 				err = 0;
2510 				break;
2511 			}
2512 		}
2513 	}
2514 	return err;
2515 queue_err:
2516 	vmxnet3_tq_destroy_all(adapter);
2517 	return err;
2518 }
2519 
2520 static int
2521 vmxnet3_open(struct net_device *netdev)
2522 {
2523 	struct vmxnet3_adapter *adapter;
2524 	int err, i;
2525 
2526 	adapter = netdev_priv(netdev);
2527 
2528 	for (i = 0; i < adapter->num_tx_queues; i++)
2529 		spin_lock_init(&adapter->tx_queue[i].tx_lock);
2530 
2531 	err = vmxnet3_create_queues(adapter, VMXNET3_DEF_TX_RING_SIZE,
2532 				    VMXNET3_DEF_RX_RING_SIZE,
2533 				    VMXNET3_DEF_RX_RING_SIZE);
2534 	if (err)
2535 		goto queue_err;
2536 
2537 	err = vmxnet3_activate_dev(adapter);
2538 	if (err)
2539 		goto activate_err;
2540 
2541 	return 0;
2542 
2543 activate_err:
2544 	vmxnet3_rq_destroy_all(adapter);
2545 	vmxnet3_tq_destroy_all(adapter);
2546 queue_err:
2547 	return err;
2548 }
2549 
2550 
2551 static int
2552 vmxnet3_close(struct net_device *netdev)
2553 {
2554 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2555 
2556 	/*
2557 	 * Reset_work may be in the middle of resetting the device, wait for its
2558 	 * completion.
2559 	 */
2560 	while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2561 		msleep(1);
2562 
2563 	vmxnet3_quiesce_dev(adapter);
2564 
2565 	vmxnet3_rq_destroy_all(adapter);
2566 	vmxnet3_tq_destroy_all(adapter);
2567 
2568 	clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2569 
2570 
2571 	return 0;
2572 }
2573 
2574 
2575 void
2576 vmxnet3_force_close(struct vmxnet3_adapter *adapter)
2577 {
2578 	int i;
2579 
2580 	/*
2581 	 * we must clear VMXNET3_STATE_BIT_RESETTING, otherwise
2582 	 * vmxnet3_close() will deadlock.
2583 	 */
2584 	BUG_ON(test_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state));
2585 
2586 	/* we need to enable NAPI, otherwise dev_close will deadlock */
2587 	for (i = 0; i < adapter->num_rx_queues; i++)
2588 		napi_enable(&adapter->rx_queue[i].napi);
2589 	dev_close(adapter->netdev);
2590 }
2591 
2592 
2593 static int
2594 vmxnet3_change_mtu(struct net_device *netdev, int new_mtu)
2595 {
2596 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2597 	int err = 0;
2598 
2599 	if (new_mtu < VMXNET3_MIN_MTU || new_mtu > VMXNET3_MAX_MTU)
2600 		return -EINVAL;
2601 
2602 	netdev->mtu = new_mtu;
2603 
2604 	/*
2605 	 * Reset_work may be in the middle of resetting the device, wait for its
2606 	 * completion.
2607 	 */
2608 	while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2609 		msleep(1);
2610 
2611 	if (netif_running(netdev)) {
2612 		vmxnet3_quiesce_dev(adapter);
2613 		vmxnet3_reset_dev(adapter);
2614 
2615 		/* we need to re-create the rx queue based on the new mtu */
2616 		vmxnet3_rq_destroy_all(adapter);
2617 		vmxnet3_adjust_rx_ring_size(adapter);
2618 		err = vmxnet3_rq_create_all(adapter);
2619 		if (err) {
2620 			printk(KERN_ERR "%s: failed to re-create rx queues,"
2621 				" error %d. Closing it.\n", netdev->name, err);
2622 			goto out;
2623 		}
2624 
2625 		err = vmxnet3_activate_dev(adapter);
2626 		if (err) {
2627 			printk(KERN_ERR "%s: failed to re-activate, error %d. "
2628 				"Closing it\n", netdev->name, err);
2629 			goto out;
2630 		}
2631 	}
2632 
2633 out:
2634 	clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2635 	if (err)
2636 		vmxnet3_force_close(adapter);
2637 
2638 	return err;
2639 }
2640 
2641 
2642 static void
2643 vmxnet3_declare_features(struct vmxnet3_adapter *adapter, bool dma64)
2644 {
2645 	struct net_device *netdev = adapter->netdev;
2646 
2647 	netdev->hw_features = NETIF_F_SG | NETIF_F_RXCSUM |
2648 		NETIF_F_HW_CSUM | NETIF_F_HW_VLAN_TX |
2649 		NETIF_F_HW_VLAN_RX | NETIF_F_TSO | NETIF_F_TSO6 |
2650 		NETIF_F_LRO;
2651 	if (dma64)
2652 		netdev->hw_features |= NETIF_F_HIGHDMA;
2653 	netdev->vlan_features = netdev->hw_features &
2654 				~(NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX);
2655 	netdev->features = netdev->hw_features | NETIF_F_HW_VLAN_FILTER;
2656 
2657 	netdev_info(adapter->netdev,
2658 		"features: sg csum vlan jf tso tsoIPv6 lro%s\n",
2659 		dma64 ? " highDMA" : "");
2660 }
2661 
2662 
2663 static void
2664 vmxnet3_read_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac)
2665 {
2666 	u32 tmp;
2667 
2668 	tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACL);
2669 	*(u32 *)mac = tmp;
2670 
2671 	tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACH);
2672 	mac[4] = tmp & 0xff;
2673 	mac[5] = (tmp >> 8) & 0xff;
2674 }
2675 
2676 #ifdef CONFIG_PCI_MSI
2677 
2678 /*
2679  * Enable MSIx vectors.
2680  * Returns :
2681  *	0 on successful enabling of required vectors,
2682  *	VMXNET3_LINUX_MIN_MSIX_VECT when only minimum number of vectors required
2683  *	 could be enabled.
2684  *	number of vectors which can be enabled otherwise (this number is smaller
2685  *	 than VMXNET3_LINUX_MIN_MSIX_VECT)
2686  */
2687 
2688 static int
2689 vmxnet3_acquire_msix_vectors(struct vmxnet3_adapter *adapter,
2690 			     int vectors)
2691 {
2692 	int err = 0, vector_threshold;
2693 	vector_threshold = VMXNET3_LINUX_MIN_MSIX_VECT;
2694 
2695 	while (vectors >= vector_threshold) {
2696 		err = pci_enable_msix(adapter->pdev, adapter->intr.msix_entries,
2697 				      vectors);
2698 		if (!err) {
2699 			adapter->intr.num_intrs = vectors;
2700 			return 0;
2701 		} else if (err < 0) {
2702 			netdev_err(adapter->netdev,
2703 				   "Failed to enable MSI-X, error: %d\n", err);
2704 			vectors = 0;
2705 		} else if (err < vector_threshold) {
2706 			break;
2707 		} else {
2708 			/* If fails to enable required number of MSI-x vectors
2709 			 * try enabling minimum number of vectors required.
2710 			 */
2711 			netdev_err(adapter->netdev,
2712 				   "Failed to enable %d MSI-X, trying %d instead\n",
2713 				    vectors, vector_threshold);
2714 			vectors = vector_threshold;
2715 		}
2716 	}
2717 
2718 	netdev_info(adapter->netdev,
2719 		    "Number of MSI-X interrupts which can be allocated are lower than min threshold required.\n");
2720 	return err;
2721 }
2722 
2723 
2724 #endif /* CONFIG_PCI_MSI */
2725 
2726 static void
2727 vmxnet3_alloc_intr_resources(struct vmxnet3_adapter *adapter)
2728 {
2729 	u32 cfg;
2730 	unsigned long flags;
2731 
2732 	/* intr settings */
2733 	spin_lock_irqsave(&adapter->cmd_lock, flags);
2734 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2735 			       VMXNET3_CMD_GET_CONF_INTR);
2736 	cfg = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
2737 	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2738 	adapter->intr.type = cfg & 0x3;
2739 	adapter->intr.mask_mode = (cfg >> 2) & 0x3;
2740 
2741 	if (adapter->intr.type == VMXNET3_IT_AUTO) {
2742 		adapter->intr.type = VMXNET3_IT_MSIX;
2743 	}
2744 
2745 #ifdef CONFIG_PCI_MSI
2746 	if (adapter->intr.type == VMXNET3_IT_MSIX) {
2747 		int vector, err = 0;
2748 
2749 		adapter->intr.num_intrs = (adapter->share_intr ==
2750 					   VMXNET3_INTR_TXSHARE) ? 1 :
2751 					   adapter->num_tx_queues;
2752 		adapter->intr.num_intrs += (adapter->share_intr ==
2753 					   VMXNET3_INTR_BUDDYSHARE) ? 0 :
2754 					   adapter->num_rx_queues;
2755 		adapter->intr.num_intrs += 1;		/* for link event */
2756 
2757 		adapter->intr.num_intrs = (adapter->intr.num_intrs >
2758 					   VMXNET3_LINUX_MIN_MSIX_VECT
2759 					   ? adapter->intr.num_intrs :
2760 					   VMXNET3_LINUX_MIN_MSIX_VECT);
2761 
2762 		for (vector = 0; vector < adapter->intr.num_intrs; vector++)
2763 			adapter->intr.msix_entries[vector].entry = vector;
2764 
2765 		err = vmxnet3_acquire_msix_vectors(adapter,
2766 						   adapter->intr.num_intrs);
2767 		/* If we cannot allocate one MSIx vector per queue
2768 		 * then limit the number of rx queues to 1
2769 		 */
2770 		if (err == VMXNET3_LINUX_MIN_MSIX_VECT) {
2771 			if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE
2772 			    || adapter->num_rx_queues != 1) {
2773 				adapter->share_intr = VMXNET3_INTR_TXSHARE;
2774 				printk(KERN_ERR "Number of rx queues : 1\n");
2775 				adapter->num_rx_queues = 1;
2776 				adapter->intr.num_intrs =
2777 						VMXNET3_LINUX_MIN_MSIX_VECT;
2778 			}
2779 			return;
2780 		}
2781 		if (!err)
2782 			return;
2783 
2784 		/* If we cannot allocate MSIx vectors use only one rx queue */
2785 		netdev_info(adapter->netdev,
2786 			    "Failed to enable MSI-X, error %d . Limiting #rx queues to 1, try MSI.\n",
2787 			    err);
2788 
2789 		adapter->intr.type = VMXNET3_IT_MSI;
2790 	}
2791 
2792 	if (adapter->intr.type == VMXNET3_IT_MSI) {
2793 		int err;
2794 		err = pci_enable_msi(adapter->pdev);
2795 		if (!err) {
2796 			adapter->num_rx_queues = 1;
2797 			adapter->intr.num_intrs = 1;
2798 			return;
2799 		}
2800 	}
2801 #endif /* CONFIG_PCI_MSI */
2802 
2803 	adapter->num_rx_queues = 1;
2804 	printk(KERN_INFO "Using INTx interrupt, #Rx queues: 1.\n");
2805 	adapter->intr.type = VMXNET3_IT_INTX;
2806 
2807 	/* INT-X related setting */
2808 	adapter->intr.num_intrs = 1;
2809 }
2810 
2811 
2812 static void
2813 vmxnet3_free_intr_resources(struct vmxnet3_adapter *adapter)
2814 {
2815 	if (adapter->intr.type == VMXNET3_IT_MSIX)
2816 		pci_disable_msix(adapter->pdev);
2817 	else if (adapter->intr.type == VMXNET3_IT_MSI)
2818 		pci_disable_msi(adapter->pdev);
2819 	else
2820 		BUG_ON(adapter->intr.type != VMXNET3_IT_INTX);
2821 }
2822 
2823 
2824 static void
2825 vmxnet3_tx_timeout(struct net_device *netdev)
2826 {
2827 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2828 	adapter->tx_timeout_count++;
2829 
2830 	printk(KERN_ERR "%s: tx hang\n", adapter->netdev->name);
2831 	schedule_work(&adapter->work);
2832 	netif_wake_queue(adapter->netdev);
2833 }
2834 
2835 
2836 static void
2837 vmxnet3_reset_work(struct work_struct *data)
2838 {
2839 	struct vmxnet3_adapter *adapter;
2840 
2841 	adapter = container_of(data, struct vmxnet3_adapter, work);
2842 
2843 	/* if another thread is resetting the device, no need to proceed */
2844 	if (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2845 		return;
2846 
2847 	/* if the device is closed, we must leave it alone */
2848 	rtnl_lock();
2849 	if (netif_running(adapter->netdev)) {
2850 		printk(KERN_INFO "%s: resetting\n", adapter->netdev->name);
2851 		vmxnet3_quiesce_dev(adapter);
2852 		vmxnet3_reset_dev(adapter);
2853 		vmxnet3_activate_dev(adapter);
2854 	} else {
2855 		printk(KERN_INFO "%s: already closed\n", adapter->netdev->name);
2856 	}
2857 	rtnl_unlock();
2858 
2859 	clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2860 }
2861 
2862 
2863 static int __devinit
2864 vmxnet3_probe_device(struct pci_dev *pdev,
2865 		     const struct pci_device_id *id)
2866 {
2867 	static const struct net_device_ops vmxnet3_netdev_ops = {
2868 		.ndo_open = vmxnet3_open,
2869 		.ndo_stop = vmxnet3_close,
2870 		.ndo_start_xmit = vmxnet3_xmit_frame,
2871 		.ndo_set_mac_address = vmxnet3_set_mac_addr,
2872 		.ndo_change_mtu = vmxnet3_change_mtu,
2873 		.ndo_set_features = vmxnet3_set_features,
2874 		.ndo_get_stats64 = vmxnet3_get_stats64,
2875 		.ndo_tx_timeout = vmxnet3_tx_timeout,
2876 		.ndo_set_rx_mode = vmxnet3_set_mc,
2877 		.ndo_vlan_rx_add_vid = vmxnet3_vlan_rx_add_vid,
2878 		.ndo_vlan_rx_kill_vid = vmxnet3_vlan_rx_kill_vid,
2879 #ifdef CONFIG_NET_POLL_CONTROLLER
2880 		.ndo_poll_controller = vmxnet3_netpoll,
2881 #endif
2882 	};
2883 	int err;
2884 	bool dma64 = false; /* stupid gcc */
2885 	u32 ver;
2886 	struct net_device *netdev;
2887 	struct vmxnet3_adapter *adapter;
2888 	u8 mac[ETH_ALEN];
2889 	int size;
2890 	int num_tx_queues;
2891 	int num_rx_queues;
2892 
2893 	if (!pci_msi_enabled())
2894 		enable_mq = 0;
2895 
2896 #ifdef VMXNET3_RSS
2897 	if (enable_mq)
2898 		num_rx_queues = min(VMXNET3_DEVICE_MAX_RX_QUEUES,
2899 				    (int)num_online_cpus());
2900 	else
2901 #endif
2902 		num_rx_queues = 1;
2903 	num_rx_queues = rounddown_pow_of_two(num_rx_queues);
2904 
2905 	if (enable_mq)
2906 		num_tx_queues = min(VMXNET3_DEVICE_MAX_TX_QUEUES,
2907 				    (int)num_online_cpus());
2908 	else
2909 		num_tx_queues = 1;
2910 
2911 	num_tx_queues = rounddown_pow_of_two(num_tx_queues);
2912 	netdev = alloc_etherdev_mq(sizeof(struct vmxnet3_adapter),
2913 				   max(num_tx_queues, num_rx_queues));
2914 	printk(KERN_INFO "# of Tx queues : %d, # of Rx queues : %d\n",
2915 	       num_tx_queues, num_rx_queues);
2916 
2917 	if (!netdev)
2918 		return -ENOMEM;
2919 
2920 	pci_set_drvdata(pdev, netdev);
2921 	adapter = netdev_priv(netdev);
2922 	adapter->netdev = netdev;
2923 	adapter->pdev = pdev;
2924 
2925 	spin_lock_init(&adapter->cmd_lock);
2926 	adapter->shared = pci_alloc_consistent(adapter->pdev,
2927 			  sizeof(struct Vmxnet3_DriverShared),
2928 			  &adapter->shared_pa);
2929 	if (!adapter->shared) {
2930 		printk(KERN_ERR "Failed to allocate memory for %s\n",
2931 			pci_name(pdev));
2932 		err = -ENOMEM;
2933 		goto err_alloc_shared;
2934 	}
2935 
2936 	adapter->num_rx_queues = num_rx_queues;
2937 	adapter->num_tx_queues = num_tx_queues;
2938 
2939 	size = sizeof(struct Vmxnet3_TxQueueDesc) * adapter->num_tx_queues;
2940 	size += sizeof(struct Vmxnet3_RxQueueDesc) * adapter->num_rx_queues;
2941 	adapter->tqd_start = pci_alloc_consistent(adapter->pdev, size,
2942 			     &adapter->queue_desc_pa);
2943 
2944 	if (!adapter->tqd_start) {
2945 		printk(KERN_ERR "Failed to allocate memory for %s\n",
2946 			pci_name(pdev));
2947 		err = -ENOMEM;
2948 		goto err_alloc_queue_desc;
2949 	}
2950 	adapter->rqd_start = (struct Vmxnet3_RxQueueDesc *)(adapter->tqd_start +
2951 							adapter->num_tx_queues);
2952 
2953 	adapter->pm_conf = kmalloc(sizeof(struct Vmxnet3_PMConf), GFP_KERNEL);
2954 	if (adapter->pm_conf == NULL) {
2955 		err = -ENOMEM;
2956 		goto err_alloc_pm;
2957 	}
2958 
2959 #ifdef VMXNET3_RSS
2960 
2961 	adapter->rss_conf = kmalloc(sizeof(struct UPT1_RSSConf), GFP_KERNEL);
2962 	if (adapter->rss_conf == NULL) {
2963 		err = -ENOMEM;
2964 		goto err_alloc_rss;
2965 	}
2966 #endif /* VMXNET3_RSS */
2967 
2968 	err = vmxnet3_alloc_pci_resources(adapter, &dma64);
2969 	if (err < 0)
2970 		goto err_alloc_pci;
2971 
2972 	ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_VRRS);
2973 	if (ver & 1) {
2974 		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_VRRS, 1);
2975 	} else {
2976 		printk(KERN_ERR "Incompatible h/w version (0x%x) for adapter"
2977 		       " %s\n",	ver, pci_name(pdev));
2978 		err = -EBUSY;
2979 		goto err_ver;
2980 	}
2981 
2982 	ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_UVRS);
2983 	if (ver & 1) {
2984 		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_UVRS, 1);
2985 	} else {
2986 		printk(KERN_ERR "Incompatible upt version (0x%x) for "
2987 		       "adapter %s\n", ver, pci_name(pdev));
2988 		err = -EBUSY;
2989 		goto err_ver;
2990 	}
2991 
2992 	SET_NETDEV_DEV(netdev, &pdev->dev);
2993 	vmxnet3_declare_features(adapter, dma64);
2994 
2995 	adapter->dev_number = atomic_read(&devices_found);
2996 
2997 	 adapter->share_intr = irq_share_mode;
2998 	if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE &&
2999 	    adapter->num_tx_queues != adapter->num_rx_queues)
3000 		adapter->share_intr = VMXNET3_INTR_DONTSHARE;
3001 
3002 	vmxnet3_alloc_intr_resources(adapter);
3003 
3004 #ifdef VMXNET3_RSS
3005 	if (adapter->num_rx_queues > 1 &&
3006 	    adapter->intr.type == VMXNET3_IT_MSIX) {
3007 		adapter->rss = true;
3008 		printk(KERN_INFO "RSS is enabled.\n");
3009 	} else {
3010 		adapter->rss = false;
3011 	}
3012 #endif
3013 
3014 	vmxnet3_read_mac_addr(adapter, mac);
3015 	memcpy(netdev->dev_addr,  mac, netdev->addr_len);
3016 
3017 	netdev->netdev_ops = &vmxnet3_netdev_ops;
3018 	vmxnet3_set_ethtool_ops(netdev);
3019 	netdev->watchdog_timeo = 5 * HZ;
3020 
3021 	INIT_WORK(&adapter->work, vmxnet3_reset_work);
3022 
3023 	if (adapter->intr.type == VMXNET3_IT_MSIX) {
3024 		int i;
3025 		for (i = 0; i < adapter->num_rx_queues; i++) {
3026 			netif_napi_add(adapter->netdev,
3027 				       &adapter->rx_queue[i].napi,
3028 				       vmxnet3_poll_rx_only, 64);
3029 		}
3030 	} else {
3031 		netif_napi_add(adapter->netdev, &adapter->rx_queue[0].napi,
3032 			       vmxnet3_poll, 64);
3033 	}
3034 
3035 	netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
3036 	netif_set_real_num_rx_queues(adapter->netdev, adapter->num_rx_queues);
3037 
3038 	err = register_netdev(netdev);
3039 
3040 	if (err) {
3041 		printk(KERN_ERR "Failed to register adapter %s\n",
3042 			pci_name(pdev));
3043 		goto err_register;
3044 	}
3045 
3046 	set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
3047 	vmxnet3_check_link(adapter, false);
3048 	atomic_inc(&devices_found);
3049 	return 0;
3050 
3051 err_register:
3052 	vmxnet3_free_intr_resources(adapter);
3053 err_ver:
3054 	vmxnet3_free_pci_resources(adapter);
3055 err_alloc_pci:
3056 #ifdef VMXNET3_RSS
3057 	kfree(adapter->rss_conf);
3058 err_alloc_rss:
3059 #endif
3060 	kfree(adapter->pm_conf);
3061 err_alloc_pm:
3062 	pci_free_consistent(adapter->pdev, size, adapter->tqd_start,
3063 			    adapter->queue_desc_pa);
3064 err_alloc_queue_desc:
3065 	pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_DriverShared),
3066 			    adapter->shared, adapter->shared_pa);
3067 err_alloc_shared:
3068 	pci_set_drvdata(pdev, NULL);
3069 	free_netdev(netdev);
3070 	return err;
3071 }
3072 
3073 
3074 static void __devexit
3075 vmxnet3_remove_device(struct pci_dev *pdev)
3076 {
3077 	struct net_device *netdev = pci_get_drvdata(pdev);
3078 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3079 	int size = 0;
3080 	int num_rx_queues;
3081 
3082 #ifdef VMXNET3_RSS
3083 	if (enable_mq)
3084 		num_rx_queues = min(VMXNET3_DEVICE_MAX_RX_QUEUES,
3085 				    (int)num_online_cpus());
3086 	else
3087 #endif
3088 		num_rx_queues = 1;
3089 	num_rx_queues = rounddown_pow_of_two(num_rx_queues);
3090 
3091 	cancel_work_sync(&adapter->work);
3092 
3093 	unregister_netdev(netdev);
3094 
3095 	vmxnet3_free_intr_resources(adapter);
3096 	vmxnet3_free_pci_resources(adapter);
3097 #ifdef VMXNET3_RSS
3098 	kfree(adapter->rss_conf);
3099 #endif
3100 	kfree(adapter->pm_conf);
3101 
3102 	size = sizeof(struct Vmxnet3_TxQueueDesc) * adapter->num_tx_queues;
3103 	size += sizeof(struct Vmxnet3_RxQueueDesc) * num_rx_queues;
3104 	pci_free_consistent(adapter->pdev, size, adapter->tqd_start,
3105 			    adapter->queue_desc_pa);
3106 	pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_DriverShared),
3107 			    adapter->shared, adapter->shared_pa);
3108 	free_netdev(netdev);
3109 }
3110 
3111 
3112 #ifdef CONFIG_PM
3113 
3114 static int
3115 vmxnet3_suspend(struct device *device)
3116 {
3117 	struct pci_dev *pdev = to_pci_dev(device);
3118 	struct net_device *netdev = pci_get_drvdata(pdev);
3119 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3120 	struct Vmxnet3_PMConf *pmConf;
3121 	struct ethhdr *ehdr;
3122 	struct arphdr *ahdr;
3123 	u8 *arpreq;
3124 	struct in_device *in_dev;
3125 	struct in_ifaddr *ifa;
3126 	unsigned long flags;
3127 	int i = 0;
3128 
3129 	if (!netif_running(netdev))
3130 		return 0;
3131 
3132 	for (i = 0; i < adapter->num_rx_queues; i++)
3133 		napi_disable(&adapter->rx_queue[i].napi);
3134 
3135 	vmxnet3_disable_all_intrs(adapter);
3136 	vmxnet3_free_irqs(adapter);
3137 	vmxnet3_free_intr_resources(adapter);
3138 
3139 	netif_device_detach(netdev);
3140 	netif_tx_stop_all_queues(netdev);
3141 
3142 	/* Create wake-up filters. */
3143 	pmConf = adapter->pm_conf;
3144 	memset(pmConf, 0, sizeof(*pmConf));
3145 
3146 	if (adapter->wol & WAKE_UCAST) {
3147 		pmConf->filters[i].patternSize = ETH_ALEN;
3148 		pmConf->filters[i].maskSize = 1;
3149 		memcpy(pmConf->filters[i].pattern, netdev->dev_addr, ETH_ALEN);
3150 		pmConf->filters[i].mask[0] = 0x3F; /* LSB ETH_ALEN bits */
3151 
3152 		pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_FILTER;
3153 		i++;
3154 	}
3155 
3156 	if (adapter->wol & WAKE_ARP) {
3157 		in_dev = in_dev_get(netdev);
3158 		if (!in_dev)
3159 			goto skip_arp;
3160 
3161 		ifa = (struct in_ifaddr *)in_dev->ifa_list;
3162 		if (!ifa)
3163 			goto skip_arp;
3164 
3165 		pmConf->filters[i].patternSize = ETH_HLEN + /* Ethernet header*/
3166 			sizeof(struct arphdr) +		/* ARP header */
3167 			2 * ETH_ALEN +		/* 2 Ethernet addresses*/
3168 			2 * sizeof(u32);	/*2 IPv4 addresses */
3169 		pmConf->filters[i].maskSize =
3170 			(pmConf->filters[i].patternSize - 1) / 8 + 1;
3171 
3172 		/* ETH_P_ARP in Ethernet header. */
3173 		ehdr = (struct ethhdr *)pmConf->filters[i].pattern;
3174 		ehdr->h_proto = htons(ETH_P_ARP);
3175 
3176 		/* ARPOP_REQUEST in ARP header. */
3177 		ahdr = (struct arphdr *)&pmConf->filters[i].pattern[ETH_HLEN];
3178 		ahdr->ar_op = htons(ARPOP_REQUEST);
3179 		arpreq = (u8 *)(ahdr + 1);
3180 
3181 		/* The Unicast IPv4 address in 'tip' field. */
3182 		arpreq += 2 * ETH_ALEN + sizeof(u32);
3183 		*(u32 *)arpreq = ifa->ifa_address;
3184 
3185 		/* The mask for the relevant bits. */
3186 		pmConf->filters[i].mask[0] = 0x00;
3187 		pmConf->filters[i].mask[1] = 0x30; /* ETH_P_ARP */
3188 		pmConf->filters[i].mask[2] = 0x30; /* ARPOP_REQUEST */
3189 		pmConf->filters[i].mask[3] = 0x00;
3190 		pmConf->filters[i].mask[4] = 0xC0; /* IPv4 TIP */
3191 		pmConf->filters[i].mask[5] = 0x03; /* IPv4 TIP */
3192 		in_dev_put(in_dev);
3193 
3194 		pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_FILTER;
3195 		i++;
3196 	}
3197 
3198 skip_arp:
3199 	if (adapter->wol & WAKE_MAGIC)
3200 		pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_MAGIC;
3201 
3202 	pmConf->numFilters = i;
3203 
3204 	adapter->shared->devRead.pmConfDesc.confVer = cpu_to_le32(1);
3205 	adapter->shared->devRead.pmConfDesc.confLen = cpu_to_le32(sizeof(
3206 								  *pmConf));
3207 	adapter->shared->devRead.pmConfDesc.confPA = cpu_to_le64(virt_to_phys(
3208 								 pmConf));
3209 
3210 	spin_lock_irqsave(&adapter->cmd_lock, flags);
3211 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3212 			       VMXNET3_CMD_UPDATE_PMCFG);
3213 	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3214 
3215 	pci_save_state(pdev);
3216 	pci_enable_wake(pdev, pci_choose_state(pdev, PMSG_SUSPEND),
3217 			adapter->wol);
3218 	pci_disable_device(pdev);
3219 	pci_set_power_state(pdev, pci_choose_state(pdev, PMSG_SUSPEND));
3220 
3221 	return 0;
3222 }
3223 
3224 
3225 static int
3226 vmxnet3_resume(struct device *device)
3227 {
3228 	int err, i = 0;
3229 	unsigned long flags;
3230 	struct pci_dev *pdev = to_pci_dev(device);
3231 	struct net_device *netdev = pci_get_drvdata(pdev);
3232 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3233 	struct Vmxnet3_PMConf *pmConf;
3234 
3235 	if (!netif_running(netdev))
3236 		return 0;
3237 
3238 	/* Destroy wake-up filters. */
3239 	pmConf = adapter->pm_conf;
3240 	memset(pmConf, 0, sizeof(*pmConf));
3241 
3242 	adapter->shared->devRead.pmConfDesc.confVer = cpu_to_le32(1);
3243 	adapter->shared->devRead.pmConfDesc.confLen = cpu_to_le32(sizeof(
3244 								  *pmConf));
3245 	adapter->shared->devRead.pmConfDesc.confPA = cpu_to_le64(virt_to_phys(
3246 								 pmConf));
3247 
3248 	netif_device_attach(netdev);
3249 	pci_set_power_state(pdev, PCI_D0);
3250 	pci_restore_state(pdev);
3251 	err = pci_enable_device_mem(pdev);
3252 	if (err != 0)
3253 		return err;
3254 
3255 	pci_enable_wake(pdev, PCI_D0, 0);
3256 
3257 	spin_lock_irqsave(&adapter->cmd_lock, flags);
3258 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3259 			       VMXNET3_CMD_UPDATE_PMCFG);
3260 	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3261 	vmxnet3_alloc_intr_resources(adapter);
3262 	vmxnet3_request_irqs(adapter);
3263 	for (i = 0; i < adapter->num_rx_queues; i++)
3264 		napi_enable(&adapter->rx_queue[i].napi);
3265 	vmxnet3_enable_all_intrs(adapter);
3266 
3267 	return 0;
3268 }
3269 
3270 static const struct dev_pm_ops vmxnet3_pm_ops = {
3271 	.suspend = vmxnet3_suspend,
3272 	.resume = vmxnet3_resume,
3273 };
3274 #endif
3275 
3276 static struct pci_driver vmxnet3_driver = {
3277 	.name		= vmxnet3_driver_name,
3278 	.id_table	= vmxnet3_pciid_table,
3279 	.probe		= vmxnet3_probe_device,
3280 	.remove		= __devexit_p(vmxnet3_remove_device),
3281 #ifdef CONFIG_PM
3282 	.driver.pm	= &vmxnet3_pm_ops,
3283 #endif
3284 };
3285 
3286 
3287 static int __init
3288 vmxnet3_init_module(void)
3289 {
3290 	printk(KERN_INFO "%s - version %s\n", VMXNET3_DRIVER_DESC,
3291 		VMXNET3_DRIVER_VERSION_REPORT);
3292 	return pci_register_driver(&vmxnet3_driver);
3293 }
3294 
3295 module_init(vmxnet3_init_module);
3296 
3297 
3298 static void
3299 vmxnet3_exit_module(void)
3300 {
3301 	pci_unregister_driver(&vmxnet3_driver);
3302 }
3303 
3304 module_exit(vmxnet3_exit_module);
3305 
3306 MODULE_AUTHOR("VMware, Inc.");
3307 MODULE_DESCRIPTION(VMXNET3_DRIVER_DESC);
3308 MODULE_LICENSE("GPL v2");
3309 MODULE_VERSION(VMXNET3_DRIVER_VERSION_STRING);
3310