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 <net/ip6_checksum.h>
28 
29 #include "vmxnet3_int.h"
30 
31 char vmxnet3_driver_name[] = "vmxnet3";
32 #define VMXNET3_DRIVER_DESC "VMware vmxnet3 virtual NIC driver"
33 
34 /*
35  * PCI Device ID Table
36  * Last entry must be all 0s
37  */
38 static const struct pci_device_id vmxnet3_pciid_table[] = {
39 	{PCI_VDEVICE(VMWARE, PCI_DEVICE_ID_VMWARE_VMXNET3)},
40 	{0}
41 };
42 
43 MODULE_DEVICE_TABLE(pci, vmxnet3_pciid_table);
44 
45 static atomic_t devices_found;
46 
47 
48 /*
49  *    Enable/Disable the given intr
50  */
51 static void
52 vmxnet3_enable_intr(struct vmxnet3_adapter *adapter, unsigned intr_idx)
53 {
54 	VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_IMR + intr_idx * 8, 0);
55 }
56 
57 
58 static void
59 vmxnet3_disable_intr(struct vmxnet3_adapter *adapter, unsigned intr_idx)
60 {
61 	VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_IMR + intr_idx * 8, 1);
62 }
63 
64 
65 /*
66  *    Enable/Disable all intrs used by the device
67  */
68 static void
69 vmxnet3_enable_all_intrs(struct vmxnet3_adapter *adapter)
70 {
71 	int i;
72 
73 	for (i = 0; i < adapter->intr.num_intrs; i++)
74 		vmxnet3_enable_intr(adapter, i);
75 }
76 
77 
78 static void
79 vmxnet3_disable_all_intrs(struct vmxnet3_adapter *adapter)
80 {
81 	int i;
82 
83 	for (i = 0; i < adapter->intr.num_intrs; i++)
84 		vmxnet3_disable_intr(adapter, i);
85 }
86 
87 
88 static void
89 vmxnet3_ack_events(struct vmxnet3_adapter *adapter, u32 events)
90 {
91 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_ECR, events);
92 }
93 
94 
95 static bool
96 vmxnet3_tq_stopped(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
97 {
98 	return netif_queue_stopped(adapter->netdev);
99 }
100 
101 
102 static void
103 vmxnet3_tq_start(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
104 {
105 	tq->stopped = false;
106 	netif_start_queue(adapter->netdev);
107 }
108 
109 
110 static void
111 vmxnet3_tq_wake(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
112 {
113 	tq->stopped = false;
114 	netif_wake_queue(adapter->netdev);
115 }
116 
117 
118 static void
119 vmxnet3_tq_stop(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
120 {
121 	tq->stopped = true;
122 	tq->num_stop++;
123 	netif_stop_queue(adapter->netdev);
124 }
125 
126 
127 /*
128  * Check the link state. This may start or stop the tx queue.
129  */
130 static void
131 vmxnet3_check_link(struct vmxnet3_adapter *adapter)
132 {
133 	u32 ret;
134 
135 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, VMXNET3_CMD_GET_LINK);
136 	ret = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
137 	adapter->link_speed = ret >> 16;
138 	if (ret & 1) { /* Link is up. */
139 		printk(KERN_INFO "%s: NIC Link is Up %d Mbps\n",
140 		       adapter->netdev->name, adapter->link_speed);
141 		if (!netif_carrier_ok(adapter->netdev))
142 			netif_carrier_on(adapter->netdev);
143 
144 		vmxnet3_tq_start(&adapter->tx_queue, adapter);
145 	} else {
146 		printk(KERN_INFO "%s: NIC Link is Down\n",
147 		       adapter->netdev->name);
148 		if (netif_carrier_ok(adapter->netdev))
149 			netif_carrier_off(adapter->netdev);
150 
151 		vmxnet3_tq_stop(&adapter->tx_queue, adapter);
152 	}
153 }
154 
155 static void
156 vmxnet3_process_events(struct vmxnet3_adapter *adapter)
157 {
158 	u32 events = le32_to_cpu(adapter->shared->ecr);
159 	if (!events)
160 		return;
161 
162 	vmxnet3_ack_events(adapter, events);
163 
164 	/* Check if link state has changed */
165 	if (events & VMXNET3_ECR_LINK)
166 		vmxnet3_check_link(adapter);
167 
168 	/* Check if there is an error on xmit/recv queues */
169 	if (events & (VMXNET3_ECR_TQERR | VMXNET3_ECR_RQERR)) {
170 		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
171 				       VMXNET3_CMD_GET_QUEUE_STATUS);
172 
173 		if (adapter->tqd_start->status.stopped) {
174 			printk(KERN_ERR "%s: tq error 0x%x\n",
175 			       adapter->netdev->name,
176 			       le32_to_cpu(adapter->tqd_start->status.error));
177 		}
178 		if (adapter->rqd_start->status.stopped) {
179 			printk(KERN_ERR "%s: rq error 0x%x\n",
180 			       adapter->netdev->name,
181 			       adapter->rqd_start->status.error);
182 		}
183 
184 		schedule_work(&adapter->work);
185 	}
186 }
187 
188 #ifdef __BIG_ENDIAN_BITFIELD
189 /*
190  * The device expects the bitfields in shared structures to be written in
191  * little endian. When CPU is big endian, the following routines are used to
192  * correctly read and write into ABI.
193  * The general technique used here is : double word bitfields are defined in
194  * opposite order for big endian architecture. Then before reading them in
195  * driver the complete double word is translated using le32_to_cpu. Similarly
196  * After the driver writes into bitfields, cpu_to_le32 is used to translate the
197  * double words into required format.
198  * In order to avoid touching bits in shared structure more than once, temporary
199  * descriptors are used. These are passed as srcDesc to following functions.
200  */
201 static void vmxnet3_RxDescToCPU(const struct Vmxnet3_RxDesc *srcDesc,
202 				struct Vmxnet3_RxDesc *dstDesc)
203 {
204 	u32 *src = (u32 *)srcDesc + 2;
205 	u32 *dst = (u32 *)dstDesc + 2;
206 	dstDesc->addr = le64_to_cpu(srcDesc->addr);
207 	*dst = le32_to_cpu(*src);
208 	dstDesc->ext1 = le32_to_cpu(srcDesc->ext1);
209 }
210 
211 static void vmxnet3_TxDescToLe(const struct Vmxnet3_TxDesc *srcDesc,
212 			       struct Vmxnet3_TxDesc *dstDesc)
213 {
214 	int i;
215 	u32 *src = (u32 *)(srcDesc + 1);
216 	u32 *dst = (u32 *)(dstDesc + 1);
217 
218 	/* Working backwards so that the gen bit is set at the end. */
219 	for (i = 2; i > 0; i--) {
220 		src--;
221 		dst--;
222 		*dst = cpu_to_le32(*src);
223 	}
224 }
225 
226 
227 static void vmxnet3_RxCompToCPU(const struct Vmxnet3_RxCompDesc *srcDesc,
228 				struct Vmxnet3_RxCompDesc *dstDesc)
229 {
230 	int i = 0;
231 	u32 *src = (u32 *)srcDesc;
232 	u32 *dst = (u32 *)dstDesc;
233 	for (i = 0; i < sizeof(struct Vmxnet3_RxCompDesc) / sizeof(u32); i++) {
234 		*dst = le32_to_cpu(*src);
235 		src++;
236 		dst++;
237 	}
238 }
239 
240 
241 /* Used to read bitfield values from double words. */
242 static u32 get_bitfield32(const __le32 *bitfield, u32 pos, u32 size)
243 {
244 	u32 temp = le32_to_cpu(*bitfield);
245 	u32 mask = ((1 << size) - 1) << pos;
246 	temp &= mask;
247 	temp >>= pos;
248 	return temp;
249 }
250 
251 
252 
253 #endif  /* __BIG_ENDIAN_BITFIELD */
254 
255 #ifdef __BIG_ENDIAN_BITFIELD
256 
257 #   define VMXNET3_TXDESC_GET_GEN(txdesc) get_bitfield32(((const __le32 *) \
258 			txdesc) + VMXNET3_TXD_GEN_DWORD_SHIFT, \
259 			VMXNET3_TXD_GEN_SHIFT, VMXNET3_TXD_GEN_SIZE)
260 #   define VMXNET3_TXDESC_GET_EOP(txdesc) get_bitfield32(((const __le32 *) \
261 			txdesc) + VMXNET3_TXD_EOP_DWORD_SHIFT, \
262 			VMXNET3_TXD_EOP_SHIFT, VMXNET3_TXD_EOP_SIZE)
263 #   define VMXNET3_TCD_GET_GEN(tcd) get_bitfield32(((const __le32 *)tcd) + \
264 			VMXNET3_TCD_GEN_DWORD_SHIFT, VMXNET3_TCD_GEN_SHIFT, \
265 			VMXNET3_TCD_GEN_SIZE)
266 #   define VMXNET3_TCD_GET_TXIDX(tcd) get_bitfield32((const __le32 *)tcd, \
267 			VMXNET3_TCD_TXIDX_SHIFT, VMXNET3_TCD_TXIDX_SIZE)
268 #   define vmxnet3_getRxComp(dstrcd, rcd, tmp) do { \
269 			(dstrcd) = (tmp); \
270 			vmxnet3_RxCompToCPU((rcd), (tmp)); \
271 		} while (0)
272 #   define vmxnet3_getRxDesc(dstrxd, rxd, tmp) do { \
273 			(dstrxd) = (tmp); \
274 			vmxnet3_RxDescToCPU((rxd), (tmp)); \
275 		} while (0)
276 
277 #else
278 
279 #   define VMXNET3_TXDESC_GET_GEN(txdesc) ((txdesc)->gen)
280 #   define VMXNET3_TXDESC_GET_EOP(txdesc) ((txdesc)->eop)
281 #   define VMXNET3_TCD_GET_GEN(tcd) ((tcd)->gen)
282 #   define VMXNET3_TCD_GET_TXIDX(tcd) ((tcd)->txdIdx)
283 #   define vmxnet3_getRxComp(dstrcd, rcd, tmp) (dstrcd) = (rcd)
284 #   define vmxnet3_getRxDesc(dstrxd, rxd, tmp) (dstrxd) = (rxd)
285 
286 #endif /* __BIG_ENDIAN_BITFIELD  */
287 
288 
289 static void
290 vmxnet3_unmap_tx_buf(struct vmxnet3_tx_buf_info *tbi,
291 		     struct pci_dev *pdev)
292 {
293 	if (tbi->map_type == VMXNET3_MAP_SINGLE)
294 		pci_unmap_single(pdev, tbi->dma_addr, tbi->len,
295 				 PCI_DMA_TODEVICE);
296 	else if (tbi->map_type == VMXNET3_MAP_PAGE)
297 		pci_unmap_page(pdev, tbi->dma_addr, tbi->len,
298 			       PCI_DMA_TODEVICE);
299 	else
300 		BUG_ON(tbi->map_type != VMXNET3_MAP_NONE);
301 
302 	tbi->map_type = VMXNET3_MAP_NONE; /* to help debugging */
303 }
304 
305 
306 static int
307 vmxnet3_unmap_pkt(u32 eop_idx, struct vmxnet3_tx_queue *tq,
308 		  struct pci_dev *pdev,	struct vmxnet3_adapter *adapter)
309 {
310 	struct sk_buff *skb;
311 	int entries = 0;
312 
313 	/* no out of order completion */
314 	BUG_ON(tq->buf_info[eop_idx].sop_idx != tq->tx_ring.next2comp);
315 	BUG_ON(VMXNET3_TXDESC_GET_EOP(&(tq->tx_ring.base[eop_idx].txd)) != 1);
316 
317 	skb = tq->buf_info[eop_idx].skb;
318 	BUG_ON(skb == NULL);
319 	tq->buf_info[eop_idx].skb = NULL;
320 
321 	VMXNET3_INC_RING_IDX_ONLY(eop_idx, tq->tx_ring.size);
322 
323 	while (tq->tx_ring.next2comp != eop_idx) {
324 		vmxnet3_unmap_tx_buf(tq->buf_info + tq->tx_ring.next2comp,
325 				     pdev);
326 
327 		/* update next2comp w/o tx_lock. Since we are marking more,
328 		 * instead of less, tx ring entries avail, the worst case is
329 		 * that the tx routine incorrectly re-queues a pkt due to
330 		 * insufficient tx ring entries.
331 		 */
332 		vmxnet3_cmd_ring_adv_next2comp(&tq->tx_ring);
333 		entries++;
334 	}
335 
336 	dev_kfree_skb_any(skb);
337 	return entries;
338 }
339 
340 
341 static int
342 vmxnet3_tq_tx_complete(struct vmxnet3_tx_queue *tq,
343 			struct vmxnet3_adapter *adapter)
344 {
345 	int completed = 0;
346 	union Vmxnet3_GenericDesc *gdesc;
347 
348 	gdesc = tq->comp_ring.base + tq->comp_ring.next2proc;
349 	while (VMXNET3_TCD_GET_GEN(&gdesc->tcd) == tq->comp_ring.gen) {
350 		completed += vmxnet3_unmap_pkt(VMXNET3_TCD_GET_TXIDX(
351 					       &gdesc->tcd), tq, adapter->pdev,
352 					       adapter);
353 
354 		vmxnet3_comp_ring_adv_next2proc(&tq->comp_ring);
355 		gdesc = tq->comp_ring.base + tq->comp_ring.next2proc;
356 	}
357 
358 	if (completed) {
359 		spin_lock(&tq->tx_lock);
360 		if (unlikely(vmxnet3_tq_stopped(tq, adapter) &&
361 			     vmxnet3_cmd_ring_desc_avail(&tq->tx_ring) >
362 			     VMXNET3_WAKE_QUEUE_THRESHOLD(tq) &&
363 			     netif_carrier_ok(adapter->netdev))) {
364 			vmxnet3_tq_wake(tq, adapter);
365 		}
366 		spin_unlock(&tq->tx_lock);
367 	}
368 	return completed;
369 }
370 
371 
372 static void
373 vmxnet3_tq_cleanup(struct vmxnet3_tx_queue *tq,
374 		   struct vmxnet3_adapter *adapter)
375 {
376 	int i;
377 
378 	while (tq->tx_ring.next2comp != tq->tx_ring.next2fill) {
379 		struct vmxnet3_tx_buf_info *tbi;
380 		union Vmxnet3_GenericDesc *gdesc;
381 
382 		tbi = tq->buf_info + tq->tx_ring.next2comp;
383 		gdesc = tq->tx_ring.base + tq->tx_ring.next2comp;
384 
385 		vmxnet3_unmap_tx_buf(tbi, adapter->pdev);
386 		if (tbi->skb) {
387 			dev_kfree_skb_any(tbi->skb);
388 			tbi->skb = NULL;
389 		}
390 		vmxnet3_cmd_ring_adv_next2comp(&tq->tx_ring);
391 	}
392 
393 	/* sanity check, verify all buffers are indeed unmapped and freed */
394 	for (i = 0; i < tq->tx_ring.size; i++) {
395 		BUG_ON(tq->buf_info[i].skb != NULL ||
396 		       tq->buf_info[i].map_type != VMXNET3_MAP_NONE);
397 	}
398 
399 	tq->tx_ring.gen = VMXNET3_INIT_GEN;
400 	tq->tx_ring.next2fill = tq->tx_ring.next2comp = 0;
401 
402 	tq->comp_ring.gen = VMXNET3_INIT_GEN;
403 	tq->comp_ring.next2proc = 0;
404 }
405 
406 
407 void
408 vmxnet3_tq_destroy(struct vmxnet3_tx_queue *tq,
409 		   struct vmxnet3_adapter *adapter)
410 {
411 	if (tq->tx_ring.base) {
412 		pci_free_consistent(adapter->pdev, tq->tx_ring.size *
413 				    sizeof(struct Vmxnet3_TxDesc),
414 				    tq->tx_ring.base, tq->tx_ring.basePA);
415 		tq->tx_ring.base = NULL;
416 	}
417 	if (tq->data_ring.base) {
418 		pci_free_consistent(adapter->pdev, tq->data_ring.size *
419 				    sizeof(struct Vmxnet3_TxDataDesc),
420 				    tq->data_ring.base, tq->data_ring.basePA);
421 		tq->data_ring.base = NULL;
422 	}
423 	if (tq->comp_ring.base) {
424 		pci_free_consistent(adapter->pdev, tq->comp_ring.size *
425 				    sizeof(struct Vmxnet3_TxCompDesc),
426 				    tq->comp_ring.base, tq->comp_ring.basePA);
427 		tq->comp_ring.base = NULL;
428 	}
429 	kfree(tq->buf_info);
430 	tq->buf_info = NULL;
431 }
432 
433 
434 static void
435 vmxnet3_tq_init(struct vmxnet3_tx_queue *tq,
436 		struct vmxnet3_adapter *adapter)
437 {
438 	int i;
439 
440 	/* reset the tx ring contents to 0 and reset the tx ring states */
441 	memset(tq->tx_ring.base, 0, tq->tx_ring.size *
442 	       sizeof(struct Vmxnet3_TxDesc));
443 	tq->tx_ring.next2fill = tq->tx_ring.next2comp = 0;
444 	tq->tx_ring.gen = VMXNET3_INIT_GEN;
445 
446 	memset(tq->data_ring.base, 0, tq->data_ring.size *
447 	       sizeof(struct Vmxnet3_TxDataDesc));
448 
449 	/* reset the tx comp ring contents to 0 and reset comp ring states */
450 	memset(tq->comp_ring.base, 0, tq->comp_ring.size *
451 	       sizeof(struct Vmxnet3_TxCompDesc));
452 	tq->comp_ring.next2proc = 0;
453 	tq->comp_ring.gen = VMXNET3_INIT_GEN;
454 
455 	/* reset the bookkeeping data */
456 	memset(tq->buf_info, 0, sizeof(tq->buf_info[0]) * tq->tx_ring.size);
457 	for (i = 0; i < tq->tx_ring.size; i++)
458 		tq->buf_info[i].map_type = VMXNET3_MAP_NONE;
459 
460 	/* stats are not reset */
461 }
462 
463 
464 static int
465 vmxnet3_tq_create(struct vmxnet3_tx_queue *tq,
466 		  struct vmxnet3_adapter *adapter)
467 {
468 	BUG_ON(tq->tx_ring.base || tq->data_ring.base ||
469 	       tq->comp_ring.base || tq->buf_info);
470 
471 	tq->tx_ring.base = pci_alloc_consistent(adapter->pdev, tq->tx_ring.size
472 			   * sizeof(struct Vmxnet3_TxDesc),
473 			   &tq->tx_ring.basePA);
474 	if (!tq->tx_ring.base) {
475 		printk(KERN_ERR "%s: failed to allocate tx ring\n",
476 		       adapter->netdev->name);
477 		goto err;
478 	}
479 
480 	tq->data_ring.base = pci_alloc_consistent(adapter->pdev,
481 			     tq->data_ring.size *
482 			     sizeof(struct Vmxnet3_TxDataDesc),
483 			     &tq->data_ring.basePA);
484 	if (!tq->data_ring.base) {
485 		printk(KERN_ERR "%s: failed to allocate data ring\n",
486 		       adapter->netdev->name);
487 		goto err;
488 	}
489 
490 	tq->comp_ring.base = pci_alloc_consistent(adapter->pdev,
491 			     tq->comp_ring.size *
492 			     sizeof(struct Vmxnet3_TxCompDesc),
493 			     &tq->comp_ring.basePA);
494 	if (!tq->comp_ring.base) {
495 		printk(KERN_ERR "%s: failed to allocate tx comp ring\n",
496 		       adapter->netdev->name);
497 		goto err;
498 	}
499 
500 	tq->buf_info = kcalloc(tq->tx_ring.size, sizeof(tq->buf_info[0]),
501 			       GFP_KERNEL);
502 	if (!tq->buf_info) {
503 		printk(KERN_ERR "%s: failed to allocate tx bufinfo\n",
504 		       adapter->netdev->name);
505 		goto err;
506 	}
507 
508 	return 0;
509 
510 err:
511 	vmxnet3_tq_destroy(tq, adapter);
512 	return -ENOMEM;
513 }
514 
515 
516 /*
517  *    starting from ring->next2fill, allocate rx buffers for the given ring
518  *    of the rx queue and update the rx desc. stop after @num_to_alloc buffers
519  *    are allocated or allocation fails
520  */
521 
522 static int
523 vmxnet3_rq_alloc_rx_buf(struct vmxnet3_rx_queue *rq, u32 ring_idx,
524 			int num_to_alloc, struct vmxnet3_adapter *adapter)
525 {
526 	int num_allocated = 0;
527 	struct vmxnet3_rx_buf_info *rbi_base = rq->buf_info[ring_idx];
528 	struct vmxnet3_cmd_ring *ring = &rq->rx_ring[ring_idx];
529 	u32 val;
530 
531 	while (num_allocated < num_to_alloc) {
532 		struct vmxnet3_rx_buf_info *rbi;
533 		union Vmxnet3_GenericDesc *gd;
534 
535 		rbi = rbi_base + ring->next2fill;
536 		gd = ring->base + ring->next2fill;
537 
538 		if (rbi->buf_type == VMXNET3_RX_BUF_SKB) {
539 			if (rbi->skb == NULL) {
540 				rbi->skb = dev_alloc_skb(rbi->len +
541 							 NET_IP_ALIGN);
542 				if (unlikely(rbi->skb == NULL)) {
543 					rq->stats.rx_buf_alloc_failure++;
544 					break;
545 				}
546 				rbi->skb->dev = adapter->netdev;
547 
548 				skb_reserve(rbi->skb, NET_IP_ALIGN);
549 				rbi->dma_addr = pci_map_single(adapter->pdev,
550 						rbi->skb->data, rbi->len,
551 						PCI_DMA_FROMDEVICE);
552 			} else {
553 				/* rx buffer skipped by the device */
554 			}
555 			val = VMXNET3_RXD_BTYPE_HEAD << VMXNET3_RXD_BTYPE_SHIFT;
556 		} else {
557 			BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_PAGE ||
558 			       rbi->len  != PAGE_SIZE);
559 
560 			if (rbi->page == NULL) {
561 				rbi->page = alloc_page(GFP_ATOMIC);
562 				if (unlikely(rbi->page == NULL)) {
563 					rq->stats.rx_buf_alloc_failure++;
564 					break;
565 				}
566 				rbi->dma_addr = pci_map_page(adapter->pdev,
567 						rbi->page, 0, PAGE_SIZE,
568 						PCI_DMA_FROMDEVICE);
569 			} else {
570 				/* rx buffers skipped by the device */
571 			}
572 			val = VMXNET3_RXD_BTYPE_BODY << VMXNET3_RXD_BTYPE_SHIFT;
573 		}
574 
575 		BUG_ON(rbi->dma_addr == 0);
576 		gd->rxd.addr = cpu_to_le64(rbi->dma_addr);
577 		gd->dword[2] = cpu_to_le32((ring->gen << VMXNET3_RXD_GEN_SHIFT)
578 					   | val | rbi->len);
579 
580 		num_allocated++;
581 		vmxnet3_cmd_ring_adv_next2fill(ring);
582 	}
583 	rq->uncommitted[ring_idx] += num_allocated;
584 
585 	dev_dbg(&adapter->netdev->dev,
586 		"alloc_rx_buf: %d allocated, next2fill %u, next2comp "
587 		"%u, uncommited %u\n", num_allocated, ring->next2fill,
588 		ring->next2comp, rq->uncommitted[ring_idx]);
589 
590 	/* so that the device can distinguish a full ring and an empty ring */
591 	BUG_ON(num_allocated != 0 && ring->next2fill == ring->next2comp);
592 
593 	return num_allocated;
594 }
595 
596 
597 static void
598 vmxnet3_append_frag(struct sk_buff *skb, struct Vmxnet3_RxCompDesc *rcd,
599 		    struct vmxnet3_rx_buf_info *rbi)
600 {
601 	struct skb_frag_struct *frag = skb_shinfo(skb)->frags +
602 		skb_shinfo(skb)->nr_frags;
603 
604 	BUG_ON(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS);
605 
606 	frag->page = rbi->page;
607 	frag->page_offset = 0;
608 	frag->size = rcd->len;
609 	skb->data_len += frag->size;
610 	skb_shinfo(skb)->nr_frags++;
611 }
612 
613 
614 static void
615 vmxnet3_map_pkt(struct sk_buff *skb, struct vmxnet3_tx_ctx *ctx,
616 		struct vmxnet3_tx_queue *tq, struct pci_dev *pdev,
617 		struct vmxnet3_adapter *adapter)
618 {
619 	u32 dw2, len;
620 	unsigned long buf_offset;
621 	int i;
622 	union Vmxnet3_GenericDesc *gdesc;
623 	struct vmxnet3_tx_buf_info *tbi = NULL;
624 
625 	BUG_ON(ctx->copy_size > skb_headlen(skb));
626 
627 	/* use the previous gen bit for the SOP desc */
628 	dw2 = (tq->tx_ring.gen ^ 0x1) << VMXNET3_TXD_GEN_SHIFT;
629 
630 	ctx->sop_txd = tq->tx_ring.base + tq->tx_ring.next2fill;
631 	gdesc = ctx->sop_txd; /* both loops below can be skipped */
632 
633 	/* no need to map the buffer if headers are copied */
634 	if (ctx->copy_size) {
635 		ctx->sop_txd->txd.addr = cpu_to_le64(tq->data_ring.basePA +
636 					tq->tx_ring.next2fill *
637 					sizeof(struct Vmxnet3_TxDataDesc));
638 		ctx->sop_txd->dword[2] = cpu_to_le32(dw2 | ctx->copy_size);
639 		ctx->sop_txd->dword[3] = 0;
640 
641 		tbi = tq->buf_info + tq->tx_ring.next2fill;
642 		tbi->map_type = VMXNET3_MAP_NONE;
643 
644 		dev_dbg(&adapter->netdev->dev,
645 			"txd[%u]: 0x%Lx 0x%x 0x%x\n",
646 			tq->tx_ring.next2fill,
647 			le64_to_cpu(ctx->sop_txd->txd.addr),
648 			ctx->sop_txd->dword[2], ctx->sop_txd->dword[3]);
649 		vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
650 
651 		/* use the right gen for non-SOP desc */
652 		dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
653 	}
654 
655 	/* linear part can use multiple tx desc if it's big */
656 	len = skb_headlen(skb) - ctx->copy_size;
657 	buf_offset = ctx->copy_size;
658 	while (len) {
659 		u32 buf_size;
660 
661 		buf_size = len > VMXNET3_MAX_TX_BUF_SIZE ?
662 			   VMXNET3_MAX_TX_BUF_SIZE : len;
663 
664 		tbi = tq->buf_info + tq->tx_ring.next2fill;
665 		tbi->map_type = VMXNET3_MAP_SINGLE;
666 		tbi->dma_addr = pci_map_single(adapter->pdev,
667 				skb->data + buf_offset, buf_size,
668 				PCI_DMA_TODEVICE);
669 
670 		tbi->len = buf_size; /* this automatically convert 2^14 to 0 */
671 
672 		gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
673 		BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
674 
675 		gdesc->txd.addr = cpu_to_le64(tbi->dma_addr);
676 		gdesc->dword[2] = cpu_to_le32(dw2 | buf_size);
677 		gdesc->dword[3] = 0;
678 
679 		dev_dbg(&adapter->netdev->dev,
680 			"txd[%u]: 0x%Lx 0x%x 0x%x\n",
681 			tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr),
682 			le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]);
683 		vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
684 		dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
685 
686 		len -= buf_size;
687 		buf_offset += buf_size;
688 	}
689 
690 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
691 		struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
692 
693 		tbi = tq->buf_info + tq->tx_ring.next2fill;
694 		tbi->map_type = VMXNET3_MAP_PAGE;
695 		tbi->dma_addr = pci_map_page(adapter->pdev, frag->page,
696 					     frag->page_offset, frag->size,
697 					     PCI_DMA_TODEVICE);
698 
699 		tbi->len = frag->size;
700 
701 		gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
702 		BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
703 
704 		gdesc->txd.addr = cpu_to_le64(tbi->dma_addr);
705 		gdesc->dword[2] = cpu_to_le32(dw2 | frag->size);
706 		gdesc->dword[3] = 0;
707 
708 		dev_dbg(&adapter->netdev->dev,
709 			"txd[%u]: 0x%llu %u %u\n",
710 			tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr),
711 			le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]);
712 		vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
713 		dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
714 	}
715 
716 	ctx->eop_txd = gdesc;
717 
718 	/* set the last buf_info for the pkt */
719 	tbi->skb = skb;
720 	tbi->sop_idx = ctx->sop_txd - tq->tx_ring.base;
721 }
722 
723 
724 /*
725  *    parse and copy relevant protocol headers:
726  *      For a tso pkt, relevant headers are L2/3/4 including options
727  *      For a pkt requesting csum offloading, they are L2/3 and may include L4
728  *      if it's a TCP/UDP pkt
729  *
730  * Returns:
731  *    -1:  error happens during parsing
732  *     0:  protocol headers parsed, but too big to be copied
733  *     1:  protocol headers parsed and copied
734  *
735  * Other effects:
736  *    1. related *ctx fields are updated.
737  *    2. ctx->copy_size is # of bytes copied
738  *    3. the portion copied is guaranteed to be in the linear part
739  *
740  */
741 static int
742 vmxnet3_parse_and_copy_hdr(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
743 			   struct vmxnet3_tx_ctx *ctx,
744 			   struct vmxnet3_adapter *adapter)
745 {
746 	struct Vmxnet3_TxDataDesc *tdd;
747 
748 	if (ctx->mss) {
749 		ctx->eth_ip_hdr_size = skb_transport_offset(skb);
750 		ctx->l4_hdr_size = ((struct tcphdr *)
751 				   skb_transport_header(skb))->doff * 4;
752 		ctx->copy_size = ctx->eth_ip_hdr_size + ctx->l4_hdr_size;
753 	} else {
754 		unsigned int pull_size;
755 
756 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
757 			ctx->eth_ip_hdr_size = skb_transport_offset(skb);
758 
759 			if (ctx->ipv4) {
760 				struct iphdr *iph = (struct iphdr *)
761 						    skb_network_header(skb);
762 				if (iph->protocol == IPPROTO_TCP) {
763 					pull_size = ctx->eth_ip_hdr_size +
764 						    sizeof(struct tcphdr);
765 
766 					if (unlikely(!pskb_may_pull(skb,
767 								pull_size))) {
768 						goto err;
769 					}
770 					ctx->l4_hdr_size = ((struct tcphdr *)
771 					   skb_transport_header(skb))->doff * 4;
772 				} else if (iph->protocol == IPPROTO_UDP) {
773 					ctx->l4_hdr_size =
774 							sizeof(struct udphdr);
775 				} else {
776 					ctx->l4_hdr_size = 0;
777 				}
778 			} else {
779 				/* for simplicity, don't copy L4 headers */
780 				ctx->l4_hdr_size = 0;
781 			}
782 			ctx->copy_size = ctx->eth_ip_hdr_size +
783 					 ctx->l4_hdr_size;
784 		} else {
785 			ctx->eth_ip_hdr_size = 0;
786 			ctx->l4_hdr_size = 0;
787 			/* copy as much as allowed */
788 			ctx->copy_size = min((unsigned int)VMXNET3_HDR_COPY_SIZE
789 					     , skb_headlen(skb));
790 		}
791 
792 		/* make sure headers are accessible directly */
793 		if (unlikely(!pskb_may_pull(skb, ctx->copy_size)))
794 			goto err;
795 	}
796 
797 	if (unlikely(ctx->copy_size > VMXNET3_HDR_COPY_SIZE)) {
798 		tq->stats.oversized_hdr++;
799 		ctx->copy_size = 0;
800 		return 0;
801 	}
802 
803 	tdd = tq->data_ring.base + tq->tx_ring.next2fill;
804 
805 	memcpy(tdd->data, skb->data, ctx->copy_size);
806 	dev_dbg(&adapter->netdev->dev,
807 		"copy %u bytes to dataRing[%u]\n",
808 		ctx->copy_size, tq->tx_ring.next2fill);
809 	return 1;
810 
811 err:
812 	return -1;
813 }
814 
815 
816 static void
817 vmxnet3_prepare_tso(struct sk_buff *skb,
818 		    struct vmxnet3_tx_ctx *ctx)
819 {
820 	struct tcphdr *tcph = (struct tcphdr *)skb_transport_header(skb);
821 	if (ctx->ipv4) {
822 		struct iphdr *iph = (struct iphdr *)skb_network_header(skb);
823 		iph->check = 0;
824 		tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, 0,
825 						 IPPROTO_TCP, 0);
826 	} else {
827 		struct ipv6hdr *iph = (struct ipv6hdr *)skb_network_header(skb);
828 		tcph->check = ~csum_ipv6_magic(&iph->saddr, &iph->daddr, 0,
829 					       IPPROTO_TCP, 0);
830 	}
831 }
832 
833 
834 /*
835  * Transmits a pkt thru a given tq
836  * Returns:
837  *    NETDEV_TX_OK:      descriptors are setup successfully
838  *    NETDEV_TX_OK:      error occured, the pkt is dropped
839  *    NETDEV_TX_BUSY:    tx ring is full, queue is stopped
840  *
841  * Side-effects:
842  *    1. tx ring may be changed
843  *    2. tq stats may be updated accordingly
844  *    3. shared->txNumDeferred may be updated
845  */
846 
847 static int
848 vmxnet3_tq_xmit(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
849 		struct vmxnet3_adapter *adapter, struct net_device *netdev)
850 {
851 	int ret;
852 	u32 count;
853 	unsigned long flags;
854 	struct vmxnet3_tx_ctx ctx;
855 	union Vmxnet3_GenericDesc *gdesc;
856 #ifdef __BIG_ENDIAN_BITFIELD
857 	/* Use temporary descriptor to avoid touching bits multiple times */
858 	union Vmxnet3_GenericDesc tempTxDesc;
859 #endif
860 
861 	/* conservatively estimate # of descriptors to use */
862 	count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) +
863 		skb_shinfo(skb)->nr_frags + 1;
864 
865 	ctx.ipv4 = (skb->protocol == __constant_ntohs(ETH_P_IP));
866 
867 	ctx.mss = skb_shinfo(skb)->gso_size;
868 	if (ctx.mss) {
869 		if (skb_header_cloned(skb)) {
870 			if (unlikely(pskb_expand_head(skb, 0, 0,
871 						      GFP_ATOMIC) != 0)) {
872 				tq->stats.drop_tso++;
873 				goto drop_pkt;
874 			}
875 			tq->stats.copy_skb_header++;
876 		}
877 		vmxnet3_prepare_tso(skb, &ctx);
878 	} else {
879 		if (unlikely(count > VMXNET3_MAX_TXD_PER_PKT)) {
880 
881 			/* non-tso pkts must not use more than
882 			 * VMXNET3_MAX_TXD_PER_PKT entries
883 			 */
884 			if (skb_linearize(skb) != 0) {
885 				tq->stats.drop_too_many_frags++;
886 				goto drop_pkt;
887 			}
888 			tq->stats.linearized++;
889 
890 			/* recalculate the # of descriptors to use */
891 			count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) + 1;
892 		}
893 	}
894 
895 	ret = vmxnet3_parse_and_copy_hdr(skb, tq, &ctx, adapter);
896 	if (ret >= 0) {
897 		BUG_ON(ret <= 0 && ctx.copy_size != 0);
898 		/* hdrs parsed, check against other limits */
899 		if (ctx.mss) {
900 			if (unlikely(ctx.eth_ip_hdr_size + ctx.l4_hdr_size >
901 				     VMXNET3_MAX_TX_BUF_SIZE)) {
902 				goto hdr_too_big;
903 			}
904 		} else {
905 			if (skb->ip_summed == CHECKSUM_PARTIAL) {
906 				if (unlikely(ctx.eth_ip_hdr_size +
907 					     skb->csum_offset >
908 					     VMXNET3_MAX_CSUM_OFFSET)) {
909 					goto hdr_too_big;
910 				}
911 			}
912 		}
913 	} else {
914 		tq->stats.drop_hdr_inspect_err++;
915 		goto drop_pkt;
916 	}
917 
918 	spin_lock_irqsave(&tq->tx_lock, flags);
919 
920 	if (count > vmxnet3_cmd_ring_desc_avail(&tq->tx_ring)) {
921 		tq->stats.tx_ring_full++;
922 		dev_dbg(&adapter->netdev->dev,
923 			"tx queue stopped on %s, next2comp %u"
924 			" next2fill %u\n", adapter->netdev->name,
925 			tq->tx_ring.next2comp, tq->tx_ring.next2fill);
926 
927 		vmxnet3_tq_stop(tq, adapter);
928 		spin_unlock_irqrestore(&tq->tx_lock, flags);
929 		return NETDEV_TX_BUSY;
930 	}
931 
932 	/* fill tx descs related to addr & len */
933 	vmxnet3_map_pkt(skb, &ctx, tq, adapter->pdev, adapter);
934 
935 	/* setup the EOP desc */
936 	ctx.eop_txd->dword[3] = cpu_to_le32(VMXNET3_TXD_CQ | VMXNET3_TXD_EOP);
937 
938 	/* setup the SOP desc */
939 #ifdef __BIG_ENDIAN_BITFIELD
940 	gdesc = &tempTxDesc;
941 	gdesc->dword[2] = ctx.sop_txd->dword[2];
942 	gdesc->dword[3] = ctx.sop_txd->dword[3];
943 #else
944 	gdesc = ctx.sop_txd;
945 #endif
946 	if (ctx.mss) {
947 		gdesc->txd.hlen = ctx.eth_ip_hdr_size + ctx.l4_hdr_size;
948 		gdesc->txd.om = VMXNET3_OM_TSO;
949 		gdesc->txd.msscof = ctx.mss;
950 		le32_add_cpu(&tq->shared->txNumDeferred, (skb->len -
951 			     gdesc->txd.hlen + ctx.mss - 1) / ctx.mss);
952 	} else {
953 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
954 			gdesc->txd.hlen = ctx.eth_ip_hdr_size;
955 			gdesc->txd.om = VMXNET3_OM_CSUM;
956 			gdesc->txd.msscof = ctx.eth_ip_hdr_size +
957 					    skb->csum_offset;
958 		} else {
959 			gdesc->txd.om = 0;
960 			gdesc->txd.msscof = 0;
961 		}
962 		le32_add_cpu(&tq->shared->txNumDeferred, 1);
963 	}
964 
965 	if (vlan_tx_tag_present(skb)) {
966 		gdesc->txd.ti = 1;
967 		gdesc->txd.tci = vlan_tx_tag_get(skb);
968 	}
969 
970 	/* finally flips the GEN bit of the SOP desc. */
971 	gdesc->dword[2] = cpu_to_le32(le32_to_cpu(gdesc->dword[2]) ^
972 						  VMXNET3_TXD_GEN);
973 #ifdef __BIG_ENDIAN_BITFIELD
974 	/* Finished updating in bitfields of Tx Desc, so write them in original
975 	 * place.
976 	 */
977 	vmxnet3_TxDescToLe((struct Vmxnet3_TxDesc *)gdesc,
978 			   (struct Vmxnet3_TxDesc *)ctx.sop_txd);
979 	gdesc = ctx.sop_txd;
980 #endif
981 	dev_dbg(&adapter->netdev->dev,
982 		"txd[%u]: SOP 0x%Lx 0x%x 0x%x\n",
983 		(u32)((union Vmxnet3_GenericDesc *)ctx.sop_txd -
984 		tq->tx_ring.base), le64_to_cpu(gdesc->txd.addr),
985 		le32_to_cpu(gdesc->dword[2]), le32_to_cpu(gdesc->dword[3]));
986 
987 	spin_unlock_irqrestore(&tq->tx_lock, flags);
988 
989 	if (le32_to_cpu(tq->shared->txNumDeferred) >=
990 					le32_to_cpu(tq->shared->txThreshold)) {
991 		tq->shared->txNumDeferred = 0;
992 		VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_TXPROD,
993 				       tq->tx_ring.next2fill);
994 	}
995 	netdev->trans_start = jiffies;
996 
997 	return NETDEV_TX_OK;
998 
999 hdr_too_big:
1000 	tq->stats.drop_oversized_hdr++;
1001 drop_pkt:
1002 	tq->stats.drop_total++;
1003 	dev_kfree_skb(skb);
1004 	return NETDEV_TX_OK;
1005 }
1006 
1007 
1008 static netdev_tx_t
1009 vmxnet3_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
1010 {
1011 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1012 
1013 	return vmxnet3_tq_xmit(skb, &adapter->tx_queue, adapter, netdev);
1014 }
1015 
1016 
1017 static void
1018 vmxnet3_rx_csum(struct vmxnet3_adapter *adapter,
1019 		struct sk_buff *skb,
1020 		union Vmxnet3_GenericDesc *gdesc)
1021 {
1022 	if (!gdesc->rcd.cnc && adapter->rxcsum) {
1023 		/* typical case: TCP/UDP over IP and both csums are correct */
1024 		if ((le32_to_cpu(gdesc->dword[3]) & VMXNET3_RCD_CSUM_OK) ==
1025 							VMXNET3_RCD_CSUM_OK) {
1026 			skb->ip_summed = CHECKSUM_UNNECESSARY;
1027 			BUG_ON(!(gdesc->rcd.tcp || gdesc->rcd.udp));
1028 			BUG_ON(!(gdesc->rcd.v4  || gdesc->rcd.v6));
1029 			BUG_ON(gdesc->rcd.frg);
1030 		} else {
1031 			if (gdesc->rcd.csum) {
1032 				skb->csum = htons(gdesc->rcd.csum);
1033 				skb->ip_summed = CHECKSUM_PARTIAL;
1034 			} else {
1035 				skb->ip_summed = CHECKSUM_NONE;
1036 			}
1037 		}
1038 	} else {
1039 		skb->ip_summed = CHECKSUM_NONE;
1040 	}
1041 }
1042 
1043 
1044 static void
1045 vmxnet3_rx_error(struct vmxnet3_rx_queue *rq, struct Vmxnet3_RxCompDesc *rcd,
1046 		 struct vmxnet3_rx_ctx *ctx,  struct vmxnet3_adapter *adapter)
1047 {
1048 	rq->stats.drop_err++;
1049 	if (!rcd->fcs)
1050 		rq->stats.drop_fcs++;
1051 
1052 	rq->stats.drop_total++;
1053 
1054 	/*
1055 	 * We do not unmap and chain the rx buffer to the skb.
1056 	 * We basically pretend this buffer is not used and will be recycled
1057 	 * by vmxnet3_rq_alloc_rx_buf()
1058 	 */
1059 
1060 	/*
1061 	 * ctx->skb may be NULL if this is the first and the only one
1062 	 * desc for the pkt
1063 	 */
1064 	if (ctx->skb)
1065 		dev_kfree_skb_irq(ctx->skb);
1066 
1067 	ctx->skb = NULL;
1068 }
1069 
1070 
1071 static int
1072 vmxnet3_rq_rx_complete(struct vmxnet3_rx_queue *rq,
1073 		       struct vmxnet3_adapter *adapter, int quota)
1074 {
1075 	static u32 rxprod_reg[2] = {VMXNET3_REG_RXPROD, VMXNET3_REG_RXPROD2};
1076 	u32 num_rxd = 0;
1077 	struct Vmxnet3_RxCompDesc *rcd;
1078 	struct vmxnet3_rx_ctx *ctx = &rq->rx_ctx;
1079 #ifdef __BIG_ENDIAN_BITFIELD
1080 	struct Vmxnet3_RxDesc rxCmdDesc;
1081 	struct Vmxnet3_RxCompDesc rxComp;
1082 #endif
1083 	vmxnet3_getRxComp(rcd, &rq->comp_ring.base[rq->comp_ring.next2proc].rcd,
1084 			  &rxComp);
1085 	while (rcd->gen == rq->comp_ring.gen) {
1086 		struct vmxnet3_rx_buf_info *rbi;
1087 		struct sk_buff *skb;
1088 		int num_to_alloc;
1089 		struct Vmxnet3_RxDesc *rxd;
1090 		u32 idx, ring_idx;
1091 
1092 		if (num_rxd >= quota) {
1093 			/* we may stop even before we see the EOP desc of
1094 			 * the current pkt
1095 			 */
1096 			break;
1097 		}
1098 		num_rxd++;
1099 
1100 		idx = rcd->rxdIdx;
1101 		ring_idx = rcd->rqID == rq->qid ? 0 : 1;
1102 		vmxnet3_getRxDesc(rxd, &rq->rx_ring[ring_idx].base[idx].rxd,
1103 				  &rxCmdDesc);
1104 		rbi = rq->buf_info[ring_idx] + idx;
1105 
1106 		BUG_ON(rxd->addr != rbi->dma_addr ||
1107 		       rxd->len != rbi->len);
1108 
1109 		if (unlikely(rcd->eop && rcd->err)) {
1110 			vmxnet3_rx_error(rq, rcd, ctx, adapter);
1111 			goto rcd_done;
1112 		}
1113 
1114 		if (rcd->sop) { /* first buf of the pkt */
1115 			BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_HEAD ||
1116 			       rcd->rqID != rq->qid);
1117 
1118 			BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_SKB);
1119 			BUG_ON(ctx->skb != NULL || rbi->skb == NULL);
1120 
1121 			if (unlikely(rcd->len == 0)) {
1122 				/* Pretend the rx buffer is skipped. */
1123 				BUG_ON(!(rcd->sop && rcd->eop));
1124 				dev_dbg(&adapter->netdev->dev,
1125 					"rxRing[%u][%u] 0 length\n",
1126 					ring_idx, idx);
1127 				goto rcd_done;
1128 			}
1129 
1130 			ctx->skb = rbi->skb;
1131 			rbi->skb = NULL;
1132 
1133 			pci_unmap_single(adapter->pdev, rbi->dma_addr, rbi->len,
1134 					 PCI_DMA_FROMDEVICE);
1135 
1136 			skb_put(ctx->skb, rcd->len);
1137 		} else {
1138 			BUG_ON(ctx->skb == NULL);
1139 			/* non SOP buffer must be type 1 in most cases */
1140 			if (rbi->buf_type == VMXNET3_RX_BUF_PAGE) {
1141 				BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_BODY);
1142 
1143 				if (rcd->len) {
1144 					pci_unmap_page(adapter->pdev,
1145 						       rbi->dma_addr, rbi->len,
1146 						       PCI_DMA_FROMDEVICE);
1147 
1148 					vmxnet3_append_frag(ctx->skb, rcd, rbi);
1149 					rbi->page = NULL;
1150 				}
1151 			} else {
1152 				/*
1153 				 * The only time a non-SOP buffer is type 0 is
1154 				 * when it's EOP and error flag is raised, which
1155 				 * has already been handled.
1156 				 */
1157 				BUG_ON(true);
1158 			}
1159 		}
1160 
1161 		skb = ctx->skb;
1162 		if (rcd->eop) {
1163 			skb->len += skb->data_len;
1164 			skb->truesize += skb->data_len;
1165 
1166 			vmxnet3_rx_csum(adapter, skb,
1167 					(union Vmxnet3_GenericDesc *)rcd);
1168 			skb->protocol = eth_type_trans(skb, adapter->netdev);
1169 
1170 			if (unlikely(adapter->vlan_grp && rcd->ts)) {
1171 				vlan_hwaccel_receive_skb(skb,
1172 						adapter->vlan_grp, rcd->tci);
1173 			} else {
1174 				netif_receive_skb(skb);
1175 			}
1176 
1177 			adapter->netdev->last_rx = jiffies;
1178 			ctx->skb = NULL;
1179 		}
1180 
1181 rcd_done:
1182 		/* device may skip some rx descs */
1183 		rq->rx_ring[ring_idx].next2comp = idx;
1184 		VMXNET3_INC_RING_IDX_ONLY(rq->rx_ring[ring_idx].next2comp,
1185 					  rq->rx_ring[ring_idx].size);
1186 
1187 		/* refill rx buffers frequently to avoid starving the h/w */
1188 		num_to_alloc = vmxnet3_cmd_ring_desc_avail(rq->rx_ring +
1189 							   ring_idx);
1190 		if (unlikely(num_to_alloc > VMXNET3_RX_ALLOC_THRESHOLD(rq,
1191 							ring_idx, adapter))) {
1192 			vmxnet3_rq_alloc_rx_buf(rq, ring_idx, num_to_alloc,
1193 						adapter);
1194 
1195 			/* if needed, update the register */
1196 			if (unlikely(rq->shared->updateRxProd)) {
1197 				VMXNET3_WRITE_BAR0_REG(adapter,
1198 					rxprod_reg[ring_idx] + rq->qid * 8,
1199 					rq->rx_ring[ring_idx].next2fill);
1200 				rq->uncommitted[ring_idx] = 0;
1201 			}
1202 		}
1203 
1204 		vmxnet3_comp_ring_adv_next2proc(&rq->comp_ring);
1205 		vmxnet3_getRxComp(rcd,
1206 		     &rq->comp_ring.base[rq->comp_ring.next2proc].rcd, &rxComp);
1207 	}
1208 
1209 	return num_rxd;
1210 }
1211 
1212 
1213 static void
1214 vmxnet3_rq_cleanup(struct vmxnet3_rx_queue *rq,
1215 		   struct vmxnet3_adapter *adapter)
1216 {
1217 	u32 i, ring_idx;
1218 	struct Vmxnet3_RxDesc *rxd;
1219 
1220 	for (ring_idx = 0; ring_idx < 2; ring_idx++) {
1221 		for (i = 0; i < rq->rx_ring[ring_idx].size; i++) {
1222 #ifdef __BIG_ENDIAN_BITFIELD
1223 			struct Vmxnet3_RxDesc rxDesc;
1224 #endif
1225 			vmxnet3_getRxDesc(rxd,
1226 				&rq->rx_ring[ring_idx].base[i].rxd, &rxDesc);
1227 
1228 			if (rxd->btype == VMXNET3_RXD_BTYPE_HEAD &&
1229 					rq->buf_info[ring_idx][i].skb) {
1230 				pci_unmap_single(adapter->pdev, rxd->addr,
1231 						 rxd->len, PCI_DMA_FROMDEVICE);
1232 				dev_kfree_skb(rq->buf_info[ring_idx][i].skb);
1233 				rq->buf_info[ring_idx][i].skb = NULL;
1234 			} else if (rxd->btype == VMXNET3_RXD_BTYPE_BODY &&
1235 					rq->buf_info[ring_idx][i].page) {
1236 				pci_unmap_page(adapter->pdev, rxd->addr,
1237 					       rxd->len, PCI_DMA_FROMDEVICE);
1238 				put_page(rq->buf_info[ring_idx][i].page);
1239 				rq->buf_info[ring_idx][i].page = NULL;
1240 			}
1241 		}
1242 
1243 		rq->rx_ring[ring_idx].gen = VMXNET3_INIT_GEN;
1244 		rq->rx_ring[ring_idx].next2fill =
1245 					rq->rx_ring[ring_idx].next2comp = 0;
1246 		rq->uncommitted[ring_idx] = 0;
1247 	}
1248 
1249 	rq->comp_ring.gen = VMXNET3_INIT_GEN;
1250 	rq->comp_ring.next2proc = 0;
1251 }
1252 
1253 
1254 void vmxnet3_rq_destroy(struct vmxnet3_rx_queue *rq,
1255 			struct vmxnet3_adapter *adapter)
1256 {
1257 	int i;
1258 	int j;
1259 
1260 	/* all rx buffers must have already been freed */
1261 	for (i = 0; i < 2; i++) {
1262 		if (rq->buf_info[i]) {
1263 			for (j = 0; j < rq->rx_ring[i].size; j++)
1264 				BUG_ON(rq->buf_info[i][j].page != NULL);
1265 		}
1266 	}
1267 
1268 
1269 	kfree(rq->buf_info[0]);
1270 
1271 	for (i = 0; i < 2; i++) {
1272 		if (rq->rx_ring[i].base) {
1273 			pci_free_consistent(adapter->pdev, rq->rx_ring[i].size
1274 					    * sizeof(struct Vmxnet3_RxDesc),
1275 					    rq->rx_ring[i].base,
1276 					    rq->rx_ring[i].basePA);
1277 			rq->rx_ring[i].base = NULL;
1278 		}
1279 		rq->buf_info[i] = NULL;
1280 	}
1281 
1282 	if (rq->comp_ring.base) {
1283 		pci_free_consistent(adapter->pdev, rq->comp_ring.size *
1284 				    sizeof(struct Vmxnet3_RxCompDesc),
1285 				    rq->comp_ring.base, rq->comp_ring.basePA);
1286 		rq->comp_ring.base = NULL;
1287 	}
1288 }
1289 
1290 
1291 static int
1292 vmxnet3_rq_init(struct vmxnet3_rx_queue *rq,
1293 		struct vmxnet3_adapter  *adapter)
1294 {
1295 	int i;
1296 
1297 	/* initialize buf_info */
1298 	for (i = 0; i < rq->rx_ring[0].size; i++) {
1299 
1300 		/* 1st buf for a pkt is skbuff */
1301 		if (i % adapter->rx_buf_per_pkt == 0) {
1302 			rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_SKB;
1303 			rq->buf_info[0][i].len = adapter->skb_buf_size;
1304 		} else { /* subsequent bufs for a pkt is frag */
1305 			rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_PAGE;
1306 			rq->buf_info[0][i].len = PAGE_SIZE;
1307 		}
1308 	}
1309 	for (i = 0; i < rq->rx_ring[1].size; i++) {
1310 		rq->buf_info[1][i].buf_type = VMXNET3_RX_BUF_PAGE;
1311 		rq->buf_info[1][i].len = PAGE_SIZE;
1312 	}
1313 
1314 	/* reset internal state and allocate buffers for both rings */
1315 	for (i = 0; i < 2; i++) {
1316 		rq->rx_ring[i].next2fill = rq->rx_ring[i].next2comp = 0;
1317 		rq->uncommitted[i] = 0;
1318 
1319 		memset(rq->rx_ring[i].base, 0, rq->rx_ring[i].size *
1320 		       sizeof(struct Vmxnet3_RxDesc));
1321 		rq->rx_ring[i].gen = VMXNET3_INIT_GEN;
1322 	}
1323 	if (vmxnet3_rq_alloc_rx_buf(rq, 0, rq->rx_ring[0].size - 1,
1324 				    adapter) == 0) {
1325 		/* at least has 1 rx buffer for the 1st ring */
1326 		return -ENOMEM;
1327 	}
1328 	vmxnet3_rq_alloc_rx_buf(rq, 1, rq->rx_ring[1].size - 1, adapter);
1329 
1330 	/* reset the comp ring */
1331 	rq->comp_ring.next2proc = 0;
1332 	memset(rq->comp_ring.base, 0, rq->comp_ring.size *
1333 	       sizeof(struct Vmxnet3_RxCompDesc));
1334 	rq->comp_ring.gen = VMXNET3_INIT_GEN;
1335 
1336 	/* reset rxctx */
1337 	rq->rx_ctx.skb = NULL;
1338 
1339 	/* stats are not reset */
1340 	return 0;
1341 }
1342 
1343 
1344 static int
1345 vmxnet3_rq_create(struct vmxnet3_rx_queue *rq, struct vmxnet3_adapter *adapter)
1346 {
1347 	int i;
1348 	size_t sz;
1349 	struct vmxnet3_rx_buf_info *bi;
1350 
1351 	for (i = 0; i < 2; i++) {
1352 
1353 		sz = rq->rx_ring[i].size * sizeof(struct Vmxnet3_RxDesc);
1354 		rq->rx_ring[i].base = pci_alloc_consistent(adapter->pdev, sz,
1355 							&rq->rx_ring[i].basePA);
1356 		if (!rq->rx_ring[i].base) {
1357 			printk(KERN_ERR "%s: failed to allocate rx ring %d\n",
1358 			       adapter->netdev->name, i);
1359 			goto err;
1360 		}
1361 	}
1362 
1363 	sz = rq->comp_ring.size * sizeof(struct Vmxnet3_RxCompDesc);
1364 	rq->comp_ring.base = pci_alloc_consistent(adapter->pdev, sz,
1365 						  &rq->comp_ring.basePA);
1366 	if (!rq->comp_ring.base) {
1367 		printk(KERN_ERR "%s: failed to allocate rx comp ring\n",
1368 		       adapter->netdev->name);
1369 		goto err;
1370 	}
1371 
1372 	sz = sizeof(struct vmxnet3_rx_buf_info) * (rq->rx_ring[0].size +
1373 						   rq->rx_ring[1].size);
1374 	bi = kmalloc(sz, GFP_KERNEL);
1375 	if (!bi) {
1376 		printk(KERN_ERR "%s: failed to allocate rx bufinfo\n",
1377 		       adapter->netdev->name);
1378 		goto err;
1379 	}
1380 	memset(bi, 0, sz);
1381 	rq->buf_info[0] = bi;
1382 	rq->buf_info[1] = bi + rq->rx_ring[0].size;
1383 
1384 	return 0;
1385 
1386 err:
1387 	vmxnet3_rq_destroy(rq, adapter);
1388 	return -ENOMEM;
1389 }
1390 
1391 
1392 static int
1393 vmxnet3_do_poll(struct vmxnet3_adapter *adapter, int budget)
1394 {
1395 	if (unlikely(adapter->shared->ecr))
1396 		vmxnet3_process_events(adapter);
1397 
1398 	vmxnet3_tq_tx_complete(&adapter->tx_queue, adapter);
1399 	return vmxnet3_rq_rx_complete(&adapter->rx_queue, adapter, budget);
1400 }
1401 
1402 
1403 static int
1404 vmxnet3_poll(struct napi_struct *napi, int budget)
1405 {
1406 	struct vmxnet3_adapter *adapter = container_of(napi,
1407 					  struct vmxnet3_adapter, napi);
1408 	int rxd_done;
1409 
1410 	rxd_done = vmxnet3_do_poll(adapter, budget);
1411 
1412 	if (rxd_done < budget) {
1413 		napi_complete(napi);
1414 		vmxnet3_enable_intr(adapter, 0);
1415 	}
1416 	return rxd_done;
1417 }
1418 
1419 
1420 /* Interrupt handler for vmxnet3  */
1421 static irqreturn_t
1422 vmxnet3_intr(int irq, void *dev_id)
1423 {
1424 	struct net_device *dev = dev_id;
1425 	struct vmxnet3_adapter *adapter = netdev_priv(dev);
1426 
1427 	if (unlikely(adapter->intr.type == VMXNET3_IT_INTX)) {
1428 		u32 icr = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_ICR);
1429 		if (unlikely(icr == 0))
1430 			/* not ours */
1431 			return IRQ_NONE;
1432 	}
1433 
1434 
1435 	/* disable intr if needed */
1436 	if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1437 		vmxnet3_disable_intr(adapter, 0);
1438 
1439 	napi_schedule(&adapter->napi);
1440 
1441 	return IRQ_HANDLED;
1442 }
1443 
1444 #ifdef CONFIG_NET_POLL_CONTROLLER
1445 
1446 
1447 /* netpoll callback. */
1448 static void
1449 vmxnet3_netpoll(struct net_device *netdev)
1450 {
1451 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1452 	int irq;
1453 
1454 #ifdef CONFIG_PCI_MSI
1455 	if (adapter->intr.type == VMXNET3_IT_MSIX)
1456 		irq = adapter->intr.msix_entries[0].vector;
1457 	else
1458 #endif
1459 		irq = adapter->pdev->irq;
1460 
1461 	disable_irq(irq);
1462 	vmxnet3_intr(irq, netdev);
1463 	enable_irq(irq);
1464 }
1465 #endif
1466 
1467 static int
1468 vmxnet3_request_irqs(struct vmxnet3_adapter *adapter)
1469 {
1470 	int err;
1471 
1472 #ifdef CONFIG_PCI_MSI
1473 	if (adapter->intr.type == VMXNET3_IT_MSIX) {
1474 		/* we only use 1 MSI-X vector */
1475 		err = request_irq(adapter->intr.msix_entries[0].vector,
1476 				  vmxnet3_intr, 0, adapter->netdev->name,
1477 				  adapter->netdev);
1478 	} else if (adapter->intr.type == VMXNET3_IT_MSI) {
1479 		err = request_irq(adapter->pdev->irq, vmxnet3_intr, 0,
1480 				  adapter->netdev->name, adapter->netdev);
1481 	} else
1482 #endif
1483 	{
1484 		err = request_irq(adapter->pdev->irq, vmxnet3_intr,
1485 				  IRQF_SHARED, adapter->netdev->name,
1486 				  adapter->netdev);
1487 	}
1488 
1489 	if (err)
1490 		printk(KERN_ERR "Failed to request irq %s (intr type:%d), error"
1491 		       ":%d\n", adapter->netdev->name, adapter->intr.type, err);
1492 
1493 
1494 	if (!err) {
1495 		int i;
1496 		/* init our intr settings */
1497 		for (i = 0; i < adapter->intr.num_intrs; i++)
1498 			adapter->intr.mod_levels[i] = UPT1_IML_ADAPTIVE;
1499 
1500 		/* next setup intr index for all intr sources */
1501 		adapter->tx_queue.comp_ring.intr_idx = 0;
1502 		adapter->rx_queue.comp_ring.intr_idx = 0;
1503 		adapter->intr.event_intr_idx = 0;
1504 
1505 		printk(KERN_INFO "%s: intr type %u, mode %u, %u vectors "
1506 		       "allocated\n", adapter->netdev->name, adapter->intr.type,
1507 		       adapter->intr.mask_mode, adapter->intr.num_intrs);
1508 	}
1509 
1510 	return err;
1511 }
1512 
1513 
1514 static void
1515 vmxnet3_free_irqs(struct vmxnet3_adapter *adapter)
1516 {
1517 	BUG_ON(adapter->intr.type == VMXNET3_IT_AUTO ||
1518 	       adapter->intr.num_intrs <= 0);
1519 
1520 	switch (adapter->intr.type) {
1521 #ifdef CONFIG_PCI_MSI
1522 	case VMXNET3_IT_MSIX:
1523 	{
1524 		int i;
1525 
1526 		for (i = 0; i < adapter->intr.num_intrs; i++)
1527 			free_irq(adapter->intr.msix_entries[i].vector,
1528 				 adapter->netdev);
1529 		break;
1530 	}
1531 #endif
1532 	case VMXNET3_IT_MSI:
1533 		free_irq(adapter->pdev->irq, adapter->netdev);
1534 		break;
1535 	case VMXNET3_IT_INTX:
1536 		free_irq(adapter->pdev->irq, adapter->netdev);
1537 		break;
1538 	default:
1539 		BUG_ON(true);
1540 	}
1541 }
1542 
1543 
1544 inline void set_flag_le16(__le16 *data, u16 flag)
1545 {
1546 	*data = cpu_to_le16(le16_to_cpu(*data) | flag);
1547 }
1548 
1549 inline void set_flag_le64(__le64 *data, u64 flag)
1550 {
1551 	*data = cpu_to_le64(le64_to_cpu(*data) | flag);
1552 }
1553 
1554 inline void reset_flag_le64(__le64 *data, u64 flag)
1555 {
1556 	*data = cpu_to_le64(le64_to_cpu(*data) & ~flag);
1557 }
1558 
1559 
1560 static void
1561 vmxnet3_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp)
1562 {
1563 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1564 	struct Vmxnet3_DriverShared *shared = adapter->shared;
1565 	u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1566 
1567 	if (grp) {
1568 		/* add vlan rx stripping. */
1569 		if (adapter->netdev->features & NETIF_F_HW_VLAN_RX) {
1570 			int i;
1571 			struct Vmxnet3_DSDevRead *devRead = &shared->devRead;
1572 			adapter->vlan_grp = grp;
1573 
1574 			/* update FEATURES to device */
1575 			set_flag_le64(&devRead->misc.uptFeatures,
1576 				      UPT1_F_RXVLAN);
1577 			VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1578 					       VMXNET3_CMD_UPDATE_FEATURE);
1579 			/*
1580 			 *  Clear entire vfTable; then enable untagged pkts.
1581 			 *  Note: setting one entry in vfTable to non-zero turns
1582 			 *  on VLAN rx filtering.
1583 			 */
1584 			for (i = 0; i < VMXNET3_VFT_SIZE; i++)
1585 				vfTable[i] = 0;
1586 
1587 			VMXNET3_SET_VFTABLE_ENTRY(vfTable, 0);
1588 			VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1589 					       VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1590 		} else {
1591 			printk(KERN_ERR "%s: vlan_rx_register when device has "
1592 			       "no NETIF_F_HW_VLAN_RX\n", netdev->name);
1593 		}
1594 	} else {
1595 		/* remove vlan rx stripping. */
1596 		struct Vmxnet3_DSDevRead *devRead = &shared->devRead;
1597 		adapter->vlan_grp = NULL;
1598 
1599 		if (le64_to_cpu(devRead->misc.uptFeatures) & UPT1_F_RXVLAN) {
1600 			int i;
1601 
1602 			for (i = 0; i < VMXNET3_VFT_SIZE; i++) {
1603 				/* clear entire vfTable; this also disables
1604 				 * VLAN rx filtering
1605 				 */
1606 				vfTable[i] = 0;
1607 			}
1608 			VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1609 					       VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1610 
1611 			/* update FEATURES to device */
1612 			reset_flag_le64(&devRead->misc.uptFeatures,
1613 					UPT1_F_RXVLAN);
1614 			VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1615 					       VMXNET3_CMD_UPDATE_FEATURE);
1616 		}
1617 	}
1618 }
1619 
1620 
1621 static void
1622 vmxnet3_restore_vlan(struct vmxnet3_adapter *adapter)
1623 {
1624 	if (adapter->vlan_grp) {
1625 		u16 vid;
1626 		u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1627 		bool activeVlan = false;
1628 
1629 		for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
1630 			if (vlan_group_get_device(adapter->vlan_grp, vid)) {
1631 				VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid);
1632 				activeVlan = true;
1633 			}
1634 		}
1635 		if (activeVlan) {
1636 			/* continue to allow untagged pkts */
1637 			VMXNET3_SET_VFTABLE_ENTRY(vfTable, 0);
1638 		}
1639 	}
1640 }
1641 
1642 
1643 static void
1644 vmxnet3_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1645 {
1646 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1647 	u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1648 
1649 	VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid);
1650 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1651 			       VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1652 }
1653 
1654 
1655 static void
1656 vmxnet3_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1657 {
1658 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1659 	u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1660 
1661 	VMXNET3_CLEAR_VFTABLE_ENTRY(vfTable, vid);
1662 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1663 			       VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1664 }
1665 
1666 
1667 static u8 *
1668 vmxnet3_copy_mc(struct net_device *netdev)
1669 {
1670 	u8 *buf = NULL;
1671 	u32 sz = netdev->mc_count * ETH_ALEN;
1672 
1673 	/* struct Vmxnet3_RxFilterConf.mfTableLen is u16. */
1674 	if (sz <= 0xffff) {
1675 		/* We may be called with BH disabled */
1676 		buf = kmalloc(sz, GFP_ATOMIC);
1677 		if (buf) {
1678 			int i;
1679 			struct dev_mc_list *mc = netdev->mc_list;
1680 
1681 			for (i = 0; i < netdev->mc_count; i++) {
1682 				BUG_ON(!mc);
1683 				memcpy(buf + i * ETH_ALEN, mc->dmi_addr,
1684 				       ETH_ALEN);
1685 				mc = mc->next;
1686 			}
1687 		}
1688 	}
1689 	return buf;
1690 }
1691 
1692 
1693 static void
1694 vmxnet3_set_mc(struct net_device *netdev)
1695 {
1696 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1697 	struct Vmxnet3_RxFilterConf *rxConf =
1698 					&adapter->shared->devRead.rxFilterConf;
1699 	u8 *new_table = NULL;
1700 	u32 new_mode = VMXNET3_RXM_UCAST;
1701 
1702 	if (netdev->flags & IFF_PROMISC)
1703 		new_mode |= VMXNET3_RXM_PROMISC;
1704 
1705 	if (netdev->flags & IFF_BROADCAST)
1706 		new_mode |= VMXNET3_RXM_BCAST;
1707 
1708 	if (netdev->flags & IFF_ALLMULTI)
1709 		new_mode |= VMXNET3_RXM_ALL_MULTI;
1710 	else
1711 		if (netdev->mc_count > 0) {
1712 			new_table = vmxnet3_copy_mc(netdev);
1713 			if (new_table) {
1714 				new_mode |= VMXNET3_RXM_MCAST;
1715 				rxConf->mfTableLen = cpu_to_le16(
1716 						netdev->mc_count * ETH_ALEN);
1717 				rxConf->mfTablePA = cpu_to_le64(virt_to_phys(
1718 						    new_table));
1719 			} else {
1720 				printk(KERN_INFO "%s: failed to copy mcast list"
1721 				       ", setting ALL_MULTI\n", netdev->name);
1722 				new_mode |= VMXNET3_RXM_ALL_MULTI;
1723 			}
1724 		}
1725 
1726 
1727 	if (!(new_mode & VMXNET3_RXM_MCAST)) {
1728 		rxConf->mfTableLen = 0;
1729 		rxConf->mfTablePA = 0;
1730 	}
1731 
1732 	if (new_mode != rxConf->rxMode) {
1733 		rxConf->rxMode = cpu_to_le32(new_mode);
1734 		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1735 				       VMXNET3_CMD_UPDATE_RX_MODE);
1736 	}
1737 
1738 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1739 			       VMXNET3_CMD_UPDATE_MAC_FILTERS);
1740 
1741 	kfree(new_table);
1742 }
1743 
1744 
1745 /*
1746  *   Set up driver_shared based on settings in adapter.
1747  */
1748 
1749 static void
1750 vmxnet3_setup_driver_shared(struct vmxnet3_adapter *adapter)
1751 {
1752 	struct Vmxnet3_DriverShared *shared = adapter->shared;
1753 	struct Vmxnet3_DSDevRead *devRead = &shared->devRead;
1754 	struct Vmxnet3_TxQueueConf *tqc;
1755 	struct Vmxnet3_RxQueueConf *rqc;
1756 	int i;
1757 
1758 	memset(shared, 0, sizeof(*shared));
1759 
1760 	/* driver settings */
1761 	shared->magic = cpu_to_le32(VMXNET3_REV1_MAGIC);
1762 	devRead->misc.driverInfo.version = cpu_to_le32(
1763 						VMXNET3_DRIVER_VERSION_NUM);
1764 	devRead->misc.driverInfo.gos.gosBits = (sizeof(void *) == 4 ?
1765 				VMXNET3_GOS_BITS_32 : VMXNET3_GOS_BITS_64);
1766 	devRead->misc.driverInfo.gos.gosType = VMXNET3_GOS_TYPE_LINUX;
1767 	*((u32 *)&devRead->misc.driverInfo.gos) = cpu_to_le32(
1768 				*((u32 *)&devRead->misc.driverInfo.gos));
1769 	devRead->misc.driverInfo.vmxnet3RevSpt = cpu_to_le32(1);
1770 	devRead->misc.driverInfo.uptVerSpt = cpu_to_le32(1);
1771 
1772 	devRead->misc.ddPA = cpu_to_le64(virt_to_phys(adapter));
1773 	devRead->misc.ddLen = cpu_to_le32(sizeof(struct vmxnet3_adapter));
1774 
1775 	/* set up feature flags */
1776 	if (adapter->rxcsum)
1777 		set_flag_le64(&devRead->misc.uptFeatures, UPT1_F_RXCSUM);
1778 
1779 	if (adapter->lro) {
1780 		set_flag_le64(&devRead->misc.uptFeatures, UPT1_F_LRO);
1781 		devRead->misc.maxNumRxSG = cpu_to_le16(1 + MAX_SKB_FRAGS);
1782 	}
1783 	if ((adapter->netdev->features & NETIF_F_HW_VLAN_RX) &&
1784 	    adapter->vlan_grp) {
1785 		set_flag_le64(&devRead->misc.uptFeatures, UPT1_F_RXVLAN);
1786 	}
1787 
1788 	devRead->misc.mtu = cpu_to_le32(adapter->netdev->mtu);
1789 	devRead->misc.queueDescPA = cpu_to_le64(adapter->queue_desc_pa);
1790 	devRead->misc.queueDescLen = cpu_to_le32(
1791 				     sizeof(struct Vmxnet3_TxQueueDesc) +
1792 				     sizeof(struct Vmxnet3_RxQueueDesc));
1793 
1794 	/* tx queue settings */
1795 	BUG_ON(adapter->tx_queue.tx_ring.base == NULL);
1796 
1797 	devRead->misc.numTxQueues = 1;
1798 	tqc = &adapter->tqd_start->conf;
1799 	tqc->txRingBasePA   = cpu_to_le64(adapter->tx_queue.tx_ring.basePA);
1800 	tqc->dataRingBasePA = cpu_to_le64(adapter->tx_queue.data_ring.basePA);
1801 	tqc->compRingBasePA = cpu_to_le64(adapter->tx_queue.comp_ring.basePA);
1802 	tqc->ddPA           = cpu_to_le64(virt_to_phys(
1803 						adapter->tx_queue.buf_info));
1804 	tqc->txRingSize     = cpu_to_le32(adapter->tx_queue.tx_ring.size);
1805 	tqc->dataRingSize   = cpu_to_le32(adapter->tx_queue.data_ring.size);
1806 	tqc->compRingSize   = cpu_to_le32(adapter->tx_queue.comp_ring.size);
1807 	tqc->ddLen          = cpu_to_le32(sizeof(struct vmxnet3_tx_buf_info) *
1808 			      tqc->txRingSize);
1809 	tqc->intrIdx        = adapter->tx_queue.comp_ring.intr_idx;
1810 
1811 	/* rx queue settings */
1812 	devRead->misc.numRxQueues = 1;
1813 	rqc = &adapter->rqd_start->conf;
1814 	rqc->rxRingBasePA[0] = cpu_to_le64(adapter->rx_queue.rx_ring[0].basePA);
1815 	rqc->rxRingBasePA[1] = cpu_to_le64(adapter->rx_queue.rx_ring[1].basePA);
1816 	rqc->compRingBasePA  = cpu_to_le64(adapter->rx_queue.comp_ring.basePA);
1817 	rqc->ddPA            = cpu_to_le64(virt_to_phys(
1818 						adapter->rx_queue.buf_info));
1819 	rqc->rxRingSize[0]   = cpu_to_le32(adapter->rx_queue.rx_ring[0].size);
1820 	rqc->rxRingSize[1]   = cpu_to_le32(adapter->rx_queue.rx_ring[1].size);
1821 	rqc->compRingSize    = cpu_to_le32(adapter->rx_queue.comp_ring.size);
1822 	rqc->ddLen           = cpu_to_le32(sizeof(struct vmxnet3_rx_buf_info) *
1823 			       (rqc->rxRingSize[0] + rqc->rxRingSize[1]));
1824 	rqc->intrIdx         = adapter->rx_queue.comp_ring.intr_idx;
1825 
1826 	/* intr settings */
1827 	devRead->intrConf.autoMask = adapter->intr.mask_mode ==
1828 				     VMXNET3_IMM_AUTO;
1829 	devRead->intrConf.numIntrs = adapter->intr.num_intrs;
1830 	for (i = 0; i < adapter->intr.num_intrs; i++)
1831 		devRead->intrConf.modLevels[i] = adapter->intr.mod_levels[i];
1832 
1833 	devRead->intrConf.eventIntrIdx = adapter->intr.event_intr_idx;
1834 
1835 	/* rx filter settings */
1836 	devRead->rxFilterConf.rxMode = 0;
1837 	vmxnet3_restore_vlan(adapter);
1838 	/* the rest are already zeroed */
1839 }
1840 
1841 
1842 int
1843 vmxnet3_activate_dev(struct vmxnet3_adapter *adapter)
1844 {
1845 	int err;
1846 	u32 ret;
1847 
1848 	dev_dbg(&adapter->netdev->dev,
1849 		"%s: skb_buf_size %d, rx_buf_per_pkt %d, ring sizes"
1850 		" %u %u %u\n", adapter->netdev->name, adapter->skb_buf_size,
1851 		adapter->rx_buf_per_pkt, adapter->tx_queue.tx_ring.size,
1852 		adapter->rx_queue.rx_ring[0].size,
1853 		adapter->rx_queue.rx_ring[1].size);
1854 
1855 	vmxnet3_tq_init(&adapter->tx_queue, adapter);
1856 	err = vmxnet3_rq_init(&adapter->rx_queue, adapter);
1857 	if (err) {
1858 		printk(KERN_ERR "Failed to init rx queue for %s: error %d\n",
1859 		       adapter->netdev->name, err);
1860 		goto rq_err;
1861 	}
1862 
1863 	err = vmxnet3_request_irqs(adapter);
1864 	if (err) {
1865 		printk(KERN_ERR "Failed to setup irq for %s: error %d\n",
1866 		       adapter->netdev->name, err);
1867 		goto irq_err;
1868 	}
1869 
1870 	vmxnet3_setup_driver_shared(adapter);
1871 
1872 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, VMXNET3_GET_ADDR_LO(
1873 			       adapter->shared_pa));
1874 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, VMXNET3_GET_ADDR_HI(
1875 			       adapter->shared_pa));
1876 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1877 			       VMXNET3_CMD_ACTIVATE_DEV);
1878 	ret = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
1879 
1880 	if (ret != 0) {
1881 		printk(KERN_ERR "Failed to activate dev %s: error %u\n",
1882 		       adapter->netdev->name, ret);
1883 		err = -EINVAL;
1884 		goto activate_err;
1885 	}
1886 	VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_RXPROD,
1887 			       adapter->rx_queue.rx_ring[0].next2fill);
1888 	VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_RXPROD2,
1889 			       adapter->rx_queue.rx_ring[1].next2fill);
1890 
1891 	/* Apply the rx filter settins last. */
1892 	vmxnet3_set_mc(adapter->netdev);
1893 
1894 	/*
1895 	 * Check link state when first activating device. It will start the
1896 	 * tx queue if the link is up.
1897 	 */
1898 	vmxnet3_check_link(adapter);
1899 
1900 	napi_enable(&adapter->napi);
1901 	vmxnet3_enable_all_intrs(adapter);
1902 	clear_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
1903 	return 0;
1904 
1905 activate_err:
1906 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, 0);
1907 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, 0);
1908 	vmxnet3_free_irqs(adapter);
1909 irq_err:
1910 rq_err:
1911 	/* free up buffers we allocated */
1912 	vmxnet3_rq_cleanup(&adapter->rx_queue, adapter);
1913 	return err;
1914 }
1915 
1916 
1917 void
1918 vmxnet3_reset_dev(struct vmxnet3_adapter *adapter)
1919 {
1920 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, VMXNET3_CMD_RESET_DEV);
1921 }
1922 
1923 
1924 int
1925 vmxnet3_quiesce_dev(struct vmxnet3_adapter *adapter)
1926 {
1927 	if (test_and_set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state))
1928 		return 0;
1929 
1930 
1931 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1932 			       VMXNET3_CMD_QUIESCE_DEV);
1933 	vmxnet3_disable_all_intrs(adapter);
1934 
1935 	napi_disable(&adapter->napi);
1936 	netif_tx_disable(adapter->netdev);
1937 	adapter->link_speed = 0;
1938 	netif_carrier_off(adapter->netdev);
1939 
1940 	vmxnet3_tq_cleanup(&adapter->tx_queue, adapter);
1941 	vmxnet3_rq_cleanup(&adapter->rx_queue, adapter);
1942 	vmxnet3_free_irqs(adapter);
1943 	return 0;
1944 }
1945 
1946 
1947 static void
1948 vmxnet3_write_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac)
1949 {
1950 	u32 tmp;
1951 
1952 	tmp = *(u32 *)mac;
1953 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACL, tmp);
1954 
1955 	tmp = (mac[5] << 8) | mac[4];
1956 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACH, tmp);
1957 }
1958 
1959 
1960 static int
1961 vmxnet3_set_mac_addr(struct net_device *netdev, void *p)
1962 {
1963 	struct sockaddr *addr = p;
1964 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1965 
1966 	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1967 	vmxnet3_write_mac_addr(adapter, addr->sa_data);
1968 
1969 	return 0;
1970 }
1971 
1972 
1973 /* ==================== initialization and cleanup routines ============ */
1974 
1975 static int
1976 vmxnet3_alloc_pci_resources(struct vmxnet3_adapter *adapter, bool *dma64)
1977 {
1978 	int err;
1979 	unsigned long mmio_start, mmio_len;
1980 	struct pci_dev *pdev = adapter->pdev;
1981 
1982 	err = pci_enable_device(pdev);
1983 	if (err) {
1984 		printk(KERN_ERR "Failed to enable adapter %s: error %d\n",
1985 		       pci_name(pdev), err);
1986 		return err;
1987 	}
1988 
1989 	if (pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) == 0) {
1990 		if (pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64)) != 0) {
1991 			printk(KERN_ERR "pci_set_consistent_dma_mask failed "
1992 			       "for adapter %s\n", pci_name(pdev));
1993 			err = -EIO;
1994 			goto err_set_mask;
1995 		}
1996 		*dma64 = true;
1997 	} else {
1998 		if (pci_set_dma_mask(pdev, DMA_BIT_MASK(32)) != 0) {
1999 			printk(KERN_ERR "pci_set_dma_mask failed for adapter "
2000 			       "%s\n",	pci_name(pdev));
2001 			err = -EIO;
2002 			goto err_set_mask;
2003 		}
2004 		*dma64 = false;
2005 	}
2006 
2007 	err = pci_request_selected_regions(pdev, (1 << 2) - 1,
2008 					   vmxnet3_driver_name);
2009 	if (err) {
2010 		printk(KERN_ERR "Failed to request region for adapter %s: "
2011 		       "error %d\n", pci_name(pdev), err);
2012 		goto err_set_mask;
2013 	}
2014 
2015 	pci_set_master(pdev);
2016 
2017 	mmio_start = pci_resource_start(pdev, 0);
2018 	mmio_len = pci_resource_len(pdev, 0);
2019 	adapter->hw_addr0 = ioremap(mmio_start, mmio_len);
2020 	if (!adapter->hw_addr0) {
2021 		printk(KERN_ERR "Failed to map bar0 for adapter %s\n",
2022 		       pci_name(pdev));
2023 		err = -EIO;
2024 		goto err_ioremap;
2025 	}
2026 
2027 	mmio_start = pci_resource_start(pdev, 1);
2028 	mmio_len = pci_resource_len(pdev, 1);
2029 	adapter->hw_addr1 = ioremap(mmio_start, mmio_len);
2030 	if (!adapter->hw_addr1) {
2031 		printk(KERN_ERR "Failed to map bar1 for adapter %s\n",
2032 		       pci_name(pdev));
2033 		err = -EIO;
2034 		goto err_bar1;
2035 	}
2036 	return 0;
2037 
2038 err_bar1:
2039 	iounmap(adapter->hw_addr0);
2040 err_ioremap:
2041 	pci_release_selected_regions(pdev, (1 << 2) - 1);
2042 err_set_mask:
2043 	pci_disable_device(pdev);
2044 	return err;
2045 }
2046 
2047 
2048 static void
2049 vmxnet3_free_pci_resources(struct vmxnet3_adapter *adapter)
2050 {
2051 	BUG_ON(!adapter->pdev);
2052 
2053 	iounmap(adapter->hw_addr0);
2054 	iounmap(adapter->hw_addr1);
2055 	pci_release_selected_regions(adapter->pdev, (1 << 2) - 1);
2056 	pci_disable_device(adapter->pdev);
2057 }
2058 
2059 
2060 static void
2061 vmxnet3_adjust_rx_ring_size(struct vmxnet3_adapter *adapter)
2062 {
2063 	size_t sz;
2064 
2065 	if (adapter->netdev->mtu <= VMXNET3_MAX_SKB_BUF_SIZE -
2066 				    VMXNET3_MAX_ETH_HDR_SIZE) {
2067 		adapter->skb_buf_size = adapter->netdev->mtu +
2068 					VMXNET3_MAX_ETH_HDR_SIZE;
2069 		if (adapter->skb_buf_size < VMXNET3_MIN_T0_BUF_SIZE)
2070 			adapter->skb_buf_size = VMXNET3_MIN_T0_BUF_SIZE;
2071 
2072 		adapter->rx_buf_per_pkt = 1;
2073 	} else {
2074 		adapter->skb_buf_size = VMXNET3_MAX_SKB_BUF_SIZE;
2075 		sz = adapter->netdev->mtu - VMXNET3_MAX_SKB_BUF_SIZE +
2076 					    VMXNET3_MAX_ETH_HDR_SIZE;
2077 		adapter->rx_buf_per_pkt = 1 + (sz + PAGE_SIZE - 1) / PAGE_SIZE;
2078 	}
2079 
2080 	/*
2081 	 * for simplicity, force the ring0 size to be a multiple of
2082 	 * rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN
2083 	 */
2084 	sz = adapter->rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN;
2085 	adapter->rx_queue.rx_ring[0].size = (adapter->rx_queue.rx_ring[0].size +
2086 					     sz - 1) / sz * sz;
2087 	adapter->rx_queue.rx_ring[0].size = min_t(u32,
2088 					    adapter->rx_queue.rx_ring[0].size,
2089 					    VMXNET3_RX_RING_MAX_SIZE / sz * sz);
2090 }
2091 
2092 
2093 int
2094 vmxnet3_create_queues(struct vmxnet3_adapter *adapter, u32 tx_ring_size,
2095 		      u32 rx_ring_size, u32 rx_ring2_size)
2096 {
2097 	int err;
2098 
2099 	adapter->tx_queue.tx_ring.size   = tx_ring_size;
2100 	adapter->tx_queue.data_ring.size = tx_ring_size;
2101 	adapter->tx_queue.comp_ring.size = tx_ring_size;
2102 	adapter->tx_queue.shared = &adapter->tqd_start->ctrl;
2103 	adapter->tx_queue.stopped = true;
2104 	err = vmxnet3_tq_create(&adapter->tx_queue, adapter);
2105 	if (err)
2106 		return err;
2107 
2108 	adapter->rx_queue.rx_ring[0].size = rx_ring_size;
2109 	adapter->rx_queue.rx_ring[1].size = rx_ring2_size;
2110 	vmxnet3_adjust_rx_ring_size(adapter);
2111 	adapter->rx_queue.comp_ring.size  = adapter->rx_queue.rx_ring[0].size +
2112 					    adapter->rx_queue.rx_ring[1].size;
2113 	adapter->rx_queue.qid  = 0;
2114 	adapter->rx_queue.qid2 = 1;
2115 	adapter->rx_queue.shared = &adapter->rqd_start->ctrl;
2116 	err = vmxnet3_rq_create(&adapter->rx_queue, adapter);
2117 	if (err)
2118 		vmxnet3_tq_destroy(&adapter->tx_queue, adapter);
2119 
2120 	return err;
2121 }
2122 
2123 static int
2124 vmxnet3_open(struct net_device *netdev)
2125 {
2126 	struct vmxnet3_adapter *adapter;
2127 	int err;
2128 
2129 	adapter = netdev_priv(netdev);
2130 
2131 	spin_lock_init(&adapter->tx_queue.tx_lock);
2132 
2133 	err = vmxnet3_create_queues(adapter, VMXNET3_DEF_TX_RING_SIZE,
2134 				    VMXNET3_DEF_RX_RING_SIZE,
2135 				    VMXNET3_DEF_RX_RING_SIZE);
2136 	if (err)
2137 		goto queue_err;
2138 
2139 	err = vmxnet3_activate_dev(adapter);
2140 	if (err)
2141 		goto activate_err;
2142 
2143 	return 0;
2144 
2145 activate_err:
2146 	vmxnet3_rq_destroy(&adapter->rx_queue, adapter);
2147 	vmxnet3_tq_destroy(&adapter->tx_queue, adapter);
2148 queue_err:
2149 	return err;
2150 }
2151 
2152 
2153 static int
2154 vmxnet3_close(struct net_device *netdev)
2155 {
2156 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2157 
2158 	/*
2159 	 * Reset_work may be in the middle of resetting the device, wait for its
2160 	 * completion.
2161 	 */
2162 	while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2163 		msleep(1);
2164 
2165 	vmxnet3_quiesce_dev(adapter);
2166 
2167 	vmxnet3_rq_destroy(&adapter->rx_queue, adapter);
2168 	vmxnet3_tq_destroy(&adapter->tx_queue, adapter);
2169 
2170 	clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2171 
2172 
2173 	return 0;
2174 }
2175 
2176 
2177 void
2178 vmxnet3_force_close(struct vmxnet3_adapter *adapter)
2179 {
2180 	/*
2181 	 * we must clear VMXNET3_STATE_BIT_RESETTING, otherwise
2182 	 * vmxnet3_close() will deadlock.
2183 	 */
2184 	BUG_ON(test_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state));
2185 
2186 	/* we need to enable NAPI, otherwise dev_close will deadlock */
2187 	napi_enable(&adapter->napi);
2188 	dev_close(adapter->netdev);
2189 }
2190 
2191 
2192 static int
2193 vmxnet3_change_mtu(struct net_device *netdev, int new_mtu)
2194 {
2195 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2196 	int err = 0;
2197 
2198 	if (new_mtu < VMXNET3_MIN_MTU || new_mtu > VMXNET3_MAX_MTU)
2199 		return -EINVAL;
2200 
2201 	if (new_mtu > 1500 && !adapter->jumbo_frame)
2202 		return -EINVAL;
2203 
2204 	netdev->mtu = new_mtu;
2205 
2206 	/*
2207 	 * Reset_work may be in the middle of resetting the device, wait for its
2208 	 * completion.
2209 	 */
2210 	while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2211 		msleep(1);
2212 
2213 	if (netif_running(netdev)) {
2214 		vmxnet3_quiesce_dev(adapter);
2215 		vmxnet3_reset_dev(adapter);
2216 
2217 		/* we need to re-create the rx queue based on the new mtu */
2218 		vmxnet3_rq_destroy(&adapter->rx_queue, adapter);
2219 		vmxnet3_adjust_rx_ring_size(adapter);
2220 		adapter->rx_queue.comp_ring.size  =
2221 					adapter->rx_queue.rx_ring[0].size +
2222 					adapter->rx_queue.rx_ring[1].size;
2223 		err = vmxnet3_rq_create(&adapter->rx_queue, adapter);
2224 		if (err) {
2225 			printk(KERN_ERR "%s: failed to re-create rx queue,"
2226 				" error %d. Closing it.\n", netdev->name, err);
2227 			goto out;
2228 		}
2229 
2230 		err = vmxnet3_activate_dev(adapter);
2231 		if (err) {
2232 			printk(KERN_ERR "%s: failed to re-activate, error %d. "
2233 				"Closing it\n", netdev->name, err);
2234 			goto out;
2235 		}
2236 	}
2237 
2238 out:
2239 	clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2240 	if (err)
2241 		vmxnet3_force_close(adapter);
2242 
2243 	return err;
2244 }
2245 
2246 
2247 static void
2248 vmxnet3_declare_features(struct vmxnet3_adapter *adapter, bool dma64)
2249 {
2250 	struct net_device *netdev = adapter->netdev;
2251 
2252 	netdev->features = NETIF_F_SG |
2253 		NETIF_F_HW_CSUM |
2254 		NETIF_F_HW_VLAN_TX |
2255 		NETIF_F_HW_VLAN_RX |
2256 		NETIF_F_HW_VLAN_FILTER |
2257 		NETIF_F_TSO |
2258 		NETIF_F_TSO6 |
2259 		NETIF_F_LRO;
2260 
2261 	printk(KERN_INFO "features: sg csum vlan jf tso tsoIPv6 lro");
2262 
2263 	adapter->rxcsum = true;
2264 	adapter->jumbo_frame = true;
2265 	adapter->lro = true;
2266 
2267 	if (dma64) {
2268 		netdev->features |= NETIF_F_HIGHDMA;
2269 		printk(" highDMA");
2270 	}
2271 
2272 	netdev->vlan_features = netdev->features;
2273 	printk("\n");
2274 }
2275 
2276 
2277 static void
2278 vmxnet3_read_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac)
2279 {
2280 	u32 tmp;
2281 
2282 	tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACL);
2283 	*(u32 *)mac = tmp;
2284 
2285 	tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACH);
2286 	mac[4] = tmp & 0xff;
2287 	mac[5] = (tmp >> 8) & 0xff;
2288 }
2289 
2290 
2291 static void
2292 vmxnet3_alloc_intr_resources(struct vmxnet3_adapter *adapter)
2293 {
2294 	u32 cfg;
2295 
2296 	/* intr settings */
2297 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2298 			       VMXNET3_CMD_GET_CONF_INTR);
2299 	cfg = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
2300 	adapter->intr.type = cfg & 0x3;
2301 	adapter->intr.mask_mode = (cfg >> 2) & 0x3;
2302 
2303 	if (adapter->intr.type == VMXNET3_IT_AUTO) {
2304 		int err;
2305 
2306 #ifdef CONFIG_PCI_MSI
2307 		adapter->intr.msix_entries[0].entry = 0;
2308 		err = pci_enable_msix(adapter->pdev, adapter->intr.msix_entries,
2309 				      VMXNET3_LINUX_MAX_MSIX_VECT);
2310 		if (!err) {
2311 			adapter->intr.num_intrs = 1;
2312 			adapter->intr.type = VMXNET3_IT_MSIX;
2313 			return;
2314 		}
2315 #endif
2316 
2317 		err = pci_enable_msi(adapter->pdev);
2318 		if (!err) {
2319 			adapter->intr.num_intrs = 1;
2320 			adapter->intr.type = VMXNET3_IT_MSI;
2321 			return;
2322 		}
2323 	}
2324 
2325 	adapter->intr.type = VMXNET3_IT_INTX;
2326 
2327 	/* INT-X related setting */
2328 	adapter->intr.num_intrs = 1;
2329 }
2330 
2331 
2332 static void
2333 vmxnet3_free_intr_resources(struct vmxnet3_adapter *adapter)
2334 {
2335 	if (adapter->intr.type == VMXNET3_IT_MSIX)
2336 		pci_disable_msix(adapter->pdev);
2337 	else if (adapter->intr.type == VMXNET3_IT_MSI)
2338 		pci_disable_msi(adapter->pdev);
2339 	else
2340 		BUG_ON(adapter->intr.type != VMXNET3_IT_INTX);
2341 }
2342 
2343 
2344 static void
2345 vmxnet3_tx_timeout(struct net_device *netdev)
2346 {
2347 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2348 	adapter->tx_timeout_count++;
2349 
2350 	printk(KERN_ERR "%s: tx hang\n", adapter->netdev->name);
2351 	schedule_work(&adapter->work);
2352 }
2353 
2354 
2355 static void
2356 vmxnet3_reset_work(struct work_struct *data)
2357 {
2358 	struct vmxnet3_adapter *adapter;
2359 
2360 	adapter = container_of(data, struct vmxnet3_adapter, work);
2361 
2362 	/* if another thread is resetting the device, no need to proceed */
2363 	if (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2364 		return;
2365 
2366 	/* if the device is closed, we must leave it alone */
2367 	if (netif_running(adapter->netdev)) {
2368 		printk(KERN_INFO "%s: resetting\n", adapter->netdev->name);
2369 		vmxnet3_quiesce_dev(adapter);
2370 		vmxnet3_reset_dev(adapter);
2371 		vmxnet3_activate_dev(adapter);
2372 	} else {
2373 		printk(KERN_INFO "%s: already closed\n", adapter->netdev->name);
2374 	}
2375 
2376 	clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2377 }
2378 
2379 
2380 static int __devinit
2381 vmxnet3_probe_device(struct pci_dev *pdev,
2382 		     const struct pci_device_id *id)
2383 {
2384 	static const struct net_device_ops vmxnet3_netdev_ops = {
2385 		.ndo_open = vmxnet3_open,
2386 		.ndo_stop = vmxnet3_close,
2387 		.ndo_start_xmit = vmxnet3_xmit_frame,
2388 		.ndo_set_mac_address = vmxnet3_set_mac_addr,
2389 		.ndo_change_mtu = vmxnet3_change_mtu,
2390 		.ndo_get_stats = vmxnet3_get_stats,
2391 		.ndo_tx_timeout = vmxnet3_tx_timeout,
2392 		.ndo_set_multicast_list = vmxnet3_set_mc,
2393 		.ndo_vlan_rx_register = vmxnet3_vlan_rx_register,
2394 		.ndo_vlan_rx_add_vid = vmxnet3_vlan_rx_add_vid,
2395 		.ndo_vlan_rx_kill_vid = vmxnet3_vlan_rx_kill_vid,
2396 #ifdef CONFIG_NET_POLL_CONTROLLER
2397 		.ndo_poll_controller = vmxnet3_netpoll,
2398 #endif
2399 	};
2400 	int err;
2401 	bool dma64 = false; /* stupid gcc */
2402 	u32 ver;
2403 	struct net_device *netdev;
2404 	struct vmxnet3_adapter *adapter;
2405 	u8 mac[ETH_ALEN];
2406 
2407 	netdev = alloc_etherdev(sizeof(struct vmxnet3_adapter));
2408 	if (!netdev) {
2409 		printk(KERN_ERR "Failed to alloc ethernet device for adapter "
2410 			"%s\n",	pci_name(pdev));
2411 		return -ENOMEM;
2412 	}
2413 
2414 	pci_set_drvdata(pdev, netdev);
2415 	adapter = netdev_priv(netdev);
2416 	adapter->netdev = netdev;
2417 	adapter->pdev = pdev;
2418 
2419 	adapter->shared = pci_alloc_consistent(adapter->pdev,
2420 			  sizeof(struct Vmxnet3_DriverShared),
2421 			  &adapter->shared_pa);
2422 	if (!adapter->shared) {
2423 		printk(KERN_ERR "Failed to allocate memory for %s\n",
2424 			pci_name(pdev));
2425 		err = -ENOMEM;
2426 		goto err_alloc_shared;
2427 	}
2428 
2429 	adapter->tqd_start = pci_alloc_consistent(adapter->pdev,
2430 			     sizeof(struct Vmxnet3_TxQueueDesc) +
2431 			     sizeof(struct Vmxnet3_RxQueueDesc),
2432 			     &adapter->queue_desc_pa);
2433 
2434 	if (!adapter->tqd_start) {
2435 		printk(KERN_ERR "Failed to allocate memory for %s\n",
2436 			pci_name(pdev));
2437 		err = -ENOMEM;
2438 		goto err_alloc_queue_desc;
2439 	}
2440 	adapter->rqd_start = (struct Vmxnet3_RxQueueDesc *)(adapter->tqd_start
2441 							    + 1);
2442 
2443 	adapter->pm_conf = kmalloc(sizeof(struct Vmxnet3_PMConf), GFP_KERNEL);
2444 	if (adapter->pm_conf == NULL) {
2445 		printk(KERN_ERR "Failed to allocate memory for %s\n",
2446 			pci_name(pdev));
2447 		err = -ENOMEM;
2448 		goto err_alloc_pm;
2449 	}
2450 
2451 	err = vmxnet3_alloc_pci_resources(adapter, &dma64);
2452 	if (err < 0)
2453 		goto err_alloc_pci;
2454 
2455 	ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_VRRS);
2456 	if (ver & 1) {
2457 		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_VRRS, 1);
2458 	} else {
2459 		printk(KERN_ERR "Incompatible h/w version (0x%x) for adapter"
2460 		       " %s\n",	ver, pci_name(pdev));
2461 		err = -EBUSY;
2462 		goto err_ver;
2463 	}
2464 
2465 	ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_UVRS);
2466 	if (ver & 1) {
2467 		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_UVRS, 1);
2468 	} else {
2469 		printk(KERN_ERR "Incompatible upt version (0x%x) for "
2470 		       "adapter %s\n", ver, pci_name(pdev));
2471 		err = -EBUSY;
2472 		goto err_ver;
2473 	}
2474 
2475 	vmxnet3_declare_features(adapter, dma64);
2476 
2477 	adapter->dev_number = atomic_read(&devices_found);
2478 	vmxnet3_alloc_intr_resources(adapter);
2479 
2480 	vmxnet3_read_mac_addr(adapter, mac);
2481 	memcpy(netdev->dev_addr,  mac, netdev->addr_len);
2482 
2483 	netdev->netdev_ops = &vmxnet3_netdev_ops;
2484 	netdev->watchdog_timeo = 5 * HZ;
2485 	vmxnet3_set_ethtool_ops(netdev);
2486 
2487 	INIT_WORK(&adapter->work, vmxnet3_reset_work);
2488 
2489 	netif_napi_add(netdev, &adapter->napi, vmxnet3_poll, 64);
2490 	SET_NETDEV_DEV(netdev, &pdev->dev);
2491 	err = register_netdev(netdev);
2492 
2493 	if (err) {
2494 		printk(KERN_ERR "Failed to register adapter %s\n",
2495 			pci_name(pdev));
2496 		goto err_register;
2497 	}
2498 
2499 	set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
2500 	atomic_inc(&devices_found);
2501 	return 0;
2502 
2503 err_register:
2504 	vmxnet3_free_intr_resources(adapter);
2505 err_ver:
2506 	vmxnet3_free_pci_resources(adapter);
2507 err_alloc_pci:
2508 	kfree(adapter->pm_conf);
2509 err_alloc_pm:
2510 	pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_TxQueueDesc) +
2511 			    sizeof(struct Vmxnet3_RxQueueDesc),
2512 			    adapter->tqd_start, adapter->queue_desc_pa);
2513 err_alloc_queue_desc:
2514 	pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_DriverShared),
2515 			    adapter->shared, adapter->shared_pa);
2516 err_alloc_shared:
2517 	pci_set_drvdata(pdev, NULL);
2518 	free_netdev(netdev);
2519 	return err;
2520 }
2521 
2522 
2523 static void __devexit
2524 vmxnet3_remove_device(struct pci_dev *pdev)
2525 {
2526 	struct net_device *netdev = pci_get_drvdata(pdev);
2527 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2528 
2529 	flush_scheduled_work();
2530 
2531 	unregister_netdev(netdev);
2532 
2533 	vmxnet3_free_intr_resources(adapter);
2534 	vmxnet3_free_pci_resources(adapter);
2535 	kfree(adapter->pm_conf);
2536 	pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_TxQueueDesc) +
2537 			    sizeof(struct Vmxnet3_RxQueueDesc),
2538 			    adapter->tqd_start, adapter->queue_desc_pa);
2539 	pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_DriverShared),
2540 			    adapter->shared, adapter->shared_pa);
2541 	free_netdev(netdev);
2542 }
2543 
2544 
2545 #ifdef CONFIG_PM
2546 
2547 static int
2548 vmxnet3_suspend(struct device *device)
2549 {
2550 	struct pci_dev *pdev = to_pci_dev(device);
2551 	struct net_device *netdev = pci_get_drvdata(pdev);
2552 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2553 	struct Vmxnet3_PMConf *pmConf;
2554 	struct ethhdr *ehdr;
2555 	struct arphdr *ahdr;
2556 	u8 *arpreq;
2557 	struct in_device *in_dev;
2558 	struct in_ifaddr *ifa;
2559 	int i = 0;
2560 
2561 	if (!netif_running(netdev))
2562 		return 0;
2563 
2564 	vmxnet3_disable_all_intrs(adapter);
2565 	vmxnet3_free_irqs(adapter);
2566 	vmxnet3_free_intr_resources(adapter);
2567 
2568 	netif_device_detach(netdev);
2569 	netif_stop_queue(netdev);
2570 
2571 	/* Create wake-up filters. */
2572 	pmConf = adapter->pm_conf;
2573 	memset(pmConf, 0, sizeof(*pmConf));
2574 
2575 	if (adapter->wol & WAKE_UCAST) {
2576 		pmConf->filters[i].patternSize = ETH_ALEN;
2577 		pmConf->filters[i].maskSize = 1;
2578 		memcpy(pmConf->filters[i].pattern, netdev->dev_addr, ETH_ALEN);
2579 		pmConf->filters[i].mask[0] = 0x3F; /* LSB ETH_ALEN bits */
2580 
2581 		set_flag_le16(&pmConf->wakeUpEvents, VMXNET3_PM_WAKEUP_FILTER);
2582 		i++;
2583 	}
2584 
2585 	if (adapter->wol & WAKE_ARP) {
2586 		in_dev = in_dev_get(netdev);
2587 		if (!in_dev)
2588 			goto skip_arp;
2589 
2590 		ifa = (struct in_ifaddr *)in_dev->ifa_list;
2591 		if (!ifa)
2592 			goto skip_arp;
2593 
2594 		pmConf->filters[i].patternSize = ETH_HLEN + /* Ethernet header*/
2595 			sizeof(struct arphdr) +		/* ARP header */
2596 			2 * ETH_ALEN +		/* 2 Ethernet addresses*/
2597 			2 * sizeof(u32);	/*2 IPv4 addresses */
2598 		pmConf->filters[i].maskSize =
2599 			(pmConf->filters[i].patternSize - 1) / 8 + 1;
2600 
2601 		/* ETH_P_ARP in Ethernet header. */
2602 		ehdr = (struct ethhdr *)pmConf->filters[i].pattern;
2603 		ehdr->h_proto = htons(ETH_P_ARP);
2604 
2605 		/* ARPOP_REQUEST in ARP header. */
2606 		ahdr = (struct arphdr *)&pmConf->filters[i].pattern[ETH_HLEN];
2607 		ahdr->ar_op = htons(ARPOP_REQUEST);
2608 		arpreq = (u8 *)(ahdr + 1);
2609 
2610 		/* The Unicast IPv4 address in 'tip' field. */
2611 		arpreq += 2 * ETH_ALEN + sizeof(u32);
2612 		*(u32 *)arpreq = ifa->ifa_address;
2613 
2614 		/* The mask for the relevant bits. */
2615 		pmConf->filters[i].mask[0] = 0x00;
2616 		pmConf->filters[i].mask[1] = 0x30; /* ETH_P_ARP */
2617 		pmConf->filters[i].mask[2] = 0x30; /* ARPOP_REQUEST */
2618 		pmConf->filters[i].mask[3] = 0x00;
2619 		pmConf->filters[i].mask[4] = 0xC0; /* IPv4 TIP */
2620 		pmConf->filters[i].mask[5] = 0x03; /* IPv4 TIP */
2621 		in_dev_put(in_dev);
2622 
2623 		set_flag_le16(&pmConf->wakeUpEvents, VMXNET3_PM_WAKEUP_FILTER);
2624 		i++;
2625 	}
2626 
2627 skip_arp:
2628 	if (adapter->wol & WAKE_MAGIC)
2629 		set_flag_le16(&pmConf->wakeUpEvents, VMXNET3_PM_WAKEUP_MAGIC);
2630 
2631 	pmConf->numFilters = i;
2632 
2633 	adapter->shared->devRead.pmConfDesc.confVer = cpu_to_le32(1);
2634 	adapter->shared->devRead.pmConfDesc.confLen = cpu_to_le32(sizeof(
2635 								  *pmConf));
2636 	adapter->shared->devRead.pmConfDesc.confPA = cpu_to_le64(virt_to_phys(
2637 								 pmConf));
2638 
2639 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2640 			       VMXNET3_CMD_UPDATE_PMCFG);
2641 
2642 	pci_save_state(pdev);
2643 	pci_enable_wake(pdev, pci_choose_state(pdev, PMSG_SUSPEND),
2644 			adapter->wol);
2645 	pci_disable_device(pdev);
2646 	pci_set_power_state(pdev, pci_choose_state(pdev, PMSG_SUSPEND));
2647 
2648 	return 0;
2649 }
2650 
2651 
2652 static int
2653 vmxnet3_resume(struct device *device)
2654 {
2655 	int err;
2656 	struct pci_dev *pdev = to_pci_dev(device);
2657 	struct net_device *netdev = pci_get_drvdata(pdev);
2658 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2659 	struct Vmxnet3_PMConf *pmConf;
2660 
2661 	if (!netif_running(netdev))
2662 		return 0;
2663 
2664 	/* Destroy wake-up filters. */
2665 	pmConf = adapter->pm_conf;
2666 	memset(pmConf, 0, sizeof(*pmConf));
2667 
2668 	adapter->shared->devRead.pmConfDesc.confVer = cpu_to_le32(1);
2669 	adapter->shared->devRead.pmConfDesc.confLen = cpu_to_le32(sizeof(
2670 								  *pmConf));
2671 	adapter->shared->devRead.pmConfDesc.confPA = cpu_to_le32(virt_to_phys(
2672 								 pmConf));
2673 
2674 	netif_device_attach(netdev);
2675 	pci_set_power_state(pdev, PCI_D0);
2676 	pci_restore_state(pdev);
2677 	err = pci_enable_device_mem(pdev);
2678 	if (err != 0)
2679 		return err;
2680 
2681 	pci_enable_wake(pdev, PCI_D0, 0);
2682 
2683 	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2684 			       VMXNET3_CMD_UPDATE_PMCFG);
2685 	vmxnet3_alloc_intr_resources(adapter);
2686 	vmxnet3_request_irqs(adapter);
2687 	vmxnet3_enable_all_intrs(adapter);
2688 
2689 	return 0;
2690 }
2691 
2692 static const struct dev_pm_ops vmxnet3_pm_ops = {
2693 	.suspend = vmxnet3_suspend,
2694 	.resume = vmxnet3_resume,
2695 };
2696 #endif
2697 
2698 static struct pci_driver vmxnet3_driver = {
2699 	.name		= vmxnet3_driver_name,
2700 	.id_table	= vmxnet3_pciid_table,
2701 	.probe		= vmxnet3_probe_device,
2702 	.remove		= __devexit_p(vmxnet3_remove_device),
2703 #ifdef CONFIG_PM
2704 	.driver.pm	= &vmxnet3_pm_ops,
2705 #endif
2706 };
2707 
2708 
2709 static int __init
2710 vmxnet3_init_module(void)
2711 {
2712 	printk(KERN_INFO "%s - version %s\n", VMXNET3_DRIVER_DESC,
2713 		VMXNET3_DRIVER_VERSION_REPORT);
2714 	return pci_register_driver(&vmxnet3_driver);
2715 }
2716 
2717 module_init(vmxnet3_init_module);
2718 
2719 
2720 static void
2721 vmxnet3_exit_module(void)
2722 {
2723 	pci_unregister_driver(&vmxnet3_driver);
2724 }
2725 
2726 module_exit(vmxnet3_exit_module);
2727 
2728 MODULE_AUTHOR("VMware, Inc.");
2729 MODULE_DESCRIPTION(VMXNET3_DRIVER_DESC);
2730 MODULE_LICENSE("GPL v2");
2731 MODULE_VERSION(VMXNET3_DRIVER_VERSION_STRING);
2732