vmxnet3_drv.c revision 8f7e524ce33ca81b663711404709396165da3cbd
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 "vmxnet3_int.h"
28
29char vmxnet3_driver_name[] = "vmxnet3";
30#define VMXNET3_DRIVER_DESC "VMware vmxnet3 virtual NIC driver"
31
32
33/*
34 * PCI Device ID Table
35 * Last entry must be all 0s
36 */
37static const struct pci_device_id vmxnet3_pciid_table[] = {
38	{PCI_VDEVICE(VMWARE, PCI_DEVICE_ID_VMWARE_VMXNET3)},
39	{0}
40};
41
42MODULE_DEVICE_TABLE(pci, vmxnet3_pciid_table);
43
44static atomic_t devices_found;
45
46
47/*
48 *    Enable/Disable the given intr
49 */
50static void
51vmxnet3_enable_intr(struct vmxnet3_adapter *adapter, unsigned intr_idx)
52{
53	VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_IMR + intr_idx * 8, 0);
54}
55
56
57static void
58vmxnet3_disable_intr(struct vmxnet3_adapter *adapter, unsigned intr_idx)
59{
60	VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_IMR + intr_idx * 8, 1);
61}
62
63
64/*
65 *    Enable/Disable all intrs used by the device
66 */
67static void
68vmxnet3_enable_all_intrs(struct vmxnet3_adapter *adapter)
69{
70	int i;
71
72	for (i = 0; i < adapter->intr.num_intrs; i++)
73		vmxnet3_enable_intr(adapter, i);
74}
75
76
77static void
78vmxnet3_disable_all_intrs(struct vmxnet3_adapter *adapter)
79{
80	int i;
81
82	for (i = 0; i < adapter->intr.num_intrs; i++)
83		vmxnet3_disable_intr(adapter, i);
84}
85
86
87static void
88vmxnet3_ack_events(struct vmxnet3_adapter *adapter, u32 events)
89{
90	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_ECR, events);
91}
92
93
94static bool
95vmxnet3_tq_stopped(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
96{
97	return netif_queue_stopped(adapter->netdev);
98}
99
100
101static void
102vmxnet3_tq_start(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
103{
104	tq->stopped = false;
105	netif_start_queue(adapter->netdev);
106}
107
108
109static void
110vmxnet3_tq_wake(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
111{
112	tq->stopped = false;
113	netif_wake_queue(adapter->netdev);
114}
115
116
117static void
118vmxnet3_tq_stop(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
119{
120	tq->stopped = true;
121	tq->num_stop++;
122	netif_stop_queue(adapter->netdev);
123}
124
125
126/*
127 * Check the link state. This may start or stop the tx queue.
128 */
129static void
130vmxnet3_check_link(struct vmxnet3_adapter *adapter)
131{
132	u32 ret;
133
134	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, VMXNET3_CMD_GET_LINK);
135	ret = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
136	adapter->link_speed = ret >> 16;
137	if (ret & 1) { /* Link is up. */
138		printk(KERN_INFO "%s: NIC Link is Up %d Mbps\n",
139		       adapter->netdev->name, adapter->link_speed);
140		if (!netif_carrier_ok(adapter->netdev))
141			netif_carrier_on(adapter->netdev);
142
143		vmxnet3_tq_start(&adapter->tx_queue, adapter);
144	} else {
145		printk(KERN_INFO "%s: NIC Link is Down\n",
146		       adapter->netdev->name);
147		if (netif_carrier_ok(adapter->netdev))
148			netif_carrier_off(adapter->netdev);
149
150		vmxnet3_tq_stop(&adapter->tx_queue, adapter);
151	}
152}
153
154
155static void
156vmxnet3_process_events(struct vmxnet3_adapter *adapter)
157{
158	u32 events = 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			       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
189static void
190vmxnet3_unmap_tx_buf(struct vmxnet3_tx_buf_info *tbi,
191		     struct pci_dev *pdev)
192{
193	if (tbi->map_type == VMXNET3_MAP_SINGLE)
194		pci_unmap_single(pdev, tbi->dma_addr, tbi->len,
195				 PCI_DMA_TODEVICE);
196	else if (tbi->map_type == VMXNET3_MAP_PAGE)
197		pci_unmap_page(pdev, tbi->dma_addr, tbi->len,
198			       PCI_DMA_TODEVICE);
199	else
200		BUG_ON(tbi->map_type != VMXNET3_MAP_NONE);
201
202	tbi->map_type = VMXNET3_MAP_NONE; /* to help debugging */
203}
204
205
206static int
207vmxnet3_unmap_pkt(u32 eop_idx, struct vmxnet3_tx_queue *tq,
208		  struct pci_dev *pdev,	struct vmxnet3_adapter *adapter)
209{
210	struct sk_buff *skb;
211	int entries = 0;
212
213	/* no out of order completion */
214	BUG_ON(tq->buf_info[eop_idx].sop_idx != tq->tx_ring.next2comp);
215	BUG_ON(tq->tx_ring.base[eop_idx].txd.eop != 1);
216
217	skb = tq->buf_info[eop_idx].skb;
218	BUG_ON(skb == NULL);
219	tq->buf_info[eop_idx].skb = NULL;
220
221	VMXNET3_INC_RING_IDX_ONLY(eop_idx, tq->tx_ring.size);
222
223	while (tq->tx_ring.next2comp != eop_idx) {
224		vmxnet3_unmap_tx_buf(tq->buf_info + tq->tx_ring.next2comp,
225				     pdev);
226
227		/* update next2comp w/o tx_lock. Since we are marking more,
228		 * instead of less, tx ring entries avail, the worst case is
229		 * that the tx routine incorrectly re-queues a pkt due to
230		 * insufficient tx ring entries.
231		 */
232		vmxnet3_cmd_ring_adv_next2comp(&tq->tx_ring);
233		entries++;
234	}
235
236	dev_kfree_skb_any(skb);
237	return entries;
238}
239
240
241static int
242vmxnet3_tq_tx_complete(struct vmxnet3_tx_queue *tq,
243			struct vmxnet3_adapter *adapter)
244{
245	int completed = 0;
246	union Vmxnet3_GenericDesc *gdesc;
247
248	gdesc = tq->comp_ring.base + tq->comp_ring.next2proc;
249	while (gdesc->tcd.gen == tq->comp_ring.gen) {
250		completed += vmxnet3_unmap_pkt(gdesc->tcd.txdIdx, tq,
251					       adapter->pdev, adapter);
252
253		vmxnet3_comp_ring_adv_next2proc(&tq->comp_ring);
254		gdesc = tq->comp_ring.base + tq->comp_ring.next2proc;
255	}
256
257	if (completed) {
258		spin_lock(&tq->tx_lock);
259		if (unlikely(vmxnet3_tq_stopped(tq, adapter) &&
260			     vmxnet3_cmd_ring_desc_avail(&tq->tx_ring) >
261			     VMXNET3_WAKE_QUEUE_THRESHOLD(tq) &&
262			     netif_carrier_ok(adapter->netdev))) {
263			vmxnet3_tq_wake(tq, adapter);
264		}
265		spin_unlock(&tq->tx_lock);
266	}
267	return completed;
268}
269
270
271static void
272vmxnet3_tq_cleanup(struct vmxnet3_tx_queue *tq,
273		   struct vmxnet3_adapter *adapter)
274{
275	int i;
276
277	while (tq->tx_ring.next2comp != tq->tx_ring.next2fill) {
278		struct vmxnet3_tx_buf_info *tbi;
279		union Vmxnet3_GenericDesc *gdesc;
280
281		tbi = tq->buf_info + tq->tx_ring.next2comp;
282		gdesc = tq->tx_ring.base + tq->tx_ring.next2comp;
283
284		vmxnet3_unmap_tx_buf(tbi, adapter->pdev);
285		if (tbi->skb) {
286			dev_kfree_skb_any(tbi->skb);
287			tbi->skb = NULL;
288		}
289		vmxnet3_cmd_ring_adv_next2comp(&tq->tx_ring);
290	}
291
292	/* sanity check, verify all buffers are indeed unmapped and freed */
293	for (i = 0; i < tq->tx_ring.size; i++) {
294		BUG_ON(tq->buf_info[i].skb != NULL ||
295		       tq->buf_info[i].map_type != VMXNET3_MAP_NONE);
296	}
297
298	tq->tx_ring.gen = VMXNET3_INIT_GEN;
299	tq->tx_ring.next2fill = tq->tx_ring.next2comp = 0;
300
301	tq->comp_ring.gen = VMXNET3_INIT_GEN;
302	tq->comp_ring.next2proc = 0;
303}
304
305
306void
307vmxnet3_tq_destroy(struct vmxnet3_tx_queue *tq,
308		   struct vmxnet3_adapter *adapter)
309{
310	if (tq->tx_ring.base) {
311		pci_free_consistent(adapter->pdev, tq->tx_ring.size *
312				    sizeof(struct Vmxnet3_TxDesc),
313				    tq->tx_ring.base, tq->tx_ring.basePA);
314		tq->tx_ring.base = NULL;
315	}
316	if (tq->data_ring.base) {
317		pci_free_consistent(adapter->pdev, tq->data_ring.size *
318				    sizeof(struct Vmxnet3_TxDataDesc),
319				    tq->data_ring.base, tq->data_ring.basePA);
320		tq->data_ring.base = NULL;
321	}
322	if (tq->comp_ring.base) {
323		pci_free_consistent(adapter->pdev, tq->comp_ring.size *
324				    sizeof(struct Vmxnet3_TxCompDesc),
325				    tq->comp_ring.base, tq->comp_ring.basePA);
326		tq->comp_ring.base = NULL;
327	}
328	kfree(tq->buf_info);
329	tq->buf_info = NULL;
330}
331
332
333static void
334vmxnet3_tq_init(struct vmxnet3_tx_queue *tq,
335		struct vmxnet3_adapter *adapter)
336{
337	int i;
338
339	/* reset the tx ring contents to 0 and reset the tx ring states */
340	memset(tq->tx_ring.base, 0, tq->tx_ring.size *
341	       sizeof(struct Vmxnet3_TxDesc));
342	tq->tx_ring.next2fill = tq->tx_ring.next2comp = 0;
343	tq->tx_ring.gen = VMXNET3_INIT_GEN;
344
345	memset(tq->data_ring.base, 0, tq->data_ring.size *
346	       sizeof(struct Vmxnet3_TxDataDesc));
347
348	/* reset the tx comp ring contents to 0 and reset comp ring states */
349	memset(tq->comp_ring.base, 0, tq->comp_ring.size *
350	       sizeof(struct Vmxnet3_TxCompDesc));
351	tq->comp_ring.next2proc = 0;
352	tq->comp_ring.gen = VMXNET3_INIT_GEN;
353
354	/* reset the bookkeeping data */
355	memset(tq->buf_info, 0, sizeof(tq->buf_info[0]) * tq->tx_ring.size);
356	for (i = 0; i < tq->tx_ring.size; i++)
357		tq->buf_info[i].map_type = VMXNET3_MAP_NONE;
358
359	/* stats are not reset */
360}
361
362
363static int
364vmxnet3_tq_create(struct vmxnet3_tx_queue *tq,
365		  struct vmxnet3_adapter *adapter)
366{
367	BUG_ON(tq->tx_ring.base || tq->data_ring.base ||
368	       tq->comp_ring.base || tq->buf_info);
369
370	tq->tx_ring.base = pci_alloc_consistent(adapter->pdev, tq->tx_ring.size
371			   * sizeof(struct Vmxnet3_TxDesc),
372			   &tq->tx_ring.basePA);
373	if (!tq->tx_ring.base) {
374		printk(KERN_ERR "%s: failed to allocate tx ring\n",
375		       adapter->netdev->name);
376		goto err;
377	}
378
379	tq->data_ring.base = pci_alloc_consistent(adapter->pdev,
380			     tq->data_ring.size *
381			     sizeof(struct Vmxnet3_TxDataDesc),
382			     &tq->data_ring.basePA);
383	if (!tq->data_ring.base) {
384		printk(KERN_ERR "%s: failed to allocate data ring\n",
385		       adapter->netdev->name);
386		goto err;
387	}
388
389	tq->comp_ring.base = pci_alloc_consistent(adapter->pdev,
390			     tq->comp_ring.size *
391			     sizeof(struct Vmxnet3_TxCompDesc),
392			     &tq->comp_ring.basePA);
393	if (!tq->comp_ring.base) {
394		printk(KERN_ERR "%s: failed to allocate tx comp ring\n",
395		       adapter->netdev->name);
396		goto err;
397	}
398
399	tq->buf_info = kcalloc(tq->tx_ring.size, sizeof(tq->buf_info[0]),
400			       GFP_KERNEL);
401	if (!tq->buf_info) {
402		printk(KERN_ERR "%s: failed to allocate tx bufinfo\n",
403		       adapter->netdev->name);
404		goto err;
405	}
406
407	return 0;
408
409err:
410	vmxnet3_tq_destroy(tq, adapter);
411	return -ENOMEM;
412}
413
414
415/*
416 *    starting from ring->next2fill, allocate rx buffers for the given ring
417 *    of the rx queue and update the rx desc. stop after @num_to_alloc buffers
418 *    are allocated or allocation fails
419 */
420
421static int
422vmxnet3_rq_alloc_rx_buf(struct vmxnet3_rx_queue *rq, u32 ring_idx,
423			int num_to_alloc, struct vmxnet3_adapter *adapter)
424{
425	int num_allocated = 0;
426	struct vmxnet3_rx_buf_info *rbi_base = rq->buf_info[ring_idx];
427	struct vmxnet3_cmd_ring *ring = &rq->rx_ring[ring_idx];
428	u32 val;
429
430	while (num_allocated < num_to_alloc) {
431		struct vmxnet3_rx_buf_info *rbi;
432		union Vmxnet3_GenericDesc *gd;
433
434		rbi = rbi_base + ring->next2fill;
435		gd = ring->base + ring->next2fill;
436
437		if (rbi->buf_type == VMXNET3_RX_BUF_SKB) {
438			if (rbi->skb == NULL) {
439				rbi->skb = dev_alloc_skb(rbi->len +
440							 NET_IP_ALIGN);
441				if (unlikely(rbi->skb == NULL)) {
442					rq->stats.rx_buf_alloc_failure++;
443					break;
444				}
445				rbi->skb->dev = adapter->netdev;
446
447				skb_reserve(rbi->skb, NET_IP_ALIGN);
448				rbi->dma_addr = pci_map_single(adapter->pdev,
449						rbi->skb->data, rbi->len,
450						PCI_DMA_FROMDEVICE);
451			} else {
452				/* rx buffer skipped by the device */
453			}
454			val = VMXNET3_RXD_BTYPE_HEAD << VMXNET3_RXD_BTYPE_SHIFT;
455		} else {
456			BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_PAGE ||
457			       rbi->len  != PAGE_SIZE);
458
459			if (rbi->page == NULL) {
460				rbi->page = alloc_page(GFP_ATOMIC);
461				if (unlikely(rbi->page == NULL)) {
462					rq->stats.rx_buf_alloc_failure++;
463					break;
464				}
465				rbi->dma_addr = pci_map_page(adapter->pdev,
466						rbi->page, 0, PAGE_SIZE,
467						PCI_DMA_FROMDEVICE);
468			} else {
469				/* rx buffers skipped by the device */
470			}
471			val = VMXNET3_RXD_BTYPE_BODY << VMXNET3_RXD_BTYPE_SHIFT;
472		}
473
474		BUG_ON(rbi->dma_addr == 0);
475		gd->rxd.addr = rbi->dma_addr;
476		gd->dword[2] = (ring->gen << VMXNET3_RXD_GEN_SHIFT) | val |
477				rbi->len;
478
479		num_allocated++;
480		vmxnet3_cmd_ring_adv_next2fill(ring);
481	}
482	rq->uncommitted[ring_idx] += num_allocated;
483
484	dprintk(KERN_ERR "alloc_rx_buf: %d allocated, next2fill %u, next2comp "
485		"%u, uncommited %u\n", num_allocated, ring->next2fill,
486		ring->next2comp, rq->uncommitted[ring_idx]);
487
488	/* so that the device can distinguish a full ring and an empty ring */
489	BUG_ON(num_allocated != 0 && ring->next2fill == ring->next2comp);
490
491	return num_allocated;
492}
493
494
495static void
496vmxnet3_append_frag(struct sk_buff *skb, struct Vmxnet3_RxCompDesc *rcd,
497		    struct vmxnet3_rx_buf_info *rbi)
498{
499	struct skb_frag_struct *frag = skb_shinfo(skb)->frags +
500		skb_shinfo(skb)->nr_frags;
501
502	BUG_ON(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS);
503
504	frag->page = rbi->page;
505	frag->page_offset = 0;
506	frag->size = rcd->len;
507	skb->data_len += frag->size;
508	skb_shinfo(skb)->nr_frags++;
509}
510
511
512static void
513vmxnet3_map_pkt(struct sk_buff *skb, struct vmxnet3_tx_ctx *ctx,
514		struct vmxnet3_tx_queue *tq, struct pci_dev *pdev,
515		struct vmxnet3_adapter *adapter)
516{
517	u32 dw2, len;
518	unsigned long buf_offset;
519	int i;
520	union Vmxnet3_GenericDesc *gdesc;
521	struct vmxnet3_tx_buf_info *tbi = NULL;
522
523	BUG_ON(ctx->copy_size > skb_headlen(skb));
524
525	/* use the previous gen bit for the SOP desc */
526	dw2 = (tq->tx_ring.gen ^ 0x1) << VMXNET3_TXD_GEN_SHIFT;
527
528	ctx->sop_txd = tq->tx_ring.base + tq->tx_ring.next2fill;
529	gdesc = ctx->sop_txd; /* both loops below can be skipped */
530
531	/* no need to map the buffer if headers are copied */
532	if (ctx->copy_size) {
533		ctx->sop_txd->txd.addr = tq->data_ring.basePA +
534					tq->tx_ring.next2fill *
535					sizeof(struct Vmxnet3_TxDataDesc);
536		ctx->sop_txd->dword[2] = dw2 | ctx->copy_size;
537		ctx->sop_txd->dword[3] = 0;
538
539		tbi = tq->buf_info + tq->tx_ring.next2fill;
540		tbi->map_type = VMXNET3_MAP_NONE;
541
542		dprintk(KERN_ERR "txd[%u]: 0x%Lx 0x%x 0x%x\n",
543			tq->tx_ring.next2fill, ctx->sop_txd->txd.addr,
544			ctx->sop_txd->dword[2], ctx->sop_txd->dword[3]);
545		vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
546
547		/* use the right gen for non-SOP desc */
548		dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
549	}
550
551	/* linear part can use multiple tx desc if it's big */
552	len = skb_headlen(skb) - ctx->copy_size;
553	buf_offset = ctx->copy_size;
554	while (len) {
555		u32 buf_size;
556
557		buf_size = len > VMXNET3_MAX_TX_BUF_SIZE ?
558			   VMXNET3_MAX_TX_BUF_SIZE : len;
559
560		tbi = tq->buf_info + tq->tx_ring.next2fill;
561		tbi->map_type = VMXNET3_MAP_SINGLE;
562		tbi->dma_addr = pci_map_single(adapter->pdev,
563				skb->data + buf_offset, buf_size,
564				PCI_DMA_TODEVICE);
565
566		tbi->len = buf_size; /* this automatically convert 2^14 to 0 */
567
568		gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
569		BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
570
571		gdesc->txd.addr = tbi->dma_addr;
572		gdesc->dword[2] = dw2 | buf_size;
573		gdesc->dword[3] = 0;
574
575		dprintk(KERN_ERR "txd[%u]: 0x%Lx 0x%x 0x%x\n",
576			tq->tx_ring.next2fill, gdesc->txd.addr,
577			gdesc->dword[2], gdesc->dword[3]);
578		vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
579		dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
580
581		len -= buf_size;
582		buf_offset += buf_size;
583	}
584
585	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
586		struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
587
588		tbi = tq->buf_info + tq->tx_ring.next2fill;
589		tbi->map_type = VMXNET3_MAP_PAGE;
590		tbi->dma_addr = pci_map_page(adapter->pdev, frag->page,
591					     frag->page_offset, frag->size,
592					     PCI_DMA_TODEVICE);
593
594		tbi->len = frag->size;
595
596		gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
597		BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
598
599		gdesc->txd.addr = tbi->dma_addr;
600		gdesc->dword[2] = dw2 | frag->size;
601		gdesc->dword[3] = 0;
602
603		dprintk(KERN_ERR "txd[%u]: 0x%llu %u %u\n",
604			tq->tx_ring.next2fill, gdesc->txd.addr,
605			gdesc->dword[2], gdesc->dword[3]);
606		vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
607		dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
608	}
609
610	ctx->eop_txd = gdesc;
611
612	/* set the last buf_info for the pkt */
613	tbi->skb = skb;
614	tbi->sop_idx = ctx->sop_txd - tq->tx_ring.base;
615}
616
617
618/*
619 *    parse and copy relevant protocol headers:
620 *      For a tso pkt, relevant headers are L2/3/4 including options
621 *      For a pkt requesting csum offloading, they are L2/3 and may include L4
622 *      if it's a TCP/UDP pkt
623 *
624 * Returns:
625 *    -1:  error happens during parsing
626 *     0:  protocol headers parsed, but too big to be copied
627 *     1:  protocol headers parsed and copied
628 *
629 * Other effects:
630 *    1. related *ctx fields are updated.
631 *    2. ctx->copy_size is # of bytes copied
632 *    3. the portion copied is guaranteed to be in the linear part
633 *
634 */
635static int
636vmxnet3_parse_and_copy_hdr(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
637			   struct vmxnet3_tx_ctx *ctx,
638			   struct vmxnet3_adapter *adapter)
639{
640	struct Vmxnet3_TxDataDesc *tdd;
641
642	if (ctx->mss) {
643		ctx->eth_ip_hdr_size = skb_transport_offset(skb);
644		ctx->l4_hdr_size = ((struct tcphdr *)
645				   skb_transport_header(skb))->doff * 4;
646		ctx->copy_size = ctx->eth_ip_hdr_size + ctx->l4_hdr_size;
647	} else {
648		unsigned int pull_size;
649
650		if (skb->ip_summed == CHECKSUM_PARTIAL) {
651			ctx->eth_ip_hdr_size = skb_transport_offset(skb);
652
653			if (ctx->ipv4) {
654				struct iphdr *iph = (struct iphdr *)
655						    skb_network_header(skb);
656				if (iph->protocol == IPPROTO_TCP) {
657					pull_size = ctx->eth_ip_hdr_size +
658						    sizeof(struct tcphdr);
659
660					if (unlikely(!pskb_may_pull(skb,
661								pull_size))) {
662						goto err;
663					}
664					ctx->l4_hdr_size = ((struct tcphdr *)
665					   skb_transport_header(skb))->doff * 4;
666				} else if (iph->protocol == IPPROTO_UDP) {
667					ctx->l4_hdr_size =
668							sizeof(struct udphdr);
669				} else {
670					ctx->l4_hdr_size = 0;
671				}
672			} else {
673				/* for simplicity, don't copy L4 headers */
674				ctx->l4_hdr_size = 0;
675			}
676			ctx->copy_size = ctx->eth_ip_hdr_size +
677					 ctx->l4_hdr_size;
678		} else {
679			ctx->eth_ip_hdr_size = 0;
680			ctx->l4_hdr_size = 0;
681			/* copy as much as allowed */
682			ctx->copy_size = min((unsigned int)VMXNET3_HDR_COPY_SIZE
683					     , skb_headlen(skb));
684		}
685
686		/* make sure headers are accessible directly */
687		if (unlikely(!pskb_may_pull(skb, ctx->copy_size)))
688			goto err;
689	}
690
691	if (unlikely(ctx->copy_size > VMXNET3_HDR_COPY_SIZE)) {
692		tq->stats.oversized_hdr++;
693		ctx->copy_size = 0;
694		return 0;
695	}
696
697	tdd = tq->data_ring.base + tq->tx_ring.next2fill;
698
699	memcpy(tdd->data, skb->data, ctx->copy_size);
700	dprintk(KERN_ERR "copy %u bytes to dataRing[%u]\n",
701		ctx->copy_size, tq->tx_ring.next2fill);
702	return 1;
703
704err:
705	return -1;
706}
707
708
709static void
710vmxnet3_prepare_tso(struct sk_buff *skb,
711		    struct vmxnet3_tx_ctx *ctx)
712{
713	struct tcphdr *tcph = (struct tcphdr *)skb_transport_header(skb);
714	if (ctx->ipv4) {
715		struct iphdr *iph = (struct iphdr *)skb_network_header(skb);
716		iph->check = 0;
717		tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, 0,
718						 IPPROTO_TCP, 0);
719	} else {
720		struct ipv6hdr *iph = (struct ipv6hdr *)skb_network_header(skb);
721		tcph->check = ~csum_ipv6_magic(&iph->saddr, &iph->daddr, 0,
722					       IPPROTO_TCP, 0);
723	}
724}
725
726
727/*
728 * Transmits a pkt thru a given tq
729 * Returns:
730 *    NETDEV_TX_OK:      descriptors are setup successfully
731 *    NETDEV_TX_OK:      error occured, the pkt is dropped
732 *    NETDEV_TX_BUSY:    tx ring is full, queue is stopped
733 *
734 * Side-effects:
735 *    1. tx ring may be changed
736 *    2. tq stats may be updated accordingly
737 *    3. shared->txNumDeferred may be updated
738 */
739
740static int
741vmxnet3_tq_xmit(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
742		struct vmxnet3_adapter *adapter, struct net_device *netdev)
743{
744	int ret;
745	u32 count;
746	unsigned long flags;
747	struct vmxnet3_tx_ctx ctx;
748	union Vmxnet3_GenericDesc *gdesc;
749
750	/* conservatively estimate # of descriptors to use */
751	count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) +
752		skb_shinfo(skb)->nr_frags + 1;
753
754	ctx.ipv4 = (skb->protocol == __constant_ntohs(ETH_P_IP));
755
756	ctx.mss = skb_shinfo(skb)->gso_size;
757	if (ctx.mss) {
758		if (skb_header_cloned(skb)) {
759			if (unlikely(pskb_expand_head(skb, 0, 0,
760						      GFP_ATOMIC) != 0)) {
761				tq->stats.drop_tso++;
762				goto drop_pkt;
763			}
764			tq->stats.copy_skb_header++;
765		}
766		vmxnet3_prepare_tso(skb, &ctx);
767	} else {
768		if (unlikely(count > VMXNET3_MAX_TXD_PER_PKT)) {
769
770			/* non-tso pkts must not use more than
771			 * VMXNET3_MAX_TXD_PER_PKT entries
772			 */
773			if (skb_linearize(skb) != 0) {
774				tq->stats.drop_too_many_frags++;
775				goto drop_pkt;
776			}
777			tq->stats.linearized++;
778
779			/* recalculate the # of descriptors to use */
780			count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) + 1;
781		}
782	}
783
784	ret = vmxnet3_parse_and_copy_hdr(skb, tq, &ctx, adapter);
785	if (ret >= 0) {
786		BUG_ON(ret <= 0 && ctx.copy_size != 0);
787		/* hdrs parsed, check against other limits */
788		if (ctx.mss) {
789			if (unlikely(ctx.eth_ip_hdr_size + ctx.l4_hdr_size >
790				     VMXNET3_MAX_TX_BUF_SIZE)) {
791				goto hdr_too_big;
792			}
793		} else {
794			if (skb->ip_summed == CHECKSUM_PARTIAL) {
795				if (unlikely(ctx.eth_ip_hdr_size +
796					     skb->csum_offset >
797					     VMXNET3_MAX_CSUM_OFFSET)) {
798					goto hdr_too_big;
799				}
800			}
801		}
802	} else {
803		tq->stats.drop_hdr_inspect_err++;
804		goto drop_pkt;
805	}
806
807	spin_lock_irqsave(&tq->tx_lock, flags);
808
809	if (count > vmxnet3_cmd_ring_desc_avail(&tq->tx_ring)) {
810		tq->stats.tx_ring_full++;
811		dprintk(KERN_ERR "tx queue stopped on %s, next2comp %u"
812			" next2fill %u\n", adapter->netdev->name,
813			tq->tx_ring.next2comp, tq->tx_ring.next2fill);
814
815		vmxnet3_tq_stop(tq, adapter);
816		spin_unlock_irqrestore(&tq->tx_lock, flags);
817		return NETDEV_TX_BUSY;
818	}
819
820	/* fill tx descs related to addr & len */
821	vmxnet3_map_pkt(skb, &ctx, tq, adapter->pdev, adapter);
822
823	/* setup the EOP desc */
824	ctx.eop_txd->dword[3] = VMXNET3_TXD_CQ | VMXNET3_TXD_EOP;
825
826	/* setup the SOP desc */
827	gdesc = ctx.sop_txd;
828	if (ctx.mss) {
829		gdesc->txd.hlen = ctx.eth_ip_hdr_size + ctx.l4_hdr_size;
830		gdesc->txd.om = VMXNET3_OM_TSO;
831		gdesc->txd.msscof = ctx.mss;
832		tq->shared->txNumDeferred += (skb->len - gdesc->txd.hlen +
833					     ctx.mss - 1) / ctx.mss;
834	} else {
835		if (skb->ip_summed == CHECKSUM_PARTIAL) {
836			gdesc->txd.hlen = ctx.eth_ip_hdr_size;
837			gdesc->txd.om = VMXNET3_OM_CSUM;
838			gdesc->txd.msscof = ctx.eth_ip_hdr_size +
839					    skb->csum_offset;
840		} else {
841			gdesc->txd.om = 0;
842			gdesc->txd.msscof = 0;
843		}
844		tq->shared->txNumDeferred++;
845	}
846
847	if (vlan_tx_tag_present(skb)) {
848		gdesc->txd.ti = 1;
849		gdesc->txd.tci = vlan_tx_tag_get(skb);
850	}
851
852	wmb();
853
854	/* finally flips the GEN bit of the SOP desc */
855	gdesc->dword[2] ^= VMXNET3_TXD_GEN;
856	dprintk(KERN_ERR "txd[%u]: SOP 0x%Lx 0x%x 0x%x\n",
857		(u32)((union Vmxnet3_GenericDesc *)ctx.sop_txd -
858		tq->tx_ring.base), gdesc->txd.addr, gdesc->dword[2],
859		gdesc->dword[3]);
860
861	spin_unlock_irqrestore(&tq->tx_lock, flags);
862
863	if (tq->shared->txNumDeferred >= tq->shared->txThreshold) {
864		tq->shared->txNumDeferred = 0;
865		VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_TXPROD,
866				       tq->tx_ring.next2fill);
867	}
868	netdev->trans_start = jiffies;
869
870	return NETDEV_TX_OK;
871
872hdr_too_big:
873	tq->stats.drop_oversized_hdr++;
874drop_pkt:
875	tq->stats.drop_total++;
876	dev_kfree_skb(skb);
877	return NETDEV_TX_OK;
878}
879
880
881static netdev_tx_t
882vmxnet3_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
883{
884	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
885	struct vmxnet3_tx_queue *tq = &adapter->tx_queue;
886
887	return vmxnet3_tq_xmit(skb, tq, adapter, netdev);
888}
889
890
891static void
892vmxnet3_rx_csum(struct vmxnet3_adapter *adapter,
893		struct sk_buff *skb,
894		union Vmxnet3_GenericDesc *gdesc)
895{
896	if (!gdesc->rcd.cnc && adapter->rxcsum) {
897		/* typical case: TCP/UDP over IP and both csums are correct */
898		if ((gdesc->dword[3] & VMXNET3_RCD_CSUM_OK) ==
899							VMXNET3_RCD_CSUM_OK) {
900			skb->ip_summed = CHECKSUM_UNNECESSARY;
901			BUG_ON(!(gdesc->rcd.tcp || gdesc->rcd.udp));
902			BUG_ON(!(gdesc->rcd.v4  || gdesc->rcd.v6));
903			BUG_ON(gdesc->rcd.frg);
904		} else {
905			if (gdesc->rcd.csum) {
906				skb->csum = htons(gdesc->rcd.csum);
907				skb->ip_summed = CHECKSUM_PARTIAL;
908			} else {
909				skb->ip_summed = CHECKSUM_NONE;
910			}
911		}
912	} else {
913		skb->ip_summed = CHECKSUM_NONE;
914	}
915}
916
917
918static void
919vmxnet3_rx_error(struct vmxnet3_rx_queue *rq, struct Vmxnet3_RxCompDesc *rcd,
920		 struct vmxnet3_rx_ctx *ctx,  struct vmxnet3_adapter *adapter)
921{
922	rq->stats.drop_err++;
923	if (!rcd->fcs)
924		rq->stats.drop_fcs++;
925
926	rq->stats.drop_total++;
927
928	/*
929	 * We do not unmap and chain the rx buffer to the skb.
930	 * We basically pretend this buffer is not used and will be recycled
931	 * by vmxnet3_rq_alloc_rx_buf()
932	 */
933
934	/*
935	 * ctx->skb may be NULL if this is the first and the only one
936	 * desc for the pkt
937	 */
938	if (ctx->skb)
939		dev_kfree_skb_irq(ctx->skb);
940
941	ctx->skb = NULL;
942}
943
944
945static int
946vmxnet3_rq_rx_complete(struct vmxnet3_rx_queue *rq,
947		       struct vmxnet3_adapter *adapter, int quota)
948{
949	static u32 rxprod_reg[2] = {VMXNET3_REG_RXPROD, VMXNET3_REG_RXPROD2};
950	u32 num_rxd = 0;
951	struct Vmxnet3_RxCompDesc *rcd;
952	struct vmxnet3_rx_ctx *ctx = &rq->rx_ctx;
953
954	rcd = &rq->comp_ring.base[rq->comp_ring.next2proc].rcd;
955	while (rcd->gen == rq->comp_ring.gen) {
956		struct vmxnet3_rx_buf_info *rbi;
957		struct sk_buff *skb;
958		int num_to_alloc;
959		struct Vmxnet3_RxDesc *rxd;
960		u32 idx, ring_idx;
961
962		if (num_rxd >= quota) {
963			/* we may stop even before we see the EOP desc of
964			 * the current pkt
965			 */
966			break;
967		}
968		num_rxd++;
969
970		idx = rcd->rxdIdx;
971		ring_idx = rcd->rqID == rq->qid ? 0 : 1;
972
973		rxd = &rq->rx_ring[ring_idx].base[idx].rxd;
974		rbi = rq->buf_info[ring_idx] + idx;
975
976		BUG_ON(rxd->addr != rbi->dma_addr || rxd->len != rbi->len);
977
978		if (unlikely(rcd->eop && rcd->err)) {
979			vmxnet3_rx_error(rq, rcd, ctx, adapter);
980			goto rcd_done;
981		}
982
983		if (rcd->sop) { /* first buf of the pkt */
984			BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_HEAD ||
985			       rcd->rqID != rq->qid);
986
987			BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_SKB);
988			BUG_ON(ctx->skb != NULL || rbi->skb == NULL);
989
990			if (unlikely(rcd->len == 0)) {
991				/* Pretend the rx buffer is skipped. */
992				BUG_ON(!(rcd->sop && rcd->eop));
993				dprintk(KERN_ERR "rxRing[%u][%u] 0 length\n",
994					ring_idx, idx);
995				goto rcd_done;
996			}
997
998			ctx->skb = rbi->skb;
999			rbi->skb = NULL;
1000
1001			pci_unmap_single(adapter->pdev, rbi->dma_addr, rbi->len,
1002					 PCI_DMA_FROMDEVICE);
1003
1004			skb_put(ctx->skb, rcd->len);
1005		} else {
1006			BUG_ON(ctx->skb == NULL);
1007			/* non SOP buffer must be type 1 in most cases */
1008			if (rbi->buf_type == VMXNET3_RX_BUF_PAGE) {
1009				BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_BODY);
1010
1011				if (rcd->len) {
1012					pci_unmap_page(adapter->pdev,
1013						       rbi->dma_addr, rbi->len,
1014						       PCI_DMA_FROMDEVICE);
1015
1016					vmxnet3_append_frag(ctx->skb, rcd, rbi);
1017					rbi->page = NULL;
1018				}
1019			} else {
1020				/*
1021				 * The only time a non-SOP buffer is type 0 is
1022				 * when it's EOP and error flag is raised, which
1023				 * has already been handled.
1024				 */
1025				BUG_ON(true);
1026			}
1027		}
1028
1029		skb = ctx->skb;
1030		if (rcd->eop) {
1031			skb->len += skb->data_len;
1032			skb->truesize += skb->data_len;
1033
1034			vmxnet3_rx_csum(adapter, skb,
1035					(union Vmxnet3_GenericDesc *)rcd);
1036			skb->protocol = eth_type_trans(skb, adapter->netdev);
1037
1038			if (unlikely(adapter->vlan_grp && rcd->ts)) {
1039				vlan_hwaccel_receive_skb(skb,
1040						adapter->vlan_grp, rcd->tci);
1041			} else {
1042				netif_receive_skb(skb);
1043			}
1044
1045			adapter->netdev->last_rx = jiffies;
1046			ctx->skb = NULL;
1047		}
1048
1049rcd_done:
1050		/* device may skip some rx descs */
1051		rq->rx_ring[ring_idx].next2comp = idx;
1052		VMXNET3_INC_RING_IDX_ONLY(rq->rx_ring[ring_idx].next2comp,
1053					  rq->rx_ring[ring_idx].size);
1054
1055		/* refill rx buffers frequently to avoid starving the h/w */
1056		num_to_alloc = vmxnet3_cmd_ring_desc_avail(rq->rx_ring +
1057							   ring_idx);
1058		if (unlikely(num_to_alloc > VMXNET3_RX_ALLOC_THRESHOLD(rq,
1059							ring_idx, adapter))) {
1060			vmxnet3_rq_alloc_rx_buf(rq, ring_idx, num_to_alloc,
1061						adapter);
1062
1063			/* if needed, update the register */
1064			if (unlikely(rq->shared->updateRxProd)) {
1065				VMXNET3_WRITE_BAR0_REG(adapter,
1066					rxprod_reg[ring_idx] + rq->qid * 8,
1067					rq->rx_ring[ring_idx].next2fill);
1068				rq->uncommitted[ring_idx] = 0;
1069			}
1070		}
1071
1072		vmxnet3_comp_ring_adv_next2proc(&rq->comp_ring);
1073		rcd = &rq->comp_ring.base[rq->comp_ring.next2proc].rcd;
1074	}
1075
1076	return num_rxd;
1077}
1078
1079
1080static void
1081vmxnet3_rq_cleanup(struct vmxnet3_rx_queue *rq,
1082		   struct vmxnet3_adapter *adapter)
1083{
1084	u32 i, ring_idx;
1085	struct Vmxnet3_RxDesc *rxd;
1086
1087	for (ring_idx = 0; ring_idx < 2; ring_idx++) {
1088		for (i = 0; i < rq->rx_ring[ring_idx].size; i++) {
1089			rxd = &rq->rx_ring[ring_idx].base[i].rxd;
1090
1091			if (rxd->btype == VMXNET3_RXD_BTYPE_HEAD &&
1092					rq->buf_info[ring_idx][i].skb) {
1093				pci_unmap_single(adapter->pdev, rxd->addr,
1094						 rxd->len, PCI_DMA_FROMDEVICE);
1095				dev_kfree_skb(rq->buf_info[ring_idx][i].skb);
1096				rq->buf_info[ring_idx][i].skb = NULL;
1097			} else if (rxd->btype == VMXNET3_RXD_BTYPE_BODY &&
1098					rq->buf_info[ring_idx][i].page) {
1099				pci_unmap_page(adapter->pdev, rxd->addr,
1100					       rxd->len, PCI_DMA_FROMDEVICE);
1101				put_page(rq->buf_info[ring_idx][i].page);
1102				rq->buf_info[ring_idx][i].page = NULL;
1103			}
1104		}
1105
1106		rq->rx_ring[ring_idx].gen = VMXNET3_INIT_GEN;
1107		rq->rx_ring[ring_idx].next2fill =
1108					rq->rx_ring[ring_idx].next2comp = 0;
1109		rq->uncommitted[ring_idx] = 0;
1110	}
1111
1112	rq->comp_ring.gen = VMXNET3_INIT_GEN;
1113	rq->comp_ring.next2proc = 0;
1114}
1115
1116
1117void vmxnet3_rq_destroy(struct vmxnet3_rx_queue *rq,
1118			struct vmxnet3_adapter *adapter)
1119{
1120	int i;
1121	int j;
1122
1123	/* all rx buffers must have already been freed */
1124	for (i = 0; i < 2; i++) {
1125		if (rq->buf_info[i]) {
1126			for (j = 0; j < rq->rx_ring[i].size; j++)
1127				BUG_ON(rq->buf_info[i][j].page != NULL);
1128		}
1129	}
1130
1131
1132	kfree(rq->buf_info[0]);
1133
1134	for (i = 0; i < 2; i++) {
1135		if (rq->rx_ring[i].base) {
1136			pci_free_consistent(adapter->pdev, rq->rx_ring[i].size
1137					    * sizeof(struct Vmxnet3_RxDesc),
1138					    rq->rx_ring[i].base,
1139					    rq->rx_ring[i].basePA);
1140			rq->rx_ring[i].base = NULL;
1141		}
1142		rq->buf_info[i] = NULL;
1143	}
1144
1145	if (rq->comp_ring.base) {
1146		pci_free_consistent(adapter->pdev, rq->comp_ring.size *
1147				    sizeof(struct Vmxnet3_RxCompDesc),
1148				    rq->comp_ring.base, rq->comp_ring.basePA);
1149		rq->comp_ring.base = NULL;
1150	}
1151}
1152
1153
1154static int
1155vmxnet3_rq_init(struct vmxnet3_rx_queue *rq,
1156		struct vmxnet3_adapter  *adapter)
1157{
1158	int i;
1159
1160	/* initialize buf_info */
1161	for (i = 0; i < rq->rx_ring[0].size; i++) {
1162
1163		/* 1st buf for a pkt is skbuff */
1164		if (i % adapter->rx_buf_per_pkt == 0) {
1165			rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_SKB;
1166			rq->buf_info[0][i].len = adapter->skb_buf_size;
1167		} else { /* subsequent bufs for a pkt is frag */
1168			rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_PAGE;
1169			rq->buf_info[0][i].len = PAGE_SIZE;
1170		}
1171	}
1172	for (i = 0; i < rq->rx_ring[1].size; i++) {
1173		rq->buf_info[1][i].buf_type = VMXNET3_RX_BUF_PAGE;
1174		rq->buf_info[1][i].len = PAGE_SIZE;
1175	}
1176
1177	/* reset internal state and allocate buffers for both rings */
1178	for (i = 0; i < 2; i++) {
1179		rq->rx_ring[i].next2fill = rq->rx_ring[i].next2comp = 0;
1180		rq->uncommitted[i] = 0;
1181
1182		memset(rq->rx_ring[i].base, 0, rq->rx_ring[i].size *
1183		       sizeof(struct Vmxnet3_RxDesc));
1184		rq->rx_ring[i].gen = VMXNET3_INIT_GEN;
1185	}
1186	if (vmxnet3_rq_alloc_rx_buf(rq, 0, rq->rx_ring[0].size - 1,
1187				    adapter) == 0) {
1188		/* at least has 1 rx buffer for the 1st ring */
1189		return -ENOMEM;
1190	}
1191	vmxnet3_rq_alloc_rx_buf(rq, 1, rq->rx_ring[1].size - 1, adapter);
1192
1193	/* reset the comp ring */
1194	rq->comp_ring.next2proc = 0;
1195	memset(rq->comp_ring.base, 0, rq->comp_ring.size *
1196	       sizeof(struct Vmxnet3_RxCompDesc));
1197	rq->comp_ring.gen = VMXNET3_INIT_GEN;
1198
1199	/* reset rxctx */
1200	rq->rx_ctx.skb = NULL;
1201
1202	/* stats are not reset */
1203	return 0;
1204}
1205
1206
1207static int
1208vmxnet3_rq_create(struct vmxnet3_rx_queue *rq, struct vmxnet3_adapter *adapter)
1209{
1210	int i;
1211	size_t sz;
1212	struct vmxnet3_rx_buf_info *bi;
1213
1214	for (i = 0; i < 2; i++) {
1215
1216		sz = rq->rx_ring[i].size * sizeof(struct Vmxnet3_RxDesc);
1217		rq->rx_ring[i].base = pci_alloc_consistent(adapter->pdev, sz,
1218							&rq->rx_ring[i].basePA);
1219		if (!rq->rx_ring[i].base) {
1220			printk(KERN_ERR "%s: failed to allocate rx ring %d\n",
1221			       adapter->netdev->name, i);
1222			goto err;
1223		}
1224	}
1225
1226	sz = rq->comp_ring.size * sizeof(struct Vmxnet3_RxCompDesc);
1227	rq->comp_ring.base = pci_alloc_consistent(adapter->pdev, sz,
1228						  &rq->comp_ring.basePA);
1229	if (!rq->comp_ring.base) {
1230		printk(KERN_ERR "%s: failed to allocate rx comp ring\n",
1231		       adapter->netdev->name);
1232		goto err;
1233	}
1234
1235	sz = sizeof(struct vmxnet3_rx_buf_info) * (rq->rx_ring[0].size +
1236						   rq->rx_ring[1].size);
1237	bi = kmalloc(sz, GFP_KERNEL);
1238	if (!bi) {
1239		printk(KERN_ERR "%s: failed to allocate rx bufinfo\n",
1240		       adapter->netdev->name);
1241		goto err;
1242	}
1243	memset(bi, 0, sz);
1244	rq->buf_info[0] = bi;
1245	rq->buf_info[1] = bi + rq->rx_ring[0].size;
1246
1247	return 0;
1248
1249err:
1250	vmxnet3_rq_destroy(rq, adapter);
1251	return -ENOMEM;
1252}
1253
1254
1255static int
1256vmxnet3_do_poll(struct vmxnet3_adapter *adapter, int budget)
1257{
1258	if (unlikely(adapter->shared->ecr))
1259		vmxnet3_process_events(adapter);
1260
1261	vmxnet3_tq_tx_complete(&adapter->tx_queue, adapter);
1262	return vmxnet3_rq_rx_complete(&adapter->rx_queue, adapter, budget);
1263}
1264
1265
1266static int
1267vmxnet3_poll(struct napi_struct *napi, int budget)
1268{
1269	struct vmxnet3_adapter *adapter = container_of(napi,
1270					  struct vmxnet3_adapter, napi);
1271	int rxd_done;
1272
1273	rxd_done = vmxnet3_do_poll(adapter, budget);
1274
1275	if (rxd_done < budget) {
1276		napi_complete(napi);
1277		vmxnet3_enable_intr(adapter, 0);
1278	}
1279	return rxd_done;
1280}
1281
1282
1283/* Interrupt handler for vmxnet3  */
1284static irqreturn_t
1285vmxnet3_intr(int irq, void *dev_id)
1286{
1287	struct net_device *dev = dev_id;
1288	struct vmxnet3_adapter *adapter = netdev_priv(dev);
1289
1290	if (unlikely(adapter->intr.type == VMXNET3_IT_INTX)) {
1291		u32 icr = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_ICR);
1292		if (unlikely(icr == 0))
1293			/* not ours */
1294			return IRQ_NONE;
1295	}
1296
1297
1298	/* disable intr if needed */
1299	if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1300		vmxnet3_disable_intr(adapter, 0);
1301
1302	napi_schedule(&adapter->napi);
1303
1304	return IRQ_HANDLED;
1305}
1306
1307#ifdef CONFIG_NET_POLL_CONTROLLER
1308
1309
1310/* netpoll callback. */
1311static void
1312vmxnet3_netpoll(struct net_device *netdev)
1313{
1314	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1315	int irq;
1316
1317#ifdef CONFIG_PCI_MSI
1318	if (adapter->intr.type == VMXNET3_IT_MSIX)
1319		irq = adapter->intr.msix_entries[0].vector;
1320	else
1321#endif
1322		irq = adapter->pdev->irq;
1323
1324	disable_irq(irq);
1325	vmxnet3_intr(irq, netdev);
1326	enable_irq(irq);
1327}
1328#endif
1329
1330static int
1331vmxnet3_request_irqs(struct vmxnet3_adapter *adapter)
1332{
1333	int err;
1334
1335#ifdef CONFIG_PCI_MSI
1336	if (adapter->intr.type == VMXNET3_IT_MSIX) {
1337		/* we only use 1 MSI-X vector */
1338		err = request_irq(adapter->intr.msix_entries[0].vector,
1339				  vmxnet3_intr, 0, adapter->netdev->name,
1340				  adapter->netdev);
1341	} else
1342#endif
1343	if (adapter->intr.type == VMXNET3_IT_MSI) {
1344		err = request_irq(adapter->pdev->irq, vmxnet3_intr, 0,
1345				  adapter->netdev->name, adapter->netdev);
1346	} else {
1347		err = request_irq(adapter->pdev->irq, vmxnet3_intr,
1348				  IRQF_SHARED, adapter->netdev->name,
1349				  adapter->netdev);
1350	}
1351
1352	if (err)
1353		printk(KERN_ERR "Failed to request irq %s (intr type:%d), error"
1354		       ":%d\n", adapter->netdev->name, adapter->intr.type, err);
1355
1356
1357	if (!err) {
1358		int i;
1359		/* init our intr settings */
1360		for (i = 0; i < adapter->intr.num_intrs; i++)
1361			adapter->intr.mod_levels[i] = UPT1_IML_ADAPTIVE;
1362
1363		/* next setup intr index for all intr sources */
1364		adapter->tx_queue.comp_ring.intr_idx = 0;
1365		adapter->rx_queue.comp_ring.intr_idx = 0;
1366		adapter->intr.event_intr_idx = 0;
1367
1368		printk(KERN_INFO "%s: intr type %u, mode %u, %u vectors "
1369		       "allocated\n", adapter->netdev->name, adapter->intr.type,
1370		       adapter->intr.mask_mode, adapter->intr.num_intrs);
1371	}
1372
1373	return err;
1374}
1375
1376
1377static void
1378vmxnet3_free_irqs(struct vmxnet3_adapter *adapter)
1379{
1380	BUG_ON(adapter->intr.type == VMXNET3_IT_AUTO ||
1381	       adapter->intr.num_intrs <= 0);
1382
1383	switch (adapter->intr.type) {
1384#ifdef CONFIG_PCI_MSI
1385	case VMXNET3_IT_MSIX:
1386	{
1387		int i;
1388
1389		for (i = 0; i < adapter->intr.num_intrs; i++)
1390			free_irq(adapter->intr.msix_entries[i].vector,
1391				 adapter->netdev);
1392		break;
1393	}
1394#endif
1395	case VMXNET3_IT_MSI:
1396		free_irq(adapter->pdev->irq, adapter->netdev);
1397		break;
1398	case VMXNET3_IT_INTX:
1399		free_irq(adapter->pdev->irq, adapter->netdev);
1400		break;
1401	default:
1402		BUG_ON(true);
1403	}
1404}
1405
1406
1407static void
1408vmxnet3_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp)
1409{
1410	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1411	struct Vmxnet3_DriverShared *shared = adapter->shared;
1412	u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1413
1414	if (grp) {
1415		/* add vlan rx stripping. */
1416		if (adapter->netdev->features & NETIF_F_HW_VLAN_RX) {
1417			int i;
1418			struct Vmxnet3_DSDevRead *devRead = &shared->devRead;
1419			adapter->vlan_grp = grp;
1420
1421			/* update FEATURES to device */
1422			devRead->misc.uptFeatures |= UPT1_F_RXVLAN;
1423			VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1424					       VMXNET3_CMD_UPDATE_FEATURE);
1425			/*
1426			 *  Clear entire vfTable; then enable untagged pkts.
1427			 *  Note: setting one entry in vfTable to non-zero turns
1428			 *  on VLAN rx filtering.
1429			 */
1430			for (i = 0; i < VMXNET3_VFT_SIZE; i++)
1431				vfTable[i] = 0;
1432
1433			VMXNET3_SET_VFTABLE_ENTRY(vfTable, 0);
1434			VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1435					       VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1436		} else {
1437			printk(KERN_ERR "%s: vlan_rx_register when device has "
1438			       "no NETIF_F_HW_VLAN_RX\n", netdev->name);
1439		}
1440	} else {
1441		/* remove vlan rx stripping. */
1442		struct Vmxnet3_DSDevRead *devRead = &shared->devRead;
1443		adapter->vlan_grp = NULL;
1444
1445		if (devRead->misc.uptFeatures & UPT1_F_RXVLAN) {
1446			int i;
1447
1448			for (i = 0; i < VMXNET3_VFT_SIZE; i++) {
1449				/* clear entire vfTable; this also disables
1450				 * VLAN rx filtering
1451				 */
1452				vfTable[i] = 0;
1453			}
1454			VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1455					       VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1456
1457			/* update FEATURES to device */
1458			devRead->misc.uptFeatures &= ~UPT1_F_RXVLAN;
1459			VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1460					       VMXNET3_CMD_UPDATE_FEATURE);
1461		}
1462	}
1463}
1464
1465
1466static void
1467vmxnet3_restore_vlan(struct vmxnet3_adapter *adapter)
1468{
1469	if (adapter->vlan_grp) {
1470		u16 vid;
1471		u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1472		bool activeVlan = false;
1473
1474		for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
1475			if (vlan_group_get_device(adapter->vlan_grp, vid)) {
1476				VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid);
1477				activeVlan = true;
1478			}
1479		}
1480		if (activeVlan) {
1481			/* continue to allow untagged pkts */
1482			VMXNET3_SET_VFTABLE_ENTRY(vfTable, 0);
1483		}
1484	}
1485}
1486
1487
1488static void
1489vmxnet3_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1490{
1491	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1492	u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1493
1494	VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid);
1495	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1496			       VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1497}
1498
1499
1500static void
1501vmxnet3_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1502{
1503	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1504	u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1505
1506	VMXNET3_CLEAR_VFTABLE_ENTRY(vfTable, vid);
1507	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1508			       VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1509}
1510
1511
1512static u8 *
1513vmxnet3_copy_mc(struct net_device *netdev)
1514{
1515	u8 *buf = NULL;
1516	u32 sz = netdev->mc_count * ETH_ALEN;
1517
1518	/* struct Vmxnet3_RxFilterConf.mfTableLen is u16. */
1519	if (sz <= 0xffff) {
1520		/* We may be called with BH disabled */
1521		buf = kmalloc(sz, GFP_ATOMIC);
1522		if (buf) {
1523			int i;
1524			struct dev_mc_list *mc = netdev->mc_list;
1525
1526			for (i = 0; i < netdev->mc_count; i++) {
1527				BUG_ON(!mc);
1528				memcpy(buf + i * ETH_ALEN, mc->dmi_addr,
1529				       ETH_ALEN);
1530				mc = mc->next;
1531			}
1532		}
1533	}
1534	return buf;
1535}
1536
1537
1538static void
1539vmxnet3_set_mc(struct net_device *netdev)
1540{
1541	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1542	struct Vmxnet3_RxFilterConf *rxConf =
1543					&adapter->shared->devRead.rxFilterConf;
1544	u8 *new_table = NULL;
1545	u32 new_mode = VMXNET3_RXM_UCAST;
1546
1547	if (netdev->flags & IFF_PROMISC)
1548		new_mode |= VMXNET3_RXM_PROMISC;
1549
1550	if (netdev->flags & IFF_BROADCAST)
1551		new_mode |= VMXNET3_RXM_BCAST;
1552
1553	if (netdev->flags & IFF_ALLMULTI)
1554		new_mode |= VMXNET3_RXM_ALL_MULTI;
1555	else
1556		if (netdev->mc_count > 0) {
1557			new_table = vmxnet3_copy_mc(netdev);
1558			if (new_table) {
1559				new_mode |= VMXNET3_RXM_MCAST;
1560				rxConf->mfTableLen = netdev->mc_count *
1561						     ETH_ALEN;
1562				rxConf->mfTablePA = virt_to_phys(new_table);
1563			} else {
1564				printk(KERN_INFO "%s: failed to copy mcast list"
1565				       ", setting ALL_MULTI\n", netdev->name);
1566				new_mode |= VMXNET3_RXM_ALL_MULTI;
1567			}
1568		}
1569
1570
1571	if (!(new_mode & VMXNET3_RXM_MCAST)) {
1572		rxConf->mfTableLen = 0;
1573		rxConf->mfTablePA = 0;
1574	}
1575
1576	if (new_mode != rxConf->rxMode) {
1577		rxConf->rxMode = new_mode;
1578		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1579				       VMXNET3_CMD_UPDATE_RX_MODE);
1580	}
1581
1582	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1583			       VMXNET3_CMD_UPDATE_MAC_FILTERS);
1584
1585	kfree(new_table);
1586}
1587
1588
1589/*
1590 *   Set up driver_shared based on settings in adapter.
1591 */
1592
1593static void
1594vmxnet3_setup_driver_shared(struct vmxnet3_adapter *adapter)
1595{
1596	struct Vmxnet3_DriverShared *shared = adapter->shared;
1597	struct Vmxnet3_DSDevRead *devRead = &shared->devRead;
1598	struct Vmxnet3_TxQueueConf *tqc;
1599	struct Vmxnet3_RxQueueConf *rqc;
1600	int i;
1601
1602	memset(shared, 0, sizeof(*shared));
1603
1604	/* driver settings */
1605	shared->magic = VMXNET3_REV1_MAGIC;
1606	devRead->misc.driverInfo.version = VMXNET3_DRIVER_VERSION_NUM;
1607	devRead->misc.driverInfo.gos.gosBits = (sizeof(void *) == 4 ?
1608				VMXNET3_GOS_BITS_32 : VMXNET3_GOS_BITS_64);
1609	devRead->misc.driverInfo.gos.gosType = VMXNET3_GOS_TYPE_LINUX;
1610	devRead->misc.driverInfo.vmxnet3RevSpt = 1;
1611	devRead->misc.driverInfo.uptVerSpt = 1;
1612
1613	devRead->misc.ddPA = virt_to_phys(adapter);
1614	devRead->misc.ddLen = sizeof(struct vmxnet3_adapter);
1615
1616	/* set up feature flags */
1617	if (adapter->rxcsum)
1618		devRead->misc.uptFeatures |= UPT1_F_RXCSUM;
1619
1620	if (adapter->lro) {
1621		devRead->misc.uptFeatures |= UPT1_F_LRO;
1622		devRead->misc.maxNumRxSG = 1 + MAX_SKB_FRAGS;
1623	}
1624	if ((adapter->netdev->features & NETIF_F_HW_VLAN_RX)
1625			&& adapter->vlan_grp) {
1626		devRead->misc.uptFeatures |= UPT1_F_RXVLAN;
1627	}
1628
1629	devRead->misc.mtu = adapter->netdev->mtu;
1630	devRead->misc.queueDescPA = adapter->queue_desc_pa;
1631	devRead->misc.queueDescLen = sizeof(struct Vmxnet3_TxQueueDesc) +
1632				     sizeof(struct Vmxnet3_RxQueueDesc);
1633
1634	/* tx queue settings */
1635	BUG_ON(adapter->tx_queue.tx_ring.base == NULL);
1636
1637	devRead->misc.numTxQueues = 1;
1638	tqc = &adapter->tqd_start->conf;
1639	tqc->txRingBasePA   = adapter->tx_queue.tx_ring.basePA;
1640	tqc->dataRingBasePA = adapter->tx_queue.data_ring.basePA;
1641	tqc->compRingBasePA = adapter->tx_queue.comp_ring.basePA;
1642	tqc->ddPA           = virt_to_phys(adapter->tx_queue.buf_info);
1643	tqc->txRingSize     = adapter->tx_queue.tx_ring.size;
1644	tqc->dataRingSize   = adapter->tx_queue.data_ring.size;
1645	tqc->compRingSize   = adapter->tx_queue.comp_ring.size;
1646	tqc->ddLen          = sizeof(struct vmxnet3_tx_buf_info) *
1647			      tqc->txRingSize;
1648	tqc->intrIdx        = adapter->tx_queue.comp_ring.intr_idx;
1649
1650	/* rx queue settings */
1651	devRead->misc.numRxQueues = 1;
1652	rqc = &adapter->rqd_start->conf;
1653	rqc->rxRingBasePA[0] = adapter->rx_queue.rx_ring[0].basePA;
1654	rqc->rxRingBasePA[1] = adapter->rx_queue.rx_ring[1].basePA;
1655	rqc->compRingBasePA  = adapter->rx_queue.comp_ring.basePA;
1656	rqc->ddPA            = virt_to_phys(adapter->rx_queue.buf_info);
1657	rqc->rxRingSize[0]   = adapter->rx_queue.rx_ring[0].size;
1658	rqc->rxRingSize[1]   = adapter->rx_queue.rx_ring[1].size;
1659	rqc->compRingSize    = adapter->rx_queue.comp_ring.size;
1660	rqc->ddLen           = sizeof(struct vmxnet3_rx_buf_info) *
1661			       (rqc->rxRingSize[0] + rqc->rxRingSize[1]);
1662	rqc->intrIdx         = adapter->rx_queue.comp_ring.intr_idx;
1663
1664	/* intr settings */
1665	devRead->intrConf.autoMask = adapter->intr.mask_mode ==
1666				     VMXNET3_IMM_AUTO;
1667	devRead->intrConf.numIntrs = adapter->intr.num_intrs;
1668	for (i = 0; i < adapter->intr.num_intrs; i++)
1669		devRead->intrConf.modLevels[i] = adapter->intr.mod_levels[i];
1670
1671	devRead->intrConf.eventIntrIdx = adapter->intr.event_intr_idx;
1672
1673	/* rx filter settings */
1674	devRead->rxFilterConf.rxMode = 0;
1675	vmxnet3_restore_vlan(adapter);
1676	/* the rest are already zeroed */
1677}
1678
1679
1680int
1681vmxnet3_activate_dev(struct vmxnet3_adapter *adapter)
1682{
1683	int err;
1684	u32 ret;
1685
1686	dprintk(KERN_ERR "%s: skb_buf_size %d, rx_buf_per_pkt %d, ring sizes"
1687		" %u %u %u\n", adapter->netdev->name, adapter->skb_buf_size,
1688		adapter->rx_buf_per_pkt, adapter->tx_queue.tx_ring.size,
1689		adapter->rx_queue.rx_ring[0].size,
1690		adapter->rx_queue.rx_ring[1].size);
1691
1692	vmxnet3_tq_init(&adapter->tx_queue, adapter);
1693	err = vmxnet3_rq_init(&adapter->rx_queue, adapter);
1694	if (err) {
1695		printk(KERN_ERR "Failed to init rx queue for %s: error %d\n",
1696		       adapter->netdev->name, err);
1697		goto rq_err;
1698	}
1699
1700	err = vmxnet3_request_irqs(adapter);
1701	if (err) {
1702		printk(KERN_ERR "Failed to setup irq for %s: error %d\n",
1703		       adapter->netdev->name, err);
1704		goto irq_err;
1705	}
1706
1707	vmxnet3_setup_driver_shared(adapter);
1708
1709	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL,
1710			       VMXNET3_GET_ADDR_LO(adapter->shared_pa));
1711	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH,
1712			       VMXNET3_GET_ADDR_HI(adapter->shared_pa));
1713
1714	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1715			       VMXNET3_CMD_ACTIVATE_DEV);
1716	ret = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
1717
1718	if (ret != 0) {
1719		printk(KERN_ERR "Failed to activate dev %s: error %u\n",
1720		       adapter->netdev->name, ret);
1721		err = -EINVAL;
1722		goto activate_err;
1723	}
1724	VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_RXPROD,
1725			       adapter->rx_queue.rx_ring[0].next2fill);
1726	VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_RXPROD2,
1727			       adapter->rx_queue.rx_ring[1].next2fill);
1728
1729	/* Apply the rx filter settins last. */
1730	vmxnet3_set_mc(adapter->netdev);
1731
1732	/*
1733	 * Check link state when first activating device. It will start the
1734	 * tx queue if the link is up.
1735	 */
1736	vmxnet3_check_link(adapter);
1737
1738	napi_enable(&adapter->napi);
1739	vmxnet3_enable_all_intrs(adapter);
1740	clear_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
1741	return 0;
1742
1743activate_err:
1744	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, 0);
1745	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, 0);
1746	vmxnet3_free_irqs(adapter);
1747irq_err:
1748rq_err:
1749	/* free up buffers we allocated */
1750	vmxnet3_rq_cleanup(&adapter->rx_queue, adapter);
1751	return err;
1752}
1753
1754
1755void
1756vmxnet3_reset_dev(struct vmxnet3_adapter *adapter)
1757{
1758	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, VMXNET3_CMD_RESET_DEV);
1759}
1760
1761
1762int
1763vmxnet3_quiesce_dev(struct vmxnet3_adapter *adapter)
1764{
1765	if (test_and_set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state))
1766		return 0;
1767
1768
1769	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1770			       VMXNET3_CMD_QUIESCE_DEV);
1771	vmxnet3_disable_all_intrs(adapter);
1772
1773	napi_disable(&adapter->napi);
1774	netif_tx_disable(adapter->netdev);
1775	adapter->link_speed = 0;
1776	netif_carrier_off(adapter->netdev);
1777
1778	vmxnet3_tq_cleanup(&adapter->tx_queue, adapter);
1779	vmxnet3_rq_cleanup(&adapter->rx_queue, adapter);
1780	vmxnet3_free_irqs(adapter);
1781	return 0;
1782}
1783
1784
1785static void
1786vmxnet3_write_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac)
1787{
1788	u32 tmp;
1789
1790	tmp = *(u32 *)mac;
1791	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACL, tmp);
1792
1793	tmp = (mac[5] << 8) | mac[4];
1794	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACH, tmp);
1795}
1796
1797
1798static int
1799vmxnet3_set_mac_addr(struct net_device *netdev, void *p)
1800{
1801	struct sockaddr *addr = p;
1802	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1803
1804	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1805	vmxnet3_write_mac_addr(adapter, addr->sa_data);
1806
1807	return 0;
1808}
1809
1810
1811/* ==================== initialization and cleanup routines ============ */
1812
1813static int
1814vmxnet3_alloc_pci_resources(struct vmxnet3_adapter *adapter, bool *dma64)
1815{
1816	int err;
1817	unsigned long mmio_start, mmio_len;
1818	struct pci_dev *pdev = adapter->pdev;
1819
1820	err = pci_enable_device(pdev);
1821	if (err) {
1822		printk(KERN_ERR "Failed to enable adapter %s: error %d\n",
1823		       pci_name(pdev), err);
1824		return err;
1825	}
1826
1827	if (pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) == 0) {
1828		if (pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64)) != 0) {
1829			printk(KERN_ERR "pci_set_consistent_dma_mask failed "
1830			       "for adapter %s\n", pci_name(pdev));
1831			err = -EIO;
1832			goto err_set_mask;
1833		}
1834		*dma64 = true;
1835	} else {
1836		if (pci_set_dma_mask(pdev, DMA_BIT_MASK(32)) != 0) {
1837			printk(KERN_ERR "pci_set_dma_mask failed for adapter "
1838			       "%s\n",	pci_name(pdev));
1839			err = -EIO;
1840			goto err_set_mask;
1841		}
1842		*dma64 = false;
1843	}
1844
1845	err = pci_request_selected_regions(pdev, (1 << 2) - 1,
1846					   vmxnet3_driver_name);
1847	if (err) {
1848		printk(KERN_ERR "Failed to request region for adapter %s: "
1849		       "error %d\n", pci_name(pdev), err);
1850		goto err_set_mask;
1851	}
1852
1853	pci_set_master(pdev);
1854
1855	mmio_start = pci_resource_start(pdev, 0);
1856	mmio_len = pci_resource_len(pdev, 0);
1857	adapter->hw_addr0 = ioremap(mmio_start, mmio_len);
1858	if (!adapter->hw_addr0) {
1859		printk(KERN_ERR "Failed to map bar0 for adapter %s\n",
1860		       pci_name(pdev));
1861		err = -EIO;
1862		goto err_ioremap;
1863	}
1864
1865	mmio_start = pci_resource_start(pdev, 1);
1866	mmio_len = pci_resource_len(pdev, 1);
1867	adapter->hw_addr1 = ioremap(mmio_start, mmio_len);
1868	if (!adapter->hw_addr1) {
1869		printk(KERN_ERR "Failed to map bar1 for adapter %s\n",
1870		       pci_name(pdev));
1871		err = -EIO;
1872		goto err_bar1;
1873	}
1874	return 0;
1875
1876err_bar1:
1877	iounmap(adapter->hw_addr0);
1878err_ioremap:
1879	pci_release_selected_regions(pdev, (1 << 2) - 1);
1880err_set_mask:
1881	pci_disable_device(pdev);
1882	return err;
1883}
1884
1885
1886static void
1887vmxnet3_free_pci_resources(struct vmxnet3_adapter *adapter)
1888{
1889	BUG_ON(!adapter->pdev);
1890
1891	iounmap(adapter->hw_addr0);
1892	iounmap(adapter->hw_addr1);
1893	pci_release_selected_regions(adapter->pdev, (1 << 2) - 1);
1894	pci_disable_device(adapter->pdev);
1895}
1896
1897
1898static void
1899vmxnet3_adjust_rx_ring_size(struct vmxnet3_adapter *adapter)
1900{
1901	size_t sz;
1902
1903	if (adapter->netdev->mtu <= VMXNET3_MAX_SKB_BUF_SIZE -
1904				    VMXNET3_MAX_ETH_HDR_SIZE) {
1905		adapter->skb_buf_size = adapter->netdev->mtu +
1906					VMXNET3_MAX_ETH_HDR_SIZE;
1907		if (adapter->skb_buf_size < VMXNET3_MIN_T0_BUF_SIZE)
1908			adapter->skb_buf_size = VMXNET3_MIN_T0_BUF_SIZE;
1909
1910		adapter->rx_buf_per_pkt = 1;
1911	} else {
1912		adapter->skb_buf_size = VMXNET3_MAX_SKB_BUF_SIZE;
1913		sz = adapter->netdev->mtu - VMXNET3_MAX_SKB_BUF_SIZE +
1914					    VMXNET3_MAX_ETH_HDR_SIZE;
1915		adapter->rx_buf_per_pkt = 1 + (sz + PAGE_SIZE - 1) / PAGE_SIZE;
1916	}
1917
1918	/*
1919	 * for simplicity, force the ring0 size to be a multiple of
1920	 * rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN
1921	 */
1922	sz = adapter->rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN;
1923	adapter->rx_queue.rx_ring[0].size = (adapter->rx_queue.rx_ring[0].size +
1924					     sz - 1) / sz * sz;
1925	adapter->rx_queue.rx_ring[0].size = min_t(u32,
1926					    adapter->rx_queue.rx_ring[0].size,
1927					    VMXNET3_RX_RING_MAX_SIZE / sz * sz);
1928}
1929
1930
1931int
1932vmxnet3_create_queues(struct vmxnet3_adapter *adapter, u32 tx_ring_size,
1933		      u32 rx_ring_size, u32 rx_ring2_size)
1934{
1935	int err;
1936
1937	adapter->tx_queue.tx_ring.size   = tx_ring_size;
1938	adapter->tx_queue.data_ring.size = tx_ring_size;
1939	adapter->tx_queue.comp_ring.size = tx_ring_size;
1940	adapter->tx_queue.shared = &adapter->tqd_start->ctrl;
1941	adapter->tx_queue.stopped = true;
1942	err = vmxnet3_tq_create(&adapter->tx_queue, adapter);
1943	if (err)
1944		return err;
1945
1946	adapter->rx_queue.rx_ring[0].size = rx_ring_size;
1947	adapter->rx_queue.rx_ring[1].size = rx_ring2_size;
1948	vmxnet3_adjust_rx_ring_size(adapter);
1949	adapter->rx_queue.comp_ring.size  = adapter->rx_queue.rx_ring[0].size +
1950					    adapter->rx_queue.rx_ring[1].size;
1951	adapter->rx_queue.qid  = 0;
1952	adapter->rx_queue.qid2 = 1;
1953	adapter->rx_queue.shared = &adapter->rqd_start->ctrl;
1954	err = vmxnet3_rq_create(&adapter->rx_queue, adapter);
1955	if (err)
1956		vmxnet3_tq_destroy(&adapter->tx_queue, adapter);
1957
1958	return err;
1959}
1960
1961static int
1962vmxnet3_open(struct net_device *netdev)
1963{
1964	struct vmxnet3_adapter *adapter;
1965	int err;
1966
1967	adapter = netdev_priv(netdev);
1968
1969	spin_lock_init(&adapter->tx_queue.tx_lock);
1970
1971	err = vmxnet3_create_queues(adapter, VMXNET3_DEF_TX_RING_SIZE,
1972				    VMXNET3_DEF_RX_RING_SIZE,
1973				    VMXNET3_DEF_RX_RING_SIZE);
1974	if (err)
1975		goto queue_err;
1976
1977	err = vmxnet3_activate_dev(adapter);
1978	if (err)
1979		goto activate_err;
1980
1981	return 0;
1982
1983activate_err:
1984	vmxnet3_rq_destroy(&adapter->rx_queue, adapter);
1985	vmxnet3_tq_destroy(&adapter->tx_queue, adapter);
1986queue_err:
1987	return err;
1988}
1989
1990
1991static int
1992vmxnet3_close(struct net_device *netdev)
1993{
1994	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1995
1996	/*
1997	 * Reset_work may be in the middle of resetting the device, wait for its
1998	 * completion.
1999	 */
2000	while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2001		msleep(1);
2002
2003	vmxnet3_quiesce_dev(adapter);
2004
2005	vmxnet3_rq_destroy(&adapter->rx_queue, adapter);
2006	vmxnet3_tq_destroy(&adapter->tx_queue, adapter);
2007
2008	clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2009
2010
2011	return 0;
2012}
2013
2014
2015void
2016vmxnet3_force_close(struct vmxnet3_adapter *adapter)
2017{
2018	/*
2019	 * we must clear VMXNET3_STATE_BIT_RESETTING, otherwise
2020	 * vmxnet3_close() will deadlock.
2021	 */
2022	BUG_ON(test_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state));
2023
2024	/* we need to enable NAPI, otherwise dev_close will deadlock */
2025	napi_enable(&adapter->napi);
2026	dev_close(adapter->netdev);
2027}
2028
2029
2030static int
2031vmxnet3_change_mtu(struct net_device *netdev, int new_mtu)
2032{
2033	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2034	int err = 0;
2035
2036	if (new_mtu < VMXNET3_MIN_MTU || new_mtu > VMXNET3_MAX_MTU)
2037		return -EINVAL;
2038
2039	if (new_mtu > 1500 && !adapter->jumbo_frame)
2040		return -EINVAL;
2041
2042	netdev->mtu = new_mtu;
2043
2044	/*
2045	 * Reset_work may be in the middle of resetting the device, wait for its
2046	 * completion.
2047	 */
2048	while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2049		msleep(1);
2050
2051	if (netif_running(netdev)) {
2052		vmxnet3_quiesce_dev(adapter);
2053		vmxnet3_reset_dev(adapter);
2054
2055		/* we need to re-create the rx queue based on the new mtu */
2056		vmxnet3_rq_destroy(&adapter->rx_queue, adapter);
2057		vmxnet3_adjust_rx_ring_size(adapter);
2058		adapter->rx_queue.comp_ring.size  =
2059					adapter->rx_queue.rx_ring[0].size +
2060					adapter->rx_queue.rx_ring[1].size;
2061		err = vmxnet3_rq_create(&adapter->rx_queue, adapter);
2062		if (err) {
2063			printk(KERN_ERR "%s: failed to re-create rx queue,"
2064				" error %d. Closing it.\n", netdev->name, err);
2065			goto out;
2066		}
2067
2068		err = vmxnet3_activate_dev(adapter);
2069		if (err) {
2070			printk(KERN_ERR "%s: failed to re-activate, error %d. "
2071				"Closing it\n", netdev->name, err);
2072			goto out;
2073		}
2074	}
2075
2076out:
2077	clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2078	if (err)
2079		vmxnet3_force_close(adapter);
2080
2081	return err;
2082}
2083
2084
2085static void
2086vmxnet3_declare_features(struct vmxnet3_adapter *adapter, bool dma64)
2087{
2088	struct net_device *netdev = adapter->netdev;
2089
2090	netdev->features = NETIF_F_SG |
2091		NETIF_F_HW_CSUM |
2092		NETIF_F_HW_VLAN_TX |
2093		NETIF_F_HW_VLAN_RX |
2094		NETIF_F_HW_VLAN_FILTER |
2095		NETIF_F_TSO |
2096		NETIF_F_TSO6 |
2097		NETIF_F_LRO;
2098
2099	printk(KERN_INFO "features: sg csum vlan jf tso tsoIPv6 lro");
2100
2101	adapter->rxcsum = true;
2102	adapter->jumbo_frame = true;
2103	adapter->lro = true;
2104
2105	if (dma64) {
2106		netdev->features |= NETIF_F_HIGHDMA;
2107		printk(" highDMA");
2108	}
2109
2110	netdev->vlan_features = netdev->features;
2111	printk("\n");
2112}
2113
2114
2115static void
2116vmxnet3_read_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac)
2117{
2118	u32 tmp;
2119
2120	tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACL);
2121	*(u32 *)mac = tmp;
2122
2123	tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACH);
2124	mac[4] = tmp & 0xff;
2125	mac[5] = (tmp >> 8) & 0xff;
2126}
2127
2128
2129static void
2130vmxnet3_alloc_intr_resources(struct vmxnet3_adapter *adapter)
2131{
2132	u32 cfg;
2133
2134	/* intr settings */
2135	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2136			       VMXNET3_CMD_GET_CONF_INTR);
2137	cfg = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
2138	adapter->intr.type = cfg & 0x3;
2139	adapter->intr.mask_mode = (cfg >> 2) & 0x3;
2140
2141	if (adapter->intr.type == VMXNET3_IT_AUTO) {
2142		int err;
2143
2144#ifdef CONFIG_PCI_MSI
2145		adapter->intr.msix_entries[0].entry = 0;
2146		err = pci_enable_msix(adapter->pdev, adapter->intr.msix_entries,
2147				      VMXNET3_LINUX_MAX_MSIX_VECT);
2148		if (!err) {
2149			adapter->intr.num_intrs = 1;
2150			adapter->intr.type = VMXNET3_IT_MSIX;
2151			return;
2152		}
2153#endif
2154
2155		err = pci_enable_msi(adapter->pdev);
2156		if (!err) {
2157			adapter->intr.num_intrs = 1;
2158			adapter->intr.type = VMXNET3_IT_MSI;
2159			return;
2160		}
2161	}
2162
2163	adapter->intr.type = VMXNET3_IT_INTX;
2164
2165	/* INT-X related setting */
2166	adapter->intr.num_intrs = 1;
2167}
2168
2169
2170static void
2171vmxnet3_free_intr_resources(struct vmxnet3_adapter *adapter)
2172{
2173	if (adapter->intr.type == VMXNET3_IT_MSIX)
2174		pci_disable_msix(adapter->pdev);
2175	else if (adapter->intr.type == VMXNET3_IT_MSI)
2176		pci_disable_msi(adapter->pdev);
2177	else
2178		BUG_ON(adapter->intr.type != VMXNET3_IT_INTX);
2179}
2180
2181
2182static void
2183vmxnet3_tx_timeout(struct net_device *netdev)
2184{
2185	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2186	adapter->tx_timeout_count++;
2187
2188	printk(KERN_ERR "%s: tx hang\n", adapter->netdev->name);
2189	schedule_work(&adapter->work);
2190}
2191
2192
2193static void
2194vmxnet3_reset_work(struct work_struct *data)
2195{
2196	struct vmxnet3_adapter *adapter;
2197
2198	adapter = container_of(data, struct vmxnet3_adapter, work);
2199
2200	/* if another thread is resetting the device, no need to proceed */
2201	if (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2202		return;
2203
2204	/* if the device is closed, we must leave it alone */
2205	if (netif_running(adapter->netdev)) {
2206		printk(KERN_INFO "%s: resetting\n", adapter->netdev->name);
2207		vmxnet3_quiesce_dev(adapter);
2208		vmxnet3_reset_dev(adapter);
2209		vmxnet3_activate_dev(adapter);
2210	} else {
2211		printk(KERN_INFO "%s: already closed\n", adapter->netdev->name);
2212	}
2213
2214	clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2215}
2216
2217
2218static int __devinit
2219vmxnet3_probe_device(struct pci_dev *pdev,
2220		     const struct pci_device_id *id)
2221{
2222	static const struct net_device_ops vmxnet3_netdev_ops = {
2223		.ndo_open = vmxnet3_open,
2224		.ndo_stop = vmxnet3_close,
2225		.ndo_start_xmit = vmxnet3_xmit_frame,
2226		.ndo_set_mac_address = vmxnet3_set_mac_addr,
2227		.ndo_change_mtu = vmxnet3_change_mtu,
2228		.ndo_get_stats = vmxnet3_get_stats,
2229		.ndo_tx_timeout = vmxnet3_tx_timeout,
2230		.ndo_set_multicast_list = vmxnet3_set_mc,
2231		.ndo_vlan_rx_register = vmxnet3_vlan_rx_register,
2232		.ndo_vlan_rx_add_vid = vmxnet3_vlan_rx_add_vid,
2233		.ndo_vlan_rx_kill_vid = vmxnet3_vlan_rx_kill_vid,
2234#ifdef CONFIG_NET_POLL_CONTROLLER
2235		.ndo_poll_controller = vmxnet3_netpoll,
2236#endif
2237	};
2238	int err;
2239	bool dma64 = false; /* stupid gcc */
2240	u32 ver;
2241	struct net_device *netdev;
2242	struct vmxnet3_adapter *adapter;
2243	u8 mac[ETH_ALEN];
2244
2245	netdev = alloc_etherdev(sizeof(struct vmxnet3_adapter));
2246	if (!netdev) {
2247		printk(KERN_ERR "Failed to alloc ethernet device for adapter "
2248			"%s\n",	pci_name(pdev));
2249		return -ENOMEM;
2250	}
2251
2252	pci_set_drvdata(pdev, netdev);
2253	adapter = netdev_priv(netdev);
2254	adapter->netdev = netdev;
2255	adapter->pdev = pdev;
2256
2257	adapter->shared = pci_alloc_consistent(adapter->pdev,
2258			  sizeof(struct Vmxnet3_DriverShared),
2259			  &adapter->shared_pa);
2260	if (!adapter->shared) {
2261		printk(KERN_ERR "Failed to allocate memory for %s\n",
2262			pci_name(pdev));
2263		err = -ENOMEM;
2264		goto err_alloc_shared;
2265	}
2266
2267	adapter->tqd_start = pci_alloc_consistent(adapter->pdev,
2268			     sizeof(struct Vmxnet3_TxQueueDesc) +
2269			     sizeof(struct Vmxnet3_RxQueueDesc),
2270			     &adapter->queue_desc_pa);
2271
2272	if (!adapter->tqd_start) {
2273		printk(KERN_ERR "Failed to allocate memory for %s\n",
2274			pci_name(pdev));
2275		err = -ENOMEM;
2276		goto err_alloc_queue_desc;
2277	}
2278	adapter->rqd_start = (struct Vmxnet3_RxQueueDesc *)(adapter->tqd_start
2279							    + 1);
2280
2281	adapter->pm_conf = kmalloc(sizeof(struct Vmxnet3_PMConf), GFP_KERNEL);
2282	if (adapter->pm_conf == NULL) {
2283		printk(KERN_ERR "Failed to allocate memory for %s\n",
2284			pci_name(pdev));
2285		err = -ENOMEM;
2286		goto err_alloc_pm;
2287	}
2288
2289	err = vmxnet3_alloc_pci_resources(adapter, &dma64);
2290	if (err < 0)
2291		goto err_alloc_pci;
2292
2293	ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_VRRS);
2294	if (ver & 1) {
2295		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_VRRS, 1);
2296	} else {
2297		printk(KERN_ERR "Incompatible h/w version (0x%x) for adapter"
2298		       " %s\n",	ver, pci_name(pdev));
2299		err = -EBUSY;
2300		goto err_ver;
2301	}
2302
2303	ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_UVRS);
2304	if (ver & 1) {
2305		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_UVRS, 1);
2306	} else {
2307		printk(KERN_ERR "Incompatible upt version (0x%x) for "
2308		       "adapter %s\n", ver, pci_name(pdev));
2309		err = -EBUSY;
2310		goto err_ver;
2311	}
2312
2313	vmxnet3_declare_features(adapter, dma64);
2314
2315	adapter->dev_number = atomic_read(&devices_found);
2316	vmxnet3_alloc_intr_resources(adapter);
2317
2318	vmxnet3_read_mac_addr(adapter, mac);
2319	memcpy(netdev->dev_addr,  mac, netdev->addr_len);
2320
2321	netdev->netdev_ops = &vmxnet3_netdev_ops;
2322	netdev->watchdog_timeo = 5 * HZ;
2323	vmxnet3_set_ethtool_ops(netdev);
2324
2325	INIT_WORK(&adapter->work, vmxnet3_reset_work);
2326
2327	netif_napi_add(netdev, &adapter->napi, vmxnet3_poll, 64);
2328	SET_NETDEV_DEV(netdev, &pdev->dev);
2329	err = register_netdev(netdev);
2330
2331	if (err) {
2332		printk(KERN_ERR "Failed to register adapter %s\n",
2333			pci_name(pdev));
2334		goto err_register;
2335	}
2336
2337	set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
2338	atomic_inc(&devices_found);
2339	return 0;
2340
2341err_register:
2342	vmxnet3_free_intr_resources(adapter);
2343err_ver:
2344	vmxnet3_free_pci_resources(adapter);
2345err_alloc_pci:
2346	kfree(adapter->pm_conf);
2347err_alloc_pm:
2348	pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_TxQueueDesc) +
2349			    sizeof(struct Vmxnet3_RxQueueDesc),
2350			    adapter->tqd_start, adapter->queue_desc_pa);
2351err_alloc_queue_desc:
2352	pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_DriverShared),
2353			    adapter->shared, adapter->shared_pa);
2354err_alloc_shared:
2355	pci_set_drvdata(pdev, NULL);
2356	free_netdev(netdev);
2357	return err;
2358}
2359
2360
2361static void __devexit
2362vmxnet3_remove_device(struct pci_dev *pdev)
2363{
2364	struct net_device *netdev = pci_get_drvdata(pdev);
2365	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2366
2367	flush_scheduled_work();
2368
2369	unregister_netdev(netdev);
2370
2371	vmxnet3_free_intr_resources(adapter);
2372	vmxnet3_free_pci_resources(adapter);
2373	kfree(adapter->pm_conf);
2374	pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_TxQueueDesc) +
2375			    sizeof(struct Vmxnet3_RxQueueDesc),
2376			    adapter->tqd_start, adapter->queue_desc_pa);
2377	pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_DriverShared),
2378			    adapter->shared, adapter->shared_pa);
2379	free_netdev(netdev);
2380}
2381
2382
2383#ifdef CONFIG_PM
2384
2385static int
2386vmxnet3_suspend(struct device *device)
2387{
2388	struct pci_dev *pdev = to_pci_dev(device);
2389	struct net_device *netdev = pci_get_drvdata(pdev);
2390	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2391	struct Vmxnet3_PMConf *pmConf;
2392	struct ethhdr *ehdr;
2393	struct arphdr *ahdr;
2394	u8 *arpreq;
2395	struct in_device *in_dev;
2396	struct in_ifaddr *ifa;
2397	int i = 0;
2398
2399	if (!netif_running(netdev))
2400		return 0;
2401
2402	vmxnet3_disable_all_intrs(adapter);
2403	vmxnet3_free_irqs(adapter);
2404	vmxnet3_free_intr_resources(adapter);
2405
2406	netif_device_detach(netdev);
2407	netif_stop_queue(netdev);
2408
2409	/* Create wake-up filters. */
2410	pmConf = adapter->pm_conf;
2411	memset(pmConf, 0, sizeof(*pmConf));
2412
2413	if (adapter->wol & WAKE_UCAST) {
2414		pmConf->filters[i].patternSize = ETH_ALEN;
2415		pmConf->filters[i].maskSize = 1;
2416		memcpy(pmConf->filters[i].pattern, netdev->dev_addr, ETH_ALEN);
2417		pmConf->filters[i].mask[0] = 0x3F; /* LSB ETH_ALEN bits */
2418
2419		pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_FILTER;
2420		i++;
2421	}
2422
2423	if (adapter->wol & WAKE_ARP) {
2424		in_dev = in_dev_get(netdev);
2425		if (!in_dev)
2426			goto skip_arp;
2427
2428		ifa = (struct in_ifaddr *)in_dev->ifa_list;
2429		if (!ifa)
2430			goto skip_arp;
2431
2432		pmConf->filters[i].patternSize = ETH_HLEN + /* Ethernet header*/
2433			sizeof(struct arphdr) +		/* ARP header */
2434			2 * ETH_ALEN +		/* 2 Ethernet addresses*/
2435			2 * sizeof(u32);	/*2 IPv4 addresses */
2436		pmConf->filters[i].maskSize =
2437			(pmConf->filters[i].patternSize - 1) / 8 + 1;
2438
2439		/* ETH_P_ARP in Ethernet header. */
2440		ehdr = (struct ethhdr *)pmConf->filters[i].pattern;
2441		ehdr->h_proto = htons(ETH_P_ARP);
2442
2443		/* ARPOP_REQUEST in ARP header. */
2444		ahdr = (struct arphdr *)&pmConf->filters[i].pattern[ETH_HLEN];
2445		ahdr->ar_op = htons(ARPOP_REQUEST);
2446		arpreq = (u8 *)(ahdr + 1);
2447
2448		/* The Unicast IPv4 address in 'tip' field. */
2449		arpreq += 2 * ETH_ALEN + sizeof(u32);
2450		*(u32 *)arpreq = ifa->ifa_address;
2451
2452		/* The mask for the relevant bits. */
2453		pmConf->filters[i].mask[0] = 0x00;
2454		pmConf->filters[i].mask[1] = 0x30; /* ETH_P_ARP */
2455		pmConf->filters[i].mask[2] = 0x30; /* ARPOP_REQUEST */
2456		pmConf->filters[i].mask[3] = 0x00;
2457		pmConf->filters[i].mask[4] = 0xC0; /* IPv4 TIP */
2458		pmConf->filters[i].mask[5] = 0x03; /* IPv4 TIP */
2459		in_dev_put(in_dev);
2460
2461		pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_FILTER;
2462		i++;
2463	}
2464
2465skip_arp:
2466	if (adapter->wol & WAKE_MAGIC)
2467		pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_MAGIC;
2468
2469	pmConf->numFilters = i;
2470
2471	adapter->shared->devRead.pmConfDesc.confVer = 1;
2472	adapter->shared->devRead.pmConfDesc.confLen = sizeof(*pmConf);
2473	adapter->shared->devRead.pmConfDesc.confPA = virt_to_phys(pmConf);
2474
2475	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2476			       VMXNET3_CMD_UPDATE_PMCFG);
2477
2478	pci_save_state(pdev);
2479	pci_enable_wake(pdev, pci_choose_state(pdev, PMSG_SUSPEND),
2480			adapter->wol);
2481	pci_disable_device(pdev);
2482	pci_set_power_state(pdev, pci_choose_state(pdev, PMSG_SUSPEND));
2483
2484	return 0;
2485}
2486
2487
2488static int
2489vmxnet3_resume(struct device *device)
2490{
2491	int err;
2492	struct pci_dev *pdev = to_pci_dev(device);
2493	struct net_device *netdev = pci_get_drvdata(pdev);
2494	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2495	struct Vmxnet3_PMConf *pmConf;
2496
2497	if (!netif_running(netdev))
2498		return 0;
2499
2500	/* Destroy wake-up filters. */
2501	pmConf = adapter->pm_conf;
2502	memset(pmConf, 0, sizeof(*pmConf));
2503
2504	adapter->shared->devRead.pmConfDesc.confVer = 1;
2505	adapter->shared->devRead.pmConfDesc.confLen = sizeof(*pmConf);
2506	adapter->shared->devRead.pmConfDesc.confPA = virt_to_phys(pmConf);
2507
2508	netif_device_attach(netdev);
2509	pci_set_power_state(pdev, PCI_D0);
2510	pci_restore_state(pdev);
2511	err = pci_enable_device_mem(pdev);
2512	if (err != 0)
2513		return err;
2514
2515	pci_enable_wake(pdev, PCI_D0, 0);
2516
2517	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2518			       VMXNET3_CMD_UPDATE_PMCFG);
2519	vmxnet3_alloc_intr_resources(adapter);
2520	vmxnet3_request_irqs(adapter);
2521	vmxnet3_enable_all_intrs(adapter);
2522
2523	return 0;
2524}
2525
2526static struct dev_pm_ops vmxnet3_pm_ops = {
2527	.suspend = vmxnet3_suspend,
2528	.resume = vmxnet3_resume,
2529};
2530#endif
2531
2532static struct pci_driver vmxnet3_driver = {
2533	.name		= vmxnet3_driver_name,
2534	.id_table	= vmxnet3_pciid_table,
2535	.probe		= vmxnet3_probe_device,
2536	.remove		= __devexit_p(vmxnet3_remove_device),
2537#ifdef CONFIG_PM
2538	.driver.pm	= &vmxnet3_pm_ops,
2539#endif
2540};
2541
2542
2543static int __init
2544vmxnet3_init_module(void)
2545{
2546	printk(KERN_INFO "%s - version %s\n", VMXNET3_DRIVER_DESC,
2547		VMXNET3_DRIVER_VERSION_REPORT);
2548	return pci_register_driver(&vmxnet3_driver);
2549}
2550
2551module_init(vmxnet3_init_module);
2552
2553
2554static void
2555vmxnet3_exit_module(void)
2556{
2557	pci_unregister_driver(&vmxnet3_driver);
2558}
2559
2560module_exit(vmxnet3_exit_module);
2561
2562MODULE_AUTHOR("VMware, Inc.");
2563MODULE_DESCRIPTION(VMXNET3_DRIVER_DESC);
2564MODULE_LICENSE("GPL v2");
2565MODULE_VERSION(VMXNET3_DRIVER_VERSION_STRING);
2566