1/*
2 * Linux driver for VMware's vmxnet3 ethernet NIC.
3 *
4 * Copyright (C) 2008-2020, VMware, Inc. All Rights Reserved.
5 *
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License as published by the
8 * Free Software Foundation; version 2 of the License and no later version.
9 *
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
13 * NON INFRINGEMENT. See the GNU General Public License for more
14 * details.
15 *
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * The full GNU General Public License is included in this distribution in
21 * the file called "COPYING".
22 *
23 * Maintained by: pv-drivers@vmware.com
24 *
25 */
26
27#include <linux/module.h>
28#include <net/ip6_checksum.h>
29
30#include "vmxnet3_int.h"
31
32char vmxnet3_driver_name[] = "vmxnet3";
33#define VMXNET3_DRIVER_DESC "VMware vmxnet3 virtual NIC driver"
34
35/*
36 * PCI Device ID Table
37 * Last entry must be all 0s
38 */
39static const struct pci_device_id vmxnet3_pciid_table[] = {
40	{PCI_VDEVICE(VMWARE, PCI_DEVICE_ID_VMWARE_VMXNET3)},
41	{0}
42};
43
44MODULE_DEVICE_TABLE(pci, vmxnet3_pciid_table);
45
46static int enable_mq = 1;
47
48static void
49vmxnet3_write_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac);
50
51/*
52 *    Enable/Disable the given intr
53 */
54static void
55vmxnet3_enable_intr(struct vmxnet3_adapter *adapter, unsigned intr_idx)
56{
57	VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_IMR + intr_idx * 8, 0);
58}
59
60
61static void
62vmxnet3_disable_intr(struct vmxnet3_adapter *adapter, unsigned intr_idx)
63{
64	VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_IMR + intr_idx * 8, 1);
65}
66
67
68/*
69 *    Enable/Disable all intrs used by the device
70 */
71static void
72vmxnet3_enable_all_intrs(struct vmxnet3_adapter *adapter)
73{
74	int i;
75
76	for (i = 0; i < adapter->intr.num_intrs; i++)
77		vmxnet3_enable_intr(adapter, i);
78	adapter->shared->devRead.intrConf.intrCtrl &=
79					cpu_to_le32(~VMXNET3_IC_DISABLE_ALL);
80}
81
82
83static void
84vmxnet3_disable_all_intrs(struct vmxnet3_adapter *adapter)
85{
86	int i;
87
88	adapter->shared->devRead.intrConf.intrCtrl |=
89					cpu_to_le32(VMXNET3_IC_DISABLE_ALL);
90	for (i = 0; i < adapter->intr.num_intrs; i++)
91		vmxnet3_disable_intr(adapter, i);
92}
93
94
95static void
96vmxnet3_ack_events(struct vmxnet3_adapter *adapter, u32 events)
97{
98	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_ECR, events);
99}
100
101
102static bool
103vmxnet3_tq_stopped(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
104{
105	return tq->stopped;
106}
107
108
109static void
110vmxnet3_tq_start(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
111{
112	tq->stopped = false;
113	netif_start_subqueue(adapter->netdev, tq - adapter->tx_queue);
114}
115
116
117static void
118vmxnet3_tq_wake(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
119{
120	tq->stopped = false;
121	netif_wake_subqueue(adapter->netdev, (tq - adapter->tx_queue));
122}
123
124
125static void
126vmxnet3_tq_stop(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
127{
128	tq->stopped = true;
129	tq->num_stop++;
130	netif_stop_subqueue(adapter->netdev, (tq - adapter->tx_queue));
131}
132
133
134/*
135 * Check the link state. This may start or stop the tx queue.
136 */
137static void
138vmxnet3_check_link(struct vmxnet3_adapter *adapter, bool affectTxQueue)
139{
140	u32 ret;
141	int i;
142	unsigned long flags;
143
144	spin_lock_irqsave(&adapter->cmd_lock, flags);
145	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, VMXNET3_CMD_GET_LINK);
146	ret = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
147	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
148
149	adapter->link_speed = ret >> 16;
150	if (ret & 1) { /* Link is up. */
151		netdev_info(adapter->netdev, "NIC Link is Up %d Mbps\n",
152			    adapter->link_speed);
153		netif_carrier_on(adapter->netdev);
154
155		if (affectTxQueue) {
156			for (i = 0; i < adapter->num_tx_queues; i++)
157				vmxnet3_tq_start(&adapter->tx_queue[i],
158						 adapter);
159		}
160	} else {
161		netdev_info(adapter->netdev, "NIC Link is Down\n");
162		netif_carrier_off(adapter->netdev);
163
164		if (affectTxQueue) {
165			for (i = 0; i < adapter->num_tx_queues; i++)
166				vmxnet3_tq_stop(&adapter->tx_queue[i], adapter);
167		}
168	}
169}
170
171static void
172vmxnet3_process_events(struct vmxnet3_adapter *adapter)
173{
174	int i;
175	unsigned long flags;
176	u32 events = le32_to_cpu(adapter->shared->ecr);
177	if (!events)
178		return;
179
180	vmxnet3_ack_events(adapter, events);
181
182	/* Check if link state has changed */
183	if (events & VMXNET3_ECR_LINK)
184		vmxnet3_check_link(adapter, true);
185
186	/* Check if there is an error on xmit/recv queues */
187	if (events & (VMXNET3_ECR_TQERR | VMXNET3_ECR_RQERR)) {
188		spin_lock_irqsave(&adapter->cmd_lock, flags);
189		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
190				       VMXNET3_CMD_GET_QUEUE_STATUS);
191		spin_unlock_irqrestore(&adapter->cmd_lock, flags);
192
193		for (i = 0; i < adapter->num_tx_queues; i++)
194			if (adapter->tqd_start[i].status.stopped)
195				dev_err(&adapter->netdev->dev,
196					"%s: tq[%d] error 0x%x\n",
197					adapter->netdev->name, i, le32_to_cpu(
198					adapter->tqd_start[i].status.error));
199		for (i = 0; i < adapter->num_rx_queues; i++)
200			if (adapter->rqd_start[i].status.stopped)
201				dev_err(&adapter->netdev->dev,
202					"%s: rq[%d] error 0x%x\n",
203					adapter->netdev->name, i,
204					adapter->rqd_start[i].status.error);
205
206		schedule_work(&adapter->work);
207	}
208}
209
210#ifdef __BIG_ENDIAN_BITFIELD
211/*
212 * The device expects the bitfields in shared structures to be written in
213 * little endian. When CPU is big endian, the following routines are used to
214 * correctly read and write into ABI.
215 * The general technique used here is : double word bitfields are defined in
216 * opposite order for big endian architecture. Then before reading them in
217 * driver the complete double word is translated using le32_to_cpu. Similarly
218 * After the driver writes into bitfields, cpu_to_le32 is used to translate the
219 * double words into required format.
220 * In order to avoid touching bits in shared structure more than once, temporary
221 * descriptors are used. These are passed as srcDesc to following functions.
222 */
223static void vmxnet3_RxDescToCPU(const struct Vmxnet3_RxDesc *srcDesc,
224				struct Vmxnet3_RxDesc *dstDesc)
225{
226	u32 *src = (u32 *)srcDesc + 2;
227	u32 *dst = (u32 *)dstDesc + 2;
228	dstDesc->addr = le64_to_cpu(srcDesc->addr);
229	*dst = le32_to_cpu(*src);
230	dstDesc->ext1 = le32_to_cpu(srcDesc->ext1);
231}
232
233static void vmxnet3_TxDescToLe(const struct Vmxnet3_TxDesc *srcDesc,
234			       struct Vmxnet3_TxDesc *dstDesc)
235{
236	int i;
237	u32 *src = (u32 *)(srcDesc + 1);
238	u32 *dst = (u32 *)(dstDesc + 1);
239
240	/* Working backwards so that the gen bit is set at the end. */
241	for (i = 2; i > 0; i--) {
242		src--;
243		dst--;
244		*dst = cpu_to_le32(*src);
245	}
246}
247
248
249static void vmxnet3_RxCompToCPU(const struct Vmxnet3_RxCompDesc *srcDesc,
250				struct Vmxnet3_RxCompDesc *dstDesc)
251{
252	int i = 0;
253	u32 *src = (u32 *)srcDesc;
254	u32 *dst = (u32 *)dstDesc;
255	for (i = 0; i < sizeof(struct Vmxnet3_RxCompDesc) / sizeof(u32); i++) {
256		*dst = le32_to_cpu(*src);
257		src++;
258		dst++;
259	}
260}
261
262
263/* Used to read bitfield values from double words. */
264static u32 get_bitfield32(const __le32 *bitfield, u32 pos, u32 size)
265{
266	u32 temp = le32_to_cpu(*bitfield);
267	u32 mask = ((1 << size) - 1) << pos;
268	temp &= mask;
269	temp >>= pos;
270	return temp;
271}
272
273
274
275#endif  /* __BIG_ENDIAN_BITFIELD */
276
277#ifdef __BIG_ENDIAN_BITFIELD
278
279#   define VMXNET3_TXDESC_GET_GEN(txdesc) get_bitfield32(((const __le32 *) \
280			txdesc) + VMXNET3_TXD_GEN_DWORD_SHIFT, \
281			VMXNET3_TXD_GEN_SHIFT, VMXNET3_TXD_GEN_SIZE)
282#   define VMXNET3_TXDESC_GET_EOP(txdesc) get_bitfield32(((const __le32 *) \
283			txdesc) + VMXNET3_TXD_EOP_DWORD_SHIFT, \
284			VMXNET3_TXD_EOP_SHIFT, VMXNET3_TXD_EOP_SIZE)
285#   define VMXNET3_TCD_GET_GEN(tcd) get_bitfield32(((const __le32 *)tcd) + \
286			VMXNET3_TCD_GEN_DWORD_SHIFT, VMXNET3_TCD_GEN_SHIFT, \
287			VMXNET3_TCD_GEN_SIZE)
288#   define VMXNET3_TCD_GET_TXIDX(tcd) get_bitfield32((const __le32 *)tcd, \
289			VMXNET3_TCD_TXIDX_SHIFT, VMXNET3_TCD_TXIDX_SIZE)
290#   define vmxnet3_getRxComp(dstrcd, rcd, tmp) do { \
291			(dstrcd) = (tmp); \
292			vmxnet3_RxCompToCPU((rcd), (tmp)); \
293		} while (0)
294#   define vmxnet3_getRxDesc(dstrxd, rxd, tmp) do { \
295			(dstrxd) = (tmp); \
296			vmxnet3_RxDescToCPU((rxd), (tmp)); \
297		} while (0)
298
299#else
300
301#   define VMXNET3_TXDESC_GET_GEN(txdesc) ((txdesc)->gen)
302#   define VMXNET3_TXDESC_GET_EOP(txdesc) ((txdesc)->eop)
303#   define VMXNET3_TCD_GET_GEN(tcd) ((tcd)->gen)
304#   define VMXNET3_TCD_GET_TXIDX(tcd) ((tcd)->txdIdx)
305#   define vmxnet3_getRxComp(dstrcd, rcd, tmp) (dstrcd) = (rcd)
306#   define vmxnet3_getRxDesc(dstrxd, rxd, tmp) (dstrxd) = (rxd)
307
308#endif /* __BIG_ENDIAN_BITFIELD  */
309
310
311static void
312vmxnet3_unmap_tx_buf(struct vmxnet3_tx_buf_info *tbi,
313		     struct pci_dev *pdev)
314{
315	if (tbi->map_type == VMXNET3_MAP_SINGLE)
316		dma_unmap_single(&pdev->dev, tbi->dma_addr, tbi->len,
317				 PCI_DMA_TODEVICE);
318	else if (tbi->map_type == VMXNET3_MAP_PAGE)
319		dma_unmap_page(&pdev->dev, tbi->dma_addr, tbi->len,
320			       PCI_DMA_TODEVICE);
321	else
322		BUG_ON(tbi->map_type != VMXNET3_MAP_NONE);
323
324	tbi->map_type = VMXNET3_MAP_NONE; /* to help debugging */
325}
326
327
328static int
329vmxnet3_unmap_pkt(u32 eop_idx, struct vmxnet3_tx_queue *tq,
330		  struct pci_dev *pdev,	struct vmxnet3_adapter *adapter)
331{
332	struct sk_buff *skb;
333	int entries = 0;
334
335	/* no out of order completion */
336	BUG_ON(tq->buf_info[eop_idx].sop_idx != tq->tx_ring.next2comp);
337	BUG_ON(VMXNET3_TXDESC_GET_EOP(&(tq->tx_ring.base[eop_idx].txd)) != 1);
338
339	skb = tq->buf_info[eop_idx].skb;
340	BUG_ON(skb == NULL);
341	tq->buf_info[eop_idx].skb = NULL;
342
343	VMXNET3_INC_RING_IDX_ONLY(eop_idx, tq->tx_ring.size);
344
345	while (tq->tx_ring.next2comp != eop_idx) {
346		vmxnet3_unmap_tx_buf(tq->buf_info + tq->tx_ring.next2comp,
347				     pdev);
348
349		/* update next2comp w/o tx_lock. Since we are marking more,
350		 * instead of less, tx ring entries avail, the worst case is
351		 * that the tx routine incorrectly re-queues a pkt due to
352		 * insufficient tx ring entries.
353		 */
354		vmxnet3_cmd_ring_adv_next2comp(&tq->tx_ring);
355		entries++;
356	}
357
358	dev_kfree_skb_any(skb);
359	return entries;
360}
361
362
363static int
364vmxnet3_tq_tx_complete(struct vmxnet3_tx_queue *tq,
365			struct vmxnet3_adapter *adapter)
366{
367	int completed = 0;
368	union Vmxnet3_GenericDesc *gdesc;
369
370	gdesc = tq->comp_ring.base + tq->comp_ring.next2proc;
371	while (VMXNET3_TCD_GET_GEN(&gdesc->tcd) == tq->comp_ring.gen) {
372		/* Prevent any &gdesc->tcd field from being (speculatively)
373		 * read before (&gdesc->tcd)->gen is read.
374		 */
375		dma_rmb();
376
377		completed += vmxnet3_unmap_pkt(VMXNET3_TCD_GET_TXIDX(
378					       &gdesc->tcd), tq, adapter->pdev,
379					       adapter);
380
381		vmxnet3_comp_ring_adv_next2proc(&tq->comp_ring);
382		gdesc = tq->comp_ring.base + tq->comp_ring.next2proc;
383	}
384
385	if (completed) {
386		spin_lock(&tq->tx_lock);
387		if (unlikely(vmxnet3_tq_stopped(tq, adapter) &&
388			     vmxnet3_cmd_ring_desc_avail(&tq->tx_ring) >
389			     VMXNET3_WAKE_QUEUE_THRESHOLD(tq) &&
390			     netif_carrier_ok(adapter->netdev))) {
391			vmxnet3_tq_wake(tq, adapter);
392		}
393		spin_unlock(&tq->tx_lock);
394	}
395	return completed;
396}
397
398
399static void
400vmxnet3_tq_cleanup(struct vmxnet3_tx_queue *tq,
401		   struct vmxnet3_adapter *adapter)
402{
403	int i;
404
405	while (tq->tx_ring.next2comp != tq->tx_ring.next2fill) {
406		struct vmxnet3_tx_buf_info *tbi;
407
408		tbi = tq->buf_info + tq->tx_ring.next2comp;
409
410		vmxnet3_unmap_tx_buf(tbi, adapter->pdev);
411		if (tbi->skb) {
412			dev_kfree_skb_any(tbi->skb);
413			tbi->skb = NULL;
414		}
415		vmxnet3_cmd_ring_adv_next2comp(&tq->tx_ring);
416	}
417
418	/* sanity check, verify all buffers are indeed unmapped and freed */
419	for (i = 0; i < tq->tx_ring.size; i++) {
420		BUG_ON(tq->buf_info[i].skb != NULL ||
421		       tq->buf_info[i].map_type != VMXNET3_MAP_NONE);
422	}
423
424	tq->tx_ring.gen = VMXNET3_INIT_GEN;
425	tq->tx_ring.next2fill = tq->tx_ring.next2comp = 0;
426
427	tq->comp_ring.gen = VMXNET3_INIT_GEN;
428	tq->comp_ring.next2proc = 0;
429}
430
431
432static void
433vmxnet3_tq_destroy(struct vmxnet3_tx_queue *tq,
434		   struct vmxnet3_adapter *adapter)
435{
436	if (tq->tx_ring.base) {
437		dma_free_coherent(&adapter->pdev->dev, tq->tx_ring.size *
438				  sizeof(struct Vmxnet3_TxDesc),
439				  tq->tx_ring.base, tq->tx_ring.basePA);
440		tq->tx_ring.base = NULL;
441	}
442	if (tq->data_ring.base) {
443		dma_free_coherent(&adapter->pdev->dev,
444				  tq->data_ring.size * tq->txdata_desc_size,
445				  tq->data_ring.base, tq->data_ring.basePA);
446		tq->data_ring.base = NULL;
447	}
448	if (tq->comp_ring.base) {
449		dma_free_coherent(&adapter->pdev->dev, tq->comp_ring.size *
450				  sizeof(struct Vmxnet3_TxCompDesc),
451				  tq->comp_ring.base, tq->comp_ring.basePA);
452		tq->comp_ring.base = NULL;
453	}
454	if (tq->buf_info) {
455		dma_free_coherent(&adapter->pdev->dev,
456				  tq->tx_ring.size * sizeof(tq->buf_info[0]),
457				  tq->buf_info, tq->buf_info_pa);
458		tq->buf_info = NULL;
459	}
460}
461
462
463/* Destroy all tx queues */
464void
465vmxnet3_tq_destroy_all(struct vmxnet3_adapter *adapter)
466{
467	int i;
468
469	for (i = 0; i < adapter->num_tx_queues; i++)
470		vmxnet3_tq_destroy(&adapter->tx_queue[i], adapter);
471}
472
473
474static void
475vmxnet3_tq_init(struct vmxnet3_tx_queue *tq,
476		struct vmxnet3_adapter *adapter)
477{
478	int i;
479
480	/* reset the tx ring contents to 0 and reset the tx ring states */
481	memset(tq->tx_ring.base, 0, tq->tx_ring.size *
482	       sizeof(struct Vmxnet3_TxDesc));
483	tq->tx_ring.next2fill = tq->tx_ring.next2comp = 0;
484	tq->tx_ring.gen = VMXNET3_INIT_GEN;
485
486	memset(tq->data_ring.base, 0,
487	       tq->data_ring.size * tq->txdata_desc_size);
488
489	/* reset the tx comp ring contents to 0 and reset comp ring states */
490	memset(tq->comp_ring.base, 0, tq->comp_ring.size *
491	       sizeof(struct Vmxnet3_TxCompDesc));
492	tq->comp_ring.next2proc = 0;
493	tq->comp_ring.gen = VMXNET3_INIT_GEN;
494
495	/* reset the bookkeeping data */
496	memset(tq->buf_info, 0, sizeof(tq->buf_info[0]) * tq->tx_ring.size);
497	for (i = 0; i < tq->tx_ring.size; i++)
498		tq->buf_info[i].map_type = VMXNET3_MAP_NONE;
499
500	/* stats are not reset */
501}
502
503
504static int
505vmxnet3_tq_create(struct vmxnet3_tx_queue *tq,
506		  struct vmxnet3_adapter *adapter)
507{
508	size_t sz;
509
510	BUG_ON(tq->tx_ring.base || tq->data_ring.base ||
511	       tq->comp_ring.base || tq->buf_info);
512
513	tq->tx_ring.base = dma_alloc_coherent(&adapter->pdev->dev,
514			tq->tx_ring.size * sizeof(struct Vmxnet3_TxDesc),
515			&tq->tx_ring.basePA, GFP_KERNEL);
516	if (!tq->tx_ring.base) {
517		netdev_err(adapter->netdev, "failed to allocate tx ring\n");
518		goto err;
519	}
520
521	tq->data_ring.base = dma_alloc_coherent(&adapter->pdev->dev,
522			tq->data_ring.size * tq->txdata_desc_size,
523			&tq->data_ring.basePA, GFP_KERNEL);
524	if (!tq->data_ring.base) {
525		netdev_err(adapter->netdev, "failed to allocate tx data ring\n");
526		goto err;
527	}
528
529	tq->comp_ring.base = dma_alloc_coherent(&adapter->pdev->dev,
530			tq->comp_ring.size * sizeof(struct Vmxnet3_TxCompDesc),
531			&tq->comp_ring.basePA, GFP_KERNEL);
532	if (!tq->comp_ring.base) {
533		netdev_err(adapter->netdev, "failed to allocate tx comp ring\n");
534		goto err;
535	}
536
537	sz = tq->tx_ring.size * sizeof(tq->buf_info[0]);
538	tq->buf_info = dma_alloc_coherent(&adapter->pdev->dev, sz,
539					  &tq->buf_info_pa, GFP_KERNEL);
540	if (!tq->buf_info)
541		goto err;
542
543	return 0;
544
545err:
546	vmxnet3_tq_destroy(tq, adapter);
547	return -ENOMEM;
548}
549
550static void
551vmxnet3_tq_cleanup_all(struct vmxnet3_adapter *adapter)
552{
553	int i;
554
555	for (i = 0; i < adapter->num_tx_queues; i++)
556		vmxnet3_tq_cleanup(&adapter->tx_queue[i], adapter);
557}
558
559/*
560 *    starting from ring->next2fill, allocate rx buffers for the given ring
561 *    of the rx queue and update the rx desc. stop after @num_to_alloc buffers
562 *    are allocated or allocation fails
563 */
564
565static int
566vmxnet3_rq_alloc_rx_buf(struct vmxnet3_rx_queue *rq, u32 ring_idx,
567			int num_to_alloc, struct vmxnet3_adapter *adapter)
568{
569	int num_allocated = 0;
570	struct vmxnet3_rx_buf_info *rbi_base = rq->buf_info[ring_idx];
571	struct vmxnet3_cmd_ring *ring = &rq->rx_ring[ring_idx];
572	u32 val;
573
574	while (num_allocated <= num_to_alloc) {
575		struct vmxnet3_rx_buf_info *rbi;
576		union Vmxnet3_GenericDesc *gd;
577
578		rbi = rbi_base + ring->next2fill;
579		gd = ring->base + ring->next2fill;
580
581		if (rbi->buf_type == VMXNET3_RX_BUF_SKB) {
582			if (rbi->skb == NULL) {
583				rbi->skb = __netdev_alloc_skb_ip_align(adapter->netdev,
584								       rbi->len,
585								       GFP_KERNEL);
586				if (unlikely(rbi->skb == NULL)) {
587					rq->stats.rx_buf_alloc_failure++;
588					break;
589				}
590
591				rbi->dma_addr = dma_map_single(
592						&adapter->pdev->dev,
593						rbi->skb->data, rbi->len,
594						PCI_DMA_FROMDEVICE);
595				if (dma_mapping_error(&adapter->pdev->dev,
596						      rbi->dma_addr)) {
597					dev_kfree_skb_any(rbi->skb);
598					rbi->skb = NULL;
599					rq->stats.rx_buf_alloc_failure++;
600					break;
601				}
602			} else {
603				/* rx buffer skipped by the device */
604			}
605			val = VMXNET3_RXD_BTYPE_HEAD << VMXNET3_RXD_BTYPE_SHIFT;
606		} else {
607			BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_PAGE ||
608			       rbi->len  != PAGE_SIZE);
609
610			if (rbi->page == NULL) {
611				rbi->page = alloc_page(GFP_ATOMIC);
612				if (unlikely(rbi->page == NULL)) {
613					rq->stats.rx_buf_alloc_failure++;
614					break;
615				}
616				rbi->dma_addr = dma_map_page(
617						&adapter->pdev->dev,
618						rbi->page, 0, PAGE_SIZE,
619						PCI_DMA_FROMDEVICE);
620				if (dma_mapping_error(&adapter->pdev->dev,
621						      rbi->dma_addr)) {
622					put_page(rbi->page);
623					rbi->page = NULL;
624					rq->stats.rx_buf_alloc_failure++;
625					break;
626				}
627			} else {
628				/* rx buffers skipped by the device */
629			}
630			val = VMXNET3_RXD_BTYPE_BODY << VMXNET3_RXD_BTYPE_SHIFT;
631		}
632
633		gd->rxd.addr = cpu_to_le64(rbi->dma_addr);
634		gd->dword[2] = cpu_to_le32((!ring->gen << VMXNET3_RXD_GEN_SHIFT)
635					   | val | rbi->len);
636
637		/* Fill the last buffer but dont mark it ready, or else the
638		 * device will think that the queue is full */
639		if (num_allocated == num_to_alloc)
640			break;
641
642		gd->dword[2] |= cpu_to_le32(ring->gen << VMXNET3_RXD_GEN_SHIFT);
643		num_allocated++;
644		vmxnet3_cmd_ring_adv_next2fill(ring);
645	}
646
647	netdev_dbg(adapter->netdev,
648		"alloc_rx_buf: %d allocated, next2fill %u, next2comp %u\n",
649		num_allocated, ring->next2fill, ring->next2comp);
650
651	/* so that the device can distinguish a full ring and an empty ring */
652	BUG_ON(num_allocated != 0 && ring->next2fill == ring->next2comp);
653
654	return num_allocated;
655}
656
657
658static void
659vmxnet3_append_frag(struct sk_buff *skb, struct Vmxnet3_RxCompDesc *rcd,
660		    struct vmxnet3_rx_buf_info *rbi)
661{
662	skb_frag_t *frag = skb_shinfo(skb)->frags + skb_shinfo(skb)->nr_frags;
663
664	BUG_ON(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS);
665
666	__skb_frag_set_page(frag, rbi->page);
667	skb_frag_off_set(frag, 0);
668	skb_frag_size_set(frag, rcd->len);
669	skb->data_len += rcd->len;
670	skb->truesize += PAGE_SIZE;
671	skb_shinfo(skb)->nr_frags++;
672}
673
674
675static int
676vmxnet3_map_pkt(struct sk_buff *skb, struct vmxnet3_tx_ctx *ctx,
677		struct vmxnet3_tx_queue *tq, struct pci_dev *pdev,
678		struct vmxnet3_adapter *adapter)
679{
680	u32 dw2, len;
681	unsigned long buf_offset;
682	int i;
683	union Vmxnet3_GenericDesc *gdesc;
684	struct vmxnet3_tx_buf_info *tbi = NULL;
685
686	BUG_ON(ctx->copy_size > skb_headlen(skb));
687
688	/* use the previous gen bit for the SOP desc */
689	dw2 = (tq->tx_ring.gen ^ 0x1) << VMXNET3_TXD_GEN_SHIFT;
690
691	ctx->sop_txd = tq->tx_ring.base + tq->tx_ring.next2fill;
692	gdesc = ctx->sop_txd; /* both loops below can be skipped */
693
694	/* no need to map the buffer if headers are copied */
695	if (ctx->copy_size) {
696		ctx->sop_txd->txd.addr = cpu_to_le64(tq->data_ring.basePA +
697					tq->tx_ring.next2fill *
698					tq->txdata_desc_size);
699		ctx->sop_txd->dword[2] = cpu_to_le32(dw2 | ctx->copy_size);
700		ctx->sop_txd->dword[3] = 0;
701
702		tbi = tq->buf_info + tq->tx_ring.next2fill;
703		tbi->map_type = VMXNET3_MAP_NONE;
704
705		netdev_dbg(adapter->netdev,
706			"txd[%u]: 0x%Lx 0x%x 0x%x\n",
707			tq->tx_ring.next2fill,
708			le64_to_cpu(ctx->sop_txd->txd.addr),
709			ctx->sop_txd->dword[2], ctx->sop_txd->dword[3]);
710		vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
711
712		/* use the right gen for non-SOP desc */
713		dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
714	}
715
716	/* linear part can use multiple tx desc if it's big */
717	len = skb_headlen(skb) - ctx->copy_size;
718	buf_offset = ctx->copy_size;
719	while (len) {
720		u32 buf_size;
721
722		if (len < VMXNET3_MAX_TX_BUF_SIZE) {
723			buf_size = len;
724			dw2 |= len;
725		} else {
726			buf_size = VMXNET3_MAX_TX_BUF_SIZE;
727			/* spec says that for TxDesc.len, 0 == 2^14 */
728		}
729
730		tbi = tq->buf_info + tq->tx_ring.next2fill;
731		tbi->map_type = VMXNET3_MAP_SINGLE;
732		tbi->dma_addr = dma_map_single(&adapter->pdev->dev,
733				skb->data + buf_offset, buf_size,
734				PCI_DMA_TODEVICE);
735		if (dma_mapping_error(&adapter->pdev->dev, tbi->dma_addr))
736			return -EFAULT;
737
738		tbi->len = buf_size;
739
740		gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
741		BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
742
743		gdesc->txd.addr = cpu_to_le64(tbi->dma_addr);
744		gdesc->dword[2] = cpu_to_le32(dw2);
745		gdesc->dword[3] = 0;
746
747		netdev_dbg(adapter->netdev,
748			"txd[%u]: 0x%Lx 0x%x 0x%x\n",
749			tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr),
750			le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]);
751		vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
752		dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
753
754		len -= buf_size;
755		buf_offset += buf_size;
756	}
757
758	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
759		const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
760		u32 buf_size;
761
762		buf_offset = 0;
763		len = skb_frag_size(frag);
764		while (len) {
765			tbi = tq->buf_info + tq->tx_ring.next2fill;
766			if (len < VMXNET3_MAX_TX_BUF_SIZE) {
767				buf_size = len;
768				dw2 |= len;
769			} else {
770				buf_size = VMXNET3_MAX_TX_BUF_SIZE;
771				/* spec says that for TxDesc.len, 0 == 2^14 */
772			}
773			tbi->map_type = VMXNET3_MAP_PAGE;
774			tbi->dma_addr = skb_frag_dma_map(&adapter->pdev->dev, frag,
775							 buf_offset, buf_size,
776							 DMA_TO_DEVICE);
777			if (dma_mapping_error(&adapter->pdev->dev, tbi->dma_addr))
778				return -EFAULT;
779
780			tbi->len = buf_size;
781
782			gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
783			BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
784
785			gdesc->txd.addr = cpu_to_le64(tbi->dma_addr);
786			gdesc->dword[2] = cpu_to_le32(dw2);
787			gdesc->dword[3] = 0;
788
789			netdev_dbg(adapter->netdev,
790				"txd[%u]: 0x%llx %u %u\n",
791				tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr),
792				le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]);
793			vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
794			dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
795
796			len -= buf_size;
797			buf_offset += buf_size;
798		}
799	}
800
801	ctx->eop_txd = gdesc;
802
803	/* set the last buf_info for the pkt */
804	tbi->skb = skb;
805	tbi->sop_idx = ctx->sop_txd - tq->tx_ring.base;
806
807	return 0;
808}
809
810
811/* Init all tx queues */
812static void
813vmxnet3_tq_init_all(struct vmxnet3_adapter *adapter)
814{
815	int i;
816
817	for (i = 0; i < adapter->num_tx_queues; i++)
818		vmxnet3_tq_init(&adapter->tx_queue[i], adapter);
819}
820
821
822/*
823 *    parse relevant protocol headers:
824 *      For a tso pkt, relevant headers are L2/3/4 including options
825 *      For a pkt requesting csum offloading, they are L2/3 and may include L4
826 *      if it's a TCP/UDP pkt
827 *
828 * Returns:
829 *    -1:  error happens during parsing
830 *     0:  protocol headers parsed, but too big to be copied
831 *     1:  protocol headers parsed and copied
832 *
833 * Other effects:
834 *    1. related *ctx fields are updated.
835 *    2. ctx->copy_size is # of bytes copied
836 *    3. the portion to be copied is guaranteed to be in the linear part
837 *
838 */
839static int
840vmxnet3_parse_hdr(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
841		  struct vmxnet3_tx_ctx *ctx,
842		  struct vmxnet3_adapter *adapter)
843{
844	u8 protocol = 0;
845
846	if (ctx->mss) {	/* TSO */
847		if (VMXNET3_VERSION_GE_4(adapter) && skb->encapsulation) {
848			ctx->l4_offset = skb_inner_transport_offset(skb);
849			ctx->l4_hdr_size = inner_tcp_hdrlen(skb);
850			ctx->copy_size = ctx->l4_offset + ctx->l4_hdr_size;
851		} else {
852			ctx->l4_offset = skb_transport_offset(skb);
853			ctx->l4_hdr_size = tcp_hdrlen(skb);
854			ctx->copy_size = ctx->l4_offset + ctx->l4_hdr_size;
855		}
856	} else {
857		if (skb->ip_summed == CHECKSUM_PARTIAL) {
858			/* For encap packets, skb_checksum_start_offset refers
859			 * to inner L4 offset. Thus, below works for encap as
860			 * well as non-encap case
861			 */
862			ctx->l4_offset = skb_checksum_start_offset(skb);
863
864			if (VMXNET3_VERSION_GE_4(adapter) &&
865			    skb->encapsulation) {
866				struct iphdr *iph = inner_ip_hdr(skb);
867
868				if (iph->version == 4) {
869					protocol = iph->protocol;
870				} else {
871					const struct ipv6hdr *ipv6h;
872
873					ipv6h = inner_ipv6_hdr(skb);
874					protocol = ipv6h->nexthdr;
875				}
876			} else {
877				if (ctx->ipv4) {
878					const struct iphdr *iph = ip_hdr(skb);
879
880					protocol = iph->protocol;
881				} else if (ctx->ipv6) {
882					const struct ipv6hdr *ipv6h;
883
884					ipv6h = ipv6_hdr(skb);
885					protocol = ipv6h->nexthdr;
886				}
887			}
888
889			switch (protocol) {
890			case IPPROTO_TCP:
891				ctx->l4_hdr_size = skb->encapsulation ? inner_tcp_hdrlen(skb) :
892						   tcp_hdrlen(skb);
893				break;
894			case IPPROTO_UDP:
895				ctx->l4_hdr_size = sizeof(struct udphdr);
896				break;
897			default:
898				ctx->l4_hdr_size = 0;
899				break;
900			}
901
902			ctx->copy_size = min(ctx->l4_offset +
903					 ctx->l4_hdr_size, skb->len);
904		} else {
905			ctx->l4_offset = 0;
906			ctx->l4_hdr_size = 0;
907			/* copy as much as allowed */
908			ctx->copy_size = min_t(unsigned int,
909					       tq->txdata_desc_size,
910					       skb_headlen(skb));
911		}
912
913		if (skb->len <= VMXNET3_HDR_COPY_SIZE)
914			ctx->copy_size = skb->len;
915
916		/* make sure headers are accessible directly */
917		if (unlikely(!pskb_may_pull(skb, ctx->copy_size)))
918			goto err;
919	}
920
921	if (unlikely(ctx->copy_size > tq->txdata_desc_size)) {
922		tq->stats.oversized_hdr++;
923		ctx->copy_size = 0;
924		return 0;
925	}
926
927	return 1;
928err:
929	return -1;
930}
931
932/*
933 *    copy relevant protocol headers to the transmit ring:
934 *      For a tso pkt, relevant headers are L2/3/4 including options
935 *      For a pkt requesting csum offloading, they are L2/3 and may include L4
936 *      if it's a TCP/UDP pkt
937 *
938 *
939 *    Note that this requires that vmxnet3_parse_hdr be called first to set the
940 *      appropriate bits in ctx first
941 */
942static void
943vmxnet3_copy_hdr(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
944		 struct vmxnet3_tx_ctx *ctx,
945		 struct vmxnet3_adapter *adapter)
946{
947	struct Vmxnet3_TxDataDesc *tdd;
948
949	tdd = (struct Vmxnet3_TxDataDesc *)((u8 *)tq->data_ring.base +
950					    tq->tx_ring.next2fill *
951					    tq->txdata_desc_size);
952
953	memcpy(tdd->data, skb->data, ctx->copy_size);
954	netdev_dbg(adapter->netdev,
955		"copy %u bytes to dataRing[%u]\n",
956		ctx->copy_size, tq->tx_ring.next2fill);
957}
958
959
960static void
961vmxnet3_prepare_inner_tso(struct sk_buff *skb,
962			  struct vmxnet3_tx_ctx *ctx)
963{
964	struct tcphdr *tcph = inner_tcp_hdr(skb);
965	struct iphdr *iph = inner_ip_hdr(skb);
966
967	if (iph->version == 4) {
968		iph->check = 0;
969		tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, 0,
970						 IPPROTO_TCP, 0);
971	} else {
972		struct ipv6hdr *iph = inner_ipv6_hdr(skb);
973
974		tcph->check = ~csum_ipv6_magic(&iph->saddr, &iph->daddr, 0,
975					       IPPROTO_TCP, 0);
976	}
977}
978
979static void
980vmxnet3_prepare_tso(struct sk_buff *skb,
981		    struct vmxnet3_tx_ctx *ctx)
982{
983	struct tcphdr *tcph = tcp_hdr(skb);
984
985	if (ctx->ipv4) {
986		struct iphdr *iph = ip_hdr(skb);
987
988		iph->check = 0;
989		tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, 0,
990						 IPPROTO_TCP, 0);
991	} else if (ctx->ipv6) {
992		tcp_v6_gso_csum_prep(skb);
993	}
994}
995
996static int txd_estimate(const struct sk_buff *skb)
997{
998	int count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) + 1;
999	int i;
1000
1001	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1002		const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1003
1004		count += VMXNET3_TXD_NEEDED(skb_frag_size(frag));
1005	}
1006	return count;
1007}
1008
1009/*
1010 * Transmits a pkt thru a given tq
1011 * Returns:
1012 *    NETDEV_TX_OK:      descriptors are setup successfully
1013 *    NETDEV_TX_OK:      error occurred, the pkt is dropped
1014 *    NETDEV_TX_BUSY:    tx ring is full, queue is stopped
1015 *
1016 * Side-effects:
1017 *    1. tx ring may be changed
1018 *    2. tq stats may be updated accordingly
1019 *    3. shared->txNumDeferred may be updated
1020 */
1021
1022static int
1023vmxnet3_tq_xmit(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
1024		struct vmxnet3_adapter *adapter, struct net_device *netdev)
1025{
1026	int ret;
1027	u32 count;
1028	int num_pkts;
1029	int tx_num_deferred;
1030	unsigned long flags;
1031	struct vmxnet3_tx_ctx ctx;
1032	union Vmxnet3_GenericDesc *gdesc;
1033#ifdef __BIG_ENDIAN_BITFIELD
1034	/* Use temporary descriptor to avoid touching bits multiple times */
1035	union Vmxnet3_GenericDesc tempTxDesc;
1036#endif
1037
1038	count = txd_estimate(skb);
1039
1040	ctx.ipv4 = (vlan_get_protocol(skb) == cpu_to_be16(ETH_P_IP));
1041	ctx.ipv6 = (vlan_get_protocol(skb) == cpu_to_be16(ETH_P_IPV6));
1042
1043	ctx.mss = skb_shinfo(skb)->gso_size;
1044	if (ctx.mss) {
1045		if (skb_header_cloned(skb)) {
1046			if (unlikely(pskb_expand_head(skb, 0, 0,
1047						      GFP_ATOMIC) != 0)) {
1048				tq->stats.drop_tso++;
1049				goto drop_pkt;
1050			}
1051			tq->stats.copy_skb_header++;
1052		}
1053		if (skb->encapsulation) {
1054			vmxnet3_prepare_inner_tso(skb, &ctx);
1055		} else {
1056			vmxnet3_prepare_tso(skb, &ctx);
1057		}
1058	} else {
1059		if (unlikely(count > VMXNET3_MAX_TXD_PER_PKT)) {
1060
1061			/* non-tso pkts must not use more than
1062			 * VMXNET3_MAX_TXD_PER_PKT entries
1063			 */
1064			if (skb_linearize(skb) != 0) {
1065				tq->stats.drop_too_many_frags++;
1066				goto drop_pkt;
1067			}
1068			tq->stats.linearized++;
1069
1070			/* recalculate the # of descriptors to use */
1071			count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) + 1;
1072		}
1073	}
1074
1075	ret = vmxnet3_parse_hdr(skb, tq, &ctx, adapter);
1076	if (ret >= 0) {
1077		BUG_ON(ret <= 0 && ctx.copy_size != 0);
1078		/* hdrs parsed, check against other limits */
1079		if (ctx.mss) {
1080			if (unlikely(ctx.l4_offset + ctx.l4_hdr_size >
1081				     VMXNET3_MAX_TX_BUF_SIZE)) {
1082				tq->stats.drop_oversized_hdr++;
1083				goto drop_pkt;
1084			}
1085		} else {
1086			if (skb->ip_summed == CHECKSUM_PARTIAL) {
1087				if (unlikely(ctx.l4_offset +
1088					     skb->csum_offset >
1089					     VMXNET3_MAX_CSUM_OFFSET)) {
1090					tq->stats.drop_oversized_hdr++;
1091					goto drop_pkt;
1092				}
1093			}
1094		}
1095	} else {
1096		tq->stats.drop_hdr_inspect_err++;
1097		goto drop_pkt;
1098	}
1099
1100	spin_lock_irqsave(&tq->tx_lock, flags);
1101
1102	if (count > vmxnet3_cmd_ring_desc_avail(&tq->tx_ring)) {
1103		tq->stats.tx_ring_full++;
1104		netdev_dbg(adapter->netdev,
1105			"tx queue stopped on %s, next2comp %u"
1106			" next2fill %u\n", adapter->netdev->name,
1107			tq->tx_ring.next2comp, tq->tx_ring.next2fill);
1108
1109		vmxnet3_tq_stop(tq, adapter);
1110		spin_unlock_irqrestore(&tq->tx_lock, flags);
1111		return NETDEV_TX_BUSY;
1112	}
1113
1114
1115	vmxnet3_copy_hdr(skb, tq, &ctx, adapter);
1116
1117	/* fill tx descs related to addr & len */
1118	if (vmxnet3_map_pkt(skb, &ctx, tq, adapter->pdev, adapter))
1119		goto unlock_drop_pkt;
1120
1121	/* setup the EOP desc */
1122	ctx.eop_txd->dword[3] = cpu_to_le32(VMXNET3_TXD_CQ | VMXNET3_TXD_EOP);
1123
1124	/* setup the SOP desc */
1125#ifdef __BIG_ENDIAN_BITFIELD
1126	gdesc = &tempTxDesc;
1127	gdesc->dword[2] = ctx.sop_txd->dword[2];
1128	gdesc->dword[3] = ctx.sop_txd->dword[3];
1129#else
1130	gdesc = ctx.sop_txd;
1131#endif
1132	tx_num_deferred = le32_to_cpu(tq->shared->txNumDeferred);
1133	if (ctx.mss) {
1134		if (VMXNET3_VERSION_GE_4(adapter) && skb->encapsulation) {
1135			gdesc->txd.hlen = ctx.l4_offset + ctx.l4_hdr_size;
1136			gdesc->txd.om = VMXNET3_OM_ENCAP;
1137			gdesc->txd.msscof = ctx.mss;
1138
1139			if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL_CSUM)
1140				gdesc->txd.oco = 1;
1141		} else {
1142			gdesc->txd.hlen = ctx.l4_offset + ctx.l4_hdr_size;
1143			gdesc->txd.om = VMXNET3_OM_TSO;
1144			gdesc->txd.msscof = ctx.mss;
1145		}
1146		num_pkts = (skb->len - gdesc->txd.hlen + ctx.mss - 1) / ctx.mss;
1147	} else {
1148		if (skb->ip_summed == CHECKSUM_PARTIAL) {
1149			if (VMXNET3_VERSION_GE_4(adapter) &&
1150			    skb->encapsulation) {
1151				gdesc->txd.hlen = ctx.l4_offset +
1152						  ctx.l4_hdr_size;
1153				gdesc->txd.om = VMXNET3_OM_ENCAP;
1154				gdesc->txd.msscof = 0;		/* Reserved */
1155			} else {
1156				gdesc->txd.hlen = ctx.l4_offset;
1157				gdesc->txd.om = VMXNET3_OM_CSUM;
1158				gdesc->txd.msscof = ctx.l4_offset +
1159						    skb->csum_offset;
1160			}
1161		} else {
1162			gdesc->txd.om = 0;
1163			gdesc->txd.msscof = 0;
1164		}
1165		num_pkts = 1;
1166	}
1167	le32_add_cpu(&tq->shared->txNumDeferred, num_pkts);
1168	tx_num_deferred += num_pkts;
1169
1170	if (skb_vlan_tag_present(skb)) {
1171		gdesc->txd.ti = 1;
1172		gdesc->txd.tci = skb_vlan_tag_get(skb);
1173	}
1174
1175	/* Ensure that the write to (&gdesc->txd)->gen will be observed after
1176	 * all other writes to &gdesc->txd.
1177	 */
1178	dma_wmb();
1179
1180	/* finally flips the GEN bit of the SOP desc. */
1181	gdesc->dword[2] = cpu_to_le32(le32_to_cpu(gdesc->dword[2]) ^
1182						  VMXNET3_TXD_GEN);
1183#ifdef __BIG_ENDIAN_BITFIELD
1184	/* Finished updating in bitfields of Tx Desc, so write them in original
1185	 * place.
1186	 */
1187	vmxnet3_TxDescToLe((struct Vmxnet3_TxDesc *)gdesc,
1188			   (struct Vmxnet3_TxDesc *)ctx.sop_txd);
1189	gdesc = ctx.sop_txd;
1190#endif
1191	netdev_dbg(adapter->netdev,
1192		"txd[%u]: SOP 0x%Lx 0x%x 0x%x\n",
1193		(u32)(ctx.sop_txd -
1194		tq->tx_ring.base), le64_to_cpu(gdesc->txd.addr),
1195		le32_to_cpu(gdesc->dword[2]), le32_to_cpu(gdesc->dword[3]));
1196
1197	spin_unlock_irqrestore(&tq->tx_lock, flags);
1198
1199	if (tx_num_deferred >= le32_to_cpu(tq->shared->txThreshold)) {
1200		tq->shared->txNumDeferred = 0;
1201		VMXNET3_WRITE_BAR0_REG(adapter,
1202				       VMXNET3_REG_TXPROD + tq->qid * 8,
1203				       tq->tx_ring.next2fill);
1204	}
1205
1206	return NETDEV_TX_OK;
1207
1208unlock_drop_pkt:
1209	spin_unlock_irqrestore(&tq->tx_lock, flags);
1210drop_pkt:
1211	tq->stats.drop_total++;
1212	dev_kfree_skb_any(skb);
1213	return NETDEV_TX_OK;
1214}
1215
1216
1217static netdev_tx_t
1218vmxnet3_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
1219{
1220	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1221
1222	BUG_ON(skb->queue_mapping > adapter->num_tx_queues);
1223	return vmxnet3_tq_xmit(skb,
1224			       &adapter->tx_queue[skb->queue_mapping],
1225			       adapter, netdev);
1226}
1227
1228
1229static void
1230vmxnet3_rx_csum(struct vmxnet3_adapter *adapter,
1231		struct sk_buff *skb,
1232		union Vmxnet3_GenericDesc *gdesc)
1233{
1234	if (!gdesc->rcd.cnc && adapter->netdev->features & NETIF_F_RXCSUM) {
1235		if (gdesc->rcd.v4 &&
1236		    (le32_to_cpu(gdesc->dword[3]) &
1237		     VMXNET3_RCD_CSUM_OK) == VMXNET3_RCD_CSUM_OK) {
1238			skb->ip_summed = CHECKSUM_UNNECESSARY;
1239			if ((le32_to_cpu(gdesc->dword[0]) &
1240				     (1UL << VMXNET3_RCD_HDR_INNER_SHIFT))) {
1241				skb->csum_level = 1;
1242			}
1243			WARN_ON_ONCE(!(gdesc->rcd.tcp || gdesc->rcd.udp) &&
1244				     !(le32_to_cpu(gdesc->dword[0]) &
1245				     (1UL << VMXNET3_RCD_HDR_INNER_SHIFT)));
1246			WARN_ON_ONCE(gdesc->rcd.frg &&
1247				     !(le32_to_cpu(gdesc->dword[0]) &
1248				     (1UL << VMXNET3_RCD_HDR_INNER_SHIFT)));
1249		} else if (gdesc->rcd.v6 && (le32_to_cpu(gdesc->dword[3]) &
1250					     (1 << VMXNET3_RCD_TUC_SHIFT))) {
1251			skb->ip_summed = CHECKSUM_UNNECESSARY;
1252			if ((le32_to_cpu(gdesc->dword[0]) &
1253				     (1UL << VMXNET3_RCD_HDR_INNER_SHIFT))) {
1254				skb->csum_level = 1;
1255			}
1256			WARN_ON_ONCE(!(gdesc->rcd.tcp || gdesc->rcd.udp) &&
1257				     !(le32_to_cpu(gdesc->dword[0]) &
1258				     (1UL << VMXNET3_RCD_HDR_INNER_SHIFT)));
1259			WARN_ON_ONCE(gdesc->rcd.frg &&
1260				     !(le32_to_cpu(gdesc->dword[0]) &
1261				     (1UL << VMXNET3_RCD_HDR_INNER_SHIFT)));
1262		} else {
1263			if (gdesc->rcd.csum) {
1264				skb->csum = htons(gdesc->rcd.csum);
1265				skb->ip_summed = CHECKSUM_PARTIAL;
1266			} else {
1267				skb_checksum_none_assert(skb);
1268			}
1269		}
1270	} else {
1271		skb_checksum_none_assert(skb);
1272	}
1273}
1274
1275
1276static void
1277vmxnet3_rx_error(struct vmxnet3_rx_queue *rq, struct Vmxnet3_RxCompDesc *rcd,
1278		 struct vmxnet3_rx_ctx *ctx,  struct vmxnet3_adapter *adapter)
1279{
1280	rq->stats.drop_err++;
1281	if (!rcd->fcs)
1282		rq->stats.drop_fcs++;
1283
1284	rq->stats.drop_total++;
1285
1286	/*
1287	 * We do not unmap and chain the rx buffer to the skb.
1288	 * We basically pretend this buffer is not used and will be recycled
1289	 * by vmxnet3_rq_alloc_rx_buf()
1290	 */
1291
1292	/*
1293	 * ctx->skb may be NULL if this is the first and the only one
1294	 * desc for the pkt
1295	 */
1296	if (ctx->skb)
1297		dev_kfree_skb_irq(ctx->skb);
1298
1299	ctx->skb = NULL;
1300}
1301
1302
1303static u32
1304vmxnet3_get_hdr_len(struct vmxnet3_adapter *adapter, struct sk_buff *skb,
1305		    union Vmxnet3_GenericDesc *gdesc)
1306{
1307	u32 hlen, maplen;
1308	union {
1309		void *ptr;
1310		struct ethhdr *eth;
1311		struct vlan_ethhdr *veth;
1312		struct iphdr *ipv4;
1313		struct ipv6hdr *ipv6;
1314		struct tcphdr *tcp;
1315	} hdr;
1316	BUG_ON(gdesc->rcd.tcp == 0);
1317
1318	maplen = skb_headlen(skb);
1319	if (unlikely(sizeof(struct iphdr) + sizeof(struct tcphdr) > maplen))
1320		return 0;
1321
1322	if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
1323	    skb->protocol == cpu_to_be16(ETH_P_8021AD))
1324		hlen = sizeof(struct vlan_ethhdr);
1325	else
1326		hlen = sizeof(struct ethhdr);
1327
1328	hdr.eth = eth_hdr(skb);
1329	if (gdesc->rcd.v4) {
1330		BUG_ON(hdr.eth->h_proto != htons(ETH_P_IP) &&
1331		       hdr.veth->h_vlan_encapsulated_proto != htons(ETH_P_IP));
1332		hdr.ptr += hlen;
1333		BUG_ON(hdr.ipv4->protocol != IPPROTO_TCP);
1334		hlen = hdr.ipv4->ihl << 2;
1335		hdr.ptr += hdr.ipv4->ihl << 2;
1336	} else if (gdesc->rcd.v6) {
1337		BUG_ON(hdr.eth->h_proto != htons(ETH_P_IPV6) &&
1338		       hdr.veth->h_vlan_encapsulated_proto != htons(ETH_P_IPV6));
1339		hdr.ptr += hlen;
1340		/* Use an estimated value, since we also need to handle
1341		 * TSO case.
1342		 */
1343		if (hdr.ipv6->nexthdr != IPPROTO_TCP)
1344			return sizeof(struct ipv6hdr) + sizeof(struct tcphdr);
1345		hlen = sizeof(struct ipv6hdr);
1346		hdr.ptr += sizeof(struct ipv6hdr);
1347	} else {
1348		/* Non-IP pkt, dont estimate header length */
1349		return 0;
1350	}
1351
1352	if (hlen + sizeof(struct tcphdr) > maplen)
1353		return 0;
1354
1355	return (hlen + (hdr.tcp->doff << 2));
1356}
1357
1358static int
1359vmxnet3_rq_rx_complete(struct vmxnet3_rx_queue *rq,
1360		       struct vmxnet3_adapter *adapter, int quota)
1361{
1362	static const u32 rxprod_reg[2] = {
1363		VMXNET3_REG_RXPROD, VMXNET3_REG_RXPROD2
1364	};
1365	u32 num_pkts = 0;
1366	bool skip_page_frags = false;
1367	bool encap_lro = false;
1368	struct Vmxnet3_RxCompDesc *rcd;
1369	struct vmxnet3_rx_ctx *ctx = &rq->rx_ctx;
1370	u16 segCnt = 0, mss = 0;
1371#ifdef __BIG_ENDIAN_BITFIELD
1372	struct Vmxnet3_RxDesc rxCmdDesc;
1373	struct Vmxnet3_RxCompDesc rxComp;
1374#endif
1375	vmxnet3_getRxComp(rcd, &rq->comp_ring.base[rq->comp_ring.next2proc].rcd,
1376			  &rxComp);
1377	while (rcd->gen == rq->comp_ring.gen) {
1378		struct vmxnet3_rx_buf_info *rbi;
1379		struct sk_buff *skb, *new_skb = NULL;
1380		struct page *new_page = NULL;
1381		dma_addr_t new_dma_addr;
1382		int num_to_alloc;
1383		struct Vmxnet3_RxDesc *rxd;
1384		u32 idx, ring_idx;
1385		struct vmxnet3_cmd_ring	*ring = NULL;
1386		if (num_pkts >= quota) {
1387			/* we may stop even before we see the EOP desc of
1388			 * the current pkt
1389			 */
1390			break;
1391		}
1392
1393		/* Prevent any rcd field from being (speculatively) read before
1394		 * rcd->gen is read.
1395		 */
1396		dma_rmb();
1397
1398		BUG_ON(rcd->rqID != rq->qid && rcd->rqID != rq->qid2 &&
1399		       rcd->rqID != rq->dataRingQid);
1400		idx = rcd->rxdIdx;
1401		ring_idx = VMXNET3_GET_RING_IDX(adapter, rcd->rqID);
1402		ring = rq->rx_ring + ring_idx;
1403		vmxnet3_getRxDesc(rxd, &rq->rx_ring[ring_idx].base[idx].rxd,
1404				  &rxCmdDesc);
1405		rbi = rq->buf_info[ring_idx] + idx;
1406
1407		BUG_ON(rxd->addr != rbi->dma_addr ||
1408		       rxd->len != rbi->len);
1409
1410		if (unlikely(rcd->eop && rcd->err)) {
1411			vmxnet3_rx_error(rq, rcd, ctx, adapter);
1412			goto rcd_done;
1413		}
1414
1415		if (rcd->sop) { /* first buf of the pkt */
1416			bool rxDataRingUsed;
1417			u16 len;
1418
1419			BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_HEAD ||
1420			       (rcd->rqID != rq->qid &&
1421				rcd->rqID != rq->dataRingQid));
1422
1423			BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_SKB);
1424			BUG_ON(ctx->skb != NULL || rbi->skb == NULL);
1425
1426			if (unlikely(rcd->len == 0)) {
1427				/* Pretend the rx buffer is skipped. */
1428				BUG_ON(!(rcd->sop && rcd->eop));
1429				netdev_dbg(adapter->netdev,
1430					"rxRing[%u][%u] 0 length\n",
1431					ring_idx, idx);
1432				goto rcd_done;
1433			}
1434
1435			skip_page_frags = false;
1436			ctx->skb = rbi->skb;
1437
1438			rxDataRingUsed =
1439				VMXNET3_RX_DATA_RING(adapter, rcd->rqID);
1440			len = rxDataRingUsed ? rcd->len : rbi->len;
1441			new_skb = netdev_alloc_skb_ip_align(adapter->netdev,
1442							    len);
1443			if (new_skb == NULL) {
1444				/* Skb allocation failed, do not handover this
1445				 * skb to stack. Reuse it. Drop the existing pkt
1446				 */
1447				rq->stats.rx_buf_alloc_failure++;
1448				ctx->skb = NULL;
1449				rq->stats.drop_total++;
1450				skip_page_frags = true;
1451				goto rcd_done;
1452			}
1453
1454			if (rxDataRingUsed) {
1455				size_t sz;
1456
1457				BUG_ON(rcd->len > rq->data_ring.desc_size);
1458
1459				ctx->skb = new_skb;
1460				sz = rcd->rxdIdx * rq->data_ring.desc_size;
1461				memcpy(new_skb->data,
1462				       &rq->data_ring.base[sz], rcd->len);
1463			} else {
1464				ctx->skb = rbi->skb;
1465
1466				new_dma_addr =
1467					dma_map_single(&adapter->pdev->dev,
1468						       new_skb->data, rbi->len,
1469						       PCI_DMA_FROMDEVICE);
1470				if (dma_mapping_error(&adapter->pdev->dev,
1471						      new_dma_addr)) {
1472					dev_kfree_skb(new_skb);
1473					/* Skb allocation failed, do not
1474					 * handover this skb to stack. Reuse
1475					 * it. Drop the existing pkt.
1476					 */
1477					rq->stats.rx_buf_alloc_failure++;
1478					ctx->skb = NULL;
1479					rq->stats.drop_total++;
1480					skip_page_frags = true;
1481					goto rcd_done;
1482				}
1483
1484				dma_unmap_single(&adapter->pdev->dev,
1485						 rbi->dma_addr,
1486						 rbi->len,
1487						 PCI_DMA_FROMDEVICE);
1488
1489				/* Immediate refill */
1490				rbi->skb = new_skb;
1491				rbi->dma_addr = new_dma_addr;
1492				rxd->addr = cpu_to_le64(rbi->dma_addr);
1493				rxd->len = rbi->len;
1494			}
1495
1496#ifdef VMXNET3_RSS
1497			if (rcd->rssType != VMXNET3_RCD_RSS_TYPE_NONE &&
1498			    (adapter->netdev->features & NETIF_F_RXHASH))
1499				skb_set_hash(ctx->skb,
1500					     le32_to_cpu(rcd->rssHash),
1501					     PKT_HASH_TYPE_L3);
1502#endif
1503			skb_put(ctx->skb, rcd->len);
1504
1505			if (VMXNET3_VERSION_GE_2(adapter) &&
1506			    rcd->type == VMXNET3_CDTYPE_RXCOMP_LRO) {
1507				struct Vmxnet3_RxCompDescExt *rcdlro;
1508				union Vmxnet3_GenericDesc *gdesc;
1509
1510				rcdlro = (struct Vmxnet3_RxCompDescExt *)rcd;
1511				gdesc = (union Vmxnet3_GenericDesc *)rcd;
1512
1513				segCnt = rcdlro->segCnt;
1514				WARN_ON_ONCE(segCnt == 0);
1515				mss = rcdlro->mss;
1516				if (unlikely(segCnt <= 1))
1517					segCnt = 0;
1518				encap_lro = (le32_to_cpu(gdesc->dword[0]) &
1519					(1UL << VMXNET3_RCD_HDR_INNER_SHIFT));
1520			} else {
1521				segCnt = 0;
1522			}
1523		} else {
1524			BUG_ON(ctx->skb == NULL && !skip_page_frags);
1525
1526			/* non SOP buffer must be type 1 in most cases */
1527			BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_PAGE);
1528			BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_BODY);
1529
1530			/* If an sop buffer was dropped, skip all
1531			 * following non-sop fragments. They will be reused.
1532			 */
1533			if (skip_page_frags)
1534				goto rcd_done;
1535
1536			if (rcd->len) {
1537				new_page = alloc_page(GFP_ATOMIC);
1538				/* Replacement page frag could not be allocated.
1539				 * Reuse this page. Drop the pkt and free the
1540				 * skb which contained this page as a frag. Skip
1541				 * processing all the following non-sop frags.
1542				 */
1543				if (unlikely(!new_page)) {
1544					rq->stats.rx_buf_alloc_failure++;
1545					dev_kfree_skb(ctx->skb);
1546					ctx->skb = NULL;
1547					skip_page_frags = true;
1548					goto rcd_done;
1549				}
1550				new_dma_addr = dma_map_page(&adapter->pdev->dev,
1551							    new_page,
1552							    0, PAGE_SIZE,
1553							    PCI_DMA_FROMDEVICE);
1554				if (dma_mapping_error(&adapter->pdev->dev,
1555						      new_dma_addr)) {
1556					put_page(new_page);
1557					rq->stats.rx_buf_alloc_failure++;
1558					dev_kfree_skb(ctx->skb);
1559					ctx->skb = NULL;
1560					skip_page_frags = true;
1561					goto rcd_done;
1562				}
1563
1564				dma_unmap_page(&adapter->pdev->dev,
1565					       rbi->dma_addr, rbi->len,
1566					       PCI_DMA_FROMDEVICE);
1567
1568				vmxnet3_append_frag(ctx->skb, rcd, rbi);
1569
1570				/* Immediate refill */
1571				rbi->page = new_page;
1572				rbi->dma_addr = new_dma_addr;
1573				rxd->addr = cpu_to_le64(rbi->dma_addr);
1574				rxd->len = rbi->len;
1575			}
1576		}
1577
1578
1579		skb = ctx->skb;
1580		if (rcd->eop) {
1581			u32 mtu = adapter->netdev->mtu;
1582			skb->len += skb->data_len;
1583
1584			vmxnet3_rx_csum(adapter, skb,
1585					(union Vmxnet3_GenericDesc *)rcd);
1586			skb->protocol = eth_type_trans(skb, adapter->netdev);
1587			if ((!rcd->tcp && !encap_lro) ||
1588			    !(adapter->netdev->features & NETIF_F_LRO))
1589				goto not_lro;
1590
1591			if (segCnt != 0 && mss != 0) {
1592				skb_shinfo(skb)->gso_type = rcd->v4 ?
1593					SKB_GSO_TCPV4 : SKB_GSO_TCPV6;
1594				skb_shinfo(skb)->gso_size = mss;
1595				skb_shinfo(skb)->gso_segs = segCnt;
1596			} else if ((segCnt != 0 || skb->len > mtu) && !encap_lro) {
1597				u32 hlen;
1598
1599				hlen = vmxnet3_get_hdr_len(adapter, skb,
1600					(union Vmxnet3_GenericDesc *)rcd);
1601				if (hlen == 0)
1602					goto not_lro;
1603
1604				skb_shinfo(skb)->gso_type =
1605					rcd->v4 ? SKB_GSO_TCPV4 : SKB_GSO_TCPV6;
1606				if (segCnt != 0) {
1607					skb_shinfo(skb)->gso_segs = segCnt;
1608					skb_shinfo(skb)->gso_size =
1609						DIV_ROUND_UP(skb->len -
1610							hlen, segCnt);
1611				} else {
1612					skb_shinfo(skb)->gso_size = mtu - hlen;
1613				}
1614			}
1615not_lro:
1616			if (unlikely(rcd->ts))
1617				__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), rcd->tci);
1618
1619			if (adapter->netdev->features & NETIF_F_LRO)
1620				netif_receive_skb(skb);
1621			else
1622				napi_gro_receive(&rq->napi, skb);
1623
1624			ctx->skb = NULL;
1625			encap_lro = false;
1626			num_pkts++;
1627		}
1628
1629rcd_done:
1630		/* device may have skipped some rx descs */
1631		ring->next2comp = idx;
1632		num_to_alloc = vmxnet3_cmd_ring_desc_avail(ring);
1633		ring = rq->rx_ring + ring_idx;
1634
1635		/* Ensure that the writes to rxd->gen bits will be observed
1636		 * after all other writes to rxd objects.
1637		 */
1638		dma_wmb();
1639
1640		while (num_to_alloc) {
1641			vmxnet3_getRxDesc(rxd, &ring->base[ring->next2fill].rxd,
1642					  &rxCmdDesc);
1643			BUG_ON(!rxd->addr);
1644
1645			/* Recv desc is ready to be used by the device */
1646			rxd->gen = ring->gen;
1647			vmxnet3_cmd_ring_adv_next2fill(ring);
1648			num_to_alloc--;
1649		}
1650
1651		/* if needed, update the register */
1652		if (unlikely(rq->shared->updateRxProd)) {
1653			VMXNET3_WRITE_BAR0_REG(adapter,
1654					       rxprod_reg[ring_idx] + rq->qid * 8,
1655					       ring->next2fill);
1656		}
1657
1658		vmxnet3_comp_ring_adv_next2proc(&rq->comp_ring);
1659		vmxnet3_getRxComp(rcd,
1660				  &rq->comp_ring.base[rq->comp_ring.next2proc].rcd, &rxComp);
1661	}
1662
1663	return num_pkts;
1664}
1665
1666
1667static void
1668vmxnet3_rq_cleanup(struct vmxnet3_rx_queue *rq,
1669		   struct vmxnet3_adapter *adapter)
1670{
1671	u32 i, ring_idx;
1672	struct Vmxnet3_RxDesc *rxd;
1673
1674	/* ring has already been cleaned up */
1675	if (!rq->rx_ring[0].base)
1676		return;
1677
1678	for (ring_idx = 0; ring_idx < 2; ring_idx++) {
1679		for (i = 0; i < rq->rx_ring[ring_idx].size; i++) {
1680#ifdef __BIG_ENDIAN_BITFIELD
1681			struct Vmxnet3_RxDesc rxDesc;
1682#endif
1683			vmxnet3_getRxDesc(rxd,
1684				&rq->rx_ring[ring_idx].base[i].rxd, &rxDesc);
1685
1686			if (rxd->btype == VMXNET3_RXD_BTYPE_HEAD &&
1687					rq->buf_info[ring_idx][i].skb) {
1688				dma_unmap_single(&adapter->pdev->dev, rxd->addr,
1689						 rxd->len, PCI_DMA_FROMDEVICE);
1690				dev_kfree_skb(rq->buf_info[ring_idx][i].skb);
1691				rq->buf_info[ring_idx][i].skb = NULL;
1692			} else if (rxd->btype == VMXNET3_RXD_BTYPE_BODY &&
1693					rq->buf_info[ring_idx][i].page) {
1694				dma_unmap_page(&adapter->pdev->dev, rxd->addr,
1695					       rxd->len, PCI_DMA_FROMDEVICE);
1696				put_page(rq->buf_info[ring_idx][i].page);
1697				rq->buf_info[ring_idx][i].page = NULL;
1698			}
1699		}
1700
1701		rq->rx_ring[ring_idx].gen = VMXNET3_INIT_GEN;
1702		rq->rx_ring[ring_idx].next2fill =
1703					rq->rx_ring[ring_idx].next2comp = 0;
1704	}
1705
1706	rq->comp_ring.gen = VMXNET3_INIT_GEN;
1707	rq->comp_ring.next2proc = 0;
1708}
1709
1710
1711static void
1712vmxnet3_rq_cleanup_all(struct vmxnet3_adapter *adapter)
1713{
1714	int i;
1715
1716	for (i = 0; i < adapter->num_rx_queues; i++)
1717		vmxnet3_rq_cleanup(&adapter->rx_queue[i], adapter);
1718}
1719
1720
1721static void vmxnet3_rq_destroy(struct vmxnet3_rx_queue *rq,
1722			       struct vmxnet3_adapter *adapter)
1723{
1724	int i;
1725	int j;
1726
1727	/* all rx buffers must have already been freed */
1728	for (i = 0; i < 2; i++) {
1729		if (rq->buf_info[i]) {
1730			for (j = 0; j < rq->rx_ring[i].size; j++)
1731				BUG_ON(rq->buf_info[i][j].page != NULL);
1732		}
1733	}
1734
1735
1736	for (i = 0; i < 2; i++) {
1737		if (rq->rx_ring[i].base) {
1738			dma_free_coherent(&adapter->pdev->dev,
1739					  rq->rx_ring[i].size
1740					  * sizeof(struct Vmxnet3_RxDesc),
1741					  rq->rx_ring[i].base,
1742					  rq->rx_ring[i].basePA);
1743			rq->rx_ring[i].base = NULL;
1744		}
1745	}
1746
1747	if (rq->data_ring.base) {
1748		dma_free_coherent(&adapter->pdev->dev,
1749				  rq->rx_ring[0].size * rq->data_ring.desc_size,
1750				  rq->data_ring.base, rq->data_ring.basePA);
1751		rq->data_ring.base = NULL;
1752	}
1753
1754	if (rq->comp_ring.base) {
1755		dma_free_coherent(&adapter->pdev->dev, rq->comp_ring.size
1756				  * sizeof(struct Vmxnet3_RxCompDesc),
1757				  rq->comp_ring.base, rq->comp_ring.basePA);
1758		rq->comp_ring.base = NULL;
1759	}
1760
1761	if (rq->buf_info[0]) {
1762		size_t sz = sizeof(struct vmxnet3_rx_buf_info) *
1763			(rq->rx_ring[0].size + rq->rx_ring[1].size);
1764		dma_free_coherent(&adapter->pdev->dev, sz, rq->buf_info[0],
1765				  rq->buf_info_pa);
1766		rq->buf_info[0] = rq->buf_info[1] = NULL;
1767	}
1768}
1769
1770static void
1771vmxnet3_rq_destroy_all_rxdataring(struct vmxnet3_adapter *adapter)
1772{
1773	int i;
1774
1775	for (i = 0; i < adapter->num_rx_queues; i++) {
1776		struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
1777
1778		if (rq->data_ring.base) {
1779			dma_free_coherent(&adapter->pdev->dev,
1780					  (rq->rx_ring[0].size *
1781					  rq->data_ring.desc_size),
1782					  rq->data_ring.base,
1783					  rq->data_ring.basePA);
1784			rq->data_ring.base = NULL;
1785			rq->data_ring.desc_size = 0;
1786		}
1787	}
1788}
1789
1790static int
1791vmxnet3_rq_init(struct vmxnet3_rx_queue *rq,
1792		struct vmxnet3_adapter  *adapter)
1793{
1794	int i;
1795
1796	/* initialize buf_info */
1797	for (i = 0; i < rq->rx_ring[0].size; i++) {
1798
1799		/* 1st buf for a pkt is skbuff */
1800		if (i % adapter->rx_buf_per_pkt == 0) {
1801			rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_SKB;
1802			rq->buf_info[0][i].len = adapter->skb_buf_size;
1803		} else { /* subsequent bufs for a pkt is frag */
1804			rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_PAGE;
1805			rq->buf_info[0][i].len = PAGE_SIZE;
1806		}
1807	}
1808	for (i = 0; i < rq->rx_ring[1].size; i++) {
1809		rq->buf_info[1][i].buf_type = VMXNET3_RX_BUF_PAGE;
1810		rq->buf_info[1][i].len = PAGE_SIZE;
1811	}
1812
1813	/* reset internal state and allocate buffers for both rings */
1814	for (i = 0; i < 2; i++) {
1815		rq->rx_ring[i].next2fill = rq->rx_ring[i].next2comp = 0;
1816
1817		memset(rq->rx_ring[i].base, 0, rq->rx_ring[i].size *
1818		       sizeof(struct Vmxnet3_RxDesc));
1819		rq->rx_ring[i].gen = VMXNET3_INIT_GEN;
1820	}
1821	if (vmxnet3_rq_alloc_rx_buf(rq, 0, rq->rx_ring[0].size - 1,
1822				    adapter) == 0) {
1823		/* at least has 1 rx buffer for the 1st ring */
1824		return -ENOMEM;
1825	}
1826	vmxnet3_rq_alloc_rx_buf(rq, 1, rq->rx_ring[1].size - 1, adapter);
1827
1828	/* reset the comp ring */
1829	rq->comp_ring.next2proc = 0;
1830	memset(rq->comp_ring.base, 0, rq->comp_ring.size *
1831	       sizeof(struct Vmxnet3_RxCompDesc));
1832	rq->comp_ring.gen = VMXNET3_INIT_GEN;
1833
1834	/* reset rxctx */
1835	rq->rx_ctx.skb = NULL;
1836
1837	/* stats are not reset */
1838	return 0;
1839}
1840
1841
1842static int
1843vmxnet3_rq_init_all(struct vmxnet3_adapter *adapter)
1844{
1845	int i, err = 0;
1846
1847	for (i = 0; i < adapter->num_rx_queues; i++) {
1848		err = vmxnet3_rq_init(&adapter->rx_queue[i], adapter);
1849		if (unlikely(err)) {
1850			dev_err(&adapter->netdev->dev, "%s: failed to "
1851				"initialize rx queue%i\n",
1852				adapter->netdev->name, i);
1853			break;
1854		}
1855	}
1856	return err;
1857
1858}
1859
1860
1861static int
1862vmxnet3_rq_create(struct vmxnet3_rx_queue *rq, struct vmxnet3_adapter *adapter)
1863{
1864	int i;
1865	size_t sz;
1866	struct vmxnet3_rx_buf_info *bi;
1867
1868	for (i = 0; i < 2; i++) {
1869
1870		sz = rq->rx_ring[i].size * sizeof(struct Vmxnet3_RxDesc);
1871		rq->rx_ring[i].base = dma_alloc_coherent(
1872						&adapter->pdev->dev, sz,
1873						&rq->rx_ring[i].basePA,
1874						GFP_KERNEL);
1875		if (!rq->rx_ring[i].base) {
1876			netdev_err(adapter->netdev,
1877				   "failed to allocate rx ring %d\n", i);
1878			goto err;
1879		}
1880	}
1881
1882	if ((adapter->rxdataring_enabled) && (rq->data_ring.desc_size != 0)) {
1883		sz = rq->rx_ring[0].size * rq->data_ring.desc_size;
1884		rq->data_ring.base =
1885			dma_alloc_coherent(&adapter->pdev->dev, sz,
1886					   &rq->data_ring.basePA,
1887					   GFP_KERNEL);
1888		if (!rq->data_ring.base) {
1889			netdev_err(adapter->netdev,
1890				   "rx data ring will be disabled\n");
1891			adapter->rxdataring_enabled = false;
1892		}
1893	} else {
1894		rq->data_ring.base = NULL;
1895		rq->data_ring.desc_size = 0;
1896	}
1897
1898	sz = rq->comp_ring.size * sizeof(struct Vmxnet3_RxCompDesc);
1899	rq->comp_ring.base = dma_alloc_coherent(&adapter->pdev->dev, sz,
1900						&rq->comp_ring.basePA,
1901						GFP_KERNEL);
1902	if (!rq->comp_ring.base) {
1903		netdev_err(adapter->netdev, "failed to allocate rx comp ring\n");
1904		goto err;
1905	}
1906
1907	sz = sizeof(struct vmxnet3_rx_buf_info) * (rq->rx_ring[0].size +
1908						   rq->rx_ring[1].size);
1909	bi = dma_alloc_coherent(&adapter->pdev->dev, sz, &rq->buf_info_pa,
1910				GFP_KERNEL);
1911	if (!bi)
1912		goto err;
1913
1914	rq->buf_info[0] = bi;
1915	rq->buf_info[1] = bi + rq->rx_ring[0].size;
1916
1917	return 0;
1918
1919err:
1920	vmxnet3_rq_destroy(rq, adapter);
1921	return -ENOMEM;
1922}
1923
1924
1925static int
1926vmxnet3_rq_create_all(struct vmxnet3_adapter *adapter)
1927{
1928	int i, err = 0;
1929
1930	adapter->rxdataring_enabled = VMXNET3_VERSION_GE_3(adapter);
1931
1932	for (i = 0; i < adapter->num_rx_queues; i++) {
1933		err = vmxnet3_rq_create(&adapter->rx_queue[i], adapter);
1934		if (unlikely(err)) {
1935			dev_err(&adapter->netdev->dev,
1936				"%s: failed to create rx queue%i\n",
1937				adapter->netdev->name, i);
1938			goto err_out;
1939		}
1940	}
1941
1942	if (!adapter->rxdataring_enabled)
1943		vmxnet3_rq_destroy_all_rxdataring(adapter);
1944
1945	return err;
1946err_out:
1947	vmxnet3_rq_destroy_all(adapter);
1948	return err;
1949
1950}
1951
1952/* Multiple queue aware polling function for tx and rx */
1953
1954static int
1955vmxnet3_do_poll(struct vmxnet3_adapter *adapter, int budget)
1956{
1957	int rcd_done = 0, i;
1958	if (unlikely(adapter->shared->ecr))
1959		vmxnet3_process_events(adapter);
1960	for (i = 0; i < adapter->num_tx_queues; i++)
1961		vmxnet3_tq_tx_complete(&adapter->tx_queue[i], adapter);
1962
1963	for (i = 0; i < adapter->num_rx_queues; i++)
1964		rcd_done += vmxnet3_rq_rx_complete(&adapter->rx_queue[i],
1965						   adapter, budget);
1966	return rcd_done;
1967}
1968
1969
1970static int
1971vmxnet3_poll(struct napi_struct *napi, int budget)
1972{
1973	struct vmxnet3_rx_queue *rx_queue = container_of(napi,
1974					  struct vmxnet3_rx_queue, napi);
1975	int rxd_done;
1976
1977	rxd_done = vmxnet3_do_poll(rx_queue->adapter, budget);
1978
1979	if (rxd_done < budget) {
1980		napi_complete_done(napi, rxd_done);
1981		vmxnet3_enable_all_intrs(rx_queue->adapter);
1982	}
1983	return rxd_done;
1984}
1985
1986/*
1987 * NAPI polling function for MSI-X mode with multiple Rx queues
1988 * Returns the # of the NAPI credit consumed (# of rx descriptors processed)
1989 */
1990
1991static int
1992vmxnet3_poll_rx_only(struct napi_struct *napi, int budget)
1993{
1994	struct vmxnet3_rx_queue *rq = container_of(napi,
1995						struct vmxnet3_rx_queue, napi);
1996	struct vmxnet3_adapter *adapter = rq->adapter;
1997	int rxd_done;
1998
1999	/* When sharing interrupt with corresponding tx queue, process
2000	 * tx completions in that queue as well
2001	 */
2002	if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE) {
2003		struct vmxnet3_tx_queue *tq =
2004				&adapter->tx_queue[rq - adapter->rx_queue];
2005		vmxnet3_tq_tx_complete(tq, adapter);
2006	}
2007
2008	rxd_done = vmxnet3_rq_rx_complete(rq, adapter, budget);
2009
2010	if (rxd_done < budget) {
2011		napi_complete_done(napi, rxd_done);
2012		vmxnet3_enable_intr(adapter, rq->comp_ring.intr_idx);
2013	}
2014	return rxd_done;
2015}
2016
2017
2018#ifdef CONFIG_PCI_MSI
2019
2020/*
2021 * Handle completion interrupts on tx queues
2022 * Returns whether or not the intr is handled
2023 */
2024
2025static irqreturn_t
2026vmxnet3_msix_tx(int irq, void *data)
2027{
2028	struct vmxnet3_tx_queue *tq = data;
2029	struct vmxnet3_adapter *adapter = tq->adapter;
2030
2031	if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
2032		vmxnet3_disable_intr(adapter, tq->comp_ring.intr_idx);
2033
2034	/* Handle the case where only one irq is allocate for all tx queues */
2035	if (adapter->share_intr == VMXNET3_INTR_TXSHARE) {
2036		int i;
2037		for (i = 0; i < adapter->num_tx_queues; i++) {
2038			struct vmxnet3_tx_queue *txq = &adapter->tx_queue[i];
2039			vmxnet3_tq_tx_complete(txq, adapter);
2040		}
2041	} else {
2042		vmxnet3_tq_tx_complete(tq, adapter);
2043	}
2044	vmxnet3_enable_intr(adapter, tq->comp_ring.intr_idx);
2045
2046	return IRQ_HANDLED;
2047}
2048
2049
2050/*
2051 * Handle completion interrupts on rx queues. Returns whether or not the
2052 * intr is handled
2053 */
2054
2055static irqreturn_t
2056vmxnet3_msix_rx(int irq, void *data)
2057{
2058	struct vmxnet3_rx_queue *rq = data;
2059	struct vmxnet3_adapter *adapter = rq->adapter;
2060
2061	/* disable intr if needed */
2062	if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
2063		vmxnet3_disable_intr(adapter, rq->comp_ring.intr_idx);
2064	napi_schedule(&rq->napi);
2065
2066	return IRQ_HANDLED;
2067}
2068
2069/*
2070 *----------------------------------------------------------------------------
2071 *
2072 * vmxnet3_msix_event --
2073 *
2074 *    vmxnet3 msix event intr handler
2075 *
2076 * Result:
2077 *    whether or not the intr is handled
2078 *
2079 *----------------------------------------------------------------------------
2080 */
2081
2082static irqreturn_t
2083vmxnet3_msix_event(int irq, void *data)
2084{
2085	struct net_device *dev = data;
2086	struct vmxnet3_adapter *adapter = netdev_priv(dev);
2087
2088	/* disable intr if needed */
2089	if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
2090		vmxnet3_disable_intr(adapter, adapter->intr.event_intr_idx);
2091
2092	if (adapter->shared->ecr)
2093		vmxnet3_process_events(adapter);
2094
2095	vmxnet3_enable_intr(adapter, adapter->intr.event_intr_idx);
2096
2097	return IRQ_HANDLED;
2098}
2099
2100#endif /* CONFIG_PCI_MSI  */
2101
2102
2103/* Interrupt handler for vmxnet3  */
2104static irqreturn_t
2105vmxnet3_intr(int irq, void *dev_id)
2106{
2107	struct net_device *dev = dev_id;
2108	struct vmxnet3_adapter *adapter = netdev_priv(dev);
2109
2110	if (adapter->intr.type == VMXNET3_IT_INTX) {
2111		u32 icr = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_ICR);
2112		if (unlikely(icr == 0))
2113			/* not ours */
2114			return IRQ_NONE;
2115	}
2116
2117
2118	/* disable intr if needed */
2119	if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
2120		vmxnet3_disable_all_intrs(adapter);
2121
2122	napi_schedule(&adapter->rx_queue[0].napi);
2123
2124	return IRQ_HANDLED;
2125}
2126
2127#ifdef CONFIG_NET_POLL_CONTROLLER
2128
2129/* netpoll callback. */
2130static void
2131vmxnet3_netpoll(struct net_device *netdev)
2132{
2133	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2134
2135	switch (adapter->intr.type) {
2136#ifdef CONFIG_PCI_MSI
2137	case VMXNET3_IT_MSIX: {
2138		int i;
2139		for (i = 0; i < adapter->num_rx_queues; i++)
2140			vmxnet3_msix_rx(0, &adapter->rx_queue[i]);
2141		break;
2142	}
2143#endif
2144	case VMXNET3_IT_MSI:
2145	default:
2146		vmxnet3_intr(0, adapter->netdev);
2147		break;
2148	}
2149
2150}
2151#endif	/* CONFIG_NET_POLL_CONTROLLER */
2152
2153static int
2154vmxnet3_request_irqs(struct vmxnet3_adapter *adapter)
2155{
2156	struct vmxnet3_intr *intr = &adapter->intr;
2157	int err = 0, i;
2158	int vector = 0;
2159
2160#ifdef CONFIG_PCI_MSI
2161	if (adapter->intr.type == VMXNET3_IT_MSIX) {
2162		for (i = 0; i < adapter->num_tx_queues; i++) {
2163			if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE) {
2164				sprintf(adapter->tx_queue[i].name, "%s-tx-%d",
2165					adapter->netdev->name, vector);
2166				err = request_irq(
2167					      intr->msix_entries[vector].vector,
2168					      vmxnet3_msix_tx, 0,
2169					      adapter->tx_queue[i].name,
2170					      &adapter->tx_queue[i]);
2171			} else {
2172				sprintf(adapter->tx_queue[i].name, "%s-rxtx-%d",
2173					adapter->netdev->name, vector);
2174			}
2175			if (err) {
2176				dev_err(&adapter->netdev->dev,
2177					"Failed to request irq for MSIX, %s, "
2178					"error %d\n",
2179					adapter->tx_queue[i].name, err);
2180				return err;
2181			}
2182
2183			/* Handle the case where only 1 MSIx was allocated for
2184			 * all tx queues */
2185			if (adapter->share_intr == VMXNET3_INTR_TXSHARE) {
2186				for (; i < adapter->num_tx_queues; i++)
2187					adapter->tx_queue[i].comp_ring.intr_idx
2188								= vector;
2189				vector++;
2190				break;
2191			} else {
2192				adapter->tx_queue[i].comp_ring.intr_idx
2193								= vector++;
2194			}
2195		}
2196		if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE)
2197			vector = 0;
2198
2199		for (i = 0; i < adapter->num_rx_queues; i++) {
2200			if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE)
2201				sprintf(adapter->rx_queue[i].name, "%s-rx-%d",
2202					adapter->netdev->name, vector);
2203			else
2204				sprintf(adapter->rx_queue[i].name, "%s-rxtx-%d",
2205					adapter->netdev->name, vector);
2206			err = request_irq(intr->msix_entries[vector].vector,
2207					  vmxnet3_msix_rx, 0,
2208					  adapter->rx_queue[i].name,
2209					  &(adapter->rx_queue[i]));
2210			if (err) {
2211				netdev_err(adapter->netdev,
2212					   "Failed to request irq for MSIX, "
2213					   "%s, error %d\n",
2214					   adapter->rx_queue[i].name, err);
2215				return err;
2216			}
2217
2218			adapter->rx_queue[i].comp_ring.intr_idx = vector++;
2219		}
2220
2221		sprintf(intr->event_msi_vector_name, "%s-event-%d",
2222			adapter->netdev->name, vector);
2223		err = request_irq(intr->msix_entries[vector].vector,
2224				  vmxnet3_msix_event, 0,
2225				  intr->event_msi_vector_name, adapter->netdev);
2226		intr->event_intr_idx = vector;
2227
2228	} else if (intr->type == VMXNET3_IT_MSI) {
2229		adapter->num_rx_queues = 1;
2230		err = request_irq(adapter->pdev->irq, vmxnet3_intr, 0,
2231				  adapter->netdev->name, adapter->netdev);
2232	} else {
2233#endif
2234		adapter->num_rx_queues = 1;
2235		err = request_irq(adapter->pdev->irq, vmxnet3_intr,
2236				  IRQF_SHARED, adapter->netdev->name,
2237				  adapter->netdev);
2238#ifdef CONFIG_PCI_MSI
2239	}
2240#endif
2241	intr->num_intrs = vector + 1;
2242	if (err) {
2243		netdev_err(adapter->netdev,
2244			   "Failed to request irq (intr type:%d), error %d\n",
2245			   intr->type, err);
2246	} else {
2247		/* Number of rx queues will not change after this */
2248		for (i = 0; i < adapter->num_rx_queues; i++) {
2249			struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
2250			rq->qid = i;
2251			rq->qid2 = i + adapter->num_rx_queues;
2252			rq->dataRingQid = i + 2 * adapter->num_rx_queues;
2253		}
2254
2255		/* init our intr settings */
2256		for (i = 0; i < intr->num_intrs; i++)
2257			intr->mod_levels[i] = UPT1_IML_ADAPTIVE;
2258		if (adapter->intr.type != VMXNET3_IT_MSIX) {
2259			adapter->intr.event_intr_idx = 0;
2260			for (i = 0; i < adapter->num_tx_queues; i++)
2261				adapter->tx_queue[i].comp_ring.intr_idx = 0;
2262			adapter->rx_queue[0].comp_ring.intr_idx = 0;
2263		}
2264
2265		netdev_info(adapter->netdev,
2266			    "intr type %u, mode %u, %u vectors allocated\n",
2267			    intr->type, intr->mask_mode, intr->num_intrs);
2268	}
2269
2270	return err;
2271}
2272
2273
2274static void
2275vmxnet3_free_irqs(struct vmxnet3_adapter *adapter)
2276{
2277	struct vmxnet3_intr *intr = &adapter->intr;
2278	BUG_ON(intr->type == VMXNET3_IT_AUTO || intr->num_intrs <= 0);
2279
2280	switch (intr->type) {
2281#ifdef CONFIG_PCI_MSI
2282	case VMXNET3_IT_MSIX:
2283	{
2284		int i, vector = 0;
2285
2286		if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE) {
2287			for (i = 0; i < adapter->num_tx_queues; i++) {
2288				free_irq(intr->msix_entries[vector++].vector,
2289					 &(adapter->tx_queue[i]));
2290				if (adapter->share_intr == VMXNET3_INTR_TXSHARE)
2291					break;
2292			}
2293		}
2294
2295		for (i = 0; i < adapter->num_rx_queues; i++) {
2296			free_irq(intr->msix_entries[vector++].vector,
2297				 &(adapter->rx_queue[i]));
2298		}
2299
2300		free_irq(intr->msix_entries[vector].vector,
2301			 adapter->netdev);
2302		BUG_ON(vector >= intr->num_intrs);
2303		break;
2304	}
2305#endif
2306	case VMXNET3_IT_MSI:
2307		free_irq(adapter->pdev->irq, adapter->netdev);
2308		break;
2309	case VMXNET3_IT_INTX:
2310		free_irq(adapter->pdev->irq, adapter->netdev);
2311		break;
2312	default:
2313		BUG();
2314	}
2315}
2316
2317
2318static void
2319vmxnet3_restore_vlan(struct vmxnet3_adapter *adapter)
2320{
2321	u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
2322	u16 vid;
2323
2324	/* allow untagged pkts */
2325	VMXNET3_SET_VFTABLE_ENTRY(vfTable, 0);
2326
2327	for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
2328		VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid);
2329}
2330
2331
2332static int
2333vmxnet3_vlan_rx_add_vid(struct net_device *netdev, __be16 proto, u16 vid)
2334{
2335	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2336
2337	if (!(netdev->flags & IFF_PROMISC)) {
2338		u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
2339		unsigned long flags;
2340
2341		VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid);
2342		spin_lock_irqsave(&adapter->cmd_lock, flags);
2343		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2344				       VMXNET3_CMD_UPDATE_VLAN_FILTERS);
2345		spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2346	}
2347
2348	set_bit(vid, adapter->active_vlans);
2349
2350	return 0;
2351}
2352
2353
2354static int
2355vmxnet3_vlan_rx_kill_vid(struct net_device *netdev, __be16 proto, u16 vid)
2356{
2357	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2358
2359	if (!(netdev->flags & IFF_PROMISC)) {
2360		u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
2361		unsigned long flags;
2362
2363		VMXNET3_CLEAR_VFTABLE_ENTRY(vfTable, vid);
2364		spin_lock_irqsave(&adapter->cmd_lock, flags);
2365		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2366				       VMXNET3_CMD_UPDATE_VLAN_FILTERS);
2367		spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2368	}
2369
2370	clear_bit(vid, adapter->active_vlans);
2371
2372	return 0;
2373}
2374
2375
2376static u8 *
2377vmxnet3_copy_mc(struct net_device *netdev)
2378{
2379	u8 *buf = NULL;
2380	u32 sz = netdev_mc_count(netdev) * ETH_ALEN;
2381
2382	/* struct Vmxnet3_RxFilterConf.mfTableLen is u16. */
2383	if (sz <= 0xffff) {
2384		/* We may be called with BH disabled */
2385		buf = kmalloc(sz, GFP_ATOMIC);
2386		if (buf) {
2387			struct netdev_hw_addr *ha;
2388			int i = 0;
2389
2390			netdev_for_each_mc_addr(ha, netdev)
2391				memcpy(buf + i++ * ETH_ALEN, ha->addr,
2392				       ETH_ALEN);
2393		}
2394	}
2395	return buf;
2396}
2397
2398
2399static void
2400vmxnet3_set_mc(struct net_device *netdev)
2401{
2402	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2403	unsigned long flags;
2404	struct Vmxnet3_RxFilterConf *rxConf =
2405					&adapter->shared->devRead.rxFilterConf;
2406	u8 *new_table = NULL;
2407	dma_addr_t new_table_pa = 0;
2408	bool new_table_pa_valid = false;
2409	u32 new_mode = VMXNET3_RXM_UCAST;
2410
2411	if (netdev->flags & IFF_PROMISC) {
2412		u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
2413		memset(vfTable, 0, VMXNET3_VFT_SIZE * sizeof(*vfTable));
2414
2415		new_mode |= VMXNET3_RXM_PROMISC;
2416	} else {
2417		vmxnet3_restore_vlan(adapter);
2418	}
2419
2420	if (netdev->flags & IFF_BROADCAST)
2421		new_mode |= VMXNET3_RXM_BCAST;
2422
2423	if (netdev->flags & IFF_ALLMULTI)
2424		new_mode |= VMXNET3_RXM_ALL_MULTI;
2425	else
2426		if (!netdev_mc_empty(netdev)) {
2427			new_table = vmxnet3_copy_mc(netdev);
2428			if (new_table) {
2429				size_t sz = netdev_mc_count(netdev) * ETH_ALEN;
2430
2431				rxConf->mfTableLen = cpu_to_le16(sz);
2432				new_table_pa = dma_map_single(
2433							&adapter->pdev->dev,
2434							new_table,
2435							sz,
2436							PCI_DMA_TODEVICE);
2437				if (!dma_mapping_error(&adapter->pdev->dev,
2438						       new_table_pa)) {
2439					new_mode |= VMXNET3_RXM_MCAST;
2440					new_table_pa_valid = true;
2441					rxConf->mfTablePA = cpu_to_le64(
2442								new_table_pa);
2443				}
2444			}
2445			if (!new_table_pa_valid) {
2446				netdev_info(netdev,
2447					    "failed to copy mcast list, setting ALL_MULTI\n");
2448				new_mode |= VMXNET3_RXM_ALL_MULTI;
2449			}
2450		}
2451
2452	if (!(new_mode & VMXNET3_RXM_MCAST)) {
2453		rxConf->mfTableLen = 0;
2454		rxConf->mfTablePA = 0;
2455	}
2456
2457	spin_lock_irqsave(&adapter->cmd_lock, flags);
2458	if (new_mode != rxConf->rxMode) {
2459		rxConf->rxMode = cpu_to_le32(new_mode);
2460		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2461				       VMXNET3_CMD_UPDATE_RX_MODE);
2462		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2463				       VMXNET3_CMD_UPDATE_VLAN_FILTERS);
2464	}
2465
2466	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2467			       VMXNET3_CMD_UPDATE_MAC_FILTERS);
2468	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2469
2470	if (new_table_pa_valid)
2471		dma_unmap_single(&adapter->pdev->dev, new_table_pa,
2472				 rxConf->mfTableLen, PCI_DMA_TODEVICE);
2473	kfree(new_table);
2474}
2475
2476void
2477vmxnet3_rq_destroy_all(struct vmxnet3_adapter *adapter)
2478{
2479	int i;
2480
2481	for (i = 0; i < adapter->num_rx_queues; i++)
2482		vmxnet3_rq_destroy(&adapter->rx_queue[i], adapter);
2483}
2484
2485
2486/*
2487 *   Set up driver_shared based on settings in adapter.
2488 */
2489
2490static void
2491vmxnet3_setup_driver_shared(struct vmxnet3_adapter *adapter)
2492{
2493	struct Vmxnet3_DriverShared *shared = adapter->shared;
2494	struct Vmxnet3_DSDevRead *devRead = &shared->devRead;
2495	struct Vmxnet3_TxQueueConf *tqc;
2496	struct Vmxnet3_RxQueueConf *rqc;
2497	int i;
2498
2499	memset(shared, 0, sizeof(*shared));
2500
2501	/* driver settings */
2502	shared->magic = cpu_to_le32(VMXNET3_REV1_MAGIC);
2503	devRead->misc.driverInfo.version = cpu_to_le32(
2504						VMXNET3_DRIVER_VERSION_NUM);
2505	devRead->misc.driverInfo.gos.gosBits = (sizeof(void *) == 4 ?
2506				VMXNET3_GOS_BITS_32 : VMXNET3_GOS_BITS_64);
2507	devRead->misc.driverInfo.gos.gosType = VMXNET3_GOS_TYPE_LINUX;
2508	*((u32 *)&devRead->misc.driverInfo.gos) = cpu_to_le32(
2509				*((u32 *)&devRead->misc.driverInfo.gos));
2510	devRead->misc.driverInfo.vmxnet3RevSpt = cpu_to_le32(1);
2511	devRead->misc.driverInfo.uptVerSpt = cpu_to_le32(1);
2512
2513	devRead->misc.ddPA = cpu_to_le64(adapter->adapter_pa);
2514	devRead->misc.ddLen = cpu_to_le32(sizeof(struct vmxnet3_adapter));
2515
2516	/* set up feature flags */
2517	if (adapter->netdev->features & NETIF_F_RXCSUM)
2518		devRead->misc.uptFeatures |= UPT1_F_RXCSUM;
2519
2520	if (adapter->netdev->features & NETIF_F_LRO) {
2521		devRead->misc.uptFeatures |= UPT1_F_LRO;
2522		devRead->misc.maxNumRxSG = cpu_to_le16(1 + MAX_SKB_FRAGS);
2523	}
2524	if (adapter->netdev->features & NETIF_F_HW_VLAN_CTAG_RX)
2525		devRead->misc.uptFeatures |= UPT1_F_RXVLAN;
2526
2527	if (adapter->netdev->features & (NETIF_F_GSO_UDP_TUNNEL |
2528					 NETIF_F_GSO_UDP_TUNNEL_CSUM))
2529		devRead->misc.uptFeatures |= UPT1_F_RXINNEROFLD;
2530
2531	devRead->misc.mtu = cpu_to_le32(adapter->netdev->mtu);
2532	devRead->misc.queueDescPA = cpu_to_le64(adapter->queue_desc_pa);
2533	devRead->misc.queueDescLen = cpu_to_le32(
2534		adapter->num_tx_queues * sizeof(struct Vmxnet3_TxQueueDesc) +
2535		adapter->num_rx_queues * sizeof(struct Vmxnet3_RxQueueDesc));
2536
2537	/* tx queue settings */
2538	devRead->misc.numTxQueues =  adapter->num_tx_queues;
2539	for (i = 0; i < adapter->num_tx_queues; i++) {
2540		struct vmxnet3_tx_queue	*tq = &adapter->tx_queue[i];
2541		BUG_ON(adapter->tx_queue[i].tx_ring.base == NULL);
2542		tqc = &adapter->tqd_start[i].conf;
2543		tqc->txRingBasePA   = cpu_to_le64(tq->tx_ring.basePA);
2544		tqc->dataRingBasePA = cpu_to_le64(tq->data_ring.basePA);
2545		tqc->compRingBasePA = cpu_to_le64(tq->comp_ring.basePA);
2546		tqc->ddPA           = cpu_to_le64(tq->buf_info_pa);
2547		tqc->txRingSize     = cpu_to_le32(tq->tx_ring.size);
2548		tqc->dataRingSize   = cpu_to_le32(tq->data_ring.size);
2549		tqc->txDataRingDescSize = cpu_to_le32(tq->txdata_desc_size);
2550		tqc->compRingSize   = cpu_to_le32(tq->comp_ring.size);
2551		tqc->ddLen          = cpu_to_le32(
2552					sizeof(struct vmxnet3_tx_buf_info) *
2553					tqc->txRingSize);
2554		tqc->intrIdx        = tq->comp_ring.intr_idx;
2555	}
2556
2557	/* rx queue settings */
2558	devRead->misc.numRxQueues = adapter->num_rx_queues;
2559	for (i = 0; i < adapter->num_rx_queues; i++) {
2560		struct vmxnet3_rx_queue	*rq = &adapter->rx_queue[i];
2561		rqc = &adapter->rqd_start[i].conf;
2562		rqc->rxRingBasePA[0] = cpu_to_le64(rq->rx_ring[0].basePA);
2563		rqc->rxRingBasePA[1] = cpu_to_le64(rq->rx_ring[1].basePA);
2564		rqc->compRingBasePA  = cpu_to_le64(rq->comp_ring.basePA);
2565		rqc->ddPA            = cpu_to_le64(rq->buf_info_pa);
2566		rqc->rxRingSize[0]   = cpu_to_le32(rq->rx_ring[0].size);
2567		rqc->rxRingSize[1]   = cpu_to_le32(rq->rx_ring[1].size);
2568		rqc->compRingSize    = cpu_to_le32(rq->comp_ring.size);
2569		rqc->ddLen           = cpu_to_le32(
2570					sizeof(struct vmxnet3_rx_buf_info) *
2571					(rqc->rxRingSize[0] +
2572					 rqc->rxRingSize[1]));
2573		rqc->intrIdx         = rq->comp_ring.intr_idx;
2574		if (VMXNET3_VERSION_GE_3(adapter)) {
2575			rqc->rxDataRingBasePA =
2576				cpu_to_le64(rq->data_ring.basePA);
2577			rqc->rxDataRingDescSize =
2578				cpu_to_le16(rq->data_ring.desc_size);
2579		}
2580	}
2581
2582#ifdef VMXNET3_RSS
2583	memset(adapter->rss_conf, 0, sizeof(*adapter->rss_conf));
2584
2585	if (adapter->rss) {
2586		struct UPT1_RSSConf *rssConf = adapter->rss_conf;
2587
2588		devRead->misc.uptFeatures |= UPT1_F_RSS;
2589		devRead->misc.numRxQueues = adapter->num_rx_queues;
2590		rssConf->hashType = UPT1_RSS_HASH_TYPE_TCP_IPV4 |
2591				    UPT1_RSS_HASH_TYPE_IPV4 |
2592				    UPT1_RSS_HASH_TYPE_TCP_IPV6 |
2593				    UPT1_RSS_HASH_TYPE_IPV6;
2594		rssConf->hashFunc = UPT1_RSS_HASH_FUNC_TOEPLITZ;
2595		rssConf->hashKeySize = UPT1_RSS_MAX_KEY_SIZE;
2596		rssConf->indTableSize = VMXNET3_RSS_IND_TABLE_SIZE;
2597		netdev_rss_key_fill(rssConf->hashKey, sizeof(rssConf->hashKey));
2598
2599		for (i = 0; i < rssConf->indTableSize; i++)
2600			rssConf->indTable[i] = ethtool_rxfh_indir_default(
2601				i, adapter->num_rx_queues);
2602
2603		devRead->rssConfDesc.confVer = 1;
2604		devRead->rssConfDesc.confLen = cpu_to_le32(sizeof(*rssConf));
2605		devRead->rssConfDesc.confPA =
2606			cpu_to_le64(adapter->rss_conf_pa);
2607	}
2608
2609#endif /* VMXNET3_RSS */
2610
2611	/* intr settings */
2612	devRead->intrConf.autoMask = adapter->intr.mask_mode ==
2613				     VMXNET3_IMM_AUTO;
2614	devRead->intrConf.numIntrs = adapter->intr.num_intrs;
2615	for (i = 0; i < adapter->intr.num_intrs; i++)
2616		devRead->intrConf.modLevels[i] = adapter->intr.mod_levels[i];
2617
2618	devRead->intrConf.eventIntrIdx = adapter->intr.event_intr_idx;
2619	devRead->intrConf.intrCtrl |= cpu_to_le32(VMXNET3_IC_DISABLE_ALL);
2620
2621	/* rx filter settings */
2622	devRead->rxFilterConf.rxMode = 0;
2623	vmxnet3_restore_vlan(adapter);
2624	vmxnet3_write_mac_addr(adapter, adapter->netdev->dev_addr);
2625
2626	/* the rest are already zeroed */
2627}
2628
2629static void
2630vmxnet3_init_coalesce(struct vmxnet3_adapter *adapter)
2631{
2632	struct Vmxnet3_DriverShared *shared = adapter->shared;
2633	union Vmxnet3_CmdInfo *cmdInfo = &shared->cu.cmdInfo;
2634	unsigned long flags;
2635
2636	if (!VMXNET3_VERSION_GE_3(adapter))
2637		return;
2638
2639	spin_lock_irqsave(&adapter->cmd_lock, flags);
2640	cmdInfo->varConf.confVer = 1;
2641	cmdInfo->varConf.confLen =
2642		cpu_to_le32(sizeof(*adapter->coal_conf));
2643	cmdInfo->varConf.confPA  = cpu_to_le64(adapter->coal_conf_pa);
2644
2645	if (adapter->default_coal_mode) {
2646		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2647				       VMXNET3_CMD_GET_COALESCE);
2648	} else {
2649		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2650				       VMXNET3_CMD_SET_COALESCE);
2651	}
2652
2653	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2654}
2655
2656static void
2657vmxnet3_init_rssfields(struct vmxnet3_adapter *adapter)
2658{
2659	struct Vmxnet3_DriverShared *shared = adapter->shared;
2660	union Vmxnet3_CmdInfo *cmdInfo = &shared->cu.cmdInfo;
2661	unsigned long flags;
2662
2663	if (!VMXNET3_VERSION_GE_4(adapter))
2664		return;
2665
2666	spin_lock_irqsave(&adapter->cmd_lock, flags);
2667
2668	if (adapter->default_rss_fields) {
2669		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2670				       VMXNET3_CMD_GET_RSS_FIELDS);
2671		adapter->rss_fields =
2672			VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
2673	} else {
2674		cmdInfo->setRssFields = adapter->rss_fields;
2675		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2676				       VMXNET3_CMD_SET_RSS_FIELDS);
2677		/* Not all requested RSS may get applied, so get and
2678		 * cache what was actually applied.
2679		 */
2680		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2681				       VMXNET3_CMD_GET_RSS_FIELDS);
2682		adapter->rss_fields =
2683			VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
2684	}
2685
2686	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2687}
2688
2689int
2690vmxnet3_activate_dev(struct vmxnet3_adapter *adapter)
2691{
2692	int err, i;
2693	u32 ret;
2694	unsigned long flags;
2695
2696	netdev_dbg(adapter->netdev, "%s: skb_buf_size %d, rx_buf_per_pkt %d,"
2697		" ring sizes %u %u %u\n", adapter->netdev->name,
2698		adapter->skb_buf_size, adapter->rx_buf_per_pkt,
2699		adapter->tx_queue[0].tx_ring.size,
2700		adapter->rx_queue[0].rx_ring[0].size,
2701		adapter->rx_queue[0].rx_ring[1].size);
2702
2703	vmxnet3_tq_init_all(adapter);
2704	err = vmxnet3_rq_init_all(adapter);
2705	if (err) {
2706		netdev_err(adapter->netdev,
2707			   "Failed to init rx queue error %d\n", err);
2708		goto rq_err;
2709	}
2710
2711	err = vmxnet3_request_irqs(adapter);
2712	if (err) {
2713		netdev_err(adapter->netdev,
2714			   "Failed to setup irq for error %d\n", err);
2715		goto irq_err;
2716	}
2717
2718	vmxnet3_setup_driver_shared(adapter);
2719
2720	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, VMXNET3_GET_ADDR_LO(
2721			       adapter->shared_pa));
2722	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, VMXNET3_GET_ADDR_HI(
2723			       adapter->shared_pa));
2724	spin_lock_irqsave(&adapter->cmd_lock, flags);
2725	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2726			       VMXNET3_CMD_ACTIVATE_DEV);
2727	ret = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
2728	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2729
2730	if (ret != 0) {
2731		netdev_err(adapter->netdev,
2732			   "Failed to activate dev: error %u\n", ret);
2733		err = -EINVAL;
2734		goto activate_err;
2735	}
2736
2737	vmxnet3_init_coalesce(adapter);
2738	vmxnet3_init_rssfields(adapter);
2739
2740	for (i = 0; i < adapter->num_rx_queues; i++) {
2741		VMXNET3_WRITE_BAR0_REG(adapter,
2742				VMXNET3_REG_RXPROD + i * VMXNET3_REG_ALIGN,
2743				adapter->rx_queue[i].rx_ring[0].next2fill);
2744		VMXNET3_WRITE_BAR0_REG(adapter, (VMXNET3_REG_RXPROD2 +
2745				(i * VMXNET3_REG_ALIGN)),
2746				adapter->rx_queue[i].rx_ring[1].next2fill);
2747	}
2748
2749	/* Apply the rx filter settins last. */
2750	vmxnet3_set_mc(adapter->netdev);
2751
2752	/*
2753	 * Check link state when first activating device. It will start the
2754	 * tx queue if the link is up.
2755	 */
2756	vmxnet3_check_link(adapter, true);
2757	for (i = 0; i < adapter->num_rx_queues; i++)
2758		napi_enable(&adapter->rx_queue[i].napi);
2759	vmxnet3_enable_all_intrs(adapter);
2760	clear_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
2761	return 0;
2762
2763activate_err:
2764	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, 0);
2765	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, 0);
2766	vmxnet3_free_irqs(adapter);
2767irq_err:
2768rq_err:
2769	/* free up buffers we allocated */
2770	vmxnet3_rq_cleanup_all(adapter);
2771	return err;
2772}
2773
2774
2775void
2776vmxnet3_reset_dev(struct vmxnet3_adapter *adapter)
2777{
2778	unsigned long flags;
2779	spin_lock_irqsave(&adapter->cmd_lock, flags);
2780	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, VMXNET3_CMD_RESET_DEV);
2781	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2782}
2783
2784
2785int
2786vmxnet3_quiesce_dev(struct vmxnet3_adapter *adapter)
2787{
2788	int i;
2789	unsigned long flags;
2790	if (test_and_set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state))
2791		return 0;
2792
2793
2794	spin_lock_irqsave(&adapter->cmd_lock, flags);
2795	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2796			       VMXNET3_CMD_QUIESCE_DEV);
2797	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2798	vmxnet3_disable_all_intrs(adapter);
2799
2800	for (i = 0; i < adapter->num_rx_queues; i++)
2801		napi_disable(&adapter->rx_queue[i].napi);
2802	netif_tx_disable(adapter->netdev);
2803	adapter->link_speed = 0;
2804	netif_carrier_off(adapter->netdev);
2805
2806	vmxnet3_tq_cleanup_all(adapter);
2807	vmxnet3_rq_cleanup_all(adapter);
2808	vmxnet3_free_irqs(adapter);
2809	return 0;
2810}
2811
2812
2813static void
2814vmxnet3_write_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac)
2815{
2816	u32 tmp;
2817
2818	tmp = *(u32 *)mac;
2819	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACL, tmp);
2820
2821	tmp = (mac[5] << 8) | mac[4];
2822	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACH, tmp);
2823}
2824
2825
2826static int
2827vmxnet3_set_mac_addr(struct net_device *netdev, void *p)
2828{
2829	struct sockaddr *addr = p;
2830	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2831
2832	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2833	vmxnet3_write_mac_addr(adapter, addr->sa_data);
2834
2835	return 0;
2836}
2837
2838
2839/* ==================== initialization and cleanup routines ============ */
2840
2841static int
2842vmxnet3_alloc_pci_resources(struct vmxnet3_adapter *adapter)
2843{
2844	int err;
2845	unsigned long mmio_start, mmio_len;
2846	struct pci_dev *pdev = adapter->pdev;
2847
2848	err = pci_enable_device(pdev);
2849	if (err) {
2850		dev_err(&pdev->dev, "Failed to enable adapter: error %d\n", err);
2851		return err;
2852	}
2853
2854	err = pci_request_selected_regions(pdev, (1 << 2) - 1,
2855					   vmxnet3_driver_name);
2856	if (err) {
2857		dev_err(&pdev->dev,
2858			"Failed to request region for adapter: error %d\n", err);
2859		goto err_enable_device;
2860	}
2861
2862	pci_set_master(pdev);
2863
2864	mmio_start = pci_resource_start(pdev, 0);
2865	mmio_len = pci_resource_len(pdev, 0);
2866	adapter->hw_addr0 = ioremap(mmio_start, mmio_len);
2867	if (!adapter->hw_addr0) {
2868		dev_err(&pdev->dev, "Failed to map bar0\n");
2869		err = -EIO;
2870		goto err_ioremap;
2871	}
2872
2873	mmio_start = pci_resource_start(pdev, 1);
2874	mmio_len = pci_resource_len(pdev, 1);
2875	adapter->hw_addr1 = ioremap(mmio_start, mmio_len);
2876	if (!adapter->hw_addr1) {
2877		dev_err(&pdev->dev, "Failed to map bar1\n");
2878		err = -EIO;
2879		goto err_bar1;
2880	}
2881	return 0;
2882
2883err_bar1:
2884	iounmap(adapter->hw_addr0);
2885err_ioremap:
2886	pci_release_selected_regions(pdev, (1 << 2) - 1);
2887err_enable_device:
2888	pci_disable_device(pdev);
2889	return err;
2890}
2891
2892
2893static void
2894vmxnet3_free_pci_resources(struct vmxnet3_adapter *adapter)
2895{
2896	BUG_ON(!adapter->pdev);
2897
2898	iounmap(adapter->hw_addr0);
2899	iounmap(adapter->hw_addr1);
2900	pci_release_selected_regions(adapter->pdev, (1 << 2) - 1);
2901	pci_disable_device(adapter->pdev);
2902}
2903
2904
2905static void
2906vmxnet3_adjust_rx_ring_size(struct vmxnet3_adapter *adapter)
2907{
2908	size_t sz, i, ring0_size, ring1_size, comp_size;
2909	if (adapter->netdev->mtu <= VMXNET3_MAX_SKB_BUF_SIZE -
2910				    VMXNET3_MAX_ETH_HDR_SIZE) {
2911		adapter->skb_buf_size = adapter->netdev->mtu +
2912					VMXNET3_MAX_ETH_HDR_SIZE;
2913		if (adapter->skb_buf_size < VMXNET3_MIN_T0_BUF_SIZE)
2914			adapter->skb_buf_size = VMXNET3_MIN_T0_BUF_SIZE;
2915
2916		adapter->rx_buf_per_pkt = 1;
2917	} else {
2918		adapter->skb_buf_size = VMXNET3_MAX_SKB_BUF_SIZE;
2919		sz = adapter->netdev->mtu - VMXNET3_MAX_SKB_BUF_SIZE +
2920					    VMXNET3_MAX_ETH_HDR_SIZE;
2921		adapter->rx_buf_per_pkt = 1 + (sz + PAGE_SIZE - 1) / PAGE_SIZE;
2922	}
2923
2924	/*
2925	 * for simplicity, force the ring0 size to be a multiple of
2926	 * rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN
2927	 */
2928	sz = adapter->rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN;
2929	ring0_size = adapter->rx_queue[0].rx_ring[0].size;
2930	ring0_size = (ring0_size + sz - 1) / sz * sz;
2931	ring0_size = min_t(u32, ring0_size, VMXNET3_RX_RING_MAX_SIZE /
2932			   sz * sz);
2933	ring1_size = adapter->rx_queue[0].rx_ring[1].size;
2934	ring1_size = (ring1_size + sz - 1) / sz * sz;
2935	ring1_size = min_t(u32, ring1_size, VMXNET3_RX_RING2_MAX_SIZE /
2936			   sz * sz);
2937	comp_size = ring0_size + ring1_size;
2938
2939	for (i = 0; i < adapter->num_rx_queues; i++) {
2940		struct vmxnet3_rx_queue	*rq = &adapter->rx_queue[i];
2941
2942		rq->rx_ring[0].size = ring0_size;
2943		rq->rx_ring[1].size = ring1_size;
2944		rq->comp_ring.size = comp_size;
2945	}
2946}
2947
2948
2949int
2950vmxnet3_create_queues(struct vmxnet3_adapter *adapter, u32 tx_ring_size,
2951		      u32 rx_ring_size, u32 rx_ring2_size,
2952		      u16 txdata_desc_size, u16 rxdata_desc_size)
2953{
2954	int err = 0, i;
2955
2956	for (i = 0; i < adapter->num_tx_queues; i++) {
2957		struct vmxnet3_tx_queue	*tq = &adapter->tx_queue[i];
2958		tq->tx_ring.size   = tx_ring_size;
2959		tq->data_ring.size = tx_ring_size;
2960		tq->comp_ring.size = tx_ring_size;
2961		tq->txdata_desc_size = txdata_desc_size;
2962		tq->shared = &adapter->tqd_start[i].ctrl;
2963		tq->stopped = true;
2964		tq->adapter = adapter;
2965		tq->qid = i;
2966		err = vmxnet3_tq_create(tq, adapter);
2967		/*
2968		 * Too late to change num_tx_queues. We cannot do away with
2969		 * lesser number of queues than what we asked for
2970		 */
2971		if (err)
2972			goto queue_err;
2973	}
2974
2975	adapter->rx_queue[0].rx_ring[0].size = rx_ring_size;
2976	adapter->rx_queue[0].rx_ring[1].size = rx_ring2_size;
2977	vmxnet3_adjust_rx_ring_size(adapter);
2978
2979	adapter->rxdataring_enabled = VMXNET3_VERSION_GE_3(adapter);
2980	for (i = 0; i < adapter->num_rx_queues; i++) {
2981		struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
2982		/* qid and qid2 for rx queues will be assigned later when num
2983		 * of rx queues is finalized after allocating intrs */
2984		rq->shared = &adapter->rqd_start[i].ctrl;
2985		rq->adapter = adapter;
2986		rq->data_ring.desc_size = rxdata_desc_size;
2987		err = vmxnet3_rq_create(rq, adapter);
2988		if (err) {
2989			if (i == 0) {
2990				netdev_err(adapter->netdev,
2991					   "Could not allocate any rx queues. "
2992					   "Aborting.\n");
2993				goto queue_err;
2994			} else {
2995				netdev_info(adapter->netdev,
2996					    "Number of rx queues changed "
2997					    "to : %d.\n", i);
2998				adapter->num_rx_queues = i;
2999				err = 0;
3000				break;
3001			}
3002		}
3003	}
3004
3005	if (!adapter->rxdataring_enabled)
3006		vmxnet3_rq_destroy_all_rxdataring(adapter);
3007
3008	return err;
3009queue_err:
3010	vmxnet3_tq_destroy_all(adapter);
3011	return err;
3012}
3013
3014static int
3015vmxnet3_open(struct net_device *netdev)
3016{
3017	struct vmxnet3_adapter *adapter;
3018	int err, i;
3019
3020	adapter = netdev_priv(netdev);
3021
3022	for (i = 0; i < adapter->num_tx_queues; i++)
3023		spin_lock_init(&adapter->tx_queue[i].tx_lock);
3024
3025	if (VMXNET3_VERSION_GE_3(adapter)) {
3026		unsigned long flags;
3027		u16 txdata_desc_size;
3028
3029		spin_lock_irqsave(&adapter->cmd_lock, flags);
3030		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3031				       VMXNET3_CMD_GET_TXDATA_DESC_SIZE);
3032		txdata_desc_size = VMXNET3_READ_BAR1_REG(adapter,
3033							 VMXNET3_REG_CMD);
3034		spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3035
3036		if ((txdata_desc_size < VMXNET3_TXDATA_DESC_MIN_SIZE) ||
3037		    (txdata_desc_size > VMXNET3_TXDATA_DESC_MAX_SIZE) ||
3038		    (txdata_desc_size & VMXNET3_TXDATA_DESC_SIZE_MASK)) {
3039			adapter->txdata_desc_size =
3040				sizeof(struct Vmxnet3_TxDataDesc);
3041		} else {
3042			adapter->txdata_desc_size = txdata_desc_size;
3043		}
3044	} else {
3045		adapter->txdata_desc_size = sizeof(struct Vmxnet3_TxDataDesc);
3046	}
3047
3048	err = vmxnet3_create_queues(adapter,
3049				    adapter->tx_ring_size,
3050				    adapter->rx_ring_size,
3051				    adapter->rx_ring2_size,
3052				    adapter->txdata_desc_size,
3053				    adapter->rxdata_desc_size);
3054	if (err)
3055		goto queue_err;
3056
3057	err = vmxnet3_activate_dev(adapter);
3058	if (err)
3059		goto activate_err;
3060
3061	return 0;
3062
3063activate_err:
3064	vmxnet3_rq_destroy_all(adapter);
3065	vmxnet3_tq_destroy_all(adapter);
3066queue_err:
3067	return err;
3068}
3069
3070
3071static int
3072vmxnet3_close(struct net_device *netdev)
3073{
3074	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3075
3076	/*
3077	 * Reset_work may be in the middle of resetting the device, wait for its
3078	 * completion.
3079	 */
3080	while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
3081		usleep_range(1000, 2000);
3082
3083	vmxnet3_quiesce_dev(adapter);
3084
3085	vmxnet3_rq_destroy_all(adapter);
3086	vmxnet3_tq_destroy_all(adapter);
3087
3088	clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
3089
3090
3091	return 0;
3092}
3093
3094
3095void
3096vmxnet3_force_close(struct vmxnet3_adapter *adapter)
3097{
3098	int i;
3099
3100	/*
3101	 * we must clear VMXNET3_STATE_BIT_RESETTING, otherwise
3102	 * vmxnet3_close() will deadlock.
3103	 */
3104	BUG_ON(test_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state));
3105
3106	/* we need to enable NAPI, otherwise dev_close will deadlock */
3107	for (i = 0; i < adapter->num_rx_queues; i++)
3108		napi_enable(&adapter->rx_queue[i].napi);
3109	/*
3110	 * Need to clear the quiesce bit to ensure that vmxnet3_close
3111	 * can quiesce the device properly
3112	 */
3113	clear_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
3114	dev_close(adapter->netdev);
3115}
3116
3117
3118static int
3119vmxnet3_change_mtu(struct net_device *netdev, int new_mtu)
3120{
3121	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3122	int err = 0;
3123
3124	netdev->mtu = new_mtu;
3125
3126	/*
3127	 * Reset_work may be in the middle of resetting the device, wait for its
3128	 * completion.
3129	 */
3130	while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
3131		usleep_range(1000, 2000);
3132
3133	if (netif_running(netdev)) {
3134		vmxnet3_quiesce_dev(adapter);
3135		vmxnet3_reset_dev(adapter);
3136
3137		/* we need to re-create the rx queue based on the new mtu */
3138		vmxnet3_rq_destroy_all(adapter);
3139		vmxnet3_adjust_rx_ring_size(adapter);
3140		err = vmxnet3_rq_create_all(adapter);
3141		if (err) {
3142			netdev_err(netdev,
3143				   "failed to re-create rx queues, "
3144				   " error %d. Closing it.\n", err);
3145			goto out;
3146		}
3147
3148		err = vmxnet3_activate_dev(adapter);
3149		if (err) {
3150			netdev_err(netdev,
3151				   "failed to re-activate, error %d. "
3152				   "Closing it\n", err);
3153			goto out;
3154		}
3155	}
3156
3157out:
3158	clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
3159	if (err)
3160		vmxnet3_force_close(adapter);
3161
3162	return err;
3163}
3164
3165
3166static void
3167vmxnet3_declare_features(struct vmxnet3_adapter *adapter, bool dma64)
3168{
3169	struct net_device *netdev = adapter->netdev;
3170
3171	netdev->hw_features = NETIF_F_SG | NETIF_F_RXCSUM |
3172		NETIF_F_HW_CSUM | NETIF_F_HW_VLAN_CTAG_TX |
3173		NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_TSO | NETIF_F_TSO6 |
3174		NETIF_F_LRO;
3175
3176	if (VMXNET3_VERSION_GE_4(adapter)) {
3177		netdev->hw_features |= NETIF_F_GSO_UDP_TUNNEL |
3178				NETIF_F_GSO_UDP_TUNNEL_CSUM;
3179
3180		netdev->hw_enc_features = NETIF_F_SG | NETIF_F_RXCSUM |
3181			NETIF_F_HW_CSUM | NETIF_F_HW_VLAN_CTAG_TX |
3182			NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_TSO | NETIF_F_TSO6 |
3183			NETIF_F_LRO | NETIF_F_GSO_UDP_TUNNEL |
3184			NETIF_F_GSO_UDP_TUNNEL_CSUM;
3185	}
3186
3187	if (dma64)
3188		netdev->hw_features |= NETIF_F_HIGHDMA;
3189	netdev->vlan_features = netdev->hw_features &
3190				~(NETIF_F_HW_VLAN_CTAG_TX |
3191				  NETIF_F_HW_VLAN_CTAG_RX);
3192	netdev->features = netdev->hw_features | NETIF_F_HW_VLAN_CTAG_FILTER;
3193}
3194
3195
3196static void
3197vmxnet3_read_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac)
3198{
3199	u32 tmp;
3200
3201	tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACL);
3202	*(u32 *)mac = tmp;
3203
3204	tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACH);
3205	mac[4] = tmp & 0xff;
3206	mac[5] = (tmp >> 8) & 0xff;
3207}
3208
3209#ifdef CONFIG_PCI_MSI
3210
3211/*
3212 * Enable MSIx vectors.
3213 * Returns :
3214 *	VMXNET3_LINUX_MIN_MSIX_VECT when only minimum number of vectors required
3215 *	 were enabled.
3216 *	number of vectors which were enabled otherwise (this number is greater
3217 *	 than VMXNET3_LINUX_MIN_MSIX_VECT)
3218 */
3219
3220static int
3221vmxnet3_acquire_msix_vectors(struct vmxnet3_adapter *adapter, int nvec)
3222{
3223	int ret = pci_enable_msix_range(adapter->pdev,
3224					adapter->intr.msix_entries, nvec, nvec);
3225
3226	if (ret == -ENOSPC && nvec > VMXNET3_LINUX_MIN_MSIX_VECT) {
3227		dev_err(&adapter->netdev->dev,
3228			"Failed to enable %d MSI-X, trying %d\n",
3229			nvec, VMXNET3_LINUX_MIN_MSIX_VECT);
3230
3231		ret = pci_enable_msix_range(adapter->pdev,
3232					    adapter->intr.msix_entries,
3233					    VMXNET3_LINUX_MIN_MSIX_VECT,
3234					    VMXNET3_LINUX_MIN_MSIX_VECT);
3235	}
3236
3237	if (ret < 0) {
3238		dev_err(&adapter->netdev->dev,
3239			"Failed to enable MSI-X, error: %d\n", ret);
3240	}
3241
3242	return ret;
3243}
3244
3245
3246#endif /* CONFIG_PCI_MSI */
3247
3248static void
3249vmxnet3_alloc_intr_resources(struct vmxnet3_adapter *adapter)
3250{
3251	u32 cfg;
3252	unsigned long flags;
3253
3254	/* intr settings */
3255	spin_lock_irqsave(&adapter->cmd_lock, flags);
3256	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3257			       VMXNET3_CMD_GET_CONF_INTR);
3258	cfg = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
3259	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3260	adapter->intr.type = cfg & 0x3;
3261	adapter->intr.mask_mode = (cfg >> 2) & 0x3;
3262
3263	if (adapter->intr.type == VMXNET3_IT_AUTO) {
3264		adapter->intr.type = VMXNET3_IT_MSIX;
3265	}
3266
3267#ifdef CONFIG_PCI_MSI
3268	if (adapter->intr.type == VMXNET3_IT_MSIX) {
3269		int i, nvec;
3270
3271		nvec  = adapter->share_intr == VMXNET3_INTR_TXSHARE ?
3272			1 : adapter->num_tx_queues;
3273		nvec += adapter->share_intr == VMXNET3_INTR_BUDDYSHARE ?
3274			0 : adapter->num_rx_queues;
3275		nvec += 1;	/* for link event */
3276		nvec = nvec > VMXNET3_LINUX_MIN_MSIX_VECT ?
3277		       nvec : VMXNET3_LINUX_MIN_MSIX_VECT;
3278
3279		for (i = 0; i < nvec; i++)
3280			adapter->intr.msix_entries[i].entry = i;
3281
3282		nvec = vmxnet3_acquire_msix_vectors(adapter, nvec);
3283		if (nvec < 0)
3284			goto msix_err;
3285
3286		/* If we cannot allocate one MSIx vector per queue
3287		 * then limit the number of rx queues to 1
3288		 */
3289		if (nvec == VMXNET3_LINUX_MIN_MSIX_VECT) {
3290			if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE
3291			    || adapter->num_rx_queues != 1) {
3292				adapter->share_intr = VMXNET3_INTR_TXSHARE;
3293				netdev_err(adapter->netdev,
3294					   "Number of rx queues : 1\n");
3295				adapter->num_rx_queues = 1;
3296			}
3297		}
3298
3299		adapter->intr.num_intrs = nvec;
3300		return;
3301
3302msix_err:
3303		/* If we cannot allocate MSIx vectors use only one rx queue */
3304		dev_info(&adapter->pdev->dev,
3305			 "Failed to enable MSI-X, error %d. "
3306			 "Limiting #rx queues to 1, try MSI.\n", nvec);
3307
3308		adapter->intr.type = VMXNET3_IT_MSI;
3309	}
3310
3311	if (adapter->intr.type == VMXNET3_IT_MSI) {
3312		if (!pci_enable_msi(adapter->pdev)) {
3313			adapter->num_rx_queues = 1;
3314			adapter->intr.num_intrs = 1;
3315			return;
3316		}
3317	}
3318#endif /* CONFIG_PCI_MSI */
3319
3320	adapter->num_rx_queues = 1;
3321	dev_info(&adapter->netdev->dev,
3322		 "Using INTx interrupt, #Rx queues: 1.\n");
3323	adapter->intr.type = VMXNET3_IT_INTX;
3324
3325	/* INT-X related setting */
3326	adapter->intr.num_intrs = 1;
3327}
3328
3329
3330static void
3331vmxnet3_free_intr_resources(struct vmxnet3_adapter *adapter)
3332{
3333	if (adapter->intr.type == VMXNET3_IT_MSIX)
3334		pci_disable_msix(adapter->pdev);
3335	else if (adapter->intr.type == VMXNET3_IT_MSI)
3336		pci_disable_msi(adapter->pdev);
3337	else
3338		BUG_ON(adapter->intr.type != VMXNET3_IT_INTX);
3339}
3340
3341
3342static void
3343vmxnet3_tx_timeout(struct net_device *netdev, unsigned int txqueue)
3344{
3345	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3346	adapter->tx_timeout_count++;
3347
3348	netdev_err(adapter->netdev, "tx hang\n");
3349	schedule_work(&adapter->work);
3350}
3351
3352
3353static void
3354vmxnet3_reset_work(struct work_struct *data)
3355{
3356	struct vmxnet3_adapter *adapter;
3357
3358	adapter = container_of(data, struct vmxnet3_adapter, work);
3359
3360	/* if another thread is resetting the device, no need to proceed */
3361	if (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
3362		return;
3363
3364	/* if the device is closed, we must leave it alone */
3365	rtnl_lock();
3366	if (netif_running(adapter->netdev)) {
3367		netdev_notice(adapter->netdev, "resetting\n");
3368		vmxnet3_quiesce_dev(adapter);
3369		vmxnet3_reset_dev(adapter);
3370		vmxnet3_activate_dev(adapter);
3371	} else {
3372		netdev_info(adapter->netdev, "already closed\n");
3373	}
3374	rtnl_unlock();
3375
3376	netif_wake_queue(adapter->netdev);
3377	clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
3378}
3379
3380
3381static int
3382vmxnet3_probe_device(struct pci_dev *pdev,
3383		     const struct pci_device_id *id)
3384{
3385	static const struct net_device_ops vmxnet3_netdev_ops = {
3386		.ndo_open = vmxnet3_open,
3387		.ndo_stop = vmxnet3_close,
3388		.ndo_start_xmit = vmxnet3_xmit_frame,
3389		.ndo_set_mac_address = vmxnet3_set_mac_addr,
3390		.ndo_change_mtu = vmxnet3_change_mtu,
3391		.ndo_fix_features = vmxnet3_fix_features,
3392		.ndo_set_features = vmxnet3_set_features,
3393		.ndo_features_check = vmxnet3_features_check,
3394		.ndo_get_stats64 = vmxnet3_get_stats64,
3395		.ndo_tx_timeout = vmxnet3_tx_timeout,
3396		.ndo_set_rx_mode = vmxnet3_set_mc,
3397		.ndo_vlan_rx_add_vid = vmxnet3_vlan_rx_add_vid,
3398		.ndo_vlan_rx_kill_vid = vmxnet3_vlan_rx_kill_vid,
3399#ifdef CONFIG_NET_POLL_CONTROLLER
3400		.ndo_poll_controller = vmxnet3_netpoll,
3401#endif
3402	};
3403	int err;
3404	bool dma64;
3405	u32 ver;
3406	struct net_device *netdev;
3407	struct vmxnet3_adapter *adapter;
3408	u8 mac[ETH_ALEN];
3409	int size;
3410	int num_tx_queues;
3411	int num_rx_queues;
3412
3413	if (!pci_msi_enabled())
3414		enable_mq = 0;
3415
3416#ifdef VMXNET3_RSS
3417	if (enable_mq)
3418		num_rx_queues = min(VMXNET3_DEVICE_MAX_RX_QUEUES,
3419				    (int)num_online_cpus());
3420	else
3421#endif
3422		num_rx_queues = 1;
3423	num_rx_queues = rounddown_pow_of_two(num_rx_queues);
3424
3425	if (enable_mq)
3426		num_tx_queues = min(VMXNET3_DEVICE_MAX_TX_QUEUES,
3427				    (int)num_online_cpus());
3428	else
3429		num_tx_queues = 1;
3430
3431	num_tx_queues = rounddown_pow_of_two(num_tx_queues);
3432	netdev = alloc_etherdev_mq(sizeof(struct vmxnet3_adapter),
3433				   max(num_tx_queues, num_rx_queues));
3434	dev_info(&pdev->dev,
3435		 "# of Tx queues : %d, # of Rx queues : %d\n",
3436		 num_tx_queues, num_rx_queues);
3437
3438	if (!netdev)
3439		return -ENOMEM;
3440
3441	pci_set_drvdata(pdev, netdev);
3442	adapter = netdev_priv(netdev);
3443	adapter->netdev = netdev;
3444	adapter->pdev = pdev;
3445
3446	adapter->tx_ring_size = VMXNET3_DEF_TX_RING_SIZE;
3447	adapter->rx_ring_size = VMXNET3_DEF_RX_RING_SIZE;
3448	adapter->rx_ring2_size = VMXNET3_DEF_RX_RING2_SIZE;
3449
3450	if (pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) == 0) {
3451		if (pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64)) != 0) {
3452			dev_err(&pdev->dev,
3453				"pci_set_consistent_dma_mask failed\n");
3454			err = -EIO;
3455			goto err_set_mask;
3456		}
3457		dma64 = true;
3458	} else {
3459		if (pci_set_dma_mask(pdev, DMA_BIT_MASK(32)) != 0) {
3460			dev_err(&pdev->dev,
3461				"pci_set_dma_mask failed\n");
3462			err = -EIO;
3463			goto err_set_mask;
3464		}
3465		dma64 = false;
3466	}
3467
3468	spin_lock_init(&adapter->cmd_lock);
3469	adapter->adapter_pa = dma_map_single(&adapter->pdev->dev, adapter,
3470					     sizeof(struct vmxnet3_adapter),
3471					     PCI_DMA_TODEVICE);
3472	if (dma_mapping_error(&adapter->pdev->dev, adapter->adapter_pa)) {
3473		dev_err(&pdev->dev, "Failed to map dma\n");
3474		err = -EFAULT;
3475		goto err_set_mask;
3476	}
3477	adapter->shared = dma_alloc_coherent(
3478				&adapter->pdev->dev,
3479				sizeof(struct Vmxnet3_DriverShared),
3480				&adapter->shared_pa, GFP_KERNEL);
3481	if (!adapter->shared) {
3482		dev_err(&pdev->dev, "Failed to allocate memory\n");
3483		err = -ENOMEM;
3484		goto err_alloc_shared;
3485	}
3486
3487	adapter->num_rx_queues = num_rx_queues;
3488	adapter->num_tx_queues = num_tx_queues;
3489	adapter->rx_buf_per_pkt = 1;
3490
3491	size = sizeof(struct Vmxnet3_TxQueueDesc) * adapter->num_tx_queues;
3492	size += sizeof(struct Vmxnet3_RxQueueDesc) * adapter->num_rx_queues;
3493	adapter->tqd_start = dma_alloc_coherent(&adapter->pdev->dev, size,
3494						&adapter->queue_desc_pa,
3495						GFP_KERNEL);
3496
3497	if (!adapter->tqd_start) {
3498		dev_err(&pdev->dev, "Failed to allocate memory\n");
3499		err = -ENOMEM;
3500		goto err_alloc_queue_desc;
3501	}
3502	adapter->rqd_start = (struct Vmxnet3_RxQueueDesc *)(adapter->tqd_start +
3503							    adapter->num_tx_queues);
3504
3505	adapter->pm_conf = dma_alloc_coherent(&adapter->pdev->dev,
3506					      sizeof(struct Vmxnet3_PMConf),
3507					      &adapter->pm_conf_pa,
3508					      GFP_KERNEL);
3509	if (adapter->pm_conf == NULL) {
3510		err = -ENOMEM;
3511		goto err_alloc_pm;
3512	}
3513
3514#ifdef VMXNET3_RSS
3515
3516	adapter->rss_conf = dma_alloc_coherent(&adapter->pdev->dev,
3517					       sizeof(struct UPT1_RSSConf),
3518					       &adapter->rss_conf_pa,
3519					       GFP_KERNEL);
3520	if (adapter->rss_conf == NULL) {
3521		err = -ENOMEM;
3522		goto err_alloc_rss;
3523	}
3524#endif /* VMXNET3_RSS */
3525
3526	err = vmxnet3_alloc_pci_resources(adapter);
3527	if (err < 0)
3528		goto err_alloc_pci;
3529
3530	ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_VRRS);
3531	if (ver & (1 << VMXNET3_REV_4)) {
3532		VMXNET3_WRITE_BAR1_REG(adapter,
3533				       VMXNET3_REG_VRRS,
3534				       1 << VMXNET3_REV_4);
3535		adapter->version = VMXNET3_REV_4 + 1;
3536	} else if (ver & (1 << VMXNET3_REV_3)) {
3537		VMXNET3_WRITE_BAR1_REG(adapter,
3538				       VMXNET3_REG_VRRS,
3539				       1 << VMXNET3_REV_3);
3540		adapter->version = VMXNET3_REV_3 + 1;
3541	} else if (ver & (1 << VMXNET3_REV_2)) {
3542		VMXNET3_WRITE_BAR1_REG(adapter,
3543				       VMXNET3_REG_VRRS,
3544				       1 << VMXNET3_REV_2);
3545		adapter->version = VMXNET3_REV_2 + 1;
3546	} else if (ver & (1 << VMXNET3_REV_1)) {
3547		VMXNET3_WRITE_BAR1_REG(adapter,
3548				       VMXNET3_REG_VRRS,
3549				       1 << VMXNET3_REV_1);
3550		adapter->version = VMXNET3_REV_1 + 1;
3551	} else {
3552		dev_err(&pdev->dev,
3553			"Incompatible h/w version (0x%x) for adapter\n", ver);
3554		err = -EBUSY;
3555		goto err_ver;
3556	}
3557	dev_dbg(&pdev->dev, "Using device version %d\n", adapter->version);
3558
3559	ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_UVRS);
3560	if (ver & 1) {
3561		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_UVRS, 1);
3562	} else {
3563		dev_err(&pdev->dev,
3564			"Incompatible upt version (0x%x) for adapter\n", ver);
3565		err = -EBUSY;
3566		goto err_ver;
3567	}
3568
3569	if (VMXNET3_VERSION_GE_3(adapter)) {
3570		adapter->coal_conf =
3571			dma_alloc_coherent(&adapter->pdev->dev,
3572					   sizeof(struct Vmxnet3_CoalesceScheme)
3573					   ,
3574					   &adapter->coal_conf_pa,
3575					   GFP_KERNEL);
3576		if (!adapter->coal_conf) {
3577			err = -ENOMEM;
3578			goto err_ver;
3579		}
3580		adapter->coal_conf->coalMode = VMXNET3_COALESCE_DISABLED;
3581		adapter->default_coal_mode = true;
3582	}
3583
3584	if (VMXNET3_VERSION_GE_4(adapter)) {
3585		adapter->default_rss_fields = true;
3586		adapter->rss_fields = VMXNET3_RSS_FIELDS_DEFAULT;
3587	}
3588
3589	SET_NETDEV_DEV(netdev, &pdev->dev);
3590	vmxnet3_declare_features(adapter, dma64);
3591
3592	adapter->rxdata_desc_size = VMXNET3_VERSION_GE_3(adapter) ?
3593		VMXNET3_DEF_RXDATA_DESC_SIZE : 0;
3594
3595	if (adapter->num_tx_queues == adapter->num_rx_queues)
3596		adapter->share_intr = VMXNET3_INTR_BUDDYSHARE;
3597	else
3598		adapter->share_intr = VMXNET3_INTR_DONTSHARE;
3599
3600	vmxnet3_alloc_intr_resources(adapter);
3601
3602#ifdef VMXNET3_RSS
3603	if (adapter->num_rx_queues > 1 &&
3604	    adapter->intr.type == VMXNET3_IT_MSIX) {
3605		adapter->rss = true;
3606		netdev->hw_features |= NETIF_F_RXHASH;
3607		netdev->features |= NETIF_F_RXHASH;
3608		dev_dbg(&pdev->dev, "RSS is enabled.\n");
3609	} else {
3610		adapter->rss = false;
3611	}
3612#endif
3613
3614	vmxnet3_read_mac_addr(adapter, mac);
3615	memcpy(netdev->dev_addr,  mac, netdev->addr_len);
3616
3617	netdev->netdev_ops = &vmxnet3_netdev_ops;
3618	vmxnet3_set_ethtool_ops(netdev);
3619	netdev->watchdog_timeo = 5 * HZ;
3620
3621	/* MTU range: 60 - 9000 */
3622	netdev->min_mtu = VMXNET3_MIN_MTU;
3623	netdev->max_mtu = VMXNET3_MAX_MTU;
3624
3625	INIT_WORK(&adapter->work, vmxnet3_reset_work);
3626	set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
3627
3628	if (adapter->intr.type == VMXNET3_IT_MSIX) {
3629		int i;
3630		for (i = 0; i < adapter->num_rx_queues; i++) {
3631			netif_napi_add(adapter->netdev,
3632				       &adapter->rx_queue[i].napi,
3633				       vmxnet3_poll_rx_only, 64);
3634		}
3635	} else {
3636		netif_napi_add(adapter->netdev, &adapter->rx_queue[0].napi,
3637			       vmxnet3_poll, 64);
3638	}
3639
3640	netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
3641	netif_set_real_num_rx_queues(adapter->netdev, adapter->num_rx_queues);
3642
3643	netif_carrier_off(netdev);
3644	err = register_netdev(netdev);
3645
3646	if (err) {
3647		dev_err(&pdev->dev, "Failed to register adapter\n");
3648		goto err_register;
3649	}
3650
3651	vmxnet3_check_link(adapter, false);
3652	return 0;
3653
3654err_register:
3655	if (VMXNET3_VERSION_GE_3(adapter)) {
3656		dma_free_coherent(&adapter->pdev->dev,
3657				  sizeof(struct Vmxnet3_CoalesceScheme),
3658				  adapter->coal_conf, adapter->coal_conf_pa);
3659	}
3660	vmxnet3_free_intr_resources(adapter);
3661err_ver:
3662	vmxnet3_free_pci_resources(adapter);
3663err_alloc_pci:
3664#ifdef VMXNET3_RSS
3665	dma_free_coherent(&adapter->pdev->dev, sizeof(struct UPT1_RSSConf),
3666			  adapter->rss_conf, adapter->rss_conf_pa);
3667err_alloc_rss:
3668#endif
3669	dma_free_coherent(&adapter->pdev->dev, sizeof(struct Vmxnet3_PMConf),
3670			  adapter->pm_conf, adapter->pm_conf_pa);
3671err_alloc_pm:
3672	dma_free_coherent(&adapter->pdev->dev, size, adapter->tqd_start,
3673			  adapter->queue_desc_pa);
3674err_alloc_queue_desc:
3675	dma_free_coherent(&adapter->pdev->dev,
3676			  sizeof(struct Vmxnet3_DriverShared),
3677			  adapter->shared, adapter->shared_pa);
3678err_alloc_shared:
3679	dma_unmap_single(&adapter->pdev->dev, adapter->adapter_pa,
3680			 sizeof(struct vmxnet3_adapter), PCI_DMA_TODEVICE);
3681err_set_mask:
3682	free_netdev(netdev);
3683	return err;
3684}
3685
3686
3687static void
3688vmxnet3_remove_device(struct pci_dev *pdev)
3689{
3690	struct net_device *netdev = pci_get_drvdata(pdev);
3691	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3692	int size = 0;
3693	int num_rx_queues;
3694
3695#ifdef VMXNET3_RSS
3696	if (enable_mq)
3697		num_rx_queues = min(VMXNET3_DEVICE_MAX_RX_QUEUES,
3698				    (int)num_online_cpus());
3699	else
3700#endif
3701		num_rx_queues = 1;
3702	num_rx_queues = rounddown_pow_of_two(num_rx_queues);
3703
3704	cancel_work_sync(&adapter->work);
3705
3706	unregister_netdev(netdev);
3707
3708	vmxnet3_free_intr_resources(adapter);
3709	vmxnet3_free_pci_resources(adapter);
3710	if (VMXNET3_VERSION_GE_3(adapter)) {
3711		dma_free_coherent(&adapter->pdev->dev,
3712				  sizeof(struct Vmxnet3_CoalesceScheme),
3713				  adapter->coal_conf, adapter->coal_conf_pa);
3714	}
3715#ifdef VMXNET3_RSS
3716	dma_free_coherent(&adapter->pdev->dev, sizeof(struct UPT1_RSSConf),
3717			  adapter->rss_conf, adapter->rss_conf_pa);
3718#endif
3719	dma_free_coherent(&adapter->pdev->dev, sizeof(struct Vmxnet3_PMConf),
3720			  adapter->pm_conf, adapter->pm_conf_pa);
3721
3722	size = sizeof(struct Vmxnet3_TxQueueDesc) * adapter->num_tx_queues;
3723	size += sizeof(struct Vmxnet3_RxQueueDesc) * num_rx_queues;
3724	dma_free_coherent(&adapter->pdev->dev, size, adapter->tqd_start,
3725			  adapter->queue_desc_pa);
3726	dma_free_coherent(&adapter->pdev->dev,
3727			  sizeof(struct Vmxnet3_DriverShared),
3728			  adapter->shared, adapter->shared_pa);
3729	dma_unmap_single(&adapter->pdev->dev, adapter->adapter_pa,
3730			 sizeof(struct vmxnet3_adapter), PCI_DMA_TODEVICE);
3731	free_netdev(netdev);
3732}
3733
3734static void vmxnet3_shutdown_device(struct pci_dev *pdev)
3735{
3736	struct net_device *netdev = pci_get_drvdata(pdev);
3737	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3738	unsigned long flags;
3739
3740	/* Reset_work may be in the middle of resetting the device, wait for its
3741	 * completion.
3742	 */
3743	while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
3744		usleep_range(1000, 2000);
3745
3746	if (test_and_set_bit(VMXNET3_STATE_BIT_QUIESCED,
3747			     &adapter->state)) {
3748		clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
3749		return;
3750	}
3751	spin_lock_irqsave(&adapter->cmd_lock, flags);
3752	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3753			       VMXNET3_CMD_QUIESCE_DEV);
3754	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3755	vmxnet3_disable_all_intrs(adapter);
3756
3757	clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
3758}
3759
3760
3761#ifdef CONFIG_PM
3762
3763static int
3764vmxnet3_suspend(struct device *device)
3765{
3766	struct pci_dev *pdev = to_pci_dev(device);
3767	struct net_device *netdev = pci_get_drvdata(pdev);
3768	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3769	struct Vmxnet3_PMConf *pmConf;
3770	struct ethhdr *ehdr;
3771	struct arphdr *ahdr;
3772	u8 *arpreq;
3773	struct in_device *in_dev;
3774	struct in_ifaddr *ifa;
3775	unsigned long flags;
3776	int i = 0;
3777
3778	if (!netif_running(netdev))
3779		return 0;
3780
3781	for (i = 0; i < adapter->num_rx_queues; i++)
3782		napi_disable(&adapter->rx_queue[i].napi);
3783
3784	vmxnet3_disable_all_intrs(adapter);
3785	vmxnet3_free_irqs(adapter);
3786	vmxnet3_free_intr_resources(adapter);
3787
3788	netif_device_detach(netdev);
3789
3790	/* Create wake-up filters. */
3791	pmConf = adapter->pm_conf;
3792	memset(pmConf, 0, sizeof(*pmConf));
3793
3794	if (adapter->wol & WAKE_UCAST) {
3795		pmConf->filters[i].patternSize = ETH_ALEN;
3796		pmConf->filters[i].maskSize = 1;
3797		memcpy(pmConf->filters[i].pattern, netdev->dev_addr, ETH_ALEN);
3798		pmConf->filters[i].mask[0] = 0x3F; /* LSB ETH_ALEN bits */
3799
3800		pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_FILTER;
3801		i++;
3802	}
3803
3804	if (adapter->wol & WAKE_ARP) {
3805		rcu_read_lock();
3806
3807		in_dev = __in_dev_get_rcu(netdev);
3808		if (!in_dev) {
3809			rcu_read_unlock();
3810			goto skip_arp;
3811		}
3812
3813		ifa = rcu_dereference(in_dev->ifa_list);
3814		if (!ifa) {
3815			rcu_read_unlock();
3816			goto skip_arp;
3817		}
3818
3819		pmConf->filters[i].patternSize = ETH_HLEN + /* Ethernet header*/
3820			sizeof(struct arphdr) +		/* ARP header */
3821			2 * ETH_ALEN +		/* 2 Ethernet addresses*/
3822			2 * sizeof(u32);	/*2 IPv4 addresses */
3823		pmConf->filters[i].maskSize =
3824			(pmConf->filters[i].patternSize - 1) / 8 + 1;
3825
3826		/* ETH_P_ARP in Ethernet header. */
3827		ehdr = (struct ethhdr *)pmConf->filters[i].pattern;
3828		ehdr->h_proto = htons(ETH_P_ARP);
3829
3830		/* ARPOP_REQUEST in ARP header. */
3831		ahdr = (struct arphdr *)&pmConf->filters[i].pattern[ETH_HLEN];
3832		ahdr->ar_op = htons(ARPOP_REQUEST);
3833		arpreq = (u8 *)(ahdr + 1);
3834
3835		/* The Unicast IPv4 address in 'tip' field. */
3836		arpreq += 2 * ETH_ALEN + sizeof(u32);
3837		*(__be32 *)arpreq = ifa->ifa_address;
3838
3839		rcu_read_unlock();
3840
3841		/* The mask for the relevant bits. */
3842		pmConf->filters[i].mask[0] = 0x00;
3843		pmConf->filters[i].mask[1] = 0x30; /* ETH_P_ARP */
3844		pmConf->filters[i].mask[2] = 0x30; /* ARPOP_REQUEST */
3845		pmConf->filters[i].mask[3] = 0x00;
3846		pmConf->filters[i].mask[4] = 0xC0; /* IPv4 TIP */
3847		pmConf->filters[i].mask[5] = 0x03; /* IPv4 TIP */
3848
3849		pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_FILTER;
3850		i++;
3851	}
3852
3853skip_arp:
3854	if (adapter->wol & WAKE_MAGIC)
3855		pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_MAGIC;
3856
3857	pmConf->numFilters = i;
3858
3859	adapter->shared->devRead.pmConfDesc.confVer = cpu_to_le32(1);
3860	adapter->shared->devRead.pmConfDesc.confLen = cpu_to_le32(sizeof(
3861								  *pmConf));
3862	adapter->shared->devRead.pmConfDesc.confPA =
3863		cpu_to_le64(adapter->pm_conf_pa);
3864
3865	spin_lock_irqsave(&adapter->cmd_lock, flags);
3866	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3867			       VMXNET3_CMD_UPDATE_PMCFG);
3868	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3869
3870	pci_save_state(pdev);
3871	pci_enable_wake(pdev, pci_choose_state(pdev, PMSG_SUSPEND),
3872			adapter->wol);
3873	pci_disable_device(pdev);
3874	pci_set_power_state(pdev, pci_choose_state(pdev, PMSG_SUSPEND));
3875
3876	return 0;
3877}
3878
3879
3880static int
3881vmxnet3_resume(struct device *device)
3882{
3883	int err;
3884	unsigned long flags;
3885	struct pci_dev *pdev = to_pci_dev(device);
3886	struct net_device *netdev = pci_get_drvdata(pdev);
3887	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3888
3889	if (!netif_running(netdev))
3890		return 0;
3891
3892	pci_set_power_state(pdev, PCI_D0);
3893	pci_restore_state(pdev);
3894	err = pci_enable_device_mem(pdev);
3895	if (err != 0)
3896		return err;
3897
3898	pci_enable_wake(pdev, PCI_D0, 0);
3899
3900	vmxnet3_alloc_intr_resources(adapter);
3901
3902	/* During hibernate and suspend, device has to be reinitialized as the
3903	 * device state need not be preserved.
3904	 */
3905
3906	/* Need not check adapter state as other reset tasks cannot run during
3907	 * device resume.
3908	 */
3909	spin_lock_irqsave(&adapter->cmd_lock, flags);
3910	VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3911			       VMXNET3_CMD_QUIESCE_DEV);
3912	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3913	vmxnet3_tq_cleanup_all(adapter);
3914	vmxnet3_rq_cleanup_all(adapter);
3915
3916	vmxnet3_reset_dev(adapter);
3917	err = vmxnet3_activate_dev(adapter);
3918	if (err != 0) {
3919		netdev_err(netdev,
3920			   "failed to re-activate on resume, error: %d", err);
3921		vmxnet3_force_close(adapter);
3922		return err;
3923	}
3924	netif_device_attach(netdev);
3925
3926	return 0;
3927}
3928
3929static const struct dev_pm_ops vmxnet3_pm_ops = {
3930	.suspend = vmxnet3_suspend,
3931	.resume = vmxnet3_resume,
3932	.freeze = vmxnet3_suspend,
3933	.restore = vmxnet3_resume,
3934};
3935#endif
3936
3937static struct pci_driver vmxnet3_driver = {
3938	.name		= vmxnet3_driver_name,
3939	.id_table	= vmxnet3_pciid_table,
3940	.probe		= vmxnet3_probe_device,
3941	.remove		= vmxnet3_remove_device,
3942	.shutdown	= vmxnet3_shutdown_device,
3943#ifdef CONFIG_PM
3944	.driver.pm	= &vmxnet3_pm_ops,
3945#endif
3946};
3947
3948
3949static int __init
3950vmxnet3_init_module(void)
3951{
3952	pr_info("%s - version %s\n", VMXNET3_DRIVER_DESC,
3953		VMXNET3_DRIVER_VERSION_REPORT);
3954	return pci_register_driver(&vmxnet3_driver);
3955}
3956
3957module_init(vmxnet3_init_module);
3958
3959
3960static void
3961vmxnet3_exit_module(void)
3962{
3963	pci_unregister_driver(&vmxnet3_driver);
3964}
3965
3966module_exit(vmxnet3_exit_module);
3967
3968MODULE_AUTHOR("VMware, Inc.");
3969MODULE_DESCRIPTION(VMXNET3_DRIVER_DESC);
3970MODULE_LICENSE("GPL v2");
3971MODULE_VERSION(VMXNET3_DRIVER_VERSION_STRING);
3972