1 /*
2  * Copyright 2008-2010 Cisco Systems, Inc.  All rights reserved.
3  * Copyright 2007 Nuova Systems, Inc.  All rights reserved.
4  *
5  * This program is free software; you may redistribute it and/or modify
6  * it under the terms of the GNU General Public License as published by
7  * the Free Software Foundation; version 2 of the License.
8  *
9  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
10  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
11  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
12  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
13  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
14  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
15  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
16  * SOFTWARE.
17  *
18  */
19 
20 #include <linux/module.h>
21 #include <linux/kernel.h>
22 #include <linux/string.h>
23 #include <linux/errno.h>
24 #include <linux/types.h>
25 #include <linux/init.h>
26 #include <linux/interrupt.h>
27 #include <linux/workqueue.h>
28 #include <linux/pci.h>
29 #include <linux/netdevice.h>
30 #include <linux/etherdevice.h>
31 #include <linux/if.h>
32 #include <linux/if_ether.h>
33 #include <linux/if_vlan.h>
34 #include <linux/in.h>
35 #include <linux/ip.h>
36 #include <linux/ipv6.h>
37 #include <linux/tcp.h>
38 #include <linux/rtnetlink.h>
39 #include <linux/prefetch.h>
40 #include <net/ip6_checksum.h>
41 #include <linux/ktime.h>
42 #ifdef CONFIG_RFS_ACCEL
43 #include <linux/cpu_rmap.h>
44 #endif
45 #ifdef CONFIG_NET_RX_BUSY_POLL
46 #include <net/busy_poll.h>
47 #endif
48 #include <linux/crash_dump.h>
49 
50 #include "cq_enet_desc.h"
51 #include "vnic_dev.h"
52 #include "vnic_intr.h"
53 #include "vnic_stats.h"
54 #include "vnic_vic.h"
55 #include "enic_res.h"
56 #include "enic.h"
57 #include "enic_dev.h"
58 #include "enic_pp.h"
59 #include "enic_clsf.h"
60 
61 #define ENIC_NOTIFY_TIMER_PERIOD	(2 * HZ)
62 #define WQ_ENET_MAX_DESC_LEN		(1 << WQ_ENET_LEN_BITS)
63 #define MAX_TSO				(1 << 16)
64 #define ENIC_DESC_MAX_SPLITS		(MAX_TSO / WQ_ENET_MAX_DESC_LEN + 1)
65 
66 #define PCI_DEVICE_ID_CISCO_VIC_ENET         0x0043  /* ethernet vnic */
67 #define PCI_DEVICE_ID_CISCO_VIC_ENET_DYN     0x0044  /* enet dynamic vnic */
68 #define PCI_DEVICE_ID_CISCO_VIC_ENET_VF      0x0071  /* enet SRIOV VF */
69 
70 #define RX_COPYBREAK_DEFAULT		256
71 
72 /* Supported devices */
73 static const struct pci_device_id enic_id_table[] = {
74 	{ PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET) },
75 	{ PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET_DYN) },
76 	{ PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET_VF) },
77 	{ 0, }	/* end of table */
78 };
79 
80 MODULE_DESCRIPTION(DRV_DESCRIPTION);
81 MODULE_AUTHOR("Scott Feldman <scofeldm@cisco.com>");
82 MODULE_LICENSE("GPL");
83 MODULE_VERSION(DRV_VERSION);
84 MODULE_DEVICE_TABLE(pci, enic_id_table);
85 
86 #define ENIC_LARGE_PKT_THRESHOLD		1000
87 #define ENIC_MAX_COALESCE_TIMERS		10
88 /*  Interrupt moderation table, which will be used to decide the
89  *  coalescing timer values
90  *  {rx_rate in Mbps, mapping percentage of the range}
91  */
92 static struct enic_intr_mod_table mod_table[ENIC_MAX_COALESCE_TIMERS + 1] = {
93 	{4000,  0},
94 	{4400, 10},
95 	{5060, 20},
96 	{5230, 30},
97 	{5540, 40},
98 	{5820, 50},
99 	{6120, 60},
100 	{6435, 70},
101 	{6745, 80},
102 	{7000, 90},
103 	{0xFFFFFFFF, 100}
104 };
105 
106 /* This table helps the driver to pick different ranges for rx coalescing
107  * timer depending on the link speed.
108  */
109 static struct enic_intr_mod_range mod_range[ENIC_MAX_LINK_SPEEDS] = {
110 	{0,  0}, /* 0  - 4  Gbps */
111 	{0,  3}, /* 4  - 10 Gbps */
112 	{3,  6}, /* 10 - 40 Gbps */
113 };
114 
115 int enic_is_dynamic(struct enic *enic)
116 {
117 	return enic->pdev->device == PCI_DEVICE_ID_CISCO_VIC_ENET_DYN;
118 }
119 
120 int enic_sriov_enabled(struct enic *enic)
121 {
122 	return (enic->priv_flags & ENIC_SRIOV_ENABLED) ? 1 : 0;
123 }
124 
125 static int enic_is_sriov_vf(struct enic *enic)
126 {
127 	return enic->pdev->device == PCI_DEVICE_ID_CISCO_VIC_ENET_VF;
128 }
129 
130 int enic_is_valid_vf(struct enic *enic, int vf)
131 {
132 #ifdef CONFIG_PCI_IOV
133 	return vf >= 0 && vf < enic->num_vfs;
134 #else
135 	return 0;
136 #endif
137 }
138 
139 static void enic_free_wq_buf(struct vnic_wq *wq, struct vnic_wq_buf *buf)
140 {
141 	struct enic *enic = vnic_dev_priv(wq->vdev);
142 
143 	if (buf->sop)
144 		pci_unmap_single(enic->pdev, buf->dma_addr,
145 			buf->len, PCI_DMA_TODEVICE);
146 	else
147 		pci_unmap_page(enic->pdev, buf->dma_addr,
148 			buf->len, PCI_DMA_TODEVICE);
149 
150 	if (buf->os_buf)
151 		dev_kfree_skb_any(buf->os_buf);
152 }
153 
154 static void enic_wq_free_buf(struct vnic_wq *wq,
155 	struct cq_desc *cq_desc, struct vnic_wq_buf *buf, void *opaque)
156 {
157 	enic_free_wq_buf(wq, buf);
158 }
159 
160 static int enic_wq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
161 	u8 type, u16 q_number, u16 completed_index, void *opaque)
162 {
163 	struct enic *enic = vnic_dev_priv(vdev);
164 
165 	spin_lock(&enic->wq_lock[q_number]);
166 
167 	vnic_wq_service(&enic->wq[q_number], cq_desc,
168 		completed_index, enic_wq_free_buf,
169 		opaque);
170 
171 	if (netif_tx_queue_stopped(netdev_get_tx_queue(enic->netdev, q_number)) &&
172 	    vnic_wq_desc_avail(&enic->wq[q_number]) >=
173 	    (MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS))
174 		netif_wake_subqueue(enic->netdev, q_number);
175 
176 	spin_unlock(&enic->wq_lock[q_number]);
177 
178 	return 0;
179 }
180 
181 static void enic_log_q_error(struct enic *enic)
182 {
183 	unsigned int i;
184 	u32 error_status;
185 
186 	for (i = 0; i < enic->wq_count; i++) {
187 		error_status = vnic_wq_error_status(&enic->wq[i]);
188 		if (error_status)
189 			netdev_err(enic->netdev, "WQ[%d] error_status %d\n",
190 				i, error_status);
191 	}
192 
193 	for (i = 0; i < enic->rq_count; i++) {
194 		error_status = vnic_rq_error_status(&enic->rq[i]);
195 		if (error_status)
196 			netdev_err(enic->netdev, "RQ[%d] error_status %d\n",
197 				i, error_status);
198 	}
199 }
200 
201 static void enic_msglvl_check(struct enic *enic)
202 {
203 	u32 msg_enable = vnic_dev_msg_lvl(enic->vdev);
204 
205 	if (msg_enable != enic->msg_enable) {
206 		netdev_info(enic->netdev, "msg lvl changed from 0x%x to 0x%x\n",
207 			enic->msg_enable, msg_enable);
208 		enic->msg_enable = msg_enable;
209 	}
210 }
211 
212 static void enic_mtu_check(struct enic *enic)
213 {
214 	u32 mtu = vnic_dev_mtu(enic->vdev);
215 	struct net_device *netdev = enic->netdev;
216 
217 	if (mtu && mtu != enic->port_mtu) {
218 		enic->port_mtu = mtu;
219 		if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic)) {
220 			mtu = max_t(int, ENIC_MIN_MTU,
221 				min_t(int, ENIC_MAX_MTU, mtu));
222 			if (mtu != netdev->mtu)
223 				schedule_work(&enic->change_mtu_work);
224 		} else {
225 			if (mtu < netdev->mtu)
226 				netdev_warn(netdev,
227 					"interface MTU (%d) set higher "
228 					"than switch port MTU (%d)\n",
229 					netdev->mtu, mtu);
230 		}
231 	}
232 }
233 
234 static void enic_link_check(struct enic *enic)
235 {
236 	int link_status = vnic_dev_link_status(enic->vdev);
237 	int carrier_ok = netif_carrier_ok(enic->netdev);
238 
239 	if (link_status && !carrier_ok) {
240 		netdev_info(enic->netdev, "Link UP\n");
241 		netif_carrier_on(enic->netdev);
242 	} else if (!link_status && carrier_ok) {
243 		netdev_info(enic->netdev, "Link DOWN\n");
244 		netif_carrier_off(enic->netdev);
245 	}
246 }
247 
248 static void enic_notify_check(struct enic *enic)
249 {
250 	enic_msglvl_check(enic);
251 	enic_mtu_check(enic);
252 	enic_link_check(enic);
253 }
254 
255 #define ENIC_TEST_INTR(pba, i) (pba & (1 << i))
256 
257 static irqreturn_t enic_isr_legacy(int irq, void *data)
258 {
259 	struct net_device *netdev = data;
260 	struct enic *enic = netdev_priv(netdev);
261 	unsigned int io_intr = enic_legacy_io_intr();
262 	unsigned int err_intr = enic_legacy_err_intr();
263 	unsigned int notify_intr = enic_legacy_notify_intr();
264 	u32 pba;
265 
266 	vnic_intr_mask(&enic->intr[io_intr]);
267 
268 	pba = vnic_intr_legacy_pba(enic->legacy_pba);
269 	if (!pba) {
270 		vnic_intr_unmask(&enic->intr[io_intr]);
271 		return IRQ_NONE;	/* not our interrupt */
272 	}
273 
274 	if (ENIC_TEST_INTR(pba, notify_intr)) {
275 		enic_notify_check(enic);
276 		vnic_intr_return_all_credits(&enic->intr[notify_intr]);
277 	}
278 
279 	if (ENIC_TEST_INTR(pba, err_intr)) {
280 		vnic_intr_return_all_credits(&enic->intr[err_intr]);
281 		enic_log_q_error(enic);
282 		/* schedule recovery from WQ/RQ error */
283 		schedule_work(&enic->reset);
284 		return IRQ_HANDLED;
285 	}
286 
287 	if (ENIC_TEST_INTR(pba, io_intr))
288 		napi_schedule_irqoff(&enic->napi[0]);
289 	else
290 		vnic_intr_unmask(&enic->intr[io_intr]);
291 
292 	return IRQ_HANDLED;
293 }
294 
295 static irqreturn_t enic_isr_msi(int irq, void *data)
296 {
297 	struct enic *enic = data;
298 
299 	/* With MSI, there is no sharing of interrupts, so this is
300 	 * our interrupt and there is no need to ack it.  The device
301 	 * is not providing per-vector masking, so the OS will not
302 	 * write to PCI config space to mask/unmask the interrupt.
303 	 * We're using mask_on_assertion for MSI, so the device
304 	 * automatically masks the interrupt when the interrupt is
305 	 * generated.  Later, when exiting polling, the interrupt
306 	 * will be unmasked (see enic_poll).
307 	 *
308 	 * Also, the device uses the same PCIe Traffic Class (TC)
309 	 * for Memory Write data and MSI, so there are no ordering
310 	 * issues; the MSI will always arrive at the Root Complex
311 	 * _after_ corresponding Memory Writes (i.e. descriptor
312 	 * writes).
313 	 */
314 
315 	napi_schedule_irqoff(&enic->napi[0]);
316 
317 	return IRQ_HANDLED;
318 }
319 
320 static irqreturn_t enic_isr_msix(int irq, void *data)
321 {
322 	struct napi_struct *napi = data;
323 
324 	napi_schedule_irqoff(napi);
325 
326 	return IRQ_HANDLED;
327 }
328 
329 static irqreturn_t enic_isr_msix_err(int irq, void *data)
330 {
331 	struct enic *enic = data;
332 	unsigned int intr = enic_msix_err_intr(enic);
333 
334 	vnic_intr_return_all_credits(&enic->intr[intr]);
335 
336 	enic_log_q_error(enic);
337 
338 	/* schedule recovery from WQ/RQ error */
339 	schedule_work(&enic->reset);
340 
341 	return IRQ_HANDLED;
342 }
343 
344 static irqreturn_t enic_isr_msix_notify(int irq, void *data)
345 {
346 	struct enic *enic = data;
347 	unsigned int intr = enic_msix_notify_intr(enic);
348 
349 	enic_notify_check(enic);
350 	vnic_intr_return_all_credits(&enic->intr[intr]);
351 
352 	return IRQ_HANDLED;
353 }
354 
355 static int enic_queue_wq_skb_cont(struct enic *enic, struct vnic_wq *wq,
356 				  struct sk_buff *skb, unsigned int len_left,
357 				  int loopback)
358 {
359 	const skb_frag_t *frag;
360 	dma_addr_t dma_addr;
361 
362 	/* Queue additional data fragments */
363 	for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
364 		len_left -= skb_frag_size(frag);
365 		dma_addr = skb_frag_dma_map(&enic->pdev->dev, frag, 0,
366 					    skb_frag_size(frag),
367 					    DMA_TO_DEVICE);
368 		if (unlikely(enic_dma_map_check(enic, dma_addr)))
369 			return -ENOMEM;
370 		enic_queue_wq_desc_cont(wq, skb, dma_addr, skb_frag_size(frag),
371 					(len_left == 0),	/* EOP? */
372 					loopback);
373 	}
374 
375 	return 0;
376 }
377 
378 static int enic_queue_wq_skb_vlan(struct enic *enic, struct vnic_wq *wq,
379 				  struct sk_buff *skb, int vlan_tag_insert,
380 				  unsigned int vlan_tag, int loopback)
381 {
382 	unsigned int head_len = skb_headlen(skb);
383 	unsigned int len_left = skb->len - head_len;
384 	int eop = (len_left == 0);
385 	dma_addr_t dma_addr;
386 	int err = 0;
387 
388 	dma_addr = pci_map_single(enic->pdev, skb->data, head_len,
389 				  PCI_DMA_TODEVICE);
390 	if (unlikely(enic_dma_map_check(enic, dma_addr)))
391 		return -ENOMEM;
392 
393 	/* Queue the main skb fragment. The fragments are no larger
394 	 * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less
395 	 * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor
396 	 * per fragment is queued.
397 	 */
398 	enic_queue_wq_desc(wq, skb, dma_addr, head_len,	vlan_tag_insert,
399 			   vlan_tag, eop, loopback);
400 
401 	if (!eop)
402 		err = enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback);
403 
404 	return err;
405 }
406 
407 static int enic_queue_wq_skb_csum_l4(struct enic *enic, struct vnic_wq *wq,
408 				     struct sk_buff *skb, int vlan_tag_insert,
409 				     unsigned int vlan_tag, int loopback)
410 {
411 	unsigned int head_len = skb_headlen(skb);
412 	unsigned int len_left = skb->len - head_len;
413 	unsigned int hdr_len = skb_checksum_start_offset(skb);
414 	unsigned int csum_offset = hdr_len + skb->csum_offset;
415 	int eop = (len_left == 0);
416 	dma_addr_t dma_addr;
417 	int err = 0;
418 
419 	dma_addr = pci_map_single(enic->pdev, skb->data, head_len,
420 				  PCI_DMA_TODEVICE);
421 	if (unlikely(enic_dma_map_check(enic, dma_addr)))
422 		return -ENOMEM;
423 
424 	/* Queue the main skb fragment. The fragments are no larger
425 	 * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less
426 	 * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor
427 	 * per fragment is queued.
428 	 */
429 	enic_queue_wq_desc_csum_l4(wq, skb, dma_addr, head_len,	csum_offset,
430 				   hdr_len, vlan_tag_insert, vlan_tag, eop,
431 				   loopback);
432 
433 	if (!eop)
434 		err = enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback);
435 
436 	return err;
437 }
438 
439 static int enic_queue_wq_skb_tso(struct enic *enic, struct vnic_wq *wq,
440 				 struct sk_buff *skb, unsigned int mss,
441 				 int vlan_tag_insert, unsigned int vlan_tag,
442 				 int loopback)
443 {
444 	unsigned int frag_len_left = skb_headlen(skb);
445 	unsigned int len_left = skb->len - frag_len_left;
446 	unsigned int hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
447 	int eop = (len_left == 0);
448 	unsigned int len;
449 	dma_addr_t dma_addr;
450 	unsigned int offset = 0;
451 	skb_frag_t *frag;
452 
453 	/* Preload TCP csum field with IP pseudo hdr calculated
454 	 * with IP length set to zero.  HW will later add in length
455 	 * to each TCP segment resulting from the TSO.
456 	 */
457 
458 	if (skb->protocol == cpu_to_be16(ETH_P_IP)) {
459 		ip_hdr(skb)->check = 0;
460 		tcp_hdr(skb)->check = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
461 			ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
462 	} else if (skb->protocol == cpu_to_be16(ETH_P_IPV6)) {
463 		tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
464 			&ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
465 	}
466 
467 	/* Queue WQ_ENET_MAX_DESC_LEN length descriptors
468 	 * for the main skb fragment
469 	 */
470 	while (frag_len_left) {
471 		len = min(frag_len_left, (unsigned int)WQ_ENET_MAX_DESC_LEN);
472 		dma_addr = pci_map_single(enic->pdev, skb->data + offset, len,
473 					  PCI_DMA_TODEVICE);
474 		if (unlikely(enic_dma_map_check(enic, dma_addr)))
475 			return -ENOMEM;
476 		enic_queue_wq_desc_tso(wq, skb, dma_addr, len, mss, hdr_len,
477 				       vlan_tag_insert, vlan_tag,
478 				       eop && (len == frag_len_left), loopback);
479 		frag_len_left -= len;
480 		offset += len;
481 	}
482 
483 	if (eop)
484 		return 0;
485 
486 	/* Queue WQ_ENET_MAX_DESC_LEN length descriptors
487 	 * for additional data fragments
488 	 */
489 	for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
490 		len_left -= skb_frag_size(frag);
491 		frag_len_left = skb_frag_size(frag);
492 		offset = 0;
493 
494 		while (frag_len_left) {
495 			len = min(frag_len_left,
496 				(unsigned int)WQ_ENET_MAX_DESC_LEN);
497 			dma_addr = skb_frag_dma_map(&enic->pdev->dev, frag,
498 						    offset, len,
499 						    DMA_TO_DEVICE);
500 			if (unlikely(enic_dma_map_check(enic, dma_addr)))
501 				return -ENOMEM;
502 			enic_queue_wq_desc_cont(wq, skb, dma_addr, len,
503 						(len_left == 0) &&
504 						 (len == frag_len_left),/*EOP*/
505 						loopback);
506 			frag_len_left -= len;
507 			offset += len;
508 		}
509 	}
510 
511 	return 0;
512 }
513 
514 static inline void enic_queue_wq_skb(struct enic *enic,
515 	struct vnic_wq *wq, struct sk_buff *skb)
516 {
517 	unsigned int mss = skb_shinfo(skb)->gso_size;
518 	unsigned int vlan_tag = 0;
519 	int vlan_tag_insert = 0;
520 	int loopback = 0;
521 	int err;
522 
523 	if (skb_vlan_tag_present(skb)) {
524 		/* VLAN tag from trunking driver */
525 		vlan_tag_insert = 1;
526 		vlan_tag = skb_vlan_tag_get(skb);
527 	} else if (enic->loop_enable) {
528 		vlan_tag = enic->loop_tag;
529 		loopback = 1;
530 	}
531 
532 	if (mss)
533 		err = enic_queue_wq_skb_tso(enic, wq, skb, mss,
534 					    vlan_tag_insert, vlan_tag,
535 					    loopback);
536 	else if	(skb->ip_summed == CHECKSUM_PARTIAL)
537 		err = enic_queue_wq_skb_csum_l4(enic, wq, skb, vlan_tag_insert,
538 						vlan_tag, loopback);
539 	else
540 		err = enic_queue_wq_skb_vlan(enic, wq, skb, vlan_tag_insert,
541 					     vlan_tag, loopback);
542 	if (unlikely(err)) {
543 		struct vnic_wq_buf *buf;
544 
545 		buf = wq->to_use->prev;
546 		/* while not EOP of previous pkt && queue not empty.
547 		 * For all non EOP bufs, os_buf is NULL.
548 		 */
549 		while (!buf->os_buf && (buf->next != wq->to_clean)) {
550 			enic_free_wq_buf(wq, buf);
551 			wq->ring.desc_avail++;
552 			buf = buf->prev;
553 		}
554 		wq->to_use = buf->next;
555 		dev_kfree_skb(skb);
556 	}
557 }
558 
559 /* netif_tx_lock held, process context with BHs disabled, or BH */
560 static netdev_tx_t enic_hard_start_xmit(struct sk_buff *skb,
561 	struct net_device *netdev)
562 {
563 	struct enic *enic = netdev_priv(netdev);
564 	struct vnic_wq *wq;
565 	unsigned int txq_map;
566 	struct netdev_queue *txq;
567 
568 	if (skb->len <= 0) {
569 		dev_kfree_skb_any(skb);
570 		return NETDEV_TX_OK;
571 	}
572 
573 	txq_map = skb_get_queue_mapping(skb) % enic->wq_count;
574 	wq = &enic->wq[txq_map];
575 	txq = netdev_get_tx_queue(netdev, txq_map);
576 
577 	/* Non-TSO sends must fit within ENIC_NON_TSO_MAX_DESC descs,
578 	 * which is very likely.  In the off chance it's going to take
579 	 * more than * ENIC_NON_TSO_MAX_DESC, linearize the skb.
580 	 */
581 
582 	if (skb_shinfo(skb)->gso_size == 0 &&
583 	    skb_shinfo(skb)->nr_frags + 1 > ENIC_NON_TSO_MAX_DESC &&
584 	    skb_linearize(skb)) {
585 		dev_kfree_skb_any(skb);
586 		return NETDEV_TX_OK;
587 	}
588 
589 	spin_lock(&enic->wq_lock[txq_map]);
590 
591 	if (vnic_wq_desc_avail(wq) <
592 	    skb_shinfo(skb)->nr_frags + ENIC_DESC_MAX_SPLITS) {
593 		netif_tx_stop_queue(txq);
594 		/* This is a hard error, log it */
595 		netdev_err(netdev, "BUG! Tx ring full when queue awake!\n");
596 		spin_unlock(&enic->wq_lock[txq_map]);
597 		return NETDEV_TX_BUSY;
598 	}
599 
600 	enic_queue_wq_skb(enic, wq, skb);
601 
602 	if (vnic_wq_desc_avail(wq) < MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS)
603 		netif_tx_stop_queue(txq);
604 	if (!skb->xmit_more || netif_xmit_stopped(txq))
605 		vnic_wq_doorbell(wq);
606 
607 	spin_unlock(&enic->wq_lock[txq_map]);
608 
609 	return NETDEV_TX_OK;
610 }
611 
612 /* dev_base_lock rwlock held, nominally process context */
613 static struct rtnl_link_stats64 *enic_get_stats(struct net_device *netdev,
614 						struct rtnl_link_stats64 *net_stats)
615 {
616 	struct enic *enic = netdev_priv(netdev);
617 	struct vnic_stats *stats;
618 	int err;
619 
620 	err = enic_dev_stats_dump(enic, &stats);
621 	/* return only when pci_zalloc_consistent fails in vnic_dev_stats_dump
622 	 * For other failures, like devcmd failure, we return previously
623 	 * recorded stats.
624 	 */
625 	if (err == -ENOMEM)
626 		return net_stats;
627 
628 	net_stats->tx_packets = stats->tx.tx_frames_ok;
629 	net_stats->tx_bytes = stats->tx.tx_bytes_ok;
630 	net_stats->tx_errors = stats->tx.tx_errors;
631 	net_stats->tx_dropped = stats->tx.tx_drops;
632 
633 	net_stats->rx_packets = stats->rx.rx_frames_ok;
634 	net_stats->rx_bytes = stats->rx.rx_bytes_ok;
635 	net_stats->rx_errors = stats->rx.rx_errors;
636 	net_stats->multicast = stats->rx.rx_multicast_frames_ok;
637 	net_stats->rx_over_errors = enic->rq_truncated_pkts;
638 	net_stats->rx_crc_errors = enic->rq_bad_fcs;
639 	net_stats->rx_dropped = stats->rx.rx_no_bufs + stats->rx.rx_drop;
640 
641 	return net_stats;
642 }
643 
644 static int enic_mc_sync(struct net_device *netdev, const u8 *mc_addr)
645 {
646 	struct enic *enic = netdev_priv(netdev);
647 
648 	if (enic->mc_count == ENIC_MULTICAST_PERFECT_FILTERS) {
649 		unsigned int mc_count = netdev_mc_count(netdev);
650 
651 		netdev_warn(netdev, "Registering only %d out of %d multicast addresses\n",
652 			    ENIC_MULTICAST_PERFECT_FILTERS, mc_count);
653 
654 		return -ENOSPC;
655 	}
656 
657 	enic_dev_add_addr(enic, mc_addr);
658 	enic->mc_count++;
659 
660 	return 0;
661 }
662 
663 static int enic_mc_unsync(struct net_device *netdev, const u8 *mc_addr)
664 {
665 	struct enic *enic = netdev_priv(netdev);
666 
667 	enic_dev_del_addr(enic, mc_addr);
668 	enic->mc_count--;
669 
670 	return 0;
671 }
672 
673 static int enic_uc_sync(struct net_device *netdev, const u8 *uc_addr)
674 {
675 	struct enic *enic = netdev_priv(netdev);
676 
677 	if (enic->uc_count == ENIC_UNICAST_PERFECT_FILTERS) {
678 		unsigned int uc_count = netdev_uc_count(netdev);
679 
680 		netdev_warn(netdev, "Registering only %d out of %d unicast addresses\n",
681 			    ENIC_UNICAST_PERFECT_FILTERS, uc_count);
682 
683 		return -ENOSPC;
684 	}
685 
686 	enic_dev_add_addr(enic, uc_addr);
687 	enic->uc_count++;
688 
689 	return 0;
690 }
691 
692 static int enic_uc_unsync(struct net_device *netdev, const u8 *uc_addr)
693 {
694 	struct enic *enic = netdev_priv(netdev);
695 
696 	enic_dev_del_addr(enic, uc_addr);
697 	enic->uc_count--;
698 
699 	return 0;
700 }
701 
702 void enic_reset_addr_lists(struct enic *enic)
703 {
704 	struct net_device *netdev = enic->netdev;
705 
706 	__dev_uc_unsync(netdev, NULL);
707 	__dev_mc_unsync(netdev, NULL);
708 
709 	enic->mc_count = 0;
710 	enic->uc_count = 0;
711 	enic->flags = 0;
712 }
713 
714 static int enic_set_mac_addr(struct net_device *netdev, char *addr)
715 {
716 	struct enic *enic = netdev_priv(netdev);
717 
718 	if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic)) {
719 		if (!is_valid_ether_addr(addr) && !is_zero_ether_addr(addr))
720 			return -EADDRNOTAVAIL;
721 	} else {
722 		if (!is_valid_ether_addr(addr))
723 			return -EADDRNOTAVAIL;
724 	}
725 
726 	memcpy(netdev->dev_addr, addr, netdev->addr_len);
727 
728 	return 0;
729 }
730 
731 static int enic_set_mac_address_dynamic(struct net_device *netdev, void *p)
732 {
733 	struct enic *enic = netdev_priv(netdev);
734 	struct sockaddr *saddr = p;
735 	char *addr = saddr->sa_data;
736 	int err;
737 
738 	if (netif_running(enic->netdev)) {
739 		err = enic_dev_del_station_addr(enic);
740 		if (err)
741 			return err;
742 	}
743 
744 	err = enic_set_mac_addr(netdev, addr);
745 	if (err)
746 		return err;
747 
748 	if (netif_running(enic->netdev)) {
749 		err = enic_dev_add_station_addr(enic);
750 		if (err)
751 			return err;
752 	}
753 
754 	return err;
755 }
756 
757 static int enic_set_mac_address(struct net_device *netdev, void *p)
758 {
759 	struct sockaddr *saddr = p;
760 	char *addr = saddr->sa_data;
761 	struct enic *enic = netdev_priv(netdev);
762 	int err;
763 
764 	err = enic_dev_del_station_addr(enic);
765 	if (err)
766 		return err;
767 
768 	err = enic_set_mac_addr(netdev, addr);
769 	if (err)
770 		return err;
771 
772 	return enic_dev_add_station_addr(enic);
773 }
774 
775 /* netif_tx_lock held, BHs disabled */
776 static void enic_set_rx_mode(struct net_device *netdev)
777 {
778 	struct enic *enic = netdev_priv(netdev);
779 	int directed = 1;
780 	int multicast = (netdev->flags & IFF_MULTICAST) ? 1 : 0;
781 	int broadcast = (netdev->flags & IFF_BROADCAST) ? 1 : 0;
782 	int promisc = (netdev->flags & IFF_PROMISC) ||
783 		netdev_uc_count(netdev) > ENIC_UNICAST_PERFECT_FILTERS;
784 	int allmulti = (netdev->flags & IFF_ALLMULTI) ||
785 		netdev_mc_count(netdev) > ENIC_MULTICAST_PERFECT_FILTERS;
786 	unsigned int flags = netdev->flags |
787 		(allmulti ? IFF_ALLMULTI : 0) |
788 		(promisc ? IFF_PROMISC : 0);
789 
790 	if (enic->flags != flags) {
791 		enic->flags = flags;
792 		enic_dev_packet_filter(enic, directed,
793 			multicast, broadcast, promisc, allmulti);
794 	}
795 
796 	if (!promisc) {
797 		__dev_uc_sync(netdev, enic_uc_sync, enic_uc_unsync);
798 		if (!allmulti)
799 			__dev_mc_sync(netdev, enic_mc_sync, enic_mc_unsync);
800 	}
801 }
802 
803 /* netif_tx_lock held, BHs disabled */
804 static void enic_tx_timeout(struct net_device *netdev)
805 {
806 	struct enic *enic = netdev_priv(netdev);
807 	schedule_work(&enic->reset);
808 }
809 
810 static int enic_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
811 {
812 	struct enic *enic = netdev_priv(netdev);
813 	struct enic_port_profile *pp;
814 	int err;
815 
816 	ENIC_PP_BY_INDEX(enic, vf, pp, &err);
817 	if (err)
818 		return err;
819 
820 	if (is_valid_ether_addr(mac) || is_zero_ether_addr(mac)) {
821 		if (vf == PORT_SELF_VF) {
822 			memcpy(pp->vf_mac, mac, ETH_ALEN);
823 			return 0;
824 		} else {
825 			/*
826 			 * For sriov vf's set the mac in hw
827 			 */
828 			ENIC_DEVCMD_PROXY_BY_INDEX(vf, err, enic,
829 				vnic_dev_set_mac_addr, mac);
830 			return enic_dev_status_to_errno(err);
831 		}
832 	} else
833 		return -EINVAL;
834 }
835 
836 static int enic_set_vf_port(struct net_device *netdev, int vf,
837 	struct nlattr *port[])
838 {
839 	struct enic *enic = netdev_priv(netdev);
840 	struct enic_port_profile prev_pp;
841 	struct enic_port_profile *pp;
842 	int err = 0, restore_pp = 1;
843 
844 	ENIC_PP_BY_INDEX(enic, vf, pp, &err);
845 	if (err)
846 		return err;
847 
848 	if (!port[IFLA_PORT_REQUEST])
849 		return -EOPNOTSUPP;
850 
851 	memcpy(&prev_pp, pp, sizeof(*enic->pp));
852 	memset(pp, 0, sizeof(*enic->pp));
853 
854 	pp->set |= ENIC_SET_REQUEST;
855 	pp->request = nla_get_u8(port[IFLA_PORT_REQUEST]);
856 
857 	if (port[IFLA_PORT_PROFILE]) {
858 		pp->set |= ENIC_SET_NAME;
859 		memcpy(pp->name, nla_data(port[IFLA_PORT_PROFILE]),
860 			PORT_PROFILE_MAX);
861 	}
862 
863 	if (port[IFLA_PORT_INSTANCE_UUID]) {
864 		pp->set |= ENIC_SET_INSTANCE;
865 		memcpy(pp->instance_uuid,
866 			nla_data(port[IFLA_PORT_INSTANCE_UUID]), PORT_UUID_MAX);
867 	}
868 
869 	if (port[IFLA_PORT_HOST_UUID]) {
870 		pp->set |= ENIC_SET_HOST;
871 		memcpy(pp->host_uuid,
872 			nla_data(port[IFLA_PORT_HOST_UUID]), PORT_UUID_MAX);
873 	}
874 
875 	if (vf == PORT_SELF_VF) {
876 		/* Special case handling: mac came from IFLA_VF_MAC */
877 		if (!is_zero_ether_addr(prev_pp.vf_mac))
878 			memcpy(pp->mac_addr, prev_pp.vf_mac, ETH_ALEN);
879 
880 		if (is_zero_ether_addr(netdev->dev_addr))
881 			eth_hw_addr_random(netdev);
882 	} else {
883 		/* SR-IOV VF: get mac from adapter */
884 		ENIC_DEVCMD_PROXY_BY_INDEX(vf, err, enic,
885 			vnic_dev_get_mac_addr, pp->mac_addr);
886 		if (err) {
887 			netdev_err(netdev, "Error getting mac for vf %d\n", vf);
888 			memcpy(pp, &prev_pp, sizeof(*pp));
889 			return enic_dev_status_to_errno(err);
890 		}
891 	}
892 
893 	err = enic_process_set_pp_request(enic, vf, &prev_pp, &restore_pp);
894 	if (err) {
895 		if (restore_pp) {
896 			/* Things are still the way they were: Implicit
897 			 * DISASSOCIATE failed
898 			 */
899 			memcpy(pp, &prev_pp, sizeof(*pp));
900 		} else {
901 			memset(pp, 0, sizeof(*pp));
902 			if (vf == PORT_SELF_VF)
903 				eth_zero_addr(netdev->dev_addr);
904 		}
905 	} else {
906 		/* Set flag to indicate that the port assoc/disassoc
907 		 * request has been sent out to fw
908 		 */
909 		pp->set |= ENIC_PORT_REQUEST_APPLIED;
910 
911 		/* If DISASSOCIATE, clean up all assigned/saved macaddresses */
912 		if (pp->request == PORT_REQUEST_DISASSOCIATE) {
913 			eth_zero_addr(pp->mac_addr);
914 			if (vf == PORT_SELF_VF)
915 				eth_zero_addr(netdev->dev_addr);
916 		}
917 	}
918 
919 	if (vf == PORT_SELF_VF)
920 		eth_zero_addr(pp->vf_mac);
921 
922 	return err;
923 }
924 
925 static int enic_get_vf_port(struct net_device *netdev, int vf,
926 	struct sk_buff *skb)
927 {
928 	struct enic *enic = netdev_priv(netdev);
929 	u16 response = PORT_PROFILE_RESPONSE_SUCCESS;
930 	struct enic_port_profile *pp;
931 	int err;
932 
933 	ENIC_PP_BY_INDEX(enic, vf, pp, &err);
934 	if (err)
935 		return err;
936 
937 	if (!(pp->set & ENIC_PORT_REQUEST_APPLIED))
938 		return -ENODATA;
939 
940 	err = enic_process_get_pp_request(enic, vf, pp->request, &response);
941 	if (err)
942 		return err;
943 
944 	if (nla_put_u16(skb, IFLA_PORT_REQUEST, pp->request) ||
945 	    nla_put_u16(skb, IFLA_PORT_RESPONSE, response) ||
946 	    ((pp->set & ENIC_SET_NAME) &&
947 	     nla_put(skb, IFLA_PORT_PROFILE, PORT_PROFILE_MAX, pp->name)) ||
948 	    ((pp->set & ENIC_SET_INSTANCE) &&
949 	     nla_put(skb, IFLA_PORT_INSTANCE_UUID, PORT_UUID_MAX,
950 		     pp->instance_uuid)) ||
951 	    ((pp->set & ENIC_SET_HOST) &&
952 	     nla_put(skb, IFLA_PORT_HOST_UUID, PORT_UUID_MAX, pp->host_uuid)))
953 		goto nla_put_failure;
954 	return 0;
955 
956 nla_put_failure:
957 	return -EMSGSIZE;
958 }
959 
960 static void enic_free_rq_buf(struct vnic_rq *rq, struct vnic_rq_buf *buf)
961 {
962 	struct enic *enic = vnic_dev_priv(rq->vdev);
963 
964 	if (!buf->os_buf)
965 		return;
966 
967 	pci_unmap_single(enic->pdev, buf->dma_addr,
968 		buf->len, PCI_DMA_FROMDEVICE);
969 	dev_kfree_skb_any(buf->os_buf);
970 	buf->os_buf = NULL;
971 }
972 
973 static int enic_rq_alloc_buf(struct vnic_rq *rq)
974 {
975 	struct enic *enic = vnic_dev_priv(rq->vdev);
976 	struct net_device *netdev = enic->netdev;
977 	struct sk_buff *skb;
978 	unsigned int len = netdev->mtu + VLAN_ETH_HLEN;
979 	unsigned int os_buf_index = 0;
980 	dma_addr_t dma_addr;
981 	struct vnic_rq_buf *buf = rq->to_use;
982 
983 	if (buf->os_buf) {
984 		enic_queue_rq_desc(rq, buf->os_buf, os_buf_index, buf->dma_addr,
985 				   buf->len);
986 
987 		return 0;
988 	}
989 	skb = netdev_alloc_skb_ip_align(netdev, len);
990 	if (!skb)
991 		return -ENOMEM;
992 
993 	dma_addr = pci_map_single(enic->pdev, skb->data, len,
994 				  PCI_DMA_FROMDEVICE);
995 	if (unlikely(enic_dma_map_check(enic, dma_addr))) {
996 		dev_kfree_skb(skb);
997 		return -ENOMEM;
998 	}
999 
1000 	enic_queue_rq_desc(rq, skb, os_buf_index,
1001 		dma_addr, len);
1002 
1003 	return 0;
1004 }
1005 
1006 static void enic_intr_update_pkt_size(struct vnic_rx_bytes_counter *pkt_size,
1007 				      u32 pkt_len)
1008 {
1009 	if (ENIC_LARGE_PKT_THRESHOLD <= pkt_len)
1010 		pkt_size->large_pkt_bytes_cnt += pkt_len;
1011 	else
1012 		pkt_size->small_pkt_bytes_cnt += pkt_len;
1013 }
1014 
1015 static bool enic_rxcopybreak(struct net_device *netdev, struct sk_buff **skb,
1016 			     struct vnic_rq_buf *buf, u16 len)
1017 {
1018 	struct enic *enic = netdev_priv(netdev);
1019 	struct sk_buff *new_skb;
1020 
1021 	if (len > enic->rx_copybreak)
1022 		return false;
1023 	new_skb = netdev_alloc_skb_ip_align(netdev, len);
1024 	if (!new_skb)
1025 		return false;
1026 	pci_dma_sync_single_for_cpu(enic->pdev, buf->dma_addr, len,
1027 				    DMA_FROM_DEVICE);
1028 	memcpy(new_skb->data, (*skb)->data, len);
1029 	*skb = new_skb;
1030 
1031 	return true;
1032 }
1033 
1034 static void enic_rq_indicate_buf(struct vnic_rq *rq,
1035 	struct cq_desc *cq_desc, struct vnic_rq_buf *buf,
1036 	int skipped, void *opaque)
1037 {
1038 	struct enic *enic = vnic_dev_priv(rq->vdev);
1039 	struct net_device *netdev = enic->netdev;
1040 	struct sk_buff *skb;
1041 	struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
1042 
1043 	u8 type, color, eop, sop, ingress_port, vlan_stripped;
1044 	u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof;
1045 	u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok;
1046 	u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc;
1047 	u8 packet_error;
1048 	u16 q_number, completed_index, bytes_written, vlan_tci, checksum;
1049 	u32 rss_hash;
1050 
1051 	if (skipped)
1052 		return;
1053 
1054 	skb = buf->os_buf;
1055 
1056 	cq_enet_rq_desc_dec((struct cq_enet_rq_desc *)cq_desc,
1057 		&type, &color, &q_number, &completed_index,
1058 		&ingress_port, &fcoe, &eop, &sop, &rss_type,
1059 		&csum_not_calc, &rss_hash, &bytes_written,
1060 		&packet_error, &vlan_stripped, &vlan_tci, &checksum,
1061 		&fcoe_sof, &fcoe_fc_crc_ok, &fcoe_enc_error,
1062 		&fcoe_eof, &tcp_udp_csum_ok, &udp, &tcp,
1063 		&ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment,
1064 		&fcs_ok);
1065 
1066 	if (packet_error) {
1067 
1068 		if (!fcs_ok) {
1069 			if (bytes_written > 0)
1070 				enic->rq_bad_fcs++;
1071 			else if (bytes_written == 0)
1072 				enic->rq_truncated_pkts++;
1073 		}
1074 
1075 		pci_unmap_single(enic->pdev, buf->dma_addr, buf->len,
1076 				 PCI_DMA_FROMDEVICE);
1077 		dev_kfree_skb_any(skb);
1078 		buf->os_buf = NULL;
1079 
1080 		return;
1081 	}
1082 
1083 	if (eop && bytes_written > 0) {
1084 
1085 		/* Good receive
1086 		 */
1087 
1088 		if (!enic_rxcopybreak(netdev, &skb, buf, bytes_written)) {
1089 			buf->os_buf = NULL;
1090 			pci_unmap_single(enic->pdev, buf->dma_addr, buf->len,
1091 					 PCI_DMA_FROMDEVICE);
1092 		}
1093 		prefetch(skb->data - NET_IP_ALIGN);
1094 
1095 		skb_put(skb, bytes_written);
1096 		skb->protocol = eth_type_trans(skb, netdev);
1097 		skb_record_rx_queue(skb, q_number);
1098 		if (netdev->features & NETIF_F_RXHASH) {
1099 			skb_set_hash(skb, rss_hash,
1100 				     (rss_type &
1101 				      (NIC_CFG_RSS_HASH_TYPE_TCP_IPV6_EX |
1102 				       NIC_CFG_RSS_HASH_TYPE_TCP_IPV6 |
1103 				       NIC_CFG_RSS_HASH_TYPE_TCP_IPV4)) ?
1104 				     PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3);
1105 		}
1106 
1107 		/* Hardware does not provide whole packet checksum. It only
1108 		 * provides pseudo checksum. Since hw validates the packet
1109 		 * checksum but not provide us the checksum value. use
1110 		 * CHECSUM_UNNECESSARY.
1111 		 */
1112 		if ((netdev->features & NETIF_F_RXCSUM) && tcp_udp_csum_ok &&
1113 		    ipv4_csum_ok)
1114 			skb->ip_summed = CHECKSUM_UNNECESSARY;
1115 
1116 		if (vlan_stripped)
1117 			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tci);
1118 
1119 		skb_mark_napi_id(skb, &enic->napi[rq->index]);
1120 		if (enic_poll_busy_polling(rq) ||
1121 		    !(netdev->features & NETIF_F_GRO))
1122 			netif_receive_skb(skb);
1123 		else
1124 			napi_gro_receive(&enic->napi[q_number], skb);
1125 		if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1126 			enic_intr_update_pkt_size(&cq->pkt_size_counter,
1127 						  bytes_written);
1128 	} else {
1129 
1130 		/* Buffer overflow
1131 		 */
1132 
1133 		pci_unmap_single(enic->pdev, buf->dma_addr, buf->len,
1134 				 PCI_DMA_FROMDEVICE);
1135 		dev_kfree_skb_any(skb);
1136 		buf->os_buf = NULL;
1137 	}
1138 }
1139 
1140 static int enic_rq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
1141 	u8 type, u16 q_number, u16 completed_index, void *opaque)
1142 {
1143 	struct enic *enic = vnic_dev_priv(vdev);
1144 
1145 	vnic_rq_service(&enic->rq[q_number], cq_desc,
1146 		completed_index, VNIC_RQ_RETURN_DESC,
1147 		enic_rq_indicate_buf, opaque);
1148 
1149 	return 0;
1150 }
1151 
1152 static int enic_poll(struct napi_struct *napi, int budget)
1153 {
1154 	struct net_device *netdev = napi->dev;
1155 	struct enic *enic = netdev_priv(netdev);
1156 	unsigned int cq_rq = enic_cq_rq(enic, 0);
1157 	unsigned int cq_wq = enic_cq_wq(enic, 0);
1158 	unsigned int intr = enic_legacy_io_intr();
1159 	unsigned int rq_work_to_do = budget;
1160 	unsigned int wq_work_to_do = -1; /* no limit */
1161 	unsigned int  work_done, rq_work_done = 0, wq_work_done;
1162 	int err;
1163 
1164 	wq_work_done = vnic_cq_service(&enic->cq[cq_wq], wq_work_to_do,
1165 				       enic_wq_service, NULL);
1166 
1167 	if (!enic_poll_lock_napi(&enic->rq[cq_rq])) {
1168 		if (wq_work_done > 0)
1169 			vnic_intr_return_credits(&enic->intr[intr],
1170 						 wq_work_done,
1171 						 0 /* dont unmask intr */,
1172 						 0 /* dont reset intr timer */);
1173 		return budget;
1174 	}
1175 
1176 	if (budget > 0)
1177 		rq_work_done = vnic_cq_service(&enic->cq[cq_rq],
1178 			rq_work_to_do, enic_rq_service, NULL);
1179 
1180 	/* Accumulate intr event credits for this polling
1181 	 * cycle.  An intr event is the completion of a
1182 	 * a WQ or RQ packet.
1183 	 */
1184 
1185 	work_done = rq_work_done + wq_work_done;
1186 
1187 	if (work_done > 0)
1188 		vnic_intr_return_credits(&enic->intr[intr],
1189 			work_done,
1190 			0 /* don't unmask intr */,
1191 			0 /* don't reset intr timer */);
1192 
1193 	err = vnic_rq_fill(&enic->rq[0], enic_rq_alloc_buf);
1194 	enic_poll_unlock_napi(&enic->rq[cq_rq], napi);
1195 
1196 	/* Buffer allocation failed. Stay in polling
1197 	 * mode so we can try to fill the ring again.
1198 	 */
1199 
1200 	if (err)
1201 		rq_work_done = rq_work_to_do;
1202 
1203 	if (rq_work_done < rq_work_to_do) {
1204 
1205 		/* Some work done, but not enough to stay in polling,
1206 		 * exit polling
1207 		 */
1208 
1209 		napi_complete(napi);
1210 		vnic_intr_unmask(&enic->intr[intr]);
1211 	}
1212 
1213 	return rq_work_done;
1214 }
1215 
1216 static void enic_set_int_moderation(struct enic *enic, struct vnic_rq *rq)
1217 {
1218 	unsigned int intr = enic_msix_rq_intr(enic, rq->index);
1219 	struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
1220 	u32 timer = cq->tobe_rx_coal_timeval;
1221 
1222 	if (cq->tobe_rx_coal_timeval != cq->cur_rx_coal_timeval) {
1223 		vnic_intr_coalescing_timer_set(&enic->intr[intr], timer);
1224 		cq->cur_rx_coal_timeval = cq->tobe_rx_coal_timeval;
1225 	}
1226 }
1227 
1228 static void enic_calc_int_moderation(struct enic *enic, struct vnic_rq *rq)
1229 {
1230 	struct enic_rx_coal *rx_coal = &enic->rx_coalesce_setting;
1231 	struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
1232 	struct vnic_rx_bytes_counter *pkt_size_counter = &cq->pkt_size_counter;
1233 	int index;
1234 	u32 timer;
1235 	u32 range_start;
1236 	u32 traffic;
1237 	u64 delta;
1238 	ktime_t now = ktime_get();
1239 
1240 	delta = ktime_us_delta(now, cq->prev_ts);
1241 	if (delta < ENIC_AIC_TS_BREAK)
1242 		return;
1243 	cq->prev_ts = now;
1244 
1245 	traffic = pkt_size_counter->large_pkt_bytes_cnt +
1246 		  pkt_size_counter->small_pkt_bytes_cnt;
1247 	/* The table takes Mbps
1248 	 * traffic *= 8    => bits
1249 	 * traffic *= (10^6 / delta)    => bps
1250 	 * traffic /= 10^6     => Mbps
1251 	 *
1252 	 * Combining, traffic *= (8 / delta)
1253 	 */
1254 
1255 	traffic <<= 3;
1256 	traffic = delta > UINT_MAX ? 0 : traffic / (u32)delta;
1257 
1258 	for (index = 0; index < ENIC_MAX_COALESCE_TIMERS; index++)
1259 		if (traffic < mod_table[index].rx_rate)
1260 			break;
1261 	range_start = (pkt_size_counter->small_pkt_bytes_cnt >
1262 		       pkt_size_counter->large_pkt_bytes_cnt << 1) ?
1263 		      rx_coal->small_pkt_range_start :
1264 		      rx_coal->large_pkt_range_start;
1265 	timer = range_start + ((rx_coal->range_end - range_start) *
1266 			       mod_table[index].range_percent / 100);
1267 	/* Damping */
1268 	cq->tobe_rx_coal_timeval = (timer + cq->tobe_rx_coal_timeval) >> 1;
1269 
1270 	pkt_size_counter->large_pkt_bytes_cnt = 0;
1271 	pkt_size_counter->small_pkt_bytes_cnt = 0;
1272 }
1273 
1274 #ifdef CONFIG_RFS_ACCEL
1275 static void enic_free_rx_cpu_rmap(struct enic *enic)
1276 {
1277 	free_irq_cpu_rmap(enic->netdev->rx_cpu_rmap);
1278 	enic->netdev->rx_cpu_rmap = NULL;
1279 }
1280 
1281 static void enic_set_rx_cpu_rmap(struct enic *enic)
1282 {
1283 	int i, res;
1284 
1285 	if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX) {
1286 		enic->netdev->rx_cpu_rmap = alloc_irq_cpu_rmap(enic->rq_count);
1287 		if (unlikely(!enic->netdev->rx_cpu_rmap))
1288 			return;
1289 		for (i = 0; i < enic->rq_count; i++) {
1290 			res = irq_cpu_rmap_add(enic->netdev->rx_cpu_rmap,
1291 					       enic->msix_entry[i].vector);
1292 			if (unlikely(res)) {
1293 				enic_free_rx_cpu_rmap(enic);
1294 				return;
1295 			}
1296 		}
1297 	}
1298 }
1299 
1300 #else
1301 
1302 static void enic_free_rx_cpu_rmap(struct enic *enic)
1303 {
1304 }
1305 
1306 static void enic_set_rx_cpu_rmap(struct enic *enic)
1307 {
1308 }
1309 
1310 #endif /* CONFIG_RFS_ACCEL */
1311 
1312 #ifdef CONFIG_NET_RX_BUSY_POLL
1313 static int enic_busy_poll(struct napi_struct *napi)
1314 {
1315 	struct net_device *netdev = napi->dev;
1316 	struct enic *enic = netdev_priv(netdev);
1317 	unsigned int rq = (napi - &enic->napi[0]);
1318 	unsigned int cq = enic_cq_rq(enic, rq);
1319 	unsigned int intr = enic_msix_rq_intr(enic, rq);
1320 	unsigned int work_to_do = -1; /* clean all pkts possible */
1321 	unsigned int work_done;
1322 
1323 	if (!enic_poll_lock_poll(&enic->rq[rq]))
1324 		return LL_FLUSH_BUSY;
1325 	work_done = vnic_cq_service(&enic->cq[cq], work_to_do,
1326 				    enic_rq_service, NULL);
1327 
1328 	if (work_done > 0)
1329 		vnic_intr_return_credits(&enic->intr[intr],
1330 					 work_done, 0, 0);
1331 	vnic_rq_fill(&enic->rq[rq], enic_rq_alloc_buf);
1332 	if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1333 		enic_calc_int_moderation(enic, &enic->rq[rq]);
1334 	enic_poll_unlock_poll(&enic->rq[rq]);
1335 
1336 	return work_done;
1337 }
1338 #endif /* CONFIG_NET_RX_BUSY_POLL */
1339 
1340 static int enic_poll_msix_wq(struct napi_struct *napi, int budget)
1341 {
1342 	struct net_device *netdev = napi->dev;
1343 	struct enic *enic = netdev_priv(netdev);
1344 	unsigned int wq_index = (napi - &enic->napi[0]) - enic->rq_count;
1345 	struct vnic_wq *wq = &enic->wq[wq_index];
1346 	unsigned int cq;
1347 	unsigned int intr;
1348 	unsigned int wq_work_to_do = -1; /* clean all desc possible */
1349 	unsigned int wq_work_done;
1350 	unsigned int wq_irq;
1351 
1352 	wq_irq = wq->index;
1353 	cq = enic_cq_wq(enic, wq_irq);
1354 	intr = enic_msix_wq_intr(enic, wq_irq);
1355 	wq_work_done = vnic_cq_service(&enic->cq[cq], wq_work_to_do,
1356 				       enic_wq_service, NULL);
1357 
1358 	vnic_intr_return_credits(&enic->intr[intr], wq_work_done,
1359 				 0 /* don't unmask intr */,
1360 				 1 /* reset intr timer */);
1361 	if (!wq_work_done) {
1362 		napi_complete(napi);
1363 		vnic_intr_unmask(&enic->intr[intr]);
1364 		return 0;
1365 	}
1366 
1367 	return budget;
1368 }
1369 
1370 static int enic_poll_msix_rq(struct napi_struct *napi, int budget)
1371 {
1372 	struct net_device *netdev = napi->dev;
1373 	struct enic *enic = netdev_priv(netdev);
1374 	unsigned int rq = (napi - &enic->napi[0]);
1375 	unsigned int cq = enic_cq_rq(enic, rq);
1376 	unsigned int intr = enic_msix_rq_intr(enic, rq);
1377 	unsigned int work_to_do = budget;
1378 	unsigned int work_done = 0;
1379 	int err;
1380 
1381 	if (!enic_poll_lock_napi(&enic->rq[rq]))
1382 		return budget;
1383 	/* Service RQ
1384 	 */
1385 
1386 	if (budget > 0)
1387 		work_done = vnic_cq_service(&enic->cq[cq],
1388 			work_to_do, enic_rq_service, NULL);
1389 
1390 	/* Return intr event credits for this polling
1391 	 * cycle.  An intr event is the completion of a
1392 	 * RQ packet.
1393 	 */
1394 
1395 	if (work_done > 0)
1396 		vnic_intr_return_credits(&enic->intr[intr],
1397 			work_done,
1398 			0 /* don't unmask intr */,
1399 			0 /* don't reset intr timer */);
1400 
1401 	err = vnic_rq_fill(&enic->rq[rq], enic_rq_alloc_buf);
1402 
1403 	/* Buffer allocation failed. Stay in polling mode
1404 	 * so we can try to fill the ring again.
1405 	 */
1406 
1407 	if (err)
1408 		work_done = work_to_do;
1409 	if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1410 		/* Call the function which refreshes
1411 		 * the intr coalescing timer value based on
1412 		 * the traffic.  This is supported only in
1413 		 * the case of MSI-x mode
1414 		 */
1415 		enic_calc_int_moderation(enic, &enic->rq[rq]);
1416 
1417 	enic_poll_unlock_napi(&enic->rq[rq], napi);
1418 	if (work_done < work_to_do) {
1419 
1420 		/* Some work done, but not enough to stay in polling,
1421 		 * exit polling
1422 		 */
1423 
1424 		napi_complete(napi);
1425 		if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1426 			enic_set_int_moderation(enic, &enic->rq[rq]);
1427 		vnic_intr_unmask(&enic->intr[intr]);
1428 	}
1429 
1430 	return work_done;
1431 }
1432 
1433 static void enic_notify_timer(unsigned long data)
1434 {
1435 	struct enic *enic = (struct enic *)data;
1436 
1437 	enic_notify_check(enic);
1438 
1439 	mod_timer(&enic->notify_timer,
1440 		round_jiffies(jiffies + ENIC_NOTIFY_TIMER_PERIOD));
1441 }
1442 
1443 static void enic_free_intr(struct enic *enic)
1444 {
1445 	struct net_device *netdev = enic->netdev;
1446 	unsigned int i;
1447 
1448 	enic_free_rx_cpu_rmap(enic);
1449 	switch (vnic_dev_get_intr_mode(enic->vdev)) {
1450 	case VNIC_DEV_INTR_MODE_INTX:
1451 		free_irq(enic->pdev->irq, netdev);
1452 		break;
1453 	case VNIC_DEV_INTR_MODE_MSI:
1454 		free_irq(enic->pdev->irq, enic);
1455 		break;
1456 	case VNIC_DEV_INTR_MODE_MSIX:
1457 		for (i = 0; i < ARRAY_SIZE(enic->msix); i++)
1458 			if (enic->msix[i].requested)
1459 				free_irq(enic->msix_entry[i].vector,
1460 					enic->msix[i].devid);
1461 		break;
1462 	default:
1463 		break;
1464 	}
1465 }
1466 
1467 static int enic_request_intr(struct enic *enic)
1468 {
1469 	struct net_device *netdev = enic->netdev;
1470 	unsigned int i, intr;
1471 	int err = 0;
1472 
1473 	enic_set_rx_cpu_rmap(enic);
1474 	switch (vnic_dev_get_intr_mode(enic->vdev)) {
1475 
1476 	case VNIC_DEV_INTR_MODE_INTX:
1477 
1478 		err = request_irq(enic->pdev->irq, enic_isr_legacy,
1479 			IRQF_SHARED, netdev->name, netdev);
1480 		break;
1481 
1482 	case VNIC_DEV_INTR_MODE_MSI:
1483 
1484 		err = request_irq(enic->pdev->irq, enic_isr_msi,
1485 			0, netdev->name, enic);
1486 		break;
1487 
1488 	case VNIC_DEV_INTR_MODE_MSIX:
1489 
1490 		for (i = 0; i < enic->rq_count; i++) {
1491 			intr = enic_msix_rq_intr(enic, i);
1492 			snprintf(enic->msix[intr].devname,
1493 				sizeof(enic->msix[intr].devname),
1494 				"%.11s-rx-%d", netdev->name, i);
1495 			enic->msix[intr].isr = enic_isr_msix;
1496 			enic->msix[intr].devid = &enic->napi[i];
1497 		}
1498 
1499 		for (i = 0; i < enic->wq_count; i++) {
1500 			int wq = enic_cq_wq(enic, i);
1501 
1502 			intr = enic_msix_wq_intr(enic, i);
1503 			snprintf(enic->msix[intr].devname,
1504 				sizeof(enic->msix[intr].devname),
1505 				"%.11s-tx-%d", netdev->name, i);
1506 			enic->msix[intr].isr = enic_isr_msix;
1507 			enic->msix[intr].devid = &enic->napi[wq];
1508 		}
1509 
1510 		intr = enic_msix_err_intr(enic);
1511 		snprintf(enic->msix[intr].devname,
1512 			sizeof(enic->msix[intr].devname),
1513 			"%.11s-err", netdev->name);
1514 		enic->msix[intr].isr = enic_isr_msix_err;
1515 		enic->msix[intr].devid = enic;
1516 
1517 		intr = enic_msix_notify_intr(enic);
1518 		snprintf(enic->msix[intr].devname,
1519 			sizeof(enic->msix[intr].devname),
1520 			"%.11s-notify", netdev->name);
1521 		enic->msix[intr].isr = enic_isr_msix_notify;
1522 		enic->msix[intr].devid = enic;
1523 
1524 		for (i = 0; i < ARRAY_SIZE(enic->msix); i++)
1525 			enic->msix[i].requested = 0;
1526 
1527 		for (i = 0; i < enic->intr_count; i++) {
1528 			err = request_irq(enic->msix_entry[i].vector,
1529 				enic->msix[i].isr, 0,
1530 				enic->msix[i].devname,
1531 				enic->msix[i].devid);
1532 			if (err) {
1533 				enic_free_intr(enic);
1534 				break;
1535 			}
1536 			enic->msix[i].requested = 1;
1537 		}
1538 
1539 		break;
1540 
1541 	default:
1542 		break;
1543 	}
1544 
1545 	return err;
1546 }
1547 
1548 static void enic_synchronize_irqs(struct enic *enic)
1549 {
1550 	unsigned int i;
1551 
1552 	switch (vnic_dev_get_intr_mode(enic->vdev)) {
1553 	case VNIC_DEV_INTR_MODE_INTX:
1554 	case VNIC_DEV_INTR_MODE_MSI:
1555 		synchronize_irq(enic->pdev->irq);
1556 		break;
1557 	case VNIC_DEV_INTR_MODE_MSIX:
1558 		for (i = 0; i < enic->intr_count; i++)
1559 			synchronize_irq(enic->msix_entry[i].vector);
1560 		break;
1561 	default:
1562 		break;
1563 	}
1564 }
1565 
1566 static void enic_set_rx_coal_setting(struct enic *enic)
1567 {
1568 	unsigned int speed;
1569 	int index = -1;
1570 	struct enic_rx_coal *rx_coal = &enic->rx_coalesce_setting;
1571 
1572 	/* If intr mode is not MSIX, do not do adaptive coalescing */
1573 	if (VNIC_DEV_INTR_MODE_MSIX != vnic_dev_get_intr_mode(enic->vdev)) {
1574 		netdev_info(enic->netdev, "INTR mode is not MSIX, Not initializing adaptive coalescing");
1575 		return;
1576 	}
1577 
1578 	/* 1. Read the link speed from fw
1579 	 * 2. Pick the default range for the speed
1580 	 * 3. Update it in enic->rx_coalesce_setting
1581 	 */
1582 	speed = vnic_dev_port_speed(enic->vdev);
1583 	if (ENIC_LINK_SPEED_10G < speed)
1584 		index = ENIC_LINK_40G_INDEX;
1585 	else if (ENIC_LINK_SPEED_4G < speed)
1586 		index = ENIC_LINK_10G_INDEX;
1587 	else
1588 		index = ENIC_LINK_4G_INDEX;
1589 
1590 	rx_coal->small_pkt_range_start = mod_range[index].small_pkt_range_start;
1591 	rx_coal->large_pkt_range_start = mod_range[index].large_pkt_range_start;
1592 	rx_coal->range_end = ENIC_RX_COALESCE_RANGE_END;
1593 
1594 	/* Start with the value provided by UCSM */
1595 	for (index = 0; index < enic->rq_count; index++)
1596 		enic->cq[index].cur_rx_coal_timeval =
1597 				enic->config.intr_timer_usec;
1598 
1599 	rx_coal->use_adaptive_rx_coalesce = 1;
1600 }
1601 
1602 static int enic_dev_notify_set(struct enic *enic)
1603 {
1604 	int err;
1605 
1606 	spin_lock_bh(&enic->devcmd_lock);
1607 	switch (vnic_dev_get_intr_mode(enic->vdev)) {
1608 	case VNIC_DEV_INTR_MODE_INTX:
1609 		err = vnic_dev_notify_set(enic->vdev,
1610 			enic_legacy_notify_intr());
1611 		break;
1612 	case VNIC_DEV_INTR_MODE_MSIX:
1613 		err = vnic_dev_notify_set(enic->vdev,
1614 			enic_msix_notify_intr(enic));
1615 		break;
1616 	default:
1617 		err = vnic_dev_notify_set(enic->vdev, -1 /* no intr */);
1618 		break;
1619 	}
1620 	spin_unlock_bh(&enic->devcmd_lock);
1621 
1622 	return err;
1623 }
1624 
1625 static void enic_notify_timer_start(struct enic *enic)
1626 {
1627 	switch (vnic_dev_get_intr_mode(enic->vdev)) {
1628 	case VNIC_DEV_INTR_MODE_MSI:
1629 		mod_timer(&enic->notify_timer, jiffies);
1630 		break;
1631 	default:
1632 		/* Using intr for notification for INTx/MSI-X */
1633 		break;
1634 	}
1635 }
1636 
1637 /* rtnl lock is held, process context */
1638 static int enic_open(struct net_device *netdev)
1639 {
1640 	struct enic *enic = netdev_priv(netdev);
1641 	unsigned int i;
1642 	int err;
1643 
1644 	err = enic_request_intr(enic);
1645 	if (err) {
1646 		netdev_err(netdev, "Unable to request irq.\n");
1647 		return err;
1648 	}
1649 
1650 	err = enic_dev_notify_set(enic);
1651 	if (err) {
1652 		netdev_err(netdev,
1653 			"Failed to alloc notify buffer, aborting.\n");
1654 		goto err_out_free_intr;
1655 	}
1656 
1657 	for (i = 0; i < enic->rq_count; i++) {
1658 		vnic_rq_fill(&enic->rq[i], enic_rq_alloc_buf);
1659 		/* Need at least one buffer on ring to get going */
1660 		if (vnic_rq_desc_used(&enic->rq[i]) == 0) {
1661 			netdev_err(netdev, "Unable to alloc receive buffers\n");
1662 			err = -ENOMEM;
1663 			goto err_out_free_rq;
1664 		}
1665 	}
1666 
1667 	for (i = 0; i < enic->wq_count; i++)
1668 		vnic_wq_enable(&enic->wq[i]);
1669 	for (i = 0; i < enic->rq_count; i++)
1670 		vnic_rq_enable(&enic->rq[i]);
1671 
1672 	if (!enic_is_dynamic(enic) && !enic_is_sriov_vf(enic))
1673 		enic_dev_add_station_addr(enic);
1674 
1675 	enic_set_rx_mode(netdev);
1676 
1677 	netif_tx_wake_all_queues(netdev);
1678 
1679 	for (i = 0; i < enic->rq_count; i++) {
1680 		enic_busy_poll_init_lock(&enic->rq[i]);
1681 		napi_enable(&enic->napi[i]);
1682 	}
1683 	if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX)
1684 		for (i = 0; i < enic->wq_count; i++)
1685 			napi_enable(&enic->napi[enic_cq_wq(enic, i)]);
1686 	enic_dev_enable(enic);
1687 
1688 	for (i = 0; i < enic->intr_count; i++)
1689 		vnic_intr_unmask(&enic->intr[i]);
1690 
1691 	enic_notify_timer_start(enic);
1692 	enic_rfs_flw_tbl_init(enic);
1693 
1694 	return 0;
1695 
1696 err_out_free_rq:
1697 	for (i = 0; i < enic->rq_count; i++)
1698 		vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
1699 	enic_dev_notify_unset(enic);
1700 err_out_free_intr:
1701 	enic_free_intr(enic);
1702 
1703 	return err;
1704 }
1705 
1706 /* rtnl lock is held, process context */
1707 static int enic_stop(struct net_device *netdev)
1708 {
1709 	struct enic *enic = netdev_priv(netdev);
1710 	unsigned int i;
1711 	int err;
1712 
1713 	for (i = 0; i < enic->intr_count; i++) {
1714 		vnic_intr_mask(&enic->intr[i]);
1715 		(void)vnic_intr_masked(&enic->intr[i]); /* flush write */
1716 	}
1717 
1718 	enic_synchronize_irqs(enic);
1719 
1720 	del_timer_sync(&enic->notify_timer);
1721 	enic_rfs_flw_tbl_free(enic);
1722 
1723 	enic_dev_disable(enic);
1724 
1725 	for (i = 0; i < enic->rq_count; i++) {
1726 		napi_disable(&enic->napi[i]);
1727 		local_bh_disable();
1728 		while (!enic_poll_lock_napi(&enic->rq[i]))
1729 			mdelay(1);
1730 		local_bh_enable();
1731 	}
1732 
1733 	netif_carrier_off(netdev);
1734 	netif_tx_disable(netdev);
1735 	if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX)
1736 		for (i = 0; i < enic->wq_count; i++)
1737 			napi_disable(&enic->napi[enic_cq_wq(enic, i)]);
1738 
1739 	if (!enic_is_dynamic(enic) && !enic_is_sriov_vf(enic))
1740 		enic_dev_del_station_addr(enic);
1741 
1742 	for (i = 0; i < enic->wq_count; i++) {
1743 		err = vnic_wq_disable(&enic->wq[i]);
1744 		if (err)
1745 			return err;
1746 	}
1747 	for (i = 0; i < enic->rq_count; i++) {
1748 		err = vnic_rq_disable(&enic->rq[i]);
1749 		if (err)
1750 			return err;
1751 	}
1752 
1753 	enic_dev_notify_unset(enic);
1754 	enic_free_intr(enic);
1755 
1756 	for (i = 0; i < enic->wq_count; i++)
1757 		vnic_wq_clean(&enic->wq[i], enic_free_wq_buf);
1758 	for (i = 0; i < enic->rq_count; i++)
1759 		vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
1760 	for (i = 0; i < enic->cq_count; i++)
1761 		vnic_cq_clean(&enic->cq[i]);
1762 	for (i = 0; i < enic->intr_count; i++)
1763 		vnic_intr_clean(&enic->intr[i]);
1764 
1765 	return 0;
1766 }
1767 
1768 static int enic_change_mtu(struct net_device *netdev, int new_mtu)
1769 {
1770 	struct enic *enic = netdev_priv(netdev);
1771 	int running = netif_running(netdev);
1772 
1773 	if (new_mtu < ENIC_MIN_MTU || new_mtu > ENIC_MAX_MTU)
1774 		return -EINVAL;
1775 
1776 	if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic))
1777 		return -EOPNOTSUPP;
1778 
1779 	if (running)
1780 		enic_stop(netdev);
1781 
1782 	netdev->mtu = new_mtu;
1783 
1784 	if (netdev->mtu > enic->port_mtu)
1785 		netdev_warn(netdev,
1786 			"interface MTU (%d) set higher than port MTU (%d)\n",
1787 			netdev->mtu, enic->port_mtu);
1788 
1789 	if (running)
1790 		enic_open(netdev);
1791 
1792 	return 0;
1793 }
1794 
1795 static void enic_change_mtu_work(struct work_struct *work)
1796 {
1797 	struct enic *enic = container_of(work, struct enic, change_mtu_work);
1798 	struct net_device *netdev = enic->netdev;
1799 	int new_mtu = vnic_dev_mtu(enic->vdev);
1800 	int err;
1801 	unsigned int i;
1802 
1803 	new_mtu = max_t(int, ENIC_MIN_MTU, min_t(int, ENIC_MAX_MTU, new_mtu));
1804 
1805 	rtnl_lock();
1806 
1807 	/* Stop RQ */
1808 	del_timer_sync(&enic->notify_timer);
1809 
1810 	for (i = 0; i < enic->rq_count; i++)
1811 		napi_disable(&enic->napi[i]);
1812 
1813 	vnic_intr_mask(&enic->intr[0]);
1814 	enic_synchronize_irqs(enic);
1815 	err = vnic_rq_disable(&enic->rq[0]);
1816 	if (err) {
1817 		rtnl_unlock();
1818 		netdev_err(netdev, "Unable to disable RQ.\n");
1819 		return;
1820 	}
1821 	vnic_rq_clean(&enic->rq[0], enic_free_rq_buf);
1822 	vnic_cq_clean(&enic->cq[0]);
1823 	vnic_intr_clean(&enic->intr[0]);
1824 
1825 	/* Fill RQ with new_mtu-sized buffers */
1826 	netdev->mtu = new_mtu;
1827 	vnic_rq_fill(&enic->rq[0], enic_rq_alloc_buf);
1828 	/* Need at least one buffer on ring to get going */
1829 	if (vnic_rq_desc_used(&enic->rq[0]) == 0) {
1830 		rtnl_unlock();
1831 		netdev_err(netdev, "Unable to alloc receive buffers.\n");
1832 		return;
1833 	}
1834 
1835 	/* Start RQ */
1836 	vnic_rq_enable(&enic->rq[0]);
1837 	napi_enable(&enic->napi[0]);
1838 	vnic_intr_unmask(&enic->intr[0]);
1839 	enic_notify_timer_start(enic);
1840 
1841 	rtnl_unlock();
1842 
1843 	netdev_info(netdev, "interface MTU set as %d\n", netdev->mtu);
1844 }
1845 
1846 #ifdef CONFIG_NET_POLL_CONTROLLER
1847 static void enic_poll_controller(struct net_device *netdev)
1848 {
1849 	struct enic *enic = netdev_priv(netdev);
1850 	struct vnic_dev *vdev = enic->vdev;
1851 	unsigned int i, intr;
1852 
1853 	switch (vnic_dev_get_intr_mode(vdev)) {
1854 	case VNIC_DEV_INTR_MODE_MSIX:
1855 		for (i = 0; i < enic->rq_count; i++) {
1856 			intr = enic_msix_rq_intr(enic, i);
1857 			enic_isr_msix(enic->msix_entry[intr].vector,
1858 				      &enic->napi[i]);
1859 		}
1860 
1861 		for (i = 0; i < enic->wq_count; i++) {
1862 			intr = enic_msix_wq_intr(enic, i);
1863 			enic_isr_msix(enic->msix_entry[intr].vector,
1864 				      &enic->napi[enic_cq_wq(enic, i)]);
1865 		}
1866 
1867 		break;
1868 	case VNIC_DEV_INTR_MODE_MSI:
1869 		enic_isr_msi(enic->pdev->irq, enic);
1870 		break;
1871 	case VNIC_DEV_INTR_MODE_INTX:
1872 		enic_isr_legacy(enic->pdev->irq, netdev);
1873 		break;
1874 	default:
1875 		break;
1876 	}
1877 }
1878 #endif
1879 
1880 static int enic_dev_wait(struct vnic_dev *vdev,
1881 	int (*start)(struct vnic_dev *, int),
1882 	int (*finished)(struct vnic_dev *, int *),
1883 	int arg)
1884 {
1885 	unsigned long time;
1886 	int done;
1887 	int err;
1888 
1889 	BUG_ON(in_interrupt());
1890 
1891 	err = start(vdev, arg);
1892 	if (err)
1893 		return err;
1894 
1895 	/* Wait for func to complete...2 seconds max
1896 	 */
1897 
1898 	time = jiffies + (HZ * 2);
1899 	do {
1900 
1901 		err = finished(vdev, &done);
1902 		if (err)
1903 			return err;
1904 
1905 		if (done)
1906 			return 0;
1907 
1908 		schedule_timeout_uninterruptible(HZ / 10);
1909 
1910 	} while (time_after(time, jiffies));
1911 
1912 	return -ETIMEDOUT;
1913 }
1914 
1915 static int enic_dev_open(struct enic *enic)
1916 {
1917 	int err;
1918 
1919 	err = enic_dev_wait(enic->vdev, vnic_dev_open,
1920 		vnic_dev_open_done, 0);
1921 	if (err)
1922 		dev_err(enic_get_dev(enic), "vNIC device open failed, err %d\n",
1923 			err);
1924 
1925 	return err;
1926 }
1927 
1928 static int enic_dev_hang_reset(struct enic *enic)
1929 {
1930 	int err;
1931 
1932 	err = enic_dev_wait(enic->vdev, vnic_dev_hang_reset,
1933 		vnic_dev_hang_reset_done, 0);
1934 	if (err)
1935 		netdev_err(enic->netdev, "vNIC hang reset failed, err %d\n",
1936 			err);
1937 
1938 	return err;
1939 }
1940 
1941 int __enic_set_rsskey(struct enic *enic)
1942 {
1943 	union vnic_rss_key *rss_key_buf_va;
1944 	dma_addr_t rss_key_buf_pa;
1945 	int i, kidx, bidx, err;
1946 
1947 	rss_key_buf_va = pci_zalloc_consistent(enic->pdev,
1948 					       sizeof(union vnic_rss_key),
1949 					       &rss_key_buf_pa);
1950 	if (!rss_key_buf_va)
1951 		return -ENOMEM;
1952 
1953 	for (i = 0; i < ENIC_RSS_LEN; i++) {
1954 		kidx = i / ENIC_RSS_BYTES_PER_KEY;
1955 		bidx = i % ENIC_RSS_BYTES_PER_KEY;
1956 		rss_key_buf_va->key[kidx].b[bidx] = enic->rss_key[i];
1957 	}
1958 	spin_lock_bh(&enic->devcmd_lock);
1959 	err = enic_set_rss_key(enic,
1960 		rss_key_buf_pa,
1961 		sizeof(union vnic_rss_key));
1962 	spin_unlock_bh(&enic->devcmd_lock);
1963 
1964 	pci_free_consistent(enic->pdev, sizeof(union vnic_rss_key),
1965 		rss_key_buf_va, rss_key_buf_pa);
1966 
1967 	return err;
1968 }
1969 
1970 static int enic_set_rsskey(struct enic *enic)
1971 {
1972 	netdev_rss_key_fill(enic->rss_key, ENIC_RSS_LEN);
1973 
1974 	return __enic_set_rsskey(enic);
1975 }
1976 
1977 static int enic_set_rsscpu(struct enic *enic, u8 rss_hash_bits)
1978 {
1979 	dma_addr_t rss_cpu_buf_pa;
1980 	union vnic_rss_cpu *rss_cpu_buf_va = NULL;
1981 	unsigned int i;
1982 	int err;
1983 
1984 	rss_cpu_buf_va = pci_alloc_consistent(enic->pdev,
1985 		sizeof(union vnic_rss_cpu), &rss_cpu_buf_pa);
1986 	if (!rss_cpu_buf_va)
1987 		return -ENOMEM;
1988 
1989 	for (i = 0; i < (1 << rss_hash_bits); i++)
1990 		(*rss_cpu_buf_va).cpu[i/4].b[i%4] = i % enic->rq_count;
1991 
1992 	spin_lock_bh(&enic->devcmd_lock);
1993 	err = enic_set_rss_cpu(enic,
1994 		rss_cpu_buf_pa,
1995 		sizeof(union vnic_rss_cpu));
1996 	spin_unlock_bh(&enic->devcmd_lock);
1997 
1998 	pci_free_consistent(enic->pdev, sizeof(union vnic_rss_cpu),
1999 		rss_cpu_buf_va, rss_cpu_buf_pa);
2000 
2001 	return err;
2002 }
2003 
2004 static int enic_set_niccfg(struct enic *enic, u8 rss_default_cpu,
2005 	u8 rss_hash_type, u8 rss_hash_bits, u8 rss_base_cpu, u8 rss_enable)
2006 {
2007 	const u8 tso_ipid_split_en = 0;
2008 	const u8 ig_vlan_strip_en = 1;
2009 	int err;
2010 
2011 	/* Enable VLAN tag stripping.
2012 	*/
2013 
2014 	spin_lock_bh(&enic->devcmd_lock);
2015 	err = enic_set_nic_cfg(enic,
2016 		rss_default_cpu, rss_hash_type,
2017 		rss_hash_bits, rss_base_cpu,
2018 		rss_enable, tso_ipid_split_en,
2019 		ig_vlan_strip_en);
2020 	spin_unlock_bh(&enic->devcmd_lock);
2021 
2022 	return err;
2023 }
2024 
2025 static int enic_set_rss_nic_cfg(struct enic *enic)
2026 {
2027 	struct device *dev = enic_get_dev(enic);
2028 	const u8 rss_default_cpu = 0;
2029 	const u8 rss_hash_type = NIC_CFG_RSS_HASH_TYPE_IPV4 |
2030 		NIC_CFG_RSS_HASH_TYPE_TCP_IPV4 |
2031 		NIC_CFG_RSS_HASH_TYPE_IPV6 |
2032 		NIC_CFG_RSS_HASH_TYPE_TCP_IPV6;
2033 	const u8 rss_hash_bits = 7;
2034 	const u8 rss_base_cpu = 0;
2035 	u8 rss_enable = ENIC_SETTING(enic, RSS) && (enic->rq_count > 1);
2036 
2037 	if (rss_enable) {
2038 		if (!enic_set_rsskey(enic)) {
2039 			if (enic_set_rsscpu(enic, rss_hash_bits)) {
2040 				rss_enable = 0;
2041 				dev_warn(dev, "RSS disabled, "
2042 					"Failed to set RSS cpu indirection table.");
2043 			}
2044 		} else {
2045 			rss_enable = 0;
2046 			dev_warn(dev, "RSS disabled, Failed to set RSS key.\n");
2047 		}
2048 	}
2049 
2050 	return enic_set_niccfg(enic, rss_default_cpu, rss_hash_type,
2051 		rss_hash_bits, rss_base_cpu, rss_enable);
2052 }
2053 
2054 static void enic_reset(struct work_struct *work)
2055 {
2056 	struct enic *enic = container_of(work, struct enic, reset);
2057 
2058 	if (!netif_running(enic->netdev))
2059 		return;
2060 
2061 	rtnl_lock();
2062 
2063 	spin_lock(&enic->enic_api_lock);
2064 	enic_dev_hang_notify(enic);
2065 	enic_stop(enic->netdev);
2066 	enic_dev_hang_reset(enic);
2067 	enic_reset_addr_lists(enic);
2068 	enic_init_vnic_resources(enic);
2069 	enic_set_rss_nic_cfg(enic);
2070 	enic_dev_set_ig_vlan_rewrite_mode(enic);
2071 	enic_open(enic->netdev);
2072 	spin_unlock(&enic->enic_api_lock);
2073 	call_netdevice_notifiers(NETDEV_REBOOT, enic->netdev);
2074 
2075 	rtnl_unlock();
2076 }
2077 
2078 static int enic_set_intr_mode(struct enic *enic)
2079 {
2080 	unsigned int n = min_t(unsigned int, enic->rq_count, ENIC_RQ_MAX);
2081 	unsigned int m = min_t(unsigned int, enic->wq_count, ENIC_WQ_MAX);
2082 	unsigned int i;
2083 
2084 	/* Set interrupt mode (INTx, MSI, MSI-X) depending
2085 	 * on system capabilities.
2086 	 *
2087 	 * Try MSI-X first
2088 	 *
2089 	 * We need n RQs, m WQs, n+m CQs, and n+m+2 INTRs
2090 	 * (the second to last INTR is used for WQ/RQ errors)
2091 	 * (the last INTR is used for notifications)
2092 	 */
2093 
2094 	BUG_ON(ARRAY_SIZE(enic->msix_entry) < n + m + 2);
2095 	for (i = 0; i < n + m + 2; i++)
2096 		enic->msix_entry[i].entry = i;
2097 
2098 	/* Use multiple RQs if RSS is enabled
2099 	 */
2100 
2101 	if (ENIC_SETTING(enic, RSS) &&
2102 	    enic->config.intr_mode < 1 &&
2103 	    enic->rq_count >= n &&
2104 	    enic->wq_count >= m &&
2105 	    enic->cq_count >= n + m &&
2106 	    enic->intr_count >= n + m + 2) {
2107 
2108 		if (pci_enable_msix_range(enic->pdev, enic->msix_entry,
2109 					  n + m + 2, n + m + 2) > 0) {
2110 
2111 			enic->rq_count = n;
2112 			enic->wq_count = m;
2113 			enic->cq_count = n + m;
2114 			enic->intr_count = n + m + 2;
2115 
2116 			vnic_dev_set_intr_mode(enic->vdev,
2117 				VNIC_DEV_INTR_MODE_MSIX);
2118 
2119 			return 0;
2120 		}
2121 	}
2122 
2123 	if (enic->config.intr_mode < 1 &&
2124 	    enic->rq_count >= 1 &&
2125 	    enic->wq_count >= m &&
2126 	    enic->cq_count >= 1 + m &&
2127 	    enic->intr_count >= 1 + m + 2) {
2128 		if (pci_enable_msix_range(enic->pdev, enic->msix_entry,
2129 					  1 + m + 2, 1 + m + 2) > 0) {
2130 
2131 			enic->rq_count = 1;
2132 			enic->wq_count = m;
2133 			enic->cq_count = 1 + m;
2134 			enic->intr_count = 1 + m + 2;
2135 
2136 			vnic_dev_set_intr_mode(enic->vdev,
2137 				VNIC_DEV_INTR_MODE_MSIX);
2138 
2139 			return 0;
2140 		}
2141 	}
2142 
2143 	/* Next try MSI
2144 	 *
2145 	 * We need 1 RQ, 1 WQ, 2 CQs, and 1 INTR
2146 	 */
2147 
2148 	if (enic->config.intr_mode < 2 &&
2149 	    enic->rq_count >= 1 &&
2150 	    enic->wq_count >= 1 &&
2151 	    enic->cq_count >= 2 &&
2152 	    enic->intr_count >= 1 &&
2153 	    !pci_enable_msi(enic->pdev)) {
2154 
2155 		enic->rq_count = 1;
2156 		enic->wq_count = 1;
2157 		enic->cq_count = 2;
2158 		enic->intr_count = 1;
2159 
2160 		vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSI);
2161 
2162 		return 0;
2163 	}
2164 
2165 	/* Next try INTx
2166 	 *
2167 	 * We need 1 RQ, 1 WQ, 2 CQs, and 3 INTRs
2168 	 * (the first INTR is used for WQ/RQ)
2169 	 * (the second INTR is used for WQ/RQ errors)
2170 	 * (the last INTR is used for notifications)
2171 	 */
2172 
2173 	if (enic->config.intr_mode < 3 &&
2174 	    enic->rq_count >= 1 &&
2175 	    enic->wq_count >= 1 &&
2176 	    enic->cq_count >= 2 &&
2177 	    enic->intr_count >= 3) {
2178 
2179 		enic->rq_count = 1;
2180 		enic->wq_count = 1;
2181 		enic->cq_count = 2;
2182 		enic->intr_count = 3;
2183 
2184 		vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_INTX);
2185 
2186 		return 0;
2187 	}
2188 
2189 	vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
2190 
2191 	return -EINVAL;
2192 }
2193 
2194 static void enic_clear_intr_mode(struct enic *enic)
2195 {
2196 	switch (vnic_dev_get_intr_mode(enic->vdev)) {
2197 	case VNIC_DEV_INTR_MODE_MSIX:
2198 		pci_disable_msix(enic->pdev);
2199 		break;
2200 	case VNIC_DEV_INTR_MODE_MSI:
2201 		pci_disable_msi(enic->pdev);
2202 		break;
2203 	default:
2204 		break;
2205 	}
2206 
2207 	vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
2208 }
2209 
2210 static const struct net_device_ops enic_netdev_dynamic_ops = {
2211 	.ndo_open		= enic_open,
2212 	.ndo_stop		= enic_stop,
2213 	.ndo_start_xmit		= enic_hard_start_xmit,
2214 	.ndo_get_stats64	= enic_get_stats,
2215 	.ndo_validate_addr	= eth_validate_addr,
2216 	.ndo_set_rx_mode	= enic_set_rx_mode,
2217 	.ndo_set_mac_address	= enic_set_mac_address_dynamic,
2218 	.ndo_change_mtu		= enic_change_mtu,
2219 	.ndo_vlan_rx_add_vid	= enic_vlan_rx_add_vid,
2220 	.ndo_vlan_rx_kill_vid	= enic_vlan_rx_kill_vid,
2221 	.ndo_tx_timeout		= enic_tx_timeout,
2222 	.ndo_set_vf_port	= enic_set_vf_port,
2223 	.ndo_get_vf_port	= enic_get_vf_port,
2224 	.ndo_set_vf_mac		= enic_set_vf_mac,
2225 #ifdef CONFIG_NET_POLL_CONTROLLER
2226 	.ndo_poll_controller	= enic_poll_controller,
2227 #endif
2228 #ifdef CONFIG_RFS_ACCEL
2229 	.ndo_rx_flow_steer	= enic_rx_flow_steer,
2230 #endif
2231 #ifdef CONFIG_NET_RX_BUSY_POLL
2232 	.ndo_busy_poll		= enic_busy_poll,
2233 #endif
2234 };
2235 
2236 static const struct net_device_ops enic_netdev_ops = {
2237 	.ndo_open		= enic_open,
2238 	.ndo_stop		= enic_stop,
2239 	.ndo_start_xmit		= enic_hard_start_xmit,
2240 	.ndo_get_stats64	= enic_get_stats,
2241 	.ndo_validate_addr	= eth_validate_addr,
2242 	.ndo_set_mac_address	= enic_set_mac_address,
2243 	.ndo_set_rx_mode	= enic_set_rx_mode,
2244 	.ndo_change_mtu		= enic_change_mtu,
2245 	.ndo_vlan_rx_add_vid	= enic_vlan_rx_add_vid,
2246 	.ndo_vlan_rx_kill_vid	= enic_vlan_rx_kill_vid,
2247 	.ndo_tx_timeout		= enic_tx_timeout,
2248 	.ndo_set_vf_port	= enic_set_vf_port,
2249 	.ndo_get_vf_port	= enic_get_vf_port,
2250 	.ndo_set_vf_mac		= enic_set_vf_mac,
2251 #ifdef CONFIG_NET_POLL_CONTROLLER
2252 	.ndo_poll_controller	= enic_poll_controller,
2253 #endif
2254 #ifdef CONFIG_RFS_ACCEL
2255 	.ndo_rx_flow_steer	= enic_rx_flow_steer,
2256 #endif
2257 #ifdef CONFIG_NET_RX_BUSY_POLL
2258 	.ndo_busy_poll		= enic_busy_poll,
2259 #endif
2260 };
2261 
2262 static void enic_dev_deinit(struct enic *enic)
2263 {
2264 	unsigned int i;
2265 
2266 	for (i = 0; i < enic->rq_count; i++) {
2267 		napi_hash_del(&enic->napi[i]);
2268 		netif_napi_del(&enic->napi[i]);
2269 	}
2270 	if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX)
2271 		for (i = 0; i < enic->wq_count; i++)
2272 			netif_napi_del(&enic->napi[enic_cq_wq(enic, i)]);
2273 
2274 	enic_free_vnic_resources(enic);
2275 	enic_clear_intr_mode(enic);
2276 }
2277 
2278 static void enic_kdump_kernel_config(struct enic *enic)
2279 {
2280 	if (is_kdump_kernel()) {
2281 		dev_info(enic_get_dev(enic), "Running from within kdump kernel. Using minimal resources\n");
2282 		enic->rq_count = 1;
2283 		enic->wq_count = 1;
2284 		enic->config.rq_desc_count = ENIC_MIN_RQ_DESCS;
2285 		enic->config.wq_desc_count = ENIC_MIN_WQ_DESCS;
2286 		enic->config.mtu = min_t(u16, 1500, enic->config.mtu);
2287 	}
2288 }
2289 
2290 static int enic_dev_init(struct enic *enic)
2291 {
2292 	struct device *dev = enic_get_dev(enic);
2293 	struct net_device *netdev = enic->netdev;
2294 	unsigned int i;
2295 	int err;
2296 
2297 	/* Get interrupt coalesce timer info */
2298 	err = enic_dev_intr_coal_timer_info(enic);
2299 	if (err) {
2300 		dev_warn(dev, "Using default conversion factor for "
2301 			"interrupt coalesce timer\n");
2302 		vnic_dev_intr_coal_timer_info_default(enic->vdev);
2303 	}
2304 
2305 	/* Get vNIC configuration
2306 	 */
2307 
2308 	err = enic_get_vnic_config(enic);
2309 	if (err) {
2310 		dev_err(dev, "Get vNIC configuration failed, aborting\n");
2311 		return err;
2312 	}
2313 
2314 	/* Get available resource counts
2315 	 */
2316 
2317 	enic_get_res_counts(enic);
2318 
2319 	/* modify resource count if we are in kdump_kernel
2320 	 */
2321 	enic_kdump_kernel_config(enic);
2322 
2323 	/* Set interrupt mode based on resource counts and system
2324 	 * capabilities
2325 	 */
2326 
2327 	err = enic_set_intr_mode(enic);
2328 	if (err) {
2329 		dev_err(dev, "Failed to set intr mode based on resource "
2330 			"counts and system capabilities, aborting\n");
2331 		return err;
2332 	}
2333 
2334 	/* Allocate and configure vNIC resources
2335 	 */
2336 
2337 	err = enic_alloc_vnic_resources(enic);
2338 	if (err) {
2339 		dev_err(dev, "Failed to alloc vNIC resources, aborting\n");
2340 		goto err_out_free_vnic_resources;
2341 	}
2342 
2343 	enic_init_vnic_resources(enic);
2344 
2345 	err = enic_set_rss_nic_cfg(enic);
2346 	if (err) {
2347 		dev_err(dev, "Failed to config nic, aborting\n");
2348 		goto err_out_free_vnic_resources;
2349 	}
2350 
2351 	switch (vnic_dev_get_intr_mode(enic->vdev)) {
2352 	default:
2353 		netif_napi_add(netdev, &enic->napi[0], enic_poll, 64);
2354 		napi_hash_add(&enic->napi[0]);
2355 		break;
2356 	case VNIC_DEV_INTR_MODE_MSIX:
2357 		for (i = 0; i < enic->rq_count; i++) {
2358 			netif_napi_add(netdev, &enic->napi[i],
2359 				enic_poll_msix_rq, NAPI_POLL_WEIGHT);
2360 			napi_hash_add(&enic->napi[i]);
2361 		}
2362 		for (i = 0; i < enic->wq_count; i++)
2363 			netif_napi_add(netdev, &enic->napi[enic_cq_wq(enic, i)],
2364 				       enic_poll_msix_wq, NAPI_POLL_WEIGHT);
2365 		break;
2366 	}
2367 
2368 	return 0;
2369 
2370 err_out_free_vnic_resources:
2371 	enic_clear_intr_mode(enic);
2372 	enic_free_vnic_resources(enic);
2373 
2374 	return err;
2375 }
2376 
2377 static void enic_iounmap(struct enic *enic)
2378 {
2379 	unsigned int i;
2380 
2381 	for (i = 0; i < ARRAY_SIZE(enic->bar); i++)
2382 		if (enic->bar[i].vaddr)
2383 			iounmap(enic->bar[i].vaddr);
2384 }
2385 
2386 static int enic_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2387 {
2388 	struct device *dev = &pdev->dev;
2389 	struct net_device *netdev;
2390 	struct enic *enic;
2391 	int using_dac = 0;
2392 	unsigned int i;
2393 	int err;
2394 #ifdef CONFIG_PCI_IOV
2395 	int pos = 0;
2396 #endif
2397 	int num_pps = 1;
2398 
2399 	/* Allocate net device structure and initialize.  Private
2400 	 * instance data is initialized to zero.
2401 	 */
2402 
2403 	netdev = alloc_etherdev_mqs(sizeof(struct enic),
2404 				    ENIC_RQ_MAX, ENIC_WQ_MAX);
2405 	if (!netdev)
2406 		return -ENOMEM;
2407 
2408 	pci_set_drvdata(pdev, netdev);
2409 
2410 	SET_NETDEV_DEV(netdev, &pdev->dev);
2411 
2412 	enic = netdev_priv(netdev);
2413 	enic->netdev = netdev;
2414 	enic->pdev = pdev;
2415 
2416 	/* Setup PCI resources
2417 	 */
2418 
2419 	err = pci_enable_device_mem(pdev);
2420 	if (err) {
2421 		dev_err(dev, "Cannot enable PCI device, aborting\n");
2422 		goto err_out_free_netdev;
2423 	}
2424 
2425 	err = pci_request_regions(pdev, DRV_NAME);
2426 	if (err) {
2427 		dev_err(dev, "Cannot request PCI regions, aborting\n");
2428 		goto err_out_disable_device;
2429 	}
2430 
2431 	pci_set_master(pdev);
2432 
2433 	/* Query PCI controller on system for DMA addressing
2434 	 * limitation for the device.  Try 64-bit first, and
2435 	 * fail to 32-bit.
2436 	 */
2437 
2438 	err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
2439 	if (err) {
2440 		err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
2441 		if (err) {
2442 			dev_err(dev, "No usable DMA configuration, aborting\n");
2443 			goto err_out_release_regions;
2444 		}
2445 		err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
2446 		if (err) {
2447 			dev_err(dev, "Unable to obtain %u-bit DMA "
2448 				"for consistent allocations, aborting\n", 32);
2449 			goto err_out_release_regions;
2450 		}
2451 	} else {
2452 		err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
2453 		if (err) {
2454 			dev_err(dev, "Unable to obtain %u-bit DMA "
2455 				"for consistent allocations, aborting\n", 64);
2456 			goto err_out_release_regions;
2457 		}
2458 		using_dac = 1;
2459 	}
2460 
2461 	/* Map vNIC resources from BAR0-5
2462 	 */
2463 
2464 	for (i = 0; i < ARRAY_SIZE(enic->bar); i++) {
2465 		if (!(pci_resource_flags(pdev, i) & IORESOURCE_MEM))
2466 			continue;
2467 		enic->bar[i].len = pci_resource_len(pdev, i);
2468 		enic->bar[i].vaddr = pci_iomap(pdev, i, enic->bar[i].len);
2469 		if (!enic->bar[i].vaddr) {
2470 			dev_err(dev, "Cannot memory-map BAR %d, aborting\n", i);
2471 			err = -ENODEV;
2472 			goto err_out_iounmap;
2473 		}
2474 		enic->bar[i].bus_addr = pci_resource_start(pdev, i);
2475 	}
2476 
2477 	/* Register vNIC device
2478 	 */
2479 
2480 	enic->vdev = vnic_dev_register(NULL, enic, pdev, enic->bar,
2481 		ARRAY_SIZE(enic->bar));
2482 	if (!enic->vdev) {
2483 		dev_err(dev, "vNIC registration failed, aborting\n");
2484 		err = -ENODEV;
2485 		goto err_out_iounmap;
2486 	}
2487 
2488 #ifdef CONFIG_PCI_IOV
2489 	/* Get number of subvnics */
2490 	pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_SRIOV);
2491 	if (pos) {
2492 		pci_read_config_word(pdev, pos + PCI_SRIOV_TOTAL_VF,
2493 			&enic->num_vfs);
2494 		if (enic->num_vfs) {
2495 			err = pci_enable_sriov(pdev, enic->num_vfs);
2496 			if (err) {
2497 				dev_err(dev, "SRIOV enable failed, aborting."
2498 					" pci_enable_sriov() returned %d\n",
2499 					err);
2500 				goto err_out_vnic_unregister;
2501 			}
2502 			enic->priv_flags |= ENIC_SRIOV_ENABLED;
2503 			num_pps = enic->num_vfs;
2504 		}
2505 	}
2506 #endif
2507 
2508 	/* Allocate structure for port profiles */
2509 	enic->pp = kcalloc(num_pps, sizeof(*enic->pp), GFP_KERNEL);
2510 	if (!enic->pp) {
2511 		err = -ENOMEM;
2512 		goto err_out_disable_sriov_pp;
2513 	}
2514 
2515 	/* Issue device open to get device in known state
2516 	 */
2517 
2518 	err = enic_dev_open(enic);
2519 	if (err) {
2520 		dev_err(dev, "vNIC dev open failed, aborting\n");
2521 		goto err_out_disable_sriov;
2522 	}
2523 
2524 	/* Setup devcmd lock
2525 	 */
2526 
2527 	spin_lock_init(&enic->devcmd_lock);
2528 	spin_lock_init(&enic->enic_api_lock);
2529 
2530 	/*
2531 	 * Set ingress vlan rewrite mode before vnic initialization
2532 	 */
2533 
2534 	err = enic_dev_set_ig_vlan_rewrite_mode(enic);
2535 	if (err) {
2536 		dev_err(dev,
2537 			"Failed to set ingress vlan rewrite mode, aborting.\n");
2538 		goto err_out_dev_close;
2539 	}
2540 
2541 	/* Issue device init to initialize the vnic-to-switch link.
2542 	 * We'll start with carrier off and wait for link UP
2543 	 * notification later to turn on carrier.  We don't need
2544 	 * to wait here for the vnic-to-switch link initialization
2545 	 * to complete; link UP notification is the indication that
2546 	 * the process is complete.
2547 	 */
2548 
2549 	netif_carrier_off(netdev);
2550 
2551 	/* Do not call dev_init for a dynamic vnic.
2552 	 * For a dynamic vnic, init_prov_info will be
2553 	 * called later by an upper layer.
2554 	 */
2555 
2556 	if (!enic_is_dynamic(enic)) {
2557 		err = vnic_dev_init(enic->vdev, 0);
2558 		if (err) {
2559 			dev_err(dev, "vNIC dev init failed, aborting\n");
2560 			goto err_out_dev_close;
2561 		}
2562 	}
2563 
2564 	err = enic_dev_init(enic);
2565 	if (err) {
2566 		dev_err(dev, "Device initialization failed, aborting\n");
2567 		goto err_out_dev_close;
2568 	}
2569 
2570 	netif_set_real_num_tx_queues(netdev, enic->wq_count);
2571 	netif_set_real_num_rx_queues(netdev, enic->rq_count);
2572 
2573 	/* Setup notification timer, HW reset task, and wq locks
2574 	 */
2575 
2576 	init_timer(&enic->notify_timer);
2577 	enic->notify_timer.function = enic_notify_timer;
2578 	enic->notify_timer.data = (unsigned long)enic;
2579 
2580 	enic_set_rx_coal_setting(enic);
2581 	INIT_WORK(&enic->reset, enic_reset);
2582 	INIT_WORK(&enic->change_mtu_work, enic_change_mtu_work);
2583 
2584 	for (i = 0; i < enic->wq_count; i++)
2585 		spin_lock_init(&enic->wq_lock[i]);
2586 
2587 	/* Register net device
2588 	 */
2589 
2590 	enic->port_mtu = enic->config.mtu;
2591 	(void)enic_change_mtu(netdev, enic->port_mtu);
2592 
2593 	err = enic_set_mac_addr(netdev, enic->mac_addr);
2594 	if (err) {
2595 		dev_err(dev, "Invalid MAC address, aborting\n");
2596 		goto err_out_dev_deinit;
2597 	}
2598 
2599 	enic->tx_coalesce_usecs = enic->config.intr_timer_usec;
2600 	/* rx coalesce time already got initialized. This gets used
2601 	 * if adaptive coal is turned off
2602 	 */
2603 	enic->rx_coalesce_usecs = enic->tx_coalesce_usecs;
2604 
2605 	if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic))
2606 		netdev->netdev_ops = &enic_netdev_dynamic_ops;
2607 	else
2608 		netdev->netdev_ops = &enic_netdev_ops;
2609 
2610 	netdev->watchdog_timeo = 2 * HZ;
2611 	enic_set_ethtool_ops(netdev);
2612 
2613 	netdev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
2614 	if (ENIC_SETTING(enic, LOOP)) {
2615 		netdev->features &= ~NETIF_F_HW_VLAN_CTAG_TX;
2616 		enic->loop_enable = 1;
2617 		enic->loop_tag = enic->config.loop_tag;
2618 		dev_info(dev, "loopback tag=0x%04x\n", enic->loop_tag);
2619 	}
2620 	if (ENIC_SETTING(enic, TXCSUM))
2621 		netdev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM;
2622 	if (ENIC_SETTING(enic, TSO))
2623 		netdev->hw_features |= NETIF_F_TSO |
2624 			NETIF_F_TSO6 | NETIF_F_TSO_ECN;
2625 	if (ENIC_SETTING(enic, RSS))
2626 		netdev->hw_features |= NETIF_F_RXHASH;
2627 	if (ENIC_SETTING(enic, RXCSUM))
2628 		netdev->hw_features |= NETIF_F_RXCSUM;
2629 
2630 	netdev->features |= netdev->hw_features;
2631 
2632 #ifdef CONFIG_RFS_ACCEL
2633 	netdev->hw_features |= NETIF_F_NTUPLE;
2634 #endif
2635 
2636 	if (using_dac)
2637 		netdev->features |= NETIF_F_HIGHDMA;
2638 
2639 	netdev->priv_flags |= IFF_UNICAST_FLT;
2640 
2641 	err = register_netdev(netdev);
2642 	if (err) {
2643 		dev_err(dev, "Cannot register net device, aborting\n");
2644 		goto err_out_dev_deinit;
2645 	}
2646 	enic->rx_copybreak = RX_COPYBREAK_DEFAULT;
2647 
2648 	return 0;
2649 
2650 err_out_dev_deinit:
2651 	enic_dev_deinit(enic);
2652 err_out_dev_close:
2653 	vnic_dev_close(enic->vdev);
2654 err_out_disable_sriov:
2655 	kfree(enic->pp);
2656 err_out_disable_sriov_pp:
2657 #ifdef CONFIG_PCI_IOV
2658 	if (enic_sriov_enabled(enic)) {
2659 		pci_disable_sriov(pdev);
2660 		enic->priv_flags &= ~ENIC_SRIOV_ENABLED;
2661 	}
2662 err_out_vnic_unregister:
2663 #endif
2664 	vnic_dev_unregister(enic->vdev);
2665 err_out_iounmap:
2666 	enic_iounmap(enic);
2667 err_out_release_regions:
2668 	pci_release_regions(pdev);
2669 err_out_disable_device:
2670 	pci_disable_device(pdev);
2671 err_out_free_netdev:
2672 	free_netdev(netdev);
2673 
2674 	return err;
2675 }
2676 
2677 static void enic_remove(struct pci_dev *pdev)
2678 {
2679 	struct net_device *netdev = pci_get_drvdata(pdev);
2680 
2681 	if (netdev) {
2682 		struct enic *enic = netdev_priv(netdev);
2683 
2684 		cancel_work_sync(&enic->reset);
2685 		cancel_work_sync(&enic->change_mtu_work);
2686 		unregister_netdev(netdev);
2687 		enic_dev_deinit(enic);
2688 		vnic_dev_close(enic->vdev);
2689 #ifdef CONFIG_PCI_IOV
2690 		if (enic_sriov_enabled(enic)) {
2691 			pci_disable_sriov(pdev);
2692 			enic->priv_flags &= ~ENIC_SRIOV_ENABLED;
2693 		}
2694 #endif
2695 		kfree(enic->pp);
2696 		vnic_dev_unregister(enic->vdev);
2697 		enic_iounmap(enic);
2698 		pci_release_regions(pdev);
2699 		pci_disable_device(pdev);
2700 		free_netdev(netdev);
2701 	}
2702 }
2703 
2704 static struct pci_driver enic_driver = {
2705 	.name = DRV_NAME,
2706 	.id_table = enic_id_table,
2707 	.probe = enic_probe,
2708 	.remove = enic_remove,
2709 };
2710 
2711 static int __init enic_init_module(void)
2712 {
2713 	pr_info("%s, ver %s\n", DRV_DESCRIPTION, DRV_VERSION);
2714 
2715 	return pci_register_driver(&enic_driver);
2716 }
2717 
2718 static void __exit enic_cleanup_module(void)
2719 {
2720 	pci_unregister_driver(&enic_driver);
2721 }
2722 
2723 module_init(enic_init_module);
2724 module_exit(enic_cleanup_module);
2725