xref: /linux/drivers/net/ethernet/cisco/enic/enic_main.c (revision e0bf6c5ca2d3281f231c5f0c9bf145e9513644de)
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 
619 	enic_dev_stats_dump(enic, &stats);
620 
621 	net_stats->tx_packets = stats->tx.tx_frames_ok;
622 	net_stats->tx_bytes = stats->tx.tx_bytes_ok;
623 	net_stats->tx_errors = stats->tx.tx_errors;
624 	net_stats->tx_dropped = stats->tx.tx_drops;
625 
626 	net_stats->rx_packets = stats->rx.rx_frames_ok;
627 	net_stats->rx_bytes = stats->rx.rx_bytes_ok;
628 	net_stats->rx_errors = stats->rx.rx_errors;
629 	net_stats->multicast = stats->rx.rx_multicast_frames_ok;
630 	net_stats->rx_over_errors = enic->rq_truncated_pkts;
631 	net_stats->rx_crc_errors = enic->rq_bad_fcs;
632 	net_stats->rx_dropped = stats->rx.rx_no_bufs + stats->rx.rx_drop;
633 
634 	return net_stats;
635 }
636 
637 static int enic_mc_sync(struct net_device *netdev, const u8 *mc_addr)
638 {
639 	struct enic *enic = netdev_priv(netdev);
640 
641 	if (enic->mc_count == ENIC_MULTICAST_PERFECT_FILTERS) {
642 		unsigned int mc_count = netdev_mc_count(netdev);
643 
644 		netdev_warn(netdev, "Registering only %d out of %d multicast addresses\n",
645 			    ENIC_MULTICAST_PERFECT_FILTERS, mc_count);
646 
647 		return -ENOSPC;
648 	}
649 
650 	enic_dev_add_addr(enic, mc_addr);
651 	enic->mc_count++;
652 
653 	return 0;
654 }
655 
656 static int enic_mc_unsync(struct net_device *netdev, const u8 *mc_addr)
657 {
658 	struct enic *enic = netdev_priv(netdev);
659 
660 	enic_dev_del_addr(enic, mc_addr);
661 	enic->mc_count--;
662 
663 	return 0;
664 }
665 
666 static int enic_uc_sync(struct net_device *netdev, const u8 *uc_addr)
667 {
668 	struct enic *enic = netdev_priv(netdev);
669 
670 	if (enic->uc_count == ENIC_UNICAST_PERFECT_FILTERS) {
671 		unsigned int uc_count = netdev_uc_count(netdev);
672 
673 		netdev_warn(netdev, "Registering only %d out of %d unicast addresses\n",
674 			    ENIC_UNICAST_PERFECT_FILTERS, uc_count);
675 
676 		return -ENOSPC;
677 	}
678 
679 	enic_dev_add_addr(enic, uc_addr);
680 	enic->uc_count++;
681 
682 	return 0;
683 }
684 
685 static int enic_uc_unsync(struct net_device *netdev, const u8 *uc_addr)
686 {
687 	struct enic *enic = netdev_priv(netdev);
688 
689 	enic_dev_del_addr(enic, uc_addr);
690 	enic->uc_count--;
691 
692 	return 0;
693 }
694 
695 void enic_reset_addr_lists(struct enic *enic)
696 {
697 	struct net_device *netdev = enic->netdev;
698 
699 	__dev_uc_unsync(netdev, NULL);
700 	__dev_mc_unsync(netdev, NULL);
701 
702 	enic->mc_count = 0;
703 	enic->uc_count = 0;
704 	enic->flags = 0;
705 }
706 
707 static int enic_set_mac_addr(struct net_device *netdev, char *addr)
708 {
709 	struct enic *enic = netdev_priv(netdev);
710 
711 	if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic)) {
712 		if (!is_valid_ether_addr(addr) && !is_zero_ether_addr(addr))
713 			return -EADDRNOTAVAIL;
714 	} else {
715 		if (!is_valid_ether_addr(addr))
716 			return -EADDRNOTAVAIL;
717 	}
718 
719 	memcpy(netdev->dev_addr, addr, netdev->addr_len);
720 
721 	return 0;
722 }
723 
724 static int enic_set_mac_address_dynamic(struct net_device *netdev, void *p)
725 {
726 	struct enic *enic = netdev_priv(netdev);
727 	struct sockaddr *saddr = p;
728 	char *addr = saddr->sa_data;
729 	int err;
730 
731 	if (netif_running(enic->netdev)) {
732 		err = enic_dev_del_station_addr(enic);
733 		if (err)
734 			return err;
735 	}
736 
737 	err = enic_set_mac_addr(netdev, addr);
738 	if (err)
739 		return err;
740 
741 	if (netif_running(enic->netdev)) {
742 		err = enic_dev_add_station_addr(enic);
743 		if (err)
744 			return err;
745 	}
746 
747 	return err;
748 }
749 
750 static int enic_set_mac_address(struct net_device *netdev, void *p)
751 {
752 	struct sockaddr *saddr = p;
753 	char *addr = saddr->sa_data;
754 	struct enic *enic = netdev_priv(netdev);
755 	int err;
756 
757 	err = enic_dev_del_station_addr(enic);
758 	if (err)
759 		return err;
760 
761 	err = enic_set_mac_addr(netdev, addr);
762 	if (err)
763 		return err;
764 
765 	return enic_dev_add_station_addr(enic);
766 }
767 
768 /* netif_tx_lock held, BHs disabled */
769 static void enic_set_rx_mode(struct net_device *netdev)
770 {
771 	struct enic *enic = netdev_priv(netdev);
772 	int directed = 1;
773 	int multicast = (netdev->flags & IFF_MULTICAST) ? 1 : 0;
774 	int broadcast = (netdev->flags & IFF_BROADCAST) ? 1 : 0;
775 	int promisc = (netdev->flags & IFF_PROMISC) ||
776 		netdev_uc_count(netdev) > ENIC_UNICAST_PERFECT_FILTERS;
777 	int allmulti = (netdev->flags & IFF_ALLMULTI) ||
778 		netdev_mc_count(netdev) > ENIC_MULTICAST_PERFECT_FILTERS;
779 	unsigned int flags = netdev->flags |
780 		(allmulti ? IFF_ALLMULTI : 0) |
781 		(promisc ? IFF_PROMISC : 0);
782 
783 	if (enic->flags != flags) {
784 		enic->flags = flags;
785 		enic_dev_packet_filter(enic, directed,
786 			multicast, broadcast, promisc, allmulti);
787 	}
788 
789 	if (!promisc) {
790 		__dev_uc_sync(netdev, enic_uc_sync, enic_uc_unsync);
791 		if (!allmulti)
792 			__dev_mc_sync(netdev, enic_mc_sync, enic_mc_unsync);
793 	}
794 }
795 
796 /* netif_tx_lock held, BHs disabled */
797 static void enic_tx_timeout(struct net_device *netdev)
798 {
799 	struct enic *enic = netdev_priv(netdev);
800 	schedule_work(&enic->reset);
801 }
802 
803 static int enic_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
804 {
805 	struct enic *enic = netdev_priv(netdev);
806 	struct enic_port_profile *pp;
807 	int err;
808 
809 	ENIC_PP_BY_INDEX(enic, vf, pp, &err);
810 	if (err)
811 		return err;
812 
813 	if (is_valid_ether_addr(mac) || is_zero_ether_addr(mac)) {
814 		if (vf == PORT_SELF_VF) {
815 			memcpy(pp->vf_mac, mac, ETH_ALEN);
816 			return 0;
817 		} else {
818 			/*
819 			 * For sriov vf's set the mac in hw
820 			 */
821 			ENIC_DEVCMD_PROXY_BY_INDEX(vf, err, enic,
822 				vnic_dev_set_mac_addr, mac);
823 			return enic_dev_status_to_errno(err);
824 		}
825 	} else
826 		return -EINVAL;
827 }
828 
829 static int enic_set_vf_port(struct net_device *netdev, int vf,
830 	struct nlattr *port[])
831 {
832 	struct enic *enic = netdev_priv(netdev);
833 	struct enic_port_profile prev_pp;
834 	struct enic_port_profile *pp;
835 	int err = 0, restore_pp = 1;
836 
837 	ENIC_PP_BY_INDEX(enic, vf, pp, &err);
838 	if (err)
839 		return err;
840 
841 	if (!port[IFLA_PORT_REQUEST])
842 		return -EOPNOTSUPP;
843 
844 	memcpy(&prev_pp, pp, sizeof(*enic->pp));
845 	memset(pp, 0, sizeof(*enic->pp));
846 
847 	pp->set |= ENIC_SET_REQUEST;
848 	pp->request = nla_get_u8(port[IFLA_PORT_REQUEST]);
849 
850 	if (port[IFLA_PORT_PROFILE]) {
851 		pp->set |= ENIC_SET_NAME;
852 		memcpy(pp->name, nla_data(port[IFLA_PORT_PROFILE]),
853 			PORT_PROFILE_MAX);
854 	}
855 
856 	if (port[IFLA_PORT_INSTANCE_UUID]) {
857 		pp->set |= ENIC_SET_INSTANCE;
858 		memcpy(pp->instance_uuid,
859 			nla_data(port[IFLA_PORT_INSTANCE_UUID]), PORT_UUID_MAX);
860 	}
861 
862 	if (port[IFLA_PORT_HOST_UUID]) {
863 		pp->set |= ENIC_SET_HOST;
864 		memcpy(pp->host_uuid,
865 			nla_data(port[IFLA_PORT_HOST_UUID]), PORT_UUID_MAX);
866 	}
867 
868 	if (vf == PORT_SELF_VF) {
869 		/* Special case handling: mac came from IFLA_VF_MAC */
870 		if (!is_zero_ether_addr(prev_pp.vf_mac))
871 			memcpy(pp->mac_addr, prev_pp.vf_mac, ETH_ALEN);
872 
873 		if (is_zero_ether_addr(netdev->dev_addr))
874 			eth_hw_addr_random(netdev);
875 	} else {
876 		/* SR-IOV VF: get mac from adapter */
877 		ENIC_DEVCMD_PROXY_BY_INDEX(vf, err, enic,
878 			vnic_dev_get_mac_addr, pp->mac_addr);
879 		if (err) {
880 			netdev_err(netdev, "Error getting mac for vf %d\n", vf);
881 			memcpy(pp, &prev_pp, sizeof(*pp));
882 			return enic_dev_status_to_errno(err);
883 		}
884 	}
885 
886 	err = enic_process_set_pp_request(enic, vf, &prev_pp, &restore_pp);
887 	if (err) {
888 		if (restore_pp) {
889 			/* Things are still the way they were: Implicit
890 			 * DISASSOCIATE failed
891 			 */
892 			memcpy(pp, &prev_pp, sizeof(*pp));
893 		} else {
894 			memset(pp, 0, sizeof(*pp));
895 			if (vf == PORT_SELF_VF)
896 				memset(netdev->dev_addr, 0, ETH_ALEN);
897 		}
898 	} else {
899 		/* Set flag to indicate that the port assoc/disassoc
900 		 * request has been sent out to fw
901 		 */
902 		pp->set |= ENIC_PORT_REQUEST_APPLIED;
903 
904 		/* If DISASSOCIATE, clean up all assigned/saved macaddresses */
905 		if (pp->request == PORT_REQUEST_DISASSOCIATE) {
906 			memset(pp->mac_addr, 0, ETH_ALEN);
907 			if (vf == PORT_SELF_VF)
908 				memset(netdev->dev_addr, 0, ETH_ALEN);
909 		}
910 	}
911 
912 	if (vf == PORT_SELF_VF)
913 		memset(pp->vf_mac, 0, ETH_ALEN);
914 
915 	return err;
916 }
917 
918 static int enic_get_vf_port(struct net_device *netdev, int vf,
919 	struct sk_buff *skb)
920 {
921 	struct enic *enic = netdev_priv(netdev);
922 	u16 response = PORT_PROFILE_RESPONSE_SUCCESS;
923 	struct enic_port_profile *pp;
924 	int err;
925 
926 	ENIC_PP_BY_INDEX(enic, vf, pp, &err);
927 	if (err)
928 		return err;
929 
930 	if (!(pp->set & ENIC_PORT_REQUEST_APPLIED))
931 		return -ENODATA;
932 
933 	err = enic_process_get_pp_request(enic, vf, pp->request, &response);
934 	if (err)
935 		return err;
936 
937 	if (nla_put_u16(skb, IFLA_PORT_REQUEST, pp->request) ||
938 	    nla_put_u16(skb, IFLA_PORT_RESPONSE, response) ||
939 	    ((pp->set & ENIC_SET_NAME) &&
940 	     nla_put(skb, IFLA_PORT_PROFILE, PORT_PROFILE_MAX, pp->name)) ||
941 	    ((pp->set & ENIC_SET_INSTANCE) &&
942 	     nla_put(skb, IFLA_PORT_INSTANCE_UUID, PORT_UUID_MAX,
943 		     pp->instance_uuid)) ||
944 	    ((pp->set & ENIC_SET_HOST) &&
945 	     nla_put(skb, IFLA_PORT_HOST_UUID, PORT_UUID_MAX, pp->host_uuid)))
946 		goto nla_put_failure;
947 	return 0;
948 
949 nla_put_failure:
950 	return -EMSGSIZE;
951 }
952 
953 static void enic_free_rq_buf(struct vnic_rq *rq, struct vnic_rq_buf *buf)
954 {
955 	struct enic *enic = vnic_dev_priv(rq->vdev);
956 
957 	if (!buf->os_buf)
958 		return;
959 
960 	pci_unmap_single(enic->pdev, buf->dma_addr,
961 		buf->len, PCI_DMA_FROMDEVICE);
962 	dev_kfree_skb_any(buf->os_buf);
963 	buf->os_buf = NULL;
964 }
965 
966 static int enic_rq_alloc_buf(struct vnic_rq *rq)
967 {
968 	struct enic *enic = vnic_dev_priv(rq->vdev);
969 	struct net_device *netdev = enic->netdev;
970 	struct sk_buff *skb;
971 	unsigned int len = netdev->mtu + VLAN_ETH_HLEN;
972 	unsigned int os_buf_index = 0;
973 	dma_addr_t dma_addr;
974 	struct vnic_rq_buf *buf = rq->to_use;
975 
976 	if (buf->os_buf) {
977 		enic_queue_rq_desc(rq, buf->os_buf, os_buf_index, buf->dma_addr,
978 				   buf->len);
979 
980 		return 0;
981 	}
982 	skb = netdev_alloc_skb_ip_align(netdev, len);
983 	if (!skb)
984 		return -ENOMEM;
985 
986 	dma_addr = pci_map_single(enic->pdev, skb->data, len,
987 				  PCI_DMA_FROMDEVICE);
988 	if (unlikely(enic_dma_map_check(enic, dma_addr))) {
989 		dev_kfree_skb(skb);
990 		return -ENOMEM;
991 	}
992 
993 	enic_queue_rq_desc(rq, skb, os_buf_index,
994 		dma_addr, len);
995 
996 	return 0;
997 }
998 
999 static void enic_intr_update_pkt_size(struct vnic_rx_bytes_counter *pkt_size,
1000 				      u32 pkt_len)
1001 {
1002 	if (ENIC_LARGE_PKT_THRESHOLD <= pkt_len)
1003 		pkt_size->large_pkt_bytes_cnt += pkt_len;
1004 	else
1005 		pkt_size->small_pkt_bytes_cnt += pkt_len;
1006 }
1007 
1008 static bool enic_rxcopybreak(struct net_device *netdev, struct sk_buff **skb,
1009 			     struct vnic_rq_buf *buf, u16 len)
1010 {
1011 	struct enic *enic = netdev_priv(netdev);
1012 	struct sk_buff *new_skb;
1013 
1014 	if (len > enic->rx_copybreak)
1015 		return false;
1016 	new_skb = netdev_alloc_skb_ip_align(netdev, len);
1017 	if (!new_skb)
1018 		return false;
1019 	pci_dma_sync_single_for_cpu(enic->pdev, buf->dma_addr, len,
1020 				    DMA_FROM_DEVICE);
1021 	memcpy(new_skb->data, (*skb)->data, len);
1022 	*skb = new_skb;
1023 
1024 	return true;
1025 }
1026 
1027 static void enic_rq_indicate_buf(struct vnic_rq *rq,
1028 	struct cq_desc *cq_desc, struct vnic_rq_buf *buf,
1029 	int skipped, void *opaque)
1030 {
1031 	struct enic *enic = vnic_dev_priv(rq->vdev);
1032 	struct net_device *netdev = enic->netdev;
1033 	struct sk_buff *skb;
1034 	struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
1035 
1036 	u8 type, color, eop, sop, ingress_port, vlan_stripped;
1037 	u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof;
1038 	u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok;
1039 	u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc;
1040 	u8 packet_error;
1041 	u16 q_number, completed_index, bytes_written, vlan_tci, checksum;
1042 	u32 rss_hash;
1043 
1044 	if (skipped)
1045 		return;
1046 
1047 	skb = buf->os_buf;
1048 
1049 	cq_enet_rq_desc_dec((struct cq_enet_rq_desc *)cq_desc,
1050 		&type, &color, &q_number, &completed_index,
1051 		&ingress_port, &fcoe, &eop, &sop, &rss_type,
1052 		&csum_not_calc, &rss_hash, &bytes_written,
1053 		&packet_error, &vlan_stripped, &vlan_tci, &checksum,
1054 		&fcoe_sof, &fcoe_fc_crc_ok, &fcoe_enc_error,
1055 		&fcoe_eof, &tcp_udp_csum_ok, &udp, &tcp,
1056 		&ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment,
1057 		&fcs_ok);
1058 
1059 	if (packet_error) {
1060 
1061 		if (!fcs_ok) {
1062 			if (bytes_written > 0)
1063 				enic->rq_bad_fcs++;
1064 			else if (bytes_written == 0)
1065 				enic->rq_truncated_pkts++;
1066 		}
1067 
1068 		pci_unmap_single(enic->pdev, buf->dma_addr, buf->len,
1069 				 PCI_DMA_FROMDEVICE);
1070 		dev_kfree_skb_any(skb);
1071 		buf->os_buf = NULL;
1072 
1073 		return;
1074 	}
1075 
1076 	if (eop && bytes_written > 0) {
1077 
1078 		/* Good receive
1079 		 */
1080 
1081 		if (!enic_rxcopybreak(netdev, &skb, buf, bytes_written)) {
1082 			buf->os_buf = NULL;
1083 			pci_unmap_single(enic->pdev, buf->dma_addr, buf->len,
1084 					 PCI_DMA_FROMDEVICE);
1085 		}
1086 		prefetch(skb->data - NET_IP_ALIGN);
1087 
1088 		skb_put(skb, bytes_written);
1089 		skb->protocol = eth_type_trans(skb, netdev);
1090 		skb_record_rx_queue(skb, q_number);
1091 		if (netdev->features & NETIF_F_RXHASH) {
1092 			skb_set_hash(skb, rss_hash,
1093 				     (rss_type &
1094 				      (NIC_CFG_RSS_HASH_TYPE_TCP_IPV6_EX |
1095 				       NIC_CFG_RSS_HASH_TYPE_TCP_IPV6 |
1096 				       NIC_CFG_RSS_HASH_TYPE_TCP_IPV4)) ?
1097 				     PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3);
1098 		}
1099 
1100 		/* Hardware does not provide whole packet checksum. It only
1101 		 * provides pseudo checksum. Since hw validates the packet
1102 		 * checksum but not provide us the checksum value. use
1103 		 * CHECSUM_UNNECESSARY.
1104 		 */
1105 		if ((netdev->features & NETIF_F_RXCSUM) && tcp_udp_csum_ok &&
1106 		    ipv4_csum_ok)
1107 			skb->ip_summed = CHECKSUM_UNNECESSARY;
1108 
1109 		if (vlan_stripped)
1110 			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tci);
1111 
1112 		skb_mark_napi_id(skb, &enic->napi[rq->index]);
1113 		if (enic_poll_busy_polling(rq) ||
1114 		    !(netdev->features & NETIF_F_GRO))
1115 			netif_receive_skb(skb);
1116 		else
1117 			napi_gro_receive(&enic->napi[q_number], skb);
1118 		if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1119 			enic_intr_update_pkt_size(&cq->pkt_size_counter,
1120 						  bytes_written);
1121 	} else {
1122 
1123 		/* Buffer overflow
1124 		 */
1125 
1126 		pci_unmap_single(enic->pdev, buf->dma_addr, buf->len,
1127 				 PCI_DMA_FROMDEVICE);
1128 		dev_kfree_skb_any(skb);
1129 		buf->os_buf = NULL;
1130 	}
1131 }
1132 
1133 static int enic_rq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
1134 	u8 type, u16 q_number, u16 completed_index, void *opaque)
1135 {
1136 	struct enic *enic = vnic_dev_priv(vdev);
1137 
1138 	vnic_rq_service(&enic->rq[q_number], cq_desc,
1139 		completed_index, VNIC_RQ_RETURN_DESC,
1140 		enic_rq_indicate_buf, opaque);
1141 
1142 	return 0;
1143 }
1144 
1145 static int enic_poll(struct napi_struct *napi, int budget)
1146 {
1147 	struct net_device *netdev = napi->dev;
1148 	struct enic *enic = netdev_priv(netdev);
1149 	unsigned int cq_rq = enic_cq_rq(enic, 0);
1150 	unsigned int cq_wq = enic_cq_wq(enic, 0);
1151 	unsigned int intr = enic_legacy_io_intr();
1152 	unsigned int rq_work_to_do = budget;
1153 	unsigned int wq_work_to_do = -1; /* no limit */
1154 	unsigned int  work_done, rq_work_done = 0, wq_work_done;
1155 	int err;
1156 
1157 	wq_work_done = vnic_cq_service(&enic->cq[cq_wq], wq_work_to_do,
1158 				       enic_wq_service, NULL);
1159 
1160 	if (!enic_poll_lock_napi(&enic->rq[cq_rq])) {
1161 		if (wq_work_done > 0)
1162 			vnic_intr_return_credits(&enic->intr[intr],
1163 						 wq_work_done,
1164 						 0 /* dont unmask intr */,
1165 						 0 /* dont reset intr timer */);
1166 		return rq_work_done;
1167 	}
1168 
1169 	if (budget > 0)
1170 		rq_work_done = vnic_cq_service(&enic->cq[cq_rq],
1171 			rq_work_to_do, enic_rq_service, NULL);
1172 
1173 	/* Accumulate intr event credits for this polling
1174 	 * cycle.  An intr event is the completion of a
1175 	 * a WQ or RQ packet.
1176 	 */
1177 
1178 	work_done = rq_work_done + wq_work_done;
1179 
1180 	if (work_done > 0)
1181 		vnic_intr_return_credits(&enic->intr[intr],
1182 			work_done,
1183 			0 /* don't unmask intr */,
1184 			0 /* don't reset intr timer */);
1185 
1186 	err = vnic_rq_fill(&enic->rq[0], enic_rq_alloc_buf);
1187 
1188 	/* Buffer allocation failed. Stay in polling
1189 	 * mode so we can try to fill the ring again.
1190 	 */
1191 
1192 	if (err)
1193 		rq_work_done = rq_work_to_do;
1194 
1195 	if (rq_work_done < rq_work_to_do) {
1196 
1197 		/* Some work done, but not enough to stay in polling,
1198 		 * exit polling
1199 		 */
1200 
1201 		napi_complete(napi);
1202 		vnic_intr_unmask(&enic->intr[intr]);
1203 	}
1204 	enic_poll_unlock_napi(&enic->rq[cq_rq]);
1205 
1206 	return rq_work_done;
1207 }
1208 
1209 static void enic_set_int_moderation(struct enic *enic, struct vnic_rq *rq)
1210 {
1211 	unsigned int intr = enic_msix_rq_intr(enic, rq->index);
1212 	struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
1213 	u32 timer = cq->tobe_rx_coal_timeval;
1214 
1215 	if (cq->tobe_rx_coal_timeval != cq->cur_rx_coal_timeval) {
1216 		vnic_intr_coalescing_timer_set(&enic->intr[intr], timer);
1217 		cq->cur_rx_coal_timeval = cq->tobe_rx_coal_timeval;
1218 	}
1219 }
1220 
1221 static void enic_calc_int_moderation(struct enic *enic, struct vnic_rq *rq)
1222 {
1223 	struct enic_rx_coal *rx_coal = &enic->rx_coalesce_setting;
1224 	struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
1225 	struct vnic_rx_bytes_counter *pkt_size_counter = &cq->pkt_size_counter;
1226 	int index;
1227 	u32 timer;
1228 	u32 range_start;
1229 	u32 traffic;
1230 	u64 delta;
1231 	ktime_t now = ktime_get();
1232 
1233 	delta = ktime_us_delta(now, cq->prev_ts);
1234 	if (delta < ENIC_AIC_TS_BREAK)
1235 		return;
1236 	cq->prev_ts = now;
1237 
1238 	traffic = pkt_size_counter->large_pkt_bytes_cnt +
1239 		  pkt_size_counter->small_pkt_bytes_cnt;
1240 	/* The table takes Mbps
1241 	 * traffic *= 8    => bits
1242 	 * traffic *= (10^6 / delta)    => bps
1243 	 * traffic /= 10^6     => Mbps
1244 	 *
1245 	 * Combining, traffic *= (8 / delta)
1246 	 */
1247 
1248 	traffic <<= 3;
1249 	traffic = delta > UINT_MAX ? 0 : traffic / (u32)delta;
1250 
1251 	for (index = 0; index < ENIC_MAX_COALESCE_TIMERS; index++)
1252 		if (traffic < mod_table[index].rx_rate)
1253 			break;
1254 	range_start = (pkt_size_counter->small_pkt_bytes_cnt >
1255 		       pkt_size_counter->large_pkt_bytes_cnt << 1) ?
1256 		      rx_coal->small_pkt_range_start :
1257 		      rx_coal->large_pkt_range_start;
1258 	timer = range_start + ((rx_coal->range_end - range_start) *
1259 			       mod_table[index].range_percent / 100);
1260 	/* Damping */
1261 	cq->tobe_rx_coal_timeval = (timer + cq->tobe_rx_coal_timeval) >> 1;
1262 
1263 	pkt_size_counter->large_pkt_bytes_cnt = 0;
1264 	pkt_size_counter->small_pkt_bytes_cnt = 0;
1265 }
1266 
1267 #ifdef CONFIG_RFS_ACCEL
1268 static void enic_free_rx_cpu_rmap(struct enic *enic)
1269 {
1270 	free_irq_cpu_rmap(enic->netdev->rx_cpu_rmap);
1271 	enic->netdev->rx_cpu_rmap = NULL;
1272 }
1273 
1274 static void enic_set_rx_cpu_rmap(struct enic *enic)
1275 {
1276 	int i, res;
1277 
1278 	if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX) {
1279 		enic->netdev->rx_cpu_rmap = alloc_irq_cpu_rmap(enic->rq_count);
1280 		if (unlikely(!enic->netdev->rx_cpu_rmap))
1281 			return;
1282 		for (i = 0; i < enic->rq_count; i++) {
1283 			res = irq_cpu_rmap_add(enic->netdev->rx_cpu_rmap,
1284 					       enic->msix_entry[i].vector);
1285 			if (unlikely(res)) {
1286 				enic_free_rx_cpu_rmap(enic);
1287 				return;
1288 			}
1289 		}
1290 	}
1291 }
1292 
1293 #else
1294 
1295 static void enic_free_rx_cpu_rmap(struct enic *enic)
1296 {
1297 }
1298 
1299 static void enic_set_rx_cpu_rmap(struct enic *enic)
1300 {
1301 }
1302 
1303 #endif /* CONFIG_RFS_ACCEL */
1304 
1305 #ifdef CONFIG_NET_RX_BUSY_POLL
1306 static int enic_busy_poll(struct napi_struct *napi)
1307 {
1308 	struct net_device *netdev = napi->dev;
1309 	struct enic *enic = netdev_priv(netdev);
1310 	unsigned int rq = (napi - &enic->napi[0]);
1311 	unsigned int cq = enic_cq_rq(enic, rq);
1312 	unsigned int intr = enic_msix_rq_intr(enic, rq);
1313 	unsigned int work_to_do = -1; /* clean all pkts possible */
1314 	unsigned int work_done;
1315 
1316 	if (!enic_poll_lock_poll(&enic->rq[rq]))
1317 		return LL_FLUSH_BUSY;
1318 	work_done = vnic_cq_service(&enic->cq[cq], work_to_do,
1319 				    enic_rq_service, NULL);
1320 
1321 	if (work_done > 0)
1322 		vnic_intr_return_credits(&enic->intr[intr],
1323 					 work_done, 0, 0);
1324 	vnic_rq_fill(&enic->rq[rq], enic_rq_alloc_buf);
1325 	if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1326 		enic_calc_int_moderation(enic, &enic->rq[rq]);
1327 	enic_poll_unlock_poll(&enic->rq[rq]);
1328 
1329 	return work_done;
1330 }
1331 #endif /* CONFIG_NET_RX_BUSY_POLL */
1332 
1333 static int enic_poll_msix_wq(struct napi_struct *napi, int budget)
1334 {
1335 	struct net_device *netdev = napi->dev;
1336 	struct enic *enic = netdev_priv(netdev);
1337 	unsigned int wq_index = (napi - &enic->napi[0]) - enic->rq_count;
1338 	struct vnic_wq *wq = &enic->wq[wq_index];
1339 	unsigned int cq;
1340 	unsigned int intr;
1341 	unsigned int wq_work_to_do = -1; /* clean all desc possible */
1342 	unsigned int wq_work_done;
1343 	unsigned int wq_irq;
1344 
1345 	wq_irq = wq->index;
1346 	cq = enic_cq_wq(enic, wq_irq);
1347 	intr = enic_msix_wq_intr(enic, wq_irq);
1348 	wq_work_done = vnic_cq_service(&enic->cq[cq], wq_work_to_do,
1349 				       enic_wq_service, NULL);
1350 
1351 	vnic_intr_return_credits(&enic->intr[intr], wq_work_done,
1352 				 0 /* don't unmask intr */,
1353 				 1 /* reset intr timer */);
1354 	if (!wq_work_done) {
1355 		napi_complete(napi);
1356 		vnic_intr_unmask(&enic->intr[intr]);
1357 		return 0;
1358 	}
1359 
1360 	return budget;
1361 }
1362 
1363 static int enic_poll_msix_rq(struct napi_struct *napi, int budget)
1364 {
1365 	struct net_device *netdev = napi->dev;
1366 	struct enic *enic = netdev_priv(netdev);
1367 	unsigned int rq = (napi - &enic->napi[0]);
1368 	unsigned int cq = enic_cq_rq(enic, rq);
1369 	unsigned int intr = enic_msix_rq_intr(enic, rq);
1370 	unsigned int work_to_do = budget;
1371 	unsigned int work_done = 0;
1372 	int err;
1373 
1374 	if (!enic_poll_lock_napi(&enic->rq[rq]))
1375 		return budget;
1376 	/* Service RQ
1377 	 */
1378 
1379 	if (budget > 0)
1380 		work_done = vnic_cq_service(&enic->cq[cq],
1381 			work_to_do, enic_rq_service, NULL);
1382 
1383 	/* Return intr event credits for this polling
1384 	 * cycle.  An intr event is the completion of a
1385 	 * RQ packet.
1386 	 */
1387 
1388 	if (work_done > 0)
1389 		vnic_intr_return_credits(&enic->intr[intr],
1390 			work_done,
1391 			0 /* don't unmask intr */,
1392 			0 /* don't reset intr timer */);
1393 
1394 	err = vnic_rq_fill(&enic->rq[rq], enic_rq_alloc_buf);
1395 
1396 	/* Buffer allocation failed. Stay in polling mode
1397 	 * so we can try to fill the ring again.
1398 	 */
1399 
1400 	if (err)
1401 		work_done = work_to_do;
1402 	if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1403 		/* Call the function which refreshes
1404 		 * the intr coalescing timer value based on
1405 		 * the traffic.  This is supported only in
1406 		 * the case of MSI-x mode
1407 		 */
1408 		enic_calc_int_moderation(enic, &enic->rq[rq]);
1409 
1410 	if (work_done < work_to_do) {
1411 
1412 		/* Some work done, but not enough to stay in polling,
1413 		 * exit polling
1414 		 */
1415 
1416 		napi_complete(napi);
1417 		if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1418 			enic_set_int_moderation(enic, &enic->rq[rq]);
1419 		vnic_intr_unmask(&enic->intr[intr]);
1420 	}
1421 	enic_poll_unlock_napi(&enic->rq[rq]);
1422 
1423 	return work_done;
1424 }
1425 
1426 static void enic_notify_timer(unsigned long data)
1427 {
1428 	struct enic *enic = (struct enic *)data;
1429 
1430 	enic_notify_check(enic);
1431 
1432 	mod_timer(&enic->notify_timer,
1433 		round_jiffies(jiffies + ENIC_NOTIFY_TIMER_PERIOD));
1434 }
1435 
1436 static void enic_free_intr(struct enic *enic)
1437 {
1438 	struct net_device *netdev = enic->netdev;
1439 	unsigned int i;
1440 
1441 	enic_free_rx_cpu_rmap(enic);
1442 	switch (vnic_dev_get_intr_mode(enic->vdev)) {
1443 	case VNIC_DEV_INTR_MODE_INTX:
1444 		free_irq(enic->pdev->irq, netdev);
1445 		break;
1446 	case VNIC_DEV_INTR_MODE_MSI:
1447 		free_irq(enic->pdev->irq, enic);
1448 		break;
1449 	case VNIC_DEV_INTR_MODE_MSIX:
1450 		for (i = 0; i < ARRAY_SIZE(enic->msix); i++)
1451 			if (enic->msix[i].requested)
1452 				free_irq(enic->msix_entry[i].vector,
1453 					enic->msix[i].devid);
1454 		break;
1455 	default:
1456 		break;
1457 	}
1458 }
1459 
1460 static int enic_request_intr(struct enic *enic)
1461 {
1462 	struct net_device *netdev = enic->netdev;
1463 	unsigned int i, intr;
1464 	int err = 0;
1465 
1466 	enic_set_rx_cpu_rmap(enic);
1467 	switch (vnic_dev_get_intr_mode(enic->vdev)) {
1468 
1469 	case VNIC_DEV_INTR_MODE_INTX:
1470 
1471 		err = request_irq(enic->pdev->irq, enic_isr_legacy,
1472 			IRQF_SHARED, netdev->name, netdev);
1473 		break;
1474 
1475 	case VNIC_DEV_INTR_MODE_MSI:
1476 
1477 		err = request_irq(enic->pdev->irq, enic_isr_msi,
1478 			0, netdev->name, enic);
1479 		break;
1480 
1481 	case VNIC_DEV_INTR_MODE_MSIX:
1482 
1483 		for (i = 0; i < enic->rq_count; i++) {
1484 			intr = enic_msix_rq_intr(enic, i);
1485 			snprintf(enic->msix[intr].devname,
1486 				sizeof(enic->msix[intr].devname),
1487 				"%.11s-rx-%d", netdev->name, i);
1488 			enic->msix[intr].isr = enic_isr_msix;
1489 			enic->msix[intr].devid = &enic->napi[i];
1490 		}
1491 
1492 		for (i = 0; i < enic->wq_count; i++) {
1493 			int wq = enic_cq_wq(enic, i);
1494 
1495 			intr = enic_msix_wq_intr(enic, i);
1496 			snprintf(enic->msix[intr].devname,
1497 				sizeof(enic->msix[intr].devname),
1498 				"%.11s-tx-%d", netdev->name, i);
1499 			enic->msix[intr].isr = enic_isr_msix;
1500 			enic->msix[intr].devid = &enic->napi[wq];
1501 		}
1502 
1503 		intr = enic_msix_err_intr(enic);
1504 		snprintf(enic->msix[intr].devname,
1505 			sizeof(enic->msix[intr].devname),
1506 			"%.11s-err", netdev->name);
1507 		enic->msix[intr].isr = enic_isr_msix_err;
1508 		enic->msix[intr].devid = enic;
1509 
1510 		intr = enic_msix_notify_intr(enic);
1511 		snprintf(enic->msix[intr].devname,
1512 			sizeof(enic->msix[intr].devname),
1513 			"%.11s-notify", netdev->name);
1514 		enic->msix[intr].isr = enic_isr_msix_notify;
1515 		enic->msix[intr].devid = enic;
1516 
1517 		for (i = 0; i < ARRAY_SIZE(enic->msix); i++)
1518 			enic->msix[i].requested = 0;
1519 
1520 		for (i = 0; i < enic->intr_count; i++) {
1521 			err = request_irq(enic->msix_entry[i].vector,
1522 				enic->msix[i].isr, 0,
1523 				enic->msix[i].devname,
1524 				enic->msix[i].devid);
1525 			if (err) {
1526 				enic_free_intr(enic);
1527 				break;
1528 			}
1529 			enic->msix[i].requested = 1;
1530 		}
1531 
1532 		break;
1533 
1534 	default:
1535 		break;
1536 	}
1537 
1538 	return err;
1539 }
1540 
1541 static void enic_synchronize_irqs(struct enic *enic)
1542 {
1543 	unsigned int i;
1544 
1545 	switch (vnic_dev_get_intr_mode(enic->vdev)) {
1546 	case VNIC_DEV_INTR_MODE_INTX:
1547 	case VNIC_DEV_INTR_MODE_MSI:
1548 		synchronize_irq(enic->pdev->irq);
1549 		break;
1550 	case VNIC_DEV_INTR_MODE_MSIX:
1551 		for (i = 0; i < enic->intr_count; i++)
1552 			synchronize_irq(enic->msix_entry[i].vector);
1553 		break;
1554 	default:
1555 		break;
1556 	}
1557 }
1558 
1559 static void enic_set_rx_coal_setting(struct enic *enic)
1560 {
1561 	unsigned int speed;
1562 	int index = -1;
1563 	struct enic_rx_coal *rx_coal = &enic->rx_coalesce_setting;
1564 
1565 	/* If intr mode is not MSIX, do not do adaptive coalescing */
1566 	if (VNIC_DEV_INTR_MODE_MSIX != vnic_dev_get_intr_mode(enic->vdev)) {
1567 		netdev_info(enic->netdev, "INTR mode is not MSIX, Not initializing adaptive coalescing");
1568 		return;
1569 	}
1570 
1571 	/* 1. Read the link speed from fw
1572 	 * 2. Pick the default range for the speed
1573 	 * 3. Update it in enic->rx_coalesce_setting
1574 	 */
1575 	speed = vnic_dev_port_speed(enic->vdev);
1576 	if (ENIC_LINK_SPEED_10G < speed)
1577 		index = ENIC_LINK_40G_INDEX;
1578 	else if (ENIC_LINK_SPEED_4G < speed)
1579 		index = ENIC_LINK_10G_INDEX;
1580 	else
1581 		index = ENIC_LINK_4G_INDEX;
1582 
1583 	rx_coal->small_pkt_range_start = mod_range[index].small_pkt_range_start;
1584 	rx_coal->large_pkt_range_start = mod_range[index].large_pkt_range_start;
1585 	rx_coal->range_end = ENIC_RX_COALESCE_RANGE_END;
1586 
1587 	/* Start with the value provided by UCSM */
1588 	for (index = 0; index < enic->rq_count; index++)
1589 		enic->cq[index].cur_rx_coal_timeval =
1590 				enic->config.intr_timer_usec;
1591 
1592 	rx_coal->use_adaptive_rx_coalesce = 1;
1593 }
1594 
1595 static int enic_dev_notify_set(struct enic *enic)
1596 {
1597 	int err;
1598 
1599 	spin_lock_bh(&enic->devcmd_lock);
1600 	switch (vnic_dev_get_intr_mode(enic->vdev)) {
1601 	case VNIC_DEV_INTR_MODE_INTX:
1602 		err = vnic_dev_notify_set(enic->vdev,
1603 			enic_legacy_notify_intr());
1604 		break;
1605 	case VNIC_DEV_INTR_MODE_MSIX:
1606 		err = vnic_dev_notify_set(enic->vdev,
1607 			enic_msix_notify_intr(enic));
1608 		break;
1609 	default:
1610 		err = vnic_dev_notify_set(enic->vdev, -1 /* no intr */);
1611 		break;
1612 	}
1613 	spin_unlock_bh(&enic->devcmd_lock);
1614 
1615 	return err;
1616 }
1617 
1618 static void enic_notify_timer_start(struct enic *enic)
1619 {
1620 	switch (vnic_dev_get_intr_mode(enic->vdev)) {
1621 	case VNIC_DEV_INTR_MODE_MSI:
1622 		mod_timer(&enic->notify_timer, jiffies);
1623 		break;
1624 	default:
1625 		/* Using intr for notification for INTx/MSI-X */
1626 		break;
1627 	}
1628 }
1629 
1630 /* rtnl lock is held, process context */
1631 static int enic_open(struct net_device *netdev)
1632 {
1633 	struct enic *enic = netdev_priv(netdev);
1634 	unsigned int i;
1635 	int err;
1636 
1637 	err = enic_request_intr(enic);
1638 	if (err) {
1639 		netdev_err(netdev, "Unable to request irq.\n");
1640 		return err;
1641 	}
1642 
1643 	err = enic_dev_notify_set(enic);
1644 	if (err) {
1645 		netdev_err(netdev,
1646 			"Failed to alloc notify buffer, aborting.\n");
1647 		goto err_out_free_intr;
1648 	}
1649 
1650 	for (i = 0; i < enic->rq_count; i++) {
1651 		vnic_rq_fill(&enic->rq[i], enic_rq_alloc_buf);
1652 		/* Need at least one buffer on ring to get going */
1653 		if (vnic_rq_desc_used(&enic->rq[i]) == 0) {
1654 			netdev_err(netdev, "Unable to alloc receive buffers\n");
1655 			err = -ENOMEM;
1656 			goto err_out_free_rq;
1657 		}
1658 	}
1659 
1660 	for (i = 0; i < enic->wq_count; i++)
1661 		vnic_wq_enable(&enic->wq[i]);
1662 	for (i = 0; i < enic->rq_count; i++)
1663 		vnic_rq_enable(&enic->rq[i]);
1664 
1665 	if (!enic_is_dynamic(enic) && !enic_is_sriov_vf(enic))
1666 		enic_dev_add_station_addr(enic);
1667 
1668 	enic_set_rx_mode(netdev);
1669 
1670 	netif_tx_wake_all_queues(netdev);
1671 
1672 	for (i = 0; i < enic->rq_count; i++) {
1673 		enic_busy_poll_init_lock(&enic->rq[i]);
1674 		napi_enable(&enic->napi[i]);
1675 	}
1676 	if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX)
1677 		for (i = 0; i < enic->wq_count; i++)
1678 			napi_enable(&enic->napi[enic_cq_wq(enic, i)]);
1679 	enic_dev_enable(enic);
1680 
1681 	for (i = 0; i < enic->intr_count; i++)
1682 		vnic_intr_unmask(&enic->intr[i]);
1683 
1684 	enic_notify_timer_start(enic);
1685 	enic_rfs_flw_tbl_init(enic);
1686 
1687 	return 0;
1688 
1689 err_out_free_rq:
1690 	for (i = 0; i < enic->rq_count; i++)
1691 		vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
1692 	enic_dev_notify_unset(enic);
1693 err_out_free_intr:
1694 	enic_free_intr(enic);
1695 
1696 	return err;
1697 }
1698 
1699 /* rtnl lock is held, process context */
1700 static int enic_stop(struct net_device *netdev)
1701 {
1702 	struct enic *enic = netdev_priv(netdev);
1703 	unsigned int i;
1704 	int err;
1705 
1706 	for (i = 0; i < enic->intr_count; i++) {
1707 		vnic_intr_mask(&enic->intr[i]);
1708 		(void)vnic_intr_masked(&enic->intr[i]); /* flush write */
1709 	}
1710 
1711 	enic_synchronize_irqs(enic);
1712 
1713 	del_timer_sync(&enic->notify_timer);
1714 	enic_rfs_flw_tbl_free(enic);
1715 
1716 	enic_dev_disable(enic);
1717 
1718 	for (i = 0; i < enic->rq_count; i++) {
1719 		napi_disable(&enic->napi[i]);
1720 		local_bh_disable();
1721 		while (!enic_poll_lock_napi(&enic->rq[i]))
1722 			mdelay(1);
1723 		local_bh_enable();
1724 	}
1725 
1726 	netif_carrier_off(netdev);
1727 	netif_tx_disable(netdev);
1728 	if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX)
1729 		for (i = 0; i < enic->wq_count; i++)
1730 			napi_disable(&enic->napi[enic_cq_wq(enic, i)]);
1731 
1732 	if (!enic_is_dynamic(enic) && !enic_is_sriov_vf(enic))
1733 		enic_dev_del_station_addr(enic);
1734 
1735 	for (i = 0; i < enic->wq_count; i++) {
1736 		err = vnic_wq_disable(&enic->wq[i]);
1737 		if (err)
1738 			return err;
1739 	}
1740 	for (i = 0; i < enic->rq_count; i++) {
1741 		err = vnic_rq_disable(&enic->rq[i]);
1742 		if (err)
1743 			return err;
1744 	}
1745 
1746 	enic_dev_notify_unset(enic);
1747 	enic_free_intr(enic);
1748 
1749 	for (i = 0; i < enic->wq_count; i++)
1750 		vnic_wq_clean(&enic->wq[i], enic_free_wq_buf);
1751 	for (i = 0; i < enic->rq_count; i++)
1752 		vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
1753 	for (i = 0; i < enic->cq_count; i++)
1754 		vnic_cq_clean(&enic->cq[i]);
1755 	for (i = 0; i < enic->intr_count; i++)
1756 		vnic_intr_clean(&enic->intr[i]);
1757 
1758 	return 0;
1759 }
1760 
1761 static int enic_change_mtu(struct net_device *netdev, int new_mtu)
1762 {
1763 	struct enic *enic = netdev_priv(netdev);
1764 	int running = netif_running(netdev);
1765 
1766 	if (new_mtu < ENIC_MIN_MTU || new_mtu > ENIC_MAX_MTU)
1767 		return -EINVAL;
1768 
1769 	if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic))
1770 		return -EOPNOTSUPP;
1771 
1772 	if (running)
1773 		enic_stop(netdev);
1774 
1775 	netdev->mtu = new_mtu;
1776 
1777 	if (netdev->mtu > enic->port_mtu)
1778 		netdev_warn(netdev,
1779 			"interface MTU (%d) set higher than port MTU (%d)\n",
1780 			netdev->mtu, enic->port_mtu);
1781 
1782 	if (running)
1783 		enic_open(netdev);
1784 
1785 	return 0;
1786 }
1787 
1788 static void enic_change_mtu_work(struct work_struct *work)
1789 {
1790 	struct enic *enic = container_of(work, struct enic, change_mtu_work);
1791 	struct net_device *netdev = enic->netdev;
1792 	int new_mtu = vnic_dev_mtu(enic->vdev);
1793 	int err;
1794 	unsigned int i;
1795 
1796 	new_mtu = max_t(int, ENIC_MIN_MTU, min_t(int, ENIC_MAX_MTU, new_mtu));
1797 
1798 	rtnl_lock();
1799 
1800 	/* Stop RQ */
1801 	del_timer_sync(&enic->notify_timer);
1802 
1803 	for (i = 0; i < enic->rq_count; i++)
1804 		napi_disable(&enic->napi[i]);
1805 
1806 	vnic_intr_mask(&enic->intr[0]);
1807 	enic_synchronize_irqs(enic);
1808 	err = vnic_rq_disable(&enic->rq[0]);
1809 	if (err) {
1810 		rtnl_unlock();
1811 		netdev_err(netdev, "Unable to disable RQ.\n");
1812 		return;
1813 	}
1814 	vnic_rq_clean(&enic->rq[0], enic_free_rq_buf);
1815 	vnic_cq_clean(&enic->cq[0]);
1816 	vnic_intr_clean(&enic->intr[0]);
1817 
1818 	/* Fill RQ with new_mtu-sized buffers */
1819 	netdev->mtu = new_mtu;
1820 	vnic_rq_fill(&enic->rq[0], enic_rq_alloc_buf);
1821 	/* Need at least one buffer on ring to get going */
1822 	if (vnic_rq_desc_used(&enic->rq[0]) == 0) {
1823 		rtnl_unlock();
1824 		netdev_err(netdev, "Unable to alloc receive buffers.\n");
1825 		return;
1826 	}
1827 
1828 	/* Start RQ */
1829 	vnic_rq_enable(&enic->rq[0]);
1830 	napi_enable(&enic->napi[0]);
1831 	vnic_intr_unmask(&enic->intr[0]);
1832 	enic_notify_timer_start(enic);
1833 
1834 	rtnl_unlock();
1835 
1836 	netdev_info(netdev, "interface MTU set as %d\n", netdev->mtu);
1837 }
1838 
1839 #ifdef CONFIG_NET_POLL_CONTROLLER
1840 static void enic_poll_controller(struct net_device *netdev)
1841 {
1842 	struct enic *enic = netdev_priv(netdev);
1843 	struct vnic_dev *vdev = enic->vdev;
1844 	unsigned int i, intr;
1845 
1846 	switch (vnic_dev_get_intr_mode(vdev)) {
1847 	case VNIC_DEV_INTR_MODE_MSIX:
1848 		for (i = 0; i < enic->rq_count; i++) {
1849 			intr = enic_msix_rq_intr(enic, i);
1850 			enic_isr_msix(enic->msix_entry[intr].vector,
1851 				      &enic->napi[i]);
1852 		}
1853 
1854 		for (i = 0; i < enic->wq_count; i++) {
1855 			intr = enic_msix_wq_intr(enic, i);
1856 			enic_isr_msix(enic->msix_entry[intr].vector,
1857 				      &enic->napi[enic_cq_wq(enic, i)]);
1858 		}
1859 
1860 		break;
1861 	case VNIC_DEV_INTR_MODE_MSI:
1862 		enic_isr_msi(enic->pdev->irq, enic);
1863 		break;
1864 	case VNIC_DEV_INTR_MODE_INTX:
1865 		enic_isr_legacy(enic->pdev->irq, netdev);
1866 		break;
1867 	default:
1868 		break;
1869 	}
1870 }
1871 #endif
1872 
1873 static int enic_dev_wait(struct vnic_dev *vdev,
1874 	int (*start)(struct vnic_dev *, int),
1875 	int (*finished)(struct vnic_dev *, int *),
1876 	int arg)
1877 {
1878 	unsigned long time;
1879 	int done;
1880 	int err;
1881 
1882 	BUG_ON(in_interrupt());
1883 
1884 	err = start(vdev, arg);
1885 	if (err)
1886 		return err;
1887 
1888 	/* Wait for func to complete...2 seconds max
1889 	 */
1890 
1891 	time = jiffies + (HZ * 2);
1892 	do {
1893 
1894 		err = finished(vdev, &done);
1895 		if (err)
1896 			return err;
1897 
1898 		if (done)
1899 			return 0;
1900 
1901 		schedule_timeout_uninterruptible(HZ / 10);
1902 
1903 	} while (time_after(time, jiffies));
1904 
1905 	return -ETIMEDOUT;
1906 }
1907 
1908 static int enic_dev_open(struct enic *enic)
1909 {
1910 	int err;
1911 
1912 	err = enic_dev_wait(enic->vdev, vnic_dev_open,
1913 		vnic_dev_open_done, 0);
1914 	if (err)
1915 		dev_err(enic_get_dev(enic), "vNIC device open failed, err %d\n",
1916 			err);
1917 
1918 	return err;
1919 }
1920 
1921 static int enic_dev_hang_reset(struct enic *enic)
1922 {
1923 	int err;
1924 
1925 	err = enic_dev_wait(enic->vdev, vnic_dev_hang_reset,
1926 		vnic_dev_hang_reset_done, 0);
1927 	if (err)
1928 		netdev_err(enic->netdev, "vNIC hang reset failed, err %d\n",
1929 			err);
1930 
1931 	return err;
1932 }
1933 
1934 int __enic_set_rsskey(struct enic *enic)
1935 {
1936 	union vnic_rss_key *rss_key_buf_va;
1937 	dma_addr_t rss_key_buf_pa;
1938 	int i, kidx, bidx, err;
1939 
1940 	rss_key_buf_va = pci_zalloc_consistent(enic->pdev,
1941 					       sizeof(union vnic_rss_key),
1942 					       &rss_key_buf_pa);
1943 	if (!rss_key_buf_va)
1944 		return -ENOMEM;
1945 
1946 	for (i = 0; i < ENIC_RSS_LEN; i++) {
1947 		kidx = i / ENIC_RSS_BYTES_PER_KEY;
1948 		bidx = i % ENIC_RSS_BYTES_PER_KEY;
1949 		rss_key_buf_va->key[kidx].b[bidx] = enic->rss_key[i];
1950 	}
1951 	spin_lock_bh(&enic->devcmd_lock);
1952 	err = enic_set_rss_key(enic,
1953 		rss_key_buf_pa,
1954 		sizeof(union vnic_rss_key));
1955 	spin_unlock_bh(&enic->devcmd_lock);
1956 
1957 	pci_free_consistent(enic->pdev, sizeof(union vnic_rss_key),
1958 		rss_key_buf_va, rss_key_buf_pa);
1959 
1960 	return err;
1961 }
1962 
1963 static int enic_set_rsskey(struct enic *enic)
1964 {
1965 	netdev_rss_key_fill(enic->rss_key, ENIC_RSS_LEN);
1966 
1967 	return __enic_set_rsskey(enic);
1968 }
1969 
1970 static int enic_set_rsscpu(struct enic *enic, u8 rss_hash_bits)
1971 {
1972 	dma_addr_t rss_cpu_buf_pa;
1973 	union vnic_rss_cpu *rss_cpu_buf_va = NULL;
1974 	unsigned int i;
1975 	int err;
1976 
1977 	rss_cpu_buf_va = pci_alloc_consistent(enic->pdev,
1978 		sizeof(union vnic_rss_cpu), &rss_cpu_buf_pa);
1979 	if (!rss_cpu_buf_va)
1980 		return -ENOMEM;
1981 
1982 	for (i = 0; i < (1 << rss_hash_bits); i++)
1983 		(*rss_cpu_buf_va).cpu[i/4].b[i%4] = i % enic->rq_count;
1984 
1985 	spin_lock_bh(&enic->devcmd_lock);
1986 	err = enic_set_rss_cpu(enic,
1987 		rss_cpu_buf_pa,
1988 		sizeof(union vnic_rss_cpu));
1989 	spin_unlock_bh(&enic->devcmd_lock);
1990 
1991 	pci_free_consistent(enic->pdev, sizeof(union vnic_rss_cpu),
1992 		rss_cpu_buf_va, rss_cpu_buf_pa);
1993 
1994 	return err;
1995 }
1996 
1997 static int enic_set_niccfg(struct enic *enic, u8 rss_default_cpu,
1998 	u8 rss_hash_type, u8 rss_hash_bits, u8 rss_base_cpu, u8 rss_enable)
1999 {
2000 	const u8 tso_ipid_split_en = 0;
2001 	const u8 ig_vlan_strip_en = 1;
2002 	int err;
2003 
2004 	/* Enable VLAN tag stripping.
2005 	*/
2006 
2007 	spin_lock_bh(&enic->devcmd_lock);
2008 	err = enic_set_nic_cfg(enic,
2009 		rss_default_cpu, rss_hash_type,
2010 		rss_hash_bits, rss_base_cpu,
2011 		rss_enable, tso_ipid_split_en,
2012 		ig_vlan_strip_en);
2013 	spin_unlock_bh(&enic->devcmd_lock);
2014 
2015 	return err;
2016 }
2017 
2018 static int enic_set_rss_nic_cfg(struct enic *enic)
2019 {
2020 	struct device *dev = enic_get_dev(enic);
2021 	const u8 rss_default_cpu = 0;
2022 	const u8 rss_hash_type = NIC_CFG_RSS_HASH_TYPE_IPV4 |
2023 		NIC_CFG_RSS_HASH_TYPE_TCP_IPV4 |
2024 		NIC_CFG_RSS_HASH_TYPE_IPV6 |
2025 		NIC_CFG_RSS_HASH_TYPE_TCP_IPV6;
2026 	const u8 rss_hash_bits = 7;
2027 	const u8 rss_base_cpu = 0;
2028 	u8 rss_enable = ENIC_SETTING(enic, RSS) && (enic->rq_count > 1);
2029 
2030 	if (rss_enable) {
2031 		if (!enic_set_rsskey(enic)) {
2032 			if (enic_set_rsscpu(enic, rss_hash_bits)) {
2033 				rss_enable = 0;
2034 				dev_warn(dev, "RSS disabled, "
2035 					"Failed to set RSS cpu indirection table.");
2036 			}
2037 		} else {
2038 			rss_enable = 0;
2039 			dev_warn(dev, "RSS disabled, Failed to set RSS key.\n");
2040 		}
2041 	}
2042 
2043 	return enic_set_niccfg(enic, rss_default_cpu, rss_hash_type,
2044 		rss_hash_bits, rss_base_cpu, rss_enable);
2045 }
2046 
2047 static void enic_reset(struct work_struct *work)
2048 {
2049 	struct enic *enic = container_of(work, struct enic, reset);
2050 
2051 	if (!netif_running(enic->netdev))
2052 		return;
2053 
2054 	rtnl_lock();
2055 
2056 	spin_lock(&enic->enic_api_lock);
2057 	enic_dev_hang_notify(enic);
2058 	enic_stop(enic->netdev);
2059 	enic_dev_hang_reset(enic);
2060 	enic_reset_addr_lists(enic);
2061 	enic_init_vnic_resources(enic);
2062 	enic_set_rss_nic_cfg(enic);
2063 	enic_dev_set_ig_vlan_rewrite_mode(enic);
2064 	enic_open(enic->netdev);
2065 	spin_unlock(&enic->enic_api_lock);
2066 	call_netdevice_notifiers(NETDEV_REBOOT, enic->netdev);
2067 
2068 	rtnl_unlock();
2069 }
2070 
2071 static int enic_set_intr_mode(struct enic *enic)
2072 {
2073 	unsigned int n = min_t(unsigned int, enic->rq_count, ENIC_RQ_MAX);
2074 	unsigned int m = min_t(unsigned int, enic->wq_count, ENIC_WQ_MAX);
2075 	unsigned int i;
2076 
2077 	/* Set interrupt mode (INTx, MSI, MSI-X) depending
2078 	 * on system capabilities.
2079 	 *
2080 	 * Try MSI-X first
2081 	 *
2082 	 * We need n RQs, m WQs, n+m CQs, and n+m+2 INTRs
2083 	 * (the second to last INTR is used for WQ/RQ errors)
2084 	 * (the last INTR is used for notifications)
2085 	 */
2086 
2087 	BUG_ON(ARRAY_SIZE(enic->msix_entry) < n + m + 2);
2088 	for (i = 0; i < n + m + 2; i++)
2089 		enic->msix_entry[i].entry = i;
2090 
2091 	/* Use multiple RQs if RSS is enabled
2092 	 */
2093 
2094 	if (ENIC_SETTING(enic, RSS) &&
2095 	    enic->config.intr_mode < 1 &&
2096 	    enic->rq_count >= n &&
2097 	    enic->wq_count >= m &&
2098 	    enic->cq_count >= n + m &&
2099 	    enic->intr_count >= n + m + 2) {
2100 
2101 		if (pci_enable_msix_range(enic->pdev, enic->msix_entry,
2102 					  n + m + 2, n + m + 2) > 0) {
2103 
2104 			enic->rq_count = n;
2105 			enic->wq_count = m;
2106 			enic->cq_count = n + m;
2107 			enic->intr_count = n + m + 2;
2108 
2109 			vnic_dev_set_intr_mode(enic->vdev,
2110 				VNIC_DEV_INTR_MODE_MSIX);
2111 
2112 			return 0;
2113 		}
2114 	}
2115 
2116 	if (enic->config.intr_mode < 1 &&
2117 	    enic->rq_count >= 1 &&
2118 	    enic->wq_count >= m &&
2119 	    enic->cq_count >= 1 + m &&
2120 	    enic->intr_count >= 1 + m + 2) {
2121 		if (pci_enable_msix_range(enic->pdev, enic->msix_entry,
2122 					  1 + m + 2, 1 + m + 2) > 0) {
2123 
2124 			enic->rq_count = 1;
2125 			enic->wq_count = m;
2126 			enic->cq_count = 1 + m;
2127 			enic->intr_count = 1 + m + 2;
2128 
2129 			vnic_dev_set_intr_mode(enic->vdev,
2130 				VNIC_DEV_INTR_MODE_MSIX);
2131 
2132 			return 0;
2133 		}
2134 	}
2135 
2136 	/* Next try MSI
2137 	 *
2138 	 * We need 1 RQ, 1 WQ, 2 CQs, and 1 INTR
2139 	 */
2140 
2141 	if (enic->config.intr_mode < 2 &&
2142 	    enic->rq_count >= 1 &&
2143 	    enic->wq_count >= 1 &&
2144 	    enic->cq_count >= 2 &&
2145 	    enic->intr_count >= 1 &&
2146 	    !pci_enable_msi(enic->pdev)) {
2147 
2148 		enic->rq_count = 1;
2149 		enic->wq_count = 1;
2150 		enic->cq_count = 2;
2151 		enic->intr_count = 1;
2152 
2153 		vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSI);
2154 
2155 		return 0;
2156 	}
2157 
2158 	/* Next try INTx
2159 	 *
2160 	 * We need 1 RQ, 1 WQ, 2 CQs, and 3 INTRs
2161 	 * (the first INTR is used for WQ/RQ)
2162 	 * (the second INTR is used for WQ/RQ errors)
2163 	 * (the last INTR is used for notifications)
2164 	 */
2165 
2166 	if (enic->config.intr_mode < 3 &&
2167 	    enic->rq_count >= 1 &&
2168 	    enic->wq_count >= 1 &&
2169 	    enic->cq_count >= 2 &&
2170 	    enic->intr_count >= 3) {
2171 
2172 		enic->rq_count = 1;
2173 		enic->wq_count = 1;
2174 		enic->cq_count = 2;
2175 		enic->intr_count = 3;
2176 
2177 		vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_INTX);
2178 
2179 		return 0;
2180 	}
2181 
2182 	vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
2183 
2184 	return -EINVAL;
2185 }
2186 
2187 static void enic_clear_intr_mode(struct enic *enic)
2188 {
2189 	switch (vnic_dev_get_intr_mode(enic->vdev)) {
2190 	case VNIC_DEV_INTR_MODE_MSIX:
2191 		pci_disable_msix(enic->pdev);
2192 		break;
2193 	case VNIC_DEV_INTR_MODE_MSI:
2194 		pci_disable_msi(enic->pdev);
2195 		break;
2196 	default:
2197 		break;
2198 	}
2199 
2200 	vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
2201 }
2202 
2203 static const struct net_device_ops enic_netdev_dynamic_ops = {
2204 	.ndo_open		= enic_open,
2205 	.ndo_stop		= enic_stop,
2206 	.ndo_start_xmit		= enic_hard_start_xmit,
2207 	.ndo_get_stats64	= enic_get_stats,
2208 	.ndo_validate_addr	= eth_validate_addr,
2209 	.ndo_set_rx_mode	= enic_set_rx_mode,
2210 	.ndo_set_mac_address	= enic_set_mac_address_dynamic,
2211 	.ndo_change_mtu		= enic_change_mtu,
2212 	.ndo_vlan_rx_add_vid	= enic_vlan_rx_add_vid,
2213 	.ndo_vlan_rx_kill_vid	= enic_vlan_rx_kill_vid,
2214 	.ndo_tx_timeout		= enic_tx_timeout,
2215 	.ndo_set_vf_port	= enic_set_vf_port,
2216 	.ndo_get_vf_port	= enic_get_vf_port,
2217 	.ndo_set_vf_mac		= enic_set_vf_mac,
2218 #ifdef CONFIG_NET_POLL_CONTROLLER
2219 	.ndo_poll_controller	= enic_poll_controller,
2220 #endif
2221 #ifdef CONFIG_RFS_ACCEL
2222 	.ndo_rx_flow_steer	= enic_rx_flow_steer,
2223 #endif
2224 #ifdef CONFIG_NET_RX_BUSY_POLL
2225 	.ndo_busy_poll		= enic_busy_poll,
2226 #endif
2227 };
2228 
2229 static const struct net_device_ops enic_netdev_ops = {
2230 	.ndo_open		= enic_open,
2231 	.ndo_stop		= enic_stop,
2232 	.ndo_start_xmit		= enic_hard_start_xmit,
2233 	.ndo_get_stats64	= enic_get_stats,
2234 	.ndo_validate_addr	= eth_validate_addr,
2235 	.ndo_set_mac_address	= enic_set_mac_address,
2236 	.ndo_set_rx_mode	= enic_set_rx_mode,
2237 	.ndo_change_mtu		= enic_change_mtu,
2238 	.ndo_vlan_rx_add_vid	= enic_vlan_rx_add_vid,
2239 	.ndo_vlan_rx_kill_vid	= enic_vlan_rx_kill_vid,
2240 	.ndo_tx_timeout		= enic_tx_timeout,
2241 	.ndo_set_vf_port	= enic_set_vf_port,
2242 	.ndo_get_vf_port	= enic_get_vf_port,
2243 	.ndo_set_vf_mac		= enic_set_vf_mac,
2244 #ifdef CONFIG_NET_POLL_CONTROLLER
2245 	.ndo_poll_controller	= enic_poll_controller,
2246 #endif
2247 #ifdef CONFIG_RFS_ACCEL
2248 	.ndo_rx_flow_steer	= enic_rx_flow_steer,
2249 #endif
2250 #ifdef CONFIG_NET_RX_BUSY_POLL
2251 	.ndo_busy_poll		= enic_busy_poll,
2252 #endif
2253 };
2254 
2255 static void enic_dev_deinit(struct enic *enic)
2256 {
2257 	unsigned int i;
2258 
2259 	for (i = 0; i < enic->rq_count; i++) {
2260 		napi_hash_del(&enic->napi[i]);
2261 		netif_napi_del(&enic->napi[i]);
2262 	}
2263 	if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX)
2264 		for (i = 0; i < enic->wq_count; i++)
2265 			netif_napi_del(&enic->napi[enic_cq_wq(enic, i)]);
2266 
2267 	enic_free_vnic_resources(enic);
2268 	enic_clear_intr_mode(enic);
2269 }
2270 
2271 static void enic_kdump_kernel_config(struct enic *enic)
2272 {
2273 	if (is_kdump_kernel()) {
2274 		dev_info(enic_get_dev(enic), "Running from within kdump kernel. Using minimal resources\n");
2275 		enic->rq_count = 1;
2276 		enic->wq_count = 1;
2277 		enic->config.rq_desc_count = ENIC_MIN_RQ_DESCS;
2278 		enic->config.wq_desc_count = ENIC_MIN_WQ_DESCS;
2279 		enic->config.mtu = min_t(u16, 1500, enic->config.mtu);
2280 	}
2281 }
2282 
2283 static int enic_dev_init(struct enic *enic)
2284 {
2285 	struct device *dev = enic_get_dev(enic);
2286 	struct net_device *netdev = enic->netdev;
2287 	unsigned int i;
2288 	int err;
2289 
2290 	/* Get interrupt coalesce timer info */
2291 	err = enic_dev_intr_coal_timer_info(enic);
2292 	if (err) {
2293 		dev_warn(dev, "Using default conversion factor for "
2294 			"interrupt coalesce timer\n");
2295 		vnic_dev_intr_coal_timer_info_default(enic->vdev);
2296 	}
2297 
2298 	/* Get vNIC configuration
2299 	 */
2300 
2301 	err = enic_get_vnic_config(enic);
2302 	if (err) {
2303 		dev_err(dev, "Get vNIC configuration failed, aborting\n");
2304 		return err;
2305 	}
2306 
2307 	/* Get available resource counts
2308 	 */
2309 
2310 	enic_get_res_counts(enic);
2311 
2312 	/* modify resource count if we are in kdump_kernel
2313 	 */
2314 	enic_kdump_kernel_config(enic);
2315 
2316 	/* Set interrupt mode based on resource counts and system
2317 	 * capabilities
2318 	 */
2319 
2320 	err = enic_set_intr_mode(enic);
2321 	if (err) {
2322 		dev_err(dev, "Failed to set intr mode based on resource "
2323 			"counts and system capabilities, aborting\n");
2324 		return err;
2325 	}
2326 
2327 	/* Allocate and configure vNIC resources
2328 	 */
2329 
2330 	err = enic_alloc_vnic_resources(enic);
2331 	if (err) {
2332 		dev_err(dev, "Failed to alloc vNIC resources, aborting\n");
2333 		goto err_out_free_vnic_resources;
2334 	}
2335 
2336 	enic_init_vnic_resources(enic);
2337 
2338 	err = enic_set_rss_nic_cfg(enic);
2339 	if (err) {
2340 		dev_err(dev, "Failed to config nic, aborting\n");
2341 		goto err_out_free_vnic_resources;
2342 	}
2343 
2344 	switch (vnic_dev_get_intr_mode(enic->vdev)) {
2345 	default:
2346 		netif_napi_add(netdev, &enic->napi[0], enic_poll, 64);
2347 		napi_hash_add(&enic->napi[0]);
2348 		break;
2349 	case VNIC_DEV_INTR_MODE_MSIX:
2350 		for (i = 0; i < enic->rq_count; i++) {
2351 			netif_napi_add(netdev, &enic->napi[i],
2352 				enic_poll_msix_rq, NAPI_POLL_WEIGHT);
2353 			napi_hash_add(&enic->napi[i]);
2354 		}
2355 		for (i = 0; i < enic->wq_count; i++)
2356 			netif_napi_add(netdev, &enic->napi[enic_cq_wq(enic, i)],
2357 				       enic_poll_msix_wq, NAPI_POLL_WEIGHT);
2358 		break;
2359 	}
2360 
2361 	return 0;
2362 
2363 err_out_free_vnic_resources:
2364 	enic_clear_intr_mode(enic);
2365 	enic_free_vnic_resources(enic);
2366 
2367 	return err;
2368 }
2369 
2370 static void enic_iounmap(struct enic *enic)
2371 {
2372 	unsigned int i;
2373 
2374 	for (i = 0; i < ARRAY_SIZE(enic->bar); i++)
2375 		if (enic->bar[i].vaddr)
2376 			iounmap(enic->bar[i].vaddr);
2377 }
2378 
2379 static int enic_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2380 {
2381 	struct device *dev = &pdev->dev;
2382 	struct net_device *netdev;
2383 	struct enic *enic;
2384 	int using_dac = 0;
2385 	unsigned int i;
2386 	int err;
2387 #ifdef CONFIG_PCI_IOV
2388 	int pos = 0;
2389 #endif
2390 	int num_pps = 1;
2391 
2392 	/* Allocate net device structure and initialize.  Private
2393 	 * instance data is initialized to zero.
2394 	 */
2395 
2396 	netdev = alloc_etherdev_mqs(sizeof(struct enic),
2397 				    ENIC_RQ_MAX, ENIC_WQ_MAX);
2398 	if (!netdev)
2399 		return -ENOMEM;
2400 
2401 	pci_set_drvdata(pdev, netdev);
2402 
2403 	SET_NETDEV_DEV(netdev, &pdev->dev);
2404 
2405 	enic = netdev_priv(netdev);
2406 	enic->netdev = netdev;
2407 	enic->pdev = pdev;
2408 
2409 	/* Setup PCI resources
2410 	 */
2411 
2412 	err = pci_enable_device_mem(pdev);
2413 	if (err) {
2414 		dev_err(dev, "Cannot enable PCI device, aborting\n");
2415 		goto err_out_free_netdev;
2416 	}
2417 
2418 	err = pci_request_regions(pdev, DRV_NAME);
2419 	if (err) {
2420 		dev_err(dev, "Cannot request PCI regions, aborting\n");
2421 		goto err_out_disable_device;
2422 	}
2423 
2424 	pci_set_master(pdev);
2425 
2426 	/* Query PCI controller on system for DMA addressing
2427 	 * limitation for the device.  Try 64-bit first, and
2428 	 * fail to 32-bit.
2429 	 */
2430 
2431 	err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
2432 	if (err) {
2433 		err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
2434 		if (err) {
2435 			dev_err(dev, "No usable DMA configuration, aborting\n");
2436 			goto err_out_release_regions;
2437 		}
2438 		err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
2439 		if (err) {
2440 			dev_err(dev, "Unable to obtain %u-bit DMA "
2441 				"for consistent allocations, aborting\n", 32);
2442 			goto err_out_release_regions;
2443 		}
2444 	} else {
2445 		err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
2446 		if (err) {
2447 			dev_err(dev, "Unable to obtain %u-bit DMA "
2448 				"for consistent allocations, aborting\n", 64);
2449 			goto err_out_release_regions;
2450 		}
2451 		using_dac = 1;
2452 	}
2453 
2454 	/* Map vNIC resources from BAR0-5
2455 	 */
2456 
2457 	for (i = 0; i < ARRAY_SIZE(enic->bar); i++) {
2458 		if (!(pci_resource_flags(pdev, i) & IORESOURCE_MEM))
2459 			continue;
2460 		enic->bar[i].len = pci_resource_len(pdev, i);
2461 		enic->bar[i].vaddr = pci_iomap(pdev, i, enic->bar[i].len);
2462 		if (!enic->bar[i].vaddr) {
2463 			dev_err(dev, "Cannot memory-map BAR %d, aborting\n", i);
2464 			err = -ENODEV;
2465 			goto err_out_iounmap;
2466 		}
2467 		enic->bar[i].bus_addr = pci_resource_start(pdev, i);
2468 	}
2469 
2470 	/* Register vNIC device
2471 	 */
2472 
2473 	enic->vdev = vnic_dev_register(NULL, enic, pdev, enic->bar,
2474 		ARRAY_SIZE(enic->bar));
2475 	if (!enic->vdev) {
2476 		dev_err(dev, "vNIC registration failed, aborting\n");
2477 		err = -ENODEV;
2478 		goto err_out_iounmap;
2479 	}
2480 
2481 #ifdef CONFIG_PCI_IOV
2482 	/* Get number of subvnics */
2483 	pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_SRIOV);
2484 	if (pos) {
2485 		pci_read_config_word(pdev, pos + PCI_SRIOV_TOTAL_VF,
2486 			&enic->num_vfs);
2487 		if (enic->num_vfs) {
2488 			err = pci_enable_sriov(pdev, enic->num_vfs);
2489 			if (err) {
2490 				dev_err(dev, "SRIOV enable failed, aborting."
2491 					" pci_enable_sriov() returned %d\n",
2492 					err);
2493 				goto err_out_vnic_unregister;
2494 			}
2495 			enic->priv_flags |= ENIC_SRIOV_ENABLED;
2496 			num_pps = enic->num_vfs;
2497 		}
2498 	}
2499 #endif
2500 
2501 	/* Allocate structure for port profiles */
2502 	enic->pp = kcalloc(num_pps, sizeof(*enic->pp), GFP_KERNEL);
2503 	if (!enic->pp) {
2504 		err = -ENOMEM;
2505 		goto err_out_disable_sriov_pp;
2506 	}
2507 
2508 	/* Issue device open to get device in known state
2509 	 */
2510 
2511 	err = enic_dev_open(enic);
2512 	if (err) {
2513 		dev_err(dev, "vNIC dev open failed, aborting\n");
2514 		goto err_out_disable_sriov;
2515 	}
2516 
2517 	/* Setup devcmd lock
2518 	 */
2519 
2520 	spin_lock_init(&enic->devcmd_lock);
2521 	spin_lock_init(&enic->enic_api_lock);
2522 
2523 	/*
2524 	 * Set ingress vlan rewrite mode before vnic initialization
2525 	 */
2526 
2527 	err = enic_dev_set_ig_vlan_rewrite_mode(enic);
2528 	if (err) {
2529 		dev_err(dev,
2530 			"Failed to set ingress vlan rewrite mode, aborting.\n");
2531 		goto err_out_dev_close;
2532 	}
2533 
2534 	/* Issue device init to initialize the vnic-to-switch link.
2535 	 * We'll start with carrier off and wait for link UP
2536 	 * notification later to turn on carrier.  We don't need
2537 	 * to wait here for the vnic-to-switch link initialization
2538 	 * to complete; link UP notification is the indication that
2539 	 * the process is complete.
2540 	 */
2541 
2542 	netif_carrier_off(netdev);
2543 
2544 	/* Do not call dev_init for a dynamic vnic.
2545 	 * For a dynamic vnic, init_prov_info will be
2546 	 * called later by an upper layer.
2547 	 */
2548 
2549 	if (!enic_is_dynamic(enic)) {
2550 		err = vnic_dev_init(enic->vdev, 0);
2551 		if (err) {
2552 			dev_err(dev, "vNIC dev init failed, aborting\n");
2553 			goto err_out_dev_close;
2554 		}
2555 	}
2556 
2557 	err = enic_dev_init(enic);
2558 	if (err) {
2559 		dev_err(dev, "Device initialization failed, aborting\n");
2560 		goto err_out_dev_close;
2561 	}
2562 
2563 	netif_set_real_num_tx_queues(netdev, enic->wq_count);
2564 	netif_set_real_num_rx_queues(netdev, enic->rq_count);
2565 
2566 	/* Setup notification timer, HW reset task, and wq locks
2567 	 */
2568 
2569 	init_timer(&enic->notify_timer);
2570 	enic->notify_timer.function = enic_notify_timer;
2571 	enic->notify_timer.data = (unsigned long)enic;
2572 
2573 	enic_set_rx_coal_setting(enic);
2574 	INIT_WORK(&enic->reset, enic_reset);
2575 	INIT_WORK(&enic->change_mtu_work, enic_change_mtu_work);
2576 
2577 	for (i = 0; i < enic->wq_count; i++)
2578 		spin_lock_init(&enic->wq_lock[i]);
2579 
2580 	/* Register net device
2581 	 */
2582 
2583 	enic->port_mtu = enic->config.mtu;
2584 	(void)enic_change_mtu(netdev, enic->port_mtu);
2585 
2586 	err = enic_set_mac_addr(netdev, enic->mac_addr);
2587 	if (err) {
2588 		dev_err(dev, "Invalid MAC address, aborting\n");
2589 		goto err_out_dev_deinit;
2590 	}
2591 
2592 	enic->tx_coalesce_usecs = enic->config.intr_timer_usec;
2593 	/* rx coalesce time already got initialized. This gets used
2594 	 * if adaptive coal is turned off
2595 	 */
2596 	enic->rx_coalesce_usecs = enic->tx_coalesce_usecs;
2597 
2598 	if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic))
2599 		netdev->netdev_ops = &enic_netdev_dynamic_ops;
2600 	else
2601 		netdev->netdev_ops = &enic_netdev_ops;
2602 
2603 	netdev->watchdog_timeo = 2 * HZ;
2604 	enic_set_ethtool_ops(netdev);
2605 
2606 	netdev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
2607 	if (ENIC_SETTING(enic, LOOP)) {
2608 		netdev->features &= ~NETIF_F_HW_VLAN_CTAG_TX;
2609 		enic->loop_enable = 1;
2610 		enic->loop_tag = enic->config.loop_tag;
2611 		dev_info(dev, "loopback tag=0x%04x\n", enic->loop_tag);
2612 	}
2613 	if (ENIC_SETTING(enic, TXCSUM))
2614 		netdev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM;
2615 	if (ENIC_SETTING(enic, TSO))
2616 		netdev->hw_features |= NETIF_F_TSO |
2617 			NETIF_F_TSO6 | NETIF_F_TSO_ECN;
2618 	if (ENIC_SETTING(enic, RSS))
2619 		netdev->hw_features |= NETIF_F_RXHASH;
2620 	if (ENIC_SETTING(enic, RXCSUM))
2621 		netdev->hw_features |= NETIF_F_RXCSUM;
2622 
2623 	netdev->features |= netdev->hw_features;
2624 
2625 #ifdef CONFIG_RFS_ACCEL
2626 	netdev->hw_features |= NETIF_F_NTUPLE;
2627 #endif
2628 
2629 	if (using_dac)
2630 		netdev->features |= NETIF_F_HIGHDMA;
2631 
2632 	netdev->priv_flags |= IFF_UNICAST_FLT;
2633 
2634 	err = register_netdev(netdev);
2635 	if (err) {
2636 		dev_err(dev, "Cannot register net device, aborting\n");
2637 		goto err_out_dev_deinit;
2638 	}
2639 	enic->rx_copybreak = RX_COPYBREAK_DEFAULT;
2640 
2641 	return 0;
2642 
2643 err_out_dev_deinit:
2644 	enic_dev_deinit(enic);
2645 err_out_dev_close:
2646 	vnic_dev_close(enic->vdev);
2647 err_out_disable_sriov:
2648 	kfree(enic->pp);
2649 err_out_disable_sriov_pp:
2650 #ifdef CONFIG_PCI_IOV
2651 	if (enic_sriov_enabled(enic)) {
2652 		pci_disable_sriov(pdev);
2653 		enic->priv_flags &= ~ENIC_SRIOV_ENABLED;
2654 	}
2655 err_out_vnic_unregister:
2656 #endif
2657 	vnic_dev_unregister(enic->vdev);
2658 err_out_iounmap:
2659 	enic_iounmap(enic);
2660 err_out_release_regions:
2661 	pci_release_regions(pdev);
2662 err_out_disable_device:
2663 	pci_disable_device(pdev);
2664 err_out_free_netdev:
2665 	free_netdev(netdev);
2666 
2667 	return err;
2668 }
2669 
2670 static void enic_remove(struct pci_dev *pdev)
2671 {
2672 	struct net_device *netdev = pci_get_drvdata(pdev);
2673 
2674 	if (netdev) {
2675 		struct enic *enic = netdev_priv(netdev);
2676 
2677 		cancel_work_sync(&enic->reset);
2678 		cancel_work_sync(&enic->change_mtu_work);
2679 		unregister_netdev(netdev);
2680 		enic_dev_deinit(enic);
2681 		vnic_dev_close(enic->vdev);
2682 #ifdef CONFIG_PCI_IOV
2683 		if (enic_sriov_enabled(enic)) {
2684 			pci_disable_sriov(pdev);
2685 			enic->priv_flags &= ~ENIC_SRIOV_ENABLED;
2686 		}
2687 #endif
2688 		kfree(enic->pp);
2689 		vnic_dev_unregister(enic->vdev);
2690 		enic_iounmap(enic);
2691 		pci_release_regions(pdev);
2692 		pci_disable_device(pdev);
2693 		free_netdev(netdev);
2694 	}
2695 }
2696 
2697 static struct pci_driver enic_driver = {
2698 	.name = DRV_NAME,
2699 	.id_table = enic_id_table,
2700 	.probe = enic_probe,
2701 	.remove = enic_remove,
2702 };
2703 
2704 static int __init enic_init_module(void)
2705 {
2706 	pr_info("%s, ver %s\n", DRV_DESCRIPTION, DRV_VERSION);
2707 
2708 	return pci_register_driver(&enic_driver);
2709 }
2710 
2711 static void __exit enic_cleanup_module(void)
2712 {
2713 	pci_unregister_driver(&enic_driver);
2714 }
2715 
2716 module_init(enic_init_module);
2717 module_exit(enic_cleanup_module);
2718