xref: /openbmc/linux/drivers/net/ethernet/cisco/enic/enic_main.c (revision c51d39010a1bccc9c1294e2d7c00005aefeb2b5c)
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 #include <linux/numa.h>
43 #ifdef CONFIG_RFS_ACCEL
44 #include <linux/cpu_rmap.h>
45 #endif
46 #ifdef CONFIG_NET_RX_BUSY_POLL
47 #include <net/busy_poll.h>
48 #endif
49 #include <linux/crash_dump.h>
50 
51 #include "cq_enet_desc.h"
52 #include "vnic_dev.h"
53 #include "vnic_intr.h"
54 #include "vnic_stats.h"
55 #include "vnic_vic.h"
56 #include "enic_res.h"
57 #include "enic.h"
58 #include "enic_dev.h"
59 #include "enic_pp.h"
60 #include "enic_clsf.h"
61 
62 #define ENIC_NOTIFY_TIMER_PERIOD	(2 * HZ)
63 #define WQ_ENET_MAX_DESC_LEN		(1 << WQ_ENET_LEN_BITS)
64 #define MAX_TSO				(1 << 16)
65 #define ENIC_DESC_MAX_SPLITS		(MAX_TSO / WQ_ENET_MAX_DESC_LEN + 1)
66 
67 #define PCI_DEVICE_ID_CISCO_VIC_ENET         0x0043  /* ethernet vnic */
68 #define PCI_DEVICE_ID_CISCO_VIC_ENET_DYN     0x0044  /* enet dynamic vnic */
69 #define PCI_DEVICE_ID_CISCO_VIC_ENET_VF      0x0071  /* enet SRIOV VF */
70 
71 #define RX_COPYBREAK_DEFAULT		256
72 
73 /* Supported devices */
74 static const struct pci_device_id enic_id_table[] = {
75 	{ PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET) },
76 	{ PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET_DYN) },
77 	{ PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET_VF) },
78 	{ 0, }	/* end of table */
79 };
80 
81 MODULE_DESCRIPTION(DRV_DESCRIPTION);
82 MODULE_AUTHOR("Scott Feldman <scofeldm@cisco.com>");
83 MODULE_LICENSE("GPL");
84 MODULE_VERSION(DRV_VERSION);
85 MODULE_DEVICE_TABLE(pci, enic_id_table);
86 
87 #define ENIC_LARGE_PKT_THRESHOLD		1000
88 #define ENIC_MAX_COALESCE_TIMERS		10
89 /*  Interrupt moderation table, which will be used to decide the
90  *  coalescing timer values
91  *  {rx_rate in Mbps, mapping percentage of the range}
92  */
93 static struct enic_intr_mod_table mod_table[ENIC_MAX_COALESCE_TIMERS + 1] = {
94 	{4000,  0},
95 	{4400, 10},
96 	{5060, 20},
97 	{5230, 30},
98 	{5540, 40},
99 	{5820, 50},
100 	{6120, 60},
101 	{6435, 70},
102 	{6745, 80},
103 	{7000, 90},
104 	{0xFFFFFFFF, 100}
105 };
106 
107 /* This table helps the driver to pick different ranges for rx coalescing
108  * timer depending on the link speed.
109  */
110 static struct enic_intr_mod_range mod_range[ENIC_MAX_LINK_SPEEDS] = {
111 	{0,  0}, /* 0  - 4  Gbps */
112 	{0,  3}, /* 4  - 10 Gbps */
113 	{3,  6}, /* 10 - 40 Gbps */
114 };
115 
116 static void enic_init_affinity_hint(struct enic *enic)
117 {
118 	int numa_node = dev_to_node(&enic->pdev->dev);
119 	int i;
120 
121 	for (i = 0; i < enic->intr_count; i++) {
122 		if (enic_is_err_intr(enic, i) || enic_is_notify_intr(enic, i) ||
123 		    (enic->msix[i].affinity_mask &&
124 		     !cpumask_empty(enic->msix[i].affinity_mask)))
125 			continue;
126 		if (zalloc_cpumask_var(&enic->msix[i].affinity_mask,
127 				       GFP_KERNEL))
128 			cpumask_set_cpu(cpumask_local_spread(i, numa_node),
129 					enic->msix[i].affinity_mask);
130 	}
131 }
132 
133 static void enic_free_affinity_hint(struct enic *enic)
134 {
135 	int i;
136 
137 	for (i = 0; i < enic->intr_count; i++) {
138 		if (enic_is_err_intr(enic, i) || enic_is_notify_intr(enic, i))
139 			continue;
140 		free_cpumask_var(enic->msix[i].affinity_mask);
141 	}
142 }
143 
144 static void enic_set_affinity_hint(struct enic *enic)
145 {
146 	int i;
147 	int err;
148 
149 	for (i = 0; i < enic->intr_count; i++) {
150 		if (enic_is_err_intr(enic, i)		||
151 		    enic_is_notify_intr(enic, i)	||
152 		    !enic->msix[i].affinity_mask	||
153 		    cpumask_empty(enic->msix[i].affinity_mask))
154 			continue;
155 		err = irq_set_affinity_hint(enic->msix_entry[i].vector,
156 					    enic->msix[i].affinity_mask);
157 		if (err)
158 			netdev_warn(enic->netdev, "irq_set_affinity_hint failed, err %d\n",
159 				    err);
160 	}
161 
162 	for (i = 0; i < enic->wq_count; i++) {
163 		int wq_intr = enic_msix_wq_intr(enic, i);
164 
165 		if (enic->msix[wq_intr].affinity_mask &&
166 		    !cpumask_empty(enic->msix[wq_intr].affinity_mask))
167 			netif_set_xps_queue(enic->netdev,
168 					    enic->msix[wq_intr].affinity_mask,
169 					    i);
170 	}
171 }
172 
173 static void enic_unset_affinity_hint(struct enic *enic)
174 {
175 	int i;
176 
177 	for (i = 0; i < enic->intr_count; i++)
178 		irq_set_affinity_hint(enic->msix_entry[i].vector, NULL);
179 }
180 
181 int enic_is_dynamic(struct enic *enic)
182 {
183 	return enic->pdev->device == PCI_DEVICE_ID_CISCO_VIC_ENET_DYN;
184 }
185 
186 int enic_sriov_enabled(struct enic *enic)
187 {
188 	return (enic->priv_flags & ENIC_SRIOV_ENABLED) ? 1 : 0;
189 }
190 
191 static int enic_is_sriov_vf(struct enic *enic)
192 {
193 	return enic->pdev->device == PCI_DEVICE_ID_CISCO_VIC_ENET_VF;
194 }
195 
196 int enic_is_valid_vf(struct enic *enic, int vf)
197 {
198 #ifdef CONFIG_PCI_IOV
199 	return vf >= 0 && vf < enic->num_vfs;
200 #else
201 	return 0;
202 #endif
203 }
204 
205 static void enic_free_wq_buf(struct vnic_wq *wq, struct vnic_wq_buf *buf)
206 {
207 	struct enic *enic = vnic_dev_priv(wq->vdev);
208 
209 	if (buf->sop)
210 		pci_unmap_single(enic->pdev, buf->dma_addr,
211 			buf->len, PCI_DMA_TODEVICE);
212 	else
213 		pci_unmap_page(enic->pdev, buf->dma_addr,
214 			buf->len, PCI_DMA_TODEVICE);
215 
216 	if (buf->os_buf)
217 		dev_kfree_skb_any(buf->os_buf);
218 }
219 
220 static void enic_wq_free_buf(struct vnic_wq *wq,
221 	struct cq_desc *cq_desc, struct vnic_wq_buf *buf, void *opaque)
222 {
223 	enic_free_wq_buf(wq, buf);
224 }
225 
226 static int enic_wq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
227 	u8 type, u16 q_number, u16 completed_index, void *opaque)
228 {
229 	struct enic *enic = vnic_dev_priv(vdev);
230 
231 	spin_lock(&enic->wq_lock[q_number]);
232 
233 	vnic_wq_service(&enic->wq[q_number], cq_desc,
234 		completed_index, enic_wq_free_buf,
235 		opaque);
236 
237 	if (netif_tx_queue_stopped(netdev_get_tx_queue(enic->netdev, q_number)) &&
238 	    vnic_wq_desc_avail(&enic->wq[q_number]) >=
239 	    (MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS))
240 		netif_wake_subqueue(enic->netdev, q_number);
241 
242 	spin_unlock(&enic->wq_lock[q_number]);
243 
244 	return 0;
245 }
246 
247 static bool enic_log_q_error(struct enic *enic)
248 {
249 	unsigned int i;
250 	u32 error_status;
251 	bool err = false;
252 
253 	for (i = 0; i < enic->wq_count; i++) {
254 		error_status = vnic_wq_error_status(&enic->wq[i]);
255 		err |= error_status;
256 		if (error_status)
257 			netdev_err(enic->netdev, "WQ[%d] error_status %d\n",
258 				i, error_status);
259 	}
260 
261 	for (i = 0; i < enic->rq_count; i++) {
262 		error_status = vnic_rq_error_status(&enic->rq[i]);
263 		err |= error_status;
264 		if (error_status)
265 			netdev_err(enic->netdev, "RQ[%d] error_status %d\n",
266 				i, error_status);
267 	}
268 
269 	return err;
270 }
271 
272 static void enic_msglvl_check(struct enic *enic)
273 {
274 	u32 msg_enable = vnic_dev_msg_lvl(enic->vdev);
275 
276 	if (msg_enable != enic->msg_enable) {
277 		netdev_info(enic->netdev, "msg lvl changed from 0x%x to 0x%x\n",
278 			enic->msg_enable, msg_enable);
279 		enic->msg_enable = msg_enable;
280 	}
281 }
282 
283 static void enic_mtu_check(struct enic *enic)
284 {
285 	u32 mtu = vnic_dev_mtu(enic->vdev);
286 	struct net_device *netdev = enic->netdev;
287 
288 	if (mtu && mtu != enic->port_mtu) {
289 		enic->port_mtu = mtu;
290 		if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic)) {
291 			mtu = max_t(int, ENIC_MIN_MTU,
292 				min_t(int, ENIC_MAX_MTU, mtu));
293 			if (mtu != netdev->mtu)
294 				schedule_work(&enic->change_mtu_work);
295 		} else {
296 			if (mtu < netdev->mtu)
297 				netdev_warn(netdev,
298 					"interface MTU (%d) set higher "
299 					"than switch port MTU (%d)\n",
300 					netdev->mtu, mtu);
301 		}
302 	}
303 }
304 
305 static void enic_link_check(struct enic *enic)
306 {
307 	int link_status = vnic_dev_link_status(enic->vdev);
308 	int carrier_ok = netif_carrier_ok(enic->netdev);
309 
310 	if (link_status && !carrier_ok) {
311 		netdev_info(enic->netdev, "Link UP\n");
312 		netif_carrier_on(enic->netdev);
313 	} else if (!link_status && carrier_ok) {
314 		netdev_info(enic->netdev, "Link DOWN\n");
315 		netif_carrier_off(enic->netdev);
316 	}
317 }
318 
319 static void enic_notify_check(struct enic *enic)
320 {
321 	enic_msglvl_check(enic);
322 	enic_mtu_check(enic);
323 	enic_link_check(enic);
324 }
325 
326 #define ENIC_TEST_INTR(pba, i) (pba & (1 << i))
327 
328 static irqreturn_t enic_isr_legacy(int irq, void *data)
329 {
330 	struct net_device *netdev = data;
331 	struct enic *enic = netdev_priv(netdev);
332 	unsigned int io_intr = enic_legacy_io_intr();
333 	unsigned int err_intr = enic_legacy_err_intr();
334 	unsigned int notify_intr = enic_legacy_notify_intr();
335 	u32 pba;
336 
337 	vnic_intr_mask(&enic->intr[io_intr]);
338 
339 	pba = vnic_intr_legacy_pba(enic->legacy_pba);
340 	if (!pba) {
341 		vnic_intr_unmask(&enic->intr[io_intr]);
342 		return IRQ_NONE;	/* not our interrupt */
343 	}
344 
345 	if (ENIC_TEST_INTR(pba, notify_intr)) {
346 		enic_notify_check(enic);
347 		vnic_intr_return_all_credits(&enic->intr[notify_intr]);
348 	}
349 
350 	if (ENIC_TEST_INTR(pba, err_intr)) {
351 		vnic_intr_return_all_credits(&enic->intr[err_intr]);
352 		enic_log_q_error(enic);
353 		/* schedule recovery from WQ/RQ error */
354 		schedule_work(&enic->reset);
355 		return IRQ_HANDLED;
356 	}
357 
358 	if (ENIC_TEST_INTR(pba, io_intr))
359 		napi_schedule_irqoff(&enic->napi[0]);
360 	else
361 		vnic_intr_unmask(&enic->intr[io_intr]);
362 
363 	return IRQ_HANDLED;
364 }
365 
366 static irqreturn_t enic_isr_msi(int irq, void *data)
367 {
368 	struct enic *enic = data;
369 
370 	/* With MSI, there is no sharing of interrupts, so this is
371 	 * our interrupt and there is no need to ack it.  The device
372 	 * is not providing per-vector masking, so the OS will not
373 	 * write to PCI config space to mask/unmask the interrupt.
374 	 * We're using mask_on_assertion for MSI, so the device
375 	 * automatically masks the interrupt when the interrupt is
376 	 * generated.  Later, when exiting polling, the interrupt
377 	 * will be unmasked (see enic_poll).
378 	 *
379 	 * Also, the device uses the same PCIe Traffic Class (TC)
380 	 * for Memory Write data and MSI, so there are no ordering
381 	 * issues; the MSI will always arrive at the Root Complex
382 	 * _after_ corresponding Memory Writes (i.e. descriptor
383 	 * writes).
384 	 */
385 
386 	napi_schedule_irqoff(&enic->napi[0]);
387 
388 	return IRQ_HANDLED;
389 }
390 
391 static irqreturn_t enic_isr_msix(int irq, void *data)
392 {
393 	struct napi_struct *napi = data;
394 
395 	napi_schedule_irqoff(napi);
396 
397 	return IRQ_HANDLED;
398 }
399 
400 static irqreturn_t enic_isr_msix_err(int irq, void *data)
401 {
402 	struct enic *enic = data;
403 	unsigned int intr = enic_msix_err_intr(enic);
404 
405 	vnic_intr_return_all_credits(&enic->intr[intr]);
406 
407 	if (enic_log_q_error(enic))
408 		/* schedule recovery from WQ/RQ error */
409 		schedule_work(&enic->reset);
410 
411 	return IRQ_HANDLED;
412 }
413 
414 static irqreturn_t enic_isr_msix_notify(int irq, void *data)
415 {
416 	struct enic *enic = data;
417 	unsigned int intr = enic_msix_notify_intr(enic);
418 
419 	enic_notify_check(enic);
420 	vnic_intr_return_all_credits(&enic->intr[intr]);
421 
422 	return IRQ_HANDLED;
423 }
424 
425 static int enic_queue_wq_skb_cont(struct enic *enic, struct vnic_wq *wq,
426 				  struct sk_buff *skb, unsigned int len_left,
427 				  int loopback)
428 {
429 	const skb_frag_t *frag;
430 	dma_addr_t dma_addr;
431 
432 	/* Queue additional data fragments */
433 	for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
434 		len_left -= skb_frag_size(frag);
435 		dma_addr = skb_frag_dma_map(&enic->pdev->dev, frag, 0,
436 					    skb_frag_size(frag),
437 					    DMA_TO_DEVICE);
438 		if (unlikely(enic_dma_map_check(enic, dma_addr)))
439 			return -ENOMEM;
440 		enic_queue_wq_desc_cont(wq, skb, dma_addr, skb_frag_size(frag),
441 					(len_left == 0),	/* EOP? */
442 					loopback);
443 	}
444 
445 	return 0;
446 }
447 
448 static int enic_queue_wq_skb_vlan(struct enic *enic, struct vnic_wq *wq,
449 				  struct sk_buff *skb, int vlan_tag_insert,
450 				  unsigned int vlan_tag, int loopback)
451 {
452 	unsigned int head_len = skb_headlen(skb);
453 	unsigned int len_left = skb->len - head_len;
454 	int eop = (len_left == 0);
455 	dma_addr_t dma_addr;
456 	int err = 0;
457 
458 	dma_addr = pci_map_single(enic->pdev, skb->data, head_len,
459 				  PCI_DMA_TODEVICE);
460 	if (unlikely(enic_dma_map_check(enic, dma_addr)))
461 		return -ENOMEM;
462 
463 	/* Queue the main skb fragment. The fragments are no larger
464 	 * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less
465 	 * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor
466 	 * per fragment is queued.
467 	 */
468 	enic_queue_wq_desc(wq, skb, dma_addr, head_len,	vlan_tag_insert,
469 			   vlan_tag, eop, loopback);
470 
471 	if (!eop)
472 		err = enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback);
473 
474 	return err;
475 }
476 
477 static int enic_queue_wq_skb_csum_l4(struct enic *enic, struct vnic_wq *wq,
478 				     struct sk_buff *skb, int vlan_tag_insert,
479 				     unsigned int vlan_tag, int loopback)
480 {
481 	unsigned int head_len = skb_headlen(skb);
482 	unsigned int len_left = skb->len - head_len;
483 	unsigned int hdr_len = skb_checksum_start_offset(skb);
484 	unsigned int csum_offset = hdr_len + skb->csum_offset;
485 	int eop = (len_left == 0);
486 	dma_addr_t dma_addr;
487 	int err = 0;
488 
489 	dma_addr = pci_map_single(enic->pdev, skb->data, head_len,
490 				  PCI_DMA_TODEVICE);
491 	if (unlikely(enic_dma_map_check(enic, dma_addr)))
492 		return -ENOMEM;
493 
494 	/* Queue the main skb fragment. The fragments are no larger
495 	 * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less
496 	 * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor
497 	 * per fragment is queued.
498 	 */
499 	enic_queue_wq_desc_csum_l4(wq, skb, dma_addr, head_len,	csum_offset,
500 				   hdr_len, vlan_tag_insert, vlan_tag, eop,
501 				   loopback);
502 
503 	if (!eop)
504 		err = enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback);
505 
506 	return err;
507 }
508 
509 static int enic_queue_wq_skb_tso(struct enic *enic, struct vnic_wq *wq,
510 				 struct sk_buff *skb, unsigned int mss,
511 				 int vlan_tag_insert, unsigned int vlan_tag,
512 				 int loopback)
513 {
514 	unsigned int frag_len_left = skb_headlen(skb);
515 	unsigned int len_left = skb->len - frag_len_left;
516 	unsigned int hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
517 	int eop = (len_left == 0);
518 	unsigned int len;
519 	dma_addr_t dma_addr;
520 	unsigned int offset = 0;
521 	skb_frag_t *frag;
522 
523 	/* Preload TCP csum field with IP pseudo hdr calculated
524 	 * with IP length set to zero.  HW will later add in length
525 	 * to each TCP segment resulting from the TSO.
526 	 */
527 
528 	if (skb->protocol == cpu_to_be16(ETH_P_IP)) {
529 		ip_hdr(skb)->check = 0;
530 		tcp_hdr(skb)->check = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
531 			ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
532 	} else if (skb->protocol == cpu_to_be16(ETH_P_IPV6)) {
533 		tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
534 			&ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
535 	}
536 
537 	/* Queue WQ_ENET_MAX_DESC_LEN length descriptors
538 	 * for the main skb fragment
539 	 */
540 	while (frag_len_left) {
541 		len = min(frag_len_left, (unsigned int)WQ_ENET_MAX_DESC_LEN);
542 		dma_addr = pci_map_single(enic->pdev, skb->data + offset, len,
543 					  PCI_DMA_TODEVICE);
544 		if (unlikely(enic_dma_map_check(enic, dma_addr)))
545 			return -ENOMEM;
546 		enic_queue_wq_desc_tso(wq, skb, dma_addr, len, mss, hdr_len,
547 				       vlan_tag_insert, vlan_tag,
548 				       eop && (len == frag_len_left), loopback);
549 		frag_len_left -= len;
550 		offset += len;
551 	}
552 
553 	if (eop)
554 		return 0;
555 
556 	/* Queue WQ_ENET_MAX_DESC_LEN length descriptors
557 	 * for additional data fragments
558 	 */
559 	for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
560 		len_left -= skb_frag_size(frag);
561 		frag_len_left = skb_frag_size(frag);
562 		offset = 0;
563 
564 		while (frag_len_left) {
565 			len = min(frag_len_left,
566 				(unsigned int)WQ_ENET_MAX_DESC_LEN);
567 			dma_addr = skb_frag_dma_map(&enic->pdev->dev, frag,
568 						    offset, len,
569 						    DMA_TO_DEVICE);
570 			if (unlikely(enic_dma_map_check(enic, dma_addr)))
571 				return -ENOMEM;
572 			enic_queue_wq_desc_cont(wq, skb, dma_addr, len,
573 						(len_left == 0) &&
574 						 (len == frag_len_left),/*EOP*/
575 						loopback);
576 			frag_len_left -= len;
577 			offset += len;
578 		}
579 	}
580 
581 	return 0;
582 }
583 
584 static inline void enic_queue_wq_skb(struct enic *enic,
585 	struct vnic_wq *wq, struct sk_buff *skb)
586 {
587 	unsigned int mss = skb_shinfo(skb)->gso_size;
588 	unsigned int vlan_tag = 0;
589 	int vlan_tag_insert = 0;
590 	int loopback = 0;
591 	int err;
592 
593 	if (skb_vlan_tag_present(skb)) {
594 		/* VLAN tag from trunking driver */
595 		vlan_tag_insert = 1;
596 		vlan_tag = skb_vlan_tag_get(skb);
597 	} else if (enic->loop_enable) {
598 		vlan_tag = enic->loop_tag;
599 		loopback = 1;
600 	}
601 
602 	if (mss)
603 		err = enic_queue_wq_skb_tso(enic, wq, skb, mss,
604 					    vlan_tag_insert, vlan_tag,
605 					    loopback);
606 	else if	(skb->ip_summed == CHECKSUM_PARTIAL)
607 		err = enic_queue_wq_skb_csum_l4(enic, wq, skb, vlan_tag_insert,
608 						vlan_tag, loopback);
609 	else
610 		err = enic_queue_wq_skb_vlan(enic, wq, skb, vlan_tag_insert,
611 					     vlan_tag, loopback);
612 	if (unlikely(err)) {
613 		struct vnic_wq_buf *buf;
614 
615 		buf = wq->to_use->prev;
616 		/* while not EOP of previous pkt && queue not empty.
617 		 * For all non EOP bufs, os_buf is NULL.
618 		 */
619 		while (!buf->os_buf && (buf->next != wq->to_clean)) {
620 			enic_free_wq_buf(wq, buf);
621 			wq->ring.desc_avail++;
622 			buf = buf->prev;
623 		}
624 		wq->to_use = buf->next;
625 		dev_kfree_skb(skb);
626 	}
627 }
628 
629 /* netif_tx_lock held, process context with BHs disabled, or BH */
630 static netdev_tx_t enic_hard_start_xmit(struct sk_buff *skb,
631 	struct net_device *netdev)
632 {
633 	struct enic *enic = netdev_priv(netdev);
634 	struct vnic_wq *wq;
635 	unsigned int txq_map;
636 	struct netdev_queue *txq;
637 
638 	if (skb->len <= 0) {
639 		dev_kfree_skb_any(skb);
640 		return NETDEV_TX_OK;
641 	}
642 
643 	txq_map = skb_get_queue_mapping(skb) % enic->wq_count;
644 	wq = &enic->wq[txq_map];
645 	txq = netdev_get_tx_queue(netdev, txq_map);
646 
647 	/* Non-TSO sends must fit within ENIC_NON_TSO_MAX_DESC descs,
648 	 * which is very likely.  In the off chance it's going to take
649 	 * more than * ENIC_NON_TSO_MAX_DESC, linearize the skb.
650 	 */
651 
652 	if (skb_shinfo(skb)->gso_size == 0 &&
653 	    skb_shinfo(skb)->nr_frags + 1 > ENIC_NON_TSO_MAX_DESC &&
654 	    skb_linearize(skb)) {
655 		dev_kfree_skb_any(skb);
656 		return NETDEV_TX_OK;
657 	}
658 
659 	spin_lock(&enic->wq_lock[txq_map]);
660 
661 	if (vnic_wq_desc_avail(wq) <
662 	    skb_shinfo(skb)->nr_frags + ENIC_DESC_MAX_SPLITS) {
663 		netif_tx_stop_queue(txq);
664 		/* This is a hard error, log it */
665 		netdev_err(netdev, "BUG! Tx ring full when queue awake!\n");
666 		spin_unlock(&enic->wq_lock[txq_map]);
667 		return NETDEV_TX_BUSY;
668 	}
669 
670 	enic_queue_wq_skb(enic, wq, skb);
671 
672 	if (vnic_wq_desc_avail(wq) < MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS)
673 		netif_tx_stop_queue(txq);
674 	if (!skb->xmit_more || netif_xmit_stopped(txq))
675 		vnic_wq_doorbell(wq);
676 
677 	spin_unlock(&enic->wq_lock[txq_map]);
678 
679 	return NETDEV_TX_OK;
680 }
681 
682 /* dev_base_lock rwlock held, nominally process context */
683 static struct rtnl_link_stats64 *enic_get_stats(struct net_device *netdev,
684 						struct rtnl_link_stats64 *net_stats)
685 {
686 	struct enic *enic = netdev_priv(netdev);
687 	struct vnic_stats *stats;
688 	int err;
689 
690 	err = enic_dev_stats_dump(enic, &stats);
691 	/* return only when pci_zalloc_consistent fails in vnic_dev_stats_dump
692 	 * For other failures, like devcmd failure, we return previously
693 	 * recorded stats.
694 	 */
695 	if (err == -ENOMEM)
696 		return net_stats;
697 
698 	net_stats->tx_packets = stats->tx.tx_frames_ok;
699 	net_stats->tx_bytes = stats->tx.tx_bytes_ok;
700 	net_stats->tx_errors = stats->tx.tx_errors;
701 	net_stats->tx_dropped = stats->tx.tx_drops;
702 
703 	net_stats->rx_packets = stats->rx.rx_frames_ok;
704 	net_stats->rx_bytes = stats->rx.rx_bytes_ok;
705 	net_stats->rx_errors = stats->rx.rx_errors;
706 	net_stats->multicast = stats->rx.rx_multicast_frames_ok;
707 	net_stats->rx_over_errors = enic->rq_truncated_pkts;
708 	net_stats->rx_crc_errors = enic->rq_bad_fcs;
709 	net_stats->rx_dropped = stats->rx.rx_no_bufs + stats->rx.rx_drop;
710 
711 	return net_stats;
712 }
713 
714 static int enic_mc_sync(struct net_device *netdev, const u8 *mc_addr)
715 {
716 	struct enic *enic = netdev_priv(netdev);
717 
718 	if (enic->mc_count == ENIC_MULTICAST_PERFECT_FILTERS) {
719 		unsigned int mc_count = netdev_mc_count(netdev);
720 
721 		netdev_warn(netdev, "Registering only %d out of %d multicast addresses\n",
722 			    ENIC_MULTICAST_PERFECT_FILTERS, mc_count);
723 
724 		return -ENOSPC;
725 	}
726 
727 	enic_dev_add_addr(enic, mc_addr);
728 	enic->mc_count++;
729 
730 	return 0;
731 }
732 
733 static int enic_mc_unsync(struct net_device *netdev, const u8 *mc_addr)
734 {
735 	struct enic *enic = netdev_priv(netdev);
736 
737 	enic_dev_del_addr(enic, mc_addr);
738 	enic->mc_count--;
739 
740 	return 0;
741 }
742 
743 static int enic_uc_sync(struct net_device *netdev, const u8 *uc_addr)
744 {
745 	struct enic *enic = netdev_priv(netdev);
746 
747 	if (enic->uc_count == ENIC_UNICAST_PERFECT_FILTERS) {
748 		unsigned int uc_count = netdev_uc_count(netdev);
749 
750 		netdev_warn(netdev, "Registering only %d out of %d unicast addresses\n",
751 			    ENIC_UNICAST_PERFECT_FILTERS, uc_count);
752 
753 		return -ENOSPC;
754 	}
755 
756 	enic_dev_add_addr(enic, uc_addr);
757 	enic->uc_count++;
758 
759 	return 0;
760 }
761 
762 static int enic_uc_unsync(struct net_device *netdev, const u8 *uc_addr)
763 {
764 	struct enic *enic = netdev_priv(netdev);
765 
766 	enic_dev_del_addr(enic, uc_addr);
767 	enic->uc_count--;
768 
769 	return 0;
770 }
771 
772 void enic_reset_addr_lists(struct enic *enic)
773 {
774 	struct net_device *netdev = enic->netdev;
775 
776 	__dev_uc_unsync(netdev, NULL);
777 	__dev_mc_unsync(netdev, NULL);
778 
779 	enic->mc_count = 0;
780 	enic->uc_count = 0;
781 	enic->flags = 0;
782 }
783 
784 static int enic_set_mac_addr(struct net_device *netdev, char *addr)
785 {
786 	struct enic *enic = netdev_priv(netdev);
787 
788 	if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic)) {
789 		if (!is_valid_ether_addr(addr) && !is_zero_ether_addr(addr))
790 			return -EADDRNOTAVAIL;
791 	} else {
792 		if (!is_valid_ether_addr(addr))
793 			return -EADDRNOTAVAIL;
794 	}
795 
796 	memcpy(netdev->dev_addr, addr, netdev->addr_len);
797 
798 	return 0;
799 }
800 
801 static int enic_set_mac_address_dynamic(struct net_device *netdev, void *p)
802 {
803 	struct enic *enic = netdev_priv(netdev);
804 	struct sockaddr *saddr = p;
805 	char *addr = saddr->sa_data;
806 	int err;
807 
808 	if (netif_running(enic->netdev)) {
809 		err = enic_dev_del_station_addr(enic);
810 		if (err)
811 			return err;
812 	}
813 
814 	err = enic_set_mac_addr(netdev, addr);
815 	if (err)
816 		return err;
817 
818 	if (netif_running(enic->netdev)) {
819 		err = enic_dev_add_station_addr(enic);
820 		if (err)
821 			return err;
822 	}
823 
824 	return err;
825 }
826 
827 static int enic_set_mac_address(struct net_device *netdev, void *p)
828 {
829 	struct sockaddr *saddr = p;
830 	char *addr = saddr->sa_data;
831 	struct enic *enic = netdev_priv(netdev);
832 	int err;
833 
834 	err = enic_dev_del_station_addr(enic);
835 	if (err)
836 		return err;
837 
838 	err = enic_set_mac_addr(netdev, addr);
839 	if (err)
840 		return err;
841 
842 	return enic_dev_add_station_addr(enic);
843 }
844 
845 /* netif_tx_lock held, BHs disabled */
846 static void enic_set_rx_mode(struct net_device *netdev)
847 {
848 	struct enic *enic = netdev_priv(netdev);
849 	int directed = 1;
850 	int multicast = (netdev->flags & IFF_MULTICAST) ? 1 : 0;
851 	int broadcast = (netdev->flags & IFF_BROADCAST) ? 1 : 0;
852 	int promisc = (netdev->flags & IFF_PROMISC) ||
853 		netdev_uc_count(netdev) > ENIC_UNICAST_PERFECT_FILTERS;
854 	int allmulti = (netdev->flags & IFF_ALLMULTI) ||
855 		netdev_mc_count(netdev) > ENIC_MULTICAST_PERFECT_FILTERS;
856 	unsigned int flags = netdev->flags |
857 		(allmulti ? IFF_ALLMULTI : 0) |
858 		(promisc ? IFF_PROMISC : 0);
859 
860 	if (enic->flags != flags) {
861 		enic->flags = flags;
862 		enic_dev_packet_filter(enic, directed,
863 			multicast, broadcast, promisc, allmulti);
864 	}
865 
866 	if (!promisc) {
867 		__dev_uc_sync(netdev, enic_uc_sync, enic_uc_unsync);
868 		if (!allmulti)
869 			__dev_mc_sync(netdev, enic_mc_sync, enic_mc_unsync);
870 	}
871 }
872 
873 /* netif_tx_lock held, BHs disabled */
874 static void enic_tx_timeout(struct net_device *netdev)
875 {
876 	struct enic *enic = netdev_priv(netdev);
877 	schedule_work(&enic->tx_hang_reset);
878 }
879 
880 static int enic_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
881 {
882 	struct enic *enic = netdev_priv(netdev);
883 	struct enic_port_profile *pp;
884 	int err;
885 
886 	ENIC_PP_BY_INDEX(enic, vf, pp, &err);
887 	if (err)
888 		return err;
889 
890 	if (is_valid_ether_addr(mac) || is_zero_ether_addr(mac)) {
891 		if (vf == PORT_SELF_VF) {
892 			memcpy(pp->vf_mac, mac, ETH_ALEN);
893 			return 0;
894 		} else {
895 			/*
896 			 * For sriov vf's set the mac in hw
897 			 */
898 			ENIC_DEVCMD_PROXY_BY_INDEX(vf, err, enic,
899 				vnic_dev_set_mac_addr, mac);
900 			return enic_dev_status_to_errno(err);
901 		}
902 	} else
903 		return -EINVAL;
904 }
905 
906 static int enic_set_vf_port(struct net_device *netdev, int vf,
907 	struct nlattr *port[])
908 {
909 	struct enic *enic = netdev_priv(netdev);
910 	struct enic_port_profile prev_pp;
911 	struct enic_port_profile *pp;
912 	int err = 0, restore_pp = 1;
913 
914 	ENIC_PP_BY_INDEX(enic, vf, pp, &err);
915 	if (err)
916 		return err;
917 
918 	if (!port[IFLA_PORT_REQUEST])
919 		return -EOPNOTSUPP;
920 
921 	memcpy(&prev_pp, pp, sizeof(*enic->pp));
922 	memset(pp, 0, sizeof(*enic->pp));
923 
924 	pp->set |= ENIC_SET_REQUEST;
925 	pp->request = nla_get_u8(port[IFLA_PORT_REQUEST]);
926 
927 	if (port[IFLA_PORT_PROFILE]) {
928 		pp->set |= ENIC_SET_NAME;
929 		memcpy(pp->name, nla_data(port[IFLA_PORT_PROFILE]),
930 			PORT_PROFILE_MAX);
931 	}
932 
933 	if (port[IFLA_PORT_INSTANCE_UUID]) {
934 		pp->set |= ENIC_SET_INSTANCE;
935 		memcpy(pp->instance_uuid,
936 			nla_data(port[IFLA_PORT_INSTANCE_UUID]), PORT_UUID_MAX);
937 	}
938 
939 	if (port[IFLA_PORT_HOST_UUID]) {
940 		pp->set |= ENIC_SET_HOST;
941 		memcpy(pp->host_uuid,
942 			nla_data(port[IFLA_PORT_HOST_UUID]), PORT_UUID_MAX);
943 	}
944 
945 	if (vf == PORT_SELF_VF) {
946 		/* Special case handling: mac came from IFLA_VF_MAC */
947 		if (!is_zero_ether_addr(prev_pp.vf_mac))
948 			memcpy(pp->mac_addr, prev_pp.vf_mac, ETH_ALEN);
949 
950 		if (is_zero_ether_addr(netdev->dev_addr))
951 			eth_hw_addr_random(netdev);
952 	} else {
953 		/* SR-IOV VF: get mac from adapter */
954 		ENIC_DEVCMD_PROXY_BY_INDEX(vf, err, enic,
955 			vnic_dev_get_mac_addr, pp->mac_addr);
956 		if (err) {
957 			netdev_err(netdev, "Error getting mac for vf %d\n", vf);
958 			memcpy(pp, &prev_pp, sizeof(*pp));
959 			return enic_dev_status_to_errno(err);
960 		}
961 	}
962 
963 	err = enic_process_set_pp_request(enic, vf, &prev_pp, &restore_pp);
964 	if (err) {
965 		if (restore_pp) {
966 			/* Things are still the way they were: Implicit
967 			 * DISASSOCIATE failed
968 			 */
969 			memcpy(pp, &prev_pp, sizeof(*pp));
970 		} else {
971 			memset(pp, 0, sizeof(*pp));
972 			if (vf == PORT_SELF_VF)
973 				eth_zero_addr(netdev->dev_addr);
974 		}
975 	} else {
976 		/* Set flag to indicate that the port assoc/disassoc
977 		 * request has been sent out to fw
978 		 */
979 		pp->set |= ENIC_PORT_REQUEST_APPLIED;
980 
981 		/* If DISASSOCIATE, clean up all assigned/saved macaddresses */
982 		if (pp->request == PORT_REQUEST_DISASSOCIATE) {
983 			eth_zero_addr(pp->mac_addr);
984 			if (vf == PORT_SELF_VF)
985 				eth_zero_addr(netdev->dev_addr);
986 		}
987 	}
988 
989 	if (vf == PORT_SELF_VF)
990 		eth_zero_addr(pp->vf_mac);
991 
992 	return err;
993 }
994 
995 static int enic_get_vf_port(struct net_device *netdev, int vf,
996 	struct sk_buff *skb)
997 {
998 	struct enic *enic = netdev_priv(netdev);
999 	u16 response = PORT_PROFILE_RESPONSE_SUCCESS;
1000 	struct enic_port_profile *pp;
1001 	int err;
1002 
1003 	ENIC_PP_BY_INDEX(enic, vf, pp, &err);
1004 	if (err)
1005 		return err;
1006 
1007 	if (!(pp->set & ENIC_PORT_REQUEST_APPLIED))
1008 		return -ENODATA;
1009 
1010 	err = enic_process_get_pp_request(enic, vf, pp->request, &response);
1011 	if (err)
1012 		return err;
1013 
1014 	if (nla_put_u16(skb, IFLA_PORT_REQUEST, pp->request) ||
1015 	    nla_put_u16(skb, IFLA_PORT_RESPONSE, response) ||
1016 	    ((pp->set & ENIC_SET_NAME) &&
1017 	     nla_put(skb, IFLA_PORT_PROFILE, PORT_PROFILE_MAX, pp->name)) ||
1018 	    ((pp->set & ENIC_SET_INSTANCE) &&
1019 	     nla_put(skb, IFLA_PORT_INSTANCE_UUID, PORT_UUID_MAX,
1020 		     pp->instance_uuid)) ||
1021 	    ((pp->set & ENIC_SET_HOST) &&
1022 	     nla_put(skb, IFLA_PORT_HOST_UUID, PORT_UUID_MAX, pp->host_uuid)))
1023 		goto nla_put_failure;
1024 	return 0;
1025 
1026 nla_put_failure:
1027 	return -EMSGSIZE;
1028 }
1029 
1030 static void enic_free_rq_buf(struct vnic_rq *rq, struct vnic_rq_buf *buf)
1031 {
1032 	struct enic *enic = vnic_dev_priv(rq->vdev);
1033 
1034 	if (!buf->os_buf)
1035 		return;
1036 
1037 	pci_unmap_single(enic->pdev, buf->dma_addr,
1038 		buf->len, PCI_DMA_FROMDEVICE);
1039 	dev_kfree_skb_any(buf->os_buf);
1040 	buf->os_buf = NULL;
1041 }
1042 
1043 static int enic_rq_alloc_buf(struct vnic_rq *rq)
1044 {
1045 	struct enic *enic = vnic_dev_priv(rq->vdev);
1046 	struct net_device *netdev = enic->netdev;
1047 	struct sk_buff *skb;
1048 	unsigned int len = netdev->mtu + VLAN_ETH_HLEN;
1049 	unsigned int os_buf_index = 0;
1050 	dma_addr_t dma_addr;
1051 	struct vnic_rq_buf *buf = rq->to_use;
1052 
1053 	if (buf->os_buf) {
1054 		enic_queue_rq_desc(rq, buf->os_buf, os_buf_index, buf->dma_addr,
1055 				   buf->len);
1056 
1057 		return 0;
1058 	}
1059 	skb = netdev_alloc_skb_ip_align(netdev, len);
1060 	if (!skb)
1061 		return -ENOMEM;
1062 
1063 	dma_addr = pci_map_single(enic->pdev, skb->data, len,
1064 				  PCI_DMA_FROMDEVICE);
1065 	if (unlikely(enic_dma_map_check(enic, dma_addr))) {
1066 		dev_kfree_skb(skb);
1067 		return -ENOMEM;
1068 	}
1069 
1070 	enic_queue_rq_desc(rq, skb, os_buf_index,
1071 		dma_addr, len);
1072 
1073 	return 0;
1074 }
1075 
1076 static void enic_intr_update_pkt_size(struct vnic_rx_bytes_counter *pkt_size,
1077 				      u32 pkt_len)
1078 {
1079 	if (ENIC_LARGE_PKT_THRESHOLD <= pkt_len)
1080 		pkt_size->large_pkt_bytes_cnt += pkt_len;
1081 	else
1082 		pkt_size->small_pkt_bytes_cnt += pkt_len;
1083 }
1084 
1085 static bool enic_rxcopybreak(struct net_device *netdev, struct sk_buff **skb,
1086 			     struct vnic_rq_buf *buf, u16 len)
1087 {
1088 	struct enic *enic = netdev_priv(netdev);
1089 	struct sk_buff *new_skb;
1090 
1091 	if (len > enic->rx_copybreak)
1092 		return false;
1093 	new_skb = netdev_alloc_skb_ip_align(netdev, len);
1094 	if (!new_skb)
1095 		return false;
1096 	pci_dma_sync_single_for_cpu(enic->pdev, buf->dma_addr, len,
1097 				    DMA_FROM_DEVICE);
1098 	memcpy(new_skb->data, (*skb)->data, len);
1099 	*skb = new_skb;
1100 
1101 	return true;
1102 }
1103 
1104 static void enic_rq_indicate_buf(struct vnic_rq *rq,
1105 	struct cq_desc *cq_desc, struct vnic_rq_buf *buf,
1106 	int skipped, void *opaque)
1107 {
1108 	struct enic *enic = vnic_dev_priv(rq->vdev);
1109 	struct net_device *netdev = enic->netdev;
1110 	struct sk_buff *skb;
1111 	struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
1112 
1113 	u8 type, color, eop, sop, ingress_port, vlan_stripped;
1114 	u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof;
1115 	u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok;
1116 	u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc;
1117 	u8 packet_error;
1118 	u16 q_number, completed_index, bytes_written, vlan_tci, checksum;
1119 	u32 rss_hash;
1120 
1121 	if (skipped)
1122 		return;
1123 
1124 	skb = buf->os_buf;
1125 
1126 	cq_enet_rq_desc_dec((struct cq_enet_rq_desc *)cq_desc,
1127 		&type, &color, &q_number, &completed_index,
1128 		&ingress_port, &fcoe, &eop, &sop, &rss_type,
1129 		&csum_not_calc, &rss_hash, &bytes_written,
1130 		&packet_error, &vlan_stripped, &vlan_tci, &checksum,
1131 		&fcoe_sof, &fcoe_fc_crc_ok, &fcoe_enc_error,
1132 		&fcoe_eof, &tcp_udp_csum_ok, &udp, &tcp,
1133 		&ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment,
1134 		&fcs_ok);
1135 
1136 	if (packet_error) {
1137 
1138 		if (!fcs_ok) {
1139 			if (bytes_written > 0)
1140 				enic->rq_bad_fcs++;
1141 			else if (bytes_written == 0)
1142 				enic->rq_truncated_pkts++;
1143 		}
1144 
1145 		pci_unmap_single(enic->pdev, buf->dma_addr, buf->len,
1146 				 PCI_DMA_FROMDEVICE);
1147 		dev_kfree_skb_any(skb);
1148 		buf->os_buf = NULL;
1149 
1150 		return;
1151 	}
1152 
1153 	if (eop && bytes_written > 0) {
1154 
1155 		/* Good receive
1156 		 */
1157 
1158 		if (!enic_rxcopybreak(netdev, &skb, buf, bytes_written)) {
1159 			buf->os_buf = NULL;
1160 			pci_unmap_single(enic->pdev, buf->dma_addr, buf->len,
1161 					 PCI_DMA_FROMDEVICE);
1162 		}
1163 		prefetch(skb->data - NET_IP_ALIGN);
1164 
1165 		skb_put(skb, bytes_written);
1166 		skb->protocol = eth_type_trans(skb, netdev);
1167 		skb_record_rx_queue(skb, q_number);
1168 		if (netdev->features & NETIF_F_RXHASH) {
1169 			switch (rss_type) {
1170 			case CQ_ENET_RQ_DESC_RSS_TYPE_TCP_IPv4:
1171 			case CQ_ENET_RQ_DESC_RSS_TYPE_TCP_IPv6:
1172 			case CQ_ENET_RQ_DESC_RSS_TYPE_TCP_IPv6_EX:
1173 				skb_set_hash(skb, rss_hash, PKT_HASH_TYPE_L4);
1174 				break;
1175 			case CQ_ENET_RQ_DESC_RSS_TYPE_IPv4:
1176 			case CQ_ENET_RQ_DESC_RSS_TYPE_IPv6:
1177 			case CQ_ENET_RQ_DESC_RSS_TYPE_IPv6_EX:
1178 				skb_set_hash(skb, rss_hash, PKT_HASH_TYPE_L3);
1179 				break;
1180 			}
1181 		}
1182 
1183 		/* Hardware does not provide whole packet checksum. It only
1184 		 * provides pseudo checksum. Since hw validates the packet
1185 		 * checksum but not provide us the checksum value. use
1186 		 * CHECSUM_UNNECESSARY.
1187 		 */
1188 		if ((netdev->features & NETIF_F_RXCSUM) && tcp_udp_csum_ok &&
1189 		    ipv4_csum_ok)
1190 			skb->ip_summed = CHECKSUM_UNNECESSARY;
1191 
1192 		if (vlan_stripped)
1193 			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tci);
1194 
1195 		skb_mark_napi_id(skb, &enic->napi[rq->index]);
1196 		if (enic_poll_busy_polling(rq) ||
1197 		    !(netdev->features & NETIF_F_GRO))
1198 			netif_receive_skb(skb);
1199 		else
1200 			napi_gro_receive(&enic->napi[q_number], skb);
1201 		if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1202 			enic_intr_update_pkt_size(&cq->pkt_size_counter,
1203 						  bytes_written);
1204 	} else {
1205 
1206 		/* Buffer overflow
1207 		 */
1208 
1209 		pci_unmap_single(enic->pdev, buf->dma_addr, buf->len,
1210 				 PCI_DMA_FROMDEVICE);
1211 		dev_kfree_skb_any(skb);
1212 		buf->os_buf = NULL;
1213 	}
1214 }
1215 
1216 static int enic_rq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
1217 	u8 type, u16 q_number, u16 completed_index, void *opaque)
1218 {
1219 	struct enic *enic = vnic_dev_priv(vdev);
1220 
1221 	vnic_rq_service(&enic->rq[q_number], cq_desc,
1222 		completed_index, VNIC_RQ_RETURN_DESC,
1223 		enic_rq_indicate_buf, opaque);
1224 
1225 	return 0;
1226 }
1227 
1228 static void enic_set_int_moderation(struct enic *enic, struct vnic_rq *rq)
1229 {
1230 	unsigned int intr = enic_msix_rq_intr(enic, rq->index);
1231 	struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
1232 	u32 timer = cq->tobe_rx_coal_timeval;
1233 
1234 	if (cq->tobe_rx_coal_timeval != cq->cur_rx_coal_timeval) {
1235 		vnic_intr_coalescing_timer_set(&enic->intr[intr], timer);
1236 		cq->cur_rx_coal_timeval = cq->tobe_rx_coal_timeval;
1237 	}
1238 }
1239 
1240 static void enic_calc_int_moderation(struct enic *enic, struct vnic_rq *rq)
1241 {
1242 	struct enic_rx_coal *rx_coal = &enic->rx_coalesce_setting;
1243 	struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
1244 	struct vnic_rx_bytes_counter *pkt_size_counter = &cq->pkt_size_counter;
1245 	int index;
1246 	u32 timer;
1247 	u32 range_start;
1248 	u32 traffic;
1249 	u64 delta;
1250 	ktime_t now = ktime_get();
1251 
1252 	delta = ktime_us_delta(now, cq->prev_ts);
1253 	if (delta < ENIC_AIC_TS_BREAK)
1254 		return;
1255 	cq->prev_ts = now;
1256 
1257 	traffic = pkt_size_counter->large_pkt_bytes_cnt +
1258 		  pkt_size_counter->small_pkt_bytes_cnt;
1259 	/* The table takes Mbps
1260 	 * traffic *= 8    => bits
1261 	 * traffic *= (10^6 / delta)    => bps
1262 	 * traffic /= 10^6     => Mbps
1263 	 *
1264 	 * Combining, traffic *= (8 / delta)
1265 	 */
1266 
1267 	traffic <<= 3;
1268 	traffic = delta > UINT_MAX ? 0 : traffic / (u32)delta;
1269 
1270 	for (index = 0; index < ENIC_MAX_COALESCE_TIMERS; index++)
1271 		if (traffic < mod_table[index].rx_rate)
1272 			break;
1273 	range_start = (pkt_size_counter->small_pkt_bytes_cnt >
1274 		       pkt_size_counter->large_pkt_bytes_cnt << 1) ?
1275 		      rx_coal->small_pkt_range_start :
1276 		      rx_coal->large_pkt_range_start;
1277 	timer = range_start + ((rx_coal->range_end - range_start) *
1278 			       mod_table[index].range_percent / 100);
1279 	/* Damping */
1280 	cq->tobe_rx_coal_timeval = (timer + cq->tobe_rx_coal_timeval) >> 1;
1281 
1282 	pkt_size_counter->large_pkt_bytes_cnt = 0;
1283 	pkt_size_counter->small_pkt_bytes_cnt = 0;
1284 }
1285 
1286 static int enic_poll(struct napi_struct *napi, int budget)
1287 {
1288 	struct net_device *netdev = napi->dev;
1289 	struct enic *enic = netdev_priv(netdev);
1290 	unsigned int cq_rq = enic_cq_rq(enic, 0);
1291 	unsigned int cq_wq = enic_cq_wq(enic, 0);
1292 	unsigned int intr = enic_legacy_io_intr();
1293 	unsigned int rq_work_to_do = budget;
1294 	unsigned int wq_work_to_do = -1; /* no limit */
1295 	unsigned int  work_done, rq_work_done = 0, wq_work_done;
1296 	int err;
1297 
1298 	wq_work_done = vnic_cq_service(&enic->cq[cq_wq], wq_work_to_do,
1299 				       enic_wq_service, NULL);
1300 
1301 	if (!enic_poll_lock_napi(&enic->rq[cq_rq])) {
1302 		if (wq_work_done > 0)
1303 			vnic_intr_return_credits(&enic->intr[intr],
1304 						 wq_work_done,
1305 						 0 /* dont unmask intr */,
1306 						 0 /* dont reset intr timer */);
1307 		return budget;
1308 	}
1309 
1310 	if (budget > 0)
1311 		rq_work_done = vnic_cq_service(&enic->cq[cq_rq],
1312 			rq_work_to_do, enic_rq_service, NULL);
1313 
1314 	/* Accumulate intr event credits for this polling
1315 	 * cycle.  An intr event is the completion of a
1316 	 * a WQ or RQ packet.
1317 	 */
1318 
1319 	work_done = rq_work_done + wq_work_done;
1320 
1321 	if (work_done > 0)
1322 		vnic_intr_return_credits(&enic->intr[intr],
1323 			work_done,
1324 			0 /* don't unmask intr */,
1325 			0 /* don't reset intr timer */);
1326 
1327 	err = vnic_rq_fill(&enic->rq[0], enic_rq_alloc_buf);
1328 	enic_poll_unlock_napi(&enic->rq[cq_rq], napi);
1329 
1330 	/* Buffer allocation failed. Stay in polling
1331 	 * mode so we can try to fill the ring again.
1332 	 */
1333 
1334 	if (err)
1335 		rq_work_done = rq_work_to_do;
1336 	if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1337 		/* Call the function which refreshes the intr coalescing timer
1338 		 * value based on the traffic.
1339 		 */
1340 		enic_calc_int_moderation(enic, &enic->rq[0]);
1341 
1342 	if (rq_work_done < rq_work_to_do) {
1343 
1344 		/* Some work done, but not enough to stay in polling,
1345 		 * exit polling
1346 		 */
1347 
1348 		napi_complete(napi);
1349 		if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1350 			enic_set_int_moderation(enic, &enic->rq[0]);
1351 		vnic_intr_unmask(&enic->intr[intr]);
1352 	}
1353 
1354 	return rq_work_done;
1355 }
1356 
1357 #ifdef CONFIG_RFS_ACCEL
1358 static void enic_free_rx_cpu_rmap(struct enic *enic)
1359 {
1360 	free_irq_cpu_rmap(enic->netdev->rx_cpu_rmap);
1361 	enic->netdev->rx_cpu_rmap = NULL;
1362 }
1363 
1364 static void enic_set_rx_cpu_rmap(struct enic *enic)
1365 {
1366 	int i, res;
1367 
1368 	if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX) {
1369 		enic->netdev->rx_cpu_rmap = alloc_irq_cpu_rmap(enic->rq_count);
1370 		if (unlikely(!enic->netdev->rx_cpu_rmap))
1371 			return;
1372 		for (i = 0; i < enic->rq_count; i++) {
1373 			res = irq_cpu_rmap_add(enic->netdev->rx_cpu_rmap,
1374 					       enic->msix_entry[i].vector);
1375 			if (unlikely(res)) {
1376 				enic_free_rx_cpu_rmap(enic);
1377 				return;
1378 			}
1379 		}
1380 	}
1381 }
1382 
1383 #else
1384 
1385 static void enic_free_rx_cpu_rmap(struct enic *enic)
1386 {
1387 }
1388 
1389 static void enic_set_rx_cpu_rmap(struct enic *enic)
1390 {
1391 }
1392 
1393 #endif /* CONFIG_RFS_ACCEL */
1394 
1395 #ifdef CONFIG_NET_RX_BUSY_POLL
1396 static int enic_busy_poll(struct napi_struct *napi)
1397 {
1398 	struct net_device *netdev = napi->dev;
1399 	struct enic *enic = netdev_priv(netdev);
1400 	unsigned int rq = (napi - &enic->napi[0]);
1401 	unsigned int cq = enic_cq_rq(enic, rq);
1402 	unsigned int intr = enic_msix_rq_intr(enic, rq);
1403 	unsigned int work_to_do = -1; /* clean all pkts possible */
1404 	unsigned int work_done;
1405 
1406 	if (!enic_poll_lock_poll(&enic->rq[rq]))
1407 		return LL_FLUSH_BUSY;
1408 	work_done = vnic_cq_service(&enic->cq[cq], work_to_do,
1409 				    enic_rq_service, NULL);
1410 
1411 	if (work_done > 0)
1412 		vnic_intr_return_credits(&enic->intr[intr],
1413 					 work_done, 0, 0);
1414 	vnic_rq_fill(&enic->rq[rq], enic_rq_alloc_buf);
1415 	if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1416 		enic_calc_int_moderation(enic, &enic->rq[rq]);
1417 	enic_poll_unlock_poll(&enic->rq[rq]);
1418 
1419 	return work_done;
1420 }
1421 #endif /* CONFIG_NET_RX_BUSY_POLL */
1422 
1423 static int enic_poll_msix_wq(struct napi_struct *napi, int budget)
1424 {
1425 	struct net_device *netdev = napi->dev;
1426 	struct enic *enic = netdev_priv(netdev);
1427 	unsigned int wq_index = (napi - &enic->napi[0]) - enic->rq_count;
1428 	struct vnic_wq *wq = &enic->wq[wq_index];
1429 	unsigned int cq;
1430 	unsigned int intr;
1431 	unsigned int wq_work_to_do = -1; /* clean all desc possible */
1432 	unsigned int wq_work_done;
1433 	unsigned int wq_irq;
1434 
1435 	wq_irq = wq->index;
1436 	cq = enic_cq_wq(enic, wq_irq);
1437 	intr = enic_msix_wq_intr(enic, wq_irq);
1438 	wq_work_done = vnic_cq_service(&enic->cq[cq], wq_work_to_do,
1439 				       enic_wq_service, NULL);
1440 
1441 	vnic_intr_return_credits(&enic->intr[intr], wq_work_done,
1442 				 0 /* don't unmask intr */,
1443 				 1 /* reset intr timer */);
1444 	if (!wq_work_done) {
1445 		napi_complete(napi);
1446 		vnic_intr_unmask(&enic->intr[intr]);
1447 		return 0;
1448 	}
1449 
1450 	return budget;
1451 }
1452 
1453 static int enic_poll_msix_rq(struct napi_struct *napi, int budget)
1454 {
1455 	struct net_device *netdev = napi->dev;
1456 	struct enic *enic = netdev_priv(netdev);
1457 	unsigned int rq = (napi - &enic->napi[0]);
1458 	unsigned int cq = enic_cq_rq(enic, rq);
1459 	unsigned int intr = enic_msix_rq_intr(enic, rq);
1460 	unsigned int work_to_do = budget;
1461 	unsigned int work_done = 0;
1462 	int err;
1463 
1464 	if (!enic_poll_lock_napi(&enic->rq[rq]))
1465 		return budget;
1466 	/* Service RQ
1467 	 */
1468 
1469 	if (budget > 0)
1470 		work_done = vnic_cq_service(&enic->cq[cq],
1471 			work_to_do, enic_rq_service, NULL);
1472 
1473 	/* Return intr event credits for this polling
1474 	 * cycle.  An intr event is the completion of a
1475 	 * RQ packet.
1476 	 */
1477 
1478 	if (work_done > 0)
1479 		vnic_intr_return_credits(&enic->intr[intr],
1480 			work_done,
1481 			0 /* don't unmask intr */,
1482 			0 /* don't reset intr timer */);
1483 
1484 	err = vnic_rq_fill(&enic->rq[rq], enic_rq_alloc_buf);
1485 
1486 	/* Buffer allocation failed. Stay in polling mode
1487 	 * so we can try to fill the ring again.
1488 	 */
1489 
1490 	if (err)
1491 		work_done = work_to_do;
1492 	if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1493 		/* Call the function which refreshes the intr coalescing timer
1494 		 * value based on the traffic.
1495 		 */
1496 		enic_calc_int_moderation(enic, &enic->rq[rq]);
1497 
1498 	enic_poll_unlock_napi(&enic->rq[rq], napi);
1499 	if (work_done < work_to_do) {
1500 
1501 		/* Some work done, but not enough to stay in polling,
1502 		 * exit polling
1503 		 */
1504 
1505 		napi_complete(napi);
1506 		if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1507 			enic_set_int_moderation(enic, &enic->rq[rq]);
1508 		vnic_intr_unmask(&enic->intr[intr]);
1509 	}
1510 
1511 	return work_done;
1512 }
1513 
1514 static void enic_notify_timer(unsigned long data)
1515 {
1516 	struct enic *enic = (struct enic *)data;
1517 
1518 	enic_notify_check(enic);
1519 
1520 	mod_timer(&enic->notify_timer,
1521 		round_jiffies(jiffies + ENIC_NOTIFY_TIMER_PERIOD));
1522 }
1523 
1524 static void enic_free_intr(struct enic *enic)
1525 {
1526 	struct net_device *netdev = enic->netdev;
1527 	unsigned int i;
1528 
1529 	enic_free_rx_cpu_rmap(enic);
1530 	switch (vnic_dev_get_intr_mode(enic->vdev)) {
1531 	case VNIC_DEV_INTR_MODE_INTX:
1532 		free_irq(enic->pdev->irq, netdev);
1533 		break;
1534 	case VNIC_DEV_INTR_MODE_MSI:
1535 		free_irq(enic->pdev->irq, enic);
1536 		break;
1537 	case VNIC_DEV_INTR_MODE_MSIX:
1538 		for (i = 0; i < ARRAY_SIZE(enic->msix); i++)
1539 			if (enic->msix[i].requested)
1540 				free_irq(enic->msix_entry[i].vector,
1541 					enic->msix[i].devid);
1542 		break;
1543 	default:
1544 		break;
1545 	}
1546 }
1547 
1548 static int enic_request_intr(struct enic *enic)
1549 {
1550 	struct net_device *netdev = enic->netdev;
1551 	unsigned int i, intr;
1552 	int err = 0;
1553 
1554 	enic_set_rx_cpu_rmap(enic);
1555 	switch (vnic_dev_get_intr_mode(enic->vdev)) {
1556 
1557 	case VNIC_DEV_INTR_MODE_INTX:
1558 
1559 		err = request_irq(enic->pdev->irq, enic_isr_legacy,
1560 			IRQF_SHARED, netdev->name, netdev);
1561 		break;
1562 
1563 	case VNIC_DEV_INTR_MODE_MSI:
1564 
1565 		err = request_irq(enic->pdev->irq, enic_isr_msi,
1566 			0, netdev->name, enic);
1567 		break;
1568 
1569 	case VNIC_DEV_INTR_MODE_MSIX:
1570 
1571 		for (i = 0; i < enic->rq_count; i++) {
1572 			intr = enic_msix_rq_intr(enic, i);
1573 			snprintf(enic->msix[intr].devname,
1574 				sizeof(enic->msix[intr].devname),
1575 				"%.11s-rx-%u", netdev->name, i);
1576 			enic->msix[intr].isr = enic_isr_msix;
1577 			enic->msix[intr].devid = &enic->napi[i];
1578 		}
1579 
1580 		for (i = 0; i < enic->wq_count; i++) {
1581 			int wq = enic_cq_wq(enic, i);
1582 
1583 			intr = enic_msix_wq_intr(enic, i);
1584 			snprintf(enic->msix[intr].devname,
1585 				sizeof(enic->msix[intr].devname),
1586 				"%.11s-tx-%u", netdev->name, i);
1587 			enic->msix[intr].isr = enic_isr_msix;
1588 			enic->msix[intr].devid = &enic->napi[wq];
1589 		}
1590 
1591 		intr = enic_msix_err_intr(enic);
1592 		snprintf(enic->msix[intr].devname,
1593 			sizeof(enic->msix[intr].devname),
1594 			"%.11s-err", netdev->name);
1595 		enic->msix[intr].isr = enic_isr_msix_err;
1596 		enic->msix[intr].devid = enic;
1597 
1598 		intr = enic_msix_notify_intr(enic);
1599 		snprintf(enic->msix[intr].devname,
1600 			sizeof(enic->msix[intr].devname),
1601 			"%.11s-notify", netdev->name);
1602 		enic->msix[intr].isr = enic_isr_msix_notify;
1603 		enic->msix[intr].devid = enic;
1604 
1605 		for (i = 0; i < ARRAY_SIZE(enic->msix); i++)
1606 			enic->msix[i].requested = 0;
1607 
1608 		for (i = 0; i < enic->intr_count; i++) {
1609 			err = request_irq(enic->msix_entry[i].vector,
1610 				enic->msix[i].isr, 0,
1611 				enic->msix[i].devname,
1612 				enic->msix[i].devid);
1613 			if (err) {
1614 				enic_free_intr(enic);
1615 				break;
1616 			}
1617 			enic->msix[i].requested = 1;
1618 		}
1619 
1620 		break;
1621 
1622 	default:
1623 		break;
1624 	}
1625 
1626 	return err;
1627 }
1628 
1629 static void enic_synchronize_irqs(struct enic *enic)
1630 {
1631 	unsigned int i;
1632 
1633 	switch (vnic_dev_get_intr_mode(enic->vdev)) {
1634 	case VNIC_DEV_INTR_MODE_INTX:
1635 	case VNIC_DEV_INTR_MODE_MSI:
1636 		synchronize_irq(enic->pdev->irq);
1637 		break;
1638 	case VNIC_DEV_INTR_MODE_MSIX:
1639 		for (i = 0; i < enic->intr_count; i++)
1640 			synchronize_irq(enic->msix_entry[i].vector);
1641 		break;
1642 	default:
1643 		break;
1644 	}
1645 }
1646 
1647 static void enic_set_rx_coal_setting(struct enic *enic)
1648 {
1649 	unsigned int speed;
1650 	int index = -1;
1651 	struct enic_rx_coal *rx_coal = &enic->rx_coalesce_setting;
1652 
1653 	/* 1. Read the link speed from fw
1654 	 * 2. Pick the default range for the speed
1655 	 * 3. Update it in enic->rx_coalesce_setting
1656 	 */
1657 	speed = vnic_dev_port_speed(enic->vdev);
1658 	if (ENIC_LINK_SPEED_10G < speed)
1659 		index = ENIC_LINK_40G_INDEX;
1660 	else if (ENIC_LINK_SPEED_4G < speed)
1661 		index = ENIC_LINK_10G_INDEX;
1662 	else
1663 		index = ENIC_LINK_4G_INDEX;
1664 
1665 	rx_coal->small_pkt_range_start = mod_range[index].small_pkt_range_start;
1666 	rx_coal->large_pkt_range_start = mod_range[index].large_pkt_range_start;
1667 	rx_coal->range_end = ENIC_RX_COALESCE_RANGE_END;
1668 
1669 	/* Start with the value provided by UCSM */
1670 	for (index = 0; index < enic->rq_count; index++)
1671 		enic->cq[index].cur_rx_coal_timeval =
1672 				enic->config.intr_timer_usec;
1673 
1674 	rx_coal->use_adaptive_rx_coalesce = 1;
1675 }
1676 
1677 static int enic_dev_notify_set(struct enic *enic)
1678 {
1679 	int err;
1680 
1681 	spin_lock_bh(&enic->devcmd_lock);
1682 	switch (vnic_dev_get_intr_mode(enic->vdev)) {
1683 	case VNIC_DEV_INTR_MODE_INTX:
1684 		err = vnic_dev_notify_set(enic->vdev,
1685 			enic_legacy_notify_intr());
1686 		break;
1687 	case VNIC_DEV_INTR_MODE_MSIX:
1688 		err = vnic_dev_notify_set(enic->vdev,
1689 			enic_msix_notify_intr(enic));
1690 		break;
1691 	default:
1692 		err = vnic_dev_notify_set(enic->vdev, -1 /* no intr */);
1693 		break;
1694 	}
1695 	spin_unlock_bh(&enic->devcmd_lock);
1696 
1697 	return err;
1698 }
1699 
1700 static void enic_notify_timer_start(struct enic *enic)
1701 {
1702 	switch (vnic_dev_get_intr_mode(enic->vdev)) {
1703 	case VNIC_DEV_INTR_MODE_MSI:
1704 		mod_timer(&enic->notify_timer, jiffies);
1705 		break;
1706 	default:
1707 		/* Using intr for notification for INTx/MSI-X */
1708 		break;
1709 	}
1710 }
1711 
1712 /* rtnl lock is held, process context */
1713 static int enic_open(struct net_device *netdev)
1714 {
1715 	struct enic *enic = netdev_priv(netdev);
1716 	unsigned int i;
1717 	int err;
1718 
1719 	err = enic_request_intr(enic);
1720 	if (err) {
1721 		netdev_err(netdev, "Unable to request irq.\n");
1722 		return err;
1723 	}
1724 	enic_init_affinity_hint(enic);
1725 	enic_set_affinity_hint(enic);
1726 
1727 	err = enic_dev_notify_set(enic);
1728 	if (err) {
1729 		netdev_err(netdev,
1730 			"Failed to alloc notify buffer, aborting.\n");
1731 		goto err_out_free_intr;
1732 	}
1733 
1734 	for (i = 0; i < enic->rq_count; i++) {
1735 		vnic_rq_fill(&enic->rq[i], enic_rq_alloc_buf);
1736 		/* Need at least one buffer on ring to get going */
1737 		if (vnic_rq_desc_used(&enic->rq[i]) == 0) {
1738 			netdev_err(netdev, "Unable to alloc receive buffers\n");
1739 			err = -ENOMEM;
1740 			goto err_out_free_rq;
1741 		}
1742 	}
1743 
1744 	for (i = 0; i < enic->wq_count; i++)
1745 		vnic_wq_enable(&enic->wq[i]);
1746 	for (i = 0; i < enic->rq_count; i++)
1747 		vnic_rq_enable(&enic->rq[i]);
1748 
1749 	if (!enic_is_dynamic(enic) && !enic_is_sriov_vf(enic))
1750 		enic_dev_add_station_addr(enic);
1751 
1752 	enic_set_rx_mode(netdev);
1753 
1754 	netif_tx_wake_all_queues(netdev);
1755 
1756 	for (i = 0; i < enic->rq_count; i++) {
1757 		enic_busy_poll_init_lock(&enic->rq[i]);
1758 		napi_enable(&enic->napi[i]);
1759 	}
1760 	if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX)
1761 		for (i = 0; i < enic->wq_count; i++)
1762 			napi_enable(&enic->napi[enic_cq_wq(enic, i)]);
1763 	enic_dev_enable(enic);
1764 
1765 	for (i = 0; i < enic->intr_count; i++)
1766 		vnic_intr_unmask(&enic->intr[i]);
1767 
1768 	enic_notify_timer_start(enic);
1769 	enic_rfs_flw_tbl_init(enic);
1770 
1771 	return 0;
1772 
1773 err_out_free_rq:
1774 	for (i = 0; i < enic->rq_count; i++)
1775 		vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
1776 	enic_dev_notify_unset(enic);
1777 err_out_free_intr:
1778 	enic_unset_affinity_hint(enic);
1779 	enic_free_intr(enic);
1780 
1781 	return err;
1782 }
1783 
1784 /* rtnl lock is held, process context */
1785 static int enic_stop(struct net_device *netdev)
1786 {
1787 	struct enic *enic = netdev_priv(netdev);
1788 	unsigned int i;
1789 	int err;
1790 
1791 	for (i = 0; i < enic->intr_count; i++) {
1792 		vnic_intr_mask(&enic->intr[i]);
1793 		(void)vnic_intr_masked(&enic->intr[i]); /* flush write */
1794 	}
1795 
1796 	enic_synchronize_irqs(enic);
1797 
1798 	del_timer_sync(&enic->notify_timer);
1799 	enic_rfs_flw_tbl_free(enic);
1800 
1801 	enic_dev_disable(enic);
1802 
1803 	for (i = 0; i < enic->rq_count; i++) {
1804 		napi_disable(&enic->napi[i]);
1805 		local_bh_disable();
1806 		while (!enic_poll_lock_napi(&enic->rq[i]))
1807 			mdelay(1);
1808 		local_bh_enable();
1809 	}
1810 
1811 	netif_carrier_off(netdev);
1812 	netif_tx_disable(netdev);
1813 	if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX)
1814 		for (i = 0; i < enic->wq_count; i++)
1815 			napi_disable(&enic->napi[enic_cq_wq(enic, i)]);
1816 
1817 	if (!enic_is_dynamic(enic) && !enic_is_sriov_vf(enic))
1818 		enic_dev_del_station_addr(enic);
1819 
1820 	for (i = 0; i < enic->wq_count; i++) {
1821 		err = vnic_wq_disable(&enic->wq[i]);
1822 		if (err)
1823 			return err;
1824 	}
1825 	for (i = 0; i < enic->rq_count; i++) {
1826 		err = vnic_rq_disable(&enic->rq[i]);
1827 		if (err)
1828 			return err;
1829 	}
1830 
1831 	enic_dev_notify_unset(enic);
1832 	enic_unset_affinity_hint(enic);
1833 	enic_free_intr(enic);
1834 
1835 	for (i = 0; i < enic->wq_count; i++)
1836 		vnic_wq_clean(&enic->wq[i], enic_free_wq_buf);
1837 	for (i = 0; i < enic->rq_count; i++)
1838 		vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
1839 	for (i = 0; i < enic->cq_count; i++)
1840 		vnic_cq_clean(&enic->cq[i]);
1841 	for (i = 0; i < enic->intr_count; i++)
1842 		vnic_intr_clean(&enic->intr[i]);
1843 
1844 	return 0;
1845 }
1846 
1847 static int enic_change_mtu(struct net_device *netdev, int new_mtu)
1848 {
1849 	struct enic *enic = netdev_priv(netdev);
1850 	int running = netif_running(netdev);
1851 
1852 	if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic))
1853 		return -EOPNOTSUPP;
1854 
1855 	if (running)
1856 		enic_stop(netdev);
1857 
1858 	netdev->mtu = new_mtu;
1859 
1860 	if (netdev->mtu > enic->port_mtu)
1861 		netdev_warn(netdev,
1862 			"interface MTU (%d) set higher than port MTU (%d)\n",
1863 			netdev->mtu, enic->port_mtu);
1864 
1865 	if (running)
1866 		enic_open(netdev);
1867 
1868 	return 0;
1869 }
1870 
1871 static void enic_change_mtu_work(struct work_struct *work)
1872 {
1873 	struct enic *enic = container_of(work, struct enic, change_mtu_work);
1874 	struct net_device *netdev = enic->netdev;
1875 	int new_mtu = vnic_dev_mtu(enic->vdev);
1876 	int err;
1877 	unsigned int i;
1878 
1879 	new_mtu = max_t(int, ENIC_MIN_MTU, min_t(int, ENIC_MAX_MTU, new_mtu));
1880 
1881 	rtnl_lock();
1882 
1883 	/* Stop RQ */
1884 	del_timer_sync(&enic->notify_timer);
1885 
1886 	for (i = 0; i < enic->rq_count; i++)
1887 		napi_disable(&enic->napi[i]);
1888 
1889 	vnic_intr_mask(&enic->intr[0]);
1890 	enic_synchronize_irqs(enic);
1891 	err = vnic_rq_disable(&enic->rq[0]);
1892 	if (err) {
1893 		rtnl_unlock();
1894 		netdev_err(netdev, "Unable to disable RQ.\n");
1895 		return;
1896 	}
1897 	vnic_rq_clean(&enic->rq[0], enic_free_rq_buf);
1898 	vnic_cq_clean(&enic->cq[0]);
1899 	vnic_intr_clean(&enic->intr[0]);
1900 
1901 	/* Fill RQ with new_mtu-sized buffers */
1902 	netdev->mtu = new_mtu;
1903 	vnic_rq_fill(&enic->rq[0], enic_rq_alloc_buf);
1904 	/* Need at least one buffer on ring to get going */
1905 	if (vnic_rq_desc_used(&enic->rq[0]) == 0) {
1906 		rtnl_unlock();
1907 		netdev_err(netdev, "Unable to alloc receive buffers.\n");
1908 		return;
1909 	}
1910 
1911 	/* Start RQ */
1912 	vnic_rq_enable(&enic->rq[0]);
1913 	napi_enable(&enic->napi[0]);
1914 	vnic_intr_unmask(&enic->intr[0]);
1915 	enic_notify_timer_start(enic);
1916 
1917 	rtnl_unlock();
1918 
1919 	netdev_info(netdev, "interface MTU set as %d\n", netdev->mtu);
1920 }
1921 
1922 #ifdef CONFIG_NET_POLL_CONTROLLER
1923 static void enic_poll_controller(struct net_device *netdev)
1924 {
1925 	struct enic *enic = netdev_priv(netdev);
1926 	struct vnic_dev *vdev = enic->vdev;
1927 	unsigned int i, intr;
1928 
1929 	switch (vnic_dev_get_intr_mode(vdev)) {
1930 	case VNIC_DEV_INTR_MODE_MSIX:
1931 		for (i = 0; i < enic->rq_count; i++) {
1932 			intr = enic_msix_rq_intr(enic, i);
1933 			enic_isr_msix(enic->msix_entry[intr].vector,
1934 				      &enic->napi[i]);
1935 		}
1936 
1937 		for (i = 0; i < enic->wq_count; i++) {
1938 			intr = enic_msix_wq_intr(enic, i);
1939 			enic_isr_msix(enic->msix_entry[intr].vector,
1940 				      &enic->napi[enic_cq_wq(enic, i)]);
1941 		}
1942 
1943 		break;
1944 	case VNIC_DEV_INTR_MODE_MSI:
1945 		enic_isr_msi(enic->pdev->irq, enic);
1946 		break;
1947 	case VNIC_DEV_INTR_MODE_INTX:
1948 		enic_isr_legacy(enic->pdev->irq, netdev);
1949 		break;
1950 	default:
1951 		break;
1952 	}
1953 }
1954 #endif
1955 
1956 static int enic_dev_wait(struct vnic_dev *vdev,
1957 	int (*start)(struct vnic_dev *, int),
1958 	int (*finished)(struct vnic_dev *, int *),
1959 	int arg)
1960 {
1961 	unsigned long time;
1962 	int done;
1963 	int err;
1964 
1965 	BUG_ON(in_interrupt());
1966 
1967 	err = start(vdev, arg);
1968 	if (err)
1969 		return err;
1970 
1971 	/* Wait for func to complete...2 seconds max
1972 	 */
1973 
1974 	time = jiffies + (HZ * 2);
1975 	do {
1976 
1977 		err = finished(vdev, &done);
1978 		if (err)
1979 			return err;
1980 
1981 		if (done)
1982 			return 0;
1983 
1984 		schedule_timeout_uninterruptible(HZ / 10);
1985 
1986 	} while (time_after(time, jiffies));
1987 
1988 	return -ETIMEDOUT;
1989 }
1990 
1991 static int enic_dev_open(struct enic *enic)
1992 {
1993 	int err;
1994 
1995 	err = enic_dev_wait(enic->vdev, vnic_dev_open,
1996 		vnic_dev_open_done, 0);
1997 	if (err)
1998 		dev_err(enic_get_dev(enic), "vNIC device open failed, err %d\n",
1999 			err);
2000 
2001 	return err;
2002 }
2003 
2004 static int enic_dev_soft_reset(struct enic *enic)
2005 {
2006 	int err;
2007 
2008 	err = enic_dev_wait(enic->vdev, vnic_dev_soft_reset,
2009 			    vnic_dev_soft_reset_done, 0);
2010 	if (err)
2011 		netdev_err(enic->netdev, "vNIC soft reset failed, err %d\n",
2012 			   err);
2013 
2014 	return err;
2015 }
2016 
2017 static int enic_dev_hang_reset(struct enic *enic)
2018 {
2019 	int err;
2020 
2021 	err = enic_dev_wait(enic->vdev, vnic_dev_hang_reset,
2022 		vnic_dev_hang_reset_done, 0);
2023 	if (err)
2024 		netdev_err(enic->netdev, "vNIC hang reset failed, err %d\n",
2025 			err);
2026 
2027 	return err;
2028 }
2029 
2030 int __enic_set_rsskey(struct enic *enic)
2031 {
2032 	union vnic_rss_key *rss_key_buf_va;
2033 	dma_addr_t rss_key_buf_pa;
2034 	int i, kidx, bidx, err;
2035 
2036 	rss_key_buf_va = pci_zalloc_consistent(enic->pdev,
2037 					       sizeof(union vnic_rss_key),
2038 					       &rss_key_buf_pa);
2039 	if (!rss_key_buf_va)
2040 		return -ENOMEM;
2041 
2042 	for (i = 0; i < ENIC_RSS_LEN; i++) {
2043 		kidx = i / ENIC_RSS_BYTES_PER_KEY;
2044 		bidx = i % ENIC_RSS_BYTES_PER_KEY;
2045 		rss_key_buf_va->key[kidx].b[bidx] = enic->rss_key[i];
2046 	}
2047 	spin_lock_bh(&enic->devcmd_lock);
2048 	err = enic_set_rss_key(enic,
2049 		rss_key_buf_pa,
2050 		sizeof(union vnic_rss_key));
2051 	spin_unlock_bh(&enic->devcmd_lock);
2052 
2053 	pci_free_consistent(enic->pdev, sizeof(union vnic_rss_key),
2054 		rss_key_buf_va, rss_key_buf_pa);
2055 
2056 	return err;
2057 }
2058 
2059 static int enic_set_rsskey(struct enic *enic)
2060 {
2061 	netdev_rss_key_fill(enic->rss_key, ENIC_RSS_LEN);
2062 
2063 	return __enic_set_rsskey(enic);
2064 }
2065 
2066 static int enic_set_rsscpu(struct enic *enic, u8 rss_hash_bits)
2067 {
2068 	dma_addr_t rss_cpu_buf_pa;
2069 	union vnic_rss_cpu *rss_cpu_buf_va = NULL;
2070 	unsigned int i;
2071 	int err;
2072 
2073 	rss_cpu_buf_va = pci_alloc_consistent(enic->pdev,
2074 		sizeof(union vnic_rss_cpu), &rss_cpu_buf_pa);
2075 	if (!rss_cpu_buf_va)
2076 		return -ENOMEM;
2077 
2078 	for (i = 0; i < (1 << rss_hash_bits); i++)
2079 		(*rss_cpu_buf_va).cpu[i/4].b[i%4] = i % enic->rq_count;
2080 
2081 	spin_lock_bh(&enic->devcmd_lock);
2082 	err = enic_set_rss_cpu(enic,
2083 		rss_cpu_buf_pa,
2084 		sizeof(union vnic_rss_cpu));
2085 	spin_unlock_bh(&enic->devcmd_lock);
2086 
2087 	pci_free_consistent(enic->pdev, sizeof(union vnic_rss_cpu),
2088 		rss_cpu_buf_va, rss_cpu_buf_pa);
2089 
2090 	return err;
2091 }
2092 
2093 static int enic_set_niccfg(struct enic *enic, u8 rss_default_cpu,
2094 	u8 rss_hash_type, u8 rss_hash_bits, u8 rss_base_cpu, u8 rss_enable)
2095 {
2096 	const u8 tso_ipid_split_en = 0;
2097 	const u8 ig_vlan_strip_en = 1;
2098 	int err;
2099 
2100 	/* Enable VLAN tag stripping.
2101 	*/
2102 
2103 	spin_lock_bh(&enic->devcmd_lock);
2104 	err = enic_set_nic_cfg(enic,
2105 		rss_default_cpu, rss_hash_type,
2106 		rss_hash_bits, rss_base_cpu,
2107 		rss_enable, tso_ipid_split_en,
2108 		ig_vlan_strip_en);
2109 	spin_unlock_bh(&enic->devcmd_lock);
2110 
2111 	return err;
2112 }
2113 
2114 static int enic_set_rss_nic_cfg(struct enic *enic)
2115 {
2116 	struct device *dev = enic_get_dev(enic);
2117 	const u8 rss_default_cpu = 0;
2118 	const u8 rss_hash_type = NIC_CFG_RSS_HASH_TYPE_IPV4 |
2119 		NIC_CFG_RSS_HASH_TYPE_TCP_IPV4 |
2120 		NIC_CFG_RSS_HASH_TYPE_IPV6 |
2121 		NIC_CFG_RSS_HASH_TYPE_TCP_IPV6;
2122 	const u8 rss_hash_bits = 7;
2123 	const u8 rss_base_cpu = 0;
2124 	u8 rss_enable = ENIC_SETTING(enic, RSS) && (enic->rq_count > 1);
2125 
2126 	if (rss_enable) {
2127 		if (!enic_set_rsskey(enic)) {
2128 			if (enic_set_rsscpu(enic, rss_hash_bits)) {
2129 				rss_enable = 0;
2130 				dev_warn(dev, "RSS disabled, "
2131 					"Failed to set RSS cpu indirection table.");
2132 			}
2133 		} else {
2134 			rss_enable = 0;
2135 			dev_warn(dev, "RSS disabled, Failed to set RSS key.\n");
2136 		}
2137 	}
2138 
2139 	return enic_set_niccfg(enic, rss_default_cpu, rss_hash_type,
2140 		rss_hash_bits, rss_base_cpu, rss_enable);
2141 }
2142 
2143 static void enic_reset(struct work_struct *work)
2144 {
2145 	struct enic *enic = container_of(work, struct enic, reset);
2146 
2147 	if (!netif_running(enic->netdev))
2148 		return;
2149 
2150 	rtnl_lock();
2151 
2152 	spin_lock(&enic->enic_api_lock);
2153 	enic_stop(enic->netdev);
2154 	enic_dev_soft_reset(enic);
2155 	enic_reset_addr_lists(enic);
2156 	enic_init_vnic_resources(enic);
2157 	enic_set_rss_nic_cfg(enic);
2158 	enic_dev_set_ig_vlan_rewrite_mode(enic);
2159 	enic_open(enic->netdev);
2160 	spin_unlock(&enic->enic_api_lock);
2161 	call_netdevice_notifiers(NETDEV_REBOOT, enic->netdev);
2162 
2163 	rtnl_unlock();
2164 }
2165 
2166 static void enic_tx_hang_reset(struct work_struct *work)
2167 {
2168 	struct enic *enic = container_of(work, struct enic, tx_hang_reset);
2169 
2170 	rtnl_lock();
2171 
2172 	spin_lock(&enic->enic_api_lock);
2173 	enic_dev_hang_notify(enic);
2174 	enic_stop(enic->netdev);
2175 	enic_dev_hang_reset(enic);
2176 	enic_reset_addr_lists(enic);
2177 	enic_init_vnic_resources(enic);
2178 	enic_set_rss_nic_cfg(enic);
2179 	enic_dev_set_ig_vlan_rewrite_mode(enic);
2180 	enic_open(enic->netdev);
2181 	spin_unlock(&enic->enic_api_lock);
2182 	call_netdevice_notifiers(NETDEV_REBOOT, enic->netdev);
2183 
2184 	rtnl_unlock();
2185 }
2186 
2187 static int enic_set_intr_mode(struct enic *enic)
2188 {
2189 	unsigned int n = min_t(unsigned int, enic->rq_count, ENIC_RQ_MAX);
2190 	unsigned int m = min_t(unsigned int, enic->wq_count, ENIC_WQ_MAX);
2191 	unsigned int i;
2192 
2193 	/* Set interrupt mode (INTx, MSI, MSI-X) depending
2194 	 * on system capabilities.
2195 	 *
2196 	 * Try MSI-X first
2197 	 *
2198 	 * We need n RQs, m WQs, n+m CQs, and n+m+2 INTRs
2199 	 * (the second to last INTR is used for WQ/RQ errors)
2200 	 * (the last INTR is used for notifications)
2201 	 */
2202 
2203 	BUG_ON(ARRAY_SIZE(enic->msix_entry) < n + m + 2);
2204 	for (i = 0; i < n + m + 2; i++)
2205 		enic->msix_entry[i].entry = i;
2206 
2207 	/* Use multiple RQs if RSS is enabled
2208 	 */
2209 
2210 	if (ENIC_SETTING(enic, RSS) &&
2211 	    enic->config.intr_mode < 1 &&
2212 	    enic->rq_count >= n &&
2213 	    enic->wq_count >= m &&
2214 	    enic->cq_count >= n + m &&
2215 	    enic->intr_count >= n + m + 2) {
2216 
2217 		if (pci_enable_msix_range(enic->pdev, enic->msix_entry,
2218 					  n + m + 2, n + m + 2) > 0) {
2219 
2220 			enic->rq_count = n;
2221 			enic->wq_count = m;
2222 			enic->cq_count = n + m;
2223 			enic->intr_count = n + m + 2;
2224 
2225 			vnic_dev_set_intr_mode(enic->vdev,
2226 				VNIC_DEV_INTR_MODE_MSIX);
2227 
2228 			return 0;
2229 		}
2230 	}
2231 
2232 	if (enic->config.intr_mode < 1 &&
2233 	    enic->rq_count >= 1 &&
2234 	    enic->wq_count >= m &&
2235 	    enic->cq_count >= 1 + m &&
2236 	    enic->intr_count >= 1 + m + 2) {
2237 		if (pci_enable_msix_range(enic->pdev, enic->msix_entry,
2238 					  1 + m + 2, 1 + m + 2) > 0) {
2239 
2240 			enic->rq_count = 1;
2241 			enic->wq_count = m;
2242 			enic->cq_count = 1 + m;
2243 			enic->intr_count = 1 + m + 2;
2244 
2245 			vnic_dev_set_intr_mode(enic->vdev,
2246 				VNIC_DEV_INTR_MODE_MSIX);
2247 
2248 			return 0;
2249 		}
2250 	}
2251 
2252 	/* Next try MSI
2253 	 *
2254 	 * We need 1 RQ, 1 WQ, 2 CQs, and 1 INTR
2255 	 */
2256 
2257 	if (enic->config.intr_mode < 2 &&
2258 	    enic->rq_count >= 1 &&
2259 	    enic->wq_count >= 1 &&
2260 	    enic->cq_count >= 2 &&
2261 	    enic->intr_count >= 1 &&
2262 	    !pci_enable_msi(enic->pdev)) {
2263 
2264 		enic->rq_count = 1;
2265 		enic->wq_count = 1;
2266 		enic->cq_count = 2;
2267 		enic->intr_count = 1;
2268 
2269 		vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSI);
2270 
2271 		return 0;
2272 	}
2273 
2274 	/* Next try INTx
2275 	 *
2276 	 * We need 1 RQ, 1 WQ, 2 CQs, and 3 INTRs
2277 	 * (the first INTR is used for WQ/RQ)
2278 	 * (the second INTR is used for WQ/RQ errors)
2279 	 * (the last INTR is used for notifications)
2280 	 */
2281 
2282 	if (enic->config.intr_mode < 3 &&
2283 	    enic->rq_count >= 1 &&
2284 	    enic->wq_count >= 1 &&
2285 	    enic->cq_count >= 2 &&
2286 	    enic->intr_count >= 3) {
2287 
2288 		enic->rq_count = 1;
2289 		enic->wq_count = 1;
2290 		enic->cq_count = 2;
2291 		enic->intr_count = 3;
2292 
2293 		vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_INTX);
2294 
2295 		return 0;
2296 	}
2297 
2298 	vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
2299 
2300 	return -EINVAL;
2301 }
2302 
2303 static void enic_clear_intr_mode(struct enic *enic)
2304 {
2305 	switch (vnic_dev_get_intr_mode(enic->vdev)) {
2306 	case VNIC_DEV_INTR_MODE_MSIX:
2307 		pci_disable_msix(enic->pdev);
2308 		break;
2309 	case VNIC_DEV_INTR_MODE_MSI:
2310 		pci_disable_msi(enic->pdev);
2311 		break;
2312 	default:
2313 		break;
2314 	}
2315 
2316 	vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
2317 }
2318 
2319 static const struct net_device_ops enic_netdev_dynamic_ops = {
2320 	.ndo_open		= enic_open,
2321 	.ndo_stop		= enic_stop,
2322 	.ndo_start_xmit		= enic_hard_start_xmit,
2323 	.ndo_get_stats64	= enic_get_stats,
2324 	.ndo_validate_addr	= eth_validate_addr,
2325 	.ndo_set_rx_mode	= enic_set_rx_mode,
2326 	.ndo_set_mac_address	= enic_set_mac_address_dynamic,
2327 	.ndo_change_mtu		= enic_change_mtu,
2328 	.ndo_vlan_rx_add_vid	= enic_vlan_rx_add_vid,
2329 	.ndo_vlan_rx_kill_vid	= enic_vlan_rx_kill_vid,
2330 	.ndo_tx_timeout		= enic_tx_timeout,
2331 	.ndo_set_vf_port	= enic_set_vf_port,
2332 	.ndo_get_vf_port	= enic_get_vf_port,
2333 	.ndo_set_vf_mac		= enic_set_vf_mac,
2334 #ifdef CONFIG_NET_POLL_CONTROLLER
2335 	.ndo_poll_controller	= enic_poll_controller,
2336 #endif
2337 #ifdef CONFIG_RFS_ACCEL
2338 	.ndo_rx_flow_steer	= enic_rx_flow_steer,
2339 #endif
2340 #ifdef CONFIG_NET_RX_BUSY_POLL
2341 	.ndo_busy_poll		= enic_busy_poll,
2342 #endif
2343 };
2344 
2345 static const struct net_device_ops enic_netdev_ops = {
2346 	.ndo_open		= enic_open,
2347 	.ndo_stop		= enic_stop,
2348 	.ndo_start_xmit		= enic_hard_start_xmit,
2349 	.ndo_get_stats64	= enic_get_stats,
2350 	.ndo_validate_addr	= eth_validate_addr,
2351 	.ndo_set_mac_address	= enic_set_mac_address,
2352 	.ndo_set_rx_mode	= enic_set_rx_mode,
2353 	.ndo_change_mtu		= enic_change_mtu,
2354 	.ndo_vlan_rx_add_vid	= enic_vlan_rx_add_vid,
2355 	.ndo_vlan_rx_kill_vid	= enic_vlan_rx_kill_vid,
2356 	.ndo_tx_timeout		= enic_tx_timeout,
2357 	.ndo_set_vf_port	= enic_set_vf_port,
2358 	.ndo_get_vf_port	= enic_get_vf_port,
2359 	.ndo_set_vf_mac		= enic_set_vf_mac,
2360 #ifdef CONFIG_NET_POLL_CONTROLLER
2361 	.ndo_poll_controller	= enic_poll_controller,
2362 #endif
2363 #ifdef CONFIG_RFS_ACCEL
2364 	.ndo_rx_flow_steer	= enic_rx_flow_steer,
2365 #endif
2366 #ifdef CONFIG_NET_RX_BUSY_POLL
2367 	.ndo_busy_poll		= enic_busy_poll,
2368 #endif
2369 };
2370 
2371 static void enic_dev_deinit(struct enic *enic)
2372 {
2373 	unsigned int i;
2374 
2375 	for (i = 0; i < enic->rq_count; i++) {
2376 		napi_hash_del(&enic->napi[i]);
2377 		netif_napi_del(&enic->napi[i]);
2378 	}
2379 	if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX)
2380 		for (i = 0; i < enic->wq_count; i++)
2381 			netif_napi_del(&enic->napi[enic_cq_wq(enic, i)]);
2382 
2383 	enic_free_vnic_resources(enic);
2384 	enic_clear_intr_mode(enic);
2385 	enic_free_affinity_hint(enic);
2386 }
2387 
2388 static void enic_kdump_kernel_config(struct enic *enic)
2389 {
2390 	if (is_kdump_kernel()) {
2391 		dev_info(enic_get_dev(enic), "Running from within kdump kernel. Using minimal resources\n");
2392 		enic->rq_count = 1;
2393 		enic->wq_count = 1;
2394 		enic->config.rq_desc_count = ENIC_MIN_RQ_DESCS;
2395 		enic->config.wq_desc_count = ENIC_MIN_WQ_DESCS;
2396 		enic->config.mtu = min_t(u16, 1500, enic->config.mtu);
2397 	}
2398 }
2399 
2400 static int enic_dev_init(struct enic *enic)
2401 {
2402 	struct device *dev = enic_get_dev(enic);
2403 	struct net_device *netdev = enic->netdev;
2404 	unsigned int i;
2405 	int err;
2406 
2407 	/* Get interrupt coalesce timer info */
2408 	err = enic_dev_intr_coal_timer_info(enic);
2409 	if (err) {
2410 		dev_warn(dev, "Using default conversion factor for "
2411 			"interrupt coalesce timer\n");
2412 		vnic_dev_intr_coal_timer_info_default(enic->vdev);
2413 	}
2414 
2415 	/* Get vNIC configuration
2416 	 */
2417 
2418 	err = enic_get_vnic_config(enic);
2419 	if (err) {
2420 		dev_err(dev, "Get vNIC configuration failed, aborting\n");
2421 		return err;
2422 	}
2423 
2424 	/* Get available resource counts
2425 	 */
2426 
2427 	enic_get_res_counts(enic);
2428 
2429 	/* modify resource count if we are in kdump_kernel
2430 	 */
2431 	enic_kdump_kernel_config(enic);
2432 
2433 	/* Set interrupt mode based on resource counts and system
2434 	 * capabilities
2435 	 */
2436 
2437 	err = enic_set_intr_mode(enic);
2438 	if (err) {
2439 		dev_err(dev, "Failed to set intr mode based on resource "
2440 			"counts and system capabilities, aborting\n");
2441 		return err;
2442 	}
2443 
2444 	/* Allocate and configure vNIC resources
2445 	 */
2446 
2447 	err = enic_alloc_vnic_resources(enic);
2448 	if (err) {
2449 		dev_err(dev, "Failed to alloc vNIC resources, aborting\n");
2450 		goto err_out_free_vnic_resources;
2451 	}
2452 
2453 	enic_init_vnic_resources(enic);
2454 
2455 	err = enic_set_rss_nic_cfg(enic);
2456 	if (err) {
2457 		dev_err(dev, "Failed to config nic, aborting\n");
2458 		goto err_out_free_vnic_resources;
2459 	}
2460 
2461 	switch (vnic_dev_get_intr_mode(enic->vdev)) {
2462 	default:
2463 		netif_napi_add(netdev, &enic->napi[0], enic_poll, 64);
2464 		break;
2465 	case VNIC_DEV_INTR_MODE_MSIX:
2466 		for (i = 0; i < enic->rq_count; i++) {
2467 			netif_napi_add(netdev, &enic->napi[i],
2468 				enic_poll_msix_rq, NAPI_POLL_WEIGHT);
2469 		}
2470 		for (i = 0; i < enic->wq_count; i++)
2471 			netif_napi_add(netdev, &enic->napi[enic_cq_wq(enic, i)],
2472 				       enic_poll_msix_wq, NAPI_POLL_WEIGHT);
2473 		break;
2474 	}
2475 
2476 	return 0;
2477 
2478 err_out_free_vnic_resources:
2479 	enic_free_affinity_hint(enic);
2480 	enic_clear_intr_mode(enic);
2481 	enic_free_vnic_resources(enic);
2482 
2483 	return err;
2484 }
2485 
2486 static void enic_iounmap(struct enic *enic)
2487 {
2488 	unsigned int i;
2489 
2490 	for (i = 0; i < ARRAY_SIZE(enic->bar); i++)
2491 		if (enic->bar[i].vaddr)
2492 			iounmap(enic->bar[i].vaddr);
2493 }
2494 
2495 static int enic_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2496 {
2497 	struct device *dev = &pdev->dev;
2498 	struct net_device *netdev;
2499 	struct enic *enic;
2500 	int using_dac = 0;
2501 	unsigned int i;
2502 	int err;
2503 #ifdef CONFIG_PCI_IOV
2504 	int pos = 0;
2505 #endif
2506 	int num_pps = 1;
2507 
2508 	/* Allocate net device structure and initialize.  Private
2509 	 * instance data is initialized to zero.
2510 	 */
2511 
2512 	netdev = alloc_etherdev_mqs(sizeof(struct enic),
2513 				    ENIC_RQ_MAX, ENIC_WQ_MAX);
2514 	if (!netdev)
2515 		return -ENOMEM;
2516 
2517 	pci_set_drvdata(pdev, netdev);
2518 
2519 	SET_NETDEV_DEV(netdev, &pdev->dev);
2520 
2521 	enic = netdev_priv(netdev);
2522 	enic->netdev = netdev;
2523 	enic->pdev = pdev;
2524 
2525 	/* Setup PCI resources
2526 	 */
2527 
2528 	err = pci_enable_device_mem(pdev);
2529 	if (err) {
2530 		dev_err(dev, "Cannot enable PCI device, aborting\n");
2531 		goto err_out_free_netdev;
2532 	}
2533 
2534 	err = pci_request_regions(pdev, DRV_NAME);
2535 	if (err) {
2536 		dev_err(dev, "Cannot request PCI regions, aborting\n");
2537 		goto err_out_disable_device;
2538 	}
2539 
2540 	pci_set_master(pdev);
2541 
2542 	/* Query PCI controller on system for DMA addressing
2543 	 * limitation for the device.  Try 64-bit first, and
2544 	 * fail to 32-bit.
2545 	 */
2546 
2547 	err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
2548 	if (err) {
2549 		err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
2550 		if (err) {
2551 			dev_err(dev, "No usable DMA configuration, aborting\n");
2552 			goto err_out_release_regions;
2553 		}
2554 		err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
2555 		if (err) {
2556 			dev_err(dev, "Unable to obtain %u-bit DMA "
2557 				"for consistent allocations, aborting\n", 32);
2558 			goto err_out_release_regions;
2559 		}
2560 	} else {
2561 		err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
2562 		if (err) {
2563 			dev_err(dev, "Unable to obtain %u-bit DMA "
2564 				"for consistent allocations, aborting\n", 64);
2565 			goto err_out_release_regions;
2566 		}
2567 		using_dac = 1;
2568 	}
2569 
2570 	/* Map vNIC resources from BAR0-5
2571 	 */
2572 
2573 	for (i = 0; i < ARRAY_SIZE(enic->bar); i++) {
2574 		if (!(pci_resource_flags(pdev, i) & IORESOURCE_MEM))
2575 			continue;
2576 		enic->bar[i].len = pci_resource_len(pdev, i);
2577 		enic->bar[i].vaddr = pci_iomap(pdev, i, enic->bar[i].len);
2578 		if (!enic->bar[i].vaddr) {
2579 			dev_err(dev, "Cannot memory-map BAR %d, aborting\n", i);
2580 			err = -ENODEV;
2581 			goto err_out_iounmap;
2582 		}
2583 		enic->bar[i].bus_addr = pci_resource_start(pdev, i);
2584 	}
2585 
2586 	/* Register vNIC device
2587 	 */
2588 
2589 	enic->vdev = vnic_dev_register(NULL, enic, pdev, enic->bar,
2590 		ARRAY_SIZE(enic->bar));
2591 	if (!enic->vdev) {
2592 		dev_err(dev, "vNIC registration failed, aborting\n");
2593 		err = -ENODEV;
2594 		goto err_out_iounmap;
2595 	}
2596 
2597 	err = vnic_devcmd_init(enic->vdev);
2598 
2599 	if (err)
2600 		goto err_out_vnic_unregister;
2601 
2602 #ifdef CONFIG_PCI_IOV
2603 	/* Get number of subvnics */
2604 	pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_SRIOV);
2605 	if (pos) {
2606 		pci_read_config_word(pdev, pos + PCI_SRIOV_TOTAL_VF,
2607 			&enic->num_vfs);
2608 		if (enic->num_vfs) {
2609 			err = pci_enable_sriov(pdev, enic->num_vfs);
2610 			if (err) {
2611 				dev_err(dev, "SRIOV enable failed, aborting."
2612 					" pci_enable_sriov() returned %d\n",
2613 					err);
2614 				goto err_out_vnic_unregister;
2615 			}
2616 			enic->priv_flags |= ENIC_SRIOV_ENABLED;
2617 			num_pps = enic->num_vfs;
2618 		}
2619 	}
2620 #endif
2621 
2622 	/* Allocate structure for port profiles */
2623 	enic->pp = kcalloc(num_pps, sizeof(*enic->pp), GFP_KERNEL);
2624 	if (!enic->pp) {
2625 		err = -ENOMEM;
2626 		goto err_out_disable_sriov_pp;
2627 	}
2628 
2629 	/* Issue device open to get device in known state
2630 	 */
2631 
2632 	err = enic_dev_open(enic);
2633 	if (err) {
2634 		dev_err(dev, "vNIC dev open failed, aborting\n");
2635 		goto err_out_disable_sriov;
2636 	}
2637 
2638 	/* Setup devcmd lock
2639 	 */
2640 
2641 	spin_lock_init(&enic->devcmd_lock);
2642 	spin_lock_init(&enic->enic_api_lock);
2643 
2644 	/*
2645 	 * Set ingress vlan rewrite mode before vnic initialization
2646 	 */
2647 
2648 	err = enic_dev_set_ig_vlan_rewrite_mode(enic);
2649 	if (err) {
2650 		dev_err(dev,
2651 			"Failed to set ingress vlan rewrite mode, aborting.\n");
2652 		goto err_out_dev_close;
2653 	}
2654 
2655 	/* Issue device init to initialize the vnic-to-switch link.
2656 	 * We'll start with carrier off and wait for link UP
2657 	 * notification later to turn on carrier.  We don't need
2658 	 * to wait here for the vnic-to-switch link initialization
2659 	 * to complete; link UP notification is the indication that
2660 	 * the process is complete.
2661 	 */
2662 
2663 	netif_carrier_off(netdev);
2664 
2665 	/* Do not call dev_init for a dynamic vnic.
2666 	 * For a dynamic vnic, init_prov_info will be
2667 	 * called later by an upper layer.
2668 	 */
2669 
2670 	if (!enic_is_dynamic(enic)) {
2671 		err = vnic_dev_init(enic->vdev, 0);
2672 		if (err) {
2673 			dev_err(dev, "vNIC dev init failed, aborting\n");
2674 			goto err_out_dev_close;
2675 		}
2676 	}
2677 
2678 	err = enic_dev_init(enic);
2679 	if (err) {
2680 		dev_err(dev, "Device initialization failed, aborting\n");
2681 		goto err_out_dev_close;
2682 	}
2683 
2684 	netif_set_real_num_tx_queues(netdev, enic->wq_count);
2685 	netif_set_real_num_rx_queues(netdev, enic->rq_count);
2686 
2687 	/* Setup notification timer, HW reset task, and wq locks
2688 	 */
2689 
2690 	init_timer(&enic->notify_timer);
2691 	enic->notify_timer.function = enic_notify_timer;
2692 	enic->notify_timer.data = (unsigned long)enic;
2693 
2694 	enic_set_rx_coal_setting(enic);
2695 	INIT_WORK(&enic->reset, enic_reset);
2696 	INIT_WORK(&enic->tx_hang_reset, enic_tx_hang_reset);
2697 	INIT_WORK(&enic->change_mtu_work, enic_change_mtu_work);
2698 
2699 	for (i = 0; i < enic->wq_count; i++)
2700 		spin_lock_init(&enic->wq_lock[i]);
2701 
2702 	/* Register net device
2703 	 */
2704 
2705 	enic->port_mtu = enic->config.mtu;
2706 	(void)enic_change_mtu(netdev, enic->port_mtu);
2707 
2708 	err = enic_set_mac_addr(netdev, enic->mac_addr);
2709 	if (err) {
2710 		dev_err(dev, "Invalid MAC address, aborting\n");
2711 		goto err_out_dev_deinit;
2712 	}
2713 
2714 	enic->tx_coalesce_usecs = enic->config.intr_timer_usec;
2715 	/* rx coalesce time already got initialized. This gets used
2716 	 * if adaptive coal is turned off
2717 	 */
2718 	enic->rx_coalesce_usecs = enic->tx_coalesce_usecs;
2719 
2720 	if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic))
2721 		netdev->netdev_ops = &enic_netdev_dynamic_ops;
2722 	else
2723 		netdev->netdev_ops = &enic_netdev_ops;
2724 
2725 	netdev->watchdog_timeo = 2 * HZ;
2726 	enic_set_ethtool_ops(netdev);
2727 
2728 	netdev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
2729 	if (ENIC_SETTING(enic, LOOP)) {
2730 		netdev->features &= ~NETIF_F_HW_VLAN_CTAG_TX;
2731 		enic->loop_enable = 1;
2732 		enic->loop_tag = enic->config.loop_tag;
2733 		dev_info(dev, "loopback tag=0x%04x\n", enic->loop_tag);
2734 	}
2735 	if (ENIC_SETTING(enic, TXCSUM))
2736 		netdev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM;
2737 	if (ENIC_SETTING(enic, TSO))
2738 		netdev->hw_features |= NETIF_F_TSO |
2739 			NETIF_F_TSO6 | NETIF_F_TSO_ECN;
2740 	if (ENIC_SETTING(enic, RSS))
2741 		netdev->hw_features |= NETIF_F_RXHASH;
2742 	if (ENIC_SETTING(enic, RXCSUM))
2743 		netdev->hw_features |= NETIF_F_RXCSUM;
2744 
2745 	netdev->features |= netdev->hw_features;
2746 	netdev->vlan_features |= netdev->features;
2747 
2748 #ifdef CONFIG_RFS_ACCEL
2749 	netdev->hw_features |= NETIF_F_NTUPLE;
2750 #endif
2751 
2752 	if (using_dac)
2753 		netdev->features |= NETIF_F_HIGHDMA;
2754 
2755 	netdev->priv_flags |= IFF_UNICAST_FLT;
2756 
2757 	/* MTU range: 68 - 9000 */
2758 	netdev->min_mtu = ENIC_MIN_MTU;
2759 	netdev->max_mtu = ENIC_MAX_MTU;
2760 
2761 	err = register_netdev(netdev);
2762 	if (err) {
2763 		dev_err(dev, "Cannot register net device, aborting\n");
2764 		goto err_out_dev_deinit;
2765 	}
2766 	enic->rx_copybreak = RX_COPYBREAK_DEFAULT;
2767 
2768 	return 0;
2769 
2770 err_out_dev_deinit:
2771 	enic_dev_deinit(enic);
2772 err_out_dev_close:
2773 	vnic_dev_close(enic->vdev);
2774 err_out_disable_sriov:
2775 	kfree(enic->pp);
2776 err_out_disable_sriov_pp:
2777 #ifdef CONFIG_PCI_IOV
2778 	if (enic_sriov_enabled(enic)) {
2779 		pci_disable_sriov(pdev);
2780 		enic->priv_flags &= ~ENIC_SRIOV_ENABLED;
2781 	}
2782 #endif
2783 err_out_vnic_unregister:
2784 	vnic_dev_unregister(enic->vdev);
2785 err_out_iounmap:
2786 	enic_iounmap(enic);
2787 err_out_release_regions:
2788 	pci_release_regions(pdev);
2789 err_out_disable_device:
2790 	pci_disable_device(pdev);
2791 err_out_free_netdev:
2792 	free_netdev(netdev);
2793 
2794 	return err;
2795 }
2796 
2797 static void enic_remove(struct pci_dev *pdev)
2798 {
2799 	struct net_device *netdev = pci_get_drvdata(pdev);
2800 
2801 	if (netdev) {
2802 		struct enic *enic = netdev_priv(netdev);
2803 
2804 		cancel_work_sync(&enic->reset);
2805 		cancel_work_sync(&enic->change_mtu_work);
2806 		unregister_netdev(netdev);
2807 		enic_dev_deinit(enic);
2808 		vnic_dev_close(enic->vdev);
2809 #ifdef CONFIG_PCI_IOV
2810 		if (enic_sriov_enabled(enic)) {
2811 			pci_disable_sriov(pdev);
2812 			enic->priv_flags &= ~ENIC_SRIOV_ENABLED;
2813 		}
2814 #endif
2815 		kfree(enic->pp);
2816 		vnic_dev_unregister(enic->vdev);
2817 		enic_iounmap(enic);
2818 		pci_release_regions(pdev);
2819 		pci_disable_device(pdev);
2820 		free_netdev(netdev);
2821 	}
2822 }
2823 
2824 static struct pci_driver enic_driver = {
2825 	.name = DRV_NAME,
2826 	.id_table = enic_id_table,
2827 	.probe = enic_probe,
2828 	.remove = enic_remove,
2829 };
2830 
2831 static int __init enic_init_module(void)
2832 {
2833 	pr_info("%s, ver %s\n", DRV_DESCRIPTION, DRV_VERSION);
2834 
2835 	return pci_register_driver(&enic_driver);
2836 }
2837 
2838 static void __exit enic_cleanup_module(void)
2839 {
2840 	pci_unregister_driver(&enic_driver);
2841 }
2842 
2843 module_init(enic_init_module);
2844 module_exit(enic_cleanup_module);
2845