1 /*
2  * Copyright 2008-2010 Cisco Systems, Inc.  All rights reserved.
3  * Copyright 2007 Nuova Systems, Inc.  All rights reserved.
4  *
5  * This program is free software; you may redistribute it and/or modify
6  * it under the terms of the GNU General Public License as published by
7  * the Free Software Foundation; version 2 of the License.
8  *
9  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
10  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
11  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
12  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
13  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
14  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
15  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
16  * SOFTWARE.
17  *
18  */
19 
20 #include <linux/module.h>
21 #include <linux/kernel.h>
22 #include <linux/string.h>
23 #include <linux/errno.h>
24 #include <linux/types.h>
25 #include <linux/init.h>
26 #include <linux/interrupt.h>
27 #include <linux/workqueue.h>
28 #include <linux/pci.h>
29 #include <linux/netdevice.h>
30 #include <linux/etherdevice.h>
31 #include <linux/if.h>
32 #include <linux/if_ether.h>
33 #include <linux/if_vlan.h>
34 #include <linux/in.h>
35 #include <linux/ip.h>
36 #include <linux/ipv6.h>
37 #include <linux/tcp.h>
38 #include <linux/rtnetlink.h>
39 #include <linux/prefetch.h>
40 #include <net/ip6_checksum.h>
41 #include <linux/ktime.h>
42 #ifdef CONFIG_RFS_ACCEL
43 #include <linux/cpu_rmap.h>
44 #endif
45 #ifdef CONFIG_NET_RX_BUSY_POLL
46 #include <net/busy_poll.h>
47 #endif
48 #include <linux/crash_dump.h>
49 
50 #include "cq_enet_desc.h"
51 #include "vnic_dev.h"
52 #include "vnic_intr.h"
53 #include "vnic_stats.h"
54 #include "vnic_vic.h"
55 #include "enic_res.h"
56 #include "enic.h"
57 #include "enic_dev.h"
58 #include "enic_pp.h"
59 #include "enic_clsf.h"
60 
61 #define ENIC_NOTIFY_TIMER_PERIOD	(2 * HZ)
62 #define WQ_ENET_MAX_DESC_LEN		(1 << WQ_ENET_LEN_BITS)
63 #define MAX_TSO				(1 << 16)
64 #define ENIC_DESC_MAX_SPLITS		(MAX_TSO / WQ_ENET_MAX_DESC_LEN + 1)
65 
66 #define PCI_DEVICE_ID_CISCO_VIC_ENET         0x0043  /* ethernet vnic */
67 #define PCI_DEVICE_ID_CISCO_VIC_ENET_DYN     0x0044  /* enet dynamic vnic */
68 #define PCI_DEVICE_ID_CISCO_VIC_ENET_VF      0x0071  /* enet SRIOV VF */
69 
70 #define RX_COPYBREAK_DEFAULT		256
71 
72 /* Supported devices */
73 static const struct pci_device_id enic_id_table[] = {
74 	{ PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET) },
75 	{ PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET_DYN) },
76 	{ PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET_VF) },
77 	{ 0, }	/* end of table */
78 };
79 
80 MODULE_DESCRIPTION(DRV_DESCRIPTION);
81 MODULE_AUTHOR("Scott Feldman <scofeldm@cisco.com>");
82 MODULE_LICENSE("GPL");
83 MODULE_VERSION(DRV_VERSION);
84 MODULE_DEVICE_TABLE(pci, enic_id_table);
85 
86 #define ENIC_LARGE_PKT_THRESHOLD		1000
87 #define ENIC_MAX_COALESCE_TIMERS		10
88 /*  Interrupt moderation table, which will be used to decide the
89  *  coalescing timer values
90  *  {rx_rate in Mbps, mapping percentage of the range}
91  */
92 static struct enic_intr_mod_table mod_table[ENIC_MAX_COALESCE_TIMERS + 1] = {
93 	{4000,  0},
94 	{4400, 10},
95 	{5060, 20},
96 	{5230, 30},
97 	{5540, 40},
98 	{5820, 50},
99 	{6120, 60},
100 	{6435, 70},
101 	{6745, 80},
102 	{7000, 90},
103 	{0xFFFFFFFF, 100}
104 };
105 
106 /* This table helps the driver to pick different ranges for rx coalescing
107  * timer depending on the link speed.
108  */
109 static struct enic_intr_mod_range mod_range[ENIC_MAX_LINK_SPEEDS] = {
110 	{0,  0}, /* 0  - 4  Gbps */
111 	{0,  3}, /* 4  - 10 Gbps */
112 	{3,  6}, /* 10 - 40 Gbps */
113 };
114 
115 int enic_is_dynamic(struct enic *enic)
116 {
117 	return enic->pdev->device == PCI_DEVICE_ID_CISCO_VIC_ENET_DYN;
118 }
119 
120 int enic_sriov_enabled(struct enic *enic)
121 {
122 	return (enic->priv_flags & ENIC_SRIOV_ENABLED) ? 1 : 0;
123 }
124 
125 static int enic_is_sriov_vf(struct enic *enic)
126 {
127 	return enic->pdev->device == PCI_DEVICE_ID_CISCO_VIC_ENET_VF;
128 }
129 
130 int enic_is_valid_vf(struct enic *enic, int vf)
131 {
132 #ifdef CONFIG_PCI_IOV
133 	return vf >= 0 && vf < enic->num_vfs;
134 #else
135 	return 0;
136 #endif
137 }
138 
139 static void enic_free_wq_buf(struct vnic_wq *wq, struct vnic_wq_buf *buf)
140 {
141 	struct enic *enic = vnic_dev_priv(wq->vdev);
142 
143 	if (buf->sop)
144 		pci_unmap_single(enic->pdev, buf->dma_addr,
145 			buf->len, PCI_DMA_TODEVICE);
146 	else
147 		pci_unmap_page(enic->pdev, buf->dma_addr,
148 			buf->len, PCI_DMA_TODEVICE);
149 
150 	if (buf->os_buf)
151 		dev_kfree_skb_any(buf->os_buf);
152 }
153 
154 static void enic_wq_free_buf(struct vnic_wq *wq,
155 	struct cq_desc *cq_desc, struct vnic_wq_buf *buf, void *opaque)
156 {
157 	enic_free_wq_buf(wq, buf);
158 }
159 
160 static int enic_wq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
161 	u8 type, u16 q_number, u16 completed_index, void *opaque)
162 {
163 	struct enic *enic = vnic_dev_priv(vdev);
164 
165 	spin_lock(&enic->wq_lock[q_number]);
166 
167 	vnic_wq_service(&enic->wq[q_number], cq_desc,
168 		completed_index, enic_wq_free_buf,
169 		opaque);
170 
171 	if (netif_tx_queue_stopped(netdev_get_tx_queue(enic->netdev, q_number)) &&
172 	    vnic_wq_desc_avail(&enic->wq[q_number]) >=
173 	    (MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS))
174 		netif_wake_subqueue(enic->netdev, q_number);
175 
176 	spin_unlock(&enic->wq_lock[q_number]);
177 
178 	return 0;
179 }
180 
181 static void enic_log_q_error(struct enic *enic)
182 {
183 	unsigned int i;
184 	u32 error_status;
185 
186 	for (i = 0; i < enic->wq_count; i++) {
187 		error_status = vnic_wq_error_status(&enic->wq[i]);
188 		if (error_status)
189 			netdev_err(enic->netdev, "WQ[%d] error_status %d\n",
190 				i, error_status);
191 	}
192 
193 	for (i = 0; i < enic->rq_count; i++) {
194 		error_status = vnic_rq_error_status(&enic->rq[i]);
195 		if (error_status)
196 			netdev_err(enic->netdev, "RQ[%d] error_status %d\n",
197 				i, error_status);
198 	}
199 }
200 
201 static void enic_msglvl_check(struct enic *enic)
202 {
203 	u32 msg_enable = vnic_dev_msg_lvl(enic->vdev);
204 
205 	if (msg_enable != enic->msg_enable) {
206 		netdev_info(enic->netdev, "msg lvl changed from 0x%x to 0x%x\n",
207 			enic->msg_enable, msg_enable);
208 		enic->msg_enable = msg_enable;
209 	}
210 }
211 
212 static void enic_mtu_check(struct enic *enic)
213 {
214 	u32 mtu = vnic_dev_mtu(enic->vdev);
215 	struct net_device *netdev = enic->netdev;
216 
217 	if (mtu && mtu != enic->port_mtu) {
218 		enic->port_mtu = mtu;
219 		if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic)) {
220 			mtu = max_t(int, ENIC_MIN_MTU,
221 				min_t(int, ENIC_MAX_MTU, mtu));
222 			if (mtu != netdev->mtu)
223 				schedule_work(&enic->change_mtu_work);
224 		} else {
225 			if (mtu < netdev->mtu)
226 				netdev_warn(netdev,
227 					"interface MTU (%d) set higher "
228 					"than switch port MTU (%d)\n",
229 					netdev->mtu, mtu);
230 		}
231 	}
232 }
233 
234 static void enic_link_check(struct enic *enic)
235 {
236 	int link_status = vnic_dev_link_status(enic->vdev);
237 	int carrier_ok = netif_carrier_ok(enic->netdev);
238 
239 	if (link_status && !carrier_ok) {
240 		netdev_info(enic->netdev, "Link UP\n");
241 		netif_carrier_on(enic->netdev);
242 	} else if (!link_status && carrier_ok) {
243 		netdev_info(enic->netdev, "Link DOWN\n");
244 		netif_carrier_off(enic->netdev);
245 	}
246 }
247 
248 static void enic_notify_check(struct enic *enic)
249 {
250 	enic_msglvl_check(enic);
251 	enic_mtu_check(enic);
252 	enic_link_check(enic);
253 }
254 
255 #define ENIC_TEST_INTR(pba, i) (pba & (1 << i))
256 
257 static irqreturn_t enic_isr_legacy(int irq, void *data)
258 {
259 	struct net_device *netdev = data;
260 	struct enic *enic = netdev_priv(netdev);
261 	unsigned int io_intr = enic_legacy_io_intr();
262 	unsigned int err_intr = enic_legacy_err_intr();
263 	unsigned int notify_intr = enic_legacy_notify_intr();
264 	u32 pba;
265 
266 	vnic_intr_mask(&enic->intr[io_intr]);
267 
268 	pba = vnic_intr_legacy_pba(enic->legacy_pba);
269 	if (!pba) {
270 		vnic_intr_unmask(&enic->intr[io_intr]);
271 		return IRQ_NONE;	/* not our interrupt */
272 	}
273 
274 	if (ENIC_TEST_INTR(pba, notify_intr)) {
275 		enic_notify_check(enic);
276 		vnic_intr_return_all_credits(&enic->intr[notify_intr]);
277 	}
278 
279 	if (ENIC_TEST_INTR(pba, err_intr)) {
280 		vnic_intr_return_all_credits(&enic->intr[err_intr]);
281 		enic_log_q_error(enic);
282 		/* schedule recovery from WQ/RQ error */
283 		schedule_work(&enic->reset);
284 		return IRQ_HANDLED;
285 	}
286 
287 	if (ENIC_TEST_INTR(pba, io_intr))
288 		napi_schedule_irqoff(&enic->napi[0]);
289 	else
290 		vnic_intr_unmask(&enic->intr[io_intr]);
291 
292 	return IRQ_HANDLED;
293 }
294 
295 static irqreturn_t enic_isr_msi(int irq, void *data)
296 {
297 	struct enic *enic = data;
298 
299 	/* With MSI, there is no sharing of interrupts, so this is
300 	 * our interrupt and there is no need to ack it.  The device
301 	 * is not providing per-vector masking, so the OS will not
302 	 * write to PCI config space to mask/unmask the interrupt.
303 	 * We're using mask_on_assertion for MSI, so the device
304 	 * automatically masks the interrupt when the interrupt is
305 	 * generated.  Later, when exiting polling, the interrupt
306 	 * will be unmasked (see enic_poll).
307 	 *
308 	 * Also, the device uses the same PCIe Traffic Class (TC)
309 	 * for Memory Write data and MSI, so there are no ordering
310 	 * issues; the MSI will always arrive at the Root Complex
311 	 * _after_ corresponding Memory Writes (i.e. descriptor
312 	 * writes).
313 	 */
314 
315 	napi_schedule_irqoff(&enic->napi[0]);
316 
317 	return IRQ_HANDLED;
318 }
319 
320 static irqreturn_t enic_isr_msix(int irq, void *data)
321 {
322 	struct napi_struct *napi = data;
323 
324 	napi_schedule_irqoff(napi);
325 
326 	return IRQ_HANDLED;
327 }
328 
329 static irqreturn_t enic_isr_msix_err(int irq, void *data)
330 {
331 	struct enic *enic = data;
332 	unsigned int intr = enic_msix_err_intr(enic);
333 
334 	vnic_intr_return_all_credits(&enic->intr[intr]);
335 
336 	enic_log_q_error(enic);
337 
338 	/* schedule recovery from WQ/RQ error */
339 	schedule_work(&enic->reset);
340 
341 	return IRQ_HANDLED;
342 }
343 
344 static irqreturn_t enic_isr_msix_notify(int irq, void *data)
345 {
346 	struct enic *enic = data;
347 	unsigned int intr = enic_msix_notify_intr(enic);
348 
349 	enic_notify_check(enic);
350 	vnic_intr_return_all_credits(&enic->intr[intr]);
351 
352 	return IRQ_HANDLED;
353 }
354 
355 static int enic_queue_wq_skb_cont(struct enic *enic, struct vnic_wq *wq,
356 				  struct sk_buff *skb, unsigned int len_left,
357 				  int loopback)
358 {
359 	const skb_frag_t *frag;
360 	dma_addr_t dma_addr;
361 
362 	/* Queue additional data fragments */
363 	for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
364 		len_left -= skb_frag_size(frag);
365 		dma_addr = skb_frag_dma_map(&enic->pdev->dev, frag, 0,
366 					    skb_frag_size(frag),
367 					    DMA_TO_DEVICE);
368 		if (unlikely(enic_dma_map_check(enic, dma_addr)))
369 			return -ENOMEM;
370 		enic_queue_wq_desc_cont(wq, skb, dma_addr, skb_frag_size(frag),
371 					(len_left == 0),	/* EOP? */
372 					loopback);
373 	}
374 
375 	return 0;
376 }
377 
378 static int enic_queue_wq_skb_vlan(struct enic *enic, struct vnic_wq *wq,
379 				  struct sk_buff *skb, int vlan_tag_insert,
380 				  unsigned int vlan_tag, int loopback)
381 {
382 	unsigned int head_len = skb_headlen(skb);
383 	unsigned int len_left = skb->len - head_len;
384 	int eop = (len_left == 0);
385 	dma_addr_t dma_addr;
386 	int err = 0;
387 
388 	dma_addr = pci_map_single(enic->pdev, skb->data, head_len,
389 				  PCI_DMA_TODEVICE);
390 	if (unlikely(enic_dma_map_check(enic, dma_addr)))
391 		return -ENOMEM;
392 
393 	/* Queue the main skb fragment. The fragments are no larger
394 	 * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less
395 	 * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor
396 	 * per fragment is queued.
397 	 */
398 	enic_queue_wq_desc(wq, skb, dma_addr, head_len,	vlan_tag_insert,
399 			   vlan_tag, eop, loopback);
400 
401 	if (!eop)
402 		err = enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback);
403 
404 	return err;
405 }
406 
407 static int enic_queue_wq_skb_csum_l4(struct enic *enic, struct vnic_wq *wq,
408 				     struct sk_buff *skb, int vlan_tag_insert,
409 				     unsigned int vlan_tag, int loopback)
410 {
411 	unsigned int head_len = skb_headlen(skb);
412 	unsigned int len_left = skb->len - head_len;
413 	unsigned int hdr_len = skb_checksum_start_offset(skb);
414 	unsigned int csum_offset = hdr_len + skb->csum_offset;
415 	int eop = (len_left == 0);
416 	dma_addr_t dma_addr;
417 	int err = 0;
418 
419 	dma_addr = pci_map_single(enic->pdev, skb->data, head_len,
420 				  PCI_DMA_TODEVICE);
421 	if (unlikely(enic_dma_map_check(enic, dma_addr)))
422 		return -ENOMEM;
423 
424 	/* Queue the main skb fragment. The fragments are no larger
425 	 * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less
426 	 * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor
427 	 * per fragment is queued.
428 	 */
429 	enic_queue_wq_desc_csum_l4(wq, skb, dma_addr, head_len,	csum_offset,
430 				   hdr_len, vlan_tag_insert, vlan_tag, eop,
431 				   loopback);
432 
433 	if (!eop)
434 		err = enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback);
435 
436 	return err;
437 }
438 
439 static int enic_queue_wq_skb_tso(struct enic *enic, struct vnic_wq *wq,
440 				 struct sk_buff *skb, unsigned int mss,
441 				 int vlan_tag_insert, unsigned int vlan_tag,
442 				 int loopback)
443 {
444 	unsigned int frag_len_left = skb_headlen(skb);
445 	unsigned int len_left = skb->len - frag_len_left;
446 	unsigned int hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
447 	int eop = (len_left == 0);
448 	unsigned int len;
449 	dma_addr_t dma_addr;
450 	unsigned int offset = 0;
451 	skb_frag_t *frag;
452 
453 	/* Preload TCP csum field with IP pseudo hdr calculated
454 	 * with IP length set to zero.  HW will later add in length
455 	 * to each TCP segment resulting from the TSO.
456 	 */
457 
458 	if (skb->protocol == cpu_to_be16(ETH_P_IP)) {
459 		ip_hdr(skb)->check = 0;
460 		tcp_hdr(skb)->check = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
461 			ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
462 	} else if (skb->protocol == cpu_to_be16(ETH_P_IPV6)) {
463 		tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
464 			&ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
465 	}
466 
467 	/* Queue WQ_ENET_MAX_DESC_LEN length descriptors
468 	 * for the main skb fragment
469 	 */
470 	while (frag_len_left) {
471 		len = min(frag_len_left, (unsigned int)WQ_ENET_MAX_DESC_LEN);
472 		dma_addr = pci_map_single(enic->pdev, skb->data + offset, len,
473 					  PCI_DMA_TODEVICE);
474 		if (unlikely(enic_dma_map_check(enic, dma_addr)))
475 			return -ENOMEM;
476 		enic_queue_wq_desc_tso(wq, skb, dma_addr, len, mss, hdr_len,
477 				       vlan_tag_insert, vlan_tag,
478 				       eop && (len == frag_len_left), loopback);
479 		frag_len_left -= len;
480 		offset += len;
481 	}
482 
483 	if (eop)
484 		return 0;
485 
486 	/* Queue WQ_ENET_MAX_DESC_LEN length descriptors
487 	 * for additional data fragments
488 	 */
489 	for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
490 		len_left -= skb_frag_size(frag);
491 		frag_len_left = skb_frag_size(frag);
492 		offset = 0;
493 
494 		while (frag_len_left) {
495 			len = min(frag_len_left,
496 				(unsigned int)WQ_ENET_MAX_DESC_LEN);
497 			dma_addr = skb_frag_dma_map(&enic->pdev->dev, frag,
498 						    offset, len,
499 						    DMA_TO_DEVICE);
500 			if (unlikely(enic_dma_map_check(enic, dma_addr)))
501 				return -ENOMEM;
502 			enic_queue_wq_desc_cont(wq, skb, dma_addr, len,
503 						(len_left == 0) &&
504 						 (len == frag_len_left),/*EOP*/
505 						loopback);
506 			frag_len_left -= len;
507 			offset += len;
508 		}
509 	}
510 
511 	return 0;
512 }
513 
514 static inline void enic_queue_wq_skb(struct enic *enic,
515 	struct vnic_wq *wq, struct sk_buff *skb)
516 {
517 	unsigned int mss = skb_shinfo(skb)->gso_size;
518 	unsigned int vlan_tag = 0;
519 	int vlan_tag_insert = 0;
520 	int loopback = 0;
521 	int err;
522 
523 	if (skb_vlan_tag_present(skb)) {
524 		/* VLAN tag from trunking driver */
525 		vlan_tag_insert = 1;
526 		vlan_tag = skb_vlan_tag_get(skb);
527 	} else if (enic->loop_enable) {
528 		vlan_tag = enic->loop_tag;
529 		loopback = 1;
530 	}
531 
532 	if (mss)
533 		err = enic_queue_wq_skb_tso(enic, wq, skb, mss,
534 					    vlan_tag_insert, vlan_tag,
535 					    loopback);
536 	else if	(skb->ip_summed == CHECKSUM_PARTIAL)
537 		err = enic_queue_wq_skb_csum_l4(enic, wq, skb, vlan_tag_insert,
538 						vlan_tag, loopback);
539 	else
540 		err = enic_queue_wq_skb_vlan(enic, wq, skb, vlan_tag_insert,
541 					     vlan_tag, loopback);
542 	if (unlikely(err)) {
543 		struct vnic_wq_buf *buf;
544 
545 		buf = wq->to_use->prev;
546 		/* while not EOP of previous pkt && queue not empty.
547 		 * For all non EOP bufs, os_buf is NULL.
548 		 */
549 		while (!buf->os_buf && (buf->next != wq->to_clean)) {
550 			enic_free_wq_buf(wq, buf);
551 			wq->ring.desc_avail++;
552 			buf = buf->prev;
553 		}
554 		wq->to_use = buf->next;
555 		dev_kfree_skb(skb);
556 	}
557 }
558 
559 /* netif_tx_lock held, process context with BHs disabled, or BH */
560 static netdev_tx_t enic_hard_start_xmit(struct sk_buff *skb,
561 	struct net_device *netdev)
562 {
563 	struct enic *enic = netdev_priv(netdev);
564 	struct vnic_wq *wq;
565 	unsigned int txq_map;
566 	struct netdev_queue *txq;
567 
568 	if (skb->len <= 0) {
569 		dev_kfree_skb_any(skb);
570 		return NETDEV_TX_OK;
571 	}
572 
573 	txq_map = skb_get_queue_mapping(skb) % enic->wq_count;
574 	wq = &enic->wq[txq_map];
575 	txq = netdev_get_tx_queue(netdev, txq_map);
576 
577 	/* Non-TSO sends must fit within ENIC_NON_TSO_MAX_DESC descs,
578 	 * which is very likely.  In the off chance it's going to take
579 	 * more than * ENIC_NON_TSO_MAX_DESC, linearize the skb.
580 	 */
581 
582 	if (skb_shinfo(skb)->gso_size == 0 &&
583 	    skb_shinfo(skb)->nr_frags + 1 > ENIC_NON_TSO_MAX_DESC &&
584 	    skb_linearize(skb)) {
585 		dev_kfree_skb_any(skb);
586 		return NETDEV_TX_OK;
587 	}
588 
589 	spin_lock(&enic->wq_lock[txq_map]);
590 
591 	if (vnic_wq_desc_avail(wq) <
592 	    skb_shinfo(skb)->nr_frags + ENIC_DESC_MAX_SPLITS) {
593 		netif_tx_stop_queue(txq);
594 		/* This is a hard error, log it */
595 		netdev_err(netdev, "BUG! Tx ring full when queue awake!\n");
596 		spin_unlock(&enic->wq_lock[txq_map]);
597 		return NETDEV_TX_BUSY;
598 	}
599 
600 	enic_queue_wq_skb(enic, wq, skb);
601 
602 	if (vnic_wq_desc_avail(wq) < MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS)
603 		netif_tx_stop_queue(txq);
604 	if (!skb->xmit_more || netif_xmit_stopped(txq))
605 		vnic_wq_doorbell(wq);
606 
607 	spin_unlock(&enic->wq_lock[txq_map]);
608 
609 	return NETDEV_TX_OK;
610 }
611 
612 /* dev_base_lock rwlock held, nominally process context */
613 static struct rtnl_link_stats64 *enic_get_stats(struct net_device *netdev,
614 						struct rtnl_link_stats64 *net_stats)
615 {
616 	struct enic *enic = netdev_priv(netdev);
617 	struct vnic_stats *stats;
618 	int err;
619 
620 	err = enic_dev_stats_dump(enic, &stats);
621 	/* return only when pci_zalloc_consistent fails in vnic_dev_stats_dump
622 	 * For other failures, like devcmd failure, we return previously
623 	 * recorded stats.
624 	 */
625 	if (err == -ENOMEM)
626 		return net_stats;
627 
628 	net_stats->tx_packets = stats->tx.tx_frames_ok;
629 	net_stats->tx_bytes = stats->tx.tx_bytes_ok;
630 	net_stats->tx_errors = stats->tx.tx_errors;
631 	net_stats->tx_dropped = stats->tx.tx_drops;
632 
633 	net_stats->rx_packets = stats->rx.rx_frames_ok;
634 	net_stats->rx_bytes = stats->rx.rx_bytes_ok;
635 	net_stats->rx_errors = stats->rx.rx_errors;
636 	net_stats->multicast = stats->rx.rx_multicast_frames_ok;
637 	net_stats->rx_over_errors = enic->rq_truncated_pkts;
638 	net_stats->rx_crc_errors = enic->rq_bad_fcs;
639 	net_stats->rx_dropped = stats->rx.rx_no_bufs + stats->rx.rx_drop;
640 
641 	return net_stats;
642 }
643 
644 static int enic_mc_sync(struct net_device *netdev, const u8 *mc_addr)
645 {
646 	struct enic *enic = netdev_priv(netdev);
647 
648 	if (enic->mc_count == ENIC_MULTICAST_PERFECT_FILTERS) {
649 		unsigned int mc_count = netdev_mc_count(netdev);
650 
651 		netdev_warn(netdev, "Registering only %d out of %d multicast addresses\n",
652 			    ENIC_MULTICAST_PERFECT_FILTERS, mc_count);
653 
654 		return -ENOSPC;
655 	}
656 
657 	enic_dev_add_addr(enic, mc_addr);
658 	enic->mc_count++;
659 
660 	return 0;
661 }
662 
663 static int enic_mc_unsync(struct net_device *netdev, const u8 *mc_addr)
664 {
665 	struct enic *enic = netdev_priv(netdev);
666 
667 	enic_dev_del_addr(enic, mc_addr);
668 	enic->mc_count--;
669 
670 	return 0;
671 }
672 
673 static int enic_uc_sync(struct net_device *netdev, const u8 *uc_addr)
674 {
675 	struct enic *enic = netdev_priv(netdev);
676 
677 	if (enic->uc_count == ENIC_UNICAST_PERFECT_FILTERS) {
678 		unsigned int uc_count = netdev_uc_count(netdev);
679 
680 		netdev_warn(netdev, "Registering only %d out of %d unicast addresses\n",
681 			    ENIC_UNICAST_PERFECT_FILTERS, uc_count);
682 
683 		return -ENOSPC;
684 	}
685 
686 	enic_dev_add_addr(enic, uc_addr);
687 	enic->uc_count++;
688 
689 	return 0;
690 }
691 
692 static int enic_uc_unsync(struct net_device *netdev, const u8 *uc_addr)
693 {
694 	struct enic *enic = netdev_priv(netdev);
695 
696 	enic_dev_del_addr(enic, uc_addr);
697 	enic->uc_count--;
698 
699 	return 0;
700 }
701 
702 void enic_reset_addr_lists(struct enic *enic)
703 {
704 	struct net_device *netdev = enic->netdev;
705 
706 	__dev_uc_unsync(netdev, NULL);
707 	__dev_mc_unsync(netdev, NULL);
708 
709 	enic->mc_count = 0;
710 	enic->uc_count = 0;
711 	enic->flags = 0;
712 }
713 
714 static int enic_set_mac_addr(struct net_device *netdev, char *addr)
715 {
716 	struct enic *enic = netdev_priv(netdev);
717 
718 	if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic)) {
719 		if (!is_valid_ether_addr(addr) && !is_zero_ether_addr(addr))
720 			return -EADDRNOTAVAIL;
721 	} else {
722 		if (!is_valid_ether_addr(addr))
723 			return -EADDRNOTAVAIL;
724 	}
725 
726 	memcpy(netdev->dev_addr, addr, netdev->addr_len);
727 
728 	return 0;
729 }
730 
731 static int enic_set_mac_address_dynamic(struct net_device *netdev, void *p)
732 {
733 	struct enic *enic = netdev_priv(netdev);
734 	struct sockaddr *saddr = p;
735 	char *addr = saddr->sa_data;
736 	int err;
737 
738 	if (netif_running(enic->netdev)) {
739 		err = enic_dev_del_station_addr(enic);
740 		if (err)
741 			return err;
742 	}
743 
744 	err = enic_set_mac_addr(netdev, addr);
745 	if (err)
746 		return err;
747 
748 	if (netif_running(enic->netdev)) {
749 		err = enic_dev_add_station_addr(enic);
750 		if (err)
751 			return err;
752 	}
753 
754 	return err;
755 }
756 
757 static int enic_set_mac_address(struct net_device *netdev, void *p)
758 {
759 	struct sockaddr *saddr = p;
760 	char *addr = saddr->sa_data;
761 	struct enic *enic = netdev_priv(netdev);
762 	int err;
763 
764 	err = enic_dev_del_station_addr(enic);
765 	if (err)
766 		return err;
767 
768 	err = enic_set_mac_addr(netdev, addr);
769 	if (err)
770 		return err;
771 
772 	return enic_dev_add_station_addr(enic);
773 }
774 
775 /* netif_tx_lock held, BHs disabled */
776 static void enic_set_rx_mode(struct net_device *netdev)
777 {
778 	struct enic *enic = netdev_priv(netdev);
779 	int directed = 1;
780 	int multicast = (netdev->flags & IFF_MULTICAST) ? 1 : 0;
781 	int broadcast = (netdev->flags & IFF_BROADCAST) ? 1 : 0;
782 	int promisc = (netdev->flags & IFF_PROMISC) ||
783 		netdev_uc_count(netdev) > ENIC_UNICAST_PERFECT_FILTERS;
784 	int allmulti = (netdev->flags & IFF_ALLMULTI) ||
785 		netdev_mc_count(netdev) > ENIC_MULTICAST_PERFECT_FILTERS;
786 	unsigned int flags = netdev->flags |
787 		(allmulti ? IFF_ALLMULTI : 0) |
788 		(promisc ? IFF_PROMISC : 0);
789 
790 	if (enic->flags != flags) {
791 		enic->flags = flags;
792 		enic_dev_packet_filter(enic, directed,
793 			multicast, broadcast, promisc, allmulti);
794 	}
795 
796 	if (!promisc) {
797 		__dev_uc_sync(netdev, enic_uc_sync, enic_uc_unsync);
798 		if (!allmulti)
799 			__dev_mc_sync(netdev, enic_mc_sync, enic_mc_unsync);
800 	}
801 }
802 
803 /* netif_tx_lock held, BHs disabled */
804 static void enic_tx_timeout(struct net_device *netdev)
805 {
806 	struct enic *enic = netdev_priv(netdev);
807 	schedule_work(&enic->reset);
808 }
809 
810 static int enic_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
811 {
812 	struct enic *enic = netdev_priv(netdev);
813 	struct enic_port_profile *pp;
814 	int err;
815 
816 	ENIC_PP_BY_INDEX(enic, vf, pp, &err);
817 	if (err)
818 		return err;
819 
820 	if (is_valid_ether_addr(mac) || is_zero_ether_addr(mac)) {
821 		if (vf == PORT_SELF_VF) {
822 			memcpy(pp->vf_mac, mac, ETH_ALEN);
823 			return 0;
824 		} else {
825 			/*
826 			 * For sriov vf's set the mac in hw
827 			 */
828 			ENIC_DEVCMD_PROXY_BY_INDEX(vf, err, enic,
829 				vnic_dev_set_mac_addr, mac);
830 			return enic_dev_status_to_errno(err);
831 		}
832 	} else
833 		return -EINVAL;
834 }
835 
836 static int enic_set_vf_port(struct net_device *netdev, int vf,
837 	struct nlattr *port[])
838 {
839 	struct enic *enic = netdev_priv(netdev);
840 	struct enic_port_profile prev_pp;
841 	struct enic_port_profile *pp;
842 	int err = 0, restore_pp = 1;
843 
844 	ENIC_PP_BY_INDEX(enic, vf, pp, &err);
845 	if (err)
846 		return err;
847 
848 	if (!port[IFLA_PORT_REQUEST])
849 		return -EOPNOTSUPP;
850 
851 	memcpy(&prev_pp, pp, sizeof(*enic->pp));
852 	memset(pp, 0, sizeof(*enic->pp));
853 
854 	pp->set |= ENIC_SET_REQUEST;
855 	pp->request = nla_get_u8(port[IFLA_PORT_REQUEST]);
856 
857 	if (port[IFLA_PORT_PROFILE]) {
858 		pp->set |= ENIC_SET_NAME;
859 		memcpy(pp->name, nla_data(port[IFLA_PORT_PROFILE]),
860 			PORT_PROFILE_MAX);
861 	}
862 
863 	if (port[IFLA_PORT_INSTANCE_UUID]) {
864 		pp->set |= ENIC_SET_INSTANCE;
865 		memcpy(pp->instance_uuid,
866 			nla_data(port[IFLA_PORT_INSTANCE_UUID]), PORT_UUID_MAX);
867 	}
868 
869 	if (port[IFLA_PORT_HOST_UUID]) {
870 		pp->set |= ENIC_SET_HOST;
871 		memcpy(pp->host_uuid,
872 			nla_data(port[IFLA_PORT_HOST_UUID]), PORT_UUID_MAX);
873 	}
874 
875 	if (vf == PORT_SELF_VF) {
876 		/* Special case handling: mac came from IFLA_VF_MAC */
877 		if (!is_zero_ether_addr(prev_pp.vf_mac))
878 			memcpy(pp->mac_addr, prev_pp.vf_mac, ETH_ALEN);
879 
880 		if (is_zero_ether_addr(netdev->dev_addr))
881 			eth_hw_addr_random(netdev);
882 	} else {
883 		/* SR-IOV VF: get mac from adapter */
884 		ENIC_DEVCMD_PROXY_BY_INDEX(vf, err, enic,
885 			vnic_dev_get_mac_addr, pp->mac_addr);
886 		if (err) {
887 			netdev_err(netdev, "Error getting mac for vf %d\n", vf);
888 			memcpy(pp, &prev_pp, sizeof(*pp));
889 			return enic_dev_status_to_errno(err);
890 		}
891 	}
892 
893 	err = enic_process_set_pp_request(enic, vf, &prev_pp, &restore_pp);
894 	if (err) {
895 		if (restore_pp) {
896 			/* Things are still the way they were: Implicit
897 			 * DISASSOCIATE failed
898 			 */
899 			memcpy(pp, &prev_pp, sizeof(*pp));
900 		} else {
901 			memset(pp, 0, sizeof(*pp));
902 			if (vf == PORT_SELF_VF)
903 				eth_zero_addr(netdev->dev_addr);
904 		}
905 	} else {
906 		/* Set flag to indicate that the port assoc/disassoc
907 		 * request has been sent out to fw
908 		 */
909 		pp->set |= ENIC_PORT_REQUEST_APPLIED;
910 
911 		/* If DISASSOCIATE, clean up all assigned/saved macaddresses */
912 		if (pp->request == PORT_REQUEST_DISASSOCIATE) {
913 			eth_zero_addr(pp->mac_addr);
914 			if (vf == PORT_SELF_VF)
915 				eth_zero_addr(netdev->dev_addr);
916 		}
917 	}
918 
919 	if (vf == PORT_SELF_VF)
920 		eth_zero_addr(pp->vf_mac);
921 
922 	return err;
923 }
924 
925 static int enic_get_vf_port(struct net_device *netdev, int vf,
926 	struct sk_buff *skb)
927 {
928 	struct enic *enic = netdev_priv(netdev);
929 	u16 response = PORT_PROFILE_RESPONSE_SUCCESS;
930 	struct enic_port_profile *pp;
931 	int err;
932 
933 	ENIC_PP_BY_INDEX(enic, vf, pp, &err);
934 	if (err)
935 		return err;
936 
937 	if (!(pp->set & ENIC_PORT_REQUEST_APPLIED))
938 		return -ENODATA;
939 
940 	err = enic_process_get_pp_request(enic, vf, pp->request, &response);
941 	if (err)
942 		return err;
943 
944 	if (nla_put_u16(skb, IFLA_PORT_REQUEST, pp->request) ||
945 	    nla_put_u16(skb, IFLA_PORT_RESPONSE, response) ||
946 	    ((pp->set & ENIC_SET_NAME) &&
947 	     nla_put(skb, IFLA_PORT_PROFILE, PORT_PROFILE_MAX, pp->name)) ||
948 	    ((pp->set & ENIC_SET_INSTANCE) &&
949 	     nla_put(skb, IFLA_PORT_INSTANCE_UUID, PORT_UUID_MAX,
950 		     pp->instance_uuid)) ||
951 	    ((pp->set & ENIC_SET_HOST) &&
952 	     nla_put(skb, IFLA_PORT_HOST_UUID, PORT_UUID_MAX, pp->host_uuid)))
953 		goto nla_put_failure;
954 	return 0;
955 
956 nla_put_failure:
957 	return -EMSGSIZE;
958 }
959 
960 static void enic_free_rq_buf(struct vnic_rq *rq, struct vnic_rq_buf *buf)
961 {
962 	struct enic *enic = vnic_dev_priv(rq->vdev);
963 
964 	if (!buf->os_buf)
965 		return;
966 
967 	pci_unmap_single(enic->pdev, buf->dma_addr,
968 		buf->len, PCI_DMA_FROMDEVICE);
969 	dev_kfree_skb_any(buf->os_buf);
970 	buf->os_buf = NULL;
971 }
972 
973 static int enic_rq_alloc_buf(struct vnic_rq *rq)
974 {
975 	struct enic *enic = vnic_dev_priv(rq->vdev);
976 	struct net_device *netdev = enic->netdev;
977 	struct sk_buff *skb;
978 	unsigned int len = netdev->mtu + VLAN_ETH_HLEN;
979 	unsigned int os_buf_index = 0;
980 	dma_addr_t dma_addr;
981 	struct vnic_rq_buf *buf = rq->to_use;
982 
983 	if (buf->os_buf) {
984 		enic_queue_rq_desc(rq, buf->os_buf, os_buf_index, buf->dma_addr,
985 				   buf->len);
986 
987 		return 0;
988 	}
989 	skb = netdev_alloc_skb_ip_align(netdev, len);
990 	if (!skb)
991 		return -ENOMEM;
992 
993 	dma_addr = pci_map_single(enic->pdev, skb->data, len,
994 				  PCI_DMA_FROMDEVICE);
995 	if (unlikely(enic_dma_map_check(enic, dma_addr))) {
996 		dev_kfree_skb(skb);
997 		return -ENOMEM;
998 	}
999 
1000 	enic_queue_rq_desc(rq, skb, os_buf_index,
1001 		dma_addr, len);
1002 
1003 	return 0;
1004 }
1005 
1006 static void enic_intr_update_pkt_size(struct vnic_rx_bytes_counter *pkt_size,
1007 				      u32 pkt_len)
1008 {
1009 	if (ENIC_LARGE_PKT_THRESHOLD <= pkt_len)
1010 		pkt_size->large_pkt_bytes_cnt += pkt_len;
1011 	else
1012 		pkt_size->small_pkt_bytes_cnt += pkt_len;
1013 }
1014 
1015 static bool enic_rxcopybreak(struct net_device *netdev, struct sk_buff **skb,
1016 			     struct vnic_rq_buf *buf, u16 len)
1017 {
1018 	struct enic *enic = netdev_priv(netdev);
1019 	struct sk_buff *new_skb;
1020 
1021 	if (len > enic->rx_copybreak)
1022 		return false;
1023 	new_skb = netdev_alloc_skb_ip_align(netdev, len);
1024 	if (!new_skb)
1025 		return false;
1026 	pci_dma_sync_single_for_cpu(enic->pdev, buf->dma_addr, len,
1027 				    DMA_FROM_DEVICE);
1028 	memcpy(new_skb->data, (*skb)->data, len);
1029 	*skb = new_skb;
1030 
1031 	return true;
1032 }
1033 
1034 static void enic_rq_indicate_buf(struct vnic_rq *rq,
1035 	struct cq_desc *cq_desc, struct vnic_rq_buf *buf,
1036 	int skipped, void *opaque)
1037 {
1038 	struct enic *enic = vnic_dev_priv(rq->vdev);
1039 	struct net_device *netdev = enic->netdev;
1040 	struct sk_buff *skb;
1041 	struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
1042 
1043 	u8 type, color, eop, sop, ingress_port, vlan_stripped;
1044 	u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof;
1045 	u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok;
1046 	u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc;
1047 	u8 packet_error;
1048 	u16 q_number, completed_index, bytes_written, vlan_tci, checksum;
1049 	u32 rss_hash;
1050 
1051 	if (skipped)
1052 		return;
1053 
1054 	skb = buf->os_buf;
1055 
1056 	cq_enet_rq_desc_dec((struct cq_enet_rq_desc *)cq_desc,
1057 		&type, &color, &q_number, &completed_index,
1058 		&ingress_port, &fcoe, &eop, &sop, &rss_type,
1059 		&csum_not_calc, &rss_hash, &bytes_written,
1060 		&packet_error, &vlan_stripped, &vlan_tci, &checksum,
1061 		&fcoe_sof, &fcoe_fc_crc_ok, &fcoe_enc_error,
1062 		&fcoe_eof, &tcp_udp_csum_ok, &udp, &tcp,
1063 		&ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment,
1064 		&fcs_ok);
1065 
1066 	if (packet_error) {
1067 
1068 		if (!fcs_ok) {
1069 			if (bytes_written > 0)
1070 				enic->rq_bad_fcs++;
1071 			else if (bytes_written == 0)
1072 				enic->rq_truncated_pkts++;
1073 		}
1074 
1075 		pci_unmap_single(enic->pdev, buf->dma_addr, buf->len,
1076 				 PCI_DMA_FROMDEVICE);
1077 		dev_kfree_skb_any(skb);
1078 		buf->os_buf = NULL;
1079 
1080 		return;
1081 	}
1082 
1083 	if (eop && bytes_written > 0) {
1084 
1085 		/* Good receive
1086 		 */
1087 
1088 		if (!enic_rxcopybreak(netdev, &skb, buf, bytes_written)) {
1089 			buf->os_buf = NULL;
1090 			pci_unmap_single(enic->pdev, buf->dma_addr, buf->len,
1091 					 PCI_DMA_FROMDEVICE);
1092 		}
1093 		prefetch(skb->data - NET_IP_ALIGN);
1094 
1095 		skb_put(skb, bytes_written);
1096 		skb->protocol = eth_type_trans(skb, netdev);
1097 		skb_record_rx_queue(skb, q_number);
1098 		if (netdev->features & NETIF_F_RXHASH) {
1099 			skb_set_hash(skb, rss_hash,
1100 				     (rss_type &
1101 				      (NIC_CFG_RSS_HASH_TYPE_TCP_IPV6_EX |
1102 				       NIC_CFG_RSS_HASH_TYPE_TCP_IPV6 |
1103 				       NIC_CFG_RSS_HASH_TYPE_TCP_IPV4)) ?
1104 				     PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3);
1105 		}
1106 
1107 		/* Hardware does not provide whole packet checksum. It only
1108 		 * provides pseudo checksum. Since hw validates the packet
1109 		 * checksum but not provide us the checksum value. use
1110 		 * CHECSUM_UNNECESSARY.
1111 		 */
1112 		if ((netdev->features & NETIF_F_RXCSUM) && tcp_udp_csum_ok &&
1113 		    ipv4_csum_ok)
1114 			skb->ip_summed = CHECKSUM_UNNECESSARY;
1115 
1116 		if (vlan_stripped)
1117 			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tci);
1118 
1119 		skb_mark_napi_id(skb, &enic->napi[rq->index]);
1120 		if (enic_poll_busy_polling(rq) ||
1121 		    !(netdev->features & NETIF_F_GRO))
1122 			netif_receive_skb(skb);
1123 		else
1124 			napi_gro_receive(&enic->napi[q_number], skb);
1125 		if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1126 			enic_intr_update_pkt_size(&cq->pkt_size_counter,
1127 						  bytes_written);
1128 	} else {
1129 
1130 		/* Buffer overflow
1131 		 */
1132 
1133 		pci_unmap_single(enic->pdev, buf->dma_addr, buf->len,
1134 				 PCI_DMA_FROMDEVICE);
1135 		dev_kfree_skb_any(skb);
1136 		buf->os_buf = NULL;
1137 	}
1138 }
1139 
1140 static int enic_rq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
1141 	u8 type, u16 q_number, u16 completed_index, void *opaque)
1142 {
1143 	struct enic *enic = vnic_dev_priv(vdev);
1144 
1145 	vnic_rq_service(&enic->rq[q_number], cq_desc,
1146 		completed_index, VNIC_RQ_RETURN_DESC,
1147 		enic_rq_indicate_buf, opaque);
1148 
1149 	return 0;
1150 }
1151 
1152 static void enic_set_int_moderation(struct enic *enic, struct vnic_rq *rq)
1153 {
1154 	unsigned int intr = enic_msix_rq_intr(enic, rq->index);
1155 	struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
1156 	u32 timer = cq->tobe_rx_coal_timeval;
1157 
1158 	if (cq->tobe_rx_coal_timeval != cq->cur_rx_coal_timeval) {
1159 		vnic_intr_coalescing_timer_set(&enic->intr[intr], timer);
1160 		cq->cur_rx_coal_timeval = cq->tobe_rx_coal_timeval;
1161 	}
1162 }
1163 
1164 static void enic_calc_int_moderation(struct enic *enic, struct vnic_rq *rq)
1165 {
1166 	struct enic_rx_coal *rx_coal = &enic->rx_coalesce_setting;
1167 	struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
1168 	struct vnic_rx_bytes_counter *pkt_size_counter = &cq->pkt_size_counter;
1169 	int index;
1170 	u32 timer;
1171 	u32 range_start;
1172 	u32 traffic;
1173 	u64 delta;
1174 	ktime_t now = ktime_get();
1175 
1176 	delta = ktime_us_delta(now, cq->prev_ts);
1177 	if (delta < ENIC_AIC_TS_BREAK)
1178 		return;
1179 	cq->prev_ts = now;
1180 
1181 	traffic = pkt_size_counter->large_pkt_bytes_cnt +
1182 		  pkt_size_counter->small_pkt_bytes_cnt;
1183 	/* The table takes Mbps
1184 	 * traffic *= 8    => bits
1185 	 * traffic *= (10^6 / delta)    => bps
1186 	 * traffic /= 10^6     => Mbps
1187 	 *
1188 	 * Combining, traffic *= (8 / delta)
1189 	 */
1190 
1191 	traffic <<= 3;
1192 	traffic = delta > UINT_MAX ? 0 : traffic / (u32)delta;
1193 
1194 	for (index = 0; index < ENIC_MAX_COALESCE_TIMERS; index++)
1195 		if (traffic < mod_table[index].rx_rate)
1196 			break;
1197 	range_start = (pkt_size_counter->small_pkt_bytes_cnt >
1198 		       pkt_size_counter->large_pkt_bytes_cnt << 1) ?
1199 		      rx_coal->small_pkt_range_start :
1200 		      rx_coal->large_pkt_range_start;
1201 	timer = range_start + ((rx_coal->range_end - range_start) *
1202 			       mod_table[index].range_percent / 100);
1203 	/* Damping */
1204 	cq->tobe_rx_coal_timeval = (timer + cq->tobe_rx_coal_timeval) >> 1;
1205 
1206 	pkt_size_counter->large_pkt_bytes_cnt = 0;
1207 	pkt_size_counter->small_pkt_bytes_cnt = 0;
1208 }
1209 
1210 static int enic_poll(struct napi_struct *napi, int budget)
1211 {
1212 	struct net_device *netdev = napi->dev;
1213 	struct enic *enic = netdev_priv(netdev);
1214 	unsigned int cq_rq = enic_cq_rq(enic, 0);
1215 	unsigned int cq_wq = enic_cq_wq(enic, 0);
1216 	unsigned int intr = enic_legacy_io_intr();
1217 	unsigned int rq_work_to_do = budget;
1218 	unsigned int wq_work_to_do = -1; /* no limit */
1219 	unsigned int  work_done, rq_work_done = 0, wq_work_done;
1220 	int err;
1221 
1222 	wq_work_done = vnic_cq_service(&enic->cq[cq_wq], wq_work_to_do,
1223 				       enic_wq_service, NULL);
1224 
1225 	if (!enic_poll_lock_napi(&enic->rq[cq_rq])) {
1226 		if (wq_work_done > 0)
1227 			vnic_intr_return_credits(&enic->intr[intr],
1228 						 wq_work_done,
1229 						 0 /* dont unmask intr */,
1230 						 0 /* dont reset intr timer */);
1231 		return budget;
1232 	}
1233 
1234 	if (budget > 0)
1235 		rq_work_done = vnic_cq_service(&enic->cq[cq_rq],
1236 			rq_work_to_do, enic_rq_service, NULL);
1237 
1238 	/* Accumulate intr event credits for this polling
1239 	 * cycle.  An intr event is the completion of a
1240 	 * a WQ or RQ packet.
1241 	 */
1242 
1243 	work_done = rq_work_done + wq_work_done;
1244 
1245 	if (work_done > 0)
1246 		vnic_intr_return_credits(&enic->intr[intr],
1247 			work_done,
1248 			0 /* don't unmask intr */,
1249 			0 /* don't reset intr timer */);
1250 
1251 	err = vnic_rq_fill(&enic->rq[0], enic_rq_alloc_buf);
1252 	enic_poll_unlock_napi(&enic->rq[cq_rq], napi);
1253 
1254 	/* Buffer allocation failed. Stay in polling
1255 	 * mode so we can try to fill the ring again.
1256 	 */
1257 
1258 	if (err)
1259 		rq_work_done = rq_work_to_do;
1260 	if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1261 		/* Call the function which refreshes the intr coalescing timer
1262 		 * value based on the traffic.
1263 		 */
1264 		enic_calc_int_moderation(enic, &enic->rq[0]);
1265 
1266 	if (rq_work_done < rq_work_to_do) {
1267 
1268 		/* Some work done, but not enough to stay in polling,
1269 		 * exit polling
1270 		 */
1271 
1272 		napi_complete(napi);
1273 		if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1274 			enic_set_int_moderation(enic, &enic->rq[0]);
1275 		vnic_intr_unmask(&enic->intr[intr]);
1276 	}
1277 
1278 	return rq_work_done;
1279 }
1280 
1281 #ifdef CONFIG_RFS_ACCEL
1282 static void enic_free_rx_cpu_rmap(struct enic *enic)
1283 {
1284 	free_irq_cpu_rmap(enic->netdev->rx_cpu_rmap);
1285 	enic->netdev->rx_cpu_rmap = NULL;
1286 }
1287 
1288 static void enic_set_rx_cpu_rmap(struct enic *enic)
1289 {
1290 	int i, res;
1291 
1292 	if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX) {
1293 		enic->netdev->rx_cpu_rmap = alloc_irq_cpu_rmap(enic->rq_count);
1294 		if (unlikely(!enic->netdev->rx_cpu_rmap))
1295 			return;
1296 		for (i = 0; i < enic->rq_count; i++) {
1297 			res = irq_cpu_rmap_add(enic->netdev->rx_cpu_rmap,
1298 					       enic->msix_entry[i].vector);
1299 			if (unlikely(res)) {
1300 				enic_free_rx_cpu_rmap(enic);
1301 				return;
1302 			}
1303 		}
1304 	}
1305 }
1306 
1307 #else
1308 
1309 static void enic_free_rx_cpu_rmap(struct enic *enic)
1310 {
1311 }
1312 
1313 static void enic_set_rx_cpu_rmap(struct enic *enic)
1314 {
1315 }
1316 
1317 #endif /* CONFIG_RFS_ACCEL */
1318 
1319 #ifdef CONFIG_NET_RX_BUSY_POLL
1320 static int enic_busy_poll(struct napi_struct *napi)
1321 {
1322 	struct net_device *netdev = napi->dev;
1323 	struct enic *enic = netdev_priv(netdev);
1324 	unsigned int rq = (napi - &enic->napi[0]);
1325 	unsigned int cq = enic_cq_rq(enic, rq);
1326 	unsigned int intr = enic_msix_rq_intr(enic, rq);
1327 	unsigned int work_to_do = -1; /* clean all pkts possible */
1328 	unsigned int work_done;
1329 
1330 	if (!enic_poll_lock_poll(&enic->rq[rq]))
1331 		return LL_FLUSH_BUSY;
1332 	work_done = vnic_cq_service(&enic->cq[cq], work_to_do,
1333 				    enic_rq_service, NULL);
1334 
1335 	if (work_done > 0)
1336 		vnic_intr_return_credits(&enic->intr[intr],
1337 					 work_done, 0, 0);
1338 	vnic_rq_fill(&enic->rq[rq], enic_rq_alloc_buf);
1339 	if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1340 		enic_calc_int_moderation(enic, &enic->rq[rq]);
1341 	enic_poll_unlock_poll(&enic->rq[rq]);
1342 
1343 	return work_done;
1344 }
1345 #endif /* CONFIG_NET_RX_BUSY_POLL */
1346 
1347 static int enic_poll_msix_wq(struct napi_struct *napi, int budget)
1348 {
1349 	struct net_device *netdev = napi->dev;
1350 	struct enic *enic = netdev_priv(netdev);
1351 	unsigned int wq_index = (napi - &enic->napi[0]) - enic->rq_count;
1352 	struct vnic_wq *wq = &enic->wq[wq_index];
1353 	unsigned int cq;
1354 	unsigned int intr;
1355 	unsigned int wq_work_to_do = -1; /* clean all desc possible */
1356 	unsigned int wq_work_done;
1357 	unsigned int wq_irq;
1358 
1359 	wq_irq = wq->index;
1360 	cq = enic_cq_wq(enic, wq_irq);
1361 	intr = enic_msix_wq_intr(enic, wq_irq);
1362 	wq_work_done = vnic_cq_service(&enic->cq[cq], wq_work_to_do,
1363 				       enic_wq_service, NULL);
1364 
1365 	vnic_intr_return_credits(&enic->intr[intr], wq_work_done,
1366 				 0 /* don't unmask intr */,
1367 				 1 /* reset intr timer */);
1368 	if (!wq_work_done) {
1369 		napi_complete(napi);
1370 		vnic_intr_unmask(&enic->intr[intr]);
1371 		return 0;
1372 	}
1373 
1374 	return budget;
1375 }
1376 
1377 static int enic_poll_msix_rq(struct napi_struct *napi, int budget)
1378 {
1379 	struct net_device *netdev = napi->dev;
1380 	struct enic *enic = netdev_priv(netdev);
1381 	unsigned int rq = (napi - &enic->napi[0]);
1382 	unsigned int cq = enic_cq_rq(enic, rq);
1383 	unsigned int intr = enic_msix_rq_intr(enic, rq);
1384 	unsigned int work_to_do = budget;
1385 	unsigned int work_done = 0;
1386 	int err;
1387 
1388 	if (!enic_poll_lock_napi(&enic->rq[rq]))
1389 		return budget;
1390 	/* Service RQ
1391 	 */
1392 
1393 	if (budget > 0)
1394 		work_done = vnic_cq_service(&enic->cq[cq],
1395 			work_to_do, enic_rq_service, NULL);
1396 
1397 	/* Return intr event credits for this polling
1398 	 * cycle.  An intr event is the completion of a
1399 	 * RQ packet.
1400 	 */
1401 
1402 	if (work_done > 0)
1403 		vnic_intr_return_credits(&enic->intr[intr],
1404 			work_done,
1405 			0 /* don't unmask intr */,
1406 			0 /* don't reset intr timer */);
1407 
1408 	err = vnic_rq_fill(&enic->rq[rq], enic_rq_alloc_buf);
1409 
1410 	/* Buffer allocation failed. Stay in polling mode
1411 	 * so we can try to fill the ring again.
1412 	 */
1413 
1414 	if (err)
1415 		work_done = work_to_do;
1416 	if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1417 		/* Call the function which refreshes the intr coalescing timer
1418 		 * value based on the traffic.
1419 		 */
1420 		enic_calc_int_moderation(enic, &enic->rq[rq]);
1421 
1422 	enic_poll_unlock_napi(&enic->rq[rq], napi);
1423 	if (work_done < work_to_do) {
1424 
1425 		/* Some work done, but not enough to stay in polling,
1426 		 * exit polling
1427 		 */
1428 
1429 		napi_complete(napi);
1430 		if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1431 			enic_set_int_moderation(enic, &enic->rq[rq]);
1432 		vnic_intr_unmask(&enic->intr[intr]);
1433 	}
1434 
1435 	return work_done;
1436 }
1437 
1438 static void enic_notify_timer(unsigned long data)
1439 {
1440 	struct enic *enic = (struct enic *)data;
1441 
1442 	enic_notify_check(enic);
1443 
1444 	mod_timer(&enic->notify_timer,
1445 		round_jiffies(jiffies + ENIC_NOTIFY_TIMER_PERIOD));
1446 }
1447 
1448 static void enic_free_intr(struct enic *enic)
1449 {
1450 	struct net_device *netdev = enic->netdev;
1451 	unsigned int i;
1452 
1453 	enic_free_rx_cpu_rmap(enic);
1454 	switch (vnic_dev_get_intr_mode(enic->vdev)) {
1455 	case VNIC_DEV_INTR_MODE_INTX:
1456 		free_irq(enic->pdev->irq, netdev);
1457 		break;
1458 	case VNIC_DEV_INTR_MODE_MSI:
1459 		free_irq(enic->pdev->irq, enic);
1460 		break;
1461 	case VNIC_DEV_INTR_MODE_MSIX:
1462 		for (i = 0; i < ARRAY_SIZE(enic->msix); i++)
1463 			if (enic->msix[i].requested)
1464 				free_irq(enic->msix_entry[i].vector,
1465 					enic->msix[i].devid);
1466 		break;
1467 	default:
1468 		break;
1469 	}
1470 }
1471 
1472 static int enic_request_intr(struct enic *enic)
1473 {
1474 	struct net_device *netdev = enic->netdev;
1475 	unsigned int i, intr;
1476 	int err = 0;
1477 
1478 	enic_set_rx_cpu_rmap(enic);
1479 	switch (vnic_dev_get_intr_mode(enic->vdev)) {
1480 
1481 	case VNIC_DEV_INTR_MODE_INTX:
1482 
1483 		err = request_irq(enic->pdev->irq, enic_isr_legacy,
1484 			IRQF_SHARED, netdev->name, netdev);
1485 		break;
1486 
1487 	case VNIC_DEV_INTR_MODE_MSI:
1488 
1489 		err = request_irq(enic->pdev->irq, enic_isr_msi,
1490 			0, netdev->name, enic);
1491 		break;
1492 
1493 	case VNIC_DEV_INTR_MODE_MSIX:
1494 
1495 		for (i = 0; i < enic->rq_count; i++) {
1496 			intr = enic_msix_rq_intr(enic, i);
1497 			snprintf(enic->msix[intr].devname,
1498 				sizeof(enic->msix[intr].devname),
1499 				"%.11s-rx-%d", netdev->name, i);
1500 			enic->msix[intr].isr = enic_isr_msix;
1501 			enic->msix[intr].devid = &enic->napi[i];
1502 		}
1503 
1504 		for (i = 0; i < enic->wq_count; i++) {
1505 			int wq = enic_cq_wq(enic, i);
1506 
1507 			intr = enic_msix_wq_intr(enic, i);
1508 			snprintf(enic->msix[intr].devname,
1509 				sizeof(enic->msix[intr].devname),
1510 				"%.11s-tx-%d", netdev->name, i);
1511 			enic->msix[intr].isr = enic_isr_msix;
1512 			enic->msix[intr].devid = &enic->napi[wq];
1513 		}
1514 
1515 		intr = enic_msix_err_intr(enic);
1516 		snprintf(enic->msix[intr].devname,
1517 			sizeof(enic->msix[intr].devname),
1518 			"%.11s-err", netdev->name);
1519 		enic->msix[intr].isr = enic_isr_msix_err;
1520 		enic->msix[intr].devid = enic;
1521 
1522 		intr = enic_msix_notify_intr(enic);
1523 		snprintf(enic->msix[intr].devname,
1524 			sizeof(enic->msix[intr].devname),
1525 			"%.11s-notify", netdev->name);
1526 		enic->msix[intr].isr = enic_isr_msix_notify;
1527 		enic->msix[intr].devid = enic;
1528 
1529 		for (i = 0; i < ARRAY_SIZE(enic->msix); i++)
1530 			enic->msix[i].requested = 0;
1531 
1532 		for (i = 0; i < enic->intr_count; i++) {
1533 			err = request_irq(enic->msix_entry[i].vector,
1534 				enic->msix[i].isr, 0,
1535 				enic->msix[i].devname,
1536 				enic->msix[i].devid);
1537 			if (err) {
1538 				enic_free_intr(enic);
1539 				break;
1540 			}
1541 			enic->msix[i].requested = 1;
1542 		}
1543 
1544 		break;
1545 
1546 	default:
1547 		break;
1548 	}
1549 
1550 	return err;
1551 }
1552 
1553 static void enic_synchronize_irqs(struct enic *enic)
1554 {
1555 	unsigned int i;
1556 
1557 	switch (vnic_dev_get_intr_mode(enic->vdev)) {
1558 	case VNIC_DEV_INTR_MODE_INTX:
1559 	case VNIC_DEV_INTR_MODE_MSI:
1560 		synchronize_irq(enic->pdev->irq);
1561 		break;
1562 	case VNIC_DEV_INTR_MODE_MSIX:
1563 		for (i = 0; i < enic->intr_count; i++)
1564 			synchronize_irq(enic->msix_entry[i].vector);
1565 		break;
1566 	default:
1567 		break;
1568 	}
1569 }
1570 
1571 static void enic_set_rx_coal_setting(struct enic *enic)
1572 {
1573 	unsigned int speed;
1574 	int index = -1;
1575 	struct enic_rx_coal *rx_coal = &enic->rx_coalesce_setting;
1576 
1577 	/* 1. Read the link speed from fw
1578 	 * 2. Pick the default range for the speed
1579 	 * 3. Update it in enic->rx_coalesce_setting
1580 	 */
1581 	speed = vnic_dev_port_speed(enic->vdev);
1582 	if (ENIC_LINK_SPEED_10G < speed)
1583 		index = ENIC_LINK_40G_INDEX;
1584 	else if (ENIC_LINK_SPEED_4G < speed)
1585 		index = ENIC_LINK_10G_INDEX;
1586 	else
1587 		index = ENIC_LINK_4G_INDEX;
1588 
1589 	rx_coal->small_pkt_range_start = mod_range[index].small_pkt_range_start;
1590 	rx_coal->large_pkt_range_start = mod_range[index].large_pkt_range_start;
1591 	rx_coal->range_end = ENIC_RX_COALESCE_RANGE_END;
1592 
1593 	/* Start with the value provided by UCSM */
1594 	for (index = 0; index < enic->rq_count; index++)
1595 		enic->cq[index].cur_rx_coal_timeval =
1596 				enic->config.intr_timer_usec;
1597 
1598 	rx_coal->use_adaptive_rx_coalesce = 1;
1599 }
1600 
1601 static int enic_dev_notify_set(struct enic *enic)
1602 {
1603 	int err;
1604 
1605 	spin_lock_bh(&enic->devcmd_lock);
1606 	switch (vnic_dev_get_intr_mode(enic->vdev)) {
1607 	case VNIC_DEV_INTR_MODE_INTX:
1608 		err = vnic_dev_notify_set(enic->vdev,
1609 			enic_legacy_notify_intr());
1610 		break;
1611 	case VNIC_DEV_INTR_MODE_MSIX:
1612 		err = vnic_dev_notify_set(enic->vdev,
1613 			enic_msix_notify_intr(enic));
1614 		break;
1615 	default:
1616 		err = vnic_dev_notify_set(enic->vdev, -1 /* no intr */);
1617 		break;
1618 	}
1619 	spin_unlock_bh(&enic->devcmd_lock);
1620 
1621 	return err;
1622 }
1623 
1624 static void enic_notify_timer_start(struct enic *enic)
1625 {
1626 	switch (vnic_dev_get_intr_mode(enic->vdev)) {
1627 	case VNIC_DEV_INTR_MODE_MSI:
1628 		mod_timer(&enic->notify_timer, jiffies);
1629 		break;
1630 	default:
1631 		/* Using intr for notification for INTx/MSI-X */
1632 		break;
1633 	}
1634 }
1635 
1636 /* rtnl lock is held, process context */
1637 static int enic_open(struct net_device *netdev)
1638 {
1639 	struct enic *enic = netdev_priv(netdev);
1640 	unsigned int i;
1641 	int err;
1642 
1643 	err = enic_request_intr(enic);
1644 	if (err) {
1645 		netdev_err(netdev, "Unable to request irq.\n");
1646 		return err;
1647 	}
1648 
1649 	err = enic_dev_notify_set(enic);
1650 	if (err) {
1651 		netdev_err(netdev,
1652 			"Failed to alloc notify buffer, aborting.\n");
1653 		goto err_out_free_intr;
1654 	}
1655 
1656 	for (i = 0; i < enic->rq_count; i++) {
1657 		vnic_rq_fill(&enic->rq[i], enic_rq_alloc_buf);
1658 		/* Need at least one buffer on ring to get going */
1659 		if (vnic_rq_desc_used(&enic->rq[i]) == 0) {
1660 			netdev_err(netdev, "Unable to alloc receive buffers\n");
1661 			err = -ENOMEM;
1662 			goto err_out_free_rq;
1663 		}
1664 	}
1665 
1666 	for (i = 0; i < enic->wq_count; i++)
1667 		vnic_wq_enable(&enic->wq[i]);
1668 	for (i = 0; i < enic->rq_count; i++)
1669 		vnic_rq_enable(&enic->rq[i]);
1670 
1671 	if (!enic_is_dynamic(enic) && !enic_is_sriov_vf(enic))
1672 		enic_dev_add_station_addr(enic);
1673 
1674 	enic_set_rx_mode(netdev);
1675 
1676 	netif_tx_wake_all_queues(netdev);
1677 
1678 	for (i = 0; i < enic->rq_count; i++) {
1679 		enic_busy_poll_init_lock(&enic->rq[i]);
1680 		napi_enable(&enic->napi[i]);
1681 	}
1682 	if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX)
1683 		for (i = 0; i < enic->wq_count; i++)
1684 			napi_enable(&enic->napi[enic_cq_wq(enic, i)]);
1685 	enic_dev_enable(enic);
1686 
1687 	for (i = 0; i < enic->intr_count; i++)
1688 		vnic_intr_unmask(&enic->intr[i]);
1689 
1690 	enic_notify_timer_start(enic);
1691 	enic_rfs_flw_tbl_init(enic);
1692 
1693 	return 0;
1694 
1695 err_out_free_rq:
1696 	for (i = 0; i < enic->rq_count; i++)
1697 		vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
1698 	enic_dev_notify_unset(enic);
1699 err_out_free_intr:
1700 	enic_free_intr(enic);
1701 
1702 	return err;
1703 }
1704 
1705 /* rtnl lock is held, process context */
1706 static int enic_stop(struct net_device *netdev)
1707 {
1708 	struct enic *enic = netdev_priv(netdev);
1709 	unsigned int i;
1710 	int err;
1711 
1712 	for (i = 0; i < enic->intr_count; i++) {
1713 		vnic_intr_mask(&enic->intr[i]);
1714 		(void)vnic_intr_masked(&enic->intr[i]); /* flush write */
1715 	}
1716 
1717 	enic_synchronize_irqs(enic);
1718 
1719 	del_timer_sync(&enic->notify_timer);
1720 	enic_rfs_flw_tbl_free(enic);
1721 
1722 	enic_dev_disable(enic);
1723 
1724 	for (i = 0; i < enic->rq_count; i++) {
1725 		napi_disable(&enic->napi[i]);
1726 		local_bh_disable();
1727 		while (!enic_poll_lock_napi(&enic->rq[i]))
1728 			mdelay(1);
1729 		local_bh_enable();
1730 	}
1731 
1732 	netif_carrier_off(netdev);
1733 	netif_tx_disable(netdev);
1734 	if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX)
1735 		for (i = 0; i < enic->wq_count; i++)
1736 			napi_disable(&enic->napi[enic_cq_wq(enic, i)]);
1737 
1738 	if (!enic_is_dynamic(enic) && !enic_is_sriov_vf(enic))
1739 		enic_dev_del_station_addr(enic);
1740 
1741 	for (i = 0; i < enic->wq_count; i++) {
1742 		err = vnic_wq_disable(&enic->wq[i]);
1743 		if (err)
1744 			return err;
1745 	}
1746 	for (i = 0; i < enic->rq_count; i++) {
1747 		err = vnic_rq_disable(&enic->rq[i]);
1748 		if (err)
1749 			return err;
1750 	}
1751 
1752 	enic_dev_notify_unset(enic);
1753 	enic_free_intr(enic);
1754 
1755 	for (i = 0; i < enic->wq_count; i++)
1756 		vnic_wq_clean(&enic->wq[i], enic_free_wq_buf);
1757 	for (i = 0; i < enic->rq_count; i++)
1758 		vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
1759 	for (i = 0; i < enic->cq_count; i++)
1760 		vnic_cq_clean(&enic->cq[i]);
1761 	for (i = 0; i < enic->intr_count; i++)
1762 		vnic_intr_clean(&enic->intr[i]);
1763 
1764 	return 0;
1765 }
1766 
1767 static int enic_change_mtu(struct net_device *netdev, int new_mtu)
1768 {
1769 	struct enic *enic = netdev_priv(netdev);
1770 	int running = netif_running(netdev);
1771 
1772 	if (new_mtu < ENIC_MIN_MTU || new_mtu > ENIC_MAX_MTU)
1773 		return -EINVAL;
1774 
1775 	if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic))
1776 		return -EOPNOTSUPP;
1777 
1778 	if (running)
1779 		enic_stop(netdev);
1780 
1781 	netdev->mtu = new_mtu;
1782 
1783 	if (netdev->mtu > enic->port_mtu)
1784 		netdev_warn(netdev,
1785 			"interface MTU (%d) set higher than port MTU (%d)\n",
1786 			netdev->mtu, enic->port_mtu);
1787 
1788 	if (running)
1789 		enic_open(netdev);
1790 
1791 	return 0;
1792 }
1793 
1794 static void enic_change_mtu_work(struct work_struct *work)
1795 {
1796 	struct enic *enic = container_of(work, struct enic, change_mtu_work);
1797 	struct net_device *netdev = enic->netdev;
1798 	int new_mtu = vnic_dev_mtu(enic->vdev);
1799 	int err;
1800 	unsigned int i;
1801 
1802 	new_mtu = max_t(int, ENIC_MIN_MTU, min_t(int, ENIC_MAX_MTU, new_mtu));
1803 
1804 	rtnl_lock();
1805 
1806 	/* Stop RQ */
1807 	del_timer_sync(&enic->notify_timer);
1808 
1809 	for (i = 0; i < enic->rq_count; i++)
1810 		napi_disable(&enic->napi[i]);
1811 
1812 	vnic_intr_mask(&enic->intr[0]);
1813 	enic_synchronize_irqs(enic);
1814 	err = vnic_rq_disable(&enic->rq[0]);
1815 	if (err) {
1816 		rtnl_unlock();
1817 		netdev_err(netdev, "Unable to disable RQ.\n");
1818 		return;
1819 	}
1820 	vnic_rq_clean(&enic->rq[0], enic_free_rq_buf);
1821 	vnic_cq_clean(&enic->cq[0]);
1822 	vnic_intr_clean(&enic->intr[0]);
1823 
1824 	/* Fill RQ with new_mtu-sized buffers */
1825 	netdev->mtu = new_mtu;
1826 	vnic_rq_fill(&enic->rq[0], enic_rq_alloc_buf);
1827 	/* Need at least one buffer on ring to get going */
1828 	if (vnic_rq_desc_used(&enic->rq[0]) == 0) {
1829 		rtnl_unlock();
1830 		netdev_err(netdev, "Unable to alloc receive buffers.\n");
1831 		return;
1832 	}
1833 
1834 	/* Start RQ */
1835 	vnic_rq_enable(&enic->rq[0]);
1836 	napi_enable(&enic->napi[0]);
1837 	vnic_intr_unmask(&enic->intr[0]);
1838 	enic_notify_timer_start(enic);
1839 
1840 	rtnl_unlock();
1841 
1842 	netdev_info(netdev, "interface MTU set as %d\n", netdev->mtu);
1843 }
1844 
1845 #ifdef CONFIG_NET_POLL_CONTROLLER
1846 static void enic_poll_controller(struct net_device *netdev)
1847 {
1848 	struct enic *enic = netdev_priv(netdev);
1849 	struct vnic_dev *vdev = enic->vdev;
1850 	unsigned int i, intr;
1851 
1852 	switch (vnic_dev_get_intr_mode(vdev)) {
1853 	case VNIC_DEV_INTR_MODE_MSIX:
1854 		for (i = 0; i < enic->rq_count; i++) {
1855 			intr = enic_msix_rq_intr(enic, i);
1856 			enic_isr_msix(enic->msix_entry[intr].vector,
1857 				      &enic->napi[i]);
1858 		}
1859 
1860 		for (i = 0; i < enic->wq_count; i++) {
1861 			intr = enic_msix_wq_intr(enic, i);
1862 			enic_isr_msix(enic->msix_entry[intr].vector,
1863 				      &enic->napi[enic_cq_wq(enic, i)]);
1864 		}
1865 
1866 		break;
1867 	case VNIC_DEV_INTR_MODE_MSI:
1868 		enic_isr_msi(enic->pdev->irq, enic);
1869 		break;
1870 	case VNIC_DEV_INTR_MODE_INTX:
1871 		enic_isr_legacy(enic->pdev->irq, netdev);
1872 		break;
1873 	default:
1874 		break;
1875 	}
1876 }
1877 #endif
1878 
1879 static int enic_dev_wait(struct vnic_dev *vdev,
1880 	int (*start)(struct vnic_dev *, int),
1881 	int (*finished)(struct vnic_dev *, int *),
1882 	int arg)
1883 {
1884 	unsigned long time;
1885 	int done;
1886 	int err;
1887 
1888 	BUG_ON(in_interrupt());
1889 
1890 	err = start(vdev, arg);
1891 	if (err)
1892 		return err;
1893 
1894 	/* Wait for func to complete...2 seconds max
1895 	 */
1896 
1897 	time = jiffies + (HZ * 2);
1898 	do {
1899 
1900 		err = finished(vdev, &done);
1901 		if (err)
1902 			return err;
1903 
1904 		if (done)
1905 			return 0;
1906 
1907 		schedule_timeout_uninterruptible(HZ / 10);
1908 
1909 	} while (time_after(time, jiffies));
1910 
1911 	return -ETIMEDOUT;
1912 }
1913 
1914 static int enic_dev_open(struct enic *enic)
1915 {
1916 	int err;
1917 
1918 	err = enic_dev_wait(enic->vdev, vnic_dev_open,
1919 		vnic_dev_open_done, 0);
1920 	if (err)
1921 		dev_err(enic_get_dev(enic), "vNIC device open failed, err %d\n",
1922 			err);
1923 
1924 	return err;
1925 }
1926 
1927 static int enic_dev_hang_reset(struct enic *enic)
1928 {
1929 	int err;
1930 
1931 	err = enic_dev_wait(enic->vdev, vnic_dev_hang_reset,
1932 		vnic_dev_hang_reset_done, 0);
1933 	if (err)
1934 		netdev_err(enic->netdev, "vNIC hang reset failed, err %d\n",
1935 			err);
1936 
1937 	return err;
1938 }
1939 
1940 int __enic_set_rsskey(struct enic *enic)
1941 {
1942 	union vnic_rss_key *rss_key_buf_va;
1943 	dma_addr_t rss_key_buf_pa;
1944 	int i, kidx, bidx, err;
1945 
1946 	rss_key_buf_va = pci_zalloc_consistent(enic->pdev,
1947 					       sizeof(union vnic_rss_key),
1948 					       &rss_key_buf_pa);
1949 	if (!rss_key_buf_va)
1950 		return -ENOMEM;
1951 
1952 	for (i = 0; i < ENIC_RSS_LEN; i++) {
1953 		kidx = i / ENIC_RSS_BYTES_PER_KEY;
1954 		bidx = i % ENIC_RSS_BYTES_PER_KEY;
1955 		rss_key_buf_va->key[kidx].b[bidx] = enic->rss_key[i];
1956 	}
1957 	spin_lock_bh(&enic->devcmd_lock);
1958 	err = enic_set_rss_key(enic,
1959 		rss_key_buf_pa,
1960 		sizeof(union vnic_rss_key));
1961 	spin_unlock_bh(&enic->devcmd_lock);
1962 
1963 	pci_free_consistent(enic->pdev, sizeof(union vnic_rss_key),
1964 		rss_key_buf_va, rss_key_buf_pa);
1965 
1966 	return err;
1967 }
1968 
1969 static int enic_set_rsskey(struct enic *enic)
1970 {
1971 	netdev_rss_key_fill(enic->rss_key, ENIC_RSS_LEN);
1972 
1973 	return __enic_set_rsskey(enic);
1974 }
1975 
1976 static int enic_set_rsscpu(struct enic *enic, u8 rss_hash_bits)
1977 {
1978 	dma_addr_t rss_cpu_buf_pa;
1979 	union vnic_rss_cpu *rss_cpu_buf_va = NULL;
1980 	unsigned int i;
1981 	int err;
1982 
1983 	rss_cpu_buf_va = pci_alloc_consistent(enic->pdev,
1984 		sizeof(union vnic_rss_cpu), &rss_cpu_buf_pa);
1985 	if (!rss_cpu_buf_va)
1986 		return -ENOMEM;
1987 
1988 	for (i = 0; i < (1 << rss_hash_bits); i++)
1989 		(*rss_cpu_buf_va).cpu[i/4].b[i%4] = i % enic->rq_count;
1990 
1991 	spin_lock_bh(&enic->devcmd_lock);
1992 	err = enic_set_rss_cpu(enic,
1993 		rss_cpu_buf_pa,
1994 		sizeof(union vnic_rss_cpu));
1995 	spin_unlock_bh(&enic->devcmd_lock);
1996 
1997 	pci_free_consistent(enic->pdev, sizeof(union vnic_rss_cpu),
1998 		rss_cpu_buf_va, rss_cpu_buf_pa);
1999 
2000 	return err;
2001 }
2002 
2003 static int enic_set_niccfg(struct enic *enic, u8 rss_default_cpu,
2004 	u8 rss_hash_type, u8 rss_hash_bits, u8 rss_base_cpu, u8 rss_enable)
2005 {
2006 	const u8 tso_ipid_split_en = 0;
2007 	const u8 ig_vlan_strip_en = 1;
2008 	int err;
2009 
2010 	/* Enable VLAN tag stripping.
2011 	*/
2012 
2013 	spin_lock_bh(&enic->devcmd_lock);
2014 	err = enic_set_nic_cfg(enic,
2015 		rss_default_cpu, rss_hash_type,
2016 		rss_hash_bits, rss_base_cpu,
2017 		rss_enable, tso_ipid_split_en,
2018 		ig_vlan_strip_en);
2019 	spin_unlock_bh(&enic->devcmd_lock);
2020 
2021 	return err;
2022 }
2023 
2024 static int enic_set_rss_nic_cfg(struct enic *enic)
2025 {
2026 	struct device *dev = enic_get_dev(enic);
2027 	const u8 rss_default_cpu = 0;
2028 	const u8 rss_hash_type = NIC_CFG_RSS_HASH_TYPE_IPV4 |
2029 		NIC_CFG_RSS_HASH_TYPE_TCP_IPV4 |
2030 		NIC_CFG_RSS_HASH_TYPE_IPV6 |
2031 		NIC_CFG_RSS_HASH_TYPE_TCP_IPV6;
2032 	const u8 rss_hash_bits = 7;
2033 	const u8 rss_base_cpu = 0;
2034 	u8 rss_enable = ENIC_SETTING(enic, RSS) && (enic->rq_count > 1);
2035 
2036 	if (rss_enable) {
2037 		if (!enic_set_rsskey(enic)) {
2038 			if (enic_set_rsscpu(enic, rss_hash_bits)) {
2039 				rss_enable = 0;
2040 				dev_warn(dev, "RSS disabled, "
2041 					"Failed to set RSS cpu indirection table.");
2042 			}
2043 		} else {
2044 			rss_enable = 0;
2045 			dev_warn(dev, "RSS disabled, Failed to set RSS key.\n");
2046 		}
2047 	}
2048 
2049 	return enic_set_niccfg(enic, rss_default_cpu, rss_hash_type,
2050 		rss_hash_bits, rss_base_cpu, rss_enable);
2051 }
2052 
2053 static void enic_reset(struct work_struct *work)
2054 {
2055 	struct enic *enic = container_of(work, struct enic, reset);
2056 
2057 	if (!netif_running(enic->netdev))
2058 		return;
2059 
2060 	rtnl_lock();
2061 
2062 	spin_lock(&enic->enic_api_lock);
2063 	enic_dev_hang_notify(enic);
2064 	enic_stop(enic->netdev);
2065 	enic_dev_hang_reset(enic);
2066 	enic_reset_addr_lists(enic);
2067 	enic_init_vnic_resources(enic);
2068 	enic_set_rss_nic_cfg(enic);
2069 	enic_dev_set_ig_vlan_rewrite_mode(enic);
2070 	enic_open(enic->netdev);
2071 	spin_unlock(&enic->enic_api_lock);
2072 	call_netdevice_notifiers(NETDEV_REBOOT, enic->netdev);
2073 
2074 	rtnl_unlock();
2075 }
2076 
2077 static int enic_set_intr_mode(struct enic *enic)
2078 {
2079 	unsigned int n = min_t(unsigned int, enic->rq_count, ENIC_RQ_MAX);
2080 	unsigned int m = min_t(unsigned int, enic->wq_count, ENIC_WQ_MAX);
2081 	unsigned int i;
2082 
2083 	/* Set interrupt mode (INTx, MSI, MSI-X) depending
2084 	 * on system capabilities.
2085 	 *
2086 	 * Try MSI-X first
2087 	 *
2088 	 * We need n RQs, m WQs, n+m CQs, and n+m+2 INTRs
2089 	 * (the second to last INTR is used for WQ/RQ errors)
2090 	 * (the last INTR is used for notifications)
2091 	 */
2092 
2093 	BUG_ON(ARRAY_SIZE(enic->msix_entry) < n + m + 2);
2094 	for (i = 0; i < n + m + 2; i++)
2095 		enic->msix_entry[i].entry = i;
2096 
2097 	/* Use multiple RQs if RSS is enabled
2098 	 */
2099 
2100 	if (ENIC_SETTING(enic, RSS) &&
2101 	    enic->config.intr_mode < 1 &&
2102 	    enic->rq_count >= n &&
2103 	    enic->wq_count >= m &&
2104 	    enic->cq_count >= n + m &&
2105 	    enic->intr_count >= n + m + 2) {
2106 
2107 		if (pci_enable_msix_range(enic->pdev, enic->msix_entry,
2108 					  n + m + 2, n + m + 2) > 0) {
2109 
2110 			enic->rq_count = n;
2111 			enic->wq_count = m;
2112 			enic->cq_count = n + m;
2113 			enic->intr_count = n + m + 2;
2114 
2115 			vnic_dev_set_intr_mode(enic->vdev,
2116 				VNIC_DEV_INTR_MODE_MSIX);
2117 
2118 			return 0;
2119 		}
2120 	}
2121 
2122 	if (enic->config.intr_mode < 1 &&
2123 	    enic->rq_count >= 1 &&
2124 	    enic->wq_count >= m &&
2125 	    enic->cq_count >= 1 + m &&
2126 	    enic->intr_count >= 1 + m + 2) {
2127 		if (pci_enable_msix_range(enic->pdev, enic->msix_entry,
2128 					  1 + m + 2, 1 + m + 2) > 0) {
2129 
2130 			enic->rq_count = 1;
2131 			enic->wq_count = m;
2132 			enic->cq_count = 1 + m;
2133 			enic->intr_count = 1 + m + 2;
2134 
2135 			vnic_dev_set_intr_mode(enic->vdev,
2136 				VNIC_DEV_INTR_MODE_MSIX);
2137 
2138 			return 0;
2139 		}
2140 	}
2141 
2142 	/* Next try MSI
2143 	 *
2144 	 * We need 1 RQ, 1 WQ, 2 CQs, and 1 INTR
2145 	 */
2146 
2147 	if (enic->config.intr_mode < 2 &&
2148 	    enic->rq_count >= 1 &&
2149 	    enic->wq_count >= 1 &&
2150 	    enic->cq_count >= 2 &&
2151 	    enic->intr_count >= 1 &&
2152 	    !pci_enable_msi(enic->pdev)) {
2153 
2154 		enic->rq_count = 1;
2155 		enic->wq_count = 1;
2156 		enic->cq_count = 2;
2157 		enic->intr_count = 1;
2158 
2159 		vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSI);
2160 
2161 		return 0;
2162 	}
2163 
2164 	/* Next try INTx
2165 	 *
2166 	 * We need 1 RQ, 1 WQ, 2 CQs, and 3 INTRs
2167 	 * (the first INTR is used for WQ/RQ)
2168 	 * (the second INTR is used for WQ/RQ errors)
2169 	 * (the last INTR is used for notifications)
2170 	 */
2171 
2172 	if (enic->config.intr_mode < 3 &&
2173 	    enic->rq_count >= 1 &&
2174 	    enic->wq_count >= 1 &&
2175 	    enic->cq_count >= 2 &&
2176 	    enic->intr_count >= 3) {
2177 
2178 		enic->rq_count = 1;
2179 		enic->wq_count = 1;
2180 		enic->cq_count = 2;
2181 		enic->intr_count = 3;
2182 
2183 		vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_INTX);
2184 
2185 		return 0;
2186 	}
2187 
2188 	vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
2189 
2190 	return -EINVAL;
2191 }
2192 
2193 static void enic_clear_intr_mode(struct enic *enic)
2194 {
2195 	switch (vnic_dev_get_intr_mode(enic->vdev)) {
2196 	case VNIC_DEV_INTR_MODE_MSIX:
2197 		pci_disable_msix(enic->pdev);
2198 		break;
2199 	case VNIC_DEV_INTR_MODE_MSI:
2200 		pci_disable_msi(enic->pdev);
2201 		break;
2202 	default:
2203 		break;
2204 	}
2205 
2206 	vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
2207 }
2208 
2209 static const struct net_device_ops enic_netdev_dynamic_ops = {
2210 	.ndo_open		= enic_open,
2211 	.ndo_stop		= enic_stop,
2212 	.ndo_start_xmit		= enic_hard_start_xmit,
2213 	.ndo_get_stats64	= enic_get_stats,
2214 	.ndo_validate_addr	= eth_validate_addr,
2215 	.ndo_set_rx_mode	= enic_set_rx_mode,
2216 	.ndo_set_mac_address	= enic_set_mac_address_dynamic,
2217 	.ndo_change_mtu		= enic_change_mtu,
2218 	.ndo_vlan_rx_add_vid	= enic_vlan_rx_add_vid,
2219 	.ndo_vlan_rx_kill_vid	= enic_vlan_rx_kill_vid,
2220 	.ndo_tx_timeout		= enic_tx_timeout,
2221 	.ndo_set_vf_port	= enic_set_vf_port,
2222 	.ndo_get_vf_port	= enic_get_vf_port,
2223 	.ndo_set_vf_mac		= enic_set_vf_mac,
2224 #ifdef CONFIG_NET_POLL_CONTROLLER
2225 	.ndo_poll_controller	= enic_poll_controller,
2226 #endif
2227 #ifdef CONFIG_RFS_ACCEL
2228 	.ndo_rx_flow_steer	= enic_rx_flow_steer,
2229 #endif
2230 #ifdef CONFIG_NET_RX_BUSY_POLL
2231 	.ndo_busy_poll		= enic_busy_poll,
2232 #endif
2233 };
2234 
2235 static const struct net_device_ops enic_netdev_ops = {
2236 	.ndo_open		= enic_open,
2237 	.ndo_stop		= enic_stop,
2238 	.ndo_start_xmit		= enic_hard_start_xmit,
2239 	.ndo_get_stats64	= enic_get_stats,
2240 	.ndo_validate_addr	= eth_validate_addr,
2241 	.ndo_set_mac_address	= enic_set_mac_address,
2242 	.ndo_set_rx_mode	= enic_set_rx_mode,
2243 	.ndo_change_mtu		= enic_change_mtu,
2244 	.ndo_vlan_rx_add_vid	= enic_vlan_rx_add_vid,
2245 	.ndo_vlan_rx_kill_vid	= enic_vlan_rx_kill_vid,
2246 	.ndo_tx_timeout		= enic_tx_timeout,
2247 	.ndo_set_vf_port	= enic_set_vf_port,
2248 	.ndo_get_vf_port	= enic_get_vf_port,
2249 	.ndo_set_vf_mac		= enic_set_vf_mac,
2250 #ifdef CONFIG_NET_POLL_CONTROLLER
2251 	.ndo_poll_controller	= enic_poll_controller,
2252 #endif
2253 #ifdef CONFIG_RFS_ACCEL
2254 	.ndo_rx_flow_steer	= enic_rx_flow_steer,
2255 #endif
2256 #ifdef CONFIG_NET_RX_BUSY_POLL
2257 	.ndo_busy_poll		= enic_busy_poll,
2258 #endif
2259 };
2260 
2261 static void enic_dev_deinit(struct enic *enic)
2262 {
2263 	unsigned int i;
2264 
2265 	for (i = 0; i < enic->rq_count; i++) {
2266 		napi_hash_del(&enic->napi[i]);
2267 		netif_napi_del(&enic->napi[i]);
2268 	}
2269 	if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX)
2270 		for (i = 0; i < enic->wq_count; i++)
2271 			netif_napi_del(&enic->napi[enic_cq_wq(enic, i)]);
2272 
2273 	enic_free_vnic_resources(enic);
2274 	enic_clear_intr_mode(enic);
2275 }
2276 
2277 static void enic_kdump_kernel_config(struct enic *enic)
2278 {
2279 	if (is_kdump_kernel()) {
2280 		dev_info(enic_get_dev(enic), "Running from within kdump kernel. Using minimal resources\n");
2281 		enic->rq_count = 1;
2282 		enic->wq_count = 1;
2283 		enic->config.rq_desc_count = ENIC_MIN_RQ_DESCS;
2284 		enic->config.wq_desc_count = ENIC_MIN_WQ_DESCS;
2285 		enic->config.mtu = min_t(u16, 1500, enic->config.mtu);
2286 	}
2287 }
2288 
2289 static int enic_dev_init(struct enic *enic)
2290 {
2291 	struct device *dev = enic_get_dev(enic);
2292 	struct net_device *netdev = enic->netdev;
2293 	unsigned int i;
2294 	int err;
2295 
2296 	/* Get interrupt coalesce timer info */
2297 	err = enic_dev_intr_coal_timer_info(enic);
2298 	if (err) {
2299 		dev_warn(dev, "Using default conversion factor for "
2300 			"interrupt coalesce timer\n");
2301 		vnic_dev_intr_coal_timer_info_default(enic->vdev);
2302 	}
2303 
2304 	/* Get vNIC configuration
2305 	 */
2306 
2307 	err = enic_get_vnic_config(enic);
2308 	if (err) {
2309 		dev_err(dev, "Get vNIC configuration failed, aborting\n");
2310 		return err;
2311 	}
2312 
2313 	/* Get available resource counts
2314 	 */
2315 
2316 	enic_get_res_counts(enic);
2317 
2318 	/* modify resource count if we are in kdump_kernel
2319 	 */
2320 	enic_kdump_kernel_config(enic);
2321 
2322 	/* Set interrupt mode based on resource counts and system
2323 	 * capabilities
2324 	 */
2325 
2326 	err = enic_set_intr_mode(enic);
2327 	if (err) {
2328 		dev_err(dev, "Failed to set intr mode based on resource "
2329 			"counts and system capabilities, aborting\n");
2330 		return err;
2331 	}
2332 
2333 	/* Allocate and configure vNIC resources
2334 	 */
2335 
2336 	err = enic_alloc_vnic_resources(enic);
2337 	if (err) {
2338 		dev_err(dev, "Failed to alloc vNIC resources, aborting\n");
2339 		goto err_out_free_vnic_resources;
2340 	}
2341 
2342 	enic_init_vnic_resources(enic);
2343 
2344 	err = enic_set_rss_nic_cfg(enic);
2345 	if (err) {
2346 		dev_err(dev, "Failed to config nic, aborting\n");
2347 		goto err_out_free_vnic_resources;
2348 	}
2349 
2350 	switch (vnic_dev_get_intr_mode(enic->vdev)) {
2351 	default:
2352 		netif_napi_add(netdev, &enic->napi[0], enic_poll, 64);
2353 		napi_hash_add(&enic->napi[0]);
2354 		break;
2355 	case VNIC_DEV_INTR_MODE_MSIX:
2356 		for (i = 0; i < enic->rq_count; i++) {
2357 			netif_napi_add(netdev, &enic->napi[i],
2358 				enic_poll_msix_rq, NAPI_POLL_WEIGHT);
2359 			napi_hash_add(&enic->napi[i]);
2360 		}
2361 		for (i = 0; i < enic->wq_count; i++)
2362 			netif_napi_add(netdev, &enic->napi[enic_cq_wq(enic, i)],
2363 				       enic_poll_msix_wq, NAPI_POLL_WEIGHT);
2364 		break;
2365 	}
2366 
2367 	return 0;
2368 
2369 err_out_free_vnic_resources:
2370 	enic_clear_intr_mode(enic);
2371 	enic_free_vnic_resources(enic);
2372 
2373 	return err;
2374 }
2375 
2376 static void enic_iounmap(struct enic *enic)
2377 {
2378 	unsigned int i;
2379 
2380 	for (i = 0; i < ARRAY_SIZE(enic->bar); i++)
2381 		if (enic->bar[i].vaddr)
2382 			iounmap(enic->bar[i].vaddr);
2383 }
2384 
2385 static int enic_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2386 {
2387 	struct device *dev = &pdev->dev;
2388 	struct net_device *netdev;
2389 	struct enic *enic;
2390 	int using_dac = 0;
2391 	unsigned int i;
2392 	int err;
2393 #ifdef CONFIG_PCI_IOV
2394 	int pos = 0;
2395 #endif
2396 	int num_pps = 1;
2397 
2398 	/* Allocate net device structure and initialize.  Private
2399 	 * instance data is initialized to zero.
2400 	 */
2401 
2402 	netdev = alloc_etherdev_mqs(sizeof(struct enic),
2403 				    ENIC_RQ_MAX, ENIC_WQ_MAX);
2404 	if (!netdev)
2405 		return -ENOMEM;
2406 
2407 	pci_set_drvdata(pdev, netdev);
2408 
2409 	SET_NETDEV_DEV(netdev, &pdev->dev);
2410 
2411 	enic = netdev_priv(netdev);
2412 	enic->netdev = netdev;
2413 	enic->pdev = pdev;
2414 
2415 	/* Setup PCI resources
2416 	 */
2417 
2418 	err = pci_enable_device_mem(pdev);
2419 	if (err) {
2420 		dev_err(dev, "Cannot enable PCI device, aborting\n");
2421 		goto err_out_free_netdev;
2422 	}
2423 
2424 	err = pci_request_regions(pdev, DRV_NAME);
2425 	if (err) {
2426 		dev_err(dev, "Cannot request PCI regions, aborting\n");
2427 		goto err_out_disable_device;
2428 	}
2429 
2430 	pci_set_master(pdev);
2431 
2432 	/* Query PCI controller on system for DMA addressing
2433 	 * limitation for the device.  Try 64-bit first, and
2434 	 * fail to 32-bit.
2435 	 */
2436 
2437 	err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
2438 	if (err) {
2439 		err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
2440 		if (err) {
2441 			dev_err(dev, "No usable DMA configuration, aborting\n");
2442 			goto err_out_release_regions;
2443 		}
2444 		err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
2445 		if (err) {
2446 			dev_err(dev, "Unable to obtain %u-bit DMA "
2447 				"for consistent allocations, aborting\n", 32);
2448 			goto err_out_release_regions;
2449 		}
2450 	} else {
2451 		err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
2452 		if (err) {
2453 			dev_err(dev, "Unable to obtain %u-bit DMA "
2454 				"for consistent allocations, aborting\n", 64);
2455 			goto err_out_release_regions;
2456 		}
2457 		using_dac = 1;
2458 	}
2459 
2460 	/* Map vNIC resources from BAR0-5
2461 	 */
2462 
2463 	for (i = 0; i < ARRAY_SIZE(enic->bar); i++) {
2464 		if (!(pci_resource_flags(pdev, i) & IORESOURCE_MEM))
2465 			continue;
2466 		enic->bar[i].len = pci_resource_len(pdev, i);
2467 		enic->bar[i].vaddr = pci_iomap(pdev, i, enic->bar[i].len);
2468 		if (!enic->bar[i].vaddr) {
2469 			dev_err(dev, "Cannot memory-map BAR %d, aborting\n", i);
2470 			err = -ENODEV;
2471 			goto err_out_iounmap;
2472 		}
2473 		enic->bar[i].bus_addr = pci_resource_start(pdev, i);
2474 	}
2475 
2476 	/* Register vNIC device
2477 	 */
2478 
2479 	enic->vdev = vnic_dev_register(NULL, enic, pdev, enic->bar,
2480 		ARRAY_SIZE(enic->bar));
2481 	if (!enic->vdev) {
2482 		dev_err(dev, "vNIC registration failed, aborting\n");
2483 		err = -ENODEV;
2484 		goto err_out_iounmap;
2485 	}
2486 
2487 	err = vnic_devcmd_init(enic->vdev);
2488 
2489 	if (err)
2490 		goto err_out_vnic_unregister;
2491 
2492 #ifdef CONFIG_PCI_IOV
2493 	/* Get number of subvnics */
2494 	pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_SRIOV);
2495 	if (pos) {
2496 		pci_read_config_word(pdev, pos + PCI_SRIOV_TOTAL_VF,
2497 			&enic->num_vfs);
2498 		if (enic->num_vfs) {
2499 			err = pci_enable_sriov(pdev, enic->num_vfs);
2500 			if (err) {
2501 				dev_err(dev, "SRIOV enable failed, aborting."
2502 					" pci_enable_sriov() returned %d\n",
2503 					err);
2504 				goto err_out_vnic_unregister;
2505 			}
2506 			enic->priv_flags |= ENIC_SRIOV_ENABLED;
2507 			num_pps = enic->num_vfs;
2508 		}
2509 	}
2510 #endif
2511 
2512 	/* Allocate structure for port profiles */
2513 	enic->pp = kcalloc(num_pps, sizeof(*enic->pp), GFP_KERNEL);
2514 	if (!enic->pp) {
2515 		err = -ENOMEM;
2516 		goto err_out_disable_sriov_pp;
2517 	}
2518 
2519 	/* Issue device open to get device in known state
2520 	 */
2521 
2522 	err = enic_dev_open(enic);
2523 	if (err) {
2524 		dev_err(dev, "vNIC dev open failed, aborting\n");
2525 		goto err_out_disable_sriov;
2526 	}
2527 
2528 	/* Setup devcmd lock
2529 	 */
2530 
2531 	spin_lock_init(&enic->devcmd_lock);
2532 	spin_lock_init(&enic->enic_api_lock);
2533 
2534 	/*
2535 	 * Set ingress vlan rewrite mode before vnic initialization
2536 	 */
2537 
2538 	err = enic_dev_set_ig_vlan_rewrite_mode(enic);
2539 	if (err) {
2540 		dev_err(dev,
2541 			"Failed to set ingress vlan rewrite mode, aborting.\n");
2542 		goto err_out_dev_close;
2543 	}
2544 
2545 	/* Issue device init to initialize the vnic-to-switch link.
2546 	 * We'll start with carrier off and wait for link UP
2547 	 * notification later to turn on carrier.  We don't need
2548 	 * to wait here for the vnic-to-switch link initialization
2549 	 * to complete; link UP notification is the indication that
2550 	 * the process is complete.
2551 	 */
2552 
2553 	netif_carrier_off(netdev);
2554 
2555 	/* Do not call dev_init for a dynamic vnic.
2556 	 * For a dynamic vnic, init_prov_info will be
2557 	 * called later by an upper layer.
2558 	 */
2559 
2560 	if (!enic_is_dynamic(enic)) {
2561 		err = vnic_dev_init(enic->vdev, 0);
2562 		if (err) {
2563 			dev_err(dev, "vNIC dev init failed, aborting\n");
2564 			goto err_out_dev_close;
2565 		}
2566 	}
2567 
2568 	err = enic_dev_init(enic);
2569 	if (err) {
2570 		dev_err(dev, "Device initialization failed, aborting\n");
2571 		goto err_out_dev_close;
2572 	}
2573 
2574 	netif_set_real_num_tx_queues(netdev, enic->wq_count);
2575 	netif_set_real_num_rx_queues(netdev, enic->rq_count);
2576 
2577 	/* Setup notification timer, HW reset task, and wq locks
2578 	 */
2579 
2580 	init_timer(&enic->notify_timer);
2581 	enic->notify_timer.function = enic_notify_timer;
2582 	enic->notify_timer.data = (unsigned long)enic;
2583 
2584 	enic_set_rx_coal_setting(enic);
2585 	INIT_WORK(&enic->reset, enic_reset);
2586 	INIT_WORK(&enic->change_mtu_work, enic_change_mtu_work);
2587 
2588 	for (i = 0; i < enic->wq_count; i++)
2589 		spin_lock_init(&enic->wq_lock[i]);
2590 
2591 	/* Register net device
2592 	 */
2593 
2594 	enic->port_mtu = enic->config.mtu;
2595 	(void)enic_change_mtu(netdev, enic->port_mtu);
2596 
2597 	err = enic_set_mac_addr(netdev, enic->mac_addr);
2598 	if (err) {
2599 		dev_err(dev, "Invalid MAC address, aborting\n");
2600 		goto err_out_dev_deinit;
2601 	}
2602 
2603 	enic->tx_coalesce_usecs = enic->config.intr_timer_usec;
2604 	/* rx coalesce time already got initialized. This gets used
2605 	 * if adaptive coal is turned off
2606 	 */
2607 	enic->rx_coalesce_usecs = enic->tx_coalesce_usecs;
2608 
2609 	if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic))
2610 		netdev->netdev_ops = &enic_netdev_dynamic_ops;
2611 	else
2612 		netdev->netdev_ops = &enic_netdev_ops;
2613 
2614 	netdev->watchdog_timeo = 2 * HZ;
2615 	enic_set_ethtool_ops(netdev);
2616 
2617 	netdev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
2618 	if (ENIC_SETTING(enic, LOOP)) {
2619 		netdev->features &= ~NETIF_F_HW_VLAN_CTAG_TX;
2620 		enic->loop_enable = 1;
2621 		enic->loop_tag = enic->config.loop_tag;
2622 		dev_info(dev, "loopback tag=0x%04x\n", enic->loop_tag);
2623 	}
2624 	if (ENIC_SETTING(enic, TXCSUM))
2625 		netdev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM;
2626 	if (ENIC_SETTING(enic, TSO))
2627 		netdev->hw_features |= NETIF_F_TSO |
2628 			NETIF_F_TSO6 | NETIF_F_TSO_ECN;
2629 	if (ENIC_SETTING(enic, RSS))
2630 		netdev->hw_features |= NETIF_F_RXHASH;
2631 	if (ENIC_SETTING(enic, RXCSUM))
2632 		netdev->hw_features |= NETIF_F_RXCSUM;
2633 
2634 	netdev->features |= netdev->hw_features;
2635 
2636 #ifdef CONFIG_RFS_ACCEL
2637 	netdev->hw_features |= NETIF_F_NTUPLE;
2638 #endif
2639 
2640 	if (using_dac)
2641 		netdev->features |= NETIF_F_HIGHDMA;
2642 
2643 	netdev->priv_flags |= IFF_UNICAST_FLT;
2644 
2645 	err = register_netdev(netdev);
2646 	if (err) {
2647 		dev_err(dev, "Cannot register net device, aborting\n");
2648 		goto err_out_dev_deinit;
2649 	}
2650 	enic->rx_copybreak = RX_COPYBREAK_DEFAULT;
2651 
2652 	return 0;
2653 
2654 err_out_dev_deinit:
2655 	enic_dev_deinit(enic);
2656 err_out_dev_close:
2657 	vnic_dev_close(enic->vdev);
2658 err_out_disable_sriov:
2659 	kfree(enic->pp);
2660 err_out_disable_sriov_pp:
2661 #ifdef CONFIG_PCI_IOV
2662 	if (enic_sriov_enabled(enic)) {
2663 		pci_disable_sriov(pdev);
2664 		enic->priv_flags &= ~ENIC_SRIOV_ENABLED;
2665 	}
2666 #endif
2667 err_out_vnic_unregister:
2668 	vnic_dev_unregister(enic->vdev);
2669 err_out_iounmap:
2670 	enic_iounmap(enic);
2671 err_out_release_regions:
2672 	pci_release_regions(pdev);
2673 err_out_disable_device:
2674 	pci_disable_device(pdev);
2675 err_out_free_netdev:
2676 	free_netdev(netdev);
2677 
2678 	return err;
2679 }
2680 
2681 static void enic_remove(struct pci_dev *pdev)
2682 {
2683 	struct net_device *netdev = pci_get_drvdata(pdev);
2684 
2685 	if (netdev) {
2686 		struct enic *enic = netdev_priv(netdev);
2687 
2688 		cancel_work_sync(&enic->reset);
2689 		cancel_work_sync(&enic->change_mtu_work);
2690 		unregister_netdev(netdev);
2691 		enic_dev_deinit(enic);
2692 		vnic_dev_close(enic->vdev);
2693 #ifdef CONFIG_PCI_IOV
2694 		if (enic_sriov_enabled(enic)) {
2695 			pci_disable_sriov(pdev);
2696 			enic->priv_flags &= ~ENIC_SRIOV_ENABLED;
2697 		}
2698 #endif
2699 		kfree(enic->pp);
2700 		vnic_dev_unregister(enic->vdev);
2701 		enic_iounmap(enic);
2702 		pci_release_regions(pdev);
2703 		pci_disable_device(pdev);
2704 		free_netdev(netdev);
2705 	}
2706 }
2707 
2708 static struct pci_driver enic_driver = {
2709 	.name = DRV_NAME,
2710 	.id_table = enic_id_table,
2711 	.probe = enic_probe,
2712 	.remove = enic_remove,
2713 };
2714 
2715 static int __init enic_init_module(void)
2716 {
2717 	pr_info("%s, ver %s\n", DRV_DESCRIPTION, DRV_VERSION);
2718 
2719 	return pci_register_driver(&enic_driver);
2720 }
2721 
2722 static void __exit enic_cleanup_module(void)
2723 {
2724 	pci_unregister_driver(&enic_driver);
2725 }
2726 
2727 module_init(enic_init_module);
2728 module_exit(enic_cleanup_module);
2729