1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2015 Cavium, Inc.
4  */
5 
6 #include <linux/module.h>
7 #include <linux/interrupt.h>
8 #include <linux/pci.h>
9 #include <linux/netdevice.h>
10 #include <linux/if_vlan.h>
11 #include <linux/etherdevice.h>
12 #include <linux/ethtool.h>
13 #include <linux/log2.h>
14 #include <linux/prefetch.h>
15 #include <linux/irq.h>
16 #include <linux/iommu.h>
17 #include <linux/bpf.h>
18 #include <linux/bpf_trace.h>
19 #include <linux/filter.h>
20 #include <linux/net_tstamp.h>
21 #include <linux/workqueue.h>
22 
23 #include "nic_reg.h"
24 #include "nic.h"
25 #include "nicvf_queues.h"
26 #include "thunder_bgx.h"
27 #include "../common/cavium_ptp.h"
28 
29 #define DRV_NAME	"nicvf"
30 #define DRV_VERSION	"1.0"
31 
32 /* NOTE: Packets bigger than 1530 are split across multiple pages and XDP needs
33  * the buffer to be contiguous. Allow XDP to be set up only if we don't exceed
34  * this value, keeping headroom for the 14 byte Ethernet header and two
35  * VLAN tags (for QinQ)
36  */
37 #define MAX_XDP_MTU	(1530 - ETH_HLEN - VLAN_HLEN * 2)
38 
39 /* Supported devices */
40 static const struct pci_device_id nicvf_id_table[] = {
41 	{ PCI_DEVICE_SUB(PCI_VENDOR_ID_CAVIUM,
42 			 PCI_DEVICE_ID_THUNDER_NIC_VF,
43 			 PCI_VENDOR_ID_CAVIUM,
44 			 PCI_SUBSYS_DEVID_88XX_NIC_VF) },
45 	{ PCI_DEVICE_SUB(PCI_VENDOR_ID_CAVIUM,
46 			 PCI_DEVICE_ID_THUNDER_PASS1_NIC_VF,
47 			 PCI_VENDOR_ID_CAVIUM,
48 			 PCI_SUBSYS_DEVID_88XX_PASS1_NIC_VF) },
49 	{ PCI_DEVICE_SUB(PCI_VENDOR_ID_CAVIUM,
50 			 PCI_DEVICE_ID_THUNDER_NIC_VF,
51 			 PCI_VENDOR_ID_CAVIUM,
52 			 PCI_SUBSYS_DEVID_81XX_NIC_VF) },
53 	{ PCI_DEVICE_SUB(PCI_VENDOR_ID_CAVIUM,
54 			 PCI_DEVICE_ID_THUNDER_NIC_VF,
55 			 PCI_VENDOR_ID_CAVIUM,
56 			 PCI_SUBSYS_DEVID_83XX_NIC_VF) },
57 	{ 0, }  /* end of table */
58 };
59 
60 MODULE_AUTHOR("Sunil Goutham");
61 MODULE_DESCRIPTION("Cavium Thunder NIC Virtual Function Driver");
62 MODULE_LICENSE("GPL v2");
63 MODULE_VERSION(DRV_VERSION);
64 MODULE_DEVICE_TABLE(pci, nicvf_id_table);
65 
66 static int debug = 0x00;
67 module_param(debug, int, 0644);
68 MODULE_PARM_DESC(debug, "Debug message level bitmap");
69 
70 static int cpi_alg = CPI_ALG_NONE;
71 module_param(cpi_alg, int, 0444);
72 MODULE_PARM_DESC(cpi_alg,
73 		 "PFC algorithm (0=none, 1=VLAN, 2=VLAN16, 3=IP Diffserv)");
74 
75 static inline u8 nicvf_netdev_qidx(struct nicvf *nic, u8 qidx)
76 {
77 	if (nic->sqs_mode)
78 		return qidx + ((nic->sqs_id + 1) * MAX_CMP_QUEUES_PER_QS);
79 	else
80 		return qidx;
81 }
82 
83 /* The Cavium ThunderX network controller can *only* be found in SoCs
84  * containing the ThunderX ARM64 CPU implementation.  All accesses to the device
85  * registers on this platform are implicitly strongly ordered with respect
86  * to memory accesses. So writeq_relaxed() and readq_relaxed() are safe to use
87  * with no memory barriers in this driver.  The readq()/writeq() functions add
88  * explicit ordering operation which in this case are redundant, and only
89  * add overhead.
90  */
91 
92 /* Register read/write APIs */
93 void nicvf_reg_write(struct nicvf *nic, u64 offset, u64 val)
94 {
95 	writeq_relaxed(val, nic->reg_base + offset);
96 }
97 
98 u64 nicvf_reg_read(struct nicvf *nic, u64 offset)
99 {
100 	return readq_relaxed(nic->reg_base + offset);
101 }
102 
103 void nicvf_queue_reg_write(struct nicvf *nic, u64 offset,
104 			   u64 qidx, u64 val)
105 {
106 	void __iomem *addr = nic->reg_base + offset;
107 
108 	writeq_relaxed(val, addr + (qidx << NIC_Q_NUM_SHIFT));
109 }
110 
111 u64 nicvf_queue_reg_read(struct nicvf *nic, u64 offset, u64 qidx)
112 {
113 	void __iomem *addr = nic->reg_base + offset;
114 
115 	return readq_relaxed(addr + (qidx << NIC_Q_NUM_SHIFT));
116 }
117 
118 /* VF -> PF mailbox communication */
119 static void nicvf_write_to_mbx(struct nicvf *nic, union nic_mbx *mbx)
120 {
121 	u64 *msg = (u64 *)mbx;
122 
123 	nicvf_reg_write(nic, NIC_VF_PF_MAILBOX_0_1 + 0, msg[0]);
124 	nicvf_reg_write(nic, NIC_VF_PF_MAILBOX_0_1 + 8, msg[1]);
125 }
126 
127 int nicvf_send_msg_to_pf(struct nicvf *nic, union nic_mbx *mbx)
128 {
129 	unsigned long timeout;
130 	int ret = 0;
131 
132 	mutex_lock(&nic->rx_mode_mtx);
133 
134 	nic->pf_acked = false;
135 	nic->pf_nacked = false;
136 
137 	nicvf_write_to_mbx(nic, mbx);
138 
139 	timeout = jiffies + msecs_to_jiffies(NIC_MBOX_MSG_TIMEOUT);
140 	/* Wait for previous message to be acked, timeout 2sec */
141 	while (!nic->pf_acked) {
142 		if (nic->pf_nacked) {
143 			netdev_err(nic->netdev,
144 				   "PF NACK to mbox msg 0x%02x from VF%d\n",
145 				   (mbx->msg.msg & 0xFF), nic->vf_id);
146 			ret = -EINVAL;
147 			break;
148 		}
149 		usleep_range(8000, 10000);
150 		if (nic->pf_acked)
151 			break;
152 		if (time_after(jiffies, timeout)) {
153 			netdev_err(nic->netdev,
154 				   "PF didn't ACK to mbox msg 0x%02x from VF%d\n",
155 				   (mbx->msg.msg & 0xFF), nic->vf_id);
156 			ret = -EBUSY;
157 			break;
158 		}
159 	}
160 	mutex_unlock(&nic->rx_mode_mtx);
161 	return ret;
162 }
163 
164 /* Checks if VF is able to comminicate with PF
165 * and also gets the VNIC number this VF is associated to.
166 */
167 static int nicvf_check_pf_ready(struct nicvf *nic)
168 {
169 	union nic_mbx mbx = {};
170 
171 	mbx.msg.msg = NIC_MBOX_MSG_READY;
172 	if (nicvf_send_msg_to_pf(nic, &mbx)) {
173 		netdev_err(nic->netdev,
174 			   "PF didn't respond to READY msg\n");
175 		return 0;
176 	}
177 
178 	return 1;
179 }
180 
181 static void nicvf_send_cfg_done(struct nicvf *nic)
182 {
183 	union nic_mbx mbx = {};
184 
185 	mbx.msg.msg = NIC_MBOX_MSG_CFG_DONE;
186 	if (nicvf_send_msg_to_pf(nic, &mbx)) {
187 		netdev_err(nic->netdev,
188 			   "PF didn't respond to CFG DONE msg\n");
189 	}
190 }
191 
192 static void nicvf_read_bgx_stats(struct nicvf *nic, struct bgx_stats_msg *bgx)
193 {
194 	if (bgx->rx)
195 		nic->bgx_stats.rx_stats[bgx->idx] = bgx->stats;
196 	else
197 		nic->bgx_stats.tx_stats[bgx->idx] = bgx->stats;
198 }
199 
200 static void  nicvf_handle_mbx_intr(struct nicvf *nic)
201 {
202 	union nic_mbx mbx = {};
203 	u64 *mbx_data;
204 	u64 mbx_addr;
205 	int i;
206 
207 	mbx_addr = NIC_VF_PF_MAILBOX_0_1;
208 	mbx_data = (u64 *)&mbx;
209 
210 	for (i = 0; i < NIC_PF_VF_MAILBOX_SIZE; i++) {
211 		*mbx_data = nicvf_reg_read(nic, mbx_addr);
212 		mbx_data++;
213 		mbx_addr += sizeof(u64);
214 	}
215 
216 	netdev_dbg(nic->netdev, "Mbox message: msg: 0x%x\n", mbx.msg.msg);
217 	switch (mbx.msg.msg) {
218 	case NIC_MBOX_MSG_READY:
219 		nic->pf_acked = true;
220 		nic->vf_id = mbx.nic_cfg.vf_id & 0x7F;
221 		nic->tns_mode = mbx.nic_cfg.tns_mode & 0x7F;
222 		nic->node = mbx.nic_cfg.node_id;
223 		if (!nic->set_mac_pending)
224 			ether_addr_copy(nic->netdev->dev_addr,
225 					mbx.nic_cfg.mac_addr);
226 		nic->sqs_mode = mbx.nic_cfg.sqs_mode;
227 		nic->loopback_supported = mbx.nic_cfg.loopback_supported;
228 		nic->link_up = false;
229 		nic->duplex = 0;
230 		nic->speed = 0;
231 		break;
232 	case NIC_MBOX_MSG_ACK:
233 		nic->pf_acked = true;
234 		break;
235 	case NIC_MBOX_MSG_NACK:
236 		nic->pf_nacked = true;
237 		break;
238 	case NIC_MBOX_MSG_RSS_SIZE:
239 		nic->rss_info.rss_size = mbx.rss_size.ind_tbl_size;
240 		nic->pf_acked = true;
241 		break;
242 	case NIC_MBOX_MSG_BGX_STATS:
243 		nicvf_read_bgx_stats(nic, &mbx.bgx_stats);
244 		nic->pf_acked = true;
245 		break;
246 	case NIC_MBOX_MSG_BGX_LINK_CHANGE:
247 		nic->pf_acked = true;
248 		if (nic->link_up != mbx.link_status.link_up) {
249 			nic->link_up = mbx.link_status.link_up;
250 			nic->duplex = mbx.link_status.duplex;
251 			nic->speed = mbx.link_status.speed;
252 			nic->mac_type = mbx.link_status.mac_type;
253 			if (nic->link_up) {
254 				netdev_info(nic->netdev,
255 					    "Link is Up %d Mbps %s duplex\n",
256 					    nic->speed,
257 					    nic->duplex == DUPLEX_FULL ?
258 					    "Full" : "Half");
259 				netif_carrier_on(nic->netdev);
260 				netif_tx_start_all_queues(nic->netdev);
261 			} else {
262 				netdev_info(nic->netdev, "Link is Down\n");
263 				netif_carrier_off(nic->netdev);
264 				netif_tx_stop_all_queues(nic->netdev);
265 			}
266 		}
267 		break;
268 	case NIC_MBOX_MSG_ALLOC_SQS:
269 		nic->sqs_count = mbx.sqs_alloc.qs_count;
270 		nic->pf_acked = true;
271 		break;
272 	case NIC_MBOX_MSG_SNICVF_PTR:
273 		/* Primary VF: make note of secondary VF's pointer
274 		 * to be used while packet transmission.
275 		 */
276 		nic->snicvf[mbx.nicvf.sqs_id] =
277 			(struct nicvf *)mbx.nicvf.nicvf;
278 		nic->pf_acked = true;
279 		break;
280 	case NIC_MBOX_MSG_PNICVF_PTR:
281 		/* Secondary VF/Qset: make note of primary VF's pointer
282 		 * to be used while packet reception, to handover packet
283 		 * to primary VF's netdev.
284 		 */
285 		nic->pnicvf = (struct nicvf *)mbx.nicvf.nicvf;
286 		nic->pf_acked = true;
287 		break;
288 	case NIC_MBOX_MSG_PFC:
289 		nic->pfc.autoneg = mbx.pfc.autoneg;
290 		nic->pfc.fc_rx = mbx.pfc.fc_rx;
291 		nic->pfc.fc_tx = mbx.pfc.fc_tx;
292 		nic->pf_acked = true;
293 		break;
294 	default:
295 		netdev_err(nic->netdev,
296 			   "Invalid message from PF, msg 0x%x\n", mbx.msg.msg);
297 		break;
298 	}
299 	nicvf_clear_intr(nic, NICVF_INTR_MBOX, 0);
300 }
301 
302 static int nicvf_hw_set_mac_addr(struct nicvf *nic, struct net_device *netdev)
303 {
304 	union nic_mbx mbx = {};
305 
306 	mbx.mac.msg = NIC_MBOX_MSG_SET_MAC;
307 	mbx.mac.vf_id = nic->vf_id;
308 	ether_addr_copy(mbx.mac.mac_addr, netdev->dev_addr);
309 
310 	return nicvf_send_msg_to_pf(nic, &mbx);
311 }
312 
313 static void nicvf_config_cpi(struct nicvf *nic)
314 {
315 	union nic_mbx mbx = {};
316 
317 	mbx.cpi_cfg.msg = NIC_MBOX_MSG_CPI_CFG;
318 	mbx.cpi_cfg.vf_id = nic->vf_id;
319 	mbx.cpi_cfg.cpi_alg = nic->cpi_alg;
320 	mbx.cpi_cfg.rq_cnt = nic->qs->rq_cnt;
321 
322 	nicvf_send_msg_to_pf(nic, &mbx);
323 }
324 
325 static void nicvf_get_rss_size(struct nicvf *nic)
326 {
327 	union nic_mbx mbx = {};
328 
329 	mbx.rss_size.msg = NIC_MBOX_MSG_RSS_SIZE;
330 	mbx.rss_size.vf_id = nic->vf_id;
331 	nicvf_send_msg_to_pf(nic, &mbx);
332 }
333 
334 void nicvf_config_rss(struct nicvf *nic)
335 {
336 	union nic_mbx mbx = {};
337 	struct nicvf_rss_info *rss = &nic->rss_info;
338 	int ind_tbl_len = rss->rss_size;
339 	int i, nextq = 0;
340 
341 	mbx.rss_cfg.vf_id = nic->vf_id;
342 	mbx.rss_cfg.hash_bits = rss->hash_bits;
343 	while (ind_tbl_len) {
344 		mbx.rss_cfg.tbl_offset = nextq;
345 		mbx.rss_cfg.tbl_len = min(ind_tbl_len,
346 					       RSS_IND_TBL_LEN_PER_MBX_MSG);
347 		mbx.rss_cfg.msg = mbx.rss_cfg.tbl_offset ?
348 			  NIC_MBOX_MSG_RSS_CFG_CONT : NIC_MBOX_MSG_RSS_CFG;
349 
350 		for (i = 0; i < mbx.rss_cfg.tbl_len; i++)
351 			mbx.rss_cfg.ind_tbl[i] = rss->ind_tbl[nextq++];
352 
353 		nicvf_send_msg_to_pf(nic, &mbx);
354 
355 		ind_tbl_len -= mbx.rss_cfg.tbl_len;
356 	}
357 }
358 
359 void nicvf_set_rss_key(struct nicvf *nic)
360 {
361 	struct nicvf_rss_info *rss = &nic->rss_info;
362 	u64 key_addr = NIC_VNIC_RSS_KEY_0_4;
363 	int idx;
364 
365 	for (idx = 0; idx < RSS_HASH_KEY_SIZE; idx++) {
366 		nicvf_reg_write(nic, key_addr, rss->key[idx]);
367 		key_addr += sizeof(u64);
368 	}
369 }
370 
371 static int nicvf_rss_init(struct nicvf *nic)
372 {
373 	struct nicvf_rss_info *rss = &nic->rss_info;
374 	int idx;
375 
376 	nicvf_get_rss_size(nic);
377 
378 	if (cpi_alg != CPI_ALG_NONE) {
379 		rss->enable = false;
380 		rss->hash_bits = 0;
381 		return 0;
382 	}
383 
384 	rss->enable = true;
385 
386 	netdev_rss_key_fill(rss->key, RSS_HASH_KEY_SIZE * sizeof(u64));
387 	nicvf_set_rss_key(nic);
388 
389 	rss->cfg = RSS_IP_HASH_ENA | RSS_TCP_HASH_ENA | RSS_UDP_HASH_ENA;
390 	nicvf_reg_write(nic, NIC_VNIC_RSS_CFG, rss->cfg);
391 
392 	rss->hash_bits =  ilog2(rounddown_pow_of_two(rss->rss_size));
393 
394 	for (idx = 0; idx < rss->rss_size; idx++)
395 		rss->ind_tbl[idx] = ethtool_rxfh_indir_default(idx,
396 							       nic->rx_queues);
397 	nicvf_config_rss(nic);
398 	return 1;
399 }
400 
401 /* Request PF to allocate additional Qsets */
402 static void nicvf_request_sqs(struct nicvf *nic)
403 {
404 	union nic_mbx mbx = {};
405 	int sqs;
406 	int sqs_count = nic->sqs_count;
407 	int rx_queues = 0, tx_queues = 0;
408 
409 	/* Only primary VF should request */
410 	if (nic->sqs_mode ||  !nic->sqs_count)
411 		return;
412 
413 	mbx.sqs_alloc.msg = NIC_MBOX_MSG_ALLOC_SQS;
414 	mbx.sqs_alloc.vf_id = nic->vf_id;
415 	mbx.sqs_alloc.qs_count = nic->sqs_count;
416 	if (nicvf_send_msg_to_pf(nic, &mbx)) {
417 		/* No response from PF */
418 		nic->sqs_count = 0;
419 		return;
420 	}
421 
422 	/* Return if no Secondary Qsets available */
423 	if (!nic->sqs_count)
424 		return;
425 
426 	if (nic->rx_queues > MAX_RCV_QUEUES_PER_QS)
427 		rx_queues = nic->rx_queues - MAX_RCV_QUEUES_PER_QS;
428 
429 	tx_queues = nic->tx_queues + nic->xdp_tx_queues;
430 	if (tx_queues > MAX_SND_QUEUES_PER_QS)
431 		tx_queues = tx_queues - MAX_SND_QUEUES_PER_QS;
432 
433 	/* Set no of Rx/Tx queues in each of the SQsets */
434 	for (sqs = 0; sqs < nic->sqs_count; sqs++) {
435 		mbx.nicvf.msg = NIC_MBOX_MSG_SNICVF_PTR;
436 		mbx.nicvf.vf_id = nic->vf_id;
437 		mbx.nicvf.sqs_id = sqs;
438 		nicvf_send_msg_to_pf(nic, &mbx);
439 
440 		nic->snicvf[sqs]->sqs_id = sqs;
441 		if (rx_queues > MAX_RCV_QUEUES_PER_QS) {
442 			nic->snicvf[sqs]->qs->rq_cnt = MAX_RCV_QUEUES_PER_QS;
443 			rx_queues -= MAX_RCV_QUEUES_PER_QS;
444 		} else {
445 			nic->snicvf[sqs]->qs->rq_cnt = rx_queues;
446 			rx_queues = 0;
447 		}
448 
449 		if (tx_queues > MAX_SND_QUEUES_PER_QS) {
450 			nic->snicvf[sqs]->qs->sq_cnt = MAX_SND_QUEUES_PER_QS;
451 			tx_queues -= MAX_SND_QUEUES_PER_QS;
452 		} else {
453 			nic->snicvf[sqs]->qs->sq_cnt = tx_queues;
454 			tx_queues = 0;
455 		}
456 
457 		nic->snicvf[sqs]->qs->cq_cnt =
458 		max(nic->snicvf[sqs]->qs->rq_cnt, nic->snicvf[sqs]->qs->sq_cnt);
459 
460 		/* Initialize secondary Qset's queues and its interrupts */
461 		nicvf_open(nic->snicvf[sqs]->netdev);
462 	}
463 
464 	/* Update stack with actual Rx/Tx queue count allocated */
465 	if (sqs_count != nic->sqs_count)
466 		nicvf_set_real_num_queues(nic->netdev,
467 					  nic->tx_queues, nic->rx_queues);
468 }
469 
470 /* Send this Qset's nicvf pointer to PF.
471  * PF inturn sends primary VF's nicvf struct to secondary Qsets/VFs
472  * so that packets received by these Qsets can use primary VF's netdev
473  */
474 static void nicvf_send_vf_struct(struct nicvf *nic)
475 {
476 	union nic_mbx mbx = {};
477 
478 	mbx.nicvf.msg = NIC_MBOX_MSG_NICVF_PTR;
479 	mbx.nicvf.sqs_mode = nic->sqs_mode;
480 	mbx.nicvf.nicvf = (u64)nic;
481 	nicvf_send_msg_to_pf(nic, &mbx);
482 }
483 
484 static void nicvf_get_primary_vf_struct(struct nicvf *nic)
485 {
486 	union nic_mbx mbx = {};
487 
488 	mbx.nicvf.msg = NIC_MBOX_MSG_PNICVF_PTR;
489 	nicvf_send_msg_to_pf(nic, &mbx);
490 }
491 
492 int nicvf_set_real_num_queues(struct net_device *netdev,
493 			      int tx_queues, int rx_queues)
494 {
495 	int err = 0;
496 
497 	err = netif_set_real_num_tx_queues(netdev, tx_queues);
498 	if (err) {
499 		netdev_err(netdev,
500 			   "Failed to set no of Tx queues: %d\n", tx_queues);
501 		return err;
502 	}
503 
504 	err = netif_set_real_num_rx_queues(netdev, rx_queues);
505 	if (err)
506 		netdev_err(netdev,
507 			   "Failed to set no of Rx queues: %d\n", rx_queues);
508 	return err;
509 }
510 
511 static int nicvf_init_resources(struct nicvf *nic)
512 {
513 	int err;
514 
515 	/* Enable Qset */
516 	nicvf_qset_config(nic, true);
517 
518 	/* Initialize queues and HW for data transfer */
519 	err = nicvf_config_data_transfer(nic, true);
520 	if (err) {
521 		netdev_err(nic->netdev,
522 			   "Failed to alloc/config VF's QSet resources\n");
523 		return err;
524 	}
525 
526 	return 0;
527 }
528 
529 static inline bool nicvf_xdp_rx(struct nicvf *nic, struct bpf_prog *prog,
530 				struct cqe_rx_t *cqe_rx, struct snd_queue *sq,
531 				struct rcv_queue *rq, struct sk_buff **skb)
532 {
533 	struct xdp_buff xdp;
534 	struct page *page;
535 	u32 action;
536 	u16 len, offset = 0;
537 	u64 dma_addr, cpu_addr;
538 	void *orig_data;
539 
540 	/* Retrieve packet buffer's DMA address and length */
541 	len = *((u16 *)((void *)cqe_rx + (3 * sizeof(u64))));
542 	dma_addr = *((u64 *)((void *)cqe_rx + (7 * sizeof(u64))));
543 
544 	cpu_addr = nicvf_iova_to_phys(nic, dma_addr);
545 	if (!cpu_addr)
546 		return false;
547 	cpu_addr = (u64)phys_to_virt(cpu_addr);
548 	page = virt_to_page((void *)cpu_addr);
549 
550 	xdp.data_hard_start = page_address(page);
551 	xdp.data = (void *)cpu_addr;
552 	xdp_set_data_meta_invalid(&xdp);
553 	xdp.data_end = xdp.data + len;
554 	xdp.rxq = &rq->xdp_rxq;
555 	xdp.frame_sz = RCV_FRAG_LEN + XDP_PACKET_HEADROOM;
556 	orig_data = xdp.data;
557 
558 	rcu_read_lock();
559 	action = bpf_prog_run_xdp(prog, &xdp);
560 	rcu_read_unlock();
561 
562 	len = xdp.data_end - xdp.data;
563 	/* Check if XDP program has changed headers */
564 	if (orig_data != xdp.data) {
565 		offset = orig_data - xdp.data;
566 		dma_addr -= offset;
567 	}
568 
569 	switch (action) {
570 	case XDP_PASS:
571 		/* Check if it's a recycled page, if not
572 		 * unmap the DMA mapping.
573 		 *
574 		 * Recycled page holds an extra reference.
575 		 */
576 		if (page_ref_count(page) == 1) {
577 			dma_addr &= PAGE_MASK;
578 			dma_unmap_page_attrs(&nic->pdev->dev, dma_addr,
579 					     RCV_FRAG_LEN + XDP_PACKET_HEADROOM,
580 					     DMA_FROM_DEVICE,
581 					     DMA_ATTR_SKIP_CPU_SYNC);
582 		}
583 
584 		/* Build SKB and pass on packet to network stack */
585 		*skb = build_skb(xdp.data,
586 				 RCV_FRAG_LEN - cqe_rx->align_pad + offset);
587 		if (!*skb)
588 			put_page(page);
589 		else
590 			skb_put(*skb, len);
591 		return false;
592 	case XDP_TX:
593 		nicvf_xdp_sq_append_pkt(nic, sq, (u64)xdp.data, dma_addr, len);
594 		return true;
595 	default:
596 		bpf_warn_invalid_xdp_action(action);
597 		fallthrough;
598 	case XDP_ABORTED:
599 		trace_xdp_exception(nic->netdev, prog, action);
600 		fallthrough;
601 	case XDP_DROP:
602 		/* Check if it's a recycled page, if not
603 		 * unmap the DMA mapping.
604 		 *
605 		 * Recycled page holds an extra reference.
606 		 */
607 		if (page_ref_count(page) == 1) {
608 			dma_addr &= PAGE_MASK;
609 			dma_unmap_page_attrs(&nic->pdev->dev, dma_addr,
610 					     RCV_FRAG_LEN + XDP_PACKET_HEADROOM,
611 					     DMA_FROM_DEVICE,
612 					     DMA_ATTR_SKIP_CPU_SYNC);
613 		}
614 		put_page(page);
615 		return true;
616 	}
617 	return false;
618 }
619 
620 static void nicvf_snd_ptp_handler(struct net_device *netdev,
621 				  struct cqe_send_t *cqe_tx)
622 {
623 	struct nicvf *nic = netdev_priv(netdev);
624 	struct skb_shared_hwtstamps ts;
625 	u64 ns;
626 
627 	nic = nic->pnicvf;
628 
629 	/* Sync for 'ptp_skb' */
630 	smp_rmb();
631 
632 	/* New timestamp request can be queued now */
633 	atomic_set(&nic->tx_ptp_skbs, 0);
634 
635 	/* Check for timestamp requested skb */
636 	if (!nic->ptp_skb)
637 		return;
638 
639 	/* Check if timestamping is timedout, which is set to 10us */
640 	if (cqe_tx->send_status == CQ_TX_ERROP_TSTMP_TIMEOUT ||
641 	    cqe_tx->send_status == CQ_TX_ERROP_TSTMP_CONFLICT)
642 		goto no_tstamp;
643 
644 	/* Get the timestamp */
645 	memset(&ts, 0, sizeof(ts));
646 	ns = cavium_ptp_tstamp2time(nic->ptp_clock, cqe_tx->ptp_timestamp);
647 	ts.hwtstamp = ns_to_ktime(ns);
648 	skb_tstamp_tx(nic->ptp_skb, &ts);
649 
650 no_tstamp:
651 	/* Free the original skb */
652 	dev_kfree_skb_any(nic->ptp_skb);
653 	nic->ptp_skb = NULL;
654 	/* Sync 'ptp_skb' */
655 	smp_wmb();
656 }
657 
658 static void nicvf_snd_pkt_handler(struct net_device *netdev,
659 				  struct cqe_send_t *cqe_tx,
660 				  int budget, int *subdesc_cnt,
661 				  unsigned int *tx_pkts, unsigned int *tx_bytes)
662 {
663 	struct sk_buff *skb = NULL;
664 	struct page *page;
665 	struct nicvf *nic = netdev_priv(netdev);
666 	struct snd_queue *sq;
667 	struct sq_hdr_subdesc *hdr;
668 	struct sq_hdr_subdesc *tso_sqe;
669 
670 	sq = &nic->qs->sq[cqe_tx->sq_idx];
671 
672 	hdr = (struct sq_hdr_subdesc *)GET_SQ_DESC(sq, cqe_tx->sqe_ptr);
673 	if (hdr->subdesc_type != SQ_DESC_TYPE_HEADER)
674 		return;
675 
676 	/* Check for errors */
677 	if (cqe_tx->send_status)
678 		nicvf_check_cqe_tx_errs(nic->pnicvf, cqe_tx);
679 
680 	/* Is this a XDP designated Tx queue */
681 	if (sq->is_xdp) {
682 		page = (struct page *)sq->xdp_page[cqe_tx->sqe_ptr];
683 		/* Check if it's recycled page or else unmap DMA mapping */
684 		if (page && (page_ref_count(page) == 1))
685 			nicvf_unmap_sndq_buffers(nic, sq, cqe_tx->sqe_ptr,
686 						 hdr->subdesc_cnt);
687 
688 		/* Release page reference for recycling */
689 		if (page)
690 			put_page(page);
691 		sq->xdp_page[cqe_tx->sqe_ptr] = (u64)NULL;
692 		*subdesc_cnt += hdr->subdesc_cnt + 1;
693 		return;
694 	}
695 
696 	skb = (struct sk_buff *)sq->skbuff[cqe_tx->sqe_ptr];
697 	if (skb) {
698 		/* Check for dummy descriptor used for HW TSO offload on 88xx */
699 		if (hdr->dont_send) {
700 			/* Get actual TSO descriptors and free them */
701 			tso_sqe =
702 			 (struct sq_hdr_subdesc *)GET_SQ_DESC(sq, hdr->rsvd2);
703 			nicvf_unmap_sndq_buffers(nic, sq, hdr->rsvd2,
704 						 tso_sqe->subdesc_cnt);
705 			*subdesc_cnt += tso_sqe->subdesc_cnt + 1;
706 		} else {
707 			nicvf_unmap_sndq_buffers(nic, sq, cqe_tx->sqe_ptr,
708 						 hdr->subdesc_cnt);
709 		}
710 		*subdesc_cnt += hdr->subdesc_cnt + 1;
711 		prefetch(skb);
712 		(*tx_pkts)++;
713 		*tx_bytes += skb->len;
714 		/* If timestamp is requested for this skb, don't free it */
715 		if (skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS &&
716 		    !nic->pnicvf->ptp_skb)
717 			nic->pnicvf->ptp_skb = skb;
718 		else
719 			napi_consume_skb(skb, budget);
720 		sq->skbuff[cqe_tx->sqe_ptr] = (u64)NULL;
721 	} else {
722 		/* In case of SW TSO on 88xx, only last segment will have
723 		 * a SKB attached, so just free SQEs here.
724 		 */
725 		if (!nic->hw_tso)
726 			*subdesc_cnt += hdr->subdesc_cnt + 1;
727 	}
728 }
729 
730 static inline void nicvf_set_rxhash(struct net_device *netdev,
731 				    struct cqe_rx_t *cqe_rx,
732 				    struct sk_buff *skb)
733 {
734 	u8 hash_type;
735 	u32 hash;
736 
737 	if (!(netdev->features & NETIF_F_RXHASH))
738 		return;
739 
740 	switch (cqe_rx->rss_alg) {
741 	case RSS_ALG_TCP_IP:
742 	case RSS_ALG_UDP_IP:
743 		hash_type = PKT_HASH_TYPE_L4;
744 		hash = cqe_rx->rss_tag;
745 		break;
746 	case RSS_ALG_IP:
747 		hash_type = PKT_HASH_TYPE_L3;
748 		hash = cqe_rx->rss_tag;
749 		break;
750 	default:
751 		hash_type = PKT_HASH_TYPE_NONE;
752 		hash = 0;
753 	}
754 
755 	skb_set_hash(skb, hash, hash_type);
756 }
757 
758 static inline void nicvf_set_rxtstamp(struct nicvf *nic, struct sk_buff *skb)
759 {
760 	u64 ns;
761 
762 	if (!nic->ptp_clock || !nic->hw_rx_tstamp)
763 		return;
764 
765 	/* The first 8 bytes is the timestamp */
766 	ns = cavium_ptp_tstamp2time(nic->ptp_clock,
767 				    be64_to_cpu(*(__be64 *)skb->data));
768 	skb_hwtstamps(skb)->hwtstamp = ns_to_ktime(ns);
769 
770 	__skb_pull(skb, 8);
771 }
772 
773 static void nicvf_rcv_pkt_handler(struct net_device *netdev,
774 				  struct napi_struct *napi,
775 				  struct cqe_rx_t *cqe_rx,
776 				  struct snd_queue *sq, struct rcv_queue *rq)
777 {
778 	struct sk_buff *skb = NULL;
779 	struct nicvf *nic = netdev_priv(netdev);
780 	struct nicvf *snic = nic;
781 	int err = 0;
782 	int rq_idx;
783 
784 	rq_idx = nicvf_netdev_qidx(nic, cqe_rx->rq_idx);
785 
786 	if (nic->sqs_mode) {
787 		/* Use primary VF's 'nicvf' struct */
788 		nic = nic->pnicvf;
789 		netdev = nic->netdev;
790 	}
791 
792 	/* Check for errors */
793 	if (cqe_rx->err_level || cqe_rx->err_opcode) {
794 		err = nicvf_check_cqe_rx_errs(nic, cqe_rx);
795 		if (err && !cqe_rx->rb_cnt)
796 			return;
797 	}
798 
799 	/* For XDP, ignore pkts spanning multiple pages */
800 	if (nic->xdp_prog && (cqe_rx->rb_cnt == 1)) {
801 		/* Packet consumed by XDP */
802 		if (nicvf_xdp_rx(snic, nic->xdp_prog, cqe_rx, sq, rq, &skb))
803 			return;
804 	} else {
805 		skb = nicvf_get_rcv_skb(snic, cqe_rx,
806 					nic->xdp_prog ? true : false);
807 	}
808 
809 	if (!skb)
810 		return;
811 
812 	if (netif_msg_pktdata(nic)) {
813 		netdev_info(nic->netdev, "skb 0x%p, len=%d\n", skb, skb->len);
814 		print_hex_dump(KERN_INFO, "", DUMP_PREFIX_OFFSET, 16, 1,
815 			       skb->data, skb->len, true);
816 	}
817 
818 	/* If error packet, drop it here */
819 	if (err) {
820 		dev_kfree_skb_any(skb);
821 		return;
822 	}
823 
824 	nicvf_set_rxtstamp(nic, skb);
825 	nicvf_set_rxhash(netdev, cqe_rx, skb);
826 
827 	skb_record_rx_queue(skb, rq_idx);
828 	if (netdev->hw_features & NETIF_F_RXCSUM) {
829 		/* HW by default verifies TCP/UDP/SCTP checksums */
830 		skb->ip_summed = CHECKSUM_UNNECESSARY;
831 	} else {
832 		skb_checksum_none_assert(skb);
833 	}
834 
835 	skb->protocol = eth_type_trans(skb, netdev);
836 
837 	/* Check for stripped VLAN */
838 	if (cqe_rx->vlan_found && cqe_rx->vlan_stripped)
839 		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
840 				       ntohs((__force __be16)cqe_rx->vlan_tci));
841 
842 	if (napi && (netdev->features & NETIF_F_GRO))
843 		napi_gro_receive(napi, skb);
844 	else
845 		netif_receive_skb(skb);
846 }
847 
848 static int nicvf_cq_intr_handler(struct net_device *netdev, u8 cq_idx,
849 				 struct napi_struct *napi, int budget)
850 {
851 	int processed_cqe, work_done = 0, tx_done = 0;
852 	int cqe_count, cqe_head;
853 	int subdesc_cnt = 0;
854 	struct nicvf *nic = netdev_priv(netdev);
855 	struct queue_set *qs = nic->qs;
856 	struct cmp_queue *cq = &qs->cq[cq_idx];
857 	struct cqe_rx_t *cq_desc;
858 	struct netdev_queue *txq;
859 	struct snd_queue *sq = &qs->sq[cq_idx];
860 	struct rcv_queue *rq = &qs->rq[cq_idx];
861 	unsigned int tx_pkts = 0, tx_bytes = 0, txq_idx;
862 
863 	spin_lock_bh(&cq->lock);
864 loop:
865 	processed_cqe = 0;
866 	/* Get no of valid CQ entries to process */
867 	cqe_count = nicvf_queue_reg_read(nic, NIC_QSET_CQ_0_7_STATUS, cq_idx);
868 	cqe_count &= CQ_CQE_COUNT;
869 	if (!cqe_count)
870 		goto done;
871 
872 	/* Get head of the valid CQ entries */
873 	cqe_head = nicvf_queue_reg_read(nic, NIC_QSET_CQ_0_7_HEAD, cq_idx) >> 9;
874 	cqe_head &= 0xFFFF;
875 
876 	while (processed_cqe < cqe_count) {
877 		/* Get the CQ descriptor */
878 		cq_desc = (struct cqe_rx_t *)GET_CQ_DESC(cq, cqe_head);
879 		cqe_head++;
880 		cqe_head &= (cq->dmem.q_len - 1);
881 		/* Initiate prefetch for next descriptor */
882 		prefetch((struct cqe_rx_t *)GET_CQ_DESC(cq, cqe_head));
883 
884 		if ((work_done >= budget) && napi &&
885 		    (cq_desc->cqe_type != CQE_TYPE_SEND)) {
886 			break;
887 		}
888 
889 		switch (cq_desc->cqe_type) {
890 		case CQE_TYPE_RX:
891 			nicvf_rcv_pkt_handler(netdev, napi, cq_desc, sq, rq);
892 			work_done++;
893 		break;
894 		case CQE_TYPE_SEND:
895 			nicvf_snd_pkt_handler(netdev, (void *)cq_desc,
896 					      budget, &subdesc_cnt,
897 					      &tx_pkts, &tx_bytes);
898 			tx_done++;
899 		break;
900 		case CQE_TYPE_SEND_PTP:
901 			nicvf_snd_ptp_handler(netdev, (void *)cq_desc);
902 		break;
903 		case CQE_TYPE_INVALID:
904 		case CQE_TYPE_RX_SPLIT:
905 		case CQE_TYPE_RX_TCP:
906 			/* Ignore for now */
907 		break;
908 		}
909 		processed_cqe++;
910 	}
911 
912 	/* Ring doorbell to inform H/W to reuse processed CQEs */
913 	nicvf_queue_reg_write(nic, NIC_QSET_CQ_0_7_DOOR,
914 			      cq_idx, processed_cqe);
915 
916 	if ((work_done < budget) && napi)
917 		goto loop;
918 
919 done:
920 	/* Update SQ's descriptor free count */
921 	if (subdesc_cnt)
922 		nicvf_put_sq_desc(sq, subdesc_cnt);
923 
924 	txq_idx = nicvf_netdev_qidx(nic, cq_idx);
925 	/* Handle XDP TX queues */
926 	if (nic->pnicvf->xdp_prog) {
927 		if (txq_idx < nic->pnicvf->xdp_tx_queues) {
928 			nicvf_xdp_sq_doorbell(nic, sq, cq_idx);
929 			goto out;
930 		}
931 		nic = nic->pnicvf;
932 		txq_idx -= nic->pnicvf->xdp_tx_queues;
933 	}
934 
935 	/* Wakeup TXQ if its stopped earlier due to SQ full */
936 	if (tx_done ||
937 	    (atomic_read(&sq->free_cnt) >= MIN_SQ_DESC_PER_PKT_XMIT)) {
938 		netdev = nic->pnicvf->netdev;
939 		txq = netdev_get_tx_queue(netdev, txq_idx);
940 		if (tx_pkts)
941 			netdev_tx_completed_queue(txq, tx_pkts, tx_bytes);
942 
943 		/* To read updated queue and carrier status */
944 		smp_mb();
945 		if (netif_tx_queue_stopped(txq) && netif_carrier_ok(netdev)) {
946 			netif_tx_wake_queue(txq);
947 			nic = nic->pnicvf;
948 			this_cpu_inc(nic->drv_stats->txq_wake);
949 			netif_warn(nic, tx_err, netdev,
950 				   "Transmit queue wakeup SQ%d\n", txq_idx);
951 		}
952 	}
953 
954 out:
955 	spin_unlock_bh(&cq->lock);
956 	return work_done;
957 }
958 
959 static int nicvf_poll(struct napi_struct *napi, int budget)
960 {
961 	u64  cq_head;
962 	int  work_done = 0;
963 	struct net_device *netdev = napi->dev;
964 	struct nicvf *nic = netdev_priv(netdev);
965 	struct nicvf_cq_poll *cq;
966 
967 	cq = container_of(napi, struct nicvf_cq_poll, napi);
968 	work_done = nicvf_cq_intr_handler(netdev, cq->cq_idx, napi, budget);
969 
970 	if (work_done < budget) {
971 		/* Slow packet rate, exit polling */
972 		napi_complete_done(napi, work_done);
973 		/* Re-enable interrupts */
974 		cq_head = nicvf_queue_reg_read(nic, NIC_QSET_CQ_0_7_HEAD,
975 					       cq->cq_idx);
976 		nicvf_clear_intr(nic, NICVF_INTR_CQ, cq->cq_idx);
977 		nicvf_queue_reg_write(nic, NIC_QSET_CQ_0_7_HEAD,
978 				      cq->cq_idx, cq_head);
979 		nicvf_enable_intr(nic, NICVF_INTR_CQ, cq->cq_idx);
980 	}
981 	return work_done;
982 }
983 
984 /* Qset error interrupt handler
985  *
986  * As of now only CQ errors are handled
987  */
988 static void nicvf_handle_qs_err(struct tasklet_struct *t)
989 {
990 	struct nicvf *nic = from_tasklet(nic, t, qs_err_task);
991 	struct queue_set *qs = nic->qs;
992 	int qidx;
993 	u64 status;
994 
995 	netif_tx_disable(nic->netdev);
996 
997 	/* Check if it is CQ err */
998 	for (qidx = 0; qidx < qs->cq_cnt; qidx++) {
999 		status = nicvf_queue_reg_read(nic, NIC_QSET_CQ_0_7_STATUS,
1000 					      qidx);
1001 		if (!(status & CQ_ERR_MASK))
1002 			continue;
1003 		/* Process already queued CQEs and reconfig CQ */
1004 		nicvf_disable_intr(nic, NICVF_INTR_CQ, qidx);
1005 		nicvf_sq_disable(nic, qidx);
1006 		nicvf_cq_intr_handler(nic->netdev, qidx, NULL, 0);
1007 		nicvf_cmp_queue_config(nic, qs, qidx, true);
1008 		nicvf_sq_free_used_descs(nic->netdev, &qs->sq[qidx], qidx);
1009 		nicvf_sq_enable(nic, &qs->sq[qidx], qidx);
1010 
1011 		nicvf_enable_intr(nic, NICVF_INTR_CQ, qidx);
1012 	}
1013 
1014 	netif_tx_start_all_queues(nic->netdev);
1015 	/* Re-enable Qset error interrupt */
1016 	nicvf_enable_intr(nic, NICVF_INTR_QS_ERR, 0);
1017 }
1018 
1019 static void nicvf_dump_intr_status(struct nicvf *nic)
1020 {
1021 	netif_info(nic, intr, nic->netdev, "interrupt status 0x%llx\n",
1022 		   nicvf_reg_read(nic, NIC_VF_INT));
1023 }
1024 
1025 static irqreturn_t nicvf_misc_intr_handler(int irq, void *nicvf_irq)
1026 {
1027 	struct nicvf *nic = (struct nicvf *)nicvf_irq;
1028 	u64 intr;
1029 
1030 	nicvf_dump_intr_status(nic);
1031 
1032 	intr = nicvf_reg_read(nic, NIC_VF_INT);
1033 	/* Check for spurious interrupt */
1034 	if (!(intr & NICVF_INTR_MBOX_MASK))
1035 		return IRQ_HANDLED;
1036 
1037 	nicvf_handle_mbx_intr(nic);
1038 
1039 	return IRQ_HANDLED;
1040 }
1041 
1042 static irqreturn_t nicvf_intr_handler(int irq, void *cq_irq)
1043 {
1044 	struct nicvf_cq_poll *cq_poll = (struct nicvf_cq_poll *)cq_irq;
1045 	struct nicvf *nic = cq_poll->nicvf;
1046 	int qidx = cq_poll->cq_idx;
1047 
1048 	nicvf_dump_intr_status(nic);
1049 
1050 	/* Disable interrupts */
1051 	nicvf_disable_intr(nic, NICVF_INTR_CQ, qidx);
1052 
1053 	/* Schedule NAPI */
1054 	napi_schedule_irqoff(&cq_poll->napi);
1055 
1056 	/* Clear interrupt */
1057 	nicvf_clear_intr(nic, NICVF_INTR_CQ, qidx);
1058 
1059 	return IRQ_HANDLED;
1060 }
1061 
1062 static irqreturn_t nicvf_rbdr_intr_handler(int irq, void *nicvf_irq)
1063 {
1064 	struct nicvf *nic = (struct nicvf *)nicvf_irq;
1065 	u8 qidx;
1066 
1067 
1068 	nicvf_dump_intr_status(nic);
1069 
1070 	/* Disable RBDR interrupt and schedule softirq */
1071 	for (qidx = 0; qidx < nic->qs->rbdr_cnt; qidx++) {
1072 		if (!nicvf_is_intr_enabled(nic, NICVF_INTR_RBDR, qidx))
1073 			continue;
1074 		nicvf_disable_intr(nic, NICVF_INTR_RBDR, qidx);
1075 		tasklet_hi_schedule(&nic->rbdr_task);
1076 		/* Clear interrupt */
1077 		nicvf_clear_intr(nic, NICVF_INTR_RBDR, qidx);
1078 	}
1079 
1080 	return IRQ_HANDLED;
1081 }
1082 
1083 static irqreturn_t nicvf_qs_err_intr_handler(int irq, void *nicvf_irq)
1084 {
1085 	struct nicvf *nic = (struct nicvf *)nicvf_irq;
1086 
1087 	nicvf_dump_intr_status(nic);
1088 
1089 	/* Disable Qset err interrupt and schedule softirq */
1090 	nicvf_disable_intr(nic, NICVF_INTR_QS_ERR, 0);
1091 	tasklet_hi_schedule(&nic->qs_err_task);
1092 	nicvf_clear_intr(nic, NICVF_INTR_QS_ERR, 0);
1093 
1094 	return IRQ_HANDLED;
1095 }
1096 
1097 static void nicvf_set_irq_affinity(struct nicvf *nic)
1098 {
1099 	int vec, cpu;
1100 
1101 	for (vec = 0; vec < nic->num_vec; vec++) {
1102 		if (!nic->irq_allocated[vec])
1103 			continue;
1104 
1105 		if (!zalloc_cpumask_var(&nic->affinity_mask[vec], GFP_KERNEL))
1106 			return;
1107 		 /* CQ interrupts */
1108 		if (vec < NICVF_INTR_ID_SQ)
1109 			/* Leave CPU0 for RBDR and other interrupts */
1110 			cpu = nicvf_netdev_qidx(nic, vec) + 1;
1111 		else
1112 			cpu = 0;
1113 
1114 		cpumask_set_cpu(cpumask_local_spread(cpu, nic->node),
1115 				nic->affinity_mask[vec]);
1116 		irq_set_affinity_hint(pci_irq_vector(nic->pdev, vec),
1117 				      nic->affinity_mask[vec]);
1118 	}
1119 }
1120 
1121 static int nicvf_register_interrupts(struct nicvf *nic)
1122 {
1123 	int irq, ret = 0;
1124 
1125 	for_each_cq_irq(irq)
1126 		sprintf(nic->irq_name[irq], "%s-rxtx-%d",
1127 			nic->pnicvf->netdev->name,
1128 			nicvf_netdev_qidx(nic, irq));
1129 
1130 	for_each_sq_irq(irq)
1131 		sprintf(nic->irq_name[irq], "%s-sq-%d",
1132 			nic->pnicvf->netdev->name,
1133 			nicvf_netdev_qidx(nic, irq - NICVF_INTR_ID_SQ));
1134 
1135 	for_each_rbdr_irq(irq)
1136 		sprintf(nic->irq_name[irq], "%s-rbdr-%d",
1137 			nic->pnicvf->netdev->name,
1138 			nic->sqs_mode ? (nic->sqs_id + 1) : 0);
1139 
1140 	/* Register CQ interrupts */
1141 	for (irq = 0; irq < nic->qs->cq_cnt; irq++) {
1142 		ret = request_irq(pci_irq_vector(nic->pdev, irq),
1143 				  nicvf_intr_handler,
1144 				  0, nic->irq_name[irq], nic->napi[irq]);
1145 		if (ret)
1146 			goto err;
1147 		nic->irq_allocated[irq] = true;
1148 	}
1149 
1150 	/* Register RBDR interrupt */
1151 	for (irq = NICVF_INTR_ID_RBDR;
1152 	     irq < (NICVF_INTR_ID_RBDR + nic->qs->rbdr_cnt); irq++) {
1153 		ret = request_irq(pci_irq_vector(nic->pdev, irq),
1154 				  nicvf_rbdr_intr_handler,
1155 				  0, nic->irq_name[irq], nic);
1156 		if (ret)
1157 			goto err;
1158 		nic->irq_allocated[irq] = true;
1159 	}
1160 
1161 	/* Register QS error interrupt */
1162 	sprintf(nic->irq_name[NICVF_INTR_ID_QS_ERR], "%s-qset-err-%d",
1163 		nic->pnicvf->netdev->name,
1164 		nic->sqs_mode ? (nic->sqs_id + 1) : 0);
1165 	irq = NICVF_INTR_ID_QS_ERR;
1166 	ret = request_irq(pci_irq_vector(nic->pdev, irq),
1167 			  nicvf_qs_err_intr_handler,
1168 			  0, nic->irq_name[irq], nic);
1169 	if (ret)
1170 		goto err;
1171 
1172 	nic->irq_allocated[irq] = true;
1173 
1174 	/* Set IRQ affinities */
1175 	nicvf_set_irq_affinity(nic);
1176 
1177 err:
1178 	if (ret)
1179 		netdev_err(nic->netdev, "request_irq failed, vector %d\n", irq);
1180 
1181 	return ret;
1182 }
1183 
1184 static void nicvf_unregister_interrupts(struct nicvf *nic)
1185 {
1186 	struct pci_dev *pdev = nic->pdev;
1187 	int irq;
1188 
1189 	/* Free registered interrupts */
1190 	for (irq = 0; irq < nic->num_vec; irq++) {
1191 		if (!nic->irq_allocated[irq])
1192 			continue;
1193 
1194 		irq_set_affinity_hint(pci_irq_vector(pdev, irq), NULL);
1195 		free_cpumask_var(nic->affinity_mask[irq]);
1196 
1197 		if (irq < NICVF_INTR_ID_SQ)
1198 			free_irq(pci_irq_vector(pdev, irq), nic->napi[irq]);
1199 		else
1200 			free_irq(pci_irq_vector(pdev, irq), nic);
1201 
1202 		nic->irq_allocated[irq] = false;
1203 	}
1204 
1205 	/* Disable MSI-X */
1206 	pci_free_irq_vectors(pdev);
1207 	nic->num_vec = 0;
1208 }
1209 
1210 /* Initialize MSIX vectors and register MISC interrupt.
1211  * Send READY message to PF to check if its alive
1212  */
1213 static int nicvf_register_misc_interrupt(struct nicvf *nic)
1214 {
1215 	int ret = 0;
1216 	int irq = NICVF_INTR_ID_MISC;
1217 
1218 	/* Return if mailbox interrupt is already registered */
1219 	if (nic->pdev->msix_enabled)
1220 		return 0;
1221 
1222 	/* Enable MSI-X */
1223 	nic->num_vec = pci_msix_vec_count(nic->pdev);
1224 	ret = pci_alloc_irq_vectors(nic->pdev, nic->num_vec, nic->num_vec,
1225 				    PCI_IRQ_MSIX);
1226 	if (ret < 0) {
1227 		netdev_err(nic->netdev,
1228 			   "Req for #%d msix vectors failed\n", nic->num_vec);
1229 		return 1;
1230 	}
1231 
1232 	sprintf(nic->irq_name[irq], "%s Mbox", "NICVF");
1233 	/* Register Misc interrupt */
1234 	ret = request_irq(pci_irq_vector(nic->pdev, irq),
1235 			  nicvf_misc_intr_handler, 0, nic->irq_name[irq], nic);
1236 
1237 	if (ret)
1238 		return ret;
1239 	nic->irq_allocated[irq] = true;
1240 
1241 	/* Enable mailbox interrupt */
1242 	nicvf_enable_intr(nic, NICVF_INTR_MBOX, 0);
1243 
1244 	/* Check if VF is able to communicate with PF */
1245 	if (!nicvf_check_pf_ready(nic)) {
1246 		nicvf_disable_intr(nic, NICVF_INTR_MBOX, 0);
1247 		nicvf_unregister_interrupts(nic);
1248 		return 1;
1249 	}
1250 
1251 	return 0;
1252 }
1253 
1254 static netdev_tx_t nicvf_xmit(struct sk_buff *skb, struct net_device *netdev)
1255 {
1256 	struct nicvf *nic = netdev_priv(netdev);
1257 	int qid = skb_get_queue_mapping(skb);
1258 	struct netdev_queue *txq = netdev_get_tx_queue(netdev, qid);
1259 	struct nicvf *snic;
1260 	struct snd_queue *sq;
1261 	int tmp;
1262 
1263 	/* Check for minimum packet length */
1264 	if (skb->len <= ETH_HLEN) {
1265 		dev_kfree_skb(skb);
1266 		return NETDEV_TX_OK;
1267 	}
1268 
1269 	/* In XDP case, initial HW tx queues are used for XDP,
1270 	 * but stack's queue mapping starts at '0', so skip the
1271 	 * Tx queues attached to Rx queues for XDP.
1272 	 */
1273 	if (nic->xdp_prog)
1274 		qid += nic->xdp_tx_queues;
1275 
1276 	snic = nic;
1277 	/* Get secondary Qset's SQ structure */
1278 	if (qid >= MAX_SND_QUEUES_PER_QS) {
1279 		tmp = qid / MAX_SND_QUEUES_PER_QS;
1280 		snic = (struct nicvf *)nic->snicvf[tmp - 1];
1281 		if (!snic) {
1282 			netdev_warn(nic->netdev,
1283 				    "Secondary Qset#%d's ptr not initialized\n",
1284 				    tmp - 1);
1285 			dev_kfree_skb(skb);
1286 			return NETDEV_TX_OK;
1287 		}
1288 		qid = qid % MAX_SND_QUEUES_PER_QS;
1289 	}
1290 
1291 	sq = &snic->qs->sq[qid];
1292 	if (!netif_tx_queue_stopped(txq) &&
1293 	    !nicvf_sq_append_skb(snic, sq, skb, qid)) {
1294 		netif_tx_stop_queue(txq);
1295 
1296 		/* Barrier, so that stop_queue visible to other cpus */
1297 		smp_mb();
1298 
1299 		/* Check again, incase another cpu freed descriptors */
1300 		if (atomic_read(&sq->free_cnt) > MIN_SQ_DESC_PER_PKT_XMIT) {
1301 			netif_tx_wake_queue(txq);
1302 		} else {
1303 			this_cpu_inc(nic->drv_stats->txq_stop);
1304 			netif_warn(nic, tx_err, netdev,
1305 				   "Transmit ring full, stopping SQ%d\n", qid);
1306 		}
1307 		return NETDEV_TX_BUSY;
1308 	}
1309 
1310 	return NETDEV_TX_OK;
1311 }
1312 
1313 static inline void nicvf_free_cq_poll(struct nicvf *nic)
1314 {
1315 	struct nicvf_cq_poll *cq_poll;
1316 	int qidx;
1317 
1318 	for (qidx = 0; qidx < nic->qs->cq_cnt; qidx++) {
1319 		cq_poll = nic->napi[qidx];
1320 		if (!cq_poll)
1321 			continue;
1322 		nic->napi[qidx] = NULL;
1323 		kfree(cq_poll);
1324 	}
1325 }
1326 
1327 int nicvf_stop(struct net_device *netdev)
1328 {
1329 	int irq, qidx;
1330 	struct nicvf *nic = netdev_priv(netdev);
1331 	struct queue_set *qs = nic->qs;
1332 	struct nicvf_cq_poll *cq_poll = NULL;
1333 	union nic_mbx mbx = {};
1334 
1335 	/* wait till all queued set_rx_mode tasks completes */
1336 	if (nic->nicvf_rx_mode_wq) {
1337 		cancel_delayed_work_sync(&nic->link_change_work);
1338 		drain_workqueue(nic->nicvf_rx_mode_wq);
1339 	}
1340 
1341 	mbx.msg.msg = NIC_MBOX_MSG_SHUTDOWN;
1342 	nicvf_send_msg_to_pf(nic, &mbx);
1343 
1344 	netif_carrier_off(netdev);
1345 	netif_tx_stop_all_queues(nic->netdev);
1346 	nic->link_up = false;
1347 
1348 	/* Teardown secondary qsets first */
1349 	if (!nic->sqs_mode) {
1350 		for (qidx = 0; qidx < nic->sqs_count; qidx++) {
1351 			if (!nic->snicvf[qidx])
1352 				continue;
1353 			nicvf_stop(nic->snicvf[qidx]->netdev);
1354 			nic->snicvf[qidx] = NULL;
1355 		}
1356 	}
1357 
1358 	/* Disable RBDR & QS error interrupts */
1359 	for (qidx = 0; qidx < qs->rbdr_cnt; qidx++) {
1360 		nicvf_disable_intr(nic, NICVF_INTR_RBDR, qidx);
1361 		nicvf_clear_intr(nic, NICVF_INTR_RBDR, qidx);
1362 	}
1363 	nicvf_disable_intr(nic, NICVF_INTR_QS_ERR, 0);
1364 	nicvf_clear_intr(nic, NICVF_INTR_QS_ERR, 0);
1365 
1366 	/* Wait for pending IRQ handlers to finish */
1367 	for (irq = 0; irq < nic->num_vec; irq++)
1368 		synchronize_irq(pci_irq_vector(nic->pdev, irq));
1369 
1370 	tasklet_kill(&nic->rbdr_task);
1371 	tasklet_kill(&nic->qs_err_task);
1372 	if (nic->rb_work_scheduled)
1373 		cancel_delayed_work_sync(&nic->rbdr_work);
1374 
1375 	for (qidx = 0; qidx < nic->qs->cq_cnt; qidx++) {
1376 		cq_poll = nic->napi[qidx];
1377 		if (!cq_poll)
1378 			continue;
1379 		napi_synchronize(&cq_poll->napi);
1380 		/* CQ intr is enabled while napi_complete,
1381 		 * so disable it now
1382 		 */
1383 		nicvf_disable_intr(nic, NICVF_INTR_CQ, qidx);
1384 		nicvf_clear_intr(nic, NICVF_INTR_CQ, qidx);
1385 		napi_disable(&cq_poll->napi);
1386 		netif_napi_del(&cq_poll->napi);
1387 	}
1388 
1389 	netif_tx_disable(netdev);
1390 
1391 	for (qidx = 0; qidx < netdev->num_tx_queues; qidx++)
1392 		netdev_tx_reset_queue(netdev_get_tx_queue(netdev, qidx));
1393 
1394 	/* Free resources */
1395 	nicvf_config_data_transfer(nic, false);
1396 
1397 	/* Disable HW Qset */
1398 	nicvf_qset_config(nic, false);
1399 
1400 	/* disable mailbox interrupt */
1401 	nicvf_disable_intr(nic, NICVF_INTR_MBOX, 0);
1402 
1403 	nicvf_unregister_interrupts(nic);
1404 
1405 	nicvf_free_cq_poll(nic);
1406 
1407 	/* Free any pending SKB saved to receive timestamp */
1408 	if (nic->ptp_skb) {
1409 		dev_kfree_skb_any(nic->ptp_skb);
1410 		nic->ptp_skb = NULL;
1411 	}
1412 
1413 	/* Clear multiqset info */
1414 	nic->pnicvf = nic;
1415 
1416 	return 0;
1417 }
1418 
1419 static int nicvf_config_hw_rx_tstamp(struct nicvf *nic, bool enable)
1420 {
1421 	union nic_mbx mbx = {};
1422 
1423 	mbx.ptp.msg = NIC_MBOX_MSG_PTP_CFG;
1424 	mbx.ptp.enable = enable;
1425 
1426 	return nicvf_send_msg_to_pf(nic, &mbx);
1427 }
1428 
1429 static int nicvf_update_hw_max_frs(struct nicvf *nic, int mtu)
1430 {
1431 	union nic_mbx mbx = {};
1432 
1433 	mbx.frs.msg = NIC_MBOX_MSG_SET_MAX_FRS;
1434 	mbx.frs.max_frs = mtu;
1435 	mbx.frs.vf_id = nic->vf_id;
1436 
1437 	return nicvf_send_msg_to_pf(nic, &mbx);
1438 }
1439 
1440 static void nicvf_link_status_check_task(struct work_struct *work_arg)
1441 {
1442 	struct nicvf *nic = container_of(work_arg,
1443 					 struct nicvf,
1444 					 link_change_work.work);
1445 	union nic_mbx mbx = {};
1446 	mbx.msg.msg = NIC_MBOX_MSG_BGX_LINK_CHANGE;
1447 	nicvf_send_msg_to_pf(nic, &mbx);
1448 	queue_delayed_work(nic->nicvf_rx_mode_wq,
1449 			   &nic->link_change_work, 2 * HZ);
1450 }
1451 
1452 int nicvf_open(struct net_device *netdev)
1453 {
1454 	int cpu, err, qidx;
1455 	struct nicvf *nic = netdev_priv(netdev);
1456 	struct queue_set *qs = nic->qs;
1457 	struct nicvf_cq_poll *cq_poll = NULL;
1458 
1459 	/* wait till all queued set_rx_mode tasks completes if any */
1460 	if (nic->nicvf_rx_mode_wq)
1461 		drain_workqueue(nic->nicvf_rx_mode_wq);
1462 
1463 	netif_carrier_off(netdev);
1464 
1465 	err = nicvf_register_misc_interrupt(nic);
1466 	if (err)
1467 		return err;
1468 
1469 	/* Register NAPI handler for processing CQEs */
1470 	for (qidx = 0; qidx < qs->cq_cnt; qidx++) {
1471 		cq_poll = kzalloc(sizeof(*cq_poll), GFP_KERNEL);
1472 		if (!cq_poll) {
1473 			err = -ENOMEM;
1474 			goto napi_del;
1475 		}
1476 		cq_poll->cq_idx = qidx;
1477 		cq_poll->nicvf = nic;
1478 		netif_napi_add(netdev, &cq_poll->napi, nicvf_poll,
1479 			       NAPI_POLL_WEIGHT);
1480 		napi_enable(&cq_poll->napi);
1481 		nic->napi[qidx] = cq_poll;
1482 	}
1483 
1484 	/* Check if we got MAC address from PF or else generate a radom MAC */
1485 	if (!nic->sqs_mode && is_zero_ether_addr(netdev->dev_addr)) {
1486 		eth_hw_addr_random(netdev);
1487 		nicvf_hw_set_mac_addr(nic, netdev);
1488 	}
1489 
1490 	if (nic->set_mac_pending) {
1491 		nic->set_mac_pending = false;
1492 		nicvf_hw_set_mac_addr(nic, netdev);
1493 	}
1494 
1495 	/* Init tasklet for handling Qset err interrupt */
1496 	tasklet_setup(&nic->qs_err_task, nicvf_handle_qs_err);
1497 
1498 	/* Init RBDR tasklet which will refill RBDR */
1499 	tasklet_setup(&nic->rbdr_task, nicvf_rbdr_task);
1500 	INIT_DELAYED_WORK(&nic->rbdr_work, nicvf_rbdr_work);
1501 
1502 	/* Configure CPI alorithm */
1503 	nic->cpi_alg = cpi_alg;
1504 	if (!nic->sqs_mode)
1505 		nicvf_config_cpi(nic);
1506 
1507 	nicvf_request_sqs(nic);
1508 	if (nic->sqs_mode)
1509 		nicvf_get_primary_vf_struct(nic);
1510 
1511 	/* Configure PTP timestamp */
1512 	if (nic->ptp_clock)
1513 		nicvf_config_hw_rx_tstamp(nic, nic->hw_rx_tstamp);
1514 	atomic_set(&nic->tx_ptp_skbs, 0);
1515 	nic->ptp_skb = NULL;
1516 
1517 	/* Configure receive side scaling and MTU */
1518 	if (!nic->sqs_mode) {
1519 		nicvf_rss_init(nic);
1520 		err = nicvf_update_hw_max_frs(nic, netdev->mtu);
1521 		if (err)
1522 			goto cleanup;
1523 
1524 		/* Clear percpu stats */
1525 		for_each_possible_cpu(cpu)
1526 			memset(per_cpu_ptr(nic->drv_stats, cpu), 0,
1527 			       sizeof(struct nicvf_drv_stats));
1528 	}
1529 
1530 	err = nicvf_register_interrupts(nic);
1531 	if (err)
1532 		goto cleanup;
1533 
1534 	/* Initialize the queues */
1535 	err = nicvf_init_resources(nic);
1536 	if (err)
1537 		goto cleanup;
1538 
1539 	/* Make sure queue initialization is written */
1540 	wmb();
1541 
1542 	nicvf_reg_write(nic, NIC_VF_INT, -1);
1543 	/* Enable Qset err interrupt */
1544 	nicvf_enable_intr(nic, NICVF_INTR_QS_ERR, 0);
1545 
1546 	/* Enable completion queue interrupt */
1547 	for (qidx = 0; qidx < qs->cq_cnt; qidx++)
1548 		nicvf_enable_intr(nic, NICVF_INTR_CQ, qidx);
1549 
1550 	/* Enable RBDR threshold interrupt */
1551 	for (qidx = 0; qidx < qs->rbdr_cnt; qidx++)
1552 		nicvf_enable_intr(nic, NICVF_INTR_RBDR, qidx);
1553 
1554 	/* Send VF config done msg to PF */
1555 	nicvf_send_cfg_done(nic);
1556 
1557 	if (nic->nicvf_rx_mode_wq) {
1558 		INIT_DELAYED_WORK(&nic->link_change_work,
1559 				  nicvf_link_status_check_task);
1560 		queue_delayed_work(nic->nicvf_rx_mode_wq,
1561 				   &nic->link_change_work, 0);
1562 	}
1563 
1564 	return 0;
1565 cleanup:
1566 	nicvf_disable_intr(nic, NICVF_INTR_MBOX, 0);
1567 	nicvf_unregister_interrupts(nic);
1568 	tasklet_kill(&nic->qs_err_task);
1569 	tasklet_kill(&nic->rbdr_task);
1570 napi_del:
1571 	for (qidx = 0; qidx < qs->cq_cnt; qidx++) {
1572 		cq_poll = nic->napi[qidx];
1573 		if (!cq_poll)
1574 			continue;
1575 		napi_disable(&cq_poll->napi);
1576 		netif_napi_del(&cq_poll->napi);
1577 	}
1578 	nicvf_free_cq_poll(nic);
1579 	return err;
1580 }
1581 
1582 static int nicvf_change_mtu(struct net_device *netdev, int new_mtu)
1583 {
1584 	struct nicvf *nic = netdev_priv(netdev);
1585 	int orig_mtu = netdev->mtu;
1586 
1587 	/* For now just support only the usual MTU sized frames,
1588 	 * plus some headroom for VLAN, QinQ.
1589 	 */
1590 	if (nic->xdp_prog && new_mtu > MAX_XDP_MTU) {
1591 		netdev_warn(netdev, "Jumbo frames not yet supported with XDP, current MTU %d.\n",
1592 			    netdev->mtu);
1593 		return -EINVAL;
1594 	}
1595 
1596 	netdev->mtu = new_mtu;
1597 
1598 	if (!netif_running(netdev))
1599 		return 0;
1600 
1601 	if (nicvf_update_hw_max_frs(nic, new_mtu)) {
1602 		netdev->mtu = orig_mtu;
1603 		return -EINVAL;
1604 	}
1605 
1606 	return 0;
1607 }
1608 
1609 static int nicvf_set_mac_address(struct net_device *netdev, void *p)
1610 {
1611 	struct sockaddr *addr = p;
1612 	struct nicvf *nic = netdev_priv(netdev);
1613 
1614 	if (!is_valid_ether_addr(addr->sa_data))
1615 		return -EADDRNOTAVAIL;
1616 
1617 	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1618 
1619 	if (nic->pdev->msix_enabled) {
1620 		if (nicvf_hw_set_mac_addr(nic, netdev))
1621 			return -EBUSY;
1622 	} else {
1623 		nic->set_mac_pending = true;
1624 	}
1625 
1626 	return 0;
1627 }
1628 
1629 void nicvf_update_lmac_stats(struct nicvf *nic)
1630 {
1631 	int stat = 0;
1632 	union nic_mbx mbx = {};
1633 
1634 	if (!netif_running(nic->netdev))
1635 		return;
1636 
1637 	mbx.bgx_stats.msg = NIC_MBOX_MSG_BGX_STATS;
1638 	mbx.bgx_stats.vf_id = nic->vf_id;
1639 	/* Rx stats */
1640 	mbx.bgx_stats.rx = 1;
1641 	while (stat < BGX_RX_STATS_COUNT) {
1642 		mbx.bgx_stats.idx = stat;
1643 		if (nicvf_send_msg_to_pf(nic, &mbx))
1644 			return;
1645 		stat++;
1646 	}
1647 
1648 	stat = 0;
1649 
1650 	/* Tx stats */
1651 	mbx.bgx_stats.rx = 0;
1652 	while (stat < BGX_TX_STATS_COUNT) {
1653 		mbx.bgx_stats.idx = stat;
1654 		if (nicvf_send_msg_to_pf(nic, &mbx))
1655 			return;
1656 		stat++;
1657 	}
1658 }
1659 
1660 void nicvf_update_stats(struct nicvf *nic)
1661 {
1662 	int qidx, cpu;
1663 	u64 tmp_stats = 0;
1664 	struct nicvf_hw_stats *stats = &nic->hw_stats;
1665 	struct nicvf_drv_stats *drv_stats;
1666 	struct queue_set *qs = nic->qs;
1667 
1668 #define GET_RX_STATS(reg) \
1669 	nicvf_reg_read(nic, NIC_VNIC_RX_STAT_0_13 | (reg << 3))
1670 #define GET_TX_STATS(reg) \
1671 	nicvf_reg_read(nic, NIC_VNIC_TX_STAT_0_4 | (reg << 3))
1672 
1673 	stats->rx_bytes = GET_RX_STATS(RX_OCTS);
1674 	stats->rx_ucast_frames = GET_RX_STATS(RX_UCAST);
1675 	stats->rx_bcast_frames = GET_RX_STATS(RX_BCAST);
1676 	stats->rx_mcast_frames = GET_RX_STATS(RX_MCAST);
1677 	stats->rx_fcs_errors = GET_RX_STATS(RX_FCS);
1678 	stats->rx_l2_errors = GET_RX_STATS(RX_L2ERR);
1679 	stats->rx_drop_red = GET_RX_STATS(RX_RED);
1680 	stats->rx_drop_red_bytes = GET_RX_STATS(RX_RED_OCTS);
1681 	stats->rx_drop_overrun = GET_RX_STATS(RX_ORUN);
1682 	stats->rx_drop_overrun_bytes = GET_RX_STATS(RX_ORUN_OCTS);
1683 	stats->rx_drop_bcast = GET_RX_STATS(RX_DRP_BCAST);
1684 	stats->rx_drop_mcast = GET_RX_STATS(RX_DRP_MCAST);
1685 	stats->rx_drop_l3_bcast = GET_RX_STATS(RX_DRP_L3BCAST);
1686 	stats->rx_drop_l3_mcast = GET_RX_STATS(RX_DRP_L3MCAST);
1687 
1688 	stats->tx_bytes = GET_TX_STATS(TX_OCTS);
1689 	stats->tx_ucast_frames = GET_TX_STATS(TX_UCAST);
1690 	stats->tx_bcast_frames = GET_TX_STATS(TX_BCAST);
1691 	stats->tx_mcast_frames = GET_TX_STATS(TX_MCAST);
1692 	stats->tx_drops = GET_TX_STATS(TX_DROP);
1693 
1694 	/* On T88 pass 2.0, the dummy SQE added for TSO notification
1695 	 * via CQE has 'dont_send' set. Hence HW drops the pkt pointed
1696 	 * pointed by dummy SQE and results in tx_drops counter being
1697 	 * incremented. Subtracting it from tx_tso counter will give
1698 	 * exact tx_drops counter.
1699 	 */
1700 	if (nic->t88 && nic->hw_tso) {
1701 		for_each_possible_cpu(cpu) {
1702 			drv_stats = per_cpu_ptr(nic->drv_stats, cpu);
1703 			tmp_stats += drv_stats->tx_tso;
1704 		}
1705 		stats->tx_drops = tmp_stats - stats->tx_drops;
1706 	}
1707 	stats->tx_frames = stats->tx_ucast_frames +
1708 			   stats->tx_bcast_frames +
1709 			   stats->tx_mcast_frames;
1710 	stats->rx_frames = stats->rx_ucast_frames +
1711 			   stats->rx_bcast_frames +
1712 			   stats->rx_mcast_frames;
1713 	stats->rx_drops = stats->rx_drop_red +
1714 			  stats->rx_drop_overrun;
1715 
1716 	/* Update RQ and SQ stats */
1717 	for (qidx = 0; qidx < qs->rq_cnt; qidx++)
1718 		nicvf_update_rq_stats(nic, qidx);
1719 	for (qidx = 0; qidx < qs->sq_cnt; qidx++)
1720 		nicvf_update_sq_stats(nic, qidx);
1721 }
1722 
1723 static void nicvf_get_stats64(struct net_device *netdev,
1724 			      struct rtnl_link_stats64 *stats)
1725 {
1726 	struct nicvf *nic = netdev_priv(netdev);
1727 	struct nicvf_hw_stats *hw_stats = &nic->hw_stats;
1728 
1729 	nicvf_update_stats(nic);
1730 
1731 	stats->rx_bytes = hw_stats->rx_bytes;
1732 	stats->rx_packets = hw_stats->rx_frames;
1733 	stats->rx_dropped = hw_stats->rx_drops;
1734 	stats->multicast = hw_stats->rx_mcast_frames;
1735 
1736 	stats->tx_bytes = hw_stats->tx_bytes;
1737 	stats->tx_packets = hw_stats->tx_frames;
1738 	stats->tx_dropped = hw_stats->tx_drops;
1739 
1740 }
1741 
1742 static void nicvf_tx_timeout(struct net_device *dev, unsigned int txqueue)
1743 {
1744 	struct nicvf *nic = netdev_priv(dev);
1745 
1746 	netif_warn(nic, tx_err, dev, "Transmit timed out, resetting\n");
1747 
1748 	this_cpu_inc(nic->drv_stats->tx_timeout);
1749 	schedule_work(&nic->reset_task);
1750 }
1751 
1752 static void nicvf_reset_task(struct work_struct *work)
1753 {
1754 	struct nicvf *nic;
1755 
1756 	nic = container_of(work, struct nicvf, reset_task);
1757 
1758 	if (!netif_running(nic->netdev))
1759 		return;
1760 
1761 	nicvf_stop(nic->netdev);
1762 	nicvf_open(nic->netdev);
1763 	netif_trans_update(nic->netdev);
1764 }
1765 
1766 static int nicvf_config_loopback(struct nicvf *nic,
1767 				 netdev_features_t features)
1768 {
1769 	union nic_mbx mbx = {};
1770 
1771 	mbx.lbk.msg = NIC_MBOX_MSG_LOOPBACK;
1772 	mbx.lbk.vf_id = nic->vf_id;
1773 	mbx.lbk.enable = (features & NETIF_F_LOOPBACK) != 0;
1774 
1775 	return nicvf_send_msg_to_pf(nic, &mbx);
1776 }
1777 
1778 static netdev_features_t nicvf_fix_features(struct net_device *netdev,
1779 					    netdev_features_t features)
1780 {
1781 	struct nicvf *nic = netdev_priv(netdev);
1782 
1783 	if ((features & NETIF_F_LOOPBACK) &&
1784 	    netif_running(netdev) && !nic->loopback_supported)
1785 		features &= ~NETIF_F_LOOPBACK;
1786 
1787 	return features;
1788 }
1789 
1790 static int nicvf_set_features(struct net_device *netdev,
1791 			      netdev_features_t features)
1792 {
1793 	struct nicvf *nic = netdev_priv(netdev);
1794 	netdev_features_t changed = features ^ netdev->features;
1795 
1796 	if (changed & NETIF_F_HW_VLAN_CTAG_RX)
1797 		nicvf_config_vlan_stripping(nic, features);
1798 
1799 	if ((changed & NETIF_F_LOOPBACK) && netif_running(netdev))
1800 		return nicvf_config_loopback(nic, features);
1801 
1802 	return 0;
1803 }
1804 
1805 static void nicvf_set_xdp_queues(struct nicvf *nic, bool bpf_attached)
1806 {
1807 	u8 cq_count, txq_count;
1808 
1809 	/* Set XDP Tx queue count same as Rx queue count */
1810 	if (!bpf_attached)
1811 		nic->xdp_tx_queues = 0;
1812 	else
1813 		nic->xdp_tx_queues = nic->rx_queues;
1814 
1815 	/* If queue count > MAX_CMP_QUEUES_PER_QS, then additional qsets
1816 	 * needs to be allocated, check how many.
1817 	 */
1818 	txq_count = nic->xdp_tx_queues + nic->tx_queues;
1819 	cq_count = max(nic->rx_queues, txq_count);
1820 	if (cq_count > MAX_CMP_QUEUES_PER_QS) {
1821 		nic->sqs_count = roundup(cq_count, MAX_CMP_QUEUES_PER_QS);
1822 		nic->sqs_count = (nic->sqs_count / MAX_CMP_QUEUES_PER_QS) - 1;
1823 	} else {
1824 		nic->sqs_count = 0;
1825 	}
1826 
1827 	/* Set primary Qset's resources */
1828 	nic->qs->rq_cnt = min_t(u8, nic->rx_queues, MAX_RCV_QUEUES_PER_QS);
1829 	nic->qs->sq_cnt = min_t(u8, txq_count, MAX_SND_QUEUES_PER_QS);
1830 	nic->qs->cq_cnt = max_t(u8, nic->qs->rq_cnt, nic->qs->sq_cnt);
1831 
1832 	/* Update stack */
1833 	nicvf_set_real_num_queues(nic->netdev, nic->tx_queues, nic->rx_queues);
1834 }
1835 
1836 static int nicvf_xdp_setup(struct nicvf *nic, struct bpf_prog *prog)
1837 {
1838 	struct net_device *dev = nic->netdev;
1839 	bool if_up = netif_running(nic->netdev);
1840 	struct bpf_prog *old_prog;
1841 	bool bpf_attached = false;
1842 	int ret = 0;
1843 
1844 	/* For now just support only the usual MTU sized frames,
1845 	 * plus some headroom for VLAN, QinQ.
1846 	 */
1847 	if (prog && dev->mtu > MAX_XDP_MTU) {
1848 		netdev_warn(dev, "Jumbo frames not yet supported with XDP, current MTU %d.\n",
1849 			    dev->mtu);
1850 		return -EOPNOTSUPP;
1851 	}
1852 
1853 	/* ALL SQs attached to CQs i.e same as RQs, are treated as
1854 	 * XDP Tx queues and more Tx queues are allocated for
1855 	 * network stack to send pkts out.
1856 	 *
1857 	 * No of Tx queues are either same as Rx queues or whatever
1858 	 * is left in max no of queues possible.
1859 	 */
1860 	if ((nic->rx_queues + nic->tx_queues) > nic->max_queues) {
1861 		netdev_warn(dev,
1862 			    "Failed to attach BPF prog, RXQs + TXQs > Max %d\n",
1863 			    nic->max_queues);
1864 		return -ENOMEM;
1865 	}
1866 
1867 	if (if_up)
1868 		nicvf_stop(nic->netdev);
1869 
1870 	old_prog = xchg(&nic->xdp_prog, prog);
1871 	/* Detach old prog, if any */
1872 	if (old_prog)
1873 		bpf_prog_put(old_prog);
1874 
1875 	if (nic->xdp_prog) {
1876 		/* Attach BPF program */
1877 		bpf_prog_add(nic->xdp_prog, nic->rx_queues - 1);
1878 		bpf_attached = true;
1879 	}
1880 
1881 	/* Calculate Tx queues needed for XDP and network stack */
1882 	nicvf_set_xdp_queues(nic, bpf_attached);
1883 
1884 	if (if_up) {
1885 		/* Reinitialize interface, clean slate */
1886 		nicvf_open(nic->netdev);
1887 		netif_trans_update(nic->netdev);
1888 	}
1889 
1890 	return ret;
1891 }
1892 
1893 static int nicvf_xdp(struct net_device *netdev, struct netdev_bpf *xdp)
1894 {
1895 	struct nicvf *nic = netdev_priv(netdev);
1896 
1897 	/* To avoid checks while retrieving buffer address from CQE_RX,
1898 	 * do not support XDP for T88 pass1.x silicons which are anyway
1899 	 * not in use widely.
1900 	 */
1901 	if (pass1_silicon(nic->pdev))
1902 		return -EOPNOTSUPP;
1903 
1904 	switch (xdp->command) {
1905 	case XDP_SETUP_PROG:
1906 		return nicvf_xdp_setup(nic, xdp->prog);
1907 	default:
1908 		return -EINVAL;
1909 	}
1910 }
1911 
1912 static int nicvf_config_hwtstamp(struct net_device *netdev, struct ifreq *ifr)
1913 {
1914 	struct hwtstamp_config config;
1915 	struct nicvf *nic = netdev_priv(netdev);
1916 
1917 	if (!nic->ptp_clock)
1918 		return -ENODEV;
1919 
1920 	if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
1921 		return -EFAULT;
1922 
1923 	/* reserved for future extensions */
1924 	if (config.flags)
1925 		return -EINVAL;
1926 
1927 	switch (config.tx_type) {
1928 	case HWTSTAMP_TX_OFF:
1929 	case HWTSTAMP_TX_ON:
1930 		break;
1931 	default:
1932 		return -ERANGE;
1933 	}
1934 
1935 	switch (config.rx_filter) {
1936 	case HWTSTAMP_FILTER_NONE:
1937 		nic->hw_rx_tstamp = false;
1938 		break;
1939 	case HWTSTAMP_FILTER_ALL:
1940 	case HWTSTAMP_FILTER_SOME:
1941 	case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
1942 	case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
1943 	case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
1944 	case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
1945 	case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
1946 	case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
1947 	case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
1948 	case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
1949 	case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
1950 	case HWTSTAMP_FILTER_PTP_V2_EVENT:
1951 	case HWTSTAMP_FILTER_PTP_V2_SYNC:
1952 	case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
1953 		nic->hw_rx_tstamp = true;
1954 		config.rx_filter = HWTSTAMP_FILTER_ALL;
1955 		break;
1956 	default:
1957 		return -ERANGE;
1958 	}
1959 
1960 	if (netif_running(netdev))
1961 		nicvf_config_hw_rx_tstamp(nic, nic->hw_rx_tstamp);
1962 
1963 	if (copy_to_user(ifr->ifr_data, &config, sizeof(config)))
1964 		return -EFAULT;
1965 
1966 	return 0;
1967 }
1968 
1969 static int nicvf_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
1970 {
1971 	switch (cmd) {
1972 	case SIOCSHWTSTAMP:
1973 		return nicvf_config_hwtstamp(netdev, req);
1974 	default:
1975 		return -EOPNOTSUPP;
1976 	}
1977 }
1978 
1979 static void __nicvf_set_rx_mode_task(u8 mode, struct xcast_addr_list *mc_addrs,
1980 				     struct nicvf *nic)
1981 {
1982 	union nic_mbx mbx = {};
1983 	int idx;
1984 
1985 	/* From the inside of VM code flow we have only 128 bits memory
1986 	 * available to send message to host's PF, so send all mc addrs
1987 	 * one by one, starting from flush command in case if kernel
1988 	 * requests to configure specific MAC filtering
1989 	 */
1990 
1991 	/* flush DMAC filters and reset RX mode */
1992 	mbx.xcast.msg = NIC_MBOX_MSG_RESET_XCAST;
1993 	if (nicvf_send_msg_to_pf(nic, &mbx) < 0)
1994 		goto free_mc;
1995 
1996 	if (mode & BGX_XCAST_MCAST_FILTER) {
1997 		/* once enabling filtering, we need to signal to PF to add
1998 		 * its' own LMAC to the filter to accept packets for it.
1999 		 */
2000 		mbx.xcast.msg = NIC_MBOX_MSG_ADD_MCAST;
2001 		mbx.xcast.mac = 0;
2002 		if (nicvf_send_msg_to_pf(nic, &mbx) < 0)
2003 			goto free_mc;
2004 	}
2005 
2006 	/* check if we have any specific MACs to be added to PF DMAC filter */
2007 	if (mc_addrs) {
2008 		/* now go through kernel list of MACs and add them one by one */
2009 		for (idx = 0; idx < mc_addrs->count; idx++) {
2010 			mbx.xcast.msg = NIC_MBOX_MSG_ADD_MCAST;
2011 			mbx.xcast.mac = mc_addrs->mc[idx];
2012 			if (nicvf_send_msg_to_pf(nic, &mbx) < 0)
2013 				goto free_mc;
2014 		}
2015 	}
2016 
2017 	/* and finally set rx mode for PF accordingly */
2018 	mbx.xcast.msg = NIC_MBOX_MSG_SET_XCAST;
2019 	mbx.xcast.mode = mode;
2020 
2021 	nicvf_send_msg_to_pf(nic, &mbx);
2022 free_mc:
2023 	kfree(mc_addrs);
2024 }
2025 
2026 static void nicvf_set_rx_mode_task(struct work_struct *work_arg)
2027 {
2028 	struct nicvf_work *vf_work = container_of(work_arg, struct nicvf_work,
2029 						  work);
2030 	struct nicvf *nic = container_of(vf_work, struct nicvf, rx_mode_work);
2031 	u8 mode;
2032 	struct xcast_addr_list *mc;
2033 
2034 	if (!vf_work)
2035 		return;
2036 
2037 	/* Save message data locally to prevent them from
2038 	 * being overwritten by next ndo_set_rx_mode call().
2039 	 */
2040 	spin_lock_bh(&nic->rx_mode_wq_lock);
2041 	mode = vf_work->mode;
2042 	mc = vf_work->mc;
2043 	vf_work->mc = NULL;
2044 	spin_unlock_bh(&nic->rx_mode_wq_lock);
2045 
2046 	__nicvf_set_rx_mode_task(mode, mc, nic);
2047 }
2048 
2049 static void nicvf_set_rx_mode(struct net_device *netdev)
2050 {
2051 	struct nicvf *nic = netdev_priv(netdev);
2052 	struct netdev_hw_addr *ha;
2053 	struct xcast_addr_list *mc_list = NULL;
2054 	u8 mode = 0;
2055 
2056 	if (netdev->flags & IFF_PROMISC) {
2057 		mode = BGX_XCAST_BCAST_ACCEPT | BGX_XCAST_MCAST_ACCEPT;
2058 	} else {
2059 		if (netdev->flags & IFF_BROADCAST)
2060 			mode |= BGX_XCAST_BCAST_ACCEPT;
2061 
2062 		if (netdev->flags & IFF_ALLMULTI) {
2063 			mode |= BGX_XCAST_MCAST_ACCEPT;
2064 		} else if (netdev->flags & IFF_MULTICAST) {
2065 			mode |= BGX_XCAST_MCAST_FILTER;
2066 			/* here we need to copy mc addrs */
2067 			if (netdev_mc_count(netdev)) {
2068 				mc_list = kmalloc(struct_size(mc_list, mc,
2069 							      netdev_mc_count(netdev)),
2070 						  GFP_ATOMIC);
2071 				if (unlikely(!mc_list))
2072 					return;
2073 				mc_list->count = 0;
2074 				netdev_hw_addr_list_for_each(ha, &netdev->mc) {
2075 					mc_list->mc[mc_list->count] =
2076 						ether_addr_to_u64(ha->addr);
2077 					mc_list->count++;
2078 				}
2079 			}
2080 		}
2081 	}
2082 	spin_lock(&nic->rx_mode_wq_lock);
2083 	kfree(nic->rx_mode_work.mc);
2084 	nic->rx_mode_work.mc = mc_list;
2085 	nic->rx_mode_work.mode = mode;
2086 	queue_work(nic->nicvf_rx_mode_wq, &nic->rx_mode_work.work);
2087 	spin_unlock(&nic->rx_mode_wq_lock);
2088 }
2089 
2090 static const struct net_device_ops nicvf_netdev_ops = {
2091 	.ndo_open		= nicvf_open,
2092 	.ndo_stop		= nicvf_stop,
2093 	.ndo_start_xmit		= nicvf_xmit,
2094 	.ndo_change_mtu		= nicvf_change_mtu,
2095 	.ndo_set_mac_address	= nicvf_set_mac_address,
2096 	.ndo_get_stats64	= nicvf_get_stats64,
2097 	.ndo_tx_timeout         = nicvf_tx_timeout,
2098 	.ndo_fix_features       = nicvf_fix_features,
2099 	.ndo_set_features       = nicvf_set_features,
2100 	.ndo_bpf		= nicvf_xdp,
2101 	.ndo_do_ioctl           = nicvf_ioctl,
2102 	.ndo_set_rx_mode        = nicvf_set_rx_mode,
2103 };
2104 
2105 static int nicvf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2106 {
2107 	struct device *dev = &pdev->dev;
2108 	struct net_device *netdev;
2109 	struct nicvf *nic;
2110 	int    err, qcount;
2111 	u16    sdevid;
2112 	struct cavium_ptp *ptp_clock;
2113 
2114 	ptp_clock = cavium_ptp_get();
2115 	if (IS_ERR(ptp_clock)) {
2116 		if (PTR_ERR(ptp_clock) == -ENODEV)
2117 			/* In virtualized environment we proceed without ptp */
2118 			ptp_clock = NULL;
2119 		else
2120 			return PTR_ERR(ptp_clock);
2121 	}
2122 
2123 	err = pci_enable_device(pdev);
2124 	if (err) {
2125 		dev_err(dev, "Failed to enable PCI device\n");
2126 		return err;
2127 	}
2128 
2129 	err = pci_request_regions(pdev, DRV_NAME);
2130 	if (err) {
2131 		dev_err(dev, "PCI request regions failed 0x%x\n", err);
2132 		goto err_disable_device;
2133 	}
2134 
2135 	err = pci_set_dma_mask(pdev, DMA_BIT_MASK(48));
2136 	if (err) {
2137 		dev_err(dev, "Unable to get usable DMA configuration\n");
2138 		goto err_release_regions;
2139 	}
2140 
2141 	err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(48));
2142 	if (err) {
2143 		dev_err(dev, "unable to get 48-bit DMA for consistent allocations\n");
2144 		goto err_release_regions;
2145 	}
2146 
2147 	qcount = netif_get_num_default_rss_queues();
2148 
2149 	/* Restrict multiqset support only for host bound VFs */
2150 	if (pdev->is_virtfn) {
2151 		/* Set max number of queues per VF */
2152 		qcount = min_t(int, num_online_cpus(),
2153 			       (MAX_SQS_PER_VF + 1) * MAX_CMP_QUEUES_PER_QS);
2154 	}
2155 
2156 	netdev = alloc_etherdev_mqs(sizeof(struct nicvf), qcount, qcount);
2157 	if (!netdev) {
2158 		err = -ENOMEM;
2159 		goto err_release_regions;
2160 	}
2161 
2162 	pci_set_drvdata(pdev, netdev);
2163 
2164 	SET_NETDEV_DEV(netdev, &pdev->dev);
2165 
2166 	nic = netdev_priv(netdev);
2167 	nic->netdev = netdev;
2168 	nic->pdev = pdev;
2169 	nic->pnicvf = nic;
2170 	nic->max_queues = qcount;
2171 	/* If no of CPUs are too low, there won't be any queues left
2172 	 * for XDP_TX, hence double it.
2173 	 */
2174 	if (!nic->t88)
2175 		nic->max_queues *= 2;
2176 	nic->ptp_clock = ptp_clock;
2177 
2178 	/* Initialize mutex that serializes usage of VF's mailbox */
2179 	mutex_init(&nic->rx_mode_mtx);
2180 
2181 	/* MAP VF's configuration registers */
2182 	nic->reg_base = pcim_iomap(pdev, PCI_CFG_REG_BAR_NUM, 0);
2183 	if (!nic->reg_base) {
2184 		dev_err(dev, "Cannot map config register space, aborting\n");
2185 		err = -ENOMEM;
2186 		goto err_free_netdev;
2187 	}
2188 
2189 	nic->drv_stats = netdev_alloc_pcpu_stats(struct nicvf_drv_stats);
2190 	if (!nic->drv_stats) {
2191 		err = -ENOMEM;
2192 		goto err_free_netdev;
2193 	}
2194 
2195 	err = nicvf_set_qset_resources(nic);
2196 	if (err)
2197 		goto err_free_netdev;
2198 
2199 	/* Check if PF is alive and get MAC address for this VF */
2200 	err = nicvf_register_misc_interrupt(nic);
2201 	if (err)
2202 		goto err_free_netdev;
2203 
2204 	nicvf_send_vf_struct(nic);
2205 
2206 	if (!pass1_silicon(nic->pdev))
2207 		nic->hw_tso = true;
2208 
2209 	/* Get iommu domain for iova to physical addr conversion */
2210 	nic->iommu_domain = iommu_get_domain_for_dev(dev);
2211 
2212 	pci_read_config_word(nic->pdev, PCI_SUBSYSTEM_ID, &sdevid);
2213 	if (sdevid == 0xA134)
2214 		nic->t88 = true;
2215 
2216 	/* Check if this VF is in QS only mode */
2217 	if (nic->sqs_mode)
2218 		return 0;
2219 
2220 	err = nicvf_set_real_num_queues(netdev, nic->tx_queues, nic->rx_queues);
2221 	if (err)
2222 		goto err_unregister_interrupts;
2223 
2224 	netdev->hw_features = (NETIF_F_RXCSUM | NETIF_F_SG |
2225 			       NETIF_F_TSO | NETIF_F_GRO | NETIF_F_TSO6 |
2226 			       NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2227 			       NETIF_F_HW_VLAN_CTAG_RX);
2228 
2229 	netdev->hw_features |= NETIF_F_RXHASH;
2230 
2231 	netdev->features |= netdev->hw_features;
2232 	netdev->hw_features |= NETIF_F_LOOPBACK;
2233 
2234 	netdev->vlan_features = NETIF_F_SG | NETIF_F_IP_CSUM |
2235 				NETIF_F_IPV6_CSUM | NETIF_F_TSO | NETIF_F_TSO6;
2236 
2237 	netdev->netdev_ops = &nicvf_netdev_ops;
2238 	netdev->watchdog_timeo = NICVF_TX_TIMEOUT;
2239 
2240 	/* MTU range: 64 - 9200 */
2241 	netdev->min_mtu = NIC_HW_MIN_FRS;
2242 	netdev->max_mtu = NIC_HW_MAX_FRS;
2243 
2244 	INIT_WORK(&nic->reset_task, nicvf_reset_task);
2245 
2246 	nic->nicvf_rx_mode_wq = alloc_ordered_workqueue("nicvf_rx_mode_wq_VF%d",
2247 							WQ_MEM_RECLAIM,
2248 							nic->vf_id);
2249 	if (!nic->nicvf_rx_mode_wq) {
2250 		err = -ENOMEM;
2251 		dev_err(dev, "Failed to allocate work queue\n");
2252 		goto err_unregister_interrupts;
2253 	}
2254 
2255 	INIT_WORK(&nic->rx_mode_work.work, nicvf_set_rx_mode_task);
2256 	spin_lock_init(&nic->rx_mode_wq_lock);
2257 
2258 	err = register_netdev(netdev);
2259 	if (err) {
2260 		dev_err(dev, "Failed to register netdevice\n");
2261 		goto err_unregister_interrupts;
2262 	}
2263 
2264 	nic->msg_enable = debug;
2265 
2266 	nicvf_set_ethtool_ops(netdev);
2267 
2268 	return 0;
2269 
2270 err_unregister_interrupts:
2271 	nicvf_unregister_interrupts(nic);
2272 err_free_netdev:
2273 	pci_set_drvdata(pdev, NULL);
2274 	if (nic->drv_stats)
2275 		free_percpu(nic->drv_stats);
2276 	free_netdev(netdev);
2277 err_release_regions:
2278 	pci_release_regions(pdev);
2279 err_disable_device:
2280 	pci_disable_device(pdev);
2281 	return err;
2282 }
2283 
2284 static void nicvf_remove(struct pci_dev *pdev)
2285 {
2286 	struct net_device *netdev = pci_get_drvdata(pdev);
2287 	struct nicvf *nic;
2288 	struct net_device *pnetdev;
2289 
2290 	if (!netdev)
2291 		return;
2292 
2293 	nic = netdev_priv(netdev);
2294 	pnetdev = nic->pnicvf->netdev;
2295 
2296 	/* Check if this Qset is assigned to different VF.
2297 	 * If yes, clean primary and all secondary Qsets.
2298 	 */
2299 	if (pnetdev && (pnetdev->reg_state == NETREG_REGISTERED))
2300 		unregister_netdev(pnetdev);
2301 	if (nic->nicvf_rx_mode_wq) {
2302 		destroy_workqueue(nic->nicvf_rx_mode_wq);
2303 		nic->nicvf_rx_mode_wq = NULL;
2304 	}
2305 	nicvf_unregister_interrupts(nic);
2306 	pci_set_drvdata(pdev, NULL);
2307 	if (nic->drv_stats)
2308 		free_percpu(nic->drv_stats);
2309 	cavium_ptp_put(nic->ptp_clock);
2310 	free_netdev(netdev);
2311 	pci_release_regions(pdev);
2312 	pci_disable_device(pdev);
2313 }
2314 
2315 static void nicvf_shutdown(struct pci_dev *pdev)
2316 {
2317 	nicvf_remove(pdev);
2318 }
2319 
2320 static struct pci_driver nicvf_driver = {
2321 	.name = DRV_NAME,
2322 	.id_table = nicvf_id_table,
2323 	.probe = nicvf_probe,
2324 	.remove = nicvf_remove,
2325 	.shutdown = nicvf_shutdown,
2326 };
2327 
2328 static int __init nicvf_init_module(void)
2329 {
2330 	pr_info("%s, ver %s\n", DRV_NAME, DRV_VERSION);
2331 	return pci_register_driver(&nicvf_driver);
2332 }
2333 
2334 static void __exit nicvf_cleanup_module(void)
2335 {
2336 	pci_unregister_driver(&nicvf_driver);
2337 }
2338 
2339 module_init(nicvf_init_module);
2340 module_exit(nicvf_cleanup_module);
2341