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 	int timeout = NIC_MBOX_MSG_TIMEOUT;
130 	int sleep = 10;
131 	int ret = 0;
132 
133 	mutex_lock(&nic->rx_mode_mtx);
134 
135 	nic->pf_acked = false;
136 	nic->pf_nacked = false;
137 
138 	nicvf_write_to_mbx(nic, mbx);
139 
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 		msleep(sleep);
150 		if (nic->pf_acked)
151 			break;
152 		timeout -= sleep;
153 		if (!timeout) {
154 			netdev_err(nic->netdev,
155 				   "PF didn't ACK to mbox msg 0x%02x from VF%d\n",
156 				   (mbx->msg.msg & 0xFF), nic->vf_id);
157 			ret = -EBUSY;
158 			break;
159 		}
160 	}
161 	mutex_unlock(&nic->rx_mode_mtx);
162 	return ret;
163 }
164 
165 /* Checks if VF is able to comminicate with PF
166 * and also gets the VNIC number this VF is associated to.
167 */
168 static int nicvf_check_pf_ready(struct nicvf *nic)
169 {
170 	union nic_mbx mbx = {};
171 
172 	mbx.msg.msg = NIC_MBOX_MSG_READY;
173 	if (nicvf_send_msg_to_pf(nic, &mbx)) {
174 		netdev_err(nic->netdev,
175 			   "PF didn't respond to READY msg\n");
176 		return 0;
177 	}
178 
179 	return 1;
180 }
181 
182 static void nicvf_send_cfg_done(struct nicvf *nic)
183 {
184 	union nic_mbx mbx = {};
185 
186 	mbx.msg.msg = NIC_MBOX_MSG_CFG_DONE;
187 	if (nicvf_send_msg_to_pf(nic, &mbx)) {
188 		netdev_err(nic->netdev,
189 			   "PF didn't respond to CFG DONE msg\n");
190 	}
191 }
192 
193 static void nicvf_read_bgx_stats(struct nicvf *nic, struct bgx_stats_msg *bgx)
194 {
195 	if (bgx->rx)
196 		nic->bgx_stats.rx_stats[bgx->idx] = bgx->stats;
197 	else
198 		nic->bgx_stats.tx_stats[bgx->idx] = bgx->stats;
199 }
200 
201 static void  nicvf_handle_mbx_intr(struct nicvf *nic)
202 {
203 	union nic_mbx mbx = {};
204 	u64 *mbx_data;
205 	u64 mbx_addr;
206 	int i;
207 
208 	mbx_addr = NIC_VF_PF_MAILBOX_0_1;
209 	mbx_data = (u64 *)&mbx;
210 
211 	for (i = 0; i < NIC_PF_VF_MAILBOX_SIZE; i++) {
212 		*mbx_data = nicvf_reg_read(nic, mbx_addr);
213 		mbx_data++;
214 		mbx_addr += sizeof(u64);
215 	}
216 
217 	netdev_dbg(nic->netdev, "Mbox message: msg: 0x%x\n", mbx.msg.msg);
218 	switch (mbx.msg.msg) {
219 	case NIC_MBOX_MSG_READY:
220 		nic->pf_acked = true;
221 		nic->vf_id = mbx.nic_cfg.vf_id & 0x7F;
222 		nic->tns_mode = mbx.nic_cfg.tns_mode & 0x7F;
223 		nic->node = mbx.nic_cfg.node_id;
224 		if (!nic->set_mac_pending)
225 			ether_addr_copy(nic->netdev->dev_addr,
226 					mbx.nic_cfg.mac_addr);
227 		nic->sqs_mode = mbx.nic_cfg.sqs_mode;
228 		nic->loopback_supported = mbx.nic_cfg.loopback_supported;
229 		nic->link_up = false;
230 		nic->duplex = 0;
231 		nic->speed = 0;
232 		break;
233 	case NIC_MBOX_MSG_ACK:
234 		nic->pf_acked = true;
235 		break;
236 	case NIC_MBOX_MSG_NACK:
237 		nic->pf_nacked = true;
238 		break;
239 	case NIC_MBOX_MSG_RSS_SIZE:
240 		nic->rss_info.rss_size = mbx.rss_size.ind_tbl_size;
241 		nic->pf_acked = true;
242 		break;
243 	case NIC_MBOX_MSG_BGX_STATS:
244 		nicvf_read_bgx_stats(nic, &mbx.bgx_stats);
245 		nic->pf_acked = true;
246 		break;
247 	case NIC_MBOX_MSG_BGX_LINK_CHANGE:
248 		nic->pf_acked = true;
249 		if (nic->link_up != mbx.link_status.link_up) {
250 			nic->link_up = mbx.link_status.link_up;
251 			nic->duplex = mbx.link_status.duplex;
252 			nic->speed = mbx.link_status.speed;
253 			nic->mac_type = mbx.link_status.mac_type;
254 			if (nic->link_up) {
255 				netdev_info(nic->netdev,
256 					    "Link is Up %d Mbps %s duplex\n",
257 					    nic->speed,
258 					    nic->duplex == DUPLEX_FULL ?
259 					    "Full" : "Half");
260 				netif_carrier_on(nic->netdev);
261 				netif_tx_start_all_queues(nic->netdev);
262 			} else {
263 				netdev_info(nic->netdev, "Link is Down\n");
264 				netif_carrier_off(nic->netdev);
265 				netif_tx_stop_all_queues(nic->netdev);
266 			}
267 		}
268 		break;
269 	case NIC_MBOX_MSG_ALLOC_SQS:
270 		nic->sqs_count = mbx.sqs_alloc.qs_count;
271 		nic->pf_acked = true;
272 		break;
273 	case NIC_MBOX_MSG_SNICVF_PTR:
274 		/* Primary VF: make note of secondary VF's pointer
275 		 * to be used while packet transmission.
276 		 */
277 		nic->snicvf[mbx.nicvf.sqs_id] =
278 			(struct nicvf *)mbx.nicvf.nicvf;
279 		nic->pf_acked = true;
280 		break;
281 	case NIC_MBOX_MSG_PNICVF_PTR:
282 		/* Secondary VF/Qset: make note of primary VF's pointer
283 		 * to be used while packet reception, to handover packet
284 		 * to primary VF's netdev.
285 		 */
286 		nic->pnicvf = (struct nicvf *)mbx.nicvf.nicvf;
287 		nic->pf_acked = true;
288 		break;
289 	case NIC_MBOX_MSG_PFC:
290 		nic->pfc.autoneg = mbx.pfc.autoneg;
291 		nic->pfc.fc_rx = mbx.pfc.fc_rx;
292 		nic->pfc.fc_tx = mbx.pfc.fc_tx;
293 		nic->pf_acked = true;
294 		break;
295 	default:
296 		netdev_err(nic->netdev,
297 			   "Invalid message from PF, msg 0x%x\n", mbx.msg.msg);
298 		break;
299 	}
300 	nicvf_clear_intr(nic, NICVF_INTR_MBOX, 0);
301 }
302 
303 static int nicvf_hw_set_mac_addr(struct nicvf *nic, struct net_device *netdev)
304 {
305 	union nic_mbx mbx = {};
306 
307 	mbx.mac.msg = NIC_MBOX_MSG_SET_MAC;
308 	mbx.mac.vf_id = nic->vf_id;
309 	ether_addr_copy(mbx.mac.mac_addr, netdev->dev_addr);
310 
311 	return nicvf_send_msg_to_pf(nic, &mbx);
312 }
313 
314 static void nicvf_config_cpi(struct nicvf *nic)
315 {
316 	union nic_mbx mbx = {};
317 
318 	mbx.cpi_cfg.msg = NIC_MBOX_MSG_CPI_CFG;
319 	mbx.cpi_cfg.vf_id = nic->vf_id;
320 	mbx.cpi_cfg.cpi_alg = nic->cpi_alg;
321 	mbx.cpi_cfg.rq_cnt = nic->qs->rq_cnt;
322 
323 	nicvf_send_msg_to_pf(nic, &mbx);
324 }
325 
326 static void nicvf_get_rss_size(struct nicvf *nic)
327 {
328 	union nic_mbx mbx = {};
329 
330 	mbx.rss_size.msg = NIC_MBOX_MSG_RSS_SIZE;
331 	mbx.rss_size.vf_id = nic->vf_id;
332 	nicvf_send_msg_to_pf(nic, &mbx);
333 }
334 
335 void nicvf_config_rss(struct nicvf *nic)
336 {
337 	union nic_mbx mbx = {};
338 	struct nicvf_rss_info *rss = &nic->rss_info;
339 	int ind_tbl_len = rss->rss_size;
340 	int i, nextq = 0;
341 
342 	mbx.rss_cfg.vf_id = nic->vf_id;
343 	mbx.rss_cfg.hash_bits = rss->hash_bits;
344 	while (ind_tbl_len) {
345 		mbx.rss_cfg.tbl_offset = nextq;
346 		mbx.rss_cfg.tbl_len = min(ind_tbl_len,
347 					       RSS_IND_TBL_LEN_PER_MBX_MSG);
348 		mbx.rss_cfg.msg = mbx.rss_cfg.tbl_offset ?
349 			  NIC_MBOX_MSG_RSS_CFG_CONT : NIC_MBOX_MSG_RSS_CFG;
350 
351 		for (i = 0; i < mbx.rss_cfg.tbl_len; i++)
352 			mbx.rss_cfg.ind_tbl[i] = rss->ind_tbl[nextq++];
353 
354 		nicvf_send_msg_to_pf(nic, &mbx);
355 
356 		ind_tbl_len -= mbx.rss_cfg.tbl_len;
357 	}
358 }
359 
360 void nicvf_set_rss_key(struct nicvf *nic)
361 {
362 	struct nicvf_rss_info *rss = &nic->rss_info;
363 	u64 key_addr = NIC_VNIC_RSS_KEY_0_4;
364 	int idx;
365 
366 	for (idx = 0; idx < RSS_HASH_KEY_SIZE; idx++) {
367 		nicvf_reg_write(nic, key_addr, rss->key[idx]);
368 		key_addr += sizeof(u64);
369 	}
370 }
371 
372 static int nicvf_rss_init(struct nicvf *nic)
373 {
374 	struct nicvf_rss_info *rss = &nic->rss_info;
375 	int idx;
376 
377 	nicvf_get_rss_size(nic);
378 
379 	if (cpi_alg != CPI_ALG_NONE) {
380 		rss->enable = false;
381 		rss->hash_bits = 0;
382 		return 0;
383 	}
384 
385 	rss->enable = true;
386 
387 	netdev_rss_key_fill(rss->key, RSS_HASH_KEY_SIZE * sizeof(u64));
388 	nicvf_set_rss_key(nic);
389 
390 	rss->cfg = RSS_IP_HASH_ENA | RSS_TCP_HASH_ENA | RSS_UDP_HASH_ENA;
391 	nicvf_reg_write(nic, NIC_VNIC_RSS_CFG, rss->cfg);
392 
393 	rss->hash_bits =  ilog2(rounddown_pow_of_two(rss->rss_size));
394 
395 	for (idx = 0; idx < rss->rss_size; idx++)
396 		rss->ind_tbl[idx] = ethtool_rxfh_indir_default(idx,
397 							       nic->rx_queues);
398 	nicvf_config_rss(nic);
399 	return 1;
400 }
401 
402 /* Request PF to allocate additional Qsets */
403 static void nicvf_request_sqs(struct nicvf *nic)
404 {
405 	union nic_mbx mbx = {};
406 	int sqs;
407 	int sqs_count = nic->sqs_count;
408 	int rx_queues = 0, tx_queues = 0;
409 
410 	/* Only primary VF should request */
411 	if (nic->sqs_mode ||  !nic->sqs_count)
412 		return;
413 
414 	mbx.sqs_alloc.msg = NIC_MBOX_MSG_ALLOC_SQS;
415 	mbx.sqs_alloc.vf_id = nic->vf_id;
416 	mbx.sqs_alloc.qs_count = nic->sqs_count;
417 	if (nicvf_send_msg_to_pf(nic, &mbx)) {
418 		/* No response from PF */
419 		nic->sqs_count = 0;
420 		return;
421 	}
422 
423 	/* Return if no Secondary Qsets available */
424 	if (!nic->sqs_count)
425 		return;
426 
427 	if (nic->rx_queues > MAX_RCV_QUEUES_PER_QS)
428 		rx_queues = nic->rx_queues - MAX_RCV_QUEUES_PER_QS;
429 
430 	tx_queues = nic->tx_queues + nic->xdp_tx_queues;
431 	if (tx_queues > MAX_SND_QUEUES_PER_QS)
432 		tx_queues = tx_queues - MAX_SND_QUEUES_PER_QS;
433 
434 	/* Set no of Rx/Tx queues in each of the SQsets */
435 	for (sqs = 0; sqs < nic->sqs_count; sqs++) {
436 		mbx.nicvf.msg = NIC_MBOX_MSG_SNICVF_PTR;
437 		mbx.nicvf.vf_id = nic->vf_id;
438 		mbx.nicvf.sqs_id = sqs;
439 		nicvf_send_msg_to_pf(nic, &mbx);
440 
441 		nic->snicvf[sqs]->sqs_id = sqs;
442 		if (rx_queues > MAX_RCV_QUEUES_PER_QS) {
443 			nic->snicvf[sqs]->qs->rq_cnt = MAX_RCV_QUEUES_PER_QS;
444 			rx_queues -= MAX_RCV_QUEUES_PER_QS;
445 		} else {
446 			nic->snicvf[sqs]->qs->rq_cnt = rx_queues;
447 			rx_queues = 0;
448 		}
449 
450 		if (tx_queues > MAX_SND_QUEUES_PER_QS) {
451 			nic->snicvf[sqs]->qs->sq_cnt = MAX_SND_QUEUES_PER_QS;
452 			tx_queues -= MAX_SND_QUEUES_PER_QS;
453 		} else {
454 			nic->snicvf[sqs]->qs->sq_cnt = tx_queues;
455 			tx_queues = 0;
456 		}
457 
458 		nic->snicvf[sqs]->qs->cq_cnt =
459 		max(nic->snicvf[sqs]->qs->rq_cnt, nic->snicvf[sqs]->qs->sq_cnt);
460 
461 		/* Initialize secondary Qset's queues and its interrupts */
462 		nicvf_open(nic->snicvf[sqs]->netdev);
463 	}
464 
465 	/* Update stack with actual Rx/Tx queue count allocated */
466 	if (sqs_count != nic->sqs_count)
467 		nicvf_set_real_num_queues(nic->netdev,
468 					  nic->tx_queues, nic->rx_queues);
469 }
470 
471 /* Send this Qset's nicvf pointer to PF.
472  * PF inturn sends primary VF's nicvf struct to secondary Qsets/VFs
473  * so that packets received by these Qsets can use primary VF's netdev
474  */
475 static void nicvf_send_vf_struct(struct nicvf *nic)
476 {
477 	union nic_mbx mbx = {};
478 
479 	mbx.nicvf.msg = NIC_MBOX_MSG_NICVF_PTR;
480 	mbx.nicvf.sqs_mode = nic->sqs_mode;
481 	mbx.nicvf.nicvf = (u64)nic;
482 	nicvf_send_msg_to_pf(nic, &mbx);
483 }
484 
485 static void nicvf_get_primary_vf_struct(struct nicvf *nic)
486 {
487 	union nic_mbx mbx = {};
488 
489 	mbx.nicvf.msg = NIC_MBOX_MSG_PNICVF_PTR;
490 	nicvf_send_msg_to_pf(nic, &mbx);
491 }
492 
493 int nicvf_set_real_num_queues(struct net_device *netdev,
494 			      int tx_queues, int rx_queues)
495 {
496 	int err = 0;
497 
498 	err = netif_set_real_num_tx_queues(netdev, tx_queues);
499 	if (err) {
500 		netdev_err(netdev,
501 			   "Failed to set no of Tx queues: %d\n", tx_queues);
502 		return err;
503 	}
504 
505 	err = netif_set_real_num_rx_queues(netdev, rx_queues);
506 	if (err)
507 		netdev_err(netdev,
508 			   "Failed to set no of Rx queues: %d\n", rx_queues);
509 	return err;
510 }
511 
512 static int nicvf_init_resources(struct nicvf *nic)
513 {
514 	int err;
515 
516 	/* Enable Qset */
517 	nicvf_qset_config(nic, true);
518 
519 	/* Initialize queues and HW for data transfer */
520 	err = nicvf_config_data_transfer(nic, true);
521 	if (err) {
522 		netdev_err(nic->netdev,
523 			   "Failed to alloc/config VF's QSet resources\n");
524 		return err;
525 	}
526 
527 	return 0;
528 }
529 
530 static inline bool nicvf_xdp_rx(struct nicvf *nic, struct bpf_prog *prog,
531 				struct cqe_rx_t *cqe_rx, struct snd_queue *sq,
532 				struct rcv_queue *rq, struct sk_buff **skb)
533 {
534 	struct xdp_buff xdp;
535 	struct page *page;
536 	u32 action;
537 	u16 len, offset = 0;
538 	u64 dma_addr, cpu_addr;
539 	void *orig_data;
540 
541 	/* Retrieve packet buffer's DMA address and length */
542 	len = *((u16 *)((void *)cqe_rx + (3 * sizeof(u64))));
543 	dma_addr = *((u64 *)((void *)cqe_rx + (7 * sizeof(u64))));
544 
545 	cpu_addr = nicvf_iova_to_phys(nic, dma_addr);
546 	if (!cpu_addr)
547 		return false;
548 	cpu_addr = (u64)phys_to_virt(cpu_addr);
549 	page = virt_to_page((void *)cpu_addr);
550 
551 	xdp.data_hard_start = page_address(page);
552 	xdp.data = (void *)cpu_addr;
553 	xdp_set_data_meta_invalid(&xdp);
554 	xdp.data_end = xdp.data + len;
555 	xdp.rxq = &rq->xdp_rxq;
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 		/* fall through */
598 	case XDP_ABORTED:
599 		trace_xdp_exception(nic->netdev, prog, action);
600 		/* fall through */
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(unsigned long data)
989 {
990 	struct nicvf *nic = (struct nicvf *)data;
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_init(&nic->qs_err_task, nicvf_handle_qs_err,
1497 		     (unsigned long)nic);
1498 
1499 	/* Init RBDR tasklet which will refill RBDR */
1500 	tasklet_init(&nic->rbdr_task, nicvf_rbdr_task,
1501 		     (unsigned long)nic);
1502 	INIT_DELAYED_WORK(&nic->rbdr_work, nicvf_rbdr_work);
1503 
1504 	/* Configure CPI alorithm */
1505 	nic->cpi_alg = cpi_alg;
1506 	if (!nic->sqs_mode)
1507 		nicvf_config_cpi(nic);
1508 
1509 	nicvf_request_sqs(nic);
1510 	if (nic->sqs_mode)
1511 		nicvf_get_primary_vf_struct(nic);
1512 
1513 	/* Configure PTP timestamp */
1514 	if (nic->ptp_clock)
1515 		nicvf_config_hw_rx_tstamp(nic, nic->hw_rx_tstamp);
1516 	atomic_set(&nic->tx_ptp_skbs, 0);
1517 	nic->ptp_skb = NULL;
1518 
1519 	/* Configure receive side scaling and MTU */
1520 	if (!nic->sqs_mode) {
1521 		nicvf_rss_init(nic);
1522 		err = nicvf_update_hw_max_frs(nic, netdev->mtu);
1523 		if (err)
1524 			goto cleanup;
1525 
1526 		/* Clear percpu stats */
1527 		for_each_possible_cpu(cpu)
1528 			memset(per_cpu_ptr(nic->drv_stats, cpu), 0,
1529 			       sizeof(struct nicvf_drv_stats));
1530 	}
1531 
1532 	err = nicvf_register_interrupts(nic);
1533 	if (err)
1534 		goto cleanup;
1535 
1536 	/* Initialize the queues */
1537 	err = nicvf_init_resources(nic);
1538 	if (err)
1539 		goto cleanup;
1540 
1541 	/* Make sure queue initialization is written */
1542 	wmb();
1543 
1544 	nicvf_reg_write(nic, NIC_VF_INT, -1);
1545 	/* Enable Qset err interrupt */
1546 	nicvf_enable_intr(nic, NICVF_INTR_QS_ERR, 0);
1547 
1548 	/* Enable completion queue interrupt */
1549 	for (qidx = 0; qidx < qs->cq_cnt; qidx++)
1550 		nicvf_enable_intr(nic, NICVF_INTR_CQ, qidx);
1551 
1552 	/* Enable RBDR threshold interrupt */
1553 	for (qidx = 0; qidx < qs->rbdr_cnt; qidx++)
1554 		nicvf_enable_intr(nic, NICVF_INTR_RBDR, qidx);
1555 
1556 	/* Send VF config done msg to PF */
1557 	nicvf_send_cfg_done(nic);
1558 
1559 	if (nic->nicvf_rx_mode_wq) {
1560 		INIT_DELAYED_WORK(&nic->link_change_work,
1561 				  nicvf_link_status_check_task);
1562 		queue_delayed_work(nic->nicvf_rx_mode_wq,
1563 				   &nic->link_change_work, 0);
1564 	}
1565 
1566 	return 0;
1567 cleanup:
1568 	nicvf_disable_intr(nic, NICVF_INTR_MBOX, 0);
1569 	nicvf_unregister_interrupts(nic);
1570 	tasklet_kill(&nic->qs_err_task);
1571 	tasklet_kill(&nic->rbdr_task);
1572 napi_del:
1573 	for (qidx = 0; qidx < qs->cq_cnt; qidx++) {
1574 		cq_poll = nic->napi[qidx];
1575 		if (!cq_poll)
1576 			continue;
1577 		napi_disable(&cq_poll->napi);
1578 		netif_napi_del(&cq_poll->napi);
1579 	}
1580 	nicvf_free_cq_poll(nic);
1581 	return err;
1582 }
1583 
1584 static int nicvf_change_mtu(struct net_device *netdev, int new_mtu)
1585 {
1586 	struct nicvf *nic = netdev_priv(netdev);
1587 	int orig_mtu = netdev->mtu;
1588 
1589 	/* For now just support only the usual MTU sized frames,
1590 	 * plus some headroom for VLAN, QinQ.
1591 	 */
1592 	if (nic->xdp_prog && new_mtu > MAX_XDP_MTU) {
1593 		netdev_warn(netdev, "Jumbo frames not yet supported with XDP, current MTU %d.\n",
1594 			    netdev->mtu);
1595 		return -EINVAL;
1596 	}
1597 
1598 	netdev->mtu = new_mtu;
1599 
1600 	if (!netif_running(netdev))
1601 		return 0;
1602 
1603 	if (nicvf_update_hw_max_frs(nic, new_mtu)) {
1604 		netdev->mtu = orig_mtu;
1605 		return -EINVAL;
1606 	}
1607 
1608 	return 0;
1609 }
1610 
1611 static int nicvf_set_mac_address(struct net_device *netdev, void *p)
1612 {
1613 	struct sockaddr *addr = p;
1614 	struct nicvf *nic = netdev_priv(netdev);
1615 
1616 	if (!is_valid_ether_addr(addr->sa_data))
1617 		return -EADDRNOTAVAIL;
1618 
1619 	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1620 
1621 	if (nic->pdev->msix_enabled) {
1622 		if (nicvf_hw_set_mac_addr(nic, netdev))
1623 			return -EBUSY;
1624 	} else {
1625 		nic->set_mac_pending = true;
1626 	}
1627 
1628 	return 0;
1629 }
1630 
1631 void nicvf_update_lmac_stats(struct nicvf *nic)
1632 {
1633 	int stat = 0;
1634 	union nic_mbx mbx = {};
1635 
1636 	if (!netif_running(nic->netdev))
1637 		return;
1638 
1639 	mbx.bgx_stats.msg = NIC_MBOX_MSG_BGX_STATS;
1640 	mbx.bgx_stats.vf_id = nic->vf_id;
1641 	/* Rx stats */
1642 	mbx.bgx_stats.rx = 1;
1643 	while (stat < BGX_RX_STATS_COUNT) {
1644 		mbx.bgx_stats.idx = stat;
1645 		if (nicvf_send_msg_to_pf(nic, &mbx))
1646 			return;
1647 		stat++;
1648 	}
1649 
1650 	stat = 0;
1651 
1652 	/* Tx stats */
1653 	mbx.bgx_stats.rx = 0;
1654 	while (stat < BGX_TX_STATS_COUNT) {
1655 		mbx.bgx_stats.idx = stat;
1656 		if (nicvf_send_msg_to_pf(nic, &mbx))
1657 			return;
1658 		stat++;
1659 	}
1660 }
1661 
1662 void nicvf_update_stats(struct nicvf *nic)
1663 {
1664 	int qidx, cpu;
1665 	u64 tmp_stats = 0;
1666 	struct nicvf_hw_stats *stats = &nic->hw_stats;
1667 	struct nicvf_drv_stats *drv_stats;
1668 	struct queue_set *qs = nic->qs;
1669 
1670 #define GET_RX_STATS(reg) \
1671 	nicvf_reg_read(nic, NIC_VNIC_RX_STAT_0_13 | (reg << 3))
1672 #define GET_TX_STATS(reg) \
1673 	nicvf_reg_read(nic, NIC_VNIC_TX_STAT_0_4 | (reg << 3))
1674 
1675 	stats->rx_bytes = GET_RX_STATS(RX_OCTS);
1676 	stats->rx_ucast_frames = GET_RX_STATS(RX_UCAST);
1677 	stats->rx_bcast_frames = GET_RX_STATS(RX_BCAST);
1678 	stats->rx_mcast_frames = GET_RX_STATS(RX_MCAST);
1679 	stats->rx_fcs_errors = GET_RX_STATS(RX_FCS);
1680 	stats->rx_l2_errors = GET_RX_STATS(RX_L2ERR);
1681 	stats->rx_drop_red = GET_RX_STATS(RX_RED);
1682 	stats->rx_drop_red_bytes = GET_RX_STATS(RX_RED_OCTS);
1683 	stats->rx_drop_overrun = GET_RX_STATS(RX_ORUN);
1684 	stats->rx_drop_overrun_bytes = GET_RX_STATS(RX_ORUN_OCTS);
1685 	stats->rx_drop_bcast = GET_RX_STATS(RX_DRP_BCAST);
1686 	stats->rx_drop_mcast = GET_RX_STATS(RX_DRP_MCAST);
1687 	stats->rx_drop_l3_bcast = GET_RX_STATS(RX_DRP_L3BCAST);
1688 	stats->rx_drop_l3_mcast = GET_RX_STATS(RX_DRP_L3MCAST);
1689 
1690 	stats->tx_bytes = GET_TX_STATS(TX_OCTS);
1691 	stats->tx_ucast_frames = GET_TX_STATS(TX_UCAST);
1692 	stats->tx_bcast_frames = GET_TX_STATS(TX_BCAST);
1693 	stats->tx_mcast_frames = GET_TX_STATS(TX_MCAST);
1694 	stats->tx_drops = GET_TX_STATS(TX_DROP);
1695 
1696 	/* On T88 pass 2.0, the dummy SQE added for TSO notification
1697 	 * via CQE has 'dont_send' set. Hence HW drops the pkt pointed
1698 	 * pointed by dummy SQE and results in tx_drops counter being
1699 	 * incremented. Subtracting it from tx_tso counter will give
1700 	 * exact tx_drops counter.
1701 	 */
1702 	if (nic->t88 && nic->hw_tso) {
1703 		for_each_possible_cpu(cpu) {
1704 			drv_stats = per_cpu_ptr(nic->drv_stats, cpu);
1705 			tmp_stats += drv_stats->tx_tso;
1706 		}
1707 		stats->tx_drops = tmp_stats - stats->tx_drops;
1708 	}
1709 	stats->tx_frames = stats->tx_ucast_frames +
1710 			   stats->tx_bcast_frames +
1711 			   stats->tx_mcast_frames;
1712 	stats->rx_frames = stats->rx_ucast_frames +
1713 			   stats->rx_bcast_frames +
1714 			   stats->rx_mcast_frames;
1715 	stats->rx_drops = stats->rx_drop_red +
1716 			  stats->rx_drop_overrun;
1717 
1718 	/* Update RQ and SQ stats */
1719 	for (qidx = 0; qidx < qs->rq_cnt; qidx++)
1720 		nicvf_update_rq_stats(nic, qidx);
1721 	for (qidx = 0; qidx < qs->sq_cnt; qidx++)
1722 		nicvf_update_sq_stats(nic, qidx);
1723 }
1724 
1725 static void nicvf_get_stats64(struct net_device *netdev,
1726 			      struct rtnl_link_stats64 *stats)
1727 {
1728 	struct nicvf *nic = netdev_priv(netdev);
1729 	struct nicvf_hw_stats *hw_stats = &nic->hw_stats;
1730 
1731 	nicvf_update_stats(nic);
1732 
1733 	stats->rx_bytes = hw_stats->rx_bytes;
1734 	stats->rx_packets = hw_stats->rx_frames;
1735 	stats->rx_dropped = hw_stats->rx_drops;
1736 	stats->multicast = hw_stats->rx_mcast_frames;
1737 
1738 	stats->tx_bytes = hw_stats->tx_bytes;
1739 	stats->tx_packets = hw_stats->tx_frames;
1740 	stats->tx_dropped = hw_stats->tx_drops;
1741 
1742 }
1743 
1744 static void nicvf_tx_timeout(struct net_device *dev)
1745 {
1746 	struct nicvf *nic = netdev_priv(dev);
1747 
1748 	netif_warn(nic, tx_err, dev, "Transmit timed out, resetting\n");
1749 
1750 	this_cpu_inc(nic->drv_stats->tx_timeout);
1751 	schedule_work(&nic->reset_task);
1752 }
1753 
1754 static void nicvf_reset_task(struct work_struct *work)
1755 {
1756 	struct nicvf *nic;
1757 
1758 	nic = container_of(work, struct nicvf, reset_task);
1759 
1760 	if (!netif_running(nic->netdev))
1761 		return;
1762 
1763 	nicvf_stop(nic->netdev);
1764 	nicvf_open(nic->netdev);
1765 	netif_trans_update(nic->netdev);
1766 }
1767 
1768 static int nicvf_config_loopback(struct nicvf *nic,
1769 				 netdev_features_t features)
1770 {
1771 	union nic_mbx mbx = {};
1772 
1773 	mbx.lbk.msg = NIC_MBOX_MSG_LOOPBACK;
1774 	mbx.lbk.vf_id = nic->vf_id;
1775 	mbx.lbk.enable = (features & NETIF_F_LOOPBACK) != 0;
1776 
1777 	return nicvf_send_msg_to_pf(nic, &mbx);
1778 }
1779 
1780 static netdev_features_t nicvf_fix_features(struct net_device *netdev,
1781 					    netdev_features_t features)
1782 {
1783 	struct nicvf *nic = netdev_priv(netdev);
1784 
1785 	if ((features & NETIF_F_LOOPBACK) &&
1786 	    netif_running(netdev) && !nic->loopback_supported)
1787 		features &= ~NETIF_F_LOOPBACK;
1788 
1789 	return features;
1790 }
1791 
1792 static int nicvf_set_features(struct net_device *netdev,
1793 			      netdev_features_t features)
1794 {
1795 	struct nicvf *nic = netdev_priv(netdev);
1796 	netdev_features_t changed = features ^ netdev->features;
1797 
1798 	if (changed & NETIF_F_HW_VLAN_CTAG_RX)
1799 		nicvf_config_vlan_stripping(nic, features);
1800 
1801 	if ((changed & NETIF_F_LOOPBACK) && netif_running(netdev))
1802 		return nicvf_config_loopback(nic, features);
1803 
1804 	return 0;
1805 }
1806 
1807 static void nicvf_set_xdp_queues(struct nicvf *nic, bool bpf_attached)
1808 {
1809 	u8 cq_count, txq_count;
1810 
1811 	/* Set XDP Tx queue count same as Rx queue count */
1812 	if (!bpf_attached)
1813 		nic->xdp_tx_queues = 0;
1814 	else
1815 		nic->xdp_tx_queues = nic->rx_queues;
1816 
1817 	/* If queue count > MAX_CMP_QUEUES_PER_QS, then additional qsets
1818 	 * needs to be allocated, check how many.
1819 	 */
1820 	txq_count = nic->xdp_tx_queues + nic->tx_queues;
1821 	cq_count = max(nic->rx_queues, txq_count);
1822 	if (cq_count > MAX_CMP_QUEUES_PER_QS) {
1823 		nic->sqs_count = roundup(cq_count, MAX_CMP_QUEUES_PER_QS);
1824 		nic->sqs_count = (nic->sqs_count / MAX_CMP_QUEUES_PER_QS) - 1;
1825 	} else {
1826 		nic->sqs_count = 0;
1827 	}
1828 
1829 	/* Set primary Qset's resources */
1830 	nic->qs->rq_cnt = min_t(u8, nic->rx_queues, MAX_RCV_QUEUES_PER_QS);
1831 	nic->qs->sq_cnt = min_t(u8, txq_count, MAX_SND_QUEUES_PER_QS);
1832 	nic->qs->cq_cnt = max_t(u8, nic->qs->rq_cnt, nic->qs->sq_cnt);
1833 
1834 	/* Update stack */
1835 	nicvf_set_real_num_queues(nic->netdev, nic->tx_queues, nic->rx_queues);
1836 }
1837 
1838 static int nicvf_xdp_setup(struct nicvf *nic, struct bpf_prog *prog)
1839 {
1840 	struct net_device *dev = nic->netdev;
1841 	bool if_up = netif_running(nic->netdev);
1842 	struct bpf_prog *old_prog;
1843 	bool bpf_attached = false;
1844 	int ret = 0;
1845 
1846 	/* For now just support only the usual MTU sized frames,
1847 	 * plus some headroom for VLAN, QinQ.
1848 	 */
1849 	if (prog && dev->mtu > MAX_XDP_MTU) {
1850 		netdev_warn(dev, "Jumbo frames not yet supported with XDP, current MTU %d.\n",
1851 			    dev->mtu);
1852 		return -EOPNOTSUPP;
1853 	}
1854 
1855 	/* ALL SQs attached to CQs i.e same as RQs, are treated as
1856 	 * XDP Tx queues and more Tx queues are allocated for
1857 	 * network stack to send pkts out.
1858 	 *
1859 	 * No of Tx queues are either same as Rx queues or whatever
1860 	 * is left in max no of queues possible.
1861 	 */
1862 	if ((nic->rx_queues + nic->tx_queues) > nic->max_queues) {
1863 		netdev_warn(dev,
1864 			    "Failed to attach BPF prog, RXQs + TXQs > Max %d\n",
1865 			    nic->max_queues);
1866 		return -ENOMEM;
1867 	}
1868 
1869 	if (if_up)
1870 		nicvf_stop(nic->netdev);
1871 
1872 	old_prog = xchg(&nic->xdp_prog, prog);
1873 	/* Detach old prog, if any */
1874 	if (old_prog)
1875 		bpf_prog_put(old_prog);
1876 
1877 	if (nic->xdp_prog) {
1878 		/* Attach BPF program */
1879 		nic->xdp_prog = bpf_prog_add(nic->xdp_prog, nic->rx_queues - 1);
1880 		if (!IS_ERR(nic->xdp_prog)) {
1881 			bpf_attached = true;
1882 		} else {
1883 			ret = PTR_ERR(nic->xdp_prog);
1884 			nic->xdp_prog = NULL;
1885 		}
1886 	}
1887 
1888 	/* Calculate Tx queues needed for XDP and network stack */
1889 	nicvf_set_xdp_queues(nic, bpf_attached);
1890 
1891 	if (if_up) {
1892 		/* Reinitialize interface, clean slate */
1893 		nicvf_open(nic->netdev);
1894 		netif_trans_update(nic->netdev);
1895 	}
1896 
1897 	return ret;
1898 }
1899 
1900 static int nicvf_xdp(struct net_device *netdev, struct netdev_bpf *xdp)
1901 {
1902 	struct nicvf *nic = netdev_priv(netdev);
1903 
1904 	/* To avoid checks while retrieving buffer address from CQE_RX,
1905 	 * do not support XDP for T88 pass1.x silicons which are anyway
1906 	 * not in use widely.
1907 	 */
1908 	if (pass1_silicon(nic->pdev))
1909 		return -EOPNOTSUPP;
1910 
1911 	switch (xdp->command) {
1912 	case XDP_SETUP_PROG:
1913 		return nicvf_xdp_setup(nic, xdp->prog);
1914 	case XDP_QUERY_PROG:
1915 		xdp->prog_id = nic->xdp_prog ? nic->xdp_prog->aux->id : 0;
1916 		return 0;
1917 	default:
1918 		return -EINVAL;
1919 	}
1920 }
1921 
1922 static int nicvf_config_hwtstamp(struct net_device *netdev, struct ifreq *ifr)
1923 {
1924 	struct hwtstamp_config config;
1925 	struct nicvf *nic = netdev_priv(netdev);
1926 
1927 	if (!nic->ptp_clock)
1928 		return -ENODEV;
1929 
1930 	if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
1931 		return -EFAULT;
1932 
1933 	/* reserved for future extensions */
1934 	if (config.flags)
1935 		return -EINVAL;
1936 
1937 	switch (config.tx_type) {
1938 	case HWTSTAMP_TX_OFF:
1939 	case HWTSTAMP_TX_ON:
1940 		break;
1941 	default:
1942 		return -ERANGE;
1943 	}
1944 
1945 	switch (config.rx_filter) {
1946 	case HWTSTAMP_FILTER_NONE:
1947 		nic->hw_rx_tstamp = false;
1948 		break;
1949 	case HWTSTAMP_FILTER_ALL:
1950 	case HWTSTAMP_FILTER_SOME:
1951 	case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
1952 	case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
1953 	case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
1954 	case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
1955 	case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
1956 	case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
1957 	case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
1958 	case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
1959 	case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
1960 	case HWTSTAMP_FILTER_PTP_V2_EVENT:
1961 	case HWTSTAMP_FILTER_PTP_V2_SYNC:
1962 	case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
1963 		nic->hw_rx_tstamp = true;
1964 		config.rx_filter = HWTSTAMP_FILTER_ALL;
1965 		break;
1966 	default:
1967 		return -ERANGE;
1968 	}
1969 
1970 	if (netif_running(netdev))
1971 		nicvf_config_hw_rx_tstamp(nic, nic->hw_rx_tstamp);
1972 
1973 	if (copy_to_user(ifr->ifr_data, &config, sizeof(config)))
1974 		return -EFAULT;
1975 
1976 	return 0;
1977 }
1978 
1979 static int nicvf_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
1980 {
1981 	switch (cmd) {
1982 	case SIOCSHWTSTAMP:
1983 		return nicvf_config_hwtstamp(netdev, req);
1984 	default:
1985 		return -EOPNOTSUPP;
1986 	}
1987 }
1988 
1989 static void __nicvf_set_rx_mode_task(u8 mode, struct xcast_addr_list *mc_addrs,
1990 				     struct nicvf *nic)
1991 {
1992 	union nic_mbx mbx = {};
1993 	int idx;
1994 
1995 	/* From the inside of VM code flow we have only 128 bits memory
1996 	 * available to send message to host's PF, so send all mc addrs
1997 	 * one by one, starting from flush command in case if kernel
1998 	 * requests to configure specific MAC filtering
1999 	 */
2000 
2001 	/* flush DMAC filters and reset RX mode */
2002 	mbx.xcast.msg = NIC_MBOX_MSG_RESET_XCAST;
2003 	if (nicvf_send_msg_to_pf(nic, &mbx) < 0)
2004 		goto free_mc;
2005 
2006 	if (mode & BGX_XCAST_MCAST_FILTER) {
2007 		/* once enabling filtering, we need to signal to PF to add
2008 		 * its' own LMAC to the filter to accept packets for it.
2009 		 */
2010 		mbx.xcast.msg = NIC_MBOX_MSG_ADD_MCAST;
2011 		mbx.xcast.mac = 0;
2012 		if (nicvf_send_msg_to_pf(nic, &mbx) < 0)
2013 			goto free_mc;
2014 	}
2015 
2016 	/* check if we have any specific MACs to be added to PF DMAC filter */
2017 	if (mc_addrs) {
2018 		/* now go through kernel list of MACs and add them one by one */
2019 		for (idx = 0; idx < mc_addrs->count; idx++) {
2020 			mbx.xcast.msg = NIC_MBOX_MSG_ADD_MCAST;
2021 			mbx.xcast.mac = mc_addrs->mc[idx];
2022 			if (nicvf_send_msg_to_pf(nic, &mbx) < 0)
2023 				goto free_mc;
2024 		}
2025 	}
2026 
2027 	/* and finally set rx mode for PF accordingly */
2028 	mbx.xcast.msg = NIC_MBOX_MSG_SET_XCAST;
2029 	mbx.xcast.mode = mode;
2030 
2031 	nicvf_send_msg_to_pf(nic, &mbx);
2032 free_mc:
2033 	kfree(mc_addrs);
2034 }
2035 
2036 static void nicvf_set_rx_mode_task(struct work_struct *work_arg)
2037 {
2038 	struct nicvf_work *vf_work = container_of(work_arg, struct nicvf_work,
2039 						  work);
2040 	struct nicvf *nic = container_of(vf_work, struct nicvf, rx_mode_work);
2041 	u8 mode;
2042 	struct xcast_addr_list *mc;
2043 
2044 	if (!vf_work)
2045 		return;
2046 
2047 	/* Save message data locally to prevent them from
2048 	 * being overwritten by next ndo_set_rx_mode call().
2049 	 */
2050 	spin_lock(&nic->rx_mode_wq_lock);
2051 	mode = vf_work->mode;
2052 	mc = vf_work->mc;
2053 	vf_work->mc = NULL;
2054 	spin_unlock(&nic->rx_mode_wq_lock);
2055 
2056 	__nicvf_set_rx_mode_task(mode, mc, nic);
2057 }
2058 
2059 static void nicvf_set_rx_mode(struct net_device *netdev)
2060 {
2061 	struct nicvf *nic = netdev_priv(netdev);
2062 	struct netdev_hw_addr *ha;
2063 	struct xcast_addr_list *mc_list = NULL;
2064 	u8 mode = 0;
2065 
2066 	if (netdev->flags & IFF_PROMISC) {
2067 		mode = BGX_XCAST_BCAST_ACCEPT | BGX_XCAST_MCAST_ACCEPT;
2068 	} else {
2069 		if (netdev->flags & IFF_BROADCAST)
2070 			mode |= BGX_XCAST_BCAST_ACCEPT;
2071 
2072 		if (netdev->flags & IFF_ALLMULTI) {
2073 			mode |= BGX_XCAST_MCAST_ACCEPT;
2074 		} else if (netdev->flags & IFF_MULTICAST) {
2075 			mode |= BGX_XCAST_MCAST_FILTER;
2076 			/* here we need to copy mc addrs */
2077 			if (netdev_mc_count(netdev)) {
2078 				mc_list = kmalloc(offsetof(typeof(*mc_list),
2079 							   mc[netdev_mc_count(netdev)]),
2080 						  GFP_ATOMIC);
2081 				if (unlikely(!mc_list))
2082 					return;
2083 				mc_list->count = 0;
2084 				netdev_hw_addr_list_for_each(ha, &netdev->mc) {
2085 					mc_list->mc[mc_list->count] =
2086 						ether_addr_to_u64(ha->addr);
2087 					mc_list->count++;
2088 				}
2089 			}
2090 		}
2091 	}
2092 	spin_lock(&nic->rx_mode_wq_lock);
2093 	kfree(nic->rx_mode_work.mc);
2094 	nic->rx_mode_work.mc = mc_list;
2095 	nic->rx_mode_work.mode = mode;
2096 	queue_work(nic->nicvf_rx_mode_wq, &nic->rx_mode_work.work);
2097 	spin_unlock(&nic->rx_mode_wq_lock);
2098 }
2099 
2100 static const struct net_device_ops nicvf_netdev_ops = {
2101 	.ndo_open		= nicvf_open,
2102 	.ndo_stop		= nicvf_stop,
2103 	.ndo_start_xmit		= nicvf_xmit,
2104 	.ndo_change_mtu		= nicvf_change_mtu,
2105 	.ndo_set_mac_address	= nicvf_set_mac_address,
2106 	.ndo_get_stats64	= nicvf_get_stats64,
2107 	.ndo_tx_timeout         = nicvf_tx_timeout,
2108 	.ndo_fix_features       = nicvf_fix_features,
2109 	.ndo_set_features       = nicvf_set_features,
2110 	.ndo_bpf		= nicvf_xdp,
2111 	.ndo_do_ioctl           = nicvf_ioctl,
2112 	.ndo_set_rx_mode        = nicvf_set_rx_mode,
2113 };
2114 
2115 static int nicvf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2116 {
2117 	struct device *dev = &pdev->dev;
2118 	struct net_device *netdev;
2119 	struct nicvf *nic;
2120 	int    err, qcount;
2121 	u16    sdevid;
2122 	struct cavium_ptp *ptp_clock;
2123 
2124 	ptp_clock = cavium_ptp_get();
2125 	if (IS_ERR(ptp_clock)) {
2126 		if (PTR_ERR(ptp_clock) == -ENODEV)
2127 			/* In virtualized environment we proceed without ptp */
2128 			ptp_clock = NULL;
2129 		else
2130 			return PTR_ERR(ptp_clock);
2131 	}
2132 
2133 	err = pci_enable_device(pdev);
2134 	if (err) {
2135 		dev_err(dev, "Failed to enable PCI device\n");
2136 		return err;
2137 	}
2138 
2139 	err = pci_request_regions(pdev, DRV_NAME);
2140 	if (err) {
2141 		dev_err(dev, "PCI request regions failed 0x%x\n", err);
2142 		goto err_disable_device;
2143 	}
2144 
2145 	err = pci_set_dma_mask(pdev, DMA_BIT_MASK(48));
2146 	if (err) {
2147 		dev_err(dev, "Unable to get usable DMA configuration\n");
2148 		goto err_release_regions;
2149 	}
2150 
2151 	err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(48));
2152 	if (err) {
2153 		dev_err(dev, "unable to get 48-bit DMA for consistent allocations\n");
2154 		goto err_release_regions;
2155 	}
2156 
2157 	qcount = netif_get_num_default_rss_queues();
2158 
2159 	/* Restrict multiqset support only for host bound VFs */
2160 	if (pdev->is_virtfn) {
2161 		/* Set max number of queues per VF */
2162 		qcount = min_t(int, num_online_cpus(),
2163 			       (MAX_SQS_PER_VF + 1) * MAX_CMP_QUEUES_PER_QS);
2164 	}
2165 
2166 	netdev = alloc_etherdev_mqs(sizeof(struct nicvf), qcount, qcount);
2167 	if (!netdev) {
2168 		err = -ENOMEM;
2169 		goto err_release_regions;
2170 	}
2171 
2172 	pci_set_drvdata(pdev, netdev);
2173 
2174 	SET_NETDEV_DEV(netdev, &pdev->dev);
2175 
2176 	nic = netdev_priv(netdev);
2177 	nic->netdev = netdev;
2178 	nic->pdev = pdev;
2179 	nic->pnicvf = nic;
2180 	nic->max_queues = qcount;
2181 	/* If no of CPUs are too low, there won't be any queues left
2182 	 * for XDP_TX, hence double it.
2183 	 */
2184 	if (!nic->t88)
2185 		nic->max_queues *= 2;
2186 	nic->ptp_clock = ptp_clock;
2187 
2188 	/* MAP VF's configuration registers */
2189 	nic->reg_base = pcim_iomap(pdev, PCI_CFG_REG_BAR_NUM, 0);
2190 	if (!nic->reg_base) {
2191 		dev_err(dev, "Cannot map config register space, aborting\n");
2192 		err = -ENOMEM;
2193 		goto err_free_netdev;
2194 	}
2195 
2196 	nic->drv_stats = netdev_alloc_pcpu_stats(struct nicvf_drv_stats);
2197 	if (!nic->drv_stats) {
2198 		err = -ENOMEM;
2199 		goto err_free_netdev;
2200 	}
2201 
2202 	err = nicvf_set_qset_resources(nic);
2203 	if (err)
2204 		goto err_free_netdev;
2205 
2206 	/* Check if PF is alive and get MAC address for this VF */
2207 	err = nicvf_register_misc_interrupt(nic);
2208 	if (err)
2209 		goto err_free_netdev;
2210 
2211 	nicvf_send_vf_struct(nic);
2212 
2213 	if (!pass1_silicon(nic->pdev))
2214 		nic->hw_tso = true;
2215 
2216 	/* Get iommu domain for iova to physical addr conversion */
2217 	nic->iommu_domain = iommu_get_domain_for_dev(dev);
2218 
2219 	pci_read_config_word(nic->pdev, PCI_SUBSYSTEM_ID, &sdevid);
2220 	if (sdevid == 0xA134)
2221 		nic->t88 = true;
2222 
2223 	/* Check if this VF is in QS only mode */
2224 	if (nic->sqs_mode)
2225 		return 0;
2226 
2227 	err = nicvf_set_real_num_queues(netdev, nic->tx_queues, nic->rx_queues);
2228 	if (err)
2229 		goto err_unregister_interrupts;
2230 
2231 	netdev->hw_features = (NETIF_F_RXCSUM | NETIF_F_SG |
2232 			       NETIF_F_TSO | NETIF_F_GRO | NETIF_F_TSO6 |
2233 			       NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2234 			       NETIF_F_HW_VLAN_CTAG_RX);
2235 
2236 	netdev->hw_features |= NETIF_F_RXHASH;
2237 
2238 	netdev->features |= netdev->hw_features;
2239 	netdev->hw_features |= NETIF_F_LOOPBACK;
2240 
2241 	netdev->vlan_features = NETIF_F_SG | NETIF_F_IP_CSUM |
2242 				NETIF_F_IPV6_CSUM | NETIF_F_TSO | NETIF_F_TSO6;
2243 
2244 	netdev->netdev_ops = &nicvf_netdev_ops;
2245 	netdev->watchdog_timeo = NICVF_TX_TIMEOUT;
2246 
2247 	/* MTU range: 64 - 9200 */
2248 	netdev->min_mtu = NIC_HW_MIN_FRS;
2249 	netdev->max_mtu = NIC_HW_MAX_FRS;
2250 
2251 	INIT_WORK(&nic->reset_task, nicvf_reset_task);
2252 
2253 	nic->nicvf_rx_mode_wq = alloc_ordered_workqueue("nicvf_rx_mode_wq_VF%d",
2254 							WQ_MEM_RECLAIM,
2255 							nic->vf_id);
2256 	if (!nic->nicvf_rx_mode_wq) {
2257 		err = -ENOMEM;
2258 		dev_err(dev, "Failed to allocate work queue\n");
2259 		goto err_unregister_interrupts;
2260 	}
2261 
2262 	INIT_WORK(&nic->rx_mode_work.work, nicvf_set_rx_mode_task);
2263 	spin_lock_init(&nic->rx_mode_wq_lock);
2264 	mutex_init(&nic->rx_mode_mtx);
2265 
2266 	err = register_netdev(netdev);
2267 	if (err) {
2268 		dev_err(dev, "Failed to register netdevice\n");
2269 		goto err_unregister_interrupts;
2270 	}
2271 
2272 	nic->msg_enable = debug;
2273 
2274 	nicvf_set_ethtool_ops(netdev);
2275 
2276 	return 0;
2277 
2278 err_unregister_interrupts:
2279 	nicvf_unregister_interrupts(nic);
2280 err_free_netdev:
2281 	pci_set_drvdata(pdev, NULL);
2282 	if (nic->drv_stats)
2283 		free_percpu(nic->drv_stats);
2284 	free_netdev(netdev);
2285 err_release_regions:
2286 	pci_release_regions(pdev);
2287 err_disable_device:
2288 	pci_disable_device(pdev);
2289 	return err;
2290 }
2291 
2292 static void nicvf_remove(struct pci_dev *pdev)
2293 {
2294 	struct net_device *netdev = pci_get_drvdata(pdev);
2295 	struct nicvf *nic;
2296 	struct net_device *pnetdev;
2297 
2298 	if (!netdev)
2299 		return;
2300 
2301 	nic = netdev_priv(netdev);
2302 	pnetdev = nic->pnicvf->netdev;
2303 
2304 	/* Check if this Qset is assigned to different VF.
2305 	 * If yes, clean primary and all secondary Qsets.
2306 	 */
2307 	if (pnetdev && (pnetdev->reg_state == NETREG_REGISTERED))
2308 		unregister_netdev(pnetdev);
2309 	if (nic->nicvf_rx_mode_wq) {
2310 		destroy_workqueue(nic->nicvf_rx_mode_wq);
2311 		nic->nicvf_rx_mode_wq = NULL;
2312 	}
2313 	nicvf_unregister_interrupts(nic);
2314 	pci_set_drvdata(pdev, NULL);
2315 	if (nic->drv_stats)
2316 		free_percpu(nic->drv_stats);
2317 	cavium_ptp_put(nic->ptp_clock);
2318 	free_netdev(netdev);
2319 	pci_release_regions(pdev);
2320 	pci_disable_device(pdev);
2321 }
2322 
2323 static void nicvf_shutdown(struct pci_dev *pdev)
2324 {
2325 	nicvf_remove(pdev);
2326 }
2327 
2328 static struct pci_driver nicvf_driver = {
2329 	.name = DRV_NAME,
2330 	.id_table = nicvf_id_table,
2331 	.probe = nicvf_probe,
2332 	.remove = nicvf_remove,
2333 	.shutdown = nicvf_shutdown,
2334 };
2335 
2336 static int __init nicvf_init_module(void)
2337 {
2338 	pr_info("%s, ver %s\n", DRV_NAME, DRV_VERSION);
2339 	return pci_register_driver(&nicvf_driver);
2340 }
2341 
2342 static void __exit nicvf_cleanup_module(void)
2343 {
2344 	pci_unregister_driver(&nicvf_driver);
2345 }
2346 
2347 module_init(nicvf_init_module);
2348 module_exit(nicvf_cleanup_module);
2349