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