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