xref: /openbmc/linux/drivers/net/ethernet/cavium/thunder/nicvf_main.c (revision 9977a8c3497a8f7f7f951994f298a8e4d961234f)
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 struct nicvf_xdp_tx {
71 	u64 dma_addr;
72 	u8  qidx;
73 };
74 
75 static inline u8 nicvf_netdev_qidx(struct nicvf *nic, u8 qidx)
76 {
77 	if (nic->sqs_mode)
78 		return qidx + ((nic->sqs_id + 1) * MAX_CMP_QUEUES_PER_QS);
79 	else
80 		return qidx;
81 }
82 
83 /* The Cavium ThunderX network controller can *only* be found in SoCs
84  * containing the ThunderX ARM64 CPU implementation.  All accesses to the device
85  * registers on this platform are implicitly strongly ordered with respect
86  * to memory accesses. So writeq_relaxed() and readq_relaxed() are safe to use
87  * with no memory barriers in this driver.  The readq()/writeq() functions add
88  * explicit ordering operation which in this case are redundant, and only
89  * add overhead.
90  */
91 
92 /* Register read/write APIs */
93 void nicvf_reg_write(struct nicvf *nic, u64 offset, u64 val)
94 {
95 	writeq_relaxed(val, nic->reg_base + offset);
96 }
97 
98 u64 nicvf_reg_read(struct nicvf *nic, u64 offset)
99 {
100 	return readq_relaxed(nic->reg_base + offset);
101 }
102 
103 void nicvf_queue_reg_write(struct nicvf *nic, u64 offset,
104 			   u64 qidx, u64 val)
105 {
106 	void __iomem *addr = nic->reg_base + offset;
107 
108 	writeq_relaxed(val, addr + (qidx << NIC_Q_NUM_SHIFT));
109 }
110 
111 u64 nicvf_queue_reg_read(struct nicvf *nic, u64 offset, u64 qidx)
112 {
113 	void __iomem *addr = nic->reg_base + offset;
114 
115 	return readq_relaxed(addr + (qidx << NIC_Q_NUM_SHIFT));
116 }
117 
118 /* VF -> PF mailbox communication */
119 static void nicvf_write_to_mbx(struct nicvf *nic, union nic_mbx *mbx)
120 {
121 	u64 *msg = (u64 *)mbx;
122 
123 	nicvf_reg_write(nic, NIC_VF_PF_MAILBOX_0_1 + 0, msg[0]);
124 	nicvf_reg_write(nic, NIC_VF_PF_MAILBOX_0_1 + 8, msg[1]);
125 }
126 
127 int nicvf_send_msg_to_pf(struct nicvf *nic, union nic_mbx *mbx)
128 {
129 	int timeout = NIC_MBOX_MSG_TIMEOUT;
130 	int sleep = 10;
131 
132 	nic->pf_acked = false;
133 	nic->pf_nacked = false;
134 
135 	nicvf_write_to_mbx(nic, mbx);
136 
137 	/* Wait for previous message to be acked, timeout 2sec */
138 	while (!nic->pf_acked) {
139 		if (nic->pf_nacked) {
140 			netdev_err(nic->netdev,
141 				   "PF NACK to mbox msg 0x%02x from VF%d\n",
142 				   (mbx->msg.msg & 0xFF), nic->vf_id);
143 			return -EINVAL;
144 		}
145 		msleep(sleep);
146 		if (nic->pf_acked)
147 			break;
148 		timeout -= sleep;
149 		if (!timeout) {
150 			netdev_err(nic->netdev,
151 				   "PF didn't ACK to mbox msg 0x%02x from VF%d\n",
152 				   (mbx->msg.msg & 0xFF), nic->vf_id);
153 			return -EBUSY;
154 		}
155 	}
156 	return 0;
157 }
158 
159 /* Checks if VF is able to comminicate with PF
160 * and also gets the VNIC number this VF is associated to.
161 */
162 static int nicvf_check_pf_ready(struct nicvf *nic)
163 {
164 	union nic_mbx mbx = {};
165 
166 	mbx.msg.msg = NIC_MBOX_MSG_READY;
167 	if (nicvf_send_msg_to_pf(nic, &mbx)) {
168 		netdev_err(nic->netdev,
169 			   "PF didn't respond to READY msg\n");
170 		return 0;
171 	}
172 
173 	return 1;
174 }
175 
176 static void nicvf_read_bgx_stats(struct nicvf *nic, struct bgx_stats_msg *bgx)
177 {
178 	if (bgx->rx)
179 		nic->bgx_stats.rx_stats[bgx->idx] = bgx->stats;
180 	else
181 		nic->bgx_stats.tx_stats[bgx->idx] = bgx->stats;
182 }
183 
184 static void  nicvf_handle_mbx_intr(struct nicvf *nic)
185 {
186 	union nic_mbx mbx = {};
187 	u64 *mbx_data;
188 	u64 mbx_addr;
189 	int i;
190 
191 	mbx_addr = NIC_VF_PF_MAILBOX_0_1;
192 	mbx_data = (u64 *)&mbx;
193 
194 	for (i = 0; i < NIC_PF_VF_MAILBOX_SIZE; i++) {
195 		*mbx_data = nicvf_reg_read(nic, mbx_addr);
196 		mbx_data++;
197 		mbx_addr += sizeof(u64);
198 	}
199 
200 	netdev_dbg(nic->netdev, "Mbox message: msg: 0x%x\n", mbx.msg.msg);
201 	switch (mbx.msg.msg) {
202 	case NIC_MBOX_MSG_READY:
203 		nic->pf_acked = true;
204 		nic->vf_id = mbx.nic_cfg.vf_id & 0x7F;
205 		nic->tns_mode = mbx.nic_cfg.tns_mode & 0x7F;
206 		nic->node = mbx.nic_cfg.node_id;
207 		if (!nic->set_mac_pending)
208 			ether_addr_copy(nic->netdev->dev_addr,
209 					mbx.nic_cfg.mac_addr);
210 		nic->sqs_mode = mbx.nic_cfg.sqs_mode;
211 		nic->loopback_supported = mbx.nic_cfg.loopback_supported;
212 		nic->link_up = false;
213 		nic->duplex = 0;
214 		nic->speed = 0;
215 		break;
216 	case NIC_MBOX_MSG_ACK:
217 		nic->pf_acked = true;
218 		break;
219 	case NIC_MBOX_MSG_NACK:
220 		nic->pf_nacked = true;
221 		break;
222 	case NIC_MBOX_MSG_RSS_SIZE:
223 		nic->rss_info.rss_size = mbx.rss_size.ind_tbl_size;
224 		nic->pf_acked = true;
225 		break;
226 	case NIC_MBOX_MSG_BGX_STATS:
227 		nicvf_read_bgx_stats(nic, &mbx.bgx_stats);
228 		nic->pf_acked = true;
229 		break;
230 	case NIC_MBOX_MSG_BGX_LINK_CHANGE:
231 		nic->pf_acked = true;
232 		nic->link_up = mbx.link_status.link_up;
233 		nic->duplex = mbx.link_status.duplex;
234 		nic->speed = mbx.link_status.speed;
235 		nic->mac_type = mbx.link_status.mac_type;
236 		if (nic->link_up) {
237 			netdev_info(nic->netdev, "Link is Up %d Mbps %s duplex\n",
238 				    nic->speed,
239 				    nic->duplex == DUPLEX_FULL ?
240 				    "Full" : "Half");
241 			netif_carrier_on(nic->netdev);
242 			netif_tx_start_all_queues(nic->netdev);
243 		} else {
244 			netdev_info(nic->netdev, "Link is Down\n");
245 			netif_carrier_off(nic->netdev);
246 			netif_tx_stop_all_queues(nic->netdev);
247 		}
248 		break;
249 	case NIC_MBOX_MSG_ALLOC_SQS:
250 		nic->sqs_count = mbx.sqs_alloc.qs_count;
251 		nic->pf_acked = true;
252 		break;
253 	case NIC_MBOX_MSG_SNICVF_PTR:
254 		/* Primary VF: make note of secondary VF's pointer
255 		 * to be used while packet transmission.
256 		 */
257 		nic->snicvf[mbx.nicvf.sqs_id] =
258 			(struct nicvf *)mbx.nicvf.nicvf;
259 		nic->pf_acked = true;
260 		break;
261 	case NIC_MBOX_MSG_PNICVF_PTR:
262 		/* Secondary VF/Qset: make note of primary VF's pointer
263 		 * to be used while packet reception, to handover packet
264 		 * to primary VF's netdev.
265 		 */
266 		nic->pnicvf = (struct nicvf *)mbx.nicvf.nicvf;
267 		nic->pf_acked = true;
268 		break;
269 	case NIC_MBOX_MSG_PFC:
270 		nic->pfc.autoneg = mbx.pfc.autoneg;
271 		nic->pfc.fc_rx = mbx.pfc.fc_rx;
272 		nic->pfc.fc_tx = mbx.pfc.fc_tx;
273 		nic->pf_acked = true;
274 		break;
275 	default:
276 		netdev_err(nic->netdev,
277 			   "Invalid message from PF, msg 0x%x\n", mbx.msg.msg);
278 		break;
279 	}
280 	nicvf_clear_intr(nic, NICVF_INTR_MBOX, 0);
281 }
282 
283 static int nicvf_hw_set_mac_addr(struct nicvf *nic, struct net_device *netdev)
284 {
285 	union nic_mbx mbx = {};
286 
287 	mbx.mac.msg = NIC_MBOX_MSG_SET_MAC;
288 	mbx.mac.vf_id = nic->vf_id;
289 	ether_addr_copy(mbx.mac.mac_addr, netdev->dev_addr);
290 
291 	return nicvf_send_msg_to_pf(nic, &mbx);
292 }
293 
294 static void nicvf_config_cpi(struct nicvf *nic)
295 {
296 	union nic_mbx mbx = {};
297 
298 	mbx.cpi_cfg.msg = NIC_MBOX_MSG_CPI_CFG;
299 	mbx.cpi_cfg.vf_id = nic->vf_id;
300 	mbx.cpi_cfg.cpi_alg = nic->cpi_alg;
301 	mbx.cpi_cfg.rq_cnt = nic->qs->rq_cnt;
302 
303 	nicvf_send_msg_to_pf(nic, &mbx);
304 }
305 
306 static void nicvf_get_rss_size(struct nicvf *nic)
307 {
308 	union nic_mbx mbx = {};
309 
310 	mbx.rss_size.msg = NIC_MBOX_MSG_RSS_SIZE;
311 	mbx.rss_size.vf_id = nic->vf_id;
312 	nicvf_send_msg_to_pf(nic, &mbx);
313 }
314 
315 void nicvf_config_rss(struct nicvf *nic)
316 {
317 	union nic_mbx mbx = {};
318 	struct nicvf_rss_info *rss = &nic->rss_info;
319 	int ind_tbl_len = rss->rss_size;
320 	int i, nextq = 0;
321 
322 	mbx.rss_cfg.vf_id = nic->vf_id;
323 	mbx.rss_cfg.hash_bits = rss->hash_bits;
324 	while (ind_tbl_len) {
325 		mbx.rss_cfg.tbl_offset = nextq;
326 		mbx.rss_cfg.tbl_len = min(ind_tbl_len,
327 					       RSS_IND_TBL_LEN_PER_MBX_MSG);
328 		mbx.rss_cfg.msg = mbx.rss_cfg.tbl_offset ?
329 			  NIC_MBOX_MSG_RSS_CFG_CONT : NIC_MBOX_MSG_RSS_CFG;
330 
331 		for (i = 0; i < mbx.rss_cfg.tbl_len; i++)
332 			mbx.rss_cfg.ind_tbl[i] = rss->ind_tbl[nextq++];
333 
334 		nicvf_send_msg_to_pf(nic, &mbx);
335 
336 		ind_tbl_len -= mbx.rss_cfg.tbl_len;
337 	}
338 }
339 
340 void nicvf_set_rss_key(struct nicvf *nic)
341 {
342 	struct nicvf_rss_info *rss = &nic->rss_info;
343 	u64 key_addr = NIC_VNIC_RSS_KEY_0_4;
344 	int idx;
345 
346 	for (idx = 0; idx < RSS_HASH_KEY_SIZE; idx++) {
347 		nicvf_reg_write(nic, key_addr, rss->key[idx]);
348 		key_addr += sizeof(u64);
349 	}
350 }
351 
352 static int nicvf_rss_init(struct nicvf *nic)
353 {
354 	struct nicvf_rss_info *rss = &nic->rss_info;
355 	int idx;
356 
357 	nicvf_get_rss_size(nic);
358 
359 	if (cpi_alg != CPI_ALG_NONE) {
360 		rss->enable = false;
361 		rss->hash_bits = 0;
362 		return 0;
363 	}
364 
365 	rss->enable = true;
366 
367 	netdev_rss_key_fill(rss->key, RSS_HASH_KEY_SIZE * sizeof(u64));
368 	nicvf_set_rss_key(nic);
369 
370 	rss->cfg = RSS_IP_HASH_ENA | RSS_TCP_HASH_ENA | RSS_UDP_HASH_ENA;
371 	nicvf_reg_write(nic, NIC_VNIC_RSS_CFG, rss->cfg);
372 
373 	rss->hash_bits =  ilog2(rounddown_pow_of_two(rss->rss_size));
374 
375 	for (idx = 0; idx < rss->rss_size; idx++)
376 		rss->ind_tbl[idx] = ethtool_rxfh_indir_default(idx,
377 							       nic->rx_queues);
378 	nicvf_config_rss(nic);
379 	return 1;
380 }
381 
382 /* Request PF to allocate additional Qsets */
383 static void nicvf_request_sqs(struct nicvf *nic)
384 {
385 	union nic_mbx mbx = {};
386 	int sqs;
387 	int sqs_count = nic->sqs_count;
388 	int rx_queues = 0, tx_queues = 0;
389 
390 	/* Only primary VF should request */
391 	if (nic->sqs_mode ||  !nic->sqs_count)
392 		return;
393 
394 	mbx.sqs_alloc.msg = NIC_MBOX_MSG_ALLOC_SQS;
395 	mbx.sqs_alloc.vf_id = nic->vf_id;
396 	mbx.sqs_alloc.qs_count = nic->sqs_count;
397 	if (nicvf_send_msg_to_pf(nic, &mbx)) {
398 		/* No response from PF */
399 		nic->sqs_count = 0;
400 		return;
401 	}
402 
403 	/* Return if no Secondary Qsets available */
404 	if (!nic->sqs_count)
405 		return;
406 
407 	if (nic->rx_queues > MAX_RCV_QUEUES_PER_QS)
408 		rx_queues = nic->rx_queues - MAX_RCV_QUEUES_PER_QS;
409 
410 	tx_queues = nic->tx_queues + nic->xdp_tx_queues;
411 	if (tx_queues > MAX_SND_QUEUES_PER_QS)
412 		tx_queues = tx_queues - MAX_SND_QUEUES_PER_QS;
413 
414 	/* Set no of Rx/Tx queues in each of the SQsets */
415 	for (sqs = 0; sqs < nic->sqs_count; sqs++) {
416 		mbx.nicvf.msg = NIC_MBOX_MSG_SNICVF_PTR;
417 		mbx.nicvf.vf_id = nic->vf_id;
418 		mbx.nicvf.sqs_id = sqs;
419 		nicvf_send_msg_to_pf(nic, &mbx);
420 
421 		nic->snicvf[sqs]->sqs_id = sqs;
422 		if (rx_queues > MAX_RCV_QUEUES_PER_QS) {
423 			nic->snicvf[sqs]->qs->rq_cnt = MAX_RCV_QUEUES_PER_QS;
424 			rx_queues -= MAX_RCV_QUEUES_PER_QS;
425 		} else {
426 			nic->snicvf[sqs]->qs->rq_cnt = rx_queues;
427 			rx_queues = 0;
428 		}
429 
430 		if (tx_queues > MAX_SND_QUEUES_PER_QS) {
431 			nic->snicvf[sqs]->qs->sq_cnt = MAX_SND_QUEUES_PER_QS;
432 			tx_queues -= MAX_SND_QUEUES_PER_QS;
433 		} else {
434 			nic->snicvf[sqs]->qs->sq_cnt = tx_queues;
435 			tx_queues = 0;
436 		}
437 
438 		nic->snicvf[sqs]->qs->cq_cnt =
439 		max(nic->snicvf[sqs]->qs->rq_cnt, nic->snicvf[sqs]->qs->sq_cnt);
440 
441 		/* Initialize secondary Qset's queues and its interrupts */
442 		nicvf_open(nic->snicvf[sqs]->netdev);
443 	}
444 
445 	/* Update stack with actual Rx/Tx queue count allocated */
446 	if (sqs_count != nic->sqs_count)
447 		nicvf_set_real_num_queues(nic->netdev,
448 					  nic->tx_queues, nic->rx_queues);
449 }
450 
451 /* Send this Qset's nicvf pointer to PF.
452  * PF inturn sends primary VF's nicvf struct to secondary Qsets/VFs
453  * so that packets received by these Qsets can use primary VF's netdev
454  */
455 static void nicvf_send_vf_struct(struct nicvf *nic)
456 {
457 	union nic_mbx mbx = {};
458 
459 	mbx.nicvf.msg = NIC_MBOX_MSG_NICVF_PTR;
460 	mbx.nicvf.sqs_mode = nic->sqs_mode;
461 	mbx.nicvf.nicvf = (u64)nic;
462 	nicvf_send_msg_to_pf(nic, &mbx);
463 }
464 
465 static void nicvf_get_primary_vf_struct(struct nicvf *nic)
466 {
467 	union nic_mbx mbx = {};
468 
469 	mbx.nicvf.msg = NIC_MBOX_MSG_PNICVF_PTR;
470 	nicvf_send_msg_to_pf(nic, &mbx);
471 }
472 
473 int nicvf_set_real_num_queues(struct net_device *netdev,
474 			      int tx_queues, int rx_queues)
475 {
476 	int err = 0;
477 
478 	err = netif_set_real_num_tx_queues(netdev, tx_queues);
479 	if (err) {
480 		netdev_err(netdev,
481 			   "Failed to set no of Tx queues: %d\n", tx_queues);
482 		return err;
483 	}
484 
485 	err = netif_set_real_num_rx_queues(netdev, rx_queues);
486 	if (err)
487 		netdev_err(netdev,
488 			   "Failed to set no of Rx queues: %d\n", rx_queues);
489 	return err;
490 }
491 
492 static int nicvf_init_resources(struct nicvf *nic)
493 {
494 	int err;
495 
496 	/* Enable Qset */
497 	nicvf_qset_config(nic, true);
498 
499 	/* Initialize queues and HW for data transfer */
500 	err = nicvf_config_data_transfer(nic, true);
501 	if (err) {
502 		netdev_err(nic->netdev,
503 			   "Failed to alloc/config VF's QSet resources\n");
504 		return err;
505 	}
506 
507 	return 0;
508 }
509 
510 static void nicvf_unmap_page(struct nicvf *nic, struct page *page, u64 dma_addr)
511 {
512 	/* Check if it's a recycled page, if not unmap the DMA mapping.
513 	 * Recycled page holds an extra reference.
514 	 */
515 	if (page_ref_count(page) == 1) {
516 		dma_addr &= PAGE_MASK;
517 		dma_unmap_page_attrs(&nic->pdev->dev, dma_addr,
518 				     RCV_FRAG_LEN + XDP_HEADROOM,
519 				     DMA_FROM_DEVICE,
520 				     DMA_ATTR_SKIP_CPU_SYNC);
521 	}
522 }
523 
524 static inline bool nicvf_xdp_rx(struct nicvf *nic, struct bpf_prog *prog,
525 				struct cqe_rx_t *cqe_rx, struct snd_queue *sq,
526 				struct rcv_queue *rq, struct sk_buff **skb)
527 {
528 	struct xdp_buff xdp;
529 	struct page *page;
530 	struct nicvf_xdp_tx *xdp_tx = NULL;
531 	u32 action;
532 	u16 len, err, offset = 0;
533 	u64 dma_addr, cpu_addr;
534 	void *orig_data;
535 
536 	/* Retrieve packet buffer's DMA address and length */
537 	len = *((u16 *)((void *)cqe_rx + (3 * sizeof(u64))));
538 	dma_addr = *((u64 *)((void *)cqe_rx + (7 * sizeof(u64))));
539 
540 	cpu_addr = nicvf_iova_to_phys(nic, dma_addr);
541 	if (!cpu_addr)
542 		return false;
543 	cpu_addr = (u64)phys_to_virt(cpu_addr);
544 	page = virt_to_page((void *)cpu_addr);
545 
546 	xdp.data_hard_start = page_address(page) + RCV_BUF_HEADROOM;
547 	xdp.data = (void *)cpu_addr;
548 	xdp_set_data_meta_invalid(&xdp);
549 	xdp.data_end = xdp.data + len;
550 	xdp.rxq = &rq->xdp_rxq;
551 	orig_data = xdp.data;
552 
553 	rcu_read_lock();
554 	action = bpf_prog_run_xdp(prog, &xdp);
555 	rcu_read_unlock();
556 
557 	/* Check if XDP program has changed headers */
558 	if (orig_data != xdp.data) {
559 		len = xdp.data_end - xdp.data;
560 		offset = orig_data - xdp.data;
561 		dma_addr -= offset;
562 	}
563 
564 	switch (action) {
565 	case XDP_PASS:
566 		nicvf_unmap_page(nic, page, dma_addr);
567 
568 		/* Build SKB and pass on packet to network stack */
569 		*skb = build_skb(xdp.data,
570 				 RCV_FRAG_LEN - cqe_rx->align_pad + offset);
571 		if (!*skb)
572 			put_page(page);
573 		else
574 			skb_put(*skb, len);
575 		return false;
576 	case XDP_TX:
577 		nicvf_xdp_sq_append_pkt(nic, sq, (u64)xdp.data, dma_addr, len);
578 		return true;
579 	case XDP_REDIRECT:
580 		/* Save DMA address for use while transmitting */
581 		xdp_tx = (struct nicvf_xdp_tx *)page_address(page);
582 		xdp_tx->dma_addr = dma_addr;
583 		xdp_tx->qidx = nicvf_netdev_qidx(nic, cqe_rx->rq_idx);
584 
585 		err = xdp_do_redirect(nic->pnicvf->netdev, &xdp, prog);
586 		if (!err)
587 			return true;
588 
589 		/* Free the page on error */
590 		nicvf_unmap_page(nic, page, dma_addr);
591 		put_page(page);
592 		break;
593 	default:
594 		bpf_warn_invalid_xdp_action(action);
595 		/* fall through */
596 	case XDP_ABORTED:
597 		trace_xdp_exception(nic->netdev, prog, action);
598 		/* fall through */
599 	case XDP_DROP:
600 		nicvf_unmap_page(nic, page, dma_addr);
601 		put_page(page);
602 		return true;
603 	}
604 	return false;
605 }
606 
607 static void nicvf_snd_ptp_handler(struct net_device *netdev,
608 				  struct cqe_send_t *cqe_tx)
609 {
610 	struct nicvf *nic = netdev_priv(netdev);
611 	struct skb_shared_hwtstamps ts;
612 	u64 ns;
613 
614 	nic = nic->pnicvf;
615 
616 	/* Sync for 'ptp_skb' */
617 	smp_rmb();
618 
619 	/* New timestamp request can be queued now */
620 	atomic_set(&nic->tx_ptp_skbs, 0);
621 
622 	/* Check for timestamp requested skb */
623 	if (!nic->ptp_skb)
624 		return;
625 
626 	/* Check if timestamping is timedout, which is set to 10us */
627 	if (cqe_tx->send_status == CQ_TX_ERROP_TSTMP_TIMEOUT ||
628 	    cqe_tx->send_status == CQ_TX_ERROP_TSTMP_CONFLICT)
629 		goto no_tstamp;
630 
631 	/* Get the timestamp */
632 	memset(&ts, 0, sizeof(ts));
633 	ns = cavium_ptp_tstamp2time(nic->ptp_clock, cqe_tx->ptp_timestamp);
634 	ts.hwtstamp = ns_to_ktime(ns);
635 	skb_tstamp_tx(nic->ptp_skb, &ts);
636 
637 no_tstamp:
638 	/* Free the original skb */
639 	dev_kfree_skb_any(nic->ptp_skb);
640 	nic->ptp_skb = NULL;
641 	/* Sync 'ptp_skb' */
642 	smp_wmb();
643 }
644 
645 static void nicvf_snd_pkt_handler(struct net_device *netdev,
646 				  struct cqe_send_t *cqe_tx,
647 				  int budget, int *subdesc_cnt,
648 				  unsigned int *tx_pkts, unsigned int *tx_bytes)
649 {
650 	struct sk_buff *skb = NULL;
651 	struct page *page;
652 	struct nicvf *nic = netdev_priv(netdev);
653 	struct snd_queue *sq;
654 	struct sq_hdr_subdesc *hdr;
655 	struct sq_hdr_subdesc *tso_sqe;
656 
657 	sq = &nic->qs->sq[cqe_tx->sq_idx];
658 
659 	hdr = (struct sq_hdr_subdesc *)GET_SQ_DESC(sq, cqe_tx->sqe_ptr);
660 	if (hdr->subdesc_type != SQ_DESC_TYPE_HEADER)
661 		return;
662 
663 	/* Check for errors */
664 	if (cqe_tx->send_status)
665 		nicvf_check_cqe_tx_errs(nic->pnicvf, cqe_tx);
666 
667 	/* Is this a XDP designated Tx queue */
668 	if (sq->is_xdp) {
669 		page = (struct page *)sq->xdp_page[cqe_tx->sqe_ptr];
670 		/* Check if it's recycled page or else unmap DMA mapping */
671 		if (page && (page_ref_count(page) == 1))
672 			nicvf_unmap_sndq_buffers(nic, sq, cqe_tx->sqe_ptr,
673 						 hdr->subdesc_cnt);
674 
675 		/* Release page reference for recycling */
676 		if (page)
677 			put_page(page);
678 		sq->xdp_page[cqe_tx->sqe_ptr] = (u64)NULL;
679 		*subdesc_cnt += hdr->subdesc_cnt + 1;
680 		return;
681 	}
682 
683 	skb = (struct sk_buff *)sq->skbuff[cqe_tx->sqe_ptr];
684 	if (skb) {
685 		/* Check for dummy descriptor used for HW TSO offload on 88xx */
686 		if (hdr->dont_send) {
687 			/* Get actual TSO descriptors and free them */
688 			tso_sqe =
689 			 (struct sq_hdr_subdesc *)GET_SQ_DESC(sq, hdr->rsvd2);
690 			nicvf_unmap_sndq_buffers(nic, sq, hdr->rsvd2,
691 						 tso_sqe->subdesc_cnt);
692 			*subdesc_cnt += tso_sqe->subdesc_cnt + 1;
693 		} else {
694 			nicvf_unmap_sndq_buffers(nic, sq, cqe_tx->sqe_ptr,
695 						 hdr->subdesc_cnt);
696 		}
697 		*subdesc_cnt += hdr->subdesc_cnt + 1;
698 		prefetch(skb);
699 		(*tx_pkts)++;
700 		*tx_bytes += skb->len;
701 		/* If timestamp is requested for this skb, don't free it */
702 		if (skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS &&
703 		    !nic->pnicvf->ptp_skb)
704 			nic->pnicvf->ptp_skb = skb;
705 		else
706 			napi_consume_skb(skb, budget);
707 		sq->skbuff[cqe_tx->sqe_ptr] = (u64)NULL;
708 	} else {
709 		/* In case of SW TSO on 88xx, only last segment will have
710 		 * a SKB attached, so just free SQEs here.
711 		 */
712 		if (!nic->hw_tso)
713 			*subdesc_cnt += hdr->subdesc_cnt + 1;
714 	}
715 }
716 
717 static inline void nicvf_set_rxhash(struct net_device *netdev,
718 				    struct cqe_rx_t *cqe_rx,
719 				    struct sk_buff *skb)
720 {
721 	u8 hash_type;
722 	u32 hash;
723 
724 	if (!(netdev->features & NETIF_F_RXHASH))
725 		return;
726 
727 	switch (cqe_rx->rss_alg) {
728 	case RSS_ALG_TCP_IP:
729 	case RSS_ALG_UDP_IP:
730 		hash_type = PKT_HASH_TYPE_L4;
731 		hash = cqe_rx->rss_tag;
732 		break;
733 	case RSS_ALG_IP:
734 		hash_type = PKT_HASH_TYPE_L3;
735 		hash = cqe_rx->rss_tag;
736 		break;
737 	default:
738 		hash_type = PKT_HASH_TYPE_NONE;
739 		hash = 0;
740 	}
741 
742 	skb_set_hash(skb, hash, hash_type);
743 }
744 
745 static inline void nicvf_set_rxtstamp(struct nicvf *nic, struct sk_buff *skb)
746 {
747 	u64 ns;
748 
749 	if (!nic->ptp_clock || !nic->hw_rx_tstamp)
750 		return;
751 
752 	/* The first 8 bytes is the timestamp */
753 	ns = cavium_ptp_tstamp2time(nic->ptp_clock,
754 				    be64_to_cpu(*(__be64 *)skb->data));
755 	skb_hwtstamps(skb)->hwtstamp = ns_to_ktime(ns);
756 
757 	__skb_pull(skb, 8);
758 }
759 
760 static void nicvf_rcv_pkt_handler(struct net_device *netdev,
761 				  struct napi_struct *napi,
762 				  struct cqe_rx_t *cqe_rx,
763 				  struct snd_queue *sq, struct rcv_queue *rq)
764 {
765 	struct sk_buff *skb = NULL;
766 	struct nicvf *nic = netdev_priv(netdev);
767 	struct nicvf *snic = nic;
768 	int err = 0;
769 	int rq_idx;
770 
771 	rq_idx = nicvf_netdev_qidx(nic, cqe_rx->rq_idx);
772 
773 	if (nic->sqs_mode) {
774 		/* Use primary VF's 'nicvf' struct */
775 		nic = nic->pnicvf;
776 		netdev = nic->netdev;
777 	}
778 
779 	/* Check for errors */
780 	if (cqe_rx->err_level || cqe_rx->err_opcode) {
781 		err = nicvf_check_cqe_rx_errs(nic, cqe_rx);
782 		if (err && !cqe_rx->rb_cnt)
783 			return;
784 	}
785 
786 	/* For XDP, ignore pkts spanning multiple pages */
787 	if (nic->xdp_prog && (cqe_rx->rb_cnt == 1)) {
788 		/* Packet consumed by XDP */
789 		if (nicvf_xdp_rx(snic, nic->xdp_prog, cqe_rx, sq, rq, &skb))
790 			return;
791 	} else {
792 		skb = nicvf_get_rcv_skb(snic, cqe_rx,
793 					nic->xdp_prog ? true : false);
794 	}
795 
796 	if (!skb)
797 		return;
798 
799 	if (netif_msg_pktdata(nic)) {
800 		netdev_info(nic->netdev, "skb 0x%p, len=%d\n", skb, skb->len);
801 		print_hex_dump(KERN_INFO, "", DUMP_PREFIX_OFFSET, 16, 1,
802 			       skb->data, skb->len, true);
803 	}
804 
805 	/* If error packet, drop it here */
806 	if (err) {
807 		dev_kfree_skb_any(skb);
808 		return;
809 	}
810 
811 	nicvf_set_rxtstamp(nic, skb);
812 	nicvf_set_rxhash(netdev, cqe_rx, skb);
813 
814 	skb_record_rx_queue(skb, rq_idx);
815 	if (netdev->hw_features & NETIF_F_RXCSUM) {
816 		/* HW by default verifies TCP/UDP/SCTP checksums */
817 		skb->ip_summed = CHECKSUM_UNNECESSARY;
818 	} else {
819 		skb_checksum_none_assert(skb);
820 	}
821 
822 	skb->protocol = eth_type_trans(skb, netdev);
823 
824 	/* Check for stripped VLAN */
825 	if (cqe_rx->vlan_found && cqe_rx->vlan_stripped)
826 		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
827 				       ntohs((__force __be16)cqe_rx->vlan_tci));
828 
829 	if (napi && (netdev->features & NETIF_F_GRO))
830 		napi_gro_receive(napi, skb);
831 	else
832 		netif_receive_skb(skb);
833 }
834 
835 static int nicvf_cq_intr_handler(struct net_device *netdev, u8 cq_idx,
836 				 struct napi_struct *napi, int budget)
837 {
838 	int processed_cqe, work_done = 0, tx_done = 0;
839 	int cqe_count, cqe_head;
840 	int subdesc_cnt = 0;
841 	struct nicvf *nic = netdev_priv(netdev);
842 	struct queue_set *qs = nic->qs;
843 	struct cmp_queue *cq = &qs->cq[cq_idx];
844 	struct cqe_rx_t *cq_desc;
845 	struct netdev_queue *txq;
846 	struct snd_queue *sq = &qs->sq[cq_idx];
847 	struct rcv_queue *rq = &qs->rq[cq_idx];
848 	unsigned int tx_pkts = 0, tx_bytes = 0, txq_idx;
849 
850 	spin_lock_bh(&cq->lock);
851 loop:
852 	processed_cqe = 0;
853 	/* Get no of valid CQ entries to process */
854 	cqe_count = nicvf_queue_reg_read(nic, NIC_QSET_CQ_0_7_STATUS, cq_idx);
855 	cqe_count &= CQ_CQE_COUNT;
856 	if (!cqe_count)
857 		goto done;
858 
859 	/* Get head of the valid CQ entries */
860 	cqe_head = nicvf_queue_reg_read(nic, NIC_QSET_CQ_0_7_HEAD, cq_idx) >> 9;
861 	cqe_head &= 0xFFFF;
862 
863 	while (processed_cqe < cqe_count) {
864 		/* Get the CQ descriptor */
865 		cq_desc = (struct cqe_rx_t *)GET_CQ_DESC(cq, cqe_head);
866 		cqe_head++;
867 		cqe_head &= (cq->dmem.q_len - 1);
868 		/* Initiate prefetch for next descriptor */
869 		prefetch((struct cqe_rx_t *)GET_CQ_DESC(cq, cqe_head));
870 
871 		if ((work_done >= budget) && napi &&
872 		    (cq_desc->cqe_type != CQE_TYPE_SEND)) {
873 			break;
874 		}
875 
876 		switch (cq_desc->cqe_type) {
877 		case CQE_TYPE_RX:
878 			nicvf_rcv_pkt_handler(netdev, napi, cq_desc, sq, rq);
879 			work_done++;
880 		break;
881 		case CQE_TYPE_SEND:
882 			nicvf_snd_pkt_handler(netdev, (void *)cq_desc,
883 					      budget, &subdesc_cnt,
884 					      &tx_pkts, &tx_bytes);
885 			tx_done++;
886 		break;
887 		case CQE_TYPE_SEND_PTP:
888 			nicvf_snd_ptp_handler(netdev, (void *)cq_desc);
889 		break;
890 		case CQE_TYPE_INVALID:
891 		case CQE_TYPE_RX_SPLIT:
892 		case CQE_TYPE_RX_TCP:
893 			/* Ignore for now */
894 		break;
895 		}
896 		processed_cqe++;
897 	}
898 
899 	/* Ring doorbell to inform H/W to reuse processed CQEs */
900 	nicvf_queue_reg_write(nic, NIC_QSET_CQ_0_7_DOOR,
901 			      cq_idx, processed_cqe);
902 
903 	if ((work_done < budget) && napi)
904 		goto loop;
905 
906 done:
907 	/* Update SQ's descriptor free count */
908 	if (subdesc_cnt)
909 		nicvf_put_sq_desc(sq, subdesc_cnt);
910 
911 	txq_idx = nicvf_netdev_qidx(nic, cq_idx);
912 	/* Handle XDP TX queues */
913 	if (nic->pnicvf->xdp_prog) {
914 		if (txq_idx < nic->pnicvf->xdp_tx_queues) {
915 			nicvf_xdp_sq_doorbell(nic, sq, cq_idx);
916 			goto out;
917 		}
918 		nic = nic->pnicvf;
919 		txq_idx -= nic->pnicvf->xdp_tx_queues;
920 	}
921 
922 	/* Wakeup TXQ if its stopped earlier due to SQ full */
923 	if (tx_done ||
924 	    (atomic_read(&sq->free_cnt) >= MIN_SQ_DESC_PER_PKT_XMIT)) {
925 		netdev = nic->pnicvf->netdev;
926 		txq = netdev_get_tx_queue(netdev, txq_idx);
927 		if (tx_pkts)
928 			netdev_tx_completed_queue(txq, tx_pkts, tx_bytes);
929 
930 		/* To read updated queue and carrier status */
931 		smp_mb();
932 		if (netif_tx_queue_stopped(txq) && netif_carrier_ok(netdev)) {
933 			netif_tx_wake_queue(txq);
934 			nic = nic->pnicvf;
935 			this_cpu_inc(nic->drv_stats->txq_wake);
936 			netif_warn(nic, tx_err, netdev,
937 				   "Transmit queue wakeup SQ%d\n", txq_idx);
938 		}
939 	}
940 
941 out:
942 	spin_unlock_bh(&cq->lock);
943 	return work_done;
944 }
945 
946 static int nicvf_poll(struct napi_struct *napi, int budget)
947 {
948 	u64  cq_head;
949 	int  work_done = 0;
950 	struct net_device *netdev = napi->dev;
951 	struct nicvf *nic = netdev_priv(netdev);
952 	struct nicvf_cq_poll *cq;
953 
954 	cq = container_of(napi, struct nicvf_cq_poll, napi);
955 	work_done = nicvf_cq_intr_handler(netdev, cq->cq_idx, napi, budget);
956 
957 	if (work_done < budget) {
958 		/* Slow packet rate, exit polling */
959 		napi_complete_done(napi, work_done);
960 		/* Re-enable interrupts */
961 		cq_head = nicvf_queue_reg_read(nic, NIC_QSET_CQ_0_7_HEAD,
962 					       cq->cq_idx);
963 		nicvf_clear_intr(nic, NICVF_INTR_CQ, cq->cq_idx);
964 		nicvf_queue_reg_write(nic, NIC_QSET_CQ_0_7_HEAD,
965 				      cq->cq_idx, cq_head);
966 		nicvf_enable_intr(nic, NICVF_INTR_CQ, cq->cq_idx);
967 	}
968 	return work_done;
969 }
970 
971 /* Qset error interrupt handler
972  *
973  * As of now only CQ errors are handled
974  */
975 static void nicvf_handle_qs_err(unsigned long data)
976 {
977 	struct nicvf *nic = (struct nicvf *)data;
978 	struct queue_set *qs = nic->qs;
979 	int qidx;
980 	u64 status;
981 
982 	netif_tx_disable(nic->netdev);
983 
984 	/* Check if it is CQ err */
985 	for (qidx = 0; qidx < qs->cq_cnt; qidx++) {
986 		status = nicvf_queue_reg_read(nic, NIC_QSET_CQ_0_7_STATUS,
987 					      qidx);
988 		if (!(status & CQ_ERR_MASK))
989 			continue;
990 		/* Process already queued CQEs and reconfig CQ */
991 		nicvf_disable_intr(nic, NICVF_INTR_CQ, qidx);
992 		nicvf_sq_disable(nic, qidx);
993 		nicvf_cq_intr_handler(nic->netdev, qidx, NULL, 0);
994 		nicvf_cmp_queue_config(nic, qs, qidx, true);
995 		nicvf_sq_free_used_descs(nic->netdev, &qs->sq[qidx], qidx);
996 		nicvf_sq_enable(nic, &qs->sq[qidx], qidx);
997 
998 		nicvf_enable_intr(nic, NICVF_INTR_CQ, qidx);
999 	}
1000 
1001 	netif_tx_start_all_queues(nic->netdev);
1002 	/* Re-enable Qset error interrupt */
1003 	nicvf_enable_intr(nic, NICVF_INTR_QS_ERR, 0);
1004 }
1005 
1006 static void nicvf_dump_intr_status(struct nicvf *nic)
1007 {
1008 	netif_info(nic, intr, nic->netdev, "interrupt status 0x%llx\n",
1009 		   nicvf_reg_read(nic, NIC_VF_INT));
1010 }
1011 
1012 static irqreturn_t nicvf_misc_intr_handler(int irq, void *nicvf_irq)
1013 {
1014 	struct nicvf *nic = (struct nicvf *)nicvf_irq;
1015 	u64 intr;
1016 
1017 	nicvf_dump_intr_status(nic);
1018 
1019 	intr = nicvf_reg_read(nic, NIC_VF_INT);
1020 	/* Check for spurious interrupt */
1021 	if (!(intr & NICVF_INTR_MBOX_MASK))
1022 		return IRQ_HANDLED;
1023 
1024 	nicvf_handle_mbx_intr(nic);
1025 
1026 	return IRQ_HANDLED;
1027 }
1028 
1029 static irqreturn_t nicvf_intr_handler(int irq, void *cq_irq)
1030 {
1031 	struct nicvf_cq_poll *cq_poll = (struct nicvf_cq_poll *)cq_irq;
1032 	struct nicvf *nic = cq_poll->nicvf;
1033 	int qidx = cq_poll->cq_idx;
1034 
1035 	nicvf_dump_intr_status(nic);
1036 
1037 	/* Disable interrupts */
1038 	nicvf_disable_intr(nic, NICVF_INTR_CQ, qidx);
1039 
1040 	/* Schedule NAPI */
1041 	napi_schedule_irqoff(&cq_poll->napi);
1042 
1043 	/* Clear interrupt */
1044 	nicvf_clear_intr(nic, NICVF_INTR_CQ, qidx);
1045 
1046 	return IRQ_HANDLED;
1047 }
1048 
1049 static irqreturn_t nicvf_rbdr_intr_handler(int irq, void *nicvf_irq)
1050 {
1051 	struct nicvf *nic = (struct nicvf *)nicvf_irq;
1052 	u8 qidx;
1053 
1054 
1055 	nicvf_dump_intr_status(nic);
1056 
1057 	/* Disable RBDR interrupt and schedule softirq */
1058 	for (qidx = 0; qidx < nic->qs->rbdr_cnt; qidx++) {
1059 		if (!nicvf_is_intr_enabled(nic, NICVF_INTR_RBDR, qidx))
1060 			continue;
1061 		nicvf_disable_intr(nic, NICVF_INTR_RBDR, qidx);
1062 		tasklet_hi_schedule(&nic->rbdr_task);
1063 		/* Clear interrupt */
1064 		nicvf_clear_intr(nic, NICVF_INTR_RBDR, qidx);
1065 	}
1066 
1067 	return IRQ_HANDLED;
1068 }
1069 
1070 static irqreturn_t nicvf_qs_err_intr_handler(int irq, void *nicvf_irq)
1071 {
1072 	struct nicvf *nic = (struct nicvf *)nicvf_irq;
1073 
1074 	nicvf_dump_intr_status(nic);
1075 
1076 	/* Disable Qset err interrupt and schedule softirq */
1077 	nicvf_disable_intr(nic, NICVF_INTR_QS_ERR, 0);
1078 	tasklet_hi_schedule(&nic->qs_err_task);
1079 	nicvf_clear_intr(nic, NICVF_INTR_QS_ERR, 0);
1080 
1081 	return IRQ_HANDLED;
1082 }
1083 
1084 static void nicvf_set_irq_affinity(struct nicvf *nic)
1085 {
1086 	int vec, cpu;
1087 
1088 	for (vec = 0; vec < nic->num_vec; vec++) {
1089 		if (!nic->irq_allocated[vec])
1090 			continue;
1091 
1092 		if (!zalloc_cpumask_var(&nic->affinity_mask[vec], GFP_KERNEL))
1093 			return;
1094 		 /* CQ interrupts */
1095 		if (vec < NICVF_INTR_ID_SQ)
1096 			/* Leave CPU0 for RBDR and other interrupts */
1097 			cpu = nicvf_netdev_qidx(nic, vec) + 1;
1098 		else
1099 			cpu = 0;
1100 
1101 		cpumask_set_cpu(cpumask_local_spread(cpu, nic->node),
1102 				nic->affinity_mask[vec]);
1103 		irq_set_affinity_hint(pci_irq_vector(nic->pdev, vec),
1104 				      nic->affinity_mask[vec]);
1105 	}
1106 }
1107 
1108 static int nicvf_register_interrupts(struct nicvf *nic)
1109 {
1110 	int irq, ret = 0;
1111 
1112 	for_each_cq_irq(irq)
1113 		sprintf(nic->irq_name[irq], "%s-rxtx-%d",
1114 			nic->pnicvf->netdev->name,
1115 			nicvf_netdev_qidx(nic, irq));
1116 
1117 	for_each_sq_irq(irq)
1118 		sprintf(nic->irq_name[irq], "%s-sq-%d",
1119 			nic->pnicvf->netdev->name,
1120 			nicvf_netdev_qidx(nic, irq - NICVF_INTR_ID_SQ));
1121 
1122 	for_each_rbdr_irq(irq)
1123 		sprintf(nic->irq_name[irq], "%s-rbdr-%d",
1124 			nic->pnicvf->netdev->name,
1125 			nic->sqs_mode ? (nic->sqs_id + 1) : 0);
1126 
1127 	/* Register CQ interrupts */
1128 	for (irq = 0; irq < nic->qs->cq_cnt; irq++) {
1129 		ret = request_irq(pci_irq_vector(nic->pdev, irq),
1130 				  nicvf_intr_handler,
1131 				  0, nic->irq_name[irq], nic->napi[irq]);
1132 		if (ret)
1133 			goto err;
1134 		nic->irq_allocated[irq] = true;
1135 	}
1136 
1137 	/* Register RBDR interrupt */
1138 	for (irq = NICVF_INTR_ID_RBDR;
1139 	     irq < (NICVF_INTR_ID_RBDR + nic->qs->rbdr_cnt); irq++) {
1140 		ret = request_irq(pci_irq_vector(nic->pdev, irq),
1141 				  nicvf_rbdr_intr_handler,
1142 				  0, nic->irq_name[irq], nic);
1143 		if (ret)
1144 			goto err;
1145 		nic->irq_allocated[irq] = true;
1146 	}
1147 
1148 	/* Register QS error interrupt */
1149 	sprintf(nic->irq_name[NICVF_INTR_ID_QS_ERR], "%s-qset-err-%d",
1150 		nic->pnicvf->netdev->name,
1151 		nic->sqs_mode ? (nic->sqs_id + 1) : 0);
1152 	irq = NICVF_INTR_ID_QS_ERR;
1153 	ret = request_irq(pci_irq_vector(nic->pdev, irq),
1154 			  nicvf_qs_err_intr_handler,
1155 			  0, nic->irq_name[irq], nic);
1156 	if (ret)
1157 		goto err;
1158 
1159 	nic->irq_allocated[irq] = true;
1160 
1161 	/* Set IRQ affinities */
1162 	nicvf_set_irq_affinity(nic);
1163 
1164 err:
1165 	if (ret)
1166 		netdev_err(nic->netdev, "request_irq failed, vector %d\n", irq);
1167 
1168 	return ret;
1169 }
1170 
1171 static void nicvf_unregister_interrupts(struct nicvf *nic)
1172 {
1173 	struct pci_dev *pdev = nic->pdev;
1174 	int irq;
1175 
1176 	/* Free registered interrupts */
1177 	for (irq = 0; irq < nic->num_vec; irq++) {
1178 		if (!nic->irq_allocated[irq])
1179 			continue;
1180 
1181 		irq_set_affinity_hint(pci_irq_vector(pdev, irq), NULL);
1182 		free_cpumask_var(nic->affinity_mask[irq]);
1183 
1184 		if (irq < NICVF_INTR_ID_SQ)
1185 			free_irq(pci_irq_vector(pdev, irq), nic->napi[irq]);
1186 		else
1187 			free_irq(pci_irq_vector(pdev, irq), nic);
1188 
1189 		nic->irq_allocated[irq] = false;
1190 	}
1191 
1192 	/* Disable MSI-X */
1193 	pci_free_irq_vectors(pdev);
1194 	nic->num_vec = 0;
1195 }
1196 
1197 /* Initialize MSIX vectors and register MISC interrupt.
1198  * Send READY message to PF to check if its alive
1199  */
1200 static int nicvf_register_misc_interrupt(struct nicvf *nic)
1201 {
1202 	int ret = 0;
1203 	int irq = NICVF_INTR_ID_MISC;
1204 
1205 	/* Return if mailbox interrupt is already registered */
1206 	if (nic->pdev->msix_enabled)
1207 		return 0;
1208 
1209 	/* Enable MSI-X */
1210 	nic->num_vec = pci_msix_vec_count(nic->pdev);
1211 	ret = pci_alloc_irq_vectors(nic->pdev, nic->num_vec, nic->num_vec,
1212 				    PCI_IRQ_MSIX);
1213 	if (ret < 0) {
1214 		netdev_err(nic->netdev,
1215 			   "Req for #%d msix vectors failed\n", nic->num_vec);
1216 		return 1;
1217 	}
1218 
1219 	sprintf(nic->irq_name[irq], "%s Mbox", "NICVF");
1220 	/* Register Misc interrupt */
1221 	ret = request_irq(pci_irq_vector(nic->pdev, irq),
1222 			  nicvf_misc_intr_handler, 0, nic->irq_name[irq], nic);
1223 
1224 	if (ret)
1225 		return ret;
1226 	nic->irq_allocated[irq] = true;
1227 
1228 	/* Enable mailbox interrupt */
1229 	nicvf_enable_intr(nic, NICVF_INTR_MBOX, 0);
1230 
1231 	/* Check if VF is able to communicate with PF */
1232 	if (!nicvf_check_pf_ready(nic)) {
1233 		nicvf_disable_intr(nic, NICVF_INTR_MBOX, 0);
1234 		nicvf_unregister_interrupts(nic);
1235 		return 1;
1236 	}
1237 
1238 	return 0;
1239 }
1240 
1241 static netdev_tx_t nicvf_xmit(struct sk_buff *skb, struct net_device *netdev)
1242 {
1243 	struct nicvf *nic = netdev_priv(netdev);
1244 	int qid = skb_get_queue_mapping(skb);
1245 	struct netdev_queue *txq = netdev_get_tx_queue(netdev, qid);
1246 	struct nicvf *snic;
1247 	struct snd_queue *sq;
1248 	int tmp;
1249 
1250 	/* Check for minimum packet length */
1251 	if (skb->len <= ETH_HLEN) {
1252 		dev_kfree_skb(skb);
1253 		return NETDEV_TX_OK;
1254 	}
1255 
1256 	/* In XDP case, initial HW tx queues are used for XDP,
1257 	 * but stack's queue mapping starts at '0', so skip the
1258 	 * Tx queues attached to Rx queues for XDP.
1259 	 */
1260 	if (nic->xdp_prog)
1261 		qid += nic->xdp_tx_queues;
1262 
1263 	snic = nic;
1264 	/* Get secondary Qset's SQ structure */
1265 	if (qid >= MAX_SND_QUEUES_PER_QS) {
1266 		tmp = qid / MAX_SND_QUEUES_PER_QS;
1267 		snic = (struct nicvf *)nic->snicvf[tmp - 1];
1268 		if (!snic) {
1269 			netdev_warn(nic->netdev,
1270 				    "Secondary Qset#%d's ptr not initialized\n",
1271 				    tmp - 1);
1272 			dev_kfree_skb(skb);
1273 			return NETDEV_TX_OK;
1274 		}
1275 		qid = qid % MAX_SND_QUEUES_PER_QS;
1276 	}
1277 
1278 	sq = &snic->qs->sq[qid];
1279 	if (!netif_tx_queue_stopped(txq) &&
1280 	    !nicvf_sq_append_skb(snic, sq, skb, qid)) {
1281 		netif_tx_stop_queue(txq);
1282 
1283 		/* Barrier, so that stop_queue visible to other cpus */
1284 		smp_mb();
1285 
1286 		/* Check again, incase another cpu freed descriptors */
1287 		if (atomic_read(&sq->free_cnt) > MIN_SQ_DESC_PER_PKT_XMIT) {
1288 			netif_tx_wake_queue(txq);
1289 		} else {
1290 			this_cpu_inc(nic->drv_stats->txq_stop);
1291 			netif_warn(nic, tx_err, netdev,
1292 				   "Transmit ring full, stopping SQ%d\n", qid);
1293 		}
1294 		return NETDEV_TX_BUSY;
1295 	}
1296 
1297 	return NETDEV_TX_OK;
1298 }
1299 
1300 static inline void nicvf_free_cq_poll(struct nicvf *nic)
1301 {
1302 	struct nicvf_cq_poll *cq_poll;
1303 	int qidx;
1304 
1305 	for (qidx = 0; qidx < nic->qs->cq_cnt; qidx++) {
1306 		cq_poll = nic->napi[qidx];
1307 		if (!cq_poll)
1308 			continue;
1309 		nic->napi[qidx] = NULL;
1310 		kfree(cq_poll);
1311 	}
1312 }
1313 
1314 int nicvf_stop(struct net_device *netdev)
1315 {
1316 	int irq, qidx;
1317 	struct nicvf *nic = netdev_priv(netdev);
1318 	struct queue_set *qs = nic->qs;
1319 	struct nicvf_cq_poll *cq_poll = NULL;
1320 	union nic_mbx mbx = {};
1321 
1322 	mbx.msg.msg = NIC_MBOX_MSG_SHUTDOWN;
1323 	nicvf_send_msg_to_pf(nic, &mbx);
1324 
1325 	netif_carrier_off(netdev);
1326 	netif_tx_stop_all_queues(nic->netdev);
1327 	nic->link_up = false;
1328 
1329 	/* Teardown secondary qsets first */
1330 	if (!nic->sqs_mode) {
1331 		for (qidx = 0; qidx < nic->sqs_count; qidx++) {
1332 			if (!nic->snicvf[qidx])
1333 				continue;
1334 			nicvf_stop(nic->snicvf[qidx]->netdev);
1335 			nic->snicvf[qidx] = NULL;
1336 		}
1337 	}
1338 
1339 	/* Disable RBDR & QS error interrupts */
1340 	for (qidx = 0; qidx < qs->rbdr_cnt; qidx++) {
1341 		nicvf_disable_intr(nic, NICVF_INTR_RBDR, qidx);
1342 		nicvf_clear_intr(nic, NICVF_INTR_RBDR, qidx);
1343 	}
1344 	nicvf_disable_intr(nic, NICVF_INTR_QS_ERR, 0);
1345 	nicvf_clear_intr(nic, NICVF_INTR_QS_ERR, 0);
1346 
1347 	/* Wait for pending IRQ handlers to finish */
1348 	for (irq = 0; irq < nic->num_vec; irq++)
1349 		synchronize_irq(pci_irq_vector(nic->pdev, irq));
1350 
1351 	tasklet_kill(&nic->rbdr_task);
1352 	tasklet_kill(&nic->qs_err_task);
1353 	if (nic->rb_work_scheduled)
1354 		cancel_delayed_work_sync(&nic->rbdr_work);
1355 
1356 	for (qidx = 0; qidx < nic->qs->cq_cnt; qidx++) {
1357 		cq_poll = nic->napi[qidx];
1358 		if (!cq_poll)
1359 			continue;
1360 		napi_synchronize(&cq_poll->napi);
1361 		/* CQ intr is enabled while napi_complete,
1362 		 * so disable it now
1363 		 */
1364 		nicvf_disable_intr(nic, NICVF_INTR_CQ, qidx);
1365 		nicvf_clear_intr(nic, NICVF_INTR_CQ, qidx);
1366 		napi_disable(&cq_poll->napi);
1367 		netif_napi_del(&cq_poll->napi);
1368 	}
1369 
1370 	netif_tx_disable(netdev);
1371 
1372 	for (qidx = 0; qidx < netdev->num_tx_queues; qidx++)
1373 		netdev_tx_reset_queue(netdev_get_tx_queue(netdev, qidx));
1374 
1375 	/* Free resources */
1376 	nicvf_config_data_transfer(nic, false);
1377 
1378 	/* Disable HW Qset */
1379 	nicvf_qset_config(nic, false);
1380 
1381 	/* disable mailbox interrupt */
1382 	nicvf_disable_intr(nic, NICVF_INTR_MBOX, 0);
1383 
1384 	nicvf_unregister_interrupts(nic);
1385 
1386 	nicvf_free_cq_poll(nic);
1387 
1388 	/* Free any pending SKB saved to receive timestamp */
1389 	if (nic->ptp_skb) {
1390 		dev_kfree_skb_any(nic->ptp_skb);
1391 		nic->ptp_skb = NULL;
1392 	}
1393 
1394 	/* Clear multiqset info */
1395 	nic->pnicvf = nic;
1396 
1397 	return 0;
1398 }
1399 
1400 static int nicvf_config_hw_rx_tstamp(struct nicvf *nic, bool enable)
1401 {
1402 	union nic_mbx mbx = {};
1403 
1404 	mbx.ptp.msg = NIC_MBOX_MSG_PTP_CFG;
1405 	mbx.ptp.enable = enable;
1406 
1407 	return nicvf_send_msg_to_pf(nic, &mbx);
1408 }
1409 
1410 static int nicvf_update_hw_max_frs(struct nicvf *nic, int mtu)
1411 {
1412 	union nic_mbx mbx = {};
1413 
1414 	mbx.frs.msg = NIC_MBOX_MSG_SET_MAX_FRS;
1415 	mbx.frs.max_frs = mtu;
1416 	mbx.frs.vf_id = nic->vf_id;
1417 
1418 	return nicvf_send_msg_to_pf(nic, &mbx);
1419 }
1420 
1421 int nicvf_open(struct net_device *netdev)
1422 {
1423 	int cpu, err, qidx;
1424 	struct nicvf *nic = netdev_priv(netdev);
1425 	struct queue_set *qs = nic->qs;
1426 	struct nicvf_cq_poll *cq_poll = NULL;
1427 	union nic_mbx mbx = {};
1428 
1429 	netif_carrier_off(netdev);
1430 
1431 	err = nicvf_register_misc_interrupt(nic);
1432 	if (err)
1433 		return err;
1434 
1435 	/* Register NAPI handler for processing CQEs */
1436 	for (qidx = 0; qidx < qs->cq_cnt; qidx++) {
1437 		cq_poll = kzalloc(sizeof(*cq_poll), GFP_KERNEL);
1438 		if (!cq_poll) {
1439 			err = -ENOMEM;
1440 			goto napi_del;
1441 		}
1442 		cq_poll->cq_idx = qidx;
1443 		cq_poll->nicvf = nic;
1444 		netif_napi_add(netdev, &cq_poll->napi, nicvf_poll,
1445 			       NAPI_POLL_WEIGHT);
1446 		napi_enable(&cq_poll->napi);
1447 		nic->napi[qidx] = cq_poll;
1448 	}
1449 
1450 	/* Check if we got MAC address from PF or else generate a radom MAC */
1451 	if (!nic->sqs_mode && is_zero_ether_addr(netdev->dev_addr)) {
1452 		eth_hw_addr_random(netdev);
1453 		nicvf_hw_set_mac_addr(nic, netdev);
1454 	}
1455 
1456 	if (nic->set_mac_pending) {
1457 		nic->set_mac_pending = false;
1458 		nicvf_hw_set_mac_addr(nic, netdev);
1459 	}
1460 
1461 	/* Init tasklet for handling Qset err interrupt */
1462 	tasklet_init(&nic->qs_err_task, nicvf_handle_qs_err,
1463 		     (unsigned long)nic);
1464 
1465 	/* Init RBDR tasklet which will refill RBDR */
1466 	tasklet_init(&nic->rbdr_task, nicvf_rbdr_task,
1467 		     (unsigned long)nic);
1468 	INIT_DELAYED_WORK(&nic->rbdr_work, nicvf_rbdr_work);
1469 
1470 	/* Configure CPI alorithm */
1471 	nic->cpi_alg = cpi_alg;
1472 	if (!nic->sqs_mode)
1473 		nicvf_config_cpi(nic);
1474 
1475 	nicvf_request_sqs(nic);
1476 	if (nic->sqs_mode)
1477 		nicvf_get_primary_vf_struct(nic);
1478 
1479 	/* Configure PTP timestamp */
1480 	if (nic->ptp_clock)
1481 		nicvf_config_hw_rx_tstamp(nic, nic->hw_rx_tstamp);
1482 	atomic_set(&nic->tx_ptp_skbs, 0);
1483 	nic->ptp_skb = NULL;
1484 
1485 	/* Configure receive side scaling and MTU */
1486 	if (!nic->sqs_mode) {
1487 		nicvf_rss_init(nic);
1488 		err = nicvf_update_hw_max_frs(nic, netdev->mtu);
1489 		if (err)
1490 			goto cleanup;
1491 
1492 		/* Clear percpu stats */
1493 		for_each_possible_cpu(cpu)
1494 			memset(per_cpu_ptr(nic->drv_stats, cpu), 0,
1495 			       sizeof(struct nicvf_drv_stats));
1496 	}
1497 
1498 	err = nicvf_register_interrupts(nic);
1499 	if (err)
1500 		goto cleanup;
1501 
1502 	/* Initialize the queues */
1503 	err = nicvf_init_resources(nic);
1504 	if (err)
1505 		goto cleanup;
1506 
1507 	/* Make sure queue initialization is written */
1508 	wmb();
1509 
1510 	nicvf_reg_write(nic, NIC_VF_INT, -1);
1511 	/* Enable Qset err interrupt */
1512 	nicvf_enable_intr(nic, NICVF_INTR_QS_ERR, 0);
1513 
1514 	/* Enable completion queue interrupt */
1515 	for (qidx = 0; qidx < qs->cq_cnt; qidx++)
1516 		nicvf_enable_intr(nic, NICVF_INTR_CQ, qidx);
1517 
1518 	/* Enable RBDR threshold interrupt */
1519 	for (qidx = 0; qidx < qs->rbdr_cnt; qidx++)
1520 		nicvf_enable_intr(nic, NICVF_INTR_RBDR, qidx);
1521 
1522 	/* Send VF config done msg to PF */
1523 	mbx.msg.msg = NIC_MBOX_MSG_CFG_DONE;
1524 	nicvf_write_to_mbx(nic, &mbx);
1525 
1526 	return 0;
1527 cleanup:
1528 	nicvf_disable_intr(nic, NICVF_INTR_MBOX, 0);
1529 	nicvf_unregister_interrupts(nic);
1530 	tasklet_kill(&nic->qs_err_task);
1531 	tasklet_kill(&nic->rbdr_task);
1532 napi_del:
1533 	for (qidx = 0; qidx < qs->cq_cnt; qidx++) {
1534 		cq_poll = nic->napi[qidx];
1535 		if (!cq_poll)
1536 			continue;
1537 		napi_disable(&cq_poll->napi);
1538 		netif_napi_del(&cq_poll->napi);
1539 	}
1540 	nicvf_free_cq_poll(nic);
1541 	return err;
1542 }
1543 
1544 static int nicvf_change_mtu(struct net_device *netdev, int new_mtu)
1545 {
1546 	struct nicvf *nic = netdev_priv(netdev);
1547 	int orig_mtu = netdev->mtu;
1548 
1549 	netdev->mtu = new_mtu;
1550 
1551 	if (!netif_running(netdev))
1552 		return 0;
1553 
1554 	if (nicvf_update_hw_max_frs(nic, new_mtu)) {
1555 		netdev->mtu = orig_mtu;
1556 		return -EINVAL;
1557 	}
1558 
1559 	return 0;
1560 }
1561 
1562 static int nicvf_set_mac_address(struct net_device *netdev, void *p)
1563 {
1564 	struct sockaddr *addr = p;
1565 	struct nicvf *nic = netdev_priv(netdev);
1566 
1567 	if (!is_valid_ether_addr(addr->sa_data))
1568 		return -EADDRNOTAVAIL;
1569 
1570 	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1571 
1572 	if (nic->pdev->msix_enabled) {
1573 		if (nicvf_hw_set_mac_addr(nic, netdev))
1574 			return -EBUSY;
1575 	} else {
1576 		nic->set_mac_pending = true;
1577 	}
1578 
1579 	return 0;
1580 }
1581 
1582 void nicvf_update_lmac_stats(struct nicvf *nic)
1583 {
1584 	int stat = 0;
1585 	union nic_mbx mbx = {};
1586 
1587 	if (!netif_running(nic->netdev))
1588 		return;
1589 
1590 	mbx.bgx_stats.msg = NIC_MBOX_MSG_BGX_STATS;
1591 	mbx.bgx_stats.vf_id = nic->vf_id;
1592 	/* Rx stats */
1593 	mbx.bgx_stats.rx = 1;
1594 	while (stat < BGX_RX_STATS_COUNT) {
1595 		mbx.bgx_stats.idx = stat;
1596 		if (nicvf_send_msg_to_pf(nic, &mbx))
1597 			return;
1598 		stat++;
1599 	}
1600 
1601 	stat = 0;
1602 
1603 	/* Tx stats */
1604 	mbx.bgx_stats.rx = 0;
1605 	while (stat < BGX_TX_STATS_COUNT) {
1606 		mbx.bgx_stats.idx = stat;
1607 		if (nicvf_send_msg_to_pf(nic, &mbx))
1608 			return;
1609 		stat++;
1610 	}
1611 }
1612 
1613 void nicvf_update_stats(struct nicvf *nic)
1614 {
1615 	int qidx, cpu;
1616 	u64 tmp_stats = 0;
1617 	struct nicvf_hw_stats *stats = &nic->hw_stats;
1618 	struct nicvf_drv_stats *drv_stats;
1619 	struct queue_set *qs = nic->qs;
1620 
1621 #define GET_RX_STATS(reg) \
1622 	nicvf_reg_read(nic, NIC_VNIC_RX_STAT_0_13 | (reg << 3))
1623 #define GET_TX_STATS(reg) \
1624 	nicvf_reg_read(nic, NIC_VNIC_TX_STAT_0_4 | (reg << 3))
1625 
1626 	stats->rx_bytes = GET_RX_STATS(RX_OCTS);
1627 	stats->rx_ucast_frames = GET_RX_STATS(RX_UCAST);
1628 	stats->rx_bcast_frames = GET_RX_STATS(RX_BCAST);
1629 	stats->rx_mcast_frames = GET_RX_STATS(RX_MCAST);
1630 	stats->rx_fcs_errors = GET_RX_STATS(RX_FCS);
1631 	stats->rx_l2_errors = GET_RX_STATS(RX_L2ERR);
1632 	stats->rx_drop_red = GET_RX_STATS(RX_RED);
1633 	stats->rx_drop_red_bytes = GET_RX_STATS(RX_RED_OCTS);
1634 	stats->rx_drop_overrun = GET_RX_STATS(RX_ORUN);
1635 	stats->rx_drop_overrun_bytes = GET_RX_STATS(RX_ORUN_OCTS);
1636 	stats->rx_drop_bcast = GET_RX_STATS(RX_DRP_BCAST);
1637 	stats->rx_drop_mcast = GET_RX_STATS(RX_DRP_MCAST);
1638 	stats->rx_drop_l3_bcast = GET_RX_STATS(RX_DRP_L3BCAST);
1639 	stats->rx_drop_l3_mcast = GET_RX_STATS(RX_DRP_L3MCAST);
1640 
1641 	stats->tx_bytes = GET_TX_STATS(TX_OCTS);
1642 	stats->tx_ucast_frames = GET_TX_STATS(TX_UCAST);
1643 	stats->tx_bcast_frames = GET_TX_STATS(TX_BCAST);
1644 	stats->tx_mcast_frames = GET_TX_STATS(TX_MCAST);
1645 	stats->tx_drops = GET_TX_STATS(TX_DROP);
1646 
1647 	/* On T88 pass 2.0, the dummy SQE added for TSO notification
1648 	 * via CQE has 'dont_send' set. Hence HW drops the pkt pointed
1649 	 * pointed by dummy SQE and results in tx_drops counter being
1650 	 * incremented. Subtracting it from tx_tso counter will give
1651 	 * exact tx_drops counter.
1652 	 */
1653 	if (nic->t88 && nic->hw_tso) {
1654 		for_each_possible_cpu(cpu) {
1655 			drv_stats = per_cpu_ptr(nic->drv_stats, cpu);
1656 			tmp_stats += drv_stats->tx_tso;
1657 		}
1658 		stats->tx_drops = tmp_stats - stats->tx_drops;
1659 	}
1660 	stats->tx_frames = stats->tx_ucast_frames +
1661 			   stats->tx_bcast_frames +
1662 			   stats->tx_mcast_frames;
1663 	stats->rx_frames = stats->rx_ucast_frames +
1664 			   stats->rx_bcast_frames +
1665 			   stats->rx_mcast_frames;
1666 	stats->rx_drops = stats->rx_drop_red +
1667 			  stats->rx_drop_overrun;
1668 
1669 	/* Update RQ and SQ stats */
1670 	for (qidx = 0; qidx < qs->rq_cnt; qidx++)
1671 		nicvf_update_rq_stats(nic, qidx);
1672 	for (qidx = 0; qidx < qs->sq_cnt; qidx++)
1673 		nicvf_update_sq_stats(nic, qidx);
1674 }
1675 
1676 static void nicvf_get_stats64(struct net_device *netdev,
1677 			      struct rtnl_link_stats64 *stats)
1678 {
1679 	struct nicvf *nic = netdev_priv(netdev);
1680 	struct nicvf_hw_stats *hw_stats = &nic->hw_stats;
1681 
1682 	nicvf_update_stats(nic);
1683 
1684 	stats->rx_bytes = hw_stats->rx_bytes;
1685 	stats->rx_packets = hw_stats->rx_frames;
1686 	stats->rx_dropped = hw_stats->rx_drops;
1687 	stats->multicast = hw_stats->rx_mcast_frames;
1688 
1689 	stats->tx_bytes = hw_stats->tx_bytes;
1690 	stats->tx_packets = hw_stats->tx_frames;
1691 	stats->tx_dropped = hw_stats->tx_drops;
1692 
1693 }
1694 
1695 static void nicvf_tx_timeout(struct net_device *dev)
1696 {
1697 	struct nicvf *nic = netdev_priv(dev);
1698 
1699 	netif_warn(nic, tx_err, dev, "Transmit timed out, resetting\n");
1700 
1701 	this_cpu_inc(nic->drv_stats->tx_timeout);
1702 	schedule_work(&nic->reset_task);
1703 }
1704 
1705 static void nicvf_reset_task(struct work_struct *work)
1706 {
1707 	struct nicvf *nic;
1708 
1709 	nic = container_of(work, struct nicvf, reset_task);
1710 
1711 	if (!netif_running(nic->netdev))
1712 		return;
1713 
1714 	nicvf_stop(nic->netdev);
1715 	nicvf_open(nic->netdev);
1716 	netif_trans_update(nic->netdev);
1717 }
1718 
1719 static int nicvf_config_loopback(struct nicvf *nic,
1720 				 netdev_features_t features)
1721 {
1722 	union nic_mbx mbx = {};
1723 
1724 	mbx.lbk.msg = NIC_MBOX_MSG_LOOPBACK;
1725 	mbx.lbk.vf_id = nic->vf_id;
1726 	mbx.lbk.enable = (features & NETIF_F_LOOPBACK) != 0;
1727 
1728 	return nicvf_send_msg_to_pf(nic, &mbx);
1729 }
1730 
1731 static netdev_features_t nicvf_fix_features(struct net_device *netdev,
1732 					    netdev_features_t features)
1733 {
1734 	struct nicvf *nic = netdev_priv(netdev);
1735 
1736 	if ((features & NETIF_F_LOOPBACK) &&
1737 	    netif_running(netdev) && !nic->loopback_supported)
1738 		features &= ~NETIF_F_LOOPBACK;
1739 
1740 	return features;
1741 }
1742 
1743 static int nicvf_set_features(struct net_device *netdev,
1744 			      netdev_features_t features)
1745 {
1746 	struct nicvf *nic = netdev_priv(netdev);
1747 	netdev_features_t changed = features ^ netdev->features;
1748 
1749 	if (changed & NETIF_F_HW_VLAN_CTAG_RX)
1750 		nicvf_config_vlan_stripping(nic, features);
1751 
1752 	if ((changed & NETIF_F_LOOPBACK) && netif_running(netdev))
1753 		return nicvf_config_loopback(nic, features);
1754 
1755 	return 0;
1756 }
1757 
1758 static void nicvf_set_xdp_queues(struct nicvf *nic, bool bpf_attached)
1759 {
1760 	u8 cq_count, txq_count;
1761 
1762 	/* Set XDP Tx queue count same as Rx queue count */
1763 	if (!bpf_attached)
1764 		nic->xdp_tx_queues = 0;
1765 	else
1766 		nic->xdp_tx_queues = nic->rx_queues;
1767 
1768 	/* If queue count > MAX_CMP_QUEUES_PER_QS, then additional qsets
1769 	 * needs to be allocated, check how many.
1770 	 */
1771 	txq_count = nic->xdp_tx_queues + nic->tx_queues;
1772 	cq_count = max(nic->rx_queues, txq_count);
1773 	if (cq_count > MAX_CMP_QUEUES_PER_QS) {
1774 		nic->sqs_count = roundup(cq_count, MAX_CMP_QUEUES_PER_QS);
1775 		nic->sqs_count = (nic->sqs_count / MAX_CMP_QUEUES_PER_QS) - 1;
1776 	} else {
1777 		nic->sqs_count = 0;
1778 	}
1779 
1780 	/* Set primary Qset's resources */
1781 	nic->qs->rq_cnt = min_t(u8, nic->rx_queues, MAX_RCV_QUEUES_PER_QS);
1782 	nic->qs->sq_cnt = min_t(u8, txq_count, MAX_SND_QUEUES_PER_QS);
1783 	nic->qs->cq_cnt = max_t(u8, nic->qs->rq_cnt, nic->qs->sq_cnt);
1784 
1785 	/* Update stack */
1786 	nicvf_set_real_num_queues(nic->netdev, nic->tx_queues, nic->rx_queues);
1787 }
1788 
1789 static int nicvf_xdp_setup(struct nicvf *nic, struct bpf_prog *prog)
1790 {
1791 	struct net_device *dev = nic->netdev;
1792 	bool if_up = netif_running(nic->netdev);
1793 	struct bpf_prog *old_prog;
1794 	bool bpf_attached = false;
1795 
1796 	/* For now just support only the usual MTU sized frames */
1797 	if (prog && (dev->mtu > 1500)) {
1798 		netdev_warn(dev, "Jumbo frames not yet supported with XDP, current MTU %d.\n",
1799 			    dev->mtu);
1800 		return -EOPNOTSUPP;
1801 	}
1802 
1803 	/* ALL SQs attached to CQs i.e same as RQs, are treated as
1804 	 * XDP Tx queues and more Tx queues are allocated for
1805 	 * network stack to send pkts out.
1806 	 *
1807 	 * No of Tx queues are either same as Rx queues or whatever
1808 	 * is left in max no of queues possible.
1809 	 */
1810 	if ((nic->rx_queues + nic->tx_queues) > nic->max_queues) {
1811 		netdev_warn(dev,
1812 			    "Failed to attach BPF prog, RXQs + TXQs > Max %d\n",
1813 			    nic->max_queues);
1814 		return -ENOMEM;
1815 	}
1816 
1817 	if (if_up)
1818 		nicvf_stop(nic->netdev);
1819 
1820 	old_prog = xchg(&nic->xdp_prog, prog);
1821 	/* Detach old prog, if any */
1822 	if (old_prog)
1823 		bpf_prog_put(old_prog);
1824 
1825 	if (nic->xdp_prog) {
1826 		/* Attach BPF program */
1827 		nic->xdp_prog = bpf_prog_add(nic->xdp_prog, nic->rx_queues - 1);
1828 		if (!IS_ERR(nic->xdp_prog))
1829 			bpf_attached = true;
1830 	}
1831 
1832 	/* Calculate Tx queues needed for XDP and network stack */
1833 	nicvf_set_xdp_queues(nic, bpf_attached);
1834 
1835 	if (if_up) {
1836 		/* Reinitialize interface, clean slate */
1837 		nicvf_open(nic->netdev);
1838 		netif_trans_update(nic->netdev);
1839 	}
1840 
1841 	return 0;
1842 }
1843 
1844 static int nicvf_xdp(struct net_device *netdev, struct netdev_bpf *xdp)
1845 {
1846 	struct nicvf *nic = netdev_priv(netdev);
1847 
1848 	/* To avoid checks while retrieving buffer address from CQE_RX,
1849 	 * do not support XDP for T88 pass1.x silicons which are anyway
1850 	 * not in use widely.
1851 	 */
1852 	if (pass1_silicon(nic->pdev))
1853 		return -EOPNOTSUPP;
1854 
1855 	switch (xdp->command) {
1856 	case XDP_SETUP_PROG:
1857 		return nicvf_xdp_setup(nic, xdp->prog);
1858 	case XDP_QUERY_PROG:
1859 		xdp->prog_attached = !!nic->xdp_prog;
1860 		xdp->prog_id = nic->xdp_prog ? nic->xdp_prog->aux->id : 0;
1861 		return 0;
1862 	default:
1863 		return -EINVAL;
1864 	}
1865 }
1866 
1867 static int nicvf_xdp_xmit(struct net_device *netdev, struct xdp_buff *xdp)
1868 {
1869 	struct nicvf *nic = netdev_priv(netdev);
1870 	struct nicvf *snic = nic;
1871 	struct nicvf_xdp_tx *xdp_tx;
1872 	struct snd_queue *sq;
1873 	struct page *page;
1874 	int err, qidx;
1875 
1876 	if (!netif_running(netdev) || !nic->xdp_prog)
1877 		return -EINVAL;
1878 
1879 	page = virt_to_page(xdp->data);
1880 	xdp_tx = (struct nicvf_xdp_tx *)page_address(page);
1881 	qidx = xdp_tx->qidx;
1882 
1883 	if (xdp_tx->qidx >= nic->xdp_tx_queues)
1884 		return -EINVAL;
1885 
1886 	/* Get secondary Qset's info */
1887 	if (xdp_tx->qidx >= MAX_SND_QUEUES_PER_QS) {
1888 		qidx = xdp_tx->qidx / MAX_SND_QUEUES_PER_QS;
1889 		snic = (struct nicvf *)nic->snicvf[qidx - 1];
1890 		if (!snic)
1891 			return -EINVAL;
1892 		qidx = xdp_tx->qidx % MAX_SND_QUEUES_PER_QS;
1893 	}
1894 
1895 	sq = &snic->qs->sq[qidx];
1896 	err = nicvf_xdp_sq_append_pkt(snic, sq, (u64)xdp->data,
1897 				      xdp_tx->dma_addr,
1898 				      xdp->data_end - xdp->data);
1899 	if (err)
1900 		return -ENOMEM;
1901 
1902 	nicvf_xdp_sq_doorbell(snic, sq, qidx);
1903 	return 0;
1904 }
1905 
1906 static void nicvf_xdp_flush(struct net_device *dev)
1907 {
1908 	return;
1909 }
1910 
1911 static int nicvf_config_hwtstamp(struct net_device *netdev, struct ifreq *ifr)
1912 {
1913 	struct hwtstamp_config config;
1914 	struct nicvf *nic = netdev_priv(netdev);
1915 
1916 	if (!nic->ptp_clock)
1917 		return -ENODEV;
1918 
1919 	if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
1920 		return -EFAULT;
1921 
1922 	/* reserved for future extensions */
1923 	if (config.flags)
1924 		return -EINVAL;
1925 
1926 	switch (config.tx_type) {
1927 	case HWTSTAMP_TX_OFF:
1928 	case HWTSTAMP_TX_ON:
1929 		break;
1930 	default:
1931 		return -ERANGE;
1932 	}
1933 
1934 	switch (config.rx_filter) {
1935 	case HWTSTAMP_FILTER_NONE:
1936 		nic->hw_rx_tstamp = false;
1937 		break;
1938 	case HWTSTAMP_FILTER_ALL:
1939 	case HWTSTAMP_FILTER_SOME:
1940 	case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
1941 	case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
1942 	case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
1943 	case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
1944 	case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
1945 	case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
1946 	case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
1947 	case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
1948 	case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
1949 	case HWTSTAMP_FILTER_PTP_V2_EVENT:
1950 	case HWTSTAMP_FILTER_PTP_V2_SYNC:
1951 	case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
1952 		nic->hw_rx_tstamp = true;
1953 		config.rx_filter = HWTSTAMP_FILTER_ALL;
1954 		break;
1955 	default:
1956 		return -ERANGE;
1957 	}
1958 
1959 	if (netif_running(netdev))
1960 		nicvf_config_hw_rx_tstamp(nic, nic->hw_rx_tstamp);
1961 
1962 	if (copy_to_user(ifr->ifr_data, &config, sizeof(config)))
1963 		return -EFAULT;
1964 
1965 	return 0;
1966 }
1967 
1968 static int nicvf_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
1969 {
1970 	switch (cmd) {
1971 	case SIOCSHWTSTAMP:
1972 		return nicvf_config_hwtstamp(netdev, req);
1973 	default:
1974 		return -EOPNOTSUPP;
1975 	}
1976 }
1977 
1978 static const struct net_device_ops nicvf_netdev_ops = {
1979 	.ndo_open		= nicvf_open,
1980 	.ndo_stop		= nicvf_stop,
1981 	.ndo_start_xmit		= nicvf_xmit,
1982 	.ndo_change_mtu		= nicvf_change_mtu,
1983 	.ndo_set_mac_address	= nicvf_set_mac_address,
1984 	.ndo_get_stats64	= nicvf_get_stats64,
1985 	.ndo_tx_timeout         = nicvf_tx_timeout,
1986 	.ndo_fix_features       = nicvf_fix_features,
1987 	.ndo_set_features       = nicvf_set_features,
1988 	.ndo_bpf		= nicvf_xdp,
1989 	.ndo_xdp_xmit		= nicvf_xdp_xmit,
1990 	.ndo_xdp_flush          = nicvf_xdp_flush,
1991 	.ndo_do_ioctl           = nicvf_ioctl,
1992 };
1993 
1994 static int nicvf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
1995 {
1996 	struct device *dev = &pdev->dev;
1997 	struct net_device *netdev;
1998 	struct nicvf *nic;
1999 	int    err, qcount;
2000 	u16    sdevid;
2001 	struct cavium_ptp *ptp_clock;
2002 
2003 	ptp_clock = cavium_ptp_get();
2004 	if (IS_ERR(ptp_clock)) {
2005 		if (PTR_ERR(ptp_clock) == -ENODEV)
2006 			/* In virtualized environment we proceed without ptp */
2007 			ptp_clock = NULL;
2008 		else
2009 			return PTR_ERR(ptp_clock);
2010 	}
2011 
2012 	err = pci_enable_device(pdev);
2013 	if (err) {
2014 		dev_err(dev, "Failed to enable PCI device\n");
2015 		return err;
2016 	}
2017 
2018 	err = pci_request_regions(pdev, DRV_NAME);
2019 	if (err) {
2020 		dev_err(dev, "PCI request regions failed 0x%x\n", err);
2021 		goto err_disable_device;
2022 	}
2023 
2024 	err = pci_set_dma_mask(pdev, DMA_BIT_MASK(48));
2025 	if (err) {
2026 		dev_err(dev, "Unable to get usable DMA configuration\n");
2027 		goto err_release_regions;
2028 	}
2029 
2030 	err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(48));
2031 	if (err) {
2032 		dev_err(dev, "unable to get 48-bit DMA for consistent allocations\n");
2033 		goto err_release_regions;
2034 	}
2035 
2036 	qcount = netif_get_num_default_rss_queues();
2037 
2038 	/* Restrict multiqset support only for host bound VFs */
2039 	if (pdev->is_virtfn) {
2040 		/* Set max number of queues per VF */
2041 		qcount = min_t(int, num_online_cpus(),
2042 			       (MAX_SQS_PER_VF + 1) * MAX_CMP_QUEUES_PER_QS);
2043 	}
2044 
2045 	netdev = alloc_etherdev_mqs(sizeof(struct nicvf), qcount, qcount);
2046 	if (!netdev) {
2047 		err = -ENOMEM;
2048 		goto err_release_regions;
2049 	}
2050 
2051 	pci_set_drvdata(pdev, netdev);
2052 
2053 	SET_NETDEV_DEV(netdev, &pdev->dev);
2054 
2055 	nic = netdev_priv(netdev);
2056 	nic->netdev = netdev;
2057 	nic->pdev = pdev;
2058 	nic->pnicvf = nic;
2059 	nic->max_queues = qcount;
2060 	/* If no of CPUs are too low, there won't be any queues left
2061 	 * for XDP_TX, hence double it.
2062 	 */
2063 	if (!nic->t88)
2064 		nic->max_queues *= 2;
2065 	nic->ptp_clock = ptp_clock;
2066 
2067 	/* MAP VF's configuration registers */
2068 	nic->reg_base = pcim_iomap(pdev, PCI_CFG_REG_BAR_NUM, 0);
2069 	if (!nic->reg_base) {
2070 		dev_err(dev, "Cannot map config register space, aborting\n");
2071 		err = -ENOMEM;
2072 		goto err_free_netdev;
2073 	}
2074 
2075 	nic->drv_stats = netdev_alloc_pcpu_stats(struct nicvf_drv_stats);
2076 	if (!nic->drv_stats) {
2077 		err = -ENOMEM;
2078 		goto err_free_netdev;
2079 	}
2080 
2081 	err = nicvf_set_qset_resources(nic);
2082 	if (err)
2083 		goto err_free_netdev;
2084 
2085 	/* Check if PF is alive and get MAC address for this VF */
2086 	err = nicvf_register_misc_interrupt(nic);
2087 	if (err)
2088 		goto err_free_netdev;
2089 
2090 	nicvf_send_vf_struct(nic);
2091 
2092 	if (!pass1_silicon(nic->pdev))
2093 		nic->hw_tso = true;
2094 
2095 	/* Get iommu domain for iova to physical addr conversion */
2096 	nic->iommu_domain = iommu_get_domain_for_dev(dev);
2097 
2098 	pci_read_config_word(nic->pdev, PCI_SUBSYSTEM_ID, &sdevid);
2099 	if (sdevid == 0xA134)
2100 		nic->t88 = true;
2101 
2102 	/* Check if this VF is in QS only mode */
2103 	if (nic->sqs_mode)
2104 		return 0;
2105 
2106 	err = nicvf_set_real_num_queues(netdev, nic->tx_queues, nic->rx_queues);
2107 	if (err)
2108 		goto err_unregister_interrupts;
2109 
2110 	netdev->hw_features = (NETIF_F_RXCSUM | NETIF_F_SG |
2111 			       NETIF_F_TSO | NETIF_F_GRO | NETIF_F_TSO6 |
2112 			       NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2113 			       NETIF_F_HW_VLAN_CTAG_RX);
2114 
2115 	netdev->hw_features |= NETIF_F_RXHASH;
2116 
2117 	netdev->features |= netdev->hw_features;
2118 	netdev->hw_features |= NETIF_F_LOOPBACK;
2119 
2120 	netdev->vlan_features = NETIF_F_SG | NETIF_F_IP_CSUM |
2121 				NETIF_F_IPV6_CSUM | NETIF_F_TSO | NETIF_F_TSO6;
2122 
2123 	netdev->netdev_ops = &nicvf_netdev_ops;
2124 	netdev->watchdog_timeo = NICVF_TX_TIMEOUT;
2125 
2126 	/* MTU range: 64 - 9200 */
2127 	netdev->min_mtu = NIC_HW_MIN_FRS;
2128 	netdev->max_mtu = NIC_HW_MAX_FRS;
2129 
2130 	INIT_WORK(&nic->reset_task, nicvf_reset_task);
2131 
2132 	err = register_netdev(netdev);
2133 	if (err) {
2134 		dev_err(dev, "Failed to register netdevice\n");
2135 		goto err_unregister_interrupts;
2136 	}
2137 
2138 	nic->msg_enable = debug;
2139 
2140 	nicvf_set_ethtool_ops(netdev);
2141 
2142 	return 0;
2143 
2144 err_unregister_interrupts:
2145 	nicvf_unregister_interrupts(nic);
2146 err_free_netdev:
2147 	pci_set_drvdata(pdev, NULL);
2148 	if (nic->drv_stats)
2149 		free_percpu(nic->drv_stats);
2150 	free_netdev(netdev);
2151 err_release_regions:
2152 	pci_release_regions(pdev);
2153 err_disable_device:
2154 	pci_disable_device(pdev);
2155 	return err;
2156 }
2157 
2158 static void nicvf_remove(struct pci_dev *pdev)
2159 {
2160 	struct net_device *netdev = pci_get_drvdata(pdev);
2161 	struct nicvf *nic;
2162 	struct net_device *pnetdev;
2163 
2164 	if (!netdev)
2165 		return;
2166 
2167 	nic = netdev_priv(netdev);
2168 	pnetdev = nic->pnicvf->netdev;
2169 
2170 	/* Check if this Qset is assigned to different VF.
2171 	 * If yes, clean primary and all secondary Qsets.
2172 	 */
2173 	if (pnetdev && (pnetdev->reg_state == NETREG_REGISTERED))
2174 		unregister_netdev(pnetdev);
2175 	nicvf_unregister_interrupts(nic);
2176 	pci_set_drvdata(pdev, NULL);
2177 	if (nic->drv_stats)
2178 		free_percpu(nic->drv_stats);
2179 	cavium_ptp_put(nic->ptp_clock);
2180 	free_netdev(netdev);
2181 	pci_release_regions(pdev);
2182 	pci_disable_device(pdev);
2183 }
2184 
2185 static void nicvf_shutdown(struct pci_dev *pdev)
2186 {
2187 	nicvf_remove(pdev);
2188 }
2189 
2190 static struct pci_driver nicvf_driver = {
2191 	.name = DRV_NAME,
2192 	.id_table = nicvf_id_table,
2193 	.probe = nicvf_probe,
2194 	.remove = nicvf_remove,
2195 	.shutdown = nicvf_shutdown,
2196 };
2197 
2198 static int __init nicvf_init_module(void)
2199 {
2200 	pr_info("%s, ver %s\n", DRV_NAME, DRV_VERSION);
2201 
2202 	return pci_register_driver(&nicvf_driver);
2203 }
2204 
2205 static void __exit nicvf_cleanup_module(void)
2206 {
2207 	pci_unregister_driver(&nicvf_driver);
2208 }
2209 
2210 module_init(nicvf_init_module);
2211 module_exit(nicvf_cleanup_module);
2212