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
20#include "nic_reg.h"
21#include "nic.h"
22#include "nicvf_queues.h"
23#include "thunder_bgx.h"
24
25#define DRV_NAME	"thunder-nicvf"
26#define DRV_VERSION	"1.0"
27
28/* Supported devices */
29static const struct pci_device_id nicvf_id_table[] = {
30	{ PCI_DEVICE_SUB(PCI_VENDOR_ID_CAVIUM,
31			 PCI_DEVICE_ID_THUNDER_NIC_VF,
32			 PCI_VENDOR_ID_CAVIUM, 0xA134) },
33	{ PCI_DEVICE_SUB(PCI_VENDOR_ID_CAVIUM,
34			 PCI_DEVICE_ID_THUNDER_PASS1_NIC_VF,
35			 PCI_VENDOR_ID_CAVIUM, 0xA11E) },
36	{ 0, }  /* end of table */
37};
38
39MODULE_AUTHOR("Sunil Goutham");
40MODULE_DESCRIPTION("Cavium Thunder NIC Virtual Function Driver");
41MODULE_LICENSE("GPL v2");
42MODULE_VERSION(DRV_VERSION);
43MODULE_DEVICE_TABLE(pci, nicvf_id_table);
44
45static int debug = 0x00;
46module_param(debug, int, 0644);
47MODULE_PARM_DESC(debug, "Debug message level bitmap");
48
49static int cpi_alg = CPI_ALG_NONE;
50module_param(cpi_alg, int, S_IRUGO);
51MODULE_PARM_DESC(cpi_alg,
52		 "PFC algorithm (0=none, 1=VLAN, 2=VLAN16, 3=IP Diffserv)");
53
54static inline u8 nicvf_netdev_qidx(struct nicvf *nic, u8 qidx)
55{
56	if (nic->sqs_mode)
57		return qidx + ((nic->sqs_id + 1) * MAX_CMP_QUEUES_PER_QS);
58	else
59		return qidx;
60}
61
62static inline void nicvf_set_rx_frame_cnt(struct nicvf *nic,
63					  struct sk_buff *skb)
64{
65	if (skb->len <= 64)
66		nic->drv_stats.rx_frames_64++;
67	else if (skb->len <= 127)
68		nic->drv_stats.rx_frames_127++;
69	else if (skb->len <= 255)
70		nic->drv_stats.rx_frames_255++;
71	else if (skb->len <= 511)
72		nic->drv_stats.rx_frames_511++;
73	else if (skb->len <= 1023)
74		nic->drv_stats.rx_frames_1023++;
75	else if (skb->len <= 1518)
76		nic->drv_stats.rx_frames_1518++;
77	else
78		nic->drv_stats.rx_frames_jumbo++;
79}
80
81/* The Cavium ThunderX network controller can *only* be found in SoCs
82 * containing the ThunderX ARM64 CPU implementation.  All accesses to the device
83 * registers on this platform are implicitly strongly ordered with respect
84 * to memory accesses. So writeq_relaxed() and readq_relaxed() are safe to use
85 * with no memory barriers in this driver.  The readq()/writeq() functions add
86 * explicit ordering operation which in this case are redundant, and only
87 * add overhead.
88 */
89
90/* Register read/write APIs */
91void nicvf_reg_write(struct nicvf *nic, u64 offset, u64 val)
92{
93	writeq_relaxed(val, nic->reg_base + offset);
94}
95
96u64 nicvf_reg_read(struct nicvf *nic, u64 offset)
97{
98	return readq_relaxed(nic->reg_base + offset);
99}
100
101void nicvf_queue_reg_write(struct nicvf *nic, u64 offset,
102			   u64 qidx, u64 val)
103{
104	void __iomem *addr = nic->reg_base + offset;
105
106	writeq_relaxed(val, addr + (qidx << NIC_Q_NUM_SHIFT));
107}
108
109u64 nicvf_queue_reg_read(struct nicvf *nic, u64 offset, u64 qidx)
110{
111	void __iomem *addr = nic->reg_base + offset;
112
113	return readq_relaxed(addr + (qidx << NIC_Q_NUM_SHIFT));
114}
115
116/* VF -> PF mailbox communication */
117static void nicvf_write_to_mbx(struct nicvf *nic, union nic_mbx *mbx)
118{
119	u64 *msg = (u64 *)mbx;
120
121	nicvf_reg_write(nic, NIC_VF_PF_MAILBOX_0_1 + 0, msg[0]);
122	nicvf_reg_write(nic, NIC_VF_PF_MAILBOX_0_1 + 8, msg[1]);
123}
124
125int nicvf_send_msg_to_pf(struct nicvf *nic, union nic_mbx *mbx)
126{
127	int timeout = NIC_MBOX_MSG_TIMEOUT;
128	int sleep = 10;
129
130	nic->pf_acked = false;
131	nic->pf_nacked = false;
132
133	nicvf_write_to_mbx(nic, mbx);
134
135	/* Wait for previous message to be acked, timeout 2sec */
136	while (!nic->pf_acked) {
137		if (nic->pf_nacked)
138			return -EINVAL;
139		msleep(sleep);
140		if (nic->pf_acked)
141			break;
142		timeout -= sleep;
143		if (!timeout) {
144			netdev_err(nic->netdev,
145				   "PF didn't ack to mbox msg %d from VF%d\n",
146				   (mbx->msg.msg & 0xFF), nic->vf_id);
147			return -EBUSY;
148		}
149	}
150	return 0;
151}
152
153/* Checks if VF is able to comminicate with PF
154* and also gets the VNIC number this VF is associated to.
155*/
156static int nicvf_check_pf_ready(struct nicvf *nic)
157{
158	union nic_mbx mbx = {};
159
160	mbx.msg.msg = NIC_MBOX_MSG_READY;
161	if (nicvf_send_msg_to_pf(nic, &mbx)) {
162		netdev_err(nic->netdev,
163			   "PF didn't respond to READY msg\n");
164		return 0;
165	}
166
167	return 1;
168}
169
170static void nicvf_read_bgx_stats(struct nicvf *nic, struct bgx_stats_msg *bgx)
171{
172	if (bgx->rx)
173		nic->bgx_stats.rx_stats[bgx->idx] = bgx->stats;
174	else
175		nic->bgx_stats.tx_stats[bgx->idx] = bgx->stats;
176}
177
178static void  nicvf_handle_mbx_intr(struct nicvf *nic)
179{
180	union nic_mbx mbx = {};
181	u64 *mbx_data;
182	u64 mbx_addr;
183	int i;
184
185	mbx_addr = NIC_VF_PF_MAILBOX_0_1;
186	mbx_data = (u64 *)&mbx;
187
188	for (i = 0; i < NIC_PF_VF_MAILBOX_SIZE; i++) {
189		*mbx_data = nicvf_reg_read(nic, mbx_addr);
190		mbx_data++;
191		mbx_addr += sizeof(u64);
192	}
193
194	netdev_dbg(nic->netdev, "Mbox message: msg: 0x%x\n", mbx.msg.msg);
195	switch (mbx.msg.msg) {
196	case NIC_MBOX_MSG_READY:
197		nic->pf_acked = true;
198		nic->vf_id = mbx.nic_cfg.vf_id & 0x7F;
199		nic->tns_mode = mbx.nic_cfg.tns_mode & 0x7F;
200		nic->node = mbx.nic_cfg.node_id;
201		if (!nic->set_mac_pending)
202			ether_addr_copy(nic->netdev->dev_addr,
203					mbx.nic_cfg.mac_addr);
204		nic->sqs_mode = mbx.nic_cfg.sqs_mode;
205		nic->loopback_supported = mbx.nic_cfg.loopback_supported;
206		nic->link_up = false;
207		nic->duplex = 0;
208		nic->speed = 0;
209		break;
210	case NIC_MBOX_MSG_ACK:
211		nic->pf_acked = true;
212		break;
213	case NIC_MBOX_MSG_NACK:
214		nic->pf_nacked = true;
215		break;
216	case NIC_MBOX_MSG_RSS_SIZE:
217		nic->rss_info.rss_size = mbx.rss_size.ind_tbl_size;
218		nic->pf_acked = true;
219		break;
220	case NIC_MBOX_MSG_BGX_STATS:
221		nicvf_read_bgx_stats(nic, &mbx.bgx_stats);
222		nic->pf_acked = true;
223		break;
224	case NIC_MBOX_MSG_BGX_LINK_CHANGE:
225		nic->pf_acked = true;
226		nic->link_up = mbx.link_status.link_up;
227		nic->duplex = mbx.link_status.duplex;
228		nic->speed = mbx.link_status.speed;
229		if (nic->link_up) {
230			netdev_info(nic->netdev, "%s: Link is Up %d Mbps %s\n",
231				    nic->netdev->name, nic->speed,
232				    nic->duplex == DUPLEX_FULL ?
233				"Full duplex" : "Half duplex");
234			netif_carrier_on(nic->netdev);
235			netif_tx_start_all_queues(nic->netdev);
236		} else {
237			netdev_info(nic->netdev, "%s: Link is Down\n",
238				    nic->netdev->name);
239			netif_carrier_off(nic->netdev);
240			netif_tx_stop_all_queues(nic->netdev);
241		}
242		break;
243	case NIC_MBOX_MSG_ALLOC_SQS:
244		nic->sqs_count = mbx.sqs_alloc.qs_count;
245		nic->pf_acked = true;
246		break;
247	case NIC_MBOX_MSG_SNICVF_PTR:
248		/* Primary VF: make note of secondary VF's pointer
249		 * to be used while packet transmission.
250		 */
251		nic->snicvf[mbx.nicvf.sqs_id] =
252			(struct nicvf *)mbx.nicvf.nicvf;
253		nic->pf_acked = true;
254		break;
255	case NIC_MBOX_MSG_PNICVF_PTR:
256		/* Secondary VF/Qset: make note of primary VF's pointer
257		 * to be used while packet reception, to handover packet
258		 * to primary VF's netdev.
259		 */
260		nic->pnicvf = (struct nicvf *)mbx.nicvf.nicvf;
261		nic->pf_acked = true;
262		break;
263	default:
264		netdev_err(nic->netdev,
265			   "Invalid message from PF, msg 0x%x\n", mbx.msg.msg);
266		break;
267	}
268	nicvf_clear_intr(nic, NICVF_INTR_MBOX, 0);
269}
270
271static int nicvf_hw_set_mac_addr(struct nicvf *nic, struct net_device *netdev)
272{
273	union nic_mbx mbx = {};
274
275	mbx.mac.msg = NIC_MBOX_MSG_SET_MAC;
276	mbx.mac.vf_id = nic->vf_id;
277	ether_addr_copy(mbx.mac.mac_addr, netdev->dev_addr);
278
279	return nicvf_send_msg_to_pf(nic, &mbx);
280}
281
282static void nicvf_config_cpi(struct nicvf *nic)
283{
284	union nic_mbx mbx = {};
285
286	mbx.cpi_cfg.msg = NIC_MBOX_MSG_CPI_CFG;
287	mbx.cpi_cfg.vf_id = nic->vf_id;
288	mbx.cpi_cfg.cpi_alg = nic->cpi_alg;
289	mbx.cpi_cfg.rq_cnt = nic->qs->rq_cnt;
290
291	nicvf_send_msg_to_pf(nic, &mbx);
292}
293
294static void nicvf_get_rss_size(struct nicvf *nic)
295{
296	union nic_mbx mbx = {};
297
298	mbx.rss_size.msg = NIC_MBOX_MSG_RSS_SIZE;
299	mbx.rss_size.vf_id = nic->vf_id;
300	nicvf_send_msg_to_pf(nic, &mbx);
301}
302
303void nicvf_config_rss(struct nicvf *nic)
304{
305	union nic_mbx mbx = {};
306	struct nicvf_rss_info *rss = &nic->rss_info;
307	int ind_tbl_len = rss->rss_size;
308	int i, nextq = 0;
309
310	mbx.rss_cfg.vf_id = nic->vf_id;
311	mbx.rss_cfg.hash_bits = rss->hash_bits;
312	while (ind_tbl_len) {
313		mbx.rss_cfg.tbl_offset = nextq;
314		mbx.rss_cfg.tbl_len = min(ind_tbl_len,
315					       RSS_IND_TBL_LEN_PER_MBX_MSG);
316		mbx.rss_cfg.msg = mbx.rss_cfg.tbl_offset ?
317			  NIC_MBOX_MSG_RSS_CFG_CONT : NIC_MBOX_MSG_RSS_CFG;
318
319		for (i = 0; i < mbx.rss_cfg.tbl_len; i++)
320			mbx.rss_cfg.ind_tbl[i] = rss->ind_tbl[nextq++];
321
322		nicvf_send_msg_to_pf(nic, &mbx);
323
324		ind_tbl_len -= mbx.rss_cfg.tbl_len;
325	}
326}
327
328void nicvf_set_rss_key(struct nicvf *nic)
329{
330	struct nicvf_rss_info *rss = &nic->rss_info;
331	u64 key_addr = NIC_VNIC_RSS_KEY_0_4;
332	int idx;
333
334	for (idx = 0; idx < RSS_HASH_KEY_SIZE; idx++) {
335		nicvf_reg_write(nic, key_addr, rss->key[idx]);
336		key_addr += sizeof(u64);
337	}
338}
339
340static int nicvf_rss_init(struct nicvf *nic)
341{
342	struct nicvf_rss_info *rss = &nic->rss_info;
343	int idx;
344
345	nicvf_get_rss_size(nic);
346
347	if (cpi_alg != CPI_ALG_NONE) {
348		rss->enable = false;
349		rss->hash_bits = 0;
350		return 0;
351	}
352
353	rss->enable = true;
354
355	/* Using the HW reset value for now */
356	rss->key[0] = 0xFEED0BADFEED0BADULL;
357	rss->key[1] = 0xFEED0BADFEED0BADULL;
358	rss->key[2] = 0xFEED0BADFEED0BADULL;
359	rss->key[3] = 0xFEED0BADFEED0BADULL;
360	rss->key[4] = 0xFEED0BADFEED0BADULL;
361
362	nicvf_set_rss_key(nic);
363
364	rss->cfg = RSS_IP_HASH_ENA | RSS_TCP_HASH_ENA | RSS_UDP_HASH_ENA;
365	nicvf_reg_write(nic, NIC_VNIC_RSS_CFG, rss->cfg);
366
367	rss->hash_bits =  ilog2(rounddown_pow_of_two(rss->rss_size));
368
369	for (idx = 0; idx < rss->rss_size; idx++)
370		rss->ind_tbl[idx] = ethtool_rxfh_indir_default(idx,
371							       nic->rx_queues);
372	nicvf_config_rss(nic);
373	return 1;
374}
375
376/* Request PF to allocate additional Qsets */
377static void nicvf_request_sqs(struct nicvf *nic)
378{
379	union nic_mbx mbx = {};
380	int sqs;
381	int sqs_count = nic->sqs_count;
382	int rx_queues = 0, tx_queues = 0;
383
384	/* Only primary VF should request */
385	if (nic->sqs_mode ||  !nic->sqs_count)
386		return;
387
388	mbx.sqs_alloc.msg = NIC_MBOX_MSG_ALLOC_SQS;
389	mbx.sqs_alloc.vf_id = nic->vf_id;
390	mbx.sqs_alloc.qs_count = nic->sqs_count;
391	if (nicvf_send_msg_to_pf(nic, &mbx)) {
392		/* No response from PF */
393		nic->sqs_count = 0;
394		return;
395	}
396
397	/* Return if no Secondary Qsets available */
398	if (!nic->sqs_count)
399		return;
400
401	if (nic->rx_queues > MAX_RCV_QUEUES_PER_QS)
402		rx_queues = nic->rx_queues - MAX_RCV_QUEUES_PER_QS;
403	if (nic->tx_queues > MAX_SND_QUEUES_PER_QS)
404		tx_queues = nic->tx_queues - MAX_SND_QUEUES_PER_QS;
405
406	/* Set no of Rx/Tx queues in each of the SQsets */
407	for (sqs = 0; sqs < nic->sqs_count; sqs++) {
408		mbx.nicvf.msg = NIC_MBOX_MSG_SNICVF_PTR;
409		mbx.nicvf.vf_id = nic->vf_id;
410		mbx.nicvf.sqs_id = sqs;
411		nicvf_send_msg_to_pf(nic, &mbx);
412
413		nic->snicvf[sqs]->sqs_id = sqs;
414		if (rx_queues > MAX_RCV_QUEUES_PER_QS) {
415			nic->snicvf[sqs]->qs->rq_cnt = MAX_RCV_QUEUES_PER_QS;
416			rx_queues -= MAX_RCV_QUEUES_PER_QS;
417		} else {
418			nic->snicvf[sqs]->qs->rq_cnt = rx_queues;
419			rx_queues = 0;
420		}
421
422		if (tx_queues > MAX_SND_QUEUES_PER_QS) {
423			nic->snicvf[sqs]->qs->sq_cnt = MAX_SND_QUEUES_PER_QS;
424			tx_queues -= MAX_SND_QUEUES_PER_QS;
425		} else {
426			nic->snicvf[sqs]->qs->sq_cnt = tx_queues;
427			tx_queues = 0;
428		}
429
430		nic->snicvf[sqs]->qs->cq_cnt =
431		max(nic->snicvf[sqs]->qs->rq_cnt, nic->snicvf[sqs]->qs->sq_cnt);
432
433		/* Initialize secondary Qset's queues and its interrupts */
434		nicvf_open(nic->snicvf[sqs]->netdev);
435	}
436
437	/* Update stack with actual Rx/Tx queue count allocated */
438	if (sqs_count != nic->sqs_count)
439		nicvf_set_real_num_queues(nic->netdev,
440					  nic->tx_queues, nic->rx_queues);
441}
442
443/* Send this Qset's nicvf pointer to PF.
444 * PF inturn sends primary VF's nicvf struct to secondary Qsets/VFs
445 * so that packets received by these Qsets can use primary VF's netdev
446 */
447static void nicvf_send_vf_struct(struct nicvf *nic)
448{
449	union nic_mbx mbx = {};
450
451	mbx.nicvf.msg = NIC_MBOX_MSG_NICVF_PTR;
452	mbx.nicvf.sqs_mode = nic->sqs_mode;
453	mbx.nicvf.nicvf = (u64)nic;
454	nicvf_send_msg_to_pf(nic, &mbx);
455}
456
457static void nicvf_get_primary_vf_struct(struct nicvf *nic)
458{
459	union nic_mbx mbx = {};
460
461	mbx.nicvf.msg = NIC_MBOX_MSG_PNICVF_PTR;
462	nicvf_send_msg_to_pf(nic, &mbx);
463}
464
465int nicvf_set_real_num_queues(struct net_device *netdev,
466			      int tx_queues, int rx_queues)
467{
468	int err = 0;
469
470	err = netif_set_real_num_tx_queues(netdev, tx_queues);
471	if (err) {
472		netdev_err(netdev,
473			   "Failed to set no of Tx queues: %d\n", tx_queues);
474		return err;
475	}
476
477	err = netif_set_real_num_rx_queues(netdev, rx_queues);
478	if (err)
479		netdev_err(netdev,
480			   "Failed to set no of Rx queues: %d\n", rx_queues);
481	return err;
482}
483
484static int nicvf_init_resources(struct nicvf *nic)
485{
486	int err;
487	union nic_mbx mbx = {};
488
489	mbx.msg.msg = NIC_MBOX_MSG_CFG_DONE;
490
491	/* Enable Qset */
492	nicvf_qset_config(nic, true);
493
494	/* Initialize queues and HW for data transfer */
495	err = nicvf_config_data_transfer(nic, true);
496	if (err) {
497		netdev_err(nic->netdev,
498			   "Failed to alloc/config VF's QSet resources\n");
499		return err;
500	}
501
502	/* Send VF config done msg to PF */
503	nicvf_write_to_mbx(nic, &mbx);
504
505	return 0;
506}
507
508static void nicvf_snd_pkt_handler(struct net_device *netdev,
509				  struct cmp_queue *cq,
510				  struct cqe_send_t *cqe_tx, int cqe_type)
511{
512	struct sk_buff *skb = NULL;
513	struct nicvf *nic = netdev_priv(netdev);
514	struct snd_queue *sq;
515	struct sq_hdr_subdesc *hdr;
516
517	sq = &nic->qs->sq[cqe_tx->sq_idx];
518
519	hdr = (struct sq_hdr_subdesc *)GET_SQ_DESC(sq, cqe_tx->sqe_ptr);
520	if (hdr->subdesc_type != SQ_DESC_TYPE_HEADER)
521		return;
522
523	netdev_dbg(nic->netdev,
524		   "%s Qset #%d SQ #%d SQ ptr #%d subdesc count %d\n",
525		   __func__, cqe_tx->sq_qs, cqe_tx->sq_idx,
526		   cqe_tx->sqe_ptr, hdr->subdesc_cnt);
527
528	nicvf_put_sq_desc(sq, hdr->subdesc_cnt + 1);
529	nicvf_check_cqe_tx_errs(nic, cq, cqe_tx);
530	skb = (struct sk_buff *)sq->skbuff[cqe_tx->sqe_ptr];
531	/* For TSO offloaded packets only one head SKB needs to be freed */
532	if (skb) {
533		prefetch(skb);
534		dev_consume_skb_any(skb);
535		sq->skbuff[cqe_tx->sqe_ptr] = (u64)NULL;
536	}
537}
538
539static inline void nicvf_set_rxhash(struct net_device *netdev,
540				    struct cqe_rx_t *cqe_rx,
541				    struct sk_buff *skb)
542{
543	u8 hash_type;
544	u32 hash;
545
546	if (!(netdev->features & NETIF_F_RXHASH))
547		return;
548
549	switch (cqe_rx->rss_alg) {
550	case RSS_ALG_TCP_IP:
551	case RSS_ALG_UDP_IP:
552		hash_type = PKT_HASH_TYPE_L4;
553		hash = cqe_rx->rss_tag;
554		break;
555	case RSS_ALG_IP:
556		hash_type = PKT_HASH_TYPE_L3;
557		hash = cqe_rx->rss_tag;
558		break;
559	default:
560		hash_type = PKT_HASH_TYPE_NONE;
561		hash = 0;
562	}
563
564	skb_set_hash(skb, hash, hash_type);
565}
566
567static void nicvf_rcv_pkt_handler(struct net_device *netdev,
568				  struct napi_struct *napi,
569				  struct cmp_queue *cq,
570				  struct cqe_rx_t *cqe_rx, int cqe_type)
571{
572	struct sk_buff *skb;
573	struct nicvf *nic = netdev_priv(netdev);
574	int err = 0;
575	int rq_idx;
576
577	rq_idx = nicvf_netdev_qidx(nic, cqe_rx->rq_idx);
578
579	if (nic->sqs_mode) {
580		/* Use primary VF's 'nicvf' struct */
581		nic = nic->pnicvf;
582		netdev = nic->netdev;
583	}
584
585	/* Check for errors */
586	err = nicvf_check_cqe_rx_errs(nic, cq, cqe_rx);
587	if (err && !cqe_rx->rb_cnt)
588		return;
589
590	skb = nicvf_get_rcv_skb(nic, cqe_rx);
591	if (!skb) {
592		netdev_dbg(nic->netdev, "Packet not received\n");
593		return;
594	}
595
596	if (netif_msg_pktdata(nic)) {
597		netdev_info(nic->netdev, "%s: skb 0x%p, len=%d\n", netdev->name,
598			    skb, skb->len);
599		print_hex_dump(KERN_INFO, "", DUMP_PREFIX_OFFSET, 16, 1,
600			       skb->data, skb->len, true);
601	}
602
603	/* If error packet, drop it here */
604	if (err) {
605		dev_kfree_skb_any(skb);
606		return;
607	}
608
609	nicvf_set_rx_frame_cnt(nic, skb);
610
611	nicvf_set_rxhash(netdev, cqe_rx, skb);
612
613	skb_record_rx_queue(skb, rq_idx);
614	if (netdev->hw_features & NETIF_F_RXCSUM) {
615		/* HW by default verifies TCP/UDP/SCTP checksums */
616		skb->ip_summed = CHECKSUM_UNNECESSARY;
617	} else {
618		skb_checksum_none_assert(skb);
619	}
620
621	skb->protocol = eth_type_trans(skb, netdev);
622
623	/* Check for stripped VLAN */
624	if (cqe_rx->vlan_found && cqe_rx->vlan_stripped)
625		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
626				       ntohs((__force __be16)cqe_rx->vlan_tci));
627
628	if (napi && (netdev->features & NETIF_F_GRO))
629		napi_gro_receive(napi, skb);
630	else
631		netif_receive_skb(skb);
632}
633
634static int nicvf_cq_intr_handler(struct net_device *netdev, u8 cq_idx,
635				 struct napi_struct *napi, int budget)
636{
637	int processed_cqe, work_done = 0, tx_done = 0;
638	int cqe_count, cqe_head;
639	struct nicvf *nic = netdev_priv(netdev);
640	struct queue_set *qs = nic->qs;
641	struct cmp_queue *cq = &qs->cq[cq_idx];
642	struct cqe_rx_t *cq_desc;
643	struct netdev_queue *txq;
644
645	spin_lock_bh(&cq->lock);
646loop:
647	processed_cqe = 0;
648	/* Get no of valid CQ entries to process */
649	cqe_count = nicvf_queue_reg_read(nic, NIC_QSET_CQ_0_7_STATUS, cq_idx);
650	cqe_count &= CQ_CQE_COUNT;
651	if (!cqe_count)
652		goto done;
653
654	/* Get head of the valid CQ entries */
655	cqe_head = nicvf_queue_reg_read(nic, NIC_QSET_CQ_0_7_HEAD, cq_idx) >> 9;
656	cqe_head &= 0xFFFF;
657
658	netdev_dbg(nic->netdev, "%s CQ%d cqe_count %d cqe_head %d\n",
659		   __func__, cq_idx, cqe_count, cqe_head);
660	while (processed_cqe < cqe_count) {
661		/* Get the CQ descriptor */
662		cq_desc = (struct cqe_rx_t *)GET_CQ_DESC(cq, cqe_head);
663		cqe_head++;
664		cqe_head &= (cq->dmem.q_len - 1);
665		/* Initiate prefetch for next descriptor */
666		prefetch((struct cqe_rx_t *)GET_CQ_DESC(cq, cqe_head));
667
668		if ((work_done >= budget) && napi &&
669		    (cq_desc->cqe_type != CQE_TYPE_SEND)) {
670			break;
671		}
672
673		netdev_dbg(nic->netdev, "CQ%d cq_desc->cqe_type %d\n",
674			   cq_idx, cq_desc->cqe_type);
675		switch (cq_desc->cqe_type) {
676		case CQE_TYPE_RX:
677			nicvf_rcv_pkt_handler(netdev, napi, cq,
678					      cq_desc, CQE_TYPE_RX);
679			work_done++;
680		break;
681		case CQE_TYPE_SEND:
682			nicvf_snd_pkt_handler(netdev, cq,
683					      (void *)cq_desc, CQE_TYPE_SEND);
684			tx_done++;
685		break;
686		case CQE_TYPE_INVALID:
687		case CQE_TYPE_RX_SPLIT:
688		case CQE_TYPE_RX_TCP:
689		case CQE_TYPE_SEND_PTP:
690			/* Ignore for now */
691		break;
692		}
693		processed_cqe++;
694	}
695	netdev_dbg(nic->netdev,
696		   "%s CQ%d processed_cqe %d work_done %d budget %d\n",
697		   __func__, cq_idx, processed_cqe, work_done, budget);
698
699	/* Ring doorbell to inform H/W to reuse processed CQEs */
700	nicvf_queue_reg_write(nic, NIC_QSET_CQ_0_7_DOOR,
701			      cq_idx, processed_cqe);
702
703	if ((work_done < budget) && napi)
704		goto loop;
705
706done:
707	/* Wakeup TXQ if its stopped earlier due to SQ full */
708	if (tx_done) {
709		netdev = nic->pnicvf->netdev;
710		txq = netdev_get_tx_queue(netdev,
711					  nicvf_netdev_qidx(nic, cq_idx));
712		nic = nic->pnicvf;
713		if (netif_tx_queue_stopped(txq) && netif_carrier_ok(netdev)) {
714			netif_tx_start_queue(txq);
715			nic->drv_stats.txq_wake++;
716			if (netif_msg_tx_err(nic))
717				netdev_warn(netdev,
718					    "%s: Transmit queue wakeup SQ%d\n",
719					    netdev->name, cq_idx);
720		}
721	}
722
723	spin_unlock_bh(&cq->lock);
724	return work_done;
725}
726
727static int nicvf_poll(struct napi_struct *napi, int budget)
728{
729	u64  cq_head;
730	int  work_done = 0;
731	struct net_device *netdev = napi->dev;
732	struct nicvf *nic = netdev_priv(netdev);
733	struct nicvf_cq_poll *cq;
734
735	cq = container_of(napi, struct nicvf_cq_poll, napi);
736	work_done = nicvf_cq_intr_handler(netdev, cq->cq_idx, napi, budget);
737
738	if (work_done < budget) {
739		/* Slow packet rate, exit polling */
740		napi_complete(napi);
741		/* Re-enable interrupts */
742		cq_head = nicvf_queue_reg_read(nic, NIC_QSET_CQ_0_7_HEAD,
743					       cq->cq_idx);
744		nicvf_clear_intr(nic, NICVF_INTR_CQ, cq->cq_idx);
745		nicvf_queue_reg_write(nic, NIC_QSET_CQ_0_7_HEAD,
746				      cq->cq_idx, cq_head);
747		nicvf_enable_intr(nic, NICVF_INTR_CQ, cq->cq_idx);
748	}
749	return work_done;
750}
751
752/* Qset error interrupt handler
753 *
754 * As of now only CQ errors are handled
755 */
756static void nicvf_handle_qs_err(unsigned long data)
757{
758	struct nicvf *nic = (struct nicvf *)data;
759	struct queue_set *qs = nic->qs;
760	int qidx;
761	u64 status;
762
763	netif_tx_disable(nic->netdev);
764
765	/* Check if it is CQ err */
766	for (qidx = 0; qidx < qs->cq_cnt; qidx++) {
767		status = nicvf_queue_reg_read(nic, NIC_QSET_CQ_0_7_STATUS,
768					      qidx);
769		if (!(status & CQ_ERR_MASK))
770			continue;
771		/* Process already queued CQEs and reconfig CQ */
772		nicvf_disable_intr(nic, NICVF_INTR_CQ, qidx);
773		nicvf_sq_disable(nic, qidx);
774		nicvf_cq_intr_handler(nic->netdev, qidx, NULL, 0);
775		nicvf_cmp_queue_config(nic, qs, qidx, true);
776		nicvf_sq_free_used_descs(nic->netdev, &qs->sq[qidx], qidx);
777		nicvf_sq_enable(nic, &qs->sq[qidx], qidx);
778
779		nicvf_enable_intr(nic, NICVF_INTR_CQ, qidx);
780	}
781
782	netif_tx_start_all_queues(nic->netdev);
783	/* Re-enable Qset error interrupt */
784	nicvf_enable_intr(nic, NICVF_INTR_QS_ERR, 0);
785}
786
787static void nicvf_dump_intr_status(struct nicvf *nic)
788{
789	if (netif_msg_intr(nic))
790		netdev_info(nic->netdev, "%s: interrupt status 0x%llx\n",
791			    nic->netdev->name, nicvf_reg_read(nic, NIC_VF_INT));
792}
793
794static irqreturn_t nicvf_misc_intr_handler(int irq, void *nicvf_irq)
795{
796	struct nicvf *nic = (struct nicvf *)nicvf_irq;
797	u64 intr;
798
799	nicvf_dump_intr_status(nic);
800
801	intr = nicvf_reg_read(nic, NIC_VF_INT);
802	/* Check for spurious interrupt */
803	if (!(intr & NICVF_INTR_MBOX_MASK))
804		return IRQ_HANDLED;
805
806	nicvf_handle_mbx_intr(nic);
807
808	return IRQ_HANDLED;
809}
810
811static irqreturn_t nicvf_intr_handler(int irq, void *cq_irq)
812{
813	struct nicvf_cq_poll *cq_poll = (struct nicvf_cq_poll *)cq_irq;
814	struct nicvf *nic = cq_poll->nicvf;
815	int qidx = cq_poll->cq_idx;
816
817	nicvf_dump_intr_status(nic);
818
819	/* Disable interrupts */
820	nicvf_disable_intr(nic, NICVF_INTR_CQ, qidx);
821
822	/* Schedule NAPI */
823	napi_schedule(&cq_poll->napi);
824
825	/* Clear interrupt */
826	nicvf_clear_intr(nic, NICVF_INTR_CQ, qidx);
827
828	return IRQ_HANDLED;
829}
830
831static irqreturn_t nicvf_rbdr_intr_handler(int irq, void *nicvf_irq)
832{
833	struct nicvf *nic = (struct nicvf *)nicvf_irq;
834	u8 qidx;
835
836
837	nicvf_dump_intr_status(nic);
838
839	/* Disable RBDR interrupt and schedule softirq */
840	for (qidx = 0; qidx < nic->qs->rbdr_cnt; qidx++) {
841		if (!nicvf_is_intr_enabled(nic, NICVF_INTR_RBDR, qidx))
842			continue;
843		nicvf_disable_intr(nic, NICVF_INTR_RBDR, qidx);
844		tasklet_hi_schedule(&nic->rbdr_task);
845		/* Clear interrupt */
846		nicvf_clear_intr(nic, NICVF_INTR_RBDR, qidx);
847	}
848
849	return IRQ_HANDLED;
850}
851
852static irqreturn_t nicvf_qs_err_intr_handler(int irq, void *nicvf_irq)
853{
854	struct nicvf *nic = (struct nicvf *)nicvf_irq;
855
856	nicvf_dump_intr_status(nic);
857
858	/* Disable Qset err interrupt and schedule softirq */
859	nicvf_disable_intr(nic, NICVF_INTR_QS_ERR, 0);
860	tasklet_hi_schedule(&nic->qs_err_task);
861	nicvf_clear_intr(nic, NICVF_INTR_QS_ERR, 0);
862
863	return IRQ_HANDLED;
864}
865
866static int nicvf_enable_msix(struct nicvf *nic)
867{
868	int ret, vec;
869
870	nic->num_vec = NIC_VF_MSIX_VECTORS;
871
872	for (vec = 0; vec < nic->num_vec; vec++)
873		nic->msix_entries[vec].entry = vec;
874
875	ret = pci_enable_msix(nic->pdev, nic->msix_entries, nic->num_vec);
876	if (ret) {
877		netdev_err(nic->netdev,
878			   "Req for #%d msix vectors failed\n", nic->num_vec);
879		return 0;
880	}
881	nic->msix_enabled = 1;
882	return 1;
883}
884
885static void nicvf_disable_msix(struct nicvf *nic)
886{
887	if (nic->msix_enabled) {
888		pci_disable_msix(nic->pdev);
889		nic->msix_enabled = 0;
890		nic->num_vec = 0;
891	}
892}
893
894static int nicvf_register_interrupts(struct nicvf *nic)
895{
896	int irq, ret = 0;
897	int vector;
898
899	for_each_cq_irq(irq)
900		sprintf(nic->irq_name[irq], "NICVF%d CQ%d",
901			nic->vf_id, irq);
902
903	for_each_sq_irq(irq)
904		sprintf(nic->irq_name[irq], "NICVF%d SQ%d",
905			nic->vf_id, irq - NICVF_INTR_ID_SQ);
906
907	for_each_rbdr_irq(irq)
908		sprintf(nic->irq_name[irq], "NICVF%d RBDR%d",
909			nic->vf_id, irq - NICVF_INTR_ID_RBDR);
910
911	/* Register CQ interrupts */
912	for (irq = 0; irq < nic->qs->cq_cnt; irq++) {
913		vector = nic->msix_entries[irq].vector;
914		ret = request_irq(vector, nicvf_intr_handler,
915				  0, nic->irq_name[irq], nic->napi[irq]);
916		if (ret)
917			goto err;
918		nic->irq_allocated[irq] = true;
919	}
920
921	/* Register RBDR interrupt */
922	for (irq = NICVF_INTR_ID_RBDR;
923	     irq < (NICVF_INTR_ID_RBDR + nic->qs->rbdr_cnt); irq++) {
924		vector = nic->msix_entries[irq].vector;
925		ret = request_irq(vector, nicvf_rbdr_intr_handler,
926				  0, nic->irq_name[irq], nic);
927		if (ret)
928			goto err;
929		nic->irq_allocated[irq] = true;
930	}
931
932	/* Register QS error interrupt */
933	sprintf(nic->irq_name[NICVF_INTR_ID_QS_ERR],
934		"NICVF%d Qset error", nic->vf_id);
935	irq = NICVF_INTR_ID_QS_ERR;
936	ret = request_irq(nic->msix_entries[irq].vector,
937			  nicvf_qs_err_intr_handler,
938			  0, nic->irq_name[irq], nic);
939	if (!ret)
940		nic->irq_allocated[irq] = true;
941
942err:
943	if (ret)
944		netdev_err(nic->netdev, "request_irq failed, vector %d\n", irq);
945
946	return ret;
947}
948
949static void nicvf_unregister_interrupts(struct nicvf *nic)
950{
951	int irq;
952
953	/* Free registered interrupts */
954	for (irq = 0; irq < nic->num_vec; irq++) {
955		if (!nic->irq_allocated[irq])
956			continue;
957
958		if (irq < NICVF_INTR_ID_SQ)
959			free_irq(nic->msix_entries[irq].vector, nic->napi[irq]);
960		else
961			free_irq(nic->msix_entries[irq].vector, nic);
962
963		nic->irq_allocated[irq] = false;
964	}
965
966	/* Disable MSI-X */
967	nicvf_disable_msix(nic);
968}
969
970/* Initialize MSIX vectors and register MISC interrupt.
971 * Send READY message to PF to check if its alive
972 */
973static int nicvf_register_misc_interrupt(struct nicvf *nic)
974{
975	int ret = 0;
976	int irq = NICVF_INTR_ID_MISC;
977
978	/* Return if mailbox interrupt is already registered */
979	if (nic->msix_enabled)
980		return 0;
981
982	/* Enable MSI-X */
983	if (!nicvf_enable_msix(nic))
984		return 1;
985
986	sprintf(nic->irq_name[irq], "%s Mbox", "NICVF");
987	/* Register Misc interrupt */
988	ret = request_irq(nic->msix_entries[irq].vector,
989			  nicvf_misc_intr_handler, 0, nic->irq_name[irq], nic);
990
991	if (ret)
992		return ret;
993	nic->irq_allocated[irq] = true;
994
995	/* Enable mailbox interrupt */
996	nicvf_enable_intr(nic, NICVF_INTR_MBOX, 0);
997
998	/* Check if VF is able to communicate with PF */
999	if (!nicvf_check_pf_ready(nic)) {
1000		nicvf_disable_intr(nic, NICVF_INTR_MBOX, 0);
1001		nicvf_unregister_interrupts(nic);
1002		return 1;
1003	}
1004
1005	return 0;
1006}
1007
1008static netdev_tx_t nicvf_xmit(struct sk_buff *skb, struct net_device *netdev)
1009{
1010	struct nicvf *nic = netdev_priv(netdev);
1011	int qid = skb_get_queue_mapping(skb);
1012	struct netdev_queue *txq = netdev_get_tx_queue(netdev, qid);
1013
1014	/* Check for minimum packet length */
1015	if (skb->len <= ETH_HLEN) {
1016		dev_kfree_skb(skb);
1017		return NETDEV_TX_OK;
1018	}
1019
1020	if (!netif_tx_queue_stopped(txq) && !nicvf_sq_append_skb(nic, skb)) {
1021		netif_tx_stop_queue(txq);
1022		nic->drv_stats.txq_stop++;
1023		if (netif_msg_tx_err(nic))
1024			netdev_warn(netdev,
1025				    "%s: Transmit ring full, stopping SQ%d\n",
1026				    netdev->name, qid);
1027		return NETDEV_TX_BUSY;
1028	}
1029
1030	return NETDEV_TX_OK;
1031}
1032
1033static inline void nicvf_free_cq_poll(struct nicvf *nic)
1034{
1035	struct nicvf_cq_poll *cq_poll;
1036	int qidx;
1037
1038	for (qidx = 0; qidx < nic->qs->cq_cnt; qidx++) {
1039		cq_poll = nic->napi[qidx];
1040		if (!cq_poll)
1041			continue;
1042		nic->napi[qidx] = NULL;
1043		kfree(cq_poll);
1044	}
1045}
1046
1047int nicvf_stop(struct net_device *netdev)
1048{
1049	int irq, qidx;
1050	struct nicvf *nic = netdev_priv(netdev);
1051	struct queue_set *qs = nic->qs;
1052	struct nicvf_cq_poll *cq_poll = NULL;
1053	union nic_mbx mbx = {};
1054
1055	mbx.msg.msg = NIC_MBOX_MSG_SHUTDOWN;
1056	nicvf_send_msg_to_pf(nic, &mbx);
1057
1058	netif_carrier_off(netdev);
1059	netif_tx_stop_all_queues(nic->netdev);
1060	nic->link_up = false;
1061
1062	/* Teardown secondary qsets first */
1063	if (!nic->sqs_mode) {
1064		for (qidx = 0; qidx < nic->sqs_count; qidx++) {
1065			if (!nic->snicvf[qidx])
1066				continue;
1067			nicvf_stop(nic->snicvf[qidx]->netdev);
1068			nic->snicvf[qidx] = NULL;
1069		}
1070	}
1071
1072	/* Disable RBDR & QS error interrupts */
1073	for (qidx = 0; qidx < qs->rbdr_cnt; qidx++) {
1074		nicvf_disable_intr(nic, NICVF_INTR_RBDR, qidx);
1075		nicvf_clear_intr(nic, NICVF_INTR_RBDR, qidx);
1076	}
1077	nicvf_disable_intr(nic, NICVF_INTR_QS_ERR, 0);
1078	nicvf_clear_intr(nic, NICVF_INTR_QS_ERR, 0);
1079
1080	/* Wait for pending IRQ handlers to finish */
1081	for (irq = 0; irq < nic->num_vec; irq++)
1082		synchronize_irq(nic->msix_entries[irq].vector);
1083
1084	tasklet_kill(&nic->rbdr_task);
1085	tasklet_kill(&nic->qs_err_task);
1086	if (nic->rb_work_scheduled)
1087		cancel_delayed_work_sync(&nic->rbdr_work);
1088
1089	for (qidx = 0; qidx < nic->qs->cq_cnt; qidx++) {
1090		cq_poll = nic->napi[qidx];
1091		if (!cq_poll)
1092			continue;
1093		napi_synchronize(&cq_poll->napi);
1094		/* CQ intr is enabled while napi_complete,
1095		 * so disable it now
1096		 */
1097		nicvf_disable_intr(nic, NICVF_INTR_CQ, qidx);
1098		nicvf_clear_intr(nic, NICVF_INTR_CQ, qidx);
1099		napi_disable(&cq_poll->napi);
1100		netif_napi_del(&cq_poll->napi);
1101	}
1102
1103	netif_tx_disable(netdev);
1104
1105	/* Free resources */
1106	nicvf_config_data_transfer(nic, false);
1107
1108	/* Disable HW Qset */
1109	nicvf_qset_config(nic, false);
1110
1111	/* disable mailbox interrupt */
1112	nicvf_disable_intr(nic, NICVF_INTR_MBOX, 0);
1113
1114	nicvf_unregister_interrupts(nic);
1115
1116	nicvf_free_cq_poll(nic);
1117
1118	/* Clear multiqset info */
1119	nic->pnicvf = nic;
1120	nic->sqs_count = 0;
1121
1122	return 0;
1123}
1124
1125int nicvf_open(struct net_device *netdev)
1126{
1127	int err, qidx;
1128	struct nicvf *nic = netdev_priv(netdev);
1129	struct queue_set *qs = nic->qs;
1130	struct nicvf_cq_poll *cq_poll = NULL;
1131
1132	nic->mtu = netdev->mtu;
1133
1134	netif_carrier_off(netdev);
1135
1136	err = nicvf_register_misc_interrupt(nic);
1137	if (err)
1138		return err;
1139
1140	/* Register NAPI handler for processing CQEs */
1141	for (qidx = 0; qidx < qs->cq_cnt; qidx++) {
1142		cq_poll = kzalloc(sizeof(*cq_poll), GFP_KERNEL);
1143		if (!cq_poll) {
1144			err = -ENOMEM;
1145			goto napi_del;
1146		}
1147		cq_poll->cq_idx = qidx;
1148		cq_poll->nicvf = nic;
1149		netif_napi_add(netdev, &cq_poll->napi, nicvf_poll,
1150			       NAPI_POLL_WEIGHT);
1151		napi_enable(&cq_poll->napi);
1152		nic->napi[qidx] = cq_poll;
1153	}
1154
1155	/* Check if we got MAC address from PF or else generate a radom MAC */
1156	if (is_zero_ether_addr(netdev->dev_addr)) {
1157		eth_hw_addr_random(netdev);
1158		nicvf_hw_set_mac_addr(nic, netdev);
1159	}
1160
1161	if (nic->set_mac_pending) {
1162		nic->set_mac_pending = false;
1163		nicvf_hw_set_mac_addr(nic, netdev);
1164	}
1165
1166	/* Init tasklet for handling Qset err interrupt */
1167	tasklet_init(&nic->qs_err_task, nicvf_handle_qs_err,
1168		     (unsigned long)nic);
1169
1170	/* Init RBDR tasklet which will refill RBDR */
1171	tasklet_init(&nic->rbdr_task, nicvf_rbdr_task,
1172		     (unsigned long)nic);
1173	INIT_DELAYED_WORK(&nic->rbdr_work, nicvf_rbdr_work);
1174
1175	/* Configure CPI alorithm */
1176	nic->cpi_alg = cpi_alg;
1177	if (!nic->sqs_mode)
1178		nicvf_config_cpi(nic);
1179
1180	nicvf_request_sqs(nic);
1181	if (nic->sqs_mode)
1182		nicvf_get_primary_vf_struct(nic);
1183
1184	/* Configure receive side scaling */
1185	if (!nic->sqs_mode)
1186		nicvf_rss_init(nic);
1187
1188	err = nicvf_register_interrupts(nic);
1189	if (err)
1190		goto cleanup;
1191
1192	/* Initialize the queues */
1193	err = nicvf_init_resources(nic);
1194	if (err)
1195		goto cleanup;
1196
1197	/* Make sure queue initialization is written */
1198	wmb();
1199
1200	nicvf_reg_write(nic, NIC_VF_INT, -1);
1201	/* Enable Qset err interrupt */
1202	nicvf_enable_intr(nic, NICVF_INTR_QS_ERR, 0);
1203
1204	/* Enable completion queue interrupt */
1205	for (qidx = 0; qidx < qs->cq_cnt; qidx++)
1206		nicvf_enable_intr(nic, NICVF_INTR_CQ, qidx);
1207
1208	/* Enable RBDR threshold interrupt */
1209	for (qidx = 0; qidx < qs->rbdr_cnt; qidx++)
1210		nicvf_enable_intr(nic, NICVF_INTR_RBDR, qidx);
1211
1212	nic->drv_stats.txq_stop = 0;
1213	nic->drv_stats.txq_wake = 0;
1214
1215	return 0;
1216cleanup:
1217	nicvf_disable_intr(nic, NICVF_INTR_MBOX, 0);
1218	nicvf_unregister_interrupts(nic);
1219	tasklet_kill(&nic->qs_err_task);
1220	tasklet_kill(&nic->rbdr_task);
1221napi_del:
1222	for (qidx = 0; qidx < qs->cq_cnt; qidx++) {
1223		cq_poll = nic->napi[qidx];
1224		if (!cq_poll)
1225			continue;
1226		napi_disable(&cq_poll->napi);
1227		netif_napi_del(&cq_poll->napi);
1228	}
1229	nicvf_free_cq_poll(nic);
1230	return err;
1231}
1232
1233static int nicvf_update_hw_max_frs(struct nicvf *nic, int mtu)
1234{
1235	union nic_mbx mbx = {};
1236
1237	mbx.frs.msg = NIC_MBOX_MSG_SET_MAX_FRS;
1238	mbx.frs.max_frs = mtu;
1239	mbx.frs.vf_id = nic->vf_id;
1240
1241	return nicvf_send_msg_to_pf(nic, &mbx);
1242}
1243
1244static int nicvf_change_mtu(struct net_device *netdev, int new_mtu)
1245{
1246	struct nicvf *nic = netdev_priv(netdev);
1247
1248	if (new_mtu > NIC_HW_MAX_FRS)
1249		return -EINVAL;
1250
1251	if (new_mtu < NIC_HW_MIN_FRS)
1252		return -EINVAL;
1253
1254	if (nicvf_update_hw_max_frs(nic, new_mtu))
1255		return -EINVAL;
1256	netdev->mtu = new_mtu;
1257	nic->mtu = new_mtu;
1258
1259	return 0;
1260}
1261
1262static int nicvf_set_mac_address(struct net_device *netdev, void *p)
1263{
1264	struct sockaddr *addr = p;
1265	struct nicvf *nic = netdev_priv(netdev);
1266
1267	if (!is_valid_ether_addr(addr->sa_data))
1268		return -EADDRNOTAVAIL;
1269
1270	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1271
1272	if (nic->msix_enabled) {
1273		if (nicvf_hw_set_mac_addr(nic, netdev))
1274			return -EBUSY;
1275	} else {
1276		nic->set_mac_pending = true;
1277	}
1278
1279	return 0;
1280}
1281
1282void nicvf_update_lmac_stats(struct nicvf *nic)
1283{
1284	int stat = 0;
1285	union nic_mbx mbx = {};
1286
1287	if (!netif_running(nic->netdev))
1288		return;
1289
1290	mbx.bgx_stats.msg = NIC_MBOX_MSG_BGX_STATS;
1291	mbx.bgx_stats.vf_id = nic->vf_id;
1292	/* Rx stats */
1293	mbx.bgx_stats.rx = 1;
1294	while (stat < BGX_RX_STATS_COUNT) {
1295		mbx.bgx_stats.idx = stat;
1296		if (nicvf_send_msg_to_pf(nic, &mbx))
1297			return;
1298		stat++;
1299	}
1300
1301	stat = 0;
1302
1303	/* Tx stats */
1304	mbx.bgx_stats.rx = 0;
1305	while (stat < BGX_TX_STATS_COUNT) {
1306		mbx.bgx_stats.idx = stat;
1307		if (nicvf_send_msg_to_pf(nic, &mbx))
1308			return;
1309		stat++;
1310	}
1311}
1312
1313void nicvf_update_stats(struct nicvf *nic)
1314{
1315	int qidx;
1316	struct nicvf_hw_stats *stats = &nic->hw_stats;
1317	struct nicvf_drv_stats *drv_stats = &nic->drv_stats;
1318	struct queue_set *qs = nic->qs;
1319
1320#define GET_RX_STATS(reg) \
1321	nicvf_reg_read(nic, NIC_VNIC_RX_STAT_0_13 | (reg << 3))
1322#define GET_TX_STATS(reg) \
1323	nicvf_reg_read(nic, NIC_VNIC_TX_STAT_0_4 | (reg << 3))
1324
1325	stats->rx_bytes = GET_RX_STATS(RX_OCTS);
1326	stats->rx_ucast_frames = GET_RX_STATS(RX_UCAST);
1327	stats->rx_bcast_frames = GET_RX_STATS(RX_BCAST);
1328	stats->rx_mcast_frames = GET_RX_STATS(RX_MCAST);
1329	stats->rx_fcs_errors = GET_RX_STATS(RX_FCS);
1330	stats->rx_l2_errors = GET_RX_STATS(RX_L2ERR);
1331	stats->rx_drop_red = GET_RX_STATS(RX_RED);
1332	stats->rx_drop_red_bytes = GET_RX_STATS(RX_RED_OCTS);
1333	stats->rx_drop_overrun = GET_RX_STATS(RX_ORUN);
1334	stats->rx_drop_overrun_bytes = GET_RX_STATS(RX_ORUN_OCTS);
1335	stats->rx_drop_bcast = GET_RX_STATS(RX_DRP_BCAST);
1336	stats->rx_drop_mcast = GET_RX_STATS(RX_DRP_MCAST);
1337	stats->rx_drop_l3_bcast = GET_RX_STATS(RX_DRP_L3BCAST);
1338	stats->rx_drop_l3_mcast = GET_RX_STATS(RX_DRP_L3MCAST);
1339
1340	stats->tx_bytes_ok = GET_TX_STATS(TX_OCTS);
1341	stats->tx_ucast_frames_ok = GET_TX_STATS(TX_UCAST);
1342	stats->tx_bcast_frames_ok = GET_TX_STATS(TX_BCAST);
1343	stats->tx_mcast_frames_ok = GET_TX_STATS(TX_MCAST);
1344	stats->tx_drops = GET_TX_STATS(TX_DROP);
1345
1346	drv_stats->tx_frames_ok = stats->tx_ucast_frames_ok +
1347				  stats->tx_bcast_frames_ok +
1348				  stats->tx_mcast_frames_ok;
1349	drv_stats->rx_drops = stats->rx_drop_red +
1350			      stats->rx_drop_overrun;
1351	drv_stats->tx_drops = stats->tx_drops;
1352
1353	/* Update RQ and SQ stats */
1354	for (qidx = 0; qidx < qs->rq_cnt; qidx++)
1355		nicvf_update_rq_stats(nic, qidx);
1356	for (qidx = 0; qidx < qs->sq_cnt; qidx++)
1357		nicvf_update_sq_stats(nic, qidx);
1358}
1359
1360static struct rtnl_link_stats64 *nicvf_get_stats64(struct net_device *netdev,
1361					    struct rtnl_link_stats64 *stats)
1362{
1363	struct nicvf *nic = netdev_priv(netdev);
1364	struct nicvf_hw_stats *hw_stats = &nic->hw_stats;
1365	struct nicvf_drv_stats *drv_stats = &nic->drv_stats;
1366
1367	nicvf_update_stats(nic);
1368
1369	stats->rx_bytes = hw_stats->rx_bytes;
1370	stats->rx_packets = drv_stats->rx_frames_ok;
1371	stats->rx_dropped = drv_stats->rx_drops;
1372	stats->multicast = hw_stats->rx_mcast_frames;
1373
1374	stats->tx_bytes = hw_stats->tx_bytes_ok;
1375	stats->tx_packets = drv_stats->tx_frames_ok;
1376	stats->tx_dropped = drv_stats->tx_drops;
1377
1378	return stats;
1379}
1380
1381static void nicvf_tx_timeout(struct net_device *dev)
1382{
1383	struct nicvf *nic = netdev_priv(dev);
1384
1385	if (netif_msg_tx_err(nic))
1386		netdev_warn(dev, "%s: Transmit timed out, resetting\n",
1387			    dev->name);
1388
1389	schedule_work(&nic->reset_task);
1390}
1391
1392static void nicvf_reset_task(struct work_struct *work)
1393{
1394	struct nicvf *nic;
1395
1396	nic = container_of(work, struct nicvf, reset_task);
1397
1398	if (!netif_running(nic->netdev))
1399		return;
1400
1401	nicvf_stop(nic->netdev);
1402	nicvf_open(nic->netdev);
1403	nic->netdev->trans_start = jiffies;
1404}
1405
1406static int nicvf_config_loopback(struct nicvf *nic,
1407				 netdev_features_t features)
1408{
1409	union nic_mbx mbx = {};
1410
1411	mbx.lbk.msg = NIC_MBOX_MSG_LOOPBACK;
1412	mbx.lbk.vf_id = nic->vf_id;
1413	mbx.lbk.enable = (features & NETIF_F_LOOPBACK) != 0;
1414
1415	return nicvf_send_msg_to_pf(nic, &mbx);
1416}
1417
1418static netdev_features_t nicvf_fix_features(struct net_device *netdev,
1419					    netdev_features_t features)
1420{
1421	struct nicvf *nic = netdev_priv(netdev);
1422
1423	if ((features & NETIF_F_LOOPBACK) &&
1424	    netif_running(netdev) && !nic->loopback_supported)
1425		features &= ~NETIF_F_LOOPBACK;
1426
1427	return features;
1428}
1429
1430static int nicvf_set_features(struct net_device *netdev,
1431			      netdev_features_t features)
1432{
1433	struct nicvf *nic = netdev_priv(netdev);
1434	netdev_features_t changed = features ^ netdev->features;
1435
1436	if (changed & NETIF_F_HW_VLAN_CTAG_RX)
1437		nicvf_config_vlan_stripping(nic, features);
1438
1439	if ((changed & NETIF_F_LOOPBACK) && netif_running(netdev))
1440		return nicvf_config_loopback(nic, features);
1441
1442	return 0;
1443}
1444
1445static const struct net_device_ops nicvf_netdev_ops = {
1446	.ndo_open		= nicvf_open,
1447	.ndo_stop		= nicvf_stop,
1448	.ndo_start_xmit		= nicvf_xmit,
1449	.ndo_change_mtu		= nicvf_change_mtu,
1450	.ndo_set_mac_address	= nicvf_set_mac_address,
1451	.ndo_get_stats64	= nicvf_get_stats64,
1452	.ndo_tx_timeout         = nicvf_tx_timeout,
1453	.ndo_fix_features       = nicvf_fix_features,
1454	.ndo_set_features       = nicvf_set_features,
1455};
1456
1457static int nicvf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
1458{
1459	struct device *dev = &pdev->dev;
1460	struct net_device *netdev;
1461	struct nicvf *nic;
1462	int    err, qcount;
1463
1464	err = pci_enable_device(pdev);
1465	if (err) {
1466		dev_err(dev, "Failed to enable PCI device\n");
1467		return err;
1468	}
1469
1470	err = pci_request_regions(pdev, DRV_NAME);
1471	if (err) {
1472		dev_err(dev, "PCI request regions failed 0x%x\n", err);
1473		goto err_disable_device;
1474	}
1475
1476	err = pci_set_dma_mask(pdev, DMA_BIT_MASK(48));
1477	if (err) {
1478		dev_err(dev, "Unable to get usable DMA configuration\n");
1479		goto err_release_regions;
1480	}
1481
1482	err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(48));
1483	if (err) {
1484		dev_err(dev, "unable to get 48-bit DMA for consistent allocations\n");
1485		goto err_release_regions;
1486	}
1487
1488	qcount = MAX_CMP_QUEUES_PER_QS;
1489
1490	/* Restrict multiqset support only for host bound VFs */
1491	if (pdev->is_virtfn) {
1492		/* Set max number of queues per VF */
1493		qcount = roundup(num_online_cpus(), MAX_CMP_QUEUES_PER_QS);
1494		qcount = min(qcount,
1495			     (MAX_SQS_PER_VF + 1) * MAX_CMP_QUEUES_PER_QS);
1496	}
1497
1498	netdev = alloc_etherdev_mqs(sizeof(struct nicvf), qcount, qcount);
1499	if (!netdev) {
1500		err = -ENOMEM;
1501		goto err_release_regions;
1502	}
1503
1504	pci_set_drvdata(pdev, netdev);
1505
1506	SET_NETDEV_DEV(netdev, &pdev->dev);
1507
1508	nic = netdev_priv(netdev);
1509	nic->netdev = netdev;
1510	nic->pdev = pdev;
1511	nic->pnicvf = nic;
1512	nic->max_queues = qcount;
1513
1514	/* MAP VF's configuration registers */
1515	nic->reg_base = pcim_iomap(pdev, PCI_CFG_REG_BAR_NUM, 0);
1516	if (!nic->reg_base) {
1517		dev_err(dev, "Cannot map config register space, aborting\n");
1518		err = -ENOMEM;
1519		goto err_free_netdev;
1520	}
1521
1522	err = nicvf_set_qset_resources(nic);
1523	if (err)
1524		goto err_free_netdev;
1525
1526	/* Check if PF is alive and get MAC address for this VF */
1527	err = nicvf_register_misc_interrupt(nic);
1528	if (err)
1529		goto err_free_netdev;
1530
1531	nicvf_send_vf_struct(nic);
1532
1533	/* Check if this VF is in QS only mode */
1534	if (nic->sqs_mode)
1535		return 0;
1536
1537	err = nicvf_set_real_num_queues(netdev, nic->tx_queues, nic->rx_queues);
1538	if (err)
1539		goto err_unregister_interrupts;
1540
1541	netdev->hw_features = (NETIF_F_RXCSUM | NETIF_F_IP_CSUM | NETIF_F_SG |
1542			       NETIF_F_TSO | NETIF_F_GRO |
1543			       NETIF_F_HW_VLAN_CTAG_RX);
1544
1545	netdev->hw_features |= NETIF_F_RXHASH;
1546
1547	netdev->features |= netdev->hw_features;
1548	netdev->hw_features |= NETIF_F_LOOPBACK;
1549
1550	netdev->vlan_features = NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_TSO;
1551
1552	netdev->netdev_ops = &nicvf_netdev_ops;
1553	netdev->watchdog_timeo = NICVF_TX_TIMEOUT;
1554
1555	INIT_WORK(&nic->reset_task, nicvf_reset_task);
1556
1557	err = register_netdev(netdev);
1558	if (err) {
1559		dev_err(dev, "Failed to register netdevice\n");
1560		goto err_unregister_interrupts;
1561	}
1562
1563	nic->msg_enable = debug;
1564
1565	nicvf_set_ethtool_ops(netdev);
1566
1567	return 0;
1568
1569err_unregister_interrupts:
1570	nicvf_unregister_interrupts(nic);
1571err_free_netdev:
1572	pci_set_drvdata(pdev, NULL);
1573	free_netdev(netdev);
1574err_release_regions:
1575	pci_release_regions(pdev);
1576err_disable_device:
1577	pci_disable_device(pdev);
1578	return err;
1579}
1580
1581static void nicvf_remove(struct pci_dev *pdev)
1582{
1583	struct net_device *netdev = pci_get_drvdata(pdev);
1584	struct nicvf *nic;
1585	struct net_device *pnetdev;
1586
1587	if (!netdev)
1588		return;
1589
1590	nic = netdev_priv(netdev);
1591	pnetdev = nic->pnicvf->netdev;
1592
1593	/* Check if this Qset is assigned to different VF.
1594	 * If yes, clean primary and all secondary Qsets.
1595	 */
1596	if (pnetdev && (pnetdev->reg_state == NETREG_REGISTERED))
1597		unregister_netdev(pnetdev);
1598	nicvf_unregister_interrupts(nic);
1599	pci_set_drvdata(pdev, NULL);
1600	free_netdev(netdev);
1601	pci_release_regions(pdev);
1602	pci_disable_device(pdev);
1603}
1604
1605static void nicvf_shutdown(struct pci_dev *pdev)
1606{
1607	nicvf_remove(pdev);
1608}
1609
1610static struct pci_driver nicvf_driver = {
1611	.name = DRV_NAME,
1612	.id_table = nicvf_id_table,
1613	.probe = nicvf_probe,
1614	.remove = nicvf_remove,
1615	.shutdown = nicvf_shutdown,
1616};
1617
1618static int __init nicvf_init_module(void)
1619{
1620	pr_info("%s, ver %s\n", DRV_NAME, DRV_VERSION);
1621
1622	return pci_register_driver(&nicvf_driver);
1623}
1624
1625static void __exit nicvf_cleanup_module(void)
1626{
1627	pci_unregister_driver(&nicvf_driver);
1628}
1629
1630module_init(nicvf_init_module);
1631module_exit(nicvf_cleanup_module);
1632