2 * Copyright (C) 2015 Cavium, Inc.
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.
9 #include <linux/module.h>
10 #include <linux/interrupt.h>
11 #include <linux/pci.h>
12 #include <linux/netdevice.h>
13 #include <linux/etherdevice.h>
14 #include <linux/ethtool.h>
15 #include <linux/log2.h>
16 #include <linux/prefetch.h>
17 #include <linux/irq.h>
21 #include "nicvf_queues.h"
22 #include "thunder_bgx.h"
24 #define DRV_NAME "thunder-nicvf"
25 #define DRV_VERSION "1.0"
27 /* Supported devices */
28 static const struct pci_device_id nicvf_id_table[] = {
29 { PCI_DEVICE_SUB(PCI_VENDOR_ID_CAVIUM,
30 PCI_DEVICE_ID_THUNDER_NIC_VF,
31 PCI_VENDOR_ID_CAVIUM, 0xA11E) },
32 { PCI_DEVICE_SUB(PCI_VENDOR_ID_CAVIUM,
33 PCI_DEVICE_ID_THUNDER_PASS1_NIC_VF,
34 PCI_VENDOR_ID_CAVIUM, 0xA11E) },
35 { 0, } /* end of table */
38 MODULE_AUTHOR("Sunil Goutham");
39 MODULE_DESCRIPTION("Cavium Thunder NIC Virtual Function Driver");
40 MODULE_LICENSE("GPL v2");
41 MODULE_VERSION(DRV_VERSION);
42 MODULE_DEVICE_TABLE(pci, nicvf_id_table);
44 static int debug = 0x00;
45 module_param(debug, int, 0644);
46 MODULE_PARM_DESC(debug, "Debug message level bitmap");
48 static int cpi_alg = CPI_ALG_NONE;
49 module_param(cpi_alg, int, S_IRUGO);
50 MODULE_PARM_DESC(cpi_alg,
51 "PFC algorithm (0=none, 1=VLAN, 2=VLAN16, 3=IP Diffserv)");
53 static inline void nicvf_set_rx_frame_cnt(struct nicvf *nic,
57 nic->drv_stats.rx_frames_64++;
58 else if (skb->len <= 127)
59 nic->drv_stats.rx_frames_127++;
60 else if (skb->len <= 255)
61 nic->drv_stats.rx_frames_255++;
62 else if (skb->len <= 511)
63 nic->drv_stats.rx_frames_511++;
64 else if (skb->len <= 1023)
65 nic->drv_stats.rx_frames_1023++;
66 else if (skb->len <= 1518)
67 nic->drv_stats.rx_frames_1518++;
69 nic->drv_stats.rx_frames_jumbo++;
72 /* The Cavium ThunderX network controller can *only* be found in SoCs
73 * containing the ThunderX ARM64 CPU implementation. All accesses to the device
74 * registers on this platform are implicitly strongly ordered with respect
75 * to memory accesses. So writeq_relaxed() and readq_relaxed() are safe to use
76 * with no memory barriers in this driver. The readq()/writeq() functions add
77 * explicit ordering operation which in this case are redundant, and only
81 /* Register read/write APIs */
82 void nicvf_reg_write(struct nicvf *nic, u64 offset, u64 val)
84 writeq_relaxed(val, nic->reg_base + offset);
87 u64 nicvf_reg_read(struct nicvf *nic, u64 offset)
89 return readq_relaxed(nic->reg_base + offset);
92 void nicvf_queue_reg_write(struct nicvf *nic, u64 offset,
95 void __iomem *addr = nic->reg_base + offset;
97 writeq_relaxed(val, addr + (qidx << NIC_Q_NUM_SHIFT));
100 u64 nicvf_queue_reg_read(struct nicvf *nic, u64 offset, u64 qidx)
102 void __iomem *addr = nic->reg_base + offset;
104 return readq_relaxed(addr + (qidx << NIC_Q_NUM_SHIFT));
107 /* VF -> PF mailbox communication */
109 static void nicvf_write_to_mbx(struct nicvf *nic, union nic_mbx *mbx)
111 u64 *msg = (u64 *)mbx;
113 nicvf_reg_write(nic, NIC_VF_PF_MAILBOX_0_1 + 0, msg[0]);
114 nicvf_reg_write(nic, NIC_VF_PF_MAILBOX_0_1 + 8, msg[1]);
117 int nicvf_send_msg_to_pf(struct nicvf *nic, union nic_mbx *mbx)
119 int timeout = NIC_MBOX_MSG_TIMEOUT;
122 nic->pf_acked = false;
123 nic->pf_nacked = false;
125 nicvf_write_to_mbx(nic, mbx);
127 /* Wait for previous message to be acked, timeout 2sec */
128 while (!nic->pf_acked) {
136 netdev_err(nic->netdev,
137 "PF didn't ack to mbox msg %d from VF%d\n",
138 (mbx->msg.msg & 0xFF), nic->vf_id);
145 /* Checks if VF is able to comminicate with PF
146 * and also gets the VNIC number this VF is associated to.
148 static int nicvf_check_pf_ready(struct nicvf *nic)
150 int timeout = 5000, sleep = 20;
151 union nic_mbx mbx = {};
153 mbx.msg.msg = NIC_MBOX_MSG_READY;
155 nic->pf_ready_to_rcv_msg = false;
157 nicvf_write_to_mbx(nic, &mbx);
159 while (!nic->pf_ready_to_rcv_msg) {
161 if (nic->pf_ready_to_rcv_msg)
165 netdev_err(nic->netdev,
166 "PF didn't respond to READY msg\n");
173 static void nicvf_read_bgx_stats(struct nicvf *nic, struct bgx_stats_msg *bgx)
176 nic->bgx_stats.rx_stats[bgx->idx] = bgx->stats;
178 nic->bgx_stats.tx_stats[bgx->idx] = bgx->stats;
181 static void nicvf_handle_mbx_intr(struct nicvf *nic)
183 union nic_mbx mbx = {};
188 mbx_addr = NIC_VF_PF_MAILBOX_0_1;
189 mbx_data = (u64 *)&mbx;
191 for (i = 0; i < NIC_PF_VF_MAILBOX_SIZE; i++) {
192 *mbx_data = nicvf_reg_read(nic, mbx_addr);
194 mbx_addr += sizeof(u64);
197 netdev_dbg(nic->netdev, "Mbox message: msg: 0x%x\n", mbx.msg.msg);
198 switch (mbx.msg.msg) {
199 case NIC_MBOX_MSG_READY:
200 nic->pf_ready_to_rcv_msg = true;
201 nic->vf_id = mbx.nic_cfg.vf_id & 0x7F;
202 nic->tns_mode = mbx.nic_cfg.tns_mode & 0x7F;
203 nic->node = mbx.nic_cfg.node_id;
204 if (!nic->set_mac_pending)
205 ether_addr_copy(nic->netdev->dev_addr,
206 mbx.nic_cfg.mac_addr);
207 nic->link_up = false;
211 case NIC_MBOX_MSG_ACK:
212 nic->pf_acked = true;
214 case NIC_MBOX_MSG_NACK:
215 nic->pf_nacked = true;
217 case NIC_MBOX_MSG_RSS_SIZE:
218 nic->rss_info.rss_size = mbx.rss_size.ind_tbl_size;
219 nic->pf_acked = true;
221 case NIC_MBOX_MSG_BGX_STATS:
222 nicvf_read_bgx_stats(nic, &mbx.bgx_stats);
223 nic->pf_acked = true;
224 nic->bgx_stats_acked = true;
226 case NIC_MBOX_MSG_BGX_LINK_CHANGE:
227 nic->pf_acked = true;
228 nic->link_up = mbx.link_status.link_up;
229 nic->duplex = mbx.link_status.duplex;
230 nic->speed = mbx.link_status.speed;
232 netdev_info(nic->netdev, "%s: Link is Up %d Mbps %s\n",
233 nic->netdev->name, nic->speed,
234 nic->duplex == DUPLEX_FULL ?
235 "Full duplex" : "Half duplex");
236 netif_carrier_on(nic->netdev);
237 netif_tx_start_all_queues(nic->netdev);
239 netdev_info(nic->netdev, "%s: Link is Down\n",
241 netif_carrier_off(nic->netdev);
242 netif_tx_stop_all_queues(nic->netdev);
246 netdev_err(nic->netdev,
247 "Invalid message from PF, msg 0x%x\n", mbx.msg.msg);
250 nicvf_clear_intr(nic, NICVF_INTR_MBOX, 0);
253 static int nicvf_hw_set_mac_addr(struct nicvf *nic, struct net_device *netdev)
255 union nic_mbx mbx = {};
257 mbx.mac.msg = NIC_MBOX_MSG_SET_MAC;
258 mbx.mac.vf_id = nic->vf_id;
259 ether_addr_copy(mbx.mac.mac_addr, netdev->dev_addr);
261 return nicvf_send_msg_to_pf(nic, &mbx);
264 static void nicvf_config_cpi(struct nicvf *nic)
266 union nic_mbx mbx = {};
268 mbx.cpi_cfg.msg = NIC_MBOX_MSG_CPI_CFG;
269 mbx.cpi_cfg.vf_id = nic->vf_id;
270 mbx.cpi_cfg.cpi_alg = nic->cpi_alg;
271 mbx.cpi_cfg.rq_cnt = nic->qs->rq_cnt;
273 nicvf_send_msg_to_pf(nic, &mbx);
276 static void nicvf_get_rss_size(struct nicvf *nic)
278 union nic_mbx mbx = {};
280 mbx.rss_size.msg = NIC_MBOX_MSG_RSS_SIZE;
281 mbx.rss_size.vf_id = nic->vf_id;
282 nicvf_send_msg_to_pf(nic, &mbx);
285 void nicvf_config_rss(struct nicvf *nic)
287 union nic_mbx mbx = {};
288 struct nicvf_rss_info *rss = &nic->rss_info;
289 int ind_tbl_len = rss->rss_size;
292 mbx.rss_cfg.vf_id = nic->vf_id;
293 mbx.rss_cfg.hash_bits = rss->hash_bits;
294 while (ind_tbl_len) {
295 mbx.rss_cfg.tbl_offset = nextq;
296 mbx.rss_cfg.tbl_len = min(ind_tbl_len,
297 RSS_IND_TBL_LEN_PER_MBX_MSG);
298 mbx.rss_cfg.msg = mbx.rss_cfg.tbl_offset ?
299 NIC_MBOX_MSG_RSS_CFG_CONT : NIC_MBOX_MSG_RSS_CFG;
301 for (i = 0; i < mbx.rss_cfg.tbl_len; i++)
302 mbx.rss_cfg.ind_tbl[i] = rss->ind_tbl[nextq++];
304 nicvf_send_msg_to_pf(nic, &mbx);
306 ind_tbl_len -= mbx.rss_cfg.tbl_len;
310 void nicvf_set_rss_key(struct nicvf *nic)
312 struct nicvf_rss_info *rss = &nic->rss_info;
313 u64 key_addr = NIC_VNIC_RSS_KEY_0_4;
316 for (idx = 0; idx < RSS_HASH_KEY_SIZE; idx++) {
317 nicvf_reg_write(nic, key_addr, rss->key[idx]);
318 key_addr += sizeof(u64);
322 static int nicvf_rss_init(struct nicvf *nic)
324 struct nicvf_rss_info *rss = &nic->rss_info;
327 nicvf_get_rss_size(nic);
329 if ((nic->qs->rq_cnt <= 1) || (cpi_alg != CPI_ALG_NONE)) {
337 /* Using the HW reset value for now */
338 rss->key[0] = 0xFEED0BADFEED0BADULL;
339 rss->key[1] = 0xFEED0BADFEED0BADULL;
340 rss->key[2] = 0xFEED0BADFEED0BADULL;
341 rss->key[3] = 0xFEED0BADFEED0BADULL;
342 rss->key[4] = 0xFEED0BADFEED0BADULL;
344 nicvf_set_rss_key(nic);
346 rss->cfg = RSS_IP_HASH_ENA | RSS_TCP_HASH_ENA | RSS_UDP_HASH_ENA;
347 nicvf_reg_write(nic, NIC_VNIC_RSS_CFG, rss->cfg);
349 rss->hash_bits = ilog2(rounddown_pow_of_two(rss->rss_size));
351 for (idx = 0; idx < rss->rss_size; idx++)
352 rss->ind_tbl[idx] = ethtool_rxfh_indir_default(idx,
354 nicvf_config_rss(nic);
358 int nicvf_set_real_num_queues(struct net_device *netdev,
359 int tx_queues, int rx_queues)
363 err = netif_set_real_num_tx_queues(netdev, tx_queues);
366 "Failed to set no of Tx queues: %d\n", tx_queues);
370 err = netif_set_real_num_rx_queues(netdev, rx_queues);
373 "Failed to set no of Rx queues: %d\n", rx_queues);
377 static int nicvf_init_resources(struct nicvf *nic)
380 union nic_mbx mbx = {};
382 mbx.msg.msg = NIC_MBOX_MSG_CFG_DONE;
385 nicvf_qset_config(nic, true);
387 /* Initialize queues and HW for data transfer */
388 err = nicvf_config_data_transfer(nic, true);
390 netdev_err(nic->netdev,
391 "Failed to alloc/config VF's QSet resources\n");
395 /* Send VF config done msg to PF */
396 nicvf_write_to_mbx(nic, &mbx);
401 static void nicvf_snd_pkt_handler(struct net_device *netdev,
402 struct cmp_queue *cq,
403 struct cqe_send_t *cqe_tx, int cqe_type)
405 struct sk_buff *skb = NULL;
406 struct nicvf *nic = netdev_priv(netdev);
407 struct snd_queue *sq;
408 struct sq_hdr_subdesc *hdr;
410 sq = &nic->qs->sq[cqe_tx->sq_idx];
412 hdr = (struct sq_hdr_subdesc *)GET_SQ_DESC(sq, cqe_tx->sqe_ptr);
413 if (hdr->subdesc_type != SQ_DESC_TYPE_HEADER)
416 netdev_dbg(nic->netdev,
417 "%s Qset #%d SQ #%d SQ ptr #%d subdesc count %d\n",
418 __func__, cqe_tx->sq_qs, cqe_tx->sq_idx,
419 cqe_tx->sqe_ptr, hdr->subdesc_cnt);
421 nicvf_put_sq_desc(sq, hdr->subdesc_cnt + 1);
422 nicvf_check_cqe_tx_errs(nic, cq, cqe_tx);
423 skb = (struct sk_buff *)sq->skbuff[cqe_tx->sqe_ptr];
424 /* For TSO offloaded packets only one head SKB needs to be freed */
427 dev_consume_skb_any(skb);
428 sq->skbuff[cqe_tx->sqe_ptr] = (u64)NULL;
432 static void nicvf_rcv_pkt_handler(struct net_device *netdev,
433 struct napi_struct *napi,
434 struct cmp_queue *cq,
435 struct cqe_rx_t *cqe_rx, int cqe_type)
438 struct nicvf *nic = netdev_priv(netdev);
441 /* Check for errors */
442 err = nicvf_check_cqe_rx_errs(nic, cq, cqe_rx);
443 if (err && !cqe_rx->rb_cnt)
446 skb = nicvf_get_rcv_skb(nic, cqe_rx);
448 netdev_dbg(nic->netdev, "Packet not received\n");
452 if (netif_msg_pktdata(nic)) {
453 netdev_info(nic->netdev, "%s: skb 0x%p, len=%d\n", netdev->name,
455 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_OFFSET, 16, 1,
456 skb->data, skb->len, true);
459 nicvf_set_rx_frame_cnt(nic, skb);
461 skb_record_rx_queue(skb, cqe_rx->rq_idx);
462 if (netdev->hw_features & NETIF_F_RXCSUM) {
463 /* HW by default verifies TCP/UDP/SCTP checksums */
464 skb->ip_summed = CHECKSUM_UNNECESSARY;
466 skb_checksum_none_assert(skb);
469 skb->protocol = eth_type_trans(skb, netdev);
471 if (napi && (netdev->features & NETIF_F_GRO))
472 napi_gro_receive(napi, skb);
474 netif_receive_skb(skb);
477 static int nicvf_cq_intr_handler(struct net_device *netdev, u8 cq_idx,
478 struct napi_struct *napi, int budget)
480 int processed_cqe, work_done = 0, tx_done = 0;
481 int cqe_count, cqe_head;
482 struct nicvf *nic = netdev_priv(netdev);
483 struct queue_set *qs = nic->qs;
484 struct cmp_queue *cq = &qs->cq[cq_idx];
485 struct cqe_rx_t *cq_desc;
486 struct netdev_queue *txq;
488 spin_lock_bh(&cq->lock);
491 /* Get no of valid CQ entries to process */
492 cqe_count = nicvf_queue_reg_read(nic, NIC_QSET_CQ_0_7_STATUS, cq_idx);
493 cqe_count &= CQ_CQE_COUNT;
497 /* Get head of the valid CQ entries */
498 cqe_head = nicvf_queue_reg_read(nic, NIC_QSET_CQ_0_7_HEAD, cq_idx) >> 9;
501 netdev_dbg(nic->netdev, "%s CQ%d cqe_count %d cqe_head %d\n",
502 __func__, cq_idx, cqe_count, cqe_head);
503 while (processed_cqe < cqe_count) {
504 /* Get the CQ descriptor */
505 cq_desc = (struct cqe_rx_t *)GET_CQ_DESC(cq, cqe_head);
507 cqe_head &= (cq->dmem.q_len - 1);
508 /* Initiate prefetch for next descriptor */
509 prefetch((struct cqe_rx_t *)GET_CQ_DESC(cq, cqe_head));
511 if ((work_done >= budget) && napi &&
512 (cq_desc->cqe_type != CQE_TYPE_SEND)) {
516 netdev_dbg(nic->netdev, "CQ%d cq_desc->cqe_type %d\n",
517 cq_idx, cq_desc->cqe_type);
518 switch (cq_desc->cqe_type) {
520 nicvf_rcv_pkt_handler(netdev, napi, cq,
521 cq_desc, CQE_TYPE_RX);
525 nicvf_snd_pkt_handler(netdev, cq,
526 (void *)cq_desc, CQE_TYPE_SEND);
529 case CQE_TYPE_INVALID:
530 case CQE_TYPE_RX_SPLIT:
531 case CQE_TYPE_RX_TCP:
532 case CQE_TYPE_SEND_PTP:
538 netdev_dbg(nic->netdev,
539 "%s CQ%d processed_cqe %d work_done %d budget %d\n",
540 __func__, cq_idx, processed_cqe, work_done, budget);
542 /* Ring doorbell to inform H/W to reuse processed CQEs */
543 nicvf_queue_reg_write(nic, NIC_QSET_CQ_0_7_DOOR,
544 cq_idx, processed_cqe);
546 if ((work_done < budget) && napi)
550 /* Wakeup TXQ if its stopped earlier due to SQ full */
552 txq = netdev_get_tx_queue(netdev, cq_idx);
553 if (netif_tx_queue_stopped(txq)) {
554 netif_tx_start_queue(txq);
555 nic->drv_stats.txq_wake++;
556 if (netif_msg_tx_err(nic))
558 "%s: Transmit queue wakeup SQ%d\n",
559 netdev->name, cq_idx);
563 spin_unlock_bh(&cq->lock);
567 static int nicvf_poll(struct napi_struct *napi, int budget)
571 struct net_device *netdev = napi->dev;
572 struct nicvf *nic = netdev_priv(netdev);
573 struct nicvf_cq_poll *cq;
575 cq = container_of(napi, struct nicvf_cq_poll, napi);
576 work_done = nicvf_cq_intr_handler(netdev, cq->cq_idx, napi, budget);
578 if (work_done < budget) {
579 /* Slow packet rate, exit polling */
581 /* Re-enable interrupts */
582 cq_head = nicvf_queue_reg_read(nic, NIC_QSET_CQ_0_7_HEAD,
584 nicvf_clear_intr(nic, NICVF_INTR_CQ, cq->cq_idx);
585 nicvf_queue_reg_write(nic, NIC_QSET_CQ_0_7_HEAD,
586 cq->cq_idx, cq_head);
587 nicvf_enable_intr(nic, NICVF_INTR_CQ, cq->cq_idx);
592 /* Qset error interrupt handler
594 * As of now only CQ errors are handled
596 static void nicvf_handle_qs_err(unsigned long data)
598 struct nicvf *nic = (struct nicvf *)data;
599 struct queue_set *qs = nic->qs;
603 netif_tx_disable(nic->netdev);
605 /* Check if it is CQ err */
606 for (qidx = 0; qidx < qs->cq_cnt; qidx++) {
607 status = nicvf_queue_reg_read(nic, NIC_QSET_CQ_0_7_STATUS,
609 if (!(status & CQ_ERR_MASK))
611 /* Process already queued CQEs and reconfig CQ */
612 nicvf_disable_intr(nic, NICVF_INTR_CQ, qidx);
613 nicvf_sq_disable(nic, qidx);
614 nicvf_cq_intr_handler(nic->netdev, qidx, NULL, 0);
615 nicvf_cmp_queue_config(nic, qs, qidx, true);
616 nicvf_sq_free_used_descs(nic->netdev, &qs->sq[qidx], qidx);
617 nicvf_sq_enable(nic, &qs->sq[qidx], qidx);
619 nicvf_enable_intr(nic, NICVF_INTR_CQ, qidx);
622 netif_tx_start_all_queues(nic->netdev);
623 /* Re-enable Qset error interrupt */
624 nicvf_enable_intr(nic, NICVF_INTR_QS_ERR, 0);
627 static irqreturn_t nicvf_misc_intr_handler(int irq, void *nicvf_irq)
629 struct nicvf *nic = (struct nicvf *)nicvf_irq;
632 intr = nicvf_reg_read(nic, NIC_VF_INT);
633 /* Check for spurious interrupt */
634 if (!(intr & NICVF_INTR_MBOX_MASK))
637 nicvf_handle_mbx_intr(nic);
642 static irqreturn_t nicvf_intr_handler(int irq, void *nicvf_irq)
644 u64 qidx, intr, clear_intr = 0;
645 u64 cq_intr, rbdr_intr, qs_err_intr;
646 struct nicvf *nic = (struct nicvf *)nicvf_irq;
647 struct queue_set *qs = nic->qs;
648 struct nicvf_cq_poll *cq_poll = NULL;
650 intr = nicvf_reg_read(nic, NIC_VF_INT);
651 if (netif_msg_intr(nic))
652 netdev_info(nic->netdev, "%s: interrupt status 0x%llx\n",
653 nic->netdev->name, intr);
655 qs_err_intr = intr & NICVF_INTR_QS_ERR_MASK;
657 /* Disable Qset err interrupt and schedule softirq */
658 nicvf_disable_intr(nic, NICVF_INTR_QS_ERR, 0);
659 tasklet_hi_schedule(&nic->qs_err_task);
660 clear_intr |= qs_err_intr;
663 /* Disable interrupts and start polling */
664 cq_intr = (intr & NICVF_INTR_CQ_MASK) >> NICVF_INTR_CQ_SHIFT;
665 for (qidx = 0; qidx < qs->cq_cnt; qidx++) {
666 if (!(cq_intr & (1 << qidx)))
668 if (!nicvf_is_intr_enabled(nic, NICVF_INTR_CQ, qidx))
671 nicvf_disable_intr(nic, NICVF_INTR_CQ, qidx);
672 clear_intr |= ((1 << qidx) << NICVF_INTR_CQ_SHIFT);
674 cq_poll = nic->napi[qidx];
677 napi_schedule(&cq_poll->napi);
680 /* Handle RBDR interrupts */
681 rbdr_intr = (intr & NICVF_INTR_RBDR_MASK) >> NICVF_INTR_RBDR_SHIFT;
683 /* Disable RBDR interrupt and schedule softirq */
684 for (qidx = 0; qidx < qs->rbdr_cnt; qidx++) {
685 if (!nicvf_is_intr_enabled(nic, NICVF_INTR_RBDR, qidx))
687 nicvf_disable_intr(nic, NICVF_INTR_RBDR, qidx);
688 tasklet_hi_schedule(&nic->rbdr_task);
689 clear_intr |= ((1 << qidx) << NICVF_INTR_RBDR_SHIFT);
693 /* Clear interrupts */
694 nicvf_reg_write(nic, NIC_VF_INT, clear_intr);
698 static int nicvf_enable_msix(struct nicvf *nic)
702 nic->num_vec = NIC_VF_MSIX_VECTORS;
704 for (vec = 0; vec < nic->num_vec; vec++)
705 nic->msix_entries[vec].entry = vec;
707 ret = pci_enable_msix(nic->pdev, nic->msix_entries, nic->num_vec);
709 netdev_err(nic->netdev,
710 "Req for #%d msix vectors failed\n", nic->num_vec);
713 nic->msix_enabled = 1;
717 static void nicvf_disable_msix(struct nicvf *nic)
719 if (nic->msix_enabled) {
720 pci_disable_msix(nic->pdev);
721 nic->msix_enabled = 0;
726 static int nicvf_register_interrupts(struct nicvf *nic)
728 int irq, free, ret = 0;
732 sprintf(nic->irq_name[irq], "NICVF%d CQ%d",
736 sprintf(nic->irq_name[irq], "NICVF%d SQ%d",
737 nic->vf_id, irq - NICVF_INTR_ID_SQ);
739 for_each_rbdr_irq(irq)
740 sprintf(nic->irq_name[irq], "NICVF%d RBDR%d",
741 nic->vf_id, irq - NICVF_INTR_ID_RBDR);
743 /* Register all interrupts except mailbox */
744 for (irq = 0; irq < NICVF_INTR_ID_SQ; irq++) {
745 vector = nic->msix_entries[irq].vector;
746 ret = request_irq(vector, nicvf_intr_handler,
747 0, nic->irq_name[irq], nic);
750 nic->irq_allocated[irq] = true;
753 for (irq = NICVF_INTR_ID_SQ; irq < NICVF_INTR_ID_MISC; irq++) {
754 vector = nic->msix_entries[irq].vector;
755 ret = request_irq(vector, nicvf_intr_handler,
756 0, nic->irq_name[irq], nic);
759 nic->irq_allocated[irq] = true;
762 sprintf(nic->irq_name[NICVF_INTR_ID_QS_ERR],
763 "NICVF%d Qset error", nic->vf_id);
765 vector = nic->msix_entries[NICVF_INTR_ID_QS_ERR].vector;
766 irq = NICVF_INTR_ID_QS_ERR;
767 ret = request_irq(vector, nicvf_intr_handler,
768 0, nic->irq_name[irq], nic);
770 nic->irq_allocated[irq] = true;
774 netdev_err(nic->netdev, "Request irq failed\n");
775 for (free = 0; free < irq; free++)
776 free_irq(nic->msix_entries[free].vector, nic);
783 static void nicvf_unregister_interrupts(struct nicvf *nic)
787 /* Free registered interrupts */
788 for (irq = 0; irq < nic->num_vec; irq++) {
789 if (nic->irq_allocated[irq])
790 free_irq(nic->msix_entries[irq].vector, nic);
791 nic->irq_allocated[irq] = false;
795 nicvf_disable_msix(nic);
798 /* Initialize MSIX vectors and register MISC interrupt.
799 * Send READY message to PF to check if its alive
801 static int nicvf_register_misc_interrupt(struct nicvf *nic)
804 int irq = NICVF_INTR_ID_MISC;
806 /* Return if mailbox interrupt is already registered */
807 if (nic->msix_enabled)
811 if (!nicvf_enable_msix(nic))
814 sprintf(nic->irq_name[irq], "%s Mbox", "NICVF");
815 /* Register Misc interrupt */
816 ret = request_irq(nic->msix_entries[irq].vector,
817 nicvf_misc_intr_handler, 0, nic->irq_name[irq], nic);
821 nic->irq_allocated[irq] = true;
823 /* Enable mailbox interrupt */
824 nicvf_enable_intr(nic, NICVF_INTR_MBOX, 0);
826 /* Check if VF is able to communicate with PF */
827 if (!nicvf_check_pf_ready(nic)) {
828 nicvf_disable_intr(nic, NICVF_INTR_MBOX, 0);
829 nicvf_unregister_interrupts(nic);
836 static netdev_tx_t nicvf_xmit(struct sk_buff *skb, struct net_device *netdev)
838 struct nicvf *nic = netdev_priv(netdev);
839 int qid = skb_get_queue_mapping(skb);
840 struct netdev_queue *txq = netdev_get_tx_queue(netdev, qid);
842 /* Check for minimum packet length */
843 if (skb->len <= ETH_HLEN) {
848 if (!netif_tx_queue_stopped(txq) && !nicvf_sq_append_skb(nic, skb)) {
849 netif_tx_stop_queue(txq);
850 nic->drv_stats.txq_stop++;
851 if (netif_msg_tx_err(nic))
853 "%s: Transmit ring full, stopping SQ%d\n",
856 return NETDEV_TX_BUSY;
862 int nicvf_stop(struct net_device *netdev)
865 struct nicvf *nic = netdev_priv(netdev);
866 struct queue_set *qs = nic->qs;
867 struct nicvf_cq_poll *cq_poll = NULL;
868 union nic_mbx mbx = {};
870 mbx.msg.msg = NIC_MBOX_MSG_SHUTDOWN;
871 nicvf_send_msg_to_pf(nic, &mbx);
873 netif_carrier_off(netdev);
875 /* Disable RBDR & QS error interrupts */
876 for (qidx = 0; qidx < qs->rbdr_cnt; qidx++) {
877 nicvf_disable_intr(nic, NICVF_INTR_RBDR, qidx);
878 nicvf_clear_intr(nic, NICVF_INTR_RBDR, qidx);
880 nicvf_disable_intr(nic, NICVF_INTR_QS_ERR, 0);
881 nicvf_clear_intr(nic, NICVF_INTR_QS_ERR, 0);
883 /* Wait for pending IRQ handlers to finish */
884 for (irq = 0; irq < nic->num_vec; irq++)
885 synchronize_irq(nic->msix_entries[irq].vector);
887 tasklet_kill(&nic->rbdr_task);
888 tasklet_kill(&nic->qs_err_task);
889 if (nic->rb_work_scheduled)
890 cancel_delayed_work_sync(&nic->rbdr_work);
892 for (qidx = 0; qidx < nic->qs->cq_cnt; qidx++) {
893 cq_poll = nic->napi[qidx];
896 nic->napi[qidx] = NULL;
897 napi_synchronize(&cq_poll->napi);
898 /* CQ intr is enabled while napi_complete,
901 nicvf_disable_intr(nic, NICVF_INTR_CQ, qidx);
902 nicvf_clear_intr(nic, NICVF_INTR_CQ, qidx);
903 napi_disable(&cq_poll->napi);
904 netif_napi_del(&cq_poll->napi);
908 netif_tx_disable(netdev);
911 nicvf_config_data_transfer(nic, false);
913 /* Disable HW Qset */
914 nicvf_qset_config(nic, false);
916 /* disable mailbox interrupt */
917 nicvf_disable_intr(nic, NICVF_INTR_MBOX, 0);
919 nicvf_unregister_interrupts(nic);
924 int nicvf_open(struct net_device *netdev)
927 struct nicvf *nic = netdev_priv(netdev);
928 struct queue_set *qs = nic->qs;
929 struct nicvf_cq_poll *cq_poll = NULL;
931 nic->mtu = netdev->mtu;
933 netif_carrier_off(netdev);
935 err = nicvf_register_misc_interrupt(nic);
939 /* Register NAPI handler for processing CQEs */
940 for (qidx = 0; qidx < qs->cq_cnt; qidx++) {
941 cq_poll = kzalloc(sizeof(*cq_poll), GFP_KERNEL);
946 cq_poll->cq_idx = qidx;
947 netif_napi_add(netdev, &cq_poll->napi, nicvf_poll,
949 napi_enable(&cq_poll->napi);
950 nic->napi[qidx] = cq_poll;
953 /* Check if we got MAC address from PF or else generate a radom MAC */
954 if (is_zero_ether_addr(netdev->dev_addr)) {
955 eth_hw_addr_random(netdev);
956 nicvf_hw_set_mac_addr(nic, netdev);
959 if (nic->set_mac_pending) {
960 nic->set_mac_pending = false;
961 nicvf_hw_set_mac_addr(nic, netdev);
964 /* Init tasklet for handling Qset err interrupt */
965 tasklet_init(&nic->qs_err_task, nicvf_handle_qs_err,
968 /* Init RBDR tasklet which will refill RBDR */
969 tasklet_init(&nic->rbdr_task, nicvf_rbdr_task,
971 INIT_DELAYED_WORK(&nic->rbdr_work, nicvf_rbdr_work);
973 /* Configure CPI alorithm */
974 nic->cpi_alg = cpi_alg;
975 nicvf_config_cpi(nic);
977 /* Configure receive side scaling */
980 err = nicvf_register_interrupts(nic);
984 /* Initialize the queues */
985 err = nicvf_init_resources(nic);
989 /* Make sure queue initialization is written */
992 nicvf_reg_write(nic, NIC_VF_INT, -1);
993 /* Enable Qset err interrupt */
994 nicvf_enable_intr(nic, NICVF_INTR_QS_ERR, 0);
996 /* Enable completion queue interrupt */
997 for (qidx = 0; qidx < qs->cq_cnt; qidx++)
998 nicvf_enable_intr(nic, NICVF_INTR_CQ, qidx);
1000 /* Enable RBDR threshold interrupt */
1001 for (qidx = 0; qidx < qs->rbdr_cnt; qidx++)
1002 nicvf_enable_intr(nic, NICVF_INTR_RBDR, qidx);
1004 nic->drv_stats.txq_stop = 0;
1005 nic->drv_stats.txq_wake = 0;
1007 netif_carrier_on(netdev);
1008 netif_tx_start_all_queues(netdev);
1012 nicvf_disable_intr(nic, NICVF_INTR_MBOX, 0);
1013 nicvf_unregister_interrupts(nic);
1015 for (qidx = 0; qidx < qs->cq_cnt; qidx++) {
1016 cq_poll = nic->napi[qidx];
1019 napi_disable(&cq_poll->napi);
1020 netif_napi_del(&cq_poll->napi);
1022 nic->napi[qidx] = NULL;
1027 static int nicvf_update_hw_max_frs(struct nicvf *nic, int mtu)
1029 union nic_mbx mbx = {};
1031 mbx.frs.msg = NIC_MBOX_MSG_SET_MAX_FRS;
1032 mbx.frs.max_frs = mtu;
1033 mbx.frs.vf_id = nic->vf_id;
1035 return nicvf_send_msg_to_pf(nic, &mbx);
1038 static int nicvf_change_mtu(struct net_device *netdev, int new_mtu)
1040 struct nicvf *nic = netdev_priv(netdev);
1042 if (new_mtu > NIC_HW_MAX_FRS)
1045 if (new_mtu < NIC_HW_MIN_FRS)
1048 if (nicvf_update_hw_max_frs(nic, new_mtu))
1050 netdev->mtu = new_mtu;
1056 static int nicvf_set_mac_address(struct net_device *netdev, void *p)
1058 struct sockaddr *addr = p;
1059 struct nicvf *nic = netdev_priv(netdev);
1061 if (!is_valid_ether_addr(addr->sa_data))
1062 return -EADDRNOTAVAIL;
1064 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1066 if (nic->msix_enabled) {
1067 if (nicvf_hw_set_mac_addr(nic, netdev))
1070 nic->set_mac_pending = true;
1076 void nicvf_update_lmac_stats(struct nicvf *nic)
1079 union nic_mbx mbx = {};
1082 if (!netif_running(nic->netdev))
1085 mbx.bgx_stats.msg = NIC_MBOX_MSG_BGX_STATS;
1086 mbx.bgx_stats.vf_id = nic->vf_id;
1088 mbx.bgx_stats.rx = 1;
1089 while (stat < BGX_RX_STATS_COUNT) {
1090 nic->bgx_stats_acked = 0;
1091 mbx.bgx_stats.idx = stat;
1092 nicvf_send_msg_to_pf(nic, &mbx);
1094 while ((!nic->bgx_stats_acked) && (timeout < 10)) {
1104 mbx.bgx_stats.rx = 0;
1105 while (stat < BGX_TX_STATS_COUNT) {
1106 nic->bgx_stats_acked = 0;
1107 mbx.bgx_stats.idx = stat;
1108 nicvf_send_msg_to_pf(nic, &mbx);
1110 while ((!nic->bgx_stats_acked) && (timeout < 10)) {
1118 void nicvf_update_stats(struct nicvf *nic)
1121 struct nicvf_hw_stats *stats = &nic->stats;
1122 struct nicvf_drv_stats *drv_stats = &nic->drv_stats;
1123 struct queue_set *qs = nic->qs;
1125 #define GET_RX_STATS(reg) \
1126 nicvf_reg_read(nic, NIC_VNIC_RX_STAT_0_13 | (reg << 3))
1127 #define GET_TX_STATS(reg) \
1128 nicvf_reg_read(nic, NIC_VNIC_TX_STAT_0_4 | (reg << 3))
1130 stats->rx_bytes_ok = GET_RX_STATS(RX_OCTS);
1131 stats->rx_ucast_frames_ok = GET_RX_STATS(RX_UCAST);
1132 stats->rx_bcast_frames_ok = GET_RX_STATS(RX_BCAST);
1133 stats->rx_mcast_frames_ok = GET_RX_STATS(RX_MCAST);
1134 stats->rx_fcs_errors = GET_RX_STATS(RX_FCS);
1135 stats->rx_l2_errors = GET_RX_STATS(RX_L2ERR);
1136 stats->rx_drop_red = GET_RX_STATS(RX_RED);
1137 stats->rx_drop_overrun = GET_RX_STATS(RX_ORUN);
1138 stats->rx_drop_bcast = GET_RX_STATS(RX_DRP_BCAST);
1139 stats->rx_drop_mcast = GET_RX_STATS(RX_DRP_MCAST);
1140 stats->rx_drop_l3_bcast = GET_RX_STATS(RX_DRP_L3BCAST);
1141 stats->rx_drop_l3_mcast = GET_RX_STATS(RX_DRP_L3MCAST);
1143 stats->tx_bytes_ok = GET_TX_STATS(TX_OCTS);
1144 stats->tx_ucast_frames_ok = GET_TX_STATS(TX_UCAST);
1145 stats->tx_bcast_frames_ok = GET_TX_STATS(TX_BCAST);
1146 stats->tx_mcast_frames_ok = GET_TX_STATS(TX_MCAST);
1147 stats->tx_drops = GET_TX_STATS(TX_DROP);
1149 drv_stats->rx_frames_ok = stats->rx_ucast_frames_ok +
1150 stats->rx_bcast_frames_ok +
1151 stats->rx_mcast_frames_ok;
1152 drv_stats->tx_frames_ok = stats->tx_ucast_frames_ok +
1153 stats->tx_bcast_frames_ok +
1154 stats->tx_mcast_frames_ok;
1155 drv_stats->rx_drops = stats->rx_drop_red +
1156 stats->rx_drop_overrun;
1157 drv_stats->tx_drops = stats->tx_drops;
1159 /* Update RQ and SQ stats */
1160 for (qidx = 0; qidx < qs->rq_cnt; qidx++)
1161 nicvf_update_rq_stats(nic, qidx);
1162 for (qidx = 0; qidx < qs->sq_cnt; qidx++)
1163 nicvf_update_sq_stats(nic, qidx);
1166 static struct rtnl_link_stats64 *nicvf_get_stats64(struct net_device *netdev,
1167 struct rtnl_link_stats64 *stats)
1169 struct nicvf *nic = netdev_priv(netdev);
1170 struct nicvf_hw_stats *hw_stats = &nic->stats;
1171 struct nicvf_drv_stats *drv_stats = &nic->drv_stats;
1173 nicvf_update_stats(nic);
1175 stats->rx_bytes = hw_stats->rx_bytes_ok;
1176 stats->rx_packets = drv_stats->rx_frames_ok;
1177 stats->rx_dropped = drv_stats->rx_drops;
1179 stats->tx_bytes = hw_stats->tx_bytes_ok;
1180 stats->tx_packets = drv_stats->tx_frames_ok;
1181 stats->tx_dropped = drv_stats->tx_drops;
1186 static void nicvf_tx_timeout(struct net_device *dev)
1188 struct nicvf *nic = netdev_priv(dev);
1190 if (netif_msg_tx_err(nic))
1191 netdev_warn(dev, "%s: Transmit timed out, resetting\n",
1194 schedule_work(&nic->reset_task);
1197 static void nicvf_reset_task(struct work_struct *work)
1201 nic = container_of(work, struct nicvf, reset_task);
1203 if (!netif_running(nic->netdev))
1206 nicvf_stop(nic->netdev);
1207 nicvf_open(nic->netdev);
1208 nic->netdev->trans_start = jiffies;
1211 static const struct net_device_ops nicvf_netdev_ops = {
1212 .ndo_open = nicvf_open,
1213 .ndo_stop = nicvf_stop,
1214 .ndo_start_xmit = nicvf_xmit,
1215 .ndo_change_mtu = nicvf_change_mtu,
1216 .ndo_set_mac_address = nicvf_set_mac_address,
1217 .ndo_get_stats64 = nicvf_get_stats64,
1218 .ndo_tx_timeout = nicvf_tx_timeout,
1221 static int nicvf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
1223 struct device *dev = &pdev->dev;
1224 struct net_device *netdev;
1226 struct queue_set *qs;
1229 err = pci_enable_device(pdev);
1231 dev_err(dev, "Failed to enable PCI device\n");
1235 err = pci_request_regions(pdev, DRV_NAME);
1237 dev_err(dev, "PCI request regions failed 0x%x\n", err);
1238 goto err_disable_device;
1241 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(48));
1243 dev_err(dev, "Unable to get usable DMA configuration\n");
1244 goto err_release_regions;
1247 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(48));
1249 dev_err(dev, "unable to get 48-bit DMA for consistent allocations\n");
1250 goto err_release_regions;
1253 netdev = alloc_etherdev_mqs(sizeof(struct nicvf),
1254 MAX_RCV_QUEUES_PER_QS,
1255 MAX_SND_QUEUES_PER_QS);
1258 goto err_release_regions;
1261 pci_set_drvdata(pdev, netdev);
1263 SET_NETDEV_DEV(netdev, &pdev->dev);
1265 nic = netdev_priv(netdev);
1266 nic->netdev = netdev;
1269 /* MAP VF's configuration registers */
1270 nic->reg_base = pcim_iomap(pdev, PCI_CFG_REG_BAR_NUM, 0);
1271 if (!nic->reg_base) {
1272 dev_err(dev, "Cannot map config register space, aborting\n");
1274 goto err_free_netdev;
1277 err = nicvf_set_qset_resources(nic);
1279 goto err_free_netdev;
1283 err = nicvf_set_real_num_queues(netdev, qs->sq_cnt, qs->rq_cnt);
1285 goto err_free_netdev;
1287 /* Check if PF is alive and get MAC address for this VF */
1288 err = nicvf_register_misc_interrupt(nic);
1290 goto err_free_netdev;
1292 netdev->features |= (NETIF_F_RXCSUM | NETIF_F_IP_CSUM | NETIF_F_SG |
1293 NETIF_F_TSO | NETIF_F_GRO);
1294 netdev->hw_features = netdev->features;
1296 netdev->netdev_ops = &nicvf_netdev_ops;
1297 netdev->watchdog_timeo = NICVF_TX_TIMEOUT;
1299 INIT_WORK(&nic->reset_task, nicvf_reset_task);
1301 err = register_netdev(netdev);
1303 dev_err(dev, "Failed to register netdevice\n");
1304 goto err_unregister_interrupts;
1307 nic->msg_enable = debug;
1309 nicvf_set_ethtool_ops(netdev);
1313 err_unregister_interrupts:
1314 nicvf_unregister_interrupts(nic);
1316 pci_set_drvdata(pdev, NULL);
1317 free_netdev(netdev);
1318 err_release_regions:
1319 pci_release_regions(pdev);
1321 pci_disable_device(pdev);
1325 static void nicvf_remove(struct pci_dev *pdev)
1327 struct net_device *netdev = pci_get_drvdata(pdev);
1328 struct nicvf *nic = netdev_priv(netdev);
1330 unregister_netdev(netdev);
1331 nicvf_unregister_interrupts(nic);
1332 pci_set_drvdata(pdev, NULL);
1333 free_netdev(netdev);
1334 pci_release_regions(pdev);
1335 pci_disable_device(pdev);
1338 static void nicvf_shutdown(struct pci_dev *pdev)
1343 static struct pci_driver nicvf_driver = {
1345 .id_table = nicvf_id_table,
1346 .probe = nicvf_probe,
1347 .remove = nicvf_remove,
1348 .shutdown = nicvf_shutdown,
1351 static int __init nicvf_init_module(void)
1353 pr_info("%s, ver %s\n", DRV_NAME, DRV_VERSION);
1355 return pci_register_driver(&nicvf_driver);
1358 static void __exit nicvf_cleanup_module(void)
1360 pci_unregister_driver(&nicvf_driver);
1363 module_init(nicvf_init_module);
1364 module_exit(nicvf_cleanup_module);