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/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 #include <linux/workqueue.h>
28 #include "nicvf_queues.h"
29 #include "thunder_bgx.h"
30 #include "../common/cavium_ptp.h"
32 #define DRV_NAME "nicvf"
33 #define DRV_VERSION "1.0"
35 /* NOTE: Packets bigger than 1530 are split across multiple pages and XDP needs
36 * the buffer to be contiguous. Allow XDP to be set up only if we don't exceed
37 * this value, keeping headroom for the 14 byte Ethernet header and two
38 * VLAN tags (for QinQ)
40 #define MAX_XDP_MTU (1530 - ETH_HLEN - VLAN_HLEN * 2)
42 /* Supported devices */
43 static const struct pci_device_id nicvf_id_table[] = {
44 { PCI_DEVICE_SUB(PCI_VENDOR_ID_CAVIUM,
45 PCI_DEVICE_ID_THUNDER_NIC_VF,
47 PCI_SUBSYS_DEVID_88XX_NIC_VF) },
48 { PCI_DEVICE_SUB(PCI_VENDOR_ID_CAVIUM,
49 PCI_DEVICE_ID_THUNDER_PASS1_NIC_VF,
51 PCI_SUBSYS_DEVID_88XX_PASS1_NIC_VF) },
52 { PCI_DEVICE_SUB(PCI_VENDOR_ID_CAVIUM,
53 PCI_DEVICE_ID_THUNDER_NIC_VF,
55 PCI_SUBSYS_DEVID_81XX_NIC_VF) },
56 { PCI_DEVICE_SUB(PCI_VENDOR_ID_CAVIUM,
57 PCI_DEVICE_ID_THUNDER_NIC_VF,
59 PCI_SUBSYS_DEVID_83XX_NIC_VF) },
60 { 0, } /* end of table */
63 MODULE_AUTHOR("Sunil Goutham");
64 MODULE_DESCRIPTION("Cavium Thunder NIC Virtual Function Driver");
65 MODULE_LICENSE("GPL v2");
66 MODULE_VERSION(DRV_VERSION);
67 MODULE_DEVICE_TABLE(pci, nicvf_id_table);
69 static int debug = 0x00;
70 module_param(debug, int, 0644);
71 MODULE_PARM_DESC(debug, "Debug message level bitmap");
73 static int cpi_alg = CPI_ALG_NONE;
74 module_param(cpi_alg, int, 0444);
75 MODULE_PARM_DESC(cpi_alg,
76 "PFC algorithm (0=none, 1=VLAN, 2=VLAN16, 3=IP Diffserv)");
78 static inline u8 nicvf_netdev_qidx(struct nicvf *nic, u8 qidx)
81 return qidx + ((nic->sqs_id + 1) * MAX_CMP_QUEUES_PER_QS);
86 /* The Cavium ThunderX network controller can *only* be found in SoCs
87 * containing the ThunderX ARM64 CPU implementation. All accesses to the device
88 * registers on this platform are implicitly strongly ordered with respect
89 * to memory accesses. So writeq_relaxed() and readq_relaxed() are safe to use
90 * with no memory barriers in this driver. The readq()/writeq() functions add
91 * explicit ordering operation which in this case are redundant, and only
95 /* Register read/write APIs */
96 void nicvf_reg_write(struct nicvf *nic, u64 offset, u64 val)
98 writeq_relaxed(val, nic->reg_base + offset);
101 u64 nicvf_reg_read(struct nicvf *nic, u64 offset)
103 return readq_relaxed(nic->reg_base + offset);
106 void nicvf_queue_reg_write(struct nicvf *nic, u64 offset,
109 void __iomem *addr = nic->reg_base + offset;
111 writeq_relaxed(val, addr + (qidx << NIC_Q_NUM_SHIFT));
114 u64 nicvf_queue_reg_read(struct nicvf *nic, u64 offset, u64 qidx)
116 void __iomem *addr = nic->reg_base + offset;
118 return readq_relaxed(addr + (qidx << NIC_Q_NUM_SHIFT));
121 /* VF -> PF mailbox communication */
122 static void nicvf_write_to_mbx(struct nicvf *nic, union nic_mbx *mbx)
124 u64 *msg = (u64 *)mbx;
126 nicvf_reg_write(nic, NIC_VF_PF_MAILBOX_0_1 + 0, msg[0]);
127 nicvf_reg_write(nic, NIC_VF_PF_MAILBOX_0_1 + 8, msg[1]);
130 int nicvf_send_msg_to_pf(struct nicvf *nic, union nic_mbx *mbx)
132 int timeout = NIC_MBOX_MSG_TIMEOUT;
136 mutex_lock(&nic->rx_mode_mtx);
138 nic->pf_acked = false;
139 nic->pf_nacked = false;
141 nicvf_write_to_mbx(nic, mbx);
143 /* Wait for previous message to be acked, timeout 2sec */
144 while (!nic->pf_acked) {
145 if (nic->pf_nacked) {
146 netdev_err(nic->netdev,
147 "PF NACK to mbox msg 0x%02x from VF%d\n",
148 (mbx->msg.msg & 0xFF), nic->vf_id);
157 netdev_err(nic->netdev,
158 "PF didn't ACK to mbox msg 0x%02x from VF%d\n",
159 (mbx->msg.msg & 0xFF), nic->vf_id);
164 mutex_unlock(&nic->rx_mode_mtx);
168 /* Checks if VF is able to comminicate with PF
169 * and also gets the VNIC number this VF is associated to.
171 static int nicvf_check_pf_ready(struct nicvf *nic)
173 union nic_mbx mbx = {};
175 mbx.msg.msg = NIC_MBOX_MSG_READY;
176 if (nicvf_send_msg_to_pf(nic, &mbx)) {
177 netdev_err(nic->netdev,
178 "PF didn't respond to READY msg\n");
185 static void nicvf_send_cfg_done(struct nicvf *nic)
187 union nic_mbx mbx = {};
189 mbx.msg.msg = NIC_MBOX_MSG_CFG_DONE;
190 if (nicvf_send_msg_to_pf(nic, &mbx)) {
191 netdev_err(nic->netdev,
192 "PF didn't respond to CFG DONE msg\n");
196 static void nicvf_read_bgx_stats(struct nicvf *nic, struct bgx_stats_msg *bgx)
199 nic->bgx_stats.rx_stats[bgx->idx] = bgx->stats;
201 nic->bgx_stats.tx_stats[bgx->idx] = bgx->stats;
204 static void nicvf_handle_mbx_intr(struct nicvf *nic)
206 union nic_mbx mbx = {};
211 mbx_addr = NIC_VF_PF_MAILBOX_0_1;
212 mbx_data = (u64 *)&mbx;
214 for (i = 0; i < NIC_PF_VF_MAILBOX_SIZE; i++) {
215 *mbx_data = nicvf_reg_read(nic, mbx_addr);
217 mbx_addr += sizeof(u64);
220 netdev_dbg(nic->netdev, "Mbox message: msg: 0x%x\n", mbx.msg.msg);
221 switch (mbx.msg.msg) {
222 case NIC_MBOX_MSG_READY:
223 nic->pf_acked = true;
224 nic->vf_id = mbx.nic_cfg.vf_id & 0x7F;
225 nic->tns_mode = mbx.nic_cfg.tns_mode & 0x7F;
226 nic->node = mbx.nic_cfg.node_id;
227 if (!nic->set_mac_pending)
228 ether_addr_copy(nic->netdev->dev_addr,
229 mbx.nic_cfg.mac_addr);
230 nic->sqs_mode = mbx.nic_cfg.sqs_mode;
231 nic->loopback_supported = mbx.nic_cfg.loopback_supported;
232 nic->link_up = false;
236 case NIC_MBOX_MSG_ACK:
237 nic->pf_acked = true;
239 case NIC_MBOX_MSG_NACK:
240 nic->pf_nacked = true;
242 case NIC_MBOX_MSG_RSS_SIZE:
243 nic->rss_info.rss_size = mbx.rss_size.ind_tbl_size;
244 nic->pf_acked = true;
246 case NIC_MBOX_MSG_BGX_STATS:
247 nicvf_read_bgx_stats(nic, &mbx.bgx_stats);
248 nic->pf_acked = true;
250 case NIC_MBOX_MSG_BGX_LINK_CHANGE:
251 nic->pf_acked = true;
252 if (nic->link_up != mbx.link_status.link_up) {
253 nic->link_up = mbx.link_status.link_up;
254 nic->duplex = mbx.link_status.duplex;
255 nic->speed = mbx.link_status.speed;
256 nic->mac_type = mbx.link_status.mac_type;
258 netdev_info(nic->netdev,
259 "Link is Up %d Mbps %s duplex\n",
261 nic->duplex == DUPLEX_FULL ?
263 netif_carrier_on(nic->netdev);
264 netif_tx_start_all_queues(nic->netdev);
266 netdev_info(nic->netdev, "Link is Down\n");
267 netif_carrier_off(nic->netdev);
268 netif_tx_stop_all_queues(nic->netdev);
272 case NIC_MBOX_MSG_ALLOC_SQS:
273 nic->sqs_count = mbx.sqs_alloc.qs_count;
274 nic->pf_acked = true;
276 case NIC_MBOX_MSG_SNICVF_PTR:
277 /* Primary VF: make note of secondary VF's pointer
278 * to be used while packet transmission.
280 nic->snicvf[mbx.nicvf.sqs_id] =
281 (struct nicvf *)mbx.nicvf.nicvf;
282 nic->pf_acked = true;
284 case NIC_MBOX_MSG_PNICVF_PTR:
285 /* Secondary VF/Qset: make note of primary VF's pointer
286 * to be used while packet reception, to handover packet
287 * to primary VF's netdev.
289 nic->pnicvf = (struct nicvf *)mbx.nicvf.nicvf;
290 nic->pf_acked = true;
292 case NIC_MBOX_MSG_PFC:
293 nic->pfc.autoneg = mbx.pfc.autoneg;
294 nic->pfc.fc_rx = mbx.pfc.fc_rx;
295 nic->pfc.fc_tx = mbx.pfc.fc_tx;
296 nic->pf_acked = true;
299 netdev_err(nic->netdev,
300 "Invalid message from PF, msg 0x%x\n", mbx.msg.msg);
303 nicvf_clear_intr(nic, NICVF_INTR_MBOX, 0);
306 static int nicvf_hw_set_mac_addr(struct nicvf *nic, struct net_device *netdev)
308 union nic_mbx mbx = {};
310 mbx.mac.msg = NIC_MBOX_MSG_SET_MAC;
311 mbx.mac.vf_id = nic->vf_id;
312 ether_addr_copy(mbx.mac.mac_addr, netdev->dev_addr);
314 return nicvf_send_msg_to_pf(nic, &mbx);
317 static void nicvf_config_cpi(struct nicvf *nic)
319 union nic_mbx mbx = {};
321 mbx.cpi_cfg.msg = NIC_MBOX_MSG_CPI_CFG;
322 mbx.cpi_cfg.vf_id = nic->vf_id;
323 mbx.cpi_cfg.cpi_alg = nic->cpi_alg;
324 mbx.cpi_cfg.rq_cnt = nic->qs->rq_cnt;
326 nicvf_send_msg_to_pf(nic, &mbx);
329 static void nicvf_get_rss_size(struct nicvf *nic)
331 union nic_mbx mbx = {};
333 mbx.rss_size.msg = NIC_MBOX_MSG_RSS_SIZE;
334 mbx.rss_size.vf_id = nic->vf_id;
335 nicvf_send_msg_to_pf(nic, &mbx);
338 void nicvf_config_rss(struct nicvf *nic)
340 union nic_mbx mbx = {};
341 struct nicvf_rss_info *rss = &nic->rss_info;
342 int ind_tbl_len = rss->rss_size;
345 mbx.rss_cfg.vf_id = nic->vf_id;
346 mbx.rss_cfg.hash_bits = rss->hash_bits;
347 while (ind_tbl_len) {
348 mbx.rss_cfg.tbl_offset = nextq;
349 mbx.rss_cfg.tbl_len = min(ind_tbl_len,
350 RSS_IND_TBL_LEN_PER_MBX_MSG);
351 mbx.rss_cfg.msg = mbx.rss_cfg.tbl_offset ?
352 NIC_MBOX_MSG_RSS_CFG_CONT : NIC_MBOX_MSG_RSS_CFG;
354 for (i = 0; i < mbx.rss_cfg.tbl_len; i++)
355 mbx.rss_cfg.ind_tbl[i] = rss->ind_tbl[nextq++];
357 nicvf_send_msg_to_pf(nic, &mbx);
359 ind_tbl_len -= mbx.rss_cfg.tbl_len;
363 void nicvf_set_rss_key(struct nicvf *nic)
365 struct nicvf_rss_info *rss = &nic->rss_info;
366 u64 key_addr = NIC_VNIC_RSS_KEY_0_4;
369 for (idx = 0; idx < RSS_HASH_KEY_SIZE; idx++) {
370 nicvf_reg_write(nic, key_addr, rss->key[idx]);
371 key_addr += sizeof(u64);
375 static int nicvf_rss_init(struct nicvf *nic)
377 struct nicvf_rss_info *rss = &nic->rss_info;
380 nicvf_get_rss_size(nic);
382 if (cpi_alg != CPI_ALG_NONE) {
390 netdev_rss_key_fill(rss->key, RSS_HASH_KEY_SIZE * sizeof(u64));
391 nicvf_set_rss_key(nic);
393 rss->cfg = RSS_IP_HASH_ENA | RSS_TCP_HASH_ENA | RSS_UDP_HASH_ENA;
394 nicvf_reg_write(nic, NIC_VNIC_RSS_CFG, rss->cfg);
396 rss->hash_bits = ilog2(rounddown_pow_of_two(rss->rss_size));
398 for (idx = 0; idx < rss->rss_size; idx++)
399 rss->ind_tbl[idx] = ethtool_rxfh_indir_default(idx,
401 nicvf_config_rss(nic);
405 /* Request PF to allocate additional Qsets */
406 static void nicvf_request_sqs(struct nicvf *nic)
408 union nic_mbx mbx = {};
410 int sqs_count = nic->sqs_count;
411 int rx_queues = 0, tx_queues = 0;
413 /* Only primary VF should request */
414 if (nic->sqs_mode || !nic->sqs_count)
417 mbx.sqs_alloc.msg = NIC_MBOX_MSG_ALLOC_SQS;
418 mbx.sqs_alloc.vf_id = nic->vf_id;
419 mbx.sqs_alloc.qs_count = nic->sqs_count;
420 if (nicvf_send_msg_to_pf(nic, &mbx)) {
421 /* No response from PF */
426 /* Return if no Secondary Qsets available */
430 if (nic->rx_queues > MAX_RCV_QUEUES_PER_QS)
431 rx_queues = nic->rx_queues - MAX_RCV_QUEUES_PER_QS;
433 tx_queues = nic->tx_queues + nic->xdp_tx_queues;
434 if (tx_queues > MAX_SND_QUEUES_PER_QS)
435 tx_queues = tx_queues - MAX_SND_QUEUES_PER_QS;
437 /* Set no of Rx/Tx queues in each of the SQsets */
438 for (sqs = 0; sqs < nic->sqs_count; sqs++) {
439 mbx.nicvf.msg = NIC_MBOX_MSG_SNICVF_PTR;
440 mbx.nicvf.vf_id = nic->vf_id;
441 mbx.nicvf.sqs_id = sqs;
442 nicvf_send_msg_to_pf(nic, &mbx);
444 nic->snicvf[sqs]->sqs_id = sqs;
445 if (rx_queues > MAX_RCV_QUEUES_PER_QS) {
446 nic->snicvf[sqs]->qs->rq_cnt = MAX_RCV_QUEUES_PER_QS;
447 rx_queues -= MAX_RCV_QUEUES_PER_QS;
449 nic->snicvf[sqs]->qs->rq_cnt = rx_queues;
453 if (tx_queues > MAX_SND_QUEUES_PER_QS) {
454 nic->snicvf[sqs]->qs->sq_cnt = MAX_SND_QUEUES_PER_QS;
455 tx_queues -= MAX_SND_QUEUES_PER_QS;
457 nic->snicvf[sqs]->qs->sq_cnt = tx_queues;
461 nic->snicvf[sqs]->qs->cq_cnt =
462 max(nic->snicvf[sqs]->qs->rq_cnt, nic->snicvf[sqs]->qs->sq_cnt);
464 /* Initialize secondary Qset's queues and its interrupts */
465 nicvf_open(nic->snicvf[sqs]->netdev);
468 /* Update stack with actual Rx/Tx queue count allocated */
469 if (sqs_count != nic->sqs_count)
470 nicvf_set_real_num_queues(nic->netdev,
471 nic->tx_queues, nic->rx_queues);
474 /* Send this Qset's nicvf pointer to PF.
475 * PF inturn sends primary VF's nicvf struct to secondary Qsets/VFs
476 * so that packets received by these Qsets can use primary VF's netdev
478 static void nicvf_send_vf_struct(struct nicvf *nic)
480 union nic_mbx mbx = {};
482 mbx.nicvf.msg = NIC_MBOX_MSG_NICVF_PTR;
483 mbx.nicvf.sqs_mode = nic->sqs_mode;
484 mbx.nicvf.nicvf = (u64)nic;
485 nicvf_send_msg_to_pf(nic, &mbx);
488 static void nicvf_get_primary_vf_struct(struct nicvf *nic)
490 union nic_mbx mbx = {};
492 mbx.nicvf.msg = NIC_MBOX_MSG_PNICVF_PTR;
493 nicvf_send_msg_to_pf(nic, &mbx);
496 int nicvf_set_real_num_queues(struct net_device *netdev,
497 int tx_queues, int rx_queues)
501 err = netif_set_real_num_tx_queues(netdev, tx_queues);
504 "Failed to set no of Tx queues: %d\n", tx_queues);
508 err = netif_set_real_num_rx_queues(netdev, rx_queues);
511 "Failed to set no of Rx queues: %d\n", rx_queues);
515 static int nicvf_init_resources(struct nicvf *nic)
520 nicvf_qset_config(nic, true);
522 /* Initialize queues and HW for data transfer */
523 err = nicvf_config_data_transfer(nic, true);
525 netdev_err(nic->netdev,
526 "Failed to alloc/config VF's QSet resources\n");
533 static inline bool nicvf_xdp_rx(struct nicvf *nic, struct bpf_prog *prog,
534 struct cqe_rx_t *cqe_rx, struct snd_queue *sq,
535 struct rcv_queue *rq, struct sk_buff **skb)
541 u64 dma_addr, cpu_addr;
544 /* Retrieve packet buffer's DMA address and length */
545 len = *((u16 *)((void *)cqe_rx + (3 * sizeof(u64))));
546 dma_addr = *((u64 *)((void *)cqe_rx + (7 * sizeof(u64))));
548 cpu_addr = nicvf_iova_to_phys(nic, dma_addr);
551 cpu_addr = (u64)phys_to_virt(cpu_addr);
552 page = virt_to_page((void *)cpu_addr);
554 xdp.data_hard_start = page_address(page);
555 xdp.data = (void *)cpu_addr;
556 xdp_set_data_meta_invalid(&xdp);
557 xdp.data_end = xdp.data + len;
558 xdp.rxq = &rq->xdp_rxq;
559 orig_data = xdp.data;
562 action = bpf_prog_run_xdp(prog, &xdp);
565 len = xdp.data_end - xdp.data;
566 /* Check if XDP program has changed headers */
567 if (orig_data != xdp.data) {
568 offset = orig_data - xdp.data;
574 /* Check if it's a recycled page, if not
575 * unmap the DMA mapping.
577 * Recycled page holds an extra reference.
579 if (page_ref_count(page) == 1) {
580 dma_addr &= PAGE_MASK;
581 dma_unmap_page_attrs(&nic->pdev->dev, dma_addr,
582 RCV_FRAG_LEN + XDP_PACKET_HEADROOM,
584 DMA_ATTR_SKIP_CPU_SYNC);
587 /* Build SKB and pass on packet to network stack */
588 *skb = build_skb(xdp.data,
589 RCV_FRAG_LEN - cqe_rx->align_pad + offset);
596 nicvf_xdp_sq_append_pkt(nic, sq, (u64)xdp.data, dma_addr, len);
599 bpf_warn_invalid_xdp_action(action);
602 trace_xdp_exception(nic->netdev, prog, action);
605 /* Check if it's a recycled page, if not
606 * unmap the DMA mapping.
608 * Recycled page holds an extra reference.
610 if (page_ref_count(page) == 1) {
611 dma_addr &= PAGE_MASK;
612 dma_unmap_page_attrs(&nic->pdev->dev, dma_addr,
613 RCV_FRAG_LEN + XDP_PACKET_HEADROOM,
615 DMA_ATTR_SKIP_CPU_SYNC);
623 static void nicvf_snd_ptp_handler(struct net_device *netdev,
624 struct cqe_send_t *cqe_tx)
626 struct nicvf *nic = netdev_priv(netdev);
627 struct skb_shared_hwtstamps ts;
632 /* Sync for 'ptp_skb' */
635 /* New timestamp request can be queued now */
636 atomic_set(&nic->tx_ptp_skbs, 0);
638 /* Check for timestamp requested skb */
642 /* Check if timestamping is timedout, which is set to 10us */
643 if (cqe_tx->send_status == CQ_TX_ERROP_TSTMP_TIMEOUT ||
644 cqe_tx->send_status == CQ_TX_ERROP_TSTMP_CONFLICT)
647 /* Get the timestamp */
648 memset(&ts, 0, sizeof(ts));
649 ns = cavium_ptp_tstamp2time(nic->ptp_clock, cqe_tx->ptp_timestamp);
650 ts.hwtstamp = ns_to_ktime(ns);
651 skb_tstamp_tx(nic->ptp_skb, &ts);
654 /* Free the original skb */
655 dev_kfree_skb_any(nic->ptp_skb);
661 static void nicvf_snd_pkt_handler(struct net_device *netdev,
662 struct cqe_send_t *cqe_tx,
663 int budget, int *subdesc_cnt,
664 unsigned int *tx_pkts, unsigned int *tx_bytes)
666 struct sk_buff *skb = NULL;
668 struct nicvf *nic = netdev_priv(netdev);
669 struct snd_queue *sq;
670 struct sq_hdr_subdesc *hdr;
671 struct sq_hdr_subdesc *tso_sqe;
673 sq = &nic->qs->sq[cqe_tx->sq_idx];
675 hdr = (struct sq_hdr_subdesc *)GET_SQ_DESC(sq, cqe_tx->sqe_ptr);
676 if (hdr->subdesc_type != SQ_DESC_TYPE_HEADER)
679 /* Check for errors */
680 if (cqe_tx->send_status)
681 nicvf_check_cqe_tx_errs(nic->pnicvf, cqe_tx);
683 /* Is this a XDP designated Tx queue */
685 page = (struct page *)sq->xdp_page[cqe_tx->sqe_ptr];
686 /* Check if it's recycled page or else unmap DMA mapping */
687 if (page && (page_ref_count(page) == 1))
688 nicvf_unmap_sndq_buffers(nic, sq, cqe_tx->sqe_ptr,
691 /* Release page reference for recycling */
694 sq->xdp_page[cqe_tx->sqe_ptr] = (u64)NULL;
695 *subdesc_cnt += hdr->subdesc_cnt + 1;
699 skb = (struct sk_buff *)sq->skbuff[cqe_tx->sqe_ptr];
701 /* Check for dummy descriptor used for HW TSO offload on 88xx */
702 if (hdr->dont_send) {
703 /* Get actual TSO descriptors and free them */
705 (struct sq_hdr_subdesc *)GET_SQ_DESC(sq, hdr->rsvd2);
706 nicvf_unmap_sndq_buffers(nic, sq, hdr->rsvd2,
707 tso_sqe->subdesc_cnt);
708 *subdesc_cnt += tso_sqe->subdesc_cnt + 1;
710 nicvf_unmap_sndq_buffers(nic, sq, cqe_tx->sqe_ptr,
713 *subdesc_cnt += hdr->subdesc_cnt + 1;
716 *tx_bytes += skb->len;
717 /* If timestamp is requested for this skb, don't free it */
718 if (skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS &&
719 !nic->pnicvf->ptp_skb)
720 nic->pnicvf->ptp_skb = skb;
722 napi_consume_skb(skb, budget);
723 sq->skbuff[cqe_tx->sqe_ptr] = (u64)NULL;
725 /* In case of SW TSO on 88xx, only last segment will have
726 * a SKB attached, so just free SQEs here.
729 *subdesc_cnt += hdr->subdesc_cnt + 1;
733 static inline void nicvf_set_rxhash(struct net_device *netdev,
734 struct cqe_rx_t *cqe_rx,
740 if (!(netdev->features & NETIF_F_RXHASH))
743 switch (cqe_rx->rss_alg) {
746 hash_type = PKT_HASH_TYPE_L4;
747 hash = cqe_rx->rss_tag;
750 hash_type = PKT_HASH_TYPE_L3;
751 hash = cqe_rx->rss_tag;
754 hash_type = PKT_HASH_TYPE_NONE;
758 skb_set_hash(skb, hash, hash_type);
761 static inline void nicvf_set_rxtstamp(struct nicvf *nic, struct sk_buff *skb)
765 if (!nic->ptp_clock || !nic->hw_rx_tstamp)
768 /* The first 8 bytes is the timestamp */
769 ns = cavium_ptp_tstamp2time(nic->ptp_clock,
770 be64_to_cpu(*(__be64 *)skb->data));
771 skb_hwtstamps(skb)->hwtstamp = ns_to_ktime(ns);
776 static void nicvf_rcv_pkt_handler(struct net_device *netdev,
777 struct napi_struct *napi,
778 struct cqe_rx_t *cqe_rx,
779 struct snd_queue *sq, struct rcv_queue *rq)
781 struct sk_buff *skb = NULL;
782 struct nicvf *nic = netdev_priv(netdev);
783 struct nicvf *snic = nic;
787 rq_idx = nicvf_netdev_qidx(nic, cqe_rx->rq_idx);
790 /* Use primary VF's 'nicvf' struct */
792 netdev = nic->netdev;
795 /* Check for errors */
796 if (cqe_rx->err_level || cqe_rx->err_opcode) {
797 err = nicvf_check_cqe_rx_errs(nic, cqe_rx);
798 if (err && !cqe_rx->rb_cnt)
802 /* For XDP, ignore pkts spanning multiple pages */
803 if (nic->xdp_prog && (cqe_rx->rb_cnt == 1)) {
804 /* Packet consumed by XDP */
805 if (nicvf_xdp_rx(snic, nic->xdp_prog, cqe_rx, sq, rq, &skb))
808 skb = nicvf_get_rcv_skb(snic, cqe_rx,
809 nic->xdp_prog ? true : false);
815 if (netif_msg_pktdata(nic)) {
816 netdev_info(nic->netdev, "skb 0x%p, len=%d\n", skb, skb->len);
817 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_OFFSET, 16, 1,
818 skb->data, skb->len, true);
821 /* If error packet, drop it here */
823 dev_kfree_skb_any(skb);
827 nicvf_set_rxtstamp(nic, skb);
828 nicvf_set_rxhash(netdev, cqe_rx, skb);
830 skb_record_rx_queue(skb, rq_idx);
831 if (netdev->hw_features & NETIF_F_RXCSUM) {
832 /* HW by default verifies TCP/UDP/SCTP checksums */
833 skb->ip_summed = CHECKSUM_UNNECESSARY;
835 skb_checksum_none_assert(skb);
838 skb->protocol = eth_type_trans(skb, netdev);
840 /* Check for stripped VLAN */
841 if (cqe_rx->vlan_found && cqe_rx->vlan_stripped)
842 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
843 ntohs((__force __be16)cqe_rx->vlan_tci));
845 if (napi && (netdev->features & NETIF_F_GRO))
846 napi_gro_receive(napi, skb);
848 netif_receive_skb(skb);
851 static int nicvf_cq_intr_handler(struct net_device *netdev, u8 cq_idx,
852 struct napi_struct *napi, int budget)
854 int processed_cqe, work_done = 0, tx_done = 0;
855 int cqe_count, cqe_head;
857 struct nicvf *nic = netdev_priv(netdev);
858 struct queue_set *qs = nic->qs;
859 struct cmp_queue *cq = &qs->cq[cq_idx];
860 struct cqe_rx_t *cq_desc;
861 struct netdev_queue *txq;
862 struct snd_queue *sq = &qs->sq[cq_idx];
863 struct rcv_queue *rq = &qs->rq[cq_idx];
864 unsigned int tx_pkts = 0, tx_bytes = 0, txq_idx;
866 spin_lock_bh(&cq->lock);
869 /* Get no of valid CQ entries to process */
870 cqe_count = nicvf_queue_reg_read(nic, NIC_QSET_CQ_0_7_STATUS, cq_idx);
871 cqe_count &= CQ_CQE_COUNT;
875 /* Get head of the valid CQ entries */
876 cqe_head = nicvf_queue_reg_read(nic, NIC_QSET_CQ_0_7_HEAD, cq_idx) >> 9;
879 while (processed_cqe < cqe_count) {
880 /* Get the CQ descriptor */
881 cq_desc = (struct cqe_rx_t *)GET_CQ_DESC(cq, cqe_head);
883 cqe_head &= (cq->dmem.q_len - 1);
884 /* Initiate prefetch for next descriptor */
885 prefetch((struct cqe_rx_t *)GET_CQ_DESC(cq, cqe_head));
887 if ((work_done >= budget) && napi &&
888 (cq_desc->cqe_type != CQE_TYPE_SEND)) {
892 switch (cq_desc->cqe_type) {
894 nicvf_rcv_pkt_handler(netdev, napi, cq_desc, sq, rq);
898 nicvf_snd_pkt_handler(netdev, (void *)cq_desc,
899 budget, &subdesc_cnt,
900 &tx_pkts, &tx_bytes);
903 case CQE_TYPE_SEND_PTP:
904 nicvf_snd_ptp_handler(netdev, (void *)cq_desc);
906 case CQE_TYPE_INVALID:
907 case CQE_TYPE_RX_SPLIT:
908 case CQE_TYPE_RX_TCP:
915 /* Ring doorbell to inform H/W to reuse processed CQEs */
916 nicvf_queue_reg_write(nic, NIC_QSET_CQ_0_7_DOOR,
917 cq_idx, processed_cqe);
919 if ((work_done < budget) && napi)
923 /* Update SQ's descriptor free count */
925 nicvf_put_sq_desc(sq, subdesc_cnt);
927 txq_idx = nicvf_netdev_qidx(nic, cq_idx);
928 /* Handle XDP TX queues */
929 if (nic->pnicvf->xdp_prog) {
930 if (txq_idx < nic->pnicvf->xdp_tx_queues) {
931 nicvf_xdp_sq_doorbell(nic, sq, cq_idx);
935 txq_idx -= nic->pnicvf->xdp_tx_queues;
938 /* Wakeup TXQ if its stopped earlier due to SQ full */
940 (atomic_read(&sq->free_cnt) >= MIN_SQ_DESC_PER_PKT_XMIT)) {
941 netdev = nic->pnicvf->netdev;
942 txq = netdev_get_tx_queue(netdev, txq_idx);
944 netdev_tx_completed_queue(txq, tx_pkts, tx_bytes);
946 /* To read updated queue and carrier status */
948 if (netif_tx_queue_stopped(txq) && netif_carrier_ok(netdev)) {
949 netif_tx_wake_queue(txq);
951 this_cpu_inc(nic->drv_stats->txq_wake);
952 netif_warn(nic, tx_err, netdev,
953 "Transmit queue wakeup SQ%d\n", txq_idx);
958 spin_unlock_bh(&cq->lock);
962 static int nicvf_poll(struct napi_struct *napi, int budget)
966 struct net_device *netdev = napi->dev;
967 struct nicvf *nic = netdev_priv(netdev);
968 struct nicvf_cq_poll *cq;
970 cq = container_of(napi, struct nicvf_cq_poll, napi);
971 work_done = nicvf_cq_intr_handler(netdev, cq->cq_idx, napi, budget);
973 if (work_done < budget) {
974 /* Slow packet rate, exit polling */
975 napi_complete_done(napi, work_done);
976 /* Re-enable interrupts */
977 cq_head = nicvf_queue_reg_read(nic, NIC_QSET_CQ_0_7_HEAD,
979 nicvf_clear_intr(nic, NICVF_INTR_CQ, cq->cq_idx);
980 nicvf_queue_reg_write(nic, NIC_QSET_CQ_0_7_HEAD,
981 cq->cq_idx, cq_head);
982 nicvf_enable_intr(nic, NICVF_INTR_CQ, cq->cq_idx);
987 /* Qset error interrupt handler
989 * As of now only CQ errors are handled
991 static void nicvf_handle_qs_err(unsigned long data)
993 struct nicvf *nic = (struct nicvf *)data;
994 struct queue_set *qs = nic->qs;
998 netif_tx_disable(nic->netdev);
1000 /* Check if it is CQ err */
1001 for (qidx = 0; qidx < qs->cq_cnt; qidx++) {
1002 status = nicvf_queue_reg_read(nic, NIC_QSET_CQ_0_7_STATUS,
1004 if (!(status & CQ_ERR_MASK))
1006 /* Process already queued CQEs and reconfig CQ */
1007 nicvf_disable_intr(nic, NICVF_INTR_CQ, qidx);
1008 nicvf_sq_disable(nic, qidx);
1009 nicvf_cq_intr_handler(nic->netdev, qidx, NULL, 0);
1010 nicvf_cmp_queue_config(nic, qs, qidx, true);
1011 nicvf_sq_free_used_descs(nic->netdev, &qs->sq[qidx], qidx);
1012 nicvf_sq_enable(nic, &qs->sq[qidx], qidx);
1014 nicvf_enable_intr(nic, NICVF_INTR_CQ, qidx);
1017 netif_tx_start_all_queues(nic->netdev);
1018 /* Re-enable Qset error interrupt */
1019 nicvf_enable_intr(nic, NICVF_INTR_QS_ERR, 0);
1022 static void nicvf_dump_intr_status(struct nicvf *nic)
1024 netif_info(nic, intr, nic->netdev, "interrupt status 0x%llx\n",
1025 nicvf_reg_read(nic, NIC_VF_INT));
1028 static irqreturn_t nicvf_misc_intr_handler(int irq, void *nicvf_irq)
1030 struct nicvf *nic = (struct nicvf *)nicvf_irq;
1033 nicvf_dump_intr_status(nic);
1035 intr = nicvf_reg_read(nic, NIC_VF_INT);
1036 /* Check for spurious interrupt */
1037 if (!(intr & NICVF_INTR_MBOX_MASK))
1040 nicvf_handle_mbx_intr(nic);
1045 static irqreturn_t nicvf_intr_handler(int irq, void *cq_irq)
1047 struct nicvf_cq_poll *cq_poll = (struct nicvf_cq_poll *)cq_irq;
1048 struct nicvf *nic = cq_poll->nicvf;
1049 int qidx = cq_poll->cq_idx;
1051 nicvf_dump_intr_status(nic);
1053 /* Disable interrupts */
1054 nicvf_disable_intr(nic, NICVF_INTR_CQ, qidx);
1057 napi_schedule_irqoff(&cq_poll->napi);
1059 /* Clear interrupt */
1060 nicvf_clear_intr(nic, NICVF_INTR_CQ, qidx);
1065 static irqreturn_t nicvf_rbdr_intr_handler(int irq, void *nicvf_irq)
1067 struct nicvf *nic = (struct nicvf *)nicvf_irq;
1071 nicvf_dump_intr_status(nic);
1073 /* Disable RBDR interrupt and schedule softirq */
1074 for (qidx = 0; qidx < nic->qs->rbdr_cnt; qidx++) {
1075 if (!nicvf_is_intr_enabled(nic, NICVF_INTR_RBDR, qidx))
1077 nicvf_disable_intr(nic, NICVF_INTR_RBDR, qidx);
1078 tasklet_hi_schedule(&nic->rbdr_task);
1079 /* Clear interrupt */
1080 nicvf_clear_intr(nic, NICVF_INTR_RBDR, qidx);
1086 static irqreturn_t nicvf_qs_err_intr_handler(int irq, void *nicvf_irq)
1088 struct nicvf *nic = (struct nicvf *)nicvf_irq;
1090 nicvf_dump_intr_status(nic);
1092 /* Disable Qset err interrupt and schedule softirq */
1093 nicvf_disable_intr(nic, NICVF_INTR_QS_ERR, 0);
1094 tasklet_hi_schedule(&nic->qs_err_task);
1095 nicvf_clear_intr(nic, NICVF_INTR_QS_ERR, 0);
1100 static void nicvf_set_irq_affinity(struct nicvf *nic)
1104 for (vec = 0; vec < nic->num_vec; vec++) {
1105 if (!nic->irq_allocated[vec])
1108 if (!zalloc_cpumask_var(&nic->affinity_mask[vec], GFP_KERNEL))
1111 if (vec < NICVF_INTR_ID_SQ)
1112 /* Leave CPU0 for RBDR and other interrupts */
1113 cpu = nicvf_netdev_qidx(nic, vec) + 1;
1117 cpumask_set_cpu(cpumask_local_spread(cpu, nic->node),
1118 nic->affinity_mask[vec]);
1119 irq_set_affinity_hint(pci_irq_vector(nic->pdev, vec),
1120 nic->affinity_mask[vec]);
1124 static int nicvf_register_interrupts(struct nicvf *nic)
1128 for_each_cq_irq(irq)
1129 sprintf(nic->irq_name[irq], "%s-rxtx-%d",
1130 nic->pnicvf->netdev->name,
1131 nicvf_netdev_qidx(nic, irq));
1133 for_each_sq_irq(irq)
1134 sprintf(nic->irq_name[irq], "%s-sq-%d",
1135 nic->pnicvf->netdev->name,
1136 nicvf_netdev_qidx(nic, irq - NICVF_INTR_ID_SQ));
1138 for_each_rbdr_irq(irq)
1139 sprintf(nic->irq_name[irq], "%s-rbdr-%d",
1140 nic->pnicvf->netdev->name,
1141 nic->sqs_mode ? (nic->sqs_id + 1) : 0);
1143 /* Register CQ interrupts */
1144 for (irq = 0; irq < nic->qs->cq_cnt; irq++) {
1145 ret = request_irq(pci_irq_vector(nic->pdev, irq),
1147 0, nic->irq_name[irq], nic->napi[irq]);
1150 nic->irq_allocated[irq] = true;
1153 /* Register RBDR interrupt */
1154 for (irq = NICVF_INTR_ID_RBDR;
1155 irq < (NICVF_INTR_ID_RBDR + nic->qs->rbdr_cnt); irq++) {
1156 ret = request_irq(pci_irq_vector(nic->pdev, irq),
1157 nicvf_rbdr_intr_handler,
1158 0, nic->irq_name[irq], nic);
1161 nic->irq_allocated[irq] = true;
1164 /* Register QS error interrupt */
1165 sprintf(nic->irq_name[NICVF_INTR_ID_QS_ERR], "%s-qset-err-%d",
1166 nic->pnicvf->netdev->name,
1167 nic->sqs_mode ? (nic->sqs_id + 1) : 0);
1168 irq = NICVF_INTR_ID_QS_ERR;
1169 ret = request_irq(pci_irq_vector(nic->pdev, irq),
1170 nicvf_qs_err_intr_handler,
1171 0, nic->irq_name[irq], nic);
1175 nic->irq_allocated[irq] = true;
1177 /* Set IRQ affinities */
1178 nicvf_set_irq_affinity(nic);
1182 netdev_err(nic->netdev, "request_irq failed, vector %d\n", irq);
1187 static void nicvf_unregister_interrupts(struct nicvf *nic)
1189 struct pci_dev *pdev = nic->pdev;
1192 /* Free registered interrupts */
1193 for (irq = 0; irq < nic->num_vec; irq++) {
1194 if (!nic->irq_allocated[irq])
1197 irq_set_affinity_hint(pci_irq_vector(pdev, irq), NULL);
1198 free_cpumask_var(nic->affinity_mask[irq]);
1200 if (irq < NICVF_INTR_ID_SQ)
1201 free_irq(pci_irq_vector(pdev, irq), nic->napi[irq]);
1203 free_irq(pci_irq_vector(pdev, irq), nic);
1205 nic->irq_allocated[irq] = false;
1209 pci_free_irq_vectors(pdev);
1213 /* Initialize MSIX vectors and register MISC interrupt.
1214 * Send READY message to PF to check if its alive
1216 static int nicvf_register_misc_interrupt(struct nicvf *nic)
1219 int irq = NICVF_INTR_ID_MISC;
1221 /* Return if mailbox interrupt is already registered */
1222 if (nic->pdev->msix_enabled)
1226 nic->num_vec = pci_msix_vec_count(nic->pdev);
1227 ret = pci_alloc_irq_vectors(nic->pdev, nic->num_vec, nic->num_vec,
1230 netdev_err(nic->netdev,
1231 "Req for #%d msix vectors failed\n", nic->num_vec);
1235 sprintf(nic->irq_name[irq], "%s Mbox", "NICVF");
1236 /* Register Misc interrupt */
1237 ret = request_irq(pci_irq_vector(nic->pdev, irq),
1238 nicvf_misc_intr_handler, 0, nic->irq_name[irq], nic);
1242 nic->irq_allocated[irq] = true;
1244 /* Enable mailbox interrupt */
1245 nicvf_enable_intr(nic, NICVF_INTR_MBOX, 0);
1247 /* Check if VF is able to communicate with PF */
1248 if (!nicvf_check_pf_ready(nic)) {
1249 nicvf_disable_intr(nic, NICVF_INTR_MBOX, 0);
1250 nicvf_unregister_interrupts(nic);
1257 static netdev_tx_t nicvf_xmit(struct sk_buff *skb, struct net_device *netdev)
1259 struct nicvf *nic = netdev_priv(netdev);
1260 int qid = skb_get_queue_mapping(skb);
1261 struct netdev_queue *txq = netdev_get_tx_queue(netdev, qid);
1263 struct snd_queue *sq;
1266 /* Check for minimum packet length */
1267 if (skb->len <= ETH_HLEN) {
1269 return NETDEV_TX_OK;
1272 /* In XDP case, initial HW tx queues are used for XDP,
1273 * but stack's queue mapping starts at '0', so skip the
1274 * Tx queues attached to Rx queues for XDP.
1277 qid += nic->xdp_tx_queues;
1280 /* Get secondary Qset's SQ structure */
1281 if (qid >= MAX_SND_QUEUES_PER_QS) {
1282 tmp = qid / MAX_SND_QUEUES_PER_QS;
1283 snic = (struct nicvf *)nic->snicvf[tmp - 1];
1285 netdev_warn(nic->netdev,
1286 "Secondary Qset#%d's ptr not initialized\n",
1289 return NETDEV_TX_OK;
1291 qid = qid % MAX_SND_QUEUES_PER_QS;
1294 sq = &snic->qs->sq[qid];
1295 if (!netif_tx_queue_stopped(txq) &&
1296 !nicvf_sq_append_skb(snic, sq, skb, qid)) {
1297 netif_tx_stop_queue(txq);
1299 /* Barrier, so that stop_queue visible to other cpus */
1302 /* Check again, incase another cpu freed descriptors */
1303 if (atomic_read(&sq->free_cnt) > MIN_SQ_DESC_PER_PKT_XMIT) {
1304 netif_tx_wake_queue(txq);
1306 this_cpu_inc(nic->drv_stats->txq_stop);
1307 netif_warn(nic, tx_err, netdev,
1308 "Transmit ring full, stopping SQ%d\n", qid);
1310 return NETDEV_TX_BUSY;
1313 return NETDEV_TX_OK;
1316 static inline void nicvf_free_cq_poll(struct nicvf *nic)
1318 struct nicvf_cq_poll *cq_poll;
1321 for (qidx = 0; qidx < nic->qs->cq_cnt; qidx++) {
1322 cq_poll = nic->napi[qidx];
1325 nic->napi[qidx] = NULL;
1330 int nicvf_stop(struct net_device *netdev)
1333 struct nicvf *nic = netdev_priv(netdev);
1334 struct queue_set *qs = nic->qs;
1335 struct nicvf_cq_poll *cq_poll = NULL;
1336 union nic_mbx mbx = {};
1338 /* wait till all queued set_rx_mode tasks completes */
1339 if (nic->nicvf_rx_mode_wq) {
1340 cancel_delayed_work_sync(&nic->link_change_work);
1341 drain_workqueue(nic->nicvf_rx_mode_wq);
1344 mbx.msg.msg = NIC_MBOX_MSG_SHUTDOWN;
1345 nicvf_send_msg_to_pf(nic, &mbx);
1347 netif_carrier_off(netdev);
1348 netif_tx_stop_all_queues(nic->netdev);
1349 nic->link_up = false;
1351 /* Teardown secondary qsets first */
1352 if (!nic->sqs_mode) {
1353 for (qidx = 0; qidx < nic->sqs_count; qidx++) {
1354 if (!nic->snicvf[qidx])
1356 nicvf_stop(nic->snicvf[qidx]->netdev);
1357 nic->snicvf[qidx] = NULL;
1361 /* Disable RBDR & QS error interrupts */
1362 for (qidx = 0; qidx < qs->rbdr_cnt; qidx++) {
1363 nicvf_disable_intr(nic, NICVF_INTR_RBDR, qidx);
1364 nicvf_clear_intr(nic, NICVF_INTR_RBDR, qidx);
1366 nicvf_disable_intr(nic, NICVF_INTR_QS_ERR, 0);
1367 nicvf_clear_intr(nic, NICVF_INTR_QS_ERR, 0);
1369 /* Wait for pending IRQ handlers to finish */
1370 for (irq = 0; irq < nic->num_vec; irq++)
1371 synchronize_irq(pci_irq_vector(nic->pdev, irq));
1373 tasklet_kill(&nic->rbdr_task);
1374 tasklet_kill(&nic->qs_err_task);
1375 if (nic->rb_work_scheduled)
1376 cancel_delayed_work_sync(&nic->rbdr_work);
1378 for (qidx = 0; qidx < nic->qs->cq_cnt; qidx++) {
1379 cq_poll = nic->napi[qidx];
1382 napi_synchronize(&cq_poll->napi);
1383 /* CQ intr is enabled while napi_complete,
1386 nicvf_disable_intr(nic, NICVF_INTR_CQ, qidx);
1387 nicvf_clear_intr(nic, NICVF_INTR_CQ, qidx);
1388 napi_disable(&cq_poll->napi);
1389 netif_napi_del(&cq_poll->napi);
1392 netif_tx_disable(netdev);
1394 for (qidx = 0; qidx < netdev->num_tx_queues; qidx++)
1395 netdev_tx_reset_queue(netdev_get_tx_queue(netdev, qidx));
1397 /* Free resources */
1398 nicvf_config_data_transfer(nic, false);
1400 /* Disable HW Qset */
1401 nicvf_qset_config(nic, false);
1403 /* disable mailbox interrupt */
1404 nicvf_disable_intr(nic, NICVF_INTR_MBOX, 0);
1406 nicvf_unregister_interrupts(nic);
1408 nicvf_free_cq_poll(nic);
1410 /* Free any pending SKB saved to receive timestamp */
1412 dev_kfree_skb_any(nic->ptp_skb);
1413 nic->ptp_skb = NULL;
1416 /* Clear multiqset info */
1422 static int nicvf_config_hw_rx_tstamp(struct nicvf *nic, bool enable)
1424 union nic_mbx mbx = {};
1426 mbx.ptp.msg = NIC_MBOX_MSG_PTP_CFG;
1427 mbx.ptp.enable = enable;
1429 return nicvf_send_msg_to_pf(nic, &mbx);
1432 static int nicvf_update_hw_max_frs(struct nicvf *nic, int mtu)
1434 union nic_mbx mbx = {};
1436 mbx.frs.msg = NIC_MBOX_MSG_SET_MAX_FRS;
1437 mbx.frs.max_frs = mtu;
1438 mbx.frs.vf_id = nic->vf_id;
1440 return nicvf_send_msg_to_pf(nic, &mbx);
1443 static void nicvf_link_status_check_task(struct work_struct *work_arg)
1445 struct nicvf *nic = container_of(work_arg,
1447 link_change_work.work);
1448 union nic_mbx mbx = {};
1449 mbx.msg.msg = NIC_MBOX_MSG_BGX_LINK_CHANGE;
1450 nicvf_send_msg_to_pf(nic, &mbx);
1451 queue_delayed_work(nic->nicvf_rx_mode_wq,
1452 &nic->link_change_work, 2 * HZ);
1455 int nicvf_open(struct net_device *netdev)
1458 struct nicvf *nic = netdev_priv(netdev);
1459 struct queue_set *qs = nic->qs;
1460 struct nicvf_cq_poll *cq_poll = NULL;
1462 /* wait till all queued set_rx_mode tasks completes if any */
1463 if (nic->nicvf_rx_mode_wq)
1464 drain_workqueue(nic->nicvf_rx_mode_wq);
1466 netif_carrier_off(netdev);
1468 err = nicvf_register_misc_interrupt(nic);
1472 /* Register NAPI handler for processing CQEs */
1473 for (qidx = 0; qidx < qs->cq_cnt; qidx++) {
1474 cq_poll = kzalloc(sizeof(*cq_poll), GFP_KERNEL);
1479 cq_poll->cq_idx = qidx;
1480 cq_poll->nicvf = nic;
1481 netif_napi_add(netdev, &cq_poll->napi, nicvf_poll,
1483 napi_enable(&cq_poll->napi);
1484 nic->napi[qidx] = cq_poll;
1487 /* Check if we got MAC address from PF or else generate a radom MAC */
1488 if (!nic->sqs_mode && is_zero_ether_addr(netdev->dev_addr)) {
1489 eth_hw_addr_random(netdev);
1490 nicvf_hw_set_mac_addr(nic, netdev);
1493 if (nic->set_mac_pending) {
1494 nic->set_mac_pending = false;
1495 nicvf_hw_set_mac_addr(nic, netdev);
1498 /* Init tasklet for handling Qset err interrupt */
1499 tasklet_init(&nic->qs_err_task, nicvf_handle_qs_err,
1500 (unsigned long)nic);
1502 /* Init RBDR tasklet which will refill RBDR */
1503 tasklet_init(&nic->rbdr_task, nicvf_rbdr_task,
1504 (unsigned long)nic);
1505 INIT_DELAYED_WORK(&nic->rbdr_work, nicvf_rbdr_work);
1507 /* Configure CPI alorithm */
1508 nic->cpi_alg = cpi_alg;
1510 nicvf_config_cpi(nic);
1512 nicvf_request_sqs(nic);
1514 nicvf_get_primary_vf_struct(nic);
1516 /* Configure PTP timestamp */
1518 nicvf_config_hw_rx_tstamp(nic, nic->hw_rx_tstamp);
1519 atomic_set(&nic->tx_ptp_skbs, 0);
1520 nic->ptp_skb = NULL;
1522 /* Configure receive side scaling and MTU */
1523 if (!nic->sqs_mode) {
1524 nicvf_rss_init(nic);
1525 err = nicvf_update_hw_max_frs(nic, netdev->mtu);
1529 /* Clear percpu stats */
1530 for_each_possible_cpu(cpu)
1531 memset(per_cpu_ptr(nic->drv_stats, cpu), 0,
1532 sizeof(struct nicvf_drv_stats));
1535 err = nicvf_register_interrupts(nic);
1539 /* Initialize the queues */
1540 err = nicvf_init_resources(nic);
1544 /* Make sure queue initialization is written */
1547 nicvf_reg_write(nic, NIC_VF_INT, -1);
1548 /* Enable Qset err interrupt */
1549 nicvf_enable_intr(nic, NICVF_INTR_QS_ERR, 0);
1551 /* Enable completion queue interrupt */
1552 for (qidx = 0; qidx < qs->cq_cnt; qidx++)
1553 nicvf_enable_intr(nic, NICVF_INTR_CQ, qidx);
1555 /* Enable RBDR threshold interrupt */
1556 for (qidx = 0; qidx < qs->rbdr_cnt; qidx++)
1557 nicvf_enable_intr(nic, NICVF_INTR_RBDR, qidx);
1559 /* Send VF config done msg to PF */
1560 nicvf_send_cfg_done(nic);
1562 if (nic->nicvf_rx_mode_wq) {
1563 INIT_DELAYED_WORK(&nic->link_change_work,
1564 nicvf_link_status_check_task);
1565 queue_delayed_work(nic->nicvf_rx_mode_wq,
1566 &nic->link_change_work, 0);
1571 nicvf_disable_intr(nic, NICVF_INTR_MBOX, 0);
1572 nicvf_unregister_interrupts(nic);
1573 tasklet_kill(&nic->qs_err_task);
1574 tasklet_kill(&nic->rbdr_task);
1576 for (qidx = 0; qidx < qs->cq_cnt; qidx++) {
1577 cq_poll = nic->napi[qidx];
1580 napi_disable(&cq_poll->napi);
1581 netif_napi_del(&cq_poll->napi);
1583 nicvf_free_cq_poll(nic);
1587 static int nicvf_change_mtu(struct net_device *netdev, int new_mtu)
1589 struct nicvf *nic = netdev_priv(netdev);
1590 int orig_mtu = netdev->mtu;
1592 /* For now just support only the usual MTU sized frames,
1593 * plus some headroom for VLAN, QinQ.
1595 if (nic->xdp_prog && new_mtu > MAX_XDP_MTU) {
1596 netdev_warn(netdev, "Jumbo frames not yet supported with XDP, current MTU %d.\n",
1601 netdev->mtu = new_mtu;
1603 if (!netif_running(netdev))
1606 if (nicvf_update_hw_max_frs(nic, new_mtu)) {
1607 netdev->mtu = orig_mtu;
1614 static int nicvf_set_mac_address(struct net_device *netdev, void *p)
1616 struct sockaddr *addr = p;
1617 struct nicvf *nic = netdev_priv(netdev);
1619 if (!is_valid_ether_addr(addr->sa_data))
1620 return -EADDRNOTAVAIL;
1622 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1624 if (nic->pdev->msix_enabled) {
1625 if (nicvf_hw_set_mac_addr(nic, netdev))
1628 nic->set_mac_pending = true;
1634 void nicvf_update_lmac_stats(struct nicvf *nic)
1637 union nic_mbx mbx = {};
1639 if (!netif_running(nic->netdev))
1642 mbx.bgx_stats.msg = NIC_MBOX_MSG_BGX_STATS;
1643 mbx.bgx_stats.vf_id = nic->vf_id;
1645 mbx.bgx_stats.rx = 1;
1646 while (stat < BGX_RX_STATS_COUNT) {
1647 mbx.bgx_stats.idx = stat;
1648 if (nicvf_send_msg_to_pf(nic, &mbx))
1656 mbx.bgx_stats.rx = 0;
1657 while (stat < BGX_TX_STATS_COUNT) {
1658 mbx.bgx_stats.idx = stat;
1659 if (nicvf_send_msg_to_pf(nic, &mbx))
1665 void nicvf_update_stats(struct nicvf *nic)
1669 struct nicvf_hw_stats *stats = &nic->hw_stats;
1670 struct nicvf_drv_stats *drv_stats;
1671 struct queue_set *qs = nic->qs;
1673 #define GET_RX_STATS(reg) \
1674 nicvf_reg_read(nic, NIC_VNIC_RX_STAT_0_13 | (reg << 3))
1675 #define GET_TX_STATS(reg) \
1676 nicvf_reg_read(nic, NIC_VNIC_TX_STAT_0_4 | (reg << 3))
1678 stats->rx_bytes = GET_RX_STATS(RX_OCTS);
1679 stats->rx_ucast_frames = GET_RX_STATS(RX_UCAST);
1680 stats->rx_bcast_frames = GET_RX_STATS(RX_BCAST);
1681 stats->rx_mcast_frames = GET_RX_STATS(RX_MCAST);
1682 stats->rx_fcs_errors = GET_RX_STATS(RX_FCS);
1683 stats->rx_l2_errors = GET_RX_STATS(RX_L2ERR);
1684 stats->rx_drop_red = GET_RX_STATS(RX_RED);
1685 stats->rx_drop_red_bytes = GET_RX_STATS(RX_RED_OCTS);
1686 stats->rx_drop_overrun = GET_RX_STATS(RX_ORUN);
1687 stats->rx_drop_overrun_bytes = GET_RX_STATS(RX_ORUN_OCTS);
1688 stats->rx_drop_bcast = GET_RX_STATS(RX_DRP_BCAST);
1689 stats->rx_drop_mcast = GET_RX_STATS(RX_DRP_MCAST);
1690 stats->rx_drop_l3_bcast = GET_RX_STATS(RX_DRP_L3BCAST);
1691 stats->rx_drop_l3_mcast = GET_RX_STATS(RX_DRP_L3MCAST);
1693 stats->tx_bytes = GET_TX_STATS(TX_OCTS);
1694 stats->tx_ucast_frames = GET_TX_STATS(TX_UCAST);
1695 stats->tx_bcast_frames = GET_TX_STATS(TX_BCAST);
1696 stats->tx_mcast_frames = GET_TX_STATS(TX_MCAST);
1697 stats->tx_drops = GET_TX_STATS(TX_DROP);
1699 /* On T88 pass 2.0, the dummy SQE added for TSO notification
1700 * via CQE has 'dont_send' set. Hence HW drops the pkt pointed
1701 * pointed by dummy SQE and results in tx_drops counter being
1702 * incremented. Subtracting it from tx_tso counter will give
1703 * exact tx_drops counter.
1705 if (nic->t88 && nic->hw_tso) {
1706 for_each_possible_cpu(cpu) {
1707 drv_stats = per_cpu_ptr(nic->drv_stats, cpu);
1708 tmp_stats += drv_stats->tx_tso;
1710 stats->tx_drops = tmp_stats - stats->tx_drops;
1712 stats->tx_frames = stats->tx_ucast_frames +
1713 stats->tx_bcast_frames +
1714 stats->tx_mcast_frames;
1715 stats->rx_frames = stats->rx_ucast_frames +
1716 stats->rx_bcast_frames +
1717 stats->rx_mcast_frames;
1718 stats->rx_drops = stats->rx_drop_red +
1719 stats->rx_drop_overrun;
1721 /* Update RQ and SQ stats */
1722 for (qidx = 0; qidx < qs->rq_cnt; qidx++)
1723 nicvf_update_rq_stats(nic, qidx);
1724 for (qidx = 0; qidx < qs->sq_cnt; qidx++)
1725 nicvf_update_sq_stats(nic, qidx);
1728 static void nicvf_get_stats64(struct net_device *netdev,
1729 struct rtnl_link_stats64 *stats)
1731 struct nicvf *nic = netdev_priv(netdev);
1732 struct nicvf_hw_stats *hw_stats = &nic->hw_stats;
1734 nicvf_update_stats(nic);
1736 stats->rx_bytes = hw_stats->rx_bytes;
1737 stats->rx_packets = hw_stats->rx_frames;
1738 stats->rx_dropped = hw_stats->rx_drops;
1739 stats->multicast = hw_stats->rx_mcast_frames;
1741 stats->tx_bytes = hw_stats->tx_bytes;
1742 stats->tx_packets = hw_stats->tx_frames;
1743 stats->tx_dropped = hw_stats->tx_drops;
1747 static void nicvf_tx_timeout(struct net_device *dev)
1749 struct nicvf *nic = netdev_priv(dev);
1751 netif_warn(nic, tx_err, dev, "Transmit timed out, resetting\n");
1753 this_cpu_inc(nic->drv_stats->tx_timeout);
1754 schedule_work(&nic->reset_task);
1757 static void nicvf_reset_task(struct work_struct *work)
1761 nic = container_of(work, struct nicvf, reset_task);
1763 if (!netif_running(nic->netdev))
1766 nicvf_stop(nic->netdev);
1767 nicvf_open(nic->netdev);
1768 netif_trans_update(nic->netdev);
1771 static int nicvf_config_loopback(struct nicvf *nic,
1772 netdev_features_t features)
1774 union nic_mbx mbx = {};
1776 mbx.lbk.msg = NIC_MBOX_MSG_LOOPBACK;
1777 mbx.lbk.vf_id = nic->vf_id;
1778 mbx.lbk.enable = (features & NETIF_F_LOOPBACK) != 0;
1780 return nicvf_send_msg_to_pf(nic, &mbx);
1783 static netdev_features_t nicvf_fix_features(struct net_device *netdev,
1784 netdev_features_t features)
1786 struct nicvf *nic = netdev_priv(netdev);
1788 if ((features & NETIF_F_LOOPBACK) &&
1789 netif_running(netdev) && !nic->loopback_supported)
1790 features &= ~NETIF_F_LOOPBACK;
1795 static int nicvf_set_features(struct net_device *netdev,
1796 netdev_features_t features)
1798 struct nicvf *nic = netdev_priv(netdev);
1799 netdev_features_t changed = features ^ netdev->features;
1801 if (changed & NETIF_F_HW_VLAN_CTAG_RX)
1802 nicvf_config_vlan_stripping(nic, features);
1804 if ((changed & NETIF_F_LOOPBACK) && netif_running(netdev))
1805 return nicvf_config_loopback(nic, features);
1810 static void nicvf_set_xdp_queues(struct nicvf *nic, bool bpf_attached)
1812 u8 cq_count, txq_count;
1814 /* Set XDP Tx queue count same as Rx queue count */
1816 nic->xdp_tx_queues = 0;
1818 nic->xdp_tx_queues = nic->rx_queues;
1820 /* If queue count > MAX_CMP_QUEUES_PER_QS, then additional qsets
1821 * needs to be allocated, check how many.
1823 txq_count = nic->xdp_tx_queues + nic->tx_queues;
1824 cq_count = max(nic->rx_queues, txq_count);
1825 if (cq_count > MAX_CMP_QUEUES_PER_QS) {
1826 nic->sqs_count = roundup(cq_count, MAX_CMP_QUEUES_PER_QS);
1827 nic->sqs_count = (nic->sqs_count / MAX_CMP_QUEUES_PER_QS) - 1;
1832 /* Set primary Qset's resources */
1833 nic->qs->rq_cnt = min_t(u8, nic->rx_queues, MAX_RCV_QUEUES_PER_QS);
1834 nic->qs->sq_cnt = min_t(u8, txq_count, MAX_SND_QUEUES_PER_QS);
1835 nic->qs->cq_cnt = max_t(u8, nic->qs->rq_cnt, nic->qs->sq_cnt);
1838 nicvf_set_real_num_queues(nic->netdev, nic->tx_queues, nic->rx_queues);
1841 static int nicvf_xdp_setup(struct nicvf *nic, struct bpf_prog *prog)
1843 struct net_device *dev = nic->netdev;
1844 bool if_up = netif_running(nic->netdev);
1845 struct bpf_prog *old_prog;
1846 bool bpf_attached = false;
1849 /* For now just support only the usual MTU sized frames,
1850 * plus some headroom for VLAN, QinQ.
1852 if (prog && dev->mtu > MAX_XDP_MTU) {
1853 netdev_warn(dev, "Jumbo frames not yet supported with XDP, current MTU %d.\n",
1858 /* ALL SQs attached to CQs i.e same as RQs, are treated as
1859 * XDP Tx queues and more Tx queues are allocated for
1860 * network stack to send pkts out.
1862 * No of Tx queues are either same as Rx queues or whatever
1863 * is left in max no of queues possible.
1865 if ((nic->rx_queues + nic->tx_queues) > nic->max_queues) {
1867 "Failed to attach BPF prog, RXQs + TXQs > Max %d\n",
1873 nicvf_stop(nic->netdev);
1875 old_prog = xchg(&nic->xdp_prog, prog);
1876 /* Detach old prog, if any */
1878 bpf_prog_put(old_prog);
1880 if (nic->xdp_prog) {
1881 /* Attach BPF program */
1882 nic->xdp_prog = bpf_prog_add(nic->xdp_prog, nic->rx_queues - 1);
1883 if (!IS_ERR(nic->xdp_prog)) {
1884 bpf_attached = true;
1886 ret = PTR_ERR(nic->xdp_prog);
1887 nic->xdp_prog = NULL;
1891 /* Calculate Tx queues needed for XDP and network stack */
1892 nicvf_set_xdp_queues(nic, bpf_attached);
1895 /* Reinitialize interface, clean slate */
1896 nicvf_open(nic->netdev);
1897 netif_trans_update(nic->netdev);
1903 static int nicvf_xdp(struct net_device *netdev, struct netdev_bpf *xdp)
1905 struct nicvf *nic = netdev_priv(netdev);
1907 /* To avoid checks while retrieving buffer address from CQE_RX,
1908 * do not support XDP for T88 pass1.x silicons which are anyway
1909 * not in use widely.
1911 if (pass1_silicon(nic->pdev))
1914 switch (xdp->command) {
1915 case XDP_SETUP_PROG:
1916 return nicvf_xdp_setup(nic, xdp->prog);
1917 case XDP_QUERY_PROG:
1918 xdp->prog_id = nic->xdp_prog ? nic->xdp_prog->aux->id : 0;
1925 static int nicvf_config_hwtstamp(struct net_device *netdev, struct ifreq *ifr)
1927 struct hwtstamp_config config;
1928 struct nicvf *nic = netdev_priv(netdev);
1930 if (!nic->ptp_clock)
1933 if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
1936 /* reserved for future extensions */
1940 switch (config.tx_type) {
1941 case HWTSTAMP_TX_OFF:
1942 case HWTSTAMP_TX_ON:
1948 switch (config.rx_filter) {
1949 case HWTSTAMP_FILTER_NONE:
1950 nic->hw_rx_tstamp = false;
1952 case HWTSTAMP_FILTER_ALL:
1953 case HWTSTAMP_FILTER_SOME:
1954 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
1955 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
1956 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
1957 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
1958 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
1959 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
1960 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
1961 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
1962 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
1963 case HWTSTAMP_FILTER_PTP_V2_EVENT:
1964 case HWTSTAMP_FILTER_PTP_V2_SYNC:
1965 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
1966 nic->hw_rx_tstamp = true;
1967 config.rx_filter = HWTSTAMP_FILTER_ALL;
1973 if (netif_running(netdev))
1974 nicvf_config_hw_rx_tstamp(nic, nic->hw_rx_tstamp);
1976 if (copy_to_user(ifr->ifr_data, &config, sizeof(config)))
1982 static int nicvf_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
1986 return nicvf_config_hwtstamp(netdev, req);
1992 static void __nicvf_set_rx_mode_task(u8 mode, struct xcast_addr_list *mc_addrs,
1995 union nic_mbx mbx = {};
1998 /* From the inside of VM code flow we have only 128 bits memory
1999 * available to send message to host's PF, so send all mc addrs
2000 * one by one, starting from flush command in case if kernel
2001 * requests to configure specific MAC filtering
2004 /* flush DMAC filters and reset RX mode */
2005 mbx.xcast.msg = NIC_MBOX_MSG_RESET_XCAST;
2006 if (nicvf_send_msg_to_pf(nic, &mbx) < 0)
2009 if (mode & BGX_XCAST_MCAST_FILTER) {
2010 /* once enabling filtering, we need to signal to PF to add
2011 * its' own LMAC to the filter to accept packets for it.
2013 mbx.xcast.msg = NIC_MBOX_MSG_ADD_MCAST;
2015 if (nicvf_send_msg_to_pf(nic, &mbx) < 0)
2019 /* check if we have any specific MACs to be added to PF DMAC filter */
2021 /* now go through kernel list of MACs and add them one by one */
2022 for (idx = 0; idx < mc_addrs->count; idx++) {
2023 mbx.xcast.msg = NIC_MBOX_MSG_ADD_MCAST;
2024 mbx.xcast.mac = mc_addrs->mc[idx];
2025 if (nicvf_send_msg_to_pf(nic, &mbx) < 0)
2030 /* and finally set rx mode for PF accordingly */
2031 mbx.xcast.msg = NIC_MBOX_MSG_SET_XCAST;
2032 mbx.xcast.mode = mode;
2034 nicvf_send_msg_to_pf(nic, &mbx);
2039 static void nicvf_set_rx_mode_task(struct work_struct *work_arg)
2041 struct nicvf_work *vf_work = container_of(work_arg, struct nicvf_work,
2043 struct nicvf *nic = container_of(vf_work, struct nicvf, rx_mode_work);
2045 struct xcast_addr_list *mc;
2050 /* Save message data locally to prevent them from
2051 * being overwritten by next ndo_set_rx_mode call().
2053 spin_lock(&nic->rx_mode_wq_lock);
2054 mode = vf_work->mode;
2057 spin_unlock(&nic->rx_mode_wq_lock);
2059 __nicvf_set_rx_mode_task(mode, mc, nic);
2062 static void nicvf_set_rx_mode(struct net_device *netdev)
2064 struct nicvf *nic = netdev_priv(netdev);
2065 struct netdev_hw_addr *ha;
2066 struct xcast_addr_list *mc_list = NULL;
2069 if (netdev->flags & IFF_PROMISC) {
2070 mode = BGX_XCAST_BCAST_ACCEPT | BGX_XCAST_MCAST_ACCEPT;
2072 if (netdev->flags & IFF_BROADCAST)
2073 mode |= BGX_XCAST_BCAST_ACCEPT;
2075 if (netdev->flags & IFF_ALLMULTI) {
2076 mode |= BGX_XCAST_MCAST_ACCEPT;
2077 } else if (netdev->flags & IFF_MULTICAST) {
2078 mode |= BGX_XCAST_MCAST_FILTER;
2079 /* here we need to copy mc addrs */
2080 if (netdev_mc_count(netdev)) {
2081 mc_list = kmalloc(offsetof(typeof(*mc_list),
2082 mc[netdev_mc_count(netdev)]),
2084 if (unlikely(!mc_list))
2087 netdev_hw_addr_list_for_each(ha, &netdev->mc) {
2088 mc_list->mc[mc_list->count] =
2089 ether_addr_to_u64(ha->addr);
2095 spin_lock(&nic->rx_mode_wq_lock);
2096 kfree(nic->rx_mode_work.mc);
2097 nic->rx_mode_work.mc = mc_list;
2098 nic->rx_mode_work.mode = mode;
2099 queue_work(nic->nicvf_rx_mode_wq, &nic->rx_mode_work.work);
2100 spin_unlock(&nic->rx_mode_wq_lock);
2103 static const struct net_device_ops nicvf_netdev_ops = {
2104 .ndo_open = nicvf_open,
2105 .ndo_stop = nicvf_stop,
2106 .ndo_start_xmit = nicvf_xmit,
2107 .ndo_change_mtu = nicvf_change_mtu,
2108 .ndo_set_mac_address = nicvf_set_mac_address,
2109 .ndo_get_stats64 = nicvf_get_stats64,
2110 .ndo_tx_timeout = nicvf_tx_timeout,
2111 .ndo_fix_features = nicvf_fix_features,
2112 .ndo_set_features = nicvf_set_features,
2113 .ndo_bpf = nicvf_xdp,
2114 .ndo_do_ioctl = nicvf_ioctl,
2115 .ndo_set_rx_mode = nicvf_set_rx_mode,
2118 static int nicvf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2120 struct device *dev = &pdev->dev;
2121 struct net_device *netdev;
2125 struct cavium_ptp *ptp_clock;
2127 ptp_clock = cavium_ptp_get();
2128 if (IS_ERR(ptp_clock)) {
2129 if (PTR_ERR(ptp_clock) == -ENODEV)
2130 /* In virtualized environment we proceed without ptp */
2133 return PTR_ERR(ptp_clock);
2136 err = pci_enable_device(pdev);
2138 dev_err(dev, "Failed to enable PCI device\n");
2142 err = pci_request_regions(pdev, DRV_NAME);
2144 dev_err(dev, "PCI request regions failed 0x%x\n", err);
2145 goto err_disable_device;
2148 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(48));
2150 dev_err(dev, "Unable to get usable DMA configuration\n");
2151 goto err_release_regions;
2154 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(48));
2156 dev_err(dev, "unable to get 48-bit DMA for consistent allocations\n");
2157 goto err_release_regions;
2160 qcount = netif_get_num_default_rss_queues();
2162 /* Restrict multiqset support only for host bound VFs */
2163 if (pdev->is_virtfn) {
2164 /* Set max number of queues per VF */
2165 qcount = min_t(int, num_online_cpus(),
2166 (MAX_SQS_PER_VF + 1) * MAX_CMP_QUEUES_PER_QS);
2169 netdev = alloc_etherdev_mqs(sizeof(struct nicvf), qcount, qcount);
2172 goto err_release_regions;
2175 pci_set_drvdata(pdev, netdev);
2177 SET_NETDEV_DEV(netdev, &pdev->dev);
2179 nic = netdev_priv(netdev);
2180 nic->netdev = netdev;
2183 nic->max_queues = qcount;
2184 /* If no of CPUs are too low, there won't be any queues left
2185 * for XDP_TX, hence double it.
2188 nic->max_queues *= 2;
2189 nic->ptp_clock = ptp_clock;
2191 /* MAP VF's configuration registers */
2192 nic->reg_base = pcim_iomap(pdev, PCI_CFG_REG_BAR_NUM, 0);
2193 if (!nic->reg_base) {
2194 dev_err(dev, "Cannot map config register space, aborting\n");
2196 goto err_free_netdev;
2199 nic->drv_stats = netdev_alloc_pcpu_stats(struct nicvf_drv_stats);
2200 if (!nic->drv_stats) {
2202 goto err_free_netdev;
2205 err = nicvf_set_qset_resources(nic);
2207 goto err_free_netdev;
2209 /* Check if PF is alive and get MAC address for this VF */
2210 err = nicvf_register_misc_interrupt(nic);
2212 goto err_free_netdev;
2214 nicvf_send_vf_struct(nic);
2216 if (!pass1_silicon(nic->pdev))
2219 /* Get iommu domain for iova to physical addr conversion */
2220 nic->iommu_domain = iommu_get_domain_for_dev(dev);
2222 pci_read_config_word(nic->pdev, PCI_SUBSYSTEM_ID, &sdevid);
2223 if (sdevid == 0xA134)
2226 /* Check if this VF is in QS only mode */
2230 err = nicvf_set_real_num_queues(netdev, nic->tx_queues, nic->rx_queues);
2232 goto err_unregister_interrupts;
2234 netdev->hw_features = (NETIF_F_RXCSUM | NETIF_F_SG |
2235 NETIF_F_TSO | NETIF_F_GRO | NETIF_F_TSO6 |
2236 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2237 NETIF_F_HW_VLAN_CTAG_RX);
2239 netdev->hw_features |= NETIF_F_RXHASH;
2241 netdev->features |= netdev->hw_features;
2242 netdev->hw_features |= NETIF_F_LOOPBACK;
2244 netdev->vlan_features = NETIF_F_SG | NETIF_F_IP_CSUM |
2245 NETIF_F_IPV6_CSUM | NETIF_F_TSO | NETIF_F_TSO6;
2247 netdev->netdev_ops = &nicvf_netdev_ops;
2248 netdev->watchdog_timeo = NICVF_TX_TIMEOUT;
2250 /* MTU range: 64 - 9200 */
2251 netdev->min_mtu = NIC_HW_MIN_FRS;
2252 netdev->max_mtu = NIC_HW_MAX_FRS;
2254 INIT_WORK(&nic->reset_task, nicvf_reset_task);
2256 nic->nicvf_rx_mode_wq = alloc_ordered_workqueue("nicvf_rx_mode_wq_VF%d",
2259 if (!nic->nicvf_rx_mode_wq) {
2261 dev_err(dev, "Failed to allocate work queue\n");
2262 goto err_unregister_interrupts;
2265 INIT_WORK(&nic->rx_mode_work.work, nicvf_set_rx_mode_task);
2266 spin_lock_init(&nic->rx_mode_wq_lock);
2267 mutex_init(&nic->rx_mode_mtx);
2269 err = register_netdev(netdev);
2271 dev_err(dev, "Failed to register netdevice\n");
2272 goto err_unregister_interrupts;
2275 nic->msg_enable = debug;
2277 nicvf_set_ethtool_ops(netdev);
2281 err_unregister_interrupts:
2282 nicvf_unregister_interrupts(nic);
2284 pci_set_drvdata(pdev, NULL);
2286 free_percpu(nic->drv_stats);
2287 free_netdev(netdev);
2288 err_release_regions:
2289 pci_release_regions(pdev);
2291 pci_disable_device(pdev);
2295 static void nicvf_remove(struct pci_dev *pdev)
2297 struct net_device *netdev = pci_get_drvdata(pdev);
2299 struct net_device *pnetdev;
2304 nic = netdev_priv(netdev);
2305 pnetdev = nic->pnicvf->netdev;
2307 /* Check if this Qset is assigned to different VF.
2308 * If yes, clean primary and all secondary Qsets.
2310 if (pnetdev && (pnetdev->reg_state == NETREG_REGISTERED))
2311 unregister_netdev(pnetdev);
2312 if (nic->nicvf_rx_mode_wq) {
2313 destroy_workqueue(nic->nicvf_rx_mode_wq);
2314 nic->nicvf_rx_mode_wq = NULL;
2316 nicvf_unregister_interrupts(nic);
2317 pci_set_drvdata(pdev, NULL);
2319 free_percpu(nic->drv_stats);
2320 cavium_ptp_put(nic->ptp_clock);
2321 free_netdev(netdev);
2322 pci_release_regions(pdev);
2323 pci_disable_device(pdev);
2326 static void nicvf_shutdown(struct pci_dev *pdev)
2331 static struct pci_driver nicvf_driver = {
2333 .id_table = nicvf_id_table,
2334 .probe = nicvf_probe,
2335 .remove = nicvf_remove,
2336 .shutdown = nicvf_shutdown,
2339 static int __init nicvf_init_module(void)
2341 pr_info("%s, ver %s\n", DRV_NAME, DRV_VERSION);
2342 return pci_register_driver(&nicvf_driver);
2345 static void __exit nicvf_cleanup_module(void)
2347 pci_unregister_driver(&nicvf_driver);
2350 module_init(nicvf_init_module);
2351 module_exit(nicvf_cleanup_module);