1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2016 Intel Corporation
11 #include <ethdev_driver.h>
12 #include <ethdev_pci.h>
13 #include <rte_memcpy.h>
14 #include <rte_string_fns.h>
15 #include <rte_memzone.h>
16 #include <rte_malloc.h>
17 #include <rte_branch_prediction.h>
19 #include <rte_bus_pci.h>
20 #include <rte_ether.h>
23 #include <rte_common.h>
24 #include <rte_errno.h>
25 #include <rte_cpuflags.h>
28 #include <rte_memory.h>
31 #include <rte_cycles.h>
32 #include <rte_kvargs.h>
34 #include "virtio_ethdev.h"
35 #include "virtio_pci.h"
36 #include "virtio_logs.h"
37 #include "virtqueue.h"
38 #include "virtio_rxtx.h"
39 #include "virtio_user/virtio_user_dev.h"
41 static int virtio_dev_configure(struct rte_eth_dev *dev);
42 static int virtio_dev_start(struct rte_eth_dev *dev);
43 static int virtio_dev_promiscuous_enable(struct rte_eth_dev *dev);
44 static int virtio_dev_promiscuous_disable(struct rte_eth_dev *dev);
45 static int virtio_dev_allmulticast_enable(struct rte_eth_dev *dev);
46 static int virtio_dev_allmulticast_disable(struct rte_eth_dev *dev);
47 static uint32_t virtio_dev_speed_capa_get(uint32_t speed);
48 static int virtio_dev_devargs_parse(struct rte_devargs *devargs,
51 static int virtio_dev_info_get(struct rte_eth_dev *dev,
52 struct rte_eth_dev_info *dev_info);
53 static int virtio_dev_link_update(struct rte_eth_dev *dev,
54 int wait_to_complete);
55 static int virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask);
57 static void virtio_set_hwaddr(struct virtio_hw *hw);
58 static void virtio_get_hwaddr(struct virtio_hw *hw);
60 static int virtio_dev_stats_get(struct rte_eth_dev *dev,
61 struct rte_eth_stats *stats);
62 static int virtio_dev_xstats_get(struct rte_eth_dev *dev,
63 struct rte_eth_xstat *xstats, unsigned n);
64 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
65 struct rte_eth_xstat_name *xstats_names,
67 static int virtio_dev_stats_reset(struct rte_eth_dev *dev);
68 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev);
69 static int virtio_vlan_filter_set(struct rte_eth_dev *dev,
70 uint16_t vlan_id, int on);
71 static int virtio_mac_addr_add(struct rte_eth_dev *dev,
72 struct rte_ether_addr *mac_addr,
73 uint32_t index, uint32_t vmdq);
74 static void virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index);
75 static int virtio_mac_addr_set(struct rte_eth_dev *dev,
76 struct rte_ether_addr *mac_addr);
78 static int virtio_intr_disable(struct rte_eth_dev *dev);
80 static int virtio_dev_queue_stats_mapping_set(
81 struct rte_eth_dev *eth_dev,
86 static void virtio_notify_peers(struct rte_eth_dev *dev);
87 static void virtio_ack_link_announce(struct rte_eth_dev *dev);
89 struct rte_virtio_xstats_name_off {
90 char name[RTE_ETH_XSTATS_NAME_SIZE];
94 /* [rt]x_qX_ is prepended to the name string here */
95 static const struct rte_virtio_xstats_name_off rte_virtio_rxq_stat_strings[] = {
96 {"good_packets", offsetof(struct virtnet_rx, stats.packets)},
97 {"good_bytes", offsetof(struct virtnet_rx, stats.bytes)},
98 {"errors", offsetof(struct virtnet_rx, stats.errors)},
99 {"multicast_packets", offsetof(struct virtnet_rx, stats.multicast)},
100 {"broadcast_packets", offsetof(struct virtnet_rx, stats.broadcast)},
101 {"undersize_packets", offsetof(struct virtnet_rx, stats.size_bins[0])},
102 {"size_64_packets", offsetof(struct virtnet_rx, stats.size_bins[1])},
103 {"size_65_127_packets", offsetof(struct virtnet_rx, stats.size_bins[2])},
104 {"size_128_255_packets", offsetof(struct virtnet_rx, stats.size_bins[3])},
105 {"size_256_511_packets", offsetof(struct virtnet_rx, stats.size_bins[4])},
106 {"size_512_1023_packets", offsetof(struct virtnet_rx, stats.size_bins[5])},
107 {"size_1024_1518_packets", offsetof(struct virtnet_rx, stats.size_bins[6])},
108 {"size_1519_max_packets", offsetof(struct virtnet_rx, stats.size_bins[7])},
111 /* [rt]x_qX_ is prepended to the name string here */
112 static const struct rte_virtio_xstats_name_off rte_virtio_txq_stat_strings[] = {
113 {"good_packets", offsetof(struct virtnet_tx, stats.packets)},
114 {"good_bytes", offsetof(struct virtnet_tx, stats.bytes)},
115 {"multicast_packets", offsetof(struct virtnet_tx, stats.multicast)},
116 {"broadcast_packets", offsetof(struct virtnet_tx, stats.broadcast)},
117 {"undersize_packets", offsetof(struct virtnet_tx, stats.size_bins[0])},
118 {"size_64_packets", offsetof(struct virtnet_tx, stats.size_bins[1])},
119 {"size_65_127_packets", offsetof(struct virtnet_tx, stats.size_bins[2])},
120 {"size_128_255_packets", offsetof(struct virtnet_tx, stats.size_bins[3])},
121 {"size_256_511_packets", offsetof(struct virtnet_tx, stats.size_bins[4])},
122 {"size_512_1023_packets", offsetof(struct virtnet_tx, stats.size_bins[5])},
123 {"size_1024_1518_packets", offsetof(struct virtnet_tx, stats.size_bins[6])},
124 {"size_1519_max_packets", offsetof(struct virtnet_tx, stats.size_bins[7])},
127 #define VIRTIO_NB_RXQ_XSTATS (sizeof(rte_virtio_rxq_stat_strings) / \
128 sizeof(rte_virtio_rxq_stat_strings[0]))
129 #define VIRTIO_NB_TXQ_XSTATS (sizeof(rte_virtio_txq_stat_strings) / \
130 sizeof(rte_virtio_txq_stat_strings[0]))
132 struct virtio_hw_internal virtio_hw_internal[RTE_MAX_ETHPORTS];
134 static struct virtio_pmd_ctrl *
135 virtio_send_command_packed(struct virtnet_ctl *cvq,
136 struct virtio_pmd_ctrl *ctrl,
137 int *dlen, int pkt_num)
139 struct virtqueue *vq = cvq->vq;
141 struct vring_packed_desc *desc = vq->vq_packed.ring.desc;
142 struct virtio_pmd_ctrl *result;
149 * Format is enforced in qemu code:
150 * One TX packet for header;
151 * At least one TX packet per argument;
152 * One RX packet for ACK.
154 head = vq->vq_avail_idx;
155 flags = vq->vq_packed.cached_flags;
156 desc[head].addr = cvq->virtio_net_hdr_mem;
157 desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
160 if (++vq->vq_avail_idx >= vq->vq_nentries) {
161 vq->vq_avail_idx -= vq->vq_nentries;
162 vq->vq_packed.cached_flags ^= VRING_PACKED_DESC_F_AVAIL_USED;
165 for (k = 0; k < pkt_num; k++) {
166 desc[vq->vq_avail_idx].addr = cvq->virtio_net_hdr_mem
167 + sizeof(struct virtio_net_ctrl_hdr)
168 + sizeof(ctrl->status) + sizeof(uint8_t) * sum;
169 desc[vq->vq_avail_idx].len = dlen[k];
170 desc[vq->vq_avail_idx].flags = VRING_DESC_F_NEXT |
171 vq->vq_packed.cached_flags;
175 if (++vq->vq_avail_idx >= vq->vq_nentries) {
176 vq->vq_avail_idx -= vq->vq_nentries;
177 vq->vq_packed.cached_flags ^=
178 VRING_PACKED_DESC_F_AVAIL_USED;
182 desc[vq->vq_avail_idx].addr = cvq->virtio_net_hdr_mem
183 + sizeof(struct virtio_net_ctrl_hdr);
184 desc[vq->vq_avail_idx].len = sizeof(ctrl->status);
185 desc[vq->vq_avail_idx].flags = VRING_DESC_F_WRITE |
186 vq->vq_packed.cached_flags;
189 if (++vq->vq_avail_idx >= vq->vq_nentries) {
190 vq->vq_avail_idx -= vq->vq_nentries;
191 vq->vq_packed.cached_flags ^= VRING_PACKED_DESC_F_AVAIL_USED;
194 virtqueue_store_flags_packed(&desc[head], VRING_DESC_F_NEXT | flags,
195 vq->hw->weak_barriers);
197 virtio_wmb(vq->hw->weak_barriers);
198 virtqueue_notify(vq);
200 /* wait for used desc in virtqueue
201 * desc_is_used has a load-acquire or rte_io_rmb inside
203 while (!desc_is_used(&desc[head], vq))
206 /* now get used descriptors */
207 vq->vq_free_cnt += nb_descs;
208 vq->vq_used_cons_idx += nb_descs;
209 if (vq->vq_used_cons_idx >= vq->vq_nentries) {
210 vq->vq_used_cons_idx -= vq->vq_nentries;
211 vq->vq_packed.used_wrap_counter ^= 1;
214 PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\n"
215 "vq->vq_avail_idx=%d\n"
216 "vq->vq_used_cons_idx=%d\n"
217 "vq->vq_packed.cached_flags=0x%x\n"
218 "vq->vq_packed.used_wrap_counter=%d\n",
221 vq->vq_used_cons_idx,
222 vq->vq_packed.cached_flags,
223 vq->vq_packed.used_wrap_counter);
225 result = cvq->virtio_net_hdr_mz->addr;
229 static struct virtio_pmd_ctrl *
230 virtio_send_command_split(struct virtnet_ctl *cvq,
231 struct virtio_pmd_ctrl *ctrl,
232 int *dlen, int pkt_num)
234 struct virtio_pmd_ctrl *result;
235 struct virtqueue *vq = cvq->vq;
239 head = vq->vq_desc_head_idx;
242 * Format is enforced in qemu code:
243 * One TX packet for header;
244 * At least one TX packet per argument;
245 * One RX packet for ACK.
247 vq->vq_split.ring.desc[head].flags = VRING_DESC_F_NEXT;
248 vq->vq_split.ring.desc[head].addr = cvq->virtio_net_hdr_mem;
249 vq->vq_split.ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
251 i = vq->vq_split.ring.desc[head].next;
253 for (k = 0; k < pkt_num; k++) {
254 vq->vq_split.ring.desc[i].flags = VRING_DESC_F_NEXT;
255 vq->vq_split.ring.desc[i].addr = cvq->virtio_net_hdr_mem
256 + sizeof(struct virtio_net_ctrl_hdr)
257 + sizeof(ctrl->status) + sizeof(uint8_t)*sum;
258 vq->vq_split.ring.desc[i].len = dlen[k];
261 i = vq->vq_split.ring.desc[i].next;
264 vq->vq_split.ring.desc[i].flags = VRING_DESC_F_WRITE;
265 vq->vq_split.ring.desc[i].addr = cvq->virtio_net_hdr_mem
266 + sizeof(struct virtio_net_ctrl_hdr);
267 vq->vq_split.ring.desc[i].len = sizeof(ctrl->status);
270 vq->vq_desc_head_idx = vq->vq_split.ring.desc[i].next;
272 vq_update_avail_ring(vq, head);
273 vq_update_avail_idx(vq);
275 PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
277 virtqueue_notify(vq);
279 while (virtqueue_nused(vq) == 0)
282 while (virtqueue_nused(vq)) {
283 uint32_t idx, desc_idx, used_idx;
284 struct vring_used_elem *uep;
286 used_idx = (uint32_t)(vq->vq_used_cons_idx
287 & (vq->vq_nentries - 1));
288 uep = &vq->vq_split.ring.used->ring[used_idx];
289 idx = (uint32_t) uep->id;
292 while (vq->vq_split.ring.desc[desc_idx].flags &
294 desc_idx = vq->vq_split.ring.desc[desc_idx].next;
298 vq->vq_split.ring.desc[desc_idx].next = vq->vq_desc_head_idx;
299 vq->vq_desc_head_idx = idx;
301 vq->vq_used_cons_idx++;
305 PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
306 vq->vq_free_cnt, vq->vq_desc_head_idx);
308 result = cvq->virtio_net_hdr_mz->addr;
313 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
314 int *dlen, int pkt_num)
316 virtio_net_ctrl_ack status = ~0;
317 struct virtio_pmd_ctrl *result;
318 struct virtqueue *vq;
320 ctrl->status = status;
322 if (!cvq || !cvq->vq) {
323 PMD_INIT_LOG(ERR, "Control queue is not supported.");
327 rte_spinlock_lock(&cvq->lock);
330 PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
331 "vq->hw->cvq = %p vq = %p",
332 vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
334 if (vq->vq_free_cnt < pkt_num + 2 || pkt_num < 1) {
335 rte_spinlock_unlock(&cvq->lock);
339 memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
340 sizeof(struct virtio_pmd_ctrl));
342 if (vtpci_packed_queue(vq->hw))
343 result = virtio_send_command_packed(cvq, ctrl, dlen, pkt_num);
345 result = virtio_send_command_split(cvq, ctrl, dlen, pkt_num);
347 rte_spinlock_unlock(&cvq->lock);
348 return result->status;
352 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
354 struct virtio_hw *hw = dev->data->dev_private;
355 struct virtio_pmd_ctrl ctrl;
359 ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
360 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
361 memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
363 dlen[0] = sizeof(uint16_t);
365 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
367 PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
368 "failed, this is too late now...");
376 virtio_dev_queue_release(void *queue __rte_unused)
382 virtio_get_nr_vq(struct virtio_hw *hw)
384 uint16_t nr_vq = hw->max_queue_pairs * 2;
386 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
393 virtio_init_vring(struct virtqueue *vq)
395 int size = vq->vq_nentries;
396 uint8_t *ring_mem = vq->vq_ring_virt_mem;
398 PMD_INIT_FUNC_TRACE();
400 memset(ring_mem, 0, vq->vq_ring_size);
402 vq->vq_used_cons_idx = 0;
403 vq->vq_desc_head_idx = 0;
404 vq->vq_avail_idx = 0;
405 vq->vq_desc_tail_idx = (uint16_t)(vq->vq_nentries - 1);
406 vq->vq_free_cnt = vq->vq_nentries;
407 memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries);
408 if (vtpci_packed_queue(vq->hw)) {
409 vring_init_packed(&vq->vq_packed.ring, ring_mem,
410 VIRTIO_PCI_VRING_ALIGN, size);
411 vring_desc_init_packed(vq, size);
413 struct vring *vr = &vq->vq_split.ring;
415 vring_init_split(vr, ring_mem, VIRTIO_PCI_VRING_ALIGN, size);
416 vring_desc_init_split(vr->desc, size);
419 * Disable device(host) interrupting guest
421 virtqueue_disable_intr(vq);
425 virtio_init_queue(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx)
427 char vq_name[VIRTQUEUE_MAX_NAME_SZ];
428 char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
429 const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
430 unsigned int vq_size, size;
431 struct virtio_hw *hw = dev->data->dev_private;
432 struct virtnet_rx *rxvq = NULL;
433 struct virtnet_tx *txvq = NULL;
434 struct virtnet_ctl *cvq = NULL;
435 struct virtqueue *vq;
436 size_t sz_hdr_mz = 0;
437 void *sw_ring = NULL;
438 int queue_type = virtio_get_queue_type(hw, vtpci_queue_idx);
440 int numa_node = dev->device->numa_node;
442 PMD_INIT_LOG(INFO, "setting up queue: %u on NUMA node %d",
443 vtpci_queue_idx, numa_node);
446 * Read the virtqueue size from the Queue Size field
447 * Always power of 2 and if 0 virtqueue does not exist
449 vq_size = VTPCI_OPS(hw)->get_queue_num(hw, vtpci_queue_idx);
450 PMD_INIT_LOG(DEBUG, "vq_size: %u", vq_size);
452 PMD_INIT_LOG(ERR, "virtqueue does not exist");
456 if (!vtpci_packed_queue(hw) && !rte_is_power_of_2(vq_size)) {
457 PMD_INIT_LOG(ERR, "split virtqueue size is not power of 2");
461 snprintf(vq_name, sizeof(vq_name), "port%d_vq%d",
462 dev->data->port_id, vtpci_queue_idx);
464 size = RTE_ALIGN_CEIL(sizeof(*vq) +
465 vq_size * sizeof(struct vq_desc_extra),
466 RTE_CACHE_LINE_SIZE);
467 if (queue_type == VTNET_TQ) {
469 * For each xmit packet, allocate a virtio_net_hdr
470 * and indirect ring elements
472 sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
473 } else if (queue_type == VTNET_CQ) {
474 /* Allocate a page for control vq command, data and status */
475 sz_hdr_mz = PAGE_SIZE;
478 vq = rte_zmalloc_socket(vq_name, size, RTE_CACHE_LINE_SIZE,
481 PMD_INIT_LOG(ERR, "can not allocate vq");
484 hw->vqs[vtpci_queue_idx] = vq;
487 vq->vq_queue_index = vtpci_queue_idx;
488 vq->vq_nentries = vq_size;
489 if (vtpci_packed_queue(hw)) {
490 vq->vq_packed.used_wrap_counter = 1;
491 vq->vq_packed.cached_flags = VRING_PACKED_DESC_F_AVAIL;
492 vq->vq_packed.event_flags_shadow = 0;
493 if (queue_type == VTNET_RQ)
494 vq->vq_packed.cached_flags |= VRING_DESC_F_WRITE;
498 * Reserve a memzone for vring elements
500 size = vring_size(hw, vq_size, VIRTIO_PCI_VRING_ALIGN);
501 vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
502 PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
503 size, vq->vq_ring_size);
505 mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
506 numa_node, RTE_MEMZONE_IOVA_CONTIG,
507 VIRTIO_PCI_VRING_ALIGN);
509 if (rte_errno == EEXIST)
510 mz = rte_memzone_lookup(vq_name);
517 memset(mz->addr, 0, mz->len);
519 vq->vq_ring_mem = mz->iova;
520 vq->vq_ring_virt_mem = mz->addr;
521 PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem: 0x%" PRIx64,
523 PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64,
524 (uint64_t)(uintptr_t)mz->addr);
526 virtio_init_vring(vq);
529 snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_vq%d_hdr",
530 dev->data->port_id, vtpci_queue_idx);
531 hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
532 numa_node, RTE_MEMZONE_IOVA_CONTIG,
533 RTE_CACHE_LINE_SIZE);
534 if (hdr_mz == NULL) {
535 if (rte_errno == EEXIST)
536 hdr_mz = rte_memzone_lookup(vq_hdr_name);
537 if (hdr_mz == NULL) {
544 if (queue_type == VTNET_RQ) {
545 size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
546 sizeof(vq->sw_ring[0]);
548 sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
549 RTE_CACHE_LINE_SIZE, numa_node);
551 PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
556 vq->sw_ring = sw_ring;
559 rxvq->port_id = dev->data->port_id;
561 } else if (queue_type == VTNET_TQ) {
564 txvq->port_id = dev->data->port_id;
566 txvq->virtio_net_hdr_mz = hdr_mz;
567 txvq->virtio_net_hdr_mem = hdr_mz->iova;
568 } else if (queue_type == VTNET_CQ) {
572 cvq->virtio_net_hdr_mz = hdr_mz;
573 cvq->virtio_net_hdr_mem = hdr_mz->iova;
574 memset(cvq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
579 /* For virtio_user case (that is when hw->virtio_user_dev is not NULL),
580 * we use virtual address. And we need properly set _offset_, please see
581 * VIRTIO_MBUF_DATA_DMA_ADDR in virtqueue.h for more information.
583 if (hw->bus_type == VIRTIO_BUS_PCI_LEGACY || hw->bus_type == VIRTIO_BUS_PCI_MODERN) {
584 vq->offset = offsetof(struct rte_mbuf, buf_iova);
585 } else if (hw->bus_type == VIRTIO_BUS_USER) {
586 vq->vq_ring_mem = (uintptr_t)mz->addr;
587 vq->offset = offsetof(struct rte_mbuf, buf_addr);
588 if (queue_type == VTNET_TQ)
589 txvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
590 else if (queue_type == VTNET_CQ)
591 cvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
594 if (queue_type == VTNET_TQ) {
595 struct virtio_tx_region *txr;
599 memset(txr, 0, vq_size * sizeof(*txr));
600 for (i = 0; i < vq_size; i++) {
601 /* first indirect descriptor is always the tx header */
602 if (!vtpci_packed_queue(hw)) {
603 struct vring_desc *start_dp = txr[i].tx_indir;
604 vring_desc_init_split(start_dp,
605 RTE_DIM(txr[i].tx_indir));
606 start_dp->addr = txvq->virtio_net_hdr_mem
608 + offsetof(struct virtio_tx_region,
610 start_dp->len = hw->vtnet_hdr_size;
611 start_dp->flags = VRING_DESC_F_NEXT;
613 struct vring_packed_desc *start_dp =
614 txr[i].tx_packed_indir;
615 vring_desc_init_indirect_packed(start_dp,
616 RTE_DIM(txr[i].tx_packed_indir));
617 start_dp->addr = txvq->virtio_net_hdr_mem
619 + offsetof(struct virtio_tx_region,
621 start_dp->len = hw->vtnet_hdr_size;
626 if (VTPCI_OPS(hw)->setup_queue(hw, vq) < 0) {
627 PMD_INIT_LOG(ERR, "setup_queue failed");
635 rte_memzone_free(hdr_mz);
636 rte_memzone_free(mz);
643 virtio_free_queues(struct virtio_hw *hw)
645 uint16_t nr_vq = virtio_get_nr_vq(hw);
646 struct virtqueue *vq;
653 for (i = 0; i < nr_vq; i++) {
658 queue_type = virtio_get_queue_type(hw, i);
659 if (queue_type == VTNET_RQ) {
660 rte_free(vq->sw_ring);
661 rte_memzone_free(vq->rxq.mz);
662 } else if (queue_type == VTNET_TQ) {
663 rte_memzone_free(vq->txq.mz);
664 rte_memzone_free(vq->txq.virtio_net_hdr_mz);
666 rte_memzone_free(vq->cq.mz);
667 rte_memzone_free(vq->cq.virtio_net_hdr_mz);
679 virtio_alloc_queues(struct rte_eth_dev *dev)
681 struct virtio_hw *hw = dev->data->dev_private;
682 uint16_t nr_vq = virtio_get_nr_vq(hw);
686 hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
688 PMD_INIT_LOG(ERR, "failed to allocate vqs");
692 for (i = 0; i < nr_vq; i++) {
693 ret = virtio_init_queue(dev, i);
695 virtio_free_queues(hw);
703 static void virtio_queues_unbind_intr(struct rte_eth_dev *dev);
706 virtio_dev_close(struct rte_eth_dev *dev)
708 struct virtio_hw *hw = dev->data->dev_private;
709 struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
711 PMD_INIT_LOG(DEBUG, "virtio_dev_close");
712 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
720 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
721 VTPCI_OPS(hw)->set_config_irq(hw, VIRTIO_MSI_NO_VECTOR);
723 virtio_queues_unbind_intr(dev);
725 if (intr_conf->lsc || intr_conf->rxq) {
726 virtio_intr_disable(dev);
727 rte_intr_efd_disable(dev->intr_handle);
728 rte_free(dev->intr_handle->intr_vec);
729 dev->intr_handle->intr_vec = NULL;
733 virtio_dev_free_mbufs(dev);
734 virtio_free_queues(hw);
736 #ifdef RTE_VIRTIO_USER
737 if (hw->bus_type == VIRTIO_BUS_USER)
738 virtio_user_dev_uninit(dev->data->dev_private);
742 rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(dev));
743 if (hw->bus_type == VIRTIO_BUS_PCI_LEGACY)
744 rte_pci_ioport_unmap(VTPCI_IO(hw));
751 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
753 struct virtio_hw *hw = dev->data->dev_private;
754 struct virtio_pmd_ctrl ctrl;
758 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
759 PMD_INIT_LOG(INFO, "host does not support rx control");
763 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
764 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
768 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
770 PMD_INIT_LOG(ERR, "Failed to enable promisc");
778 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
780 struct virtio_hw *hw = dev->data->dev_private;
781 struct virtio_pmd_ctrl ctrl;
785 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
786 PMD_INIT_LOG(INFO, "host does not support rx control");
790 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
791 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
795 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
797 PMD_INIT_LOG(ERR, "Failed to disable promisc");
805 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
807 struct virtio_hw *hw = dev->data->dev_private;
808 struct virtio_pmd_ctrl ctrl;
812 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
813 PMD_INIT_LOG(INFO, "host does not support rx control");
817 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
818 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
822 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
824 PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
832 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
834 struct virtio_hw *hw = dev->data->dev_private;
835 struct virtio_pmd_ctrl ctrl;
839 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
840 PMD_INIT_LOG(INFO, "host does not support rx control");
844 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
845 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
849 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
851 PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
858 #define VLAN_TAG_LEN 4 /* 802.3ac tag (not DMA'd) */
860 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
862 struct virtio_hw *hw = dev->data->dev_private;
863 uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
865 uint32_t frame_size = mtu + ether_hdr_len;
866 uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
868 max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
870 if (mtu < RTE_ETHER_MIN_MTU || frame_size > max_frame_size) {
871 PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
872 RTE_ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
879 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
881 struct virtio_hw *hw = dev->data->dev_private;
882 struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
883 struct virtqueue *vq = rxvq->vq;
885 virtqueue_enable_intr(vq);
886 virtio_mb(hw->weak_barriers);
891 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
893 struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
894 struct virtqueue *vq = rxvq->vq;
896 virtqueue_disable_intr(vq);
901 * dev_ops for virtio, bare necessities for basic operation
903 static const struct eth_dev_ops virtio_eth_dev_ops = {
904 .dev_configure = virtio_dev_configure,
905 .dev_start = virtio_dev_start,
906 .dev_stop = virtio_dev_stop,
907 .dev_close = virtio_dev_close,
908 .promiscuous_enable = virtio_dev_promiscuous_enable,
909 .promiscuous_disable = virtio_dev_promiscuous_disable,
910 .allmulticast_enable = virtio_dev_allmulticast_enable,
911 .allmulticast_disable = virtio_dev_allmulticast_disable,
912 .mtu_set = virtio_mtu_set,
913 .dev_infos_get = virtio_dev_info_get,
914 .stats_get = virtio_dev_stats_get,
915 .xstats_get = virtio_dev_xstats_get,
916 .xstats_get_names = virtio_dev_xstats_get_names,
917 .stats_reset = virtio_dev_stats_reset,
918 .xstats_reset = virtio_dev_stats_reset,
919 .link_update = virtio_dev_link_update,
920 .vlan_offload_set = virtio_dev_vlan_offload_set,
921 .rx_queue_setup = virtio_dev_rx_queue_setup,
922 .rx_queue_intr_enable = virtio_dev_rx_queue_intr_enable,
923 .rx_queue_intr_disable = virtio_dev_rx_queue_intr_disable,
924 .rx_queue_release = virtio_dev_queue_release,
925 .tx_queue_setup = virtio_dev_tx_queue_setup,
926 .tx_queue_release = virtio_dev_queue_release,
927 /* collect stats per queue */
928 .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
929 .vlan_filter_set = virtio_vlan_filter_set,
930 .mac_addr_add = virtio_mac_addr_add,
931 .mac_addr_remove = virtio_mac_addr_remove,
932 .mac_addr_set = virtio_mac_addr_set,
936 * dev_ops for virtio-user in secondary processes, as we just have
937 * some limited supports currently.
939 const struct eth_dev_ops virtio_user_secondary_eth_dev_ops = {
940 .dev_infos_get = virtio_dev_info_get,
941 .stats_get = virtio_dev_stats_get,
942 .xstats_get = virtio_dev_xstats_get,
943 .xstats_get_names = virtio_dev_xstats_get_names,
944 .stats_reset = virtio_dev_stats_reset,
945 .xstats_reset = virtio_dev_stats_reset,
946 /* collect stats per queue */
947 .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
951 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
955 for (i = 0; i < dev->data->nb_tx_queues; i++) {
956 const struct virtnet_tx *txvq = dev->data->tx_queues[i];
960 stats->opackets += txvq->stats.packets;
961 stats->obytes += txvq->stats.bytes;
963 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
964 stats->q_opackets[i] = txvq->stats.packets;
965 stats->q_obytes[i] = txvq->stats.bytes;
969 for (i = 0; i < dev->data->nb_rx_queues; i++) {
970 const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
974 stats->ipackets += rxvq->stats.packets;
975 stats->ibytes += rxvq->stats.bytes;
976 stats->ierrors += rxvq->stats.errors;
978 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
979 stats->q_ipackets[i] = rxvq->stats.packets;
980 stats->q_ibytes[i] = rxvq->stats.bytes;
984 stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
987 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
988 struct rte_eth_xstat_name *xstats_names,
989 __rte_unused unsigned limit)
995 unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
996 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
998 if (xstats_names != NULL) {
999 /* Note: limit checked in rte_eth_xstats_names() */
1001 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1002 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1005 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1006 snprintf(xstats_names[count].name,
1007 sizeof(xstats_names[count].name),
1009 rte_virtio_rxq_stat_strings[t].name);
1014 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1015 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1018 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1019 snprintf(xstats_names[count].name,
1020 sizeof(xstats_names[count].name),
1022 rte_virtio_txq_stat_strings[t].name);
1032 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
1038 unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
1039 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
1044 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1045 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1052 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1053 xstats[count].value = *(uint64_t *)(((char *)rxvq) +
1054 rte_virtio_rxq_stat_strings[t].offset);
1055 xstats[count].id = count;
1060 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1061 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1068 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1069 xstats[count].value = *(uint64_t *)(((char *)txvq) +
1070 rte_virtio_txq_stat_strings[t].offset);
1071 xstats[count].id = count;
1080 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1082 virtio_update_stats(dev, stats);
1088 virtio_dev_stats_reset(struct rte_eth_dev *dev)
1092 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1093 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1097 txvq->stats.packets = 0;
1098 txvq->stats.bytes = 0;
1099 txvq->stats.multicast = 0;
1100 txvq->stats.broadcast = 0;
1101 memset(txvq->stats.size_bins, 0,
1102 sizeof(txvq->stats.size_bins[0]) * 8);
1105 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1106 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1110 rxvq->stats.packets = 0;
1111 rxvq->stats.bytes = 0;
1112 rxvq->stats.errors = 0;
1113 rxvq->stats.multicast = 0;
1114 rxvq->stats.broadcast = 0;
1115 memset(rxvq->stats.size_bins, 0,
1116 sizeof(rxvq->stats.size_bins[0]) * 8);
1123 virtio_set_hwaddr(struct virtio_hw *hw)
1125 vtpci_write_dev_config(hw,
1126 offsetof(struct virtio_net_config, mac),
1127 &hw->mac_addr, RTE_ETHER_ADDR_LEN);
1131 virtio_get_hwaddr(struct virtio_hw *hw)
1133 if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
1134 vtpci_read_dev_config(hw,
1135 offsetof(struct virtio_net_config, mac),
1136 &hw->mac_addr, RTE_ETHER_ADDR_LEN);
1138 rte_eth_random_addr(&hw->mac_addr[0]);
1139 virtio_set_hwaddr(hw);
1144 virtio_mac_table_set(struct virtio_hw *hw,
1145 const struct virtio_net_ctrl_mac *uc,
1146 const struct virtio_net_ctrl_mac *mc)
1148 struct virtio_pmd_ctrl ctrl;
1151 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1152 PMD_DRV_LOG(INFO, "host does not support mac table");
1156 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1157 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1159 len[0] = uc->entries * RTE_ETHER_ADDR_LEN + sizeof(uc->entries);
1160 memcpy(ctrl.data, uc, len[0]);
1162 len[1] = mc->entries * RTE_ETHER_ADDR_LEN + sizeof(mc->entries);
1163 memcpy(ctrl.data + len[0], mc, len[1]);
1165 err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1167 PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1172 virtio_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
1173 uint32_t index, uint32_t vmdq __rte_unused)
1175 struct virtio_hw *hw = dev->data->dev_private;
1176 const struct rte_ether_addr *addrs = dev->data->mac_addrs;
1178 struct virtio_net_ctrl_mac *uc, *mc;
1180 if (index >= VIRTIO_MAX_MAC_ADDRS) {
1181 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1185 uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1186 sizeof(uc->entries));
1188 mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1189 sizeof(mc->entries));
1192 for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1193 const struct rte_ether_addr *addr
1194 = (i == index) ? mac_addr : addrs + i;
1195 struct virtio_net_ctrl_mac *tbl
1196 = rte_is_multicast_ether_addr(addr) ? mc : uc;
1198 memcpy(&tbl->macs[tbl->entries++], addr, RTE_ETHER_ADDR_LEN);
1201 return virtio_mac_table_set(hw, uc, mc);
1205 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1207 struct virtio_hw *hw = dev->data->dev_private;
1208 struct rte_ether_addr *addrs = dev->data->mac_addrs;
1209 struct virtio_net_ctrl_mac *uc, *mc;
1212 if (index >= VIRTIO_MAX_MAC_ADDRS) {
1213 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1217 uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1218 sizeof(uc->entries));
1220 mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1221 sizeof(mc->entries));
1224 for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1225 struct virtio_net_ctrl_mac *tbl;
1227 if (i == index || rte_is_zero_ether_addr(addrs + i))
1230 tbl = rte_is_multicast_ether_addr(addrs + i) ? mc : uc;
1231 memcpy(&tbl->macs[tbl->entries++], addrs + i,
1232 RTE_ETHER_ADDR_LEN);
1235 virtio_mac_table_set(hw, uc, mc);
1239 virtio_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1241 struct virtio_hw *hw = dev->data->dev_private;
1243 memcpy(hw->mac_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1245 /* Use atomic update if available */
1246 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1247 struct virtio_pmd_ctrl ctrl;
1248 int len = RTE_ETHER_ADDR_LEN;
1250 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1251 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1253 memcpy(ctrl.data, mac_addr, RTE_ETHER_ADDR_LEN);
1254 return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1257 if (!vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1260 virtio_set_hwaddr(hw);
1265 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1267 struct virtio_hw *hw = dev->data->dev_private;
1268 struct virtio_pmd_ctrl ctrl;
1271 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1274 ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1275 ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1276 memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1277 len = sizeof(vlan_id);
1279 return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1283 virtio_intr_unmask(struct rte_eth_dev *dev)
1285 struct virtio_hw *hw = dev->data->dev_private;
1287 if (rte_intr_ack(dev->intr_handle) < 0)
1290 if (hw->bus_type == VIRTIO_BUS_PCI_LEGACY || hw->bus_type == VIRTIO_BUS_PCI_MODERN)
1291 hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1297 virtio_intr_enable(struct rte_eth_dev *dev)
1299 struct virtio_hw *hw = dev->data->dev_private;
1301 if (rte_intr_enable(dev->intr_handle) < 0)
1304 if (hw->bus_type == VIRTIO_BUS_PCI_LEGACY || hw->bus_type == VIRTIO_BUS_PCI_MODERN)
1305 hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1311 virtio_intr_disable(struct rte_eth_dev *dev)
1313 struct virtio_hw *hw = dev->data->dev_private;
1315 if (rte_intr_disable(dev->intr_handle) < 0)
1318 if (hw->bus_type == VIRTIO_BUS_PCI_LEGACY || hw->bus_type == VIRTIO_BUS_PCI_MODERN)
1319 hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1325 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1327 uint64_t host_features;
1329 /* Prepare guest_features: feature that driver wants to support */
1330 PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1333 /* Read device(host) feature bits */
1334 host_features = VTPCI_OPS(hw)->get_features(hw);
1335 PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1338 /* If supported, ensure MTU value is valid before acknowledging it. */
1339 if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1340 struct virtio_net_config config;
1342 vtpci_read_dev_config(hw,
1343 offsetof(struct virtio_net_config, mtu),
1344 &config.mtu, sizeof(config.mtu));
1346 if (config.mtu < RTE_ETHER_MIN_MTU)
1347 req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1351 * Negotiate features: Subset of device feature bits are written back
1352 * guest feature bits.
1354 hw->guest_features = req_features;
1355 hw->guest_features = vtpci_negotiate_features(hw, host_features);
1356 PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1357 hw->guest_features);
1359 if (hw->bus_type == VIRTIO_BUS_PCI_MODERN && !vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1361 "VIRTIO_F_VERSION_1 features is not enabled.");
1365 if (hw->bus_type == VIRTIO_BUS_PCI_MODERN || hw->bus_type == VIRTIO_BUS_USER) {
1366 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1367 if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1369 "failed to set FEATURES_OK status!");
1374 hw->req_guest_features = req_features;
1380 virtio_dev_pause(struct rte_eth_dev *dev)
1382 struct virtio_hw *hw = dev->data->dev_private;
1384 rte_spinlock_lock(&hw->state_lock);
1386 if (hw->started == 0) {
1387 /* Device is just stopped. */
1388 rte_spinlock_unlock(&hw->state_lock);
1393 * Prevent the worker threads from touching queues to avoid contention,
1394 * 1 ms should be enough for the ongoing Tx function to finish.
1401 * Recover hw state to let the worker threads continue.
1404 virtio_dev_resume(struct rte_eth_dev *dev)
1406 struct virtio_hw *hw = dev->data->dev_private;
1409 rte_spinlock_unlock(&hw->state_lock);
1413 * Should be called only after device is paused.
1416 virtio_inject_pkts(struct rte_eth_dev *dev, struct rte_mbuf **tx_pkts,
1419 struct virtio_hw *hw = dev->data->dev_private;
1420 struct virtnet_tx *txvq = dev->data->tx_queues[0];
1423 hw->inject_pkts = tx_pkts;
1424 ret = dev->tx_pkt_burst(txvq, tx_pkts, nb_pkts);
1425 hw->inject_pkts = NULL;
1431 virtio_notify_peers(struct rte_eth_dev *dev)
1433 struct virtio_hw *hw = dev->data->dev_private;
1434 struct virtnet_rx *rxvq;
1435 struct rte_mbuf *rarp_mbuf;
1437 if (!dev->data->rx_queues)
1440 rxvq = dev->data->rx_queues[0];
1444 rarp_mbuf = rte_net_make_rarp_packet(rxvq->mpool,
1445 (struct rte_ether_addr *)hw->mac_addr);
1446 if (rarp_mbuf == NULL) {
1447 PMD_DRV_LOG(ERR, "failed to make RARP packet.");
1451 /* If virtio port just stopped, no need to send RARP */
1452 if (virtio_dev_pause(dev) < 0) {
1453 rte_pktmbuf_free(rarp_mbuf);
1457 virtio_inject_pkts(dev, &rarp_mbuf, 1);
1458 virtio_dev_resume(dev);
1462 virtio_ack_link_announce(struct rte_eth_dev *dev)
1464 struct virtio_hw *hw = dev->data->dev_private;
1465 struct virtio_pmd_ctrl ctrl;
1467 ctrl.hdr.class = VIRTIO_NET_CTRL_ANNOUNCE;
1468 ctrl.hdr.cmd = VIRTIO_NET_CTRL_ANNOUNCE_ACK;
1470 virtio_send_command(hw->cvq, &ctrl, NULL, 0);
1474 * Process virtio config changed interrupt. Call the callback
1475 * if link state changed, generate gratuitous RARP packet if
1476 * the status indicates an ANNOUNCE.
1479 virtio_interrupt_handler(void *param)
1481 struct rte_eth_dev *dev = param;
1482 struct virtio_hw *hw = dev->data->dev_private;
1486 /* Read interrupt status which clears interrupt */
1487 isr = vtpci_isr(hw);
1488 PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1490 if (virtio_intr_unmask(dev) < 0)
1491 PMD_DRV_LOG(ERR, "interrupt enable failed");
1493 if (isr & VIRTIO_PCI_ISR_CONFIG) {
1494 if (virtio_dev_link_update(dev, 0) == 0)
1495 rte_eth_dev_callback_process(dev,
1496 RTE_ETH_EVENT_INTR_LSC,
1499 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1500 vtpci_read_dev_config(hw,
1501 offsetof(struct virtio_net_config, status),
1502 &status, sizeof(status));
1503 if (status & VIRTIO_NET_S_ANNOUNCE) {
1504 virtio_notify_peers(dev);
1506 virtio_ack_link_announce(dev);
1512 /* set rx and tx handlers according to what is supported */
1514 set_rxtx_funcs(struct rte_eth_dev *eth_dev)
1516 struct virtio_hw *hw = eth_dev->data->dev_private;
1518 eth_dev->tx_pkt_prepare = virtio_xmit_pkts_prepare;
1519 if (vtpci_packed_queue(hw)) {
1521 "virtio: using packed ring %s Tx path on port %u",
1522 hw->use_vec_tx ? "vectorized" : "standard",
1523 eth_dev->data->port_id);
1525 eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed_vec;
1527 eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed;
1529 if (hw->use_inorder_tx) {
1530 PMD_INIT_LOG(INFO, "virtio: using inorder Tx path on port %u",
1531 eth_dev->data->port_id);
1532 eth_dev->tx_pkt_burst = virtio_xmit_pkts_inorder;
1534 PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
1535 eth_dev->data->port_id);
1536 eth_dev->tx_pkt_burst = virtio_xmit_pkts;
1540 if (vtpci_packed_queue(hw)) {
1541 if (hw->use_vec_rx) {
1543 "virtio: using packed ring vectorized Rx path on port %u",
1544 eth_dev->data->port_id);
1545 eth_dev->rx_pkt_burst =
1546 &virtio_recv_pkts_packed_vec;
1547 } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1549 "virtio: using packed ring mergeable buffer Rx path on port %u",
1550 eth_dev->data->port_id);
1551 eth_dev->rx_pkt_burst =
1552 &virtio_recv_mergeable_pkts_packed;
1555 "virtio: using packed ring standard Rx path on port %u",
1556 eth_dev->data->port_id);
1557 eth_dev->rx_pkt_burst = &virtio_recv_pkts_packed;
1560 if (hw->use_vec_rx) {
1561 PMD_INIT_LOG(INFO, "virtio: using vectorized Rx path on port %u",
1562 eth_dev->data->port_id);
1563 eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1564 } else if (hw->use_inorder_rx) {
1566 "virtio: using inorder Rx path on port %u",
1567 eth_dev->data->port_id);
1568 eth_dev->rx_pkt_burst = &virtio_recv_pkts_inorder;
1569 } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1571 "virtio: using mergeable buffer Rx path on port %u",
1572 eth_dev->data->port_id);
1573 eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1575 PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
1576 eth_dev->data->port_id);
1577 eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1583 /* Only support 1:1 queue/interrupt mapping so far.
1584 * TODO: support n:1 queue/interrupt mapping when there are limited number of
1585 * interrupt vectors (<N+1).
1588 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1591 struct virtio_hw *hw = dev->data->dev_private;
1593 PMD_INIT_LOG(INFO, "queue/interrupt binding");
1594 for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1595 dev->intr_handle->intr_vec[i] = i + 1;
1596 if (VTPCI_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1597 VIRTIO_MSI_NO_VECTOR) {
1598 PMD_DRV_LOG(ERR, "failed to set queue vector");
1607 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1610 struct virtio_hw *hw = dev->data->dev_private;
1612 PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1613 for (i = 0; i < dev->data->nb_rx_queues; ++i)
1614 VTPCI_OPS(hw)->set_queue_irq(hw,
1615 hw->vqs[i * VTNET_CQ],
1616 VIRTIO_MSI_NO_VECTOR);
1620 virtio_configure_intr(struct rte_eth_dev *dev)
1622 struct virtio_hw *hw = dev->data->dev_private;
1624 if (!rte_intr_cap_multiple(dev->intr_handle)) {
1625 PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1629 if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1630 PMD_INIT_LOG(ERR, "Fail to create eventfd");
1634 if (!dev->intr_handle->intr_vec) {
1635 dev->intr_handle->intr_vec =
1636 rte_zmalloc("intr_vec",
1637 hw->max_queue_pairs * sizeof(int), 0);
1638 if (!dev->intr_handle->intr_vec) {
1639 PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1640 hw->max_queue_pairs);
1645 /* Re-register callback to update max_intr */
1646 rte_intr_callback_unregister(dev->intr_handle,
1647 virtio_interrupt_handler,
1649 rte_intr_callback_register(dev->intr_handle,
1650 virtio_interrupt_handler,
1653 /* DO NOT try to remove this! This function will enable msix, or QEMU
1654 * will encounter SIGSEGV when DRIVER_OK is sent.
1655 * And for legacy devices, this should be done before queue/vec binding
1656 * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1657 * (22) will be ignored.
1659 if (virtio_intr_enable(dev) < 0) {
1660 PMD_DRV_LOG(ERR, "interrupt enable failed");
1664 if (virtio_queues_bind_intr(dev) < 0) {
1665 PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1671 #define DUPLEX_UNKNOWN 0xff
1672 /* reset device and renegotiate features if needed */
1674 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1676 struct virtio_hw *hw = eth_dev->data->dev_private;
1677 struct virtio_net_config *config;
1678 struct virtio_net_config local_config;
1679 struct rte_pci_device *pci_dev = NULL;
1682 /* Reset the device although not necessary at startup */
1686 virtio_dev_free_mbufs(eth_dev);
1687 virtio_free_queues(hw);
1690 /* Tell the host we've noticed this device. */
1691 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1693 /* Tell the host we've known how to drive the device. */
1694 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1695 if (virtio_negotiate_features(hw, req_features) < 0)
1698 hw->weak_barriers = !vtpci_with_feature(hw, VIRTIO_F_ORDER_PLATFORM);
1700 if (hw->bus_type == VIRTIO_BUS_PCI_LEGACY || hw->bus_type == VIRTIO_BUS_PCI_MODERN)
1701 pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1703 /* If host does not support both status and MSI-X then disable LSC */
1704 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS) &&
1705 hw->use_msix != VIRTIO_MSIX_NONE)
1706 eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1708 eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1710 eth_dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
1712 /* Setting up rx_header size for the device */
1713 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1714 vtpci_with_feature(hw, VIRTIO_F_VERSION_1) ||
1715 vtpci_with_feature(hw, VIRTIO_F_RING_PACKED))
1716 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1718 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1720 /* Copy the permanent MAC address to: virtio_hw */
1721 virtio_get_hwaddr(hw);
1722 rte_ether_addr_copy((struct rte_ether_addr *)hw->mac_addr,
1723 ð_dev->data->mac_addrs[0]);
1725 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1726 hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1727 hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1729 if (hw->speed == ETH_SPEED_NUM_UNKNOWN) {
1730 if (vtpci_with_feature(hw, VIRTIO_NET_F_SPEED_DUPLEX)) {
1731 config = &local_config;
1732 vtpci_read_dev_config(hw,
1733 offsetof(struct virtio_net_config, speed),
1734 &config->speed, sizeof(config->speed));
1735 vtpci_read_dev_config(hw,
1736 offsetof(struct virtio_net_config, duplex),
1737 &config->duplex, sizeof(config->duplex));
1738 hw->speed = config->speed;
1739 hw->duplex = config->duplex;
1742 if (hw->duplex == DUPLEX_UNKNOWN)
1743 hw->duplex = ETH_LINK_FULL_DUPLEX;
1744 PMD_INIT_LOG(DEBUG, "link speed = %d, duplex = %d",
1745 hw->speed, hw->duplex);
1746 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1747 config = &local_config;
1749 vtpci_read_dev_config(hw,
1750 offsetof(struct virtio_net_config, mac),
1751 &config->mac, sizeof(config->mac));
1753 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1754 vtpci_read_dev_config(hw,
1755 offsetof(struct virtio_net_config, status),
1756 &config->status, sizeof(config->status));
1759 "VIRTIO_NET_F_STATUS is not supported");
1763 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1764 vtpci_read_dev_config(hw,
1765 offsetof(struct virtio_net_config, max_virtqueue_pairs),
1766 &config->max_virtqueue_pairs,
1767 sizeof(config->max_virtqueue_pairs));
1770 "VIRTIO_NET_F_MQ is not supported");
1771 config->max_virtqueue_pairs = 1;
1774 hw->max_queue_pairs = config->max_virtqueue_pairs;
1776 if (vtpci_with_feature(hw, VIRTIO_NET_F_MTU)) {
1777 vtpci_read_dev_config(hw,
1778 offsetof(struct virtio_net_config, mtu),
1780 sizeof(config->mtu));
1783 * MTU value has already been checked at negotiation
1784 * time, but check again in case it has changed since
1785 * then, which should not happen.
1787 if (config->mtu < RTE_ETHER_MIN_MTU) {
1788 PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1793 hw->max_mtu = config->mtu;
1794 /* Set initial MTU to maximum one supported by vhost */
1795 eth_dev->data->mtu = config->mtu;
1798 hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1799 VLAN_TAG_LEN - hw->vtnet_hdr_size;
1802 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1803 config->max_virtqueue_pairs);
1804 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1806 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1807 config->mac[0], config->mac[1],
1808 config->mac[2], config->mac[3],
1809 config->mac[4], config->mac[5]);
1811 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1812 hw->max_queue_pairs = 1;
1813 hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1814 VLAN_TAG_LEN - hw->vtnet_hdr_size;
1817 ret = virtio_alloc_queues(eth_dev);
1821 if (eth_dev->data->dev_conf.intr_conf.rxq) {
1822 if (virtio_configure_intr(eth_dev) < 0) {
1823 PMD_INIT_LOG(ERR, "failed to configure interrupt");
1824 virtio_free_queues(hw);
1829 vtpci_reinit_complete(hw);
1832 PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1833 eth_dev->data->port_id, pci_dev->id.vendor_id,
1834 pci_dev->id.device_id);
1840 * Remap the PCI device again (IO port map for legacy device and
1841 * memory map for modern device), so that the secondary process
1842 * could have the PCI initiated correctly.
1845 virtio_remap_pci(struct rte_pci_device *pci_dev, struct virtio_hw *hw)
1847 if (hw->bus_type == VIRTIO_BUS_PCI_MODERN) {
1849 * We don't have to re-parse the PCI config space, since
1850 * rte_pci_map_device() makes sure the mapped address
1851 * in secondary process would equal to the one mapped in
1852 * the primary process: error will be returned if that
1853 * requirement is not met.
1855 * That said, we could simply reuse all cap pointers
1856 * (such as dev_cfg, common_cfg, etc.) parsed from the
1857 * primary process, which is stored in shared memory.
1859 if (rte_pci_map_device(pci_dev)) {
1860 PMD_INIT_LOG(DEBUG, "failed to map pci device!");
1863 } else if (hw->bus_type == VIRTIO_BUS_PCI_LEGACY) {
1864 if (rte_pci_ioport_map(pci_dev, 0, VTPCI_IO(hw)) < 0)
1872 virtio_set_vtpci_ops(struct virtio_hw *hw)
1874 #ifdef RTE_VIRTIO_USER
1875 if (hw->bus_type == VIRTIO_BUS_USER)
1876 VTPCI_OPS(hw) = &virtio_user_ops;
1879 if (hw->bus_type == VIRTIO_BUS_PCI_MODERN)
1880 VTPCI_OPS(hw) = &modern_ops;
1881 else if (hw->bus_type == VIRTIO_BUS_PCI_LEGACY)
1882 VTPCI_OPS(hw) = &legacy_ops;
1888 * This function is based on probe() function in virtio_pci.c
1889 * It returns 0 on success.
1892 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1894 struct virtio_hw *hw = eth_dev->data->dev_private;
1895 uint32_t speed = ETH_SPEED_NUM_UNKNOWN;
1899 if (sizeof(struct virtio_net_hdr_mrg_rxbuf) > RTE_PKTMBUF_HEADROOM) {
1901 "Not sufficient headroom required = %d, avail = %d",
1902 (int)sizeof(struct virtio_net_hdr_mrg_rxbuf),
1903 RTE_PKTMBUF_HEADROOM);
1908 eth_dev->dev_ops = &virtio_eth_dev_ops;
1909 eth_dev->rx_descriptor_done = virtio_dev_rx_queue_done;
1911 if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1912 if (hw->bus_type != VIRTIO_BUS_USER) {
1913 ret = virtio_remap_pci(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1918 virtio_set_vtpci_ops(hw);
1919 set_rxtx_funcs(eth_dev);
1923 ret = virtio_dev_devargs_parse(eth_dev->device->devargs, &speed, &vectorized);
1928 /* Allocate memory for storing MAC addresses */
1929 eth_dev->data->mac_addrs = rte_zmalloc("virtio",
1930 VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN, 0);
1931 if (eth_dev->data->mac_addrs == NULL) {
1933 "Failed to allocate %d bytes needed to store MAC addresses",
1934 VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN);
1938 hw->port_id = eth_dev->data->port_id;
1939 /* For virtio_user case the hw->virtio_user_dev is populated by
1940 * virtio_user_eth_dev_alloc() before eth_virtio_dev_init() is called.
1942 if (hw->bus_type != VIRTIO_BUS_USER) {
1943 ret = vtpci_init(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1945 goto err_vtpci_init;
1948 rte_spinlock_init(&hw->state_lock);
1950 /* reset device and negotiate default features */
1951 ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1953 goto err_virtio_init;
1956 if (!vtpci_packed_queue(hw)) {
1959 #if defined(CC_AVX512_SUPPORT) || defined(RTE_ARCH_ARM)
1964 "building environment do not support packed ring vectorized");
1974 if (hw->bus_type == VIRTIO_BUS_PCI_MODERN || hw->bus_type == VIRTIO_BUS_PCI_LEGACY) {
1975 rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(eth_dev));
1976 if (hw->bus_type == VIRTIO_BUS_PCI_LEGACY)
1977 rte_pci_ioport_unmap(VTPCI_IO(hw));
1980 rte_free(eth_dev->data->mac_addrs);
1981 eth_dev->data->mac_addrs = NULL;
1986 virtio_dev_speed_capa_get(uint32_t speed)
1989 case ETH_SPEED_NUM_10G:
1990 return ETH_LINK_SPEED_10G;
1991 case ETH_SPEED_NUM_20G:
1992 return ETH_LINK_SPEED_20G;
1993 case ETH_SPEED_NUM_25G:
1994 return ETH_LINK_SPEED_25G;
1995 case ETH_SPEED_NUM_40G:
1996 return ETH_LINK_SPEED_40G;
1997 case ETH_SPEED_NUM_50G:
1998 return ETH_LINK_SPEED_50G;
1999 case ETH_SPEED_NUM_56G:
2000 return ETH_LINK_SPEED_56G;
2001 case ETH_SPEED_NUM_100G:
2002 return ETH_LINK_SPEED_100G;
2003 case ETH_SPEED_NUM_200G:
2004 return ETH_LINK_SPEED_200G;
2010 static int vectorized_check_handler(__rte_unused const char *key,
2011 const char *value, void *ret_val)
2013 if (strcmp(value, "1") == 0)
2014 *(int *)ret_val = 1;
2016 *(int *)ret_val = 0;
2021 #define VIRTIO_ARG_SPEED "speed"
2022 #define VIRTIO_ARG_VECTORIZED "vectorized"
2025 link_speed_handler(const char *key __rte_unused,
2026 const char *value, void *ret_val)
2029 if (!value || !ret_val)
2031 val = strtoul(value, NULL, 0);
2032 /* validate input */
2033 if (virtio_dev_speed_capa_get(val) == 0)
2035 *(uint32_t *)ret_val = val;
2042 virtio_dev_devargs_parse(struct rte_devargs *devargs, uint32_t *speed, int *vectorized)
2044 struct rte_kvargs *kvlist;
2047 if (devargs == NULL)
2050 kvlist = rte_kvargs_parse(devargs->args, NULL);
2051 if (kvlist == NULL) {
2052 PMD_INIT_LOG(ERR, "error when parsing param");
2056 if (speed && rte_kvargs_count(kvlist, VIRTIO_ARG_SPEED) == 1) {
2057 ret = rte_kvargs_process(kvlist,
2059 link_speed_handler, speed);
2061 PMD_INIT_LOG(ERR, "Failed to parse %s",
2068 rte_kvargs_count(kvlist, VIRTIO_ARG_VECTORIZED) == 1) {
2069 ret = rte_kvargs_process(kvlist,
2070 VIRTIO_ARG_VECTORIZED,
2071 vectorized_check_handler, vectorized);
2073 PMD_INIT_LOG(ERR, "Failed to parse %s",
2074 VIRTIO_ARG_VECTORIZED);
2080 rte_kvargs_free(kvlist);
2085 rx_offload_enabled(struct virtio_hw *hw)
2087 return vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM) ||
2088 vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2089 vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6);
2093 tx_offload_enabled(struct virtio_hw *hw)
2095 return vtpci_with_feature(hw, VIRTIO_NET_F_CSUM) ||
2096 vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO4) ||
2097 vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO6);
2101 * Configure virtio device
2102 * It returns 0 on success.
2105 virtio_dev_configure(struct rte_eth_dev *dev)
2107 const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2108 const struct rte_eth_txmode *txmode = &dev->data->dev_conf.txmode;
2109 struct virtio_hw *hw = dev->data->dev_private;
2110 uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
2112 uint64_t rx_offloads = rxmode->offloads;
2113 uint64_t tx_offloads = txmode->offloads;
2114 uint64_t req_features;
2117 PMD_INIT_LOG(DEBUG, "configure");
2118 req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
2120 if (rxmode->mq_mode != ETH_MQ_RX_NONE) {
2122 "Unsupported Rx multi queue mode %d",
2127 if (txmode->mq_mode != ETH_MQ_TX_NONE) {
2129 "Unsupported Tx multi queue mode %d",
2134 if (dev->data->dev_conf.intr_conf.rxq) {
2135 ret = virtio_init_device(dev, hw->req_guest_features);
2140 if (rxmode->max_rx_pkt_len > hw->max_mtu + ether_hdr_len)
2141 req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
2143 if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2144 DEV_RX_OFFLOAD_TCP_CKSUM))
2145 req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
2147 if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO)
2149 (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2150 (1ULL << VIRTIO_NET_F_GUEST_TSO6);
2152 if (tx_offloads & (DEV_TX_OFFLOAD_UDP_CKSUM |
2153 DEV_TX_OFFLOAD_TCP_CKSUM))
2154 req_features |= (1ULL << VIRTIO_NET_F_CSUM);
2156 if (tx_offloads & DEV_TX_OFFLOAD_TCP_TSO)
2158 (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2159 (1ULL << VIRTIO_NET_F_HOST_TSO6);
2161 /* if request features changed, reinit the device */
2162 if (req_features != hw->req_guest_features) {
2163 ret = virtio_init_device(dev, req_features);
2168 if ((rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2169 DEV_RX_OFFLOAD_TCP_CKSUM)) &&
2170 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
2172 "rx checksum not available on this host");
2176 if ((rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) &&
2177 (!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2178 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6))) {
2180 "Large Receive Offload not available on this host");
2184 /* start control queue */
2185 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
2186 virtio_dev_cq_start(dev);
2188 if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
2191 if ((rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
2192 && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2194 "vlan filtering not available on this host");
2198 hw->has_tx_offload = tx_offload_enabled(hw);
2199 hw->has_rx_offload = rx_offload_enabled(hw);
2201 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2202 /* Enable vector (0) for Link State Intrerrupt */
2203 if (VTPCI_OPS(hw)->set_config_irq(hw, 0) ==
2204 VIRTIO_MSI_NO_VECTOR) {
2205 PMD_DRV_LOG(ERR, "failed to set config vector");
2209 if (vtpci_packed_queue(hw)) {
2210 #if defined(RTE_ARCH_X86_64) && defined(CC_AVX512_SUPPORT)
2211 if ((hw->use_vec_rx || hw->use_vec_tx) &&
2212 (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX512F) ||
2213 !vtpci_with_feature(hw, VIRTIO_F_IN_ORDER) ||
2214 !vtpci_with_feature(hw, VIRTIO_F_VERSION_1) ||
2215 rte_vect_get_max_simd_bitwidth() < RTE_VECT_SIMD_512)) {
2217 "disabled packed ring vectorized path for requirements not met");
2221 #elif defined(RTE_ARCH_ARM)
2222 if ((hw->use_vec_rx || hw->use_vec_tx) &&
2223 (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON) ||
2224 !vtpci_with_feature(hw, VIRTIO_F_IN_ORDER) ||
2225 !vtpci_with_feature(hw, VIRTIO_F_VERSION_1) ||
2226 rte_vect_get_max_simd_bitwidth() < RTE_VECT_SIMD_128)) {
2228 "disabled packed ring vectorized path for requirements not met");
2237 if (hw->use_vec_rx) {
2238 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2240 "disabled packed ring vectorized rx for mrg_rxbuf enabled");
2244 if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) {
2246 "disabled packed ring vectorized rx for TCP_LRO enabled");
2251 if (vtpci_with_feature(hw, VIRTIO_F_IN_ORDER)) {
2252 hw->use_inorder_tx = 1;
2253 hw->use_inorder_rx = 1;
2257 if (hw->use_vec_rx) {
2258 #if defined RTE_ARCH_ARM
2259 if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
2261 "disabled split ring vectorized path for requirement not met");
2265 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2267 "disabled split ring vectorized rx for mrg_rxbuf enabled");
2271 if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2272 DEV_RX_OFFLOAD_TCP_CKSUM |
2273 DEV_RX_OFFLOAD_TCP_LRO |
2274 DEV_RX_OFFLOAD_VLAN_STRIP)) {
2276 "disabled split ring vectorized rx for offloading enabled");
2280 if (rte_vect_get_max_simd_bitwidth() < RTE_VECT_SIMD_128) {
2282 "disabled split ring vectorized rx, max SIMD bitwidth too low");
2293 virtio_dev_start(struct rte_eth_dev *dev)
2295 uint16_t nb_queues, i;
2296 struct virtnet_rx *rxvq;
2297 struct virtnet_tx *txvq __rte_unused;
2298 struct virtio_hw *hw = dev->data->dev_private;
2301 /* Finish the initialization of the queues */
2302 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2303 ret = virtio_dev_rx_queue_setup_finish(dev, i);
2307 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2308 ret = virtio_dev_tx_queue_setup_finish(dev, i);
2313 /* check if lsc interrupt feature is enabled */
2314 if (dev->data->dev_conf.intr_conf.lsc) {
2315 if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
2316 PMD_DRV_LOG(ERR, "link status not supported by host");
2321 /* Enable uio/vfio intr/eventfd mapping: althrough we already did that
2322 * in device configure, but it could be unmapped when device is
2325 if (dev->data->dev_conf.intr_conf.lsc ||
2326 dev->data->dev_conf.intr_conf.rxq) {
2327 virtio_intr_disable(dev);
2329 /* Setup interrupt callback */
2330 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2331 rte_intr_callback_register(dev->intr_handle,
2332 virtio_interrupt_handler,
2335 if (virtio_intr_enable(dev) < 0) {
2336 PMD_DRV_LOG(ERR, "interrupt enable failed");
2341 /*Notify the backend
2342 *Otherwise the tap backend might already stop its queue due to fullness.
2343 *vhost backend will have no chance to be waked up
2345 nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
2346 if (hw->max_queue_pairs > 1) {
2347 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
2351 PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
2353 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2354 rxvq = dev->data->rx_queues[i];
2355 /* Flush the old packets */
2356 virtqueue_rxvq_flush(rxvq->vq);
2357 virtqueue_notify(rxvq->vq);
2360 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2361 txvq = dev->data->tx_queues[i];
2362 virtqueue_notify(txvq->vq);
2365 PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
2367 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2368 rxvq = dev->data->rx_queues[i];
2369 VIRTQUEUE_DUMP(rxvq->vq);
2372 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2373 txvq = dev->data->tx_queues[i];
2374 VIRTQUEUE_DUMP(txvq->vq);
2377 set_rxtx_funcs(dev);
2380 /* Initialize Link state */
2381 virtio_dev_link_update(dev, 0);
2386 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
2388 struct virtio_hw *hw = dev->data->dev_private;
2389 uint16_t nr_vq = virtio_get_nr_vq(hw);
2390 const char *type __rte_unused;
2391 unsigned int i, mbuf_num = 0;
2392 struct virtqueue *vq;
2393 struct rte_mbuf *buf;
2396 if (hw->vqs == NULL)
2399 for (i = 0; i < nr_vq; i++) {
2404 queue_type = virtio_get_queue_type(hw, i);
2405 if (queue_type == VTNET_RQ)
2407 else if (queue_type == VTNET_TQ)
2413 "Before freeing %s[%d] used and unused buf",
2417 while ((buf = virtqueue_detach_unused(vq)) != NULL) {
2418 rte_pktmbuf_free(buf);
2423 "After freeing %s[%d] used and unused buf",
2428 PMD_INIT_LOG(DEBUG, "%d mbufs freed", mbuf_num);
2432 * Stop device: disable interrupt and mark link down
2435 virtio_dev_stop(struct rte_eth_dev *dev)
2437 struct virtio_hw *hw = dev->data->dev_private;
2438 struct rte_eth_link link;
2439 struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
2441 PMD_INIT_LOG(DEBUG, "stop");
2442 dev->data->dev_started = 0;
2444 rte_spinlock_lock(&hw->state_lock);
2447 hw->started = false;
2449 if (intr_conf->lsc || intr_conf->rxq) {
2450 virtio_intr_disable(dev);
2452 /* Reset interrupt callback */
2453 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
2454 rte_intr_callback_unregister(dev->intr_handle,
2455 virtio_interrupt_handler,
2460 memset(&link, 0, sizeof(link));
2461 rte_eth_linkstatus_set(dev, &link);
2463 rte_spinlock_unlock(&hw->state_lock);
2469 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
2471 struct rte_eth_link link;
2473 struct virtio_hw *hw = dev->data->dev_private;
2475 memset(&link, 0, sizeof(link));
2476 link.link_duplex = hw->duplex;
2477 link.link_speed = hw->speed;
2478 link.link_autoneg = ETH_LINK_AUTONEG;
2481 link.link_status = ETH_LINK_DOWN;
2482 link.link_speed = ETH_SPEED_NUM_NONE;
2483 } else if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
2484 PMD_INIT_LOG(DEBUG, "Get link status from hw");
2485 vtpci_read_dev_config(hw,
2486 offsetof(struct virtio_net_config, status),
2487 &status, sizeof(status));
2488 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
2489 link.link_status = ETH_LINK_DOWN;
2490 link.link_speed = ETH_SPEED_NUM_NONE;
2491 PMD_INIT_LOG(DEBUG, "Port %d is down",
2492 dev->data->port_id);
2494 link.link_status = ETH_LINK_UP;
2495 PMD_INIT_LOG(DEBUG, "Port %d is up",
2496 dev->data->port_id);
2499 link.link_status = ETH_LINK_UP;
2502 return rte_eth_linkstatus_set(dev, &link);
2506 virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
2508 const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2509 struct virtio_hw *hw = dev->data->dev_private;
2510 uint64_t offloads = rxmode->offloads;
2512 if (mask & ETH_VLAN_FILTER_MASK) {
2513 if ((offloads & DEV_RX_OFFLOAD_VLAN_FILTER) &&
2514 !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2517 "vlan filtering not available on this host");
2523 if (mask & ETH_VLAN_STRIP_MASK)
2524 hw->vlan_strip = !!(offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
2530 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2532 uint64_t tso_mask, host_features;
2533 struct virtio_hw *hw = dev->data->dev_private;
2534 dev_info->speed_capa = virtio_dev_speed_capa_get(hw->speed);
2536 dev_info->max_rx_queues =
2537 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
2538 dev_info->max_tx_queues =
2539 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
2540 dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
2541 dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
2542 dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
2544 host_features = VTPCI_OPS(hw)->get_features(hw);
2545 dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
2546 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_JUMBO_FRAME;
2547 if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
2548 dev_info->rx_offload_capa |=
2549 DEV_RX_OFFLOAD_TCP_CKSUM |
2550 DEV_RX_OFFLOAD_UDP_CKSUM;
2552 if (host_features & (1ULL << VIRTIO_NET_F_CTRL_VLAN))
2553 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_VLAN_FILTER;
2554 tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2555 (1ULL << VIRTIO_NET_F_GUEST_TSO6);
2556 if ((host_features & tso_mask) == tso_mask)
2557 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
2559 dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
2560 DEV_TX_OFFLOAD_VLAN_INSERT;
2561 if (host_features & (1ULL << VIRTIO_NET_F_CSUM)) {
2562 dev_info->tx_offload_capa |=
2563 DEV_TX_OFFLOAD_UDP_CKSUM |
2564 DEV_TX_OFFLOAD_TCP_CKSUM;
2566 tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2567 (1ULL << VIRTIO_NET_F_HOST_TSO6);
2568 if ((host_features & tso_mask) == tso_mask)
2569 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
2575 * It enables testpmd to collect per queue stats.
2578 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
2579 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
2580 __rte_unused uint8_t is_rx)
2585 RTE_LOG_REGISTER(virtio_logtype_init, pmd.net.virtio.init, NOTICE);
2586 RTE_LOG_REGISTER(virtio_logtype_driver, pmd.net.virtio.driver, NOTICE);