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 if (queue_type == VTNET_TQ) {
580 struct virtio_tx_region *txr;
584 memset(txr, 0, vq_size * sizeof(*txr));
585 for (i = 0; i < vq_size; i++) {
586 /* first indirect descriptor is always the tx header */
587 if (!vtpci_packed_queue(hw)) {
588 struct vring_desc *start_dp = txr[i].tx_indir;
589 vring_desc_init_split(start_dp,
590 RTE_DIM(txr[i].tx_indir));
591 start_dp->addr = txvq->virtio_net_hdr_mem
593 + offsetof(struct virtio_tx_region,
595 start_dp->len = hw->vtnet_hdr_size;
596 start_dp->flags = VRING_DESC_F_NEXT;
598 struct vring_packed_desc *start_dp =
599 txr[i].tx_packed_indir;
600 vring_desc_init_indirect_packed(start_dp,
601 RTE_DIM(txr[i].tx_packed_indir));
602 start_dp->addr = txvq->virtio_net_hdr_mem
604 + offsetof(struct virtio_tx_region,
606 start_dp->len = hw->vtnet_hdr_size;
611 if (VTPCI_OPS(hw)->setup_queue(hw, vq) < 0) {
612 PMD_INIT_LOG(ERR, "setup_queue failed");
620 rte_memzone_free(hdr_mz);
621 rte_memzone_free(mz);
628 virtio_free_queues(struct virtio_hw *hw)
630 uint16_t nr_vq = virtio_get_nr_vq(hw);
631 struct virtqueue *vq;
638 for (i = 0; i < nr_vq; i++) {
643 queue_type = virtio_get_queue_type(hw, i);
644 if (queue_type == VTNET_RQ) {
645 rte_free(vq->sw_ring);
646 rte_memzone_free(vq->rxq.mz);
647 } else if (queue_type == VTNET_TQ) {
648 rte_memzone_free(vq->txq.mz);
649 rte_memzone_free(vq->txq.virtio_net_hdr_mz);
651 rte_memzone_free(vq->cq.mz);
652 rte_memzone_free(vq->cq.virtio_net_hdr_mz);
664 virtio_alloc_queues(struct rte_eth_dev *dev)
666 struct virtio_hw *hw = dev->data->dev_private;
667 uint16_t nr_vq = virtio_get_nr_vq(hw);
671 hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
673 PMD_INIT_LOG(ERR, "failed to allocate vqs");
677 for (i = 0; i < nr_vq; i++) {
678 ret = virtio_init_queue(dev, i);
680 virtio_free_queues(hw);
688 static void virtio_queues_unbind_intr(struct rte_eth_dev *dev);
691 virtio_dev_close(struct rte_eth_dev *dev)
693 struct virtio_hw *hw = dev->data->dev_private;
694 struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
696 PMD_INIT_LOG(DEBUG, "virtio_dev_close");
697 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
705 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
706 VTPCI_OPS(hw)->set_config_irq(hw, VIRTIO_MSI_NO_VECTOR);
708 virtio_queues_unbind_intr(dev);
710 if (intr_conf->lsc || intr_conf->rxq) {
711 virtio_intr_disable(dev);
712 rte_intr_efd_disable(dev->intr_handle);
713 rte_free(dev->intr_handle->intr_vec);
714 dev->intr_handle->intr_vec = NULL;
718 virtio_dev_free_mbufs(dev);
719 virtio_free_queues(hw);
721 #ifdef RTE_VIRTIO_USER
722 if (hw->bus_type == VIRTIO_BUS_USER)
723 virtio_user_dev_uninit(dev->data->dev_private);
727 rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(dev));
728 if (hw->bus_type == VIRTIO_BUS_PCI_LEGACY)
729 rte_pci_ioport_unmap(VTPCI_IO(hw));
736 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
738 struct virtio_hw *hw = dev->data->dev_private;
739 struct virtio_pmd_ctrl ctrl;
743 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
744 PMD_INIT_LOG(INFO, "host does not support rx control");
748 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
749 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
753 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
755 PMD_INIT_LOG(ERR, "Failed to enable promisc");
763 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
765 struct virtio_hw *hw = dev->data->dev_private;
766 struct virtio_pmd_ctrl ctrl;
770 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
771 PMD_INIT_LOG(INFO, "host does not support rx control");
775 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
776 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
780 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
782 PMD_INIT_LOG(ERR, "Failed to disable promisc");
790 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
792 struct virtio_hw *hw = dev->data->dev_private;
793 struct virtio_pmd_ctrl ctrl;
797 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
798 PMD_INIT_LOG(INFO, "host does not support rx control");
802 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
803 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
807 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
809 PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
817 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
819 struct virtio_hw *hw = dev->data->dev_private;
820 struct virtio_pmd_ctrl ctrl;
824 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
825 PMD_INIT_LOG(INFO, "host does not support rx control");
829 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
830 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
834 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
836 PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
843 #define VLAN_TAG_LEN 4 /* 802.3ac tag (not DMA'd) */
845 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
847 struct virtio_hw *hw = dev->data->dev_private;
848 uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
850 uint32_t frame_size = mtu + ether_hdr_len;
851 uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
853 max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
855 if (mtu < RTE_ETHER_MIN_MTU || frame_size > max_frame_size) {
856 PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
857 RTE_ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
864 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
866 struct virtio_hw *hw = dev->data->dev_private;
867 struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
868 struct virtqueue *vq = rxvq->vq;
870 virtqueue_enable_intr(vq);
871 virtio_mb(hw->weak_barriers);
876 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
878 struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
879 struct virtqueue *vq = rxvq->vq;
881 virtqueue_disable_intr(vq);
886 * dev_ops for virtio, bare necessities for basic operation
888 static const struct eth_dev_ops virtio_eth_dev_ops = {
889 .dev_configure = virtio_dev_configure,
890 .dev_start = virtio_dev_start,
891 .dev_stop = virtio_dev_stop,
892 .dev_close = virtio_dev_close,
893 .promiscuous_enable = virtio_dev_promiscuous_enable,
894 .promiscuous_disable = virtio_dev_promiscuous_disable,
895 .allmulticast_enable = virtio_dev_allmulticast_enable,
896 .allmulticast_disable = virtio_dev_allmulticast_disable,
897 .mtu_set = virtio_mtu_set,
898 .dev_infos_get = virtio_dev_info_get,
899 .stats_get = virtio_dev_stats_get,
900 .xstats_get = virtio_dev_xstats_get,
901 .xstats_get_names = virtio_dev_xstats_get_names,
902 .stats_reset = virtio_dev_stats_reset,
903 .xstats_reset = virtio_dev_stats_reset,
904 .link_update = virtio_dev_link_update,
905 .vlan_offload_set = virtio_dev_vlan_offload_set,
906 .rx_queue_setup = virtio_dev_rx_queue_setup,
907 .rx_queue_intr_enable = virtio_dev_rx_queue_intr_enable,
908 .rx_queue_intr_disable = virtio_dev_rx_queue_intr_disable,
909 .rx_queue_release = virtio_dev_queue_release,
910 .tx_queue_setup = virtio_dev_tx_queue_setup,
911 .tx_queue_release = virtio_dev_queue_release,
912 /* collect stats per queue */
913 .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
914 .vlan_filter_set = virtio_vlan_filter_set,
915 .mac_addr_add = virtio_mac_addr_add,
916 .mac_addr_remove = virtio_mac_addr_remove,
917 .mac_addr_set = virtio_mac_addr_set,
921 * dev_ops for virtio-user in secondary processes, as we just have
922 * some limited supports currently.
924 const struct eth_dev_ops virtio_user_secondary_eth_dev_ops = {
925 .dev_infos_get = virtio_dev_info_get,
926 .stats_get = virtio_dev_stats_get,
927 .xstats_get = virtio_dev_xstats_get,
928 .xstats_get_names = virtio_dev_xstats_get_names,
929 .stats_reset = virtio_dev_stats_reset,
930 .xstats_reset = virtio_dev_stats_reset,
931 /* collect stats per queue */
932 .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
936 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
940 for (i = 0; i < dev->data->nb_tx_queues; i++) {
941 const struct virtnet_tx *txvq = dev->data->tx_queues[i];
945 stats->opackets += txvq->stats.packets;
946 stats->obytes += txvq->stats.bytes;
948 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
949 stats->q_opackets[i] = txvq->stats.packets;
950 stats->q_obytes[i] = txvq->stats.bytes;
954 for (i = 0; i < dev->data->nb_rx_queues; i++) {
955 const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
959 stats->ipackets += rxvq->stats.packets;
960 stats->ibytes += rxvq->stats.bytes;
961 stats->ierrors += rxvq->stats.errors;
963 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
964 stats->q_ipackets[i] = rxvq->stats.packets;
965 stats->q_ibytes[i] = rxvq->stats.bytes;
969 stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
972 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
973 struct rte_eth_xstat_name *xstats_names,
974 __rte_unused unsigned limit)
980 unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
981 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
983 if (xstats_names != NULL) {
984 /* Note: limit checked in rte_eth_xstats_names() */
986 for (i = 0; i < dev->data->nb_rx_queues; i++) {
987 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
990 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
991 snprintf(xstats_names[count].name,
992 sizeof(xstats_names[count].name),
994 rte_virtio_rxq_stat_strings[t].name);
999 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1000 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1003 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1004 snprintf(xstats_names[count].name,
1005 sizeof(xstats_names[count].name),
1007 rte_virtio_txq_stat_strings[t].name);
1017 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
1023 unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
1024 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
1029 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1030 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1037 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1038 xstats[count].value = *(uint64_t *)(((char *)rxvq) +
1039 rte_virtio_rxq_stat_strings[t].offset);
1040 xstats[count].id = count;
1045 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1046 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1053 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1054 xstats[count].value = *(uint64_t *)(((char *)txvq) +
1055 rte_virtio_txq_stat_strings[t].offset);
1056 xstats[count].id = count;
1065 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1067 virtio_update_stats(dev, stats);
1073 virtio_dev_stats_reset(struct rte_eth_dev *dev)
1077 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1078 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1082 txvq->stats.packets = 0;
1083 txvq->stats.bytes = 0;
1084 txvq->stats.multicast = 0;
1085 txvq->stats.broadcast = 0;
1086 memset(txvq->stats.size_bins, 0,
1087 sizeof(txvq->stats.size_bins[0]) * 8);
1090 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1091 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1095 rxvq->stats.packets = 0;
1096 rxvq->stats.bytes = 0;
1097 rxvq->stats.errors = 0;
1098 rxvq->stats.multicast = 0;
1099 rxvq->stats.broadcast = 0;
1100 memset(rxvq->stats.size_bins, 0,
1101 sizeof(rxvq->stats.size_bins[0]) * 8);
1108 virtio_set_hwaddr(struct virtio_hw *hw)
1110 vtpci_write_dev_config(hw,
1111 offsetof(struct virtio_net_config, mac),
1112 &hw->mac_addr, RTE_ETHER_ADDR_LEN);
1116 virtio_get_hwaddr(struct virtio_hw *hw)
1118 if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
1119 vtpci_read_dev_config(hw,
1120 offsetof(struct virtio_net_config, mac),
1121 &hw->mac_addr, RTE_ETHER_ADDR_LEN);
1123 rte_eth_random_addr(&hw->mac_addr[0]);
1124 virtio_set_hwaddr(hw);
1129 virtio_mac_table_set(struct virtio_hw *hw,
1130 const struct virtio_net_ctrl_mac *uc,
1131 const struct virtio_net_ctrl_mac *mc)
1133 struct virtio_pmd_ctrl ctrl;
1136 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1137 PMD_DRV_LOG(INFO, "host does not support mac table");
1141 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1142 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1144 len[0] = uc->entries * RTE_ETHER_ADDR_LEN + sizeof(uc->entries);
1145 memcpy(ctrl.data, uc, len[0]);
1147 len[1] = mc->entries * RTE_ETHER_ADDR_LEN + sizeof(mc->entries);
1148 memcpy(ctrl.data + len[0], mc, len[1]);
1150 err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1152 PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1157 virtio_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
1158 uint32_t index, uint32_t vmdq __rte_unused)
1160 struct virtio_hw *hw = dev->data->dev_private;
1161 const struct rte_ether_addr *addrs = dev->data->mac_addrs;
1163 struct virtio_net_ctrl_mac *uc, *mc;
1165 if (index >= VIRTIO_MAX_MAC_ADDRS) {
1166 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1170 uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1171 sizeof(uc->entries));
1173 mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1174 sizeof(mc->entries));
1177 for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1178 const struct rte_ether_addr *addr
1179 = (i == index) ? mac_addr : addrs + i;
1180 struct virtio_net_ctrl_mac *tbl
1181 = rte_is_multicast_ether_addr(addr) ? mc : uc;
1183 memcpy(&tbl->macs[tbl->entries++], addr, RTE_ETHER_ADDR_LEN);
1186 return virtio_mac_table_set(hw, uc, mc);
1190 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1192 struct virtio_hw *hw = dev->data->dev_private;
1193 struct rte_ether_addr *addrs = dev->data->mac_addrs;
1194 struct virtio_net_ctrl_mac *uc, *mc;
1197 if (index >= VIRTIO_MAX_MAC_ADDRS) {
1198 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1202 uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1203 sizeof(uc->entries));
1205 mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1206 sizeof(mc->entries));
1209 for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1210 struct virtio_net_ctrl_mac *tbl;
1212 if (i == index || rte_is_zero_ether_addr(addrs + i))
1215 tbl = rte_is_multicast_ether_addr(addrs + i) ? mc : uc;
1216 memcpy(&tbl->macs[tbl->entries++], addrs + i,
1217 RTE_ETHER_ADDR_LEN);
1220 virtio_mac_table_set(hw, uc, mc);
1224 virtio_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1226 struct virtio_hw *hw = dev->data->dev_private;
1228 memcpy(hw->mac_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1230 /* Use atomic update if available */
1231 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1232 struct virtio_pmd_ctrl ctrl;
1233 int len = RTE_ETHER_ADDR_LEN;
1235 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1236 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1238 memcpy(ctrl.data, mac_addr, RTE_ETHER_ADDR_LEN);
1239 return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1242 if (!vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1245 virtio_set_hwaddr(hw);
1250 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1252 struct virtio_hw *hw = dev->data->dev_private;
1253 struct virtio_pmd_ctrl ctrl;
1256 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1259 ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1260 ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1261 memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1262 len = sizeof(vlan_id);
1264 return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1268 virtio_intr_unmask(struct rte_eth_dev *dev)
1270 struct virtio_hw *hw = dev->data->dev_private;
1272 if (rte_intr_ack(dev->intr_handle) < 0)
1275 if (VTPCI_OPS(hw)->intr_detect)
1276 VTPCI_OPS(hw)->intr_detect(hw);
1282 virtio_intr_enable(struct rte_eth_dev *dev)
1284 struct virtio_hw *hw = dev->data->dev_private;
1286 if (rte_intr_enable(dev->intr_handle) < 0)
1289 if (VTPCI_OPS(hw)->intr_detect)
1290 VTPCI_OPS(hw)->intr_detect(hw);
1296 virtio_intr_disable(struct rte_eth_dev *dev)
1298 struct virtio_hw *hw = dev->data->dev_private;
1300 if (rte_intr_disable(dev->intr_handle) < 0)
1303 if (VTPCI_OPS(hw)->intr_detect)
1304 VTPCI_OPS(hw)->intr_detect(hw);
1310 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1312 uint64_t host_features;
1314 /* Prepare guest_features: feature that driver wants to support */
1315 PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1318 /* Read device(host) feature bits */
1319 host_features = VTPCI_OPS(hw)->get_features(hw);
1320 PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1323 /* If supported, ensure MTU value is valid before acknowledging it. */
1324 if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1325 struct virtio_net_config config;
1327 vtpci_read_dev_config(hw,
1328 offsetof(struct virtio_net_config, mtu),
1329 &config.mtu, sizeof(config.mtu));
1331 if (config.mtu < RTE_ETHER_MIN_MTU)
1332 req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1336 * Negotiate features: Subset of device feature bits are written back
1337 * guest feature bits.
1339 hw->guest_features = req_features;
1340 hw->guest_features = vtpci_negotiate_features(hw, host_features);
1341 PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1342 hw->guest_features);
1344 if (hw->bus_type == VIRTIO_BUS_PCI_MODERN && !vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1346 "VIRTIO_F_VERSION_1 features is not enabled.");
1350 if (hw->bus_type == VIRTIO_BUS_PCI_MODERN || hw->bus_type == VIRTIO_BUS_USER) {
1351 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1352 if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1354 "failed to set FEATURES_OK status!");
1359 hw->req_guest_features = req_features;
1365 virtio_dev_pause(struct rte_eth_dev *dev)
1367 struct virtio_hw *hw = dev->data->dev_private;
1369 rte_spinlock_lock(&hw->state_lock);
1371 if (hw->started == 0) {
1372 /* Device is just stopped. */
1373 rte_spinlock_unlock(&hw->state_lock);
1378 * Prevent the worker threads from touching queues to avoid contention,
1379 * 1 ms should be enough for the ongoing Tx function to finish.
1386 * Recover hw state to let the worker threads continue.
1389 virtio_dev_resume(struct rte_eth_dev *dev)
1391 struct virtio_hw *hw = dev->data->dev_private;
1394 rte_spinlock_unlock(&hw->state_lock);
1398 * Should be called only after device is paused.
1401 virtio_inject_pkts(struct rte_eth_dev *dev, struct rte_mbuf **tx_pkts,
1404 struct virtio_hw *hw = dev->data->dev_private;
1405 struct virtnet_tx *txvq = dev->data->tx_queues[0];
1408 hw->inject_pkts = tx_pkts;
1409 ret = dev->tx_pkt_burst(txvq, tx_pkts, nb_pkts);
1410 hw->inject_pkts = NULL;
1416 virtio_notify_peers(struct rte_eth_dev *dev)
1418 struct virtio_hw *hw = dev->data->dev_private;
1419 struct virtnet_rx *rxvq;
1420 struct rte_mbuf *rarp_mbuf;
1422 if (!dev->data->rx_queues)
1425 rxvq = dev->data->rx_queues[0];
1429 rarp_mbuf = rte_net_make_rarp_packet(rxvq->mpool,
1430 (struct rte_ether_addr *)hw->mac_addr);
1431 if (rarp_mbuf == NULL) {
1432 PMD_DRV_LOG(ERR, "failed to make RARP packet.");
1436 /* If virtio port just stopped, no need to send RARP */
1437 if (virtio_dev_pause(dev) < 0) {
1438 rte_pktmbuf_free(rarp_mbuf);
1442 virtio_inject_pkts(dev, &rarp_mbuf, 1);
1443 virtio_dev_resume(dev);
1447 virtio_ack_link_announce(struct rte_eth_dev *dev)
1449 struct virtio_hw *hw = dev->data->dev_private;
1450 struct virtio_pmd_ctrl ctrl;
1452 ctrl.hdr.class = VIRTIO_NET_CTRL_ANNOUNCE;
1453 ctrl.hdr.cmd = VIRTIO_NET_CTRL_ANNOUNCE_ACK;
1455 virtio_send_command(hw->cvq, &ctrl, NULL, 0);
1459 * Process virtio config changed interrupt. Call the callback
1460 * if link state changed, generate gratuitous RARP packet if
1461 * the status indicates an ANNOUNCE.
1464 virtio_interrupt_handler(void *param)
1466 struct rte_eth_dev *dev = param;
1467 struct virtio_hw *hw = dev->data->dev_private;
1471 /* Read interrupt status which clears interrupt */
1472 isr = vtpci_isr(hw);
1473 PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1475 if (virtio_intr_unmask(dev) < 0)
1476 PMD_DRV_LOG(ERR, "interrupt enable failed");
1478 if (isr & VIRTIO_PCI_ISR_CONFIG) {
1479 if (virtio_dev_link_update(dev, 0) == 0)
1480 rte_eth_dev_callback_process(dev,
1481 RTE_ETH_EVENT_INTR_LSC,
1484 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1485 vtpci_read_dev_config(hw,
1486 offsetof(struct virtio_net_config, status),
1487 &status, sizeof(status));
1488 if (status & VIRTIO_NET_S_ANNOUNCE) {
1489 virtio_notify_peers(dev);
1491 virtio_ack_link_announce(dev);
1497 /* set rx and tx handlers according to what is supported */
1499 set_rxtx_funcs(struct rte_eth_dev *eth_dev)
1501 struct virtio_hw *hw = eth_dev->data->dev_private;
1503 eth_dev->tx_pkt_prepare = virtio_xmit_pkts_prepare;
1504 if (vtpci_packed_queue(hw)) {
1506 "virtio: using packed ring %s Tx path on port %u",
1507 hw->use_vec_tx ? "vectorized" : "standard",
1508 eth_dev->data->port_id);
1510 eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed_vec;
1512 eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed;
1514 if (hw->use_inorder_tx) {
1515 PMD_INIT_LOG(INFO, "virtio: using inorder Tx path on port %u",
1516 eth_dev->data->port_id);
1517 eth_dev->tx_pkt_burst = virtio_xmit_pkts_inorder;
1519 PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
1520 eth_dev->data->port_id);
1521 eth_dev->tx_pkt_burst = virtio_xmit_pkts;
1525 if (vtpci_packed_queue(hw)) {
1526 if (hw->use_vec_rx) {
1528 "virtio: using packed ring vectorized Rx path on port %u",
1529 eth_dev->data->port_id);
1530 eth_dev->rx_pkt_burst =
1531 &virtio_recv_pkts_packed_vec;
1532 } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1534 "virtio: using packed ring mergeable buffer Rx path on port %u",
1535 eth_dev->data->port_id);
1536 eth_dev->rx_pkt_burst =
1537 &virtio_recv_mergeable_pkts_packed;
1540 "virtio: using packed ring standard Rx path on port %u",
1541 eth_dev->data->port_id);
1542 eth_dev->rx_pkt_burst = &virtio_recv_pkts_packed;
1545 if (hw->use_vec_rx) {
1546 PMD_INIT_LOG(INFO, "virtio: using vectorized Rx path on port %u",
1547 eth_dev->data->port_id);
1548 eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1549 } else if (hw->use_inorder_rx) {
1551 "virtio: using inorder Rx path on port %u",
1552 eth_dev->data->port_id);
1553 eth_dev->rx_pkt_burst = &virtio_recv_pkts_inorder;
1554 } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1556 "virtio: using mergeable buffer Rx path on port %u",
1557 eth_dev->data->port_id);
1558 eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1560 PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
1561 eth_dev->data->port_id);
1562 eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1568 /* Only support 1:1 queue/interrupt mapping so far.
1569 * TODO: support n:1 queue/interrupt mapping when there are limited number of
1570 * interrupt vectors (<N+1).
1573 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1576 struct virtio_hw *hw = dev->data->dev_private;
1578 PMD_INIT_LOG(INFO, "queue/interrupt binding");
1579 for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1580 dev->intr_handle->intr_vec[i] = i + 1;
1581 if (VTPCI_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1582 VIRTIO_MSI_NO_VECTOR) {
1583 PMD_DRV_LOG(ERR, "failed to set queue vector");
1592 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1595 struct virtio_hw *hw = dev->data->dev_private;
1597 PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1598 for (i = 0; i < dev->data->nb_rx_queues; ++i)
1599 VTPCI_OPS(hw)->set_queue_irq(hw,
1600 hw->vqs[i * VTNET_CQ],
1601 VIRTIO_MSI_NO_VECTOR);
1605 virtio_configure_intr(struct rte_eth_dev *dev)
1607 struct virtio_hw *hw = dev->data->dev_private;
1609 if (!rte_intr_cap_multiple(dev->intr_handle)) {
1610 PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1614 if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1615 PMD_INIT_LOG(ERR, "Fail to create eventfd");
1619 if (!dev->intr_handle->intr_vec) {
1620 dev->intr_handle->intr_vec =
1621 rte_zmalloc("intr_vec",
1622 hw->max_queue_pairs * sizeof(int), 0);
1623 if (!dev->intr_handle->intr_vec) {
1624 PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1625 hw->max_queue_pairs);
1630 /* Re-register callback to update max_intr */
1631 rte_intr_callback_unregister(dev->intr_handle,
1632 virtio_interrupt_handler,
1634 rte_intr_callback_register(dev->intr_handle,
1635 virtio_interrupt_handler,
1638 /* DO NOT try to remove this! This function will enable msix, or QEMU
1639 * will encounter SIGSEGV when DRIVER_OK is sent.
1640 * And for legacy devices, this should be done before queue/vec binding
1641 * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1642 * (22) will be ignored.
1644 if (virtio_intr_enable(dev) < 0) {
1645 PMD_DRV_LOG(ERR, "interrupt enable failed");
1649 if (virtio_queues_bind_intr(dev) < 0) {
1650 PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1656 #define DUPLEX_UNKNOWN 0xff
1657 /* reset device and renegotiate features if needed */
1659 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1661 struct virtio_hw *hw = eth_dev->data->dev_private;
1662 struct virtio_net_config *config;
1663 struct virtio_net_config local_config;
1666 /* Reset the device although not necessary at startup */
1670 virtio_dev_free_mbufs(eth_dev);
1671 virtio_free_queues(hw);
1674 /* Tell the host we've noticed this device. */
1675 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1677 /* Tell the host we've known how to drive the device. */
1678 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1679 if (virtio_negotiate_features(hw, req_features) < 0)
1682 hw->weak_barriers = !vtpci_with_feature(hw, VIRTIO_F_ORDER_PLATFORM);
1684 /* If host does not support both status and MSI-X then disable LSC */
1685 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS) &&
1686 hw->use_msix != VIRTIO_MSIX_NONE)
1687 eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1689 eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1691 eth_dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
1693 /* Setting up rx_header size for the device */
1694 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1695 vtpci_with_feature(hw, VIRTIO_F_VERSION_1) ||
1696 vtpci_with_feature(hw, VIRTIO_F_RING_PACKED))
1697 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1699 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1701 /* Copy the permanent MAC address to: virtio_hw */
1702 virtio_get_hwaddr(hw);
1703 rte_ether_addr_copy((struct rte_ether_addr *)hw->mac_addr,
1704 ð_dev->data->mac_addrs[0]);
1706 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1707 hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1708 hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1710 if (hw->speed == ETH_SPEED_NUM_UNKNOWN) {
1711 if (vtpci_with_feature(hw, VIRTIO_NET_F_SPEED_DUPLEX)) {
1712 config = &local_config;
1713 vtpci_read_dev_config(hw,
1714 offsetof(struct virtio_net_config, speed),
1715 &config->speed, sizeof(config->speed));
1716 vtpci_read_dev_config(hw,
1717 offsetof(struct virtio_net_config, duplex),
1718 &config->duplex, sizeof(config->duplex));
1719 hw->speed = config->speed;
1720 hw->duplex = config->duplex;
1723 if (hw->duplex == DUPLEX_UNKNOWN)
1724 hw->duplex = ETH_LINK_FULL_DUPLEX;
1725 PMD_INIT_LOG(DEBUG, "link speed = %d, duplex = %d",
1726 hw->speed, hw->duplex);
1727 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1728 config = &local_config;
1730 vtpci_read_dev_config(hw,
1731 offsetof(struct virtio_net_config, mac),
1732 &config->mac, sizeof(config->mac));
1734 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1735 vtpci_read_dev_config(hw,
1736 offsetof(struct virtio_net_config, status),
1737 &config->status, sizeof(config->status));
1740 "VIRTIO_NET_F_STATUS is not supported");
1744 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1745 vtpci_read_dev_config(hw,
1746 offsetof(struct virtio_net_config, max_virtqueue_pairs),
1747 &config->max_virtqueue_pairs,
1748 sizeof(config->max_virtqueue_pairs));
1751 "VIRTIO_NET_F_MQ is not supported");
1752 config->max_virtqueue_pairs = 1;
1755 hw->max_queue_pairs = config->max_virtqueue_pairs;
1757 if (vtpci_with_feature(hw, VIRTIO_NET_F_MTU)) {
1758 vtpci_read_dev_config(hw,
1759 offsetof(struct virtio_net_config, mtu),
1761 sizeof(config->mtu));
1764 * MTU value has already been checked at negotiation
1765 * time, but check again in case it has changed since
1766 * then, which should not happen.
1768 if (config->mtu < RTE_ETHER_MIN_MTU) {
1769 PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1774 hw->max_mtu = config->mtu;
1775 /* Set initial MTU to maximum one supported by vhost */
1776 eth_dev->data->mtu = config->mtu;
1779 hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1780 VLAN_TAG_LEN - hw->vtnet_hdr_size;
1783 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1784 config->max_virtqueue_pairs);
1785 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1787 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1788 config->mac[0], config->mac[1],
1789 config->mac[2], config->mac[3],
1790 config->mac[4], config->mac[5]);
1792 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1793 hw->max_queue_pairs = 1;
1794 hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1795 VLAN_TAG_LEN - hw->vtnet_hdr_size;
1798 ret = virtio_alloc_queues(eth_dev);
1802 if (eth_dev->data->dev_conf.intr_conf.rxq) {
1803 if (virtio_configure_intr(eth_dev) < 0) {
1804 PMD_INIT_LOG(ERR, "failed to configure interrupt");
1805 virtio_free_queues(hw);
1810 vtpci_reinit_complete(hw);
1817 virtio_set_vtpci_ops(struct virtio_hw *hw)
1819 #ifdef RTE_VIRTIO_USER
1820 if (hw->bus_type == VIRTIO_BUS_USER)
1821 VTPCI_OPS(hw) = &virtio_user_ops;
1824 if (hw->bus_type == VIRTIO_BUS_PCI_MODERN)
1825 VTPCI_OPS(hw) = &modern_ops;
1826 else if (hw->bus_type == VIRTIO_BUS_PCI_LEGACY)
1827 VTPCI_OPS(hw) = &legacy_ops;
1833 * This function is based on probe() function in virtio_pci.c
1834 * It returns 0 on success.
1837 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1839 struct virtio_hw *hw = eth_dev->data->dev_private;
1840 uint32_t speed = ETH_SPEED_NUM_UNKNOWN;
1844 if (sizeof(struct virtio_net_hdr_mrg_rxbuf) > RTE_PKTMBUF_HEADROOM) {
1846 "Not sufficient headroom required = %d, avail = %d",
1847 (int)sizeof(struct virtio_net_hdr_mrg_rxbuf),
1848 RTE_PKTMBUF_HEADROOM);
1853 eth_dev->dev_ops = &virtio_eth_dev_ops;
1854 eth_dev->rx_descriptor_done = virtio_dev_rx_queue_done;
1856 if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1857 virtio_set_vtpci_ops(hw);
1858 set_rxtx_funcs(eth_dev);
1862 ret = virtio_dev_devargs_parse(eth_dev->device->devargs, &speed, &vectorized);
1867 /* Allocate memory for storing MAC addresses */
1868 eth_dev->data->mac_addrs = rte_zmalloc("virtio",
1869 VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN, 0);
1870 if (eth_dev->data->mac_addrs == NULL) {
1872 "Failed to allocate %d bytes needed to store MAC addresses",
1873 VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN);
1877 hw->port_id = eth_dev->data->port_id;
1878 rte_spinlock_init(&hw->state_lock);
1880 /* reset device and negotiate default features */
1881 ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1883 goto err_virtio_init;
1886 if (!vtpci_packed_queue(hw)) {
1889 #if defined(CC_AVX512_SUPPORT) || defined(RTE_ARCH_ARM)
1894 "building environment do not support packed ring vectorized");
1904 rte_free(eth_dev->data->mac_addrs);
1905 eth_dev->data->mac_addrs = NULL;
1910 virtio_dev_speed_capa_get(uint32_t speed)
1913 case ETH_SPEED_NUM_10G:
1914 return ETH_LINK_SPEED_10G;
1915 case ETH_SPEED_NUM_20G:
1916 return ETH_LINK_SPEED_20G;
1917 case ETH_SPEED_NUM_25G:
1918 return ETH_LINK_SPEED_25G;
1919 case ETH_SPEED_NUM_40G:
1920 return ETH_LINK_SPEED_40G;
1921 case ETH_SPEED_NUM_50G:
1922 return ETH_LINK_SPEED_50G;
1923 case ETH_SPEED_NUM_56G:
1924 return ETH_LINK_SPEED_56G;
1925 case ETH_SPEED_NUM_100G:
1926 return ETH_LINK_SPEED_100G;
1927 case ETH_SPEED_NUM_200G:
1928 return ETH_LINK_SPEED_200G;
1934 static int vectorized_check_handler(__rte_unused const char *key,
1935 const char *value, void *ret_val)
1937 if (strcmp(value, "1") == 0)
1938 *(int *)ret_val = 1;
1940 *(int *)ret_val = 0;
1945 #define VIRTIO_ARG_SPEED "speed"
1946 #define VIRTIO_ARG_VECTORIZED "vectorized"
1949 link_speed_handler(const char *key __rte_unused,
1950 const char *value, void *ret_val)
1953 if (!value || !ret_val)
1955 val = strtoul(value, NULL, 0);
1956 /* validate input */
1957 if (virtio_dev_speed_capa_get(val) == 0)
1959 *(uint32_t *)ret_val = val;
1966 virtio_dev_devargs_parse(struct rte_devargs *devargs, uint32_t *speed, int *vectorized)
1968 struct rte_kvargs *kvlist;
1971 if (devargs == NULL)
1974 kvlist = rte_kvargs_parse(devargs->args, NULL);
1975 if (kvlist == NULL) {
1976 PMD_INIT_LOG(ERR, "error when parsing param");
1980 if (speed && rte_kvargs_count(kvlist, VIRTIO_ARG_SPEED) == 1) {
1981 ret = rte_kvargs_process(kvlist,
1983 link_speed_handler, speed);
1985 PMD_INIT_LOG(ERR, "Failed to parse %s",
1992 rte_kvargs_count(kvlist, VIRTIO_ARG_VECTORIZED) == 1) {
1993 ret = rte_kvargs_process(kvlist,
1994 VIRTIO_ARG_VECTORIZED,
1995 vectorized_check_handler, vectorized);
1997 PMD_INIT_LOG(ERR, "Failed to parse %s",
1998 VIRTIO_ARG_VECTORIZED);
2004 rte_kvargs_free(kvlist);
2009 rx_offload_enabled(struct virtio_hw *hw)
2011 return vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM) ||
2012 vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2013 vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6);
2017 tx_offload_enabled(struct virtio_hw *hw)
2019 return vtpci_with_feature(hw, VIRTIO_NET_F_CSUM) ||
2020 vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO4) ||
2021 vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO6);
2025 * Configure virtio device
2026 * It returns 0 on success.
2029 virtio_dev_configure(struct rte_eth_dev *dev)
2031 const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2032 const struct rte_eth_txmode *txmode = &dev->data->dev_conf.txmode;
2033 struct virtio_hw *hw = dev->data->dev_private;
2034 uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
2036 uint64_t rx_offloads = rxmode->offloads;
2037 uint64_t tx_offloads = txmode->offloads;
2038 uint64_t req_features;
2041 PMD_INIT_LOG(DEBUG, "configure");
2042 req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
2044 if (rxmode->mq_mode != ETH_MQ_RX_NONE) {
2046 "Unsupported Rx multi queue mode %d",
2051 if (txmode->mq_mode != ETH_MQ_TX_NONE) {
2053 "Unsupported Tx multi queue mode %d",
2058 if (dev->data->dev_conf.intr_conf.rxq) {
2059 ret = virtio_init_device(dev, hw->req_guest_features);
2064 if (rxmode->max_rx_pkt_len > hw->max_mtu + ether_hdr_len)
2065 req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
2067 if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2068 DEV_RX_OFFLOAD_TCP_CKSUM))
2069 req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
2071 if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO)
2073 (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2074 (1ULL << VIRTIO_NET_F_GUEST_TSO6);
2076 if (tx_offloads & (DEV_TX_OFFLOAD_UDP_CKSUM |
2077 DEV_TX_OFFLOAD_TCP_CKSUM))
2078 req_features |= (1ULL << VIRTIO_NET_F_CSUM);
2080 if (tx_offloads & DEV_TX_OFFLOAD_TCP_TSO)
2082 (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2083 (1ULL << VIRTIO_NET_F_HOST_TSO6);
2085 /* if request features changed, reinit the device */
2086 if (req_features != hw->req_guest_features) {
2087 ret = virtio_init_device(dev, req_features);
2092 if ((rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2093 DEV_RX_OFFLOAD_TCP_CKSUM)) &&
2094 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
2096 "rx checksum not available on this host");
2100 if ((rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) &&
2101 (!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2102 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6))) {
2104 "Large Receive Offload not available on this host");
2108 /* start control queue */
2109 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
2110 virtio_dev_cq_start(dev);
2112 if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
2115 if ((rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
2116 && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2118 "vlan filtering not available on this host");
2122 hw->has_tx_offload = tx_offload_enabled(hw);
2123 hw->has_rx_offload = rx_offload_enabled(hw);
2125 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2126 /* Enable vector (0) for Link State Intrerrupt */
2127 if (VTPCI_OPS(hw)->set_config_irq(hw, 0) ==
2128 VIRTIO_MSI_NO_VECTOR) {
2129 PMD_DRV_LOG(ERR, "failed to set config vector");
2133 if (vtpci_packed_queue(hw)) {
2134 #if defined(RTE_ARCH_X86_64) && defined(CC_AVX512_SUPPORT)
2135 if ((hw->use_vec_rx || hw->use_vec_tx) &&
2136 (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX512F) ||
2137 !vtpci_with_feature(hw, VIRTIO_F_IN_ORDER) ||
2138 !vtpci_with_feature(hw, VIRTIO_F_VERSION_1) ||
2139 rte_vect_get_max_simd_bitwidth() < RTE_VECT_SIMD_512)) {
2141 "disabled packed ring vectorized path for requirements not met");
2145 #elif defined(RTE_ARCH_ARM)
2146 if ((hw->use_vec_rx || hw->use_vec_tx) &&
2147 (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON) ||
2148 !vtpci_with_feature(hw, VIRTIO_F_IN_ORDER) ||
2149 !vtpci_with_feature(hw, VIRTIO_F_VERSION_1) ||
2150 rte_vect_get_max_simd_bitwidth() < RTE_VECT_SIMD_128)) {
2152 "disabled packed ring vectorized path for requirements not met");
2161 if (hw->use_vec_rx) {
2162 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2164 "disabled packed ring vectorized rx for mrg_rxbuf enabled");
2168 if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) {
2170 "disabled packed ring vectorized rx for TCP_LRO enabled");
2175 if (vtpci_with_feature(hw, VIRTIO_F_IN_ORDER)) {
2176 hw->use_inorder_tx = 1;
2177 hw->use_inorder_rx = 1;
2181 if (hw->use_vec_rx) {
2182 #if defined RTE_ARCH_ARM
2183 if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
2185 "disabled split ring vectorized path for requirement not met");
2189 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2191 "disabled split ring vectorized rx for mrg_rxbuf enabled");
2195 if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2196 DEV_RX_OFFLOAD_TCP_CKSUM |
2197 DEV_RX_OFFLOAD_TCP_LRO |
2198 DEV_RX_OFFLOAD_VLAN_STRIP)) {
2200 "disabled split ring vectorized rx for offloading enabled");
2204 if (rte_vect_get_max_simd_bitwidth() < RTE_VECT_SIMD_128) {
2206 "disabled split ring vectorized rx, max SIMD bitwidth too low");
2217 virtio_dev_start(struct rte_eth_dev *dev)
2219 uint16_t nb_queues, i;
2220 struct virtnet_rx *rxvq;
2221 struct virtnet_tx *txvq __rte_unused;
2222 struct virtio_hw *hw = dev->data->dev_private;
2225 /* Finish the initialization of the queues */
2226 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2227 ret = virtio_dev_rx_queue_setup_finish(dev, i);
2231 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2232 ret = virtio_dev_tx_queue_setup_finish(dev, i);
2237 /* check if lsc interrupt feature is enabled */
2238 if (dev->data->dev_conf.intr_conf.lsc) {
2239 if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
2240 PMD_DRV_LOG(ERR, "link status not supported by host");
2245 /* Enable uio/vfio intr/eventfd mapping: althrough we already did that
2246 * in device configure, but it could be unmapped when device is
2249 if (dev->data->dev_conf.intr_conf.lsc ||
2250 dev->data->dev_conf.intr_conf.rxq) {
2251 virtio_intr_disable(dev);
2253 /* Setup interrupt callback */
2254 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2255 rte_intr_callback_register(dev->intr_handle,
2256 virtio_interrupt_handler,
2259 if (virtio_intr_enable(dev) < 0) {
2260 PMD_DRV_LOG(ERR, "interrupt enable failed");
2265 /*Notify the backend
2266 *Otherwise the tap backend might already stop its queue due to fullness.
2267 *vhost backend will have no chance to be waked up
2269 nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
2270 if (hw->max_queue_pairs > 1) {
2271 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
2275 PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
2277 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2278 rxvq = dev->data->rx_queues[i];
2279 /* Flush the old packets */
2280 virtqueue_rxvq_flush(rxvq->vq);
2281 virtqueue_notify(rxvq->vq);
2284 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2285 txvq = dev->data->tx_queues[i];
2286 virtqueue_notify(txvq->vq);
2289 PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
2291 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2292 rxvq = dev->data->rx_queues[i];
2293 VIRTQUEUE_DUMP(rxvq->vq);
2296 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2297 txvq = dev->data->tx_queues[i];
2298 VIRTQUEUE_DUMP(txvq->vq);
2301 set_rxtx_funcs(dev);
2304 /* Initialize Link state */
2305 virtio_dev_link_update(dev, 0);
2310 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
2312 struct virtio_hw *hw = dev->data->dev_private;
2313 uint16_t nr_vq = virtio_get_nr_vq(hw);
2314 const char *type __rte_unused;
2315 unsigned int i, mbuf_num = 0;
2316 struct virtqueue *vq;
2317 struct rte_mbuf *buf;
2320 if (hw->vqs == NULL)
2323 for (i = 0; i < nr_vq; i++) {
2328 queue_type = virtio_get_queue_type(hw, i);
2329 if (queue_type == VTNET_RQ)
2331 else if (queue_type == VTNET_TQ)
2337 "Before freeing %s[%d] used and unused buf",
2341 while ((buf = virtqueue_detach_unused(vq)) != NULL) {
2342 rte_pktmbuf_free(buf);
2347 "After freeing %s[%d] used and unused buf",
2352 PMD_INIT_LOG(DEBUG, "%d mbufs freed", mbuf_num);
2356 * Stop device: disable interrupt and mark link down
2359 virtio_dev_stop(struct rte_eth_dev *dev)
2361 struct virtio_hw *hw = dev->data->dev_private;
2362 struct rte_eth_link link;
2363 struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
2365 PMD_INIT_LOG(DEBUG, "stop");
2366 dev->data->dev_started = 0;
2368 rte_spinlock_lock(&hw->state_lock);
2371 hw->started = false;
2373 if (intr_conf->lsc || intr_conf->rxq) {
2374 virtio_intr_disable(dev);
2376 /* Reset interrupt callback */
2377 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
2378 rte_intr_callback_unregister(dev->intr_handle,
2379 virtio_interrupt_handler,
2384 memset(&link, 0, sizeof(link));
2385 rte_eth_linkstatus_set(dev, &link);
2387 rte_spinlock_unlock(&hw->state_lock);
2393 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
2395 struct rte_eth_link link;
2397 struct virtio_hw *hw = dev->data->dev_private;
2399 memset(&link, 0, sizeof(link));
2400 link.link_duplex = hw->duplex;
2401 link.link_speed = hw->speed;
2402 link.link_autoneg = ETH_LINK_AUTONEG;
2405 link.link_status = ETH_LINK_DOWN;
2406 link.link_speed = ETH_SPEED_NUM_NONE;
2407 } else if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
2408 PMD_INIT_LOG(DEBUG, "Get link status from hw");
2409 vtpci_read_dev_config(hw,
2410 offsetof(struct virtio_net_config, status),
2411 &status, sizeof(status));
2412 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
2413 link.link_status = ETH_LINK_DOWN;
2414 link.link_speed = ETH_SPEED_NUM_NONE;
2415 PMD_INIT_LOG(DEBUG, "Port %d is down",
2416 dev->data->port_id);
2418 link.link_status = ETH_LINK_UP;
2419 PMD_INIT_LOG(DEBUG, "Port %d is up",
2420 dev->data->port_id);
2423 link.link_status = ETH_LINK_UP;
2426 return rte_eth_linkstatus_set(dev, &link);
2430 virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
2432 const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2433 struct virtio_hw *hw = dev->data->dev_private;
2434 uint64_t offloads = rxmode->offloads;
2436 if (mask & ETH_VLAN_FILTER_MASK) {
2437 if ((offloads & DEV_RX_OFFLOAD_VLAN_FILTER) &&
2438 !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2441 "vlan filtering not available on this host");
2447 if (mask & ETH_VLAN_STRIP_MASK)
2448 hw->vlan_strip = !!(offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
2454 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2456 uint64_t tso_mask, host_features;
2457 struct virtio_hw *hw = dev->data->dev_private;
2458 dev_info->speed_capa = virtio_dev_speed_capa_get(hw->speed);
2460 dev_info->max_rx_queues =
2461 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
2462 dev_info->max_tx_queues =
2463 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
2464 dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
2465 dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
2466 dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
2468 host_features = VTPCI_OPS(hw)->get_features(hw);
2469 dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
2470 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_JUMBO_FRAME;
2471 if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
2472 dev_info->rx_offload_capa |=
2473 DEV_RX_OFFLOAD_TCP_CKSUM |
2474 DEV_RX_OFFLOAD_UDP_CKSUM;
2476 if (host_features & (1ULL << VIRTIO_NET_F_CTRL_VLAN))
2477 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_VLAN_FILTER;
2478 tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2479 (1ULL << VIRTIO_NET_F_GUEST_TSO6);
2480 if ((host_features & tso_mask) == tso_mask)
2481 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
2483 dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
2484 DEV_TX_OFFLOAD_VLAN_INSERT;
2485 if (host_features & (1ULL << VIRTIO_NET_F_CSUM)) {
2486 dev_info->tx_offload_capa |=
2487 DEV_TX_OFFLOAD_UDP_CKSUM |
2488 DEV_TX_OFFLOAD_TCP_CKSUM;
2490 tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2491 (1ULL << VIRTIO_NET_F_HOST_TSO6);
2492 if ((host_features & tso_mask) == tso_mask)
2493 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
2499 * It enables testpmd to collect per queue stats.
2502 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
2503 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
2504 __rte_unused uint8_t is_rx)
2509 RTE_LOG_REGISTER(virtio_logtype_init, pmd.net.virtio.init, NOTICE);
2510 RTE_LOG_REGISTER(virtio_logtype_driver, pmd.net.virtio.driver, NOTICE);