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 return VTPCI_OPS(hw)->dev_close(hw);
725 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
727 struct virtio_hw *hw = dev->data->dev_private;
728 struct virtio_pmd_ctrl ctrl;
732 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
733 PMD_INIT_LOG(INFO, "host does not support rx control");
737 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
738 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
742 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
744 PMD_INIT_LOG(ERR, "Failed to enable promisc");
752 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
754 struct virtio_hw *hw = dev->data->dev_private;
755 struct virtio_pmd_ctrl ctrl;
759 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
760 PMD_INIT_LOG(INFO, "host does not support rx control");
764 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
765 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
769 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
771 PMD_INIT_LOG(ERR, "Failed to disable promisc");
779 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
781 struct virtio_hw *hw = dev->data->dev_private;
782 struct virtio_pmd_ctrl ctrl;
786 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
787 PMD_INIT_LOG(INFO, "host does not support rx control");
791 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
792 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
796 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
798 PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
806 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
808 struct virtio_hw *hw = dev->data->dev_private;
809 struct virtio_pmd_ctrl ctrl;
813 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
814 PMD_INIT_LOG(INFO, "host does not support rx control");
818 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
819 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
823 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
825 PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
832 #define VLAN_TAG_LEN 4 /* 802.3ac tag (not DMA'd) */
834 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
836 struct virtio_hw *hw = dev->data->dev_private;
837 uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
839 uint32_t frame_size = mtu + ether_hdr_len;
840 uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
842 max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
844 if (mtu < RTE_ETHER_MIN_MTU || frame_size > max_frame_size) {
845 PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
846 RTE_ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
853 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
855 struct virtio_hw *hw = dev->data->dev_private;
856 struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
857 struct virtqueue *vq = rxvq->vq;
859 virtqueue_enable_intr(vq);
860 virtio_mb(hw->weak_barriers);
865 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
867 struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
868 struct virtqueue *vq = rxvq->vq;
870 virtqueue_disable_intr(vq);
875 * dev_ops for virtio, bare necessities for basic operation
877 static const struct eth_dev_ops virtio_eth_dev_ops = {
878 .dev_configure = virtio_dev_configure,
879 .dev_start = virtio_dev_start,
880 .dev_stop = virtio_dev_stop,
881 .dev_close = virtio_dev_close,
882 .promiscuous_enable = virtio_dev_promiscuous_enable,
883 .promiscuous_disable = virtio_dev_promiscuous_disable,
884 .allmulticast_enable = virtio_dev_allmulticast_enable,
885 .allmulticast_disable = virtio_dev_allmulticast_disable,
886 .mtu_set = virtio_mtu_set,
887 .dev_infos_get = virtio_dev_info_get,
888 .stats_get = virtio_dev_stats_get,
889 .xstats_get = virtio_dev_xstats_get,
890 .xstats_get_names = virtio_dev_xstats_get_names,
891 .stats_reset = virtio_dev_stats_reset,
892 .xstats_reset = virtio_dev_stats_reset,
893 .link_update = virtio_dev_link_update,
894 .vlan_offload_set = virtio_dev_vlan_offload_set,
895 .rx_queue_setup = virtio_dev_rx_queue_setup,
896 .rx_queue_intr_enable = virtio_dev_rx_queue_intr_enable,
897 .rx_queue_intr_disable = virtio_dev_rx_queue_intr_disable,
898 .rx_queue_release = virtio_dev_queue_release,
899 .tx_queue_setup = virtio_dev_tx_queue_setup,
900 .tx_queue_release = virtio_dev_queue_release,
901 /* collect stats per queue */
902 .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
903 .vlan_filter_set = virtio_vlan_filter_set,
904 .mac_addr_add = virtio_mac_addr_add,
905 .mac_addr_remove = virtio_mac_addr_remove,
906 .mac_addr_set = virtio_mac_addr_set,
910 * dev_ops for virtio-user in secondary processes, as we just have
911 * some limited supports currently.
913 const struct eth_dev_ops virtio_user_secondary_eth_dev_ops = {
914 .dev_infos_get = virtio_dev_info_get,
915 .stats_get = virtio_dev_stats_get,
916 .xstats_get = virtio_dev_xstats_get,
917 .xstats_get_names = virtio_dev_xstats_get_names,
918 .stats_reset = virtio_dev_stats_reset,
919 .xstats_reset = virtio_dev_stats_reset,
920 /* collect stats per queue */
921 .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
925 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
929 for (i = 0; i < dev->data->nb_tx_queues; i++) {
930 const struct virtnet_tx *txvq = dev->data->tx_queues[i];
934 stats->opackets += txvq->stats.packets;
935 stats->obytes += txvq->stats.bytes;
937 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
938 stats->q_opackets[i] = txvq->stats.packets;
939 stats->q_obytes[i] = txvq->stats.bytes;
943 for (i = 0; i < dev->data->nb_rx_queues; i++) {
944 const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
948 stats->ipackets += rxvq->stats.packets;
949 stats->ibytes += rxvq->stats.bytes;
950 stats->ierrors += rxvq->stats.errors;
952 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
953 stats->q_ipackets[i] = rxvq->stats.packets;
954 stats->q_ibytes[i] = rxvq->stats.bytes;
958 stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
961 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
962 struct rte_eth_xstat_name *xstats_names,
963 __rte_unused unsigned limit)
969 unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
970 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
972 if (xstats_names != NULL) {
973 /* Note: limit checked in rte_eth_xstats_names() */
975 for (i = 0; i < dev->data->nb_rx_queues; i++) {
976 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
979 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
980 snprintf(xstats_names[count].name,
981 sizeof(xstats_names[count].name),
983 rte_virtio_rxq_stat_strings[t].name);
988 for (i = 0; i < dev->data->nb_tx_queues; i++) {
989 struct virtnet_tx *txvq = dev->data->tx_queues[i];
992 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
993 snprintf(xstats_names[count].name,
994 sizeof(xstats_names[count].name),
996 rte_virtio_txq_stat_strings[t].name);
1006 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
1012 unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
1013 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
1018 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1019 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1026 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1027 xstats[count].value = *(uint64_t *)(((char *)rxvq) +
1028 rte_virtio_rxq_stat_strings[t].offset);
1029 xstats[count].id = count;
1034 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1035 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1042 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1043 xstats[count].value = *(uint64_t *)(((char *)txvq) +
1044 rte_virtio_txq_stat_strings[t].offset);
1045 xstats[count].id = count;
1054 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1056 virtio_update_stats(dev, stats);
1062 virtio_dev_stats_reset(struct rte_eth_dev *dev)
1066 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1067 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1071 txvq->stats.packets = 0;
1072 txvq->stats.bytes = 0;
1073 txvq->stats.multicast = 0;
1074 txvq->stats.broadcast = 0;
1075 memset(txvq->stats.size_bins, 0,
1076 sizeof(txvq->stats.size_bins[0]) * 8);
1079 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1080 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1084 rxvq->stats.packets = 0;
1085 rxvq->stats.bytes = 0;
1086 rxvq->stats.errors = 0;
1087 rxvq->stats.multicast = 0;
1088 rxvq->stats.broadcast = 0;
1089 memset(rxvq->stats.size_bins, 0,
1090 sizeof(rxvq->stats.size_bins[0]) * 8);
1097 virtio_set_hwaddr(struct virtio_hw *hw)
1099 vtpci_write_dev_config(hw,
1100 offsetof(struct virtio_net_config, mac),
1101 &hw->mac_addr, RTE_ETHER_ADDR_LEN);
1105 virtio_get_hwaddr(struct virtio_hw *hw)
1107 if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
1108 vtpci_read_dev_config(hw,
1109 offsetof(struct virtio_net_config, mac),
1110 &hw->mac_addr, RTE_ETHER_ADDR_LEN);
1112 rte_eth_random_addr(&hw->mac_addr[0]);
1113 virtio_set_hwaddr(hw);
1118 virtio_mac_table_set(struct virtio_hw *hw,
1119 const struct virtio_net_ctrl_mac *uc,
1120 const struct virtio_net_ctrl_mac *mc)
1122 struct virtio_pmd_ctrl ctrl;
1125 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1126 PMD_DRV_LOG(INFO, "host does not support mac table");
1130 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1131 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1133 len[0] = uc->entries * RTE_ETHER_ADDR_LEN + sizeof(uc->entries);
1134 memcpy(ctrl.data, uc, len[0]);
1136 len[1] = mc->entries * RTE_ETHER_ADDR_LEN + sizeof(mc->entries);
1137 memcpy(ctrl.data + len[0], mc, len[1]);
1139 err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1141 PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1146 virtio_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
1147 uint32_t index, uint32_t vmdq __rte_unused)
1149 struct virtio_hw *hw = dev->data->dev_private;
1150 const struct rte_ether_addr *addrs = dev->data->mac_addrs;
1152 struct virtio_net_ctrl_mac *uc, *mc;
1154 if (index >= VIRTIO_MAX_MAC_ADDRS) {
1155 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1159 uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1160 sizeof(uc->entries));
1162 mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1163 sizeof(mc->entries));
1166 for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1167 const struct rte_ether_addr *addr
1168 = (i == index) ? mac_addr : addrs + i;
1169 struct virtio_net_ctrl_mac *tbl
1170 = rte_is_multicast_ether_addr(addr) ? mc : uc;
1172 memcpy(&tbl->macs[tbl->entries++], addr, RTE_ETHER_ADDR_LEN);
1175 return virtio_mac_table_set(hw, uc, mc);
1179 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1181 struct virtio_hw *hw = dev->data->dev_private;
1182 struct rte_ether_addr *addrs = dev->data->mac_addrs;
1183 struct virtio_net_ctrl_mac *uc, *mc;
1186 if (index >= VIRTIO_MAX_MAC_ADDRS) {
1187 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1191 uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1192 sizeof(uc->entries));
1194 mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1195 sizeof(mc->entries));
1198 for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1199 struct virtio_net_ctrl_mac *tbl;
1201 if (i == index || rte_is_zero_ether_addr(addrs + i))
1204 tbl = rte_is_multicast_ether_addr(addrs + i) ? mc : uc;
1205 memcpy(&tbl->macs[tbl->entries++], addrs + i,
1206 RTE_ETHER_ADDR_LEN);
1209 virtio_mac_table_set(hw, uc, mc);
1213 virtio_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1215 struct virtio_hw *hw = dev->data->dev_private;
1217 memcpy(hw->mac_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1219 /* Use atomic update if available */
1220 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1221 struct virtio_pmd_ctrl ctrl;
1222 int len = RTE_ETHER_ADDR_LEN;
1224 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1225 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1227 memcpy(ctrl.data, mac_addr, RTE_ETHER_ADDR_LEN);
1228 return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1231 if (!vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1234 virtio_set_hwaddr(hw);
1239 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1241 struct virtio_hw *hw = dev->data->dev_private;
1242 struct virtio_pmd_ctrl ctrl;
1245 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1248 ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1249 ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1250 memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1251 len = sizeof(vlan_id);
1253 return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1257 virtio_intr_unmask(struct rte_eth_dev *dev)
1259 struct virtio_hw *hw = dev->data->dev_private;
1261 if (rte_intr_ack(dev->intr_handle) < 0)
1264 if (VTPCI_OPS(hw)->intr_detect)
1265 VTPCI_OPS(hw)->intr_detect(hw);
1271 virtio_intr_enable(struct rte_eth_dev *dev)
1273 struct virtio_hw *hw = dev->data->dev_private;
1275 if (rte_intr_enable(dev->intr_handle) < 0)
1278 if (VTPCI_OPS(hw)->intr_detect)
1279 VTPCI_OPS(hw)->intr_detect(hw);
1285 virtio_intr_disable(struct rte_eth_dev *dev)
1287 struct virtio_hw *hw = dev->data->dev_private;
1289 if (rte_intr_disable(dev->intr_handle) < 0)
1292 if (VTPCI_OPS(hw)->intr_detect)
1293 VTPCI_OPS(hw)->intr_detect(hw);
1299 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1301 uint64_t host_features;
1303 /* Prepare guest_features: feature that driver wants to support */
1304 PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1307 /* Read device(host) feature bits */
1308 host_features = VTPCI_OPS(hw)->get_features(hw);
1309 PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1312 /* If supported, ensure MTU value is valid before acknowledging it. */
1313 if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1314 struct virtio_net_config config;
1316 vtpci_read_dev_config(hw,
1317 offsetof(struct virtio_net_config, mtu),
1318 &config.mtu, sizeof(config.mtu));
1320 if (config.mtu < RTE_ETHER_MIN_MTU)
1321 req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1325 * Negotiate features: Subset of device feature bits are written back
1326 * guest feature bits.
1328 hw->guest_features = req_features;
1329 hw->guest_features = vtpci_negotiate_features(hw, host_features);
1330 PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1331 hw->guest_features);
1333 if (VTPCI_OPS(hw)->features_ok(hw) < 0)
1336 if (vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1337 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1339 if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1340 PMD_INIT_LOG(ERR, "Failed to set FEATURES_OK status!");
1345 hw->req_guest_features = req_features;
1351 virtio_dev_pause(struct rte_eth_dev *dev)
1353 struct virtio_hw *hw = dev->data->dev_private;
1355 rte_spinlock_lock(&hw->state_lock);
1357 if (hw->started == 0) {
1358 /* Device is just stopped. */
1359 rte_spinlock_unlock(&hw->state_lock);
1364 * Prevent the worker threads from touching queues to avoid contention,
1365 * 1 ms should be enough for the ongoing Tx function to finish.
1372 * Recover hw state to let the worker threads continue.
1375 virtio_dev_resume(struct rte_eth_dev *dev)
1377 struct virtio_hw *hw = dev->data->dev_private;
1380 rte_spinlock_unlock(&hw->state_lock);
1384 * Should be called only after device is paused.
1387 virtio_inject_pkts(struct rte_eth_dev *dev, struct rte_mbuf **tx_pkts,
1390 struct virtio_hw *hw = dev->data->dev_private;
1391 struct virtnet_tx *txvq = dev->data->tx_queues[0];
1394 hw->inject_pkts = tx_pkts;
1395 ret = dev->tx_pkt_burst(txvq, tx_pkts, nb_pkts);
1396 hw->inject_pkts = NULL;
1402 virtio_notify_peers(struct rte_eth_dev *dev)
1404 struct virtio_hw *hw = dev->data->dev_private;
1405 struct virtnet_rx *rxvq;
1406 struct rte_mbuf *rarp_mbuf;
1408 if (!dev->data->rx_queues)
1411 rxvq = dev->data->rx_queues[0];
1415 rarp_mbuf = rte_net_make_rarp_packet(rxvq->mpool,
1416 (struct rte_ether_addr *)hw->mac_addr);
1417 if (rarp_mbuf == NULL) {
1418 PMD_DRV_LOG(ERR, "failed to make RARP packet.");
1422 /* If virtio port just stopped, no need to send RARP */
1423 if (virtio_dev_pause(dev) < 0) {
1424 rte_pktmbuf_free(rarp_mbuf);
1428 virtio_inject_pkts(dev, &rarp_mbuf, 1);
1429 virtio_dev_resume(dev);
1433 virtio_ack_link_announce(struct rte_eth_dev *dev)
1435 struct virtio_hw *hw = dev->data->dev_private;
1436 struct virtio_pmd_ctrl ctrl;
1438 ctrl.hdr.class = VIRTIO_NET_CTRL_ANNOUNCE;
1439 ctrl.hdr.cmd = VIRTIO_NET_CTRL_ANNOUNCE_ACK;
1441 virtio_send_command(hw->cvq, &ctrl, NULL, 0);
1445 * Process virtio config changed interrupt. Call the callback
1446 * if link state changed, generate gratuitous RARP packet if
1447 * the status indicates an ANNOUNCE.
1450 virtio_interrupt_handler(void *param)
1452 struct rte_eth_dev *dev = param;
1453 struct virtio_hw *hw = dev->data->dev_private;
1457 /* Read interrupt status which clears interrupt */
1458 isr = vtpci_isr(hw);
1459 PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1461 if (virtio_intr_unmask(dev) < 0)
1462 PMD_DRV_LOG(ERR, "interrupt enable failed");
1464 if (isr & VIRTIO_PCI_ISR_CONFIG) {
1465 if (virtio_dev_link_update(dev, 0) == 0)
1466 rte_eth_dev_callback_process(dev,
1467 RTE_ETH_EVENT_INTR_LSC,
1470 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1471 vtpci_read_dev_config(hw,
1472 offsetof(struct virtio_net_config, status),
1473 &status, sizeof(status));
1474 if (status & VIRTIO_NET_S_ANNOUNCE) {
1475 virtio_notify_peers(dev);
1477 virtio_ack_link_announce(dev);
1483 /* set rx and tx handlers according to what is supported */
1485 set_rxtx_funcs(struct rte_eth_dev *eth_dev)
1487 struct virtio_hw *hw = eth_dev->data->dev_private;
1489 eth_dev->tx_pkt_prepare = virtio_xmit_pkts_prepare;
1490 if (vtpci_packed_queue(hw)) {
1492 "virtio: using packed ring %s Tx path on port %u",
1493 hw->use_vec_tx ? "vectorized" : "standard",
1494 eth_dev->data->port_id);
1496 eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed_vec;
1498 eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed;
1500 if (hw->use_inorder_tx) {
1501 PMD_INIT_LOG(INFO, "virtio: using inorder Tx path on port %u",
1502 eth_dev->data->port_id);
1503 eth_dev->tx_pkt_burst = virtio_xmit_pkts_inorder;
1505 PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
1506 eth_dev->data->port_id);
1507 eth_dev->tx_pkt_burst = virtio_xmit_pkts;
1511 if (vtpci_packed_queue(hw)) {
1512 if (hw->use_vec_rx) {
1514 "virtio: using packed ring vectorized Rx path on port %u",
1515 eth_dev->data->port_id);
1516 eth_dev->rx_pkt_burst =
1517 &virtio_recv_pkts_packed_vec;
1518 } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1520 "virtio: using packed ring mergeable buffer Rx path on port %u",
1521 eth_dev->data->port_id);
1522 eth_dev->rx_pkt_burst =
1523 &virtio_recv_mergeable_pkts_packed;
1526 "virtio: using packed ring standard Rx path on port %u",
1527 eth_dev->data->port_id);
1528 eth_dev->rx_pkt_burst = &virtio_recv_pkts_packed;
1531 if (hw->use_vec_rx) {
1532 PMD_INIT_LOG(INFO, "virtio: using vectorized Rx path on port %u",
1533 eth_dev->data->port_id);
1534 eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1535 } else if (hw->use_inorder_rx) {
1537 "virtio: using inorder Rx path on port %u",
1538 eth_dev->data->port_id);
1539 eth_dev->rx_pkt_burst = &virtio_recv_pkts_inorder;
1540 } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1542 "virtio: using mergeable buffer Rx path on port %u",
1543 eth_dev->data->port_id);
1544 eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1546 PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
1547 eth_dev->data->port_id);
1548 eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1554 /* Only support 1:1 queue/interrupt mapping so far.
1555 * TODO: support n:1 queue/interrupt mapping when there are limited number of
1556 * interrupt vectors (<N+1).
1559 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1562 struct virtio_hw *hw = dev->data->dev_private;
1564 PMD_INIT_LOG(INFO, "queue/interrupt binding");
1565 for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1566 dev->intr_handle->intr_vec[i] = i + 1;
1567 if (VTPCI_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1568 VIRTIO_MSI_NO_VECTOR) {
1569 PMD_DRV_LOG(ERR, "failed to set queue vector");
1578 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1581 struct virtio_hw *hw = dev->data->dev_private;
1583 PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1584 for (i = 0; i < dev->data->nb_rx_queues; ++i)
1585 VTPCI_OPS(hw)->set_queue_irq(hw,
1586 hw->vqs[i * VTNET_CQ],
1587 VIRTIO_MSI_NO_VECTOR);
1591 virtio_configure_intr(struct rte_eth_dev *dev)
1593 struct virtio_hw *hw = dev->data->dev_private;
1595 if (!rte_intr_cap_multiple(dev->intr_handle)) {
1596 PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1600 if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1601 PMD_INIT_LOG(ERR, "Fail to create eventfd");
1605 if (!dev->intr_handle->intr_vec) {
1606 dev->intr_handle->intr_vec =
1607 rte_zmalloc("intr_vec",
1608 hw->max_queue_pairs * sizeof(int), 0);
1609 if (!dev->intr_handle->intr_vec) {
1610 PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1611 hw->max_queue_pairs);
1616 /* Re-register callback to update max_intr */
1617 rte_intr_callback_unregister(dev->intr_handle,
1618 virtio_interrupt_handler,
1620 rte_intr_callback_register(dev->intr_handle,
1621 virtio_interrupt_handler,
1624 /* DO NOT try to remove this! This function will enable msix, or QEMU
1625 * will encounter SIGSEGV when DRIVER_OK is sent.
1626 * And for legacy devices, this should be done before queue/vec binding
1627 * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1628 * (22) will be ignored.
1630 if (virtio_intr_enable(dev) < 0) {
1631 PMD_DRV_LOG(ERR, "interrupt enable failed");
1635 if (virtio_queues_bind_intr(dev) < 0) {
1636 PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1642 #define DUPLEX_UNKNOWN 0xff
1643 /* reset device and renegotiate features if needed */
1645 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1647 struct virtio_hw *hw = eth_dev->data->dev_private;
1648 struct virtio_net_config *config;
1649 struct virtio_net_config local_config;
1652 /* Reset the device although not necessary at startup */
1656 virtio_dev_free_mbufs(eth_dev);
1657 virtio_free_queues(hw);
1660 /* Tell the host we've noticed this device. */
1661 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1663 /* Tell the host we've known how to drive the device. */
1664 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1665 if (virtio_negotiate_features(hw, req_features) < 0)
1668 hw->weak_barriers = !vtpci_with_feature(hw, VIRTIO_F_ORDER_PLATFORM);
1670 /* If host does not support both status and MSI-X then disable LSC */
1671 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS) &&
1672 hw->use_msix != VIRTIO_MSIX_NONE)
1673 eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1675 eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1677 eth_dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
1679 /* Setting up rx_header size for the device */
1680 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1681 vtpci_with_feature(hw, VIRTIO_F_VERSION_1) ||
1682 vtpci_with_feature(hw, VIRTIO_F_RING_PACKED))
1683 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1685 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1687 /* Copy the permanent MAC address to: virtio_hw */
1688 virtio_get_hwaddr(hw);
1689 rte_ether_addr_copy((struct rte_ether_addr *)hw->mac_addr,
1690 ð_dev->data->mac_addrs[0]);
1692 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1693 hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1694 hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1696 if (hw->speed == ETH_SPEED_NUM_UNKNOWN) {
1697 if (vtpci_with_feature(hw, VIRTIO_NET_F_SPEED_DUPLEX)) {
1698 config = &local_config;
1699 vtpci_read_dev_config(hw,
1700 offsetof(struct virtio_net_config, speed),
1701 &config->speed, sizeof(config->speed));
1702 vtpci_read_dev_config(hw,
1703 offsetof(struct virtio_net_config, duplex),
1704 &config->duplex, sizeof(config->duplex));
1705 hw->speed = config->speed;
1706 hw->duplex = config->duplex;
1709 if (hw->duplex == DUPLEX_UNKNOWN)
1710 hw->duplex = ETH_LINK_FULL_DUPLEX;
1711 PMD_INIT_LOG(DEBUG, "link speed = %d, duplex = %d",
1712 hw->speed, hw->duplex);
1713 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1714 config = &local_config;
1716 vtpci_read_dev_config(hw,
1717 offsetof(struct virtio_net_config, mac),
1718 &config->mac, sizeof(config->mac));
1720 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1721 vtpci_read_dev_config(hw,
1722 offsetof(struct virtio_net_config, status),
1723 &config->status, sizeof(config->status));
1726 "VIRTIO_NET_F_STATUS is not supported");
1730 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1731 vtpci_read_dev_config(hw,
1732 offsetof(struct virtio_net_config, max_virtqueue_pairs),
1733 &config->max_virtqueue_pairs,
1734 sizeof(config->max_virtqueue_pairs));
1737 "VIRTIO_NET_F_MQ is not supported");
1738 config->max_virtqueue_pairs = 1;
1741 hw->max_queue_pairs = config->max_virtqueue_pairs;
1743 if (vtpci_with_feature(hw, VIRTIO_NET_F_MTU)) {
1744 vtpci_read_dev_config(hw,
1745 offsetof(struct virtio_net_config, mtu),
1747 sizeof(config->mtu));
1750 * MTU value has already been checked at negotiation
1751 * time, but check again in case it has changed since
1752 * then, which should not happen.
1754 if (config->mtu < RTE_ETHER_MIN_MTU) {
1755 PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1760 hw->max_mtu = config->mtu;
1761 /* Set initial MTU to maximum one supported by vhost */
1762 eth_dev->data->mtu = config->mtu;
1765 hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1766 VLAN_TAG_LEN - hw->vtnet_hdr_size;
1769 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1770 config->max_virtqueue_pairs);
1771 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1773 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1774 config->mac[0], config->mac[1],
1775 config->mac[2], config->mac[3],
1776 config->mac[4], config->mac[5]);
1778 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1779 hw->max_queue_pairs = 1;
1780 hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1781 VLAN_TAG_LEN - hw->vtnet_hdr_size;
1784 ret = virtio_alloc_queues(eth_dev);
1788 if (eth_dev->data->dev_conf.intr_conf.rxq) {
1789 if (virtio_configure_intr(eth_dev) < 0) {
1790 PMD_INIT_LOG(ERR, "failed to configure interrupt");
1791 virtio_free_queues(hw);
1796 vtpci_reinit_complete(hw);
1802 * This function is based on probe() function in virtio_pci.c
1803 * It returns 0 on success.
1806 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1808 struct virtio_hw *hw = eth_dev->data->dev_private;
1809 uint32_t speed = ETH_SPEED_NUM_UNKNOWN;
1813 if (sizeof(struct virtio_net_hdr_mrg_rxbuf) > RTE_PKTMBUF_HEADROOM) {
1815 "Not sufficient headroom required = %d, avail = %d",
1816 (int)sizeof(struct virtio_net_hdr_mrg_rxbuf),
1817 RTE_PKTMBUF_HEADROOM);
1822 eth_dev->dev_ops = &virtio_eth_dev_ops;
1823 eth_dev->rx_descriptor_done = virtio_dev_rx_queue_done;
1825 if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1826 set_rxtx_funcs(eth_dev);
1830 ret = virtio_dev_devargs_parse(eth_dev->device->devargs, &speed, &vectorized);
1835 /* Allocate memory for storing MAC addresses */
1836 eth_dev->data->mac_addrs = rte_zmalloc("virtio",
1837 VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN, 0);
1838 if (eth_dev->data->mac_addrs == NULL) {
1840 "Failed to allocate %d bytes needed to store MAC addresses",
1841 VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN);
1845 hw->port_id = eth_dev->data->port_id;
1846 rte_spinlock_init(&hw->state_lock);
1848 /* reset device and negotiate default features */
1849 ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1851 goto err_virtio_init;
1854 if (!vtpci_packed_queue(hw)) {
1857 #if defined(CC_AVX512_SUPPORT) || defined(RTE_ARCH_ARM)
1862 "building environment do not support packed ring vectorized");
1872 rte_free(eth_dev->data->mac_addrs);
1873 eth_dev->data->mac_addrs = NULL;
1878 virtio_dev_speed_capa_get(uint32_t speed)
1881 case ETH_SPEED_NUM_10G:
1882 return ETH_LINK_SPEED_10G;
1883 case ETH_SPEED_NUM_20G:
1884 return ETH_LINK_SPEED_20G;
1885 case ETH_SPEED_NUM_25G:
1886 return ETH_LINK_SPEED_25G;
1887 case ETH_SPEED_NUM_40G:
1888 return ETH_LINK_SPEED_40G;
1889 case ETH_SPEED_NUM_50G:
1890 return ETH_LINK_SPEED_50G;
1891 case ETH_SPEED_NUM_56G:
1892 return ETH_LINK_SPEED_56G;
1893 case ETH_SPEED_NUM_100G:
1894 return ETH_LINK_SPEED_100G;
1895 case ETH_SPEED_NUM_200G:
1896 return ETH_LINK_SPEED_200G;
1902 static int vectorized_check_handler(__rte_unused const char *key,
1903 const char *value, void *ret_val)
1905 if (strcmp(value, "1") == 0)
1906 *(int *)ret_val = 1;
1908 *(int *)ret_val = 0;
1913 #define VIRTIO_ARG_SPEED "speed"
1914 #define VIRTIO_ARG_VECTORIZED "vectorized"
1917 link_speed_handler(const char *key __rte_unused,
1918 const char *value, void *ret_val)
1921 if (!value || !ret_val)
1923 val = strtoul(value, NULL, 0);
1924 /* validate input */
1925 if (virtio_dev_speed_capa_get(val) == 0)
1927 *(uint32_t *)ret_val = val;
1934 virtio_dev_devargs_parse(struct rte_devargs *devargs, uint32_t *speed, int *vectorized)
1936 struct rte_kvargs *kvlist;
1939 if (devargs == NULL)
1942 kvlist = rte_kvargs_parse(devargs->args, NULL);
1943 if (kvlist == NULL) {
1944 PMD_INIT_LOG(ERR, "error when parsing param");
1948 if (speed && rte_kvargs_count(kvlist, VIRTIO_ARG_SPEED) == 1) {
1949 ret = rte_kvargs_process(kvlist,
1951 link_speed_handler, speed);
1953 PMD_INIT_LOG(ERR, "Failed to parse %s",
1960 rte_kvargs_count(kvlist, VIRTIO_ARG_VECTORIZED) == 1) {
1961 ret = rte_kvargs_process(kvlist,
1962 VIRTIO_ARG_VECTORIZED,
1963 vectorized_check_handler, vectorized);
1965 PMD_INIT_LOG(ERR, "Failed to parse %s",
1966 VIRTIO_ARG_VECTORIZED);
1972 rte_kvargs_free(kvlist);
1977 rx_offload_enabled(struct virtio_hw *hw)
1979 return vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM) ||
1980 vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
1981 vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6);
1985 tx_offload_enabled(struct virtio_hw *hw)
1987 return vtpci_with_feature(hw, VIRTIO_NET_F_CSUM) ||
1988 vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO4) ||
1989 vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO6);
1993 * Configure virtio device
1994 * It returns 0 on success.
1997 virtio_dev_configure(struct rte_eth_dev *dev)
1999 const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2000 const struct rte_eth_txmode *txmode = &dev->data->dev_conf.txmode;
2001 struct virtio_hw *hw = dev->data->dev_private;
2002 uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
2004 uint64_t rx_offloads = rxmode->offloads;
2005 uint64_t tx_offloads = txmode->offloads;
2006 uint64_t req_features;
2009 PMD_INIT_LOG(DEBUG, "configure");
2010 req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
2012 if (rxmode->mq_mode != ETH_MQ_RX_NONE) {
2014 "Unsupported Rx multi queue mode %d",
2019 if (txmode->mq_mode != ETH_MQ_TX_NONE) {
2021 "Unsupported Tx multi queue mode %d",
2026 if (dev->data->dev_conf.intr_conf.rxq) {
2027 ret = virtio_init_device(dev, hw->req_guest_features);
2032 if (rxmode->max_rx_pkt_len > hw->max_mtu + ether_hdr_len)
2033 req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
2035 if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2036 DEV_RX_OFFLOAD_TCP_CKSUM))
2037 req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
2039 if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO)
2041 (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2042 (1ULL << VIRTIO_NET_F_GUEST_TSO6);
2044 if (tx_offloads & (DEV_TX_OFFLOAD_UDP_CKSUM |
2045 DEV_TX_OFFLOAD_TCP_CKSUM))
2046 req_features |= (1ULL << VIRTIO_NET_F_CSUM);
2048 if (tx_offloads & DEV_TX_OFFLOAD_TCP_TSO)
2050 (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2051 (1ULL << VIRTIO_NET_F_HOST_TSO6);
2053 /* if request features changed, reinit the device */
2054 if (req_features != hw->req_guest_features) {
2055 ret = virtio_init_device(dev, req_features);
2060 if ((rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2061 DEV_RX_OFFLOAD_TCP_CKSUM)) &&
2062 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
2064 "rx checksum not available on this host");
2068 if ((rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) &&
2069 (!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2070 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6))) {
2072 "Large Receive Offload not available on this host");
2076 /* start control queue */
2077 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
2078 virtio_dev_cq_start(dev);
2080 if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
2083 if ((rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
2084 && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2086 "vlan filtering not available on this host");
2090 hw->has_tx_offload = tx_offload_enabled(hw);
2091 hw->has_rx_offload = rx_offload_enabled(hw);
2093 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2094 /* Enable vector (0) for Link State Intrerrupt */
2095 if (VTPCI_OPS(hw)->set_config_irq(hw, 0) ==
2096 VIRTIO_MSI_NO_VECTOR) {
2097 PMD_DRV_LOG(ERR, "failed to set config vector");
2101 if (vtpci_packed_queue(hw)) {
2102 #if defined(RTE_ARCH_X86_64) && defined(CC_AVX512_SUPPORT)
2103 if ((hw->use_vec_rx || hw->use_vec_tx) &&
2104 (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX512F) ||
2105 !vtpci_with_feature(hw, VIRTIO_F_IN_ORDER) ||
2106 !vtpci_with_feature(hw, VIRTIO_F_VERSION_1) ||
2107 rte_vect_get_max_simd_bitwidth() < RTE_VECT_SIMD_512)) {
2109 "disabled packed ring vectorized path for requirements not met");
2113 #elif defined(RTE_ARCH_ARM)
2114 if ((hw->use_vec_rx || hw->use_vec_tx) &&
2115 (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON) ||
2116 !vtpci_with_feature(hw, VIRTIO_F_IN_ORDER) ||
2117 !vtpci_with_feature(hw, VIRTIO_F_VERSION_1) ||
2118 rte_vect_get_max_simd_bitwidth() < RTE_VECT_SIMD_128)) {
2120 "disabled packed ring vectorized path for requirements not met");
2129 if (hw->use_vec_rx) {
2130 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2132 "disabled packed ring vectorized rx for mrg_rxbuf enabled");
2136 if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) {
2138 "disabled packed ring vectorized rx for TCP_LRO enabled");
2143 if (vtpci_with_feature(hw, VIRTIO_F_IN_ORDER)) {
2144 hw->use_inorder_tx = 1;
2145 hw->use_inorder_rx = 1;
2149 if (hw->use_vec_rx) {
2150 #if defined RTE_ARCH_ARM
2151 if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
2153 "disabled split ring vectorized path for requirement not met");
2157 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2159 "disabled split ring vectorized rx for mrg_rxbuf enabled");
2163 if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2164 DEV_RX_OFFLOAD_TCP_CKSUM |
2165 DEV_RX_OFFLOAD_TCP_LRO |
2166 DEV_RX_OFFLOAD_VLAN_STRIP)) {
2168 "disabled split ring vectorized rx for offloading enabled");
2172 if (rte_vect_get_max_simd_bitwidth() < RTE_VECT_SIMD_128) {
2174 "disabled split ring vectorized rx, max SIMD bitwidth too low");
2185 virtio_dev_start(struct rte_eth_dev *dev)
2187 uint16_t nb_queues, i;
2188 struct virtnet_rx *rxvq;
2189 struct virtnet_tx *txvq __rte_unused;
2190 struct virtio_hw *hw = dev->data->dev_private;
2193 /* Finish the initialization of the queues */
2194 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2195 ret = virtio_dev_rx_queue_setup_finish(dev, i);
2199 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2200 ret = virtio_dev_tx_queue_setup_finish(dev, i);
2205 /* check if lsc interrupt feature is enabled */
2206 if (dev->data->dev_conf.intr_conf.lsc) {
2207 if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
2208 PMD_DRV_LOG(ERR, "link status not supported by host");
2213 /* Enable uio/vfio intr/eventfd mapping: althrough we already did that
2214 * in device configure, but it could be unmapped when device is
2217 if (dev->data->dev_conf.intr_conf.lsc ||
2218 dev->data->dev_conf.intr_conf.rxq) {
2219 virtio_intr_disable(dev);
2221 /* Setup interrupt callback */
2222 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2223 rte_intr_callback_register(dev->intr_handle,
2224 virtio_interrupt_handler,
2227 if (virtio_intr_enable(dev) < 0) {
2228 PMD_DRV_LOG(ERR, "interrupt enable failed");
2233 /*Notify the backend
2234 *Otherwise the tap backend might already stop its queue due to fullness.
2235 *vhost backend will have no chance to be waked up
2237 nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
2238 if (hw->max_queue_pairs > 1) {
2239 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
2243 PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
2245 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2246 rxvq = dev->data->rx_queues[i];
2247 /* Flush the old packets */
2248 virtqueue_rxvq_flush(rxvq->vq);
2249 virtqueue_notify(rxvq->vq);
2252 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2253 txvq = dev->data->tx_queues[i];
2254 virtqueue_notify(txvq->vq);
2257 PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
2259 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2260 rxvq = dev->data->rx_queues[i];
2261 VIRTQUEUE_DUMP(rxvq->vq);
2264 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2265 txvq = dev->data->tx_queues[i];
2266 VIRTQUEUE_DUMP(txvq->vq);
2269 set_rxtx_funcs(dev);
2272 /* Initialize Link state */
2273 virtio_dev_link_update(dev, 0);
2278 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
2280 struct virtio_hw *hw = dev->data->dev_private;
2281 uint16_t nr_vq = virtio_get_nr_vq(hw);
2282 const char *type __rte_unused;
2283 unsigned int i, mbuf_num = 0;
2284 struct virtqueue *vq;
2285 struct rte_mbuf *buf;
2288 if (hw->vqs == NULL)
2291 for (i = 0; i < nr_vq; i++) {
2296 queue_type = virtio_get_queue_type(hw, i);
2297 if (queue_type == VTNET_RQ)
2299 else if (queue_type == VTNET_TQ)
2305 "Before freeing %s[%d] used and unused buf",
2309 while ((buf = virtqueue_detach_unused(vq)) != NULL) {
2310 rte_pktmbuf_free(buf);
2315 "After freeing %s[%d] used and unused buf",
2320 PMD_INIT_LOG(DEBUG, "%d mbufs freed", mbuf_num);
2324 * Stop device: disable interrupt and mark link down
2327 virtio_dev_stop(struct rte_eth_dev *dev)
2329 struct virtio_hw *hw = dev->data->dev_private;
2330 struct rte_eth_link link;
2331 struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
2333 PMD_INIT_LOG(DEBUG, "stop");
2334 dev->data->dev_started = 0;
2336 rte_spinlock_lock(&hw->state_lock);
2341 if (intr_conf->lsc || intr_conf->rxq) {
2342 virtio_intr_disable(dev);
2344 /* Reset interrupt callback */
2345 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
2346 rte_intr_callback_unregister(dev->intr_handle,
2347 virtio_interrupt_handler,
2352 memset(&link, 0, sizeof(link));
2353 rte_eth_linkstatus_set(dev, &link);
2355 rte_spinlock_unlock(&hw->state_lock);
2361 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
2363 struct rte_eth_link link;
2365 struct virtio_hw *hw = dev->data->dev_private;
2367 memset(&link, 0, sizeof(link));
2368 link.link_duplex = hw->duplex;
2369 link.link_speed = hw->speed;
2370 link.link_autoneg = ETH_LINK_AUTONEG;
2373 link.link_status = ETH_LINK_DOWN;
2374 link.link_speed = ETH_SPEED_NUM_NONE;
2375 } else if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
2376 PMD_INIT_LOG(DEBUG, "Get link status from hw");
2377 vtpci_read_dev_config(hw,
2378 offsetof(struct virtio_net_config, status),
2379 &status, sizeof(status));
2380 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
2381 link.link_status = ETH_LINK_DOWN;
2382 link.link_speed = ETH_SPEED_NUM_NONE;
2383 PMD_INIT_LOG(DEBUG, "Port %d is down",
2384 dev->data->port_id);
2386 link.link_status = ETH_LINK_UP;
2387 PMD_INIT_LOG(DEBUG, "Port %d is up",
2388 dev->data->port_id);
2391 link.link_status = ETH_LINK_UP;
2394 return rte_eth_linkstatus_set(dev, &link);
2398 virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
2400 const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2401 struct virtio_hw *hw = dev->data->dev_private;
2402 uint64_t offloads = rxmode->offloads;
2404 if (mask & ETH_VLAN_FILTER_MASK) {
2405 if ((offloads & DEV_RX_OFFLOAD_VLAN_FILTER) &&
2406 !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2409 "vlan filtering not available on this host");
2415 if (mask & ETH_VLAN_STRIP_MASK)
2416 hw->vlan_strip = !!(offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
2422 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2424 uint64_t tso_mask, host_features;
2425 struct virtio_hw *hw = dev->data->dev_private;
2426 dev_info->speed_capa = virtio_dev_speed_capa_get(hw->speed);
2428 dev_info->max_rx_queues =
2429 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
2430 dev_info->max_tx_queues =
2431 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
2432 dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
2433 dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
2434 dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
2436 host_features = VTPCI_OPS(hw)->get_features(hw);
2437 dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
2438 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_JUMBO_FRAME;
2439 if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
2440 dev_info->rx_offload_capa |=
2441 DEV_RX_OFFLOAD_TCP_CKSUM |
2442 DEV_RX_OFFLOAD_UDP_CKSUM;
2444 if (host_features & (1ULL << VIRTIO_NET_F_CTRL_VLAN))
2445 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_VLAN_FILTER;
2446 tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2447 (1ULL << VIRTIO_NET_F_GUEST_TSO6);
2448 if ((host_features & tso_mask) == tso_mask)
2449 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
2451 dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
2452 DEV_TX_OFFLOAD_VLAN_INSERT;
2453 if (host_features & (1ULL << VIRTIO_NET_F_CSUM)) {
2454 dev_info->tx_offload_capa |=
2455 DEV_TX_OFFLOAD_UDP_CKSUM |
2456 DEV_TX_OFFLOAD_TCP_CKSUM;
2458 tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2459 (1ULL << VIRTIO_NET_F_HOST_TSO6);
2460 if ((host_features & tso_mask) == tso_mask)
2461 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
2467 * It enables testpmd to collect per queue stats.
2470 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
2471 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
2472 __rte_unused uint8_t is_rx)
2477 RTE_LOG_REGISTER(virtio_logtype_init, pmd.net.virtio.init, NOTICE);
2478 RTE_LOG_REGISTER(virtio_logtype_driver, pmd.net.virtio.driver, NOTICE);