1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2016 Intel Corporation
11 #include <ethdev_driver.h>
12 #include <rte_memcpy.h>
13 #include <rte_string_fns.h>
14 #include <rte_memzone.h>
15 #include <rte_malloc.h>
16 #include <rte_branch_prediction.h>
17 #include <rte_ether.h>
20 #include <rte_common.h>
21 #include <rte_errno.h>
22 #include <rte_cpuflags.h>
24 #include <rte_memory.h>
25 #include <rte_eal_paging.h>
28 #include <rte_cycles.h>
29 #include <rte_kvargs.h>
31 #include "virtio_ethdev.h"
33 #include "virtio_logs.h"
34 #include "virtqueue.h"
35 #include "virtio_rxtx.h"
36 #include "virtio_user/virtio_user_dev.h"
38 static int virtio_dev_configure(struct rte_eth_dev *dev);
39 static int virtio_dev_start(struct rte_eth_dev *dev);
40 static int virtio_dev_promiscuous_enable(struct rte_eth_dev *dev);
41 static int virtio_dev_promiscuous_disable(struct rte_eth_dev *dev);
42 static int virtio_dev_allmulticast_enable(struct rte_eth_dev *dev);
43 static int virtio_dev_allmulticast_disable(struct rte_eth_dev *dev);
44 static uint32_t virtio_dev_speed_capa_get(uint32_t speed);
45 static int virtio_dev_devargs_parse(struct rte_devargs *devargs,
48 static int virtio_dev_info_get(struct rte_eth_dev *dev,
49 struct rte_eth_dev_info *dev_info);
50 static int virtio_dev_link_update(struct rte_eth_dev *dev,
51 int wait_to_complete);
52 static int virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask);
54 static void virtio_set_hwaddr(struct virtio_hw *hw);
55 static void virtio_get_hwaddr(struct virtio_hw *hw);
57 static int virtio_dev_stats_get(struct rte_eth_dev *dev,
58 struct rte_eth_stats *stats);
59 static int virtio_dev_xstats_get(struct rte_eth_dev *dev,
60 struct rte_eth_xstat *xstats, unsigned n);
61 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
62 struct rte_eth_xstat_name *xstats_names,
64 static int virtio_dev_stats_reset(struct rte_eth_dev *dev);
65 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev);
66 static int virtio_vlan_filter_set(struct rte_eth_dev *dev,
67 uint16_t vlan_id, int on);
68 static int virtio_mac_addr_add(struct rte_eth_dev *dev,
69 struct rte_ether_addr *mac_addr,
70 uint32_t index, uint32_t vmdq);
71 static void virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index);
72 static int virtio_mac_addr_set(struct rte_eth_dev *dev,
73 struct rte_ether_addr *mac_addr);
75 static int virtio_intr_disable(struct rte_eth_dev *dev);
77 static int virtio_dev_queue_stats_mapping_set(
78 struct rte_eth_dev *eth_dev,
83 static void virtio_notify_peers(struct rte_eth_dev *dev);
84 static void virtio_ack_link_announce(struct rte_eth_dev *dev);
86 struct rte_virtio_xstats_name_off {
87 char name[RTE_ETH_XSTATS_NAME_SIZE];
91 /* [rt]x_qX_ is prepended to the name string here */
92 static const struct rte_virtio_xstats_name_off rte_virtio_rxq_stat_strings[] = {
93 {"good_packets", offsetof(struct virtnet_rx, stats.packets)},
94 {"good_bytes", offsetof(struct virtnet_rx, stats.bytes)},
95 {"errors", offsetof(struct virtnet_rx, stats.errors)},
96 {"multicast_packets", offsetof(struct virtnet_rx, stats.multicast)},
97 {"broadcast_packets", offsetof(struct virtnet_rx, stats.broadcast)},
98 {"undersize_packets", offsetof(struct virtnet_rx, stats.size_bins[0])},
99 {"size_64_packets", offsetof(struct virtnet_rx, stats.size_bins[1])},
100 {"size_65_127_packets", offsetof(struct virtnet_rx, stats.size_bins[2])},
101 {"size_128_255_packets", offsetof(struct virtnet_rx, stats.size_bins[3])},
102 {"size_256_511_packets", offsetof(struct virtnet_rx, stats.size_bins[4])},
103 {"size_512_1023_packets", offsetof(struct virtnet_rx, stats.size_bins[5])},
104 {"size_1024_1518_packets", offsetof(struct virtnet_rx, stats.size_bins[6])},
105 {"size_1519_max_packets", offsetof(struct virtnet_rx, stats.size_bins[7])},
108 /* [rt]x_qX_ is prepended to the name string here */
109 static const struct rte_virtio_xstats_name_off rte_virtio_txq_stat_strings[] = {
110 {"good_packets", offsetof(struct virtnet_tx, stats.packets)},
111 {"good_bytes", offsetof(struct virtnet_tx, stats.bytes)},
112 {"multicast_packets", offsetof(struct virtnet_tx, stats.multicast)},
113 {"broadcast_packets", offsetof(struct virtnet_tx, stats.broadcast)},
114 {"undersize_packets", offsetof(struct virtnet_tx, stats.size_bins[0])},
115 {"size_64_packets", offsetof(struct virtnet_tx, stats.size_bins[1])},
116 {"size_65_127_packets", offsetof(struct virtnet_tx, stats.size_bins[2])},
117 {"size_128_255_packets", offsetof(struct virtnet_tx, stats.size_bins[3])},
118 {"size_256_511_packets", offsetof(struct virtnet_tx, stats.size_bins[4])},
119 {"size_512_1023_packets", offsetof(struct virtnet_tx, stats.size_bins[5])},
120 {"size_1024_1518_packets", offsetof(struct virtnet_tx, stats.size_bins[6])},
121 {"size_1519_max_packets", offsetof(struct virtnet_tx, stats.size_bins[7])},
124 #define VIRTIO_NB_RXQ_XSTATS (sizeof(rte_virtio_rxq_stat_strings) / \
125 sizeof(rte_virtio_rxq_stat_strings[0]))
126 #define VIRTIO_NB_TXQ_XSTATS (sizeof(rte_virtio_txq_stat_strings) / \
127 sizeof(rte_virtio_txq_stat_strings[0]))
129 struct virtio_hw_internal virtio_hw_internal[RTE_MAX_ETHPORTS];
131 static struct virtio_pmd_ctrl *
132 virtio_send_command_packed(struct virtnet_ctl *cvq,
133 struct virtio_pmd_ctrl *ctrl,
134 int *dlen, int pkt_num)
136 struct virtqueue *vq = virtnet_cq_to_vq(cvq);
138 struct vring_packed_desc *desc = vq->vq_packed.ring.desc;
139 struct virtio_pmd_ctrl *result;
146 * Format is enforced in qemu code:
147 * One TX packet for header;
148 * At least one TX packet per argument;
149 * One RX packet for ACK.
151 head = vq->vq_avail_idx;
152 flags = vq->vq_packed.cached_flags;
153 desc[head].addr = cvq->virtio_net_hdr_mem;
154 desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
157 if (++vq->vq_avail_idx >= vq->vq_nentries) {
158 vq->vq_avail_idx -= vq->vq_nentries;
159 vq->vq_packed.cached_flags ^= VRING_PACKED_DESC_F_AVAIL_USED;
162 for (k = 0; k < pkt_num; k++) {
163 desc[vq->vq_avail_idx].addr = cvq->virtio_net_hdr_mem
164 + sizeof(struct virtio_net_ctrl_hdr)
165 + sizeof(ctrl->status) + sizeof(uint8_t) * sum;
166 desc[vq->vq_avail_idx].len = dlen[k];
167 desc[vq->vq_avail_idx].flags = VRING_DESC_F_NEXT |
168 vq->vq_packed.cached_flags;
172 if (++vq->vq_avail_idx >= vq->vq_nentries) {
173 vq->vq_avail_idx -= vq->vq_nentries;
174 vq->vq_packed.cached_flags ^=
175 VRING_PACKED_DESC_F_AVAIL_USED;
179 desc[vq->vq_avail_idx].addr = cvq->virtio_net_hdr_mem
180 + sizeof(struct virtio_net_ctrl_hdr);
181 desc[vq->vq_avail_idx].len = sizeof(ctrl->status);
182 desc[vq->vq_avail_idx].flags = VRING_DESC_F_WRITE |
183 vq->vq_packed.cached_flags;
186 if (++vq->vq_avail_idx >= vq->vq_nentries) {
187 vq->vq_avail_idx -= vq->vq_nentries;
188 vq->vq_packed.cached_flags ^= VRING_PACKED_DESC_F_AVAIL_USED;
191 virtqueue_store_flags_packed(&desc[head], VRING_DESC_F_NEXT | flags,
192 vq->hw->weak_barriers);
194 virtio_wmb(vq->hw->weak_barriers);
195 virtqueue_notify(vq);
197 /* wait for used desc in virtqueue
198 * desc_is_used has a load-acquire or rte_io_rmb inside
200 while (!desc_is_used(&desc[head], vq))
203 /* now get used descriptors */
204 vq->vq_free_cnt += nb_descs;
205 vq->vq_used_cons_idx += nb_descs;
206 if (vq->vq_used_cons_idx >= vq->vq_nentries) {
207 vq->vq_used_cons_idx -= vq->vq_nentries;
208 vq->vq_packed.used_wrap_counter ^= 1;
211 PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\n"
212 "vq->vq_avail_idx=%d\n"
213 "vq->vq_used_cons_idx=%d\n"
214 "vq->vq_packed.cached_flags=0x%x\n"
215 "vq->vq_packed.used_wrap_counter=%d\n",
218 vq->vq_used_cons_idx,
219 vq->vq_packed.cached_flags,
220 vq->vq_packed.used_wrap_counter);
222 result = cvq->virtio_net_hdr_mz->addr;
226 static struct virtio_pmd_ctrl *
227 virtio_send_command_split(struct virtnet_ctl *cvq,
228 struct virtio_pmd_ctrl *ctrl,
229 int *dlen, int pkt_num)
231 struct virtio_pmd_ctrl *result;
232 struct virtqueue *vq = virtnet_cq_to_vq(cvq);
236 head = vq->vq_desc_head_idx;
239 * Format is enforced in qemu code:
240 * One TX packet for header;
241 * At least one TX packet per argument;
242 * One RX packet for ACK.
244 vq->vq_split.ring.desc[head].flags = VRING_DESC_F_NEXT;
245 vq->vq_split.ring.desc[head].addr = cvq->virtio_net_hdr_mem;
246 vq->vq_split.ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
248 i = vq->vq_split.ring.desc[head].next;
250 for (k = 0; k < pkt_num; k++) {
251 vq->vq_split.ring.desc[i].flags = VRING_DESC_F_NEXT;
252 vq->vq_split.ring.desc[i].addr = cvq->virtio_net_hdr_mem
253 + sizeof(struct virtio_net_ctrl_hdr)
254 + sizeof(ctrl->status) + sizeof(uint8_t)*sum;
255 vq->vq_split.ring.desc[i].len = dlen[k];
258 i = vq->vq_split.ring.desc[i].next;
261 vq->vq_split.ring.desc[i].flags = VRING_DESC_F_WRITE;
262 vq->vq_split.ring.desc[i].addr = cvq->virtio_net_hdr_mem
263 + sizeof(struct virtio_net_ctrl_hdr);
264 vq->vq_split.ring.desc[i].len = sizeof(ctrl->status);
267 vq->vq_desc_head_idx = vq->vq_split.ring.desc[i].next;
269 vq_update_avail_ring(vq, head);
270 vq_update_avail_idx(vq);
272 PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
274 virtqueue_notify(vq);
276 while (virtqueue_nused(vq) == 0)
279 while (virtqueue_nused(vq)) {
280 uint32_t idx, desc_idx, used_idx;
281 struct vring_used_elem *uep;
283 used_idx = (uint32_t)(vq->vq_used_cons_idx
284 & (vq->vq_nentries - 1));
285 uep = &vq->vq_split.ring.used->ring[used_idx];
286 idx = (uint32_t) uep->id;
289 while (vq->vq_split.ring.desc[desc_idx].flags &
291 desc_idx = vq->vq_split.ring.desc[desc_idx].next;
295 vq->vq_split.ring.desc[desc_idx].next = vq->vq_desc_head_idx;
296 vq->vq_desc_head_idx = idx;
298 vq->vq_used_cons_idx++;
302 PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
303 vq->vq_free_cnt, vq->vq_desc_head_idx);
305 result = cvq->virtio_net_hdr_mz->addr;
310 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
311 int *dlen, int pkt_num)
313 virtio_net_ctrl_ack status = ~0;
314 struct virtio_pmd_ctrl *result;
315 struct virtqueue *vq;
317 ctrl->status = status;
320 PMD_INIT_LOG(ERR, "Control queue is not supported.");
324 rte_spinlock_lock(&cvq->lock);
325 vq = virtnet_cq_to_vq(cvq);
327 PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
328 "vq->hw->cvq = %p vq = %p",
329 vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
331 if (vq->vq_free_cnt < pkt_num + 2 || pkt_num < 1) {
332 rte_spinlock_unlock(&cvq->lock);
336 memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
337 sizeof(struct virtio_pmd_ctrl));
339 if (virtio_with_packed_queue(vq->hw))
340 result = virtio_send_command_packed(cvq, ctrl, dlen, pkt_num);
342 result = virtio_send_command_split(cvq, ctrl, dlen, pkt_num);
344 rte_spinlock_unlock(&cvq->lock);
345 return result->status;
349 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
351 struct virtio_hw *hw = dev->data->dev_private;
352 struct virtio_pmd_ctrl ctrl;
356 ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
357 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
358 memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
360 dlen[0] = sizeof(uint16_t);
362 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
364 PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
365 "failed, this is too late now...");
373 virtio_dev_queue_release(void *queue __rte_unused)
379 virtio_get_nr_vq(struct virtio_hw *hw)
381 uint16_t nr_vq = hw->max_queue_pairs * 2;
383 if (virtio_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
390 virtio_init_vring(struct virtqueue *vq)
392 int size = vq->vq_nentries;
393 uint8_t *ring_mem = vq->vq_ring_virt_mem;
395 PMD_INIT_FUNC_TRACE();
397 memset(ring_mem, 0, vq->vq_ring_size);
399 vq->vq_used_cons_idx = 0;
400 vq->vq_desc_head_idx = 0;
401 vq->vq_avail_idx = 0;
402 vq->vq_desc_tail_idx = (uint16_t)(vq->vq_nentries - 1);
403 vq->vq_free_cnt = vq->vq_nentries;
404 memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries);
405 if (virtio_with_packed_queue(vq->hw)) {
406 vring_init_packed(&vq->vq_packed.ring, ring_mem,
407 VIRTIO_VRING_ALIGN, size);
408 vring_desc_init_packed(vq, size);
410 struct vring *vr = &vq->vq_split.ring;
412 vring_init_split(vr, ring_mem, VIRTIO_VRING_ALIGN, size);
413 vring_desc_init_split(vr->desc, size);
416 * Disable device(host) interrupting guest
418 virtqueue_disable_intr(vq);
422 virtio_init_queue(struct rte_eth_dev *dev, uint16_t queue_idx)
424 char vq_name[VIRTQUEUE_MAX_NAME_SZ];
425 char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
426 const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
427 unsigned int vq_size, size;
428 struct virtio_hw *hw = dev->data->dev_private;
429 struct virtnet_rx *rxvq = NULL;
430 struct virtnet_tx *txvq = NULL;
431 struct virtnet_ctl *cvq = NULL;
432 struct virtqueue *vq;
433 size_t sz_hdr_mz = 0;
434 void *sw_ring = NULL;
435 int queue_type = virtio_get_queue_type(hw, queue_idx);
437 int numa_node = dev->device->numa_node;
438 struct rte_mbuf *fake_mbuf = NULL;
440 PMD_INIT_LOG(INFO, "setting up queue: %u on NUMA node %d",
441 queue_idx, numa_node);
444 * Read the virtqueue size from the Queue Size field
445 * Always power of 2 and if 0 virtqueue does not exist
447 vq_size = VIRTIO_OPS(hw)->get_queue_num(hw, queue_idx);
448 PMD_INIT_LOG(DEBUG, "vq_size: %u", vq_size);
450 PMD_INIT_LOG(ERR, "virtqueue does not exist");
454 if (!virtio_with_packed_queue(hw) && !rte_is_power_of_2(vq_size)) {
455 PMD_INIT_LOG(ERR, "split virtqueue size is not power of 2");
459 snprintf(vq_name, sizeof(vq_name), "port%d_vq%d",
460 dev->data->port_id, queue_idx);
462 size = RTE_ALIGN_CEIL(sizeof(*vq) +
463 vq_size * sizeof(struct vq_desc_extra),
464 RTE_CACHE_LINE_SIZE);
465 if (queue_type == VTNET_TQ) {
467 * For each xmit packet, allocate a virtio_net_hdr
468 * and indirect ring elements
470 sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
471 } else if (queue_type == VTNET_CQ) {
472 /* Allocate a page for control vq command, data and status */
473 sz_hdr_mz = rte_mem_page_size();
476 vq = rte_zmalloc_socket(vq_name, size, RTE_CACHE_LINE_SIZE,
479 PMD_INIT_LOG(ERR, "can not allocate vq");
482 hw->vqs[queue_idx] = vq;
485 vq->vq_queue_index = queue_idx;
486 vq->vq_nentries = vq_size;
487 if (virtio_with_packed_queue(hw)) {
488 vq->vq_packed.used_wrap_counter = 1;
489 vq->vq_packed.cached_flags = VRING_PACKED_DESC_F_AVAIL;
490 vq->vq_packed.event_flags_shadow = 0;
491 if (queue_type == VTNET_RQ)
492 vq->vq_packed.cached_flags |= VRING_DESC_F_WRITE;
496 * Reserve a memzone for vring elements
498 size = vring_size(hw, vq_size, VIRTIO_VRING_ALIGN);
499 vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_VRING_ALIGN);
500 PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
501 size, vq->vq_ring_size);
503 mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
504 numa_node, RTE_MEMZONE_IOVA_CONTIG,
507 if (rte_errno == EEXIST)
508 mz = rte_memzone_lookup(vq_name);
515 memset(mz->addr, 0, mz->len);
517 vq->vq_ring_mem = mz->iova;
518 vq->vq_ring_virt_mem = mz->addr;
519 PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem: 0x%" PRIx64,
521 PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64,
522 (uint64_t)(uintptr_t)mz->addr);
524 virtio_init_vring(vq);
527 snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_vq%d_hdr",
528 dev->data->port_id, queue_idx);
529 hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
530 numa_node, RTE_MEMZONE_IOVA_CONTIG,
531 RTE_CACHE_LINE_SIZE);
532 if (hdr_mz == NULL) {
533 if (rte_errno == EEXIST)
534 hdr_mz = rte_memzone_lookup(vq_hdr_name);
535 if (hdr_mz == NULL) {
542 if (queue_type == VTNET_RQ) {
543 size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
544 sizeof(vq->sw_ring[0]);
546 sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
547 RTE_CACHE_LINE_SIZE, numa_node);
549 PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
554 fake_mbuf = rte_zmalloc_socket("sw_ring", sizeof(*fake_mbuf),
555 RTE_CACHE_LINE_SIZE, numa_node);
557 PMD_INIT_LOG(ERR, "can not allocate fake mbuf");
562 vq->sw_ring = sw_ring;
564 rxvq->port_id = dev->data->port_id;
566 rxvq->fake_mbuf = fake_mbuf;
567 } else if (queue_type == VTNET_TQ) {
569 txvq->port_id = dev->data->port_id;
571 txvq->virtio_net_hdr_mz = hdr_mz;
572 txvq->virtio_net_hdr_mem = hdr_mz->iova;
573 } else if (queue_type == VTNET_CQ) {
576 cvq->virtio_net_hdr_mz = hdr_mz;
577 cvq->virtio_net_hdr_mem = hdr_mz->iova;
578 memset(cvq->virtio_net_hdr_mz->addr, 0, rte_mem_page_size());
583 if (queue_type == VTNET_TQ) {
584 struct virtio_tx_region *txr;
588 memset(txr, 0, vq_size * sizeof(*txr));
589 for (i = 0; i < vq_size; i++) {
590 /* first indirect descriptor is always the tx header */
591 if (!virtio_with_packed_queue(hw)) {
592 struct vring_desc *start_dp = txr[i].tx_indir;
593 vring_desc_init_split(start_dp,
594 RTE_DIM(txr[i].tx_indir));
595 start_dp->addr = txvq->virtio_net_hdr_mem
597 + offsetof(struct virtio_tx_region,
599 start_dp->len = hw->vtnet_hdr_size;
600 start_dp->flags = VRING_DESC_F_NEXT;
602 struct vring_packed_desc *start_dp =
603 txr[i].tx_packed_indir;
604 vring_desc_init_indirect_packed(start_dp,
605 RTE_DIM(txr[i].tx_packed_indir));
606 start_dp->addr = txvq->virtio_net_hdr_mem
608 + offsetof(struct virtio_tx_region,
610 start_dp->len = hw->vtnet_hdr_size;
615 if (VIRTIO_OPS(hw)->setup_queue(hw, vq) < 0) {
616 PMD_INIT_LOG(ERR, "setup_queue failed");
629 rte_memzone_free(hdr_mz);
631 rte_memzone_free(mz);
639 virtio_free_queues(struct virtio_hw *hw)
641 uint16_t nr_vq = virtio_get_nr_vq(hw);
642 struct virtqueue *vq;
649 for (i = 0; i < nr_vq; i++) {
654 queue_type = virtio_get_queue_type(hw, i);
655 if (queue_type == VTNET_RQ) {
656 rte_free(vq->rxq.fake_mbuf);
657 rte_free(vq->sw_ring);
658 rte_memzone_free(vq->rxq.mz);
659 } else if (queue_type == VTNET_TQ) {
660 rte_memzone_free(vq->txq.mz);
661 rte_memzone_free(vq->txq.virtio_net_hdr_mz);
663 rte_memzone_free(vq->cq.mz);
664 rte_memzone_free(vq->cq.virtio_net_hdr_mz);
676 virtio_alloc_queues(struct rte_eth_dev *dev)
678 struct virtio_hw *hw = dev->data->dev_private;
679 uint16_t nr_vq = virtio_get_nr_vq(hw);
683 hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
685 PMD_INIT_LOG(ERR, "failed to allocate vqs");
689 for (i = 0; i < nr_vq; i++) {
690 ret = virtio_init_queue(dev, i);
692 virtio_free_queues(hw);
700 static void virtio_queues_unbind_intr(struct rte_eth_dev *dev);
703 virtio_dev_close(struct rte_eth_dev *dev)
705 struct virtio_hw *hw = dev->data->dev_private;
706 struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
708 PMD_INIT_LOG(DEBUG, "virtio_dev_close");
709 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
717 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
718 VIRTIO_OPS(hw)->set_config_irq(hw, VIRTIO_MSI_NO_VECTOR);
720 virtio_queues_unbind_intr(dev);
722 if (intr_conf->lsc || intr_conf->rxq) {
723 virtio_intr_disable(dev);
724 rte_intr_efd_disable(dev->intr_handle);
725 rte_free(dev->intr_handle->intr_vec);
726 dev->intr_handle->intr_vec = NULL;
730 virtio_dev_free_mbufs(dev);
731 virtio_free_queues(hw);
733 return VIRTIO_OPS(hw)->dev_close(hw);
737 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
739 struct virtio_hw *hw = dev->data->dev_private;
740 struct virtio_pmd_ctrl ctrl;
744 if (!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
745 PMD_INIT_LOG(INFO, "host does not support rx control");
749 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
750 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
754 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
756 PMD_INIT_LOG(ERR, "Failed to enable promisc");
764 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
766 struct virtio_hw *hw = dev->data->dev_private;
767 struct virtio_pmd_ctrl ctrl;
771 if (!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
772 PMD_INIT_LOG(INFO, "host does not support rx control");
776 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
777 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
781 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
783 PMD_INIT_LOG(ERR, "Failed to disable promisc");
791 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
793 struct virtio_hw *hw = dev->data->dev_private;
794 struct virtio_pmd_ctrl ctrl;
798 if (!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
799 PMD_INIT_LOG(INFO, "host does not support rx control");
803 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
804 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
808 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
810 PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
818 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
820 struct virtio_hw *hw = dev->data->dev_private;
821 struct virtio_pmd_ctrl ctrl;
825 if (!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
826 PMD_INIT_LOG(INFO, "host does not support rx control");
830 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
831 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
835 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
837 PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
845 virtio_rx_mem_pool_buf_size(struct rte_mempool *mp)
847 return rte_pktmbuf_data_room_size(mp) - RTE_PKTMBUF_HEADROOM;
851 virtio_rx_check_scatter(uint16_t max_rx_pkt_len, uint16_t rx_buf_size,
852 bool rx_scatter_enabled, const char **error)
854 if (!rx_scatter_enabled && max_rx_pkt_len > rx_buf_size) {
855 *error = "Rx scatter is disabled and RxQ mbuf pool object size is too small";
863 virtio_check_scatter_on_all_rx_queues(struct rte_eth_dev *dev,
866 struct virtio_hw *hw = dev->data->dev_private;
867 struct virtnet_rx *rxvq;
868 struct virtqueue *vq;
876 for (qidx = 0; (vq = hw->vqs[2 * qidx + VTNET_SQ_RQ_QUEUE_IDX]) != NULL;
879 if (rxvq->mpool == NULL)
881 buf_size = virtio_rx_mem_pool_buf_size(rxvq->mpool);
883 if (!virtio_rx_check_scatter(frame_size, buf_size,
884 hw->rx_ol_scatter, &error)) {
885 PMD_INIT_LOG(ERR, "MTU check for RxQ %u failed: %s",
894 #define VLAN_TAG_LEN 4 /* 802.3ac tag (not DMA'd) */
896 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
898 struct virtio_hw *hw = dev->data->dev_private;
899 uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
901 uint32_t frame_size = mtu + ether_hdr_len;
902 uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
904 max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
906 if (mtu < RTE_ETHER_MIN_MTU || frame_size > max_frame_size) {
907 PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
908 RTE_ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
912 if (!virtio_check_scatter_on_all_rx_queues(dev, frame_size)) {
913 PMD_INIT_LOG(ERR, "MTU vs Rx scatter and Rx buffers check failed");
917 hw->max_rx_pkt_len = frame_size;
918 dev->data->dev_conf.rxmode.max_rx_pkt_len = hw->max_rx_pkt_len;
924 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
926 struct virtio_hw *hw = dev->data->dev_private;
927 struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
928 struct virtqueue *vq = virtnet_rxq_to_vq(rxvq);
930 virtqueue_enable_intr(vq);
931 virtio_mb(hw->weak_barriers);
936 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
938 struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
939 struct virtqueue *vq = virtnet_rxq_to_vq(rxvq);
941 virtqueue_disable_intr(vq);
946 * dev_ops for virtio, bare necessities for basic operation
948 static const struct eth_dev_ops virtio_eth_dev_ops = {
949 .dev_configure = virtio_dev_configure,
950 .dev_start = virtio_dev_start,
951 .dev_stop = virtio_dev_stop,
952 .dev_close = virtio_dev_close,
953 .promiscuous_enable = virtio_dev_promiscuous_enable,
954 .promiscuous_disable = virtio_dev_promiscuous_disable,
955 .allmulticast_enable = virtio_dev_allmulticast_enable,
956 .allmulticast_disable = virtio_dev_allmulticast_disable,
957 .mtu_set = virtio_mtu_set,
958 .dev_infos_get = virtio_dev_info_get,
959 .stats_get = virtio_dev_stats_get,
960 .xstats_get = virtio_dev_xstats_get,
961 .xstats_get_names = virtio_dev_xstats_get_names,
962 .stats_reset = virtio_dev_stats_reset,
963 .xstats_reset = virtio_dev_stats_reset,
964 .link_update = virtio_dev_link_update,
965 .vlan_offload_set = virtio_dev_vlan_offload_set,
966 .rx_queue_setup = virtio_dev_rx_queue_setup,
967 .rx_queue_intr_enable = virtio_dev_rx_queue_intr_enable,
968 .rx_queue_intr_disable = virtio_dev_rx_queue_intr_disable,
969 .rx_queue_release = virtio_dev_queue_release,
970 .tx_queue_setup = virtio_dev_tx_queue_setup,
971 .tx_queue_release = virtio_dev_queue_release,
972 /* collect stats per queue */
973 .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
974 .vlan_filter_set = virtio_vlan_filter_set,
975 .mac_addr_add = virtio_mac_addr_add,
976 .mac_addr_remove = virtio_mac_addr_remove,
977 .mac_addr_set = virtio_mac_addr_set,
981 * dev_ops for virtio-user in secondary processes, as we just have
982 * some limited supports currently.
984 const struct eth_dev_ops virtio_user_secondary_eth_dev_ops = {
985 .dev_infos_get = virtio_dev_info_get,
986 .stats_get = virtio_dev_stats_get,
987 .xstats_get = virtio_dev_xstats_get,
988 .xstats_get_names = virtio_dev_xstats_get_names,
989 .stats_reset = virtio_dev_stats_reset,
990 .xstats_reset = virtio_dev_stats_reset,
991 /* collect stats per queue */
992 .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
996 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1000 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1001 const struct virtnet_tx *txvq = dev->data->tx_queues[i];
1005 stats->opackets += txvq->stats.packets;
1006 stats->obytes += txvq->stats.bytes;
1008 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
1009 stats->q_opackets[i] = txvq->stats.packets;
1010 stats->q_obytes[i] = txvq->stats.bytes;
1014 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1015 const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1019 stats->ipackets += rxvq->stats.packets;
1020 stats->ibytes += rxvq->stats.bytes;
1021 stats->ierrors += rxvq->stats.errors;
1023 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
1024 stats->q_ipackets[i] = rxvq->stats.packets;
1025 stats->q_ibytes[i] = rxvq->stats.bytes;
1029 stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
1032 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
1033 struct rte_eth_xstat_name *xstats_names,
1034 __rte_unused unsigned limit)
1040 unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
1041 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
1043 if (xstats_names != NULL) {
1044 /* Note: limit checked in rte_eth_xstats_names() */
1046 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1047 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1050 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1051 snprintf(xstats_names[count].name,
1052 sizeof(xstats_names[count].name),
1054 rte_virtio_rxq_stat_strings[t].name);
1059 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1060 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1063 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1064 snprintf(xstats_names[count].name,
1065 sizeof(xstats_names[count].name),
1067 rte_virtio_txq_stat_strings[t].name);
1077 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
1083 unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
1084 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
1089 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1090 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1097 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1098 xstats[count].value = *(uint64_t *)(((char *)rxvq) +
1099 rte_virtio_rxq_stat_strings[t].offset);
1100 xstats[count].id = count;
1105 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1106 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1113 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1114 xstats[count].value = *(uint64_t *)(((char *)txvq) +
1115 rte_virtio_txq_stat_strings[t].offset);
1116 xstats[count].id = count;
1125 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1127 virtio_update_stats(dev, stats);
1133 virtio_dev_stats_reset(struct rte_eth_dev *dev)
1137 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1138 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1142 txvq->stats.packets = 0;
1143 txvq->stats.bytes = 0;
1144 txvq->stats.multicast = 0;
1145 txvq->stats.broadcast = 0;
1146 memset(txvq->stats.size_bins, 0,
1147 sizeof(txvq->stats.size_bins[0]) * 8);
1150 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1151 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1155 rxvq->stats.packets = 0;
1156 rxvq->stats.bytes = 0;
1157 rxvq->stats.errors = 0;
1158 rxvq->stats.multicast = 0;
1159 rxvq->stats.broadcast = 0;
1160 memset(rxvq->stats.size_bins, 0,
1161 sizeof(rxvq->stats.size_bins[0]) * 8);
1168 virtio_set_hwaddr(struct virtio_hw *hw)
1170 virtio_write_dev_config(hw,
1171 offsetof(struct virtio_net_config, mac),
1172 &hw->mac_addr, RTE_ETHER_ADDR_LEN);
1176 virtio_get_hwaddr(struct virtio_hw *hw)
1178 if (virtio_with_feature(hw, VIRTIO_NET_F_MAC)) {
1179 virtio_read_dev_config(hw,
1180 offsetof(struct virtio_net_config, mac),
1181 &hw->mac_addr, RTE_ETHER_ADDR_LEN);
1183 rte_eth_random_addr(&hw->mac_addr[0]);
1184 virtio_set_hwaddr(hw);
1189 virtio_mac_table_set(struct virtio_hw *hw,
1190 const struct virtio_net_ctrl_mac *uc,
1191 const struct virtio_net_ctrl_mac *mc)
1193 struct virtio_pmd_ctrl ctrl;
1196 if (!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1197 PMD_DRV_LOG(INFO, "host does not support mac table");
1201 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1202 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1204 len[0] = uc->entries * RTE_ETHER_ADDR_LEN + sizeof(uc->entries);
1205 memcpy(ctrl.data, uc, len[0]);
1207 len[1] = mc->entries * RTE_ETHER_ADDR_LEN + sizeof(mc->entries);
1208 memcpy(ctrl.data + len[0], mc, len[1]);
1210 err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1212 PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1217 virtio_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
1218 uint32_t index, uint32_t vmdq __rte_unused)
1220 struct virtio_hw *hw = dev->data->dev_private;
1221 const struct rte_ether_addr *addrs = dev->data->mac_addrs;
1223 struct virtio_net_ctrl_mac *uc, *mc;
1225 if (index >= VIRTIO_MAX_MAC_ADDRS) {
1226 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1230 uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1231 sizeof(uc->entries));
1233 mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1234 sizeof(mc->entries));
1237 for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1238 const struct rte_ether_addr *addr
1239 = (i == index) ? mac_addr : addrs + i;
1240 struct virtio_net_ctrl_mac *tbl
1241 = rte_is_multicast_ether_addr(addr) ? mc : uc;
1243 memcpy(&tbl->macs[tbl->entries++], addr, RTE_ETHER_ADDR_LEN);
1246 return virtio_mac_table_set(hw, uc, mc);
1250 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1252 struct virtio_hw *hw = dev->data->dev_private;
1253 struct rte_ether_addr *addrs = dev->data->mac_addrs;
1254 struct virtio_net_ctrl_mac *uc, *mc;
1257 if (index >= VIRTIO_MAX_MAC_ADDRS) {
1258 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1262 uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1263 sizeof(uc->entries));
1265 mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1266 sizeof(mc->entries));
1269 for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1270 struct virtio_net_ctrl_mac *tbl;
1272 if (i == index || rte_is_zero_ether_addr(addrs + i))
1275 tbl = rte_is_multicast_ether_addr(addrs + i) ? mc : uc;
1276 memcpy(&tbl->macs[tbl->entries++], addrs + i,
1277 RTE_ETHER_ADDR_LEN);
1280 virtio_mac_table_set(hw, uc, mc);
1284 virtio_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1286 struct virtio_hw *hw = dev->data->dev_private;
1288 memcpy(hw->mac_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1290 /* Use atomic update if available */
1291 if (virtio_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1292 struct virtio_pmd_ctrl ctrl;
1293 int len = RTE_ETHER_ADDR_LEN;
1295 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1296 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1298 memcpy(ctrl.data, mac_addr, RTE_ETHER_ADDR_LEN);
1299 return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1302 if (!virtio_with_feature(hw, VIRTIO_NET_F_MAC))
1305 virtio_set_hwaddr(hw);
1310 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1312 struct virtio_hw *hw = dev->data->dev_private;
1313 struct virtio_pmd_ctrl ctrl;
1316 if (!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1319 ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1320 ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1321 memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1322 len = sizeof(vlan_id);
1324 return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1328 virtio_intr_unmask(struct rte_eth_dev *dev)
1330 struct virtio_hw *hw = dev->data->dev_private;
1332 if (rte_intr_ack(dev->intr_handle) < 0)
1335 if (VIRTIO_OPS(hw)->intr_detect)
1336 VIRTIO_OPS(hw)->intr_detect(hw);
1342 virtio_intr_enable(struct rte_eth_dev *dev)
1344 struct virtio_hw *hw = dev->data->dev_private;
1346 if (rte_intr_enable(dev->intr_handle) < 0)
1349 if (VIRTIO_OPS(hw)->intr_detect)
1350 VIRTIO_OPS(hw)->intr_detect(hw);
1356 virtio_intr_disable(struct rte_eth_dev *dev)
1358 struct virtio_hw *hw = dev->data->dev_private;
1360 if (rte_intr_disable(dev->intr_handle) < 0)
1363 if (VIRTIO_OPS(hw)->intr_detect)
1364 VIRTIO_OPS(hw)->intr_detect(hw);
1370 virtio_ethdev_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1372 uint64_t host_features;
1374 /* Prepare guest_features: feature that driver wants to support */
1375 PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1378 /* Read device(host) feature bits */
1379 host_features = VIRTIO_OPS(hw)->get_features(hw);
1380 PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1383 /* If supported, ensure MTU value is valid before acknowledging it. */
1384 if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1385 struct virtio_net_config config;
1387 virtio_read_dev_config(hw,
1388 offsetof(struct virtio_net_config, mtu),
1389 &config.mtu, sizeof(config.mtu));
1391 if (config.mtu < RTE_ETHER_MIN_MTU)
1392 req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1396 * Negotiate features: Subset of device feature bits are written back
1397 * guest feature bits.
1399 hw->guest_features = req_features;
1400 hw->guest_features = virtio_negotiate_features(hw, host_features);
1401 PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1402 hw->guest_features);
1404 if (VIRTIO_OPS(hw)->features_ok(hw) < 0)
1407 if (virtio_with_feature(hw, VIRTIO_F_VERSION_1)) {
1408 virtio_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1410 if (!(virtio_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1411 PMD_INIT_LOG(ERR, "Failed to set FEATURES_OK status!");
1416 hw->req_guest_features = req_features;
1422 virtio_dev_pause(struct rte_eth_dev *dev)
1424 struct virtio_hw *hw = dev->data->dev_private;
1426 rte_spinlock_lock(&hw->state_lock);
1428 if (hw->started == 0) {
1429 /* Device is just stopped. */
1430 rte_spinlock_unlock(&hw->state_lock);
1435 * Prevent the worker threads from touching queues to avoid contention,
1436 * 1 ms should be enough for the ongoing Tx function to finish.
1443 * Recover hw state to let the worker threads continue.
1446 virtio_dev_resume(struct rte_eth_dev *dev)
1448 struct virtio_hw *hw = dev->data->dev_private;
1451 rte_spinlock_unlock(&hw->state_lock);
1455 * Should be called only after device is paused.
1458 virtio_inject_pkts(struct rte_eth_dev *dev, struct rte_mbuf **tx_pkts,
1461 struct virtio_hw *hw = dev->data->dev_private;
1462 struct virtnet_tx *txvq = dev->data->tx_queues[0];
1465 hw->inject_pkts = tx_pkts;
1466 ret = dev->tx_pkt_burst(txvq, tx_pkts, nb_pkts);
1467 hw->inject_pkts = NULL;
1473 virtio_notify_peers(struct rte_eth_dev *dev)
1475 struct virtio_hw *hw = dev->data->dev_private;
1476 struct virtnet_rx *rxvq;
1477 struct rte_mbuf *rarp_mbuf;
1479 if (!dev->data->rx_queues)
1482 rxvq = dev->data->rx_queues[0];
1486 rarp_mbuf = rte_net_make_rarp_packet(rxvq->mpool,
1487 (struct rte_ether_addr *)hw->mac_addr);
1488 if (rarp_mbuf == NULL) {
1489 PMD_DRV_LOG(ERR, "failed to make RARP packet.");
1493 /* If virtio port just stopped, no need to send RARP */
1494 if (virtio_dev_pause(dev) < 0) {
1495 rte_pktmbuf_free(rarp_mbuf);
1499 virtio_inject_pkts(dev, &rarp_mbuf, 1);
1500 virtio_dev_resume(dev);
1504 virtio_ack_link_announce(struct rte_eth_dev *dev)
1506 struct virtio_hw *hw = dev->data->dev_private;
1507 struct virtio_pmd_ctrl ctrl;
1509 ctrl.hdr.class = VIRTIO_NET_CTRL_ANNOUNCE;
1510 ctrl.hdr.cmd = VIRTIO_NET_CTRL_ANNOUNCE_ACK;
1512 virtio_send_command(hw->cvq, &ctrl, NULL, 0);
1516 * Process virtio config changed interrupt. Call the callback
1517 * if link state changed, generate gratuitous RARP packet if
1518 * the status indicates an ANNOUNCE.
1521 virtio_interrupt_handler(void *param)
1523 struct rte_eth_dev *dev = param;
1524 struct virtio_hw *hw = dev->data->dev_private;
1528 /* Read interrupt status which clears interrupt */
1529 isr = virtio_get_isr(hw);
1530 PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1532 if (virtio_intr_unmask(dev) < 0)
1533 PMD_DRV_LOG(ERR, "interrupt enable failed");
1535 if (isr & VIRTIO_ISR_CONFIG) {
1536 if (virtio_dev_link_update(dev, 0) == 0)
1537 rte_eth_dev_callback_process(dev,
1538 RTE_ETH_EVENT_INTR_LSC,
1541 if (virtio_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1542 virtio_read_dev_config(hw,
1543 offsetof(struct virtio_net_config, status),
1544 &status, sizeof(status));
1545 if (status & VIRTIO_NET_S_ANNOUNCE) {
1546 virtio_notify_peers(dev);
1548 virtio_ack_link_announce(dev);
1554 /* set rx and tx handlers according to what is supported */
1556 set_rxtx_funcs(struct rte_eth_dev *eth_dev)
1558 struct virtio_hw *hw = eth_dev->data->dev_private;
1560 eth_dev->tx_pkt_prepare = virtio_xmit_pkts_prepare;
1561 if (virtio_with_packed_queue(hw)) {
1563 "virtio: using packed ring %s Tx path on port %u",
1564 hw->use_vec_tx ? "vectorized" : "standard",
1565 eth_dev->data->port_id);
1567 eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed_vec;
1569 eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed;
1571 if (hw->use_inorder_tx) {
1572 PMD_INIT_LOG(INFO, "virtio: using inorder Tx path on port %u",
1573 eth_dev->data->port_id);
1574 eth_dev->tx_pkt_burst = virtio_xmit_pkts_inorder;
1576 PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
1577 eth_dev->data->port_id);
1578 eth_dev->tx_pkt_burst = virtio_xmit_pkts;
1582 if (virtio_with_packed_queue(hw)) {
1583 if (hw->use_vec_rx) {
1585 "virtio: using packed ring vectorized Rx path on port %u",
1586 eth_dev->data->port_id);
1587 eth_dev->rx_pkt_burst =
1588 &virtio_recv_pkts_packed_vec;
1589 } else if (virtio_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1591 "virtio: using packed ring mergeable buffer Rx path on port %u",
1592 eth_dev->data->port_id);
1593 eth_dev->rx_pkt_burst =
1594 &virtio_recv_mergeable_pkts_packed;
1597 "virtio: using packed ring standard Rx path on port %u",
1598 eth_dev->data->port_id);
1599 eth_dev->rx_pkt_burst = &virtio_recv_pkts_packed;
1602 if (hw->use_vec_rx) {
1603 PMD_INIT_LOG(INFO, "virtio: using vectorized Rx path on port %u",
1604 eth_dev->data->port_id);
1605 eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1606 } else if (hw->use_inorder_rx) {
1608 "virtio: using inorder Rx path on port %u",
1609 eth_dev->data->port_id);
1610 eth_dev->rx_pkt_burst = &virtio_recv_pkts_inorder;
1611 } else if (virtio_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1613 "virtio: using mergeable buffer Rx path on port %u",
1614 eth_dev->data->port_id);
1615 eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1617 PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
1618 eth_dev->data->port_id);
1619 eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1625 /* Only support 1:1 queue/interrupt mapping so far.
1626 * TODO: support n:1 queue/interrupt mapping when there are limited number of
1627 * interrupt vectors (<N+1).
1630 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1633 struct virtio_hw *hw = dev->data->dev_private;
1635 PMD_INIT_LOG(INFO, "queue/interrupt binding");
1636 for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1637 dev->intr_handle->intr_vec[i] = i + 1;
1638 if (VIRTIO_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1639 VIRTIO_MSI_NO_VECTOR) {
1640 PMD_DRV_LOG(ERR, "failed to set queue vector");
1649 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1652 struct virtio_hw *hw = dev->data->dev_private;
1654 PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1655 for (i = 0; i < dev->data->nb_rx_queues; ++i)
1656 VIRTIO_OPS(hw)->set_queue_irq(hw,
1657 hw->vqs[i * VTNET_CQ],
1658 VIRTIO_MSI_NO_VECTOR);
1662 virtio_configure_intr(struct rte_eth_dev *dev)
1664 struct virtio_hw *hw = dev->data->dev_private;
1666 if (!rte_intr_cap_multiple(dev->intr_handle)) {
1667 PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1671 if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1672 PMD_INIT_LOG(ERR, "Fail to create eventfd");
1676 if (!dev->intr_handle->intr_vec) {
1677 dev->intr_handle->intr_vec =
1678 rte_zmalloc("intr_vec",
1679 hw->max_queue_pairs * sizeof(int), 0);
1680 if (!dev->intr_handle->intr_vec) {
1681 PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1682 hw->max_queue_pairs);
1687 /* Re-register callback to update max_intr */
1688 rte_intr_callback_unregister(dev->intr_handle,
1689 virtio_interrupt_handler,
1691 rte_intr_callback_register(dev->intr_handle,
1692 virtio_interrupt_handler,
1695 /* DO NOT try to remove this! This function will enable msix, or QEMU
1696 * will encounter SIGSEGV when DRIVER_OK is sent.
1697 * And for legacy devices, this should be done before queue/vec binding
1698 * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1699 * (22) will be ignored.
1701 if (virtio_intr_enable(dev) < 0) {
1702 PMD_DRV_LOG(ERR, "interrupt enable failed");
1706 if (virtio_queues_bind_intr(dev) < 0) {
1707 PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1713 #define DUPLEX_UNKNOWN 0xff
1714 /* reset device and renegotiate features if needed */
1716 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1718 struct virtio_hw *hw = eth_dev->data->dev_private;
1719 struct virtio_net_config *config;
1720 struct virtio_net_config local_config;
1723 /* Reset the device although not necessary at startup */
1727 virtio_dev_free_mbufs(eth_dev);
1728 virtio_free_queues(hw);
1731 /* Tell the host we've noticed this device. */
1732 virtio_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1734 /* Tell the host we've known how to drive the device. */
1735 virtio_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1736 if (virtio_ethdev_negotiate_features(hw, req_features) < 0)
1739 hw->weak_barriers = !virtio_with_feature(hw, VIRTIO_F_ORDER_PLATFORM);
1741 /* If host does not support both status and MSI-X then disable LSC */
1742 if (virtio_with_feature(hw, VIRTIO_NET_F_STATUS) && hw->intr_lsc)
1743 eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1745 eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1747 eth_dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
1749 /* Setting up rx_header size for the device */
1750 if (virtio_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1751 virtio_with_feature(hw, VIRTIO_F_VERSION_1) ||
1752 virtio_with_packed_queue(hw))
1753 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1755 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1757 /* Copy the permanent MAC address to: virtio_hw */
1758 virtio_get_hwaddr(hw);
1759 rte_ether_addr_copy((struct rte_ether_addr *)hw->mac_addr,
1760 ð_dev->data->mac_addrs[0]);
1762 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1763 hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1764 hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1766 if (hw->speed == ETH_SPEED_NUM_UNKNOWN) {
1767 if (virtio_with_feature(hw, VIRTIO_NET_F_SPEED_DUPLEX)) {
1768 config = &local_config;
1769 virtio_read_dev_config(hw,
1770 offsetof(struct virtio_net_config, speed),
1771 &config->speed, sizeof(config->speed));
1772 virtio_read_dev_config(hw,
1773 offsetof(struct virtio_net_config, duplex),
1774 &config->duplex, sizeof(config->duplex));
1775 hw->speed = config->speed;
1776 hw->duplex = config->duplex;
1779 if (hw->duplex == DUPLEX_UNKNOWN)
1780 hw->duplex = ETH_LINK_FULL_DUPLEX;
1781 PMD_INIT_LOG(DEBUG, "link speed = %d, duplex = %d",
1782 hw->speed, hw->duplex);
1783 if (virtio_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1784 config = &local_config;
1786 virtio_read_dev_config(hw,
1787 offsetof(struct virtio_net_config, mac),
1788 &config->mac, sizeof(config->mac));
1790 if (virtio_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1791 virtio_read_dev_config(hw,
1792 offsetof(struct virtio_net_config, status),
1793 &config->status, sizeof(config->status));
1796 "VIRTIO_NET_F_STATUS is not supported");
1800 if (virtio_with_feature(hw, VIRTIO_NET_F_MQ)) {
1801 virtio_read_dev_config(hw,
1802 offsetof(struct virtio_net_config, max_virtqueue_pairs),
1803 &config->max_virtqueue_pairs,
1804 sizeof(config->max_virtqueue_pairs));
1807 "VIRTIO_NET_F_MQ is not supported");
1808 config->max_virtqueue_pairs = 1;
1811 hw->max_queue_pairs = config->max_virtqueue_pairs;
1813 if (virtio_with_feature(hw, VIRTIO_NET_F_MTU)) {
1814 virtio_read_dev_config(hw,
1815 offsetof(struct virtio_net_config, mtu),
1817 sizeof(config->mtu));
1820 * MTU value has already been checked at negotiation
1821 * time, but check again in case it has changed since
1822 * then, which should not happen.
1824 if (config->mtu < RTE_ETHER_MIN_MTU) {
1825 PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1830 hw->max_mtu = config->mtu;
1831 /* Set initial MTU to maximum one supported by vhost */
1832 eth_dev->data->mtu = config->mtu;
1835 hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1836 VLAN_TAG_LEN - hw->vtnet_hdr_size;
1839 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1840 config->max_virtqueue_pairs);
1841 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1843 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1844 config->mac[0], config->mac[1],
1845 config->mac[2], config->mac[3],
1846 config->mac[4], config->mac[5]);
1848 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1849 hw->max_queue_pairs = 1;
1850 hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1851 VLAN_TAG_LEN - hw->vtnet_hdr_size;
1854 ret = virtio_alloc_queues(eth_dev);
1858 if (eth_dev->data->dev_conf.intr_conf.rxq) {
1859 if (virtio_configure_intr(eth_dev) < 0) {
1860 PMD_INIT_LOG(ERR, "failed to configure interrupt");
1861 virtio_free_queues(hw);
1866 virtio_reinit_complete(hw);
1872 * This function is based on probe() function in virtio_pci.c
1873 * It returns 0 on success.
1876 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1878 struct virtio_hw *hw = eth_dev->data->dev_private;
1879 uint32_t speed = ETH_SPEED_NUM_UNKNOWN;
1883 if (sizeof(struct virtio_net_hdr_mrg_rxbuf) > RTE_PKTMBUF_HEADROOM) {
1885 "Not sufficient headroom required = %d, avail = %d",
1886 (int)sizeof(struct virtio_net_hdr_mrg_rxbuf),
1887 RTE_PKTMBUF_HEADROOM);
1892 eth_dev->dev_ops = &virtio_eth_dev_ops;
1893 eth_dev->rx_descriptor_done = virtio_dev_rx_queue_done;
1895 if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1896 set_rxtx_funcs(eth_dev);
1900 ret = virtio_dev_devargs_parse(eth_dev->device->devargs, &speed, &vectorized);
1904 hw->duplex = DUPLEX_UNKNOWN;
1906 /* Allocate memory for storing MAC addresses */
1907 eth_dev->data->mac_addrs = rte_zmalloc("virtio",
1908 VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN, 0);
1909 if (eth_dev->data->mac_addrs == NULL) {
1911 "Failed to allocate %d bytes needed to store MAC addresses",
1912 VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN);
1916 rte_spinlock_init(&hw->state_lock);
1918 /* reset device and negotiate default features */
1919 ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1921 goto err_virtio_init;
1924 if (!virtio_with_packed_queue(hw)) {
1927 #if defined(CC_AVX512_SUPPORT) || defined(RTE_ARCH_ARM)
1932 "building environment do not support packed ring vectorized");
1942 rte_free(eth_dev->data->mac_addrs);
1943 eth_dev->data->mac_addrs = NULL;
1948 virtio_dev_speed_capa_get(uint32_t speed)
1951 case ETH_SPEED_NUM_10G:
1952 return ETH_LINK_SPEED_10G;
1953 case ETH_SPEED_NUM_20G:
1954 return ETH_LINK_SPEED_20G;
1955 case ETH_SPEED_NUM_25G:
1956 return ETH_LINK_SPEED_25G;
1957 case ETH_SPEED_NUM_40G:
1958 return ETH_LINK_SPEED_40G;
1959 case ETH_SPEED_NUM_50G:
1960 return ETH_LINK_SPEED_50G;
1961 case ETH_SPEED_NUM_56G:
1962 return ETH_LINK_SPEED_56G;
1963 case ETH_SPEED_NUM_100G:
1964 return ETH_LINK_SPEED_100G;
1965 case ETH_SPEED_NUM_200G:
1966 return ETH_LINK_SPEED_200G;
1972 static int vectorized_check_handler(__rte_unused const char *key,
1973 const char *value, void *ret_val)
1975 if (strcmp(value, "1") == 0)
1976 *(int *)ret_val = 1;
1978 *(int *)ret_val = 0;
1983 #define VIRTIO_ARG_SPEED "speed"
1984 #define VIRTIO_ARG_VECTORIZED "vectorized"
1987 link_speed_handler(const char *key __rte_unused,
1988 const char *value, void *ret_val)
1991 if (!value || !ret_val)
1993 val = strtoul(value, NULL, 0);
1994 /* validate input */
1995 if (virtio_dev_speed_capa_get(val) == 0)
1997 *(uint32_t *)ret_val = val;
2004 virtio_dev_devargs_parse(struct rte_devargs *devargs, uint32_t *speed, int *vectorized)
2006 struct rte_kvargs *kvlist;
2009 if (devargs == NULL)
2012 kvlist = rte_kvargs_parse(devargs->args, NULL);
2013 if (kvlist == NULL) {
2014 PMD_INIT_LOG(ERR, "error when parsing param");
2018 if (speed && rte_kvargs_count(kvlist, VIRTIO_ARG_SPEED) == 1) {
2019 ret = rte_kvargs_process(kvlist,
2021 link_speed_handler, speed);
2023 PMD_INIT_LOG(ERR, "Failed to parse %s",
2030 rte_kvargs_count(kvlist, VIRTIO_ARG_VECTORIZED) == 1) {
2031 ret = rte_kvargs_process(kvlist,
2032 VIRTIO_ARG_VECTORIZED,
2033 vectorized_check_handler, vectorized);
2035 PMD_INIT_LOG(ERR, "Failed to parse %s",
2036 VIRTIO_ARG_VECTORIZED);
2042 rte_kvargs_free(kvlist);
2047 rx_offload_enabled(struct virtio_hw *hw)
2049 return virtio_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM) ||
2050 virtio_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2051 virtio_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6);
2055 tx_offload_enabled(struct virtio_hw *hw)
2057 return virtio_with_feature(hw, VIRTIO_NET_F_CSUM) ||
2058 virtio_with_feature(hw, VIRTIO_NET_F_HOST_TSO4) ||
2059 virtio_with_feature(hw, VIRTIO_NET_F_HOST_TSO6);
2063 * Configure virtio device
2064 * It returns 0 on success.
2067 virtio_dev_configure(struct rte_eth_dev *dev)
2069 const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2070 const struct rte_eth_txmode *txmode = &dev->data->dev_conf.txmode;
2071 struct virtio_hw *hw = dev->data->dev_private;
2072 uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
2074 uint64_t rx_offloads = rxmode->offloads;
2075 uint64_t tx_offloads = txmode->offloads;
2076 uint64_t req_features;
2079 PMD_INIT_LOG(DEBUG, "configure");
2080 req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
2082 if (rxmode->mq_mode != ETH_MQ_RX_NONE) {
2084 "Unsupported Rx multi queue mode %d",
2089 if (txmode->mq_mode != ETH_MQ_TX_NONE) {
2091 "Unsupported Tx multi queue mode %d",
2096 if (dev->data->dev_conf.intr_conf.rxq) {
2097 ret = virtio_init_device(dev, hw->req_guest_features);
2102 if (rxmode->max_rx_pkt_len > hw->max_mtu + ether_hdr_len)
2103 req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
2105 hw->max_rx_pkt_len = rxmode->max_rx_pkt_len;
2107 if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2108 DEV_RX_OFFLOAD_TCP_CKSUM))
2109 req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
2111 if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO)
2113 (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2114 (1ULL << VIRTIO_NET_F_GUEST_TSO6);
2116 if (tx_offloads & (DEV_TX_OFFLOAD_UDP_CKSUM |
2117 DEV_TX_OFFLOAD_TCP_CKSUM))
2118 req_features |= (1ULL << VIRTIO_NET_F_CSUM);
2120 if (tx_offloads & DEV_TX_OFFLOAD_TCP_TSO)
2122 (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2123 (1ULL << VIRTIO_NET_F_HOST_TSO6);
2125 /* if request features changed, reinit the device */
2126 if (req_features != hw->req_guest_features) {
2127 ret = virtio_init_device(dev, req_features);
2132 if ((rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2133 DEV_RX_OFFLOAD_TCP_CKSUM)) &&
2134 !virtio_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
2136 "rx checksum not available on this host");
2140 if ((rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) &&
2141 (!virtio_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2142 !virtio_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6))) {
2144 "Large Receive Offload not available on this host");
2148 /* start control queue */
2149 if (virtio_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
2150 virtio_dev_cq_start(dev);
2152 if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
2155 hw->rx_ol_scatter = (rx_offloads & DEV_RX_OFFLOAD_SCATTER);
2157 if ((rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER) &&
2158 !virtio_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2160 "vlan filtering not available on this host");
2164 hw->has_tx_offload = tx_offload_enabled(hw);
2165 hw->has_rx_offload = rx_offload_enabled(hw);
2167 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2168 /* Enable vector (0) for Link State Intrerrupt */
2169 if (VIRTIO_OPS(hw)->set_config_irq(hw, 0) ==
2170 VIRTIO_MSI_NO_VECTOR) {
2171 PMD_DRV_LOG(ERR, "failed to set config vector");
2175 if (virtio_with_packed_queue(hw)) {
2176 #if defined(RTE_ARCH_X86_64) && defined(CC_AVX512_SUPPORT)
2177 if ((hw->use_vec_rx || hw->use_vec_tx) &&
2178 (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX512F) ||
2179 !virtio_with_feature(hw, VIRTIO_F_IN_ORDER) ||
2180 !virtio_with_feature(hw, VIRTIO_F_VERSION_1) ||
2181 rte_vect_get_max_simd_bitwidth() < RTE_VECT_SIMD_512)) {
2183 "disabled packed ring vectorized path for requirements not met");
2187 #elif defined(RTE_ARCH_ARM)
2188 if ((hw->use_vec_rx || hw->use_vec_tx) &&
2189 (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON) ||
2190 !virtio_with_feature(hw, VIRTIO_F_IN_ORDER) ||
2191 !virtio_with_feature(hw, VIRTIO_F_VERSION_1) ||
2192 rte_vect_get_max_simd_bitwidth() < RTE_VECT_SIMD_128)) {
2194 "disabled packed ring vectorized path for requirements not met");
2203 if (hw->use_vec_rx) {
2204 if (virtio_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2206 "disabled packed ring vectorized rx for mrg_rxbuf enabled");
2210 if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) {
2212 "disabled packed ring vectorized rx for TCP_LRO enabled");
2217 if (virtio_with_feature(hw, VIRTIO_F_IN_ORDER)) {
2218 hw->use_inorder_tx = 1;
2219 hw->use_inorder_rx = 1;
2223 if (hw->use_vec_rx) {
2224 #if defined RTE_ARCH_ARM
2225 if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
2227 "disabled split ring vectorized path for requirement not met");
2231 if (virtio_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2233 "disabled split ring vectorized rx for mrg_rxbuf enabled");
2237 if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2238 DEV_RX_OFFLOAD_TCP_CKSUM |
2239 DEV_RX_OFFLOAD_TCP_LRO |
2240 DEV_RX_OFFLOAD_VLAN_STRIP)) {
2242 "disabled split ring vectorized rx for offloading enabled");
2246 if (rte_vect_get_max_simd_bitwidth() < RTE_VECT_SIMD_128) {
2248 "disabled split ring vectorized rx, max SIMD bitwidth too low");
2259 virtio_dev_start(struct rte_eth_dev *dev)
2261 uint16_t nb_queues, i;
2262 struct virtqueue *vq;
2263 struct virtio_hw *hw = dev->data->dev_private;
2266 /* Finish the initialization of the queues */
2267 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2268 ret = virtio_dev_rx_queue_setup_finish(dev, i);
2272 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2273 ret = virtio_dev_tx_queue_setup_finish(dev, i);
2278 /* check if lsc interrupt feature is enabled */
2279 if (dev->data->dev_conf.intr_conf.lsc) {
2280 if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
2281 PMD_DRV_LOG(ERR, "link status not supported by host");
2286 /* Enable uio/vfio intr/eventfd mapping: althrough we already did that
2287 * in device configure, but it could be unmapped when device is
2290 if (dev->data->dev_conf.intr_conf.lsc ||
2291 dev->data->dev_conf.intr_conf.rxq) {
2292 virtio_intr_disable(dev);
2294 /* Setup interrupt callback */
2295 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2296 rte_intr_callback_register(dev->intr_handle,
2297 virtio_interrupt_handler,
2300 if (virtio_intr_enable(dev) < 0) {
2301 PMD_DRV_LOG(ERR, "interrupt enable failed");
2306 /*Notify the backend
2307 *Otherwise the tap backend might already stop its queue due to fullness.
2308 *vhost backend will have no chance to be waked up
2310 nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
2311 if (hw->max_queue_pairs > 1) {
2312 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
2316 PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
2318 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2319 vq = virtnet_rxq_to_vq(dev->data->rx_queues[i]);
2320 /* Flush the old packets */
2321 virtqueue_rxvq_flush(vq);
2322 virtqueue_notify(vq);
2325 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2326 vq = virtnet_txq_to_vq(dev->data->tx_queues[i]);
2327 virtqueue_notify(vq);
2330 PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
2332 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2333 vq = virtnet_rxq_to_vq(dev->data->rx_queues[i]);
2337 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2338 vq = virtnet_txq_to_vq(dev->data->tx_queues[i]);
2342 set_rxtx_funcs(dev);
2345 /* Initialize Link state */
2346 virtio_dev_link_update(dev, 0);
2351 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
2353 struct virtio_hw *hw = dev->data->dev_private;
2354 uint16_t nr_vq = virtio_get_nr_vq(hw);
2355 const char *type __rte_unused;
2356 unsigned int i, mbuf_num = 0;
2357 struct virtqueue *vq;
2358 struct rte_mbuf *buf;
2361 if (hw->vqs == NULL)
2364 for (i = 0; i < nr_vq; i++) {
2369 queue_type = virtio_get_queue_type(hw, i);
2370 if (queue_type == VTNET_RQ)
2372 else if (queue_type == VTNET_TQ)
2378 "Before freeing %s[%d] used and unused buf",
2382 while ((buf = virtqueue_detach_unused(vq)) != NULL) {
2383 rte_pktmbuf_free(buf);
2388 "After freeing %s[%d] used and unused buf",
2393 PMD_INIT_LOG(DEBUG, "%d mbufs freed", mbuf_num);
2397 * Stop device: disable interrupt and mark link down
2400 virtio_dev_stop(struct rte_eth_dev *dev)
2402 struct virtio_hw *hw = dev->data->dev_private;
2403 struct rte_eth_link link;
2404 struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
2406 PMD_INIT_LOG(DEBUG, "stop");
2407 dev->data->dev_started = 0;
2409 rte_spinlock_lock(&hw->state_lock);
2414 if (intr_conf->lsc || intr_conf->rxq) {
2415 virtio_intr_disable(dev);
2417 /* Reset interrupt callback */
2418 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
2419 rte_intr_callback_unregister(dev->intr_handle,
2420 virtio_interrupt_handler,
2425 memset(&link, 0, sizeof(link));
2426 rte_eth_linkstatus_set(dev, &link);
2428 rte_spinlock_unlock(&hw->state_lock);
2434 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
2436 struct rte_eth_link link;
2438 struct virtio_hw *hw = dev->data->dev_private;
2440 memset(&link, 0, sizeof(link));
2441 link.link_duplex = hw->duplex;
2442 link.link_speed = hw->speed;
2443 link.link_autoneg = ETH_LINK_AUTONEG;
2446 link.link_status = ETH_LINK_DOWN;
2447 link.link_speed = ETH_SPEED_NUM_NONE;
2448 } else if (virtio_with_feature(hw, VIRTIO_NET_F_STATUS)) {
2449 PMD_INIT_LOG(DEBUG, "Get link status from hw");
2450 virtio_read_dev_config(hw,
2451 offsetof(struct virtio_net_config, status),
2452 &status, sizeof(status));
2453 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
2454 link.link_status = ETH_LINK_DOWN;
2455 link.link_speed = ETH_SPEED_NUM_NONE;
2456 PMD_INIT_LOG(DEBUG, "Port %d is down",
2457 dev->data->port_id);
2459 link.link_status = ETH_LINK_UP;
2460 PMD_INIT_LOG(DEBUG, "Port %d is up",
2461 dev->data->port_id);
2464 link.link_status = ETH_LINK_UP;
2467 return rte_eth_linkstatus_set(dev, &link);
2471 virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
2473 const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2474 struct virtio_hw *hw = dev->data->dev_private;
2475 uint64_t offloads = rxmode->offloads;
2477 if (mask & ETH_VLAN_FILTER_MASK) {
2478 if ((offloads & DEV_RX_OFFLOAD_VLAN_FILTER) &&
2479 !virtio_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2482 "vlan filtering not available on this host");
2488 if (mask & ETH_VLAN_STRIP_MASK)
2489 hw->vlan_strip = !!(offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
2495 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2497 uint64_t tso_mask, host_features;
2498 struct virtio_hw *hw = dev->data->dev_private;
2499 dev_info->speed_capa = virtio_dev_speed_capa_get(hw->speed);
2501 dev_info->max_rx_queues =
2502 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
2503 dev_info->max_tx_queues =
2504 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
2505 dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
2506 dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
2507 dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
2508 dev_info->max_mtu = hw->max_mtu;
2510 host_features = VIRTIO_OPS(hw)->get_features(hw);
2511 dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
2512 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_JUMBO_FRAME;
2513 if (host_features & (1ULL << VIRTIO_NET_F_MRG_RXBUF))
2514 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_SCATTER;
2515 if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
2516 dev_info->rx_offload_capa |=
2517 DEV_RX_OFFLOAD_TCP_CKSUM |
2518 DEV_RX_OFFLOAD_UDP_CKSUM;
2520 if (host_features & (1ULL << VIRTIO_NET_F_CTRL_VLAN))
2521 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_VLAN_FILTER;
2522 tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2523 (1ULL << VIRTIO_NET_F_GUEST_TSO6);
2524 if ((host_features & tso_mask) == tso_mask)
2525 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
2527 dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
2528 DEV_TX_OFFLOAD_VLAN_INSERT;
2529 if (host_features & (1ULL << VIRTIO_NET_F_CSUM)) {
2530 dev_info->tx_offload_capa |=
2531 DEV_TX_OFFLOAD_UDP_CKSUM |
2532 DEV_TX_OFFLOAD_TCP_CKSUM;
2534 tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2535 (1ULL << VIRTIO_NET_F_HOST_TSO6);
2536 if ((host_features & tso_mask) == tso_mask)
2537 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
2543 * It enables testpmd to collect per queue stats.
2546 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
2547 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
2548 __rte_unused uint8_t is_rx)
2553 RTE_LOG_REGISTER_SUFFIX(virtio_logtype_init, init, NOTICE);
2554 RTE_LOG_REGISTER_SUFFIX(virtio_logtype_driver, driver, NOTICE);