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_rxtx_simple.h"
37 #include "virtio_user/virtio_user_dev.h"
39 static int virtio_dev_configure(struct rte_eth_dev *dev);
40 static int virtio_dev_start(struct rte_eth_dev *dev);
41 static int virtio_dev_promiscuous_enable(struct rte_eth_dev *dev);
42 static int virtio_dev_promiscuous_disable(struct rte_eth_dev *dev);
43 static int virtio_dev_allmulticast_enable(struct rte_eth_dev *dev);
44 static int virtio_dev_allmulticast_disable(struct rte_eth_dev *dev);
45 static uint32_t virtio_dev_speed_capa_get(uint32_t speed);
46 static int virtio_dev_devargs_parse(struct rte_devargs *devargs,
49 static int virtio_dev_info_get(struct rte_eth_dev *dev,
50 struct rte_eth_dev_info *dev_info);
51 static int virtio_dev_link_update(struct rte_eth_dev *dev,
52 int wait_to_complete);
53 static int virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask);
55 static void virtio_set_hwaddr(struct virtio_hw *hw);
56 static void virtio_get_hwaddr(struct virtio_hw *hw);
58 static int virtio_dev_stats_get(struct rte_eth_dev *dev,
59 struct rte_eth_stats *stats);
60 static int virtio_dev_xstats_get(struct rte_eth_dev *dev,
61 struct rte_eth_xstat *xstats, unsigned n);
62 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
63 struct rte_eth_xstat_name *xstats_names,
65 static int virtio_dev_stats_reset(struct rte_eth_dev *dev);
66 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev);
67 static int virtio_vlan_filter_set(struct rte_eth_dev *dev,
68 uint16_t vlan_id, int on);
69 static int virtio_mac_addr_add(struct rte_eth_dev *dev,
70 struct rte_ether_addr *mac_addr,
71 uint32_t index, uint32_t vmdq);
72 static void virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index);
73 static int virtio_mac_addr_set(struct rte_eth_dev *dev,
74 struct rte_ether_addr *mac_addr);
76 static int virtio_intr_disable(struct rte_eth_dev *dev);
78 static int virtio_dev_queue_stats_mapping_set(
79 struct rte_eth_dev *eth_dev,
84 static void virtio_notify_peers(struct rte_eth_dev *dev);
85 static void virtio_ack_link_announce(struct rte_eth_dev *dev);
87 struct rte_virtio_xstats_name_off {
88 char name[RTE_ETH_XSTATS_NAME_SIZE];
92 /* [rt]x_qX_ is prepended to the name string here */
93 static const struct rte_virtio_xstats_name_off rte_virtio_rxq_stat_strings[] = {
94 {"good_packets", offsetof(struct virtnet_rx, stats.packets)},
95 {"good_bytes", offsetof(struct virtnet_rx, stats.bytes)},
96 {"errors", offsetof(struct virtnet_rx, stats.errors)},
97 {"multicast_packets", offsetof(struct virtnet_rx, stats.multicast)},
98 {"broadcast_packets", offsetof(struct virtnet_rx, stats.broadcast)},
99 {"undersize_packets", offsetof(struct virtnet_rx, stats.size_bins[0])},
100 {"size_64_packets", offsetof(struct virtnet_rx, stats.size_bins[1])},
101 {"size_65_127_packets", offsetof(struct virtnet_rx, stats.size_bins[2])},
102 {"size_128_255_packets", offsetof(struct virtnet_rx, stats.size_bins[3])},
103 {"size_256_511_packets", offsetof(struct virtnet_rx, stats.size_bins[4])},
104 {"size_512_1023_packets", offsetof(struct virtnet_rx, stats.size_bins[5])},
105 {"size_1024_1518_packets", offsetof(struct virtnet_rx, stats.size_bins[6])},
106 {"size_1519_max_packets", offsetof(struct virtnet_rx, stats.size_bins[7])},
109 /* [rt]x_qX_ is prepended to the name string here */
110 static const struct rte_virtio_xstats_name_off rte_virtio_txq_stat_strings[] = {
111 {"good_packets", offsetof(struct virtnet_tx, stats.packets)},
112 {"good_bytes", offsetof(struct virtnet_tx, stats.bytes)},
113 {"multicast_packets", offsetof(struct virtnet_tx, stats.multicast)},
114 {"broadcast_packets", offsetof(struct virtnet_tx, stats.broadcast)},
115 {"undersize_packets", offsetof(struct virtnet_tx, stats.size_bins[0])},
116 {"size_64_packets", offsetof(struct virtnet_tx, stats.size_bins[1])},
117 {"size_65_127_packets", offsetof(struct virtnet_tx, stats.size_bins[2])},
118 {"size_128_255_packets", offsetof(struct virtnet_tx, stats.size_bins[3])},
119 {"size_256_511_packets", offsetof(struct virtnet_tx, stats.size_bins[4])},
120 {"size_512_1023_packets", offsetof(struct virtnet_tx, stats.size_bins[5])},
121 {"size_1024_1518_packets", offsetof(struct virtnet_tx, stats.size_bins[6])},
122 {"size_1519_max_packets", offsetof(struct virtnet_tx, stats.size_bins[7])},
125 #define VIRTIO_NB_RXQ_XSTATS (sizeof(rte_virtio_rxq_stat_strings) / \
126 sizeof(rte_virtio_rxq_stat_strings[0]))
127 #define VIRTIO_NB_TXQ_XSTATS (sizeof(rte_virtio_txq_stat_strings) / \
128 sizeof(rte_virtio_txq_stat_strings[0]))
130 struct virtio_hw_internal virtio_hw_internal[RTE_MAX_ETHPORTS];
132 static struct virtio_pmd_ctrl *
133 virtio_send_command_packed(struct virtnet_ctl *cvq,
134 struct virtio_pmd_ctrl *ctrl,
135 int *dlen, int pkt_num)
137 struct virtqueue *vq = virtnet_cq_to_vq(cvq);
139 struct vring_packed_desc *desc = vq->vq_packed.ring.desc;
140 struct virtio_pmd_ctrl *result;
147 * Format is enforced in qemu code:
148 * One TX packet for header;
149 * At least one TX packet per argument;
150 * One RX packet for ACK.
152 head = vq->vq_avail_idx;
153 flags = vq->vq_packed.cached_flags;
154 desc[head].addr = cvq->virtio_net_hdr_mem;
155 desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
158 if (++vq->vq_avail_idx >= vq->vq_nentries) {
159 vq->vq_avail_idx -= vq->vq_nentries;
160 vq->vq_packed.cached_flags ^= VRING_PACKED_DESC_F_AVAIL_USED;
163 for (k = 0; k < pkt_num; k++) {
164 desc[vq->vq_avail_idx].addr = cvq->virtio_net_hdr_mem
165 + sizeof(struct virtio_net_ctrl_hdr)
166 + sizeof(ctrl->status) + sizeof(uint8_t) * sum;
167 desc[vq->vq_avail_idx].len = dlen[k];
168 desc[vq->vq_avail_idx].flags = VRING_DESC_F_NEXT |
169 vq->vq_packed.cached_flags;
173 if (++vq->vq_avail_idx >= vq->vq_nentries) {
174 vq->vq_avail_idx -= vq->vq_nentries;
175 vq->vq_packed.cached_flags ^=
176 VRING_PACKED_DESC_F_AVAIL_USED;
180 desc[vq->vq_avail_idx].addr = cvq->virtio_net_hdr_mem
181 + sizeof(struct virtio_net_ctrl_hdr);
182 desc[vq->vq_avail_idx].len = sizeof(ctrl->status);
183 desc[vq->vq_avail_idx].flags = VRING_DESC_F_WRITE |
184 vq->vq_packed.cached_flags;
187 if (++vq->vq_avail_idx >= vq->vq_nentries) {
188 vq->vq_avail_idx -= vq->vq_nentries;
189 vq->vq_packed.cached_flags ^= VRING_PACKED_DESC_F_AVAIL_USED;
192 virtqueue_store_flags_packed(&desc[head], VRING_DESC_F_NEXT | flags,
193 vq->hw->weak_barriers);
195 virtio_wmb(vq->hw->weak_barriers);
196 virtqueue_notify(vq);
198 /* wait for used desc in virtqueue
199 * desc_is_used has a load-acquire or rte_io_rmb inside
201 while (!desc_is_used(&desc[head], vq))
204 /* now get used descriptors */
205 vq->vq_free_cnt += nb_descs;
206 vq->vq_used_cons_idx += nb_descs;
207 if (vq->vq_used_cons_idx >= vq->vq_nentries) {
208 vq->vq_used_cons_idx -= vq->vq_nentries;
209 vq->vq_packed.used_wrap_counter ^= 1;
212 PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\n"
213 "vq->vq_avail_idx=%d\n"
214 "vq->vq_used_cons_idx=%d\n"
215 "vq->vq_packed.cached_flags=0x%x\n"
216 "vq->vq_packed.used_wrap_counter=%d\n",
219 vq->vq_used_cons_idx,
220 vq->vq_packed.cached_flags,
221 vq->vq_packed.used_wrap_counter);
223 result = cvq->virtio_net_hdr_mz->addr;
227 static struct virtio_pmd_ctrl *
228 virtio_send_command_split(struct virtnet_ctl *cvq,
229 struct virtio_pmd_ctrl *ctrl,
230 int *dlen, int pkt_num)
232 struct virtio_pmd_ctrl *result;
233 struct virtqueue *vq = virtnet_cq_to_vq(cvq);
237 head = vq->vq_desc_head_idx;
240 * Format is enforced in qemu code:
241 * One TX packet for header;
242 * At least one TX packet per argument;
243 * One RX packet for ACK.
245 vq->vq_split.ring.desc[head].flags = VRING_DESC_F_NEXT;
246 vq->vq_split.ring.desc[head].addr = cvq->virtio_net_hdr_mem;
247 vq->vq_split.ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
249 i = vq->vq_split.ring.desc[head].next;
251 for (k = 0; k < pkt_num; k++) {
252 vq->vq_split.ring.desc[i].flags = VRING_DESC_F_NEXT;
253 vq->vq_split.ring.desc[i].addr = cvq->virtio_net_hdr_mem
254 + sizeof(struct virtio_net_ctrl_hdr)
255 + sizeof(ctrl->status) + sizeof(uint8_t)*sum;
256 vq->vq_split.ring.desc[i].len = dlen[k];
259 i = vq->vq_split.ring.desc[i].next;
262 vq->vq_split.ring.desc[i].flags = VRING_DESC_F_WRITE;
263 vq->vq_split.ring.desc[i].addr = cvq->virtio_net_hdr_mem
264 + sizeof(struct virtio_net_ctrl_hdr);
265 vq->vq_split.ring.desc[i].len = sizeof(ctrl->status);
268 vq->vq_desc_head_idx = vq->vq_split.ring.desc[i].next;
270 vq_update_avail_ring(vq, head);
271 vq_update_avail_idx(vq);
273 PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
275 virtqueue_notify(vq);
277 while (virtqueue_nused(vq) == 0)
280 while (virtqueue_nused(vq)) {
281 uint32_t idx, desc_idx, used_idx;
282 struct vring_used_elem *uep;
284 used_idx = (uint32_t)(vq->vq_used_cons_idx
285 & (vq->vq_nentries - 1));
286 uep = &vq->vq_split.ring.used->ring[used_idx];
287 idx = (uint32_t) uep->id;
290 while (vq->vq_split.ring.desc[desc_idx].flags &
292 desc_idx = vq->vq_split.ring.desc[desc_idx].next;
296 vq->vq_split.ring.desc[desc_idx].next = vq->vq_desc_head_idx;
297 vq->vq_desc_head_idx = idx;
299 vq->vq_used_cons_idx++;
303 PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
304 vq->vq_free_cnt, vq->vq_desc_head_idx);
306 result = cvq->virtio_net_hdr_mz->addr;
311 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
312 int *dlen, int pkt_num)
314 virtio_net_ctrl_ack status = ~0;
315 struct virtio_pmd_ctrl *result;
316 struct virtqueue *vq;
318 ctrl->status = status;
321 PMD_INIT_LOG(ERR, "Control queue is not supported.");
325 rte_spinlock_lock(&cvq->lock);
326 vq = virtnet_cq_to_vq(cvq);
328 PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
329 "vq->hw->cvq = %p vq = %p",
330 vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
332 if (vq->vq_free_cnt < pkt_num + 2 || pkt_num < 1) {
333 rte_spinlock_unlock(&cvq->lock);
337 memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
338 sizeof(struct virtio_pmd_ctrl));
340 if (virtio_with_packed_queue(vq->hw))
341 result = virtio_send_command_packed(cvq, ctrl, dlen, pkt_num);
343 result = virtio_send_command_split(cvq, ctrl, dlen, pkt_num);
345 rte_spinlock_unlock(&cvq->lock);
346 return result->status;
350 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
352 struct virtio_hw *hw = dev->data->dev_private;
353 struct virtio_pmd_ctrl ctrl;
357 ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
358 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
359 memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
361 dlen[0] = sizeof(uint16_t);
363 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
365 PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
366 "failed, this is too late now...");
374 virtio_dev_queue_release(void *queue __rte_unused)
380 virtio_get_nr_vq(struct virtio_hw *hw)
382 uint16_t nr_vq = hw->max_queue_pairs * 2;
384 if (virtio_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
391 virtio_init_vring(struct virtqueue *vq)
393 int size = vq->vq_nentries;
394 uint8_t *ring_mem = vq->vq_ring_virt_mem;
396 PMD_INIT_FUNC_TRACE();
398 memset(ring_mem, 0, vq->vq_ring_size);
400 vq->vq_used_cons_idx = 0;
401 vq->vq_desc_head_idx = 0;
402 vq->vq_avail_idx = 0;
403 vq->vq_desc_tail_idx = (uint16_t)(vq->vq_nentries - 1);
404 vq->vq_free_cnt = vq->vq_nentries;
405 memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries);
406 if (virtio_with_packed_queue(vq->hw)) {
407 vring_init_packed(&vq->vq_packed.ring, ring_mem,
408 VIRTIO_VRING_ALIGN, size);
409 vring_desc_init_packed(vq, size);
411 struct vring *vr = &vq->vq_split.ring;
413 vring_init_split(vr, ring_mem, VIRTIO_VRING_ALIGN, size);
414 vring_desc_init_split(vr->desc, size);
417 * Disable device(host) interrupting guest
419 virtqueue_disable_intr(vq);
423 virtio_init_queue(struct rte_eth_dev *dev, uint16_t queue_idx)
425 char vq_name[VIRTQUEUE_MAX_NAME_SZ];
426 char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
427 const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
428 unsigned int vq_size, size;
429 struct virtio_hw *hw = dev->data->dev_private;
430 struct virtnet_rx *rxvq = NULL;
431 struct virtnet_tx *txvq = NULL;
432 struct virtnet_ctl *cvq = NULL;
433 struct virtqueue *vq;
434 size_t sz_hdr_mz = 0;
435 void *sw_ring = NULL;
436 int queue_type = virtio_get_queue_type(hw, queue_idx);
438 int numa_node = dev->device->numa_node;
439 struct rte_mbuf *fake_mbuf = NULL;
441 PMD_INIT_LOG(INFO, "setting up queue: %u on NUMA node %d",
442 queue_idx, numa_node);
445 * Read the virtqueue size from the Queue Size field
446 * Always power of 2 and if 0 virtqueue does not exist
448 vq_size = VIRTIO_OPS(hw)->get_queue_num(hw, queue_idx);
449 PMD_INIT_LOG(DEBUG, "vq_size: %u", vq_size);
451 PMD_INIT_LOG(ERR, "virtqueue does not exist");
455 if (!virtio_with_packed_queue(hw) && !rte_is_power_of_2(vq_size)) {
456 PMD_INIT_LOG(ERR, "split virtqueue size is not power of 2");
460 snprintf(vq_name, sizeof(vq_name), "port%d_vq%d",
461 dev->data->port_id, queue_idx);
463 size = RTE_ALIGN_CEIL(sizeof(*vq) +
464 vq_size * sizeof(struct vq_desc_extra),
465 RTE_CACHE_LINE_SIZE);
466 if (queue_type == VTNET_TQ) {
468 * For each xmit packet, allocate a virtio_net_hdr
469 * and indirect ring elements
471 sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
472 } else if (queue_type == VTNET_CQ) {
473 /* Allocate a page for control vq command, data and status */
474 sz_hdr_mz = rte_mem_page_size();
477 vq = rte_zmalloc_socket(vq_name, size, RTE_CACHE_LINE_SIZE,
480 PMD_INIT_LOG(ERR, "can not allocate vq");
483 hw->vqs[queue_idx] = vq;
486 vq->vq_queue_index = queue_idx;
487 vq->vq_nentries = vq_size;
488 if (virtio_with_packed_queue(hw)) {
489 vq->vq_packed.used_wrap_counter = 1;
490 vq->vq_packed.cached_flags = VRING_PACKED_DESC_F_AVAIL;
491 vq->vq_packed.event_flags_shadow = 0;
492 if (queue_type == VTNET_RQ)
493 vq->vq_packed.cached_flags |= VRING_DESC_F_WRITE;
497 * Reserve a memzone for vring elements
499 size = vring_size(hw, vq_size, VIRTIO_VRING_ALIGN);
500 vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_VRING_ALIGN);
501 PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
502 size, vq->vq_ring_size);
504 mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
505 numa_node, RTE_MEMZONE_IOVA_CONTIG,
508 if (rte_errno == EEXIST)
509 mz = rte_memzone_lookup(vq_name);
516 memset(mz->addr, 0, mz->len);
518 vq->vq_ring_mem = mz->iova;
519 vq->vq_ring_virt_mem = mz->addr;
520 PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem: 0x%" PRIx64,
522 PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64,
523 (uint64_t)(uintptr_t)mz->addr);
525 virtio_init_vring(vq);
528 snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_vq%d_hdr",
529 dev->data->port_id, queue_idx);
530 hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
531 numa_node, RTE_MEMZONE_IOVA_CONTIG,
532 RTE_CACHE_LINE_SIZE);
533 if (hdr_mz == NULL) {
534 if (rte_errno == EEXIST)
535 hdr_mz = rte_memzone_lookup(vq_hdr_name);
536 if (hdr_mz == NULL) {
543 if (queue_type == VTNET_RQ) {
544 size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
545 sizeof(vq->sw_ring[0]);
547 sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
548 RTE_CACHE_LINE_SIZE, numa_node);
550 PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
555 fake_mbuf = rte_zmalloc_socket("sw_ring", sizeof(*fake_mbuf),
556 RTE_CACHE_LINE_SIZE, numa_node);
558 PMD_INIT_LOG(ERR, "can not allocate fake mbuf");
563 vq->sw_ring = sw_ring;
565 rxvq->port_id = dev->data->port_id;
567 rxvq->fake_mbuf = fake_mbuf;
568 } else if (queue_type == VTNET_TQ) {
570 txvq->port_id = dev->data->port_id;
572 txvq->virtio_net_hdr_mz = hdr_mz;
573 txvq->virtio_net_hdr_mem = hdr_mz->iova;
574 } else if (queue_type == VTNET_CQ) {
577 cvq->virtio_net_hdr_mz = hdr_mz;
578 cvq->virtio_net_hdr_mem = hdr_mz->iova;
579 memset(cvq->virtio_net_hdr_mz->addr, 0, rte_mem_page_size());
584 if (queue_type == VTNET_TQ) {
585 struct virtio_tx_region *txr;
589 memset(txr, 0, vq_size * sizeof(*txr));
590 for (i = 0; i < vq_size; i++) {
591 /* first indirect descriptor is always the tx header */
592 if (!virtio_with_packed_queue(hw)) {
593 struct vring_desc *start_dp = txr[i].tx_indir;
594 vring_desc_init_split(start_dp,
595 RTE_DIM(txr[i].tx_indir));
596 start_dp->addr = txvq->virtio_net_hdr_mem
598 + offsetof(struct virtio_tx_region,
600 start_dp->len = hw->vtnet_hdr_size;
601 start_dp->flags = VRING_DESC_F_NEXT;
603 struct vring_packed_desc *start_dp =
604 txr[i].tx_packed_indir;
605 vring_desc_init_indirect_packed(start_dp,
606 RTE_DIM(txr[i].tx_packed_indir));
607 start_dp->addr = txvq->virtio_net_hdr_mem
609 + offsetof(struct virtio_tx_region,
611 start_dp->len = hw->vtnet_hdr_size;
616 if (VIRTIO_OPS(hw)->setup_queue(hw, vq) < 0) {
617 PMD_INIT_LOG(ERR, "setup_queue failed");
630 rte_memzone_free(hdr_mz);
632 rte_memzone_free(mz);
640 virtio_free_queues(struct virtio_hw *hw)
642 uint16_t nr_vq = virtio_get_nr_vq(hw);
643 struct virtqueue *vq;
650 for (i = 0; i < nr_vq; i++) {
655 queue_type = virtio_get_queue_type(hw, i);
656 if (queue_type == VTNET_RQ) {
657 rte_free(vq->rxq.fake_mbuf);
658 rte_free(vq->sw_ring);
659 rte_memzone_free(vq->rxq.mz);
660 } else if (queue_type == VTNET_TQ) {
661 rte_memzone_free(vq->txq.mz);
662 rte_memzone_free(vq->txq.virtio_net_hdr_mz);
664 rte_memzone_free(vq->cq.mz);
665 rte_memzone_free(vq->cq.virtio_net_hdr_mz);
677 virtio_alloc_queues(struct rte_eth_dev *dev)
679 struct virtio_hw *hw = dev->data->dev_private;
680 uint16_t nr_vq = virtio_get_nr_vq(hw);
684 hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
686 PMD_INIT_LOG(ERR, "failed to allocate vqs");
690 for (i = 0; i < nr_vq; i++) {
691 ret = virtio_init_queue(dev, i);
693 virtio_free_queues(hw);
701 static void virtio_queues_unbind_intr(struct rte_eth_dev *dev);
704 virtio_dev_close(struct rte_eth_dev *dev)
706 struct virtio_hw *hw = dev->data->dev_private;
707 struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
709 PMD_INIT_LOG(DEBUG, "virtio_dev_close");
710 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
718 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
719 VIRTIO_OPS(hw)->set_config_irq(hw, VIRTIO_MSI_NO_VECTOR);
721 virtio_queues_unbind_intr(dev);
723 if (intr_conf->lsc || intr_conf->rxq) {
724 virtio_intr_disable(dev);
725 rte_intr_efd_disable(dev->intr_handle);
726 rte_free(dev->intr_handle->intr_vec);
727 dev->intr_handle->intr_vec = NULL;
731 virtio_dev_free_mbufs(dev);
732 virtio_free_queues(hw);
734 return VIRTIO_OPS(hw)->dev_close(hw);
738 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
740 struct virtio_hw *hw = dev->data->dev_private;
741 struct virtio_pmd_ctrl ctrl;
745 if (!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
746 PMD_INIT_LOG(INFO, "host does not support rx control");
750 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
751 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
755 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
757 PMD_INIT_LOG(ERR, "Failed to enable promisc");
765 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
767 struct virtio_hw *hw = dev->data->dev_private;
768 struct virtio_pmd_ctrl ctrl;
772 if (!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
773 PMD_INIT_LOG(INFO, "host does not support rx control");
777 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
778 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
782 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
784 PMD_INIT_LOG(ERR, "Failed to disable promisc");
792 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
794 struct virtio_hw *hw = dev->data->dev_private;
795 struct virtio_pmd_ctrl ctrl;
799 if (!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
800 PMD_INIT_LOG(INFO, "host does not support rx control");
804 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
805 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
809 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
811 PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
819 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
821 struct virtio_hw *hw = dev->data->dev_private;
822 struct virtio_pmd_ctrl ctrl;
826 if (!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
827 PMD_INIT_LOG(INFO, "host does not support rx control");
831 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
832 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
836 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
838 PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
846 virtio_rx_mem_pool_buf_size(struct rte_mempool *mp)
848 return rte_pktmbuf_data_room_size(mp) - RTE_PKTMBUF_HEADROOM;
852 virtio_rx_check_scatter(uint16_t max_rx_pkt_len, uint16_t rx_buf_size,
853 bool rx_scatter_enabled, const char **error)
855 if (!rx_scatter_enabled && max_rx_pkt_len > rx_buf_size) {
856 *error = "Rx scatter is disabled and RxQ mbuf pool object size is too small";
864 virtio_check_scatter_on_all_rx_queues(struct rte_eth_dev *dev,
867 struct virtio_hw *hw = dev->data->dev_private;
868 struct virtnet_rx *rxvq;
869 struct virtqueue *vq;
877 for (qidx = 0; (vq = hw->vqs[2 * qidx + VTNET_SQ_RQ_QUEUE_IDX]) != NULL;
880 if (rxvq->mpool == NULL)
882 buf_size = virtio_rx_mem_pool_buf_size(rxvq->mpool);
884 if (!virtio_rx_check_scatter(frame_size, buf_size,
885 hw->rx_ol_scatter, &error)) {
886 PMD_INIT_LOG(ERR, "MTU check for RxQ %u failed: %s",
895 #define VLAN_TAG_LEN 4 /* 802.3ac tag (not DMA'd) */
897 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
899 struct virtio_hw *hw = dev->data->dev_private;
900 uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
902 uint32_t frame_size = mtu + ether_hdr_len;
903 uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
905 max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
907 if (mtu < RTE_ETHER_MIN_MTU || frame_size > max_frame_size) {
908 PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
909 RTE_ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
913 if (!virtio_check_scatter_on_all_rx_queues(dev, frame_size)) {
914 PMD_INIT_LOG(ERR, "MTU vs Rx scatter and Rx buffers check failed");
918 hw->max_rx_pkt_len = frame_size;
919 dev->data->dev_conf.rxmode.max_rx_pkt_len = hw->max_rx_pkt_len;
925 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
927 struct virtio_hw *hw = dev->data->dev_private;
928 struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
929 struct virtqueue *vq = virtnet_rxq_to_vq(rxvq);
931 virtqueue_enable_intr(vq);
932 virtio_mb(hw->weak_barriers);
937 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
939 struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
940 struct virtqueue *vq = virtnet_rxq_to_vq(rxvq);
942 virtqueue_disable_intr(vq);
947 * dev_ops for virtio, bare necessities for basic operation
949 static const struct eth_dev_ops virtio_eth_dev_ops = {
950 .dev_configure = virtio_dev_configure,
951 .dev_start = virtio_dev_start,
952 .dev_stop = virtio_dev_stop,
953 .dev_close = virtio_dev_close,
954 .promiscuous_enable = virtio_dev_promiscuous_enable,
955 .promiscuous_disable = virtio_dev_promiscuous_disable,
956 .allmulticast_enable = virtio_dev_allmulticast_enable,
957 .allmulticast_disable = virtio_dev_allmulticast_disable,
958 .mtu_set = virtio_mtu_set,
959 .dev_infos_get = virtio_dev_info_get,
960 .stats_get = virtio_dev_stats_get,
961 .xstats_get = virtio_dev_xstats_get,
962 .xstats_get_names = virtio_dev_xstats_get_names,
963 .stats_reset = virtio_dev_stats_reset,
964 .xstats_reset = virtio_dev_stats_reset,
965 .link_update = virtio_dev_link_update,
966 .vlan_offload_set = virtio_dev_vlan_offload_set,
967 .rx_queue_setup = virtio_dev_rx_queue_setup,
968 .rx_queue_intr_enable = virtio_dev_rx_queue_intr_enable,
969 .rx_queue_intr_disable = virtio_dev_rx_queue_intr_disable,
970 .rx_queue_release = virtio_dev_queue_release,
971 .tx_queue_setup = virtio_dev_tx_queue_setup,
972 .tx_queue_release = virtio_dev_queue_release,
973 /* collect stats per queue */
974 .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
975 .vlan_filter_set = virtio_vlan_filter_set,
976 .mac_addr_add = virtio_mac_addr_add,
977 .mac_addr_remove = virtio_mac_addr_remove,
978 .mac_addr_set = virtio_mac_addr_set,
982 * dev_ops for virtio-user in secondary processes, as we just have
983 * some limited supports currently.
985 const struct eth_dev_ops virtio_user_secondary_eth_dev_ops = {
986 .dev_infos_get = virtio_dev_info_get,
987 .stats_get = virtio_dev_stats_get,
988 .xstats_get = virtio_dev_xstats_get,
989 .xstats_get_names = virtio_dev_xstats_get_names,
990 .stats_reset = virtio_dev_stats_reset,
991 .xstats_reset = virtio_dev_stats_reset,
992 /* collect stats per queue */
993 .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
997 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1001 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1002 const struct virtnet_tx *txvq = dev->data->tx_queues[i];
1006 stats->opackets += txvq->stats.packets;
1007 stats->obytes += txvq->stats.bytes;
1009 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
1010 stats->q_opackets[i] = txvq->stats.packets;
1011 stats->q_obytes[i] = txvq->stats.bytes;
1015 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1016 const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1020 stats->ipackets += rxvq->stats.packets;
1021 stats->ibytes += rxvq->stats.bytes;
1022 stats->ierrors += rxvq->stats.errors;
1024 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
1025 stats->q_ipackets[i] = rxvq->stats.packets;
1026 stats->q_ibytes[i] = rxvq->stats.bytes;
1030 stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
1033 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
1034 struct rte_eth_xstat_name *xstats_names,
1035 __rte_unused unsigned limit)
1041 unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
1042 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
1044 if (xstats_names != NULL) {
1045 /* Note: limit checked in rte_eth_xstats_names() */
1047 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1048 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1051 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1052 snprintf(xstats_names[count].name,
1053 sizeof(xstats_names[count].name),
1055 rte_virtio_rxq_stat_strings[t].name);
1060 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1061 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1064 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1065 snprintf(xstats_names[count].name,
1066 sizeof(xstats_names[count].name),
1068 rte_virtio_txq_stat_strings[t].name);
1078 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
1084 unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
1085 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
1090 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1091 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1098 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1099 xstats[count].value = *(uint64_t *)(((char *)rxvq) +
1100 rte_virtio_rxq_stat_strings[t].offset);
1101 xstats[count].id = count;
1106 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1107 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1114 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1115 xstats[count].value = *(uint64_t *)(((char *)txvq) +
1116 rte_virtio_txq_stat_strings[t].offset);
1117 xstats[count].id = count;
1126 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1128 virtio_update_stats(dev, stats);
1134 virtio_dev_stats_reset(struct rte_eth_dev *dev)
1138 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1139 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1143 txvq->stats.packets = 0;
1144 txvq->stats.bytes = 0;
1145 txvq->stats.multicast = 0;
1146 txvq->stats.broadcast = 0;
1147 memset(txvq->stats.size_bins, 0,
1148 sizeof(txvq->stats.size_bins[0]) * 8);
1151 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1152 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1156 rxvq->stats.packets = 0;
1157 rxvq->stats.bytes = 0;
1158 rxvq->stats.errors = 0;
1159 rxvq->stats.multicast = 0;
1160 rxvq->stats.broadcast = 0;
1161 memset(rxvq->stats.size_bins, 0,
1162 sizeof(rxvq->stats.size_bins[0]) * 8);
1169 virtio_set_hwaddr(struct virtio_hw *hw)
1171 virtio_write_dev_config(hw,
1172 offsetof(struct virtio_net_config, mac),
1173 &hw->mac_addr, RTE_ETHER_ADDR_LEN);
1177 virtio_get_hwaddr(struct virtio_hw *hw)
1179 if (virtio_with_feature(hw, VIRTIO_NET_F_MAC)) {
1180 virtio_read_dev_config(hw,
1181 offsetof(struct virtio_net_config, mac),
1182 &hw->mac_addr, RTE_ETHER_ADDR_LEN);
1184 rte_eth_random_addr(&hw->mac_addr[0]);
1185 virtio_set_hwaddr(hw);
1190 virtio_mac_table_set(struct virtio_hw *hw,
1191 const struct virtio_net_ctrl_mac *uc,
1192 const struct virtio_net_ctrl_mac *mc)
1194 struct virtio_pmd_ctrl ctrl;
1197 if (!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1198 PMD_DRV_LOG(INFO, "host does not support mac table");
1202 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1203 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1205 len[0] = uc->entries * RTE_ETHER_ADDR_LEN + sizeof(uc->entries);
1206 memcpy(ctrl.data, uc, len[0]);
1208 len[1] = mc->entries * RTE_ETHER_ADDR_LEN + sizeof(mc->entries);
1209 memcpy(ctrl.data + len[0], mc, len[1]);
1211 err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1213 PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1218 virtio_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
1219 uint32_t index, uint32_t vmdq __rte_unused)
1221 struct virtio_hw *hw = dev->data->dev_private;
1222 const struct rte_ether_addr *addrs = dev->data->mac_addrs;
1224 struct virtio_net_ctrl_mac *uc, *mc;
1226 if (index >= VIRTIO_MAX_MAC_ADDRS) {
1227 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1231 uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1232 sizeof(uc->entries));
1234 mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1235 sizeof(mc->entries));
1238 for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1239 const struct rte_ether_addr *addr
1240 = (i == index) ? mac_addr : addrs + i;
1241 struct virtio_net_ctrl_mac *tbl
1242 = rte_is_multicast_ether_addr(addr) ? mc : uc;
1244 memcpy(&tbl->macs[tbl->entries++], addr, RTE_ETHER_ADDR_LEN);
1247 return virtio_mac_table_set(hw, uc, mc);
1251 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1253 struct virtio_hw *hw = dev->data->dev_private;
1254 struct rte_ether_addr *addrs = dev->data->mac_addrs;
1255 struct virtio_net_ctrl_mac *uc, *mc;
1258 if (index >= VIRTIO_MAX_MAC_ADDRS) {
1259 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1263 uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1264 sizeof(uc->entries));
1266 mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1267 sizeof(mc->entries));
1270 for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1271 struct virtio_net_ctrl_mac *tbl;
1273 if (i == index || rte_is_zero_ether_addr(addrs + i))
1276 tbl = rte_is_multicast_ether_addr(addrs + i) ? mc : uc;
1277 memcpy(&tbl->macs[tbl->entries++], addrs + i,
1278 RTE_ETHER_ADDR_LEN);
1281 virtio_mac_table_set(hw, uc, mc);
1285 virtio_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1287 struct virtio_hw *hw = dev->data->dev_private;
1289 memcpy(hw->mac_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1291 /* Use atomic update if available */
1292 if (virtio_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1293 struct virtio_pmd_ctrl ctrl;
1294 int len = RTE_ETHER_ADDR_LEN;
1296 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1297 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1299 memcpy(ctrl.data, mac_addr, RTE_ETHER_ADDR_LEN);
1300 return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1303 if (!virtio_with_feature(hw, VIRTIO_NET_F_MAC))
1306 virtio_set_hwaddr(hw);
1311 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1313 struct virtio_hw *hw = dev->data->dev_private;
1314 struct virtio_pmd_ctrl ctrl;
1317 if (!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1320 ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1321 ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1322 memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1323 len = sizeof(vlan_id);
1325 return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1329 virtio_intr_unmask(struct rte_eth_dev *dev)
1331 struct virtio_hw *hw = dev->data->dev_private;
1333 if (rte_intr_ack(dev->intr_handle) < 0)
1336 if (VIRTIO_OPS(hw)->intr_detect)
1337 VIRTIO_OPS(hw)->intr_detect(hw);
1343 virtio_intr_enable(struct rte_eth_dev *dev)
1345 struct virtio_hw *hw = dev->data->dev_private;
1347 if (rte_intr_enable(dev->intr_handle) < 0)
1350 if (VIRTIO_OPS(hw)->intr_detect)
1351 VIRTIO_OPS(hw)->intr_detect(hw);
1357 virtio_intr_disable(struct rte_eth_dev *dev)
1359 struct virtio_hw *hw = dev->data->dev_private;
1361 if (rte_intr_disable(dev->intr_handle) < 0)
1364 if (VIRTIO_OPS(hw)->intr_detect)
1365 VIRTIO_OPS(hw)->intr_detect(hw);
1371 virtio_ethdev_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1373 uint64_t host_features;
1375 /* Prepare guest_features: feature that driver wants to support */
1376 PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1379 /* Read device(host) feature bits */
1380 host_features = VIRTIO_OPS(hw)->get_features(hw);
1381 PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1384 /* If supported, ensure MTU value is valid before acknowledging it. */
1385 if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1386 struct virtio_net_config config;
1388 virtio_read_dev_config(hw,
1389 offsetof(struct virtio_net_config, mtu),
1390 &config.mtu, sizeof(config.mtu));
1392 if (config.mtu < RTE_ETHER_MIN_MTU)
1393 req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1397 * Negotiate features: Subset of device feature bits are written back
1398 * guest feature bits.
1400 hw->guest_features = req_features;
1401 hw->guest_features = virtio_negotiate_features(hw, host_features);
1402 PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1403 hw->guest_features);
1405 if (VIRTIO_OPS(hw)->features_ok(hw) < 0)
1408 if (virtio_with_feature(hw, VIRTIO_F_VERSION_1)) {
1409 virtio_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1411 if (!(virtio_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1412 PMD_INIT_LOG(ERR, "Failed to set FEATURES_OK status!");
1417 hw->req_guest_features = req_features;
1423 virtio_dev_pause(struct rte_eth_dev *dev)
1425 struct virtio_hw *hw = dev->data->dev_private;
1427 rte_spinlock_lock(&hw->state_lock);
1429 if (hw->started == 0) {
1430 /* Device is just stopped. */
1431 rte_spinlock_unlock(&hw->state_lock);
1436 * Prevent the worker threads from touching queues to avoid contention,
1437 * 1 ms should be enough for the ongoing Tx function to finish.
1444 * Recover hw state to let the worker threads continue.
1447 virtio_dev_resume(struct rte_eth_dev *dev)
1449 struct virtio_hw *hw = dev->data->dev_private;
1452 rte_spinlock_unlock(&hw->state_lock);
1456 * Should be called only after device is paused.
1459 virtio_inject_pkts(struct rte_eth_dev *dev, struct rte_mbuf **tx_pkts,
1462 struct virtio_hw *hw = dev->data->dev_private;
1463 struct virtnet_tx *txvq = dev->data->tx_queues[0];
1466 hw->inject_pkts = tx_pkts;
1467 ret = dev->tx_pkt_burst(txvq, tx_pkts, nb_pkts);
1468 hw->inject_pkts = NULL;
1474 virtio_notify_peers(struct rte_eth_dev *dev)
1476 struct virtio_hw *hw = dev->data->dev_private;
1477 struct virtnet_rx *rxvq;
1478 struct rte_mbuf *rarp_mbuf;
1480 if (!dev->data->rx_queues)
1483 rxvq = dev->data->rx_queues[0];
1487 rarp_mbuf = rte_net_make_rarp_packet(rxvq->mpool,
1488 (struct rte_ether_addr *)hw->mac_addr);
1489 if (rarp_mbuf == NULL) {
1490 PMD_DRV_LOG(ERR, "failed to make RARP packet.");
1494 /* If virtio port just stopped, no need to send RARP */
1495 if (virtio_dev_pause(dev) < 0) {
1496 rte_pktmbuf_free(rarp_mbuf);
1500 virtio_inject_pkts(dev, &rarp_mbuf, 1);
1501 virtio_dev_resume(dev);
1505 virtio_ack_link_announce(struct rte_eth_dev *dev)
1507 struct virtio_hw *hw = dev->data->dev_private;
1508 struct virtio_pmd_ctrl ctrl;
1510 ctrl.hdr.class = VIRTIO_NET_CTRL_ANNOUNCE;
1511 ctrl.hdr.cmd = VIRTIO_NET_CTRL_ANNOUNCE_ACK;
1513 virtio_send_command(hw->cvq, &ctrl, NULL, 0);
1517 * Process virtio config changed interrupt. Call the callback
1518 * if link state changed, generate gratuitous RARP packet if
1519 * the status indicates an ANNOUNCE.
1522 virtio_interrupt_handler(void *param)
1524 struct rte_eth_dev *dev = param;
1525 struct virtio_hw *hw = dev->data->dev_private;
1529 /* Read interrupt status which clears interrupt */
1530 isr = virtio_get_isr(hw);
1531 PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1533 if (virtio_intr_unmask(dev) < 0)
1534 PMD_DRV_LOG(ERR, "interrupt enable failed");
1536 if (isr & VIRTIO_ISR_CONFIG) {
1537 if (virtio_dev_link_update(dev, 0) == 0)
1538 rte_eth_dev_callback_process(dev,
1539 RTE_ETH_EVENT_INTR_LSC,
1542 if (virtio_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1543 virtio_read_dev_config(hw,
1544 offsetof(struct virtio_net_config, status),
1545 &status, sizeof(status));
1546 if (status & VIRTIO_NET_S_ANNOUNCE) {
1547 virtio_notify_peers(dev);
1549 virtio_ack_link_announce(dev);
1555 /* set rx and tx handlers according to what is supported */
1557 set_rxtx_funcs(struct rte_eth_dev *eth_dev)
1559 struct virtio_hw *hw = eth_dev->data->dev_private;
1561 eth_dev->tx_pkt_prepare = virtio_xmit_pkts_prepare;
1562 if (virtio_with_packed_queue(hw)) {
1564 "virtio: using packed ring %s Tx path on port %u",
1565 hw->use_vec_tx ? "vectorized" : "standard",
1566 eth_dev->data->port_id);
1568 eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed_vec;
1570 eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed;
1572 if (hw->use_inorder_tx) {
1573 PMD_INIT_LOG(INFO, "virtio: using inorder Tx path on port %u",
1574 eth_dev->data->port_id);
1575 eth_dev->tx_pkt_burst = virtio_xmit_pkts_inorder;
1577 PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
1578 eth_dev->data->port_id);
1579 eth_dev->tx_pkt_burst = virtio_xmit_pkts;
1583 if (virtio_with_packed_queue(hw)) {
1584 if (hw->use_vec_rx) {
1586 "virtio: using packed ring vectorized Rx path on port %u",
1587 eth_dev->data->port_id);
1588 eth_dev->rx_pkt_burst =
1589 &virtio_recv_pkts_packed_vec;
1590 } else if (virtio_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1592 "virtio: using packed ring mergeable buffer Rx path on port %u",
1593 eth_dev->data->port_id);
1594 eth_dev->rx_pkt_burst =
1595 &virtio_recv_mergeable_pkts_packed;
1598 "virtio: using packed ring standard Rx path on port %u",
1599 eth_dev->data->port_id);
1600 eth_dev->rx_pkt_burst = &virtio_recv_pkts_packed;
1603 if (hw->use_vec_rx) {
1604 PMD_INIT_LOG(INFO, "virtio: using vectorized Rx path on port %u",
1605 eth_dev->data->port_id);
1606 eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1607 } else if (hw->use_inorder_rx) {
1609 "virtio: using inorder Rx path on port %u",
1610 eth_dev->data->port_id);
1611 eth_dev->rx_pkt_burst = &virtio_recv_pkts_inorder;
1612 } else if (virtio_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1614 "virtio: using mergeable buffer Rx path on port %u",
1615 eth_dev->data->port_id);
1616 eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1618 PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
1619 eth_dev->data->port_id);
1620 eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1626 /* Only support 1:1 queue/interrupt mapping so far.
1627 * TODO: support n:1 queue/interrupt mapping when there are limited number of
1628 * interrupt vectors (<N+1).
1631 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1634 struct virtio_hw *hw = dev->data->dev_private;
1636 PMD_INIT_LOG(INFO, "queue/interrupt binding");
1637 for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1638 dev->intr_handle->intr_vec[i] = i + 1;
1639 if (VIRTIO_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1640 VIRTIO_MSI_NO_VECTOR) {
1641 PMD_DRV_LOG(ERR, "failed to set queue vector");
1650 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1653 struct virtio_hw *hw = dev->data->dev_private;
1655 PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1656 for (i = 0; i < dev->data->nb_rx_queues; ++i)
1657 VIRTIO_OPS(hw)->set_queue_irq(hw,
1658 hw->vqs[i * VTNET_CQ],
1659 VIRTIO_MSI_NO_VECTOR);
1663 virtio_configure_intr(struct rte_eth_dev *dev)
1665 struct virtio_hw *hw = dev->data->dev_private;
1667 if (!rte_intr_cap_multiple(dev->intr_handle)) {
1668 PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1672 if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1673 PMD_INIT_LOG(ERR, "Fail to create eventfd");
1677 if (!dev->intr_handle->intr_vec) {
1678 dev->intr_handle->intr_vec =
1679 rte_zmalloc("intr_vec",
1680 hw->max_queue_pairs * sizeof(int), 0);
1681 if (!dev->intr_handle->intr_vec) {
1682 PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1683 hw->max_queue_pairs);
1688 /* Re-register callback to update max_intr */
1689 rte_intr_callback_unregister(dev->intr_handle,
1690 virtio_interrupt_handler,
1692 rte_intr_callback_register(dev->intr_handle,
1693 virtio_interrupt_handler,
1696 /* DO NOT try to remove this! This function will enable msix, or QEMU
1697 * will encounter SIGSEGV when DRIVER_OK is sent.
1698 * And for legacy devices, this should be done before queue/vec binding
1699 * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1700 * (22) will be ignored.
1702 if (virtio_intr_enable(dev) < 0) {
1703 PMD_DRV_LOG(ERR, "interrupt enable failed");
1707 if (virtio_queues_bind_intr(dev) < 0) {
1708 PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1714 #define DUPLEX_UNKNOWN 0xff
1715 /* reset device and renegotiate features if needed */
1717 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1719 struct virtio_hw *hw = eth_dev->data->dev_private;
1720 struct virtio_net_config *config;
1721 struct virtio_net_config local_config;
1724 /* Reset the device although not necessary at startup */
1728 virtio_dev_free_mbufs(eth_dev);
1729 virtio_free_queues(hw);
1732 /* Tell the host we've noticed this device. */
1733 virtio_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1735 /* Tell the host we've known how to drive the device. */
1736 virtio_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1737 if (virtio_ethdev_negotiate_features(hw, req_features) < 0)
1740 hw->weak_barriers = !virtio_with_feature(hw, VIRTIO_F_ORDER_PLATFORM);
1742 /* If host does not support both status and MSI-X then disable LSC */
1743 if (virtio_with_feature(hw, VIRTIO_NET_F_STATUS) && hw->intr_lsc)
1744 eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1746 eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1748 eth_dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
1750 /* Setting up rx_header size for the device */
1751 if (virtio_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1752 virtio_with_feature(hw, VIRTIO_F_VERSION_1) ||
1753 virtio_with_packed_queue(hw))
1754 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1756 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1758 /* Copy the permanent MAC address to: virtio_hw */
1759 virtio_get_hwaddr(hw);
1760 rte_ether_addr_copy((struct rte_ether_addr *)hw->mac_addr,
1761 ð_dev->data->mac_addrs[0]);
1763 "PORT MAC: " RTE_ETHER_ADDR_PRT_FMT,
1764 hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1765 hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1767 if (hw->speed == ETH_SPEED_NUM_UNKNOWN) {
1768 if (virtio_with_feature(hw, VIRTIO_NET_F_SPEED_DUPLEX)) {
1769 config = &local_config;
1770 virtio_read_dev_config(hw,
1771 offsetof(struct virtio_net_config, speed),
1772 &config->speed, sizeof(config->speed));
1773 virtio_read_dev_config(hw,
1774 offsetof(struct virtio_net_config, duplex),
1775 &config->duplex, sizeof(config->duplex));
1776 hw->speed = config->speed;
1777 hw->duplex = config->duplex;
1780 if (hw->duplex == DUPLEX_UNKNOWN)
1781 hw->duplex = ETH_LINK_FULL_DUPLEX;
1782 PMD_INIT_LOG(DEBUG, "link speed = %d, duplex = %d",
1783 hw->speed, hw->duplex);
1784 if (virtio_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1785 config = &local_config;
1787 virtio_read_dev_config(hw,
1788 offsetof(struct virtio_net_config, mac),
1789 &config->mac, sizeof(config->mac));
1791 if (virtio_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1792 virtio_read_dev_config(hw,
1793 offsetof(struct virtio_net_config, status),
1794 &config->status, sizeof(config->status));
1797 "VIRTIO_NET_F_STATUS is not supported");
1801 if (virtio_with_feature(hw, VIRTIO_NET_F_MQ)) {
1802 virtio_read_dev_config(hw,
1803 offsetof(struct virtio_net_config, max_virtqueue_pairs),
1804 &config->max_virtqueue_pairs,
1805 sizeof(config->max_virtqueue_pairs));
1808 "VIRTIO_NET_F_MQ is not supported");
1809 config->max_virtqueue_pairs = 1;
1812 hw->max_queue_pairs = config->max_virtqueue_pairs;
1814 if (virtio_with_feature(hw, VIRTIO_NET_F_MTU)) {
1815 virtio_read_dev_config(hw,
1816 offsetof(struct virtio_net_config, mtu),
1818 sizeof(config->mtu));
1821 * MTU value has already been checked at negotiation
1822 * time, but check again in case it has changed since
1823 * then, which should not happen.
1825 if (config->mtu < RTE_ETHER_MIN_MTU) {
1826 PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1831 hw->max_mtu = config->mtu;
1832 /* Set initial MTU to maximum one supported by vhost */
1833 eth_dev->data->mtu = config->mtu;
1836 hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1837 VLAN_TAG_LEN - hw->vtnet_hdr_size;
1840 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1841 config->max_virtqueue_pairs);
1842 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1844 "PORT MAC: " RTE_ETHER_ADDR_PRT_FMT,
1845 config->mac[0], config->mac[1],
1846 config->mac[2], config->mac[3],
1847 config->mac[4], config->mac[5]);
1849 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1850 hw->max_queue_pairs = 1;
1851 hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1852 VLAN_TAG_LEN - hw->vtnet_hdr_size;
1855 ret = virtio_alloc_queues(eth_dev);
1859 if (eth_dev->data->dev_conf.intr_conf.rxq) {
1860 if (virtio_configure_intr(eth_dev) < 0) {
1861 PMD_INIT_LOG(ERR, "failed to configure interrupt");
1862 virtio_free_queues(hw);
1867 virtio_reinit_complete(hw);
1873 * This function is based on probe() function in virtio_pci.c
1874 * It returns 0 on success.
1877 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1879 struct virtio_hw *hw = eth_dev->data->dev_private;
1880 uint32_t speed = ETH_SPEED_NUM_UNKNOWN;
1884 if (sizeof(struct virtio_net_hdr_mrg_rxbuf) > RTE_PKTMBUF_HEADROOM) {
1886 "Not sufficient headroom required = %d, avail = %d",
1887 (int)sizeof(struct virtio_net_hdr_mrg_rxbuf),
1888 RTE_PKTMBUF_HEADROOM);
1893 eth_dev->dev_ops = &virtio_eth_dev_ops;
1894 eth_dev->rx_descriptor_done = virtio_dev_rx_queue_done;
1896 if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1897 set_rxtx_funcs(eth_dev);
1901 ret = virtio_dev_devargs_parse(eth_dev->device->devargs, &speed, &vectorized);
1905 hw->duplex = DUPLEX_UNKNOWN;
1907 /* Allocate memory for storing MAC addresses */
1908 eth_dev->data->mac_addrs = rte_zmalloc("virtio",
1909 VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN, 0);
1910 if (eth_dev->data->mac_addrs == NULL) {
1912 "Failed to allocate %d bytes needed to store MAC addresses",
1913 VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN);
1917 rte_spinlock_init(&hw->state_lock);
1919 /* reset device and negotiate default features */
1920 ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1922 goto err_virtio_init;
1925 if (!virtio_with_packed_queue(hw)) {
1928 #if defined(CC_AVX512_SUPPORT) || defined(RTE_ARCH_ARM)
1933 "building environment do not support packed ring vectorized");
1943 rte_free(eth_dev->data->mac_addrs);
1944 eth_dev->data->mac_addrs = NULL;
1949 virtio_dev_speed_capa_get(uint32_t speed)
1952 case ETH_SPEED_NUM_10G:
1953 return ETH_LINK_SPEED_10G;
1954 case ETH_SPEED_NUM_20G:
1955 return ETH_LINK_SPEED_20G;
1956 case ETH_SPEED_NUM_25G:
1957 return ETH_LINK_SPEED_25G;
1958 case ETH_SPEED_NUM_40G:
1959 return ETH_LINK_SPEED_40G;
1960 case ETH_SPEED_NUM_50G:
1961 return ETH_LINK_SPEED_50G;
1962 case ETH_SPEED_NUM_56G:
1963 return ETH_LINK_SPEED_56G;
1964 case ETH_SPEED_NUM_100G:
1965 return ETH_LINK_SPEED_100G;
1966 case ETH_SPEED_NUM_200G:
1967 return ETH_LINK_SPEED_200G;
1973 static int vectorized_check_handler(__rte_unused const char *key,
1974 const char *value, void *ret_val)
1976 if (strcmp(value, "1") == 0)
1977 *(int *)ret_val = 1;
1979 *(int *)ret_val = 0;
1984 #define VIRTIO_ARG_SPEED "speed"
1985 #define VIRTIO_ARG_VECTORIZED "vectorized"
1988 link_speed_handler(const char *key __rte_unused,
1989 const char *value, void *ret_val)
1992 if (!value || !ret_val)
1994 val = strtoul(value, NULL, 0);
1995 /* validate input */
1996 if (virtio_dev_speed_capa_get(val) == 0)
1998 *(uint32_t *)ret_val = val;
2005 virtio_dev_devargs_parse(struct rte_devargs *devargs, uint32_t *speed, int *vectorized)
2007 struct rte_kvargs *kvlist;
2010 if (devargs == NULL)
2013 kvlist = rte_kvargs_parse(devargs->args, NULL);
2014 if (kvlist == NULL) {
2015 PMD_INIT_LOG(ERR, "error when parsing param");
2019 if (speed && rte_kvargs_count(kvlist, VIRTIO_ARG_SPEED) == 1) {
2020 ret = rte_kvargs_process(kvlist,
2022 link_speed_handler, speed);
2024 PMD_INIT_LOG(ERR, "Failed to parse %s",
2031 rte_kvargs_count(kvlist, VIRTIO_ARG_VECTORIZED) == 1) {
2032 ret = rte_kvargs_process(kvlist,
2033 VIRTIO_ARG_VECTORIZED,
2034 vectorized_check_handler, vectorized);
2036 PMD_INIT_LOG(ERR, "Failed to parse %s",
2037 VIRTIO_ARG_VECTORIZED);
2043 rte_kvargs_free(kvlist);
2048 rx_offload_enabled(struct virtio_hw *hw)
2050 return virtio_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM) ||
2051 virtio_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2052 virtio_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6);
2056 tx_offload_enabled(struct virtio_hw *hw)
2058 return virtio_with_feature(hw, VIRTIO_NET_F_CSUM) ||
2059 virtio_with_feature(hw, VIRTIO_NET_F_HOST_TSO4) ||
2060 virtio_with_feature(hw, VIRTIO_NET_F_HOST_TSO6);
2064 * Configure virtio device
2065 * It returns 0 on success.
2068 virtio_dev_configure(struct rte_eth_dev *dev)
2070 const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2071 const struct rte_eth_txmode *txmode = &dev->data->dev_conf.txmode;
2072 struct virtio_hw *hw = dev->data->dev_private;
2073 uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
2075 uint64_t rx_offloads = rxmode->offloads;
2076 uint64_t tx_offloads = txmode->offloads;
2077 uint64_t req_features;
2080 PMD_INIT_LOG(DEBUG, "configure");
2081 req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
2083 if (rxmode->mq_mode != ETH_MQ_RX_NONE) {
2085 "Unsupported Rx multi queue mode %d",
2090 if (txmode->mq_mode != ETH_MQ_TX_NONE) {
2092 "Unsupported Tx multi queue mode %d",
2097 if (dev->data->dev_conf.intr_conf.rxq) {
2098 ret = virtio_init_device(dev, hw->req_guest_features);
2103 if (rxmode->max_rx_pkt_len > hw->max_mtu + ether_hdr_len)
2104 req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
2106 hw->max_rx_pkt_len = rxmode->max_rx_pkt_len;
2108 if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2109 DEV_RX_OFFLOAD_TCP_CKSUM))
2110 req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
2112 if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO)
2114 (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2115 (1ULL << VIRTIO_NET_F_GUEST_TSO6);
2117 if (tx_offloads & (DEV_TX_OFFLOAD_UDP_CKSUM |
2118 DEV_TX_OFFLOAD_TCP_CKSUM))
2119 req_features |= (1ULL << VIRTIO_NET_F_CSUM);
2121 if (tx_offloads & DEV_TX_OFFLOAD_TCP_TSO)
2123 (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2124 (1ULL << VIRTIO_NET_F_HOST_TSO6);
2126 /* if request features changed, reinit the device */
2127 if (req_features != hw->req_guest_features) {
2128 ret = virtio_init_device(dev, req_features);
2133 if ((rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2134 DEV_RX_OFFLOAD_TCP_CKSUM)) &&
2135 !virtio_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
2137 "rx checksum not available on this host");
2141 if ((rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) &&
2142 (!virtio_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2143 !virtio_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6))) {
2145 "Large Receive Offload not available on this host");
2149 /* start control queue */
2150 if (virtio_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
2151 virtio_dev_cq_start(dev);
2153 if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
2156 hw->rx_ol_scatter = (rx_offloads & DEV_RX_OFFLOAD_SCATTER);
2158 if ((rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER) &&
2159 !virtio_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2161 "vlan filtering not available on this host");
2165 hw->has_tx_offload = tx_offload_enabled(hw);
2166 hw->has_rx_offload = rx_offload_enabled(hw);
2168 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2169 /* Enable vector (0) for Link State Intrerrupt */
2170 if (VIRTIO_OPS(hw)->set_config_irq(hw, 0) ==
2171 VIRTIO_MSI_NO_VECTOR) {
2172 PMD_DRV_LOG(ERR, "failed to set config vector");
2176 if (virtio_with_packed_queue(hw)) {
2177 #if defined(RTE_ARCH_X86_64) && defined(CC_AVX512_SUPPORT)
2178 if ((hw->use_vec_rx || hw->use_vec_tx) &&
2179 (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX512F) ||
2180 !virtio_with_feature(hw, VIRTIO_F_IN_ORDER) ||
2181 !virtio_with_feature(hw, VIRTIO_F_VERSION_1) ||
2182 rte_vect_get_max_simd_bitwidth() < RTE_VECT_SIMD_512)) {
2184 "disabled packed ring vectorized path for requirements not met");
2188 #elif defined(RTE_ARCH_ARM)
2189 if ((hw->use_vec_rx || hw->use_vec_tx) &&
2190 (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON) ||
2191 !virtio_with_feature(hw, VIRTIO_F_IN_ORDER) ||
2192 !virtio_with_feature(hw, VIRTIO_F_VERSION_1) ||
2193 rte_vect_get_max_simd_bitwidth() < RTE_VECT_SIMD_128)) {
2195 "disabled packed ring vectorized path for requirements not met");
2204 if (hw->use_vec_rx) {
2205 if (virtio_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2207 "disabled packed ring vectorized rx for mrg_rxbuf enabled");
2211 if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) {
2213 "disabled packed ring vectorized rx for TCP_LRO enabled");
2218 if (virtio_with_feature(hw, VIRTIO_F_IN_ORDER)) {
2219 hw->use_inorder_tx = 1;
2220 hw->use_inorder_rx = 1;
2224 if (hw->use_vec_rx) {
2225 #if defined RTE_ARCH_ARM
2226 if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
2228 "disabled split ring vectorized path for requirement not met");
2232 if (virtio_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2234 "disabled split ring vectorized rx for mrg_rxbuf enabled");
2238 if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2239 DEV_RX_OFFLOAD_TCP_CKSUM |
2240 DEV_RX_OFFLOAD_TCP_LRO |
2241 DEV_RX_OFFLOAD_VLAN_STRIP)) {
2243 "disabled split ring vectorized rx for offloading enabled");
2247 if (rte_vect_get_max_simd_bitwidth() < RTE_VECT_SIMD_128) {
2249 "disabled split ring vectorized rx, max SIMD bitwidth too low");
2260 virtio_dev_start(struct rte_eth_dev *dev)
2262 uint16_t nb_queues, i;
2263 struct virtqueue *vq;
2264 struct virtio_hw *hw = dev->data->dev_private;
2267 /* Finish the initialization of the queues */
2268 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2269 ret = virtio_dev_rx_queue_setup_finish(dev, i);
2273 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2274 ret = virtio_dev_tx_queue_setup_finish(dev, i);
2279 /* check if lsc interrupt feature is enabled */
2280 if (dev->data->dev_conf.intr_conf.lsc) {
2281 if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
2282 PMD_DRV_LOG(ERR, "link status not supported by host");
2287 /* Enable uio/vfio intr/eventfd mapping: althrough we already did that
2288 * in device configure, but it could be unmapped when device is
2291 if (dev->data->dev_conf.intr_conf.lsc ||
2292 dev->data->dev_conf.intr_conf.rxq) {
2293 virtio_intr_disable(dev);
2295 /* Setup interrupt callback */
2296 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2297 rte_intr_callback_register(dev->intr_handle,
2298 virtio_interrupt_handler,
2301 if (virtio_intr_enable(dev) < 0) {
2302 PMD_DRV_LOG(ERR, "interrupt enable failed");
2307 /*Notify the backend
2308 *Otherwise the tap backend might already stop its queue due to fullness.
2309 *vhost backend will have no chance to be waked up
2311 nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
2312 if (hw->max_queue_pairs > 1) {
2313 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
2317 PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
2319 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2320 vq = virtnet_rxq_to_vq(dev->data->rx_queues[i]);
2321 /* Flush the old packets */
2322 virtqueue_rxvq_flush(vq);
2323 virtqueue_notify(vq);
2326 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2327 vq = virtnet_txq_to_vq(dev->data->tx_queues[i]);
2328 virtqueue_notify(vq);
2331 PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
2333 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2334 vq = virtnet_rxq_to_vq(dev->data->rx_queues[i]);
2338 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2339 vq = virtnet_txq_to_vq(dev->data->tx_queues[i]);
2343 set_rxtx_funcs(dev);
2346 /* Initialize Link state */
2347 virtio_dev_link_update(dev, 0);
2352 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
2354 struct virtio_hw *hw = dev->data->dev_private;
2355 uint16_t nr_vq = virtio_get_nr_vq(hw);
2356 const char *type __rte_unused;
2357 unsigned int i, mbuf_num = 0;
2358 struct virtqueue *vq;
2359 struct rte_mbuf *buf;
2362 if (hw->vqs == NULL)
2365 for (i = 0; i < nr_vq; i++) {
2370 queue_type = virtio_get_queue_type(hw, i);
2371 if (queue_type == VTNET_RQ)
2373 else if (queue_type == VTNET_TQ)
2379 "Before freeing %s[%d] used and unused buf",
2383 while ((buf = virtqueue_detach_unused(vq)) != NULL) {
2384 rte_pktmbuf_free(buf);
2389 "After freeing %s[%d] used and unused buf",
2394 PMD_INIT_LOG(DEBUG, "%d mbufs freed", mbuf_num);
2398 * Stop device: disable interrupt and mark link down
2401 virtio_dev_stop(struct rte_eth_dev *dev)
2403 struct virtio_hw *hw = dev->data->dev_private;
2404 struct rte_eth_link link;
2405 struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
2407 PMD_INIT_LOG(DEBUG, "stop");
2408 dev->data->dev_started = 0;
2410 rte_spinlock_lock(&hw->state_lock);
2415 if (intr_conf->lsc || intr_conf->rxq) {
2416 virtio_intr_disable(dev);
2418 /* Reset interrupt callback */
2419 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
2420 rte_intr_callback_unregister(dev->intr_handle,
2421 virtio_interrupt_handler,
2426 memset(&link, 0, sizeof(link));
2427 rte_eth_linkstatus_set(dev, &link);
2429 rte_spinlock_unlock(&hw->state_lock);
2435 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
2437 struct rte_eth_link link;
2439 struct virtio_hw *hw = dev->data->dev_private;
2441 memset(&link, 0, sizeof(link));
2442 link.link_duplex = hw->duplex;
2443 link.link_speed = hw->speed;
2444 link.link_autoneg = ETH_LINK_AUTONEG;
2447 link.link_status = ETH_LINK_DOWN;
2448 link.link_speed = ETH_SPEED_NUM_NONE;
2449 } else if (virtio_with_feature(hw, VIRTIO_NET_F_STATUS)) {
2450 PMD_INIT_LOG(DEBUG, "Get link status from hw");
2451 virtio_read_dev_config(hw,
2452 offsetof(struct virtio_net_config, status),
2453 &status, sizeof(status));
2454 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
2455 link.link_status = ETH_LINK_DOWN;
2456 link.link_speed = ETH_SPEED_NUM_NONE;
2457 PMD_INIT_LOG(DEBUG, "Port %d is down",
2458 dev->data->port_id);
2460 link.link_status = ETH_LINK_UP;
2461 PMD_INIT_LOG(DEBUG, "Port %d is up",
2462 dev->data->port_id);
2465 link.link_status = ETH_LINK_UP;
2468 return rte_eth_linkstatus_set(dev, &link);
2472 virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
2474 const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2475 struct virtio_hw *hw = dev->data->dev_private;
2476 uint64_t offloads = rxmode->offloads;
2478 if (mask & ETH_VLAN_FILTER_MASK) {
2479 if ((offloads & DEV_RX_OFFLOAD_VLAN_FILTER) &&
2480 !virtio_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2483 "vlan filtering not available on this host");
2489 if (mask & ETH_VLAN_STRIP_MASK)
2490 hw->vlan_strip = !!(offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
2496 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2498 uint64_t tso_mask, host_features;
2499 struct virtio_hw *hw = dev->data->dev_private;
2500 dev_info->speed_capa = virtio_dev_speed_capa_get(hw->speed);
2502 dev_info->max_rx_queues =
2503 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
2504 dev_info->max_tx_queues =
2505 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
2506 dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
2507 dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
2508 dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
2509 dev_info->max_mtu = hw->max_mtu;
2511 host_features = VIRTIO_OPS(hw)->get_features(hw);
2512 dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
2513 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_JUMBO_FRAME;
2514 if (host_features & (1ULL << VIRTIO_NET_F_MRG_RXBUF))
2515 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_SCATTER;
2516 if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
2517 dev_info->rx_offload_capa |=
2518 DEV_RX_OFFLOAD_TCP_CKSUM |
2519 DEV_RX_OFFLOAD_UDP_CKSUM;
2521 if (host_features & (1ULL << VIRTIO_NET_F_CTRL_VLAN))
2522 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_VLAN_FILTER;
2523 tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2524 (1ULL << VIRTIO_NET_F_GUEST_TSO6);
2525 if ((host_features & tso_mask) == tso_mask)
2526 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
2528 dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
2529 DEV_TX_OFFLOAD_VLAN_INSERT;
2530 if (host_features & (1ULL << VIRTIO_NET_F_CSUM)) {
2531 dev_info->tx_offload_capa |=
2532 DEV_TX_OFFLOAD_UDP_CKSUM |
2533 DEV_TX_OFFLOAD_TCP_CKSUM;
2535 tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2536 (1ULL << VIRTIO_NET_F_HOST_TSO6);
2537 if ((host_features & tso_mask) == tso_mask)
2538 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
2540 if (host_features & (1ULL << VIRTIO_F_RING_PACKED)) {
2542 * According to 2.7 Packed Virtqueues,
2543 * 2.7.10.1 Structure Size and Alignment:
2544 * The Queue Size value does not have to be a power of 2.
2546 dev_info->rx_desc_lim.nb_max = UINT16_MAX;
2549 * According to 2.6 Split Virtqueues:
2550 * Queue Size value is always a power of 2. The maximum Queue
2551 * Size value is 32768.
2553 dev_info->rx_desc_lim.nb_max = 32768;
2556 * Actual minimum is not the same for virtqueues of different kinds,
2557 * but to avoid tangling the code with separate branches, rely on
2558 * default thresholds since desc number must be at least of their size.
2560 dev_info->rx_desc_lim.nb_min = RTE_MAX(DEFAULT_RX_FREE_THRESH,
2561 RTE_VIRTIO_VPMD_RX_REARM_THRESH);
2562 dev_info->rx_desc_lim.nb_align = 1;
2568 * It enables testpmd to collect per queue stats.
2571 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
2572 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
2573 __rte_unused uint8_t is_rx)
2578 RTE_LOG_REGISTER_SUFFIX(virtio_logtype_init, init, NOTICE);
2579 RTE_LOG_REGISTER_SUFFIX(virtio_logtype_driver, driver, NOTICE);