1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2016 Intel Corporation
11 #include <rte_ethdev_driver.h>
12 #include <rte_ethdev_pci.h>
13 #include <rte_memcpy.h>
14 #include <rte_string_fns.h>
15 #include <rte_memzone.h>
16 #include <rte_malloc.h>
17 #include <rte_branch_prediction.h>
19 #include <rte_bus_pci.h>
20 #include <rte_ether.h>
23 #include <rte_common.h>
24 #include <rte_errno.h>
25 #include <rte_cpuflags.h>
27 #include <rte_memory.h>
30 #include <rte_cycles.h>
31 #include <rte_kvargs.h>
33 #include "virtio_ethdev.h"
34 #include "virtio_pci.h"
35 #include "virtio_logs.h"
36 #include "virtqueue.h"
37 #include "virtio_rxtx.h"
38 #include "virtio_user/virtio_user_dev.h"
40 static int eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev);
41 static int virtio_dev_configure(struct rte_eth_dev *dev);
42 static int virtio_dev_start(struct rte_eth_dev *dev);
43 static void virtio_dev_stop(struct rte_eth_dev *dev);
44 static int virtio_dev_promiscuous_enable(struct rte_eth_dev *dev);
45 static int virtio_dev_promiscuous_disable(struct rte_eth_dev *dev);
46 static int virtio_dev_allmulticast_enable(struct rte_eth_dev *dev);
47 static int virtio_dev_allmulticast_disable(struct rte_eth_dev *dev);
48 static uint32_t virtio_dev_speed_capa_get(uint32_t speed);
49 static int virtio_dev_devargs_parse(struct rte_devargs *devargs,
53 static int virtio_dev_info_get(struct rte_eth_dev *dev,
54 struct rte_eth_dev_info *dev_info);
55 static int virtio_dev_link_update(struct rte_eth_dev *dev,
56 int wait_to_complete);
57 static int virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask);
59 static void virtio_set_hwaddr(struct virtio_hw *hw);
60 static void virtio_get_hwaddr(struct virtio_hw *hw);
62 static int virtio_dev_stats_get(struct rte_eth_dev *dev,
63 struct rte_eth_stats *stats);
64 static int virtio_dev_xstats_get(struct rte_eth_dev *dev,
65 struct rte_eth_xstat *xstats, unsigned n);
66 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
67 struct rte_eth_xstat_name *xstats_names,
69 static int virtio_dev_stats_reset(struct rte_eth_dev *dev);
70 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev);
71 static int virtio_vlan_filter_set(struct rte_eth_dev *dev,
72 uint16_t vlan_id, int on);
73 static int virtio_mac_addr_add(struct rte_eth_dev *dev,
74 struct rte_ether_addr *mac_addr,
75 uint32_t index, uint32_t vmdq);
76 static void virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index);
77 static int virtio_mac_addr_set(struct rte_eth_dev *dev,
78 struct rte_ether_addr *mac_addr);
80 static int virtio_intr_disable(struct rte_eth_dev *dev);
82 static int virtio_dev_queue_stats_mapping_set(
83 struct rte_eth_dev *eth_dev,
88 int virtio_logtype_init;
89 int virtio_logtype_driver;
91 static void virtio_notify_peers(struct rte_eth_dev *dev);
92 static void virtio_ack_link_announce(struct rte_eth_dev *dev);
95 * The set of PCI devices this driver supports
97 static const struct rte_pci_id pci_id_virtio_map[] = {
98 { RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_LEGACY_DEVICEID_NET) },
99 { RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_MODERN_DEVICEID_NET) },
100 { .vendor_id = 0, /* sentinel */ },
103 struct rte_virtio_xstats_name_off {
104 char name[RTE_ETH_XSTATS_NAME_SIZE];
108 /* [rt]x_qX_ is prepended to the name string here */
109 static const struct rte_virtio_xstats_name_off rte_virtio_rxq_stat_strings[] = {
110 {"good_packets", offsetof(struct virtnet_rx, stats.packets)},
111 {"good_bytes", offsetof(struct virtnet_rx, stats.bytes)},
112 {"errors", offsetof(struct virtnet_rx, stats.errors)},
113 {"multicast_packets", offsetof(struct virtnet_rx, stats.multicast)},
114 {"broadcast_packets", offsetof(struct virtnet_rx, stats.broadcast)},
115 {"undersize_packets", offsetof(struct virtnet_rx, stats.size_bins[0])},
116 {"size_64_packets", offsetof(struct virtnet_rx, stats.size_bins[1])},
117 {"size_65_127_packets", offsetof(struct virtnet_rx, stats.size_bins[2])},
118 {"size_128_255_packets", offsetof(struct virtnet_rx, stats.size_bins[3])},
119 {"size_256_511_packets", offsetof(struct virtnet_rx, stats.size_bins[4])},
120 {"size_512_1023_packets", offsetof(struct virtnet_rx, stats.size_bins[5])},
121 {"size_1024_1518_packets", offsetof(struct virtnet_rx, stats.size_bins[6])},
122 {"size_1519_max_packets", offsetof(struct virtnet_rx, stats.size_bins[7])},
125 /* [rt]x_qX_ is prepended to the name string here */
126 static const struct rte_virtio_xstats_name_off rte_virtio_txq_stat_strings[] = {
127 {"good_packets", offsetof(struct virtnet_tx, stats.packets)},
128 {"good_bytes", offsetof(struct virtnet_tx, stats.bytes)},
129 {"multicast_packets", offsetof(struct virtnet_tx, stats.multicast)},
130 {"broadcast_packets", offsetof(struct virtnet_tx, stats.broadcast)},
131 {"undersize_packets", offsetof(struct virtnet_tx, stats.size_bins[0])},
132 {"size_64_packets", offsetof(struct virtnet_tx, stats.size_bins[1])},
133 {"size_65_127_packets", offsetof(struct virtnet_tx, stats.size_bins[2])},
134 {"size_128_255_packets", offsetof(struct virtnet_tx, stats.size_bins[3])},
135 {"size_256_511_packets", offsetof(struct virtnet_tx, stats.size_bins[4])},
136 {"size_512_1023_packets", offsetof(struct virtnet_tx, stats.size_bins[5])},
137 {"size_1024_1518_packets", offsetof(struct virtnet_tx, stats.size_bins[6])},
138 {"size_1519_max_packets", offsetof(struct virtnet_tx, stats.size_bins[7])},
141 #define VIRTIO_NB_RXQ_XSTATS (sizeof(rte_virtio_rxq_stat_strings) / \
142 sizeof(rte_virtio_rxq_stat_strings[0]))
143 #define VIRTIO_NB_TXQ_XSTATS (sizeof(rte_virtio_txq_stat_strings) / \
144 sizeof(rte_virtio_txq_stat_strings[0]))
146 struct virtio_hw_internal virtio_hw_internal[RTE_MAX_ETHPORTS];
148 static struct virtio_pmd_ctrl *
149 virtio_send_command_packed(struct virtnet_ctl *cvq,
150 struct virtio_pmd_ctrl *ctrl,
151 int *dlen, int pkt_num)
153 struct virtqueue *vq = cvq->vq;
155 struct vring_packed_desc *desc = vq->vq_packed.ring.desc;
156 struct virtio_pmd_ctrl *result;
163 * Format is enforced in qemu code:
164 * One TX packet for header;
165 * At least one TX packet per argument;
166 * One RX packet for ACK.
168 head = vq->vq_avail_idx;
169 flags = vq->vq_packed.cached_flags;
170 desc[head].addr = cvq->virtio_net_hdr_mem;
171 desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
174 if (++vq->vq_avail_idx >= vq->vq_nentries) {
175 vq->vq_avail_idx -= vq->vq_nentries;
176 vq->vq_packed.cached_flags ^= VRING_PACKED_DESC_F_AVAIL_USED;
179 for (k = 0; k < pkt_num; k++) {
180 desc[vq->vq_avail_idx].addr = cvq->virtio_net_hdr_mem
181 + sizeof(struct virtio_net_ctrl_hdr)
182 + sizeof(ctrl->status) + sizeof(uint8_t) * sum;
183 desc[vq->vq_avail_idx].len = dlen[k];
184 desc[vq->vq_avail_idx].flags = VRING_DESC_F_NEXT |
185 vq->vq_packed.cached_flags;
189 if (++vq->vq_avail_idx >= vq->vq_nentries) {
190 vq->vq_avail_idx -= vq->vq_nentries;
191 vq->vq_packed.cached_flags ^=
192 VRING_PACKED_DESC_F_AVAIL_USED;
196 desc[vq->vq_avail_idx].addr = cvq->virtio_net_hdr_mem
197 + sizeof(struct virtio_net_ctrl_hdr);
198 desc[vq->vq_avail_idx].len = sizeof(ctrl->status);
199 desc[vq->vq_avail_idx].flags = VRING_DESC_F_WRITE |
200 vq->vq_packed.cached_flags;
203 if (++vq->vq_avail_idx >= vq->vq_nentries) {
204 vq->vq_avail_idx -= vq->vq_nentries;
205 vq->vq_packed.cached_flags ^= VRING_PACKED_DESC_F_AVAIL_USED;
208 virtio_wmb(vq->hw->weak_barriers);
209 desc[head].flags = VRING_DESC_F_NEXT | flags;
211 virtio_wmb(vq->hw->weak_barriers);
212 virtqueue_notify(vq);
214 /* wait for used descriptors in virtqueue */
215 while (!desc_is_used(&desc[head], vq))
218 virtio_rmb(vq->hw->weak_barriers);
220 /* now get used descriptors */
221 vq->vq_free_cnt += nb_descs;
222 vq->vq_used_cons_idx += nb_descs;
223 if (vq->vq_used_cons_idx >= vq->vq_nentries) {
224 vq->vq_used_cons_idx -= vq->vq_nentries;
225 vq->vq_packed.used_wrap_counter ^= 1;
228 PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\n"
229 "vq->vq_avail_idx=%d\n"
230 "vq->vq_used_cons_idx=%d\n"
231 "vq->vq_packed.cached_flags=0x%x\n"
232 "vq->vq_packed.used_wrap_counter=%d\n",
235 vq->vq_used_cons_idx,
236 vq->vq_packed.cached_flags,
237 vq->vq_packed.used_wrap_counter);
239 result = cvq->virtio_net_hdr_mz->addr;
243 static struct virtio_pmd_ctrl *
244 virtio_send_command_split(struct virtnet_ctl *cvq,
245 struct virtio_pmd_ctrl *ctrl,
246 int *dlen, int pkt_num)
248 struct virtio_pmd_ctrl *result;
249 struct virtqueue *vq = cvq->vq;
253 head = vq->vq_desc_head_idx;
256 * Format is enforced in qemu code:
257 * One TX packet for header;
258 * At least one TX packet per argument;
259 * One RX packet for ACK.
261 vq->vq_split.ring.desc[head].flags = VRING_DESC_F_NEXT;
262 vq->vq_split.ring.desc[head].addr = cvq->virtio_net_hdr_mem;
263 vq->vq_split.ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
265 i = vq->vq_split.ring.desc[head].next;
267 for (k = 0; k < pkt_num; k++) {
268 vq->vq_split.ring.desc[i].flags = VRING_DESC_F_NEXT;
269 vq->vq_split.ring.desc[i].addr = cvq->virtio_net_hdr_mem
270 + sizeof(struct virtio_net_ctrl_hdr)
271 + sizeof(ctrl->status) + sizeof(uint8_t)*sum;
272 vq->vq_split.ring.desc[i].len = dlen[k];
275 i = vq->vq_split.ring.desc[i].next;
278 vq->vq_split.ring.desc[i].flags = VRING_DESC_F_WRITE;
279 vq->vq_split.ring.desc[i].addr = cvq->virtio_net_hdr_mem
280 + sizeof(struct virtio_net_ctrl_hdr);
281 vq->vq_split.ring.desc[i].len = sizeof(ctrl->status);
284 vq->vq_desc_head_idx = vq->vq_split.ring.desc[i].next;
286 vq_update_avail_ring(vq, head);
287 vq_update_avail_idx(vq);
289 PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
291 virtqueue_notify(vq);
293 while (virtqueue_nused(vq) == 0)
296 while (virtqueue_nused(vq)) {
297 uint32_t idx, desc_idx, used_idx;
298 struct vring_used_elem *uep;
300 used_idx = (uint32_t)(vq->vq_used_cons_idx
301 & (vq->vq_nentries - 1));
302 uep = &vq->vq_split.ring.used->ring[used_idx];
303 idx = (uint32_t) uep->id;
306 while (vq->vq_split.ring.desc[desc_idx].flags &
308 desc_idx = vq->vq_split.ring.desc[desc_idx].next;
312 vq->vq_split.ring.desc[desc_idx].next = vq->vq_desc_head_idx;
313 vq->vq_desc_head_idx = idx;
315 vq->vq_used_cons_idx++;
319 PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
320 vq->vq_free_cnt, vq->vq_desc_head_idx);
322 result = cvq->virtio_net_hdr_mz->addr;
327 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
328 int *dlen, int pkt_num)
330 virtio_net_ctrl_ack status = ~0;
331 struct virtio_pmd_ctrl *result;
332 struct virtqueue *vq;
334 ctrl->status = status;
336 if (!cvq || !cvq->vq) {
337 PMD_INIT_LOG(ERR, "Control queue is not supported.");
341 rte_spinlock_lock(&cvq->lock);
344 PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
345 "vq->hw->cvq = %p vq = %p",
346 vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
348 if (vq->vq_free_cnt < pkt_num + 2 || pkt_num < 1) {
349 rte_spinlock_unlock(&cvq->lock);
353 memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
354 sizeof(struct virtio_pmd_ctrl));
356 if (vtpci_packed_queue(vq->hw))
357 result = virtio_send_command_packed(cvq, ctrl, dlen, pkt_num);
359 result = virtio_send_command_split(cvq, ctrl, dlen, pkt_num);
361 rte_spinlock_unlock(&cvq->lock);
362 return result->status;
366 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
368 struct virtio_hw *hw = dev->data->dev_private;
369 struct virtio_pmd_ctrl ctrl;
373 ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
374 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
375 memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
377 dlen[0] = sizeof(uint16_t);
379 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
381 PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
382 "failed, this is too late now...");
390 virtio_dev_queue_release(void *queue __rte_unused)
396 virtio_get_nr_vq(struct virtio_hw *hw)
398 uint16_t nr_vq = hw->max_queue_pairs * 2;
400 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
407 virtio_init_vring(struct virtqueue *vq)
409 int size = vq->vq_nentries;
410 uint8_t *ring_mem = vq->vq_ring_virt_mem;
412 PMD_INIT_FUNC_TRACE();
414 memset(ring_mem, 0, vq->vq_ring_size);
416 vq->vq_used_cons_idx = 0;
417 vq->vq_desc_head_idx = 0;
418 vq->vq_avail_idx = 0;
419 vq->vq_desc_tail_idx = (uint16_t)(vq->vq_nentries - 1);
420 vq->vq_free_cnt = vq->vq_nentries;
421 memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries);
422 if (vtpci_packed_queue(vq->hw)) {
423 vring_init_packed(&vq->vq_packed.ring, ring_mem,
424 VIRTIO_PCI_VRING_ALIGN, size);
425 vring_desc_init_packed(vq, size);
427 struct vring *vr = &vq->vq_split.ring;
429 vring_init_split(vr, ring_mem, VIRTIO_PCI_VRING_ALIGN, size);
430 vring_desc_init_split(vr->desc, size);
433 * Disable device(host) interrupting guest
435 virtqueue_disable_intr(vq);
439 virtio_init_queue(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx)
441 char vq_name[VIRTQUEUE_MAX_NAME_SZ];
442 char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
443 const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
444 unsigned int vq_size, size;
445 struct virtio_hw *hw = dev->data->dev_private;
446 struct virtnet_rx *rxvq = NULL;
447 struct virtnet_tx *txvq = NULL;
448 struct virtnet_ctl *cvq = NULL;
449 struct virtqueue *vq;
450 size_t sz_hdr_mz = 0;
451 void *sw_ring = NULL;
452 int queue_type = virtio_get_queue_type(hw, vtpci_queue_idx);
454 int numa_node = dev->device->numa_node;
456 PMD_INIT_LOG(INFO, "setting up queue: %u on NUMA node %d",
457 vtpci_queue_idx, numa_node);
460 * Read the virtqueue size from the Queue Size field
461 * Always power of 2 and if 0 virtqueue does not exist
463 vq_size = VTPCI_OPS(hw)->get_queue_num(hw, vtpci_queue_idx);
464 PMD_INIT_LOG(DEBUG, "vq_size: %u", vq_size);
466 PMD_INIT_LOG(ERR, "virtqueue does not exist");
470 if (!vtpci_packed_queue(hw) && !rte_is_power_of_2(vq_size)) {
471 PMD_INIT_LOG(ERR, "split virtqueue size is not power of 2");
475 snprintf(vq_name, sizeof(vq_name), "port%d_vq%d",
476 dev->data->port_id, vtpci_queue_idx);
478 size = RTE_ALIGN_CEIL(sizeof(*vq) +
479 vq_size * sizeof(struct vq_desc_extra),
480 RTE_CACHE_LINE_SIZE);
481 if (queue_type == VTNET_TQ) {
483 * For each xmit packet, allocate a virtio_net_hdr
484 * and indirect ring elements
486 sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
487 } else if (queue_type == VTNET_CQ) {
488 /* Allocate a page for control vq command, data and status */
489 sz_hdr_mz = PAGE_SIZE;
492 vq = rte_zmalloc_socket(vq_name, size, RTE_CACHE_LINE_SIZE,
495 PMD_INIT_LOG(ERR, "can not allocate vq");
498 hw->vqs[vtpci_queue_idx] = vq;
501 vq->vq_queue_index = vtpci_queue_idx;
502 vq->vq_nentries = vq_size;
503 if (vtpci_packed_queue(hw)) {
504 vq->vq_packed.used_wrap_counter = 1;
505 vq->vq_packed.cached_flags = VRING_PACKED_DESC_F_AVAIL;
506 vq->vq_packed.event_flags_shadow = 0;
507 if (queue_type == VTNET_RQ)
508 vq->vq_packed.cached_flags |= VRING_DESC_F_WRITE;
512 * Reserve a memzone for vring elements
514 size = vring_size(hw, vq_size, VIRTIO_PCI_VRING_ALIGN);
515 vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
516 PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
517 size, vq->vq_ring_size);
519 mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
520 numa_node, RTE_MEMZONE_IOVA_CONTIG,
521 VIRTIO_PCI_VRING_ALIGN);
523 if (rte_errno == EEXIST)
524 mz = rte_memzone_lookup(vq_name);
531 memset(mz->addr, 0, mz->len);
533 vq->vq_ring_mem = mz->iova;
534 vq->vq_ring_virt_mem = mz->addr;
535 PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem: 0x%" PRIx64,
537 PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64,
538 (uint64_t)(uintptr_t)mz->addr);
540 virtio_init_vring(vq);
543 snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_vq%d_hdr",
544 dev->data->port_id, vtpci_queue_idx);
545 hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
546 numa_node, RTE_MEMZONE_IOVA_CONTIG,
547 RTE_CACHE_LINE_SIZE);
548 if (hdr_mz == NULL) {
549 if (rte_errno == EEXIST)
550 hdr_mz = rte_memzone_lookup(vq_hdr_name);
551 if (hdr_mz == NULL) {
558 if (queue_type == VTNET_RQ) {
559 size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
560 sizeof(vq->sw_ring[0]);
562 sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
563 RTE_CACHE_LINE_SIZE, numa_node);
565 PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
570 vq->sw_ring = sw_ring;
573 rxvq->port_id = dev->data->port_id;
575 } else if (queue_type == VTNET_TQ) {
578 txvq->port_id = dev->data->port_id;
580 txvq->virtio_net_hdr_mz = hdr_mz;
581 txvq->virtio_net_hdr_mem = hdr_mz->iova;
582 } else if (queue_type == VTNET_CQ) {
586 cvq->virtio_net_hdr_mz = hdr_mz;
587 cvq->virtio_net_hdr_mem = hdr_mz->iova;
588 memset(cvq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
593 /* For virtio_user case (that is when hw->virtio_user_dev is not NULL),
594 * we use virtual address. And we need properly set _offset_, please see
595 * VIRTIO_MBUF_DATA_DMA_ADDR in virtqueue.h for more information.
597 if (!hw->virtio_user_dev)
598 vq->offset = offsetof(struct rte_mbuf, buf_iova);
600 vq->vq_ring_mem = (uintptr_t)mz->addr;
601 vq->offset = offsetof(struct rte_mbuf, buf_addr);
602 if (queue_type == VTNET_TQ)
603 txvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
604 else if (queue_type == VTNET_CQ)
605 cvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
608 if (queue_type == VTNET_TQ) {
609 struct virtio_tx_region *txr;
613 memset(txr, 0, vq_size * sizeof(*txr));
614 for (i = 0; i < vq_size; i++) {
615 struct vring_desc *start_dp = txr[i].tx_indir;
617 /* first indirect descriptor is always the tx header */
618 if (!vtpci_packed_queue(hw)) {
619 vring_desc_init_split(start_dp,
620 RTE_DIM(txr[i].tx_indir));
621 start_dp->addr = txvq->virtio_net_hdr_mem
623 + offsetof(struct virtio_tx_region,
625 start_dp->len = hw->vtnet_hdr_size;
626 start_dp->flags = VRING_DESC_F_NEXT;
631 if (VTPCI_OPS(hw)->setup_queue(hw, vq) < 0) {
632 PMD_INIT_LOG(ERR, "setup_queue failed");
640 rte_memzone_free(hdr_mz);
641 rte_memzone_free(mz);
648 virtio_free_queues(struct virtio_hw *hw)
650 uint16_t nr_vq = virtio_get_nr_vq(hw);
651 struct virtqueue *vq;
658 for (i = 0; i < nr_vq; i++) {
663 queue_type = virtio_get_queue_type(hw, i);
664 if (queue_type == VTNET_RQ) {
665 rte_free(vq->sw_ring);
666 rte_memzone_free(vq->rxq.mz);
667 } else if (queue_type == VTNET_TQ) {
668 rte_memzone_free(vq->txq.mz);
669 rte_memzone_free(vq->txq.virtio_net_hdr_mz);
671 rte_memzone_free(vq->cq.mz);
672 rte_memzone_free(vq->cq.virtio_net_hdr_mz);
684 virtio_alloc_queues(struct rte_eth_dev *dev)
686 struct virtio_hw *hw = dev->data->dev_private;
687 uint16_t nr_vq = virtio_get_nr_vq(hw);
691 hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
693 PMD_INIT_LOG(ERR, "failed to allocate vqs");
697 for (i = 0; i < nr_vq; i++) {
698 ret = virtio_init_queue(dev, i);
700 virtio_free_queues(hw);
708 static void virtio_queues_unbind_intr(struct rte_eth_dev *dev);
711 virtio_dev_close(struct rte_eth_dev *dev)
713 struct virtio_hw *hw = dev->data->dev_private;
714 struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
716 PMD_INIT_LOG(DEBUG, "virtio_dev_close");
723 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
724 VTPCI_OPS(hw)->set_config_irq(hw, VIRTIO_MSI_NO_VECTOR);
726 virtio_queues_unbind_intr(dev);
728 if (intr_conf->lsc || intr_conf->rxq) {
729 virtio_intr_disable(dev);
730 rte_intr_efd_disable(dev->intr_handle);
731 rte_free(dev->intr_handle->intr_vec);
732 dev->intr_handle->intr_vec = NULL;
736 virtio_dev_free_mbufs(dev);
737 virtio_free_queues(hw);
739 #ifdef RTE_VIRTIO_USER
740 if (hw->virtio_user_dev)
741 virtio_user_dev_uninit(hw->virtio_user_dev);
745 rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(dev));
747 rte_pci_ioport_unmap(VTPCI_IO(hw));
752 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
754 struct virtio_hw *hw = dev->data->dev_private;
755 struct virtio_pmd_ctrl ctrl;
759 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
760 PMD_INIT_LOG(INFO, "host does not support rx control");
764 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
765 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
769 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
771 PMD_INIT_LOG(ERR, "Failed to enable promisc");
779 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
781 struct virtio_hw *hw = dev->data->dev_private;
782 struct virtio_pmd_ctrl ctrl;
786 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
787 PMD_INIT_LOG(INFO, "host does not support rx control");
791 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
792 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
796 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
798 PMD_INIT_LOG(ERR, "Failed to disable promisc");
806 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
808 struct virtio_hw *hw = dev->data->dev_private;
809 struct virtio_pmd_ctrl ctrl;
813 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
814 PMD_INIT_LOG(INFO, "host does not support rx control");
818 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
819 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
823 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
825 PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
833 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
835 struct virtio_hw *hw = dev->data->dev_private;
836 struct virtio_pmd_ctrl ctrl;
840 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
841 PMD_INIT_LOG(INFO, "host does not support rx control");
845 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
846 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
850 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
852 PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
859 #define VLAN_TAG_LEN 4 /* 802.3ac tag (not DMA'd) */
861 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
863 struct virtio_hw *hw = dev->data->dev_private;
864 uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
866 uint32_t frame_size = mtu + ether_hdr_len;
867 uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
869 max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
871 if (mtu < RTE_ETHER_MIN_MTU || frame_size > max_frame_size) {
872 PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
873 RTE_ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
880 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
882 struct virtio_hw *hw = dev->data->dev_private;
883 struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
884 struct virtqueue *vq = rxvq->vq;
886 virtqueue_enable_intr(vq);
887 virtio_mb(hw->weak_barriers);
892 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
894 struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
895 struct virtqueue *vq = rxvq->vq;
897 virtqueue_disable_intr(vq);
902 * dev_ops for virtio, bare necessities for basic operation
904 static const struct eth_dev_ops virtio_eth_dev_ops = {
905 .dev_configure = virtio_dev_configure,
906 .dev_start = virtio_dev_start,
907 .dev_stop = virtio_dev_stop,
908 .dev_close = virtio_dev_close,
909 .promiscuous_enable = virtio_dev_promiscuous_enable,
910 .promiscuous_disable = virtio_dev_promiscuous_disable,
911 .allmulticast_enable = virtio_dev_allmulticast_enable,
912 .allmulticast_disable = virtio_dev_allmulticast_disable,
913 .mtu_set = virtio_mtu_set,
914 .dev_infos_get = virtio_dev_info_get,
915 .stats_get = virtio_dev_stats_get,
916 .xstats_get = virtio_dev_xstats_get,
917 .xstats_get_names = virtio_dev_xstats_get_names,
918 .stats_reset = virtio_dev_stats_reset,
919 .xstats_reset = virtio_dev_stats_reset,
920 .link_update = virtio_dev_link_update,
921 .vlan_offload_set = virtio_dev_vlan_offload_set,
922 .rx_queue_setup = virtio_dev_rx_queue_setup,
923 .rx_queue_intr_enable = virtio_dev_rx_queue_intr_enable,
924 .rx_queue_intr_disable = virtio_dev_rx_queue_intr_disable,
925 .rx_queue_release = virtio_dev_queue_release,
926 .rx_descriptor_done = virtio_dev_rx_queue_done,
927 .tx_queue_setup = virtio_dev_tx_queue_setup,
928 .tx_queue_release = virtio_dev_queue_release,
929 /* collect stats per queue */
930 .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
931 .vlan_filter_set = virtio_vlan_filter_set,
932 .mac_addr_add = virtio_mac_addr_add,
933 .mac_addr_remove = virtio_mac_addr_remove,
934 .mac_addr_set = virtio_mac_addr_set,
938 * dev_ops for virtio-user in secondary processes, as we just have
939 * some limited supports currently.
941 const struct eth_dev_ops virtio_user_secondary_eth_dev_ops = {
942 .dev_infos_get = virtio_dev_info_get,
943 .stats_get = virtio_dev_stats_get,
944 .xstats_get = virtio_dev_xstats_get,
945 .xstats_get_names = virtio_dev_xstats_get_names,
946 .stats_reset = virtio_dev_stats_reset,
947 .xstats_reset = virtio_dev_stats_reset,
948 /* collect stats per queue */
949 .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
953 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
957 for (i = 0; i < dev->data->nb_tx_queues; i++) {
958 const struct virtnet_tx *txvq = dev->data->tx_queues[i];
962 stats->opackets += txvq->stats.packets;
963 stats->obytes += txvq->stats.bytes;
965 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
966 stats->q_opackets[i] = txvq->stats.packets;
967 stats->q_obytes[i] = txvq->stats.bytes;
971 for (i = 0; i < dev->data->nb_rx_queues; i++) {
972 const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
976 stats->ipackets += rxvq->stats.packets;
977 stats->ibytes += rxvq->stats.bytes;
978 stats->ierrors += rxvq->stats.errors;
980 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
981 stats->q_ipackets[i] = rxvq->stats.packets;
982 stats->q_ibytes[i] = rxvq->stats.bytes;
986 stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
989 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
990 struct rte_eth_xstat_name *xstats_names,
991 __rte_unused unsigned limit)
997 unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
998 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
1000 if (xstats_names != NULL) {
1001 /* Note: limit checked in rte_eth_xstats_names() */
1003 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1004 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1007 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1008 snprintf(xstats_names[count].name,
1009 sizeof(xstats_names[count].name),
1011 rte_virtio_rxq_stat_strings[t].name);
1016 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1017 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1020 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1021 snprintf(xstats_names[count].name,
1022 sizeof(xstats_names[count].name),
1024 rte_virtio_txq_stat_strings[t].name);
1034 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
1040 unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
1041 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
1046 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1047 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1054 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1055 xstats[count].value = *(uint64_t *)(((char *)rxvq) +
1056 rte_virtio_rxq_stat_strings[t].offset);
1057 xstats[count].id = count;
1062 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1063 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1070 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1071 xstats[count].value = *(uint64_t *)(((char *)txvq) +
1072 rte_virtio_txq_stat_strings[t].offset);
1073 xstats[count].id = count;
1082 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1084 virtio_update_stats(dev, stats);
1090 virtio_dev_stats_reset(struct rte_eth_dev *dev)
1094 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1095 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1099 txvq->stats.packets = 0;
1100 txvq->stats.bytes = 0;
1101 txvq->stats.multicast = 0;
1102 txvq->stats.broadcast = 0;
1103 memset(txvq->stats.size_bins, 0,
1104 sizeof(txvq->stats.size_bins[0]) * 8);
1107 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1108 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1112 rxvq->stats.packets = 0;
1113 rxvq->stats.bytes = 0;
1114 rxvq->stats.errors = 0;
1115 rxvq->stats.multicast = 0;
1116 rxvq->stats.broadcast = 0;
1117 memset(rxvq->stats.size_bins, 0,
1118 sizeof(rxvq->stats.size_bins[0]) * 8);
1125 virtio_set_hwaddr(struct virtio_hw *hw)
1127 vtpci_write_dev_config(hw,
1128 offsetof(struct virtio_net_config, mac),
1129 &hw->mac_addr, RTE_ETHER_ADDR_LEN);
1133 virtio_get_hwaddr(struct virtio_hw *hw)
1135 if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
1136 vtpci_read_dev_config(hw,
1137 offsetof(struct virtio_net_config, mac),
1138 &hw->mac_addr, RTE_ETHER_ADDR_LEN);
1140 rte_eth_random_addr(&hw->mac_addr[0]);
1141 virtio_set_hwaddr(hw);
1146 virtio_mac_table_set(struct virtio_hw *hw,
1147 const struct virtio_net_ctrl_mac *uc,
1148 const struct virtio_net_ctrl_mac *mc)
1150 struct virtio_pmd_ctrl ctrl;
1153 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1154 PMD_DRV_LOG(INFO, "host does not support mac table");
1158 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1159 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1161 len[0] = uc->entries * RTE_ETHER_ADDR_LEN + sizeof(uc->entries);
1162 memcpy(ctrl.data, uc, len[0]);
1164 len[1] = mc->entries * RTE_ETHER_ADDR_LEN + sizeof(mc->entries);
1165 memcpy(ctrl.data + len[0], mc, len[1]);
1167 err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1169 PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1174 virtio_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
1175 uint32_t index, uint32_t vmdq __rte_unused)
1177 struct virtio_hw *hw = dev->data->dev_private;
1178 const struct rte_ether_addr *addrs = dev->data->mac_addrs;
1180 struct virtio_net_ctrl_mac *uc, *mc;
1182 if (index >= VIRTIO_MAX_MAC_ADDRS) {
1183 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1187 uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1188 sizeof(uc->entries));
1190 mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1191 sizeof(mc->entries));
1194 for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1195 const struct rte_ether_addr *addr
1196 = (i == index) ? mac_addr : addrs + i;
1197 struct virtio_net_ctrl_mac *tbl
1198 = rte_is_multicast_ether_addr(addr) ? mc : uc;
1200 memcpy(&tbl->macs[tbl->entries++], addr, RTE_ETHER_ADDR_LEN);
1203 return virtio_mac_table_set(hw, uc, mc);
1207 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1209 struct virtio_hw *hw = dev->data->dev_private;
1210 struct rte_ether_addr *addrs = dev->data->mac_addrs;
1211 struct virtio_net_ctrl_mac *uc, *mc;
1214 if (index >= VIRTIO_MAX_MAC_ADDRS) {
1215 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1219 uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1220 sizeof(uc->entries));
1222 mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1223 sizeof(mc->entries));
1226 for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1227 struct virtio_net_ctrl_mac *tbl;
1229 if (i == index || rte_is_zero_ether_addr(addrs + i))
1232 tbl = rte_is_multicast_ether_addr(addrs + i) ? mc : uc;
1233 memcpy(&tbl->macs[tbl->entries++], addrs + i,
1234 RTE_ETHER_ADDR_LEN);
1237 virtio_mac_table_set(hw, uc, mc);
1241 virtio_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1243 struct virtio_hw *hw = dev->data->dev_private;
1245 memcpy(hw->mac_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1247 /* Use atomic update if available */
1248 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1249 struct virtio_pmd_ctrl ctrl;
1250 int len = RTE_ETHER_ADDR_LEN;
1252 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1253 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1255 memcpy(ctrl.data, mac_addr, RTE_ETHER_ADDR_LEN);
1256 return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1259 if (!vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1262 virtio_set_hwaddr(hw);
1267 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1269 struct virtio_hw *hw = dev->data->dev_private;
1270 struct virtio_pmd_ctrl ctrl;
1273 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1276 ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1277 ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1278 memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1279 len = sizeof(vlan_id);
1281 return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1285 virtio_intr_unmask(struct rte_eth_dev *dev)
1287 struct virtio_hw *hw = dev->data->dev_private;
1289 if (rte_intr_ack(dev->intr_handle) < 0)
1292 if (!hw->virtio_user_dev)
1293 hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1299 virtio_intr_enable(struct rte_eth_dev *dev)
1301 struct virtio_hw *hw = dev->data->dev_private;
1303 if (rte_intr_enable(dev->intr_handle) < 0)
1306 if (!hw->virtio_user_dev)
1307 hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1313 virtio_intr_disable(struct rte_eth_dev *dev)
1315 struct virtio_hw *hw = dev->data->dev_private;
1317 if (rte_intr_disable(dev->intr_handle) < 0)
1320 if (!hw->virtio_user_dev)
1321 hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1327 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1329 uint64_t host_features;
1331 /* Prepare guest_features: feature that driver wants to support */
1332 PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1335 /* Read device(host) feature bits */
1336 host_features = VTPCI_OPS(hw)->get_features(hw);
1337 PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1340 /* If supported, ensure MTU value is valid before acknowledging it. */
1341 if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1342 struct virtio_net_config config;
1344 vtpci_read_dev_config(hw,
1345 offsetof(struct virtio_net_config, mtu),
1346 &config.mtu, sizeof(config.mtu));
1348 if (config.mtu < RTE_ETHER_MIN_MTU)
1349 req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1353 * Negotiate features: Subset of device feature bits are written back
1354 * guest feature bits.
1356 hw->guest_features = req_features;
1357 hw->guest_features = vtpci_negotiate_features(hw, host_features);
1358 PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1359 hw->guest_features);
1362 if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1364 "VIRTIO_F_VERSION_1 features is not enabled.");
1367 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1368 if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1370 "failed to set FEATURES_OK status!");
1375 hw->req_guest_features = req_features;
1381 virtio_dev_pause(struct rte_eth_dev *dev)
1383 struct virtio_hw *hw = dev->data->dev_private;
1385 rte_spinlock_lock(&hw->state_lock);
1387 if (hw->started == 0) {
1388 /* Device is just stopped. */
1389 rte_spinlock_unlock(&hw->state_lock);
1394 * Prevent the worker threads from touching queues to avoid contention,
1395 * 1 ms should be enough for the ongoing Tx function to finish.
1402 * Recover hw state to let the worker threads continue.
1405 virtio_dev_resume(struct rte_eth_dev *dev)
1407 struct virtio_hw *hw = dev->data->dev_private;
1410 rte_spinlock_unlock(&hw->state_lock);
1414 * Should be called only after device is paused.
1417 virtio_inject_pkts(struct rte_eth_dev *dev, struct rte_mbuf **tx_pkts,
1420 struct virtio_hw *hw = dev->data->dev_private;
1421 struct virtnet_tx *txvq = dev->data->tx_queues[0];
1424 hw->inject_pkts = tx_pkts;
1425 ret = dev->tx_pkt_burst(txvq, tx_pkts, nb_pkts);
1426 hw->inject_pkts = NULL;
1432 virtio_notify_peers(struct rte_eth_dev *dev)
1434 struct virtio_hw *hw = dev->data->dev_private;
1435 struct virtnet_rx *rxvq;
1436 struct rte_mbuf *rarp_mbuf;
1438 if (!dev->data->rx_queues)
1441 rxvq = dev->data->rx_queues[0];
1445 rarp_mbuf = rte_net_make_rarp_packet(rxvq->mpool,
1446 (struct rte_ether_addr *)hw->mac_addr);
1447 if (rarp_mbuf == NULL) {
1448 PMD_DRV_LOG(ERR, "failed to make RARP packet.");
1452 /* If virtio port just stopped, no need to send RARP */
1453 if (virtio_dev_pause(dev) < 0) {
1454 rte_pktmbuf_free(rarp_mbuf);
1458 virtio_inject_pkts(dev, &rarp_mbuf, 1);
1459 virtio_dev_resume(dev);
1463 virtio_ack_link_announce(struct rte_eth_dev *dev)
1465 struct virtio_hw *hw = dev->data->dev_private;
1466 struct virtio_pmd_ctrl ctrl;
1468 ctrl.hdr.class = VIRTIO_NET_CTRL_ANNOUNCE;
1469 ctrl.hdr.cmd = VIRTIO_NET_CTRL_ANNOUNCE_ACK;
1471 virtio_send_command(hw->cvq, &ctrl, NULL, 0);
1475 * Process virtio config changed interrupt. Call the callback
1476 * if link state changed, generate gratuitous RARP packet if
1477 * the status indicates an ANNOUNCE.
1480 virtio_interrupt_handler(void *param)
1482 struct rte_eth_dev *dev = param;
1483 struct virtio_hw *hw = dev->data->dev_private;
1487 /* Read interrupt status which clears interrupt */
1488 isr = vtpci_isr(hw);
1489 PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1491 if (virtio_intr_unmask(dev) < 0)
1492 PMD_DRV_LOG(ERR, "interrupt enable failed");
1494 if (isr & VIRTIO_PCI_ISR_CONFIG) {
1495 if (virtio_dev_link_update(dev, 0) == 0)
1496 _rte_eth_dev_callback_process(dev,
1497 RTE_ETH_EVENT_INTR_LSC,
1500 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1501 vtpci_read_dev_config(hw,
1502 offsetof(struct virtio_net_config, status),
1503 &status, sizeof(status));
1504 if (status & VIRTIO_NET_S_ANNOUNCE) {
1505 virtio_notify_peers(dev);
1507 virtio_ack_link_announce(dev);
1513 /* set rx and tx handlers according to what is supported */
1515 set_rxtx_funcs(struct rte_eth_dev *eth_dev)
1517 struct virtio_hw *hw = eth_dev->data->dev_private;
1519 eth_dev->tx_pkt_prepare = virtio_xmit_pkts_prepare;
1520 if (vtpci_packed_queue(hw)) {
1522 "virtio: using packed ring %s Tx path on port %u",
1523 hw->use_vec_tx ? "vectorized" : "standard",
1524 eth_dev->data->port_id);
1526 eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed_vec;
1528 eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed;
1530 if (hw->use_inorder_tx) {
1531 PMD_INIT_LOG(INFO, "virtio: using inorder Tx path on port %u",
1532 eth_dev->data->port_id);
1533 eth_dev->tx_pkt_burst = virtio_xmit_pkts_inorder;
1535 PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
1536 eth_dev->data->port_id);
1537 eth_dev->tx_pkt_burst = virtio_xmit_pkts;
1541 if (vtpci_packed_queue(hw)) {
1542 if (hw->use_vec_rx) {
1544 "virtio: using packed ring vectorized Rx path on port %u",
1545 eth_dev->data->port_id);
1546 eth_dev->rx_pkt_burst =
1547 &virtio_recv_pkts_packed_vec;
1548 } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1550 "virtio: using packed ring mergeable buffer Rx path on port %u",
1551 eth_dev->data->port_id);
1552 eth_dev->rx_pkt_burst =
1553 &virtio_recv_mergeable_pkts_packed;
1556 "virtio: using packed ring standard Rx path on port %u",
1557 eth_dev->data->port_id);
1558 eth_dev->rx_pkt_burst = &virtio_recv_pkts_packed;
1561 if (hw->use_vec_rx) {
1562 PMD_INIT_LOG(INFO, "virtio: using vectorized Rx path on port %u",
1563 eth_dev->data->port_id);
1564 eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1565 } else if (hw->use_inorder_rx) {
1567 "virtio: using inorder Rx path on port %u",
1568 eth_dev->data->port_id);
1569 eth_dev->rx_pkt_burst = &virtio_recv_pkts_inorder;
1570 } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1572 "virtio: using mergeable buffer Rx path on port %u",
1573 eth_dev->data->port_id);
1574 eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1576 PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
1577 eth_dev->data->port_id);
1578 eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1584 /* Only support 1:1 queue/interrupt mapping so far.
1585 * TODO: support n:1 queue/interrupt mapping when there are limited number of
1586 * interrupt vectors (<N+1).
1589 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1592 struct virtio_hw *hw = dev->data->dev_private;
1594 PMD_INIT_LOG(INFO, "queue/interrupt binding");
1595 for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1596 dev->intr_handle->intr_vec[i] = i + 1;
1597 if (VTPCI_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1598 VIRTIO_MSI_NO_VECTOR) {
1599 PMD_DRV_LOG(ERR, "failed to set queue vector");
1608 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1611 struct virtio_hw *hw = dev->data->dev_private;
1613 PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1614 for (i = 0; i < dev->data->nb_rx_queues; ++i)
1615 VTPCI_OPS(hw)->set_queue_irq(hw,
1616 hw->vqs[i * VTNET_CQ],
1617 VIRTIO_MSI_NO_VECTOR);
1621 virtio_configure_intr(struct rte_eth_dev *dev)
1623 struct virtio_hw *hw = dev->data->dev_private;
1625 if (!rte_intr_cap_multiple(dev->intr_handle)) {
1626 PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1630 if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1631 PMD_INIT_LOG(ERR, "Fail to create eventfd");
1635 if (!dev->intr_handle->intr_vec) {
1636 dev->intr_handle->intr_vec =
1637 rte_zmalloc("intr_vec",
1638 hw->max_queue_pairs * sizeof(int), 0);
1639 if (!dev->intr_handle->intr_vec) {
1640 PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1641 hw->max_queue_pairs);
1646 /* Re-register callback to update max_intr */
1647 rte_intr_callback_unregister(dev->intr_handle,
1648 virtio_interrupt_handler,
1650 rte_intr_callback_register(dev->intr_handle,
1651 virtio_interrupt_handler,
1654 /* DO NOT try to remove this! This function will enable msix, or QEMU
1655 * will encounter SIGSEGV when DRIVER_OK is sent.
1656 * And for legacy devices, this should be done before queue/vec binding
1657 * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1658 * (22) will be ignored.
1660 if (virtio_intr_enable(dev) < 0) {
1661 PMD_DRV_LOG(ERR, "interrupt enable failed");
1665 if (virtio_queues_bind_intr(dev) < 0) {
1666 PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1672 #define SPEED_UNKNOWN 0xffffffff
1673 #define DUPLEX_UNKNOWN 0xff
1674 /* reset device and renegotiate features if needed */
1676 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1678 struct virtio_hw *hw = eth_dev->data->dev_private;
1679 struct virtio_net_config *config;
1680 struct virtio_net_config local_config;
1681 struct rte_pci_device *pci_dev = NULL;
1684 /* Reset the device although not necessary at startup */
1688 virtio_dev_free_mbufs(eth_dev);
1689 virtio_free_queues(hw);
1692 /* Tell the host we've noticed this device. */
1693 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1695 /* Tell the host we've known how to drive the device. */
1696 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1697 if (virtio_negotiate_features(hw, req_features) < 0)
1700 hw->weak_barriers = !vtpci_with_feature(hw, VIRTIO_F_ORDER_PLATFORM);
1702 if (!hw->virtio_user_dev)
1703 pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1705 /* If host does not support both status and MSI-X then disable LSC */
1706 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS) &&
1707 hw->use_msix != VIRTIO_MSIX_NONE)
1708 eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1710 eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1712 /* Setting up rx_header size for the device */
1713 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1714 vtpci_with_feature(hw, VIRTIO_F_VERSION_1) ||
1715 vtpci_with_feature(hw, VIRTIO_F_RING_PACKED))
1716 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1718 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1720 /* Copy the permanent MAC address to: virtio_hw */
1721 virtio_get_hwaddr(hw);
1722 rte_ether_addr_copy((struct rte_ether_addr *)hw->mac_addr,
1723 ð_dev->data->mac_addrs[0]);
1725 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1726 hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1727 hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1729 if (hw->speed == SPEED_UNKNOWN) {
1730 if (vtpci_with_feature(hw, VIRTIO_NET_F_SPEED_DUPLEX)) {
1731 config = &local_config;
1732 vtpci_read_dev_config(hw,
1733 offsetof(struct virtio_net_config, speed),
1734 &config->speed, sizeof(config->speed));
1735 vtpci_read_dev_config(hw,
1736 offsetof(struct virtio_net_config, duplex),
1737 &config->duplex, sizeof(config->duplex));
1738 hw->speed = config->speed;
1739 hw->duplex = config->duplex;
1742 if (hw->speed == SPEED_UNKNOWN)
1743 hw->speed = ETH_SPEED_NUM_10G;
1744 if (hw->duplex == DUPLEX_UNKNOWN)
1745 hw->duplex = ETH_LINK_FULL_DUPLEX;
1746 PMD_INIT_LOG(DEBUG, "link speed = %d, duplex = %d",
1747 hw->speed, hw->duplex);
1748 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1749 config = &local_config;
1751 vtpci_read_dev_config(hw,
1752 offsetof(struct virtio_net_config, mac),
1753 &config->mac, sizeof(config->mac));
1755 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1756 vtpci_read_dev_config(hw,
1757 offsetof(struct virtio_net_config, status),
1758 &config->status, sizeof(config->status));
1761 "VIRTIO_NET_F_STATUS is not supported");
1765 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1766 vtpci_read_dev_config(hw,
1767 offsetof(struct virtio_net_config, max_virtqueue_pairs),
1768 &config->max_virtqueue_pairs,
1769 sizeof(config->max_virtqueue_pairs));
1772 "VIRTIO_NET_F_MQ is not supported");
1773 config->max_virtqueue_pairs = 1;
1776 hw->max_queue_pairs = config->max_virtqueue_pairs;
1778 if (vtpci_with_feature(hw, VIRTIO_NET_F_MTU)) {
1779 vtpci_read_dev_config(hw,
1780 offsetof(struct virtio_net_config, mtu),
1782 sizeof(config->mtu));
1785 * MTU value has already been checked at negotiation
1786 * time, but check again in case it has changed since
1787 * then, which should not happen.
1789 if (config->mtu < RTE_ETHER_MIN_MTU) {
1790 PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1795 hw->max_mtu = config->mtu;
1796 /* Set initial MTU to maximum one supported by vhost */
1797 eth_dev->data->mtu = config->mtu;
1800 hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1801 VLAN_TAG_LEN - hw->vtnet_hdr_size;
1804 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1805 config->max_virtqueue_pairs);
1806 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1808 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1809 config->mac[0], config->mac[1],
1810 config->mac[2], config->mac[3],
1811 config->mac[4], config->mac[5]);
1813 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1814 hw->max_queue_pairs = 1;
1815 hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1816 VLAN_TAG_LEN - hw->vtnet_hdr_size;
1819 ret = virtio_alloc_queues(eth_dev);
1823 if (eth_dev->data->dev_conf.intr_conf.rxq) {
1824 if (virtio_configure_intr(eth_dev) < 0) {
1825 PMD_INIT_LOG(ERR, "failed to configure interrupt");
1826 virtio_free_queues(hw);
1831 vtpci_reinit_complete(hw);
1834 PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1835 eth_dev->data->port_id, pci_dev->id.vendor_id,
1836 pci_dev->id.device_id);
1842 * Remap the PCI device again (IO port map for legacy device and
1843 * memory map for modern device), so that the secondary process
1844 * could have the PCI initiated correctly.
1847 virtio_remap_pci(struct rte_pci_device *pci_dev, struct virtio_hw *hw)
1851 * We don't have to re-parse the PCI config space, since
1852 * rte_pci_map_device() makes sure the mapped address
1853 * in secondary process would equal to the one mapped in
1854 * the primary process: error will be returned if that
1855 * requirement is not met.
1857 * That said, we could simply reuse all cap pointers
1858 * (such as dev_cfg, common_cfg, etc.) parsed from the
1859 * primary process, which is stored in shared memory.
1861 if (rte_pci_map_device(pci_dev)) {
1862 PMD_INIT_LOG(DEBUG, "failed to map pci device!");
1866 if (rte_pci_ioport_map(pci_dev, 0, VTPCI_IO(hw)) < 0)
1874 virtio_set_vtpci_ops(struct virtio_hw *hw)
1876 #ifdef RTE_VIRTIO_USER
1877 if (hw->virtio_user_dev)
1878 VTPCI_OPS(hw) = &virtio_user_ops;
1882 VTPCI_OPS(hw) = &modern_ops;
1884 VTPCI_OPS(hw) = &legacy_ops;
1888 * This function is based on probe() function in virtio_pci.c
1889 * It returns 0 on success.
1892 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1894 struct virtio_hw *hw = eth_dev->data->dev_private;
1895 uint32_t speed = SPEED_UNKNOWN;
1899 if (sizeof(struct virtio_net_hdr_mrg_rxbuf) > RTE_PKTMBUF_HEADROOM) {
1901 "Not sufficient headroom required = %d, avail = %d",
1902 (int)sizeof(struct virtio_net_hdr_mrg_rxbuf),
1903 RTE_PKTMBUF_HEADROOM);
1908 eth_dev->dev_ops = &virtio_eth_dev_ops;
1910 if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1911 if (!hw->virtio_user_dev) {
1912 ret = virtio_remap_pci(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1917 virtio_set_vtpci_ops(hw);
1918 set_rxtx_funcs(eth_dev);
1922 ret = virtio_dev_devargs_parse(eth_dev->device->devargs,
1923 NULL, &speed, &vectorized);
1928 * Pass the information to the rte_eth_dev_close() that it should also
1929 * release the private port resources.
1931 eth_dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
1933 /* Allocate memory for storing MAC addresses */
1934 eth_dev->data->mac_addrs = rte_zmalloc("virtio",
1935 VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN, 0);
1936 if (eth_dev->data->mac_addrs == NULL) {
1938 "Failed to allocate %d bytes needed to store MAC addresses",
1939 VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN);
1943 hw->port_id = eth_dev->data->port_id;
1944 /* For virtio_user case the hw->virtio_user_dev is populated by
1945 * virtio_user_eth_dev_alloc() before eth_virtio_dev_init() is called.
1947 if (!hw->virtio_user_dev) {
1948 ret = vtpci_init(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1950 goto err_vtpci_init;
1953 rte_spinlock_init(&hw->state_lock);
1955 /* reset device and negotiate default features */
1956 ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1958 goto err_virtio_init;
1961 if (!vtpci_packed_queue(hw)) {
1964 #if !defined(CC_AVX512_SUPPORT)
1966 "building environment do not support packed ring vectorized");
1979 if (!hw->virtio_user_dev) {
1980 rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(eth_dev));
1982 rte_pci_ioport_unmap(VTPCI_IO(hw));
1985 rte_free(eth_dev->data->mac_addrs);
1986 eth_dev->data->mac_addrs = NULL;
1991 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1993 PMD_INIT_FUNC_TRACE();
1995 if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1998 virtio_dev_stop(eth_dev);
1999 virtio_dev_close(eth_dev);
2001 eth_dev->dev_ops = NULL;
2002 eth_dev->tx_pkt_burst = NULL;
2003 eth_dev->rx_pkt_burst = NULL;
2005 PMD_INIT_LOG(DEBUG, "dev_uninit completed");
2011 static int vdpa_check_handler(__rte_unused const char *key,
2012 const char *value, void *ret_val)
2014 if (strcmp(value, "1") == 0)
2015 *(int *)ret_val = 1;
2017 *(int *)ret_val = 0;
2024 virtio_dev_speed_capa_get(uint32_t speed)
2027 case ETH_SPEED_NUM_10G:
2028 return ETH_LINK_SPEED_10G;
2029 case ETH_SPEED_NUM_20G:
2030 return ETH_LINK_SPEED_20G;
2031 case ETH_SPEED_NUM_25G:
2032 return ETH_LINK_SPEED_25G;
2033 case ETH_SPEED_NUM_40G:
2034 return ETH_LINK_SPEED_40G;
2035 case ETH_SPEED_NUM_50G:
2036 return ETH_LINK_SPEED_50G;
2037 case ETH_SPEED_NUM_56G:
2038 return ETH_LINK_SPEED_56G;
2039 case ETH_SPEED_NUM_100G:
2040 return ETH_LINK_SPEED_100G;
2046 static int vectorized_check_handler(__rte_unused const char *key,
2047 const char *value, void *ret_val)
2049 if (strcmp(value, "1") == 0)
2050 *(int *)ret_val = 1;
2052 *(int *)ret_val = 0;
2057 #define VIRTIO_ARG_SPEED "speed"
2058 #define VIRTIO_ARG_VDPA "vdpa"
2059 #define VIRTIO_ARG_VECTORIZED "vectorized"
2063 link_speed_handler(const char *key __rte_unused,
2064 const char *value, void *ret_val)
2067 if (!value || !ret_val)
2069 val = strtoul(value, NULL, 0);
2070 /* validate input */
2071 if (virtio_dev_speed_capa_get(val) == 0)
2073 *(uint32_t *)ret_val = val;
2080 virtio_dev_devargs_parse(struct rte_devargs *devargs, int *vdpa,
2081 uint32_t *speed, int *vectorized)
2083 struct rte_kvargs *kvlist;
2086 if (devargs == NULL)
2089 kvlist = rte_kvargs_parse(devargs->args, NULL);
2090 if (kvlist == NULL) {
2091 PMD_INIT_LOG(ERR, "error when parsing param");
2094 if (vdpa && rte_kvargs_count(kvlist, VIRTIO_ARG_VDPA) == 1) {
2095 /* vdpa mode selected when there's a key-value pair:
2098 ret = rte_kvargs_process(kvlist, VIRTIO_ARG_VDPA,
2099 vdpa_check_handler, vdpa);
2101 PMD_INIT_LOG(ERR, "Failed to parse %s",
2106 if (speed && rte_kvargs_count(kvlist, VIRTIO_ARG_SPEED) == 1) {
2107 ret = rte_kvargs_process(kvlist,
2109 link_speed_handler, speed);
2111 PMD_INIT_LOG(ERR, "Failed to parse %s",
2118 rte_kvargs_count(kvlist, VIRTIO_ARG_VECTORIZED) == 1) {
2119 ret = rte_kvargs_process(kvlist,
2120 VIRTIO_ARG_VECTORIZED,
2121 vectorized_check_handler, vectorized);
2123 PMD_INIT_LOG(ERR, "Failed to parse %s",
2124 VIRTIO_ARG_VECTORIZED);
2130 rte_kvargs_free(kvlist);
2134 static int eth_virtio_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2135 struct rte_pci_device *pci_dev)
2140 ret = virtio_dev_devargs_parse(pci_dev->device.devargs, &vdpa, NULL,
2143 PMD_INIT_LOG(ERR, "devargs parsing is failed");
2146 /* virtio pmd skips probe if device needs to work in vdpa mode */
2150 return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct virtio_hw),
2151 eth_virtio_dev_init);
2154 static int eth_virtio_pci_remove(struct rte_pci_device *pci_dev)
2158 ret = rte_eth_dev_pci_generic_remove(pci_dev, eth_virtio_dev_uninit);
2159 /* Port has already been released by close. */
2165 static struct rte_pci_driver rte_virtio_pmd = {
2167 .name = "net_virtio",
2169 .id_table = pci_id_virtio_map,
2171 .probe = eth_virtio_pci_probe,
2172 .remove = eth_virtio_pci_remove,
2175 RTE_INIT(rte_virtio_pmd_init)
2177 rte_eal_iopl_init();
2178 rte_pci_register(&rte_virtio_pmd);
2182 rx_offload_enabled(struct virtio_hw *hw)
2184 return vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM) ||
2185 vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2186 vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6);
2190 tx_offload_enabled(struct virtio_hw *hw)
2192 return vtpci_with_feature(hw, VIRTIO_NET_F_CSUM) ||
2193 vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO4) ||
2194 vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO6);
2198 * Configure virtio device
2199 * It returns 0 on success.
2202 virtio_dev_configure(struct rte_eth_dev *dev)
2204 const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2205 const struct rte_eth_txmode *txmode = &dev->data->dev_conf.txmode;
2206 struct virtio_hw *hw = dev->data->dev_private;
2207 uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
2209 uint64_t rx_offloads = rxmode->offloads;
2210 uint64_t tx_offloads = txmode->offloads;
2211 uint64_t req_features;
2214 PMD_INIT_LOG(DEBUG, "configure");
2215 req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
2217 if (rxmode->mq_mode != ETH_MQ_RX_NONE) {
2219 "Unsupported Rx multi queue mode %d",
2224 if (txmode->mq_mode != ETH_MQ_TX_NONE) {
2226 "Unsupported Tx multi queue mode %d",
2231 if (dev->data->dev_conf.intr_conf.rxq) {
2232 ret = virtio_init_device(dev, hw->req_guest_features);
2237 if (rxmode->max_rx_pkt_len > hw->max_mtu + ether_hdr_len)
2238 req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
2240 if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2241 DEV_RX_OFFLOAD_TCP_CKSUM))
2242 req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
2244 if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO)
2246 (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2247 (1ULL << VIRTIO_NET_F_GUEST_TSO6);
2249 if (tx_offloads & (DEV_TX_OFFLOAD_UDP_CKSUM |
2250 DEV_TX_OFFLOAD_TCP_CKSUM))
2251 req_features |= (1ULL << VIRTIO_NET_F_CSUM);
2253 if (tx_offloads & DEV_TX_OFFLOAD_TCP_TSO)
2255 (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2256 (1ULL << VIRTIO_NET_F_HOST_TSO6);
2258 /* if request features changed, reinit the device */
2259 if (req_features != hw->req_guest_features) {
2260 ret = virtio_init_device(dev, req_features);
2265 if ((rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2266 DEV_RX_OFFLOAD_TCP_CKSUM)) &&
2267 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
2269 "rx checksum not available on this host");
2273 if ((rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) &&
2274 (!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2275 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6))) {
2277 "Large Receive Offload not available on this host");
2281 /* start control queue */
2282 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
2283 virtio_dev_cq_start(dev);
2285 if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
2288 if ((rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
2289 && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2291 "vlan filtering not available on this host");
2295 hw->has_tx_offload = tx_offload_enabled(hw);
2296 hw->has_rx_offload = rx_offload_enabled(hw);
2298 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2299 /* Enable vector (0) for Link State Intrerrupt */
2300 if (VTPCI_OPS(hw)->set_config_irq(hw, 0) ==
2301 VIRTIO_MSI_NO_VECTOR) {
2302 PMD_DRV_LOG(ERR, "failed to set config vector");
2306 if (vtpci_packed_queue(hw)) {
2307 #if defined(RTE_ARCH_X86_64) && defined(CC_AVX512_SUPPORT)
2308 if ((hw->use_vec_rx || hw->use_vec_tx) &&
2309 (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX512F) ||
2310 !vtpci_with_feature(hw, VIRTIO_F_IN_ORDER) ||
2311 !vtpci_with_feature(hw, VIRTIO_F_VERSION_1))) {
2313 "disabled packed ring vectorized path for requirements not met");
2322 if (hw->use_vec_rx) {
2323 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2325 "disabled packed ring vectorized rx for mrg_rxbuf enabled");
2329 if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) {
2331 "disabled packed ring vectorized rx for TCP_LRO enabled");
2336 if (vtpci_with_feature(hw, VIRTIO_F_IN_ORDER)) {
2337 hw->use_inorder_tx = 1;
2338 hw->use_inorder_rx = 1;
2342 if (hw->use_vec_rx) {
2343 #if defined RTE_ARCH_ARM64 || defined RTE_ARCH_ARM
2344 if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
2346 "disabled split ring vectorized path for requirement not met");
2350 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2352 "disabled split ring vectorized rx for mrg_rxbuf enabled");
2356 if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2357 DEV_RX_OFFLOAD_TCP_CKSUM |
2358 DEV_RX_OFFLOAD_TCP_LRO |
2359 DEV_RX_OFFLOAD_VLAN_STRIP)) {
2361 "disabled split ring vectorized rx for offloading enabled");
2372 virtio_dev_start(struct rte_eth_dev *dev)
2374 uint16_t nb_queues, i;
2375 struct virtnet_rx *rxvq;
2376 struct virtnet_tx *txvq __rte_unused;
2377 struct virtio_hw *hw = dev->data->dev_private;
2380 /* Finish the initialization of the queues */
2381 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2382 ret = virtio_dev_rx_queue_setup_finish(dev, i);
2386 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2387 ret = virtio_dev_tx_queue_setup_finish(dev, i);
2392 /* check if lsc interrupt feature is enabled */
2393 if (dev->data->dev_conf.intr_conf.lsc) {
2394 if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
2395 PMD_DRV_LOG(ERR, "link status not supported by host");
2400 /* Enable uio/vfio intr/eventfd mapping: althrough we already did that
2401 * in device configure, but it could be unmapped when device is
2404 if (dev->data->dev_conf.intr_conf.lsc ||
2405 dev->data->dev_conf.intr_conf.rxq) {
2406 virtio_intr_disable(dev);
2408 /* Setup interrupt callback */
2409 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2410 rte_intr_callback_register(dev->intr_handle,
2411 virtio_interrupt_handler,
2414 if (virtio_intr_enable(dev) < 0) {
2415 PMD_DRV_LOG(ERR, "interrupt enable failed");
2420 /*Notify the backend
2421 *Otherwise the tap backend might already stop its queue due to fullness.
2422 *vhost backend will have no chance to be waked up
2424 nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
2425 if (hw->max_queue_pairs > 1) {
2426 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
2430 PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
2432 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2433 rxvq = dev->data->rx_queues[i];
2434 /* Flush the old packets */
2435 virtqueue_rxvq_flush(rxvq->vq);
2436 virtqueue_notify(rxvq->vq);
2439 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2440 txvq = dev->data->tx_queues[i];
2441 virtqueue_notify(txvq->vq);
2444 PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
2446 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2447 rxvq = dev->data->rx_queues[i];
2448 VIRTQUEUE_DUMP(rxvq->vq);
2451 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2452 txvq = dev->data->tx_queues[i];
2453 VIRTQUEUE_DUMP(txvq->vq);
2456 set_rxtx_funcs(dev);
2459 /* Initialize Link state */
2460 virtio_dev_link_update(dev, 0);
2465 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
2467 struct virtio_hw *hw = dev->data->dev_private;
2468 uint16_t nr_vq = virtio_get_nr_vq(hw);
2469 const char *type __rte_unused;
2470 unsigned int i, mbuf_num = 0;
2471 struct virtqueue *vq;
2472 struct rte_mbuf *buf;
2475 if (hw->vqs == NULL)
2478 for (i = 0; i < nr_vq; i++) {
2483 queue_type = virtio_get_queue_type(hw, i);
2484 if (queue_type == VTNET_RQ)
2486 else if (queue_type == VTNET_TQ)
2492 "Before freeing %s[%d] used and unused buf",
2496 while ((buf = virtqueue_detach_unused(vq)) != NULL) {
2497 rte_pktmbuf_free(buf);
2502 "After freeing %s[%d] used and unused buf",
2507 PMD_INIT_LOG(DEBUG, "%d mbufs freed", mbuf_num);
2511 * Stop device: disable interrupt and mark link down
2514 virtio_dev_stop(struct rte_eth_dev *dev)
2516 struct virtio_hw *hw = dev->data->dev_private;
2517 struct rte_eth_link link;
2518 struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
2520 PMD_INIT_LOG(DEBUG, "stop");
2522 rte_spinlock_lock(&hw->state_lock);
2525 hw->started = false;
2527 if (intr_conf->lsc || intr_conf->rxq) {
2528 virtio_intr_disable(dev);
2530 /* Reset interrupt callback */
2531 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
2532 rte_intr_callback_unregister(dev->intr_handle,
2533 virtio_interrupt_handler,
2538 memset(&link, 0, sizeof(link));
2539 rte_eth_linkstatus_set(dev, &link);
2541 rte_spinlock_unlock(&hw->state_lock);
2545 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
2547 struct rte_eth_link link;
2549 struct virtio_hw *hw = dev->data->dev_private;
2551 memset(&link, 0, sizeof(link));
2552 link.link_duplex = hw->duplex;
2553 link.link_speed = hw->speed;
2554 link.link_autoneg = ETH_LINK_AUTONEG;
2557 link.link_status = ETH_LINK_DOWN;
2558 } else if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
2559 PMD_INIT_LOG(DEBUG, "Get link status from hw");
2560 vtpci_read_dev_config(hw,
2561 offsetof(struct virtio_net_config, status),
2562 &status, sizeof(status));
2563 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
2564 link.link_status = ETH_LINK_DOWN;
2565 PMD_INIT_LOG(DEBUG, "Port %d is down",
2566 dev->data->port_id);
2568 link.link_status = ETH_LINK_UP;
2569 PMD_INIT_LOG(DEBUG, "Port %d is up",
2570 dev->data->port_id);
2573 link.link_status = ETH_LINK_UP;
2576 return rte_eth_linkstatus_set(dev, &link);
2580 virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
2582 const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2583 struct virtio_hw *hw = dev->data->dev_private;
2584 uint64_t offloads = rxmode->offloads;
2586 if (mask & ETH_VLAN_FILTER_MASK) {
2587 if ((offloads & DEV_RX_OFFLOAD_VLAN_FILTER) &&
2588 !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2591 "vlan filtering not available on this host");
2597 if (mask & ETH_VLAN_STRIP_MASK)
2598 hw->vlan_strip = !!(offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
2604 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2606 uint64_t tso_mask, host_features;
2607 struct virtio_hw *hw = dev->data->dev_private;
2608 dev_info->speed_capa = virtio_dev_speed_capa_get(hw->speed);
2610 dev_info->max_rx_queues =
2611 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
2612 dev_info->max_tx_queues =
2613 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
2614 dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
2615 dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
2616 dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
2618 host_features = VTPCI_OPS(hw)->get_features(hw);
2619 dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
2620 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_JUMBO_FRAME;
2621 if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
2622 dev_info->rx_offload_capa |=
2623 DEV_RX_OFFLOAD_TCP_CKSUM |
2624 DEV_RX_OFFLOAD_UDP_CKSUM;
2626 if (host_features & (1ULL << VIRTIO_NET_F_CTRL_VLAN))
2627 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_VLAN_FILTER;
2628 tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2629 (1ULL << VIRTIO_NET_F_GUEST_TSO6);
2630 if ((host_features & tso_mask) == tso_mask)
2631 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
2633 dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
2634 DEV_TX_OFFLOAD_VLAN_INSERT;
2635 if (host_features & (1ULL << VIRTIO_NET_F_CSUM)) {
2636 dev_info->tx_offload_capa |=
2637 DEV_TX_OFFLOAD_UDP_CKSUM |
2638 DEV_TX_OFFLOAD_TCP_CKSUM;
2640 tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2641 (1ULL << VIRTIO_NET_F_HOST_TSO6);
2642 if ((host_features & tso_mask) == tso_mask)
2643 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
2649 * It enables testpmd to collect per queue stats.
2652 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
2653 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
2654 __rte_unused uint8_t is_rx)
2659 RTE_PMD_EXPORT_NAME(net_virtio, __COUNTER__);
2660 RTE_PMD_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map);
2661 RTE_PMD_REGISTER_KMOD_DEP(net_virtio, "* igb_uio | uio_pci_generic | vfio-pci");
2663 RTE_INIT(virtio_init_log)
2665 virtio_logtype_init = rte_log_register("pmd.net.virtio.init");
2666 if (virtio_logtype_init >= 0)
2667 rte_log_set_level(virtio_logtype_init, RTE_LOG_NOTICE);
2668 virtio_logtype_driver = rte_log_register("pmd.net.virtio.driver");
2669 if (virtio_logtype_driver >= 0)
2670 rte_log_set_level(virtio_logtype_driver, RTE_LOG_NOTICE);