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);
294 while (VIRTQUEUE_NUSED(vq) == 0) {
299 while (VIRTQUEUE_NUSED(vq)) {
300 uint32_t idx, desc_idx, used_idx;
301 struct vring_used_elem *uep;
303 used_idx = (uint32_t)(vq->vq_used_cons_idx
304 & (vq->vq_nentries - 1));
305 uep = &vq->vq_split.ring.used->ring[used_idx];
306 idx = (uint32_t) uep->id;
309 while (vq->vq_split.ring.desc[desc_idx].flags &
311 desc_idx = vq->vq_split.ring.desc[desc_idx].next;
315 vq->vq_split.ring.desc[desc_idx].next = vq->vq_desc_head_idx;
316 vq->vq_desc_head_idx = idx;
318 vq->vq_used_cons_idx++;
322 PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
323 vq->vq_free_cnt, vq->vq_desc_head_idx);
325 result = cvq->virtio_net_hdr_mz->addr;
330 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
331 int *dlen, int pkt_num)
333 virtio_net_ctrl_ack status = ~0;
334 struct virtio_pmd_ctrl *result;
335 struct virtqueue *vq;
337 ctrl->status = status;
339 if (!cvq || !cvq->vq) {
340 PMD_INIT_LOG(ERR, "Control queue is not supported.");
344 rte_spinlock_lock(&cvq->lock);
347 PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
348 "vq->hw->cvq = %p vq = %p",
349 vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
351 if (vq->vq_free_cnt < pkt_num + 2 || pkt_num < 1) {
352 rte_spinlock_unlock(&cvq->lock);
356 memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
357 sizeof(struct virtio_pmd_ctrl));
359 if (vtpci_packed_queue(vq->hw))
360 result = virtio_send_command_packed(cvq, ctrl, dlen, pkt_num);
362 result = virtio_send_command_split(cvq, ctrl, dlen, pkt_num);
364 rte_spinlock_unlock(&cvq->lock);
365 return result->status;
369 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
371 struct virtio_hw *hw = dev->data->dev_private;
372 struct virtio_pmd_ctrl ctrl;
376 ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
377 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
378 memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
380 dlen[0] = sizeof(uint16_t);
382 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
384 PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
385 "failed, this is too late now...");
393 virtio_dev_queue_release(void *queue __rte_unused)
399 virtio_get_nr_vq(struct virtio_hw *hw)
401 uint16_t nr_vq = hw->max_queue_pairs * 2;
403 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
410 virtio_init_vring(struct virtqueue *vq)
412 int size = vq->vq_nentries;
413 uint8_t *ring_mem = vq->vq_ring_virt_mem;
415 PMD_INIT_FUNC_TRACE();
417 memset(ring_mem, 0, vq->vq_ring_size);
419 vq->vq_used_cons_idx = 0;
420 vq->vq_desc_head_idx = 0;
421 vq->vq_avail_idx = 0;
422 vq->vq_desc_tail_idx = (uint16_t)(vq->vq_nentries - 1);
423 vq->vq_free_cnt = vq->vq_nentries;
424 memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries);
425 if (vtpci_packed_queue(vq->hw)) {
426 vring_init_packed(&vq->vq_packed.ring, ring_mem,
427 VIRTIO_PCI_VRING_ALIGN, size);
428 vring_desc_init_packed(vq, size);
430 struct vring *vr = &vq->vq_split.ring;
432 vring_init_split(vr, ring_mem, VIRTIO_PCI_VRING_ALIGN, size);
433 vring_desc_init_split(vr->desc, size);
436 * Disable device(host) interrupting guest
438 virtqueue_disable_intr(vq);
442 virtio_init_queue(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx)
444 char vq_name[VIRTQUEUE_MAX_NAME_SZ];
445 char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
446 const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
447 unsigned int vq_size, size;
448 struct virtio_hw *hw = dev->data->dev_private;
449 struct virtnet_rx *rxvq = NULL;
450 struct virtnet_tx *txvq = NULL;
451 struct virtnet_ctl *cvq = NULL;
452 struct virtqueue *vq;
453 size_t sz_hdr_mz = 0;
454 void *sw_ring = NULL;
455 int queue_type = virtio_get_queue_type(hw, vtpci_queue_idx);
457 int numa_node = dev->device->numa_node;
459 PMD_INIT_LOG(INFO, "setting up queue: %u on NUMA node %d",
460 vtpci_queue_idx, numa_node);
463 * Read the virtqueue size from the Queue Size field
464 * Always power of 2 and if 0 virtqueue does not exist
466 vq_size = VTPCI_OPS(hw)->get_queue_num(hw, vtpci_queue_idx);
467 PMD_INIT_LOG(DEBUG, "vq_size: %u", vq_size);
469 PMD_INIT_LOG(ERR, "virtqueue does not exist");
473 if (!vtpci_packed_queue(hw) && !rte_is_power_of_2(vq_size)) {
474 PMD_INIT_LOG(ERR, "split virtqueue size is not power of 2");
478 snprintf(vq_name, sizeof(vq_name), "port%d_vq%d",
479 dev->data->port_id, vtpci_queue_idx);
481 size = RTE_ALIGN_CEIL(sizeof(*vq) +
482 vq_size * sizeof(struct vq_desc_extra),
483 RTE_CACHE_LINE_SIZE);
484 if (queue_type == VTNET_TQ) {
486 * For each xmit packet, allocate a virtio_net_hdr
487 * and indirect ring elements
489 sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
490 } else if (queue_type == VTNET_CQ) {
491 /* Allocate a page for control vq command, data and status */
492 sz_hdr_mz = PAGE_SIZE;
495 vq = rte_zmalloc_socket(vq_name, size, RTE_CACHE_LINE_SIZE,
498 PMD_INIT_LOG(ERR, "can not allocate vq");
501 hw->vqs[vtpci_queue_idx] = vq;
504 vq->vq_queue_index = vtpci_queue_idx;
505 vq->vq_nentries = vq_size;
506 if (vtpci_packed_queue(hw)) {
507 vq->vq_packed.used_wrap_counter = 1;
508 vq->vq_packed.cached_flags = VRING_PACKED_DESC_F_AVAIL;
509 vq->vq_packed.event_flags_shadow = 0;
510 if (queue_type == VTNET_RQ)
511 vq->vq_packed.cached_flags |= VRING_DESC_F_WRITE;
515 * Reserve a memzone for vring elements
517 size = vring_size(hw, vq_size, VIRTIO_PCI_VRING_ALIGN);
518 vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
519 PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
520 size, vq->vq_ring_size);
522 mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
523 numa_node, RTE_MEMZONE_IOVA_CONTIG,
524 VIRTIO_PCI_VRING_ALIGN);
526 if (rte_errno == EEXIST)
527 mz = rte_memzone_lookup(vq_name);
534 memset(mz->addr, 0, mz->len);
536 vq->vq_ring_mem = mz->iova;
537 vq->vq_ring_virt_mem = mz->addr;
538 PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem: 0x%" PRIx64,
540 PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64,
541 (uint64_t)(uintptr_t)mz->addr);
543 virtio_init_vring(vq);
546 snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_vq%d_hdr",
547 dev->data->port_id, vtpci_queue_idx);
548 hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
549 numa_node, RTE_MEMZONE_IOVA_CONTIG,
550 RTE_CACHE_LINE_SIZE);
551 if (hdr_mz == NULL) {
552 if (rte_errno == EEXIST)
553 hdr_mz = rte_memzone_lookup(vq_hdr_name);
554 if (hdr_mz == NULL) {
561 if (queue_type == VTNET_RQ) {
562 size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
563 sizeof(vq->sw_ring[0]);
565 sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
566 RTE_CACHE_LINE_SIZE, numa_node);
568 PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
573 vq->sw_ring = sw_ring;
576 rxvq->port_id = dev->data->port_id;
578 } else if (queue_type == VTNET_TQ) {
581 txvq->port_id = dev->data->port_id;
583 txvq->virtio_net_hdr_mz = hdr_mz;
584 txvq->virtio_net_hdr_mem = hdr_mz->iova;
585 } else if (queue_type == VTNET_CQ) {
589 cvq->virtio_net_hdr_mz = hdr_mz;
590 cvq->virtio_net_hdr_mem = hdr_mz->iova;
591 memset(cvq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
596 /* For virtio_user case (that is when hw->virtio_user_dev is not NULL),
597 * we use virtual address. And we need properly set _offset_, please see
598 * VIRTIO_MBUF_DATA_DMA_ADDR in virtqueue.h for more information.
600 if (!hw->virtio_user_dev)
601 vq->offset = offsetof(struct rte_mbuf, buf_iova);
603 vq->vq_ring_mem = (uintptr_t)mz->addr;
604 vq->offset = offsetof(struct rte_mbuf, buf_addr);
605 if (queue_type == VTNET_TQ)
606 txvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
607 else if (queue_type == VTNET_CQ)
608 cvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
611 if (queue_type == VTNET_TQ) {
612 struct virtio_tx_region *txr;
616 memset(txr, 0, vq_size * sizeof(*txr));
617 for (i = 0; i < vq_size; i++) {
618 struct vring_desc *start_dp = txr[i].tx_indir;
620 /* first indirect descriptor is always the tx header */
621 if (!vtpci_packed_queue(hw)) {
622 vring_desc_init_split(start_dp,
623 RTE_DIM(txr[i].tx_indir));
624 start_dp->addr = txvq->virtio_net_hdr_mem
626 + offsetof(struct virtio_tx_region,
628 start_dp->len = hw->vtnet_hdr_size;
629 start_dp->flags = VRING_DESC_F_NEXT;
634 if (VTPCI_OPS(hw)->setup_queue(hw, vq) < 0) {
635 PMD_INIT_LOG(ERR, "setup_queue failed");
643 rte_memzone_free(hdr_mz);
644 rte_memzone_free(mz);
651 virtio_free_queues(struct virtio_hw *hw)
653 uint16_t nr_vq = virtio_get_nr_vq(hw);
654 struct virtqueue *vq;
661 for (i = 0; i < nr_vq; i++) {
666 queue_type = virtio_get_queue_type(hw, i);
667 if (queue_type == VTNET_RQ) {
668 rte_free(vq->sw_ring);
669 rte_memzone_free(vq->rxq.mz);
670 } else if (queue_type == VTNET_TQ) {
671 rte_memzone_free(vq->txq.mz);
672 rte_memzone_free(vq->txq.virtio_net_hdr_mz);
674 rte_memzone_free(vq->cq.mz);
675 rte_memzone_free(vq->cq.virtio_net_hdr_mz);
687 virtio_alloc_queues(struct rte_eth_dev *dev)
689 struct virtio_hw *hw = dev->data->dev_private;
690 uint16_t nr_vq = virtio_get_nr_vq(hw);
694 hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
696 PMD_INIT_LOG(ERR, "failed to allocate vqs");
700 for (i = 0; i < nr_vq; i++) {
701 ret = virtio_init_queue(dev, i);
703 virtio_free_queues(hw);
711 static void virtio_queues_unbind_intr(struct rte_eth_dev *dev);
714 virtio_dev_close(struct rte_eth_dev *dev)
716 struct virtio_hw *hw = dev->data->dev_private;
717 struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
719 PMD_INIT_LOG(DEBUG, "virtio_dev_close");
726 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
727 VTPCI_OPS(hw)->set_config_irq(hw, VIRTIO_MSI_NO_VECTOR);
729 virtio_queues_unbind_intr(dev);
731 if (intr_conf->lsc || intr_conf->rxq) {
732 virtio_intr_disable(dev);
733 rte_intr_efd_disable(dev->intr_handle);
734 rte_free(dev->intr_handle->intr_vec);
735 dev->intr_handle->intr_vec = NULL;
739 virtio_dev_free_mbufs(dev);
740 virtio_free_queues(hw);
742 #ifdef RTE_VIRTIO_USER
743 if (hw->virtio_user_dev)
744 virtio_user_dev_uninit(hw->virtio_user_dev);
748 rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(dev));
750 rte_pci_ioport_unmap(VTPCI_IO(hw));
755 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
757 struct virtio_hw *hw = dev->data->dev_private;
758 struct virtio_pmd_ctrl ctrl;
762 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
763 PMD_INIT_LOG(INFO, "host does not support rx control");
767 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
768 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
772 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
774 PMD_INIT_LOG(ERR, "Failed to enable promisc");
782 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
784 struct virtio_hw *hw = dev->data->dev_private;
785 struct virtio_pmd_ctrl ctrl;
789 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
790 PMD_INIT_LOG(INFO, "host does not support rx control");
794 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
795 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
799 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
801 PMD_INIT_LOG(ERR, "Failed to disable promisc");
809 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
811 struct virtio_hw *hw = dev->data->dev_private;
812 struct virtio_pmd_ctrl ctrl;
816 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
817 PMD_INIT_LOG(INFO, "host does not support rx control");
821 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
822 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
826 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
828 PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
836 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
838 struct virtio_hw *hw = dev->data->dev_private;
839 struct virtio_pmd_ctrl ctrl;
843 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
844 PMD_INIT_LOG(INFO, "host does not support rx control");
848 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
849 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
853 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
855 PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
862 #define VLAN_TAG_LEN 4 /* 802.3ac tag (not DMA'd) */
864 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
866 struct virtio_hw *hw = dev->data->dev_private;
867 uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
869 uint32_t frame_size = mtu + ether_hdr_len;
870 uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
872 max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
874 if (mtu < RTE_ETHER_MIN_MTU || frame_size > max_frame_size) {
875 PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
876 RTE_ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
883 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
885 struct virtio_hw *hw = dev->data->dev_private;
886 struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
887 struct virtqueue *vq = rxvq->vq;
889 virtqueue_enable_intr(vq);
890 virtio_mb(hw->weak_barriers);
895 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
897 struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
898 struct virtqueue *vq = rxvq->vq;
900 virtqueue_disable_intr(vq);
905 * dev_ops for virtio, bare necessities for basic operation
907 static const struct eth_dev_ops virtio_eth_dev_ops = {
908 .dev_configure = virtio_dev_configure,
909 .dev_start = virtio_dev_start,
910 .dev_stop = virtio_dev_stop,
911 .dev_close = virtio_dev_close,
912 .promiscuous_enable = virtio_dev_promiscuous_enable,
913 .promiscuous_disable = virtio_dev_promiscuous_disable,
914 .allmulticast_enable = virtio_dev_allmulticast_enable,
915 .allmulticast_disable = virtio_dev_allmulticast_disable,
916 .mtu_set = virtio_mtu_set,
917 .dev_infos_get = virtio_dev_info_get,
918 .stats_get = virtio_dev_stats_get,
919 .xstats_get = virtio_dev_xstats_get,
920 .xstats_get_names = virtio_dev_xstats_get_names,
921 .stats_reset = virtio_dev_stats_reset,
922 .xstats_reset = virtio_dev_stats_reset,
923 .link_update = virtio_dev_link_update,
924 .vlan_offload_set = virtio_dev_vlan_offload_set,
925 .rx_queue_setup = virtio_dev_rx_queue_setup,
926 .rx_queue_intr_enable = virtio_dev_rx_queue_intr_enable,
927 .rx_queue_intr_disable = virtio_dev_rx_queue_intr_disable,
928 .rx_queue_release = virtio_dev_queue_release,
929 .rx_descriptor_done = virtio_dev_rx_queue_done,
930 .tx_queue_setup = virtio_dev_tx_queue_setup,
931 .tx_queue_release = virtio_dev_queue_release,
932 /* collect stats per queue */
933 .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
934 .vlan_filter_set = virtio_vlan_filter_set,
935 .mac_addr_add = virtio_mac_addr_add,
936 .mac_addr_remove = virtio_mac_addr_remove,
937 .mac_addr_set = virtio_mac_addr_set,
941 * dev_ops for virtio-user in secondary processes, as we just have
942 * some limited supports currently.
944 const struct eth_dev_ops virtio_user_secondary_eth_dev_ops = {
945 .dev_infos_get = virtio_dev_info_get,
946 .stats_get = virtio_dev_stats_get,
947 .xstats_get = virtio_dev_xstats_get,
948 .xstats_get_names = virtio_dev_xstats_get_names,
949 .stats_reset = virtio_dev_stats_reset,
950 .xstats_reset = virtio_dev_stats_reset,
951 /* collect stats per queue */
952 .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
956 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
960 for (i = 0; i < dev->data->nb_tx_queues; i++) {
961 const struct virtnet_tx *txvq = dev->data->tx_queues[i];
965 stats->opackets += txvq->stats.packets;
966 stats->obytes += txvq->stats.bytes;
968 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
969 stats->q_opackets[i] = txvq->stats.packets;
970 stats->q_obytes[i] = txvq->stats.bytes;
974 for (i = 0; i < dev->data->nb_rx_queues; i++) {
975 const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
979 stats->ipackets += rxvq->stats.packets;
980 stats->ibytes += rxvq->stats.bytes;
981 stats->ierrors += rxvq->stats.errors;
983 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
984 stats->q_ipackets[i] = rxvq->stats.packets;
985 stats->q_ibytes[i] = rxvq->stats.bytes;
989 stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
992 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
993 struct rte_eth_xstat_name *xstats_names,
994 __rte_unused unsigned limit)
1000 unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
1001 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
1003 if (xstats_names != NULL) {
1004 /* Note: limit checked in rte_eth_xstats_names() */
1006 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1007 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1010 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1011 snprintf(xstats_names[count].name,
1012 sizeof(xstats_names[count].name),
1014 rte_virtio_rxq_stat_strings[t].name);
1019 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1020 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1023 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1024 snprintf(xstats_names[count].name,
1025 sizeof(xstats_names[count].name),
1027 rte_virtio_txq_stat_strings[t].name);
1037 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
1043 unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
1044 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
1049 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1050 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1057 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1058 xstats[count].value = *(uint64_t *)(((char *)rxvq) +
1059 rte_virtio_rxq_stat_strings[t].offset);
1060 xstats[count].id = count;
1065 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1066 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1073 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1074 xstats[count].value = *(uint64_t *)(((char *)txvq) +
1075 rte_virtio_txq_stat_strings[t].offset);
1076 xstats[count].id = count;
1085 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1087 virtio_update_stats(dev, stats);
1093 virtio_dev_stats_reset(struct rte_eth_dev *dev)
1097 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1098 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1102 txvq->stats.packets = 0;
1103 txvq->stats.bytes = 0;
1104 txvq->stats.multicast = 0;
1105 txvq->stats.broadcast = 0;
1106 memset(txvq->stats.size_bins, 0,
1107 sizeof(txvq->stats.size_bins[0]) * 8);
1110 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1111 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1115 rxvq->stats.packets = 0;
1116 rxvq->stats.bytes = 0;
1117 rxvq->stats.errors = 0;
1118 rxvq->stats.multicast = 0;
1119 rxvq->stats.broadcast = 0;
1120 memset(rxvq->stats.size_bins, 0,
1121 sizeof(rxvq->stats.size_bins[0]) * 8);
1128 virtio_set_hwaddr(struct virtio_hw *hw)
1130 vtpci_write_dev_config(hw,
1131 offsetof(struct virtio_net_config, mac),
1132 &hw->mac_addr, RTE_ETHER_ADDR_LEN);
1136 virtio_get_hwaddr(struct virtio_hw *hw)
1138 if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
1139 vtpci_read_dev_config(hw,
1140 offsetof(struct virtio_net_config, mac),
1141 &hw->mac_addr, RTE_ETHER_ADDR_LEN);
1143 rte_eth_random_addr(&hw->mac_addr[0]);
1144 virtio_set_hwaddr(hw);
1149 virtio_mac_table_set(struct virtio_hw *hw,
1150 const struct virtio_net_ctrl_mac *uc,
1151 const struct virtio_net_ctrl_mac *mc)
1153 struct virtio_pmd_ctrl ctrl;
1156 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1157 PMD_DRV_LOG(INFO, "host does not support mac table");
1161 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1162 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1164 len[0] = uc->entries * RTE_ETHER_ADDR_LEN + sizeof(uc->entries);
1165 memcpy(ctrl.data, uc, len[0]);
1167 len[1] = mc->entries * RTE_ETHER_ADDR_LEN + sizeof(mc->entries);
1168 memcpy(ctrl.data + len[0], mc, len[1]);
1170 err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1172 PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1177 virtio_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
1178 uint32_t index, uint32_t vmdq __rte_unused)
1180 struct virtio_hw *hw = dev->data->dev_private;
1181 const struct rte_ether_addr *addrs = dev->data->mac_addrs;
1183 struct virtio_net_ctrl_mac *uc, *mc;
1185 if (index >= VIRTIO_MAX_MAC_ADDRS) {
1186 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1190 uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1191 sizeof(uc->entries));
1193 mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1194 sizeof(mc->entries));
1197 for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1198 const struct rte_ether_addr *addr
1199 = (i == index) ? mac_addr : addrs + i;
1200 struct virtio_net_ctrl_mac *tbl
1201 = rte_is_multicast_ether_addr(addr) ? mc : uc;
1203 memcpy(&tbl->macs[tbl->entries++], addr, RTE_ETHER_ADDR_LEN);
1206 return virtio_mac_table_set(hw, uc, mc);
1210 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1212 struct virtio_hw *hw = dev->data->dev_private;
1213 struct rte_ether_addr *addrs = dev->data->mac_addrs;
1214 struct virtio_net_ctrl_mac *uc, *mc;
1217 if (index >= VIRTIO_MAX_MAC_ADDRS) {
1218 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1222 uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1223 sizeof(uc->entries));
1225 mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1226 sizeof(mc->entries));
1229 for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1230 struct virtio_net_ctrl_mac *tbl;
1232 if (i == index || rte_is_zero_ether_addr(addrs + i))
1235 tbl = rte_is_multicast_ether_addr(addrs + i) ? mc : uc;
1236 memcpy(&tbl->macs[tbl->entries++], addrs + i,
1237 RTE_ETHER_ADDR_LEN);
1240 virtio_mac_table_set(hw, uc, mc);
1244 virtio_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1246 struct virtio_hw *hw = dev->data->dev_private;
1248 memcpy(hw->mac_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1250 /* Use atomic update if available */
1251 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1252 struct virtio_pmd_ctrl ctrl;
1253 int len = RTE_ETHER_ADDR_LEN;
1255 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1256 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1258 memcpy(ctrl.data, mac_addr, RTE_ETHER_ADDR_LEN);
1259 return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1262 if (!vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1265 virtio_set_hwaddr(hw);
1270 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1272 struct virtio_hw *hw = dev->data->dev_private;
1273 struct virtio_pmd_ctrl ctrl;
1276 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1279 ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1280 ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1281 memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1282 len = sizeof(vlan_id);
1284 return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1288 virtio_intr_unmask(struct rte_eth_dev *dev)
1290 struct virtio_hw *hw = dev->data->dev_private;
1292 if (rte_intr_ack(dev->intr_handle) < 0)
1295 if (!hw->virtio_user_dev)
1296 hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1302 virtio_intr_enable(struct rte_eth_dev *dev)
1304 struct virtio_hw *hw = dev->data->dev_private;
1306 if (rte_intr_enable(dev->intr_handle) < 0)
1309 if (!hw->virtio_user_dev)
1310 hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1316 virtio_intr_disable(struct rte_eth_dev *dev)
1318 struct virtio_hw *hw = dev->data->dev_private;
1320 if (rte_intr_disable(dev->intr_handle) < 0)
1323 if (!hw->virtio_user_dev)
1324 hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1330 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1332 uint64_t host_features;
1334 /* Prepare guest_features: feature that driver wants to support */
1335 PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1338 /* Read device(host) feature bits */
1339 host_features = VTPCI_OPS(hw)->get_features(hw);
1340 PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1343 /* If supported, ensure MTU value is valid before acknowledging it. */
1344 if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1345 struct virtio_net_config config;
1347 vtpci_read_dev_config(hw,
1348 offsetof(struct virtio_net_config, mtu),
1349 &config.mtu, sizeof(config.mtu));
1351 if (config.mtu < RTE_ETHER_MIN_MTU)
1352 req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1356 * Negotiate features: Subset of device feature bits are written back
1357 * guest feature bits.
1359 hw->guest_features = req_features;
1360 hw->guest_features = vtpci_negotiate_features(hw, host_features);
1361 PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1362 hw->guest_features);
1365 if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1367 "VIRTIO_F_VERSION_1 features is not enabled.");
1370 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1371 if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1373 "failed to set FEATURES_OK status!");
1378 hw->req_guest_features = req_features;
1384 virtio_dev_pause(struct rte_eth_dev *dev)
1386 struct virtio_hw *hw = dev->data->dev_private;
1388 rte_spinlock_lock(&hw->state_lock);
1390 if (hw->started == 0) {
1391 /* Device is just stopped. */
1392 rte_spinlock_unlock(&hw->state_lock);
1397 * Prevent the worker threads from touching queues to avoid contention,
1398 * 1 ms should be enough for the ongoing Tx function to finish.
1405 * Recover hw state to let the worker threads continue.
1408 virtio_dev_resume(struct rte_eth_dev *dev)
1410 struct virtio_hw *hw = dev->data->dev_private;
1413 rte_spinlock_unlock(&hw->state_lock);
1417 * Should be called only after device is paused.
1420 virtio_inject_pkts(struct rte_eth_dev *dev, struct rte_mbuf **tx_pkts,
1423 struct virtio_hw *hw = dev->data->dev_private;
1424 struct virtnet_tx *txvq = dev->data->tx_queues[0];
1427 hw->inject_pkts = tx_pkts;
1428 ret = dev->tx_pkt_burst(txvq, tx_pkts, nb_pkts);
1429 hw->inject_pkts = NULL;
1435 virtio_notify_peers(struct rte_eth_dev *dev)
1437 struct virtio_hw *hw = dev->data->dev_private;
1438 struct virtnet_rx *rxvq;
1439 struct rte_mbuf *rarp_mbuf;
1441 if (!dev->data->rx_queues)
1444 rxvq = dev->data->rx_queues[0];
1448 rarp_mbuf = rte_net_make_rarp_packet(rxvq->mpool,
1449 (struct rte_ether_addr *)hw->mac_addr);
1450 if (rarp_mbuf == NULL) {
1451 PMD_DRV_LOG(ERR, "failed to make RARP packet.");
1455 /* If virtio port just stopped, no need to send RARP */
1456 if (virtio_dev_pause(dev) < 0) {
1457 rte_pktmbuf_free(rarp_mbuf);
1461 virtio_inject_pkts(dev, &rarp_mbuf, 1);
1462 virtio_dev_resume(dev);
1466 virtio_ack_link_announce(struct rte_eth_dev *dev)
1468 struct virtio_hw *hw = dev->data->dev_private;
1469 struct virtio_pmd_ctrl ctrl;
1471 ctrl.hdr.class = VIRTIO_NET_CTRL_ANNOUNCE;
1472 ctrl.hdr.cmd = VIRTIO_NET_CTRL_ANNOUNCE_ACK;
1474 virtio_send_command(hw->cvq, &ctrl, NULL, 0);
1478 * Process virtio config changed interrupt. Call the callback
1479 * if link state changed, generate gratuitous RARP packet if
1480 * the status indicates an ANNOUNCE.
1483 virtio_interrupt_handler(void *param)
1485 struct rte_eth_dev *dev = param;
1486 struct virtio_hw *hw = dev->data->dev_private;
1490 /* Read interrupt status which clears interrupt */
1491 isr = vtpci_isr(hw);
1492 PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1494 if (virtio_intr_unmask(dev) < 0)
1495 PMD_DRV_LOG(ERR, "interrupt enable failed");
1497 if (isr & VIRTIO_PCI_ISR_CONFIG) {
1498 if (virtio_dev_link_update(dev, 0) == 0)
1499 _rte_eth_dev_callback_process(dev,
1500 RTE_ETH_EVENT_INTR_LSC,
1503 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1504 vtpci_read_dev_config(hw,
1505 offsetof(struct virtio_net_config, status),
1506 &status, sizeof(status));
1507 if (status & VIRTIO_NET_S_ANNOUNCE) {
1508 virtio_notify_peers(dev);
1510 virtio_ack_link_announce(dev);
1516 /* set rx and tx handlers according to what is supported */
1518 set_rxtx_funcs(struct rte_eth_dev *eth_dev)
1520 struct virtio_hw *hw = eth_dev->data->dev_private;
1522 eth_dev->tx_pkt_prepare = virtio_xmit_pkts_prepare;
1523 if (vtpci_packed_queue(hw)) {
1525 "virtio: using packed ring %s Tx path on port %u",
1526 hw->use_vec_tx ? "vectorized" : "standard",
1527 eth_dev->data->port_id);
1529 eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed_vec;
1531 eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed;
1533 if (hw->use_inorder_tx) {
1534 PMD_INIT_LOG(INFO, "virtio: using inorder Tx path on port %u",
1535 eth_dev->data->port_id);
1536 eth_dev->tx_pkt_burst = virtio_xmit_pkts_inorder;
1538 PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
1539 eth_dev->data->port_id);
1540 eth_dev->tx_pkt_burst = virtio_xmit_pkts;
1544 if (vtpci_packed_queue(hw)) {
1545 if (hw->use_vec_rx) {
1547 "virtio: using packed ring vectorized Rx path on port %u",
1548 eth_dev->data->port_id);
1549 eth_dev->rx_pkt_burst =
1550 &virtio_recv_pkts_packed_vec;
1551 } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1553 "virtio: using packed ring mergeable buffer Rx path on port %u",
1554 eth_dev->data->port_id);
1555 eth_dev->rx_pkt_burst =
1556 &virtio_recv_mergeable_pkts_packed;
1559 "virtio: using packed ring standard Rx path on port %u",
1560 eth_dev->data->port_id);
1561 eth_dev->rx_pkt_burst = &virtio_recv_pkts_packed;
1564 if (hw->use_vec_rx) {
1565 PMD_INIT_LOG(INFO, "virtio: using vectorized Rx path on port %u",
1566 eth_dev->data->port_id);
1567 eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1568 } else if (hw->use_inorder_rx) {
1570 "virtio: using inorder Rx path on port %u",
1571 eth_dev->data->port_id);
1572 eth_dev->rx_pkt_burst = &virtio_recv_pkts_inorder;
1573 } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1575 "virtio: using mergeable buffer Rx path on port %u",
1576 eth_dev->data->port_id);
1577 eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1579 PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
1580 eth_dev->data->port_id);
1581 eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1587 /* Only support 1:1 queue/interrupt mapping so far.
1588 * TODO: support n:1 queue/interrupt mapping when there are limited number of
1589 * interrupt vectors (<N+1).
1592 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1595 struct virtio_hw *hw = dev->data->dev_private;
1597 PMD_INIT_LOG(INFO, "queue/interrupt binding");
1598 for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1599 dev->intr_handle->intr_vec[i] = i + 1;
1600 if (VTPCI_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1601 VIRTIO_MSI_NO_VECTOR) {
1602 PMD_DRV_LOG(ERR, "failed to set queue vector");
1611 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1614 struct virtio_hw *hw = dev->data->dev_private;
1616 PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1617 for (i = 0; i < dev->data->nb_rx_queues; ++i)
1618 VTPCI_OPS(hw)->set_queue_irq(hw,
1619 hw->vqs[i * VTNET_CQ],
1620 VIRTIO_MSI_NO_VECTOR);
1624 virtio_configure_intr(struct rte_eth_dev *dev)
1626 struct virtio_hw *hw = dev->data->dev_private;
1628 if (!rte_intr_cap_multiple(dev->intr_handle)) {
1629 PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1633 if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1634 PMD_INIT_LOG(ERR, "Fail to create eventfd");
1638 if (!dev->intr_handle->intr_vec) {
1639 dev->intr_handle->intr_vec =
1640 rte_zmalloc("intr_vec",
1641 hw->max_queue_pairs * sizeof(int), 0);
1642 if (!dev->intr_handle->intr_vec) {
1643 PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1644 hw->max_queue_pairs);
1649 /* Re-register callback to update max_intr */
1650 rte_intr_callback_unregister(dev->intr_handle,
1651 virtio_interrupt_handler,
1653 rte_intr_callback_register(dev->intr_handle,
1654 virtio_interrupt_handler,
1657 /* DO NOT try to remove this! This function will enable msix, or QEMU
1658 * will encounter SIGSEGV when DRIVER_OK is sent.
1659 * And for legacy devices, this should be done before queue/vec binding
1660 * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1661 * (22) will be ignored.
1663 if (virtio_intr_enable(dev) < 0) {
1664 PMD_DRV_LOG(ERR, "interrupt enable failed");
1668 if (virtio_queues_bind_intr(dev) < 0) {
1669 PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1675 #define SPEED_UNKNOWN 0xffffffff
1676 #define DUPLEX_UNKNOWN 0xff
1677 /* reset device and renegotiate features if needed */
1679 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1681 struct virtio_hw *hw = eth_dev->data->dev_private;
1682 struct virtio_net_config *config;
1683 struct virtio_net_config local_config;
1684 struct rte_pci_device *pci_dev = NULL;
1687 /* Reset the device although not necessary at startup */
1691 virtio_dev_free_mbufs(eth_dev);
1692 virtio_free_queues(hw);
1695 /* Tell the host we've noticed this device. */
1696 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1698 /* Tell the host we've known how to drive the device. */
1699 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1700 if (virtio_negotiate_features(hw, req_features) < 0)
1703 hw->weak_barriers = !vtpci_with_feature(hw, VIRTIO_F_ORDER_PLATFORM);
1705 if (!hw->virtio_user_dev)
1706 pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1708 /* If host does not support both status and MSI-X then disable LSC */
1709 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS) &&
1710 hw->use_msix != VIRTIO_MSIX_NONE)
1711 eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1713 eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1715 /* Setting up rx_header size for the device */
1716 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1717 vtpci_with_feature(hw, VIRTIO_F_VERSION_1) ||
1718 vtpci_with_feature(hw, VIRTIO_F_RING_PACKED))
1719 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1721 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1723 /* Copy the permanent MAC address to: virtio_hw */
1724 virtio_get_hwaddr(hw);
1725 rte_ether_addr_copy((struct rte_ether_addr *)hw->mac_addr,
1726 ð_dev->data->mac_addrs[0]);
1728 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1729 hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1730 hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1732 if (hw->speed == SPEED_UNKNOWN) {
1733 if (vtpci_with_feature(hw, VIRTIO_NET_F_SPEED_DUPLEX)) {
1734 config = &local_config;
1735 vtpci_read_dev_config(hw,
1736 offsetof(struct virtio_net_config, speed),
1737 &config->speed, sizeof(config->speed));
1738 vtpci_read_dev_config(hw,
1739 offsetof(struct virtio_net_config, duplex),
1740 &config->duplex, sizeof(config->duplex));
1741 hw->speed = config->speed;
1742 hw->duplex = config->duplex;
1745 if (hw->speed == SPEED_UNKNOWN)
1746 hw->speed = ETH_SPEED_NUM_10G;
1747 if (hw->duplex == DUPLEX_UNKNOWN)
1748 hw->duplex = ETH_LINK_FULL_DUPLEX;
1749 PMD_INIT_LOG(DEBUG, "link speed = %d, duplex = %d",
1750 hw->speed, hw->duplex);
1751 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1752 config = &local_config;
1754 vtpci_read_dev_config(hw,
1755 offsetof(struct virtio_net_config, mac),
1756 &config->mac, sizeof(config->mac));
1758 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1759 vtpci_read_dev_config(hw,
1760 offsetof(struct virtio_net_config, status),
1761 &config->status, sizeof(config->status));
1764 "VIRTIO_NET_F_STATUS is not supported");
1768 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1769 vtpci_read_dev_config(hw,
1770 offsetof(struct virtio_net_config, max_virtqueue_pairs),
1771 &config->max_virtqueue_pairs,
1772 sizeof(config->max_virtqueue_pairs));
1775 "VIRTIO_NET_F_MQ is not supported");
1776 config->max_virtqueue_pairs = 1;
1779 hw->max_queue_pairs = config->max_virtqueue_pairs;
1781 if (vtpci_with_feature(hw, VIRTIO_NET_F_MTU)) {
1782 vtpci_read_dev_config(hw,
1783 offsetof(struct virtio_net_config, mtu),
1785 sizeof(config->mtu));
1788 * MTU value has already been checked at negotiation
1789 * time, but check again in case it has changed since
1790 * then, which should not happen.
1792 if (config->mtu < RTE_ETHER_MIN_MTU) {
1793 PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1798 hw->max_mtu = config->mtu;
1799 /* Set initial MTU to maximum one supported by vhost */
1800 eth_dev->data->mtu = config->mtu;
1803 hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1804 VLAN_TAG_LEN - hw->vtnet_hdr_size;
1807 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1808 config->max_virtqueue_pairs);
1809 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1811 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1812 config->mac[0], config->mac[1],
1813 config->mac[2], config->mac[3],
1814 config->mac[4], config->mac[5]);
1816 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1817 hw->max_queue_pairs = 1;
1818 hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1819 VLAN_TAG_LEN - hw->vtnet_hdr_size;
1822 ret = virtio_alloc_queues(eth_dev);
1826 if (eth_dev->data->dev_conf.intr_conf.rxq) {
1827 if (virtio_configure_intr(eth_dev) < 0) {
1828 PMD_INIT_LOG(ERR, "failed to configure interrupt");
1829 virtio_free_queues(hw);
1834 vtpci_reinit_complete(hw);
1837 PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1838 eth_dev->data->port_id, pci_dev->id.vendor_id,
1839 pci_dev->id.device_id);
1845 * Remap the PCI device again (IO port map for legacy device and
1846 * memory map for modern device), so that the secondary process
1847 * could have the PCI initiated correctly.
1850 virtio_remap_pci(struct rte_pci_device *pci_dev, struct virtio_hw *hw)
1854 * We don't have to re-parse the PCI config space, since
1855 * rte_pci_map_device() makes sure the mapped address
1856 * in secondary process would equal to the one mapped in
1857 * the primary process: error will be returned if that
1858 * requirement is not met.
1860 * That said, we could simply reuse all cap pointers
1861 * (such as dev_cfg, common_cfg, etc.) parsed from the
1862 * primary process, which is stored in shared memory.
1864 if (rte_pci_map_device(pci_dev)) {
1865 PMD_INIT_LOG(DEBUG, "failed to map pci device!");
1869 if (rte_pci_ioport_map(pci_dev, 0, VTPCI_IO(hw)) < 0)
1877 virtio_set_vtpci_ops(struct virtio_hw *hw)
1879 #ifdef RTE_VIRTIO_USER
1880 if (hw->virtio_user_dev)
1881 VTPCI_OPS(hw) = &virtio_user_ops;
1885 VTPCI_OPS(hw) = &modern_ops;
1887 VTPCI_OPS(hw) = &legacy_ops;
1891 * This function is based on probe() function in virtio_pci.c
1892 * It returns 0 on success.
1895 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1897 struct virtio_hw *hw = eth_dev->data->dev_private;
1898 uint32_t speed = SPEED_UNKNOWN;
1902 if (sizeof(struct virtio_net_hdr_mrg_rxbuf) > RTE_PKTMBUF_HEADROOM) {
1904 "Not sufficient headroom required = %d, avail = %d",
1905 (int)sizeof(struct virtio_net_hdr_mrg_rxbuf),
1906 RTE_PKTMBUF_HEADROOM);
1911 eth_dev->dev_ops = &virtio_eth_dev_ops;
1913 if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1914 if (!hw->virtio_user_dev) {
1915 ret = virtio_remap_pci(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1920 virtio_set_vtpci_ops(hw);
1921 set_rxtx_funcs(eth_dev);
1925 ret = virtio_dev_devargs_parse(eth_dev->device->devargs,
1926 NULL, &speed, &vectorized);
1931 * Pass the information to the rte_eth_dev_close() that it should also
1932 * release the private port resources.
1934 eth_dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
1936 /* Allocate memory for storing MAC addresses */
1937 eth_dev->data->mac_addrs = rte_zmalloc("virtio",
1938 VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN, 0);
1939 if (eth_dev->data->mac_addrs == NULL) {
1941 "Failed to allocate %d bytes needed to store MAC addresses",
1942 VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN);
1946 hw->port_id = eth_dev->data->port_id;
1947 /* For virtio_user case the hw->virtio_user_dev is populated by
1948 * virtio_user_eth_dev_alloc() before eth_virtio_dev_init() is called.
1950 if (!hw->virtio_user_dev) {
1951 ret = vtpci_init(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1953 goto err_vtpci_init;
1956 rte_spinlock_init(&hw->state_lock);
1958 /* reset device and negotiate default features */
1959 ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1961 goto err_virtio_init;
1964 if (!vtpci_packed_queue(hw)) {
1967 #if !defined(CC_AVX512_SUPPORT)
1969 "building environment do not support packed ring vectorized");
1982 if (!hw->virtio_user_dev) {
1983 rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(eth_dev));
1985 rte_pci_ioport_unmap(VTPCI_IO(hw));
1988 rte_free(eth_dev->data->mac_addrs);
1989 eth_dev->data->mac_addrs = NULL;
1994 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1996 PMD_INIT_FUNC_TRACE();
1998 if (rte_eal_process_type() == RTE_PROC_SECONDARY)
2001 virtio_dev_stop(eth_dev);
2002 virtio_dev_close(eth_dev);
2004 eth_dev->dev_ops = NULL;
2005 eth_dev->tx_pkt_burst = NULL;
2006 eth_dev->rx_pkt_burst = NULL;
2008 PMD_INIT_LOG(DEBUG, "dev_uninit completed");
2014 static int vdpa_check_handler(__rte_unused const char *key,
2015 const char *value, void *ret_val)
2017 if (strcmp(value, "1") == 0)
2018 *(int *)ret_val = 1;
2020 *(int *)ret_val = 0;
2027 virtio_dev_speed_capa_get(uint32_t speed)
2030 case ETH_SPEED_NUM_10G:
2031 return ETH_LINK_SPEED_10G;
2032 case ETH_SPEED_NUM_20G:
2033 return ETH_LINK_SPEED_20G;
2034 case ETH_SPEED_NUM_25G:
2035 return ETH_LINK_SPEED_25G;
2036 case ETH_SPEED_NUM_40G:
2037 return ETH_LINK_SPEED_40G;
2038 case ETH_SPEED_NUM_50G:
2039 return ETH_LINK_SPEED_50G;
2040 case ETH_SPEED_NUM_56G:
2041 return ETH_LINK_SPEED_56G;
2042 case ETH_SPEED_NUM_100G:
2043 return ETH_LINK_SPEED_100G;
2049 static int vectorized_check_handler(__rte_unused const char *key,
2050 const char *value, void *ret_val)
2052 if (strcmp(value, "1") == 0)
2053 *(int *)ret_val = 1;
2055 *(int *)ret_val = 0;
2060 #define VIRTIO_ARG_SPEED "speed"
2061 #define VIRTIO_ARG_VDPA "vdpa"
2062 #define VIRTIO_ARG_VECTORIZED "vectorized"
2066 link_speed_handler(const char *key __rte_unused,
2067 const char *value, void *ret_val)
2070 if (!value || !ret_val)
2072 val = strtoul(value, NULL, 0);
2073 /* validate input */
2074 if (virtio_dev_speed_capa_get(val) == 0)
2076 *(uint32_t *)ret_val = val;
2083 virtio_dev_devargs_parse(struct rte_devargs *devargs, int *vdpa,
2084 uint32_t *speed, int *vectorized)
2086 struct rte_kvargs *kvlist;
2089 if (devargs == NULL)
2092 kvlist = rte_kvargs_parse(devargs->args, NULL);
2093 if (kvlist == NULL) {
2094 PMD_INIT_LOG(ERR, "error when parsing param");
2097 if (vdpa && rte_kvargs_count(kvlist, VIRTIO_ARG_VDPA) == 1) {
2098 /* vdpa mode selected when there's a key-value pair:
2101 ret = rte_kvargs_process(kvlist, VIRTIO_ARG_VDPA,
2102 vdpa_check_handler, vdpa);
2104 PMD_INIT_LOG(ERR, "Failed to parse %s",
2109 if (speed && rte_kvargs_count(kvlist, VIRTIO_ARG_SPEED) == 1) {
2110 ret = rte_kvargs_process(kvlist,
2112 link_speed_handler, speed);
2114 PMD_INIT_LOG(ERR, "Failed to parse %s",
2121 rte_kvargs_count(kvlist, VIRTIO_ARG_VECTORIZED) == 1) {
2122 ret = rte_kvargs_process(kvlist,
2123 VIRTIO_ARG_VECTORIZED,
2124 vectorized_check_handler, vectorized);
2126 PMD_INIT_LOG(ERR, "Failed to parse %s",
2127 VIRTIO_ARG_VECTORIZED);
2133 rte_kvargs_free(kvlist);
2137 static int eth_virtio_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2138 struct rte_pci_device *pci_dev)
2143 ret = virtio_dev_devargs_parse(pci_dev->device.devargs, &vdpa, NULL,
2146 PMD_INIT_LOG(ERR, "devargs parsing is failed");
2149 /* virtio pmd skips probe if device needs to work in vdpa mode */
2153 return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct virtio_hw),
2154 eth_virtio_dev_init);
2157 static int eth_virtio_pci_remove(struct rte_pci_device *pci_dev)
2161 ret = rte_eth_dev_pci_generic_remove(pci_dev, eth_virtio_dev_uninit);
2162 /* Port has already been released by close. */
2168 static struct rte_pci_driver rte_virtio_pmd = {
2170 .name = "net_virtio",
2172 .id_table = pci_id_virtio_map,
2174 .probe = eth_virtio_pci_probe,
2175 .remove = eth_virtio_pci_remove,
2178 RTE_INIT(rte_virtio_pmd_init)
2180 rte_eal_iopl_init();
2181 rte_pci_register(&rte_virtio_pmd);
2185 rx_offload_enabled(struct virtio_hw *hw)
2187 return vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM) ||
2188 vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2189 vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6);
2193 tx_offload_enabled(struct virtio_hw *hw)
2195 return vtpci_with_feature(hw, VIRTIO_NET_F_CSUM) ||
2196 vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO4) ||
2197 vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO6);
2201 * Configure virtio device
2202 * It returns 0 on success.
2205 virtio_dev_configure(struct rte_eth_dev *dev)
2207 const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2208 const struct rte_eth_txmode *txmode = &dev->data->dev_conf.txmode;
2209 struct virtio_hw *hw = dev->data->dev_private;
2210 uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
2212 uint64_t rx_offloads = rxmode->offloads;
2213 uint64_t tx_offloads = txmode->offloads;
2214 uint64_t req_features;
2217 PMD_INIT_LOG(DEBUG, "configure");
2218 req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
2220 if (rxmode->mq_mode != ETH_MQ_RX_NONE) {
2222 "Unsupported Rx multi queue mode %d",
2227 if (txmode->mq_mode != ETH_MQ_TX_NONE) {
2229 "Unsupported Tx multi queue mode %d",
2234 if (dev->data->dev_conf.intr_conf.rxq) {
2235 ret = virtio_init_device(dev, hw->req_guest_features);
2240 if (rxmode->max_rx_pkt_len > hw->max_mtu + ether_hdr_len)
2241 req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
2243 if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2244 DEV_RX_OFFLOAD_TCP_CKSUM))
2245 req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
2247 if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO)
2249 (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2250 (1ULL << VIRTIO_NET_F_GUEST_TSO6);
2252 if (tx_offloads & (DEV_TX_OFFLOAD_UDP_CKSUM |
2253 DEV_TX_OFFLOAD_TCP_CKSUM))
2254 req_features |= (1ULL << VIRTIO_NET_F_CSUM);
2256 if (tx_offloads & DEV_TX_OFFLOAD_TCP_TSO)
2258 (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2259 (1ULL << VIRTIO_NET_F_HOST_TSO6);
2261 /* if request features changed, reinit the device */
2262 if (req_features != hw->req_guest_features) {
2263 ret = virtio_init_device(dev, req_features);
2268 if ((rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2269 DEV_RX_OFFLOAD_TCP_CKSUM)) &&
2270 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
2272 "rx checksum not available on this host");
2276 if ((rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) &&
2277 (!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2278 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6))) {
2280 "Large Receive Offload not available on this host");
2284 /* start control queue */
2285 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
2286 virtio_dev_cq_start(dev);
2288 if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
2291 if ((rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
2292 && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2294 "vlan filtering not available on this host");
2298 hw->has_tx_offload = tx_offload_enabled(hw);
2299 hw->has_rx_offload = rx_offload_enabled(hw);
2301 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2302 /* Enable vector (0) for Link State Intrerrupt */
2303 if (VTPCI_OPS(hw)->set_config_irq(hw, 0) ==
2304 VIRTIO_MSI_NO_VECTOR) {
2305 PMD_DRV_LOG(ERR, "failed to set config vector");
2309 if (vtpci_packed_queue(hw)) {
2310 #if defined(RTE_ARCH_X86_64) && defined(CC_AVX512_SUPPORT)
2311 if ((hw->use_vec_rx || hw->use_vec_tx) &&
2312 (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX512F) ||
2313 !vtpci_with_feature(hw, VIRTIO_F_IN_ORDER) ||
2314 !vtpci_with_feature(hw, VIRTIO_F_VERSION_1))) {
2316 "disabled packed ring vectorized path for requirements not met");
2325 if (hw->use_vec_rx) {
2326 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2328 "disabled packed ring vectorized rx for mrg_rxbuf enabled");
2332 if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) {
2334 "disabled packed ring vectorized rx for TCP_LRO enabled");
2339 if (vtpci_with_feature(hw, VIRTIO_F_IN_ORDER)) {
2340 hw->use_inorder_tx = 1;
2341 hw->use_inorder_rx = 1;
2345 if (hw->use_vec_rx) {
2346 #if defined RTE_ARCH_ARM64 || defined RTE_ARCH_ARM
2347 if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
2349 "disabled split ring vectorized path for requirement not met");
2353 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2355 "disabled split ring vectorized rx for mrg_rxbuf enabled");
2359 if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2360 DEV_RX_OFFLOAD_TCP_CKSUM |
2361 DEV_RX_OFFLOAD_TCP_LRO |
2362 DEV_RX_OFFLOAD_VLAN_STRIP)) {
2364 "disabled split ring vectorized rx for offloading enabled");
2375 virtio_dev_start(struct rte_eth_dev *dev)
2377 uint16_t nb_queues, i;
2378 struct virtnet_rx *rxvq;
2379 struct virtnet_tx *txvq __rte_unused;
2380 struct virtio_hw *hw = dev->data->dev_private;
2383 /* Finish the initialization of the queues */
2384 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2385 ret = virtio_dev_rx_queue_setup_finish(dev, i);
2389 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2390 ret = virtio_dev_tx_queue_setup_finish(dev, i);
2395 /* check if lsc interrupt feature is enabled */
2396 if (dev->data->dev_conf.intr_conf.lsc) {
2397 if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
2398 PMD_DRV_LOG(ERR, "link status not supported by host");
2403 /* Enable uio/vfio intr/eventfd mapping: althrough we already did that
2404 * in device configure, but it could be unmapped when device is
2407 if (dev->data->dev_conf.intr_conf.lsc ||
2408 dev->data->dev_conf.intr_conf.rxq) {
2409 virtio_intr_disable(dev);
2411 /* Setup interrupt callback */
2412 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2413 rte_intr_callback_register(dev->intr_handle,
2414 virtio_interrupt_handler,
2417 if (virtio_intr_enable(dev) < 0) {
2418 PMD_DRV_LOG(ERR, "interrupt enable failed");
2423 /*Notify the backend
2424 *Otherwise the tap backend might already stop its queue due to fullness.
2425 *vhost backend will have no chance to be waked up
2427 nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
2428 if (hw->max_queue_pairs > 1) {
2429 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
2433 PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
2435 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2436 rxvq = dev->data->rx_queues[i];
2437 /* Flush the old packets */
2438 virtqueue_rxvq_flush(rxvq->vq);
2439 virtqueue_notify(rxvq->vq);
2442 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2443 txvq = dev->data->tx_queues[i];
2444 virtqueue_notify(txvq->vq);
2447 PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
2449 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2450 rxvq = dev->data->rx_queues[i];
2451 VIRTQUEUE_DUMP(rxvq->vq);
2454 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2455 txvq = dev->data->tx_queues[i];
2456 VIRTQUEUE_DUMP(txvq->vq);
2459 set_rxtx_funcs(dev);
2462 /* Initialize Link state */
2463 virtio_dev_link_update(dev, 0);
2468 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
2470 struct virtio_hw *hw = dev->data->dev_private;
2471 uint16_t nr_vq = virtio_get_nr_vq(hw);
2472 const char *type __rte_unused;
2473 unsigned int i, mbuf_num = 0;
2474 struct virtqueue *vq;
2475 struct rte_mbuf *buf;
2478 if (hw->vqs == NULL)
2481 for (i = 0; i < nr_vq; i++) {
2486 queue_type = virtio_get_queue_type(hw, i);
2487 if (queue_type == VTNET_RQ)
2489 else if (queue_type == VTNET_TQ)
2495 "Before freeing %s[%d] used and unused buf",
2499 while ((buf = virtqueue_detach_unused(vq)) != NULL) {
2500 rte_pktmbuf_free(buf);
2505 "After freeing %s[%d] used and unused buf",
2510 PMD_INIT_LOG(DEBUG, "%d mbufs freed", mbuf_num);
2514 * Stop device: disable interrupt and mark link down
2517 virtio_dev_stop(struct rte_eth_dev *dev)
2519 struct virtio_hw *hw = dev->data->dev_private;
2520 struct rte_eth_link link;
2521 struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
2523 PMD_INIT_LOG(DEBUG, "stop");
2525 rte_spinlock_lock(&hw->state_lock);
2528 hw->started = false;
2530 if (intr_conf->lsc || intr_conf->rxq) {
2531 virtio_intr_disable(dev);
2533 /* Reset interrupt callback */
2534 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
2535 rte_intr_callback_unregister(dev->intr_handle,
2536 virtio_interrupt_handler,
2541 memset(&link, 0, sizeof(link));
2542 rte_eth_linkstatus_set(dev, &link);
2544 rte_spinlock_unlock(&hw->state_lock);
2548 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
2550 struct rte_eth_link link;
2552 struct virtio_hw *hw = dev->data->dev_private;
2554 memset(&link, 0, sizeof(link));
2555 link.link_duplex = hw->duplex;
2556 link.link_speed = hw->speed;
2557 link.link_autoneg = ETH_LINK_AUTONEG;
2560 link.link_status = ETH_LINK_DOWN;
2561 } else if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
2562 PMD_INIT_LOG(DEBUG, "Get link status from hw");
2563 vtpci_read_dev_config(hw,
2564 offsetof(struct virtio_net_config, status),
2565 &status, sizeof(status));
2566 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
2567 link.link_status = ETH_LINK_DOWN;
2568 PMD_INIT_LOG(DEBUG, "Port %d is down",
2569 dev->data->port_id);
2571 link.link_status = ETH_LINK_UP;
2572 PMD_INIT_LOG(DEBUG, "Port %d is up",
2573 dev->data->port_id);
2576 link.link_status = ETH_LINK_UP;
2579 return rte_eth_linkstatus_set(dev, &link);
2583 virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
2585 const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2586 struct virtio_hw *hw = dev->data->dev_private;
2587 uint64_t offloads = rxmode->offloads;
2589 if (mask & ETH_VLAN_FILTER_MASK) {
2590 if ((offloads & DEV_RX_OFFLOAD_VLAN_FILTER) &&
2591 !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2594 "vlan filtering not available on this host");
2600 if (mask & ETH_VLAN_STRIP_MASK)
2601 hw->vlan_strip = !!(offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
2607 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2609 uint64_t tso_mask, host_features;
2610 struct virtio_hw *hw = dev->data->dev_private;
2611 dev_info->speed_capa = virtio_dev_speed_capa_get(hw->speed);
2613 dev_info->max_rx_queues =
2614 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
2615 dev_info->max_tx_queues =
2616 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
2617 dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
2618 dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
2619 dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
2621 host_features = VTPCI_OPS(hw)->get_features(hw);
2622 dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
2623 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_JUMBO_FRAME;
2624 if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
2625 dev_info->rx_offload_capa |=
2626 DEV_RX_OFFLOAD_TCP_CKSUM |
2627 DEV_RX_OFFLOAD_UDP_CKSUM;
2629 if (host_features & (1ULL << VIRTIO_NET_F_CTRL_VLAN))
2630 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_VLAN_FILTER;
2631 tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2632 (1ULL << VIRTIO_NET_F_GUEST_TSO6);
2633 if ((host_features & tso_mask) == tso_mask)
2634 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
2636 dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
2637 DEV_TX_OFFLOAD_VLAN_INSERT;
2638 if (host_features & (1ULL << VIRTIO_NET_F_CSUM)) {
2639 dev_info->tx_offload_capa |=
2640 DEV_TX_OFFLOAD_UDP_CKSUM |
2641 DEV_TX_OFFLOAD_TCP_CKSUM;
2643 tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2644 (1ULL << VIRTIO_NET_F_HOST_TSO6);
2645 if ((host_features & tso_mask) == tso_mask)
2646 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
2652 * It enables testpmd to collect per queue stats.
2655 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
2656 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
2657 __rte_unused uint8_t is_rx)
2662 RTE_PMD_EXPORT_NAME(net_virtio, __COUNTER__);
2663 RTE_PMD_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map);
2664 RTE_PMD_REGISTER_KMOD_DEP(net_virtio, "* igb_uio | uio_pci_generic | vfio-pci");
2666 RTE_INIT(virtio_init_log)
2668 virtio_logtype_init = rte_log_register("pmd.net.virtio.init");
2669 if (virtio_logtype_init >= 0)
2670 rte_log_set_level(virtio_logtype_init, RTE_LOG_NOTICE);
2671 virtio_logtype_driver = rte_log_register("pmd.net.virtio.driver");
2672 if (virtio_logtype_driver >= 0)
2673 rte_log_set_level(virtio_logtype_driver, RTE_LOG_NOTICE);