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 static void virtio_notify_peers(struct rte_eth_dev *dev);
89 static void virtio_ack_link_announce(struct rte_eth_dev *dev);
92 * The set of PCI devices this driver supports
94 static const struct rte_pci_id pci_id_virtio_map[] = {
95 { RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_LEGACY_DEVICEID_NET) },
96 { RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_MODERN_DEVICEID_NET) },
97 { .vendor_id = 0, /* sentinel */ },
100 struct rte_virtio_xstats_name_off {
101 char name[RTE_ETH_XSTATS_NAME_SIZE];
105 /* [rt]x_qX_ is prepended to the name string here */
106 static const struct rte_virtio_xstats_name_off rte_virtio_rxq_stat_strings[] = {
107 {"good_packets", offsetof(struct virtnet_rx, stats.packets)},
108 {"good_bytes", offsetof(struct virtnet_rx, stats.bytes)},
109 {"errors", offsetof(struct virtnet_rx, stats.errors)},
110 {"multicast_packets", offsetof(struct virtnet_rx, stats.multicast)},
111 {"broadcast_packets", offsetof(struct virtnet_rx, stats.broadcast)},
112 {"undersize_packets", offsetof(struct virtnet_rx, stats.size_bins[0])},
113 {"size_64_packets", offsetof(struct virtnet_rx, stats.size_bins[1])},
114 {"size_65_127_packets", offsetof(struct virtnet_rx, stats.size_bins[2])},
115 {"size_128_255_packets", offsetof(struct virtnet_rx, stats.size_bins[3])},
116 {"size_256_511_packets", offsetof(struct virtnet_rx, stats.size_bins[4])},
117 {"size_512_1023_packets", offsetof(struct virtnet_rx, stats.size_bins[5])},
118 {"size_1024_1518_packets", offsetof(struct virtnet_rx, stats.size_bins[6])},
119 {"size_1519_max_packets", offsetof(struct virtnet_rx, stats.size_bins[7])},
122 /* [rt]x_qX_ is prepended to the name string here */
123 static const struct rte_virtio_xstats_name_off rte_virtio_txq_stat_strings[] = {
124 {"good_packets", offsetof(struct virtnet_tx, stats.packets)},
125 {"good_bytes", offsetof(struct virtnet_tx, stats.bytes)},
126 {"multicast_packets", offsetof(struct virtnet_tx, stats.multicast)},
127 {"broadcast_packets", offsetof(struct virtnet_tx, stats.broadcast)},
128 {"undersize_packets", offsetof(struct virtnet_tx, stats.size_bins[0])},
129 {"size_64_packets", offsetof(struct virtnet_tx, stats.size_bins[1])},
130 {"size_65_127_packets", offsetof(struct virtnet_tx, stats.size_bins[2])},
131 {"size_128_255_packets", offsetof(struct virtnet_tx, stats.size_bins[3])},
132 {"size_256_511_packets", offsetof(struct virtnet_tx, stats.size_bins[4])},
133 {"size_512_1023_packets", offsetof(struct virtnet_tx, stats.size_bins[5])},
134 {"size_1024_1518_packets", offsetof(struct virtnet_tx, stats.size_bins[6])},
135 {"size_1519_max_packets", offsetof(struct virtnet_tx, stats.size_bins[7])},
138 #define VIRTIO_NB_RXQ_XSTATS (sizeof(rte_virtio_rxq_stat_strings) / \
139 sizeof(rte_virtio_rxq_stat_strings[0]))
140 #define VIRTIO_NB_TXQ_XSTATS (sizeof(rte_virtio_txq_stat_strings) / \
141 sizeof(rte_virtio_txq_stat_strings[0]))
143 struct virtio_hw_internal virtio_hw_internal[RTE_MAX_ETHPORTS];
145 static struct virtio_pmd_ctrl *
146 virtio_send_command_packed(struct virtnet_ctl *cvq,
147 struct virtio_pmd_ctrl *ctrl,
148 int *dlen, int pkt_num)
150 struct virtqueue *vq = cvq->vq;
152 struct vring_packed_desc *desc = vq->vq_packed.ring.desc;
153 struct virtio_pmd_ctrl *result;
160 * Format is enforced in qemu code:
161 * One TX packet for header;
162 * At least one TX packet per argument;
163 * One RX packet for ACK.
165 head = vq->vq_avail_idx;
166 flags = vq->vq_packed.cached_flags;
167 desc[head].addr = cvq->virtio_net_hdr_mem;
168 desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
171 if (++vq->vq_avail_idx >= vq->vq_nentries) {
172 vq->vq_avail_idx -= vq->vq_nentries;
173 vq->vq_packed.cached_flags ^= VRING_PACKED_DESC_F_AVAIL_USED;
176 for (k = 0; k < pkt_num; k++) {
177 desc[vq->vq_avail_idx].addr = cvq->virtio_net_hdr_mem
178 + sizeof(struct virtio_net_ctrl_hdr)
179 + sizeof(ctrl->status) + sizeof(uint8_t) * sum;
180 desc[vq->vq_avail_idx].len = dlen[k];
181 desc[vq->vq_avail_idx].flags = VRING_DESC_F_NEXT |
182 vq->vq_packed.cached_flags;
186 if (++vq->vq_avail_idx >= vq->vq_nentries) {
187 vq->vq_avail_idx -= vq->vq_nentries;
188 vq->vq_packed.cached_flags ^=
189 VRING_PACKED_DESC_F_AVAIL_USED;
193 desc[vq->vq_avail_idx].addr = cvq->virtio_net_hdr_mem
194 + sizeof(struct virtio_net_ctrl_hdr);
195 desc[vq->vq_avail_idx].len = sizeof(ctrl->status);
196 desc[vq->vq_avail_idx].flags = VRING_DESC_F_WRITE |
197 vq->vq_packed.cached_flags;
200 if (++vq->vq_avail_idx >= vq->vq_nentries) {
201 vq->vq_avail_idx -= vq->vq_nentries;
202 vq->vq_packed.cached_flags ^= VRING_PACKED_DESC_F_AVAIL_USED;
205 virtio_wmb(vq->hw->weak_barriers);
206 desc[head].flags = VRING_DESC_F_NEXT | flags;
208 virtio_wmb(vq->hw->weak_barriers);
209 virtqueue_notify(vq);
211 /* wait for used descriptors in virtqueue */
212 while (!desc_is_used(&desc[head], vq))
215 virtio_rmb(vq->hw->weak_barriers);
217 /* now get used descriptors */
218 vq->vq_free_cnt += nb_descs;
219 vq->vq_used_cons_idx += nb_descs;
220 if (vq->vq_used_cons_idx >= vq->vq_nentries) {
221 vq->vq_used_cons_idx -= vq->vq_nentries;
222 vq->vq_packed.used_wrap_counter ^= 1;
225 PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\n"
226 "vq->vq_avail_idx=%d\n"
227 "vq->vq_used_cons_idx=%d\n"
228 "vq->vq_packed.cached_flags=0x%x\n"
229 "vq->vq_packed.used_wrap_counter=%d\n",
232 vq->vq_used_cons_idx,
233 vq->vq_packed.cached_flags,
234 vq->vq_packed.used_wrap_counter);
236 result = cvq->virtio_net_hdr_mz->addr;
240 static struct virtio_pmd_ctrl *
241 virtio_send_command_split(struct virtnet_ctl *cvq,
242 struct virtio_pmd_ctrl *ctrl,
243 int *dlen, int pkt_num)
245 struct virtio_pmd_ctrl *result;
246 struct virtqueue *vq = cvq->vq;
250 head = vq->vq_desc_head_idx;
253 * Format is enforced in qemu code:
254 * One TX packet for header;
255 * At least one TX packet per argument;
256 * One RX packet for ACK.
258 vq->vq_split.ring.desc[head].flags = VRING_DESC_F_NEXT;
259 vq->vq_split.ring.desc[head].addr = cvq->virtio_net_hdr_mem;
260 vq->vq_split.ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
262 i = vq->vq_split.ring.desc[head].next;
264 for (k = 0; k < pkt_num; k++) {
265 vq->vq_split.ring.desc[i].flags = VRING_DESC_F_NEXT;
266 vq->vq_split.ring.desc[i].addr = cvq->virtio_net_hdr_mem
267 + sizeof(struct virtio_net_ctrl_hdr)
268 + sizeof(ctrl->status) + sizeof(uint8_t)*sum;
269 vq->vq_split.ring.desc[i].len = dlen[k];
272 i = vq->vq_split.ring.desc[i].next;
275 vq->vq_split.ring.desc[i].flags = VRING_DESC_F_WRITE;
276 vq->vq_split.ring.desc[i].addr = cvq->virtio_net_hdr_mem
277 + sizeof(struct virtio_net_ctrl_hdr);
278 vq->vq_split.ring.desc[i].len = sizeof(ctrl->status);
281 vq->vq_desc_head_idx = vq->vq_split.ring.desc[i].next;
283 vq_update_avail_ring(vq, head);
284 vq_update_avail_idx(vq);
286 PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
288 virtqueue_notify(vq);
290 while (virtqueue_nused(vq) == 0)
293 while (virtqueue_nused(vq)) {
294 uint32_t idx, desc_idx, used_idx;
295 struct vring_used_elem *uep;
297 used_idx = (uint32_t)(vq->vq_used_cons_idx
298 & (vq->vq_nentries - 1));
299 uep = &vq->vq_split.ring.used->ring[used_idx];
300 idx = (uint32_t) uep->id;
303 while (vq->vq_split.ring.desc[desc_idx].flags &
305 desc_idx = vq->vq_split.ring.desc[desc_idx].next;
309 vq->vq_split.ring.desc[desc_idx].next = vq->vq_desc_head_idx;
310 vq->vq_desc_head_idx = idx;
312 vq->vq_used_cons_idx++;
316 PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
317 vq->vq_free_cnt, vq->vq_desc_head_idx);
319 result = cvq->virtio_net_hdr_mz->addr;
324 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
325 int *dlen, int pkt_num)
327 virtio_net_ctrl_ack status = ~0;
328 struct virtio_pmd_ctrl *result;
329 struct virtqueue *vq;
331 ctrl->status = status;
333 if (!cvq || !cvq->vq) {
334 PMD_INIT_LOG(ERR, "Control queue is not supported.");
338 rte_spinlock_lock(&cvq->lock);
341 PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
342 "vq->hw->cvq = %p vq = %p",
343 vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
345 if (vq->vq_free_cnt < pkt_num + 2 || pkt_num < 1) {
346 rte_spinlock_unlock(&cvq->lock);
350 memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
351 sizeof(struct virtio_pmd_ctrl));
353 if (vtpci_packed_queue(vq->hw))
354 result = virtio_send_command_packed(cvq, ctrl, dlen, pkt_num);
356 result = virtio_send_command_split(cvq, ctrl, dlen, pkt_num);
358 rte_spinlock_unlock(&cvq->lock);
359 return result->status;
363 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
365 struct virtio_hw *hw = dev->data->dev_private;
366 struct virtio_pmd_ctrl ctrl;
370 ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
371 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
372 memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
374 dlen[0] = sizeof(uint16_t);
376 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
378 PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
379 "failed, this is too late now...");
387 virtio_dev_queue_release(void *queue __rte_unused)
393 virtio_get_nr_vq(struct virtio_hw *hw)
395 uint16_t nr_vq = hw->max_queue_pairs * 2;
397 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
404 virtio_init_vring(struct virtqueue *vq)
406 int size = vq->vq_nentries;
407 uint8_t *ring_mem = vq->vq_ring_virt_mem;
409 PMD_INIT_FUNC_TRACE();
411 memset(ring_mem, 0, vq->vq_ring_size);
413 vq->vq_used_cons_idx = 0;
414 vq->vq_desc_head_idx = 0;
415 vq->vq_avail_idx = 0;
416 vq->vq_desc_tail_idx = (uint16_t)(vq->vq_nentries - 1);
417 vq->vq_free_cnt = vq->vq_nentries;
418 memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries);
419 if (vtpci_packed_queue(vq->hw)) {
420 vring_init_packed(&vq->vq_packed.ring, ring_mem,
421 VIRTIO_PCI_VRING_ALIGN, size);
422 vring_desc_init_packed(vq, size);
424 struct vring *vr = &vq->vq_split.ring;
426 vring_init_split(vr, ring_mem, VIRTIO_PCI_VRING_ALIGN, size);
427 vring_desc_init_split(vr->desc, size);
430 * Disable device(host) interrupting guest
432 virtqueue_disable_intr(vq);
436 virtio_init_queue(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx)
438 char vq_name[VIRTQUEUE_MAX_NAME_SZ];
439 char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
440 const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
441 unsigned int vq_size, size;
442 struct virtio_hw *hw = dev->data->dev_private;
443 struct virtnet_rx *rxvq = NULL;
444 struct virtnet_tx *txvq = NULL;
445 struct virtnet_ctl *cvq = NULL;
446 struct virtqueue *vq;
447 size_t sz_hdr_mz = 0;
448 void *sw_ring = NULL;
449 int queue_type = virtio_get_queue_type(hw, vtpci_queue_idx);
451 int numa_node = dev->device->numa_node;
453 PMD_INIT_LOG(INFO, "setting up queue: %u on NUMA node %d",
454 vtpci_queue_idx, numa_node);
457 * Read the virtqueue size from the Queue Size field
458 * Always power of 2 and if 0 virtqueue does not exist
460 vq_size = VTPCI_OPS(hw)->get_queue_num(hw, vtpci_queue_idx);
461 PMD_INIT_LOG(DEBUG, "vq_size: %u", vq_size);
463 PMD_INIT_LOG(ERR, "virtqueue does not exist");
467 if (!vtpci_packed_queue(hw) && !rte_is_power_of_2(vq_size)) {
468 PMD_INIT_LOG(ERR, "split virtqueue size is not power of 2");
472 snprintf(vq_name, sizeof(vq_name), "port%d_vq%d",
473 dev->data->port_id, vtpci_queue_idx);
475 size = RTE_ALIGN_CEIL(sizeof(*vq) +
476 vq_size * sizeof(struct vq_desc_extra),
477 RTE_CACHE_LINE_SIZE);
478 if (queue_type == VTNET_TQ) {
480 * For each xmit packet, allocate a virtio_net_hdr
481 * and indirect ring elements
483 sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
484 } else if (queue_type == VTNET_CQ) {
485 /* Allocate a page for control vq command, data and status */
486 sz_hdr_mz = PAGE_SIZE;
489 vq = rte_zmalloc_socket(vq_name, size, RTE_CACHE_LINE_SIZE,
492 PMD_INIT_LOG(ERR, "can not allocate vq");
495 hw->vqs[vtpci_queue_idx] = vq;
498 vq->vq_queue_index = vtpci_queue_idx;
499 vq->vq_nentries = vq_size;
500 if (vtpci_packed_queue(hw)) {
501 vq->vq_packed.used_wrap_counter = 1;
502 vq->vq_packed.cached_flags = VRING_PACKED_DESC_F_AVAIL;
503 vq->vq_packed.event_flags_shadow = 0;
504 if (queue_type == VTNET_RQ)
505 vq->vq_packed.cached_flags |= VRING_DESC_F_WRITE;
509 * Reserve a memzone for vring elements
511 size = vring_size(hw, vq_size, VIRTIO_PCI_VRING_ALIGN);
512 vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
513 PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
514 size, vq->vq_ring_size);
516 mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
517 numa_node, RTE_MEMZONE_IOVA_CONTIG,
518 VIRTIO_PCI_VRING_ALIGN);
520 if (rte_errno == EEXIST)
521 mz = rte_memzone_lookup(vq_name);
528 memset(mz->addr, 0, mz->len);
530 vq->vq_ring_mem = mz->iova;
531 vq->vq_ring_virt_mem = mz->addr;
532 PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem: 0x%" PRIx64,
534 PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64,
535 (uint64_t)(uintptr_t)mz->addr);
537 virtio_init_vring(vq);
540 snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_vq%d_hdr",
541 dev->data->port_id, vtpci_queue_idx);
542 hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
543 numa_node, RTE_MEMZONE_IOVA_CONTIG,
544 RTE_CACHE_LINE_SIZE);
545 if (hdr_mz == NULL) {
546 if (rte_errno == EEXIST)
547 hdr_mz = rte_memzone_lookup(vq_hdr_name);
548 if (hdr_mz == NULL) {
555 if (queue_type == VTNET_RQ) {
556 size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
557 sizeof(vq->sw_ring[0]);
559 sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
560 RTE_CACHE_LINE_SIZE, numa_node);
562 PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
567 vq->sw_ring = sw_ring;
570 rxvq->port_id = dev->data->port_id;
572 } else if (queue_type == VTNET_TQ) {
575 txvq->port_id = dev->data->port_id;
577 txvq->virtio_net_hdr_mz = hdr_mz;
578 txvq->virtio_net_hdr_mem = hdr_mz->iova;
579 } else if (queue_type == VTNET_CQ) {
583 cvq->virtio_net_hdr_mz = hdr_mz;
584 cvq->virtio_net_hdr_mem = hdr_mz->iova;
585 memset(cvq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
590 /* For virtio_user case (that is when hw->virtio_user_dev is not NULL),
591 * we use virtual address. And we need properly set _offset_, please see
592 * VIRTIO_MBUF_DATA_DMA_ADDR in virtqueue.h for more information.
594 if (!hw->virtio_user_dev)
595 vq->offset = offsetof(struct rte_mbuf, buf_iova);
597 vq->vq_ring_mem = (uintptr_t)mz->addr;
598 vq->offset = offsetof(struct rte_mbuf, buf_addr);
599 if (queue_type == VTNET_TQ)
600 txvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
601 else if (queue_type == VTNET_CQ)
602 cvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
605 if (queue_type == VTNET_TQ) {
606 struct virtio_tx_region *txr;
610 memset(txr, 0, vq_size * sizeof(*txr));
611 for (i = 0; i < vq_size; i++) {
612 struct vring_desc *start_dp = txr[i].tx_indir;
614 /* first indirect descriptor is always the tx header */
615 if (!vtpci_packed_queue(hw)) {
616 vring_desc_init_split(start_dp,
617 RTE_DIM(txr[i].tx_indir));
618 start_dp->addr = txvq->virtio_net_hdr_mem
620 + offsetof(struct virtio_tx_region,
622 start_dp->len = hw->vtnet_hdr_size;
623 start_dp->flags = VRING_DESC_F_NEXT;
628 if (VTPCI_OPS(hw)->setup_queue(hw, vq) < 0) {
629 PMD_INIT_LOG(ERR, "setup_queue failed");
637 rte_memzone_free(hdr_mz);
638 rte_memzone_free(mz);
645 virtio_free_queues(struct virtio_hw *hw)
647 uint16_t nr_vq = virtio_get_nr_vq(hw);
648 struct virtqueue *vq;
655 for (i = 0; i < nr_vq; i++) {
660 queue_type = virtio_get_queue_type(hw, i);
661 if (queue_type == VTNET_RQ) {
662 rte_free(vq->sw_ring);
663 rte_memzone_free(vq->rxq.mz);
664 } else if (queue_type == VTNET_TQ) {
665 rte_memzone_free(vq->txq.mz);
666 rte_memzone_free(vq->txq.virtio_net_hdr_mz);
668 rte_memzone_free(vq->cq.mz);
669 rte_memzone_free(vq->cq.virtio_net_hdr_mz);
681 virtio_alloc_queues(struct rte_eth_dev *dev)
683 struct virtio_hw *hw = dev->data->dev_private;
684 uint16_t nr_vq = virtio_get_nr_vq(hw);
688 hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
690 PMD_INIT_LOG(ERR, "failed to allocate vqs");
694 for (i = 0; i < nr_vq; i++) {
695 ret = virtio_init_queue(dev, i);
697 virtio_free_queues(hw);
705 static void virtio_queues_unbind_intr(struct rte_eth_dev *dev);
708 virtio_dev_close(struct rte_eth_dev *dev)
710 struct virtio_hw *hw = dev->data->dev_private;
711 struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
713 PMD_INIT_LOG(DEBUG, "virtio_dev_close");
714 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
722 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
723 VTPCI_OPS(hw)->set_config_irq(hw, VIRTIO_MSI_NO_VECTOR);
725 virtio_queues_unbind_intr(dev);
727 if (intr_conf->lsc || intr_conf->rxq) {
728 virtio_intr_disable(dev);
729 rte_intr_efd_disable(dev->intr_handle);
730 rte_free(dev->intr_handle->intr_vec);
731 dev->intr_handle->intr_vec = NULL;
735 virtio_dev_free_mbufs(dev);
736 virtio_free_queues(hw);
738 #ifdef RTE_VIRTIO_USER
739 if (hw->virtio_user_dev)
740 virtio_user_dev_uninit(hw->virtio_user_dev);
744 rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(dev));
746 rte_pci_ioport_unmap(VTPCI_IO(hw));
753 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
755 struct virtio_hw *hw = dev->data->dev_private;
756 struct virtio_pmd_ctrl ctrl;
760 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
761 PMD_INIT_LOG(INFO, "host does not support rx control");
765 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
766 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
770 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
772 PMD_INIT_LOG(ERR, "Failed to enable promisc");
780 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
782 struct virtio_hw *hw = dev->data->dev_private;
783 struct virtio_pmd_ctrl ctrl;
787 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
788 PMD_INIT_LOG(INFO, "host does not support rx control");
792 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
793 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
797 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
799 PMD_INIT_LOG(ERR, "Failed to disable promisc");
807 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
809 struct virtio_hw *hw = dev->data->dev_private;
810 struct virtio_pmd_ctrl ctrl;
814 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
815 PMD_INIT_LOG(INFO, "host does not support rx control");
819 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
820 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
824 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
826 PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
834 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
836 struct virtio_hw *hw = dev->data->dev_private;
837 struct virtio_pmd_ctrl ctrl;
841 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
842 PMD_INIT_LOG(INFO, "host does not support rx control");
846 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
847 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
851 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
853 PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
860 #define VLAN_TAG_LEN 4 /* 802.3ac tag (not DMA'd) */
862 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
864 struct virtio_hw *hw = dev->data->dev_private;
865 uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
867 uint32_t frame_size = mtu + ether_hdr_len;
868 uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
870 max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
872 if (mtu < RTE_ETHER_MIN_MTU || frame_size > max_frame_size) {
873 PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
874 RTE_ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
881 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
883 struct virtio_hw *hw = dev->data->dev_private;
884 struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
885 struct virtqueue *vq = rxvq->vq;
887 virtqueue_enable_intr(vq);
888 virtio_mb(hw->weak_barriers);
893 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
895 struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
896 struct virtqueue *vq = rxvq->vq;
898 virtqueue_disable_intr(vq);
903 * dev_ops for virtio, bare necessities for basic operation
905 static const struct eth_dev_ops virtio_eth_dev_ops = {
906 .dev_configure = virtio_dev_configure,
907 .dev_start = virtio_dev_start,
908 .dev_stop = virtio_dev_stop,
909 .dev_close = virtio_dev_close,
910 .promiscuous_enable = virtio_dev_promiscuous_enable,
911 .promiscuous_disable = virtio_dev_promiscuous_disable,
912 .allmulticast_enable = virtio_dev_allmulticast_enable,
913 .allmulticast_disable = virtio_dev_allmulticast_disable,
914 .mtu_set = virtio_mtu_set,
915 .dev_infos_get = virtio_dev_info_get,
916 .stats_get = virtio_dev_stats_get,
917 .xstats_get = virtio_dev_xstats_get,
918 .xstats_get_names = virtio_dev_xstats_get_names,
919 .stats_reset = virtio_dev_stats_reset,
920 .xstats_reset = virtio_dev_stats_reset,
921 .link_update = virtio_dev_link_update,
922 .vlan_offload_set = virtio_dev_vlan_offload_set,
923 .rx_queue_setup = virtio_dev_rx_queue_setup,
924 .rx_queue_intr_enable = virtio_dev_rx_queue_intr_enable,
925 .rx_queue_intr_disable = virtio_dev_rx_queue_intr_disable,
926 .rx_queue_release = virtio_dev_queue_release,
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);
1361 if (hw->modern && !vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1363 "VIRTIO_F_VERSION_1 features is not enabled.");
1367 if (hw->modern || hw->virtio_user_dev) {
1368 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1369 if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1371 "failed to set FEATURES_OK status!");
1376 hw->req_guest_features = req_features;
1382 virtio_dev_pause(struct rte_eth_dev *dev)
1384 struct virtio_hw *hw = dev->data->dev_private;
1386 rte_spinlock_lock(&hw->state_lock);
1388 if (hw->started == 0) {
1389 /* Device is just stopped. */
1390 rte_spinlock_unlock(&hw->state_lock);
1395 * Prevent the worker threads from touching queues to avoid contention,
1396 * 1 ms should be enough for the ongoing Tx function to finish.
1403 * Recover hw state to let the worker threads continue.
1406 virtio_dev_resume(struct rte_eth_dev *dev)
1408 struct virtio_hw *hw = dev->data->dev_private;
1411 rte_spinlock_unlock(&hw->state_lock);
1415 * Should be called only after device is paused.
1418 virtio_inject_pkts(struct rte_eth_dev *dev, struct rte_mbuf **tx_pkts,
1421 struct virtio_hw *hw = dev->data->dev_private;
1422 struct virtnet_tx *txvq = dev->data->tx_queues[0];
1425 hw->inject_pkts = tx_pkts;
1426 ret = dev->tx_pkt_burst(txvq, tx_pkts, nb_pkts);
1427 hw->inject_pkts = NULL;
1433 virtio_notify_peers(struct rte_eth_dev *dev)
1435 struct virtio_hw *hw = dev->data->dev_private;
1436 struct virtnet_rx *rxvq;
1437 struct rte_mbuf *rarp_mbuf;
1439 if (!dev->data->rx_queues)
1442 rxvq = dev->data->rx_queues[0];
1446 rarp_mbuf = rte_net_make_rarp_packet(rxvq->mpool,
1447 (struct rte_ether_addr *)hw->mac_addr);
1448 if (rarp_mbuf == NULL) {
1449 PMD_DRV_LOG(ERR, "failed to make RARP packet.");
1453 /* If virtio port just stopped, no need to send RARP */
1454 if (virtio_dev_pause(dev) < 0) {
1455 rte_pktmbuf_free(rarp_mbuf);
1459 virtio_inject_pkts(dev, &rarp_mbuf, 1);
1460 virtio_dev_resume(dev);
1464 virtio_ack_link_announce(struct rte_eth_dev *dev)
1466 struct virtio_hw *hw = dev->data->dev_private;
1467 struct virtio_pmd_ctrl ctrl;
1469 ctrl.hdr.class = VIRTIO_NET_CTRL_ANNOUNCE;
1470 ctrl.hdr.cmd = VIRTIO_NET_CTRL_ANNOUNCE_ACK;
1472 virtio_send_command(hw->cvq, &ctrl, NULL, 0);
1476 * Process virtio config changed interrupt. Call the callback
1477 * if link state changed, generate gratuitous RARP packet if
1478 * the status indicates an ANNOUNCE.
1481 virtio_interrupt_handler(void *param)
1483 struct rte_eth_dev *dev = param;
1484 struct virtio_hw *hw = dev->data->dev_private;
1488 /* Read interrupt status which clears interrupt */
1489 isr = vtpci_isr(hw);
1490 PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1492 if (virtio_intr_unmask(dev) < 0)
1493 PMD_DRV_LOG(ERR, "interrupt enable failed");
1495 if (isr & VIRTIO_PCI_ISR_CONFIG) {
1496 if (virtio_dev_link_update(dev, 0) == 0)
1497 rte_eth_dev_callback_process(dev,
1498 RTE_ETH_EVENT_INTR_LSC,
1501 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1502 vtpci_read_dev_config(hw,
1503 offsetof(struct virtio_net_config, status),
1504 &status, sizeof(status));
1505 if (status & VIRTIO_NET_S_ANNOUNCE) {
1506 virtio_notify_peers(dev);
1508 virtio_ack_link_announce(dev);
1514 /* set rx and tx handlers according to what is supported */
1516 set_rxtx_funcs(struct rte_eth_dev *eth_dev)
1518 struct virtio_hw *hw = eth_dev->data->dev_private;
1520 eth_dev->tx_pkt_prepare = virtio_xmit_pkts_prepare;
1521 if (vtpci_packed_queue(hw)) {
1523 "virtio: using packed ring %s Tx path on port %u",
1524 hw->use_vec_tx ? "vectorized" : "standard",
1525 eth_dev->data->port_id);
1527 eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed_vec;
1529 eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed;
1531 if (hw->use_inorder_tx) {
1532 PMD_INIT_LOG(INFO, "virtio: using inorder Tx path on port %u",
1533 eth_dev->data->port_id);
1534 eth_dev->tx_pkt_burst = virtio_xmit_pkts_inorder;
1536 PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
1537 eth_dev->data->port_id);
1538 eth_dev->tx_pkt_burst = virtio_xmit_pkts;
1542 if (vtpci_packed_queue(hw)) {
1543 if (hw->use_vec_rx) {
1545 "virtio: using packed ring vectorized Rx path on port %u",
1546 eth_dev->data->port_id);
1547 eth_dev->rx_pkt_burst =
1548 &virtio_recv_pkts_packed_vec;
1549 } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1551 "virtio: using packed ring mergeable buffer Rx path on port %u",
1552 eth_dev->data->port_id);
1553 eth_dev->rx_pkt_burst =
1554 &virtio_recv_mergeable_pkts_packed;
1557 "virtio: using packed ring standard Rx path on port %u",
1558 eth_dev->data->port_id);
1559 eth_dev->rx_pkt_burst = &virtio_recv_pkts_packed;
1562 if (hw->use_vec_rx) {
1563 PMD_INIT_LOG(INFO, "virtio: using vectorized Rx path on port %u",
1564 eth_dev->data->port_id);
1565 eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1566 } else if (hw->use_inorder_rx) {
1568 "virtio: using inorder Rx path on port %u",
1569 eth_dev->data->port_id);
1570 eth_dev->rx_pkt_burst = &virtio_recv_pkts_inorder;
1571 } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1573 "virtio: using mergeable buffer Rx path on port %u",
1574 eth_dev->data->port_id);
1575 eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1577 PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
1578 eth_dev->data->port_id);
1579 eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1585 /* Only support 1:1 queue/interrupt mapping so far.
1586 * TODO: support n:1 queue/interrupt mapping when there are limited number of
1587 * interrupt vectors (<N+1).
1590 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1593 struct virtio_hw *hw = dev->data->dev_private;
1595 PMD_INIT_LOG(INFO, "queue/interrupt binding");
1596 for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1597 dev->intr_handle->intr_vec[i] = i + 1;
1598 if (VTPCI_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1599 VIRTIO_MSI_NO_VECTOR) {
1600 PMD_DRV_LOG(ERR, "failed to set queue vector");
1609 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1612 struct virtio_hw *hw = dev->data->dev_private;
1614 PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1615 for (i = 0; i < dev->data->nb_rx_queues; ++i)
1616 VTPCI_OPS(hw)->set_queue_irq(hw,
1617 hw->vqs[i * VTNET_CQ],
1618 VIRTIO_MSI_NO_VECTOR);
1622 virtio_configure_intr(struct rte_eth_dev *dev)
1624 struct virtio_hw *hw = dev->data->dev_private;
1626 if (!rte_intr_cap_multiple(dev->intr_handle)) {
1627 PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1631 if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1632 PMD_INIT_LOG(ERR, "Fail to create eventfd");
1636 if (!dev->intr_handle->intr_vec) {
1637 dev->intr_handle->intr_vec =
1638 rte_zmalloc("intr_vec",
1639 hw->max_queue_pairs * sizeof(int), 0);
1640 if (!dev->intr_handle->intr_vec) {
1641 PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1642 hw->max_queue_pairs);
1647 /* Re-register callback to update max_intr */
1648 rte_intr_callback_unregister(dev->intr_handle,
1649 virtio_interrupt_handler,
1651 rte_intr_callback_register(dev->intr_handle,
1652 virtio_interrupt_handler,
1655 /* DO NOT try to remove this! This function will enable msix, or QEMU
1656 * will encounter SIGSEGV when DRIVER_OK is sent.
1657 * And for legacy devices, this should be done before queue/vec binding
1658 * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1659 * (22) will be ignored.
1661 if (virtio_intr_enable(dev) < 0) {
1662 PMD_DRV_LOG(ERR, "interrupt enable failed");
1666 if (virtio_queues_bind_intr(dev) < 0) {
1667 PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1673 #define SPEED_UNKNOWN 0xffffffff
1674 #define DUPLEX_UNKNOWN 0xff
1675 /* reset device and renegotiate features if needed */
1677 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1679 struct virtio_hw *hw = eth_dev->data->dev_private;
1680 struct virtio_net_config *config;
1681 struct virtio_net_config local_config;
1682 struct rte_pci_device *pci_dev = NULL;
1685 /* Reset the device although not necessary at startup */
1689 virtio_dev_free_mbufs(eth_dev);
1690 virtio_free_queues(hw);
1693 /* Tell the host we've noticed this device. */
1694 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1696 /* Tell the host we've known how to drive the device. */
1697 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1698 if (virtio_negotiate_features(hw, req_features) < 0)
1701 hw->weak_barriers = !vtpci_with_feature(hw, VIRTIO_F_ORDER_PLATFORM);
1703 if (!hw->virtio_user_dev)
1704 pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1706 /* If host does not support both status and MSI-X then disable LSC */
1707 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS) &&
1708 hw->use_msix != VIRTIO_MSIX_NONE)
1709 eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1711 eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1713 /* Setting up rx_header size for the device */
1714 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1715 vtpci_with_feature(hw, VIRTIO_F_VERSION_1) ||
1716 vtpci_with_feature(hw, VIRTIO_F_RING_PACKED))
1717 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1719 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1721 /* Copy the permanent MAC address to: virtio_hw */
1722 virtio_get_hwaddr(hw);
1723 rte_ether_addr_copy((struct rte_ether_addr *)hw->mac_addr,
1724 ð_dev->data->mac_addrs[0]);
1726 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1727 hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1728 hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1730 if (hw->speed == SPEED_UNKNOWN) {
1731 if (vtpci_with_feature(hw, VIRTIO_NET_F_SPEED_DUPLEX)) {
1732 config = &local_config;
1733 vtpci_read_dev_config(hw,
1734 offsetof(struct virtio_net_config, speed),
1735 &config->speed, sizeof(config->speed));
1736 vtpci_read_dev_config(hw,
1737 offsetof(struct virtio_net_config, duplex),
1738 &config->duplex, sizeof(config->duplex));
1739 hw->speed = config->speed;
1740 hw->duplex = config->duplex;
1743 if (hw->speed == SPEED_UNKNOWN)
1744 hw->speed = ETH_SPEED_NUM_10G;
1745 if (hw->duplex == DUPLEX_UNKNOWN)
1746 hw->duplex = ETH_LINK_FULL_DUPLEX;
1747 PMD_INIT_LOG(DEBUG, "link speed = %d, duplex = %d",
1748 hw->speed, hw->duplex);
1749 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1750 config = &local_config;
1752 vtpci_read_dev_config(hw,
1753 offsetof(struct virtio_net_config, mac),
1754 &config->mac, sizeof(config->mac));
1756 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1757 vtpci_read_dev_config(hw,
1758 offsetof(struct virtio_net_config, status),
1759 &config->status, sizeof(config->status));
1762 "VIRTIO_NET_F_STATUS is not supported");
1766 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1767 vtpci_read_dev_config(hw,
1768 offsetof(struct virtio_net_config, max_virtqueue_pairs),
1769 &config->max_virtqueue_pairs,
1770 sizeof(config->max_virtqueue_pairs));
1773 "VIRTIO_NET_F_MQ is not supported");
1774 config->max_virtqueue_pairs = 1;
1777 hw->max_queue_pairs = config->max_virtqueue_pairs;
1779 if (vtpci_with_feature(hw, VIRTIO_NET_F_MTU)) {
1780 vtpci_read_dev_config(hw,
1781 offsetof(struct virtio_net_config, mtu),
1783 sizeof(config->mtu));
1786 * MTU value has already been checked at negotiation
1787 * time, but check again in case it has changed since
1788 * then, which should not happen.
1790 if (config->mtu < RTE_ETHER_MIN_MTU) {
1791 PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1796 hw->max_mtu = config->mtu;
1797 /* Set initial MTU to maximum one supported by vhost */
1798 eth_dev->data->mtu = config->mtu;
1801 hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1802 VLAN_TAG_LEN - hw->vtnet_hdr_size;
1805 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1806 config->max_virtqueue_pairs);
1807 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1809 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1810 config->mac[0], config->mac[1],
1811 config->mac[2], config->mac[3],
1812 config->mac[4], config->mac[5]);
1814 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1815 hw->max_queue_pairs = 1;
1816 hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1817 VLAN_TAG_LEN - hw->vtnet_hdr_size;
1820 ret = virtio_alloc_queues(eth_dev);
1824 if (eth_dev->data->dev_conf.intr_conf.rxq) {
1825 if (virtio_configure_intr(eth_dev) < 0) {
1826 PMD_INIT_LOG(ERR, "failed to configure interrupt");
1827 virtio_free_queues(hw);
1832 vtpci_reinit_complete(hw);
1835 PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1836 eth_dev->data->port_id, pci_dev->id.vendor_id,
1837 pci_dev->id.device_id);
1843 * Remap the PCI device again (IO port map for legacy device and
1844 * memory map for modern device), so that the secondary process
1845 * could have the PCI initiated correctly.
1848 virtio_remap_pci(struct rte_pci_device *pci_dev, struct virtio_hw *hw)
1852 * We don't have to re-parse the PCI config space, since
1853 * rte_pci_map_device() makes sure the mapped address
1854 * in secondary process would equal to the one mapped in
1855 * the primary process: error will be returned if that
1856 * requirement is not met.
1858 * That said, we could simply reuse all cap pointers
1859 * (such as dev_cfg, common_cfg, etc.) parsed from the
1860 * primary process, which is stored in shared memory.
1862 if (rte_pci_map_device(pci_dev)) {
1863 PMD_INIT_LOG(DEBUG, "failed to map pci device!");
1867 if (rte_pci_ioport_map(pci_dev, 0, VTPCI_IO(hw)) < 0)
1875 virtio_set_vtpci_ops(struct virtio_hw *hw)
1877 #ifdef RTE_VIRTIO_USER
1878 if (hw->virtio_user_dev)
1879 VTPCI_OPS(hw) = &virtio_user_ops;
1883 VTPCI_OPS(hw) = &modern_ops;
1885 VTPCI_OPS(hw) = &legacy_ops;
1889 * This function is based on probe() function in virtio_pci.c
1890 * It returns 0 on success.
1893 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1895 struct virtio_hw *hw = eth_dev->data->dev_private;
1896 uint32_t speed = SPEED_UNKNOWN;
1900 if (sizeof(struct virtio_net_hdr_mrg_rxbuf) > RTE_PKTMBUF_HEADROOM) {
1902 "Not sufficient headroom required = %d, avail = %d",
1903 (int)sizeof(struct virtio_net_hdr_mrg_rxbuf),
1904 RTE_PKTMBUF_HEADROOM);
1909 eth_dev->dev_ops = &virtio_eth_dev_ops;
1910 eth_dev->rx_descriptor_done = virtio_dev_rx_queue_done;
1912 if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1913 if (!hw->virtio_user_dev) {
1914 ret = virtio_remap_pci(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1919 virtio_set_vtpci_ops(hw);
1920 set_rxtx_funcs(eth_dev);
1924 ret = virtio_dev_devargs_parse(eth_dev->device->devargs,
1925 NULL, &speed, &vectorized);
1930 /* Allocate memory for storing MAC addresses */
1931 eth_dev->data->mac_addrs = rte_zmalloc("virtio",
1932 VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN, 0);
1933 if (eth_dev->data->mac_addrs == NULL) {
1935 "Failed to allocate %d bytes needed to store MAC addresses",
1936 VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN);
1940 hw->port_id = eth_dev->data->port_id;
1941 /* For virtio_user case the hw->virtio_user_dev is populated by
1942 * virtio_user_eth_dev_alloc() before eth_virtio_dev_init() is called.
1944 if (!hw->virtio_user_dev) {
1945 ret = vtpci_init(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1947 goto err_vtpci_init;
1950 rte_spinlock_init(&hw->state_lock);
1952 /* reset device and negotiate default features */
1953 ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1955 goto err_virtio_init;
1958 if (!vtpci_packed_queue(hw)) {
1961 #if !defined(CC_AVX512_SUPPORT)
1963 "building environment do not support packed ring vectorized");
1976 if (!hw->virtio_user_dev) {
1977 rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(eth_dev));
1979 rte_pci_ioport_unmap(VTPCI_IO(hw));
1982 rte_free(eth_dev->data->mac_addrs);
1983 eth_dev->data->mac_addrs = NULL;
1988 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1990 PMD_INIT_FUNC_TRACE();
1992 if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1995 virtio_dev_stop(eth_dev);
1996 virtio_dev_close(eth_dev);
1998 eth_dev->dev_ops = NULL;
1999 eth_dev->tx_pkt_burst = NULL;
2000 eth_dev->rx_pkt_burst = NULL;
2002 PMD_INIT_LOG(DEBUG, "dev_uninit completed");
2008 static int vdpa_check_handler(__rte_unused const char *key,
2009 const char *value, void *ret_val)
2011 if (strcmp(value, "1") == 0)
2012 *(int *)ret_val = 1;
2014 *(int *)ret_val = 0;
2021 virtio_dev_speed_capa_get(uint32_t speed)
2024 case ETH_SPEED_NUM_10G:
2025 return ETH_LINK_SPEED_10G;
2026 case ETH_SPEED_NUM_20G:
2027 return ETH_LINK_SPEED_20G;
2028 case ETH_SPEED_NUM_25G:
2029 return ETH_LINK_SPEED_25G;
2030 case ETH_SPEED_NUM_40G:
2031 return ETH_LINK_SPEED_40G;
2032 case ETH_SPEED_NUM_50G:
2033 return ETH_LINK_SPEED_50G;
2034 case ETH_SPEED_NUM_56G:
2035 return ETH_LINK_SPEED_56G;
2036 case ETH_SPEED_NUM_100G:
2037 return ETH_LINK_SPEED_100G;
2043 static int vectorized_check_handler(__rte_unused const char *key,
2044 const char *value, void *ret_val)
2046 if (strcmp(value, "1") == 0)
2047 *(int *)ret_val = 1;
2049 *(int *)ret_val = 0;
2054 #define VIRTIO_ARG_SPEED "speed"
2055 #define VIRTIO_ARG_VDPA "vdpa"
2056 #define VIRTIO_ARG_VECTORIZED "vectorized"
2060 link_speed_handler(const char *key __rte_unused,
2061 const char *value, void *ret_val)
2064 if (!value || !ret_val)
2066 val = strtoul(value, NULL, 0);
2067 /* validate input */
2068 if (virtio_dev_speed_capa_get(val) == 0)
2070 *(uint32_t *)ret_val = val;
2077 virtio_dev_devargs_parse(struct rte_devargs *devargs, int *vdpa,
2078 uint32_t *speed, int *vectorized)
2080 struct rte_kvargs *kvlist;
2083 if (devargs == NULL)
2086 kvlist = rte_kvargs_parse(devargs->args, NULL);
2087 if (kvlist == NULL) {
2088 PMD_INIT_LOG(ERR, "error when parsing param");
2091 if (vdpa && rte_kvargs_count(kvlist, VIRTIO_ARG_VDPA) == 1) {
2092 /* vdpa mode selected when there's a key-value pair:
2095 ret = rte_kvargs_process(kvlist, VIRTIO_ARG_VDPA,
2096 vdpa_check_handler, vdpa);
2098 PMD_INIT_LOG(ERR, "Failed to parse %s",
2103 if (speed && rte_kvargs_count(kvlist, VIRTIO_ARG_SPEED) == 1) {
2104 ret = rte_kvargs_process(kvlist,
2106 link_speed_handler, speed);
2108 PMD_INIT_LOG(ERR, "Failed to parse %s",
2115 rte_kvargs_count(kvlist, VIRTIO_ARG_VECTORIZED) == 1) {
2116 ret = rte_kvargs_process(kvlist,
2117 VIRTIO_ARG_VECTORIZED,
2118 vectorized_check_handler, vectorized);
2120 PMD_INIT_LOG(ERR, "Failed to parse %s",
2121 VIRTIO_ARG_VECTORIZED);
2127 rte_kvargs_free(kvlist);
2131 static int eth_virtio_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2132 struct rte_pci_device *pci_dev)
2137 ret = virtio_dev_devargs_parse(pci_dev->device.devargs, &vdpa, NULL,
2140 PMD_INIT_LOG(ERR, "devargs parsing is failed");
2143 /* virtio pmd skips probe if device needs to work in vdpa mode */
2147 return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct virtio_hw),
2148 eth_virtio_dev_init);
2151 static int eth_virtio_pci_remove(struct rte_pci_device *pci_dev)
2155 ret = rte_eth_dev_pci_generic_remove(pci_dev, eth_virtio_dev_uninit);
2156 /* Port has already been released by close. */
2162 static struct rte_pci_driver rte_virtio_pmd = {
2164 .name = "net_virtio",
2166 .id_table = pci_id_virtio_map,
2168 .probe = eth_virtio_pci_probe,
2169 .remove = eth_virtio_pci_remove,
2172 RTE_INIT(rte_virtio_pmd_init)
2174 rte_eal_iopl_init();
2175 rte_pci_register(&rte_virtio_pmd);
2179 rx_offload_enabled(struct virtio_hw *hw)
2181 return vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM) ||
2182 vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2183 vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6);
2187 tx_offload_enabled(struct virtio_hw *hw)
2189 return vtpci_with_feature(hw, VIRTIO_NET_F_CSUM) ||
2190 vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO4) ||
2191 vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO6);
2195 * Configure virtio device
2196 * It returns 0 on success.
2199 virtio_dev_configure(struct rte_eth_dev *dev)
2201 const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2202 const struct rte_eth_txmode *txmode = &dev->data->dev_conf.txmode;
2203 struct virtio_hw *hw = dev->data->dev_private;
2204 uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
2206 uint64_t rx_offloads = rxmode->offloads;
2207 uint64_t tx_offloads = txmode->offloads;
2208 uint64_t req_features;
2211 PMD_INIT_LOG(DEBUG, "configure");
2212 req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
2214 if (rxmode->mq_mode != ETH_MQ_RX_NONE) {
2216 "Unsupported Rx multi queue mode %d",
2221 if (txmode->mq_mode != ETH_MQ_TX_NONE) {
2223 "Unsupported Tx multi queue mode %d",
2228 if (dev->data->dev_conf.intr_conf.rxq) {
2229 ret = virtio_init_device(dev, hw->req_guest_features);
2234 if (rxmode->max_rx_pkt_len > hw->max_mtu + ether_hdr_len)
2235 req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
2237 if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2238 DEV_RX_OFFLOAD_TCP_CKSUM))
2239 req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
2241 if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO)
2243 (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2244 (1ULL << VIRTIO_NET_F_GUEST_TSO6);
2246 if (tx_offloads & (DEV_TX_OFFLOAD_UDP_CKSUM |
2247 DEV_TX_OFFLOAD_TCP_CKSUM))
2248 req_features |= (1ULL << VIRTIO_NET_F_CSUM);
2250 if (tx_offloads & DEV_TX_OFFLOAD_TCP_TSO)
2252 (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2253 (1ULL << VIRTIO_NET_F_HOST_TSO6);
2255 /* if request features changed, reinit the device */
2256 if (req_features != hw->req_guest_features) {
2257 ret = virtio_init_device(dev, req_features);
2262 if ((rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2263 DEV_RX_OFFLOAD_TCP_CKSUM)) &&
2264 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
2266 "rx checksum not available on this host");
2270 if ((rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) &&
2271 (!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2272 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6))) {
2274 "Large Receive Offload not available on this host");
2278 /* start control queue */
2279 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
2280 virtio_dev_cq_start(dev);
2282 if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
2285 if ((rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
2286 && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2288 "vlan filtering not available on this host");
2292 hw->has_tx_offload = tx_offload_enabled(hw);
2293 hw->has_rx_offload = rx_offload_enabled(hw);
2295 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2296 /* Enable vector (0) for Link State Intrerrupt */
2297 if (VTPCI_OPS(hw)->set_config_irq(hw, 0) ==
2298 VIRTIO_MSI_NO_VECTOR) {
2299 PMD_DRV_LOG(ERR, "failed to set config vector");
2303 if (vtpci_packed_queue(hw)) {
2304 #if defined(RTE_ARCH_X86_64) && defined(CC_AVX512_SUPPORT)
2305 if ((hw->use_vec_rx || hw->use_vec_tx) &&
2306 (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX512F) ||
2307 !vtpci_with_feature(hw, VIRTIO_F_IN_ORDER) ||
2308 !vtpci_with_feature(hw, VIRTIO_F_VERSION_1))) {
2310 "disabled packed ring vectorized path for requirements not met");
2319 if (hw->use_vec_rx) {
2320 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2322 "disabled packed ring vectorized rx for mrg_rxbuf enabled");
2326 if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) {
2328 "disabled packed ring vectorized rx for TCP_LRO enabled");
2333 if (vtpci_with_feature(hw, VIRTIO_F_IN_ORDER)) {
2334 hw->use_inorder_tx = 1;
2335 hw->use_inorder_rx = 1;
2339 if (hw->use_vec_rx) {
2340 #if defined RTE_ARCH_ARM64 || defined RTE_ARCH_ARM
2341 if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
2343 "disabled split ring vectorized path for requirement not met");
2347 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2349 "disabled split ring vectorized rx for mrg_rxbuf enabled");
2353 if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2354 DEV_RX_OFFLOAD_TCP_CKSUM |
2355 DEV_RX_OFFLOAD_TCP_LRO |
2356 DEV_RX_OFFLOAD_VLAN_STRIP)) {
2358 "disabled split ring vectorized rx for offloading enabled");
2369 virtio_dev_start(struct rte_eth_dev *dev)
2371 uint16_t nb_queues, i;
2372 struct virtnet_rx *rxvq;
2373 struct virtnet_tx *txvq __rte_unused;
2374 struct virtio_hw *hw = dev->data->dev_private;
2377 /* Finish the initialization of the queues */
2378 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2379 ret = virtio_dev_rx_queue_setup_finish(dev, i);
2383 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2384 ret = virtio_dev_tx_queue_setup_finish(dev, i);
2389 /* check if lsc interrupt feature is enabled */
2390 if (dev->data->dev_conf.intr_conf.lsc) {
2391 if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
2392 PMD_DRV_LOG(ERR, "link status not supported by host");
2397 /* Enable uio/vfio intr/eventfd mapping: althrough we already did that
2398 * in device configure, but it could be unmapped when device is
2401 if (dev->data->dev_conf.intr_conf.lsc ||
2402 dev->data->dev_conf.intr_conf.rxq) {
2403 virtio_intr_disable(dev);
2405 /* Setup interrupt callback */
2406 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2407 rte_intr_callback_register(dev->intr_handle,
2408 virtio_interrupt_handler,
2411 if (virtio_intr_enable(dev) < 0) {
2412 PMD_DRV_LOG(ERR, "interrupt enable failed");
2417 /*Notify the backend
2418 *Otherwise the tap backend might already stop its queue due to fullness.
2419 *vhost backend will have no chance to be waked up
2421 nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
2422 if (hw->max_queue_pairs > 1) {
2423 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
2427 PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
2429 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2430 rxvq = dev->data->rx_queues[i];
2431 /* Flush the old packets */
2432 virtqueue_rxvq_flush(rxvq->vq);
2433 virtqueue_notify(rxvq->vq);
2436 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2437 txvq = dev->data->tx_queues[i];
2438 virtqueue_notify(txvq->vq);
2441 PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
2443 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2444 rxvq = dev->data->rx_queues[i];
2445 VIRTQUEUE_DUMP(rxvq->vq);
2448 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2449 txvq = dev->data->tx_queues[i];
2450 VIRTQUEUE_DUMP(txvq->vq);
2453 set_rxtx_funcs(dev);
2456 /* Initialize Link state */
2457 virtio_dev_link_update(dev, 0);
2462 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
2464 struct virtio_hw *hw = dev->data->dev_private;
2465 uint16_t nr_vq = virtio_get_nr_vq(hw);
2466 const char *type __rte_unused;
2467 unsigned int i, mbuf_num = 0;
2468 struct virtqueue *vq;
2469 struct rte_mbuf *buf;
2472 if (hw->vqs == NULL)
2475 for (i = 0; i < nr_vq; i++) {
2480 queue_type = virtio_get_queue_type(hw, i);
2481 if (queue_type == VTNET_RQ)
2483 else if (queue_type == VTNET_TQ)
2489 "Before freeing %s[%d] used and unused buf",
2493 while ((buf = virtqueue_detach_unused(vq)) != NULL) {
2494 rte_pktmbuf_free(buf);
2499 "After freeing %s[%d] used and unused buf",
2504 PMD_INIT_LOG(DEBUG, "%d mbufs freed", mbuf_num);
2508 * Stop device: disable interrupt and mark link down
2511 virtio_dev_stop(struct rte_eth_dev *dev)
2513 struct virtio_hw *hw = dev->data->dev_private;
2514 struct rte_eth_link link;
2515 struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
2517 PMD_INIT_LOG(DEBUG, "stop");
2519 rte_spinlock_lock(&hw->state_lock);
2522 hw->started = false;
2524 if (intr_conf->lsc || intr_conf->rxq) {
2525 virtio_intr_disable(dev);
2527 /* Reset interrupt callback */
2528 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
2529 rte_intr_callback_unregister(dev->intr_handle,
2530 virtio_interrupt_handler,
2535 memset(&link, 0, sizeof(link));
2536 rte_eth_linkstatus_set(dev, &link);
2538 rte_spinlock_unlock(&hw->state_lock);
2542 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
2544 struct rte_eth_link link;
2546 struct virtio_hw *hw = dev->data->dev_private;
2548 memset(&link, 0, sizeof(link));
2549 link.link_duplex = hw->duplex;
2550 link.link_speed = hw->speed;
2551 link.link_autoneg = ETH_LINK_AUTONEG;
2554 link.link_status = ETH_LINK_DOWN;
2555 } else if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
2556 PMD_INIT_LOG(DEBUG, "Get link status from hw");
2557 vtpci_read_dev_config(hw,
2558 offsetof(struct virtio_net_config, status),
2559 &status, sizeof(status));
2560 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
2561 link.link_status = ETH_LINK_DOWN;
2562 PMD_INIT_LOG(DEBUG, "Port %d is down",
2563 dev->data->port_id);
2565 link.link_status = ETH_LINK_UP;
2566 PMD_INIT_LOG(DEBUG, "Port %d is up",
2567 dev->data->port_id);
2570 link.link_status = ETH_LINK_UP;
2573 return rte_eth_linkstatus_set(dev, &link);
2577 virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
2579 const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2580 struct virtio_hw *hw = dev->data->dev_private;
2581 uint64_t offloads = rxmode->offloads;
2583 if (mask & ETH_VLAN_FILTER_MASK) {
2584 if ((offloads & DEV_RX_OFFLOAD_VLAN_FILTER) &&
2585 !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2588 "vlan filtering not available on this host");
2594 if (mask & ETH_VLAN_STRIP_MASK)
2595 hw->vlan_strip = !!(offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
2601 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2603 uint64_t tso_mask, host_features;
2604 struct virtio_hw *hw = dev->data->dev_private;
2605 dev_info->speed_capa = virtio_dev_speed_capa_get(hw->speed);
2607 dev_info->max_rx_queues =
2608 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
2609 dev_info->max_tx_queues =
2610 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
2611 dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
2612 dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
2613 dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
2615 host_features = VTPCI_OPS(hw)->get_features(hw);
2616 dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
2617 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_JUMBO_FRAME;
2618 if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
2619 dev_info->rx_offload_capa |=
2620 DEV_RX_OFFLOAD_TCP_CKSUM |
2621 DEV_RX_OFFLOAD_UDP_CKSUM;
2623 if (host_features & (1ULL << VIRTIO_NET_F_CTRL_VLAN))
2624 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_VLAN_FILTER;
2625 tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2626 (1ULL << VIRTIO_NET_F_GUEST_TSO6);
2627 if ((host_features & tso_mask) == tso_mask)
2628 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
2630 dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
2631 DEV_TX_OFFLOAD_VLAN_INSERT;
2632 if (host_features & (1ULL << VIRTIO_NET_F_CSUM)) {
2633 dev_info->tx_offload_capa |=
2634 DEV_TX_OFFLOAD_UDP_CKSUM |
2635 DEV_TX_OFFLOAD_TCP_CKSUM;
2637 tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2638 (1ULL << VIRTIO_NET_F_HOST_TSO6);
2639 if ((host_features & tso_mask) == tso_mask)
2640 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
2646 * It enables testpmd to collect per queue stats.
2649 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
2650 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
2651 __rte_unused uint8_t is_rx)
2656 RTE_PMD_EXPORT_NAME(net_virtio, __COUNTER__);
2657 RTE_PMD_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map);
2658 RTE_PMD_REGISTER_KMOD_DEP(net_virtio, "* igb_uio | uio_pci_generic | vfio-pci");
2659 RTE_LOG_REGISTER(virtio_logtype_init, pmd.net.virtio.init, NOTICE);
2660 RTE_LOG_REGISTER(virtio_logtype_driver, pmd.net.virtio.driver, NOTICE);