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 /* first indirect descriptor is always the tx header */
613 if (!vtpci_packed_queue(hw)) {
614 struct vring_desc *start_dp = txr[i].tx_indir;
615 vring_desc_init_split(start_dp,
616 RTE_DIM(txr[i].tx_indir));
617 start_dp->addr = txvq->virtio_net_hdr_mem
619 + offsetof(struct virtio_tx_region,
621 start_dp->len = hw->vtnet_hdr_size;
622 start_dp->flags = VRING_DESC_F_NEXT;
624 struct vring_packed_desc *start_dp =
625 txr[i].tx_packed_indir;
626 vring_desc_init_indirect_packed(start_dp,
627 RTE_DIM(txr[i].tx_packed_indir));
628 start_dp->addr = txvq->virtio_net_hdr_mem
630 + offsetof(struct virtio_tx_region,
632 start_dp->len = hw->vtnet_hdr_size;
637 if (VTPCI_OPS(hw)->setup_queue(hw, vq) < 0) {
638 PMD_INIT_LOG(ERR, "setup_queue failed");
646 rte_memzone_free(hdr_mz);
647 rte_memzone_free(mz);
654 virtio_free_queues(struct virtio_hw *hw)
656 uint16_t nr_vq = virtio_get_nr_vq(hw);
657 struct virtqueue *vq;
664 for (i = 0; i < nr_vq; i++) {
669 queue_type = virtio_get_queue_type(hw, i);
670 if (queue_type == VTNET_RQ) {
671 rte_free(vq->sw_ring);
672 rte_memzone_free(vq->rxq.mz);
673 } else if (queue_type == VTNET_TQ) {
674 rte_memzone_free(vq->txq.mz);
675 rte_memzone_free(vq->txq.virtio_net_hdr_mz);
677 rte_memzone_free(vq->cq.mz);
678 rte_memzone_free(vq->cq.virtio_net_hdr_mz);
690 virtio_alloc_queues(struct rte_eth_dev *dev)
692 struct virtio_hw *hw = dev->data->dev_private;
693 uint16_t nr_vq = virtio_get_nr_vq(hw);
697 hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
699 PMD_INIT_LOG(ERR, "failed to allocate vqs");
703 for (i = 0; i < nr_vq; i++) {
704 ret = virtio_init_queue(dev, i);
706 virtio_free_queues(hw);
714 static void virtio_queues_unbind_intr(struct rte_eth_dev *dev);
717 virtio_dev_close(struct rte_eth_dev *dev)
719 struct virtio_hw *hw = dev->data->dev_private;
720 struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
722 PMD_INIT_LOG(DEBUG, "virtio_dev_close");
723 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
731 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
732 VTPCI_OPS(hw)->set_config_irq(hw, VIRTIO_MSI_NO_VECTOR);
734 virtio_queues_unbind_intr(dev);
736 if (intr_conf->lsc || intr_conf->rxq) {
737 virtio_intr_disable(dev);
738 rte_intr_efd_disable(dev->intr_handle);
739 rte_free(dev->intr_handle->intr_vec);
740 dev->intr_handle->intr_vec = NULL;
744 virtio_dev_free_mbufs(dev);
745 virtio_free_queues(hw);
747 #ifdef RTE_VIRTIO_USER
748 if (hw->virtio_user_dev)
749 virtio_user_dev_uninit(hw->virtio_user_dev);
753 rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(dev));
755 rte_pci_ioport_unmap(VTPCI_IO(hw));
762 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
764 struct virtio_hw *hw = dev->data->dev_private;
765 struct virtio_pmd_ctrl ctrl;
769 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
770 PMD_INIT_LOG(INFO, "host does not support rx control");
774 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
775 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
779 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
781 PMD_INIT_LOG(ERR, "Failed to enable promisc");
789 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
791 struct virtio_hw *hw = dev->data->dev_private;
792 struct virtio_pmd_ctrl ctrl;
796 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
797 PMD_INIT_LOG(INFO, "host does not support rx control");
801 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
802 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
806 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
808 PMD_INIT_LOG(ERR, "Failed to disable promisc");
816 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
818 struct virtio_hw *hw = dev->data->dev_private;
819 struct virtio_pmd_ctrl ctrl;
823 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
824 PMD_INIT_LOG(INFO, "host does not support rx control");
828 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
829 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
833 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
835 PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
843 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
845 struct virtio_hw *hw = dev->data->dev_private;
846 struct virtio_pmd_ctrl ctrl;
850 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
851 PMD_INIT_LOG(INFO, "host does not support rx control");
855 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
856 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
860 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
862 PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
869 #define VLAN_TAG_LEN 4 /* 802.3ac tag (not DMA'd) */
871 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
873 struct virtio_hw *hw = dev->data->dev_private;
874 uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
876 uint32_t frame_size = mtu + ether_hdr_len;
877 uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
879 max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
881 if (mtu < RTE_ETHER_MIN_MTU || frame_size > max_frame_size) {
882 PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
883 RTE_ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
890 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
892 struct virtio_hw *hw = dev->data->dev_private;
893 struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
894 struct virtqueue *vq = rxvq->vq;
896 virtqueue_enable_intr(vq);
897 virtio_mb(hw->weak_barriers);
902 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
904 struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
905 struct virtqueue *vq = rxvq->vq;
907 virtqueue_disable_intr(vq);
912 * dev_ops for virtio, bare necessities for basic operation
914 static const struct eth_dev_ops virtio_eth_dev_ops = {
915 .dev_configure = virtio_dev_configure,
916 .dev_start = virtio_dev_start,
917 .dev_stop = virtio_dev_stop,
918 .dev_close = virtio_dev_close,
919 .promiscuous_enable = virtio_dev_promiscuous_enable,
920 .promiscuous_disable = virtio_dev_promiscuous_disable,
921 .allmulticast_enable = virtio_dev_allmulticast_enable,
922 .allmulticast_disable = virtio_dev_allmulticast_disable,
923 .mtu_set = virtio_mtu_set,
924 .dev_infos_get = virtio_dev_info_get,
925 .stats_get = virtio_dev_stats_get,
926 .xstats_get = virtio_dev_xstats_get,
927 .xstats_get_names = virtio_dev_xstats_get_names,
928 .stats_reset = virtio_dev_stats_reset,
929 .xstats_reset = virtio_dev_stats_reset,
930 .link_update = virtio_dev_link_update,
931 .vlan_offload_set = virtio_dev_vlan_offload_set,
932 .rx_queue_setup = virtio_dev_rx_queue_setup,
933 .rx_queue_intr_enable = virtio_dev_rx_queue_intr_enable,
934 .rx_queue_intr_disable = virtio_dev_rx_queue_intr_disable,
935 .rx_queue_release = virtio_dev_queue_release,
936 .tx_queue_setup = virtio_dev_tx_queue_setup,
937 .tx_queue_release = virtio_dev_queue_release,
938 /* collect stats per queue */
939 .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
940 .vlan_filter_set = virtio_vlan_filter_set,
941 .mac_addr_add = virtio_mac_addr_add,
942 .mac_addr_remove = virtio_mac_addr_remove,
943 .mac_addr_set = virtio_mac_addr_set,
947 * dev_ops for virtio-user in secondary processes, as we just have
948 * some limited supports currently.
950 const struct eth_dev_ops virtio_user_secondary_eth_dev_ops = {
951 .dev_infos_get = virtio_dev_info_get,
952 .stats_get = virtio_dev_stats_get,
953 .xstats_get = virtio_dev_xstats_get,
954 .xstats_get_names = virtio_dev_xstats_get_names,
955 .stats_reset = virtio_dev_stats_reset,
956 .xstats_reset = virtio_dev_stats_reset,
957 /* collect stats per queue */
958 .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
962 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
966 for (i = 0; i < dev->data->nb_tx_queues; i++) {
967 const struct virtnet_tx *txvq = dev->data->tx_queues[i];
971 stats->opackets += txvq->stats.packets;
972 stats->obytes += txvq->stats.bytes;
974 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
975 stats->q_opackets[i] = txvq->stats.packets;
976 stats->q_obytes[i] = txvq->stats.bytes;
980 for (i = 0; i < dev->data->nb_rx_queues; i++) {
981 const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
985 stats->ipackets += rxvq->stats.packets;
986 stats->ibytes += rxvq->stats.bytes;
987 stats->ierrors += rxvq->stats.errors;
989 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
990 stats->q_ipackets[i] = rxvq->stats.packets;
991 stats->q_ibytes[i] = rxvq->stats.bytes;
995 stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
998 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
999 struct rte_eth_xstat_name *xstats_names,
1000 __rte_unused unsigned limit)
1006 unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
1007 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
1009 if (xstats_names != NULL) {
1010 /* Note: limit checked in rte_eth_xstats_names() */
1012 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1013 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1016 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1017 snprintf(xstats_names[count].name,
1018 sizeof(xstats_names[count].name),
1020 rte_virtio_rxq_stat_strings[t].name);
1025 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1026 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1029 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1030 snprintf(xstats_names[count].name,
1031 sizeof(xstats_names[count].name),
1033 rte_virtio_txq_stat_strings[t].name);
1043 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
1049 unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
1050 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
1055 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1056 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1063 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1064 xstats[count].value = *(uint64_t *)(((char *)rxvq) +
1065 rte_virtio_rxq_stat_strings[t].offset);
1066 xstats[count].id = count;
1071 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1072 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1079 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1080 xstats[count].value = *(uint64_t *)(((char *)txvq) +
1081 rte_virtio_txq_stat_strings[t].offset);
1082 xstats[count].id = count;
1091 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1093 virtio_update_stats(dev, stats);
1099 virtio_dev_stats_reset(struct rte_eth_dev *dev)
1103 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1104 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1108 txvq->stats.packets = 0;
1109 txvq->stats.bytes = 0;
1110 txvq->stats.multicast = 0;
1111 txvq->stats.broadcast = 0;
1112 memset(txvq->stats.size_bins, 0,
1113 sizeof(txvq->stats.size_bins[0]) * 8);
1116 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1117 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1121 rxvq->stats.packets = 0;
1122 rxvq->stats.bytes = 0;
1123 rxvq->stats.errors = 0;
1124 rxvq->stats.multicast = 0;
1125 rxvq->stats.broadcast = 0;
1126 memset(rxvq->stats.size_bins, 0,
1127 sizeof(rxvq->stats.size_bins[0]) * 8);
1134 virtio_set_hwaddr(struct virtio_hw *hw)
1136 vtpci_write_dev_config(hw,
1137 offsetof(struct virtio_net_config, mac),
1138 &hw->mac_addr, RTE_ETHER_ADDR_LEN);
1142 virtio_get_hwaddr(struct virtio_hw *hw)
1144 if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
1145 vtpci_read_dev_config(hw,
1146 offsetof(struct virtio_net_config, mac),
1147 &hw->mac_addr, RTE_ETHER_ADDR_LEN);
1149 rte_eth_random_addr(&hw->mac_addr[0]);
1150 virtio_set_hwaddr(hw);
1155 virtio_mac_table_set(struct virtio_hw *hw,
1156 const struct virtio_net_ctrl_mac *uc,
1157 const struct virtio_net_ctrl_mac *mc)
1159 struct virtio_pmd_ctrl ctrl;
1162 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1163 PMD_DRV_LOG(INFO, "host does not support mac table");
1167 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1168 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1170 len[0] = uc->entries * RTE_ETHER_ADDR_LEN + sizeof(uc->entries);
1171 memcpy(ctrl.data, uc, len[0]);
1173 len[1] = mc->entries * RTE_ETHER_ADDR_LEN + sizeof(mc->entries);
1174 memcpy(ctrl.data + len[0], mc, len[1]);
1176 err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1178 PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1183 virtio_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
1184 uint32_t index, uint32_t vmdq __rte_unused)
1186 struct virtio_hw *hw = dev->data->dev_private;
1187 const struct rte_ether_addr *addrs = dev->data->mac_addrs;
1189 struct virtio_net_ctrl_mac *uc, *mc;
1191 if (index >= VIRTIO_MAX_MAC_ADDRS) {
1192 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1196 uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1197 sizeof(uc->entries));
1199 mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1200 sizeof(mc->entries));
1203 for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1204 const struct rte_ether_addr *addr
1205 = (i == index) ? mac_addr : addrs + i;
1206 struct virtio_net_ctrl_mac *tbl
1207 = rte_is_multicast_ether_addr(addr) ? mc : uc;
1209 memcpy(&tbl->macs[tbl->entries++], addr, RTE_ETHER_ADDR_LEN);
1212 return virtio_mac_table_set(hw, uc, mc);
1216 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1218 struct virtio_hw *hw = dev->data->dev_private;
1219 struct rte_ether_addr *addrs = dev->data->mac_addrs;
1220 struct virtio_net_ctrl_mac *uc, *mc;
1223 if (index >= VIRTIO_MAX_MAC_ADDRS) {
1224 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1228 uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1229 sizeof(uc->entries));
1231 mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1232 sizeof(mc->entries));
1235 for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1236 struct virtio_net_ctrl_mac *tbl;
1238 if (i == index || rte_is_zero_ether_addr(addrs + i))
1241 tbl = rte_is_multicast_ether_addr(addrs + i) ? mc : uc;
1242 memcpy(&tbl->macs[tbl->entries++], addrs + i,
1243 RTE_ETHER_ADDR_LEN);
1246 virtio_mac_table_set(hw, uc, mc);
1250 virtio_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1252 struct virtio_hw *hw = dev->data->dev_private;
1254 memcpy(hw->mac_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1256 /* Use atomic update if available */
1257 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1258 struct virtio_pmd_ctrl ctrl;
1259 int len = RTE_ETHER_ADDR_LEN;
1261 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1262 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1264 memcpy(ctrl.data, mac_addr, RTE_ETHER_ADDR_LEN);
1265 return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1268 if (!vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1271 virtio_set_hwaddr(hw);
1276 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1278 struct virtio_hw *hw = dev->data->dev_private;
1279 struct virtio_pmd_ctrl ctrl;
1282 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1285 ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1286 ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1287 memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1288 len = sizeof(vlan_id);
1290 return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1294 virtio_intr_unmask(struct rte_eth_dev *dev)
1296 struct virtio_hw *hw = dev->data->dev_private;
1298 if (rte_intr_ack(dev->intr_handle) < 0)
1301 if (!hw->virtio_user_dev)
1302 hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1308 virtio_intr_enable(struct rte_eth_dev *dev)
1310 struct virtio_hw *hw = dev->data->dev_private;
1312 if (rte_intr_enable(dev->intr_handle) < 0)
1315 if (!hw->virtio_user_dev)
1316 hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1322 virtio_intr_disable(struct rte_eth_dev *dev)
1324 struct virtio_hw *hw = dev->data->dev_private;
1326 if (rte_intr_disable(dev->intr_handle) < 0)
1329 if (!hw->virtio_user_dev)
1330 hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1336 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1338 uint64_t host_features;
1340 /* Prepare guest_features: feature that driver wants to support */
1341 PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1344 /* Read device(host) feature bits */
1345 host_features = VTPCI_OPS(hw)->get_features(hw);
1346 PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1349 /* If supported, ensure MTU value is valid before acknowledging it. */
1350 if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1351 struct virtio_net_config config;
1353 vtpci_read_dev_config(hw,
1354 offsetof(struct virtio_net_config, mtu),
1355 &config.mtu, sizeof(config.mtu));
1357 if (config.mtu < RTE_ETHER_MIN_MTU)
1358 req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1362 * Negotiate features: Subset of device feature bits are written back
1363 * guest feature bits.
1365 hw->guest_features = req_features;
1366 hw->guest_features = vtpci_negotiate_features(hw, host_features);
1367 PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1368 hw->guest_features);
1370 if (hw->modern && !vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1372 "VIRTIO_F_VERSION_1 features is not enabled.");
1376 if (hw->modern || hw->virtio_user_dev) {
1377 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1378 if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1380 "failed to set FEATURES_OK status!");
1385 hw->req_guest_features = req_features;
1391 virtio_dev_pause(struct rte_eth_dev *dev)
1393 struct virtio_hw *hw = dev->data->dev_private;
1395 rte_spinlock_lock(&hw->state_lock);
1397 if (hw->started == 0) {
1398 /* Device is just stopped. */
1399 rte_spinlock_unlock(&hw->state_lock);
1404 * Prevent the worker threads from touching queues to avoid contention,
1405 * 1 ms should be enough for the ongoing Tx function to finish.
1412 * Recover hw state to let the worker threads continue.
1415 virtio_dev_resume(struct rte_eth_dev *dev)
1417 struct virtio_hw *hw = dev->data->dev_private;
1420 rte_spinlock_unlock(&hw->state_lock);
1424 * Should be called only after device is paused.
1427 virtio_inject_pkts(struct rte_eth_dev *dev, struct rte_mbuf **tx_pkts,
1430 struct virtio_hw *hw = dev->data->dev_private;
1431 struct virtnet_tx *txvq = dev->data->tx_queues[0];
1434 hw->inject_pkts = tx_pkts;
1435 ret = dev->tx_pkt_burst(txvq, tx_pkts, nb_pkts);
1436 hw->inject_pkts = NULL;
1442 virtio_notify_peers(struct rte_eth_dev *dev)
1444 struct virtio_hw *hw = dev->data->dev_private;
1445 struct virtnet_rx *rxvq;
1446 struct rte_mbuf *rarp_mbuf;
1448 if (!dev->data->rx_queues)
1451 rxvq = dev->data->rx_queues[0];
1455 rarp_mbuf = rte_net_make_rarp_packet(rxvq->mpool,
1456 (struct rte_ether_addr *)hw->mac_addr);
1457 if (rarp_mbuf == NULL) {
1458 PMD_DRV_LOG(ERR, "failed to make RARP packet.");
1462 /* If virtio port just stopped, no need to send RARP */
1463 if (virtio_dev_pause(dev) < 0) {
1464 rte_pktmbuf_free(rarp_mbuf);
1468 virtio_inject_pkts(dev, &rarp_mbuf, 1);
1469 virtio_dev_resume(dev);
1473 virtio_ack_link_announce(struct rte_eth_dev *dev)
1475 struct virtio_hw *hw = dev->data->dev_private;
1476 struct virtio_pmd_ctrl ctrl;
1478 ctrl.hdr.class = VIRTIO_NET_CTRL_ANNOUNCE;
1479 ctrl.hdr.cmd = VIRTIO_NET_CTRL_ANNOUNCE_ACK;
1481 virtio_send_command(hw->cvq, &ctrl, NULL, 0);
1485 * Process virtio config changed interrupt. Call the callback
1486 * if link state changed, generate gratuitous RARP packet if
1487 * the status indicates an ANNOUNCE.
1490 virtio_interrupt_handler(void *param)
1492 struct rte_eth_dev *dev = param;
1493 struct virtio_hw *hw = dev->data->dev_private;
1497 /* Read interrupt status which clears interrupt */
1498 isr = vtpci_isr(hw);
1499 PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1501 if (virtio_intr_unmask(dev) < 0)
1502 PMD_DRV_LOG(ERR, "interrupt enable failed");
1504 if (isr & VIRTIO_PCI_ISR_CONFIG) {
1505 if (virtio_dev_link_update(dev, 0) == 0)
1506 rte_eth_dev_callback_process(dev,
1507 RTE_ETH_EVENT_INTR_LSC,
1510 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1511 vtpci_read_dev_config(hw,
1512 offsetof(struct virtio_net_config, status),
1513 &status, sizeof(status));
1514 if (status & VIRTIO_NET_S_ANNOUNCE) {
1515 virtio_notify_peers(dev);
1517 virtio_ack_link_announce(dev);
1523 /* set rx and tx handlers according to what is supported */
1525 set_rxtx_funcs(struct rte_eth_dev *eth_dev)
1527 struct virtio_hw *hw = eth_dev->data->dev_private;
1529 eth_dev->tx_pkt_prepare = virtio_xmit_pkts_prepare;
1530 if (vtpci_packed_queue(hw)) {
1532 "virtio: using packed ring %s Tx path on port %u",
1533 hw->use_vec_tx ? "vectorized" : "standard",
1534 eth_dev->data->port_id);
1536 eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed_vec;
1538 eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed;
1540 if (hw->use_inorder_tx) {
1541 PMD_INIT_LOG(INFO, "virtio: using inorder Tx path on port %u",
1542 eth_dev->data->port_id);
1543 eth_dev->tx_pkt_burst = virtio_xmit_pkts_inorder;
1545 PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
1546 eth_dev->data->port_id);
1547 eth_dev->tx_pkt_burst = virtio_xmit_pkts;
1551 if (vtpci_packed_queue(hw)) {
1552 if (hw->use_vec_rx) {
1554 "virtio: using packed ring vectorized Rx path on port %u",
1555 eth_dev->data->port_id);
1556 eth_dev->rx_pkt_burst =
1557 &virtio_recv_pkts_packed_vec;
1558 } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1560 "virtio: using packed ring mergeable buffer Rx path on port %u",
1561 eth_dev->data->port_id);
1562 eth_dev->rx_pkt_burst =
1563 &virtio_recv_mergeable_pkts_packed;
1566 "virtio: using packed ring standard Rx path on port %u",
1567 eth_dev->data->port_id);
1568 eth_dev->rx_pkt_burst = &virtio_recv_pkts_packed;
1571 if (hw->use_vec_rx) {
1572 PMD_INIT_LOG(INFO, "virtio: using vectorized Rx path on port %u",
1573 eth_dev->data->port_id);
1574 eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1575 } else if (hw->use_inorder_rx) {
1577 "virtio: using inorder Rx path on port %u",
1578 eth_dev->data->port_id);
1579 eth_dev->rx_pkt_burst = &virtio_recv_pkts_inorder;
1580 } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1582 "virtio: using mergeable buffer Rx path on port %u",
1583 eth_dev->data->port_id);
1584 eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1586 PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
1587 eth_dev->data->port_id);
1588 eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1594 /* Only support 1:1 queue/interrupt mapping so far.
1595 * TODO: support n:1 queue/interrupt mapping when there are limited number of
1596 * interrupt vectors (<N+1).
1599 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1602 struct virtio_hw *hw = dev->data->dev_private;
1604 PMD_INIT_LOG(INFO, "queue/interrupt binding");
1605 for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1606 dev->intr_handle->intr_vec[i] = i + 1;
1607 if (VTPCI_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1608 VIRTIO_MSI_NO_VECTOR) {
1609 PMD_DRV_LOG(ERR, "failed to set queue vector");
1618 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1621 struct virtio_hw *hw = dev->data->dev_private;
1623 PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1624 for (i = 0; i < dev->data->nb_rx_queues; ++i)
1625 VTPCI_OPS(hw)->set_queue_irq(hw,
1626 hw->vqs[i * VTNET_CQ],
1627 VIRTIO_MSI_NO_VECTOR);
1631 virtio_configure_intr(struct rte_eth_dev *dev)
1633 struct virtio_hw *hw = dev->data->dev_private;
1635 if (!rte_intr_cap_multiple(dev->intr_handle)) {
1636 PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1640 if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1641 PMD_INIT_LOG(ERR, "Fail to create eventfd");
1645 if (!dev->intr_handle->intr_vec) {
1646 dev->intr_handle->intr_vec =
1647 rte_zmalloc("intr_vec",
1648 hw->max_queue_pairs * sizeof(int), 0);
1649 if (!dev->intr_handle->intr_vec) {
1650 PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1651 hw->max_queue_pairs);
1656 /* Re-register callback to update max_intr */
1657 rte_intr_callback_unregister(dev->intr_handle,
1658 virtio_interrupt_handler,
1660 rte_intr_callback_register(dev->intr_handle,
1661 virtio_interrupt_handler,
1664 /* DO NOT try to remove this! This function will enable msix, or QEMU
1665 * will encounter SIGSEGV when DRIVER_OK is sent.
1666 * And for legacy devices, this should be done before queue/vec binding
1667 * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1668 * (22) will be ignored.
1670 if (virtio_intr_enable(dev) < 0) {
1671 PMD_DRV_LOG(ERR, "interrupt enable failed");
1675 if (virtio_queues_bind_intr(dev) < 0) {
1676 PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1682 #define DUPLEX_UNKNOWN 0xff
1683 /* reset device and renegotiate features if needed */
1685 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1687 struct virtio_hw *hw = eth_dev->data->dev_private;
1688 struct virtio_net_config *config;
1689 struct virtio_net_config local_config;
1690 struct rte_pci_device *pci_dev = NULL;
1693 /* Reset the device although not necessary at startup */
1697 virtio_dev_free_mbufs(eth_dev);
1698 virtio_free_queues(hw);
1701 /* Tell the host we've noticed this device. */
1702 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1704 /* Tell the host we've known how to drive the device. */
1705 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1706 if (virtio_negotiate_features(hw, req_features) < 0)
1709 hw->weak_barriers = !vtpci_with_feature(hw, VIRTIO_F_ORDER_PLATFORM);
1711 if (!hw->virtio_user_dev)
1712 pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1714 /* If host does not support both status and MSI-X then disable LSC */
1715 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS) &&
1716 hw->use_msix != VIRTIO_MSIX_NONE)
1717 eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1719 eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1721 /* Setting up rx_header size for the device */
1722 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1723 vtpci_with_feature(hw, VIRTIO_F_VERSION_1) ||
1724 vtpci_with_feature(hw, VIRTIO_F_RING_PACKED))
1725 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1727 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1729 /* Copy the permanent MAC address to: virtio_hw */
1730 virtio_get_hwaddr(hw);
1731 rte_ether_addr_copy((struct rte_ether_addr *)hw->mac_addr,
1732 ð_dev->data->mac_addrs[0]);
1734 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1735 hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1736 hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1738 if (hw->speed == ETH_SPEED_NUM_UNKNOWN) {
1739 if (vtpci_with_feature(hw, VIRTIO_NET_F_SPEED_DUPLEX)) {
1740 config = &local_config;
1741 vtpci_read_dev_config(hw,
1742 offsetof(struct virtio_net_config, speed),
1743 &config->speed, sizeof(config->speed));
1744 vtpci_read_dev_config(hw,
1745 offsetof(struct virtio_net_config, duplex),
1746 &config->duplex, sizeof(config->duplex));
1747 hw->speed = config->speed;
1748 hw->duplex = config->duplex;
1751 if (hw->duplex == DUPLEX_UNKNOWN)
1752 hw->duplex = ETH_LINK_FULL_DUPLEX;
1753 PMD_INIT_LOG(DEBUG, "link speed = %d, duplex = %d",
1754 hw->speed, hw->duplex);
1755 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1756 config = &local_config;
1758 vtpci_read_dev_config(hw,
1759 offsetof(struct virtio_net_config, mac),
1760 &config->mac, sizeof(config->mac));
1762 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1763 vtpci_read_dev_config(hw,
1764 offsetof(struct virtio_net_config, status),
1765 &config->status, sizeof(config->status));
1768 "VIRTIO_NET_F_STATUS is not supported");
1772 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1773 vtpci_read_dev_config(hw,
1774 offsetof(struct virtio_net_config, max_virtqueue_pairs),
1775 &config->max_virtqueue_pairs,
1776 sizeof(config->max_virtqueue_pairs));
1779 "VIRTIO_NET_F_MQ is not supported");
1780 config->max_virtqueue_pairs = 1;
1783 hw->max_queue_pairs = config->max_virtqueue_pairs;
1785 if (vtpci_with_feature(hw, VIRTIO_NET_F_MTU)) {
1786 vtpci_read_dev_config(hw,
1787 offsetof(struct virtio_net_config, mtu),
1789 sizeof(config->mtu));
1792 * MTU value has already been checked at negotiation
1793 * time, but check again in case it has changed since
1794 * then, which should not happen.
1796 if (config->mtu < RTE_ETHER_MIN_MTU) {
1797 PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1802 hw->max_mtu = config->mtu;
1803 /* Set initial MTU to maximum one supported by vhost */
1804 eth_dev->data->mtu = config->mtu;
1807 hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1808 VLAN_TAG_LEN - hw->vtnet_hdr_size;
1811 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1812 config->max_virtqueue_pairs);
1813 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1815 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1816 config->mac[0], config->mac[1],
1817 config->mac[2], config->mac[3],
1818 config->mac[4], config->mac[5]);
1820 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1821 hw->max_queue_pairs = 1;
1822 hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1823 VLAN_TAG_LEN - hw->vtnet_hdr_size;
1826 ret = virtio_alloc_queues(eth_dev);
1830 if (eth_dev->data->dev_conf.intr_conf.rxq) {
1831 if (virtio_configure_intr(eth_dev) < 0) {
1832 PMD_INIT_LOG(ERR, "failed to configure interrupt");
1833 virtio_free_queues(hw);
1838 vtpci_reinit_complete(hw);
1841 PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1842 eth_dev->data->port_id, pci_dev->id.vendor_id,
1843 pci_dev->id.device_id);
1849 * Remap the PCI device again (IO port map for legacy device and
1850 * memory map for modern device), so that the secondary process
1851 * could have the PCI initiated correctly.
1854 virtio_remap_pci(struct rte_pci_device *pci_dev, struct virtio_hw *hw)
1858 * We don't have to re-parse the PCI config space, since
1859 * rte_pci_map_device() makes sure the mapped address
1860 * in secondary process would equal to the one mapped in
1861 * the primary process: error will be returned if that
1862 * requirement is not met.
1864 * That said, we could simply reuse all cap pointers
1865 * (such as dev_cfg, common_cfg, etc.) parsed from the
1866 * primary process, which is stored in shared memory.
1868 if (rte_pci_map_device(pci_dev)) {
1869 PMD_INIT_LOG(DEBUG, "failed to map pci device!");
1873 if (rte_pci_ioport_map(pci_dev, 0, VTPCI_IO(hw)) < 0)
1881 virtio_set_vtpci_ops(struct virtio_hw *hw)
1883 #ifdef RTE_VIRTIO_USER
1884 if (hw->virtio_user_dev)
1885 VTPCI_OPS(hw) = &virtio_user_ops;
1889 VTPCI_OPS(hw) = &modern_ops;
1891 VTPCI_OPS(hw) = &legacy_ops;
1895 * This function is based on probe() function in virtio_pci.c
1896 * It returns 0 on success.
1899 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1901 struct virtio_hw *hw = eth_dev->data->dev_private;
1902 uint32_t speed = ETH_SPEED_NUM_UNKNOWN;
1906 if (sizeof(struct virtio_net_hdr_mrg_rxbuf) > RTE_PKTMBUF_HEADROOM) {
1908 "Not sufficient headroom required = %d, avail = %d",
1909 (int)sizeof(struct virtio_net_hdr_mrg_rxbuf),
1910 RTE_PKTMBUF_HEADROOM);
1915 eth_dev->dev_ops = &virtio_eth_dev_ops;
1916 eth_dev->rx_descriptor_done = virtio_dev_rx_queue_done;
1918 if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1919 if (!hw->virtio_user_dev) {
1920 ret = virtio_remap_pci(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1925 virtio_set_vtpci_ops(hw);
1926 set_rxtx_funcs(eth_dev);
1930 ret = virtio_dev_devargs_parse(eth_dev->device->devargs,
1931 NULL, &speed, &vectorized);
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;
2044 case ETH_SPEED_NUM_200G:
2045 return ETH_LINK_SPEED_200G;
2051 static int vectorized_check_handler(__rte_unused const char *key,
2052 const char *value, void *ret_val)
2054 if (strcmp(value, "1") == 0)
2055 *(int *)ret_val = 1;
2057 *(int *)ret_val = 0;
2062 #define VIRTIO_ARG_SPEED "speed"
2063 #define VIRTIO_ARG_VDPA "vdpa"
2064 #define VIRTIO_ARG_VECTORIZED "vectorized"
2068 link_speed_handler(const char *key __rte_unused,
2069 const char *value, void *ret_val)
2072 if (!value || !ret_val)
2074 val = strtoul(value, NULL, 0);
2075 /* validate input */
2076 if (virtio_dev_speed_capa_get(val) == 0)
2078 *(uint32_t *)ret_val = val;
2085 virtio_dev_devargs_parse(struct rte_devargs *devargs, int *vdpa,
2086 uint32_t *speed, int *vectorized)
2088 struct rte_kvargs *kvlist;
2091 if (devargs == NULL)
2094 kvlist = rte_kvargs_parse(devargs->args, NULL);
2095 if (kvlist == NULL) {
2096 PMD_INIT_LOG(ERR, "error when parsing param");
2099 if (vdpa && rte_kvargs_count(kvlist, VIRTIO_ARG_VDPA) == 1) {
2100 /* vdpa mode selected when there's a key-value pair:
2103 ret = rte_kvargs_process(kvlist, VIRTIO_ARG_VDPA,
2104 vdpa_check_handler, vdpa);
2106 PMD_INIT_LOG(ERR, "Failed to parse %s",
2111 if (speed && rte_kvargs_count(kvlist, VIRTIO_ARG_SPEED) == 1) {
2112 ret = rte_kvargs_process(kvlist,
2114 link_speed_handler, speed);
2116 PMD_INIT_LOG(ERR, "Failed to parse %s",
2123 rte_kvargs_count(kvlist, VIRTIO_ARG_VECTORIZED) == 1) {
2124 ret = rte_kvargs_process(kvlist,
2125 VIRTIO_ARG_VECTORIZED,
2126 vectorized_check_handler, vectorized);
2128 PMD_INIT_LOG(ERR, "Failed to parse %s",
2129 VIRTIO_ARG_VECTORIZED);
2135 rte_kvargs_free(kvlist);
2139 static int eth_virtio_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2140 struct rte_pci_device *pci_dev)
2145 ret = virtio_dev_devargs_parse(pci_dev->device.devargs, &vdpa, NULL,
2148 PMD_INIT_LOG(ERR, "devargs parsing is failed");
2151 /* virtio pmd skips probe if device needs to work in vdpa mode */
2155 return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct virtio_hw),
2156 eth_virtio_dev_init);
2159 static int eth_virtio_pci_remove(struct rte_pci_device *pci_dev)
2163 ret = rte_eth_dev_pci_generic_remove(pci_dev, eth_virtio_dev_uninit);
2164 /* Port has already been released by close. */
2170 static struct rte_pci_driver rte_virtio_pmd = {
2172 .name = "net_virtio",
2174 .id_table = pci_id_virtio_map,
2176 .probe = eth_virtio_pci_probe,
2177 .remove = eth_virtio_pci_remove,
2180 RTE_INIT(rte_virtio_pmd_init)
2182 rte_eal_iopl_init();
2183 rte_pci_register(&rte_virtio_pmd);
2187 rx_offload_enabled(struct virtio_hw *hw)
2189 return vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM) ||
2190 vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2191 vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6);
2195 tx_offload_enabled(struct virtio_hw *hw)
2197 return vtpci_with_feature(hw, VIRTIO_NET_F_CSUM) ||
2198 vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO4) ||
2199 vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO6);
2203 * Configure virtio device
2204 * It returns 0 on success.
2207 virtio_dev_configure(struct rte_eth_dev *dev)
2209 const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2210 const struct rte_eth_txmode *txmode = &dev->data->dev_conf.txmode;
2211 struct virtio_hw *hw = dev->data->dev_private;
2212 uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
2214 uint64_t rx_offloads = rxmode->offloads;
2215 uint64_t tx_offloads = txmode->offloads;
2216 uint64_t req_features;
2219 PMD_INIT_LOG(DEBUG, "configure");
2220 req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
2222 if (rxmode->mq_mode != ETH_MQ_RX_NONE) {
2224 "Unsupported Rx multi queue mode %d",
2229 if (txmode->mq_mode != ETH_MQ_TX_NONE) {
2231 "Unsupported Tx multi queue mode %d",
2236 if (dev->data->dev_conf.intr_conf.rxq) {
2237 ret = virtio_init_device(dev, hw->req_guest_features);
2242 if (rxmode->max_rx_pkt_len > hw->max_mtu + ether_hdr_len)
2243 req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
2245 if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2246 DEV_RX_OFFLOAD_TCP_CKSUM))
2247 req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
2249 if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO)
2251 (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2252 (1ULL << VIRTIO_NET_F_GUEST_TSO6);
2254 if (tx_offloads & (DEV_TX_OFFLOAD_UDP_CKSUM |
2255 DEV_TX_OFFLOAD_TCP_CKSUM))
2256 req_features |= (1ULL << VIRTIO_NET_F_CSUM);
2258 if (tx_offloads & DEV_TX_OFFLOAD_TCP_TSO)
2260 (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2261 (1ULL << VIRTIO_NET_F_HOST_TSO6);
2263 /* if request features changed, reinit the device */
2264 if (req_features != hw->req_guest_features) {
2265 ret = virtio_init_device(dev, req_features);
2270 if ((rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2271 DEV_RX_OFFLOAD_TCP_CKSUM)) &&
2272 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
2274 "rx checksum not available on this host");
2278 if ((rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) &&
2279 (!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2280 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6))) {
2282 "Large Receive Offload not available on this host");
2286 /* start control queue */
2287 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
2288 virtio_dev_cq_start(dev);
2290 if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
2293 if ((rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
2294 && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2296 "vlan filtering not available on this host");
2300 hw->has_tx_offload = tx_offload_enabled(hw);
2301 hw->has_rx_offload = rx_offload_enabled(hw);
2303 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2304 /* Enable vector (0) for Link State Intrerrupt */
2305 if (VTPCI_OPS(hw)->set_config_irq(hw, 0) ==
2306 VIRTIO_MSI_NO_VECTOR) {
2307 PMD_DRV_LOG(ERR, "failed to set config vector");
2311 if (vtpci_packed_queue(hw)) {
2312 #if defined(RTE_ARCH_X86_64) && defined(CC_AVX512_SUPPORT)
2313 if ((hw->use_vec_rx || hw->use_vec_tx) &&
2314 (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX512F) ||
2315 !vtpci_with_feature(hw, VIRTIO_F_IN_ORDER) ||
2316 !vtpci_with_feature(hw, VIRTIO_F_VERSION_1))) {
2318 "disabled packed ring vectorized path for requirements not met");
2327 if (hw->use_vec_rx) {
2328 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2330 "disabled packed ring vectorized rx for mrg_rxbuf enabled");
2334 if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) {
2336 "disabled packed ring vectorized rx for TCP_LRO enabled");
2341 if (vtpci_with_feature(hw, VIRTIO_F_IN_ORDER)) {
2342 hw->use_inorder_tx = 1;
2343 hw->use_inorder_rx = 1;
2347 if (hw->use_vec_rx) {
2348 #if defined RTE_ARCH_ARM
2349 if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
2351 "disabled split ring vectorized path for requirement not met");
2355 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2357 "disabled split ring vectorized rx for mrg_rxbuf enabled");
2361 if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2362 DEV_RX_OFFLOAD_TCP_CKSUM |
2363 DEV_RX_OFFLOAD_TCP_LRO |
2364 DEV_RX_OFFLOAD_VLAN_STRIP)) {
2366 "disabled split ring vectorized rx for offloading enabled");
2377 virtio_dev_start(struct rte_eth_dev *dev)
2379 uint16_t nb_queues, i;
2380 struct virtnet_rx *rxvq;
2381 struct virtnet_tx *txvq __rte_unused;
2382 struct virtio_hw *hw = dev->data->dev_private;
2385 /* Finish the initialization of the queues */
2386 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2387 ret = virtio_dev_rx_queue_setup_finish(dev, i);
2391 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2392 ret = virtio_dev_tx_queue_setup_finish(dev, i);
2397 /* check if lsc interrupt feature is enabled */
2398 if (dev->data->dev_conf.intr_conf.lsc) {
2399 if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
2400 PMD_DRV_LOG(ERR, "link status not supported by host");
2405 /* Enable uio/vfio intr/eventfd mapping: althrough we already did that
2406 * in device configure, but it could be unmapped when device is
2409 if (dev->data->dev_conf.intr_conf.lsc ||
2410 dev->data->dev_conf.intr_conf.rxq) {
2411 virtio_intr_disable(dev);
2413 /* Setup interrupt callback */
2414 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2415 rte_intr_callback_register(dev->intr_handle,
2416 virtio_interrupt_handler,
2419 if (virtio_intr_enable(dev) < 0) {
2420 PMD_DRV_LOG(ERR, "interrupt enable failed");
2425 /*Notify the backend
2426 *Otherwise the tap backend might already stop its queue due to fullness.
2427 *vhost backend will have no chance to be waked up
2429 nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
2430 if (hw->max_queue_pairs > 1) {
2431 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
2435 PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
2437 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2438 rxvq = dev->data->rx_queues[i];
2439 /* Flush the old packets */
2440 virtqueue_rxvq_flush(rxvq->vq);
2441 virtqueue_notify(rxvq->vq);
2444 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2445 txvq = dev->data->tx_queues[i];
2446 virtqueue_notify(txvq->vq);
2449 PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
2451 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2452 rxvq = dev->data->rx_queues[i];
2453 VIRTQUEUE_DUMP(rxvq->vq);
2456 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2457 txvq = dev->data->tx_queues[i];
2458 VIRTQUEUE_DUMP(txvq->vq);
2461 set_rxtx_funcs(dev);
2464 /* Initialize Link state */
2465 virtio_dev_link_update(dev, 0);
2470 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
2472 struct virtio_hw *hw = dev->data->dev_private;
2473 uint16_t nr_vq = virtio_get_nr_vq(hw);
2474 const char *type __rte_unused;
2475 unsigned int i, mbuf_num = 0;
2476 struct virtqueue *vq;
2477 struct rte_mbuf *buf;
2480 if (hw->vqs == NULL)
2483 for (i = 0; i < nr_vq; i++) {
2488 queue_type = virtio_get_queue_type(hw, i);
2489 if (queue_type == VTNET_RQ)
2491 else if (queue_type == VTNET_TQ)
2497 "Before freeing %s[%d] used and unused buf",
2501 while ((buf = virtqueue_detach_unused(vq)) != NULL) {
2502 rte_pktmbuf_free(buf);
2507 "After freeing %s[%d] used and unused buf",
2512 PMD_INIT_LOG(DEBUG, "%d mbufs freed", mbuf_num);
2516 * Stop device: disable interrupt and mark link down
2519 virtio_dev_stop(struct rte_eth_dev *dev)
2521 struct virtio_hw *hw = dev->data->dev_private;
2522 struct rte_eth_link link;
2523 struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
2525 PMD_INIT_LOG(DEBUG, "stop");
2527 rte_spinlock_lock(&hw->state_lock);
2530 hw->started = false;
2532 if (intr_conf->lsc || intr_conf->rxq) {
2533 virtio_intr_disable(dev);
2535 /* Reset interrupt callback */
2536 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
2537 rte_intr_callback_unregister(dev->intr_handle,
2538 virtio_interrupt_handler,
2543 memset(&link, 0, sizeof(link));
2544 rte_eth_linkstatus_set(dev, &link);
2546 rte_spinlock_unlock(&hw->state_lock);
2550 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
2552 struct rte_eth_link link;
2554 struct virtio_hw *hw = dev->data->dev_private;
2556 memset(&link, 0, sizeof(link));
2557 link.link_duplex = hw->duplex;
2558 link.link_speed = hw->speed;
2559 link.link_autoneg = ETH_LINK_AUTONEG;
2562 link.link_status = ETH_LINK_DOWN;
2563 link.link_speed = ETH_SPEED_NUM_NONE;
2564 } else if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
2565 PMD_INIT_LOG(DEBUG, "Get link status from hw");
2566 vtpci_read_dev_config(hw,
2567 offsetof(struct virtio_net_config, status),
2568 &status, sizeof(status));
2569 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
2570 link.link_status = ETH_LINK_DOWN;
2571 link.link_speed = ETH_SPEED_NUM_NONE;
2572 PMD_INIT_LOG(DEBUG, "Port %d is down",
2573 dev->data->port_id);
2575 link.link_status = ETH_LINK_UP;
2576 PMD_INIT_LOG(DEBUG, "Port %d is up",
2577 dev->data->port_id);
2580 link.link_status = ETH_LINK_UP;
2583 return rte_eth_linkstatus_set(dev, &link);
2587 virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
2589 const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2590 struct virtio_hw *hw = dev->data->dev_private;
2591 uint64_t offloads = rxmode->offloads;
2593 if (mask & ETH_VLAN_FILTER_MASK) {
2594 if ((offloads & DEV_RX_OFFLOAD_VLAN_FILTER) &&
2595 !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2598 "vlan filtering not available on this host");
2604 if (mask & ETH_VLAN_STRIP_MASK)
2605 hw->vlan_strip = !!(offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
2611 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2613 uint64_t tso_mask, host_features;
2614 struct virtio_hw *hw = dev->data->dev_private;
2615 dev_info->speed_capa = virtio_dev_speed_capa_get(hw->speed);
2617 dev_info->max_rx_queues =
2618 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
2619 dev_info->max_tx_queues =
2620 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
2621 dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
2622 dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
2623 dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
2625 host_features = VTPCI_OPS(hw)->get_features(hw);
2626 dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
2627 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_JUMBO_FRAME;
2628 if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
2629 dev_info->rx_offload_capa |=
2630 DEV_RX_OFFLOAD_TCP_CKSUM |
2631 DEV_RX_OFFLOAD_UDP_CKSUM;
2633 if (host_features & (1ULL << VIRTIO_NET_F_CTRL_VLAN))
2634 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_VLAN_FILTER;
2635 tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2636 (1ULL << VIRTIO_NET_F_GUEST_TSO6);
2637 if ((host_features & tso_mask) == tso_mask)
2638 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
2640 dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
2641 DEV_TX_OFFLOAD_VLAN_INSERT;
2642 if (host_features & (1ULL << VIRTIO_NET_F_CSUM)) {
2643 dev_info->tx_offload_capa |=
2644 DEV_TX_OFFLOAD_UDP_CKSUM |
2645 DEV_TX_OFFLOAD_TCP_CKSUM;
2647 tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2648 (1ULL << VIRTIO_NET_F_HOST_TSO6);
2649 if ((host_features & tso_mask) == tso_mask)
2650 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
2656 * It enables testpmd to collect per queue stats.
2659 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
2660 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
2661 __rte_unused uint8_t is_rx)
2666 RTE_PMD_EXPORT_NAME(net_virtio, __COUNTER__);
2667 RTE_PMD_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map);
2668 RTE_PMD_REGISTER_KMOD_DEP(net_virtio, "* igb_uio | uio_pci_generic | vfio-pci");
2669 RTE_LOG_REGISTER(virtio_logtype_init, pmd.net.virtio.init, NOTICE);
2670 RTE_LOG_REGISTER(virtio_logtype_driver, pmd.net.virtio.driver, NOTICE);