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 void virtio_dev_promiscuous_enable(struct rte_eth_dev *dev);
45 static void virtio_dev_promiscuous_disable(struct rte_eth_dev *dev);
46 static void virtio_dev_allmulticast_enable(struct rte_eth_dev *dev);
47 static void virtio_dev_allmulticast_disable(struct rte_eth_dev *dev);
48 static void virtio_dev_info_get(struct rte_eth_dev *dev,
49 struct rte_eth_dev_info *dev_info);
50 static int virtio_dev_link_update(struct rte_eth_dev *dev,
51 int wait_to_complete);
52 static int virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask);
54 static void virtio_set_hwaddr(struct virtio_hw *hw);
55 static void virtio_get_hwaddr(struct virtio_hw *hw);
57 static int virtio_dev_stats_get(struct rte_eth_dev *dev,
58 struct rte_eth_stats *stats);
59 static int virtio_dev_xstats_get(struct rte_eth_dev *dev,
60 struct rte_eth_xstat *xstats, unsigned n);
61 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
62 struct rte_eth_xstat_name *xstats_names,
64 static void virtio_dev_stats_reset(struct rte_eth_dev *dev);
65 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev);
66 static int virtio_vlan_filter_set(struct rte_eth_dev *dev,
67 uint16_t vlan_id, int on);
68 static int virtio_mac_addr_add(struct rte_eth_dev *dev,
69 struct rte_ether_addr *mac_addr,
70 uint32_t index, uint32_t vmdq);
71 static void virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index);
72 static int virtio_mac_addr_set(struct rte_eth_dev *dev,
73 struct rte_ether_addr *mac_addr);
75 static int virtio_intr_disable(struct rte_eth_dev *dev);
77 static int virtio_dev_queue_stats_mapping_set(
78 struct rte_eth_dev *eth_dev,
83 int virtio_logtype_init;
84 int virtio_logtype_driver;
86 static void virtio_notify_peers(struct rte_eth_dev *dev);
87 static void virtio_ack_link_announce(struct rte_eth_dev *dev);
90 * The set of PCI devices this driver supports
92 static const struct rte_pci_id pci_id_virtio_map[] = {
93 { RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_LEGACY_DEVICEID_NET) },
94 { RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_MODERN_DEVICEID_NET) },
95 { .vendor_id = 0, /* sentinel */ },
98 struct rte_virtio_xstats_name_off {
99 char name[RTE_ETH_XSTATS_NAME_SIZE];
103 /* [rt]x_qX_ is prepended to the name string here */
104 static const struct rte_virtio_xstats_name_off rte_virtio_rxq_stat_strings[] = {
105 {"good_packets", offsetof(struct virtnet_rx, stats.packets)},
106 {"good_bytes", offsetof(struct virtnet_rx, stats.bytes)},
107 {"errors", offsetof(struct virtnet_rx, stats.errors)},
108 {"multicast_packets", offsetof(struct virtnet_rx, stats.multicast)},
109 {"broadcast_packets", offsetof(struct virtnet_rx, stats.broadcast)},
110 {"undersize_packets", offsetof(struct virtnet_rx, stats.size_bins[0])},
111 {"size_64_packets", offsetof(struct virtnet_rx, stats.size_bins[1])},
112 {"size_65_127_packets", offsetof(struct virtnet_rx, stats.size_bins[2])},
113 {"size_128_255_packets", offsetof(struct virtnet_rx, stats.size_bins[3])},
114 {"size_256_511_packets", offsetof(struct virtnet_rx, stats.size_bins[4])},
115 {"size_512_1023_packets", offsetof(struct virtnet_rx, stats.size_bins[5])},
116 {"size_1024_1518_packets", offsetof(struct virtnet_rx, stats.size_bins[6])},
117 {"size_1519_max_packets", offsetof(struct virtnet_rx, stats.size_bins[7])},
120 /* [rt]x_qX_ is prepended to the name string here */
121 static const struct rte_virtio_xstats_name_off rte_virtio_txq_stat_strings[] = {
122 {"good_packets", offsetof(struct virtnet_tx, stats.packets)},
123 {"good_bytes", offsetof(struct virtnet_tx, stats.bytes)},
124 {"errors", offsetof(struct virtnet_tx, stats.errors)},
125 {"multicast_packets", offsetof(struct virtnet_tx, stats.multicast)},
126 {"broadcast_packets", offsetof(struct virtnet_tx, stats.broadcast)},
127 {"undersize_packets", offsetof(struct virtnet_tx, stats.size_bins[0])},
128 {"size_64_packets", offsetof(struct virtnet_tx, stats.size_bins[1])},
129 {"size_65_127_packets", offsetof(struct virtnet_tx, stats.size_bins[2])},
130 {"size_128_255_packets", offsetof(struct virtnet_tx, stats.size_bins[3])},
131 {"size_256_511_packets", offsetof(struct virtnet_tx, stats.size_bins[4])},
132 {"size_512_1023_packets", offsetof(struct virtnet_tx, stats.size_bins[5])},
133 {"size_1024_1518_packets", offsetof(struct virtnet_tx, stats.size_bins[6])},
134 {"size_1519_max_packets", offsetof(struct virtnet_tx, stats.size_bins[7])},
137 #define VIRTIO_NB_RXQ_XSTATS (sizeof(rte_virtio_rxq_stat_strings) / \
138 sizeof(rte_virtio_rxq_stat_strings[0]))
139 #define VIRTIO_NB_TXQ_XSTATS (sizeof(rte_virtio_txq_stat_strings) / \
140 sizeof(rte_virtio_txq_stat_strings[0]))
142 struct virtio_hw_internal virtio_hw_internal[RTE_MAX_ETHPORTS];
144 static struct virtio_pmd_ctrl *
145 virtio_send_command_packed(struct virtnet_ctl *cvq,
146 struct virtio_pmd_ctrl *ctrl,
147 int *dlen, int pkt_num)
149 struct virtqueue *vq = cvq->vq;
151 struct vring_packed_desc *desc = vq->vq_packed.ring.desc;
152 struct virtio_pmd_ctrl *result;
159 * Format is enforced in qemu code:
160 * One TX packet for header;
161 * At least one TX packet per argument;
162 * One RX packet for ACK.
164 head = vq->vq_avail_idx;
165 flags = vq->vq_packed.cached_flags;
166 desc[head].addr = cvq->virtio_net_hdr_mem;
167 desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
170 if (++vq->vq_avail_idx >= vq->vq_nentries) {
171 vq->vq_avail_idx -= vq->vq_nentries;
172 vq->vq_packed.cached_flags ^= VRING_PACKED_DESC_F_AVAIL_USED;
175 for (k = 0; k < pkt_num; k++) {
176 desc[vq->vq_avail_idx].addr = cvq->virtio_net_hdr_mem
177 + sizeof(struct virtio_net_ctrl_hdr)
178 + sizeof(ctrl->status) + sizeof(uint8_t) * sum;
179 desc[vq->vq_avail_idx].len = dlen[k];
180 desc[vq->vq_avail_idx].flags = VRING_DESC_F_NEXT |
181 vq->vq_packed.cached_flags;
185 if (++vq->vq_avail_idx >= vq->vq_nentries) {
186 vq->vq_avail_idx -= vq->vq_nentries;
187 vq->vq_packed.cached_flags ^=
188 VRING_PACKED_DESC_F_AVAIL_USED;
192 desc[vq->vq_avail_idx].addr = cvq->virtio_net_hdr_mem
193 + sizeof(struct virtio_net_ctrl_hdr);
194 desc[vq->vq_avail_idx].len = sizeof(ctrl->status);
195 desc[vq->vq_avail_idx].flags = VRING_DESC_F_WRITE |
196 vq->vq_packed.cached_flags;
199 if (++vq->vq_avail_idx >= vq->vq_nentries) {
200 vq->vq_avail_idx -= vq->vq_nentries;
201 vq->vq_packed.cached_flags ^= VRING_PACKED_DESC_F_AVAIL_USED;
204 virtio_wmb(vq->hw->weak_barriers);
205 desc[head].flags = VRING_DESC_F_NEXT | flags;
207 virtio_wmb(vq->hw->weak_barriers);
208 virtqueue_notify(vq);
210 /* wait for used descriptors in virtqueue */
211 while (!desc_is_used(&desc[head], vq))
214 virtio_rmb(vq->hw->weak_barriers);
216 /* now get used descriptors */
217 vq->vq_free_cnt += nb_descs;
218 vq->vq_used_cons_idx += nb_descs;
219 if (vq->vq_used_cons_idx >= vq->vq_nentries) {
220 vq->vq_used_cons_idx -= vq->vq_nentries;
221 vq->vq_packed.used_wrap_counter ^= 1;
224 PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\n"
225 "vq->vq_avail_idx=%d\n"
226 "vq->vq_used_cons_idx=%d\n"
227 "vq->vq_packed.cached_flags=0x%x\n"
228 "vq->vq_packed.used_wrap_counter=%d\n",
231 vq->vq_used_cons_idx,
232 vq->vq_packed.cached_flags,
233 vq->vq_packed.used_wrap_counter);
235 result = cvq->virtio_net_hdr_mz->addr;
239 static struct virtio_pmd_ctrl *
240 virtio_send_command_split(struct virtnet_ctl *cvq,
241 struct virtio_pmd_ctrl *ctrl,
242 int *dlen, int pkt_num)
244 struct virtio_pmd_ctrl *result;
245 struct virtqueue *vq = cvq->vq;
249 head = vq->vq_desc_head_idx;
252 * Format is enforced in qemu code:
253 * One TX packet for header;
254 * At least one TX packet per argument;
255 * One RX packet for ACK.
257 vq->vq_split.ring.desc[head].flags = VRING_DESC_F_NEXT;
258 vq->vq_split.ring.desc[head].addr = cvq->virtio_net_hdr_mem;
259 vq->vq_split.ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
261 i = vq->vq_split.ring.desc[head].next;
263 for (k = 0; k < pkt_num; k++) {
264 vq->vq_split.ring.desc[i].flags = VRING_DESC_F_NEXT;
265 vq->vq_split.ring.desc[i].addr = cvq->virtio_net_hdr_mem
266 + sizeof(struct virtio_net_ctrl_hdr)
267 + sizeof(ctrl->status) + sizeof(uint8_t)*sum;
268 vq->vq_split.ring.desc[i].len = dlen[k];
271 i = vq->vq_split.ring.desc[i].next;
274 vq->vq_split.ring.desc[i].flags = VRING_DESC_F_WRITE;
275 vq->vq_split.ring.desc[i].addr = cvq->virtio_net_hdr_mem
276 + sizeof(struct virtio_net_ctrl_hdr);
277 vq->vq_split.ring.desc[i].len = sizeof(ctrl->status);
280 vq->vq_desc_head_idx = vq->vq_split.ring.desc[i].next;
282 vq_update_avail_ring(vq, head);
283 vq_update_avail_idx(vq);
285 PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
287 virtqueue_notify(vq);
290 while (VIRTQUEUE_NUSED(vq) == 0) {
295 while (VIRTQUEUE_NUSED(vq)) {
296 uint32_t idx, desc_idx, used_idx;
297 struct vring_used_elem *uep;
299 used_idx = (uint32_t)(vq->vq_used_cons_idx
300 & (vq->vq_nentries - 1));
301 uep = &vq->vq_split.ring.used->ring[used_idx];
302 idx = (uint32_t) uep->id;
305 while (vq->vq_split.ring.desc[desc_idx].flags &
307 desc_idx = vq->vq_split.ring.desc[desc_idx].next;
311 vq->vq_split.ring.desc[desc_idx].next = vq->vq_desc_head_idx;
312 vq->vq_desc_head_idx = idx;
314 vq->vq_used_cons_idx++;
318 PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
319 vq->vq_free_cnt, vq->vq_desc_head_idx);
321 result = cvq->virtio_net_hdr_mz->addr;
326 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
327 int *dlen, int pkt_num)
329 virtio_net_ctrl_ack status = ~0;
330 struct virtio_pmd_ctrl *result;
331 struct virtqueue *vq;
333 ctrl->status = status;
335 if (!cvq || !cvq->vq) {
336 PMD_INIT_LOG(ERR, "Control queue is not supported.");
340 rte_spinlock_lock(&cvq->lock);
343 PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
344 "vq->hw->cvq = %p vq = %p",
345 vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
347 if (vq->vq_free_cnt < pkt_num + 2 || pkt_num < 1) {
348 rte_spinlock_unlock(&cvq->lock);
352 memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
353 sizeof(struct virtio_pmd_ctrl));
355 if (vtpci_packed_queue(vq->hw))
356 result = virtio_send_command_packed(cvq, ctrl, dlen, pkt_num);
358 result = virtio_send_command_split(cvq, ctrl, dlen, pkt_num);
360 rte_spinlock_unlock(&cvq->lock);
361 return result->status;
365 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
367 struct virtio_hw *hw = dev->data->dev_private;
368 struct virtio_pmd_ctrl ctrl;
372 ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
373 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
374 memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
376 dlen[0] = sizeof(uint16_t);
378 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
380 PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
381 "failed, this is too late now...");
389 virtio_dev_queue_release(void *queue __rte_unused)
395 virtio_get_nr_vq(struct virtio_hw *hw)
397 uint16_t nr_vq = hw->max_queue_pairs * 2;
399 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
406 virtio_init_vring(struct virtqueue *vq)
408 int size = vq->vq_nentries;
409 uint8_t *ring_mem = vq->vq_ring_virt_mem;
411 PMD_INIT_FUNC_TRACE();
413 memset(ring_mem, 0, vq->vq_ring_size);
415 vq->vq_used_cons_idx = 0;
416 vq->vq_desc_head_idx = 0;
417 vq->vq_avail_idx = 0;
418 vq->vq_desc_tail_idx = (uint16_t)(vq->vq_nentries - 1);
419 vq->vq_free_cnt = vq->vq_nentries;
420 memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries);
421 if (vtpci_packed_queue(vq->hw)) {
422 vring_init_packed(&vq->vq_packed.ring, ring_mem,
423 VIRTIO_PCI_VRING_ALIGN, size);
424 vring_desc_init_packed(vq, size);
426 struct vring *vr = &vq->vq_split.ring;
428 vring_init_split(vr, ring_mem, VIRTIO_PCI_VRING_ALIGN, size);
429 vring_desc_init_split(vr->desc, size);
432 * Disable device(host) interrupting guest
434 virtqueue_disable_intr(vq);
438 virtio_init_queue(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx)
440 char vq_name[VIRTQUEUE_MAX_NAME_SZ];
441 char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
442 const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
443 unsigned int vq_size, size;
444 struct virtio_hw *hw = dev->data->dev_private;
445 struct virtnet_rx *rxvq = NULL;
446 struct virtnet_tx *txvq = NULL;
447 struct virtnet_ctl *cvq = NULL;
448 struct virtqueue *vq;
449 size_t sz_hdr_mz = 0;
450 void *sw_ring = NULL;
451 int queue_type = virtio_get_queue_type(hw, vtpci_queue_idx);
453 int numa_node = dev->device->numa_node;
455 PMD_INIT_LOG(INFO, "setting up queue: %u on NUMA node %d",
456 vtpci_queue_idx, numa_node);
459 * Read the virtqueue size from the Queue Size field
460 * Always power of 2 and if 0 virtqueue does not exist
462 vq_size = VTPCI_OPS(hw)->get_queue_num(hw, vtpci_queue_idx);
463 PMD_INIT_LOG(DEBUG, "vq_size: %u", vq_size);
465 PMD_INIT_LOG(ERR, "virtqueue does not exist");
469 if (!rte_is_power_of_2(vq_size)) {
470 PMD_INIT_LOG(ERR, "virtqueue size is not powerof 2");
474 snprintf(vq_name, sizeof(vq_name), "port%d_vq%d",
475 dev->data->port_id, vtpci_queue_idx);
477 size = RTE_ALIGN_CEIL(sizeof(*vq) +
478 vq_size * sizeof(struct vq_desc_extra),
479 RTE_CACHE_LINE_SIZE);
480 if (queue_type == VTNET_TQ) {
482 * For each xmit packet, allocate a virtio_net_hdr
483 * and indirect ring elements
485 sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
486 } else if (queue_type == VTNET_CQ) {
487 /* Allocate a page for control vq command, data and status */
488 sz_hdr_mz = PAGE_SIZE;
491 vq = rte_zmalloc_socket(vq_name, size, RTE_CACHE_LINE_SIZE,
494 PMD_INIT_LOG(ERR, "can not allocate vq");
497 hw->vqs[vtpci_queue_idx] = vq;
500 vq->vq_queue_index = vtpci_queue_idx;
501 vq->vq_nentries = vq_size;
502 if (vtpci_packed_queue(hw)) {
503 vq->vq_packed.used_wrap_counter = 1;
504 vq->vq_packed.cached_flags = VRING_PACKED_DESC_F_AVAIL;
505 vq->vq_packed.event_flags_shadow = 0;
506 if (queue_type == VTNET_RQ)
507 vq->vq_packed.cached_flags |= VRING_DESC_F_WRITE;
511 * Reserve a memzone for vring elements
513 size = vring_size(hw, vq_size, VIRTIO_PCI_VRING_ALIGN);
514 vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
515 PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
516 size, vq->vq_ring_size);
518 mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
519 numa_node, RTE_MEMZONE_IOVA_CONTIG,
520 VIRTIO_PCI_VRING_ALIGN);
522 if (rte_errno == EEXIST)
523 mz = rte_memzone_lookup(vq_name);
530 memset(mz->addr, 0, mz->len);
532 vq->vq_ring_mem = mz->iova;
533 vq->vq_ring_virt_mem = mz->addr;
534 PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem: 0x%" PRIx64,
536 PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64,
537 (uint64_t)(uintptr_t)mz->addr);
539 virtio_init_vring(vq);
542 snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_vq%d_hdr",
543 dev->data->port_id, vtpci_queue_idx);
544 hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
545 numa_node, RTE_MEMZONE_IOVA_CONTIG,
546 RTE_CACHE_LINE_SIZE);
547 if (hdr_mz == NULL) {
548 if (rte_errno == EEXIST)
549 hdr_mz = rte_memzone_lookup(vq_hdr_name);
550 if (hdr_mz == NULL) {
557 if (queue_type == VTNET_RQ) {
558 size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
559 sizeof(vq->sw_ring[0]);
561 sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
562 RTE_CACHE_LINE_SIZE, numa_node);
564 PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
569 vq->sw_ring = sw_ring;
572 rxvq->port_id = dev->data->port_id;
574 } else if (queue_type == VTNET_TQ) {
577 txvq->port_id = dev->data->port_id;
579 txvq->virtio_net_hdr_mz = hdr_mz;
580 txvq->virtio_net_hdr_mem = hdr_mz->iova;
581 } else if (queue_type == VTNET_CQ) {
585 cvq->virtio_net_hdr_mz = hdr_mz;
586 cvq->virtio_net_hdr_mem = hdr_mz->iova;
587 memset(cvq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
592 /* For virtio_user case (that is when hw->dev is NULL), we use
593 * virtual address. And we need properly set _offset_, please see
594 * VIRTIO_MBUF_DATA_DMA_ADDR in virtqueue.h for more information.
596 if (!hw->virtio_user_dev)
597 vq->offset = offsetof(struct rte_mbuf, buf_iova);
599 vq->vq_ring_mem = (uintptr_t)mz->addr;
600 vq->offset = offsetof(struct rte_mbuf, buf_addr);
601 if (queue_type == VTNET_TQ)
602 txvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
603 else if (queue_type == VTNET_CQ)
604 cvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
607 if (queue_type == VTNET_TQ) {
608 struct virtio_tx_region *txr;
612 memset(txr, 0, vq_size * sizeof(*txr));
613 for (i = 0; i < vq_size; i++) {
614 struct vring_desc *start_dp = txr[i].tx_indir;
616 /* first indirect descriptor is always the tx header */
617 if (!vtpci_packed_queue(hw)) {
618 vring_desc_init_split(start_dp,
619 RTE_DIM(txr[i].tx_indir));
620 start_dp->addr = txvq->virtio_net_hdr_mem
622 + offsetof(struct virtio_tx_region,
624 start_dp->len = hw->vtnet_hdr_size;
625 start_dp->flags = VRING_DESC_F_NEXT;
630 if (VTPCI_OPS(hw)->setup_queue(hw, vq) < 0) {
631 PMD_INIT_LOG(ERR, "setup_queue failed");
639 rte_memzone_free(hdr_mz);
640 rte_memzone_free(mz);
647 virtio_free_queues(struct virtio_hw *hw)
649 uint16_t nr_vq = virtio_get_nr_vq(hw);
650 struct virtqueue *vq;
657 for (i = 0; i < nr_vq; i++) {
662 queue_type = virtio_get_queue_type(hw, i);
663 if (queue_type == VTNET_RQ) {
664 rte_free(vq->sw_ring);
665 rte_memzone_free(vq->rxq.mz);
666 } else if (queue_type == VTNET_TQ) {
667 rte_memzone_free(vq->txq.mz);
668 rte_memzone_free(vq->txq.virtio_net_hdr_mz);
670 rte_memzone_free(vq->cq.mz);
671 rte_memzone_free(vq->cq.virtio_net_hdr_mz);
683 virtio_alloc_queues(struct rte_eth_dev *dev)
685 struct virtio_hw *hw = dev->data->dev_private;
686 uint16_t nr_vq = virtio_get_nr_vq(hw);
690 hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
692 PMD_INIT_LOG(ERR, "failed to allocate vqs");
696 for (i = 0; i < nr_vq; i++) {
697 ret = virtio_init_queue(dev, i);
699 virtio_free_queues(hw);
707 static void virtio_queues_unbind_intr(struct rte_eth_dev *dev);
710 virtio_dev_close(struct rte_eth_dev *dev)
712 struct virtio_hw *hw = dev->data->dev_private;
713 struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
715 PMD_INIT_LOG(DEBUG, "virtio_dev_close");
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));
751 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
753 struct virtio_hw *hw = dev->data->dev_private;
754 struct virtio_pmd_ctrl ctrl;
758 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
759 PMD_INIT_LOG(INFO, "host does not support rx control");
763 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
764 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
768 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
770 PMD_INIT_LOG(ERR, "Failed to enable promisc");
774 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
776 struct virtio_hw *hw = dev->data->dev_private;
777 struct virtio_pmd_ctrl ctrl;
781 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
782 PMD_INIT_LOG(INFO, "host does not support rx control");
786 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
787 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
791 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
793 PMD_INIT_LOG(ERR, "Failed to disable promisc");
797 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
799 struct virtio_hw *hw = dev->data->dev_private;
800 struct virtio_pmd_ctrl ctrl;
804 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
805 PMD_INIT_LOG(INFO, "host does not support rx control");
809 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
810 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
814 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
816 PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
820 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
822 struct virtio_hw *hw = dev->data->dev_private;
823 struct virtio_pmd_ctrl ctrl;
827 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
828 PMD_INIT_LOG(INFO, "host does not support rx control");
832 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
833 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
837 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
839 PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
842 #define VLAN_TAG_LEN 4 /* 802.3ac tag (not DMA'd) */
844 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
846 struct virtio_hw *hw = dev->data->dev_private;
847 uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
849 uint32_t frame_size = mtu + ether_hdr_len;
850 uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
852 max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
854 if (mtu < RTE_ETHER_MIN_MTU || frame_size > max_frame_size) {
855 PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
856 RTE_ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
863 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
865 struct virtio_hw *hw = dev->data->dev_private;
866 struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
867 struct virtqueue *vq = rxvq->vq;
869 virtqueue_enable_intr(vq);
870 virtio_mb(hw->weak_barriers);
875 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
877 struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
878 struct virtqueue *vq = rxvq->vq;
880 virtqueue_disable_intr(vq);
885 * dev_ops for virtio, bare necessities for basic operation
887 static const struct eth_dev_ops virtio_eth_dev_ops = {
888 .dev_configure = virtio_dev_configure,
889 .dev_start = virtio_dev_start,
890 .dev_stop = virtio_dev_stop,
891 .dev_close = virtio_dev_close,
892 .promiscuous_enable = virtio_dev_promiscuous_enable,
893 .promiscuous_disable = virtio_dev_promiscuous_disable,
894 .allmulticast_enable = virtio_dev_allmulticast_enable,
895 .allmulticast_disable = virtio_dev_allmulticast_disable,
896 .mtu_set = virtio_mtu_set,
897 .dev_infos_get = virtio_dev_info_get,
898 .stats_get = virtio_dev_stats_get,
899 .xstats_get = virtio_dev_xstats_get,
900 .xstats_get_names = virtio_dev_xstats_get_names,
901 .stats_reset = virtio_dev_stats_reset,
902 .xstats_reset = virtio_dev_stats_reset,
903 .link_update = virtio_dev_link_update,
904 .vlan_offload_set = virtio_dev_vlan_offload_set,
905 .rx_queue_setup = virtio_dev_rx_queue_setup,
906 .rx_queue_intr_enable = virtio_dev_rx_queue_intr_enable,
907 .rx_queue_intr_disable = virtio_dev_rx_queue_intr_disable,
908 .rx_queue_release = virtio_dev_queue_release,
909 .rx_descriptor_done = virtio_dev_rx_queue_done,
910 .tx_queue_setup = virtio_dev_tx_queue_setup,
911 .tx_queue_release = virtio_dev_queue_release,
912 /* collect stats per queue */
913 .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
914 .vlan_filter_set = virtio_vlan_filter_set,
915 .mac_addr_add = virtio_mac_addr_add,
916 .mac_addr_remove = virtio_mac_addr_remove,
917 .mac_addr_set = virtio_mac_addr_set,
921 * dev_ops for virtio-user in secondary processes, as we just have
922 * some limited supports currently.
924 const struct eth_dev_ops virtio_user_secondary_eth_dev_ops = {
925 .dev_infos_get = virtio_dev_info_get,
926 .stats_get = virtio_dev_stats_get,
927 .xstats_get = virtio_dev_xstats_get,
928 .xstats_get_names = virtio_dev_xstats_get_names,
929 .stats_reset = virtio_dev_stats_reset,
930 .xstats_reset = virtio_dev_stats_reset,
931 /* collect stats per queue */
932 .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
936 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
940 for (i = 0; i < dev->data->nb_tx_queues; i++) {
941 const struct virtnet_tx *txvq = dev->data->tx_queues[i];
945 stats->opackets += txvq->stats.packets;
946 stats->obytes += txvq->stats.bytes;
947 stats->oerrors += txvq->stats.errors;
949 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
950 stats->q_opackets[i] = txvq->stats.packets;
951 stats->q_obytes[i] = txvq->stats.bytes;
955 for (i = 0; i < dev->data->nb_rx_queues; i++) {
956 const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
960 stats->ipackets += rxvq->stats.packets;
961 stats->ibytes += rxvq->stats.bytes;
962 stats->ierrors += rxvq->stats.errors;
964 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
965 stats->q_ipackets[i] = rxvq->stats.packets;
966 stats->q_ibytes[i] = rxvq->stats.bytes;
970 stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
973 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
974 struct rte_eth_xstat_name *xstats_names,
975 __rte_unused unsigned limit)
981 unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
982 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
984 if (xstats_names != NULL) {
985 /* Note: limit checked in rte_eth_xstats_names() */
987 for (i = 0; i < dev->data->nb_rx_queues; i++) {
988 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
991 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
992 snprintf(xstats_names[count].name,
993 sizeof(xstats_names[count].name),
995 rte_virtio_rxq_stat_strings[t].name);
1000 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1001 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1004 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1005 snprintf(xstats_names[count].name,
1006 sizeof(xstats_names[count].name),
1008 rte_virtio_txq_stat_strings[t].name);
1018 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
1024 unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
1025 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
1030 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1031 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1038 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1039 xstats[count].value = *(uint64_t *)(((char *)rxvq) +
1040 rte_virtio_rxq_stat_strings[t].offset);
1041 xstats[count].id = count;
1046 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1047 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1054 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1055 xstats[count].value = *(uint64_t *)(((char *)txvq) +
1056 rte_virtio_txq_stat_strings[t].offset);
1057 xstats[count].id = count;
1066 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1068 virtio_update_stats(dev, stats);
1074 virtio_dev_stats_reset(struct rte_eth_dev *dev)
1078 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1079 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1083 txvq->stats.packets = 0;
1084 txvq->stats.bytes = 0;
1085 txvq->stats.errors = 0;
1086 txvq->stats.multicast = 0;
1087 txvq->stats.broadcast = 0;
1088 memset(txvq->stats.size_bins, 0,
1089 sizeof(txvq->stats.size_bins[0]) * 8);
1092 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1093 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1097 rxvq->stats.packets = 0;
1098 rxvq->stats.bytes = 0;
1099 rxvq->stats.errors = 0;
1100 rxvq->stats.multicast = 0;
1101 rxvq->stats.broadcast = 0;
1102 memset(rxvq->stats.size_bins, 0,
1103 sizeof(rxvq->stats.size_bins[0]) * 8);
1108 virtio_set_hwaddr(struct virtio_hw *hw)
1110 vtpci_write_dev_config(hw,
1111 offsetof(struct virtio_net_config, mac),
1112 &hw->mac_addr, RTE_ETHER_ADDR_LEN);
1116 virtio_get_hwaddr(struct virtio_hw *hw)
1118 if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
1119 vtpci_read_dev_config(hw,
1120 offsetof(struct virtio_net_config, mac),
1121 &hw->mac_addr, RTE_ETHER_ADDR_LEN);
1123 rte_eth_random_addr(&hw->mac_addr[0]);
1124 virtio_set_hwaddr(hw);
1129 virtio_mac_table_set(struct virtio_hw *hw,
1130 const struct virtio_net_ctrl_mac *uc,
1131 const struct virtio_net_ctrl_mac *mc)
1133 struct virtio_pmd_ctrl ctrl;
1136 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1137 PMD_DRV_LOG(INFO, "host does not support mac table");
1141 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1142 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1144 len[0] = uc->entries * RTE_ETHER_ADDR_LEN + sizeof(uc->entries);
1145 memcpy(ctrl.data, uc, len[0]);
1147 len[1] = mc->entries * RTE_ETHER_ADDR_LEN + sizeof(mc->entries);
1148 memcpy(ctrl.data + len[0], mc, len[1]);
1150 err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1152 PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1157 virtio_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
1158 uint32_t index, uint32_t vmdq __rte_unused)
1160 struct virtio_hw *hw = dev->data->dev_private;
1161 const struct rte_ether_addr *addrs = dev->data->mac_addrs;
1163 struct virtio_net_ctrl_mac *uc, *mc;
1165 if (index >= VIRTIO_MAX_MAC_ADDRS) {
1166 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1170 uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1171 sizeof(uc->entries));
1173 mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1174 sizeof(mc->entries));
1177 for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1178 const struct rte_ether_addr *addr
1179 = (i == index) ? mac_addr : addrs + i;
1180 struct virtio_net_ctrl_mac *tbl
1181 = rte_is_multicast_ether_addr(addr) ? mc : uc;
1183 memcpy(&tbl->macs[tbl->entries++], addr, RTE_ETHER_ADDR_LEN);
1186 return virtio_mac_table_set(hw, uc, mc);
1190 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1192 struct virtio_hw *hw = dev->data->dev_private;
1193 struct rte_ether_addr *addrs = dev->data->mac_addrs;
1194 struct virtio_net_ctrl_mac *uc, *mc;
1197 if (index >= VIRTIO_MAX_MAC_ADDRS) {
1198 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1202 uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1203 sizeof(uc->entries));
1205 mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1206 sizeof(mc->entries));
1209 for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1210 struct virtio_net_ctrl_mac *tbl;
1212 if (i == index || rte_is_zero_ether_addr(addrs + i))
1215 tbl = rte_is_multicast_ether_addr(addrs + i) ? mc : uc;
1216 memcpy(&tbl->macs[tbl->entries++], addrs + i,
1217 RTE_ETHER_ADDR_LEN);
1220 virtio_mac_table_set(hw, uc, mc);
1224 virtio_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1226 struct virtio_hw *hw = dev->data->dev_private;
1228 memcpy(hw->mac_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1230 /* Use atomic update if available */
1231 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1232 struct virtio_pmd_ctrl ctrl;
1233 int len = RTE_ETHER_ADDR_LEN;
1235 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1236 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1238 memcpy(ctrl.data, mac_addr, RTE_ETHER_ADDR_LEN);
1239 return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1242 if (!vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1245 virtio_set_hwaddr(hw);
1250 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1252 struct virtio_hw *hw = dev->data->dev_private;
1253 struct virtio_pmd_ctrl ctrl;
1256 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1259 ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1260 ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1261 memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1262 len = sizeof(vlan_id);
1264 return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1268 virtio_intr_unmask(struct rte_eth_dev *dev)
1270 struct virtio_hw *hw = dev->data->dev_private;
1272 if (rte_intr_ack(dev->intr_handle) < 0)
1275 if (!hw->virtio_user_dev)
1276 hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1282 virtio_intr_enable(struct rte_eth_dev *dev)
1284 struct virtio_hw *hw = dev->data->dev_private;
1286 if (rte_intr_enable(dev->intr_handle) < 0)
1289 if (!hw->virtio_user_dev)
1290 hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1296 virtio_intr_disable(struct rte_eth_dev *dev)
1298 struct virtio_hw *hw = dev->data->dev_private;
1300 if (rte_intr_disable(dev->intr_handle) < 0)
1303 if (!hw->virtio_user_dev)
1304 hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1310 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1312 uint64_t host_features;
1314 /* Prepare guest_features: feature that driver wants to support */
1315 PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1318 /* Read device(host) feature bits */
1319 host_features = VTPCI_OPS(hw)->get_features(hw);
1320 PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1323 /* If supported, ensure MTU value is valid before acknowledging it. */
1324 if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1325 struct virtio_net_config config;
1327 vtpci_read_dev_config(hw,
1328 offsetof(struct virtio_net_config, mtu),
1329 &config.mtu, sizeof(config.mtu));
1331 if (config.mtu < RTE_ETHER_MIN_MTU)
1332 req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1336 * Negotiate features: Subset of device feature bits are written back
1337 * guest feature bits.
1339 hw->guest_features = req_features;
1340 hw->guest_features = vtpci_negotiate_features(hw, host_features);
1341 PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1342 hw->guest_features);
1345 if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1347 "VIRTIO_F_VERSION_1 features is not enabled.");
1350 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1351 if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1353 "failed to set FEATURES_OK status!");
1358 hw->req_guest_features = req_features;
1364 virtio_dev_pause(struct rte_eth_dev *dev)
1366 struct virtio_hw *hw = dev->data->dev_private;
1368 rte_spinlock_lock(&hw->state_lock);
1370 if (hw->started == 0) {
1371 /* Device is just stopped. */
1372 rte_spinlock_unlock(&hw->state_lock);
1377 * Prevent the worker threads from touching queues to avoid contention,
1378 * 1 ms should be enough for the ongoing Tx function to finish.
1385 * Recover hw state to let the worker threads continue.
1388 virtio_dev_resume(struct rte_eth_dev *dev)
1390 struct virtio_hw *hw = dev->data->dev_private;
1393 rte_spinlock_unlock(&hw->state_lock);
1397 * Should be called only after device is paused.
1400 virtio_inject_pkts(struct rte_eth_dev *dev, struct rte_mbuf **tx_pkts,
1403 struct virtio_hw *hw = dev->data->dev_private;
1404 struct virtnet_tx *txvq = dev->data->tx_queues[0];
1407 hw->inject_pkts = tx_pkts;
1408 ret = dev->tx_pkt_burst(txvq, tx_pkts, nb_pkts);
1409 hw->inject_pkts = NULL;
1415 virtio_notify_peers(struct rte_eth_dev *dev)
1417 struct virtio_hw *hw = dev->data->dev_private;
1418 struct virtnet_rx *rxvq;
1419 struct rte_mbuf *rarp_mbuf;
1421 if (!dev->data->rx_queues)
1424 rxvq = dev->data->rx_queues[0];
1428 rarp_mbuf = rte_net_make_rarp_packet(rxvq->mpool,
1429 (struct rte_ether_addr *)hw->mac_addr);
1430 if (rarp_mbuf == NULL) {
1431 PMD_DRV_LOG(ERR, "failed to make RARP packet.");
1435 /* If virtio port just stopped, no need to send RARP */
1436 if (virtio_dev_pause(dev) < 0) {
1437 rte_pktmbuf_free(rarp_mbuf);
1441 virtio_inject_pkts(dev, &rarp_mbuf, 1);
1442 virtio_dev_resume(dev);
1446 virtio_ack_link_announce(struct rte_eth_dev *dev)
1448 struct virtio_hw *hw = dev->data->dev_private;
1449 struct virtio_pmd_ctrl ctrl;
1451 ctrl.hdr.class = VIRTIO_NET_CTRL_ANNOUNCE;
1452 ctrl.hdr.cmd = VIRTIO_NET_CTRL_ANNOUNCE_ACK;
1454 virtio_send_command(hw->cvq, &ctrl, NULL, 0);
1458 * Process virtio config changed interrupt. Call the callback
1459 * if link state changed, generate gratuitous RARP packet if
1460 * the status indicates an ANNOUNCE.
1463 virtio_interrupt_handler(void *param)
1465 struct rte_eth_dev *dev = param;
1466 struct virtio_hw *hw = dev->data->dev_private;
1470 /* Read interrupt status which clears interrupt */
1471 isr = vtpci_isr(hw);
1472 PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1474 if (virtio_intr_unmask(dev) < 0)
1475 PMD_DRV_LOG(ERR, "interrupt enable failed");
1477 if (isr & VIRTIO_PCI_ISR_CONFIG) {
1478 if (virtio_dev_link_update(dev, 0) == 0)
1479 _rte_eth_dev_callback_process(dev,
1480 RTE_ETH_EVENT_INTR_LSC,
1483 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1484 vtpci_read_dev_config(hw,
1485 offsetof(struct virtio_net_config, status),
1486 &status, sizeof(status));
1487 if (status & VIRTIO_NET_S_ANNOUNCE) {
1488 virtio_notify_peers(dev);
1490 virtio_ack_link_announce(dev);
1496 /* set rx and tx handlers according to what is supported */
1498 set_rxtx_funcs(struct rte_eth_dev *eth_dev)
1500 struct virtio_hw *hw = eth_dev->data->dev_private;
1502 eth_dev->tx_pkt_prepare = virtio_xmit_pkts_prepare;
1503 if (vtpci_packed_queue(hw)) {
1505 "virtio: using packed ring %s Tx path on port %u",
1506 hw->use_inorder_tx ? "inorder" : "standard",
1507 eth_dev->data->port_id);
1508 eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed;
1510 if (hw->use_inorder_tx) {
1511 PMD_INIT_LOG(INFO, "virtio: using inorder Tx path on port %u",
1512 eth_dev->data->port_id);
1513 eth_dev->tx_pkt_burst = virtio_xmit_pkts_inorder;
1515 PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
1516 eth_dev->data->port_id);
1517 eth_dev->tx_pkt_burst = virtio_xmit_pkts;
1521 if (vtpci_packed_queue(hw)) {
1522 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1524 "virtio: using packed ring mergeable buffer Rx path on port %u",
1525 eth_dev->data->port_id);
1526 eth_dev->rx_pkt_burst =
1527 &virtio_recv_mergeable_pkts_packed;
1530 "virtio: using packed ring standard Rx path on port %u",
1531 eth_dev->data->port_id);
1532 eth_dev->rx_pkt_burst = &virtio_recv_pkts_packed;
1535 if (hw->use_simple_rx) {
1536 PMD_INIT_LOG(INFO, "virtio: using simple Rx path on port %u",
1537 eth_dev->data->port_id);
1538 eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1539 } else if (hw->use_inorder_rx) {
1541 "virtio: using inorder Rx path on port %u",
1542 eth_dev->data->port_id);
1543 eth_dev->rx_pkt_burst = &virtio_recv_pkts_inorder;
1544 } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1546 "virtio: using mergeable buffer Rx path on port %u",
1547 eth_dev->data->port_id);
1548 eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1550 PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
1551 eth_dev->data->port_id);
1552 eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1558 /* Only support 1:1 queue/interrupt mapping so far.
1559 * TODO: support n:1 queue/interrupt mapping when there are limited number of
1560 * interrupt vectors (<N+1).
1563 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1566 struct virtio_hw *hw = dev->data->dev_private;
1568 PMD_INIT_LOG(INFO, "queue/interrupt binding");
1569 for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1570 dev->intr_handle->intr_vec[i] = i + 1;
1571 if (VTPCI_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1572 VIRTIO_MSI_NO_VECTOR) {
1573 PMD_DRV_LOG(ERR, "failed to set queue vector");
1582 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1585 struct virtio_hw *hw = dev->data->dev_private;
1587 PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1588 for (i = 0; i < dev->data->nb_rx_queues; ++i)
1589 VTPCI_OPS(hw)->set_queue_irq(hw,
1590 hw->vqs[i * VTNET_CQ],
1591 VIRTIO_MSI_NO_VECTOR);
1595 virtio_configure_intr(struct rte_eth_dev *dev)
1597 struct virtio_hw *hw = dev->data->dev_private;
1599 if (!rte_intr_cap_multiple(dev->intr_handle)) {
1600 PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1604 if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1605 PMD_INIT_LOG(ERR, "Fail to create eventfd");
1609 if (!dev->intr_handle->intr_vec) {
1610 dev->intr_handle->intr_vec =
1611 rte_zmalloc("intr_vec",
1612 hw->max_queue_pairs * sizeof(int), 0);
1613 if (!dev->intr_handle->intr_vec) {
1614 PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1615 hw->max_queue_pairs);
1620 /* Re-register callback to update max_intr */
1621 rte_intr_callback_unregister(dev->intr_handle,
1622 virtio_interrupt_handler,
1624 rte_intr_callback_register(dev->intr_handle,
1625 virtio_interrupt_handler,
1628 /* DO NOT try to remove this! This function will enable msix, or QEMU
1629 * will encounter SIGSEGV when DRIVER_OK is sent.
1630 * And for legacy devices, this should be done before queue/vec binding
1631 * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1632 * (22) will be ignored.
1634 if (virtio_intr_enable(dev) < 0) {
1635 PMD_DRV_LOG(ERR, "interrupt enable failed");
1639 if (virtio_queues_bind_intr(dev) < 0) {
1640 PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1647 /* reset device and renegotiate features if needed */
1649 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1651 struct virtio_hw *hw = eth_dev->data->dev_private;
1652 struct virtio_net_config *config;
1653 struct virtio_net_config local_config;
1654 struct rte_pci_device *pci_dev = NULL;
1657 /* Reset the device although not necessary at startup */
1661 virtio_dev_free_mbufs(eth_dev);
1662 virtio_free_queues(hw);
1665 /* Tell the host we've noticed this device. */
1666 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1668 /* Tell the host we've known how to drive the device. */
1669 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1670 if (virtio_negotiate_features(hw, req_features) < 0)
1673 hw->weak_barriers = !vtpci_with_feature(hw, VIRTIO_F_ORDER_PLATFORM);
1675 if (!hw->virtio_user_dev)
1676 pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1678 /* If host does not support both status and MSI-X then disable LSC */
1679 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS) &&
1680 hw->use_msix != VIRTIO_MSIX_NONE)
1681 eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1683 eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1685 /* Setting up rx_header size for the device */
1686 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1687 vtpci_with_feature(hw, VIRTIO_F_VERSION_1) ||
1688 vtpci_with_feature(hw, VIRTIO_F_RING_PACKED))
1689 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1691 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1693 /* Copy the permanent MAC address to: virtio_hw */
1694 virtio_get_hwaddr(hw);
1695 rte_ether_addr_copy((struct rte_ether_addr *)hw->mac_addr,
1696 ð_dev->data->mac_addrs[0]);
1698 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1699 hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1700 hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1702 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1703 config = &local_config;
1705 vtpci_read_dev_config(hw,
1706 offsetof(struct virtio_net_config, mac),
1707 &config->mac, sizeof(config->mac));
1709 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1710 vtpci_read_dev_config(hw,
1711 offsetof(struct virtio_net_config, status),
1712 &config->status, sizeof(config->status));
1715 "VIRTIO_NET_F_STATUS is not supported");
1719 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1720 vtpci_read_dev_config(hw,
1721 offsetof(struct virtio_net_config, max_virtqueue_pairs),
1722 &config->max_virtqueue_pairs,
1723 sizeof(config->max_virtqueue_pairs));
1726 "VIRTIO_NET_F_MQ is not supported");
1727 config->max_virtqueue_pairs = 1;
1730 hw->max_queue_pairs = config->max_virtqueue_pairs;
1732 if (vtpci_with_feature(hw, VIRTIO_NET_F_MTU)) {
1733 vtpci_read_dev_config(hw,
1734 offsetof(struct virtio_net_config, mtu),
1736 sizeof(config->mtu));
1739 * MTU value has already been checked at negotiation
1740 * time, but check again in case it has changed since
1741 * then, which should not happen.
1743 if (config->mtu < RTE_ETHER_MIN_MTU) {
1744 PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1749 hw->max_mtu = config->mtu;
1750 /* Set initial MTU to maximum one supported by vhost */
1751 eth_dev->data->mtu = config->mtu;
1754 hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1755 VLAN_TAG_LEN - hw->vtnet_hdr_size;
1758 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1759 config->max_virtqueue_pairs);
1760 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1762 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1763 config->mac[0], config->mac[1],
1764 config->mac[2], config->mac[3],
1765 config->mac[4], config->mac[5]);
1767 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1768 hw->max_queue_pairs = 1;
1769 hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1770 VLAN_TAG_LEN - hw->vtnet_hdr_size;
1773 ret = virtio_alloc_queues(eth_dev);
1777 if (eth_dev->data->dev_conf.intr_conf.rxq) {
1778 if (virtio_configure_intr(eth_dev) < 0) {
1779 PMD_INIT_LOG(ERR, "failed to configure interrupt");
1780 virtio_free_queues(hw);
1785 vtpci_reinit_complete(hw);
1788 PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1789 eth_dev->data->port_id, pci_dev->id.vendor_id,
1790 pci_dev->id.device_id);
1796 * Remap the PCI device again (IO port map for legacy device and
1797 * memory map for modern device), so that the secondary process
1798 * could have the PCI initiated correctly.
1801 virtio_remap_pci(struct rte_pci_device *pci_dev, struct virtio_hw *hw)
1805 * We don't have to re-parse the PCI config space, since
1806 * rte_pci_map_device() makes sure the mapped address
1807 * in secondary process would equal to the one mapped in
1808 * the primary process: error will be returned if that
1809 * requirement is not met.
1811 * That said, we could simply reuse all cap pointers
1812 * (such as dev_cfg, common_cfg, etc.) parsed from the
1813 * primary process, which is stored in shared memory.
1815 if (rte_pci_map_device(pci_dev)) {
1816 PMD_INIT_LOG(DEBUG, "failed to map pci device!");
1820 if (rte_pci_ioport_map(pci_dev, 0, VTPCI_IO(hw)) < 0)
1828 virtio_set_vtpci_ops(struct virtio_hw *hw)
1830 #ifdef RTE_VIRTIO_USER
1831 if (hw->virtio_user_dev)
1832 VTPCI_OPS(hw) = &virtio_user_ops;
1836 VTPCI_OPS(hw) = &modern_ops;
1838 VTPCI_OPS(hw) = &legacy_ops;
1842 * This function is based on probe() function in virtio_pci.c
1843 * It returns 0 on success.
1846 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1848 struct virtio_hw *hw = eth_dev->data->dev_private;
1851 RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr_mrg_rxbuf));
1853 eth_dev->dev_ops = &virtio_eth_dev_ops;
1855 if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1856 if (!hw->virtio_user_dev) {
1857 ret = virtio_remap_pci(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1862 virtio_set_vtpci_ops(hw);
1863 set_rxtx_funcs(eth_dev);
1869 * Pass the information to the rte_eth_dev_close() that it should also
1870 * release the private port resources.
1872 eth_dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
1874 /* Allocate memory for storing MAC addresses */
1875 eth_dev->data->mac_addrs = rte_zmalloc("virtio",
1876 VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN, 0);
1877 if (eth_dev->data->mac_addrs == NULL) {
1879 "Failed to allocate %d bytes needed to store MAC addresses",
1880 VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN);
1884 hw->port_id = eth_dev->data->port_id;
1885 /* For virtio_user case the hw->virtio_user_dev is populated by
1886 * virtio_user_eth_dev_alloc() before eth_virtio_dev_init() is called.
1888 if (!hw->virtio_user_dev) {
1889 ret = vtpci_init(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1891 goto err_vtpci_init;
1894 /* reset device and negotiate default features */
1895 ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1897 goto err_virtio_init;
1904 if (!hw->virtio_user_dev) {
1905 rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(eth_dev));
1907 rte_pci_ioport_unmap(VTPCI_IO(hw));
1910 rte_free(eth_dev->data->mac_addrs);
1911 eth_dev->data->mac_addrs = NULL;
1916 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1918 PMD_INIT_FUNC_TRACE();
1920 if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1923 virtio_dev_stop(eth_dev);
1924 virtio_dev_close(eth_dev);
1926 eth_dev->dev_ops = NULL;
1927 eth_dev->tx_pkt_burst = NULL;
1928 eth_dev->rx_pkt_burst = NULL;
1930 PMD_INIT_LOG(DEBUG, "dev_uninit completed");
1935 static int vdpa_check_handler(__rte_unused const char *key,
1936 const char *value, __rte_unused void *opaque)
1938 if (strcmp(value, "1"))
1945 vdpa_mode_selected(struct rte_devargs *devargs)
1947 struct rte_kvargs *kvlist;
1948 const char *key = "vdpa";
1951 if (devargs == NULL)
1954 kvlist = rte_kvargs_parse(devargs->args, NULL);
1958 if (!rte_kvargs_count(kvlist, key))
1961 /* vdpa mode selected when there's a key-value pair: vdpa=1 */
1962 if (rte_kvargs_process(kvlist, key,
1963 vdpa_check_handler, NULL) < 0) {
1969 rte_kvargs_free(kvlist);
1973 static int eth_virtio_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1974 struct rte_pci_device *pci_dev)
1976 if (rte_eal_iopl_init() != 0) {
1977 PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1981 /* virtio pmd skips probe if device needs to work in vdpa mode */
1982 if (vdpa_mode_selected(pci_dev->device.devargs))
1985 return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct virtio_hw),
1986 eth_virtio_dev_init);
1989 static int eth_virtio_pci_remove(struct rte_pci_device *pci_dev)
1993 ret = rte_eth_dev_pci_generic_remove(pci_dev, eth_virtio_dev_uninit);
1994 /* Port has already been released by close. */
2000 static struct rte_pci_driver rte_virtio_pmd = {
2002 .name = "net_virtio",
2004 .id_table = pci_id_virtio_map,
2006 .probe = eth_virtio_pci_probe,
2007 .remove = eth_virtio_pci_remove,
2010 RTE_INIT(rte_virtio_pmd_init)
2012 rte_eal_iopl_init();
2013 rte_pci_register(&rte_virtio_pmd);
2017 rx_offload_enabled(struct virtio_hw *hw)
2019 return vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM) ||
2020 vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2021 vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6);
2025 tx_offload_enabled(struct virtio_hw *hw)
2027 return vtpci_with_feature(hw, VIRTIO_NET_F_CSUM) ||
2028 vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO4) ||
2029 vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO6);
2033 * Configure virtio device
2034 * It returns 0 on success.
2037 virtio_dev_configure(struct rte_eth_dev *dev)
2039 const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2040 const struct rte_eth_txmode *txmode = &dev->data->dev_conf.txmode;
2041 struct virtio_hw *hw = dev->data->dev_private;
2042 uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
2044 uint64_t rx_offloads = rxmode->offloads;
2045 uint64_t tx_offloads = txmode->offloads;
2046 uint64_t req_features;
2049 PMD_INIT_LOG(DEBUG, "configure");
2050 req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
2052 if (dev->data->dev_conf.intr_conf.rxq) {
2053 ret = virtio_init_device(dev, hw->req_guest_features);
2058 if (rxmode->max_rx_pkt_len > hw->max_mtu + ether_hdr_len)
2059 req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
2061 if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2062 DEV_RX_OFFLOAD_TCP_CKSUM))
2063 req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
2065 if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO)
2067 (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2068 (1ULL << VIRTIO_NET_F_GUEST_TSO6);
2070 if (tx_offloads & (DEV_TX_OFFLOAD_UDP_CKSUM |
2071 DEV_TX_OFFLOAD_TCP_CKSUM))
2072 req_features |= (1ULL << VIRTIO_NET_F_CSUM);
2074 if (tx_offloads & DEV_TX_OFFLOAD_TCP_TSO)
2076 (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2077 (1ULL << VIRTIO_NET_F_HOST_TSO6);
2079 /* if request features changed, reinit the device */
2080 if (req_features != hw->req_guest_features) {
2081 ret = virtio_init_device(dev, req_features);
2086 if ((rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2087 DEV_RX_OFFLOAD_TCP_CKSUM)) &&
2088 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
2090 "rx checksum not available on this host");
2094 if ((rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) &&
2095 (!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2096 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6))) {
2098 "Large Receive Offload not available on this host");
2102 /* start control queue */
2103 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
2104 virtio_dev_cq_start(dev);
2106 if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
2109 if ((rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
2110 && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2112 "vlan filtering not available on this host");
2116 hw->has_tx_offload = tx_offload_enabled(hw);
2117 hw->has_rx_offload = rx_offload_enabled(hw);
2119 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2120 /* Enable vector (0) for Link State Intrerrupt */
2121 if (VTPCI_OPS(hw)->set_config_irq(hw, 0) ==
2122 VIRTIO_MSI_NO_VECTOR) {
2123 PMD_DRV_LOG(ERR, "failed to set config vector");
2127 rte_spinlock_init(&hw->state_lock);
2129 hw->use_simple_rx = 1;
2131 if (vtpci_with_feature(hw, VIRTIO_F_IN_ORDER)) {
2132 hw->use_inorder_tx = 1;
2133 hw->use_inorder_rx = 1;
2134 hw->use_simple_rx = 0;
2137 if (vtpci_packed_queue(hw)) {
2138 hw->use_simple_rx = 0;
2139 hw->use_inorder_rx = 0;
2142 #if defined RTE_ARCH_ARM64 || defined RTE_ARCH_ARM
2143 if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
2144 hw->use_simple_rx = 0;
2147 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2148 hw->use_simple_rx = 0;
2151 if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2152 DEV_RX_OFFLOAD_TCP_CKSUM |
2153 DEV_RX_OFFLOAD_TCP_LRO |
2154 DEV_RX_OFFLOAD_VLAN_STRIP))
2155 hw->use_simple_rx = 0;
2162 virtio_dev_start(struct rte_eth_dev *dev)
2164 uint16_t nb_queues, i;
2165 struct virtnet_rx *rxvq;
2166 struct virtnet_tx *txvq __rte_unused;
2167 struct virtio_hw *hw = dev->data->dev_private;
2170 /* Finish the initialization of the queues */
2171 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2172 ret = virtio_dev_rx_queue_setup_finish(dev, i);
2176 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2177 ret = virtio_dev_tx_queue_setup_finish(dev, i);
2182 /* check if lsc interrupt feature is enabled */
2183 if (dev->data->dev_conf.intr_conf.lsc) {
2184 if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
2185 PMD_DRV_LOG(ERR, "link status not supported by host");
2190 /* Enable uio/vfio intr/eventfd mapping: althrough we already did that
2191 * in device configure, but it could be unmapped when device is
2194 if (dev->data->dev_conf.intr_conf.lsc ||
2195 dev->data->dev_conf.intr_conf.rxq) {
2196 virtio_intr_disable(dev);
2198 /* Setup interrupt callback */
2199 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2200 rte_intr_callback_register(dev->intr_handle,
2201 virtio_interrupt_handler,
2204 if (virtio_intr_enable(dev) < 0) {
2205 PMD_DRV_LOG(ERR, "interrupt enable failed");
2210 /*Notify the backend
2211 *Otherwise the tap backend might already stop its queue due to fullness.
2212 *vhost backend will have no chance to be waked up
2214 nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
2215 if (hw->max_queue_pairs > 1) {
2216 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
2220 PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
2222 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2223 rxvq = dev->data->rx_queues[i];
2224 /* Flush the old packets */
2225 virtqueue_rxvq_flush(rxvq->vq);
2226 virtqueue_notify(rxvq->vq);
2229 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2230 txvq = dev->data->tx_queues[i];
2231 virtqueue_notify(txvq->vq);
2234 PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
2236 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2237 rxvq = dev->data->rx_queues[i];
2238 VIRTQUEUE_DUMP(rxvq->vq);
2241 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2242 txvq = dev->data->tx_queues[i];
2243 VIRTQUEUE_DUMP(txvq->vq);
2246 set_rxtx_funcs(dev);
2249 /* Initialize Link state */
2250 virtio_dev_link_update(dev, 0);
2255 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
2257 struct virtio_hw *hw = dev->data->dev_private;
2258 uint16_t nr_vq = virtio_get_nr_vq(hw);
2259 const char *type __rte_unused;
2260 unsigned int i, mbuf_num = 0;
2261 struct virtqueue *vq;
2262 struct rte_mbuf *buf;
2265 if (hw->vqs == NULL)
2268 for (i = 0; i < nr_vq; i++) {
2273 queue_type = virtio_get_queue_type(hw, i);
2274 if (queue_type == VTNET_RQ)
2276 else if (queue_type == VTNET_TQ)
2282 "Before freeing %s[%d] used and unused buf",
2286 while ((buf = virtqueue_detach_unused(vq)) != NULL) {
2287 rte_pktmbuf_free(buf);
2292 "After freeing %s[%d] used and unused buf",
2297 PMD_INIT_LOG(DEBUG, "%d mbufs freed", mbuf_num);
2301 * Stop device: disable interrupt and mark link down
2304 virtio_dev_stop(struct rte_eth_dev *dev)
2306 struct virtio_hw *hw = dev->data->dev_private;
2307 struct rte_eth_link link;
2308 struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
2310 PMD_INIT_LOG(DEBUG, "stop");
2312 rte_spinlock_lock(&hw->state_lock);
2315 hw->started = false;
2317 if (intr_conf->lsc || intr_conf->rxq) {
2318 virtio_intr_disable(dev);
2320 /* Reset interrupt callback */
2321 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
2322 rte_intr_callback_unregister(dev->intr_handle,
2323 virtio_interrupt_handler,
2328 memset(&link, 0, sizeof(link));
2329 rte_eth_linkstatus_set(dev, &link);
2331 rte_spinlock_unlock(&hw->state_lock);
2335 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
2337 struct rte_eth_link link;
2339 struct virtio_hw *hw = dev->data->dev_private;
2341 memset(&link, 0, sizeof(link));
2342 link.link_duplex = ETH_LINK_FULL_DUPLEX;
2343 link.link_speed = ETH_SPEED_NUM_10G;
2344 link.link_autoneg = ETH_LINK_FIXED;
2347 link.link_status = ETH_LINK_DOWN;
2348 } else if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
2349 PMD_INIT_LOG(DEBUG, "Get link status from hw");
2350 vtpci_read_dev_config(hw,
2351 offsetof(struct virtio_net_config, status),
2352 &status, sizeof(status));
2353 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
2354 link.link_status = ETH_LINK_DOWN;
2355 PMD_INIT_LOG(DEBUG, "Port %d is down",
2356 dev->data->port_id);
2358 link.link_status = ETH_LINK_UP;
2359 PMD_INIT_LOG(DEBUG, "Port %d is up",
2360 dev->data->port_id);
2363 link.link_status = ETH_LINK_UP;
2366 return rte_eth_linkstatus_set(dev, &link);
2370 virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
2372 const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2373 struct virtio_hw *hw = dev->data->dev_private;
2374 uint64_t offloads = rxmode->offloads;
2376 if (mask & ETH_VLAN_FILTER_MASK) {
2377 if ((offloads & DEV_RX_OFFLOAD_VLAN_FILTER) &&
2378 !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2381 "vlan filtering not available on this host");
2387 if (mask & ETH_VLAN_STRIP_MASK)
2388 hw->vlan_strip = !!(offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
2394 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2396 uint64_t tso_mask, host_features;
2397 struct virtio_hw *hw = dev->data->dev_private;
2399 dev_info->speed_capa = ETH_LINK_SPEED_10G; /* fake value */
2401 dev_info->max_rx_queues =
2402 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
2403 dev_info->max_tx_queues =
2404 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
2405 dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
2406 dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
2407 dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
2409 host_features = VTPCI_OPS(hw)->get_features(hw);
2410 dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
2411 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_JUMBO_FRAME;
2412 if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
2413 dev_info->rx_offload_capa |=
2414 DEV_RX_OFFLOAD_TCP_CKSUM |
2415 DEV_RX_OFFLOAD_UDP_CKSUM;
2417 if (host_features & (1ULL << VIRTIO_NET_F_CTRL_VLAN))
2418 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_VLAN_FILTER;
2419 tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2420 (1ULL << VIRTIO_NET_F_GUEST_TSO6);
2421 if ((host_features & tso_mask) == tso_mask)
2422 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
2424 dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
2425 DEV_TX_OFFLOAD_VLAN_INSERT;
2426 if (host_features & (1ULL << VIRTIO_NET_F_CSUM)) {
2427 dev_info->tx_offload_capa |=
2428 DEV_TX_OFFLOAD_UDP_CKSUM |
2429 DEV_TX_OFFLOAD_TCP_CKSUM;
2431 tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2432 (1ULL << VIRTIO_NET_F_HOST_TSO6);
2433 if ((host_features & tso_mask) == tso_mask)
2434 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
2438 * It enables testpmd to collect per queue stats.
2441 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
2442 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
2443 __rte_unused uint8_t is_rx)
2448 RTE_PMD_EXPORT_NAME(net_virtio, __COUNTER__);
2449 RTE_PMD_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map);
2450 RTE_PMD_REGISTER_KMOD_DEP(net_virtio, "* igb_uio | uio_pci_generic | vfio-pci");
2452 RTE_INIT(virtio_init_log)
2454 virtio_logtype_init = rte_log_register("pmd.net.virtio.init");
2455 if (virtio_logtype_init >= 0)
2456 rte_log_set_level(virtio_logtype_init, RTE_LOG_NOTICE);
2457 virtio_logtype_driver = rte_log_register("pmd.net.virtio.driver");
2458 if (virtio_logtype_driver >= 0)
2459 rte_log_set_level(virtio_logtype_driver, RTE_LOG_NOTICE);