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 int 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 int 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 {"multicast_packets", offsetof(struct virtnet_tx, stats.multicast)},
125 {"broadcast_packets", offsetof(struct virtnet_tx, stats.broadcast)},
126 {"undersize_packets", offsetof(struct virtnet_tx, stats.size_bins[0])},
127 {"size_64_packets", offsetof(struct virtnet_tx, stats.size_bins[1])},
128 {"size_65_127_packets", offsetof(struct virtnet_tx, stats.size_bins[2])},
129 {"size_128_255_packets", offsetof(struct virtnet_tx, stats.size_bins[3])},
130 {"size_256_511_packets", offsetof(struct virtnet_tx, stats.size_bins[4])},
131 {"size_512_1023_packets", offsetof(struct virtnet_tx, stats.size_bins[5])},
132 {"size_1024_1518_packets", offsetof(struct virtnet_tx, stats.size_bins[6])},
133 {"size_1519_max_packets", offsetof(struct virtnet_tx, stats.size_bins[7])},
136 #define VIRTIO_NB_RXQ_XSTATS (sizeof(rte_virtio_rxq_stat_strings) / \
137 sizeof(rte_virtio_rxq_stat_strings[0]))
138 #define VIRTIO_NB_TXQ_XSTATS (sizeof(rte_virtio_txq_stat_strings) / \
139 sizeof(rte_virtio_txq_stat_strings[0]))
141 struct virtio_hw_internal virtio_hw_internal[RTE_MAX_ETHPORTS];
143 static struct virtio_pmd_ctrl *
144 virtio_send_command_packed(struct virtnet_ctl *cvq,
145 struct virtio_pmd_ctrl *ctrl,
146 int *dlen, int pkt_num)
148 struct virtqueue *vq = cvq->vq;
150 struct vring_packed_desc *desc = vq->vq_packed.ring.desc;
151 struct virtio_pmd_ctrl *result;
158 * Format is enforced in qemu code:
159 * One TX packet for header;
160 * At least one TX packet per argument;
161 * One RX packet for ACK.
163 head = vq->vq_avail_idx;
164 flags = vq->vq_packed.cached_flags;
165 desc[head].addr = cvq->virtio_net_hdr_mem;
166 desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
169 if (++vq->vq_avail_idx >= vq->vq_nentries) {
170 vq->vq_avail_idx -= vq->vq_nentries;
171 vq->vq_packed.cached_flags ^= VRING_PACKED_DESC_F_AVAIL_USED;
174 for (k = 0; k < pkt_num; k++) {
175 desc[vq->vq_avail_idx].addr = cvq->virtio_net_hdr_mem
176 + sizeof(struct virtio_net_ctrl_hdr)
177 + sizeof(ctrl->status) + sizeof(uint8_t) * sum;
178 desc[vq->vq_avail_idx].len = dlen[k];
179 desc[vq->vq_avail_idx].flags = VRING_DESC_F_NEXT |
180 vq->vq_packed.cached_flags;
184 if (++vq->vq_avail_idx >= vq->vq_nentries) {
185 vq->vq_avail_idx -= vq->vq_nentries;
186 vq->vq_packed.cached_flags ^=
187 VRING_PACKED_DESC_F_AVAIL_USED;
191 desc[vq->vq_avail_idx].addr = cvq->virtio_net_hdr_mem
192 + sizeof(struct virtio_net_ctrl_hdr);
193 desc[vq->vq_avail_idx].len = sizeof(ctrl->status);
194 desc[vq->vq_avail_idx].flags = VRING_DESC_F_WRITE |
195 vq->vq_packed.cached_flags;
198 if (++vq->vq_avail_idx >= vq->vq_nentries) {
199 vq->vq_avail_idx -= vq->vq_nentries;
200 vq->vq_packed.cached_flags ^= VRING_PACKED_DESC_F_AVAIL_USED;
203 virtio_wmb(vq->hw->weak_barriers);
204 desc[head].flags = VRING_DESC_F_NEXT | flags;
206 virtio_wmb(vq->hw->weak_barriers);
207 virtqueue_notify(vq);
209 /* wait for used descriptors in virtqueue */
210 while (!desc_is_used(&desc[head], vq))
213 virtio_rmb(vq->hw->weak_barriers);
215 /* now get used descriptors */
216 vq->vq_free_cnt += nb_descs;
217 vq->vq_used_cons_idx += nb_descs;
218 if (vq->vq_used_cons_idx >= vq->vq_nentries) {
219 vq->vq_used_cons_idx -= vq->vq_nentries;
220 vq->vq_packed.used_wrap_counter ^= 1;
223 PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\n"
224 "vq->vq_avail_idx=%d\n"
225 "vq->vq_used_cons_idx=%d\n"
226 "vq->vq_packed.cached_flags=0x%x\n"
227 "vq->vq_packed.used_wrap_counter=%d\n",
230 vq->vq_used_cons_idx,
231 vq->vq_packed.cached_flags,
232 vq->vq_packed.used_wrap_counter);
234 result = cvq->virtio_net_hdr_mz->addr;
238 static struct virtio_pmd_ctrl *
239 virtio_send_command_split(struct virtnet_ctl *cvq,
240 struct virtio_pmd_ctrl *ctrl,
241 int *dlen, int pkt_num)
243 struct virtio_pmd_ctrl *result;
244 struct virtqueue *vq = cvq->vq;
248 head = vq->vq_desc_head_idx;
251 * Format is enforced in qemu code:
252 * One TX packet for header;
253 * At least one TX packet per argument;
254 * One RX packet for ACK.
256 vq->vq_split.ring.desc[head].flags = VRING_DESC_F_NEXT;
257 vq->vq_split.ring.desc[head].addr = cvq->virtio_net_hdr_mem;
258 vq->vq_split.ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
260 i = vq->vq_split.ring.desc[head].next;
262 for (k = 0; k < pkt_num; k++) {
263 vq->vq_split.ring.desc[i].flags = VRING_DESC_F_NEXT;
264 vq->vq_split.ring.desc[i].addr = cvq->virtio_net_hdr_mem
265 + sizeof(struct virtio_net_ctrl_hdr)
266 + sizeof(ctrl->status) + sizeof(uint8_t)*sum;
267 vq->vq_split.ring.desc[i].len = dlen[k];
270 i = vq->vq_split.ring.desc[i].next;
273 vq->vq_split.ring.desc[i].flags = VRING_DESC_F_WRITE;
274 vq->vq_split.ring.desc[i].addr = cvq->virtio_net_hdr_mem
275 + sizeof(struct virtio_net_ctrl_hdr);
276 vq->vq_split.ring.desc[i].len = sizeof(ctrl->status);
279 vq->vq_desc_head_idx = vq->vq_split.ring.desc[i].next;
281 vq_update_avail_ring(vq, head);
282 vq_update_avail_idx(vq);
284 PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
286 virtqueue_notify(vq);
289 while (VIRTQUEUE_NUSED(vq) == 0) {
294 while (VIRTQUEUE_NUSED(vq)) {
295 uint32_t idx, desc_idx, used_idx;
296 struct vring_used_elem *uep;
298 used_idx = (uint32_t)(vq->vq_used_cons_idx
299 & (vq->vq_nentries - 1));
300 uep = &vq->vq_split.ring.used->ring[used_idx];
301 idx = (uint32_t) uep->id;
304 while (vq->vq_split.ring.desc[desc_idx].flags &
306 desc_idx = vq->vq_split.ring.desc[desc_idx].next;
310 vq->vq_split.ring.desc[desc_idx].next = vq->vq_desc_head_idx;
311 vq->vq_desc_head_idx = idx;
313 vq->vq_used_cons_idx++;
317 PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
318 vq->vq_free_cnt, vq->vq_desc_head_idx);
320 result = cvq->virtio_net_hdr_mz->addr;
325 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
326 int *dlen, int pkt_num)
328 virtio_net_ctrl_ack status = ~0;
329 struct virtio_pmd_ctrl *result;
330 struct virtqueue *vq;
332 ctrl->status = status;
334 if (!cvq || !cvq->vq) {
335 PMD_INIT_LOG(ERR, "Control queue is not supported.");
339 rte_spinlock_lock(&cvq->lock);
342 PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
343 "vq->hw->cvq = %p vq = %p",
344 vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
346 if (vq->vq_free_cnt < pkt_num + 2 || pkt_num < 1) {
347 rte_spinlock_unlock(&cvq->lock);
351 memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
352 sizeof(struct virtio_pmd_ctrl));
354 if (vtpci_packed_queue(vq->hw))
355 result = virtio_send_command_packed(cvq, ctrl, dlen, pkt_num);
357 result = virtio_send_command_split(cvq, ctrl, dlen, pkt_num);
359 rte_spinlock_unlock(&cvq->lock);
360 return result->status;
364 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
366 struct virtio_hw *hw = dev->data->dev_private;
367 struct virtio_pmd_ctrl ctrl;
371 ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
372 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
373 memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
375 dlen[0] = sizeof(uint16_t);
377 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
379 PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
380 "failed, this is too late now...");
388 virtio_dev_queue_release(void *queue __rte_unused)
394 virtio_get_nr_vq(struct virtio_hw *hw)
396 uint16_t nr_vq = hw->max_queue_pairs * 2;
398 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
405 virtio_init_vring(struct virtqueue *vq)
407 int size = vq->vq_nentries;
408 uint8_t *ring_mem = vq->vq_ring_virt_mem;
410 PMD_INIT_FUNC_TRACE();
412 memset(ring_mem, 0, vq->vq_ring_size);
414 vq->vq_used_cons_idx = 0;
415 vq->vq_desc_head_idx = 0;
416 vq->vq_avail_idx = 0;
417 vq->vq_desc_tail_idx = (uint16_t)(vq->vq_nentries - 1);
418 vq->vq_free_cnt = vq->vq_nentries;
419 memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries);
420 if (vtpci_packed_queue(vq->hw)) {
421 vring_init_packed(&vq->vq_packed.ring, ring_mem,
422 VIRTIO_PCI_VRING_ALIGN, size);
423 vring_desc_init_packed(vq, size);
425 struct vring *vr = &vq->vq_split.ring;
427 vring_init_split(vr, ring_mem, VIRTIO_PCI_VRING_ALIGN, size);
428 vring_desc_init_split(vr->desc, size);
431 * Disable device(host) interrupting guest
433 virtqueue_disable_intr(vq);
437 virtio_init_queue(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx)
439 char vq_name[VIRTQUEUE_MAX_NAME_SZ];
440 char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
441 const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
442 unsigned int vq_size, size;
443 struct virtio_hw *hw = dev->data->dev_private;
444 struct virtnet_rx *rxvq = NULL;
445 struct virtnet_tx *txvq = NULL;
446 struct virtnet_ctl *cvq = NULL;
447 struct virtqueue *vq;
448 size_t sz_hdr_mz = 0;
449 void *sw_ring = NULL;
450 int queue_type = virtio_get_queue_type(hw, vtpci_queue_idx);
452 int numa_node = dev->device->numa_node;
454 PMD_INIT_LOG(INFO, "setting up queue: %u on NUMA node %d",
455 vtpci_queue_idx, numa_node);
458 * Read the virtqueue size from the Queue Size field
459 * Always power of 2 and if 0 virtqueue does not exist
461 vq_size = VTPCI_OPS(hw)->get_queue_num(hw, vtpci_queue_idx);
462 PMD_INIT_LOG(DEBUG, "vq_size: %u", vq_size);
464 PMD_INIT_LOG(ERR, "virtqueue does not exist");
468 if (!rte_is_power_of_2(vq_size)) {
469 PMD_INIT_LOG(ERR, "virtqueue size is not powerof 2");
473 snprintf(vq_name, sizeof(vq_name), "port%d_vq%d",
474 dev->data->port_id, vtpci_queue_idx);
476 size = RTE_ALIGN_CEIL(sizeof(*vq) +
477 vq_size * sizeof(struct vq_desc_extra),
478 RTE_CACHE_LINE_SIZE);
479 if (queue_type == VTNET_TQ) {
481 * For each xmit packet, allocate a virtio_net_hdr
482 * and indirect ring elements
484 sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
485 } else if (queue_type == VTNET_CQ) {
486 /* Allocate a page for control vq command, data and status */
487 sz_hdr_mz = PAGE_SIZE;
490 vq = rte_zmalloc_socket(vq_name, size, RTE_CACHE_LINE_SIZE,
493 PMD_INIT_LOG(ERR, "can not allocate vq");
496 hw->vqs[vtpci_queue_idx] = vq;
499 vq->vq_queue_index = vtpci_queue_idx;
500 vq->vq_nentries = vq_size;
501 if (vtpci_packed_queue(hw)) {
502 vq->vq_packed.used_wrap_counter = 1;
503 vq->vq_packed.cached_flags = VRING_PACKED_DESC_F_AVAIL;
504 vq->vq_packed.event_flags_shadow = 0;
505 if (queue_type == VTNET_RQ)
506 vq->vq_packed.cached_flags |= VRING_DESC_F_WRITE;
510 * Reserve a memzone for vring elements
512 size = vring_size(hw, vq_size, VIRTIO_PCI_VRING_ALIGN);
513 vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
514 PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
515 size, vq->vq_ring_size);
517 mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
518 numa_node, RTE_MEMZONE_IOVA_CONTIG,
519 VIRTIO_PCI_VRING_ALIGN);
521 if (rte_errno == EEXIST)
522 mz = rte_memzone_lookup(vq_name);
529 memset(mz->addr, 0, mz->len);
531 vq->vq_ring_mem = mz->iova;
532 vq->vq_ring_virt_mem = mz->addr;
533 PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem: 0x%" PRIx64,
535 PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64,
536 (uint64_t)(uintptr_t)mz->addr);
538 virtio_init_vring(vq);
541 snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_vq%d_hdr",
542 dev->data->port_id, vtpci_queue_idx);
543 hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
544 numa_node, RTE_MEMZONE_IOVA_CONTIG,
545 RTE_CACHE_LINE_SIZE);
546 if (hdr_mz == NULL) {
547 if (rte_errno == EEXIST)
548 hdr_mz = rte_memzone_lookup(vq_hdr_name);
549 if (hdr_mz == NULL) {
556 if (queue_type == VTNET_RQ) {
557 size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
558 sizeof(vq->sw_ring[0]);
560 sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
561 RTE_CACHE_LINE_SIZE, numa_node);
563 PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
568 vq->sw_ring = sw_ring;
571 rxvq->port_id = dev->data->port_id;
573 } else if (queue_type == VTNET_TQ) {
576 txvq->port_id = dev->data->port_id;
578 txvq->virtio_net_hdr_mz = hdr_mz;
579 txvq->virtio_net_hdr_mem = hdr_mz->iova;
580 } else if (queue_type == VTNET_CQ) {
584 cvq->virtio_net_hdr_mz = hdr_mz;
585 cvq->virtio_net_hdr_mem = hdr_mz->iova;
586 memset(cvq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
591 /* For virtio_user case (that is when hw->dev is NULL), we use
592 * virtual address. And we need properly set _offset_, please see
593 * VIRTIO_MBUF_DATA_DMA_ADDR in virtqueue.h for more information.
595 if (!hw->virtio_user_dev)
596 vq->offset = offsetof(struct rte_mbuf, buf_iova);
598 vq->vq_ring_mem = (uintptr_t)mz->addr;
599 vq->offset = offsetof(struct rte_mbuf, buf_addr);
600 if (queue_type == VTNET_TQ)
601 txvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
602 else if (queue_type == VTNET_CQ)
603 cvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
606 if (queue_type == VTNET_TQ) {
607 struct virtio_tx_region *txr;
611 memset(txr, 0, vq_size * sizeof(*txr));
612 for (i = 0; i < vq_size; i++) {
613 struct vring_desc *start_dp = txr[i].tx_indir;
615 /* first indirect descriptor is always the tx header */
616 if (!vtpci_packed_queue(hw)) {
617 vring_desc_init_split(start_dp,
618 RTE_DIM(txr[i].tx_indir));
619 start_dp->addr = txvq->virtio_net_hdr_mem
621 + offsetof(struct virtio_tx_region,
623 start_dp->len = hw->vtnet_hdr_size;
624 start_dp->flags = VRING_DESC_F_NEXT;
629 if (VTPCI_OPS(hw)->setup_queue(hw, vq) < 0) {
630 PMD_INIT_LOG(ERR, "setup_queue failed");
638 rte_memzone_free(hdr_mz);
639 rte_memzone_free(mz);
646 virtio_free_queues(struct virtio_hw *hw)
648 uint16_t nr_vq = virtio_get_nr_vq(hw);
649 struct virtqueue *vq;
656 for (i = 0; i < nr_vq; i++) {
661 queue_type = virtio_get_queue_type(hw, i);
662 if (queue_type == VTNET_RQ) {
663 rte_free(vq->sw_ring);
664 rte_memzone_free(vq->rxq.mz);
665 } else if (queue_type == VTNET_TQ) {
666 rte_memzone_free(vq->txq.mz);
667 rte_memzone_free(vq->txq.virtio_net_hdr_mz);
669 rte_memzone_free(vq->cq.mz);
670 rte_memzone_free(vq->cq.virtio_net_hdr_mz);
682 virtio_alloc_queues(struct rte_eth_dev *dev)
684 struct virtio_hw *hw = dev->data->dev_private;
685 uint16_t nr_vq = virtio_get_nr_vq(hw);
689 hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
691 PMD_INIT_LOG(ERR, "failed to allocate vqs");
695 for (i = 0; i < nr_vq; i++) {
696 ret = virtio_init_queue(dev, i);
698 virtio_free_queues(hw);
706 static void virtio_queues_unbind_intr(struct rte_eth_dev *dev);
709 virtio_dev_close(struct rte_eth_dev *dev)
711 struct virtio_hw *hw = dev->data->dev_private;
712 struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
714 PMD_INIT_LOG(DEBUG, "virtio_dev_close");
721 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
722 VTPCI_OPS(hw)->set_config_irq(hw, VIRTIO_MSI_NO_VECTOR);
724 virtio_queues_unbind_intr(dev);
726 if (intr_conf->lsc || intr_conf->rxq) {
727 virtio_intr_disable(dev);
728 rte_intr_efd_disable(dev->intr_handle);
729 rte_free(dev->intr_handle->intr_vec);
730 dev->intr_handle->intr_vec = NULL;
734 virtio_dev_free_mbufs(dev);
735 virtio_free_queues(hw);
737 #ifdef RTE_VIRTIO_USER
738 if (hw->virtio_user_dev)
739 virtio_user_dev_uninit(hw->virtio_user_dev);
743 rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(dev));
745 rte_pci_ioport_unmap(VTPCI_IO(hw));
750 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
752 struct virtio_hw *hw = dev->data->dev_private;
753 struct virtio_pmd_ctrl ctrl;
757 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
758 PMD_INIT_LOG(INFO, "host does not support rx control");
762 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
763 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
767 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
769 PMD_INIT_LOG(ERR, "Failed to enable promisc");
777 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
779 struct virtio_hw *hw = dev->data->dev_private;
780 struct virtio_pmd_ctrl ctrl;
784 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
785 PMD_INIT_LOG(INFO, "host does not support rx control");
789 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
790 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
794 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
796 PMD_INIT_LOG(ERR, "Failed to disable promisc");
804 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
806 struct virtio_hw *hw = dev->data->dev_private;
807 struct virtio_pmd_ctrl ctrl;
811 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
812 PMD_INIT_LOG(INFO, "host does not support rx control");
816 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
817 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
821 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
823 PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
831 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
833 struct virtio_hw *hw = dev->data->dev_private;
834 struct virtio_pmd_ctrl ctrl;
838 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
839 PMD_INIT_LOG(INFO, "host does not support rx control");
843 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
844 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
848 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
850 PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
857 #define VLAN_TAG_LEN 4 /* 802.3ac tag (not DMA'd) */
859 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
861 struct virtio_hw *hw = dev->data->dev_private;
862 uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
864 uint32_t frame_size = mtu + ether_hdr_len;
865 uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
867 max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
869 if (mtu < RTE_ETHER_MIN_MTU || frame_size > max_frame_size) {
870 PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
871 RTE_ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
878 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
880 struct virtio_hw *hw = dev->data->dev_private;
881 struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
882 struct virtqueue *vq = rxvq->vq;
884 virtqueue_enable_intr(vq);
885 virtio_mb(hw->weak_barriers);
890 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
892 struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
893 struct virtqueue *vq = rxvq->vq;
895 virtqueue_disable_intr(vq);
900 * dev_ops for virtio, bare necessities for basic operation
902 static const struct eth_dev_ops virtio_eth_dev_ops = {
903 .dev_configure = virtio_dev_configure,
904 .dev_start = virtio_dev_start,
905 .dev_stop = virtio_dev_stop,
906 .dev_close = virtio_dev_close,
907 .promiscuous_enable = virtio_dev_promiscuous_enable,
908 .promiscuous_disable = virtio_dev_promiscuous_disable,
909 .allmulticast_enable = virtio_dev_allmulticast_enable,
910 .allmulticast_disable = virtio_dev_allmulticast_disable,
911 .mtu_set = virtio_mtu_set,
912 .dev_infos_get = virtio_dev_info_get,
913 .stats_get = virtio_dev_stats_get,
914 .xstats_get = virtio_dev_xstats_get,
915 .xstats_get_names = virtio_dev_xstats_get_names,
916 .stats_reset = virtio_dev_stats_reset,
917 .xstats_reset = virtio_dev_stats_reset,
918 .link_update = virtio_dev_link_update,
919 .vlan_offload_set = virtio_dev_vlan_offload_set,
920 .rx_queue_setup = virtio_dev_rx_queue_setup,
921 .rx_queue_intr_enable = virtio_dev_rx_queue_intr_enable,
922 .rx_queue_intr_disable = virtio_dev_rx_queue_intr_disable,
923 .rx_queue_release = virtio_dev_queue_release,
924 .rx_descriptor_done = virtio_dev_rx_queue_done,
925 .tx_queue_setup = virtio_dev_tx_queue_setup,
926 .tx_queue_release = virtio_dev_queue_release,
927 /* collect stats per queue */
928 .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
929 .vlan_filter_set = virtio_vlan_filter_set,
930 .mac_addr_add = virtio_mac_addr_add,
931 .mac_addr_remove = virtio_mac_addr_remove,
932 .mac_addr_set = virtio_mac_addr_set,
936 * dev_ops for virtio-user in secondary processes, as we just have
937 * some limited supports currently.
939 const struct eth_dev_ops virtio_user_secondary_eth_dev_ops = {
940 .dev_infos_get = virtio_dev_info_get,
941 .stats_get = virtio_dev_stats_get,
942 .xstats_get = virtio_dev_xstats_get,
943 .xstats_get_names = virtio_dev_xstats_get_names,
944 .stats_reset = virtio_dev_stats_reset,
945 .xstats_reset = virtio_dev_stats_reset,
946 /* collect stats per queue */
947 .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
951 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
955 for (i = 0; i < dev->data->nb_tx_queues; i++) {
956 const struct virtnet_tx *txvq = dev->data->tx_queues[i];
960 stats->opackets += txvq->stats.packets;
961 stats->obytes += txvq->stats.bytes;
963 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
964 stats->q_opackets[i] = txvq->stats.packets;
965 stats->q_obytes[i] = txvq->stats.bytes;
969 for (i = 0; i < dev->data->nb_rx_queues; i++) {
970 const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
974 stats->ipackets += rxvq->stats.packets;
975 stats->ibytes += rxvq->stats.bytes;
976 stats->ierrors += rxvq->stats.errors;
978 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
979 stats->q_ipackets[i] = rxvq->stats.packets;
980 stats->q_ibytes[i] = rxvq->stats.bytes;
984 stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
987 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
988 struct rte_eth_xstat_name *xstats_names,
989 __rte_unused unsigned limit)
995 unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
996 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
998 if (xstats_names != NULL) {
999 /* Note: limit checked in rte_eth_xstats_names() */
1001 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1002 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1005 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1006 snprintf(xstats_names[count].name,
1007 sizeof(xstats_names[count].name),
1009 rte_virtio_rxq_stat_strings[t].name);
1014 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1015 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1018 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1019 snprintf(xstats_names[count].name,
1020 sizeof(xstats_names[count].name),
1022 rte_virtio_txq_stat_strings[t].name);
1032 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
1038 unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
1039 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
1044 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1045 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1052 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1053 xstats[count].value = *(uint64_t *)(((char *)rxvq) +
1054 rte_virtio_rxq_stat_strings[t].offset);
1055 xstats[count].id = count;
1060 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1061 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1068 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1069 xstats[count].value = *(uint64_t *)(((char *)txvq) +
1070 rte_virtio_txq_stat_strings[t].offset);
1071 xstats[count].id = count;
1080 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1082 virtio_update_stats(dev, stats);
1088 virtio_dev_stats_reset(struct rte_eth_dev *dev)
1092 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1093 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1097 txvq->stats.packets = 0;
1098 txvq->stats.bytes = 0;
1099 txvq->stats.multicast = 0;
1100 txvq->stats.broadcast = 0;
1101 memset(txvq->stats.size_bins, 0,
1102 sizeof(txvq->stats.size_bins[0]) * 8);
1105 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1106 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1110 rxvq->stats.packets = 0;
1111 rxvq->stats.bytes = 0;
1112 rxvq->stats.errors = 0;
1113 rxvq->stats.multicast = 0;
1114 rxvq->stats.broadcast = 0;
1115 memset(rxvq->stats.size_bins, 0,
1116 sizeof(rxvq->stats.size_bins[0]) * 8);
1123 virtio_set_hwaddr(struct virtio_hw *hw)
1125 vtpci_write_dev_config(hw,
1126 offsetof(struct virtio_net_config, mac),
1127 &hw->mac_addr, RTE_ETHER_ADDR_LEN);
1131 virtio_get_hwaddr(struct virtio_hw *hw)
1133 if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
1134 vtpci_read_dev_config(hw,
1135 offsetof(struct virtio_net_config, mac),
1136 &hw->mac_addr, RTE_ETHER_ADDR_LEN);
1138 rte_eth_random_addr(&hw->mac_addr[0]);
1139 virtio_set_hwaddr(hw);
1144 virtio_mac_table_set(struct virtio_hw *hw,
1145 const struct virtio_net_ctrl_mac *uc,
1146 const struct virtio_net_ctrl_mac *mc)
1148 struct virtio_pmd_ctrl ctrl;
1151 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1152 PMD_DRV_LOG(INFO, "host does not support mac table");
1156 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1157 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1159 len[0] = uc->entries * RTE_ETHER_ADDR_LEN + sizeof(uc->entries);
1160 memcpy(ctrl.data, uc, len[0]);
1162 len[1] = mc->entries * RTE_ETHER_ADDR_LEN + sizeof(mc->entries);
1163 memcpy(ctrl.data + len[0], mc, len[1]);
1165 err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1167 PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1172 virtio_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
1173 uint32_t index, uint32_t vmdq __rte_unused)
1175 struct virtio_hw *hw = dev->data->dev_private;
1176 const struct rte_ether_addr *addrs = dev->data->mac_addrs;
1178 struct virtio_net_ctrl_mac *uc, *mc;
1180 if (index >= VIRTIO_MAX_MAC_ADDRS) {
1181 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1185 uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1186 sizeof(uc->entries));
1188 mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1189 sizeof(mc->entries));
1192 for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1193 const struct rte_ether_addr *addr
1194 = (i == index) ? mac_addr : addrs + i;
1195 struct virtio_net_ctrl_mac *tbl
1196 = rte_is_multicast_ether_addr(addr) ? mc : uc;
1198 memcpy(&tbl->macs[tbl->entries++], addr, RTE_ETHER_ADDR_LEN);
1201 return virtio_mac_table_set(hw, uc, mc);
1205 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1207 struct virtio_hw *hw = dev->data->dev_private;
1208 struct rte_ether_addr *addrs = dev->data->mac_addrs;
1209 struct virtio_net_ctrl_mac *uc, *mc;
1212 if (index >= VIRTIO_MAX_MAC_ADDRS) {
1213 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1217 uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1218 sizeof(uc->entries));
1220 mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1221 sizeof(mc->entries));
1224 for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1225 struct virtio_net_ctrl_mac *tbl;
1227 if (i == index || rte_is_zero_ether_addr(addrs + i))
1230 tbl = rte_is_multicast_ether_addr(addrs + i) ? mc : uc;
1231 memcpy(&tbl->macs[tbl->entries++], addrs + i,
1232 RTE_ETHER_ADDR_LEN);
1235 virtio_mac_table_set(hw, uc, mc);
1239 virtio_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1241 struct virtio_hw *hw = dev->data->dev_private;
1243 memcpy(hw->mac_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1245 /* Use atomic update if available */
1246 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1247 struct virtio_pmd_ctrl ctrl;
1248 int len = RTE_ETHER_ADDR_LEN;
1250 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1251 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1253 memcpy(ctrl.data, mac_addr, RTE_ETHER_ADDR_LEN);
1254 return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1257 if (!vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1260 virtio_set_hwaddr(hw);
1265 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1267 struct virtio_hw *hw = dev->data->dev_private;
1268 struct virtio_pmd_ctrl ctrl;
1271 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1274 ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1275 ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1276 memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1277 len = sizeof(vlan_id);
1279 return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1283 virtio_intr_unmask(struct rte_eth_dev *dev)
1285 struct virtio_hw *hw = dev->data->dev_private;
1287 if (rte_intr_ack(dev->intr_handle) < 0)
1290 if (!hw->virtio_user_dev)
1291 hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1297 virtio_intr_enable(struct rte_eth_dev *dev)
1299 struct virtio_hw *hw = dev->data->dev_private;
1301 if (rte_intr_enable(dev->intr_handle) < 0)
1304 if (!hw->virtio_user_dev)
1305 hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1311 virtio_intr_disable(struct rte_eth_dev *dev)
1313 struct virtio_hw *hw = dev->data->dev_private;
1315 if (rte_intr_disable(dev->intr_handle) < 0)
1318 if (!hw->virtio_user_dev)
1319 hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1325 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1327 uint64_t host_features;
1329 /* Prepare guest_features: feature that driver wants to support */
1330 PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1333 /* Read device(host) feature bits */
1334 host_features = VTPCI_OPS(hw)->get_features(hw);
1335 PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1338 /* If supported, ensure MTU value is valid before acknowledging it. */
1339 if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1340 struct virtio_net_config config;
1342 vtpci_read_dev_config(hw,
1343 offsetof(struct virtio_net_config, mtu),
1344 &config.mtu, sizeof(config.mtu));
1346 if (config.mtu < RTE_ETHER_MIN_MTU)
1347 req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1351 * Negotiate features: Subset of device feature bits are written back
1352 * guest feature bits.
1354 hw->guest_features = req_features;
1355 hw->guest_features = vtpci_negotiate_features(hw, host_features);
1356 PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1357 hw->guest_features);
1360 if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1362 "VIRTIO_F_VERSION_1 features is not enabled.");
1365 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1366 if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1368 "failed to set FEATURES_OK status!");
1373 hw->req_guest_features = req_features;
1379 virtio_dev_pause(struct rte_eth_dev *dev)
1381 struct virtio_hw *hw = dev->data->dev_private;
1383 rte_spinlock_lock(&hw->state_lock);
1385 if (hw->started == 0) {
1386 /* Device is just stopped. */
1387 rte_spinlock_unlock(&hw->state_lock);
1392 * Prevent the worker threads from touching queues to avoid contention,
1393 * 1 ms should be enough for the ongoing Tx function to finish.
1400 * Recover hw state to let the worker threads continue.
1403 virtio_dev_resume(struct rte_eth_dev *dev)
1405 struct virtio_hw *hw = dev->data->dev_private;
1408 rte_spinlock_unlock(&hw->state_lock);
1412 * Should be called only after device is paused.
1415 virtio_inject_pkts(struct rte_eth_dev *dev, struct rte_mbuf **tx_pkts,
1418 struct virtio_hw *hw = dev->data->dev_private;
1419 struct virtnet_tx *txvq = dev->data->tx_queues[0];
1422 hw->inject_pkts = tx_pkts;
1423 ret = dev->tx_pkt_burst(txvq, tx_pkts, nb_pkts);
1424 hw->inject_pkts = NULL;
1430 virtio_notify_peers(struct rte_eth_dev *dev)
1432 struct virtio_hw *hw = dev->data->dev_private;
1433 struct virtnet_rx *rxvq;
1434 struct rte_mbuf *rarp_mbuf;
1436 if (!dev->data->rx_queues)
1439 rxvq = dev->data->rx_queues[0];
1443 rarp_mbuf = rte_net_make_rarp_packet(rxvq->mpool,
1444 (struct rte_ether_addr *)hw->mac_addr);
1445 if (rarp_mbuf == NULL) {
1446 PMD_DRV_LOG(ERR, "failed to make RARP packet.");
1450 /* If virtio port just stopped, no need to send RARP */
1451 if (virtio_dev_pause(dev) < 0) {
1452 rte_pktmbuf_free(rarp_mbuf);
1456 virtio_inject_pkts(dev, &rarp_mbuf, 1);
1457 virtio_dev_resume(dev);
1461 virtio_ack_link_announce(struct rte_eth_dev *dev)
1463 struct virtio_hw *hw = dev->data->dev_private;
1464 struct virtio_pmd_ctrl ctrl;
1466 ctrl.hdr.class = VIRTIO_NET_CTRL_ANNOUNCE;
1467 ctrl.hdr.cmd = VIRTIO_NET_CTRL_ANNOUNCE_ACK;
1469 virtio_send_command(hw->cvq, &ctrl, NULL, 0);
1473 * Process virtio config changed interrupt. Call the callback
1474 * if link state changed, generate gratuitous RARP packet if
1475 * the status indicates an ANNOUNCE.
1478 virtio_interrupt_handler(void *param)
1480 struct rte_eth_dev *dev = param;
1481 struct virtio_hw *hw = dev->data->dev_private;
1485 /* Read interrupt status which clears interrupt */
1486 isr = vtpci_isr(hw);
1487 PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1489 if (virtio_intr_unmask(dev) < 0)
1490 PMD_DRV_LOG(ERR, "interrupt enable failed");
1492 if (isr & VIRTIO_PCI_ISR_CONFIG) {
1493 if (virtio_dev_link_update(dev, 0) == 0)
1494 _rte_eth_dev_callback_process(dev,
1495 RTE_ETH_EVENT_INTR_LSC,
1498 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1499 vtpci_read_dev_config(hw,
1500 offsetof(struct virtio_net_config, status),
1501 &status, sizeof(status));
1502 if (status & VIRTIO_NET_S_ANNOUNCE) {
1503 virtio_notify_peers(dev);
1505 virtio_ack_link_announce(dev);
1511 /* set rx and tx handlers according to what is supported */
1513 set_rxtx_funcs(struct rte_eth_dev *eth_dev)
1515 struct virtio_hw *hw = eth_dev->data->dev_private;
1517 eth_dev->tx_pkt_prepare = virtio_xmit_pkts_prepare;
1518 if (vtpci_packed_queue(hw)) {
1520 "virtio: using packed ring %s Tx path on port %u",
1521 hw->use_inorder_tx ? "inorder" : "standard",
1522 eth_dev->data->port_id);
1523 eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed;
1525 if (hw->use_inorder_tx) {
1526 PMD_INIT_LOG(INFO, "virtio: using inorder Tx path on port %u",
1527 eth_dev->data->port_id);
1528 eth_dev->tx_pkt_burst = virtio_xmit_pkts_inorder;
1530 PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
1531 eth_dev->data->port_id);
1532 eth_dev->tx_pkt_burst = virtio_xmit_pkts;
1536 if (vtpci_packed_queue(hw)) {
1537 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1539 "virtio: using packed ring mergeable buffer Rx path on port %u",
1540 eth_dev->data->port_id);
1541 eth_dev->rx_pkt_burst =
1542 &virtio_recv_mergeable_pkts_packed;
1545 "virtio: using packed ring standard Rx path on port %u",
1546 eth_dev->data->port_id);
1547 eth_dev->rx_pkt_burst = &virtio_recv_pkts_packed;
1550 if (hw->use_simple_rx) {
1551 PMD_INIT_LOG(INFO, "virtio: using simple Rx path on port %u",
1552 eth_dev->data->port_id);
1553 eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1554 } else if (hw->use_inorder_rx) {
1556 "virtio: using inorder Rx path on port %u",
1557 eth_dev->data->port_id);
1558 eth_dev->rx_pkt_burst = &virtio_recv_pkts_inorder;
1559 } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1561 "virtio: using mergeable buffer Rx path on port %u",
1562 eth_dev->data->port_id);
1563 eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1565 PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
1566 eth_dev->data->port_id);
1567 eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1573 /* Only support 1:1 queue/interrupt mapping so far.
1574 * TODO: support n:1 queue/interrupt mapping when there are limited number of
1575 * interrupt vectors (<N+1).
1578 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1581 struct virtio_hw *hw = dev->data->dev_private;
1583 PMD_INIT_LOG(INFO, "queue/interrupt binding");
1584 for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1585 dev->intr_handle->intr_vec[i] = i + 1;
1586 if (VTPCI_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1587 VIRTIO_MSI_NO_VECTOR) {
1588 PMD_DRV_LOG(ERR, "failed to set queue vector");
1597 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1600 struct virtio_hw *hw = dev->data->dev_private;
1602 PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1603 for (i = 0; i < dev->data->nb_rx_queues; ++i)
1604 VTPCI_OPS(hw)->set_queue_irq(hw,
1605 hw->vqs[i * VTNET_CQ],
1606 VIRTIO_MSI_NO_VECTOR);
1610 virtio_configure_intr(struct rte_eth_dev *dev)
1612 struct virtio_hw *hw = dev->data->dev_private;
1614 if (!rte_intr_cap_multiple(dev->intr_handle)) {
1615 PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1619 if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1620 PMD_INIT_LOG(ERR, "Fail to create eventfd");
1624 if (!dev->intr_handle->intr_vec) {
1625 dev->intr_handle->intr_vec =
1626 rte_zmalloc("intr_vec",
1627 hw->max_queue_pairs * sizeof(int), 0);
1628 if (!dev->intr_handle->intr_vec) {
1629 PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1630 hw->max_queue_pairs);
1635 /* Re-register callback to update max_intr */
1636 rte_intr_callback_unregister(dev->intr_handle,
1637 virtio_interrupt_handler,
1639 rte_intr_callback_register(dev->intr_handle,
1640 virtio_interrupt_handler,
1643 /* DO NOT try to remove this! This function will enable msix, or QEMU
1644 * will encounter SIGSEGV when DRIVER_OK is sent.
1645 * And for legacy devices, this should be done before queue/vec binding
1646 * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1647 * (22) will be ignored.
1649 if (virtio_intr_enable(dev) < 0) {
1650 PMD_DRV_LOG(ERR, "interrupt enable failed");
1654 if (virtio_queues_bind_intr(dev) < 0) {
1655 PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1662 /* reset device and renegotiate features if needed */
1664 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1666 struct virtio_hw *hw = eth_dev->data->dev_private;
1667 struct virtio_net_config *config;
1668 struct virtio_net_config local_config;
1669 struct rte_pci_device *pci_dev = NULL;
1672 /* Reset the device although not necessary at startup */
1676 virtio_dev_free_mbufs(eth_dev);
1677 virtio_free_queues(hw);
1680 /* Tell the host we've noticed this device. */
1681 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1683 /* Tell the host we've known how to drive the device. */
1684 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1685 if (virtio_negotiate_features(hw, req_features) < 0)
1688 hw->weak_barriers = !vtpci_with_feature(hw, VIRTIO_F_ORDER_PLATFORM);
1690 if (!hw->virtio_user_dev)
1691 pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1693 /* If host does not support both status and MSI-X then disable LSC */
1694 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS) &&
1695 hw->use_msix != VIRTIO_MSIX_NONE)
1696 eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1698 eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1700 /* Setting up rx_header size for the device */
1701 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1702 vtpci_with_feature(hw, VIRTIO_F_VERSION_1) ||
1703 vtpci_with_feature(hw, VIRTIO_F_RING_PACKED))
1704 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1706 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1708 /* Copy the permanent MAC address to: virtio_hw */
1709 virtio_get_hwaddr(hw);
1710 rte_ether_addr_copy((struct rte_ether_addr *)hw->mac_addr,
1711 ð_dev->data->mac_addrs[0]);
1713 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1714 hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1715 hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1717 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1718 config = &local_config;
1720 vtpci_read_dev_config(hw,
1721 offsetof(struct virtio_net_config, mac),
1722 &config->mac, sizeof(config->mac));
1724 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1725 vtpci_read_dev_config(hw,
1726 offsetof(struct virtio_net_config, status),
1727 &config->status, sizeof(config->status));
1730 "VIRTIO_NET_F_STATUS is not supported");
1734 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1735 vtpci_read_dev_config(hw,
1736 offsetof(struct virtio_net_config, max_virtqueue_pairs),
1737 &config->max_virtqueue_pairs,
1738 sizeof(config->max_virtqueue_pairs));
1741 "VIRTIO_NET_F_MQ is not supported");
1742 config->max_virtqueue_pairs = 1;
1745 hw->max_queue_pairs = config->max_virtqueue_pairs;
1747 if (vtpci_with_feature(hw, VIRTIO_NET_F_MTU)) {
1748 vtpci_read_dev_config(hw,
1749 offsetof(struct virtio_net_config, mtu),
1751 sizeof(config->mtu));
1754 * MTU value has already been checked at negotiation
1755 * time, but check again in case it has changed since
1756 * then, which should not happen.
1758 if (config->mtu < RTE_ETHER_MIN_MTU) {
1759 PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1764 hw->max_mtu = config->mtu;
1765 /* Set initial MTU to maximum one supported by vhost */
1766 eth_dev->data->mtu = config->mtu;
1769 hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1770 VLAN_TAG_LEN - hw->vtnet_hdr_size;
1773 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1774 config->max_virtqueue_pairs);
1775 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1777 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1778 config->mac[0], config->mac[1],
1779 config->mac[2], config->mac[3],
1780 config->mac[4], config->mac[5]);
1782 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1783 hw->max_queue_pairs = 1;
1784 hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1785 VLAN_TAG_LEN - hw->vtnet_hdr_size;
1788 ret = virtio_alloc_queues(eth_dev);
1792 if (eth_dev->data->dev_conf.intr_conf.rxq) {
1793 if (virtio_configure_intr(eth_dev) < 0) {
1794 PMD_INIT_LOG(ERR, "failed to configure interrupt");
1795 virtio_free_queues(hw);
1800 vtpci_reinit_complete(hw);
1803 PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1804 eth_dev->data->port_id, pci_dev->id.vendor_id,
1805 pci_dev->id.device_id);
1811 * Remap the PCI device again (IO port map for legacy device and
1812 * memory map for modern device), so that the secondary process
1813 * could have the PCI initiated correctly.
1816 virtio_remap_pci(struct rte_pci_device *pci_dev, struct virtio_hw *hw)
1820 * We don't have to re-parse the PCI config space, since
1821 * rte_pci_map_device() makes sure the mapped address
1822 * in secondary process would equal to the one mapped in
1823 * the primary process: error will be returned if that
1824 * requirement is not met.
1826 * That said, we could simply reuse all cap pointers
1827 * (such as dev_cfg, common_cfg, etc.) parsed from the
1828 * primary process, which is stored in shared memory.
1830 if (rte_pci_map_device(pci_dev)) {
1831 PMD_INIT_LOG(DEBUG, "failed to map pci device!");
1835 if (rte_pci_ioport_map(pci_dev, 0, VTPCI_IO(hw)) < 0)
1843 virtio_set_vtpci_ops(struct virtio_hw *hw)
1845 #ifdef RTE_VIRTIO_USER
1846 if (hw->virtio_user_dev)
1847 VTPCI_OPS(hw) = &virtio_user_ops;
1851 VTPCI_OPS(hw) = &modern_ops;
1853 VTPCI_OPS(hw) = &legacy_ops;
1857 * This function is based on probe() function in virtio_pci.c
1858 * It returns 0 on success.
1861 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1863 struct virtio_hw *hw = eth_dev->data->dev_private;
1866 if (sizeof(struct virtio_net_hdr_mrg_rxbuf) > RTE_PKTMBUF_HEADROOM) {
1868 "Not sufficient headroom required = %d, avail = %d",
1869 (int)sizeof(struct virtio_net_hdr_mrg_rxbuf),
1870 RTE_PKTMBUF_HEADROOM);
1875 eth_dev->dev_ops = &virtio_eth_dev_ops;
1877 if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1878 if (!hw->virtio_user_dev) {
1879 ret = virtio_remap_pci(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1884 virtio_set_vtpci_ops(hw);
1885 set_rxtx_funcs(eth_dev);
1891 * Pass the information to the rte_eth_dev_close() that it should also
1892 * release the private port resources.
1894 eth_dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
1896 /* Allocate memory for storing MAC addresses */
1897 eth_dev->data->mac_addrs = rte_zmalloc("virtio",
1898 VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN, 0);
1899 if (eth_dev->data->mac_addrs == NULL) {
1901 "Failed to allocate %d bytes needed to store MAC addresses",
1902 VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN);
1906 hw->port_id = eth_dev->data->port_id;
1907 /* For virtio_user case the hw->virtio_user_dev is populated by
1908 * virtio_user_eth_dev_alloc() before eth_virtio_dev_init() is called.
1910 if (!hw->virtio_user_dev) {
1911 ret = vtpci_init(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1913 goto err_vtpci_init;
1916 /* reset device and negotiate default features */
1917 ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1919 goto err_virtio_init;
1926 if (!hw->virtio_user_dev) {
1927 rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(eth_dev));
1929 rte_pci_ioport_unmap(VTPCI_IO(hw));
1932 rte_free(eth_dev->data->mac_addrs);
1933 eth_dev->data->mac_addrs = NULL;
1938 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1940 PMD_INIT_FUNC_TRACE();
1942 if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1945 virtio_dev_stop(eth_dev);
1946 virtio_dev_close(eth_dev);
1948 eth_dev->dev_ops = NULL;
1949 eth_dev->tx_pkt_burst = NULL;
1950 eth_dev->rx_pkt_burst = NULL;
1952 PMD_INIT_LOG(DEBUG, "dev_uninit completed");
1957 static int vdpa_check_handler(__rte_unused const char *key,
1958 const char *value, __rte_unused void *opaque)
1960 if (strcmp(value, "1"))
1967 vdpa_mode_selected(struct rte_devargs *devargs)
1969 struct rte_kvargs *kvlist;
1970 const char *key = "vdpa";
1973 if (devargs == NULL)
1976 kvlist = rte_kvargs_parse(devargs->args, NULL);
1980 if (!rte_kvargs_count(kvlist, key))
1983 /* vdpa mode selected when there's a key-value pair: vdpa=1 */
1984 if (rte_kvargs_process(kvlist, key,
1985 vdpa_check_handler, NULL) < 0) {
1991 rte_kvargs_free(kvlist);
1995 static int eth_virtio_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1996 struct rte_pci_device *pci_dev)
1998 if (rte_eal_iopl_init() != 0) {
1999 PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
2003 /* virtio pmd skips probe if device needs to work in vdpa mode */
2004 if (vdpa_mode_selected(pci_dev->device.devargs))
2007 return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct virtio_hw),
2008 eth_virtio_dev_init);
2011 static int eth_virtio_pci_remove(struct rte_pci_device *pci_dev)
2015 ret = rte_eth_dev_pci_generic_remove(pci_dev, eth_virtio_dev_uninit);
2016 /* Port has already been released by close. */
2022 static struct rte_pci_driver rte_virtio_pmd = {
2024 .name = "net_virtio",
2026 .id_table = pci_id_virtio_map,
2028 .probe = eth_virtio_pci_probe,
2029 .remove = eth_virtio_pci_remove,
2032 RTE_INIT(rte_virtio_pmd_init)
2034 rte_eal_iopl_init();
2035 rte_pci_register(&rte_virtio_pmd);
2039 rx_offload_enabled(struct virtio_hw *hw)
2041 return vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM) ||
2042 vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2043 vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6);
2047 tx_offload_enabled(struct virtio_hw *hw)
2049 return vtpci_with_feature(hw, VIRTIO_NET_F_CSUM) ||
2050 vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO4) ||
2051 vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO6);
2055 * Configure virtio device
2056 * It returns 0 on success.
2059 virtio_dev_configure(struct rte_eth_dev *dev)
2061 const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2062 const struct rte_eth_txmode *txmode = &dev->data->dev_conf.txmode;
2063 struct virtio_hw *hw = dev->data->dev_private;
2064 uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
2066 uint64_t rx_offloads = rxmode->offloads;
2067 uint64_t tx_offloads = txmode->offloads;
2068 uint64_t req_features;
2071 PMD_INIT_LOG(DEBUG, "configure");
2072 req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
2074 if (dev->data->dev_conf.intr_conf.rxq) {
2075 ret = virtio_init_device(dev, hw->req_guest_features);
2080 if (rxmode->max_rx_pkt_len > hw->max_mtu + ether_hdr_len)
2081 req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
2083 if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2084 DEV_RX_OFFLOAD_TCP_CKSUM))
2085 req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
2087 if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO)
2089 (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2090 (1ULL << VIRTIO_NET_F_GUEST_TSO6);
2092 if (tx_offloads & (DEV_TX_OFFLOAD_UDP_CKSUM |
2093 DEV_TX_OFFLOAD_TCP_CKSUM))
2094 req_features |= (1ULL << VIRTIO_NET_F_CSUM);
2096 if (tx_offloads & DEV_TX_OFFLOAD_TCP_TSO)
2098 (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2099 (1ULL << VIRTIO_NET_F_HOST_TSO6);
2101 /* if request features changed, reinit the device */
2102 if (req_features != hw->req_guest_features) {
2103 ret = virtio_init_device(dev, req_features);
2108 if ((rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2109 DEV_RX_OFFLOAD_TCP_CKSUM)) &&
2110 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
2112 "rx checksum not available on this host");
2116 if ((rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) &&
2117 (!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2118 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6))) {
2120 "Large Receive Offload not available on this host");
2124 /* start control queue */
2125 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
2126 virtio_dev_cq_start(dev);
2128 if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
2131 if ((rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
2132 && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2134 "vlan filtering not available on this host");
2138 hw->has_tx_offload = tx_offload_enabled(hw);
2139 hw->has_rx_offload = rx_offload_enabled(hw);
2141 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2142 /* Enable vector (0) for Link State Intrerrupt */
2143 if (VTPCI_OPS(hw)->set_config_irq(hw, 0) ==
2144 VIRTIO_MSI_NO_VECTOR) {
2145 PMD_DRV_LOG(ERR, "failed to set config vector");
2149 rte_spinlock_init(&hw->state_lock);
2151 hw->use_simple_rx = 1;
2153 if (vtpci_with_feature(hw, VIRTIO_F_IN_ORDER)) {
2154 hw->use_inorder_tx = 1;
2155 hw->use_inorder_rx = 1;
2156 hw->use_simple_rx = 0;
2159 if (vtpci_packed_queue(hw)) {
2160 hw->use_simple_rx = 0;
2161 hw->use_inorder_rx = 0;
2164 #if defined RTE_ARCH_ARM64 || defined RTE_ARCH_ARM
2165 if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
2166 hw->use_simple_rx = 0;
2169 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2170 hw->use_simple_rx = 0;
2173 if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2174 DEV_RX_OFFLOAD_TCP_CKSUM |
2175 DEV_RX_OFFLOAD_TCP_LRO |
2176 DEV_RX_OFFLOAD_VLAN_STRIP))
2177 hw->use_simple_rx = 0;
2184 virtio_dev_start(struct rte_eth_dev *dev)
2186 uint16_t nb_queues, i;
2187 struct virtnet_rx *rxvq;
2188 struct virtnet_tx *txvq __rte_unused;
2189 struct virtio_hw *hw = dev->data->dev_private;
2192 /* Finish the initialization of the queues */
2193 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2194 ret = virtio_dev_rx_queue_setup_finish(dev, i);
2198 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2199 ret = virtio_dev_tx_queue_setup_finish(dev, i);
2204 /* check if lsc interrupt feature is enabled */
2205 if (dev->data->dev_conf.intr_conf.lsc) {
2206 if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
2207 PMD_DRV_LOG(ERR, "link status not supported by host");
2212 /* Enable uio/vfio intr/eventfd mapping: althrough we already did that
2213 * in device configure, but it could be unmapped when device is
2216 if (dev->data->dev_conf.intr_conf.lsc ||
2217 dev->data->dev_conf.intr_conf.rxq) {
2218 virtio_intr_disable(dev);
2220 /* Setup interrupt callback */
2221 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2222 rte_intr_callback_register(dev->intr_handle,
2223 virtio_interrupt_handler,
2226 if (virtio_intr_enable(dev) < 0) {
2227 PMD_DRV_LOG(ERR, "interrupt enable failed");
2232 /*Notify the backend
2233 *Otherwise the tap backend might already stop its queue due to fullness.
2234 *vhost backend will have no chance to be waked up
2236 nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
2237 if (hw->max_queue_pairs > 1) {
2238 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
2242 PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
2244 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2245 rxvq = dev->data->rx_queues[i];
2246 /* Flush the old packets */
2247 virtqueue_rxvq_flush(rxvq->vq);
2248 virtqueue_notify(rxvq->vq);
2251 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2252 txvq = dev->data->tx_queues[i];
2253 virtqueue_notify(txvq->vq);
2256 PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
2258 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2259 rxvq = dev->data->rx_queues[i];
2260 VIRTQUEUE_DUMP(rxvq->vq);
2263 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2264 txvq = dev->data->tx_queues[i];
2265 VIRTQUEUE_DUMP(txvq->vq);
2268 set_rxtx_funcs(dev);
2271 /* Initialize Link state */
2272 virtio_dev_link_update(dev, 0);
2277 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
2279 struct virtio_hw *hw = dev->data->dev_private;
2280 uint16_t nr_vq = virtio_get_nr_vq(hw);
2281 const char *type __rte_unused;
2282 unsigned int i, mbuf_num = 0;
2283 struct virtqueue *vq;
2284 struct rte_mbuf *buf;
2287 if (hw->vqs == NULL)
2290 for (i = 0; i < nr_vq; i++) {
2295 queue_type = virtio_get_queue_type(hw, i);
2296 if (queue_type == VTNET_RQ)
2298 else if (queue_type == VTNET_TQ)
2304 "Before freeing %s[%d] used and unused buf",
2308 while ((buf = virtqueue_detach_unused(vq)) != NULL) {
2309 rte_pktmbuf_free(buf);
2314 "After freeing %s[%d] used and unused buf",
2319 PMD_INIT_LOG(DEBUG, "%d mbufs freed", mbuf_num);
2323 * Stop device: disable interrupt and mark link down
2326 virtio_dev_stop(struct rte_eth_dev *dev)
2328 struct virtio_hw *hw = dev->data->dev_private;
2329 struct rte_eth_link link;
2330 struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
2332 PMD_INIT_LOG(DEBUG, "stop");
2334 rte_spinlock_lock(&hw->state_lock);
2337 hw->started = false;
2339 if (intr_conf->lsc || intr_conf->rxq) {
2340 virtio_intr_disable(dev);
2342 /* Reset interrupt callback */
2343 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
2344 rte_intr_callback_unregister(dev->intr_handle,
2345 virtio_interrupt_handler,
2350 memset(&link, 0, sizeof(link));
2351 rte_eth_linkstatus_set(dev, &link);
2353 rte_spinlock_unlock(&hw->state_lock);
2357 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
2359 struct rte_eth_link link;
2361 struct virtio_hw *hw = dev->data->dev_private;
2363 memset(&link, 0, sizeof(link));
2364 link.link_duplex = ETH_LINK_FULL_DUPLEX;
2365 link.link_speed = ETH_SPEED_NUM_10G;
2366 link.link_autoneg = ETH_LINK_FIXED;
2369 link.link_status = ETH_LINK_DOWN;
2370 } else if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
2371 PMD_INIT_LOG(DEBUG, "Get link status from hw");
2372 vtpci_read_dev_config(hw,
2373 offsetof(struct virtio_net_config, status),
2374 &status, sizeof(status));
2375 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
2376 link.link_status = ETH_LINK_DOWN;
2377 PMD_INIT_LOG(DEBUG, "Port %d is down",
2378 dev->data->port_id);
2380 link.link_status = ETH_LINK_UP;
2381 PMD_INIT_LOG(DEBUG, "Port %d is up",
2382 dev->data->port_id);
2385 link.link_status = ETH_LINK_UP;
2388 return rte_eth_linkstatus_set(dev, &link);
2392 virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
2394 const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2395 struct virtio_hw *hw = dev->data->dev_private;
2396 uint64_t offloads = rxmode->offloads;
2398 if (mask & ETH_VLAN_FILTER_MASK) {
2399 if ((offloads & DEV_RX_OFFLOAD_VLAN_FILTER) &&
2400 !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2403 "vlan filtering not available on this host");
2409 if (mask & ETH_VLAN_STRIP_MASK)
2410 hw->vlan_strip = !!(offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
2416 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2418 uint64_t tso_mask, host_features;
2419 struct virtio_hw *hw = dev->data->dev_private;
2421 dev_info->speed_capa = ETH_LINK_SPEED_10G; /* fake value */
2423 dev_info->max_rx_queues =
2424 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
2425 dev_info->max_tx_queues =
2426 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
2427 dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
2428 dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
2429 dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
2431 host_features = VTPCI_OPS(hw)->get_features(hw);
2432 dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
2433 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_JUMBO_FRAME;
2434 if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
2435 dev_info->rx_offload_capa |=
2436 DEV_RX_OFFLOAD_TCP_CKSUM |
2437 DEV_RX_OFFLOAD_UDP_CKSUM;
2439 if (host_features & (1ULL << VIRTIO_NET_F_CTRL_VLAN))
2440 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_VLAN_FILTER;
2441 tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2442 (1ULL << VIRTIO_NET_F_GUEST_TSO6);
2443 if ((host_features & tso_mask) == tso_mask)
2444 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
2446 dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
2447 DEV_TX_OFFLOAD_VLAN_INSERT;
2448 if (host_features & (1ULL << VIRTIO_NET_F_CSUM)) {
2449 dev_info->tx_offload_capa |=
2450 DEV_TX_OFFLOAD_UDP_CKSUM |
2451 DEV_TX_OFFLOAD_TCP_CKSUM;
2453 tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2454 (1ULL << VIRTIO_NET_F_HOST_TSO6);
2455 if ((host_features & tso_mask) == tso_mask)
2456 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
2462 * It enables testpmd to collect per queue stats.
2465 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
2466 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
2467 __rte_unused uint8_t is_rx)
2472 RTE_PMD_EXPORT_NAME(net_virtio, __COUNTER__);
2473 RTE_PMD_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map);
2474 RTE_PMD_REGISTER_KMOD_DEP(net_virtio, "* igb_uio | uio_pci_generic | vfio-pci");
2476 RTE_INIT(virtio_init_log)
2478 virtio_logtype_init = rte_log_register("pmd.net.virtio.init");
2479 if (virtio_logtype_init >= 0)
2480 rte_log_set_level(virtio_logtype_init, RTE_LOG_NOTICE);
2481 virtio_logtype_driver = rte_log_register("pmd.net.virtio.driver");
2482 if (virtio_logtype_driver >= 0)
2483 rte_log_set_level(virtio_logtype_driver, RTE_LOG_NOTICE);