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 void virtio_dev_allmulticast_enable(struct rte_eth_dev *dev);
47 static void 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");
827 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
829 struct virtio_hw *hw = dev->data->dev_private;
830 struct virtio_pmd_ctrl ctrl;
834 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
835 PMD_INIT_LOG(INFO, "host does not support rx control");
839 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
840 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
844 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
846 PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
849 #define VLAN_TAG_LEN 4 /* 802.3ac tag (not DMA'd) */
851 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
853 struct virtio_hw *hw = dev->data->dev_private;
854 uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
856 uint32_t frame_size = mtu + ether_hdr_len;
857 uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
859 max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
861 if (mtu < RTE_ETHER_MIN_MTU || frame_size > max_frame_size) {
862 PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
863 RTE_ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
870 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
872 struct virtio_hw *hw = dev->data->dev_private;
873 struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
874 struct virtqueue *vq = rxvq->vq;
876 virtqueue_enable_intr(vq);
877 virtio_mb(hw->weak_barriers);
882 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
884 struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
885 struct virtqueue *vq = rxvq->vq;
887 virtqueue_disable_intr(vq);
892 * dev_ops for virtio, bare necessities for basic operation
894 static const struct eth_dev_ops virtio_eth_dev_ops = {
895 .dev_configure = virtio_dev_configure,
896 .dev_start = virtio_dev_start,
897 .dev_stop = virtio_dev_stop,
898 .dev_close = virtio_dev_close,
899 .promiscuous_enable = virtio_dev_promiscuous_enable,
900 .promiscuous_disable = virtio_dev_promiscuous_disable,
901 .allmulticast_enable = virtio_dev_allmulticast_enable,
902 .allmulticast_disable = virtio_dev_allmulticast_disable,
903 .mtu_set = virtio_mtu_set,
904 .dev_infos_get = virtio_dev_info_get,
905 .stats_get = virtio_dev_stats_get,
906 .xstats_get = virtio_dev_xstats_get,
907 .xstats_get_names = virtio_dev_xstats_get_names,
908 .stats_reset = virtio_dev_stats_reset,
909 .xstats_reset = virtio_dev_stats_reset,
910 .link_update = virtio_dev_link_update,
911 .vlan_offload_set = virtio_dev_vlan_offload_set,
912 .rx_queue_setup = virtio_dev_rx_queue_setup,
913 .rx_queue_intr_enable = virtio_dev_rx_queue_intr_enable,
914 .rx_queue_intr_disable = virtio_dev_rx_queue_intr_disable,
915 .rx_queue_release = virtio_dev_queue_release,
916 .rx_descriptor_done = virtio_dev_rx_queue_done,
917 .tx_queue_setup = virtio_dev_tx_queue_setup,
918 .tx_queue_release = virtio_dev_queue_release,
919 /* collect stats per queue */
920 .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
921 .vlan_filter_set = virtio_vlan_filter_set,
922 .mac_addr_add = virtio_mac_addr_add,
923 .mac_addr_remove = virtio_mac_addr_remove,
924 .mac_addr_set = virtio_mac_addr_set,
928 * dev_ops for virtio-user in secondary processes, as we just have
929 * some limited supports currently.
931 const struct eth_dev_ops virtio_user_secondary_eth_dev_ops = {
932 .dev_infos_get = virtio_dev_info_get,
933 .stats_get = virtio_dev_stats_get,
934 .xstats_get = virtio_dev_xstats_get,
935 .xstats_get_names = virtio_dev_xstats_get_names,
936 .stats_reset = virtio_dev_stats_reset,
937 .xstats_reset = virtio_dev_stats_reset,
938 /* collect stats per queue */
939 .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
943 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
947 for (i = 0; i < dev->data->nb_tx_queues; i++) {
948 const struct virtnet_tx *txvq = dev->data->tx_queues[i];
952 stats->opackets += txvq->stats.packets;
953 stats->obytes += txvq->stats.bytes;
955 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
956 stats->q_opackets[i] = txvq->stats.packets;
957 stats->q_obytes[i] = txvq->stats.bytes;
961 for (i = 0; i < dev->data->nb_rx_queues; i++) {
962 const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
966 stats->ipackets += rxvq->stats.packets;
967 stats->ibytes += rxvq->stats.bytes;
968 stats->ierrors += rxvq->stats.errors;
970 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
971 stats->q_ipackets[i] = rxvq->stats.packets;
972 stats->q_ibytes[i] = rxvq->stats.bytes;
976 stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
979 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
980 struct rte_eth_xstat_name *xstats_names,
981 __rte_unused unsigned limit)
987 unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
988 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
990 if (xstats_names != NULL) {
991 /* Note: limit checked in rte_eth_xstats_names() */
993 for (i = 0; i < dev->data->nb_rx_queues; i++) {
994 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
997 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
998 snprintf(xstats_names[count].name,
999 sizeof(xstats_names[count].name),
1001 rte_virtio_rxq_stat_strings[t].name);
1006 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1007 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1010 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1011 snprintf(xstats_names[count].name,
1012 sizeof(xstats_names[count].name),
1014 rte_virtio_txq_stat_strings[t].name);
1024 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
1030 unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
1031 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
1036 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1037 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1044 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1045 xstats[count].value = *(uint64_t *)(((char *)rxvq) +
1046 rte_virtio_rxq_stat_strings[t].offset);
1047 xstats[count].id = count;
1052 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1053 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1060 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1061 xstats[count].value = *(uint64_t *)(((char *)txvq) +
1062 rte_virtio_txq_stat_strings[t].offset);
1063 xstats[count].id = count;
1072 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1074 virtio_update_stats(dev, stats);
1080 virtio_dev_stats_reset(struct rte_eth_dev *dev)
1084 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1085 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1089 txvq->stats.packets = 0;
1090 txvq->stats.bytes = 0;
1091 txvq->stats.multicast = 0;
1092 txvq->stats.broadcast = 0;
1093 memset(txvq->stats.size_bins, 0,
1094 sizeof(txvq->stats.size_bins[0]) * 8);
1097 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1098 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1102 rxvq->stats.packets = 0;
1103 rxvq->stats.bytes = 0;
1104 rxvq->stats.errors = 0;
1105 rxvq->stats.multicast = 0;
1106 rxvq->stats.broadcast = 0;
1107 memset(rxvq->stats.size_bins, 0,
1108 sizeof(rxvq->stats.size_bins[0]) * 8);
1115 virtio_set_hwaddr(struct virtio_hw *hw)
1117 vtpci_write_dev_config(hw,
1118 offsetof(struct virtio_net_config, mac),
1119 &hw->mac_addr, RTE_ETHER_ADDR_LEN);
1123 virtio_get_hwaddr(struct virtio_hw *hw)
1125 if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
1126 vtpci_read_dev_config(hw,
1127 offsetof(struct virtio_net_config, mac),
1128 &hw->mac_addr, RTE_ETHER_ADDR_LEN);
1130 rte_eth_random_addr(&hw->mac_addr[0]);
1131 virtio_set_hwaddr(hw);
1136 virtio_mac_table_set(struct virtio_hw *hw,
1137 const struct virtio_net_ctrl_mac *uc,
1138 const struct virtio_net_ctrl_mac *mc)
1140 struct virtio_pmd_ctrl ctrl;
1143 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1144 PMD_DRV_LOG(INFO, "host does not support mac table");
1148 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1149 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1151 len[0] = uc->entries * RTE_ETHER_ADDR_LEN + sizeof(uc->entries);
1152 memcpy(ctrl.data, uc, len[0]);
1154 len[1] = mc->entries * RTE_ETHER_ADDR_LEN + sizeof(mc->entries);
1155 memcpy(ctrl.data + len[0], mc, len[1]);
1157 err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1159 PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1164 virtio_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
1165 uint32_t index, uint32_t vmdq __rte_unused)
1167 struct virtio_hw *hw = dev->data->dev_private;
1168 const struct rte_ether_addr *addrs = dev->data->mac_addrs;
1170 struct virtio_net_ctrl_mac *uc, *mc;
1172 if (index >= VIRTIO_MAX_MAC_ADDRS) {
1173 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1177 uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1178 sizeof(uc->entries));
1180 mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1181 sizeof(mc->entries));
1184 for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1185 const struct rte_ether_addr *addr
1186 = (i == index) ? mac_addr : addrs + i;
1187 struct virtio_net_ctrl_mac *tbl
1188 = rte_is_multicast_ether_addr(addr) ? mc : uc;
1190 memcpy(&tbl->macs[tbl->entries++], addr, RTE_ETHER_ADDR_LEN);
1193 return virtio_mac_table_set(hw, uc, mc);
1197 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1199 struct virtio_hw *hw = dev->data->dev_private;
1200 struct rte_ether_addr *addrs = dev->data->mac_addrs;
1201 struct virtio_net_ctrl_mac *uc, *mc;
1204 if (index >= VIRTIO_MAX_MAC_ADDRS) {
1205 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1209 uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1210 sizeof(uc->entries));
1212 mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1213 sizeof(mc->entries));
1216 for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1217 struct virtio_net_ctrl_mac *tbl;
1219 if (i == index || rte_is_zero_ether_addr(addrs + i))
1222 tbl = rte_is_multicast_ether_addr(addrs + i) ? mc : uc;
1223 memcpy(&tbl->macs[tbl->entries++], addrs + i,
1224 RTE_ETHER_ADDR_LEN);
1227 virtio_mac_table_set(hw, uc, mc);
1231 virtio_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1233 struct virtio_hw *hw = dev->data->dev_private;
1235 memcpy(hw->mac_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1237 /* Use atomic update if available */
1238 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1239 struct virtio_pmd_ctrl ctrl;
1240 int len = RTE_ETHER_ADDR_LEN;
1242 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1243 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1245 memcpy(ctrl.data, mac_addr, RTE_ETHER_ADDR_LEN);
1246 return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1249 if (!vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1252 virtio_set_hwaddr(hw);
1257 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1259 struct virtio_hw *hw = dev->data->dev_private;
1260 struct virtio_pmd_ctrl ctrl;
1263 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1266 ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1267 ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1268 memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1269 len = sizeof(vlan_id);
1271 return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1275 virtio_intr_unmask(struct rte_eth_dev *dev)
1277 struct virtio_hw *hw = dev->data->dev_private;
1279 if (rte_intr_ack(dev->intr_handle) < 0)
1282 if (!hw->virtio_user_dev)
1283 hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1289 virtio_intr_enable(struct rte_eth_dev *dev)
1291 struct virtio_hw *hw = dev->data->dev_private;
1293 if (rte_intr_enable(dev->intr_handle) < 0)
1296 if (!hw->virtio_user_dev)
1297 hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1303 virtio_intr_disable(struct rte_eth_dev *dev)
1305 struct virtio_hw *hw = dev->data->dev_private;
1307 if (rte_intr_disable(dev->intr_handle) < 0)
1310 if (!hw->virtio_user_dev)
1311 hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1317 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1319 uint64_t host_features;
1321 /* Prepare guest_features: feature that driver wants to support */
1322 PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1325 /* Read device(host) feature bits */
1326 host_features = VTPCI_OPS(hw)->get_features(hw);
1327 PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1330 /* If supported, ensure MTU value is valid before acknowledging it. */
1331 if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1332 struct virtio_net_config config;
1334 vtpci_read_dev_config(hw,
1335 offsetof(struct virtio_net_config, mtu),
1336 &config.mtu, sizeof(config.mtu));
1338 if (config.mtu < RTE_ETHER_MIN_MTU)
1339 req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1343 * Negotiate features: Subset of device feature bits are written back
1344 * guest feature bits.
1346 hw->guest_features = req_features;
1347 hw->guest_features = vtpci_negotiate_features(hw, host_features);
1348 PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1349 hw->guest_features);
1352 if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1354 "VIRTIO_F_VERSION_1 features is not enabled.");
1357 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1358 if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1360 "failed to set FEATURES_OK status!");
1365 hw->req_guest_features = req_features;
1371 virtio_dev_pause(struct rte_eth_dev *dev)
1373 struct virtio_hw *hw = dev->data->dev_private;
1375 rte_spinlock_lock(&hw->state_lock);
1377 if (hw->started == 0) {
1378 /* Device is just stopped. */
1379 rte_spinlock_unlock(&hw->state_lock);
1384 * Prevent the worker threads from touching queues to avoid contention,
1385 * 1 ms should be enough for the ongoing Tx function to finish.
1392 * Recover hw state to let the worker threads continue.
1395 virtio_dev_resume(struct rte_eth_dev *dev)
1397 struct virtio_hw *hw = dev->data->dev_private;
1400 rte_spinlock_unlock(&hw->state_lock);
1404 * Should be called only after device is paused.
1407 virtio_inject_pkts(struct rte_eth_dev *dev, struct rte_mbuf **tx_pkts,
1410 struct virtio_hw *hw = dev->data->dev_private;
1411 struct virtnet_tx *txvq = dev->data->tx_queues[0];
1414 hw->inject_pkts = tx_pkts;
1415 ret = dev->tx_pkt_burst(txvq, tx_pkts, nb_pkts);
1416 hw->inject_pkts = NULL;
1422 virtio_notify_peers(struct rte_eth_dev *dev)
1424 struct virtio_hw *hw = dev->data->dev_private;
1425 struct virtnet_rx *rxvq;
1426 struct rte_mbuf *rarp_mbuf;
1428 if (!dev->data->rx_queues)
1431 rxvq = dev->data->rx_queues[0];
1435 rarp_mbuf = rte_net_make_rarp_packet(rxvq->mpool,
1436 (struct rte_ether_addr *)hw->mac_addr);
1437 if (rarp_mbuf == NULL) {
1438 PMD_DRV_LOG(ERR, "failed to make RARP packet.");
1442 /* If virtio port just stopped, no need to send RARP */
1443 if (virtio_dev_pause(dev) < 0) {
1444 rte_pktmbuf_free(rarp_mbuf);
1448 virtio_inject_pkts(dev, &rarp_mbuf, 1);
1449 virtio_dev_resume(dev);
1453 virtio_ack_link_announce(struct rte_eth_dev *dev)
1455 struct virtio_hw *hw = dev->data->dev_private;
1456 struct virtio_pmd_ctrl ctrl;
1458 ctrl.hdr.class = VIRTIO_NET_CTRL_ANNOUNCE;
1459 ctrl.hdr.cmd = VIRTIO_NET_CTRL_ANNOUNCE_ACK;
1461 virtio_send_command(hw->cvq, &ctrl, NULL, 0);
1465 * Process virtio config changed interrupt. Call the callback
1466 * if link state changed, generate gratuitous RARP packet if
1467 * the status indicates an ANNOUNCE.
1470 virtio_interrupt_handler(void *param)
1472 struct rte_eth_dev *dev = param;
1473 struct virtio_hw *hw = dev->data->dev_private;
1477 /* Read interrupt status which clears interrupt */
1478 isr = vtpci_isr(hw);
1479 PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1481 if (virtio_intr_unmask(dev) < 0)
1482 PMD_DRV_LOG(ERR, "interrupt enable failed");
1484 if (isr & VIRTIO_PCI_ISR_CONFIG) {
1485 if (virtio_dev_link_update(dev, 0) == 0)
1486 _rte_eth_dev_callback_process(dev,
1487 RTE_ETH_EVENT_INTR_LSC,
1490 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1491 vtpci_read_dev_config(hw,
1492 offsetof(struct virtio_net_config, status),
1493 &status, sizeof(status));
1494 if (status & VIRTIO_NET_S_ANNOUNCE) {
1495 virtio_notify_peers(dev);
1497 virtio_ack_link_announce(dev);
1503 /* set rx and tx handlers according to what is supported */
1505 set_rxtx_funcs(struct rte_eth_dev *eth_dev)
1507 struct virtio_hw *hw = eth_dev->data->dev_private;
1509 eth_dev->tx_pkt_prepare = virtio_xmit_pkts_prepare;
1510 if (vtpci_packed_queue(hw)) {
1512 "virtio: using packed ring %s Tx path on port %u",
1513 hw->use_inorder_tx ? "inorder" : "standard",
1514 eth_dev->data->port_id);
1515 eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed;
1517 if (hw->use_inorder_tx) {
1518 PMD_INIT_LOG(INFO, "virtio: using inorder Tx path on port %u",
1519 eth_dev->data->port_id);
1520 eth_dev->tx_pkt_burst = virtio_xmit_pkts_inorder;
1522 PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
1523 eth_dev->data->port_id);
1524 eth_dev->tx_pkt_burst = virtio_xmit_pkts;
1528 if (vtpci_packed_queue(hw)) {
1529 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1531 "virtio: using packed ring mergeable buffer Rx path on port %u",
1532 eth_dev->data->port_id);
1533 eth_dev->rx_pkt_burst =
1534 &virtio_recv_mergeable_pkts_packed;
1537 "virtio: using packed ring standard Rx path on port %u",
1538 eth_dev->data->port_id);
1539 eth_dev->rx_pkt_burst = &virtio_recv_pkts_packed;
1542 if (hw->use_simple_rx) {
1543 PMD_INIT_LOG(INFO, "virtio: using simple Rx path on port %u",
1544 eth_dev->data->port_id);
1545 eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1546 } else if (hw->use_inorder_rx) {
1548 "virtio: using inorder Rx path on port %u",
1549 eth_dev->data->port_id);
1550 eth_dev->rx_pkt_burst = &virtio_recv_pkts_inorder;
1551 } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1553 "virtio: using mergeable buffer Rx path on port %u",
1554 eth_dev->data->port_id);
1555 eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1557 PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
1558 eth_dev->data->port_id);
1559 eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1565 /* Only support 1:1 queue/interrupt mapping so far.
1566 * TODO: support n:1 queue/interrupt mapping when there are limited number of
1567 * interrupt vectors (<N+1).
1570 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1573 struct virtio_hw *hw = dev->data->dev_private;
1575 PMD_INIT_LOG(INFO, "queue/interrupt binding");
1576 for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1577 dev->intr_handle->intr_vec[i] = i + 1;
1578 if (VTPCI_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1579 VIRTIO_MSI_NO_VECTOR) {
1580 PMD_DRV_LOG(ERR, "failed to set queue vector");
1589 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1592 struct virtio_hw *hw = dev->data->dev_private;
1594 PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1595 for (i = 0; i < dev->data->nb_rx_queues; ++i)
1596 VTPCI_OPS(hw)->set_queue_irq(hw,
1597 hw->vqs[i * VTNET_CQ],
1598 VIRTIO_MSI_NO_VECTOR);
1602 virtio_configure_intr(struct rte_eth_dev *dev)
1604 struct virtio_hw *hw = dev->data->dev_private;
1606 if (!rte_intr_cap_multiple(dev->intr_handle)) {
1607 PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1611 if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1612 PMD_INIT_LOG(ERR, "Fail to create eventfd");
1616 if (!dev->intr_handle->intr_vec) {
1617 dev->intr_handle->intr_vec =
1618 rte_zmalloc("intr_vec",
1619 hw->max_queue_pairs * sizeof(int), 0);
1620 if (!dev->intr_handle->intr_vec) {
1621 PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1622 hw->max_queue_pairs);
1627 /* Re-register callback to update max_intr */
1628 rte_intr_callback_unregister(dev->intr_handle,
1629 virtio_interrupt_handler,
1631 rte_intr_callback_register(dev->intr_handle,
1632 virtio_interrupt_handler,
1635 /* DO NOT try to remove this! This function will enable msix, or QEMU
1636 * will encounter SIGSEGV when DRIVER_OK is sent.
1637 * And for legacy devices, this should be done before queue/vec binding
1638 * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1639 * (22) will be ignored.
1641 if (virtio_intr_enable(dev) < 0) {
1642 PMD_DRV_LOG(ERR, "interrupt enable failed");
1646 if (virtio_queues_bind_intr(dev) < 0) {
1647 PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1654 /* reset device and renegotiate features if needed */
1656 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1658 struct virtio_hw *hw = eth_dev->data->dev_private;
1659 struct virtio_net_config *config;
1660 struct virtio_net_config local_config;
1661 struct rte_pci_device *pci_dev = NULL;
1664 /* Reset the device although not necessary at startup */
1668 virtio_dev_free_mbufs(eth_dev);
1669 virtio_free_queues(hw);
1672 /* Tell the host we've noticed this device. */
1673 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1675 /* Tell the host we've known how to drive the device. */
1676 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1677 if (virtio_negotiate_features(hw, req_features) < 0)
1680 hw->weak_barriers = !vtpci_with_feature(hw, VIRTIO_F_ORDER_PLATFORM);
1682 if (!hw->virtio_user_dev)
1683 pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1685 /* If host does not support both status and MSI-X then disable LSC */
1686 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS) &&
1687 hw->use_msix != VIRTIO_MSIX_NONE)
1688 eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1690 eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1692 /* Setting up rx_header size for the device */
1693 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1694 vtpci_with_feature(hw, VIRTIO_F_VERSION_1) ||
1695 vtpci_with_feature(hw, VIRTIO_F_RING_PACKED))
1696 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1698 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1700 /* Copy the permanent MAC address to: virtio_hw */
1701 virtio_get_hwaddr(hw);
1702 rte_ether_addr_copy((struct rte_ether_addr *)hw->mac_addr,
1703 ð_dev->data->mac_addrs[0]);
1705 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1706 hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1707 hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1709 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1710 config = &local_config;
1712 vtpci_read_dev_config(hw,
1713 offsetof(struct virtio_net_config, mac),
1714 &config->mac, sizeof(config->mac));
1716 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1717 vtpci_read_dev_config(hw,
1718 offsetof(struct virtio_net_config, status),
1719 &config->status, sizeof(config->status));
1722 "VIRTIO_NET_F_STATUS is not supported");
1726 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1727 vtpci_read_dev_config(hw,
1728 offsetof(struct virtio_net_config, max_virtqueue_pairs),
1729 &config->max_virtqueue_pairs,
1730 sizeof(config->max_virtqueue_pairs));
1733 "VIRTIO_NET_F_MQ is not supported");
1734 config->max_virtqueue_pairs = 1;
1737 hw->max_queue_pairs = config->max_virtqueue_pairs;
1739 if (vtpci_with_feature(hw, VIRTIO_NET_F_MTU)) {
1740 vtpci_read_dev_config(hw,
1741 offsetof(struct virtio_net_config, mtu),
1743 sizeof(config->mtu));
1746 * MTU value has already been checked at negotiation
1747 * time, but check again in case it has changed since
1748 * then, which should not happen.
1750 if (config->mtu < RTE_ETHER_MIN_MTU) {
1751 PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1756 hw->max_mtu = config->mtu;
1757 /* Set initial MTU to maximum one supported by vhost */
1758 eth_dev->data->mtu = config->mtu;
1761 hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1762 VLAN_TAG_LEN - hw->vtnet_hdr_size;
1765 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1766 config->max_virtqueue_pairs);
1767 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1769 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1770 config->mac[0], config->mac[1],
1771 config->mac[2], config->mac[3],
1772 config->mac[4], config->mac[5]);
1774 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1775 hw->max_queue_pairs = 1;
1776 hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1777 VLAN_TAG_LEN - hw->vtnet_hdr_size;
1780 ret = virtio_alloc_queues(eth_dev);
1784 if (eth_dev->data->dev_conf.intr_conf.rxq) {
1785 if (virtio_configure_intr(eth_dev) < 0) {
1786 PMD_INIT_LOG(ERR, "failed to configure interrupt");
1787 virtio_free_queues(hw);
1792 vtpci_reinit_complete(hw);
1795 PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1796 eth_dev->data->port_id, pci_dev->id.vendor_id,
1797 pci_dev->id.device_id);
1803 * Remap the PCI device again (IO port map for legacy device and
1804 * memory map for modern device), so that the secondary process
1805 * could have the PCI initiated correctly.
1808 virtio_remap_pci(struct rte_pci_device *pci_dev, struct virtio_hw *hw)
1812 * We don't have to re-parse the PCI config space, since
1813 * rte_pci_map_device() makes sure the mapped address
1814 * in secondary process would equal to the one mapped in
1815 * the primary process: error will be returned if that
1816 * requirement is not met.
1818 * That said, we could simply reuse all cap pointers
1819 * (such as dev_cfg, common_cfg, etc.) parsed from the
1820 * primary process, which is stored in shared memory.
1822 if (rte_pci_map_device(pci_dev)) {
1823 PMD_INIT_LOG(DEBUG, "failed to map pci device!");
1827 if (rte_pci_ioport_map(pci_dev, 0, VTPCI_IO(hw)) < 0)
1835 virtio_set_vtpci_ops(struct virtio_hw *hw)
1837 #ifdef RTE_VIRTIO_USER
1838 if (hw->virtio_user_dev)
1839 VTPCI_OPS(hw) = &virtio_user_ops;
1843 VTPCI_OPS(hw) = &modern_ops;
1845 VTPCI_OPS(hw) = &legacy_ops;
1849 * This function is based on probe() function in virtio_pci.c
1850 * It returns 0 on success.
1853 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1855 struct virtio_hw *hw = eth_dev->data->dev_private;
1858 if (sizeof(struct virtio_net_hdr_mrg_rxbuf) > RTE_PKTMBUF_HEADROOM) {
1860 "Not sufficient headroom required = %d, avail = %d",
1861 (int)sizeof(struct virtio_net_hdr_mrg_rxbuf),
1862 RTE_PKTMBUF_HEADROOM);
1867 eth_dev->dev_ops = &virtio_eth_dev_ops;
1869 if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1870 if (!hw->virtio_user_dev) {
1871 ret = virtio_remap_pci(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1876 virtio_set_vtpci_ops(hw);
1877 set_rxtx_funcs(eth_dev);
1883 * Pass the information to the rte_eth_dev_close() that it should also
1884 * release the private port resources.
1886 eth_dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
1888 /* Allocate memory for storing MAC addresses */
1889 eth_dev->data->mac_addrs = rte_zmalloc("virtio",
1890 VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN, 0);
1891 if (eth_dev->data->mac_addrs == NULL) {
1893 "Failed to allocate %d bytes needed to store MAC addresses",
1894 VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN);
1898 hw->port_id = eth_dev->data->port_id;
1899 /* For virtio_user case the hw->virtio_user_dev is populated by
1900 * virtio_user_eth_dev_alloc() before eth_virtio_dev_init() is called.
1902 if (!hw->virtio_user_dev) {
1903 ret = vtpci_init(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1905 goto err_vtpci_init;
1908 /* reset device and negotiate default features */
1909 ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1911 goto err_virtio_init;
1918 if (!hw->virtio_user_dev) {
1919 rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(eth_dev));
1921 rte_pci_ioport_unmap(VTPCI_IO(hw));
1924 rte_free(eth_dev->data->mac_addrs);
1925 eth_dev->data->mac_addrs = NULL;
1930 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1932 PMD_INIT_FUNC_TRACE();
1934 if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1937 virtio_dev_stop(eth_dev);
1938 virtio_dev_close(eth_dev);
1940 eth_dev->dev_ops = NULL;
1941 eth_dev->tx_pkt_burst = NULL;
1942 eth_dev->rx_pkt_burst = NULL;
1944 PMD_INIT_LOG(DEBUG, "dev_uninit completed");
1949 static int vdpa_check_handler(__rte_unused const char *key,
1950 const char *value, __rte_unused void *opaque)
1952 if (strcmp(value, "1"))
1959 vdpa_mode_selected(struct rte_devargs *devargs)
1961 struct rte_kvargs *kvlist;
1962 const char *key = "vdpa";
1965 if (devargs == NULL)
1968 kvlist = rte_kvargs_parse(devargs->args, NULL);
1972 if (!rte_kvargs_count(kvlist, key))
1975 /* vdpa mode selected when there's a key-value pair: vdpa=1 */
1976 if (rte_kvargs_process(kvlist, key,
1977 vdpa_check_handler, NULL) < 0) {
1983 rte_kvargs_free(kvlist);
1987 static int eth_virtio_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1988 struct rte_pci_device *pci_dev)
1990 if (rte_eal_iopl_init() != 0) {
1991 PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1995 /* virtio pmd skips probe if device needs to work in vdpa mode */
1996 if (vdpa_mode_selected(pci_dev->device.devargs))
1999 return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct virtio_hw),
2000 eth_virtio_dev_init);
2003 static int eth_virtio_pci_remove(struct rte_pci_device *pci_dev)
2007 ret = rte_eth_dev_pci_generic_remove(pci_dev, eth_virtio_dev_uninit);
2008 /* Port has already been released by close. */
2014 static struct rte_pci_driver rte_virtio_pmd = {
2016 .name = "net_virtio",
2018 .id_table = pci_id_virtio_map,
2020 .probe = eth_virtio_pci_probe,
2021 .remove = eth_virtio_pci_remove,
2024 RTE_INIT(rte_virtio_pmd_init)
2026 rte_eal_iopl_init();
2027 rte_pci_register(&rte_virtio_pmd);
2031 rx_offload_enabled(struct virtio_hw *hw)
2033 return vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM) ||
2034 vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2035 vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6);
2039 tx_offload_enabled(struct virtio_hw *hw)
2041 return vtpci_with_feature(hw, VIRTIO_NET_F_CSUM) ||
2042 vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO4) ||
2043 vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO6);
2047 * Configure virtio device
2048 * It returns 0 on success.
2051 virtio_dev_configure(struct rte_eth_dev *dev)
2053 const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2054 const struct rte_eth_txmode *txmode = &dev->data->dev_conf.txmode;
2055 struct virtio_hw *hw = dev->data->dev_private;
2056 uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
2058 uint64_t rx_offloads = rxmode->offloads;
2059 uint64_t tx_offloads = txmode->offloads;
2060 uint64_t req_features;
2063 PMD_INIT_LOG(DEBUG, "configure");
2064 req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
2066 if (dev->data->dev_conf.intr_conf.rxq) {
2067 ret = virtio_init_device(dev, hw->req_guest_features);
2072 if (rxmode->max_rx_pkt_len > hw->max_mtu + ether_hdr_len)
2073 req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
2075 if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2076 DEV_RX_OFFLOAD_TCP_CKSUM))
2077 req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
2079 if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO)
2081 (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2082 (1ULL << VIRTIO_NET_F_GUEST_TSO6);
2084 if (tx_offloads & (DEV_TX_OFFLOAD_UDP_CKSUM |
2085 DEV_TX_OFFLOAD_TCP_CKSUM))
2086 req_features |= (1ULL << VIRTIO_NET_F_CSUM);
2088 if (tx_offloads & DEV_TX_OFFLOAD_TCP_TSO)
2090 (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2091 (1ULL << VIRTIO_NET_F_HOST_TSO6);
2093 /* if request features changed, reinit the device */
2094 if (req_features != hw->req_guest_features) {
2095 ret = virtio_init_device(dev, req_features);
2100 if ((rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2101 DEV_RX_OFFLOAD_TCP_CKSUM)) &&
2102 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
2104 "rx checksum not available on this host");
2108 if ((rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) &&
2109 (!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2110 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6))) {
2112 "Large Receive Offload not available on this host");
2116 /* start control queue */
2117 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
2118 virtio_dev_cq_start(dev);
2120 if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
2123 if ((rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
2124 && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2126 "vlan filtering not available on this host");
2130 hw->has_tx_offload = tx_offload_enabled(hw);
2131 hw->has_rx_offload = rx_offload_enabled(hw);
2133 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2134 /* Enable vector (0) for Link State Intrerrupt */
2135 if (VTPCI_OPS(hw)->set_config_irq(hw, 0) ==
2136 VIRTIO_MSI_NO_VECTOR) {
2137 PMD_DRV_LOG(ERR, "failed to set config vector");
2141 rte_spinlock_init(&hw->state_lock);
2143 hw->use_simple_rx = 1;
2145 if (vtpci_with_feature(hw, VIRTIO_F_IN_ORDER)) {
2146 hw->use_inorder_tx = 1;
2147 hw->use_inorder_rx = 1;
2148 hw->use_simple_rx = 0;
2151 if (vtpci_packed_queue(hw)) {
2152 hw->use_simple_rx = 0;
2153 hw->use_inorder_rx = 0;
2156 #if defined RTE_ARCH_ARM64 || defined RTE_ARCH_ARM
2157 if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
2158 hw->use_simple_rx = 0;
2161 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2162 hw->use_simple_rx = 0;
2165 if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2166 DEV_RX_OFFLOAD_TCP_CKSUM |
2167 DEV_RX_OFFLOAD_TCP_LRO |
2168 DEV_RX_OFFLOAD_VLAN_STRIP))
2169 hw->use_simple_rx = 0;
2176 virtio_dev_start(struct rte_eth_dev *dev)
2178 uint16_t nb_queues, i;
2179 struct virtnet_rx *rxvq;
2180 struct virtnet_tx *txvq __rte_unused;
2181 struct virtio_hw *hw = dev->data->dev_private;
2184 /* Finish the initialization of the queues */
2185 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2186 ret = virtio_dev_rx_queue_setup_finish(dev, i);
2190 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2191 ret = virtio_dev_tx_queue_setup_finish(dev, i);
2196 /* check if lsc interrupt feature is enabled */
2197 if (dev->data->dev_conf.intr_conf.lsc) {
2198 if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
2199 PMD_DRV_LOG(ERR, "link status not supported by host");
2204 /* Enable uio/vfio intr/eventfd mapping: althrough we already did that
2205 * in device configure, but it could be unmapped when device is
2208 if (dev->data->dev_conf.intr_conf.lsc ||
2209 dev->data->dev_conf.intr_conf.rxq) {
2210 virtio_intr_disable(dev);
2212 /* Setup interrupt callback */
2213 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2214 rte_intr_callback_register(dev->intr_handle,
2215 virtio_interrupt_handler,
2218 if (virtio_intr_enable(dev) < 0) {
2219 PMD_DRV_LOG(ERR, "interrupt enable failed");
2224 /*Notify the backend
2225 *Otherwise the tap backend might already stop its queue due to fullness.
2226 *vhost backend will have no chance to be waked up
2228 nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
2229 if (hw->max_queue_pairs > 1) {
2230 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
2234 PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
2236 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2237 rxvq = dev->data->rx_queues[i];
2238 /* Flush the old packets */
2239 virtqueue_rxvq_flush(rxvq->vq);
2240 virtqueue_notify(rxvq->vq);
2243 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2244 txvq = dev->data->tx_queues[i];
2245 virtqueue_notify(txvq->vq);
2248 PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
2250 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2251 rxvq = dev->data->rx_queues[i];
2252 VIRTQUEUE_DUMP(rxvq->vq);
2255 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2256 txvq = dev->data->tx_queues[i];
2257 VIRTQUEUE_DUMP(txvq->vq);
2260 set_rxtx_funcs(dev);
2263 /* Initialize Link state */
2264 virtio_dev_link_update(dev, 0);
2269 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
2271 struct virtio_hw *hw = dev->data->dev_private;
2272 uint16_t nr_vq = virtio_get_nr_vq(hw);
2273 const char *type __rte_unused;
2274 unsigned int i, mbuf_num = 0;
2275 struct virtqueue *vq;
2276 struct rte_mbuf *buf;
2279 if (hw->vqs == NULL)
2282 for (i = 0; i < nr_vq; i++) {
2287 queue_type = virtio_get_queue_type(hw, i);
2288 if (queue_type == VTNET_RQ)
2290 else if (queue_type == VTNET_TQ)
2296 "Before freeing %s[%d] used and unused buf",
2300 while ((buf = virtqueue_detach_unused(vq)) != NULL) {
2301 rte_pktmbuf_free(buf);
2306 "After freeing %s[%d] used and unused buf",
2311 PMD_INIT_LOG(DEBUG, "%d mbufs freed", mbuf_num);
2315 * Stop device: disable interrupt and mark link down
2318 virtio_dev_stop(struct rte_eth_dev *dev)
2320 struct virtio_hw *hw = dev->data->dev_private;
2321 struct rte_eth_link link;
2322 struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
2324 PMD_INIT_LOG(DEBUG, "stop");
2326 rte_spinlock_lock(&hw->state_lock);
2329 hw->started = false;
2331 if (intr_conf->lsc || intr_conf->rxq) {
2332 virtio_intr_disable(dev);
2334 /* Reset interrupt callback */
2335 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
2336 rte_intr_callback_unregister(dev->intr_handle,
2337 virtio_interrupt_handler,
2342 memset(&link, 0, sizeof(link));
2343 rte_eth_linkstatus_set(dev, &link);
2345 rte_spinlock_unlock(&hw->state_lock);
2349 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
2351 struct rte_eth_link link;
2353 struct virtio_hw *hw = dev->data->dev_private;
2355 memset(&link, 0, sizeof(link));
2356 link.link_duplex = ETH_LINK_FULL_DUPLEX;
2357 link.link_speed = ETH_SPEED_NUM_10G;
2358 link.link_autoneg = ETH_LINK_FIXED;
2361 link.link_status = ETH_LINK_DOWN;
2362 } else if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
2363 PMD_INIT_LOG(DEBUG, "Get link status from hw");
2364 vtpci_read_dev_config(hw,
2365 offsetof(struct virtio_net_config, status),
2366 &status, sizeof(status));
2367 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
2368 link.link_status = ETH_LINK_DOWN;
2369 PMD_INIT_LOG(DEBUG, "Port %d is down",
2370 dev->data->port_id);
2372 link.link_status = ETH_LINK_UP;
2373 PMD_INIT_LOG(DEBUG, "Port %d is up",
2374 dev->data->port_id);
2377 link.link_status = ETH_LINK_UP;
2380 return rte_eth_linkstatus_set(dev, &link);
2384 virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
2386 const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2387 struct virtio_hw *hw = dev->data->dev_private;
2388 uint64_t offloads = rxmode->offloads;
2390 if (mask & ETH_VLAN_FILTER_MASK) {
2391 if ((offloads & DEV_RX_OFFLOAD_VLAN_FILTER) &&
2392 !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2395 "vlan filtering not available on this host");
2401 if (mask & ETH_VLAN_STRIP_MASK)
2402 hw->vlan_strip = !!(offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
2408 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2410 uint64_t tso_mask, host_features;
2411 struct virtio_hw *hw = dev->data->dev_private;
2413 dev_info->speed_capa = ETH_LINK_SPEED_10G; /* fake value */
2415 dev_info->max_rx_queues =
2416 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
2417 dev_info->max_tx_queues =
2418 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
2419 dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
2420 dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
2421 dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
2423 host_features = VTPCI_OPS(hw)->get_features(hw);
2424 dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
2425 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_JUMBO_FRAME;
2426 if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
2427 dev_info->rx_offload_capa |=
2428 DEV_RX_OFFLOAD_TCP_CKSUM |
2429 DEV_RX_OFFLOAD_UDP_CKSUM;
2431 if (host_features & (1ULL << VIRTIO_NET_F_CTRL_VLAN))
2432 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_VLAN_FILTER;
2433 tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2434 (1ULL << VIRTIO_NET_F_GUEST_TSO6);
2435 if ((host_features & tso_mask) == tso_mask)
2436 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
2438 dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
2439 DEV_TX_OFFLOAD_VLAN_INSERT;
2440 if (host_features & (1ULL << VIRTIO_NET_F_CSUM)) {
2441 dev_info->tx_offload_capa |=
2442 DEV_TX_OFFLOAD_UDP_CKSUM |
2443 DEV_TX_OFFLOAD_TCP_CKSUM;
2445 tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2446 (1ULL << VIRTIO_NET_F_HOST_TSO6);
2447 if ((host_features & tso_mask) == tso_mask)
2448 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
2454 * It enables testpmd to collect per queue stats.
2457 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
2458 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
2459 __rte_unused uint8_t is_rx)
2464 RTE_PMD_EXPORT_NAME(net_virtio, __COUNTER__);
2465 RTE_PMD_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map);
2466 RTE_PMD_REGISTER_KMOD_DEP(net_virtio, "* igb_uio | uio_pci_generic | vfio-pci");
2468 RTE_INIT(virtio_init_log)
2470 virtio_logtype_init = rte_log_register("pmd.net.virtio.init");
2471 if (virtio_logtype_init >= 0)
2472 rte_log_set_level(virtio_logtype_init, RTE_LOG_NOTICE);
2473 virtio_logtype_driver = rte_log_register("pmd.net.virtio.driver");
2474 if (virtio_logtype_driver >= 0)
2475 rte_log_set_level(virtio_logtype_driver, RTE_LOG_NOTICE);