1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2016 Intel Corporation
11 #include <ethdev_driver.h>
12 #include <rte_memcpy.h>
13 #include <rte_string_fns.h>
14 #include <rte_memzone.h>
15 #include <rte_malloc.h>
16 #include <rte_branch_prediction.h>
17 #include <rte_ether.h>
20 #include <rte_common.h>
21 #include <rte_errno.h>
22 #include <rte_cpuflags.h>
24 #include <rte_memory.h>
25 #include <rte_eal_paging.h>
28 #include <rte_cycles.h>
29 #include <rte_kvargs.h>
31 #include "virtio_ethdev.h"
33 #include "virtio_logs.h"
34 #include "virtqueue.h"
35 #include "virtio_rxtx.h"
36 #include "virtio_rxtx_simple.h"
37 #include "virtio_user/virtio_user_dev.h"
39 static int virtio_dev_configure(struct rte_eth_dev *dev);
40 static int virtio_dev_start(struct rte_eth_dev *dev);
41 static int virtio_dev_promiscuous_enable(struct rte_eth_dev *dev);
42 static int virtio_dev_promiscuous_disable(struct rte_eth_dev *dev);
43 static int virtio_dev_allmulticast_enable(struct rte_eth_dev *dev);
44 static int virtio_dev_allmulticast_disable(struct rte_eth_dev *dev);
45 static uint32_t virtio_dev_speed_capa_get(uint32_t speed);
46 static int virtio_dev_devargs_parse(struct rte_devargs *devargs,
49 static int virtio_dev_info_get(struct rte_eth_dev *dev,
50 struct rte_eth_dev_info *dev_info);
51 static int virtio_dev_link_update(struct rte_eth_dev *dev,
52 int wait_to_complete);
53 static int virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask);
55 static void virtio_set_hwaddr(struct virtio_hw *hw);
56 static void virtio_get_hwaddr(struct virtio_hw *hw);
58 static int virtio_dev_stats_get(struct rte_eth_dev *dev,
59 struct rte_eth_stats *stats);
60 static int virtio_dev_xstats_get(struct rte_eth_dev *dev,
61 struct rte_eth_xstat *xstats, unsigned n);
62 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
63 struct rte_eth_xstat_name *xstats_names,
65 static int virtio_dev_stats_reset(struct rte_eth_dev *dev);
66 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev);
67 static int virtio_vlan_filter_set(struct rte_eth_dev *dev,
68 uint16_t vlan_id, int on);
69 static int virtio_mac_addr_add(struct rte_eth_dev *dev,
70 struct rte_ether_addr *mac_addr,
71 uint32_t index, uint32_t vmdq);
72 static void virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index);
73 static int virtio_mac_addr_set(struct rte_eth_dev *dev,
74 struct rte_ether_addr *mac_addr);
76 static int virtio_intr_disable(struct rte_eth_dev *dev);
78 static int virtio_dev_queue_stats_mapping_set(
79 struct rte_eth_dev *eth_dev,
84 static void virtio_notify_peers(struct rte_eth_dev *dev);
85 static void virtio_ack_link_announce(struct rte_eth_dev *dev);
87 struct rte_virtio_xstats_name_off {
88 char name[RTE_ETH_XSTATS_NAME_SIZE];
92 /* [rt]x_qX_ is prepended to the name string here */
93 static const struct rte_virtio_xstats_name_off rte_virtio_rxq_stat_strings[] = {
94 {"good_packets", offsetof(struct virtnet_rx, stats.packets)},
95 {"good_bytes", offsetof(struct virtnet_rx, stats.bytes)},
96 {"errors", offsetof(struct virtnet_rx, stats.errors)},
97 {"multicast_packets", offsetof(struct virtnet_rx, stats.multicast)},
98 {"broadcast_packets", offsetof(struct virtnet_rx, stats.broadcast)},
99 {"undersize_packets", offsetof(struct virtnet_rx, stats.size_bins[0])},
100 {"size_64_packets", offsetof(struct virtnet_rx, stats.size_bins[1])},
101 {"size_65_127_packets", offsetof(struct virtnet_rx, stats.size_bins[2])},
102 {"size_128_255_packets", offsetof(struct virtnet_rx, stats.size_bins[3])},
103 {"size_256_511_packets", offsetof(struct virtnet_rx, stats.size_bins[4])},
104 {"size_512_1023_packets", offsetof(struct virtnet_rx, stats.size_bins[5])},
105 {"size_1024_1518_packets", offsetof(struct virtnet_rx, stats.size_bins[6])},
106 {"size_1519_max_packets", offsetof(struct virtnet_rx, stats.size_bins[7])},
109 /* [rt]x_qX_ is prepended to the name string here */
110 static const struct rte_virtio_xstats_name_off rte_virtio_txq_stat_strings[] = {
111 {"good_packets", offsetof(struct virtnet_tx, stats.packets)},
112 {"good_bytes", offsetof(struct virtnet_tx, stats.bytes)},
113 {"multicast_packets", offsetof(struct virtnet_tx, stats.multicast)},
114 {"broadcast_packets", offsetof(struct virtnet_tx, stats.broadcast)},
115 {"undersize_packets", offsetof(struct virtnet_tx, stats.size_bins[0])},
116 {"size_64_packets", offsetof(struct virtnet_tx, stats.size_bins[1])},
117 {"size_65_127_packets", offsetof(struct virtnet_tx, stats.size_bins[2])},
118 {"size_128_255_packets", offsetof(struct virtnet_tx, stats.size_bins[3])},
119 {"size_256_511_packets", offsetof(struct virtnet_tx, stats.size_bins[4])},
120 {"size_512_1023_packets", offsetof(struct virtnet_tx, stats.size_bins[5])},
121 {"size_1024_1518_packets", offsetof(struct virtnet_tx, stats.size_bins[6])},
122 {"size_1519_max_packets", offsetof(struct virtnet_tx, stats.size_bins[7])},
125 #define VIRTIO_NB_RXQ_XSTATS (sizeof(rte_virtio_rxq_stat_strings) / \
126 sizeof(rte_virtio_rxq_stat_strings[0]))
127 #define VIRTIO_NB_TXQ_XSTATS (sizeof(rte_virtio_txq_stat_strings) / \
128 sizeof(rte_virtio_txq_stat_strings[0]))
130 struct virtio_hw_internal virtio_hw_internal[RTE_MAX_ETHPORTS];
132 static struct virtio_pmd_ctrl *
133 virtio_send_command_packed(struct virtnet_ctl *cvq,
134 struct virtio_pmd_ctrl *ctrl,
135 int *dlen, int pkt_num)
137 struct virtqueue *vq = virtnet_cq_to_vq(cvq);
139 struct vring_packed_desc *desc = vq->vq_packed.ring.desc;
140 struct virtio_pmd_ctrl *result;
147 * Format is enforced in qemu code:
148 * One TX packet for header;
149 * At least one TX packet per argument;
150 * One RX packet for ACK.
152 head = vq->vq_avail_idx;
153 flags = vq->vq_packed.cached_flags;
154 desc[head].addr = cvq->virtio_net_hdr_mem;
155 desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
158 if (++vq->vq_avail_idx >= vq->vq_nentries) {
159 vq->vq_avail_idx -= vq->vq_nentries;
160 vq->vq_packed.cached_flags ^= VRING_PACKED_DESC_F_AVAIL_USED;
163 for (k = 0; k < pkt_num; k++) {
164 desc[vq->vq_avail_idx].addr = cvq->virtio_net_hdr_mem
165 + sizeof(struct virtio_net_ctrl_hdr)
166 + sizeof(ctrl->status) + sizeof(uint8_t) * sum;
167 desc[vq->vq_avail_idx].len = dlen[k];
168 desc[vq->vq_avail_idx].flags = VRING_DESC_F_NEXT |
169 vq->vq_packed.cached_flags;
173 if (++vq->vq_avail_idx >= vq->vq_nentries) {
174 vq->vq_avail_idx -= vq->vq_nentries;
175 vq->vq_packed.cached_flags ^=
176 VRING_PACKED_DESC_F_AVAIL_USED;
180 desc[vq->vq_avail_idx].addr = cvq->virtio_net_hdr_mem
181 + sizeof(struct virtio_net_ctrl_hdr);
182 desc[vq->vq_avail_idx].len = sizeof(ctrl->status);
183 desc[vq->vq_avail_idx].flags = VRING_DESC_F_WRITE |
184 vq->vq_packed.cached_flags;
187 if (++vq->vq_avail_idx >= vq->vq_nentries) {
188 vq->vq_avail_idx -= vq->vq_nentries;
189 vq->vq_packed.cached_flags ^= VRING_PACKED_DESC_F_AVAIL_USED;
192 virtqueue_store_flags_packed(&desc[head], VRING_DESC_F_NEXT | flags,
193 vq->hw->weak_barriers);
195 virtio_wmb(vq->hw->weak_barriers);
196 virtqueue_notify(vq);
198 /* wait for used desc in virtqueue
199 * desc_is_used has a load-acquire or rte_io_rmb inside
201 while (!desc_is_used(&desc[head], vq))
204 /* now get used descriptors */
205 vq->vq_free_cnt += nb_descs;
206 vq->vq_used_cons_idx += nb_descs;
207 if (vq->vq_used_cons_idx >= vq->vq_nentries) {
208 vq->vq_used_cons_idx -= vq->vq_nentries;
209 vq->vq_packed.used_wrap_counter ^= 1;
212 PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\n"
213 "vq->vq_avail_idx=%d\n"
214 "vq->vq_used_cons_idx=%d\n"
215 "vq->vq_packed.cached_flags=0x%x\n"
216 "vq->vq_packed.used_wrap_counter=%d",
219 vq->vq_used_cons_idx,
220 vq->vq_packed.cached_flags,
221 vq->vq_packed.used_wrap_counter);
223 result = cvq->virtio_net_hdr_mz->addr;
227 static struct virtio_pmd_ctrl *
228 virtio_send_command_split(struct virtnet_ctl *cvq,
229 struct virtio_pmd_ctrl *ctrl,
230 int *dlen, int pkt_num)
232 struct virtio_pmd_ctrl *result;
233 struct virtqueue *vq = virtnet_cq_to_vq(cvq);
237 head = vq->vq_desc_head_idx;
240 * Format is enforced in qemu code:
241 * One TX packet for header;
242 * At least one TX packet per argument;
243 * One RX packet for ACK.
245 vq->vq_split.ring.desc[head].flags = VRING_DESC_F_NEXT;
246 vq->vq_split.ring.desc[head].addr = cvq->virtio_net_hdr_mem;
247 vq->vq_split.ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
249 i = vq->vq_split.ring.desc[head].next;
251 for (k = 0; k < pkt_num; k++) {
252 vq->vq_split.ring.desc[i].flags = VRING_DESC_F_NEXT;
253 vq->vq_split.ring.desc[i].addr = cvq->virtio_net_hdr_mem
254 + sizeof(struct virtio_net_ctrl_hdr)
255 + sizeof(ctrl->status) + sizeof(uint8_t)*sum;
256 vq->vq_split.ring.desc[i].len = dlen[k];
259 i = vq->vq_split.ring.desc[i].next;
262 vq->vq_split.ring.desc[i].flags = VRING_DESC_F_WRITE;
263 vq->vq_split.ring.desc[i].addr = cvq->virtio_net_hdr_mem
264 + sizeof(struct virtio_net_ctrl_hdr);
265 vq->vq_split.ring.desc[i].len = sizeof(ctrl->status);
268 vq->vq_desc_head_idx = vq->vq_split.ring.desc[i].next;
270 vq_update_avail_ring(vq, head);
271 vq_update_avail_idx(vq);
273 PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
275 virtqueue_notify(vq);
277 while (virtqueue_nused(vq) == 0)
280 while (virtqueue_nused(vq)) {
281 uint32_t idx, desc_idx, used_idx;
282 struct vring_used_elem *uep;
284 used_idx = (uint32_t)(vq->vq_used_cons_idx
285 & (vq->vq_nentries - 1));
286 uep = &vq->vq_split.ring.used->ring[used_idx];
287 idx = (uint32_t) uep->id;
290 while (vq->vq_split.ring.desc[desc_idx].flags &
292 desc_idx = vq->vq_split.ring.desc[desc_idx].next;
296 vq->vq_split.ring.desc[desc_idx].next = vq->vq_desc_head_idx;
297 vq->vq_desc_head_idx = idx;
299 vq->vq_used_cons_idx++;
303 PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
304 vq->vq_free_cnt, vq->vq_desc_head_idx);
306 result = cvq->virtio_net_hdr_mz->addr;
311 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
312 int *dlen, int pkt_num)
314 virtio_net_ctrl_ack status = ~0;
315 struct virtio_pmd_ctrl *result;
316 struct virtqueue *vq;
318 ctrl->status = status;
321 PMD_INIT_LOG(ERR, "Control queue is not supported.");
325 rte_spinlock_lock(&cvq->lock);
326 vq = virtnet_cq_to_vq(cvq);
328 PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
329 "vq->hw->cvq = %p vq = %p",
330 vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
332 if (vq->vq_free_cnt < pkt_num + 2 || pkt_num < 1) {
333 rte_spinlock_unlock(&cvq->lock);
337 memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
338 sizeof(struct virtio_pmd_ctrl));
340 if (virtio_with_packed_queue(vq->hw))
341 result = virtio_send_command_packed(cvq, ctrl, dlen, pkt_num);
343 result = virtio_send_command_split(cvq, ctrl, dlen, pkt_num);
345 rte_spinlock_unlock(&cvq->lock);
346 return result->status;
350 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
352 struct virtio_hw *hw = dev->data->dev_private;
353 struct virtio_pmd_ctrl ctrl;
357 ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
358 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
359 memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
361 dlen[0] = sizeof(uint16_t);
363 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
365 PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
366 "failed, this is too late now...");
374 virtio_get_nr_vq(struct virtio_hw *hw)
376 uint16_t nr_vq = hw->max_queue_pairs * 2;
378 if (virtio_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
385 virtio_init_vring(struct virtqueue *vq)
387 int size = vq->vq_nentries;
388 uint8_t *ring_mem = vq->vq_ring_virt_mem;
390 PMD_INIT_FUNC_TRACE();
392 memset(ring_mem, 0, vq->vq_ring_size);
394 vq->vq_used_cons_idx = 0;
395 vq->vq_desc_head_idx = 0;
396 vq->vq_avail_idx = 0;
397 vq->vq_desc_tail_idx = (uint16_t)(vq->vq_nentries - 1);
398 vq->vq_free_cnt = vq->vq_nentries;
399 memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries);
400 if (virtio_with_packed_queue(vq->hw)) {
401 vring_init_packed(&vq->vq_packed.ring, ring_mem,
402 VIRTIO_VRING_ALIGN, size);
403 vring_desc_init_packed(vq, size);
405 struct vring *vr = &vq->vq_split.ring;
407 vring_init_split(vr, ring_mem, VIRTIO_VRING_ALIGN, size);
408 vring_desc_init_split(vr->desc, size);
411 * Disable device(host) interrupting guest
413 virtqueue_disable_intr(vq);
417 virtio_init_queue(struct rte_eth_dev *dev, uint16_t queue_idx)
419 char vq_name[VIRTQUEUE_MAX_NAME_SZ];
420 char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
421 const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
422 unsigned int vq_size, size;
423 struct virtio_hw *hw = dev->data->dev_private;
424 struct virtnet_rx *rxvq = NULL;
425 struct virtnet_tx *txvq = NULL;
426 struct virtnet_ctl *cvq = NULL;
427 struct virtqueue *vq;
428 size_t sz_hdr_mz = 0;
429 void *sw_ring = NULL;
430 int queue_type = virtio_get_queue_type(hw, queue_idx);
432 int numa_node = dev->device->numa_node;
433 struct rte_mbuf *fake_mbuf = NULL;
435 PMD_INIT_LOG(INFO, "setting up queue: %u on NUMA node %d",
436 queue_idx, numa_node);
439 * Read the virtqueue size from the Queue Size field
440 * Always power of 2 and if 0 virtqueue does not exist
442 vq_size = VIRTIO_OPS(hw)->get_queue_num(hw, queue_idx);
443 PMD_INIT_LOG(DEBUG, "vq_size: %u", vq_size);
445 PMD_INIT_LOG(ERR, "virtqueue does not exist");
449 if (!virtio_with_packed_queue(hw) && !rte_is_power_of_2(vq_size)) {
450 PMD_INIT_LOG(ERR, "split virtqueue size is not power of 2");
454 snprintf(vq_name, sizeof(vq_name), "port%d_vq%d",
455 dev->data->port_id, queue_idx);
457 size = RTE_ALIGN_CEIL(sizeof(*vq) +
458 vq_size * sizeof(struct vq_desc_extra),
459 RTE_CACHE_LINE_SIZE);
460 if (queue_type == VTNET_TQ) {
462 * For each xmit packet, allocate a virtio_net_hdr
463 * and indirect ring elements
465 sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
466 } else if (queue_type == VTNET_CQ) {
467 /* Allocate a page for control vq command, data and status */
468 sz_hdr_mz = rte_mem_page_size();
471 vq = rte_zmalloc_socket(vq_name, size, RTE_CACHE_LINE_SIZE,
474 PMD_INIT_LOG(ERR, "can not allocate vq");
477 hw->vqs[queue_idx] = vq;
480 vq->vq_queue_index = queue_idx;
481 vq->vq_nentries = vq_size;
482 if (virtio_with_packed_queue(hw)) {
483 vq->vq_packed.used_wrap_counter = 1;
484 vq->vq_packed.cached_flags = VRING_PACKED_DESC_F_AVAIL;
485 vq->vq_packed.event_flags_shadow = 0;
486 if (queue_type == VTNET_RQ)
487 vq->vq_packed.cached_flags |= VRING_DESC_F_WRITE;
491 * Reserve a memzone for vring elements
493 size = vring_size(hw, vq_size, VIRTIO_VRING_ALIGN);
494 vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_VRING_ALIGN);
495 PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
496 size, vq->vq_ring_size);
498 mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
499 numa_node, RTE_MEMZONE_IOVA_CONTIG,
502 if (rte_errno == EEXIST)
503 mz = rte_memzone_lookup(vq_name);
510 memset(mz->addr, 0, mz->len);
513 vq->vq_ring_mem = (uintptr_t)mz->addr;
515 vq->vq_ring_mem = mz->iova;
517 vq->vq_ring_virt_mem = mz->addr;
518 PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem: 0x%" PRIx64, vq->vq_ring_mem);
519 PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: %p", vq->vq_ring_virt_mem);
521 virtio_init_vring(vq);
524 snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_vq%d_hdr",
525 dev->data->port_id, queue_idx);
526 hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
527 numa_node, RTE_MEMZONE_IOVA_CONTIG,
528 RTE_CACHE_LINE_SIZE);
529 if (hdr_mz == NULL) {
530 if (rte_errno == EEXIST)
531 hdr_mz = rte_memzone_lookup(vq_hdr_name);
532 if (hdr_mz == NULL) {
539 if (queue_type == VTNET_RQ) {
540 size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
541 sizeof(vq->sw_ring[0]);
543 sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
544 RTE_CACHE_LINE_SIZE, numa_node);
546 PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
551 fake_mbuf = rte_zmalloc_socket("sw_ring", sizeof(*fake_mbuf),
552 RTE_CACHE_LINE_SIZE, numa_node);
554 PMD_INIT_LOG(ERR, "can not allocate fake mbuf");
559 vq->sw_ring = sw_ring;
561 rxvq->port_id = dev->data->port_id;
563 rxvq->fake_mbuf = fake_mbuf;
564 } else if (queue_type == VTNET_TQ) {
566 txvq->port_id = dev->data->port_id;
568 txvq->virtio_net_hdr_mz = hdr_mz;
570 txvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
572 txvq->virtio_net_hdr_mem = hdr_mz->iova;
573 } else if (queue_type == VTNET_CQ) {
576 cvq->virtio_net_hdr_mz = hdr_mz;
578 cvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
580 cvq->virtio_net_hdr_mem = hdr_mz->iova;
581 memset(cvq->virtio_net_hdr_mz->addr, 0, rte_mem_page_size());
587 vq->mbuf_addr_offset = offsetof(struct rte_mbuf, buf_addr);
589 vq->mbuf_addr_offset = offsetof(struct rte_mbuf, buf_iova);
591 if (queue_type == VTNET_TQ) {
592 struct virtio_tx_region *txr;
596 memset(txr, 0, vq_size * sizeof(*txr));
597 for (i = 0; i < vq_size; i++) {
598 /* first indirect descriptor is always the tx header */
599 if (!virtio_with_packed_queue(hw)) {
600 struct vring_desc *start_dp = txr[i].tx_indir;
601 vring_desc_init_split(start_dp,
602 RTE_DIM(txr[i].tx_indir));
603 start_dp->addr = txvq->virtio_net_hdr_mem
605 + offsetof(struct virtio_tx_region,
607 start_dp->len = hw->vtnet_hdr_size;
608 start_dp->flags = VRING_DESC_F_NEXT;
610 struct vring_packed_desc *start_dp =
611 txr[i].tx_packed_indir;
612 vring_desc_init_indirect_packed(start_dp,
613 RTE_DIM(txr[i].tx_packed_indir));
614 start_dp->addr = txvq->virtio_net_hdr_mem
616 + offsetof(struct virtio_tx_region,
618 start_dp->len = hw->vtnet_hdr_size;
623 if (VIRTIO_OPS(hw)->setup_queue(hw, vq) < 0) {
624 PMD_INIT_LOG(ERR, "setup_queue failed");
637 rte_memzone_free(hdr_mz);
639 rte_memzone_free(mz);
642 hw->vqs[queue_idx] = NULL;
648 virtio_free_queues(struct virtio_hw *hw)
650 uint16_t nr_vq = virtio_get_nr_vq(hw);
651 struct virtqueue *vq;
658 for (i = 0; i < nr_vq; i++) {
663 queue_type = virtio_get_queue_type(hw, i);
664 if (queue_type == VTNET_RQ) {
665 rte_free(vq->rxq.fake_mbuf);
666 rte_free(vq->sw_ring);
667 rte_memzone_free(vq->rxq.mz);
668 } else if (queue_type == VTNET_TQ) {
669 rte_memzone_free(vq->txq.mz);
670 rte_memzone_free(vq->txq.virtio_net_hdr_mz);
672 rte_memzone_free(vq->cq.mz);
673 rte_memzone_free(vq->cq.virtio_net_hdr_mz);
685 virtio_alloc_queues(struct rte_eth_dev *dev)
687 struct virtio_hw *hw = dev->data->dev_private;
688 uint16_t nr_vq = virtio_get_nr_vq(hw);
692 hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
694 PMD_INIT_LOG(ERR, "failed to allocate vqs");
698 for (i = 0; i < nr_vq; i++) {
699 ret = virtio_init_queue(dev, i);
701 virtio_free_queues(hw);
709 static void virtio_queues_unbind_intr(struct rte_eth_dev *dev);
712 virtio_dev_close(struct rte_eth_dev *dev)
714 struct virtio_hw *hw = dev->data->dev_private;
715 struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
717 PMD_INIT_LOG(DEBUG, "virtio_dev_close");
718 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
726 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
727 VIRTIO_OPS(hw)->set_config_irq(hw, VIRTIO_MSI_NO_VECTOR);
729 virtio_queues_unbind_intr(dev);
731 if (intr_conf->lsc || intr_conf->rxq) {
732 virtio_intr_disable(dev);
733 rte_intr_efd_disable(dev->intr_handle);
734 rte_free(dev->intr_handle->intr_vec);
735 dev->intr_handle->intr_vec = NULL;
739 virtio_dev_free_mbufs(dev);
740 virtio_free_queues(hw);
742 return VIRTIO_OPS(hw)->dev_close(hw);
746 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
748 struct virtio_hw *hw = dev->data->dev_private;
749 struct virtio_pmd_ctrl ctrl;
753 if (!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
754 PMD_INIT_LOG(INFO, "host does not support rx control");
758 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
759 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
763 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
765 PMD_INIT_LOG(ERR, "Failed to enable promisc");
773 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
775 struct virtio_hw *hw = dev->data->dev_private;
776 struct virtio_pmd_ctrl ctrl;
780 if (!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
781 PMD_INIT_LOG(INFO, "host does not support rx control");
785 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
786 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
790 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
792 PMD_INIT_LOG(ERR, "Failed to disable promisc");
800 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
802 struct virtio_hw *hw = dev->data->dev_private;
803 struct virtio_pmd_ctrl ctrl;
807 if (!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
808 PMD_INIT_LOG(INFO, "host does not support rx control");
812 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
813 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
817 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
819 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 (!virtio_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");
854 virtio_rx_mem_pool_buf_size(struct rte_mempool *mp)
856 return rte_pktmbuf_data_room_size(mp) - RTE_PKTMBUF_HEADROOM;
860 virtio_rx_check_scatter(uint16_t max_rx_pkt_len, uint16_t rx_buf_size,
861 bool rx_scatter_enabled, const char **error)
863 if (!rx_scatter_enabled && max_rx_pkt_len > rx_buf_size) {
864 *error = "Rx scatter is disabled and RxQ mbuf pool object size is too small";
872 virtio_check_scatter_on_all_rx_queues(struct rte_eth_dev *dev,
875 struct virtio_hw *hw = dev->data->dev_private;
876 struct virtnet_rx *rxvq;
877 struct virtqueue *vq;
885 for (qidx = 0; (vq = hw->vqs[2 * qidx + VTNET_SQ_RQ_QUEUE_IDX]) != NULL;
888 if (rxvq->mpool == NULL)
890 buf_size = virtio_rx_mem_pool_buf_size(rxvq->mpool);
892 if (!virtio_rx_check_scatter(frame_size, buf_size,
893 hw->rx_ol_scatter, &error)) {
894 PMD_INIT_LOG(ERR, "MTU check for RxQ %u failed: %s",
903 #define VLAN_TAG_LEN 4 /* 802.3ac tag (not DMA'd) */
905 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
907 struct virtio_hw *hw = dev->data->dev_private;
908 uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
910 uint32_t frame_size = mtu + ether_hdr_len;
911 uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
913 max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
915 if (mtu < RTE_ETHER_MIN_MTU || frame_size > max_frame_size) {
916 PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
917 RTE_ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
921 if (!virtio_check_scatter_on_all_rx_queues(dev, frame_size)) {
922 PMD_INIT_LOG(ERR, "MTU vs Rx scatter and Rx buffers check failed");
926 hw->max_rx_pkt_len = frame_size;
932 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
934 struct virtio_hw *hw = dev->data->dev_private;
935 struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
936 struct virtqueue *vq = virtnet_rxq_to_vq(rxvq);
938 virtqueue_enable_intr(vq);
939 virtio_mb(hw->weak_barriers);
944 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
946 struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
947 struct virtqueue *vq = virtnet_rxq_to_vq(rxvq);
949 virtqueue_disable_intr(vq);
954 * dev_ops for virtio, bare necessities for basic operation
956 static const struct eth_dev_ops virtio_eth_dev_ops = {
957 .dev_configure = virtio_dev_configure,
958 .dev_start = virtio_dev_start,
959 .dev_stop = virtio_dev_stop,
960 .dev_close = virtio_dev_close,
961 .promiscuous_enable = virtio_dev_promiscuous_enable,
962 .promiscuous_disable = virtio_dev_promiscuous_disable,
963 .allmulticast_enable = virtio_dev_allmulticast_enable,
964 .allmulticast_disable = virtio_dev_allmulticast_disable,
965 .mtu_set = virtio_mtu_set,
966 .dev_infos_get = virtio_dev_info_get,
967 .stats_get = virtio_dev_stats_get,
968 .xstats_get = virtio_dev_xstats_get,
969 .xstats_get_names = virtio_dev_xstats_get_names,
970 .stats_reset = virtio_dev_stats_reset,
971 .xstats_reset = virtio_dev_stats_reset,
972 .link_update = virtio_dev_link_update,
973 .vlan_offload_set = virtio_dev_vlan_offload_set,
974 .rx_queue_setup = virtio_dev_rx_queue_setup,
975 .rx_queue_intr_enable = virtio_dev_rx_queue_intr_enable,
976 .rx_queue_intr_disable = virtio_dev_rx_queue_intr_disable,
977 .tx_queue_setup = virtio_dev_tx_queue_setup,
978 /* collect stats per queue */
979 .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
980 .vlan_filter_set = virtio_vlan_filter_set,
981 .mac_addr_add = virtio_mac_addr_add,
982 .mac_addr_remove = virtio_mac_addr_remove,
983 .mac_addr_set = virtio_mac_addr_set,
987 * dev_ops for virtio-user in secondary processes, as we just have
988 * some limited supports currently.
990 const struct eth_dev_ops virtio_user_secondary_eth_dev_ops = {
991 .dev_infos_get = virtio_dev_info_get,
992 .stats_get = virtio_dev_stats_get,
993 .xstats_get = virtio_dev_xstats_get,
994 .xstats_get_names = virtio_dev_xstats_get_names,
995 .stats_reset = virtio_dev_stats_reset,
996 .xstats_reset = virtio_dev_stats_reset,
997 /* collect stats per queue */
998 .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
1002 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1006 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1007 const struct virtnet_tx *txvq = dev->data->tx_queues[i];
1011 stats->opackets += txvq->stats.packets;
1012 stats->obytes += txvq->stats.bytes;
1014 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
1015 stats->q_opackets[i] = txvq->stats.packets;
1016 stats->q_obytes[i] = txvq->stats.bytes;
1020 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1021 const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1025 stats->ipackets += rxvq->stats.packets;
1026 stats->ibytes += rxvq->stats.bytes;
1027 stats->ierrors += rxvq->stats.errors;
1029 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
1030 stats->q_ipackets[i] = rxvq->stats.packets;
1031 stats->q_ibytes[i] = rxvq->stats.bytes;
1035 stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
1038 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
1039 struct rte_eth_xstat_name *xstats_names,
1040 __rte_unused unsigned limit)
1046 unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
1047 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
1049 if (xstats_names != NULL) {
1050 /* Note: limit checked in rte_eth_xstats_names() */
1052 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1053 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1056 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1057 snprintf(xstats_names[count].name,
1058 sizeof(xstats_names[count].name),
1060 rte_virtio_rxq_stat_strings[t].name);
1065 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1066 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1069 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1070 snprintf(xstats_names[count].name,
1071 sizeof(xstats_names[count].name),
1073 rte_virtio_txq_stat_strings[t].name);
1083 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
1089 unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
1090 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
1095 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1096 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1103 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1104 xstats[count].value = *(uint64_t *)(((char *)rxvq) +
1105 rte_virtio_rxq_stat_strings[t].offset);
1106 xstats[count].id = count;
1111 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1112 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1119 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1120 xstats[count].value = *(uint64_t *)(((char *)txvq) +
1121 rte_virtio_txq_stat_strings[t].offset);
1122 xstats[count].id = count;
1131 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1133 virtio_update_stats(dev, stats);
1139 virtio_dev_stats_reset(struct rte_eth_dev *dev)
1143 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1144 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1148 txvq->stats.packets = 0;
1149 txvq->stats.bytes = 0;
1150 txvq->stats.multicast = 0;
1151 txvq->stats.broadcast = 0;
1152 memset(txvq->stats.size_bins, 0,
1153 sizeof(txvq->stats.size_bins[0]) * 8);
1156 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1157 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1161 rxvq->stats.packets = 0;
1162 rxvq->stats.bytes = 0;
1163 rxvq->stats.errors = 0;
1164 rxvq->stats.multicast = 0;
1165 rxvq->stats.broadcast = 0;
1166 memset(rxvq->stats.size_bins, 0,
1167 sizeof(rxvq->stats.size_bins[0]) * 8);
1174 virtio_set_hwaddr(struct virtio_hw *hw)
1176 virtio_write_dev_config(hw,
1177 offsetof(struct virtio_net_config, mac),
1178 &hw->mac_addr, RTE_ETHER_ADDR_LEN);
1182 virtio_get_hwaddr(struct virtio_hw *hw)
1184 if (virtio_with_feature(hw, VIRTIO_NET_F_MAC)) {
1185 virtio_read_dev_config(hw,
1186 offsetof(struct virtio_net_config, mac),
1187 &hw->mac_addr, RTE_ETHER_ADDR_LEN);
1189 rte_eth_random_addr(&hw->mac_addr[0]);
1190 virtio_set_hwaddr(hw);
1195 virtio_mac_table_set(struct virtio_hw *hw,
1196 const struct virtio_net_ctrl_mac *uc,
1197 const struct virtio_net_ctrl_mac *mc)
1199 struct virtio_pmd_ctrl ctrl;
1202 if (!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1203 PMD_DRV_LOG(INFO, "host does not support mac table");
1207 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1208 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1210 len[0] = uc->entries * RTE_ETHER_ADDR_LEN + sizeof(uc->entries);
1211 memcpy(ctrl.data, uc, len[0]);
1213 len[1] = mc->entries * RTE_ETHER_ADDR_LEN + sizeof(mc->entries);
1214 memcpy(ctrl.data + len[0], mc, len[1]);
1216 err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1218 PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1223 virtio_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
1224 uint32_t index, uint32_t vmdq __rte_unused)
1226 struct virtio_hw *hw = dev->data->dev_private;
1227 const struct rte_ether_addr *addrs = dev->data->mac_addrs;
1229 struct virtio_net_ctrl_mac *uc, *mc;
1231 if (index >= VIRTIO_MAX_MAC_ADDRS) {
1232 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1236 uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1237 sizeof(uc->entries));
1239 mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1240 sizeof(mc->entries));
1243 for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1244 const struct rte_ether_addr *addr
1245 = (i == index) ? mac_addr : addrs + i;
1246 struct virtio_net_ctrl_mac *tbl
1247 = rte_is_multicast_ether_addr(addr) ? mc : uc;
1249 memcpy(&tbl->macs[tbl->entries++], addr, RTE_ETHER_ADDR_LEN);
1252 return virtio_mac_table_set(hw, uc, mc);
1256 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1258 struct virtio_hw *hw = dev->data->dev_private;
1259 struct rte_ether_addr *addrs = dev->data->mac_addrs;
1260 struct virtio_net_ctrl_mac *uc, *mc;
1263 if (index >= VIRTIO_MAX_MAC_ADDRS) {
1264 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1268 uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1269 sizeof(uc->entries));
1271 mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1272 sizeof(mc->entries));
1275 for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1276 struct virtio_net_ctrl_mac *tbl;
1278 if (i == index || rte_is_zero_ether_addr(addrs + i))
1281 tbl = rte_is_multicast_ether_addr(addrs + i) ? mc : uc;
1282 memcpy(&tbl->macs[tbl->entries++], addrs + i,
1283 RTE_ETHER_ADDR_LEN);
1286 virtio_mac_table_set(hw, uc, mc);
1290 virtio_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1292 struct virtio_hw *hw = dev->data->dev_private;
1294 memcpy(hw->mac_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1296 /* Use atomic update if available */
1297 if (virtio_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1298 struct virtio_pmd_ctrl ctrl;
1299 int len = RTE_ETHER_ADDR_LEN;
1301 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1302 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1304 memcpy(ctrl.data, mac_addr, RTE_ETHER_ADDR_LEN);
1305 return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1308 if (!virtio_with_feature(hw, VIRTIO_NET_F_MAC))
1311 virtio_set_hwaddr(hw);
1316 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1318 struct virtio_hw *hw = dev->data->dev_private;
1319 struct virtio_pmd_ctrl ctrl;
1322 if (!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1325 ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1326 ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1327 memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1328 len = sizeof(vlan_id);
1330 return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1334 virtio_intr_unmask(struct rte_eth_dev *dev)
1336 struct virtio_hw *hw = dev->data->dev_private;
1338 if (rte_intr_ack(dev->intr_handle) < 0)
1341 if (VIRTIO_OPS(hw)->intr_detect)
1342 VIRTIO_OPS(hw)->intr_detect(hw);
1348 virtio_intr_enable(struct rte_eth_dev *dev)
1350 struct virtio_hw *hw = dev->data->dev_private;
1352 if (rte_intr_enable(dev->intr_handle) < 0)
1355 if (VIRTIO_OPS(hw)->intr_detect)
1356 VIRTIO_OPS(hw)->intr_detect(hw);
1362 virtio_intr_disable(struct rte_eth_dev *dev)
1364 struct virtio_hw *hw = dev->data->dev_private;
1366 if (rte_intr_disable(dev->intr_handle) < 0)
1369 if (VIRTIO_OPS(hw)->intr_detect)
1370 VIRTIO_OPS(hw)->intr_detect(hw);
1376 virtio_ethdev_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1378 uint64_t host_features;
1380 /* Prepare guest_features: feature that driver wants to support */
1381 PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1384 /* Read device(host) feature bits */
1385 host_features = VIRTIO_OPS(hw)->get_features(hw);
1386 PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1389 /* If supported, ensure MTU value is valid before acknowledging it. */
1390 if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1391 struct virtio_net_config config;
1393 virtio_read_dev_config(hw,
1394 offsetof(struct virtio_net_config, mtu),
1395 &config.mtu, sizeof(config.mtu));
1397 if (config.mtu < RTE_ETHER_MIN_MTU)
1398 req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1402 * Negotiate features: Subset of device feature bits are written back
1403 * guest feature bits.
1405 hw->guest_features = req_features;
1406 hw->guest_features = virtio_negotiate_features(hw, host_features);
1407 PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1408 hw->guest_features);
1410 if (VIRTIO_OPS(hw)->features_ok(hw) < 0)
1413 if (virtio_with_feature(hw, VIRTIO_F_VERSION_1)) {
1414 virtio_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1416 if (!(virtio_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1417 PMD_INIT_LOG(ERR, "Failed to set FEATURES_OK status!");
1422 hw->req_guest_features = req_features;
1428 virtio_dev_pause(struct rte_eth_dev *dev)
1430 struct virtio_hw *hw = dev->data->dev_private;
1432 rte_spinlock_lock(&hw->state_lock);
1434 if (hw->started == 0) {
1435 /* Device is just stopped. */
1436 rte_spinlock_unlock(&hw->state_lock);
1441 * Prevent the worker threads from touching queues to avoid contention,
1442 * 1 ms should be enough for the ongoing Tx function to finish.
1449 * Recover hw state to let the worker threads continue.
1452 virtio_dev_resume(struct rte_eth_dev *dev)
1454 struct virtio_hw *hw = dev->data->dev_private;
1457 rte_spinlock_unlock(&hw->state_lock);
1461 * Should be called only after device is paused.
1464 virtio_inject_pkts(struct rte_eth_dev *dev, struct rte_mbuf **tx_pkts,
1467 struct virtio_hw *hw = dev->data->dev_private;
1468 struct virtnet_tx *txvq = dev->data->tx_queues[0];
1471 hw->inject_pkts = tx_pkts;
1472 ret = dev->tx_pkt_burst(txvq, tx_pkts, nb_pkts);
1473 hw->inject_pkts = NULL;
1479 virtio_notify_peers(struct rte_eth_dev *dev)
1481 struct virtio_hw *hw = dev->data->dev_private;
1482 struct virtnet_rx *rxvq;
1483 struct rte_mbuf *rarp_mbuf;
1485 if (!dev->data->rx_queues)
1488 rxvq = dev->data->rx_queues[0];
1492 rarp_mbuf = rte_net_make_rarp_packet(rxvq->mpool,
1493 (struct rte_ether_addr *)hw->mac_addr);
1494 if (rarp_mbuf == NULL) {
1495 PMD_DRV_LOG(ERR, "failed to make RARP packet.");
1499 /* If virtio port just stopped, no need to send RARP */
1500 if (virtio_dev_pause(dev) < 0) {
1501 rte_pktmbuf_free(rarp_mbuf);
1505 virtio_inject_pkts(dev, &rarp_mbuf, 1);
1506 virtio_dev_resume(dev);
1510 virtio_ack_link_announce(struct rte_eth_dev *dev)
1512 struct virtio_hw *hw = dev->data->dev_private;
1513 struct virtio_pmd_ctrl ctrl;
1515 ctrl.hdr.class = VIRTIO_NET_CTRL_ANNOUNCE;
1516 ctrl.hdr.cmd = VIRTIO_NET_CTRL_ANNOUNCE_ACK;
1518 virtio_send_command(hw->cvq, &ctrl, NULL, 0);
1522 * Process virtio config changed interrupt. Call the callback
1523 * if link state changed, generate gratuitous RARP packet if
1524 * the status indicates an ANNOUNCE.
1527 virtio_interrupt_handler(void *param)
1529 struct rte_eth_dev *dev = param;
1530 struct virtio_hw *hw = dev->data->dev_private;
1534 /* Read interrupt status which clears interrupt */
1535 isr = virtio_get_isr(hw);
1536 PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1538 if (virtio_intr_unmask(dev) < 0)
1539 PMD_DRV_LOG(ERR, "interrupt enable failed");
1541 if (isr & VIRTIO_ISR_CONFIG) {
1542 if (virtio_dev_link_update(dev, 0) == 0)
1543 rte_eth_dev_callback_process(dev,
1544 RTE_ETH_EVENT_INTR_LSC,
1547 if (virtio_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1548 virtio_read_dev_config(hw,
1549 offsetof(struct virtio_net_config, status),
1550 &status, sizeof(status));
1551 if (status & VIRTIO_NET_S_ANNOUNCE) {
1552 virtio_notify_peers(dev);
1554 virtio_ack_link_announce(dev);
1560 /* set rx and tx handlers according to what is supported */
1562 set_rxtx_funcs(struct rte_eth_dev *eth_dev)
1564 struct virtio_hw *hw = eth_dev->data->dev_private;
1566 eth_dev->tx_pkt_prepare = virtio_xmit_pkts_prepare;
1567 if (virtio_with_packed_queue(hw)) {
1569 "virtio: using packed ring %s Tx path on port %u",
1570 hw->use_vec_tx ? "vectorized" : "standard",
1571 eth_dev->data->port_id);
1573 eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed_vec;
1575 eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed;
1577 if (hw->use_inorder_tx) {
1578 PMD_INIT_LOG(INFO, "virtio: using inorder Tx path on port %u",
1579 eth_dev->data->port_id);
1580 eth_dev->tx_pkt_burst = virtio_xmit_pkts_inorder;
1582 PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
1583 eth_dev->data->port_id);
1584 eth_dev->tx_pkt_burst = virtio_xmit_pkts;
1588 if (virtio_with_packed_queue(hw)) {
1589 if (hw->use_vec_rx) {
1591 "virtio: using packed ring vectorized Rx path on port %u",
1592 eth_dev->data->port_id);
1593 eth_dev->rx_pkt_burst =
1594 &virtio_recv_pkts_packed_vec;
1595 } else if (virtio_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1597 "virtio: using packed ring mergeable buffer Rx path on port %u",
1598 eth_dev->data->port_id);
1599 eth_dev->rx_pkt_burst =
1600 &virtio_recv_mergeable_pkts_packed;
1603 "virtio: using packed ring standard Rx path on port %u",
1604 eth_dev->data->port_id);
1605 eth_dev->rx_pkt_burst = &virtio_recv_pkts_packed;
1608 if (hw->use_vec_rx) {
1609 PMD_INIT_LOG(INFO, "virtio: using vectorized Rx path on port %u",
1610 eth_dev->data->port_id);
1611 eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1612 } else if (hw->use_inorder_rx) {
1614 "virtio: using inorder Rx path on port %u",
1615 eth_dev->data->port_id);
1616 eth_dev->rx_pkt_burst = &virtio_recv_pkts_inorder;
1617 } else if (virtio_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1619 "virtio: using mergeable buffer Rx path on port %u",
1620 eth_dev->data->port_id);
1621 eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1623 PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
1624 eth_dev->data->port_id);
1625 eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1631 /* Only support 1:1 queue/interrupt mapping so far.
1632 * TODO: support n:1 queue/interrupt mapping when there are limited number of
1633 * interrupt vectors (<N+1).
1636 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1639 struct virtio_hw *hw = dev->data->dev_private;
1641 PMD_INIT_LOG(INFO, "queue/interrupt binding");
1642 for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1643 dev->intr_handle->intr_vec[i] = i + 1;
1644 if (VIRTIO_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1645 VIRTIO_MSI_NO_VECTOR) {
1646 PMD_DRV_LOG(ERR, "failed to set queue vector");
1655 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1658 struct virtio_hw *hw = dev->data->dev_private;
1660 PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1661 for (i = 0; i < dev->data->nb_rx_queues; ++i)
1662 VIRTIO_OPS(hw)->set_queue_irq(hw,
1663 hw->vqs[i * VTNET_CQ],
1664 VIRTIO_MSI_NO_VECTOR);
1668 virtio_configure_intr(struct rte_eth_dev *dev)
1670 struct virtio_hw *hw = dev->data->dev_private;
1672 if (!rte_intr_cap_multiple(dev->intr_handle)) {
1673 PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1677 if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1678 PMD_INIT_LOG(ERR, "Fail to create eventfd");
1682 if (!dev->intr_handle->intr_vec) {
1683 dev->intr_handle->intr_vec =
1684 rte_zmalloc("intr_vec",
1685 hw->max_queue_pairs * sizeof(int), 0);
1686 if (!dev->intr_handle->intr_vec) {
1687 PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1688 hw->max_queue_pairs);
1693 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
1694 /* Re-register callback to update max_intr */
1695 rte_intr_callback_unregister(dev->intr_handle,
1696 virtio_interrupt_handler,
1698 rte_intr_callback_register(dev->intr_handle,
1699 virtio_interrupt_handler,
1703 /* DO NOT try to remove this! This function will enable msix, or QEMU
1704 * will encounter SIGSEGV when DRIVER_OK is sent.
1705 * And for legacy devices, this should be done before queue/vec binding
1706 * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1707 * (22) will be ignored.
1709 if (virtio_intr_enable(dev) < 0) {
1710 PMD_DRV_LOG(ERR, "interrupt enable failed");
1714 if (virtio_queues_bind_intr(dev) < 0) {
1715 PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1721 #define DUPLEX_UNKNOWN 0xff
1722 /* reset device and renegotiate features if needed */
1724 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1726 struct virtio_hw *hw = eth_dev->data->dev_private;
1727 struct virtio_net_config *config;
1728 struct virtio_net_config local_config;
1731 /* Reset the device although not necessary at startup */
1735 virtio_dev_free_mbufs(eth_dev);
1736 virtio_free_queues(hw);
1739 /* Tell the host we've noticed this device. */
1740 virtio_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1742 /* Tell the host we've known how to drive the device. */
1743 virtio_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1744 if (virtio_ethdev_negotiate_features(hw, req_features) < 0)
1747 hw->weak_barriers = !virtio_with_feature(hw, VIRTIO_F_ORDER_PLATFORM);
1749 /* If host does not support both status and MSI-X then disable LSC */
1750 if (virtio_with_feature(hw, VIRTIO_NET_F_STATUS) && hw->intr_lsc)
1751 eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1753 eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1755 eth_dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
1757 /* Setting up rx_header size for the device */
1758 if (virtio_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1759 virtio_with_feature(hw, VIRTIO_F_VERSION_1) ||
1760 virtio_with_packed_queue(hw))
1761 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1763 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1765 /* Copy the permanent MAC address to: virtio_hw */
1766 virtio_get_hwaddr(hw);
1767 rte_ether_addr_copy((struct rte_ether_addr *)hw->mac_addr,
1768 ð_dev->data->mac_addrs[0]);
1770 "PORT MAC: " RTE_ETHER_ADDR_PRT_FMT,
1771 hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1772 hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1774 if (hw->speed == ETH_SPEED_NUM_UNKNOWN) {
1775 if (virtio_with_feature(hw, VIRTIO_NET_F_SPEED_DUPLEX)) {
1776 config = &local_config;
1777 virtio_read_dev_config(hw,
1778 offsetof(struct virtio_net_config, speed),
1779 &config->speed, sizeof(config->speed));
1780 virtio_read_dev_config(hw,
1781 offsetof(struct virtio_net_config, duplex),
1782 &config->duplex, sizeof(config->duplex));
1783 hw->speed = config->speed;
1784 hw->duplex = config->duplex;
1787 if (hw->duplex == DUPLEX_UNKNOWN)
1788 hw->duplex = ETH_LINK_FULL_DUPLEX;
1789 PMD_INIT_LOG(DEBUG, "link speed = %d, duplex = %d",
1790 hw->speed, hw->duplex);
1791 if (virtio_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1792 config = &local_config;
1794 virtio_read_dev_config(hw,
1795 offsetof(struct virtio_net_config, mac),
1796 &config->mac, sizeof(config->mac));
1798 if (virtio_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1799 virtio_read_dev_config(hw,
1800 offsetof(struct virtio_net_config, status),
1801 &config->status, sizeof(config->status));
1804 "VIRTIO_NET_F_STATUS is not supported");
1808 if (virtio_with_feature(hw, VIRTIO_NET_F_MQ)) {
1809 virtio_read_dev_config(hw,
1810 offsetof(struct virtio_net_config, max_virtqueue_pairs),
1811 &config->max_virtqueue_pairs,
1812 sizeof(config->max_virtqueue_pairs));
1815 "VIRTIO_NET_F_MQ is not supported");
1816 config->max_virtqueue_pairs = 1;
1819 hw->max_queue_pairs = config->max_virtqueue_pairs;
1821 if (virtio_with_feature(hw, VIRTIO_NET_F_MTU)) {
1822 virtio_read_dev_config(hw,
1823 offsetof(struct virtio_net_config, mtu),
1825 sizeof(config->mtu));
1828 * MTU value has already been checked at negotiation
1829 * time, but check again in case it has changed since
1830 * then, which should not happen.
1832 if (config->mtu < RTE_ETHER_MIN_MTU) {
1833 PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1838 hw->max_mtu = config->mtu;
1839 /* Set initial MTU to maximum one supported by vhost */
1840 eth_dev->data->mtu = config->mtu;
1843 hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1844 VLAN_TAG_LEN - hw->vtnet_hdr_size;
1847 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1848 config->max_virtqueue_pairs);
1849 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1851 "PORT MAC: " RTE_ETHER_ADDR_PRT_FMT,
1852 config->mac[0], config->mac[1],
1853 config->mac[2], config->mac[3],
1854 config->mac[4], config->mac[5]);
1856 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1857 hw->max_queue_pairs = 1;
1858 hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1859 VLAN_TAG_LEN - hw->vtnet_hdr_size;
1862 ret = virtio_alloc_queues(eth_dev);
1866 if (eth_dev->data->dev_conf.intr_conf.rxq) {
1867 if (virtio_configure_intr(eth_dev) < 0) {
1868 PMD_INIT_LOG(ERR, "failed to configure interrupt");
1869 virtio_free_queues(hw);
1874 virtio_reinit_complete(hw);
1880 * This function is based on probe() function in virtio_pci.c
1881 * It returns 0 on success.
1884 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1886 struct virtio_hw *hw = eth_dev->data->dev_private;
1887 uint32_t speed = ETH_SPEED_NUM_UNKNOWN;
1891 if (sizeof(struct virtio_net_hdr_mrg_rxbuf) > RTE_PKTMBUF_HEADROOM) {
1893 "Not sufficient headroom required = %d, avail = %d",
1894 (int)sizeof(struct virtio_net_hdr_mrg_rxbuf),
1895 RTE_PKTMBUF_HEADROOM);
1900 eth_dev->dev_ops = &virtio_eth_dev_ops;
1902 if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1903 set_rxtx_funcs(eth_dev);
1907 ret = virtio_dev_devargs_parse(eth_dev->device->devargs, &speed, &vectorized);
1911 hw->duplex = DUPLEX_UNKNOWN;
1913 /* Allocate memory for storing MAC addresses */
1914 eth_dev->data->mac_addrs = rte_zmalloc("virtio",
1915 VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN, 0);
1916 if (eth_dev->data->mac_addrs == NULL) {
1918 "Failed to allocate %d bytes needed to store MAC addresses",
1919 VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN);
1923 rte_spinlock_init(&hw->state_lock);
1925 /* reset device and negotiate default features */
1926 ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1928 goto err_virtio_init;
1931 if (!virtio_with_packed_queue(hw)) {
1934 #if defined(CC_AVX512_SUPPORT) || defined(RTE_ARCH_ARM)
1939 "building environment do not support packed ring vectorized");
1949 rte_free(eth_dev->data->mac_addrs);
1950 eth_dev->data->mac_addrs = NULL;
1955 virtio_dev_speed_capa_get(uint32_t speed)
1958 case ETH_SPEED_NUM_10G:
1959 return ETH_LINK_SPEED_10G;
1960 case ETH_SPEED_NUM_20G:
1961 return ETH_LINK_SPEED_20G;
1962 case ETH_SPEED_NUM_25G:
1963 return ETH_LINK_SPEED_25G;
1964 case ETH_SPEED_NUM_40G:
1965 return ETH_LINK_SPEED_40G;
1966 case ETH_SPEED_NUM_50G:
1967 return ETH_LINK_SPEED_50G;
1968 case ETH_SPEED_NUM_56G:
1969 return ETH_LINK_SPEED_56G;
1970 case ETH_SPEED_NUM_100G:
1971 return ETH_LINK_SPEED_100G;
1972 case ETH_SPEED_NUM_200G:
1973 return ETH_LINK_SPEED_200G;
1979 static int vectorized_check_handler(__rte_unused const char *key,
1980 const char *value, void *ret_val)
1982 if (strcmp(value, "1") == 0)
1983 *(int *)ret_val = 1;
1985 *(int *)ret_val = 0;
1990 #define VIRTIO_ARG_SPEED "speed"
1991 #define VIRTIO_ARG_VECTORIZED "vectorized"
1994 link_speed_handler(const char *key __rte_unused,
1995 const char *value, void *ret_val)
1998 if (!value || !ret_val)
2000 val = strtoul(value, NULL, 0);
2001 /* validate input */
2002 if (virtio_dev_speed_capa_get(val) == 0)
2004 *(uint32_t *)ret_val = val;
2011 virtio_dev_devargs_parse(struct rte_devargs *devargs, uint32_t *speed, int *vectorized)
2013 struct rte_kvargs *kvlist;
2016 if (devargs == NULL)
2019 kvlist = rte_kvargs_parse(devargs->args, NULL);
2020 if (kvlist == NULL) {
2021 PMD_INIT_LOG(ERR, "error when parsing param");
2025 if (speed && rte_kvargs_count(kvlist, VIRTIO_ARG_SPEED) == 1) {
2026 ret = rte_kvargs_process(kvlist,
2028 link_speed_handler, speed);
2030 PMD_INIT_LOG(ERR, "Failed to parse %s",
2037 rte_kvargs_count(kvlist, VIRTIO_ARG_VECTORIZED) == 1) {
2038 ret = rte_kvargs_process(kvlist,
2039 VIRTIO_ARG_VECTORIZED,
2040 vectorized_check_handler, vectorized);
2042 PMD_INIT_LOG(ERR, "Failed to parse %s",
2043 VIRTIO_ARG_VECTORIZED);
2049 rte_kvargs_free(kvlist);
2054 rx_offload_enabled(struct virtio_hw *hw)
2056 return virtio_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM) ||
2057 virtio_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2058 virtio_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6);
2062 tx_offload_enabled(struct virtio_hw *hw)
2064 return virtio_with_feature(hw, VIRTIO_NET_F_CSUM) ||
2065 virtio_with_feature(hw, VIRTIO_NET_F_HOST_TSO4) ||
2066 virtio_with_feature(hw, VIRTIO_NET_F_HOST_TSO6);
2070 * Configure virtio device
2071 * It returns 0 on success.
2074 virtio_dev_configure(struct rte_eth_dev *dev)
2076 const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2077 const struct rte_eth_txmode *txmode = &dev->data->dev_conf.txmode;
2078 struct virtio_hw *hw = dev->data->dev_private;
2079 uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
2081 uint64_t rx_offloads = rxmode->offloads;
2082 uint64_t tx_offloads = txmode->offloads;
2083 uint64_t req_features;
2086 PMD_INIT_LOG(DEBUG, "configure");
2087 req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
2089 if (rxmode->mq_mode != ETH_MQ_RX_NONE) {
2091 "Unsupported Rx multi queue mode %d",
2096 if (txmode->mq_mode != ETH_MQ_TX_NONE) {
2098 "Unsupported Tx multi queue mode %d",
2103 if (dev->data->dev_conf.intr_conf.rxq) {
2104 ret = virtio_init_device(dev, hw->req_guest_features);
2109 if (rxmode->mtu > hw->max_mtu)
2110 req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
2112 hw->max_rx_pkt_len = ether_hdr_len + rxmode->mtu;
2114 if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2115 DEV_RX_OFFLOAD_TCP_CKSUM))
2116 req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
2118 if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO)
2120 (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2121 (1ULL << VIRTIO_NET_F_GUEST_TSO6);
2123 if (tx_offloads & (DEV_TX_OFFLOAD_UDP_CKSUM |
2124 DEV_TX_OFFLOAD_TCP_CKSUM))
2125 req_features |= (1ULL << VIRTIO_NET_F_CSUM);
2127 if (tx_offloads & DEV_TX_OFFLOAD_TCP_TSO)
2129 (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2130 (1ULL << VIRTIO_NET_F_HOST_TSO6);
2132 /* if request features changed, reinit the device */
2133 if (req_features != hw->req_guest_features) {
2134 ret = virtio_init_device(dev, req_features);
2139 if ((rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2140 DEV_RX_OFFLOAD_TCP_CKSUM)) &&
2141 !virtio_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
2143 "rx checksum not available on this host");
2147 if ((rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) &&
2148 (!virtio_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2149 !virtio_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6))) {
2151 "Large Receive Offload not available on this host");
2155 /* start control queue */
2156 if (virtio_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
2157 virtio_dev_cq_start(dev);
2159 if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
2162 hw->rx_ol_scatter = (rx_offloads & DEV_RX_OFFLOAD_SCATTER);
2164 if ((rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER) &&
2165 !virtio_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2167 "vlan filtering not available on this host");
2171 hw->has_tx_offload = tx_offload_enabled(hw);
2172 hw->has_rx_offload = rx_offload_enabled(hw);
2174 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2175 /* Enable vector (0) for Link State Intrerrupt */
2176 if (VIRTIO_OPS(hw)->set_config_irq(hw, 0) ==
2177 VIRTIO_MSI_NO_VECTOR) {
2178 PMD_DRV_LOG(ERR, "failed to set config vector");
2182 if (virtio_with_packed_queue(hw)) {
2183 #if defined(RTE_ARCH_X86_64) && defined(CC_AVX512_SUPPORT)
2184 if ((hw->use_vec_rx || hw->use_vec_tx) &&
2185 (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX512F) ||
2186 !virtio_with_feature(hw, VIRTIO_F_IN_ORDER) ||
2187 !virtio_with_feature(hw, VIRTIO_F_VERSION_1) ||
2188 rte_vect_get_max_simd_bitwidth() < RTE_VECT_SIMD_512)) {
2190 "disabled packed ring vectorized path for requirements not met");
2194 #elif defined(RTE_ARCH_ARM)
2195 if ((hw->use_vec_rx || hw->use_vec_tx) &&
2196 (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON) ||
2197 !virtio_with_feature(hw, VIRTIO_F_IN_ORDER) ||
2198 !virtio_with_feature(hw, VIRTIO_F_VERSION_1) ||
2199 rte_vect_get_max_simd_bitwidth() < RTE_VECT_SIMD_128)) {
2201 "disabled packed ring vectorized path for requirements not met");
2210 if (hw->use_vec_rx) {
2211 if (virtio_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2213 "disabled packed ring vectorized rx for mrg_rxbuf enabled");
2217 if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) {
2219 "disabled packed ring vectorized rx for TCP_LRO enabled");
2224 if (virtio_with_feature(hw, VIRTIO_F_IN_ORDER)) {
2225 hw->use_inorder_tx = 1;
2226 hw->use_inorder_rx = 1;
2230 if (hw->use_vec_rx) {
2231 #if defined RTE_ARCH_ARM
2232 if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
2234 "disabled split ring vectorized path for requirement not met");
2238 if (virtio_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2240 "disabled split ring vectorized rx for mrg_rxbuf enabled");
2244 if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2245 DEV_RX_OFFLOAD_TCP_CKSUM |
2246 DEV_RX_OFFLOAD_TCP_LRO |
2247 DEV_RX_OFFLOAD_VLAN_STRIP)) {
2249 "disabled split ring vectorized rx for offloading enabled");
2253 if (rte_vect_get_max_simd_bitwidth() < RTE_VECT_SIMD_128) {
2255 "disabled split ring vectorized rx, max SIMD bitwidth too low");
2266 virtio_dev_start(struct rte_eth_dev *dev)
2268 uint16_t nb_queues, i;
2269 struct virtqueue *vq;
2270 struct virtio_hw *hw = dev->data->dev_private;
2273 /* Finish the initialization of the queues */
2274 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2275 ret = virtio_dev_rx_queue_setup_finish(dev, i);
2279 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2280 ret = virtio_dev_tx_queue_setup_finish(dev, i);
2285 /* check if lsc interrupt feature is enabled */
2286 if (dev->data->dev_conf.intr_conf.lsc) {
2287 if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
2288 PMD_DRV_LOG(ERR, "link status not supported by host");
2293 /* Enable uio/vfio intr/eventfd mapping: althrough we already did that
2294 * in device configure, but it could be unmapped when device is
2297 if (dev->data->dev_conf.intr_conf.lsc ||
2298 dev->data->dev_conf.intr_conf.rxq) {
2299 virtio_intr_disable(dev);
2301 /* Setup interrupt callback */
2302 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2303 rte_intr_callback_register(dev->intr_handle,
2304 virtio_interrupt_handler,
2307 if (virtio_intr_enable(dev) < 0) {
2308 PMD_DRV_LOG(ERR, "interrupt enable failed");
2313 /*Notify the backend
2314 *Otherwise the tap backend might already stop its queue due to fullness.
2315 *vhost backend will have no chance to be waked up
2317 nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
2318 if (hw->max_queue_pairs > 1) {
2319 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
2323 PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
2325 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2326 vq = virtnet_rxq_to_vq(dev->data->rx_queues[i]);
2327 /* Flush the old packets */
2328 virtqueue_rxvq_flush(vq);
2329 virtqueue_notify(vq);
2332 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2333 vq = virtnet_txq_to_vq(dev->data->tx_queues[i]);
2334 virtqueue_notify(vq);
2337 PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
2339 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2340 vq = virtnet_rxq_to_vq(dev->data->rx_queues[i]);
2344 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2345 vq = virtnet_txq_to_vq(dev->data->tx_queues[i]);
2349 set_rxtx_funcs(dev);
2352 /* Initialize Link state */
2353 virtio_dev_link_update(dev, 0);
2358 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
2360 struct virtio_hw *hw = dev->data->dev_private;
2361 uint16_t nr_vq = virtio_get_nr_vq(hw);
2362 const char *type __rte_unused;
2363 unsigned int i, mbuf_num = 0;
2364 struct virtqueue *vq;
2365 struct rte_mbuf *buf;
2368 if (hw->vqs == NULL)
2371 for (i = 0; i < nr_vq; i++) {
2376 queue_type = virtio_get_queue_type(hw, i);
2377 if (queue_type == VTNET_RQ)
2379 else if (queue_type == VTNET_TQ)
2385 "Before freeing %s[%d] used and unused buf",
2389 while ((buf = virtqueue_detach_unused(vq)) != NULL) {
2390 rte_pktmbuf_free(buf);
2395 "After freeing %s[%d] used and unused buf",
2400 PMD_INIT_LOG(DEBUG, "%d mbufs freed", mbuf_num);
2404 virtio_tx_completed_cleanup(struct rte_eth_dev *dev)
2406 struct virtio_hw *hw = dev->data->dev_private;
2407 struct virtqueue *vq;
2409 void (*xmit_cleanup)(struct virtqueue *vq, uint16_t nb_used);
2411 if (virtio_with_packed_queue(hw)) {
2413 xmit_cleanup = &virtio_xmit_cleanup_inorder_packed;
2414 else if (virtio_with_feature(hw, VIRTIO_F_IN_ORDER))
2415 xmit_cleanup = &virtio_xmit_cleanup_inorder_packed;
2417 xmit_cleanup = &virtio_xmit_cleanup_normal_packed;
2419 if (hw->use_inorder_tx)
2420 xmit_cleanup = &virtio_xmit_cleanup_inorder;
2422 xmit_cleanup = &virtio_xmit_cleanup;
2425 for (qidx = 0; qidx < hw->max_queue_pairs; qidx++) {
2426 vq = hw->vqs[2 * qidx + VTNET_SQ_TQ_QUEUE_IDX];
2428 xmit_cleanup(vq, virtqueue_nused(vq));
2433 * Stop device: disable interrupt and mark link down
2436 virtio_dev_stop(struct rte_eth_dev *dev)
2438 struct virtio_hw *hw = dev->data->dev_private;
2439 struct rte_eth_link link;
2440 struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
2442 PMD_INIT_LOG(DEBUG, "stop");
2443 dev->data->dev_started = 0;
2445 rte_spinlock_lock(&hw->state_lock);
2450 virtio_tx_completed_cleanup(dev);
2452 if (intr_conf->lsc || intr_conf->rxq) {
2453 virtio_intr_disable(dev);
2455 /* Reset interrupt callback */
2456 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
2457 rte_intr_callback_unregister(dev->intr_handle,
2458 virtio_interrupt_handler,
2463 memset(&link, 0, sizeof(link));
2464 rte_eth_linkstatus_set(dev, &link);
2466 rte_spinlock_unlock(&hw->state_lock);
2472 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
2474 struct rte_eth_link link;
2476 struct virtio_hw *hw = dev->data->dev_private;
2478 memset(&link, 0, sizeof(link));
2479 link.link_duplex = hw->duplex;
2480 link.link_speed = hw->speed;
2481 link.link_autoneg = ETH_LINK_AUTONEG;
2484 link.link_status = ETH_LINK_DOWN;
2485 link.link_speed = ETH_SPEED_NUM_NONE;
2486 } else if (virtio_with_feature(hw, VIRTIO_NET_F_STATUS)) {
2487 PMD_INIT_LOG(DEBUG, "Get link status from hw");
2488 virtio_read_dev_config(hw,
2489 offsetof(struct virtio_net_config, status),
2490 &status, sizeof(status));
2491 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
2492 link.link_status = ETH_LINK_DOWN;
2493 link.link_speed = ETH_SPEED_NUM_NONE;
2494 PMD_INIT_LOG(DEBUG, "Port %d is down",
2495 dev->data->port_id);
2497 link.link_status = ETH_LINK_UP;
2498 PMD_INIT_LOG(DEBUG, "Port %d is up",
2499 dev->data->port_id);
2502 link.link_status = ETH_LINK_UP;
2505 return rte_eth_linkstatus_set(dev, &link);
2509 virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
2511 const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2512 struct virtio_hw *hw = dev->data->dev_private;
2513 uint64_t offloads = rxmode->offloads;
2515 if (mask & ETH_VLAN_FILTER_MASK) {
2516 if ((offloads & DEV_RX_OFFLOAD_VLAN_FILTER) &&
2517 !virtio_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2520 "vlan filtering not available on this host");
2526 if (mask & ETH_VLAN_STRIP_MASK)
2527 hw->vlan_strip = !!(offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
2533 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2535 uint64_t tso_mask, host_features;
2536 struct virtio_hw *hw = dev->data->dev_private;
2537 dev_info->speed_capa = virtio_dev_speed_capa_get(hw->speed);
2539 dev_info->max_rx_queues =
2540 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
2541 dev_info->max_tx_queues =
2542 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
2543 dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
2544 dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
2545 dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
2546 dev_info->max_mtu = hw->max_mtu;
2548 host_features = VIRTIO_OPS(hw)->get_features(hw);
2549 dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
2550 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_JUMBO_FRAME;
2551 if (host_features & (1ULL << VIRTIO_NET_F_MRG_RXBUF))
2552 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_SCATTER;
2553 if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
2554 dev_info->rx_offload_capa |=
2555 DEV_RX_OFFLOAD_TCP_CKSUM |
2556 DEV_RX_OFFLOAD_UDP_CKSUM;
2558 if (host_features & (1ULL << VIRTIO_NET_F_CTRL_VLAN))
2559 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_VLAN_FILTER;
2560 tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2561 (1ULL << VIRTIO_NET_F_GUEST_TSO6);
2562 if ((host_features & tso_mask) == tso_mask)
2563 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
2565 dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
2566 DEV_TX_OFFLOAD_VLAN_INSERT;
2567 if (host_features & (1ULL << VIRTIO_NET_F_CSUM)) {
2568 dev_info->tx_offload_capa |=
2569 DEV_TX_OFFLOAD_UDP_CKSUM |
2570 DEV_TX_OFFLOAD_TCP_CKSUM;
2572 tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2573 (1ULL << VIRTIO_NET_F_HOST_TSO6);
2574 if ((host_features & tso_mask) == tso_mask)
2575 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
2577 if (host_features & (1ULL << VIRTIO_F_RING_PACKED)) {
2579 * According to 2.7 Packed Virtqueues,
2580 * 2.7.10.1 Structure Size and Alignment:
2581 * The Queue Size value does not have to be a power of 2.
2583 dev_info->rx_desc_lim.nb_max = UINT16_MAX;
2584 dev_info->tx_desc_lim.nb_max = UINT16_MAX;
2587 * According to 2.6 Split Virtqueues:
2588 * Queue Size value is always a power of 2. The maximum Queue
2589 * Size value is 32768.
2591 dev_info->rx_desc_lim.nb_max = 32768;
2592 dev_info->tx_desc_lim.nb_max = 32768;
2595 * Actual minimum is not the same for virtqueues of different kinds,
2596 * but to avoid tangling the code with separate branches, rely on
2597 * default thresholds since desc number must be at least of their size.
2599 dev_info->rx_desc_lim.nb_min = RTE_MAX(DEFAULT_RX_FREE_THRESH,
2600 RTE_VIRTIO_VPMD_RX_REARM_THRESH);
2601 dev_info->tx_desc_lim.nb_min = DEFAULT_TX_FREE_THRESH;
2602 dev_info->rx_desc_lim.nb_align = 1;
2603 dev_info->tx_desc_lim.nb_align = 1;
2609 * It enables testpmd to collect per queue stats.
2612 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
2613 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
2614 __rte_unused uint8_t is_rx)
2619 RTE_LOG_REGISTER_SUFFIX(virtio_logtype_init, init, NOTICE);
2620 RTE_LOG_REGISTER_SUFFIX(virtio_logtype_driver, driver, NOTICE);