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);
54 static int virtio_dev_rss_hash_update(struct rte_eth_dev *dev,
55 struct rte_eth_rss_conf *rss_conf);
56 static int virtio_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
57 struct rte_eth_rss_conf *rss_conf);
58 static int virtio_dev_rss_reta_update(struct rte_eth_dev *dev,
59 struct rte_eth_rss_reta_entry64 *reta_conf,
61 static int virtio_dev_rss_reta_query(struct rte_eth_dev *dev,
62 struct rte_eth_rss_reta_entry64 *reta_conf,
65 static void virtio_set_hwaddr(struct virtio_hw *hw);
66 static void virtio_get_hwaddr(struct virtio_hw *hw);
68 static int virtio_dev_stats_get(struct rte_eth_dev *dev,
69 struct rte_eth_stats *stats);
70 static int virtio_dev_xstats_get(struct rte_eth_dev *dev,
71 struct rte_eth_xstat *xstats, unsigned n);
72 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
73 struct rte_eth_xstat_name *xstats_names,
75 static int virtio_dev_stats_reset(struct rte_eth_dev *dev);
76 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev);
77 static int virtio_vlan_filter_set(struct rte_eth_dev *dev,
78 uint16_t vlan_id, int on);
79 static int virtio_mac_addr_add(struct rte_eth_dev *dev,
80 struct rte_ether_addr *mac_addr,
81 uint32_t index, uint32_t vmdq);
82 static void virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index);
83 static int virtio_mac_addr_set(struct rte_eth_dev *dev,
84 struct rte_ether_addr *mac_addr);
86 static int virtio_intr_disable(struct rte_eth_dev *dev);
87 static int virtio_get_monitor_addr(void *rx_queue,
88 struct rte_power_monitor_cond *pmc);
90 static int virtio_dev_queue_stats_mapping_set(
91 struct rte_eth_dev *eth_dev,
96 static void virtio_notify_peers(struct rte_eth_dev *dev);
97 static void virtio_ack_link_announce(struct rte_eth_dev *dev);
99 struct rte_virtio_xstats_name_off {
100 char name[RTE_ETH_XSTATS_NAME_SIZE];
104 /* [rt]x_qX_ is prepended to the name string here */
105 static const struct rte_virtio_xstats_name_off rte_virtio_rxq_stat_strings[] = {
106 {"good_packets", offsetof(struct virtnet_rx, stats.packets)},
107 {"good_bytes", offsetof(struct virtnet_rx, stats.bytes)},
108 {"errors", offsetof(struct virtnet_rx, stats.errors)},
109 {"multicast_packets", offsetof(struct virtnet_rx, stats.multicast)},
110 {"broadcast_packets", offsetof(struct virtnet_rx, stats.broadcast)},
111 {"undersize_packets", offsetof(struct virtnet_rx, stats.size_bins[0])},
112 {"size_64_packets", offsetof(struct virtnet_rx, stats.size_bins[1])},
113 {"size_65_127_packets", offsetof(struct virtnet_rx, stats.size_bins[2])},
114 {"size_128_255_packets", offsetof(struct virtnet_rx, stats.size_bins[3])},
115 {"size_256_511_packets", offsetof(struct virtnet_rx, stats.size_bins[4])},
116 {"size_512_1023_packets", offsetof(struct virtnet_rx, stats.size_bins[5])},
117 {"size_1024_1518_packets", offsetof(struct virtnet_rx, stats.size_bins[6])},
118 {"size_1519_max_packets", offsetof(struct virtnet_rx, stats.size_bins[7])},
121 /* [rt]x_qX_ is prepended to the name string here */
122 static const struct rte_virtio_xstats_name_off rte_virtio_txq_stat_strings[] = {
123 {"good_packets", offsetof(struct virtnet_tx, stats.packets)},
124 {"good_bytes", offsetof(struct virtnet_tx, stats.bytes)},
125 {"multicast_packets", offsetof(struct virtnet_tx, stats.multicast)},
126 {"broadcast_packets", offsetof(struct virtnet_tx, stats.broadcast)},
127 {"undersize_packets", offsetof(struct virtnet_tx, stats.size_bins[0])},
128 {"size_64_packets", offsetof(struct virtnet_tx, stats.size_bins[1])},
129 {"size_65_127_packets", offsetof(struct virtnet_tx, stats.size_bins[2])},
130 {"size_128_255_packets", offsetof(struct virtnet_tx, stats.size_bins[3])},
131 {"size_256_511_packets", offsetof(struct virtnet_tx, stats.size_bins[4])},
132 {"size_512_1023_packets", offsetof(struct virtnet_tx, stats.size_bins[5])},
133 {"size_1024_1518_packets", offsetof(struct virtnet_tx, stats.size_bins[6])},
134 {"size_1519_max_packets", offsetof(struct virtnet_tx, stats.size_bins[7])},
137 #define VIRTIO_NB_RXQ_XSTATS (sizeof(rte_virtio_rxq_stat_strings) / \
138 sizeof(rte_virtio_rxq_stat_strings[0]))
139 #define VIRTIO_NB_TXQ_XSTATS (sizeof(rte_virtio_txq_stat_strings) / \
140 sizeof(rte_virtio_txq_stat_strings[0]))
142 struct virtio_hw_internal virtio_hw_internal[RTE_MAX_ETHPORTS];
144 static struct virtio_pmd_ctrl *
145 virtio_send_command_packed(struct virtnet_ctl *cvq,
146 struct virtio_pmd_ctrl *ctrl,
147 int *dlen, int pkt_num)
149 struct virtqueue *vq = virtnet_cq_to_vq(cvq);
151 struct vring_packed_desc *desc = vq->vq_packed.ring.desc;
152 struct virtio_pmd_ctrl *result;
159 * Format is enforced in qemu code:
160 * One TX packet for header;
161 * At least one TX packet per argument;
162 * One RX packet for ACK.
164 head = vq->vq_avail_idx;
165 flags = vq->vq_packed.cached_flags;
166 desc[head].addr = cvq->virtio_net_hdr_mem;
167 desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
170 if (++vq->vq_avail_idx >= vq->vq_nentries) {
171 vq->vq_avail_idx -= vq->vq_nentries;
172 vq->vq_packed.cached_flags ^= VRING_PACKED_DESC_F_AVAIL_USED;
175 for (k = 0; k < pkt_num; k++) {
176 desc[vq->vq_avail_idx].addr = cvq->virtio_net_hdr_mem
177 + sizeof(struct virtio_net_ctrl_hdr)
178 + sizeof(ctrl->status) + sizeof(uint8_t) * sum;
179 desc[vq->vq_avail_idx].len = dlen[k];
180 desc[vq->vq_avail_idx].flags = VRING_DESC_F_NEXT |
181 vq->vq_packed.cached_flags;
185 if (++vq->vq_avail_idx >= vq->vq_nentries) {
186 vq->vq_avail_idx -= vq->vq_nentries;
187 vq->vq_packed.cached_flags ^=
188 VRING_PACKED_DESC_F_AVAIL_USED;
192 desc[vq->vq_avail_idx].addr = cvq->virtio_net_hdr_mem
193 + sizeof(struct virtio_net_ctrl_hdr);
194 desc[vq->vq_avail_idx].len = sizeof(ctrl->status);
195 desc[vq->vq_avail_idx].flags = VRING_DESC_F_WRITE |
196 vq->vq_packed.cached_flags;
199 if (++vq->vq_avail_idx >= vq->vq_nentries) {
200 vq->vq_avail_idx -= vq->vq_nentries;
201 vq->vq_packed.cached_flags ^= VRING_PACKED_DESC_F_AVAIL_USED;
204 virtqueue_store_flags_packed(&desc[head], VRING_DESC_F_NEXT | flags,
205 vq->hw->weak_barriers);
207 virtio_wmb(vq->hw->weak_barriers);
208 virtqueue_notify(vq);
210 /* wait for used desc in virtqueue
211 * desc_is_used has a load-acquire or rte_io_rmb inside
213 while (!desc_is_used(&desc[head], vq))
216 /* now get used descriptors */
217 vq->vq_free_cnt += nb_descs;
218 vq->vq_used_cons_idx += nb_descs;
219 if (vq->vq_used_cons_idx >= vq->vq_nentries) {
220 vq->vq_used_cons_idx -= vq->vq_nentries;
221 vq->vq_packed.used_wrap_counter ^= 1;
224 PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\n"
225 "vq->vq_avail_idx=%d\n"
226 "vq->vq_used_cons_idx=%d\n"
227 "vq->vq_packed.cached_flags=0x%x\n"
228 "vq->vq_packed.used_wrap_counter=%d",
231 vq->vq_used_cons_idx,
232 vq->vq_packed.cached_flags,
233 vq->vq_packed.used_wrap_counter);
235 result = cvq->virtio_net_hdr_mz->addr;
239 static struct virtio_pmd_ctrl *
240 virtio_send_command_split(struct virtnet_ctl *cvq,
241 struct virtio_pmd_ctrl *ctrl,
242 int *dlen, int pkt_num)
244 struct virtio_pmd_ctrl *result;
245 struct virtqueue *vq = virtnet_cq_to_vq(cvq);
249 head = vq->vq_desc_head_idx;
252 * Format is enforced in qemu code:
253 * One TX packet for header;
254 * At least one TX packet per argument;
255 * One RX packet for ACK.
257 vq->vq_split.ring.desc[head].flags = VRING_DESC_F_NEXT;
258 vq->vq_split.ring.desc[head].addr = cvq->virtio_net_hdr_mem;
259 vq->vq_split.ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
261 i = vq->vq_split.ring.desc[head].next;
263 for (k = 0; k < pkt_num; k++) {
264 vq->vq_split.ring.desc[i].flags = VRING_DESC_F_NEXT;
265 vq->vq_split.ring.desc[i].addr = cvq->virtio_net_hdr_mem
266 + sizeof(struct virtio_net_ctrl_hdr)
267 + sizeof(ctrl->status) + sizeof(uint8_t)*sum;
268 vq->vq_split.ring.desc[i].len = dlen[k];
271 i = vq->vq_split.ring.desc[i].next;
274 vq->vq_split.ring.desc[i].flags = VRING_DESC_F_WRITE;
275 vq->vq_split.ring.desc[i].addr = cvq->virtio_net_hdr_mem
276 + sizeof(struct virtio_net_ctrl_hdr);
277 vq->vq_split.ring.desc[i].len = sizeof(ctrl->status);
280 vq->vq_desc_head_idx = vq->vq_split.ring.desc[i].next;
282 vq_update_avail_ring(vq, head);
283 vq_update_avail_idx(vq);
285 PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
287 virtqueue_notify(vq);
289 while (virtqueue_nused(vq) == 0)
292 while (virtqueue_nused(vq)) {
293 uint32_t idx, desc_idx, used_idx;
294 struct vring_used_elem *uep;
296 used_idx = (uint32_t)(vq->vq_used_cons_idx
297 & (vq->vq_nentries - 1));
298 uep = &vq->vq_split.ring.used->ring[used_idx];
299 idx = (uint32_t) uep->id;
302 while (vq->vq_split.ring.desc[desc_idx].flags &
304 desc_idx = vq->vq_split.ring.desc[desc_idx].next;
308 vq->vq_split.ring.desc[desc_idx].next = vq->vq_desc_head_idx;
309 vq->vq_desc_head_idx = idx;
311 vq->vq_used_cons_idx++;
315 PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
316 vq->vq_free_cnt, vq->vq_desc_head_idx);
318 result = cvq->virtio_net_hdr_mz->addr;
323 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
324 int *dlen, int pkt_num)
326 virtio_net_ctrl_ack status = ~0;
327 struct virtio_pmd_ctrl *result;
328 struct virtqueue *vq;
330 ctrl->status = status;
333 PMD_INIT_LOG(ERR, "Control queue is not supported.");
337 rte_spinlock_lock(&cvq->lock);
338 vq = virtnet_cq_to_vq(cvq);
340 PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
341 "vq->hw->cvq = %p vq = %p",
342 vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
344 if (vq->vq_free_cnt < pkt_num + 2 || pkt_num < 1) {
345 rte_spinlock_unlock(&cvq->lock);
349 memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
350 sizeof(struct virtio_pmd_ctrl));
352 if (virtio_with_packed_queue(vq->hw))
353 result = virtio_send_command_packed(cvq, ctrl, dlen, pkt_num);
355 result = virtio_send_command_split(cvq, ctrl, dlen, pkt_num);
357 rte_spinlock_unlock(&cvq->lock);
358 return result->status;
362 virtio_set_multiple_queues_rss(struct rte_eth_dev *dev, uint16_t nb_queues)
364 struct virtio_hw *hw = dev->data->dev_private;
365 struct virtio_pmd_ctrl ctrl;
366 struct virtio_net_ctrl_rss rss;
369 rss.hash_types = hw->rss_hash_types & VIRTIO_NET_HASH_TYPE_MASK;
370 RTE_BUILD_BUG_ON(!RTE_IS_POWER_OF_2(VIRTIO_NET_RSS_RETA_SIZE));
371 rss.indirection_table_mask = VIRTIO_NET_RSS_RETA_SIZE - 1;
372 rss.unclassified_queue = 0;
373 memcpy(rss.indirection_table, hw->rss_reta, VIRTIO_NET_RSS_RETA_SIZE * sizeof(uint16_t));
374 rss.max_tx_vq = nb_queues;
375 rss.hash_key_length = VIRTIO_NET_RSS_KEY_SIZE;
376 memcpy(rss.hash_key_data, hw->rss_key, VIRTIO_NET_RSS_KEY_SIZE);
378 ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
379 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_RSS_CONFIG;
380 memcpy(ctrl.data, &rss, sizeof(rss));
384 ret = virtio_send_command(hw->cvq, &ctrl, &dlen, 1);
386 PMD_INIT_LOG(ERR, "RSS multiqueue configured but send command failed");
394 virtio_set_multiple_queues_auto(struct rte_eth_dev *dev, uint16_t nb_queues)
396 struct virtio_hw *hw = dev->data->dev_private;
397 struct virtio_pmd_ctrl ctrl;
401 ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
402 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
403 memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
405 dlen = sizeof(uint16_t);
407 ret = virtio_send_command(hw->cvq, &ctrl, &dlen, 1);
409 PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
410 "failed, this is too late now...");
418 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
420 struct virtio_hw *hw = dev->data->dev_private;
422 if (virtio_with_feature(hw, VIRTIO_NET_F_RSS))
423 return virtio_set_multiple_queues_rss(dev, nb_queues);
425 return virtio_set_multiple_queues_auto(dev, nb_queues);
429 virtio_get_nr_vq(struct virtio_hw *hw)
431 uint16_t nr_vq = hw->max_queue_pairs * 2;
433 if (virtio_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
440 virtio_init_vring(struct virtqueue *vq)
442 int size = vq->vq_nentries;
443 uint8_t *ring_mem = vq->vq_ring_virt_mem;
445 PMD_INIT_FUNC_TRACE();
447 memset(ring_mem, 0, vq->vq_ring_size);
449 vq->vq_used_cons_idx = 0;
450 vq->vq_desc_head_idx = 0;
451 vq->vq_avail_idx = 0;
452 vq->vq_desc_tail_idx = (uint16_t)(vq->vq_nentries - 1);
453 vq->vq_free_cnt = vq->vq_nentries;
454 memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries);
455 if (virtio_with_packed_queue(vq->hw)) {
456 vring_init_packed(&vq->vq_packed.ring, ring_mem,
457 VIRTIO_VRING_ALIGN, size);
458 vring_desc_init_packed(vq, size);
460 struct vring *vr = &vq->vq_split.ring;
462 vring_init_split(vr, ring_mem, VIRTIO_VRING_ALIGN, size);
463 vring_desc_init_split(vr->desc, size);
466 * Disable device(host) interrupting guest
468 virtqueue_disable_intr(vq);
472 virtio_init_queue(struct rte_eth_dev *dev, uint16_t queue_idx)
474 char vq_name[VIRTQUEUE_MAX_NAME_SZ];
475 char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
476 const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
477 unsigned int vq_size, size;
478 struct virtio_hw *hw = dev->data->dev_private;
479 struct virtnet_rx *rxvq = NULL;
480 struct virtnet_tx *txvq = NULL;
481 struct virtnet_ctl *cvq = NULL;
482 struct virtqueue *vq;
483 size_t sz_hdr_mz = 0;
484 void *sw_ring = NULL;
485 int queue_type = virtio_get_queue_type(hw, queue_idx);
487 int numa_node = dev->device->numa_node;
488 struct rte_mbuf *fake_mbuf = NULL;
490 PMD_INIT_LOG(INFO, "setting up queue: %u on NUMA node %d",
491 queue_idx, numa_node);
494 * Read the virtqueue size from the Queue Size field
495 * Always power of 2 and if 0 virtqueue does not exist
497 vq_size = VIRTIO_OPS(hw)->get_queue_num(hw, queue_idx);
498 PMD_INIT_LOG(DEBUG, "vq_size: %u", vq_size);
500 PMD_INIT_LOG(ERR, "virtqueue does not exist");
504 if (!virtio_with_packed_queue(hw) && !rte_is_power_of_2(vq_size)) {
505 PMD_INIT_LOG(ERR, "split virtqueue size is not power of 2");
509 snprintf(vq_name, sizeof(vq_name), "port%d_vq%d",
510 dev->data->port_id, queue_idx);
512 size = RTE_ALIGN_CEIL(sizeof(*vq) +
513 vq_size * sizeof(struct vq_desc_extra),
514 RTE_CACHE_LINE_SIZE);
515 if (queue_type == VTNET_TQ) {
517 * For each xmit packet, allocate a virtio_net_hdr
518 * and indirect ring elements
520 sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
521 } else if (queue_type == VTNET_CQ) {
522 /* Allocate a page for control vq command, data and status */
523 sz_hdr_mz = rte_mem_page_size();
526 vq = rte_zmalloc_socket(vq_name, size, RTE_CACHE_LINE_SIZE,
529 PMD_INIT_LOG(ERR, "can not allocate vq");
532 hw->vqs[queue_idx] = vq;
535 vq->vq_queue_index = queue_idx;
536 vq->vq_nentries = vq_size;
537 if (virtio_with_packed_queue(hw)) {
538 vq->vq_packed.used_wrap_counter = 1;
539 vq->vq_packed.cached_flags = VRING_PACKED_DESC_F_AVAIL;
540 vq->vq_packed.event_flags_shadow = 0;
541 if (queue_type == VTNET_RQ)
542 vq->vq_packed.cached_flags |= VRING_DESC_F_WRITE;
546 * Reserve a memzone for vring elements
548 size = vring_size(hw, vq_size, VIRTIO_VRING_ALIGN);
549 vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_VRING_ALIGN);
550 PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
551 size, vq->vq_ring_size);
553 mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
554 numa_node, RTE_MEMZONE_IOVA_CONTIG,
557 if (rte_errno == EEXIST)
558 mz = rte_memzone_lookup(vq_name);
565 memset(mz->addr, 0, mz->len);
568 vq->vq_ring_mem = (uintptr_t)mz->addr;
570 vq->vq_ring_mem = mz->iova;
572 vq->vq_ring_virt_mem = mz->addr;
573 PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem: 0x%" PRIx64, vq->vq_ring_mem);
574 PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: %p", vq->vq_ring_virt_mem);
576 virtio_init_vring(vq);
579 snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_vq%d_hdr",
580 dev->data->port_id, queue_idx);
581 hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
582 numa_node, RTE_MEMZONE_IOVA_CONTIG,
583 RTE_CACHE_LINE_SIZE);
584 if (hdr_mz == NULL) {
585 if (rte_errno == EEXIST)
586 hdr_mz = rte_memzone_lookup(vq_hdr_name);
587 if (hdr_mz == NULL) {
594 if (queue_type == VTNET_RQ) {
595 size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
596 sizeof(vq->sw_ring[0]);
598 sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
599 RTE_CACHE_LINE_SIZE, numa_node);
601 PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
606 fake_mbuf = rte_zmalloc_socket("sw_ring", sizeof(*fake_mbuf),
607 RTE_CACHE_LINE_SIZE, numa_node);
609 PMD_INIT_LOG(ERR, "can not allocate fake mbuf");
614 vq->sw_ring = sw_ring;
616 rxvq->port_id = dev->data->port_id;
618 rxvq->fake_mbuf = fake_mbuf;
619 } else if (queue_type == VTNET_TQ) {
621 txvq->port_id = dev->data->port_id;
623 txvq->virtio_net_hdr_mz = hdr_mz;
625 txvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
627 txvq->virtio_net_hdr_mem = hdr_mz->iova;
628 } else if (queue_type == VTNET_CQ) {
631 cvq->virtio_net_hdr_mz = hdr_mz;
633 cvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
635 cvq->virtio_net_hdr_mem = hdr_mz->iova;
636 memset(cvq->virtio_net_hdr_mz->addr, 0, rte_mem_page_size());
642 vq->mbuf_addr_offset = offsetof(struct rte_mbuf, buf_addr);
644 vq->mbuf_addr_offset = offsetof(struct rte_mbuf, buf_iova);
646 if (queue_type == VTNET_TQ) {
647 struct virtio_tx_region *txr;
651 memset(txr, 0, vq_size * sizeof(*txr));
652 for (i = 0; i < vq_size; i++) {
653 /* first indirect descriptor is always the tx header */
654 if (!virtio_with_packed_queue(hw)) {
655 struct vring_desc *start_dp = txr[i].tx_indir;
656 vring_desc_init_split(start_dp,
657 RTE_DIM(txr[i].tx_indir));
658 start_dp->addr = txvq->virtio_net_hdr_mem
660 + offsetof(struct virtio_tx_region,
662 start_dp->len = hw->vtnet_hdr_size;
663 start_dp->flags = VRING_DESC_F_NEXT;
665 struct vring_packed_desc *start_dp =
666 txr[i].tx_packed_indir;
667 vring_desc_init_indirect_packed(start_dp,
668 RTE_DIM(txr[i].tx_packed_indir));
669 start_dp->addr = txvq->virtio_net_hdr_mem
671 + offsetof(struct virtio_tx_region,
673 start_dp->len = hw->vtnet_hdr_size;
678 if (VIRTIO_OPS(hw)->setup_queue(hw, vq) < 0) {
679 PMD_INIT_LOG(ERR, "setup_queue failed");
692 rte_memzone_free(hdr_mz);
694 rte_memzone_free(mz);
697 hw->vqs[queue_idx] = NULL;
703 virtio_free_queues(struct virtio_hw *hw)
705 uint16_t nr_vq = virtio_get_nr_vq(hw);
706 struct virtqueue *vq;
713 for (i = 0; i < nr_vq; i++) {
718 queue_type = virtio_get_queue_type(hw, i);
719 if (queue_type == VTNET_RQ) {
720 rte_free(vq->rxq.fake_mbuf);
721 rte_free(vq->sw_ring);
722 rte_memzone_free(vq->rxq.mz);
723 } else if (queue_type == VTNET_TQ) {
724 rte_memzone_free(vq->txq.mz);
725 rte_memzone_free(vq->txq.virtio_net_hdr_mz);
727 rte_memzone_free(vq->cq.mz);
728 rte_memzone_free(vq->cq.virtio_net_hdr_mz);
740 virtio_alloc_queues(struct rte_eth_dev *dev)
742 struct virtio_hw *hw = dev->data->dev_private;
743 uint16_t nr_vq = virtio_get_nr_vq(hw);
747 hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
749 PMD_INIT_LOG(ERR, "failed to allocate vqs");
753 for (i = 0; i < nr_vq; i++) {
754 ret = virtio_init_queue(dev, i);
756 virtio_free_queues(hw);
764 static void virtio_queues_unbind_intr(struct rte_eth_dev *dev);
767 virtio_free_rss(struct virtio_hw *hw)
769 rte_free(hw->rss_key);
772 rte_free(hw->rss_reta);
777 virtio_dev_close(struct rte_eth_dev *dev)
779 struct virtio_hw *hw = dev->data->dev_private;
780 struct rte_eth_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
782 PMD_INIT_LOG(DEBUG, "virtio_dev_close");
783 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
791 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
792 VIRTIO_OPS(hw)->set_config_irq(hw, VIRTIO_MSI_NO_VECTOR);
794 virtio_queues_unbind_intr(dev);
796 if (intr_conf->lsc || intr_conf->rxq) {
797 virtio_intr_disable(dev);
798 rte_intr_efd_disable(dev->intr_handle);
799 rte_intr_vec_list_free(dev->intr_handle);
803 virtio_dev_free_mbufs(dev);
804 virtio_free_queues(hw);
807 return VIRTIO_OPS(hw)->dev_close(hw);
811 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
813 struct virtio_hw *hw = dev->data->dev_private;
814 struct virtio_pmd_ctrl ctrl;
818 if (!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
819 PMD_INIT_LOG(INFO, "host does not support rx control");
823 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
824 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
828 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
830 PMD_INIT_LOG(ERR, "Failed to enable promisc");
838 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
840 struct virtio_hw *hw = dev->data->dev_private;
841 struct virtio_pmd_ctrl ctrl;
845 if (!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
846 PMD_INIT_LOG(INFO, "host does not support rx control");
850 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
851 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
855 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
857 PMD_INIT_LOG(ERR, "Failed to disable promisc");
865 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
867 struct virtio_hw *hw = dev->data->dev_private;
868 struct virtio_pmd_ctrl ctrl;
872 if (!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
873 PMD_INIT_LOG(INFO, "host does not support rx control");
877 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
878 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
882 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
884 PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
892 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
894 struct virtio_hw *hw = dev->data->dev_private;
895 struct virtio_pmd_ctrl ctrl;
899 if (!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
900 PMD_INIT_LOG(INFO, "host does not support rx control");
904 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
905 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
909 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
911 PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
919 virtio_rx_mem_pool_buf_size(struct rte_mempool *mp)
921 return rte_pktmbuf_data_room_size(mp) - RTE_PKTMBUF_HEADROOM;
925 virtio_rx_check_scatter(uint16_t max_rx_pkt_len, uint16_t rx_buf_size,
926 bool rx_scatter_enabled, const char **error)
928 if (!rx_scatter_enabled && max_rx_pkt_len > rx_buf_size) {
929 *error = "Rx scatter is disabled and RxQ mbuf pool object size is too small";
937 virtio_check_scatter_on_all_rx_queues(struct rte_eth_dev *dev,
940 struct virtio_hw *hw = dev->data->dev_private;
941 struct virtnet_rx *rxvq;
942 struct virtqueue *vq;
950 for (qidx = 0; qidx < hw->max_queue_pairs; qidx++) {
951 vq = hw->vqs[2 * qidx + VTNET_SQ_RQ_QUEUE_IDX];
956 if (rxvq->mpool == NULL)
958 buf_size = virtio_rx_mem_pool_buf_size(rxvq->mpool);
960 if (!virtio_rx_check_scatter(frame_size, buf_size,
961 hw->rx_ol_scatter, &error)) {
962 PMD_INIT_LOG(ERR, "MTU check for RxQ %u failed: %s",
971 #define VLAN_TAG_LEN 4 /* 802.3ac tag (not DMA'd) */
973 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
975 struct virtio_hw *hw = dev->data->dev_private;
976 uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
978 uint32_t frame_size = mtu + ether_hdr_len;
979 uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
981 max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
983 if (mtu < RTE_ETHER_MIN_MTU || frame_size > max_frame_size) {
984 PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
985 RTE_ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
989 if (!virtio_check_scatter_on_all_rx_queues(dev, frame_size)) {
990 PMD_INIT_LOG(ERR, "MTU vs Rx scatter and Rx buffers check failed");
994 hw->max_rx_pkt_len = frame_size;
1000 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
1002 struct virtio_hw *hw = dev->data->dev_private;
1003 struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
1004 struct virtqueue *vq = virtnet_rxq_to_vq(rxvq);
1006 virtqueue_enable_intr(vq);
1007 virtio_mb(hw->weak_barriers);
1012 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
1014 struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
1015 struct virtqueue *vq = virtnet_rxq_to_vq(rxvq);
1017 virtqueue_disable_intr(vq);
1022 * dev_ops for virtio, bare necessities for basic operation
1024 static const struct eth_dev_ops virtio_eth_dev_ops = {
1025 .dev_configure = virtio_dev_configure,
1026 .dev_start = virtio_dev_start,
1027 .dev_stop = virtio_dev_stop,
1028 .dev_close = virtio_dev_close,
1029 .promiscuous_enable = virtio_dev_promiscuous_enable,
1030 .promiscuous_disable = virtio_dev_promiscuous_disable,
1031 .allmulticast_enable = virtio_dev_allmulticast_enable,
1032 .allmulticast_disable = virtio_dev_allmulticast_disable,
1033 .mtu_set = virtio_mtu_set,
1034 .dev_infos_get = virtio_dev_info_get,
1035 .stats_get = virtio_dev_stats_get,
1036 .xstats_get = virtio_dev_xstats_get,
1037 .xstats_get_names = virtio_dev_xstats_get_names,
1038 .stats_reset = virtio_dev_stats_reset,
1039 .xstats_reset = virtio_dev_stats_reset,
1040 .link_update = virtio_dev_link_update,
1041 .vlan_offload_set = virtio_dev_vlan_offload_set,
1042 .rx_queue_setup = virtio_dev_rx_queue_setup,
1043 .rx_queue_intr_enable = virtio_dev_rx_queue_intr_enable,
1044 .rx_queue_intr_disable = virtio_dev_rx_queue_intr_disable,
1045 .tx_queue_setup = virtio_dev_tx_queue_setup,
1046 .rss_hash_update = virtio_dev_rss_hash_update,
1047 .rss_hash_conf_get = virtio_dev_rss_hash_conf_get,
1048 .reta_update = virtio_dev_rss_reta_update,
1049 .reta_query = virtio_dev_rss_reta_query,
1050 /* collect stats per queue */
1051 .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
1052 .vlan_filter_set = virtio_vlan_filter_set,
1053 .mac_addr_add = virtio_mac_addr_add,
1054 .mac_addr_remove = virtio_mac_addr_remove,
1055 .mac_addr_set = virtio_mac_addr_set,
1056 .get_monitor_addr = virtio_get_monitor_addr,
1060 * dev_ops for virtio-user in secondary processes, as we just have
1061 * some limited supports currently.
1063 const struct eth_dev_ops virtio_user_secondary_eth_dev_ops = {
1064 .dev_infos_get = virtio_dev_info_get,
1065 .stats_get = virtio_dev_stats_get,
1066 .xstats_get = virtio_dev_xstats_get,
1067 .xstats_get_names = virtio_dev_xstats_get_names,
1068 .stats_reset = virtio_dev_stats_reset,
1069 .xstats_reset = virtio_dev_stats_reset,
1070 /* collect stats per queue */
1071 .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
1075 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1079 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1080 const struct virtnet_tx *txvq = dev->data->tx_queues[i];
1084 stats->opackets += txvq->stats.packets;
1085 stats->obytes += txvq->stats.bytes;
1087 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
1088 stats->q_opackets[i] = txvq->stats.packets;
1089 stats->q_obytes[i] = txvq->stats.bytes;
1093 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1094 const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1098 stats->ipackets += rxvq->stats.packets;
1099 stats->ibytes += rxvq->stats.bytes;
1100 stats->ierrors += rxvq->stats.errors;
1102 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
1103 stats->q_ipackets[i] = rxvq->stats.packets;
1104 stats->q_ibytes[i] = rxvq->stats.bytes;
1108 stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
1111 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
1112 struct rte_eth_xstat_name *xstats_names,
1113 __rte_unused unsigned limit)
1119 unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
1120 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
1122 if (xstats_names != NULL) {
1123 /* Note: limit checked in rte_eth_xstats_names() */
1125 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1126 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1129 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1130 snprintf(xstats_names[count].name,
1131 sizeof(xstats_names[count].name),
1133 rte_virtio_rxq_stat_strings[t].name);
1138 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1139 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1142 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1143 snprintf(xstats_names[count].name,
1144 sizeof(xstats_names[count].name),
1146 rte_virtio_txq_stat_strings[t].name);
1156 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
1162 unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
1163 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
1168 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1169 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1176 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1177 xstats[count].value = *(uint64_t *)(((char *)rxvq) +
1178 rte_virtio_rxq_stat_strings[t].offset);
1179 xstats[count].id = count;
1184 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1185 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1192 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1193 xstats[count].value = *(uint64_t *)(((char *)txvq) +
1194 rte_virtio_txq_stat_strings[t].offset);
1195 xstats[count].id = count;
1204 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1206 virtio_update_stats(dev, stats);
1212 virtio_dev_stats_reset(struct rte_eth_dev *dev)
1216 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1217 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1221 txvq->stats.packets = 0;
1222 txvq->stats.bytes = 0;
1223 txvq->stats.multicast = 0;
1224 txvq->stats.broadcast = 0;
1225 memset(txvq->stats.size_bins, 0,
1226 sizeof(txvq->stats.size_bins[0]) * 8);
1229 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1230 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1234 rxvq->stats.packets = 0;
1235 rxvq->stats.bytes = 0;
1236 rxvq->stats.errors = 0;
1237 rxvq->stats.multicast = 0;
1238 rxvq->stats.broadcast = 0;
1239 memset(rxvq->stats.size_bins, 0,
1240 sizeof(rxvq->stats.size_bins[0]) * 8);
1247 virtio_set_hwaddr(struct virtio_hw *hw)
1249 virtio_write_dev_config(hw,
1250 offsetof(struct virtio_net_config, mac),
1251 &hw->mac_addr, RTE_ETHER_ADDR_LEN);
1255 virtio_get_hwaddr(struct virtio_hw *hw)
1257 if (virtio_with_feature(hw, VIRTIO_NET_F_MAC)) {
1258 virtio_read_dev_config(hw,
1259 offsetof(struct virtio_net_config, mac),
1260 &hw->mac_addr, RTE_ETHER_ADDR_LEN);
1262 rte_eth_random_addr(&hw->mac_addr[0]);
1263 virtio_set_hwaddr(hw);
1268 virtio_mac_table_set(struct virtio_hw *hw,
1269 const struct virtio_net_ctrl_mac *uc,
1270 const struct virtio_net_ctrl_mac *mc)
1272 struct virtio_pmd_ctrl ctrl;
1275 if (!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1276 PMD_DRV_LOG(INFO, "host does not support mac table");
1280 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1281 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1283 len[0] = uc->entries * RTE_ETHER_ADDR_LEN + sizeof(uc->entries);
1284 memcpy(ctrl.data, uc, len[0]);
1286 len[1] = mc->entries * RTE_ETHER_ADDR_LEN + sizeof(mc->entries);
1287 memcpy(ctrl.data + len[0], mc, len[1]);
1289 err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1291 PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1296 virtio_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
1297 uint32_t index, uint32_t vmdq __rte_unused)
1299 struct virtio_hw *hw = dev->data->dev_private;
1300 const struct rte_ether_addr *addrs = dev->data->mac_addrs;
1302 struct virtio_net_ctrl_mac *uc, *mc;
1304 if (index >= VIRTIO_MAX_MAC_ADDRS) {
1305 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1309 uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1310 sizeof(uc->entries));
1312 mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1313 sizeof(mc->entries));
1316 for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1317 const struct rte_ether_addr *addr
1318 = (i == index) ? mac_addr : addrs + i;
1319 struct virtio_net_ctrl_mac *tbl
1320 = rte_is_multicast_ether_addr(addr) ? mc : uc;
1322 memcpy(&tbl->macs[tbl->entries++], addr, RTE_ETHER_ADDR_LEN);
1325 return virtio_mac_table_set(hw, uc, mc);
1329 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1331 struct virtio_hw *hw = dev->data->dev_private;
1332 struct rte_ether_addr *addrs = dev->data->mac_addrs;
1333 struct virtio_net_ctrl_mac *uc, *mc;
1336 if (index >= VIRTIO_MAX_MAC_ADDRS) {
1337 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1341 uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1342 sizeof(uc->entries));
1344 mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1345 sizeof(mc->entries));
1348 for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1349 struct virtio_net_ctrl_mac *tbl;
1351 if (i == index || rte_is_zero_ether_addr(addrs + i))
1354 tbl = rte_is_multicast_ether_addr(addrs + i) ? mc : uc;
1355 memcpy(&tbl->macs[tbl->entries++], addrs + i,
1356 RTE_ETHER_ADDR_LEN);
1359 virtio_mac_table_set(hw, uc, mc);
1363 virtio_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1365 struct virtio_hw *hw = dev->data->dev_private;
1367 memcpy(hw->mac_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1369 /* Use atomic update if available */
1370 if (virtio_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1371 struct virtio_pmd_ctrl ctrl;
1372 int len = RTE_ETHER_ADDR_LEN;
1374 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1375 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1377 memcpy(ctrl.data, mac_addr, RTE_ETHER_ADDR_LEN);
1378 return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1381 if (!virtio_with_feature(hw, VIRTIO_NET_F_MAC))
1384 virtio_set_hwaddr(hw);
1388 #define CLB_VAL_IDX 0
1389 #define CLB_MSK_IDX 1
1390 #define CLB_MATCH_IDX 2
1392 virtio_monitor_callback(const uint64_t value,
1393 const uint64_t opaque[RTE_POWER_MONITOR_OPAQUE_SZ])
1395 const uint64_t m = opaque[CLB_MSK_IDX];
1396 const uint64_t v = opaque[CLB_VAL_IDX];
1397 const uint64_t c = opaque[CLB_MATCH_IDX];
1400 return (value & m) == v ? -1 : 0;
1402 return (value & m) == v ? 0 : -1;
1406 virtio_get_monitor_addr(void *rx_queue, struct rte_power_monitor_cond *pmc)
1408 struct virtnet_rx *rxvq = rx_queue;
1409 struct virtqueue *vq = virtnet_rxq_to_vq(rxvq);
1410 struct virtio_hw *hw;
1416 if (virtio_with_packed_queue(hw)) {
1417 struct vring_packed_desc *desc;
1418 desc = vq->vq_packed.ring.desc;
1419 pmc->addr = &desc[vq->vq_used_cons_idx].flags;
1420 if (vq->vq_packed.used_wrap_counter)
1421 pmc->opaque[CLB_VAL_IDX] =
1422 VRING_PACKED_DESC_F_AVAIL_USED;
1424 pmc->opaque[CLB_VAL_IDX] = 0;
1425 pmc->opaque[CLB_MSK_IDX] = VRING_PACKED_DESC_F_AVAIL_USED;
1426 pmc->opaque[CLB_MATCH_IDX] = 1;
1427 pmc->size = sizeof(desc[vq->vq_used_cons_idx].flags);
1429 pmc->addr = &vq->vq_split.ring.used->idx;
1430 pmc->opaque[CLB_VAL_IDX] = vq->vq_used_cons_idx
1431 & (vq->vq_nentries - 1);
1432 pmc->opaque[CLB_MSK_IDX] = vq->vq_nentries - 1;
1433 pmc->opaque[CLB_MATCH_IDX] = 0;
1434 pmc->size = sizeof(vq->vq_split.ring.used->idx);
1436 pmc->fn = virtio_monitor_callback;
1442 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1444 struct virtio_hw *hw = dev->data->dev_private;
1445 struct virtio_pmd_ctrl ctrl;
1448 if (!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1451 ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1452 ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1453 memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1454 len = sizeof(vlan_id);
1456 return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1460 virtio_intr_unmask(struct rte_eth_dev *dev)
1462 struct virtio_hw *hw = dev->data->dev_private;
1464 if (rte_intr_ack(dev->intr_handle) < 0)
1467 if (VIRTIO_OPS(hw)->intr_detect)
1468 VIRTIO_OPS(hw)->intr_detect(hw);
1474 virtio_intr_enable(struct rte_eth_dev *dev)
1476 struct virtio_hw *hw = dev->data->dev_private;
1478 if (rte_intr_enable(dev->intr_handle) < 0)
1481 if (VIRTIO_OPS(hw)->intr_detect)
1482 VIRTIO_OPS(hw)->intr_detect(hw);
1488 virtio_intr_disable(struct rte_eth_dev *dev)
1490 struct virtio_hw *hw = dev->data->dev_private;
1492 if (rte_intr_disable(dev->intr_handle) < 0)
1495 if (VIRTIO_OPS(hw)->intr_detect)
1496 VIRTIO_OPS(hw)->intr_detect(hw);
1502 virtio_ethdev_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1504 uint64_t host_features;
1506 /* Prepare guest_features: feature that driver wants to support */
1507 PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1510 /* Read device(host) feature bits */
1511 host_features = VIRTIO_OPS(hw)->get_features(hw);
1512 PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1515 /* If supported, ensure MTU value is valid before acknowledging it. */
1516 if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1517 struct virtio_net_config config;
1519 virtio_read_dev_config(hw,
1520 offsetof(struct virtio_net_config, mtu),
1521 &config.mtu, sizeof(config.mtu));
1523 if (config.mtu < RTE_ETHER_MIN_MTU)
1524 req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1528 * Negotiate features: Subset of device feature bits are written back
1529 * guest feature bits.
1531 hw->guest_features = req_features;
1532 hw->guest_features = virtio_negotiate_features(hw, host_features);
1533 PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1534 hw->guest_features);
1536 if (VIRTIO_OPS(hw)->features_ok(hw) < 0)
1539 if (virtio_with_feature(hw, VIRTIO_F_VERSION_1)) {
1540 virtio_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1542 if (!(virtio_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1543 PMD_INIT_LOG(ERR, "Failed to set FEATURES_OK status!");
1548 hw->req_guest_features = req_features;
1554 virtio_dev_pause(struct rte_eth_dev *dev)
1556 struct virtio_hw *hw = dev->data->dev_private;
1558 rte_spinlock_lock(&hw->state_lock);
1560 if (hw->started == 0) {
1561 /* Device is just stopped. */
1562 rte_spinlock_unlock(&hw->state_lock);
1567 * Prevent the worker threads from touching queues to avoid contention,
1568 * 1 ms should be enough for the ongoing Tx function to finish.
1575 * Recover hw state to let the worker threads continue.
1578 virtio_dev_resume(struct rte_eth_dev *dev)
1580 struct virtio_hw *hw = dev->data->dev_private;
1583 rte_spinlock_unlock(&hw->state_lock);
1587 * Should be called only after device is paused.
1590 virtio_inject_pkts(struct rte_eth_dev *dev, struct rte_mbuf **tx_pkts,
1593 struct virtio_hw *hw = dev->data->dev_private;
1594 struct virtnet_tx *txvq = dev->data->tx_queues[0];
1597 hw->inject_pkts = tx_pkts;
1598 ret = dev->tx_pkt_burst(txvq, tx_pkts, nb_pkts);
1599 hw->inject_pkts = NULL;
1605 virtio_notify_peers(struct rte_eth_dev *dev)
1607 struct virtio_hw *hw = dev->data->dev_private;
1608 struct virtnet_rx *rxvq;
1609 struct rte_mbuf *rarp_mbuf;
1611 if (!dev->data->rx_queues)
1614 rxvq = dev->data->rx_queues[0];
1618 rarp_mbuf = rte_net_make_rarp_packet(rxvq->mpool,
1619 (struct rte_ether_addr *)hw->mac_addr);
1620 if (rarp_mbuf == NULL) {
1621 PMD_DRV_LOG(ERR, "failed to make RARP packet.");
1625 /* If virtio port just stopped, no need to send RARP */
1626 if (virtio_dev_pause(dev) < 0) {
1627 rte_pktmbuf_free(rarp_mbuf);
1631 virtio_inject_pkts(dev, &rarp_mbuf, 1);
1632 virtio_dev_resume(dev);
1636 virtio_ack_link_announce(struct rte_eth_dev *dev)
1638 struct virtio_hw *hw = dev->data->dev_private;
1639 struct virtio_pmd_ctrl ctrl;
1641 ctrl.hdr.class = VIRTIO_NET_CTRL_ANNOUNCE;
1642 ctrl.hdr.cmd = VIRTIO_NET_CTRL_ANNOUNCE_ACK;
1644 virtio_send_command(hw->cvq, &ctrl, NULL, 0);
1648 * Process virtio config changed interrupt. Call the callback
1649 * if link state changed, generate gratuitous RARP packet if
1650 * the status indicates an ANNOUNCE.
1653 virtio_interrupt_handler(void *param)
1655 struct rte_eth_dev *dev = param;
1656 struct virtio_hw *hw = dev->data->dev_private;
1660 /* Read interrupt status which clears interrupt */
1661 isr = virtio_get_isr(hw);
1662 PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1664 if (virtio_intr_unmask(dev) < 0)
1665 PMD_DRV_LOG(ERR, "interrupt enable failed");
1667 if (isr & VIRTIO_ISR_CONFIG) {
1668 if (virtio_dev_link_update(dev, 0) == 0)
1669 rte_eth_dev_callback_process(dev,
1670 RTE_ETH_EVENT_INTR_LSC,
1673 if (virtio_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1674 virtio_read_dev_config(hw,
1675 offsetof(struct virtio_net_config, status),
1676 &status, sizeof(status));
1677 if (status & VIRTIO_NET_S_ANNOUNCE) {
1678 virtio_notify_peers(dev);
1680 virtio_ack_link_announce(dev);
1686 /* set rx and tx handlers according to what is supported */
1688 set_rxtx_funcs(struct rte_eth_dev *eth_dev)
1690 struct virtio_hw *hw = eth_dev->data->dev_private;
1692 eth_dev->tx_pkt_prepare = virtio_xmit_pkts_prepare;
1693 if (virtio_with_packed_queue(hw)) {
1695 "virtio: using packed ring %s Tx path on port %u",
1696 hw->use_vec_tx ? "vectorized" : "standard",
1697 eth_dev->data->port_id);
1699 eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed_vec;
1701 eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed;
1703 if (hw->use_inorder_tx) {
1704 PMD_INIT_LOG(INFO, "virtio: using inorder Tx path on port %u",
1705 eth_dev->data->port_id);
1706 eth_dev->tx_pkt_burst = virtio_xmit_pkts_inorder;
1708 PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
1709 eth_dev->data->port_id);
1710 eth_dev->tx_pkt_burst = virtio_xmit_pkts;
1714 if (virtio_with_packed_queue(hw)) {
1715 if (hw->use_vec_rx) {
1717 "virtio: using packed ring vectorized Rx path on port %u",
1718 eth_dev->data->port_id);
1719 eth_dev->rx_pkt_burst =
1720 &virtio_recv_pkts_packed_vec;
1721 } else if (virtio_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1723 "virtio: using packed ring mergeable buffer Rx path on port %u",
1724 eth_dev->data->port_id);
1725 eth_dev->rx_pkt_burst =
1726 &virtio_recv_mergeable_pkts_packed;
1729 "virtio: using packed ring standard Rx path on port %u",
1730 eth_dev->data->port_id);
1731 eth_dev->rx_pkt_burst = &virtio_recv_pkts_packed;
1734 if (hw->use_vec_rx) {
1735 PMD_INIT_LOG(INFO, "virtio: using vectorized Rx path on port %u",
1736 eth_dev->data->port_id);
1737 eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1738 } else if (hw->use_inorder_rx) {
1740 "virtio: using inorder Rx path on port %u",
1741 eth_dev->data->port_id);
1742 eth_dev->rx_pkt_burst = &virtio_recv_pkts_inorder;
1743 } else if (virtio_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1745 "virtio: using mergeable buffer Rx path on port %u",
1746 eth_dev->data->port_id);
1747 eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1749 PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
1750 eth_dev->data->port_id);
1751 eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1757 /* Only support 1:1 queue/interrupt mapping so far.
1758 * TODO: support n:1 queue/interrupt mapping when there are limited number of
1759 * interrupt vectors (<N+1).
1762 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1765 struct virtio_hw *hw = dev->data->dev_private;
1767 PMD_INIT_LOG(INFO, "queue/interrupt binding");
1768 for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1769 if (rte_intr_vec_list_index_set(dev->intr_handle, i,
1772 if (VIRTIO_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1773 VIRTIO_MSI_NO_VECTOR) {
1774 PMD_DRV_LOG(ERR, "failed to set queue vector");
1783 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1786 struct virtio_hw *hw = dev->data->dev_private;
1788 PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1789 for (i = 0; i < dev->data->nb_rx_queues; ++i)
1790 VIRTIO_OPS(hw)->set_queue_irq(hw,
1791 hw->vqs[i * VTNET_CQ],
1792 VIRTIO_MSI_NO_VECTOR);
1796 virtio_configure_intr(struct rte_eth_dev *dev)
1798 struct virtio_hw *hw = dev->data->dev_private;
1800 if (!rte_intr_cap_multiple(dev->intr_handle)) {
1801 PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1805 if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1806 PMD_INIT_LOG(ERR, "Fail to create eventfd");
1810 if (rte_intr_vec_list_alloc(dev->intr_handle, "intr_vec",
1811 hw->max_queue_pairs)) {
1812 PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1813 hw->max_queue_pairs);
1817 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
1818 /* Re-register callback to update max_intr */
1819 rte_intr_callback_unregister(dev->intr_handle,
1820 virtio_interrupt_handler,
1822 rte_intr_callback_register(dev->intr_handle,
1823 virtio_interrupt_handler,
1827 /* DO NOT try to remove this! This function will enable msix, or QEMU
1828 * will encounter SIGSEGV when DRIVER_OK is sent.
1829 * And for legacy devices, this should be done before queue/vec binding
1830 * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1831 * (22) will be ignored.
1833 if (virtio_intr_enable(dev) < 0) {
1834 PMD_DRV_LOG(ERR, "interrupt enable failed");
1838 if (virtio_queues_bind_intr(dev) < 0) {
1839 PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1847 virtio_get_speed_duplex(struct rte_eth_dev *eth_dev,
1848 struct rte_eth_link *link)
1850 struct virtio_hw *hw = eth_dev->data->dev_private;
1851 struct virtio_net_config *config;
1852 struct virtio_net_config local_config;
1854 config = &local_config;
1855 virtio_read_dev_config(hw,
1856 offsetof(struct virtio_net_config, speed),
1857 &config->speed, sizeof(config->speed));
1858 virtio_read_dev_config(hw,
1859 offsetof(struct virtio_net_config, duplex),
1860 &config->duplex, sizeof(config->duplex));
1861 hw->speed = config->speed;
1862 hw->duplex = config->duplex;
1864 link->link_duplex = hw->duplex;
1865 link->link_speed = hw->speed;
1867 PMD_INIT_LOG(DEBUG, "link speed = %d, duplex = %d",
1868 hw->speed, hw->duplex);
1872 ethdev_to_virtio_rss_offloads(uint64_t ethdev_hash_types)
1874 uint64_t virtio_hash_types = 0;
1876 if (ethdev_hash_types & (RTE_ETH_RSS_IPV4 | RTE_ETH_RSS_FRAG_IPV4 |
1877 RTE_ETH_RSS_NONFRAG_IPV4_OTHER))
1878 virtio_hash_types |= VIRTIO_NET_HASH_TYPE_IPV4;
1880 if (ethdev_hash_types & RTE_ETH_RSS_NONFRAG_IPV4_TCP)
1881 virtio_hash_types |= VIRTIO_NET_HASH_TYPE_TCPV4;
1883 if (ethdev_hash_types & RTE_ETH_RSS_NONFRAG_IPV4_UDP)
1884 virtio_hash_types |= VIRTIO_NET_HASH_TYPE_UDPV4;
1886 if (ethdev_hash_types & (RTE_ETH_RSS_IPV6 | RTE_ETH_RSS_FRAG_IPV6 |
1887 RTE_ETH_RSS_NONFRAG_IPV6_OTHER))
1888 virtio_hash_types |= VIRTIO_NET_HASH_TYPE_IPV6;
1890 if (ethdev_hash_types & RTE_ETH_RSS_NONFRAG_IPV6_TCP)
1891 virtio_hash_types |= VIRTIO_NET_HASH_TYPE_TCPV6;
1893 if (ethdev_hash_types & RTE_ETH_RSS_NONFRAG_IPV6_UDP)
1894 virtio_hash_types |= VIRTIO_NET_HASH_TYPE_UDPV6;
1896 if (ethdev_hash_types & RTE_ETH_RSS_IPV6_EX)
1897 virtio_hash_types |= VIRTIO_NET_HASH_TYPE_IP_EX;
1899 if (ethdev_hash_types & RTE_ETH_RSS_IPV6_TCP_EX)
1900 virtio_hash_types |= VIRTIO_NET_HASH_TYPE_TCP_EX;
1902 if (ethdev_hash_types & RTE_ETH_RSS_IPV6_UDP_EX)
1903 virtio_hash_types |= VIRTIO_NET_HASH_TYPE_UDP_EX;
1905 return virtio_hash_types;
1909 virtio_to_ethdev_rss_offloads(uint64_t virtio_hash_types)
1911 uint64_t rss_offloads = 0;
1913 if (virtio_hash_types & VIRTIO_NET_HASH_TYPE_IPV4)
1914 rss_offloads |= RTE_ETH_RSS_IPV4 | RTE_ETH_RSS_FRAG_IPV4 |
1915 RTE_ETH_RSS_NONFRAG_IPV4_OTHER;
1917 if (virtio_hash_types & VIRTIO_NET_HASH_TYPE_TCPV4)
1918 rss_offloads |= RTE_ETH_RSS_NONFRAG_IPV4_TCP;
1920 if (virtio_hash_types & VIRTIO_NET_HASH_TYPE_UDPV4)
1921 rss_offloads |= RTE_ETH_RSS_NONFRAG_IPV4_UDP;
1923 if (virtio_hash_types & VIRTIO_NET_HASH_TYPE_IPV6)
1924 rss_offloads |= RTE_ETH_RSS_IPV6 | RTE_ETH_RSS_FRAG_IPV6 |
1925 RTE_ETH_RSS_NONFRAG_IPV6_OTHER;
1927 if (virtio_hash_types & VIRTIO_NET_HASH_TYPE_TCPV6)
1928 rss_offloads |= RTE_ETH_RSS_NONFRAG_IPV6_TCP;
1930 if (virtio_hash_types & VIRTIO_NET_HASH_TYPE_UDPV6)
1931 rss_offloads |= RTE_ETH_RSS_NONFRAG_IPV6_UDP;
1933 if (virtio_hash_types & VIRTIO_NET_HASH_TYPE_IP_EX)
1934 rss_offloads |= RTE_ETH_RSS_IPV6_EX;
1936 if (virtio_hash_types & VIRTIO_NET_HASH_TYPE_TCP_EX)
1937 rss_offloads |= RTE_ETH_RSS_IPV6_TCP_EX;
1939 if (virtio_hash_types & VIRTIO_NET_HASH_TYPE_UDP_EX)
1940 rss_offloads |= RTE_ETH_RSS_IPV6_UDP_EX;
1942 return rss_offloads;
1946 virtio_dev_get_rss_config(struct virtio_hw *hw, uint32_t *rss_hash_types)
1948 struct virtio_net_config local_config;
1949 struct virtio_net_config *config = &local_config;
1951 virtio_read_dev_config(hw,
1952 offsetof(struct virtio_net_config, rss_max_key_size),
1953 &config->rss_max_key_size,
1954 sizeof(config->rss_max_key_size));
1955 if (config->rss_max_key_size < VIRTIO_NET_RSS_KEY_SIZE) {
1956 PMD_INIT_LOG(ERR, "Invalid device RSS max key size (%u)",
1957 config->rss_max_key_size);
1961 virtio_read_dev_config(hw,
1962 offsetof(struct virtio_net_config,
1963 rss_max_indirection_table_length),
1964 &config->rss_max_indirection_table_length,
1965 sizeof(config->rss_max_indirection_table_length));
1966 if (config->rss_max_indirection_table_length < VIRTIO_NET_RSS_RETA_SIZE) {
1967 PMD_INIT_LOG(ERR, "Invalid device RSS max reta size (%u)",
1968 config->rss_max_indirection_table_length);
1972 virtio_read_dev_config(hw,
1973 offsetof(struct virtio_net_config, supported_hash_types),
1974 &config->supported_hash_types,
1975 sizeof(config->supported_hash_types));
1976 if ((config->supported_hash_types & VIRTIO_NET_HASH_TYPE_MASK) == 0) {
1977 PMD_INIT_LOG(ERR, "Invalid device RSS hash types (0x%x)",
1978 config->supported_hash_types);
1982 *rss_hash_types = config->supported_hash_types & VIRTIO_NET_HASH_TYPE_MASK;
1984 PMD_INIT_LOG(DEBUG, "Device RSS config:");
1985 PMD_INIT_LOG(DEBUG, "\t-Max key size: %u", config->rss_max_key_size);
1986 PMD_INIT_LOG(DEBUG, "\t-Max reta size: %u", config->rss_max_indirection_table_length);
1987 PMD_INIT_LOG(DEBUG, "\t-Supported hash types: 0x%x", *rss_hash_types);
1993 virtio_dev_rss_hash_update(struct rte_eth_dev *dev,
1994 struct rte_eth_rss_conf *rss_conf)
1996 struct virtio_hw *hw = dev->data->dev_private;
1997 char old_rss_key[VIRTIO_NET_RSS_KEY_SIZE];
1998 uint32_t old_hash_types;
2002 if (!virtio_with_feature(hw, VIRTIO_NET_F_RSS))
2005 if (rss_conf->rss_hf & ~virtio_to_ethdev_rss_offloads(VIRTIO_NET_HASH_TYPE_MASK))
2008 old_hash_types = hw->rss_hash_types;
2009 hw->rss_hash_types = ethdev_to_virtio_rss_offloads(rss_conf->rss_hf);
2011 if (rss_conf->rss_key && rss_conf->rss_key_len) {
2012 if (rss_conf->rss_key_len != VIRTIO_NET_RSS_KEY_SIZE) {
2013 PMD_INIT_LOG(ERR, "Driver only supports %u RSS key length",
2014 VIRTIO_NET_RSS_KEY_SIZE);
2018 memcpy(old_rss_key, hw->rss_key, VIRTIO_NET_RSS_KEY_SIZE);
2019 memcpy(hw->rss_key, rss_conf->rss_key, VIRTIO_NET_RSS_KEY_SIZE);
2022 nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
2023 ret = virtio_set_multiple_queues_rss(dev, nb_queues);
2025 PMD_INIT_LOG(ERR, "Failed to apply new RSS config to the device");
2031 if (rss_conf->rss_key && rss_conf->rss_key_len)
2032 memcpy(hw->rss_key, old_rss_key, VIRTIO_NET_RSS_KEY_SIZE);
2034 hw->rss_hash_types = old_hash_types;
2040 virtio_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
2041 struct rte_eth_rss_conf *rss_conf)
2043 struct virtio_hw *hw = dev->data->dev_private;
2045 if (!virtio_with_feature(hw, VIRTIO_NET_F_RSS))
2048 if (rss_conf->rss_key && rss_conf->rss_key_len >= VIRTIO_NET_RSS_KEY_SIZE)
2049 memcpy(rss_conf->rss_key, hw->rss_key, VIRTIO_NET_RSS_KEY_SIZE);
2050 rss_conf->rss_key_len = VIRTIO_NET_RSS_KEY_SIZE;
2051 rss_conf->rss_hf = virtio_to_ethdev_rss_offloads(hw->rss_hash_types);
2056 static int virtio_dev_rss_reta_update(struct rte_eth_dev *dev,
2057 struct rte_eth_rss_reta_entry64 *reta_conf,
2060 struct virtio_hw *hw = dev->data->dev_private;
2062 uint16_t old_reta[VIRTIO_NET_RSS_RETA_SIZE];
2063 int idx, pos, i, ret;
2065 if (!virtio_with_feature(hw, VIRTIO_NET_F_RSS))
2068 if (reta_size != VIRTIO_NET_RSS_RETA_SIZE)
2071 memcpy(old_reta, hw->rss_reta, sizeof(old_reta));
2073 for (i = 0; i < reta_size; i++) {
2074 idx = i / RTE_ETH_RETA_GROUP_SIZE;
2075 pos = i % RTE_ETH_RETA_GROUP_SIZE;
2077 if (((reta_conf[idx].mask >> pos) & 0x1) == 0)
2080 hw->rss_reta[i] = reta_conf[idx].reta[pos];
2083 nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
2084 ret = virtio_set_multiple_queues_rss(dev, nb_queues);
2086 PMD_INIT_LOG(ERR, "Failed to apply new RETA to the device");
2087 memcpy(hw->rss_reta, old_reta, sizeof(old_reta));
2090 hw->rss_rx_queues = dev->data->nb_rx_queues;
2095 static int virtio_dev_rss_reta_query(struct rte_eth_dev *dev,
2096 struct rte_eth_rss_reta_entry64 *reta_conf,
2099 struct virtio_hw *hw = dev->data->dev_private;
2102 if (!virtio_with_feature(hw, VIRTIO_NET_F_RSS))
2105 if (reta_size != VIRTIO_NET_RSS_RETA_SIZE)
2108 for (i = 0; i < reta_size; i++) {
2109 idx = i / RTE_ETH_RETA_GROUP_SIZE;
2110 reta_conf[idx].reta[i % RTE_ETH_RETA_GROUP_SIZE] = hw->rss_reta[i];
2117 * As default RSS hash key, it uses the default key of the
2118 * Intel IXGBE devices. It can be updated by the application
2119 * with any 40B key value.
2121 static uint8_t rss_intel_key[VIRTIO_NET_RSS_KEY_SIZE] = {
2122 0x6D, 0x5A, 0x56, 0xDA, 0x25, 0x5B, 0x0E, 0xC2,
2123 0x41, 0x67, 0x25, 0x3D, 0x43, 0xA3, 0x8F, 0xB0,
2124 0xD0, 0xCA, 0x2B, 0xCB, 0xAE, 0x7B, 0x30, 0xB4,
2125 0x77, 0xCB, 0x2D, 0xA3, 0x80, 0x30, 0xF2, 0x0C,
2126 0x6A, 0x42, 0xB7, 0x3B, 0xBE, 0xAC, 0x01, 0xFA,
2130 virtio_dev_rss_init(struct rte_eth_dev *eth_dev)
2132 struct virtio_hw *hw = eth_dev->data->dev_private;
2133 uint16_t nb_rx_queues = eth_dev->data->nb_rx_queues;
2134 struct rte_eth_rss_conf *rss_conf;
2137 if (!nb_rx_queues) {
2138 PMD_INIT_LOG(ERR, "Cannot init RSS if no Rx queues");
2142 rss_conf = ð_dev->data->dev_conf.rx_adv_conf.rss_conf;
2144 ret = virtio_dev_get_rss_config(hw, &hw->rss_hash_types);
2148 if (rss_conf->rss_hf) {
2149 /* Ensure requested hash types are supported by the device */
2150 if (rss_conf->rss_hf & ~virtio_to_ethdev_rss_offloads(hw->rss_hash_types))
2153 hw->rss_hash_types = ethdev_to_virtio_rss_offloads(rss_conf->rss_hf);
2157 /* Setup default RSS key if not already setup by the user */
2158 hw->rss_key = rte_malloc_socket("rss_key",
2159 VIRTIO_NET_RSS_KEY_SIZE, 0,
2160 eth_dev->device->numa_node);
2162 PMD_INIT_LOG(ERR, "Failed to allocate RSS key");
2167 if (rss_conf->rss_key && rss_conf->rss_key_len) {
2168 if (rss_conf->rss_key_len != VIRTIO_NET_RSS_KEY_SIZE) {
2169 PMD_INIT_LOG(ERR, "Driver only supports %u RSS key length",
2170 VIRTIO_NET_RSS_KEY_SIZE);
2173 memcpy(hw->rss_key, rss_conf->rss_key, VIRTIO_NET_RSS_KEY_SIZE);
2175 memcpy(hw->rss_key, rss_intel_key, VIRTIO_NET_RSS_KEY_SIZE);
2178 if (!hw->rss_reta) {
2179 /* Setup default RSS reta if not already setup by the user */
2180 hw->rss_reta = rte_zmalloc_socket("rss_reta",
2181 VIRTIO_NET_RSS_RETA_SIZE * sizeof(uint16_t), 0,
2182 eth_dev->device->numa_node);
2183 if (!hw->rss_reta) {
2184 PMD_INIT_LOG(ERR, "Failed to allocate RSS reta");
2188 hw->rss_rx_queues = 0;
2191 /* Re-initialize the RSS reta if the number of RX queues has changed */
2192 if (hw->rss_rx_queues != nb_rx_queues) {
2193 for (i = 0; i < VIRTIO_NET_RSS_RETA_SIZE; i++)
2194 hw->rss_reta[i] = i % nb_rx_queues;
2195 hw->rss_rx_queues = nb_rx_queues;
2201 #define DUPLEX_UNKNOWN 0xff
2202 /* reset device and renegotiate features if needed */
2204 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
2206 struct virtio_hw *hw = eth_dev->data->dev_private;
2207 struct virtio_net_config *config;
2208 struct virtio_net_config local_config;
2211 /* Reset the device although not necessary at startup */
2215 virtio_dev_free_mbufs(eth_dev);
2216 virtio_free_queues(hw);
2219 /* Tell the host we've noticed this device. */
2220 virtio_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
2222 /* Tell the host we've known how to drive the device. */
2223 virtio_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
2224 if (virtio_ethdev_negotiate_features(hw, req_features) < 0)
2227 hw->weak_barriers = !virtio_with_feature(hw, VIRTIO_F_ORDER_PLATFORM);
2229 /* If host does not support both status and MSI-X then disable LSC */
2230 if (virtio_with_feature(hw, VIRTIO_NET_F_STATUS) && hw->intr_lsc)
2231 eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
2233 eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
2235 eth_dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
2237 /* Setting up rx_header size for the device */
2238 if (virtio_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
2239 virtio_with_feature(hw, VIRTIO_F_VERSION_1) ||
2240 virtio_with_packed_queue(hw))
2241 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
2243 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
2245 /* Copy the permanent MAC address to: virtio_hw */
2246 virtio_get_hwaddr(hw);
2247 rte_ether_addr_copy((struct rte_ether_addr *)hw->mac_addr,
2248 ð_dev->data->mac_addrs[0]);
2250 "PORT MAC: " RTE_ETHER_ADDR_PRT_FMT,
2251 hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
2252 hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
2254 hw->get_speed_via_feat = hw->speed == RTE_ETH_SPEED_NUM_UNKNOWN &&
2255 virtio_with_feature(hw, VIRTIO_NET_F_SPEED_DUPLEX);
2256 if (hw->get_speed_via_feat)
2257 virtio_get_speed_duplex(eth_dev, NULL);
2258 if (hw->duplex == DUPLEX_UNKNOWN)
2259 hw->duplex = RTE_ETH_LINK_FULL_DUPLEX;
2260 PMD_INIT_LOG(DEBUG, "link speed = %d, duplex = %d",
2261 hw->speed, hw->duplex);
2262 if (virtio_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
2263 config = &local_config;
2265 virtio_read_dev_config(hw,
2266 offsetof(struct virtio_net_config, mac),
2267 &config->mac, sizeof(config->mac));
2269 if (virtio_with_feature(hw, VIRTIO_NET_F_STATUS)) {
2270 virtio_read_dev_config(hw,
2271 offsetof(struct virtio_net_config, status),
2272 &config->status, sizeof(config->status));
2275 "VIRTIO_NET_F_STATUS is not supported");
2279 if (virtio_with_feature(hw, VIRTIO_NET_F_MQ) ||
2280 virtio_with_feature(hw, VIRTIO_NET_F_RSS)) {
2281 virtio_read_dev_config(hw,
2282 offsetof(struct virtio_net_config, max_virtqueue_pairs),
2283 &config->max_virtqueue_pairs,
2284 sizeof(config->max_virtqueue_pairs));
2287 "Neither VIRTIO_NET_F_MQ nor VIRTIO_NET_F_RSS are supported");
2288 config->max_virtqueue_pairs = 1;
2291 hw->max_queue_pairs = config->max_virtqueue_pairs;
2293 if (virtio_with_feature(hw, VIRTIO_NET_F_MTU)) {
2294 virtio_read_dev_config(hw,
2295 offsetof(struct virtio_net_config, mtu),
2297 sizeof(config->mtu));
2300 * MTU value has already been checked at negotiation
2301 * time, but check again in case it has changed since
2302 * then, which should not happen.
2304 if (config->mtu < RTE_ETHER_MIN_MTU) {
2305 PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
2310 hw->max_mtu = config->mtu;
2311 /* Set initial MTU to maximum one supported by vhost */
2312 eth_dev->data->mtu = config->mtu;
2315 hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
2316 VLAN_TAG_LEN - hw->vtnet_hdr_size;
2319 hw->rss_hash_types = 0;
2320 if (virtio_with_feature(hw, VIRTIO_NET_F_RSS))
2321 if (virtio_dev_rss_init(eth_dev))
2324 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
2325 config->max_virtqueue_pairs);
2326 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
2328 "PORT MAC: " RTE_ETHER_ADDR_PRT_FMT,
2329 config->mac[0], config->mac[1],
2330 config->mac[2], config->mac[3],
2331 config->mac[4], config->mac[5]);
2333 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
2334 hw->max_queue_pairs = 1;
2335 hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
2336 VLAN_TAG_LEN - hw->vtnet_hdr_size;
2339 ret = virtio_alloc_queues(eth_dev);
2343 if (eth_dev->data->dev_conf.intr_conf.rxq) {
2344 if (virtio_configure_intr(eth_dev) < 0) {
2345 PMD_INIT_LOG(ERR, "failed to configure interrupt");
2346 virtio_free_queues(hw);
2351 virtio_reinit_complete(hw);
2357 * This function is based on probe() function in virtio_pci.c
2358 * It returns 0 on success.
2361 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
2363 struct virtio_hw *hw = eth_dev->data->dev_private;
2364 uint32_t speed = RTE_ETH_SPEED_NUM_UNKNOWN;
2368 if (sizeof(struct virtio_net_hdr_mrg_rxbuf) > RTE_PKTMBUF_HEADROOM) {
2370 "Not sufficient headroom required = %d, avail = %d",
2371 (int)sizeof(struct virtio_net_hdr_mrg_rxbuf),
2372 RTE_PKTMBUF_HEADROOM);
2377 eth_dev->dev_ops = &virtio_eth_dev_ops;
2379 if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
2380 set_rxtx_funcs(eth_dev);
2384 ret = virtio_dev_devargs_parse(eth_dev->device->devargs, &speed, &vectorized);
2388 hw->duplex = DUPLEX_UNKNOWN;
2390 /* Allocate memory for storing MAC addresses */
2391 eth_dev->data->mac_addrs = rte_zmalloc("virtio",
2392 VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN, 0);
2393 if (eth_dev->data->mac_addrs == NULL) {
2395 "Failed to allocate %d bytes needed to store MAC addresses",
2396 VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN);
2400 rte_spinlock_init(&hw->state_lock);
2402 /* reset device and negotiate default features */
2403 ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
2405 goto err_virtio_init;
2408 if (!virtio_with_packed_queue(hw)) {
2411 #if defined(CC_AVX512_SUPPORT) || defined(RTE_ARCH_ARM)
2416 "building environment do not support packed ring vectorized");
2426 rte_free(eth_dev->data->mac_addrs);
2427 eth_dev->data->mac_addrs = NULL;
2432 virtio_dev_speed_capa_get(uint32_t speed)
2435 case RTE_ETH_SPEED_NUM_10G:
2436 return RTE_ETH_LINK_SPEED_10G;
2437 case RTE_ETH_SPEED_NUM_20G:
2438 return RTE_ETH_LINK_SPEED_20G;
2439 case RTE_ETH_SPEED_NUM_25G:
2440 return RTE_ETH_LINK_SPEED_25G;
2441 case RTE_ETH_SPEED_NUM_40G:
2442 return RTE_ETH_LINK_SPEED_40G;
2443 case RTE_ETH_SPEED_NUM_50G:
2444 return RTE_ETH_LINK_SPEED_50G;
2445 case RTE_ETH_SPEED_NUM_56G:
2446 return RTE_ETH_LINK_SPEED_56G;
2447 case RTE_ETH_SPEED_NUM_100G:
2448 return RTE_ETH_LINK_SPEED_100G;
2449 case RTE_ETH_SPEED_NUM_200G:
2450 return RTE_ETH_LINK_SPEED_200G;
2456 static int vectorized_check_handler(__rte_unused const char *key,
2457 const char *value, void *ret_val)
2459 if (strcmp(value, "1") == 0)
2460 *(int *)ret_val = 1;
2462 *(int *)ret_val = 0;
2467 #define VIRTIO_ARG_SPEED "speed"
2468 #define VIRTIO_ARG_VECTORIZED "vectorized"
2471 link_speed_handler(const char *key __rte_unused,
2472 const char *value, void *ret_val)
2475 if (!value || !ret_val)
2477 val = strtoul(value, NULL, 0);
2478 /* validate input */
2479 if (virtio_dev_speed_capa_get(val) == 0)
2481 *(uint32_t *)ret_val = val;
2488 virtio_dev_devargs_parse(struct rte_devargs *devargs, uint32_t *speed, int *vectorized)
2490 struct rte_kvargs *kvlist;
2493 if (devargs == NULL)
2496 kvlist = rte_kvargs_parse(devargs->args, NULL);
2497 if (kvlist == NULL) {
2498 PMD_INIT_LOG(ERR, "error when parsing param");
2502 if (speed && rte_kvargs_count(kvlist, VIRTIO_ARG_SPEED) == 1) {
2503 ret = rte_kvargs_process(kvlist,
2505 link_speed_handler, speed);
2507 PMD_INIT_LOG(ERR, "Failed to parse %s",
2514 rte_kvargs_count(kvlist, VIRTIO_ARG_VECTORIZED) == 1) {
2515 ret = rte_kvargs_process(kvlist,
2516 VIRTIO_ARG_VECTORIZED,
2517 vectorized_check_handler, vectorized);
2519 PMD_INIT_LOG(ERR, "Failed to parse %s",
2520 VIRTIO_ARG_VECTORIZED);
2526 rte_kvargs_free(kvlist);
2531 rx_offload_enabled(struct virtio_hw *hw)
2533 return virtio_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM) ||
2534 virtio_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2535 virtio_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6);
2539 tx_offload_enabled(struct virtio_hw *hw)
2541 return virtio_with_feature(hw, VIRTIO_NET_F_CSUM) ||
2542 virtio_with_feature(hw, VIRTIO_NET_F_HOST_TSO4) ||
2543 virtio_with_feature(hw, VIRTIO_NET_F_HOST_TSO6);
2547 * Configure virtio device
2548 * It returns 0 on success.
2551 virtio_dev_configure(struct rte_eth_dev *dev)
2553 const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2554 const struct rte_eth_txmode *txmode = &dev->data->dev_conf.txmode;
2555 struct virtio_hw *hw = dev->data->dev_private;
2556 uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
2558 uint64_t rx_offloads = rxmode->offloads;
2559 uint64_t tx_offloads = txmode->offloads;
2560 uint64_t req_features;
2563 PMD_INIT_LOG(DEBUG, "configure");
2564 req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
2566 if (rxmode->mq_mode != RTE_ETH_MQ_RX_NONE && rxmode->mq_mode != RTE_ETH_MQ_RX_RSS) {
2568 "Unsupported Rx multi queue mode %d",
2573 if (txmode->mq_mode != RTE_ETH_MQ_TX_NONE) {
2575 "Unsupported Tx multi queue mode %d",
2580 if (dev->data->dev_conf.intr_conf.rxq) {
2581 ret = virtio_init_device(dev, hw->req_guest_features);
2586 if (rxmode->mq_mode == RTE_ETH_MQ_RX_RSS)
2587 req_features |= (1ULL << VIRTIO_NET_F_RSS);
2589 if (rxmode->mtu > hw->max_mtu)
2590 req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
2592 hw->max_rx_pkt_len = ether_hdr_len + rxmode->mtu;
2594 if (rx_offloads & (RTE_ETH_RX_OFFLOAD_UDP_CKSUM |
2595 RTE_ETH_RX_OFFLOAD_TCP_CKSUM))
2596 req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
2598 if (rx_offloads & RTE_ETH_RX_OFFLOAD_TCP_LRO)
2600 (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2601 (1ULL << VIRTIO_NET_F_GUEST_TSO6);
2603 if (tx_offloads & (RTE_ETH_TX_OFFLOAD_UDP_CKSUM |
2604 RTE_ETH_TX_OFFLOAD_TCP_CKSUM))
2605 req_features |= (1ULL << VIRTIO_NET_F_CSUM);
2607 if (tx_offloads & RTE_ETH_TX_OFFLOAD_TCP_TSO)
2609 (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2610 (1ULL << VIRTIO_NET_F_HOST_TSO6);
2612 /* if request features changed, reinit the device */
2613 if (req_features != hw->req_guest_features) {
2614 ret = virtio_init_device(dev, req_features);
2619 if ((rxmode->mq_mode & RTE_ETH_MQ_RX_RSS_FLAG) &&
2620 !virtio_with_feature(hw, VIRTIO_NET_F_RSS)) {
2621 PMD_DRV_LOG(ERR, "RSS support requested but not supported by the device");
2625 if ((rx_offloads & (RTE_ETH_RX_OFFLOAD_UDP_CKSUM |
2626 RTE_ETH_RX_OFFLOAD_TCP_CKSUM)) &&
2627 !virtio_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
2629 "rx checksum not available on this host");
2633 if ((rx_offloads & RTE_ETH_RX_OFFLOAD_TCP_LRO) &&
2634 (!virtio_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2635 !virtio_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6))) {
2637 "Large Receive Offload not available on this host");
2641 /* start control queue */
2642 if (virtio_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
2643 virtio_dev_cq_start(dev);
2645 if (rx_offloads & RTE_ETH_RX_OFFLOAD_VLAN_STRIP)
2648 hw->rx_ol_scatter = (rx_offloads & RTE_ETH_RX_OFFLOAD_SCATTER);
2650 if ((rx_offloads & RTE_ETH_RX_OFFLOAD_VLAN_FILTER) &&
2651 !virtio_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2653 "vlan filtering not available on this host");
2657 hw->has_tx_offload = tx_offload_enabled(hw);
2658 hw->has_rx_offload = rx_offload_enabled(hw);
2660 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2661 /* Enable vector (0) for Link State Interrupt */
2662 if (VIRTIO_OPS(hw)->set_config_irq(hw, 0) ==
2663 VIRTIO_MSI_NO_VECTOR) {
2664 PMD_DRV_LOG(ERR, "failed to set config vector");
2668 if (virtio_with_packed_queue(hw)) {
2669 #if defined(RTE_ARCH_X86_64) && defined(CC_AVX512_SUPPORT)
2670 if ((hw->use_vec_rx || hw->use_vec_tx) &&
2671 (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX512F) ||
2672 !virtio_with_feature(hw, VIRTIO_F_IN_ORDER) ||
2673 !virtio_with_feature(hw, VIRTIO_F_VERSION_1) ||
2674 rte_vect_get_max_simd_bitwidth() < RTE_VECT_SIMD_512)) {
2676 "disabled packed ring vectorized path for requirements not met");
2680 #elif defined(RTE_ARCH_ARM)
2681 if ((hw->use_vec_rx || hw->use_vec_tx) &&
2682 (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON) ||
2683 !virtio_with_feature(hw, VIRTIO_F_IN_ORDER) ||
2684 !virtio_with_feature(hw, VIRTIO_F_VERSION_1) ||
2685 rte_vect_get_max_simd_bitwidth() < RTE_VECT_SIMD_128)) {
2687 "disabled packed ring vectorized path for requirements not met");
2696 if (hw->use_vec_rx) {
2697 if (virtio_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2699 "disabled packed ring vectorized rx for mrg_rxbuf enabled");
2703 if (rx_offloads & RTE_ETH_RX_OFFLOAD_TCP_LRO) {
2705 "disabled packed ring vectorized rx for TCP_LRO enabled");
2710 if (virtio_with_feature(hw, VIRTIO_F_IN_ORDER)) {
2711 hw->use_inorder_tx = 1;
2712 hw->use_inorder_rx = 1;
2716 if (hw->use_vec_rx) {
2717 #if defined RTE_ARCH_ARM
2718 if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
2720 "disabled split ring vectorized path for requirement not met");
2724 if (virtio_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2726 "disabled split ring vectorized rx for mrg_rxbuf enabled");
2730 if (rx_offloads & (RTE_ETH_RX_OFFLOAD_UDP_CKSUM |
2731 RTE_ETH_RX_OFFLOAD_TCP_CKSUM |
2732 RTE_ETH_RX_OFFLOAD_TCP_LRO |
2733 RTE_ETH_RX_OFFLOAD_VLAN_STRIP)) {
2735 "disabled split ring vectorized rx for offloading enabled");
2739 if (rte_vect_get_max_simd_bitwidth() < RTE_VECT_SIMD_128) {
2741 "disabled split ring vectorized rx, max SIMD bitwidth too low");
2752 virtio_dev_start(struct rte_eth_dev *dev)
2754 uint16_t nb_queues, i;
2755 struct virtqueue *vq;
2756 struct virtio_hw *hw = dev->data->dev_private;
2759 /* Finish the initialization of the queues */
2760 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2761 ret = virtio_dev_rx_queue_setup_finish(dev, i);
2765 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2766 ret = virtio_dev_tx_queue_setup_finish(dev, i);
2771 /* check if lsc interrupt feature is enabled */
2772 if (dev->data->dev_conf.intr_conf.lsc) {
2773 if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
2774 PMD_DRV_LOG(ERR, "link status not supported by host");
2779 /* Enable uio/vfio intr/eventfd mapping: although we already did that
2780 * in device configure, but it could be unmapped when device is
2783 if (dev->data->dev_conf.intr_conf.lsc ||
2784 dev->data->dev_conf.intr_conf.rxq) {
2785 virtio_intr_disable(dev);
2787 /* Setup interrupt callback */
2788 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2789 rte_intr_callback_register(dev->intr_handle,
2790 virtio_interrupt_handler,
2793 if (virtio_intr_enable(dev) < 0) {
2794 PMD_DRV_LOG(ERR, "interrupt enable failed");
2799 /*Notify the backend
2800 *Otherwise the tap backend might already stop its queue due to fullness.
2801 *vhost backend will have no chance to be waked up
2803 nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
2804 if (hw->max_queue_pairs > 1) {
2805 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
2809 PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
2811 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2812 vq = virtnet_rxq_to_vq(dev->data->rx_queues[i]);
2813 /* Flush the old packets */
2814 virtqueue_rxvq_flush(vq);
2815 virtqueue_notify(vq);
2818 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2819 vq = virtnet_txq_to_vq(dev->data->tx_queues[i]);
2820 virtqueue_notify(vq);
2823 PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
2825 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2826 vq = virtnet_rxq_to_vq(dev->data->rx_queues[i]);
2830 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2831 vq = virtnet_txq_to_vq(dev->data->tx_queues[i]);
2835 set_rxtx_funcs(dev);
2838 /* Initialize Link state */
2839 virtio_dev_link_update(dev, 0);
2844 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
2846 struct virtio_hw *hw = dev->data->dev_private;
2847 uint16_t nr_vq = virtio_get_nr_vq(hw);
2848 const char *type __rte_unused;
2849 unsigned int i, mbuf_num = 0;
2850 struct virtqueue *vq;
2851 struct rte_mbuf *buf;
2854 if (hw->vqs == NULL)
2857 for (i = 0; i < nr_vq; i++) {
2862 queue_type = virtio_get_queue_type(hw, i);
2863 if (queue_type == VTNET_RQ)
2865 else if (queue_type == VTNET_TQ)
2871 "Before freeing %s[%d] used and unused buf",
2875 while ((buf = virtqueue_detach_unused(vq)) != NULL) {
2876 rte_pktmbuf_free(buf);
2881 "After freeing %s[%d] used and unused buf",
2886 PMD_INIT_LOG(DEBUG, "%d mbufs freed", mbuf_num);
2890 virtio_tx_completed_cleanup(struct rte_eth_dev *dev)
2892 struct virtio_hw *hw = dev->data->dev_private;
2893 struct virtqueue *vq;
2895 void (*xmit_cleanup)(struct virtqueue *vq, uint16_t nb_used);
2897 if (virtio_with_packed_queue(hw)) {
2899 xmit_cleanup = &virtio_xmit_cleanup_inorder_packed;
2900 else if (virtio_with_feature(hw, VIRTIO_F_IN_ORDER))
2901 xmit_cleanup = &virtio_xmit_cleanup_inorder_packed;
2903 xmit_cleanup = &virtio_xmit_cleanup_normal_packed;
2905 if (hw->use_inorder_tx)
2906 xmit_cleanup = &virtio_xmit_cleanup_inorder;
2908 xmit_cleanup = &virtio_xmit_cleanup;
2911 for (qidx = 0; qidx < hw->max_queue_pairs; qidx++) {
2912 vq = hw->vqs[2 * qidx + VTNET_SQ_TQ_QUEUE_IDX];
2914 xmit_cleanup(vq, virtqueue_nused(vq));
2919 * Stop device: disable interrupt and mark link down
2922 virtio_dev_stop(struct rte_eth_dev *dev)
2924 struct virtio_hw *hw = dev->data->dev_private;
2925 struct rte_eth_link link;
2926 struct rte_eth_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
2928 PMD_INIT_LOG(DEBUG, "stop");
2929 dev->data->dev_started = 0;
2931 rte_spinlock_lock(&hw->state_lock);
2936 virtio_tx_completed_cleanup(dev);
2938 if (intr_conf->lsc || intr_conf->rxq) {
2939 virtio_intr_disable(dev);
2941 /* Reset interrupt callback */
2942 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
2943 rte_intr_callback_unregister(dev->intr_handle,
2944 virtio_interrupt_handler,
2949 memset(&link, 0, sizeof(link));
2950 rte_eth_linkstatus_set(dev, &link);
2952 rte_spinlock_unlock(&hw->state_lock);
2958 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
2960 struct rte_eth_link link;
2962 struct virtio_hw *hw = dev->data->dev_private;
2964 memset(&link, 0, sizeof(link));
2965 link.link_duplex = hw->duplex;
2966 link.link_speed = hw->speed;
2967 link.link_autoneg = RTE_ETH_LINK_AUTONEG;
2970 link.link_status = RTE_ETH_LINK_DOWN;
2971 link.link_speed = RTE_ETH_SPEED_NUM_NONE;
2972 } else if (virtio_with_feature(hw, VIRTIO_NET_F_STATUS)) {
2973 PMD_INIT_LOG(DEBUG, "Get link status from hw");
2974 virtio_read_dev_config(hw,
2975 offsetof(struct virtio_net_config, status),
2976 &status, sizeof(status));
2977 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
2978 link.link_status = RTE_ETH_LINK_DOWN;
2979 link.link_speed = RTE_ETH_SPEED_NUM_NONE;
2980 PMD_INIT_LOG(DEBUG, "Port %d is down",
2981 dev->data->port_id);
2983 link.link_status = RTE_ETH_LINK_UP;
2984 if (hw->get_speed_via_feat)
2985 virtio_get_speed_duplex(dev, &link);
2986 PMD_INIT_LOG(DEBUG, "Port %d is up",
2987 dev->data->port_id);
2990 link.link_status = RTE_ETH_LINK_UP;
2991 if (hw->get_speed_via_feat)
2992 virtio_get_speed_duplex(dev, &link);
2995 return rte_eth_linkstatus_set(dev, &link);
2999 virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
3001 const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
3002 struct virtio_hw *hw = dev->data->dev_private;
3003 uint64_t offloads = rxmode->offloads;
3005 if (mask & RTE_ETH_VLAN_FILTER_MASK) {
3006 if ((offloads & RTE_ETH_RX_OFFLOAD_VLAN_FILTER) &&
3007 !virtio_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
3010 "vlan filtering not available on this host");
3016 if (mask & RTE_ETH_VLAN_STRIP_MASK)
3017 hw->vlan_strip = !!(offloads & RTE_ETH_RX_OFFLOAD_VLAN_STRIP);
3023 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
3025 uint64_t tso_mask, host_features;
3026 uint32_t rss_hash_types = 0;
3027 struct virtio_hw *hw = dev->data->dev_private;
3028 dev_info->speed_capa = virtio_dev_speed_capa_get(hw->speed);
3030 dev_info->max_rx_queues =
3031 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
3032 dev_info->max_tx_queues =
3033 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
3034 dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
3035 dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
3036 dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
3037 dev_info->max_mtu = hw->max_mtu;
3039 host_features = VIRTIO_OPS(hw)->get_features(hw);
3040 dev_info->rx_offload_capa = RTE_ETH_RX_OFFLOAD_VLAN_STRIP;
3041 if (host_features & (1ULL << VIRTIO_NET_F_MRG_RXBUF))
3042 dev_info->rx_offload_capa |= RTE_ETH_RX_OFFLOAD_SCATTER;
3043 if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
3044 dev_info->rx_offload_capa |=
3045 RTE_ETH_RX_OFFLOAD_TCP_CKSUM |
3046 RTE_ETH_RX_OFFLOAD_UDP_CKSUM;
3048 if (host_features & (1ULL << VIRTIO_NET_F_CTRL_VLAN))
3049 dev_info->rx_offload_capa |= RTE_ETH_RX_OFFLOAD_VLAN_FILTER;
3050 tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
3051 (1ULL << VIRTIO_NET_F_GUEST_TSO6);
3052 if ((host_features & tso_mask) == tso_mask)
3053 dev_info->rx_offload_capa |= RTE_ETH_RX_OFFLOAD_TCP_LRO;
3055 dev_info->tx_offload_capa = RTE_ETH_TX_OFFLOAD_MULTI_SEGS |
3056 RTE_ETH_TX_OFFLOAD_VLAN_INSERT;
3057 if (host_features & (1ULL << VIRTIO_NET_F_CSUM)) {
3058 dev_info->tx_offload_capa |=
3059 RTE_ETH_TX_OFFLOAD_UDP_CKSUM |
3060 RTE_ETH_TX_OFFLOAD_TCP_CKSUM;
3062 tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
3063 (1ULL << VIRTIO_NET_F_HOST_TSO6);
3064 if ((host_features & tso_mask) == tso_mask)
3065 dev_info->tx_offload_capa |= RTE_ETH_TX_OFFLOAD_TCP_TSO;
3067 if (host_features & (1ULL << VIRTIO_NET_F_RSS)) {
3068 virtio_dev_get_rss_config(hw, &rss_hash_types);
3069 dev_info->hash_key_size = VIRTIO_NET_RSS_KEY_SIZE;
3070 dev_info->reta_size = VIRTIO_NET_RSS_RETA_SIZE;
3071 dev_info->flow_type_rss_offloads =
3072 virtio_to_ethdev_rss_offloads(rss_hash_types);
3074 dev_info->hash_key_size = 0;
3075 dev_info->reta_size = 0;
3076 dev_info->flow_type_rss_offloads = 0;
3079 if (host_features & (1ULL << VIRTIO_F_RING_PACKED)) {
3081 * According to 2.7 Packed Virtqueues,
3082 * 2.7.10.1 Structure Size and Alignment:
3083 * The Queue Size value does not have to be a power of 2.
3085 dev_info->rx_desc_lim.nb_max = UINT16_MAX;
3086 dev_info->tx_desc_lim.nb_max = UINT16_MAX;
3089 * According to 2.6 Split Virtqueues:
3090 * Queue Size value is always a power of 2. The maximum Queue
3091 * Size value is 32768.
3093 dev_info->rx_desc_lim.nb_max = 32768;
3094 dev_info->tx_desc_lim.nb_max = 32768;
3097 * Actual minimum is not the same for virtqueues of different kinds,
3098 * but to avoid tangling the code with separate branches, rely on
3099 * default thresholds since desc number must be at least of their size.
3101 dev_info->rx_desc_lim.nb_min = RTE_MAX(DEFAULT_RX_FREE_THRESH,
3102 RTE_VIRTIO_VPMD_RX_REARM_THRESH);
3103 dev_info->tx_desc_lim.nb_min = DEFAULT_TX_FREE_THRESH;
3104 dev_info->rx_desc_lim.nb_align = 1;
3105 dev_info->tx_desc_lim.nb_align = 1;
3111 * It enables testpmd to collect per queue stats.
3114 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
3115 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
3116 __rte_unused uint8_t is_rx)
3121 RTE_LOG_REGISTER_SUFFIX(virtio_logtype_init, init, NOTICE);
3122 RTE_LOG_REGISTER_SUFFIX(virtio_logtype_driver, driver, NOTICE);