4 * Copyright(c) 2010-2016 Intel Corporation. All rights reserved.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * * Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * * Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * * Neither the name of Intel Corporation nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
40 #include <rte_ethdev.h>
41 #include <rte_memcpy.h>
42 #include <rte_string_fns.h>
43 #include <rte_memzone.h>
44 #include <rte_malloc.h>
45 #include <rte_atomic.h>
46 #include <rte_branch_prediction.h>
48 #include <rte_ether.h>
49 #include <rte_common.h>
50 #include <rte_errno.h>
52 #include <rte_memory.h>
56 #include "virtio_ethdev.h"
57 #include "virtio_pci.h"
58 #include "virtio_logs.h"
59 #include "virtqueue.h"
60 #include "virtio_rxtx.h"
62 static int eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev);
63 static int virtio_dev_configure(struct rte_eth_dev *dev);
64 static int virtio_dev_start(struct rte_eth_dev *dev);
65 static void virtio_dev_stop(struct rte_eth_dev *dev);
66 static void virtio_dev_promiscuous_enable(struct rte_eth_dev *dev);
67 static void virtio_dev_promiscuous_disable(struct rte_eth_dev *dev);
68 static void virtio_dev_allmulticast_enable(struct rte_eth_dev *dev);
69 static void virtio_dev_allmulticast_disable(struct rte_eth_dev *dev);
70 static void virtio_dev_info_get(struct rte_eth_dev *dev,
71 struct rte_eth_dev_info *dev_info);
72 static int virtio_dev_link_update(struct rte_eth_dev *dev,
73 __rte_unused int wait_to_complete);
75 static void virtio_set_hwaddr(struct virtio_hw *hw);
76 static void virtio_get_hwaddr(struct virtio_hw *hw);
78 static void virtio_dev_stats_get(struct rte_eth_dev *dev,
79 struct rte_eth_stats *stats);
80 static int virtio_dev_xstats_get(struct rte_eth_dev *dev,
81 struct rte_eth_xstat *xstats, unsigned n);
82 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
83 struct rte_eth_xstat_name *xstats_names,
85 static void virtio_dev_stats_reset(struct rte_eth_dev *dev);
86 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev);
87 static int virtio_vlan_filter_set(struct rte_eth_dev *dev,
88 uint16_t vlan_id, int on);
89 static void virtio_mac_addr_add(struct rte_eth_dev *dev,
90 struct ether_addr *mac_addr,
91 uint32_t index, uint32_t vmdq __rte_unused);
92 static void virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index);
93 static void virtio_mac_addr_set(struct rte_eth_dev *dev,
94 struct ether_addr *mac_addr);
96 static int virtio_dev_queue_stats_mapping_set(
97 __rte_unused struct rte_eth_dev *eth_dev,
98 __rte_unused uint16_t queue_id,
99 __rte_unused uint8_t stat_idx,
100 __rte_unused uint8_t is_rx);
103 * The set of PCI devices this driver supports
105 static const struct rte_pci_id pci_id_virtio_map[] = {
106 { RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_LEGACY_DEVICEID_NET) },
107 { RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_MODERN_DEVICEID_NET) },
108 { .vendor_id = 0, /* sentinel */ },
111 struct rte_virtio_xstats_name_off {
112 char name[RTE_ETH_XSTATS_NAME_SIZE];
116 /* [rt]x_qX_ is prepended to the name string here */
117 static const struct rte_virtio_xstats_name_off rte_virtio_rxq_stat_strings[] = {
118 {"good_packets", offsetof(struct virtnet_rx, stats.packets)},
119 {"good_bytes", offsetof(struct virtnet_rx, stats.bytes)},
120 {"errors", offsetof(struct virtnet_rx, stats.errors)},
121 {"multicast_packets", offsetof(struct virtnet_rx, stats.multicast)},
122 {"broadcast_packets", offsetof(struct virtnet_rx, stats.broadcast)},
123 {"undersize_packets", offsetof(struct virtnet_rx, stats.size_bins[0])},
124 {"size_64_packets", offsetof(struct virtnet_rx, stats.size_bins[1])},
125 {"size_65_127_packets", offsetof(struct virtnet_rx, stats.size_bins[2])},
126 {"size_128_255_packets", offsetof(struct virtnet_rx, stats.size_bins[3])},
127 {"size_256_511_packets", offsetof(struct virtnet_rx, stats.size_bins[4])},
128 {"size_512_1023_packets", offsetof(struct virtnet_rx, stats.size_bins[5])},
129 {"size_1024_1518_packets", offsetof(struct virtnet_rx, stats.size_bins[6])},
130 {"size_1519_max_packets", offsetof(struct virtnet_rx, stats.size_bins[7])},
133 /* [rt]x_qX_ is prepended to the name string here */
134 static const struct rte_virtio_xstats_name_off rte_virtio_txq_stat_strings[] = {
135 {"good_packets", offsetof(struct virtnet_tx, stats.packets)},
136 {"good_bytes", offsetof(struct virtnet_tx, stats.bytes)},
137 {"errors", offsetof(struct virtnet_tx, stats.errors)},
138 {"multicast_packets", offsetof(struct virtnet_tx, stats.multicast)},
139 {"broadcast_packets", offsetof(struct virtnet_tx, stats.broadcast)},
140 {"undersize_packets", offsetof(struct virtnet_tx, stats.size_bins[0])},
141 {"size_64_packets", offsetof(struct virtnet_tx, stats.size_bins[1])},
142 {"size_65_127_packets", offsetof(struct virtnet_tx, stats.size_bins[2])},
143 {"size_128_255_packets", offsetof(struct virtnet_tx, stats.size_bins[3])},
144 {"size_256_511_packets", offsetof(struct virtnet_tx, stats.size_bins[4])},
145 {"size_512_1023_packets", offsetof(struct virtnet_tx, stats.size_bins[5])},
146 {"size_1024_1518_packets", offsetof(struct virtnet_tx, stats.size_bins[6])},
147 {"size_1519_max_packets", offsetof(struct virtnet_tx, stats.size_bins[7])},
150 #define VIRTIO_NB_RXQ_XSTATS (sizeof(rte_virtio_rxq_stat_strings) / \
151 sizeof(rte_virtio_rxq_stat_strings[0]))
152 #define VIRTIO_NB_TXQ_XSTATS (sizeof(rte_virtio_txq_stat_strings) / \
153 sizeof(rte_virtio_txq_stat_strings[0]))
156 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
157 int *dlen, int pkt_num)
161 virtio_net_ctrl_ack status = ~0;
162 struct virtio_pmd_ctrl result;
163 struct virtqueue *vq;
165 ctrl->status = status;
167 if (!cvq || !cvq->vq) {
168 PMD_INIT_LOG(ERR, "Control queue is not supported.");
172 head = vq->vq_desc_head_idx;
174 PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
175 "vq->hw->cvq = %p vq = %p",
176 vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
178 if ((vq->vq_free_cnt < ((uint32_t)pkt_num + 2)) || (pkt_num < 1))
181 memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
182 sizeof(struct virtio_pmd_ctrl));
185 * Format is enforced in qemu code:
186 * One TX packet for header;
187 * At least one TX packet per argument;
188 * One RX packet for ACK.
190 vq->vq_ring.desc[head].flags = VRING_DESC_F_NEXT;
191 vq->vq_ring.desc[head].addr = cvq->virtio_net_hdr_mem;
192 vq->vq_ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
194 i = vq->vq_ring.desc[head].next;
196 for (k = 0; k < pkt_num; k++) {
197 vq->vq_ring.desc[i].flags = VRING_DESC_F_NEXT;
198 vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem
199 + sizeof(struct virtio_net_ctrl_hdr)
200 + sizeof(ctrl->status) + sizeof(uint8_t)*sum;
201 vq->vq_ring.desc[i].len = dlen[k];
204 i = vq->vq_ring.desc[i].next;
207 vq->vq_ring.desc[i].flags = VRING_DESC_F_WRITE;
208 vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem
209 + sizeof(struct virtio_net_ctrl_hdr);
210 vq->vq_ring.desc[i].len = sizeof(ctrl->status);
213 vq->vq_desc_head_idx = vq->vq_ring.desc[i].next;
215 vq_update_avail_ring(vq, head);
216 vq_update_avail_idx(vq);
218 PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
220 virtqueue_notify(vq);
223 while (VIRTQUEUE_NUSED(vq) == 0) {
228 while (VIRTQUEUE_NUSED(vq)) {
229 uint32_t idx, desc_idx, used_idx;
230 struct vring_used_elem *uep;
232 used_idx = (uint32_t)(vq->vq_used_cons_idx
233 & (vq->vq_nentries - 1));
234 uep = &vq->vq_ring.used->ring[used_idx];
235 idx = (uint32_t) uep->id;
238 while (vq->vq_ring.desc[desc_idx].flags & VRING_DESC_F_NEXT) {
239 desc_idx = vq->vq_ring.desc[desc_idx].next;
243 vq->vq_ring.desc[desc_idx].next = vq->vq_desc_head_idx;
244 vq->vq_desc_head_idx = idx;
246 vq->vq_used_cons_idx++;
250 PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
251 vq->vq_free_cnt, vq->vq_desc_head_idx);
253 memcpy(&result, cvq->virtio_net_hdr_mz->addr,
254 sizeof(struct virtio_pmd_ctrl));
256 return result.status;
260 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
262 struct virtio_hw *hw = dev->data->dev_private;
263 struct virtio_pmd_ctrl ctrl;
267 ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
268 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
269 memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
271 dlen[0] = sizeof(uint16_t);
273 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
275 PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
276 "failed, this is too late now...");
284 virtio_dev_queue_release(void *queue __rte_unused)
290 virtio_get_queue_type(struct virtio_hw *hw, uint16_t vtpci_queue_idx)
292 if (vtpci_queue_idx == hw->max_queue_pairs * 2)
294 else if (vtpci_queue_idx % 2 == 0)
301 virtio_get_nr_vq(struct virtio_hw *hw)
303 uint16_t nr_vq = hw->max_queue_pairs * 2;
305 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
312 virtio_init_vring(struct virtqueue *vq)
314 int size = vq->vq_nentries;
315 struct vring *vr = &vq->vq_ring;
316 uint8_t *ring_mem = vq->vq_ring_virt_mem;
318 PMD_INIT_FUNC_TRACE();
321 * Reinitialise since virtio port might have been stopped and restarted
323 memset(ring_mem, 0, vq->vq_ring_size);
324 vring_init(vr, size, ring_mem, VIRTIO_PCI_VRING_ALIGN);
325 vq->vq_used_cons_idx = 0;
326 vq->vq_desc_head_idx = 0;
327 vq->vq_avail_idx = 0;
328 vq->vq_desc_tail_idx = (uint16_t)(vq->vq_nentries - 1);
329 vq->vq_free_cnt = vq->vq_nentries;
330 memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries);
332 vring_desc_init(vr->desc, size);
335 * Disable device(host) interrupting guest
337 virtqueue_disable_intr(vq);
341 virtio_init_queue(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx)
343 char vq_name[VIRTQUEUE_MAX_NAME_SZ];
344 char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
345 const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
346 unsigned int vq_size, size;
347 struct virtio_hw *hw = dev->data->dev_private;
348 struct virtnet_rx *rxvq = NULL;
349 struct virtnet_tx *txvq = NULL;
350 struct virtnet_ctl *cvq = NULL;
351 struct virtqueue *vq;
352 size_t sz_hdr_mz = 0;
353 void *sw_ring = NULL;
354 int queue_type = virtio_get_queue_type(hw, vtpci_queue_idx);
357 PMD_INIT_LOG(DEBUG, "setting up queue: %u", vtpci_queue_idx);
360 * Read the virtqueue size from the Queue Size field
361 * Always power of 2 and if 0 virtqueue does not exist
363 vq_size = hw->vtpci_ops->get_queue_num(hw, vtpci_queue_idx);
364 PMD_INIT_LOG(DEBUG, "vq_size: %u", vq_size);
366 PMD_INIT_LOG(ERR, "virtqueue does not exist");
370 if (!rte_is_power_of_2(vq_size)) {
371 PMD_INIT_LOG(ERR, "virtqueue size is not powerof 2");
375 snprintf(vq_name, sizeof(vq_name), "port%d_vq%d",
376 dev->data->port_id, vtpci_queue_idx);
378 size = RTE_ALIGN_CEIL(sizeof(*vq) +
379 vq_size * sizeof(struct vq_desc_extra),
380 RTE_CACHE_LINE_SIZE);
381 if (queue_type == VTNET_TQ) {
383 * For each xmit packet, allocate a virtio_net_hdr
384 * and indirect ring elements
386 sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
387 } else if (queue_type == VTNET_CQ) {
388 /* Allocate a page for control vq command, data and status */
389 sz_hdr_mz = PAGE_SIZE;
392 vq = rte_zmalloc_socket(vq_name, size, RTE_CACHE_LINE_SIZE,
395 PMD_INIT_LOG(ERR, "can not allocate vq");
398 hw->vqs[vtpci_queue_idx] = vq;
401 vq->vq_queue_index = vtpci_queue_idx;
402 vq->vq_nentries = vq_size;
405 * Reserve a memzone for vring elements
407 size = vring_size(vq_size, VIRTIO_PCI_VRING_ALIGN);
408 vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
409 PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
410 size, vq->vq_ring_size);
412 mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
414 0, VIRTIO_PCI_VRING_ALIGN);
416 if (rte_errno == EEXIST)
417 mz = rte_memzone_lookup(vq_name);
424 memset(mz->addr, 0, sizeof(mz->len));
426 vq->vq_ring_mem = mz->phys_addr;
427 vq->vq_ring_virt_mem = mz->addr;
428 PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem: 0x%" PRIx64,
429 (uint64_t)mz->phys_addr);
430 PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64,
431 (uint64_t)(uintptr_t)mz->addr);
433 virtio_init_vring(vq);
436 snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_vq%d_hdr",
437 dev->data->port_id, vtpci_queue_idx);
438 hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
440 RTE_CACHE_LINE_SIZE);
441 if (hdr_mz == NULL) {
442 if (rte_errno == EEXIST)
443 hdr_mz = rte_memzone_lookup(vq_hdr_name);
444 if (hdr_mz == NULL) {
451 if (queue_type == VTNET_RQ) {
452 size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
453 sizeof(vq->sw_ring[0]);
455 sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
456 RTE_CACHE_LINE_SIZE, SOCKET_ID_ANY);
458 PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
463 vq->sw_ring = sw_ring;
466 rxvq->port_id = dev->data->port_id;
468 } else if (queue_type == VTNET_TQ) {
471 txvq->port_id = dev->data->port_id;
473 txvq->virtio_net_hdr_mz = hdr_mz;
474 txvq->virtio_net_hdr_mem = hdr_mz->phys_addr;
475 } else if (queue_type == VTNET_CQ) {
479 cvq->virtio_net_hdr_mz = hdr_mz;
480 cvq->virtio_net_hdr_mem = hdr_mz->phys_addr;
481 memset(cvq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
486 /* For virtio_user case (that is when hw->dev is NULL), we use
487 * virtual address. And we need properly set _offset_, please see
488 * VIRTIO_MBUF_DATA_DMA_ADDR in virtqueue.h for more information.
491 vq->offset = offsetof(struct rte_mbuf, buf_physaddr);
493 vq->vq_ring_mem = (uintptr_t)mz->addr;
494 vq->offset = offsetof(struct rte_mbuf, buf_addr);
495 if (queue_type == VTNET_TQ)
496 txvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
497 else if (queue_type == VTNET_CQ)
498 cvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
501 if (queue_type == VTNET_TQ) {
502 struct virtio_tx_region *txr;
506 memset(txr, 0, vq_size * sizeof(*txr));
507 for (i = 0; i < vq_size; i++) {
508 struct vring_desc *start_dp = txr[i].tx_indir;
510 vring_desc_init(start_dp, RTE_DIM(txr[i].tx_indir));
512 /* first indirect descriptor is always the tx header */
513 start_dp->addr = txvq->virtio_net_hdr_mem
515 + offsetof(struct virtio_tx_region, tx_hdr);
517 start_dp->len = hw->vtnet_hdr_size;
518 start_dp->flags = VRING_DESC_F_NEXT;
522 if (hw->vtpci_ops->setup_queue(hw, vq) < 0) {
523 PMD_INIT_LOG(ERR, "setup_queue failed");
531 rte_memzone_free(hdr_mz);
532 rte_memzone_free(mz);
539 virtio_free_queues(struct virtio_hw *hw)
541 uint16_t nr_vq = virtio_get_nr_vq(hw);
542 struct virtqueue *vq;
546 for (i = 0; i < nr_vq; i++) {
551 queue_type = virtio_get_queue_type(hw, i);
552 if (queue_type == VTNET_RQ) {
553 rte_free(vq->sw_ring);
554 rte_memzone_free(vq->rxq.mz);
555 } else if (queue_type == VTNET_TQ) {
556 rte_memzone_free(vq->txq.mz);
557 rte_memzone_free(vq->txq.virtio_net_hdr_mz);
559 rte_memzone_free(vq->cq.mz);
560 rte_memzone_free(vq->cq.virtio_net_hdr_mz);
570 virtio_alloc_queues(struct rte_eth_dev *dev)
572 struct virtio_hw *hw = dev->data->dev_private;
573 uint16_t nr_vq = virtio_get_nr_vq(hw);
577 hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
579 PMD_INIT_LOG(ERR, "failed to allocate vqs");
583 for (i = 0; i < nr_vq; i++) {
584 ret = virtio_init_queue(dev, i);
586 virtio_free_queues(hw);
595 virtio_dev_close(struct rte_eth_dev *dev)
597 struct virtio_hw *hw = dev->data->dev_private;
599 PMD_INIT_LOG(DEBUG, "virtio_dev_close");
602 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
603 vtpci_irq_config(hw, VIRTIO_MSI_NO_VECTOR);
605 virtio_dev_free_mbufs(dev);
606 virtio_free_queues(hw);
610 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
612 struct virtio_hw *hw = dev->data->dev_private;
613 struct virtio_pmd_ctrl ctrl;
617 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
618 PMD_INIT_LOG(INFO, "host does not support rx control\n");
622 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
623 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
627 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
629 PMD_INIT_LOG(ERR, "Failed to enable promisc");
633 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
635 struct virtio_hw *hw = dev->data->dev_private;
636 struct virtio_pmd_ctrl ctrl;
640 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
641 PMD_INIT_LOG(INFO, "host does not support rx control\n");
645 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
646 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
650 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
652 PMD_INIT_LOG(ERR, "Failed to disable promisc");
656 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
658 struct virtio_hw *hw = dev->data->dev_private;
659 struct virtio_pmd_ctrl ctrl;
663 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
664 PMD_INIT_LOG(INFO, "host does not support rx control\n");
668 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
669 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
673 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
675 PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
679 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
681 struct virtio_hw *hw = dev->data->dev_private;
682 struct virtio_pmd_ctrl ctrl;
686 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
687 PMD_INIT_LOG(INFO, "host does not support rx control\n");
691 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
692 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
696 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
698 PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
701 #define VLAN_TAG_LEN 4 /* 802.3ac tag (not DMA'd) */
703 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
705 struct virtio_hw *hw = dev->data->dev_private;
706 uint32_t ether_hdr_len = ETHER_HDR_LEN + VLAN_TAG_LEN +
708 uint32_t frame_size = mtu + ether_hdr_len;
710 if (mtu < ETHER_MIN_MTU || frame_size > VIRTIO_MAX_RX_PKTLEN) {
711 PMD_INIT_LOG(ERR, "MTU should be between %d and %d\n",
712 ETHER_MIN_MTU, VIRTIO_MAX_RX_PKTLEN - ether_hdr_len);
719 * dev_ops for virtio, bare necessities for basic operation
721 static const struct eth_dev_ops virtio_eth_dev_ops = {
722 .dev_configure = virtio_dev_configure,
723 .dev_start = virtio_dev_start,
724 .dev_stop = virtio_dev_stop,
725 .dev_close = virtio_dev_close,
726 .promiscuous_enable = virtio_dev_promiscuous_enable,
727 .promiscuous_disable = virtio_dev_promiscuous_disable,
728 .allmulticast_enable = virtio_dev_allmulticast_enable,
729 .allmulticast_disable = virtio_dev_allmulticast_disable,
730 .mtu_set = virtio_mtu_set,
731 .dev_infos_get = virtio_dev_info_get,
732 .stats_get = virtio_dev_stats_get,
733 .xstats_get = virtio_dev_xstats_get,
734 .xstats_get_names = virtio_dev_xstats_get_names,
735 .stats_reset = virtio_dev_stats_reset,
736 .xstats_reset = virtio_dev_stats_reset,
737 .link_update = virtio_dev_link_update,
738 .rx_queue_setup = virtio_dev_rx_queue_setup,
739 .rx_queue_release = virtio_dev_queue_release,
740 .tx_queue_setup = virtio_dev_tx_queue_setup,
741 .tx_queue_release = virtio_dev_queue_release,
742 /* collect stats per queue */
743 .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
744 .vlan_filter_set = virtio_vlan_filter_set,
745 .mac_addr_add = virtio_mac_addr_add,
746 .mac_addr_remove = virtio_mac_addr_remove,
747 .mac_addr_set = virtio_mac_addr_set,
751 virtio_dev_atomic_read_link_status(struct rte_eth_dev *dev,
752 struct rte_eth_link *link)
754 struct rte_eth_link *dst = link;
755 struct rte_eth_link *src = &(dev->data->dev_link);
757 if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
758 *(uint64_t *)src) == 0)
765 * Atomically writes the link status information into global
766 * structure rte_eth_dev.
769 * - Pointer to the structure rte_eth_dev to read from.
770 * - Pointer to the buffer to be saved with the link status.
773 * - On success, zero.
774 * - On failure, negative value.
777 virtio_dev_atomic_write_link_status(struct rte_eth_dev *dev,
778 struct rte_eth_link *link)
780 struct rte_eth_link *dst = &(dev->data->dev_link);
781 struct rte_eth_link *src = link;
783 if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
784 *(uint64_t *)src) == 0)
791 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
795 for (i = 0; i < dev->data->nb_tx_queues; i++) {
796 const struct virtnet_tx *txvq = dev->data->tx_queues[i];
800 stats->opackets += txvq->stats.packets;
801 stats->obytes += txvq->stats.bytes;
802 stats->oerrors += txvq->stats.errors;
804 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
805 stats->q_opackets[i] = txvq->stats.packets;
806 stats->q_obytes[i] = txvq->stats.bytes;
810 for (i = 0; i < dev->data->nb_rx_queues; i++) {
811 const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
815 stats->ipackets += rxvq->stats.packets;
816 stats->ibytes += rxvq->stats.bytes;
817 stats->ierrors += rxvq->stats.errors;
819 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
820 stats->q_ipackets[i] = rxvq->stats.packets;
821 stats->q_ibytes[i] = rxvq->stats.bytes;
825 stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
828 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
829 struct rte_eth_xstat_name *xstats_names,
830 __rte_unused unsigned limit)
836 unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
837 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
839 if (xstats_names != NULL) {
840 /* Note: limit checked in rte_eth_xstats_names() */
842 for (i = 0; i < dev->data->nb_rx_queues; i++) {
843 struct virtqueue *rxvq = dev->data->rx_queues[i];
846 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
847 snprintf(xstats_names[count].name,
848 sizeof(xstats_names[count].name),
850 rte_virtio_rxq_stat_strings[t].name);
855 for (i = 0; i < dev->data->nb_tx_queues; i++) {
856 struct virtqueue *txvq = dev->data->tx_queues[i];
859 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
860 snprintf(xstats_names[count].name,
861 sizeof(xstats_names[count].name),
863 rte_virtio_txq_stat_strings[t].name);
873 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
879 unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
880 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
885 for (i = 0; i < dev->data->nb_rx_queues; i++) {
886 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
893 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
894 xstats[count].value = *(uint64_t *)(((char *)rxvq) +
895 rte_virtio_rxq_stat_strings[t].offset);
896 xstats[count].id = count;
901 for (i = 0; i < dev->data->nb_tx_queues; i++) {
902 struct virtnet_tx *txvq = dev->data->tx_queues[i];
909 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
910 xstats[count].value = *(uint64_t *)(((char *)txvq) +
911 rte_virtio_txq_stat_strings[t].offset);
912 xstats[count].id = count;
921 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
923 virtio_update_stats(dev, stats);
927 virtio_dev_stats_reset(struct rte_eth_dev *dev)
931 for (i = 0; i < dev->data->nb_tx_queues; i++) {
932 struct virtnet_tx *txvq = dev->data->tx_queues[i];
936 txvq->stats.packets = 0;
937 txvq->stats.bytes = 0;
938 txvq->stats.errors = 0;
939 txvq->stats.multicast = 0;
940 txvq->stats.broadcast = 0;
941 memset(txvq->stats.size_bins, 0,
942 sizeof(txvq->stats.size_bins[0]) * 8);
945 for (i = 0; i < dev->data->nb_rx_queues; i++) {
946 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
950 rxvq->stats.packets = 0;
951 rxvq->stats.bytes = 0;
952 rxvq->stats.errors = 0;
953 rxvq->stats.multicast = 0;
954 rxvq->stats.broadcast = 0;
955 memset(rxvq->stats.size_bins, 0,
956 sizeof(rxvq->stats.size_bins[0]) * 8);
961 virtio_set_hwaddr(struct virtio_hw *hw)
963 vtpci_write_dev_config(hw,
964 offsetof(struct virtio_net_config, mac),
965 &hw->mac_addr, ETHER_ADDR_LEN);
969 virtio_get_hwaddr(struct virtio_hw *hw)
971 if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
972 vtpci_read_dev_config(hw,
973 offsetof(struct virtio_net_config, mac),
974 &hw->mac_addr, ETHER_ADDR_LEN);
976 eth_random_addr(&hw->mac_addr[0]);
977 virtio_set_hwaddr(hw);
982 virtio_mac_table_set(struct virtio_hw *hw,
983 const struct virtio_net_ctrl_mac *uc,
984 const struct virtio_net_ctrl_mac *mc)
986 struct virtio_pmd_ctrl ctrl;
989 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
990 PMD_DRV_LOG(INFO, "host does not support mac table");
994 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
995 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
997 len[0] = uc->entries * ETHER_ADDR_LEN + sizeof(uc->entries);
998 memcpy(ctrl.data, uc, len[0]);
1000 len[1] = mc->entries * ETHER_ADDR_LEN + sizeof(mc->entries);
1001 memcpy(ctrl.data + len[0], mc, len[1]);
1003 err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1005 PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1009 virtio_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
1010 uint32_t index, uint32_t vmdq __rte_unused)
1012 struct virtio_hw *hw = dev->data->dev_private;
1013 const struct ether_addr *addrs = dev->data->mac_addrs;
1015 struct virtio_net_ctrl_mac *uc, *mc;
1017 if (index >= VIRTIO_MAX_MAC_ADDRS) {
1018 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1022 uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1024 mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1027 for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1028 const struct ether_addr *addr
1029 = (i == index) ? mac_addr : addrs + i;
1030 struct virtio_net_ctrl_mac *tbl
1031 = is_multicast_ether_addr(addr) ? mc : uc;
1033 memcpy(&tbl->macs[tbl->entries++], addr, ETHER_ADDR_LEN);
1036 virtio_mac_table_set(hw, uc, mc);
1040 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1042 struct virtio_hw *hw = dev->data->dev_private;
1043 struct ether_addr *addrs = dev->data->mac_addrs;
1044 struct virtio_net_ctrl_mac *uc, *mc;
1047 if (index >= VIRTIO_MAX_MAC_ADDRS) {
1048 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1052 uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1054 mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1057 for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1058 struct virtio_net_ctrl_mac *tbl;
1060 if (i == index || is_zero_ether_addr(addrs + i))
1063 tbl = is_multicast_ether_addr(addrs + i) ? mc : uc;
1064 memcpy(&tbl->macs[tbl->entries++], addrs + i, ETHER_ADDR_LEN);
1067 virtio_mac_table_set(hw, uc, mc);
1071 virtio_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
1073 struct virtio_hw *hw = dev->data->dev_private;
1075 memcpy(hw->mac_addr, mac_addr, ETHER_ADDR_LEN);
1077 /* Use atomic update if available */
1078 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1079 struct virtio_pmd_ctrl ctrl;
1080 int len = ETHER_ADDR_LEN;
1082 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1083 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1085 memcpy(ctrl.data, mac_addr, ETHER_ADDR_LEN);
1086 virtio_send_command(hw->cvq, &ctrl, &len, 1);
1087 } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1088 virtio_set_hwaddr(hw);
1092 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1094 struct virtio_hw *hw = dev->data->dev_private;
1095 struct virtio_pmd_ctrl ctrl;
1098 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1101 ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1102 ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1103 memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1104 len = sizeof(vlan_id);
1106 return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1110 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1112 uint64_t host_features;
1114 /* Prepare guest_features: feature that driver wants to support */
1115 PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1118 /* Read device(host) feature bits */
1119 host_features = hw->vtpci_ops->get_features(hw);
1120 PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1124 * Negotiate features: Subset of device feature bits are written back
1125 * guest feature bits.
1127 hw->guest_features = req_features;
1128 hw->guest_features = vtpci_negotiate_features(hw, host_features);
1129 PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1130 hw->guest_features);
1133 if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1135 "VIRTIO_F_VERSION_1 features is not enabled.");
1138 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1139 if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1141 "failed to set FEATURES_OK status!");
1146 hw->req_guest_features = req_features;
1152 * Process Virtio Config changed interrupt and call the callback
1153 * if link state changed.
1156 virtio_interrupt_handler(struct rte_intr_handle *handle,
1159 struct rte_eth_dev *dev = param;
1160 struct virtio_hw *hw = dev->data->dev_private;
1163 /* Read interrupt status which clears interrupt */
1164 isr = vtpci_isr(hw);
1165 PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1167 if (rte_intr_enable(handle) < 0)
1168 PMD_DRV_LOG(ERR, "interrupt enable failed");
1170 if (isr & VIRTIO_PCI_ISR_CONFIG) {
1171 if (virtio_dev_link_update(dev, 0) == 0)
1172 _rte_eth_dev_callback_process(dev,
1173 RTE_ETH_EVENT_INTR_LSC, NULL);
1179 rx_func_get(struct rte_eth_dev *eth_dev)
1181 struct virtio_hw *hw = eth_dev->data->dev_private;
1182 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF))
1183 eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1185 eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1188 /* reset device and renegotiate features if needed */
1190 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1192 struct virtio_hw *hw = eth_dev->data->dev_private;
1193 struct virtio_net_config *config;
1194 struct virtio_net_config local_config;
1195 struct rte_pci_device *pci_dev = hw->dev;
1198 /* Reset the device although not necessary at startup */
1201 /* Tell the host we've noticed this device. */
1202 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1204 /* Tell the host we've known how to drive the device. */
1205 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1206 if (virtio_negotiate_features(hw, req_features) < 0)
1209 /* If host does not support status then disable LSC */
1210 if (!vtpci_with_feature(hw, VIRTIO_NET_F_STATUS))
1211 eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1213 eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1216 rte_eth_copy_pci_info(eth_dev, pci_dev);
1218 rx_func_get(eth_dev);
1220 /* Setting up rx_header size for the device */
1221 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1222 vtpci_with_feature(hw, VIRTIO_F_VERSION_1))
1223 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1225 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1227 /* Copy the permanent MAC address to: virtio_hw */
1228 virtio_get_hwaddr(hw);
1229 ether_addr_copy((struct ether_addr *) hw->mac_addr,
1230 ð_dev->data->mac_addrs[0]);
1232 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1233 hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1234 hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1236 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1237 config = &local_config;
1239 vtpci_read_dev_config(hw,
1240 offsetof(struct virtio_net_config, mac),
1241 &config->mac, sizeof(config->mac));
1243 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1244 vtpci_read_dev_config(hw,
1245 offsetof(struct virtio_net_config, status),
1246 &config->status, sizeof(config->status));
1249 "VIRTIO_NET_F_STATUS is not supported");
1253 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1254 vtpci_read_dev_config(hw,
1255 offsetof(struct virtio_net_config, max_virtqueue_pairs),
1256 &config->max_virtqueue_pairs,
1257 sizeof(config->max_virtqueue_pairs));
1260 "VIRTIO_NET_F_MQ is not supported");
1261 config->max_virtqueue_pairs = 1;
1264 hw->max_queue_pairs = config->max_virtqueue_pairs;
1266 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1267 config->max_virtqueue_pairs);
1268 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1270 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1271 config->mac[0], config->mac[1],
1272 config->mac[2], config->mac[3],
1273 config->mac[4], config->mac[5]);
1275 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1276 hw->max_queue_pairs = 1;
1279 ret = virtio_alloc_queues(eth_dev);
1282 vtpci_reinit_complete(hw);
1285 PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1286 eth_dev->data->port_id, pci_dev->id.vendor_id,
1287 pci_dev->id.device_id);
1293 * This function is based on probe() function in virtio_pci.c
1294 * It returns 0 on success.
1297 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1299 struct virtio_hw *hw = eth_dev->data->dev_private;
1300 uint32_t dev_flags = RTE_ETH_DEV_DETACHABLE;
1303 RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr_mrg_rxbuf));
1305 eth_dev->dev_ops = &virtio_eth_dev_ops;
1306 eth_dev->tx_pkt_burst = &virtio_xmit_pkts;
1308 if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1309 rx_func_get(eth_dev);
1313 /* Allocate memory for storing MAC addresses */
1314 eth_dev->data->mac_addrs = rte_zmalloc("virtio", VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN, 0);
1315 if (eth_dev->data->mac_addrs == NULL) {
1317 "Failed to allocate %d bytes needed to store MAC addresses",
1318 VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN);
1322 /* For virtio_user case the hw->virtio_user_dev is populated by
1323 * virtio_user_eth_dev_alloc() before eth_virtio_dev_init() is called.
1325 if (!hw->virtio_user_dev) {
1326 ret = vtpci_init(RTE_DEV_TO_PCI(eth_dev->device), hw,
1332 eth_dev->data->dev_flags = dev_flags;
1334 /* reset device and negotiate default features */
1335 ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1339 /* Setup interrupt callback */
1340 if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1341 rte_intr_callback_register(vtpci_intr_handle(hw),
1342 virtio_interrupt_handler, eth_dev);
1348 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1350 struct virtio_hw *hw = eth_dev->data->dev_private;
1352 PMD_INIT_FUNC_TRACE();
1354 if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1357 virtio_dev_stop(eth_dev);
1358 virtio_dev_close(eth_dev);
1360 eth_dev->dev_ops = NULL;
1361 eth_dev->tx_pkt_burst = NULL;
1362 eth_dev->rx_pkt_burst = NULL;
1364 rte_free(eth_dev->data->mac_addrs);
1365 eth_dev->data->mac_addrs = NULL;
1367 /* reset interrupt callback */
1368 if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1369 rte_intr_callback_unregister(vtpci_intr_handle(hw),
1370 virtio_interrupt_handler,
1373 rte_eal_pci_unmap_device(hw->dev);
1375 PMD_INIT_LOG(DEBUG, "dev_uninit completed");
1380 static struct eth_driver rte_virtio_pmd = {
1383 .name = "net_virtio",
1385 .id_table = pci_id_virtio_map,
1386 .drv_flags = RTE_PCI_DRV_DETACHABLE,
1387 .probe = rte_eth_dev_pci_probe,
1388 .remove = rte_eth_dev_pci_remove,
1390 .eth_dev_init = eth_virtio_dev_init,
1391 .eth_dev_uninit = eth_virtio_dev_uninit,
1392 .dev_private_size = sizeof(struct virtio_hw),
1395 RTE_INIT(rte_virtio_pmd_init);
1397 rte_virtio_pmd_init(void)
1399 if (rte_eal_iopl_init() != 0) {
1400 PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1404 rte_eal_pci_register(&rte_virtio_pmd.pci_drv);
1408 * Configure virtio device
1409 * It returns 0 on success.
1412 virtio_dev_configure(struct rte_eth_dev *dev)
1414 const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1415 struct virtio_hw *hw = dev->data->dev_private;
1416 uint64_t req_features;
1419 PMD_INIT_LOG(DEBUG, "configure");
1420 req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
1421 if (rxmode->hw_ip_checksum)
1422 req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
1423 if (rxmode->enable_lro)
1425 (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
1426 (1ULL << VIRTIO_NET_F_GUEST_TSO6);
1428 /* if request features changed, reinit the device */
1429 if (req_features != hw->req_guest_features) {
1430 ret = virtio_init_device(dev, req_features);
1435 if (rxmode->hw_ip_checksum &&
1436 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
1438 "rx ip checksum not available on this host");
1442 if (rxmode->enable_lro &&
1443 (!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
1444 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4))) {
1446 "lro not available on this host");
1450 /* start control queue */
1451 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
1452 virtio_dev_cq_start(dev);
1454 hw->vlan_strip = rxmode->hw_vlan_strip;
1456 if (rxmode->hw_vlan_filter
1457 && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
1459 "vlan filtering not available on this host");
1463 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1464 if (vtpci_irq_config(hw, 0) == VIRTIO_MSI_NO_VECTOR) {
1465 PMD_DRV_LOG(ERR, "failed to set config vector");
1474 virtio_dev_start(struct rte_eth_dev *dev)
1476 uint16_t nb_queues, i;
1477 struct virtnet_rx *rxvq;
1478 struct virtnet_tx *txvq __rte_unused;
1479 struct virtio_hw *hw = dev->data->dev_private;
1481 /* check if lsc interrupt feature is enabled */
1482 if (dev->data->dev_conf.intr_conf.lsc) {
1483 if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
1484 PMD_DRV_LOG(ERR, "link status not supported by host");
1488 if (rte_intr_enable(vtpci_intr_handle(hw)) < 0) {
1489 PMD_DRV_LOG(ERR, "interrupt enable failed");
1494 /* Initialize Link state */
1495 virtio_dev_link_update(dev, 0);
1497 /*Notify the backend
1498 *Otherwise the tap backend might already stop its queue due to fullness.
1499 *vhost backend will have no chance to be waked up
1501 nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
1502 if (hw->max_queue_pairs > 1) {
1503 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
1507 PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
1509 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1510 rxvq = dev->data->rx_queues[i];
1511 virtqueue_notify(rxvq->vq);
1514 PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
1516 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1517 rxvq = dev->data->rx_queues[i];
1518 VIRTQUEUE_DUMP(rxvq->vq);
1521 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1522 txvq = dev->data->tx_queues[i];
1523 VIRTQUEUE_DUMP(txvq->vq);
1529 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
1531 struct rte_mbuf *buf;
1532 int i, mbuf_num = 0;
1534 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1535 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1538 "Before freeing rxq[%d] used and unused buf", i);
1539 VIRTQUEUE_DUMP(rxvq->vq);
1541 PMD_INIT_LOG(DEBUG, "rx_queues[%d]=%p", i, rxvq);
1542 while ((buf = virtqueue_detatch_unused(rxvq->vq)) != NULL) {
1543 rte_pktmbuf_free(buf);
1547 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1549 "After freeing rxq[%d] used and unused buf", i);
1550 VIRTQUEUE_DUMP(rxvq->vq);
1553 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1554 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1557 "Before freeing txq[%d] used and unused bufs",
1559 VIRTQUEUE_DUMP(txvq->vq);
1562 while ((buf = virtqueue_detatch_unused(txvq->vq)) != NULL) {
1563 rte_pktmbuf_free(buf);
1567 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1569 "After freeing txq[%d] used and unused buf", i);
1570 VIRTQUEUE_DUMP(txvq->vq);
1575 * Stop device: disable interrupt and mark link down
1578 virtio_dev_stop(struct rte_eth_dev *dev)
1580 struct virtio_hw *hw = dev->data->dev_private;
1581 struct rte_eth_link link;
1583 PMD_INIT_LOG(DEBUG, "stop");
1585 if (dev->data->dev_conf.intr_conf.lsc)
1586 rte_intr_disable(vtpci_intr_handle(hw));
1588 memset(&link, 0, sizeof(link));
1589 virtio_dev_atomic_write_link_status(dev, &link);
1593 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
1595 struct rte_eth_link link, old;
1597 struct virtio_hw *hw = dev->data->dev_private;
1598 memset(&link, 0, sizeof(link));
1599 virtio_dev_atomic_read_link_status(dev, &link);
1601 link.link_duplex = ETH_LINK_FULL_DUPLEX;
1602 link.link_speed = SPEED_10G;
1604 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1605 PMD_INIT_LOG(DEBUG, "Get link status from hw");
1606 vtpci_read_dev_config(hw,
1607 offsetof(struct virtio_net_config, status),
1608 &status, sizeof(status));
1609 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
1610 link.link_status = ETH_LINK_DOWN;
1611 PMD_INIT_LOG(DEBUG, "Port %d is down",
1612 dev->data->port_id);
1614 link.link_status = ETH_LINK_UP;
1615 PMD_INIT_LOG(DEBUG, "Port %d is up",
1616 dev->data->port_id);
1619 link.link_status = ETH_LINK_UP;
1621 virtio_dev_atomic_write_link_status(dev, &link);
1623 return (old.link_status == link.link_status) ? -1 : 0;
1627 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1630 struct virtio_hw *hw = dev->data->dev_private;
1632 dev_info->pci_dev = hw->dev;
1633 dev_info->max_rx_queues =
1634 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
1635 dev_info->max_tx_queues =
1636 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
1637 dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
1638 dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
1639 dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
1640 dev_info->default_txconf = (struct rte_eth_txconf) {
1641 .txq_flags = ETH_TXQ_FLAGS_NOOFFLOADS
1643 dev_info->rx_offload_capa =
1644 DEV_RX_OFFLOAD_TCP_CKSUM |
1645 DEV_RX_OFFLOAD_UDP_CKSUM |
1646 DEV_RX_OFFLOAD_TCP_LRO;
1647 dev_info->tx_offload_capa = 0;
1649 if (hw->guest_features & (1ULL << VIRTIO_NET_F_CSUM)) {
1650 dev_info->tx_offload_capa |=
1651 DEV_TX_OFFLOAD_UDP_CKSUM |
1652 DEV_TX_OFFLOAD_TCP_CKSUM;
1655 tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
1656 (1ULL << VIRTIO_NET_F_HOST_TSO6);
1657 if ((hw->guest_features & tso_mask) == tso_mask)
1658 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
1662 * It enables testpmd to collect per queue stats.
1665 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
1666 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
1667 __rte_unused uint8_t is_rx)
1672 RTE_PMD_EXPORT_NAME(net_virtio, __COUNTER__);
1673 RTE_PMD_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map);
1674 RTE_PMD_REGISTER_KMOD_DEP(net_virtio, "* igb_uio | uio_pci_generic | vfio");