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]))
155 struct virtio_hw_internal virtio_hw_internal[RTE_MAX_ETHPORTS];
158 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
159 int *dlen, int pkt_num)
163 virtio_net_ctrl_ack status = ~0;
164 struct virtio_pmd_ctrl result;
165 struct virtqueue *vq;
167 ctrl->status = status;
169 if (!cvq || !cvq->vq) {
170 PMD_INIT_LOG(ERR, "Control queue is not supported.");
174 head = vq->vq_desc_head_idx;
176 PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
177 "vq->hw->cvq = %p vq = %p",
178 vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
180 if ((vq->vq_free_cnt < ((uint32_t)pkt_num + 2)) || (pkt_num < 1))
183 memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
184 sizeof(struct virtio_pmd_ctrl));
187 * Format is enforced in qemu code:
188 * One TX packet for header;
189 * At least one TX packet per argument;
190 * One RX packet for ACK.
192 vq->vq_ring.desc[head].flags = VRING_DESC_F_NEXT;
193 vq->vq_ring.desc[head].addr = cvq->virtio_net_hdr_mem;
194 vq->vq_ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
196 i = vq->vq_ring.desc[head].next;
198 for (k = 0; k < pkt_num; k++) {
199 vq->vq_ring.desc[i].flags = VRING_DESC_F_NEXT;
200 vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem
201 + sizeof(struct virtio_net_ctrl_hdr)
202 + sizeof(ctrl->status) + sizeof(uint8_t)*sum;
203 vq->vq_ring.desc[i].len = dlen[k];
206 i = vq->vq_ring.desc[i].next;
209 vq->vq_ring.desc[i].flags = VRING_DESC_F_WRITE;
210 vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem
211 + sizeof(struct virtio_net_ctrl_hdr);
212 vq->vq_ring.desc[i].len = sizeof(ctrl->status);
215 vq->vq_desc_head_idx = vq->vq_ring.desc[i].next;
217 vq_update_avail_ring(vq, head);
218 vq_update_avail_idx(vq);
220 PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
222 virtqueue_notify(vq);
225 while (VIRTQUEUE_NUSED(vq) == 0) {
230 while (VIRTQUEUE_NUSED(vq)) {
231 uint32_t idx, desc_idx, used_idx;
232 struct vring_used_elem *uep;
234 used_idx = (uint32_t)(vq->vq_used_cons_idx
235 & (vq->vq_nentries - 1));
236 uep = &vq->vq_ring.used->ring[used_idx];
237 idx = (uint32_t) uep->id;
240 while (vq->vq_ring.desc[desc_idx].flags & VRING_DESC_F_NEXT) {
241 desc_idx = vq->vq_ring.desc[desc_idx].next;
245 vq->vq_ring.desc[desc_idx].next = vq->vq_desc_head_idx;
246 vq->vq_desc_head_idx = idx;
248 vq->vq_used_cons_idx++;
252 PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
253 vq->vq_free_cnt, vq->vq_desc_head_idx);
255 memcpy(&result, cvq->virtio_net_hdr_mz->addr,
256 sizeof(struct virtio_pmd_ctrl));
258 return result.status;
262 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
264 struct virtio_hw *hw = dev->data->dev_private;
265 struct virtio_pmd_ctrl ctrl;
269 ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
270 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
271 memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
273 dlen[0] = sizeof(uint16_t);
275 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
277 PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
278 "failed, this is too late now...");
286 virtio_dev_queue_release(void *queue __rte_unused)
292 virtio_get_queue_type(struct virtio_hw *hw, uint16_t vtpci_queue_idx)
294 if (vtpci_queue_idx == hw->max_queue_pairs * 2)
296 else if (vtpci_queue_idx % 2 == 0)
303 virtio_get_nr_vq(struct virtio_hw *hw)
305 uint16_t nr_vq = hw->max_queue_pairs * 2;
307 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
314 virtio_init_vring(struct virtqueue *vq)
316 int size = vq->vq_nentries;
317 struct vring *vr = &vq->vq_ring;
318 uint8_t *ring_mem = vq->vq_ring_virt_mem;
320 PMD_INIT_FUNC_TRACE();
323 * Reinitialise since virtio port might have been stopped and restarted
325 memset(ring_mem, 0, vq->vq_ring_size);
326 vring_init(vr, size, ring_mem, VIRTIO_PCI_VRING_ALIGN);
327 vq->vq_used_cons_idx = 0;
328 vq->vq_desc_head_idx = 0;
329 vq->vq_avail_idx = 0;
330 vq->vq_desc_tail_idx = (uint16_t)(vq->vq_nentries - 1);
331 vq->vq_free_cnt = vq->vq_nentries;
332 memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries);
334 vring_desc_init(vr->desc, size);
337 * Disable device(host) interrupting guest
339 virtqueue_disable_intr(vq);
343 virtio_init_queue(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx)
345 char vq_name[VIRTQUEUE_MAX_NAME_SZ];
346 char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
347 const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
348 unsigned int vq_size, size;
349 struct virtio_hw *hw = dev->data->dev_private;
350 struct virtnet_rx *rxvq = NULL;
351 struct virtnet_tx *txvq = NULL;
352 struct virtnet_ctl *cvq = NULL;
353 struct virtqueue *vq;
354 size_t sz_hdr_mz = 0;
355 void *sw_ring = NULL;
356 int queue_type = virtio_get_queue_type(hw, vtpci_queue_idx);
359 PMD_INIT_LOG(DEBUG, "setting up queue: %u", vtpci_queue_idx);
362 * Read the virtqueue size from the Queue Size field
363 * Always power of 2 and if 0 virtqueue does not exist
365 vq_size = VTPCI_OPS(hw)->get_queue_num(hw, vtpci_queue_idx);
366 PMD_INIT_LOG(DEBUG, "vq_size: %u", vq_size);
368 PMD_INIT_LOG(ERR, "virtqueue does not exist");
372 if (!rte_is_power_of_2(vq_size)) {
373 PMD_INIT_LOG(ERR, "virtqueue size is not powerof 2");
377 snprintf(vq_name, sizeof(vq_name), "port%d_vq%d",
378 dev->data->port_id, vtpci_queue_idx);
380 size = RTE_ALIGN_CEIL(sizeof(*vq) +
381 vq_size * sizeof(struct vq_desc_extra),
382 RTE_CACHE_LINE_SIZE);
383 if (queue_type == VTNET_TQ) {
385 * For each xmit packet, allocate a virtio_net_hdr
386 * and indirect ring elements
388 sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
389 } else if (queue_type == VTNET_CQ) {
390 /* Allocate a page for control vq command, data and status */
391 sz_hdr_mz = PAGE_SIZE;
394 vq = rte_zmalloc_socket(vq_name, size, RTE_CACHE_LINE_SIZE,
397 PMD_INIT_LOG(ERR, "can not allocate vq");
400 hw->vqs[vtpci_queue_idx] = vq;
403 vq->vq_queue_index = vtpci_queue_idx;
404 vq->vq_nentries = vq_size;
407 * Reserve a memzone for vring elements
409 size = vring_size(vq_size, VIRTIO_PCI_VRING_ALIGN);
410 vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
411 PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
412 size, vq->vq_ring_size);
414 mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
416 0, VIRTIO_PCI_VRING_ALIGN);
418 if (rte_errno == EEXIST)
419 mz = rte_memzone_lookup(vq_name);
426 memset(mz->addr, 0, sizeof(mz->len));
428 vq->vq_ring_mem = mz->phys_addr;
429 vq->vq_ring_virt_mem = mz->addr;
430 PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem: 0x%" PRIx64,
431 (uint64_t)mz->phys_addr);
432 PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64,
433 (uint64_t)(uintptr_t)mz->addr);
435 virtio_init_vring(vq);
438 snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_vq%d_hdr",
439 dev->data->port_id, vtpci_queue_idx);
440 hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
442 RTE_CACHE_LINE_SIZE);
443 if (hdr_mz == NULL) {
444 if (rte_errno == EEXIST)
445 hdr_mz = rte_memzone_lookup(vq_hdr_name);
446 if (hdr_mz == NULL) {
453 if (queue_type == VTNET_RQ) {
454 size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
455 sizeof(vq->sw_ring[0]);
457 sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
458 RTE_CACHE_LINE_SIZE, SOCKET_ID_ANY);
460 PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
465 vq->sw_ring = sw_ring;
468 rxvq->port_id = dev->data->port_id;
470 } else if (queue_type == VTNET_TQ) {
473 txvq->port_id = dev->data->port_id;
475 txvq->virtio_net_hdr_mz = hdr_mz;
476 txvq->virtio_net_hdr_mem = hdr_mz->phys_addr;
477 } else if (queue_type == VTNET_CQ) {
481 cvq->virtio_net_hdr_mz = hdr_mz;
482 cvq->virtio_net_hdr_mem = hdr_mz->phys_addr;
483 memset(cvq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
488 /* For virtio_user case (that is when hw->dev is NULL), we use
489 * virtual address. And we need properly set _offset_, please see
490 * VIRTIO_MBUF_DATA_DMA_ADDR in virtqueue.h for more information.
492 if (!hw->virtio_user_dev)
493 vq->offset = offsetof(struct rte_mbuf, buf_physaddr);
495 vq->vq_ring_mem = (uintptr_t)mz->addr;
496 vq->offset = offsetof(struct rte_mbuf, buf_addr);
497 if (queue_type == VTNET_TQ)
498 txvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
499 else if (queue_type == VTNET_CQ)
500 cvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
503 if (queue_type == VTNET_TQ) {
504 struct virtio_tx_region *txr;
508 memset(txr, 0, vq_size * sizeof(*txr));
509 for (i = 0; i < vq_size; i++) {
510 struct vring_desc *start_dp = txr[i].tx_indir;
512 vring_desc_init(start_dp, RTE_DIM(txr[i].tx_indir));
514 /* first indirect descriptor is always the tx header */
515 start_dp->addr = txvq->virtio_net_hdr_mem
517 + offsetof(struct virtio_tx_region, tx_hdr);
519 start_dp->len = hw->vtnet_hdr_size;
520 start_dp->flags = VRING_DESC_F_NEXT;
524 if (VTPCI_OPS(hw)->setup_queue(hw, vq) < 0) {
525 PMD_INIT_LOG(ERR, "setup_queue failed");
533 rte_memzone_free(hdr_mz);
534 rte_memzone_free(mz);
541 virtio_free_queues(struct virtio_hw *hw)
543 uint16_t nr_vq = virtio_get_nr_vq(hw);
544 struct virtqueue *vq;
548 for (i = 0; i < nr_vq; i++) {
553 queue_type = virtio_get_queue_type(hw, i);
554 if (queue_type == VTNET_RQ) {
555 rte_free(vq->sw_ring);
556 rte_memzone_free(vq->rxq.mz);
557 } else if (queue_type == VTNET_TQ) {
558 rte_memzone_free(vq->txq.mz);
559 rte_memzone_free(vq->txq.virtio_net_hdr_mz);
561 rte_memzone_free(vq->cq.mz);
562 rte_memzone_free(vq->cq.virtio_net_hdr_mz);
572 virtio_alloc_queues(struct rte_eth_dev *dev)
574 struct virtio_hw *hw = dev->data->dev_private;
575 uint16_t nr_vq = virtio_get_nr_vq(hw);
579 hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
581 PMD_INIT_LOG(ERR, "failed to allocate vqs");
585 for (i = 0; i < nr_vq; i++) {
586 ret = virtio_init_queue(dev, i);
588 virtio_free_queues(hw);
596 static void virtio_queues_unbind_intr(struct rte_eth_dev *dev);
599 virtio_dev_close(struct rte_eth_dev *dev)
601 struct virtio_hw *hw = dev->data->dev_private;
602 struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
604 PMD_INIT_LOG(DEBUG, "virtio_dev_close");
607 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
608 VTPCI_OPS(hw)->set_config_irq(hw, VIRTIO_MSI_NO_VECTOR);
610 virtio_queues_unbind_intr(dev);
612 if (intr_conf->lsc || intr_conf->rxq) {
613 rte_intr_disable(dev->intr_handle);
614 rte_intr_efd_disable(dev->intr_handle);
615 rte_free(dev->intr_handle->intr_vec);
616 dev->intr_handle->intr_vec = NULL;
620 virtio_dev_free_mbufs(dev);
621 virtio_free_queues(hw);
625 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
627 struct virtio_hw *hw = dev->data->dev_private;
628 struct virtio_pmd_ctrl ctrl;
632 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
633 PMD_INIT_LOG(INFO, "host does not support rx control\n");
637 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
638 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
642 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
644 PMD_INIT_LOG(ERR, "Failed to enable promisc");
648 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
650 struct virtio_hw *hw = dev->data->dev_private;
651 struct virtio_pmd_ctrl ctrl;
655 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
656 PMD_INIT_LOG(INFO, "host does not support rx control\n");
660 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
661 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
665 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
667 PMD_INIT_LOG(ERR, "Failed to disable promisc");
671 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
673 struct virtio_hw *hw = dev->data->dev_private;
674 struct virtio_pmd_ctrl ctrl;
678 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
679 PMD_INIT_LOG(INFO, "host does not support rx control\n");
683 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
684 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
688 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
690 PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
694 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
696 struct virtio_hw *hw = dev->data->dev_private;
697 struct virtio_pmd_ctrl ctrl;
701 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
702 PMD_INIT_LOG(INFO, "host does not support rx control\n");
706 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
707 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
711 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
713 PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
716 #define VLAN_TAG_LEN 4 /* 802.3ac tag (not DMA'd) */
718 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
720 struct virtio_hw *hw = dev->data->dev_private;
721 uint32_t ether_hdr_len = ETHER_HDR_LEN + VLAN_TAG_LEN +
723 uint32_t frame_size = mtu + ether_hdr_len;
725 if (mtu < ETHER_MIN_MTU || frame_size > VIRTIO_MAX_RX_PKTLEN) {
726 PMD_INIT_LOG(ERR, "MTU should be between %d and %d\n",
727 ETHER_MIN_MTU, VIRTIO_MAX_RX_PKTLEN - ether_hdr_len);
734 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
736 struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
737 struct virtqueue *vq = rxvq->vq;
739 virtqueue_enable_intr(vq);
744 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
746 struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
747 struct virtqueue *vq = rxvq->vq;
749 virtqueue_disable_intr(vq);
754 * dev_ops for virtio, bare necessities for basic operation
756 static const struct eth_dev_ops virtio_eth_dev_ops = {
757 .dev_configure = virtio_dev_configure,
758 .dev_start = virtio_dev_start,
759 .dev_stop = virtio_dev_stop,
760 .dev_close = virtio_dev_close,
761 .promiscuous_enable = virtio_dev_promiscuous_enable,
762 .promiscuous_disable = virtio_dev_promiscuous_disable,
763 .allmulticast_enable = virtio_dev_allmulticast_enable,
764 .allmulticast_disable = virtio_dev_allmulticast_disable,
765 .mtu_set = virtio_mtu_set,
766 .dev_infos_get = virtio_dev_info_get,
767 .stats_get = virtio_dev_stats_get,
768 .xstats_get = virtio_dev_xstats_get,
769 .xstats_get_names = virtio_dev_xstats_get_names,
770 .stats_reset = virtio_dev_stats_reset,
771 .xstats_reset = virtio_dev_stats_reset,
772 .link_update = virtio_dev_link_update,
773 .rx_queue_setup = virtio_dev_rx_queue_setup,
774 .rx_queue_intr_enable = virtio_dev_rx_queue_intr_enable,
775 .rx_queue_intr_disable = virtio_dev_rx_queue_intr_disable,
776 .rx_queue_release = virtio_dev_queue_release,
777 .rx_descriptor_done = virtio_dev_rx_queue_done,
778 .tx_queue_setup = virtio_dev_tx_queue_setup,
779 .tx_queue_release = virtio_dev_queue_release,
780 /* collect stats per queue */
781 .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
782 .vlan_filter_set = virtio_vlan_filter_set,
783 .mac_addr_add = virtio_mac_addr_add,
784 .mac_addr_remove = virtio_mac_addr_remove,
785 .mac_addr_set = virtio_mac_addr_set,
789 virtio_dev_atomic_read_link_status(struct rte_eth_dev *dev,
790 struct rte_eth_link *link)
792 struct rte_eth_link *dst = link;
793 struct rte_eth_link *src = &(dev->data->dev_link);
795 if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
796 *(uint64_t *)src) == 0)
803 * Atomically writes the link status information into global
804 * structure rte_eth_dev.
807 * - Pointer to the structure rte_eth_dev to read from.
808 * - Pointer to the buffer to be saved with the link status.
811 * - On success, zero.
812 * - On failure, negative value.
815 virtio_dev_atomic_write_link_status(struct rte_eth_dev *dev,
816 struct rte_eth_link *link)
818 struct rte_eth_link *dst = &(dev->data->dev_link);
819 struct rte_eth_link *src = link;
821 if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
822 *(uint64_t *)src) == 0)
829 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
833 for (i = 0; i < dev->data->nb_tx_queues; i++) {
834 const struct virtnet_tx *txvq = dev->data->tx_queues[i];
838 stats->opackets += txvq->stats.packets;
839 stats->obytes += txvq->stats.bytes;
840 stats->oerrors += txvq->stats.errors;
842 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
843 stats->q_opackets[i] = txvq->stats.packets;
844 stats->q_obytes[i] = txvq->stats.bytes;
848 for (i = 0; i < dev->data->nb_rx_queues; i++) {
849 const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
853 stats->ipackets += rxvq->stats.packets;
854 stats->ibytes += rxvq->stats.bytes;
855 stats->ierrors += rxvq->stats.errors;
857 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
858 stats->q_ipackets[i] = rxvq->stats.packets;
859 stats->q_ibytes[i] = rxvq->stats.bytes;
863 stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
866 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
867 struct rte_eth_xstat_name *xstats_names,
868 __rte_unused unsigned limit)
874 unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
875 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
877 if (xstats_names != NULL) {
878 /* Note: limit checked in rte_eth_xstats_names() */
880 for (i = 0; i < dev->data->nb_rx_queues; i++) {
881 struct virtqueue *rxvq = dev->data->rx_queues[i];
884 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
885 snprintf(xstats_names[count].name,
886 sizeof(xstats_names[count].name),
888 rte_virtio_rxq_stat_strings[t].name);
893 for (i = 0; i < dev->data->nb_tx_queues; i++) {
894 struct virtqueue *txvq = dev->data->tx_queues[i];
897 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
898 snprintf(xstats_names[count].name,
899 sizeof(xstats_names[count].name),
901 rte_virtio_txq_stat_strings[t].name);
911 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
917 unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
918 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
923 for (i = 0; i < dev->data->nb_rx_queues; i++) {
924 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
931 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
932 xstats[count].value = *(uint64_t *)(((char *)rxvq) +
933 rte_virtio_rxq_stat_strings[t].offset);
934 xstats[count].id = count;
939 for (i = 0; i < dev->data->nb_tx_queues; i++) {
940 struct virtnet_tx *txvq = dev->data->tx_queues[i];
947 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
948 xstats[count].value = *(uint64_t *)(((char *)txvq) +
949 rte_virtio_txq_stat_strings[t].offset);
950 xstats[count].id = count;
959 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
961 virtio_update_stats(dev, stats);
965 virtio_dev_stats_reset(struct rte_eth_dev *dev)
969 for (i = 0; i < dev->data->nb_tx_queues; i++) {
970 struct virtnet_tx *txvq = dev->data->tx_queues[i];
974 txvq->stats.packets = 0;
975 txvq->stats.bytes = 0;
976 txvq->stats.errors = 0;
977 txvq->stats.multicast = 0;
978 txvq->stats.broadcast = 0;
979 memset(txvq->stats.size_bins, 0,
980 sizeof(txvq->stats.size_bins[0]) * 8);
983 for (i = 0; i < dev->data->nb_rx_queues; i++) {
984 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
988 rxvq->stats.packets = 0;
989 rxvq->stats.bytes = 0;
990 rxvq->stats.errors = 0;
991 rxvq->stats.multicast = 0;
992 rxvq->stats.broadcast = 0;
993 memset(rxvq->stats.size_bins, 0,
994 sizeof(rxvq->stats.size_bins[0]) * 8);
999 virtio_set_hwaddr(struct virtio_hw *hw)
1001 vtpci_write_dev_config(hw,
1002 offsetof(struct virtio_net_config, mac),
1003 &hw->mac_addr, ETHER_ADDR_LEN);
1007 virtio_get_hwaddr(struct virtio_hw *hw)
1009 if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
1010 vtpci_read_dev_config(hw,
1011 offsetof(struct virtio_net_config, mac),
1012 &hw->mac_addr, ETHER_ADDR_LEN);
1014 eth_random_addr(&hw->mac_addr[0]);
1015 virtio_set_hwaddr(hw);
1020 virtio_mac_table_set(struct virtio_hw *hw,
1021 const struct virtio_net_ctrl_mac *uc,
1022 const struct virtio_net_ctrl_mac *mc)
1024 struct virtio_pmd_ctrl ctrl;
1027 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1028 PMD_DRV_LOG(INFO, "host does not support mac table");
1032 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1033 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1035 len[0] = uc->entries * ETHER_ADDR_LEN + sizeof(uc->entries);
1036 memcpy(ctrl.data, uc, len[0]);
1038 len[1] = mc->entries * ETHER_ADDR_LEN + sizeof(mc->entries);
1039 memcpy(ctrl.data + len[0], mc, len[1]);
1041 err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1043 PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1047 virtio_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
1048 uint32_t index, uint32_t vmdq __rte_unused)
1050 struct virtio_hw *hw = dev->data->dev_private;
1051 const struct ether_addr *addrs = dev->data->mac_addrs;
1053 struct virtio_net_ctrl_mac *uc, *mc;
1055 if (index >= VIRTIO_MAX_MAC_ADDRS) {
1056 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1060 uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1062 mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1065 for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1066 const struct ether_addr *addr
1067 = (i == index) ? mac_addr : addrs + i;
1068 struct virtio_net_ctrl_mac *tbl
1069 = is_multicast_ether_addr(addr) ? mc : uc;
1071 memcpy(&tbl->macs[tbl->entries++], addr, ETHER_ADDR_LEN);
1074 virtio_mac_table_set(hw, uc, mc);
1078 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1080 struct virtio_hw *hw = dev->data->dev_private;
1081 struct ether_addr *addrs = dev->data->mac_addrs;
1082 struct virtio_net_ctrl_mac *uc, *mc;
1085 if (index >= VIRTIO_MAX_MAC_ADDRS) {
1086 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1090 uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1092 mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1095 for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1096 struct virtio_net_ctrl_mac *tbl;
1098 if (i == index || is_zero_ether_addr(addrs + i))
1101 tbl = is_multicast_ether_addr(addrs + i) ? mc : uc;
1102 memcpy(&tbl->macs[tbl->entries++], addrs + i, ETHER_ADDR_LEN);
1105 virtio_mac_table_set(hw, uc, mc);
1109 virtio_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
1111 struct virtio_hw *hw = dev->data->dev_private;
1113 memcpy(hw->mac_addr, mac_addr, ETHER_ADDR_LEN);
1115 /* Use atomic update if available */
1116 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1117 struct virtio_pmd_ctrl ctrl;
1118 int len = ETHER_ADDR_LEN;
1120 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1121 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1123 memcpy(ctrl.data, mac_addr, ETHER_ADDR_LEN);
1124 virtio_send_command(hw->cvq, &ctrl, &len, 1);
1125 } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1126 virtio_set_hwaddr(hw);
1130 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1132 struct virtio_hw *hw = dev->data->dev_private;
1133 struct virtio_pmd_ctrl ctrl;
1136 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1139 ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1140 ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1141 memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1142 len = sizeof(vlan_id);
1144 return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1148 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1150 uint64_t host_features;
1152 /* Prepare guest_features: feature that driver wants to support */
1153 PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1156 /* Read device(host) feature bits */
1157 host_features = VTPCI_OPS(hw)->get_features(hw);
1158 PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1162 * Negotiate features: Subset of device feature bits are written back
1163 * guest feature bits.
1165 hw->guest_features = req_features;
1166 hw->guest_features = vtpci_negotiate_features(hw, host_features);
1167 PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1168 hw->guest_features);
1171 if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1173 "VIRTIO_F_VERSION_1 features is not enabled.");
1176 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1177 if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1179 "failed to set FEATURES_OK status!");
1184 hw->req_guest_features = req_features;
1190 * Process Virtio Config changed interrupt and call the callback
1191 * if link state changed.
1194 virtio_interrupt_handler(struct rte_intr_handle *handle,
1197 struct rte_eth_dev *dev = param;
1198 struct virtio_hw *hw = dev->data->dev_private;
1201 /* Read interrupt status which clears interrupt */
1202 isr = vtpci_isr(hw);
1203 PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1205 if (rte_intr_enable(handle) < 0)
1206 PMD_DRV_LOG(ERR, "interrupt enable failed");
1208 if (isr & VIRTIO_PCI_ISR_CONFIG) {
1209 if (virtio_dev_link_update(dev, 0) == 0)
1210 _rte_eth_dev_callback_process(dev,
1211 RTE_ETH_EVENT_INTR_LSC, NULL);
1217 rx_func_get(struct rte_eth_dev *eth_dev)
1219 struct virtio_hw *hw = eth_dev->data->dev_private;
1220 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF))
1221 eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1223 eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1226 /* Only support 1:1 queue/interrupt mapping so far.
1227 * TODO: support n:1 queue/interrupt mapping when there are limited number of
1228 * interrupt vectors (<N+1).
1231 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1234 struct virtio_hw *hw = dev->data->dev_private;
1236 PMD_INIT_LOG(INFO, "queue/interrupt binding\n");
1237 for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1238 dev->intr_handle->intr_vec[i] = i + 1;
1239 if (VTPCI_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1240 VIRTIO_MSI_NO_VECTOR) {
1241 PMD_DRV_LOG(ERR, "failed to set queue vector");
1250 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1253 struct virtio_hw *hw = dev->data->dev_private;
1255 PMD_INIT_LOG(INFO, "queue/interrupt unbinding\n");
1256 for (i = 0; i < dev->data->nb_rx_queues; ++i)
1257 VTPCI_OPS(hw)->set_queue_irq(hw,
1258 hw->vqs[i * VTNET_CQ],
1259 VIRTIO_MSI_NO_VECTOR);
1263 virtio_configure_intr(struct rte_eth_dev *dev)
1265 struct virtio_hw *hw = dev->data->dev_private;
1267 if (!rte_intr_cap_multiple(dev->intr_handle)) {
1268 PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1272 if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1273 PMD_INIT_LOG(ERR, "Fail to create eventfd");
1277 if (!dev->intr_handle->intr_vec) {
1278 dev->intr_handle->intr_vec =
1279 rte_zmalloc("intr_vec",
1280 hw->max_queue_pairs * sizeof(int), 0);
1281 if (!dev->intr_handle->intr_vec) {
1282 PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1283 hw->max_queue_pairs);
1288 /* Re-register callback to update max_intr */
1289 rte_intr_callback_unregister(dev->intr_handle,
1290 virtio_interrupt_handler,
1292 rte_intr_callback_register(dev->intr_handle,
1293 virtio_interrupt_handler,
1296 /* DO NOT try to remove this! This function will enable msix, or QEMU
1297 * will encounter SIGSEGV when DRIVER_OK is sent.
1298 * And for legacy devices, this should be done before queue/vec binding
1299 * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1300 * (22) will be ignored.
1302 if (rte_intr_enable(dev->intr_handle) < 0) {
1303 PMD_DRV_LOG(ERR, "interrupt enable failed");
1307 if (virtio_queues_bind_intr(dev) < 0) {
1308 PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1315 /* reset device and renegotiate features if needed */
1317 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1319 struct virtio_hw *hw = eth_dev->data->dev_private;
1320 struct virtio_net_config *config;
1321 struct virtio_net_config local_config;
1322 struct rte_pci_device *pci_dev = NULL;
1325 /* Reset the device although not necessary at startup */
1328 /* Tell the host we've noticed this device. */
1329 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1331 /* Tell the host we've known how to drive the device. */
1332 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1333 if (virtio_negotiate_features(hw, req_features) < 0)
1336 if (eth_dev->device) {
1337 pci_dev = RTE_DEV_TO_PCI(eth_dev->device);
1338 rte_eth_copy_pci_info(eth_dev, pci_dev);
1341 /* If host does not support status then disable LSC */
1342 if (!vtpci_with_feature(hw, VIRTIO_NET_F_STATUS))
1343 eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1345 eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1347 rx_func_get(eth_dev);
1349 /* Setting up rx_header size for the device */
1350 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1351 vtpci_with_feature(hw, VIRTIO_F_VERSION_1))
1352 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1354 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1356 /* Copy the permanent MAC address to: virtio_hw */
1357 virtio_get_hwaddr(hw);
1358 ether_addr_copy((struct ether_addr *) hw->mac_addr,
1359 ð_dev->data->mac_addrs[0]);
1361 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1362 hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1363 hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1365 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1366 config = &local_config;
1368 vtpci_read_dev_config(hw,
1369 offsetof(struct virtio_net_config, mac),
1370 &config->mac, sizeof(config->mac));
1372 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1373 vtpci_read_dev_config(hw,
1374 offsetof(struct virtio_net_config, status),
1375 &config->status, sizeof(config->status));
1378 "VIRTIO_NET_F_STATUS is not supported");
1382 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1383 vtpci_read_dev_config(hw,
1384 offsetof(struct virtio_net_config, max_virtqueue_pairs),
1385 &config->max_virtqueue_pairs,
1386 sizeof(config->max_virtqueue_pairs));
1389 "VIRTIO_NET_F_MQ is not supported");
1390 config->max_virtqueue_pairs = 1;
1393 hw->max_queue_pairs = config->max_virtqueue_pairs;
1395 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1396 config->max_virtqueue_pairs);
1397 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1399 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1400 config->mac[0], config->mac[1],
1401 config->mac[2], config->mac[3],
1402 config->mac[4], config->mac[5]);
1404 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1405 hw->max_queue_pairs = 1;
1408 ret = virtio_alloc_queues(eth_dev);
1412 if (eth_dev->data->dev_conf.intr_conf.rxq) {
1413 if (virtio_configure_intr(eth_dev) < 0) {
1414 PMD_INIT_LOG(ERR, "failed to configure interrupt");
1419 vtpci_reinit_complete(hw);
1422 PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1423 eth_dev->data->port_id, pci_dev->id.vendor_id,
1424 pci_dev->id.device_id);
1430 * Remap the PCI device again (IO port map for legacy device and
1431 * memory map for modern device), so that the secondary process
1432 * could have the PCI initiated correctly.
1435 virtio_remap_pci(struct rte_pci_device *pci_dev, struct virtio_hw *hw)
1439 * We don't have to re-parse the PCI config space, since
1440 * rte_eal_pci_map_device() makes sure the mapped address
1441 * in secondary process would equal to the one mapped in
1442 * the primary process: error will be returned if that
1443 * requirement is not met.
1445 * That said, we could simply reuse all cap pointers
1446 * (such as dev_cfg, common_cfg, etc.) parsed from the
1447 * primary process, which is stored in shared memory.
1449 if (rte_eal_pci_map_device(pci_dev)) {
1450 PMD_INIT_LOG(DEBUG, "failed to map pci device!");
1454 if (rte_eal_pci_ioport_map(pci_dev, 0, VTPCI_IO(hw)) < 0)
1462 virtio_set_vtpci_ops(struct virtio_hw *hw)
1464 #ifdef RTE_VIRTIO_USER
1465 if (hw->virtio_user_dev)
1466 VTPCI_OPS(hw) = &virtio_user_ops;
1470 VTPCI_OPS(hw) = &modern_ops;
1472 VTPCI_OPS(hw) = &legacy_ops;
1476 * This function is based on probe() function in virtio_pci.c
1477 * It returns 0 on success.
1480 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1482 struct virtio_hw *hw = eth_dev->data->dev_private;
1483 uint32_t dev_flags = RTE_ETH_DEV_DETACHABLE;
1486 RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr_mrg_rxbuf));
1488 eth_dev->dev_ops = &virtio_eth_dev_ops;
1489 eth_dev->tx_pkt_burst = &virtio_xmit_pkts;
1491 if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1492 if (!hw->virtio_user_dev) {
1493 ret = virtio_remap_pci(RTE_DEV_TO_PCI(eth_dev->device),
1499 virtio_set_vtpci_ops(hw);
1500 if (hw->use_simple_rxtx) {
1501 eth_dev->tx_pkt_burst = virtio_xmit_pkts_simple;
1502 eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1504 rx_func_get(eth_dev);
1509 /* Allocate memory for storing MAC addresses */
1510 eth_dev->data->mac_addrs = rte_zmalloc("virtio", VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN, 0);
1511 if (eth_dev->data->mac_addrs == NULL) {
1513 "Failed to allocate %d bytes needed to store MAC addresses",
1514 VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN);
1518 hw->port_id = eth_dev->data->port_id;
1519 /* For virtio_user case the hw->virtio_user_dev is populated by
1520 * virtio_user_eth_dev_alloc() before eth_virtio_dev_init() is called.
1522 if (!hw->virtio_user_dev) {
1523 ret = vtpci_init(RTE_DEV_TO_PCI(eth_dev->device), hw,
1529 eth_dev->data->dev_flags = dev_flags;
1531 /* reset device and negotiate default features */
1532 ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1536 /* Setup interrupt callback */
1537 if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1538 rte_intr_callback_register(eth_dev->intr_handle,
1539 virtio_interrupt_handler, eth_dev);
1545 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1547 PMD_INIT_FUNC_TRACE();
1549 if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1552 virtio_dev_stop(eth_dev);
1553 virtio_dev_close(eth_dev);
1555 eth_dev->dev_ops = NULL;
1556 eth_dev->tx_pkt_burst = NULL;
1557 eth_dev->rx_pkt_burst = NULL;
1559 rte_free(eth_dev->data->mac_addrs);
1560 eth_dev->data->mac_addrs = NULL;
1562 /* reset interrupt callback */
1563 if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1564 rte_intr_callback_unregister(eth_dev->intr_handle,
1565 virtio_interrupt_handler,
1567 if (eth_dev->device)
1568 rte_eal_pci_unmap_device(RTE_DEV_TO_PCI(eth_dev->device));
1570 PMD_INIT_LOG(DEBUG, "dev_uninit completed");
1575 static struct eth_driver rte_virtio_pmd = {
1578 .name = "net_virtio",
1580 .id_table = pci_id_virtio_map,
1582 .probe = rte_eth_dev_pci_probe,
1583 .remove = rte_eth_dev_pci_remove,
1585 .eth_dev_init = eth_virtio_dev_init,
1586 .eth_dev_uninit = eth_virtio_dev_uninit,
1587 .dev_private_size = sizeof(struct virtio_hw),
1590 RTE_INIT(rte_virtio_pmd_init);
1592 rte_virtio_pmd_init(void)
1594 if (rte_eal_iopl_init() != 0) {
1595 PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1599 rte_eal_pci_register(&rte_virtio_pmd.pci_drv);
1603 * Configure virtio device
1604 * It returns 0 on success.
1607 virtio_dev_configure(struct rte_eth_dev *dev)
1609 const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1610 struct virtio_hw *hw = dev->data->dev_private;
1611 uint64_t req_features;
1614 PMD_INIT_LOG(DEBUG, "configure");
1615 req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
1616 if (rxmode->hw_ip_checksum)
1617 req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
1618 if (rxmode->enable_lro)
1620 (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
1621 (1ULL << VIRTIO_NET_F_GUEST_TSO6);
1623 /* if request features changed, reinit the device */
1624 if (req_features != hw->req_guest_features) {
1625 ret = virtio_init_device(dev, req_features);
1630 if (rxmode->hw_ip_checksum &&
1631 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
1633 "rx ip checksum not available on this host");
1637 if (rxmode->enable_lro &&
1638 (!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
1639 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4))) {
1641 "lro not available on this host");
1645 /* start control queue */
1646 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
1647 virtio_dev_cq_start(dev);
1649 hw->vlan_strip = rxmode->hw_vlan_strip;
1651 if (rxmode->hw_vlan_filter
1652 && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
1654 "vlan filtering not available on this host");
1658 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1659 /* Enable vector (0) for Link State Intrerrupt */
1660 if (VTPCI_OPS(hw)->set_config_irq(hw, 0) ==
1661 VIRTIO_MSI_NO_VECTOR) {
1662 PMD_DRV_LOG(ERR, "failed to set config vector");
1671 virtio_dev_start(struct rte_eth_dev *dev)
1673 uint16_t nb_queues, i;
1674 struct virtnet_rx *rxvq;
1675 struct virtnet_tx *txvq __rte_unused;
1676 struct virtio_hw *hw = dev->data->dev_private;
1678 /* check if lsc interrupt feature is enabled */
1679 if (dev->data->dev_conf.intr_conf.lsc) {
1680 if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
1681 PMD_DRV_LOG(ERR, "link status not supported by host");
1686 /* Enable uio/vfio intr/eventfd mapping: althrough we already did that
1687 * in device configure, but it could be unmapped when device is
1690 if (dev->data->dev_conf.intr_conf.lsc ||
1691 dev->data->dev_conf.intr_conf.rxq) {
1692 rte_intr_disable(dev->intr_handle);
1694 if (rte_intr_enable(dev->intr_handle) < 0) {
1695 PMD_DRV_LOG(ERR, "interrupt enable failed");
1700 /* Initialize Link state */
1701 virtio_dev_link_update(dev, 0);
1703 /*Notify the backend
1704 *Otherwise the tap backend might already stop its queue due to fullness.
1705 *vhost backend will have no chance to be waked up
1707 nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
1708 if (hw->max_queue_pairs > 1) {
1709 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
1713 PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
1715 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1716 rxvq = dev->data->rx_queues[i];
1717 virtqueue_notify(rxvq->vq);
1720 PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
1722 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1723 rxvq = dev->data->rx_queues[i];
1724 VIRTQUEUE_DUMP(rxvq->vq);
1727 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1728 txvq = dev->data->tx_queues[i];
1729 VIRTQUEUE_DUMP(txvq->vq);
1735 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
1737 struct rte_mbuf *buf;
1738 int i, mbuf_num = 0;
1740 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1741 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1744 "Before freeing rxq[%d] used and unused buf", i);
1745 VIRTQUEUE_DUMP(rxvq->vq);
1747 PMD_INIT_LOG(DEBUG, "rx_queues[%d]=%p", i, rxvq);
1748 while ((buf = virtqueue_detatch_unused(rxvq->vq)) != NULL) {
1749 rte_pktmbuf_free(buf);
1753 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1755 "After freeing rxq[%d] used and unused buf", i);
1756 VIRTQUEUE_DUMP(rxvq->vq);
1759 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1760 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1763 "Before freeing txq[%d] used and unused bufs",
1765 VIRTQUEUE_DUMP(txvq->vq);
1768 while ((buf = virtqueue_detatch_unused(txvq->vq)) != NULL) {
1769 rte_pktmbuf_free(buf);
1773 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1775 "After freeing txq[%d] used and unused buf", i);
1776 VIRTQUEUE_DUMP(txvq->vq);
1781 * Stop device: disable interrupt and mark link down
1784 virtio_dev_stop(struct rte_eth_dev *dev)
1786 struct rte_eth_link link;
1787 struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
1789 PMD_INIT_LOG(DEBUG, "stop");
1791 if (intr_conf->lsc || intr_conf->rxq)
1792 rte_intr_disable(dev->intr_handle);
1794 memset(&link, 0, sizeof(link));
1795 virtio_dev_atomic_write_link_status(dev, &link);
1799 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
1801 struct rte_eth_link link, old;
1803 struct virtio_hw *hw = dev->data->dev_private;
1804 memset(&link, 0, sizeof(link));
1805 virtio_dev_atomic_read_link_status(dev, &link);
1807 link.link_duplex = ETH_LINK_FULL_DUPLEX;
1808 link.link_speed = SPEED_10G;
1810 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1811 PMD_INIT_LOG(DEBUG, "Get link status from hw");
1812 vtpci_read_dev_config(hw,
1813 offsetof(struct virtio_net_config, status),
1814 &status, sizeof(status));
1815 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
1816 link.link_status = ETH_LINK_DOWN;
1817 PMD_INIT_LOG(DEBUG, "Port %d is down",
1818 dev->data->port_id);
1820 link.link_status = ETH_LINK_UP;
1821 PMD_INIT_LOG(DEBUG, "Port %d is up",
1822 dev->data->port_id);
1825 link.link_status = ETH_LINK_UP;
1827 virtio_dev_atomic_write_link_status(dev, &link);
1829 return (old.link_status == link.link_status) ? -1 : 0;
1833 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1835 uint64_t tso_mask, host_features;
1836 struct virtio_hw *hw = dev->data->dev_private;
1838 dev_info->pci_dev = dev->device ? RTE_DEV_TO_PCI(dev->device) : NULL;
1839 dev_info->max_rx_queues =
1840 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
1841 dev_info->max_tx_queues =
1842 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
1843 dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
1844 dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
1845 dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
1846 dev_info->default_txconf = (struct rte_eth_txconf) {
1847 .txq_flags = ETH_TXQ_FLAGS_NOOFFLOADS
1850 host_features = VTPCI_OPS(hw)->get_features(hw);
1851 dev_info->rx_offload_capa = 0;
1852 if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
1853 dev_info->rx_offload_capa |=
1854 DEV_RX_OFFLOAD_TCP_CKSUM |
1855 DEV_RX_OFFLOAD_UDP_CKSUM;
1857 tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
1858 (1ULL << VIRTIO_NET_F_GUEST_TSO6);
1859 if ((host_features & tso_mask) == tso_mask)
1860 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
1862 dev_info->tx_offload_capa = 0;
1863 if (hw->guest_features & (1ULL << VIRTIO_NET_F_CSUM)) {
1864 dev_info->tx_offload_capa |=
1865 DEV_TX_OFFLOAD_UDP_CKSUM |
1866 DEV_TX_OFFLOAD_TCP_CKSUM;
1868 tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
1869 (1ULL << VIRTIO_NET_F_HOST_TSO6);
1870 if ((hw->guest_features & tso_mask) == tso_mask)
1871 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
1875 * It enables testpmd to collect per queue stats.
1878 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
1879 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
1880 __rte_unused uint8_t is_rx)
1885 RTE_PMD_EXPORT_NAME(net_virtio, __COUNTER__);
1886 RTE_PMD_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map);
1887 RTE_PMD_REGISTER_KMOD_DEP(net_virtio, "* igb_uio | uio_pci_generic | vfio");