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_ethdev_pci.h>
42 #include <rte_memcpy.h>
43 #include <rte_string_fns.h>
44 #include <rte_memzone.h>
45 #include <rte_malloc.h>
46 #include <rte_atomic.h>
47 #include <rte_branch_prediction.h>
49 #include <rte_ether.h>
50 #include <rte_common.h>
51 #include <rte_errno.h>
53 #include <rte_memory.h>
57 #include "virtio_ethdev.h"
58 #include "virtio_pci.h"
59 #include "virtio_logs.h"
60 #include "virtqueue.h"
61 #include "virtio_rxtx.h"
63 static int eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev);
64 static int virtio_dev_configure(struct rte_eth_dev *dev);
65 static int virtio_dev_start(struct rte_eth_dev *dev);
66 static void virtio_dev_stop(struct rte_eth_dev *dev);
67 static void virtio_dev_promiscuous_enable(struct rte_eth_dev *dev);
68 static void virtio_dev_promiscuous_disable(struct rte_eth_dev *dev);
69 static void virtio_dev_allmulticast_enable(struct rte_eth_dev *dev);
70 static void virtio_dev_allmulticast_disable(struct rte_eth_dev *dev);
71 static void virtio_dev_info_get(struct rte_eth_dev *dev,
72 struct rte_eth_dev_info *dev_info);
73 static int virtio_dev_link_update(struct rte_eth_dev *dev,
74 int wait_to_complete);
76 static void virtio_set_hwaddr(struct virtio_hw *hw);
77 static void virtio_get_hwaddr(struct virtio_hw *hw);
79 static void virtio_dev_stats_get(struct rte_eth_dev *dev,
80 struct rte_eth_stats *stats);
81 static int virtio_dev_xstats_get(struct rte_eth_dev *dev,
82 struct rte_eth_xstat *xstats, unsigned n);
83 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
84 struct rte_eth_xstat_name *xstats_names,
86 static void virtio_dev_stats_reset(struct rte_eth_dev *dev);
87 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev);
88 static int virtio_vlan_filter_set(struct rte_eth_dev *dev,
89 uint16_t vlan_id, int on);
90 static int virtio_mac_addr_add(struct rte_eth_dev *dev,
91 struct ether_addr *mac_addr,
92 uint32_t index, uint32_t vmdq);
93 static void virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index);
94 static void virtio_mac_addr_set(struct rte_eth_dev *dev,
95 struct ether_addr *mac_addr);
97 static int virtio_dev_queue_stats_mapping_set(
98 struct rte_eth_dev *eth_dev,
104 * The set of PCI devices this driver supports
106 static const struct rte_pci_id pci_id_virtio_map[] = {
107 { RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_LEGACY_DEVICEID_NET) },
108 { RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_MODERN_DEVICEID_NET) },
109 { .vendor_id = 0, /* sentinel */ },
112 struct rte_virtio_xstats_name_off {
113 char name[RTE_ETH_XSTATS_NAME_SIZE];
117 /* [rt]x_qX_ is prepended to the name string here */
118 static const struct rte_virtio_xstats_name_off rte_virtio_rxq_stat_strings[] = {
119 {"good_packets", offsetof(struct virtnet_rx, stats.packets)},
120 {"good_bytes", offsetof(struct virtnet_rx, stats.bytes)},
121 {"errors", offsetof(struct virtnet_rx, stats.errors)},
122 {"multicast_packets", offsetof(struct virtnet_rx, stats.multicast)},
123 {"broadcast_packets", offsetof(struct virtnet_rx, stats.broadcast)},
124 {"undersize_packets", offsetof(struct virtnet_rx, stats.size_bins[0])},
125 {"size_64_packets", offsetof(struct virtnet_rx, stats.size_bins[1])},
126 {"size_65_127_packets", offsetof(struct virtnet_rx, stats.size_bins[2])},
127 {"size_128_255_packets", offsetof(struct virtnet_rx, stats.size_bins[3])},
128 {"size_256_511_packets", offsetof(struct virtnet_rx, stats.size_bins[4])},
129 {"size_512_1023_packets", offsetof(struct virtnet_rx, stats.size_bins[5])},
130 {"size_1024_1518_packets", offsetof(struct virtnet_rx, stats.size_bins[6])},
131 {"size_1519_max_packets", offsetof(struct virtnet_rx, stats.size_bins[7])},
134 /* [rt]x_qX_ is prepended to the name string here */
135 static const struct rte_virtio_xstats_name_off rte_virtio_txq_stat_strings[] = {
136 {"good_packets", offsetof(struct virtnet_tx, stats.packets)},
137 {"good_bytes", offsetof(struct virtnet_tx, stats.bytes)},
138 {"errors", offsetof(struct virtnet_tx, stats.errors)},
139 {"multicast_packets", offsetof(struct virtnet_tx, stats.multicast)},
140 {"broadcast_packets", offsetof(struct virtnet_tx, stats.broadcast)},
141 {"undersize_packets", offsetof(struct virtnet_tx, stats.size_bins[0])},
142 {"size_64_packets", offsetof(struct virtnet_tx, stats.size_bins[1])},
143 {"size_65_127_packets", offsetof(struct virtnet_tx, stats.size_bins[2])},
144 {"size_128_255_packets", offsetof(struct virtnet_tx, stats.size_bins[3])},
145 {"size_256_511_packets", offsetof(struct virtnet_tx, stats.size_bins[4])},
146 {"size_512_1023_packets", offsetof(struct virtnet_tx, stats.size_bins[5])},
147 {"size_1024_1518_packets", offsetof(struct virtnet_tx, stats.size_bins[6])},
148 {"size_1519_max_packets", offsetof(struct virtnet_tx, stats.size_bins[7])},
151 #define VIRTIO_NB_RXQ_XSTATS (sizeof(rte_virtio_rxq_stat_strings) / \
152 sizeof(rte_virtio_rxq_stat_strings[0]))
153 #define VIRTIO_NB_TXQ_XSTATS (sizeof(rte_virtio_txq_stat_strings) / \
154 sizeof(rte_virtio_txq_stat_strings[0]))
156 struct virtio_hw_internal virtio_hw_internal[RTE_MAX_ETHPORTS];
159 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
160 int *dlen, int pkt_num)
164 virtio_net_ctrl_ack status = ~0;
165 struct virtio_pmd_ctrl result;
166 struct virtqueue *vq;
168 ctrl->status = status;
170 if (!cvq || !cvq->vq) {
171 PMD_INIT_LOG(ERR, "Control queue is not supported.");
175 head = vq->vq_desc_head_idx;
177 PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
178 "vq->hw->cvq = %p vq = %p",
179 vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
181 if ((vq->vq_free_cnt < ((uint32_t)pkt_num + 2)) || (pkt_num < 1))
184 memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
185 sizeof(struct virtio_pmd_ctrl));
188 * Format is enforced in qemu code:
189 * One TX packet for header;
190 * At least one TX packet per argument;
191 * One RX packet for ACK.
193 vq->vq_ring.desc[head].flags = VRING_DESC_F_NEXT;
194 vq->vq_ring.desc[head].addr = cvq->virtio_net_hdr_mem;
195 vq->vq_ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
197 i = vq->vq_ring.desc[head].next;
199 for (k = 0; k < pkt_num; k++) {
200 vq->vq_ring.desc[i].flags = VRING_DESC_F_NEXT;
201 vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem
202 + sizeof(struct virtio_net_ctrl_hdr)
203 + sizeof(ctrl->status) + sizeof(uint8_t)*sum;
204 vq->vq_ring.desc[i].len = dlen[k];
207 i = vq->vq_ring.desc[i].next;
210 vq->vq_ring.desc[i].flags = VRING_DESC_F_WRITE;
211 vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem
212 + sizeof(struct virtio_net_ctrl_hdr);
213 vq->vq_ring.desc[i].len = sizeof(ctrl->status);
216 vq->vq_desc_head_idx = vq->vq_ring.desc[i].next;
218 vq_update_avail_ring(vq, head);
219 vq_update_avail_idx(vq);
221 PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
223 virtqueue_notify(vq);
226 while (VIRTQUEUE_NUSED(vq) == 0) {
231 while (VIRTQUEUE_NUSED(vq)) {
232 uint32_t idx, desc_idx, used_idx;
233 struct vring_used_elem *uep;
235 used_idx = (uint32_t)(vq->vq_used_cons_idx
236 & (vq->vq_nentries - 1));
237 uep = &vq->vq_ring.used->ring[used_idx];
238 idx = (uint32_t) uep->id;
241 while (vq->vq_ring.desc[desc_idx].flags & VRING_DESC_F_NEXT) {
242 desc_idx = vq->vq_ring.desc[desc_idx].next;
246 vq->vq_ring.desc[desc_idx].next = vq->vq_desc_head_idx;
247 vq->vq_desc_head_idx = idx;
249 vq->vq_used_cons_idx++;
253 PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
254 vq->vq_free_cnt, vq->vq_desc_head_idx);
256 memcpy(&result, cvq->virtio_net_hdr_mz->addr,
257 sizeof(struct virtio_pmd_ctrl));
259 return result.status;
263 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
265 struct virtio_hw *hw = dev->data->dev_private;
266 struct virtio_pmd_ctrl ctrl;
270 ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
271 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
272 memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
274 dlen[0] = sizeof(uint16_t);
276 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
278 PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
279 "failed, this is too late now...");
287 virtio_dev_queue_release(void *queue __rte_unused)
293 virtio_get_queue_type(struct virtio_hw *hw, uint16_t vtpci_queue_idx)
295 if (vtpci_queue_idx == hw->max_queue_pairs * 2)
297 else if (vtpci_queue_idx % 2 == 0)
304 virtio_get_nr_vq(struct virtio_hw *hw)
306 uint16_t nr_vq = hw->max_queue_pairs * 2;
308 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
315 virtio_init_vring(struct virtqueue *vq)
317 int size = vq->vq_nentries;
318 struct vring *vr = &vq->vq_ring;
319 uint8_t *ring_mem = vq->vq_ring_virt_mem;
321 PMD_INIT_FUNC_TRACE();
324 * Reinitialise since virtio port might have been stopped and restarted
326 memset(ring_mem, 0, vq->vq_ring_size);
327 vring_init(vr, size, ring_mem, VIRTIO_PCI_VRING_ALIGN);
328 vq->vq_used_cons_idx = 0;
329 vq->vq_desc_head_idx = 0;
330 vq->vq_avail_idx = 0;
331 vq->vq_desc_tail_idx = (uint16_t)(vq->vq_nentries - 1);
332 vq->vq_free_cnt = vq->vq_nentries;
333 memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries);
335 vring_desc_init(vr->desc, size);
338 * Disable device(host) interrupting guest
340 virtqueue_disable_intr(vq);
344 virtio_init_queue(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx)
346 char vq_name[VIRTQUEUE_MAX_NAME_SZ];
347 char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
348 const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
349 unsigned int vq_size, size;
350 struct virtio_hw *hw = dev->data->dev_private;
351 struct virtnet_rx *rxvq = NULL;
352 struct virtnet_tx *txvq = NULL;
353 struct virtnet_ctl *cvq = NULL;
354 struct virtqueue *vq;
355 size_t sz_hdr_mz = 0;
356 void *sw_ring = NULL;
357 int queue_type = virtio_get_queue_type(hw, vtpci_queue_idx);
360 PMD_INIT_LOG(DEBUG, "setting up queue: %u", vtpci_queue_idx);
363 * Read the virtqueue size from the Queue Size field
364 * Always power of 2 and if 0 virtqueue does not exist
366 vq_size = VTPCI_OPS(hw)->get_queue_num(hw, vtpci_queue_idx);
367 PMD_INIT_LOG(DEBUG, "vq_size: %u", vq_size);
369 PMD_INIT_LOG(ERR, "virtqueue does not exist");
373 if (!rte_is_power_of_2(vq_size)) {
374 PMD_INIT_LOG(ERR, "virtqueue size is not powerof 2");
378 snprintf(vq_name, sizeof(vq_name), "port%d_vq%d",
379 dev->data->port_id, vtpci_queue_idx);
381 size = RTE_ALIGN_CEIL(sizeof(*vq) +
382 vq_size * sizeof(struct vq_desc_extra),
383 RTE_CACHE_LINE_SIZE);
384 if (queue_type == VTNET_TQ) {
386 * For each xmit packet, allocate a virtio_net_hdr
387 * and indirect ring elements
389 sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
390 } else if (queue_type == VTNET_CQ) {
391 /* Allocate a page for control vq command, data and status */
392 sz_hdr_mz = PAGE_SIZE;
395 vq = rte_zmalloc_socket(vq_name, size, RTE_CACHE_LINE_SIZE,
398 PMD_INIT_LOG(ERR, "can not allocate vq");
401 hw->vqs[vtpci_queue_idx] = vq;
404 vq->vq_queue_index = vtpci_queue_idx;
405 vq->vq_nentries = vq_size;
408 * Reserve a memzone for vring elements
410 size = vring_size(vq_size, VIRTIO_PCI_VRING_ALIGN);
411 vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
412 PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
413 size, vq->vq_ring_size);
415 mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
417 0, VIRTIO_PCI_VRING_ALIGN);
419 if (rte_errno == EEXIST)
420 mz = rte_memzone_lookup(vq_name);
427 memset(mz->addr, 0, mz->len);
429 vq->vq_ring_mem = mz->phys_addr;
430 vq->vq_ring_virt_mem = mz->addr;
431 PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem: 0x%" PRIx64,
432 (uint64_t)mz->phys_addr);
433 PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64,
434 (uint64_t)(uintptr_t)mz->addr);
436 virtio_init_vring(vq);
439 snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_vq%d_hdr",
440 dev->data->port_id, vtpci_queue_idx);
441 hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
443 RTE_CACHE_LINE_SIZE);
444 if (hdr_mz == NULL) {
445 if (rte_errno == EEXIST)
446 hdr_mz = rte_memzone_lookup(vq_hdr_name);
447 if (hdr_mz == NULL) {
454 if (queue_type == VTNET_RQ) {
455 size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
456 sizeof(vq->sw_ring[0]);
458 sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
459 RTE_CACHE_LINE_SIZE, SOCKET_ID_ANY);
461 PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
466 vq->sw_ring = sw_ring;
469 rxvq->port_id = dev->data->port_id;
471 } else if (queue_type == VTNET_TQ) {
474 txvq->port_id = dev->data->port_id;
476 txvq->virtio_net_hdr_mz = hdr_mz;
477 txvq->virtio_net_hdr_mem = hdr_mz->phys_addr;
478 } else if (queue_type == VTNET_CQ) {
482 cvq->virtio_net_hdr_mz = hdr_mz;
483 cvq->virtio_net_hdr_mem = hdr_mz->phys_addr;
484 memset(cvq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
489 /* For virtio_user case (that is when hw->dev is NULL), we use
490 * virtual address. And we need properly set _offset_, please see
491 * VIRTIO_MBUF_DATA_DMA_ADDR in virtqueue.h for more information.
493 if (!hw->virtio_user_dev)
494 vq->offset = offsetof(struct rte_mbuf, buf_physaddr);
496 vq->vq_ring_mem = (uintptr_t)mz->addr;
497 vq->offset = offsetof(struct rte_mbuf, buf_addr);
498 if (queue_type == VTNET_TQ)
499 txvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
500 else if (queue_type == VTNET_CQ)
501 cvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
504 if (queue_type == VTNET_TQ) {
505 struct virtio_tx_region *txr;
509 memset(txr, 0, vq_size * sizeof(*txr));
510 for (i = 0; i < vq_size; i++) {
511 struct vring_desc *start_dp = txr[i].tx_indir;
513 vring_desc_init(start_dp, RTE_DIM(txr[i].tx_indir));
515 /* first indirect descriptor is always the tx header */
516 start_dp->addr = txvq->virtio_net_hdr_mem
518 + offsetof(struct virtio_tx_region, tx_hdr);
520 start_dp->len = hw->vtnet_hdr_size;
521 start_dp->flags = VRING_DESC_F_NEXT;
525 if (VTPCI_OPS(hw)->setup_queue(hw, vq) < 0) {
526 PMD_INIT_LOG(ERR, "setup_queue failed");
534 rte_memzone_free(hdr_mz);
535 rte_memzone_free(mz);
542 virtio_free_queues(struct virtio_hw *hw)
544 uint16_t nr_vq = virtio_get_nr_vq(hw);
545 struct virtqueue *vq;
552 for (i = 0; i < nr_vq; i++) {
557 queue_type = virtio_get_queue_type(hw, i);
558 if (queue_type == VTNET_RQ) {
559 rte_free(vq->sw_ring);
560 rte_memzone_free(vq->rxq.mz);
561 } else if (queue_type == VTNET_TQ) {
562 rte_memzone_free(vq->txq.mz);
563 rte_memzone_free(vq->txq.virtio_net_hdr_mz);
565 rte_memzone_free(vq->cq.mz);
566 rte_memzone_free(vq->cq.virtio_net_hdr_mz);
578 virtio_alloc_queues(struct rte_eth_dev *dev)
580 struct virtio_hw *hw = dev->data->dev_private;
581 uint16_t nr_vq = virtio_get_nr_vq(hw);
585 hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
587 PMD_INIT_LOG(ERR, "failed to allocate vqs");
591 for (i = 0; i < nr_vq; i++) {
592 ret = virtio_init_queue(dev, i);
594 virtio_free_queues(hw);
602 static void virtio_queues_unbind_intr(struct rte_eth_dev *dev);
605 virtio_dev_close(struct rte_eth_dev *dev)
607 struct virtio_hw *hw = dev->data->dev_private;
608 struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
610 PMD_INIT_LOG(DEBUG, "virtio_dev_close");
613 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
614 VTPCI_OPS(hw)->set_config_irq(hw, VIRTIO_MSI_NO_VECTOR);
616 virtio_queues_unbind_intr(dev);
618 if (intr_conf->lsc || intr_conf->rxq) {
619 rte_intr_disable(dev->intr_handle);
620 rte_intr_efd_disable(dev->intr_handle);
621 rte_free(dev->intr_handle->intr_vec);
622 dev->intr_handle->intr_vec = NULL;
626 virtio_dev_free_mbufs(dev);
627 virtio_free_queues(hw);
631 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
633 struct virtio_hw *hw = dev->data->dev_private;
634 struct virtio_pmd_ctrl ctrl;
638 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
639 PMD_INIT_LOG(INFO, "host does not support rx control");
643 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
644 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
648 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
650 PMD_INIT_LOG(ERR, "Failed to enable promisc");
654 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
656 struct virtio_hw *hw = dev->data->dev_private;
657 struct virtio_pmd_ctrl ctrl;
661 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
662 PMD_INIT_LOG(INFO, "host does not support rx control");
666 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
667 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
671 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
673 PMD_INIT_LOG(ERR, "Failed to disable promisc");
677 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
679 struct virtio_hw *hw = dev->data->dev_private;
680 struct virtio_pmd_ctrl ctrl;
684 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
685 PMD_INIT_LOG(INFO, "host does not support rx control");
689 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
690 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
694 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
696 PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
700 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
702 struct virtio_hw *hw = dev->data->dev_private;
703 struct virtio_pmd_ctrl ctrl;
707 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
708 PMD_INIT_LOG(INFO, "host does not support rx control");
712 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
713 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
717 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
719 PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
722 #define VLAN_TAG_LEN 4 /* 802.3ac tag (not DMA'd) */
724 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
726 struct virtio_hw *hw = dev->data->dev_private;
727 uint32_t ether_hdr_len = ETHER_HDR_LEN + VLAN_TAG_LEN +
729 uint32_t frame_size = mtu + ether_hdr_len;
730 uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
732 max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
734 if (mtu < ETHER_MIN_MTU || frame_size > max_frame_size) {
735 PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
736 ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
743 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
745 struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
746 struct virtqueue *vq = rxvq->vq;
748 virtqueue_enable_intr(vq);
753 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
755 struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
756 struct virtqueue *vq = rxvq->vq;
758 virtqueue_disable_intr(vq);
763 * dev_ops for virtio, bare necessities for basic operation
765 static const struct eth_dev_ops virtio_eth_dev_ops = {
766 .dev_configure = virtio_dev_configure,
767 .dev_start = virtio_dev_start,
768 .dev_stop = virtio_dev_stop,
769 .dev_close = virtio_dev_close,
770 .promiscuous_enable = virtio_dev_promiscuous_enable,
771 .promiscuous_disable = virtio_dev_promiscuous_disable,
772 .allmulticast_enable = virtio_dev_allmulticast_enable,
773 .allmulticast_disable = virtio_dev_allmulticast_disable,
774 .mtu_set = virtio_mtu_set,
775 .dev_infos_get = virtio_dev_info_get,
776 .stats_get = virtio_dev_stats_get,
777 .xstats_get = virtio_dev_xstats_get,
778 .xstats_get_names = virtio_dev_xstats_get_names,
779 .stats_reset = virtio_dev_stats_reset,
780 .xstats_reset = virtio_dev_stats_reset,
781 .link_update = virtio_dev_link_update,
782 .rx_queue_setup = virtio_dev_rx_queue_setup,
783 .rx_queue_intr_enable = virtio_dev_rx_queue_intr_enable,
784 .rx_queue_intr_disable = virtio_dev_rx_queue_intr_disable,
785 .rx_queue_release = virtio_dev_queue_release,
786 .rx_descriptor_done = virtio_dev_rx_queue_done,
787 .tx_queue_setup = virtio_dev_tx_queue_setup,
788 .tx_queue_release = virtio_dev_queue_release,
789 /* collect stats per queue */
790 .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
791 .vlan_filter_set = virtio_vlan_filter_set,
792 .mac_addr_add = virtio_mac_addr_add,
793 .mac_addr_remove = virtio_mac_addr_remove,
794 .mac_addr_set = virtio_mac_addr_set,
798 virtio_dev_atomic_read_link_status(struct rte_eth_dev *dev,
799 struct rte_eth_link *link)
801 struct rte_eth_link *dst = link;
802 struct rte_eth_link *src = &(dev->data->dev_link);
804 if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
805 *(uint64_t *)src) == 0)
812 * Atomically writes the link status information into global
813 * structure rte_eth_dev.
816 * - Pointer to the structure rte_eth_dev to read from.
817 * - Pointer to the buffer to be saved with the link status.
820 * - On success, zero.
821 * - On failure, negative value.
824 virtio_dev_atomic_write_link_status(struct rte_eth_dev *dev,
825 struct rte_eth_link *link)
827 struct rte_eth_link *dst = &(dev->data->dev_link);
828 struct rte_eth_link *src = link;
830 if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
831 *(uint64_t *)src) == 0)
838 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
842 for (i = 0; i < dev->data->nb_tx_queues; i++) {
843 const struct virtnet_tx *txvq = dev->data->tx_queues[i];
847 stats->opackets += txvq->stats.packets;
848 stats->obytes += txvq->stats.bytes;
849 stats->oerrors += txvq->stats.errors;
851 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
852 stats->q_opackets[i] = txvq->stats.packets;
853 stats->q_obytes[i] = txvq->stats.bytes;
857 for (i = 0; i < dev->data->nb_rx_queues; i++) {
858 const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
862 stats->ipackets += rxvq->stats.packets;
863 stats->ibytes += rxvq->stats.bytes;
864 stats->ierrors += rxvq->stats.errors;
866 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
867 stats->q_ipackets[i] = rxvq->stats.packets;
868 stats->q_ibytes[i] = rxvq->stats.bytes;
872 stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
875 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
876 struct rte_eth_xstat_name *xstats_names,
877 __rte_unused unsigned limit)
883 unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
884 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
886 if (xstats_names != NULL) {
887 /* Note: limit checked in rte_eth_xstats_names() */
889 for (i = 0; i < dev->data->nb_rx_queues; i++) {
890 struct virtqueue *rxvq = dev->data->rx_queues[i];
893 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
894 snprintf(xstats_names[count].name,
895 sizeof(xstats_names[count].name),
897 rte_virtio_rxq_stat_strings[t].name);
902 for (i = 0; i < dev->data->nb_tx_queues; i++) {
903 struct virtqueue *txvq = dev->data->tx_queues[i];
906 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
907 snprintf(xstats_names[count].name,
908 sizeof(xstats_names[count].name),
910 rte_virtio_txq_stat_strings[t].name);
920 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
926 unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
927 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
932 for (i = 0; i < dev->data->nb_rx_queues; i++) {
933 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
940 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
941 xstats[count].value = *(uint64_t *)(((char *)rxvq) +
942 rte_virtio_rxq_stat_strings[t].offset);
943 xstats[count].id = count;
948 for (i = 0; i < dev->data->nb_tx_queues; i++) {
949 struct virtnet_tx *txvq = dev->data->tx_queues[i];
956 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
957 xstats[count].value = *(uint64_t *)(((char *)txvq) +
958 rte_virtio_txq_stat_strings[t].offset);
959 xstats[count].id = count;
968 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
970 virtio_update_stats(dev, stats);
974 virtio_dev_stats_reset(struct rte_eth_dev *dev)
978 for (i = 0; i < dev->data->nb_tx_queues; i++) {
979 struct virtnet_tx *txvq = dev->data->tx_queues[i];
983 txvq->stats.packets = 0;
984 txvq->stats.bytes = 0;
985 txvq->stats.errors = 0;
986 txvq->stats.multicast = 0;
987 txvq->stats.broadcast = 0;
988 memset(txvq->stats.size_bins, 0,
989 sizeof(txvq->stats.size_bins[0]) * 8);
992 for (i = 0; i < dev->data->nb_rx_queues; i++) {
993 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
997 rxvq->stats.packets = 0;
998 rxvq->stats.bytes = 0;
999 rxvq->stats.errors = 0;
1000 rxvq->stats.multicast = 0;
1001 rxvq->stats.broadcast = 0;
1002 memset(rxvq->stats.size_bins, 0,
1003 sizeof(rxvq->stats.size_bins[0]) * 8);
1008 virtio_set_hwaddr(struct virtio_hw *hw)
1010 vtpci_write_dev_config(hw,
1011 offsetof(struct virtio_net_config, mac),
1012 &hw->mac_addr, ETHER_ADDR_LEN);
1016 virtio_get_hwaddr(struct virtio_hw *hw)
1018 if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
1019 vtpci_read_dev_config(hw,
1020 offsetof(struct virtio_net_config, mac),
1021 &hw->mac_addr, ETHER_ADDR_LEN);
1023 eth_random_addr(&hw->mac_addr[0]);
1024 virtio_set_hwaddr(hw);
1029 virtio_mac_table_set(struct virtio_hw *hw,
1030 const struct virtio_net_ctrl_mac *uc,
1031 const struct virtio_net_ctrl_mac *mc)
1033 struct virtio_pmd_ctrl ctrl;
1036 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1037 PMD_DRV_LOG(INFO, "host does not support mac table");
1041 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1042 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1044 len[0] = uc->entries * ETHER_ADDR_LEN + sizeof(uc->entries);
1045 memcpy(ctrl.data, uc, len[0]);
1047 len[1] = mc->entries * ETHER_ADDR_LEN + sizeof(mc->entries);
1048 memcpy(ctrl.data + len[0], mc, len[1]);
1050 err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1052 PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1057 virtio_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
1058 uint32_t index, uint32_t vmdq __rte_unused)
1060 struct virtio_hw *hw = dev->data->dev_private;
1061 const struct ether_addr *addrs = dev->data->mac_addrs;
1063 struct virtio_net_ctrl_mac *uc, *mc;
1065 if (index >= VIRTIO_MAX_MAC_ADDRS) {
1066 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1070 uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1072 mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1075 for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1076 const struct ether_addr *addr
1077 = (i == index) ? mac_addr : addrs + i;
1078 struct virtio_net_ctrl_mac *tbl
1079 = is_multicast_ether_addr(addr) ? mc : uc;
1081 memcpy(&tbl->macs[tbl->entries++], addr, ETHER_ADDR_LEN);
1084 return virtio_mac_table_set(hw, uc, mc);
1088 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1090 struct virtio_hw *hw = dev->data->dev_private;
1091 struct ether_addr *addrs = dev->data->mac_addrs;
1092 struct virtio_net_ctrl_mac *uc, *mc;
1095 if (index >= VIRTIO_MAX_MAC_ADDRS) {
1096 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1100 uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1102 mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1105 for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1106 struct virtio_net_ctrl_mac *tbl;
1108 if (i == index || is_zero_ether_addr(addrs + i))
1111 tbl = is_multicast_ether_addr(addrs + i) ? mc : uc;
1112 memcpy(&tbl->macs[tbl->entries++], addrs + i, ETHER_ADDR_LEN);
1115 virtio_mac_table_set(hw, uc, mc);
1119 virtio_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
1121 struct virtio_hw *hw = dev->data->dev_private;
1123 memcpy(hw->mac_addr, mac_addr, ETHER_ADDR_LEN);
1125 /* Use atomic update if available */
1126 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1127 struct virtio_pmd_ctrl ctrl;
1128 int len = ETHER_ADDR_LEN;
1130 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1131 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1133 memcpy(ctrl.data, mac_addr, ETHER_ADDR_LEN);
1134 virtio_send_command(hw->cvq, &ctrl, &len, 1);
1135 } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1136 virtio_set_hwaddr(hw);
1140 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1142 struct virtio_hw *hw = dev->data->dev_private;
1143 struct virtio_pmd_ctrl ctrl;
1146 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1149 ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1150 ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1151 memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1152 len = sizeof(vlan_id);
1154 return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1158 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1160 uint64_t host_features;
1162 /* Prepare guest_features: feature that driver wants to support */
1163 PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1166 /* Read device(host) feature bits */
1167 host_features = VTPCI_OPS(hw)->get_features(hw);
1168 PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1171 /* If supported, ensure MTU value is valid before acknowledging it. */
1172 if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1173 struct virtio_net_config config;
1175 vtpci_read_dev_config(hw,
1176 offsetof(struct virtio_net_config, mtu),
1177 &config.mtu, sizeof(config.mtu));
1179 if (config.mtu < ETHER_MIN_MTU)
1180 req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1184 * Negotiate features: Subset of device feature bits are written back
1185 * guest feature bits.
1187 hw->guest_features = req_features;
1188 hw->guest_features = vtpci_negotiate_features(hw, host_features);
1189 PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1190 hw->guest_features);
1193 if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1195 "VIRTIO_F_VERSION_1 features is not enabled.");
1198 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1199 if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1201 "failed to set FEATURES_OK status!");
1206 hw->req_guest_features = req_features;
1212 * Process Virtio Config changed interrupt and call the callback
1213 * if link state changed.
1216 virtio_interrupt_handler(void *param)
1218 struct rte_eth_dev *dev = param;
1219 struct virtio_hw *hw = dev->data->dev_private;
1222 /* Read interrupt status which clears interrupt */
1223 isr = vtpci_isr(hw);
1224 PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1226 if (rte_intr_enable(dev->intr_handle) < 0)
1227 PMD_DRV_LOG(ERR, "interrupt enable failed");
1229 if (isr & VIRTIO_PCI_ISR_CONFIG) {
1230 if (virtio_dev_link_update(dev, 0) == 0)
1231 _rte_eth_dev_callback_process(dev,
1232 RTE_ETH_EVENT_INTR_LSC,
1239 rx_func_get(struct rte_eth_dev *eth_dev)
1241 struct virtio_hw *hw = eth_dev->data->dev_private;
1242 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF))
1243 eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1245 eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1248 /* Only support 1:1 queue/interrupt mapping so far.
1249 * TODO: support n:1 queue/interrupt mapping when there are limited number of
1250 * interrupt vectors (<N+1).
1253 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1256 struct virtio_hw *hw = dev->data->dev_private;
1258 PMD_INIT_LOG(INFO, "queue/interrupt binding");
1259 for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1260 dev->intr_handle->intr_vec[i] = i + 1;
1261 if (VTPCI_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1262 VIRTIO_MSI_NO_VECTOR) {
1263 PMD_DRV_LOG(ERR, "failed to set queue vector");
1272 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1275 struct virtio_hw *hw = dev->data->dev_private;
1277 PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1278 for (i = 0; i < dev->data->nb_rx_queues; ++i)
1279 VTPCI_OPS(hw)->set_queue_irq(hw,
1280 hw->vqs[i * VTNET_CQ],
1281 VIRTIO_MSI_NO_VECTOR);
1285 virtio_configure_intr(struct rte_eth_dev *dev)
1287 struct virtio_hw *hw = dev->data->dev_private;
1289 if (!rte_intr_cap_multiple(dev->intr_handle)) {
1290 PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1294 if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1295 PMD_INIT_LOG(ERR, "Fail to create eventfd");
1299 if (!dev->intr_handle->intr_vec) {
1300 dev->intr_handle->intr_vec =
1301 rte_zmalloc("intr_vec",
1302 hw->max_queue_pairs * sizeof(int), 0);
1303 if (!dev->intr_handle->intr_vec) {
1304 PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1305 hw->max_queue_pairs);
1310 /* Re-register callback to update max_intr */
1311 rte_intr_callback_unregister(dev->intr_handle,
1312 virtio_interrupt_handler,
1314 rte_intr_callback_register(dev->intr_handle,
1315 virtio_interrupt_handler,
1318 /* DO NOT try to remove this! This function will enable msix, or QEMU
1319 * will encounter SIGSEGV when DRIVER_OK is sent.
1320 * And for legacy devices, this should be done before queue/vec binding
1321 * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1322 * (22) will be ignored.
1324 if (rte_intr_enable(dev->intr_handle) < 0) {
1325 PMD_DRV_LOG(ERR, "interrupt enable failed");
1329 if (virtio_queues_bind_intr(dev) < 0) {
1330 PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1337 /* reset device and renegotiate features if needed */
1339 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1341 struct virtio_hw *hw = eth_dev->data->dev_private;
1342 struct virtio_net_config *config;
1343 struct virtio_net_config local_config;
1344 struct rte_pci_device *pci_dev = NULL;
1347 /* Reset the device although not necessary at startup */
1350 /* Tell the host we've noticed this device. */
1351 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1353 /* Tell the host we've known how to drive the device. */
1354 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1355 if (virtio_negotiate_features(hw, req_features) < 0)
1358 if (!hw->virtio_user_dev) {
1359 pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1360 rte_eth_copy_pci_info(eth_dev, pci_dev);
1363 eth_dev->data->dev_flags = RTE_ETH_DEV_DETACHABLE;
1364 /* If host does not support both status and MSI-X then disable LSC */
1365 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS) && hw->use_msix)
1366 eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1368 eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1370 rx_func_get(eth_dev);
1372 /* Setting up rx_header size for the device */
1373 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1374 vtpci_with_feature(hw, VIRTIO_F_VERSION_1))
1375 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1377 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1379 /* Copy the permanent MAC address to: virtio_hw */
1380 virtio_get_hwaddr(hw);
1381 ether_addr_copy((struct ether_addr *) hw->mac_addr,
1382 ð_dev->data->mac_addrs[0]);
1384 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1385 hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1386 hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1388 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1389 config = &local_config;
1391 vtpci_read_dev_config(hw,
1392 offsetof(struct virtio_net_config, mac),
1393 &config->mac, sizeof(config->mac));
1395 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1396 vtpci_read_dev_config(hw,
1397 offsetof(struct virtio_net_config, status),
1398 &config->status, sizeof(config->status));
1401 "VIRTIO_NET_F_STATUS is not supported");
1405 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1406 vtpci_read_dev_config(hw,
1407 offsetof(struct virtio_net_config, max_virtqueue_pairs),
1408 &config->max_virtqueue_pairs,
1409 sizeof(config->max_virtqueue_pairs));
1412 "VIRTIO_NET_F_MQ is not supported");
1413 config->max_virtqueue_pairs = 1;
1416 hw->max_queue_pairs = config->max_virtqueue_pairs;
1418 if (vtpci_with_feature(hw, VIRTIO_NET_F_MTU)) {
1419 vtpci_read_dev_config(hw,
1420 offsetof(struct virtio_net_config, mtu),
1422 sizeof(config->mtu));
1425 * MTU value has already been checked at negotiation
1426 * time, but check again in case it has changed since
1427 * then, which should not happen.
1429 if (config->mtu < ETHER_MIN_MTU) {
1430 PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1435 hw->max_mtu = config->mtu;
1436 /* Set initial MTU to maximum one supported by vhost */
1437 eth_dev->data->mtu = config->mtu;
1440 hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - ETHER_HDR_LEN -
1441 VLAN_TAG_LEN - hw->vtnet_hdr_size;
1444 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1445 config->max_virtqueue_pairs);
1446 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1448 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1449 config->mac[0], config->mac[1],
1450 config->mac[2], config->mac[3],
1451 config->mac[4], config->mac[5]);
1453 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1454 hw->max_queue_pairs = 1;
1457 ret = virtio_alloc_queues(eth_dev);
1461 if (eth_dev->data->dev_conf.intr_conf.rxq) {
1462 if (virtio_configure_intr(eth_dev) < 0) {
1463 PMD_INIT_LOG(ERR, "failed to configure interrupt");
1468 vtpci_reinit_complete(hw);
1471 PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1472 eth_dev->data->port_id, pci_dev->id.vendor_id,
1473 pci_dev->id.device_id);
1479 * Remap the PCI device again (IO port map for legacy device and
1480 * memory map for modern device), so that the secondary process
1481 * could have the PCI initiated correctly.
1484 virtio_remap_pci(struct rte_pci_device *pci_dev, struct virtio_hw *hw)
1488 * We don't have to re-parse the PCI config space, since
1489 * rte_pci_map_device() makes sure the mapped address
1490 * in secondary process would equal to the one mapped in
1491 * the primary process: error will be returned if that
1492 * requirement is not met.
1494 * That said, we could simply reuse all cap pointers
1495 * (such as dev_cfg, common_cfg, etc.) parsed from the
1496 * primary process, which is stored in shared memory.
1498 if (rte_pci_map_device(pci_dev)) {
1499 PMD_INIT_LOG(DEBUG, "failed to map pci device!");
1503 if (rte_pci_ioport_map(pci_dev, 0, VTPCI_IO(hw)) < 0)
1511 virtio_set_vtpci_ops(struct virtio_hw *hw)
1513 #ifdef RTE_VIRTIO_USER
1514 if (hw->virtio_user_dev)
1515 VTPCI_OPS(hw) = &virtio_user_ops;
1519 VTPCI_OPS(hw) = &modern_ops;
1521 VTPCI_OPS(hw) = &legacy_ops;
1525 * This function is based on probe() function in virtio_pci.c
1526 * It returns 0 on success.
1529 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1531 struct virtio_hw *hw = eth_dev->data->dev_private;
1534 RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr_mrg_rxbuf));
1536 eth_dev->dev_ops = &virtio_eth_dev_ops;
1537 eth_dev->tx_pkt_burst = &virtio_xmit_pkts;
1539 if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1540 if (!hw->virtio_user_dev) {
1541 ret = virtio_remap_pci(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1546 virtio_set_vtpci_ops(hw);
1547 if (hw->use_simple_rxtx) {
1548 eth_dev->tx_pkt_burst = virtio_xmit_pkts_simple;
1549 eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1551 rx_func_get(eth_dev);
1556 /* Allocate memory for storing MAC addresses */
1557 eth_dev->data->mac_addrs = rte_zmalloc("virtio", VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN, 0);
1558 if (eth_dev->data->mac_addrs == NULL) {
1560 "Failed to allocate %d bytes needed to store MAC addresses",
1561 VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN);
1565 hw->port_id = eth_dev->data->port_id;
1566 /* For virtio_user case the hw->virtio_user_dev is populated by
1567 * virtio_user_eth_dev_alloc() before eth_virtio_dev_init() is called.
1569 if (!hw->virtio_user_dev) {
1570 ret = vtpci_init(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1575 /* reset device and negotiate default features */
1576 ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1580 /* Setup interrupt callback */
1581 if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1582 rte_intr_callback_register(eth_dev->intr_handle,
1583 virtio_interrupt_handler, eth_dev);
1589 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1591 PMD_INIT_FUNC_TRACE();
1593 if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1596 virtio_dev_stop(eth_dev);
1597 virtio_dev_close(eth_dev);
1599 eth_dev->dev_ops = NULL;
1600 eth_dev->tx_pkt_burst = NULL;
1601 eth_dev->rx_pkt_burst = NULL;
1603 rte_free(eth_dev->data->mac_addrs);
1604 eth_dev->data->mac_addrs = NULL;
1606 /* reset interrupt callback */
1607 if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1608 rte_intr_callback_unregister(eth_dev->intr_handle,
1609 virtio_interrupt_handler,
1611 if (eth_dev->device)
1612 rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(eth_dev));
1614 PMD_INIT_LOG(DEBUG, "dev_uninit completed");
1619 static int eth_virtio_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1620 struct rte_pci_device *pci_dev)
1622 return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct virtio_hw),
1623 eth_virtio_dev_init);
1626 static int eth_virtio_pci_remove(struct rte_pci_device *pci_dev)
1628 return rte_eth_dev_pci_generic_remove(pci_dev, eth_virtio_dev_uninit);
1631 static struct rte_pci_driver rte_virtio_pmd = {
1633 .name = "net_virtio",
1635 .id_table = pci_id_virtio_map,
1637 .probe = eth_virtio_pci_probe,
1638 .remove = eth_virtio_pci_remove,
1641 RTE_INIT(rte_virtio_pmd_init);
1643 rte_virtio_pmd_init(void)
1645 if (rte_eal_iopl_init() != 0) {
1646 PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1650 rte_pci_register(&rte_virtio_pmd);
1654 * Configure virtio device
1655 * It returns 0 on success.
1658 virtio_dev_configure(struct rte_eth_dev *dev)
1660 const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1661 struct virtio_hw *hw = dev->data->dev_private;
1664 PMD_INIT_LOG(DEBUG, "configure");
1666 if (dev->data->dev_conf.intr_conf.rxq) {
1667 ret = virtio_init_device(dev, hw->req_guest_features);
1672 /* Virtio does L4 checksum but not L3! */
1673 if (rxmode->hw_ip_checksum) {
1675 "virtio does not support IP checksum");
1679 if (rxmode->enable_lro) {
1681 "virtio does not support Large Receive Offload");
1685 /* start control queue */
1686 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
1687 virtio_dev_cq_start(dev);
1689 hw->vlan_strip = rxmode->hw_vlan_strip;
1691 if (rxmode->hw_vlan_filter
1692 && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
1694 "vlan filtering not available on this host");
1698 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1699 /* Enable vector (0) for Link State Intrerrupt */
1700 if (VTPCI_OPS(hw)->set_config_irq(hw, 0) ==
1701 VIRTIO_MSI_NO_VECTOR) {
1702 PMD_DRV_LOG(ERR, "failed to set config vector");
1711 virtio_dev_start(struct rte_eth_dev *dev)
1713 uint16_t nb_queues, i;
1714 struct virtnet_rx *rxvq;
1715 struct virtnet_tx *txvq __rte_unused;
1716 struct virtio_hw *hw = dev->data->dev_private;
1718 /* check if lsc interrupt feature is enabled */
1719 if (dev->data->dev_conf.intr_conf.lsc) {
1720 if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
1721 PMD_DRV_LOG(ERR, "link status not supported by host");
1726 /* Enable uio/vfio intr/eventfd mapping: althrough we already did that
1727 * in device configure, but it could be unmapped when device is
1730 if (dev->data->dev_conf.intr_conf.lsc ||
1731 dev->data->dev_conf.intr_conf.rxq) {
1732 rte_intr_disable(dev->intr_handle);
1734 if (rte_intr_enable(dev->intr_handle) < 0) {
1735 PMD_DRV_LOG(ERR, "interrupt enable failed");
1740 /*Notify the backend
1741 *Otherwise the tap backend might already stop its queue due to fullness.
1742 *vhost backend will have no chance to be waked up
1744 nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
1745 if (hw->max_queue_pairs > 1) {
1746 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
1750 PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
1752 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1753 rxvq = dev->data->rx_queues[i];
1754 virtqueue_notify(rxvq->vq);
1757 PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
1759 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1760 rxvq = dev->data->rx_queues[i];
1761 VIRTQUEUE_DUMP(rxvq->vq);
1764 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1765 txvq = dev->data->tx_queues[i];
1766 VIRTQUEUE_DUMP(txvq->vq);
1771 /* Initialize Link state */
1772 virtio_dev_link_update(dev, 0);
1777 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
1779 struct rte_mbuf *buf;
1780 int i, mbuf_num = 0;
1782 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1783 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1786 "Before freeing rxq[%d] used and unused buf", i);
1787 VIRTQUEUE_DUMP(rxvq->vq);
1789 PMD_INIT_LOG(DEBUG, "rx_queues[%d]=%p", i, rxvq);
1790 while ((buf = virtqueue_detatch_unused(rxvq->vq)) != NULL) {
1791 rte_pktmbuf_free(buf);
1795 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1797 "After freeing rxq[%d] used and unused buf", i);
1798 VIRTQUEUE_DUMP(rxvq->vq);
1801 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1802 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1805 "Before freeing txq[%d] used and unused bufs",
1807 VIRTQUEUE_DUMP(txvq->vq);
1810 while ((buf = virtqueue_detatch_unused(txvq->vq)) != NULL) {
1811 rte_pktmbuf_free(buf);
1815 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1817 "After freeing txq[%d] used and unused buf", i);
1818 VIRTQUEUE_DUMP(txvq->vq);
1823 * Stop device: disable interrupt and mark link down
1826 virtio_dev_stop(struct rte_eth_dev *dev)
1828 struct virtio_hw *hw = dev->data->dev_private;
1829 struct rte_eth_link link;
1830 struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
1832 PMD_INIT_LOG(DEBUG, "stop");
1834 if (intr_conf->lsc || intr_conf->rxq)
1835 rte_intr_disable(dev->intr_handle);
1838 memset(&link, 0, sizeof(link));
1839 virtio_dev_atomic_write_link_status(dev, &link);
1843 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
1845 struct rte_eth_link link, old;
1847 struct virtio_hw *hw = dev->data->dev_private;
1848 memset(&link, 0, sizeof(link));
1849 virtio_dev_atomic_read_link_status(dev, &link);
1851 link.link_duplex = ETH_LINK_FULL_DUPLEX;
1852 link.link_speed = SPEED_10G;
1854 if (hw->started == 0) {
1855 link.link_status = ETH_LINK_DOWN;
1856 } else if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1857 PMD_INIT_LOG(DEBUG, "Get link status from hw");
1858 vtpci_read_dev_config(hw,
1859 offsetof(struct virtio_net_config, status),
1860 &status, sizeof(status));
1861 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
1862 link.link_status = ETH_LINK_DOWN;
1863 PMD_INIT_LOG(DEBUG, "Port %d is down",
1864 dev->data->port_id);
1866 link.link_status = ETH_LINK_UP;
1867 PMD_INIT_LOG(DEBUG, "Port %d is up",
1868 dev->data->port_id);
1871 link.link_status = ETH_LINK_UP;
1873 virtio_dev_atomic_write_link_status(dev, &link);
1875 return (old.link_status == link.link_status) ? -1 : 0;
1879 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1881 uint64_t tso_mask, host_features;
1882 struct virtio_hw *hw = dev->data->dev_private;
1884 dev_info->speed_capa = ETH_LINK_SPEED_10G; /* fake value */
1886 dev_info->pci_dev = dev->device ? RTE_ETH_DEV_TO_PCI(dev) : NULL;
1887 dev_info->max_rx_queues =
1888 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
1889 dev_info->max_tx_queues =
1890 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
1891 dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
1892 dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
1893 dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
1894 dev_info->default_txconf = (struct rte_eth_txconf) {
1895 .txq_flags = ETH_TXQ_FLAGS_NOOFFLOADS
1898 host_features = VTPCI_OPS(hw)->get_features(hw);
1899 dev_info->rx_offload_capa = 0;
1900 if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
1901 dev_info->rx_offload_capa |=
1902 DEV_RX_OFFLOAD_TCP_CKSUM |
1903 DEV_RX_OFFLOAD_UDP_CKSUM;
1905 tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
1906 (1ULL << VIRTIO_NET_F_GUEST_TSO6);
1908 dev_info->tx_offload_capa = 0;
1909 if (hw->guest_features & (1ULL << VIRTIO_NET_F_CSUM)) {
1910 dev_info->tx_offload_capa |=
1911 DEV_TX_OFFLOAD_UDP_CKSUM |
1912 DEV_TX_OFFLOAD_TCP_CKSUM;
1914 tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
1915 (1ULL << VIRTIO_NET_F_HOST_TSO6);
1916 if ((hw->guest_features & tso_mask) == tso_mask)
1917 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
1921 * It enables testpmd to collect per queue stats.
1924 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
1925 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
1926 __rte_unused uint8_t is_rx)
1931 RTE_PMD_EXPORT_NAME(net_virtio, __COUNTER__);
1932 RTE_PMD_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map);
1933 RTE_PMD_REGISTER_KMOD_DEP(net_virtio, "* igb_uio | uio_pci_generic | vfio-pci");