4 * Copyright(c) 2010-2014 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.
39 #ifdef RTE_EXEC_ENV_LINUXAPP
43 #include <rte_ethdev.h>
44 #include <rte_memcpy.h>
45 #include <rte_string_fns.h>
46 #include <rte_memzone.h>
47 #include <rte_malloc.h>
48 #include <rte_atomic.h>
49 #include <rte_branch_prediction.h>
51 #include <rte_ether.h>
52 #include <rte_common.h>
54 #include <rte_memory.h>
58 #include "virtio_ethdev.h"
59 #include "virtio_pci.h"
60 #include "virtio_logs.h"
61 #include "virtqueue.h"
64 static int eth_virtio_dev_init(struct eth_driver *eth_drv,
65 struct rte_eth_dev *eth_dev);
66 static int virtio_dev_configure(struct rte_eth_dev *dev);
67 static int virtio_dev_start(struct rte_eth_dev *dev);
68 static void virtio_dev_stop(struct rte_eth_dev *dev);
69 static void virtio_dev_promiscuous_enable(struct rte_eth_dev *dev);
70 static void virtio_dev_promiscuous_disable(struct rte_eth_dev *dev);
71 static void virtio_dev_allmulticast_enable(struct rte_eth_dev *dev);
72 static void virtio_dev_allmulticast_disable(struct rte_eth_dev *dev);
73 static void virtio_dev_info_get(struct rte_eth_dev *dev,
74 struct rte_eth_dev_info *dev_info);
75 static int virtio_dev_link_update(struct rte_eth_dev *dev,
76 __rte_unused int wait_to_complete);
78 static void virtio_set_hwaddr(struct virtio_hw *hw);
79 static void virtio_get_hwaddr(struct virtio_hw *hw);
81 static void virtio_dev_rx_queue_release(__rte_unused void *rxq);
82 static void virtio_dev_tx_queue_release(__rte_unused void *txq);
84 static void virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats);
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 struct rte_pci_id pci_id_virtio_map[] = {
107 #define RTE_PCI_DEV_ID_DECL_VIRTIO(vend, dev) {RTE_PCI_DEVICE(vend, dev)},
108 #include "rte_pci_dev_ids.h"
110 { .vendor_id = 0, /* sentinel */ },
114 virtio_send_command(struct virtqueue *vq, struct virtio_pmd_ctrl *ctrl,
115 int *dlen, int pkt_num)
117 uint32_t head = vq->vq_desc_head_idx, i;
119 virtio_net_ctrl_ack status = ~0;
120 struct virtio_pmd_ctrl result;
122 ctrl->status = status;
126 "%s(): Control queue is not supported.",
131 PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
132 "vq->hw->cvq = %p vq = %p",
133 vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
135 if ((vq->vq_free_cnt < ((uint32_t)pkt_num + 2)) || (pkt_num < 1))
138 memcpy(vq->virtio_net_hdr_mz->addr, ctrl,
139 sizeof(struct virtio_pmd_ctrl));
142 * Format is enforced in qemu code:
143 * One TX packet for header;
144 * At least one TX packet per argument;
145 * One RX packet for ACK.
147 vq->vq_ring.desc[head].flags = VRING_DESC_F_NEXT;
148 vq->vq_ring.desc[head].addr = vq->virtio_net_hdr_mz->phys_addr;
149 vq->vq_ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
151 i = vq->vq_ring.desc[head].next;
153 for (k = 0; k < pkt_num; k++) {
154 vq->vq_ring.desc[i].flags = VRING_DESC_F_NEXT;
155 vq->vq_ring.desc[i].addr = vq->virtio_net_hdr_mz->phys_addr
156 + sizeof(struct virtio_net_ctrl_hdr)
157 + sizeof(ctrl->status) + sizeof(uint8_t)*sum;
158 vq->vq_ring.desc[i].len = dlen[k];
161 i = vq->vq_ring.desc[i].next;
164 vq->vq_ring.desc[i].flags = VRING_DESC_F_WRITE;
165 vq->vq_ring.desc[i].addr = vq->virtio_net_hdr_mz->phys_addr
166 + sizeof(struct virtio_net_ctrl_hdr);
167 vq->vq_ring.desc[i].len = sizeof(ctrl->status);
170 vq->vq_desc_head_idx = vq->vq_ring.desc[i].next;
172 vq_update_avail_ring(vq, head);
173 vq_update_avail_idx(vq);
175 PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
177 virtqueue_notify(vq);
180 while (vq->vq_used_cons_idx == vq->vq_ring.used->idx) {
185 while (vq->vq_used_cons_idx != vq->vq_ring.used->idx) {
186 uint32_t idx, desc_idx, used_idx;
187 struct vring_used_elem *uep;
189 used_idx = (uint32_t)(vq->vq_used_cons_idx
190 & (vq->vq_nentries - 1));
191 uep = &vq->vq_ring.used->ring[used_idx];
192 idx = (uint32_t) uep->id;
195 while (vq->vq_ring.desc[desc_idx].flags & VRING_DESC_F_NEXT) {
196 desc_idx = vq->vq_ring.desc[desc_idx].next;
200 vq->vq_ring.desc[desc_idx].next = vq->vq_desc_head_idx;
201 vq->vq_desc_head_idx = idx;
203 vq->vq_used_cons_idx++;
207 PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
208 vq->vq_free_cnt, vq->vq_desc_head_idx);
210 memcpy(&result, vq->virtio_net_hdr_mz->addr,
211 sizeof(struct virtio_pmd_ctrl));
213 return result.status;
217 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
219 struct virtio_hw *hw = dev->data->dev_private;
220 struct virtio_pmd_ctrl ctrl;
224 ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
225 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
226 memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
228 dlen[0] = sizeof(uint16_t);
230 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
233 PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
234 "failed, this is too late now...");
241 int virtio_dev_queue_setup(struct rte_eth_dev *dev,
244 uint8_t vtpci_queue_idx,
246 unsigned int socket_id,
247 struct virtqueue **pvq)
249 char vq_name[VIRTQUEUE_MAX_NAME_SZ];
250 const struct rte_memzone *mz;
253 struct virtio_hw *hw = dev->data->dev_private;
254 struct virtqueue *vq = NULL;
256 /* Write the virtqueue index to the Queue Select Field */
257 VIRTIO_WRITE_REG_2(hw, VIRTIO_PCI_QUEUE_SEL, vtpci_queue_idx);
258 PMD_INIT_LOG(DEBUG, "selecting queue: %d", vtpci_queue_idx);
261 * Read the virtqueue size from the Queue Size field
262 * Always power of 2 and if 0 virtqueue does not exist
264 vq_size = VIRTIO_READ_REG_2(hw, VIRTIO_PCI_QUEUE_NUM);
265 PMD_INIT_LOG(DEBUG, "vq_size: %d nb_desc:%d", vq_size, nb_desc);
269 PMD_INIT_LOG(ERR, "%s: virtqueue does not exist", __func__);
271 } else if (!rte_is_power_of_2(vq_size)) {
272 PMD_INIT_LOG(ERR, "%s: virtqueue size is not powerof 2", __func__);
274 } else if (nb_desc != vq_size) {
275 PMD_INIT_LOG(ERR, "Warning: nb_desc(%d) is not equal to vq size (%d), fall to vq size",
280 if (queue_type == VTNET_RQ) {
281 snprintf(vq_name, sizeof(vq_name), "port%d_rvq%d",
282 dev->data->port_id, queue_idx);
283 vq = rte_zmalloc(vq_name, sizeof(struct virtqueue) +
284 vq_size * sizeof(struct vq_desc_extra), RTE_CACHE_LINE_SIZE);
285 } else if (queue_type == VTNET_TQ) {
286 snprintf(vq_name, sizeof(vq_name), "port%d_tvq%d",
287 dev->data->port_id, queue_idx);
288 vq = rte_zmalloc(vq_name, sizeof(struct virtqueue) +
289 vq_size * sizeof(struct vq_desc_extra), RTE_CACHE_LINE_SIZE);
290 } else if (queue_type == VTNET_CQ) {
291 snprintf(vq_name, sizeof(vq_name), "port%d_cvq",
293 vq = rte_zmalloc(vq_name, sizeof(struct virtqueue) +
294 vq_size * sizeof(struct vq_desc_extra),
295 RTE_CACHE_LINE_SIZE);
298 PMD_INIT_LOG(ERR, "%s: Can not allocate virtqueue", __func__);
303 vq->port_id = dev->data->port_id;
304 vq->queue_id = queue_idx;
305 vq->vq_queue_index = vtpci_queue_idx;
306 vq->vq_nentries = vq_size;
307 vq->vq_free_cnt = vq_size;
310 * Reserve a memzone for vring elements
312 size = vring_size(vq_size, VIRTIO_PCI_VRING_ALIGN);
313 vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
314 PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d", size, vq->vq_ring_size);
316 mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
317 socket_id, 0, VIRTIO_PCI_VRING_ALIGN);
324 * Virtio PCI device VIRTIO_PCI_QUEUE_PF register is 32bit,
325 * and only accepts 32 bit page frame number.
326 * Check if the allocated physical memory exceeds 16TB.
328 if ((mz->phys_addr + vq->vq_ring_size - 1) >> (VIRTIO_PCI_QUEUE_ADDR_SHIFT + 32)) {
329 PMD_INIT_LOG(ERR, "vring address shouldn't be above 16TB!");
334 memset(mz->addr, 0, sizeof(mz->len));
336 vq->vq_ring_mem = mz->phys_addr;
337 vq->vq_ring_virt_mem = mz->addr;
338 PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem: 0x%"PRIx64, (uint64_t)mz->phys_addr);
339 PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%"PRIx64, (uint64_t)mz->addr);
340 vq->virtio_net_hdr_mz = NULL;
341 vq->virtio_net_hdr_mem = 0;
343 if (queue_type == VTNET_TQ) {
345 * For each xmit packet, allocate a virtio_net_hdr
347 snprintf(vq_name, sizeof(vq_name), "port%d_tvq%d_hdrzone",
348 dev->data->port_id, queue_idx);
349 vq->virtio_net_hdr_mz = rte_memzone_reserve_aligned(vq_name,
350 vq_size * hw->vtnet_hdr_size,
351 socket_id, 0, RTE_CACHE_LINE_SIZE);
352 if (vq->virtio_net_hdr_mz == NULL) {
356 vq->virtio_net_hdr_mem =
357 vq->virtio_net_hdr_mz->phys_addr;
358 memset(vq->virtio_net_hdr_mz->addr, 0,
359 vq_size * hw->vtnet_hdr_size);
360 } else if (queue_type == VTNET_CQ) {
361 /* Allocate a page for control vq command, data and status */
362 snprintf(vq_name, sizeof(vq_name), "port%d_cvq_hdrzone",
364 vq->virtio_net_hdr_mz = rte_memzone_reserve_aligned(vq_name,
365 PAGE_SIZE, socket_id, 0, RTE_CACHE_LINE_SIZE);
366 if (vq->virtio_net_hdr_mz == NULL) {
370 vq->virtio_net_hdr_mem =
371 vq->virtio_net_hdr_mz->phys_addr;
372 memset(vq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
376 * Set guest physical address of the virtqueue
377 * in VIRTIO_PCI_QUEUE_PFN config register of device
379 VIRTIO_WRITE_REG_4(hw, VIRTIO_PCI_QUEUE_PFN,
380 mz->phys_addr >> VIRTIO_PCI_QUEUE_ADDR_SHIFT);
386 virtio_dev_cq_queue_setup(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx,
389 struct virtqueue *vq;
390 uint16_t nb_desc = 0;
392 struct virtio_hw *hw = dev->data->dev_private;
394 PMD_INIT_FUNC_TRACE();
395 ret = virtio_dev_queue_setup(dev, VTNET_CQ, VTNET_SQ_CQ_QUEUE_IDX,
396 vtpci_queue_idx, nb_desc, socket_id, &vq);
399 PMD_INIT_LOG(ERR, "control vq initialization failed");
408 virtio_dev_close(struct rte_eth_dev *dev)
410 struct virtio_hw *hw = dev->data->dev_private;
412 PMD_INIT_LOG(DEBUG, "virtio_dev_close");
415 vtpci_irq_config(hw, VIRTIO_MSI_NO_VECTOR);
418 virtio_dev_free_mbufs(dev);
422 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
424 struct virtio_hw *hw = dev->data->dev_private;
425 struct virtio_pmd_ctrl ctrl;
429 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
430 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
434 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
437 PMD_INIT_LOG(ERR, "Failed to enable promisc");
441 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
443 struct virtio_hw *hw = dev->data->dev_private;
444 struct virtio_pmd_ctrl ctrl;
448 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
449 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
453 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
456 PMD_INIT_LOG(ERR, "Failed to disable promisc");
460 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
462 struct virtio_hw *hw = dev->data->dev_private;
463 struct virtio_pmd_ctrl ctrl;
467 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
468 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
472 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
475 PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
479 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
481 struct virtio_hw *hw = dev->data->dev_private;
482 struct virtio_pmd_ctrl ctrl;
486 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
487 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
491 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
494 PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
498 * dev_ops for virtio, bare necessities for basic operation
500 static struct eth_dev_ops virtio_eth_dev_ops = {
501 .dev_configure = virtio_dev_configure,
502 .dev_start = virtio_dev_start,
503 .dev_stop = virtio_dev_stop,
504 .dev_close = virtio_dev_close,
505 .promiscuous_enable = virtio_dev_promiscuous_enable,
506 .promiscuous_disable = virtio_dev_promiscuous_disable,
507 .allmulticast_enable = virtio_dev_allmulticast_enable,
508 .allmulticast_disable = virtio_dev_allmulticast_disable,
510 .dev_infos_get = virtio_dev_info_get,
511 .stats_get = virtio_dev_stats_get,
512 .stats_reset = virtio_dev_stats_reset,
513 .link_update = virtio_dev_link_update,
514 .rx_queue_setup = virtio_dev_rx_queue_setup,
515 /* meaningfull only to multiple queue */
516 .rx_queue_release = virtio_dev_rx_queue_release,
517 .tx_queue_setup = virtio_dev_tx_queue_setup,
518 /* meaningfull only to multiple queue */
519 .tx_queue_release = virtio_dev_tx_queue_release,
520 /* collect stats per queue */
521 .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
522 .vlan_filter_set = virtio_vlan_filter_set,
523 .mac_addr_add = virtio_mac_addr_add,
524 .mac_addr_remove = virtio_mac_addr_remove,
525 .mac_addr_set = virtio_mac_addr_set,
529 virtio_dev_atomic_read_link_status(struct rte_eth_dev *dev,
530 struct rte_eth_link *link)
532 struct rte_eth_link *dst = link;
533 struct rte_eth_link *src = &(dev->data->dev_link);
535 if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
536 *(uint64_t *)src) == 0)
543 * Atomically writes the link status information into global
544 * structure rte_eth_dev.
547 * - Pointer to the structure rte_eth_dev to read from.
548 * - Pointer to the buffer to be saved with the link status.
551 * - On success, zero.
552 * - On failure, negative value.
555 virtio_dev_atomic_write_link_status(struct rte_eth_dev *dev,
556 struct rte_eth_link *link)
558 struct rte_eth_link *dst = &(dev->data->dev_link);
559 struct rte_eth_link *src = link;
561 if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
562 *(uint64_t *)src) == 0)
569 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
573 for (i = 0; i < dev->data->nb_tx_queues; i++) {
574 const struct virtqueue *txvq = dev->data->tx_queues[i];
578 stats->opackets += txvq->packets;
579 stats->obytes += txvq->bytes;
580 stats->oerrors += txvq->errors;
582 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
583 stats->q_opackets[i] = txvq->packets;
584 stats->q_obytes[i] = txvq->bytes;
588 for (i = 0; i < dev->data->nb_rx_queues; i++) {
589 const struct virtqueue *rxvq = dev->data->rx_queues[i];
593 stats->ipackets += rxvq->packets;
594 stats->ibytes += rxvq->bytes;
595 stats->ierrors += rxvq->errors;
597 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
598 stats->q_ipackets[i] = rxvq->packets;
599 stats->q_ibytes[i] = rxvq->bytes;
603 stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
607 virtio_dev_stats_reset(struct rte_eth_dev *dev)
611 for (i = 0; i < dev->data->nb_tx_queues; i++) {
612 struct virtqueue *txvq = dev->data->tx_queues[i];
621 for (i = 0; i < dev->data->nb_rx_queues; i++) {
622 struct virtqueue *rxvq = dev->data->rx_queues[i];
631 dev->data->rx_mbuf_alloc_failed = 0;
635 virtio_set_hwaddr(struct virtio_hw *hw)
637 vtpci_write_dev_config(hw,
638 offsetof(struct virtio_net_config, mac),
639 &hw->mac_addr, ETHER_ADDR_LEN);
643 virtio_get_hwaddr(struct virtio_hw *hw)
645 if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
646 vtpci_read_dev_config(hw,
647 offsetof(struct virtio_net_config, mac),
648 &hw->mac_addr, ETHER_ADDR_LEN);
650 eth_random_addr(&hw->mac_addr[0]);
651 virtio_set_hwaddr(hw);
656 virtio_mac_table_set(struct virtio_hw *hw,
657 const struct virtio_net_ctrl_mac *uc,
658 const struct virtio_net_ctrl_mac *mc)
660 struct virtio_pmd_ctrl ctrl;
663 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
664 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
666 len[0] = uc->entries * ETHER_ADDR_LEN + sizeof(uc->entries);
667 memcpy(ctrl.data, uc, len[0]);
669 len[1] = mc->entries * ETHER_ADDR_LEN + sizeof(mc->entries);
670 memcpy(ctrl.data + len[0], mc, len[1]);
672 err = virtio_send_command(hw->cvq, &ctrl, len, 2);
674 PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
680 virtio_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
681 uint32_t index, uint32_t vmdq __rte_unused)
683 struct virtio_hw *hw = dev->data->dev_private;
684 const struct ether_addr *addrs = dev->data->mac_addrs;
686 struct virtio_net_ctrl_mac *uc, *mc;
688 if (index >= VIRTIO_MAX_MAC_ADDRS) {
689 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
693 uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
695 mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
698 for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
699 const struct ether_addr *addr
700 = (i == index) ? mac_addr : addrs + i;
701 struct virtio_net_ctrl_mac *tbl
702 = is_multicast_ether_addr(addr) ? mc : uc;
704 memcpy(&tbl->macs[tbl->entries++], addr, ETHER_ADDR_LEN);
707 virtio_mac_table_set(hw, uc, mc);
711 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
713 struct virtio_hw *hw = dev->data->dev_private;
714 struct ether_addr *addrs = dev->data->mac_addrs;
715 struct virtio_net_ctrl_mac *uc, *mc;
718 if (index >= VIRTIO_MAX_MAC_ADDRS) {
719 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
723 uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
725 mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
728 for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
729 struct virtio_net_ctrl_mac *tbl;
731 if (i == index || is_zero_ether_addr(addrs + i))
734 tbl = is_multicast_ether_addr(addrs + i) ? mc : uc;
735 memcpy(&tbl->macs[tbl->entries++], addrs + i, ETHER_ADDR_LEN);
738 virtio_mac_table_set(hw, uc, mc);
742 virtio_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
744 struct virtio_hw *hw = dev->data->dev_private;
746 memcpy(hw->mac_addr, mac_addr, ETHER_ADDR_LEN);
748 /* Use atomic update if available */
749 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
750 struct virtio_pmd_ctrl ctrl;
751 int len = ETHER_ADDR_LEN;
753 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
754 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
756 memcpy(ctrl.data, mac_addr, ETHER_ADDR_LEN);
757 virtio_send_command(hw->cvq, &ctrl, &len, 1);
758 } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
759 virtio_set_hwaddr(hw);
763 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
765 struct virtio_hw *hw = dev->data->dev_private;
766 struct virtio_pmd_ctrl ctrl;
769 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
772 ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
773 ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
774 memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
775 len = sizeof(vlan_id);
777 return virtio_send_command(hw->cvq, &ctrl, &len, 1);
781 virtio_negotiate_features(struct virtio_hw *hw)
783 uint32_t host_features, mask;
785 /* checksum offload not implemented */
786 mask = VIRTIO_NET_F_CSUM | VIRTIO_NET_F_GUEST_CSUM;
788 /* TSO and LRO are only available when their corresponding
789 * checksum offload feature is also negotiated.
791 mask |= VIRTIO_NET_F_HOST_TSO4 | VIRTIO_NET_F_HOST_TSO6 | VIRTIO_NET_F_HOST_ECN;
792 mask |= VIRTIO_NET_F_GUEST_TSO4 | VIRTIO_NET_F_GUEST_TSO6 | VIRTIO_NET_F_GUEST_ECN;
793 mask |= VTNET_LRO_FEATURES;
795 /* not negotiating INDIRECT descriptor table support */
796 mask |= VIRTIO_RING_F_INDIRECT_DESC;
798 /* Prepare guest_features: feature that driver wants to support */
799 hw->guest_features = VTNET_FEATURES & ~mask;
800 PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %x",
803 /* Read device(host) feature bits */
804 host_features = VIRTIO_READ_REG_4(hw, VIRTIO_PCI_HOST_FEATURES);
805 PMD_INIT_LOG(DEBUG, "host_features before negotiate = %x",
809 * Negotiate features: Subset of device feature bits are written back
810 * guest feature bits.
812 hw->guest_features = vtpci_negotiate_features(hw, host_features);
813 PMD_INIT_LOG(DEBUG, "features after negotiate = %x",
817 #ifdef RTE_EXEC_ENV_LINUXAPP
819 parse_sysfs_value(const char *filename, unsigned long *val)
825 f = fopen(filename, "r");
827 PMD_INIT_LOG(ERR, "%s(): cannot open sysfs value %s",
832 if (fgets(buf, sizeof(buf), f) == NULL) {
833 PMD_INIT_LOG(ERR, "%s(): cannot read sysfs value %s",
838 *val = strtoul(buf, &end, 0);
839 if ((buf[0] == '\0') || (end == NULL) || (*end != '\n')) {
840 PMD_INIT_LOG(ERR, "%s(): cannot parse sysfs value %s",
849 static int get_uio_dev(struct rte_pci_addr *loc, char *buf, unsigned int buflen)
851 unsigned int uio_num;
854 char dirname[PATH_MAX];
856 /* depending on kernel version, uio can be located in uio/uioX
858 snprintf(dirname, sizeof(dirname),
859 SYSFS_PCI_DEVICES "/" PCI_PRI_FMT "/uio",
860 loc->domain, loc->bus, loc->devid, loc->function);
861 dir = opendir(dirname);
863 /* retry with the parent directory */
864 snprintf(dirname, sizeof(dirname),
865 SYSFS_PCI_DEVICES "/" PCI_PRI_FMT,
866 loc->domain, loc->bus, loc->devid, loc->function);
867 dir = opendir(dirname);
870 PMD_INIT_LOG(ERR, "Cannot opendir %s", dirname);
875 /* take the first file starting with "uio" */
876 while ((e = readdir(dir)) != NULL) {
877 /* format could be uio%d ...*/
878 int shortprefix_len = sizeof("uio") - 1;
879 /* ... or uio:uio%d */
880 int longprefix_len = sizeof("uio:uio") - 1;
883 if (strncmp(e->d_name, "uio", 3) != 0)
886 /* first try uio%d */
888 uio_num = strtoull(e->d_name + shortprefix_len, &endptr, 10);
889 if (errno == 0 && endptr != (e->d_name + shortprefix_len)) {
890 snprintf(buf, buflen, "%s/uio%u", dirname, uio_num);
894 /* then try uio:uio%d */
896 uio_num = strtoull(e->d_name + longprefix_len, &endptr, 10);
897 if (errno == 0 && endptr != (e->d_name + longprefix_len)) {
898 snprintf(buf, buflen, "%s/uio:uio%u", dirname,
905 /* No uio resource found */
907 PMD_INIT_LOG(ERR, "Could not find uio resource");
915 virtio_has_msix(const struct rte_pci_addr *loc)
918 char dirname[PATH_MAX];
920 snprintf(dirname, sizeof(dirname),
921 SYSFS_PCI_DEVICES "/" PCI_PRI_FMT "/msi_irqs",
922 loc->domain, loc->bus, loc->devid, loc->function);
924 d = opendir(dirname);
931 /* Extract I/O port numbers from sysfs */
932 static int virtio_resource_init(struct rte_pci_device *pci_dev)
934 char dirname[PATH_MAX];
935 char filename[PATH_MAX];
936 unsigned long start, size;
938 if (get_uio_dev(&pci_dev->addr, dirname, sizeof(dirname)) < 0)
941 /* get portio size */
942 snprintf(filename, sizeof(filename),
943 "%s/portio/port0/size", dirname);
944 if (parse_sysfs_value(filename, &size) < 0) {
945 PMD_INIT_LOG(ERR, "%s(): cannot parse size",
950 /* get portio start */
951 snprintf(filename, sizeof(filename),
952 "%s/portio/port0/start", dirname);
953 if (parse_sysfs_value(filename, &start) < 0) {
954 PMD_INIT_LOG(ERR, "%s(): cannot parse portio start",
958 pci_dev->mem_resource[0].addr = (void *)(uintptr_t)start;
959 pci_dev->mem_resource[0].len = (uint64_t)size;
961 "PCI Port IO found start=0x%lx with size=0x%lx",
967 virtio_has_msix(const struct rte_pci_addr *loc __rte_unused)
969 /* nic_uio does not enable interrupts, return 0 (false). */
973 static int virtio_resource_init(struct rte_pci_device *pci_dev __rte_unused)
975 /* no setup required */
981 * Process Virtio Config changed interrupt and call the callback
982 * if link state changed.
985 virtio_interrupt_handler(__rte_unused struct rte_intr_handle *handle,
988 struct rte_eth_dev *dev = param;
989 struct virtio_hw *hw = dev->data->dev_private;
992 /* Read interrupt status which clears interrupt */
994 PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
996 if (rte_intr_enable(&dev->pci_dev->intr_handle) < 0)
997 PMD_DRV_LOG(ERR, "interrupt enable failed");
999 if (isr & VIRTIO_PCI_ISR_CONFIG) {
1000 if (virtio_dev_link_update(dev, 0) == 0)
1001 _rte_eth_dev_callback_process(dev,
1002 RTE_ETH_EVENT_INTR_LSC);
1008 * This function is based on probe() function in virtio_pci.c
1009 * It returns 0 on success.
1012 eth_virtio_dev_init(__rte_unused struct eth_driver *eth_drv,
1013 struct rte_eth_dev *eth_dev)
1015 struct virtio_hw *hw = eth_dev->data->dev_private;
1016 struct virtio_net_config *config;
1017 struct virtio_net_config local_config;
1018 uint32_t offset_conf = sizeof(config->mac);
1019 struct rte_pci_device *pci_dev;
1021 RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr));
1023 eth_dev->dev_ops = &virtio_eth_dev_ops;
1024 eth_dev->tx_pkt_burst = &virtio_xmit_pkts;
1026 if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1029 /* Allocate memory for storing MAC addresses */
1030 eth_dev->data->mac_addrs = rte_zmalloc("virtio", ETHER_ADDR_LEN, 0);
1031 if (eth_dev->data->mac_addrs == NULL) {
1033 "Failed to allocate %d bytes needed to store MAC addresses",
1038 pci_dev = eth_dev->pci_dev;
1039 if (virtio_resource_init(pci_dev) < 0)
1042 hw->use_msix = virtio_has_msix(&pci_dev->addr);
1043 hw->io_base = (uint32_t)(uintptr_t)pci_dev->mem_resource[0].addr;
1045 /* Reset the device although not necessary at startup */
1048 /* Tell the host we've noticed this device. */
1049 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1051 /* Tell the host we've known how to drive the device. */
1052 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1053 virtio_negotiate_features(hw);
1055 /* Setting up rx_header size for the device */
1056 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1057 eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1058 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1060 eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1061 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1064 /* Copy the permanent MAC address to: virtio_hw */
1065 virtio_get_hwaddr(hw);
1066 ether_addr_copy((struct ether_addr *) hw->mac_addr,
1067 ð_dev->data->mac_addrs[0]);
1069 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1070 hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1071 hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1073 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1074 config = &local_config;
1076 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1077 offset_conf += sizeof(config->status);
1080 "VIRTIO_NET_F_STATUS is not supported");
1084 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1085 offset_conf += sizeof(config->max_virtqueue_pairs);
1088 "VIRTIO_NET_F_MQ is not supported");
1089 config->max_virtqueue_pairs = 1;
1092 vtpci_read_dev_config(hw, 0, (uint8_t *)config, offset_conf);
1095 (VIRTIO_MAX_RX_QUEUES < config->max_virtqueue_pairs) ?
1096 VIRTIO_MAX_RX_QUEUES : config->max_virtqueue_pairs;
1098 (VIRTIO_MAX_TX_QUEUES < config->max_virtqueue_pairs) ?
1099 VIRTIO_MAX_TX_QUEUES : config->max_virtqueue_pairs;
1101 virtio_dev_cq_queue_setup(eth_dev,
1102 config->max_virtqueue_pairs * 2,
1105 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1106 config->max_virtqueue_pairs);
1107 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1109 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1110 config->mac[0], config->mac[1],
1111 config->mac[2], config->mac[3],
1112 config->mac[4], config->mac[5]);
1114 hw->max_rx_queues = 1;
1115 hw->max_tx_queues = 1;
1118 eth_dev->data->nb_rx_queues = hw->max_rx_queues;
1119 eth_dev->data->nb_tx_queues = hw->max_tx_queues;
1121 PMD_INIT_LOG(DEBUG, "hw->max_rx_queues=%d hw->max_tx_queues=%d",
1122 hw->max_rx_queues, hw->max_tx_queues);
1123 PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1124 eth_dev->data->port_id, pci_dev->id.vendor_id,
1125 pci_dev->id.device_id);
1127 /* Setup interrupt callback */
1128 rte_intr_callback_register(&pci_dev->intr_handle,
1129 virtio_interrupt_handler, eth_dev);
1131 virtio_dev_cq_start(eth_dev);
1136 static struct eth_driver rte_virtio_pmd = {
1138 .name = "rte_virtio_pmd",
1139 .id_table = pci_id_virtio_map,
1140 .drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC,
1142 .eth_dev_init = eth_virtio_dev_init,
1143 .dev_private_size = sizeof(struct virtio_hw),
1147 * Driver initialization routine.
1148 * Invoked once at EAL init time.
1149 * Register itself as the [Poll Mode] Driver of PCI virtio devices.
1150 * Returns 0 on success.
1153 rte_virtio_pmd_init(const char *name __rte_unused,
1154 const char *param __rte_unused)
1156 if (rte_eal_iopl_init() != 0) {
1157 PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1161 rte_eth_driver_register(&rte_virtio_pmd);
1166 * Only 1 queue is supported, no queue release related operation
1169 virtio_dev_rx_queue_release(__rte_unused void *rxq)
1174 virtio_dev_tx_queue_release(__rte_unused void *txq)
1179 * Configure virtio device
1180 * It returns 0 on success.
1183 virtio_dev_configure(struct rte_eth_dev *dev)
1185 const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1186 struct virtio_hw *hw = dev->data->dev_private;
1188 PMD_INIT_LOG(DEBUG, "configure");
1190 if (rxmode->hw_ip_checksum) {
1191 PMD_DRV_LOG(ERR, "HW IP checksum not supported");
1195 hw->vlan_strip = rxmode->hw_vlan_strip;
1197 if (rxmode->hw_vlan_filter
1198 && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
1200 "vlan filtering not available on this host");
1204 if (vtpci_irq_config(hw, 0) == VIRTIO_MSI_NO_VECTOR) {
1205 PMD_DRV_LOG(ERR, "failed to set config vector");
1214 virtio_dev_start(struct rte_eth_dev *dev)
1216 uint16_t nb_queues, i;
1217 struct virtio_hw *hw = dev->data->dev_private;
1219 /* check if lsc interrupt feature is enabled */
1220 if (dev->data->dev_conf.intr_conf.lsc) {
1221 if (!vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1222 PMD_DRV_LOG(ERR, "link status not supported by host");
1226 if (rte_intr_enable(&dev->pci_dev->intr_handle) < 0) {
1227 PMD_DRV_LOG(ERR, "interrupt enable failed");
1232 /* Initialize Link state */
1233 virtio_dev_link_update(dev, 0);
1235 /* On restart after stop do not touch queues */
1239 /* Do final configuration before rx/tx engine starts */
1240 virtio_dev_rxtx_start(dev);
1241 vtpci_reinit_complete(hw);
1245 /*Notify the backend
1246 *Otherwise the tap backend might already stop its queue due to fullness.
1247 *vhost backend will have no chance to be waked up
1249 nb_queues = dev->data->nb_rx_queues;
1250 if (nb_queues > 1) {
1251 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
1255 PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
1257 for (i = 0; i < nb_queues; i++)
1258 virtqueue_notify(dev->data->rx_queues[i]);
1260 PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
1262 for (i = 0; i < dev->data->nb_rx_queues; i++)
1263 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
1265 for (i = 0; i < dev->data->nb_tx_queues; i++)
1266 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
1271 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
1273 struct rte_mbuf *buf;
1274 int i, mbuf_num = 0;
1276 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1278 "Before freeing rxq[%d] used and unused buf", i);
1279 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
1281 while ((buf = (struct rte_mbuf *)virtqueue_detatch_unused(
1282 dev->data->rx_queues[i])) != NULL) {
1283 rte_pktmbuf_free(buf);
1287 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1289 "After freeing rxq[%d] used and unused buf", i);
1290 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
1293 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1295 "Before freeing txq[%d] used and unused bufs",
1297 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
1300 while ((buf = (struct rte_mbuf *)virtqueue_detatch_unused(
1301 dev->data->tx_queues[i])) != NULL) {
1302 rte_pktmbuf_free(buf);
1307 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1309 "After freeing txq[%d] used and unused buf", i);
1310 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
1315 * Stop device: disable interrupt and mark link down
1318 virtio_dev_stop(struct rte_eth_dev *dev)
1320 struct rte_eth_link link;
1322 PMD_INIT_LOG(DEBUG, "stop");
1324 if (dev->data->dev_conf.intr_conf.lsc)
1325 rte_intr_disable(&dev->pci_dev->intr_handle);
1327 memset(&link, 0, sizeof(link));
1328 virtio_dev_atomic_write_link_status(dev, &link);
1332 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
1334 struct rte_eth_link link, old;
1336 struct virtio_hw *hw = dev->data->dev_private;
1337 memset(&link, 0, sizeof(link));
1338 virtio_dev_atomic_read_link_status(dev, &link);
1340 link.link_duplex = FULL_DUPLEX;
1341 link.link_speed = SPEED_10G;
1343 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1344 PMD_INIT_LOG(DEBUG, "Get link status from hw");
1345 vtpci_read_dev_config(hw,
1346 offsetof(struct virtio_net_config, status),
1347 &status, sizeof(status));
1348 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
1349 link.link_status = 0;
1350 PMD_INIT_LOG(DEBUG, "Port %d is down",
1351 dev->data->port_id);
1353 link.link_status = 1;
1354 PMD_INIT_LOG(DEBUG, "Port %d is up",
1355 dev->data->port_id);
1358 link.link_status = 1; /* Link up */
1360 virtio_dev_atomic_write_link_status(dev, &link);
1362 return (old.link_status == link.link_status) ? -1 : 0;
1366 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1368 struct virtio_hw *hw = dev->data->dev_private;
1370 dev_info->driver_name = dev->driver->pci_drv.name;
1371 dev_info->max_rx_queues = (uint16_t)hw->max_rx_queues;
1372 dev_info->max_tx_queues = (uint16_t)hw->max_tx_queues;
1373 dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
1374 dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
1375 dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
1379 * It enables testpmd to collect per queue stats.
1382 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
1383 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
1384 __rte_unused uint8_t is_rx)
1389 static struct rte_driver rte_virtio_driver = {
1391 .init = rte_virtio_pmd_init,
1394 PMD_REGISTER_DRIVER(rte_virtio_driver);