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);
88 static int virtio_dev_queue_stats_mapping_set(
89 __rte_unused struct rte_eth_dev *eth_dev,
90 __rte_unused uint16_t queue_id,
91 __rte_unused uint8_t stat_idx,
92 __rte_unused uint8_t is_rx);
95 * The set of PCI devices this driver supports
97 static struct rte_pci_id pci_id_virtio_map[] = {
99 #define RTE_PCI_DEV_ID_DECL_VIRTIO(vend, dev) {RTE_PCI_DEVICE(vend, dev)},
100 #include "rte_pci_dev_ids.h"
102 { .vendor_id = 0, /* sentinel */ },
106 virtio_send_command(struct virtqueue *vq, struct virtio_pmd_ctrl *ctrl,
107 int *dlen, int pkt_num)
109 uint32_t head = vq->vq_desc_head_idx, i;
111 virtio_net_ctrl_ack status = ~0;
112 struct virtio_pmd_ctrl result;
114 ctrl->status = status;
118 "%s(): Control queue is not supported.",
123 PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
124 "vq->hw->cvq = %p vq = %p",
125 vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
127 if ((vq->vq_free_cnt < ((uint32_t)pkt_num + 2)) || (pkt_num < 1))
130 memcpy(vq->virtio_net_hdr_mz->addr, ctrl,
131 sizeof(struct virtio_pmd_ctrl));
134 * Format is enforced in qemu code:
135 * One TX packet for header;
136 * At least one TX packet per argument;
137 * One RX packet for ACK.
139 vq->vq_ring.desc[head].flags = VRING_DESC_F_NEXT;
140 vq->vq_ring.desc[head].addr = vq->virtio_net_hdr_mz->phys_addr;
141 vq->vq_ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
143 i = vq->vq_ring.desc[head].next;
145 for (k = 0; k < pkt_num; k++) {
146 vq->vq_ring.desc[i].flags = VRING_DESC_F_NEXT;
147 vq->vq_ring.desc[i].addr = vq->virtio_net_hdr_mz->phys_addr
148 + sizeof(struct virtio_net_ctrl_hdr)
149 + sizeof(ctrl->status) + sizeof(uint8_t)*sum;
150 vq->vq_ring.desc[i].len = dlen[k];
153 i = vq->vq_ring.desc[i].next;
156 vq->vq_ring.desc[i].flags = VRING_DESC_F_WRITE;
157 vq->vq_ring.desc[i].addr = vq->virtio_net_hdr_mz->phys_addr
158 + sizeof(struct virtio_net_ctrl_hdr);
159 vq->vq_ring.desc[i].len = sizeof(ctrl->status);
162 vq->vq_desc_head_idx = vq->vq_ring.desc[i].next;
164 vq_update_avail_ring(vq, head);
165 vq_update_avail_idx(vq);
167 PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
169 virtqueue_notify(vq);
171 while (vq->vq_used_cons_idx == vq->vq_ring.used->idx)
174 while (vq->vq_used_cons_idx != vq->vq_ring.used->idx) {
175 uint32_t idx, desc_idx, used_idx;
176 struct vring_used_elem *uep;
180 used_idx = (uint32_t)(vq->vq_used_cons_idx
181 & (vq->vq_nentries - 1));
182 uep = &vq->vq_ring.used->ring[used_idx];
183 idx = (uint32_t) uep->id;
186 while (vq->vq_ring.desc[desc_idx].flags & VRING_DESC_F_NEXT) {
187 desc_idx = vq->vq_ring.desc[desc_idx].next;
191 vq->vq_ring.desc[desc_idx].next = vq->vq_desc_head_idx;
192 vq->vq_desc_head_idx = idx;
194 vq->vq_used_cons_idx++;
198 PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
199 vq->vq_free_cnt, vq->vq_desc_head_idx);
201 memcpy(&result, vq->virtio_net_hdr_mz->addr,
202 sizeof(struct virtio_pmd_ctrl));
204 return result.status;
208 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
210 struct virtio_hw *hw = dev->data->dev_private;
211 struct virtio_pmd_ctrl ctrl;
215 ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
216 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
217 memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
219 dlen[0] = sizeof(uint16_t);
221 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
224 PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
225 "failed, this is too late now...");
232 int virtio_dev_queue_setup(struct rte_eth_dev *dev,
235 uint8_t vtpci_queue_idx,
237 unsigned int socket_id,
238 struct virtqueue **pvq)
240 char vq_name[VIRTQUEUE_MAX_NAME_SZ];
241 const struct rte_memzone *mz;
244 struct virtio_hw *hw = dev->data->dev_private;
245 struct virtqueue *vq = NULL;
247 /* Write the virtqueue index to the Queue Select Field */
248 VIRTIO_WRITE_REG_2(hw, VIRTIO_PCI_QUEUE_SEL, vtpci_queue_idx);
249 PMD_INIT_LOG(DEBUG, "selecting queue: %d", vtpci_queue_idx);
252 * Read the virtqueue size from the Queue Size field
253 * Always power of 2 and if 0 virtqueue does not exist
255 vq_size = VIRTIO_READ_REG_2(hw, VIRTIO_PCI_QUEUE_NUM);
256 PMD_INIT_LOG(DEBUG, "vq_size: %d nb_desc:%d", vq_size, nb_desc);
260 PMD_INIT_LOG(ERR, "%s: virtqueue does not exist", __func__);
262 } else if (!rte_is_power_of_2(vq_size)) {
263 PMD_INIT_LOG(ERR, "%s: virtqueue size is not powerof 2", __func__);
265 } else if (nb_desc != vq_size) {
266 PMD_INIT_LOG(ERR, "Warning: nb_desc(%d) is not equal to vq size (%d), fall to vq size",
271 if (queue_type == VTNET_RQ) {
272 snprintf(vq_name, sizeof(vq_name), "port%d_rvq%d",
273 dev->data->port_id, queue_idx);
274 vq = rte_zmalloc(vq_name, sizeof(struct virtqueue) +
275 vq_size * sizeof(struct vq_desc_extra), RTE_CACHE_LINE_SIZE);
276 } else if (queue_type == VTNET_TQ) {
277 snprintf(vq_name, sizeof(vq_name), "port%d_tvq%d",
278 dev->data->port_id, queue_idx);
279 vq = rte_zmalloc(vq_name, sizeof(struct virtqueue) +
280 vq_size * sizeof(struct vq_desc_extra), RTE_CACHE_LINE_SIZE);
281 } else if (queue_type == VTNET_CQ) {
282 snprintf(vq_name, sizeof(vq_name), "port%d_cvq",
284 vq = rte_zmalloc(vq_name, sizeof(struct virtqueue) +
285 vq_size * sizeof(struct vq_desc_extra),
286 RTE_CACHE_LINE_SIZE);
289 PMD_INIT_LOG(ERR, "%s: Can not allocate virtqueue", __func__);
294 vq->port_id = dev->data->port_id;
295 vq->queue_id = queue_idx;
296 vq->vq_queue_index = vtpci_queue_idx;
297 vq->vq_nentries = vq_size;
298 vq->vq_free_cnt = vq_size;
301 * Reserve a memzone for vring elements
303 size = vring_size(vq_size, VIRTIO_PCI_VRING_ALIGN);
304 vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
305 PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d", size, vq->vq_ring_size);
307 mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
308 socket_id, 0, VIRTIO_PCI_VRING_ALIGN);
315 * Virtio PCI device VIRTIO_PCI_QUEUE_PF register is 32bit,
316 * and only accepts 32 bit page frame number.
317 * Check if the allocated physical memory exceeds 16TB.
319 if ((mz->phys_addr + vq->vq_ring_size - 1) >> (VIRTIO_PCI_QUEUE_ADDR_SHIFT + 32)) {
320 PMD_INIT_LOG(ERR, "vring address shouldn't be above 16TB!");
325 memset(mz->addr, 0, sizeof(mz->len));
327 vq->vq_ring_mem = mz->phys_addr;
328 vq->vq_ring_virt_mem = mz->addr;
329 PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem: 0x%"PRIx64, (uint64_t)mz->phys_addr);
330 PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%"PRIx64, (uint64_t)mz->addr);
331 vq->virtio_net_hdr_mz = NULL;
332 vq->virtio_net_hdr_mem = 0;
334 if (queue_type == VTNET_TQ) {
336 * For each xmit packet, allocate a virtio_net_hdr
338 snprintf(vq_name, sizeof(vq_name), "port%d_tvq%d_hdrzone",
339 dev->data->port_id, queue_idx);
340 vq->virtio_net_hdr_mz = rte_memzone_reserve_aligned(vq_name,
341 vq_size * hw->vtnet_hdr_size,
342 socket_id, 0, RTE_CACHE_LINE_SIZE);
343 if (vq->virtio_net_hdr_mz == NULL) {
347 vq->virtio_net_hdr_mem =
348 vq->virtio_net_hdr_mz->phys_addr;
349 memset(vq->virtio_net_hdr_mz->addr, 0,
350 vq_size * hw->vtnet_hdr_size);
351 } else if (queue_type == VTNET_CQ) {
352 /* Allocate a page for control vq command, data and status */
353 snprintf(vq_name, sizeof(vq_name), "port%d_cvq_hdrzone",
355 vq->virtio_net_hdr_mz = rte_memzone_reserve_aligned(vq_name,
356 PAGE_SIZE, socket_id, 0, RTE_CACHE_LINE_SIZE);
357 if (vq->virtio_net_hdr_mz == NULL) {
361 vq->virtio_net_hdr_mem =
362 vq->virtio_net_hdr_mz->phys_addr;
363 memset(vq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
367 * Set guest physical address of the virtqueue
368 * in VIRTIO_PCI_QUEUE_PFN config register of device
370 VIRTIO_WRITE_REG_4(hw, VIRTIO_PCI_QUEUE_PFN,
371 mz->phys_addr >> VIRTIO_PCI_QUEUE_ADDR_SHIFT);
377 virtio_dev_cq_queue_setup(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx,
380 struct virtqueue *vq;
381 uint16_t nb_desc = 0;
383 struct virtio_hw *hw = dev->data->dev_private;
385 PMD_INIT_FUNC_TRACE();
386 ret = virtio_dev_queue_setup(dev, VTNET_CQ, VTNET_SQ_CQ_QUEUE_IDX,
387 vtpci_queue_idx, nb_desc, socket_id, &vq);
390 PMD_INIT_LOG(ERR, "control vq initialization failed");
399 virtio_dev_close(struct rte_eth_dev *dev)
401 PMD_INIT_LOG(DEBUG, "virtio_dev_close");
403 virtio_dev_stop(dev);
407 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
409 struct virtio_hw *hw = dev->data->dev_private;
410 struct virtio_pmd_ctrl ctrl;
414 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
415 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
419 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
422 PMD_INIT_LOG(ERR, "Failed to enable promisc");
426 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
428 struct virtio_hw *hw = dev->data->dev_private;
429 struct virtio_pmd_ctrl ctrl;
433 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
434 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
438 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
441 PMD_INIT_LOG(ERR, "Failed to disable promisc");
445 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
447 struct virtio_hw *hw = dev->data->dev_private;
448 struct virtio_pmd_ctrl ctrl;
452 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
453 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
457 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
460 PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
464 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
466 struct virtio_hw *hw = dev->data->dev_private;
467 struct virtio_pmd_ctrl ctrl;
471 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
472 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
476 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
479 PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
483 * dev_ops for virtio, bare necessities for basic operation
485 static struct eth_dev_ops virtio_eth_dev_ops = {
486 .dev_configure = virtio_dev_configure,
487 .dev_start = virtio_dev_start,
488 .dev_stop = virtio_dev_stop,
489 .dev_close = virtio_dev_close,
490 .promiscuous_enable = virtio_dev_promiscuous_enable,
491 .promiscuous_disable = virtio_dev_promiscuous_disable,
492 .allmulticast_enable = virtio_dev_allmulticast_enable,
493 .allmulticast_disable = virtio_dev_allmulticast_disable,
495 .dev_infos_get = virtio_dev_info_get,
496 .stats_get = virtio_dev_stats_get,
497 .stats_reset = virtio_dev_stats_reset,
498 .link_update = virtio_dev_link_update,
499 .mac_addr_add = NULL,
500 .mac_addr_remove = NULL,
501 .rx_queue_setup = virtio_dev_rx_queue_setup,
502 /* meaningfull only to multiple queue */
503 .rx_queue_release = virtio_dev_rx_queue_release,
504 .tx_queue_setup = virtio_dev_tx_queue_setup,
505 /* meaningfull only to multiple queue */
506 .tx_queue_release = virtio_dev_tx_queue_release,
507 /* collect stats per queue */
508 .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
512 virtio_dev_atomic_read_link_status(struct rte_eth_dev *dev,
513 struct rte_eth_link *link)
515 struct rte_eth_link *dst = link;
516 struct rte_eth_link *src = &(dev->data->dev_link);
518 if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
519 *(uint64_t *)src) == 0)
526 * Atomically writes the link status information into global
527 * structure rte_eth_dev.
530 * - Pointer to the structure rte_eth_dev to read from.
531 * - Pointer to the buffer to be saved with the link status.
534 * - On success, zero.
535 * - On failure, negative value.
538 virtio_dev_atomic_write_link_status(struct rte_eth_dev *dev,
539 struct rte_eth_link *link)
541 struct rte_eth_link *dst = &(dev->data->dev_link);
542 struct rte_eth_link *src = link;
544 if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
545 *(uint64_t *)src) == 0)
552 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
556 for (i = 0; i < dev->data->nb_tx_queues; i++) {
557 const struct virtqueue *txvq = dev->data->tx_queues[i];
561 stats->opackets += txvq->packets;
562 stats->obytes += txvq->bytes;
563 stats->oerrors += txvq->errors;
565 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
566 stats->q_opackets[i] = txvq->packets;
567 stats->q_obytes[i] = txvq->bytes;
571 for (i = 0; i < dev->data->nb_rx_queues; i++) {
572 const struct virtqueue *rxvq = dev->data->rx_queues[i];
576 stats->ipackets += rxvq->packets;
577 stats->ibytes += rxvq->bytes;
578 stats->ierrors += rxvq->errors;
580 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
581 stats->q_ipackets[i] = rxvq->packets;
582 stats->q_ibytes[i] = rxvq->bytes;
586 stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
590 virtio_dev_stats_reset(struct rte_eth_dev *dev)
594 for (i = 0; i < dev->data->nb_tx_queues; i++) {
595 struct virtqueue *txvq = dev->data->tx_queues[i];
604 for (i = 0; i < dev->data->nb_rx_queues; i++) {
605 struct virtqueue *rxvq = dev->data->rx_queues[i];
614 dev->data->rx_mbuf_alloc_failed = 0;
618 virtio_set_hwaddr(struct virtio_hw *hw)
620 vtpci_write_dev_config(hw,
621 offsetof(struct virtio_net_config, mac),
622 &hw->mac_addr, ETHER_ADDR_LEN);
626 virtio_get_hwaddr(struct virtio_hw *hw)
628 if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
629 vtpci_read_dev_config(hw,
630 offsetof(struct virtio_net_config, mac),
631 &hw->mac_addr, ETHER_ADDR_LEN);
633 eth_random_addr(&hw->mac_addr[0]);
634 virtio_set_hwaddr(hw);
640 virtio_negotiate_features(struct virtio_hw *hw)
642 uint32_t host_features, mask;
644 mask = VIRTIO_NET_F_CTRL_VLAN;
645 mask |= VIRTIO_NET_F_CSUM | VIRTIO_NET_F_GUEST_CSUM;
647 /* TSO and LRO are only available when their corresponding
648 * checksum offload feature is also negotiated.
650 mask |= VIRTIO_NET_F_HOST_TSO4 | VIRTIO_NET_F_HOST_TSO6 | VIRTIO_NET_F_HOST_ECN;
651 mask |= VIRTIO_NET_F_GUEST_TSO4 | VIRTIO_NET_F_GUEST_TSO6 | VIRTIO_NET_F_GUEST_ECN;
652 mask |= VTNET_LRO_FEATURES;
654 /* not negotiating INDIRECT descriptor table support */
655 mask |= VIRTIO_RING_F_INDIRECT_DESC;
657 /* Prepare guest_features: feature that driver wants to support */
658 hw->guest_features = VTNET_FEATURES & ~mask;
659 PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %x",
662 /* Read device(host) feature bits */
663 host_features = VIRTIO_READ_REG_4(hw, VIRTIO_PCI_HOST_FEATURES);
664 PMD_INIT_LOG(DEBUG, "host_features before negotiate = %x",
668 * Negotiate features: Subset of device feature bits are written back
669 * guest feature bits.
671 hw->guest_features = vtpci_negotiate_features(hw, host_features);
672 PMD_INIT_LOG(DEBUG, "features after negotiate = %x",
676 #ifdef RTE_EXEC_ENV_LINUXAPP
678 parse_sysfs_value(const char *filename, unsigned long *val)
684 f = fopen(filename, "r");
686 PMD_INIT_LOG(ERR, "%s(): cannot open sysfs value %s",
691 if (fgets(buf, sizeof(buf), f) == NULL) {
692 PMD_INIT_LOG(ERR, "%s(): cannot read sysfs value %s",
697 *val = strtoul(buf, &end, 0);
698 if ((buf[0] == '\0') || (end == NULL) || (*end != '\n')) {
699 PMD_INIT_LOG(ERR, "%s(): cannot parse sysfs value %s",
708 static int get_uio_dev(struct rte_pci_addr *loc, char *buf, unsigned int buflen)
710 unsigned int uio_num;
713 char dirname[PATH_MAX];
715 /* depending on kernel version, uio can be located in uio/uioX
717 snprintf(dirname, sizeof(dirname),
718 SYSFS_PCI_DEVICES "/" PCI_PRI_FMT "/uio",
719 loc->domain, loc->bus, loc->devid, loc->function);
720 dir = opendir(dirname);
722 /* retry with the parent directory */
723 snprintf(dirname, sizeof(dirname),
724 SYSFS_PCI_DEVICES "/" PCI_PRI_FMT,
725 loc->domain, loc->bus, loc->devid, loc->function);
726 dir = opendir(dirname);
729 PMD_INIT_LOG(ERR, "Cannot opendir %s", dirname);
734 /* take the first file starting with "uio" */
735 while ((e = readdir(dir)) != NULL) {
736 /* format could be uio%d ...*/
737 int shortprefix_len = sizeof("uio") - 1;
738 /* ... or uio:uio%d */
739 int longprefix_len = sizeof("uio:uio") - 1;
742 if (strncmp(e->d_name, "uio", 3) != 0)
745 /* first try uio%d */
747 uio_num = strtoull(e->d_name + shortprefix_len, &endptr, 10);
748 if (errno == 0 && endptr != (e->d_name + shortprefix_len)) {
749 snprintf(buf, buflen, "%s/uio%u", dirname, uio_num);
753 /* then try uio:uio%d */
755 uio_num = strtoull(e->d_name + longprefix_len, &endptr, 10);
756 if (errno == 0 && endptr != (e->d_name + longprefix_len)) {
757 snprintf(buf, buflen, "%s/uio:uio%u", dirname,
764 /* No uio resource found */
766 PMD_INIT_LOG(ERR, "Could not find uio resource");
774 virtio_has_msix(const struct rte_pci_addr *loc)
777 char dirname[PATH_MAX];
779 snprintf(dirname, sizeof(dirname),
780 SYSFS_PCI_DEVICES "/" PCI_PRI_FMT "/msi_irqs",
781 loc->domain, loc->bus, loc->devid, loc->function);
783 d = opendir(dirname);
790 /* Extract I/O port numbers from sysfs */
791 static int virtio_resource_init(struct rte_pci_device *pci_dev)
793 char dirname[PATH_MAX];
794 char filename[PATH_MAX];
795 unsigned long start, size;
797 if (get_uio_dev(&pci_dev->addr, dirname, sizeof(dirname)) < 0)
800 /* get portio size */
801 snprintf(filename, sizeof(filename),
802 "%s/portio/port0/size", dirname);
803 if (parse_sysfs_value(filename, &size) < 0) {
804 PMD_INIT_LOG(ERR, "%s(): cannot parse size",
809 /* get portio start */
810 snprintf(filename, sizeof(filename),
811 "%s/portio/port0/start", dirname);
812 if (parse_sysfs_value(filename, &start) < 0) {
813 PMD_INIT_LOG(ERR, "%s(): cannot parse portio start",
817 pci_dev->mem_resource[0].addr = (void *)(uintptr_t)start;
818 pci_dev->mem_resource[0].len = (uint64_t)size;
820 "PCI Port IO found start=0x%lx with size=0x%lx",
826 virtio_has_msix(const struct rte_pci_addr *loc __rte_unused)
828 /* nic_uio does not enable interrupts, return 0 (false). */
832 static int virtio_resource_init(struct rte_pci_device *pci_dev __rte_unused)
834 /* no setup required */
840 * Process Virtio Config changed interrupt and call the callback
841 * if link state changed.
844 virtio_interrupt_handler(__rte_unused struct rte_intr_handle *handle,
847 struct rte_eth_dev *dev = param;
848 struct virtio_hw *hw = dev->data->dev_private;
851 /* Read interrupt status which clears interrupt */
853 PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
855 if (rte_intr_enable(&dev->pci_dev->intr_handle) < 0)
856 PMD_DRV_LOG(ERR, "interrupt enable failed");
858 if (isr & VIRTIO_PCI_ISR_CONFIG) {
859 if (virtio_dev_link_update(dev, 0) == 0)
860 _rte_eth_dev_callback_process(dev,
861 RTE_ETH_EVENT_INTR_LSC);
867 * This function is based on probe() function in virtio_pci.c
868 * It returns 0 on success.
871 eth_virtio_dev_init(__rte_unused struct eth_driver *eth_drv,
872 struct rte_eth_dev *eth_dev)
874 struct virtio_hw *hw = eth_dev->data->dev_private;
875 struct virtio_net_config *config;
876 struct virtio_net_config local_config;
877 uint32_t offset_conf = sizeof(config->mac);
878 struct rte_pci_device *pci_dev;
880 RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr));
882 eth_dev->dev_ops = &virtio_eth_dev_ops;
883 eth_dev->tx_pkt_burst = &virtio_xmit_pkts;
885 if (rte_eal_process_type() == RTE_PROC_SECONDARY)
888 pci_dev = eth_dev->pci_dev;
889 if (virtio_resource_init(pci_dev) < 0)
892 hw->use_msix = virtio_has_msix(&pci_dev->addr);
893 hw->io_base = (uint32_t)(uintptr_t)pci_dev->mem_resource[0].addr;
895 /* Reset the device although not necessary at startup */
898 /* Tell the host we've noticed this device. */
899 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
901 /* Tell the host we've known how to drive the device. */
902 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
903 virtio_negotiate_features(hw);
905 /* Setting up rx_header size for the device */
906 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
907 eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
908 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
910 eth_dev->rx_pkt_burst = &virtio_recv_pkts;
911 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
914 /* Allocate memory for storing MAC addresses */
915 eth_dev->data->mac_addrs = rte_zmalloc("virtio", ETHER_ADDR_LEN, 0);
916 if (eth_dev->data->mac_addrs == NULL) {
918 "Failed to allocate %d bytes needed to store MAC addresses",
923 /* Copy the permanent MAC address to: virtio_hw */
924 virtio_get_hwaddr(hw);
925 ether_addr_copy((struct ether_addr *) hw->mac_addr,
926 ð_dev->data->mac_addrs[0]);
928 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
929 hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
930 hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
932 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
933 config = &local_config;
935 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
936 offset_conf += sizeof(config->status);
939 "VIRTIO_NET_F_STATUS is not supported");
943 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
944 offset_conf += sizeof(config->max_virtqueue_pairs);
947 "VIRTIO_NET_F_MQ is not supported");
948 config->max_virtqueue_pairs = 1;
951 vtpci_read_dev_config(hw, 0, (uint8_t *)config, offset_conf);
954 (VIRTIO_MAX_RX_QUEUES < config->max_virtqueue_pairs) ?
955 VIRTIO_MAX_RX_QUEUES : config->max_virtqueue_pairs;
957 (VIRTIO_MAX_TX_QUEUES < config->max_virtqueue_pairs) ?
958 VIRTIO_MAX_TX_QUEUES : config->max_virtqueue_pairs;
960 virtio_dev_cq_queue_setup(eth_dev,
961 config->max_virtqueue_pairs * 2,
964 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
965 config->max_virtqueue_pairs);
966 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
968 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
969 config->mac[0], config->mac[1],
970 config->mac[2], config->mac[3],
971 config->mac[4], config->mac[5]);
973 hw->max_rx_queues = 1;
974 hw->max_tx_queues = 1;
977 eth_dev->data->nb_rx_queues = hw->max_rx_queues;
978 eth_dev->data->nb_tx_queues = hw->max_tx_queues;
980 PMD_INIT_LOG(DEBUG, "hw->max_rx_queues=%d hw->max_tx_queues=%d",
981 hw->max_rx_queues, hw->max_tx_queues);
982 PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
983 eth_dev->data->port_id, pci_dev->id.vendor_id,
984 pci_dev->id.device_id);
986 /* Setup interrupt callback */
987 rte_intr_callback_register(&pci_dev->intr_handle,
988 virtio_interrupt_handler, eth_dev);
992 static struct eth_driver rte_virtio_pmd = {
994 .name = "rte_virtio_pmd",
995 .id_table = pci_id_virtio_map,
996 .drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC,
998 .eth_dev_init = eth_virtio_dev_init,
999 .dev_private_size = sizeof(struct virtio_hw),
1003 * Driver initialization routine.
1004 * Invoked once at EAL init time.
1005 * Register itself as the [Poll Mode] Driver of PCI virtio devices.
1006 * Returns 0 on success.
1009 rte_virtio_pmd_init(const char *name __rte_unused,
1010 const char *param __rte_unused)
1012 if (rte_eal_iopl_init() != 0) {
1013 PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1017 rte_eth_driver_register(&rte_virtio_pmd);
1022 * Only 1 queue is supported, no queue release related operation
1025 virtio_dev_rx_queue_release(__rte_unused void *rxq)
1030 virtio_dev_tx_queue_release(__rte_unused void *txq)
1035 * Configure virtio device
1036 * It returns 0 on success.
1039 virtio_dev_configure(struct rte_eth_dev *dev)
1041 const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1042 struct virtio_hw *hw = dev->data->dev_private;
1045 PMD_INIT_LOG(DEBUG, "configure");
1047 if (rxmode->hw_ip_checksum) {
1048 PMD_DRV_LOG(ERR, "HW IP checksum not supported");
1052 ret = vtpci_irq_config(hw, 0);
1054 PMD_DRV_LOG(ERR, "failed to set config vector");
1061 virtio_dev_start(struct rte_eth_dev *dev)
1063 uint16_t nb_queues, i;
1064 struct virtio_hw *hw = dev->data->dev_private;
1066 /* Tell the host we've noticed this device. */
1067 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1069 /* Tell the host we've known how to drive the device. */
1070 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1072 virtio_dev_cq_start(dev);
1074 /* Do final configuration before rx/tx engine starts */
1075 virtio_dev_rxtx_start(dev);
1077 /* check if lsc interrupt feature is enabled */
1078 if (dev->data->dev_conf.intr_conf.lsc) {
1079 if (!vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1080 PMD_DRV_LOG(ERR, "link status not supported by host");
1084 if (rte_intr_enable(&dev->pci_dev->intr_handle) < 0) {
1085 PMD_DRV_LOG(ERR, "interrupt enable failed");
1090 /* Initialize Link state */
1091 virtio_dev_link_update(dev, 0);
1093 vtpci_reinit_complete(hw);
1095 /*Notify the backend
1096 *Otherwise the tap backend might already stop its queue due to fullness.
1097 *vhost backend will have no chance to be waked up
1099 nb_queues = dev->data->nb_rx_queues;
1100 if (nb_queues > 1) {
1101 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
1105 PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
1107 for (i = 0; i < nb_queues; i++)
1108 virtqueue_notify(dev->data->rx_queues[i]);
1110 PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
1112 for (i = 0; i < dev->data->nb_rx_queues; i++)
1113 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
1115 for (i = 0; i < dev->data->nb_tx_queues; i++)
1116 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
1121 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
1123 struct rte_mbuf *buf;
1124 int i, mbuf_num = 0;
1126 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1128 "Before freeing rxq[%d] used and unused buf", i);
1129 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
1131 while ((buf = (struct rte_mbuf *)virtqueue_detatch_unused(
1132 dev->data->rx_queues[i])) != NULL) {
1133 rte_pktmbuf_free(buf);
1137 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1139 "After freeing rxq[%d] used and unused buf", i);
1140 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
1143 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1145 "Before freeing txq[%d] used and unused bufs",
1147 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
1150 while ((buf = (struct rte_mbuf *)virtqueue_detatch_unused(
1151 dev->data->tx_queues[i])) != NULL) {
1152 rte_pktmbuf_free(buf);
1157 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1159 "After freeing txq[%d] used and unused buf", i);
1160 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
1165 * Stop device: disable rx and tx functions to allow for reconfiguring.
1168 virtio_dev_stop(struct rte_eth_dev *dev)
1170 struct virtio_hw *hw = dev->data->dev_private;
1173 vtpci_irq_config(hw, 0);
1175 virtio_dev_free_mbufs(dev);
1179 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
1181 struct rte_eth_link link, old;
1183 struct virtio_hw *hw = dev->data->dev_private;
1184 memset(&link, 0, sizeof(link));
1185 virtio_dev_atomic_read_link_status(dev, &link);
1187 link.link_duplex = FULL_DUPLEX;
1188 link.link_speed = SPEED_10G;
1190 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1191 PMD_INIT_LOG(DEBUG, "Get link status from hw");
1192 vtpci_read_dev_config(hw,
1193 offsetof(struct virtio_net_config, status),
1194 &status, sizeof(status));
1195 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
1196 link.link_status = 0;
1197 PMD_INIT_LOG(DEBUG, "Port %d is down",
1198 dev->data->port_id);
1200 link.link_status = 1;
1201 PMD_INIT_LOG(DEBUG, "Port %d is up",
1202 dev->data->port_id);
1205 link.link_status = 1; /* Link up */
1207 virtio_dev_atomic_write_link_status(dev, &link);
1209 return (old.link_status == link.link_status) ? -1 : 0;
1213 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1215 struct virtio_hw *hw = dev->data->dev_private;
1217 dev_info->driver_name = dev->driver->pci_drv.name;
1218 dev_info->max_rx_queues = (uint16_t)hw->max_rx_queues;
1219 dev_info->max_tx_queues = (uint16_t)hw->max_tx_queues;
1220 dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
1221 dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
1222 dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
1226 * It enables testpmd to collect per queue stats.
1229 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
1230 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
1231 __rte_unused uint8_t is_rx)
1236 static struct rte_driver rte_virtio_driver = {
1238 .init = rte_virtio_pmd_init,
1241 PMD_REGISTER_DRIVER(rte_virtio_driver);