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_alignment = VIRTIO_PCI_VRING_ALIGN;
298 vq->vq_nentries = vq_size;
299 vq->vq_free_cnt = vq_size;
302 * Reserve a memzone for vring elements
304 size = vring_size(vq_size, VIRTIO_PCI_VRING_ALIGN);
305 vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
306 PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d", size, vq->vq_ring_size);
308 mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
309 socket_id, 0, VIRTIO_PCI_VRING_ALIGN);
316 * Virtio PCI device VIRTIO_PCI_QUEUE_PF register is 32bit,
317 * and only accepts 32 bit page frame number.
318 * Check if the allocated physical memory exceeds 16TB.
320 if ((mz->phys_addr + vq->vq_ring_size - 1) >> (VIRTIO_PCI_QUEUE_ADDR_SHIFT + 32)) {
321 PMD_INIT_LOG(ERR, "vring address shouldn't be above 16TB!");
326 memset(mz->addr, 0, sizeof(mz->len));
328 vq->vq_ring_mem = mz->phys_addr;
329 vq->vq_ring_virt_mem = mz->addr;
330 PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem: 0x%"PRIx64, (uint64_t)mz->phys_addr);
331 PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%"PRIx64, (uint64_t)mz->addr);
332 vq->virtio_net_hdr_mz = NULL;
333 vq->virtio_net_hdr_mem = 0;
335 if (queue_type == VTNET_TQ) {
337 * For each xmit packet, allocate a virtio_net_hdr
339 snprintf(vq_name, sizeof(vq_name), "port%d_tvq%d_hdrzone",
340 dev->data->port_id, queue_idx);
341 vq->virtio_net_hdr_mz = rte_memzone_reserve_aligned(vq_name,
342 vq_size * hw->vtnet_hdr_size,
343 socket_id, 0, RTE_CACHE_LINE_SIZE);
344 if (vq->virtio_net_hdr_mz == NULL) {
348 vq->virtio_net_hdr_mem =
349 vq->virtio_net_hdr_mz->phys_addr;
350 memset(vq->virtio_net_hdr_mz->addr, 0,
351 vq_size * hw->vtnet_hdr_size);
352 } else if (queue_type == VTNET_CQ) {
353 /* Allocate a page for control vq command, data and status */
354 snprintf(vq_name, sizeof(vq_name), "port%d_cvq_hdrzone",
356 vq->virtio_net_hdr_mz = rte_memzone_reserve_aligned(vq_name,
357 PAGE_SIZE, socket_id, 0, RTE_CACHE_LINE_SIZE);
358 if (vq->virtio_net_hdr_mz == NULL) {
362 vq->virtio_net_hdr_mem =
363 vq->virtio_net_hdr_mz->phys_addr;
364 memset(vq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
368 * Set guest physical address of the virtqueue
369 * in VIRTIO_PCI_QUEUE_PFN config register of device
371 VIRTIO_WRITE_REG_4(hw, VIRTIO_PCI_QUEUE_PFN,
372 mz->phys_addr >> VIRTIO_PCI_QUEUE_ADDR_SHIFT);
378 virtio_dev_cq_queue_setup(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx,
381 struct virtqueue *vq;
382 uint16_t nb_desc = 0;
384 struct virtio_hw *hw = dev->data->dev_private;
386 PMD_INIT_FUNC_TRACE();
387 ret = virtio_dev_queue_setup(dev, VTNET_CQ, VTNET_SQ_CQ_QUEUE_IDX,
388 vtpci_queue_idx, nb_desc, socket_id, &vq);
391 PMD_INIT_LOG(ERR, "control vq initialization failed");
400 virtio_dev_close(struct rte_eth_dev *dev)
402 PMD_INIT_LOG(DEBUG, "virtio_dev_close");
404 virtio_dev_stop(dev);
408 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
410 struct virtio_hw *hw = dev->data->dev_private;
411 struct virtio_pmd_ctrl ctrl;
415 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
416 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
420 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
423 PMD_INIT_LOG(ERR, "Failed to enable promisc");
427 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
429 struct virtio_hw *hw = dev->data->dev_private;
430 struct virtio_pmd_ctrl ctrl;
434 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
435 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
439 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
442 PMD_INIT_LOG(ERR, "Failed to disable promisc");
446 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
448 struct virtio_hw *hw = dev->data->dev_private;
449 struct virtio_pmd_ctrl ctrl;
453 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
454 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
458 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
461 PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
465 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
467 struct virtio_hw *hw = dev->data->dev_private;
468 struct virtio_pmd_ctrl ctrl;
472 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
473 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
477 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
480 PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
484 * dev_ops for virtio, bare necessities for basic operation
486 static struct eth_dev_ops virtio_eth_dev_ops = {
487 .dev_configure = virtio_dev_configure,
488 .dev_start = virtio_dev_start,
489 .dev_stop = virtio_dev_stop,
490 .dev_close = virtio_dev_close,
491 .promiscuous_enable = virtio_dev_promiscuous_enable,
492 .promiscuous_disable = virtio_dev_promiscuous_disable,
493 .allmulticast_enable = virtio_dev_allmulticast_enable,
494 .allmulticast_disable = virtio_dev_allmulticast_disable,
496 .dev_infos_get = virtio_dev_info_get,
497 .stats_get = virtio_dev_stats_get,
498 .stats_reset = virtio_dev_stats_reset,
499 .link_update = virtio_dev_link_update,
500 .mac_addr_add = NULL,
501 .mac_addr_remove = NULL,
502 .rx_queue_setup = virtio_dev_rx_queue_setup,
503 /* meaningfull only to multiple queue */
504 .rx_queue_release = virtio_dev_rx_queue_release,
505 .tx_queue_setup = virtio_dev_tx_queue_setup,
506 /* meaningfull only to multiple queue */
507 .tx_queue_release = virtio_dev_tx_queue_release,
508 /* collect stats per queue */
509 .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
513 virtio_dev_atomic_read_link_status(struct rte_eth_dev *dev,
514 struct rte_eth_link *link)
516 struct rte_eth_link *dst = link;
517 struct rte_eth_link *src = &(dev->data->dev_link);
519 if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
520 *(uint64_t *)src) == 0)
527 * Atomically writes the link status information into global
528 * structure rte_eth_dev.
531 * - Pointer to the structure rte_eth_dev to read from.
532 * - Pointer to the buffer to be saved with the link status.
535 * - On success, zero.
536 * - On failure, negative value.
539 virtio_dev_atomic_write_link_status(struct rte_eth_dev *dev,
540 struct rte_eth_link *link)
542 struct rte_eth_link *dst = &(dev->data->dev_link);
543 struct rte_eth_link *src = link;
545 if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
546 *(uint64_t *)src) == 0)
553 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
557 for (i = 0; i < dev->data->nb_tx_queues; i++) {
558 const struct virtqueue *txvq = dev->data->tx_queues[i];
562 stats->opackets += txvq->packets;
563 stats->obytes += txvq->bytes;
564 stats->oerrors += txvq->errors;
566 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
567 stats->q_opackets[i] = txvq->packets;
568 stats->q_obytes[i] = txvq->bytes;
572 for (i = 0; i < dev->data->nb_rx_queues; i++) {
573 const struct virtqueue *rxvq = dev->data->rx_queues[i];
577 stats->ipackets += rxvq->packets;
578 stats->ibytes += rxvq->bytes;
579 stats->ierrors += rxvq->errors;
581 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
582 stats->q_ipackets[i] = rxvq->packets;
583 stats->q_ibytes[i] = rxvq->bytes;
587 stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
591 virtio_dev_stats_reset(struct rte_eth_dev *dev)
595 for (i = 0; i < dev->data->nb_tx_queues; i++) {
596 struct virtqueue *txvq = dev->data->tx_queues[i];
605 for (i = 0; i < dev->data->nb_rx_queues; i++) {
606 struct virtqueue *rxvq = dev->data->rx_queues[i];
615 dev->data->rx_mbuf_alloc_failed = 0;
619 virtio_set_hwaddr(struct virtio_hw *hw)
621 vtpci_write_dev_config(hw,
622 offsetof(struct virtio_net_config, mac),
623 &hw->mac_addr, ETHER_ADDR_LEN);
627 virtio_get_hwaddr(struct virtio_hw *hw)
629 if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
630 vtpci_read_dev_config(hw,
631 offsetof(struct virtio_net_config, mac),
632 &hw->mac_addr, ETHER_ADDR_LEN);
634 eth_random_addr(&hw->mac_addr[0]);
635 virtio_set_hwaddr(hw);
641 virtio_negotiate_features(struct virtio_hw *hw)
643 uint32_t host_features, mask;
645 mask = VIRTIO_NET_F_CTRL_VLAN;
646 mask |= VIRTIO_NET_F_CSUM | VIRTIO_NET_F_GUEST_CSUM;
648 /* TSO and LRO are only available when their corresponding
649 * checksum offload feature is also negotiated.
651 mask |= VIRTIO_NET_F_HOST_TSO4 | VIRTIO_NET_F_HOST_TSO6 | VIRTIO_NET_F_HOST_ECN;
652 mask |= VIRTIO_NET_F_GUEST_TSO4 | VIRTIO_NET_F_GUEST_TSO6 | VIRTIO_NET_F_GUEST_ECN;
653 mask |= VTNET_LRO_FEATURES;
655 /* not negotiating INDIRECT descriptor table support */
656 mask |= VIRTIO_RING_F_INDIRECT_DESC;
658 /* Prepare guest_features: feature that driver wants to support */
659 hw->guest_features = VTNET_FEATURES & ~mask;
660 PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %x",
663 /* Read device(host) feature bits */
664 host_features = VIRTIO_READ_REG_4(hw, VIRTIO_PCI_HOST_FEATURES);
665 PMD_INIT_LOG(DEBUG, "host_features before negotiate = %x",
669 * Negotiate features: Subset of device feature bits are written back
670 * guest feature bits.
672 hw->guest_features = vtpci_negotiate_features(hw, host_features);
673 PMD_INIT_LOG(DEBUG, "features after negotiate = %x",
677 #ifdef RTE_EXEC_ENV_LINUXAPP
679 parse_sysfs_value(const char *filename, unsigned long *val)
685 f = fopen(filename, "r");
687 PMD_INIT_LOG(ERR, "%s(): cannot open sysfs value %s",
692 if (fgets(buf, sizeof(buf), f) == NULL) {
693 PMD_INIT_LOG(ERR, "%s(): cannot read sysfs value %s",
698 *val = strtoul(buf, &end, 0);
699 if ((buf[0] == '\0') || (end == NULL) || (*end != '\n')) {
700 PMD_INIT_LOG(ERR, "%s(): cannot parse sysfs value %s",
709 static int get_uio_dev(struct rte_pci_addr *loc, char *buf, unsigned int buflen)
711 unsigned int uio_num;
714 char dirname[PATH_MAX];
716 /* depending on kernel version, uio can be located in uio/uioX
718 snprintf(dirname, sizeof(dirname),
719 SYSFS_PCI_DEVICES "/" PCI_PRI_FMT "/uio",
720 loc->domain, loc->bus, loc->devid, loc->function);
721 dir = opendir(dirname);
723 /* retry with the parent directory */
724 snprintf(dirname, sizeof(dirname),
725 SYSFS_PCI_DEVICES "/" PCI_PRI_FMT,
726 loc->domain, loc->bus, loc->devid, loc->function);
727 dir = opendir(dirname);
730 PMD_INIT_LOG(ERR, "Cannot opendir %s", dirname);
735 /* take the first file starting with "uio" */
736 while ((e = readdir(dir)) != NULL) {
737 /* format could be uio%d ...*/
738 int shortprefix_len = sizeof("uio") - 1;
739 /* ... or uio:uio%d */
740 int longprefix_len = sizeof("uio:uio") - 1;
743 if (strncmp(e->d_name, "uio", 3) != 0)
746 /* first try uio%d */
748 uio_num = strtoull(e->d_name + shortprefix_len, &endptr, 10);
749 if (errno == 0 && endptr != (e->d_name + shortprefix_len)) {
750 snprintf(buf, buflen, "%s/uio%u", dirname, uio_num);
754 /* then try uio:uio%d */
756 uio_num = strtoull(e->d_name + longprefix_len, &endptr, 10);
757 if (errno == 0 && endptr != (e->d_name + longprefix_len)) {
758 snprintf(buf, buflen, "%s/uio:uio%u", dirname,
765 /* No uio resource found */
767 PMD_INIT_LOG(ERR, "Could not find uio resource");
775 virtio_has_msix(const struct rte_pci_addr *loc)
778 char dirname[PATH_MAX];
780 snprintf(dirname, sizeof(dirname),
781 SYSFS_PCI_DEVICES "/" PCI_PRI_FMT "/msi_irqs",
782 loc->domain, loc->bus, loc->devid, loc->function);
784 d = opendir(dirname);
791 /* Extract I/O port numbers from sysfs */
792 static int virtio_resource_init(struct rte_pci_device *pci_dev)
794 char dirname[PATH_MAX];
795 char filename[PATH_MAX];
796 unsigned long start, size;
798 if (get_uio_dev(&pci_dev->addr, dirname, sizeof(dirname)) < 0)
801 /* get portio size */
802 snprintf(filename, sizeof(filename),
803 "%s/portio/port0/size", dirname);
804 if (parse_sysfs_value(filename, &size) < 0) {
805 PMD_INIT_LOG(ERR, "%s(): cannot parse size",
810 /* get portio start */
811 snprintf(filename, sizeof(filename),
812 "%s/portio/port0/start", dirname);
813 if (parse_sysfs_value(filename, &start) < 0) {
814 PMD_INIT_LOG(ERR, "%s(): cannot parse portio start",
818 pci_dev->mem_resource[0].addr = (void *)(uintptr_t)start;
819 pci_dev->mem_resource[0].len = (uint64_t)size;
821 "PCI Port IO found start=0x%lx with size=0x%lx",
827 virtio_has_msix(const struct rte_pci_addr *loc __rte_unused)
829 /* nic_uio does not enable interrupts, return 0 (false). */
833 static int virtio_resource_init(struct rte_pci_device *pci_dev __rte_unused)
835 /* no setup required */
841 * Process Virtio Config changed interrupt and call the callback
842 * if link state changed.
845 virtio_interrupt_handler(__rte_unused struct rte_intr_handle *handle,
848 struct rte_eth_dev *dev = param;
849 struct virtio_hw *hw = dev->data->dev_private;
852 /* Read interrupt status which clears interrupt */
854 PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
856 if (rte_intr_enable(&dev->pci_dev->intr_handle) < 0)
857 PMD_DRV_LOG(ERR, "interrupt enable failed");
859 if (isr & VIRTIO_PCI_ISR_CONFIG) {
860 if (virtio_dev_link_update(dev, 0) == 0)
861 _rte_eth_dev_callback_process(dev,
862 RTE_ETH_EVENT_INTR_LSC);
868 * This function is based on probe() function in virtio_pci.c
869 * It returns 0 on success.
872 eth_virtio_dev_init(__rte_unused struct eth_driver *eth_drv,
873 struct rte_eth_dev *eth_dev)
875 struct virtio_hw *hw = eth_dev->data->dev_private;
876 struct virtio_net_config *config;
877 struct virtio_net_config local_config;
878 uint32_t offset_conf = sizeof(config->mac);
879 struct rte_pci_device *pci_dev;
881 RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr));
883 eth_dev->dev_ops = &virtio_eth_dev_ops;
884 eth_dev->tx_pkt_burst = &virtio_xmit_pkts;
886 if (rte_eal_process_type() == RTE_PROC_SECONDARY)
889 pci_dev = eth_dev->pci_dev;
890 if (virtio_resource_init(pci_dev) < 0)
893 hw->use_msix = virtio_has_msix(&pci_dev->addr);
894 hw->io_base = (uint32_t)(uintptr_t)pci_dev->mem_resource[0].addr;
896 /* Reset the device although not necessary at startup */
899 /* Tell the host we've noticed this device. */
900 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
902 /* Tell the host we've known how to drive the device. */
903 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
904 virtio_negotiate_features(hw);
906 /* Setting up rx_header size for the device */
907 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
908 eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
909 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
911 eth_dev->rx_pkt_burst = &virtio_recv_pkts;
912 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
915 /* Allocate memory for storing MAC addresses */
916 eth_dev->data->mac_addrs = rte_zmalloc("virtio", ETHER_ADDR_LEN, 0);
917 if (eth_dev->data->mac_addrs == NULL) {
919 "Failed to allocate %d bytes needed to store MAC addresses",
924 /* Copy the permanent MAC address to: virtio_hw */
925 virtio_get_hwaddr(hw);
926 ether_addr_copy((struct ether_addr *) hw->mac_addr,
927 ð_dev->data->mac_addrs[0]);
929 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
930 hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
931 hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
933 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
934 config = &local_config;
936 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
937 offset_conf += sizeof(config->status);
940 "VIRTIO_NET_F_STATUS is not supported");
944 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
945 offset_conf += sizeof(config->max_virtqueue_pairs);
948 "VIRTIO_NET_F_MQ is not supported");
949 config->max_virtqueue_pairs = 1;
952 vtpci_read_dev_config(hw, 0, (uint8_t *)config, offset_conf);
955 (VIRTIO_MAX_RX_QUEUES < config->max_virtqueue_pairs) ?
956 VIRTIO_MAX_RX_QUEUES : config->max_virtqueue_pairs;
958 (VIRTIO_MAX_TX_QUEUES < config->max_virtqueue_pairs) ?
959 VIRTIO_MAX_TX_QUEUES : config->max_virtqueue_pairs;
961 virtio_dev_cq_queue_setup(eth_dev,
962 config->max_virtqueue_pairs * 2,
965 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
966 config->max_virtqueue_pairs);
967 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
969 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
970 config->mac[0], config->mac[1],
971 config->mac[2], config->mac[3],
972 config->mac[4], config->mac[5]);
974 hw->max_rx_queues = 1;
975 hw->max_tx_queues = 1;
978 eth_dev->data->nb_rx_queues = hw->max_rx_queues;
979 eth_dev->data->nb_tx_queues = hw->max_tx_queues;
981 PMD_INIT_LOG(DEBUG, "hw->max_rx_queues=%d hw->max_tx_queues=%d",
982 hw->max_rx_queues, hw->max_tx_queues);
983 PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
984 eth_dev->data->port_id, pci_dev->id.vendor_id,
985 pci_dev->id.device_id);
987 /* Setup interrupt callback */
988 rte_intr_callback_register(&pci_dev->intr_handle,
989 virtio_interrupt_handler, eth_dev);
993 static struct eth_driver rte_virtio_pmd = {
995 .name = "rte_virtio_pmd",
996 .id_table = pci_id_virtio_map,
997 .drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC,
999 .eth_dev_init = eth_virtio_dev_init,
1000 .dev_private_size = sizeof(struct virtio_hw),
1004 * Driver initialization routine.
1005 * Invoked once at EAL init time.
1006 * Register itself as the [Poll Mode] Driver of PCI virtio devices.
1007 * Returns 0 on success.
1010 rte_virtio_pmd_init(const char *name __rte_unused,
1011 const char *param __rte_unused)
1013 if (rte_eal_iopl_init() != 0) {
1014 PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1018 rte_eth_driver_register(&rte_virtio_pmd);
1023 * Only 1 queue is supported, no queue release related operation
1026 virtio_dev_rx_queue_release(__rte_unused void *rxq)
1031 virtio_dev_tx_queue_release(__rte_unused void *txq)
1036 * Configure virtio device
1037 * It returns 0 on success.
1040 virtio_dev_configure(struct rte_eth_dev *dev)
1042 const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1043 struct virtio_hw *hw = dev->data->dev_private;
1046 PMD_INIT_LOG(DEBUG, "configure");
1048 if (rxmode->hw_ip_checksum) {
1049 PMD_DRV_LOG(ERR, "HW IP checksum not supported");
1053 ret = vtpci_irq_config(hw, 0);
1055 PMD_DRV_LOG(ERR, "failed to set config vector");
1062 virtio_dev_start(struct rte_eth_dev *dev)
1064 uint16_t nb_queues, i;
1065 struct virtio_hw *hw = dev->data->dev_private;
1067 /* Tell the host we've noticed this device. */
1068 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1070 /* Tell the host we've known how to drive the device. */
1071 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1073 virtio_dev_cq_start(dev);
1075 /* Do final configuration before rx/tx engine starts */
1076 virtio_dev_rxtx_start(dev);
1078 /* check if lsc interrupt feature is enabled */
1079 if (dev->data->dev_conf.intr_conf.lsc) {
1080 if (!vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1081 PMD_DRV_LOG(ERR, "link status not supported by host");
1085 if (rte_intr_enable(&dev->pci_dev->intr_handle) < 0) {
1086 PMD_DRV_LOG(ERR, "interrupt enable failed");
1091 /* Initialize Link state */
1092 virtio_dev_link_update(dev, 0);
1094 vtpci_reinit_complete(hw);
1096 /*Notify the backend
1097 *Otherwise the tap backend might already stop its queue due to fullness.
1098 *vhost backend will have no chance to be waked up
1100 nb_queues = dev->data->nb_rx_queues;
1101 if (nb_queues > 1) {
1102 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
1106 PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
1108 for (i = 0; i < nb_queues; i++)
1109 virtqueue_notify(dev->data->rx_queues[i]);
1111 PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
1113 for (i = 0; i < dev->data->nb_rx_queues; i++)
1114 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
1116 for (i = 0; i < dev->data->nb_tx_queues; i++)
1117 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
1122 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
1124 struct rte_mbuf *buf;
1125 int i, mbuf_num = 0;
1127 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1129 "Before freeing rxq[%d] used and unused buf", i);
1130 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
1132 while ((buf = (struct rte_mbuf *)virtqueue_detatch_unused(
1133 dev->data->rx_queues[i])) != NULL) {
1134 rte_pktmbuf_free(buf);
1138 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1140 "After freeing rxq[%d] used and unused buf", i);
1141 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
1144 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1146 "Before freeing txq[%d] used and unused bufs",
1148 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
1151 while ((buf = (struct rte_mbuf *)virtqueue_detatch_unused(
1152 dev->data->tx_queues[i])) != NULL) {
1153 rte_pktmbuf_free(buf);
1158 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1160 "After freeing txq[%d] used and unused buf", i);
1161 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
1166 * Stop device: disable rx and tx functions to allow for reconfiguring.
1169 virtio_dev_stop(struct rte_eth_dev *dev)
1171 struct virtio_hw *hw = dev->data->dev_private;
1174 vtpci_irq_config(hw, 0);
1176 virtio_dev_free_mbufs(dev);
1180 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
1182 struct rte_eth_link link, old;
1184 struct virtio_hw *hw = dev->data->dev_private;
1185 memset(&link, 0, sizeof(link));
1186 virtio_dev_atomic_read_link_status(dev, &link);
1188 link.link_duplex = FULL_DUPLEX;
1189 link.link_speed = SPEED_10G;
1191 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1192 PMD_INIT_LOG(DEBUG, "Get link status from hw");
1193 vtpci_read_dev_config(hw,
1194 offsetof(struct virtio_net_config, status),
1195 &status, sizeof(status));
1196 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
1197 link.link_status = 0;
1198 PMD_INIT_LOG(DEBUG, "Port %d is down",
1199 dev->data->port_id);
1201 link.link_status = 1;
1202 PMD_INIT_LOG(DEBUG, "Port %d is up",
1203 dev->data->port_id);
1206 link.link_status = 1; /* Link up */
1208 virtio_dev_atomic_write_link_status(dev, &link);
1210 return (old.link_status == link.link_status) ? -1 : 0;
1214 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1216 struct virtio_hw *hw = dev->data->dev_private;
1218 dev_info->driver_name = dev->driver->pci_drv.name;
1219 dev_info->max_rx_queues = (uint16_t)hw->max_rx_queues;
1220 dev_info->max_tx_queues = (uint16_t)hw->max_tx_queues;
1221 dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
1222 dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
1223 dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
1227 * It enables testpmd to collect per queue stats.
1230 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
1231 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
1232 __rte_unused uint8_t is_rx)
1237 static struct rte_driver rte_virtio_driver = {
1239 .init = rte_virtio_pmd_init,
1242 PMD_REGISTER_DRIVER(rte_virtio_driver);