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)
211 = VIRTIO_DEV_PRIVATE_TO_HW(dev->data->dev_private);
212 struct virtio_pmd_ctrl ctrl;
216 ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
217 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
218 memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
220 dlen[0] = sizeof(uint16_t);
222 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
225 PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
226 "failed, this is too late now...");
233 int virtio_dev_queue_setup(struct rte_eth_dev *dev,
236 uint8_t vtpci_queue_idx,
238 unsigned int socket_id,
239 struct virtqueue **pvq)
241 char vq_name[VIRTQUEUE_MAX_NAME_SZ];
242 const struct rte_memzone *mz;
245 struct virtio_hw *hw =
246 VIRTIO_DEV_PRIVATE_TO_HW(dev->data->dev_private);
247 struct virtqueue *vq = NULL;
249 /* Write the virtqueue index to the Queue Select Field */
250 VIRTIO_WRITE_REG_2(hw, VIRTIO_PCI_QUEUE_SEL, vtpci_queue_idx);
251 PMD_INIT_LOG(DEBUG, "selecting queue: %d", vtpci_queue_idx);
254 * Read the virtqueue size from the Queue Size field
255 * Always power of 2 and if 0 virtqueue does not exist
257 vq_size = VIRTIO_READ_REG_2(hw, VIRTIO_PCI_QUEUE_NUM);
258 PMD_INIT_LOG(DEBUG, "vq_size: %d nb_desc:%d", vq_size, nb_desc);
262 PMD_INIT_LOG(ERR, "%s: virtqueue does not exist", __func__);
264 } else if (!rte_is_power_of_2(vq_size)) {
265 PMD_INIT_LOG(ERR, "%s: virtqueue size is not powerof 2", __func__);
267 } else if (nb_desc != vq_size) {
268 PMD_INIT_LOG(ERR, "Warning: nb_desc(%d) is not equal to vq size (%d), fall to vq size",
273 if (queue_type == VTNET_RQ) {
274 snprintf(vq_name, sizeof(vq_name), "port%d_rvq%d",
275 dev->data->port_id, queue_idx);
276 vq = rte_zmalloc(vq_name, sizeof(struct virtqueue) +
277 vq_size * sizeof(struct vq_desc_extra), RTE_CACHE_LINE_SIZE);
278 } else if (queue_type == VTNET_TQ) {
279 snprintf(vq_name, sizeof(vq_name), "port%d_tvq%d",
280 dev->data->port_id, queue_idx);
281 vq = rte_zmalloc(vq_name, sizeof(struct virtqueue) +
282 vq_size * sizeof(struct vq_desc_extra), RTE_CACHE_LINE_SIZE);
283 } else if (queue_type == VTNET_CQ) {
284 snprintf(vq_name, sizeof(vq_name), "port%d_cvq",
286 vq = rte_zmalloc(vq_name, sizeof(struct virtqueue) +
287 vq_size * sizeof(struct vq_desc_extra),
288 RTE_CACHE_LINE_SIZE);
291 PMD_INIT_LOG(ERR, "%s: Can not allocate virtqueue", __func__);
296 vq->port_id = dev->data->port_id;
297 vq->queue_id = queue_idx;
298 vq->vq_queue_index = vtpci_queue_idx;
299 vq->vq_alignment = VIRTIO_PCI_VRING_ALIGN;
300 vq->vq_nentries = vq_size;
301 vq->vq_free_cnt = vq_size;
304 * Reserve a memzone for vring elements
306 size = vring_size(vq_size, VIRTIO_PCI_VRING_ALIGN);
307 vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
308 PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d", size, vq->vq_ring_size);
310 mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
311 socket_id, 0, VIRTIO_PCI_VRING_ALIGN);
318 * Virtio PCI device VIRTIO_PCI_QUEUE_PF register is 32bit,
319 * and only accepts 32 bit page frame number.
320 * Check if the allocated physical memory exceeds 16TB.
322 if ((mz->phys_addr + vq->vq_ring_size - 1) >> (VIRTIO_PCI_QUEUE_ADDR_SHIFT + 32)) {
323 PMD_INIT_LOG(ERR, "vring address shouldn't be above 16TB!");
328 memset(mz->addr, 0, sizeof(mz->len));
330 vq->vq_ring_mem = mz->phys_addr;
331 vq->vq_ring_virt_mem = mz->addr;
332 PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem: 0x%"PRIx64, (uint64_t)mz->phys_addr);
333 PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%"PRIx64, (uint64_t)mz->addr);
334 vq->virtio_net_hdr_mz = NULL;
335 vq->virtio_net_hdr_mem = 0;
337 if (queue_type == VTNET_TQ) {
339 * For each xmit packet, allocate a virtio_net_hdr
341 snprintf(vq_name, sizeof(vq_name), "port%d_tvq%d_hdrzone",
342 dev->data->port_id, queue_idx);
343 vq->virtio_net_hdr_mz = rte_memzone_reserve_aligned(vq_name,
344 vq_size * hw->vtnet_hdr_size,
345 socket_id, 0, RTE_CACHE_LINE_SIZE);
346 if (vq->virtio_net_hdr_mz == NULL) {
350 vq->virtio_net_hdr_mem =
351 vq->virtio_net_hdr_mz->phys_addr;
352 memset(vq->virtio_net_hdr_mz->addr, 0,
353 vq_size * hw->vtnet_hdr_size);
354 } else if (queue_type == VTNET_CQ) {
355 /* Allocate a page for control vq command, data and status */
356 snprintf(vq_name, sizeof(vq_name), "port%d_cvq_hdrzone",
358 vq->virtio_net_hdr_mz = rte_memzone_reserve_aligned(vq_name,
359 PAGE_SIZE, socket_id, 0, RTE_CACHE_LINE_SIZE);
360 if (vq->virtio_net_hdr_mz == NULL) {
364 vq->virtio_net_hdr_mem =
365 vq->virtio_net_hdr_mz->phys_addr;
366 memset(vq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
370 * Set guest physical address of the virtqueue
371 * in VIRTIO_PCI_QUEUE_PFN config register of device
373 VIRTIO_WRITE_REG_4(hw, VIRTIO_PCI_QUEUE_PFN,
374 mz->phys_addr >> VIRTIO_PCI_QUEUE_ADDR_SHIFT);
380 virtio_dev_cq_queue_setup(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx,
383 struct virtqueue *vq;
384 uint16_t nb_desc = 0;
386 struct virtio_hw *hw =
387 VIRTIO_DEV_PRIVATE_TO_HW(dev->data->dev_private);
389 PMD_INIT_FUNC_TRACE();
390 ret = virtio_dev_queue_setup(dev, VTNET_CQ, VTNET_SQ_CQ_QUEUE_IDX,
391 vtpci_queue_idx, nb_desc, socket_id, &vq);
394 PMD_INIT_LOG(ERR, "control vq initialization failed");
403 virtio_dev_close(struct rte_eth_dev *dev)
405 PMD_INIT_LOG(DEBUG, "virtio_dev_close");
407 virtio_dev_stop(dev);
411 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
414 = VIRTIO_DEV_PRIVATE_TO_HW(dev->data->dev_private);
415 struct virtio_pmd_ctrl ctrl;
419 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
420 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
424 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
427 PMD_INIT_LOG(ERR, "Failed to enable promisc");
431 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
434 = VIRTIO_DEV_PRIVATE_TO_HW(dev->data->dev_private);
435 struct virtio_pmd_ctrl ctrl;
439 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
440 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
444 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
447 PMD_INIT_LOG(ERR, "Failed to disable promisc");
451 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
454 = VIRTIO_DEV_PRIVATE_TO_HW(dev->data->dev_private);
455 struct virtio_pmd_ctrl ctrl;
459 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
460 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
464 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
467 PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
471 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
474 = VIRTIO_DEV_PRIVATE_TO_HW(dev->data->dev_private);
475 struct virtio_pmd_ctrl ctrl;
479 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
480 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
484 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
487 PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
491 * dev_ops for virtio, bare necessities for basic operation
493 static struct eth_dev_ops virtio_eth_dev_ops = {
494 .dev_configure = virtio_dev_configure,
495 .dev_start = virtio_dev_start,
496 .dev_stop = virtio_dev_stop,
497 .dev_close = virtio_dev_close,
498 .promiscuous_enable = virtio_dev_promiscuous_enable,
499 .promiscuous_disable = virtio_dev_promiscuous_disable,
500 .allmulticast_enable = virtio_dev_allmulticast_enable,
501 .allmulticast_disable = virtio_dev_allmulticast_disable,
503 .dev_infos_get = virtio_dev_info_get,
504 .stats_get = virtio_dev_stats_get,
505 .stats_reset = virtio_dev_stats_reset,
506 .link_update = virtio_dev_link_update,
507 .mac_addr_add = NULL,
508 .mac_addr_remove = NULL,
509 .rx_queue_setup = virtio_dev_rx_queue_setup,
510 /* meaningfull only to multiple queue */
511 .rx_queue_release = virtio_dev_rx_queue_release,
512 .tx_queue_setup = virtio_dev_tx_queue_setup,
513 /* meaningfull only to multiple queue */
514 .tx_queue_release = virtio_dev_tx_queue_release,
515 /* collect stats per queue */
516 .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
520 virtio_dev_atomic_read_link_status(struct rte_eth_dev *dev,
521 struct rte_eth_link *link)
523 struct rte_eth_link *dst = link;
524 struct rte_eth_link *src = &(dev->data->dev_link);
526 if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
527 *(uint64_t *)src) == 0)
534 * Atomically writes the link status information into global
535 * structure rte_eth_dev.
538 * - Pointer to the structure rte_eth_dev to read from.
539 * - Pointer to the buffer to be saved with the link status.
542 * - On success, zero.
543 * - On failure, negative value.
546 virtio_dev_atomic_write_link_status(struct rte_eth_dev *dev,
547 struct rte_eth_link *link)
549 struct rte_eth_link *dst = &(dev->data->dev_link);
550 struct rte_eth_link *src = link;
552 if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
553 *(uint64_t *)src) == 0)
560 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
564 for (i = 0; i < dev->data->nb_tx_queues; i++) {
565 const struct virtqueue *txvq = dev->data->tx_queues[i];
569 stats->opackets += txvq->packets;
570 stats->obytes += txvq->bytes;
571 stats->oerrors += txvq->errors;
573 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
574 stats->q_opackets[i] = txvq->packets;
575 stats->q_obytes[i] = txvq->bytes;
579 for (i = 0; i < dev->data->nb_rx_queues; i++) {
580 const struct virtqueue *rxvq = dev->data->rx_queues[i];
584 stats->ipackets += rxvq->packets;
585 stats->ibytes += rxvq->bytes;
586 stats->ierrors += rxvq->errors;
588 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
589 stats->q_ipackets[i] = rxvq->packets;
590 stats->q_ibytes[i] = rxvq->bytes;
594 stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
598 virtio_dev_stats_reset(struct rte_eth_dev *dev)
602 for (i = 0; i < dev->data->nb_tx_queues; i++) {
603 struct virtqueue *txvq = dev->data->tx_queues[i];
612 for (i = 0; i < dev->data->nb_rx_queues; i++) {
613 struct virtqueue *rxvq = dev->data->rx_queues[i];
622 dev->data->rx_mbuf_alloc_failed = 0;
626 virtio_set_hwaddr(struct virtio_hw *hw)
628 vtpci_write_dev_config(hw,
629 offsetof(struct virtio_net_config, mac),
630 &hw->mac_addr, ETHER_ADDR_LEN);
634 virtio_get_hwaddr(struct virtio_hw *hw)
636 if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
637 vtpci_read_dev_config(hw,
638 offsetof(struct virtio_net_config, mac),
639 &hw->mac_addr, ETHER_ADDR_LEN);
641 eth_random_addr(&hw->mac_addr[0]);
642 virtio_set_hwaddr(hw);
648 virtio_negotiate_features(struct virtio_hw *hw)
650 uint32_t host_features, mask;
652 mask = VIRTIO_NET_F_CTRL_VLAN;
653 mask |= VIRTIO_NET_F_CSUM | VIRTIO_NET_F_GUEST_CSUM;
655 /* TSO and LRO are only available when their corresponding
656 * checksum offload feature is also negotiated.
658 mask |= VIRTIO_NET_F_HOST_TSO4 | VIRTIO_NET_F_HOST_TSO6 | VIRTIO_NET_F_HOST_ECN;
659 mask |= VIRTIO_NET_F_GUEST_TSO4 | VIRTIO_NET_F_GUEST_TSO6 | VIRTIO_NET_F_GUEST_ECN;
660 mask |= VTNET_LRO_FEATURES;
662 /* not negotiating INDIRECT descriptor table support */
663 mask |= VIRTIO_RING_F_INDIRECT_DESC;
665 /* Prepare guest_features: feature that driver wants to support */
666 hw->guest_features = VTNET_FEATURES & ~mask;
667 PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %x",
670 /* Read device(host) feature bits */
671 host_features = VIRTIO_READ_REG_4(hw, VIRTIO_PCI_HOST_FEATURES);
672 PMD_INIT_LOG(DEBUG, "host_features before negotiate = %x",
676 * Negotiate features: Subset of device feature bits are written back
677 * guest feature bits.
679 hw->guest_features = vtpci_negotiate_features(hw, host_features);
680 PMD_INIT_LOG(DEBUG, "features after negotiate = %x",
684 #ifdef RTE_EXEC_ENV_LINUXAPP
686 parse_sysfs_value(const char *filename, unsigned long *val)
692 f = fopen(filename, "r");
694 PMD_INIT_LOG(ERR, "%s(): cannot open sysfs value %s",
699 if (fgets(buf, sizeof(buf), f) == NULL) {
700 PMD_INIT_LOG(ERR, "%s(): cannot read sysfs value %s",
705 *val = strtoul(buf, &end, 0);
706 if ((buf[0] == '\0') || (end == NULL) || (*end != '\n')) {
707 PMD_INIT_LOG(ERR, "%s(): cannot parse sysfs value %s",
716 static int get_uio_dev(struct rte_pci_addr *loc, char *buf, unsigned int buflen)
718 unsigned int uio_num;
721 char dirname[PATH_MAX];
723 /* depending on kernel version, uio can be located in uio/uioX
725 snprintf(dirname, sizeof(dirname),
726 SYSFS_PCI_DEVICES "/" PCI_PRI_FMT "/uio",
727 loc->domain, loc->bus, loc->devid, loc->function);
728 dir = opendir(dirname);
730 /* retry with the parent directory */
731 snprintf(dirname, sizeof(dirname),
732 SYSFS_PCI_DEVICES "/" PCI_PRI_FMT,
733 loc->domain, loc->bus, loc->devid, loc->function);
734 dir = opendir(dirname);
737 PMD_INIT_LOG(ERR, "Cannot opendir %s", dirname);
742 /* take the first file starting with "uio" */
743 while ((e = readdir(dir)) != NULL) {
744 /* format could be uio%d ...*/
745 int shortprefix_len = sizeof("uio") - 1;
746 /* ... or uio:uio%d */
747 int longprefix_len = sizeof("uio:uio") - 1;
750 if (strncmp(e->d_name, "uio", 3) != 0)
753 /* first try uio%d */
755 uio_num = strtoull(e->d_name + shortprefix_len, &endptr, 10);
756 if (errno == 0 && endptr != (e->d_name + shortprefix_len)) {
757 snprintf(buf, buflen, "%s/uio%u", dirname, uio_num);
761 /* then try uio:uio%d */
763 uio_num = strtoull(e->d_name + longprefix_len, &endptr, 10);
764 if (errno == 0 && endptr != (e->d_name + longprefix_len)) {
765 snprintf(buf, buflen, "%s/uio:uio%u", dirname,
772 /* No uio resource found */
774 PMD_INIT_LOG(ERR, "Could not find uio resource");
782 virtio_has_msix(const struct rte_pci_addr *loc)
785 char dirname[PATH_MAX];
787 snprintf(dirname, sizeof(dirname),
788 SYSFS_PCI_DEVICES "/" PCI_PRI_FMT "/msi_irqs",
789 loc->domain, loc->bus, loc->devid, loc->function);
791 d = opendir(dirname);
799 virtio_has_msix(const struct rte_pci_addr *loc __rte_unused)
801 /* nic_uio does not enable interrupts, return 0 (false). */
807 * This function is based on probe() function in virtio_pci.c
808 * It returns 0 on success.
811 eth_virtio_dev_init(__rte_unused struct eth_driver *eth_drv,
812 struct rte_eth_dev *eth_dev)
814 struct virtio_net_config *config;
815 struct virtio_net_config local_config;
816 uint32_t offset_conf = sizeof(config->mac);
817 struct rte_pci_device *pci_dev;
818 struct virtio_hw *hw =
819 VIRTIO_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
821 if (RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr)) {
823 "MBUF HEADROOM should be enough to hold virtio net hdr\n");
827 eth_dev->dev_ops = &virtio_eth_dev_ops;
828 eth_dev->tx_pkt_burst = &virtio_xmit_pkts;
830 if (rte_eal_process_type() == RTE_PROC_SECONDARY)
833 pci_dev = eth_dev->pci_dev;
835 #ifdef RTE_EXEC_ENV_LINUXAPP
837 char dirname[PATH_MAX];
838 char filename[PATH_MAX];
839 unsigned long start, size;
841 if (get_uio_dev(&pci_dev->addr, dirname, sizeof(dirname)) < 0)
844 /* get portio size */
845 snprintf(filename, sizeof(filename),
846 "%s/portio/port0/size", dirname);
847 if (parse_sysfs_value(filename, &size) < 0) {
848 PMD_INIT_LOG(ERR, "%s(): cannot parse size",
853 /* get portio start */
854 snprintf(filename, sizeof(filename),
855 "%s/portio/port0/start", dirname);
856 if (parse_sysfs_value(filename, &start) < 0) {
857 PMD_INIT_LOG(ERR, "%s(): cannot parse portio start",
861 pci_dev->mem_resource[0].addr = (void *)(uintptr_t)start;
862 pci_dev->mem_resource[0].len = (uint64_t)size;
864 "PCI Port IO found start=0x%lx with size=0x%lx",
868 hw->use_msix = virtio_has_msix(&pci_dev->addr);
869 hw->io_base = (uint32_t)(uintptr_t)pci_dev->mem_resource[0].addr;
871 /* Reset the device although not necessary at startup */
874 /* Tell the host we've noticed this device. */
875 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
877 /* Tell the host we've known how to drive the device. */
878 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
879 virtio_negotiate_features(hw);
881 /* Setting up rx_header size for the device */
882 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
883 eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
884 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
886 eth_dev->rx_pkt_burst = &virtio_recv_pkts;
887 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
890 /* Allocate memory for storing MAC addresses */
891 eth_dev->data->mac_addrs = rte_zmalloc("virtio", ETHER_ADDR_LEN, 0);
892 if (eth_dev->data->mac_addrs == NULL) {
894 "Failed to allocate %d bytes needed to store MAC addresses",
899 /* Copy the permanent MAC address to: virtio_hw */
900 virtio_get_hwaddr(hw);
901 ether_addr_copy((struct ether_addr *) hw->mac_addr,
902 ð_dev->data->mac_addrs[0]);
904 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
905 hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
906 hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
908 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
909 config = &local_config;
911 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
912 offset_conf += sizeof(config->status);
915 "VIRTIO_NET_F_STATUS is not supported");
919 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
920 offset_conf += sizeof(config->max_virtqueue_pairs);
923 "VIRTIO_NET_F_MQ is not supported");
924 config->max_virtqueue_pairs = 1;
927 vtpci_read_dev_config(hw, 0, (uint8_t *)config, offset_conf);
930 (VIRTIO_MAX_RX_QUEUES < config->max_virtqueue_pairs) ?
931 VIRTIO_MAX_RX_QUEUES : config->max_virtqueue_pairs;
933 (VIRTIO_MAX_TX_QUEUES < config->max_virtqueue_pairs) ?
934 VIRTIO_MAX_TX_QUEUES : config->max_virtqueue_pairs;
936 virtio_dev_cq_queue_setup(eth_dev,
937 config->max_virtqueue_pairs * 2,
940 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
941 config->max_virtqueue_pairs);
942 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
944 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
945 config->mac[0], config->mac[1],
946 config->mac[2], config->mac[3],
947 config->mac[4], config->mac[5]);
949 hw->max_rx_queues = 1;
950 hw->max_tx_queues = 1;
953 eth_dev->data->nb_rx_queues = hw->max_rx_queues;
954 eth_dev->data->nb_tx_queues = hw->max_tx_queues;
956 PMD_INIT_LOG(DEBUG, "hw->max_rx_queues=%d hw->max_tx_queues=%d",
957 hw->max_rx_queues, hw->max_tx_queues);
958 PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
959 eth_dev->data->port_id, pci_dev->id.vendor_id,
960 pci_dev->id.device_id);
964 static struct eth_driver rte_virtio_pmd = {
966 .name = "rte_virtio_pmd",
967 .id_table = pci_id_virtio_map,
968 .drv_flags = RTE_PCI_DRV_NEED_MAPPING,
970 .eth_dev_init = eth_virtio_dev_init,
971 .dev_private_size = sizeof(struct virtio_adapter),
975 * Driver initialization routine.
976 * Invoked once at EAL init time.
977 * Register itself as the [Poll Mode] Driver of PCI virtio devices.
978 * Returns 0 on success.
981 rte_virtio_pmd_init(const char *name __rte_unused,
982 const char *param __rte_unused)
984 if (rte_eal_iopl_init() != 0) {
985 PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
989 rte_eth_driver_register(&rte_virtio_pmd);
994 * Only 1 queue is supported, no queue release related operation
997 virtio_dev_rx_queue_release(__rte_unused void *rxq)
1002 virtio_dev_tx_queue_release(__rte_unused void *txq)
1007 * Configure virtio device
1008 * It returns 0 on success.
1011 virtio_dev_configure(struct rte_eth_dev *dev)
1013 const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1015 PMD_INIT_LOG(DEBUG, "configure");
1017 if (rxmode->hw_ip_checksum) {
1018 PMD_DRV_LOG(ERR, "HW IP checksum not supported");
1027 virtio_dev_start(struct rte_eth_dev *dev)
1029 uint16_t nb_queues, i;
1031 struct virtio_hw *hw =
1032 VIRTIO_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1034 /* Tell the host we've noticed this device. */
1035 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1037 /* Tell the host we've known how to drive the device. */
1038 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1040 virtio_dev_cq_start(dev);
1042 /* Do final configuration before rx/tx engine starts */
1043 virtio_dev_rxtx_start(dev);
1045 /* Check VIRTIO_NET_F_STATUS for link status*/
1046 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1047 vtpci_read_dev_config(hw,
1048 offsetof(struct virtio_net_config, status),
1049 &status, sizeof(status));
1050 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
1051 PMD_INIT_LOG(ERR, "Port: %d Link is DOWN",
1052 dev->data->port_id);
1055 PMD_INIT_LOG(DEBUG, "Port: %d Link is UP",
1056 dev->data->port_id);
1059 vtpci_reinit_complete(hw);
1061 /*Notify the backend
1062 *Otherwise the tap backend might already stop its queue due to fullness.
1063 *vhost backend will have no chance to be waked up
1065 nb_queues = dev->data->nb_rx_queues;
1066 if (nb_queues > 1) {
1067 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
1071 PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
1073 for (i = 0; i < nb_queues; i++)
1074 virtqueue_notify(dev->data->rx_queues[i]);
1076 PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
1078 for (i = 0; i < dev->data->nb_rx_queues; i++)
1079 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
1081 for (i = 0; i < dev->data->nb_tx_queues; i++)
1082 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
1087 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
1089 struct rte_mbuf *buf;
1090 int i, mbuf_num = 0;
1092 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1094 "Before freeing rxq[%d] used and unused buf", i);
1095 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
1097 while ((buf = (struct rte_mbuf *)virtqueue_detatch_unused(
1098 dev->data->rx_queues[i])) != NULL) {
1099 rte_pktmbuf_free(buf);
1103 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1105 "After freeing rxq[%d] used and unused buf", i);
1106 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
1109 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1111 "Before freeing txq[%d] used and unused bufs",
1113 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
1116 while ((buf = (struct rte_mbuf *)virtqueue_detatch_unused(
1117 dev->data->tx_queues[i])) != NULL) {
1118 rte_pktmbuf_free(buf);
1123 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1125 "After freeing txq[%d] used and unused buf", i);
1126 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
1131 * Stop device: disable rx and tx functions to allow for reconfiguring.
1134 virtio_dev_stop(struct rte_eth_dev *dev)
1136 struct virtio_hw *hw =
1137 VIRTIO_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1141 virtio_dev_free_mbufs(dev);
1145 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
1147 struct rte_eth_link link, old;
1149 struct virtio_hw *hw =
1150 VIRTIO_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1151 memset(&link, 0, sizeof(link));
1152 virtio_dev_atomic_read_link_status(dev, &link);
1154 link.link_duplex = FULL_DUPLEX;
1155 link.link_speed = SPEED_10G;
1156 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1157 PMD_INIT_LOG(DEBUG, "Get link status from hw");
1158 vtpci_read_dev_config(hw,
1159 offsetof(struct virtio_net_config, status),
1160 &status, sizeof(status));
1161 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
1162 link.link_status = 0;
1163 PMD_INIT_LOG(DEBUG, "Port %d is down",
1164 dev->data->port_id);
1166 link.link_status = 1;
1167 PMD_INIT_LOG(DEBUG, "Port %d is up",
1168 dev->data->port_id);
1171 link.link_status = 1; /* Link up */
1173 virtio_dev_atomic_write_link_status(dev, &link);
1174 if (old.link_status == link.link_status)
1181 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1183 struct virtio_hw *hw = VIRTIO_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1185 dev_info->driver_name = dev->driver->pci_drv.name;
1186 dev_info->max_rx_queues = (uint16_t)hw->max_rx_queues;
1187 dev_info->max_tx_queues = (uint16_t)hw->max_tx_queues;
1188 dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
1189 dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
1190 dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
1194 * It enables testpmd to collect per queue stats.
1197 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
1198 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
1199 __rte_unused uint8_t is_rx)
1204 static struct rte_driver rte_virtio_driver = {
1206 .init = rte_virtio_pmd_init,
1209 PMD_REGISTER_DRIVER(rte_virtio_driver);