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_info_get(struct rte_eth_dev *dev,
70 struct rte_eth_dev_info *dev_info);
71 static int virtio_dev_link_update(struct rte_eth_dev *dev,
72 __rte_unused int wait_to_complete);
74 static void virtio_set_hwaddr(struct virtio_hw *hw);
75 static void virtio_get_hwaddr(struct virtio_hw *hw);
77 static void virtio_dev_rx_queue_release(__rte_unused void *rxq);
78 static void virtio_dev_tx_queue_release(__rte_unused void *txq);
80 static void virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats);
81 static void virtio_dev_stats_reset(struct rte_eth_dev *dev);
82 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev);
84 static int virtio_dev_queue_stats_mapping_set(
85 __rte_unused struct rte_eth_dev *eth_dev,
86 __rte_unused uint16_t queue_id,
87 __rte_unused uint8_t stat_idx,
88 __rte_unused uint8_t is_rx);
91 * The set of PCI devices this driver supports
93 static struct rte_pci_id pci_id_virtio_map[] = {
95 #define RTE_PCI_DEV_ID_DECL_VIRTIO(vend, dev) {RTE_PCI_DEVICE(vend, dev)},
96 #include "rte_pci_dev_ids.h"
98 { .vendor_id = 0, /* sentinel */ },
102 virtio_send_command(struct virtqueue *vq, struct virtio_pmd_ctrl *ctrl,
103 int *dlen, int pkt_num)
105 uint32_t head = vq->vq_desc_head_idx, i;
107 virtio_net_ctrl_ack status = ~0;
108 struct virtio_pmd_ctrl result;
110 ctrl->status = status;
114 "%s(): Control queue is not supported.",
119 PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
120 "vq->hw->cvq = %p vq = %p",
121 vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
123 if ((vq->vq_free_cnt < ((uint32_t)pkt_num + 2)) || (pkt_num < 1))
126 memcpy(vq->virtio_net_hdr_mz->addr, ctrl,
127 sizeof(struct virtio_pmd_ctrl));
130 * Format is enforced in qemu code:
131 * One TX packet for header;
132 * At least one TX packet per argument;
133 * One RX packet for ACK.
135 vq->vq_ring.desc[head].flags = VRING_DESC_F_NEXT;
136 vq->vq_ring.desc[head].addr = vq->virtio_net_hdr_mz->phys_addr;
137 vq->vq_ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
139 i = vq->vq_ring.desc[head].next;
141 for (k = 0; k < pkt_num; k++) {
142 vq->vq_ring.desc[i].flags = VRING_DESC_F_NEXT;
143 vq->vq_ring.desc[i].addr = vq->virtio_net_hdr_mz->phys_addr
144 + sizeof(struct virtio_net_ctrl_hdr)
145 + sizeof(ctrl->status) + sizeof(uint8_t)*sum;
146 vq->vq_ring.desc[i].len = dlen[k];
149 i = vq->vq_ring.desc[i].next;
152 vq->vq_ring.desc[i].flags = VRING_DESC_F_WRITE;
153 vq->vq_ring.desc[i].addr = vq->virtio_net_hdr_mz->phys_addr
154 + sizeof(struct virtio_net_ctrl_hdr);
155 vq->vq_ring.desc[i].len = sizeof(ctrl->status);
158 vq->vq_desc_head_idx = vq->vq_ring.desc[i].next;
160 vq_update_avail_ring(vq, head);
161 vq_update_avail_idx(vq);
163 PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
165 virtqueue_notify(vq);
167 while (vq->vq_used_cons_idx == vq->vq_ring.used->idx)
170 while (vq->vq_used_cons_idx != vq->vq_ring.used->idx) {
171 uint32_t idx, desc_idx, used_idx;
172 struct vring_used_elem *uep;
176 used_idx = (uint32_t)(vq->vq_used_cons_idx
177 & (vq->vq_nentries - 1));
178 uep = &vq->vq_ring.used->ring[used_idx];
179 idx = (uint32_t) uep->id;
182 while (vq->vq_ring.desc[desc_idx].flags & VRING_DESC_F_NEXT) {
183 desc_idx = vq->vq_ring.desc[desc_idx].next;
187 vq->vq_ring.desc[desc_idx].next = vq->vq_desc_head_idx;
188 vq->vq_desc_head_idx = idx;
190 vq->vq_used_cons_idx++;
194 PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
195 vq->vq_free_cnt, vq->vq_desc_head_idx);
197 memcpy(&result, vq->virtio_net_hdr_mz->addr,
198 sizeof(struct virtio_pmd_ctrl));
200 return result.status;
204 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
207 = VIRTIO_DEV_PRIVATE_TO_HW(dev->data->dev_private);
208 struct virtio_pmd_ctrl ctrl;
212 ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
213 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
214 memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
216 dlen[0] = sizeof(uint16_t);
218 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
221 PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
222 "failed, this is too late now...");
229 int virtio_dev_queue_setup(struct rte_eth_dev *dev,
232 uint8_t vtpci_queue_idx,
234 unsigned int socket_id,
235 struct virtqueue **pvq)
237 char vq_name[VIRTQUEUE_MAX_NAME_SZ];
238 const struct rte_memzone *mz;
241 struct virtio_hw *hw =
242 VIRTIO_DEV_PRIVATE_TO_HW(dev->data->dev_private);
243 struct virtqueue *vq = NULL;
245 /* Write the virtqueue index to the Queue Select Field */
246 VIRTIO_WRITE_REG_2(hw, VIRTIO_PCI_QUEUE_SEL, vtpci_queue_idx);
247 PMD_INIT_LOG(DEBUG, "selecting queue: %d", vtpci_queue_idx);
250 * Read the virtqueue size from the Queue Size field
251 * Always power of 2 and if 0 virtqueue does not exist
253 vq_size = VIRTIO_READ_REG_2(hw, VIRTIO_PCI_QUEUE_NUM);
254 PMD_INIT_LOG(DEBUG, "vq_size: %d nb_desc:%d", vq_size, nb_desc);
258 PMD_INIT_LOG(ERR, "%s: virtqueue does not exist", __func__);
260 } else if (!rte_is_power_of_2(vq_size)) {
261 PMD_INIT_LOG(ERR, "%s: virtqueue size is not powerof 2", __func__);
263 } else if (nb_desc != vq_size) {
264 PMD_INIT_LOG(ERR, "Warning: nb_desc(%d) is not equal to vq size (%d), fall to vq size",
269 if (queue_type == VTNET_RQ) {
270 snprintf(vq_name, sizeof(vq_name), "port%d_rvq%d",
271 dev->data->port_id, queue_idx);
272 vq = rte_zmalloc(vq_name, sizeof(struct virtqueue) +
273 vq_size * sizeof(struct vq_desc_extra), CACHE_LINE_SIZE);
274 } else if (queue_type == VTNET_TQ) {
275 snprintf(vq_name, sizeof(vq_name), "port%d_tvq%d",
276 dev->data->port_id, queue_idx);
277 vq = rte_zmalloc(vq_name, sizeof(struct virtqueue) +
278 vq_size * sizeof(struct vq_desc_extra), CACHE_LINE_SIZE);
279 } else if (queue_type == VTNET_CQ) {
280 snprintf(vq_name, sizeof(vq_name), "port%d_cvq",
282 vq = rte_zmalloc(vq_name, sizeof(struct virtqueue) +
283 vq_size * sizeof(struct vq_desc_extra),
287 PMD_INIT_LOG(ERR, "%s: Can not allocate virtqueue", __func__);
292 vq->port_id = dev->data->port_id;
293 vq->queue_id = queue_idx;
294 vq->vq_queue_index = vtpci_queue_idx;
295 vq->vq_alignment = VIRTIO_PCI_VRING_ALIGN;
296 vq->vq_nentries = vq_size;
297 vq->vq_free_cnt = vq_size;
300 * Reserve a memzone for vring elements
302 size = vring_size(vq_size, VIRTIO_PCI_VRING_ALIGN);
303 vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
304 PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d", size, vq->vq_ring_size);
306 mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
307 socket_id, 0, VIRTIO_PCI_VRING_ALIGN);
314 * Virtio PCI device VIRTIO_PCI_QUEUE_PF register is 32bit,
315 * and only accepts 32 bit page frame number.
316 * Check if the allocated physical memory exceeds 16TB.
318 if ((mz->phys_addr + vq->vq_ring_size - 1) >> (VIRTIO_PCI_QUEUE_ADDR_SHIFT + 32)) {
319 PMD_INIT_LOG(ERR, "vring address shouldn't be above 16TB!");
324 memset(mz->addr, 0, sizeof(mz->len));
326 vq->vq_ring_mem = mz->phys_addr;
327 vq->vq_ring_virt_mem = mz->addr;
328 PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem: 0x%"PRIx64, (uint64_t)mz->phys_addr);
329 PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%"PRIx64, (uint64_t)mz->addr);
330 vq->virtio_net_hdr_mz = NULL;
331 vq->virtio_net_hdr_mem = 0;
333 if (queue_type == VTNET_TQ) {
335 * For each xmit packet, allocate a virtio_net_hdr
337 snprintf(vq_name, sizeof(vq_name), "port%d_tvq%d_hdrzone",
338 dev->data->port_id, queue_idx);
339 vq->virtio_net_hdr_mz = rte_memzone_reserve_aligned(vq_name,
340 vq_size * sizeof(struct virtio_net_hdr),
341 socket_id, 0, CACHE_LINE_SIZE);
342 if (vq->virtio_net_hdr_mz == NULL) {
346 vq->virtio_net_hdr_mem =
347 vq->virtio_net_hdr_mz->phys_addr;
348 memset(vq->virtio_net_hdr_mz->addr, 0,
349 vq_size * sizeof(struct virtio_net_hdr));
350 } else if (queue_type == VTNET_CQ) {
351 /* Allocate a page for control vq command, data and status */
352 snprintf(vq_name, sizeof(vq_name), "port%d_cvq_hdrzone",
354 vq->virtio_net_hdr_mz = rte_memzone_reserve_aligned(vq_name,
355 PAGE_SIZE, socket_id, 0, CACHE_LINE_SIZE);
356 if (vq->virtio_net_hdr_mz == NULL) {
360 vq->virtio_net_hdr_mem =
361 vq->virtio_net_hdr_mz->phys_addr;
362 memset(vq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
366 * Set guest physical address of the virtqueue
367 * in VIRTIO_PCI_QUEUE_PFN config register of device
369 VIRTIO_WRITE_REG_4(hw, VIRTIO_PCI_QUEUE_PFN,
370 mz->phys_addr >> VIRTIO_PCI_QUEUE_ADDR_SHIFT);
376 virtio_dev_cq_queue_setup(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx,
379 struct virtqueue *vq;
380 uint16_t nb_desc = 0;
382 struct virtio_hw *hw =
383 VIRTIO_DEV_PRIVATE_TO_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 * dev_ops for virtio, bare necessities for basic operation
409 static struct eth_dev_ops virtio_eth_dev_ops = {
410 .dev_configure = virtio_dev_configure,
411 .dev_start = virtio_dev_start,
412 .dev_stop = virtio_dev_stop,
413 .dev_close = virtio_dev_close,
415 .dev_infos_get = virtio_dev_info_get,
416 .stats_get = virtio_dev_stats_get,
417 .stats_reset = virtio_dev_stats_reset,
418 .link_update = virtio_dev_link_update,
419 .mac_addr_add = NULL,
420 .mac_addr_remove = NULL,
421 .rx_queue_setup = virtio_dev_rx_queue_setup,
422 /* meaningfull only to multiple queue */
423 .rx_queue_release = virtio_dev_rx_queue_release,
424 .tx_queue_setup = virtio_dev_tx_queue_setup,
425 /* meaningfull only to multiple queue */
426 .tx_queue_release = virtio_dev_tx_queue_release,
427 /* collect stats per queue */
428 .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
432 virtio_dev_atomic_read_link_status(struct rte_eth_dev *dev,
433 struct rte_eth_link *link)
435 struct rte_eth_link *dst = link;
436 struct rte_eth_link *src = &(dev->data->dev_link);
438 if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
439 *(uint64_t *)src) == 0)
446 * Atomically writes the link status information into global
447 * structure rte_eth_dev.
450 * - Pointer to the structure rte_eth_dev to read from.
451 * - Pointer to the buffer to be saved with the link status.
454 * - On success, zero.
455 * - On failure, negative value.
458 virtio_dev_atomic_write_link_status(struct rte_eth_dev *dev,
459 struct rte_eth_link *link)
461 struct rte_eth_link *dst = &(dev->data->dev_link);
462 struct rte_eth_link *src = link;
464 if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
465 *(uint64_t *)src) == 0)
472 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
476 for (i = 0; i < dev->data->nb_tx_queues; i++) {
477 const struct virtqueue *txvq = dev->data->tx_queues[i];
481 stats->opackets += txvq->packets;
482 stats->obytes += txvq->bytes;
483 stats->oerrors += txvq->errors;
485 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
486 stats->q_opackets[i] = txvq->packets;
487 stats->q_obytes[i] = txvq->bytes;
491 for (i = 0; i < dev->data->nb_rx_queues; i++) {
492 const struct virtqueue *rxvq = dev->data->rx_queues[i];
496 stats->ipackets += rxvq->packets;
497 stats->ibytes += rxvq->bytes;
498 stats->ierrors += rxvq->errors;
500 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
501 stats->q_ipackets[i] = rxvq->packets;
502 stats->q_ibytes[i] = rxvq->bytes;
506 stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
510 virtio_dev_stats_reset(struct rte_eth_dev *dev)
514 for (i = 0; i < dev->data->nb_tx_queues; i++) {
515 struct virtqueue *txvq = dev->data->tx_queues[i];
524 for (i = 0; i < dev->data->nb_rx_queues; i++) {
525 struct virtqueue *rxvq = dev->data->rx_queues[i];
534 dev->data->rx_mbuf_alloc_failed = 0;
538 virtio_set_hwaddr(struct virtio_hw *hw)
540 vtpci_write_dev_config(hw,
541 offsetof(struct virtio_net_config, mac),
542 &hw->mac_addr, ETHER_ADDR_LEN);
546 virtio_get_hwaddr(struct virtio_hw *hw)
548 if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
549 vtpci_read_dev_config(hw,
550 offsetof(struct virtio_net_config, mac),
551 &hw->mac_addr, ETHER_ADDR_LEN);
553 eth_random_addr(&hw->mac_addr[0]);
554 virtio_set_hwaddr(hw);
560 virtio_negotiate_features(struct virtio_hw *hw)
562 uint32_t host_features, mask;
564 mask = VIRTIO_NET_F_CTRL_RX | VIRTIO_NET_F_CTRL_VLAN;
565 mask |= VIRTIO_NET_F_CSUM | VIRTIO_NET_F_GUEST_CSUM;
567 /* TSO and LRO are only available when their corresponding
568 * checksum offload feature is also negotiated.
570 mask |= VIRTIO_NET_F_HOST_TSO4 | VIRTIO_NET_F_HOST_TSO6 | VIRTIO_NET_F_HOST_ECN;
571 mask |= VIRTIO_NET_F_GUEST_TSO4 | VIRTIO_NET_F_GUEST_TSO6 | VIRTIO_NET_F_GUEST_ECN;
572 mask |= VTNET_LRO_FEATURES;
574 /* rx_mbuf should not be in multiple merged segments */
575 mask |= VIRTIO_NET_F_MRG_RXBUF;
577 /* not negotiating INDIRECT descriptor table support */
578 mask |= VIRTIO_RING_F_INDIRECT_DESC;
580 /* Prepare guest_features: feature that driver wants to support */
581 hw->guest_features = VTNET_FEATURES & ~mask;
582 PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %x",
585 /* Read device(host) feature bits */
586 host_features = VIRTIO_READ_REG_4(hw, VIRTIO_PCI_HOST_FEATURES);
587 PMD_INIT_LOG(DEBUG, "host_features before negotiate = %x",
591 * Negotiate features: Subset of device feature bits are written back
592 * guest feature bits.
594 hw->guest_features = vtpci_negotiate_features(hw, host_features);
595 PMD_INIT_LOG(DEBUG, "features after negotiate = %x",
599 #ifdef RTE_EXEC_ENV_LINUXAPP
601 parse_sysfs_value(const char *filename, unsigned long *val)
607 f = fopen(filename, "r");
609 PMD_INIT_LOG(ERR, "%s(): cannot open sysfs value %s",
614 if (fgets(buf, sizeof(buf), f) == NULL) {
615 PMD_INIT_LOG(ERR, "%s(): cannot read sysfs value %s",
620 *val = strtoul(buf, &end, 0);
621 if ((buf[0] == '\0') || (end == NULL) || (*end != '\n')) {
622 PMD_INIT_LOG(ERR, "%s(): cannot parse sysfs value %s",
631 static int get_uio_dev(struct rte_pci_addr *loc, char *buf, unsigned int buflen)
633 unsigned int uio_num;
636 char dirname[PATH_MAX];
638 /* depending on kernel version, uio can be located in uio/uioX
640 snprintf(dirname, sizeof(dirname),
641 SYSFS_PCI_DEVICES "/" PCI_PRI_FMT "/uio",
642 loc->domain, loc->bus, loc->devid, loc->function);
643 dir = opendir(dirname);
645 /* retry with the parent directory */
646 snprintf(dirname, sizeof(dirname),
647 SYSFS_PCI_DEVICES "/" PCI_PRI_FMT,
648 loc->domain, loc->bus, loc->devid, loc->function);
649 dir = opendir(dirname);
652 PMD_INIT_LOG(ERR, "Cannot opendir %s", dirname);
657 /* take the first file starting with "uio" */
658 while ((e = readdir(dir)) != NULL) {
659 /* format could be uio%d ...*/
660 int shortprefix_len = sizeof("uio") - 1;
661 /* ... or uio:uio%d */
662 int longprefix_len = sizeof("uio:uio") - 1;
665 if (strncmp(e->d_name, "uio", 3) != 0)
668 /* first try uio%d */
670 uio_num = strtoull(e->d_name + shortprefix_len, &endptr, 10);
671 if (errno == 0 && endptr != (e->d_name + shortprefix_len)) {
672 snprintf(buf, buflen, "%s/uio%u", dirname, uio_num);
676 /* then try uio:uio%d */
678 uio_num = strtoull(e->d_name + longprefix_len, &endptr, 10);
679 if (errno == 0 && endptr != (e->d_name + longprefix_len)) {
680 snprintf(buf, buflen, "%s/uio:uio%u", dirname,
687 /* No uio resource found */
689 PMD_INIT_LOG(ERR, "Could not find uio resource");
697 virtio_has_msix(const struct rte_pci_addr *loc)
700 char dirname[PATH_MAX];
702 snprintf(dirname, sizeof(dirname),
703 SYSFS_PCI_DEVICES "/" PCI_PRI_FMT "/msi_irqs",
704 loc->domain, loc->bus, loc->devid, loc->function);
706 d = opendir(dirname);
714 virtio_has_msix(const struct rte_pci_addr *loc __rte_unused)
716 /* nic_uio does not enable interrupts, return 0 (false). */
722 * This function is based on probe() function in virtio_pci.c
723 * It returns 0 on success.
726 eth_virtio_dev_init(__rte_unused struct eth_driver *eth_drv,
727 struct rte_eth_dev *eth_dev)
729 struct virtio_net_config *config;
730 struct virtio_net_config local_config;
731 uint32_t offset_conf = sizeof(config->mac);
732 struct rte_pci_device *pci_dev;
733 struct virtio_hw *hw =
734 VIRTIO_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
736 if (RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr)) {
738 "MBUF HEADROOM should be enough to hold virtio net hdr\n");
742 if (!(rte_eal_get_configuration()->flags & EAL_FLG_HIGH_IOPL)) {
744 "IOPL call failed in EAL init - cannot use virtio PMD driver\n");
748 eth_dev->dev_ops = &virtio_eth_dev_ops;
749 eth_dev->rx_pkt_burst = &virtio_recv_pkts;
750 eth_dev->tx_pkt_burst = &virtio_xmit_pkts;
752 if (rte_eal_process_type() == RTE_PROC_SECONDARY)
755 pci_dev = eth_dev->pci_dev;
757 #ifdef RTE_EXEC_ENV_LINUXAPP
759 char dirname[PATH_MAX];
760 char filename[PATH_MAX];
761 unsigned long start, size;
763 if (get_uio_dev(&pci_dev->addr, dirname, sizeof(dirname)) < 0)
766 /* get portio size */
767 snprintf(filename, sizeof(filename),
768 "%s/portio/port0/size", dirname);
769 if (parse_sysfs_value(filename, &size) < 0) {
770 PMD_INIT_LOG(ERR, "%s(): cannot parse size",
775 /* get portio start */
776 snprintf(filename, sizeof(filename),
777 "%s/portio/port0/start", dirname);
778 if (parse_sysfs_value(filename, &start) < 0) {
779 PMD_INIT_LOG(ERR, "%s(): cannot parse portio start",
783 pci_dev->mem_resource[0].addr = (void *)(uintptr_t)start;
784 pci_dev->mem_resource[0].len = (uint64_t)size;
786 "PCI Port IO found start=0x%lx with size=0x%lx",
790 hw->use_msix = virtio_has_msix(&pci_dev->addr);
791 hw->io_base = (uint32_t)(uintptr_t)pci_dev->mem_resource[0].addr;
793 /* Reset the device although not necessary at startup */
796 /* Tell the host we've noticed this device. */
797 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
799 /* Tell the host we've known how to drive the device. */
800 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
801 virtio_negotiate_features(hw);
803 /* Setting up rx_header size for the device */
804 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF))
805 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
807 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
809 /* Allocate memory for storing MAC addresses */
810 eth_dev->data->mac_addrs = rte_zmalloc("virtio", ETHER_ADDR_LEN, 0);
811 if (eth_dev->data->mac_addrs == NULL) {
813 "Failed to allocate %d bytes needed to store MAC addresses",
818 /* Copy the permanent MAC address to: virtio_hw */
819 virtio_get_hwaddr(hw);
820 ether_addr_copy((struct ether_addr *) hw->mac_addr,
821 ð_dev->data->mac_addrs[0]);
823 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
824 hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
825 hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
827 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
828 config = &local_config;
830 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
831 offset_conf += sizeof(config->status);
834 "VIRTIO_NET_F_STATUS is not supported");
838 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
839 offset_conf += sizeof(config->max_virtqueue_pairs);
842 "VIRTIO_NET_F_MQ is not supported");
843 config->max_virtqueue_pairs = 1;
846 vtpci_read_dev_config(hw, 0, (uint8_t *)config, offset_conf);
849 (VIRTIO_MAX_RX_QUEUES < config->max_virtqueue_pairs) ?
850 VIRTIO_MAX_RX_QUEUES : config->max_virtqueue_pairs;
852 (VIRTIO_MAX_TX_QUEUES < config->max_virtqueue_pairs) ?
853 VIRTIO_MAX_TX_QUEUES : config->max_virtqueue_pairs;
855 virtio_dev_cq_queue_setup(eth_dev,
856 config->max_virtqueue_pairs * 2,
859 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
860 config->max_virtqueue_pairs);
861 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
863 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
864 config->mac[0], config->mac[1],
865 config->mac[2], config->mac[3],
866 config->mac[4], config->mac[5]);
868 hw->max_rx_queues = 1;
869 hw->max_tx_queues = 1;
872 eth_dev->data->nb_rx_queues = hw->max_rx_queues;
873 eth_dev->data->nb_tx_queues = hw->max_tx_queues;
875 PMD_INIT_LOG(DEBUG, "hw->max_rx_queues=%d hw->max_tx_queues=%d",
876 hw->max_rx_queues, hw->max_tx_queues);
877 PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
878 eth_dev->data->port_id, pci_dev->id.vendor_id,
879 pci_dev->id.device_id);
883 static struct eth_driver rte_virtio_pmd = {
885 .name = "rte_virtio_pmd",
886 .id_table = pci_id_virtio_map,
887 .drv_flags = RTE_PCI_DRV_NEED_MAPPING,
889 .eth_dev_init = eth_virtio_dev_init,
890 .dev_private_size = sizeof(struct virtio_adapter),
894 * Driver initialization routine.
895 * Invoked once at EAL init time.
896 * Register itself as the [Poll Mode] Driver of PCI virtio devices.
897 * Returns 0 on success.
900 rte_virtio_pmd_init(const char *name __rte_unused, const char *param __rte_unused)
902 rte_eth_driver_register(&rte_virtio_pmd);
907 * Only 1 queue is supported, no queue release related operation
910 virtio_dev_rx_queue_release(__rte_unused void *rxq)
915 virtio_dev_tx_queue_release(__rte_unused void *txq)
920 * Configure virtio device
921 * It returns 0 on success.
924 virtio_dev_configure(struct rte_eth_dev *dev)
926 const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
928 PMD_INIT_LOG(DEBUG, "configure");
930 if (rxmode->hw_ip_checksum) {
931 PMD_DRV_LOG(ERR, "HW IP checksum not supported");
940 virtio_dev_start(struct rte_eth_dev *dev)
942 uint16_t nb_queues, i;
944 struct virtio_hw *hw =
945 VIRTIO_DEV_PRIVATE_TO_HW(dev->data->dev_private);
947 /* Tell the host we've noticed this device. */
948 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
950 /* Tell the host we've known how to drive the device. */
951 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
953 virtio_dev_cq_start(dev);
955 /* Do final configuration before rx/tx engine starts */
956 virtio_dev_rxtx_start(dev);
958 /* Check VIRTIO_NET_F_STATUS for link status*/
959 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
960 vtpci_read_dev_config(hw,
961 offsetof(struct virtio_net_config, status),
962 &status, sizeof(status));
963 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
964 PMD_INIT_LOG(ERR, "Port: %d Link is DOWN",
968 PMD_INIT_LOG(DEBUG, "Port: %d Link is UP",
972 vtpci_reinit_complete(hw);
975 *Otherwise the tap backend might already stop its queue due to fullness.
976 *vhost backend will have no chance to be waked up
978 nb_queues = dev->data->nb_rx_queues;
980 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
984 PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
986 for (i = 0; i < nb_queues; i++)
987 virtqueue_notify(dev->data->rx_queues[i]);
989 PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
991 for (i = 0; i < dev->data->nb_rx_queues; i++)
992 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
994 for (i = 0; i < dev->data->nb_tx_queues; i++)
995 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
1000 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
1002 struct rte_mbuf *buf;
1003 int i, mbuf_num = 0;
1005 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1007 "Before freeing rxq[%d] used and unused buf", i);
1008 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
1010 while ((buf = (struct rte_mbuf *)virtqueue_detatch_unused(
1011 dev->data->rx_queues[i])) != NULL) {
1012 rte_pktmbuf_free_seg(buf);
1016 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1018 "After freeing rxq[%d] used and unused buf", i);
1019 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
1022 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1024 "Before freeing txq[%d] used and unused bufs",
1026 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
1029 while ((buf = (struct rte_mbuf *)virtqueue_detatch_unused(
1030 dev->data->tx_queues[i])) != NULL) {
1031 rte_pktmbuf_free_seg(buf);
1035 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1037 "After freeing txq[%d] used and unused buf", i);
1038 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
1043 * Stop device: disable rx and tx functions to allow for reconfiguring.
1046 virtio_dev_stop(struct rte_eth_dev *dev)
1048 struct virtio_hw *hw =
1049 VIRTIO_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1053 virtio_dev_free_mbufs(dev);
1057 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
1059 struct rte_eth_link link, old;
1061 struct virtio_hw *hw =
1062 VIRTIO_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1063 memset(&link, 0, sizeof(link));
1064 virtio_dev_atomic_read_link_status(dev, &link);
1066 link.link_duplex = FULL_DUPLEX;
1067 link.link_speed = SPEED_10G;
1068 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1069 PMD_INIT_LOG(DEBUG, "Get link status from hw");
1070 vtpci_read_dev_config(hw,
1071 offsetof(struct virtio_net_config, status),
1072 &status, sizeof(status));
1073 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
1074 link.link_status = 0;
1075 PMD_INIT_LOG(DEBUG, "Port %d is down",
1076 dev->data->port_id);
1078 link.link_status = 1;
1079 PMD_INIT_LOG(DEBUG, "Port %d is up",
1080 dev->data->port_id);
1083 link.link_status = 1; /* Link up */
1085 virtio_dev_atomic_write_link_status(dev, &link);
1086 if (old.link_status == link.link_status)
1093 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1095 struct virtio_hw *hw = VIRTIO_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1097 dev_info->driver_name = dev->driver->pci_drv.name;
1098 dev_info->max_rx_queues = (uint16_t)hw->max_rx_queues;
1099 dev_info->max_tx_queues = (uint16_t)hw->max_tx_queues;
1100 dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
1101 dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
1102 dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
1106 * It enables testpmd to collect per queue stats.
1109 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
1110 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
1111 __rte_unused uint8_t is_rx)
1116 static struct rte_driver rte_virtio_driver = {
1118 .init = rte_virtio_pmd_init,
1121 PMD_REGISTER_DRIVER(rte_virtio_driver);