1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2018 Intel Corporation
10 #include <sys/epoll.h>
11 #include <linux/virtio_net.h>
14 #include <rte_eal_paging.h>
15 #include <rte_malloc.h>
16 #include <rte_memory.h>
17 #include <rte_bus_pci.h>
18 #include <rte_vhost.h>
20 #include <rte_vdpa_dev.h>
22 #include <rte_spinlock.h>
24 #include <rte_kvargs.h>
25 #include <rte_devargs.h>
27 #include "base/ifcvf.h"
29 RTE_LOG_REGISTER(ifcvf_vdpa_logtype, pmd.vdpa.ifcvf, NOTICE);
30 #define DRV_LOG(level, fmt, args...) \
31 rte_log(RTE_LOG_ ## level, ifcvf_vdpa_logtype, \
32 "IFCVF %s(): " fmt "\n", __func__, ##args)
34 #define IFCVF_USED_RING_LEN(size) \
35 ((size) * sizeof(struct vring_used_elem) + sizeof(uint16_t) * 3)
37 #define IFCVF_VDPA_MODE "vdpa"
38 #define IFCVF_SW_FALLBACK_LM "sw-live-migration"
40 #define THREAD_NAME_LEN 16
42 static const char * const ifcvf_valid_arguments[] = {
48 struct ifcvf_internal {
49 struct rte_pci_device *pdev;
52 int vfio_container_fd;
55 pthread_t tid; /* thread for notify relay */
58 struct rte_vdpa_device *vdev;
61 rte_atomic32_t started;
62 rte_atomic32_t dev_attached;
63 rte_atomic32_t running;
66 bool sw_fallback_running;
67 /* mediated vring for sw fallback */
68 struct vring m_vring[IFCVF_MAX_QUEUES * 2];
69 /* eventfd for used ring interrupt */
70 int intr_fd[IFCVF_MAX_QUEUES * 2];
73 struct internal_list {
74 TAILQ_ENTRY(internal_list) next;
75 struct ifcvf_internal *internal;
78 TAILQ_HEAD(internal_list_head, internal_list);
79 static struct internal_list_head internal_list =
80 TAILQ_HEAD_INITIALIZER(internal_list);
82 static pthread_mutex_t internal_list_lock = PTHREAD_MUTEX_INITIALIZER;
84 static void update_used_ring(struct ifcvf_internal *internal, uint16_t qid);
86 static struct internal_list *
87 find_internal_resource_by_vdev(struct rte_vdpa_device *vdev)
90 struct internal_list *list;
92 pthread_mutex_lock(&internal_list_lock);
94 TAILQ_FOREACH(list, &internal_list, next) {
95 if (vdev == list->internal->vdev) {
101 pthread_mutex_unlock(&internal_list_lock);
109 static struct internal_list *
110 find_internal_resource_by_dev(struct rte_pci_device *pdev)
113 struct internal_list *list;
115 pthread_mutex_lock(&internal_list_lock);
117 TAILQ_FOREACH(list, &internal_list, next) {
118 if (!rte_pci_addr_cmp(&pdev->addr,
119 &list->internal->pdev->addr)) {
125 pthread_mutex_unlock(&internal_list_lock);
134 ifcvf_vfio_setup(struct ifcvf_internal *internal)
136 struct rte_pci_device *dev = internal->pdev;
137 char devname[RTE_DEV_NAME_MAX_LEN] = {0};
141 internal->vfio_dev_fd = -1;
142 internal->vfio_group_fd = -1;
143 internal->vfio_container_fd = -1;
145 rte_pci_device_name(&dev->addr, devname, RTE_DEV_NAME_MAX_LEN);
146 ret = rte_vfio_get_group_num(rte_pci_get_sysfs_path(), devname,
149 DRV_LOG(ERR, "%s failed to get IOMMU group", devname);
153 internal->vfio_container_fd = rte_vfio_container_create();
154 if (internal->vfio_container_fd < 0)
157 internal->vfio_group_fd = rte_vfio_container_group_bind(
158 internal->vfio_container_fd, iommu_group_num);
159 if (internal->vfio_group_fd < 0)
162 if (rte_pci_map_device(dev))
165 internal->vfio_dev_fd = dev->intr_handle.vfio_dev_fd;
167 for (i = 0; i < RTE_MIN(PCI_MAX_RESOURCE, IFCVF_PCI_MAX_RESOURCE);
169 internal->hw.mem_resource[i].addr =
170 internal->pdev->mem_resource[i].addr;
171 internal->hw.mem_resource[i].phys_addr =
172 internal->pdev->mem_resource[i].phys_addr;
173 internal->hw.mem_resource[i].len =
174 internal->pdev->mem_resource[i].len;
180 rte_vfio_container_destroy(internal->vfio_container_fd);
185 ifcvf_dma_map(struct ifcvf_internal *internal, bool do_map)
189 struct rte_vhost_memory *mem = NULL;
190 int vfio_container_fd;
192 ret = rte_vhost_get_mem_table(internal->vid, &mem);
194 DRV_LOG(ERR, "failed to get VM memory layout.");
198 vfio_container_fd = internal->vfio_container_fd;
200 for (i = 0; i < mem->nregions; i++) {
201 struct rte_vhost_mem_region *reg;
203 reg = &mem->regions[i];
204 DRV_LOG(INFO, "%s, region %u: HVA 0x%" PRIx64 ", "
205 "GPA 0x%" PRIx64 ", size 0x%" PRIx64 ".",
206 do_map ? "DMA map" : "DMA unmap", i,
207 reg->host_user_addr, reg->guest_phys_addr, reg->size);
210 ret = rte_vfio_container_dma_map(vfio_container_fd,
211 reg->host_user_addr, reg->guest_phys_addr,
214 DRV_LOG(ERR, "DMA map failed.");
218 ret = rte_vfio_container_dma_unmap(vfio_container_fd,
219 reg->host_user_addr, reg->guest_phys_addr,
222 DRV_LOG(ERR, "DMA unmap failed.");
235 hva_to_gpa(int vid, uint64_t hva)
237 struct rte_vhost_memory *mem = NULL;
238 struct rte_vhost_mem_region *reg;
242 if (rte_vhost_get_mem_table(vid, &mem) < 0)
245 for (i = 0; i < mem->nregions; i++) {
246 reg = &mem->regions[i];
248 if (hva >= reg->host_user_addr &&
249 hva < reg->host_user_addr + reg->size) {
250 gpa = hva - reg->host_user_addr + reg->guest_phys_addr;
262 vdpa_ifcvf_start(struct ifcvf_internal *internal)
264 struct ifcvf_hw *hw = &internal->hw;
267 struct rte_vhost_vring vq;
271 nr_vring = rte_vhost_get_vring_num(vid);
272 rte_vhost_get_negotiated_features(vid, &hw->req_features);
274 for (i = 0; i < nr_vring; i++) {
275 rte_vhost_get_vhost_vring(vid, i, &vq);
276 gpa = hva_to_gpa(vid, (uint64_t)(uintptr_t)vq.desc);
278 DRV_LOG(ERR, "Fail to get GPA for descriptor ring.");
281 hw->vring[i].desc = gpa;
283 gpa = hva_to_gpa(vid, (uint64_t)(uintptr_t)vq.avail);
285 DRV_LOG(ERR, "Fail to get GPA for available ring.");
288 hw->vring[i].avail = gpa;
290 gpa = hva_to_gpa(vid, (uint64_t)(uintptr_t)vq.used);
292 DRV_LOG(ERR, "Fail to get GPA for used ring.");
295 hw->vring[i].used = gpa;
297 hw->vring[i].size = vq.size;
298 rte_vhost_get_vring_base(vid, i, &hw->vring[i].last_avail_idx,
299 &hw->vring[i].last_used_idx);
303 return ifcvf_start_hw(&internal->hw);
307 vdpa_ifcvf_stop(struct ifcvf_internal *internal)
309 struct ifcvf_hw *hw = &internal->hw;
312 uint64_t features = 0;
313 uint64_t log_base = 0, log_size = 0;
319 for (i = 0; i < hw->nr_vring; i++)
320 rte_vhost_set_vring_base(vid, i, hw->vring[i].last_avail_idx,
321 hw->vring[i].last_used_idx);
326 rte_vhost_get_negotiated_features(vid, &features);
327 if (RTE_VHOST_NEED_LOG(features)) {
328 ifcvf_disable_logging(hw);
329 rte_vhost_get_log_base(internal->vid, &log_base, &log_size);
330 rte_vfio_container_dma_unmap(internal->vfio_container_fd,
331 log_base, IFCVF_LOG_BASE, log_size);
333 * IFCVF marks dirty memory pages for only packet buffer,
334 * SW helps to mark the used ring as dirty after device stops.
336 for (i = 0; i < hw->nr_vring; i++) {
337 len = IFCVF_USED_RING_LEN(hw->vring[i].size);
338 rte_vhost_log_used_vring(vid, i, 0, len);
343 #define MSIX_IRQ_SET_BUF_LEN (sizeof(struct vfio_irq_set) + \
344 sizeof(int) * (IFCVF_MAX_QUEUES * 2 + 1))
346 vdpa_enable_vfio_intr(struct ifcvf_internal *internal, bool m_rx)
349 uint32_t i, nr_vring;
350 char irq_set_buf[MSIX_IRQ_SET_BUF_LEN];
351 struct vfio_irq_set *irq_set;
353 struct rte_vhost_vring vring;
358 nr_vring = rte_vhost_get_vring_num(internal->vid);
360 irq_set = (struct vfio_irq_set *)irq_set_buf;
361 irq_set->argsz = sizeof(irq_set_buf);
362 irq_set->count = nr_vring + 1;
363 irq_set->flags = VFIO_IRQ_SET_DATA_EVENTFD |
364 VFIO_IRQ_SET_ACTION_TRIGGER;
365 irq_set->index = VFIO_PCI_MSIX_IRQ_INDEX;
367 fd_ptr = (int *)&irq_set->data;
368 fd_ptr[RTE_INTR_VEC_ZERO_OFFSET] = internal->pdev->intr_handle.fd;
370 for (i = 0; i < nr_vring; i++)
371 internal->intr_fd[i] = -1;
373 for (i = 0; i < nr_vring; i++) {
374 rte_vhost_get_vhost_vring(internal->vid, i, &vring);
375 fd_ptr[RTE_INTR_VEC_RXTX_OFFSET + i] = vring.callfd;
376 if ((i & 1) == 0 && m_rx == true) {
377 fd = eventfd(0, EFD_NONBLOCK | EFD_CLOEXEC);
379 DRV_LOG(ERR, "can't setup eventfd: %s",
383 internal->intr_fd[i] = fd;
384 fd_ptr[RTE_INTR_VEC_RXTX_OFFSET + i] = fd;
388 ret = ioctl(internal->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
390 DRV_LOG(ERR, "Error enabling MSI-X interrupts: %s",
399 vdpa_disable_vfio_intr(struct ifcvf_internal *internal)
402 uint32_t i, nr_vring;
403 char irq_set_buf[MSIX_IRQ_SET_BUF_LEN];
404 struct vfio_irq_set *irq_set;
406 irq_set = (struct vfio_irq_set *)irq_set_buf;
407 irq_set->argsz = sizeof(irq_set_buf);
409 irq_set->flags = VFIO_IRQ_SET_DATA_NONE | VFIO_IRQ_SET_ACTION_TRIGGER;
410 irq_set->index = VFIO_PCI_MSIX_IRQ_INDEX;
413 nr_vring = rte_vhost_get_vring_num(internal->vid);
414 for (i = 0; i < nr_vring; i++) {
415 if (internal->intr_fd[i] >= 0)
416 close(internal->intr_fd[i]);
417 internal->intr_fd[i] = -1;
420 ret = ioctl(internal->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
422 DRV_LOG(ERR, "Error disabling MSI-X interrupts: %s",
431 notify_relay(void *arg)
433 int i, kickfd, epfd, nfds = 0;
435 struct epoll_event events[IFCVF_MAX_QUEUES * 2];
436 struct epoll_event ev;
439 struct rte_vhost_vring vring;
440 struct ifcvf_internal *internal = (struct ifcvf_internal *)arg;
441 struct ifcvf_hw *hw = &internal->hw;
443 q_num = rte_vhost_get_vring_num(internal->vid);
445 epfd = epoll_create(IFCVF_MAX_QUEUES * 2);
447 DRV_LOG(ERR, "failed to create epoll instance.");
450 internal->epfd = epfd;
453 for (qid = 0; qid < q_num; qid++) {
454 ev.events = EPOLLIN | EPOLLPRI;
455 rte_vhost_get_vhost_vring(internal->vid, qid, &vring);
456 ev.data.u64 = qid | (uint64_t)vring.kickfd << 32;
457 if (epoll_ctl(epfd, EPOLL_CTL_ADD, vring.kickfd, &ev) < 0) {
458 DRV_LOG(ERR, "epoll add error: %s", strerror(errno));
464 nfds = epoll_wait(epfd, events, q_num, -1);
468 DRV_LOG(ERR, "epoll_wait return fail\n");
472 for (i = 0; i < nfds; i++) {
473 qid = events[i].data.u32;
474 kickfd = (uint32_t)(events[i].data.u64 >> 32);
476 nbytes = read(kickfd, &buf, 8);
478 if (errno == EINTR ||
479 errno == EWOULDBLOCK ||
482 DRV_LOG(INFO, "Error reading "
489 ifcvf_notify_queue(hw, qid);
497 setup_notify_relay(struct ifcvf_internal *internal)
499 char name[THREAD_NAME_LEN];
502 snprintf(name, sizeof(name), "ifc-notify-%d", internal->vid);
503 ret = rte_ctrl_thread_create(&internal->tid, name, NULL, notify_relay,
506 DRV_LOG(ERR, "failed to create notify relay pthread.");
514 unset_notify_relay(struct ifcvf_internal *internal)
519 pthread_cancel(internal->tid);
520 pthread_join(internal->tid, &status);
524 if (internal->epfd >= 0)
525 close(internal->epfd);
532 update_datapath(struct ifcvf_internal *internal)
536 rte_spinlock_lock(&internal->lock);
538 if (!rte_atomic32_read(&internal->running) &&
539 (rte_atomic32_read(&internal->started) &&
540 rte_atomic32_read(&internal->dev_attached))) {
541 ret = ifcvf_dma_map(internal, true);
545 ret = vdpa_enable_vfio_intr(internal, false);
549 ret = vdpa_ifcvf_start(internal);
553 ret = setup_notify_relay(internal);
557 rte_atomic32_set(&internal->running, 1);
558 } else if (rte_atomic32_read(&internal->running) &&
559 (!rte_atomic32_read(&internal->started) ||
560 !rte_atomic32_read(&internal->dev_attached))) {
561 ret = unset_notify_relay(internal);
565 vdpa_ifcvf_stop(internal);
567 ret = vdpa_disable_vfio_intr(internal);
571 ret = ifcvf_dma_map(internal, false);
575 rte_atomic32_set(&internal->running, 0);
578 rte_spinlock_unlock(&internal->lock);
581 rte_spinlock_unlock(&internal->lock);
586 m_ifcvf_start(struct ifcvf_internal *internal)
588 struct ifcvf_hw *hw = &internal->hw;
589 uint32_t i, nr_vring;
591 struct rte_vhost_vring vq;
593 uint64_t m_vring_iova = IFCVF_MEDIATED_VRING;
597 memset(&vq, 0, sizeof(vq));
599 nr_vring = rte_vhost_get_vring_num(vid);
600 rte_vhost_get_negotiated_features(vid, &hw->req_features);
602 for (i = 0; i < nr_vring; i++) {
603 rte_vhost_get_vhost_vring(vid, i, &vq);
605 size = RTE_ALIGN_CEIL(vring_size(vq.size, rte_mem_page_size()),
606 rte_mem_page_size());
607 vring_buf = rte_zmalloc("ifcvf", size, rte_mem_page_size());
608 vring_init(&internal->m_vring[i], vq.size, vring_buf,
609 rte_mem_page_size());
611 ret = rte_vfio_container_dma_map(internal->vfio_container_fd,
612 (uint64_t)(uintptr_t)vring_buf, m_vring_iova, size);
614 DRV_LOG(ERR, "mediated vring DMA map failed.");
618 gpa = hva_to_gpa(vid, (uint64_t)(uintptr_t)vq.desc);
620 DRV_LOG(ERR, "Fail to get GPA for descriptor ring.");
623 hw->vring[i].desc = gpa;
625 gpa = hva_to_gpa(vid, (uint64_t)(uintptr_t)vq.avail);
627 DRV_LOG(ERR, "Fail to get GPA for available ring.");
630 hw->vring[i].avail = gpa;
632 /* Direct I/O for Tx queue, relay for Rx queue */
634 gpa = hva_to_gpa(vid, (uint64_t)(uintptr_t)vq.used);
636 DRV_LOG(ERR, "Fail to get GPA for used ring.");
639 hw->vring[i].used = gpa;
641 hw->vring[i].used = m_vring_iova +
642 (char *)internal->m_vring[i].used -
643 (char *)internal->m_vring[i].desc;
646 hw->vring[i].size = vq.size;
648 rte_vhost_get_vring_base(vid, i,
649 &internal->m_vring[i].avail->idx,
650 &internal->m_vring[i].used->idx);
652 rte_vhost_get_vring_base(vid, i, &hw->vring[i].last_avail_idx,
653 &hw->vring[i].last_used_idx);
655 m_vring_iova += size;
657 hw->nr_vring = nr_vring;
659 return ifcvf_start_hw(&internal->hw);
662 for (i = 0; i < nr_vring; i++)
663 if (internal->m_vring[i].desc)
664 rte_free(internal->m_vring[i].desc);
670 m_ifcvf_stop(struct ifcvf_internal *internal)
674 struct rte_vhost_vring vq;
675 struct ifcvf_hw *hw = &internal->hw;
676 uint64_t m_vring_iova = IFCVF_MEDIATED_VRING;
682 for (i = 0; i < hw->nr_vring; i++) {
683 /* synchronize remaining new used entries if any */
685 update_used_ring(internal, i);
687 rte_vhost_get_vhost_vring(vid, i, &vq);
688 len = IFCVF_USED_RING_LEN(vq.size);
689 rte_vhost_log_used_vring(vid, i, 0, len);
691 size = RTE_ALIGN_CEIL(vring_size(vq.size, rte_mem_page_size()),
692 rte_mem_page_size());
693 rte_vfio_container_dma_unmap(internal->vfio_container_fd,
694 (uint64_t)(uintptr_t)internal->m_vring[i].desc,
697 rte_vhost_set_vring_base(vid, i, hw->vring[i].last_avail_idx,
698 hw->vring[i].last_used_idx);
699 rte_free(internal->m_vring[i].desc);
700 m_vring_iova += size;
707 update_used_ring(struct ifcvf_internal *internal, uint16_t qid)
709 rte_vdpa_relay_vring_used(internal->vid, qid, &internal->m_vring[qid]);
710 rte_vhost_vring_call(internal->vid, qid);
714 vring_relay(void *arg)
716 int i, vid, epfd, fd, nfds;
717 struct ifcvf_internal *internal = (struct ifcvf_internal *)arg;
718 struct rte_vhost_vring vring;
720 struct epoll_event events[IFCVF_MAX_QUEUES * 4];
721 struct epoll_event ev;
726 q_num = rte_vhost_get_vring_num(vid);
728 /* add notify fd and interrupt fd to epoll */
729 epfd = epoll_create(IFCVF_MAX_QUEUES * 2);
731 DRV_LOG(ERR, "failed to create epoll instance.");
734 internal->epfd = epfd;
737 for (qid = 0; qid < q_num; qid++) {
738 ev.events = EPOLLIN | EPOLLPRI;
739 rte_vhost_get_vhost_vring(vid, qid, &vring);
740 ev.data.u64 = qid << 1 | (uint64_t)vring.kickfd << 32;
741 if (epoll_ctl(epfd, EPOLL_CTL_ADD, vring.kickfd, &ev) < 0) {
742 DRV_LOG(ERR, "epoll add error: %s", strerror(errno));
747 for (qid = 0; qid < q_num; qid += 2) {
748 ev.events = EPOLLIN | EPOLLPRI;
749 /* leave a flag to mark it's for interrupt */
750 ev.data.u64 = 1 | qid << 1 |
751 (uint64_t)internal->intr_fd[qid] << 32;
752 if (epoll_ctl(epfd, EPOLL_CTL_ADD, internal->intr_fd[qid], &ev)
754 DRV_LOG(ERR, "epoll add error: %s", strerror(errno));
757 update_used_ring(internal, qid);
760 /* start relay with a first kick */
761 for (qid = 0; qid < q_num; qid++)
762 ifcvf_notify_queue(&internal->hw, qid);
764 /* listen to the events and react accordingly */
766 nfds = epoll_wait(epfd, events, q_num * 2, -1);
770 DRV_LOG(ERR, "epoll_wait return fail\n");
774 for (i = 0; i < nfds; i++) {
775 fd = (uint32_t)(events[i].data.u64 >> 32);
777 nbytes = read(fd, &buf, 8);
779 if (errno == EINTR ||
780 errno == EWOULDBLOCK ||
783 DRV_LOG(INFO, "Error reading "
790 qid = events[i].data.u32 >> 1;
792 if (events[i].data.u32 & 1)
793 update_used_ring(internal, qid);
795 ifcvf_notify_queue(&internal->hw, qid);
803 setup_vring_relay(struct ifcvf_internal *internal)
805 char name[THREAD_NAME_LEN];
808 snprintf(name, sizeof(name), "ifc-vring-%d", internal->vid);
809 ret = rte_ctrl_thread_create(&internal->tid, name, NULL, vring_relay,
812 DRV_LOG(ERR, "failed to create ring relay pthread.");
820 unset_vring_relay(struct ifcvf_internal *internal)
825 pthread_cancel(internal->tid);
826 pthread_join(internal->tid, &status);
830 if (internal->epfd >= 0)
831 close(internal->epfd);
838 ifcvf_sw_fallback_switchover(struct ifcvf_internal *internal)
841 int vid = internal->vid;
843 /* stop the direct IO data path */
844 unset_notify_relay(internal);
845 vdpa_ifcvf_stop(internal);
846 vdpa_disable_vfio_intr(internal);
848 ret = rte_vhost_host_notifier_ctrl(vid, RTE_VHOST_QUEUE_ALL, false);
849 if (ret && ret != -ENOTSUP)
852 /* set up interrupt for interrupt relay */
853 ret = vdpa_enable_vfio_intr(internal, true);
858 ret = m_ifcvf_start(internal);
862 /* set up vring relay thread */
863 ret = setup_vring_relay(internal);
867 rte_vhost_host_notifier_ctrl(vid, RTE_VHOST_QUEUE_ALL, true);
869 internal->sw_fallback_running = true;
874 m_ifcvf_stop(internal);
876 vdpa_disable_vfio_intr(internal);
878 ifcvf_dma_map(internal, false);
884 ifcvf_dev_config(int vid)
886 struct rte_vdpa_device *vdev;
887 struct internal_list *list;
888 struct ifcvf_internal *internal;
890 vdev = rte_vhost_get_vdpa_device(vid);
891 list = find_internal_resource_by_vdev(vdev);
893 DRV_LOG(ERR, "Invalid vDPA device: %p", vdev);
897 internal = list->internal;
899 rte_atomic32_set(&internal->dev_attached, 1);
900 update_datapath(internal);
902 if (rte_vhost_host_notifier_ctrl(vid, RTE_VHOST_QUEUE_ALL, true) != 0)
903 DRV_LOG(NOTICE, "vDPA (%s): software relay is used.",
906 internal->configured = 1;
911 ifcvf_dev_close(int vid)
913 struct rte_vdpa_device *vdev;
914 struct internal_list *list;
915 struct ifcvf_internal *internal;
917 vdev = rte_vhost_get_vdpa_device(vid);
918 list = find_internal_resource_by_vdev(vdev);
920 DRV_LOG(ERR, "Invalid vDPA device: %p", vdev);
924 internal = list->internal;
926 if (internal->sw_fallback_running) {
927 /* unset ring relay */
928 unset_vring_relay(internal);
931 m_ifcvf_stop(internal);
933 /* remove interrupt setting */
934 vdpa_disable_vfio_intr(internal);
936 /* unset DMA map for guest memory */
937 ifcvf_dma_map(internal, false);
939 internal->sw_fallback_running = false;
941 rte_atomic32_set(&internal->dev_attached, 0);
942 update_datapath(internal);
945 internal->configured = 0;
950 ifcvf_set_features(int vid)
952 uint64_t features = 0;
953 struct rte_vdpa_device *vdev;
954 struct internal_list *list;
955 struct ifcvf_internal *internal;
956 uint64_t log_base = 0, log_size = 0;
958 vdev = rte_vhost_get_vdpa_device(vid);
959 list = find_internal_resource_by_vdev(vdev);
961 DRV_LOG(ERR, "Invalid vDPA device: %p", vdev);
965 internal = list->internal;
966 rte_vhost_get_negotiated_features(vid, &features);
968 if (!RTE_VHOST_NEED_LOG(features))
971 if (internal->sw_lm) {
972 ifcvf_sw_fallback_switchover(internal);
974 rte_vhost_get_log_base(vid, &log_base, &log_size);
975 rte_vfio_container_dma_map(internal->vfio_container_fd,
976 log_base, IFCVF_LOG_BASE, log_size);
977 ifcvf_enable_logging(&internal->hw, IFCVF_LOG_BASE, log_size);
984 ifcvf_get_vfio_group_fd(int vid)
986 struct rte_vdpa_device *vdev;
987 struct internal_list *list;
989 vdev = rte_vhost_get_vdpa_device(vid);
990 list = find_internal_resource_by_vdev(vdev);
992 DRV_LOG(ERR, "Invalid vDPA device: %p", vdev);
996 return list->internal->vfio_group_fd;
1000 ifcvf_get_vfio_device_fd(int vid)
1002 struct rte_vdpa_device *vdev;
1003 struct internal_list *list;
1005 vdev = rte_vhost_get_vdpa_device(vid);
1006 list = find_internal_resource_by_vdev(vdev);
1008 DRV_LOG(ERR, "Invalid vDPA device: %p", vdev);
1012 return list->internal->vfio_dev_fd;
1016 ifcvf_get_notify_area(int vid, int qid, uint64_t *offset, uint64_t *size)
1018 struct rte_vdpa_device *vdev;
1019 struct internal_list *list;
1020 struct ifcvf_internal *internal;
1021 struct vfio_region_info reg = { .argsz = sizeof(reg) };
1024 vdev = rte_vhost_get_vdpa_device(vid);
1025 list = find_internal_resource_by_vdev(vdev);
1027 DRV_LOG(ERR, "Invalid vDPA device: %p", vdev);
1031 internal = list->internal;
1033 reg.index = ifcvf_get_notify_region(&internal->hw);
1034 ret = ioctl(internal->vfio_dev_fd, VFIO_DEVICE_GET_REGION_INFO, ®);
1036 DRV_LOG(ERR, "Get not get device region info: %s",
1041 *offset = ifcvf_get_queue_notify_off(&internal->hw, qid) + reg.offset;
1048 ifcvf_get_queue_num(struct rte_vdpa_device *vdev, uint32_t *queue_num)
1050 struct internal_list *list;
1052 list = find_internal_resource_by_vdev(vdev);
1054 DRV_LOG(ERR, "Invalid vDPA device: %p", vdev);
1058 *queue_num = list->internal->max_queues;
1064 ifcvf_get_vdpa_features(struct rte_vdpa_device *vdev, uint64_t *features)
1066 struct internal_list *list;
1068 list = find_internal_resource_by_vdev(vdev);
1070 DRV_LOG(ERR, "Invalid vDPA device: %p", vdev);
1074 *features = list->internal->features;
1079 #define VDPA_SUPPORTED_PROTOCOL_FEATURES \
1080 (1ULL << VHOST_USER_PROTOCOL_F_REPLY_ACK | \
1081 1ULL << VHOST_USER_PROTOCOL_F_SLAVE_REQ | \
1082 1ULL << VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD | \
1083 1ULL << VHOST_USER_PROTOCOL_F_HOST_NOTIFIER | \
1084 1ULL << VHOST_USER_PROTOCOL_F_LOG_SHMFD | \
1085 1ULL << VHOST_USER_PROTOCOL_F_STATUS)
1087 ifcvf_get_protocol_features(struct rte_vdpa_device *vdev, uint64_t *features)
1091 *features = VDPA_SUPPORTED_PROTOCOL_FEATURES;
1096 ifcvf_set_vring_state(int vid, int vring, int state)
1098 struct rte_vdpa_device *vdev;
1099 struct internal_list *list;
1100 struct ifcvf_internal *internal;
1101 struct ifcvf_hw *hw;
1102 struct ifcvf_pci_common_cfg *cfg;
1105 vdev = rte_vhost_get_vdpa_device(vid);
1106 list = find_internal_resource_by_vdev(vdev);
1108 DRV_LOG(ERR, "Invalid vDPA device: %p", vdev);
1112 internal = list->internal;
1113 if (vring < 0 || vring >= internal->max_queues * 2) {
1114 DRV_LOG(ERR, "Vring index %d not correct", vring);
1119 if (!internal->configured)
1122 cfg = hw->common_cfg;
1123 IFCVF_WRITE_REG16(vring, &cfg->queue_select);
1124 IFCVF_WRITE_REG16(!!state, &cfg->queue_enable);
1126 if (!state && hw->vring[vring].enable) {
1127 ret = vdpa_disable_vfio_intr(internal);
1132 if (state && !hw->vring[vring].enable) {
1133 ret = vdpa_enable_vfio_intr(internal, false);
1139 hw->vring[vring].enable = !!state;
1143 static struct rte_vdpa_dev_ops ifcvf_ops = {
1144 .get_queue_num = ifcvf_get_queue_num,
1145 .get_features = ifcvf_get_vdpa_features,
1146 .get_protocol_features = ifcvf_get_protocol_features,
1147 .dev_conf = ifcvf_dev_config,
1148 .dev_close = ifcvf_dev_close,
1149 .set_vring_state = ifcvf_set_vring_state,
1150 .set_features = ifcvf_set_features,
1151 .migration_done = NULL,
1152 .get_vfio_group_fd = ifcvf_get_vfio_group_fd,
1153 .get_vfio_device_fd = ifcvf_get_vfio_device_fd,
1154 .get_notify_area = ifcvf_get_notify_area,
1158 open_int(const char *key __rte_unused, const char *value, void *extra_args)
1160 uint16_t *n = extra_args;
1162 if (value == NULL || extra_args == NULL)
1165 *n = (uint16_t)strtoul(value, NULL, 0);
1166 if (*n == USHRT_MAX && errno == ERANGE)
1173 ifcvf_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1174 struct rte_pci_device *pci_dev)
1177 struct ifcvf_internal *internal = NULL;
1178 struct internal_list *list = NULL;
1180 int sw_fallback_lm = 0;
1181 struct rte_kvargs *kvlist = NULL;
1184 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1187 if (!pci_dev->device.devargs)
1190 kvlist = rte_kvargs_parse(pci_dev->device.devargs->args,
1191 ifcvf_valid_arguments);
1195 /* probe only when vdpa mode is specified */
1196 if (rte_kvargs_count(kvlist, IFCVF_VDPA_MODE) == 0) {
1197 rte_kvargs_free(kvlist);
1201 ret = rte_kvargs_process(kvlist, IFCVF_VDPA_MODE, &open_int,
1203 if (ret < 0 || vdpa_mode == 0) {
1204 rte_kvargs_free(kvlist);
1208 list = rte_zmalloc("ifcvf", sizeof(*list), 0);
1212 internal = rte_zmalloc("ifcvf", sizeof(*internal), 0);
1213 if (internal == NULL)
1216 internal->pdev = pci_dev;
1217 rte_spinlock_init(&internal->lock);
1219 if (ifcvf_vfio_setup(internal) < 0) {
1220 DRV_LOG(ERR, "failed to setup device %s", pci_dev->name);
1224 if (ifcvf_init_hw(&internal->hw, internal->pdev) < 0) {
1225 DRV_LOG(ERR, "failed to init device %s", pci_dev->name);
1229 internal->configured = 0;
1230 internal->max_queues = IFCVF_MAX_QUEUES;
1231 features = ifcvf_get_features(&internal->hw);
1232 internal->features = (features &
1233 ~(1ULL << VIRTIO_F_IOMMU_PLATFORM)) |
1234 (1ULL << VIRTIO_NET_F_GUEST_ANNOUNCE) |
1235 (1ULL << VIRTIO_NET_F_CTRL_VQ) |
1236 (1ULL << VIRTIO_NET_F_STATUS) |
1237 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES) |
1238 (1ULL << VHOST_F_LOG_ALL);
1240 list->internal = internal;
1242 if (rte_kvargs_count(kvlist, IFCVF_SW_FALLBACK_LM)) {
1243 ret = rte_kvargs_process(kvlist, IFCVF_SW_FALLBACK_LM,
1244 &open_int, &sw_fallback_lm);
1248 internal->sw_lm = sw_fallback_lm;
1250 internal->vdev = rte_vdpa_register_device(&pci_dev->device, &ifcvf_ops);
1251 if (internal->vdev == NULL) {
1252 DRV_LOG(ERR, "failed to register device %s", pci_dev->name);
1256 pthread_mutex_lock(&internal_list_lock);
1257 TAILQ_INSERT_TAIL(&internal_list, list, next);
1258 pthread_mutex_unlock(&internal_list_lock);
1260 rte_atomic32_set(&internal->started, 1);
1261 update_datapath(internal);
1263 rte_kvargs_free(kvlist);
1267 rte_kvargs_free(kvlist);
1274 ifcvf_pci_remove(struct rte_pci_device *pci_dev)
1276 struct ifcvf_internal *internal;
1277 struct internal_list *list;
1279 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1282 list = find_internal_resource_by_dev(pci_dev);
1284 DRV_LOG(ERR, "Invalid device: %s", pci_dev->name);
1288 internal = list->internal;
1289 rte_atomic32_set(&internal->started, 0);
1290 update_datapath(internal);
1292 rte_pci_unmap_device(internal->pdev);
1293 rte_vfio_container_destroy(internal->vfio_container_fd);
1294 rte_vdpa_unregister_device(internal->vdev);
1296 pthread_mutex_lock(&internal_list_lock);
1297 TAILQ_REMOVE(&internal_list, list, next);
1298 pthread_mutex_unlock(&internal_list_lock);
1307 * IFCVF has the same vendor ID and device ID as virtio net PCI
1308 * device, with its specific subsystem vendor ID and device ID.
1310 static const struct rte_pci_id pci_id_ifcvf_map[] = {
1311 { .class_id = RTE_CLASS_ANY_ID,
1312 .vendor_id = IFCVF_VENDOR_ID,
1313 .device_id = IFCVF_DEVICE_ID,
1314 .subsystem_vendor_id = IFCVF_SUBSYS_VENDOR_ID,
1315 .subsystem_device_id = IFCVF_SUBSYS_DEVICE_ID,
1318 { .vendor_id = 0, /* sentinel */
1322 static struct rte_pci_driver rte_ifcvf_vdpa = {
1323 .id_table = pci_id_ifcvf_map,
1325 .probe = ifcvf_pci_probe,
1326 .remove = ifcvf_pci_remove,
1329 RTE_PMD_REGISTER_PCI(net_ifcvf, rte_ifcvf_vdpa);
1330 RTE_PMD_REGISTER_PCI_TABLE(net_ifcvf, pci_id_ifcvf_map);
1331 RTE_PMD_REGISTER_KMOD_DEP(net_ifcvf, "* vfio-pci");