4 * Copyright Red Hat, Inc. 2010
9 * This work is licensed under the terms of the GNU GPL, version 2. See
10 * the COPYING file in the top-level directory.
12 * Contributions after 2012-01-13 are licensed under the terms of the
13 * GNU GPL, version 2 or (at your option) any later version.
16 #include <sys/ioctl.h>
19 #include "qemu/range.h"
20 #include <linux/vhost.h>
21 #include "exec/address-spaces.h"
23 static void vhost_dev_sync_region(struct vhost_dev *dev,
24 MemoryRegionSection *section,
25 uint64_t mfirst, uint64_t mlast,
26 uint64_t rfirst, uint64_t rlast)
28 uint64_t start = MAX(mfirst, rfirst);
29 uint64_t end = MIN(mlast, rlast);
30 vhost_log_chunk_t *from = dev->log + start / VHOST_LOG_CHUNK;
31 vhost_log_chunk_t *to = dev->log + end / VHOST_LOG_CHUNK + 1;
32 uint64_t addr = (start / VHOST_LOG_CHUNK) * VHOST_LOG_CHUNK;
37 assert(end / VHOST_LOG_CHUNK < dev->log_size);
38 assert(start / VHOST_LOG_CHUNK < dev->log_size);
40 for (;from < to; ++from) {
41 vhost_log_chunk_t log;
43 /* We first check with non-atomic: much cheaper,
44 * and we expect non-dirty to be the common case. */
46 addr += VHOST_LOG_CHUNK;
49 /* Data must be read atomically. We don't really
50 * need the barrier semantics of __sync
51 * builtins, but it's easier to use them than
53 log = __sync_fetch_and_and(from, 0);
54 while ((bit = sizeof(log) > sizeof(int) ?
55 ffsll(log) : ffs(log))) {
57 hwaddr section_offset;
60 page_addr = addr + bit * VHOST_LOG_PAGE;
61 section_offset = page_addr - section->offset_within_address_space;
62 mr_offset = section_offset + section->offset_within_region;
63 memory_region_set_dirty(section->mr, mr_offset, VHOST_LOG_PAGE);
64 log &= ~(0x1ull << bit);
66 addr += VHOST_LOG_CHUNK;
70 static int vhost_sync_dirty_bitmap(struct vhost_dev *dev,
71 MemoryRegionSection *section,
79 if (!dev->log_enabled || !dev->started) {
82 start_addr = section->offset_within_address_space;
83 end_addr = range_get_last(start_addr, section->size);
84 start_addr = MAX(first, start_addr);
85 end_addr = MIN(last, end_addr);
87 for (i = 0; i < dev->mem->nregions; ++i) {
88 struct vhost_memory_region *reg = dev->mem->regions + i;
89 vhost_dev_sync_region(dev, section, start_addr, end_addr,
91 range_get_last(reg->guest_phys_addr,
94 for (i = 0; i < dev->nvqs; ++i) {
95 struct vhost_virtqueue *vq = dev->vqs + i;
96 vhost_dev_sync_region(dev, section, start_addr, end_addr, vq->used_phys,
97 range_get_last(vq->used_phys, vq->used_size));
102 static void vhost_log_sync(MemoryListener *listener,
103 MemoryRegionSection *section)
105 struct vhost_dev *dev = container_of(listener, struct vhost_dev,
107 vhost_sync_dirty_bitmap(dev, section, 0x0, ~0x0ULL);
110 static void vhost_log_sync_range(struct vhost_dev *dev,
111 hwaddr first, hwaddr last)
114 /* FIXME: this is N^2 in number of sections */
115 for (i = 0; i < dev->n_mem_sections; ++i) {
116 MemoryRegionSection *section = &dev->mem_sections[i];
117 vhost_sync_dirty_bitmap(dev, section, first, last);
121 /* Assign/unassign. Keep an unsorted array of non-overlapping
122 * memory regions in dev->mem. */
123 static void vhost_dev_unassign_memory(struct vhost_dev *dev,
127 int from, to, n = dev->mem->nregions;
128 /* Track overlapping/split regions for sanity checking. */
129 int overlap_start = 0, overlap_end = 0, overlap_middle = 0, split = 0;
131 for (from = 0, to = 0; from < n; ++from, ++to) {
132 struct vhost_memory_region *reg = dev->mem->regions + to;
137 /* clone old region */
139 memcpy(reg, dev->mem->regions + from, sizeof *reg);
142 /* No overlap is simple */
143 if (!ranges_overlap(reg->guest_phys_addr, reg->memory_size,
148 /* Split only happens if supplied region
149 * is in the middle of an existing one. Thus it can not
150 * overlap with any other existing region. */
153 reglast = range_get_last(reg->guest_phys_addr, reg->memory_size);
154 memlast = range_get_last(start_addr, size);
156 /* Remove whole region */
157 if (start_addr <= reg->guest_phys_addr && memlast >= reglast) {
158 --dev->mem->nregions;
165 if (memlast >= reglast) {
166 reg->memory_size = start_addr - reg->guest_phys_addr;
167 assert(reg->memory_size);
168 assert(!overlap_end);
174 if (start_addr <= reg->guest_phys_addr) {
175 change = memlast + 1 - reg->guest_phys_addr;
176 reg->memory_size -= change;
177 reg->guest_phys_addr += change;
178 reg->userspace_addr += change;
179 assert(reg->memory_size);
180 assert(!overlap_start);
185 /* This only happens if supplied region
186 * is in the middle of an existing one. Thus it can not
187 * overlap with any other existing region. */
188 assert(!overlap_start);
189 assert(!overlap_end);
190 assert(!overlap_middle);
191 /* Split region: shrink first part, shift second part. */
192 memcpy(dev->mem->regions + n, reg, sizeof *reg);
193 reg->memory_size = start_addr - reg->guest_phys_addr;
194 assert(reg->memory_size);
195 change = memlast + 1 - reg->guest_phys_addr;
196 reg = dev->mem->regions + n;
197 reg->memory_size -= change;
198 assert(reg->memory_size);
199 reg->guest_phys_addr += change;
200 reg->userspace_addr += change;
201 /* Never add more than 1 region */
202 assert(dev->mem->nregions == n);
203 ++dev->mem->nregions;
208 /* Called after unassign, so no regions overlap the given range. */
209 static void vhost_dev_assign_memory(struct vhost_dev *dev,
215 struct vhost_memory_region *merged = NULL;
216 for (from = 0, to = 0; from < dev->mem->nregions; ++from, ++to) {
217 struct vhost_memory_region *reg = dev->mem->regions + to;
218 uint64_t prlast, urlast;
219 uint64_t pmlast, umlast;
222 /* clone old region */
224 memcpy(reg, dev->mem->regions + from, sizeof *reg);
226 prlast = range_get_last(reg->guest_phys_addr, reg->memory_size);
227 pmlast = range_get_last(start_addr, size);
228 urlast = range_get_last(reg->userspace_addr, reg->memory_size);
229 umlast = range_get_last(uaddr, size);
231 /* check for overlapping regions: should never happen. */
232 assert(prlast < start_addr || pmlast < reg->guest_phys_addr);
233 /* Not an adjacent or overlapping region - do not merge. */
234 if ((prlast + 1 != start_addr || urlast + 1 != uaddr) &&
235 (pmlast + 1 != reg->guest_phys_addr ||
236 umlast + 1 != reg->userspace_addr)) {
246 u = MIN(uaddr, reg->userspace_addr);
247 s = MIN(start_addr, reg->guest_phys_addr);
248 e = MAX(pmlast, prlast);
249 uaddr = merged->userspace_addr = u;
250 start_addr = merged->guest_phys_addr = s;
251 size = merged->memory_size = e - s + 1;
252 assert(merged->memory_size);
256 struct vhost_memory_region *reg = dev->mem->regions + to;
257 memset(reg, 0, sizeof *reg);
258 reg->memory_size = size;
259 assert(reg->memory_size);
260 reg->guest_phys_addr = start_addr;
261 reg->userspace_addr = uaddr;
264 assert(to <= dev->mem->nregions + 1);
265 dev->mem->nregions = to;
268 static uint64_t vhost_get_log_size(struct vhost_dev *dev)
270 uint64_t log_size = 0;
272 for (i = 0; i < dev->mem->nregions; ++i) {
273 struct vhost_memory_region *reg = dev->mem->regions + i;
274 uint64_t last = range_get_last(reg->guest_phys_addr,
276 log_size = MAX(log_size, last / VHOST_LOG_CHUNK + 1);
278 for (i = 0; i < dev->nvqs; ++i) {
279 struct vhost_virtqueue *vq = dev->vqs + i;
280 uint64_t last = vq->used_phys + vq->used_size - 1;
281 log_size = MAX(log_size, last / VHOST_LOG_CHUNK + 1);
286 static inline void vhost_dev_log_resize(struct vhost_dev* dev, uint64_t size)
288 vhost_log_chunk_t *log;
292 log = g_malloc0(size * sizeof *log);
293 log_base = (uint64_t)(unsigned long)log;
294 r = ioctl(dev->control, VHOST_SET_LOG_BASE, &log_base);
296 /* Sync only the range covered by the old log */
298 vhost_log_sync_range(dev, 0, dev->log_size * VHOST_LOG_CHUNK - 1);
304 dev->log_size = size;
307 static int vhost_verify_ring_mappings(struct vhost_dev *dev,
312 for (i = 0; i < dev->nvqs; ++i) {
313 struct vhost_virtqueue *vq = dev->vqs + i;
317 if (!ranges_overlap(start_addr, size, vq->ring_phys, vq->ring_size)) {
321 p = cpu_physical_memory_map(vq->ring_phys, &l, 1);
322 if (!p || l != vq->ring_size) {
323 fprintf(stderr, "Unable to map ring buffer for ring %d\n", i);
327 fprintf(stderr, "Ring buffer relocated for ring %d\n", i);
330 cpu_physical_memory_unmap(p, l, 0, 0);
335 static struct vhost_memory_region *vhost_dev_find_reg(struct vhost_dev *dev,
339 int i, n = dev->mem->nregions;
340 for (i = 0; i < n; ++i) {
341 struct vhost_memory_region *reg = dev->mem->regions + i;
342 if (ranges_overlap(reg->guest_phys_addr, reg->memory_size,
350 static bool vhost_dev_cmp_memory(struct vhost_dev *dev,
355 struct vhost_memory_region *reg = vhost_dev_find_reg(dev, start_addr, size);
363 reglast = range_get_last(reg->guest_phys_addr, reg->memory_size);
364 memlast = range_get_last(start_addr, size);
366 /* Need to extend region? */
367 if (start_addr < reg->guest_phys_addr || memlast > reglast) {
370 /* userspace_addr changed? */
371 return uaddr != reg->userspace_addr + start_addr - reg->guest_phys_addr;
374 static void vhost_set_memory(MemoryListener *listener,
375 MemoryRegionSection *section,
378 struct vhost_dev *dev = container_of(listener, struct vhost_dev,
380 hwaddr start_addr = section->offset_within_address_space;
381 ram_addr_t size = section->size;
382 bool log_dirty = memory_region_is_logging(section->mr);
383 int s = offsetof(struct vhost_memory, regions) +
384 (dev->mem->nregions + 1) * sizeof dev->mem->regions[0];
389 dev->mem = g_realloc(dev->mem, s);
397 /* Optimize no-change case. At least cirrus_vga does this a lot at this time. */
398 ram = memory_region_get_ram_ptr(section->mr) + section->offset_within_region;
400 if (!vhost_dev_cmp_memory(dev, start_addr, size, (uintptr_t)ram)) {
401 /* Region exists with same address. Nothing to do. */
405 if (!vhost_dev_find_reg(dev, start_addr, size)) {
406 /* Removing region that we don't access. Nothing to do. */
411 vhost_dev_unassign_memory(dev, start_addr, size);
413 /* Add given mapping, merging adjacent regions if any */
414 vhost_dev_assign_memory(dev, start_addr, size, (uintptr_t)ram);
416 /* Remove old mapping for this memory, if any. */
417 vhost_dev_unassign_memory(dev, start_addr, size);
425 r = vhost_verify_ring_mappings(dev, start_addr, size);
429 if (!dev->log_enabled) {
430 r = ioctl(dev->control, VHOST_SET_MEM_TABLE, dev->mem);
434 log_size = vhost_get_log_size(dev);
435 /* We allocate an extra 4K bytes to log,
436 * to reduce the * number of reallocations. */
437 #define VHOST_LOG_BUFFER (0x1000 / sizeof *dev->log)
438 /* To log more, must increase log size before table update. */
439 if (dev->log_size < log_size) {
440 vhost_dev_log_resize(dev, log_size + VHOST_LOG_BUFFER);
442 r = ioctl(dev->control, VHOST_SET_MEM_TABLE, dev->mem);
444 /* To log less, can only decrease log size after table update. */
445 if (dev->log_size > log_size + VHOST_LOG_BUFFER) {
446 vhost_dev_log_resize(dev, log_size);
450 static bool vhost_section(MemoryRegionSection *section)
452 return memory_region_is_ram(section->mr);
455 static void vhost_begin(MemoryListener *listener)
459 static void vhost_commit(MemoryListener *listener)
463 static void vhost_region_add(MemoryListener *listener,
464 MemoryRegionSection *section)
466 struct vhost_dev *dev = container_of(listener, struct vhost_dev,
469 if (!vhost_section(section)) {
473 ++dev->n_mem_sections;
474 dev->mem_sections = g_renew(MemoryRegionSection, dev->mem_sections,
475 dev->n_mem_sections);
476 dev->mem_sections[dev->n_mem_sections - 1] = *section;
477 vhost_set_memory(listener, section, true);
480 static void vhost_region_del(MemoryListener *listener,
481 MemoryRegionSection *section)
483 struct vhost_dev *dev = container_of(listener, struct vhost_dev,
487 if (!vhost_section(section)) {
491 vhost_set_memory(listener, section, false);
492 for (i = 0; i < dev->n_mem_sections; ++i) {
493 if (dev->mem_sections[i].offset_within_address_space
494 == section->offset_within_address_space) {
495 --dev->n_mem_sections;
496 memmove(&dev->mem_sections[i], &dev->mem_sections[i+1],
497 (dev->n_mem_sections - i) * sizeof(*dev->mem_sections));
503 static void vhost_region_nop(MemoryListener *listener,
504 MemoryRegionSection *section)
508 static int vhost_virtqueue_set_addr(struct vhost_dev *dev,
509 struct vhost_virtqueue *vq,
510 unsigned idx, bool enable_log)
512 struct vhost_vring_addr addr = {
514 .desc_user_addr = (uint64_t)(unsigned long)vq->desc,
515 .avail_user_addr = (uint64_t)(unsigned long)vq->avail,
516 .used_user_addr = (uint64_t)(unsigned long)vq->used,
517 .log_guest_addr = vq->used_phys,
518 .flags = enable_log ? (1 << VHOST_VRING_F_LOG) : 0,
520 int r = ioctl(dev->control, VHOST_SET_VRING_ADDR, &addr);
527 static int vhost_dev_set_features(struct vhost_dev *dev, bool enable_log)
529 uint64_t features = dev->acked_features;
532 features |= 0x1 << VHOST_F_LOG_ALL;
534 r = ioctl(dev->control, VHOST_SET_FEATURES, &features);
535 return r < 0 ? -errno : 0;
538 static int vhost_dev_set_log(struct vhost_dev *dev, bool enable_log)
541 r = vhost_dev_set_features(dev, enable_log);
545 for (i = 0; i < dev->nvqs; ++i) {
546 r = vhost_virtqueue_set_addr(dev, dev->vqs + i, i,
554 for (; i >= 0; --i) {
555 t = vhost_virtqueue_set_addr(dev, dev->vqs + i, i,
559 t = vhost_dev_set_features(dev, dev->log_enabled);
565 static int vhost_migration_log(MemoryListener *listener, int enable)
567 struct vhost_dev *dev = container_of(listener, struct vhost_dev,
570 if (!!enable == dev->log_enabled) {
574 dev->log_enabled = enable;
578 r = vhost_dev_set_log(dev, false);
588 vhost_dev_log_resize(dev, vhost_get_log_size(dev));
589 r = vhost_dev_set_log(dev, true);
594 dev->log_enabled = enable;
598 static void vhost_log_global_start(MemoryListener *listener)
602 r = vhost_migration_log(listener, true);
608 static void vhost_log_global_stop(MemoryListener *listener)
612 r = vhost_migration_log(listener, false);
618 static void vhost_log_start(MemoryListener *listener,
619 MemoryRegionSection *section)
621 /* FIXME: implement */
624 static void vhost_log_stop(MemoryListener *listener,
625 MemoryRegionSection *section)
627 /* FIXME: implement */
630 static int vhost_virtqueue_start(struct vhost_dev *dev,
631 struct VirtIODevice *vdev,
632 struct vhost_virtqueue *vq,
637 int vhost_vq_index = idx - dev->vq_index;
638 struct vhost_vring_file file = {
639 .index = vhost_vq_index
641 struct vhost_vring_state state = {
642 .index = vhost_vq_index
644 struct VirtQueue *vvq = virtio_get_queue(vdev, idx);
646 assert(idx >= dev->vq_index && idx < dev->vq_index + dev->nvqs);
648 vq->num = state.num = virtio_queue_get_num(vdev, idx);
649 r = ioctl(dev->control, VHOST_SET_VRING_NUM, &state);
654 state.num = virtio_queue_get_last_avail_idx(vdev, idx);
655 r = ioctl(dev->control, VHOST_SET_VRING_BASE, &state);
660 s = l = virtio_queue_get_desc_size(vdev, idx);
661 a = virtio_queue_get_desc_addr(vdev, idx);
662 vq->desc = cpu_physical_memory_map(a, &l, 0);
663 if (!vq->desc || l != s) {
665 goto fail_alloc_desc;
667 s = l = virtio_queue_get_avail_size(vdev, idx);
668 a = virtio_queue_get_avail_addr(vdev, idx);
669 vq->avail = cpu_physical_memory_map(a, &l, 0);
670 if (!vq->avail || l != s) {
672 goto fail_alloc_avail;
674 vq->used_size = s = l = virtio_queue_get_used_size(vdev, idx);
675 vq->used_phys = a = virtio_queue_get_used_addr(vdev, idx);
676 vq->used = cpu_physical_memory_map(a, &l, 1);
677 if (!vq->used || l != s) {
679 goto fail_alloc_used;
682 vq->ring_size = s = l = virtio_queue_get_ring_size(vdev, idx);
683 vq->ring_phys = a = virtio_queue_get_ring_addr(vdev, idx);
684 vq->ring = cpu_physical_memory_map(a, &l, 1);
685 if (!vq->ring || l != s) {
687 goto fail_alloc_ring;
690 r = vhost_virtqueue_set_addr(dev, vq, vhost_vq_index, dev->log_enabled);
696 file.fd = event_notifier_get_fd(virtio_queue_get_host_notifier(vvq));
697 r = ioctl(dev->control, VHOST_SET_VRING_KICK, &file);
703 /* Clear and discard previous events if any. */
704 event_notifier_test_and_clear(&vq->masked_notifier);
710 cpu_physical_memory_unmap(vq->ring, virtio_queue_get_ring_size(vdev, idx),
713 cpu_physical_memory_unmap(vq->used, virtio_queue_get_used_size(vdev, idx),
716 cpu_physical_memory_unmap(vq->avail, virtio_queue_get_avail_size(vdev, idx),
719 cpu_physical_memory_unmap(vq->desc, virtio_queue_get_desc_size(vdev, idx),
725 static void vhost_virtqueue_stop(struct vhost_dev *dev,
726 struct VirtIODevice *vdev,
727 struct vhost_virtqueue *vq,
730 struct vhost_vring_state state = {
731 .index = idx - dev->vq_index
734 assert(idx >= dev->vq_index && idx < dev->vq_index + dev->nvqs);
735 r = ioctl(dev->control, VHOST_GET_VRING_BASE, &state);
737 fprintf(stderr, "vhost VQ %d ring restore failed: %d\n", idx, r);
740 virtio_queue_set_last_avail_idx(vdev, idx, state.num);
742 cpu_physical_memory_unmap(vq->ring, virtio_queue_get_ring_size(vdev, idx),
743 0, virtio_queue_get_ring_size(vdev, idx));
744 cpu_physical_memory_unmap(vq->used, virtio_queue_get_used_size(vdev, idx),
745 1, virtio_queue_get_used_size(vdev, idx));
746 cpu_physical_memory_unmap(vq->avail, virtio_queue_get_avail_size(vdev, idx),
747 0, virtio_queue_get_avail_size(vdev, idx));
748 cpu_physical_memory_unmap(vq->desc, virtio_queue_get_desc_size(vdev, idx),
749 0, virtio_queue_get_desc_size(vdev, idx));
752 static void vhost_eventfd_add(MemoryListener *listener,
753 MemoryRegionSection *section,
754 bool match_data, uint64_t data, EventNotifier *e)
758 static void vhost_eventfd_del(MemoryListener *listener,
759 MemoryRegionSection *section,
760 bool match_data, uint64_t data, EventNotifier *e)
764 static int vhost_virtqueue_init(struct vhost_dev *dev,
765 struct vhost_virtqueue *vq, int n)
767 struct vhost_vring_file file = {
770 int r = event_notifier_init(&vq->masked_notifier, 0);
775 file.fd = event_notifier_get_fd(&vq->masked_notifier);
776 r = ioctl(dev->control, VHOST_SET_VRING_CALL, &file);
783 event_notifier_cleanup(&vq->masked_notifier);
787 static void vhost_virtqueue_cleanup(struct vhost_virtqueue *vq)
789 event_notifier_cleanup(&vq->masked_notifier);
792 int vhost_dev_init(struct vhost_dev *hdev, int devfd, const char *devpath,
798 hdev->control = devfd;
800 hdev->control = open(devpath, O_RDWR);
801 if (hdev->control < 0) {
805 r = ioctl(hdev->control, VHOST_SET_OWNER, NULL);
810 r = ioctl(hdev->control, VHOST_GET_FEATURES, &features);
815 for (i = 0; i < hdev->nvqs; ++i) {
816 r = vhost_virtqueue_init(hdev, hdev->vqs + i, i);
821 hdev->features = features;
823 hdev->memory_listener = (MemoryListener) {
824 .begin = vhost_begin,
825 .commit = vhost_commit,
826 .region_add = vhost_region_add,
827 .region_del = vhost_region_del,
828 .region_nop = vhost_region_nop,
829 .log_start = vhost_log_start,
830 .log_stop = vhost_log_stop,
831 .log_sync = vhost_log_sync,
832 .log_global_start = vhost_log_global_start,
833 .log_global_stop = vhost_log_global_stop,
834 .eventfd_add = vhost_eventfd_add,
835 .eventfd_del = vhost_eventfd_del,
838 hdev->mem = g_malloc0(offsetof(struct vhost_memory, regions));
839 hdev->n_mem_sections = 0;
840 hdev->mem_sections = NULL;
843 hdev->log_enabled = false;
844 hdev->started = false;
845 memory_listener_register(&hdev->memory_listener, &address_space_memory);
850 vhost_virtqueue_cleanup(hdev->vqs + i);
854 close(hdev->control);
858 void vhost_dev_cleanup(struct vhost_dev *hdev)
861 for (i = 0; i < hdev->nvqs; ++i) {
862 vhost_virtqueue_cleanup(hdev->vqs + i);
864 memory_listener_unregister(&hdev->memory_listener);
866 g_free(hdev->mem_sections);
867 close(hdev->control);
870 bool vhost_dev_query(struct vhost_dev *hdev, VirtIODevice *vdev)
872 return !vdev->binding->query_guest_notifiers ||
873 vdev->binding->query_guest_notifiers(vdev->binding_opaque) ||
877 /* Stop processing guest IO notifications in qemu.
878 * Start processing them in vhost in kernel.
880 int vhost_dev_enable_notifiers(struct vhost_dev *hdev, VirtIODevice *vdev)
883 if (!vdev->binding->set_host_notifier) {
884 fprintf(stderr, "binding does not support host notifiers\n");
889 for (i = 0; i < hdev->nvqs; ++i) {
890 r = vdev->binding->set_host_notifier(vdev->binding_opaque,
894 fprintf(stderr, "vhost VQ %d notifier binding failed: %d\n", i, -r);
902 r = vdev->binding->set_host_notifier(vdev->binding_opaque,
906 fprintf(stderr, "vhost VQ %d notifier cleanup error: %d\n", i, -r);
915 /* Stop processing guest IO notifications in vhost.
916 * Start processing them in qemu.
917 * This might actually run the qemu handlers right away,
918 * so virtio in qemu must be completely setup when this is called.
920 void vhost_dev_disable_notifiers(struct vhost_dev *hdev, VirtIODevice *vdev)
924 for (i = 0; i < hdev->nvqs; ++i) {
925 r = vdev->binding->set_host_notifier(vdev->binding_opaque,
929 fprintf(stderr, "vhost VQ %d notifier cleanup failed: %d\n", i, -r);
936 /* Test and clear event pending status.
937 * Should be called after unmask to avoid losing events.
939 bool vhost_virtqueue_pending(struct vhost_dev *hdev, int n)
941 struct vhost_virtqueue *vq = hdev->vqs + n - hdev->vq_index;
942 assert(hdev->started);
943 assert(n >= hdev->vq_index && n < hdev->vq_index + hdev->nvqs);
944 return event_notifier_test_and_clear(&vq->masked_notifier);
947 /* Mask/unmask events from this vq. */
948 void vhost_virtqueue_mask(struct vhost_dev *hdev, VirtIODevice *vdev, int n,
951 struct VirtQueue *vvq = virtio_get_queue(vdev, n);
952 int r, index = n - hdev->vq_index;
954 assert(hdev->started);
955 assert(n >= hdev->vq_index && n < hdev->vq_index + hdev->nvqs);
957 struct vhost_vring_file file = {
961 file.fd = event_notifier_get_fd(&hdev->vqs[index].masked_notifier);
963 file.fd = event_notifier_get_fd(virtio_queue_get_guest_notifier(vvq));
965 r = ioctl(hdev->control, VHOST_SET_VRING_CALL, &file);
969 /* Host notifiers must be enabled at this point. */
970 int vhost_dev_start(struct vhost_dev *hdev, VirtIODevice *vdev)
974 hdev->started = true;
976 r = vhost_dev_set_features(hdev, hdev->log_enabled);
980 r = ioctl(hdev->control, VHOST_SET_MEM_TABLE, hdev->mem);
985 for (i = 0; i < hdev->nvqs; ++i) {
986 r = vhost_virtqueue_start(hdev,
995 if (hdev->log_enabled) {
996 hdev->log_size = vhost_get_log_size(hdev);
997 hdev->log = hdev->log_size ?
998 g_malloc0(hdev->log_size * sizeof *hdev->log) : NULL;
999 r = ioctl(hdev->control, VHOST_SET_LOG_BASE,
1000 (uint64_t)(unsigned long)hdev->log);
1011 vhost_virtqueue_stop(hdev,
1014 hdev->vq_index + i);
1020 hdev->started = false;
1024 /* Host notifiers must be enabled at this point. */
1025 void vhost_dev_stop(struct vhost_dev *hdev, VirtIODevice *vdev)
1029 for (i = 0; i < hdev->nvqs; ++i) {
1030 vhost_virtqueue_stop(hdev,
1033 hdev->vq_index + i);
1035 vhost_log_sync_range(hdev, 0, ~0x0ull);
1037 hdev->started = false;