4 * Copyright IBM, Corp. 2007
9 * This work is licensed under the terms of the GNU GPL, version 2. See
10 * the COPYING file in the top-level directory.
17 #include "exec/address-spaces.h"
18 #include "qemu/error-report.h"
19 #include "hw/virtio/virtio.h"
20 #include "qemu/atomic.h"
21 #include "hw/virtio/virtio-bus.h"
22 #include "migration/migration.h"
23 #include "hw/virtio/virtio-access.h"
26 * The alignment to use between consumer and producer parts of vring.
27 * x86 pagesize again. This is the default, used by transports like PCI
28 * which don't provide a means for the guest to tell the host the alignment.
30 #define VIRTIO_PCI_VRING_ALIGN 4096
32 typedef struct VRingDesc
40 typedef struct VRingAvail
47 typedef struct VRingUsedElem
53 typedef struct VRingUsed
57 VRingUsedElem ring[0];
73 uint16_t last_avail_idx;
74 /* Last used index value we have signalled on */
75 uint16_t signalled_used;
77 /* Last used index value we have signalled on */
78 bool signalled_used_valid;
80 /* Notification enabled? */
88 void (*handle_output)(VirtIODevice *vdev, VirtQueue *vq);
90 EventNotifier guest_notifier;
91 EventNotifier host_notifier;
92 QLIST_ENTRY(VirtQueue) node;
95 /* virt queue functions */
96 static void virtqueue_init(VirtQueue *vq)
101 vq->vring.avail = pa + vq->vring.num * sizeof(VRingDesc);
102 vq->vring.used = vring_align(vq->vring.avail +
103 offsetof(VRingAvail, ring[vq->vring.num]),
107 static inline uint64_t vring_desc_addr(VirtIODevice *vdev, hwaddr desc_pa,
111 pa = desc_pa + sizeof(VRingDesc) * i + offsetof(VRingDesc, addr);
112 return virtio_ldq_phys(vdev, pa);
115 static inline uint32_t vring_desc_len(VirtIODevice *vdev, hwaddr desc_pa, int i)
118 pa = desc_pa + sizeof(VRingDesc) * i + offsetof(VRingDesc, len);
119 return virtio_ldl_phys(vdev, pa);
122 static inline uint16_t vring_desc_flags(VirtIODevice *vdev, hwaddr desc_pa,
126 pa = desc_pa + sizeof(VRingDesc) * i + offsetof(VRingDesc, flags);
127 return virtio_lduw_phys(vdev, pa);
130 static inline uint16_t vring_desc_next(VirtIODevice *vdev, hwaddr desc_pa,
134 pa = desc_pa + sizeof(VRingDesc) * i + offsetof(VRingDesc, next);
135 return virtio_lduw_phys(vdev, pa);
138 static inline uint16_t vring_avail_flags(VirtQueue *vq)
141 pa = vq->vring.avail + offsetof(VRingAvail, flags);
142 return virtio_lduw_phys(vq->vdev, pa);
145 static inline uint16_t vring_avail_idx(VirtQueue *vq)
148 pa = vq->vring.avail + offsetof(VRingAvail, idx);
149 return virtio_lduw_phys(vq->vdev, pa);
152 static inline uint16_t vring_avail_ring(VirtQueue *vq, int i)
155 pa = vq->vring.avail + offsetof(VRingAvail, ring[i]);
156 return virtio_lduw_phys(vq->vdev, pa);
159 static inline uint16_t vring_get_used_event(VirtQueue *vq)
161 return vring_avail_ring(vq, vq->vring.num);
164 static inline void vring_used_ring_id(VirtQueue *vq, int i, uint32_t val)
167 pa = vq->vring.used + offsetof(VRingUsed, ring[i].id);
168 virtio_stl_phys(vq->vdev, pa, val);
171 static inline void vring_used_ring_len(VirtQueue *vq, int i, uint32_t val)
174 pa = vq->vring.used + offsetof(VRingUsed, ring[i].len);
175 virtio_stl_phys(vq->vdev, pa, val);
178 static uint16_t vring_used_idx(VirtQueue *vq)
181 pa = vq->vring.used + offsetof(VRingUsed, idx);
182 return virtio_lduw_phys(vq->vdev, pa);
185 static inline void vring_used_idx_set(VirtQueue *vq, uint16_t val)
188 pa = vq->vring.used + offsetof(VRingUsed, idx);
189 virtio_stw_phys(vq->vdev, pa, val);
192 static inline void vring_used_flags_set_bit(VirtQueue *vq, int mask)
194 VirtIODevice *vdev = vq->vdev;
196 pa = vq->vring.used + offsetof(VRingUsed, flags);
197 virtio_stw_phys(vdev, pa, virtio_lduw_phys(vdev, pa) | mask);
200 static inline void vring_used_flags_unset_bit(VirtQueue *vq, int mask)
202 VirtIODevice *vdev = vq->vdev;
204 pa = vq->vring.used + offsetof(VRingUsed, flags);
205 virtio_stw_phys(vdev, pa, virtio_lduw_phys(vdev, pa) & ~mask);
208 static inline void vring_set_avail_event(VirtQueue *vq, uint16_t val)
211 if (!vq->notification) {
214 pa = vq->vring.used + offsetof(VRingUsed, ring[vq->vring.num]);
215 virtio_stw_phys(vq->vdev, pa, val);
218 void virtio_queue_set_notification(VirtQueue *vq, int enable)
220 vq->notification = enable;
221 if (virtio_has_feature(vq->vdev, VIRTIO_RING_F_EVENT_IDX)) {
222 vring_set_avail_event(vq, vring_avail_idx(vq));
224 vring_used_flags_unset_bit(vq, VRING_USED_F_NO_NOTIFY);
226 vring_used_flags_set_bit(vq, VRING_USED_F_NO_NOTIFY);
229 /* Expose avail event/used flags before caller checks the avail idx. */
234 int virtio_queue_ready(VirtQueue *vq)
236 return vq->vring.avail != 0;
239 int virtio_queue_empty(VirtQueue *vq)
241 return vring_avail_idx(vq) == vq->last_avail_idx;
244 void virtqueue_fill(VirtQueue *vq, const VirtQueueElement *elem,
245 unsigned int len, unsigned int idx)
250 trace_virtqueue_fill(vq, elem, len, idx);
253 for (i = 0; i < elem->in_num; i++) {
254 size_t size = MIN(len - offset, elem->in_sg[i].iov_len);
256 cpu_physical_memory_unmap(elem->in_sg[i].iov_base,
257 elem->in_sg[i].iov_len,
263 for (i = 0; i < elem->out_num; i++)
264 cpu_physical_memory_unmap(elem->out_sg[i].iov_base,
265 elem->out_sg[i].iov_len,
266 0, elem->out_sg[i].iov_len);
268 idx = (idx + vring_used_idx(vq)) % vq->vring.num;
270 /* Get a pointer to the next entry in the used ring. */
271 vring_used_ring_id(vq, idx, elem->index);
272 vring_used_ring_len(vq, idx, len);
275 void virtqueue_flush(VirtQueue *vq, unsigned int count)
278 /* Make sure buffer is written before we update index. */
280 trace_virtqueue_flush(vq, count);
281 old = vring_used_idx(vq);
283 vring_used_idx_set(vq, new);
285 if (unlikely((int16_t)(new - vq->signalled_used) < (uint16_t)(new - old)))
286 vq->signalled_used_valid = false;
289 void virtqueue_push(VirtQueue *vq, const VirtQueueElement *elem,
292 virtqueue_fill(vq, elem, len, 0);
293 virtqueue_flush(vq, 1);
296 static int virtqueue_num_heads(VirtQueue *vq, unsigned int idx)
298 uint16_t num_heads = vring_avail_idx(vq) - idx;
300 /* Check it isn't doing very strange things with descriptor numbers. */
301 if (num_heads > vq->vring.num) {
302 error_report("Guest moved used index from %u to %u",
303 idx, vring_avail_idx(vq));
306 /* On success, callers read a descriptor at vq->last_avail_idx.
307 * Make sure descriptor read does not bypass avail index read. */
315 static unsigned int virtqueue_get_head(VirtQueue *vq, unsigned int idx)
319 /* Grab the next descriptor number they're advertising, and increment
320 * the index we've seen. */
321 head = vring_avail_ring(vq, idx % vq->vring.num);
323 /* If their number is silly, that's a fatal mistake. */
324 if (head >= vq->vring.num) {
325 error_report("Guest says index %u is available", head);
332 static unsigned virtqueue_next_desc(VirtIODevice *vdev, hwaddr desc_pa,
333 unsigned int i, unsigned int max)
337 /* If this descriptor says it doesn't chain, we're done. */
338 if (!(vring_desc_flags(vdev, desc_pa, i) & VRING_DESC_F_NEXT)) {
342 /* Check they're not leading us off end of descriptors. */
343 next = vring_desc_next(vdev, desc_pa, i);
344 /* Make sure compiler knows to grab that: we don't want it changing! */
348 error_report("Desc next is %u", next);
355 void virtqueue_get_avail_bytes(VirtQueue *vq, unsigned int *in_bytes,
356 unsigned int *out_bytes,
357 unsigned max_in_bytes, unsigned max_out_bytes)
360 unsigned int total_bufs, in_total, out_total;
362 idx = vq->last_avail_idx;
364 total_bufs = in_total = out_total = 0;
365 while (virtqueue_num_heads(vq, idx)) {
366 VirtIODevice *vdev = vq->vdev;
367 unsigned int max, num_bufs, indirect = 0;
372 num_bufs = total_bufs;
373 i = virtqueue_get_head(vq, idx++);
374 desc_pa = vq->vring.desc;
376 if (vring_desc_flags(vdev, desc_pa, i) & VRING_DESC_F_INDIRECT) {
377 if (vring_desc_len(vdev, desc_pa, i) % sizeof(VRingDesc)) {
378 error_report("Invalid size for indirect buffer table");
382 /* If we've got too many, that implies a descriptor loop. */
383 if (num_bufs >= max) {
384 error_report("Looped descriptor");
388 /* loop over the indirect descriptor table */
390 max = vring_desc_len(vdev, desc_pa, i) / sizeof(VRingDesc);
391 desc_pa = vring_desc_addr(vdev, desc_pa, i);
396 /* If we've got too many, that implies a descriptor loop. */
397 if (++num_bufs > max) {
398 error_report("Looped descriptor");
402 if (vring_desc_flags(vdev, desc_pa, i) & VRING_DESC_F_WRITE) {
403 in_total += vring_desc_len(vdev, desc_pa, i);
405 out_total += vring_desc_len(vdev, desc_pa, i);
407 if (in_total >= max_in_bytes && out_total >= max_out_bytes) {
410 } while ((i = virtqueue_next_desc(vdev, desc_pa, i, max)) != max);
413 total_bufs = num_bufs;
419 *in_bytes = in_total;
422 *out_bytes = out_total;
426 int virtqueue_avail_bytes(VirtQueue *vq, unsigned int in_bytes,
427 unsigned int out_bytes)
429 unsigned int in_total, out_total;
431 virtqueue_get_avail_bytes(vq, &in_total, &out_total, in_bytes, out_bytes);
432 return in_bytes <= in_total && out_bytes <= out_total;
435 void virtqueue_map_sg(struct iovec *sg, hwaddr *addr,
436 size_t num_sg, int is_write)
441 if (num_sg > VIRTQUEUE_MAX_SIZE) {
442 error_report("virtio: map attempt out of bounds: %zd > %d",
443 num_sg, VIRTQUEUE_MAX_SIZE);
447 for (i = 0; i < num_sg; i++) {
449 sg[i].iov_base = cpu_physical_memory_map(addr[i], &len, is_write);
450 if (sg[i].iov_base == NULL || len != sg[i].iov_len) {
451 error_report("virtio: error trying to map MMIO memory");
457 int virtqueue_pop(VirtQueue *vq, VirtQueueElement *elem)
459 unsigned int i, head, max;
460 hwaddr desc_pa = vq->vring.desc;
461 VirtIODevice *vdev = vq->vdev;
463 if (!virtqueue_num_heads(vq, vq->last_avail_idx))
466 /* When we start there are none of either input nor output. */
467 elem->out_num = elem->in_num = 0;
471 i = head = virtqueue_get_head(vq, vq->last_avail_idx++);
472 if (virtio_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX)) {
473 vring_set_avail_event(vq, vq->last_avail_idx);
476 if (vring_desc_flags(vdev, desc_pa, i) & VRING_DESC_F_INDIRECT) {
477 if (vring_desc_len(vdev, desc_pa, i) % sizeof(VRingDesc)) {
478 error_report("Invalid size for indirect buffer table");
482 /* loop over the indirect descriptor table */
483 max = vring_desc_len(vdev, desc_pa, i) / sizeof(VRingDesc);
484 desc_pa = vring_desc_addr(vdev, desc_pa, i);
488 /* Collect all the descriptors */
492 if (vring_desc_flags(vdev, desc_pa, i) & VRING_DESC_F_WRITE) {
493 if (elem->in_num >= ARRAY_SIZE(elem->in_sg)) {
494 error_report("Too many write descriptors in indirect table");
497 elem->in_addr[elem->in_num] = vring_desc_addr(vdev, desc_pa, i);
498 sg = &elem->in_sg[elem->in_num++];
500 if (elem->out_num >= ARRAY_SIZE(elem->out_sg)) {
501 error_report("Too many read descriptors in indirect table");
504 elem->out_addr[elem->out_num] = vring_desc_addr(vdev, desc_pa, i);
505 sg = &elem->out_sg[elem->out_num++];
508 sg->iov_len = vring_desc_len(vdev, desc_pa, i);
510 /* If we've got too many, that implies a descriptor loop. */
511 if ((elem->in_num + elem->out_num) > max) {
512 error_report("Looped descriptor");
515 } while ((i = virtqueue_next_desc(vdev, desc_pa, i, max)) != max);
517 /* Now map what we have collected */
518 virtqueue_map_sg(elem->in_sg, elem->in_addr, elem->in_num, 1);
519 virtqueue_map_sg(elem->out_sg, elem->out_addr, elem->out_num, 0);
525 trace_virtqueue_pop(vq, elem, elem->in_num, elem->out_num);
526 return elem->in_num + elem->out_num;
530 static void virtio_notify_vector(VirtIODevice *vdev, uint16_t vector)
532 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
533 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
536 k->notify(qbus->parent, vector);
540 void virtio_update_irq(VirtIODevice *vdev)
542 virtio_notify_vector(vdev, VIRTIO_NO_VECTOR);
545 void virtio_set_status(VirtIODevice *vdev, uint8_t val)
547 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
548 trace_virtio_set_status(vdev, val);
551 k->set_status(vdev, val);
556 bool target_words_bigendian(void);
557 static enum virtio_device_endian virtio_default_endian(void)
559 if (target_words_bigendian()) {
560 return VIRTIO_DEVICE_ENDIAN_BIG;
562 return VIRTIO_DEVICE_ENDIAN_LITTLE;
566 static enum virtio_device_endian virtio_current_cpu_endian(void)
568 CPUClass *cc = CPU_GET_CLASS(current_cpu);
570 if (cc->virtio_is_big_endian(current_cpu)) {
571 return VIRTIO_DEVICE_ENDIAN_BIG;
573 return VIRTIO_DEVICE_ENDIAN_LITTLE;
577 void virtio_reset(void *opaque)
579 VirtIODevice *vdev = opaque;
580 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
583 virtio_set_status(vdev, 0);
585 /* Guest initiated reset */
586 vdev->device_endian = virtio_current_cpu_endian();
589 vdev->device_endian = virtio_default_endian();
596 vdev->guest_features = 0;
600 vdev->config_vector = VIRTIO_NO_VECTOR;
601 virtio_notify_vector(vdev, vdev->config_vector);
603 for(i = 0; i < VIRTIO_PCI_QUEUE_MAX; i++) {
604 vdev->vq[i].vring.desc = 0;
605 vdev->vq[i].vring.avail = 0;
606 vdev->vq[i].vring.used = 0;
607 vdev->vq[i].last_avail_idx = 0;
609 virtio_queue_set_vector(vdev, i, VIRTIO_NO_VECTOR);
610 vdev->vq[i].signalled_used = 0;
611 vdev->vq[i].signalled_used_valid = false;
612 vdev->vq[i].notification = true;
616 uint32_t virtio_config_readb(VirtIODevice *vdev, uint32_t addr)
618 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
621 if (addr + sizeof(val) > vdev->config_len) {
625 k->get_config(vdev, vdev->config);
627 val = ldub_p(vdev->config + addr);
631 uint32_t virtio_config_readw(VirtIODevice *vdev, uint32_t addr)
633 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
636 if (addr + sizeof(val) > vdev->config_len) {
640 k->get_config(vdev, vdev->config);
642 val = lduw_p(vdev->config + addr);
646 uint32_t virtio_config_readl(VirtIODevice *vdev, uint32_t addr)
648 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
651 if (addr + sizeof(val) > vdev->config_len) {
655 k->get_config(vdev, vdev->config);
657 val = ldl_p(vdev->config + addr);
661 void virtio_config_writeb(VirtIODevice *vdev, uint32_t addr, uint32_t data)
663 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
666 if (addr + sizeof(val) > vdev->config_len) {
670 stb_p(vdev->config + addr, val);
673 k->set_config(vdev, vdev->config);
677 void virtio_config_writew(VirtIODevice *vdev, uint32_t addr, uint32_t data)
679 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
682 if (addr + sizeof(val) > vdev->config_len) {
686 stw_p(vdev->config + addr, val);
689 k->set_config(vdev, vdev->config);
693 void virtio_config_writel(VirtIODevice *vdev, uint32_t addr, uint32_t data)
695 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
698 if (addr + sizeof(val) > vdev->config_len) {
702 stl_p(vdev->config + addr, val);
705 k->set_config(vdev, vdev->config);
709 void virtio_queue_set_addr(VirtIODevice *vdev, int n, hwaddr addr)
711 vdev->vq[n].pa = addr;
712 virtqueue_init(&vdev->vq[n]);
715 hwaddr virtio_queue_get_addr(VirtIODevice *vdev, int n)
717 return vdev->vq[n].pa;
720 void virtio_queue_set_num(VirtIODevice *vdev, int n, int num)
722 /* Don't allow guest to flip queue between existent and
723 * nonexistent states, or to set it to an invalid size.
725 if (!!num != !!vdev->vq[n].vring.num ||
726 num > VIRTQUEUE_MAX_SIZE ||
730 vdev->vq[n].vring.num = num;
731 virtqueue_init(&vdev->vq[n]);
734 VirtQueue *virtio_vector_first_queue(VirtIODevice *vdev, uint16_t vector)
736 return QLIST_FIRST(&vdev->vector_queues[vector]);
739 VirtQueue *virtio_vector_next_queue(VirtQueue *vq)
741 return QLIST_NEXT(vq, node);
744 int virtio_queue_get_num(VirtIODevice *vdev, int n)
746 return vdev->vq[n].vring.num;
749 int virtio_queue_get_id(VirtQueue *vq)
751 VirtIODevice *vdev = vq->vdev;
752 assert(vq >= &vdev->vq[0] && vq < &vdev->vq[VIRTIO_PCI_QUEUE_MAX]);
753 return vq - &vdev->vq[0];
756 void virtio_queue_set_align(VirtIODevice *vdev, int n, int align)
758 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
759 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
761 /* Check that the transport told us it was going to do this
762 * (so a buggy transport will immediately assert rather than
763 * silently failing to migrate this state)
765 assert(k->has_variable_vring_alignment);
767 vdev->vq[n].vring.align = align;
768 virtqueue_init(&vdev->vq[n]);
771 void virtio_queue_notify_vq(VirtQueue *vq)
773 if (vq->vring.desc && vq->handle_output) {
774 VirtIODevice *vdev = vq->vdev;
776 trace_virtio_queue_notify(vdev, vq - vdev->vq, vq);
777 vq->handle_output(vdev, vq);
781 void virtio_queue_notify(VirtIODevice *vdev, int n)
783 virtio_queue_notify_vq(&vdev->vq[n]);
786 uint16_t virtio_queue_vector(VirtIODevice *vdev, int n)
788 return n < VIRTIO_PCI_QUEUE_MAX ? vdev->vq[n].vector :
792 void virtio_queue_set_vector(VirtIODevice *vdev, int n, uint16_t vector)
794 VirtQueue *vq = &vdev->vq[n];
796 if (n < VIRTIO_PCI_QUEUE_MAX) {
797 if (vdev->vector_queues &&
798 vdev->vq[n].vector != VIRTIO_NO_VECTOR) {
799 QLIST_REMOVE(vq, node);
801 vdev->vq[n].vector = vector;
802 if (vdev->vector_queues &&
803 vector != VIRTIO_NO_VECTOR) {
804 QLIST_INSERT_HEAD(&vdev->vector_queues[vector], vq, node);
809 VirtQueue *virtio_add_queue(VirtIODevice *vdev, int queue_size,
810 void (*handle_output)(VirtIODevice *, VirtQueue *))
814 for (i = 0; i < VIRTIO_PCI_QUEUE_MAX; i++) {
815 if (vdev->vq[i].vring.num == 0)
819 if (i == VIRTIO_PCI_QUEUE_MAX || queue_size > VIRTQUEUE_MAX_SIZE)
822 vdev->vq[i].vring.num = queue_size;
823 vdev->vq[i].vring.align = VIRTIO_PCI_VRING_ALIGN;
824 vdev->vq[i].handle_output = handle_output;
829 void virtio_del_queue(VirtIODevice *vdev, int n)
831 if (n < 0 || n >= VIRTIO_PCI_QUEUE_MAX) {
835 vdev->vq[n].vring.num = 0;
838 void virtio_irq(VirtQueue *vq)
840 trace_virtio_irq(vq);
841 vq->vdev->isr |= 0x01;
842 virtio_notify_vector(vq->vdev, vq->vector);
845 static bool vring_notify(VirtIODevice *vdev, VirtQueue *vq)
849 /* We need to expose used array entries before checking used event. */
851 /* Always notify when queue is empty (when feature acknowledge) */
852 if (virtio_has_feature(vdev, VIRTIO_F_NOTIFY_ON_EMPTY) &&
853 !vq->inuse && vring_avail_idx(vq) == vq->last_avail_idx) {
857 if (!virtio_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX)) {
858 return !(vring_avail_flags(vq) & VRING_AVAIL_F_NO_INTERRUPT);
861 v = vq->signalled_used_valid;
862 vq->signalled_used_valid = true;
863 old = vq->signalled_used;
864 new = vq->signalled_used = vring_used_idx(vq);
865 return !v || vring_need_event(vring_get_used_event(vq), new, old);
868 void virtio_notify(VirtIODevice *vdev, VirtQueue *vq)
870 if (!vring_notify(vdev, vq)) {
874 trace_virtio_notify(vdev, vq);
876 virtio_notify_vector(vdev, vq->vector);
879 void virtio_notify_config(VirtIODevice *vdev)
881 if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK))
885 virtio_notify_vector(vdev, vdev->config_vector);
888 static bool virtio_device_endian_needed(void *opaque)
890 VirtIODevice *vdev = opaque;
892 assert(vdev->device_endian != VIRTIO_DEVICE_ENDIAN_UNKNOWN);
893 return vdev->device_endian != virtio_default_endian();
896 static const VMStateDescription vmstate_virtio_device_endian = {
897 .name = "virtio/device_endian",
899 .minimum_version_id = 1,
900 .fields = (VMStateField[]) {
901 VMSTATE_UINT8(device_endian, VirtIODevice),
902 VMSTATE_END_OF_LIST()
906 static const VMStateDescription vmstate_virtio = {
909 .minimum_version_id = 1,
910 .minimum_version_id_old = 1,
911 .fields = (VMStateField[]) {
912 VMSTATE_END_OF_LIST()
914 .subsections = (VMStateSubsection[]) {
916 .vmsd = &vmstate_virtio_device_endian,
917 .needed = &virtio_device_endian_needed
923 void virtio_save(VirtIODevice *vdev, QEMUFile *f)
925 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
926 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
927 VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev);
930 if (k->save_config) {
931 k->save_config(qbus->parent, f);
934 qemu_put_8s(f, &vdev->status);
935 qemu_put_8s(f, &vdev->isr);
936 qemu_put_be16s(f, &vdev->queue_sel);
937 qemu_put_be32s(f, &vdev->guest_features);
938 qemu_put_be32(f, vdev->config_len);
939 qemu_put_buffer(f, vdev->config, vdev->config_len);
941 for (i = 0; i < VIRTIO_PCI_QUEUE_MAX; i++) {
942 if (vdev->vq[i].vring.num == 0)
948 for (i = 0; i < VIRTIO_PCI_QUEUE_MAX; i++) {
949 if (vdev->vq[i].vring.num == 0)
952 qemu_put_be32(f, vdev->vq[i].vring.num);
953 if (k->has_variable_vring_alignment) {
954 qemu_put_be32(f, vdev->vq[i].vring.align);
956 qemu_put_be64(f, vdev->vq[i].pa);
957 qemu_put_be16s(f, &vdev->vq[i].last_avail_idx);
959 k->save_queue(qbus->parent, i, f);
963 if (vdc->save != NULL) {
968 vmstate_save_state(f, &vmstate_virtio, vdev, NULL);
971 int virtio_set_features(VirtIODevice *vdev, uint32_t val)
973 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
974 VirtioBusClass *vbusk = VIRTIO_BUS_GET_CLASS(qbus);
975 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
976 uint32_t supported_features = vbusk->get_features(qbus->parent);
977 bool bad = (val & ~supported_features) != 0;
979 val &= supported_features;
980 if (k->set_features) {
981 k->set_features(vdev, val);
983 vdev->guest_features = val;
987 int virtio_load(VirtIODevice *vdev, QEMUFile *f, int version_id)
993 uint32_t supported_features;
994 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
995 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
996 VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev);
999 * We poison the endianness to ensure it does not get used before
1000 * subsections have been loaded.
1002 vdev->device_endian = VIRTIO_DEVICE_ENDIAN_UNKNOWN;
1004 if (k->load_config) {
1005 ret = k->load_config(qbus->parent, f);
1010 qemu_get_8s(f, &vdev->status);
1011 qemu_get_8s(f, &vdev->isr);
1012 qemu_get_be16s(f, &vdev->queue_sel);
1013 if (vdev->queue_sel >= VIRTIO_PCI_QUEUE_MAX) {
1016 qemu_get_be32s(f, &features);
1018 if (virtio_set_features(vdev, features) < 0) {
1019 supported_features = k->get_features(qbus->parent);
1020 error_report("Features 0x%x unsupported. Allowed features: 0x%x",
1021 features, supported_features);
1024 config_len = qemu_get_be32(f);
1027 * There are cases where the incoming config can be bigger or smaller
1028 * than what we have; so load what we have space for, and skip
1029 * any excess that's in the stream.
1031 qemu_get_buffer(f, vdev->config, MIN(config_len, vdev->config_len));
1033 while (config_len > vdev->config_len) {
1038 num = qemu_get_be32(f);
1040 if (num > VIRTIO_PCI_QUEUE_MAX) {
1041 error_report("Invalid number of PCI queues: 0x%x", num);
1045 for (i = 0; i < num; i++) {
1046 vdev->vq[i].vring.num = qemu_get_be32(f);
1047 if (k->has_variable_vring_alignment) {
1048 vdev->vq[i].vring.align = qemu_get_be32(f);
1050 vdev->vq[i].pa = qemu_get_be64(f);
1051 qemu_get_be16s(f, &vdev->vq[i].last_avail_idx);
1052 vdev->vq[i].signalled_used_valid = false;
1053 vdev->vq[i].notification = true;
1055 if (vdev->vq[i].pa) {
1056 virtqueue_init(&vdev->vq[i]);
1057 } else if (vdev->vq[i].last_avail_idx) {
1058 error_report("VQ %d address 0x0 "
1059 "inconsistent with Host index 0x%x",
1060 i, vdev->vq[i].last_avail_idx);
1063 if (k->load_queue) {
1064 ret = k->load_queue(qbus->parent, i, f);
1070 virtio_notify_vector(vdev, VIRTIO_NO_VECTOR);
1072 if (vdc->load != NULL) {
1073 ret = vdc->load(vdev, f, version_id);
1080 ret = vmstate_load_state(f, &vmstate_virtio, vdev, 1);
1085 if (vdev->device_endian == VIRTIO_DEVICE_ENDIAN_UNKNOWN) {
1086 vdev->device_endian = virtio_default_endian();
1089 for (i = 0; i < num; i++) {
1090 if (vdev->vq[i].pa) {
1092 nheads = vring_avail_idx(&vdev->vq[i]) - vdev->vq[i].last_avail_idx;
1093 /* Check it isn't doing strange things with descriptor numbers. */
1094 if (nheads > vdev->vq[i].vring.num) {
1095 error_report("VQ %d size 0x%x Guest index 0x%x "
1096 "inconsistent with Host index 0x%x: delta 0x%x",
1097 i, vdev->vq[i].vring.num,
1098 vring_avail_idx(&vdev->vq[i]),
1099 vdev->vq[i].last_avail_idx, nheads);
1108 void virtio_cleanup(VirtIODevice *vdev)
1110 qemu_del_vm_change_state_handler(vdev->vmstate);
1111 g_free(vdev->config);
1113 g_free(vdev->vector_queues);
1116 static void virtio_vmstate_change(void *opaque, int running, RunState state)
1118 VirtIODevice *vdev = opaque;
1119 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
1120 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
1121 bool backend_run = running && (vdev->status & VIRTIO_CONFIG_S_DRIVER_OK);
1122 vdev->vm_running = running;
1125 virtio_set_status(vdev, vdev->status);
1128 if (k->vmstate_change) {
1129 k->vmstate_change(qbus->parent, backend_run);
1133 virtio_set_status(vdev, vdev->status);
1137 void virtio_instance_init_common(Object *proxy_obj, void *data,
1138 size_t vdev_size, const char *vdev_name)
1140 DeviceState *vdev = data;
1142 object_initialize(vdev, vdev_size, vdev_name);
1143 object_property_add_child(proxy_obj, "virtio-backend", OBJECT(vdev), NULL);
1144 object_unref(OBJECT(vdev));
1145 qdev_alias_all_properties(vdev, proxy_obj);
1148 void virtio_init(VirtIODevice *vdev, const char *name,
1149 uint16_t device_id, size_t config_size)
1151 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
1152 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
1154 int nvectors = k->query_nvectors ? k->query_nvectors(qbus->parent) : 0;
1157 vdev->vector_queues =
1158 g_malloc0(sizeof(*vdev->vector_queues) * nvectors);
1161 vdev->device_id = device_id;
1164 vdev->queue_sel = 0;
1165 vdev->config_vector = VIRTIO_NO_VECTOR;
1166 vdev->vq = g_malloc0(sizeof(VirtQueue) * VIRTIO_PCI_QUEUE_MAX);
1167 vdev->vm_running = runstate_is_running();
1168 for (i = 0; i < VIRTIO_PCI_QUEUE_MAX; i++) {
1169 vdev->vq[i].vector = VIRTIO_NO_VECTOR;
1170 vdev->vq[i].vdev = vdev;
1171 vdev->vq[i].queue_index = i;
1175 vdev->config_len = config_size;
1176 if (vdev->config_len) {
1177 vdev->config = g_malloc0(config_size);
1179 vdev->config = NULL;
1181 vdev->vmstate = qemu_add_vm_change_state_handler(virtio_vmstate_change,
1183 vdev->device_endian = virtio_default_endian();
1186 hwaddr virtio_queue_get_desc_addr(VirtIODevice *vdev, int n)
1188 return vdev->vq[n].vring.desc;
1191 hwaddr virtio_queue_get_avail_addr(VirtIODevice *vdev, int n)
1193 return vdev->vq[n].vring.avail;
1196 hwaddr virtio_queue_get_used_addr(VirtIODevice *vdev, int n)
1198 return vdev->vq[n].vring.used;
1201 hwaddr virtio_queue_get_ring_addr(VirtIODevice *vdev, int n)
1203 return vdev->vq[n].vring.desc;
1206 hwaddr virtio_queue_get_desc_size(VirtIODevice *vdev, int n)
1208 return sizeof(VRingDesc) * vdev->vq[n].vring.num;
1211 hwaddr virtio_queue_get_avail_size(VirtIODevice *vdev, int n)
1213 return offsetof(VRingAvail, ring) +
1214 sizeof(uint64_t) * vdev->vq[n].vring.num;
1217 hwaddr virtio_queue_get_used_size(VirtIODevice *vdev, int n)
1219 return offsetof(VRingUsed, ring) +
1220 sizeof(VRingUsedElem) * vdev->vq[n].vring.num;
1223 hwaddr virtio_queue_get_ring_size(VirtIODevice *vdev, int n)
1225 return vdev->vq[n].vring.used - vdev->vq[n].vring.desc +
1226 virtio_queue_get_used_size(vdev, n);
1229 uint16_t virtio_queue_get_last_avail_idx(VirtIODevice *vdev, int n)
1231 return vdev->vq[n].last_avail_idx;
1234 void virtio_queue_set_last_avail_idx(VirtIODevice *vdev, int n, uint16_t idx)
1236 vdev->vq[n].last_avail_idx = idx;
1239 void virtio_queue_invalidate_signalled_used(VirtIODevice *vdev, int n)
1241 vdev->vq[n].signalled_used_valid = false;
1244 VirtQueue *virtio_get_queue(VirtIODevice *vdev, int n)
1246 return vdev->vq + n;
1249 uint16_t virtio_get_queue_index(VirtQueue *vq)
1251 return vq->queue_index;
1254 static void virtio_queue_guest_notifier_read(EventNotifier *n)
1256 VirtQueue *vq = container_of(n, VirtQueue, guest_notifier);
1257 if (event_notifier_test_and_clear(n)) {
1262 void virtio_queue_set_guest_notifier_fd_handler(VirtQueue *vq, bool assign,
1265 if (assign && !with_irqfd) {
1266 event_notifier_set_handler(&vq->guest_notifier,
1267 virtio_queue_guest_notifier_read);
1269 event_notifier_set_handler(&vq->guest_notifier, NULL);
1272 /* Test and clear notifier before closing it,
1273 * in case poll callback didn't have time to run. */
1274 virtio_queue_guest_notifier_read(&vq->guest_notifier);
1278 EventNotifier *virtio_queue_get_guest_notifier(VirtQueue *vq)
1280 return &vq->guest_notifier;
1283 static void virtio_queue_host_notifier_read(EventNotifier *n)
1285 VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
1286 if (event_notifier_test_and_clear(n)) {
1287 virtio_queue_notify_vq(vq);
1291 void virtio_queue_set_host_notifier_fd_handler(VirtQueue *vq, bool assign,
1294 if (assign && set_handler) {
1295 event_notifier_set_handler(&vq->host_notifier,
1296 virtio_queue_host_notifier_read);
1298 event_notifier_set_handler(&vq->host_notifier, NULL);
1301 /* Test and clear notifier before after disabling event,
1302 * in case poll callback didn't have time to run. */
1303 virtio_queue_host_notifier_read(&vq->host_notifier);
1307 EventNotifier *virtio_queue_get_host_notifier(VirtQueue *vq)
1309 return &vq->host_notifier;
1312 void virtio_device_set_child_bus_name(VirtIODevice *vdev, char *bus_name)
1314 g_free(vdev->bus_name);
1315 vdev->bus_name = g_strdup(bus_name);
1318 static void virtio_device_realize(DeviceState *dev, Error **errp)
1320 VirtIODevice *vdev = VIRTIO_DEVICE(dev);
1321 VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(dev);
1324 if (vdc->realize != NULL) {
1325 vdc->realize(dev, &err);
1327 error_propagate(errp, err);
1331 virtio_bus_device_plugged(vdev);
1334 static void virtio_device_unrealize(DeviceState *dev, Error **errp)
1336 VirtIODevice *vdev = VIRTIO_DEVICE(dev);
1337 VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(dev);
1340 virtio_bus_device_unplugged(vdev);
1342 if (vdc->unrealize != NULL) {
1343 vdc->unrealize(dev, &err);
1345 error_propagate(errp, err);
1350 g_free(vdev->bus_name);
1351 vdev->bus_name = NULL;
1354 static void virtio_device_class_init(ObjectClass *klass, void *data)
1356 /* Set the default value here. */
1357 DeviceClass *dc = DEVICE_CLASS(klass);
1359 dc->realize = virtio_device_realize;
1360 dc->unrealize = virtio_device_unrealize;
1361 dc->bus_type = TYPE_VIRTIO_BUS;
1364 static const TypeInfo virtio_device_info = {
1365 .name = TYPE_VIRTIO_DEVICE,
1366 .parent = TYPE_DEVICE,
1367 .instance_size = sizeof(VirtIODevice),
1368 .class_init = virtio_device_class_init,
1370 .class_size = sizeof(VirtioDeviceClass),
1373 static void virtio_register_types(void)
1375 type_register_static(&virtio_device_info);
1378 type_init(virtio_register_types)