4 * Copyright IBM, Corp. 2007
7 * Anthony Liguori <aliguori@us.ibm.com>
9 * This work is licensed under the terms of the GNU GPL, version 2. See
10 * the COPYING file in the top-level directory.
14 #include "qemu/osdep.h"
15 #include "qapi/error.h"
16 #include "qemu-common.h"
19 #include "exec/address-spaces.h"
20 #include "qemu/error-report.h"
21 #include "hw/virtio/virtio.h"
22 #include "qemu/atomic.h"
23 #include "hw/virtio/virtio-bus.h"
24 #include "migration/migration.h"
25 #include "hw/virtio/virtio-access.h"
26 #include "sysemu/dma.h"
29 * The alignment to use between consumer and producer parts of vring.
30 * x86 pagesize again. This is the default, used by transports like PCI
31 * which don't provide a means for the guest to tell the host the alignment.
33 #define VIRTIO_PCI_VRING_ALIGN 4096
35 typedef struct VRingDesc
43 typedef struct VRingAvail
50 typedef struct VRingUsedElem
56 typedef struct VRingUsed
60 VRingUsedElem ring
[0];
63 typedef struct VRingMemoryRegionCaches
{
65 MemoryRegionCache desc
;
66 MemoryRegionCache avail
;
67 MemoryRegionCache used
;
68 } VRingMemoryRegionCaches
;
73 unsigned int num_default
;
78 VRingMemoryRegionCaches
*caches
;
85 /* Next head to pop */
86 uint16_t last_avail_idx
;
88 /* Last avail_idx read from VQ. */
89 uint16_t shadow_avail_idx
;
93 /* Last used index value we have signalled on */
94 uint16_t signalled_used
;
96 /* Last used index value we have signalled on */
97 bool signalled_used_valid
;
99 /* Notification enabled? */
102 uint16_t queue_index
;
107 VirtIOHandleOutput handle_output
;
108 VirtIOHandleAIOOutput handle_aio_output
;
110 EventNotifier guest_notifier
;
111 EventNotifier host_notifier
;
112 QLIST_ENTRY(VirtQueue
) node
;
115 static void virtio_free_region_cache(VRingMemoryRegionCaches
*caches
)
121 address_space_cache_destroy(&caches
->desc
);
122 address_space_cache_destroy(&caches
->avail
);
123 address_space_cache_destroy(&caches
->used
);
127 static void virtio_init_region_cache(VirtIODevice
*vdev
, int n
)
129 VirtQueue
*vq
= &vdev
->vq
[n
];
130 VRingMemoryRegionCaches
*old
= vq
->vring
.caches
;
131 VRingMemoryRegionCaches
*new;
135 event_size
= virtio_vdev_has_feature(vq
->vdev
, VIRTIO_RING_F_EVENT_IDX
) ? 2 : 0;
137 addr
= vq
->vring
.desc
;
141 new = g_new0(VRingMemoryRegionCaches
, 1);
142 size
= virtio_queue_get_desc_size(vdev
, n
);
143 address_space_cache_init(&new->desc
, vdev
->dma_as
,
146 size
= virtio_queue_get_used_size(vdev
, n
) + event_size
;
147 address_space_cache_init(&new->used
, vdev
->dma_as
,
148 vq
->vring
.used
, size
, true);
150 size
= virtio_queue_get_avail_size(vdev
, n
) + event_size
;
151 address_space_cache_init(&new->avail
, vdev
->dma_as
,
152 vq
->vring
.avail
, size
, false);
154 atomic_rcu_set(&vq
->vring
.caches
, new);
156 call_rcu(old
, virtio_free_region_cache
, rcu
);
160 /* virt queue functions */
161 void virtio_queue_update_rings(VirtIODevice
*vdev
, int n
)
163 VRing
*vring
= &vdev
->vq
[n
].vring
;
166 /* not yet setup -> nothing to do */
169 vring
->avail
= vring
->desc
+ vring
->num
* sizeof(VRingDesc
);
170 vring
->used
= vring_align(vring
->avail
+
171 offsetof(VRingAvail
, ring
[vring
->num
]),
173 virtio_init_region_cache(vdev
, n
);
176 /* Called within rcu_read_lock(). */
177 static void vring_desc_read(VirtIODevice
*vdev
, VRingDesc
*desc
,
178 MemoryRegionCache
*cache
, int i
)
180 address_space_read_cached(cache
, i
* sizeof(VRingDesc
),
181 desc
, sizeof(VRingDesc
));
182 virtio_tswap64s(vdev
, &desc
->addr
);
183 virtio_tswap32s(vdev
, &desc
->len
);
184 virtio_tswap16s(vdev
, &desc
->flags
);
185 virtio_tswap16s(vdev
, &desc
->next
);
188 /* Called within rcu_read_lock(). */
189 static inline uint16_t vring_avail_flags(VirtQueue
*vq
)
191 VRingMemoryRegionCaches
*caches
= atomic_rcu_read(&vq
->vring
.caches
);
192 hwaddr pa
= offsetof(VRingAvail
, flags
);
193 return virtio_lduw_phys_cached(vq
->vdev
, &caches
->avail
, pa
);
196 /* Called within rcu_read_lock(). */
197 static inline uint16_t vring_avail_idx(VirtQueue
*vq
)
199 VRingMemoryRegionCaches
*caches
= atomic_rcu_read(&vq
->vring
.caches
);
200 hwaddr pa
= offsetof(VRingAvail
, idx
);
201 vq
->shadow_avail_idx
= virtio_lduw_phys_cached(vq
->vdev
, &caches
->avail
, pa
);
202 return vq
->shadow_avail_idx
;
205 /* Called within rcu_read_lock(). */
206 static inline uint16_t vring_avail_ring(VirtQueue
*vq
, int i
)
208 VRingMemoryRegionCaches
*caches
= atomic_rcu_read(&vq
->vring
.caches
);
209 hwaddr pa
= offsetof(VRingAvail
, ring
[i
]);
210 return virtio_lduw_phys_cached(vq
->vdev
, &caches
->avail
, pa
);
213 /* Called within rcu_read_lock(). */
214 static inline uint16_t vring_get_used_event(VirtQueue
*vq
)
216 return vring_avail_ring(vq
, vq
->vring
.num
);
219 /* Called within rcu_read_lock(). */
220 static inline void vring_used_write(VirtQueue
*vq
, VRingUsedElem
*uelem
,
223 VRingMemoryRegionCaches
*caches
= atomic_rcu_read(&vq
->vring
.caches
);
224 hwaddr pa
= offsetof(VRingUsed
, ring
[i
]);
225 virtio_tswap32s(vq
->vdev
, &uelem
->id
);
226 virtio_tswap32s(vq
->vdev
, &uelem
->len
);
227 address_space_write_cached(&caches
->used
, pa
, uelem
, sizeof(VRingUsedElem
));
228 address_space_cache_invalidate(&caches
->used
, pa
, sizeof(VRingUsedElem
));
231 /* Called within rcu_read_lock(). */
232 static uint16_t vring_used_idx(VirtQueue
*vq
)
234 VRingMemoryRegionCaches
*caches
= atomic_rcu_read(&vq
->vring
.caches
);
235 hwaddr pa
= offsetof(VRingUsed
, idx
);
236 return virtio_lduw_phys_cached(vq
->vdev
, &caches
->used
, pa
);
239 /* Called within rcu_read_lock(). */
240 static inline void vring_used_idx_set(VirtQueue
*vq
, uint16_t val
)
242 VRingMemoryRegionCaches
*caches
= atomic_rcu_read(&vq
->vring
.caches
);
243 hwaddr pa
= offsetof(VRingUsed
, idx
);
244 virtio_stw_phys_cached(vq
->vdev
, &caches
->used
, pa
, val
);
245 address_space_cache_invalidate(&caches
->used
, pa
, sizeof(val
));
249 /* Called within rcu_read_lock(). */
250 static inline void vring_used_flags_set_bit(VirtQueue
*vq
, int mask
)
252 VRingMemoryRegionCaches
*caches
= atomic_rcu_read(&vq
->vring
.caches
);
253 VirtIODevice
*vdev
= vq
->vdev
;
254 hwaddr pa
= offsetof(VRingUsed
, flags
);
255 uint16_t flags
= virtio_lduw_phys_cached(vq
->vdev
, &caches
->used
, pa
);
257 virtio_stw_phys_cached(vdev
, &caches
->used
, pa
, flags
| mask
);
258 address_space_cache_invalidate(&caches
->used
, pa
, sizeof(flags
));
261 /* Called within rcu_read_lock(). */
262 static inline void vring_used_flags_unset_bit(VirtQueue
*vq
, int mask
)
264 VRingMemoryRegionCaches
*caches
= atomic_rcu_read(&vq
->vring
.caches
);
265 VirtIODevice
*vdev
= vq
->vdev
;
266 hwaddr pa
= offsetof(VRingUsed
, flags
);
267 uint16_t flags
= virtio_lduw_phys_cached(vq
->vdev
, &caches
->used
, pa
);
269 virtio_stw_phys_cached(vdev
, &caches
->used
, pa
, flags
& ~mask
);
270 address_space_cache_invalidate(&caches
->used
, pa
, sizeof(flags
));
273 /* Called within rcu_read_lock(). */
274 static inline void vring_set_avail_event(VirtQueue
*vq
, uint16_t val
)
276 VRingMemoryRegionCaches
*caches
;
278 if (!vq
->notification
) {
282 caches
= atomic_rcu_read(&vq
->vring
.caches
);
283 pa
= offsetof(VRingUsed
, ring
[vq
->vring
.num
]);
284 virtio_stw_phys_cached(vq
->vdev
, &caches
->used
, pa
, val
);
287 void virtio_queue_set_notification(VirtQueue
*vq
, int enable
)
289 vq
->notification
= enable
;
292 if (virtio_vdev_has_feature(vq
->vdev
, VIRTIO_RING_F_EVENT_IDX
)) {
293 vring_set_avail_event(vq
, vring_avail_idx(vq
));
295 vring_used_flags_unset_bit(vq
, VRING_USED_F_NO_NOTIFY
);
297 vring_used_flags_set_bit(vq
, VRING_USED_F_NO_NOTIFY
);
300 /* Expose avail event/used flags before caller checks the avail idx. */
306 int virtio_queue_ready(VirtQueue
*vq
)
308 return vq
->vring
.avail
!= 0;
311 /* Fetch avail_idx from VQ memory only when we really need to know if
312 * guest has added some buffers.
313 * Called within rcu_read_lock(). */
314 static int virtio_queue_empty_rcu(VirtQueue
*vq
)
316 if (vq
->shadow_avail_idx
!= vq
->last_avail_idx
) {
320 return vring_avail_idx(vq
) == vq
->last_avail_idx
;
323 int virtio_queue_empty(VirtQueue
*vq
)
327 if (vq
->shadow_avail_idx
!= vq
->last_avail_idx
) {
332 empty
= vring_avail_idx(vq
) == vq
->last_avail_idx
;
337 static void virtqueue_unmap_sg(VirtQueue
*vq
, const VirtQueueElement
*elem
,
340 AddressSpace
*dma_as
= vq
->vdev
->dma_as
;
345 for (i
= 0; i
< elem
->in_num
; i
++) {
346 size_t size
= MIN(len
- offset
, elem
->in_sg
[i
].iov_len
);
348 dma_memory_unmap(dma_as
, elem
->in_sg
[i
].iov_base
,
349 elem
->in_sg
[i
].iov_len
,
350 DMA_DIRECTION_FROM_DEVICE
, size
);
355 for (i
= 0; i
< elem
->out_num
; i
++)
356 dma_memory_unmap(dma_as
, elem
->out_sg
[i
].iov_base
,
357 elem
->out_sg
[i
].iov_len
,
358 DMA_DIRECTION_TO_DEVICE
,
359 elem
->out_sg
[i
].iov_len
);
362 /* virtqueue_detach_element:
363 * @vq: The #VirtQueue
364 * @elem: The #VirtQueueElement
365 * @len: number of bytes written
367 * Detach the element from the virtqueue. This function is suitable for device
368 * reset or other situations where a #VirtQueueElement is simply freed and will
369 * not be pushed or discarded.
371 void virtqueue_detach_element(VirtQueue
*vq
, const VirtQueueElement
*elem
,
375 virtqueue_unmap_sg(vq
, elem
, len
);
379 * @vq: The #VirtQueue
380 * @elem: The #VirtQueueElement
381 * @len: number of bytes written
383 * Pretend the most recent element wasn't popped from the virtqueue. The next
384 * call to virtqueue_pop() will refetch the element.
386 void virtqueue_unpop(VirtQueue
*vq
, const VirtQueueElement
*elem
,
389 vq
->last_avail_idx
--;
390 virtqueue_detach_element(vq
, elem
, len
);
394 * @vq: The #VirtQueue
395 * @num: Number of elements to push back
397 * Pretend that elements weren't popped from the virtqueue. The next
398 * virtqueue_pop() will refetch the oldest element.
400 * Use virtqueue_unpop() instead if you have a VirtQueueElement.
402 * Returns: true on success, false if @num is greater than the number of in use
405 bool virtqueue_rewind(VirtQueue
*vq
, unsigned int num
)
407 if (num
> vq
->inuse
) {
410 vq
->last_avail_idx
-= num
;
415 /* Called within rcu_read_lock(). */
416 void virtqueue_fill(VirtQueue
*vq
, const VirtQueueElement
*elem
,
417 unsigned int len
, unsigned int idx
)
421 trace_virtqueue_fill(vq
, elem
, len
, idx
);
423 virtqueue_unmap_sg(vq
, elem
, len
);
425 if (unlikely(vq
->vdev
->broken
)) {
429 idx
= (idx
+ vq
->used_idx
) % vq
->vring
.num
;
431 uelem
.id
= elem
->index
;
433 vring_used_write(vq
, &uelem
, idx
);
436 /* Called within rcu_read_lock(). */
437 void virtqueue_flush(VirtQueue
*vq
, unsigned int count
)
441 if (unlikely(vq
->vdev
->broken
)) {
446 /* Make sure buffer is written before we update index. */
448 trace_virtqueue_flush(vq
, count
);
451 vring_used_idx_set(vq
, new);
453 if (unlikely((int16_t)(new - vq
->signalled_used
) < (uint16_t)(new - old
)))
454 vq
->signalled_used_valid
= false;
457 void virtqueue_push(VirtQueue
*vq
, const VirtQueueElement
*elem
,
461 virtqueue_fill(vq
, elem
, len
, 0);
462 virtqueue_flush(vq
, 1);
466 /* Called within rcu_read_lock(). */
467 static int virtqueue_num_heads(VirtQueue
*vq
, unsigned int idx
)
469 uint16_t num_heads
= vring_avail_idx(vq
) - idx
;
471 /* Check it isn't doing very strange things with descriptor numbers. */
472 if (num_heads
> vq
->vring
.num
) {
473 virtio_error(vq
->vdev
, "Guest moved used index from %u to %u",
474 idx
, vq
->shadow_avail_idx
);
477 /* On success, callers read a descriptor at vq->last_avail_idx.
478 * Make sure descriptor read does not bypass avail index read. */
486 /* Called within rcu_read_lock(). */
487 static bool virtqueue_get_head(VirtQueue
*vq
, unsigned int idx
,
490 /* Grab the next descriptor number they're advertising, and increment
491 * the index we've seen. */
492 *head
= vring_avail_ring(vq
, idx
% vq
->vring
.num
);
494 /* If their number is silly, that's a fatal mistake. */
495 if (*head
>= vq
->vring
.num
) {
496 virtio_error(vq
->vdev
, "Guest says index %u is available", *head
);
504 VIRTQUEUE_READ_DESC_ERROR
= -1,
505 VIRTQUEUE_READ_DESC_DONE
= 0, /* end of chain */
506 VIRTQUEUE_READ_DESC_MORE
= 1, /* more buffers in chain */
509 static int virtqueue_read_next_desc(VirtIODevice
*vdev
, VRingDesc
*desc
,
510 MemoryRegionCache
*desc_cache
, unsigned int max
,
513 /* If this descriptor says it doesn't chain, we're done. */
514 if (!(desc
->flags
& VRING_DESC_F_NEXT
)) {
515 return VIRTQUEUE_READ_DESC_DONE
;
518 /* Check they're not leading us off end of descriptors. */
520 /* Make sure compiler knows to grab that: we don't want it changing! */
524 virtio_error(vdev
, "Desc next is %u", *next
);
525 return VIRTQUEUE_READ_DESC_ERROR
;
528 vring_desc_read(vdev
, desc
, desc_cache
, *next
);
529 return VIRTQUEUE_READ_DESC_MORE
;
532 void virtqueue_get_avail_bytes(VirtQueue
*vq
, unsigned int *in_bytes
,
533 unsigned int *out_bytes
,
534 unsigned max_in_bytes
, unsigned max_out_bytes
)
536 VirtIODevice
*vdev
= vq
->vdev
;
537 unsigned int max
, idx
;
538 unsigned int total_bufs
, in_total
, out_total
;
539 VRingMemoryRegionCaches
*caches
;
540 MemoryRegionCache indirect_desc_cache
= MEMORY_REGION_CACHE_INVALID
;
545 idx
= vq
->last_avail_idx
;
546 total_bufs
= in_total
= out_total
= 0;
549 caches
= atomic_rcu_read(&vq
->vring
.caches
);
550 if (caches
->desc
.len
< max
* sizeof(VRingDesc
)) {
551 virtio_error(vdev
, "Cannot map descriptor ring");
555 while ((rc
= virtqueue_num_heads(vq
, idx
)) > 0) {
556 MemoryRegionCache
*desc_cache
= &caches
->desc
;
557 unsigned int num_bufs
;
561 num_bufs
= total_bufs
;
563 if (!virtqueue_get_head(vq
, idx
++, &i
)) {
567 vring_desc_read(vdev
, &desc
, desc_cache
, i
);
569 if (desc
.flags
& VRING_DESC_F_INDIRECT
) {
570 if (desc
.len
% sizeof(VRingDesc
)) {
571 virtio_error(vdev
, "Invalid size for indirect buffer table");
575 /* If we've got too many, that implies a descriptor loop. */
576 if (num_bufs
>= max
) {
577 virtio_error(vdev
, "Looped descriptor");
581 /* loop over the indirect descriptor table */
582 len
= address_space_cache_init(&indirect_desc_cache
,
584 desc
.addr
, desc
.len
, false);
585 desc_cache
= &indirect_desc_cache
;
586 if (len
< desc
.len
) {
587 virtio_error(vdev
, "Cannot map indirect buffer");
591 max
= desc
.len
/ sizeof(VRingDesc
);
593 vring_desc_read(vdev
, &desc
, desc_cache
, i
);
597 /* If we've got too many, that implies a descriptor loop. */
598 if (++num_bufs
> max
) {
599 virtio_error(vdev
, "Looped descriptor");
603 if (desc
.flags
& VRING_DESC_F_WRITE
) {
604 in_total
+= desc
.len
;
606 out_total
+= desc
.len
;
608 if (in_total
>= max_in_bytes
&& out_total
>= max_out_bytes
) {
612 rc
= virtqueue_read_next_desc(vdev
, &desc
, desc_cache
, max
, &i
);
613 } while (rc
== VIRTQUEUE_READ_DESC_MORE
);
615 if (rc
== VIRTQUEUE_READ_DESC_ERROR
) {
619 if (desc_cache
== &indirect_desc_cache
) {
620 address_space_cache_destroy(&indirect_desc_cache
);
623 total_bufs
= num_bufs
;
632 address_space_cache_destroy(&indirect_desc_cache
);
634 *in_bytes
= in_total
;
637 *out_bytes
= out_total
;
643 in_total
= out_total
= 0;
647 int virtqueue_avail_bytes(VirtQueue
*vq
, unsigned int in_bytes
,
648 unsigned int out_bytes
)
650 unsigned int in_total
, out_total
;
652 virtqueue_get_avail_bytes(vq
, &in_total
, &out_total
, in_bytes
, out_bytes
);
653 return in_bytes
<= in_total
&& out_bytes
<= out_total
;
656 static bool virtqueue_map_desc(VirtIODevice
*vdev
, unsigned int *p_num_sg
,
657 hwaddr
*addr
, struct iovec
*iov
,
658 unsigned int max_num_sg
, bool is_write
,
659 hwaddr pa
, size_t sz
)
662 unsigned num_sg
= *p_num_sg
;
663 assert(num_sg
<= max_num_sg
);
666 virtio_error(vdev
, "virtio: zero sized buffers are not allowed");
673 if (num_sg
== max_num_sg
) {
674 virtio_error(vdev
, "virtio: too many write descriptors in "
679 iov
[num_sg
].iov_base
= dma_memory_map(vdev
->dma_as
, pa
, &len
,
681 DMA_DIRECTION_FROM_DEVICE
:
682 DMA_DIRECTION_TO_DEVICE
);
683 if (!iov
[num_sg
].iov_base
) {
684 virtio_error(vdev
, "virtio: bogus descriptor or out of resources");
688 iov
[num_sg
].iov_len
= len
;
702 /* Only used by error code paths before we have a VirtQueueElement (therefore
703 * virtqueue_unmap_sg() can't be used). Assumes buffers weren't written to
706 static void virtqueue_undo_map_desc(unsigned int out_num
, unsigned int in_num
,
711 for (i
= 0; i
< out_num
+ in_num
; i
++) {
712 int is_write
= i
>= out_num
;
714 cpu_physical_memory_unmap(iov
->iov_base
, iov
->iov_len
, is_write
, 0);
719 static void virtqueue_map_iovec(VirtIODevice
*vdev
, struct iovec
*sg
,
720 hwaddr
*addr
, unsigned int *num_sg
,
726 for (i
= 0; i
< *num_sg
; i
++) {
728 sg
[i
].iov_base
= dma_memory_map(vdev
->dma_as
,
729 addr
[i
], &len
, is_write
?
730 DMA_DIRECTION_FROM_DEVICE
:
731 DMA_DIRECTION_TO_DEVICE
);
732 if (!sg
[i
].iov_base
) {
733 error_report("virtio: error trying to map MMIO memory");
736 if (len
!= sg
[i
].iov_len
) {
737 error_report("virtio: unexpected memory split");
743 void virtqueue_map(VirtIODevice
*vdev
, VirtQueueElement
*elem
)
745 virtqueue_map_iovec(vdev
, elem
->in_sg
, elem
->in_addr
, &elem
->in_num
, 1);
746 virtqueue_map_iovec(vdev
, elem
->out_sg
, elem
->out_addr
, &elem
->out_num
, 0);
749 static void *virtqueue_alloc_element(size_t sz
, unsigned out_num
, unsigned in_num
)
751 VirtQueueElement
*elem
;
752 size_t in_addr_ofs
= QEMU_ALIGN_UP(sz
, __alignof__(elem
->in_addr
[0]));
753 size_t out_addr_ofs
= in_addr_ofs
+ in_num
* sizeof(elem
->in_addr
[0]);
754 size_t out_addr_end
= out_addr_ofs
+ out_num
* sizeof(elem
->out_addr
[0]);
755 size_t in_sg_ofs
= QEMU_ALIGN_UP(out_addr_end
, __alignof__(elem
->in_sg
[0]));
756 size_t out_sg_ofs
= in_sg_ofs
+ in_num
* sizeof(elem
->in_sg
[0]);
757 size_t out_sg_end
= out_sg_ofs
+ out_num
* sizeof(elem
->out_sg
[0]);
759 assert(sz
>= sizeof(VirtQueueElement
));
760 elem
= g_malloc(out_sg_end
);
761 elem
->out_num
= out_num
;
762 elem
->in_num
= in_num
;
763 elem
->in_addr
= (void *)elem
+ in_addr_ofs
;
764 elem
->out_addr
= (void *)elem
+ out_addr_ofs
;
765 elem
->in_sg
= (void *)elem
+ in_sg_ofs
;
766 elem
->out_sg
= (void *)elem
+ out_sg_ofs
;
770 void *virtqueue_pop(VirtQueue
*vq
, size_t sz
)
772 unsigned int i
, head
, max
;
773 VRingMemoryRegionCaches
*caches
;
774 MemoryRegionCache indirect_desc_cache
= MEMORY_REGION_CACHE_INVALID
;
775 MemoryRegionCache
*desc_cache
;
777 VirtIODevice
*vdev
= vq
->vdev
;
778 VirtQueueElement
*elem
= NULL
;
779 unsigned out_num
, in_num
;
780 hwaddr addr
[VIRTQUEUE_MAX_SIZE
];
781 struct iovec iov
[VIRTQUEUE_MAX_SIZE
];
785 if (unlikely(vdev
->broken
)) {
789 if (virtio_queue_empty_rcu(vq
)) {
792 /* Needed after virtio_queue_empty(), see comment in
793 * virtqueue_num_heads(). */
796 /* When we start there are none of either input nor output. */
797 out_num
= in_num
= 0;
801 if (vq
->inuse
>= vq
->vring
.num
) {
802 virtio_error(vdev
, "Virtqueue size exceeded");
806 if (!virtqueue_get_head(vq
, vq
->last_avail_idx
++, &head
)) {
810 if (virtio_vdev_has_feature(vdev
, VIRTIO_RING_F_EVENT_IDX
)) {
811 vring_set_avail_event(vq
, vq
->last_avail_idx
);
816 caches
= atomic_rcu_read(&vq
->vring
.caches
);
817 if (caches
->desc
.len
< max
* sizeof(VRingDesc
)) {
818 virtio_error(vdev
, "Cannot map descriptor ring");
822 desc_cache
= &caches
->desc
;
823 vring_desc_read(vdev
, &desc
, desc_cache
, i
);
824 if (desc
.flags
& VRING_DESC_F_INDIRECT
) {
825 if (desc
.len
% sizeof(VRingDesc
)) {
826 virtio_error(vdev
, "Invalid size for indirect buffer table");
830 /* loop over the indirect descriptor table */
831 len
= address_space_cache_init(&indirect_desc_cache
, vdev
->dma_as
,
832 desc
.addr
, desc
.len
, false);
833 desc_cache
= &indirect_desc_cache
;
834 if (len
< desc
.len
) {
835 virtio_error(vdev
, "Cannot map indirect buffer");
839 max
= desc
.len
/ sizeof(VRingDesc
);
841 vring_desc_read(vdev
, &desc
, desc_cache
, i
);
844 /* Collect all the descriptors */
848 if (desc
.flags
& VRING_DESC_F_WRITE
) {
849 map_ok
= virtqueue_map_desc(vdev
, &in_num
, addr
+ out_num
,
851 VIRTQUEUE_MAX_SIZE
- out_num
, true,
852 desc
.addr
, desc
.len
);
855 virtio_error(vdev
, "Incorrect order for descriptors");
858 map_ok
= virtqueue_map_desc(vdev
, &out_num
, addr
, iov
,
859 VIRTQUEUE_MAX_SIZE
, false,
860 desc
.addr
, desc
.len
);
866 /* If we've got too many, that implies a descriptor loop. */
867 if ((in_num
+ out_num
) > max
) {
868 virtio_error(vdev
, "Looped descriptor");
872 rc
= virtqueue_read_next_desc(vdev
, &desc
, desc_cache
, max
, &i
);
873 } while (rc
== VIRTQUEUE_READ_DESC_MORE
);
875 if (rc
== VIRTQUEUE_READ_DESC_ERROR
) {
879 /* Now copy what we have collected and mapped */
880 elem
= virtqueue_alloc_element(sz
, out_num
, in_num
);
882 for (i
= 0; i
< out_num
; i
++) {
883 elem
->out_addr
[i
] = addr
[i
];
884 elem
->out_sg
[i
] = iov
[i
];
886 for (i
= 0; i
< in_num
; i
++) {
887 elem
->in_addr
[i
] = addr
[out_num
+ i
];
888 elem
->in_sg
[i
] = iov
[out_num
+ i
];
893 trace_virtqueue_pop(vq
, elem
, elem
->in_num
, elem
->out_num
);
895 address_space_cache_destroy(&indirect_desc_cache
);
901 virtqueue_undo_map_desc(out_num
, in_num
, iov
);
905 /* virtqueue_drop_all:
906 * @vq: The #VirtQueue
907 * Drops all queued buffers and indicates them to the guest
908 * as if they are done. Useful when buffers can not be
909 * processed but must be returned to the guest.
911 unsigned int virtqueue_drop_all(VirtQueue
*vq
)
913 unsigned int dropped
= 0;
914 VirtQueueElement elem
= {};
915 VirtIODevice
*vdev
= vq
->vdev
;
916 bool fEventIdx
= virtio_vdev_has_feature(vdev
, VIRTIO_RING_F_EVENT_IDX
);
918 if (unlikely(vdev
->broken
)) {
922 while (!virtio_queue_empty(vq
) && vq
->inuse
< vq
->vring
.num
) {
923 /* works similar to virtqueue_pop but does not map buffers
924 * and does not allocate any memory */
926 if (!virtqueue_get_head(vq
, vq
->last_avail_idx
, &elem
.index
)) {
930 vq
->last_avail_idx
++;
932 vring_set_avail_event(vq
, vq
->last_avail_idx
);
934 /* immediately push the element, nothing to unmap
935 * as both in_num and out_num are set to 0 */
936 virtqueue_push(vq
, &elem
, 0);
943 /* Reading and writing a structure directly to QEMUFile is *awful*, but
944 * it is what QEMU has always done by mistake. We can change it sooner
945 * or later by bumping the version number of the affected vm states.
946 * In the meanwhile, since the in-memory layout of VirtQueueElement
947 * has changed, we need to marshal to and from the layout that was
948 * used before the change.
950 typedef struct VirtQueueElementOld
{
952 unsigned int out_num
;
954 hwaddr in_addr
[VIRTQUEUE_MAX_SIZE
];
955 hwaddr out_addr
[VIRTQUEUE_MAX_SIZE
];
956 struct iovec in_sg
[VIRTQUEUE_MAX_SIZE
];
957 struct iovec out_sg
[VIRTQUEUE_MAX_SIZE
];
958 } VirtQueueElementOld
;
960 void *qemu_get_virtqueue_element(VirtIODevice
*vdev
, QEMUFile
*f
, size_t sz
)
962 VirtQueueElement
*elem
;
963 VirtQueueElementOld data
;
966 qemu_get_buffer(f
, (uint8_t *)&data
, sizeof(VirtQueueElementOld
));
968 /* TODO: teach all callers that this can fail, and return failure instead
970 * When we do, we might be able to re-enable NDEBUG below.
973 #error building with NDEBUG is not supported
975 assert(ARRAY_SIZE(data
.in_addr
) >= data
.in_num
);
976 assert(ARRAY_SIZE(data
.out_addr
) >= data
.out_num
);
978 elem
= virtqueue_alloc_element(sz
, data
.out_num
, data
.in_num
);
979 elem
->index
= data
.index
;
981 for (i
= 0; i
< elem
->in_num
; i
++) {
982 elem
->in_addr
[i
] = data
.in_addr
[i
];
985 for (i
= 0; i
< elem
->out_num
; i
++) {
986 elem
->out_addr
[i
] = data
.out_addr
[i
];
989 for (i
= 0; i
< elem
->in_num
; i
++) {
990 /* Base is overwritten by virtqueue_map. */
991 elem
->in_sg
[i
].iov_base
= 0;
992 elem
->in_sg
[i
].iov_len
= data
.in_sg
[i
].iov_len
;
995 for (i
= 0; i
< elem
->out_num
; i
++) {
996 /* Base is overwritten by virtqueue_map. */
997 elem
->out_sg
[i
].iov_base
= 0;
998 elem
->out_sg
[i
].iov_len
= data
.out_sg
[i
].iov_len
;
1001 virtqueue_map(vdev
, elem
);
1005 void qemu_put_virtqueue_element(QEMUFile
*f
, VirtQueueElement
*elem
)
1007 VirtQueueElementOld data
;
1010 memset(&data
, 0, sizeof(data
));
1011 data
.index
= elem
->index
;
1012 data
.in_num
= elem
->in_num
;
1013 data
.out_num
= elem
->out_num
;
1015 for (i
= 0; i
< elem
->in_num
; i
++) {
1016 data
.in_addr
[i
] = elem
->in_addr
[i
];
1019 for (i
= 0; i
< elem
->out_num
; i
++) {
1020 data
.out_addr
[i
] = elem
->out_addr
[i
];
1023 for (i
= 0; i
< elem
->in_num
; i
++) {
1024 /* Base is overwritten by virtqueue_map when loading. Do not
1025 * save it, as it would leak the QEMU address space layout. */
1026 data
.in_sg
[i
].iov_len
= elem
->in_sg
[i
].iov_len
;
1029 for (i
= 0; i
< elem
->out_num
; i
++) {
1030 /* Do not save iov_base as above. */
1031 data
.out_sg
[i
].iov_len
= elem
->out_sg
[i
].iov_len
;
1033 qemu_put_buffer(f
, (uint8_t *)&data
, sizeof(VirtQueueElementOld
));
1037 static void virtio_notify_vector(VirtIODevice
*vdev
, uint16_t vector
)
1039 BusState
*qbus
= qdev_get_parent_bus(DEVICE(vdev
));
1040 VirtioBusClass
*k
= VIRTIO_BUS_GET_CLASS(qbus
);
1042 if (unlikely(vdev
->broken
)) {
1047 k
->notify(qbus
->parent
, vector
);
1051 void virtio_update_irq(VirtIODevice
*vdev
)
1053 virtio_notify_vector(vdev
, VIRTIO_NO_VECTOR
);
1056 static int virtio_validate_features(VirtIODevice
*vdev
)
1058 VirtioDeviceClass
*k
= VIRTIO_DEVICE_GET_CLASS(vdev
);
1060 if (virtio_host_has_feature(vdev
, VIRTIO_F_IOMMU_PLATFORM
) &&
1061 !virtio_vdev_has_feature(vdev
, VIRTIO_F_IOMMU_PLATFORM
)) {
1065 if (k
->validate_features
) {
1066 return k
->validate_features(vdev
);
1072 int virtio_set_status(VirtIODevice
*vdev
, uint8_t val
)
1074 VirtioDeviceClass
*k
= VIRTIO_DEVICE_GET_CLASS(vdev
);
1075 trace_virtio_set_status(vdev
, val
);
1077 if (virtio_vdev_has_feature(vdev
, VIRTIO_F_VERSION_1
)) {
1078 if (!(vdev
->status
& VIRTIO_CONFIG_S_FEATURES_OK
) &&
1079 val
& VIRTIO_CONFIG_S_FEATURES_OK
) {
1080 int ret
= virtio_validate_features(vdev
);
1087 if (k
->set_status
) {
1088 k
->set_status(vdev
, val
);
1094 bool target_words_bigendian(void);
1095 static enum virtio_device_endian
virtio_default_endian(void)
1097 if (target_words_bigendian()) {
1098 return VIRTIO_DEVICE_ENDIAN_BIG
;
1100 return VIRTIO_DEVICE_ENDIAN_LITTLE
;
1104 static enum virtio_device_endian
virtio_current_cpu_endian(void)
1106 CPUClass
*cc
= CPU_GET_CLASS(current_cpu
);
1108 if (cc
->virtio_is_big_endian(current_cpu
)) {
1109 return VIRTIO_DEVICE_ENDIAN_BIG
;
1111 return VIRTIO_DEVICE_ENDIAN_LITTLE
;
1115 void virtio_reset(void *opaque
)
1117 VirtIODevice
*vdev
= opaque
;
1118 VirtioDeviceClass
*k
= VIRTIO_DEVICE_GET_CLASS(vdev
);
1121 virtio_set_status(vdev
, 0);
1123 /* Guest initiated reset */
1124 vdev
->device_endian
= virtio_current_cpu_endian();
1127 vdev
->device_endian
= virtio_default_endian();
1134 vdev
->broken
= false;
1135 vdev
->guest_features
= 0;
1136 vdev
->queue_sel
= 0;
1138 atomic_set(&vdev
->isr
, 0);
1139 vdev
->config_vector
= VIRTIO_NO_VECTOR
;
1140 virtio_notify_vector(vdev
, vdev
->config_vector
);
1142 for(i
= 0; i
< VIRTIO_QUEUE_MAX
; i
++) {
1143 vdev
->vq
[i
].vring
.desc
= 0;
1144 vdev
->vq
[i
].vring
.avail
= 0;
1145 vdev
->vq
[i
].vring
.used
= 0;
1146 vdev
->vq
[i
].last_avail_idx
= 0;
1147 vdev
->vq
[i
].shadow_avail_idx
= 0;
1148 vdev
->vq
[i
].used_idx
= 0;
1149 virtio_queue_set_vector(vdev
, i
, VIRTIO_NO_VECTOR
);
1150 vdev
->vq
[i
].signalled_used
= 0;
1151 vdev
->vq
[i
].signalled_used_valid
= false;
1152 vdev
->vq
[i
].notification
= true;
1153 vdev
->vq
[i
].vring
.num
= vdev
->vq
[i
].vring
.num_default
;
1154 vdev
->vq
[i
].inuse
= 0;
1158 uint32_t virtio_config_readb(VirtIODevice
*vdev
, uint32_t addr
)
1160 VirtioDeviceClass
*k
= VIRTIO_DEVICE_GET_CLASS(vdev
);
1163 if (addr
+ sizeof(val
) > vdev
->config_len
) {
1164 return (uint32_t)-1;
1167 k
->get_config(vdev
, vdev
->config
);
1169 val
= ldub_p(vdev
->config
+ addr
);
1173 uint32_t virtio_config_readw(VirtIODevice
*vdev
, uint32_t addr
)
1175 VirtioDeviceClass
*k
= VIRTIO_DEVICE_GET_CLASS(vdev
);
1178 if (addr
+ sizeof(val
) > vdev
->config_len
) {
1179 return (uint32_t)-1;
1182 k
->get_config(vdev
, vdev
->config
);
1184 val
= lduw_p(vdev
->config
+ addr
);
1188 uint32_t virtio_config_readl(VirtIODevice
*vdev
, uint32_t addr
)
1190 VirtioDeviceClass
*k
= VIRTIO_DEVICE_GET_CLASS(vdev
);
1193 if (addr
+ sizeof(val
) > vdev
->config_len
) {
1194 return (uint32_t)-1;
1197 k
->get_config(vdev
, vdev
->config
);
1199 val
= ldl_p(vdev
->config
+ addr
);
1203 void virtio_config_writeb(VirtIODevice
*vdev
, uint32_t addr
, uint32_t data
)
1205 VirtioDeviceClass
*k
= VIRTIO_DEVICE_GET_CLASS(vdev
);
1208 if (addr
+ sizeof(val
) > vdev
->config_len
) {
1212 stb_p(vdev
->config
+ addr
, val
);
1214 if (k
->set_config
) {
1215 k
->set_config(vdev
, vdev
->config
);
1219 void virtio_config_writew(VirtIODevice
*vdev
, uint32_t addr
, uint32_t data
)
1221 VirtioDeviceClass
*k
= VIRTIO_DEVICE_GET_CLASS(vdev
);
1222 uint16_t val
= data
;
1224 if (addr
+ sizeof(val
) > vdev
->config_len
) {
1228 stw_p(vdev
->config
+ addr
, val
);
1230 if (k
->set_config
) {
1231 k
->set_config(vdev
, vdev
->config
);
1235 void virtio_config_writel(VirtIODevice
*vdev
, uint32_t addr
, uint32_t data
)
1237 VirtioDeviceClass
*k
= VIRTIO_DEVICE_GET_CLASS(vdev
);
1238 uint32_t val
= data
;
1240 if (addr
+ sizeof(val
) > vdev
->config_len
) {
1244 stl_p(vdev
->config
+ addr
, val
);
1246 if (k
->set_config
) {
1247 k
->set_config(vdev
, vdev
->config
);
1251 uint32_t virtio_config_modern_readb(VirtIODevice
*vdev
, uint32_t addr
)
1253 VirtioDeviceClass
*k
= VIRTIO_DEVICE_GET_CLASS(vdev
);
1256 if (addr
+ sizeof(val
) > vdev
->config_len
) {
1257 return (uint32_t)-1;
1260 k
->get_config(vdev
, vdev
->config
);
1262 val
= ldub_p(vdev
->config
+ addr
);
1266 uint32_t virtio_config_modern_readw(VirtIODevice
*vdev
, uint32_t addr
)
1268 VirtioDeviceClass
*k
= VIRTIO_DEVICE_GET_CLASS(vdev
);
1271 if (addr
+ sizeof(val
) > vdev
->config_len
) {
1272 return (uint32_t)-1;
1275 k
->get_config(vdev
, vdev
->config
);
1277 val
= lduw_le_p(vdev
->config
+ addr
);
1281 uint32_t virtio_config_modern_readl(VirtIODevice
*vdev
, uint32_t addr
)
1283 VirtioDeviceClass
*k
= VIRTIO_DEVICE_GET_CLASS(vdev
);
1286 if (addr
+ sizeof(val
) > vdev
->config_len
) {
1287 return (uint32_t)-1;
1290 k
->get_config(vdev
, vdev
->config
);
1292 val
= ldl_le_p(vdev
->config
+ addr
);
1296 void virtio_config_modern_writeb(VirtIODevice
*vdev
,
1297 uint32_t addr
, uint32_t data
)
1299 VirtioDeviceClass
*k
= VIRTIO_DEVICE_GET_CLASS(vdev
);
1302 if (addr
+ sizeof(val
) > vdev
->config_len
) {
1306 stb_p(vdev
->config
+ addr
, val
);
1308 if (k
->set_config
) {
1309 k
->set_config(vdev
, vdev
->config
);
1313 void virtio_config_modern_writew(VirtIODevice
*vdev
,
1314 uint32_t addr
, uint32_t data
)
1316 VirtioDeviceClass
*k
= VIRTIO_DEVICE_GET_CLASS(vdev
);
1317 uint16_t val
= data
;
1319 if (addr
+ sizeof(val
) > vdev
->config_len
) {
1323 stw_le_p(vdev
->config
+ addr
, val
);
1325 if (k
->set_config
) {
1326 k
->set_config(vdev
, vdev
->config
);
1330 void virtio_config_modern_writel(VirtIODevice
*vdev
,
1331 uint32_t addr
, uint32_t data
)
1333 VirtioDeviceClass
*k
= VIRTIO_DEVICE_GET_CLASS(vdev
);
1334 uint32_t val
= data
;
1336 if (addr
+ sizeof(val
) > vdev
->config_len
) {
1340 stl_le_p(vdev
->config
+ addr
, val
);
1342 if (k
->set_config
) {
1343 k
->set_config(vdev
, vdev
->config
);
1347 void virtio_queue_set_addr(VirtIODevice
*vdev
, int n
, hwaddr addr
)
1349 vdev
->vq
[n
].vring
.desc
= addr
;
1350 virtio_queue_update_rings(vdev
, n
);
1353 hwaddr
virtio_queue_get_addr(VirtIODevice
*vdev
, int n
)
1355 return vdev
->vq
[n
].vring
.desc
;
1358 void virtio_queue_set_rings(VirtIODevice
*vdev
, int n
, hwaddr desc
,
1359 hwaddr avail
, hwaddr used
)
1361 vdev
->vq
[n
].vring
.desc
= desc
;
1362 vdev
->vq
[n
].vring
.avail
= avail
;
1363 vdev
->vq
[n
].vring
.used
= used
;
1364 virtio_init_region_cache(vdev
, n
);
1367 void virtio_queue_set_num(VirtIODevice
*vdev
, int n
, int num
)
1369 /* Don't allow guest to flip queue between existent and
1370 * nonexistent states, or to set it to an invalid size.
1372 if (!!num
!= !!vdev
->vq
[n
].vring
.num
||
1373 num
> VIRTQUEUE_MAX_SIZE
||
1377 vdev
->vq
[n
].vring
.num
= num
;
1380 VirtQueue
*virtio_vector_first_queue(VirtIODevice
*vdev
, uint16_t vector
)
1382 return QLIST_FIRST(&vdev
->vector_queues
[vector
]);
1385 VirtQueue
*virtio_vector_next_queue(VirtQueue
*vq
)
1387 return QLIST_NEXT(vq
, node
);
1390 int virtio_queue_get_num(VirtIODevice
*vdev
, int n
)
1392 return vdev
->vq
[n
].vring
.num
;
1395 int virtio_queue_get_max_num(VirtIODevice
*vdev
, int n
)
1397 return vdev
->vq
[n
].vring
.num_default
;
1400 int virtio_get_num_queues(VirtIODevice
*vdev
)
1404 for (i
= 0; i
< VIRTIO_QUEUE_MAX
; i
++) {
1405 if (!virtio_queue_get_num(vdev
, i
)) {
1413 void virtio_queue_set_align(VirtIODevice
*vdev
, int n
, int align
)
1415 BusState
*qbus
= qdev_get_parent_bus(DEVICE(vdev
));
1416 VirtioBusClass
*k
= VIRTIO_BUS_GET_CLASS(qbus
);
1418 /* virtio-1 compliant devices cannot change the alignment */
1419 if (virtio_vdev_has_feature(vdev
, VIRTIO_F_VERSION_1
)) {
1420 error_report("tried to modify queue alignment for virtio-1 device");
1423 /* Check that the transport told us it was going to do this
1424 * (so a buggy transport will immediately assert rather than
1425 * silently failing to migrate this state)
1427 assert(k
->has_variable_vring_alignment
);
1429 vdev
->vq
[n
].vring
.align
= align
;
1430 virtio_queue_update_rings(vdev
, n
);
1433 static bool virtio_queue_notify_aio_vq(VirtQueue
*vq
)
1435 if (vq
->vring
.desc
&& vq
->handle_aio_output
) {
1436 VirtIODevice
*vdev
= vq
->vdev
;
1438 trace_virtio_queue_notify(vdev
, vq
- vdev
->vq
, vq
);
1439 return vq
->handle_aio_output(vdev
, vq
);
1445 static void virtio_queue_notify_vq(VirtQueue
*vq
)
1447 if (vq
->vring
.desc
&& vq
->handle_output
) {
1448 VirtIODevice
*vdev
= vq
->vdev
;
1450 if (unlikely(vdev
->broken
)) {
1454 trace_virtio_queue_notify(vdev
, vq
- vdev
->vq
, vq
);
1455 vq
->handle_output(vdev
, vq
);
1459 void virtio_queue_notify(VirtIODevice
*vdev
, int n
)
1461 virtio_queue_notify_vq(&vdev
->vq
[n
]);
1464 uint16_t virtio_queue_vector(VirtIODevice
*vdev
, int n
)
1466 return n
< VIRTIO_QUEUE_MAX
? vdev
->vq
[n
].vector
:
1470 void virtio_queue_set_vector(VirtIODevice
*vdev
, int n
, uint16_t vector
)
1472 VirtQueue
*vq
= &vdev
->vq
[n
];
1474 if (n
< VIRTIO_QUEUE_MAX
) {
1475 if (vdev
->vector_queues
&&
1476 vdev
->vq
[n
].vector
!= VIRTIO_NO_VECTOR
) {
1477 QLIST_REMOVE(vq
, node
);
1479 vdev
->vq
[n
].vector
= vector
;
1480 if (vdev
->vector_queues
&&
1481 vector
!= VIRTIO_NO_VECTOR
) {
1482 QLIST_INSERT_HEAD(&vdev
->vector_queues
[vector
], vq
, node
);
1487 VirtQueue
*virtio_add_queue(VirtIODevice
*vdev
, int queue_size
,
1488 VirtIOHandleOutput handle_output
)
1492 for (i
= 0; i
< VIRTIO_QUEUE_MAX
; i
++) {
1493 if (vdev
->vq
[i
].vring
.num
== 0)
1497 if (i
== VIRTIO_QUEUE_MAX
|| queue_size
> VIRTQUEUE_MAX_SIZE
)
1500 vdev
->vq
[i
].vring
.num
= queue_size
;
1501 vdev
->vq
[i
].vring
.num_default
= queue_size
;
1502 vdev
->vq
[i
].vring
.align
= VIRTIO_PCI_VRING_ALIGN
;
1503 vdev
->vq
[i
].handle_output
= handle_output
;
1504 vdev
->vq
[i
].handle_aio_output
= NULL
;
1506 return &vdev
->vq
[i
];
1509 void virtio_del_queue(VirtIODevice
*vdev
, int n
)
1511 if (n
< 0 || n
>= VIRTIO_QUEUE_MAX
) {
1515 vdev
->vq
[n
].vring
.num
= 0;
1516 vdev
->vq
[n
].vring
.num_default
= 0;
1519 static void virtio_set_isr(VirtIODevice
*vdev
, int value
)
1521 uint8_t old
= atomic_read(&vdev
->isr
);
1523 /* Do not write ISR if it does not change, so that its cacheline remains
1524 * shared in the common case where the guest does not read it.
1526 if ((old
& value
) != value
) {
1527 atomic_or(&vdev
->isr
, value
);
1531 /* Called within rcu_read_lock(). */
1532 static bool virtio_should_notify(VirtIODevice
*vdev
, VirtQueue
*vq
)
1536 /* We need to expose used array entries before checking used event. */
1538 /* Always notify when queue is empty (when feature acknowledge) */
1539 if (virtio_vdev_has_feature(vdev
, VIRTIO_F_NOTIFY_ON_EMPTY
) &&
1540 !vq
->inuse
&& virtio_queue_empty(vq
)) {
1544 if (!virtio_vdev_has_feature(vdev
, VIRTIO_RING_F_EVENT_IDX
)) {
1545 return !(vring_avail_flags(vq
) & VRING_AVAIL_F_NO_INTERRUPT
);
1548 v
= vq
->signalled_used_valid
;
1549 vq
->signalled_used_valid
= true;
1550 old
= vq
->signalled_used
;
1551 new = vq
->signalled_used
= vq
->used_idx
;
1552 return !v
|| vring_need_event(vring_get_used_event(vq
), new, old
);
1555 void virtio_notify_irqfd(VirtIODevice
*vdev
, VirtQueue
*vq
)
1559 should_notify
= virtio_should_notify(vdev
, vq
);
1562 if (!should_notify
) {
1566 trace_virtio_notify_irqfd(vdev
, vq
);
1569 * virtio spec 1.0 says ISR bit 0 should be ignored with MSI, but
1570 * windows drivers included in virtio-win 1.8.0 (circa 2015) are
1571 * incorrectly polling this bit during crashdump and hibernation
1572 * in MSI mode, causing a hang if this bit is never updated.
1573 * Recent releases of Windows do not really shut down, but rather
1574 * log out and hibernate to make the next startup faster. Hence,
1575 * this manifested as a more serious hang during shutdown with
1577 * Next driver release from 2016 fixed this problem, so working around it
1578 * is not a must, but it's easy to do so let's do it here.
1580 * Note: it's safe to update ISR from any thread as it was switched
1581 * to an atomic operation.
1583 virtio_set_isr(vq
->vdev
, 0x1);
1584 event_notifier_set(&vq
->guest_notifier
);
1587 static void virtio_irq(VirtQueue
*vq
)
1589 virtio_set_isr(vq
->vdev
, 0x1);
1590 virtio_notify_vector(vq
->vdev
, vq
->vector
);
1593 void virtio_notify(VirtIODevice
*vdev
, VirtQueue
*vq
)
1597 should_notify
= virtio_should_notify(vdev
, vq
);
1600 if (!should_notify
) {
1604 trace_virtio_notify(vdev
, vq
);
1608 void virtio_notify_config(VirtIODevice
*vdev
)
1610 if (!(vdev
->status
& VIRTIO_CONFIG_S_DRIVER_OK
))
1613 virtio_set_isr(vdev
, 0x3);
1615 virtio_notify_vector(vdev
, vdev
->config_vector
);
1618 static bool virtio_device_endian_needed(void *opaque
)
1620 VirtIODevice
*vdev
= opaque
;
1622 assert(vdev
->device_endian
!= VIRTIO_DEVICE_ENDIAN_UNKNOWN
);
1623 if (!virtio_vdev_has_feature(vdev
, VIRTIO_F_VERSION_1
)) {
1624 return vdev
->device_endian
!= virtio_default_endian();
1626 /* Devices conforming to VIRTIO 1.0 or later are always LE. */
1627 return vdev
->device_endian
!= VIRTIO_DEVICE_ENDIAN_LITTLE
;
1630 static bool virtio_64bit_features_needed(void *opaque
)
1632 VirtIODevice
*vdev
= opaque
;
1634 return (vdev
->host_features
>> 32) != 0;
1637 static bool virtio_virtqueue_needed(void *opaque
)
1639 VirtIODevice
*vdev
= opaque
;
1641 return virtio_host_has_feature(vdev
, VIRTIO_F_VERSION_1
);
1644 static bool virtio_ringsize_needed(void *opaque
)
1646 VirtIODevice
*vdev
= opaque
;
1649 for (i
= 0; i
< VIRTIO_QUEUE_MAX
; i
++) {
1650 if (vdev
->vq
[i
].vring
.num
!= vdev
->vq
[i
].vring
.num_default
) {
1657 static bool virtio_extra_state_needed(void *opaque
)
1659 VirtIODevice
*vdev
= opaque
;
1660 BusState
*qbus
= qdev_get_parent_bus(DEVICE(vdev
));
1661 VirtioBusClass
*k
= VIRTIO_BUS_GET_CLASS(qbus
);
1663 return k
->has_extra_state
&&
1664 k
->has_extra_state(qbus
->parent
);
1667 static bool virtio_broken_needed(void *opaque
)
1669 VirtIODevice
*vdev
= opaque
;
1671 return vdev
->broken
;
1674 static const VMStateDescription vmstate_virtqueue
= {
1675 .name
= "virtqueue_state",
1677 .minimum_version_id
= 1,
1678 .fields
= (VMStateField
[]) {
1679 VMSTATE_UINT64(vring
.avail
, struct VirtQueue
),
1680 VMSTATE_UINT64(vring
.used
, struct VirtQueue
),
1681 VMSTATE_END_OF_LIST()
1685 static const VMStateDescription vmstate_virtio_virtqueues
= {
1686 .name
= "virtio/virtqueues",
1688 .minimum_version_id
= 1,
1689 .needed
= &virtio_virtqueue_needed
,
1690 .fields
= (VMStateField
[]) {
1691 VMSTATE_STRUCT_VARRAY_POINTER_KNOWN(vq
, struct VirtIODevice
,
1692 VIRTIO_QUEUE_MAX
, 0, vmstate_virtqueue
, VirtQueue
),
1693 VMSTATE_END_OF_LIST()
1697 static const VMStateDescription vmstate_ringsize
= {
1698 .name
= "ringsize_state",
1700 .minimum_version_id
= 1,
1701 .fields
= (VMStateField
[]) {
1702 VMSTATE_UINT32(vring
.num_default
, struct VirtQueue
),
1703 VMSTATE_END_OF_LIST()
1707 static const VMStateDescription vmstate_virtio_ringsize
= {
1708 .name
= "virtio/ringsize",
1710 .minimum_version_id
= 1,
1711 .needed
= &virtio_ringsize_needed
,
1712 .fields
= (VMStateField
[]) {
1713 VMSTATE_STRUCT_VARRAY_POINTER_KNOWN(vq
, struct VirtIODevice
,
1714 VIRTIO_QUEUE_MAX
, 0, vmstate_ringsize
, VirtQueue
),
1715 VMSTATE_END_OF_LIST()
1719 static int get_extra_state(QEMUFile
*f
, void *pv
, size_t size
,
1720 VMStateField
*field
)
1722 VirtIODevice
*vdev
= pv
;
1723 BusState
*qbus
= qdev_get_parent_bus(DEVICE(vdev
));
1724 VirtioBusClass
*k
= VIRTIO_BUS_GET_CLASS(qbus
);
1726 if (!k
->load_extra_state
) {
1729 return k
->load_extra_state(qbus
->parent
, f
);
1733 static int put_extra_state(QEMUFile
*f
, void *pv
, size_t size
,
1734 VMStateField
*field
, QJSON
*vmdesc
)
1736 VirtIODevice
*vdev
= pv
;
1737 BusState
*qbus
= qdev_get_parent_bus(DEVICE(vdev
));
1738 VirtioBusClass
*k
= VIRTIO_BUS_GET_CLASS(qbus
);
1740 k
->save_extra_state(qbus
->parent
, f
);
1744 static const VMStateInfo vmstate_info_extra_state
= {
1745 .name
= "virtqueue_extra_state",
1746 .get
= get_extra_state
,
1747 .put
= put_extra_state
,
1750 static const VMStateDescription vmstate_virtio_extra_state
= {
1751 .name
= "virtio/extra_state",
1753 .minimum_version_id
= 1,
1754 .needed
= &virtio_extra_state_needed
,
1755 .fields
= (VMStateField
[]) {
1757 .name
= "extra_state",
1759 .field_exists
= NULL
,
1761 .info
= &vmstate_info_extra_state
,
1762 .flags
= VMS_SINGLE
,
1765 VMSTATE_END_OF_LIST()
1769 static const VMStateDescription vmstate_virtio_device_endian
= {
1770 .name
= "virtio/device_endian",
1772 .minimum_version_id
= 1,
1773 .needed
= &virtio_device_endian_needed
,
1774 .fields
= (VMStateField
[]) {
1775 VMSTATE_UINT8(device_endian
, VirtIODevice
),
1776 VMSTATE_END_OF_LIST()
1780 static const VMStateDescription vmstate_virtio_64bit_features
= {
1781 .name
= "virtio/64bit_features",
1783 .minimum_version_id
= 1,
1784 .needed
= &virtio_64bit_features_needed
,
1785 .fields
= (VMStateField
[]) {
1786 VMSTATE_UINT64(guest_features
, VirtIODevice
),
1787 VMSTATE_END_OF_LIST()
1791 static const VMStateDescription vmstate_virtio_broken
= {
1792 .name
= "virtio/broken",
1794 .minimum_version_id
= 1,
1795 .needed
= &virtio_broken_needed
,
1796 .fields
= (VMStateField
[]) {
1797 VMSTATE_BOOL(broken
, VirtIODevice
),
1798 VMSTATE_END_OF_LIST()
1802 static const VMStateDescription vmstate_virtio
= {
1805 .minimum_version_id
= 1,
1806 .minimum_version_id_old
= 1,
1807 .fields
= (VMStateField
[]) {
1808 VMSTATE_END_OF_LIST()
1810 .subsections
= (const VMStateDescription
*[]) {
1811 &vmstate_virtio_device_endian
,
1812 &vmstate_virtio_64bit_features
,
1813 &vmstate_virtio_virtqueues
,
1814 &vmstate_virtio_ringsize
,
1815 &vmstate_virtio_broken
,
1816 &vmstate_virtio_extra_state
,
1821 void virtio_save(VirtIODevice
*vdev
, QEMUFile
*f
)
1823 BusState
*qbus
= qdev_get_parent_bus(DEVICE(vdev
));
1824 VirtioBusClass
*k
= VIRTIO_BUS_GET_CLASS(qbus
);
1825 VirtioDeviceClass
*vdc
= VIRTIO_DEVICE_GET_CLASS(vdev
);
1826 uint32_t guest_features_lo
= (vdev
->guest_features
& 0xffffffff);
1829 if (k
->save_config
) {
1830 k
->save_config(qbus
->parent
, f
);
1833 qemu_put_8s(f
, &vdev
->status
);
1834 qemu_put_8s(f
, &vdev
->isr
);
1835 qemu_put_be16s(f
, &vdev
->queue_sel
);
1836 qemu_put_be32s(f
, &guest_features_lo
);
1837 qemu_put_be32(f
, vdev
->config_len
);
1838 qemu_put_buffer(f
, vdev
->config
, vdev
->config_len
);
1840 for (i
= 0; i
< VIRTIO_QUEUE_MAX
; i
++) {
1841 if (vdev
->vq
[i
].vring
.num
== 0)
1845 qemu_put_be32(f
, i
);
1847 for (i
= 0; i
< VIRTIO_QUEUE_MAX
; i
++) {
1848 if (vdev
->vq
[i
].vring
.num
== 0)
1851 qemu_put_be32(f
, vdev
->vq
[i
].vring
.num
);
1852 if (k
->has_variable_vring_alignment
) {
1853 qemu_put_be32(f
, vdev
->vq
[i
].vring
.align
);
1855 /* XXX virtio-1 devices */
1856 qemu_put_be64(f
, vdev
->vq
[i
].vring
.desc
);
1857 qemu_put_be16s(f
, &vdev
->vq
[i
].last_avail_idx
);
1858 if (k
->save_queue
) {
1859 k
->save_queue(qbus
->parent
, i
, f
);
1863 if (vdc
->save
!= NULL
) {
1868 vmstate_save_state(f
, vdc
->vmsd
, vdev
, NULL
);
1872 vmstate_save_state(f
, &vmstate_virtio
, vdev
, NULL
);
1875 /* A wrapper for use as a VMState .put function */
1876 static int virtio_device_put(QEMUFile
*f
, void *opaque
, size_t size
,
1877 VMStateField
*field
, QJSON
*vmdesc
)
1879 virtio_save(VIRTIO_DEVICE(opaque
), f
);
1884 /* A wrapper for use as a VMState .get function */
1885 static int virtio_device_get(QEMUFile
*f
, void *opaque
, size_t size
,
1886 VMStateField
*field
)
1888 VirtIODevice
*vdev
= VIRTIO_DEVICE(opaque
);
1889 DeviceClass
*dc
= DEVICE_CLASS(VIRTIO_DEVICE_GET_CLASS(vdev
));
1891 return virtio_load(vdev
, f
, dc
->vmsd
->version_id
);
1894 const VMStateInfo virtio_vmstate_info
= {
1896 .get
= virtio_device_get
,
1897 .put
= virtio_device_put
,
1900 static int virtio_set_features_nocheck(VirtIODevice
*vdev
, uint64_t val
)
1902 VirtioDeviceClass
*k
= VIRTIO_DEVICE_GET_CLASS(vdev
);
1903 bool bad
= (val
& ~(vdev
->host_features
)) != 0;
1905 val
&= vdev
->host_features
;
1906 if (k
->set_features
) {
1907 k
->set_features(vdev
, val
);
1909 vdev
->guest_features
= val
;
1910 return bad
? -1 : 0;
1913 int virtio_set_features(VirtIODevice
*vdev
, uint64_t val
)
1916 * The driver must not attempt to set features after feature negotiation
1919 if (vdev
->status
& VIRTIO_CONFIG_S_FEATURES_OK
) {
1922 return virtio_set_features_nocheck(vdev
, val
);
1925 int virtio_load(VirtIODevice
*vdev
, QEMUFile
*f
, int version_id
)
1931 BusState
*qbus
= qdev_get_parent_bus(DEVICE(vdev
));
1932 VirtioBusClass
*k
= VIRTIO_BUS_GET_CLASS(qbus
);
1933 VirtioDeviceClass
*vdc
= VIRTIO_DEVICE_GET_CLASS(vdev
);
1936 * We poison the endianness to ensure it does not get used before
1937 * subsections have been loaded.
1939 vdev
->device_endian
= VIRTIO_DEVICE_ENDIAN_UNKNOWN
;
1941 if (k
->load_config
) {
1942 ret
= k
->load_config(qbus
->parent
, f
);
1947 qemu_get_8s(f
, &vdev
->status
);
1948 qemu_get_8s(f
, &vdev
->isr
);
1949 qemu_get_be16s(f
, &vdev
->queue_sel
);
1950 if (vdev
->queue_sel
>= VIRTIO_QUEUE_MAX
) {
1953 qemu_get_be32s(f
, &features
);
1956 * Temporarily set guest_features low bits - needed by
1957 * virtio net load code testing for VIRTIO_NET_F_CTRL_GUEST_OFFLOADS
1958 * VIRTIO_NET_F_GUEST_ANNOUNCE and VIRTIO_NET_F_CTRL_VQ.
1960 * Note: devices should always test host features in future - don't create
1961 * new dependencies like this.
1963 vdev
->guest_features
= features
;
1965 config_len
= qemu_get_be32(f
);
1968 * There are cases where the incoming config can be bigger or smaller
1969 * than what we have; so load what we have space for, and skip
1970 * any excess that's in the stream.
1972 qemu_get_buffer(f
, vdev
->config
, MIN(config_len
, vdev
->config_len
));
1974 while (config_len
> vdev
->config_len
) {
1979 num
= qemu_get_be32(f
);
1981 if (num
> VIRTIO_QUEUE_MAX
) {
1982 error_report("Invalid number of virtqueues: 0x%x", num
);
1986 for (i
= 0; i
< num
; i
++) {
1987 vdev
->vq
[i
].vring
.num
= qemu_get_be32(f
);
1988 if (k
->has_variable_vring_alignment
) {
1989 vdev
->vq
[i
].vring
.align
= qemu_get_be32(f
);
1991 vdev
->vq
[i
].vring
.desc
= qemu_get_be64(f
);
1992 qemu_get_be16s(f
, &vdev
->vq
[i
].last_avail_idx
);
1993 vdev
->vq
[i
].signalled_used_valid
= false;
1994 vdev
->vq
[i
].notification
= true;
1996 if (vdev
->vq
[i
].vring
.desc
) {
1997 /* XXX virtio-1 devices */
1998 virtio_queue_update_rings(vdev
, i
);
1999 } else if (vdev
->vq
[i
].last_avail_idx
) {
2000 error_report("VQ %d address 0x0 "
2001 "inconsistent with Host index 0x%x",
2002 i
, vdev
->vq
[i
].last_avail_idx
);
2005 if (k
->load_queue
) {
2006 ret
= k
->load_queue(qbus
->parent
, i
, f
);
2012 virtio_notify_vector(vdev
, VIRTIO_NO_VECTOR
);
2014 if (vdc
->load
!= NULL
) {
2015 ret
= vdc
->load(vdev
, f
, version_id
);
2022 ret
= vmstate_load_state(f
, vdc
->vmsd
, vdev
, version_id
);
2029 ret
= vmstate_load_state(f
, &vmstate_virtio
, vdev
, 1);
2034 if (vdev
->device_endian
== VIRTIO_DEVICE_ENDIAN_UNKNOWN
) {
2035 vdev
->device_endian
= virtio_default_endian();
2038 if (virtio_64bit_features_needed(vdev
)) {
2040 * Subsection load filled vdev->guest_features. Run them
2041 * through virtio_set_features to sanity-check them against
2044 uint64_t features64
= vdev
->guest_features
;
2045 if (virtio_set_features_nocheck(vdev
, features64
) < 0) {
2046 error_report("Features 0x%" PRIx64
" unsupported. "
2047 "Allowed features: 0x%" PRIx64
,
2048 features64
, vdev
->host_features
);
2052 if (virtio_set_features_nocheck(vdev
, features
) < 0) {
2053 error_report("Features 0x%x unsupported. "
2054 "Allowed features: 0x%" PRIx64
,
2055 features
, vdev
->host_features
);
2061 for (i
= 0; i
< num
; i
++) {
2062 if (vdev
->vq
[i
].vring
.desc
) {
2064 nheads
= vring_avail_idx(&vdev
->vq
[i
]) - vdev
->vq
[i
].last_avail_idx
;
2065 /* Check it isn't doing strange things with descriptor numbers. */
2066 if (nheads
> vdev
->vq
[i
].vring
.num
) {
2067 error_report("VQ %d size 0x%x Guest index 0x%x "
2068 "inconsistent with Host index 0x%x: delta 0x%x",
2069 i
, vdev
->vq
[i
].vring
.num
,
2070 vring_avail_idx(&vdev
->vq
[i
]),
2071 vdev
->vq
[i
].last_avail_idx
, nheads
);
2074 vdev
->vq
[i
].used_idx
= vring_used_idx(&vdev
->vq
[i
]);
2075 vdev
->vq
[i
].shadow_avail_idx
= vring_avail_idx(&vdev
->vq
[i
]);
2078 * Some devices migrate VirtQueueElements that have been popped
2079 * from the avail ring but not yet returned to the used ring.
2080 * Since max ring size < UINT16_MAX it's safe to use modulo
2081 * UINT16_MAX + 1 subtraction.
2083 vdev
->vq
[i
].inuse
= (uint16_t)(vdev
->vq
[i
].last_avail_idx
-
2084 vdev
->vq
[i
].used_idx
);
2085 if (vdev
->vq
[i
].inuse
> vdev
->vq
[i
].vring
.num
) {
2086 error_report("VQ %d size 0x%x < last_avail_idx 0x%x - "
2088 i
, vdev
->vq
[i
].vring
.num
,
2089 vdev
->vq
[i
].last_avail_idx
,
2090 vdev
->vq
[i
].used_idx
);
2100 void virtio_cleanup(VirtIODevice
*vdev
)
2102 qemu_del_vm_change_state_handler(vdev
->vmstate
);
2105 static void virtio_vmstate_change(void *opaque
, int running
, RunState state
)
2107 VirtIODevice
*vdev
= opaque
;
2108 BusState
*qbus
= qdev_get_parent_bus(DEVICE(vdev
));
2109 VirtioBusClass
*k
= VIRTIO_BUS_GET_CLASS(qbus
);
2110 bool backend_run
= running
&& (vdev
->status
& VIRTIO_CONFIG_S_DRIVER_OK
);
2111 vdev
->vm_running
= running
;
2114 virtio_set_status(vdev
, vdev
->status
);
2117 if (k
->vmstate_change
) {
2118 k
->vmstate_change(qbus
->parent
, backend_run
);
2122 virtio_set_status(vdev
, vdev
->status
);
2126 void virtio_instance_init_common(Object
*proxy_obj
, void *data
,
2127 size_t vdev_size
, const char *vdev_name
)
2129 DeviceState
*vdev
= data
;
2131 object_initialize(vdev
, vdev_size
, vdev_name
);
2132 object_property_add_child(proxy_obj
, "virtio-backend", OBJECT(vdev
), NULL
);
2133 object_unref(OBJECT(vdev
));
2134 qdev_alias_all_properties(vdev
, proxy_obj
);
2137 void virtio_init(VirtIODevice
*vdev
, const char *name
,
2138 uint16_t device_id
, size_t config_size
)
2140 BusState
*qbus
= qdev_get_parent_bus(DEVICE(vdev
));
2141 VirtioBusClass
*k
= VIRTIO_BUS_GET_CLASS(qbus
);
2143 int nvectors
= k
->query_nvectors
? k
->query_nvectors(qbus
->parent
) : 0;
2146 vdev
->vector_queues
=
2147 g_malloc0(sizeof(*vdev
->vector_queues
) * nvectors
);
2150 vdev
->device_id
= device_id
;
2152 atomic_set(&vdev
->isr
, 0);
2153 vdev
->queue_sel
= 0;
2154 vdev
->config_vector
= VIRTIO_NO_VECTOR
;
2155 vdev
->vq
= g_malloc0(sizeof(VirtQueue
) * VIRTIO_QUEUE_MAX
);
2156 vdev
->vm_running
= runstate_is_running();
2157 vdev
->broken
= false;
2158 for (i
= 0; i
< VIRTIO_QUEUE_MAX
; i
++) {
2159 vdev
->vq
[i
].vector
= VIRTIO_NO_VECTOR
;
2160 vdev
->vq
[i
].vdev
= vdev
;
2161 vdev
->vq
[i
].queue_index
= i
;
2165 vdev
->config_len
= config_size
;
2166 if (vdev
->config_len
) {
2167 vdev
->config
= g_malloc0(config_size
);
2169 vdev
->config
= NULL
;
2171 vdev
->vmstate
= qemu_add_vm_change_state_handler(virtio_vmstate_change
,
2173 vdev
->device_endian
= virtio_default_endian();
2174 vdev
->use_guest_notifier_mask
= true;
2177 hwaddr
virtio_queue_get_desc_addr(VirtIODevice
*vdev
, int n
)
2179 return vdev
->vq
[n
].vring
.desc
;
2182 hwaddr
virtio_queue_get_avail_addr(VirtIODevice
*vdev
, int n
)
2184 return vdev
->vq
[n
].vring
.avail
;
2187 hwaddr
virtio_queue_get_used_addr(VirtIODevice
*vdev
, int n
)
2189 return vdev
->vq
[n
].vring
.used
;
2192 hwaddr
virtio_queue_get_desc_size(VirtIODevice
*vdev
, int n
)
2194 return sizeof(VRingDesc
) * vdev
->vq
[n
].vring
.num
;
2197 hwaddr
virtio_queue_get_avail_size(VirtIODevice
*vdev
, int n
)
2199 return offsetof(VRingAvail
, ring
) +
2200 sizeof(uint16_t) * vdev
->vq
[n
].vring
.num
;
2203 hwaddr
virtio_queue_get_used_size(VirtIODevice
*vdev
, int n
)
2205 return offsetof(VRingUsed
, ring
) +
2206 sizeof(VRingUsedElem
) * vdev
->vq
[n
].vring
.num
;
2209 uint16_t virtio_queue_get_last_avail_idx(VirtIODevice
*vdev
, int n
)
2211 return vdev
->vq
[n
].last_avail_idx
;
2214 void virtio_queue_set_last_avail_idx(VirtIODevice
*vdev
, int n
, uint16_t idx
)
2216 vdev
->vq
[n
].last_avail_idx
= idx
;
2217 vdev
->vq
[n
].shadow_avail_idx
= idx
;
2220 void virtio_queue_update_used_idx(VirtIODevice
*vdev
, int n
)
2223 if (vdev
->vq
[n
].vring
.desc
) {
2224 vdev
->vq
[n
].used_idx
= vring_used_idx(&vdev
->vq
[n
]);
2229 void virtio_queue_invalidate_signalled_used(VirtIODevice
*vdev
, int n
)
2231 vdev
->vq
[n
].signalled_used_valid
= false;
2234 VirtQueue
*virtio_get_queue(VirtIODevice
*vdev
, int n
)
2236 return vdev
->vq
+ n
;
2239 uint16_t virtio_get_queue_index(VirtQueue
*vq
)
2241 return vq
->queue_index
;
2244 static void virtio_queue_guest_notifier_read(EventNotifier
*n
)
2246 VirtQueue
*vq
= container_of(n
, VirtQueue
, guest_notifier
);
2247 if (event_notifier_test_and_clear(n
)) {
2252 void virtio_queue_set_guest_notifier_fd_handler(VirtQueue
*vq
, bool assign
,
2255 if (assign
&& !with_irqfd
) {
2256 event_notifier_set_handler(&vq
->guest_notifier
,
2257 virtio_queue_guest_notifier_read
);
2259 event_notifier_set_handler(&vq
->guest_notifier
, NULL
);
2262 /* Test and clear notifier before closing it,
2263 * in case poll callback didn't have time to run. */
2264 virtio_queue_guest_notifier_read(&vq
->guest_notifier
);
2268 EventNotifier
*virtio_queue_get_guest_notifier(VirtQueue
*vq
)
2270 return &vq
->guest_notifier
;
2273 static void virtio_queue_host_notifier_aio_read(EventNotifier
*n
)
2275 VirtQueue
*vq
= container_of(n
, VirtQueue
, host_notifier
);
2276 if (event_notifier_test_and_clear(n
)) {
2277 virtio_queue_notify_aio_vq(vq
);
2281 static void virtio_queue_host_notifier_aio_poll_begin(EventNotifier
*n
)
2283 VirtQueue
*vq
= container_of(n
, VirtQueue
, host_notifier
);
2285 virtio_queue_set_notification(vq
, 0);
2288 static bool virtio_queue_host_notifier_aio_poll(void *opaque
)
2290 EventNotifier
*n
= opaque
;
2291 VirtQueue
*vq
= container_of(n
, VirtQueue
, host_notifier
);
2294 if (!vq
->vring
.desc
|| virtio_queue_empty(vq
)) {
2298 progress
= virtio_queue_notify_aio_vq(vq
);
2300 /* In case the handler function re-enabled notifications */
2301 virtio_queue_set_notification(vq
, 0);
2305 static void virtio_queue_host_notifier_aio_poll_end(EventNotifier
*n
)
2307 VirtQueue
*vq
= container_of(n
, VirtQueue
, host_notifier
);
2309 /* Caller polls once more after this to catch requests that race with us */
2310 virtio_queue_set_notification(vq
, 1);
2313 void virtio_queue_aio_set_host_notifier_handler(VirtQueue
*vq
, AioContext
*ctx
,
2314 VirtIOHandleAIOOutput handle_output
)
2316 if (handle_output
) {
2317 vq
->handle_aio_output
= handle_output
;
2318 aio_set_event_notifier(ctx
, &vq
->host_notifier
, true,
2319 virtio_queue_host_notifier_aio_read
,
2320 virtio_queue_host_notifier_aio_poll
);
2321 aio_set_event_notifier_poll(ctx
, &vq
->host_notifier
,
2322 virtio_queue_host_notifier_aio_poll_begin
,
2323 virtio_queue_host_notifier_aio_poll_end
);
2325 aio_set_event_notifier(ctx
, &vq
->host_notifier
, true, NULL
, NULL
);
2326 /* Test and clear notifier before after disabling event,
2327 * in case poll callback didn't have time to run. */
2328 virtio_queue_host_notifier_aio_read(&vq
->host_notifier
);
2329 vq
->handle_aio_output
= NULL
;
2333 void virtio_queue_host_notifier_read(EventNotifier
*n
)
2335 VirtQueue
*vq
= container_of(n
, VirtQueue
, host_notifier
);
2336 if (event_notifier_test_and_clear(n
)) {
2337 virtio_queue_notify_vq(vq
);
2341 EventNotifier
*virtio_queue_get_host_notifier(VirtQueue
*vq
)
2343 return &vq
->host_notifier
;
2346 void virtio_device_set_child_bus_name(VirtIODevice
*vdev
, char *bus_name
)
2348 g_free(vdev
->bus_name
);
2349 vdev
->bus_name
= g_strdup(bus_name
);
2352 void GCC_FMT_ATTR(2, 3) virtio_error(VirtIODevice
*vdev
, const char *fmt
, ...)
2357 error_vreport(fmt
, ap
);
2360 vdev
->broken
= true;
2362 if (virtio_vdev_has_feature(vdev
, VIRTIO_F_VERSION_1
)) {
2363 virtio_set_status(vdev
, vdev
->status
| VIRTIO_CONFIG_S_NEEDS_RESET
);
2364 virtio_notify_config(vdev
);
2368 static void virtio_memory_listener_commit(MemoryListener
*listener
)
2370 VirtIODevice
*vdev
= container_of(listener
, VirtIODevice
, listener
);
2373 for (i
= 0; i
< VIRTIO_QUEUE_MAX
; i
++) {
2374 if (vdev
->vq
[i
].vring
.num
== 0) {
2377 virtio_init_region_cache(vdev
, i
);
2381 static void virtio_device_realize(DeviceState
*dev
, Error
**errp
)
2383 VirtIODevice
*vdev
= VIRTIO_DEVICE(dev
);
2384 VirtioDeviceClass
*vdc
= VIRTIO_DEVICE_GET_CLASS(dev
);
2387 /* Devices should either use vmsd or the load/save methods */
2388 assert(!vdc
->vmsd
|| !vdc
->load
);
2390 if (vdc
->realize
!= NULL
) {
2391 vdc
->realize(dev
, &err
);
2393 error_propagate(errp
, err
);
2398 virtio_bus_device_plugged(vdev
, &err
);
2400 error_propagate(errp
, err
);
2404 vdev
->listener
.commit
= virtio_memory_listener_commit
;
2405 memory_listener_register(&vdev
->listener
, vdev
->dma_as
);
2408 static void virtio_device_unrealize(DeviceState
*dev
, Error
**errp
)
2410 VirtIODevice
*vdev
= VIRTIO_DEVICE(dev
);
2411 VirtioDeviceClass
*vdc
= VIRTIO_DEVICE_GET_CLASS(dev
);
2414 virtio_bus_device_unplugged(vdev
);
2416 if (vdc
->unrealize
!= NULL
) {
2417 vdc
->unrealize(dev
, &err
);
2419 error_propagate(errp
, err
);
2424 g_free(vdev
->bus_name
);
2425 vdev
->bus_name
= NULL
;
2428 static void virtio_device_free_virtqueues(VirtIODevice
*vdev
)
2435 for (i
= 0; i
< VIRTIO_QUEUE_MAX
; i
++) {
2436 VRingMemoryRegionCaches
*caches
;
2437 if (vdev
->vq
[i
].vring
.num
== 0) {
2440 caches
= atomic_read(&vdev
->vq
[i
].vring
.caches
);
2441 atomic_set(&vdev
->vq
[i
].vring
.caches
, NULL
);
2442 virtio_free_region_cache(caches
);
2447 static void virtio_device_instance_finalize(Object
*obj
)
2449 VirtIODevice
*vdev
= VIRTIO_DEVICE(obj
);
2451 memory_listener_unregister(&vdev
->listener
);
2452 virtio_device_free_virtqueues(vdev
);
2454 g_free(vdev
->config
);
2455 g_free(vdev
->vector_queues
);
2458 static Property virtio_properties
[] = {
2459 DEFINE_VIRTIO_COMMON_FEATURES(VirtIODevice
, host_features
),
2460 DEFINE_PROP_END_OF_LIST(),
2463 static int virtio_device_start_ioeventfd_impl(VirtIODevice
*vdev
)
2465 VirtioBusState
*qbus
= VIRTIO_BUS(qdev_get_parent_bus(DEVICE(vdev
)));
2468 for (n
= 0; n
< VIRTIO_QUEUE_MAX
; n
++) {
2469 VirtQueue
*vq
= &vdev
->vq
[n
];
2470 if (!virtio_queue_get_num(vdev
, n
)) {
2473 r
= virtio_bus_set_host_notifier(qbus
, n
, true);
2478 event_notifier_set_handler(&vq
->host_notifier
,
2479 virtio_queue_host_notifier_read
);
2482 for (n
= 0; n
< VIRTIO_QUEUE_MAX
; n
++) {
2483 /* Kick right away to begin processing requests already in vring */
2484 VirtQueue
*vq
= &vdev
->vq
[n
];
2485 if (!vq
->vring
.num
) {
2488 event_notifier_set(&vq
->host_notifier
);
2494 VirtQueue
*vq
= &vdev
->vq
[n
];
2495 if (!virtio_queue_get_num(vdev
, n
)) {
2499 event_notifier_set_handler(&vq
->host_notifier
, NULL
);
2500 r
= virtio_bus_set_host_notifier(qbus
, n
, false);
2506 int virtio_device_start_ioeventfd(VirtIODevice
*vdev
)
2508 BusState
*qbus
= qdev_get_parent_bus(DEVICE(vdev
));
2509 VirtioBusState
*vbus
= VIRTIO_BUS(qbus
);
2511 return virtio_bus_start_ioeventfd(vbus
);
2514 static void virtio_device_stop_ioeventfd_impl(VirtIODevice
*vdev
)
2516 VirtioBusState
*qbus
= VIRTIO_BUS(qdev_get_parent_bus(DEVICE(vdev
)));
2519 for (n
= 0; n
< VIRTIO_QUEUE_MAX
; n
++) {
2520 VirtQueue
*vq
= &vdev
->vq
[n
];
2522 if (!virtio_queue_get_num(vdev
, n
)) {
2525 event_notifier_set_handler(&vq
->host_notifier
, NULL
);
2526 r
= virtio_bus_set_host_notifier(qbus
, n
, false);
2531 void virtio_device_stop_ioeventfd(VirtIODevice
*vdev
)
2533 BusState
*qbus
= qdev_get_parent_bus(DEVICE(vdev
));
2534 VirtioBusState
*vbus
= VIRTIO_BUS(qbus
);
2536 virtio_bus_stop_ioeventfd(vbus
);
2539 int virtio_device_grab_ioeventfd(VirtIODevice
*vdev
)
2541 BusState
*qbus
= qdev_get_parent_bus(DEVICE(vdev
));
2542 VirtioBusState
*vbus
= VIRTIO_BUS(qbus
);
2544 return virtio_bus_grab_ioeventfd(vbus
);
2547 void virtio_device_release_ioeventfd(VirtIODevice
*vdev
)
2549 BusState
*qbus
= qdev_get_parent_bus(DEVICE(vdev
));
2550 VirtioBusState
*vbus
= VIRTIO_BUS(qbus
);
2552 virtio_bus_release_ioeventfd(vbus
);
2555 static void virtio_device_class_init(ObjectClass
*klass
, void *data
)
2557 /* Set the default value here. */
2558 VirtioDeviceClass
*vdc
= VIRTIO_DEVICE_CLASS(klass
);
2559 DeviceClass
*dc
= DEVICE_CLASS(klass
);
2561 dc
->realize
= virtio_device_realize
;
2562 dc
->unrealize
= virtio_device_unrealize
;
2563 dc
->bus_type
= TYPE_VIRTIO_BUS
;
2564 dc
->props
= virtio_properties
;
2565 vdc
->start_ioeventfd
= virtio_device_start_ioeventfd_impl
;
2566 vdc
->stop_ioeventfd
= virtio_device_stop_ioeventfd_impl
;
2568 vdc
->legacy_features
|= VIRTIO_LEGACY_FEATURES
;
2571 bool virtio_device_ioeventfd_enabled(VirtIODevice
*vdev
)
2573 BusState
*qbus
= qdev_get_parent_bus(DEVICE(vdev
));
2574 VirtioBusState
*vbus
= VIRTIO_BUS(qbus
);
2576 return virtio_bus_ioeventfd_enabled(vbus
);
2579 static const TypeInfo virtio_device_info
= {
2580 .name
= TYPE_VIRTIO_DEVICE
,
2581 .parent
= TYPE_DEVICE
,
2582 .instance_size
= sizeof(VirtIODevice
),
2583 .class_init
= virtio_device_class_init
,
2584 .instance_finalize
= virtio_device_instance_finalize
,
2586 .class_size
= sizeof(VirtioDeviceClass
),
2589 static void virtio_register_types(void)
2591 type_register_static(&virtio_device_info
);
2594 type_init(virtio_register_types
)