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"
28 * The alignment to use between consumer and producer parts of vring.
29 * x86 pagesize again. This is the default, used by transports like PCI
30 * which don't provide a means for the guest to tell the host the alignment.
32 #define VIRTIO_PCI_VRING_ALIGN 4096
34 typedef struct VRingDesc
42 typedef struct VRingAvail
49 typedef struct VRingUsedElem
55 typedef struct VRingUsed
59 VRingUsedElem ring
[0];
65 unsigned int num_default
;
76 /* Next head to pop */
77 uint16_t last_avail_idx
;
79 /* Last avail_idx read from VQ. */
80 uint16_t shadow_avail_idx
;
84 /* Last used index value we have signalled on */
85 uint16_t signalled_used
;
87 /* Last used index value we have signalled on */
88 bool signalled_used_valid
;
90 /* Notification enabled? */
98 VirtIOHandleOutput handle_output
;
99 VirtIOHandleOutput handle_aio_output
;
101 EventNotifier guest_notifier
;
102 EventNotifier host_notifier
;
103 QLIST_ENTRY(VirtQueue
) node
;
106 /* virt queue functions */
107 void virtio_queue_update_rings(VirtIODevice
*vdev
, int n
)
109 VRing
*vring
= &vdev
->vq
[n
].vring
;
112 /* not yet setup -> nothing to do */
115 vring
->avail
= vring
->desc
+ vring
->num
* sizeof(VRingDesc
);
116 vring
->used
= vring_align(vring
->avail
+
117 offsetof(VRingAvail
, ring
[vring
->num
]),
121 static void vring_desc_read(VirtIODevice
*vdev
, VRingDesc
*desc
,
122 hwaddr desc_pa
, int i
)
124 address_space_read(&address_space_memory
, desc_pa
+ i
* sizeof(VRingDesc
),
125 MEMTXATTRS_UNSPECIFIED
, (void *)desc
, sizeof(VRingDesc
));
126 virtio_tswap64s(vdev
, &desc
->addr
);
127 virtio_tswap32s(vdev
, &desc
->len
);
128 virtio_tswap16s(vdev
, &desc
->flags
);
129 virtio_tswap16s(vdev
, &desc
->next
);
132 static inline uint16_t vring_avail_flags(VirtQueue
*vq
)
135 pa
= vq
->vring
.avail
+ offsetof(VRingAvail
, flags
);
136 return virtio_lduw_phys(vq
->vdev
, pa
);
139 static inline uint16_t vring_avail_idx(VirtQueue
*vq
)
142 pa
= vq
->vring
.avail
+ offsetof(VRingAvail
, idx
);
143 vq
->shadow_avail_idx
= virtio_lduw_phys(vq
->vdev
, pa
);
144 return vq
->shadow_avail_idx
;
147 static inline uint16_t vring_avail_ring(VirtQueue
*vq
, int i
)
150 pa
= vq
->vring
.avail
+ offsetof(VRingAvail
, ring
[i
]);
151 return virtio_lduw_phys(vq
->vdev
, pa
);
154 static inline uint16_t vring_get_used_event(VirtQueue
*vq
)
156 return vring_avail_ring(vq
, vq
->vring
.num
);
159 static inline void vring_used_write(VirtQueue
*vq
, VRingUsedElem
*uelem
,
163 virtio_tswap32s(vq
->vdev
, &uelem
->id
);
164 virtio_tswap32s(vq
->vdev
, &uelem
->len
);
165 pa
= vq
->vring
.used
+ offsetof(VRingUsed
, ring
[i
]);
166 address_space_write(&address_space_memory
, pa
, MEMTXATTRS_UNSPECIFIED
,
167 (void *)uelem
, sizeof(VRingUsedElem
));
170 static uint16_t vring_used_idx(VirtQueue
*vq
)
173 pa
= vq
->vring
.used
+ offsetof(VRingUsed
, idx
);
174 return virtio_lduw_phys(vq
->vdev
, pa
);
177 static inline void vring_used_idx_set(VirtQueue
*vq
, uint16_t val
)
180 pa
= vq
->vring
.used
+ offsetof(VRingUsed
, idx
);
181 virtio_stw_phys(vq
->vdev
, pa
, val
);
185 static inline void vring_used_flags_set_bit(VirtQueue
*vq
, int mask
)
187 VirtIODevice
*vdev
= vq
->vdev
;
189 pa
= vq
->vring
.used
+ offsetof(VRingUsed
, flags
);
190 virtio_stw_phys(vdev
, pa
, virtio_lduw_phys(vdev
, pa
) | mask
);
193 static inline void vring_used_flags_unset_bit(VirtQueue
*vq
, int mask
)
195 VirtIODevice
*vdev
= vq
->vdev
;
197 pa
= vq
->vring
.used
+ offsetof(VRingUsed
, flags
);
198 virtio_stw_phys(vdev
, pa
, virtio_lduw_phys(vdev
, pa
) & ~mask
);
201 static inline void vring_set_avail_event(VirtQueue
*vq
, uint16_t val
)
204 if (!vq
->notification
) {
207 pa
= vq
->vring
.used
+ offsetof(VRingUsed
, ring
[vq
->vring
.num
]);
208 virtio_stw_phys(vq
->vdev
, pa
, val
);
211 void virtio_queue_set_notification(VirtQueue
*vq
, int enable
)
213 vq
->notification
= enable
;
214 if (virtio_vdev_has_feature(vq
->vdev
, VIRTIO_RING_F_EVENT_IDX
)) {
215 vring_set_avail_event(vq
, vring_avail_idx(vq
));
217 vring_used_flags_unset_bit(vq
, VRING_USED_F_NO_NOTIFY
);
219 vring_used_flags_set_bit(vq
, VRING_USED_F_NO_NOTIFY
);
222 /* Expose avail event/used flags before caller checks the avail idx. */
227 int virtio_queue_ready(VirtQueue
*vq
)
229 return vq
->vring
.avail
!= 0;
232 /* Fetch avail_idx from VQ memory only when we really need to know if
233 * guest has added some buffers. */
234 int virtio_queue_empty(VirtQueue
*vq
)
236 if (vq
->shadow_avail_idx
!= vq
->last_avail_idx
) {
240 return vring_avail_idx(vq
) == vq
->last_avail_idx
;
243 static void virtqueue_unmap_sg(VirtQueue
*vq
, const VirtQueueElement
*elem
,
250 for (i
= 0; i
< elem
->in_num
; i
++) {
251 size_t size
= MIN(len
- offset
, elem
->in_sg
[i
].iov_len
);
253 cpu_physical_memory_unmap(elem
->in_sg
[i
].iov_base
,
254 elem
->in_sg
[i
].iov_len
,
260 for (i
= 0; i
< elem
->out_num
; i
++)
261 cpu_physical_memory_unmap(elem
->out_sg
[i
].iov_base
,
262 elem
->out_sg
[i
].iov_len
,
263 0, elem
->out_sg
[i
].iov_len
);
266 /* virtqueue_detach_element:
267 * @vq: The #VirtQueue
268 * @elem: The #VirtQueueElement
269 * @len: number of bytes written
271 * Detach the element from the virtqueue. This function is suitable for device
272 * reset or other situations where a #VirtQueueElement is simply freed and will
273 * not be pushed or discarded.
275 void virtqueue_detach_element(VirtQueue
*vq
, const VirtQueueElement
*elem
,
279 virtqueue_unmap_sg(vq
, elem
, len
);
283 * @vq: The #VirtQueue
284 * @elem: The #VirtQueueElement
285 * @len: number of bytes written
287 * Pretend the most recent element wasn't popped from the virtqueue. The next
288 * call to virtqueue_pop() will refetch the element.
290 void virtqueue_unpop(VirtQueue
*vq
, const VirtQueueElement
*elem
,
293 vq
->last_avail_idx
--;
294 virtqueue_detach_element(vq
, elem
, len
);
298 * @vq: The #VirtQueue
299 * @num: Number of elements to push back
301 * Pretend that elements weren't popped from the virtqueue. The next
302 * virtqueue_pop() will refetch the oldest element.
304 * Use virtqueue_unpop() instead if you have a VirtQueueElement.
306 * Returns: true on success, false if @num is greater than the number of in use
309 bool virtqueue_rewind(VirtQueue
*vq
, unsigned int num
)
311 if (num
> vq
->inuse
) {
314 vq
->last_avail_idx
-= num
;
319 void virtqueue_fill(VirtQueue
*vq
, const VirtQueueElement
*elem
,
320 unsigned int len
, unsigned int idx
)
324 trace_virtqueue_fill(vq
, elem
, len
, idx
);
326 virtqueue_unmap_sg(vq
, elem
, len
);
328 if (unlikely(vq
->vdev
->broken
)) {
332 idx
= (idx
+ vq
->used_idx
) % vq
->vring
.num
;
334 uelem
.id
= elem
->index
;
336 vring_used_write(vq
, &uelem
, idx
);
339 void virtqueue_flush(VirtQueue
*vq
, unsigned int count
)
343 if (unlikely(vq
->vdev
->broken
)) {
348 /* Make sure buffer is written before we update index. */
350 trace_virtqueue_flush(vq
, count
);
353 vring_used_idx_set(vq
, new);
355 if (unlikely((int16_t)(new - vq
->signalled_used
) < (uint16_t)(new - old
)))
356 vq
->signalled_used_valid
= false;
359 void virtqueue_push(VirtQueue
*vq
, const VirtQueueElement
*elem
,
362 virtqueue_fill(vq
, elem
, len
, 0);
363 virtqueue_flush(vq
, 1);
366 static int virtqueue_num_heads(VirtQueue
*vq
, unsigned int idx
)
368 uint16_t num_heads
= vring_avail_idx(vq
) - idx
;
370 /* Check it isn't doing very strange things with descriptor numbers. */
371 if (num_heads
> vq
->vring
.num
) {
372 virtio_error(vq
->vdev
, "Guest moved used index from %u to %u",
373 idx
, vq
->shadow_avail_idx
);
376 /* On success, callers read a descriptor at vq->last_avail_idx.
377 * Make sure descriptor read does not bypass avail index read. */
385 static bool virtqueue_get_head(VirtQueue
*vq
, unsigned int idx
,
388 /* Grab the next descriptor number they're advertising, and increment
389 * the index we've seen. */
390 *head
= vring_avail_ring(vq
, idx
% vq
->vring
.num
);
392 /* If their number is silly, that's a fatal mistake. */
393 if (*head
>= vq
->vring
.num
) {
394 virtio_error(vq
->vdev
, "Guest says index %u is available", *head
);
402 VIRTQUEUE_READ_DESC_ERROR
= -1,
403 VIRTQUEUE_READ_DESC_DONE
= 0, /* end of chain */
404 VIRTQUEUE_READ_DESC_MORE
= 1, /* more buffers in chain */
407 static int virtqueue_read_next_desc(VirtIODevice
*vdev
, VRingDesc
*desc
,
408 hwaddr desc_pa
, unsigned int max
,
411 /* If this descriptor says it doesn't chain, we're done. */
412 if (!(desc
->flags
& VRING_DESC_F_NEXT
)) {
413 return VIRTQUEUE_READ_DESC_DONE
;
416 /* Check they're not leading us off end of descriptors. */
418 /* Make sure compiler knows to grab that: we don't want it changing! */
422 virtio_error(vdev
, "Desc next is %u", *next
);
423 return VIRTQUEUE_READ_DESC_ERROR
;
426 vring_desc_read(vdev
, desc
, desc_pa
, *next
);
427 return VIRTQUEUE_READ_DESC_MORE
;
430 void virtqueue_get_avail_bytes(VirtQueue
*vq
, unsigned int *in_bytes
,
431 unsigned int *out_bytes
,
432 unsigned max_in_bytes
, unsigned max_out_bytes
)
435 unsigned int total_bufs
, in_total
, out_total
;
438 idx
= vq
->last_avail_idx
;
440 total_bufs
= in_total
= out_total
= 0;
441 while ((rc
= virtqueue_num_heads(vq
, idx
)) > 0) {
442 VirtIODevice
*vdev
= vq
->vdev
;
443 unsigned int max
, num_bufs
, indirect
= 0;
449 num_bufs
= total_bufs
;
451 if (!virtqueue_get_head(vq
, idx
++, &i
)) {
455 desc_pa
= vq
->vring
.desc
;
456 vring_desc_read(vdev
, &desc
, desc_pa
, i
);
458 if (desc
.flags
& VRING_DESC_F_INDIRECT
) {
459 if (desc
.len
% sizeof(VRingDesc
)) {
460 virtio_error(vdev
, "Invalid size for indirect buffer table");
464 /* If we've got too many, that implies a descriptor loop. */
465 if (num_bufs
>= max
) {
466 virtio_error(vdev
, "Looped descriptor");
470 /* loop over the indirect descriptor table */
472 max
= desc
.len
/ sizeof(VRingDesc
);
475 vring_desc_read(vdev
, &desc
, desc_pa
, i
);
479 /* If we've got too many, that implies a descriptor loop. */
480 if (++num_bufs
> max
) {
481 virtio_error(vdev
, "Looped descriptor");
485 if (desc
.flags
& VRING_DESC_F_WRITE
) {
486 in_total
+= desc
.len
;
488 out_total
+= desc
.len
;
490 if (in_total
>= max_in_bytes
&& out_total
>= max_out_bytes
) {
494 rc
= virtqueue_read_next_desc(vdev
, &desc
, desc_pa
, max
, &i
);
495 } while (rc
== VIRTQUEUE_READ_DESC_MORE
);
497 if (rc
== VIRTQUEUE_READ_DESC_ERROR
) {
502 total_bufs
= num_bufs
;
513 *in_bytes
= in_total
;
516 *out_bytes
= out_total
;
521 in_total
= out_total
= 0;
525 int virtqueue_avail_bytes(VirtQueue
*vq
, unsigned int in_bytes
,
526 unsigned int out_bytes
)
528 unsigned int in_total
, out_total
;
530 virtqueue_get_avail_bytes(vq
, &in_total
, &out_total
, in_bytes
, out_bytes
);
531 return in_bytes
<= in_total
&& out_bytes
<= out_total
;
534 static bool virtqueue_map_desc(VirtIODevice
*vdev
, unsigned int *p_num_sg
,
535 hwaddr
*addr
, struct iovec
*iov
,
536 unsigned int max_num_sg
, bool is_write
,
537 hwaddr pa
, size_t sz
)
540 unsigned num_sg
= *p_num_sg
;
541 assert(num_sg
<= max_num_sg
);
544 virtio_error(vdev
, "virtio: zero sized buffers are not allowed");
551 if (num_sg
== max_num_sg
) {
552 virtio_error(vdev
, "virtio: too many write descriptors in "
557 iov
[num_sg
].iov_base
= cpu_physical_memory_map(pa
, &len
, is_write
);
558 if (!iov
[num_sg
].iov_base
) {
559 virtio_error(vdev
, "virtio: bogus descriptor or out of resources");
563 iov
[num_sg
].iov_len
= len
;
577 /* Only used by error code paths before we have a VirtQueueElement (therefore
578 * virtqueue_unmap_sg() can't be used). Assumes buffers weren't written to
581 static void virtqueue_undo_map_desc(unsigned int out_num
, unsigned int in_num
,
586 for (i
= 0; i
< out_num
+ in_num
; i
++) {
587 int is_write
= i
>= out_num
;
589 cpu_physical_memory_unmap(iov
->iov_base
, iov
->iov_len
, is_write
, 0);
594 static void virtqueue_map_iovec(struct iovec
*sg
, hwaddr
*addr
,
595 unsigned int *num_sg
, unsigned int max_size
,
601 /* Note: this function MUST validate input, some callers
602 * are passing in num_sg values received over the network.
604 /* TODO: teach all callers that this can fail, and return failure instead
606 * When we do, we might be able to re-enable NDEBUG below.
609 #error building with NDEBUG is not supported
611 assert(*num_sg
<= max_size
);
613 for (i
= 0; i
< *num_sg
; i
++) {
615 sg
[i
].iov_base
= cpu_physical_memory_map(addr
[i
], &len
, is_write
);
616 if (!sg
[i
].iov_base
) {
617 error_report("virtio: error trying to map MMIO memory");
620 if (len
!= sg
[i
].iov_len
) {
621 error_report("virtio: unexpected memory split");
627 void virtqueue_map(VirtQueueElement
*elem
)
629 virtqueue_map_iovec(elem
->in_sg
, elem
->in_addr
, &elem
->in_num
,
630 VIRTQUEUE_MAX_SIZE
, 1);
631 virtqueue_map_iovec(elem
->out_sg
, elem
->out_addr
, &elem
->out_num
,
632 VIRTQUEUE_MAX_SIZE
, 0);
635 static void *virtqueue_alloc_element(size_t sz
, unsigned out_num
, unsigned in_num
)
637 VirtQueueElement
*elem
;
638 size_t in_addr_ofs
= QEMU_ALIGN_UP(sz
, __alignof__(elem
->in_addr
[0]));
639 size_t out_addr_ofs
= in_addr_ofs
+ in_num
* sizeof(elem
->in_addr
[0]);
640 size_t out_addr_end
= out_addr_ofs
+ out_num
* sizeof(elem
->out_addr
[0]);
641 size_t in_sg_ofs
= QEMU_ALIGN_UP(out_addr_end
, __alignof__(elem
->in_sg
[0]));
642 size_t out_sg_ofs
= in_sg_ofs
+ in_num
* sizeof(elem
->in_sg
[0]);
643 size_t out_sg_end
= out_sg_ofs
+ out_num
* sizeof(elem
->out_sg
[0]);
645 assert(sz
>= sizeof(VirtQueueElement
));
646 elem
= g_malloc(out_sg_end
);
647 elem
->out_num
= out_num
;
648 elem
->in_num
= in_num
;
649 elem
->in_addr
= (void *)elem
+ in_addr_ofs
;
650 elem
->out_addr
= (void *)elem
+ out_addr_ofs
;
651 elem
->in_sg
= (void *)elem
+ in_sg_ofs
;
652 elem
->out_sg
= (void *)elem
+ out_sg_ofs
;
656 void *virtqueue_pop(VirtQueue
*vq
, size_t sz
)
658 unsigned int i
, head
, max
;
659 hwaddr desc_pa
= vq
->vring
.desc
;
660 VirtIODevice
*vdev
= vq
->vdev
;
661 VirtQueueElement
*elem
;
662 unsigned out_num
, in_num
;
663 hwaddr addr
[VIRTQUEUE_MAX_SIZE
];
664 struct iovec iov
[VIRTQUEUE_MAX_SIZE
];
668 if (unlikely(vdev
->broken
)) {
671 if (virtio_queue_empty(vq
)) {
674 /* Needed after virtio_queue_empty(), see comment in
675 * virtqueue_num_heads(). */
678 /* When we start there are none of either input nor output. */
679 out_num
= in_num
= 0;
683 if (vq
->inuse
>= vq
->vring
.num
) {
684 virtio_error(vdev
, "Virtqueue size exceeded");
688 if (!virtqueue_get_head(vq
, vq
->last_avail_idx
++, &head
)) {
692 if (virtio_vdev_has_feature(vdev
, VIRTIO_RING_F_EVENT_IDX
)) {
693 vring_set_avail_event(vq
, vq
->last_avail_idx
);
697 vring_desc_read(vdev
, &desc
, desc_pa
, i
);
698 if (desc
.flags
& VRING_DESC_F_INDIRECT
) {
699 if (desc
.len
% sizeof(VRingDesc
)) {
700 virtio_error(vdev
, "Invalid size for indirect buffer table");
704 /* loop over the indirect descriptor table */
705 max
= desc
.len
/ sizeof(VRingDesc
);
708 vring_desc_read(vdev
, &desc
, desc_pa
, i
);
711 /* Collect all the descriptors */
715 if (desc
.flags
& VRING_DESC_F_WRITE
) {
716 map_ok
= virtqueue_map_desc(vdev
, &in_num
, addr
+ out_num
,
718 VIRTQUEUE_MAX_SIZE
- out_num
, true,
719 desc
.addr
, desc
.len
);
722 virtio_error(vdev
, "Incorrect order for descriptors");
725 map_ok
= virtqueue_map_desc(vdev
, &out_num
, addr
, iov
,
726 VIRTQUEUE_MAX_SIZE
, false,
727 desc
.addr
, desc
.len
);
733 /* If we've got too many, that implies a descriptor loop. */
734 if ((in_num
+ out_num
) > max
) {
735 virtio_error(vdev
, "Looped descriptor");
739 rc
= virtqueue_read_next_desc(vdev
, &desc
, desc_pa
, max
, &i
);
740 } while (rc
== VIRTQUEUE_READ_DESC_MORE
);
742 if (rc
== VIRTQUEUE_READ_DESC_ERROR
) {
746 /* Now copy what we have collected and mapped */
747 elem
= virtqueue_alloc_element(sz
, out_num
, in_num
);
749 for (i
= 0; i
< out_num
; i
++) {
750 elem
->out_addr
[i
] = addr
[i
];
751 elem
->out_sg
[i
] = iov
[i
];
753 for (i
= 0; i
< in_num
; i
++) {
754 elem
->in_addr
[i
] = addr
[out_num
+ i
];
755 elem
->in_sg
[i
] = iov
[out_num
+ i
];
760 trace_virtqueue_pop(vq
, elem
, elem
->in_num
, elem
->out_num
);
764 virtqueue_undo_map_desc(out_num
, in_num
, iov
);
768 /* Reading and writing a structure directly to QEMUFile is *awful*, but
769 * it is what QEMU has always done by mistake. We can change it sooner
770 * or later by bumping the version number of the affected vm states.
771 * In the meanwhile, since the in-memory layout of VirtQueueElement
772 * has changed, we need to marshal to and from the layout that was
773 * used before the change.
775 typedef struct VirtQueueElementOld
{
777 unsigned int out_num
;
779 hwaddr in_addr
[VIRTQUEUE_MAX_SIZE
];
780 hwaddr out_addr
[VIRTQUEUE_MAX_SIZE
];
781 struct iovec in_sg
[VIRTQUEUE_MAX_SIZE
];
782 struct iovec out_sg
[VIRTQUEUE_MAX_SIZE
];
783 } VirtQueueElementOld
;
785 void *qemu_get_virtqueue_element(QEMUFile
*f
, size_t sz
)
787 VirtQueueElement
*elem
;
788 VirtQueueElementOld data
;
791 qemu_get_buffer(f
, (uint8_t *)&data
, sizeof(VirtQueueElementOld
));
793 elem
= virtqueue_alloc_element(sz
, data
.out_num
, data
.in_num
);
794 elem
->index
= data
.index
;
796 for (i
= 0; i
< elem
->in_num
; i
++) {
797 elem
->in_addr
[i
] = data
.in_addr
[i
];
800 for (i
= 0; i
< elem
->out_num
; i
++) {
801 elem
->out_addr
[i
] = data
.out_addr
[i
];
804 for (i
= 0; i
< elem
->in_num
; i
++) {
805 /* Base is overwritten by virtqueue_map. */
806 elem
->in_sg
[i
].iov_base
= 0;
807 elem
->in_sg
[i
].iov_len
= data
.in_sg
[i
].iov_len
;
810 for (i
= 0; i
< elem
->out_num
; i
++) {
811 /* Base is overwritten by virtqueue_map. */
812 elem
->out_sg
[i
].iov_base
= 0;
813 elem
->out_sg
[i
].iov_len
= data
.out_sg
[i
].iov_len
;
820 void qemu_put_virtqueue_element(QEMUFile
*f
, VirtQueueElement
*elem
)
822 VirtQueueElementOld data
;
825 memset(&data
, 0, sizeof(data
));
826 data
.index
= elem
->index
;
827 data
.in_num
= elem
->in_num
;
828 data
.out_num
= elem
->out_num
;
830 for (i
= 0; i
< elem
->in_num
; i
++) {
831 data
.in_addr
[i
] = elem
->in_addr
[i
];
834 for (i
= 0; i
< elem
->out_num
; i
++) {
835 data
.out_addr
[i
] = elem
->out_addr
[i
];
838 for (i
= 0; i
< elem
->in_num
; i
++) {
839 /* Base is overwritten by virtqueue_map when loading. Do not
840 * save it, as it would leak the QEMU address space layout. */
841 data
.in_sg
[i
].iov_len
= elem
->in_sg
[i
].iov_len
;
844 for (i
= 0; i
< elem
->out_num
; i
++) {
845 /* Do not save iov_base as above. */
846 data
.out_sg
[i
].iov_len
= elem
->out_sg
[i
].iov_len
;
848 qemu_put_buffer(f
, (uint8_t *)&data
, sizeof(VirtQueueElementOld
));
852 static void virtio_notify_vector(VirtIODevice
*vdev
, uint16_t vector
)
854 BusState
*qbus
= qdev_get_parent_bus(DEVICE(vdev
));
855 VirtioBusClass
*k
= VIRTIO_BUS_GET_CLASS(qbus
);
857 if (unlikely(vdev
->broken
)) {
862 k
->notify(qbus
->parent
, vector
);
866 void virtio_update_irq(VirtIODevice
*vdev
)
868 virtio_notify_vector(vdev
, VIRTIO_NO_VECTOR
);
871 static int virtio_validate_features(VirtIODevice
*vdev
)
873 VirtioDeviceClass
*k
= VIRTIO_DEVICE_GET_CLASS(vdev
);
875 if (k
->validate_features
) {
876 return k
->validate_features(vdev
);
882 int virtio_set_status(VirtIODevice
*vdev
, uint8_t val
)
884 VirtioDeviceClass
*k
= VIRTIO_DEVICE_GET_CLASS(vdev
);
885 trace_virtio_set_status(vdev
, val
);
887 if (virtio_vdev_has_feature(vdev
, VIRTIO_F_VERSION_1
)) {
888 if (!(vdev
->status
& VIRTIO_CONFIG_S_FEATURES_OK
) &&
889 val
& VIRTIO_CONFIG_S_FEATURES_OK
) {
890 int ret
= virtio_validate_features(vdev
);
898 k
->set_status(vdev
, val
);
904 bool target_words_bigendian(void);
905 static enum virtio_device_endian
virtio_default_endian(void)
907 if (target_words_bigendian()) {
908 return VIRTIO_DEVICE_ENDIAN_BIG
;
910 return VIRTIO_DEVICE_ENDIAN_LITTLE
;
914 static enum virtio_device_endian
virtio_current_cpu_endian(void)
916 CPUClass
*cc
= CPU_GET_CLASS(current_cpu
);
918 if (cc
->virtio_is_big_endian(current_cpu
)) {
919 return VIRTIO_DEVICE_ENDIAN_BIG
;
921 return VIRTIO_DEVICE_ENDIAN_LITTLE
;
925 void virtio_reset(void *opaque
)
927 VirtIODevice
*vdev
= opaque
;
928 VirtioDeviceClass
*k
= VIRTIO_DEVICE_GET_CLASS(vdev
);
931 virtio_set_status(vdev
, 0);
933 /* Guest initiated reset */
934 vdev
->device_endian
= virtio_current_cpu_endian();
937 vdev
->device_endian
= virtio_default_endian();
944 vdev
->broken
= false;
945 vdev
->guest_features
= 0;
948 atomic_set(&vdev
->isr
, 0);
949 vdev
->config_vector
= VIRTIO_NO_VECTOR
;
950 virtio_notify_vector(vdev
, vdev
->config_vector
);
952 for(i
= 0; i
< VIRTIO_QUEUE_MAX
; i
++) {
953 vdev
->vq
[i
].vring
.desc
= 0;
954 vdev
->vq
[i
].vring
.avail
= 0;
955 vdev
->vq
[i
].vring
.used
= 0;
956 vdev
->vq
[i
].last_avail_idx
= 0;
957 vdev
->vq
[i
].shadow_avail_idx
= 0;
958 vdev
->vq
[i
].used_idx
= 0;
959 virtio_queue_set_vector(vdev
, i
, VIRTIO_NO_VECTOR
);
960 vdev
->vq
[i
].signalled_used
= 0;
961 vdev
->vq
[i
].signalled_used_valid
= false;
962 vdev
->vq
[i
].notification
= true;
963 vdev
->vq
[i
].vring
.num
= vdev
->vq
[i
].vring
.num_default
;
964 vdev
->vq
[i
].inuse
= 0;
968 uint32_t virtio_config_readb(VirtIODevice
*vdev
, uint32_t addr
)
970 VirtioDeviceClass
*k
= VIRTIO_DEVICE_GET_CLASS(vdev
);
973 if (addr
+ sizeof(val
) > vdev
->config_len
) {
977 k
->get_config(vdev
, vdev
->config
);
979 val
= ldub_p(vdev
->config
+ addr
);
983 uint32_t virtio_config_readw(VirtIODevice
*vdev
, uint32_t addr
)
985 VirtioDeviceClass
*k
= VIRTIO_DEVICE_GET_CLASS(vdev
);
988 if (addr
+ sizeof(val
) > vdev
->config_len
) {
992 k
->get_config(vdev
, vdev
->config
);
994 val
= lduw_p(vdev
->config
+ addr
);
998 uint32_t virtio_config_readl(VirtIODevice
*vdev
, uint32_t addr
)
1000 VirtioDeviceClass
*k
= VIRTIO_DEVICE_GET_CLASS(vdev
);
1003 if (addr
+ sizeof(val
) > vdev
->config_len
) {
1004 return (uint32_t)-1;
1007 k
->get_config(vdev
, vdev
->config
);
1009 val
= ldl_p(vdev
->config
+ addr
);
1013 void virtio_config_writeb(VirtIODevice
*vdev
, uint32_t addr
, uint32_t data
)
1015 VirtioDeviceClass
*k
= VIRTIO_DEVICE_GET_CLASS(vdev
);
1018 if (addr
+ sizeof(val
) > vdev
->config_len
) {
1022 stb_p(vdev
->config
+ addr
, val
);
1024 if (k
->set_config
) {
1025 k
->set_config(vdev
, vdev
->config
);
1029 void virtio_config_writew(VirtIODevice
*vdev
, uint32_t addr
, uint32_t data
)
1031 VirtioDeviceClass
*k
= VIRTIO_DEVICE_GET_CLASS(vdev
);
1032 uint16_t val
= data
;
1034 if (addr
+ sizeof(val
) > vdev
->config_len
) {
1038 stw_p(vdev
->config
+ addr
, val
);
1040 if (k
->set_config
) {
1041 k
->set_config(vdev
, vdev
->config
);
1045 void virtio_config_writel(VirtIODevice
*vdev
, uint32_t addr
, uint32_t data
)
1047 VirtioDeviceClass
*k
= VIRTIO_DEVICE_GET_CLASS(vdev
);
1048 uint32_t val
= data
;
1050 if (addr
+ sizeof(val
) > vdev
->config_len
) {
1054 stl_p(vdev
->config
+ addr
, val
);
1056 if (k
->set_config
) {
1057 k
->set_config(vdev
, vdev
->config
);
1061 uint32_t virtio_config_modern_readb(VirtIODevice
*vdev
, uint32_t addr
)
1063 VirtioDeviceClass
*k
= VIRTIO_DEVICE_GET_CLASS(vdev
);
1066 if (addr
+ sizeof(val
) > vdev
->config_len
) {
1067 return (uint32_t)-1;
1070 k
->get_config(vdev
, vdev
->config
);
1072 val
= ldub_p(vdev
->config
+ addr
);
1076 uint32_t virtio_config_modern_readw(VirtIODevice
*vdev
, uint32_t addr
)
1078 VirtioDeviceClass
*k
= VIRTIO_DEVICE_GET_CLASS(vdev
);
1081 if (addr
+ sizeof(val
) > vdev
->config_len
) {
1082 return (uint32_t)-1;
1085 k
->get_config(vdev
, vdev
->config
);
1087 val
= lduw_le_p(vdev
->config
+ addr
);
1091 uint32_t virtio_config_modern_readl(VirtIODevice
*vdev
, uint32_t addr
)
1093 VirtioDeviceClass
*k
= VIRTIO_DEVICE_GET_CLASS(vdev
);
1096 if (addr
+ sizeof(val
) > vdev
->config_len
) {
1097 return (uint32_t)-1;
1100 k
->get_config(vdev
, vdev
->config
);
1102 val
= ldl_le_p(vdev
->config
+ addr
);
1106 void virtio_config_modern_writeb(VirtIODevice
*vdev
,
1107 uint32_t addr
, uint32_t data
)
1109 VirtioDeviceClass
*k
= VIRTIO_DEVICE_GET_CLASS(vdev
);
1112 if (addr
+ sizeof(val
) > vdev
->config_len
) {
1116 stb_p(vdev
->config
+ addr
, val
);
1118 if (k
->set_config
) {
1119 k
->set_config(vdev
, vdev
->config
);
1123 void virtio_config_modern_writew(VirtIODevice
*vdev
,
1124 uint32_t addr
, uint32_t data
)
1126 VirtioDeviceClass
*k
= VIRTIO_DEVICE_GET_CLASS(vdev
);
1127 uint16_t val
= data
;
1129 if (addr
+ sizeof(val
) > vdev
->config_len
) {
1133 stw_le_p(vdev
->config
+ addr
, val
);
1135 if (k
->set_config
) {
1136 k
->set_config(vdev
, vdev
->config
);
1140 void virtio_config_modern_writel(VirtIODevice
*vdev
,
1141 uint32_t addr
, uint32_t data
)
1143 VirtioDeviceClass
*k
= VIRTIO_DEVICE_GET_CLASS(vdev
);
1144 uint32_t val
= data
;
1146 if (addr
+ sizeof(val
) > vdev
->config_len
) {
1150 stl_le_p(vdev
->config
+ addr
, val
);
1152 if (k
->set_config
) {
1153 k
->set_config(vdev
, vdev
->config
);
1157 void virtio_queue_set_addr(VirtIODevice
*vdev
, int n
, hwaddr addr
)
1159 vdev
->vq
[n
].vring
.desc
= addr
;
1160 virtio_queue_update_rings(vdev
, n
);
1163 hwaddr
virtio_queue_get_addr(VirtIODevice
*vdev
, int n
)
1165 return vdev
->vq
[n
].vring
.desc
;
1168 void virtio_queue_set_rings(VirtIODevice
*vdev
, int n
, hwaddr desc
,
1169 hwaddr avail
, hwaddr used
)
1171 vdev
->vq
[n
].vring
.desc
= desc
;
1172 vdev
->vq
[n
].vring
.avail
= avail
;
1173 vdev
->vq
[n
].vring
.used
= used
;
1176 void virtio_queue_set_num(VirtIODevice
*vdev
, int n
, int num
)
1178 /* Don't allow guest to flip queue between existent and
1179 * nonexistent states, or to set it to an invalid size.
1181 if (!!num
!= !!vdev
->vq
[n
].vring
.num
||
1182 num
> VIRTQUEUE_MAX_SIZE
||
1186 vdev
->vq
[n
].vring
.num
= num
;
1189 VirtQueue
*virtio_vector_first_queue(VirtIODevice
*vdev
, uint16_t vector
)
1191 return QLIST_FIRST(&vdev
->vector_queues
[vector
]);
1194 VirtQueue
*virtio_vector_next_queue(VirtQueue
*vq
)
1196 return QLIST_NEXT(vq
, node
);
1199 int virtio_queue_get_num(VirtIODevice
*vdev
, int n
)
1201 return vdev
->vq
[n
].vring
.num
;
1204 int virtio_get_num_queues(VirtIODevice
*vdev
)
1208 for (i
= 0; i
< VIRTIO_QUEUE_MAX
; i
++) {
1209 if (!virtio_queue_get_num(vdev
, i
)) {
1217 void virtio_queue_set_align(VirtIODevice
*vdev
, int n
, int align
)
1219 BusState
*qbus
= qdev_get_parent_bus(DEVICE(vdev
));
1220 VirtioBusClass
*k
= VIRTIO_BUS_GET_CLASS(qbus
);
1222 /* virtio-1 compliant devices cannot change the alignment */
1223 if (virtio_vdev_has_feature(vdev
, VIRTIO_F_VERSION_1
)) {
1224 error_report("tried to modify queue alignment for virtio-1 device");
1227 /* Check that the transport told us it was going to do this
1228 * (so a buggy transport will immediately assert rather than
1229 * silently failing to migrate this state)
1231 assert(k
->has_variable_vring_alignment
);
1233 vdev
->vq
[n
].vring
.align
= align
;
1234 virtio_queue_update_rings(vdev
, n
);
1237 static void virtio_queue_notify_aio_vq(VirtQueue
*vq
)
1239 if (vq
->vring
.desc
&& vq
->handle_aio_output
) {
1240 VirtIODevice
*vdev
= vq
->vdev
;
1242 trace_virtio_queue_notify(vdev
, vq
- vdev
->vq
, vq
);
1243 vq
->handle_aio_output(vdev
, vq
);
1247 static void virtio_queue_notify_vq(VirtQueue
*vq
)
1249 if (vq
->vring
.desc
&& vq
->handle_output
) {
1250 VirtIODevice
*vdev
= vq
->vdev
;
1252 if (unlikely(vdev
->broken
)) {
1256 trace_virtio_queue_notify(vdev
, vq
- vdev
->vq
, vq
);
1257 vq
->handle_output(vdev
, vq
);
1261 void virtio_queue_notify(VirtIODevice
*vdev
, int n
)
1263 virtio_queue_notify_vq(&vdev
->vq
[n
]);
1266 uint16_t virtio_queue_vector(VirtIODevice
*vdev
, int n
)
1268 return n
< VIRTIO_QUEUE_MAX
? vdev
->vq
[n
].vector
:
1272 void virtio_queue_set_vector(VirtIODevice
*vdev
, int n
, uint16_t vector
)
1274 VirtQueue
*vq
= &vdev
->vq
[n
];
1276 if (n
< VIRTIO_QUEUE_MAX
) {
1277 if (vdev
->vector_queues
&&
1278 vdev
->vq
[n
].vector
!= VIRTIO_NO_VECTOR
) {
1279 QLIST_REMOVE(vq
, node
);
1281 vdev
->vq
[n
].vector
= vector
;
1282 if (vdev
->vector_queues
&&
1283 vector
!= VIRTIO_NO_VECTOR
) {
1284 QLIST_INSERT_HEAD(&vdev
->vector_queues
[vector
], vq
, node
);
1289 VirtQueue
*virtio_add_queue(VirtIODevice
*vdev
, int queue_size
,
1290 VirtIOHandleOutput handle_output
)
1294 for (i
= 0; i
< VIRTIO_QUEUE_MAX
; i
++) {
1295 if (vdev
->vq
[i
].vring
.num
== 0)
1299 if (i
== VIRTIO_QUEUE_MAX
|| queue_size
> VIRTQUEUE_MAX_SIZE
)
1302 vdev
->vq
[i
].vring
.num
= queue_size
;
1303 vdev
->vq
[i
].vring
.num_default
= queue_size
;
1304 vdev
->vq
[i
].vring
.align
= VIRTIO_PCI_VRING_ALIGN
;
1305 vdev
->vq
[i
].handle_output
= handle_output
;
1306 vdev
->vq
[i
].handle_aio_output
= NULL
;
1308 return &vdev
->vq
[i
];
1311 void virtio_del_queue(VirtIODevice
*vdev
, int n
)
1313 if (n
< 0 || n
>= VIRTIO_QUEUE_MAX
) {
1317 vdev
->vq
[n
].vring
.num
= 0;
1318 vdev
->vq
[n
].vring
.num_default
= 0;
1321 static void virtio_set_isr(VirtIODevice
*vdev
, int value
)
1323 uint8_t old
= atomic_read(&vdev
->isr
);
1325 /* Do not write ISR if it does not change, so that its cacheline remains
1326 * shared in the common case where the guest does not read it.
1328 if ((old
& value
) != value
) {
1329 atomic_or(&vdev
->isr
, value
);
1333 bool virtio_should_notify(VirtIODevice
*vdev
, VirtQueue
*vq
)
1337 /* We need to expose used array entries before checking used event. */
1339 /* Always notify when queue is empty (when feature acknowledge) */
1340 if (virtio_vdev_has_feature(vdev
, VIRTIO_F_NOTIFY_ON_EMPTY
) &&
1341 !vq
->inuse
&& virtio_queue_empty(vq
)) {
1345 if (!virtio_vdev_has_feature(vdev
, VIRTIO_RING_F_EVENT_IDX
)) {
1346 return !(vring_avail_flags(vq
) & VRING_AVAIL_F_NO_INTERRUPT
);
1349 v
= vq
->signalled_used_valid
;
1350 vq
->signalled_used_valid
= true;
1351 old
= vq
->signalled_used
;
1352 new = vq
->signalled_used
= vq
->used_idx
;
1353 return !v
|| vring_need_event(vring_get_used_event(vq
), new, old
);
1356 void virtio_notify_irqfd(VirtIODevice
*vdev
, VirtQueue
*vq
)
1358 if (!virtio_should_notify(vdev
, vq
)) {
1362 trace_virtio_notify_irqfd(vdev
, vq
);
1365 * virtio spec 1.0 says ISR bit 0 should be ignored with MSI, but
1366 * windows drivers included in virtio-win 1.8.0 (circa 2015) are
1367 * incorrectly polling this bit during crashdump and hibernation
1368 * in MSI mode, causing a hang if this bit is never updated.
1369 * Recent releases of Windows do not really shut down, but rather
1370 * log out and hibernate to make the next startup faster. Hence,
1371 * this manifested as a more serious hang during shutdown with
1373 * Next driver release from 2016 fixed this problem, so working around it
1374 * is not a must, but it's easy to do so let's do it here.
1376 * Note: it's safe to update ISR from any thread as it was switched
1377 * to an atomic operation.
1379 virtio_set_isr(vq
->vdev
, 0x1);
1380 event_notifier_set(&vq
->guest_notifier
);
1383 void virtio_notify(VirtIODevice
*vdev
, VirtQueue
*vq
)
1385 if (!virtio_should_notify(vdev
, vq
)) {
1389 trace_virtio_notify(vdev
, vq
);
1390 virtio_set_isr(vq
->vdev
, 0x1);
1391 virtio_notify_vector(vdev
, vq
->vector
);
1394 void virtio_notify_config(VirtIODevice
*vdev
)
1396 if (!(vdev
->status
& VIRTIO_CONFIG_S_DRIVER_OK
))
1399 virtio_set_isr(vdev
, 0x3);
1401 virtio_notify_vector(vdev
, vdev
->config_vector
);
1404 static bool virtio_device_endian_needed(void *opaque
)
1406 VirtIODevice
*vdev
= opaque
;
1408 assert(vdev
->device_endian
!= VIRTIO_DEVICE_ENDIAN_UNKNOWN
);
1409 if (!virtio_vdev_has_feature(vdev
, VIRTIO_F_VERSION_1
)) {
1410 return vdev
->device_endian
!= virtio_default_endian();
1412 /* Devices conforming to VIRTIO 1.0 or later are always LE. */
1413 return vdev
->device_endian
!= VIRTIO_DEVICE_ENDIAN_LITTLE
;
1416 static bool virtio_64bit_features_needed(void *opaque
)
1418 VirtIODevice
*vdev
= opaque
;
1420 return (vdev
->host_features
>> 32) != 0;
1423 static bool virtio_virtqueue_needed(void *opaque
)
1425 VirtIODevice
*vdev
= opaque
;
1427 return virtio_host_has_feature(vdev
, VIRTIO_F_VERSION_1
);
1430 static bool virtio_ringsize_needed(void *opaque
)
1432 VirtIODevice
*vdev
= opaque
;
1435 for (i
= 0; i
< VIRTIO_QUEUE_MAX
; i
++) {
1436 if (vdev
->vq
[i
].vring
.num
!= vdev
->vq
[i
].vring
.num_default
) {
1443 static bool virtio_extra_state_needed(void *opaque
)
1445 VirtIODevice
*vdev
= opaque
;
1446 BusState
*qbus
= qdev_get_parent_bus(DEVICE(vdev
));
1447 VirtioBusClass
*k
= VIRTIO_BUS_GET_CLASS(qbus
);
1449 return k
->has_extra_state
&&
1450 k
->has_extra_state(qbus
->parent
);
1453 static bool virtio_broken_needed(void *opaque
)
1455 VirtIODevice
*vdev
= opaque
;
1457 return vdev
->broken
;
1460 static const VMStateDescription vmstate_virtqueue
= {
1461 .name
= "virtqueue_state",
1463 .minimum_version_id
= 1,
1464 .fields
= (VMStateField
[]) {
1465 VMSTATE_UINT64(vring
.avail
, struct VirtQueue
),
1466 VMSTATE_UINT64(vring
.used
, struct VirtQueue
),
1467 VMSTATE_END_OF_LIST()
1471 static const VMStateDescription vmstate_virtio_virtqueues
= {
1472 .name
= "virtio/virtqueues",
1474 .minimum_version_id
= 1,
1475 .needed
= &virtio_virtqueue_needed
,
1476 .fields
= (VMStateField
[]) {
1477 VMSTATE_STRUCT_VARRAY_POINTER_KNOWN(vq
, struct VirtIODevice
,
1478 VIRTIO_QUEUE_MAX
, 0, vmstate_virtqueue
, VirtQueue
),
1479 VMSTATE_END_OF_LIST()
1483 static const VMStateDescription vmstate_ringsize
= {
1484 .name
= "ringsize_state",
1486 .minimum_version_id
= 1,
1487 .fields
= (VMStateField
[]) {
1488 VMSTATE_UINT32(vring
.num_default
, struct VirtQueue
),
1489 VMSTATE_END_OF_LIST()
1493 static const VMStateDescription vmstate_virtio_ringsize
= {
1494 .name
= "virtio/ringsize",
1496 .minimum_version_id
= 1,
1497 .needed
= &virtio_ringsize_needed
,
1498 .fields
= (VMStateField
[]) {
1499 VMSTATE_STRUCT_VARRAY_POINTER_KNOWN(vq
, struct VirtIODevice
,
1500 VIRTIO_QUEUE_MAX
, 0, vmstate_ringsize
, VirtQueue
),
1501 VMSTATE_END_OF_LIST()
1505 static int get_extra_state(QEMUFile
*f
, void *pv
, size_t size
)
1507 VirtIODevice
*vdev
= pv
;
1508 BusState
*qbus
= qdev_get_parent_bus(DEVICE(vdev
));
1509 VirtioBusClass
*k
= VIRTIO_BUS_GET_CLASS(qbus
);
1511 if (!k
->load_extra_state
) {
1514 return k
->load_extra_state(qbus
->parent
, f
);
1518 static void put_extra_state(QEMUFile
*f
, void *pv
, size_t size
)
1520 VirtIODevice
*vdev
= pv
;
1521 BusState
*qbus
= qdev_get_parent_bus(DEVICE(vdev
));
1522 VirtioBusClass
*k
= VIRTIO_BUS_GET_CLASS(qbus
);
1524 k
->save_extra_state(qbus
->parent
, f
);
1527 static const VMStateInfo vmstate_info_extra_state
= {
1528 .name
= "virtqueue_extra_state",
1529 .get
= get_extra_state
,
1530 .put
= put_extra_state
,
1533 static const VMStateDescription vmstate_virtio_extra_state
= {
1534 .name
= "virtio/extra_state",
1536 .minimum_version_id
= 1,
1537 .needed
= &virtio_extra_state_needed
,
1538 .fields
= (VMStateField
[]) {
1540 .name
= "extra_state",
1542 .field_exists
= NULL
,
1544 .info
= &vmstate_info_extra_state
,
1545 .flags
= VMS_SINGLE
,
1548 VMSTATE_END_OF_LIST()
1552 static const VMStateDescription vmstate_virtio_device_endian
= {
1553 .name
= "virtio/device_endian",
1555 .minimum_version_id
= 1,
1556 .needed
= &virtio_device_endian_needed
,
1557 .fields
= (VMStateField
[]) {
1558 VMSTATE_UINT8(device_endian
, VirtIODevice
),
1559 VMSTATE_END_OF_LIST()
1563 static const VMStateDescription vmstate_virtio_64bit_features
= {
1564 .name
= "virtio/64bit_features",
1566 .minimum_version_id
= 1,
1567 .needed
= &virtio_64bit_features_needed
,
1568 .fields
= (VMStateField
[]) {
1569 VMSTATE_UINT64(guest_features
, VirtIODevice
),
1570 VMSTATE_END_OF_LIST()
1574 static const VMStateDescription vmstate_virtio_broken
= {
1575 .name
= "virtio/broken",
1577 .minimum_version_id
= 1,
1578 .needed
= &virtio_broken_needed
,
1579 .fields
= (VMStateField
[]) {
1580 VMSTATE_BOOL(broken
, VirtIODevice
),
1581 VMSTATE_END_OF_LIST()
1585 static const VMStateDescription vmstate_virtio
= {
1588 .minimum_version_id
= 1,
1589 .minimum_version_id_old
= 1,
1590 .fields
= (VMStateField
[]) {
1591 VMSTATE_END_OF_LIST()
1593 .subsections
= (const VMStateDescription
*[]) {
1594 &vmstate_virtio_device_endian
,
1595 &vmstate_virtio_64bit_features
,
1596 &vmstate_virtio_virtqueues
,
1597 &vmstate_virtio_ringsize
,
1598 &vmstate_virtio_broken
,
1599 &vmstate_virtio_extra_state
,
1604 void virtio_save(VirtIODevice
*vdev
, QEMUFile
*f
)
1606 BusState
*qbus
= qdev_get_parent_bus(DEVICE(vdev
));
1607 VirtioBusClass
*k
= VIRTIO_BUS_GET_CLASS(qbus
);
1608 VirtioDeviceClass
*vdc
= VIRTIO_DEVICE_GET_CLASS(vdev
);
1609 uint32_t guest_features_lo
= (vdev
->guest_features
& 0xffffffff);
1612 if (k
->save_config
) {
1613 k
->save_config(qbus
->parent
, f
);
1616 qemu_put_8s(f
, &vdev
->status
);
1617 qemu_put_8s(f
, &vdev
->isr
);
1618 qemu_put_be16s(f
, &vdev
->queue_sel
);
1619 qemu_put_be32s(f
, &guest_features_lo
);
1620 qemu_put_be32(f
, vdev
->config_len
);
1621 qemu_put_buffer(f
, vdev
->config
, vdev
->config_len
);
1623 for (i
= 0; i
< VIRTIO_QUEUE_MAX
; i
++) {
1624 if (vdev
->vq
[i
].vring
.num
== 0)
1628 qemu_put_be32(f
, i
);
1630 for (i
= 0; i
< VIRTIO_QUEUE_MAX
; i
++) {
1631 if (vdev
->vq
[i
].vring
.num
== 0)
1634 qemu_put_be32(f
, vdev
->vq
[i
].vring
.num
);
1635 if (k
->has_variable_vring_alignment
) {
1636 qemu_put_be32(f
, vdev
->vq
[i
].vring
.align
);
1638 /* XXX virtio-1 devices */
1639 qemu_put_be64(f
, vdev
->vq
[i
].vring
.desc
);
1640 qemu_put_be16s(f
, &vdev
->vq
[i
].last_avail_idx
);
1641 if (k
->save_queue
) {
1642 k
->save_queue(qbus
->parent
, i
, f
);
1646 if (vdc
->save
!= NULL
) {
1651 vmstate_save_state(f
, vdc
->vmsd
, vdev
, NULL
);
1655 vmstate_save_state(f
, &vmstate_virtio
, vdev
, NULL
);
1658 /* A wrapper for use as a VMState .put function */
1659 static void virtio_device_put(QEMUFile
*f
, void *opaque
, size_t size
)
1661 virtio_save(VIRTIO_DEVICE(opaque
), f
);
1664 /* A wrapper for use as a VMState .get function */
1665 static int virtio_device_get(QEMUFile
*f
, void *opaque
, size_t size
)
1667 VirtIODevice
*vdev
= VIRTIO_DEVICE(opaque
);
1668 DeviceClass
*dc
= DEVICE_CLASS(VIRTIO_DEVICE_GET_CLASS(vdev
));
1670 return virtio_load(vdev
, f
, dc
->vmsd
->version_id
);
1673 const VMStateInfo virtio_vmstate_info
= {
1675 .get
= virtio_device_get
,
1676 .put
= virtio_device_put
,
1679 static int virtio_set_features_nocheck(VirtIODevice
*vdev
, uint64_t val
)
1681 VirtioDeviceClass
*k
= VIRTIO_DEVICE_GET_CLASS(vdev
);
1682 bool bad
= (val
& ~(vdev
->host_features
)) != 0;
1684 val
&= vdev
->host_features
;
1685 if (k
->set_features
) {
1686 k
->set_features(vdev
, val
);
1688 vdev
->guest_features
= val
;
1689 return bad
? -1 : 0;
1692 int virtio_set_features(VirtIODevice
*vdev
, uint64_t val
)
1695 * The driver must not attempt to set features after feature negotiation
1698 if (vdev
->status
& VIRTIO_CONFIG_S_FEATURES_OK
) {
1701 return virtio_set_features_nocheck(vdev
, val
);
1704 int virtio_load(VirtIODevice
*vdev
, QEMUFile
*f
, int version_id
)
1710 BusState
*qbus
= qdev_get_parent_bus(DEVICE(vdev
));
1711 VirtioBusClass
*k
= VIRTIO_BUS_GET_CLASS(qbus
);
1712 VirtioDeviceClass
*vdc
= VIRTIO_DEVICE_GET_CLASS(vdev
);
1715 * We poison the endianness to ensure it does not get used before
1716 * subsections have been loaded.
1718 vdev
->device_endian
= VIRTIO_DEVICE_ENDIAN_UNKNOWN
;
1720 if (k
->load_config
) {
1721 ret
= k
->load_config(qbus
->parent
, f
);
1726 qemu_get_8s(f
, &vdev
->status
);
1727 qemu_get_8s(f
, &vdev
->isr
);
1728 qemu_get_be16s(f
, &vdev
->queue_sel
);
1729 if (vdev
->queue_sel
>= VIRTIO_QUEUE_MAX
) {
1732 qemu_get_be32s(f
, &features
);
1735 * Temporarily set guest_features low bits - needed by
1736 * virtio net load code testing for VIRTIO_NET_F_CTRL_GUEST_OFFLOADS
1737 * VIRTIO_NET_F_GUEST_ANNOUNCE and VIRTIO_NET_F_CTRL_VQ.
1739 * Note: devices should always test host features in future - don't create
1740 * new dependencies like this.
1742 vdev
->guest_features
= features
;
1744 config_len
= qemu_get_be32(f
);
1747 * There are cases where the incoming config can be bigger or smaller
1748 * than what we have; so load what we have space for, and skip
1749 * any excess that's in the stream.
1751 qemu_get_buffer(f
, vdev
->config
, MIN(config_len
, vdev
->config_len
));
1753 while (config_len
> vdev
->config_len
) {
1758 num
= qemu_get_be32(f
);
1760 if (num
> VIRTIO_QUEUE_MAX
) {
1761 error_report("Invalid number of virtqueues: 0x%x", num
);
1765 for (i
= 0; i
< num
; i
++) {
1766 vdev
->vq
[i
].vring
.num
= qemu_get_be32(f
);
1767 if (k
->has_variable_vring_alignment
) {
1768 vdev
->vq
[i
].vring
.align
= qemu_get_be32(f
);
1770 vdev
->vq
[i
].vring
.desc
= qemu_get_be64(f
);
1771 qemu_get_be16s(f
, &vdev
->vq
[i
].last_avail_idx
);
1772 vdev
->vq
[i
].signalled_used_valid
= false;
1773 vdev
->vq
[i
].notification
= true;
1775 if (vdev
->vq
[i
].vring
.desc
) {
1776 /* XXX virtio-1 devices */
1777 virtio_queue_update_rings(vdev
, i
);
1778 } else if (vdev
->vq
[i
].last_avail_idx
) {
1779 error_report("VQ %d address 0x0 "
1780 "inconsistent with Host index 0x%x",
1781 i
, vdev
->vq
[i
].last_avail_idx
);
1784 if (k
->load_queue
) {
1785 ret
= k
->load_queue(qbus
->parent
, i
, f
);
1791 virtio_notify_vector(vdev
, VIRTIO_NO_VECTOR
);
1793 if (vdc
->load
!= NULL
) {
1794 ret
= vdc
->load(vdev
, f
, version_id
);
1801 ret
= vmstate_load_state(f
, vdc
->vmsd
, vdev
, version_id
);
1808 ret
= vmstate_load_state(f
, &vmstate_virtio
, vdev
, 1);
1813 if (vdev
->device_endian
== VIRTIO_DEVICE_ENDIAN_UNKNOWN
) {
1814 vdev
->device_endian
= virtio_default_endian();
1817 if (virtio_64bit_features_needed(vdev
)) {
1819 * Subsection load filled vdev->guest_features. Run them
1820 * through virtio_set_features to sanity-check them against
1823 uint64_t features64
= vdev
->guest_features
;
1824 if (virtio_set_features_nocheck(vdev
, features64
) < 0) {
1825 error_report("Features 0x%" PRIx64
" unsupported. "
1826 "Allowed features: 0x%" PRIx64
,
1827 features64
, vdev
->host_features
);
1831 if (virtio_set_features_nocheck(vdev
, features
) < 0) {
1832 error_report("Features 0x%x unsupported. "
1833 "Allowed features: 0x%" PRIx64
,
1834 features
, vdev
->host_features
);
1839 for (i
= 0; i
< num
; i
++) {
1840 if (vdev
->vq
[i
].vring
.desc
) {
1842 nheads
= vring_avail_idx(&vdev
->vq
[i
]) - vdev
->vq
[i
].last_avail_idx
;
1843 /* Check it isn't doing strange things with descriptor numbers. */
1844 if (nheads
> vdev
->vq
[i
].vring
.num
) {
1845 error_report("VQ %d size 0x%x Guest index 0x%x "
1846 "inconsistent with Host index 0x%x: delta 0x%x",
1847 i
, vdev
->vq
[i
].vring
.num
,
1848 vring_avail_idx(&vdev
->vq
[i
]),
1849 vdev
->vq
[i
].last_avail_idx
, nheads
);
1852 vdev
->vq
[i
].used_idx
= vring_used_idx(&vdev
->vq
[i
]);
1853 vdev
->vq
[i
].shadow_avail_idx
= vring_avail_idx(&vdev
->vq
[i
]);
1856 * Some devices migrate VirtQueueElements that have been popped
1857 * from the avail ring but not yet returned to the used ring.
1859 vdev
->vq
[i
].inuse
= vdev
->vq
[i
].last_avail_idx
-
1860 vdev
->vq
[i
].used_idx
;
1861 if (vdev
->vq
[i
].inuse
> vdev
->vq
[i
].vring
.num
) {
1862 error_report("VQ %d size 0x%x < last_avail_idx 0x%x - "
1864 i
, vdev
->vq
[i
].vring
.num
,
1865 vdev
->vq
[i
].last_avail_idx
,
1866 vdev
->vq
[i
].used_idx
);
1875 void virtio_cleanup(VirtIODevice
*vdev
)
1877 qemu_del_vm_change_state_handler(vdev
->vmstate
);
1878 g_free(vdev
->config
);
1880 g_free(vdev
->vector_queues
);
1883 static void virtio_vmstate_change(void *opaque
, int running
, RunState state
)
1885 VirtIODevice
*vdev
= opaque
;
1886 BusState
*qbus
= qdev_get_parent_bus(DEVICE(vdev
));
1887 VirtioBusClass
*k
= VIRTIO_BUS_GET_CLASS(qbus
);
1888 bool backend_run
= running
&& (vdev
->status
& VIRTIO_CONFIG_S_DRIVER_OK
);
1889 vdev
->vm_running
= running
;
1892 virtio_set_status(vdev
, vdev
->status
);
1895 if (k
->vmstate_change
) {
1896 k
->vmstate_change(qbus
->parent
, backend_run
);
1900 virtio_set_status(vdev
, vdev
->status
);
1904 void virtio_instance_init_common(Object
*proxy_obj
, void *data
,
1905 size_t vdev_size
, const char *vdev_name
)
1907 DeviceState
*vdev
= data
;
1909 object_initialize(vdev
, vdev_size
, vdev_name
);
1910 object_property_add_child(proxy_obj
, "virtio-backend", OBJECT(vdev
), NULL
);
1911 object_unref(OBJECT(vdev
));
1912 qdev_alias_all_properties(vdev
, proxy_obj
);
1915 void virtio_init(VirtIODevice
*vdev
, const char *name
,
1916 uint16_t device_id
, size_t config_size
)
1918 BusState
*qbus
= qdev_get_parent_bus(DEVICE(vdev
));
1919 VirtioBusClass
*k
= VIRTIO_BUS_GET_CLASS(qbus
);
1921 int nvectors
= k
->query_nvectors
? k
->query_nvectors(qbus
->parent
) : 0;
1924 vdev
->vector_queues
=
1925 g_malloc0(sizeof(*vdev
->vector_queues
) * nvectors
);
1928 vdev
->device_id
= device_id
;
1930 atomic_set(&vdev
->isr
, 0);
1931 vdev
->queue_sel
= 0;
1932 vdev
->config_vector
= VIRTIO_NO_VECTOR
;
1933 vdev
->vq
= g_malloc0(sizeof(VirtQueue
) * VIRTIO_QUEUE_MAX
);
1934 vdev
->vm_running
= runstate_is_running();
1935 vdev
->broken
= false;
1936 for (i
= 0; i
< VIRTIO_QUEUE_MAX
; i
++) {
1937 vdev
->vq
[i
].vector
= VIRTIO_NO_VECTOR
;
1938 vdev
->vq
[i
].vdev
= vdev
;
1939 vdev
->vq
[i
].queue_index
= i
;
1943 vdev
->config_len
= config_size
;
1944 if (vdev
->config_len
) {
1945 vdev
->config
= g_malloc0(config_size
);
1947 vdev
->config
= NULL
;
1949 vdev
->vmstate
= qemu_add_vm_change_state_handler(virtio_vmstate_change
,
1951 vdev
->device_endian
= virtio_default_endian();
1952 vdev
->use_guest_notifier_mask
= true;
1955 hwaddr
virtio_queue_get_desc_addr(VirtIODevice
*vdev
, int n
)
1957 return vdev
->vq
[n
].vring
.desc
;
1960 hwaddr
virtio_queue_get_avail_addr(VirtIODevice
*vdev
, int n
)
1962 return vdev
->vq
[n
].vring
.avail
;
1965 hwaddr
virtio_queue_get_used_addr(VirtIODevice
*vdev
, int n
)
1967 return vdev
->vq
[n
].vring
.used
;
1970 hwaddr
virtio_queue_get_desc_size(VirtIODevice
*vdev
, int n
)
1972 return sizeof(VRingDesc
) * vdev
->vq
[n
].vring
.num
;
1975 hwaddr
virtio_queue_get_avail_size(VirtIODevice
*vdev
, int n
)
1977 return offsetof(VRingAvail
, ring
) +
1978 sizeof(uint16_t) * vdev
->vq
[n
].vring
.num
;
1981 hwaddr
virtio_queue_get_used_size(VirtIODevice
*vdev
, int n
)
1983 return offsetof(VRingUsed
, ring
) +
1984 sizeof(VRingUsedElem
) * vdev
->vq
[n
].vring
.num
;
1987 uint16_t virtio_queue_get_last_avail_idx(VirtIODevice
*vdev
, int n
)
1989 return vdev
->vq
[n
].last_avail_idx
;
1992 void virtio_queue_set_last_avail_idx(VirtIODevice
*vdev
, int n
, uint16_t idx
)
1994 vdev
->vq
[n
].last_avail_idx
= idx
;
1995 vdev
->vq
[n
].shadow_avail_idx
= idx
;
1998 void virtio_queue_invalidate_signalled_used(VirtIODevice
*vdev
, int n
)
2000 vdev
->vq
[n
].signalled_used_valid
= false;
2003 VirtQueue
*virtio_get_queue(VirtIODevice
*vdev
, int n
)
2005 return vdev
->vq
+ n
;
2008 uint16_t virtio_get_queue_index(VirtQueue
*vq
)
2010 return vq
->queue_index
;
2013 static void virtio_queue_guest_notifier_read(EventNotifier
*n
)
2015 VirtQueue
*vq
= container_of(n
, VirtQueue
, guest_notifier
);
2016 if (event_notifier_test_and_clear(n
)) {
2017 virtio_notify_vector(vq
->vdev
, vq
->vector
);
2021 void virtio_queue_set_guest_notifier_fd_handler(VirtQueue
*vq
, bool assign
,
2024 if (assign
&& !with_irqfd
) {
2025 event_notifier_set_handler(&vq
->guest_notifier
, false,
2026 virtio_queue_guest_notifier_read
);
2028 event_notifier_set_handler(&vq
->guest_notifier
, false, NULL
);
2031 /* Test and clear notifier before closing it,
2032 * in case poll callback didn't have time to run. */
2033 virtio_queue_guest_notifier_read(&vq
->guest_notifier
);
2037 EventNotifier
*virtio_queue_get_guest_notifier(VirtQueue
*vq
)
2039 return &vq
->guest_notifier
;
2042 static void virtio_queue_host_notifier_aio_read(EventNotifier
*n
)
2044 VirtQueue
*vq
= container_of(n
, VirtQueue
, host_notifier
);
2045 if (event_notifier_test_and_clear(n
)) {
2046 virtio_queue_notify_aio_vq(vq
);
2050 void virtio_queue_aio_set_host_notifier_handler(VirtQueue
*vq
, AioContext
*ctx
,
2051 VirtIOHandleOutput handle_output
)
2053 if (handle_output
) {
2054 vq
->handle_aio_output
= handle_output
;
2055 aio_set_event_notifier(ctx
, &vq
->host_notifier
, true,
2056 virtio_queue_host_notifier_aio_read
);
2058 aio_set_event_notifier(ctx
, &vq
->host_notifier
, true, NULL
);
2059 /* Test and clear notifier before after disabling event,
2060 * in case poll callback didn't have time to run. */
2061 virtio_queue_host_notifier_aio_read(&vq
->host_notifier
);
2062 vq
->handle_aio_output
= NULL
;
2066 void virtio_queue_host_notifier_read(EventNotifier
*n
)
2068 VirtQueue
*vq
= container_of(n
, VirtQueue
, host_notifier
);
2069 if (event_notifier_test_and_clear(n
)) {
2070 virtio_queue_notify_vq(vq
);
2074 EventNotifier
*virtio_queue_get_host_notifier(VirtQueue
*vq
)
2076 return &vq
->host_notifier
;
2079 void virtio_device_set_child_bus_name(VirtIODevice
*vdev
, char *bus_name
)
2081 g_free(vdev
->bus_name
);
2082 vdev
->bus_name
= g_strdup(bus_name
);
2085 void GCC_FMT_ATTR(2, 3) virtio_error(VirtIODevice
*vdev
, const char *fmt
, ...)
2090 error_vreport(fmt
, ap
);
2093 vdev
->broken
= true;
2095 if (virtio_vdev_has_feature(vdev
, VIRTIO_F_VERSION_1
)) {
2096 virtio_set_status(vdev
, vdev
->status
| VIRTIO_CONFIG_S_NEEDS_RESET
);
2097 virtio_notify_config(vdev
);
2101 static void virtio_device_realize(DeviceState
*dev
, Error
**errp
)
2103 VirtIODevice
*vdev
= VIRTIO_DEVICE(dev
);
2104 VirtioDeviceClass
*vdc
= VIRTIO_DEVICE_GET_CLASS(dev
);
2107 /* Devices should either use vmsd or the load/save methods */
2108 assert(!vdc
->vmsd
|| !vdc
->load
);
2110 if (vdc
->realize
!= NULL
) {
2111 vdc
->realize(dev
, &err
);
2113 error_propagate(errp
, err
);
2118 virtio_bus_device_plugged(vdev
, &err
);
2120 error_propagate(errp
, err
);
2125 static void virtio_device_unrealize(DeviceState
*dev
, Error
**errp
)
2127 VirtIODevice
*vdev
= VIRTIO_DEVICE(dev
);
2128 VirtioDeviceClass
*vdc
= VIRTIO_DEVICE_GET_CLASS(dev
);
2131 virtio_bus_device_unplugged(vdev
);
2133 if (vdc
->unrealize
!= NULL
) {
2134 vdc
->unrealize(dev
, &err
);
2136 error_propagate(errp
, err
);
2141 g_free(vdev
->bus_name
);
2142 vdev
->bus_name
= NULL
;
2145 static Property virtio_properties
[] = {
2146 DEFINE_VIRTIO_COMMON_FEATURES(VirtIODevice
, host_features
),
2147 DEFINE_PROP_END_OF_LIST(),
2150 static int virtio_device_start_ioeventfd_impl(VirtIODevice
*vdev
)
2152 VirtioBusState
*qbus
= VIRTIO_BUS(qdev_get_parent_bus(DEVICE(vdev
)));
2155 for (n
= 0; n
< VIRTIO_QUEUE_MAX
; n
++) {
2156 VirtQueue
*vq
= &vdev
->vq
[n
];
2157 if (!virtio_queue_get_num(vdev
, n
)) {
2160 r
= virtio_bus_set_host_notifier(qbus
, n
, true);
2165 event_notifier_set_handler(&vq
->host_notifier
, true,
2166 virtio_queue_host_notifier_read
);
2169 for (n
= 0; n
< VIRTIO_QUEUE_MAX
; n
++) {
2170 /* Kick right away to begin processing requests already in vring */
2171 VirtQueue
*vq
= &vdev
->vq
[n
];
2172 if (!vq
->vring
.num
) {
2175 event_notifier_set(&vq
->host_notifier
);
2181 VirtQueue
*vq
= &vdev
->vq
[n
];
2182 if (!virtio_queue_get_num(vdev
, n
)) {
2186 event_notifier_set_handler(&vq
->host_notifier
, true, NULL
);
2187 r
= virtio_bus_set_host_notifier(qbus
, n
, false);
2193 int virtio_device_start_ioeventfd(VirtIODevice
*vdev
)
2195 BusState
*qbus
= qdev_get_parent_bus(DEVICE(vdev
));
2196 VirtioBusState
*vbus
= VIRTIO_BUS(qbus
);
2198 return virtio_bus_start_ioeventfd(vbus
);
2201 static void virtio_device_stop_ioeventfd_impl(VirtIODevice
*vdev
)
2203 VirtioBusState
*qbus
= VIRTIO_BUS(qdev_get_parent_bus(DEVICE(vdev
)));
2206 for (n
= 0; n
< VIRTIO_QUEUE_MAX
; n
++) {
2207 VirtQueue
*vq
= &vdev
->vq
[n
];
2209 if (!virtio_queue_get_num(vdev
, n
)) {
2212 event_notifier_set_handler(&vq
->host_notifier
, true, NULL
);
2213 r
= virtio_bus_set_host_notifier(qbus
, n
, false);
2218 void virtio_device_stop_ioeventfd(VirtIODevice
*vdev
)
2220 BusState
*qbus
= qdev_get_parent_bus(DEVICE(vdev
));
2221 VirtioBusState
*vbus
= VIRTIO_BUS(qbus
);
2223 virtio_bus_stop_ioeventfd(vbus
);
2226 int virtio_device_grab_ioeventfd(VirtIODevice
*vdev
)
2228 BusState
*qbus
= qdev_get_parent_bus(DEVICE(vdev
));
2229 VirtioBusState
*vbus
= VIRTIO_BUS(qbus
);
2231 return virtio_bus_grab_ioeventfd(vbus
);
2234 void virtio_device_release_ioeventfd(VirtIODevice
*vdev
)
2236 BusState
*qbus
= qdev_get_parent_bus(DEVICE(vdev
));
2237 VirtioBusState
*vbus
= VIRTIO_BUS(qbus
);
2239 virtio_bus_release_ioeventfd(vbus
);
2242 static void virtio_device_class_init(ObjectClass
*klass
, void *data
)
2244 /* Set the default value here. */
2245 VirtioDeviceClass
*vdc
= VIRTIO_DEVICE_CLASS(klass
);
2246 DeviceClass
*dc
= DEVICE_CLASS(klass
);
2248 dc
->realize
= virtio_device_realize
;
2249 dc
->unrealize
= virtio_device_unrealize
;
2250 dc
->bus_type
= TYPE_VIRTIO_BUS
;
2251 dc
->props
= virtio_properties
;
2252 vdc
->start_ioeventfd
= virtio_device_start_ioeventfd_impl
;
2253 vdc
->stop_ioeventfd
= virtio_device_stop_ioeventfd_impl
;
2255 vdc
->legacy_features
|= VIRTIO_LEGACY_FEATURES
;
2258 bool virtio_device_ioeventfd_enabled(VirtIODevice
*vdev
)
2260 BusState
*qbus
= qdev_get_parent_bus(DEVICE(vdev
));
2261 VirtioBusState
*vbus
= VIRTIO_BUS(qbus
);
2263 return virtio_bus_ioeventfd_enabled(vbus
);
2266 static const TypeInfo virtio_device_info
= {
2267 .name
= TYPE_VIRTIO_DEVICE
,
2268 .parent
= TYPE_DEVICE
,
2269 .instance_size
= sizeof(VirtIODevice
),
2270 .class_init
= virtio_device_class_init
,
2272 .class_size
= sizeof(VirtioDeviceClass
),
2275 static void virtio_register_types(void)
2277 type_register_static(&virtio_device_info
);
2280 type_init(virtio_register_types
)