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
;
102 EventNotifier guest_notifier
;
103 EventNotifier host_notifier
;
104 QLIST_ENTRY(VirtQueue
) node
;
107 /* virt queue functions */
108 void virtio_queue_update_rings(VirtIODevice
*vdev
, int n
)
110 VRing
*vring
= &vdev
->vq
[n
].vring
;
113 /* not yet setup -> nothing to do */
116 vring
->avail
= vring
->desc
+ vring
->num
* sizeof(VRingDesc
);
117 vring
->used
= vring_align(vring
->avail
+
118 offsetof(VRingAvail
, ring
[vring
->num
]),
122 static void vring_desc_read(VirtIODevice
*vdev
, VRingDesc
*desc
,
123 hwaddr desc_pa
, int i
)
125 address_space_read(&address_space_memory
, desc_pa
+ i
* sizeof(VRingDesc
),
126 MEMTXATTRS_UNSPECIFIED
, (void *)desc
, sizeof(VRingDesc
));
127 virtio_tswap64s(vdev
, &desc
->addr
);
128 virtio_tswap32s(vdev
, &desc
->len
);
129 virtio_tswap16s(vdev
, &desc
->flags
);
130 virtio_tswap16s(vdev
, &desc
->next
);
133 static inline uint16_t vring_avail_flags(VirtQueue
*vq
)
136 pa
= vq
->vring
.avail
+ offsetof(VRingAvail
, flags
);
137 return virtio_lduw_phys(vq
->vdev
, pa
);
140 static inline uint16_t vring_avail_idx(VirtQueue
*vq
)
143 pa
= vq
->vring
.avail
+ offsetof(VRingAvail
, idx
);
144 vq
->shadow_avail_idx
= virtio_lduw_phys(vq
->vdev
, pa
);
145 return vq
->shadow_avail_idx
;
148 static inline uint16_t vring_avail_ring(VirtQueue
*vq
, int i
)
151 pa
= vq
->vring
.avail
+ offsetof(VRingAvail
, ring
[i
]);
152 return virtio_lduw_phys(vq
->vdev
, pa
);
155 static inline uint16_t vring_get_used_event(VirtQueue
*vq
)
157 return vring_avail_ring(vq
, vq
->vring
.num
);
160 static inline void vring_used_write(VirtQueue
*vq
, VRingUsedElem
*uelem
,
164 virtio_tswap32s(vq
->vdev
, &uelem
->id
);
165 virtio_tswap32s(vq
->vdev
, &uelem
->len
);
166 pa
= vq
->vring
.used
+ offsetof(VRingUsed
, ring
[i
]);
167 address_space_write(&address_space_memory
, pa
, MEMTXATTRS_UNSPECIFIED
,
168 (void *)uelem
, sizeof(VRingUsedElem
));
171 static uint16_t vring_used_idx(VirtQueue
*vq
)
174 pa
= vq
->vring
.used
+ offsetof(VRingUsed
, idx
);
175 return virtio_lduw_phys(vq
->vdev
, pa
);
178 static inline void vring_used_idx_set(VirtQueue
*vq
, uint16_t val
)
181 pa
= vq
->vring
.used
+ offsetof(VRingUsed
, idx
);
182 virtio_stw_phys(vq
->vdev
, pa
, val
);
186 static inline void vring_used_flags_set_bit(VirtQueue
*vq
, int mask
)
188 VirtIODevice
*vdev
= vq
->vdev
;
190 pa
= vq
->vring
.used
+ offsetof(VRingUsed
, flags
);
191 virtio_stw_phys(vdev
, pa
, virtio_lduw_phys(vdev
, pa
) | mask
);
194 static inline void vring_used_flags_unset_bit(VirtQueue
*vq
, int mask
)
196 VirtIODevice
*vdev
= vq
->vdev
;
198 pa
= vq
->vring
.used
+ offsetof(VRingUsed
, flags
);
199 virtio_stw_phys(vdev
, pa
, virtio_lduw_phys(vdev
, pa
) & ~mask
);
202 static inline void vring_set_avail_event(VirtQueue
*vq
, uint16_t val
)
205 if (!vq
->notification
) {
208 pa
= vq
->vring
.used
+ offsetof(VRingUsed
, ring
[vq
->vring
.num
]);
209 virtio_stw_phys(vq
->vdev
, pa
, val
);
212 void virtio_queue_set_notification(VirtQueue
*vq
, int enable
)
214 vq
->notification
= enable
;
215 if (virtio_vdev_has_feature(vq
->vdev
, VIRTIO_RING_F_EVENT_IDX
)) {
216 vring_set_avail_event(vq
, vring_avail_idx(vq
));
218 vring_used_flags_unset_bit(vq
, VRING_USED_F_NO_NOTIFY
);
220 vring_used_flags_set_bit(vq
, VRING_USED_F_NO_NOTIFY
);
223 /* Expose avail event/used flags before caller checks the avail idx. */
228 int virtio_queue_ready(VirtQueue
*vq
)
230 return vq
->vring
.avail
!= 0;
233 /* Fetch avail_idx from VQ memory only when we really need to know if
234 * guest has added some buffers. */
235 int virtio_queue_empty(VirtQueue
*vq
)
237 if (vq
->shadow_avail_idx
!= vq
->last_avail_idx
) {
241 return vring_avail_idx(vq
) == vq
->last_avail_idx
;
244 static void virtqueue_unmap_sg(VirtQueue
*vq
, const VirtQueueElement
*elem
,
251 for (i
= 0; i
< elem
->in_num
; i
++) {
252 size_t size
= MIN(len
- offset
, elem
->in_sg
[i
].iov_len
);
254 cpu_physical_memory_unmap(elem
->in_sg
[i
].iov_base
,
255 elem
->in_sg
[i
].iov_len
,
261 for (i
= 0; i
< elem
->out_num
; i
++)
262 cpu_physical_memory_unmap(elem
->out_sg
[i
].iov_base
,
263 elem
->out_sg
[i
].iov_len
,
264 0, elem
->out_sg
[i
].iov_len
);
267 void virtqueue_discard(VirtQueue
*vq
, const VirtQueueElement
*elem
,
270 vq
->last_avail_idx
--;
271 virtqueue_unmap_sg(vq
, elem
, len
);
274 void virtqueue_fill(VirtQueue
*vq
, const VirtQueueElement
*elem
,
275 unsigned int len
, unsigned int idx
)
279 trace_virtqueue_fill(vq
, elem
, len
, idx
);
281 virtqueue_unmap_sg(vq
, elem
, len
);
283 idx
= (idx
+ vq
->used_idx
) % vq
->vring
.num
;
285 uelem
.id
= elem
->index
;
287 vring_used_write(vq
, &uelem
, idx
);
290 void virtqueue_flush(VirtQueue
*vq
, unsigned int count
)
293 /* Make sure buffer is written before we update index. */
295 trace_virtqueue_flush(vq
, count
);
298 vring_used_idx_set(vq
, new);
300 if (unlikely((int16_t)(new - vq
->signalled_used
) < (uint16_t)(new - old
)))
301 vq
->signalled_used_valid
= false;
304 void virtqueue_push(VirtQueue
*vq
, const VirtQueueElement
*elem
,
307 virtqueue_fill(vq
, elem
, len
, 0);
308 virtqueue_flush(vq
, 1);
311 static int virtqueue_num_heads(VirtQueue
*vq
, unsigned int idx
)
313 uint16_t num_heads
= vring_avail_idx(vq
) - idx
;
315 /* Check it isn't doing very strange things with descriptor numbers. */
316 if (num_heads
> vq
->vring
.num
) {
317 error_report("Guest moved used index from %u to %u",
318 idx
, vq
->shadow_avail_idx
);
321 /* On success, callers read a descriptor at vq->last_avail_idx.
322 * Make sure descriptor read does not bypass avail index read. */
330 static unsigned int virtqueue_get_head(VirtQueue
*vq
, unsigned int idx
)
334 /* Grab the next descriptor number they're advertising, and increment
335 * the index we've seen. */
336 head
= vring_avail_ring(vq
, idx
% vq
->vring
.num
);
338 /* If their number is silly, that's a fatal mistake. */
339 if (head
>= vq
->vring
.num
) {
340 error_report("Guest says index %u is available", head
);
347 static unsigned virtqueue_read_next_desc(VirtIODevice
*vdev
, VRingDesc
*desc
,
348 hwaddr desc_pa
, unsigned int max
)
352 /* If this descriptor says it doesn't chain, we're done. */
353 if (!(desc
->flags
& VRING_DESC_F_NEXT
)) {
357 /* Check they're not leading us off end of descriptors. */
359 /* Make sure compiler knows to grab that: we don't want it changing! */
363 error_report("Desc next is %u", next
);
367 vring_desc_read(vdev
, desc
, desc_pa
, next
);
371 void virtqueue_get_avail_bytes(VirtQueue
*vq
, unsigned int *in_bytes
,
372 unsigned int *out_bytes
,
373 unsigned max_in_bytes
, unsigned max_out_bytes
)
376 unsigned int total_bufs
, in_total
, out_total
;
378 idx
= vq
->last_avail_idx
;
380 total_bufs
= in_total
= out_total
= 0;
381 while (virtqueue_num_heads(vq
, idx
)) {
382 VirtIODevice
*vdev
= vq
->vdev
;
383 unsigned int max
, num_bufs
, indirect
= 0;
389 num_bufs
= total_bufs
;
390 i
= virtqueue_get_head(vq
, idx
++);
391 desc_pa
= vq
->vring
.desc
;
392 vring_desc_read(vdev
, &desc
, desc_pa
, i
);
394 if (desc
.flags
& VRING_DESC_F_INDIRECT
) {
395 if (desc
.len
% sizeof(VRingDesc
)) {
396 error_report("Invalid size for indirect buffer table");
400 /* If we've got too many, that implies a descriptor loop. */
401 if (num_bufs
>= max
) {
402 error_report("Looped descriptor");
406 /* loop over the indirect descriptor table */
408 max
= desc
.len
/ sizeof(VRingDesc
);
411 vring_desc_read(vdev
, &desc
, desc_pa
, i
);
415 /* If we've got too many, that implies a descriptor loop. */
416 if (++num_bufs
> max
) {
417 error_report("Looped descriptor");
421 if (desc
.flags
& VRING_DESC_F_WRITE
) {
422 in_total
+= desc
.len
;
424 out_total
+= desc
.len
;
426 if (in_total
>= max_in_bytes
&& out_total
>= max_out_bytes
) {
429 } while ((i
= virtqueue_read_next_desc(vdev
, &desc
, desc_pa
, max
)) != max
);
432 total_bufs
= num_bufs
;
438 *in_bytes
= in_total
;
441 *out_bytes
= out_total
;
445 int virtqueue_avail_bytes(VirtQueue
*vq
, unsigned int in_bytes
,
446 unsigned int out_bytes
)
448 unsigned int in_total
, out_total
;
450 virtqueue_get_avail_bytes(vq
, &in_total
, &out_total
, in_bytes
, out_bytes
);
451 return in_bytes
<= in_total
&& out_bytes
<= out_total
;
454 static void virtqueue_map_desc(unsigned int *p_num_sg
, hwaddr
*addr
, struct iovec
*iov
,
455 unsigned int max_num_sg
, bool is_write
,
456 hwaddr pa
, size_t sz
)
458 unsigned num_sg
= *p_num_sg
;
459 assert(num_sg
<= max_num_sg
);
462 error_report("virtio: zero sized buffers are not allowed");
469 if (num_sg
== max_num_sg
) {
470 error_report("virtio: too many write descriptors in indirect table");
474 iov
[num_sg
].iov_base
= cpu_physical_memory_map(pa
, &len
, is_write
);
475 iov
[num_sg
].iov_len
= len
;
485 static void virtqueue_map_iovec(struct iovec
*sg
, hwaddr
*addr
,
486 unsigned int *num_sg
, unsigned int max_size
,
492 /* Note: this function MUST validate input, some callers
493 * are passing in num_sg values received over the network.
495 /* TODO: teach all callers that this can fail, and return failure instead
497 * When we do, we might be able to re-enable NDEBUG below.
500 #error building with NDEBUG is not supported
502 assert(*num_sg
<= max_size
);
504 for (i
= 0; i
< *num_sg
; i
++) {
506 sg
[i
].iov_base
= cpu_physical_memory_map(addr
[i
], &len
, is_write
);
507 if (!sg
[i
].iov_base
) {
508 error_report("virtio: error trying to map MMIO memory");
511 if (len
!= sg
[i
].iov_len
) {
512 error_report("virtio: unexpected memory split");
518 void virtqueue_map(VirtQueueElement
*elem
)
520 virtqueue_map_iovec(elem
->in_sg
, elem
->in_addr
, &elem
->in_num
,
521 VIRTQUEUE_MAX_SIZE
, 1);
522 virtqueue_map_iovec(elem
->out_sg
, elem
->out_addr
, &elem
->out_num
,
523 VIRTQUEUE_MAX_SIZE
, 0);
526 void *virtqueue_alloc_element(size_t sz
, unsigned out_num
, unsigned in_num
)
528 VirtQueueElement
*elem
;
529 size_t in_addr_ofs
= QEMU_ALIGN_UP(sz
, __alignof__(elem
->in_addr
[0]));
530 size_t out_addr_ofs
= in_addr_ofs
+ in_num
* sizeof(elem
->in_addr
[0]);
531 size_t out_addr_end
= out_addr_ofs
+ out_num
* sizeof(elem
->out_addr
[0]);
532 size_t in_sg_ofs
= QEMU_ALIGN_UP(out_addr_end
, __alignof__(elem
->in_sg
[0]));
533 size_t out_sg_ofs
= in_sg_ofs
+ in_num
* sizeof(elem
->in_sg
[0]);
534 size_t out_sg_end
= out_sg_ofs
+ out_num
* sizeof(elem
->out_sg
[0]);
536 assert(sz
>= sizeof(VirtQueueElement
));
537 elem
= g_malloc(out_sg_end
);
538 elem
->out_num
= out_num
;
539 elem
->in_num
= in_num
;
540 elem
->in_addr
= (void *)elem
+ in_addr_ofs
;
541 elem
->out_addr
= (void *)elem
+ out_addr_ofs
;
542 elem
->in_sg
= (void *)elem
+ in_sg_ofs
;
543 elem
->out_sg
= (void *)elem
+ out_sg_ofs
;
547 void *virtqueue_pop(VirtQueue
*vq
, size_t sz
)
549 unsigned int i
, head
, max
;
550 hwaddr desc_pa
= vq
->vring
.desc
;
551 VirtIODevice
*vdev
= vq
->vdev
;
552 VirtQueueElement
*elem
;
553 unsigned out_num
, in_num
;
554 hwaddr addr
[VIRTQUEUE_MAX_SIZE
];
555 struct iovec iov
[VIRTQUEUE_MAX_SIZE
];
558 if (virtio_queue_empty(vq
)) {
561 /* Needed after virtio_queue_empty(), see comment in
562 * virtqueue_num_heads(). */
565 /* When we start there are none of either input nor output. */
566 out_num
= in_num
= 0;
570 i
= head
= virtqueue_get_head(vq
, vq
->last_avail_idx
++);
571 if (virtio_vdev_has_feature(vdev
, VIRTIO_RING_F_EVENT_IDX
)) {
572 vring_set_avail_event(vq
, vq
->last_avail_idx
);
575 vring_desc_read(vdev
, &desc
, desc_pa
, i
);
576 if (desc
.flags
& VRING_DESC_F_INDIRECT
) {
577 if (desc
.len
% sizeof(VRingDesc
)) {
578 error_report("Invalid size for indirect buffer table");
582 /* loop over the indirect descriptor table */
583 max
= desc
.len
/ sizeof(VRingDesc
);
586 vring_desc_read(vdev
, &desc
, desc_pa
, i
);
589 /* Collect all the descriptors */
591 if (desc
.flags
& VRING_DESC_F_WRITE
) {
592 virtqueue_map_desc(&in_num
, addr
+ out_num
, iov
+ out_num
,
593 VIRTQUEUE_MAX_SIZE
- out_num
, true, desc
.addr
, desc
.len
);
596 error_report("Incorrect order for descriptors");
599 virtqueue_map_desc(&out_num
, addr
, iov
,
600 VIRTQUEUE_MAX_SIZE
, false, desc
.addr
, desc
.len
);
603 /* If we've got too many, that implies a descriptor loop. */
604 if ((in_num
+ out_num
) > max
) {
605 error_report("Looped descriptor");
608 } while ((i
= virtqueue_read_next_desc(vdev
, &desc
, desc_pa
, max
)) != max
);
610 /* Now copy what we have collected and mapped */
611 elem
= virtqueue_alloc_element(sz
, out_num
, in_num
);
613 for (i
= 0; i
< out_num
; i
++) {
614 elem
->out_addr
[i
] = addr
[i
];
615 elem
->out_sg
[i
] = iov
[i
];
617 for (i
= 0; i
< in_num
; i
++) {
618 elem
->in_addr
[i
] = addr
[out_num
+ i
];
619 elem
->in_sg
[i
] = iov
[out_num
+ i
];
624 trace_virtqueue_pop(vq
, elem
, elem
->in_num
, elem
->out_num
);
628 /* Reading and writing a structure directly to QEMUFile is *awful*, but
629 * it is what QEMU has always done by mistake. We can change it sooner
630 * or later by bumping the version number of the affected vm states.
631 * In the meanwhile, since the in-memory layout of VirtQueueElement
632 * has changed, we need to marshal to and from the layout that was
633 * used before the change.
635 typedef struct VirtQueueElementOld
{
637 unsigned int out_num
;
639 hwaddr in_addr
[VIRTQUEUE_MAX_SIZE
];
640 hwaddr out_addr
[VIRTQUEUE_MAX_SIZE
];
641 struct iovec in_sg
[VIRTQUEUE_MAX_SIZE
];
642 struct iovec out_sg
[VIRTQUEUE_MAX_SIZE
];
643 } VirtQueueElementOld
;
645 void *qemu_get_virtqueue_element(QEMUFile
*f
, size_t sz
)
647 VirtQueueElement
*elem
;
648 VirtQueueElementOld data
;
651 qemu_get_buffer(f
, (uint8_t *)&data
, sizeof(VirtQueueElementOld
));
653 elem
= virtqueue_alloc_element(sz
, data
.out_num
, data
.in_num
);
654 elem
->index
= data
.index
;
656 for (i
= 0; i
< elem
->in_num
; i
++) {
657 elem
->in_addr
[i
] = data
.in_addr
[i
];
660 for (i
= 0; i
< elem
->out_num
; i
++) {
661 elem
->out_addr
[i
] = data
.out_addr
[i
];
664 for (i
= 0; i
< elem
->in_num
; i
++) {
665 /* Base is overwritten by virtqueue_map. */
666 elem
->in_sg
[i
].iov_base
= 0;
667 elem
->in_sg
[i
].iov_len
= data
.in_sg
[i
].iov_len
;
670 for (i
= 0; i
< elem
->out_num
; i
++) {
671 /* Base is overwritten by virtqueue_map. */
672 elem
->out_sg
[i
].iov_base
= 0;
673 elem
->out_sg
[i
].iov_len
= data
.out_sg
[i
].iov_len
;
680 void qemu_put_virtqueue_element(QEMUFile
*f
, VirtQueueElement
*elem
)
682 VirtQueueElementOld data
;
685 memset(&data
, 0, sizeof(data
));
686 data
.index
= elem
->index
;
687 data
.in_num
= elem
->in_num
;
688 data
.out_num
= elem
->out_num
;
690 for (i
= 0; i
< elem
->in_num
; i
++) {
691 data
.in_addr
[i
] = elem
->in_addr
[i
];
694 for (i
= 0; i
< elem
->out_num
; i
++) {
695 data
.out_addr
[i
] = elem
->out_addr
[i
];
698 for (i
= 0; i
< elem
->in_num
; i
++) {
699 /* Base is overwritten by virtqueue_map when loading. Do not
700 * save it, as it would leak the QEMU address space layout. */
701 data
.in_sg
[i
].iov_len
= elem
->in_sg
[i
].iov_len
;
704 for (i
= 0; i
< elem
->out_num
; i
++) {
705 /* Do not save iov_base as above. */
706 data
.out_sg
[i
].iov_len
= elem
->out_sg
[i
].iov_len
;
708 qemu_put_buffer(f
, (uint8_t *)&data
, sizeof(VirtQueueElementOld
));
712 static void virtio_notify_vector(VirtIODevice
*vdev
, uint16_t vector
)
714 BusState
*qbus
= qdev_get_parent_bus(DEVICE(vdev
));
715 VirtioBusClass
*k
= VIRTIO_BUS_GET_CLASS(qbus
);
718 k
->notify(qbus
->parent
, vector
);
722 void virtio_update_irq(VirtIODevice
*vdev
)
724 virtio_notify_vector(vdev
, VIRTIO_NO_VECTOR
);
727 static int virtio_validate_features(VirtIODevice
*vdev
)
729 VirtioDeviceClass
*k
= VIRTIO_DEVICE_GET_CLASS(vdev
);
731 if (k
->validate_features
) {
732 return k
->validate_features(vdev
);
738 int virtio_set_status(VirtIODevice
*vdev
, uint8_t val
)
740 VirtioDeviceClass
*k
= VIRTIO_DEVICE_GET_CLASS(vdev
);
741 trace_virtio_set_status(vdev
, val
);
743 if (virtio_vdev_has_feature(vdev
, VIRTIO_F_VERSION_1
)) {
744 if (!(vdev
->status
& VIRTIO_CONFIG_S_FEATURES_OK
) &&
745 val
& VIRTIO_CONFIG_S_FEATURES_OK
) {
746 int ret
= virtio_validate_features(vdev
);
754 k
->set_status(vdev
, val
);
760 bool target_words_bigendian(void);
761 static enum virtio_device_endian
virtio_default_endian(void)
763 if (target_words_bigendian()) {
764 return VIRTIO_DEVICE_ENDIAN_BIG
;
766 return VIRTIO_DEVICE_ENDIAN_LITTLE
;
770 static enum virtio_device_endian
virtio_current_cpu_endian(void)
772 CPUClass
*cc
= CPU_GET_CLASS(current_cpu
);
774 if (cc
->virtio_is_big_endian(current_cpu
)) {
775 return VIRTIO_DEVICE_ENDIAN_BIG
;
777 return VIRTIO_DEVICE_ENDIAN_LITTLE
;
781 void virtio_reset(void *opaque
)
783 VirtIODevice
*vdev
= opaque
;
784 VirtioDeviceClass
*k
= VIRTIO_DEVICE_GET_CLASS(vdev
);
787 virtio_set_status(vdev
, 0);
789 /* Guest initiated reset */
790 vdev
->device_endian
= virtio_current_cpu_endian();
793 vdev
->device_endian
= virtio_default_endian();
800 vdev
->guest_features
= 0;
804 vdev
->config_vector
= VIRTIO_NO_VECTOR
;
805 virtio_notify_vector(vdev
, vdev
->config_vector
);
807 for(i
= 0; i
< VIRTIO_QUEUE_MAX
; i
++) {
808 vdev
->vq
[i
].vring
.desc
= 0;
809 vdev
->vq
[i
].vring
.avail
= 0;
810 vdev
->vq
[i
].vring
.used
= 0;
811 vdev
->vq
[i
].last_avail_idx
= 0;
812 vdev
->vq
[i
].shadow_avail_idx
= 0;
813 vdev
->vq
[i
].used_idx
= 0;
814 virtio_queue_set_vector(vdev
, i
, VIRTIO_NO_VECTOR
);
815 vdev
->vq
[i
].signalled_used
= 0;
816 vdev
->vq
[i
].signalled_used_valid
= false;
817 vdev
->vq
[i
].notification
= true;
818 vdev
->vq
[i
].vring
.num
= vdev
->vq
[i
].vring
.num_default
;
822 uint32_t virtio_config_readb(VirtIODevice
*vdev
, uint32_t addr
)
824 VirtioDeviceClass
*k
= VIRTIO_DEVICE_GET_CLASS(vdev
);
827 if (addr
+ sizeof(val
) > vdev
->config_len
) {
831 k
->get_config(vdev
, vdev
->config
);
833 val
= ldub_p(vdev
->config
+ addr
);
837 uint32_t virtio_config_readw(VirtIODevice
*vdev
, uint32_t addr
)
839 VirtioDeviceClass
*k
= VIRTIO_DEVICE_GET_CLASS(vdev
);
842 if (addr
+ sizeof(val
) > vdev
->config_len
) {
846 k
->get_config(vdev
, vdev
->config
);
848 val
= lduw_p(vdev
->config
+ addr
);
852 uint32_t virtio_config_readl(VirtIODevice
*vdev
, uint32_t addr
)
854 VirtioDeviceClass
*k
= VIRTIO_DEVICE_GET_CLASS(vdev
);
857 if (addr
+ sizeof(val
) > vdev
->config_len
) {
861 k
->get_config(vdev
, vdev
->config
);
863 val
= ldl_p(vdev
->config
+ addr
);
867 void virtio_config_writeb(VirtIODevice
*vdev
, uint32_t addr
, uint32_t data
)
869 VirtioDeviceClass
*k
= VIRTIO_DEVICE_GET_CLASS(vdev
);
872 if (addr
+ sizeof(val
) > vdev
->config_len
) {
876 stb_p(vdev
->config
+ addr
, val
);
879 k
->set_config(vdev
, vdev
->config
);
883 void virtio_config_writew(VirtIODevice
*vdev
, uint32_t addr
, uint32_t data
)
885 VirtioDeviceClass
*k
= VIRTIO_DEVICE_GET_CLASS(vdev
);
888 if (addr
+ sizeof(val
) > vdev
->config_len
) {
892 stw_p(vdev
->config
+ addr
, val
);
895 k
->set_config(vdev
, vdev
->config
);
899 void virtio_config_writel(VirtIODevice
*vdev
, uint32_t addr
, uint32_t data
)
901 VirtioDeviceClass
*k
= VIRTIO_DEVICE_GET_CLASS(vdev
);
904 if (addr
+ sizeof(val
) > vdev
->config_len
) {
908 stl_p(vdev
->config
+ addr
, val
);
911 k
->set_config(vdev
, vdev
->config
);
915 uint32_t virtio_config_modern_readb(VirtIODevice
*vdev
, uint32_t addr
)
917 VirtioDeviceClass
*k
= VIRTIO_DEVICE_GET_CLASS(vdev
);
920 if (addr
+ sizeof(val
) > vdev
->config_len
) {
924 k
->get_config(vdev
, vdev
->config
);
926 val
= ldub_p(vdev
->config
+ addr
);
930 uint32_t virtio_config_modern_readw(VirtIODevice
*vdev
, uint32_t addr
)
932 VirtioDeviceClass
*k
= VIRTIO_DEVICE_GET_CLASS(vdev
);
935 if (addr
+ sizeof(val
) > vdev
->config_len
) {
939 k
->get_config(vdev
, vdev
->config
);
941 val
= lduw_le_p(vdev
->config
+ addr
);
945 uint32_t virtio_config_modern_readl(VirtIODevice
*vdev
, uint32_t addr
)
947 VirtioDeviceClass
*k
= VIRTIO_DEVICE_GET_CLASS(vdev
);
950 if (addr
+ sizeof(val
) > vdev
->config_len
) {
954 k
->get_config(vdev
, vdev
->config
);
956 val
= ldl_le_p(vdev
->config
+ addr
);
960 void virtio_config_modern_writeb(VirtIODevice
*vdev
,
961 uint32_t addr
, uint32_t data
)
963 VirtioDeviceClass
*k
= VIRTIO_DEVICE_GET_CLASS(vdev
);
966 if (addr
+ sizeof(val
) > vdev
->config_len
) {
970 stb_p(vdev
->config
+ addr
, val
);
973 k
->set_config(vdev
, vdev
->config
);
977 void virtio_config_modern_writew(VirtIODevice
*vdev
,
978 uint32_t addr
, uint32_t data
)
980 VirtioDeviceClass
*k
= VIRTIO_DEVICE_GET_CLASS(vdev
);
983 if (addr
+ sizeof(val
) > vdev
->config_len
) {
987 stw_le_p(vdev
->config
+ addr
, val
);
990 k
->set_config(vdev
, vdev
->config
);
994 void virtio_config_modern_writel(VirtIODevice
*vdev
,
995 uint32_t addr
, uint32_t data
)
997 VirtioDeviceClass
*k
= VIRTIO_DEVICE_GET_CLASS(vdev
);
1000 if (addr
+ sizeof(val
) > vdev
->config_len
) {
1004 stl_le_p(vdev
->config
+ addr
, val
);
1006 if (k
->set_config
) {
1007 k
->set_config(vdev
, vdev
->config
);
1011 void virtio_queue_set_addr(VirtIODevice
*vdev
, int n
, hwaddr addr
)
1013 vdev
->vq
[n
].vring
.desc
= addr
;
1014 virtio_queue_update_rings(vdev
, n
);
1017 hwaddr
virtio_queue_get_addr(VirtIODevice
*vdev
, int n
)
1019 return vdev
->vq
[n
].vring
.desc
;
1022 void virtio_queue_set_rings(VirtIODevice
*vdev
, int n
, hwaddr desc
,
1023 hwaddr avail
, hwaddr used
)
1025 vdev
->vq
[n
].vring
.desc
= desc
;
1026 vdev
->vq
[n
].vring
.avail
= avail
;
1027 vdev
->vq
[n
].vring
.used
= used
;
1030 void virtio_queue_set_num(VirtIODevice
*vdev
, int n
, int num
)
1032 /* Don't allow guest to flip queue between existent and
1033 * nonexistent states, or to set it to an invalid size.
1035 if (!!num
!= !!vdev
->vq
[n
].vring
.num
||
1036 num
> VIRTQUEUE_MAX_SIZE
||
1040 vdev
->vq
[n
].vring
.num
= num
;
1043 VirtQueue
*virtio_vector_first_queue(VirtIODevice
*vdev
, uint16_t vector
)
1045 return QLIST_FIRST(&vdev
->vector_queues
[vector
]);
1048 VirtQueue
*virtio_vector_next_queue(VirtQueue
*vq
)
1050 return QLIST_NEXT(vq
, node
);
1053 int virtio_queue_get_num(VirtIODevice
*vdev
, int n
)
1055 return vdev
->vq
[n
].vring
.num
;
1058 int virtio_get_num_queues(VirtIODevice
*vdev
)
1062 for (i
= 0; i
< VIRTIO_QUEUE_MAX
; i
++) {
1063 if (!virtio_queue_get_num(vdev
, i
)) {
1071 void virtio_queue_set_align(VirtIODevice
*vdev
, int n
, int align
)
1073 BusState
*qbus
= qdev_get_parent_bus(DEVICE(vdev
));
1074 VirtioBusClass
*k
= VIRTIO_BUS_GET_CLASS(qbus
);
1076 /* virtio-1 compliant devices cannot change the alignment */
1077 if (virtio_vdev_has_feature(vdev
, VIRTIO_F_VERSION_1
)) {
1078 error_report("tried to modify queue alignment for virtio-1 device");
1081 /* Check that the transport told us it was going to do this
1082 * (so a buggy transport will immediately assert rather than
1083 * silently failing to migrate this state)
1085 assert(k
->has_variable_vring_alignment
);
1087 vdev
->vq
[n
].vring
.align
= align
;
1088 virtio_queue_update_rings(vdev
, n
);
1091 static void virtio_queue_notify_aio_vq(VirtQueue
*vq
)
1093 if (vq
->vring
.desc
&& vq
->handle_aio_output
) {
1094 VirtIODevice
*vdev
= vq
->vdev
;
1096 trace_virtio_queue_notify(vdev
, vq
- vdev
->vq
, vq
);
1097 vq
->handle_aio_output(vdev
, vq
);
1101 static void virtio_queue_notify_vq(VirtQueue
*vq
)
1103 if (vq
->vring
.desc
&& vq
->handle_output
) {
1104 VirtIODevice
*vdev
= vq
->vdev
;
1106 trace_virtio_queue_notify(vdev
, vq
- vdev
->vq
, vq
);
1107 vq
->handle_output(vdev
, vq
);
1111 void virtio_queue_notify(VirtIODevice
*vdev
, int n
)
1113 virtio_queue_notify_vq(&vdev
->vq
[n
]);
1116 uint16_t virtio_queue_vector(VirtIODevice
*vdev
, int n
)
1118 return n
< VIRTIO_QUEUE_MAX
? vdev
->vq
[n
].vector
:
1122 void virtio_queue_set_vector(VirtIODevice
*vdev
, int n
, uint16_t vector
)
1124 VirtQueue
*vq
= &vdev
->vq
[n
];
1126 if (n
< VIRTIO_QUEUE_MAX
) {
1127 if (vdev
->vector_queues
&&
1128 vdev
->vq
[n
].vector
!= VIRTIO_NO_VECTOR
) {
1129 QLIST_REMOVE(vq
, node
);
1131 vdev
->vq
[n
].vector
= vector
;
1132 if (vdev
->vector_queues
&&
1133 vector
!= VIRTIO_NO_VECTOR
) {
1134 QLIST_INSERT_HEAD(&vdev
->vector_queues
[vector
], vq
, node
);
1139 static VirtQueue
*virtio_add_queue_internal(VirtIODevice
*vdev
, int queue_size
,
1140 VirtIOHandleOutput handle_output
,
1145 for (i
= 0; i
< VIRTIO_QUEUE_MAX
; i
++) {
1146 if (vdev
->vq
[i
].vring
.num
== 0)
1150 if (i
== VIRTIO_QUEUE_MAX
|| queue_size
> VIRTQUEUE_MAX_SIZE
)
1153 vdev
->vq
[i
].vring
.num
= queue_size
;
1154 vdev
->vq
[i
].vring
.num_default
= queue_size
;
1155 vdev
->vq
[i
].vring
.align
= VIRTIO_PCI_VRING_ALIGN
;
1156 vdev
->vq
[i
].handle_output
= handle_output
;
1157 vdev
->vq
[i
].handle_aio_output
= NULL
;
1158 vdev
->vq
[i
].use_aio
= use_aio
;
1160 return &vdev
->vq
[i
];
1163 /* Add a virt queue and mark AIO.
1164 * An AIO queue will use the AioContext based event interface instead of the
1165 * default IOHandler and EventNotifier interface.
1167 VirtQueue
*virtio_add_queue_aio(VirtIODevice
*vdev
, int queue_size
,
1168 VirtIOHandleOutput handle_output
)
1170 return virtio_add_queue_internal(vdev
, queue_size
, handle_output
, true);
1173 /* Add a normal virt queue (on the contrary to the AIO version above. */
1174 VirtQueue
*virtio_add_queue(VirtIODevice
*vdev
, int queue_size
,
1175 VirtIOHandleOutput handle_output
)
1177 return virtio_add_queue_internal(vdev
, queue_size
, handle_output
, false);
1180 void virtio_del_queue(VirtIODevice
*vdev
, int n
)
1182 if (n
< 0 || n
>= VIRTIO_QUEUE_MAX
) {
1186 vdev
->vq
[n
].vring
.num
= 0;
1187 vdev
->vq
[n
].vring
.num_default
= 0;
1190 void virtio_irq(VirtQueue
*vq
)
1192 trace_virtio_irq(vq
);
1193 vq
->vdev
->isr
|= 0x01;
1194 virtio_notify_vector(vq
->vdev
, vq
->vector
);
1197 bool virtio_should_notify(VirtIODevice
*vdev
, VirtQueue
*vq
)
1201 /* We need to expose used array entries before checking used event. */
1203 /* Always notify when queue is empty (when feature acknowledge) */
1204 if (virtio_vdev_has_feature(vdev
, VIRTIO_F_NOTIFY_ON_EMPTY
) &&
1205 !vq
->inuse
&& virtio_queue_empty(vq
)) {
1209 if (!virtio_vdev_has_feature(vdev
, VIRTIO_RING_F_EVENT_IDX
)) {
1210 return !(vring_avail_flags(vq
) & VRING_AVAIL_F_NO_INTERRUPT
);
1213 v
= vq
->signalled_used_valid
;
1214 vq
->signalled_used_valid
= true;
1215 old
= vq
->signalled_used
;
1216 new = vq
->signalled_used
= vq
->used_idx
;
1217 return !v
|| vring_need_event(vring_get_used_event(vq
), new, old
);
1220 void virtio_notify(VirtIODevice
*vdev
, VirtQueue
*vq
)
1222 if (!virtio_should_notify(vdev
, vq
)) {
1226 trace_virtio_notify(vdev
, vq
);
1228 virtio_notify_vector(vdev
, vq
->vector
);
1231 void virtio_notify_config(VirtIODevice
*vdev
)
1233 if (!(vdev
->status
& VIRTIO_CONFIG_S_DRIVER_OK
))
1238 virtio_notify_vector(vdev
, vdev
->config_vector
);
1241 static bool virtio_device_endian_needed(void *opaque
)
1243 VirtIODevice
*vdev
= opaque
;
1245 assert(vdev
->device_endian
!= VIRTIO_DEVICE_ENDIAN_UNKNOWN
);
1246 if (!virtio_vdev_has_feature(vdev
, VIRTIO_F_VERSION_1
)) {
1247 return vdev
->device_endian
!= virtio_default_endian();
1249 /* Devices conforming to VIRTIO 1.0 or later are always LE. */
1250 return vdev
->device_endian
!= VIRTIO_DEVICE_ENDIAN_LITTLE
;
1253 static bool virtio_64bit_features_needed(void *opaque
)
1255 VirtIODevice
*vdev
= opaque
;
1257 return (vdev
->host_features
>> 32) != 0;
1260 static bool virtio_virtqueue_needed(void *opaque
)
1262 VirtIODevice
*vdev
= opaque
;
1264 return virtio_host_has_feature(vdev
, VIRTIO_F_VERSION_1
);
1267 static bool virtio_ringsize_needed(void *opaque
)
1269 VirtIODevice
*vdev
= opaque
;
1272 for (i
= 0; i
< VIRTIO_QUEUE_MAX
; i
++) {
1273 if (vdev
->vq
[i
].vring
.num
!= vdev
->vq
[i
].vring
.num_default
) {
1280 static bool virtio_extra_state_needed(void *opaque
)
1282 VirtIODevice
*vdev
= opaque
;
1283 BusState
*qbus
= qdev_get_parent_bus(DEVICE(vdev
));
1284 VirtioBusClass
*k
= VIRTIO_BUS_GET_CLASS(qbus
);
1286 return k
->has_extra_state
&&
1287 k
->has_extra_state(qbus
->parent
);
1290 static const VMStateDescription vmstate_virtqueue
= {
1291 .name
= "virtqueue_state",
1293 .minimum_version_id
= 1,
1294 .fields
= (VMStateField
[]) {
1295 VMSTATE_UINT64(vring
.avail
, struct VirtQueue
),
1296 VMSTATE_UINT64(vring
.used
, struct VirtQueue
),
1297 VMSTATE_END_OF_LIST()
1301 static const VMStateDescription vmstate_virtio_virtqueues
= {
1302 .name
= "virtio/virtqueues",
1304 .minimum_version_id
= 1,
1305 .needed
= &virtio_virtqueue_needed
,
1306 .fields
= (VMStateField
[]) {
1307 VMSTATE_STRUCT_VARRAY_POINTER_KNOWN(vq
, struct VirtIODevice
,
1308 VIRTIO_QUEUE_MAX
, 0, vmstate_virtqueue
, VirtQueue
),
1309 VMSTATE_END_OF_LIST()
1313 static const VMStateDescription vmstate_ringsize
= {
1314 .name
= "ringsize_state",
1316 .minimum_version_id
= 1,
1317 .fields
= (VMStateField
[]) {
1318 VMSTATE_UINT32(vring
.num_default
, struct VirtQueue
),
1319 VMSTATE_END_OF_LIST()
1323 static const VMStateDescription vmstate_virtio_ringsize
= {
1324 .name
= "virtio/ringsize",
1326 .minimum_version_id
= 1,
1327 .needed
= &virtio_ringsize_needed
,
1328 .fields
= (VMStateField
[]) {
1329 VMSTATE_STRUCT_VARRAY_POINTER_KNOWN(vq
, struct VirtIODevice
,
1330 VIRTIO_QUEUE_MAX
, 0, vmstate_ringsize
, VirtQueue
),
1331 VMSTATE_END_OF_LIST()
1335 static int get_extra_state(QEMUFile
*f
, void *pv
, size_t size
)
1337 VirtIODevice
*vdev
= pv
;
1338 BusState
*qbus
= qdev_get_parent_bus(DEVICE(vdev
));
1339 VirtioBusClass
*k
= VIRTIO_BUS_GET_CLASS(qbus
);
1341 if (!k
->load_extra_state
) {
1344 return k
->load_extra_state(qbus
->parent
, f
);
1348 static void put_extra_state(QEMUFile
*f
, void *pv
, size_t size
)
1350 VirtIODevice
*vdev
= pv
;
1351 BusState
*qbus
= qdev_get_parent_bus(DEVICE(vdev
));
1352 VirtioBusClass
*k
= VIRTIO_BUS_GET_CLASS(qbus
);
1354 k
->save_extra_state(qbus
->parent
, f
);
1357 static const VMStateInfo vmstate_info_extra_state
= {
1358 .name
= "virtqueue_extra_state",
1359 .get
= get_extra_state
,
1360 .put
= put_extra_state
,
1363 static const VMStateDescription vmstate_virtio_extra_state
= {
1364 .name
= "virtio/extra_state",
1366 .minimum_version_id
= 1,
1367 .needed
= &virtio_extra_state_needed
,
1368 .fields
= (VMStateField
[]) {
1370 .name
= "extra_state",
1372 .field_exists
= NULL
,
1374 .info
= &vmstate_info_extra_state
,
1375 .flags
= VMS_SINGLE
,
1378 VMSTATE_END_OF_LIST()
1382 static const VMStateDescription vmstate_virtio_device_endian
= {
1383 .name
= "virtio/device_endian",
1385 .minimum_version_id
= 1,
1386 .needed
= &virtio_device_endian_needed
,
1387 .fields
= (VMStateField
[]) {
1388 VMSTATE_UINT8(device_endian
, VirtIODevice
),
1389 VMSTATE_END_OF_LIST()
1393 static const VMStateDescription vmstate_virtio_64bit_features
= {
1394 .name
= "virtio/64bit_features",
1396 .minimum_version_id
= 1,
1397 .needed
= &virtio_64bit_features_needed
,
1398 .fields
= (VMStateField
[]) {
1399 VMSTATE_UINT64(guest_features
, VirtIODevice
),
1400 VMSTATE_END_OF_LIST()
1404 static const VMStateDescription vmstate_virtio
= {
1407 .minimum_version_id
= 1,
1408 .minimum_version_id_old
= 1,
1409 .fields
= (VMStateField
[]) {
1410 VMSTATE_END_OF_LIST()
1412 .subsections
= (const VMStateDescription
*[]) {
1413 &vmstate_virtio_device_endian
,
1414 &vmstate_virtio_64bit_features
,
1415 &vmstate_virtio_virtqueues
,
1416 &vmstate_virtio_ringsize
,
1417 &vmstate_virtio_extra_state
,
1422 void virtio_save(VirtIODevice
*vdev
, QEMUFile
*f
)
1424 BusState
*qbus
= qdev_get_parent_bus(DEVICE(vdev
));
1425 VirtioBusClass
*k
= VIRTIO_BUS_GET_CLASS(qbus
);
1426 VirtioDeviceClass
*vdc
= VIRTIO_DEVICE_GET_CLASS(vdev
);
1427 uint32_t guest_features_lo
= (vdev
->guest_features
& 0xffffffff);
1430 if (k
->save_config
) {
1431 k
->save_config(qbus
->parent
, f
);
1434 qemu_put_8s(f
, &vdev
->status
);
1435 qemu_put_8s(f
, &vdev
->isr
);
1436 qemu_put_be16s(f
, &vdev
->queue_sel
);
1437 qemu_put_be32s(f
, &guest_features_lo
);
1438 qemu_put_be32(f
, vdev
->config_len
);
1439 qemu_put_buffer(f
, vdev
->config
, vdev
->config_len
);
1441 for (i
= 0; i
< VIRTIO_QUEUE_MAX
; i
++) {
1442 if (vdev
->vq
[i
].vring
.num
== 0)
1446 qemu_put_be32(f
, i
);
1448 for (i
= 0; i
< VIRTIO_QUEUE_MAX
; i
++) {
1449 if (vdev
->vq
[i
].vring
.num
== 0)
1452 qemu_put_be32(f
, vdev
->vq
[i
].vring
.num
);
1453 if (k
->has_variable_vring_alignment
) {
1454 qemu_put_be32(f
, vdev
->vq
[i
].vring
.align
);
1456 /* XXX virtio-1 devices */
1457 qemu_put_be64(f
, vdev
->vq
[i
].vring
.desc
);
1458 qemu_put_be16s(f
, &vdev
->vq
[i
].last_avail_idx
);
1459 if (k
->save_queue
) {
1460 k
->save_queue(qbus
->parent
, i
, f
);
1464 if (vdc
->save
!= NULL
) {
1469 vmstate_save_state(f
, &vmstate_virtio
, vdev
, NULL
);
1472 /* A wrapper for use as a VMState .put function */
1473 void virtio_vmstate_save(QEMUFile
*f
, void *opaque
, size_t size
)
1475 virtio_save(VIRTIO_DEVICE(opaque
), f
);
1478 static int virtio_set_features_nocheck(VirtIODevice
*vdev
, uint64_t val
)
1480 VirtioDeviceClass
*k
= VIRTIO_DEVICE_GET_CLASS(vdev
);
1481 bool bad
= (val
& ~(vdev
->host_features
)) != 0;
1483 val
&= vdev
->host_features
;
1484 if (k
->set_features
) {
1485 k
->set_features(vdev
, val
);
1487 vdev
->guest_features
= val
;
1488 return bad
? -1 : 0;
1491 int virtio_set_features(VirtIODevice
*vdev
, uint64_t val
)
1494 * The driver must not attempt to set features after feature negotiation
1497 if (vdev
->status
& VIRTIO_CONFIG_S_FEATURES_OK
) {
1500 return virtio_set_features_nocheck(vdev
, val
);
1503 int virtio_load(VirtIODevice
*vdev
, QEMUFile
*f
, int version_id
)
1509 BusState
*qbus
= qdev_get_parent_bus(DEVICE(vdev
));
1510 VirtioBusClass
*k
= VIRTIO_BUS_GET_CLASS(qbus
);
1511 VirtioDeviceClass
*vdc
= VIRTIO_DEVICE_GET_CLASS(vdev
);
1514 * We poison the endianness to ensure it does not get used before
1515 * subsections have been loaded.
1517 vdev
->device_endian
= VIRTIO_DEVICE_ENDIAN_UNKNOWN
;
1519 if (k
->load_config
) {
1520 ret
= k
->load_config(qbus
->parent
, f
);
1525 qemu_get_8s(f
, &vdev
->status
);
1526 qemu_get_8s(f
, &vdev
->isr
);
1527 qemu_get_be16s(f
, &vdev
->queue_sel
);
1528 if (vdev
->queue_sel
>= VIRTIO_QUEUE_MAX
) {
1531 qemu_get_be32s(f
, &features
);
1534 * Temporarily set guest_features low bits - needed by
1535 * virtio net load code testing for VIRTIO_NET_F_CTRL_GUEST_OFFLOADS
1536 * VIRTIO_NET_F_GUEST_ANNOUNCE and VIRTIO_NET_F_CTRL_VQ.
1538 * Note: devices should always test host features in future - don't create
1539 * new dependencies like this.
1541 vdev
->guest_features
= features
;
1543 config_len
= qemu_get_be32(f
);
1546 * There are cases where the incoming config can be bigger or smaller
1547 * than what we have; so load what we have space for, and skip
1548 * any excess that's in the stream.
1550 qemu_get_buffer(f
, vdev
->config
, MIN(config_len
, vdev
->config_len
));
1552 while (config_len
> vdev
->config_len
) {
1557 num
= qemu_get_be32(f
);
1559 if (num
> VIRTIO_QUEUE_MAX
) {
1560 error_report("Invalid number of virtqueues: 0x%x", num
);
1564 for (i
= 0; i
< num
; i
++) {
1565 vdev
->vq
[i
].vring
.num
= qemu_get_be32(f
);
1566 if (k
->has_variable_vring_alignment
) {
1567 vdev
->vq
[i
].vring
.align
= qemu_get_be32(f
);
1569 vdev
->vq
[i
].vring
.desc
= qemu_get_be64(f
);
1570 qemu_get_be16s(f
, &vdev
->vq
[i
].last_avail_idx
);
1571 vdev
->vq
[i
].signalled_used_valid
= false;
1572 vdev
->vq
[i
].notification
= true;
1574 if (vdev
->vq
[i
].vring
.desc
) {
1575 /* XXX virtio-1 devices */
1576 virtio_queue_update_rings(vdev
, i
);
1577 } else if (vdev
->vq
[i
].last_avail_idx
) {
1578 error_report("VQ %d address 0x0 "
1579 "inconsistent with Host index 0x%x",
1580 i
, vdev
->vq
[i
].last_avail_idx
);
1583 if (k
->load_queue
) {
1584 ret
= k
->load_queue(qbus
->parent
, i
, f
);
1590 virtio_notify_vector(vdev
, VIRTIO_NO_VECTOR
);
1592 if (vdc
->load
!= NULL
) {
1593 ret
= vdc
->load(vdev
, f
, version_id
);
1600 ret
= vmstate_load_state(f
, &vmstate_virtio
, vdev
, 1);
1605 if (vdev
->device_endian
== VIRTIO_DEVICE_ENDIAN_UNKNOWN
) {
1606 vdev
->device_endian
= virtio_default_endian();
1609 if (virtio_64bit_features_needed(vdev
)) {
1611 * Subsection load filled vdev->guest_features. Run them
1612 * through virtio_set_features to sanity-check them against
1615 uint64_t features64
= vdev
->guest_features
;
1616 if (virtio_set_features_nocheck(vdev
, features64
) < 0) {
1617 error_report("Features 0x%" PRIx64
" unsupported. "
1618 "Allowed features: 0x%" PRIx64
,
1619 features64
, vdev
->host_features
);
1623 if (virtio_set_features_nocheck(vdev
, features
) < 0) {
1624 error_report("Features 0x%x unsupported. "
1625 "Allowed features: 0x%" PRIx64
,
1626 features
, vdev
->host_features
);
1631 for (i
= 0; i
< num
; i
++) {
1632 if (vdev
->vq
[i
].vring
.desc
) {
1634 nheads
= vring_avail_idx(&vdev
->vq
[i
]) - vdev
->vq
[i
].last_avail_idx
;
1635 /* Check it isn't doing strange things with descriptor numbers. */
1636 if (nheads
> vdev
->vq
[i
].vring
.num
) {
1637 error_report("VQ %d size 0x%x Guest index 0x%x "
1638 "inconsistent with Host index 0x%x: delta 0x%x",
1639 i
, vdev
->vq
[i
].vring
.num
,
1640 vring_avail_idx(&vdev
->vq
[i
]),
1641 vdev
->vq
[i
].last_avail_idx
, nheads
);
1644 vdev
->vq
[i
].used_idx
= vring_used_idx(&vdev
->vq
[i
]);
1645 vdev
->vq
[i
].shadow_avail_idx
= vring_avail_idx(&vdev
->vq
[i
]);
1652 void virtio_cleanup(VirtIODevice
*vdev
)
1654 qemu_del_vm_change_state_handler(vdev
->vmstate
);
1655 g_free(vdev
->config
);
1657 g_free(vdev
->vector_queues
);
1660 static void virtio_vmstate_change(void *opaque
, int running
, RunState state
)
1662 VirtIODevice
*vdev
= opaque
;
1663 BusState
*qbus
= qdev_get_parent_bus(DEVICE(vdev
));
1664 VirtioBusClass
*k
= VIRTIO_BUS_GET_CLASS(qbus
);
1665 bool backend_run
= running
&& (vdev
->status
& VIRTIO_CONFIG_S_DRIVER_OK
);
1666 vdev
->vm_running
= running
;
1669 virtio_set_status(vdev
, vdev
->status
);
1672 if (k
->vmstate_change
) {
1673 k
->vmstate_change(qbus
->parent
, backend_run
);
1677 virtio_set_status(vdev
, vdev
->status
);
1681 void virtio_instance_init_common(Object
*proxy_obj
, void *data
,
1682 size_t vdev_size
, const char *vdev_name
)
1684 DeviceState
*vdev
= data
;
1686 object_initialize(vdev
, vdev_size
, vdev_name
);
1687 object_property_add_child(proxy_obj
, "virtio-backend", OBJECT(vdev
), NULL
);
1688 object_unref(OBJECT(vdev
));
1689 qdev_alias_all_properties(vdev
, proxy_obj
);
1692 void virtio_init(VirtIODevice
*vdev
, const char *name
,
1693 uint16_t device_id
, size_t config_size
)
1695 BusState
*qbus
= qdev_get_parent_bus(DEVICE(vdev
));
1696 VirtioBusClass
*k
= VIRTIO_BUS_GET_CLASS(qbus
);
1698 int nvectors
= k
->query_nvectors
? k
->query_nvectors(qbus
->parent
) : 0;
1701 vdev
->vector_queues
=
1702 g_malloc0(sizeof(*vdev
->vector_queues
) * nvectors
);
1705 vdev
->device_id
= device_id
;
1708 vdev
->queue_sel
= 0;
1709 vdev
->config_vector
= VIRTIO_NO_VECTOR
;
1710 vdev
->vq
= g_malloc0(sizeof(VirtQueue
) * VIRTIO_QUEUE_MAX
);
1711 vdev
->vm_running
= runstate_is_running();
1712 for (i
= 0; i
< VIRTIO_QUEUE_MAX
; i
++) {
1713 vdev
->vq
[i
].vector
= VIRTIO_NO_VECTOR
;
1714 vdev
->vq
[i
].vdev
= vdev
;
1715 vdev
->vq
[i
].queue_index
= i
;
1719 vdev
->config_len
= config_size
;
1720 if (vdev
->config_len
) {
1721 vdev
->config
= g_malloc0(config_size
);
1723 vdev
->config
= NULL
;
1725 vdev
->vmstate
= qemu_add_vm_change_state_handler(virtio_vmstate_change
,
1727 vdev
->device_endian
= virtio_default_endian();
1728 vdev
->use_guest_notifier_mask
= true;
1731 hwaddr
virtio_queue_get_desc_addr(VirtIODevice
*vdev
, int n
)
1733 return vdev
->vq
[n
].vring
.desc
;
1736 hwaddr
virtio_queue_get_avail_addr(VirtIODevice
*vdev
, int n
)
1738 return vdev
->vq
[n
].vring
.avail
;
1741 hwaddr
virtio_queue_get_used_addr(VirtIODevice
*vdev
, int n
)
1743 return vdev
->vq
[n
].vring
.used
;
1746 hwaddr
virtio_queue_get_ring_addr(VirtIODevice
*vdev
, int n
)
1748 return vdev
->vq
[n
].vring
.desc
;
1751 hwaddr
virtio_queue_get_desc_size(VirtIODevice
*vdev
, int n
)
1753 return sizeof(VRingDesc
) * vdev
->vq
[n
].vring
.num
;
1756 hwaddr
virtio_queue_get_avail_size(VirtIODevice
*vdev
, int n
)
1758 return offsetof(VRingAvail
, ring
) +
1759 sizeof(uint16_t) * vdev
->vq
[n
].vring
.num
;
1762 hwaddr
virtio_queue_get_used_size(VirtIODevice
*vdev
, int n
)
1764 return offsetof(VRingUsed
, ring
) +
1765 sizeof(VRingUsedElem
) * vdev
->vq
[n
].vring
.num
;
1768 hwaddr
virtio_queue_get_ring_size(VirtIODevice
*vdev
, int n
)
1770 return vdev
->vq
[n
].vring
.used
- vdev
->vq
[n
].vring
.desc
+
1771 virtio_queue_get_used_size(vdev
, n
);
1774 uint16_t virtio_queue_get_last_avail_idx(VirtIODevice
*vdev
, int n
)
1776 return vdev
->vq
[n
].last_avail_idx
;
1779 void virtio_queue_set_last_avail_idx(VirtIODevice
*vdev
, int n
, uint16_t idx
)
1781 vdev
->vq
[n
].last_avail_idx
= idx
;
1782 vdev
->vq
[n
].shadow_avail_idx
= idx
;
1785 void virtio_queue_invalidate_signalled_used(VirtIODevice
*vdev
, int n
)
1787 vdev
->vq
[n
].signalled_used_valid
= false;
1790 VirtQueue
*virtio_get_queue(VirtIODevice
*vdev
, int n
)
1792 return vdev
->vq
+ n
;
1795 uint16_t virtio_get_queue_index(VirtQueue
*vq
)
1797 return vq
->queue_index
;
1800 static void virtio_queue_guest_notifier_read(EventNotifier
*n
)
1802 VirtQueue
*vq
= container_of(n
, VirtQueue
, guest_notifier
);
1803 if (event_notifier_test_and_clear(n
)) {
1808 void virtio_queue_set_guest_notifier_fd_handler(VirtQueue
*vq
, bool assign
,
1811 if (assign
&& !with_irqfd
) {
1812 event_notifier_set_handler(&vq
->guest_notifier
, false,
1813 virtio_queue_guest_notifier_read
);
1815 event_notifier_set_handler(&vq
->guest_notifier
, false, NULL
);
1818 /* Test and clear notifier before closing it,
1819 * in case poll callback didn't have time to run. */
1820 virtio_queue_guest_notifier_read(&vq
->guest_notifier
);
1824 EventNotifier
*virtio_queue_get_guest_notifier(VirtQueue
*vq
)
1826 return &vq
->guest_notifier
;
1829 static void virtio_queue_host_notifier_aio_read(EventNotifier
*n
)
1831 VirtQueue
*vq
= container_of(n
, VirtQueue
, host_notifier
);
1832 if (event_notifier_test_and_clear(n
)) {
1833 virtio_queue_notify_aio_vq(vq
);
1837 void virtio_queue_aio_set_host_notifier_handler(VirtQueue
*vq
, AioContext
*ctx
,
1838 VirtIOHandleOutput handle_output
)
1840 if (handle_output
) {
1841 vq
->handle_aio_output
= handle_output
;
1842 aio_set_event_notifier(ctx
, &vq
->host_notifier
, true,
1843 virtio_queue_host_notifier_aio_read
);
1845 aio_set_event_notifier(ctx
, &vq
->host_notifier
, true, NULL
);
1846 /* Test and clear notifier before after disabling event,
1847 * in case poll callback didn't have time to run. */
1848 virtio_queue_host_notifier_aio_read(&vq
->host_notifier
);
1849 vq
->handle_aio_output
= NULL
;
1853 static void virtio_queue_host_notifier_read(EventNotifier
*n
)
1855 VirtQueue
*vq
= container_of(n
, VirtQueue
, host_notifier
);
1856 if (event_notifier_test_and_clear(n
)) {
1857 virtio_queue_notify_vq(vq
);
1861 void virtio_queue_set_host_notifier_fd_handler(VirtQueue
*vq
, bool assign
,
1864 AioContext
*ctx
= qemu_get_aio_context();
1865 if (assign
&& set_handler
) {
1867 aio_set_event_notifier(ctx
, &vq
->host_notifier
, true,
1868 virtio_queue_host_notifier_read
);
1870 event_notifier_set_handler(&vq
->host_notifier
, true,
1871 virtio_queue_host_notifier_read
);
1875 aio_set_event_notifier(ctx
, &vq
->host_notifier
, true, NULL
);
1877 event_notifier_set_handler(&vq
->host_notifier
, true, NULL
);
1881 /* Test and clear notifier before after disabling event,
1882 * in case poll callback didn't have time to run. */
1883 virtio_queue_host_notifier_read(&vq
->host_notifier
);
1887 EventNotifier
*virtio_queue_get_host_notifier(VirtQueue
*vq
)
1889 return &vq
->host_notifier
;
1892 void virtio_device_set_child_bus_name(VirtIODevice
*vdev
, char *bus_name
)
1894 g_free(vdev
->bus_name
);
1895 vdev
->bus_name
= g_strdup(bus_name
);
1898 static void virtio_device_realize(DeviceState
*dev
, Error
**errp
)
1900 VirtIODevice
*vdev
= VIRTIO_DEVICE(dev
);
1901 VirtioDeviceClass
*vdc
= VIRTIO_DEVICE_GET_CLASS(dev
);
1904 if (vdc
->realize
!= NULL
) {
1905 vdc
->realize(dev
, &err
);
1907 error_propagate(errp
, err
);
1912 virtio_bus_device_plugged(vdev
, &err
);
1914 error_propagate(errp
, err
);
1919 static void virtio_device_unrealize(DeviceState
*dev
, Error
**errp
)
1921 VirtIODevice
*vdev
= VIRTIO_DEVICE(dev
);
1922 VirtioDeviceClass
*vdc
= VIRTIO_DEVICE_GET_CLASS(dev
);
1925 virtio_bus_device_unplugged(vdev
);
1927 if (vdc
->unrealize
!= NULL
) {
1928 vdc
->unrealize(dev
, &err
);
1930 error_propagate(errp
, err
);
1935 g_free(vdev
->bus_name
);
1936 vdev
->bus_name
= NULL
;
1939 static Property virtio_properties
[] = {
1940 DEFINE_VIRTIO_COMMON_FEATURES(VirtIODevice
, host_features
),
1941 DEFINE_PROP_END_OF_LIST(),
1944 static void virtio_device_class_init(ObjectClass
*klass
, void *data
)
1946 /* Set the default value here. */
1947 DeviceClass
*dc
= DEVICE_CLASS(klass
);
1949 dc
->realize
= virtio_device_realize
;
1950 dc
->unrealize
= virtio_device_unrealize
;
1951 dc
->bus_type
= TYPE_VIRTIO_BUS
;
1952 dc
->props
= virtio_properties
;
1955 static const TypeInfo virtio_device_info
= {
1956 .name
= TYPE_VIRTIO_DEVICE
,
1957 .parent
= TYPE_DEVICE
,
1958 .instance_size
= sizeof(VirtIODevice
),
1959 .class_init
= virtio_device_class_init
,
1961 .class_size
= sizeof(VirtioDeviceClass
),
1964 static void virtio_register_types(void)
1966 type_register_static(&virtio_device_info
);
1969 type_init(virtio_register_types
)