?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
libqtest: Inline g_assert_no_errno()
[qemu/armbru.git] / hw / virtio / virtio.c
blob94f5c8e52a4e20f4427c21ccc47a34ee67467b4f
1 /*
2 * Virtio Support
3 *
4 * Copyright IBM, Corp. 2007
5 *
6 * Authors:
7 * Anthony Liguori <aliguori@us.ibm.com>
8 *
9 * This work is licensed under the terms of the GNU GPL, version 2. See
10 * the COPYING file in the top-level directory.
11 *
12 */
13
14 #include "qemu/osdep.h"
15 #include "qapi/error.h"
16 #include "qemu-common.h"
17 #include "cpu.h"
18 #include "trace.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 "hw/virtio/virtio-access.h"
25 #include "sysemu/dma.h"
26
27 /*
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.
31 */
32 #define VIRTIO_PCI_VRING_ALIGN 4096
33
34 typedef struct VRingDesc
35 {
36 uint64_t addr;
37 uint32_t len;
38 uint16_t flags;
39 uint16_t next;
40 } VRingDesc;
41
42 typedef struct VRingAvail
43 {
44 uint16_t flags;
45 uint16_t idx;
46 uint16_t ring[0];
47 } VRingAvail;
48
49 typedef struct VRingUsedElem
50 {
51 uint32_t id;
52 uint32_t len;
53 } VRingUsedElem;
54
55 typedef struct VRingUsed
56 {
57 uint16_t flags;
58 uint16_t idx;
59 VRingUsedElem ring[0];
60 } VRingUsed;
61
62 typedef struct VRingMemoryRegionCaches {
63 struct rcu_head rcu;
64 MemoryRegionCache desc;
65 MemoryRegionCache avail;
66 MemoryRegionCache used;
67 } VRingMemoryRegionCaches;
68
69 typedef struct VRing
70 {
71 unsigned int num;
72 unsigned int num_default;
73 unsigned int align;
74 hwaddr desc;
75 hwaddr avail;
76 hwaddr used;
77 VRingMemoryRegionCaches *caches;
78 } VRing;
79
80 struct VirtQueue
81 {
82 VRing vring;
83
84 /* Next head to pop */
85 uint16_t last_avail_idx;
86
87 /* Last avail_idx read from VQ. */
88 uint16_t shadow_avail_idx;
89
90 uint16_t used_idx;
91
92 /* Last used index value we have signalled on */
93 uint16_t signalled_used;
94
95 /* Last used index value we have signalled on */
96 bool signalled_used_valid;
97
98 /* Notification enabled? */
99 bool notification;
100
101 uint16_t queue_index;
102
103 unsigned int inuse;
104
105 uint16_t vector;
106 VirtIOHandleOutput handle_output;
107 VirtIOHandleAIOOutput handle_aio_output;
108 VirtIODevice *vdev;
109 EventNotifier guest_notifier;
110 EventNotifier host_notifier;
111 QLIST_ENTRY(VirtQueue) node;
112 };
113
114 static void virtio_free_region_cache(VRingMemoryRegionCaches *caches)
115 {
116 if (!caches) {
117 return;
118 }
119
120 address_space_cache_destroy(&caches->desc);
121 address_space_cache_destroy(&caches->avail);
122 address_space_cache_destroy(&caches->used);
123 g_free(caches);
124 }
125
126 static void virtio_virtqueue_reset_region_cache(struct VirtQueue *vq)
127 {
128 VRingMemoryRegionCaches *caches;
129
130 caches = atomic_read(&vq->vring.caches);
131 atomic_rcu_set(&vq->vring.caches, NULL);
132 if (caches) {
133 call_rcu(caches, virtio_free_region_cache, rcu);
134 }
135 }
136
137 static void virtio_init_region_cache(VirtIODevice *vdev, int n)
138 {
139 VirtQueue *vq = &vdev->vq[n];
140 VRingMemoryRegionCaches *old = vq->vring.caches;
141 VRingMemoryRegionCaches *new = NULL;
142 hwaddr addr, size;
143 int event_size;
144 int64_t len;
145
146 event_size = virtio_vdev_has_feature(vq->vdev, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0;
147
148 addr = vq->vring.desc;
149 if (!addr) {
150 goto out_no_cache;
151 }
152 new = g_new0(VRingMemoryRegionCaches, 1);
153 size = virtio_queue_get_desc_size(vdev, n);
154 len = address_space_cache_init(&new->desc, vdev->dma_as,
155 addr, size, false);
156 if (len < size) {
157 virtio_error(vdev, "Cannot map desc");
158 goto err_desc;
159 }
160
161 size = virtio_queue_get_used_size(vdev, n) + event_size;
162 len = address_space_cache_init(&new->used, vdev->dma_as,
163 vq->vring.used, size, true);
164 if (len < size) {
165 virtio_error(vdev, "Cannot map used");
166 goto err_used;
167 }
168
169 size = virtio_queue_get_avail_size(vdev, n) + event_size;
170 len = address_space_cache_init(&new->avail, vdev->dma_as,
171 vq->vring.avail, size, false);
172 if (len < size) {
173 virtio_error(vdev, "Cannot map avail");
174 goto err_avail;
175 }
176
177 atomic_rcu_set(&vq->vring.caches, new);
178 if (old) {
179 call_rcu(old, virtio_free_region_cache, rcu);
180 }
181 return;
182
183 err_avail:
184 address_space_cache_destroy(&new->avail);
185 err_used:
186 address_space_cache_destroy(&new->used);
187 err_desc:
188 address_space_cache_destroy(&new->desc);
189 out_no_cache:
190 g_free(new);
191 virtio_virtqueue_reset_region_cache(vq);
192 }
193
194 /* virt queue functions */
195 void virtio_queue_update_rings(VirtIODevice *vdev, int n)
196 {
197 VRing *vring = &vdev->vq[n].vring;
198
199 if (!vring->num || !vring->desc || !vring->align) {
200 /* not yet setup -> nothing to do */
201 return;
202 }
203 vring->avail = vring->desc + vring->num * sizeof(VRingDesc);
204 vring->used = vring_align(vring->avail +
205 offsetof(VRingAvail, ring[vring->num]),
206 vring->align);
207 virtio_init_region_cache(vdev, n);
208 }
209
210 /* Called within rcu_read_lock(). */
211 static void vring_desc_read(VirtIODevice *vdev, VRingDesc *desc,
212 MemoryRegionCache *cache, int i)
213 {
214 address_space_read_cached(cache, i * sizeof(VRingDesc),
215 desc, sizeof(VRingDesc));
216 virtio_tswap64s(vdev, &desc->addr);
217 virtio_tswap32s(vdev, &desc->len);
218 virtio_tswap16s(vdev, &desc->flags);
219 virtio_tswap16s(vdev, &desc->next);
220 }
221
222 static VRingMemoryRegionCaches *vring_get_region_caches(struct VirtQueue *vq)
223 {
224 VRingMemoryRegionCaches *caches = atomic_rcu_read(&vq->vring.caches);
225 assert(caches != NULL);
226 return caches;
227 }
228 /* Called within rcu_read_lock(). */
229 static inline uint16_t vring_avail_flags(VirtQueue *vq)
230 {
231 VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
232 hwaddr pa = offsetof(VRingAvail, flags);
233 return virtio_lduw_phys_cached(vq->vdev, &caches->avail, pa);
234 }
235
236 /* Called within rcu_read_lock(). */
237 static inline uint16_t vring_avail_idx(VirtQueue *vq)
238 {
239 VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
240 hwaddr pa = offsetof(VRingAvail, idx);
241 vq->shadow_avail_idx = virtio_lduw_phys_cached(vq->vdev, &caches->avail, pa);
242 return vq->shadow_avail_idx;
243 }
244
245 /* Called within rcu_read_lock(). */
246 static inline uint16_t vring_avail_ring(VirtQueue *vq, int i)
247 {
248 VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
249 hwaddr pa = offsetof(VRingAvail, ring[i]);
250 return virtio_lduw_phys_cached(vq->vdev, &caches->avail, pa);
251 }
252
253 /* Called within rcu_read_lock(). */
254 static inline uint16_t vring_get_used_event(VirtQueue *vq)
255 {
256 return vring_avail_ring(vq, vq->vring.num);
257 }
258
259 /* Called within rcu_read_lock(). */
260 static inline void vring_used_write(VirtQueue *vq, VRingUsedElem *uelem,
261 int i)
262 {
263 VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
264 hwaddr pa = offsetof(VRingUsed, ring[i]);
265 virtio_tswap32s(vq->vdev, &uelem->id);
266 virtio_tswap32s(vq->vdev, &uelem->len);
267 address_space_write_cached(&caches->used, pa, uelem, sizeof(VRingUsedElem));
268 address_space_cache_invalidate(&caches->used, pa, sizeof(VRingUsedElem));
269 }
270
271 /* Called within rcu_read_lock(). */
272 static uint16_t vring_used_idx(VirtQueue *vq)
273 {
274 VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
275 hwaddr pa = offsetof(VRingUsed, idx);
276 return virtio_lduw_phys_cached(vq->vdev, &caches->used, pa);
277 }
278
279 /* Called within rcu_read_lock(). */
280 static inline void vring_used_idx_set(VirtQueue *vq, uint16_t val)
281 {
282 VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
283 hwaddr pa = offsetof(VRingUsed, idx);
284 virtio_stw_phys_cached(vq->vdev, &caches->used, pa, val);
285 address_space_cache_invalidate(&caches->used, pa, sizeof(val));
286 vq->used_idx = val;
287 }
288
289 /* Called within rcu_read_lock(). */
290 static inline void vring_used_flags_set_bit(VirtQueue *vq, int mask)
291 {
292 VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
293 VirtIODevice *vdev = vq->vdev;
294 hwaddr pa = offsetof(VRingUsed, flags);
295 uint16_t flags = virtio_lduw_phys_cached(vq->vdev, &caches->used, pa);
296
297 virtio_stw_phys_cached(vdev, &caches->used, pa, flags | mask);
298 address_space_cache_invalidate(&caches->used, pa, sizeof(flags));
299 }
300
301 /* Called within rcu_read_lock(). */
302 static inline void vring_used_flags_unset_bit(VirtQueue *vq, int mask)
303 {
304 VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
305 VirtIODevice *vdev = vq->vdev;
306 hwaddr pa = offsetof(VRingUsed, flags);
307 uint16_t flags = virtio_lduw_phys_cached(vq->vdev, &caches->used, pa);
308
309 virtio_stw_phys_cached(vdev, &caches->used, pa, flags & ~mask);
310 address_space_cache_invalidate(&caches->used, pa, sizeof(flags));
311 }
312
313 /* Called within rcu_read_lock(). */
314 static inline void vring_set_avail_event(VirtQueue *vq, uint16_t val)
315 {
316 VRingMemoryRegionCaches *caches;
317 hwaddr pa;
318 if (!vq->notification) {
319 return;
320 }
321
322 caches = vring_get_region_caches(vq);
323 pa = offsetof(VRingUsed, ring[vq->vring.num]);
324 virtio_stw_phys_cached(vq->vdev, &caches->used, pa, val);
325 address_space_cache_invalidate(&caches->used, pa, sizeof(val));
326 }
327
328 void virtio_queue_set_notification(VirtQueue *vq, int enable)
329 {
330 vq->notification = enable;
331
332 if (!vq->vring.desc) {
333 return;
334 }
335
336 rcu_read_lock();
337 if (virtio_vdev_has_feature(vq->vdev, VIRTIO_RING_F_EVENT_IDX)) {
338 vring_set_avail_event(vq, vring_avail_idx(vq));
339 } else if (enable) {
340 vring_used_flags_unset_bit(vq, VRING_USED_F_NO_NOTIFY);
341 } else {
342 vring_used_flags_set_bit(vq, VRING_USED_F_NO_NOTIFY);
343 }
344 if (enable) {
345 /* Expose avail event/used flags before caller checks the avail idx. */
346 smp_mb();
347 }
348 rcu_read_unlock();
349 }
350
351 int virtio_queue_ready(VirtQueue *vq)
352 {
353 return vq->vring.avail != 0;
354 }
355
356 /* Fetch avail_idx from VQ memory only when we really need to know if
357 * guest has added some buffers.
358 * Called within rcu_read_lock(). */
359 static int virtio_queue_empty_rcu(VirtQueue *vq)
360 {
361 if (unlikely(vq->vdev->broken)) {
362 return 1;
363 }
364
365 if (unlikely(!vq->vring.avail)) {
366 return 1;
367 }
368
369 if (vq->shadow_avail_idx != vq->last_avail_idx) {
370 return 0;
371 }
372
373 return vring_avail_idx(vq) == vq->last_avail_idx;
374 }
375
376 int virtio_queue_empty(VirtQueue *vq)
377 {
378 bool empty;
379
380 if (unlikely(vq->vdev->broken)) {
381 return 1;
382 }
383
384 if (unlikely(!vq->vring.avail)) {
385 return 1;
386 }
387
388 if (vq->shadow_avail_idx != vq->last_avail_idx) {
389 return 0;
390 }
391
392 rcu_read_lock();
393 empty = vring_avail_idx(vq) == vq->last_avail_idx;
394 rcu_read_unlock();
395 return empty;
396 }
397
398 static void virtqueue_unmap_sg(VirtQueue *vq, const VirtQueueElement *elem,
399 unsigned int len)
400 {
401 AddressSpace *dma_as = vq->vdev->dma_as;
402 unsigned int offset;
403 int i;
404
405 offset = 0;
406 for (i = 0; i < elem->in_num; i++) {
407 size_t size = MIN(len - offset, elem->in_sg[i].iov_len);
408
409 dma_memory_unmap(dma_as, elem->in_sg[i].iov_base,
410 elem->in_sg[i].iov_len,
411 DMA_DIRECTION_FROM_DEVICE, size);
412
413 offset += size;
414 }
415
416 for (i = 0; i < elem->out_num; i++)
417 dma_memory_unmap(dma_as, elem->out_sg[i].iov_base,
418 elem->out_sg[i].iov_len,
419 DMA_DIRECTION_TO_DEVICE,
420 elem->out_sg[i].iov_len);
421 }
422
423 /* virtqueue_detach_element:
424 * @vq: The #VirtQueue
425 * @elem: The #VirtQueueElement
426 * @len: number of bytes written
427 *
428 * Detach the element from the virtqueue. This function is suitable for device
429 * reset or other situations where a #VirtQueueElement is simply freed and will
430 * not be pushed or discarded.
431 */
432 void virtqueue_detach_element(VirtQueue *vq, const VirtQueueElement *elem,
433 unsigned int len)
434 {
435 vq->inuse--;
436 virtqueue_unmap_sg(vq, elem, len);
437 }
438
439 /* virtqueue_unpop:
440 * @vq: The #VirtQueue
441 * @elem: The #VirtQueueElement
442 * @len: number of bytes written
443 *
444 * Pretend the most recent element wasn't popped from the virtqueue. The next
445 * call to virtqueue_pop() will refetch the element.
446 */
447 void virtqueue_unpop(VirtQueue *vq, const VirtQueueElement *elem,
448 unsigned int len)
449 {
450 vq->last_avail_idx--;
451 virtqueue_detach_element(vq, elem, len);
452 }
453
454 /* virtqueue_rewind:
455 * @vq: The #VirtQueue
456 * @num: Number of elements to push back
457 *
458 * Pretend that elements weren't popped from the virtqueue. The next
459 * virtqueue_pop() will refetch the oldest element.
460 *
461 * Use virtqueue_unpop() instead if you have a VirtQueueElement.
462 *
463 * Returns: true on success, false if @num is greater than the number of in use
464 * elements.
465 */
466 bool virtqueue_rewind(VirtQueue *vq, unsigned int num)
467 {
468 if (num > vq->inuse) {
469 return false;
470 }
471 vq->last_avail_idx -= num;
472 vq->inuse -= num;
473 return true;
474 }
475
476 /* Called within rcu_read_lock(). */
477 void virtqueue_fill(VirtQueue *vq, const VirtQueueElement *elem,
478 unsigned int len, unsigned int idx)
479 {
480 VRingUsedElem uelem;
481
482 trace_virtqueue_fill(vq, elem, len, idx);
483
484 virtqueue_unmap_sg(vq, elem, len);
485
486 if (unlikely(vq->vdev->broken)) {
487 return;
488 }
489
490 if (unlikely(!vq->vring.used)) {
491 return;
492 }
493
494 idx = (idx + vq->used_idx) % vq->vring.num;
495
496 uelem.id = elem->index;
497 uelem.len = len;
498 vring_used_write(vq, &uelem, idx);
499 }
500
501 /* Called within rcu_read_lock(). */
502 void virtqueue_flush(VirtQueue *vq, unsigned int count)
503 {
504 uint16_t old, new;
505
506 if (unlikely(vq->vdev->broken)) {
507 vq->inuse -= count;
508 return;
509 }
510
511 if (unlikely(!vq->vring.used)) {
512 return;
513 }
514
515 /* Make sure buffer is written before we update index. */
516 smp_wmb();
517 trace_virtqueue_flush(vq, count);
518 old = vq->used_idx;
519 new = old + count;
520 vring_used_idx_set(vq, new);
521 vq->inuse -= count;
522 if (unlikely((int16_t)(new - vq->signalled_used) < (uint16_t)(new - old)))
523 vq->signalled_used_valid = false;
524 }
525
526 void virtqueue_push(VirtQueue *vq, const VirtQueueElement *elem,
527 unsigned int len)
528 {
529 rcu_read_lock();
530 virtqueue_fill(vq, elem, len, 0);
531 virtqueue_flush(vq, 1);
532 rcu_read_unlock();
533 }
534
535 /* Called within rcu_read_lock(). */
536 static int virtqueue_num_heads(VirtQueue *vq, unsigned int idx)
537 {
538 uint16_t num_heads = vring_avail_idx(vq) - idx;
539
540 /* Check it isn't doing very strange things with descriptor numbers. */
541 if (num_heads > vq->vring.num) {
542 virtio_error(vq->vdev, "Guest moved used index from %u to %u",
543 idx, vq->shadow_avail_idx);
544 return -EINVAL;
545 }
546 /* On success, callers read a descriptor at vq->last_avail_idx.
547 * Make sure descriptor read does not bypass avail index read. */
548 if (num_heads) {
549 smp_rmb();
550 }
551
552 return num_heads;
553 }
554
555 /* Called within rcu_read_lock(). */
556 static bool virtqueue_get_head(VirtQueue *vq, unsigned int idx,
557 unsigned int *head)
558 {
559 /* Grab the next descriptor number they're advertising, and increment
560 * the index we've seen. */
561 *head = vring_avail_ring(vq, idx % vq->vring.num);
562
563 /* If their number is silly, that's a fatal mistake. */
564 if (*head >= vq->vring.num) {
565 virtio_error(vq->vdev, "Guest says index %u is available", *head);
566 return false;
567 }
568
569 return true;
570 }
571
572 enum {
573 VIRTQUEUE_READ_DESC_ERROR = -1,
574 VIRTQUEUE_READ_DESC_DONE = 0, /* end of chain */
575 VIRTQUEUE_READ_DESC_MORE = 1, /* more buffers in chain */
576 };
577
578 static int virtqueue_read_next_desc(VirtIODevice *vdev, VRingDesc *desc,
579 MemoryRegionCache *desc_cache, unsigned int max,
580 unsigned int *next)
581 {
582 /* If this descriptor says it doesn't chain, we're done. */
583 if (!(desc->flags & VRING_DESC_F_NEXT)) {
584 return VIRTQUEUE_READ_DESC_DONE;
585 }
586
587 /* Check they're not leading us off end of descriptors. */
588 *next = desc->next;
589 /* Make sure compiler knows to grab that: we don't want it changing! */
590 smp_wmb();
591
592 if (*next >= max) {
593 virtio_error(vdev, "Desc next is %u", *next);
594 return VIRTQUEUE_READ_DESC_ERROR;
595 }
596
597 vring_desc_read(vdev, desc, desc_cache, *next);
598 return VIRTQUEUE_READ_DESC_MORE;
599 }
600
601 void virtqueue_get_avail_bytes(VirtQueue *vq, unsigned int *in_bytes,
602 unsigned int *out_bytes,
603 unsigned max_in_bytes, unsigned max_out_bytes)
604 {
605 VirtIODevice *vdev = vq->vdev;
606 unsigned int max, idx;
607 unsigned int total_bufs, in_total, out_total;
608 VRingMemoryRegionCaches *caches;
609 MemoryRegionCache indirect_desc_cache = MEMORY_REGION_CACHE_INVALID;
610 int64_t len = 0;
611 int rc;
612
613 if (unlikely(!vq->vring.desc)) {
614 if (in_bytes) {
615 *in_bytes = 0;
616 }
617 if (out_bytes) {
618 *out_bytes = 0;
619 }
620 return;
621 }
622
623 rcu_read_lock();
624 idx = vq->last_avail_idx;
625 total_bufs = in_total = out_total = 0;
626
627 max = vq->vring.num;
628 caches = vring_get_region_caches(vq);
629 if (caches->desc.len < max * sizeof(VRingDesc)) {
630 virtio_error(vdev, "Cannot map descriptor ring");
631 goto err;
632 }
633
634 while ((rc = virtqueue_num_heads(vq, idx)) > 0) {
635 MemoryRegionCache *desc_cache = &caches->desc;
636 unsigned int num_bufs;
637 VRingDesc desc;
638 unsigned int i;
639
640 num_bufs = total_bufs;
641
642 if (!virtqueue_get_head(vq, idx++, &i)) {
643 goto err;
644 }
645
646 vring_desc_read(vdev, &desc, desc_cache, i);
647
648 if (desc.flags & VRING_DESC_F_INDIRECT) {
649 if (desc.len % sizeof(VRingDesc)) {
650 virtio_error(vdev, "Invalid size for indirect buffer table");
651 goto err;
652 }
653
654 /* If we've got too many, that implies a descriptor loop. */
655 if (num_bufs >= max) {
656 virtio_error(vdev, "Looped descriptor");
657 goto err;
658 }
659
660 /* loop over the indirect descriptor table */
661 len = address_space_cache_init(&indirect_desc_cache,
662 vdev->dma_as,
663 desc.addr, desc.len, false);
664 desc_cache = &indirect_desc_cache;
665 if (len < desc.len) {
666 virtio_error(vdev, "Cannot map indirect buffer");
667 goto err;
668 }
669
670 max = desc.len / sizeof(VRingDesc);
671 num_bufs = i = 0;
672 vring_desc_read(vdev, &desc, desc_cache, i);
673 }
674
675 do {
676 /* If we've got too many, that implies a descriptor loop. */
677 if (++num_bufs > max) {
678 virtio_error(vdev, "Looped descriptor");
679 goto err;
680 }
681
682 if (desc.flags & VRING_DESC_F_WRITE) {
683 in_total += desc.len;
684 } else {
685 out_total += desc.len;
686 }
687 if (in_total >= max_in_bytes && out_total >= max_out_bytes) {
688 goto done;
689 }
690
691 rc = virtqueue_read_next_desc(vdev, &desc, desc_cache, max, &i);
692 } while (rc == VIRTQUEUE_READ_DESC_MORE);
693
694 if (rc == VIRTQUEUE_READ_DESC_ERROR) {
695 goto err;
696 }
697
698 if (desc_cache == &indirect_desc_cache) {
699 address_space_cache_destroy(&indirect_desc_cache);
700 total_bufs++;
701 } else {
702 total_bufs = num_bufs;
703 }
704 }
705
706 if (rc < 0) {
707 goto err;
708 }
709
710 done:
711 address_space_cache_destroy(&indirect_desc_cache);
712 if (in_bytes) {
713 *in_bytes = in_total;
714 }
715 if (out_bytes) {
716 *out_bytes = out_total;
717 }
718 rcu_read_unlock();
719 return;
720
721 err:
722 in_total = out_total = 0;
723 goto done;
724 }
725
726 int virtqueue_avail_bytes(VirtQueue *vq, unsigned int in_bytes,
727 unsigned int out_bytes)
728 {
729 unsigned int in_total, out_total;
730
731 virtqueue_get_avail_bytes(vq, &in_total, &out_total, in_bytes, out_bytes);
732 return in_bytes <= in_total && out_bytes <= out_total;
733 }
734
735 static bool virtqueue_map_desc(VirtIODevice *vdev, unsigned int *p_num_sg,
736 hwaddr *addr, struct iovec *iov,
737 unsigned int max_num_sg, bool is_write,
738 hwaddr pa, size_t sz)
739 {
740 bool ok = false;
741 unsigned num_sg = *p_num_sg;
742 assert(num_sg <= max_num_sg);
743
744 if (!sz) {
745 virtio_error(vdev, "virtio: zero sized buffers are not allowed");
746 goto out;
747 }
748
749 while (sz) {
750 hwaddr len = sz;
751
752 if (num_sg == max_num_sg) {
753 virtio_error(vdev, "virtio: too many write descriptors in "
754 "indirect table");
755 goto out;
756 }
757
758 iov[num_sg].iov_base = dma_memory_map(vdev->dma_as, pa, &len,
759 is_write ?
760 DMA_DIRECTION_FROM_DEVICE :
761 DMA_DIRECTION_TO_DEVICE);
762 if (!iov[num_sg].iov_base) {
763 virtio_error(vdev, "virtio: bogus descriptor or out of resources");
764 goto out;
765 }
766
767 iov[num_sg].iov_len = len;
768 addr[num_sg] = pa;
769
770 sz -= len;
771 pa += len;
772 num_sg++;
773 }
774 ok = true;
775
776 out:
777 *p_num_sg = num_sg;
778 return ok;
779 }
780
781 /* Only used by error code paths before we have a VirtQueueElement (therefore
782 * virtqueue_unmap_sg() can't be used). Assumes buffers weren't written to
783 * yet.
784 */
785 static void virtqueue_undo_map_desc(unsigned int out_num, unsigned int in_num,
786 struct iovec *iov)
787 {
788 unsigned int i;
789
790 for (i = 0; i < out_num + in_num; i++) {
791 int is_write = i >= out_num;
792
793 cpu_physical_memory_unmap(iov->iov_base, iov->iov_len, is_write, 0);
794 iov++;
795 }
796 }
797
798 static void virtqueue_map_iovec(VirtIODevice *vdev, struct iovec *sg,
799 hwaddr *addr, unsigned int *num_sg,
800 int is_write)
801 {
802 unsigned int i;
803 hwaddr len;
804
805 for (i = 0; i < *num_sg; i++) {
806 len = sg[i].iov_len;
807 sg[i].iov_base = dma_memory_map(vdev->dma_as,
808 addr[i], &len, is_write ?
809 DMA_DIRECTION_FROM_DEVICE :
810 DMA_DIRECTION_TO_DEVICE);
811 if (!sg[i].iov_base) {
812 error_report("virtio: error trying to map MMIO memory");
813 exit(1);
814 }
815 if (len != sg[i].iov_len) {
816 error_report("virtio: unexpected memory split");
817 exit(1);
818 }
819 }
820 }
821
822 void virtqueue_map(VirtIODevice *vdev, VirtQueueElement *elem)
823 {
824 virtqueue_map_iovec(vdev, elem->in_sg, elem->in_addr, &elem->in_num, 1);
825 virtqueue_map_iovec(vdev, elem->out_sg, elem->out_addr, &elem->out_num, 0);
826 }
827
828 static void *virtqueue_alloc_element(size_t sz, unsigned out_num, unsigned in_num)
829 {
830 VirtQueueElement *elem;
831 size_t in_addr_ofs = QEMU_ALIGN_UP(sz, __alignof__(elem->in_addr[0]));
832 size_t out_addr_ofs = in_addr_ofs + in_num * sizeof(elem->in_addr[0]);
833 size_t out_addr_end = out_addr_ofs + out_num * sizeof(elem->out_addr[0]);
834 size_t in_sg_ofs = QEMU_ALIGN_UP(out_addr_end, __alignof__(elem->in_sg[0]));
835 size_t out_sg_ofs = in_sg_ofs + in_num * sizeof(elem->in_sg[0]);
836 size_t out_sg_end = out_sg_ofs + out_num * sizeof(elem->out_sg[0]);
837
838 assert(sz >= sizeof(VirtQueueElement));
839 elem = g_malloc(out_sg_end);
840 trace_virtqueue_alloc_element(elem, sz, in_num, out_num);
841 elem->out_num = out_num;
842 elem->in_num = in_num;
843 elem->in_addr = (void *)elem + in_addr_ofs;
844 elem->out_addr = (void *)elem + out_addr_ofs;
845 elem->in_sg = (void *)elem + in_sg_ofs;
846 elem->out_sg = (void *)elem + out_sg_ofs;
847 return elem;
848 }
849
850 void *virtqueue_pop(VirtQueue *vq, size_t sz)
851 {
852 unsigned int i, head, max;
853 VRingMemoryRegionCaches *caches;
854 MemoryRegionCache indirect_desc_cache = MEMORY_REGION_CACHE_INVALID;
855 MemoryRegionCache *desc_cache;
856 int64_t len;
857 VirtIODevice *vdev = vq->vdev;
858 VirtQueueElement *elem = NULL;
859 unsigned out_num, in_num, elem_entries;
860 hwaddr addr[VIRTQUEUE_MAX_SIZE];
861 struct iovec iov[VIRTQUEUE_MAX_SIZE];
862 VRingDesc desc;
863 int rc;
864
865 if (unlikely(vdev->broken)) {
866 return NULL;
867 }
868 rcu_read_lock();
869 if (virtio_queue_empty_rcu(vq)) {
870 goto done;
871 }
872 /* Needed after virtio_queue_empty(), see comment in
873 * virtqueue_num_heads(). */
874 smp_rmb();
875
876 /* When we start there are none of either input nor output. */
877 out_num = in_num = elem_entries = 0;
878
879 max = vq->vring.num;
880
881 if (vq->inuse >= vq->vring.num) {
882 virtio_error(vdev, "Virtqueue size exceeded");
883 goto done;
884 }
885
886 if (!virtqueue_get_head(vq, vq->last_avail_idx++, &head)) {
887 goto done;
888 }
889
890 if (virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX)) {
891 vring_set_avail_event(vq, vq->last_avail_idx);
892 }
893
894 i = head;
895
896 caches = vring_get_region_caches(vq);
897 if (caches->desc.len < max * sizeof(VRingDesc)) {
898 virtio_error(vdev, "Cannot map descriptor ring");
899 goto done;
900 }
901
902 desc_cache = &caches->desc;
903 vring_desc_read(vdev, &desc, desc_cache, i);
904 if (desc.flags & VRING_DESC_F_INDIRECT) {
905 if (desc.len % sizeof(VRingDesc)) {
906 virtio_error(vdev, "Invalid size for indirect buffer table");
907 goto done;
908 }
909
910 /* loop over the indirect descriptor table */
911 len = address_space_cache_init(&indirect_desc_cache, vdev->dma_as,
912 desc.addr, desc.len, false);
913 desc_cache = &indirect_desc_cache;
914 if (len < desc.len) {
915 virtio_error(vdev, "Cannot map indirect buffer");
916 goto done;
917 }
918
919 max = desc.len / sizeof(VRingDesc);
920 i = 0;
921 vring_desc_read(vdev, &desc, desc_cache, i);
922 }
923
924 /* Collect all the descriptors */
925 do {
926 bool map_ok;
927
928 if (desc.flags & VRING_DESC_F_WRITE) {
929 map_ok = virtqueue_map_desc(vdev, &in_num, addr + out_num,
930 iov + out_num,
931 VIRTQUEUE_MAX_SIZE - out_num, true,
932 desc.addr, desc.len);
933 } else {
934 if (in_num) {
935 virtio_error(vdev, "Incorrect order for descriptors");
936 goto err_undo_map;
937 }
938 map_ok = virtqueue_map_desc(vdev, &out_num, addr, iov,
939 VIRTQUEUE_MAX_SIZE, false,
940 desc.addr, desc.len);
941 }
942 if (!map_ok) {
943 goto err_undo_map;
944 }
945
946 /* If we've got too many, that implies a descriptor loop. */
947 if (++elem_entries > max) {
948 virtio_error(vdev, "Looped descriptor");
949 goto err_undo_map;
950 }
951
952 rc = virtqueue_read_next_desc(vdev, &desc, desc_cache, max, &i);
953 } while (rc == VIRTQUEUE_READ_DESC_MORE);
954
955 if (rc == VIRTQUEUE_READ_DESC_ERROR) {
956 goto err_undo_map;
957 }
958
959 /* Now copy what we have collected and mapped */
960 elem = virtqueue_alloc_element(sz, out_num, in_num);
961 elem->index = head;
962 for (i = 0; i < out_num; i++) {
963 elem->out_addr[i] = addr[i];
964 elem->out_sg[i] = iov[i];
965 }
966 for (i = 0; i < in_num; i++) {
967 elem->in_addr[i] = addr[out_num + i];
968 elem->in_sg[i] = iov[out_num + i];
969 }
970
971 vq->inuse++;
972
973 trace_virtqueue_pop(vq, elem, elem->in_num, elem->out_num);
974 done:
975 address_space_cache_destroy(&indirect_desc_cache);
976 rcu_read_unlock();
977
978 return elem;
979
980 err_undo_map:
981 virtqueue_undo_map_desc(out_num, in_num, iov);
982 goto done;
983 }
984
985 /* virtqueue_drop_all:
986 * @vq: The #VirtQueue
987 * Drops all queued buffers and indicates them to the guest
988 * as if they are done. Useful when buffers can not be
989 * processed but must be returned to the guest.
990 */
991 unsigned int virtqueue_drop_all(VirtQueue *vq)
992 {
993 unsigned int dropped = 0;
994 VirtQueueElement elem = {};
995 VirtIODevice *vdev = vq->vdev;
996 bool fEventIdx = virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX);
997
998 if (unlikely(vdev->broken)) {
999 return 0;
1000 }
1001
1002 while (!virtio_queue_empty(vq) && vq->inuse < vq->vring.num) {
1003 /* works similar to virtqueue_pop but does not map buffers
1004 * and does not allocate any memory */
1005 smp_rmb();
1006 if (!virtqueue_get_head(vq, vq->last_avail_idx, &elem.index)) {
1007 break;
1008 }
1009 vq->inuse++;
1010 vq->last_avail_idx++;
1011 if (fEventIdx) {
1012 vring_set_avail_event(vq, vq->last_avail_idx);
1013 }
1014 /* immediately push the element, nothing to unmap
1015 * as both in_num and out_num are set to 0 */
1016 virtqueue_push(vq, &elem, 0);
1017 dropped++;
1018 }
1019
1020 return dropped;
1021 }
1022
1023 /* Reading and writing a structure directly to QEMUFile is *awful*, but
1024 * it is what QEMU has always done by mistake. We can change it sooner
1025 * or later by bumping the version number of the affected vm states.
1026 * In the meanwhile, since the in-memory layout of VirtQueueElement
1027 * has changed, we need to marshal to and from the layout that was
1028 * used before the change.
1029 */
1030 typedef struct VirtQueueElementOld {
1031 unsigned int index;
1032 unsigned int out_num;
1033 unsigned int in_num;
1034 hwaddr in_addr[VIRTQUEUE_MAX_SIZE];
1035 hwaddr out_addr[VIRTQUEUE_MAX_SIZE];
1036 struct iovec in_sg[VIRTQUEUE_MAX_SIZE];
1037 struct iovec out_sg[VIRTQUEUE_MAX_SIZE];
1038 } VirtQueueElementOld;
1039
1040 void *qemu_get_virtqueue_element(VirtIODevice *vdev, QEMUFile *f, size_t sz)
1041 {
1042 VirtQueueElement *elem;
1043 VirtQueueElementOld data;
1044 int i;
1045
1046 qemu_get_buffer(f, (uint8_t *)&data, sizeof(VirtQueueElementOld));
1047
1048 /* TODO: teach all callers that this can fail, and return failure instead
1049 * of asserting here.
1050 * This is just one thing (there are probably more) that must be
1051 * fixed before we can allow NDEBUG compilation.
1052 */
1053 assert(ARRAY_SIZE(data.in_addr) >= data.in_num);
1054 assert(ARRAY_SIZE(data.out_addr) >= data.out_num);
1055
1056 elem = virtqueue_alloc_element(sz, data.out_num, data.in_num);
1057 elem->index = data.index;
1058
1059 for (i = 0; i < elem->in_num; i++) {
1060 elem->in_addr[i] = data.in_addr[i];
1061 }
1062
1063 for (i = 0; i < elem->out_num; i++) {
1064 elem->out_addr[i] = data.out_addr[i];
1065 }
1066
1067 for (i = 0; i < elem->in_num; i++) {
1068 /* Base is overwritten by virtqueue_map. */
1069 elem->in_sg[i].iov_base = 0;
1070 elem->in_sg[i].iov_len = data.in_sg[i].iov_len;
1071 }
1072
1073 for (i = 0; i < elem->out_num; i++) {
1074 /* Base is overwritten by virtqueue_map. */
1075 elem->out_sg[i].iov_base = 0;
1076 elem->out_sg[i].iov_len = data.out_sg[i].iov_len;
1077 }
1078
1079 virtqueue_map(vdev, elem);
1080 return elem;
1081 }
1082
1083 void qemu_put_virtqueue_element(QEMUFile *f, VirtQueueElement *elem)
1084 {
1085 VirtQueueElementOld data;
1086 int i;
1087
1088 memset(&data, 0, sizeof(data));
1089 data.index = elem->index;
1090 data.in_num = elem->in_num;
1091 data.out_num = elem->out_num;
1092
1093 for (i = 0; i < elem->in_num; i++) {
1094 data.in_addr[i] = elem->in_addr[i];
1095 }
1096
1097 for (i = 0; i < elem->out_num; i++) {
1098 data.out_addr[i] = elem->out_addr[i];
1099 }
1100
1101 for (i = 0; i < elem->in_num; i++) {
1102 /* Base is overwritten by virtqueue_map when loading. Do not
1103 * save it, as it would leak the QEMU address space layout. */
1104 data.in_sg[i].iov_len = elem->in_sg[i].iov_len;
1105 }
1106
1107 for (i = 0; i < elem->out_num; i++) {
1108 /* Do not save iov_base as above. */
1109 data.out_sg[i].iov_len = elem->out_sg[i].iov_len;
1110 }
1111 qemu_put_buffer(f, (uint8_t *)&data, sizeof(VirtQueueElementOld));
1112 }
1113
1114 /* virtio device */
1115 static void virtio_notify_vector(VirtIODevice *vdev, uint16_t vector)
1116 {
1117 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
1118 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
1119
1120 if (unlikely(vdev->broken)) {
1121 return;
1122 }
1123
1124 if (k->notify) {
1125 k->notify(qbus->parent, vector);
1126 }
1127 }
1128
1129 void virtio_update_irq(VirtIODevice *vdev)
1130 {
1131 virtio_notify_vector(vdev, VIRTIO_NO_VECTOR);
1132 }
1133
1134 static int virtio_validate_features(VirtIODevice *vdev)
1135 {
1136 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1137
1138 if (virtio_host_has_feature(vdev, VIRTIO_F_IOMMU_PLATFORM) &&
1139 !virtio_vdev_has_feature(vdev, VIRTIO_F_IOMMU_PLATFORM)) {
1140 return -EFAULT;
1141 }
1142
1143 if (k->validate_features) {
1144 return k->validate_features(vdev);
1145 } else {
1146 return 0;
1147 }
1148 }
1149
1150 int virtio_set_status(VirtIODevice *vdev, uint8_t val)
1151 {
1152 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1153 trace_virtio_set_status(vdev, val);
1154
1155 if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
1156 if (!(vdev->status & VIRTIO_CONFIG_S_FEATURES_OK) &&
1157 val & VIRTIO_CONFIG_S_FEATURES_OK) {
1158 int ret = virtio_validate_features(vdev);
1159
1160 if (ret) {
1161 return ret;
1162 }
1163 }
1164 }
1165 if (k->set_status) {
1166 k->set_status(vdev, val);
1167 }
1168 vdev->status = val;
1169 return 0;
1170 }
1171
1172 bool target_words_bigendian(void);
1173 static enum virtio_device_endian virtio_default_endian(void)
1174 {
1175 if (target_words_bigendian()) {
1176 return VIRTIO_DEVICE_ENDIAN_BIG;
1177 } else {
1178 return VIRTIO_DEVICE_ENDIAN_LITTLE;
1179 }
1180 }
1181
1182 static enum virtio_device_endian virtio_current_cpu_endian(void)
1183 {
1184 CPUClass *cc = CPU_GET_CLASS(current_cpu);
1185
1186 if (cc->virtio_is_big_endian(current_cpu)) {
1187 return VIRTIO_DEVICE_ENDIAN_BIG;
1188 } else {
1189 return VIRTIO_DEVICE_ENDIAN_LITTLE;
1190 }
1191 }
1192
1193 void virtio_reset(void *opaque)
1194 {
1195 VirtIODevice *vdev = opaque;
1196 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1197 int i;
1198
1199 virtio_set_status(vdev, 0);
1200 if (current_cpu) {
1201 /* Guest initiated reset */
1202 vdev->device_endian = virtio_current_cpu_endian();
1203 } else {
1204 /* System reset */
1205 vdev->device_endian = virtio_default_endian();
1206 }
1207
1208 if (k->reset) {
1209 k->reset(vdev);
1210 }
1211
1212 vdev->broken = false;
1213 vdev->guest_features = 0;
1214 vdev->queue_sel = 0;
1215 vdev->status = 0;
1216 atomic_set(&vdev->isr, 0);
1217 vdev->config_vector = VIRTIO_NO_VECTOR;
1218 virtio_notify_vector(vdev, vdev->config_vector);
1219
1220 for(i = 0; i < VIRTIO_QUEUE_MAX; i++) {
1221 vdev->vq[i].vring.desc = 0;
1222 vdev->vq[i].vring.avail = 0;
1223 vdev->vq[i].vring.used = 0;
1224 vdev->vq[i].last_avail_idx = 0;
1225 vdev->vq[i].shadow_avail_idx = 0;
1226 vdev->vq[i].used_idx = 0;
1227 virtio_queue_set_vector(vdev, i, VIRTIO_NO_VECTOR);
1228 vdev->vq[i].signalled_used = 0;
1229 vdev->vq[i].signalled_used_valid = false;
1230 vdev->vq[i].notification = true;
1231 vdev->vq[i].vring.num = vdev->vq[i].vring.num_default;
1232 vdev->vq[i].inuse = 0;
1233 virtio_virtqueue_reset_region_cache(&vdev->vq[i]);
1234 }
1235 }
1236
1237 uint32_t virtio_config_readb(VirtIODevice *vdev, uint32_t addr)
1238 {
1239 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1240 uint8_t val;
1241
1242 if (addr + sizeof(val) > vdev->config_len) {
1243 return (uint32_t)-1;
1244 }
1245
1246 k->get_config(vdev, vdev->config);
1247
1248 val = ldub_p(vdev->config + addr);
1249 return val;
1250 }
1251
1252 uint32_t virtio_config_readw(VirtIODevice *vdev, uint32_t addr)
1253 {
1254 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1255 uint16_t val;
1256
1257 if (addr + sizeof(val) > vdev->config_len) {
1258 return (uint32_t)-1;
1259 }
1260
1261 k->get_config(vdev, vdev->config);
1262
1263 val = lduw_p(vdev->config + addr);
1264 return val;
1265 }
1266
1267 uint32_t virtio_config_readl(VirtIODevice *vdev, uint32_t addr)
1268 {
1269 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1270 uint32_t val;
1271
1272 if (addr + sizeof(val) > vdev->config_len) {
1273 return (uint32_t)-1;
1274 }
1275
1276 k->get_config(vdev, vdev->config);
1277
1278 val = ldl_p(vdev->config + addr);
1279 return val;
1280 }
1281
1282 void virtio_config_writeb(VirtIODevice *vdev, uint32_t addr, uint32_t data)
1283 {
1284 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1285 uint8_t val = data;
1286
1287 if (addr + sizeof(val) > vdev->config_len) {
1288 return;
1289 }
1290
1291 stb_p(vdev->config + addr, val);
1292
1293 if (k->set_config) {
1294 k->set_config(vdev, vdev->config);
1295 }
1296 }
1297
1298 void virtio_config_writew(VirtIODevice *vdev, uint32_t addr, uint32_t data)
1299 {
1300 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1301 uint16_t val = data;
1302
1303 if (addr + sizeof(val) > vdev->config_len) {
1304 return;
1305 }
1306
1307 stw_p(vdev->config + addr, val);
1308
1309 if (k->set_config) {
1310 k->set_config(vdev, vdev->config);
1311 }
1312 }
1313
1314 void virtio_config_writel(VirtIODevice *vdev, uint32_t addr, uint32_t data)
1315 {
1316 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1317 uint32_t val = data;
1318
1319 if (addr + sizeof(val) > vdev->config_len) {
1320 return;
1321 }
1322
1323 stl_p(vdev->config + addr, val);
1324
1325 if (k->set_config) {
1326 k->set_config(vdev, vdev->config);
1327 }
1328 }
1329
1330 uint32_t virtio_config_modern_readb(VirtIODevice *vdev, uint32_t addr)
1331 {
1332 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1333 uint8_t val;
1334
1335 if (addr + sizeof(val) > vdev->config_len) {
1336 return (uint32_t)-1;
1337 }
1338
1339 k->get_config(vdev, vdev->config);
1340
1341 val = ldub_p(vdev->config + addr);
1342 return val;
1343 }
1344
1345 uint32_t virtio_config_modern_readw(VirtIODevice *vdev, uint32_t addr)
1346 {
1347 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1348 uint16_t val;
1349
1350 if (addr + sizeof(val) > vdev->config_len) {
1351 return (uint32_t)-1;
1352 }
1353
1354 k->get_config(vdev, vdev->config);
1355
1356 val = lduw_le_p(vdev->config + addr);
1357 return val;
1358 }
1359
1360 uint32_t virtio_config_modern_readl(VirtIODevice *vdev, uint32_t addr)
1361 {
1362 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1363 uint32_t val;
1364
1365 if (addr + sizeof(val) > vdev->config_len) {
1366 return (uint32_t)-1;
1367 }
1368
1369 k->get_config(vdev, vdev->config);
1370
1371 val = ldl_le_p(vdev->config + addr);
1372 return val;
1373 }
1374
1375 void virtio_config_modern_writeb(VirtIODevice *vdev,
1376 uint32_t addr, uint32_t data)
1377 {
1378 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1379 uint8_t val = data;
1380
1381 if (addr + sizeof(val) > vdev->config_len) {
1382 return;
1383 }
1384
1385 stb_p(vdev->config + addr, val);
1386
1387 if (k->set_config) {
1388 k->set_config(vdev, vdev->config);
1389 }
1390 }
1391
1392 void virtio_config_modern_writew(VirtIODevice *vdev,
1393 uint32_t addr, uint32_t data)
1394 {
1395 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1396 uint16_t val = data;
1397
1398 if (addr + sizeof(val) > vdev->config_len) {
1399 return;
1400 }
1401
1402 stw_le_p(vdev->config + addr, val);
1403
1404 if (k->set_config) {
1405 k->set_config(vdev, vdev->config);
1406 }
1407 }
1408
1409 void virtio_config_modern_writel(VirtIODevice *vdev,
1410 uint32_t addr, uint32_t data)
1411 {
1412 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1413 uint32_t val = data;
1414
1415 if (addr + sizeof(val) > vdev->config_len) {
1416 return;
1417 }
1418
1419 stl_le_p(vdev->config + addr, val);
1420
1421 if (k->set_config) {
1422 k->set_config(vdev, vdev->config);
1423 }
1424 }
1425
1426 void virtio_queue_set_addr(VirtIODevice *vdev, int n, hwaddr addr)
1427 {
1428 if (!vdev->vq[n].vring.num) {
1429 return;
1430 }
1431 vdev->vq[n].vring.desc = addr;
1432 virtio_queue_update_rings(vdev, n);
1433 }
1434
1435 hwaddr virtio_queue_get_addr(VirtIODevice *vdev, int n)
1436 {
1437 return vdev->vq[n].vring.desc;
1438 }
1439
1440 void virtio_queue_set_rings(VirtIODevice *vdev, int n, hwaddr desc,
1441 hwaddr avail, hwaddr used)
1442 {
1443 if (!vdev->vq[n].vring.num) {
1444 return;
1445 }
1446 vdev->vq[n].vring.desc = desc;
1447 vdev->vq[n].vring.avail = avail;
1448 vdev->vq[n].vring.used = used;
1449 virtio_init_region_cache(vdev, n);
1450 }
1451
1452 void virtio_queue_set_num(VirtIODevice *vdev, int n, int num)
1453 {
1454 /* Don't allow guest to flip queue between existent and
1455 * nonexistent states, or to set it to an invalid size.
1456 */
1457 if (!!num != !!vdev->vq[n].vring.num ||
1458 num > VIRTQUEUE_MAX_SIZE ||
1459 num < 0) {
1460 return;
1461 }
1462 vdev->vq[n].vring.num = num;
1463 }
1464
1465 VirtQueue *virtio_vector_first_queue(VirtIODevice *vdev, uint16_t vector)
1466 {
1467 return QLIST_FIRST(&vdev->vector_queues[vector]);
1468 }
1469
1470 VirtQueue *virtio_vector_next_queue(VirtQueue *vq)
1471 {
1472 return QLIST_NEXT(vq, node);
1473 }
1474
1475 int virtio_queue_get_num(VirtIODevice *vdev, int n)
1476 {
1477 return vdev->vq[n].vring.num;
1478 }
1479
1480 int virtio_queue_get_max_num(VirtIODevice *vdev, int n)
1481 {
1482 return vdev->vq[n].vring.num_default;
1483 }
1484
1485 int virtio_get_num_queues(VirtIODevice *vdev)
1486 {
1487 int i;
1488
1489 for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
1490 if (!virtio_queue_get_num(vdev, i)) {
1491 break;
1492 }
1493 }
1494
1495 return i;
1496 }
1497
1498 void virtio_queue_set_align(VirtIODevice *vdev, int n, int align)
1499 {
1500 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
1501 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
1502
1503 /* virtio-1 compliant devices cannot change the alignment */
1504 if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
1505 error_report("tried to modify queue alignment for virtio-1 device");
1506 return;
1507 }
1508 /* Check that the transport told us it was going to do this
1509 * (so a buggy transport will immediately assert rather than
1510 * silently failing to migrate this state)
1511 */
1512 assert(k->has_variable_vring_alignment);
1513
1514 if (align) {
1515 vdev->vq[n].vring.align = align;
1516 virtio_queue_update_rings(vdev, n);
1517 }
1518 }
1519
1520 static bool virtio_queue_notify_aio_vq(VirtQueue *vq)
1521 {
1522 if (vq->vring.desc && vq->handle_aio_output) {
1523 VirtIODevice *vdev = vq->vdev;
1524
1525 trace_virtio_queue_notify(vdev, vq - vdev->vq, vq);
1526 return vq->handle_aio_output(vdev, vq);
1527 }
1528
1529 return false;
1530 }
1531
1532 static void virtio_queue_notify_vq(VirtQueue *vq)
1533 {
1534 if (vq->vring.desc && vq->handle_output) {
1535 VirtIODevice *vdev = vq->vdev;
1536
1537 if (unlikely(vdev->broken)) {
1538 return;
1539 }
1540
1541 trace_virtio_queue_notify(vdev, vq - vdev->vq, vq);
1542 vq->handle_output(vdev, vq);
1543 }
1544 }
1545
1546 void virtio_queue_notify(VirtIODevice *vdev, int n)
1547 {
1548 VirtQueue *vq = &vdev->vq[n];
1549
1550 if (unlikely(!vq->vring.desc || vdev->broken)) {
1551 return;
1552 }
1553
1554 trace_virtio_queue_notify(vdev, vq - vdev->vq, vq);
1555 if (vq->handle_aio_output) {
1556 event_notifier_set(&vq->host_notifier);
1557 } else if (vq->handle_output) {
1558 vq->handle_output(vdev, vq);
1559 }
1560 }
1561
1562 uint16_t virtio_queue_vector(VirtIODevice *vdev, int n)
1563 {
1564 return n < VIRTIO_QUEUE_MAX ? vdev->vq[n].vector :
1565 VIRTIO_NO_VECTOR;
1566 }
1567
1568 void virtio_queue_set_vector(VirtIODevice *vdev, int n, uint16_t vector)
1569 {
1570 VirtQueue *vq = &vdev->vq[n];
1571
1572 if (n < VIRTIO_QUEUE_MAX) {
1573 if (vdev->vector_queues &&
1574 vdev->vq[n].vector != VIRTIO_NO_VECTOR) {
1575 QLIST_REMOVE(vq, node);
1576 }
1577 vdev->vq[n].vector = vector;
1578 if (vdev->vector_queues &&
1579 vector != VIRTIO_NO_VECTOR) {
1580 QLIST_INSERT_HEAD(&vdev->vector_queues[vector], vq, node);
1581 }
1582 }
1583 }
1584
1585 VirtQueue *virtio_add_queue(VirtIODevice *vdev, int queue_size,
1586 VirtIOHandleOutput handle_output)
1587 {
1588 int i;
1589
1590 for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
1591 if (vdev->vq[i].vring.num == 0)
1592 break;
1593 }
1594
1595 if (i == VIRTIO_QUEUE_MAX || queue_size > VIRTQUEUE_MAX_SIZE)
1596 abort();
1597
1598 vdev->vq[i].vring.num = queue_size;
1599 vdev->vq[i].vring.num_default = queue_size;
1600 vdev->vq[i].vring.align = VIRTIO_PCI_VRING_ALIGN;
1601 vdev->vq[i].handle_output = handle_output;
1602 vdev->vq[i].handle_aio_output = NULL;
1603
1604 return &vdev->vq[i];
1605 }
1606
1607 void virtio_del_queue(VirtIODevice *vdev, int n)
1608 {
1609 if (n < 0 || n >= VIRTIO_QUEUE_MAX) {
1610 abort();
1611 }
1612
1613 vdev->vq[n].vring.num = 0;
1614 vdev->vq[n].vring.num_default = 0;
1615 }
1616
1617 static void virtio_set_isr(VirtIODevice *vdev, int value)
1618 {
1619 uint8_t old = atomic_read(&vdev->isr);
1620
1621 /* Do not write ISR if it does not change, so that its cacheline remains
1622 * shared in the common case where the guest does not read it.
1623 */
1624 if ((old & value) != value) {
1625 atomic_or(&vdev->isr, value);
1626 }
1627 }
1628
1629 /* Called within rcu_read_lock(). */
1630 static bool virtio_should_notify(VirtIODevice *vdev, VirtQueue *vq)
1631 {
1632 uint16_t old, new;
1633 bool v;
1634 /* We need to expose used array entries before checking used event. */
1635 smp_mb();
1636 /* Always notify when queue is empty (when feature acknowledge) */
1637 if (virtio_vdev_has_feature(vdev, VIRTIO_F_NOTIFY_ON_EMPTY) &&
1638 !vq->inuse && virtio_queue_empty(vq)) {
1639 return true;
1640 }
1641
1642 if (!virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX)) {
1643 return !(vring_avail_flags(vq) & VRING_AVAIL_F_NO_INTERRUPT);
1644 }
1645
1646 v = vq->signalled_used_valid;
1647 vq->signalled_used_valid = true;
1648 old = vq->signalled_used;
1649 new = vq->signalled_used = vq->used_idx;
1650 return !v || vring_need_event(vring_get_used_event(vq), new, old);
1651 }
1652
1653 void virtio_notify_irqfd(VirtIODevice *vdev, VirtQueue *vq)
1654 {
1655 bool should_notify;
1656 rcu_read_lock();
1657 should_notify = virtio_should_notify(vdev, vq);
1658 rcu_read_unlock();
1659
1660 if (!should_notify) {
1661 return;
1662 }
1663
1664 trace_virtio_notify_irqfd(vdev, vq);
1665
1666 /*
1667 * virtio spec 1.0 says ISR bit 0 should be ignored with MSI, but
1668 * windows drivers included in virtio-win 1.8.0 (circa 2015) are
1669 * incorrectly polling this bit during crashdump and hibernation
1670 * in MSI mode, causing a hang if this bit is never updated.
1671 * Recent releases of Windows do not really shut down, but rather
1672 * log out and hibernate to make the next startup faster. Hence,
1673 * this manifested as a more serious hang during shutdown with
1674 *
1675 * Next driver release from 2016 fixed this problem, so working around it
1676 * is not a must, but it's easy to do so let's do it here.
1677 *
1678 * Note: it's safe to update ISR from any thread as it was switched
1679 * to an atomic operation.
1680 */
1681 virtio_set_isr(vq->vdev, 0x1);
1682 event_notifier_set(&vq->guest_notifier);
1683 }
1684
1685 static void virtio_irq(VirtQueue *vq)
1686 {
1687 virtio_set_isr(vq->vdev, 0x1);
1688 virtio_notify_vector(vq->vdev, vq->vector);
1689 }
1690
1691 void virtio_notify(VirtIODevice *vdev, VirtQueue *vq)
1692 {
1693 bool should_notify;
1694 rcu_read_lock();
1695 should_notify = virtio_should_notify(vdev, vq);
1696 rcu_read_unlock();
1697
1698 if (!should_notify) {
1699 return;
1700 }
1701
1702 trace_virtio_notify(vdev, vq);
1703 virtio_irq(vq);
1704 }
1705
1706 void virtio_notify_config(VirtIODevice *vdev)
1707 {
1708 if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK))
1709 return;
1710
1711 virtio_set_isr(vdev, 0x3);
1712 vdev->generation++;
1713 virtio_notify_vector(vdev, vdev->config_vector);
1714 }
1715
1716 static bool virtio_device_endian_needed(void *opaque)
1717 {
1718 VirtIODevice *vdev = opaque;
1719
1720 assert(vdev->device_endian != VIRTIO_DEVICE_ENDIAN_UNKNOWN);
1721 if (!virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
1722 return vdev->device_endian != virtio_default_endian();
1723 }
1724 /* Devices conforming to VIRTIO 1.0 or later are always LE. */
1725 return vdev->device_endian != VIRTIO_DEVICE_ENDIAN_LITTLE;
1726 }
1727
1728 static bool virtio_64bit_features_needed(void *opaque)
1729 {
1730 VirtIODevice *vdev = opaque;
1731
1732 return (vdev->host_features >> 32) != 0;
1733 }
1734
1735 static bool virtio_virtqueue_needed(void *opaque)
1736 {
1737 VirtIODevice *vdev = opaque;
1738
1739 return virtio_host_has_feature(vdev, VIRTIO_F_VERSION_1);
1740 }
1741
1742 static bool virtio_ringsize_needed(void *opaque)
1743 {
1744 VirtIODevice *vdev = opaque;
1745 int i;
1746
1747 for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
1748 if (vdev->vq[i].vring.num != vdev->vq[i].vring.num_default) {
1749 return true;
1750 }
1751 }
1752 return false;
1753 }
1754
1755 static bool virtio_extra_state_needed(void *opaque)
1756 {
1757 VirtIODevice *vdev = opaque;
1758 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
1759 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
1760
1761 return k->has_extra_state &&
1762 k->has_extra_state(qbus->parent);
1763 }
1764
1765 static bool virtio_broken_needed(void *opaque)
1766 {
1767 VirtIODevice *vdev = opaque;
1768
1769 return vdev->broken;
1770 }
1771
1772 static const VMStateDescription vmstate_virtqueue = {
1773 .name = "virtqueue_state",
1774 .version_id = 1,
1775 .minimum_version_id = 1,
1776 .fields = (VMStateField[]) {
1777 VMSTATE_UINT64(vring.avail, struct VirtQueue),
1778 VMSTATE_UINT64(vring.used, struct VirtQueue),
1779 VMSTATE_END_OF_LIST()
1780 }
1781 };
1782
1783 static const VMStateDescription vmstate_virtio_virtqueues = {
1784 .name = "virtio/virtqueues",
1785 .version_id = 1,
1786 .minimum_version_id = 1,
1787 .needed = &virtio_virtqueue_needed,
1788 .fields = (VMStateField[]) {
1789 VMSTATE_STRUCT_VARRAY_POINTER_KNOWN(vq, struct VirtIODevice,
1790 VIRTIO_QUEUE_MAX, 0, vmstate_virtqueue, VirtQueue),
1791 VMSTATE_END_OF_LIST()
1792 }
1793 };
1794
1795 static const VMStateDescription vmstate_ringsize = {
1796 .name = "ringsize_state",
1797 .version_id = 1,
1798 .minimum_version_id = 1,
1799 .fields = (VMStateField[]) {
1800 VMSTATE_UINT32(vring.num_default, struct VirtQueue),
1801 VMSTATE_END_OF_LIST()
1802 }
1803 };
1804
1805 static const VMStateDescription vmstate_virtio_ringsize = {
1806 .name = "virtio/ringsize",
1807 .version_id = 1,
1808 .minimum_version_id = 1,
1809 .needed = &virtio_ringsize_needed,
1810 .fields = (VMStateField[]) {
1811 VMSTATE_STRUCT_VARRAY_POINTER_KNOWN(vq, struct VirtIODevice,
1812 VIRTIO_QUEUE_MAX, 0, vmstate_ringsize, VirtQueue),
1813 VMSTATE_END_OF_LIST()
1814 }
1815 };
1816
1817 static int get_extra_state(QEMUFile *f, void *pv, size_t size,
1818 VMStateField *field)
1819 {
1820 VirtIODevice *vdev = pv;
1821 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
1822 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
1823
1824 if (!k->load_extra_state) {
1825 return -1;
1826 } else {
1827 return k->load_extra_state(qbus->parent, f);
1828 }
1829 }
1830
1831 static int put_extra_state(QEMUFile *f, void *pv, size_t size,
1832 VMStateField *field, QJSON *vmdesc)
1833 {
1834 VirtIODevice *vdev = pv;
1835 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
1836 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
1837
1838 k->save_extra_state(qbus->parent, f);
1839 return 0;
1840 }
1841
1842 static const VMStateInfo vmstate_info_extra_state = {
1843 .name = "virtqueue_extra_state",
1844 .get = get_extra_state,
1845 .put = put_extra_state,
1846 };
1847
1848 static const VMStateDescription vmstate_virtio_extra_state = {
1849 .name = "virtio/extra_state",
1850 .version_id = 1,
1851 .minimum_version_id = 1,
1852 .needed = &virtio_extra_state_needed,
1853 .fields = (VMStateField[]) {
1854 {
1855 .name = "extra_state",
1856 .version_id = 0,
1857 .field_exists = NULL,
1858 .size = 0,
1859 .info = &vmstate_info_extra_state,
1860 .flags = VMS_SINGLE,
1861 .offset = 0,
1862 },
1863 VMSTATE_END_OF_LIST()
1864 }
1865 };
1866
1867 static const VMStateDescription vmstate_virtio_device_endian = {
1868 .name = "virtio/device_endian",
1869 .version_id = 1,
1870 .minimum_version_id = 1,
1871 .needed = &virtio_device_endian_needed,
1872 .fields = (VMStateField[]) {
1873 VMSTATE_UINT8(device_endian, VirtIODevice),
1874 VMSTATE_END_OF_LIST()
1875 }
1876 };
1877
1878 static const VMStateDescription vmstate_virtio_64bit_features = {
1879 .name = "virtio/64bit_features",
1880 .version_id = 1,
1881 .minimum_version_id = 1,
1882 .needed = &virtio_64bit_features_needed,
1883 .fields = (VMStateField[]) {
1884 VMSTATE_UINT64(guest_features, VirtIODevice),
1885 VMSTATE_END_OF_LIST()
1886 }
1887 };
1888
1889 static const VMStateDescription vmstate_virtio_broken = {
1890 .name = "virtio/broken",
1891 .version_id = 1,
1892 .minimum_version_id = 1,
1893 .needed = &virtio_broken_needed,
1894 .fields = (VMStateField[]) {
1895 VMSTATE_BOOL(broken, VirtIODevice),
1896 VMSTATE_END_OF_LIST()
1897 }
1898 };
1899
1900 static const VMStateDescription vmstate_virtio = {
1901 .name = "virtio",
1902 .version_id = 1,
1903 .minimum_version_id = 1,
1904 .minimum_version_id_old = 1,
1905 .fields = (VMStateField[]) {
1906 VMSTATE_END_OF_LIST()
1907 },
1908 .subsections = (const VMStateDescription*[]) {
1909 &vmstate_virtio_device_endian,
1910 &vmstate_virtio_64bit_features,
1911 &vmstate_virtio_virtqueues,
1912 &vmstate_virtio_ringsize,
1913 &vmstate_virtio_broken,
1914 &vmstate_virtio_extra_state,
1915 NULL
1916 }
1917 };
1918
1919 int virtio_save(VirtIODevice *vdev, QEMUFile *f)
1920 {
1921 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
1922 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
1923 VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev);
1924 uint32_t guest_features_lo = (vdev->guest_features & 0xffffffff);
1925 int i;
1926
1927 if (k->save_config) {
1928 k->save_config(qbus->parent, f);
1929 }
1930
1931 qemu_put_8s(f, &vdev->status);
1932 qemu_put_8s(f, &vdev->isr);
1933 qemu_put_be16s(f, &vdev->queue_sel);
1934 qemu_put_be32s(f, &guest_features_lo);
1935 qemu_put_be32(f, vdev->config_len);
1936 qemu_put_buffer(f, vdev->config, vdev->config_len);
1937
1938 for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
1939 if (vdev->vq[i].vring.num == 0)
1940 break;
1941 }
1942
1943 qemu_put_be32(f, i);
1944
1945 for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
1946 if (vdev->vq[i].vring.num == 0)
1947 break;
1948
1949 qemu_put_be32(f, vdev->vq[i].vring.num);
1950 if (k->has_variable_vring_alignment) {
1951 qemu_put_be32(f, vdev->vq[i].vring.align);
1952 }
1953 /*
1954 * Save desc now, the rest of the ring addresses are saved in
1955 * subsections for VIRTIO-1 devices.
1956 */
1957 qemu_put_be64(f, vdev->vq[i].vring.desc);
1958 qemu_put_be16s(f, &vdev->vq[i].last_avail_idx);
1959 if (k->save_queue) {
1960 k->save_queue(qbus->parent, i, f);
1961 }
1962 }
1963
1964 if (vdc->save != NULL) {
1965 vdc->save(vdev, f);
1966 }
1967
1968 if (vdc->vmsd) {
1969 int ret = vmstate_save_state(f, vdc->vmsd, vdev, NULL);
1970 if (ret) {
1971 return ret;
1972 }
1973 }
1974
1975 /* Subsections */
1976 return vmstate_save_state(f, &vmstate_virtio, vdev, NULL);
1977 }
1978
1979 /* A wrapper for use as a VMState .put function */
1980 static int virtio_device_put(QEMUFile *f, void *opaque, size_t size,
1981 VMStateField *field, QJSON *vmdesc)
1982 {
1983 return virtio_save(VIRTIO_DEVICE(opaque), f);
1984 }
1985
1986 /* A wrapper for use as a VMState .get function */
1987 static int virtio_device_get(QEMUFile *f, void *opaque, size_t size,
1988 VMStateField *field)
1989 {
1990 VirtIODevice *vdev = VIRTIO_DEVICE(opaque);
1991 DeviceClass *dc = DEVICE_CLASS(VIRTIO_DEVICE_GET_CLASS(vdev));
1992
1993 return virtio_load(vdev, f, dc->vmsd->version_id);
1994 }
1995
1996 const VMStateInfo virtio_vmstate_info = {
1997 .name = "virtio",
1998 .get = virtio_device_get,
1999 .put = virtio_device_put,
2000 };
2001
2002 static int virtio_set_features_nocheck(VirtIODevice *vdev, uint64_t val)
2003 {
2004 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2005 bool bad = (val & ~(vdev->host_features)) != 0;
2006
2007 val &= vdev->host_features;
2008 if (k->set_features) {
2009 k->set_features(vdev, val);
2010 }
2011 vdev->guest_features = val;
2012 return bad ? -1 : 0;
2013 }
2014
2015 int virtio_set_features(VirtIODevice *vdev, uint64_t val)
2016 {
2017 int ret;
2018 /*
2019 * The driver must not attempt to set features after feature negotiation
2020 * has finished.
2021 */
2022 if (vdev->status & VIRTIO_CONFIG_S_FEATURES_OK) {
2023 return -EINVAL;
2024 }
2025 ret = virtio_set_features_nocheck(vdev, val);
2026 if (!ret && virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX)) {
2027 /* VIRTIO_RING_F_EVENT_IDX changes the size of the caches. */
2028 int i;
2029 for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2030 if (vdev->vq[i].vring.num != 0) {
2031 virtio_init_region_cache(vdev, i);
2032 }
2033 }
2034 }
2035 return ret;
2036 }
2037
2038 int virtio_load(VirtIODevice *vdev, QEMUFile *f, int version_id)
2039 {
2040 int i, ret;
2041 int32_t config_len;
2042 uint32_t num;
2043 uint32_t features;
2044 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2045 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
2046 VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev);
2047
2048 /*
2049 * We poison the endianness to ensure it does not get used before
2050 * subsections have been loaded.
2051 */
2052 vdev->device_endian = VIRTIO_DEVICE_ENDIAN_UNKNOWN;
2053
2054 if (k->load_config) {
2055 ret = k->load_config(qbus->parent, f);
2056 if (ret)
2057 return ret;
2058 }
2059
2060 qemu_get_8s(f, &vdev->status);
2061 qemu_get_8s(f, &vdev->isr);
2062 qemu_get_be16s(f, &vdev->queue_sel);
2063 if (vdev->queue_sel >= VIRTIO_QUEUE_MAX) {
2064 return -1;
2065 }
2066 qemu_get_be32s(f, &features);
2067
2068 /*
2069 * Temporarily set guest_features low bits - needed by
2070 * virtio net load code testing for VIRTIO_NET_F_CTRL_GUEST_OFFLOADS
2071 * VIRTIO_NET_F_GUEST_ANNOUNCE and VIRTIO_NET_F_CTRL_VQ.
2072 *
2073 * Note: devices should always test host features in future - don't create
2074 * new dependencies like this.
2075 */
2076 vdev->guest_features = features;
2077
2078 config_len = qemu_get_be32(f);
2079
2080 /*
2081 * There are cases where the incoming config can be bigger or smaller
2082 * than what we have; so load what we have space for, and skip
2083 * any excess that's in the stream.
2084 */
2085 qemu_get_buffer(f, vdev->config, MIN(config_len, vdev->config_len));
2086
2087 while (config_len > vdev->config_len) {
2088 qemu_get_byte(f);
2089 config_len--;
2090 }
2091
2092 num = qemu_get_be32(f);
2093
2094 if (num > VIRTIO_QUEUE_MAX) {
2095 error_report("Invalid number of virtqueues: 0x%x", num);
2096 return -1;
2097 }
2098
2099 for (i = 0; i < num; i++) {
2100 vdev->vq[i].vring.num = qemu_get_be32(f);
2101 if (k->has_variable_vring_alignment) {
2102 vdev->vq[i].vring.align = qemu_get_be32(f);
2103 }
2104 vdev->vq[i].vring.desc = qemu_get_be64(f);
2105 qemu_get_be16s(f, &vdev->vq[i].last_avail_idx);
2106 vdev->vq[i].signalled_used_valid = false;
2107 vdev->vq[i].notification = true;
2108
2109 if (!vdev->vq[i].vring.desc && vdev->vq[i].last_avail_idx) {
2110 error_report("VQ %d address 0x0 "
2111 "inconsistent with Host index 0x%x",
2112 i, vdev->vq[i].last_avail_idx);
2113 return -1;
2114 }
2115 if (k->load_queue) {
2116 ret = k->load_queue(qbus->parent, i, f);
2117 if (ret)
2118 return ret;
2119 }
2120 }
2121
2122 virtio_notify_vector(vdev, VIRTIO_NO_VECTOR);
2123
2124 if (vdc->load != NULL) {
2125 ret = vdc->load(vdev, f, version_id);
2126 if (ret) {
2127 return ret;
2128 }
2129 }
2130
2131 if (vdc->vmsd) {
2132 ret = vmstate_load_state(f, vdc->vmsd, vdev, version_id);
2133 if (ret) {
2134 return ret;
2135 }
2136 }
2137
2138 /* Subsections */
2139 ret = vmstate_load_state(f, &vmstate_virtio, vdev, 1);
2140 if (ret) {
2141 return ret;
2142 }
2143
2144 if (vdev->device_endian == VIRTIO_DEVICE_ENDIAN_UNKNOWN) {
2145 vdev->device_endian = virtio_default_endian();
2146 }
2147
2148 if (virtio_64bit_features_needed(vdev)) {
2149 /*
2150 * Subsection load filled vdev->guest_features. Run them
2151 * through virtio_set_features to sanity-check them against
2152 * host_features.
2153 */
2154 uint64_t features64 = vdev->guest_features;
2155 if (virtio_set_features_nocheck(vdev, features64) < 0) {
2156 error_report("Features 0x%" PRIx64 " unsupported. "
2157 "Allowed features: 0x%" PRIx64,
2158 features64, vdev->host_features);
2159 return -1;
2160 }
2161 } else {
2162 if (virtio_set_features_nocheck(vdev, features) < 0) {
2163 error_report("Features 0x%x unsupported. "
2164 "Allowed features: 0x%" PRIx64,
2165 features, vdev->host_features);
2166 return -1;
2167 }
2168 }
2169
2170 rcu_read_lock();
2171 for (i = 0; i < num; i++) {
2172 if (vdev->vq[i].vring.desc) {
2173 uint16_t nheads;
2174
2175 /*
2176 * VIRTIO-1 devices migrate desc, used, and avail ring addresses so
2177 * only the region cache needs to be set up. Legacy devices need
2178 * to calculate used and avail ring addresses based on the desc
2179 * address.
2180 */
2181 if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
2182 virtio_init_region_cache(vdev, i);
2183 } else {
2184 virtio_queue_update_rings(vdev, i);
2185 }
2186
2187 nheads = vring_avail_idx(&vdev->vq[i]) - vdev->vq[i].last_avail_idx;
2188 /* Check it isn't doing strange things with descriptor numbers. */
2189 if (nheads > vdev->vq[i].vring.num) {
2190 error_report("VQ %d size 0x%x Guest index 0x%x "
2191 "inconsistent with Host index 0x%x: delta 0x%x",
2192 i, vdev->vq[i].vring.num,
2193 vring_avail_idx(&vdev->vq[i]),
2194 vdev->vq[i].last_avail_idx, nheads);
2195 return -1;
2196 }
2197 vdev->vq[i].used_idx = vring_used_idx(&vdev->vq[i]);
2198 vdev->vq[i].shadow_avail_idx = vring_avail_idx(&vdev->vq[i]);
2199
2200 /*
2201 * Some devices migrate VirtQueueElements that have been popped
2202 * from the avail ring but not yet returned to the used ring.
2203 * Since max ring size < UINT16_MAX it's safe to use modulo
2204 * UINT16_MAX + 1 subtraction.
2205 */
2206 vdev->vq[i].inuse = (uint16_t)(vdev->vq[i].last_avail_idx -
2207 vdev->vq[i].used_idx);
2208 if (vdev->vq[i].inuse > vdev->vq[i].vring.num) {
2209 error_report("VQ %d size 0x%x < last_avail_idx 0x%x - "
2210 "used_idx 0x%x",
2211 i, vdev->vq[i].vring.num,
2212 vdev->vq[i].last_avail_idx,
2213 vdev->vq[i].used_idx);
2214 return -1;
2215 }
2216 }
2217 }
2218 rcu_read_unlock();
2219
2220 return 0;
2221 }
2222
2223 void virtio_cleanup(VirtIODevice *vdev)
2224 {
2225 qemu_del_vm_change_state_handler(vdev->vmstate);
2226 }
2227
2228 static void virtio_vmstate_change(void *opaque, int running, RunState state)
2229 {
2230 VirtIODevice *vdev = opaque;
2231 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2232 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
2233 bool backend_run = running && (vdev->status & VIRTIO_CONFIG_S_DRIVER_OK);
2234 vdev->vm_running = running;
2235
2236 if (backend_run) {
2237 virtio_set_status(vdev, vdev->status);
2238 }
2239
2240 if (k->vmstate_change) {
2241 k->vmstate_change(qbus->parent, backend_run);
2242 }
2243
2244 if (!backend_run) {
2245 virtio_set_status(vdev, vdev->status);
2246 }
2247 }
2248
2249 void virtio_instance_init_common(Object *proxy_obj, void *data,
2250 size_t vdev_size, const char *vdev_name)
2251 {
2252 DeviceState *vdev = data;
2253
2254 object_initialize(vdev, vdev_size, vdev_name);
2255 object_property_add_child(proxy_obj, "virtio-backend", OBJECT(vdev), NULL);
2256 object_unref(OBJECT(vdev));
2257 qdev_alias_all_properties(vdev, proxy_obj);
2258 }
2259
2260 void virtio_init(VirtIODevice *vdev, const char *name,
2261 uint16_t device_id, size_t config_size)
2262 {
2263 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2264 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
2265 int i;
2266 int nvectors = k->query_nvectors ? k->query_nvectors(qbus->parent) : 0;
2267
2268 if (nvectors) {
2269 vdev->vector_queues =
2270 g_malloc0(sizeof(*vdev->vector_queues) * nvectors);
2271 }
2272
2273 vdev->device_id = device_id;
2274 vdev->status = 0;
2275 atomic_set(&vdev->isr, 0);
2276 vdev->queue_sel = 0;
2277 vdev->config_vector = VIRTIO_NO_VECTOR;
2278 vdev->vq = g_malloc0(sizeof(VirtQueue) * VIRTIO_QUEUE_MAX);
2279 vdev->vm_running = runstate_is_running();
2280 vdev->broken = false;
2281 for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2282 vdev->vq[i].vector = VIRTIO_NO_VECTOR;
2283 vdev->vq[i].vdev = vdev;
2284 vdev->vq[i].queue_index = i;
2285 }
2286
2287 vdev->name = name;
2288 vdev->config_len = config_size;
2289 if (vdev->config_len) {
2290 vdev->config = g_malloc0(config_size);
2291 } else {
2292 vdev->config = NULL;
2293 }
2294 vdev->vmstate = qemu_add_vm_change_state_handler(virtio_vmstate_change,
2295 vdev);
2296 vdev->device_endian = virtio_default_endian();
2297 vdev->use_guest_notifier_mask = true;
2298 }
2299
2300 hwaddr virtio_queue_get_desc_addr(VirtIODevice *vdev, int n)
2301 {
2302 return vdev->vq[n].vring.desc;
2303 }
2304
2305 hwaddr virtio_queue_get_avail_addr(VirtIODevice *vdev, int n)
2306 {
2307 return vdev->vq[n].vring.avail;
2308 }
2309
2310 hwaddr virtio_queue_get_used_addr(VirtIODevice *vdev, int n)
2311 {
2312 return vdev->vq[n].vring.used;
2313 }
2314
2315 hwaddr virtio_queue_get_desc_size(VirtIODevice *vdev, int n)
2316 {
2317 return sizeof(VRingDesc) * vdev->vq[n].vring.num;
2318 }
2319
2320 hwaddr virtio_queue_get_avail_size(VirtIODevice *vdev, int n)
2321 {
2322 return offsetof(VRingAvail, ring) +
2323 sizeof(uint16_t) * vdev->vq[n].vring.num;
2324 }
2325
2326 hwaddr virtio_queue_get_used_size(VirtIODevice *vdev, int n)
2327 {
2328 return offsetof(VRingUsed, ring) +
2329 sizeof(VRingUsedElem) * vdev->vq[n].vring.num;
2330 }
2331
2332 uint16_t virtio_queue_get_last_avail_idx(VirtIODevice *vdev, int n)
2333 {
2334 return vdev->vq[n].last_avail_idx;
2335 }
2336
2337 void virtio_queue_set_last_avail_idx(VirtIODevice *vdev, int n, uint16_t idx)
2338 {
2339 vdev->vq[n].last_avail_idx = idx;
2340 vdev->vq[n].shadow_avail_idx = idx;
2341 }
2342
2343 void virtio_queue_restore_last_avail_idx(VirtIODevice *vdev, int n)
2344 {
2345 rcu_read_lock();
2346 if (vdev->vq[n].vring.desc) {
2347 vdev->vq[n].last_avail_idx = vring_used_idx(&vdev->vq[n]);
2348 vdev->vq[n].shadow_avail_idx = vdev->vq[n].last_avail_idx;
2349 }
2350 rcu_read_unlock();
2351 }
2352
2353 void virtio_queue_update_used_idx(VirtIODevice *vdev, int n)
2354 {
2355 rcu_read_lock();
2356 if (vdev->vq[n].vring.desc) {
2357 vdev->vq[n].used_idx = vring_used_idx(&vdev->vq[n]);
2358 }
2359 rcu_read_unlock();
2360 }
2361
2362 void virtio_queue_invalidate_signalled_used(VirtIODevice *vdev, int n)
2363 {
2364 vdev->vq[n].signalled_used_valid = false;
2365 }
2366
2367 VirtQueue *virtio_get_queue(VirtIODevice *vdev, int n)
2368 {
2369 return vdev->vq + n;
2370 }
2371
2372 uint16_t virtio_get_queue_index(VirtQueue *vq)
2373 {
2374 return vq->queue_index;
2375 }
2376
2377 static void virtio_queue_guest_notifier_read(EventNotifier *n)
2378 {
2379 VirtQueue *vq = container_of(n, VirtQueue, guest_notifier);
2380 if (event_notifier_test_and_clear(n)) {
2381 virtio_irq(vq);
2382 }
2383 }
2384
2385 void virtio_queue_set_guest_notifier_fd_handler(VirtQueue *vq, bool assign,
2386 bool with_irqfd)
2387 {
2388 if (assign && !with_irqfd) {
2389 event_notifier_set_handler(&vq->guest_notifier,
2390 virtio_queue_guest_notifier_read);
2391 } else {
2392 event_notifier_set_handler(&vq->guest_notifier, NULL);
2393 }
2394 if (!assign) {
2395 /* Test and clear notifier before closing it,
2396 * in case poll callback didn't have time to run. */
2397 virtio_queue_guest_notifier_read(&vq->guest_notifier);
2398 }
2399 }
2400
2401 EventNotifier *virtio_queue_get_guest_notifier(VirtQueue *vq)
2402 {
2403 return &vq->guest_notifier;
2404 }
2405
2406 static void virtio_queue_host_notifier_aio_read(EventNotifier *n)
2407 {
2408 VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
2409 if (event_notifier_test_and_clear(n)) {
2410 virtio_queue_notify_aio_vq(vq);
2411 }
2412 }
2413
2414 static void virtio_queue_host_notifier_aio_poll_begin(EventNotifier *n)
2415 {
2416 VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
2417
2418 virtio_queue_set_notification(vq, 0);
2419 }
2420
2421 static bool virtio_queue_host_notifier_aio_poll(void *opaque)
2422 {
2423 EventNotifier *n = opaque;
2424 VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
2425 bool progress;
2426
2427 if (!vq->vring.desc || virtio_queue_empty(vq)) {
2428 return false;
2429 }
2430
2431 progress = virtio_queue_notify_aio_vq(vq);
2432
2433 /* In case the handler function re-enabled notifications */
2434 virtio_queue_set_notification(vq, 0);
2435 return progress;
2436 }
2437
2438 static void virtio_queue_host_notifier_aio_poll_end(EventNotifier *n)
2439 {
2440 VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
2441
2442 /* Caller polls once more after this to catch requests that race with us */
2443 virtio_queue_set_notification(vq, 1);
2444 }
2445
2446 void virtio_queue_aio_set_host_notifier_handler(VirtQueue *vq, AioContext *ctx,
2447 VirtIOHandleAIOOutput handle_output)
2448 {
2449 if (handle_output) {
2450 vq->handle_aio_output = handle_output;
2451 aio_set_event_notifier(ctx, &vq->host_notifier, true,
2452 virtio_queue_host_notifier_aio_read,
2453 virtio_queue_host_notifier_aio_poll);
2454 aio_set_event_notifier_poll(ctx, &vq->host_notifier,
2455 virtio_queue_host_notifier_aio_poll_begin,
2456 virtio_queue_host_notifier_aio_poll_end);
2457 } else {
2458 aio_set_event_notifier(ctx, &vq->host_notifier, true, NULL, NULL);
2459 /* Test and clear notifier before after disabling event,
2460 * in case poll callback didn't have time to run. */
2461 virtio_queue_host_notifier_aio_read(&vq->host_notifier);
2462 vq->handle_aio_output = NULL;
2463 }
2464 }
2465
2466 void virtio_queue_host_notifier_read(EventNotifier *n)
2467 {
2468 VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
2469 if (event_notifier_test_and_clear(n)) {
2470 virtio_queue_notify_vq(vq);
2471 }
2472 }
2473
2474 EventNotifier *virtio_queue_get_host_notifier(VirtQueue *vq)
2475 {
2476 return &vq->host_notifier;
2477 }
2478
2479 int virtio_queue_set_host_notifier_mr(VirtIODevice *vdev, int n,
2480 MemoryRegion *mr, bool assign)
2481 {
2482 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2483 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
2484
2485 if (k->set_host_notifier_mr) {
2486 return k->set_host_notifier_mr(qbus->parent, n, mr, assign);
2487 }
2488
2489 return -1;
2490 }
2491
2492 void virtio_device_set_child_bus_name(VirtIODevice *vdev, char *bus_name)
2493 {
2494 g_free(vdev->bus_name);
2495 vdev->bus_name = g_strdup(bus_name);
2496 }
2497
2498 void GCC_FMT_ATTR(2, 3) virtio_error(VirtIODevice *vdev, const char *fmt, ...)
2499 {
2500 va_list ap;
2501
2502 va_start(ap, fmt);
2503 error_vreport(fmt, ap);
2504 va_end(ap);
2505
2506 if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
2507 vdev->status = vdev->status | VIRTIO_CONFIG_S_NEEDS_RESET;
2508 virtio_notify_config(vdev);
2509 }
2510
2511 vdev->broken = true;
2512 }
2513
2514 static void virtio_memory_listener_commit(MemoryListener *listener)
2515 {
2516 VirtIODevice *vdev = container_of(listener, VirtIODevice, listener);
2517 int i;
2518
2519 for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2520 if (vdev->vq[i].vring.num == 0) {
2521 break;
2522 }
2523 virtio_init_region_cache(vdev, i);
2524 }
2525 }
2526
2527 static void virtio_device_realize(DeviceState *dev, Error **errp)
2528 {
2529 VirtIODevice *vdev = VIRTIO_DEVICE(dev);
2530 VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(dev);
2531 Error *err = NULL;
2532
2533 /* Devices should either use vmsd or the load/save methods */
2534 assert(!vdc->vmsd || !vdc->load);
2535
2536 if (vdc->realize != NULL) {
2537 vdc->realize(dev, &err);
2538 if (err != NULL) {
2539 error_propagate(errp, err);
2540 return;
2541 }
2542 }
2543
2544 virtio_bus_device_plugged(vdev, &err);
2545 if (err != NULL) {
2546 error_propagate(errp, err);
2547 vdc->unrealize(dev, NULL);
2548 return;
2549 }
2550
2551 vdev->listener.commit = virtio_memory_listener_commit;
2552 memory_listener_register(&vdev->listener, vdev->dma_as);
2553 }
2554
2555 static void virtio_device_unrealize(DeviceState *dev, Error **errp)
2556 {
2557 VirtIODevice *vdev = VIRTIO_DEVICE(dev);
2558 VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(dev);
2559 Error *err = NULL;
2560
2561 virtio_bus_device_unplugged(vdev);
2562
2563 if (vdc->unrealize != NULL) {
2564 vdc->unrealize(dev, &err);
2565 if (err != NULL) {
2566 error_propagate(errp, err);
2567 return;
2568 }
2569 }
2570
2571 g_free(vdev->bus_name);
2572 vdev->bus_name = NULL;
2573 }
2574
2575 static void virtio_device_free_virtqueues(VirtIODevice *vdev)
2576 {
2577 int i;
2578 if (!vdev->vq) {
2579 return;
2580 }
2581
2582 for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2583 if (vdev->vq[i].vring.num == 0) {
2584 break;
2585 }
2586 virtio_virtqueue_reset_region_cache(&vdev->vq[i]);
2587 }
2588 g_free(vdev->vq);
2589 }
2590
2591 static void virtio_device_instance_finalize(Object *obj)
2592 {
2593 VirtIODevice *vdev = VIRTIO_DEVICE(obj);
2594
2595 memory_listener_unregister(&vdev->listener);
2596 virtio_device_free_virtqueues(vdev);
2597
2598 g_free(vdev->config);
2599 g_free(vdev->vector_queues);
2600 }
2601
2602 static Property virtio_properties[] = {
2603 DEFINE_VIRTIO_COMMON_FEATURES(VirtIODevice, host_features),
2604 DEFINE_PROP_END_OF_LIST(),
2605 };
2606
2607 static int virtio_device_start_ioeventfd_impl(VirtIODevice *vdev)
2608 {
2609 VirtioBusState *qbus = VIRTIO_BUS(qdev_get_parent_bus(DEVICE(vdev)));
2610 int i, n, r, err;
2611
2612 memory_region_transaction_begin();
2613 for (n = 0; n < VIRTIO_QUEUE_MAX; n++) {
2614 VirtQueue *vq = &vdev->vq[n];
2615 if (!virtio_queue_get_num(vdev, n)) {
2616 continue;
2617 }
2618 r = virtio_bus_set_host_notifier(qbus, n, true);
2619 if (r < 0) {
2620 err = r;
2621 goto assign_error;
2622 }
2623 event_notifier_set_handler(&vq->host_notifier,
2624 virtio_queue_host_notifier_read);
2625 }
2626
2627 for (n = 0; n < VIRTIO_QUEUE_MAX; n++) {
2628 /* Kick right away to begin processing requests already in vring */
2629 VirtQueue *vq = &vdev->vq[n];
2630 if (!vq->vring.num) {
2631 continue;
2632 }
2633 event_notifier_set(&vq->host_notifier);
2634 }
2635 memory_region_transaction_commit();
2636 return 0;
2637
2638 assign_error:
2639 i = n; /* save n for a second iteration after transaction is committed. */
2640 while (--n >= 0) {
2641 VirtQueue *vq = &vdev->vq[n];
2642 if (!virtio_queue_get_num(vdev, n)) {
2643 continue;
2644 }
2645
2646 event_notifier_set_handler(&vq->host_notifier, NULL);
2647 r = virtio_bus_set_host_notifier(qbus, n, false);
2648 assert(r >= 0);
2649 }
2650 memory_region_transaction_commit();
2651
2652 while (--i >= 0) {
2653 if (!virtio_queue_get_num(vdev, i)) {
2654 continue;
2655 }
2656 virtio_bus_cleanup_host_notifier(qbus, i);
2657 }
2658 return err;
2659 }
2660
2661 int virtio_device_start_ioeventfd(VirtIODevice *vdev)
2662 {
2663 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2664 VirtioBusState *vbus = VIRTIO_BUS(qbus);
2665
2666 return virtio_bus_start_ioeventfd(vbus);
2667 }
2668
2669 static void virtio_device_stop_ioeventfd_impl(VirtIODevice *vdev)
2670 {
2671 VirtioBusState *qbus = VIRTIO_BUS(qdev_get_parent_bus(DEVICE(vdev)));
2672 int n, r;
2673
2674 memory_region_transaction_begin();
2675 for (n = 0; n < VIRTIO_QUEUE_MAX; n++) {
2676 VirtQueue *vq = &vdev->vq[n];
2677
2678 if (!virtio_queue_get_num(vdev, n)) {
2679 continue;
2680 }
2681 event_notifier_set_handler(&vq->host_notifier, NULL);
2682 r = virtio_bus_set_host_notifier(qbus, n, false);
2683 assert(r >= 0);
2684 }
2685 memory_region_transaction_commit();
2686
2687 for (n = 0; n < VIRTIO_QUEUE_MAX; n++) {
2688 if (!virtio_queue_get_num(vdev, n)) {
2689 continue;
2690 }
2691 virtio_bus_cleanup_host_notifier(qbus, n);
2692 }
2693 }
2694
2695 void virtio_device_stop_ioeventfd(VirtIODevice *vdev)
2696 {
2697 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2698 VirtioBusState *vbus = VIRTIO_BUS(qbus);
2699
2700 virtio_bus_stop_ioeventfd(vbus);
2701 }
2702
2703 int virtio_device_grab_ioeventfd(VirtIODevice *vdev)
2704 {
2705 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2706 VirtioBusState *vbus = VIRTIO_BUS(qbus);
2707
2708 return virtio_bus_grab_ioeventfd(vbus);
2709 }
2710
2711 void virtio_device_release_ioeventfd(VirtIODevice *vdev)
2712 {
2713 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2714 VirtioBusState *vbus = VIRTIO_BUS(qbus);
2715
2716 virtio_bus_release_ioeventfd(vbus);
2717 }
2718
2719 static void virtio_device_class_init(ObjectClass *klass, void *data)
2720 {
2721 /* Set the default value here. */
2722 VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
2723 DeviceClass *dc = DEVICE_CLASS(klass);
2724
2725 dc->realize = virtio_device_realize;
2726 dc->unrealize = virtio_device_unrealize;
2727 dc->bus_type = TYPE_VIRTIO_BUS;
2728 dc->props = virtio_properties;
2729 vdc->start_ioeventfd = virtio_device_start_ioeventfd_impl;
2730 vdc->stop_ioeventfd = virtio_device_stop_ioeventfd_impl;
2731
2732 vdc->legacy_features |= VIRTIO_LEGACY_FEATURES;
2733 }
2734
2735 bool virtio_device_ioeventfd_enabled(VirtIODevice *vdev)
2736 {
2737 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2738 VirtioBusState *vbus = VIRTIO_BUS(qbus);
2739
2740 return virtio_bus_ioeventfd_enabled(vbus);
2741 }
2742
2743 static const TypeInfo virtio_device_info = {
2744 .name = TYPE_VIRTIO_DEVICE,
2745 .parent = TYPE_DEVICE,
2746 .instance_size = sizeof(VirtIODevice),
2747 .class_init = virtio_device_class_init,
2748 .instance_finalize = virtio_device_instance_finalize,
2749 .abstract = true,
2750 .class_size = sizeof(VirtioDeviceClass),
2751 };
2752
2753 static void virtio_register_types(void)
2754 {
2755 type_register_static(&virtio_device_info);
2756 }
2757
2758 type_init(virtio_register_types)
armbru's QEMU hackery