xref: /openbmc/qemu/hw/virtio/virtio.c (revision 5242ef88)
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 "cpu.h"
17 #include "trace.h"
18 #include "qemu/error-report.h"
19 #include "qemu/log.h"
20 #include "qemu/main-loop.h"
21 #include "qemu/module.h"
22 #include "hw/virtio/virtio.h"
23 #include "migration/qemu-file-types.h"
24 #include "qemu/atomic.h"
25 #include "hw/virtio/virtio-bus.h"
26 #include "hw/qdev-properties.h"
27 #include "hw/virtio/virtio-access.h"
28 #include "sysemu/dma.h"
29 #include "sysemu/runstate.h"
30 #include "standard-headers/linux/virtio_ids.h"
31 
32 /*
33  * The alignment to use between consumer and producer parts of vring.
34  * x86 pagesize again. This is the default, used by transports like PCI
35  * which don't provide a means for the guest to tell the host the alignment.
36  */
37 #define VIRTIO_PCI_VRING_ALIGN         4096
38 
39 typedef struct VRingDesc
40 {
41     uint64_t addr;
42     uint32_t len;
43     uint16_t flags;
44     uint16_t next;
45 } VRingDesc;
46 
47 typedef struct VRingPackedDesc {
48     uint64_t addr;
49     uint32_t len;
50     uint16_t id;
51     uint16_t flags;
52 } VRingPackedDesc;
53 
54 typedef struct VRingAvail
55 {
56     uint16_t flags;
57     uint16_t idx;
58     uint16_t ring[];
59 } VRingAvail;
60 
61 typedef struct VRingUsedElem
62 {
63     uint32_t id;
64     uint32_t len;
65 } VRingUsedElem;
66 
67 typedef struct VRingUsed
68 {
69     uint16_t flags;
70     uint16_t idx;
71     VRingUsedElem ring[];
72 } VRingUsed;
73 
74 typedef struct VRingMemoryRegionCaches {
75     struct rcu_head rcu;
76     MemoryRegionCache desc;
77     MemoryRegionCache avail;
78     MemoryRegionCache used;
79 } VRingMemoryRegionCaches;
80 
81 typedef struct VRing
82 {
83     unsigned int num;
84     unsigned int num_default;
85     unsigned int align;
86     hwaddr desc;
87     hwaddr avail;
88     hwaddr used;
89     VRingMemoryRegionCaches *caches;
90 } VRing;
91 
92 typedef struct VRingPackedDescEvent {
93     uint16_t off_wrap;
94     uint16_t flags;
95 } VRingPackedDescEvent ;
96 
97 struct VirtQueue
98 {
99     VRing vring;
100     VirtQueueElement *used_elems;
101 
102     /* Next head to pop */
103     uint16_t last_avail_idx;
104     bool last_avail_wrap_counter;
105 
106     /* Last avail_idx read from VQ. */
107     uint16_t shadow_avail_idx;
108     bool shadow_avail_wrap_counter;
109 
110     uint16_t used_idx;
111     bool used_wrap_counter;
112 
113     /* Last used index value we have signalled on */
114     uint16_t signalled_used;
115 
116     /* Last used index value we have signalled on */
117     bool signalled_used_valid;
118 
119     /* Notification enabled? */
120     bool notification;
121 
122     uint16_t queue_index;
123 
124     unsigned int inuse;
125 
126     uint16_t vector;
127     VirtIOHandleOutput handle_output;
128     VirtIODevice *vdev;
129     EventNotifier guest_notifier;
130     EventNotifier host_notifier;
131     bool host_notifier_enabled;
132     QLIST_ENTRY(VirtQueue) node;
133 };
134 
135 /* Called within call_rcu().  */
136 static void virtio_free_region_cache(VRingMemoryRegionCaches *caches)
137 {
138     assert(caches != NULL);
139     address_space_cache_destroy(&caches->desc);
140     address_space_cache_destroy(&caches->avail);
141     address_space_cache_destroy(&caches->used);
142     g_free(caches);
143 }
144 
145 static void virtio_virtqueue_reset_region_cache(struct VirtQueue *vq)
146 {
147     VRingMemoryRegionCaches *caches;
148 
149     caches = qatomic_read(&vq->vring.caches);
150     qatomic_rcu_set(&vq->vring.caches, NULL);
151     if (caches) {
152         call_rcu(caches, virtio_free_region_cache, rcu);
153     }
154 }
155 
156 static void virtio_init_region_cache(VirtIODevice *vdev, int n)
157 {
158     VirtQueue *vq = &vdev->vq[n];
159     VRingMemoryRegionCaches *old = vq->vring.caches;
160     VRingMemoryRegionCaches *new = NULL;
161     hwaddr addr, size;
162     int64_t len;
163     bool packed;
164 
165 
166     addr = vq->vring.desc;
167     if (!addr) {
168         goto out_no_cache;
169     }
170     new = g_new0(VRingMemoryRegionCaches, 1);
171     size = virtio_queue_get_desc_size(vdev, n);
172     packed = virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED) ?
173                                    true : false;
174     len = address_space_cache_init(&new->desc, vdev->dma_as,
175                                    addr, size, packed);
176     if (len < size) {
177         virtio_error(vdev, "Cannot map desc");
178         goto err_desc;
179     }
180 
181     size = virtio_queue_get_used_size(vdev, n);
182     len = address_space_cache_init(&new->used, vdev->dma_as,
183                                    vq->vring.used, size, true);
184     if (len < size) {
185         virtio_error(vdev, "Cannot map used");
186         goto err_used;
187     }
188 
189     size = virtio_queue_get_avail_size(vdev, n);
190     len = address_space_cache_init(&new->avail, vdev->dma_as,
191                                    vq->vring.avail, size, false);
192     if (len < size) {
193         virtio_error(vdev, "Cannot map avail");
194         goto err_avail;
195     }
196 
197     qatomic_rcu_set(&vq->vring.caches, new);
198     if (old) {
199         call_rcu(old, virtio_free_region_cache, rcu);
200     }
201     return;
202 
203 err_avail:
204     address_space_cache_destroy(&new->avail);
205 err_used:
206     address_space_cache_destroy(&new->used);
207 err_desc:
208     address_space_cache_destroy(&new->desc);
209 out_no_cache:
210     g_free(new);
211     virtio_virtqueue_reset_region_cache(vq);
212 }
213 
214 /* virt queue functions */
215 void virtio_queue_update_rings(VirtIODevice *vdev, int n)
216 {
217     VRing *vring = &vdev->vq[n].vring;
218 
219     if (!vring->num || !vring->desc || !vring->align) {
220         /* not yet setup -> nothing to do */
221         return;
222     }
223     vring->avail = vring->desc + vring->num * sizeof(VRingDesc);
224     vring->used = vring_align(vring->avail +
225                               offsetof(VRingAvail, ring[vring->num]),
226                               vring->align);
227     virtio_init_region_cache(vdev, n);
228 }
229 
230 /* Called within rcu_read_lock().  */
231 static void vring_split_desc_read(VirtIODevice *vdev, VRingDesc *desc,
232                                   MemoryRegionCache *cache, int i)
233 {
234     address_space_read_cached(cache, i * sizeof(VRingDesc),
235                               desc, sizeof(VRingDesc));
236     virtio_tswap64s(vdev, &desc->addr);
237     virtio_tswap32s(vdev, &desc->len);
238     virtio_tswap16s(vdev, &desc->flags);
239     virtio_tswap16s(vdev, &desc->next);
240 }
241 
242 static void vring_packed_event_read(VirtIODevice *vdev,
243                                     MemoryRegionCache *cache,
244                                     VRingPackedDescEvent *e)
245 {
246     hwaddr off_off = offsetof(VRingPackedDescEvent, off_wrap);
247     hwaddr off_flags = offsetof(VRingPackedDescEvent, flags);
248 
249     e->flags = virtio_lduw_phys_cached(vdev, cache, off_flags);
250     /* Make sure flags is seen before off_wrap */
251     smp_rmb();
252     e->off_wrap = virtio_lduw_phys_cached(vdev, cache, off_off);
253     virtio_tswap16s(vdev, &e->flags);
254 }
255 
256 static void vring_packed_off_wrap_write(VirtIODevice *vdev,
257                                         MemoryRegionCache *cache,
258                                         uint16_t off_wrap)
259 {
260     hwaddr off = offsetof(VRingPackedDescEvent, off_wrap);
261 
262     virtio_stw_phys_cached(vdev, cache, off, off_wrap);
263     address_space_cache_invalidate(cache, off, sizeof(off_wrap));
264 }
265 
266 static void vring_packed_flags_write(VirtIODevice *vdev,
267                                      MemoryRegionCache *cache, uint16_t flags)
268 {
269     hwaddr off = offsetof(VRingPackedDescEvent, flags);
270 
271     virtio_stw_phys_cached(vdev, cache, off, flags);
272     address_space_cache_invalidate(cache, off, sizeof(flags));
273 }
274 
275 /* Called within rcu_read_lock().  */
276 static VRingMemoryRegionCaches *vring_get_region_caches(struct VirtQueue *vq)
277 {
278     return qatomic_rcu_read(&vq->vring.caches);
279 }
280 
281 /* Called within rcu_read_lock().  */
282 static inline uint16_t vring_avail_flags(VirtQueue *vq)
283 {
284     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
285     hwaddr pa = offsetof(VRingAvail, flags);
286 
287     if (!caches) {
288         return 0;
289     }
290 
291     return virtio_lduw_phys_cached(vq->vdev, &caches->avail, pa);
292 }
293 
294 /* Called within rcu_read_lock().  */
295 static inline uint16_t vring_avail_idx(VirtQueue *vq)
296 {
297     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
298     hwaddr pa = offsetof(VRingAvail, idx);
299 
300     if (!caches) {
301         return 0;
302     }
303 
304     vq->shadow_avail_idx = virtio_lduw_phys_cached(vq->vdev, &caches->avail, pa);
305     return vq->shadow_avail_idx;
306 }
307 
308 /* Called within rcu_read_lock().  */
309 static inline uint16_t vring_avail_ring(VirtQueue *vq, int i)
310 {
311     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
312     hwaddr pa = offsetof(VRingAvail, ring[i]);
313 
314     if (!caches) {
315         return 0;
316     }
317 
318     return virtio_lduw_phys_cached(vq->vdev, &caches->avail, pa);
319 }
320 
321 /* Called within rcu_read_lock().  */
322 static inline uint16_t vring_get_used_event(VirtQueue *vq)
323 {
324     return vring_avail_ring(vq, vq->vring.num);
325 }
326 
327 /* Called within rcu_read_lock().  */
328 static inline void vring_used_write(VirtQueue *vq, VRingUsedElem *uelem,
329                                     int i)
330 {
331     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
332     hwaddr pa = offsetof(VRingUsed, ring[i]);
333 
334     if (!caches) {
335         return;
336     }
337 
338     virtio_tswap32s(vq->vdev, &uelem->id);
339     virtio_tswap32s(vq->vdev, &uelem->len);
340     address_space_write_cached(&caches->used, pa, uelem, sizeof(VRingUsedElem));
341     address_space_cache_invalidate(&caches->used, pa, sizeof(VRingUsedElem));
342 }
343 
344 /* Called within rcu_read_lock().  */
345 static uint16_t vring_used_idx(VirtQueue *vq)
346 {
347     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
348     hwaddr pa = offsetof(VRingUsed, idx);
349 
350     if (!caches) {
351         return 0;
352     }
353 
354     return virtio_lduw_phys_cached(vq->vdev, &caches->used, pa);
355 }
356 
357 /* Called within rcu_read_lock().  */
358 static inline void vring_used_idx_set(VirtQueue *vq, uint16_t val)
359 {
360     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
361     hwaddr pa = offsetof(VRingUsed, idx);
362 
363     if (caches) {
364         virtio_stw_phys_cached(vq->vdev, &caches->used, pa, val);
365         address_space_cache_invalidate(&caches->used, pa, sizeof(val));
366     }
367 
368     vq->used_idx = val;
369 }
370 
371 /* Called within rcu_read_lock().  */
372 static inline void vring_used_flags_set_bit(VirtQueue *vq, int mask)
373 {
374     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
375     VirtIODevice *vdev = vq->vdev;
376     hwaddr pa = offsetof(VRingUsed, flags);
377     uint16_t flags;
378 
379     if (!caches) {
380         return;
381     }
382 
383     flags = virtio_lduw_phys_cached(vq->vdev, &caches->used, pa);
384     virtio_stw_phys_cached(vdev, &caches->used, pa, flags | mask);
385     address_space_cache_invalidate(&caches->used, pa, sizeof(flags));
386 }
387 
388 /* Called within rcu_read_lock().  */
389 static inline void vring_used_flags_unset_bit(VirtQueue *vq, int mask)
390 {
391     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
392     VirtIODevice *vdev = vq->vdev;
393     hwaddr pa = offsetof(VRingUsed, flags);
394     uint16_t flags;
395 
396     if (!caches) {
397         return;
398     }
399 
400     flags = virtio_lduw_phys_cached(vq->vdev, &caches->used, pa);
401     virtio_stw_phys_cached(vdev, &caches->used, pa, flags & ~mask);
402     address_space_cache_invalidate(&caches->used, pa, sizeof(flags));
403 }
404 
405 /* Called within rcu_read_lock().  */
406 static inline void vring_set_avail_event(VirtQueue *vq, uint16_t val)
407 {
408     VRingMemoryRegionCaches *caches;
409     hwaddr pa;
410     if (!vq->notification) {
411         return;
412     }
413 
414     caches = vring_get_region_caches(vq);
415     if (!caches) {
416         return;
417     }
418 
419     pa = offsetof(VRingUsed, ring[vq->vring.num]);
420     virtio_stw_phys_cached(vq->vdev, &caches->used, pa, val);
421     address_space_cache_invalidate(&caches->used, pa, sizeof(val));
422 }
423 
424 static void virtio_queue_split_set_notification(VirtQueue *vq, int enable)
425 {
426     RCU_READ_LOCK_GUARD();
427 
428     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_RING_F_EVENT_IDX)) {
429         vring_set_avail_event(vq, vring_avail_idx(vq));
430     } else if (enable) {
431         vring_used_flags_unset_bit(vq, VRING_USED_F_NO_NOTIFY);
432     } else {
433         vring_used_flags_set_bit(vq, VRING_USED_F_NO_NOTIFY);
434     }
435     if (enable) {
436         /* Expose avail event/used flags before caller checks the avail idx. */
437         smp_mb();
438     }
439 }
440 
441 static void virtio_queue_packed_set_notification(VirtQueue *vq, int enable)
442 {
443     uint16_t off_wrap;
444     VRingPackedDescEvent e;
445     VRingMemoryRegionCaches *caches;
446 
447     RCU_READ_LOCK_GUARD();
448     caches = vring_get_region_caches(vq);
449     if (!caches) {
450         return;
451     }
452 
453     vring_packed_event_read(vq->vdev, &caches->used, &e);
454 
455     if (!enable) {
456         e.flags = VRING_PACKED_EVENT_FLAG_DISABLE;
457     } else if (virtio_vdev_has_feature(vq->vdev, VIRTIO_RING_F_EVENT_IDX)) {
458         off_wrap = vq->shadow_avail_idx | vq->shadow_avail_wrap_counter << 15;
459         vring_packed_off_wrap_write(vq->vdev, &caches->used, off_wrap);
460         /* Make sure off_wrap is wrote before flags */
461         smp_wmb();
462         e.flags = VRING_PACKED_EVENT_FLAG_DESC;
463     } else {
464         e.flags = VRING_PACKED_EVENT_FLAG_ENABLE;
465     }
466 
467     vring_packed_flags_write(vq->vdev, &caches->used, e.flags);
468     if (enable) {
469         /* Expose avail event/used flags before caller checks the avail idx. */
470         smp_mb();
471     }
472 }
473 
474 bool virtio_queue_get_notification(VirtQueue *vq)
475 {
476     return vq->notification;
477 }
478 
479 void virtio_queue_set_notification(VirtQueue *vq, int enable)
480 {
481     vq->notification = enable;
482 
483     if (!vq->vring.desc) {
484         return;
485     }
486 
487     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) {
488         virtio_queue_packed_set_notification(vq, enable);
489     } else {
490         virtio_queue_split_set_notification(vq, enable);
491     }
492 }
493 
494 int virtio_queue_ready(VirtQueue *vq)
495 {
496     return vq->vring.avail != 0;
497 }
498 
499 static void vring_packed_desc_read_flags(VirtIODevice *vdev,
500                                          uint16_t *flags,
501                                          MemoryRegionCache *cache,
502                                          int i)
503 {
504     hwaddr off = i * sizeof(VRingPackedDesc) + offsetof(VRingPackedDesc, flags);
505 
506     *flags = virtio_lduw_phys_cached(vdev, cache, off);
507 }
508 
509 static void vring_packed_desc_read(VirtIODevice *vdev,
510                                    VRingPackedDesc *desc,
511                                    MemoryRegionCache *cache,
512                                    int i, bool strict_order)
513 {
514     hwaddr off = i * sizeof(VRingPackedDesc);
515 
516     vring_packed_desc_read_flags(vdev, &desc->flags, cache, i);
517 
518     if (strict_order) {
519         /* Make sure flags is read before the rest fields. */
520         smp_rmb();
521     }
522 
523     address_space_read_cached(cache, off + offsetof(VRingPackedDesc, addr),
524                               &desc->addr, sizeof(desc->addr));
525     address_space_read_cached(cache, off + offsetof(VRingPackedDesc, id),
526                               &desc->id, sizeof(desc->id));
527     address_space_read_cached(cache, off + offsetof(VRingPackedDesc, len),
528                               &desc->len, sizeof(desc->len));
529     virtio_tswap64s(vdev, &desc->addr);
530     virtio_tswap16s(vdev, &desc->id);
531     virtio_tswap32s(vdev, &desc->len);
532 }
533 
534 static void vring_packed_desc_write_data(VirtIODevice *vdev,
535                                          VRingPackedDesc *desc,
536                                          MemoryRegionCache *cache,
537                                          int i)
538 {
539     hwaddr off_id = i * sizeof(VRingPackedDesc) +
540                     offsetof(VRingPackedDesc, id);
541     hwaddr off_len = i * sizeof(VRingPackedDesc) +
542                     offsetof(VRingPackedDesc, len);
543 
544     virtio_tswap32s(vdev, &desc->len);
545     virtio_tswap16s(vdev, &desc->id);
546     address_space_write_cached(cache, off_id, &desc->id, sizeof(desc->id));
547     address_space_cache_invalidate(cache, off_id, sizeof(desc->id));
548     address_space_write_cached(cache, off_len, &desc->len, sizeof(desc->len));
549     address_space_cache_invalidate(cache, off_len, sizeof(desc->len));
550 }
551 
552 static void vring_packed_desc_write_flags(VirtIODevice *vdev,
553                                           VRingPackedDesc *desc,
554                                           MemoryRegionCache *cache,
555                                           int i)
556 {
557     hwaddr off = i * sizeof(VRingPackedDesc) + offsetof(VRingPackedDesc, flags);
558 
559     virtio_stw_phys_cached(vdev, cache, off, desc->flags);
560     address_space_cache_invalidate(cache, off, sizeof(desc->flags));
561 }
562 
563 static void vring_packed_desc_write(VirtIODevice *vdev,
564                                     VRingPackedDesc *desc,
565                                     MemoryRegionCache *cache,
566                                     int i, bool strict_order)
567 {
568     vring_packed_desc_write_data(vdev, desc, cache, i);
569     if (strict_order) {
570         /* Make sure data is wrote before flags. */
571         smp_wmb();
572     }
573     vring_packed_desc_write_flags(vdev, desc, cache, i);
574 }
575 
576 static inline bool is_desc_avail(uint16_t flags, bool wrap_counter)
577 {
578     bool avail, used;
579 
580     avail = !!(flags & (1 << VRING_PACKED_DESC_F_AVAIL));
581     used = !!(flags & (1 << VRING_PACKED_DESC_F_USED));
582     return (avail != used) && (avail == wrap_counter);
583 }
584 
585 /* Fetch avail_idx from VQ memory only when we really need to know if
586  * guest has added some buffers.
587  * Called within rcu_read_lock().  */
588 static int virtio_queue_empty_rcu(VirtQueue *vq)
589 {
590     if (virtio_device_disabled(vq->vdev)) {
591         return 1;
592     }
593 
594     if (unlikely(!vq->vring.avail)) {
595         return 1;
596     }
597 
598     if (vq->shadow_avail_idx != vq->last_avail_idx) {
599         return 0;
600     }
601 
602     return vring_avail_idx(vq) == vq->last_avail_idx;
603 }
604 
605 static int virtio_queue_split_empty(VirtQueue *vq)
606 {
607     bool empty;
608 
609     if (virtio_device_disabled(vq->vdev)) {
610         return 1;
611     }
612 
613     if (unlikely(!vq->vring.avail)) {
614         return 1;
615     }
616 
617     if (vq->shadow_avail_idx != vq->last_avail_idx) {
618         return 0;
619     }
620 
621     RCU_READ_LOCK_GUARD();
622     empty = vring_avail_idx(vq) == vq->last_avail_idx;
623     return empty;
624 }
625 
626 /* Called within rcu_read_lock().  */
627 static int virtio_queue_packed_empty_rcu(VirtQueue *vq)
628 {
629     struct VRingPackedDesc desc;
630     VRingMemoryRegionCaches *cache;
631 
632     if (unlikely(!vq->vring.desc)) {
633         return 1;
634     }
635 
636     cache = vring_get_region_caches(vq);
637     if (!cache) {
638         return 1;
639     }
640 
641     vring_packed_desc_read_flags(vq->vdev, &desc.flags, &cache->desc,
642                                  vq->last_avail_idx);
643 
644     return !is_desc_avail(desc.flags, vq->last_avail_wrap_counter);
645 }
646 
647 static int virtio_queue_packed_empty(VirtQueue *vq)
648 {
649     RCU_READ_LOCK_GUARD();
650     return virtio_queue_packed_empty_rcu(vq);
651 }
652 
653 int virtio_queue_empty(VirtQueue *vq)
654 {
655     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) {
656         return virtio_queue_packed_empty(vq);
657     } else {
658         return virtio_queue_split_empty(vq);
659     }
660 }
661 
662 static void virtqueue_unmap_sg(VirtQueue *vq, const VirtQueueElement *elem,
663                                unsigned int len)
664 {
665     AddressSpace *dma_as = vq->vdev->dma_as;
666     unsigned int offset;
667     int i;
668 
669     offset = 0;
670     for (i = 0; i < elem->in_num; i++) {
671         size_t size = MIN(len - offset, elem->in_sg[i].iov_len);
672 
673         dma_memory_unmap(dma_as, elem->in_sg[i].iov_base,
674                          elem->in_sg[i].iov_len,
675                          DMA_DIRECTION_FROM_DEVICE, size);
676 
677         offset += size;
678     }
679 
680     for (i = 0; i < elem->out_num; i++)
681         dma_memory_unmap(dma_as, elem->out_sg[i].iov_base,
682                          elem->out_sg[i].iov_len,
683                          DMA_DIRECTION_TO_DEVICE,
684                          elem->out_sg[i].iov_len);
685 }
686 
687 /* virtqueue_detach_element:
688  * @vq: The #VirtQueue
689  * @elem: The #VirtQueueElement
690  * @len: number of bytes written
691  *
692  * Detach the element from the virtqueue.  This function is suitable for device
693  * reset or other situations where a #VirtQueueElement is simply freed and will
694  * not be pushed or discarded.
695  */
696 void virtqueue_detach_element(VirtQueue *vq, const VirtQueueElement *elem,
697                               unsigned int len)
698 {
699     vq->inuse -= elem->ndescs;
700     virtqueue_unmap_sg(vq, elem, len);
701 }
702 
703 static void virtqueue_split_rewind(VirtQueue *vq, unsigned int num)
704 {
705     vq->last_avail_idx -= num;
706 }
707 
708 static void virtqueue_packed_rewind(VirtQueue *vq, unsigned int num)
709 {
710     if (vq->last_avail_idx < num) {
711         vq->last_avail_idx = vq->vring.num + vq->last_avail_idx - num;
712         vq->last_avail_wrap_counter ^= 1;
713     } else {
714         vq->last_avail_idx -= num;
715     }
716 }
717 
718 /* virtqueue_unpop:
719  * @vq: The #VirtQueue
720  * @elem: The #VirtQueueElement
721  * @len: number of bytes written
722  *
723  * Pretend the most recent element wasn't popped from the virtqueue.  The next
724  * call to virtqueue_pop() will refetch the element.
725  */
726 void virtqueue_unpop(VirtQueue *vq, const VirtQueueElement *elem,
727                      unsigned int len)
728 {
729 
730     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) {
731         virtqueue_packed_rewind(vq, 1);
732     } else {
733         virtqueue_split_rewind(vq, 1);
734     }
735 
736     virtqueue_detach_element(vq, elem, len);
737 }
738 
739 /* virtqueue_rewind:
740  * @vq: The #VirtQueue
741  * @num: Number of elements to push back
742  *
743  * Pretend that elements weren't popped from the virtqueue.  The next
744  * virtqueue_pop() will refetch the oldest element.
745  *
746  * Use virtqueue_unpop() instead if you have a VirtQueueElement.
747  *
748  * Returns: true on success, false if @num is greater than the number of in use
749  * elements.
750  */
751 bool virtqueue_rewind(VirtQueue *vq, unsigned int num)
752 {
753     if (num > vq->inuse) {
754         return false;
755     }
756 
757     vq->inuse -= num;
758     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) {
759         virtqueue_packed_rewind(vq, num);
760     } else {
761         virtqueue_split_rewind(vq, num);
762     }
763     return true;
764 }
765 
766 static void virtqueue_split_fill(VirtQueue *vq, const VirtQueueElement *elem,
767                     unsigned int len, unsigned int idx)
768 {
769     VRingUsedElem uelem;
770 
771     if (unlikely(!vq->vring.used)) {
772         return;
773     }
774 
775     idx = (idx + vq->used_idx) % vq->vring.num;
776 
777     uelem.id = elem->index;
778     uelem.len = len;
779     vring_used_write(vq, &uelem, idx);
780 }
781 
782 static void virtqueue_packed_fill(VirtQueue *vq, const VirtQueueElement *elem,
783                                   unsigned int len, unsigned int idx)
784 {
785     vq->used_elems[idx].index = elem->index;
786     vq->used_elems[idx].len = len;
787     vq->used_elems[idx].ndescs = elem->ndescs;
788 }
789 
790 static void virtqueue_packed_fill_desc(VirtQueue *vq,
791                                        const VirtQueueElement *elem,
792                                        unsigned int idx,
793                                        bool strict_order)
794 {
795     uint16_t head;
796     VRingMemoryRegionCaches *caches;
797     VRingPackedDesc desc = {
798         .id = elem->index,
799         .len = elem->len,
800     };
801     bool wrap_counter = vq->used_wrap_counter;
802 
803     if (unlikely(!vq->vring.desc)) {
804         return;
805     }
806 
807     head = vq->used_idx + idx;
808     if (head >= vq->vring.num) {
809         head -= vq->vring.num;
810         wrap_counter ^= 1;
811     }
812     if (wrap_counter) {
813         desc.flags |= (1 << VRING_PACKED_DESC_F_AVAIL);
814         desc.flags |= (1 << VRING_PACKED_DESC_F_USED);
815     } else {
816         desc.flags &= ~(1 << VRING_PACKED_DESC_F_AVAIL);
817         desc.flags &= ~(1 << VRING_PACKED_DESC_F_USED);
818     }
819 
820     caches = vring_get_region_caches(vq);
821     if (!caches) {
822         return;
823     }
824 
825     vring_packed_desc_write(vq->vdev, &desc, &caches->desc, head, strict_order);
826 }
827 
828 /* Called within rcu_read_lock().  */
829 void virtqueue_fill(VirtQueue *vq, const VirtQueueElement *elem,
830                     unsigned int len, unsigned int idx)
831 {
832     trace_virtqueue_fill(vq, elem, len, idx);
833 
834     virtqueue_unmap_sg(vq, elem, len);
835 
836     if (virtio_device_disabled(vq->vdev)) {
837         return;
838     }
839 
840     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) {
841         virtqueue_packed_fill(vq, elem, len, idx);
842     } else {
843         virtqueue_split_fill(vq, elem, len, idx);
844     }
845 }
846 
847 /* Called within rcu_read_lock().  */
848 static void virtqueue_split_flush(VirtQueue *vq, unsigned int count)
849 {
850     uint16_t old, new;
851 
852     if (unlikely(!vq->vring.used)) {
853         return;
854     }
855 
856     /* Make sure buffer is written before we update index. */
857     smp_wmb();
858     trace_virtqueue_flush(vq, count);
859     old = vq->used_idx;
860     new = old + count;
861     vring_used_idx_set(vq, new);
862     vq->inuse -= count;
863     if (unlikely((int16_t)(new - vq->signalled_used) < (uint16_t)(new - old)))
864         vq->signalled_used_valid = false;
865 }
866 
867 static void virtqueue_packed_flush(VirtQueue *vq, unsigned int count)
868 {
869     unsigned int i, ndescs = 0;
870 
871     if (unlikely(!vq->vring.desc)) {
872         return;
873     }
874 
875     for (i = 1; i < count; i++) {
876         virtqueue_packed_fill_desc(vq, &vq->used_elems[i], i, false);
877         ndescs += vq->used_elems[i].ndescs;
878     }
879     virtqueue_packed_fill_desc(vq, &vq->used_elems[0], 0, true);
880     ndescs += vq->used_elems[0].ndescs;
881 
882     vq->inuse -= ndescs;
883     vq->used_idx += ndescs;
884     if (vq->used_idx >= vq->vring.num) {
885         vq->used_idx -= vq->vring.num;
886         vq->used_wrap_counter ^= 1;
887         vq->signalled_used_valid = false;
888     }
889 }
890 
891 void virtqueue_flush(VirtQueue *vq, unsigned int count)
892 {
893     if (virtio_device_disabled(vq->vdev)) {
894         vq->inuse -= count;
895         return;
896     }
897 
898     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) {
899         virtqueue_packed_flush(vq, count);
900     } else {
901         virtqueue_split_flush(vq, count);
902     }
903 }
904 
905 void virtqueue_push(VirtQueue *vq, const VirtQueueElement *elem,
906                     unsigned int len)
907 {
908     RCU_READ_LOCK_GUARD();
909     virtqueue_fill(vq, elem, len, 0);
910     virtqueue_flush(vq, 1);
911 }
912 
913 /* Called within rcu_read_lock().  */
914 static int virtqueue_num_heads(VirtQueue *vq, unsigned int idx)
915 {
916     uint16_t num_heads = vring_avail_idx(vq) - idx;
917 
918     /* Check it isn't doing very strange things with descriptor numbers. */
919     if (num_heads > vq->vring.num) {
920         virtio_error(vq->vdev, "Guest moved used index from %u to %u",
921                      idx, vq->shadow_avail_idx);
922         return -EINVAL;
923     }
924     /* On success, callers read a descriptor at vq->last_avail_idx.
925      * Make sure descriptor read does not bypass avail index read. */
926     if (num_heads) {
927         smp_rmb();
928     }
929 
930     return num_heads;
931 }
932 
933 /* Called within rcu_read_lock().  */
934 static bool virtqueue_get_head(VirtQueue *vq, unsigned int idx,
935                                unsigned int *head)
936 {
937     /* Grab the next descriptor number they're advertising, and increment
938      * the index we've seen. */
939     *head = vring_avail_ring(vq, idx % vq->vring.num);
940 
941     /* If their number is silly, that's a fatal mistake. */
942     if (*head >= vq->vring.num) {
943         virtio_error(vq->vdev, "Guest says index %u is available", *head);
944         return false;
945     }
946 
947     return true;
948 }
949 
950 enum {
951     VIRTQUEUE_READ_DESC_ERROR = -1,
952     VIRTQUEUE_READ_DESC_DONE = 0,   /* end of chain */
953     VIRTQUEUE_READ_DESC_MORE = 1,   /* more buffers in chain */
954 };
955 
956 static int virtqueue_split_read_next_desc(VirtIODevice *vdev, VRingDesc *desc,
957                                           MemoryRegionCache *desc_cache,
958                                           unsigned int max, unsigned int *next)
959 {
960     /* If this descriptor says it doesn't chain, we're done. */
961     if (!(desc->flags & VRING_DESC_F_NEXT)) {
962         return VIRTQUEUE_READ_DESC_DONE;
963     }
964 
965     /* Check they're not leading us off end of descriptors. */
966     *next = desc->next;
967     /* Make sure compiler knows to grab that: we don't want it changing! */
968     smp_wmb();
969 
970     if (*next >= max) {
971         virtio_error(vdev, "Desc next is %u", *next);
972         return VIRTQUEUE_READ_DESC_ERROR;
973     }
974 
975     vring_split_desc_read(vdev, desc, desc_cache, *next);
976     return VIRTQUEUE_READ_DESC_MORE;
977 }
978 
979 /* Called within rcu_read_lock().  */
980 static void virtqueue_split_get_avail_bytes(VirtQueue *vq,
981                             unsigned int *in_bytes, unsigned int *out_bytes,
982                             unsigned max_in_bytes, unsigned max_out_bytes,
983                             VRingMemoryRegionCaches *caches)
984 {
985     VirtIODevice *vdev = vq->vdev;
986     unsigned int max, idx;
987     unsigned int total_bufs, in_total, out_total;
988     MemoryRegionCache indirect_desc_cache = MEMORY_REGION_CACHE_INVALID;
989     int64_t len = 0;
990     int rc;
991 
992     idx = vq->last_avail_idx;
993     total_bufs = in_total = out_total = 0;
994 
995     max = vq->vring.num;
996 
997     while ((rc = virtqueue_num_heads(vq, idx)) > 0) {
998         MemoryRegionCache *desc_cache = &caches->desc;
999         unsigned int num_bufs;
1000         VRingDesc desc;
1001         unsigned int i;
1002 
1003         num_bufs = total_bufs;
1004 
1005         if (!virtqueue_get_head(vq, idx++, &i)) {
1006             goto err;
1007         }
1008 
1009         vring_split_desc_read(vdev, &desc, desc_cache, i);
1010 
1011         if (desc.flags & VRING_DESC_F_INDIRECT) {
1012             if (!desc.len || (desc.len % sizeof(VRingDesc))) {
1013                 virtio_error(vdev, "Invalid size for indirect buffer table");
1014                 goto err;
1015             }
1016 
1017             /* If we've got too many, that implies a descriptor loop. */
1018             if (num_bufs >= max) {
1019                 virtio_error(vdev, "Looped descriptor");
1020                 goto err;
1021             }
1022 
1023             /* loop over the indirect descriptor table */
1024             len = address_space_cache_init(&indirect_desc_cache,
1025                                            vdev->dma_as,
1026                                            desc.addr, desc.len, false);
1027             desc_cache = &indirect_desc_cache;
1028             if (len < desc.len) {
1029                 virtio_error(vdev, "Cannot map indirect buffer");
1030                 goto err;
1031             }
1032 
1033             max = desc.len / sizeof(VRingDesc);
1034             num_bufs = i = 0;
1035             vring_split_desc_read(vdev, &desc, desc_cache, i);
1036         }
1037 
1038         do {
1039             /* If we've got too many, that implies a descriptor loop. */
1040             if (++num_bufs > max) {
1041                 virtio_error(vdev, "Looped descriptor");
1042                 goto err;
1043             }
1044 
1045             if (desc.flags & VRING_DESC_F_WRITE) {
1046                 in_total += desc.len;
1047             } else {
1048                 out_total += desc.len;
1049             }
1050             if (in_total >= max_in_bytes && out_total >= max_out_bytes) {
1051                 goto done;
1052             }
1053 
1054             rc = virtqueue_split_read_next_desc(vdev, &desc, desc_cache, max, &i);
1055         } while (rc == VIRTQUEUE_READ_DESC_MORE);
1056 
1057         if (rc == VIRTQUEUE_READ_DESC_ERROR) {
1058             goto err;
1059         }
1060 
1061         if (desc_cache == &indirect_desc_cache) {
1062             address_space_cache_destroy(&indirect_desc_cache);
1063             total_bufs++;
1064         } else {
1065             total_bufs = num_bufs;
1066         }
1067     }
1068 
1069     if (rc < 0) {
1070         goto err;
1071     }
1072 
1073 done:
1074     address_space_cache_destroy(&indirect_desc_cache);
1075     if (in_bytes) {
1076         *in_bytes = in_total;
1077     }
1078     if (out_bytes) {
1079         *out_bytes = out_total;
1080     }
1081     return;
1082 
1083 err:
1084     in_total = out_total = 0;
1085     goto done;
1086 }
1087 
1088 static int virtqueue_packed_read_next_desc(VirtQueue *vq,
1089                                            VRingPackedDesc *desc,
1090                                            MemoryRegionCache
1091                                            *desc_cache,
1092                                            unsigned int max,
1093                                            unsigned int *next,
1094                                            bool indirect)
1095 {
1096     /* If this descriptor says it doesn't chain, we're done. */
1097     if (!indirect && !(desc->flags & VRING_DESC_F_NEXT)) {
1098         return VIRTQUEUE_READ_DESC_DONE;
1099     }
1100 
1101     ++*next;
1102     if (*next == max) {
1103         if (indirect) {
1104             return VIRTQUEUE_READ_DESC_DONE;
1105         } else {
1106             (*next) -= vq->vring.num;
1107         }
1108     }
1109 
1110     vring_packed_desc_read(vq->vdev, desc, desc_cache, *next, false);
1111     return VIRTQUEUE_READ_DESC_MORE;
1112 }
1113 
1114 /* Called within rcu_read_lock().  */
1115 static void virtqueue_packed_get_avail_bytes(VirtQueue *vq,
1116                                              unsigned int *in_bytes,
1117                                              unsigned int *out_bytes,
1118                                              unsigned max_in_bytes,
1119                                              unsigned max_out_bytes,
1120                                              VRingMemoryRegionCaches *caches)
1121 {
1122     VirtIODevice *vdev = vq->vdev;
1123     unsigned int max, idx;
1124     unsigned int total_bufs, in_total, out_total;
1125     MemoryRegionCache *desc_cache;
1126     MemoryRegionCache indirect_desc_cache = MEMORY_REGION_CACHE_INVALID;
1127     int64_t len = 0;
1128     VRingPackedDesc desc;
1129     bool wrap_counter;
1130 
1131     idx = vq->last_avail_idx;
1132     wrap_counter = vq->last_avail_wrap_counter;
1133     total_bufs = in_total = out_total = 0;
1134 
1135     max = vq->vring.num;
1136 
1137     for (;;) {
1138         unsigned int num_bufs = total_bufs;
1139         unsigned int i = idx;
1140         int rc;
1141 
1142         desc_cache = &caches->desc;
1143         vring_packed_desc_read(vdev, &desc, desc_cache, idx, true);
1144         if (!is_desc_avail(desc.flags, wrap_counter)) {
1145             break;
1146         }
1147 
1148         if (desc.flags & VRING_DESC_F_INDIRECT) {
1149             if (desc.len % sizeof(VRingPackedDesc)) {
1150                 virtio_error(vdev, "Invalid size for indirect buffer table");
1151                 goto err;
1152             }
1153 
1154             /* If we've got too many, that implies a descriptor loop. */
1155             if (num_bufs >= max) {
1156                 virtio_error(vdev, "Looped descriptor");
1157                 goto err;
1158             }
1159 
1160             /* loop over the indirect descriptor table */
1161             len = address_space_cache_init(&indirect_desc_cache,
1162                                            vdev->dma_as,
1163                                            desc.addr, desc.len, false);
1164             desc_cache = &indirect_desc_cache;
1165             if (len < desc.len) {
1166                 virtio_error(vdev, "Cannot map indirect buffer");
1167                 goto err;
1168             }
1169 
1170             max = desc.len / sizeof(VRingPackedDesc);
1171             num_bufs = i = 0;
1172             vring_packed_desc_read(vdev, &desc, desc_cache, i, false);
1173         }
1174 
1175         do {
1176             /* If we've got too many, that implies a descriptor loop. */
1177             if (++num_bufs > max) {
1178                 virtio_error(vdev, "Looped descriptor");
1179                 goto err;
1180             }
1181 
1182             if (desc.flags & VRING_DESC_F_WRITE) {
1183                 in_total += desc.len;
1184             } else {
1185                 out_total += desc.len;
1186             }
1187             if (in_total >= max_in_bytes && out_total >= max_out_bytes) {
1188                 goto done;
1189             }
1190 
1191             rc = virtqueue_packed_read_next_desc(vq, &desc, desc_cache, max,
1192                                                  &i, desc_cache ==
1193                                                  &indirect_desc_cache);
1194         } while (rc == VIRTQUEUE_READ_DESC_MORE);
1195 
1196         if (desc_cache == &indirect_desc_cache) {
1197             address_space_cache_destroy(&indirect_desc_cache);
1198             total_bufs++;
1199             idx++;
1200         } else {
1201             idx += num_bufs - total_bufs;
1202             total_bufs = num_bufs;
1203         }
1204 
1205         if (idx >= vq->vring.num) {
1206             idx -= vq->vring.num;
1207             wrap_counter ^= 1;
1208         }
1209     }
1210 
1211     /* Record the index and wrap counter for a kick we want */
1212     vq->shadow_avail_idx = idx;
1213     vq->shadow_avail_wrap_counter = wrap_counter;
1214 done:
1215     address_space_cache_destroy(&indirect_desc_cache);
1216     if (in_bytes) {
1217         *in_bytes = in_total;
1218     }
1219     if (out_bytes) {
1220         *out_bytes = out_total;
1221     }
1222     return;
1223 
1224 err:
1225     in_total = out_total = 0;
1226     goto done;
1227 }
1228 
1229 void virtqueue_get_avail_bytes(VirtQueue *vq, unsigned int *in_bytes,
1230                                unsigned int *out_bytes,
1231                                unsigned max_in_bytes, unsigned max_out_bytes)
1232 {
1233     uint16_t desc_size;
1234     VRingMemoryRegionCaches *caches;
1235 
1236     RCU_READ_LOCK_GUARD();
1237 
1238     if (unlikely(!vq->vring.desc)) {
1239         goto err;
1240     }
1241 
1242     caches = vring_get_region_caches(vq);
1243     if (!caches) {
1244         goto err;
1245     }
1246 
1247     desc_size = virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED) ?
1248                                 sizeof(VRingPackedDesc) : sizeof(VRingDesc);
1249     if (caches->desc.len < vq->vring.num * desc_size) {
1250         virtio_error(vq->vdev, "Cannot map descriptor ring");
1251         goto err;
1252     }
1253 
1254     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) {
1255         virtqueue_packed_get_avail_bytes(vq, in_bytes, out_bytes,
1256                                          max_in_bytes, max_out_bytes,
1257                                          caches);
1258     } else {
1259         virtqueue_split_get_avail_bytes(vq, in_bytes, out_bytes,
1260                                         max_in_bytes, max_out_bytes,
1261                                         caches);
1262     }
1263 
1264     return;
1265 err:
1266     if (in_bytes) {
1267         *in_bytes = 0;
1268     }
1269     if (out_bytes) {
1270         *out_bytes = 0;
1271     }
1272 }
1273 
1274 int virtqueue_avail_bytes(VirtQueue *vq, unsigned int in_bytes,
1275                           unsigned int out_bytes)
1276 {
1277     unsigned int in_total, out_total;
1278 
1279     virtqueue_get_avail_bytes(vq, &in_total, &out_total, in_bytes, out_bytes);
1280     return in_bytes <= in_total && out_bytes <= out_total;
1281 }
1282 
1283 static bool virtqueue_map_desc(VirtIODevice *vdev, unsigned int *p_num_sg,
1284                                hwaddr *addr, struct iovec *iov,
1285                                unsigned int max_num_sg, bool is_write,
1286                                hwaddr pa, size_t sz)
1287 {
1288     bool ok = false;
1289     unsigned num_sg = *p_num_sg;
1290     assert(num_sg <= max_num_sg);
1291 
1292     if (!sz) {
1293         virtio_error(vdev, "virtio: zero sized buffers are not allowed");
1294         goto out;
1295     }
1296 
1297     while (sz) {
1298         hwaddr len = sz;
1299 
1300         if (num_sg == max_num_sg) {
1301             virtio_error(vdev, "virtio: too many write descriptors in "
1302                                "indirect table");
1303             goto out;
1304         }
1305 
1306         iov[num_sg].iov_base = dma_memory_map(vdev->dma_as, pa, &len,
1307                                               is_write ?
1308                                               DMA_DIRECTION_FROM_DEVICE :
1309                                               DMA_DIRECTION_TO_DEVICE,
1310                                               MEMTXATTRS_UNSPECIFIED);
1311         if (!iov[num_sg].iov_base) {
1312             virtio_error(vdev, "virtio: bogus descriptor or out of resources");
1313             goto out;
1314         }
1315 
1316         iov[num_sg].iov_len = len;
1317         addr[num_sg] = pa;
1318 
1319         sz -= len;
1320         pa += len;
1321         num_sg++;
1322     }
1323     ok = true;
1324 
1325 out:
1326     *p_num_sg = num_sg;
1327     return ok;
1328 }
1329 
1330 /* Only used by error code paths before we have a VirtQueueElement (therefore
1331  * virtqueue_unmap_sg() can't be used).  Assumes buffers weren't written to
1332  * yet.
1333  */
1334 static void virtqueue_undo_map_desc(unsigned int out_num, unsigned int in_num,
1335                                     struct iovec *iov)
1336 {
1337     unsigned int i;
1338 
1339     for (i = 0; i < out_num + in_num; i++) {
1340         int is_write = i >= out_num;
1341 
1342         cpu_physical_memory_unmap(iov->iov_base, iov->iov_len, is_write, 0);
1343         iov++;
1344     }
1345 }
1346 
1347 static void virtqueue_map_iovec(VirtIODevice *vdev, struct iovec *sg,
1348                                 hwaddr *addr, unsigned int num_sg,
1349                                 bool is_write)
1350 {
1351     unsigned int i;
1352     hwaddr len;
1353 
1354     for (i = 0; i < num_sg; i++) {
1355         len = sg[i].iov_len;
1356         sg[i].iov_base = dma_memory_map(vdev->dma_as,
1357                                         addr[i], &len, is_write ?
1358                                         DMA_DIRECTION_FROM_DEVICE :
1359                                         DMA_DIRECTION_TO_DEVICE,
1360                                         MEMTXATTRS_UNSPECIFIED);
1361         if (!sg[i].iov_base) {
1362             error_report("virtio: error trying to map MMIO memory");
1363             exit(1);
1364         }
1365         if (len != sg[i].iov_len) {
1366             error_report("virtio: unexpected memory split");
1367             exit(1);
1368         }
1369     }
1370 }
1371 
1372 void virtqueue_map(VirtIODevice *vdev, VirtQueueElement *elem)
1373 {
1374     virtqueue_map_iovec(vdev, elem->in_sg, elem->in_addr, elem->in_num, true);
1375     virtqueue_map_iovec(vdev, elem->out_sg, elem->out_addr, elem->out_num,
1376                                                                         false);
1377 }
1378 
1379 static void *virtqueue_alloc_element(size_t sz, unsigned out_num, unsigned in_num)
1380 {
1381     VirtQueueElement *elem;
1382     size_t in_addr_ofs = QEMU_ALIGN_UP(sz, __alignof__(elem->in_addr[0]));
1383     size_t out_addr_ofs = in_addr_ofs + in_num * sizeof(elem->in_addr[0]);
1384     size_t out_addr_end = out_addr_ofs + out_num * sizeof(elem->out_addr[0]);
1385     size_t in_sg_ofs = QEMU_ALIGN_UP(out_addr_end, __alignof__(elem->in_sg[0]));
1386     size_t out_sg_ofs = in_sg_ofs + in_num * sizeof(elem->in_sg[0]);
1387     size_t out_sg_end = out_sg_ofs + out_num * sizeof(elem->out_sg[0]);
1388 
1389     assert(sz >= sizeof(VirtQueueElement));
1390     elem = g_malloc(out_sg_end);
1391     trace_virtqueue_alloc_element(elem, sz, in_num, out_num);
1392     elem->out_num = out_num;
1393     elem->in_num = in_num;
1394     elem->in_addr = (void *)elem + in_addr_ofs;
1395     elem->out_addr = (void *)elem + out_addr_ofs;
1396     elem->in_sg = (void *)elem + in_sg_ofs;
1397     elem->out_sg = (void *)elem + out_sg_ofs;
1398     return elem;
1399 }
1400 
1401 static void *virtqueue_split_pop(VirtQueue *vq, size_t sz)
1402 {
1403     unsigned int i, head, max;
1404     VRingMemoryRegionCaches *caches;
1405     MemoryRegionCache indirect_desc_cache = MEMORY_REGION_CACHE_INVALID;
1406     MemoryRegionCache *desc_cache;
1407     int64_t len;
1408     VirtIODevice *vdev = vq->vdev;
1409     VirtQueueElement *elem = NULL;
1410     unsigned out_num, in_num, elem_entries;
1411     hwaddr addr[VIRTQUEUE_MAX_SIZE];
1412     struct iovec iov[VIRTQUEUE_MAX_SIZE];
1413     VRingDesc desc;
1414     int rc;
1415 
1416     RCU_READ_LOCK_GUARD();
1417     if (virtio_queue_empty_rcu(vq)) {
1418         goto done;
1419     }
1420     /* Needed after virtio_queue_empty(), see comment in
1421      * virtqueue_num_heads(). */
1422     smp_rmb();
1423 
1424     /* When we start there are none of either input nor output. */
1425     out_num = in_num = elem_entries = 0;
1426 
1427     max = vq->vring.num;
1428 
1429     if (vq->inuse >= vq->vring.num) {
1430         virtio_error(vdev, "Virtqueue size exceeded");
1431         goto done;
1432     }
1433 
1434     if (!virtqueue_get_head(vq, vq->last_avail_idx++, &head)) {
1435         goto done;
1436     }
1437 
1438     if (virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX)) {
1439         vring_set_avail_event(vq, vq->last_avail_idx);
1440     }
1441 
1442     i = head;
1443 
1444     caches = vring_get_region_caches(vq);
1445     if (!caches) {
1446         virtio_error(vdev, "Region caches not initialized");
1447         goto done;
1448     }
1449 
1450     if (caches->desc.len < max * sizeof(VRingDesc)) {
1451         virtio_error(vdev, "Cannot map descriptor ring");
1452         goto done;
1453     }
1454 
1455     desc_cache = &caches->desc;
1456     vring_split_desc_read(vdev, &desc, desc_cache, i);
1457     if (desc.flags & VRING_DESC_F_INDIRECT) {
1458         if (!desc.len || (desc.len % sizeof(VRingDesc))) {
1459             virtio_error(vdev, "Invalid size for indirect buffer table");
1460             goto done;
1461         }
1462 
1463         /* loop over the indirect descriptor table */
1464         len = address_space_cache_init(&indirect_desc_cache, vdev->dma_as,
1465                                        desc.addr, desc.len, false);
1466         desc_cache = &indirect_desc_cache;
1467         if (len < desc.len) {
1468             virtio_error(vdev, "Cannot map indirect buffer");
1469             goto done;
1470         }
1471 
1472         max = desc.len / sizeof(VRingDesc);
1473         i = 0;
1474         vring_split_desc_read(vdev, &desc, desc_cache, i);
1475     }
1476 
1477     /* Collect all the descriptors */
1478     do {
1479         bool map_ok;
1480 
1481         if (desc.flags & VRING_DESC_F_WRITE) {
1482             map_ok = virtqueue_map_desc(vdev, &in_num, addr + out_num,
1483                                         iov + out_num,
1484                                         VIRTQUEUE_MAX_SIZE - out_num, true,
1485                                         desc.addr, desc.len);
1486         } else {
1487             if (in_num) {
1488                 virtio_error(vdev, "Incorrect order for descriptors");
1489                 goto err_undo_map;
1490             }
1491             map_ok = virtqueue_map_desc(vdev, &out_num, addr, iov,
1492                                         VIRTQUEUE_MAX_SIZE, false,
1493                                         desc.addr, desc.len);
1494         }
1495         if (!map_ok) {
1496             goto err_undo_map;
1497         }
1498 
1499         /* If we've got too many, that implies a descriptor loop. */
1500         if (++elem_entries > max) {
1501             virtio_error(vdev, "Looped descriptor");
1502             goto err_undo_map;
1503         }
1504 
1505         rc = virtqueue_split_read_next_desc(vdev, &desc, desc_cache, max, &i);
1506     } while (rc == VIRTQUEUE_READ_DESC_MORE);
1507 
1508     if (rc == VIRTQUEUE_READ_DESC_ERROR) {
1509         goto err_undo_map;
1510     }
1511 
1512     /* Now copy what we have collected and mapped */
1513     elem = virtqueue_alloc_element(sz, out_num, in_num);
1514     elem->index = head;
1515     elem->ndescs = 1;
1516     for (i = 0; i < out_num; i++) {
1517         elem->out_addr[i] = addr[i];
1518         elem->out_sg[i] = iov[i];
1519     }
1520     for (i = 0; i < in_num; i++) {
1521         elem->in_addr[i] = addr[out_num + i];
1522         elem->in_sg[i] = iov[out_num + i];
1523     }
1524 
1525     vq->inuse++;
1526 
1527     trace_virtqueue_pop(vq, elem, elem->in_num, elem->out_num);
1528 done:
1529     address_space_cache_destroy(&indirect_desc_cache);
1530 
1531     return elem;
1532 
1533 err_undo_map:
1534     virtqueue_undo_map_desc(out_num, in_num, iov);
1535     goto done;
1536 }
1537 
1538 static void *virtqueue_packed_pop(VirtQueue *vq, size_t sz)
1539 {
1540     unsigned int i, max;
1541     VRingMemoryRegionCaches *caches;
1542     MemoryRegionCache indirect_desc_cache = MEMORY_REGION_CACHE_INVALID;
1543     MemoryRegionCache *desc_cache;
1544     int64_t len;
1545     VirtIODevice *vdev = vq->vdev;
1546     VirtQueueElement *elem = NULL;
1547     unsigned out_num, in_num, elem_entries;
1548     hwaddr addr[VIRTQUEUE_MAX_SIZE];
1549     struct iovec iov[VIRTQUEUE_MAX_SIZE];
1550     VRingPackedDesc desc;
1551     uint16_t id;
1552     int rc;
1553 
1554     RCU_READ_LOCK_GUARD();
1555     if (virtio_queue_packed_empty_rcu(vq)) {
1556         goto done;
1557     }
1558 
1559     /* When we start there are none of either input nor output. */
1560     out_num = in_num = elem_entries = 0;
1561 
1562     max = vq->vring.num;
1563 
1564     if (vq->inuse >= vq->vring.num) {
1565         virtio_error(vdev, "Virtqueue size exceeded");
1566         goto done;
1567     }
1568 
1569     i = vq->last_avail_idx;
1570 
1571     caches = vring_get_region_caches(vq);
1572     if (!caches) {
1573         virtio_error(vdev, "Region caches not initialized");
1574         goto done;
1575     }
1576 
1577     if (caches->desc.len < max * sizeof(VRingDesc)) {
1578         virtio_error(vdev, "Cannot map descriptor ring");
1579         goto done;
1580     }
1581 
1582     desc_cache = &caches->desc;
1583     vring_packed_desc_read(vdev, &desc, desc_cache, i, true);
1584     id = desc.id;
1585     if (desc.flags & VRING_DESC_F_INDIRECT) {
1586         if (desc.len % sizeof(VRingPackedDesc)) {
1587             virtio_error(vdev, "Invalid size for indirect buffer table");
1588             goto done;
1589         }
1590 
1591         /* loop over the indirect descriptor table */
1592         len = address_space_cache_init(&indirect_desc_cache, vdev->dma_as,
1593                                        desc.addr, desc.len, false);
1594         desc_cache = &indirect_desc_cache;
1595         if (len < desc.len) {
1596             virtio_error(vdev, "Cannot map indirect buffer");
1597             goto done;
1598         }
1599 
1600         max = desc.len / sizeof(VRingPackedDesc);
1601         i = 0;
1602         vring_packed_desc_read(vdev, &desc, desc_cache, i, false);
1603     }
1604 
1605     /* Collect all the descriptors */
1606     do {
1607         bool map_ok;
1608 
1609         if (desc.flags & VRING_DESC_F_WRITE) {
1610             map_ok = virtqueue_map_desc(vdev, &in_num, addr + out_num,
1611                                         iov + out_num,
1612                                         VIRTQUEUE_MAX_SIZE - out_num, true,
1613                                         desc.addr, desc.len);
1614         } else {
1615             if (in_num) {
1616                 virtio_error(vdev, "Incorrect order for descriptors");
1617                 goto err_undo_map;
1618             }
1619             map_ok = virtqueue_map_desc(vdev, &out_num, addr, iov,
1620                                         VIRTQUEUE_MAX_SIZE, false,
1621                                         desc.addr, desc.len);
1622         }
1623         if (!map_ok) {
1624             goto err_undo_map;
1625         }
1626 
1627         /* If we've got too many, that implies a descriptor loop. */
1628         if (++elem_entries > max) {
1629             virtio_error(vdev, "Looped descriptor");
1630             goto err_undo_map;
1631         }
1632 
1633         rc = virtqueue_packed_read_next_desc(vq, &desc, desc_cache, max, &i,
1634                                              desc_cache ==
1635                                              &indirect_desc_cache);
1636     } while (rc == VIRTQUEUE_READ_DESC_MORE);
1637 
1638     /* Now copy what we have collected and mapped */
1639     elem = virtqueue_alloc_element(sz, out_num, in_num);
1640     for (i = 0; i < out_num; i++) {
1641         elem->out_addr[i] = addr[i];
1642         elem->out_sg[i] = iov[i];
1643     }
1644     for (i = 0; i < in_num; i++) {
1645         elem->in_addr[i] = addr[out_num + i];
1646         elem->in_sg[i] = iov[out_num + i];
1647     }
1648 
1649     elem->index = id;
1650     elem->ndescs = (desc_cache == &indirect_desc_cache) ? 1 : elem_entries;
1651     vq->last_avail_idx += elem->ndescs;
1652     vq->inuse += elem->ndescs;
1653 
1654     if (vq->last_avail_idx >= vq->vring.num) {
1655         vq->last_avail_idx -= vq->vring.num;
1656         vq->last_avail_wrap_counter ^= 1;
1657     }
1658 
1659     vq->shadow_avail_idx = vq->last_avail_idx;
1660     vq->shadow_avail_wrap_counter = vq->last_avail_wrap_counter;
1661 
1662     trace_virtqueue_pop(vq, elem, elem->in_num, elem->out_num);
1663 done:
1664     address_space_cache_destroy(&indirect_desc_cache);
1665 
1666     return elem;
1667 
1668 err_undo_map:
1669     virtqueue_undo_map_desc(out_num, in_num, iov);
1670     goto done;
1671 }
1672 
1673 void *virtqueue_pop(VirtQueue *vq, size_t sz)
1674 {
1675     if (virtio_device_disabled(vq->vdev)) {
1676         return NULL;
1677     }
1678 
1679     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) {
1680         return virtqueue_packed_pop(vq, sz);
1681     } else {
1682         return virtqueue_split_pop(vq, sz);
1683     }
1684 }
1685 
1686 static unsigned int virtqueue_packed_drop_all(VirtQueue *vq)
1687 {
1688     VRingMemoryRegionCaches *caches;
1689     MemoryRegionCache *desc_cache;
1690     unsigned int dropped = 0;
1691     VirtQueueElement elem = {};
1692     VirtIODevice *vdev = vq->vdev;
1693     VRingPackedDesc desc;
1694 
1695     RCU_READ_LOCK_GUARD();
1696 
1697     caches = vring_get_region_caches(vq);
1698     if (!caches) {
1699         return 0;
1700     }
1701 
1702     desc_cache = &caches->desc;
1703 
1704     virtio_queue_set_notification(vq, 0);
1705 
1706     while (vq->inuse < vq->vring.num) {
1707         unsigned int idx = vq->last_avail_idx;
1708         /*
1709          * works similar to virtqueue_pop but does not map buffers
1710          * and does not allocate any memory.
1711          */
1712         vring_packed_desc_read(vdev, &desc, desc_cache,
1713                                vq->last_avail_idx , true);
1714         if (!is_desc_avail(desc.flags, vq->last_avail_wrap_counter)) {
1715             break;
1716         }
1717         elem.index = desc.id;
1718         elem.ndescs = 1;
1719         while (virtqueue_packed_read_next_desc(vq, &desc, desc_cache,
1720                                                vq->vring.num, &idx, false)) {
1721             ++elem.ndescs;
1722         }
1723         /*
1724          * immediately push the element, nothing to unmap
1725          * as both in_num and out_num are set to 0.
1726          */
1727         virtqueue_push(vq, &elem, 0);
1728         dropped++;
1729         vq->last_avail_idx += elem.ndescs;
1730         if (vq->last_avail_idx >= vq->vring.num) {
1731             vq->last_avail_idx -= vq->vring.num;
1732             vq->last_avail_wrap_counter ^= 1;
1733         }
1734     }
1735 
1736     return dropped;
1737 }
1738 
1739 static unsigned int virtqueue_split_drop_all(VirtQueue *vq)
1740 {
1741     unsigned int dropped = 0;
1742     VirtQueueElement elem = {};
1743     VirtIODevice *vdev = vq->vdev;
1744     bool fEventIdx = virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX);
1745 
1746     while (!virtio_queue_empty(vq) && vq->inuse < vq->vring.num) {
1747         /* works similar to virtqueue_pop but does not map buffers
1748         * and does not allocate any memory */
1749         smp_rmb();
1750         if (!virtqueue_get_head(vq, vq->last_avail_idx, &elem.index)) {
1751             break;
1752         }
1753         vq->inuse++;
1754         vq->last_avail_idx++;
1755         if (fEventIdx) {
1756             vring_set_avail_event(vq, vq->last_avail_idx);
1757         }
1758         /* immediately push the element, nothing to unmap
1759          * as both in_num and out_num are set to 0 */
1760         virtqueue_push(vq, &elem, 0);
1761         dropped++;
1762     }
1763 
1764     return dropped;
1765 }
1766 
1767 /* virtqueue_drop_all:
1768  * @vq: The #VirtQueue
1769  * Drops all queued buffers and indicates them to the guest
1770  * as if they are done. Useful when buffers can not be
1771  * processed but must be returned to the guest.
1772  */
1773 unsigned int virtqueue_drop_all(VirtQueue *vq)
1774 {
1775     struct VirtIODevice *vdev = vq->vdev;
1776 
1777     if (virtio_device_disabled(vq->vdev)) {
1778         return 0;
1779     }
1780 
1781     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
1782         return virtqueue_packed_drop_all(vq);
1783     } else {
1784         return virtqueue_split_drop_all(vq);
1785     }
1786 }
1787 
1788 /* Reading and writing a structure directly to QEMUFile is *awful*, but
1789  * it is what QEMU has always done by mistake.  We can change it sooner
1790  * or later by bumping the version number of the affected vm states.
1791  * In the meanwhile, since the in-memory layout of VirtQueueElement
1792  * has changed, we need to marshal to and from the layout that was
1793  * used before the change.
1794  */
1795 typedef struct VirtQueueElementOld {
1796     unsigned int index;
1797     unsigned int out_num;
1798     unsigned int in_num;
1799     hwaddr in_addr[VIRTQUEUE_MAX_SIZE];
1800     hwaddr out_addr[VIRTQUEUE_MAX_SIZE];
1801     struct iovec in_sg[VIRTQUEUE_MAX_SIZE];
1802     struct iovec out_sg[VIRTQUEUE_MAX_SIZE];
1803 } VirtQueueElementOld;
1804 
1805 void *qemu_get_virtqueue_element(VirtIODevice *vdev, QEMUFile *f, size_t sz)
1806 {
1807     VirtQueueElement *elem;
1808     VirtQueueElementOld data;
1809     int i;
1810 
1811     qemu_get_buffer(f, (uint8_t *)&data, sizeof(VirtQueueElementOld));
1812 
1813     /* TODO: teach all callers that this can fail, and return failure instead
1814      * of asserting here.
1815      * This is just one thing (there are probably more) that must be
1816      * fixed before we can allow NDEBUG compilation.
1817      */
1818     assert(ARRAY_SIZE(data.in_addr) >= data.in_num);
1819     assert(ARRAY_SIZE(data.out_addr) >= data.out_num);
1820 
1821     elem = virtqueue_alloc_element(sz, data.out_num, data.in_num);
1822     elem->index = data.index;
1823 
1824     for (i = 0; i < elem->in_num; i++) {
1825         elem->in_addr[i] = data.in_addr[i];
1826     }
1827 
1828     for (i = 0; i < elem->out_num; i++) {
1829         elem->out_addr[i] = data.out_addr[i];
1830     }
1831 
1832     for (i = 0; i < elem->in_num; i++) {
1833         /* Base is overwritten by virtqueue_map.  */
1834         elem->in_sg[i].iov_base = 0;
1835         elem->in_sg[i].iov_len = data.in_sg[i].iov_len;
1836     }
1837 
1838     for (i = 0; i < elem->out_num; i++) {
1839         /* Base is overwritten by virtqueue_map.  */
1840         elem->out_sg[i].iov_base = 0;
1841         elem->out_sg[i].iov_len = data.out_sg[i].iov_len;
1842     }
1843 
1844     if (virtio_host_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
1845         qemu_get_be32s(f, &elem->ndescs);
1846     }
1847 
1848     virtqueue_map(vdev, elem);
1849     return elem;
1850 }
1851 
1852 void qemu_put_virtqueue_element(VirtIODevice *vdev, QEMUFile *f,
1853                                 VirtQueueElement *elem)
1854 {
1855     VirtQueueElementOld data;
1856     int i;
1857 
1858     memset(&data, 0, sizeof(data));
1859     data.index = elem->index;
1860     data.in_num = elem->in_num;
1861     data.out_num = elem->out_num;
1862 
1863     for (i = 0; i < elem->in_num; i++) {
1864         data.in_addr[i] = elem->in_addr[i];
1865     }
1866 
1867     for (i = 0; i < elem->out_num; i++) {
1868         data.out_addr[i] = elem->out_addr[i];
1869     }
1870 
1871     for (i = 0; i < elem->in_num; i++) {
1872         /* Base is overwritten by virtqueue_map when loading.  Do not
1873          * save it, as it would leak the QEMU address space layout.  */
1874         data.in_sg[i].iov_len = elem->in_sg[i].iov_len;
1875     }
1876 
1877     for (i = 0; i < elem->out_num; i++) {
1878         /* Do not save iov_base as above.  */
1879         data.out_sg[i].iov_len = elem->out_sg[i].iov_len;
1880     }
1881 
1882     if (virtio_host_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
1883         qemu_put_be32s(f, &elem->ndescs);
1884     }
1885 
1886     qemu_put_buffer(f, (uint8_t *)&data, sizeof(VirtQueueElementOld));
1887 }
1888 
1889 /* virtio device */
1890 static void virtio_notify_vector(VirtIODevice *vdev, uint16_t vector)
1891 {
1892     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
1893     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
1894 
1895     if (virtio_device_disabled(vdev)) {
1896         return;
1897     }
1898 
1899     if (k->notify) {
1900         k->notify(qbus->parent, vector);
1901     }
1902 }
1903 
1904 void virtio_update_irq(VirtIODevice *vdev)
1905 {
1906     virtio_notify_vector(vdev, VIRTIO_NO_VECTOR);
1907 }
1908 
1909 static int virtio_validate_features(VirtIODevice *vdev)
1910 {
1911     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1912 
1913     if (virtio_host_has_feature(vdev, VIRTIO_F_IOMMU_PLATFORM) &&
1914         !virtio_vdev_has_feature(vdev, VIRTIO_F_IOMMU_PLATFORM)) {
1915         return -EFAULT;
1916     }
1917 
1918     if (k->validate_features) {
1919         return k->validate_features(vdev);
1920     } else {
1921         return 0;
1922     }
1923 }
1924 
1925 int virtio_set_status(VirtIODevice *vdev, uint8_t val)
1926 {
1927     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1928     trace_virtio_set_status(vdev, val);
1929 
1930     if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
1931         if (!(vdev->status & VIRTIO_CONFIG_S_FEATURES_OK) &&
1932             val & VIRTIO_CONFIG_S_FEATURES_OK) {
1933             int ret = virtio_validate_features(vdev);
1934 
1935             if (ret) {
1936                 return ret;
1937             }
1938         }
1939     }
1940 
1941     if ((vdev->status & VIRTIO_CONFIG_S_DRIVER_OK) !=
1942         (val & VIRTIO_CONFIG_S_DRIVER_OK)) {
1943         virtio_set_started(vdev, val & VIRTIO_CONFIG_S_DRIVER_OK);
1944     }
1945 
1946     if (k->set_status) {
1947         k->set_status(vdev, val);
1948     }
1949     vdev->status = val;
1950 
1951     return 0;
1952 }
1953 
1954 static enum virtio_device_endian virtio_default_endian(void)
1955 {
1956     if (target_words_bigendian()) {
1957         return VIRTIO_DEVICE_ENDIAN_BIG;
1958     } else {
1959         return VIRTIO_DEVICE_ENDIAN_LITTLE;
1960     }
1961 }
1962 
1963 static enum virtio_device_endian virtio_current_cpu_endian(void)
1964 {
1965     if (cpu_virtio_is_big_endian(current_cpu)) {
1966         return VIRTIO_DEVICE_ENDIAN_BIG;
1967     } else {
1968         return VIRTIO_DEVICE_ENDIAN_LITTLE;
1969     }
1970 }
1971 
1972 void virtio_reset(void *opaque)
1973 {
1974     VirtIODevice *vdev = opaque;
1975     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1976     int i;
1977 
1978     virtio_set_status(vdev, 0);
1979     if (current_cpu) {
1980         /* Guest initiated reset */
1981         vdev->device_endian = virtio_current_cpu_endian();
1982     } else {
1983         /* System reset */
1984         vdev->device_endian = virtio_default_endian();
1985     }
1986 
1987     if (k->reset) {
1988         k->reset(vdev);
1989     }
1990 
1991     vdev->start_on_kick = false;
1992     vdev->started = false;
1993     vdev->broken = false;
1994     vdev->guest_features = 0;
1995     vdev->queue_sel = 0;
1996     vdev->status = 0;
1997     vdev->disabled = false;
1998     qatomic_set(&vdev->isr, 0);
1999     vdev->config_vector = VIRTIO_NO_VECTOR;
2000     virtio_notify_vector(vdev, vdev->config_vector);
2001 
2002     for(i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2003         vdev->vq[i].vring.desc = 0;
2004         vdev->vq[i].vring.avail = 0;
2005         vdev->vq[i].vring.used = 0;
2006         vdev->vq[i].last_avail_idx = 0;
2007         vdev->vq[i].shadow_avail_idx = 0;
2008         vdev->vq[i].used_idx = 0;
2009         vdev->vq[i].last_avail_wrap_counter = true;
2010         vdev->vq[i].shadow_avail_wrap_counter = true;
2011         vdev->vq[i].used_wrap_counter = true;
2012         virtio_queue_set_vector(vdev, i, VIRTIO_NO_VECTOR);
2013         vdev->vq[i].signalled_used = 0;
2014         vdev->vq[i].signalled_used_valid = false;
2015         vdev->vq[i].notification = true;
2016         vdev->vq[i].vring.num = vdev->vq[i].vring.num_default;
2017         vdev->vq[i].inuse = 0;
2018         virtio_virtqueue_reset_region_cache(&vdev->vq[i]);
2019     }
2020 }
2021 
2022 uint32_t virtio_config_readb(VirtIODevice *vdev, uint32_t addr)
2023 {
2024     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2025     uint8_t val;
2026 
2027     if (addr + sizeof(val) > vdev->config_len) {
2028         return (uint32_t)-1;
2029     }
2030 
2031     k->get_config(vdev, vdev->config);
2032 
2033     val = ldub_p(vdev->config + addr);
2034     return val;
2035 }
2036 
2037 uint32_t virtio_config_readw(VirtIODevice *vdev, uint32_t addr)
2038 {
2039     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2040     uint16_t val;
2041 
2042     if (addr + sizeof(val) > vdev->config_len) {
2043         return (uint32_t)-1;
2044     }
2045 
2046     k->get_config(vdev, vdev->config);
2047 
2048     val = lduw_p(vdev->config + addr);
2049     return val;
2050 }
2051 
2052 uint32_t virtio_config_readl(VirtIODevice *vdev, uint32_t addr)
2053 {
2054     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2055     uint32_t val;
2056 
2057     if (addr + sizeof(val) > vdev->config_len) {
2058         return (uint32_t)-1;
2059     }
2060 
2061     k->get_config(vdev, vdev->config);
2062 
2063     val = ldl_p(vdev->config + addr);
2064     return val;
2065 }
2066 
2067 void virtio_config_writeb(VirtIODevice *vdev, uint32_t addr, uint32_t data)
2068 {
2069     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2070     uint8_t val = data;
2071 
2072     if (addr + sizeof(val) > vdev->config_len) {
2073         return;
2074     }
2075 
2076     stb_p(vdev->config + addr, val);
2077 
2078     if (k->set_config) {
2079         k->set_config(vdev, vdev->config);
2080     }
2081 }
2082 
2083 void virtio_config_writew(VirtIODevice *vdev, uint32_t addr, uint32_t data)
2084 {
2085     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2086     uint16_t val = data;
2087 
2088     if (addr + sizeof(val) > vdev->config_len) {
2089         return;
2090     }
2091 
2092     stw_p(vdev->config + addr, val);
2093 
2094     if (k->set_config) {
2095         k->set_config(vdev, vdev->config);
2096     }
2097 }
2098 
2099 void virtio_config_writel(VirtIODevice *vdev, uint32_t addr, uint32_t data)
2100 {
2101     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2102     uint32_t val = data;
2103 
2104     if (addr + sizeof(val) > vdev->config_len) {
2105         return;
2106     }
2107 
2108     stl_p(vdev->config + addr, val);
2109 
2110     if (k->set_config) {
2111         k->set_config(vdev, vdev->config);
2112     }
2113 }
2114 
2115 uint32_t virtio_config_modern_readb(VirtIODevice *vdev, uint32_t addr)
2116 {
2117     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2118     uint8_t val;
2119 
2120     if (addr + sizeof(val) > vdev->config_len) {
2121         return (uint32_t)-1;
2122     }
2123 
2124     k->get_config(vdev, vdev->config);
2125 
2126     val = ldub_p(vdev->config + addr);
2127     return val;
2128 }
2129 
2130 uint32_t virtio_config_modern_readw(VirtIODevice *vdev, uint32_t addr)
2131 {
2132     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2133     uint16_t val;
2134 
2135     if (addr + sizeof(val) > vdev->config_len) {
2136         return (uint32_t)-1;
2137     }
2138 
2139     k->get_config(vdev, vdev->config);
2140 
2141     val = lduw_le_p(vdev->config + addr);
2142     return val;
2143 }
2144 
2145 uint32_t virtio_config_modern_readl(VirtIODevice *vdev, uint32_t addr)
2146 {
2147     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2148     uint32_t val;
2149 
2150     if (addr + sizeof(val) > vdev->config_len) {
2151         return (uint32_t)-1;
2152     }
2153 
2154     k->get_config(vdev, vdev->config);
2155 
2156     val = ldl_le_p(vdev->config + addr);
2157     return val;
2158 }
2159 
2160 void virtio_config_modern_writeb(VirtIODevice *vdev,
2161                                  uint32_t addr, uint32_t data)
2162 {
2163     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2164     uint8_t val = data;
2165 
2166     if (addr + sizeof(val) > vdev->config_len) {
2167         return;
2168     }
2169 
2170     stb_p(vdev->config + addr, val);
2171 
2172     if (k->set_config) {
2173         k->set_config(vdev, vdev->config);
2174     }
2175 }
2176 
2177 void virtio_config_modern_writew(VirtIODevice *vdev,
2178                                  uint32_t addr, uint32_t data)
2179 {
2180     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2181     uint16_t val = data;
2182 
2183     if (addr + sizeof(val) > vdev->config_len) {
2184         return;
2185     }
2186 
2187     stw_le_p(vdev->config + addr, val);
2188 
2189     if (k->set_config) {
2190         k->set_config(vdev, vdev->config);
2191     }
2192 }
2193 
2194 void virtio_config_modern_writel(VirtIODevice *vdev,
2195                                  uint32_t addr, uint32_t data)
2196 {
2197     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2198     uint32_t val = data;
2199 
2200     if (addr + sizeof(val) > vdev->config_len) {
2201         return;
2202     }
2203 
2204     stl_le_p(vdev->config + addr, val);
2205 
2206     if (k->set_config) {
2207         k->set_config(vdev, vdev->config);
2208     }
2209 }
2210 
2211 void virtio_queue_set_addr(VirtIODevice *vdev, int n, hwaddr addr)
2212 {
2213     if (!vdev->vq[n].vring.num) {
2214         return;
2215     }
2216     vdev->vq[n].vring.desc = addr;
2217     virtio_queue_update_rings(vdev, n);
2218 }
2219 
2220 hwaddr virtio_queue_get_addr(VirtIODevice *vdev, int n)
2221 {
2222     return vdev->vq[n].vring.desc;
2223 }
2224 
2225 void virtio_queue_set_rings(VirtIODevice *vdev, int n, hwaddr desc,
2226                             hwaddr avail, hwaddr used)
2227 {
2228     if (!vdev->vq[n].vring.num) {
2229         return;
2230     }
2231     vdev->vq[n].vring.desc = desc;
2232     vdev->vq[n].vring.avail = avail;
2233     vdev->vq[n].vring.used = used;
2234     virtio_init_region_cache(vdev, n);
2235 }
2236 
2237 void virtio_queue_set_num(VirtIODevice *vdev, int n, int num)
2238 {
2239     /* Don't allow guest to flip queue between existent and
2240      * nonexistent states, or to set it to an invalid size.
2241      */
2242     if (!!num != !!vdev->vq[n].vring.num ||
2243         num > VIRTQUEUE_MAX_SIZE ||
2244         num < 0) {
2245         return;
2246     }
2247     vdev->vq[n].vring.num = num;
2248 }
2249 
2250 VirtQueue *virtio_vector_first_queue(VirtIODevice *vdev, uint16_t vector)
2251 {
2252     return QLIST_FIRST(&vdev->vector_queues[vector]);
2253 }
2254 
2255 VirtQueue *virtio_vector_next_queue(VirtQueue *vq)
2256 {
2257     return QLIST_NEXT(vq, node);
2258 }
2259 
2260 int virtio_queue_get_num(VirtIODevice *vdev, int n)
2261 {
2262     return vdev->vq[n].vring.num;
2263 }
2264 
2265 int virtio_queue_get_max_num(VirtIODevice *vdev, int n)
2266 {
2267     return vdev->vq[n].vring.num_default;
2268 }
2269 
2270 int virtio_get_num_queues(VirtIODevice *vdev)
2271 {
2272     int i;
2273 
2274     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2275         if (!virtio_queue_get_num(vdev, i)) {
2276             break;
2277         }
2278     }
2279 
2280     return i;
2281 }
2282 
2283 void virtio_queue_set_align(VirtIODevice *vdev, int n, int align)
2284 {
2285     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2286     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
2287 
2288     /* virtio-1 compliant devices cannot change the alignment */
2289     if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
2290         error_report("tried to modify queue alignment for virtio-1 device");
2291         return;
2292     }
2293     /* Check that the transport told us it was going to do this
2294      * (so a buggy transport will immediately assert rather than
2295      * silently failing to migrate this state)
2296      */
2297     assert(k->has_variable_vring_alignment);
2298 
2299     if (align) {
2300         vdev->vq[n].vring.align = align;
2301         virtio_queue_update_rings(vdev, n);
2302     }
2303 }
2304 
2305 static void virtio_queue_notify_vq(VirtQueue *vq)
2306 {
2307     if (vq->vring.desc && vq->handle_output) {
2308         VirtIODevice *vdev = vq->vdev;
2309 
2310         if (unlikely(vdev->broken)) {
2311             return;
2312         }
2313 
2314         trace_virtio_queue_notify(vdev, vq - vdev->vq, vq);
2315         vq->handle_output(vdev, vq);
2316 
2317         if (unlikely(vdev->start_on_kick)) {
2318             virtio_set_started(vdev, true);
2319         }
2320     }
2321 }
2322 
2323 void virtio_queue_notify(VirtIODevice *vdev, int n)
2324 {
2325     VirtQueue *vq = &vdev->vq[n];
2326 
2327     if (unlikely(!vq->vring.desc || vdev->broken)) {
2328         return;
2329     }
2330 
2331     trace_virtio_queue_notify(vdev, vq - vdev->vq, vq);
2332     if (vq->host_notifier_enabled) {
2333         event_notifier_set(&vq->host_notifier);
2334     } else if (vq->handle_output) {
2335         vq->handle_output(vdev, vq);
2336 
2337         if (unlikely(vdev->start_on_kick)) {
2338             virtio_set_started(vdev, true);
2339         }
2340     }
2341 }
2342 
2343 uint16_t virtio_queue_vector(VirtIODevice *vdev, int n)
2344 {
2345     return n < VIRTIO_QUEUE_MAX ? vdev->vq[n].vector :
2346         VIRTIO_NO_VECTOR;
2347 }
2348 
2349 void virtio_queue_set_vector(VirtIODevice *vdev, int n, uint16_t vector)
2350 {
2351     VirtQueue *vq = &vdev->vq[n];
2352 
2353     if (n < VIRTIO_QUEUE_MAX) {
2354         if (vdev->vector_queues &&
2355             vdev->vq[n].vector != VIRTIO_NO_VECTOR) {
2356             QLIST_REMOVE(vq, node);
2357         }
2358         vdev->vq[n].vector = vector;
2359         if (vdev->vector_queues &&
2360             vector != VIRTIO_NO_VECTOR) {
2361             QLIST_INSERT_HEAD(&vdev->vector_queues[vector], vq, node);
2362         }
2363     }
2364 }
2365 
2366 VirtQueue *virtio_add_queue(VirtIODevice *vdev, int queue_size,
2367                             VirtIOHandleOutput handle_output)
2368 {
2369     int i;
2370 
2371     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2372         if (vdev->vq[i].vring.num == 0)
2373             break;
2374     }
2375 
2376     if (i == VIRTIO_QUEUE_MAX || queue_size > VIRTQUEUE_MAX_SIZE)
2377         abort();
2378 
2379     vdev->vq[i].vring.num = queue_size;
2380     vdev->vq[i].vring.num_default = queue_size;
2381     vdev->vq[i].vring.align = VIRTIO_PCI_VRING_ALIGN;
2382     vdev->vq[i].handle_output = handle_output;
2383     vdev->vq[i].used_elems = g_new0(VirtQueueElement, queue_size);
2384 
2385     return &vdev->vq[i];
2386 }
2387 
2388 void virtio_delete_queue(VirtQueue *vq)
2389 {
2390     vq->vring.num = 0;
2391     vq->vring.num_default = 0;
2392     vq->handle_output = NULL;
2393     g_free(vq->used_elems);
2394     vq->used_elems = NULL;
2395     virtio_virtqueue_reset_region_cache(vq);
2396 }
2397 
2398 void virtio_del_queue(VirtIODevice *vdev, int n)
2399 {
2400     if (n < 0 || n >= VIRTIO_QUEUE_MAX) {
2401         abort();
2402     }
2403 
2404     virtio_delete_queue(&vdev->vq[n]);
2405 }
2406 
2407 static void virtio_set_isr(VirtIODevice *vdev, int value)
2408 {
2409     uint8_t old = qatomic_read(&vdev->isr);
2410 
2411     /* Do not write ISR if it does not change, so that its cacheline remains
2412      * shared in the common case where the guest does not read it.
2413      */
2414     if ((old & value) != value) {
2415         qatomic_or(&vdev->isr, value);
2416     }
2417 }
2418 
2419 /* Called within rcu_read_lock(). */
2420 static bool virtio_split_should_notify(VirtIODevice *vdev, VirtQueue *vq)
2421 {
2422     uint16_t old, new;
2423     bool v;
2424     /* We need to expose used array entries before checking used event. */
2425     smp_mb();
2426     /* Always notify when queue is empty (when feature acknowledge) */
2427     if (virtio_vdev_has_feature(vdev, VIRTIO_F_NOTIFY_ON_EMPTY) &&
2428         !vq->inuse && virtio_queue_empty(vq)) {
2429         return true;
2430     }
2431 
2432     if (!virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX)) {
2433         return !(vring_avail_flags(vq) & VRING_AVAIL_F_NO_INTERRUPT);
2434     }
2435 
2436     v = vq->signalled_used_valid;
2437     vq->signalled_used_valid = true;
2438     old = vq->signalled_used;
2439     new = vq->signalled_used = vq->used_idx;
2440     return !v || vring_need_event(vring_get_used_event(vq), new, old);
2441 }
2442 
2443 static bool vring_packed_need_event(VirtQueue *vq, bool wrap,
2444                                     uint16_t off_wrap, uint16_t new,
2445                                     uint16_t old)
2446 {
2447     int off = off_wrap & ~(1 << 15);
2448 
2449     if (wrap != off_wrap >> 15) {
2450         off -= vq->vring.num;
2451     }
2452 
2453     return vring_need_event(off, new, old);
2454 }
2455 
2456 /* Called within rcu_read_lock(). */
2457 static bool virtio_packed_should_notify(VirtIODevice *vdev, VirtQueue *vq)
2458 {
2459     VRingPackedDescEvent e;
2460     uint16_t old, new;
2461     bool v;
2462     VRingMemoryRegionCaches *caches;
2463 
2464     caches = vring_get_region_caches(vq);
2465     if (!caches) {
2466         return false;
2467     }
2468 
2469     vring_packed_event_read(vdev, &caches->avail, &e);
2470 
2471     old = vq->signalled_used;
2472     new = vq->signalled_used = vq->used_idx;
2473     v = vq->signalled_used_valid;
2474     vq->signalled_used_valid = true;
2475 
2476     if (e.flags == VRING_PACKED_EVENT_FLAG_DISABLE) {
2477         return false;
2478     } else if (e.flags == VRING_PACKED_EVENT_FLAG_ENABLE) {
2479         return true;
2480     }
2481 
2482     return !v || vring_packed_need_event(vq, vq->used_wrap_counter,
2483                                          e.off_wrap, new, old);
2484 }
2485 
2486 /* Called within rcu_read_lock().  */
2487 static bool virtio_should_notify(VirtIODevice *vdev, VirtQueue *vq)
2488 {
2489     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
2490         return virtio_packed_should_notify(vdev, vq);
2491     } else {
2492         return virtio_split_should_notify(vdev, vq);
2493     }
2494 }
2495 
2496 void virtio_notify_irqfd(VirtIODevice *vdev, VirtQueue *vq)
2497 {
2498     WITH_RCU_READ_LOCK_GUARD() {
2499         if (!virtio_should_notify(vdev, vq)) {
2500             return;
2501         }
2502     }
2503 
2504     trace_virtio_notify_irqfd(vdev, vq);
2505 
2506     /*
2507      * virtio spec 1.0 says ISR bit 0 should be ignored with MSI, but
2508      * windows drivers included in virtio-win 1.8.0 (circa 2015) are
2509      * incorrectly polling this bit during crashdump and hibernation
2510      * in MSI mode, causing a hang if this bit is never updated.
2511      * Recent releases of Windows do not really shut down, but rather
2512      * log out and hibernate to make the next startup faster.  Hence,
2513      * this manifested as a more serious hang during shutdown with
2514      *
2515      * Next driver release from 2016 fixed this problem, so working around it
2516      * is not a must, but it's easy to do so let's do it here.
2517      *
2518      * Note: it's safe to update ISR from any thread as it was switched
2519      * to an atomic operation.
2520      */
2521     virtio_set_isr(vq->vdev, 0x1);
2522     event_notifier_set(&vq->guest_notifier);
2523 }
2524 
2525 static void virtio_irq(VirtQueue *vq)
2526 {
2527     virtio_set_isr(vq->vdev, 0x1);
2528     virtio_notify_vector(vq->vdev, vq->vector);
2529 }
2530 
2531 void virtio_notify(VirtIODevice *vdev, VirtQueue *vq)
2532 {
2533     WITH_RCU_READ_LOCK_GUARD() {
2534         if (!virtio_should_notify(vdev, vq)) {
2535             return;
2536         }
2537     }
2538 
2539     trace_virtio_notify(vdev, vq);
2540     virtio_irq(vq);
2541 }
2542 
2543 void virtio_notify_config(VirtIODevice *vdev)
2544 {
2545     if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK))
2546         return;
2547 
2548     virtio_set_isr(vdev, 0x3);
2549     vdev->generation++;
2550     virtio_notify_vector(vdev, vdev->config_vector);
2551 }
2552 
2553 static bool virtio_device_endian_needed(void *opaque)
2554 {
2555     VirtIODevice *vdev = opaque;
2556 
2557     assert(vdev->device_endian != VIRTIO_DEVICE_ENDIAN_UNKNOWN);
2558     if (!virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
2559         return vdev->device_endian != virtio_default_endian();
2560     }
2561     /* Devices conforming to VIRTIO 1.0 or later are always LE. */
2562     return vdev->device_endian != VIRTIO_DEVICE_ENDIAN_LITTLE;
2563 }
2564 
2565 static bool virtio_64bit_features_needed(void *opaque)
2566 {
2567     VirtIODevice *vdev = opaque;
2568 
2569     return (vdev->host_features >> 32) != 0;
2570 }
2571 
2572 static bool virtio_virtqueue_needed(void *opaque)
2573 {
2574     VirtIODevice *vdev = opaque;
2575 
2576     return virtio_host_has_feature(vdev, VIRTIO_F_VERSION_1);
2577 }
2578 
2579 static bool virtio_packed_virtqueue_needed(void *opaque)
2580 {
2581     VirtIODevice *vdev = opaque;
2582 
2583     return virtio_host_has_feature(vdev, VIRTIO_F_RING_PACKED);
2584 }
2585 
2586 static bool virtio_ringsize_needed(void *opaque)
2587 {
2588     VirtIODevice *vdev = opaque;
2589     int i;
2590 
2591     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2592         if (vdev->vq[i].vring.num != vdev->vq[i].vring.num_default) {
2593             return true;
2594         }
2595     }
2596     return false;
2597 }
2598 
2599 static bool virtio_extra_state_needed(void *opaque)
2600 {
2601     VirtIODevice *vdev = opaque;
2602     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2603     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
2604 
2605     return k->has_extra_state &&
2606         k->has_extra_state(qbus->parent);
2607 }
2608 
2609 static bool virtio_broken_needed(void *opaque)
2610 {
2611     VirtIODevice *vdev = opaque;
2612 
2613     return vdev->broken;
2614 }
2615 
2616 static bool virtio_started_needed(void *opaque)
2617 {
2618     VirtIODevice *vdev = opaque;
2619 
2620     return vdev->started;
2621 }
2622 
2623 static bool virtio_disabled_needed(void *opaque)
2624 {
2625     VirtIODevice *vdev = opaque;
2626 
2627     return vdev->disabled;
2628 }
2629 
2630 static const VMStateDescription vmstate_virtqueue = {
2631     .name = "virtqueue_state",
2632     .version_id = 1,
2633     .minimum_version_id = 1,
2634     .fields = (VMStateField[]) {
2635         VMSTATE_UINT64(vring.avail, struct VirtQueue),
2636         VMSTATE_UINT64(vring.used, struct VirtQueue),
2637         VMSTATE_END_OF_LIST()
2638     }
2639 };
2640 
2641 static const VMStateDescription vmstate_packed_virtqueue = {
2642     .name = "packed_virtqueue_state",
2643     .version_id = 1,
2644     .minimum_version_id = 1,
2645     .fields = (VMStateField[]) {
2646         VMSTATE_UINT16(last_avail_idx, struct VirtQueue),
2647         VMSTATE_BOOL(last_avail_wrap_counter, struct VirtQueue),
2648         VMSTATE_UINT16(used_idx, struct VirtQueue),
2649         VMSTATE_BOOL(used_wrap_counter, struct VirtQueue),
2650         VMSTATE_UINT32(inuse, struct VirtQueue),
2651         VMSTATE_END_OF_LIST()
2652     }
2653 };
2654 
2655 static const VMStateDescription vmstate_virtio_virtqueues = {
2656     .name = "virtio/virtqueues",
2657     .version_id = 1,
2658     .minimum_version_id = 1,
2659     .needed = &virtio_virtqueue_needed,
2660     .fields = (VMStateField[]) {
2661         VMSTATE_STRUCT_VARRAY_POINTER_KNOWN(vq, struct VirtIODevice,
2662                       VIRTIO_QUEUE_MAX, 0, vmstate_virtqueue, VirtQueue),
2663         VMSTATE_END_OF_LIST()
2664     }
2665 };
2666 
2667 static const VMStateDescription vmstate_virtio_packed_virtqueues = {
2668     .name = "virtio/packed_virtqueues",
2669     .version_id = 1,
2670     .minimum_version_id = 1,
2671     .needed = &virtio_packed_virtqueue_needed,
2672     .fields = (VMStateField[]) {
2673         VMSTATE_STRUCT_VARRAY_POINTER_KNOWN(vq, struct VirtIODevice,
2674                       VIRTIO_QUEUE_MAX, 0, vmstate_packed_virtqueue, VirtQueue),
2675         VMSTATE_END_OF_LIST()
2676     }
2677 };
2678 
2679 static const VMStateDescription vmstate_ringsize = {
2680     .name = "ringsize_state",
2681     .version_id = 1,
2682     .minimum_version_id = 1,
2683     .fields = (VMStateField[]) {
2684         VMSTATE_UINT32(vring.num_default, struct VirtQueue),
2685         VMSTATE_END_OF_LIST()
2686     }
2687 };
2688 
2689 static const VMStateDescription vmstate_virtio_ringsize = {
2690     .name = "virtio/ringsize",
2691     .version_id = 1,
2692     .minimum_version_id = 1,
2693     .needed = &virtio_ringsize_needed,
2694     .fields = (VMStateField[]) {
2695         VMSTATE_STRUCT_VARRAY_POINTER_KNOWN(vq, struct VirtIODevice,
2696                       VIRTIO_QUEUE_MAX, 0, vmstate_ringsize, VirtQueue),
2697         VMSTATE_END_OF_LIST()
2698     }
2699 };
2700 
2701 static int get_extra_state(QEMUFile *f, void *pv, size_t size,
2702                            const VMStateField *field)
2703 {
2704     VirtIODevice *vdev = pv;
2705     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2706     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
2707 
2708     if (!k->load_extra_state) {
2709         return -1;
2710     } else {
2711         return k->load_extra_state(qbus->parent, f);
2712     }
2713 }
2714 
2715 static int put_extra_state(QEMUFile *f, void *pv, size_t size,
2716                            const VMStateField *field, JSONWriter *vmdesc)
2717 {
2718     VirtIODevice *vdev = pv;
2719     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2720     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
2721 
2722     k->save_extra_state(qbus->parent, f);
2723     return 0;
2724 }
2725 
2726 static const VMStateInfo vmstate_info_extra_state = {
2727     .name = "virtqueue_extra_state",
2728     .get = get_extra_state,
2729     .put = put_extra_state,
2730 };
2731 
2732 static const VMStateDescription vmstate_virtio_extra_state = {
2733     .name = "virtio/extra_state",
2734     .version_id = 1,
2735     .minimum_version_id = 1,
2736     .needed = &virtio_extra_state_needed,
2737     .fields = (VMStateField[]) {
2738         {
2739             .name         = "extra_state",
2740             .version_id   = 0,
2741             .field_exists = NULL,
2742             .size         = 0,
2743             .info         = &vmstate_info_extra_state,
2744             .flags        = VMS_SINGLE,
2745             .offset       = 0,
2746         },
2747         VMSTATE_END_OF_LIST()
2748     }
2749 };
2750 
2751 static const VMStateDescription vmstate_virtio_device_endian = {
2752     .name = "virtio/device_endian",
2753     .version_id = 1,
2754     .minimum_version_id = 1,
2755     .needed = &virtio_device_endian_needed,
2756     .fields = (VMStateField[]) {
2757         VMSTATE_UINT8(device_endian, VirtIODevice),
2758         VMSTATE_END_OF_LIST()
2759     }
2760 };
2761 
2762 static const VMStateDescription vmstate_virtio_64bit_features = {
2763     .name = "virtio/64bit_features",
2764     .version_id = 1,
2765     .minimum_version_id = 1,
2766     .needed = &virtio_64bit_features_needed,
2767     .fields = (VMStateField[]) {
2768         VMSTATE_UINT64(guest_features, VirtIODevice),
2769         VMSTATE_END_OF_LIST()
2770     }
2771 };
2772 
2773 static const VMStateDescription vmstate_virtio_broken = {
2774     .name = "virtio/broken",
2775     .version_id = 1,
2776     .minimum_version_id = 1,
2777     .needed = &virtio_broken_needed,
2778     .fields = (VMStateField[]) {
2779         VMSTATE_BOOL(broken, VirtIODevice),
2780         VMSTATE_END_OF_LIST()
2781     }
2782 };
2783 
2784 static const VMStateDescription vmstate_virtio_started = {
2785     .name = "virtio/started",
2786     .version_id = 1,
2787     .minimum_version_id = 1,
2788     .needed = &virtio_started_needed,
2789     .fields = (VMStateField[]) {
2790         VMSTATE_BOOL(started, VirtIODevice),
2791         VMSTATE_END_OF_LIST()
2792     }
2793 };
2794 
2795 static const VMStateDescription vmstate_virtio_disabled = {
2796     .name = "virtio/disabled",
2797     .version_id = 1,
2798     .minimum_version_id = 1,
2799     .needed = &virtio_disabled_needed,
2800     .fields = (VMStateField[]) {
2801         VMSTATE_BOOL(disabled, VirtIODevice),
2802         VMSTATE_END_OF_LIST()
2803     }
2804 };
2805 
2806 static const VMStateDescription vmstate_virtio = {
2807     .name = "virtio",
2808     .version_id = 1,
2809     .minimum_version_id = 1,
2810     .fields = (VMStateField[]) {
2811         VMSTATE_END_OF_LIST()
2812     },
2813     .subsections = (const VMStateDescription*[]) {
2814         &vmstate_virtio_device_endian,
2815         &vmstate_virtio_64bit_features,
2816         &vmstate_virtio_virtqueues,
2817         &vmstate_virtio_ringsize,
2818         &vmstate_virtio_broken,
2819         &vmstate_virtio_extra_state,
2820         &vmstate_virtio_started,
2821         &vmstate_virtio_packed_virtqueues,
2822         &vmstate_virtio_disabled,
2823         NULL
2824     }
2825 };
2826 
2827 int virtio_save(VirtIODevice *vdev, QEMUFile *f)
2828 {
2829     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2830     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
2831     VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev);
2832     uint32_t guest_features_lo = (vdev->guest_features & 0xffffffff);
2833     int i;
2834 
2835     if (k->save_config) {
2836         k->save_config(qbus->parent, f);
2837     }
2838 
2839     qemu_put_8s(f, &vdev->status);
2840     qemu_put_8s(f, &vdev->isr);
2841     qemu_put_be16s(f, &vdev->queue_sel);
2842     qemu_put_be32s(f, &guest_features_lo);
2843     qemu_put_be32(f, vdev->config_len);
2844     qemu_put_buffer(f, vdev->config, vdev->config_len);
2845 
2846     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2847         if (vdev->vq[i].vring.num == 0)
2848             break;
2849     }
2850 
2851     qemu_put_be32(f, i);
2852 
2853     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2854         if (vdev->vq[i].vring.num == 0)
2855             break;
2856 
2857         qemu_put_be32(f, vdev->vq[i].vring.num);
2858         if (k->has_variable_vring_alignment) {
2859             qemu_put_be32(f, vdev->vq[i].vring.align);
2860         }
2861         /*
2862          * Save desc now, the rest of the ring addresses are saved in
2863          * subsections for VIRTIO-1 devices.
2864          */
2865         qemu_put_be64(f, vdev->vq[i].vring.desc);
2866         qemu_put_be16s(f, &vdev->vq[i].last_avail_idx);
2867         if (k->save_queue) {
2868             k->save_queue(qbus->parent, i, f);
2869         }
2870     }
2871 
2872     if (vdc->save != NULL) {
2873         vdc->save(vdev, f);
2874     }
2875 
2876     if (vdc->vmsd) {
2877         int ret = vmstate_save_state(f, vdc->vmsd, vdev, NULL);
2878         if (ret) {
2879             return ret;
2880         }
2881     }
2882 
2883     /* Subsections */
2884     return vmstate_save_state(f, &vmstate_virtio, vdev, NULL);
2885 }
2886 
2887 /* A wrapper for use as a VMState .put function */
2888 static int virtio_device_put(QEMUFile *f, void *opaque, size_t size,
2889                               const VMStateField *field, JSONWriter *vmdesc)
2890 {
2891     return virtio_save(VIRTIO_DEVICE(opaque), f);
2892 }
2893 
2894 /* A wrapper for use as a VMState .get function */
2895 static int virtio_device_get(QEMUFile *f, void *opaque, size_t size,
2896                              const VMStateField *field)
2897 {
2898     VirtIODevice *vdev = VIRTIO_DEVICE(opaque);
2899     DeviceClass *dc = DEVICE_CLASS(VIRTIO_DEVICE_GET_CLASS(vdev));
2900 
2901     return virtio_load(vdev, f, dc->vmsd->version_id);
2902 }
2903 
2904 const VMStateInfo  virtio_vmstate_info = {
2905     .name = "virtio",
2906     .get = virtio_device_get,
2907     .put = virtio_device_put,
2908 };
2909 
2910 static int virtio_set_features_nocheck(VirtIODevice *vdev, uint64_t val)
2911 {
2912     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2913     bool bad = (val & ~(vdev->host_features)) != 0;
2914 
2915     val &= vdev->host_features;
2916     if (k->set_features) {
2917         k->set_features(vdev, val);
2918     }
2919     vdev->guest_features = val;
2920     return bad ? -1 : 0;
2921 }
2922 
2923 int virtio_set_features(VirtIODevice *vdev, uint64_t val)
2924 {
2925     int ret;
2926     /*
2927      * The driver must not attempt to set features after feature negotiation
2928      * has finished.
2929      */
2930     if (vdev->status & VIRTIO_CONFIG_S_FEATURES_OK) {
2931         return -EINVAL;
2932     }
2933     ret = virtio_set_features_nocheck(vdev, val);
2934     if (virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX)) {
2935         /* VIRTIO_RING_F_EVENT_IDX changes the size of the caches.  */
2936         int i;
2937         for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2938             if (vdev->vq[i].vring.num != 0) {
2939                 virtio_init_region_cache(vdev, i);
2940             }
2941         }
2942     }
2943     if (!ret) {
2944         if (!virtio_device_started(vdev, vdev->status) &&
2945             !virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
2946             vdev->start_on_kick = true;
2947         }
2948     }
2949     return ret;
2950 }
2951 
2952 size_t virtio_feature_get_config_size(const VirtIOFeature *feature_sizes,
2953                                       uint64_t host_features)
2954 {
2955     size_t config_size = 0;
2956     int i;
2957 
2958     for (i = 0; feature_sizes[i].flags != 0; i++) {
2959         if (host_features & feature_sizes[i].flags) {
2960             config_size = MAX(feature_sizes[i].end, config_size);
2961         }
2962     }
2963 
2964     return config_size;
2965 }
2966 
2967 int virtio_load(VirtIODevice *vdev, QEMUFile *f, int version_id)
2968 {
2969     int i, ret;
2970     int32_t config_len;
2971     uint32_t num;
2972     uint32_t features;
2973     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2974     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
2975     VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev);
2976 
2977     /*
2978      * We poison the endianness to ensure it does not get used before
2979      * subsections have been loaded.
2980      */
2981     vdev->device_endian = VIRTIO_DEVICE_ENDIAN_UNKNOWN;
2982 
2983     if (k->load_config) {
2984         ret = k->load_config(qbus->parent, f);
2985         if (ret)
2986             return ret;
2987     }
2988 
2989     qemu_get_8s(f, &vdev->status);
2990     qemu_get_8s(f, &vdev->isr);
2991     qemu_get_be16s(f, &vdev->queue_sel);
2992     if (vdev->queue_sel >= VIRTIO_QUEUE_MAX) {
2993         return -1;
2994     }
2995     qemu_get_be32s(f, &features);
2996 
2997     /*
2998      * Temporarily set guest_features low bits - needed by
2999      * virtio net load code testing for VIRTIO_NET_F_CTRL_GUEST_OFFLOADS
3000      * VIRTIO_NET_F_GUEST_ANNOUNCE and VIRTIO_NET_F_CTRL_VQ.
3001      *
3002      * Note: devices should always test host features in future - don't create
3003      * new dependencies like this.
3004      */
3005     vdev->guest_features = features;
3006 
3007     config_len = qemu_get_be32(f);
3008 
3009     /*
3010      * There are cases where the incoming config can be bigger or smaller
3011      * than what we have; so load what we have space for, and skip
3012      * any excess that's in the stream.
3013      */
3014     qemu_get_buffer(f, vdev->config, MIN(config_len, vdev->config_len));
3015 
3016     while (config_len > vdev->config_len) {
3017         qemu_get_byte(f);
3018         config_len--;
3019     }
3020 
3021     num = qemu_get_be32(f);
3022 
3023     if (num > VIRTIO_QUEUE_MAX) {
3024         error_report("Invalid number of virtqueues: 0x%x", num);
3025         return -1;
3026     }
3027 
3028     for (i = 0; i < num; i++) {
3029         vdev->vq[i].vring.num = qemu_get_be32(f);
3030         if (k->has_variable_vring_alignment) {
3031             vdev->vq[i].vring.align = qemu_get_be32(f);
3032         }
3033         vdev->vq[i].vring.desc = qemu_get_be64(f);
3034         qemu_get_be16s(f, &vdev->vq[i].last_avail_idx);
3035         vdev->vq[i].signalled_used_valid = false;
3036         vdev->vq[i].notification = true;
3037 
3038         if (!vdev->vq[i].vring.desc && vdev->vq[i].last_avail_idx) {
3039             error_report("VQ %d address 0x0 "
3040                          "inconsistent with Host index 0x%x",
3041                          i, vdev->vq[i].last_avail_idx);
3042             return -1;
3043         }
3044         if (k->load_queue) {
3045             ret = k->load_queue(qbus->parent, i, f);
3046             if (ret)
3047                 return ret;
3048         }
3049     }
3050 
3051     virtio_notify_vector(vdev, VIRTIO_NO_VECTOR);
3052 
3053     if (vdc->load != NULL) {
3054         ret = vdc->load(vdev, f, version_id);
3055         if (ret) {
3056             return ret;
3057         }
3058     }
3059 
3060     if (vdc->vmsd) {
3061         ret = vmstate_load_state(f, vdc->vmsd, vdev, version_id);
3062         if (ret) {
3063             return ret;
3064         }
3065     }
3066 
3067     /* Subsections */
3068     ret = vmstate_load_state(f, &vmstate_virtio, vdev, 1);
3069     if (ret) {
3070         return ret;
3071     }
3072 
3073     if (vdev->device_endian == VIRTIO_DEVICE_ENDIAN_UNKNOWN) {
3074         vdev->device_endian = virtio_default_endian();
3075     }
3076 
3077     if (virtio_64bit_features_needed(vdev)) {
3078         /*
3079          * Subsection load filled vdev->guest_features.  Run them
3080          * through virtio_set_features to sanity-check them against
3081          * host_features.
3082          */
3083         uint64_t features64 = vdev->guest_features;
3084         if (virtio_set_features_nocheck(vdev, features64) < 0) {
3085             error_report("Features 0x%" PRIx64 " unsupported. "
3086                          "Allowed features: 0x%" PRIx64,
3087                          features64, vdev->host_features);
3088             return -1;
3089         }
3090     } else {
3091         if (virtio_set_features_nocheck(vdev, features) < 0) {
3092             error_report("Features 0x%x unsupported. "
3093                          "Allowed features: 0x%" PRIx64,
3094                          features, vdev->host_features);
3095             return -1;
3096         }
3097     }
3098 
3099     if (!virtio_device_started(vdev, vdev->status) &&
3100         !virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
3101         vdev->start_on_kick = true;
3102     }
3103 
3104     RCU_READ_LOCK_GUARD();
3105     for (i = 0; i < num; i++) {
3106         if (vdev->vq[i].vring.desc) {
3107             uint16_t nheads;
3108 
3109             /*
3110              * VIRTIO-1 devices migrate desc, used, and avail ring addresses so
3111              * only the region cache needs to be set up.  Legacy devices need
3112              * to calculate used and avail ring addresses based on the desc
3113              * address.
3114              */
3115             if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
3116                 virtio_init_region_cache(vdev, i);
3117             } else {
3118                 virtio_queue_update_rings(vdev, i);
3119             }
3120 
3121             if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3122                 vdev->vq[i].shadow_avail_idx = vdev->vq[i].last_avail_idx;
3123                 vdev->vq[i].shadow_avail_wrap_counter =
3124                                         vdev->vq[i].last_avail_wrap_counter;
3125                 continue;
3126             }
3127 
3128             nheads = vring_avail_idx(&vdev->vq[i]) - vdev->vq[i].last_avail_idx;
3129             /* Check it isn't doing strange things with descriptor numbers. */
3130             if (nheads > vdev->vq[i].vring.num) {
3131                 virtio_error(vdev, "VQ %d size 0x%x Guest index 0x%x "
3132                              "inconsistent with Host index 0x%x: delta 0x%x",
3133                              i, vdev->vq[i].vring.num,
3134                              vring_avail_idx(&vdev->vq[i]),
3135                              vdev->vq[i].last_avail_idx, nheads);
3136                 vdev->vq[i].used_idx = 0;
3137                 vdev->vq[i].shadow_avail_idx = 0;
3138                 vdev->vq[i].inuse = 0;
3139                 continue;
3140             }
3141             vdev->vq[i].used_idx = vring_used_idx(&vdev->vq[i]);
3142             vdev->vq[i].shadow_avail_idx = vring_avail_idx(&vdev->vq[i]);
3143 
3144             /*
3145              * Some devices migrate VirtQueueElements that have been popped
3146              * from the avail ring but not yet returned to the used ring.
3147              * Since max ring size < UINT16_MAX it's safe to use modulo
3148              * UINT16_MAX + 1 subtraction.
3149              */
3150             vdev->vq[i].inuse = (uint16_t)(vdev->vq[i].last_avail_idx -
3151                                 vdev->vq[i].used_idx);
3152             if (vdev->vq[i].inuse > vdev->vq[i].vring.num) {
3153                 error_report("VQ %d size 0x%x < last_avail_idx 0x%x - "
3154                              "used_idx 0x%x",
3155                              i, vdev->vq[i].vring.num,
3156                              vdev->vq[i].last_avail_idx,
3157                              vdev->vq[i].used_idx);
3158                 return -1;
3159             }
3160         }
3161     }
3162 
3163     if (vdc->post_load) {
3164         ret = vdc->post_load(vdev);
3165         if (ret) {
3166             return ret;
3167         }
3168     }
3169 
3170     return 0;
3171 }
3172 
3173 void virtio_cleanup(VirtIODevice *vdev)
3174 {
3175     qemu_del_vm_change_state_handler(vdev->vmstate);
3176 }
3177 
3178 static void virtio_vmstate_change(void *opaque, bool running, RunState state)
3179 {
3180     VirtIODevice *vdev = opaque;
3181     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3182     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
3183     bool backend_run = running && virtio_device_started(vdev, vdev->status);
3184     vdev->vm_running = running;
3185 
3186     if (backend_run) {
3187         virtio_set_status(vdev, vdev->status);
3188     }
3189 
3190     if (k->vmstate_change) {
3191         k->vmstate_change(qbus->parent, backend_run);
3192     }
3193 
3194     if (!backend_run) {
3195         virtio_set_status(vdev, vdev->status);
3196     }
3197 }
3198 
3199 void virtio_instance_init_common(Object *proxy_obj, void *data,
3200                                  size_t vdev_size, const char *vdev_name)
3201 {
3202     DeviceState *vdev = data;
3203 
3204     object_initialize_child_with_props(proxy_obj, "virtio-backend", vdev,
3205                                        vdev_size, vdev_name, &error_abort,
3206                                        NULL);
3207     qdev_alias_all_properties(vdev, proxy_obj);
3208 }
3209 
3210 void virtio_init(VirtIODevice *vdev, const char *name,
3211                  uint16_t device_id, size_t config_size)
3212 {
3213     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3214     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
3215     int i;
3216     int nvectors = k->query_nvectors ? k->query_nvectors(qbus->parent) : 0;
3217 
3218     if (nvectors) {
3219         vdev->vector_queues =
3220             g_malloc0(sizeof(*vdev->vector_queues) * nvectors);
3221     }
3222 
3223     vdev->start_on_kick = false;
3224     vdev->started = false;
3225     vdev->device_id = device_id;
3226     vdev->status = 0;
3227     qatomic_set(&vdev->isr, 0);
3228     vdev->queue_sel = 0;
3229     vdev->config_vector = VIRTIO_NO_VECTOR;
3230     vdev->vq = g_new0(VirtQueue, VIRTIO_QUEUE_MAX);
3231     vdev->vm_running = runstate_is_running();
3232     vdev->broken = false;
3233     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
3234         vdev->vq[i].vector = VIRTIO_NO_VECTOR;
3235         vdev->vq[i].vdev = vdev;
3236         vdev->vq[i].queue_index = i;
3237         vdev->vq[i].host_notifier_enabled = false;
3238     }
3239 
3240     vdev->name = name;
3241     vdev->config_len = config_size;
3242     if (vdev->config_len) {
3243         vdev->config = g_malloc0(config_size);
3244     } else {
3245         vdev->config = NULL;
3246     }
3247     vdev->vmstate = qdev_add_vm_change_state_handler(DEVICE(vdev),
3248             virtio_vmstate_change, vdev);
3249     vdev->device_endian = virtio_default_endian();
3250     vdev->use_guest_notifier_mask = true;
3251 }
3252 
3253 /*
3254  * Only devices that have already been around prior to defining the virtio
3255  * standard support legacy mode; this includes devices not specified in the
3256  * standard. All newer devices conform to the virtio standard only.
3257  */
3258 bool virtio_legacy_allowed(VirtIODevice *vdev)
3259 {
3260     switch (vdev->device_id) {
3261     case VIRTIO_ID_NET:
3262     case VIRTIO_ID_BLOCK:
3263     case VIRTIO_ID_CONSOLE:
3264     case VIRTIO_ID_RNG:
3265     case VIRTIO_ID_BALLOON:
3266     case VIRTIO_ID_RPMSG:
3267     case VIRTIO_ID_SCSI:
3268     case VIRTIO_ID_9P:
3269     case VIRTIO_ID_RPROC_SERIAL:
3270     case VIRTIO_ID_CAIF:
3271         return true;
3272     default:
3273         return false;
3274     }
3275 }
3276 
3277 bool virtio_legacy_check_disabled(VirtIODevice *vdev)
3278 {
3279     return vdev->disable_legacy_check;
3280 }
3281 
3282 hwaddr virtio_queue_get_desc_addr(VirtIODevice *vdev, int n)
3283 {
3284     return vdev->vq[n].vring.desc;
3285 }
3286 
3287 bool virtio_queue_enabled_legacy(VirtIODevice *vdev, int n)
3288 {
3289     return virtio_queue_get_desc_addr(vdev, n) != 0;
3290 }
3291 
3292 bool virtio_queue_enabled(VirtIODevice *vdev, int n)
3293 {
3294     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3295     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
3296 
3297     if (k->queue_enabled) {
3298         return k->queue_enabled(qbus->parent, n);
3299     }
3300     return virtio_queue_enabled_legacy(vdev, n);
3301 }
3302 
3303 hwaddr virtio_queue_get_avail_addr(VirtIODevice *vdev, int n)
3304 {
3305     return vdev->vq[n].vring.avail;
3306 }
3307 
3308 hwaddr virtio_queue_get_used_addr(VirtIODevice *vdev, int n)
3309 {
3310     return vdev->vq[n].vring.used;
3311 }
3312 
3313 hwaddr virtio_queue_get_desc_size(VirtIODevice *vdev, int n)
3314 {
3315     return sizeof(VRingDesc) * vdev->vq[n].vring.num;
3316 }
3317 
3318 hwaddr virtio_queue_get_avail_size(VirtIODevice *vdev, int n)
3319 {
3320     int s;
3321 
3322     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3323         return sizeof(struct VRingPackedDescEvent);
3324     }
3325 
3326     s = virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0;
3327     return offsetof(VRingAvail, ring) +
3328         sizeof(uint16_t) * vdev->vq[n].vring.num + s;
3329 }
3330 
3331 hwaddr virtio_queue_get_used_size(VirtIODevice *vdev, int n)
3332 {
3333     int s;
3334 
3335     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3336         return sizeof(struct VRingPackedDescEvent);
3337     }
3338 
3339     s = virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0;
3340     return offsetof(VRingUsed, ring) +
3341         sizeof(VRingUsedElem) * vdev->vq[n].vring.num + s;
3342 }
3343 
3344 static unsigned int virtio_queue_packed_get_last_avail_idx(VirtIODevice *vdev,
3345                                                            int n)
3346 {
3347     unsigned int avail, used;
3348 
3349     avail = vdev->vq[n].last_avail_idx;
3350     avail |= ((uint16_t)vdev->vq[n].last_avail_wrap_counter) << 15;
3351 
3352     used = vdev->vq[n].used_idx;
3353     used |= ((uint16_t)vdev->vq[n].used_wrap_counter) << 15;
3354 
3355     return avail | used << 16;
3356 }
3357 
3358 static uint16_t virtio_queue_split_get_last_avail_idx(VirtIODevice *vdev,
3359                                                       int n)
3360 {
3361     return vdev->vq[n].last_avail_idx;
3362 }
3363 
3364 unsigned int virtio_queue_get_last_avail_idx(VirtIODevice *vdev, int n)
3365 {
3366     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3367         return virtio_queue_packed_get_last_avail_idx(vdev, n);
3368     } else {
3369         return virtio_queue_split_get_last_avail_idx(vdev, n);
3370     }
3371 }
3372 
3373 static void virtio_queue_packed_set_last_avail_idx(VirtIODevice *vdev,
3374                                                    int n, unsigned int idx)
3375 {
3376     struct VirtQueue *vq = &vdev->vq[n];
3377 
3378     vq->last_avail_idx = vq->shadow_avail_idx = idx & 0x7fff;
3379     vq->last_avail_wrap_counter =
3380         vq->shadow_avail_wrap_counter = !!(idx & 0x8000);
3381     idx >>= 16;
3382     vq->used_idx = idx & 0x7ffff;
3383     vq->used_wrap_counter = !!(idx & 0x8000);
3384 }
3385 
3386 static void virtio_queue_split_set_last_avail_idx(VirtIODevice *vdev,
3387                                                   int n, unsigned int idx)
3388 {
3389         vdev->vq[n].last_avail_idx = idx;
3390         vdev->vq[n].shadow_avail_idx = idx;
3391 }
3392 
3393 void virtio_queue_set_last_avail_idx(VirtIODevice *vdev, int n,
3394                                      unsigned int idx)
3395 {
3396     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3397         virtio_queue_packed_set_last_avail_idx(vdev, n, idx);
3398     } else {
3399         virtio_queue_split_set_last_avail_idx(vdev, n, idx);
3400     }
3401 }
3402 
3403 static void virtio_queue_packed_restore_last_avail_idx(VirtIODevice *vdev,
3404                                                        int n)
3405 {
3406     /* We don't have a reference like avail idx in shared memory */
3407     return;
3408 }
3409 
3410 static void virtio_queue_split_restore_last_avail_idx(VirtIODevice *vdev,
3411                                                       int n)
3412 {
3413     RCU_READ_LOCK_GUARD();
3414     if (vdev->vq[n].vring.desc) {
3415         vdev->vq[n].last_avail_idx = vring_used_idx(&vdev->vq[n]);
3416         vdev->vq[n].shadow_avail_idx = vdev->vq[n].last_avail_idx;
3417     }
3418 }
3419 
3420 void virtio_queue_restore_last_avail_idx(VirtIODevice *vdev, int n)
3421 {
3422     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3423         virtio_queue_packed_restore_last_avail_idx(vdev, n);
3424     } else {
3425         virtio_queue_split_restore_last_avail_idx(vdev, n);
3426     }
3427 }
3428 
3429 static void virtio_queue_packed_update_used_idx(VirtIODevice *vdev, int n)
3430 {
3431     /* used idx was updated through set_last_avail_idx() */
3432     return;
3433 }
3434 
3435 static void virtio_split_packed_update_used_idx(VirtIODevice *vdev, int n)
3436 {
3437     RCU_READ_LOCK_GUARD();
3438     if (vdev->vq[n].vring.desc) {
3439         vdev->vq[n].used_idx = vring_used_idx(&vdev->vq[n]);
3440     }
3441 }
3442 
3443 void virtio_queue_update_used_idx(VirtIODevice *vdev, int n)
3444 {
3445     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3446         return virtio_queue_packed_update_used_idx(vdev, n);
3447     } else {
3448         return virtio_split_packed_update_used_idx(vdev, n);
3449     }
3450 }
3451 
3452 void virtio_queue_invalidate_signalled_used(VirtIODevice *vdev, int n)
3453 {
3454     vdev->vq[n].signalled_used_valid = false;
3455 }
3456 
3457 VirtQueue *virtio_get_queue(VirtIODevice *vdev, int n)
3458 {
3459     return vdev->vq + n;
3460 }
3461 
3462 uint16_t virtio_get_queue_index(VirtQueue *vq)
3463 {
3464     return vq->queue_index;
3465 }
3466 
3467 static void virtio_queue_guest_notifier_read(EventNotifier *n)
3468 {
3469     VirtQueue *vq = container_of(n, VirtQueue, guest_notifier);
3470     if (event_notifier_test_and_clear(n)) {
3471         virtio_irq(vq);
3472     }
3473 }
3474 
3475 void virtio_queue_set_guest_notifier_fd_handler(VirtQueue *vq, bool assign,
3476                                                 bool with_irqfd)
3477 {
3478     if (assign && !with_irqfd) {
3479         event_notifier_set_handler(&vq->guest_notifier,
3480                                    virtio_queue_guest_notifier_read);
3481     } else {
3482         event_notifier_set_handler(&vq->guest_notifier, NULL);
3483     }
3484     if (!assign) {
3485         /* Test and clear notifier before closing it,
3486          * in case poll callback didn't have time to run. */
3487         virtio_queue_guest_notifier_read(&vq->guest_notifier);
3488     }
3489 }
3490 
3491 EventNotifier *virtio_queue_get_guest_notifier(VirtQueue *vq)
3492 {
3493     return &vq->guest_notifier;
3494 }
3495 
3496 static void virtio_queue_host_notifier_aio_poll_begin(EventNotifier *n)
3497 {
3498     VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
3499 
3500     virtio_queue_set_notification(vq, 0);
3501 }
3502 
3503 static bool virtio_queue_host_notifier_aio_poll(void *opaque)
3504 {
3505     EventNotifier *n = opaque;
3506     VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
3507 
3508     return vq->vring.desc && !virtio_queue_empty(vq);
3509 }
3510 
3511 static void virtio_queue_host_notifier_aio_poll_ready(EventNotifier *n)
3512 {
3513     VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
3514 
3515     virtio_queue_notify_vq(vq);
3516 }
3517 
3518 static void virtio_queue_host_notifier_aio_poll_end(EventNotifier *n)
3519 {
3520     VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
3521 
3522     /* Caller polls once more after this to catch requests that race with us */
3523     virtio_queue_set_notification(vq, 1);
3524 }
3525 
3526 void virtio_queue_aio_attach_host_notifier(VirtQueue *vq, AioContext *ctx)
3527 {
3528     aio_set_event_notifier(ctx, &vq->host_notifier, true,
3529                            virtio_queue_host_notifier_read,
3530                            virtio_queue_host_notifier_aio_poll,
3531                            virtio_queue_host_notifier_aio_poll_ready);
3532     aio_set_event_notifier_poll(ctx, &vq->host_notifier,
3533                                 virtio_queue_host_notifier_aio_poll_begin,
3534                                 virtio_queue_host_notifier_aio_poll_end);
3535 }
3536 
3537 void virtio_queue_aio_detach_host_notifier(VirtQueue *vq, AioContext *ctx)
3538 {
3539     aio_set_event_notifier(ctx, &vq->host_notifier, true, NULL, NULL, NULL);
3540     /* Test and clear notifier before after disabling event,
3541      * in case poll callback didn't have time to run. */
3542     virtio_queue_host_notifier_read(&vq->host_notifier);
3543 }
3544 
3545 void virtio_queue_host_notifier_read(EventNotifier *n)
3546 {
3547     VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
3548     if (event_notifier_test_and_clear(n)) {
3549         virtio_queue_notify_vq(vq);
3550     }
3551 }
3552 
3553 EventNotifier *virtio_queue_get_host_notifier(VirtQueue *vq)
3554 {
3555     return &vq->host_notifier;
3556 }
3557 
3558 void virtio_queue_set_host_notifier_enabled(VirtQueue *vq, bool enabled)
3559 {
3560     vq->host_notifier_enabled = enabled;
3561 }
3562 
3563 int virtio_queue_set_host_notifier_mr(VirtIODevice *vdev, int n,
3564                                       MemoryRegion *mr, bool assign)
3565 {
3566     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3567     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
3568 
3569     if (k->set_host_notifier_mr) {
3570         return k->set_host_notifier_mr(qbus->parent, n, mr, assign);
3571     }
3572 
3573     return -1;
3574 }
3575 
3576 void virtio_device_set_child_bus_name(VirtIODevice *vdev, char *bus_name)
3577 {
3578     g_free(vdev->bus_name);
3579     vdev->bus_name = g_strdup(bus_name);
3580 }
3581 
3582 void G_GNUC_PRINTF(2, 3) virtio_error(VirtIODevice *vdev, const char *fmt, ...)
3583 {
3584     va_list ap;
3585 
3586     va_start(ap, fmt);
3587     error_vreport(fmt, ap);
3588     va_end(ap);
3589 
3590     if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
3591         vdev->status = vdev->status | VIRTIO_CONFIG_S_NEEDS_RESET;
3592         virtio_notify_config(vdev);
3593     }
3594 
3595     vdev->broken = true;
3596 }
3597 
3598 static void virtio_memory_listener_commit(MemoryListener *listener)
3599 {
3600     VirtIODevice *vdev = container_of(listener, VirtIODevice, listener);
3601     int i;
3602 
3603     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
3604         if (vdev->vq[i].vring.num == 0) {
3605             break;
3606         }
3607         virtio_init_region_cache(vdev, i);
3608     }
3609 }
3610 
3611 static void virtio_device_realize(DeviceState *dev, Error **errp)
3612 {
3613     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
3614     VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(dev);
3615     Error *err = NULL;
3616 
3617     /* Devices should either use vmsd or the load/save methods */
3618     assert(!vdc->vmsd || !vdc->load);
3619 
3620     if (vdc->realize != NULL) {
3621         vdc->realize(dev, &err);
3622         if (err != NULL) {
3623             error_propagate(errp, err);
3624             return;
3625         }
3626     }
3627 
3628     virtio_bus_device_plugged(vdev, &err);
3629     if (err != NULL) {
3630         error_propagate(errp, err);
3631         vdc->unrealize(dev);
3632         return;
3633     }
3634 
3635     vdev->listener.commit = virtio_memory_listener_commit;
3636     vdev->listener.name = "virtio";
3637     memory_listener_register(&vdev->listener, vdev->dma_as);
3638 }
3639 
3640 static void virtio_device_unrealize(DeviceState *dev)
3641 {
3642     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
3643     VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(dev);
3644 
3645     memory_listener_unregister(&vdev->listener);
3646     virtio_bus_device_unplugged(vdev);
3647 
3648     if (vdc->unrealize != NULL) {
3649         vdc->unrealize(dev);
3650     }
3651 
3652     g_free(vdev->bus_name);
3653     vdev->bus_name = NULL;
3654 }
3655 
3656 static void virtio_device_free_virtqueues(VirtIODevice *vdev)
3657 {
3658     int i;
3659     if (!vdev->vq) {
3660         return;
3661     }
3662 
3663     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
3664         if (vdev->vq[i].vring.num == 0) {
3665             break;
3666         }
3667         virtio_virtqueue_reset_region_cache(&vdev->vq[i]);
3668     }
3669     g_free(vdev->vq);
3670 }
3671 
3672 static void virtio_device_instance_finalize(Object *obj)
3673 {
3674     VirtIODevice *vdev = VIRTIO_DEVICE(obj);
3675 
3676     virtio_device_free_virtqueues(vdev);
3677 
3678     g_free(vdev->config);
3679     g_free(vdev->vector_queues);
3680 }
3681 
3682 static Property virtio_properties[] = {
3683     DEFINE_VIRTIO_COMMON_FEATURES(VirtIODevice, host_features),
3684     DEFINE_PROP_BOOL("use-started", VirtIODevice, use_started, true),
3685     DEFINE_PROP_BOOL("use-disabled-flag", VirtIODevice, use_disabled_flag, true),
3686     DEFINE_PROP_BOOL("x-disable-legacy-check", VirtIODevice,
3687                      disable_legacy_check, false),
3688     DEFINE_PROP_END_OF_LIST(),
3689 };
3690 
3691 static int virtio_device_start_ioeventfd_impl(VirtIODevice *vdev)
3692 {
3693     VirtioBusState *qbus = VIRTIO_BUS(qdev_get_parent_bus(DEVICE(vdev)));
3694     int i, n, r, err;
3695 
3696     /*
3697      * Batch all the host notifiers in a single transaction to avoid
3698      * quadratic time complexity in address_space_update_ioeventfds().
3699      */
3700     memory_region_transaction_begin();
3701     for (n = 0; n < VIRTIO_QUEUE_MAX; n++) {
3702         VirtQueue *vq = &vdev->vq[n];
3703         if (!virtio_queue_get_num(vdev, n)) {
3704             continue;
3705         }
3706         r = virtio_bus_set_host_notifier(qbus, n, true);
3707         if (r < 0) {
3708             err = r;
3709             goto assign_error;
3710         }
3711         event_notifier_set_handler(&vq->host_notifier,
3712                                    virtio_queue_host_notifier_read);
3713     }
3714 
3715     for (n = 0; n < VIRTIO_QUEUE_MAX; n++) {
3716         /* Kick right away to begin processing requests already in vring */
3717         VirtQueue *vq = &vdev->vq[n];
3718         if (!vq->vring.num) {
3719             continue;
3720         }
3721         event_notifier_set(&vq->host_notifier);
3722     }
3723     memory_region_transaction_commit();
3724     return 0;
3725 
3726 assign_error:
3727     i = n; /* save n for a second iteration after transaction is committed. */
3728     while (--n >= 0) {
3729         VirtQueue *vq = &vdev->vq[n];
3730         if (!virtio_queue_get_num(vdev, n)) {
3731             continue;
3732         }
3733 
3734         event_notifier_set_handler(&vq->host_notifier, NULL);
3735         r = virtio_bus_set_host_notifier(qbus, n, false);
3736         assert(r >= 0);
3737     }
3738     /*
3739      * The transaction expects the ioeventfds to be open when it
3740      * commits. Do it now, before the cleanup loop.
3741      */
3742     memory_region_transaction_commit();
3743 
3744     while (--i >= 0) {
3745         if (!virtio_queue_get_num(vdev, i)) {
3746             continue;
3747         }
3748         virtio_bus_cleanup_host_notifier(qbus, i);
3749     }
3750     return err;
3751 }
3752 
3753 int virtio_device_start_ioeventfd(VirtIODevice *vdev)
3754 {
3755     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3756     VirtioBusState *vbus = VIRTIO_BUS(qbus);
3757 
3758     return virtio_bus_start_ioeventfd(vbus);
3759 }
3760 
3761 static void virtio_device_stop_ioeventfd_impl(VirtIODevice *vdev)
3762 {
3763     VirtioBusState *qbus = VIRTIO_BUS(qdev_get_parent_bus(DEVICE(vdev)));
3764     int n, r;
3765 
3766     /*
3767      * Batch all the host notifiers in a single transaction to avoid
3768      * quadratic time complexity in address_space_update_ioeventfds().
3769      */
3770     memory_region_transaction_begin();
3771     for (n = 0; n < VIRTIO_QUEUE_MAX; n++) {
3772         VirtQueue *vq = &vdev->vq[n];
3773 
3774         if (!virtio_queue_get_num(vdev, n)) {
3775             continue;
3776         }
3777         event_notifier_set_handler(&vq->host_notifier, NULL);
3778         r = virtio_bus_set_host_notifier(qbus, n, false);
3779         assert(r >= 0);
3780     }
3781     /*
3782      * The transaction expects the ioeventfds to be open when it
3783      * commits. Do it now, before the cleanup loop.
3784      */
3785     memory_region_transaction_commit();
3786 
3787     for (n = 0; n < VIRTIO_QUEUE_MAX; n++) {
3788         if (!virtio_queue_get_num(vdev, n)) {
3789             continue;
3790         }
3791         virtio_bus_cleanup_host_notifier(qbus, n);
3792     }
3793 }
3794 
3795 int virtio_device_grab_ioeventfd(VirtIODevice *vdev)
3796 {
3797     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3798     VirtioBusState *vbus = VIRTIO_BUS(qbus);
3799 
3800     return virtio_bus_grab_ioeventfd(vbus);
3801 }
3802 
3803 void virtio_device_release_ioeventfd(VirtIODevice *vdev)
3804 {
3805     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3806     VirtioBusState *vbus = VIRTIO_BUS(qbus);
3807 
3808     virtio_bus_release_ioeventfd(vbus);
3809 }
3810 
3811 static void virtio_device_class_init(ObjectClass *klass, void *data)
3812 {
3813     /* Set the default value here. */
3814     VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
3815     DeviceClass *dc = DEVICE_CLASS(klass);
3816 
3817     dc->realize = virtio_device_realize;
3818     dc->unrealize = virtio_device_unrealize;
3819     dc->bus_type = TYPE_VIRTIO_BUS;
3820     device_class_set_props(dc, virtio_properties);
3821     vdc->start_ioeventfd = virtio_device_start_ioeventfd_impl;
3822     vdc->stop_ioeventfd = virtio_device_stop_ioeventfd_impl;
3823 
3824     vdc->legacy_features |= VIRTIO_LEGACY_FEATURES;
3825 }
3826 
3827 bool virtio_device_ioeventfd_enabled(VirtIODevice *vdev)
3828 {
3829     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3830     VirtioBusState *vbus = VIRTIO_BUS(qbus);
3831 
3832     return virtio_bus_ioeventfd_enabled(vbus);
3833 }
3834 
3835 static const TypeInfo virtio_device_info = {
3836     .name = TYPE_VIRTIO_DEVICE,
3837     .parent = TYPE_DEVICE,
3838     .instance_size = sizeof(VirtIODevice),
3839     .class_init = virtio_device_class_init,
3840     .instance_finalize = virtio_device_instance_finalize,
3841     .abstract = true,
3842     .class_size = sizeof(VirtioDeviceClass),
3843 };
3844 
3845 static void virtio_register_types(void)
3846 {
3847     type_register_static(&virtio_device_info);
3848 }
3849 
3850 type_init(virtio_register_types)
3851