xref: /openbmc/qemu/hw/virtio/virtio.c (revision e6e03dcf)
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 "exec/address-spaces.h"
19 #include "qemu/error-report.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 
31 /*
32  * The alignment to use between consumer and producer parts of vring.
33  * x86 pagesize again. This is the default, used by transports like PCI
34  * which don't provide a means for the guest to tell the host the alignment.
35  */
36 #define VIRTIO_PCI_VRING_ALIGN         4096
37 
38 typedef struct VRingDesc
39 {
40     uint64_t addr;
41     uint32_t len;
42     uint16_t flags;
43     uint16_t next;
44 } VRingDesc;
45 
46 typedef struct VRingPackedDesc {
47     uint64_t addr;
48     uint32_t len;
49     uint16_t id;
50     uint16_t flags;
51 } VRingPackedDesc;
52 
53 typedef struct VRingAvail
54 {
55     uint16_t flags;
56     uint16_t idx;
57     uint16_t ring[0];
58 } VRingAvail;
59 
60 typedef struct VRingUsedElem
61 {
62     uint32_t id;
63     uint32_t len;
64 } VRingUsedElem;
65 
66 typedef struct VRingUsed
67 {
68     uint16_t flags;
69     uint16_t idx;
70     VRingUsedElem ring[0];
71 } VRingUsed;
72 
73 typedef struct VRingMemoryRegionCaches {
74     struct rcu_head rcu;
75     MemoryRegionCache desc;
76     MemoryRegionCache avail;
77     MemoryRegionCache used;
78 } VRingMemoryRegionCaches;
79 
80 typedef struct VRing
81 {
82     unsigned int num;
83     unsigned int num_default;
84     unsigned int align;
85     hwaddr desc;
86     hwaddr avail;
87     hwaddr used;
88     VRingMemoryRegionCaches *caches;
89 } VRing;
90 
91 typedef struct VRingPackedDescEvent {
92     uint16_t off_wrap;
93     uint16_t flags;
94 } VRingPackedDescEvent ;
95 
96 struct VirtQueue
97 {
98     VRing vring;
99     VirtQueueElement *used_elems;
100 
101     /* Next head to pop */
102     uint16_t last_avail_idx;
103     bool last_avail_wrap_counter;
104 
105     /* Last avail_idx read from VQ. */
106     uint16_t shadow_avail_idx;
107     bool shadow_avail_wrap_counter;
108 
109     uint16_t used_idx;
110     bool used_wrap_counter;
111 
112     /* Last used index value we have signalled on */
113     uint16_t signalled_used;
114 
115     /* Last used index value we have signalled on */
116     bool signalled_used_valid;
117 
118     /* Notification enabled? */
119     bool notification;
120 
121     uint16_t queue_index;
122 
123     unsigned int inuse;
124 
125     uint16_t vector;
126     VirtIOHandleOutput handle_output;
127     VirtIOHandleAIOOutput handle_aio_output;
128     VirtIODevice *vdev;
129     EventNotifier guest_notifier;
130     EventNotifier host_notifier;
131     QLIST_ENTRY(VirtQueue) node;
132 };
133 
134 static void virtio_free_region_cache(VRingMemoryRegionCaches *caches)
135 {
136     if (!caches) {
137         return;
138     }
139 
140     address_space_cache_destroy(&caches->desc);
141     address_space_cache_destroy(&caches->avail);
142     address_space_cache_destroy(&caches->used);
143     g_free(caches);
144 }
145 
146 static void virtio_virtqueue_reset_region_cache(struct VirtQueue *vq)
147 {
148     VRingMemoryRegionCaches *caches;
149 
150     caches = atomic_read(&vq->vring.caches);
151     atomic_rcu_set(&vq->vring.caches, NULL);
152     if (caches) {
153         call_rcu(caches, virtio_free_region_cache, rcu);
154     }
155 }
156 
157 static void virtio_init_region_cache(VirtIODevice *vdev, int n)
158 {
159     VirtQueue *vq = &vdev->vq[n];
160     VRingMemoryRegionCaches *old = vq->vring.caches;
161     VRingMemoryRegionCaches *new = NULL;
162     hwaddr addr, size;
163     int64_t len;
164     bool packed;
165 
166 
167     addr = vq->vring.desc;
168     if (!addr) {
169         goto out_no_cache;
170     }
171     new = g_new0(VRingMemoryRegionCaches, 1);
172     size = virtio_queue_get_desc_size(vdev, n);
173     packed = virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED) ?
174                                    true : false;
175     len = address_space_cache_init(&new->desc, vdev->dma_as,
176                                    addr, size, packed);
177     if (len < size) {
178         virtio_error(vdev, "Cannot map desc");
179         goto err_desc;
180     }
181 
182     size = virtio_queue_get_used_size(vdev, n);
183     len = address_space_cache_init(&new->used, vdev->dma_as,
184                                    vq->vring.used, size, true);
185     if (len < size) {
186         virtio_error(vdev, "Cannot map used");
187         goto err_used;
188     }
189 
190     size = virtio_queue_get_avail_size(vdev, n);
191     len = address_space_cache_init(&new->avail, vdev->dma_as,
192                                    vq->vring.avail, size, false);
193     if (len < size) {
194         virtio_error(vdev, "Cannot map avail");
195         goto err_avail;
196     }
197 
198     atomic_rcu_set(&vq->vring.caches, new);
199     if (old) {
200         call_rcu(old, virtio_free_region_cache, rcu);
201     }
202     return;
203 
204 err_avail:
205     address_space_cache_destroy(&new->avail);
206 err_used:
207     address_space_cache_destroy(&new->used);
208 err_desc:
209     address_space_cache_destroy(&new->desc);
210 out_no_cache:
211     g_free(new);
212     virtio_virtqueue_reset_region_cache(vq);
213 }
214 
215 /* virt queue functions */
216 void virtio_queue_update_rings(VirtIODevice *vdev, int n)
217 {
218     VRing *vring = &vdev->vq[n].vring;
219 
220     if (!vring->num || !vring->desc || !vring->align) {
221         /* not yet setup -> nothing to do */
222         return;
223     }
224     vring->avail = vring->desc + vring->num * sizeof(VRingDesc);
225     vring->used = vring_align(vring->avail +
226                               offsetof(VRingAvail, ring[vring->num]),
227                               vring->align);
228     virtio_init_region_cache(vdev, n);
229 }
230 
231 /* Called within rcu_read_lock().  */
232 static void vring_split_desc_read(VirtIODevice *vdev, VRingDesc *desc,
233                                   MemoryRegionCache *cache, int i)
234 {
235     address_space_read_cached(cache, i * sizeof(VRingDesc),
236                               desc, sizeof(VRingDesc));
237     virtio_tswap64s(vdev, &desc->addr);
238     virtio_tswap32s(vdev, &desc->len);
239     virtio_tswap16s(vdev, &desc->flags);
240     virtio_tswap16s(vdev, &desc->next);
241 }
242 
243 static void vring_packed_event_read(VirtIODevice *vdev,
244                                     MemoryRegionCache *cache,
245                                     VRingPackedDescEvent *e)
246 {
247     hwaddr off_off = offsetof(VRingPackedDescEvent, off_wrap);
248     hwaddr off_flags = offsetof(VRingPackedDescEvent, flags);
249 
250     address_space_read_cached(cache, off_flags, &e->flags,
251                               sizeof(e->flags));
252     /* Make sure flags is seen before off_wrap */
253     smp_rmb();
254     address_space_read_cached(cache, off_off, &e->off_wrap,
255                               sizeof(e->off_wrap));
256     virtio_tswap16s(vdev, &e->off_wrap);
257     virtio_tswap16s(vdev, &e->flags);
258 }
259 
260 static void vring_packed_off_wrap_write(VirtIODevice *vdev,
261                                         MemoryRegionCache *cache,
262                                         uint16_t off_wrap)
263 {
264     hwaddr off = offsetof(VRingPackedDescEvent, off_wrap);
265 
266     virtio_tswap16s(vdev, &off_wrap);
267     address_space_write_cached(cache, off, &off_wrap, sizeof(off_wrap));
268     address_space_cache_invalidate(cache, off, sizeof(off_wrap));
269 }
270 
271 static void vring_packed_flags_write(VirtIODevice *vdev,
272                                      MemoryRegionCache *cache, uint16_t flags)
273 {
274     hwaddr off = offsetof(VRingPackedDescEvent, flags);
275 
276     virtio_tswap16s(vdev, &flags);
277     address_space_write_cached(cache, off, &flags, sizeof(flags));
278     address_space_cache_invalidate(cache, off, sizeof(flags));
279 }
280 
281 /* Called within rcu_read_lock().  */
282 static VRingMemoryRegionCaches *vring_get_region_caches(struct VirtQueue *vq)
283 {
284     VRingMemoryRegionCaches *caches = atomic_rcu_read(&vq->vring.caches);
285     assert(caches != NULL);
286     return caches;
287 }
288 /* Called within rcu_read_lock().  */
289 static inline uint16_t vring_avail_flags(VirtQueue *vq)
290 {
291     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
292     hwaddr pa = offsetof(VRingAvail, flags);
293     return virtio_lduw_phys_cached(vq->vdev, &caches->avail, pa);
294 }
295 
296 /* Called within rcu_read_lock().  */
297 static inline uint16_t vring_avail_idx(VirtQueue *vq)
298 {
299     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
300     hwaddr pa = offsetof(VRingAvail, idx);
301     vq->shadow_avail_idx = virtio_lduw_phys_cached(vq->vdev, &caches->avail, pa);
302     return vq->shadow_avail_idx;
303 }
304 
305 /* Called within rcu_read_lock().  */
306 static inline uint16_t vring_avail_ring(VirtQueue *vq, int i)
307 {
308     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
309     hwaddr pa = offsetof(VRingAvail, ring[i]);
310     return virtio_lduw_phys_cached(vq->vdev, &caches->avail, pa);
311 }
312 
313 /* Called within rcu_read_lock().  */
314 static inline uint16_t vring_get_used_event(VirtQueue *vq)
315 {
316     return vring_avail_ring(vq, vq->vring.num);
317 }
318 
319 /* Called within rcu_read_lock().  */
320 static inline void vring_used_write(VirtQueue *vq, VRingUsedElem *uelem,
321                                     int i)
322 {
323     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
324     hwaddr pa = offsetof(VRingUsed, ring[i]);
325     virtio_tswap32s(vq->vdev, &uelem->id);
326     virtio_tswap32s(vq->vdev, &uelem->len);
327     address_space_write_cached(&caches->used, pa, uelem, sizeof(VRingUsedElem));
328     address_space_cache_invalidate(&caches->used, pa, sizeof(VRingUsedElem));
329 }
330 
331 /* Called within rcu_read_lock().  */
332 static uint16_t vring_used_idx(VirtQueue *vq)
333 {
334     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
335     hwaddr pa = offsetof(VRingUsed, idx);
336     return virtio_lduw_phys_cached(vq->vdev, &caches->used, pa);
337 }
338 
339 /* Called within rcu_read_lock().  */
340 static inline void vring_used_idx_set(VirtQueue *vq, uint16_t val)
341 {
342     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
343     hwaddr pa = offsetof(VRingUsed, idx);
344     virtio_stw_phys_cached(vq->vdev, &caches->used, pa, val);
345     address_space_cache_invalidate(&caches->used, pa, sizeof(val));
346     vq->used_idx = val;
347 }
348 
349 /* Called within rcu_read_lock().  */
350 static inline void vring_used_flags_set_bit(VirtQueue *vq, int mask)
351 {
352     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
353     VirtIODevice *vdev = vq->vdev;
354     hwaddr pa = offsetof(VRingUsed, flags);
355     uint16_t flags = virtio_lduw_phys_cached(vq->vdev, &caches->used, pa);
356 
357     virtio_stw_phys_cached(vdev, &caches->used, pa, flags | mask);
358     address_space_cache_invalidate(&caches->used, pa, sizeof(flags));
359 }
360 
361 /* Called within rcu_read_lock().  */
362 static inline void vring_used_flags_unset_bit(VirtQueue *vq, int mask)
363 {
364     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
365     VirtIODevice *vdev = vq->vdev;
366     hwaddr pa = offsetof(VRingUsed, flags);
367     uint16_t flags = virtio_lduw_phys_cached(vq->vdev, &caches->used, pa);
368 
369     virtio_stw_phys_cached(vdev, &caches->used, pa, flags & ~mask);
370     address_space_cache_invalidate(&caches->used, pa, sizeof(flags));
371 }
372 
373 /* Called within rcu_read_lock().  */
374 static inline void vring_set_avail_event(VirtQueue *vq, uint16_t val)
375 {
376     VRingMemoryRegionCaches *caches;
377     hwaddr pa;
378     if (!vq->notification) {
379         return;
380     }
381 
382     caches = vring_get_region_caches(vq);
383     pa = offsetof(VRingUsed, ring[vq->vring.num]);
384     virtio_stw_phys_cached(vq->vdev, &caches->used, pa, val);
385     address_space_cache_invalidate(&caches->used, pa, sizeof(val));
386 }
387 
388 static void virtio_queue_split_set_notification(VirtQueue *vq, int enable)
389 {
390     rcu_read_lock();
391     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_RING_F_EVENT_IDX)) {
392         vring_set_avail_event(vq, vring_avail_idx(vq));
393     } else if (enable) {
394         vring_used_flags_unset_bit(vq, VRING_USED_F_NO_NOTIFY);
395     } else {
396         vring_used_flags_set_bit(vq, VRING_USED_F_NO_NOTIFY);
397     }
398     if (enable) {
399         /* Expose avail event/used flags before caller checks the avail idx. */
400         smp_mb();
401     }
402     rcu_read_unlock();
403 }
404 
405 static void virtio_queue_packed_set_notification(VirtQueue *vq, int enable)
406 {
407     uint16_t off_wrap;
408     VRingPackedDescEvent e;
409     VRingMemoryRegionCaches *caches;
410 
411     rcu_read_lock();
412     caches  = vring_get_region_caches(vq);
413     vring_packed_event_read(vq->vdev, &caches->used, &e);
414 
415     if (!enable) {
416         e.flags = VRING_PACKED_EVENT_FLAG_DISABLE;
417     } else if (virtio_vdev_has_feature(vq->vdev, VIRTIO_RING_F_EVENT_IDX)) {
418         off_wrap = vq->shadow_avail_idx | vq->shadow_avail_wrap_counter << 15;
419         vring_packed_off_wrap_write(vq->vdev, &caches->used, off_wrap);
420         /* Make sure off_wrap is wrote before flags */
421         smp_wmb();
422         e.flags = VRING_PACKED_EVENT_FLAG_DESC;
423     } else {
424         e.flags = VRING_PACKED_EVENT_FLAG_ENABLE;
425     }
426 
427     vring_packed_flags_write(vq->vdev, &caches->used, e.flags);
428     if (enable) {
429         /* Expose avail event/used flags before caller checks the avail idx. */
430         smp_mb();
431     }
432     rcu_read_unlock();
433 }
434 
435 void virtio_queue_set_notification(VirtQueue *vq, int enable)
436 {
437     vq->notification = enable;
438 
439     if (!vq->vring.desc) {
440         return;
441     }
442 
443     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) {
444         virtio_queue_packed_set_notification(vq, enable);
445     } else {
446         virtio_queue_split_set_notification(vq, enable);
447     }
448 }
449 
450 int virtio_queue_ready(VirtQueue *vq)
451 {
452     return vq->vring.avail != 0;
453 }
454 
455 static void vring_packed_desc_read_flags(VirtIODevice *vdev,
456                                          uint16_t *flags,
457                                          MemoryRegionCache *cache,
458                                          int i)
459 {
460     address_space_read_cached(cache,
461                               i * sizeof(VRingPackedDesc) +
462                               offsetof(VRingPackedDesc, flags),
463                               flags, sizeof(*flags));
464     virtio_tswap16s(vdev, flags);
465 }
466 
467 static void vring_packed_desc_read(VirtIODevice *vdev,
468                                    VRingPackedDesc *desc,
469                                    MemoryRegionCache *cache,
470                                    int i, bool strict_order)
471 {
472     hwaddr off = i * sizeof(VRingPackedDesc);
473 
474     vring_packed_desc_read_flags(vdev, &desc->flags, cache, i);
475 
476     if (strict_order) {
477         /* Make sure flags is read before the rest fields. */
478         smp_rmb();
479     }
480 
481     address_space_read_cached(cache, off + offsetof(VRingPackedDesc, addr),
482                               &desc->addr, sizeof(desc->addr));
483     address_space_read_cached(cache, off + offsetof(VRingPackedDesc, id),
484                               &desc->id, sizeof(desc->id));
485     address_space_read_cached(cache, off + offsetof(VRingPackedDesc, len),
486                               &desc->len, sizeof(desc->len));
487     virtio_tswap64s(vdev, &desc->addr);
488     virtio_tswap16s(vdev, &desc->id);
489     virtio_tswap32s(vdev, &desc->len);
490 }
491 
492 static void vring_packed_desc_write_data(VirtIODevice *vdev,
493                                          VRingPackedDesc *desc,
494                                          MemoryRegionCache *cache,
495                                          int i)
496 {
497     hwaddr off_id = i * sizeof(VRingPackedDesc) +
498                     offsetof(VRingPackedDesc, id);
499     hwaddr off_len = i * sizeof(VRingPackedDesc) +
500                     offsetof(VRingPackedDesc, len);
501 
502     virtio_tswap32s(vdev, &desc->len);
503     virtio_tswap16s(vdev, &desc->id);
504     address_space_write_cached(cache, off_id, &desc->id, sizeof(desc->id));
505     address_space_cache_invalidate(cache, off_id, sizeof(desc->id));
506     address_space_write_cached(cache, off_len, &desc->len, sizeof(desc->len));
507     address_space_cache_invalidate(cache, off_len, sizeof(desc->len));
508 }
509 
510 static void vring_packed_desc_write_flags(VirtIODevice *vdev,
511                                           VRingPackedDesc *desc,
512                                           MemoryRegionCache *cache,
513                                           int i)
514 {
515     hwaddr off = i * sizeof(VRingPackedDesc) + offsetof(VRingPackedDesc, flags);
516 
517     virtio_tswap16s(vdev, &desc->flags);
518     address_space_write_cached(cache, off, &desc->flags, sizeof(desc->flags));
519     address_space_cache_invalidate(cache, off, sizeof(desc->flags));
520 }
521 
522 static void vring_packed_desc_write(VirtIODevice *vdev,
523                                     VRingPackedDesc *desc,
524                                     MemoryRegionCache *cache,
525                                     int i, bool strict_order)
526 {
527     vring_packed_desc_write_data(vdev, desc, cache, i);
528     if (strict_order) {
529         /* Make sure data is wrote before flags. */
530         smp_wmb();
531     }
532     vring_packed_desc_write_flags(vdev, desc, cache, i);
533 }
534 
535 static inline bool is_desc_avail(uint16_t flags, bool wrap_counter)
536 {
537     bool avail, used;
538 
539     avail = !!(flags & (1 << VRING_PACKED_DESC_F_AVAIL));
540     used = !!(flags & (1 << VRING_PACKED_DESC_F_USED));
541     return (avail != used) && (avail == wrap_counter);
542 }
543 
544 /* Fetch avail_idx from VQ memory only when we really need to know if
545  * guest has added some buffers.
546  * Called within rcu_read_lock().  */
547 static int virtio_queue_empty_rcu(VirtQueue *vq)
548 {
549     if (unlikely(vq->vdev->broken)) {
550         return 1;
551     }
552 
553     if (unlikely(!vq->vring.avail)) {
554         return 1;
555     }
556 
557     if (vq->shadow_avail_idx != vq->last_avail_idx) {
558         return 0;
559     }
560 
561     return vring_avail_idx(vq) == vq->last_avail_idx;
562 }
563 
564 static int virtio_queue_split_empty(VirtQueue *vq)
565 {
566     bool empty;
567 
568     if (unlikely(vq->vdev->broken)) {
569         return 1;
570     }
571 
572     if (unlikely(!vq->vring.avail)) {
573         return 1;
574     }
575 
576     if (vq->shadow_avail_idx != vq->last_avail_idx) {
577         return 0;
578     }
579 
580     rcu_read_lock();
581     empty = vring_avail_idx(vq) == vq->last_avail_idx;
582     rcu_read_unlock();
583     return empty;
584 }
585 
586 static int virtio_queue_packed_empty_rcu(VirtQueue *vq)
587 {
588     struct VRingPackedDesc desc;
589     VRingMemoryRegionCaches *cache;
590 
591     if (unlikely(!vq->vring.desc)) {
592         return 1;
593     }
594 
595     cache = vring_get_region_caches(vq);
596     vring_packed_desc_read_flags(vq->vdev, &desc.flags, &cache->desc,
597                                  vq->last_avail_idx);
598 
599     return !is_desc_avail(desc.flags, vq->last_avail_wrap_counter);
600 }
601 
602 static int virtio_queue_packed_empty(VirtQueue *vq)
603 {
604     bool empty;
605 
606     rcu_read_lock();
607     empty = virtio_queue_packed_empty_rcu(vq);
608     rcu_read_unlock();
609     return empty;
610 }
611 
612 int virtio_queue_empty(VirtQueue *vq)
613 {
614     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) {
615         return virtio_queue_packed_empty(vq);
616     } else {
617         return virtio_queue_split_empty(vq);
618     }
619 }
620 
621 static void virtqueue_unmap_sg(VirtQueue *vq, const VirtQueueElement *elem,
622                                unsigned int len)
623 {
624     AddressSpace *dma_as = vq->vdev->dma_as;
625     unsigned int offset;
626     int i;
627 
628     offset = 0;
629     for (i = 0; i < elem->in_num; i++) {
630         size_t size = MIN(len - offset, elem->in_sg[i].iov_len);
631 
632         dma_memory_unmap(dma_as, elem->in_sg[i].iov_base,
633                          elem->in_sg[i].iov_len,
634                          DMA_DIRECTION_FROM_DEVICE, size);
635 
636         offset += size;
637     }
638 
639     for (i = 0; i < elem->out_num; i++)
640         dma_memory_unmap(dma_as, elem->out_sg[i].iov_base,
641                          elem->out_sg[i].iov_len,
642                          DMA_DIRECTION_TO_DEVICE,
643                          elem->out_sg[i].iov_len);
644 }
645 
646 /* virtqueue_detach_element:
647  * @vq: The #VirtQueue
648  * @elem: The #VirtQueueElement
649  * @len: number of bytes written
650  *
651  * Detach the element from the virtqueue.  This function is suitable for device
652  * reset or other situations where a #VirtQueueElement is simply freed and will
653  * not be pushed or discarded.
654  */
655 void virtqueue_detach_element(VirtQueue *vq, const VirtQueueElement *elem,
656                               unsigned int len)
657 {
658     vq->inuse -= elem->ndescs;
659     virtqueue_unmap_sg(vq, elem, len);
660 }
661 
662 static void virtqueue_split_rewind(VirtQueue *vq, unsigned int num)
663 {
664     vq->last_avail_idx -= num;
665 }
666 
667 static void virtqueue_packed_rewind(VirtQueue *vq, unsigned int num)
668 {
669     if (vq->last_avail_idx < num) {
670         vq->last_avail_idx = vq->vring.num + vq->last_avail_idx - num;
671         vq->last_avail_wrap_counter ^= 1;
672     } else {
673         vq->last_avail_idx -= num;
674     }
675 }
676 
677 /* virtqueue_unpop:
678  * @vq: The #VirtQueue
679  * @elem: The #VirtQueueElement
680  * @len: number of bytes written
681  *
682  * Pretend the most recent element wasn't popped from the virtqueue.  The next
683  * call to virtqueue_pop() will refetch the element.
684  */
685 void virtqueue_unpop(VirtQueue *vq, const VirtQueueElement *elem,
686                      unsigned int len)
687 {
688 
689     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) {
690         virtqueue_packed_rewind(vq, 1);
691     } else {
692         virtqueue_split_rewind(vq, 1);
693     }
694 
695     virtqueue_detach_element(vq, elem, len);
696 }
697 
698 /* virtqueue_rewind:
699  * @vq: The #VirtQueue
700  * @num: Number of elements to push back
701  *
702  * Pretend that elements weren't popped from the virtqueue.  The next
703  * virtqueue_pop() will refetch the oldest element.
704  *
705  * Use virtqueue_unpop() instead if you have a VirtQueueElement.
706  *
707  * Returns: true on success, false if @num is greater than the number of in use
708  * elements.
709  */
710 bool virtqueue_rewind(VirtQueue *vq, unsigned int num)
711 {
712     if (num > vq->inuse) {
713         return false;
714     }
715 
716     vq->inuse -= num;
717     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) {
718         virtqueue_packed_rewind(vq, num);
719     } else {
720         virtqueue_split_rewind(vq, num);
721     }
722     return true;
723 }
724 
725 static void virtqueue_split_fill(VirtQueue *vq, const VirtQueueElement *elem,
726                     unsigned int len, unsigned int idx)
727 {
728     VRingUsedElem uelem;
729 
730     if (unlikely(!vq->vring.used)) {
731         return;
732     }
733 
734     idx = (idx + vq->used_idx) % vq->vring.num;
735 
736     uelem.id = elem->index;
737     uelem.len = len;
738     vring_used_write(vq, &uelem, idx);
739 }
740 
741 static void virtqueue_packed_fill(VirtQueue *vq, const VirtQueueElement *elem,
742                                   unsigned int len, unsigned int idx)
743 {
744     vq->used_elems[idx].index = elem->index;
745     vq->used_elems[idx].len = len;
746     vq->used_elems[idx].ndescs = elem->ndescs;
747 }
748 
749 static void virtqueue_packed_fill_desc(VirtQueue *vq,
750                                        const VirtQueueElement *elem,
751                                        unsigned int idx,
752                                        bool strict_order)
753 {
754     uint16_t head;
755     VRingMemoryRegionCaches *caches;
756     VRingPackedDesc desc = {
757         .id = elem->index,
758         .len = elem->len,
759     };
760     bool wrap_counter = vq->used_wrap_counter;
761 
762     if (unlikely(!vq->vring.desc)) {
763         return;
764     }
765 
766     head = vq->used_idx + idx;
767     if (head >= vq->vring.num) {
768         head -= vq->vring.num;
769         wrap_counter ^= 1;
770     }
771     if (wrap_counter) {
772         desc.flags |= (1 << VRING_PACKED_DESC_F_AVAIL);
773         desc.flags |= (1 << VRING_PACKED_DESC_F_USED);
774     } else {
775         desc.flags &= ~(1 << VRING_PACKED_DESC_F_AVAIL);
776         desc.flags &= ~(1 << VRING_PACKED_DESC_F_USED);
777     }
778 
779     caches = vring_get_region_caches(vq);
780     vring_packed_desc_write(vq->vdev, &desc, &caches->desc, head, strict_order);
781 }
782 
783 /* Called within rcu_read_lock().  */
784 void virtqueue_fill(VirtQueue *vq, const VirtQueueElement *elem,
785                     unsigned int len, unsigned int idx)
786 {
787     trace_virtqueue_fill(vq, elem, len, idx);
788 
789     virtqueue_unmap_sg(vq, elem, len);
790 
791     if (unlikely(vq->vdev->broken)) {
792         return;
793     }
794 
795     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) {
796         virtqueue_packed_fill(vq, elem, len, idx);
797     } else {
798         virtqueue_split_fill(vq, elem, len, idx);
799     }
800 }
801 
802 /* Called within rcu_read_lock().  */
803 static void virtqueue_split_flush(VirtQueue *vq, unsigned int count)
804 {
805     uint16_t old, new;
806 
807     if (unlikely(!vq->vring.used)) {
808         return;
809     }
810 
811     /* Make sure buffer is written before we update index. */
812     smp_wmb();
813     trace_virtqueue_flush(vq, count);
814     old = vq->used_idx;
815     new = old + count;
816     vring_used_idx_set(vq, new);
817     vq->inuse -= count;
818     if (unlikely((int16_t)(new - vq->signalled_used) < (uint16_t)(new - old)))
819         vq->signalled_used_valid = false;
820 }
821 
822 static void virtqueue_packed_flush(VirtQueue *vq, unsigned int count)
823 {
824     unsigned int i, ndescs = 0;
825 
826     if (unlikely(!vq->vring.desc)) {
827         return;
828     }
829 
830     for (i = 1; i < count; i++) {
831         virtqueue_packed_fill_desc(vq, &vq->used_elems[i], i, false);
832         ndescs += vq->used_elems[i].ndescs;
833     }
834     virtqueue_packed_fill_desc(vq, &vq->used_elems[0], 0, true);
835     ndescs += vq->used_elems[0].ndescs;
836 
837     vq->inuse -= ndescs;
838     vq->used_idx += ndescs;
839     if (vq->used_idx >= vq->vring.num) {
840         vq->used_idx -= vq->vring.num;
841         vq->used_wrap_counter ^= 1;
842     }
843 }
844 
845 void virtqueue_flush(VirtQueue *vq, unsigned int count)
846 {
847     if (unlikely(vq->vdev->broken)) {
848         vq->inuse -= count;
849         return;
850     }
851 
852     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) {
853         virtqueue_packed_flush(vq, count);
854     } else {
855         virtqueue_split_flush(vq, count);
856     }
857 }
858 
859 void virtqueue_push(VirtQueue *vq, const VirtQueueElement *elem,
860                     unsigned int len)
861 {
862     rcu_read_lock();
863     virtqueue_fill(vq, elem, len, 0);
864     virtqueue_flush(vq, 1);
865     rcu_read_unlock();
866 }
867 
868 /* Called within rcu_read_lock().  */
869 static int virtqueue_num_heads(VirtQueue *vq, unsigned int idx)
870 {
871     uint16_t num_heads = vring_avail_idx(vq) - idx;
872 
873     /* Check it isn't doing very strange things with descriptor numbers. */
874     if (num_heads > vq->vring.num) {
875         virtio_error(vq->vdev, "Guest moved used index from %u to %u",
876                      idx, vq->shadow_avail_idx);
877         return -EINVAL;
878     }
879     /* On success, callers read a descriptor at vq->last_avail_idx.
880      * Make sure descriptor read does not bypass avail index read. */
881     if (num_heads) {
882         smp_rmb();
883     }
884 
885     return num_heads;
886 }
887 
888 /* Called within rcu_read_lock().  */
889 static bool virtqueue_get_head(VirtQueue *vq, unsigned int idx,
890                                unsigned int *head)
891 {
892     /* Grab the next descriptor number they're advertising, and increment
893      * the index we've seen. */
894     *head = vring_avail_ring(vq, idx % vq->vring.num);
895 
896     /* If their number is silly, that's a fatal mistake. */
897     if (*head >= vq->vring.num) {
898         virtio_error(vq->vdev, "Guest says index %u is available", *head);
899         return false;
900     }
901 
902     return true;
903 }
904 
905 enum {
906     VIRTQUEUE_READ_DESC_ERROR = -1,
907     VIRTQUEUE_READ_DESC_DONE = 0,   /* end of chain */
908     VIRTQUEUE_READ_DESC_MORE = 1,   /* more buffers in chain */
909 };
910 
911 static int virtqueue_split_read_next_desc(VirtIODevice *vdev, VRingDesc *desc,
912                                           MemoryRegionCache *desc_cache,
913                                           unsigned int max, unsigned int *next)
914 {
915     /* If this descriptor says it doesn't chain, we're done. */
916     if (!(desc->flags & VRING_DESC_F_NEXT)) {
917         return VIRTQUEUE_READ_DESC_DONE;
918     }
919 
920     /* Check they're not leading us off end of descriptors. */
921     *next = desc->next;
922     /* Make sure compiler knows to grab that: we don't want it changing! */
923     smp_wmb();
924 
925     if (*next >= max) {
926         virtio_error(vdev, "Desc next is %u", *next);
927         return VIRTQUEUE_READ_DESC_ERROR;
928     }
929 
930     vring_split_desc_read(vdev, desc, desc_cache, *next);
931     return VIRTQUEUE_READ_DESC_MORE;
932 }
933 
934 static void virtqueue_split_get_avail_bytes(VirtQueue *vq,
935                             unsigned int *in_bytes, unsigned int *out_bytes,
936                             unsigned max_in_bytes, unsigned max_out_bytes)
937 {
938     VirtIODevice *vdev = vq->vdev;
939     unsigned int max, idx;
940     unsigned int total_bufs, in_total, out_total;
941     VRingMemoryRegionCaches *caches;
942     MemoryRegionCache indirect_desc_cache = MEMORY_REGION_CACHE_INVALID;
943     int64_t len = 0;
944     int rc;
945 
946     rcu_read_lock();
947     idx = vq->last_avail_idx;
948     total_bufs = in_total = out_total = 0;
949 
950     max = vq->vring.num;
951     caches = vring_get_region_caches(vq);
952     while ((rc = virtqueue_num_heads(vq, idx)) > 0) {
953         MemoryRegionCache *desc_cache = &caches->desc;
954         unsigned int num_bufs;
955         VRingDesc desc;
956         unsigned int i;
957 
958         num_bufs = total_bufs;
959 
960         if (!virtqueue_get_head(vq, idx++, &i)) {
961             goto err;
962         }
963 
964         vring_split_desc_read(vdev, &desc, desc_cache, i);
965 
966         if (desc.flags & VRING_DESC_F_INDIRECT) {
967             if (!desc.len || (desc.len % sizeof(VRingDesc))) {
968                 virtio_error(vdev, "Invalid size for indirect buffer table");
969                 goto err;
970             }
971 
972             /* If we've got too many, that implies a descriptor loop. */
973             if (num_bufs >= max) {
974                 virtio_error(vdev, "Looped descriptor");
975                 goto err;
976             }
977 
978             /* loop over the indirect descriptor table */
979             len = address_space_cache_init(&indirect_desc_cache,
980                                            vdev->dma_as,
981                                            desc.addr, desc.len, false);
982             desc_cache = &indirect_desc_cache;
983             if (len < desc.len) {
984                 virtio_error(vdev, "Cannot map indirect buffer");
985                 goto err;
986             }
987 
988             max = desc.len / sizeof(VRingDesc);
989             num_bufs = i = 0;
990             vring_split_desc_read(vdev, &desc, desc_cache, i);
991         }
992 
993         do {
994             /* If we've got too many, that implies a descriptor loop. */
995             if (++num_bufs > max) {
996                 virtio_error(vdev, "Looped descriptor");
997                 goto err;
998             }
999 
1000             if (desc.flags & VRING_DESC_F_WRITE) {
1001                 in_total += desc.len;
1002             } else {
1003                 out_total += desc.len;
1004             }
1005             if (in_total >= max_in_bytes && out_total >= max_out_bytes) {
1006                 goto done;
1007             }
1008 
1009             rc = virtqueue_split_read_next_desc(vdev, &desc, desc_cache, max, &i);
1010         } while (rc == VIRTQUEUE_READ_DESC_MORE);
1011 
1012         if (rc == VIRTQUEUE_READ_DESC_ERROR) {
1013             goto err;
1014         }
1015 
1016         if (desc_cache == &indirect_desc_cache) {
1017             address_space_cache_destroy(&indirect_desc_cache);
1018             total_bufs++;
1019         } else {
1020             total_bufs = num_bufs;
1021         }
1022     }
1023 
1024     if (rc < 0) {
1025         goto err;
1026     }
1027 
1028 done:
1029     address_space_cache_destroy(&indirect_desc_cache);
1030     if (in_bytes) {
1031         *in_bytes = in_total;
1032     }
1033     if (out_bytes) {
1034         *out_bytes = out_total;
1035     }
1036     rcu_read_unlock();
1037     return;
1038 
1039 err:
1040     in_total = out_total = 0;
1041     goto done;
1042 }
1043 
1044 static int virtqueue_packed_read_next_desc(VirtQueue *vq,
1045                                            VRingPackedDesc *desc,
1046                                            MemoryRegionCache
1047                                            *desc_cache,
1048                                            unsigned int max,
1049                                            unsigned int *next,
1050                                            bool indirect)
1051 {
1052     /* If this descriptor says it doesn't chain, we're done. */
1053     if (!indirect && !(desc->flags & VRING_DESC_F_NEXT)) {
1054         return VIRTQUEUE_READ_DESC_DONE;
1055     }
1056 
1057     ++*next;
1058     if (*next == max) {
1059         if (indirect) {
1060             return VIRTQUEUE_READ_DESC_DONE;
1061         } else {
1062             (*next) -= vq->vring.num;
1063         }
1064     }
1065 
1066     vring_packed_desc_read(vq->vdev, desc, desc_cache, *next, false);
1067     return VIRTQUEUE_READ_DESC_MORE;
1068 }
1069 
1070 static void virtqueue_packed_get_avail_bytes(VirtQueue *vq,
1071                                              unsigned int *in_bytes,
1072                                              unsigned int *out_bytes,
1073                                              unsigned max_in_bytes,
1074                                              unsigned max_out_bytes)
1075 {
1076     VirtIODevice *vdev = vq->vdev;
1077     unsigned int max, idx;
1078     unsigned int total_bufs, in_total, out_total;
1079     MemoryRegionCache *desc_cache;
1080     VRingMemoryRegionCaches *caches;
1081     MemoryRegionCache indirect_desc_cache = MEMORY_REGION_CACHE_INVALID;
1082     int64_t len = 0;
1083     VRingPackedDesc desc;
1084     bool wrap_counter;
1085 
1086     rcu_read_lock();
1087     idx = vq->last_avail_idx;
1088     wrap_counter = vq->last_avail_wrap_counter;
1089     total_bufs = in_total = out_total = 0;
1090 
1091     max = vq->vring.num;
1092     caches = vring_get_region_caches(vq);
1093 
1094     for (;;) {
1095         unsigned int num_bufs = total_bufs;
1096         unsigned int i = idx;
1097         int rc;
1098 
1099         desc_cache = &caches->desc;
1100         vring_packed_desc_read(vdev, &desc, desc_cache, idx, true);
1101         if (!is_desc_avail(desc.flags, wrap_counter)) {
1102             break;
1103         }
1104 
1105         if (desc.flags & VRING_DESC_F_INDIRECT) {
1106             if (desc.len % sizeof(VRingPackedDesc)) {
1107                 virtio_error(vdev, "Invalid size for indirect buffer table");
1108                 goto err;
1109             }
1110 
1111             /* If we've got too many, that implies a descriptor loop. */
1112             if (num_bufs >= max) {
1113                 virtio_error(vdev, "Looped descriptor");
1114                 goto err;
1115             }
1116 
1117             /* loop over the indirect descriptor table */
1118             len = address_space_cache_init(&indirect_desc_cache,
1119                                            vdev->dma_as,
1120                                            desc.addr, desc.len, false);
1121             desc_cache = &indirect_desc_cache;
1122             if (len < desc.len) {
1123                 virtio_error(vdev, "Cannot map indirect buffer");
1124                 goto err;
1125             }
1126 
1127             max = desc.len / sizeof(VRingPackedDesc);
1128             num_bufs = i = 0;
1129             vring_packed_desc_read(vdev, &desc, desc_cache, i, false);
1130         }
1131 
1132         do {
1133             /* If we've got too many, that implies a descriptor loop. */
1134             if (++num_bufs > max) {
1135                 virtio_error(vdev, "Looped descriptor");
1136                 goto err;
1137             }
1138 
1139             if (desc.flags & VRING_DESC_F_WRITE) {
1140                 in_total += desc.len;
1141             } else {
1142                 out_total += desc.len;
1143             }
1144             if (in_total >= max_in_bytes && out_total >= max_out_bytes) {
1145                 goto done;
1146             }
1147 
1148             rc = virtqueue_packed_read_next_desc(vq, &desc, desc_cache, max,
1149                                                  &i, desc_cache ==
1150                                                  &indirect_desc_cache);
1151         } while (rc == VIRTQUEUE_READ_DESC_MORE);
1152 
1153         if (desc_cache == &indirect_desc_cache) {
1154             address_space_cache_destroy(&indirect_desc_cache);
1155             total_bufs++;
1156             idx++;
1157         } else {
1158             idx += num_bufs - total_bufs;
1159             total_bufs = num_bufs;
1160         }
1161 
1162         if (idx >= vq->vring.num) {
1163             idx -= vq->vring.num;
1164             wrap_counter ^= 1;
1165         }
1166     }
1167 
1168     /* Record the index and wrap counter for a kick we want */
1169     vq->shadow_avail_idx = idx;
1170     vq->shadow_avail_wrap_counter = wrap_counter;
1171 done:
1172     address_space_cache_destroy(&indirect_desc_cache);
1173     if (in_bytes) {
1174         *in_bytes = in_total;
1175     }
1176     if (out_bytes) {
1177         *out_bytes = out_total;
1178     }
1179     rcu_read_unlock();
1180     return;
1181 
1182 err:
1183     in_total = out_total = 0;
1184     goto done;
1185 }
1186 
1187 void virtqueue_get_avail_bytes(VirtQueue *vq, unsigned int *in_bytes,
1188                                unsigned int *out_bytes,
1189                                unsigned max_in_bytes, unsigned max_out_bytes)
1190 {
1191     uint16_t desc_size;
1192     VRingMemoryRegionCaches *caches;
1193 
1194     if (unlikely(!vq->vring.desc)) {
1195         goto err;
1196     }
1197 
1198     caches = vring_get_region_caches(vq);
1199     desc_size = virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED) ?
1200                                 sizeof(VRingPackedDesc) : sizeof(VRingDesc);
1201     if (caches->desc.len < vq->vring.num * desc_size) {
1202         virtio_error(vq->vdev, "Cannot map descriptor ring");
1203         goto err;
1204     }
1205 
1206     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) {
1207         virtqueue_packed_get_avail_bytes(vq, in_bytes, out_bytes,
1208                                          max_in_bytes, max_out_bytes);
1209     } else {
1210         virtqueue_split_get_avail_bytes(vq, in_bytes, out_bytes,
1211                                         max_in_bytes, max_out_bytes);
1212     }
1213 
1214     return;
1215 err:
1216     if (in_bytes) {
1217         *in_bytes = 0;
1218     }
1219     if (out_bytes) {
1220         *out_bytes = 0;
1221     }
1222 }
1223 
1224 int virtqueue_avail_bytes(VirtQueue *vq, unsigned int in_bytes,
1225                           unsigned int out_bytes)
1226 {
1227     unsigned int in_total, out_total;
1228 
1229     virtqueue_get_avail_bytes(vq, &in_total, &out_total, in_bytes, out_bytes);
1230     return in_bytes <= in_total && out_bytes <= out_total;
1231 }
1232 
1233 static bool virtqueue_map_desc(VirtIODevice *vdev, unsigned int *p_num_sg,
1234                                hwaddr *addr, struct iovec *iov,
1235                                unsigned int max_num_sg, bool is_write,
1236                                hwaddr pa, size_t sz)
1237 {
1238     bool ok = false;
1239     unsigned num_sg = *p_num_sg;
1240     assert(num_sg <= max_num_sg);
1241 
1242     if (!sz) {
1243         virtio_error(vdev, "virtio: zero sized buffers are not allowed");
1244         goto out;
1245     }
1246 
1247     while (sz) {
1248         hwaddr len = sz;
1249 
1250         if (num_sg == max_num_sg) {
1251             virtio_error(vdev, "virtio: too many write descriptors in "
1252                                "indirect table");
1253             goto out;
1254         }
1255 
1256         iov[num_sg].iov_base = dma_memory_map(vdev->dma_as, pa, &len,
1257                                               is_write ?
1258                                               DMA_DIRECTION_FROM_DEVICE :
1259                                               DMA_DIRECTION_TO_DEVICE);
1260         if (!iov[num_sg].iov_base) {
1261             virtio_error(vdev, "virtio: bogus descriptor or out of resources");
1262             goto out;
1263         }
1264 
1265         iov[num_sg].iov_len = len;
1266         addr[num_sg] = pa;
1267 
1268         sz -= len;
1269         pa += len;
1270         num_sg++;
1271     }
1272     ok = true;
1273 
1274 out:
1275     *p_num_sg = num_sg;
1276     return ok;
1277 }
1278 
1279 /* Only used by error code paths before we have a VirtQueueElement (therefore
1280  * virtqueue_unmap_sg() can't be used).  Assumes buffers weren't written to
1281  * yet.
1282  */
1283 static void virtqueue_undo_map_desc(unsigned int out_num, unsigned int in_num,
1284                                     struct iovec *iov)
1285 {
1286     unsigned int i;
1287 
1288     for (i = 0; i < out_num + in_num; i++) {
1289         int is_write = i >= out_num;
1290 
1291         cpu_physical_memory_unmap(iov->iov_base, iov->iov_len, is_write, 0);
1292         iov++;
1293     }
1294 }
1295 
1296 static void virtqueue_map_iovec(VirtIODevice *vdev, struct iovec *sg,
1297                                 hwaddr *addr, unsigned int num_sg,
1298                                 int is_write)
1299 {
1300     unsigned int i;
1301     hwaddr len;
1302 
1303     for (i = 0; i < num_sg; i++) {
1304         len = sg[i].iov_len;
1305         sg[i].iov_base = dma_memory_map(vdev->dma_as,
1306                                         addr[i], &len, is_write ?
1307                                         DMA_DIRECTION_FROM_DEVICE :
1308                                         DMA_DIRECTION_TO_DEVICE);
1309         if (!sg[i].iov_base) {
1310             error_report("virtio: error trying to map MMIO memory");
1311             exit(1);
1312         }
1313         if (len != sg[i].iov_len) {
1314             error_report("virtio: unexpected memory split");
1315             exit(1);
1316         }
1317     }
1318 }
1319 
1320 void virtqueue_map(VirtIODevice *vdev, VirtQueueElement *elem)
1321 {
1322     virtqueue_map_iovec(vdev, elem->in_sg, elem->in_addr, elem->in_num, 1);
1323     virtqueue_map_iovec(vdev, elem->out_sg, elem->out_addr, elem->out_num, 0);
1324 }
1325 
1326 static void *virtqueue_alloc_element(size_t sz, unsigned out_num, unsigned in_num)
1327 {
1328     VirtQueueElement *elem;
1329     size_t in_addr_ofs = QEMU_ALIGN_UP(sz, __alignof__(elem->in_addr[0]));
1330     size_t out_addr_ofs = in_addr_ofs + in_num * sizeof(elem->in_addr[0]);
1331     size_t out_addr_end = out_addr_ofs + out_num * sizeof(elem->out_addr[0]);
1332     size_t in_sg_ofs = QEMU_ALIGN_UP(out_addr_end, __alignof__(elem->in_sg[0]));
1333     size_t out_sg_ofs = in_sg_ofs + in_num * sizeof(elem->in_sg[0]);
1334     size_t out_sg_end = out_sg_ofs + out_num * sizeof(elem->out_sg[0]);
1335 
1336     assert(sz >= sizeof(VirtQueueElement));
1337     elem = g_malloc(out_sg_end);
1338     trace_virtqueue_alloc_element(elem, sz, in_num, out_num);
1339     elem->out_num = out_num;
1340     elem->in_num = in_num;
1341     elem->in_addr = (void *)elem + in_addr_ofs;
1342     elem->out_addr = (void *)elem + out_addr_ofs;
1343     elem->in_sg = (void *)elem + in_sg_ofs;
1344     elem->out_sg = (void *)elem + out_sg_ofs;
1345     return elem;
1346 }
1347 
1348 static void *virtqueue_split_pop(VirtQueue *vq, size_t sz)
1349 {
1350     unsigned int i, head, max;
1351     VRingMemoryRegionCaches *caches;
1352     MemoryRegionCache indirect_desc_cache = MEMORY_REGION_CACHE_INVALID;
1353     MemoryRegionCache *desc_cache;
1354     int64_t len;
1355     VirtIODevice *vdev = vq->vdev;
1356     VirtQueueElement *elem = NULL;
1357     unsigned out_num, in_num, elem_entries;
1358     hwaddr addr[VIRTQUEUE_MAX_SIZE];
1359     struct iovec iov[VIRTQUEUE_MAX_SIZE];
1360     VRingDesc desc;
1361     int rc;
1362 
1363     rcu_read_lock();
1364     if (virtio_queue_empty_rcu(vq)) {
1365         goto done;
1366     }
1367     /* Needed after virtio_queue_empty(), see comment in
1368      * virtqueue_num_heads(). */
1369     smp_rmb();
1370 
1371     /* When we start there are none of either input nor output. */
1372     out_num = in_num = elem_entries = 0;
1373 
1374     max = vq->vring.num;
1375 
1376     if (vq->inuse >= vq->vring.num) {
1377         virtio_error(vdev, "Virtqueue size exceeded");
1378         goto done;
1379     }
1380 
1381     if (!virtqueue_get_head(vq, vq->last_avail_idx++, &head)) {
1382         goto done;
1383     }
1384 
1385     if (virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX)) {
1386         vring_set_avail_event(vq, vq->last_avail_idx);
1387     }
1388 
1389     i = head;
1390 
1391     caches = vring_get_region_caches(vq);
1392     if (caches->desc.len < max * sizeof(VRingDesc)) {
1393         virtio_error(vdev, "Cannot map descriptor ring");
1394         goto done;
1395     }
1396 
1397     desc_cache = &caches->desc;
1398     vring_split_desc_read(vdev, &desc, desc_cache, i);
1399     if (desc.flags & VRING_DESC_F_INDIRECT) {
1400         if (!desc.len || (desc.len % sizeof(VRingDesc))) {
1401             virtio_error(vdev, "Invalid size for indirect buffer table");
1402             goto done;
1403         }
1404 
1405         /* loop over the indirect descriptor table */
1406         len = address_space_cache_init(&indirect_desc_cache, vdev->dma_as,
1407                                        desc.addr, desc.len, false);
1408         desc_cache = &indirect_desc_cache;
1409         if (len < desc.len) {
1410             virtio_error(vdev, "Cannot map indirect buffer");
1411             goto done;
1412         }
1413 
1414         max = desc.len / sizeof(VRingDesc);
1415         i = 0;
1416         vring_split_desc_read(vdev, &desc, desc_cache, i);
1417     }
1418 
1419     /* Collect all the descriptors */
1420     do {
1421         bool map_ok;
1422 
1423         if (desc.flags & VRING_DESC_F_WRITE) {
1424             map_ok = virtqueue_map_desc(vdev, &in_num, addr + out_num,
1425                                         iov + out_num,
1426                                         VIRTQUEUE_MAX_SIZE - out_num, true,
1427                                         desc.addr, desc.len);
1428         } else {
1429             if (in_num) {
1430                 virtio_error(vdev, "Incorrect order for descriptors");
1431                 goto err_undo_map;
1432             }
1433             map_ok = virtqueue_map_desc(vdev, &out_num, addr, iov,
1434                                         VIRTQUEUE_MAX_SIZE, false,
1435                                         desc.addr, desc.len);
1436         }
1437         if (!map_ok) {
1438             goto err_undo_map;
1439         }
1440 
1441         /* If we've got too many, that implies a descriptor loop. */
1442         if (++elem_entries > max) {
1443             virtio_error(vdev, "Looped descriptor");
1444             goto err_undo_map;
1445         }
1446 
1447         rc = virtqueue_split_read_next_desc(vdev, &desc, desc_cache, max, &i);
1448     } while (rc == VIRTQUEUE_READ_DESC_MORE);
1449 
1450     if (rc == VIRTQUEUE_READ_DESC_ERROR) {
1451         goto err_undo_map;
1452     }
1453 
1454     /* Now copy what we have collected and mapped */
1455     elem = virtqueue_alloc_element(sz, out_num, in_num);
1456     elem->index = head;
1457     elem->ndescs = 1;
1458     for (i = 0; i < out_num; i++) {
1459         elem->out_addr[i] = addr[i];
1460         elem->out_sg[i] = iov[i];
1461     }
1462     for (i = 0; i < in_num; i++) {
1463         elem->in_addr[i] = addr[out_num + i];
1464         elem->in_sg[i] = iov[out_num + i];
1465     }
1466 
1467     vq->inuse++;
1468 
1469     trace_virtqueue_pop(vq, elem, elem->in_num, elem->out_num);
1470 done:
1471     address_space_cache_destroy(&indirect_desc_cache);
1472     rcu_read_unlock();
1473 
1474     return elem;
1475 
1476 err_undo_map:
1477     virtqueue_undo_map_desc(out_num, in_num, iov);
1478     goto done;
1479 }
1480 
1481 static void *virtqueue_packed_pop(VirtQueue *vq, size_t sz)
1482 {
1483     unsigned int i, max;
1484     VRingMemoryRegionCaches *caches;
1485     MemoryRegionCache indirect_desc_cache = MEMORY_REGION_CACHE_INVALID;
1486     MemoryRegionCache *desc_cache;
1487     int64_t len;
1488     VirtIODevice *vdev = vq->vdev;
1489     VirtQueueElement *elem = NULL;
1490     unsigned out_num, in_num, elem_entries;
1491     hwaddr addr[VIRTQUEUE_MAX_SIZE];
1492     struct iovec iov[VIRTQUEUE_MAX_SIZE];
1493     VRingPackedDesc desc;
1494     uint16_t id;
1495     int rc;
1496 
1497     rcu_read_lock();
1498     if (virtio_queue_packed_empty_rcu(vq)) {
1499         goto done;
1500     }
1501 
1502     /* When we start there are none of either input nor output. */
1503     out_num = in_num = elem_entries = 0;
1504 
1505     max = vq->vring.num;
1506 
1507     if (vq->inuse >= vq->vring.num) {
1508         virtio_error(vdev, "Virtqueue size exceeded");
1509         goto done;
1510     }
1511 
1512     i = vq->last_avail_idx;
1513 
1514     caches = vring_get_region_caches(vq);
1515     if (caches->desc.len < max * sizeof(VRingDesc)) {
1516         virtio_error(vdev, "Cannot map descriptor ring");
1517         goto done;
1518     }
1519 
1520     desc_cache = &caches->desc;
1521     vring_packed_desc_read(vdev, &desc, desc_cache, i, true);
1522     id = desc.id;
1523     if (desc.flags & VRING_DESC_F_INDIRECT) {
1524         if (desc.len % sizeof(VRingPackedDesc)) {
1525             virtio_error(vdev, "Invalid size for indirect buffer table");
1526             goto done;
1527         }
1528 
1529         /* loop over the indirect descriptor table */
1530         len = address_space_cache_init(&indirect_desc_cache, vdev->dma_as,
1531                                        desc.addr, desc.len, false);
1532         desc_cache = &indirect_desc_cache;
1533         if (len < desc.len) {
1534             virtio_error(vdev, "Cannot map indirect buffer");
1535             goto done;
1536         }
1537 
1538         max = desc.len / sizeof(VRingPackedDesc);
1539         i = 0;
1540         vring_packed_desc_read(vdev, &desc, desc_cache, i, false);
1541     }
1542 
1543     /* Collect all the descriptors */
1544     do {
1545         bool map_ok;
1546 
1547         if (desc.flags & VRING_DESC_F_WRITE) {
1548             map_ok = virtqueue_map_desc(vdev, &in_num, addr + out_num,
1549                                         iov + out_num,
1550                                         VIRTQUEUE_MAX_SIZE - out_num, true,
1551                                         desc.addr, desc.len);
1552         } else {
1553             if (in_num) {
1554                 virtio_error(vdev, "Incorrect order for descriptors");
1555                 goto err_undo_map;
1556             }
1557             map_ok = virtqueue_map_desc(vdev, &out_num, addr, iov,
1558                                         VIRTQUEUE_MAX_SIZE, false,
1559                                         desc.addr, desc.len);
1560         }
1561         if (!map_ok) {
1562             goto err_undo_map;
1563         }
1564 
1565         /* If we've got too many, that implies a descriptor loop. */
1566         if (++elem_entries > max) {
1567             virtio_error(vdev, "Looped descriptor");
1568             goto err_undo_map;
1569         }
1570 
1571         rc = virtqueue_packed_read_next_desc(vq, &desc, desc_cache, max, &i,
1572                                              desc_cache ==
1573                                              &indirect_desc_cache);
1574     } while (rc == VIRTQUEUE_READ_DESC_MORE);
1575 
1576     /* Now copy what we have collected and mapped */
1577     elem = virtqueue_alloc_element(sz, out_num, in_num);
1578     for (i = 0; i < out_num; i++) {
1579         elem->out_addr[i] = addr[i];
1580         elem->out_sg[i] = iov[i];
1581     }
1582     for (i = 0; i < in_num; i++) {
1583         elem->in_addr[i] = addr[out_num + i];
1584         elem->in_sg[i] = iov[out_num + i];
1585     }
1586 
1587     elem->index = id;
1588     elem->ndescs = (desc_cache == &indirect_desc_cache) ? 1 : elem_entries;
1589     vq->last_avail_idx += elem->ndescs;
1590     vq->inuse += elem->ndescs;
1591 
1592     if (vq->last_avail_idx >= vq->vring.num) {
1593         vq->last_avail_idx -= vq->vring.num;
1594         vq->last_avail_wrap_counter ^= 1;
1595     }
1596 
1597     vq->shadow_avail_idx = vq->last_avail_idx;
1598     vq->shadow_avail_wrap_counter = vq->last_avail_wrap_counter;
1599 
1600     trace_virtqueue_pop(vq, elem, elem->in_num, elem->out_num);
1601 done:
1602     address_space_cache_destroy(&indirect_desc_cache);
1603     rcu_read_unlock();
1604 
1605     return elem;
1606 
1607 err_undo_map:
1608     virtqueue_undo_map_desc(out_num, in_num, iov);
1609     goto done;
1610 }
1611 
1612 void *virtqueue_pop(VirtQueue *vq, size_t sz)
1613 {
1614     if (unlikely(vq->vdev->broken)) {
1615         return NULL;
1616     }
1617 
1618     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) {
1619         return virtqueue_packed_pop(vq, sz);
1620     } else {
1621         return virtqueue_split_pop(vq, sz);
1622     }
1623 }
1624 
1625 static unsigned int virtqueue_packed_drop_all(VirtQueue *vq)
1626 {
1627     VRingMemoryRegionCaches *caches;
1628     MemoryRegionCache *desc_cache;
1629     unsigned int dropped = 0;
1630     VirtQueueElement elem = {};
1631     VirtIODevice *vdev = vq->vdev;
1632     VRingPackedDesc desc;
1633 
1634     caches = vring_get_region_caches(vq);
1635     desc_cache = &caches->desc;
1636 
1637     virtio_queue_set_notification(vq, 0);
1638 
1639     while (vq->inuse < vq->vring.num) {
1640         unsigned int idx = vq->last_avail_idx;
1641         /*
1642          * works similar to virtqueue_pop but does not map buffers
1643          * and does not allocate any memory.
1644          */
1645         vring_packed_desc_read(vdev, &desc, desc_cache,
1646                                vq->last_avail_idx , true);
1647         if (!is_desc_avail(desc.flags, vq->last_avail_wrap_counter)) {
1648             break;
1649         }
1650         elem.index = desc.id;
1651         elem.ndescs = 1;
1652         while (virtqueue_packed_read_next_desc(vq, &desc, desc_cache,
1653                                                vq->vring.num, &idx, false)) {
1654             ++elem.ndescs;
1655         }
1656         /*
1657          * immediately push the element, nothing to unmap
1658          * as both in_num and out_num are set to 0.
1659          */
1660         virtqueue_push(vq, &elem, 0);
1661         dropped++;
1662         vq->last_avail_idx += elem.ndescs;
1663         if (vq->last_avail_idx >= vq->vring.num) {
1664             vq->last_avail_idx -= vq->vring.num;
1665             vq->last_avail_wrap_counter ^= 1;
1666         }
1667     }
1668 
1669     return dropped;
1670 }
1671 
1672 static unsigned int virtqueue_split_drop_all(VirtQueue *vq)
1673 {
1674     unsigned int dropped = 0;
1675     VirtQueueElement elem = {};
1676     VirtIODevice *vdev = vq->vdev;
1677     bool fEventIdx = virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX);
1678 
1679     while (!virtio_queue_empty(vq) && vq->inuse < vq->vring.num) {
1680         /* works similar to virtqueue_pop but does not map buffers
1681         * and does not allocate any memory */
1682         smp_rmb();
1683         if (!virtqueue_get_head(vq, vq->last_avail_idx, &elem.index)) {
1684             break;
1685         }
1686         vq->inuse++;
1687         vq->last_avail_idx++;
1688         if (fEventIdx) {
1689             vring_set_avail_event(vq, vq->last_avail_idx);
1690         }
1691         /* immediately push the element, nothing to unmap
1692          * as both in_num and out_num are set to 0 */
1693         virtqueue_push(vq, &elem, 0);
1694         dropped++;
1695     }
1696 
1697     return dropped;
1698 }
1699 
1700 /* virtqueue_drop_all:
1701  * @vq: The #VirtQueue
1702  * Drops all queued buffers and indicates them to the guest
1703  * as if they are done. Useful when buffers can not be
1704  * processed but must be returned to the guest.
1705  */
1706 unsigned int virtqueue_drop_all(VirtQueue *vq)
1707 {
1708     struct VirtIODevice *vdev = vq->vdev;
1709 
1710     if (unlikely(vdev->broken)) {
1711         return 0;
1712     }
1713 
1714     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
1715         return virtqueue_packed_drop_all(vq);
1716     } else {
1717         return virtqueue_split_drop_all(vq);
1718     }
1719 }
1720 
1721 /* Reading and writing a structure directly to QEMUFile is *awful*, but
1722  * it is what QEMU has always done by mistake.  We can change it sooner
1723  * or later by bumping the version number of the affected vm states.
1724  * In the meanwhile, since the in-memory layout of VirtQueueElement
1725  * has changed, we need to marshal to and from the layout that was
1726  * used before the change.
1727  */
1728 typedef struct VirtQueueElementOld {
1729     unsigned int index;
1730     unsigned int out_num;
1731     unsigned int in_num;
1732     hwaddr in_addr[VIRTQUEUE_MAX_SIZE];
1733     hwaddr out_addr[VIRTQUEUE_MAX_SIZE];
1734     struct iovec in_sg[VIRTQUEUE_MAX_SIZE];
1735     struct iovec out_sg[VIRTQUEUE_MAX_SIZE];
1736 } VirtQueueElementOld;
1737 
1738 void *qemu_get_virtqueue_element(VirtIODevice *vdev, QEMUFile *f, size_t sz)
1739 {
1740     VirtQueueElement *elem;
1741     VirtQueueElementOld data;
1742     int i;
1743 
1744     qemu_get_buffer(f, (uint8_t *)&data, sizeof(VirtQueueElementOld));
1745 
1746     /* TODO: teach all callers that this can fail, and return failure instead
1747      * of asserting here.
1748      * This is just one thing (there are probably more) that must be
1749      * fixed before we can allow NDEBUG compilation.
1750      */
1751     assert(ARRAY_SIZE(data.in_addr) >= data.in_num);
1752     assert(ARRAY_SIZE(data.out_addr) >= data.out_num);
1753 
1754     elem = virtqueue_alloc_element(sz, data.out_num, data.in_num);
1755     elem->index = data.index;
1756 
1757     for (i = 0; i < elem->in_num; i++) {
1758         elem->in_addr[i] = data.in_addr[i];
1759     }
1760 
1761     for (i = 0; i < elem->out_num; i++) {
1762         elem->out_addr[i] = data.out_addr[i];
1763     }
1764 
1765     for (i = 0; i < elem->in_num; i++) {
1766         /* Base is overwritten by virtqueue_map.  */
1767         elem->in_sg[i].iov_base = 0;
1768         elem->in_sg[i].iov_len = data.in_sg[i].iov_len;
1769     }
1770 
1771     for (i = 0; i < elem->out_num; i++) {
1772         /* Base is overwritten by virtqueue_map.  */
1773         elem->out_sg[i].iov_base = 0;
1774         elem->out_sg[i].iov_len = data.out_sg[i].iov_len;
1775     }
1776 
1777     if (virtio_host_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
1778         qemu_get_be32s(f, &elem->ndescs);
1779     }
1780 
1781     virtqueue_map(vdev, elem);
1782     return elem;
1783 }
1784 
1785 void qemu_put_virtqueue_element(VirtIODevice *vdev, QEMUFile *f,
1786                                 VirtQueueElement *elem)
1787 {
1788     VirtQueueElementOld data;
1789     int i;
1790 
1791     memset(&data, 0, sizeof(data));
1792     data.index = elem->index;
1793     data.in_num = elem->in_num;
1794     data.out_num = elem->out_num;
1795 
1796     for (i = 0; i < elem->in_num; i++) {
1797         data.in_addr[i] = elem->in_addr[i];
1798     }
1799 
1800     for (i = 0; i < elem->out_num; i++) {
1801         data.out_addr[i] = elem->out_addr[i];
1802     }
1803 
1804     for (i = 0; i < elem->in_num; i++) {
1805         /* Base is overwritten by virtqueue_map when loading.  Do not
1806          * save it, as it would leak the QEMU address space layout.  */
1807         data.in_sg[i].iov_len = elem->in_sg[i].iov_len;
1808     }
1809 
1810     for (i = 0; i < elem->out_num; i++) {
1811         /* Do not save iov_base as above.  */
1812         data.out_sg[i].iov_len = elem->out_sg[i].iov_len;
1813     }
1814 
1815     if (virtio_host_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
1816         qemu_put_be32s(f, &elem->ndescs);
1817     }
1818 
1819     qemu_put_buffer(f, (uint8_t *)&data, sizeof(VirtQueueElementOld));
1820 }
1821 
1822 /* virtio device */
1823 static void virtio_notify_vector(VirtIODevice *vdev, uint16_t vector)
1824 {
1825     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
1826     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
1827 
1828     if (unlikely(vdev->broken)) {
1829         return;
1830     }
1831 
1832     if (k->notify) {
1833         k->notify(qbus->parent, vector);
1834     }
1835 }
1836 
1837 void virtio_update_irq(VirtIODevice *vdev)
1838 {
1839     virtio_notify_vector(vdev, VIRTIO_NO_VECTOR);
1840 }
1841 
1842 static int virtio_validate_features(VirtIODevice *vdev)
1843 {
1844     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1845 
1846     if (virtio_host_has_feature(vdev, VIRTIO_F_IOMMU_PLATFORM) &&
1847         !virtio_vdev_has_feature(vdev, VIRTIO_F_IOMMU_PLATFORM)) {
1848         return -EFAULT;
1849     }
1850 
1851     if (k->validate_features) {
1852         return k->validate_features(vdev);
1853     } else {
1854         return 0;
1855     }
1856 }
1857 
1858 int virtio_set_status(VirtIODevice *vdev, uint8_t val)
1859 {
1860     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1861     trace_virtio_set_status(vdev, val);
1862 
1863     if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
1864         if (!(vdev->status & VIRTIO_CONFIG_S_FEATURES_OK) &&
1865             val & VIRTIO_CONFIG_S_FEATURES_OK) {
1866             int ret = virtio_validate_features(vdev);
1867 
1868             if (ret) {
1869                 return ret;
1870             }
1871         }
1872     }
1873 
1874     if ((vdev->status & VIRTIO_CONFIG_S_DRIVER_OK) !=
1875         (val & VIRTIO_CONFIG_S_DRIVER_OK)) {
1876         virtio_set_started(vdev, val & VIRTIO_CONFIG_S_DRIVER_OK);
1877     }
1878 
1879     if (k->set_status) {
1880         k->set_status(vdev, val);
1881     }
1882     vdev->status = val;
1883 
1884     return 0;
1885 }
1886 
1887 static enum virtio_device_endian virtio_default_endian(void)
1888 {
1889     if (target_words_bigendian()) {
1890         return VIRTIO_DEVICE_ENDIAN_BIG;
1891     } else {
1892         return VIRTIO_DEVICE_ENDIAN_LITTLE;
1893     }
1894 }
1895 
1896 static enum virtio_device_endian virtio_current_cpu_endian(void)
1897 {
1898     CPUClass *cc = CPU_GET_CLASS(current_cpu);
1899 
1900     if (cc->virtio_is_big_endian(current_cpu)) {
1901         return VIRTIO_DEVICE_ENDIAN_BIG;
1902     } else {
1903         return VIRTIO_DEVICE_ENDIAN_LITTLE;
1904     }
1905 }
1906 
1907 void virtio_reset(void *opaque)
1908 {
1909     VirtIODevice *vdev = opaque;
1910     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1911     int i;
1912 
1913     virtio_set_status(vdev, 0);
1914     if (current_cpu) {
1915         /* Guest initiated reset */
1916         vdev->device_endian = virtio_current_cpu_endian();
1917     } else {
1918         /* System reset */
1919         vdev->device_endian = virtio_default_endian();
1920     }
1921 
1922     if (k->reset) {
1923         k->reset(vdev);
1924     }
1925 
1926     vdev->start_on_kick = false;
1927     vdev->started = false;
1928     vdev->broken = false;
1929     vdev->guest_features = 0;
1930     vdev->queue_sel = 0;
1931     vdev->status = 0;
1932     atomic_set(&vdev->isr, 0);
1933     vdev->config_vector = VIRTIO_NO_VECTOR;
1934     virtio_notify_vector(vdev, vdev->config_vector);
1935 
1936     for(i = 0; i < VIRTIO_QUEUE_MAX; i++) {
1937         vdev->vq[i].vring.desc = 0;
1938         vdev->vq[i].vring.avail = 0;
1939         vdev->vq[i].vring.used = 0;
1940         vdev->vq[i].last_avail_idx = 0;
1941         vdev->vq[i].shadow_avail_idx = 0;
1942         vdev->vq[i].used_idx = 0;
1943         vdev->vq[i].last_avail_wrap_counter = true;
1944         vdev->vq[i].shadow_avail_wrap_counter = true;
1945         vdev->vq[i].used_wrap_counter = true;
1946         virtio_queue_set_vector(vdev, i, VIRTIO_NO_VECTOR);
1947         vdev->vq[i].signalled_used = 0;
1948         vdev->vq[i].signalled_used_valid = false;
1949         vdev->vq[i].notification = true;
1950         vdev->vq[i].vring.num = vdev->vq[i].vring.num_default;
1951         vdev->vq[i].inuse = 0;
1952         virtio_virtqueue_reset_region_cache(&vdev->vq[i]);
1953     }
1954 }
1955 
1956 uint32_t virtio_config_readb(VirtIODevice *vdev, uint32_t addr)
1957 {
1958     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1959     uint8_t val;
1960 
1961     if (addr + sizeof(val) > vdev->config_len) {
1962         return (uint32_t)-1;
1963     }
1964 
1965     k->get_config(vdev, vdev->config);
1966 
1967     val = ldub_p(vdev->config + addr);
1968     return val;
1969 }
1970 
1971 uint32_t virtio_config_readw(VirtIODevice *vdev, uint32_t addr)
1972 {
1973     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1974     uint16_t val;
1975 
1976     if (addr + sizeof(val) > vdev->config_len) {
1977         return (uint32_t)-1;
1978     }
1979 
1980     k->get_config(vdev, vdev->config);
1981 
1982     val = lduw_p(vdev->config + addr);
1983     return val;
1984 }
1985 
1986 uint32_t virtio_config_readl(VirtIODevice *vdev, uint32_t addr)
1987 {
1988     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1989     uint32_t val;
1990 
1991     if (addr + sizeof(val) > vdev->config_len) {
1992         return (uint32_t)-1;
1993     }
1994 
1995     k->get_config(vdev, vdev->config);
1996 
1997     val = ldl_p(vdev->config + addr);
1998     return val;
1999 }
2000 
2001 void virtio_config_writeb(VirtIODevice *vdev, uint32_t addr, uint32_t data)
2002 {
2003     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2004     uint8_t val = data;
2005 
2006     if (addr + sizeof(val) > vdev->config_len) {
2007         return;
2008     }
2009 
2010     stb_p(vdev->config + addr, val);
2011 
2012     if (k->set_config) {
2013         k->set_config(vdev, vdev->config);
2014     }
2015 }
2016 
2017 void virtio_config_writew(VirtIODevice *vdev, uint32_t addr, uint32_t data)
2018 {
2019     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2020     uint16_t val = data;
2021 
2022     if (addr + sizeof(val) > vdev->config_len) {
2023         return;
2024     }
2025 
2026     stw_p(vdev->config + addr, val);
2027 
2028     if (k->set_config) {
2029         k->set_config(vdev, vdev->config);
2030     }
2031 }
2032 
2033 void virtio_config_writel(VirtIODevice *vdev, uint32_t addr, uint32_t data)
2034 {
2035     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2036     uint32_t val = data;
2037 
2038     if (addr + sizeof(val) > vdev->config_len) {
2039         return;
2040     }
2041 
2042     stl_p(vdev->config + addr, val);
2043 
2044     if (k->set_config) {
2045         k->set_config(vdev, vdev->config);
2046     }
2047 }
2048 
2049 uint32_t virtio_config_modern_readb(VirtIODevice *vdev, uint32_t addr)
2050 {
2051     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2052     uint8_t val;
2053 
2054     if (addr + sizeof(val) > vdev->config_len) {
2055         return (uint32_t)-1;
2056     }
2057 
2058     k->get_config(vdev, vdev->config);
2059 
2060     val = ldub_p(vdev->config + addr);
2061     return val;
2062 }
2063 
2064 uint32_t virtio_config_modern_readw(VirtIODevice *vdev, uint32_t addr)
2065 {
2066     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2067     uint16_t val;
2068 
2069     if (addr + sizeof(val) > vdev->config_len) {
2070         return (uint32_t)-1;
2071     }
2072 
2073     k->get_config(vdev, vdev->config);
2074 
2075     val = lduw_le_p(vdev->config + addr);
2076     return val;
2077 }
2078 
2079 uint32_t virtio_config_modern_readl(VirtIODevice *vdev, uint32_t addr)
2080 {
2081     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2082     uint32_t val;
2083 
2084     if (addr + sizeof(val) > vdev->config_len) {
2085         return (uint32_t)-1;
2086     }
2087 
2088     k->get_config(vdev, vdev->config);
2089 
2090     val = ldl_le_p(vdev->config + addr);
2091     return val;
2092 }
2093 
2094 void virtio_config_modern_writeb(VirtIODevice *vdev,
2095                                  uint32_t addr, uint32_t data)
2096 {
2097     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2098     uint8_t val = data;
2099 
2100     if (addr + sizeof(val) > vdev->config_len) {
2101         return;
2102     }
2103 
2104     stb_p(vdev->config + addr, val);
2105 
2106     if (k->set_config) {
2107         k->set_config(vdev, vdev->config);
2108     }
2109 }
2110 
2111 void virtio_config_modern_writew(VirtIODevice *vdev,
2112                                  uint32_t addr, uint32_t data)
2113 {
2114     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2115     uint16_t val = data;
2116 
2117     if (addr + sizeof(val) > vdev->config_len) {
2118         return;
2119     }
2120 
2121     stw_le_p(vdev->config + addr, val);
2122 
2123     if (k->set_config) {
2124         k->set_config(vdev, vdev->config);
2125     }
2126 }
2127 
2128 void virtio_config_modern_writel(VirtIODevice *vdev,
2129                                  uint32_t addr, uint32_t data)
2130 {
2131     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2132     uint32_t val = data;
2133 
2134     if (addr + sizeof(val) > vdev->config_len) {
2135         return;
2136     }
2137 
2138     stl_le_p(vdev->config + addr, val);
2139 
2140     if (k->set_config) {
2141         k->set_config(vdev, vdev->config);
2142     }
2143 }
2144 
2145 void virtio_queue_set_addr(VirtIODevice *vdev, int n, hwaddr addr)
2146 {
2147     if (!vdev->vq[n].vring.num) {
2148         return;
2149     }
2150     vdev->vq[n].vring.desc = addr;
2151     virtio_queue_update_rings(vdev, n);
2152 }
2153 
2154 hwaddr virtio_queue_get_addr(VirtIODevice *vdev, int n)
2155 {
2156     return vdev->vq[n].vring.desc;
2157 }
2158 
2159 void virtio_queue_set_rings(VirtIODevice *vdev, int n, hwaddr desc,
2160                             hwaddr avail, hwaddr used)
2161 {
2162     if (!vdev->vq[n].vring.num) {
2163         return;
2164     }
2165     vdev->vq[n].vring.desc = desc;
2166     vdev->vq[n].vring.avail = avail;
2167     vdev->vq[n].vring.used = used;
2168     virtio_init_region_cache(vdev, n);
2169 }
2170 
2171 void virtio_queue_set_num(VirtIODevice *vdev, int n, int num)
2172 {
2173     /* Don't allow guest to flip queue between existent and
2174      * nonexistent states, or to set it to an invalid size.
2175      */
2176     if (!!num != !!vdev->vq[n].vring.num ||
2177         num > VIRTQUEUE_MAX_SIZE ||
2178         num < 0) {
2179         return;
2180     }
2181     vdev->vq[n].vring.num = num;
2182 }
2183 
2184 VirtQueue *virtio_vector_first_queue(VirtIODevice *vdev, uint16_t vector)
2185 {
2186     return QLIST_FIRST(&vdev->vector_queues[vector]);
2187 }
2188 
2189 VirtQueue *virtio_vector_next_queue(VirtQueue *vq)
2190 {
2191     return QLIST_NEXT(vq, node);
2192 }
2193 
2194 int virtio_queue_get_num(VirtIODevice *vdev, int n)
2195 {
2196     return vdev->vq[n].vring.num;
2197 }
2198 
2199 int virtio_queue_get_max_num(VirtIODevice *vdev, int n)
2200 {
2201     return vdev->vq[n].vring.num_default;
2202 }
2203 
2204 int virtio_get_num_queues(VirtIODevice *vdev)
2205 {
2206     int i;
2207 
2208     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2209         if (!virtio_queue_get_num(vdev, i)) {
2210             break;
2211         }
2212     }
2213 
2214     return i;
2215 }
2216 
2217 void virtio_queue_set_align(VirtIODevice *vdev, int n, int align)
2218 {
2219     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2220     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
2221 
2222     /* virtio-1 compliant devices cannot change the alignment */
2223     if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
2224         error_report("tried to modify queue alignment for virtio-1 device");
2225         return;
2226     }
2227     /* Check that the transport told us it was going to do this
2228      * (so a buggy transport will immediately assert rather than
2229      * silently failing to migrate this state)
2230      */
2231     assert(k->has_variable_vring_alignment);
2232 
2233     if (align) {
2234         vdev->vq[n].vring.align = align;
2235         virtio_queue_update_rings(vdev, n);
2236     }
2237 }
2238 
2239 static bool virtio_queue_notify_aio_vq(VirtQueue *vq)
2240 {
2241     bool ret = false;
2242 
2243     if (vq->vring.desc && vq->handle_aio_output) {
2244         VirtIODevice *vdev = vq->vdev;
2245 
2246         trace_virtio_queue_notify(vdev, vq - vdev->vq, vq);
2247         ret = vq->handle_aio_output(vdev, vq);
2248 
2249         if (unlikely(vdev->start_on_kick)) {
2250             virtio_set_started(vdev, true);
2251         }
2252     }
2253 
2254     return ret;
2255 }
2256 
2257 static void virtio_queue_notify_vq(VirtQueue *vq)
2258 {
2259     if (vq->vring.desc && vq->handle_output) {
2260         VirtIODevice *vdev = vq->vdev;
2261 
2262         if (unlikely(vdev->broken)) {
2263             return;
2264         }
2265 
2266         trace_virtio_queue_notify(vdev, vq - vdev->vq, vq);
2267         vq->handle_output(vdev, vq);
2268 
2269         if (unlikely(vdev->start_on_kick)) {
2270             virtio_set_started(vdev, true);
2271         }
2272     }
2273 }
2274 
2275 void virtio_queue_notify(VirtIODevice *vdev, int n)
2276 {
2277     VirtQueue *vq = &vdev->vq[n];
2278 
2279     if (unlikely(!vq->vring.desc || vdev->broken)) {
2280         return;
2281     }
2282 
2283     trace_virtio_queue_notify(vdev, vq - vdev->vq, vq);
2284     if (vq->handle_aio_output) {
2285         event_notifier_set(&vq->host_notifier);
2286     } else if (vq->handle_output) {
2287         vq->handle_output(vdev, vq);
2288 
2289         if (unlikely(vdev->start_on_kick)) {
2290             virtio_set_started(vdev, true);
2291         }
2292     }
2293 }
2294 
2295 uint16_t virtio_queue_vector(VirtIODevice *vdev, int n)
2296 {
2297     return n < VIRTIO_QUEUE_MAX ? vdev->vq[n].vector :
2298         VIRTIO_NO_VECTOR;
2299 }
2300 
2301 void virtio_queue_set_vector(VirtIODevice *vdev, int n, uint16_t vector)
2302 {
2303     VirtQueue *vq = &vdev->vq[n];
2304 
2305     if (n < VIRTIO_QUEUE_MAX) {
2306         if (vdev->vector_queues &&
2307             vdev->vq[n].vector != VIRTIO_NO_VECTOR) {
2308             QLIST_REMOVE(vq, node);
2309         }
2310         vdev->vq[n].vector = vector;
2311         if (vdev->vector_queues &&
2312             vector != VIRTIO_NO_VECTOR) {
2313             QLIST_INSERT_HEAD(&vdev->vector_queues[vector], vq, node);
2314         }
2315     }
2316 }
2317 
2318 VirtQueue *virtio_add_queue(VirtIODevice *vdev, int queue_size,
2319                             VirtIOHandleOutput handle_output)
2320 {
2321     int i;
2322 
2323     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2324         if (vdev->vq[i].vring.num == 0)
2325             break;
2326     }
2327 
2328     if (i == VIRTIO_QUEUE_MAX || queue_size > VIRTQUEUE_MAX_SIZE)
2329         abort();
2330 
2331     vdev->vq[i].vring.num = queue_size;
2332     vdev->vq[i].vring.num_default = queue_size;
2333     vdev->vq[i].vring.align = VIRTIO_PCI_VRING_ALIGN;
2334     vdev->vq[i].handle_output = handle_output;
2335     vdev->vq[i].handle_aio_output = NULL;
2336     vdev->vq[i].used_elems = g_malloc0(sizeof(VirtQueueElement) *
2337                                        queue_size);
2338 
2339     return &vdev->vq[i];
2340 }
2341 
2342 void virtio_del_queue(VirtIODevice *vdev, int n)
2343 {
2344     if (n < 0 || n >= VIRTIO_QUEUE_MAX) {
2345         abort();
2346     }
2347 
2348     vdev->vq[n].vring.num = 0;
2349     vdev->vq[n].vring.num_default = 0;
2350     vdev->vq[n].handle_output = NULL;
2351     vdev->vq[n].handle_aio_output = NULL;
2352     g_free(vdev->vq[n].used_elems);
2353 }
2354 
2355 static void virtio_set_isr(VirtIODevice *vdev, int value)
2356 {
2357     uint8_t old = atomic_read(&vdev->isr);
2358 
2359     /* Do not write ISR if it does not change, so that its cacheline remains
2360      * shared in the common case where the guest does not read it.
2361      */
2362     if ((old & value) != value) {
2363         atomic_or(&vdev->isr, value);
2364     }
2365 }
2366 
2367 static bool virtio_split_should_notify(VirtIODevice *vdev, VirtQueue *vq)
2368 {
2369     uint16_t old, new;
2370     bool v;
2371     /* We need to expose used array entries before checking used event. */
2372     smp_mb();
2373     /* Always notify when queue is empty (when feature acknowledge) */
2374     if (virtio_vdev_has_feature(vdev, VIRTIO_F_NOTIFY_ON_EMPTY) &&
2375         !vq->inuse && virtio_queue_empty(vq)) {
2376         return true;
2377     }
2378 
2379     if (!virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX)) {
2380         return !(vring_avail_flags(vq) & VRING_AVAIL_F_NO_INTERRUPT);
2381     }
2382 
2383     v = vq->signalled_used_valid;
2384     vq->signalled_used_valid = true;
2385     old = vq->signalled_used;
2386     new = vq->signalled_used = vq->used_idx;
2387     return !v || vring_need_event(vring_get_used_event(vq), new, old);
2388 }
2389 
2390 static bool vring_packed_need_event(VirtQueue *vq, bool wrap,
2391                                     uint16_t off_wrap, uint16_t new,
2392                                     uint16_t old)
2393 {
2394     int off = off_wrap & ~(1 << 15);
2395 
2396     if (wrap != off_wrap >> 15) {
2397         off -= vq->vring.num;
2398     }
2399 
2400     return vring_need_event(off, new, old);
2401 }
2402 
2403 static bool virtio_packed_should_notify(VirtIODevice *vdev, VirtQueue *vq)
2404 {
2405     VRingPackedDescEvent e;
2406     uint16_t old, new;
2407     bool v;
2408     VRingMemoryRegionCaches *caches;
2409 
2410     caches = vring_get_region_caches(vq);
2411     vring_packed_event_read(vdev, &caches->avail, &e);
2412 
2413     old = vq->signalled_used;
2414     new = vq->signalled_used = vq->used_idx;
2415     v = vq->signalled_used_valid;
2416     vq->signalled_used_valid = true;
2417 
2418     if (e.flags == VRING_PACKED_EVENT_FLAG_DISABLE) {
2419         return false;
2420     } else if (e.flags == VRING_PACKED_EVENT_FLAG_ENABLE) {
2421         return true;
2422     }
2423 
2424     return !v || vring_packed_need_event(vq, vq->used_wrap_counter,
2425                                          e.off_wrap, new, old);
2426 }
2427 
2428 /* Called within rcu_read_lock().  */
2429 static bool virtio_should_notify(VirtIODevice *vdev, VirtQueue *vq)
2430 {
2431     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
2432         return virtio_packed_should_notify(vdev, vq);
2433     } else {
2434         return virtio_split_should_notify(vdev, vq);
2435     }
2436 }
2437 
2438 void virtio_notify_irqfd(VirtIODevice *vdev, VirtQueue *vq)
2439 {
2440     bool should_notify;
2441     rcu_read_lock();
2442     should_notify = virtio_should_notify(vdev, vq);
2443     rcu_read_unlock();
2444 
2445     if (!should_notify) {
2446         return;
2447     }
2448 
2449     trace_virtio_notify_irqfd(vdev, vq);
2450 
2451     /*
2452      * virtio spec 1.0 says ISR bit 0 should be ignored with MSI, but
2453      * windows drivers included in virtio-win 1.8.0 (circa 2015) are
2454      * incorrectly polling this bit during crashdump and hibernation
2455      * in MSI mode, causing a hang if this bit is never updated.
2456      * Recent releases of Windows do not really shut down, but rather
2457      * log out and hibernate to make the next startup faster.  Hence,
2458      * this manifested as a more serious hang during shutdown with
2459      *
2460      * Next driver release from 2016 fixed this problem, so working around it
2461      * is not a must, but it's easy to do so let's do it here.
2462      *
2463      * Note: it's safe to update ISR from any thread as it was switched
2464      * to an atomic operation.
2465      */
2466     virtio_set_isr(vq->vdev, 0x1);
2467     event_notifier_set(&vq->guest_notifier);
2468 }
2469 
2470 static void virtio_irq(VirtQueue *vq)
2471 {
2472     virtio_set_isr(vq->vdev, 0x1);
2473     virtio_notify_vector(vq->vdev, vq->vector);
2474 }
2475 
2476 void virtio_notify(VirtIODevice *vdev, VirtQueue *vq)
2477 {
2478     bool should_notify;
2479     rcu_read_lock();
2480     should_notify = virtio_should_notify(vdev, vq);
2481     rcu_read_unlock();
2482 
2483     if (!should_notify) {
2484         return;
2485     }
2486 
2487     trace_virtio_notify(vdev, vq);
2488     virtio_irq(vq);
2489 }
2490 
2491 void virtio_notify_config(VirtIODevice *vdev)
2492 {
2493     if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK))
2494         return;
2495 
2496     virtio_set_isr(vdev, 0x3);
2497     vdev->generation++;
2498     virtio_notify_vector(vdev, vdev->config_vector);
2499 }
2500 
2501 static bool virtio_device_endian_needed(void *opaque)
2502 {
2503     VirtIODevice *vdev = opaque;
2504 
2505     assert(vdev->device_endian != VIRTIO_DEVICE_ENDIAN_UNKNOWN);
2506     if (!virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
2507         return vdev->device_endian != virtio_default_endian();
2508     }
2509     /* Devices conforming to VIRTIO 1.0 or later are always LE. */
2510     return vdev->device_endian != VIRTIO_DEVICE_ENDIAN_LITTLE;
2511 }
2512 
2513 static bool virtio_64bit_features_needed(void *opaque)
2514 {
2515     VirtIODevice *vdev = opaque;
2516 
2517     return (vdev->host_features >> 32) != 0;
2518 }
2519 
2520 static bool virtio_virtqueue_needed(void *opaque)
2521 {
2522     VirtIODevice *vdev = opaque;
2523 
2524     return virtio_host_has_feature(vdev, VIRTIO_F_VERSION_1);
2525 }
2526 
2527 static bool virtio_packed_virtqueue_needed(void *opaque)
2528 {
2529     VirtIODevice *vdev = opaque;
2530 
2531     return virtio_host_has_feature(vdev, VIRTIO_F_RING_PACKED);
2532 }
2533 
2534 static bool virtio_ringsize_needed(void *opaque)
2535 {
2536     VirtIODevice *vdev = opaque;
2537     int i;
2538 
2539     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2540         if (vdev->vq[i].vring.num != vdev->vq[i].vring.num_default) {
2541             return true;
2542         }
2543     }
2544     return false;
2545 }
2546 
2547 static bool virtio_extra_state_needed(void *opaque)
2548 {
2549     VirtIODevice *vdev = opaque;
2550     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2551     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
2552 
2553     return k->has_extra_state &&
2554         k->has_extra_state(qbus->parent);
2555 }
2556 
2557 static bool virtio_broken_needed(void *opaque)
2558 {
2559     VirtIODevice *vdev = opaque;
2560 
2561     return vdev->broken;
2562 }
2563 
2564 static bool virtio_started_needed(void *opaque)
2565 {
2566     VirtIODevice *vdev = opaque;
2567 
2568     return vdev->started;
2569 }
2570 
2571 static const VMStateDescription vmstate_virtqueue = {
2572     .name = "virtqueue_state",
2573     .version_id = 1,
2574     .minimum_version_id = 1,
2575     .fields = (VMStateField[]) {
2576         VMSTATE_UINT64(vring.avail, struct VirtQueue),
2577         VMSTATE_UINT64(vring.used, struct VirtQueue),
2578         VMSTATE_END_OF_LIST()
2579     }
2580 };
2581 
2582 static const VMStateDescription vmstate_packed_virtqueue = {
2583     .name = "packed_virtqueue_state",
2584     .version_id = 1,
2585     .minimum_version_id = 1,
2586     .fields = (VMStateField[]) {
2587         VMSTATE_UINT16(last_avail_idx, struct VirtQueue),
2588         VMSTATE_BOOL(last_avail_wrap_counter, struct VirtQueue),
2589         VMSTATE_UINT16(used_idx, struct VirtQueue),
2590         VMSTATE_BOOL(used_wrap_counter, struct VirtQueue),
2591         VMSTATE_UINT32(inuse, struct VirtQueue),
2592         VMSTATE_END_OF_LIST()
2593     }
2594 };
2595 
2596 static const VMStateDescription vmstate_virtio_virtqueues = {
2597     .name = "virtio/virtqueues",
2598     .version_id = 1,
2599     .minimum_version_id = 1,
2600     .needed = &virtio_virtqueue_needed,
2601     .fields = (VMStateField[]) {
2602         VMSTATE_STRUCT_VARRAY_POINTER_KNOWN(vq, struct VirtIODevice,
2603                       VIRTIO_QUEUE_MAX, 0, vmstate_virtqueue, VirtQueue),
2604         VMSTATE_END_OF_LIST()
2605     }
2606 };
2607 
2608 static const VMStateDescription vmstate_virtio_packed_virtqueues = {
2609     .name = "virtio/packed_virtqueues",
2610     .version_id = 1,
2611     .minimum_version_id = 1,
2612     .needed = &virtio_packed_virtqueue_needed,
2613     .fields = (VMStateField[]) {
2614         VMSTATE_STRUCT_VARRAY_POINTER_KNOWN(vq, struct VirtIODevice,
2615                       VIRTIO_QUEUE_MAX, 0, vmstate_packed_virtqueue, VirtQueue),
2616         VMSTATE_END_OF_LIST()
2617     }
2618 };
2619 
2620 static const VMStateDescription vmstate_ringsize = {
2621     .name = "ringsize_state",
2622     .version_id = 1,
2623     .minimum_version_id = 1,
2624     .fields = (VMStateField[]) {
2625         VMSTATE_UINT32(vring.num_default, struct VirtQueue),
2626         VMSTATE_END_OF_LIST()
2627     }
2628 };
2629 
2630 static const VMStateDescription vmstate_virtio_ringsize = {
2631     .name = "virtio/ringsize",
2632     .version_id = 1,
2633     .minimum_version_id = 1,
2634     .needed = &virtio_ringsize_needed,
2635     .fields = (VMStateField[]) {
2636         VMSTATE_STRUCT_VARRAY_POINTER_KNOWN(vq, struct VirtIODevice,
2637                       VIRTIO_QUEUE_MAX, 0, vmstate_ringsize, VirtQueue),
2638         VMSTATE_END_OF_LIST()
2639     }
2640 };
2641 
2642 static int get_extra_state(QEMUFile *f, void *pv, size_t size,
2643                            const VMStateField *field)
2644 {
2645     VirtIODevice *vdev = pv;
2646     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2647     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
2648 
2649     if (!k->load_extra_state) {
2650         return -1;
2651     } else {
2652         return k->load_extra_state(qbus->parent, f);
2653     }
2654 }
2655 
2656 static int put_extra_state(QEMUFile *f, void *pv, size_t size,
2657                            const VMStateField *field, QJSON *vmdesc)
2658 {
2659     VirtIODevice *vdev = pv;
2660     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2661     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
2662 
2663     k->save_extra_state(qbus->parent, f);
2664     return 0;
2665 }
2666 
2667 static const VMStateInfo vmstate_info_extra_state = {
2668     .name = "virtqueue_extra_state",
2669     .get = get_extra_state,
2670     .put = put_extra_state,
2671 };
2672 
2673 static const VMStateDescription vmstate_virtio_extra_state = {
2674     .name = "virtio/extra_state",
2675     .version_id = 1,
2676     .minimum_version_id = 1,
2677     .needed = &virtio_extra_state_needed,
2678     .fields = (VMStateField[]) {
2679         {
2680             .name         = "extra_state",
2681             .version_id   = 0,
2682             .field_exists = NULL,
2683             .size         = 0,
2684             .info         = &vmstate_info_extra_state,
2685             .flags        = VMS_SINGLE,
2686             .offset       = 0,
2687         },
2688         VMSTATE_END_OF_LIST()
2689     }
2690 };
2691 
2692 static const VMStateDescription vmstate_virtio_device_endian = {
2693     .name = "virtio/device_endian",
2694     .version_id = 1,
2695     .minimum_version_id = 1,
2696     .needed = &virtio_device_endian_needed,
2697     .fields = (VMStateField[]) {
2698         VMSTATE_UINT8(device_endian, VirtIODevice),
2699         VMSTATE_END_OF_LIST()
2700     }
2701 };
2702 
2703 static const VMStateDescription vmstate_virtio_64bit_features = {
2704     .name = "virtio/64bit_features",
2705     .version_id = 1,
2706     .minimum_version_id = 1,
2707     .needed = &virtio_64bit_features_needed,
2708     .fields = (VMStateField[]) {
2709         VMSTATE_UINT64(guest_features, VirtIODevice),
2710         VMSTATE_END_OF_LIST()
2711     }
2712 };
2713 
2714 static const VMStateDescription vmstate_virtio_broken = {
2715     .name = "virtio/broken",
2716     .version_id = 1,
2717     .minimum_version_id = 1,
2718     .needed = &virtio_broken_needed,
2719     .fields = (VMStateField[]) {
2720         VMSTATE_BOOL(broken, VirtIODevice),
2721         VMSTATE_END_OF_LIST()
2722     }
2723 };
2724 
2725 static const VMStateDescription vmstate_virtio_started = {
2726     .name = "virtio/started",
2727     .version_id = 1,
2728     .minimum_version_id = 1,
2729     .needed = &virtio_started_needed,
2730     .fields = (VMStateField[]) {
2731         VMSTATE_BOOL(started, VirtIODevice),
2732         VMSTATE_END_OF_LIST()
2733     }
2734 };
2735 
2736 static const VMStateDescription vmstate_virtio = {
2737     .name = "virtio",
2738     .version_id = 1,
2739     .minimum_version_id = 1,
2740     .minimum_version_id_old = 1,
2741     .fields = (VMStateField[]) {
2742         VMSTATE_END_OF_LIST()
2743     },
2744     .subsections = (const VMStateDescription*[]) {
2745         &vmstate_virtio_device_endian,
2746         &vmstate_virtio_64bit_features,
2747         &vmstate_virtio_virtqueues,
2748         &vmstate_virtio_ringsize,
2749         &vmstate_virtio_broken,
2750         &vmstate_virtio_extra_state,
2751         &vmstate_virtio_started,
2752         &vmstate_virtio_packed_virtqueues,
2753         NULL
2754     }
2755 };
2756 
2757 int virtio_save(VirtIODevice *vdev, QEMUFile *f)
2758 {
2759     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2760     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
2761     VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev);
2762     uint32_t guest_features_lo = (vdev->guest_features & 0xffffffff);
2763     int i;
2764 
2765     if (k->save_config) {
2766         k->save_config(qbus->parent, f);
2767     }
2768 
2769     qemu_put_8s(f, &vdev->status);
2770     qemu_put_8s(f, &vdev->isr);
2771     qemu_put_be16s(f, &vdev->queue_sel);
2772     qemu_put_be32s(f, &guest_features_lo);
2773     qemu_put_be32(f, vdev->config_len);
2774     qemu_put_buffer(f, vdev->config, vdev->config_len);
2775 
2776     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2777         if (vdev->vq[i].vring.num == 0)
2778             break;
2779     }
2780 
2781     qemu_put_be32(f, i);
2782 
2783     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2784         if (vdev->vq[i].vring.num == 0)
2785             break;
2786 
2787         qemu_put_be32(f, vdev->vq[i].vring.num);
2788         if (k->has_variable_vring_alignment) {
2789             qemu_put_be32(f, vdev->vq[i].vring.align);
2790         }
2791         /*
2792          * Save desc now, the rest of the ring addresses are saved in
2793          * subsections for VIRTIO-1 devices.
2794          */
2795         qemu_put_be64(f, vdev->vq[i].vring.desc);
2796         qemu_put_be16s(f, &vdev->vq[i].last_avail_idx);
2797         if (k->save_queue) {
2798             k->save_queue(qbus->parent, i, f);
2799         }
2800     }
2801 
2802     if (vdc->save != NULL) {
2803         vdc->save(vdev, f);
2804     }
2805 
2806     if (vdc->vmsd) {
2807         int ret = vmstate_save_state(f, vdc->vmsd, vdev, NULL);
2808         if (ret) {
2809             return ret;
2810         }
2811     }
2812 
2813     /* Subsections */
2814     return vmstate_save_state(f, &vmstate_virtio, vdev, NULL);
2815 }
2816 
2817 /* A wrapper for use as a VMState .put function */
2818 static int virtio_device_put(QEMUFile *f, void *opaque, size_t size,
2819                               const VMStateField *field, QJSON *vmdesc)
2820 {
2821     return virtio_save(VIRTIO_DEVICE(opaque), f);
2822 }
2823 
2824 /* A wrapper for use as a VMState .get function */
2825 static int virtio_device_get(QEMUFile *f, void *opaque, size_t size,
2826                              const VMStateField *field)
2827 {
2828     VirtIODevice *vdev = VIRTIO_DEVICE(opaque);
2829     DeviceClass *dc = DEVICE_CLASS(VIRTIO_DEVICE_GET_CLASS(vdev));
2830 
2831     return virtio_load(vdev, f, dc->vmsd->version_id);
2832 }
2833 
2834 const VMStateInfo  virtio_vmstate_info = {
2835     .name = "virtio",
2836     .get = virtio_device_get,
2837     .put = virtio_device_put,
2838 };
2839 
2840 static int virtio_set_features_nocheck(VirtIODevice *vdev, uint64_t val)
2841 {
2842     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2843     bool bad = (val & ~(vdev->host_features)) != 0;
2844 
2845     val &= vdev->host_features;
2846     if (k->set_features) {
2847         k->set_features(vdev, val);
2848     }
2849     vdev->guest_features = val;
2850     return bad ? -1 : 0;
2851 }
2852 
2853 int virtio_set_features(VirtIODevice *vdev, uint64_t val)
2854 {
2855     int ret;
2856     /*
2857      * The driver must not attempt to set features after feature negotiation
2858      * has finished.
2859      */
2860     if (vdev->status & VIRTIO_CONFIG_S_FEATURES_OK) {
2861         return -EINVAL;
2862     }
2863     ret = virtio_set_features_nocheck(vdev, val);
2864     if (!ret) {
2865         if (virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX)) {
2866             /* VIRTIO_RING_F_EVENT_IDX changes the size of the caches.  */
2867             int i;
2868             for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2869                 if (vdev->vq[i].vring.num != 0) {
2870                     virtio_init_region_cache(vdev, i);
2871                 }
2872             }
2873         }
2874 
2875         if (!virtio_device_started(vdev, vdev->status) &&
2876             !virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
2877             vdev->start_on_kick = true;
2878         }
2879     }
2880     return ret;
2881 }
2882 
2883 size_t virtio_feature_get_config_size(VirtIOFeature *feature_sizes,
2884                                       uint64_t host_features)
2885 {
2886     size_t config_size = 0;
2887     int i;
2888 
2889     for (i = 0; feature_sizes[i].flags != 0; i++) {
2890         if (host_features & feature_sizes[i].flags) {
2891             config_size = MAX(feature_sizes[i].end, config_size);
2892         }
2893     }
2894 
2895     return config_size;
2896 }
2897 
2898 int virtio_load(VirtIODevice *vdev, QEMUFile *f, int version_id)
2899 {
2900     int i, ret;
2901     int32_t config_len;
2902     uint32_t num;
2903     uint32_t features;
2904     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2905     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
2906     VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev);
2907 
2908     /*
2909      * We poison the endianness to ensure it does not get used before
2910      * subsections have been loaded.
2911      */
2912     vdev->device_endian = VIRTIO_DEVICE_ENDIAN_UNKNOWN;
2913 
2914     if (k->load_config) {
2915         ret = k->load_config(qbus->parent, f);
2916         if (ret)
2917             return ret;
2918     }
2919 
2920     qemu_get_8s(f, &vdev->status);
2921     qemu_get_8s(f, &vdev->isr);
2922     qemu_get_be16s(f, &vdev->queue_sel);
2923     if (vdev->queue_sel >= VIRTIO_QUEUE_MAX) {
2924         return -1;
2925     }
2926     qemu_get_be32s(f, &features);
2927 
2928     /*
2929      * Temporarily set guest_features low bits - needed by
2930      * virtio net load code testing for VIRTIO_NET_F_CTRL_GUEST_OFFLOADS
2931      * VIRTIO_NET_F_GUEST_ANNOUNCE and VIRTIO_NET_F_CTRL_VQ.
2932      *
2933      * Note: devices should always test host features in future - don't create
2934      * new dependencies like this.
2935      */
2936     vdev->guest_features = features;
2937 
2938     config_len = qemu_get_be32(f);
2939 
2940     /*
2941      * There are cases where the incoming config can be bigger or smaller
2942      * than what we have; so load what we have space for, and skip
2943      * any excess that's in the stream.
2944      */
2945     qemu_get_buffer(f, vdev->config, MIN(config_len, vdev->config_len));
2946 
2947     while (config_len > vdev->config_len) {
2948         qemu_get_byte(f);
2949         config_len--;
2950     }
2951 
2952     num = qemu_get_be32(f);
2953 
2954     if (num > VIRTIO_QUEUE_MAX) {
2955         error_report("Invalid number of virtqueues: 0x%x", num);
2956         return -1;
2957     }
2958 
2959     for (i = 0; i < num; i++) {
2960         vdev->vq[i].vring.num = qemu_get_be32(f);
2961         if (k->has_variable_vring_alignment) {
2962             vdev->vq[i].vring.align = qemu_get_be32(f);
2963         }
2964         vdev->vq[i].vring.desc = qemu_get_be64(f);
2965         qemu_get_be16s(f, &vdev->vq[i].last_avail_idx);
2966         vdev->vq[i].signalled_used_valid = false;
2967         vdev->vq[i].notification = true;
2968 
2969         if (!vdev->vq[i].vring.desc && vdev->vq[i].last_avail_idx) {
2970             error_report("VQ %d address 0x0 "
2971                          "inconsistent with Host index 0x%x",
2972                          i, vdev->vq[i].last_avail_idx);
2973             return -1;
2974         }
2975         if (k->load_queue) {
2976             ret = k->load_queue(qbus->parent, i, f);
2977             if (ret)
2978                 return ret;
2979         }
2980     }
2981 
2982     virtio_notify_vector(vdev, VIRTIO_NO_VECTOR);
2983 
2984     if (vdc->load != NULL) {
2985         ret = vdc->load(vdev, f, version_id);
2986         if (ret) {
2987             return ret;
2988         }
2989     }
2990 
2991     if (vdc->vmsd) {
2992         ret = vmstate_load_state(f, vdc->vmsd, vdev, version_id);
2993         if (ret) {
2994             return ret;
2995         }
2996     }
2997 
2998     /* Subsections */
2999     ret = vmstate_load_state(f, &vmstate_virtio, vdev, 1);
3000     if (ret) {
3001         return ret;
3002     }
3003 
3004     if (vdev->device_endian == VIRTIO_DEVICE_ENDIAN_UNKNOWN) {
3005         vdev->device_endian = virtio_default_endian();
3006     }
3007 
3008     if (virtio_64bit_features_needed(vdev)) {
3009         /*
3010          * Subsection load filled vdev->guest_features.  Run them
3011          * through virtio_set_features to sanity-check them against
3012          * host_features.
3013          */
3014         uint64_t features64 = vdev->guest_features;
3015         if (virtio_set_features_nocheck(vdev, features64) < 0) {
3016             error_report("Features 0x%" PRIx64 " unsupported. "
3017                          "Allowed features: 0x%" PRIx64,
3018                          features64, vdev->host_features);
3019             return -1;
3020         }
3021     } else {
3022         if (virtio_set_features_nocheck(vdev, features) < 0) {
3023             error_report("Features 0x%x unsupported. "
3024                          "Allowed features: 0x%" PRIx64,
3025                          features, vdev->host_features);
3026             return -1;
3027         }
3028     }
3029 
3030     if (!virtio_device_started(vdev, vdev->status) &&
3031         !virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
3032         vdev->start_on_kick = true;
3033     }
3034 
3035     rcu_read_lock();
3036     for (i = 0; i < num; i++) {
3037         if (vdev->vq[i].vring.desc) {
3038             uint16_t nheads;
3039 
3040             /*
3041              * VIRTIO-1 devices migrate desc, used, and avail ring addresses so
3042              * only the region cache needs to be set up.  Legacy devices need
3043              * to calculate used and avail ring addresses based on the desc
3044              * address.
3045              */
3046             if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
3047                 virtio_init_region_cache(vdev, i);
3048             } else {
3049                 virtio_queue_update_rings(vdev, i);
3050             }
3051 
3052             if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3053                 vdev->vq[i].shadow_avail_idx = vdev->vq[i].last_avail_idx;
3054                 vdev->vq[i].shadow_avail_wrap_counter =
3055                                         vdev->vq[i].last_avail_wrap_counter;
3056                 continue;
3057             }
3058 
3059             nheads = vring_avail_idx(&vdev->vq[i]) - vdev->vq[i].last_avail_idx;
3060             /* Check it isn't doing strange things with descriptor numbers. */
3061             if (nheads > vdev->vq[i].vring.num) {
3062                 error_report("VQ %d size 0x%x Guest index 0x%x "
3063                              "inconsistent with Host index 0x%x: delta 0x%x",
3064                              i, vdev->vq[i].vring.num,
3065                              vring_avail_idx(&vdev->vq[i]),
3066                              vdev->vq[i].last_avail_idx, nheads);
3067                 return -1;
3068             }
3069             vdev->vq[i].used_idx = vring_used_idx(&vdev->vq[i]);
3070             vdev->vq[i].shadow_avail_idx = vring_avail_idx(&vdev->vq[i]);
3071 
3072             /*
3073              * Some devices migrate VirtQueueElements that have been popped
3074              * from the avail ring but not yet returned to the used ring.
3075              * Since max ring size < UINT16_MAX it's safe to use modulo
3076              * UINT16_MAX + 1 subtraction.
3077              */
3078             vdev->vq[i].inuse = (uint16_t)(vdev->vq[i].last_avail_idx -
3079                                 vdev->vq[i].used_idx);
3080             if (vdev->vq[i].inuse > vdev->vq[i].vring.num) {
3081                 error_report("VQ %d size 0x%x < last_avail_idx 0x%x - "
3082                              "used_idx 0x%x",
3083                              i, vdev->vq[i].vring.num,
3084                              vdev->vq[i].last_avail_idx,
3085                              vdev->vq[i].used_idx);
3086                 return -1;
3087             }
3088         }
3089     }
3090     rcu_read_unlock();
3091 
3092     return 0;
3093 }
3094 
3095 void virtio_cleanup(VirtIODevice *vdev)
3096 {
3097     qemu_del_vm_change_state_handler(vdev->vmstate);
3098 }
3099 
3100 static void virtio_vmstate_change(void *opaque, int running, RunState state)
3101 {
3102     VirtIODevice *vdev = opaque;
3103     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3104     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
3105     bool backend_run = running && virtio_device_started(vdev, vdev->status);
3106     vdev->vm_running = running;
3107 
3108     if (backend_run) {
3109         virtio_set_status(vdev, vdev->status);
3110     }
3111 
3112     if (k->vmstate_change) {
3113         k->vmstate_change(qbus->parent, backend_run);
3114     }
3115 
3116     if (!backend_run) {
3117         virtio_set_status(vdev, vdev->status);
3118     }
3119 }
3120 
3121 void virtio_instance_init_common(Object *proxy_obj, void *data,
3122                                  size_t vdev_size, const char *vdev_name)
3123 {
3124     DeviceState *vdev = data;
3125 
3126     object_initialize_child(proxy_obj, "virtio-backend", vdev, vdev_size,
3127                             vdev_name, &error_abort, NULL);
3128     qdev_alias_all_properties(vdev, proxy_obj);
3129 }
3130 
3131 void virtio_init(VirtIODevice *vdev, const char *name,
3132                  uint16_t device_id, size_t config_size)
3133 {
3134     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3135     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
3136     int i;
3137     int nvectors = k->query_nvectors ? k->query_nvectors(qbus->parent) : 0;
3138 
3139     if (nvectors) {
3140         vdev->vector_queues =
3141             g_malloc0(sizeof(*vdev->vector_queues) * nvectors);
3142     }
3143 
3144     vdev->start_on_kick = false;
3145     vdev->started = false;
3146     vdev->device_id = device_id;
3147     vdev->status = 0;
3148     atomic_set(&vdev->isr, 0);
3149     vdev->queue_sel = 0;
3150     vdev->config_vector = VIRTIO_NO_VECTOR;
3151     vdev->vq = g_malloc0(sizeof(VirtQueue) * VIRTIO_QUEUE_MAX);
3152     vdev->vm_running = runstate_is_running();
3153     vdev->broken = false;
3154     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
3155         vdev->vq[i].vector = VIRTIO_NO_VECTOR;
3156         vdev->vq[i].vdev = vdev;
3157         vdev->vq[i].queue_index = i;
3158     }
3159 
3160     vdev->name = name;
3161     vdev->config_len = config_size;
3162     if (vdev->config_len) {
3163         vdev->config = g_malloc0(config_size);
3164     } else {
3165         vdev->config = NULL;
3166     }
3167     vdev->vmstate = qdev_add_vm_change_state_handler(DEVICE(vdev),
3168             virtio_vmstate_change, vdev);
3169     vdev->device_endian = virtio_default_endian();
3170     vdev->use_guest_notifier_mask = true;
3171 }
3172 
3173 hwaddr virtio_queue_get_desc_addr(VirtIODevice *vdev, int n)
3174 {
3175     return vdev->vq[n].vring.desc;
3176 }
3177 
3178 bool virtio_queue_enabled(VirtIODevice *vdev, int n)
3179 {
3180     return virtio_queue_get_desc_addr(vdev, n) != 0;
3181 }
3182 
3183 hwaddr virtio_queue_get_avail_addr(VirtIODevice *vdev, int n)
3184 {
3185     return vdev->vq[n].vring.avail;
3186 }
3187 
3188 hwaddr virtio_queue_get_used_addr(VirtIODevice *vdev, int n)
3189 {
3190     return vdev->vq[n].vring.used;
3191 }
3192 
3193 hwaddr virtio_queue_get_desc_size(VirtIODevice *vdev, int n)
3194 {
3195     return sizeof(VRingDesc) * vdev->vq[n].vring.num;
3196 }
3197 
3198 hwaddr virtio_queue_get_avail_size(VirtIODevice *vdev, int n)
3199 {
3200     int s;
3201 
3202     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3203         return sizeof(struct VRingPackedDescEvent);
3204     }
3205 
3206     s = virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0;
3207     return offsetof(VRingAvail, ring) +
3208         sizeof(uint16_t) * vdev->vq[n].vring.num + s;
3209 }
3210 
3211 hwaddr virtio_queue_get_used_size(VirtIODevice *vdev, int n)
3212 {
3213     int s;
3214 
3215     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3216         return sizeof(struct VRingPackedDescEvent);
3217     }
3218 
3219     s = virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0;
3220     return offsetof(VRingUsed, ring) +
3221         sizeof(VRingUsedElem) * vdev->vq[n].vring.num + s;
3222 }
3223 
3224 static unsigned int virtio_queue_packed_get_last_avail_idx(VirtIODevice *vdev,
3225                                                            int n)
3226 {
3227     unsigned int avail, used;
3228 
3229     avail = vdev->vq[n].last_avail_idx;
3230     avail |= ((uint16_t)vdev->vq[n].last_avail_wrap_counter) << 15;
3231 
3232     used = vdev->vq[n].used_idx;
3233     used |= ((uint16_t)vdev->vq[n].used_wrap_counter) << 15;
3234 
3235     return avail | used << 16;
3236 }
3237 
3238 static uint16_t virtio_queue_split_get_last_avail_idx(VirtIODevice *vdev,
3239                                                       int n)
3240 {
3241     return vdev->vq[n].last_avail_idx;
3242 }
3243 
3244 unsigned int virtio_queue_get_last_avail_idx(VirtIODevice *vdev, int n)
3245 {
3246     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3247         return virtio_queue_packed_get_last_avail_idx(vdev, n);
3248     } else {
3249         return virtio_queue_split_get_last_avail_idx(vdev, n);
3250     }
3251 }
3252 
3253 static void virtio_queue_packed_set_last_avail_idx(VirtIODevice *vdev,
3254                                                    int n, unsigned int idx)
3255 {
3256     struct VirtQueue *vq = &vdev->vq[n];
3257 
3258     vq->last_avail_idx = vq->shadow_avail_idx = idx & 0x7fff;
3259     vq->last_avail_wrap_counter =
3260         vq->shadow_avail_wrap_counter = !!(idx & 0x8000);
3261     idx >>= 16;
3262     vq->used_idx = idx & 0x7ffff;
3263     vq->used_wrap_counter = !!(idx & 0x8000);
3264 }
3265 
3266 static void virtio_queue_split_set_last_avail_idx(VirtIODevice *vdev,
3267                                                   int n, unsigned int idx)
3268 {
3269         vdev->vq[n].last_avail_idx = idx;
3270         vdev->vq[n].shadow_avail_idx = idx;
3271 }
3272 
3273 void virtio_queue_set_last_avail_idx(VirtIODevice *vdev, int n,
3274                                      unsigned int idx)
3275 {
3276     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3277         virtio_queue_packed_set_last_avail_idx(vdev, n, idx);
3278     } else {
3279         virtio_queue_split_set_last_avail_idx(vdev, n, idx);
3280     }
3281 }
3282 
3283 static void virtio_queue_packed_restore_last_avail_idx(VirtIODevice *vdev,
3284                                                        int n)
3285 {
3286     /* We don't have a reference like avail idx in shared memory */
3287     return;
3288 }
3289 
3290 static void virtio_queue_split_restore_last_avail_idx(VirtIODevice *vdev,
3291                                                       int n)
3292 {
3293     rcu_read_lock();
3294     if (vdev->vq[n].vring.desc) {
3295         vdev->vq[n].last_avail_idx = vring_used_idx(&vdev->vq[n]);
3296         vdev->vq[n].shadow_avail_idx = vdev->vq[n].last_avail_idx;
3297     }
3298     rcu_read_unlock();
3299 }
3300 
3301 void virtio_queue_restore_last_avail_idx(VirtIODevice *vdev, int n)
3302 {
3303     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3304         virtio_queue_packed_restore_last_avail_idx(vdev, n);
3305     } else {
3306         virtio_queue_split_restore_last_avail_idx(vdev, n);
3307     }
3308 }
3309 
3310 static void virtio_queue_packed_update_used_idx(VirtIODevice *vdev, int n)
3311 {
3312     /* used idx was updated through set_last_avail_idx() */
3313     return;
3314 }
3315 
3316 static void virtio_split_packed_update_used_idx(VirtIODevice *vdev, int n)
3317 {
3318     rcu_read_lock();
3319     if (vdev->vq[n].vring.desc) {
3320         vdev->vq[n].used_idx = vring_used_idx(&vdev->vq[n]);
3321     }
3322     rcu_read_unlock();
3323 }
3324 
3325 void virtio_queue_update_used_idx(VirtIODevice *vdev, int n)
3326 {
3327     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3328         return virtio_queue_packed_update_used_idx(vdev, n);
3329     } else {
3330         return virtio_split_packed_update_used_idx(vdev, n);
3331     }
3332 }
3333 
3334 void virtio_queue_invalidate_signalled_used(VirtIODevice *vdev, int n)
3335 {
3336     vdev->vq[n].signalled_used_valid = false;
3337 }
3338 
3339 VirtQueue *virtio_get_queue(VirtIODevice *vdev, int n)
3340 {
3341     return vdev->vq + n;
3342 }
3343 
3344 uint16_t virtio_get_queue_index(VirtQueue *vq)
3345 {
3346     return vq->queue_index;
3347 }
3348 
3349 static void virtio_queue_guest_notifier_read(EventNotifier *n)
3350 {
3351     VirtQueue *vq = container_of(n, VirtQueue, guest_notifier);
3352     if (event_notifier_test_and_clear(n)) {
3353         virtio_irq(vq);
3354     }
3355 }
3356 
3357 void virtio_queue_set_guest_notifier_fd_handler(VirtQueue *vq, bool assign,
3358                                                 bool with_irqfd)
3359 {
3360     if (assign && !with_irqfd) {
3361         event_notifier_set_handler(&vq->guest_notifier,
3362                                    virtio_queue_guest_notifier_read);
3363     } else {
3364         event_notifier_set_handler(&vq->guest_notifier, NULL);
3365     }
3366     if (!assign) {
3367         /* Test and clear notifier before closing it,
3368          * in case poll callback didn't have time to run. */
3369         virtio_queue_guest_notifier_read(&vq->guest_notifier);
3370     }
3371 }
3372 
3373 EventNotifier *virtio_queue_get_guest_notifier(VirtQueue *vq)
3374 {
3375     return &vq->guest_notifier;
3376 }
3377 
3378 static void virtio_queue_host_notifier_aio_read(EventNotifier *n)
3379 {
3380     VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
3381     if (event_notifier_test_and_clear(n)) {
3382         virtio_queue_notify_aio_vq(vq);
3383     }
3384 }
3385 
3386 static void virtio_queue_host_notifier_aio_poll_begin(EventNotifier *n)
3387 {
3388     VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
3389 
3390     virtio_queue_set_notification(vq, 0);
3391 }
3392 
3393 static bool virtio_queue_host_notifier_aio_poll(void *opaque)
3394 {
3395     EventNotifier *n = opaque;
3396     VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
3397     bool progress;
3398 
3399     if (!vq->vring.desc || virtio_queue_empty(vq)) {
3400         return false;
3401     }
3402 
3403     progress = virtio_queue_notify_aio_vq(vq);
3404 
3405     /* In case the handler function re-enabled notifications */
3406     virtio_queue_set_notification(vq, 0);
3407     return progress;
3408 }
3409 
3410 static void virtio_queue_host_notifier_aio_poll_end(EventNotifier *n)
3411 {
3412     VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
3413 
3414     /* Caller polls once more after this to catch requests that race with us */
3415     virtio_queue_set_notification(vq, 1);
3416 }
3417 
3418 void virtio_queue_aio_set_host_notifier_handler(VirtQueue *vq, AioContext *ctx,
3419                                                 VirtIOHandleAIOOutput handle_output)
3420 {
3421     if (handle_output) {
3422         vq->handle_aio_output = handle_output;
3423         aio_set_event_notifier(ctx, &vq->host_notifier, true,
3424                                virtio_queue_host_notifier_aio_read,
3425                                virtio_queue_host_notifier_aio_poll);
3426         aio_set_event_notifier_poll(ctx, &vq->host_notifier,
3427                                     virtio_queue_host_notifier_aio_poll_begin,
3428                                     virtio_queue_host_notifier_aio_poll_end);
3429     } else {
3430         aio_set_event_notifier(ctx, &vq->host_notifier, true, NULL, NULL);
3431         /* Test and clear notifier before after disabling event,
3432          * in case poll callback didn't have time to run. */
3433         virtio_queue_host_notifier_aio_read(&vq->host_notifier);
3434         vq->handle_aio_output = NULL;
3435     }
3436 }
3437 
3438 void virtio_queue_host_notifier_read(EventNotifier *n)
3439 {
3440     VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
3441     if (event_notifier_test_and_clear(n)) {
3442         virtio_queue_notify_vq(vq);
3443     }
3444 }
3445 
3446 EventNotifier *virtio_queue_get_host_notifier(VirtQueue *vq)
3447 {
3448     return &vq->host_notifier;
3449 }
3450 
3451 int virtio_queue_set_host_notifier_mr(VirtIODevice *vdev, int n,
3452                                       MemoryRegion *mr, bool assign)
3453 {
3454     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3455     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
3456 
3457     if (k->set_host_notifier_mr) {
3458         return k->set_host_notifier_mr(qbus->parent, n, mr, assign);
3459     }
3460 
3461     return -1;
3462 }
3463 
3464 void virtio_device_set_child_bus_name(VirtIODevice *vdev, char *bus_name)
3465 {
3466     g_free(vdev->bus_name);
3467     vdev->bus_name = g_strdup(bus_name);
3468 }
3469 
3470 void GCC_FMT_ATTR(2, 3) virtio_error(VirtIODevice *vdev, const char *fmt, ...)
3471 {
3472     va_list ap;
3473 
3474     va_start(ap, fmt);
3475     error_vreport(fmt, ap);
3476     va_end(ap);
3477 
3478     if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
3479         vdev->status = vdev->status | VIRTIO_CONFIG_S_NEEDS_RESET;
3480         virtio_notify_config(vdev);
3481     }
3482 
3483     vdev->broken = true;
3484 }
3485 
3486 static void virtio_memory_listener_commit(MemoryListener *listener)
3487 {
3488     VirtIODevice *vdev = container_of(listener, VirtIODevice, listener);
3489     int i;
3490 
3491     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
3492         if (vdev->vq[i].vring.num == 0) {
3493             break;
3494         }
3495         virtio_init_region_cache(vdev, i);
3496     }
3497 }
3498 
3499 static void virtio_device_realize(DeviceState *dev, Error **errp)
3500 {
3501     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
3502     VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(dev);
3503     Error *err = NULL;
3504 
3505     /* Devices should either use vmsd or the load/save methods */
3506     assert(!vdc->vmsd || !vdc->load);
3507 
3508     if (vdc->realize != NULL) {
3509         vdc->realize(dev, &err);
3510         if (err != NULL) {
3511             error_propagate(errp, err);
3512             return;
3513         }
3514     }
3515 
3516     virtio_bus_device_plugged(vdev, &err);
3517     if (err != NULL) {
3518         error_propagate(errp, err);
3519         vdc->unrealize(dev, NULL);
3520         return;
3521     }
3522 
3523     vdev->listener.commit = virtio_memory_listener_commit;
3524     memory_listener_register(&vdev->listener, vdev->dma_as);
3525 }
3526 
3527 static void virtio_device_unrealize(DeviceState *dev, Error **errp)
3528 {
3529     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
3530     VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(dev);
3531     Error *err = NULL;
3532 
3533     virtio_bus_device_unplugged(vdev);
3534 
3535     if (vdc->unrealize != NULL) {
3536         vdc->unrealize(dev, &err);
3537         if (err != NULL) {
3538             error_propagate(errp, err);
3539             return;
3540         }
3541     }
3542 
3543     g_free(vdev->bus_name);
3544     vdev->bus_name = NULL;
3545 }
3546 
3547 static void virtio_device_free_virtqueues(VirtIODevice *vdev)
3548 {
3549     int i;
3550     if (!vdev->vq) {
3551         return;
3552     }
3553 
3554     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
3555         if (vdev->vq[i].vring.num == 0) {
3556             break;
3557         }
3558         virtio_virtqueue_reset_region_cache(&vdev->vq[i]);
3559     }
3560     g_free(vdev->vq);
3561 }
3562 
3563 static void virtio_device_instance_finalize(Object *obj)
3564 {
3565     VirtIODevice *vdev = VIRTIO_DEVICE(obj);
3566 
3567     memory_listener_unregister(&vdev->listener);
3568     virtio_device_free_virtqueues(vdev);
3569 
3570     g_free(vdev->config);
3571     g_free(vdev->vector_queues);
3572 }
3573 
3574 static Property virtio_properties[] = {
3575     DEFINE_VIRTIO_COMMON_FEATURES(VirtIODevice, host_features),
3576     DEFINE_PROP_BOOL("use-started", VirtIODevice, use_started, true),
3577     DEFINE_PROP_END_OF_LIST(),
3578 };
3579 
3580 static int virtio_device_start_ioeventfd_impl(VirtIODevice *vdev)
3581 {
3582     VirtioBusState *qbus = VIRTIO_BUS(qdev_get_parent_bus(DEVICE(vdev)));
3583     int i, n, r, err;
3584 
3585     memory_region_transaction_begin();
3586     for (n = 0; n < VIRTIO_QUEUE_MAX; n++) {
3587         VirtQueue *vq = &vdev->vq[n];
3588         if (!virtio_queue_get_num(vdev, n)) {
3589             continue;
3590         }
3591         r = virtio_bus_set_host_notifier(qbus, n, true);
3592         if (r < 0) {
3593             err = r;
3594             goto assign_error;
3595         }
3596         event_notifier_set_handler(&vq->host_notifier,
3597                                    virtio_queue_host_notifier_read);
3598     }
3599 
3600     for (n = 0; n < VIRTIO_QUEUE_MAX; n++) {
3601         /* Kick right away to begin processing requests already in vring */
3602         VirtQueue *vq = &vdev->vq[n];
3603         if (!vq->vring.num) {
3604             continue;
3605         }
3606         event_notifier_set(&vq->host_notifier);
3607     }
3608     memory_region_transaction_commit();
3609     return 0;
3610 
3611 assign_error:
3612     i = n; /* save n for a second iteration after transaction is committed. */
3613     while (--n >= 0) {
3614         VirtQueue *vq = &vdev->vq[n];
3615         if (!virtio_queue_get_num(vdev, n)) {
3616             continue;
3617         }
3618 
3619         event_notifier_set_handler(&vq->host_notifier, NULL);
3620         r = virtio_bus_set_host_notifier(qbus, n, false);
3621         assert(r >= 0);
3622     }
3623     memory_region_transaction_commit();
3624 
3625     while (--i >= 0) {
3626         if (!virtio_queue_get_num(vdev, i)) {
3627             continue;
3628         }
3629         virtio_bus_cleanup_host_notifier(qbus, i);
3630     }
3631     return err;
3632 }
3633 
3634 int virtio_device_start_ioeventfd(VirtIODevice *vdev)
3635 {
3636     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3637     VirtioBusState *vbus = VIRTIO_BUS(qbus);
3638 
3639     return virtio_bus_start_ioeventfd(vbus);
3640 }
3641 
3642 static void virtio_device_stop_ioeventfd_impl(VirtIODevice *vdev)
3643 {
3644     VirtioBusState *qbus = VIRTIO_BUS(qdev_get_parent_bus(DEVICE(vdev)));
3645     int n, r;
3646 
3647     memory_region_transaction_begin();
3648     for (n = 0; n < VIRTIO_QUEUE_MAX; n++) {
3649         VirtQueue *vq = &vdev->vq[n];
3650 
3651         if (!virtio_queue_get_num(vdev, n)) {
3652             continue;
3653         }
3654         event_notifier_set_handler(&vq->host_notifier, NULL);
3655         r = virtio_bus_set_host_notifier(qbus, n, false);
3656         assert(r >= 0);
3657     }
3658     memory_region_transaction_commit();
3659 
3660     for (n = 0; n < VIRTIO_QUEUE_MAX; n++) {
3661         if (!virtio_queue_get_num(vdev, n)) {
3662             continue;
3663         }
3664         virtio_bus_cleanup_host_notifier(qbus, n);
3665     }
3666 }
3667 
3668 int virtio_device_grab_ioeventfd(VirtIODevice *vdev)
3669 {
3670     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3671     VirtioBusState *vbus = VIRTIO_BUS(qbus);
3672 
3673     return virtio_bus_grab_ioeventfd(vbus);
3674 }
3675 
3676 void virtio_device_release_ioeventfd(VirtIODevice *vdev)
3677 {
3678     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3679     VirtioBusState *vbus = VIRTIO_BUS(qbus);
3680 
3681     virtio_bus_release_ioeventfd(vbus);
3682 }
3683 
3684 static void virtio_device_class_init(ObjectClass *klass, void *data)
3685 {
3686     /* Set the default value here. */
3687     VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
3688     DeviceClass *dc = DEVICE_CLASS(klass);
3689 
3690     dc->realize = virtio_device_realize;
3691     dc->unrealize = virtio_device_unrealize;
3692     dc->bus_type = TYPE_VIRTIO_BUS;
3693     dc->props = virtio_properties;
3694     vdc->start_ioeventfd = virtio_device_start_ioeventfd_impl;
3695     vdc->stop_ioeventfd = virtio_device_stop_ioeventfd_impl;
3696 
3697     vdc->legacy_features |= VIRTIO_LEGACY_FEATURES;
3698 }
3699 
3700 bool virtio_device_ioeventfd_enabled(VirtIODevice *vdev)
3701 {
3702     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3703     VirtioBusState *vbus = VIRTIO_BUS(qbus);
3704 
3705     return virtio_bus_ioeventfd_enabled(vbus);
3706 }
3707 
3708 static const TypeInfo virtio_device_info = {
3709     .name = TYPE_VIRTIO_DEVICE,
3710     .parent = TYPE_DEVICE,
3711     .instance_size = sizeof(VirtIODevice),
3712     .class_init = virtio_device_class_init,
3713     .instance_finalize = virtio_device_instance_finalize,
3714     .abstract = true,
3715     .class_size = sizeof(VirtioDeviceClass),
3716 };
3717 
3718 static void virtio_register_types(void)
3719 {
3720     type_register_static(&virtio_device_info);
3721 }
3722 
3723 type_init(virtio_register_types)
3724