xref: /openbmc/qemu/hw/virtio/virtio.c (revision 1c8f85d9)
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     bool host_notifier_enabled;
132     QLIST_ENTRY(VirtQueue) node;
133 };
134 
135 static void virtio_free_region_cache(VRingMemoryRegionCaches *caches)
136 {
137     if (!caches) {
138         return;
139     }
140 
141     address_space_cache_destroy(&caches->desc);
142     address_space_cache_destroy(&caches->avail);
143     address_space_cache_destroy(&caches->used);
144     g_free(caches);
145 }
146 
147 static void virtio_virtqueue_reset_region_cache(struct VirtQueue *vq)
148 {
149     VRingMemoryRegionCaches *caches;
150 
151     caches = atomic_read(&vq->vring.caches);
152     atomic_rcu_set(&vq->vring.caches, NULL);
153     if (caches) {
154         call_rcu(caches, virtio_free_region_cache, rcu);
155     }
156 }
157 
158 static void virtio_init_region_cache(VirtIODevice *vdev, int n)
159 {
160     VirtQueue *vq = &vdev->vq[n];
161     VRingMemoryRegionCaches *old = vq->vring.caches;
162     VRingMemoryRegionCaches *new = NULL;
163     hwaddr addr, size;
164     int64_t len;
165     bool packed;
166 
167 
168     addr = vq->vring.desc;
169     if (!addr) {
170         goto out_no_cache;
171     }
172     new = g_new0(VRingMemoryRegionCaches, 1);
173     size = virtio_queue_get_desc_size(vdev, n);
174     packed = virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED) ?
175                                    true : false;
176     len = address_space_cache_init(&new->desc, vdev->dma_as,
177                                    addr, size, packed);
178     if (len < size) {
179         virtio_error(vdev, "Cannot map desc");
180         goto err_desc;
181     }
182 
183     size = virtio_queue_get_used_size(vdev, n);
184     len = address_space_cache_init(&new->used, vdev->dma_as,
185                                    vq->vring.used, size, true);
186     if (len < size) {
187         virtio_error(vdev, "Cannot map used");
188         goto err_used;
189     }
190 
191     size = virtio_queue_get_avail_size(vdev, n);
192     len = address_space_cache_init(&new->avail, vdev->dma_as,
193                                    vq->vring.avail, size, false);
194     if (len < size) {
195         virtio_error(vdev, "Cannot map avail");
196         goto err_avail;
197     }
198 
199     atomic_rcu_set(&vq->vring.caches, new);
200     if (old) {
201         call_rcu(old, virtio_free_region_cache, rcu);
202     }
203     return;
204 
205 err_avail:
206     address_space_cache_destroy(&new->avail);
207 err_used:
208     address_space_cache_destroy(&new->used);
209 err_desc:
210     address_space_cache_destroy(&new->desc);
211 out_no_cache:
212     g_free(new);
213     virtio_virtqueue_reset_region_cache(vq);
214 }
215 
216 /* virt queue functions */
217 void virtio_queue_update_rings(VirtIODevice *vdev, int n)
218 {
219     VRing *vring = &vdev->vq[n].vring;
220 
221     if (!vring->num || !vring->desc || !vring->align) {
222         /* not yet setup -> nothing to do */
223         return;
224     }
225     vring->avail = vring->desc + vring->num * sizeof(VRingDesc);
226     vring->used = vring_align(vring->avail +
227                               offsetof(VRingAvail, ring[vring->num]),
228                               vring->align);
229     virtio_init_region_cache(vdev, n);
230 }
231 
232 /* Called within rcu_read_lock().  */
233 static void vring_split_desc_read(VirtIODevice *vdev, VRingDesc *desc,
234                                   MemoryRegionCache *cache, int i)
235 {
236     address_space_read_cached(cache, i * sizeof(VRingDesc),
237                               desc, sizeof(VRingDesc));
238     virtio_tswap64s(vdev, &desc->addr);
239     virtio_tswap32s(vdev, &desc->len);
240     virtio_tswap16s(vdev, &desc->flags);
241     virtio_tswap16s(vdev, &desc->next);
242 }
243 
244 static void vring_packed_event_read(VirtIODevice *vdev,
245                                     MemoryRegionCache *cache,
246                                     VRingPackedDescEvent *e)
247 {
248     hwaddr off_off = offsetof(VRingPackedDescEvent, off_wrap);
249     hwaddr off_flags = offsetof(VRingPackedDescEvent, flags);
250 
251     address_space_read_cached(cache, off_flags, &e->flags,
252                               sizeof(e->flags));
253     /* Make sure flags is seen before off_wrap */
254     smp_rmb();
255     address_space_read_cached(cache, off_off, &e->off_wrap,
256                               sizeof(e->off_wrap));
257     virtio_tswap16s(vdev, &e->off_wrap);
258     virtio_tswap16s(vdev, &e->flags);
259 }
260 
261 static void vring_packed_off_wrap_write(VirtIODevice *vdev,
262                                         MemoryRegionCache *cache,
263                                         uint16_t off_wrap)
264 {
265     hwaddr off = offsetof(VRingPackedDescEvent, off_wrap);
266 
267     virtio_tswap16s(vdev, &off_wrap);
268     address_space_write_cached(cache, off, &off_wrap, sizeof(off_wrap));
269     address_space_cache_invalidate(cache, off, sizeof(off_wrap));
270 }
271 
272 static void vring_packed_flags_write(VirtIODevice *vdev,
273                                      MemoryRegionCache *cache, uint16_t flags)
274 {
275     hwaddr off = offsetof(VRingPackedDescEvent, flags);
276 
277     virtio_tswap16s(vdev, &flags);
278     address_space_write_cached(cache, off, &flags, sizeof(flags));
279     address_space_cache_invalidate(cache, off, sizeof(flags));
280 }
281 
282 /* Called within rcu_read_lock().  */
283 static VRingMemoryRegionCaches *vring_get_region_caches(struct VirtQueue *vq)
284 {
285     VRingMemoryRegionCaches *caches = atomic_rcu_read(&vq->vring.caches);
286     assert(caches != NULL);
287     return caches;
288 }
289 /* Called within rcu_read_lock().  */
290 static inline uint16_t vring_avail_flags(VirtQueue *vq)
291 {
292     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
293     hwaddr pa = offsetof(VRingAvail, flags);
294     return virtio_lduw_phys_cached(vq->vdev, &caches->avail, pa);
295 }
296 
297 /* Called within rcu_read_lock().  */
298 static inline uint16_t vring_avail_idx(VirtQueue *vq)
299 {
300     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
301     hwaddr pa = offsetof(VRingAvail, idx);
302     vq->shadow_avail_idx = virtio_lduw_phys_cached(vq->vdev, &caches->avail, pa);
303     return vq->shadow_avail_idx;
304 }
305 
306 /* Called within rcu_read_lock().  */
307 static inline uint16_t vring_avail_ring(VirtQueue *vq, int i)
308 {
309     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
310     hwaddr pa = offsetof(VRingAvail, ring[i]);
311     return virtio_lduw_phys_cached(vq->vdev, &caches->avail, pa);
312 }
313 
314 /* Called within rcu_read_lock().  */
315 static inline uint16_t vring_get_used_event(VirtQueue *vq)
316 {
317     return vring_avail_ring(vq, vq->vring.num);
318 }
319 
320 /* Called within rcu_read_lock().  */
321 static inline void vring_used_write(VirtQueue *vq, VRingUsedElem *uelem,
322                                     int i)
323 {
324     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
325     hwaddr pa = offsetof(VRingUsed, ring[i]);
326     virtio_tswap32s(vq->vdev, &uelem->id);
327     virtio_tswap32s(vq->vdev, &uelem->len);
328     address_space_write_cached(&caches->used, pa, uelem, sizeof(VRingUsedElem));
329     address_space_cache_invalidate(&caches->used, pa, sizeof(VRingUsedElem));
330 }
331 
332 /* Called within rcu_read_lock().  */
333 static uint16_t vring_used_idx(VirtQueue *vq)
334 {
335     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
336     hwaddr pa = offsetof(VRingUsed, idx);
337     return virtio_lduw_phys_cached(vq->vdev, &caches->used, pa);
338 }
339 
340 /* Called within rcu_read_lock().  */
341 static inline void vring_used_idx_set(VirtQueue *vq, uint16_t val)
342 {
343     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
344     hwaddr pa = offsetof(VRingUsed, idx);
345     virtio_stw_phys_cached(vq->vdev, &caches->used, pa, val);
346     address_space_cache_invalidate(&caches->used, pa, sizeof(val));
347     vq->used_idx = val;
348 }
349 
350 /* Called within rcu_read_lock().  */
351 static inline void vring_used_flags_set_bit(VirtQueue *vq, int mask)
352 {
353     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
354     VirtIODevice *vdev = vq->vdev;
355     hwaddr pa = offsetof(VRingUsed, flags);
356     uint16_t flags = virtio_lduw_phys_cached(vq->vdev, &caches->used, pa);
357 
358     virtio_stw_phys_cached(vdev, &caches->used, pa, flags | mask);
359     address_space_cache_invalidate(&caches->used, pa, sizeof(flags));
360 }
361 
362 /* Called within rcu_read_lock().  */
363 static inline void vring_used_flags_unset_bit(VirtQueue *vq, int mask)
364 {
365     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
366     VirtIODevice *vdev = vq->vdev;
367     hwaddr pa = offsetof(VRingUsed, flags);
368     uint16_t flags = virtio_lduw_phys_cached(vq->vdev, &caches->used, pa);
369 
370     virtio_stw_phys_cached(vdev, &caches->used, pa, flags & ~mask);
371     address_space_cache_invalidate(&caches->used, pa, sizeof(flags));
372 }
373 
374 /* Called within rcu_read_lock().  */
375 static inline void vring_set_avail_event(VirtQueue *vq, uint16_t val)
376 {
377     VRingMemoryRegionCaches *caches;
378     hwaddr pa;
379     if (!vq->notification) {
380         return;
381     }
382 
383     caches = vring_get_region_caches(vq);
384     pa = offsetof(VRingUsed, ring[vq->vring.num]);
385     virtio_stw_phys_cached(vq->vdev, &caches->used, pa, val);
386     address_space_cache_invalidate(&caches->used, pa, sizeof(val));
387 }
388 
389 static void virtio_queue_split_set_notification(VirtQueue *vq, int enable)
390 {
391     RCU_READ_LOCK_GUARD();
392 
393     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_RING_F_EVENT_IDX)) {
394         vring_set_avail_event(vq, vring_avail_idx(vq));
395     } else if (enable) {
396         vring_used_flags_unset_bit(vq, VRING_USED_F_NO_NOTIFY);
397     } else {
398         vring_used_flags_set_bit(vq, VRING_USED_F_NO_NOTIFY);
399     }
400     if (enable) {
401         /* Expose avail event/used flags before caller checks the avail idx. */
402         smp_mb();
403     }
404 }
405 
406 static void virtio_queue_packed_set_notification(VirtQueue *vq, int enable)
407 {
408     uint16_t off_wrap;
409     VRingPackedDescEvent e;
410     VRingMemoryRegionCaches *caches;
411 
412     RCU_READ_LOCK_GUARD();
413     caches  = vring_get_region_caches(vq);
414     vring_packed_event_read(vq->vdev, &caches->used, &e);
415 
416     if (!enable) {
417         e.flags = VRING_PACKED_EVENT_FLAG_DISABLE;
418     } else if (virtio_vdev_has_feature(vq->vdev, VIRTIO_RING_F_EVENT_IDX)) {
419         off_wrap = vq->shadow_avail_idx | vq->shadow_avail_wrap_counter << 15;
420         vring_packed_off_wrap_write(vq->vdev, &caches->used, off_wrap);
421         /* Make sure off_wrap is wrote before flags */
422         smp_wmb();
423         e.flags = VRING_PACKED_EVENT_FLAG_DESC;
424     } else {
425         e.flags = VRING_PACKED_EVENT_FLAG_ENABLE;
426     }
427 
428     vring_packed_flags_write(vq->vdev, &caches->used, e.flags);
429     if (enable) {
430         /* Expose avail event/used flags before caller checks the avail idx. */
431         smp_mb();
432     }
433 }
434 
435 bool virtio_queue_get_notification(VirtQueue *vq)
436 {
437     return vq->notification;
438 }
439 
440 void virtio_queue_set_notification(VirtQueue *vq, int enable)
441 {
442     vq->notification = enable;
443 
444     if (!vq->vring.desc) {
445         return;
446     }
447 
448     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) {
449         virtio_queue_packed_set_notification(vq, enable);
450     } else {
451         virtio_queue_split_set_notification(vq, enable);
452     }
453 }
454 
455 int virtio_queue_ready(VirtQueue *vq)
456 {
457     return vq->vring.avail != 0;
458 }
459 
460 static void vring_packed_desc_read_flags(VirtIODevice *vdev,
461                                          uint16_t *flags,
462                                          MemoryRegionCache *cache,
463                                          int i)
464 {
465     address_space_read_cached(cache,
466                               i * sizeof(VRingPackedDesc) +
467                               offsetof(VRingPackedDesc, flags),
468                               flags, sizeof(*flags));
469     virtio_tswap16s(vdev, flags);
470 }
471 
472 static void vring_packed_desc_read(VirtIODevice *vdev,
473                                    VRingPackedDesc *desc,
474                                    MemoryRegionCache *cache,
475                                    int i, bool strict_order)
476 {
477     hwaddr off = i * sizeof(VRingPackedDesc);
478 
479     vring_packed_desc_read_flags(vdev, &desc->flags, cache, i);
480 
481     if (strict_order) {
482         /* Make sure flags is read before the rest fields. */
483         smp_rmb();
484     }
485 
486     address_space_read_cached(cache, off + offsetof(VRingPackedDesc, addr),
487                               &desc->addr, sizeof(desc->addr));
488     address_space_read_cached(cache, off + offsetof(VRingPackedDesc, id),
489                               &desc->id, sizeof(desc->id));
490     address_space_read_cached(cache, off + offsetof(VRingPackedDesc, len),
491                               &desc->len, sizeof(desc->len));
492     virtio_tswap64s(vdev, &desc->addr);
493     virtio_tswap16s(vdev, &desc->id);
494     virtio_tswap32s(vdev, &desc->len);
495 }
496 
497 static void vring_packed_desc_write_data(VirtIODevice *vdev,
498                                          VRingPackedDesc *desc,
499                                          MemoryRegionCache *cache,
500                                          int i)
501 {
502     hwaddr off_id = i * sizeof(VRingPackedDesc) +
503                     offsetof(VRingPackedDesc, id);
504     hwaddr off_len = i * sizeof(VRingPackedDesc) +
505                     offsetof(VRingPackedDesc, len);
506 
507     virtio_tswap32s(vdev, &desc->len);
508     virtio_tswap16s(vdev, &desc->id);
509     address_space_write_cached(cache, off_id, &desc->id, sizeof(desc->id));
510     address_space_cache_invalidate(cache, off_id, sizeof(desc->id));
511     address_space_write_cached(cache, off_len, &desc->len, sizeof(desc->len));
512     address_space_cache_invalidate(cache, off_len, sizeof(desc->len));
513 }
514 
515 static void vring_packed_desc_write_flags(VirtIODevice *vdev,
516                                           VRingPackedDesc *desc,
517                                           MemoryRegionCache *cache,
518                                           int i)
519 {
520     hwaddr off = i * sizeof(VRingPackedDesc) + offsetof(VRingPackedDesc, flags);
521 
522     virtio_tswap16s(vdev, &desc->flags);
523     address_space_write_cached(cache, off, &desc->flags, sizeof(desc->flags));
524     address_space_cache_invalidate(cache, off, sizeof(desc->flags));
525 }
526 
527 static void vring_packed_desc_write(VirtIODevice *vdev,
528                                     VRingPackedDesc *desc,
529                                     MemoryRegionCache *cache,
530                                     int i, bool strict_order)
531 {
532     vring_packed_desc_write_data(vdev, desc, cache, i);
533     if (strict_order) {
534         /* Make sure data is wrote before flags. */
535         smp_wmb();
536     }
537     vring_packed_desc_write_flags(vdev, desc, cache, i);
538 }
539 
540 static inline bool is_desc_avail(uint16_t flags, bool wrap_counter)
541 {
542     bool avail, used;
543 
544     avail = !!(flags & (1 << VRING_PACKED_DESC_F_AVAIL));
545     used = !!(flags & (1 << VRING_PACKED_DESC_F_USED));
546     return (avail != used) && (avail == wrap_counter);
547 }
548 
549 /* Fetch avail_idx from VQ memory only when we really need to know if
550  * guest has added some buffers.
551  * Called within rcu_read_lock().  */
552 static int virtio_queue_empty_rcu(VirtQueue *vq)
553 {
554     if (virtio_device_disabled(vq->vdev)) {
555         return 1;
556     }
557 
558     if (unlikely(!vq->vring.avail)) {
559         return 1;
560     }
561 
562     if (vq->shadow_avail_idx != vq->last_avail_idx) {
563         return 0;
564     }
565 
566     return vring_avail_idx(vq) == vq->last_avail_idx;
567 }
568 
569 static int virtio_queue_split_empty(VirtQueue *vq)
570 {
571     bool empty;
572 
573     if (virtio_device_disabled(vq->vdev)) {
574         return 1;
575     }
576 
577     if (unlikely(!vq->vring.avail)) {
578         return 1;
579     }
580 
581     if (vq->shadow_avail_idx != vq->last_avail_idx) {
582         return 0;
583     }
584 
585     RCU_READ_LOCK_GUARD();
586     empty = vring_avail_idx(vq) == vq->last_avail_idx;
587     return empty;
588 }
589 
590 static int virtio_queue_packed_empty_rcu(VirtQueue *vq)
591 {
592     struct VRingPackedDesc desc;
593     VRingMemoryRegionCaches *cache;
594 
595     if (unlikely(!vq->vring.desc)) {
596         return 1;
597     }
598 
599     cache = vring_get_region_caches(vq);
600     vring_packed_desc_read_flags(vq->vdev, &desc.flags, &cache->desc,
601                                  vq->last_avail_idx);
602 
603     return !is_desc_avail(desc.flags, vq->last_avail_wrap_counter);
604 }
605 
606 static int virtio_queue_packed_empty(VirtQueue *vq)
607 {
608     RCU_READ_LOCK_GUARD();
609     return virtio_queue_packed_empty_rcu(vq);
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 (virtio_device_disabled(vq->vdev)) {
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 (virtio_device_disabled(vq->vdev)) {
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_GUARD();
863     virtqueue_fill(vq, elem, len, 0);
864     virtqueue_flush(vq, 1);
865 }
866 
867 /* Called within rcu_read_lock().  */
868 static int virtqueue_num_heads(VirtQueue *vq, unsigned int idx)
869 {
870     uint16_t num_heads = vring_avail_idx(vq) - idx;
871 
872     /* Check it isn't doing very strange things with descriptor numbers. */
873     if (num_heads > vq->vring.num) {
874         virtio_error(vq->vdev, "Guest moved used index from %u to %u",
875                      idx, vq->shadow_avail_idx);
876         return -EINVAL;
877     }
878     /* On success, callers read a descriptor at vq->last_avail_idx.
879      * Make sure descriptor read does not bypass avail index read. */
880     if (num_heads) {
881         smp_rmb();
882     }
883 
884     return num_heads;
885 }
886 
887 /* Called within rcu_read_lock().  */
888 static bool virtqueue_get_head(VirtQueue *vq, unsigned int idx,
889                                unsigned int *head)
890 {
891     /* Grab the next descriptor number they're advertising, and increment
892      * the index we've seen. */
893     *head = vring_avail_ring(vq, idx % vq->vring.num);
894 
895     /* If their number is silly, that's a fatal mistake. */
896     if (*head >= vq->vring.num) {
897         virtio_error(vq->vdev, "Guest says index %u is available", *head);
898         return false;
899     }
900 
901     return true;
902 }
903 
904 enum {
905     VIRTQUEUE_READ_DESC_ERROR = -1,
906     VIRTQUEUE_READ_DESC_DONE = 0,   /* end of chain */
907     VIRTQUEUE_READ_DESC_MORE = 1,   /* more buffers in chain */
908 };
909 
910 static int virtqueue_split_read_next_desc(VirtIODevice *vdev, VRingDesc *desc,
911                                           MemoryRegionCache *desc_cache,
912                                           unsigned int max, unsigned int *next)
913 {
914     /* If this descriptor says it doesn't chain, we're done. */
915     if (!(desc->flags & VRING_DESC_F_NEXT)) {
916         return VIRTQUEUE_READ_DESC_DONE;
917     }
918 
919     /* Check they're not leading us off end of descriptors. */
920     *next = desc->next;
921     /* Make sure compiler knows to grab that: we don't want it changing! */
922     smp_wmb();
923 
924     if (*next >= max) {
925         virtio_error(vdev, "Desc next is %u", *next);
926         return VIRTQUEUE_READ_DESC_ERROR;
927     }
928 
929     vring_split_desc_read(vdev, desc, desc_cache, *next);
930     return VIRTQUEUE_READ_DESC_MORE;
931 }
932 
933 static void virtqueue_split_get_avail_bytes(VirtQueue *vq,
934                             unsigned int *in_bytes, unsigned int *out_bytes,
935                             unsigned max_in_bytes, unsigned max_out_bytes)
936 {
937     VirtIODevice *vdev = vq->vdev;
938     unsigned int max, idx;
939     unsigned int total_bufs, in_total, out_total;
940     VRingMemoryRegionCaches *caches;
941     MemoryRegionCache indirect_desc_cache = MEMORY_REGION_CACHE_INVALID;
942     int64_t len = 0;
943     int rc;
944 
945     RCU_READ_LOCK_GUARD();
946 
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     return;
1037 
1038 err:
1039     in_total = out_total = 0;
1040     goto done;
1041 }
1042 
1043 static int virtqueue_packed_read_next_desc(VirtQueue *vq,
1044                                            VRingPackedDesc *desc,
1045                                            MemoryRegionCache
1046                                            *desc_cache,
1047                                            unsigned int max,
1048                                            unsigned int *next,
1049                                            bool indirect)
1050 {
1051     /* If this descriptor says it doesn't chain, we're done. */
1052     if (!indirect && !(desc->flags & VRING_DESC_F_NEXT)) {
1053         return VIRTQUEUE_READ_DESC_DONE;
1054     }
1055 
1056     ++*next;
1057     if (*next == max) {
1058         if (indirect) {
1059             return VIRTQUEUE_READ_DESC_DONE;
1060         } else {
1061             (*next) -= vq->vring.num;
1062         }
1063     }
1064 
1065     vring_packed_desc_read(vq->vdev, desc, desc_cache, *next, false);
1066     return VIRTQUEUE_READ_DESC_MORE;
1067 }
1068 
1069 static void virtqueue_packed_get_avail_bytes(VirtQueue *vq,
1070                                              unsigned int *in_bytes,
1071                                              unsigned int *out_bytes,
1072                                              unsigned max_in_bytes,
1073                                              unsigned max_out_bytes)
1074 {
1075     VirtIODevice *vdev = vq->vdev;
1076     unsigned int max, idx;
1077     unsigned int total_bufs, in_total, out_total;
1078     MemoryRegionCache *desc_cache;
1079     VRingMemoryRegionCaches *caches;
1080     MemoryRegionCache indirect_desc_cache = MEMORY_REGION_CACHE_INVALID;
1081     int64_t len = 0;
1082     VRingPackedDesc desc;
1083     bool wrap_counter;
1084 
1085     RCU_READ_LOCK_GUARD();
1086     idx = vq->last_avail_idx;
1087     wrap_counter = vq->last_avail_wrap_counter;
1088     total_bufs = in_total = out_total = 0;
1089 
1090     max = vq->vring.num;
1091     caches = vring_get_region_caches(vq);
1092 
1093     for (;;) {
1094         unsigned int num_bufs = total_bufs;
1095         unsigned int i = idx;
1096         int rc;
1097 
1098         desc_cache = &caches->desc;
1099         vring_packed_desc_read(vdev, &desc, desc_cache, idx, true);
1100         if (!is_desc_avail(desc.flags, wrap_counter)) {
1101             break;
1102         }
1103 
1104         if (desc.flags & VRING_DESC_F_INDIRECT) {
1105             if (desc.len % sizeof(VRingPackedDesc)) {
1106                 virtio_error(vdev, "Invalid size for indirect buffer table");
1107                 goto err;
1108             }
1109 
1110             /* If we've got too many, that implies a descriptor loop. */
1111             if (num_bufs >= max) {
1112                 virtio_error(vdev, "Looped descriptor");
1113                 goto err;
1114             }
1115 
1116             /* loop over the indirect descriptor table */
1117             len = address_space_cache_init(&indirect_desc_cache,
1118                                            vdev->dma_as,
1119                                            desc.addr, desc.len, false);
1120             desc_cache = &indirect_desc_cache;
1121             if (len < desc.len) {
1122                 virtio_error(vdev, "Cannot map indirect buffer");
1123                 goto err;
1124             }
1125 
1126             max = desc.len / sizeof(VRingPackedDesc);
1127             num_bufs = i = 0;
1128             vring_packed_desc_read(vdev, &desc, desc_cache, i, false);
1129         }
1130 
1131         do {
1132             /* If we've got too many, that implies a descriptor loop. */
1133             if (++num_bufs > max) {
1134                 virtio_error(vdev, "Looped descriptor");
1135                 goto err;
1136             }
1137 
1138             if (desc.flags & VRING_DESC_F_WRITE) {
1139                 in_total += desc.len;
1140             } else {
1141                 out_total += desc.len;
1142             }
1143             if (in_total >= max_in_bytes && out_total >= max_out_bytes) {
1144                 goto done;
1145             }
1146 
1147             rc = virtqueue_packed_read_next_desc(vq, &desc, desc_cache, max,
1148                                                  &i, desc_cache ==
1149                                                  &indirect_desc_cache);
1150         } while (rc == VIRTQUEUE_READ_DESC_MORE);
1151 
1152         if (desc_cache == &indirect_desc_cache) {
1153             address_space_cache_destroy(&indirect_desc_cache);
1154             total_bufs++;
1155             idx++;
1156         } else {
1157             idx += num_bufs - total_bufs;
1158             total_bufs = num_bufs;
1159         }
1160 
1161         if (idx >= vq->vring.num) {
1162             idx -= vq->vring.num;
1163             wrap_counter ^= 1;
1164         }
1165     }
1166 
1167     /* Record the index and wrap counter for a kick we want */
1168     vq->shadow_avail_idx = idx;
1169     vq->shadow_avail_wrap_counter = wrap_counter;
1170 done:
1171     address_space_cache_destroy(&indirect_desc_cache);
1172     if (in_bytes) {
1173         *in_bytes = in_total;
1174     }
1175     if (out_bytes) {
1176         *out_bytes = out_total;
1177     }
1178     return;
1179 
1180 err:
1181     in_total = out_total = 0;
1182     goto done;
1183 }
1184 
1185 void virtqueue_get_avail_bytes(VirtQueue *vq, unsigned int *in_bytes,
1186                                unsigned int *out_bytes,
1187                                unsigned max_in_bytes, unsigned max_out_bytes)
1188 {
1189     uint16_t desc_size;
1190     VRingMemoryRegionCaches *caches;
1191 
1192     if (unlikely(!vq->vring.desc)) {
1193         goto err;
1194     }
1195 
1196     caches = vring_get_region_caches(vq);
1197     desc_size = virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED) ?
1198                                 sizeof(VRingPackedDesc) : sizeof(VRingDesc);
1199     if (caches->desc.len < vq->vring.num * desc_size) {
1200         virtio_error(vq->vdev, "Cannot map descriptor ring");
1201         goto err;
1202     }
1203 
1204     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) {
1205         virtqueue_packed_get_avail_bytes(vq, in_bytes, out_bytes,
1206                                          max_in_bytes, max_out_bytes);
1207     } else {
1208         virtqueue_split_get_avail_bytes(vq, in_bytes, out_bytes,
1209                                         max_in_bytes, max_out_bytes);
1210     }
1211 
1212     return;
1213 err:
1214     if (in_bytes) {
1215         *in_bytes = 0;
1216     }
1217     if (out_bytes) {
1218         *out_bytes = 0;
1219     }
1220 }
1221 
1222 int virtqueue_avail_bytes(VirtQueue *vq, unsigned int in_bytes,
1223                           unsigned int out_bytes)
1224 {
1225     unsigned int in_total, out_total;
1226 
1227     virtqueue_get_avail_bytes(vq, &in_total, &out_total, in_bytes, out_bytes);
1228     return in_bytes <= in_total && out_bytes <= out_total;
1229 }
1230 
1231 static bool virtqueue_map_desc(VirtIODevice *vdev, unsigned int *p_num_sg,
1232                                hwaddr *addr, struct iovec *iov,
1233                                unsigned int max_num_sg, bool is_write,
1234                                hwaddr pa, size_t sz)
1235 {
1236     bool ok = false;
1237     unsigned num_sg = *p_num_sg;
1238     assert(num_sg <= max_num_sg);
1239 
1240     if (!sz) {
1241         virtio_error(vdev, "virtio: zero sized buffers are not allowed");
1242         goto out;
1243     }
1244 
1245     while (sz) {
1246         hwaddr len = sz;
1247 
1248         if (num_sg == max_num_sg) {
1249             virtio_error(vdev, "virtio: too many write descriptors in "
1250                                "indirect table");
1251             goto out;
1252         }
1253 
1254         iov[num_sg].iov_base = dma_memory_map(vdev->dma_as, pa, &len,
1255                                               is_write ?
1256                                               DMA_DIRECTION_FROM_DEVICE :
1257                                               DMA_DIRECTION_TO_DEVICE);
1258         if (!iov[num_sg].iov_base) {
1259             virtio_error(vdev, "virtio: bogus descriptor or out of resources");
1260             goto out;
1261         }
1262 
1263         iov[num_sg].iov_len = len;
1264         addr[num_sg] = pa;
1265 
1266         sz -= len;
1267         pa += len;
1268         num_sg++;
1269     }
1270     ok = true;
1271 
1272 out:
1273     *p_num_sg = num_sg;
1274     return ok;
1275 }
1276 
1277 /* Only used by error code paths before we have a VirtQueueElement (therefore
1278  * virtqueue_unmap_sg() can't be used).  Assumes buffers weren't written to
1279  * yet.
1280  */
1281 static void virtqueue_undo_map_desc(unsigned int out_num, unsigned int in_num,
1282                                     struct iovec *iov)
1283 {
1284     unsigned int i;
1285 
1286     for (i = 0; i < out_num + in_num; i++) {
1287         int is_write = i >= out_num;
1288 
1289         cpu_physical_memory_unmap(iov->iov_base, iov->iov_len, is_write, 0);
1290         iov++;
1291     }
1292 }
1293 
1294 static void virtqueue_map_iovec(VirtIODevice *vdev, struct iovec *sg,
1295                                 hwaddr *addr, unsigned int num_sg,
1296                                 int is_write)
1297 {
1298     unsigned int i;
1299     hwaddr len;
1300 
1301     for (i = 0; i < num_sg; i++) {
1302         len = sg[i].iov_len;
1303         sg[i].iov_base = dma_memory_map(vdev->dma_as,
1304                                         addr[i], &len, is_write ?
1305                                         DMA_DIRECTION_FROM_DEVICE :
1306                                         DMA_DIRECTION_TO_DEVICE);
1307         if (!sg[i].iov_base) {
1308             error_report("virtio: error trying to map MMIO memory");
1309             exit(1);
1310         }
1311         if (len != sg[i].iov_len) {
1312             error_report("virtio: unexpected memory split");
1313             exit(1);
1314         }
1315     }
1316 }
1317 
1318 void virtqueue_map(VirtIODevice *vdev, VirtQueueElement *elem)
1319 {
1320     virtqueue_map_iovec(vdev, elem->in_sg, elem->in_addr, elem->in_num, 1);
1321     virtqueue_map_iovec(vdev, elem->out_sg, elem->out_addr, elem->out_num, 0);
1322 }
1323 
1324 static void *virtqueue_alloc_element(size_t sz, unsigned out_num, unsigned in_num)
1325 {
1326     VirtQueueElement *elem;
1327     size_t in_addr_ofs = QEMU_ALIGN_UP(sz, __alignof__(elem->in_addr[0]));
1328     size_t out_addr_ofs = in_addr_ofs + in_num * sizeof(elem->in_addr[0]);
1329     size_t out_addr_end = out_addr_ofs + out_num * sizeof(elem->out_addr[0]);
1330     size_t in_sg_ofs = QEMU_ALIGN_UP(out_addr_end, __alignof__(elem->in_sg[0]));
1331     size_t out_sg_ofs = in_sg_ofs + in_num * sizeof(elem->in_sg[0]);
1332     size_t out_sg_end = out_sg_ofs + out_num * sizeof(elem->out_sg[0]);
1333 
1334     assert(sz >= sizeof(VirtQueueElement));
1335     elem = g_malloc(out_sg_end);
1336     trace_virtqueue_alloc_element(elem, sz, in_num, out_num);
1337     elem->out_num = out_num;
1338     elem->in_num = in_num;
1339     elem->in_addr = (void *)elem + in_addr_ofs;
1340     elem->out_addr = (void *)elem + out_addr_ofs;
1341     elem->in_sg = (void *)elem + in_sg_ofs;
1342     elem->out_sg = (void *)elem + out_sg_ofs;
1343     return elem;
1344 }
1345 
1346 static void *virtqueue_split_pop(VirtQueue *vq, size_t sz)
1347 {
1348     unsigned int i, head, max;
1349     VRingMemoryRegionCaches *caches;
1350     MemoryRegionCache indirect_desc_cache = MEMORY_REGION_CACHE_INVALID;
1351     MemoryRegionCache *desc_cache;
1352     int64_t len;
1353     VirtIODevice *vdev = vq->vdev;
1354     VirtQueueElement *elem = NULL;
1355     unsigned out_num, in_num, elem_entries;
1356     hwaddr addr[VIRTQUEUE_MAX_SIZE];
1357     struct iovec iov[VIRTQUEUE_MAX_SIZE];
1358     VRingDesc desc;
1359     int rc;
1360 
1361     RCU_READ_LOCK_GUARD();
1362     if (virtio_queue_empty_rcu(vq)) {
1363         goto done;
1364     }
1365     /* Needed after virtio_queue_empty(), see comment in
1366      * virtqueue_num_heads(). */
1367     smp_rmb();
1368 
1369     /* When we start there are none of either input nor output. */
1370     out_num = in_num = elem_entries = 0;
1371 
1372     max = vq->vring.num;
1373 
1374     if (vq->inuse >= vq->vring.num) {
1375         virtio_error(vdev, "Virtqueue size exceeded");
1376         goto done;
1377     }
1378 
1379     if (!virtqueue_get_head(vq, vq->last_avail_idx++, &head)) {
1380         goto done;
1381     }
1382 
1383     if (virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX)) {
1384         vring_set_avail_event(vq, vq->last_avail_idx);
1385     }
1386 
1387     i = head;
1388 
1389     caches = vring_get_region_caches(vq);
1390     if (caches->desc.len < max * sizeof(VRingDesc)) {
1391         virtio_error(vdev, "Cannot map descriptor ring");
1392         goto done;
1393     }
1394 
1395     desc_cache = &caches->desc;
1396     vring_split_desc_read(vdev, &desc, desc_cache, i);
1397     if (desc.flags & VRING_DESC_F_INDIRECT) {
1398         if (!desc.len || (desc.len % sizeof(VRingDesc))) {
1399             virtio_error(vdev, "Invalid size for indirect buffer table");
1400             goto done;
1401         }
1402 
1403         /* loop over the indirect descriptor table */
1404         len = address_space_cache_init(&indirect_desc_cache, vdev->dma_as,
1405                                        desc.addr, desc.len, false);
1406         desc_cache = &indirect_desc_cache;
1407         if (len < desc.len) {
1408             virtio_error(vdev, "Cannot map indirect buffer");
1409             goto done;
1410         }
1411 
1412         max = desc.len / sizeof(VRingDesc);
1413         i = 0;
1414         vring_split_desc_read(vdev, &desc, desc_cache, i);
1415     }
1416 
1417     /* Collect all the descriptors */
1418     do {
1419         bool map_ok;
1420 
1421         if (desc.flags & VRING_DESC_F_WRITE) {
1422             map_ok = virtqueue_map_desc(vdev, &in_num, addr + out_num,
1423                                         iov + out_num,
1424                                         VIRTQUEUE_MAX_SIZE - out_num, true,
1425                                         desc.addr, desc.len);
1426         } else {
1427             if (in_num) {
1428                 virtio_error(vdev, "Incorrect order for descriptors");
1429                 goto err_undo_map;
1430             }
1431             map_ok = virtqueue_map_desc(vdev, &out_num, addr, iov,
1432                                         VIRTQUEUE_MAX_SIZE, false,
1433                                         desc.addr, desc.len);
1434         }
1435         if (!map_ok) {
1436             goto err_undo_map;
1437         }
1438 
1439         /* If we've got too many, that implies a descriptor loop. */
1440         if (++elem_entries > max) {
1441             virtio_error(vdev, "Looped descriptor");
1442             goto err_undo_map;
1443         }
1444 
1445         rc = virtqueue_split_read_next_desc(vdev, &desc, desc_cache, max, &i);
1446     } while (rc == VIRTQUEUE_READ_DESC_MORE);
1447 
1448     if (rc == VIRTQUEUE_READ_DESC_ERROR) {
1449         goto err_undo_map;
1450     }
1451 
1452     /* Now copy what we have collected and mapped */
1453     elem = virtqueue_alloc_element(sz, out_num, in_num);
1454     elem->index = head;
1455     elem->ndescs = 1;
1456     for (i = 0; i < out_num; i++) {
1457         elem->out_addr[i] = addr[i];
1458         elem->out_sg[i] = iov[i];
1459     }
1460     for (i = 0; i < in_num; i++) {
1461         elem->in_addr[i] = addr[out_num + i];
1462         elem->in_sg[i] = iov[out_num + i];
1463     }
1464 
1465     vq->inuse++;
1466 
1467     trace_virtqueue_pop(vq, elem, elem->in_num, elem->out_num);
1468 done:
1469     address_space_cache_destroy(&indirect_desc_cache);
1470 
1471     return elem;
1472 
1473 err_undo_map:
1474     virtqueue_undo_map_desc(out_num, in_num, iov);
1475     goto done;
1476 }
1477 
1478 static void *virtqueue_packed_pop(VirtQueue *vq, size_t sz)
1479 {
1480     unsigned int i, max;
1481     VRingMemoryRegionCaches *caches;
1482     MemoryRegionCache indirect_desc_cache = MEMORY_REGION_CACHE_INVALID;
1483     MemoryRegionCache *desc_cache;
1484     int64_t len;
1485     VirtIODevice *vdev = vq->vdev;
1486     VirtQueueElement *elem = NULL;
1487     unsigned out_num, in_num, elem_entries;
1488     hwaddr addr[VIRTQUEUE_MAX_SIZE];
1489     struct iovec iov[VIRTQUEUE_MAX_SIZE];
1490     VRingPackedDesc desc;
1491     uint16_t id;
1492     int rc;
1493 
1494     RCU_READ_LOCK_GUARD();
1495     if (virtio_queue_packed_empty_rcu(vq)) {
1496         goto done;
1497     }
1498 
1499     /* When we start there are none of either input nor output. */
1500     out_num = in_num = elem_entries = 0;
1501 
1502     max = vq->vring.num;
1503 
1504     if (vq->inuse >= vq->vring.num) {
1505         virtio_error(vdev, "Virtqueue size exceeded");
1506         goto done;
1507     }
1508 
1509     i = vq->last_avail_idx;
1510 
1511     caches = vring_get_region_caches(vq);
1512     if (caches->desc.len < max * sizeof(VRingDesc)) {
1513         virtio_error(vdev, "Cannot map descriptor ring");
1514         goto done;
1515     }
1516 
1517     desc_cache = &caches->desc;
1518     vring_packed_desc_read(vdev, &desc, desc_cache, i, true);
1519     id = desc.id;
1520     if (desc.flags & VRING_DESC_F_INDIRECT) {
1521         if (desc.len % sizeof(VRingPackedDesc)) {
1522             virtio_error(vdev, "Invalid size for indirect buffer table");
1523             goto done;
1524         }
1525 
1526         /* loop over the indirect descriptor table */
1527         len = address_space_cache_init(&indirect_desc_cache, vdev->dma_as,
1528                                        desc.addr, desc.len, false);
1529         desc_cache = &indirect_desc_cache;
1530         if (len < desc.len) {
1531             virtio_error(vdev, "Cannot map indirect buffer");
1532             goto done;
1533         }
1534 
1535         max = desc.len / sizeof(VRingPackedDesc);
1536         i = 0;
1537         vring_packed_desc_read(vdev, &desc, desc_cache, i, false);
1538     }
1539 
1540     /* Collect all the descriptors */
1541     do {
1542         bool map_ok;
1543 
1544         if (desc.flags & VRING_DESC_F_WRITE) {
1545             map_ok = virtqueue_map_desc(vdev, &in_num, addr + out_num,
1546                                         iov + out_num,
1547                                         VIRTQUEUE_MAX_SIZE - out_num, true,
1548                                         desc.addr, desc.len);
1549         } else {
1550             if (in_num) {
1551                 virtio_error(vdev, "Incorrect order for descriptors");
1552                 goto err_undo_map;
1553             }
1554             map_ok = virtqueue_map_desc(vdev, &out_num, addr, iov,
1555                                         VIRTQUEUE_MAX_SIZE, false,
1556                                         desc.addr, desc.len);
1557         }
1558         if (!map_ok) {
1559             goto err_undo_map;
1560         }
1561 
1562         /* If we've got too many, that implies a descriptor loop. */
1563         if (++elem_entries > max) {
1564             virtio_error(vdev, "Looped descriptor");
1565             goto err_undo_map;
1566         }
1567 
1568         rc = virtqueue_packed_read_next_desc(vq, &desc, desc_cache, max, &i,
1569                                              desc_cache ==
1570                                              &indirect_desc_cache);
1571     } while (rc == VIRTQUEUE_READ_DESC_MORE);
1572 
1573     /* Now copy what we have collected and mapped */
1574     elem = virtqueue_alloc_element(sz, out_num, in_num);
1575     for (i = 0; i < out_num; i++) {
1576         elem->out_addr[i] = addr[i];
1577         elem->out_sg[i] = iov[i];
1578     }
1579     for (i = 0; i < in_num; i++) {
1580         elem->in_addr[i] = addr[out_num + i];
1581         elem->in_sg[i] = iov[out_num + i];
1582     }
1583 
1584     elem->index = id;
1585     elem->ndescs = (desc_cache == &indirect_desc_cache) ? 1 : elem_entries;
1586     vq->last_avail_idx += elem->ndescs;
1587     vq->inuse += elem->ndescs;
1588 
1589     if (vq->last_avail_idx >= vq->vring.num) {
1590         vq->last_avail_idx -= vq->vring.num;
1591         vq->last_avail_wrap_counter ^= 1;
1592     }
1593 
1594     vq->shadow_avail_idx = vq->last_avail_idx;
1595     vq->shadow_avail_wrap_counter = vq->last_avail_wrap_counter;
1596 
1597     trace_virtqueue_pop(vq, elem, elem->in_num, elem->out_num);
1598 done:
1599     address_space_cache_destroy(&indirect_desc_cache);
1600 
1601     return elem;
1602 
1603 err_undo_map:
1604     virtqueue_undo_map_desc(out_num, in_num, iov);
1605     goto done;
1606 }
1607 
1608 void *virtqueue_pop(VirtQueue *vq, size_t sz)
1609 {
1610     if (virtio_device_disabled(vq->vdev)) {
1611         return NULL;
1612     }
1613 
1614     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) {
1615         return virtqueue_packed_pop(vq, sz);
1616     } else {
1617         return virtqueue_split_pop(vq, sz);
1618     }
1619 }
1620 
1621 static unsigned int virtqueue_packed_drop_all(VirtQueue *vq)
1622 {
1623     VRingMemoryRegionCaches *caches;
1624     MemoryRegionCache *desc_cache;
1625     unsigned int dropped = 0;
1626     VirtQueueElement elem = {};
1627     VirtIODevice *vdev = vq->vdev;
1628     VRingPackedDesc desc;
1629 
1630     caches = vring_get_region_caches(vq);
1631     desc_cache = &caches->desc;
1632 
1633     virtio_queue_set_notification(vq, 0);
1634 
1635     while (vq->inuse < vq->vring.num) {
1636         unsigned int idx = vq->last_avail_idx;
1637         /*
1638          * works similar to virtqueue_pop but does not map buffers
1639          * and does not allocate any memory.
1640          */
1641         vring_packed_desc_read(vdev, &desc, desc_cache,
1642                                vq->last_avail_idx , true);
1643         if (!is_desc_avail(desc.flags, vq->last_avail_wrap_counter)) {
1644             break;
1645         }
1646         elem.index = desc.id;
1647         elem.ndescs = 1;
1648         while (virtqueue_packed_read_next_desc(vq, &desc, desc_cache,
1649                                                vq->vring.num, &idx, false)) {
1650             ++elem.ndescs;
1651         }
1652         /*
1653          * immediately push the element, nothing to unmap
1654          * as both in_num and out_num are set to 0.
1655          */
1656         virtqueue_push(vq, &elem, 0);
1657         dropped++;
1658         vq->last_avail_idx += elem.ndescs;
1659         if (vq->last_avail_idx >= vq->vring.num) {
1660             vq->last_avail_idx -= vq->vring.num;
1661             vq->last_avail_wrap_counter ^= 1;
1662         }
1663     }
1664 
1665     return dropped;
1666 }
1667 
1668 static unsigned int virtqueue_split_drop_all(VirtQueue *vq)
1669 {
1670     unsigned int dropped = 0;
1671     VirtQueueElement elem = {};
1672     VirtIODevice *vdev = vq->vdev;
1673     bool fEventIdx = virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX);
1674 
1675     while (!virtio_queue_empty(vq) && vq->inuse < vq->vring.num) {
1676         /* works similar to virtqueue_pop but does not map buffers
1677         * and does not allocate any memory */
1678         smp_rmb();
1679         if (!virtqueue_get_head(vq, vq->last_avail_idx, &elem.index)) {
1680             break;
1681         }
1682         vq->inuse++;
1683         vq->last_avail_idx++;
1684         if (fEventIdx) {
1685             vring_set_avail_event(vq, vq->last_avail_idx);
1686         }
1687         /* immediately push the element, nothing to unmap
1688          * as both in_num and out_num are set to 0 */
1689         virtqueue_push(vq, &elem, 0);
1690         dropped++;
1691     }
1692 
1693     return dropped;
1694 }
1695 
1696 /* virtqueue_drop_all:
1697  * @vq: The #VirtQueue
1698  * Drops all queued buffers and indicates them to the guest
1699  * as if they are done. Useful when buffers can not be
1700  * processed but must be returned to the guest.
1701  */
1702 unsigned int virtqueue_drop_all(VirtQueue *vq)
1703 {
1704     struct VirtIODevice *vdev = vq->vdev;
1705 
1706     if (virtio_device_disabled(vq->vdev)) {
1707         return 0;
1708     }
1709 
1710     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
1711         return virtqueue_packed_drop_all(vq);
1712     } else {
1713         return virtqueue_split_drop_all(vq);
1714     }
1715 }
1716 
1717 /* Reading and writing a structure directly to QEMUFile is *awful*, but
1718  * it is what QEMU has always done by mistake.  We can change it sooner
1719  * or later by bumping the version number of the affected vm states.
1720  * In the meanwhile, since the in-memory layout of VirtQueueElement
1721  * has changed, we need to marshal to and from the layout that was
1722  * used before the change.
1723  */
1724 typedef struct VirtQueueElementOld {
1725     unsigned int index;
1726     unsigned int out_num;
1727     unsigned int in_num;
1728     hwaddr in_addr[VIRTQUEUE_MAX_SIZE];
1729     hwaddr out_addr[VIRTQUEUE_MAX_SIZE];
1730     struct iovec in_sg[VIRTQUEUE_MAX_SIZE];
1731     struct iovec out_sg[VIRTQUEUE_MAX_SIZE];
1732 } VirtQueueElementOld;
1733 
1734 void *qemu_get_virtqueue_element(VirtIODevice *vdev, QEMUFile *f, size_t sz)
1735 {
1736     VirtQueueElement *elem;
1737     VirtQueueElementOld data;
1738     int i;
1739 
1740     qemu_get_buffer(f, (uint8_t *)&data, sizeof(VirtQueueElementOld));
1741 
1742     /* TODO: teach all callers that this can fail, and return failure instead
1743      * of asserting here.
1744      * This is just one thing (there are probably more) that must be
1745      * fixed before we can allow NDEBUG compilation.
1746      */
1747     assert(ARRAY_SIZE(data.in_addr) >= data.in_num);
1748     assert(ARRAY_SIZE(data.out_addr) >= data.out_num);
1749 
1750     elem = virtqueue_alloc_element(sz, data.out_num, data.in_num);
1751     elem->index = data.index;
1752 
1753     for (i = 0; i < elem->in_num; i++) {
1754         elem->in_addr[i] = data.in_addr[i];
1755     }
1756 
1757     for (i = 0; i < elem->out_num; i++) {
1758         elem->out_addr[i] = data.out_addr[i];
1759     }
1760 
1761     for (i = 0; i < elem->in_num; i++) {
1762         /* Base is overwritten by virtqueue_map.  */
1763         elem->in_sg[i].iov_base = 0;
1764         elem->in_sg[i].iov_len = data.in_sg[i].iov_len;
1765     }
1766 
1767     for (i = 0; i < elem->out_num; i++) {
1768         /* Base is overwritten by virtqueue_map.  */
1769         elem->out_sg[i].iov_base = 0;
1770         elem->out_sg[i].iov_len = data.out_sg[i].iov_len;
1771     }
1772 
1773     if (virtio_host_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
1774         qemu_get_be32s(f, &elem->ndescs);
1775     }
1776 
1777     virtqueue_map(vdev, elem);
1778     return elem;
1779 }
1780 
1781 void qemu_put_virtqueue_element(VirtIODevice *vdev, QEMUFile *f,
1782                                 VirtQueueElement *elem)
1783 {
1784     VirtQueueElementOld data;
1785     int i;
1786 
1787     memset(&data, 0, sizeof(data));
1788     data.index = elem->index;
1789     data.in_num = elem->in_num;
1790     data.out_num = elem->out_num;
1791 
1792     for (i = 0; i < elem->in_num; i++) {
1793         data.in_addr[i] = elem->in_addr[i];
1794     }
1795 
1796     for (i = 0; i < elem->out_num; i++) {
1797         data.out_addr[i] = elem->out_addr[i];
1798     }
1799 
1800     for (i = 0; i < elem->in_num; i++) {
1801         /* Base is overwritten by virtqueue_map when loading.  Do not
1802          * save it, as it would leak the QEMU address space layout.  */
1803         data.in_sg[i].iov_len = elem->in_sg[i].iov_len;
1804     }
1805 
1806     for (i = 0; i < elem->out_num; i++) {
1807         /* Do not save iov_base as above.  */
1808         data.out_sg[i].iov_len = elem->out_sg[i].iov_len;
1809     }
1810 
1811     if (virtio_host_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
1812         qemu_put_be32s(f, &elem->ndescs);
1813     }
1814 
1815     qemu_put_buffer(f, (uint8_t *)&data, sizeof(VirtQueueElementOld));
1816 }
1817 
1818 /* virtio device */
1819 static void virtio_notify_vector(VirtIODevice *vdev, uint16_t vector)
1820 {
1821     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
1822     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
1823 
1824     if (virtio_device_disabled(vdev)) {
1825         return;
1826     }
1827 
1828     if (k->notify) {
1829         k->notify(qbus->parent, vector);
1830     }
1831 }
1832 
1833 void virtio_update_irq(VirtIODevice *vdev)
1834 {
1835     virtio_notify_vector(vdev, VIRTIO_NO_VECTOR);
1836 }
1837 
1838 static int virtio_validate_features(VirtIODevice *vdev)
1839 {
1840     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1841 
1842     if (virtio_host_has_feature(vdev, VIRTIO_F_IOMMU_PLATFORM) &&
1843         !virtio_vdev_has_feature(vdev, VIRTIO_F_IOMMU_PLATFORM)) {
1844         return -EFAULT;
1845     }
1846 
1847     if (k->validate_features) {
1848         return k->validate_features(vdev);
1849     } else {
1850         return 0;
1851     }
1852 }
1853 
1854 int virtio_set_status(VirtIODevice *vdev, uint8_t val)
1855 {
1856     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1857     trace_virtio_set_status(vdev, val);
1858 
1859     if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
1860         if (!(vdev->status & VIRTIO_CONFIG_S_FEATURES_OK) &&
1861             val & VIRTIO_CONFIG_S_FEATURES_OK) {
1862             int ret = virtio_validate_features(vdev);
1863 
1864             if (ret) {
1865                 return ret;
1866             }
1867         }
1868     }
1869 
1870     if ((vdev->status & VIRTIO_CONFIG_S_DRIVER_OK) !=
1871         (val & VIRTIO_CONFIG_S_DRIVER_OK)) {
1872         virtio_set_started(vdev, val & VIRTIO_CONFIG_S_DRIVER_OK);
1873     }
1874 
1875     if (k->set_status) {
1876         k->set_status(vdev, val);
1877     }
1878     vdev->status = val;
1879 
1880     return 0;
1881 }
1882 
1883 static enum virtio_device_endian virtio_default_endian(void)
1884 {
1885     if (target_words_bigendian()) {
1886         return VIRTIO_DEVICE_ENDIAN_BIG;
1887     } else {
1888         return VIRTIO_DEVICE_ENDIAN_LITTLE;
1889     }
1890 }
1891 
1892 static enum virtio_device_endian virtio_current_cpu_endian(void)
1893 {
1894     CPUClass *cc = CPU_GET_CLASS(current_cpu);
1895 
1896     if (cc->virtio_is_big_endian(current_cpu)) {
1897         return VIRTIO_DEVICE_ENDIAN_BIG;
1898     } else {
1899         return VIRTIO_DEVICE_ENDIAN_LITTLE;
1900     }
1901 }
1902 
1903 void virtio_reset(void *opaque)
1904 {
1905     VirtIODevice *vdev = opaque;
1906     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1907     int i;
1908 
1909     virtio_set_status(vdev, 0);
1910     if (current_cpu) {
1911         /* Guest initiated reset */
1912         vdev->device_endian = virtio_current_cpu_endian();
1913     } else {
1914         /* System reset */
1915         vdev->device_endian = virtio_default_endian();
1916     }
1917 
1918     if (k->reset) {
1919         k->reset(vdev);
1920     }
1921 
1922     vdev->start_on_kick = false;
1923     vdev->started = false;
1924     vdev->broken = false;
1925     vdev->guest_features = 0;
1926     vdev->queue_sel = 0;
1927     vdev->status = 0;
1928     vdev->disabled = false;
1929     atomic_set(&vdev->isr, 0);
1930     vdev->config_vector = VIRTIO_NO_VECTOR;
1931     virtio_notify_vector(vdev, vdev->config_vector);
1932 
1933     for(i = 0; i < VIRTIO_QUEUE_MAX; i++) {
1934         vdev->vq[i].vring.desc = 0;
1935         vdev->vq[i].vring.avail = 0;
1936         vdev->vq[i].vring.used = 0;
1937         vdev->vq[i].last_avail_idx = 0;
1938         vdev->vq[i].shadow_avail_idx = 0;
1939         vdev->vq[i].used_idx = 0;
1940         vdev->vq[i].last_avail_wrap_counter = true;
1941         vdev->vq[i].shadow_avail_wrap_counter = true;
1942         vdev->vq[i].used_wrap_counter = true;
1943         virtio_queue_set_vector(vdev, i, VIRTIO_NO_VECTOR);
1944         vdev->vq[i].signalled_used = 0;
1945         vdev->vq[i].signalled_used_valid = false;
1946         vdev->vq[i].notification = true;
1947         vdev->vq[i].vring.num = vdev->vq[i].vring.num_default;
1948         vdev->vq[i].inuse = 0;
1949         virtio_virtqueue_reset_region_cache(&vdev->vq[i]);
1950     }
1951 }
1952 
1953 uint32_t virtio_config_readb(VirtIODevice *vdev, uint32_t addr)
1954 {
1955     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1956     uint8_t val;
1957 
1958     if (addr + sizeof(val) > vdev->config_len) {
1959         return (uint32_t)-1;
1960     }
1961 
1962     k->get_config(vdev, vdev->config);
1963 
1964     val = ldub_p(vdev->config + addr);
1965     return val;
1966 }
1967 
1968 uint32_t virtio_config_readw(VirtIODevice *vdev, uint32_t addr)
1969 {
1970     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1971     uint16_t val;
1972 
1973     if (addr + sizeof(val) > vdev->config_len) {
1974         return (uint32_t)-1;
1975     }
1976 
1977     k->get_config(vdev, vdev->config);
1978 
1979     val = lduw_p(vdev->config + addr);
1980     return val;
1981 }
1982 
1983 uint32_t virtio_config_readl(VirtIODevice *vdev, uint32_t addr)
1984 {
1985     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1986     uint32_t val;
1987 
1988     if (addr + sizeof(val) > vdev->config_len) {
1989         return (uint32_t)-1;
1990     }
1991 
1992     k->get_config(vdev, vdev->config);
1993 
1994     val = ldl_p(vdev->config + addr);
1995     return val;
1996 }
1997 
1998 void virtio_config_writeb(VirtIODevice *vdev, uint32_t addr, uint32_t data)
1999 {
2000     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2001     uint8_t val = data;
2002 
2003     if (addr + sizeof(val) > vdev->config_len) {
2004         return;
2005     }
2006 
2007     stb_p(vdev->config + addr, val);
2008 
2009     if (k->set_config) {
2010         k->set_config(vdev, vdev->config);
2011     }
2012 }
2013 
2014 void virtio_config_writew(VirtIODevice *vdev, uint32_t addr, uint32_t data)
2015 {
2016     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2017     uint16_t val = data;
2018 
2019     if (addr + sizeof(val) > vdev->config_len) {
2020         return;
2021     }
2022 
2023     stw_p(vdev->config + addr, val);
2024 
2025     if (k->set_config) {
2026         k->set_config(vdev, vdev->config);
2027     }
2028 }
2029 
2030 void virtio_config_writel(VirtIODevice *vdev, uint32_t addr, uint32_t data)
2031 {
2032     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2033     uint32_t val = data;
2034 
2035     if (addr + sizeof(val) > vdev->config_len) {
2036         return;
2037     }
2038 
2039     stl_p(vdev->config + addr, val);
2040 
2041     if (k->set_config) {
2042         k->set_config(vdev, vdev->config);
2043     }
2044 }
2045 
2046 uint32_t virtio_config_modern_readb(VirtIODevice *vdev, uint32_t addr)
2047 {
2048     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2049     uint8_t val;
2050 
2051     if (addr + sizeof(val) > vdev->config_len) {
2052         return (uint32_t)-1;
2053     }
2054 
2055     k->get_config(vdev, vdev->config);
2056 
2057     val = ldub_p(vdev->config + addr);
2058     return val;
2059 }
2060 
2061 uint32_t virtio_config_modern_readw(VirtIODevice *vdev, uint32_t addr)
2062 {
2063     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2064     uint16_t val;
2065 
2066     if (addr + sizeof(val) > vdev->config_len) {
2067         return (uint32_t)-1;
2068     }
2069 
2070     k->get_config(vdev, vdev->config);
2071 
2072     val = lduw_le_p(vdev->config + addr);
2073     return val;
2074 }
2075 
2076 uint32_t virtio_config_modern_readl(VirtIODevice *vdev, uint32_t addr)
2077 {
2078     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2079     uint32_t val;
2080 
2081     if (addr + sizeof(val) > vdev->config_len) {
2082         return (uint32_t)-1;
2083     }
2084 
2085     k->get_config(vdev, vdev->config);
2086 
2087     val = ldl_le_p(vdev->config + addr);
2088     return val;
2089 }
2090 
2091 void virtio_config_modern_writeb(VirtIODevice *vdev,
2092                                  uint32_t addr, uint32_t data)
2093 {
2094     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2095     uint8_t val = data;
2096 
2097     if (addr + sizeof(val) > vdev->config_len) {
2098         return;
2099     }
2100 
2101     stb_p(vdev->config + addr, val);
2102 
2103     if (k->set_config) {
2104         k->set_config(vdev, vdev->config);
2105     }
2106 }
2107 
2108 void virtio_config_modern_writew(VirtIODevice *vdev,
2109                                  uint32_t addr, uint32_t data)
2110 {
2111     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2112     uint16_t val = data;
2113 
2114     if (addr + sizeof(val) > vdev->config_len) {
2115         return;
2116     }
2117 
2118     stw_le_p(vdev->config + addr, val);
2119 
2120     if (k->set_config) {
2121         k->set_config(vdev, vdev->config);
2122     }
2123 }
2124 
2125 void virtio_config_modern_writel(VirtIODevice *vdev,
2126                                  uint32_t addr, uint32_t data)
2127 {
2128     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2129     uint32_t val = data;
2130 
2131     if (addr + sizeof(val) > vdev->config_len) {
2132         return;
2133     }
2134 
2135     stl_le_p(vdev->config + addr, val);
2136 
2137     if (k->set_config) {
2138         k->set_config(vdev, vdev->config);
2139     }
2140 }
2141 
2142 void virtio_queue_set_addr(VirtIODevice *vdev, int n, hwaddr addr)
2143 {
2144     if (!vdev->vq[n].vring.num) {
2145         return;
2146     }
2147     vdev->vq[n].vring.desc = addr;
2148     virtio_queue_update_rings(vdev, n);
2149 }
2150 
2151 hwaddr virtio_queue_get_addr(VirtIODevice *vdev, int n)
2152 {
2153     return vdev->vq[n].vring.desc;
2154 }
2155 
2156 void virtio_queue_set_rings(VirtIODevice *vdev, int n, hwaddr desc,
2157                             hwaddr avail, hwaddr used)
2158 {
2159     if (!vdev->vq[n].vring.num) {
2160         return;
2161     }
2162     vdev->vq[n].vring.desc = desc;
2163     vdev->vq[n].vring.avail = avail;
2164     vdev->vq[n].vring.used = used;
2165     virtio_init_region_cache(vdev, n);
2166 }
2167 
2168 void virtio_queue_set_num(VirtIODevice *vdev, int n, int num)
2169 {
2170     /* Don't allow guest to flip queue between existent and
2171      * nonexistent states, or to set it to an invalid size.
2172      */
2173     if (!!num != !!vdev->vq[n].vring.num ||
2174         num > VIRTQUEUE_MAX_SIZE ||
2175         num < 0) {
2176         return;
2177     }
2178     vdev->vq[n].vring.num = num;
2179 }
2180 
2181 VirtQueue *virtio_vector_first_queue(VirtIODevice *vdev, uint16_t vector)
2182 {
2183     return QLIST_FIRST(&vdev->vector_queues[vector]);
2184 }
2185 
2186 VirtQueue *virtio_vector_next_queue(VirtQueue *vq)
2187 {
2188     return QLIST_NEXT(vq, node);
2189 }
2190 
2191 int virtio_queue_get_num(VirtIODevice *vdev, int n)
2192 {
2193     return vdev->vq[n].vring.num;
2194 }
2195 
2196 int virtio_queue_get_max_num(VirtIODevice *vdev, int n)
2197 {
2198     return vdev->vq[n].vring.num_default;
2199 }
2200 
2201 int virtio_get_num_queues(VirtIODevice *vdev)
2202 {
2203     int i;
2204 
2205     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2206         if (!virtio_queue_get_num(vdev, i)) {
2207             break;
2208         }
2209     }
2210 
2211     return i;
2212 }
2213 
2214 void virtio_queue_set_align(VirtIODevice *vdev, int n, int align)
2215 {
2216     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2217     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
2218 
2219     /* virtio-1 compliant devices cannot change the alignment */
2220     if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
2221         error_report("tried to modify queue alignment for virtio-1 device");
2222         return;
2223     }
2224     /* Check that the transport told us it was going to do this
2225      * (so a buggy transport will immediately assert rather than
2226      * silently failing to migrate this state)
2227      */
2228     assert(k->has_variable_vring_alignment);
2229 
2230     if (align) {
2231         vdev->vq[n].vring.align = align;
2232         virtio_queue_update_rings(vdev, n);
2233     }
2234 }
2235 
2236 static bool virtio_queue_notify_aio_vq(VirtQueue *vq)
2237 {
2238     bool ret = false;
2239 
2240     if (vq->vring.desc && vq->handle_aio_output) {
2241         VirtIODevice *vdev = vq->vdev;
2242 
2243         trace_virtio_queue_notify(vdev, vq - vdev->vq, vq);
2244         ret = vq->handle_aio_output(vdev, vq);
2245 
2246         if (unlikely(vdev->start_on_kick)) {
2247             virtio_set_started(vdev, true);
2248         }
2249     }
2250 
2251     return ret;
2252 }
2253 
2254 static void virtio_queue_notify_vq(VirtQueue *vq)
2255 {
2256     if (vq->vring.desc && vq->handle_output) {
2257         VirtIODevice *vdev = vq->vdev;
2258 
2259         if (unlikely(vdev->broken)) {
2260             return;
2261         }
2262 
2263         trace_virtio_queue_notify(vdev, vq - vdev->vq, vq);
2264         vq->handle_output(vdev, vq);
2265 
2266         if (unlikely(vdev->start_on_kick)) {
2267             virtio_set_started(vdev, true);
2268         }
2269     }
2270 }
2271 
2272 void virtio_queue_notify(VirtIODevice *vdev, int n)
2273 {
2274     VirtQueue *vq = &vdev->vq[n];
2275 
2276     if (unlikely(!vq->vring.desc || vdev->broken)) {
2277         return;
2278     }
2279 
2280     trace_virtio_queue_notify(vdev, vq - vdev->vq, vq);
2281     if (vq->host_notifier_enabled) {
2282         event_notifier_set(&vq->host_notifier);
2283     } else if (vq->handle_output) {
2284         vq->handle_output(vdev, vq);
2285 
2286         if (unlikely(vdev->start_on_kick)) {
2287             virtio_set_started(vdev, true);
2288         }
2289     }
2290 }
2291 
2292 uint16_t virtio_queue_vector(VirtIODevice *vdev, int n)
2293 {
2294     return n < VIRTIO_QUEUE_MAX ? vdev->vq[n].vector :
2295         VIRTIO_NO_VECTOR;
2296 }
2297 
2298 void virtio_queue_set_vector(VirtIODevice *vdev, int n, uint16_t vector)
2299 {
2300     VirtQueue *vq = &vdev->vq[n];
2301 
2302     if (n < VIRTIO_QUEUE_MAX) {
2303         if (vdev->vector_queues &&
2304             vdev->vq[n].vector != VIRTIO_NO_VECTOR) {
2305             QLIST_REMOVE(vq, node);
2306         }
2307         vdev->vq[n].vector = vector;
2308         if (vdev->vector_queues &&
2309             vector != VIRTIO_NO_VECTOR) {
2310             QLIST_INSERT_HEAD(&vdev->vector_queues[vector], vq, node);
2311         }
2312     }
2313 }
2314 
2315 VirtQueue *virtio_add_queue(VirtIODevice *vdev, int queue_size,
2316                             VirtIOHandleOutput handle_output)
2317 {
2318     int i;
2319 
2320     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2321         if (vdev->vq[i].vring.num == 0)
2322             break;
2323     }
2324 
2325     if (i == VIRTIO_QUEUE_MAX || queue_size > VIRTQUEUE_MAX_SIZE)
2326         abort();
2327 
2328     vdev->vq[i].vring.num = queue_size;
2329     vdev->vq[i].vring.num_default = queue_size;
2330     vdev->vq[i].vring.align = VIRTIO_PCI_VRING_ALIGN;
2331     vdev->vq[i].handle_output = handle_output;
2332     vdev->vq[i].handle_aio_output = NULL;
2333     vdev->vq[i].used_elems = g_malloc0(sizeof(VirtQueueElement) *
2334                                        queue_size);
2335 
2336     return &vdev->vq[i];
2337 }
2338 
2339 void virtio_delete_queue(VirtQueue *vq)
2340 {
2341     vq->vring.num = 0;
2342     vq->vring.num_default = 0;
2343     vq->handle_output = NULL;
2344     vq->handle_aio_output = NULL;
2345     g_free(vq->used_elems);
2346     vq->used_elems = NULL;
2347 }
2348 
2349 void virtio_del_queue(VirtIODevice *vdev, int n)
2350 {
2351     if (n < 0 || n >= VIRTIO_QUEUE_MAX) {
2352         abort();
2353     }
2354 
2355     virtio_delete_queue(&vdev->vq[n]);
2356 }
2357 
2358 static void virtio_set_isr(VirtIODevice *vdev, int value)
2359 {
2360     uint8_t old = atomic_read(&vdev->isr);
2361 
2362     /* Do not write ISR if it does not change, so that its cacheline remains
2363      * shared in the common case where the guest does not read it.
2364      */
2365     if ((old & value) != value) {
2366         atomic_or(&vdev->isr, value);
2367     }
2368 }
2369 
2370 static bool virtio_split_should_notify(VirtIODevice *vdev, VirtQueue *vq)
2371 {
2372     uint16_t old, new;
2373     bool v;
2374     /* We need to expose used array entries before checking used event. */
2375     smp_mb();
2376     /* Always notify when queue is empty (when feature acknowledge) */
2377     if (virtio_vdev_has_feature(vdev, VIRTIO_F_NOTIFY_ON_EMPTY) &&
2378         !vq->inuse && virtio_queue_empty(vq)) {
2379         return true;
2380     }
2381 
2382     if (!virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX)) {
2383         return !(vring_avail_flags(vq) & VRING_AVAIL_F_NO_INTERRUPT);
2384     }
2385 
2386     v = vq->signalled_used_valid;
2387     vq->signalled_used_valid = true;
2388     old = vq->signalled_used;
2389     new = vq->signalled_used = vq->used_idx;
2390     return !v || vring_need_event(vring_get_used_event(vq), new, old);
2391 }
2392 
2393 static bool vring_packed_need_event(VirtQueue *vq, bool wrap,
2394                                     uint16_t off_wrap, uint16_t new,
2395                                     uint16_t old)
2396 {
2397     int off = off_wrap & ~(1 << 15);
2398 
2399     if (wrap != off_wrap >> 15) {
2400         off -= vq->vring.num;
2401     }
2402 
2403     return vring_need_event(off, new, old);
2404 }
2405 
2406 static bool virtio_packed_should_notify(VirtIODevice *vdev, VirtQueue *vq)
2407 {
2408     VRingPackedDescEvent e;
2409     uint16_t old, new;
2410     bool v;
2411     VRingMemoryRegionCaches *caches;
2412 
2413     caches = vring_get_region_caches(vq);
2414     vring_packed_event_read(vdev, &caches->avail, &e);
2415 
2416     old = vq->signalled_used;
2417     new = vq->signalled_used = vq->used_idx;
2418     v = vq->signalled_used_valid;
2419     vq->signalled_used_valid = true;
2420 
2421     if (e.flags == VRING_PACKED_EVENT_FLAG_DISABLE) {
2422         return false;
2423     } else if (e.flags == VRING_PACKED_EVENT_FLAG_ENABLE) {
2424         return true;
2425     }
2426 
2427     return !v || vring_packed_need_event(vq, vq->used_wrap_counter,
2428                                          e.off_wrap, new, old);
2429 }
2430 
2431 /* Called within rcu_read_lock().  */
2432 static bool virtio_should_notify(VirtIODevice *vdev, VirtQueue *vq)
2433 {
2434     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
2435         return virtio_packed_should_notify(vdev, vq);
2436     } else {
2437         return virtio_split_should_notify(vdev, vq);
2438     }
2439 }
2440 
2441 void virtio_notify_irqfd(VirtIODevice *vdev, VirtQueue *vq)
2442 {
2443     WITH_RCU_READ_LOCK_GUARD() {
2444         if (!virtio_should_notify(vdev, vq)) {
2445             return;
2446         }
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     WITH_RCU_READ_LOCK_GUARD() {
2479         if (!virtio_should_notify(vdev, vq)) {
2480             return;
2481         }
2482     }
2483 
2484     trace_virtio_notify(vdev, vq);
2485     virtio_irq(vq);
2486 }
2487 
2488 void virtio_notify_config(VirtIODevice *vdev)
2489 {
2490     if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK))
2491         return;
2492 
2493     virtio_set_isr(vdev, 0x3);
2494     vdev->generation++;
2495     virtio_notify_vector(vdev, vdev->config_vector);
2496 }
2497 
2498 static bool virtio_device_endian_needed(void *opaque)
2499 {
2500     VirtIODevice *vdev = opaque;
2501 
2502     assert(vdev->device_endian != VIRTIO_DEVICE_ENDIAN_UNKNOWN);
2503     if (!virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
2504         return vdev->device_endian != virtio_default_endian();
2505     }
2506     /* Devices conforming to VIRTIO 1.0 or later are always LE. */
2507     return vdev->device_endian != VIRTIO_DEVICE_ENDIAN_LITTLE;
2508 }
2509 
2510 static bool virtio_64bit_features_needed(void *opaque)
2511 {
2512     VirtIODevice *vdev = opaque;
2513 
2514     return (vdev->host_features >> 32) != 0;
2515 }
2516 
2517 static bool virtio_virtqueue_needed(void *opaque)
2518 {
2519     VirtIODevice *vdev = opaque;
2520 
2521     return virtio_host_has_feature(vdev, VIRTIO_F_VERSION_1);
2522 }
2523 
2524 static bool virtio_packed_virtqueue_needed(void *opaque)
2525 {
2526     VirtIODevice *vdev = opaque;
2527 
2528     return virtio_host_has_feature(vdev, VIRTIO_F_RING_PACKED);
2529 }
2530 
2531 static bool virtio_ringsize_needed(void *opaque)
2532 {
2533     VirtIODevice *vdev = opaque;
2534     int i;
2535 
2536     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2537         if (vdev->vq[i].vring.num != vdev->vq[i].vring.num_default) {
2538             return true;
2539         }
2540     }
2541     return false;
2542 }
2543 
2544 static bool virtio_extra_state_needed(void *opaque)
2545 {
2546     VirtIODevice *vdev = opaque;
2547     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2548     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
2549 
2550     return k->has_extra_state &&
2551         k->has_extra_state(qbus->parent);
2552 }
2553 
2554 static bool virtio_broken_needed(void *opaque)
2555 {
2556     VirtIODevice *vdev = opaque;
2557 
2558     return vdev->broken;
2559 }
2560 
2561 static bool virtio_started_needed(void *opaque)
2562 {
2563     VirtIODevice *vdev = opaque;
2564 
2565     return vdev->started;
2566 }
2567 
2568 static bool virtio_disabled_needed(void *opaque)
2569 {
2570     VirtIODevice *vdev = opaque;
2571 
2572     return vdev->disabled;
2573 }
2574 
2575 static const VMStateDescription vmstate_virtqueue = {
2576     .name = "virtqueue_state",
2577     .version_id = 1,
2578     .minimum_version_id = 1,
2579     .fields = (VMStateField[]) {
2580         VMSTATE_UINT64(vring.avail, struct VirtQueue),
2581         VMSTATE_UINT64(vring.used, struct VirtQueue),
2582         VMSTATE_END_OF_LIST()
2583     }
2584 };
2585 
2586 static const VMStateDescription vmstate_packed_virtqueue = {
2587     .name = "packed_virtqueue_state",
2588     .version_id = 1,
2589     .minimum_version_id = 1,
2590     .fields = (VMStateField[]) {
2591         VMSTATE_UINT16(last_avail_idx, struct VirtQueue),
2592         VMSTATE_BOOL(last_avail_wrap_counter, struct VirtQueue),
2593         VMSTATE_UINT16(used_idx, struct VirtQueue),
2594         VMSTATE_BOOL(used_wrap_counter, struct VirtQueue),
2595         VMSTATE_UINT32(inuse, struct VirtQueue),
2596         VMSTATE_END_OF_LIST()
2597     }
2598 };
2599 
2600 static const VMStateDescription vmstate_virtio_virtqueues = {
2601     .name = "virtio/virtqueues",
2602     .version_id = 1,
2603     .minimum_version_id = 1,
2604     .needed = &virtio_virtqueue_needed,
2605     .fields = (VMStateField[]) {
2606         VMSTATE_STRUCT_VARRAY_POINTER_KNOWN(vq, struct VirtIODevice,
2607                       VIRTIO_QUEUE_MAX, 0, vmstate_virtqueue, VirtQueue),
2608         VMSTATE_END_OF_LIST()
2609     }
2610 };
2611 
2612 static const VMStateDescription vmstate_virtio_packed_virtqueues = {
2613     .name = "virtio/packed_virtqueues",
2614     .version_id = 1,
2615     .minimum_version_id = 1,
2616     .needed = &virtio_packed_virtqueue_needed,
2617     .fields = (VMStateField[]) {
2618         VMSTATE_STRUCT_VARRAY_POINTER_KNOWN(vq, struct VirtIODevice,
2619                       VIRTIO_QUEUE_MAX, 0, vmstate_packed_virtqueue, VirtQueue),
2620         VMSTATE_END_OF_LIST()
2621     }
2622 };
2623 
2624 static const VMStateDescription vmstate_ringsize = {
2625     .name = "ringsize_state",
2626     .version_id = 1,
2627     .minimum_version_id = 1,
2628     .fields = (VMStateField[]) {
2629         VMSTATE_UINT32(vring.num_default, struct VirtQueue),
2630         VMSTATE_END_OF_LIST()
2631     }
2632 };
2633 
2634 static const VMStateDescription vmstate_virtio_ringsize = {
2635     .name = "virtio/ringsize",
2636     .version_id = 1,
2637     .minimum_version_id = 1,
2638     .needed = &virtio_ringsize_needed,
2639     .fields = (VMStateField[]) {
2640         VMSTATE_STRUCT_VARRAY_POINTER_KNOWN(vq, struct VirtIODevice,
2641                       VIRTIO_QUEUE_MAX, 0, vmstate_ringsize, VirtQueue),
2642         VMSTATE_END_OF_LIST()
2643     }
2644 };
2645 
2646 static int get_extra_state(QEMUFile *f, void *pv, size_t size,
2647                            const VMStateField *field)
2648 {
2649     VirtIODevice *vdev = pv;
2650     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2651     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
2652 
2653     if (!k->load_extra_state) {
2654         return -1;
2655     } else {
2656         return k->load_extra_state(qbus->parent, f);
2657     }
2658 }
2659 
2660 static int put_extra_state(QEMUFile *f, void *pv, size_t size,
2661                            const VMStateField *field, QJSON *vmdesc)
2662 {
2663     VirtIODevice *vdev = pv;
2664     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2665     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
2666 
2667     k->save_extra_state(qbus->parent, f);
2668     return 0;
2669 }
2670 
2671 static const VMStateInfo vmstate_info_extra_state = {
2672     .name = "virtqueue_extra_state",
2673     .get = get_extra_state,
2674     .put = put_extra_state,
2675 };
2676 
2677 static const VMStateDescription vmstate_virtio_extra_state = {
2678     .name = "virtio/extra_state",
2679     .version_id = 1,
2680     .minimum_version_id = 1,
2681     .needed = &virtio_extra_state_needed,
2682     .fields = (VMStateField[]) {
2683         {
2684             .name         = "extra_state",
2685             .version_id   = 0,
2686             .field_exists = NULL,
2687             .size         = 0,
2688             .info         = &vmstate_info_extra_state,
2689             .flags        = VMS_SINGLE,
2690             .offset       = 0,
2691         },
2692         VMSTATE_END_OF_LIST()
2693     }
2694 };
2695 
2696 static const VMStateDescription vmstate_virtio_device_endian = {
2697     .name = "virtio/device_endian",
2698     .version_id = 1,
2699     .minimum_version_id = 1,
2700     .needed = &virtio_device_endian_needed,
2701     .fields = (VMStateField[]) {
2702         VMSTATE_UINT8(device_endian, VirtIODevice),
2703         VMSTATE_END_OF_LIST()
2704     }
2705 };
2706 
2707 static const VMStateDescription vmstate_virtio_64bit_features = {
2708     .name = "virtio/64bit_features",
2709     .version_id = 1,
2710     .minimum_version_id = 1,
2711     .needed = &virtio_64bit_features_needed,
2712     .fields = (VMStateField[]) {
2713         VMSTATE_UINT64(guest_features, VirtIODevice),
2714         VMSTATE_END_OF_LIST()
2715     }
2716 };
2717 
2718 static const VMStateDescription vmstate_virtio_broken = {
2719     .name = "virtio/broken",
2720     .version_id = 1,
2721     .minimum_version_id = 1,
2722     .needed = &virtio_broken_needed,
2723     .fields = (VMStateField[]) {
2724         VMSTATE_BOOL(broken, VirtIODevice),
2725         VMSTATE_END_OF_LIST()
2726     }
2727 };
2728 
2729 static const VMStateDescription vmstate_virtio_started = {
2730     .name = "virtio/started",
2731     .version_id = 1,
2732     .minimum_version_id = 1,
2733     .needed = &virtio_started_needed,
2734     .fields = (VMStateField[]) {
2735         VMSTATE_BOOL(started, VirtIODevice),
2736         VMSTATE_END_OF_LIST()
2737     }
2738 };
2739 
2740 static const VMStateDescription vmstate_virtio_disabled = {
2741     .name = "virtio/disabled",
2742     .version_id = 1,
2743     .minimum_version_id = 1,
2744     .needed = &virtio_disabled_needed,
2745     .fields = (VMStateField[]) {
2746         VMSTATE_BOOL(disabled, VirtIODevice),
2747         VMSTATE_END_OF_LIST()
2748     }
2749 };
2750 
2751 static const VMStateDescription vmstate_virtio = {
2752     .name = "virtio",
2753     .version_id = 1,
2754     .minimum_version_id = 1,
2755     .minimum_version_id_old = 1,
2756     .fields = (VMStateField[]) {
2757         VMSTATE_END_OF_LIST()
2758     },
2759     .subsections = (const VMStateDescription*[]) {
2760         &vmstate_virtio_device_endian,
2761         &vmstate_virtio_64bit_features,
2762         &vmstate_virtio_virtqueues,
2763         &vmstate_virtio_ringsize,
2764         &vmstate_virtio_broken,
2765         &vmstate_virtio_extra_state,
2766         &vmstate_virtio_started,
2767         &vmstate_virtio_packed_virtqueues,
2768         &vmstate_virtio_disabled,
2769         NULL
2770     }
2771 };
2772 
2773 int virtio_save(VirtIODevice *vdev, QEMUFile *f)
2774 {
2775     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2776     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
2777     VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev);
2778     uint32_t guest_features_lo = (vdev->guest_features & 0xffffffff);
2779     int i;
2780 
2781     if (k->save_config) {
2782         k->save_config(qbus->parent, f);
2783     }
2784 
2785     qemu_put_8s(f, &vdev->status);
2786     qemu_put_8s(f, &vdev->isr);
2787     qemu_put_be16s(f, &vdev->queue_sel);
2788     qemu_put_be32s(f, &guest_features_lo);
2789     qemu_put_be32(f, vdev->config_len);
2790     qemu_put_buffer(f, vdev->config, vdev->config_len);
2791 
2792     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2793         if (vdev->vq[i].vring.num == 0)
2794             break;
2795     }
2796 
2797     qemu_put_be32(f, i);
2798 
2799     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2800         if (vdev->vq[i].vring.num == 0)
2801             break;
2802 
2803         qemu_put_be32(f, vdev->vq[i].vring.num);
2804         if (k->has_variable_vring_alignment) {
2805             qemu_put_be32(f, vdev->vq[i].vring.align);
2806         }
2807         /*
2808          * Save desc now, the rest of the ring addresses are saved in
2809          * subsections for VIRTIO-1 devices.
2810          */
2811         qemu_put_be64(f, vdev->vq[i].vring.desc);
2812         qemu_put_be16s(f, &vdev->vq[i].last_avail_idx);
2813         if (k->save_queue) {
2814             k->save_queue(qbus->parent, i, f);
2815         }
2816     }
2817 
2818     if (vdc->save != NULL) {
2819         vdc->save(vdev, f);
2820     }
2821 
2822     if (vdc->vmsd) {
2823         int ret = vmstate_save_state(f, vdc->vmsd, vdev, NULL);
2824         if (ret) {
2825             return ret;
2826         }
2827     }
2828 
2829     /* Subsections */
2830     return vmstate_save_state(f, &vmstate_virtio, vdev, NULL);
2831 }
2832 
2833 /* A wrapper for use as a VMState .put function */
2834 static int virtio_device_put(QEMUFile *f, void *opaque, size_t size,
2835                               const VMStateField *field, QJSON *vmdesc)
2836 {
2837     return virtio_save(VIRTIO_DEVICE(opaque), f);
2838 }
2839 
2840 /* A wrapper for use as a VMState .get function */
2841 static int virtio_device_get(QEMUFile *f, void *opaque, size_t size,
2842                              const VMStateField *field)
2843 {
2844     VirtIODevice *vdev = VIRTIO_DEVICE(opaque);
2845     DeviceClass *dc = DEVICE_CLASS(VIRTIO_DEVICE_GET_CLASS(vdev));
2846 
2847     return virtio_load(vdev, f, dc->vmsd->version_id);
2848 }
2849 
2850 const VMStateInfo  virtio_vmstate_info = {
2851     .name = "virtio",
2852     .get = virtio_device_get,
2853     .put = virtio_device_put,
2854 };
2855 
2856 static int virtio_set_features_nocheck(VirtIODevice *vdev, uint64_t val)
2857 {
2858     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2859     bool bad = (val & ~(vdev->host_features)) != 0;
2860 
2861     val &= vdev->host_features;
2862     if (k->set_features) {
2863         k->set_features(vdev, val);
2864     }
2865     vdev->guest_features = val;
2866     return bad ? -1 : 0;
2867 }
2868 
2869 int virtio_set_features(VirtIODevice *vdev, uint64_t val)
2870 {
2871     int ret;
2872     /*
2873      * The driver must not attempt to set features after feature negotiation
2874      * has finished.
2875      */
2876     if (vdev->status & VIRTIO_CONFIG_S_FEATURES_OK) {
2877         return -EINVAL;
2878     }
2879     ret = virtio_set_features_nocheck(vdev, val);
2880     if (!ret) {
2881         if (virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX)) {
2882             /* VIRTIO_RING_F_EVENT_IDX changes the size of the caches.  */
2883             int i;
2884             for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2885                 if (vdev->vq[i].vring.num != 0) {
2886                     virtio_init_region_cache(vdev, i);
2887                 }
2888             }
2889         }
2890 
2891         if (!virtio_device_started(vdev, vdev->status) &&
2892             !virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
2893             vdev->start_on_kick = true;
2894         }
2895     }
2896     return ret;
2897 }
2898 
2899 size_t virtio_feature_get_config_size(VirtIOFeature *feature_sizes,
2900                                       uint64_t host_features)
2901 {
2902     size_t config_size = 0;
2903     int i;
2904 
2905     for (i = 0; feature_sizes[i].flags != 0; i++) {
2906         if (host_features & feature_sizes[i].flags) {
2907             config_size = MAX(feature_sizes[i].end, config_size);
2908         }
2909     }
2910 
2911     return config_size;
2912 }
2913 
2914 int virtio_load(VirtIODevice *vdev, QEMUFile *f, int version_id)
2915 {
2916     int i, ret;
2917     int32_t config_len;
2918     uint32_t num;
2919     uint32_t features;
2920     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2921     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
2922     VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev);
2923 
2924     /*
2925      * We poison the endianness to ensure it does not get used before
2926      * subsections have been loaded.
2927      */
2928     vdev->device_endian = VIRTIO_DEVICE_ENDIAN_UNKNOWN;
2929 
2930     if (k->load_config) {
2931         ret = k->load_config(qbus->parent, f);
2932         if (ret)
2933             return ret;
2934     }
2935 
2936     qemu_get_8s(f, &vdev->status);
2937     qemu_get_8s(f, &vdev->isr);
2938     qemu_get_be16s(f, &vdev->queue_sel);
2939     if (vdev->queue_sel >= VIRTIO_QUEUE_MAX) {
2940         return -1;
2941     }
2942     qemu_get_be32s(f, &features);
2943 
2944     /*
2945      * Temporarily set guest_features low bits - needed by
2946      * virtio net load code testing for VIRTIO_NET_F_CTRL_GUEST_OFFLOADS
2947      * VIRTIO_NET_F_GUEST_ANNOUNCE and VIRTIO_NET_F_CTRL_VQ.
2948      *
2949      * Note: devices should always test host features in future - don't create
2950      * new dependencies like this.
2951      */
2952     vdev->guest_features = features;
2953 
2954     config_len = qemu_get_be32(f);
2955 
2956     /*
2957      * There are cases where the incoming config can be bigger or smaller
2958      * than what we have; so load what we have space for, and skip
2959      * any excess that's in the stream.
2960      */
2961     qemu_get_buffer(f, vdev->config, MIN(config_len, vdev->config_len));
2962 
2963     while (config_len > vdev->config_len) {
2964         qemu_get_byte(f);
2965         config_len--;
2966     }
2967 
2968     num = qemu_get_be32(f);
2969 
2970     if (num > VIRTIO_QUEUE_MAX) {
2971         error_report("Invalid number of virtqueues: 0x%x", num);
2972         return -1;
2973     }
2974 
2975     for (i = 0; i < num; i++) {
2976         vdev->vq[i].vring.num = qemu_get_be32(f);
2977         if (k->has_variable_vring_alignment) {
2978             vdev->vq[i].vring.align = qemu_get_be32(f);
2979         }
2980         vdev->vq[i].vring.desc = qemu_get_be64(f);
2981         qemu_get_be16s(f, &vdev->vq[i].last_avail_idx);
2982         vdev->vq[i].signalled_used_valid = false;
2983         vdev->vq[i].notification = true;
2984 
2985         if (!vdev->vq[i].vring.desc && vdev->vq[i].last_avail_idx) {
2986             error_report("VQ %d address 0x0 "
2987                          "inconsistent with Host index 0x%x",
2988                          i, vdev->vq[i].last_avail_idx);
2989             return -1;
2990         }
2991         if (k->load_queue) {
2992             ret = k->load_queue(qbus->parent, i, f);
2993             if (ret)
2994                 return ret;
2995         }
2996     }
2997 
2998     virtio_notify_vector(vdev, VIRTIO_NO_VECTOR);
2999 
3000     if (vdc->load != NULL) {
3001         ret = vdc->load(vdev, f, version_id);
3002         if (ret) {
3003             return ret;
3004         }
3005     }
3006 
3007     if (vdc->vmsd) {
3008         ret = vmstate_load_state(f, vdc->vmsd, vdev, version_id);
3009         if (ret) {
3010             return ret;
3011         }
3012     }
3013 
3014     /* Subsections */
3015     ret = vmstate_load_state(f, &vmstate_virtio, vdev, 1);
3016     if (ret) {
3017         return ret;
3018     }
3019 
3020     if (vdev->device_endian == VIRTIO_DEVICE_ENDIAN_UNKNOWN) {
3021         vdev->device_endian = virtio_default_endian();
3022     }
3023 
3024     if (virtio_64bit_features_needed(vdev)) {
3025         /*
3026          * Subsection load filled vdev->guest_features.  Run them
3027          * through virtio_set_features to sanity-check them against
3028          * host_features.
3029          */
3030         uint64_t features64 = vdev->guest_features;
3031         if (virtio_set_features_nocheck(vdev, features64) < 0) {
3032             error_report("Features 0x%" PRIx64 " unsupported. "
3033                          "Allowed features: 0x%" PRIx64,
3034                          features64, vdev->host_features);
3035             return -1;
3036         }
3037     } else {
3038         if (virtio_set_features_nocheck(vdev, features) < 0) {
3039             error_report("Features 0x%x unsupported. "
3040                          "Allowed features: 0x%" PRIx64,
3041                          features, vdev->host_features);
3042             return -1;
3043         }
3044     }
3045 
3046     if (!virtio_device_started(vdev, vdev->status) &&
3047         !virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
3048         vdev->start_on_kick = true;
3049     }
3050 
3051     RCU_READ_LOCK_GUARD();
3052     for (i = 0; i < num; i++) {
3053         if (vdev->vq[i].vring.desc) {
3054             uint16_t nheads;
3055 
3056             /*
3057              * VIRTIO-1 devices migrate desc, used, and avail ring addresses so
3058              * only the region cache needs to be set up.  Legacy devices need
3059              * to calculate used and avail ring addresses based on the desc
3060              * address.
3061              */
3062             if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
3063                 virtio_init_region_cache(vdev, i);
3064             } else {
3065                 virtio_queue_update_rings(vdev, i);
3066             }
3067 
3068             if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3069                 vdev->vq[i].shadow_avail_idx = vdev->vq[i].last_avail_idx;
3070                 vdev->vq[i].shadow_avail_wrap_counter =
3071                                         vdev->vq[i].last_avail_wrap_counter;
3072                 continue;
3073             }
3074 
3075             nheads = vring_avail_idx(&vdev->vq[i]) - vdev->vq[i].last_avail_idx;
3076             /* Check it isn't doing strange things with descriptor numbers. */
3077             if (nheads > vdev->vq[i].vring.num) {
3078                 error_report("VQ %d size 0x%x Guest index 0x%x "
3079                              "inconsistent with Host index 0x%x: delta 0x%x",
3080                              i, vdev->vq[i].vring.num,
3081                              vring_avail_idx(&vdev->vq[i]),
3082                              vdev->vq[i].last_avail_idx, nheads);
3083                 return -1;
3084             }
3085             vdev->vq[i].used_idx = vring_used_idx(&vdev->vq[i]);
3086             vdev->vq[i].shadow_avail_idx = vring_avail_idx(&vdev->vq[i]);
3087 
3088             /*
3089              * Some devices migrate VirtQueueElements that have been popped
3090              * from the avail ring but not yet returned to the used ring.
3091              * Since max ring size < UINT16_MAX it's safe to use modulo
3092              * UINT16_MAX + 1 subtraction.
3093              */
3094             vdev->vq[i].inuse = (uint16_t)(vdev->vq[i].last_avail_idx -
3095                                 vdev->vq[i].used_idx);
3096             if (vdev->vq[i].inuse > vdev->vq[i].vring.num) {
3097                 error_report("VQ %d size 0x%x < last_avail_idx 0x%x - "
3098                              "used_idx 0x%x",
3099                              i, vdev->vq[i].vring.num,
3100                              vdev->vq[i].last_avail_idx,
3101                              vdev->vq[i].used_idx);
3102                 return -1;
3103             }
3104         }
3105     }
3106 
3107     if (vdc->post_load) {
3108         ret = vdc->post_load(vdev);
3109         if (ret) {
3110             return ret;
3111         }
3112     }
3113 
3114     return 0;
3115 }
3116 
3117 void virtio_cleanup(VirtIODevice *vdev)
3118 {
3119     qemu_del_vm_change_state_handler(vdev->vmstate);
3120 }
3121 
3122 static void virtio_vmstate_change(void *opaque, int running, RunState state)
3123 {
3124     VirtIODevice *vdev = opaque;
3125     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3126     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
3127     bool backend_run = running && virtio_device_started(vdev, vdev->status);
3128     vdev->vm_running = running;
3129 
3130     if (backend_run) {
3131         virtio_set_status(vdev, vdev->status);
3132     }
3133 
3134     if (k->vmstate_change) {
3135         k->vmstate_change(qbus->parent, backend_run);
3136     }
3137 
3138     if (!backend_run) {
3139         virtio_set_status(vdev, vdev->status);
3140     }
3141 }
3142 
3143 void virtio_instance_init_common(Object *proxy_obj, void *data,
3144                                  size_t vdev_size, const char *vdev_name)
3145 {
3146     DeviceState *vdev = data;
3147 
3148     object_initialize_child(proxy_obj, "virtio-backend", vdev, vdev_size,
3149                             vdev_name, &error_abort, NULL);
3150     qdev_alias_all_properties(vdev, proxy_obj);
3151 }
3152 
3153 void virtio_init(VirtIODevice *vdev, const char *name,
3154                  uint16_t device_id, size_t config_size)
3155 {
3156     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3157     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
3158     int i;
3159     int nvectors = k->query_nvectors ? k->query_nvectors(qbus->parent) : 0;
3160 
3161     if (nvectors) {
3162         vdev->vector_queues =
3163             g_malloc0(sizeof(*vdev->vector_queues) * nvectors);
3164     }
3165 
3166     vdev->start_on_kick = false;
3167     vdev->started = false;
3168     vdev->device_id = device_id;
3169     vdev->status = 0;
3170     atomic_set(&vdev->isr, 0);
3171     vdev->queue_sel = 0;
3172     vdev->config_vector = VIRTIO_NO_VECTOR;
3173     vdev->vq = g_malloc0(sizeof(VirtQueue) * VIRTIO_QUEUE_MAX);
3174     vdev->vm_running = runstate_is_running();
3175     vdev->broken = false;
3176     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
3177         vdev->vq[i].vector = VIRTIO_NO_VECTOR;
3178         vdev->vq[i].vdev = vdev;
3179         vdev->vq[i].queue_index = i;
3180         vdev->vq[i].host_notifier_enabled = false;
3181     }
3182 
3183     vdev->name = name;
3184     vdev->config_len = config_size;
3185     if (vdev->config_len) {
3186         vdev->config = g_malloc0(config_size);
3187     } else {
3188         vdev->config = NULL;
3189     }
3190     vdev->vmstate = qdev_add_vm_change_state_handler(DEVICE(vdev),
3191             virtio_vmstate_change, vdev);
3192     vdev->device_endian = virtio_default_endian();
3193     vdev->use_guest_notifier_mask = true;
3194 }
3195 
3196 hwaddr virtio_queue_get_desc_addr(VirtIODevice *vdev, int n)
3197 {
3198     return vdev->vq[n].vring.desc;
3199 }
3200 
3201 bool virtio_queue_enabled(VirtIODevice *vdev, int n)
3202 {
3203     return virtio_queue_get_desc_addr(vdev, n) != 0;
3204 }
3205 
3206 hwaddr virtio_queue_get_avail_addr(VirtIODevice *vdev, int n)
3207 {
3208     return vdev->vq[n].vring.avail;
3209 }
3210 
3211 hwaddr virtio_queue_get_used_addr(VirtIODevice *vdev, int n)
3212 {
3213     return vdev->vq[n].vring.used;
3214 }
3215 
3216 hwaddr virtio_queue_get_desc_size(VirtIODevice *vdev, int n)
3217 {
3218     return sizeof(VRingDesc) * vdev->vq[n].vring.num;
3219 }
3220 
3221 hwaddr virtio_queue_get_avail_size(VirtIODevice *vdev, int n)
3222 {
3223     int s;
3224 
3225     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3226         return sizeof(struct VRingPackedDescEvent);
3227     }
3228 
3229     s = virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0;
3230     return offsetof(VRingAvail, ring) +
3231         sizeof(uint16_t) * vdev->vq[n].vring.num + s;
3232 }
3233 
3234 hwaddr virtio_queue_get_used_size(VirtIODevice *vdev, int n)
3235 {
3236     int s;
3237 
3238     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3239         return sizeof(struct VRingPackedDescEvent);
3240     }
3241 
3242     s = virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0;
3243     return offsetof(VRingUsed, ring) +
3244         sizeof(VRingUsedElem) * vdev->vq[n].vring.num + s;
3245 }
3246 
3247 static unsigned int virtio_queue_packed_get_last_avail_idx(VirtIODevice *vdev,
3248                                                            int n)
3249 {
3250     unsigned int avail, used;
3251 
3252     avail = vdev->vq[n].last_avail_idx;
3253     avail |= ((uint16_t)vdev->vq[n].last_avail_wrap_counter) << 15;
3254 
3255     used = vdev->vq[n].used_idx;
3256     used |= ((uint16_t)vdev->vq[n].used_wrap_counter) << 15;
3257 
3258     return avail | used << 16;
3259 }
3260 
3261 static uint16_t virtio_queue_split_get_last_avail_idx(VirtIODevice *vdev,
3262                                                       int n)
3263 {
3264     return vdev->vq[n].last_avail_idx;
3265 }
3266 
3267 unsigned int virtio_queue_get_last_avail_idx(VirtIODevice *vdev, int n)
3268 {
3269     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3270         return virtio_queue_packed_get_last_avail_idx(vdev, n);
3271     } else {
3272         return virtio_queue_split_get_last_avail_idx(vdev, n);
3273     }
3274 }
3275 
3276 static void virtio_queue_packed_set_last_avail_idx(VirtIODevice *vdev,
3277                                                    int n, unsigned int idx)
3278 {
3279     struct VirtQueue *vq = &vdev->vq[n];
3280 
3281     vq->last_avail_idx = vq->shadow_avail_idx = idx & 0x7fff;
3282     vq->last_avail_wrap_counter =
3283         vq->shadow_avail_wrap_counter = !!(idx & 0x8000);
3284     idx >>= 16;
3285     vq->used_idx = idx & 0x7ffff;
3286     vq->used_wrap_counter = !!(idx & 0x8000);
3287 }
3288 
3289 static void virtio_queue_split_set_last_avail_idx(VirtIODevice *vdev,
3290                                                   int n, unsigned int idx)
3291 {
3292         vdev->vq[n].last_avail_idx = idx;
3293         vdev->vq[n].shadow_avail_idx = idx;
3294 }
3295 
3296 void virtio_queue_set_last_avail_idx(VirtIODevice *vdev, int n,
3297                                      unsigned int idx)
3298 {
3299     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3300         virtio_queue_packed_set_last_avail_idx(vdev, n, idx);
3301     } else {
3302         virtio_queue_split_set_last_avail_idx(vdev, n, idx);
3303     }
3304 }
3305 
3306 static void virtio_queue_packed_restore_last_avail_idx(VirtIODevice *vdev,
3307                                                        int n)
3308 {
3309     /* We don't have a reference like avail idx in shared memory */
3310     return;
3311 }
3312 
3313 static void virtio_queue_split_restore_last_avail_idx(VirtIODevice *vdev,
3314                                                       int n)
3315 {
3316     RCU_READ_LOCK_GUARD();
3317     if (vdev->vq[n].vring.desc) {
3318         vdev->vq[n].last_avail_idx = vring_used_idx(&vdev->vq[n]);
3319         vdev->vq[n].shadow_avail_idx = vdev->vq[n].last_avail_idx;
3320     }
3321 }
3322 
3323 void virtio_queue_restore_last_avail_idx(VirtIODevice *vdev, int n)
3324 {
3325     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3326         virtio_queue_packed_restore_last_avail_idx(vdev, n);
3327     } else {
3328         virtio_queue_split_restore_last_avail_idx(vdev, n);
3329     }
3330 }
3331 
3332 static void virtio_queue_packed_update_used_idx(VirtIODevice *vdev, int n)
3333 {
3334     /* used idx was updated through set_last_avail_idx() */
3335     return;
3336 }
3337 
3338 static void virtio_split_packed_update_used_idx(VirtIODevice *vdev, int n)
3339 {
3340     RCU_READ_LOCK_GUARD();
3341     if (vdev->vq[n].vring.desc) {
3342         vdev->vq[n].used_idx = vring_used_idx(&vdev->vq[n]);
3343     }
3344 }
3345 
3346 void virtio_queue_update_used_idx(VirtIODevice *vdev, int n)
3347 {
3348     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3349         return virtio_queue_packed_update_used_idx(vdev, n);
3350     } else {
3351         return virtio_split_packed_update_used_idx(vdev, n);
3352     }
3353 }
3354 
3355 void virtio_queue_invalidate_signalled_used(VirtIODevice *vdev, int n)
3356 {
3357     vdev->vq[n].signalled_used_valid = false;
3358 }
3359 
3360 VirtQueue *virtio_get_queue(VirtIODevice *vdev, int n)
3361 {
3362     return vdev->vq + n;
3363 }
3364 
3365 uint16_t virtio_get_queue_index(VirtQueue *vq)
3366 {
3367     return vq->queue_index;
3368 }
3369 
3370 static void virtio_queue_guest_notifier_read(EventNotifier *n)
3371 {
3372     VirtQueue *vq = container_of(n, VirtQueue, guest_notifier);
3373     if (event_notifier_test_and_clear(n)) {
3374         virtio_irq(vq);
3375     }
3376 }
3377 
3378 void virtio_queue_set_guest_notifier_fd_handler(VirtQueue *vq, bool assign,
3379                                                 bool with_irqfd)
3380 {
3381     if (assign && !with_irqfd) {
3382         event_notifier_set_handler(&vq->guest_notifier,
3383                                    virtio_queue_guest_notifier_read);
3384     } else {
3385         event_notifier_set_handler(&vq->guest_notifier, NULL);
3386     }
3387     if (!assign) {
3388         /* Test and clear notifier before closing it,
3389          * in case poll callback didn't have time to run. */
3390         virtio_queue_guest_notifier_read(&vq->guest_notifier);
3391     }
3392 }
3393 
3394 EventNotifier *virtio_queue_get_guest_notifier(VirtQueue *vq)
3395 {
3396     return &vq->guest_notifier;
3397 }
3398 
3399 static void virtio_queue_host_notifier_aio_read(EventNotifier *n)
3400 {
3401     VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
3402     if (event_notifier_test_and_clear(n)) {
3403         virtio_queue_notify_aio_vq(vq);
3404     }
3405 }
3406 
3407 static void virtio_queue_host_notifier_aio_poll_begin(EventNotifier *n)
3408 {
3409     VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
3410 
3411     virtio_queue_set_notification(vq, 0);
3412 }
3413 
3414 static bool virtio_queue_host_notifier_aio_poll(void *opaque)
3415 {
3416     EventNotifier *n = opaque;
3417     VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
3418 
3419     if (!vq->vring.desc || virtio_queue_empty(vq)) {
3420         return false;
3421     }
3422 
3423     return virtio_queue_notify_aio_vq(vq);
3424 }
3425 
3426 static void virtio_queue_host_notifier_aio_poll_end(EventNotifier *n)
3427 {
3428     VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
3429 
3430     /* Caller polls once more after this to catch requests that race with us */
3431     virtio_queue_set_notification(vq, 1);
3432 }
3433 
3434 void virtio_queue_aio_set_host_notifier_handler(VirtQueue *vq, AioContext *ctx,
3435                                                 VirtIOHandleAIOOutput handle_output)
3436 {
3437     if (handle_output) {
3438         vq->handle_aio_output = handle_output;
3439         aio_set_event_notifier(ctx, &vq->host_notifier, true,
3440                                virtio_queue_host_notifier_aio_read,
3441                                virtio_queue_host_notifier_aio_poll);
3442         aio_set_event_notifier_poll(ctx, &vq->host_notifier,
3443                                     virtio_queue_host_notifier_aio_poll_begin,
3444                                     virtio_queue_host_notifier_aio_poll_end);
3445     } else {
3446         aio_set_event_notifier(ctx, &vq->host_notifier, true, NULL, NULL);
3447         /* Test and clear notifier before after disabling event,
3448          * in case poll callback didn't have time to run. */
3449         virtio_queue_host_notifier_aio_read(&vq->host_notifier);
3450         vq->handle_aio_output = NULL;
3451     }
3452 }
3453 
3454 void virtio_queue_host_notifier_read(EventNotifier *n)
3455 {
3456     VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
3457     if (event_notifier_test_and_clear(n)) {
3458         virtio_queue_notify_vq(vq);
3459     }
3460 }
3461 
3462 EventNotifier *virtio_queue_get_host_notifier(VirtQueue *vq)
3463 {
3464     return &vq->host_notifier;
3465 }
3466 
3467 void virtio_queue_set_host_notifier_enabled(VirtQueue *vq, bool enabled)
3468 {
3469     vq->host_notifier_enabled = enabled;
3470 }
3471 
3472 int virtio_queue_set_host_notifier_mr(VirtIODevice *vdev, int n,
3473                                       MemoryRegion *mr, bool assign)
3474 {
3475     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3476     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
3477 
3478     if (k->set_host_notifier_mr) {
3479         return k->set_host_notifier_mr(qbus->parent, n, mr, assign);
3480     }
3481 
3482     return -1;
3483 }
3484 
3485 void virtio_device_set_child_bus_name(VirtIODevice *vdev, char *bus_name)
3486 {
3487     g_free(vdev->bus_name);
3488     vdev->bus_name = g_strdup(bus_name);
3489 }
3490 
3491 void GCC_FMT_ATTR(2, 3) virtio_error(VirtIODevice *vdev, const char *fmt, ...)
3492 {
3493     va_list ap;
3494 
3495     va_start(ap, fmt);
3496     error_vreport(fmt, ap);
3497     va_end(ap);
3498 
3499     if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
3500         vdev->status = vdev->status | VIRTIO_CONFIG_S_NEEDS_RESET;
3501         virtio_notify_config(vdev);
3502     }
3503 
3504     vdev->broken = true;
3505 }
3506 
3507 static void virtio_memory_listener_commit(MemoryListener *listener)
3508 {
3509     VirtIODevice *vdev = container_of(listener, VirtIODevice, listener);
3510     int i;
3511 
3512     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
3513         if (vdev->vq[i].vring.num == 0) {
3514             break;
3515         }
3516         virtio_init_region_cache(vdev, i);
3517     }
3518 }
3519 
3520 static void virtio_device_realize(DeviceState *dev, Error **errp)
3521 {
3522     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
3523     VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(dev);
3524     Error *err = NULL;
3525 
3526     /* Devices should either use vmsd or the load/save methods */
3527     assert(!vdc->vmsd || !vdc->load);
3528 
3529     if (vdc->realize != NULL) {
3530         vdc->realize(dev, &err);
3531         if (err != NULL) {
3532             error_propagate(errp, err);
3533             return;
3534         }
3535     }
3536 
3537     virtio_bus_device_plugged(vdev, &err);
3538     if (err != NULL) {
3539         error_propagate(errp, err);
3540         vdc->unrealize(dev, NULL);
3541         return;
3542     }
3543 
3544     vdev->listener.commit = virtio_memory_listener_commit;
3545     memory_listener_register(&vdev->listener, vdev->dma_as);
3546 }
3547 
3548 static void virtio_device_unrealize(DeviceState *dev, Error **errp)
3549 {
3550     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
3551     VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(dev);
3552     Error *err = NULL;
3553 
3554     virtio_bus_device_unplugged(vdev);
3555 
3556     if (vdc->unrealize != NULL) {
3557         vdc->unrealize(dev, &err);
3558         if (err != NULL) {
3559             error_propagate(errp, err);
3560             return;
3561         }
3562     }
3563 
3564     g_free(vdev->bus_name);
3565     vdev->bus_name = NULL;
3566 }
3567 
3568 static void virtio_device_free_virtqueues(VirtIODevice *vdev)
3569 {
3570     int i;
3571     if (!vdev->vq) {
3572         return;
3573     }
3574 
3575     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
3576         if (vdev->vq[i].vring.num == 0) {
3577             break;
3578         }
3579         virtio_virtqueue_reset_region_cache(&vdev->vq[i]);
3580     }
3581     g_free(vdev->vq);
3582 }
3583 
3584 static void virtio_device_instance_finalize(Object *obj)
3585 {
3586     VirtIODevice *vdev = VIRTIO_DEVICE(obj);
3587 
3588     memory_listener_unregister(&vdev->listener);
3589     virtio_device_free_virtqueues(vdev);
3590 
3591     g_free(vdev->config);
3592     g_free(vdev->vector_queues);
3593 }
3594 
3595 static Property virtio_properties[] = {
3596     DEFINE_VIRTIO_COMMON_FEATURES(VirtIODevice, host_features),
3597     DEFINE_PROP_BOOL("use-started", VirtIODevice, use_started, true),
3598     DEFINE_PROP_BOOL("use-disabled-flag", VirtIODevice, use_disabled_flag, true),
3599     DEFINE_PROP_END_OF_LIST(),
3600 };
3601 
3602 static int virtio_device_start_ioeventfd_impl(VirtIODevice *vdev)
3603 {
3604     VirtioBusState *qbus = VIRTIO_BUS(qdev_get_parent_bus(DEVICE(vdev)));
3605     int i, n, r, err;
3606 
3607     memory_region_transaction_begin();
3608     for (n = 0; n < VIRTIO_QUEUE_MAX; n++) {
3609         VirtQueue *vq = &vdev->vq[n];
3610         if (!virtio_queue_get_num(vdev, n)) {
3611             continue;
3612         }
3613         r = virtio_bus_set_host_notifier(qbus, n, true);
3614         if (r < 0) {
3615             err = r;
3616             goto assign_error;
3617         }
3618         event_notifier_set_handler(&vq->host_notifier,
3619                                    virtio_queue_host_notifier_read);
3620     }
3621 
3622     for (n = 0; n < VIRTIO_QUEUE_MAX; n++) {
3623         /* Kick right away to begin processing requests already in vring */
3624         VirtQueue *vq = &vdev->vq[n];
3625         if (!vq->vring.num) {
3626             continue;
3627         }
3628         event_notifier_set(&vq->host_notifier);
3629     }
3630     memory_region_transaction_commit();
3631     return 0;
3632 
3633 assign_error:
3634     i = n; /* save n for a second iteration after transaction is committed. */
3635     while (--n >= 0) {
3636         VirtQueue *vq = &vdev->vq[n];
3637         if (!virtio_queue_get_num(vdev, n)) {
3638             continue;
3639         }
3640 
3641         event_notifier_set_handler(&vq->host_notifier, NULL);
3642         r = virtio_bus_set_host_notifier(qbus, n, false);
3643         assert(r >= 0);
3644     }
3645     memory_region_transaction_commit();
3646 
3647     while (--i >= 0) {
3648         if (!virtio_queue_get_num(vdev, i)) {
3649             continue;
3650         }
3651         virtio_bus_cleanup_host_notifier(qbus, i);
3652     }
3653     return err;
3654 }
3655 
3656 int virtio_device_start_ioeventfd(VirtIODevice *vdev)
3657 {
3658     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3659     VirtioBusState *vbus = VIRTIO_BUS(qbus);
3660 
3661     return virtio_bus_start_ioeventfd(vbus);
3662 }
3663 
3664 static void virtio_device_stop_ioeventfd_impl(VirtIODevice *vdev)
3665 {
3666     VirtioBusState *qbus = VIRTIO_BUS(qdev_get_parent_bus(DEVICE(vdev)));
3667     int n, r;
3668 
3669     memory_region_transaction_begin();
3670     for (n = 0; n < VIRTIO_QUEUE_MAX; n++) {
3671         VirtQueue *vq = &vdev->vq[n];
3672 
3673         if (!virtio_queue_get_num(vdev, n)) {
3674             continue;
3675         }
3676         event_notifier_set_handler(&vq->host_notifier, NULL);
3677         r = virtio_bus_set_host_notifier(qbus, n, false);
3678         assert(r >= 0);
3679     }
3680     memory_region_transaction_commit();
3681 
3682     for (n = 0; n < VIRTIO_QUEUE_MAX; n++) {
3683         if (!virtio_queue_get_num(vdev, n)) {
3684             continue;
3685         }
3686         virtio_bus_cleanup_host_notifier(qbus, n);
3687     }
3688 }
3689 
3690 int virtio_device_grab_ioeventfd(VirtIODevice *vdev)
3691 {
3692     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3693     VirtioBusState *vbus = VIRTIO_BUS(qbus);
3694 
3695     return virtio_bus_grab_ioeventfd(vbus);
3696 }
3697 
3698 void virtio_device_release_ioeventfd(VirtIODevice *vdev)
3699 {
3700     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3701     VirtioBusState *vbus = VIRTIO_BUS(qbus);
3702 
3703     virtio_bus_release_ioeventfd(vbus);
3704 }
3705 
3706 static void virtio_device_class_init(ObjectClass *klass, void *data)
3707 {
3708     /* Set the default value here. */
3709     VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
3710     DeviceClass *dc = DEVICE_CLASS(klass);
3711 
3712     dc->realize = virtio_device_realize;
3713     dc->unrealize = virtio_device_unrealize;
3714     dc->bus_type = TYPE_VIRTIO_BUS;
3715     dc->props = virtio_properties;
3716     vdc->start_ioeventfd = virtio_device_start_ioeventfd_impl;
3717     vdc->stop_ioeventfd = virtio_device_stop_ioeventfd_impl;
3718 
3719     vdc->legacy_features |= VIRTIO_LEGACY_FEATURES;
3720 }
3721 
3722 bool virtio_device_ioeventfd_enabled(VirtIODevice *vdev)
3723 {
3724     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3725     VirtioBusState *vbus = VIRTIO_BUS(qbus);
3726 
3727     return virtio_bus_ioeventfd_enabled(vbus);
3728 }
3729 
3730 static const TypeInfo virtio_device_info = {
3731     .name = TYPE_VIRTIO_DEVICE,
3732     .parent = TYPE_DEVICE,
3733     .instance_size = sizeof(VirtIODevice),
3734     .class_init = virtio_device_class_init,
3735     .instance_finalize = virtio_device_instance_finalize,
3736     .abstract = true,
3737     .class_size = sizeof(VirtioDeviceClass),
3738 };
3739 
3740 static void virtio_register_types(void)
3741 {
3742     type_register_static(&virtio_device_info);
3743 }
3744 
3745 type_init(virtio_register_types)
3746