xref: /openbmc/qemu/hw/virtio/virtio.c (revision 159fb790)
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 "qapi/qapi-commands-virtio.h"
17 #include "trace.h"
18 #include "qemu/defer-call.h"
19 #include "qemu/error-report.h"
20 #include "qemu/log.h"
21 #include "qemu/main-loop.h"
22 #include "qemu/module.h"
23 #include "qom/object_interfaces.h"
24 #include "hw/core/cpu.h"
25 #include "hw/virtio/virtio.h"
26 #include "hw/virtio/vhost.h"
27 #include "migration/qemu-file-types.h"
28 #include "qemu/atomic.h"
29 #include "hw/virtio/virtio-bus.h"
30 #include "hw/qdev-properties.h"
31 #include "hw/virtio/virtio-access.h"
32 #include "sysemu/dma.h"
33 #include "sysemu/runstate.h"
34 #include "virtio-qmp.h"
35 
36 #include "standard-headers/linux/virtio_ids.h"
37 #include "standard-headers/linux/vhost_types.h"
38 #include "standard-headers/linux/virtio_blk.h"
39 #include "standard-headers/linux/virtio_console.h"
40 #include "standard-headers/linux/virtio_gpu.h"
41 #include "standard-headers/linux/virtio_net.h"
42 #include "standard-headers/linux/virtio_scsi.h"
43 #include "standard-headers/linux/virtio_i2c.h"
44 #include "standard-headers/linux/virtio_balloon.h"
45 #include "standard-headers/linux/virtio_iommu.h"
46 #include "standard-headers/linux/virtio_mem.h"
47 #include "standard-headers/linux/virtio_vsock.h"
48 
49 /*
50  * Maximum size of virtio device config space
51  */
52 #define VHOST_USER_MAX_CONFIG_SIZE 256
53 
54 /*
55  * The alignment to use between consumer and producer parts of vring.
56  * x86 pagesize again. This is the default, used by transports like PCI
57  * which don't provide a means for the guest to tell the host the alignment.
58  */
59 #define VIRTIO_PCI_VRING_ALIGN         4096
60 
61 typedef struct VRingDesc
62 {
63     uint64_t addr;
64     uint32_t len;
65     uint16_t flags;
66     uint16_t next;
67 } VRingDesc;
68 
69 typedef struct VRingPackedDesc {
70     uint64_t addr;
71     uint32_t len;
72     uint16_t id;
73     uint16_t flags;
74 } VRingPackedDesc;
75 
76 typedef struct VRingAvail
77 {
78     uint16_t flags;
79     uint16_t idx;
80     uint16_t ring[];
81 } VRingAvail;
82 
83 typedef struct VRingUsedElem
84 {
85     uint32_t id;
86     uint32_t len;
87 } VRingUsedElem;
88 
89 typedef struct VRingUsed
90 {
91     uint16_t flags;
92     uint16_t idx;
93     VRingUsedElem ring[];
94 } VRingUsed;
95 
96 typedef struct VRingMemoryRegionCaches {
97     struct rcu_head rcu;
98     MemoryRegionCache desc;
99     MemoryRegionCache avail;
100     MemoryRegionCache used;
101 } VRingMemoryRegionCaches;
102 
103 typedef struct VRing
104 {
105     unsigned int num;
106     unsigned int num_default;
107     unsigned int align;
108     hwaddr desc;
109     hwaddr avail;
110     hwaddr used;
111     VRingMemoryRegionCaches *caches;
112 } VRing;
113 
114 typedef struct VRingPackedDescEvent {
115     uint16_t off_wrap;
116     uint16_t flags;
117 } VRingPackedDescEvent ;
118 
119 struct VirtQueue
120 {
121     VRing vring;
122     VirtQueueElement *used_elems;
123 
124     /* Next head to pop */
125     uint16_t last_avail_idx;
126     bool last_avail_wrap_counter;
127 
128     /* Last avail_idx read from VQ. */
129     uint16_t shadow_avail_idx;
130     bool shadow_avail_wrap_counter;
131 
132     uint16_t used_idx;
133     bool used_wrap_counter;
134 
135     /* Last used index value we have signalled on */
136     uint16_t signalled_used;
137 
138     /* Last used index value we have signalled on */
139     bool signalled_used_valid;
140 
141     /* Notification enabled? */
142     bool notification;
143 
144     uint16_t queue_index;
145 
146     unsigned int inuse;
147 
148     uint16_t vector;
149     VirtIOHandleOutput handle_output;
150     VirtIODevice *vdev;
151     EventNotifier guest_notifier;
152     EventNotifier host_notifier;
153     bool host_notifier_enabled;
154     QLIST_ENTRY(VirtQueue) node;
155 };
156 
157 const char *virtio_device_names[] = {
158     [VIRTIO_ID_NET] = "virtio-net",
159     [VIRTIO_ID_BLOCK] = "virtio-blk",
160     [VIRTIO_ID_CONSOLE] = "virtio-serial",
161     [VIRTIO_ID_RNG] = "virtio-rng",
162     [VIRTIO_ID_BALLOON] = "virtio-balloon",
163     [VIRTIO_ID_IOMEM] = "virtio-iomem",
164     [VIRTIO_ID_RPMSG] = "virtio-rpmsg",
165     [VIRTIO_ID_SCSI] = "virtio-scsi",
166     [VIRTIO_ID_9P] = "virtio-9p",
167     [VIRTIO_ID_MAC80211_WLAN] = "virtio-mac-wlan",
168     [VIRTIO_ID_RPROC_SERIAL] = "virtio-rproc-serial",
169     [VIRTIO_ID_CAIF] = "virtio-caif",
170     [VIRTIO_ID_MEMORY_BALLOON] = "virtio-mem-balloon",
171     [VIRTIO_ID_GPU] = "virtio-gpu",
172     [VIRTIO_ID_CLOCK] = "virtio-clk",
173     [VIRTIO_ID_INPUT] = "virtio-input",
174     [VIRTIO_ID_VSOCK] = "vhost-vsock",
175     [VIRTIO_ID_CRYPTO] = "virtio-crypto",
176     [VIRTIO_ID_SIGNAL_DIST] = "virtio-signal",
177     [VIRTIO_ID_PSTORE] = "virtio-pstore",
178     [VIRTIO_ID_IOMMU] = "virtio-iommu",
179     [VIRTIO_ID_MEM] = "virtio-mem",
180     [VIRTIO_ID_SOUND] = "virtio-sound",
181     [VIRTIO_ID_FS] = "virtio-user-fs",
182     [VIRTIO_ID_PMEM] = "virtio-pmem",
183     [VIRTIO_ID_RPMB] = "virtio-rpmb",
184     [VIRTIO_ID_MAC80211_HWSIM] = "virtio-mac-hwsim",
185     [VIRTIO_ID_VIDEO_ENCODER] = "virtio-vid-encoder",
186     [VIRTIO_ID_VIDEO_DECODER] = "virtio-vid-decoder",
187     [VIRTIO_ID_SCMI] = "virtio-scmi",
188     [VIRTIO_ID_NITRO_SEC_MOD] = "virtio-nitro-sec-mod",
189     [VIRTIO_ID_I2C_ADAPTER] = "vhost-user-i2c",
190     [VIRTIO_ID_WATCHDOG] = "virtio-watchdog",
191     [VIRTIO_ID_CAN] = "virtio-can",
192     [VIRTIO_ID_DMABUF] = "virtio-dmabuf",
193     [VIRTIO_ID_PARAM_SERV] = "virtio-param-serv",
194     [VIRTIO_ID_AUDIO_POLICY] = "virtio-audio-pol",
195     [VIRTIO_ID_BT] = "virtio-bluetooth",
196     [VIRTIO_ID_GPIO] = "virtio-gpio"
197 };
198 
199 static const char *virtio_id_to_name(uint16_t device_id)
200 {
201     assert(device_id < G_N_ELEMENTS(virtio_device_names));
202     const char *name = virtio_device_names[device_id];
203     assert(name != NULL);
204     return name;
205 }
206 
207 /* Called within call_rcu().  */
208 static void virtio_free_region_cache(VRingMemoryRegionCaches *caches)
209 {
210     assert(caches != NULL);
211     address_space_cache_destroy(&caches->desc);
212     address_space_cache_destroy(&caches->avail);
213     address_space_cache_destroy(&caches->used);
214     g_free(caches);
215 }
216 
217 static void virtio_virtqueue_reset_region_cache(struct VirtQueue *vq)
218 {
219     VRingMemoryRegionCaches *caches;
220 
221     caches = qatomic_read(&vq->vring.caches);
222     qatomic_rcu_set(&vq->vring.caches, NULL);
223     if (caches) {
224         call_rcu(caches, virtio_free_region_cache, rcu);
225     }
226 }
227 
228 void virtio_init_region_cache(VirtIODevice *vdev, int n)
229 {
230     VirtQueue *vq = &vdev->vq[n];
231     VRingMemoryRegionCaches *old = vq->vring.caches;
232     VRingMemoryRegionCaches *new = NULL;
233     hwaddr addr, size;
234     int64_t len;
235     bool packed;
236 
237 
238     addr = vq->vring.desc;
239     if (!addr) {
240         goto out_no_cache;
241     }
242     new = g_new0(VRingMemoryRegionCaches, 1);
243     size = virtio_queue_get_desc_size(vdev, n);
244     packed = virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED) ?
245                                    true : false;
246     len = address_space_cache_init(&new->desc, vdev->dma_as,
247                                    addr, size, packed);
248     if (len < size) {
249         virtio_error(vdev, "Cannot map desc");
250         goto err_desc;
251     }
252 
253     size = virtio_queue_get_used_size(vdev, n);
254     len = address_space_cache_init(&new->used, vdev->dma_as,
255                                    vq->vring.used, size, true);
256     if (len < size) {
257         virtio_error(vdev, "Cannot map used");
258         goto err_used;
259     }
260 
261     size = virtio_queue_get_avail_size(vdev, n);
262     len = address_space_cache_init(&new->avail, vdev->dma_as,
263                                    vq->vring.avail, size, false);
264     if (len < size) {
265         virtio_error(vdev, "Cannot map avail");
266         goto err_avail;
267     }
268 
269     qatomic_rcu_set(&vq->vring.caches, new);
270     if (old) {
271         call_rcu(old, virtio_free_region_cache, rcu);
272     }
273     return;
274 
275 err_avail:
276     address_space_cache_destroy(&new->avail);
277 err_used:
278     address_space_cache_destroy(&new->used);
279 err_desc:
280     address_space_cache_destroy(&new->desc);
281 out_no_cache:
282     g_free(new);
283     virtio_virtqueue_reset_region_cache(vq);
284 }
285 
286 /* virt queue functions */
287 void virtio_queue_update_rings(VirtIODevice *vdev, int n)
288 {
289     VRing *vring = &vdev->vq[n].vring;
290 
291     if (!vring->num || !vring->desc || !vring->align) {
292         /* not yet setup -> nothing to do */
293         return;
294     }
295     vring->avail = vring->desc + vring->num * sizeof(VRingDesc);
296     vring->used = vring_align(vring->avail +
297                               offsetof(VRingAvail, ring[vring->num]),
298                               vring->align);
299     virtio_init_region_cache(vdev, n);
300 }
301 
302 /* Called within rcu_read_lock().  */
303 static void vring_split_desc_read(VirtIODevice *vdev, VRingDesc *desc,
304                                   MemoryRegionCache *cache, int i)
305 {
306     address_space_read_cached(cache, i * sizeof(VRingDesc),
307                               desc, sizeof(VRingDesc));
308     virtio_tswap64s(vdev, &desc->addr);
309     virtio_tswap32s(vdev, &desc->len);
310     virtio_tswap16s(vdev, &desc->flags);
311     virtio_tswap16s(vdev, &desc->next);
312 }
313 
314 static void vring_packed_event_read(VirtIODevice *vdev,
315                                     MemoryRegionCache *cache,
316                                     VRingPackedDescEvent *e)
317 {
318     hwaddr off_off = offsetof(VRingPackedDescEvent, off_wrap);
319     hwaddr off_flags = offsetof(VRingPackedDescEvent, flags);
320 
321     e->flags = virtio_lduw_phys_cached(vdev, cache, off_flags);
322     /* Make sure flags is seen before off_wrap */
323     smp_rmb();
324     e->off_wrap = virtio_lduw_phys_cached(vdev, cache, off_off);
325     virtio_tswap16s(vdev, &e->flags);
326 }
327 
328 static void vring_packed_off_wrap_write(VirtIODevice *vdev,
329                                         MemoryRegionCache *cache,
330                                         uint16_t off_wrap)
331 {
332     hwaddr off = offsetof(VRingPackedDescEvent, off_wrap);
333 
334     virtio_stw_phys_cached(vdev, cache, off, off_wrap);
335     address_space_cache_invalidate(cache, off, sizeof(off_wrap));
336 }
337 
338 static void vring_packed_flags_write(VirtIODevice *vdev,
339                                      MemoryRegionCache *cache, uint16_t flags)
340 {
341     hwaddr off = offsetof(VRingPackedDescEvent, flags);
342 
343     virtio_stw_phys_cached(vdev, cache, off, flags);
344     address_space_cache_invalidate(cache, off, sizeof(flags));
345 }
346 
347 /* Called within rcu_read_lock().  */
348 static VRingMemoryRegionCaches *vring_get_region_caches(struct VirtQueue *vq)
349 {
350     return qatomic_rcu_read(&vq->vring.caches);
351 }
352 
353 /* Called within rcu_read_lock().  */
354 static inline uint16_t vring_avail_flags(VirtQueue *vq)
355 {
356     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
357     hwaddr pa = offsetof(VRingAvail, flags);
358 
359     if (!caches) {
360         return 0;
361     }
362 
363     return virtio_lduw_phys_cached(vq->vdev, &caches->avail, pa);
364 }
365 
366 /* Called within rcu_read_lock().  */
367 static inline uint16_t vring_avail_idx(VirtQueue *vq)
368 {
369     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
370     hwaddr pa = offsetof(VRingAvail, idx);
371 
372     if (!caches) {
373         return 0;
374     }
375 
376     vq->shadow_avail_idx = virtio_lduw_phys_cached(vq->vdev, &caches->avail, pa);
377     return vq->shadow_avail_idx;
378 }
379 
380 /* Called within rcu_read_lock().  */
381 static inline uint16_t vring_avail_ring(VirtQueue *vq, int i)
382 {
383     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
384     hwaddr pa = offsetof(VRingAvail, ring[i]);
385 
386     if (!caches) {
387         return 0;
388     }
389 
390     return virtio_lduw_phys_cached(vq->vdev, &caches->avail, pa);
391 }
392 
393 /* Called within rcu_read_lock().  */
394 static inline uint16_t vring_get_used_event(VirtQueue *vq)
395 {
396     return vring_avail_ring(vq, vq->vring.num);
397 }
398 
399 /* Called within rcu_read_lock().  */
400 static inline void vring_used_write(VirtQueue *vq, VRingUsedElem *uelem,
401                                     int i)
402 {
403     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
404     hwaddr pa = offsetof(VRingUsed, ring[i]);
405 
406     if (!caches) {
407         return;
408     }
409 
410     virtio_tswap32s(vq->vdev, &uelem->id);
411     virtio_tswap32s(vq->vdev, &uelem->len);
412     address_space_write_cached(&caches->used, pa, uelem, sizeof(VRingUsedElem));
413     address_space_cache_invalidate(&caches->used, pa, sizeof(VRingUsedElem));
414 }
415 
416 /* Called within rcu_read_lock(). */
417 static inline uint16_t vring_used_flags(VirtQueue *vq)
418 {
419     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
420     hwaddr pa = offsetof(VRingUsed, flags);
421 
422     if (!caches) {
423         return 0;
424     }
425 
426     return virtio_lduw_phys_cached(vq->vdev, &caches->used, pa);
427 }
428 
429 /* Called within rcu_read_lock().  */
430 static uint16_t vring_used_idx(VirtQueue *vq)
431 {
432     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
433     hwaddr pa = offsetof(VRingUsed, idx);
434 
435     if (!caches) {
436         return 0;
437     }
438 
439     return virtio_lduw_phys_cached(vq->vdev, &caches->used, pa);
440 }
441 
442 /* Called within rcu_read_lock().  */
443 static inline void vring_used_idx_set(VirtQueue *vq, uint16_t val)
444 {
445     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
446     hwaddr pa = offsetof(VRingUsed, idx);
447 
448     if (caches) {
449         virtio_stw_phys_cached(vq->vdev, &caches->used, pa, val);
450         address_space_cache_invalidate(&caches->used, pa, sizeof(val));
451     }
452 
453     vq->used_idx = val;
454 }
455 
456 /* Called within rcu_read_lock().  */
457 static inline void vring_used_flags_set_bit(VirtQueue *vq, int mask)
458 {
459     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
460     VirtIODevice *vdev = vq->vdev;
461     hwaddr pa = offsetof(VRingUsed, flags);
462     uint16_t flags;
463 
464     if (!caches) {
465         return;
466     }
467 
468     flags = virtio_lduw_phys_cached(vq->vdev, &caches->used, pa);
469     virtio_stw_phys_cached(vdev, &caches->used, pa, flags | mask);
470     address_space_cache_invalidate(&caches->used, pa, sizeof(flags));
471 }
472 
473 /* Called within rcu_read_lock().  */
474 static inline void vring_used_flags_unset_bit(VirtQueue *vq, int mask)
475 {
476     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
477     VirtIODevice *vdev = vq->vdev;
478     hwaddr pa = offsetof(VRingUsed, flags);
479     uint16_t flags;
480 
481     if (!caches) {
482         return;
483     }
484 
485     flags = virtio_lduw_phys_cached(vq->vdev, &caches->used, pa);
486     virtio_stw_phys_cached(vdev, &caches->used, pa, flags & ~mask);
487     address_space_cache_invalidate(&caches->used, pa, sizeof(flags));
488 }
489 
490 /* Called within rcu_read_lock().  */
491 static inline void vring_set_avail_event(VirtQueue *vq, uint16_t val)
492 {
493     VRingMemoryRegionCaches *caches;
494     hwaddr pa;
495     if (!vq->notification) {
496         return;
497     }
498 
499     caches = vring_get_region_caches(vq);
500     if (!caches) {
501         return;
502     }
503 
504     pa = offsetof(VRingUsed, ring[vq->vring.num]);
505     virtio_stw_phys_cached(vq->vdev, &caches->used, pa, val);
506     address_space_cache_invalidate(&caches->used, pa, sizeof(val));
507 }
508 
509 static void virtio_queue_split_set_notification(VirtQueue *vq, int enable)
510 {
511     RCU_READ_LOCK_GUARD();
512 
513     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_RING_F_EVENT_IDX)) {
514         vring_set_avail_event(vq, vring_avail_idx(vq));
515     } else if (enable) {
516         vring_used_flags_unset_bit(vq, VRING_USED_F_NO_NOTIFY);
517     } else {
518         vring_used_flags_set_bit(vq, VRING_USED_F_NO_NOTIFY);
519     }
520     if (enable) {
521         /* Expose avail event/used flags before caller checks the avail idx. */
522         smp_mb();
523     }
524 }
525 
526 static void virtio_queue_packed_set_notification(VirtQueue *vq, int enable)
527 {
528     uint16_t off_wrap;
529     VRingPackedDescEvent e;
530     VRingMemoryRegionCaches *caches;
531 
532     RCU_READ_LOCK_GUARD();
533     caches = vring_get_region_caches(vq);
534     if (!caches) {
535         return;
536     }
537 
538     vring_packed_event_read(vq->vdev, &caches->used, &e);
539 
540     if (!enable) {
541         e.flags = VRING_PACKED_EVENT_FLAG_DISABLE;
542     } else if (virtio_vdev_has_feature(vq->vdev, VIRTIO_RING_F_EVENT_IDX)) {
543         off_wrap = vq->shadow_avail_idx | vq->shadow_avail_wrap_counter << 15;
544         vring_packed_off_wrap_write(vq->vdev, &caches->used, off_wrap);
545         /* Make sure off_wrap is wrote before flags */
546         smp_wmb();
547         e.flags = VRING_PACKED_EVENT_FLAG_DESC;
548     } else {
549         e.flags = VRING_PACKED_EVENT_FLAG_ENABLE;
550     }
551 
552     vring_packed_flags_write(vq->vdev, &caches->used, e.flags);
553     if (enable) {
554         /* Expose avail event/used flags before caller checks the avail idx. */
555         smp_mb();
556     }
557 }
558 
559 bool virtio_queue_get_notification(VirtQueue *vq)
560 {
561     return vq->notification;
562 }
563 
564 void virtio_queue_set_notification(VirtQueue *vq, int enable)
565 {
566     vq->notification = enable;
567 
568     if (!vq->vring.desc) {
569         return;
570     }
571 
572     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) {
573         virtio_queue_packed_set_notification(vq, enable);
574     } else {
575         virtio_queue_split_set_notification(vq, enable);
576     }
577 }
578 
579 int virtio_queue_ready(VirtQueue *vq)
580 {
581     return vq->vring.avail != 0;
582 }
583 
584 static void vring_packed_desc_read_flags(VirtIODevice *vdev,
585                                          uint16_t *flags,
586                                          MemoryRegionCache *cache,
587                                          int i)
588 {
589     hwaddr off = i * sizeof(VRingPackedDesc) + offsetof(VRingPackedDesc, flags);
590 
591     *flags = virtio_lduw_phys_cached(vdev, cache, off);
592 }
593 
594 static void vring_packed_desc_read(VirtIODevice *vdev,
595                                    VRingPackedDesc *desc,
596                                    MemoryRegionCache *cache,
597                                    int i, bool strict_order)
598 {
599     hwaddr off = i * sizeof(VRingPackedDesc);
600 
601     vring_packed_desc_read_flags(vdev, &desc->flags, cache, i);
602 
603     if (strict_order) {
604         /* Make sure flags is read before the rest fields. */
605         smp_rmb();
606     }
607 
608     address_space_read_cached(cache, off + offsetof(VRingPackedDesc, addr),
609                               &desc->addr, sizeof(desc->addr));
610     address_space_read_cached(cache, off + offsetof(VRingPackedDesc, id),
611                               &desc->id, sizeof(desc->id));
612     address_space_read_cached(cache, off + offsetof(VRingPackedDesc, len),
613                               &desc->len, sizeof(desc->len));
614     virtio_tswap64s(vdev, &desc->addr);
615     virtio_tswap16s(vdev, &desc->id);
616     virtio_tswap32s(vdev, &desc->len);
617 }
618 
619 static void vring_packed_desc_write_data(VirtIODevice *vdev,
620                                          VRingPackedDesc *desc,
621                                          MemoryRegionCache *cache,
622                                          int i)
623 {
624     hwaddr off_id = i * sizeof(VRingPackedDesc) +
625                     offsetof(VRingPackedDesc, id);
626     hwaddr off_len = i * sizeof(VRingPackedDesc) +
627                     offsetof(VRingPackedDesc, len);
628 
629     virtio_tswap32s(vdev, &desc->len);
630     virtio_tswap16s(vdev, &desc->id);
631     address_space_write_cached(cache, off_id, &desc->id, sizeof(desc->id));
632     address_space_cache_invalidate(cache, off_id, sizeof(desc->id));
633     address_space_write_cached(cache, off_len, &desc->len, sizeof(desc->len));
634     address_space_cache_invalidate(cache, off_len, sizeof(desc->len));
635 }
636 
637 static void vring_packed_desc_write_flags(VirtIODevice *vdev,
638                                           VRingPackedDesc *desc,
639                                           MemoryRegionCache *cache,
640                                           int i)
641 {
642     hwaddr off = i * sizeof(VRingPackedDesc) + offsetof(VRingPackedDesc, flags);
643 
644     virtio_stw_phys_cached(vdev, cache, off, desc->flags);
645     address_space_cache_invalidate(cache, off, sizeof(desc->flags));
646 }
647 
648 static void vring_packed_desc_write(VirtIODevice *vdev,
649                                     VRingPackedDesc *desc,
650                                     MemoryRegionCache *cache,
651                                     int i, bool strict_order)
652 {
653     vring_packed_desc_write_data(vdev, desc, cache, i);
654     if (strict_order) {
655         /* Make sure data is wrote before flags. */
656         smp_wmb();
657     }
658     vring_packed_desc_write_flags(vdev, desc, cache, i);
659 }
660 
661 static inline bool is_desc_avail(uint16_t flags, bool wrap_counter)
662 {
663     bool avail, used;
664 
665     avail = !!(flags & (1 << VRING_PACKED_DESC_F_AVAIL));
666     used = !!(flags & (1 << VRING_PACKED_DESC_F_USED));
667     return (avail != used) && (avail == wrap_counter);
668 }
669 
670 /* Fetch avail_idx from VQ memory only when we really need to know if
671  * guest has added some buffers.
672  * Called within rcu_read_lock().  */
673 static int virtio_queue_empty_rcu(VirtQueue *vq)
674 {
675     if (virtio_device_disabled(vq->vdev)) {
676         return 1;
677     }
678 
679     if (unlikely(!vq->vring.avail)) {
680         return 1;
681     }
682 
683     if (vq->shadow_avail_idx != vq->last_avail_idx) {
684         return 0;
685     }
686 
687     return vring_avail_idx(vq) == vq->last_avail_idx;
688 }
689 
690 static int virtio_queue_split_empty(VirtQueue *vq)
691 {
692     bool empty;
693 
694     if (virtio_device_disabled(vq->vdev)) {
695         return 1;
696     }
697 
698     if (unlikely(!vq->vring.avail)) {
699         return 1;
700     }
701 
702     if (vq->shadow_avail_idx != vq->last_avail_idx) {
703         return 0;
704     }
705 
706     RCU_READ_LOCK_GUARD();
707     empty = vring_avail_idx(vq) == vq->last_avail_idx;
708     return empty;
709 }
710 
711 /* Called within rcu_read_lock().  */
712 static int virtio_queue_packed_empty_rcu(VirtQueue *vq)
713 {
714     struct VRingPackedDesc desc;
715     VRingMemoryRegionCaches *cache;
716 
717     if (unlikely(!vq->vring.desc)) {
718         return 1;
719     }
720 
721     cache = vring_get_region_caches(vq);
722     if (!cache) {
723         return 1;
724     }
725 
726     vring_packed_desc_read_flags(vq->vdev, &desc.flags, &cache->desc,
727                                  vq->last_avail_idx);
728 
729     return !is_desc_avail(desc.flags, vq->last_avail_wrap_counter);
730 }
731 
732 static int virtio_queue_packed_empty(VirtQueue *vq)
733 {
734     RCU_READ_LOCK_GUARD();
735     return virtio_queue_packed_empty_rcu(vq);
736 }
737 
738 int virtio_queue_empty(VirtQueue *vq)
739 {
740     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) {
741         return virtio_queue_packed_empty(vq);
742     } else {
743         return virtio_queue_split_empty(vq);
744     }
745 }
746 
747 static void virtqueue_unmap_sg(VirtQueue *vq, const VirtQueueElement *elem,
748                                unsigned int len)
749 {
750     AddressSpace *dma_as = vq->vdev->dma_as;
751     unsigned int offset;
752     int i;
753 
754     offset = 0;
755     for (i = 0; i < elem->in_num; i++) {
756         size_t size = MIN(len - offset, elem->in_sg[i].iov_len);
757 
758         dma_memory_unmap(dma_as, elem->in_sg[i].iov_base,
759                          elem->in_sg[i].iov_len,
760                          DMA_DIRECTION_FROM_DEVICE, size);
761 
762         offset += size;
763     }
764 
765     for (i = 0; i < elem->out_num; i++)
766         dma_memory_unmap(dma_as, elem->out_sg[i].iov_base,
767                          elem->out_sg[i].iov_len,
768                          DMA_DIRECTION_TO_DEVICE,
769                          elem->out_sg[i].iov_len);
770 }
771 
772 /* virtqueue_detach_element:
773  * @vq: The #VirtQueue
774  * @elem: The #VirtQueueElement
775  * @len: number of bytes written
776  *
777  * Detach the element from the virtqueue.  This function is suitable for device
778  * reset or other situations where a #VirtQueueElement is simply freed and will
779  * not be pushed or discarded.
780  */
781 void virtqueue_detach_element(VirtQueue *vq, const VirtQueueElement *elem,
782                               unsigned int len)
783 {
784     vq->inuse -= elem->ndescs;
785     virtqueue_unmap_sg(vq, elem, len);
786 }
787 
788 static void virtqueue_split_rewind(VirtQueue *vq, unsigned int num)
789 {
790     vq->last_avail_idx -= num;
791 }
792 
793 static void virtqueue_packed_rewind(VirtQueue *vq, unsigned int num)
794 {
795     if (vq->last_avail_idx < num) {
796         vq->last_avail_idx = vq->vring.num + vq->last_avail_idx - num;
797         vq->last_avail_wrap_counter ^= 1;
798     } else {
799         vq->last_avail_idx -= num;
800     }
801 }
802 
803 /* virtqueue_unpop:
804  * @vq: The #VirtQueue
805  * @elem: The #VirtQueueElement
806  * @len: number of bytes written
807  *
808  * Pretend the most recent element wasn't popped from the virtqueue.  The next
809  * call to virtqueue_pop() will refetch the element.
810  */
811 void virtqueue_unpop(VirtQueue *vq, const VirtQueueElement *elem,
812                      unsigned int len)
813 {
814 
815     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) {
816         virtqueue_packed_rewind(vq, 1);
817     } else {
818         virtqueue_split_rewind(vq, 1);
819     }
820 
821     virtqueue_detach_element(vq, elem, len);
822 }
823 
824 /* virtqueue_rewind:
825  * @vq: The #VirtQueue
826  * @num: Number of elements to push back
827  *
828  * Pretend that elements weren't popped from the virtqueue.  The next
829  * virtqueue_pop() will refetch the oldest element.
830  *
831  * Use virtqueue_unpop() instead if you have a VirtQueueElement.
832  *
833  * Returns: true on success, false if @num is greater than the number of in use
834  * elements.
835  */
836 bool virtqueue_rewind(VirtQueue *vq, unsigned int num)
837 {
838     if (num > vq->inuse) {
839         return false;
840     }
841 
842     vq->inuse -= num;
843     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) {
844         virtqueue_packed_rewind(vq, num);
845     } else {
846         virtqueue_split_rewind(vq, num);
847     }
848     return true;
849 }
850 
851 static void virtqueue_split_fill(VirtQueue *vq, const VirtQueueElement *elem,
852                     unsigned int len, unsigned int idx)
853 {
854     VRingUsedElem uelem;
855 
856     if (unlikely(!vq->vring.used)) {
857         return;
858     }
859 
860     idx = (idx + vq->used_idx) % vq->vring.num;
861 
862     uelem.id = elem->index;
863     uelem.len = len;
864     vring_used_write(vq, &uelem, idx);
865 }
866 
867 static void virtqueue_packed_fill(VirtQueue *vq, const VirtQueueElement *elem,
868                                   unsigned int len, unsigned int idx)
869 {
870     vq->used_elems[idx].index = elem->index;
871     vq->used_elems[idx].len = len;
872     vq->used_elems[idx].ndescs = elem->ndescs;
873 }
874 
875 static void virtqueue_packed_fill_desc(VirtQueue *vq,
876                                        const VirtQueueElement *elem,
877                                        unsigned int idx,
878                                        bool strict_order)
879 {
880     uint16_t head;
881     VRingMemoryRegionCaches *caches;
882     VRingPackedDesc desc = {
883         .id = elem->index,
884         .len = elem->len,
885     };
886     bool wrap_counter = vq->used_wrap_counter;
887 
888     if (unlikely(!vq->vring.desc)) {
889         return;
890     }
891 
892     head = vq->used_idx + idx;
893     if (head >= vq->vring.num) {
894         head -= vq->vring.num;
895         wrap_counter ^= 1;
896     }
897     if (wrap_counter) {
898         desc.flags |= (1 << VRING_PACKED_DESC_F_AVAIL);
899         desc.flags |= (1 << VRING_PACKED_DESC_F_USED);
900     } else {
901         desc.flags &= ~(1 << VRING_PACKED_DESC_F_AVAIL);
902         desc.flags &= ~(1 << VRING_PACKED_DESC_F_USED);
903     }
904 
905     caches = vring_get_region_caches(vq);
906     if (!caches) {
907         return;
908     }
909 
910     vring_packed_desc_write(vq->vdev, &desc, &caches->desc, head, strict_order);
911 }
912 
913 /* Called within rcu_read_lock().  */
914 void virtqueue_fill(VirtQueue *vq, const VirtQueueElement *elem,
915                     unsigned int len, unsigned int idx)
916 {
917     trace_virtqueue_fill(vq, elem, len, idx);
918 
919     virtqueue_unmap_sg(vq, elem, len);
920 
921     if (virtio_device_disabled(vq->vdev)) {
922         return;
923     }
924 
925     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) {
926         virtqueue_packed_fill(vq, elem, len, idx);
927     } else {
928         virtqueue_split_fill(vq, elem, len, idx);
929     }
930 }
931 
932 /* Called within rcu_read_lock().  */
933 static void virtqueue_split_flush(VirtQueue *vq, unsigned int count)
934 {
935     uint16_t old, new;
936 
937     if (unlikely(!vq->vring.used)) {
938         return;
939     }
940 
941     /* Make sure buffer is written before we update index. */
942     smp_wmb();
943     trace_virtqueue_flush(vq, count);
944     old = vq->used_idx;
945     new = old + count;
946     vring_used_idx_set(vq, new);
947     vq->inuse -= count;
948     if (unlikely((int16_t)(new - vq->signalled_used) < (uint16_t)(new - old)))
949         vq->signalled_used_valid = false;
950 }
951 
952 static void virtqueue_packed_flush(VirtQueue *vq, unsigned int count)
953 {
954     unsigned int i, ndescs = 0;
955 
956     if (unlikely(!vq->vring.desc)) {
957         return;
958     }
959 
960     /*
961      * For indirect element's 'ndescs' is 1.
962      * For all other elemment's 'ndescs' is the
963      * number of descriptors chained by NEXT (as set in virtqueue_packed_pop).
964      * So When the 'elem' be filled into the descriptor ring,
965      * The 'idx' of this 'elem' shall be
966      * the value of 'vq->used_idx' plus the 'ndescs'.
967      */
968     ndescs += vq->used_elems[0].ndescs;
969     for (i = 1; i < count; i++) {
970         virtqueue_packed_fill_desc(vq, &vq->used_elems[i], ndescs, false);
971         ndescs += vq->used_elems[i].ndescs;
972     }
973     virtqueue_packed_fill_desc(vq, &vq->used_elems[0], 0, true);
974 
975     vq->inuse -= ndescs;
976     vq->used_idx += ndescs;
977     if (vq->used_idx >= vq->vring.num) {
978         vq->used_idx -= vq->vring.num;
979         vq->used_wrap_counter ^= 1;
980         vq->signalled_used_valid = false;
981     }
982 }
983 
984 void virtqueue_flush(VirtQueue *vq, unsigned int count)
985 {
986     if (virtio_device_disabled(vq->vdev)) {
987         vq->inuse -= count;
988         return;
989     }
990 
991     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) {
992         virtqueue_packed_flush(vq, count);
993     } else {
994         virtqueue_split_flush(vq, count);
995     }
996 }
997 
998 void virtqueue_push(VirtQueue *vq, const VirtQueueElement *elem,
999                     unsigned int len)
1000 {
1001     RCU_READ_LOCK_GUARD();
1002     virtqueue_fill(vq, elem, len, 0);
1003     virtqueue_flush(vq, 1);
1004 }
1005 
1006 /* Called within rcu_read_lock().  */
1007 static int virtqueue_num_heads(VirtQueue *vq, unsigned int idx)
1008 {
1009     uint16_t avail_idx, num_heads;
1010 
1011     /* Use shadow index whenever possible. */
1012     avail_idx = (vq->shadow_avail_idx != idx) ? vq->shadow_avail_idx
1013                                               : vring_avail_idx(vq);
1014     num_heads = avail_idx - idx;
1015 
1016     /* Check it isn't doing very strange things with descriptor numbers. */
1017     if (num_heads > vq->vring.num) {
1018         virtio_error(vq->vdev, "Guest moved used index from %u to %u",
1019                      idx, vq->shadow_avail_idx);
1020         return -EINVAL;
1021     }
1022     /*
1023      * On success, callers read a descriptor at vq->last_avail_idx.
1024      * Make sure descriptor read does not bypass avail index read.
1025      *
1026      * This is necessary even if we are using a shadow index, since
1027      * the shadow index could have been initialized by calling
1028      * vring_avail_idx() outside of this function, i.e., by a guest
1029      * memory read not accompanied by a barrier.
1030      */
1031     if (num_heads) {
1032         smp_rmb();
1033     }
1034 
1035     return num_heads;
1036 }
1037 
1038 /* Called within rcu_read_lock().  */
1039 static bool virtqueue_get_head(VirtQueue *vq, unsigned int idx,
1040                                unsigned int *head)
1041 {
1042     /* Grab the next descriptor number they're advertising, and increment
1043      * the index we've seen. */
1044     *head = vring_avail_ring(vq, idx % vq->vring.num);
1045 
1046     /* If their number is silly, that's a fatal mistake. */
1047     if (*head >= vq->vring.num) {
1048         virtio_error(vq->vdev, "Guest says index %u is available", *head);
1049         return false;
1050     }
1051 
1052     return true;
1053 }
1054 
1055 enum {
1056     VIRTQUEUE_READ_DESC_ERROR = -1,
1057     VIRTQUEUE_READ_DESC_DONE = 0,   /* end of chain */
1058     VIRTQUEUE_READ_DESC_MORE = 1,   /* more buffers in chain */
1059 };
1060 
1061 /* Reads the 'desc->next' descriptor into '*desc'. */
1062 static int virtqueue_split_read_next_desc(VirtIODevice *vdev, VRingDesc *desc,
1063                                           MemoryRegionCache *desc_cache,
1064                                           unsigned int max)
1065 {
1066     /* If this descriptor says it doesn't chain, we're done. */
1067     if (!(desc->flags & VRING_DESC_F_NEXT)) {
1068         return VIRTQUEUE_READ_DESC_DONE;
1069     }
1070 
1071     /* Check they're not leading us off end of descriptors. */
1072     if (desc->next >= max) {
1073         virtio_error(vdev, "Desc next is %u", desc->next);
1074         return VIRTQUEUE_READ_DESC_ERROR;
1075     }
1076 
1077     vring_split_desc_read(vdev, desc, desc_cache, desc->next);
1078     return VIRTQUEUE_READ_DESC_MORE;
1079 }
1080 
1081 /* Called within rcu_read_lock().  */
1082 static void virtqueue_split_get_avail_bytes(VirtQueue *vq,
1083                             unsigned int *in_bytes, unsigned int *out_bytes,
1084                             unsigned max_in_bytes, unsigned max_out_bytes,
1085                             VRingMemoryRegionCaches *caches)
1086 {
1087     VirtIODevice *vdev = vq->vdev;
1088     unsigned int idx;
1089     unsigned int total_bufs, in_total, out_total;
1090     MemoryRegionCache indirect_desc_cache;
1091     int64_t len = 0;
1092     int rc;
1093 
1094     address_space_cache_init_empty(&indirect_desc_cache);
1095 
1096     idx = vq->last_avail_idx;
1097     total_bufs = in_total = out_total = 0;
1098 
1099     while ((rc = virtqueue_num_heads(vq, idx)) > 0) {
1100         MemoryRegionCache *desc_cache = &caches->desc;
1101         unsigned int num_bufs;
1102         VRingDesc desc;
1103         unsigned int i;
1104         unsigned int max = vq->vring.num;
1105 
1106         num_bufs = total_bufs;
1107 
1108         if (!virtqueue_get_head(vq, idx++, &i)) {
1109             goto err;
1110         }
1111 
1112         vring_split_desc_read(vdev, &desc, desc_cache, i);
1113 
1114         if (desc.flags & VRING_DESC_F_INDIRECT) {
1115             if (!desc.len || (desc.len % sizeof(VRingDesc))) {
1116                 virtio_error(vdev, "Invalid size for indirect buffer table");
1117                 goto err;
1118             }
1119 
1120             /* If we've got too many, that implies a descriptor loop. */
1121             if (num_bufs >= max) {
1122                 virtio_error(vdev, "Looped descriptor");
1123                 goto err;
1124             }
1125 
1126             /* loop over the indirect descriptor table */
1127             len = address_space_cache_init(&indirect_desc_cache,
1128                                            vdev->dma_as,
1129                                            desc.addr, desc.len, false);
1130             desc_cache = &indirect_desc_cache;
1131             if (len < desc.len) {
1132                 virtio_error(vdev, "Cannot map indirect buffer");
1133                 goto err;
1134             }
1135 
1136             max = desc.len / sizeof(VRingDesc);
1137             num_bufs = i = 0;
1138             vring_split_desc_read(vdev, &desc, desc_cache, i);
1139         }
1140 
1141         do {
1142             /* If we've got too many, that implies a descriptor loop. */
1143             if (++num_bufs > max) {
1144                 virtio_error(vdev, "Looped descriptor");
1145                 goto err;
1146             }
1147 
1148             if (desc.flags & VRING_DESC_F_WRITE) {
1149                 in_total += desc.len;
1150             } else {
1151                 out_total += desc.len;
1152             }
1153             if (in_total >= max_in_bytes && out_total >= max_out_bytes) {
1154                 goto done;
1155             }
1156 
1157             rc = virtqueue_split_read_next_desc(vdev, &desc, desc_cache, max);
1158         } while (rc == VIRTQUEUE_READ_DESC_MORE);
1159 
1160         if (rc == VIRTQUEUE_READ_DESC_ERROR) {
1161             goto err;
1162         }
1163 
1164         if (desc_cache == &indirect_desc_cache) {
1165             address_space_cache_destroy(&indirect_desc_cache);
1166             total_bufs++;
1167         } else {
1168             total_bufs = num_bufs;
1169         }
1170     }
1171 
1172     if (rc < 0) {
1173         goto err;
1174     }
1175 
1176 done:
1177     address_space_cache_destroy(&indirect_desc_cache);
1178     if (in_bytes) {
1179         *in_bytes = in_total;
1180     }
1181     if (out_bytes) {
1182         *out_bytes = out_total;
1183     }
1184     return;
1185 
1186 err:
1187     in_total = out_total = 0;
1188     goto done;
1189 }
1190 
1191 static int virtqueue_packed_read_next_desc(VirtQueue *vq,
1192                                            VRingPackedDesc *desc,
1193                                            MemoryRegionCache
1194                                            *desc_cache,
1195                                            unsigned int max,
1196                                            unsigned int *next,
1197                                            bool indirect)
1198 {
1199     /* If this descriptor says it doesn't chain, we're done. */
1200     if (!indirect && !(desc->flags & VRING_DESC_F_NEXT)) {
1201         return VIRTQUEUE_READ_DESC_DONE;
1202     }
1203 
1204     ++*next;
1205     if (*next == max) {
1206         if (indirect) {
1207             return VIRTQUEUE_READ_DESC_DONE;
1208         } else {
1209             (*next) -= vq->vring.num;
1210         }
1211     }
1212 
1213     vring_packed_desc_read(vq->vdev, desc, desc_cache, *next, false);
1214     return VIRTQUEUE_READ_DESC_MORE;
1215 }
1216 
1217 /* Called within rcu_read_lock().  */
1218 static void virtqueue_packed_get_avail_bytes(VirtQueue *vq,
1219                                              unsigned int *in_bytes,
1220                                              unsigned int *out_bytes,
1221                                              unsigned max_in_bytes,
1222                                              unsigned max_out_bytes,
1223                                              VRingMemoryRegionCaches *caches)
1224 {
1225     VirtIODevice *vdev = vq->vdev;
1226     unsigned int idx;
1227     unsigned int total_bufs, in_total, out_total;
1228     MemoryRegionCache indirect_desc_cache;
1229     MemoryRegionCache *desc_cache;
1230     int64_t len = 0;
1231     VRingPackedDesc desc;
1232     bool wrap_counter;
1233 
1234     address_space_cache_init_empty(&indirect_desc_cache);
1235 
1236     idx = vq->last_avail_idx;
1237     wrap_counter = vq->last_avail_wrap_counter;
1238     total_bufs = in_total = out_total = 0;
1239 
1240     for (;;) {
1241         unsigned int num_bufs = total_bufs;
1242         unsigned int i = idx;
1243         int rc;
1244         unsigned int max = vq->vring.num;
1245 
1246         desc_cache = &caches->desc;
1247 
1248         vring_packed_desc_read(vdev, &desc, desc_cache, idx, true);
1249         if (!is_desc_avail(desc.flags, wrap_counter)) {
1250             break;
1251         }
1252 
1253         if (desc.flags & VRING_DESC_F_INDIRECT) {
1254             if (desc.len % sizeof(VRingPackedDesc)) {
1255                 virtio_error(vdev, "Invalid size for indirect buffer table");
1256                 goto err;
1257             }
1258 
1259             /* If we've got too many, that implies a descriptor loop. */
1260             if (num_bufs >= max) {
1261                 virtio_error(vdev, "Looped descriptor");
1262                 goto err;
1263             }
1264 
1265             /* loop over the indirect descriptor table */
1266             len = address_space_cache_init(&indirect_desc_cache,
1267                                            vdev->dma_as,
1268                                            desc.addr, desc.len, false);
1269             desc_cache = &indirect_desc_cache;
1270             if (len < desc.len) {
1271                 virtio_error(vdev, "Cannot map indirect buffer");
1272                 goto err;
1273             }
1274 
1275             max = desc.len / sizeof(VRingPackedDesc);
1276             num_bufs = i = 0;
1277             vring_packed_desc_read(vdev, &desc, desc_cache, i, false);
1278         }
1279 
1280         do {
1281             /* If we've got too many, that implies a descriptor loop. */
1282             if (++num_bufs > max) {
1283                 virtio_error(vdev, "Looped descriptor");
1284                 goto err;
1285             }
1286 
1287             if (desc.flags & VRING_DESC_F_WRITE) {
1288                 in_total += desc.len;
1289             } else {
1290                 out_total += desc.len;
1291             }
1292             if (in_total >= max_in_bytes && out_total >= max_out_bytes) {
1293                 goto done;
1294             }
1295 
1296             rc = virtqueue_packed_read_next_desc(vq, &desc, desc_cache, max,
1297                                                  &i, desc_cache ==
1298                                                  &indirect_desc_cache);
1299         } while (rc == VIRTQUEUE_READ_DESC_MORE);
1300 
1301         if (desc_cache == &indirect_desc_cache) {
1302             address_space_cache_destroy(&indirect_desc_cache);
1303             total_bufs++;
1304             idx++;
1305         } else {
1306             idx += num_bufs - total_bufs;
1307             total_bufs = num_bufs;
1308         }
1309 
1310         if (idx >= vq->vring.num) {
1311             idx -= vq->vring.num;
1312             wrap_counter ^= 1;
1313         }
1314     }
1315 
1316     /* Record the index and wrap counter for a kick we want */
1317     vq->shadow_avail_idx = idx;
1318     vq->shadow_avail_wrap_counter = wrap_counter;
1319 done:
1320     address_space_cache_destroy(&indirect_desc_cache);
1321     if (in_bytes) {
1322         *in_bytes = in_total;
1323     }
1324     if (out_bytes) {
1325         *out_bytes = out_total;
1326     }
1327     return;
1328 
1329 err:
1330     in_total = out_total = 0;
1331     goto done;
1332 }
1333 
1334 void virtqueue_get_avail_bytes(VirtQueue *vq, unsigned int *in_bytes,
1335                                unsigned int *out_bytes,
1336                                unsigned max_in_bytes, unsigned max_out_bytes)
1337 {
1338     uint16_t desc_size;
1339     VRingMemoryRegionCaches *caches;
1340 
1341     RCU_READ_LOCK_GUARD();
1342 
1343     if (unlikely(!vq->vring.desc)) {
1344         goto err;
1345     }
1346 
1347     caches = vring_get_region_caches(vq);
1348     if (!caches) {
1349         goto err;
1350     }
1351 
1352     desc_size = virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED) ?
1353                                 sizeof(VRingPackedDesc) : sizeof(VRingDesc);
1354     if (caches->desc.len < vq->vring.num * desc_size) {
1355         virtio_error(vq->vdev, "Cannot map descriptor ring");
1356         goto err;
1357     }
1358 
1359     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) {
1360         virtqueue_packed_get_avail_bytes(vq, in_bytes, out_bytes,
1361                                          max_in_bytes, max_out_bytes,
1362                                          caches);
1363     } else {
1364         virtqueue_split_get_avail_bytes(vq, in_bytes, out_bytes,
1365                                         max_in_bytes, max_out_bytes,
1366                                         caches);
1367     }
1368 
1369     return;
1370 err:
1371     if (in_bytes) {
1372         *in_bytes = 0;
1373     }
1374     if (out_bytes) {
1375         *out_bytes = 0;
1376     }
1377 }
1378 
1379 int virtqueue_avail_bytes(VirtQueue *vq, unsigned int in_bytes,
1380                           unsigned int out_bytes)
1381 {
1382     unsigned int in_total, out_total;
1383 
1384     virtqueue_get_avail_bytes(vq, &in_total, &out_total, in_bytes, out_bytes);
1385     return in_bytes <= in_total && out_bytes <= out_total;
1386 }
1387 
1388 static bool virtqueue_map_desc(VirtIODevice *vdev, unsigned int *p_num_sg,
1389                                hwaddr *addr, struct iovec *iov,
1390                                unsigned int max_num_sg, bool is_write,
1391                                hwaddr pa, size_t sz)
1392 {
1393     bool ok = false;
1394     unsigned num_sg = *p_num_sg;
1395     assert(num_sg <= max_num_sg);
1396 
1397     if (!sz) {
1398         virtio_error(vdev, "virtio: zero sized buffers are not allowed");
1399         goto out;
1400     }
1401 
1402     while (sz) {
1403         hwaddr len = sz;
1404 
1405         if (num_sg == max_num_sg) {
1406             virtio_error(vdev, "virtio: too many write descriptors in "
1407                                "indirect table");
1408             goto out;
1409         }
1410 
1411         iov[num_sg].iov_base = dma_memory_map(vdev->dma_as, pa, &len,
1412                                               is_write ?
1413                                               DMA_DIRECTION_FROM_DEVICE :
1414                                               DMA_DIRECTION_TO_DEVICE,
1415                                               MEMTXATTRS_UNSPECIFIED);
1416         if (!iov[num_sg].iov_base) {
1417             virtio_error(vdev, "virtio: bogus descriptor or out of resources");
1418             goto out;
1419         }
1420 
1421         iov[num_sg].iov_len = len;
1422         addr[num_sg] = pa;
1423 
1424         sz -= len;
1425         pa += len;
1426         num_sg++;
1427     }
1428     ok = true;
1429 
1430 out:
1431     *p_num_sg = num_sg;
1432     return ok;
1433 }
1434 
1435 /* Only used by error code paths before we have a VirtQueueElement (therefore
1436  * virtqueue_unmap_sg() can't be used).  Assumes buffers weren't written to
1437  * yet.
1438  */
1439 static void virtqueue_undo_map_desc(unsigned int out_num, unsigned int in_num,
1440                                     struct iovec *iov)
1441 {
1442     unsigned int i;
1443 
1444     for (i = 0; i < out_num + in_num; i++) {
1445         int is_write = i >= out_num;
1446 
1447         cpu_physical_memory_unmap(iov->iov_base, iov->iov_len, is_write, 0);
1448         iov++;
1449     }
1450 }
1451 
1452 static void virtqueue_map_iovec(VirtIODevice *vdev, struct iovec *sg,
1453                                 hwaddr *addr, unsigned int num_sg,
1454                                 bool is_write)
1455 {
1456     unsigned int i;
1457     hwaddr len;
1458 
1459     for (i = 0; i < num_sg; i++) {
1460         len = sg[i].iov_len;
1461         sg[i].iov_base = dma_memory_map(vdev->dma_as,
1462                                         addr[i], &len, is_write ?
1463                                         DMA_DIRECTION_FROM_DEVICE :
1464                                         DMA_DIRECTION_TO_DEVICE,
1465                                         MEMTXATTRS_UNSPECIFIED);
1466         if (!sg[i].iov_base) {
1467             error_report("virtio: error trying to map MMIO memory");
1468             exit(1);
1469         }
1470         if (len != sg[i].iov_len) {
1471             error_report("virtio: unexpected memory split");
1472             exit(1);
1473         }
1474     }
1475 }
1476 
1477 void virtqueue_map(VirtIODevice *vdev, VirtQueueElement *elem)
1478 {
1479     virtqueue_map_iovec(vdev, elem->in_sg, elem->in_addr, elem->in_num, true);
1480     virtqueue_map_iovec(vdev, elem->out_sg, elem->out_addr, elem->out_num,
1481                                                                         false);
1482 }
1483 
1484 static void *virtqueue_alloc_element(size_t sz, unsigned out_num, unsigned in_num)
1485 {
1486     VirtQueueElement *elem;
1487     size_t in_addr_ofs = QEMU_ALIGN_UP(sz, __alignof__(elem->in_addr[0]));
1488     size_t out_addr_ofs = in_addr_ofs + in_num * sizeof(elem->in_addr[0]);
1489     size_t out_addr_end = out_addr_ofs + out_num * sizeof(elem->out_addr[0]);
1490     size_t in_sg_ofs = QEMU_ALIGN_UP(out_addr_end, __alignof__(elem->in_sg[0]));
1491     size_t out_sg_ofs = in_sg_ofs + in_num * sizeof(elem->in_sg[0]);
1492     size_t out_sg_end = out_sg_ofs + out_num * sizeof(elem->out_sg[0]);
1493 
1494     assert(sz >= sizeof(VirtQueueElement));
1495     elem = g_malloc(out_sg_end);
1496     trace_virtqueue_alloc_element(elem, sz, in_num, out_num);
1497     elem->out_num = out_num;
1498     elem->in_num = in_num;
1499     elem->in_addr = (void *)elem + in_addr_ofs;
1500     elem->out_addr = (void *)elem + out_addr_ofs;
1501     elem->in_sg = (void *)elem + in_sg_ofs;
1502     elem->out_sg = (void *)elem + out_sg_ofs;
1503     return elem;
1504 }
1505 
1506 static void *virtqueue_split_pop(VirtQueue *vq, size_t sz)
1507 {
1508     unsigned int i, head, max;
1509     VRingMemoryRegionCaches *caches;
1510     MemoryRegionCache indirect_desc_cache;
1511     MemoryRegionCache *desc_cache;
1512     int64_t len;
1513     VirtIODevice *vdev = vq->vdev;
1514     VirtQueueElement *elem = NULL;
1515     unsigned out_num, in_num, elem_entries;
1516     hwaddr addr[VIRTQUEUE_MAX_SIZE];
1517     struct iovec iov[VIRTQUEUE_MAX_SIZE];
1518     VRingDesc desc;
1519     int rc;
1520 
1521     address_space_cache_init_empty(&indirect_desc_cache);
1522 
1523     RCU_READ_LOCK_GUARD();
1524     if (virtio_queue_empty_rcu(vq)) {
1525         goto done;
1526     }
1527     /* Needed after virtio_queue_empty(), see comment in
1528      * virtqueue_num_heads(). */
1529     smp_rmb();
1530 
1531     /* When we start there are none of either input nor output. */
1532     out_num = in_num = elem_entries = 0;
1533 
1534     max = vq->vring.num;
1535 
1536     if (vq->inuse >= vq->vring.num) {
1537         virtio_error(vdev, "Virtqueue size exceeded");
1538         goto done;
1539     }
1540 
1541     if (!virtqueue_get_head(vq, vq->last_avail_idx++, &head)) {
1542         goto done;
1543     }
1544 
1545     if (virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX)) {
1546         vring_set_avail_event(vq, vq->last_avail_idx);
1547     }
1548 
1549     i = head;
1550 
1551     caches = vring_get_region_caches(vq);
1552     if (!caches) {
1553         virtio_error(vdev, "Region caches not initialized");
1554         goto done;
1555     }
1556 
1557     if (caches->desc.len < max * sizeof(VRingDesc)) {
1558         virtio_error(vdev, "Cannot map descriptor ring");
1559         goto done;
1560     }
1561 
1562     desc_cache = &caches->desc;
1563     vring_split_desc_read(vdev, &desc, desc_cache, i);
1564     if (desc.flags & VRING_DESC_F_INDIRECT) {
1565         if (!desc.len || (desc.len % sizeof(VRingDesc))) {
1566             virtio_error(vdev, "Invalid size for indirect buffer table");
1567             goto done;
1568         }
1569 
1570         /* loop over the indirect descriptor table */
1571         len = address_space_cache_init(&indirect_desc_cache, vdev->dma_as,
1572                                        desc.addr, desc.len, false);
1573         desc_cache = &indirect_desc_cache;
1574         if (len < desc.len) {
1575             virtio_error(vdev, "Cannot map indirect buffer");
1576             goto done;
1577         }
1578 
1579         max = desc.len / sizeof(VRingDesc);
1580         i = 0;
1581         vring_split_desc_read(vdev, &desc, desc_cache, i);
1582     }
1583 
1584     /* Collect all the descriptors */
1585     do {
1586         bool map_ok;
1587 
1588         if (desc.flags & VRING_DESC_F_WRITE) {
1589             map_ok = virtqueue_map_desc(vdev, &in_num, addr + out_num,
1590                                         iov + out_num,
1591                                         VIRTQUEUE_MAX_SIZE - out_num, true,
1592                                         desc.addr, desc.len);
1593         } else {
1594             if (in_num) {
1595                 virtio_error(vdev, "Incorrect order for descriptors");
1596                 goto err_undo_map;
1597             }
1598             map_ok = virtqueue_map_desc(vdev, &out_num, addr, iov,
1599                                         VIRTQUEUE_MAX_SIZE, false,
1600                                         desc.addr, desc.len);
1601         }
1602         if (!map_ok) {
1603             goto err_undo_map;
1604         }
1605 
1606         /* If we've got too many, that implies a descriptor loop. */
1607         if (++elem_entries > max) {
1608             virtio_error(vdev, "Looped descriptor");
1609             goto err_undo_map;
1610         }
1611 
1612         rc = virtqueue_split_read_next_desc(vdev, &desc, desc_cache, max);
1613     } while (rc == VIRTQUEUE_READ_DESC_MORE);
1614 
1615     if (rc == VIRTQUEUE_READ_DESC_ERROR) {
1616         goto err_undo_map;
1617     }
1618 
1619     /* Now copy what we have collected and mapped */
1620     elem = virtqueue_alloc_element(sz, out_num, in_num);
1621     elem->index = head;
1622     elem->ndescs = 1;
1623     for (i = 0; i < out_num; i++) {
1624         elem->out_addr[i] = addr[i];
1625         elem->out_sg[i] = iov[i];
1626     }
1627     for (i = 0; i < in_num; i++) {
1628         elem->in_addr[i] = addr[out_num + i];
1629         elem->in_sg[i] = iov[out_num + i];
1630     }
1631 
1632     vq->inuse++;
1633 
1634     trace_virtqueue_pop(vq, elem, elem->in_num, elem->out_num);
1635 done:
1636     address_space_cache_destroy(&indirect_desc_cache);
1637 
1638     return elem;
1639 
1640 err_undo_map:
1641     virtqueue_undo_map_desc(out_num, in_num, iov);
1642     goto done;
1643 }
1644 
1645 static void *virtqueue_packed_pop(VirtQueue *vq, size_t sz)
1646 {
1647     unsigned int i, max;
1648     VRingMemoryRegionCaches *caches;
1649     MemoryRegionCache indirect_desc_cache;
1650     MemoryRegionCache *desc_cache;
1651     int64_t len;
1652     VirtIODevice *vdev = vq->vdev;
1653     VirtQueueElement *elem = NULL;
1654     unsigned out_num, in_num, elem_entries;
1655     hwaddr addr[VIRTQUEUE_MAX_SIZE];
1656     struct iovec iov[VIRTQUEUE_MAX_SIZE];
1657     VRingPackedDesc desc;
1658     uint16_t id;
1659     int rc;
1660 
1661     address_space_cache_init_empty(&indirect_desc_cache);
1662 
1663     RCU_READ_LOCK_GUARD();
1664     if (virtio_queue_packed_empty_rcu(vq)) {
1665         goto done;
1666     }
1667 
1668     /* When we start there are none of either input nor output. */
1669     out_num = in_num = elem_entries = 0;
1670 
1671     max = vq->vring.num;
1672 
1673     if (vq->inuse >= vq->vring.num) {
1674         virtio_error(vdev, "Virtqueue size exceeded");
1675         goto done;
1676     }
1677 
1678     i = vq->last_avail_idx;
1679 
1680     caches = vring_get_region_caches(vq);
1681     if (!caches) {
1682         virtio_error(vdev, "Region caches not initialized");
1683         goto done;
1684     }
1685 
1686     if (caches->desc.len < max * sizeof(VRingDesc)) {
1687         virtio_error(vdev, "Cannot map descriptor ring");
1688         goto done;
1689     }
1690 
1691     desc_cache = &caches->desc;
1692     vring_packed_desc_read(vdev, &desc, desc_cache, i, true);
1693     id = desc.id;
1694     if (desc.flags & VRING_DESC_F_INDIRECT) {
1695         if (desc.len % sizeof(VRingPackedDesc)) {
1696             virtio_error(vdev, "Invalid size for indirect buffer table");
1697             goto done;
1698         }
1699 
1700         /* loop over the indirect descriptor table */
1701         len = address_space_cache_init(&indirect_desc_cache, vdev->dma_as,
1702                                        desc.addr, desc.len, false);
1703         desc_cache = &indirect_desc_cache;
1704         if (len < desc.len) {
1705             virtio_error(vdev, "Cannot map indirect buffer");
1706             goto done;
1707         }
1708 
1709         max = desc.len / sizeof(VRingPackedDesc);
1710         i = 0;
1711         vring_packed_desc_read(vdev, &desc, desc_cache, i, false);
1712     }
1713 
1714     /* Collect all the descriptors */
1715     do {
1716         bool map_ok;
1717 
1718         if (desc.flags & VRING_DESC_F_WRITE) {
1719             map_ok = virtqueue_map_desc(vdev, &in_num, addr + out_num,
1720                                         iov + out_num,
1721                                         VIRTQUEUE_MAX_SIZE - out_num, true,
1722                                         desc.addr, desc.len);
1723         } else {
1724             if (in_num) {
1725                 virtio_error(vdev, "Incorrect order for descriptors");
1726                 goto err_undo_map;
1727             }
1728             map_ok = virtqueue_map_desc(vdev, &out_num, addr, iov,
1729                                         VIRTQUEUE_MAX_SIZE, false,
1730                                         desc.addr, desc.len);
1731         }
1732         if (!map_ok) {
1733             goto err_undo_map;
1734         }
1735 
1736         /* If we've got too many, that implies a descriptor loop. */
1737         if (++elem_entries > max) {
1738             virtio_error(vdev, "Looped descriptor");
1739             goto err_undo_map;
1740         }
1741 
1742         rc = virtqueue_packed_read_next_desc(vq, &desc, desc_cache, max, &i,
1743                                              desc_cache ==
1744                                              &indirect_desc_cache);
1745     } while (rc == VIRTQUEUE_READ_DESC_MORE);
1746 
1747     /* Now copy what we have collected and mapped */
1748     elem = virtqueue_alloc_element(sz, out_num, in_num);
1749     for (i = 0; i < out_num; i++) {
1750         elem->out_addr[i] = addr[i];
1751         elem->out_sg[i] = iov[i];
1752     }
1753     for (i = 0; i < in_num; i++) {
1754         elem->in_addr[i] = addr[out_num + i];
1755         elem->in_sg[i] = iov[out_num + i];
1756     }
1757 
1758     elem->index = id;
1759     elem->ndescs = (desc_cache == &indirect_desc_cache) ? 1 : elem_entries;
1760     vq->last_avail_idx += elem->ndescs;
1761     vq->inuse += elem->ndescs;
1762 
1763     if (vq->last_avail_idx >= vq->vring.num) {
1764         vq->last_avail_idx -= vq->vring.num;
1765         vq->last_avail_wrap_counter ^= 1;
1766     }
1767 
1768     vq->shadow_avail_idx = vq->last_avail_idx;
1769     vq->shadow_avail_wrap_counter = vq->last_avail_wrap_counter;
1770 
1771     trace_virtqueue_pop(vq, elem, elem->in_num, elem->out_num);
1772 done:
1773     address_space_cache_destroy(&indirect_desc_cache);
1774 
1775     return elem;
1776 
1777 err_undo_map:
1778     virtqueue_undo_map_desc(out_num, in_num, iov);
1779     goto done;
1780 }
1781 
1782 void *virtqueue_pop(VirtQueue *vq, size_t sz)
1783 {
1784     if (virtio_device_disabled(vq->vdev)) {
1785         return NULL;
1786     }
1787 
1788     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) {
1789         return virtqueue_packed_pop(vq, sz);
1790     } else {
1791         return virtqueue_split_pop(vq, sz);
1792     }
1793 }
1794 
1795 static unsigned int virtqueue_packed_drop_all(VirtQueue *vq)
1796 {
1797     VRingMemoryRegionCaches *caches;
1798     MemoryRegionCache *desc_cache;
1799     unsigned int dropped = 0;
1800     VirtQueueElement elem = {};
1801     VirtIODevice *vdev = vq->vdev;
1802     VRingPackedDesc desc;
1803 
1804     RCU_READ_LOCK_GUARD();
1805 
1806     caches = vring_get_region_caches(vq);
1807     if (!caches) {
1808         return 0;
1809     }
1810 
1811     desc_cache = &caches->desc;
1812 
1813     virtio_queue_set_notification(vq, 0);
1814 
1815     while (vq->inuse < vq->vring.num) {
1816         unsigned int idx = vq->last_avail_idx;
1817         /*
1818          * works similar to virtqueue_pop but does not map buffers
1819          * and does not allocate any memory.
1820          */
1821         vring_packed_desc_read(vdev, &desc, desc_cache,
1822                                vq->last_avail_idx , true);
1823         if (!is_desc_avail(desc.flags, vq->last_avail_wrap_counter)) {
1824             break;
1825         }
1826         elem.index = desc.id;
1827         elem.ndescs = 1;
1828         while (virtqueue_packed_read_next_desc(vq, &desc, desc_cache,
1829                                                vq->vring.num, &idx, false)) {
1830             ++elem.ndescs;
1831         }
1832         /*
1833          * immediately push the element, nothing to unmap
1834          * as both in_num and out_num are set to 0.
1835          */
1836         virtqueue_push(vq, &elem, 0);
1837         dropped++;
1838         vq->last_avail_idx += elem.ndescs;
1839         if (vq->last_avail_idx >= vq->vring.num) {
1840             vq->last_avail_idx -= vq->vring.num;
1841             vq->last_avail_wrap_counter ^= 1;
1842         }
1843     }
1844 
1845     return dropped;
1846 }
1847 
1848 static unsigned int virtqueue_split_drop_all(VirtQueue *vq)
1849 {
1850     unsigned int dropped = 0;
1851     VirtQueueElement elem = {};
1852     VirtIODevice *vdev = vq->vdev;
1853     bool fEventIdx = virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX);
1854 
1855     while (!virtio_queue_empty(vq) && vq->inuse < vq->vring.num) {
1856         /* works similar to virtqueue_pop but does not map buffers
1857         * and does not allocate any memory */
1858         smp_rmb();
1859         if (!virtqueue_get_head(vq, vq->last_avail_idx, &elem.index)) {
1860             break;
1861         }
1862         vq->inuse++;
1863         vq->last_avail_idx++;
1864         if (fEventIdx) {
1865             vring_set_avail_event(vq, vq->last_avail_idx);
1866         }
1867         /* immediately push the element, nothing to unmap
1868          * as both in_num and out_num are set to 0 */
1869         virtqueue_push(vq, &elem, 0);
1870         dropped++;
1871     }
1872 
1873     return dropped;
1874 }
1875 
1876 /* virtqueue_drop_all:
1877  * @vq: The #VirtQueue
1878  * Drops all queued buffers and indicates them to the guest
1879  * as if they are done. Useful when buffers can not be
1880  * processed but must be returned to the guest.
1881  */
1882 unsigned int virtqueue_drop_all(VirtQueue *vq)
1883 {
1884     struct VirtIODevice *vdev = vq->vdev;
1885 
1886     if (virtio_device_disabled(vq->vdev)) {
1887         return 0;
1888     }
1889 
1890     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
1891         return virtqueue_packed_drop_all(vq);
1892     } else {
1893         return virtqueue_split_drop_all(vq);
1894     }
1895 }
1896 
1897 /* Reading and writing a structure directly to QEMUFile is *awful*, but
1898  * it is what QEMU has always done by mistake.  We can change it sooner
1899  * or later by bumping the version number of the affected vm states.
1900  * In the meanwhile, since the in-memory layout of VirtQueueElement
1901  * has changed, we need to marshal to and from the layout that was
1902  * used before the change.
1903  */
1904 typedef struct VirtQueueElementOld {
1905     unsigned int index;
1906     unsigned int out_num;
1907     unsigned int in_num;
1908     hwaddr in_addr[VIRTQUEUE_MAX_SIZE];
1909     hwaddr out_addr[VIRTQUEUE_MAX_SIZE];
1910     struct iovec in_sg[VIRTQUEUE_MAX_SIZE];
1911     struct iovec out_sg[VIRTQUEUE_MAX_SIZE];
1912 } VirtQueueElementOld;
1913 
1914 void *qemu_get_virtqueue_element(VirtIODevice *vdev, QEMUFile *f, size_t sz)
1915 {
1916     VirtQueueElement *elem;
1917     VirtQueueElementOld data;
1918     int i;
1919 
1920     qemu_get_buffer(f, (uint8_t *)&data, sizeof(VirtQueueElementOld));
1921 
1922     /* TODO: teach all callers that this can fail, and return failure instead
1923      * of asserting here.
1924      * This is just one thing (there are probably more) that must be
1925      * fixed before we can allow NDEBUG compilation.
1926      */
1927     assert(ARRAY_SIZE(data.in_addr) >= data.in_num);
1928     assert(ARRAY_SIZE(data.out_addr) >= data.out_num);
1929 
1930     elem = virtqueue_alloc_element(sz, data.out_num, data.in_num);
1931     elem->index = data.index;
1932 
1933     for (i = 0; i < elem->in_num; i++) {
1934         elem->in_addr[i] = data.in_addr[i];
1935     }
1936 
1937     for (i = 0; i < elem->out_num; i++) {
1938         elem->out_addr[i] = data.out_addr[i];
1939     }
1940 
1941     for (i = 0; i < elem->in_num; i++) {
1942         /* Base is overwritten by virtqueue_map.  */
1943         elem->in_sg[i].iov_base = 0;
1944         elem->in_sg[i].iov_len = data.in_sg[i].iov_len;
1945     }
1946 
1947     for (i = 0; i < elem->out_num; i++) {
1948         /* Base is overwritten by virtqueue_map.  */
1949         elem->out_sg[i].iov_base = 0;
1950         elem->out_sg[i].iov_len = data.out_sg[i].iov_len;
1951     }
1952 
1953     if (virtio_host_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
1954         qemu_get_be32s(f, &elem->ndescs);
1955     }
1956 
1957     virtqueue_map(vdev, elem);
1958     return elem;
1959 }
1960 
1961 void qemu_put_virtqueue_element(VirtIODevice *vdev, QEMUFile *f,
1962                                 VirtQueueElement *elem)
1963 {
1964     VirtQueueElementOld data;
1965     int i;
1966 
1967     memset(&data, 0, sizeof(data));
1968     data.index = elem->index;
1969     data.in_num = elem->in_num;
1970     data.out_num = elem->out_num;
1971 
1972     for (i = 0; i < elem->in_num; i++) {
1973         data.in_addr[i] = elem->in_addr[i];
1974     }
1975 
1976     for (i = 0; i < elem->out_num; i++) {
1977         data.out_addr[i] = elem->out_addr[i];
1978     }
1979 
1980     for (i = 0; i < elem->in_num; i++) {
1981         /* Base is overwritten by virtqueue_map when loading.  Do not
1982          * save it, as it would leak the QEMU address space layout.  */
1983         data.in_sg[i].iov_len = elem->in_sg[i].iov_len;
1984     }
1985 
1986     for (i = 0; i < elem->out_num; i++) {
1987         /* Do not save iov_base as above.  */
1988         data.out_sg[i].iov_len = elem->out_sg[i].iov_len;
1989     }
1990 
1991     if (virtio_host_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
1992         qemu_put_be32s(f, &elem->ndescs);
1993     }
1994 
1995     qemu_put_buffer(f, (uint8_t *)&data, sizeof(VirtQueueElementOld));
1996 }
1997 
1998 /* virtio device */
1999 static void virtio_notify_vector(VirtIODevice *vdev, uint16_t vector)
2000 {
2001     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2002     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
2003 
2004     if (virtio_device_disabled(vdev)) {
2005         return;
2006     }
2007 
2008     if (k->notify) {
2009         k->notify(qbus->parent, vector);
2010     }
2011 }
2012 
2013 void virtio_update_irq(VirtIODevice *vdev)
2014 {
2015     virtio_notify_vector(vdev, VIRTIO_NO_VECTOR);
2016 }
2017 
2018 static int virtio_validate_features(VirtIODevice *vdev)
2019 {
2020     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2021 
2022     if (virtio_host_has_feature(vdev, VIRTIO_F_IOMMU_PLATFORM) &&
2023         !virtio_vdev_has_feature(vdev, VIRTIO_F_IOMMU_PLATFORM)) {
2024         return -EFAULT;
2025     }
2026 
2027     if (k->validate_features) {
2028         return k->validate_features(vdev);
2029     } else {
2030         return 0;
2031     }
2032 }
2033 
2034 int virtio_set_status(VirtIODevice *vdev, uint8_t val)
2035 {
2036     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2037     trace_virtio_set_status(vdev, val);
2038 
2039     if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
2040         if (!(vdev->status & VIRTIO_CONFIG_S_FEATURES_OK) &&
2041             val & VIRTIO_CONFIG_S_FEATURES_OK) {
2042             int ret = virtio_validate_features(vdev);
2043 
2044             if (ret) {
2045                 return ret;
2046             }
2047         }
2048     }
2049 
2050     if ((vdev->status & VIRTIO_CONFIG_S_DRIVER_OK) !=
2051         (val & VIRTIO_CONFIG_S_DRIVER_OK)) {
2052         virtio_set_started(vdev, val & VIRTIO_CONFIG_S_DRIVER_OK);
2053     }
2054 
2055     if (k->set_status) {
2056         k->set_status(vdev, val);
2057     }
2058     vdev->status = val;
2059 
2060     return 0;
2061 }
2062 
2063 static enum virtio_device_endian virtio_default_endian(void)
2064 {
2065     if (target_words_bigendian()) {
2066         return VIRTIO_DEVICE_ENDIAN_BIG;
2067     } else {
2068         return VIRTIO_DEVICE_ENDIAN_LITTLE;
2069     }
2070 }
2071 
2072 static enum virtio_device_endian virtio_current_cpu_endian(void)
2073 {
2074     if (cpu_virtio_is_big_endian(current_cpu)) {
2075         return VIRTIO_DEVICE_ENDIAN_BIG;
2076     } else {
2077         return VIRTIO_DEVICE_ENDIAN_LITTLE;
2078     }
2079 }
2080 
2081 static void __virtio_queue_reset(VirtIODevice *vdev, uint32_t i)
2082 {
2083     vdev->vq[i].vring.desc = 0;
2084     vdev->vq[i].vring.avail = 0;
2085     vdev->vq[i].vring.used = 0;
2086     vdev->vq[i].last_avail_idx = 0;
2087     vdev->vq[i].shadow_avail_idx = 0;
2088     vdev->vq[i].used_idx = 0;
2089     vdev->vq[i].last_avail_wrap_counter = true;
2090     vdev->vq[i].shadow_avail_wrap_counter = true;
2091     vdev->vq[i].used_wrap_counter = true;
2092     virtio_queue_set_vector(vdev, i, VIRTIO_NO_VECTOR);
2093     vdev->vq[i].signalled_used = 0;
2094     vdev->vq[i].signalled_used_valid = false;
2095     vdev->vq[i].notification = true;
2096     vdev->vq[i].vring.num = vdev->vq[i].vring.num_default;
2097     vdev->vq[i].inuse = 0;
2098     virtio_virtqueue_reset_region_cache(&vdev->vq[i]);
2099 }
2100 
2101 void virtio_queue_reset(VirtIODevice *vdev, uint32_t queue_index)
2102 {
2103     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2104 
2105     if (k->queue_reset) {
2106         k->queue_reset(vdev, queue_index);
2107     }
2108 
2109     __virtio_queue_reset(vdev, queue_index);
2110 }
2111 
2112 void virtio_queue_enable(VirtIODevice *vdev, uint32_t queue_index)
2113 {
2114     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2115 
2116     /*
2117      * TODO: Seabios is currently out of spec and triggering this error.
2118      * So this needs to be fixed in Seabios, then this can
2119      * be re-enabled for new machine types only, and also after
2120      * being converted to LOG_GUEST_ERROR.
2121      *
2122     if (!virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
2123         error_report("queue_enable is only supported in devices of virtio "
2124                      "1.0 or later.");
2125     }
2126     */
2127 
2128     if (k->queue_enable) {
2129         k->queue_enable(vdev, queue_index);
2130     }
2131 }
2132 
2133 void virtio_reset(void *opaque)
2134 {
2135     VirtIODevice *vdev = opaque;
2136     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2137     int i;
2138 
2139     virtio_set_status(vdev, 0);
2140     if (current_cpu) {
2141         /* Guest initiated reset */
2142         vdev->device_endian = virtio_current_cpu_endian();
2143     } else {
2144         /* System reset */
2145         vdev->device_endian = virtio_default_endian();
2146     }
2147 
2148     if (vdev->vhost_started && k->get_vhost) {
2149         vhost_reset_device(k->get_vhost(vdev));
2150     }
2151 
2152     if (k->reset) {
2153         k->reset(vdev);
2154     }
2155 
2156     vdev->start_on_kick = false;
2157     vdev->started = false;
2158     vdev->broken = false;
2159     vdev->guest_features = 0;
2160     vdev->queue_sel = 0;
2161     vdev->status = 0;
2162     vdev->disabled = false;
2163     qatomic_set(&vdev->isr, 0);
2164     vdev->config_vector = VIRTIO_NO_VECTOR;
2165     virtio_notify_vector(vdev, vdev->config_vector);
2166 
2167     for(i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2168         __virtio_queue_reset(vdev, i);
2169     }
2170 }
2171 
2172 void virtio_queue_set_addr(VirtIODevice *vdev, int n, hwaddr addr)
2173 {
2174     if (!vdev->vq[n].vring.num) {
2175         return;
2176     }
2177     vdev->vq[n].vring.desc = addr;
2178     virtio_queue_update_rings(vdev, n);
2179 }
2180 
2181 hwaddr virtio_queue_get_addr(VirtIODevice *vdev, int n)
2182 {
2183     return vdev->vq[n].vring.desc;
2184 }
2185 
2186 void virtio_queue_set_rings(VirtIODevice *vdev, int n, hwaddr desc,
2187                             hwaddr avail, hwaddr used)
2188 {
2189     if (!vdev->vq[n].vring.num) {
2190         return;
2191     }
2192     vdev->vq[n].vring.desc = desc;
2193     vdev->vq[n].vring.avail = avail;
2194     vdev->vq[n].vring.used = used;
2195     virtio_init_region_cache(vdev, n);
2196 }
2197 
2198 void virtio_queue_set_num(VirtIODevice *vdev, int n, int num)
2199 {
2200     /* Don't allow guest to flip queue between existent and
2201      * nonexistent states, or to set it to an invalid size.
2202      */
2203     if (!!num != !!vdev->vq[n].vring.num ||
2204         num > VIRTQUEUE_MAX_SIZE ||
2205         num < 0) {
2206         return;
2207     }
2208     vdev->vq[n].vring.num = num;
2209 }
2210 
2211 VirtQueue *virtio_vector_first_queue(VirtIODevice *vdev, uint16_t vector)
2212 {
2213     return QLIST_FIRST(&vdev->vector_queues[vector]);
2214 }
2215 
2216 VirtQueue *virtio_vector_next_queue(VirtQueue *vq)
2217 {
2218     return QLIST_NEXT(vq, node);
2219 }
2220 
2221 int virtio_queue_get_num(VirtIODevice *vdev, int n)
2222 {
2223     return vdev->vq[n].vring.num;
2224 }
2225 
2226 int virtio_queue_get_max_num(VirtIODevice *vdev, int n)
2227 {
2228     return vdev->vq[n].vring.num_default;
2229 }
2230 
2231 int virtio_get_num_queues(VirtIODevice *vdev)
2232 {
2233     int i;
2234 
2235     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2236         if (!virtio_queue_get_num(vdev, i)) {
2237             break;
2238         }
2239     }
2240 
2241     return i;
2242 }
2243 
2244 void virtio_queue_set_align(VirtIODevice *vdev, int n, int align)
2245 {
2246     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2247     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
2248 
2249     /* virtio-1 compliant devices cannot change the alignment */
2250     if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
2251         error_report("tried to modify queue alignment for virtio-1 device");
2252         return;
2253     }
2254     /* Check that the transport told us it was going to do this
2255      * (so a buggy transport will immediately assert rather than
2256      * silently failing to migrate this state)
2257      */
2258     assert(k->has_variable_vring_alignment);
2259 
2260     if (align) {
2261         vdev->vq[n].vring.align = align;
2262         virtio_queue_update_rings(vdev, n);
2263     }
2264 }
2265 
2266 static void virtio_queue_notify_vq(VirtQueue *vq)
2267 {
2268     if (vq->vring.desc && vq->handle_output) {
2269         VirtIODevice *vdev = vq->vdev;
2270 
2271         if (unlikely(vdev->broken)) {
2272             return;
2273         }
2274 
2275         trace_virtio_queue_notify(vdev, vq - vdev->vq, vq);
2276         vq->handle_output(vdev, vq);
2277 
2278         if (unlikely(vdev->start_on_kick)) {
2279             virtio_set_started(vdev, true);
2280         }
2281     }
2282 }
2283 
2284 void virtio_queue_notify(VirtIODevice *vdev, int n)
2285 {
2286     VirtQueue *vq = &vdev->vq[n];
2287 
2288     if (unlikely(!vq->vring.desc || vdev->broken)) {
2289         return;
2290     }
2291 
2292     trace_virtio_queue_notify(vdev, vq - vdev->vq, vq);
2293     if (vq->host_notifier_enabled) {
2294         event_notifier_set(&vq->host_notifier);
2295     } else if (vq->handle_output) {
2296         vq->handle_output(vdev, vq);
2297 
2298         if (unlikely(vdev->start_on_kick)) {
2299             virtio_set_started(vdev, true);
2300         }
2301     }
2302 }
2303 
2304 uint16_t virtio_queue_vector(VirtIODevice *vdev, int n)
2305 {
2306     return n < VIRTIO_QUEUE_MAX ? vdev->vq[n].vector :
2307         VIRTIO_NO_VECTOR;
2308 }
2309 
2310 void virtio_queue_set_vector(VirtIODevice *vdev, int n, uint16_t vector)
2311 {
2312     VirtQueue *vq = &vdev->vq[n];
2313 
2314     if (n < VIRTIO_QUEUE_MAX) {
2315         if (vdev->vector_queues &&
2316             vdev->vq[n].vector != VIRTIO_NO_VECTOR) {
2317             QLIST_REMOVE(vq, node);
2318         }
2319         vdev->vq[n].vector = vector;
2320         if (vdev->vector_queues &&
2321             vector != VIRTIO_NO_VECTOR) {
2322             QLIST_INSERT_HEAD(&vdev->vector_queues[vector], vq, node);
2323         }
2324     }
2325 }
2326 
2327 VirtQueue *virtio_add_queue(VirtIODevice *vdev, int queue_size,
2328                             VirtIOHandleOutput handle_output)
2329 {
2330     int i;
2331 
2332     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2333         if (vdev->vq[i].vring.num == 0)
2334             break;
2335     }
2336 
2337     if (i == VIRTIO_QUEUE_MAX || queue_size > VIRTQUEUE_MAX_SIZE)
2338         abort();
2339 
2340     vdev->vq[i].vring.num = queue_size;
2341     vdev->vq[i].vring.num_default = queue_size;
2342     vdev->vq[i].vring.align = VIRTIO_PCI_VRING_ALIGN;
2343     vdev->vq[i].handle_output = handle_output;
2344     vdev->vq[i].used_elems = g_new0(VirtQueueElement, queue_size);
2345 
2346     return &vdev->vq[i];
2347 }
2348 
2349 void virtio_delete_queue(VirtQueue *vq)
2350 {
2351     vq->vring.num = 0;
2352     vq->vring.num_default = 0;
2353     vq->handle_output = NULL;
2354     g_free(vq->used_elems);
2355     vq->used_elems = NULL;
2356     virtio_virtqueue_reset_region_cache(vq);
2357 }
2358 
2359 void virtio_del_queue(VirtIODevice *vdev, int n)
2360 {
2361     if (n < 0 || n >= VIRTIO_QUEUE_MAX) {
2362         abort();
2363     }
2364 
2365     virtio_delete_queue(&vdev->vq[n]);
2366 }
2367 
2368 static void virtio_set_isr(VirtIODevice *vdev, int value)
2369 {
2370     uint8_t old = qatomic_read(&vdev->isr);
2371 
2372     /* Do not write ISR if it does not change, so that its cacheline remains
2373      * shared in the common case where the guest does not read it.
2374      */
2375     if ((old & value) != value) {
2376         qatomic_or(&vdev->isr, value);
2377     }
2378 }
2379 
2380 /* Called within rcu_read_lock(). */
2381 static bool virtio_split_should_notify(VirtIODevice *vdev, VirtQueue *vq)
2382 {
2383     uint16_t old, new;
2384     bool v;
2385     /* We need to expose used array entries before checking used event. */
2386     smp_mb();
2387     /* Always notify when queue is empty (when feature acknowledge) */
2388     if (virtio_vdev_has_feature(vdev, VIRTIO_F_NOTIFY_ON_EMPTY) &&
2389         !vq->inuse && virtio_queue_empty(vq)) {
2390         return true;
2391     }
2392 
2393     if (!virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX)) {
2394         return !(vring_avail_flags(vq) & VRING_AVAIL_F_NO_INTERRUPT);
2395     }
2396 
2397     v = vq->signalled_used_valid;
2398     vq->signalled_used_valid = true;
2399     old = vq->signalled_used;
2400     new = vq->signalled_used = vq->used_idx;
2401     return !v || vring_need_event(vring_get_used_event(vq), new, old);
2402 }
2403 
2404 static bool vring_packed_need_event(VirtQueue *vq, bool wrap,
2405                                     uint16_t off_wrap, uint16_t new,
2406                                     uint16_t old)
2407 {
2408     int off = off_wrap & ~(1 << 15);
2409 
2410     if (wrap != off_wrap >> 15) {
2411         off -= vq->vring.num;
2412     }
2413 
2414     return vring_need_event(off, new, old);
2415 }
2416 
2417 /* Called within rcu_read_lock(). */
2418 static bool virtio_packed_should_notify(VirtIODevice *vdev, VirtQueue *vq)
2419 {
2420     VRingPackedDescEvent e;
2421     uint16_t old, new;
2422     bool v;
2423     VRingMemoryRegionCaches *caches;
2424 
2425     caches = vring_get_region_caches(vq);
2426     if (!caches) {
2427         return false;
2428     }
2429 
2430     vring_packed_event_read(vdev, &caches->avail, &e);
2431 
2432     old = vq->signalled_used;
2433     new = vq->signalled_used = vq->used_idx;
2434     v = vq->signalled_used_valid;
2435     vq->signalled_used_valid = true;
2436 
2437     if (e.flags == VRING_PACKED_EVENT_FLAG_DISABLE) {
2438         return false;
2439     } else if (e.flags == VRING_PACKED_EVENT_FLAG_ENABLE) {
2440         return true;
2441     }
2442 
2443     return !v || vring_packed_need_event(vq, vq->used_wrap_counter,
2444                                          e.off_wrap, new, old);
2445 }
2446 
2447 /* Called within rcu_read_lock().  */
2448 static bool virtio_should_notify(VirtIODevice *vdev, VirtQueue *vq)
2449 {
2450     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
2451         return virtio_packed_should_notify(vdev, vq);
2452     } else {
2453         return virtio_split_should_notify(vdev, vq);
2454     }
2455 }
2456 
2457 /* Batch irqs while inside a defer_call_begin()/defer_call_end() section */
2458 static void virtio_notify_irqfd_deferred_fn(void *opaque)
2459 {
2460     EventNotifier *notifier = opaque;
2461     VirtQueue *vq = container_of(notifier, VirtQueue, guest_notifier);
2462 
2463     trace_virtio_notify_irqfd_deferred_fn(vq->vdev, vq);
2464     event_notifier_set(notifier);
2465 }
2466 
2467 void virtio_notify_irqfd(VirtIODevice *vdev, VirtQueue *vq)
2468 {
2469     WITH_RCU_READ_LOCK_GUARD() {
2470         if (!virtio_should_notify(vdev, vq)) {
2471             return;
2472         }
2473     }
2474 
2475     trace_virtio_notify_irqfd(vdev, vq);
2476 
2477     /*
2478      * virtio spec 1.0 says ISR bit 0 should be ignored with MSI, but
2479      * windows drivers included in virtio-win 1.8.0 (circa 2015) are
2480      * incorrectly polling this bit during crashdump and hibernation
2481      * in MSI mode, causing a hang if this bit is never updated.
2482      * Recent releases of Windows do not really shut down, but rather
2483      * log out and hibernate to make the next startup faster.  Hence,
2484      * this manifested as a more serious hang during shutdown with
2485      *
2486      * Next driver release from 2016 fixed this problem, so working around it
2487      * is not a must, but it's easy to do so let's do it here.
2488      *
2489      * Note: it's safe to update ISR from any thread as it was switched
2490      * to an atomic operation.
2491      */
2492     virtio_set_isr(vq->vdev, 0x1);
2493     defer_call(virtio_notify_irqfd_deferred_fn, &vq->guest_notifier);
2494 }
2495 
2496 static void virtio_irq(VirtQueue *vq)
2497 {
2498     virtio_set_isr(vq->vdev, 0x1);
2499     virtio_notify_vector(vq->vdev, vq->vector);
2500 }
2501 
2502 void virtio_notify(VirtIODevice *vdev, VirtQueue *vq)
2503 {
2504     WITH_RCU_READ_LOCK_GUARD() {
2505         if (!virtio_should_notify(vdev, vq)) {
2506             return;
2507         }
2508     }
2509 
2510     trace_virtio_notify(vdev, vq);
2511     virtio_irq(vq);
2512 }
2513 
2514 void virtio_notify_config(VirtIODevice *vdev)
2515 {
2516     if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK))
2517         return;
2518 
2519     virtio_set_isr(vdev, 0x3);
2520     vdev->generation++;
2521     virtio_notify_vector(vdev, vdev->config_vector);
2522 }
2523 
2524 static bool virtio_device_endian_needed(void *opaque)
2525 {
2526     VirtIODevice *vdev = opaque;
2527 
2528     assert(vdev->device_endian != VIRTIO_DEVICE_ENDIAN_UNKNOWN);
2529     if (!virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
2530         return vdev->device_endian != virtio_default_endian();
2531     }
2532     /* Devices conforming to VIRTIO 1.0 or later are always LE. */
2533     return vdev->device_endian != VIRTIO_DEVICE_ENDIAN_LITTLE;
2534 }
2535 
2536 static bool virtio_64bit_features_needed(void *opaque)
2537 {
2538     VirtIODevice *vdev = opaque;
2539 
2540     return (vdev->host_features >> 32) != 0;
2541 }
2542 
2543 static bool virtio_virtqueue_needed(void *opaque)
2544 {
2545     VirtIODevice *vdev = opaque;
2546 
2547     return virtio_host_has_feature(vdev, VIRTIO_F_VERSION_1);
2548 }
2549 
2550 static bool virtio_packed_virtqueue_needed(void *opaque)
2551 {
2552     VirtIODevice *vdev = opaque;
2553 
2554     return virtio_host_has_feature(vdev, VIRTIO_F_RING_PACKED);
2555 }
2556 
2557 static bool virtio_ringsize_needed(void *opaque)
2558 {
2559     VirtIODevice *vdev = opaque;
2560     int i;
2561 
2562     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2563         if (vdev->vq[i].vring.num != vdev->vq[i].vring.num_default) {
2564             return true;
2565         }
2566     }
2567     return false;
2568 }
2569 
2570 static bool virtio_extra_state_needed(void *opaque)
2571 {
2572     VirtIODevice *vdev = opaque;
2573     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2574     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
2575 
2576     return k->has_extra_state &&
2577         k->has_extra_state(qbus->parent);
2578 }
2579 
2580 static bool virtio_broken_needed(void *opaque)
2581 {
2582     VirtIODevice *vdev = opaque;
2583 
2584     return vdev->broken;
2585 }
2586 
2587 static bool virtio_started_needed(void *opaque)
2588 {
2589     VirtIODevice *vdev = opaque;
2590 
2591     return vdev->started;
2592 }
2593 
2594 static bool virtio_disabled_needed(void *opaque)
2595 {
2596     VirtIODevice *vdev = opaque;
2597 
2598     return vdev->disabled;
2599 }
2600 
2601 static const VMStateDescription vmstate_virtqueue = {
2602     .name = "virtqueue_state",
2603     .version_id = 1,
2604     .minimum_version_id = 1,
2605     .fields = (const VMStateField[]) {
2606         VMSTATE_UINT64(vring.avail, struct VirtQueue),
2607         VMSTATE_UINT64(vring.used, struct VirtQueue),
2608         VMSTATE_END_OF_LIST()
2609     }
2610 };
2611 
2612 static const VMStateDescription vmstate_packed_virtqueue = {
2613     .name = "packed_virtqueue_state",
2614     .version_id = 1,
2615     .minimum_version_id = 1,
2616     .fields = (const VMStateField[]) {
2617         VMSTATE_UINT16(last_avail_idx, struct VirtQueue),
2618         VMSTATE_BOOL(last_avail_wrap_counter, struct VirtQueue),
2619         VMSTATE_UINT16(used_idx, struct VirtQueue),
2620         VMSTATE_BOOL(used_wrap_counter, struct VirtQueue),
2621         VMSTATE_UINT32(inuse, struct VirtQueue),
2622         VMSTATE_END_OF_LIST()
2623     }
2624 };
2625 
2626 static const VMStateDescription vmstate_virtio_virtqueues = {
2627     .name = "virtio/virtqueues",
2628     .version_id = 1,
2629     .minimum_version_id = 1,
2630     .needed = &virtio_virtqueue_needed,
2631     .fields = (const VMStateField[]) {
2632         VMSTATE_STRUCT_VARRAY_POINTER_KNOWN(vq, struct VirtIODevice,
2633                       VIRTIO_QUEUE_MAX, 0, vmstate_virtqueue, VirtQueue),
2634         VMSTATE_END_OF_LIST()
2635     }
2636 };
2637 
2638 static const VMStateDescription vmstate_virtio_packed_virtqueues = {
2639     .name = "virtio/packed_virtqueues",
2640     .version_id = 1,
2641     .minimum_version_id = 1,
2642     .needed = &virtio_packed_virtqueue_needed,
2643     .fields = (const VMStateField[]) {
2644         VMSTATE_STRUCT_VARRAY_POINTER_KNOWN(vq, struct VirtIODevice,
2645                       VIRTIO_QUEUE_MAX, 0, vmstate_packed_virtqueue, VirtQueue),
2646         VMSTATE_END_OF_LIST()
2647     }
2648 };
2649 
2650 static const VMStateDescription vmstate_ringsize = {
2651     .name = "ringsize_state",
2652     .version_id = 1,
2653     .minimum_version_id = 1,
2654     .fields = (const VMStateField[]) {
2655         VMSTATE_UINT32(vring.num_default, struct VirtQueue),
2656         VMSTATE_END_OF_LIST()
2657     }
2658 };
2659 
2660 static const VMStateDescription vmstate_virtio_ringsize = {
2661     .name = "virtio/ringsize",
2662     .version_id = 1,
2663     .minimum_version_id = 1,
2664     .needed = &virtio_ringsize_needed,
2665     .fields = (const VMStateField[]) {
2666         VMSTATE_STRUCT_VARRAY_POINTER_KNOWN(vq, struct VirtIODevice,
2667                       VIRTIO_QUEUE_MAX, 0, vmstate_ringsize, VirtQueue),
2668         VMSTATE_END_OF_LIST()
2669     }
2670 };
2671 
2672 static int get_extra_state(QEMUFile *f, void *pv, size_t size,
2673                            const VMStateField *field)
2674 {
2675     VirtIODevice *vdev = pv;
2676     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2677     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
2678 
2679     if (!k->load_extra_state) {
2680         return -1;
2681     } else {
2682         return k->load_extra_state(qbus->parent, f);
2683     }
2684 }
2685 
2686 static int put_extra_state(QEMUFile *f, void *pv, size_t size,
2687                            const VMStateField *field, JSONWriter *vmdesc)
2688 {
2689     VirtIODevice *vdev = pv;
2690     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2691     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
2692 
2693     k->save_extra_state(qbus->parent, f);
2694     return 0;
2695 }
2696 
2697 static const VMStateInfo vmstate_info_extra_state = {
2698     .name = "virtqueue_extra_state",
2699     .get = get_extra_state,
2700     .put = put_extra_state,
2701 };
2702 
2703 static const VMStateDescription vmstate_virtio_extra_state = {
2704     .name = "virtio/extra_state",
2705     .version_id = 1,
2706     .minimum_version_id = 1,
2707     .needed = &virtio_extra_state_needed,
2708     .fields = (const VMStateField[]) {
2709         {
2710             .name         = "extra_state",
2711             .version_id   = 0,
2712             .field_exists = NULL,
2713             .size         = 0,
2714             .info         = &vmstate_info_extra_state,
2715             .flags        = VMS_SINGLE,
2716             .offset       = 0,
2717         },
2718         VMSTATE_END_OF_LIST()
2719     }
2720 };
2721 
2722 static const VMStateDescription vmstate_virtio_device_endian = {
2723     .name = "virtio/device_endian",
2724     .version_id = 1,
2725     .minimum_version_id = 1,
2726     .needed = &virtio_device_endian_needed,
2727     .fields = (const VMStateField[]) {
2728         VMSTATE_UINT8(device_endian, VirtIODevice),
2729         VMSTATE_END_OF_LIST()
2730     }
2731 };
2732 
2733 static const VMStateDescription vmstate_virtio_64bit_features = {
2734     .name = "virtio/64bit_features",
2735     .version_id = 1,
2736     .minimum_version_id = 1,
2737     .needed = &virtio_64bit_features_needed,
2738     .fields = (const VMStateField[]) {
2739         VMSTATE_UINT64(guest_features, VirtIODevice),
2740         VMSTATE_END_OF_LIST()
2741     }
2742 };
2743 
2744 static const VMStateDescription vmstate_virtio_broken = {
2745     .name = "virtio/broken",
2746     .version_id = 1,
2747     .minimum_version_id = 1,
2748     .needed = &virtio_broken_needed,
2749     .fields = (const VMStateField[]) {
2750         VMSTATE_BOOL(broken, VirtIODevice),
2751         VMSTATE_END_OF_LIST()
2752     }
2753 };
2754 
2755 static const VMStateDescription vmstate_virtio_started = {
2756     .name = "virtio/started",
2757     .version_id = 1,
2758     .minimum_version_id = 1,
2759     .needed = &virtio_started_needed,
2760     .fields = (const VMStateField[]) {
2761         VMSTATE_BOOL(started, VirtIODevice),
2762         VMSTATE_END_OF_LIST()
2763     }
2764 };
2765 
2766 static const VMStateDescription vmstate_virtio_disabled = {
2767     .name = "virtio/disabled",
2768     .version_id = 1,
2769     .minimum_version_id = 1,
2770     .needed = &virtio_disabled_needed,
2771     .fields = (const VMStateField[]) {
2772         VMSTATE_BOOL(disabled, VirtIODevice),
2773         VMSTATE_END_OF_LIST()
2774     }
2775 };
2776 
2777 static const VMStateDescription vmstate_virtio = {
2778     .name = "virtio",
2779     .version_id = 1,
2780     .minimum_version_id = 1,
2781     .fields = (const VMStateField[]) {
2782         VMSTATE_END_OF_LIST()
2783     },
2784     .subsections = (const VMStateDescription * const []) {
2785         &vmstate_virtio_device_endian,
2786         &vmstate_virtio_64bit_features,
2787         &vmstate_virtio_virtqueues,
2788         &vmstate_virtio_ringsize,
2789         &vmstate_virtio_broken,
2790         &vmstate_virtio_extra_state,
2791         &vmstate_virtio_started,
2792         &vmstate_virtio_packed_virtqueues,
2793         &vmstate_virtio_disabled,
2794         NULL
2795     }
2796 };
2797 
2798 int virtio_save(VirtIODevice *vdev, QEMUFile *f)
2799 {
2800     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2801     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
2802     VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev);
2803     uint32_t guest_features_lo = (vdev->guest_features & 0xffffffff);
2804     int i;
2805 
2806     if (k->save_config) {
2807         k->save_config(qbus->parent, f);
2808     }
2809 
2810     qemu_put_8s(f, &vdev->status);
2811     qemu_put_8s(f, &vdev->isr);
2812     qemu_put_be16s(f, &vdev->queue_sel);
2813     qemu_put_be32s(f, &guest_features_lo);
2814     qemu_put_be32(f, vdev->config_len);
2815     qemu_put_buffer(f, vdev->config, vdev->config_len);
2816 
2817     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2818         if (vdev->vq[i].vring.num == 0)
2819             break;
2820     }
2821 
2822     qemu_put_be32(f, i);
2823 
2824     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2825         if (vdev->vq[i].vring.num == 0)
2826             break;
2827 
2828         qemu_put_be32(f, vdev->vq[i].vring.num);
2829         if (k->has_variable_vring_alignment) {
2830             qemu_put_be32(f, vdev->vq[i].vring.align);
2831         }
2832         /*
2833          * Save desc now, the rest of the ring addresses are saved in
2834          * subsections for VIRTIO-1 devices.
2835          */
2836         qemu_put_be64(f, vdev->vq[i].vring.desc);
2837         qemu_put_be16s(f, &vdev->vq[i].last_avail_idx);
2838         if (k->save_queue) {
2839             k->save_queue(qbus->parent, i, f);
2840         }
2841     }
2842 
2843     if (vdc->save != NULL) {
2844         vdc->save(vdev, f);
2845     }
2846 
2847     if (vdc->vmsd) {
2848         int ret = vmstate_save_state(f, vdc->vmsd, vdev, NULL);
2849         if (ret) {
2850             return ret;
2851         }
2852     }
2853 
2854     /* Subsections */
2855     return vmstate_save_state(f, &vmstate_virtio, vdev, NULL);
2856 }
2857 
2858 /* A wrapper for use as a VMState .put function */
2859 static int virtio_device_put(QEMUFile *f, void *opaque, size_t size,
2860                               const VMStateField *field, JSONWriter *vmdesc)
2861 {
2862     return virtio_save(VIRTIO_DEVICE(opaque), f);
2863 }
2864 
2865 /* A wrapper for use as a VMState .get function */
2866 static int coroutine_mixed_fn
2867 virtio_device_get(QEMUFile *f, void *opaque, size_t size,
2868                   const VMStateField *field)
2869 {
2870     VirtIODevice *vdev = VIRTIO_DEVICE(opaque);
2871     DeviceClass *dc = DEVICE_CLASS(VIRTIO_DEVICE_GET_CLASS(vdev));
2872 
2873     return virtio_load(vdev, f, dc->vmsd->version_id);
2874 }
2875 
2876 const VMStateInfo  virtio_vmstate_info = {
2877     .name = "virtio",
2878     .get = virtio_device_get,
2879     .put = virtio_device_put,
2880 };
2881 
2882 static int virtio_set_features_nocheck(VirtIODevice *vdev, uint64_t val)
2883 {
2884     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2885     bool bad = (val & ~(vdev->host_features)) != 0;
2886 
2887     val &= vdev->host_features;
2888     if (k->set_features) {
2889         k->set_features(vdev, val);
2890     }
2891     vdev->guest_features = val;
2892     return bad ? -1 : 0;
2893 }
2894 
2895 typedef struct VirtioSetFeaturesNocheckData {
2896     Coroutine *co;
2897     VirtIODevice *vdev;
2898     uint64_t val;
2899     int ret;
2900 } VirtioSetFeaturesNocheckData;
2901 
2902 static void virtio_set_features_nocheck_bh(void *opaque)
2903 {
2904     VirtioSetFeaturesNocheckData *data = opaque;
2905 
2906     data->ret = virtio_set_features_nocheck(data->vdev, data->val);
2907     aio_co_wake(data->co);
2908 }
2909 
2910 static int coroutine_mixed_fn
2911 virtio_set_features_nocheck_maybe_co(VirtIODevice *vdev, uint64_t val)
2912 {
2913     if (qemu_in_coroutine()) {
2914         VirtioSetFeaturesNocheckData data = {
2915             .co = qemu_coroutine_self(),
2916             .vdev = vdev,
2917             .val = val,
2918         };
2919         aio_bh_schedule_oneshot(qemu_get_current_aio_context(),
2920                                 virtio_set_features_nocheck_bh, &data);
2921         qemu_coroutine_yield();
2922         return data.ret;
2923     } else {
2924         return virtio_set_features_nocheck(vdev, val);
2925     }
2926 }
2927 
2928 int virtio_set_features(VirtIODevice *vdev, uint64_t val)
2929 {
2930     int ret;
2931     /*
2932      * The driver must not attempt to set features after feature negotiation
2933      * has finished.
2934      */
2935     if (vdev->status & VIRTIO_CONFIG_S_FEATURES_OK) {
2936         return -EINVAL;
2937     }
2938 
2939     if (val & (1ull << VIRTIO_F_BAD_FEATURE)) {
2940         qemu_log_mask(LOG_GUEST_ERROR,
2941                       "%s: guest driver for %s has enabled UNUSED(30) feature bit!\n",
2942                       __func__, vdev->name);
2943     }
2944 
2945     ret = virtio_set_features_nocheck(vdev, val);
2946     if (virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX)) {
2947         /* VIRTIO_RING_F_EVENT_IDX changes the size of the caches.  */
2948         int i;
2949         for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2950             if (vdev->vq[i].vring.num != 0) {
2951                 virtio_init_region_cache(vdev, i);
2952             }
2953         }
2954     }
2955     if (!ret) {
2956         if (!virtio_device_started(vdev, vdev->status) &&
2957             !virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
2958             vdev->start_on_kick = true;
2959         }
2960     }
2961     return ret;
2962 }
2963 
2964 size_t virtio_get_config_size(const VirtIOConfigSizeParams *params,
2965                               uint64_t host_features)
2966 {
2967     size_t config_size = params->min_size;
2968     const VirtIOFeature *feature_sizes = params->feature_sizes;
2969     size_t i;
2970 
2971     for (i = 0; feature_sizes[i].flags != 0; i++) {
2972         if (host_features & feature_sizes[i].flags) {
2973             config_size = MAX(feature_sizes[i].end, config_size);
2974         }
2975     }
2976 
2977     assert(config_size <= params->max_size);
2978     return config_size;
2979 }
2980 
2981 int coroutine_mixed_fn
2982 virtio_load(VirtIODevice *vdev, QEMUFile *f, int version_id)
2983 {
2984     int i, ret;
2985     int32_t config_len;
2986     uint32_t num;
2987     uint32_t features;
2988     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2989     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
2990     VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev);
2991 
2992     /*
2993      * We poison the endianness to ensure it does not get used before
2994      * subsections have been loaded.
2995      */
2996     vdev->device_endian = VIRTIO_DEVICE_ENDIAN_UNKNOWN;
2997 
2998     if (k->load_config) {
2999         ret = k->load_config(qbus->parent, f);
3000         if (ret)
3001             return ret;
3002     }
3003 
3004     qemu_get_8s(f, &vdev->status);
3005     qemu_get_8s(f, &vdev->isr);
3006     qemu_get_be16s(f, &vdev->queue_sel);
3007     if (vdev->queue_sel >= VIRTIO_QUEUE_MAX) {
3008         return -1;
3009     }
3010     qemu_get_be32s(f, &features);
3011 
3012     /*
3013      * Temporarily set guest_features low bits - needed by
3014      * virtio net load code testing for VIRTIO_NET_F_CTRL_GUEST_OFFLOADS
3015      * VIRTIO_NET_F_GUEST_ANNOUNCE and VIRTIO_NET_F_CTRL_VQ.
3016      *
3017      * Note: devices should always test host features in future - don't create
3018      * new dependencies like this.
3019      */
3020     vdev->guest_features = features;
3021 
3022     config_len = qemu_get_be32(f);
3023 
3024     /*
3025      * There are cases where the incoming config can be bigger or smaller
3026      * than what we have; so load what we have space for, and skip
3027      * any excess that's in the stream.
3028      */
3029     qemu_get_buffer(f, vdev->config, MIN(config_len, vdev->config_len));
3030 
3031     while (config_len > vdev->config_len) {
3032         qemu_get_byte(f);
3033         config_len--;
3034     }
3035 
3036     num = qemu_get_be32(f);
3037 
3038     if (num > VIRTIO_QUEUE_MAX) {
3039         error_report("Invalid number of virtqueues: 0x%x", num);
3040         return -1;
3041     }
3042 
3043     for (i = 0; i < num; i++) {
3044         vdev->vq[i].vring.num = qemu_get_be32(f);
3045         if (k->has_variable_vring_alignment) {
3046             vdev->vq[i].vring.align = qemu_get_be32(f);
3047         }
3048         vdev->vq[i].vring.desc = qemu_get_be64(f);
3049         qemu_get_be16s(f, &vdev->vq[i].last_avail_idx);
3050         vdev->vq[i].signalled_used_valid = false;
3051         vdev->vq[i].notification = true;
3052 
3053         if (!vdev->vq[i].vring.desc && vdev->vq[i].last_avail_idx) {
3054             error_report("VQ %d address 0x0 "
3055                          "inconsistent with Host index 0x%x",
3056                          i, vdev->vq[i].last_avail_idx);
3057             return -1;
3058         }
3059         if (k->load_queue) {
3060             ret = k->load_queue(qbus->parent, i, f);
3061             if (ret)
3062                 return ret;
3063         }
3064     }
3065 
3066     virtio_notify_vector(vdev, VIRTIO_NO_VECTOR);
3067 
3068     if (vdc->load != NULL) {
3069         ret = vdc->load(vdev, f, version_id);
3070         if (ret) {
3071             return ret;
3072         }
3073     }
3074 
3075     if (vdc->vmsd) {
3076         ret = vmstate_load_state(f, vdc->vmsd, vdev, version_id);
3077         if (ret) {
3078             return ret;
3079         }
3080     }
3081 
3082     /* Subsections */
3083     ret = vmstate_load_state(f, &vmstate_virtio, vdev, 1);
3084     if (ret) {
3085         return ret;
3086     }
3087 
3088     if (vdev->device_endian == VIRTIO_DEVICE_ENDIAN_UNKNOWN) {
3089         vdev->device_endian = virtio_default_endian();
3090     }
3091 
3092     if (virtio_64bit_features_needed(vdev)) {
3093         /*
3094          * Subsection load filled vdev->guest_features.  Run them
3095          * through virtio_set_features to sanity-check them against
3096          * host_features.
3097          */
3098         uint64_t features64 = vdev->guest_features;
3099         if (virtio_set_features_nocheck_maybe_co(vdev, features64) < 0) {
3100             error_report("Features 0x%" PRIx64 " unsupported. "
3101                          "Allowed features: 0x%" PRIx64,
3102                          features64, vdev->host_features);
3103             return -1;
3104         }
3105     } else {
3106         if (virtio_set_features_nocheck_maybe_co(vdev, features) < 0) {
3107             error_report("Features 0x%x unsupported. "
3108                          "Allowed features: 0x%" PRIx64,
3109                          features, vdev->host_features);
3110             return -1;
3111         }
3112     }
3113 
3114     if (!virtio_device_started(vdev, vdev->status) &&
3115         !virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
3116         vdev->start_on_kick = true;
3117     }
3118 
3119     RCU_READ_LOCK_GUARD();
3120     for (i = 0; i < num; i++) {
3121         if (vdev->vq[i].vring.desc) {
3122             uint16_t nheads;
3123 
3124             /*
3125              * VIRTIO-1 devices migrate desc, used, and avail ring addresses so
3126              * only the region cache needs to be set up.  Legacy devices need
3127              * to calculate used and avail ring addresses based on the desc
3128              * address.
3129              */
3130             if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
3131                 virtio_init_region_cache(vdev, i);
3132             } else {
3133                 virtio_queue_update_rings(vdev, i);
3134             }
3135 
3136             if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3137                 vdev->vq[i].shadow_avail_idx = vdev->vq[i].last_avail_idx;
3138                 vdev->vq[i].shadow_avail_wrap_counter =
3139                                         vdev->vq[i].last_avail_wrap_counter;
3140                 continue;
3141             }
3142 
3143             nheads = vring_avail_idx(&vdev->vq[i]) - vdev->vq[i].last_avail_idx;
3144             /* Check it isn't doing strange things with descriptor numbers. */
3145             if (nheads > vdev->vq[i].vring.num) {
3146                 virtio_error(vdev, "VQ %d size 0x%x Guest index 0x%x "
3147                              "inconsistent with Host index 0x%x: delta 0x%x",
3148                              i, vdev->vq[i].vring.num,
3149                              vring_avail_idx(&vdev->vq[i]),
3150                              vdev->vq[i].last_avail_idx, nheads);
3151                 vdev->vq[i].used_idx = 0;
3152                 vdev->vq[i].shadow_avail_idx = 0;
3153                 vdev->vq[i].inuse = 0;
3154                 continue;
3155             }
3156             vdev->vq[i].used_idx = vring_used_idx(&vdev->vq[i]);
3157             vdev->vq[i].shadow_avail_idx = vring_avail_idx(&vdev->vq[i]);
3158 
3159             /*
3160              * Some devices migrate VirtQueueElements that have been popped
3161              * from the avail ring but not yet returned to the used ring.
3162              * Since max ring size < UINT16_MAX it's safe to use modulo
3163              * UINT16_MAX + 1 subtraction.
3164              */
3165             vdev->vq[i].inuse = (uint16_t)(vdev->vq[i].last_avail_idx -
3166                                 vdev->vq[i].used_idx);
3167             if (vdev->vq[i].inuse > vdev->vq[i].vring.num) {
3168                 error_report("VQ %d size 0x%x < last_avail_idx 0x%x - "
3169                              "used_idx 0x%x",
3170                              i, vdev->vq[i].vring.num,
3171                              vdev->vq[i].last_avail_idx,
3172                              vdev->vq[i].used_idx);
3173                 return -1;
3174             }
3175         }
3176     }
3177 
3178     if (vdc->post_load) {
3179         ret = vdc->post_load(vdev);
3180         if (ret) {
3181             return ret;
3182         }
3183     }
3184 
3185     return 0;
3186 }
3187 
3188 void virtio_cleanup(VirtIODevice *vdev)
3189 {
3190     qemu_del_vm_change_state_handler(vdev->vmstate);
3191 }
3192 
3193 static void virtio_vmstate_change(void *opaque, bool running, RunState state)
3194 {
3195     VirtIODevice *vdev = opaque;
3196     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3197     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
3198     bool backend_run = running && virtio_device_started(vdev, vdev->status);
3199     vdev->vm_running = running;
3200 
3201     if (backend_run) {
3202         virtio_set_status(vdev, vdev->status);
3203     }
3204 
3205     if (k->vmstate_change) {
3206         k->vmstate_change(qbus->parent, backend_run);
3207     }
3208 
3209     if (!backend_run) {
3210         virtio_set_status(vdev, vdev->status);
3211     }
3212 }
3213 
3214 void virtio_instance_init_common(Object *proxy_obj, void *data,
3215                                  size_t vdev_size, const char *vdev_name)
3216 {
3217     DeviceState *vdev = data;
3218 
3219     object_initialize_child_with_props(proxy_obj, "virtio-backend", vdev,
3220                                        vdev_size, vdev_name, &error_abort,
3221                                        NULL);
3222     qdev_alias_all_properties(vdev, proxy_obj);
3223 }
3224 
3225 void virtio_init(VirtIODevice *vdev, uint16_t device_id, size_t config_size)
3226 {
3227     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3228     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
3229     int i;
3230     int nvectors = k->query_nvectors ? k->query_nvectors(qbus->parent) : 0;
3231 
3232     if (nvectors) {
3233         vdev->vector_queues =
3234             g_malloc0(sizeof(*vdev->vector_queues) * nvectors);
3235     }
3236 
3237     vdev->start_on_kick = false;
3238     vdev->started = false;
3239     vdev->vhost_started = false;
3240     vdev->device_id = device_id;
3241     vdev->status = 0;
3242     qatomic_set(&vdev->isr, 0);
3243     vdev->queue_sel = 0;
3244     vdev->config_vector = VIRTIO_NO_VECTOR;
3245     vdev->vq = g_new0(VirtQueue, VIRTIO_QUEUE_MAX);
3246     vdev->vm_running = runstate_is_running();
3247     vdev->broken = false;
3248     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
3249         vdev->vq[i].vector = VIRTIO_NO_VECTOR;
3250         vdev->vq[i].vdev = vdev;
3251         vdev->vq[i].queue_index = i;
3252         vdev->vq[i].host_notifier_enabled = false;
3253     }
3254 
3255     vdev->name = virtio_id_to_name(device_id);
3256     vdev->config_len = config_size;
3257     if (vdev->config_len) {
3258         vdev->config = g_malloc0(config_size);
3259     } else {
3260         vdev->config = NULL;
3261     }
3262     vdev->vmstate = qdev_add_vm_change_state_handler(DEVICE(vdev),
3263             virtio_vmstate_change, vdev);
3264     vdev->device_endian = virtio_default_endian();
3265     vdev->use_guest_notifier_mask = true;
3266 }
3267 
3268 /*
3269  * Only devices that have already been around prior to defining the virtio
3270  * standard support legacy mode; this includes devices not specified in the
3271  * standard. All newer devices conform to the virtio standard only.
3272  */
3273 bool virtio_legacy_allowed(VirtIODevice *vdev)
3274 {
3275     switch (vdev->device_id) {
3276     case VIRTIO_ID_NET:
3277     case VIRTIO_ID_BLOCK:
3278     case VIRTIO_ID_CONSOLE:
3279     case VIRTIO_ID_RNG:
3280     case VIRTIO_ID_BALLOON:
3281     case VIRTIO_ID_RPMSG:
3282     case VIRTIO_ID_SCSI:
3283     case VIRTIO_ID_9P:
3284     case VIRTIO_ID_RPROC_SERIAL:
3285     case VIRTIO_ID_CAIF:
3286         return true;
3287     default:
3288         return false;
3289     }
3290 }
3291 
3292 bool virtio_legacy_check_disabled(VirtIODevice *vdev)
3293 {
3294     return vdev->disable_legacy_check;
3295 }
3296 
3297 hwaddr virtio_queue_get_desc_addr(VirtIODevice *vdev, int n)
3298 {
3299     return vdev->vq[n].vring.desc;
3300 }
3301 
3302 bool virtio_queue_enabled_legacy(VirtIODevice *vdev, int n)
3303 {
3304     return virtio_queue_get_desc_addr(vdev, n) != 0;
3305 }
3306 
3307 bool virtio_queue_enabled(VirtIODevice *vdev, int n)
3308 {
3309     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3310     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
3311 
3312     if (k->queue_enabled) {
3313         return k->queue_enabled(qbus->parent, n);
3314     }
3315     return virtio_queue_enabled_legacy(vdev, n);
3316 }
3317 
3318 hwaddr virtio_queue_get_avail_addr(VirtIODevice *vdev, int n)
3319 {
3320     return vdev->vq[n].vring.avail;
3321 }
3322 
3323 hwaddr virtio_queue_get_used_addr(VirtIODevice *vdev, int n)
3324 {
3325     return vdev->vq[n].vring.used;
3326 }
3327 
3328 hwaddr virtio_queue_get_desc_size(VirtIODevice *vdev, int n)
3329 {
3330     return sizeof(VRingDesc) * vdev->vq[n].vring.num;
3331 }
3332 
3333 hwaddr virtio_queue_get_avail_size(VirtIODevice *vdev, int n)
3334 {
3335     int s;
3336 
3337     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3338         return sizeof(struct VRingPackedDescEvent);
3339     }
3340 
3341     s = virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0;
3342     return offsetof(VRingAvail, ring) +
3343         sizeof(uint16_t) * vdev->vq[n].vring.num + s;
3344 }
3345 
3346 hwaddr virtio_queue_get_used_size(VirtIODevice *vdev, int n)
3347 {
3348     int s;
3349 
3350     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3351         return sizeof(struct VRingPackedDescEvent);
3352     }
3353 
3354     s = virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0;
3355     return offsetof(VRingUsed, ring) +
3356         sizeof(VRingUsedElem) * vdev->vq[n].vring.num + s;
3357 }
3358 
3359 static unsigned int virtio_queue_packed_get_last_avail_idx(VirtIODevice *vdev,
3360                                                            int n)
3361 {
3362     unsigned int avail, used;
3363 
3364     avail = vdev->vq[n].last_avail_idx;
3365     avail |= ((uint16_t)vdev->vq[n].last_avail_wrap_counter) << 15;
3366 
3367     used = vdev->vq[n].used_idx;
3368     used |= ((uint16_t)vdev->vq[n].used_wrap_counter) << 15;
3369 
3370     return avail | used << 16;
3371 }
3372 
3373 static uint16_t virtio_queue_split_get_last_avail_idx(VirtIODevice *vdev,
3374                                                       int n)
3375 {
3376     return vdev->vq[n].last_avail_idx;
3377 }
3378 
3379 unsigned int virtio_queue_get_last_avail_idx(VirtIODevice *vdev, int n)
3380 {
3381     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3382         return virtio_queue_packed_get_last_avail_idx(vdev, n);
3383     } else {
3384         return virtio_queue_split_get_last_avail_idx(vdev, n);
3385     }
3386 }
3387 
3388 static void virtio_queue_packed_set_last_avail_idx(VirtIODevice *vdev,
3389                                                    int n, unsigned int idx)
3390 {
3391     struct VirtQueue *vq = &vdev->vq[n];
3392 
3393     vq->last_avail_idx = vq->shadow_avail_idx = idx & 0x7fff;
3394     vq->last_avail_wrap_counter =
3395         vq->shadow_avail_wrap_counter = !!(idx & 0x8000);
3396     idx >>= 16;
3397     vq->used_idx = idx & 0x7fff;
3398     vq->used_wrap_counter = !!(idx & 0x8000);
3399 }
3400 
3401 static void virtio_queue_split_set_last_avail_idx(VirtIODevice *vdev,
3402                                                   int n, unsigned int idx)
3403 {
3404         vdev->vq[n].last_avail_idx = idx;
3405         vdev->vq[n].shadow_avail_idx = idx;
3406 }
3407 
3408 void virtio_queue_set_last_avail_idx(VirtIODevice *vdev, int n,
3409                                      unsigned int idx)
3410 {
3411     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3412         virtio_queue_packed_set_last_avail_idx(vdev, n, idx);
3413     } else {
3414         virtio_queue_split_set_last_avail_idx(vdev, n, idx);
3415     }
3416 }
3417 
3418 static void virtio_queue_packed_restore_last_avail_idx(VirtIODevice *vdev,
3419                                                        int n)
3420 {
3421     /* We don't have a reference like avail idx in shared memory */
3422     return;
3423 }
3424 
3425 static void virtio_queue_split_restore_last_avail_idx(VirtIODevice *vdev,
3426                                                       int n)
3427 {
3428     RCU_READ_LOCK_GUARD();
3429     if (vdev->vq[n].vring.desc) {
3430         vdev->vq[n].last_avail_idx = vring_used_idx(&vdev->vq[n]);
3431         vdev->vq[n].shadow_avail_idx = vdev->vq[n].last_avail_idx;
3432     }
3433 }
3434 
3435 void virtio_queue_restore_last_avail_idx(VirtIODevice *vdev, int n)
3436 {
3437     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3438         virtio_queue_packed_restore_last_avail_idx(vdev, n);
3439     } else {
3440         virtio_queue_split_restore_last_avail_idx(vdev, n);
3441     }
3442 }
3443 
3444 static void virtio_queue_packed_update_used_idx(VirtIODevice *vdev, int n)
3445 {
3446     /* used idx was updated through set_last_avail_idx() */
3447     return;
3448 }
3449 
3450 static void virtio_split_packed_update_used_idx(VirtIODevice *vdev, int n)
3451 {
3452     RCU_READ_LOCK_GUARD();
3453     if (vdev->vq[n].vring.desc) {
3454         vdev->vq[n].used_idx = vring_used_idx(&vdev->vq[n]);
3455     }
3456 }
3457 
3458 void virtio_queue_update_used_idx(VirtIODevice *vdev, int n)
3459 {
3460     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3461         return virtio_queue_packed_update_used_idx(vdev, n);
3462     } else {
3463         return virtio_split_packed_update_used_idx(vdev, n);
3464     }
3465 }
3466 
3467 void virtio_queue_invalidate_signalled_used(VirtIODevice *vdev, int n)
3468 {
3469     vdev->vq[n].signalled_used_valid = false;
3470 }
3471 
3472 VirtQueue *virtio_get_queue(VirtIODevice *vdev, int n)
3473 {
3474     return vdev->vq + n;
3475 }
3476 
3477 uint16_t virtio_get_queue_index(VirtQueue *vq)
3478 {
3479     return vq->queue_index;
3480 }
3481 
3482 static void virtio_queue_guest_notifier_read(EventNotifier *n)
3483 {
3484     VirtQueue *vq = container_of(n, VirtQueue, guest_notifier);
3485     if (event_notifier_test_and_clear(n)) {
3486         virtio_irq(vq);
3487     }
3488 }
3489 static void virtio_config_guest_notifier_read(EventNotifier *n)
3490 {
3491     VirtIODevice *vdev = container_of(n, VirtIODevice, config_notifier);
3492 
3493     if (event_notifier_test_and_clear(n)) {
3494         virtio_notify_config(vdev);
3495     }
3496 }
3497 void virtio_queue_set_guest_notifier_fd_handler(VirtQueue *vq, bool assign,
3498                                                 bool with_irqfd)
3499 {
3500     if (assign && !with_irqfd) {
3501         event_notifier_set_handler(&vq->guest_notifier,
3502                                    virtio_queue_guest_notifier_read);
3503     } else {
3504         event_notifier_set_handler(&vq->guest_notifier, NULL);
3505     }
3506     if (!assign) {
3507         /* Test and clear notifier before closing it,
3508          * in case poll callback didn't have time to run. */
3509         virtio_queue_guest_notifier_read(&vq->guest_notifier);
3510     }
3511 }
3512 
3513 void virtio_config_set_guest_notifier_fd_handler(VirtIODevice *vdev,
3514                                                  bool assign, bool with_irqfd)
3515 {
3516     EventNotifier *n;
3517     n = &vdev->config_notifier;
3518     if (assign && !with_irqfd) {
3519         event_notifier_set_handler(n, virtio_config_guest_notifier_read);
3520     } else {
3521         event_notifier_set_handler(n, NULL);
3522     }
3523     if (!assign) {
3524         /* Test and clear notifier before closing it,*/
3525         /* in case poll callback didn't have time to run. */
3526         virtio_config_guest_notifier_read(n);
3527     }
3528 }
3529 
3530 EventNotifier *virtio_queue_get_guest_notifier(VirtQueue *vq)
3531 {
3532     return &vq->guest_notifier;
3533 }
3534 
3535 static void virtio_queue_host_notifier_aio_poll_begin(EventNotifier *n)
3536 {
3537     VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
3538 
3539     virtio_queue_set_notification(vq, 0);
3540 }
3541 
3542 static bool virtio_queue_host_notifier_aio_poll(void *opaque)
3543 {
3544     EventNotifier *n = opaque;
3545     VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
3546 
3547     return vq->vring.desc && !virtio_queue_empty(vq);
3548 }
3549 
3550 static void virtio_queue_host_notifier_aio_poll_ready(EventNotifier *n)
3551 {
3552     VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
3553 
3554     virtio_queue_notify_vq(vq);
3555 }
3556 
3557 static void virtio_queue_host_notifier_aio_poll_end(EventNotifier *n)
3558 {
3559     VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
3560 
3561     /* Caller polls once more after this to catch requests that race with us */
3562     virtio_queue_set_notification(vq, 1);
3563 }
3564 
3565 void virtio_queue_aio_attach_host_notifier(VirtQueue *vq, AioContext *ctx)
3566 {
3567     /*
3568      * virtio_queue_aio_detach_host_notifier() can leave notifications disabled.
3569      * Re-enable them.  (And if detach has not been used before, notifications
3570      * being enabled is still the default state while a notifier is attached;
3571      * see virtio_queue_host_notifier_aio_poll_end(), which will always leave
3572      * notifications enabled once the polling section is left.)
3573      */
3574     if (!virtio_queue_get_notification(vq)) {
3575         virtio_queue_set_notification(vq, 1);
3576     }
3577 
3578     aio_set_event_notifier(ctx, &vq->host_notifier,
3579                            virtio_queue_host_notifier_read,
3580                            virtio_queue_host_notifier_aio_poll,
3581                            virtio_queue_host_notifier_aio_poll_ready);
3582     aio_set_event_notifier_poll(ctx, &vq->host_notifier,
3583                                 virtio_queue_host_notifier_aio_poll_begin,
3584                                 virtio_queue_host_notifier_aio_poll_end);
3585 
3586     /*
3587      * We will have ignored notifications about new requests from the guest
3588      * while no notifiers were attached, so "kick" the virt queue to process
3589      * those requests now.
3590      */
3591     event_notifier_set(&vq->host_notifier);
3592 }
3593 
3594 /*
3595  * Same as virtio_queue_aio_attach_host_notifier() but without polling. Use
3596  * this for rx virtqueues and similar cases where the virtqueue handler
3597  * function does not pop all elements. When the virtqueue is left non-empty
3598  * polling consumes CPU cycles and should not be used.
3599  */
3600 void virtio_queue_aio_attach_host_notifier_no_poll(VirtQueue *vq, AioContext *ctx)
3601 {
3602     /* See virtio_queue_aio_attach_host_notifier() */
3603     if (!virtio_queue_get_notification(vq)) {
3604         virtio_queue_set_notification(vq, 1);
3605     }
3606 
3607     aio_set_event_notifier(ctx, &vq->host_notifier,
3608                            virtio_queue_host_notifier_read,
3609                            NULL, NULL);
3610 
3611     /*
3612      * See virtio_queue_aio_attach_host_notifier().
3613      * Note that this may be unnecessary for the type of virtqueues this
3614      * function is used for.  Still, it will not hurt to have a quick look into
3615      * whether we can/should process any of the virtqueue elements.
3616      */
3617     event_notifier_set(&vq->host_notifier);
3618 }
3619 
3620 void virtio_queue_aio_detach_host_notifier(VirtQueue *vq, AioContext *ctx)
3621 {
3622     aio_set_event_notifier(ctx, &vq->host_notifier, NULL, NULL, NULL);
3623 
3624     /*
3625      * aio_set_event_notifier_poll() does not guarantee whether io_poll_end()
3626      * will run after io_poll_begin(), so by removing the notifier, we do not
3627      * know whether virtio_queue_host_notifier_aio_poll_end() has run after a
3628      * previous virtio_queue_host_notifier_aio_poll_begin(), i.e. whether
3629      * notifications are enabled or disabled.  It does not really matter anyway;
3630      * we just removed the notifier, so we do not care about notifications until
3631      * we potentially re-attach it.  The attach_host_notifier functions will
3632      * ensure that notifications are enabled again when they are needed.
3633      */
3634 }
3635 
3636 void virtio_queue_host_notifier_read(EventNotifier *n)
3637 {
3638     VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
3639     if (event_notifier_test_and_clear(n)) {
3640         virtio_queue_notify_vq(vq);
3641     }
3642 }
3643 
3644 EventNotifier *virtio_queue_get_host_notifier(VirtQueue *vq)
3645 {
3646     return &vq->host_notifier;
3647 }
3648 
3649 EventNotifier *virtio_config_get_guest_notifier(VirtIODevice *vdev)
3650 {
3651     return &vdev->config_notifier;
3652 }
3653 
3654 void virtio_queue_set_host_notifier_enabled(VirtQueue *vq, bool enabled)
3655 {
3656     vq->host_notifier_enabled = enabled;
3657 }
3658 
3659 int virtio_queue_set_host_notifier_mr(VirtIODevice *vdev, int n,
3660                                       MemoryRegion *mr, bool assign)
3661 {
3662     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3663     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
3664 
3665     if (k->set_host_notifier_mr) {
3666         return k->set_host_notifier_mr(qbus->parent, n, mr, assign);
3667     }
3668 
3669     return -1;
3670 }
3671 
3672 void virtio_device_set_child_bus_name(VirtIODevice *vdev, char *bus_name)
3673 {
3674     g_free(vdev->bus_name);
3675     vdev->bus_name = g_strdup(bus_name);
3676 }
3677 
3678 void G_GNUC_PRINTF(2, 3) virtio_error(VirtIODevice *vdev, const char *fmt, ...)
3679 {
3680     va_list ap;
3681 
3682     va_start(ap, fmt);
3683     error_vreport(fmt, ap);
3684     va_end(ap);
3685 
3686     if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
3687         vdev->status = vdev->status | VIRTIO_CONFIG_S_NEEDS_RESET;
3688         virtio_notify_config(vdev);
3689     }
3690 
3691     vdev->broken = true;
3692 }
3693 
3694 static void virtio_memory_listener_commit(MemoryListener *listener)
3695 {
3696     VirtIODevice *vdev = container_of(listener, VirtIODevice, listener);
3697     int i;
3698 
3699     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
3700         if (vdev->vq[i].vring.num == 0) {
3701             break;
3702         }
3703         virtio_init_region_cache(vdev, i);
3704     }
3705 }
3706 
3707 static void virtio_device_realize(DeviceState *dev, Error **errp)
3708 {
3709     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
3710     VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(dev);
3711     Error *err = NULL;
3712 
3713     /* Devices should either use vmsd or the load/save methods */
3714     assert(!vdc->vmsd || !vdc->load);
3715 
3716     if (vdc->realize != NULL) {
3717         vdc->realize(dev, &err);
3718         if (err != NULL) {
3719             error_propagate(errp, err);
3720             return;
3721         }
3722     }
3723 
3724     virtio_bus_device_plugged(vdev, &err);
3725     if (err != NULL) {
3726         error_propagate(errp, err);
3727         vdc->unrealize(dev);
3728         return;
3729     }
3730 
3731     vdev->listener.commit = virtio_memory_listener_commit;
3732     vdev->listener.name = "virtio";
3733     memory_listener_register(&vdev->listener, vdev->dma_as);
3734 }
3735 
3736 static void virtio_device_unrealize(DeviceState *dev)
3737 {
3738     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
3739     VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(dev);
3740 
3741     memory_listener_unregister(&vdev->listener);
3742     virtio_bus_device_unplugged(vdev);
3743 
3744     if (vdc->unrealize != NULL) {
3745         vdc->unrealize(dev);
3746     }
3747 
3748     g_free(vdev->bus_name);
3749     vdev->bus_name = NULL;
3750 }
3751 
3752 static void virtio_device_free_virtqueues(VirtIODevice *vdev)
3753 {
3754     int i;
3755     if (!vdev->vq) {
3756         return;
3757     }
3758 
3759     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
3760         if (vdev->vq[i].vring.num == 0) {
3761             break;
3762         }
3763         virtio_virtqueue_reset_region_cache(&vdev->vq[i]);
3764     }
3765     g_free(vdev->vq);
3766 }
3767 
3768 static void virtio_device_instance_finalize(Object *obj)
3769 {
3770     VirtIODevice *vdev = VIRTIO_DEVICE(obj);
3771 
3772     virtio_device_free_virtqueues(vdev);
3773 
3774     g_free(vdev->config);
3775     g_free(vdev->vector_queues);
3776 }
3777 
3778 static Property virtio_properties[] = {
3779     DEFINE_VIRTIO_COMMON_FEATURES(VirtIODevice, host_features),
3780     DEFINE_PROP_BOOL("use-started", VirtIODevice, use_started, true),
3781     DEFINE_PROP_BOOL("use-disabled-flag", VirtIODevice, use_disabled_flag, true),
3782     DEFINE_PROP_BOOL("x-disable-legacy-check", VirtIODevice,
3783                      disable_legacy_check, false),
3784     DEFINE_PROP_END_OF_LIST(),
3785 };
3786 
3787 static int virtio_device_start_ioeventfd_impl(VirtIODevice *vdev)
3788 {
3789     VirtioBusState *qbus = VIRTIO_BUS(qdev_get_parent_bus(DEVICE(vdev)));
3790     int i, n, r, err;
3791 
3792     /*
3793      * Batch all the host notifiers in a single transaction to avoid
3794      * quadratic time complexity in address_space_update_ioeventfds().
3795      */
3796     memory_region_transaction_begin();
3797     for (n = 0; n < VIRTIO_QUEUE_MAX; n++) {
3798         VirtQueue *vq = &vdev->vq[n];
3799         if (!virtio_queue_get_num(vdev, n)) {
3800             continue;
3801         }
3802         r = virtio_bus_set_host_notifier(qbus, n, true);
3803         if (r < 0) {
3804             err = r;
3805             goto assign_error;
3806         }
3807         event_notifier_set_handler(&vq->host_notifier,
3808                                    virtio_queue_host_notifier_read);
3809     }
3810 
3811     for (n = 0; n < VIRTIO_QUEUE_MAX; n++) {
3812         /* Kick right away to begin processing requests already in vring */
3813         VirtQueue *vq = &vdev->vq[n];
3814         if (!vq->vring.num) {
3815             continue;
3816         }
3817         event_notifier_set(&vq->host_notifier);
3818     }
3819     memory_region_transaction_commit();
3820     return 0;
3821 
3822 assign_error:
3823     i = n; /* save n for a second iteration after transaction is committed. */
3824     while (--n >= 0) {
3825         VirtQueue *vq = &vdev->vq[n];
3826         if (!virtio_queue_get_num(vdev, n)) {
3827             continue;
3828         }
3829 
3830         event_notifier_set_handler(&vq->host_notifier, NULL);
3831         r = virtio_bus_set_host_notifier(qbus, n, false);
3832         assert(r >= 0);
3833     }
3834     /*
3835      * The transaction expects the ioeventfds to be open when it
3836      * commits. Do it now, before the cleanup loop.
3837      */
3838     memory_region_transaction_commit();
3839 
3840     while (--i >= 0) {
3841         if (!virtio_queue_get_num(vdev, i)) {
3842             continue;
3843         }
3844         virtio_bus_cleanup_host_notifier(qbus, i);
3845     }
3846     return err;
3847 }
3848 
3849 int virtio_device_start_ioeventfd(VirtIODevice *vdev)
3850 {
3851     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3852     VirtioBusState *vbus = VIRTIO_BUS(qbus);
3853 
3854     return virtio_bus_start_ioeventfd(vbus);
3855 }
3856 
3857 static void virtio_device_stop_ioeventfd_impl(VirtIODevice *vdev)
3858 {
3859     VirtioBusState *qbus = VIRTIO_BUS(qdev_get_parent_bus(DEVICE(vdev)));
3860     int n, r;
3861 
3862     /*
3863      * Batch all the host notifiers in a single transaction to avoid
3864      * quadratic time complexity in address_space_update_ioeventfds().
3865      */
3866     memory_region_transaction_begin();
3867     for (n = 0; n < VIRTIO_QUEUE_MAX; n++) {
3868         VirtQueue *vq = &vdev->vq[n];
3869 
3870         if (!virtio_queue_get_num(vdev, n)) {
3871             continue;
3872         }
3873         event_notifier_set_handler(&vq->host_notifier, NULL);
3874         r = virtio_bus_set_host_notifier(qbus, n, false);
3875         assert(r >= 0);
3876     }
3877     /*
3878      * The transaction expects the ioeventfds to be open when it
3879      * commits. Do it now, before the cleanup loop.
3880      */
3881     memory_region_transaction_commit();
3882 
3883     for (n = 0; n < VIRTIO_QUEUE_MAX; n++) {
3884         if (!virtio_queue_get_num(vdev, n)) {
3885             continue;
3886         }
3887         virtio_bus_cleanup_host_notifier(qbus, n);
3888     }
3889 }
3890 
3891 int virtio_device_grab_ioeventfd(VirtIODevice *vdev)
3892 {
3893     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3894     VirtioBusState *vbus = VIRTIO_BUS(qbus);
3895 
3896     return virtio_bus_grab_ioeventfd(vbus);
3897 }
3898 
3899 void virtio_device_release_ioeventfd(VirtIODevice *vdev)
3900 {
3901     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3902     VirtioBusState *vbus = VIRTIO_BUS(qbus);
3903 
3904     virtio_bus_release_ioeventfd(vbus);
3905 }
3906 
3907 static void virtio_device_class_init(ObjectClass *klass, void *data)
3908 {
3909     /* Set the default value here. */
3910     VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
3911     DeviceClass *dc = DEVICE_CLASS(klass);
3912 
3913     dc->realize = virtio_device_realize;
3914     dc->unrealize = virtio_device_unrealize;
3915     dc->bus_type = TYPE_VIRTIO_BUS;
3916     device_class_set_props(dc, virtio_properties);
3917     vdc->start_ioeventfd = virtio_device_start_ioeventfd_impl;
3918     vdc->stop_ioeventfd = virtio_device_stop_ioeventfd_impl;
3919 
3920     vdc->legacy_features |= VIRTIO_LEGACY_FEATURES;
3921 }
3922 
3923 bool virtio_device_ioeventfd_enabled(VirtIODevice *vdev)
3924 {
3925     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3926     VirtioBusState *vbus = VIRTIO_BUS(qbus);
3927 
3928     return virtio_bus_ioeventfd_enabled(vbus);
3929 }
3930 
3931 VirtQueueStatus *qmp_x_query_virtio_queue_status(const char *path,
3932                                                  uint16_t queue,
3933                                                  Error **errp)
3934 {
3935     VirtIODevice *vdev;
3936     VirtQueueStatus *status;
3937 
3938     vdev = qmp_find_virtio_device(path);
3939     if (vdev == NULL) {
3940         error_setg(errp, "Path %s is not a VirtIODevice", path);
3941         return NULL;
3942     }
3943 
3944     if (queue >= VIRTIO_QUEUE_MAX || !virtio_queue_get_num(vdev, queue)) {
3945         error_setg(errp, "Invalid virtqueue number %d", queue);
3946         return NULL;
3947     }
3948 
3949     status = g_new0(VirtQueueStatus, 1);
3950     status->name = g_strdup(vdev->name);
3951     status->queue_index = vdev->vq[queue].queue_index;
3952     status->inuse = vdev->vq[queue].inuse;
3953     status->vring_num = vdev->vq[queue].vring.num;
3954     status->vring_num_default = vdev->vq[queue].vring.num_default;
3955     status->vring_align = vdev->vq[queue].vring.align;
3956     status->vring_desc = vdev->vq[queue].vring.desc;
3957     status->vring_avail = vdev->vq[queue].vring.avail;
3958     status->vring_used = vdev->vq[queue].vring.used;
3959     status->used_idx = vdev->vq[queue].used_idx;
3960     status->signalled_used = vdev->vq[queue].signalled_used;
3961     status->signalled_used_valid = vdev->vq[queue].signalled_used_valid;
3962 
3963     if (vdev->vhost_started) {
3964         VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev);
3965         struct vhost_dev *hdev = vdc->get_vhost(vdev);
3966 
3967         /* check if vq index exists for vhost as well  */
3968         if (queue >= hdev->vq_index && queue < hdev->vq_index + hdev->nvqs) {
3969             status->has_last_avail_idx = true;
3970 
3971             int vhost_vq_index =
3972                 hdev->vhost_ops->vhost_get_vq_index(hdev, queue);
3973             struct vhost_vring_state state = {
3974                 .index = vhost_vq_index,
3975             };
3976 
3977             status->last_avail_idx =
3978                 hdev->vhost_ops->vhost_get_vring_base(hdev, &state);
3979         }
3980     } else {
3981         status->has_shadow_avail_idx = true;
3982         status->has_last_avail_idx = true;
3983         status->last_avail_idx = vdev->vq[queue].last_avail_idx;
3984         status->shadow_avail_idx = vdev->vq[queue].shadow_avail_idx;
3985     }
3986 
3987     return status;
3988 }
3989 
3990 static strList *qmp_decode_vring_desc_flags(uint16_t flags)
3991 {
3992     strList *list = NULL;
3993     strList *node;
3994     int i;
3995 
3996     struct {
3997         uint16_t flag;
3998         const char *value;
3999     } map[] = {
4000         { VRING_DESC_F_NEXT, "next" },
4001         { VRING_DESC_F_WRITE, "write" },
4002         { VRING_DESC_F_INDIRECT, "indirect" },
4003         { 1 << VRING_PACKED_DESC_F_AVAIL, "avail" },
4004         { 1 << VRING_PACKED_DESC_F_USED, "used" },
4005         { 0, "" }
4006     };
4007 
4008     for (i = 0; map[i].flag; i++) {
4009         if ((map[i].flag & flags) == 0) {
4010             continue;
4011         }
4012         node = g_malloc0(sizeof(strList));
4013         node->value = g_strdup(map[i].value);
4014         node->next = list;
4015         list = node;
4016     }
4017 
4018     return list;
4019 }
4020 
4021 VirtioQueueElement *qmp_x_query_virtio_queue_element(const char *path,
4022                                                      uint16_t queue,
4023                                                      bool has_index,
4024                                                      uint16_t index,
4025                                                      Error **errp)
4026 {
4027     VirtIODevice *vdev;
4028     VirtQueue *vq;
4029     VirtioQueueElement *element = NULL;
4030 
4031     vdev = qmp_find_virtio_device(path);
4032     if (vdev == NULL) {
4033         error_setg(errp, "Path %s is not a VirtIO device", path);
4034         return NULL;
4035     }
4036 
4037     if (queue >= VIRTIO_QUEUE_MAX || !virtio_queue_get_num(vdev, queue)) {
4038         error_setg(errp, "Invalid virtqueue number %d", queue);
4039         return NULL;
4040     }
4041     vq = &vdev->vq[queue];
4042 
4043     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
4044         error_setg(errp, "Packed ring not supported");
4045         return NULL;
4046     } else {
4047         unsigned int head, i, max;
4048         VRingMemoryRegionCaches *caches;
4049         MemoryRegionCache indirect_desc_cache;
4050         MemoryRegionCache *desc_cache;
4051         VRingDesc desc;
4052         VirtioRingDescList *list = NULL;
4053         VirtioRingDescList *node;
4054         int rc; int ndescs;
4055 
4056         address_space_cache_init_empty(&indirect_desc_cache);
4057 
4058         RCU_READ_LOCK_GUARD();
4059 
4060         max = vq->vring.num;
4061 
4062         if (!has_index) {
4063             head = vring_avail_ring(vq, vq->last_avail_idx % vq->vring.num);
4064         } else {
4065             head = vring_avail_ring(vq, index % vq->vring.num);
4066         }
4067         i = head;
4068 
4069         caches = vring_get_region_caches(vq);
4070         if (!caches) {
4071             error_setg(errp, "Region caches not initialized");
4072             return NULL;
4073         }
4074         if (caches->desc.len < max * sizeof(VRingDesc)) {
4075             error_setg(errp, "Cannot map descriptor ring");
4076             return NULL;
4077         }
4078 
4079         desc_cache = &caches->desc;
4080         vring_split_desc_read(vdev, &desc, desc_cache, i);
4081         if (desc.flags & VRING_DESC_F_INDIRECT) {
4082             int64_t len;
4083             len = address_space_cache_init(&indirect_desc_cache, vdev->dma_as,
4084                                            desc.addr, desc.len, false);
4085             desc_cache = &indirect_desc_cache;
4086             if (len < desc.len) {
4087                 error_setg(errp, "Cannot map indirect buffer");
4088                 goto done;
4089             }
4090 
4091             max = desc.len / sizeof(VRingDesc);
4092             i = 0;
4093             vring_split_desc_read(vdev, &desc, desc_cache, i);
4094         }
4095 
4096         element = g_new0(VirtioQueueElement, 1);
4097         element->avail = g_new0(VirtioRingAvail, 1);
4098         element->used = g_new0(VirtioRingUsed, 1);
4099         element->name = g_strdup(vdev->name);
4100         element->index = head;
4101         element->avail->flags = vring_avail_flags(vq);
4102         element->avail->idx = vring_avail_idx(vq);
4103         element->avail->ring = head;
4104         element->used->flags = vring_used_flags(vq);
4105         element->used->idx = vring_used_idx(vq);
4106         ndescs = 0;
4107 
4108         do {
4109             /* A buggy driver may produce an infinite loop */
4110             if (ndescs >= max) {
4111                 break;
4112             }
4113             node = g_new0(VirtioRingDescList, 1);
4114             node->value = g_new0(VirtioRingDesc, 1);
4115             node->value->addr = desc.addr;
4116             node->value->len = desc.len;
4117             node->value->flags = qmp_decode_vring_desc_flags(desc.flags);
4118             node->next = list;
4119             list = node;
4120 
4121             ndescs++;
4122             rc = virtqueue_split_read_next_desc(vdev, &desc, desc_cache, max);
4123         } while (rc == VIRTQUEUE_READ_DESC_MORE);
4124         element->descs = list;
4125 done:
4126         address_space_cache_destroy(&indirect_desc_cache);
4127     }
4128 
4129     return element;
4130 }
4131 
4132 static const TypeInfo virtio_device_info = {
4133     .name = TYPE_VIRTIO_DEVICE,
4134     .parent = TYPE_DEVICE,
4135     .instance_size = sizeof(VirtIODevice),
4136     .class_init = virtio_device_class_init,
4137     .instance_finalize = virtio_device_instance_finalize,
4138     .abstract = true,
4139     .class_size = sizeof(VirtioDeviceClass),
4140 };
4141 
4142 static void virtio_register_types(void)
4143 {
4144     type_register_static(&virtio_device_info);
4145 }
4146 
4147 type_init(virtio_register_types)
4148 
4149 QEMUBH *virtio_bh_new_guarded_full(DeviceState *dev,
4150                                    QEMUBHFunc *cb, void *opaque,
4151                                    const char *name)
4152 {
4153     DeviceState *transport = qdev_get_parent_bus(dev)->parent;
4154 
4155     return qemu_bh_new_full(cb, opaque, name,
4156                             &transport->mem_reentrancy_guard);
4157 }
4158