xref: /openbmc/qemu/hw/virtio/virtio.c (revision 136cb9cc)
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/error-report.h"
19 #include "qemu/log.h"
20 #include "qemu/main-loop.h"
21 #include "qemu/module.h"
22 #include "qom/object_interfaces.h"
23 #include "hw/core/cpu.h"
24 #include "hw/virtio/virtio.h"
25 #include "hw/virtio/vhost.h"
26 #include "migration/qemu-file-types.h"
27 #include "qemu/atomic.h"
28 #include "hw/virtio/virtio-bus.h"
29 #include "hw/qdev-properties.h"
30 #include "hw/virtio/virtio-access.h"
31 #include "sysemu/dma.h"
32 #include "sysemu/runstate.h"
33 #include "virtio-qmp.h"
34 
35 #include "standard-headers/linux/virtio_ids.h"
36 #include "standard-headers/linux/vhost_types.h"
37 #include "standard-headers/linux/virtio_blk.h"
38 #include "standard-headers/linux/virtio_console.h"
39 #include "standard-headers/linux/virtio_gpu.h"
40 #include "standard-headers/linux/virtio_net.h"
41 #include "standard-headers/linux/virtio_scsi.h"
42 #include "standard-headers/linux/virtio_i2c.h"
43 #include "standard-headers/linux/virtio_balloon.h"
44 #include "standard-headers/linux/virtio_iommu.h"
45 #include "standard-headers/linux/virtio_mem.h"
46 #include "standard-headers/linux/virtio_vsock.h"
47 
48 QmpVirtIODeviceList virtio_list;
49 
50 /*
51  * Maximum size of virtio device config space
52  */
53 #define VHOST_USER_MAX_CONFIG_SIZE 256
54 
55 /*
56  * The alignment to use between consumer and producer parts of vring.
57  * x86 pagesize again. This is the default, used by transports like PCI
58  * which don't provide a means for the guest to tell the host the alignment.
59  */
60 #define VIRTIO_PCI_VRING_ALIGN         4096
61 
62 typedef struct VRingDesc
63 {
64     uint64_t addr;
65     uint32_t len;
66     uint16_t flags;
67     uint16_t next;
68 } VRingDesc;
69 
70 typedef struct VRingPackedDesc {
71     uint64_t addr;
72     uint32_t len;
73     uint16_t id;
74     uint16_t flags;
75 } VRingPackedDesc;
76 
77 typedef struct VRingAvail
78 {
79     uint16_t flags;
80     uint16_t idx;
81     uint16_t ring[];
82 } VRingAvail;
83 
84 typedef struct VRingUsedElem
85 {
86     uint32_t id;
87     uint32_t len;
88 } VRingUsedElem;
89 
90 typedef struct VRingUsed
91 {
92     uint16_t flags;
93     uint16_t idx;
94     VRingUsedElem ring[];
95 } VRingUsed;
96 
97 typedef struct VRingMemoryRegionCaches {
98     struct rcu_head rcu;
99     MemoryRegionCache desc;
100     MemoryRegionCache avail;
101     MemoryRegionCache used;
102 } VRingMemoryRegionCaches;
103 
104 typedef struct VRing
105 {
106     unsigned int num;
107     unsigned int num_default;
108     unsigned int align;
109     hwaddr desc;
110     hwaddr avail;
111     hwaddr used;
112     VRingMemoryRegionCaches *caches;
113 } VRing;
114 
115 typedef struct VRingPackedDescEvent {
116     uint16_t off_wrap;
117     uint16_t flags;
118 } VRingPackedDescEvent ;
119 
120 struct VirtQueue
121 {
122     VRing vring;
123     VirtQueueElement *used_elems;
124 
125     /* Next head to pop */
126     uint16_t last_avail_idx;
127     bool last_avail_wrap_counter;
128 
129     /* Last avail_idx read from VQ. */
130     uint16_t shadow_avail_idx;
131     bool shadow_avail_wrap_counter;
132 
133     uint16_t used_idx;
134     bool used_wrap_counter;
135 
136     /* Last used index value we have signalled on */
137     uint16_t signalled_used;
138 
139     /* Last used index value we have signalled on */
140     bool signalled_used_valid;
141 
142     /* Notification enabled? */
143     bool notification;
144 
145     uint16_t queue_index;
146 
147     unsigned int inuse;
148 
149     uint16_t vector;
150     VirtIOHandleOutput handle_output;
151     VirtIODevice *vdev;
152     EventNotifier guest_notifier;
153     EventNotifier host_notifier;
154     bool host_notifier_enabled;
155     QLIST_ENTRY(VirtQueue) node;
156 };
157 
158 const char *virtio_device_names[] = {
159     [VIRTIO_ID_NET] = "virtio-net",
160     [VIRTIO_ID_BLOCK] = "virtio-blk",
161     [VIRTIO_ID_CONSOLE] = "virtio-serial",
162     [VIRTIO_ID_RNG] = "virtio-rng",
163     [VIRTIO_ID_BALLOON] = "virtio-balloon",
164     [VIRTIO_ID_IOMEM] = "virtio-iomem",
165     [VIRTIO_ID_RPMSG] = "virtio-rpmsg",
166     [VIRTIO_ID_SCSI] = "virtio-scsi",
167     [VIRTIO_ID_9P] = "virtio-9p",
168     [VIRTIO_ID_MAC80211_WLAN] = "virtio-mac-wlan",
169     [VIRTIO_ID_RPROC_SERIAL] = "virtio-rproc-serial",
170     [VIRTIO_ID_CAIF] = "virtio-caif",
171     [VIRTIO_ID_MEMORY_BALLOON] = "virtio-mem-balloon",
172     [VIRTIO_ID_GPU] = "virtio-gpu",
173     [VIRTIO_ID_CLOCK] = "virtio-clk",
174     [VIRTIO_ID_INPUT] = "virtio-input",
175     [VIRTIO_ID_VSOCK] = "vhost-vsock",
176     [VIRTIO_ID_CRYPTO] = "virtio-crypto",
177     [VIRTIO_ID_SIGNAL_DIST] = "virtio-signal",
178     [VIRTIO_ID_PSTORE] = "virtio-pstore",
179     [VIRTIO_ID_IOMMU] = "virtio-iommu",
180     [VIRTIO_ID_MEM] = "virtio-mem",
181     [VIRTIO_ID_SOUND] = "virtio-sound",
182     [VIRTIO_ID_FS] = "virtio-user-fs",
183     [VIRTIO_ID_PMEM] = "virtio-pmem",
184     [VIRTIO_ID_RPMB] = "virtio-rpmb",
185     [VIRTIO_ID_MAC80211_HWSIM] = "virtio-mac-hwsim",
186     [VIRTIO_ID_VIDEO_ENCODER] = "virtio-vid-encoder",
187     [VIRTIO_ID_VIDEO_DECODER] = "virtio-vid-decoder",
188     [VIRTIO_ID_SCMI] = "virtio-scmi",
189     [VIRTIO_ID_NITRO_SEC_MOD] = "virtio-nitro-sec-mod",
190     [VIRTIO_ID_I2C_ADAPTER] = "vhost-user-i2c",
191     [VIRTIO_ID_WATCHDOG] = "virtio-watchdog",
192     [VIRTIO_ID_CAN] = "virtio-can",
193     [VIRTIO_ID_DMABUF] = "virtio-dmabuf",
194     [VIRTIO_ID_PARAM_SERV] = "virtio-param-serv",
195     [VIRTIO_ID_AUDIO_POLICY] = "virtio-audio-pol",
196     [VIRTIO_ID_BT] = "virtio-bluetooth",
197     [VIRTIO_ID_GPIO] = "virtio-gpio"
198 };
199 
200 static const char *virtio_id_to_name(uint16_t device_id)
201 {
202     assert(device_id < G_N_ELEMENTS(virtio_device_names));
203     const char *name = virtio_device_names[device_id];
204     assert(name != NULL);
205     return name;
206 }
207 
208 /* Called within call_rcu().  */
209 static void virtio_free_region_cache(VRingMemoryRegionCaches *caches)
210 {
211     assert(caches != NULL);
212     address_space_cache_destroy(&caches->desc);
213     address_space_cache_destroy(&caches->avail);
214     address_space_cache_destroy(&caches->used);
215     g_free(caches);
216 }
217 
218 static void virtio_virtqueue_reset_region_cache(struct VirtQueue *vq)
219 {
220     VRingMemoryRegionCaches *caches;
221 
222     caches = qatomic_read(&vq->vring.caches);
223     qatomic_rcu_set(&vq->vring.caches, NULL);
224     if (caches) {
225         call_rcu(caches, virtio_free_region_cache, rcu);
226     }
227 }
228 
229 void virtio_init_region_cache(VirtIODevice *vdev, int n)
230 {
231     VirtQueue *vq = &vdev->vq[n];
232     VRingMemoryRegionCaches *old = vq->vring.caches;
233     VRingMemoryRegionCaches *new = NULL;
234     hwaddr addr, size;
235     int64_t len;
236     bool packed;
237 
238 
239     addr = vq->vring.desc;
240     if (!addr) {
241         goto out_no_cache;
242     }
243     new = g_new0(VRingMemoryRegionCaches, 1);
244     size = virtio_queue_get_desc_size(vdev, n);
245     packed = virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED) ?
246                                    true : false;
247     len = address_space_cache_init(&new->desc, vdev->dma_as,
248                                    addr, size, packed);
249     if (len < size) {
250         virtio_error(vdev, "Cannot map desc");
251         goto err_desc;
252     }
253 
254     size = virtio_queue_get_used_size(vdev, n);
255     len = address_space_cache_init(&new->used, vdev->dma_as,
256                                    vq->vring.used, size, true);
257     if (len < size) {
258         virtio_error(vdev, "Cannot map used");
259         goto err_used;
260     }
261 
262     size = virtio_queue_get_avail_size(vdev, n);
263     len = address_space_cache_init(&new->avail, vdev->dma_as,
264                                    vq->vring.avail, size, false);
265     if (len < size) {
266         virtio_error(vdev, "Cannot map avail");
267         goto err_avail;
268     }
269 
270     qatomic_rcu_set(&vq->vring.caches, new);
271     if (old) {
272         call_rcu(old, virtio_free_region_cache, rcu);
273     }
274     return;
275 
276 err_avail:
277     address_space_cache_destroy(&new->avail);
278 err_used:
279     address_space_cache_destroy(&new->used);
280 err_desc:
281     address_space_cache_destroy(&new->desc);
282 out_no_cache:
283     g_free(new);
284     virtio_virtqueue_reset_region_cache(vq);
285 }
286 
287 /* virt queue functions */
288 void virtio_queue_update_rings(VirtIODevice *vdev, int n)
289 {
290     VRing *vring = &vdev->vq[n].vring;
291 
292     if (!vring->num || !vring->desc || !vring->align) {
293         /* not yet setup -> nothing to do */
294         return;
295     }
296     vring->avail = vring->desc + vring->num * sizeof(VRingDesc);
297     vring->used = vring_align(vring->avail +
298                               offsetof(VRingAvail, ring[vring->num]),
299                               vring->align);
300     virtio_init_region_cache(vdev, n);
301 }
302 
303 /* Called within rcu_read_lock().  */
304 static void vring_split_desc_read(VirtIODevice *vdev, VRingDesc *desc,
305                                   MemoryRegionCache *cache, int i)
306 {
307     address_space_read_cached(cache, i * sizeof(VRingDesc),
308                               desc, sizeof(VRingDesc));
309     virtio_tswap64s(vdev, &desc->addr);
310     virtio_tswap32s(vdev, &desc->len);
311     virtio_tswap16s(vdev, &desc->flags);
312     virtio_tswap16s(vdev, &desc->next);
313 }
314 
315 static void vring_packed_event_read(VirtIODevice *vdev,
316                                     MemoryRegionCache *cache,
317                                     VRingPackedDescEvent *e)
318 {
319     hwaddr off_off = offsetof(VRingPackedDescEvent, off_wrap);
320     hwaddr off_flags = offsetof(VRingPackedDescEvent, flags);
321 
322     e->flags = virtio_lduw_phys_cached(vdev, cache, off_flags);
323     /* Make sure flags is seen before off_wrap */
324     smp_rmb();
325     e->off_wrap = virtio_lduw_phys_cached(vdev, cache, off_off);
326     virtio_tswap16s(vdev, &e->flags);
327 }
328 
329 static void vring_packed_off_wrap_write(VirtIODevice *vdev,
330                                         MemoryRegionCache *cache,
331                                         uint16_t off_wrap)
332 {
333     hwaddr off = offsetof(VRingPackedDescEvent, off_wrap);
334 
335     virtio_stw_phys_cached(vdev, cache, off, off_wrap);
336     address_space_cache_invalidate(cache, off, sizeof(off_wrap));
337 }
338 
339 static void vring_packed_flags_write(VirtIODevice *vdev,
340                                      MemoryRegionCache *cache, uint16_t flags)
341 {
342     hwaddr off = offsetof(VRingPackedDescEvent, flags);
343 
344     virtio_stw_phys_cached(vdev, cache, off, flags);
345     address_space_cache_invalidate(cache, off, sizeof(flags));
346 }
347 
348 /* Called within rcu_read_lock().  */
349 static VRingMemoryRegionCaches *vring_get_region_caches(struct VirtQueue *vq)
350 {
351     return qatomic_rcu_read(&vq->vring.caches);
352 }
353 
354 /* Called within rcu_read_lock().  */
355 static inline uint16_t vring_avail_flags(VirtQueue *vq)
356 {
357     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
358     hwaddr pa = offsetof(VRingAvail, flags);
359 
360     if (!caches) {
361         return 0;
362     }
363 
364     return virtio_lduw_phys_cached(vq->vdev, &caches->avail, pa);
365 }
366 
367 /* Called within rcu_read_lock().  */
368 static inline uint16_t vring_avail_idx(VirtQueue *vq)
369 {
370     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
371     hwaddr pa = offsetof(VRingAvail, idx);
372 
373     if (!caches) {
374         return 0;
375     }
376 
377     vq->shadow_avail_idx = virtio_lduw_phys_cached(vq->vdev, &caches->avail, pa);
378     return vq->shadow_avail_idx;
379 }
380 
381 /* Called within rcu_read_lock().  */
382 static inline uint16_t vring_avail_ring(VirtQueue *vq, int i)
383 {
384     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
385     hwaddr pa = offsetof(VRingAvail, ring[i]);
386 
387     if (!caches) {
388         return 0;
389     }
390 
391     return virtio_lduw_phys_cached(vq->vdev, &caches->avail, pa);
392 }
393 
394 /* Called within rcu_read_lock().  */
395 static inline uint16_t vring_get_used_event(VirtQueue *vq)
396 {
397     return vring_avail_ring(vq, vq->vring.num);
398 }
399 
400 /* Called within rcu_read_lock().  */
401 static inline void vring_used_write(VirtQueue *vq, VRingUsedElem *uelem,
402                                     int i)
403 {
404     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
405     hwaddr pa = offsetof(VRingUsed, ring[i]);
406 
407     if (!caches) {
408         return;
409     }
410 
411     virtio_tswap32s(vq->vdev, &uelem->id);
412     virtio_tswap32s(vq->vdev, &uelem->len);
413     address_space_write_cached(&caches->used, pa, uelem, sizeof(VRingUsedElem));
414     address_space_cache_invalidate(&caches->used, pa, sizeof(VRingUsedElem));
415 }
416 
417 /* Called within rcu_read_lock(). */
418 static inline uint16_t vring_used_flags(VirtQueue *vq)
419 {
420     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
421     hwaddr pa = offsetof(VRingUsed, flags);
422 
423     if (!caches) {
424         return 0;
425     }
426 
427     return virtio_lduw_phys_cached(vq->vdev, &caches->used, pa);
428 }
429 
430 /* Called within rcu_read_lock().  */
431 static uint16_t vring_used_idx(VirtQueue *vq)
432 {
433     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
434     hwaddr pa = offsetof(VRingUsed, idx);
435 
436     if (!caches) {
437         return 0;
438     }
439 
440     return virtio_lduw_phys_cached(vq->vdev, &caches->used, pa);
441 }
442 
443 /* Called within rcu_read_lock().  */
444 static inline void vring_used_idx_set(VirtQueue *vq, uint16_t val)
445 {
446     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
447     hwaddr pa = offsetof(VRingUsed, idx);
448 
449     if (caches) {
450         virtio_stw_phys_cached(vq->vdev, &caches->used, pa, val);
451         address_space_cache_invalidate(&caches->used, pa, sizeof(val));
452     }
453 
454     vq->used_idx = val;
455 }
456 
457 /* Called within rcu_read_lock().  */
458 static inline void vring_used_flags_set_bit(VirtQueue *vq, int mask)
459 {
460     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
461     VirtIODevice *vdev = vq->vdev;
462     hwaddr pa = offsetof(VRingUsed, flags);
463     uint16_t flags;
464 
465     if (!caches) {
466         return;
467     }
468 
469     flags = virtio_lduw_phys_cached(vq->vdev, &caches->used, pa);
470     virtio_stw_phys_cached(vdev, &caches->used, pa, flags | mask);
471     address_space_cache_invalidate(&caches->used, pa, sizeof(flags));
472 }
473 
474 /* Called within rcu_read_lock().  */
475 static inline void vring_used_flags_unset_bit(VirtQueue *vq, int mask)
476 {
477     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
478     VirtIODevice *vdev = vq->vdev;
479     hwaddr pa = offsetof(VRingUsed, flags);
480     uint16_t flags;
481 
482     if (!caches) {
483         return;
484     }
485 
486     flags = virtio_lduw_phys_cached(vq->vdev, &caches->used, pa);
487     virtio_stw_phys_cached(vdev, &caches->used, pa, flags & ~mask);
488     address_space_cache_invalidate(&caches->used, pa, sizeof(flags));
489 }
490 
491 /* Called within rcu_read_lock().  */
492 static inline void vring_set_avail_event(VirtQueue *vq, uint16_t val)
493 {
494     VRingMemoryRegionCaches *caches;
495     hwaddr pa;
496     if (!vq->notification) {
497         return;
498     }
499 
500     caches = vring_get_region_caches(vq);
501     if (!caches) {
502         return;
503     }
504 
505     pa = offsetof(VRingUsed, ring[vq->vring.num]);
506     virtio_stw_phys_cached(vq->vdev, &caches->used, pa, val);
507     address_space_cache_invalidate(&caches->used, pa, sizeof(val));
508 }
509 
510 static void virtio_queue_split_set_notification(VirtQueue *vq, int enable)
511 {
512     RCU_READ_LOCK_GUARD();
513 
514     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_RING_F_EVENT_IDX)) {
515         vring_set_avail_event(vq, vring_avail_idx(vq));
516     } else if (enable) {
517         vring_used_flags_unset_bit(vq, VRING_USED_F_NO_NOTIFY);
518     } else {
519         vring_used_flags_set_bit(vq, VRING_USED_F_NO_NOTIFY);
520     }
521     if (enable) {
522         /* Expose avail event/used flags before caller checks the avail idx. */
523         smp_mb();
524     }
525 }
526 
527 static void virtio_queue_packed_set_notification(VirtQueue *vq, int enable)
528 {
529     uint16_t off_wrap;
530     VRingPackedDescEvent e;
531     VRingMemoryRegionCaches *caches;
532 
533     RCU_READ_LOCK_GUARD();
534     caches = vring_get_region_caches(vq);
535     if (!caches) {
536         return;
537     }
538 
539     vring_packed_event_read(vq->vdev, &caches->used, &e);
540 
541     if (!enable) {
542         e.flags = VRING_PACKED_EVENT_FLAG_DISABLE;
543     } else if (virtio_vdev_has_feature(vq->vdev, VIRTIO_RING_F_EVENT_IDX)) {
544         off_wrap = vq->shadow_avail_idx | vq->shadow_avail_wrap_counter << 15;
545         vring_packed_off_wrap_write(vq->vdev, &caches->used, off_wrap);
546         /* Make sure off_wrap is wrote before flags */
547         smp_wmb();
548         e.flags = VRING_PACKED_EVENT_FLAG_DESC;
549     } else {
550         e.flags = VRING_PACKED_EVENT_FLAG_ENABLE;
551     }
552 
553     vring_packed_flags_write(vq->vdev, &caches->used, e.flags);
554     if (enable) {
555         /* Expose avail event/used flags before caller checks the avail idx. */
556         smp_mb();
557     }
558 }
559 
560 bool virtio_queue_get_notification(VirtQueue *vq)
561 {
562     return vq->notification;
563 }
564 
565 void virtio_queue_set_notification(VirtQueue *vq, int enable)
566 {
567     vq->notification = enable;
568 
569     if (!vq->vring.desc) {
570         return;
571     }
572 
573     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) {
574         virtio_queue_packed_set_notification(vq, enable);
575     } else {
576         virtio_queue_split_set_notification(vq, enable);
577     }
578 }
579 
580 int virtio_queue_ready(VirtQueue *vq)
581 {
582     return vq->vring.avail != 0;
583 }
584 
585 static void vring_packed_desc_read_flags(VirtIODevice *vdev,
586                                          uint16_t *flags,
587                                          MemoryRegionCache *cache,
588                                          int i)
589 {
590     hwaddr off = i * sizeof(VRingPackedDesc) + offsetof(VRingPackedDesc, flags);
591 
592     *flags = virtio_lduw_phys_cached(vdev, cache, off);
593 }
594 
595 static void vring_packed_desc_read(VirtIODevice *vdev,
596                                    VRingPackedDesc *desc,
597                                    MemoryRegionCache *cache,
598                                    int i, bool strict_order)
599 {
600     hwaddr off = i * sizeof(VRingPackedDesc);
601 
602     vring_packed_desc_read_flags(vdev, &desc->flags, cache, i);
603 
604     if (strict_order) {
605         /* Make sure flags is read before the rest fields. */
606         smp_rmb();
607     }
608 
609     address_space_read_cached(cache, off + offsetof(VRingPackedDesc, addr),
610                               &desc->addr, sizeof(desc->addr));
611     address_space_read_cached(cache, off + offsetof(VRingPackedDesc, id),
612                               &desc->id, sizeof(desc->id));
613     address_space_read_cached(cache, off + offsetof(VRingPackedDesc, len),
614                               &desc->len, sizeof(desc->len));
615     virtio_tswap64s(vdev, &desc->addr);
616     virtio_tswap16s(vdev, &desc->id);
617     virtio_tswap32s(vdev, &desc->len);
618 }
619 
620 static void vring_packed_desc_write_data(VirtIODevice *vdev,
621                                          VRingPackedDesc *desc,
622                                          MemoryRegionCache *cache,
623                                          int i)
624 {
625     hwaddr off_id = i * sizeof(VRingPackedDesc) +
626                     offsetof(VRingPackedDesc, id);
627     hwaddr off_len = i * sizeof(VRingPackedDesc) +
628                     offsetof(VRingPackedDesc, len);
629 
630     virtio_tswap32s(vdev, &desc->len);
631     virtio_tswap16s(vdev, &desc->id);
632     address_space_write_cached(cache, off_id, &desc->id, sizeof(desc->id));
633     address_space_cache_invalidate(cache, off_id, sizeof(desc->id));
634     address_space_write_cached(cache, off_len, &desc->len, sizeof(desc->len));
635     address_space_cache_invalidate(cache, off_len, sizeof(desc->len));
636 }
637 
638 static void vring_packed_desc_write_flags(VirtIODevice *vdev,
639                                           VRingPackedDesc *desc,
640                                           MemoryRegionCache *cache,
641                                           int i)
642 {
643     hwaddr off = i * sizeof(VRingPackedDesc) + offsetof(VRingPackedDesc, flags);
644 
645     virtio_stw_phys_cached(vdev, cache, off, desc->flags);
646     address_space_cache_invalidate(cache, off, sizeof(desc->flags));
647 }
648 
649 static void vring_packed_desc_write(VirtIODevice *vdev,
650                                     VRingPackedDesc *desc,
651                                     MemoryRegionCache *cache,
652                                     int i, bool strict_order)
653 {
654     vring_packed_desc_write_data(vdev, desc, cache, i);
655     if (strict_order) {
656         /* Make sure data is wrote before flags. */
657         smp_wmb();
658     }
659     vring_packed_desc_write_flags(vdev, desc, cache, i);
660 }
661 
662 static inline bool is_desc_avail(uint16_t flags, bool wrap_counter)
663 {
664     bool avail, used;
665 
666     avail = !!(flags & (1 << VRING_PACKED_DESC_F_AVAIL));
667     used = !!(flags & (1 << VRING_PACKED_DESC_F_USED));
668     return (avail != used) && (avail == wrap_counter);
669 }
670 
671 /* Fetch avail_idx from VQ memory only when we really need to know if
672  * guest has added some buffers.
673  * Called within rcu_read_lock().  */
674 static int virtio_queue_empty_rcu(VirtQueue *vq)
675 {
676     if (virtio_device_disabled(vq->vdev)) {
677         return 1;
678     }
679 
680     if (unlikely(!vq->vring.avail)) {
681         return 1;
682     }
683 
684     if (vq->shadow_avail_idx != vq->last_avail_idx) {
685         return 0;
686     }
687 
688     return vring_avail_idx(vq) == vq->last_avail_idx;
689 }
690 
691 static int virtio_queue_split_empty(VirtQueue *vq)
692 {
693     bool empty;
694 
695     if (virtio_device_disabled(vq->vdev)) {
696         return 1;
697     }
698 
699     if (unlikely(!vq->vring.avail)) {
700         return 1;
701     }
702 
703     if (vq->shadow_avail_idx != vq->last_avail_idx) {
704         return 0;
705     }
706 
707     RCU_READ_LOCK_GUARD();
708     empty = vring_avail_idx(vq) == vq->last_avail_idx;
709     return empty;
710 }
711 
712 /* Called within rcu_read_lock().  */
713 static int virtio_queue_packed_empty_rcu(VirtQueue *vq)
714 {
715     struct VRingPackedDesc desc;
716     VRingMemoryRegionCaches *cache;
717 
718     if (unlikely(!vq->vring.desc)) {
719         return 1;
720     }
721 
722     cache = vring_get_region_caches(vq);
723     if (!cache) {
724         return 1;
725     }
726 
727     vring_packed_desc_read_flags(vq->vdev, &desc.flags, &cache->desc,
728                                  vq->last_avail_idx);
729 
730     return !is_desc_avail(desc.flags, vq->last_avail_wrap_counter);
731 }
732 
733 static int virtio_queue_packed_empty(VirtQueue *vq)
734 {
735     RCU_READ_LOCK_GUARD();
736     return virtio_queue_packed_empty_rcu(vq);
737 }
738 
739 int virtio_queue_empty(VirtQueue *vq)
740 {
741     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) {
742         return virtio_queue_packed_empty(vq);
743     } else {
744         return virtio_queue_split_empty(vq);
745     }
746 }
747 
748 static void virtqueue_unmap_sg(VirtQueue *vq, const VirtQueueElement *elem,
749                                unsigned int len)
750 {
751     AddressSpace *dma_as = vq->vdev->dma_as;
752     unsigned int offset;
753     int i;
754 
755     offset = 0;
756     for (i = 0; i < elem->in_num; i++) {
757         size_t size = MIN(len - offset, elem->in_sg[i].iov_len);
758 
759         dma_memory_unmap(dma_as, elem->in_sg[i].iov_base,
760                          elem->in_sg[i].iov_len,
761                          DMA_DIRECTION_FROM_DEVICE, size);
762 
763         offset += size;
764     }
765 
766     for (i = 0; i < elem->out_num; i++)
767         dma_memory_unmap(dma_as, elem->out_sg[i].iov_base,
768                          elem->out_sg[i].iov_len,
769                          DMA_DIRECTION_TO_DEVICE,
770                          elem->out_sg[i].iov_len);
771 }
772 
773 /* virtqueue_detach_element:
774  * @vq: The #VirtQueue
775  * @elem: The #VirtQueueElement
776  * @len: number of bytes written
777  *
778  * Detach the element from the virtqueue.  This function is suitable for device
779  * reset or other situations where a #VirtQueueElement is simply freed and will
780  * not be pushed or discarded.
781  */
782 void virtqueue_detach_element(VirtQueue *vq, const VirtQueueElement *elem,
783                               unsigned int len)
784 {
785     vq->inuse -= elem->ndescs;
786     virtqueue_unmap_sg(vq, elem, len);
787 }
788 
789 static void virtqueue_split_rewind(VirtQueue *vq, unsigned int num)
790 {
791     vq->last_avail_idx -= num;
792 }
793 
794 static void virtqueue_packed_rewind(VirtQueue *vq, unsigned int num)
795 {
796     if (vq->last_avail_idx < num) {
797         vq->last_avail_idx = vq->vring.num + vq->last_avail_idx - num;
798         vq->last_avail_wrap_counter ^= 1;
799     } else {
800         vq->last_avail_idx -= num;
801     }
802 }
803 
804 /* virtqueue_unpop:
805  * @vq: The #VirtQueue
806  * @elem: The #VirtQueueElement
807  * @len: number of bytes written
808  *
809  * Pretend the most recent element wasn't popped from the virtqueue.  The next
810  * call to virtqueue_pop() will refetch the element.
811  */
812 void virtqueue_unpop(VirtQueue *vq, const VirtQueueElement *elem,
813                      unsigned int len)
814 {
815 
816     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) {
817         virtqueue_packed_rewind(vq, 1);
818     } else {
819         virtqueue_split_rewind(vq, 1);
820     }
821 
822     virtqueue_detach_element(vq, elem, len);
823 }
824 
825 /* virtqueue_rewind:
826  * @vq: The #VirtQueue
827  * @num: Number of elements to push back
828  *
829  * Pretend that elements weren't popped from the virtqueue.  The next
830  * virtqueue_pop() will refetch the oldest element.
831  *
832  * Use virtqueue_unpop() instead if you have a VirtQueueElement.
833  *
834  * Returns: true on success, false if @num is greater than the number of in use
835  * elements.
836  */
837 bool virtqueue_rewind(VirtQueue *vq, unsigned int num)
838 {
839     if (num > vq->inuse) {
840         return false;
841     }
842 
843     vq->inuse -= num;
844     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) {
845         virtqueue_packed_rewind(vq, num);
846     } else {
847         virtqueue_split_rewind(vq, num);
848     }
849     return true;
850 }
851 
852 static void virtqueue_split_fill(VirtQueue *vq, const VirtQueueElement *elem,
853                     unsigned int len, unsigned int idx)
854 {
855     VRingUsedElem uelem;
856 
857     if (unlikely(!vq->vring.used)) {
858         return;
859     }
860 
861     idx = (idx + vq->used_idx) % vq->vring.num;
862 
863     uelem.id = elem->index;
864     uelem.len = len;
865     vring_used_write(vq, &uelem, idx);
866 }
867 
868 static void virtqueue_packed_fill(VirtQueue *vq, const VirtQueueElement *elem,
869                                   unsigned int len, unsigned int idx)
870 {
871     vq->used_elems[idx].index = elem->index;
872     vq->used_elems[idx].len = len;
873     vq->used_elems[idx].ndescs = elem->ndescs;
874 }
875 
876 static void virtqueue_packed_fill_desc(VirtQueue *vq,
877                                        const VirtQueueElement *elem,
878                                        unsigned int idx,
879                                        bool strict_order)
880 {
881     uint16_t head;
882     VRingMemoryRegionCaches *caches;
883     VRingPackedDesc desc = {
884         .id = elem->index,
885         .len = elem->len,
886     };
887     bool wrap_counter = vq->used_wrap_counter;
888 
889     if (unlikely(!vq->vring.desc)) {
890         return;
891     }
892 
893     head = vq->used_idx + idx;
894     if (head >= vq->vring.num) {
895         head -= vq->vring.num;
896         wrap_counter ^= 1;
897     }
898     if (wrap_counter) {
899         desc.flags |= (1 << VRING_PACKED_DESC_F_AVAIL);
900         desc.flags |= (1 << VRING_PACKED_DESC_F_USED);
901     } else {
902         desc.flags &= ~(1 << VRING_PACKED_DESC_F_AVAIL);
903         desc.flags &= ~(1 << VRING_PACKED_DESC_F_USED);
904     }
905 
906     caches = vring_get_region_caches(vq);
907     if (!caches) {
908         return;
909     }
910 
911     vring_packed_desc_write(vq->vdev, &desc, &caches->desc, head, strict_order);
912 }
913 
914 /* Called within rcu_read_lock().  */
915 void virtqueue_fill(VirtQueue *vq, const VirtQueueElement *elem,
916                     unsigned int len, unsigned int idx)
917 {
918     trace_virtqueue_fill(vq, elem, len, idx);
919 
920     virtqueue_unmap_sg(vq, elem, len);
921 
922     if (virtio_device_disabled(vq->vdev)) {
923         return;
924     }
925 
926     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) {
927         virtqueue_packed_fill(vq, elem, len, idx);
928     } else {
929         virtqueue_split_fill(vq, elem, len, idx);
930     }
931 }
932 
933 /* Called within rcu_read_lock().  */
934 static void virtqueue_split_flush(VirtQueue *vq, unsigned int count)
935 {
936     uint16_t old, new;
937 
938     if (unlikely(!vq->vring.used)) {
939         return;
940     }
941 
942     /* Make sure buffer is written before we update index. */
943     smp_wmb();
944     trace_virtqueue_flush(vq, count);
945     old = vq->used_idx;
946     new = old + count;
947     vring_used_idx_set(vq, new);
948     vq->inuse -= count;
949     if (unlikely((int16_t)(new - vq->signalled_used) < (uint16_t)(new - old)))
950         vq->signalled_used_valid = false;
951 }
952 
953 static void virtqueue_packed_flush(VirtQueue *vq, unsigned int count)
954 {
955     unsigned int i, ndescs = 0;
956 
957     if (unlikely(!vq->vring.desc)) {
958         return;
959     }
960 
961     for (i = 1; i < count; i++) {
962         virtqueue_packed_fill_desc(vq, &vq->used_elems[i], i, false);
963         ndescs += vq->used_elems[i].ndescs;
964     }
965     virtqueue_packed_fill_desc(vq, &vq->used_elems[0], 0, true);
966     ndescs += vq->used_elems[0].ndescs;
967 
968     vq->inuse -= ndescs;
969     vq->used_idx += ndescs;
970     if (vq->used_idx >= vq->vring.num) {
971         vq->used_idx -= vq->vring.num;
972         vq->used_wrap_counter ^= 1;
973         vq->signalled_used_valid = false;
974     }
975 }
976 
977 void virtqueue_flush(VirtQueue *vq, unsigned int count)
978 {
979     if (virtio_device_disabled(vq->vdev)) {
980         vq->inuse -= count;
981         return;
982     }
983 
984     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) {
985         virtqueue_packed_flush(vq, count);
986     } else {
987         virtqueue_split_flush(vq, count);
988     }
989 }
990 
991 void virtqueue_push(VirtQueue *vq, const VirtQueueElement *elem,
992                     unsigned int len)
993 {
994     RCU_READ_LOCK_GUARD();
995     virtqueue_fill(vq, elem, len, 0);
996     virtqueue_flush(vq, 1);
997 }
998 
999 /* Called within rcu_read_lock().  */
1000 static int virtqueue_num_heads(VirtQueue *vq, unsigned int idx)
1001 {
1002     uint16_t num_heads = vring_avail_idx(vq) - idx;
1003 
1004     /* Check it isn't doing very strange things with descriptor numbers. */
1005     if (num_heads > vq->vring.num) {
1006         virtio_error(vq->vdev, "Guest moved used index from %u to %u",
1007                      idx, vq->shadow_avail_idx);
1008         return -EINVAL;
1009     }
1010     /* On success, callers read a descriptor at vq->last_avail_idx.
1011      * Make sure descriptor read does not bypass avail index read. */
1012     if (num_heads) {
1013         smp_rmb();
1014     }
1015 
1016     return num_heads;
1017 }
1018 
1019 /* Called within rcu_read_lock().  */
1020 static bool virtqueue_get_head(VirtQueue *vq, unsigned int idx,
1021                                unsigned int *head)
1022 {
1023     /* Grab the next descriptor number they're advertising, and increment
1024      * the index we've seen. */
1025     *head = vring_avail_ring(vq, idx % vq->vring.num);
1026 
1027     /* If their number is silly, that's a fatal mistake. */
1028     if (*head >= vq->vring.num) {
1029         virtio_error(vq->vdev, "Guest says index %u is available", *head);
1030         return false;
1031     }
1032 
1033     return true;
1034 }
1035 
1036 enum {
1037     VIRTQUEUE_READ_DESC_ERROR = -1,
1038     VIRTQUEUE_READ_DESC_DONE = 0,   /* end of chain */
1039     VIRTQUEUE_READ_DESC_MORE = 1,   /* more buffers in chain */
1040 };
1041 
1042 static int virtqueue_split_read_next_desc(VirtIODevice *vdev, VRingDesc *desc,
1043                                           MemoryRegionCache *desc_cache,
1044                                           unsigned int max, unsigned int *next)
1045 {
1046     /* If this descriptor says it doesn't chain, we're done. */
1047     if (!(desc->flags & VRING_DESC_F_NEXT)) {
1048         return VIRTQUEUE_READ_DESC_DONE;
1049     }
1050 
1051     /* Check they're not leading us off end of descriptors. */
1052     *next = desc->next;
1053     /* Make sure compiler knows to grab that: we don't want it changing! */
1054     smp_wmb();
1055 
1056     if (*next >= max) {
1057         virtio_error(vdev, "Desc next is %u", *next);
1058         return VIRTQUEUE_READ_DESC_ERROR;
1059     }
1060 
1061     vring_split_desc_read(vdev, desc, desc_cache, *next);
1062     return VIRTQUEUE_READ_DESC_MORE;
1063 }
1064 
1065 /* Called within rcu_read_lock().  */
1066 static void virtqueue_split_get_avail_bytes(VirtQueue *vq,
1067                             unsigned int *in_bytes, unsigned int *out_bytes,
1068                             unsigned max_in_bytes, unsigned max_out_bytes,
1069                             VRingMemoryRegionCaches *caches)
1070 {
1071     VirtIODevice *vdev = vq->vdev;
1072     unsigned int idx;
1073     unsigned int total_bufs, in_total, out_total;
1074     MemoryRegionCache indirect_desc_cache = MEMORY_REGION_CACHE_INVALID;
1075     int64_t len = 0;
1076     int rc;
1077 
1078     idx = vq->last_avail_idx;
1079     total_bufs = in_total = out_total = 0;
1080 
1081     while ((rc = virtqueue_num_heads(vq, idx)) > 0) {
1082         MemoryRegionCache *desc_cache = &caches->desc;
1083         unsigned int num_bufs;
1084         VRingDesc desc;
1085         unsigned int i;
1086         unsigned int max = vq->vring.num;
1087 
1088         num_bufs = total_bufs;
1089 
1090         if (!virtqueue_get_head(vq, idx++, &i)) {
1091             goto err;
1092         }
1093 
1094         vring_split_desc_read(vdev, &desc, desc_cache, i);
1095 
1096         if (desc.flags & VRING_DESC_F_INDIRECT) {
1097             if (!desc.len || (desc.len % sizeof(VRingDesc))) {
1098                 virtio_error(vdev, "Invalid size for indirect buffer table");
1099                 goto err;
1100             }
1101 
1102             /* If we've got too many, that implies a descriptor loop. */
1103             if (num_bufs >= max) {
1104                 virtio_error(vdev, "Looped descriptor");
1105                 goto err;
1106             }
1107 
1108             /* loop over the indirect descriptor table */
1109             len = address_space_cache_init(&indirect_desc_cache,
1110                                            vdev->dma_as,
1111                                            desc.addr, desc.len, false);
1112             desc_cache = &indirect_desc_cache;
1113             if (len < desc.len) {
1114                 virtio_error(vdev, "Cannot map indirect buffer");
1115                 goto err;
1116             }
1117 
1118             max = desc.len / sizeof(VRingDesc);
1119             num_bufs = i = 0;
1120             vring_split_desc_read(vdev, &desc, desc_cache, i);
1121         }
1122 
1123         do {
1124             /* If we've got too many, that implies a descriptor loop. */
1125             if (++num_bufs > max) {
1126                 virtio_error(vdev, "Looped descriptor");
1127                 goto err;
1128             }
1129 
1130             if (desc.flags & VRING_DESC_F_WRITE) {
1131                 in_total += desc.len;
1132             } else {
1133                 out_total += desc.len;
1134             }
1135             if (in_total >= max_in_bytes && out_total >= max_out_bytes) {
1136                 goto done;
1137             }
1138 
1139             rc = virtqueue_split_read_next_desc(vdev, &desc, desc_cache, max, &i);
1140         } while (rc == VIRTQUEUE_READ_DESC_MORE);
1141 
1142         if (rc == VIRTQUEUE_READ_DESC_ERROR) {
1143             goto err;
1144         }
1145 
1146         if (desc_cache == &indirect_desc_cache) {
1147             address_space_cache_destroy(&indirect_desc_cache);
1148             total_bufs++;
1149         } else {
1150             total_bufs = num_bufs;
1151         }
1152     }
1153 
1154     if (rc < 0) {
1155         goto err;
1156     }
1157 
1158 done:
1159     address_space_cache_destroy(&indirect_desc_cache);
1160     if (in_bytes) {
1161         *in_bytes = in_total;
1162     }
1163     if (out_bytes) {
1164         *out_bytes = out_total;
1165     }
1166     return;
1167 
1168 err:
1169     in_total = out_total = 0;
1170     goto done;
1171 }
1172 
1173 static int virtqueue_packed_read_next_desc(VirtQueue *vq,
1174                                            VRingPackedDesc *desc,
1175                                            MemoryRegionCache
1176                                            *desc_cache,
1177                                            unsigned int max,
1178                                            unsigned int *next,
1179                                            bool indirect)
1180 {
1181     /* If this descriptor says it doesn't chain, we're done. */
1182     if (!indirect && !(desc->flags & VRING_DESC_F_NEXT)) {
1183         return VIRTQUEUE_READ_DESC_DONE;
1184     }
1185 
1186     ++*next;
1187     if (*next == max) {
1188         if (indirect) {
1189             return VIRTQUEUE_READ_DESC_DONE;
1190         } else {
1191             (*next) -= vq->vring.num;
1192         }
1193     }
1194 
1195     vring_packed_desc_read(vq->vdev, desc, desc_cache, *next, false);
1196     return VIRTQUEUE_READ_DESC_MORE;
1197 }
1198 
1199 /* Called within rcu_read_lock().  */
1200 static void virtqueue_packed_get_avail_bytes(VirtQueue *vq,
1201                                              unsigned int *in_bytes,
1202                                              unsigned int *out_bytes,
1203                                              unsigned max_in_bytes,
1204                                              unsigned max_out_bytes,
1205                                              VRingMemoryRegionCaches *caches)
1206 {
1207     VirtIODevice *vdev = vq->vdev;
1208     unsigned int idx;
1209     unsigned int total_bufs, in_total, out_total;
1210     MemoryRegionCache *desc_cache;
1211     MemoryRegionCache indirect_desc_cache = MEMORY_REGION_CACHE_INVALID;
1212     int64_t len = 0;
1213     VRingPackedDesc desc;
1214     bool wrap_counter;
1215 
1216     idx = vq->last_avail_idx;
1217     wrap_counter = vq->last_avail_wrap_counter;
1218     total_bufs = in_total = out_total = 0;
1219 
1220     for (;;) {
1221         unsigned int num_bufs = total_bufs;
1222         unsigned int i = idx;
1223         int rc;
1224         unsigned int max = vq->vring.num;
1225 
1226         desc_cache = &caches->desc;
1227 
1228         vring_packed_desc_read(vdev, &desc, desc_cache, idx, true);
1229         if (!is_desc_avail(desc.flags, wrap_counter)) {
1230             break;
1231         }
1232 
1233         if (desc.flags & VRING_DESC_F_INDIRECT) {
1234             if (desc.len % sizeof(VRingPackedDesc)) {
1235                 virtio_error(vdev, "Invalid size for indirect buffer table");
1236                 goto err;
1237             }
1238 
1239             /* If we've got too many, that implies a descriptor loop. */
1240             if (num_bufs >= max) {
1241                 virtio_error(vdev, "Looped descriptor");
1242                 goto err;
1243             }
1244 
1245             /* loop over the indirect descriptor table */
1246             len = address_space_cache_init(&indirect_desc_cache,
1247                                            vdev->dma_as,
1248                                            desc.addr, desc.len, false);
1249             desc_cache = &indirect_desc_cache;
1250             if (len < desc.len) {
1251                 virtio_error(vdev, "Cannot map indirect buffer");
1252                 goto err;
1253             }
1254 
1255             max = desc.len / sizeof(VRingPackedDesc);
1256             num_bufs = i = 0;
1257             vring_packed_desc_read(vdev, &desc, desc_cache, i, false);
1258         }
1259 
1260         do {
1261             /* If we've got too many, that implies a descriptor loop. */
1262             if (++num_bufs > max) {
1263                 virtio_error(vdev, "Looped descriptor");
1264                 goto err;
1265             }
1266 
1267             if (desc.flags & VRING_DESC_F_WRITE) {
1268                 in_total += desc.len;
1269             } else {
1270                 out_total += desc.len;
1271             }
1272             if (in_total >= max_in_bytes && out_total >= max_out_bytes) {
1273                 goto done;
1274             }
1275 
1276             rc = virtqueue_packed_read_next_desc(vq, &desc, desc_cache, max,
1277                                                  &i, desc_cache ==
1278                                                  &indirect_desc_cache);
1279         } while (rc == VIRTQUEUE_READ_DESC_MORE);
1280 
1281         if (desc_cache == &indirect_desc_cache) {
1282             address_space_cache_destroy(&indirect_desc_cache);
1283             total_bufs++;
1284             idx++;
1285         } else {
1286             idx += num_bufs - total_bufs;
1287             total_bufs = num_bufs;
1288         }
1289 
1290         if (idx >= vq->vring.num) {
1291             idx -= vq->vring.num;
1292             wrap_counter ^= 1;
1293         }
1294     }
1295 
1296     /* Record the index and wrap counter for a kick we want */
1297     vq->shadow_avail_idx = idx;
1298     vq->shadow_avail_wrap_counter = wrap_counter;
1299 done:
1300     address_space_cache_destroy(&indirect_desc_cache);
1301     if (in_bytes) {
1302         *in_bytes = in_total;
1303     }
1304     if (out_bytes) {
1305         *out_bytes = out_total;
1306     }
1307     return;
1308 
1309 err:
1310     in_total = out_total = 0;
1311     goto done;
1312 }
1313 
1314 void virtqueue_get_avail_bytes(VirtQueue *vq, unsigned int *in_bytes,
1315                                unsigned int *out_bytes,
1316                                unsigned max_in_bytes, unsigned max_out_bytes)
1317 {
1318     uint16_t desc_size;
1319     VRingMemoryRegionCaches *caches;
1320 
1321     RCU_READ_LOCK_GUARD();
1322 
1323     if (unlikely(!vq->vring.desc)) {
1324         goto err;
1325     }
1326 
1327     caches = vring_get_region_caches(vq);
1328     if (!caches) {
1329         goto err;
1330     }
1331 
1332     desc_size = virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED) ?
1333                                 sizeof(VRingPackedDesc) : sizeof(VRingDesc);
1334     if (caches->desc.len < vq->vring.num * desc_size) {
1335         virtio_error(vq->vdev, "Cannot map descriptor ring");
1336         goto err;
1337     }
1338 
1339     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) {
1340         virtqueue_packed_get_avail_bytes(vq, in_bytes, out_bytes,
1341                                          max_in_bytes, max_out_bytes,
1342                                          caches);
1343     } else {
1344         virtqueue_split_get_avail_bytes(vq, in_bytes, out_bytes,
1345                                         max_in_bytes, max_out_bytes,
1346                                         caches);
1347     }
1348 
1349     return;
1350 err:
1351     if (in_bytes) {
1352         *in_bytes = 0;
1353     }
1354     if (out_bytes) {
1355         *out_bytes = 0;
1356     }
1357 }
1358 
1359 int virtqueue_avail_bytes(VirtQueue *vq, unsigned int in_bytes,
1360                           unsigned int out_bytes)
1361 {
1362     unsigned int in_total, out_total;
1363 
1364     virtqueue_get_avail_bytes(vq, &in_total, &out_total, in_bytes, out_bytes);
1365     return in_bytes <= in_total && out_bytes <= out_total;
1366 }
1367 
1368 static bool virtqueue_map_desc(VirtIODevice *vdev, unsigned int *p_num_sg,
1369                                hwaddr *addr, struct iovec *iov,
1370                                unsigned int max_num_sg, bool is_write,
1371                                hwaddr pa, size_t sz)
1372 {
1373     bool ok = false;
1374     unsigned num_sg = *p_num_sg;
1375     assert(num_sg <= max_num_sg);
1376 
1377     if (!sz) {
1378         virtio_error(vdev, "virtio: zero sized buffers are not allowed");
1379         goto out;
1380     }
1381 
1382     while (sz) {
1383         hwaddr len = sz;
1384 
1385         if (num_sg == max_num_sg) {
1386             virtio_error(vdev, "virtio: too many write descriptors in "
1387                                "indirect table");
1388             goto out;
1389         }
1390 
1391         iov[num_sg].iov_base = dma_memory_map(vdev->dma_as, pa, &len,
1392                                               is_write ?
1393                                               DMA_DIRECTION_FROM_DEVICE :
1394                                               DMA_DIRECTION_TO_DEVICE,
1395                                               MEMTXATTRS_UNSPECIFIED);
1396         if (!iov[num_sg].iov_base) {
1397             virtio_error(vdev, "virtio: bogus descriptor or out of resources");
1398             goto out;
1399         }
1400 
1401         iov[num_sg].iov_len = len;
1402         addr[num_sg] = pa;
1403 
1404         sz -= len;
1405         pa += len;
1406         num_sg++;
1407     }
1408     ok = true;
1409 
1410 out:
1411     *p_num_sg = num_sg;
1412     return ok;
1413 }
1414 
1415 /* Only used by error code paths before we have a VirtQueueElement (therefore
1416  * virtqueue_unmap_sg() can't be used).  Assumes buffers weren't written to
1417  * yet.
1418  */
1419 static void virtqueue_undo_map_desc(unsigned int out_num, unsigned int in_num,
1420                                     struct iovec *iov)
1421 {
1422     unsigned int i;
1423 
1424     for (i = 0; i < out_num + in_num; i++) {
1425         int is_write = i >= out_num;
1426 
1427         cpu_physical_memory_unmap(iov->iov_base, iov->iov_len, is_write, 0);
1428         iov++;
1429     }
1430 }
1431 
1432 static void virtqueue_map_iovec(VirtIODevice *vdev, struct iovec *sg,
1433                                 hwaddr *addr, unsigned int num_sg,
1434                                 bool is_write)
1435 {
1436     unsigned int i;
1437     hwaddr len;
1438 
1439     for (i = 0; i < num_sg; i++) {
1440         len = sg[i].iov_len;
1441         sg[i].iov_base = dma_memory_map(vdev->dma_as,
1442                                         addr[i], &len, is_write ?
1443                                         DMA_DIRECTION_FROM_DEVICE :
1444                                         DMA_DIRECTION_TO_DEVICE,
1445                                         MEMTXATTRS_UNSPECIFIED);
1446         if (!sg[i].iov_base) {
1447             error_report("virtio: error trying to map MMIO memory");
1448             exit(1);
1449         }
1450         if (len != sg[i].iov_len) {
1451             error_report("virtio: unexpected memory split");
1452             exit(1);
1453         }
1454     }
1455 }
1456 
1457 void virtqueue_map(VirtIODevice *vdev, VirtQueueElement *elem)
1458 {
1459     virtqueue_map_iovec(vdev, elem->in_sg, elem->in_addr, elem->in_num, true);
1460     virtqueue_map_iovec(vdev, elem->out_sg, elem->out_addr, elem->out_num,
1461                                                                         false);
1462 }
1463 
1464 static void *virtqueue_alloc_element(size_t sz, unsigned out_num, unsigned in_num)
1465 {
1466     VirtQueueElement *elem;
1467     size_t in_addr_ofs = QEMU_ALIGN_UP(sz, __alignof__(elem->in_addr[0]));
1468     size_t out_addr_ofs = in_addr_ofs + in_num * sizeof(elem->in_addr[0]);
1469     size_t out_addr_end = out_addr_ofs + out_num * sizeof(elem->out_addr[0]);
1470     size_t in_sg_ofs = QEMU_ALIGN_UP(out_addr_end, __alignof__(elem->in_sg[0]));
1471     size_t out_sg_ofs = in_sg_ofs + in_num * sizeof(elem->in_sg[0]);
1472     size_t out_sg_end = out_sg_ofs + out_num * sizeof(elem->out_sg[0]);
1473 
1474     assert(sz >= sizeof(VirtQueueElement));
1475     elem = g_malloc(out_sg_end);
1476     trace_virtqueue_alloc_element(elem, sz, in_num, out_num);
1477     elem->out_num = out_num;
1478     elem->in_num = in_num;
1479     elem->in_addr = (void *)elem + in_addr_ofs;
1480     elem->out_addr = (void *)elem + out_addr_ofs;
1481     elem->in_sg = (void *)elem + in_sg_ofs;
1482     elem->out_sg = (void *)elem + out_sg_ofs;
1483     return elem;
1484 }
1485 
1486 static void *virtqueue_split_pop(VirtQueue *vq, size_t sz)
1487 {
1488     unsigned int i, head, max;
1489     VRingMemoryRegionCaches *caches;
1490     MemoryRegionCache indirect_desc_cache = MEMORY_REGION_CACHE_INVALID;
1491     MemoryRegionCache *desc_cache;
1492     int64_t len;
1493     VirtIODevice *vdev = vq->vdev;
1494     VirtQueueElement *elem = NULL;
1495     unsigned out_num, in_num, elem_entries;
1496     hwaddr addr[VIRTQUEUE_MAX_SIZE];
1497     struct iovec iov[VIRTQUEUE_MAX_SIZE];
1498     VRingDesc desc;
1499     int rc;
1500 
1501     RCU_READ_LOCK_GUARD();
1502     if (virtio_queue_empty_rcu(vq)) {
1503         goto done;
1504     }
1505     /* Needed after virtio_queue_empty(), see comment in
1506      * virtqueue_num_heads(). */
1507     smp_rmb();
1508 
1509     /* When we start there are none of either input nor output. */
1510     out_num = in_num = elem_entries = 0;
1511 
1512     max = vq->vring.num;
1513 
1514     if (vq->inuse >= vq->vring.num) {
1515         virtio_error(vdev, "Virtqueue size exceeded");
1516         goto done;
1517     }
1518 
1519     if (!virtqueue_get_head(vq, vq->last_avail_idx++, &head)) {
1520         goto done;
1521     }
1522 
1523     if (virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX)) {
1524         vring_set_avail_event(vq, vq->last_avail_idx);
1525     }
1526 
1527     i = head;
1528 
1529     caches = vring_get_region_caches(vq);
1530     if (!caches) {
1531         virtio_error(vdev, "Region caches not initialized");
1532         goto done;
1533     }
1534 
1535     if (caches->desc.len < max * sizeof(VRingDesc)) {
1536         virtio_error(vdev, "Cannot map descriptor ring");
1537         goto done;
1538     }
1539 
1540     desc_cache = &caches->desc;
1541     vring_split_desc_read(vdev, &desc, desc_cache, i);
1542     if (desc.flags & VRING_DESC_F_INDIRECT) {
1543         if (!desc.len || (desc.len % sizeof(VRingDesc))) {
1544             virtio_error(vdev, "Invalid size for indirect buffer table");
1545             goto done;
1546         }
1547 
1548         /* loop over the indirect descriptor table */
1549         len = address_space_cache_init(&indirect_desc_cache, vdev->dma_as,
1550                                        desc.addr, desc.len, false);
1551         desc_cache = &indirect_desc_cache;
1552         if (len < desc.len) {
1553             virtio_error(vdev, "Cannot map indirect buffer");
1554             goto done;
1555         }
1556 
1557         max = desc.len / sizeof(VRingDesc);
1558         i = 0;
1559         vring_split_desc_read(vdev, &desc, desc_cache, i);
1560     }
1561 
1562     /* Collect all the descriptors */
1563     do {
1564         bool map_ok;
1565 
1566         if (desc.flags & VRING_DESC_F_WRITE) {
1567             map_ok = virtqueue_map_desc(vdev, &in_num, addr + out_num,
1568                                         iov + out_num,
1569                                         VIRTQUEUE_MAX_SIZE - out_num, true,
1570                                         desc.addr, desc.len);
1571         } else {
1572             if (in_num) {
1573                 virtio_error(vdev, "Incorrect order for descriptors");
1574                 goto err_undo_map;
1575             }
1576             map_ok = virtqueue_map_desc(vdev, &out_num, addr, iov,
1577                                         VIRTQUEUE_MAX_SIZE, false,
1578                                         desc.addr, desc.len);
1579         }
1580         if (!map_ok) {
1581             goto err_undo_map;
1582         }
1583 
1584         /* If we've got too many, that implies a descriptor loop. */
1585         if (++elem_entries > max) {
1586             virtio_error(vdev, "Looped descriptor");
1587             goto err_undo_map;
1588         }
1589 
1590         rc = virtqueue_split_read_next_desc(vdev, &desc, desc_cache, max, &i);
1591     } while (rc == VIRTQUEUE_READ_DESC_MORE);
1592 
1593     if (rc == VIRTQUEUE_READ_DESC_ERROR) {
1594         goto err_undo_map;
1595     }
1596 
1597     /* Now copy what we have collected and mapped */
1598     elem = virtqueue_alloc_element(sz, out_num, in_num);
1599     elem->index = head;
1600     elem->ndescs = 1;
1601     for (i = 0; i < out_num; i++) {
1602         elem->out_addr[i] = addr[i];
1603         elem->out_sg[i] = iov[i];
1604     }
1605     for (i = 0; i < in_num; i++) {
1606         elem->in_addr[i] = addr[out_num + i];
1607         elem->in_sg[i] = iov[out_num + i];
1608     }
1609 
1610     vq->inuse++;
1611 
1612     trace_virtqueue_pop(vq, elem, elem->in_num, elem->out_num);
1613 done:
1614     address_space_cache_destroy(&indirect_desc_cache);
1615 
1616     return elem;
1617 
1618 err_undo_map:
1619     virtqueue_undo_map_desc(out_num, in_num, iov);
1620     goto done;
1621 }
1622 
1623 static void *virtqueue_packed_pop(VirtQueue *vq, size_t sz)
1624 {
1625     unsigned int i, max;
1626     VRingMemoryRegionCaches *caches;
1627     MemoryRegionCache indirect_desc_cache = MEMORY_REGION_CACHE_INVALID;
1628     MemoryRegionCache *desc_cache;
1629     int64_t len;
1630     VirtIODevice *vdev = vq->vdev;
1631     VirtQueueElement *elem = NULL;
1632     unsigned out_num, in_num, elem_entries;
1633     hwaddr addr[VIRTQUEUE_MAX_SIZE];
1634     struct iovec iov[VIRTQUEUE_MAX_SIZE];
1635     VRingPackedDesc desc;
1636     uint16_t id;
1637     int rc;
1638 
1639     RCU_READ_LOCK_GUARD();
1640     if (virtio_queue_packed_empty_rcu(vq)) {
1641         goto done;
1642     }
1643 
1644     /* When we start there are none of either input nor output. */
1645     out_num = in_num = elem_entries = 0;
1646 
1647     max = vq->vring.num;
1648 
1649     if (vq->inuse >= vq->vring.num) {
1650         virtio_error(vdev, "Virtqueue size exceeded");
1651         goto done;
1652     }
1653 
1654     i = vq->last_avail_idx;
1655 
1656     caches = vring_get_region_caches(vq);
1657     if (!caches) {
1658         virtio_error(vdev, "Region caches not initialized");
1659         goto done;
1660     }
1661 
1662     if (caches->desc.len < max * sizeof(VRingDesc)) {
1663         virtio_error(vdev, "Cannot map descriptor ring");
1664         goto done;
1665     }
1666 
1667     desc_cache = &caches->desc;
1668     vring_packed_desc_read(vdev, &desc, desc_cache, i, true);
1669     id = desc.id;
1670     if (desc.flags & VRING_DESC_F_INDIRECT) {
1671         if (desc.len % sizeof(VRingPackedDesc)) {
1672             virtio_error(vdev, "Invalid size for indirect buffer table");
1673             goto done;
1674         }
1675 
1676         /* loop over the indirect descriptor table */
1677         len = address_space_cache_init(&indirect_desc_cache, vdev->dma_as,
1678                                        desc.addr, desc.len, false);
1679         desc_cache = &indirect_desc_cache;
1680         if (len < desc.len) {
1681             virtio_error(vdev, "Cannot map indirect buffer");
1682             goto done;
1683         }
1684 
1685         max = desc.len / sizeof(VRingPackedDesc);
1686         i = 0;
1687         vring_packed_desc_read(vdev, &desc, desc_cache, i, false);
1688     }
1689 
1690     /* Collect all the descriptors */
1691     do {
1692         bool map_ok;
1693 
1694         if (desc.flags & VRING_DESC_F_WRITE) {
1695             map_ok = virtqueue_map_desc(vdev, &in_num, addr + out_num,
1696                                         iov + out_num,
1697                                         VIRTQUEUE_MAX_SIZE - out_num, true,
1698                                         desc.addr, desc.len);
1699         } else {
1700             if (in_num) {
1701                 virtio_error(vdev, "Incorrect order for descriptors");
1702                 goto err_undo_map;
1703             }
1704             map_ok = virtqueue_map_desc(vdev, &out_num, addr, iov,
1705                                         VIRTQUEUE_MAX_SIZE, false,
1706                                         desc.addr, desc.len);
1707         }
1708         if (!map_ok) {
1709             goto err_undo_map;
1710         }
1711 
1712         /* If we've got too many, that implies a descriptor loop. */
1713         if (++elem_entries > max) {
1714             virtio_error(vdev, "Looped descriptor");
1715             goto err_undo_map;
1716         }
1717 
1718         rc = virtqueue_packed_read_next_desc(vq, &desc, desc_cache, max, &i,
1719                                              desc_cache ==
1720                                              &indirect_desc_cache);
1721     } while (rc == VIRTQUEUE_READ_DESC_MORE);
1722 
1723     /* Now copy what we have collected and mapped */
1724     elem = virtqueue_alloc_element(sz, out_num, in_num);
1725     for (i = 0; i < out_num; i++) {
1726         elem->out_addr[i] = addr[i];
1727         elem->out_sg[i] = iov[i];
1728     }
1729     for (i = 0; i < in_num; i++) {
1730         elem->in_addr[i] = addr[out_num + i];
1731         elem->in_sg[i] = iov[out_num + i];
1732     }
1733 
1734     elem->index = id;
1735     elem->ndescs = (desc_cache == &indirect_desc_cache) ? 1 : elem_entries;
1736     vq->last_avail_idx += elem->ndescs;
1737     vq->inuse += elem->ndescs;
1738 
1739     if (vq->last_avail_idx >= vq->vring.num) {
1740         vq->last_avail_idx -= vq->vring.num;
1741         vq->last_avail_wrap_counter ^= 1;
1742     }
1743 
1744     vq->shadow_avail_idx = vq->last_avail_idx;
1745     vq->shadow_avail_wrap_counter = vq->last_avail_wrap_counter;
1746 
1747     trace_virtqueue_pop(vq, elem, elem->in_num, elem->out_num);
1748 done:
1749     address_space_cache_destroy(&indirect_desc_cache);
1750 
1751     return elem;
1752 
1753 err_undo_map:
1754     virtqueue_undo_map_desc(out_num, in_num, iov);
1755     goto done;
1756 }
1757 
1758 void *virtqueue_pop(VirtQueue *vq, size_t sz)
1759 {
1760     if (virtio_device_disabled(vq->vdev)) {
1761         return NULL;
1762     }
1763 
1764     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) {
1765         return virtqueue_packed_pop(vq, sz);
1766     } else {
1767         return virtqueue_split_pop(vq, sz);
1768     }
1769 }
1770 
1771 static unsigned int virtqueue_packed_drop_all(VirtQueue *vq)
1772 {
1773     VRingMemoryRegionCaches *caches;
1774     MemoryRegionCache *desc_cache;
1775     unsigned int dropped = 0;
1776     VirtQueueElement elem = {};
1777     VirtIODevice *vdev = vq->vdev;
1778     VRingPackedDesc desc;
1779 
1780     RCU_READ_LOCK_GUARD();
1781 
1782     caches = vring_get_region_caches(vq);
1783     if (!caches) {
1784         return 0;
1785     }
1786 
1787     desc_cache = &caches->desc;
1788 
1789     virtio_queue_set_notification(vq, 0);
1790 
1791     while (vq->inuse < vq->vring.num) {
1792         unsigned int idx = vq->last_avail_idx;
1793         /*
1794          * works similar to virtqueue_pop but does not map buffers
1795          * and does not allocate any memory.
1796          */
1797         vring_packed_desc_read(vdev, &desc, desc_cache,
1798                                vq->last_avail_idx , true);
1799         if (!is_desc_avail(desc.flags, vq->last_avail_wrap_counter)) {
1800             break;
1801         }
1802         elem.index = desc.id;
1803         elem.ndescs = 1;
1804         while (virtqueue_packed_read_next_desc(vq, &desc, desc_cache,
1805                                                vq->vring.num, &idx, false)) {
1806             ++elem.ndescs;
1807         }
1808         /*
1809          * immediately push the element, nothing to unmap
1810          * as both in_num and out_num are set to 0.
1811          */
1812         virtqueue_push(vq, &elem, 0);
1813         dropped++;
1814         vq->last_avail_idx += elem.ndescs;
1815         if (vq->last_avail_idx >= vq->vring.num) {
1816             vq->last_avail_idx -= vq->vring.num;
1817             vq->last_avail_wrap_counter ^= 1;
1818         }
1819     }
1820 
1821     return dropped;
1822 }
1823 
1824 static unsigned int virtqueue_split_drop_all(VirtQueue *vq)
1825 {
1826     unsigned int dropped = 0;
1827     VirtQueueElement elem = {};
1828     VirtIODevice *vdev = vq->vdev;
1829     bool fEventIdx = virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX);
1830 
1831     while (!virtio_queue_empty(vq) && vq->inuse < vq->vring.num) {
1832         /* works similar to virtqueue_pop but does not map buffers
1833         * and does not allocate any memory */
1834         smp_rmb();
1835         if (!virtqueue_get_head(vq, vq->last_avail_idx, &elem.index)) {
1836             break;
1837         }
1838         vq->inuse++;
1839         vq->last_avail_idx++;
1840         if (fEventIdx) {
1841             vring_set_avail_event(vq, vq->last_avail_idx);
1842         }
1843         /* immediately push the element, nothing to unmap
1844          * as both in_num and out_num are set to 0 */
1845         virtqueue_push(vq, &elem, 0);
1846         dropped++;
1847     }
1848 
1849     return dropped;
1850 }
1851 
1852 /* virtqueue_drop_all:
1853  * @vq: The #VirtQueue
1854  * Drops all queued buffers and indicates them to the guest
1855  * as if they are done. Useful when buffers can not be
1856  * processed but must be returned to the guest.
1857  */
1858 unsigned int virtqueue_drop_all(VirtQueue *vq)
1859 {
1860     struct VirtIODevice *vdev = vq->vdev;
1861 
1862     if (virtio_device_disabled(vq->vdev)) {
1863         return 0;
1864     }
1865 
1866     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
1867         return virtqueue_packed_drop_all(vq);
1868     } else {
1869         return virtqueue_split_drop_all(vq);
1870     }
1871 }
1872 
1873 /* Reading and writing a structure directly to QEMUFile is *awful*, but
1874  * it is what QEMU has always done by mistake.  We can change it sooner
1875  * or later by bumping the version number of the affected vm states.
1876  * In the meanwhile, since the in-memory layout of VirtQueueElement
1877  * has changed, we need to marshal to and from the layout that was
1878  * used before the change.
1879  */
1880 typedef struct VirtQueueElementOld {
1881     unsigned int index;
1882     unsigned int out_num;
1883     unsigned int in_num;
1884     hwaddr in_addr[VIRTQUEUE_MAX_SIZE];
1885     hwaddr out_addr[VIRTQUEUE_MAX_SIZE];
1886     struct iovec in_sg[VIRTQUEUE_MAX_SIZE];
1887     struct iovec out_sg[VIRTQUEUE_MAX_SIZE];
1888 } VirtQueueElementOld;
1889 
1890 void *qemu_get_virtqueue_element(VirtIODevice *vdev, QEMUFile *f, size_t sz)
1891 {
1892     VirtQueueElement *elem;
1893     VirtQueueElementOld data;
1894     int i;
1895 
1896     qemu_get_buffer(f, (uint8_t *)&data, sizeof(VirtQueueElementOld));
1897 
1898     /* TODO: teach all callers that this can fail, and return failure instead
1899      * of asserting here.
1900      * This is just one thing (there are probably more) that must be
1901      * fixed before we can allow NDEBUG compilation.
1902      */
1903     assert(ARRAY_SIZE(data.in_addr) >= data.in_num);
1904     assert(ARRAY_SIZE(data.out_addr) >= data.out_num);
1905 
1906     elem = virtqueue_alloc_element(sz, data.out_num, data.in_num);
1907     elem->index = data.index;
1908 
1909     for (i = 0; i < elem->in_num; i++) {
1910         elem->in_addr[i] = data.in_addr[i];
1911     }
1912 
1913     for (i = 0; i < elem->out_num; i++) {
1914         elem->out_addr[i] = data.out_addr[i];
1915     }
1916 
1917     for (i = 0; i < elem->in_num; i++) {
1918         /* Base is overwritten by virtqueue_map.  */
1919         elem->in_sg[i].iov_base = 0;
1920         elem->in_sg[i].iov_len = data.in_sg[i].iov_len;
1921     }
1922 
1923     for (i = 0; i < elem->out_num; i++) {
1924         /* Base is overwritten by virtqueue_map.  */
1925         elem->out_sg[i].iov_base = 0;
1926         elem->out_sg[i].iov_len = data.out_sg[i].iov_len;
1927     }
1928 
1929     if (virtio_host_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
1930         qemu_get_be32s(f, &elem->ndescs);
1931     }
1932 
1933     virtqueue_map(vdev, elem);
1934     return elem;
1935 }
1936 
1937 void qemu_put_virtqueue_element(VirtIODevice *vdev, QEMUFile *f,
1938                                 VirtQueueElement *elem)
1939 {
1940     VirtQueueElementOld data;
1941     int i;
1942 
1943     memset(&data, 0, sizeof(data));
1944     data.index = elem->index;
1945     data.in_num = elem->in_num;
1946     data.out_num = elem->out_num;
1947 
1948     for (i = 0; i < elem->in_num; i++) {
1949         data.in_addr[i] = elem->in_addr[i];
1950     }
1951 
1952     for (i = 0; i < elem->out_num; i++) {
1953         data.out_addr[i] = elem->out_addr[i];
1954     }
1955 
1956     for (i = 0; i < elem->in_num; i++) {
1957         /* Base is overwritten by virtqueue_map when loading.  Do not
1958          * save it, as it would leak the QEMU address space layout.  */
1959         data.in_sg[i].iov_len = elem->in_sg[i].iov_len;
1960     }
1961 
1962     for (i = 0; i < elem->out_num; i++) {
1963         /* Do not save iov_base as above.  */
1964         data.out_sg[i].iov_len = elem->out_sg[i].iov_len;
1965     }
1966 
1967     if (virtio_host_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
1968         qemu_put_be32s(f, &elem->ndescs);
1969     }
1970 
1971     qemu_put_buffer(f, (uint8_t *)&data, sizeof(VirtQueueElementOld));
1972 }
1973 
1974 /* virtio device */
1975 static void virtio_notify_vector(VirtIODevice *vdev, uint16_t vector)
1976 {
1977     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
1978     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
1979 
1980     if (virtio_device_disabled(vdev)) {
1981         return;
1982     }
1983 
1984     if (k->notify) {
1985         k->notify(qbus->parent, vector);
1986     }
1987 }
1988 
1989 void virtio_update_irq(VirtIODevice *vdev)
1990 {
1991     virtio_notify_vector(vdev, VIRTIO_NO_VECTOR);
1992 }
1993 
1994 static int virtio_validate_features(VirtIODevice *vdev)
1995 {
1996     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1997 
1998     if (virtio_host_has_feature(vdev, VIRTIO_F_IOMMU_PLATFORM) &&
1999         !virtio_vdev_has_feature(vdev, VIRTIO_F_IOMMU_PLATFORM)) {
2000         return -EFAULT;
2001     }
2002 
2003     if (k->validate_features) {
2004         return k->validate_features(vdev);
2005     } else {
2006         return 0;
2007     }
2008 }
2009 
2010 int virtio_set_status(VirtIODevice *vdev, uint8_t val)
2011 {
2012     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2013     trace_virtio_set_status(vdev, val);
2014 
2015     if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
2016         if (!(vdev->status & VIRTIO_CONFIG_S_FEATURES_OK) &&
2017             val & VIRTIO_CONFIG_S_FEATURES_OK) {
2018             int ret = virtio_validate_features(vdev);
2019 
2020             if (ret) {
2021                 return ret;
2022             }
2023         }
2024     }
2025 
2026     if ((vdev->status & VIRTIO_CONFIG_S_DRIVER_OK) !=
2027         (val & VIRTIO_CONFIG_S_DRIVER_OK)) {
2028         virtio_set_started(vdev, val & VIRTIO_CONFIG_S_DRIVER_OK);
2029     }
2030 
2031     if (k->set_status) {
2032         k->set_status(vdev, val);
2033     }
2034     vdev->status = val;
2035 
2036     return 0;
2037 }
2038 
2039 static enum virtio_device_endian virtio_default_endian(void)
2040 {
2041     if (target_words_bigendian()) {
2042         return VIRTIO_DEVICE_ENDIAN_BIG;
2043     } else {
2044         return VIRTIO_DEVICE_ENDIAN_LITTLE;
2045     }
2046 }
2047 
2048 static enum virtio_device_endian virtio_current_cpu_endian(void)
2049 {
2050     if (cpu_virtio_is_big_endian(current_cpu)) {
2051         return VIRTIO_DEVICE_ENDIAN_BIG;
2052     } else {
2053         return VIRTIO_DEVICE_ENDIAN_LITTLE;
2054     }
2055 }
2056 
2057 static void __virtio_queue_reset(VirtIODevice *vdev, uint32_t i)
2058 {
2059     vdev->vq[i].vring.desc = 0;
2060     vdev->vq[i].vring.avail = 0;
2061     vdev->vq[i].vring.used = 0;
2062     vdev->vq[i].last_avail_idx = 0;
2063     vdev->vq[i].shadow_avail_idx = 0;
2064     vdev->vq[i].used_idx = 0;
2065     vdev->vq[i].last_avail_wrap_counter = true;
2066     vdev->vq[i].shadow_avail_wrap_counter = true;
2067     vdev->vq[i].used_wrap_counter = true;
2068     virtio_queue_set_vector(vdev, i, VIRTIO_NO_VECTOR);
2069     vdev->vq[i].signalled_used = 0;
2070     vdev->vq[i].signalled_used_valid = false;
2071     vdev->vq[i].notification = true;
2072     vdev->vq[i].vring.num = vdev->vq[i].vring.num_default;
2073     vdev->vq[i].inuse = 0;
2074     virtio_virtqueue_reset_region_cache(&vdev->vq[i]);
2075 }
2076 
2077 void virtio_queue_reset(VirtIODevice *vdev, uint32_t queue_index)
2078 {
2079     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2080 
2081     if (k->queue_reset) {
2082         k->queue_reset(vdev, queue_index);
2083     }
2084 
2085     __virtio_queue_reset(vdev, queue_index);
2086 }
2087 
2088 void virtio_queue_enable(VirtIODevice *vdev, uint32_t queue_index)
2089 {
2090     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2091 
2092     /*
2093      * TODO: Seabios is currently out of spec and triggering this error.
2094      * So this needs to be fixed in Seabios, then this can
2095      * be re-enabled for new machine types only, and also after
2096      * being converted to LOG_GUEST_ERROR.
2097      *
2098     if (!virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
2099         error_report("queue_enable is only suppported in devices of virtio "
2100                      "1.0 or later.");
2101     }
2102     */
2103 
2104     if (k->queue_enable) {
2105         k->queue_enable(vdev, queue_index);
2106     }
2107 }
2108 
2109 void virtio_reset(void *opaque)
2110 {
2111     VirtIODevice *vdev = opaque;
2112     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2113     int i;
2114 
2115     virtio_set_status(vdev, 0);
2116     if (current_cpu) {
2117         /* Guest initiated reset */
2118         vdev->device_endian = virtio_current_cpu_endian();
2119     } else {
2120         /* System reset */
2121         vdev->device_endian = virtio_default_endian();
2122     }
2123 
2124     if (k->reset) {
2125         k->reset(vdev);
2126     }
2127 
2128     vdev->start_on_kick = false;
2129     vdev->started = false;
2130     vdev->broken = false;
2131     vdev->guest_features = 0;
2132     vdev->queue_sel = 0;
2133     vdev->status = 0;
2134     vdev->disabled = false;
2135     qatomic_set(&vdev->isr, 0);
2136     vdev->config_vector = VIRTIO_NO_VECTOR;
2137     virtio_notify_vector(vdev, vdev->config_vector);
2138 
2139     for(i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2140         __virtio_queue_reset(vdev, i);
2141     }
2142 }
2143 
2144 void virtio_queue_set_addr(VirtIODevice *vdev, int n, hwaddr addr)
2145 {
2146     if (!vdev->vq[n].vring.num) {
2147         return;
2148     }
2149     vdev->vq[n].vring.desc = addr;
2150     virtio_queue_update_rings(vdev, n);
2151 }
2152 
2153 hwaddr virtio_queue_get_addr(VirtIODevice *vdev, int n)
2154 {
2155     return vdev->vq[n].vring.desc;
2156 }
2157 
2158 void virtio_queue_set_rings(VirtIODevice *vdev, int n, hwaddr desc,
2159                             hwaddr avail, hwaddr used)
2160 {
2161     if (!vdev->vq[n].vring.num) {
2162         return;
2163     }
2164     vdev->vq[n].vring.desc = desc;
2165     vdev->vq[n].vring.avail = avail;
2166     vdev->vq[n].vring.used = used;
2167     virtio_init_region_cache(vdev, n);
2168 }
2169 
2170 void virtio_queue_set_num(VirtIODevice *vdev, int n, int num)
2171 {
2172     /* Don't allow guest to flip queue between existent and
2173      * nonexistent states, or to set it to an invalid size.
2174      */
2175     if (!!num != !!vdev->vq[n].vring.num ||
2176         num > VIRTQUEUE_MAX_SIZE ||
2177         num < 0) {
2178         return;
2179     }
2180     vdev->vq[n].vring.num = num;
2181 }
2182 
2183 VirtQueue *virtio_vector_first_queue(VirtIODevice *vdev, uint16_t vector)
2184 {
2185     return QLIST_FIRST(&vdev->vector_queues[vector]);
2186 }
2187 
2188 VirtQueue *virtio_vector_next_queue(VirtQueue *vq)
2189 {
2190     return QLIST_NEXT(vq, node);
2191 }
2192 
2193 int virtio_queue_get_num(VirtIODevice *vdev, int n)
2194 {
2195     return vdev->vq[n].vring.num;
2196 }
2197 
2198 int virtio_queue_get_max_num(VirtIODevice *vdev, int n)
2199 {
2200     return vdev->vq[n].vring.num_default;
2201 }
2202 
2203 int virtio_get_num_queues(VirtIODevice *vdev)
2204 {
2205     int i;
2206 
2207     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2208         if (!virtio_queue_get_num(vdev, i)) {
2209             break;
2210         }
2211     }
2212 
2213     return i;
2214 }
2215 
2216 void virtio_queue_set_align(VirtIODevice *vdev, int n, int align)
2217 {
2218     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2219     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
2220 
2221     /* virtio-1 compliant devices cannot change the alignment */
2222     if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
2223         error_report("tried to modify queue alignment for virtio-1 device");
2224         return;
2225     }
2226     /* Check that the transport told us it was going to do this
2227      * (so a buggy transport will immediately assert rather than
2228      * silently failing to migrate this state)
2229      */
2230     assert(k->has_variable_vring_alignment);
2231 
2232     if (align) {
2233         vdev->vq[n].vring.align = align;
2234         virtio_queue_update_rings(vdev, n);
2235     }
2236 }
2237 
2238 static void virtio_queue_notify_vq(VirtQueue *vq)
2239 {
2240     if (vq->vring.desc && vq->handle_output) {
2241         VirtIODevice *vdev = vq->vdev;
2242 
2243         if (unlikely(vdev->broken)) {
2244             return;
2245         }
2246 
2247         trace_virtio_queue_notify(vdev, vq - vdev->vq, vq);
2248         vq->handle_output(vdev, vq);
2249 
2250         if (unlikely(vdev->start_on_kick)) {
2251             virtio_set_started(vdev, true);
2252         }
2253     }
2254 }
2255 
2256 void virtio_queue_notify(VirtIODevice *vdev, int n)
2257 {
2258     VirtQueue *vq = &vdev->vq[n];
2259 
2260     if (unlikely(!vq->vring.desc || vdev->broken)) {
2261         return;
2262     }
2263 
2264     trace_virtio_queue_notify(vdev, vq - vdev->vq, vq);
2265     if (vq->host_notifier_enabled) {
2266         event_notifier_set(&vq->host_notifier);
2267     } else if (vq->handle_output) {
2268         vq->handle_output(vdev, vq);
2269 
2270         if (unlikely(vdev->start_on_kick)) {
2271             virtio_set_started(vdev, true);
2272         }
2273     }
2274 }
2275 
2276 uint16_t virtio_queue_vector(VirtIODevice *vdev, int n)
2277 {
2278     return n < VIRTIO_QUEUE_MAX ? vdev->vq[n].vector :
2279         VIRTIO_NO_VECTOR;
2280 }
2281 
2282 void virtio_queue_set_vector(VirtIODevice *vdev, int n, uint16_t vector)
2283 {
2284     VirtQueue *vq = &vdev->vq[n];
2285 
2286     if (n < VIRTIO_QUEUE_MAX) {
2287         if (vdev->vector_queues &&
2288             vdev->vq[n].vector != VIRTIO_NO_VECTOR) {
2289             QLIST_REMOVE(vq, node);
2290         }
2291         vdev->vq[n].vector = vector;
2292         if (vdev->vector_queues &&
2293             vector != VIRTIO_NO_VECTOR) {
2294             QLIST_INSERT_HEAD(&vdev->vector_queues[vector], vq, node);
2295         }
2296     }
2297 }
2298 
2299 VirtQueue *virtio_add_queue(VirtIODevice *vdev, int queue_size,
2300                             VirtIOHandleOutput handle_output)
2301 {
2302     int i;
2303 
2304     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2305         if (vdev->vq[i].vring.num == 0)
2306             break;
2307     }
2308 
2309     if (i == VIRTIO_QUEUE_MAX || queue_size > VIRTQUEUE_MAX_SIZE)
2310         abort();
2311 
2312     vdev->vq[i].vring.num = queue_size;
2313     vdev->vq[i].vring.num_default = queue_size;
2314     vdev->vq[i].vring.align = VIRTIO_PCI_VRING_ALIGN;
2315     vdev->vq[i].handle_output = handle_output;
2316     vdev->vq[i].used_elems = g_new0(VirtQueueElement, queue_size);
2317 
2318     return &vdev->vq[i];
2319 }
2320 
2321 void virtio_delete_queue(VirtQueue *vq)
2322 {
2323     vq->vring.num = 0;
2324     vq->vring.num_default = 0;
2325     vq->handle_output = NULL;
2326     g_free(vq->used_elems);
2327     vq->used_elems = NULL;
2328     virtio_virtqueue_reset_region_cache(vq);
2329 }
2330 
2331 void virtio_del_queue(VirtIODevice *vdev, int n)
2332 {
2333     if (n < 0 || n >= VIRTIO_QUEUE_MAX) {
2334         abort();
2335     }
2336 
2337     virtio_delete_queue(&vdev->vq[n]);
2338 }
2339 
2340 static void virtio_set_isr(VirtIODevice *vdev, int value)
2341 {
2342     uint8_t old = qatomic_read(&vdev->isr);
2343 
2344     /* Do not write ISR if it does not change, so that its cacheline remains
2345      * shared in the common case where the guest does not read it.
2346      */
2347     if ((old & value) != value) {
2348         qatomic_or(&vdev->isr, value);
2349     }
2350 }
2351 
2352 /* Called within rcu_read_lock(). */
2353 static bool virtio_split_should_notify(VirtIODevice *vdev, VirtQueue *vq)
2354 {
2355     uint16_t old, new;
2356     bool v;
2357     /* We need to expose used array entries before checking used event. */
2358     smp_mb();
2359     /* Always notify when queue is empty (when feature acknowledge) */
2360     if (virtio_vdev_has_feature(vdev, VIRTIO_F_NOTIFY_ON_EMPTY) &&
2361         !vq->inuse && virtio_queue_empty(vq)) {
2362         return true;
2363     }
2364 
2365     if (!virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX)) {
2366         return !(vring_avail_flags(vq) & VRING_AVAIL_F_NO_INTERRUPT);
2367     }
2368 
2369     v = vq->signalled_used_valid;
2370     vq->signalled_used_valid = true;
2371     old = vq->signalled_used;
2372     new = vq->signalled_used = vq->used_idx;
2373     return !v || vring_need_event(vring_get_used_event(vq), new, old);
2374 }
2375 
2376 static bool vring_packed_need_event(VirtQueue *vq, bool wrap,
2377                                     uint16_t off_wrap, uint16_t new,
2378                                     uint16_t old)
2379 {
2380     int off = off_wrap & ~(1 << 15);
2381 
2382     if (wrap != off_wrap >> 15) {
2383         off -= vq->vring.num;
2384     }
2385 
2386     return vring_need_event(off, new, old);
2387 }
2388 
2389 /* Called within rcu_read_lock(). */
2390 static bool virtio_packed_should_notify(VirtIODevice *vdev, VirtQueue *vq)
2391 {
2392     VRingPackedDescEvent e;
2393     uint16_t old, new;
2394     bool v;
2395     VRingMemoryRegionCaches *caches;
2396 
2397     caches = vring_get_region_caches(vq);
2398     if (!caches) {
2399         return false;
2400     }
2401 
2402     vring_packed_event_read(vdev, &caches->avail, &e);
2403 
2404     old = vq->signalled_used;
2405     new = vq->signalled_used = vq->used_idx;
2406     v = vq->signalled_used_valid;
2407     vq->signalled_used_valid = true;
2408 
2409     if (e.flags == VRING_PACKED_EVENT_FLAG_DISABLE) {
2410         return false;
2411     } else if (e.flags == VRING_PACKED_EVENT_FLAG_ENABLE) {
2412         return true;
2413     }
2414 
2415     return !v || vring_packed_need_event(vq, vq->used_wrap_counter,
2416                                          e.off_wrap, new, old);
2417 }
2418 
2419 /* Called within rcu_read_lock().  */
2420 static bool virtio_should_notify(VirtIODevice *vdev, VirtQueue *vq)
2421 {
2422     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
2423         return virtio_packed_should_notify(vdev, vq);
2424     } else {
2425         return virtio_split_should_notify(vdev, vq);
2426     }
2427 }
2428 
2429 void virtio_notify_irqfd(VirtIODevice *vdev, VirtQueue *vq)
2430 {
2431     WITH_RCU_READ_LOCK_GUARD() {
2432         if (!virtio_should_notify(vdev, vq)) {
2433             return;
2434         }
2435     }
2436 
2437     trace_virtio_notify_irqfd(vdev, vq);
2438 
2439     /*
2440      * virtio spec 1.0 says ISR bit 0 should be ignored with MSI, but
2441      * windows drivers included in virtio-win 1.8.0 (circa 2015) are
2442      * incorrectly polling this bit during crashdump and hibernation
2443      * in MSI mode, causing a hang if this bit is never updated.
2444      * Recent releases of Windows do not really shut down, but rather
2445      * log out and hibernate to make the next startup faster.  Hence,
2446      * this manifested as a more serious hang during shutdown with
2447      *
2448      * Next driver release from 2016 fixed this problem, so working around it
2449      * is not a must, but it's easy to do so let's do it here.
2450      *
2451      * Note: it's safe to update ISR from any thread as it was switched
2452      * to an atomic operation.
2453      */
2454     virtio_set_isr(vq->vdev, 0x1);
2455     event_notifier_set(&vq->guest_notifier);
2456 }
2457 
2458 static void virtio_irq(VirtQueue *vq)
2459 {
2460     virtio_set_isr(vq->vdev, 0x1);
2461     virtio_notify_vector(vq->vdev, vq->vector);
2462 }
2463 
2464 void virtio_notify(VirtIODevice *vdev, VirtQueue *vq)
2465 {
2466     WITH_RCU_READ_LOCK_GUARD() {
2467         if (!virtio_should_notify(vdev, vq)) {
2468             return;
2469         }
2470     }
2471 
2472     trace_virtio_notify(vdev, vq);
2473     virtio_irq(vq);
2474 }
2475 
2476 void virtio_notify_config(VirtIODevice *vdev)
2477 {
2478     if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK))
2479         return;
2480 
2481     virtio_set_isr(vdev, 0x3);
2482     vdev->generation++;
2483     virtio_notify_vector(vdev, vdev->config_vector);
2484 }
2485 
2486 static bool virtio_device_endian_needed(void *opaque)
2487 {
2488     VirtIODevice *vdev = opaque;
2489 
2490     assert(vdev->device_endian != VIRTIO_DEVICE_ENDIAN_UNKNOWN);
2491     if (!virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
2492         return vdev->device_endian != virtio_default_endian();
2493     }
2494     /* Devices conforming to VIRTIO 1.0 or later are always LE. */
2495     return vdev->device_endian != VIRTIO_DEVICE_ENDIAN_LITTLE;
2496 }
2497 
2498 static bool virtio_64bit_features_needed(void *opaque)
2499 {
2500     VirtIODevice *vdev = opaque;
2501 
2502     return (vdev->host_features >> 32) != 0;
2503 }
2504 
2505 static bool virtio_virtqueue_needed(void *opaque)
2506 {
2507     VirtIODevice *vdev = opaque;
2508 
2509     return virtio_host_has_feature(vdev, VIRTIO_F_VERSION_1);
2510 }
2511 
2512 static bool virtio_packed_virtqueue_needed(void *opaque)
2513 {
2514     VirtIODevice *vdev = opaque;
2515 
2516     return virtio_host_has_feature(vdev, VIRTIO_F_RING_PACKED);
2517 }
2518 
2519 static bool virtio_ringsize_needed(void *opaque)
2520 {
2521     VirtIODevice *vdev = opaque;
2522     int i;
2523 
2524     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2525         if (vdev->vq[i].vring.num != vdev->vq[i].vring.num_default) {
2526             return true;
2527         }
2528     }
2529     return false;
2530 }
2531 
2532 static bool virtio_extra_state_needed(void *opaque)
2533 {
2534     VirtIODevice *vdev = opaque;
2535     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2536     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
2537 
2538     return k->has_extra_state &&
2539         k->has_extra_state(qbus->parent);
2540 }
2541 
2542 static bool virtio_broken_needed(void *opaque)
2543 {
2544     VirtIODevice *vdev = opaque;
2545 
2546     return vdev->broken;
2547 }
2548 
2549 static bool virtio_started_needed(void *opaque)
2550 {
2551     VirtIODevice *vdev = opaque;
2552 
2553     return vdev->started;
2554 }
2555 
2556 static bool virtio_disabled_needed(void *opaque)
2557 {
2558     VirtIODevice *vdev = opaque;
2559 
2560     return vdev->disabled;
2561 }
2562 
2563 static const VMStateDescription vmstate_virtqueue = {
2564     .name = "virtqueue_state",
2565     .version_id = 1,
2566     .minimum_version_id = 1,
2567     .fields = (VMStateField[]) {
2568         VMSTATE_UINT64(vring.avail, struct VirtQueue),
2569         VMSTATE_UINT64(vring.used, struct VirtQueue),
2570         VMSTATE_END_OF_LIST()
2571     }
2572 };
2573 
2574 static const VMStateDescription vmstate_packed_virtqueue = {
2575     .name = "packed_virtqueue_state",
2576     .version_id = 1,
2577     .minimum_version_id = 1,
2578     .fields = (VMStateField[]) {
2579         VMSTATE_UINT16(last_avail_idx, struct VirtQueue),
2580         VMSTATE_BOOL(last_avail_wrap_counter, struct VirtQueue),
2581         VMSTATE_UINT16(used_idx, struct VirtQueue),
2582         VMSTATE_BOOL(used_wrap_counter, struct VirtQueue),
2583         VMSTATE_UINT32(inuse, struct VirtQueue),
2584         VMSTATE_END_OF_LIST()
2585     }
2586 };
2587 
2588 static const VMStateDescription vmstate_virtio_virtqueues = {
2589     .name = "virtio/virtqueues",
2590     .version_id = 1,
2591     .minimum_version_id = 1,
2592     .needed = &virtio_virtqueue_needed,
2593     .fields = (VMStateField[]) {
2594         VMSTATE_STRUCT_VARRAY_POINTER_KNOWN(vq, struct VirtIODevice,
2595                       VIRTIO_QUEUE_MAX, 0, vmstate_virtqueue, VirtQueue),
2596         VMSTATE_END_OF_LIST()
2597     }
2598 };
2599 
2600 static const VMStateDescription vmstate_virtio_packed_virtqueues = {
2601     .name = "virtio/packed_virtqueues",
2602     .version_id = 1,
2603     .minimum_version_id = 1,
2604     .needed = &virtio_packed_virtqueue_needed,
2605     .fields = (VMStateField[]) {
2606         VMSTATE_STRUCT_VARRAY_POINTER_KNOWN(vq, struct VirtIODevice,
2607                       VIRTIO_QUEUE_MAX, 0, vmstate_packed_virtqueue, VirtQueue),
2608         VMSTATE_END_OF_LIST()
2609     }
2610 };
2611 
2612 static const VMStateDescription vmstate_ringsize = {
2613     .name = "ringsize_state",
2614     .version_id = 1,
2615     .minimum_version_id = 1,
2616     .fields = (VMStateField[]) {
2617         VMSTATE_UINT32(vring.num_default, struct VirtQueue),
2618         VMSTATE_END_OF_LIST()
2619     }
2620 };
2621 
2622 static const VMStateDescription vmstate_virtio_ringsize = {
2623     .name = "virtio/ringsize",
2624     .version_id = 1,
2625     .minimum_version_id = 1,
2626     .needed = &virtio_ringsize_needed,
2627     .fields = (VMStateField[]) {
2628         VMSTATE_STRUCT_VARRAY_POINTER_KNOWN(vq, struct VirtIODevice,
2629                       VIRTIO_QUEUE_MAX, 0, vmstate_ringsize, VirtQueue),
2630         VMSTATE_END_OF_LIST()
2631     }
2632 };
2633 
2634 static int get_extra_state(QEMUFile *f, void *pv, size_t size,
2635                            const VMStateField *field)
2636 {
2637     VirtIODevice *vdev = pv;
2638     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2639     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
2640 
2641     if (!k->load_extra_state) {
2642         return -1;
2643     } else {
2644         return k->load_extra_state(qbus->parent, f);
2645     }
2646 }
2647 
2648 static int put_extra_state(QEMUFile *f, void *pv, size_t size,
2649                            const VMStateField *field, JSONWriter *vmdesc)
2650 {
2651     VirtIODevice *vdev = pv;
2652     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2653     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
2654 
2655     k->save_extra_state(qbus->parent, f);
2656     return 0;
2657 }
2658 
2659 static const VMStateInfo vmstate_info_extra_state = {
2660     .name = "virtqueue_extra_state",
2661     .get = get_extra_state,
2662     .put = put_extra_state,
2663 };
2664 
2665 static const VMStateDescription vmstate_virtio_extra_state = {
2666     .name = "virtio/extra_state",
2667     .version_id = 1,
2668     .minimum_version_id = 1,
2669     .needed = &virtio_extra_state_needed,
2670     .fields = (VMStateField[]) {
2671         {
2672             .name         = "extra_state",
2673             .version_id   = 0,
2674             .field_exists = NULL,
2675             .size         = 0,
2676             .info         = &vmstate_info_extra_state,
2677             .flags        = VMS_SINGLE,
2678             .offset       = 0,
2679         },
2680         VMSTATE_END_OF_LIST()
2681     }
2682 };
2683 
2684 static const VMStateDescription vmstate_virtio_device_endian = {
2685     .name = "virtio/device_endian",
2686     .version_id = 1,
2687     .minimum_version_id = 1,
2688     .needed = &virtio_device_endian_needed,
2689     .fields = (VMStateField[]) {
2690         VMSTATE_UINT8(device_endian, VirtIODevice),
2691         VMSTATE_END_OF_LIST()
2692     }
2693 };
2694 
2695 static const VMStateDescription vmstate_virtio_64bit_features = {
2696     .name = "virtio/64bit_features",
2697     .version_id = 1,
2698     .minimum_version_id = 1,
2699     .needed = &virtio_64bit_features_needed,
2700     .fields = (VMStateField[]) {
2701         VMSTATE_UINT64(guest_features, VirtIODevice),
2702         VMSTATE_END_OF_LIST()
2703     }
2704 };
2705 
2706 static const VMStateDescription vmstate_virtio_broken = {
2707     .name = "virtio/broken",
2708     .version_id = 1,
2709     .minimum_version_id = 1,
2710     .needed = &virtio_broken_needed,
2711     .fields = (VMStateField[]) {
2712         VMSTATE_BOOL(broken, VirtIODevice),
2713         VMSTATE_END_OF_LIST()
2714     }
2715 };
2716 
2717 static const VMStateDescription vmstate_virtio_started = {
2718     .name = "virtio/started",
2719     .version_id = 1,
2720     .minimum_version_id = 1,
2721     .needed = &virtio_started_needed,
2722     .fields = (VMStateField[]) {
2723         VMSTATE_BOOL(started, VirtIODevice),
2724         VMSTATE_END_OF_LIST()
2725     }
2726 };
2727 
2728 static const VMStateDescription vmstate_virtio_disabled = {
2729     .name = "virtio/disabled",
2730     .version_id = 1,
2731     .minimum_version_id = 1,
2732     .needed = &virtio_disabled_needed,
2733     .fields = (VMStateField[]) {
2734         VMSTATE_BOOL(disabled, VirtIODevice),
2735         VMSTATE_END_OF_LIST()
2736     }
2737 };
2738 
2739 static const VMStateDescription vmstate_virtio = {
2740     .name = "virtio",
2741     .version_id = 1,
2742     .minimum_version_id = 1,
2743     .fields = (VMStateField[]) {
2744         VMSTATE_END_OF_LIST()
2745     },
2746     .subsections = (const VMStateDescription*[]) {
2747         &vmstate_virtio_device_endian,
2748         &vmstate_virtio_64bit_features,
2749         &vmstate_virtio_virtqueues,
2750         &vmstate_virtio_ringsize,
2751         &vmstate_virtio_broken,
2752         &vmstate_virtio_extra_state,
2753         &vmstate_virtio_started,
2754         &vmstate_virtio_packed_virtqueues,
2755         &vmstate_virtio_disabled,
2756         NULL
2757     }
2758 };
2759 
2760 int virtio_save(VirtIODevice *vdev, QEMUFile *f)
2761 {
2762     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2763     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
2764     VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev);
2765     uint32_t guest_features_lo = (vdev->guest_features & 0xffffffff);
2766     int i;
2767 
2768     if (k->save_config) {
2769         k->save_config(qbus->parent, f);
2770     }
2771 
2772     qemu_put_8s(f, &vdev->status);
2773     qemu_put_8s(f, &vdev->isr);
2774     qemu_put_be16s(f, &vdev->queue_sel);
2775     qemu_put_be32s(f, &guest_features_lo);
2776     qemu_put_be32(f, vdev->config_len);
2777     qemu_put_buffer(f, vdev->config, vdev->config_len);
2778 
2779     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2780         if (vdev->vq[i].vring.num == 0)
2781             break;
2782     }
2783 
2784     qemu_put_be32(f, i);
2785 
2786     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2787         if (vdev->vq[i].vring.num == 0)
2788             break;
2789 
2790         qemu_put_be32(f, vdev->vq[i].vring.num);
2791         if (k->has_variable_vring_alignment) {
2792             qemu_put_be32(f, vdev->vq[i].vring.align);
2793         }
2794         /*
2795          * Save desc now, the rest of the ring addresses are saved in
2796          * subsections for VIRTIO-1 devices.
2797          */
2798         qemu_put_be64(f, vdev->vq[i].vring.desc);
2799         qemu_put_be16s(f, &vdev->vq[i].last_avail_idx);
2800         if (k->save_queue) {
2801             k->save_queue(qbus->parent, i, f);
2802         }
2803     }
2804 
2805     if (vdc->save != NULL) {
2806         vdc->save(vdev, f);
2807     }
2808 
2809     if (vdc->vmsd) {
2810         int ret = vmstate_save_state(f, vdc->vmsd, vdev, NULL);
2811         if (ret) {
2812             return ret;
2813         }
2814     }
2815 
2816     /* Subsections */
2817     return vmstate_save_state(f, &vmstate_virtio, vdev, NULL);
2818 }
2819 
2820 /* A wrapper for use as a VMState .put function */
2821 static int virtio_device_put(QEMUFile *f, void *opaque, size_t size,
2822                               const VMStateField *field, JSONWriter *vmdesc)
2823 {
2824     return virtio_save(VIRTIO_DEVICE(opaque), f);
2825 }
2826 
2827 /* A wrapper for use as a VMState .get function */
2828 static int virtio_device_get(QEMUFile *f, void *opaque, size_t size,
2829                              const VMStateField *field)
2830 {
2831     VirtIODevice *vdev = VIRTIO_DEVICE(opaque);
2832     DeviceClass *dc = DEVICE_CLASS(VIRTIO_DEVICE_GET_CLASS(vdev));
2833 
2834     return virtio_load(vdev, f, dc->vmsd->version_id);
2835 }
2836 
2837 const VMStateInfo  virtio_vmstate_info = {
2838     .name = "virtio",
2839     .get = virtio_device_get,
2840     .put = virtio_device_put,
2841 };
2842 
2843 static int virtio_set_features_nocheck(VirtIODevice *vdev, uint64_t val)
2844 {
2845     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2846     bool bad = (val & ~(vdev->host_features)) != 0;
2847 
2848     val &= vdev->host_features;
2849     if (k->set_features) {
2850         k->set_features(vdev, val);
2851     }
2852     vdev->guest_features = val;
2853     return bad ? -1 : 0;
2854 }
2855 
2856 int virtio_set_features(VirtIODevice *vdev, uint64_t val)
2857 {
2858     int ret;
2859     /*
2860      * The driver must not attempt to set features after feature negotiation
2861      * has finished.
2862      */
2863     if (vdev->status & VIRTIO_CONFIG_S_FEATURES_OK) {
2864         return -EINVAL;
2865     }
2866 
2867     if (val & (1ull << VIRTIO_F_BAD_FEATURE)) {
2868         qemu_log_mask(LOG_GUEST_ERROR,
2869                       "%s: guest driver for %s has enabled UNUSED(30) feature bit!\n",
2870                       __func__, vdev->name);
2871     }
2872 
2873     ret = virtio_set_features_nocheck(vdev, val);
2874     if (virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX)) {
2875         /* VIRTIO_RING_F_EVENT_IDX changes the size of the caches.  */
2876         int i;
2877         for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2878             if (vdev->vq[i].vring.num != 0) {
2879                 virtio_init_region_cache(vdev, i);
2880             }
2881         }
2882     }
2883     if (!ret) {
2884         if (!virtio_device_started(vdev, vdev->status) &&
2885             !virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
2886             vdev->start_on_kick = true;
2887         }
2888     }
2889     return ret;
2890 }
2891 
2892 size_t virtio_get_config_size(const VirtIOConfigSizeParams *params,
2893                               uint64_t host_features)
2894 {
2895     size_t config_size = params->min_size;
2896     const VirtIOFeature *feature_sizes = params->feature_sizes;
2897     size_t i;
2898 
2899     for (i = 0; feature_sizes[i].flags != 0; i++) {
2900         if (host_features & feature_sizes[i].flags) {
2901             config_size = MAX(feature_sizes[i].end, config_size);
2902         }
2903     }
2904 
2905     assert(config_size <= params->max_size);
2906     return config_size;
2907 }
2908 
2909 int virtio_load(VirtIODevice *vdev, QEMUFile *f, int version_id)
2910 {
2911     int i, ret;
2912     int32_t config_len;
2913     uint32_t num;
2914     uint32_t features;
2915     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2916     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
2917     VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev);
2918 
2919     /*
2920      * We poison the endianness to ensure it does not get used before
2921      * subsections have been loaded.
2922      */
2923     vdev->device_endian = VIRTIO_DEVICE_ENDIAN_UNKNOWN;
2924 
2925     if (k->load_config) {
2926         ret = k->load_config(qbus->parent, f);
2927         if (ret)
2928             return ret;
2929     }
2930 
2931     qemu_get_8s(f, &vdev->status);
2932     qemu_get_8s(f, &vdev->isr);
2933     qemu_get_be16s(f, &vdev->queue_sel);
2934     if (vdev->queue_sel >= VIRTIO_QUEUE_MAX) {
2935         return -1;
2936     }
2937     qemu_get_be32s(f, &features);
2938 
2939     /*
2940      * Temporarily set guest_features low bits - needed by
2941      * virtio net load code testing for VIRTIO_NET_F_CTRL_GUEST_OFFLOADS
2942      * VIRTIO_NET_F_GUEST_ANNOUNCE and VIRTIO_NET_F_CTRL_VQ.
2943      *
2944      * Note: devices should always test host features in future - don't create
2945      * new dependencies like this.
2946      */
2947     vdev->guest_features = features;
2948 
2949     config_len = qemu_get_be32(f);
2950 
2951     /*
2952      * There are cases where the incoming config can be bigger or smaller
2953      * than what we have; so load what we have space for, and skip
2954      * any excess that's in the stream.
2955      */
2956     qemu_get_buffer(f, vdev->config, MIN(config_len, vdev->config_len));
2957 
2958     while (config_len > vdev->config_len) {
2959         qemu_get_byte(f);
2960         config_len--;
2961     }
2962 
2963     num = qemu_get_be32(f);
2964 
2965     if (num > VIRTIO_QUEUE_MAX) {
2966         error_report("Invalid number of virtqueues: 0x%x", num);
2967         return -1;
2968     }
2969 
2970     for (i = 0; i < num; i++) {
2971         vdev->vq[i].vring.num = qemu_get_be32(f);
2972         if (k->has_variable_vring_alignment) {
2973             vdev->vq[i].vring.align = qemu_get_be32(f);
2974         }
2975         vdev->vq[i].vring.desc = qemu_get_be64(f);
2976         qemu_get_be16s(f, &vdev->vq[i].last_avail_idx);
2977         vdev->vq[i].signalled_used_valid = false;
2978         vdev->vq[i].notification = true;
2979 
2980         if (!vdev->vq[i].vring.desc && vdev->vq[i].last_avail_idx) {
2981             error_report("VQ %d address 0x0 "
2982                          "inconsistent with Host index 0x%x",
2983                          i, vdev->vq[i].last_avail_idx);
2984             return -1;
2985         }
2986         if (k->load_queue) {
2987             ret = k->load_queue(qbus->parent, i, f);
2988             if (ret)
2989                 return ret;
2990         }
2991     }
2992 
2993     virtio_notify_vector(vdev, VIRTIO_NO_VECTOR);
2994 
2995     if (vdc->load != NULL) {
2996         ret = vdc->load(vdev, f, version_id);
2997         if (ret) {
2998             return ret;
2999         }
3000     }
3001 
3002     if (vdc->vmsd) {
3003         ret = vmstate_load_state(f, vdc->vmsd, vdev, version_id);
3004         if (ret) {
3005             return ret;
3006         }
3007     }
3008 
3009     /* Subsections */
3010     ret = vmstate_load_state(f, &vmstate_virtio, vdev, 1);
3011     if (ret) {
3012         return ret;
3013     }
3014 
3015     if (vdev->device_endian == VIRTIO_DEVICE_ENDIAN_UNKNOWN) {
3016         vdev->device_endian = virtio_default_endian();
3017     }
3018 
3019     if (virtio_64bit_features_needed(vdev)) {
3020         /*
3021          * Subsection load filled vdev->guest_features.  Run them
3022          * through virtio_set_features to sanity-check them against
3023          * host_features.
3024          */
3025         uint64_t features64 = vdev->guest_features;
3026         if (virtio_set_features_nocheck(vdev, features64) < 0) {
3027             error_report("Features 0x%" PRIx64 " unsupported. "
3028                          "Allowed features: 0x%" PRIx64,
3029                          features64, vdev->host_features);
3030             return -1;
3031         }
3032     } else {
3033         if (virtio_set_features_nocheck(vdev, features) < 0) {
3034             error_report("Features 0x%x unsupported. "
3035                          "Allowed features: 0x%" PRIx64,
3036                          features, vdev->host_features);
3037             return -1;
3038         }
3039     }
3040 
3041     if (!virtio_device_started(vdev, vdev->status) &&
3042         !virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
3043         vdev->start_on_kick = true;
3044     }
3045 
3046     RCU_READ_LOCK_GUARD();
3047     for (i = 0; i < num; i++) {
3048         if (vdev->vq[i].vring.desc) {
3049             uint16_t nheads;
3050 
3051             /*
3052              * VIRTIO-1 devices migrate desc, used, and avail ring addresses so
3053              * only the region cache needs to be set up.  Legacy devices need
3054              * to calculate used and avail ring addresses based on the desc
3055              * address.
3056              */
3057             if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
3058                 virtio_init_region_cache(vdev, i);
3059             } else {
3060                 virtio_queue_update_rings(vdev, i);
3061             }
3062 
3063             if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3064                 vdev->vq[i].shadow_avail_idx = vdev->vq[i].last_avail_idx;
3065                 vdev->vq[i].shadow_avail_wrap_counter =
3066                                         vdev->vq[i].last_avail_wrap_counter;
3067                 continue;
3068             }
3069 
3070             nheads = vring_avail_idx(&vdev->vq[i]) - vdev->vq[i].last_avail_idx;
3071             /* Check it isn't doing strange things with descriptor numbers. */
3072             if (nheads > vdev->vq[i].vring.num) {
3073                 virtio_error(vdev, "VQ %d size 0x%x Guest index 0x%x "
3074                              "inconsistent with Host index 0x%x: delta 0x%x",
3075                              i, vdev->vq[i].vring.num,
3076                              vring_avail_idx(&vdev->vq[i]),
3077                              vdev->vq[i].last_avail_idx, nheads);
3078                 vdev->vq[i].used_idx = 0;
3079                 vdev->vq[i].shadow_avail_idx = 0;
3080                 vdev->vq[i].inuse = 0;
3081                 continue;
3082             }
3083             vdev->vq[i].used_idx = vring_used_idx(&vdev->vq[i]);
3084             vdev->vq[i].shadow_avail_idx = vring_avail_idx(&vdev->vq[i]);
3085 
3086             /*
3087              * Some devices migrate VirtQueueElements that have been popped
3088              * from the avail ring but not yet returned to the used ring.
3089              * Since max ring size < UINT16_MAX it's safe to use modulo
3090              * UINT16_MAX + 1 subtraction.
3091              */
3092             vdev->vq[i].inuse = (uint16_t)(vdev->vq[i].last_avail_idx -
3093                                 vdev->vq[i].used_idx);
3094             if (vdev->vq[i].inuse > vdev->vq[i].vring.num) {
3095                 error_report("VQ %d size 0x%x < last_avail_idx 0x%x - "
3096                              "used_idx 0x%x",
3097                              i, vdev->vq[i].vring.num,
3098                              vdev->vq[i].last_avail_idx,
3099                              vdev->vq[i].used_idx);
3100                 return -1;
3101             }
3102         }
3103     }
3104 
3105     if (vdc->post_load) {
3106         ret = vdc->post_load(vdev);
3107         if (ret) {
3108             return ret;
3109         }
3110     }
3111 
3112     return 0;
3113 }
3114 
3115 void virtio_cleanup(VirtIODevice *vdev)
3116 {
3117     qemu_del_vm_change_state_handler(vdev->vmstate);
3118 }
3119 
3120 static void virtio_vmstate_change(void *opaque, bool running, RunState state)
3121 {
3122     VirtIODevice *vdev = opaque;
3123     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3124     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
3125     bool backend_run = running && virtio_device_started(vdev, vdev->status);
3126     vdev->vm_running = running;
3127 
3128     if (backend_run) {
3129         virtio_set_status(vdev, vdev->status);
3130     }
3131 
3132     if (k->vmstate_change) {
3133         k->vmstate_change(qbus->parent, backend_run);
3134     }
3135 
3136     if (!backend_run) {
3137         virtio_set_status(vdev, vdev->status);
3138     }
3139 }
3140 
3141 void virtio_instance_init_common(Object *proxy_obj, void *data,
3142                                  size_t vdev_size, const char *vdev_name)
3143 {
3144     DeviceState *vdev = data;
3145 
3146     object_initialize_child_with_props(proxy_obj, "virtio-backend", vdev,
3147                                        vdev_size, vdev_name, &error_abort,
3148                                        NULL);
3149     qdev_alias_all_properties(vdev, proxy_obj);
3150 }
3151 
3152 void virtio_init(VirtIODevice *vdev, uint16_t device_id, size_t config_size)
3153 {
3154     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3155     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
3156     int i;
3157     int nvectors = k->query_nvectors ? k->query_nvectors(qbus->parent) : 0;
3158 
3159     if (nvectors) {
3160         vdev->vector_queues =
3161             g_malloc0(sizeof(*vdev->vector_queues) * nvectors);
3162     }
3163 
3164     vdev->start_on_kick = false;
3165     vdev->started = false;
3166     vdev->vhost_started = false;
3167     vdev->device_id = device_id;
3168     vdev->status = 0;
3169     qatomic_set(&vdev->isr, 0);
3170     vdev->queue_sel = 0;
3171     vdev->config_vector = VIRTIO_NO_VECTOR;
3172     vdev->vq = g_new0(VirtQueue, VIRTIO_QUEUE_MAX);
3173     vdev->vm_running = runstate_is_running();
3174     vdev->broken = false;
3175     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
3176         vdev->vq[i].vector = VIRTIO_NO_VECTOR;
3177         vdev->vq[i].vdev = vdev;
3178         vdev->vq[i].queue_index = i;
3179         vdev->vq[i].host_notifier_enabled = false;
3180     }
3181 
3182     vdev->name = virtio_id_to_name(device_id);
3183     vdev->config_len = config_size;
3184     if (vdev->config_len) {
3185         vdev->config = g_malloc0(config_size);
3186     } else {
3187         vdev->config = NULL;
3188     }
3189     vdev->vmstate = qdev_add_vm_change_state_handler(DEVICE(vdev),
3190             virtio_vmstate_change, vdev);
3191     vdev->device_endian = virtio_default_endian();
3192     vdev->use_guest_notifier_mask = true;
3193 }
3194 
3195 /*
3196  * Only devices that have already been around prior to defining the virtio
3197  * standard support legacy mode; this includes devices not specified in the
3198  * standard. All newer devices conform to the virtio standard only.
3199  */
3200 bool virtio_legacy_allowed(VirtIODevice *vdev)
3201 {
3202     switch (vdev->device_id) {
3203     case VIRTIO_ID_NET:
3204     case VIRTIO_ID_BLOCK:
3205     case VIRTIO_ID_CONSOLE:
3206     case VIRTIO_ID_RNG:
3207     case VIRTIO_ID_BALLOON:
3208     case VIRTIO_ID_RPMSG:
3209     case VIRTIO_ID_SCSI:
3210     case VIRTIO_ID_9P:
3211     case VIRTIO_ID_RPROC_SERIAL:
3212     case VIRTIO_ID_CAIF:
3213         return true;
3214     default:
3215         return false;
3216     }
3217 }
3218 
3219 bool virtio_legacy_check_disabled(VirtIODevice *vdev)
3220 {
3221     return vdev->disable_legacy_check;
3222 }
3223 
3224 hwaddr virtio_queue_get_desc_addr(VirtIODevice *vdev, int n)
3225 {
3226     return vdev->vq[n].vring.desc;
3227 }
3228 
3229 bool virtio_queue_enabled_legacy(VirtIODevice *vdev, int n)
3230 {
3231     return virtio_queue_get_desc_addr(vdev, n) != 0;
3232 }
3233 
3234 bool virtio_queue_enabled(VirtIODevice *vdev, int n)
3235 {
3236     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3237     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
3238 
3239     if (k->queue_enabled) {
3240         return k->queue_enabled(qbus->parent, n);
3241     }
3242     return virtio_queue_enabled_legacy(vdev, n);
3243 }
3244 
3245 hwaddr virtio_queue_get_avail_addr(VirtIODevice *vdev, int n)
3246 {
3247     return vdev->vq[n].vring.avail;
3248 }
3249 
3250 hwaddr virtio_queue_get_used_addr(VirtIODevice *vdev, int n)
3251 {
3252     return vdev->vq[n].vring.used;
3253 }
3254 
3255 hwaddr virtio_queue_get_desc_size(VirtIODevice *vdev, int n)
3256 {
3257     return sizeof(VRingDesc) * vdev->vq[n].vring.num;
3258 }
3259 
3260 hwaddr virtio_queue_get_avail_size(VirtIODevice *vdev, int n)
3261 {
3262     int s;
3263 
3264     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3265         return sizeof(struct VRingPackedDescEvent);
3266     }
3267 
3268     s = virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0;
3269     return offsetof(VRingAvail, ring) +
3270         sizeof(uint16_t) * vdev->vq[n].vring.num + s;
3271 }
3272 
3273 hwaddr virtio_queue_get_used_size(VirtIODevice *vdev, int n)
3274 {
3275     int s;
3276 
3277     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3278         return sizeof(struct VRingPackedDescEvent);
3279     }
3280 
3281     s = virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0;
3282     return offsetof(VRingUsed, ring) +
3283         sizeof(VRingUsedElem) * vdev->vq[n].vring.num + s;
3284 }
3285 
3286 static unsigned int virtio_queue_packed_get_last_avail_idx(VirtIODevice *vdev,
3287                                                            int n)
3288 {
3289     unsigned int avail, used;
3290 
3291     avail = vdev->vq[n].last_avail_idx;
3292     avail |= ((uint16_t)vdev->vq[n].last_avail_wrap_counter) << 15;
3293 
3294     used = vdev->vq[n].used_idx;
3295     used |= ((uint16_t)vdev->vq[n].used_wrap_counter) << 15;
3296 
3297     return avail | used << 16;
3298 }
3299 
3300 static uint16_t virtio_queue_split_get_last_avail_idx(VirtIODevice *vdev,
3301                                                       int n)
3302 {
3303     return vdev->vq[n].last_avail_idx;
3304 }
3305 
3306 unsigned int virtio_queue_get_last_avail_idx(VirtIODevice *vdev, int n)
3307 {
3308     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3309         return virtio_queue_packed_get_last_avail_idx(vdev, n);
3310     } else {
3311         return virtio_queue_split_get_last_avail_idx(vdev, n);
3312     }
3313 }
3314 
3315 static void virtio_queue_packed_set_last_avail_idx(VirtIODevice *vdev,
3316                                                    int n, unsigned int idx)
3317 {
3318     struct VirtQueue *vq = &vdev->vq[n];
3319 
3320     vq->last_avail_idx = vq->shadow_avail_idx = idx & 0x7fff;
3321     vq->last_avail_wrap_counter =
3322         vq->shadow_avail_wrap_counter = !!(idx & 0x8000);
3323     idx >>= 16;
3324     vq->used_idx = idx & 0x7fff;
3325     vq->used_wrap_counter = !!(idx & 0x8000);
3326 }
3327 
3328 static void virtio_queue_split_set_last_avail_idx(VirtIODevice *vdev,
3329                                                   int n, unsigned int idx)
3330 {
3331         vdev->vq[n].last_avail_idx = idx;
3332         vdev->vq[n].shadow_avail_idx = idx;
3333 }
3334 
3335 void virtio_queue_set_last_avail_idx(VirtIODevice *vdev, int n,
3336                                      unsigned int idx)
3337 {
3338     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3339         virtio_queue_packed_set_last_avail_idx(vdev, n, idx);
3340     } else {
3341         virtio_queue_split_set_last_avail_idx(vdev, n, idx);
3342     }
3343 }
3344 
3345 static void virtio_queue_packed_restore_last_avail_idx(VirtIODevice *vdev,
3346                                                        int n)
3347 {
3348     /* We don't have a reference like avail idx in shared memory */
3349     return;
3350 }
3351 
3352 static void virtio_queue_split_restore_last_avail_idx(VirtIODevice *vdev,
3353                                                       int n)
3354 {
3355     RCU_READ_LOCK_GUARD();
3356     if (vdev->vq[n].vring.desc) {
3357         vdev->vq[n].last_avail_idx = vring_used_idx(&vdev->vq[n]);
3358         vdev->vq[n].shadow_avail_idx = vdev->vq[n].last_avail_idx;
3359     }
3360 }
3361 
3362 void virtio_queue_restore_last_avail_idx(VirtIODevice *vdev, int n)
3363 {
3364     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3365         virtio_queue_packed_restore_last_avail_idx(vdev, n);
3366     } else {
3367         virtio_queue_split_restore_last_avail_idx(vdev, n);
3368     }
3369 }
3370 
3371 static void virtio_queue_packed_update_used_idx(VirtIODevice *vdev, int n)
3372 {
3373     /* used idx was updated through set_last_avail_idx() */
3374     return;
3375 }
3376 
3377 static void virtio_split_packed_update_used_idx(VirtIODevice *vdev, int n)
3378 {
3379     RCU_READ_LOCK_GUARD();
3380     if (vdev->vq[n].vring.desc) {
3381         vdev->vq[n].used_idx = vring_used_idx(&vdev->vq[n]);
3382     }
3383 }
3384 
3385 void virtio_queue_update_used_idx(VirtIODevice *vdev, int n)
3386 {
3387     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3388         return virtio_queue_packed_update_used_idx(vdev, n);
3389     } else {
3390         return virtio_split_packed_update_used_idx(vdev, n);
3391     }
3392 }
3393 
3394 void virtio_queue_invalidate_signalled_used(VirtIODevice *vdev, int n)
3395 {
3396     vdev->vq[n].signalled_used_valid = false;
3397 }
3398 
3399 VirtQueue *virtio_get_queue(VirtIODevice *vdev, int n)
3400 {
3401     return vdev->vq + n;
3402 }
3403 
3404 uint16_t virtio_get_queue_index(VirtQueue *vq)
3405 {
3406     return vq->queue_index;
3407 }
3408 
3409 static void virtio_queue_guest_notifier_read(EventNotifier *n)
3410 {
3411     VirtQueue *vq = container_of(n, VirtQueue, guest_notifier);
3412     if (event_notifier_test_and_clear(n)) {
3413         virtio_irq(vq);
3414     }
3415 }
3416 static void virtio_config_guest_notifier_read(EventNotifier *n)
3417 {
3418     VirtIODevice *vdev = container_of(n, VirtIODevice, config_notifier);
3419 
3420     if (event_notifier_test_and_clear(n)) {
3421         virtio_notify_config(vdev);
3422     }
3423 }
3424 void virtio_queue_set_guest_notifier_fd_handler(VirtQueue *vq, bool assign,
3425                                                 bool with_irqfd)
3426 {
3427     if (assign && !with_irqfd) {
3428         event_notifier_set_handler(&vq->guest_notifier,
3429                                    virtio_queue_guest_notifier_read);
3430     } else {
3431         event_notifier_set_handler(&vq->guest_notifier, NULL);
3432     }
3433     if (!assign) {
3434         /* Test and clear notifier before closing it,
3435          * in case poll callback didn't have time to run. */
3436         virtio_queue_guest_notifier_read(&vq->guest_notifier);
3437     }
3438 }
3439 
3440 void virtio_config_set_guest_notifier_fd_handler(VirtIODevice *vdev,
3441                                                  bool assign, bool with_irqfd)
3442 {
3443     EventNotifier *n;
3444     n = &vdev->config_notifier;
3445     if (assign && !with_irqfd) {
3446         event_notifier_set_handler(n, virtio_config_guest_notifier_read);
3447     } else {
3448         event_notifier_set_handler(n, NULL);
3449     }
3450     if (!assign) {
3451         /* Test and clear notifier before closing it,*/
3452         /* in case poll callback didn't have time to run. */
3453         virtio_config_guest_notifier_read(n);
3454     }
3455 }
3456 
3457 EventNotifier *virtio_queue_get_guest_notifier(VirtQueue *vq)
3458 {
3459     return &vq->guest_notifier;
3460 }
3461 
3462 static void virtio_queue_host_notifier_aio_poll_begin(EventNotifier *n)
3463 {
3464     VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
3465 
3466     virtio_queue_set_notification(vq, 0);
3467 }
3468 
3469 static bool virtio_queue_host_notifier_aio_poll(void *opaque)
3470 {
3471     EventNotifier *n = opaque;
3472     VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
3473 
3474     return vq->vring.desc && !virtio_queue_empty(vq);
3475 }
3476 
3477 static void virtio_queue_host_notifier_aio_poll_ready(EventNotifier *n)
3478 {
3479     VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
3480 
3481     virtio_queue_notify_vq(vq);
3482 }
3483 
3484 static void virtio_queue_host_notifier_aio_poll_end(EventNotifier *n)
3485 {
3486     VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
3487 
3488     /* Caller polls once more after this to catch requests that race with us */
3489     virtio_queue_set_notification(vq, 1);
3490 }
3491 
3492 void virtio_queue_aio_attach_host_notifier(VirtQueue *vq, AioContext *ctx)
3493 {
3494     aio_set_event_notifier(ctx, &vq->host_notifier,
3495                            virtio_queue_host_notifier_read,
3496                            virtio_queue_host_notifier_aio_poll,
3497                            virtio_queue_host_notifier_aio_poll_ready);
3498     aio_set_event_notifier_poll(ctx, &vq->host_notifier,
3499                                 virtio_queue_host_notifier_aio_poll_begin,
3500                                 virtio_queue_host_notifier_aio_poll_end);
3501 }
3502 
3503 /*
3504  * Same as virtio_queue_aio_attach_host_notifier() but without polling. Use
3505  * this for rx virtqueues and similar cases where the virtqueue handler
3506  * function does not pop all elements. When the virtqueue is left non-empty
3507  * polling consumes CPU cycles and should not be used.
3508  */
3509 void virtio_queue_aio_attach_host_notifier_no_poll(VirtQueue *vq, AioContext *ctx)
3510 {
3511     aio_set_event_notifier(ctx, &vq->host_notifier,
3512                            virtio_queue_host_notifier_read,
3513                            NULL, NULL);
3514 }
3515 
3516 void virtio_queue_aio_detach_host_notifier(VirtQueue *vq, AioContext *ctx)
3517 {
3518     aio_set_event_notifier(ctx, &vq->host_notifier, NULL, NULL, NULL);
3519 }
3520 
3521 void virtio_queue_host_notifier_read(EventNotifier *n)
3522 {
3523     VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
3524     if (event_notifier_test_and_clear(n)) {
3525         virtio_queue_notify_vq(vq);
3526     }
3527 }
3528 
3529 EventNotifier *virtio_queue_get_host_notifier(VirtQueue *vq)
3530 {
3531     return &vq->host_notifier;
3532 }
3533 
3534 EventNotifier *virtio_config_get_guest_notifier(VirtIODevice *vdev)
3535 {
3536     return &vdev->config_notifier;
3537 }
3538 
3539 void virtio_queue_set_host_notifier_enabled(VirtQueue *vq, bool enabled)
3540 {
3541     vq->host_notifier_enabled = enabled;
3542 }
3543 
3544 int virtio_queue_set_host_notifier_mr(VirtIODevice *vdev, int n,
3545                                       MemoryRegion *mr, bool assign)
3546 {
3547     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3548     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
3549 
3550     if (k->set_host_notifier_mr) {
3551         return k->set_host_notifier_mr(qbus->parent, n, mr, assign);
3552     }
3553 
3554     return -1;
3555 }
3556 
3557 void virtio_device_set_child_bus_name(VirtIODevice *vdev, char *bus_name)
3558 {
3559     g_free(vdev->bus_name);
3560     vdev->bus_name = g_strdup(bus_name);
3561 }
3562 
3563 void G_GNUC_PRINTF(2, 3) virtio_error(VirtIODevice *vdev, const char *fmt, ...)
3564 {
3565     va_list ap;
3566 
3567     va_start(ap, fmt);
3568     error_vreport(fmt, ap);
3569     va_end(ap);
3570 
3571     if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
3572         vdev->status = vdev->status | VIRTIO_CONFIG_S_NEEDS_RESET;
3573         virtio_notify_config(vdev);
3574     }
3575 
3576     vdev->broken = true;
3577 }
3578 
3579 static void virtio_memory_listener_commit(MemoryListener *listener)
3580 {
3581     VirtIODevice *vdev = container_of(listener, VirtIODevice, listener);
3582     int i;
3583 
3584     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
3585         if (vdev->vq[i].vring.num == 0) {
3586             break;
3587         }
3588         virtio_init_region_cache(vdev, i);
3589     }
3590 }
3591 
3592 static void virtio_device_realize(DeviceState *dev, Error **errp)
3593 {
3594     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
3595     VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(dev);
3596     Error *err = NULL;
3597 
3598     /* Devices should either use vmsd or the load/save methods */
3599     assert(!vdc->vmsd || !vdc->load);
3600 
3601     if (vdc->realize != NULL) {
3602         vdc->realize(dev, &err);
3603         if (err != NULL) {
3604             error_propagate(errp, err);
3605             return;
3606         }
3607     }
3608 
3609     virtio_bus_device_plugged(vdev, &err);
3610     if (err != NULL) {
3611         error_propagate(errp, err);
3612         vdc->unrealize(dev);
3613         return;
3614     }
3615 
3616     vdev->listener.commit = virtio_memory_listener_commit;
3617     vdev->listener.name = "virtio";
3618     memory_listener_register(&vdev->listener, vdev->dma_as);
3619     QTAILQ_INSERT_TAIL(&virtio_list, vdev, next);
3620 }
3621 
3622 static void virtio_device_unrealize(DeviceState *dev)
3623 {
3624     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
3625     VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(dev);
3626 
3627     memory_listener_unregister(&vdev->listener);
3628     virtio_bus_device_unplugged(vdev);
3629 
3630     if (vdc->unrealize != NULL) {
3631         vdc->unrealize(dev);
3632     }
3633 
3634     QTAILQ_REMOVE(&virtio_list, vdev, next);
3635     g_free(vdev->bus_name);
3636     vdev->bus_name = NULL;
3637 }
3638 
3639 static void virtio_device_free_virtqueues(VirtIODevice *vdev)
3640 {
3641     int i;
3642     if (!vdev->vq) {
3643         return;
3644     }
3645 
3646     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
3647         if (vdev->vq[i].vring.num == 0) {
3648             break;
3649         }
3650         virtio_virtqueue_reset_region_cache(&vdev->vq[i]);
3651     }
3652     g_free(vdev->vq);
3653 }
3654 
3655 static void virtio_device_instance_finalize(Object *obj)
3656 {
3657     VirtIODevice *vdev = VIRTIO_DEVICE(obj);
3658 
3659     virtio_device_free_virtqueues(vdev);
3660 
3661     g_free(vdev->config);
3662     g_free(vdev->vector_queues);
3663 }
3664 
3665 static Property virtio_properties[] = {
3666     DEFINE_VIRTIO_COMMON_FEATURES(VirtIODevice, host_features),
3667     DEFINE_PROP_BOOL("use-started", VirtIODevice, use_started, true),
3668     DEFINE_PROP_BOOL("use-disabled-flag", VirtIODevice, use_disabled_flag, true),
3669     DEFINE_PROP_BOOL("x-disable-legacy-check", VirtIODevice,
3670                      disable_legacy_check, false),
3671     DEFINE_PROP_END_OF_LIST(),
3672 };
3673 
3674 static int virtio_device_start_ioeventfd_impl(VirtIODevice *vdev)
3675 {
3676     VirtioBusState *qbus = VIRTIO_BUS(qdev_get_parent_bus(DEVICE(vdev)));
3677     int i, n, r, err;
3678 
3679     /*
3680      * Batch all the host notifiers in a single transaction to avoid
3681      * quadratic time complexity in address_space_update_ioeventfds().
3682      */
3683     memory_region_transaction_begin();
3684     for (n = 0; n < VIRTIO_QUEUE_MAX; n++) {
3685         VirtQueue *vq = &vdev->vq[n];
3686         if (!virtio_queue_get_num(vdev, n)) {
3687             continue;
3688         }
3689         r = virtio_bus_set_host_notifier(qbus, n, true);
3690         if (r < 0) {
3691             err = r;
3692             goto assign_error;
3693         }
3694         event_notifier_set_handler(&vq->host_notifier,
3695                                    virtio_queue_host_notifier_read);
3696     }
3697 
3698     for (n = 0; n < VIRTIO_QUEUE_MAX; n++) {
3699         /* Kick right away to begin processing requests already in vring */
3700         VirtQueue *vq = &vdev->vq[n];
3701         if (!vq->vring.num) {
3702             continue;
3703         }
3704         event_notifier_set(&vq->host_notifier);
3705     }
3706     memory_region_transaction_commit();
3707     return 0;
3708 
3709 assign_error:
3710     i = n; /* save n for a second iteration after transaction is committed. */
3711     while (--n >= 0) {
3712         VirtQueue *vq = &vdev->vq[n];
3713         if (!virtio_queue_get_num(vdev, n)) {
3714             continue;
3715         }
3716 
3717         event_notifier_set_handler(&vq->host_notifier, NULL);
3718         r = virtio_bus_set_host_notifier(qbus, n, false);
3719         assert(r >= 0);
3720     }
3721     /*
3722      * The transaction expects the ioeventfds to be open when it
3723      * commits. Do it now, before the cleanup loop.
3724      */
3725     memory_region_transaction_commit();
3726 
3727     while (--i >= 0) {
3728         if (!virtio_queue_get_num(vdev, i)) {
3729             continue;
3730         }
3731         virtio_bus_cleanup_host_notifier(qbus, i);
3732     }
3733     return err;
3734 }
3735 
3736 int virtio_device_start_ioeventfd(VirtIODevice *vdev)
3737 {
3738     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3739     VirtioBusState *vbus = VIRTIO_BUS(qbus);
3740 
3741     return virtio_bus_start_ioeventfd(vbus);
3742 }
3743 
3744 static void virtio_device_stop_ioeventfd_impl(VirtIODevice *vdev)
3745 {
3746     VirtioBusState *qbus = VIRTIO_BUS(qdev_get_parent_bus(DEVICE(vdev)));
3747     int n, r;
3748 
3749     /*
3750      * Batch all the host notifiers in a single transaction to avoid
3751      * quadratic time complexity in address_space_update_ioeventfds().
3752      */
3753     memory_region_transaction_begin();
3754     for (n = 0; n < VIRTIO_QUEUE_MAX; n++) {
3755         VirtQueue *vq = &vdev->vq[n];
3756 
3757         if (!virtio_queue_get_num(vdev, n)) {
3758             continue;
3759         }
3760         event_notifier_set_handler(&vq->host_notifier, NULL);
3761         r = virtio_bus_set_host_notifier(qbus, n, false);
3762         assert(r >= 0);
3763     }
3764     /*
3765      * The transaction expects the ioeventfds to be open when it
3766      * commits. Do it now, before the cleanup loop.
3767      */
3768     memory_region_transaction_commit();
3769 
3770     for (n = 0; n < VIRTIO_QUEUE_MAX; n++) {
3771         if (!virtio_queue_get_num(vdev, n)) {
3772             continue;
3773         }
3774         virtio_bus_cleanup_host_notifier(qbus, n);
3775     }
3776 }
3777 
3778 int virtio_device_grab_ioeventfd(VirtIODevice *vdev)
3779 {
3780     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3781     VirtioBusState *vbus = VIRTIO_BUS(qbus);
3782 
3783     return virtio_bus_grab_ioeventfd(vbus);
3784 }
3785 
3786 void virtio_device_release_ioeventfd(VirtIODevice *vdev)
3787 {
3788     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3789     VirtioBusState *vbus = VIRTIO_BUS(qbus);
3790 
3791     virtio_bus_release_ioeventfd(vbus);
3792 }
3793 
3794 static void virtio_device_class_init(ObjectClass *klass, void *data)
3795 {
3796     /* Set the default value here. */
3797     VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
3798     DeviceClass *dc = DEVICE_CLASS(klass);
3799 
3800     dc->realize = virtio_device_realize;
3801     dc->unrealize = virtio_device_unrealize;
3802     dc->bus_type = TYPE_VIRTIO_BUS;
3803     device_class_set_props(dc, virtio_properties);
3804     vdc->start_ioeventfd = virtio_device_start_ioeventfd_impl;
3805     vdc->stop_ioeventfd = virtio_device_stop_ioeventfd_impl;
3806 
3807     vdc->legacy_features |= VIRTIO_LEGACY_FEATURES;
3808 
3809     QTAILQ_INIT(&virtio_list);
3810 }
3811 
3812 bool virtio_device_ioeventfd_enabled(VirtIODevice *vdev)
3813 {
3814     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3815     VirtioBusState *vbus = VIRTIO_BUS(qbus);
3816 
3817     return virtio_bus_ioeventfd_enabled(vbus);
3818 }
3819 
3820 VirtQueueStatus *qmp_x_query_virtio_queue_status(const char *path,
3821                                                  uint16_t queue,
3822                                                  Error **errp)
3823 {
3824     VirtIODevice *vdev;
3825     VirtQueueStatus *status;
3826 
3827     vdev = qmp_find_virtio_device(path);
3828     if (vdev == NULL) {
3829         error_setg(errp, "Path %s is not a VirtIODevice", path);
3830         return NULL;
3831     }
3832 
3833     if (queue >= VIRTIO_QUEUE_MAX || !virtio_queue_get_num(vdev, queue)) {
3834         error_setg(errp, "Invalid virtqueue number %d", queue);
3835         return NULL;
3836     }
3837 
3838     status = g_new0(VirtQueueStatus, 1);
3839     status->name = g_strdup(vdev->name);
3840     status->queue_index = vdev->vq[queue].queue_index;
3841     status->inuse = vdev->vq[queue].inuse;
3842     status->vring_num = vdev->vq[queue].vring.num;
3843     status->vring_num_default = vdev->vq[queue].vring.num_default;
3844     status->vring_align = vdev->vq[queue].vring.align;
3845     status->vring_desc = vdev->vq[queue].vring.desc;
3846     status->vring_avail = vdev->vq[queue].vring.avail;
3847     status->vring_used = vdev->vq[queue].vring.used;
3848     status->used_idx = vdev->vq[queue].used_idx;
3849     status->signalled_used = vdev->vq[queue].signalled_used;
3850     status->signalled_used_valid = vdev->vq[queue].signalled_used_valid;
3851 
3852     if (vdev->vhost_started) {
3853         VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev);
3854         struct vhost_dev *hdev = vdc->get_vhost(vdev);
3855 
3856         /* check if vq index exists for vhost as well  */
3857         if (queue >= hdev->vq_index && queue < hdev->vq_index + hdev->nvqs) {
3858             status->has_last_avail_idx = true;
3859 
3860             int vhost_vq_index =
3861                 hdev->vhost_ops->vhost_get_vq_index(hdev, queue);
3862             struct vhost_vring_state state = {
3863                 .index = vhost_vq_index,
3864             };
3865 
3866             status->last_avail_idx =
3867                 hdev->vhost_ops->vhost_get_vring_base(hdev, &state);
3868         }
3869     } else {
3870         status->has_shadow_avail_idx = true;
3871         status->has_last_avail_idx = true;
3872         status->last_avail_idx = vdev->vq[queue].last_avail_idx;
3873         status->shadow_avail_idx = vdev->vq[queue].shadow_avail_idx;
3874     }
3875 
3876     return status;
3877 }
3878 
3879 static strList *qmp_decode_vring_desc_flags(uint16_t flags)
3880 {
3881     strList *list = NULL;
3882     strList *node;
3883     int i;
3884 
3885     struct {
3886         uint16_t flag;
3887         const char *value;
3888     } map[] = {
3889         { VRING_DESC_F_NEXT, "next" },
3890         { VRING_DESC_F_WRITE, "write" },
3891         { VRING_DESC_F_INDIRECT, "indirect" },
3892         { 1 << VRING_PACKED_DESC_F_AVAIL, "avail" },
3893         { 1 << VRING_PACKED_DESC_F_USED, "used" },
3894         { 0, "" }
3895     };
3896 
3897     for (i = 0; map[i].flag; i++) {
3898         if ((map[i].flag & flags) == 0) {
3899             continue;
3900         }
3901         node = g_malloc0(sizeof(strList));
3902         node->value = g_strdup(map[i].value);
3903         node->next = list;
3904         list = node;
3905     }
3906 
3907     return list;
3908 }
3909 
3910 VirtioQueueElement *qmp_x_query_virtio_queue_element(const char *path,
3911                                                      uint16_t queue,
3912                                                      bool has_index,
3913                                                      uint16_t index,
3914                                                      Error **errp)
3915 {
3916     VirtIODevice *vdev;
3917     VirtQueue *vq;
3918     VirtioQueueElement *element = NULL;
3919 
3920     vdev = qmp_find_virtio_device(path);
3921     if (vdev == NULL) {
3922         error_setg(errp, "Path %s is not a VirtIO device", path);
3923         return NULL;
3924     }
3925 
3926     if (queue >= VIRTIO_QUEUE_MAX || !virtio_queue_get_num(vdev, queue)) {
3927         error_setg(errp, "Invalid virtqueue number %d", queue);
3928         return NULL;
3929     }
3930     vq = &vdev->vq[queue];
3931 
3932     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3933         error_setg(errp, "Packed ring not supported");
3934         return NULL;
3935     } else {
3936         unsigned int head, i, max;
3937         VRingMemoryRegionCaches *caches;
3938         MemoryRegionCache indirect_desc_cache = MEMORY_REGION_CACHE_INVALID;
3939         MemoryRegionCache *desc_cache;
3940         VRingDesc desc;
3941         VirtioRingDescList *list = NULL;
3942         VirtioRingDescList *node;
3943         int rc; int ndescs;
3944 
3945         RCU_READ_LOCK_GUARD();
3946 
3947         max = vq->vring.num;
3948 
3949         if (!has_index) {
3950             head = vring_avail_ring(vq, vq->last_avail_idx % vq->vring.num);
3951         } else {
3952             head = vring_avail_ring(vq, index % vq->vring.num);
3953         }
3954         i = head;
3955 
3956         caches = vring_get_region_caches(vq);
3957         if (!caches) {
3958             error_setg(errp, "Region caches not initialized");
3959             return NULL;
3960         }
3961         if (caches->desc.len < max * sizeof(VRingDesc)) {
3962             error_setg(errp, "Cannot map descriptor ring");
3963             return NULL;
3964         }
3965 
3966         desc_cache = &caches->desc;
3967         vring_split_desc_read(vdev, &desc, desc_cache, i);
3968         if (desc.flags & VRING_DESC_F_INDIRECT) {
3969             int64_t len;
3970             len = address_space_cache_init(&indirect_desc_cache, vdev->dma_as,
3971                                            desc.addr, desc.len, false);
3972             desc_cache = &indirect_desc_cache;
3973             if (len < desc.len) {
3974                 error_setg(errp, "Cannot map indirect buffer");
3975                 goto done;
3976             }
3977 
3978             max = desc.len / sizeof(VRingDesc);
3979             i = 0;
3980             vring_split_desc_read(vdev, &desc, desc_cache, i);
3981         }
3982 
3983         element = g_new0(VirtioQueueElement, 1);
3984         element->avail = g_new0(VirtioRingAvail, 1);
3985         element->used = g_new0(VirtioRingUsed, 1);
3986         element->name = g_strdup(vdev->name);
3987         element->index = head;
3988         element->avail->flags = vring_avail_flags(vq);
3989         element->avail->idx = vring_avail_idx(vq);
3990         element->avail->ring = head;
3991         element->used->flags = vring_used_flags(vq);
3992         element->used->idx = vring_used_idx(vq);
3993         ndescs = 0;
3994 
3995         do {
3996             /* A buggy driver may produce an infinite loop */
3997             if (ndescs >= max) {
3998                 break;
3999             }
4000             node = g_new0(VirtioRingDescList, 1);
4001             node->value = g_new0(VirtioRingDesc, 1);
4002             node->value->addr = desc.addr;
4003             node->value->len = desc.len;
4004             node->value->flags = qmp_decode_vring_desc_flags(desc.flags);
4005             node->next = list;
4006             list = node;
4007 
4008             ndescs++;
4009             rc = virtqueue_split_read_next_desc(vdev, &desc, desc_cache,
4010                                                 max, &i);
4011         } while (rc == VIRTQUEUE_READ_DESC_MORE);
4012         element->descs = list;
4013 done:
4014         address_space_cache_destroy(&indirect_desc_cache);
4015     }
4016 
4017     return element;
4018 }
4019 
4020 static const TypeInfo virtio_device_info = {
4021     .name = TYPE_VIRTIO_DEVICE,
4022     .parent = TYPE_DEVICE,
4023     .instance_size = sizeof(VirtIODevice),
4024     .class_init = virtio_device_class_init,
4025     .instance_finalize = virtio_device_instance_finalize,
4026     .abstract = true,
4027     .class_size = sizeof(VirtioDeviceClass),
4028 };
4029 
4030 static void virtio_register_types(void)
4031 {
4032     type_register_static(&virtio_device_info);
4033 }
4034 
4035 type_init(virtio_register_types)
4036