xref: /openbmc/qemu/hw/virtio/virtio.c (revision ffe98631)
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 static 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 max, 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     max = vq->vring.num;
1082 
1083     while ((rc = virtqueue_num_heads(vq, idx)) > 0) {
1084         MemoryRegionCache *desc_cache = &caches->desc;
1085         unsigned int num_bufs;
1086         VRingDesc desc;
1087         unsigned int i;
1088 
1089         num_bufs = total_bufs;
1090 
1091         if (!virtqueue_get_head(vq, idx++, &i)) {
1092             goto err;
1093         }
1094 
1095         vring_split_desc_read(vdev, &desc, desc_cache, i);
1096 
1097         if (desc.flags & VRING_DESC_F_INDIRECT) {
1098             if (!desc.len || (desc.len % sizeof(VRingDesc))) {
1099                 virtio_error(vdev, "Invalid size for indirect buffer table");
1100                 goto err;
1101             }
1102 
1103             /* If we've got too many, that implies a descriptor loop. */
1104             if (num_bufs >= max) {
1105                 virtio_error(vdev, "Looped descriptor");
1106                 goto err;
1107             }
1108 
1109             /* loop over the indirect descriptor table */
1110             len = address_space_cache_init(&indirect_desc_cache,
1111                                            vdev->dma_as,
1112                                            desc.addr, desc.len, false);
1113             desc_cache = &indirect_desc_cache;
1114             if (len < desc.len) {
1115                 virtio_error(vdev, "Cannot map indirect buffer");
1116                 goto err;
1117             }
1118 
1119             max = desc.len / sizeof(VRingDesc);
1120             num_bufs = i = 0;
1121             vring_split_desc_read(vdev, &desc, desc_cache, i);
1122         }
1123 
1124         do {
1125             /* If we've got too many, that implies a descriptor loop. */
1126             if (++num_bufs > max) {
1127                 virtio_error(vdev, "Looped descriptor");
1128                 goto err;
1129             }
1130 
1131             if (desc.flags & VRING_DESC_F_WRITE) {
1132                 in_total += desc.len;
1133             } else {
1134                 out_total += desc.len;
1135             }
1136             if (in_total >= max_in_bytes && out_total >= max_out_bytes) {
1137                 goto done;
1138             }
1139 
1140             rc = virtqueue_split_read_next_desc(vdev, &desc, desc_cache, max, &i);
1141         } while (rc == VIRTQUEUE_READ_DESC_MORE);
1142 
1143         if (rc == VIRTQUEUE_READ_DESC_ERROR) {
1144             goto err;
1145         }
1146 
1147         if (desc_cache == &indirect_desc_cache) {
1148             address_space_cache_destroy(&indirect_desc_cache);
1149             total_bufs++;
1150         } else {
1151             total_bufs = num_bufs;
1152         }
1153     }
1154 
1155     if (rc < 0) {
1156         goto err;
1157     }
1158 
1159 done:
1160     address_space_cache_destroy(&indirect_desc_cache);
1161     if (in_bytes) {
1162         *in_bytes = in_total;
1163     }
1164     if (out_bytes) {
1165         *out_bytes = out_total;
1166     }
1167     return;
1168 
1169 err:
1170     in_total = out_total = 0;
1171     goto done;
1172 }
1173 
1174 static int virtqueue_packed_read_next_desc(VirtQueue *vq,
1175                                            VRingPackedDesc *desc,
1176                                            MemoryRegionCache
1177                                            *desc_cache,
1178                                            unsigned int max,
1179                                            unsigned int *next,
1180                                            bool indirect)
1181 {
1182     /* If this descriptor says it doesn't chain, we're done. */
1183     if (!indirect && !(desc->flags & VRING_DESC_F_NEXT)) {
1184         return VIRTQUEUE_READ_DESC_DONE;
1185     }
1186 
1187     ++*next;
1188     if (*next == max) {
1189         if (indirect) {
1190             return VIRTQUEUE_READ_DESC_DONE;
1191         } else {
1192             (*next) -= vq->vring.num;
1193         }
1194     }
1195 
1196     vring_packed_desc_read(vq->vdev, desc, desc_cache, *next, false);
1197     return VIRTQUEUE_READ_DESC_MORE;
1198 }
1199 
1200 /* Called within rcu_read_lock().  */
1201 static void virtqueue_packed_get_avail_bytes(VirtQueue *vq,
1202                                              unsigned int *in_bytes,
1203                                              unsigned int *out_bytes,
1204                                              unsigned max_in_bytes,
1205                                              unsigned max_out_bytes,
1206                                              VRingMemoryRegionCaches *caches)
1207 {
1208     VirtIODevice *vdev = vq->vdev;
1209     unsigned int max, idx;
1210     unsigned int total_bufs, in_total, out_total;
1211     MemoryRegionCache *desc_cache;
1212     MemoryRegionCache indirect_desc_cache = MEMORY_REGION_CACHE_INVALID;
1213     int64_t len = 0;
1214     VRingPackedDesc desc;
1215     bool wrap_counter;
1216 
1217     idx = vq->last_avail_idx;
1218     wrap_counter = vq->last_avail_wrap_counter;
1219     total_bufs = in_total = out_total = 0;
1220 
1221     max = vq->vring.num;
1222 
1223     for (;;) {
1224         unsigned int num_bufs = total_bufs;
1225         unsigned int i = idx;
1226         int rc;
1227 
1228         desc_cache = &caches->desc;
1229         vring_packed_desc_read(vdev, &desc, desc_cache, idx, true);
1230         if (!is_desc_avail(desc.flags, wrap_counter)) {
1231             break;
1232         }
1233 
1234         if (desc.flags & VRING_DESC_F_INDIRECT) {
1235             if (desc.len % sizeof(VRingPackedDesc)) {
1236                 virtio_error(vdev, "Invalid size for indirect buffer table");
1237                 goto err;
1238             }
1239 
1240             /* If we've got too many, that implies a descriptor loop. */
1241             if (num_bufs >= max) {
1242                 virtio_error(vdev, "Looped descriptor");
1243                 goto err;
1244             }
1245 
1246             /* loop over the indirect descriptor table */
1247             len = address_space_cache_init(&indirect_desc_cache,
1248                                            vdev->dma_as,
1249                                            desc.addr, desc.len, false);
1250             desc_cache = &indirect_desc_cache;
1251             if (len < desc.len) {
1252                 virtio_error(vdev, "Cannot map indirect buffer");
1253                 goto err;
1254             }
1255 
1256             max = desc.len / sizeof(VRingPackedDesc);
1257             num_bufs = i = 0;
1258             vring_packed_desc_read(vdev, &desc, desc_cache, i, false);
1259         }
1260 
1261         do {
1262             /* If we've got too many, that implies a descriptor loop. */
1263             if (++num_bufs > max) {
1264                 virtio_error(vdev, "Looped descriptor");
1265                 goto err;
1266             }
1267 
1268             if (desc.flags & VRING_DESC_F_WRITE) {
1269                 in_total += desc.len;
1270             } else {
1271                 out_total += desc.len;
1272             }
1273             if (in_total >= max_in_bytes && out_total >= max_out_bytes) {
1274                 goto done;
1275             }
1276 
1277             rc = virtqueue_packed_read_next_desc(vq, &desc, desc_cache, max,
1278                                                  &i, desc_cache ==
1279                                                  &indirect_desc_cache);
1280         } while (rc == VIRTQUEUE_READ_DESC_MORE);
1281 
1282         if (desc_cache == &indirect_desc_cache) {
1283             address_space_cache_destroy(&indirect_desc_cache);
1284             total_bufs++;
1285             idx++;
1286         } else {
1287             idx += num_bufs - total_bufs;
1288             total_bufs = num_bufs;
1289         }
1290 
1291         if (idx >= vq->vring.num) {
1292             idx -= vq->vring.num;
1293             wrap_counter ^= 1;
1294         }
1295     }
1296 
1297     /* Record the index and wrap counter for a kick we want */
1298     vq->shadow_avail_idx = idx;
1299     vq->shadow_avail_wrap_counter = wrap_counter;
1300 done:
1301     address_space_cache_destroy(&indirect_desc_cache);
1302     if (in_bytes) {
1303         *in_bytes = in_total;
1304     }
1305     if (out_bytes) {
1306         *out_bytes = out_total;
1307     }
1308     return;
1309 
1310 err:
1311     in_total = out_total = 0;
1312     goto done;
1313 }
1314 
1315 void virtqueue_get_avail_bytes(VirtQueue *vq, unsigned int *in_bytes,
1316                                unsigned int *out_bytes,
1317                                unsigned max_in_bytes, unsigned max_out_bytes)
1318 {
1319     uint16_t desc_size;
1320     VRingMemoryRegionCaches *caches;
1321 
1322     RCU_READ_LOCK_GUARD();
1323 
1324     if (unlikely(!vq->vring.desc)) {
1325         goto err;
1326     }
1327 
1328     caches = vring_get_region_caches(vq);
1329     if (!caches) {
1330         goto err;
1331     }
1332 
1333     desc_size = virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED) ?
1334                                 sizeof(VRingPackedDesc) : sizeof(VRingDesc);
1335     if (caches->desc.len < vq->vring.num * desc_size) {
1336         virtio_error(vq->vdev, "Cannot map descriptor ring");
1337         goto err;
1338     }
1339 
1340     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) {
1341         virtqueue_packed_get_avail_bytes(vq, in_bytes, out_bytes,
1342                                          max_in_bytes, max_out_bytes,
1343                                          caches);
1344     } else {
1345         virtqueue_split_get_avail_bytes(vq, in_bytes, out_bytes,
1346                                         max_in_bytes, max_out_bytes,
1347                                         caches);
1348     }
1349 
1350     return;
1351 err:
1352     if (in_bytes) {
1353         *in_bytes = 0;
1354     }
1355     if (out_bytes) {
1356         *out_bytes = 0;
1357     }
1358 }
1359 
1360 int virtqueue_avail_bytes(VirtQueue *vq, unsigned int in_bytes,
1361                           unsigned int out_bytes)
1362 {
1363     unsigned int in_total, out_total;
1364 
1365     virtqueue_get_avail_bytes(vq, &in_total, &out_total, in_bytes, out_bytes);
1366     return in_bytes <= in_total && out_bytes <= out_total;
1367 }
1368 
1369 static bool virtqueue_map_desc(VirtIODevice *vdev, unsigned int *p_num_sg,
1370                                hwaddr *addr, struct iovec *iov,
1371                                unsigned int max_num_sg, bool is_write,
1372                                hwaddr pa, size_t sz)
1373 {
1374     bool ok = false;
1375     unsigned num_sg = *p_num_sg;
1376     assert(num_sg <= max_num_sg);
1377 
1378     if (!sz) {
1379         virtio_error(vdev, "virtio: zero sized buffers are not allowed");
1380         goto out;
1381     }
1382 
1383     while (sz) {
1384         hwaddr len = sz;
1385 
1386         if (num_sg == max_num_sg) {
1387             virtio_error(vdev, "virtio: too many write descriptors in "
1388                                "indirect table");
1389             goto out;
1390         }
1391 
1392         iov[num_sg].iov_base = dma_memory_map(vdev->dma_as, pa, &len,
1393                                               is_write ?
1394                                               DMA_DIRECTION_FROM_DEVICE :
1395                                               DMA_DIRECTION_TO_DEVICE,
1396                                               MEMTXATTRS_UNSPECIFIED);
1397         if (!iov[num_sg].iov_base) {
1398             virtio_error(vdev, "virtio: bogus descriptor or out of resources");
1399             goto out;
1400         }
1401 
1402         iov[num_sg].iov_len = len;
1403         addr[num_sg] = pa;
1404 
1405         sz -= len;
1406         pa += len;
1407         num_sg++;
1408     }
1409     ok = true;
1410 
1411 out:
1412     *p_num_sg = num_sg;
1413     return ok;
1414 }
1415 
1416 /* Only used by error code paths before we have a VirtQueueElement (therefore
1417  * virtqueue_unmap_sg() can't be used).  Assumes buffers weren't written to
1418  * yet.
1419  */
1420 static void virtqueue_undo_map_desc(unsigned int out_num, unsigned int in_num,
1421                                     struct iovec *iov)
1422 {
1423     unsigned int i;
1424 
1425     for (i = 0; i < out_num + in_num; i++) {
1426         int is_write = i >= out_num;
1427 
1428         cpu_physical_memory_unmap(iov->iov_base, iov->iov_len, is_write, 0);
1429         iov++;
1430     }
1431 }
1432 
1433 static void virtqueue_map_iovec(VirtIODevice *vdev, struct iovec *sg,
1434                                 hwaddr *addr, unsigned int num_sg,
1435                                 bool is_write)
1436 {
1437     unsigned int i;
1438     hwaddr len;
1439 
1440     for (i = 0; i < num_sg; i++) {
1441         len = sg[i].iov_len;
1442         sg[i].iov_base = dma_memory_map(vdev->dma_as,
1443                                         addr[i], &len, is_write ?
1444                                         DMA_DIRECTION_FROM_DEVICE :
1445                                         DMA_DIRECTION_TO_DEVICE,
1446                                         MEMTXATTRS_UNSPECIFIED);
1447         if (!sg[i].iov_base) {
1448             error_report("virtio: error trying to map MMIO memory");
1449             exit(1);
1450         }
1451         if (len != sg[i].iov_len) {
1452             error_report("virtio: unexpected memory split");
1453             exit(1);
1454         }
1455     }
1456 }
1457 
1458 void virtqueue_map(VirtIODevice *vdev, VirtQueueElement *elem)
1459 {
1460     virtqueue_map_iovec(vdev, elem->in_sg, elem->in_addr, elem->in_num, true);
1461     virtqueue_map_iovec(vdev, elem->out_sg, elem->out_addr, elem->out_num,
1462                                                                         false);
1463 }
1464 
1465 static void *virtqueue_alloc_element(size_t sz, unsigned out_num, unsigned in_num)
1466 {
1467     VirtQueueElement *elem;
1468     size_t in_addr_ofs = QEMU_ALIGN_UP(sz, __alignof__(elem->in_addr[0]));
1469     size_t out_addr_ofs = in_addr_ofs + in_num * sizeof(elem->in_addr[0]);
1470     size_t out_addr_end = out_addr_ofs + out_num * sizeof(elem->out_addr[0]);
1471     size_t in_sg_ofs = QEMU_ALIGN_UP(out_addr_end, __alignof__(elem->in_sg[0]));
1472     size_t out_sg_ofs = in_sg_ofs + in_num * sizeof(elem->in_sg[0]);
1473     size_t out_sg_end = out_sg_ofs + out_num * sizeof(elem->out_sg[0]);
1474 
1475     assert(sz >= sizeof(VirtQueueElement));
1476     elem = g_malloc(out_sg_end);
1477     trace_virtqueue_alloc_element(elem, sz, in_num, out_num);
1478     elem->out_num = out_num;
1479     elem->in_num = in_num;
1480     elem->in_addr = (void *)elem + in_addr_ofs;
1481     elem->out_addr = (void *)elem + out_addr_ofs;
1482     elem->in_sg = (void *)elem + in_sg_ofs;
1483     elem->out_sg = (void *)elem + out_sg_ofs;
1484     return elem;
1485 }
1486 
1487 static void *virtqueue_split_pop(VirtQueue *vq, size_t sz)
1488 {
1489     unsigned int i, head, max;
1490     VRingMemoryRegionCaches *caches;
1491     MemoryRegionCache indirect_desc_cache = MEMORY_REGION_CACHE_INVALID;
1492     MemoryRegionCache *desc_cache;
1493     int64_t len;
1494     VirtIODevice *vdev = vq->vdev;
1495     VirtQueueElement *elem = NULL;
1496     unsigned out_num, in_num, elem_entries;
1497     hwaddr addr[VIRTQUEUE_MAX_SIZE];
1498     struct iovec iov[VIRTQUEUE_MAX_SIZE];
1499     VRingDesc desc;
1500     int rc;
1501 
1502     RCU_READ_LOCK_GUARD();
1503     if (virtio_queue_empty_rcu(vq)) {
1504         goto done;
1505     }
1506     /* Needed after virtio_queue_empty(), see comment in
1507      * virtqueue_num_heads(). */
1508     smp_rmb();
1509 
1510     /* When we start there are none of either input nor output. */
1511     out_num = in_num = elem_entries = 0;
1512 
1513     max = vq->vring.num;
1514 
1515     if (vq->inuse >= vq->vring.num) {
1516         virtio_error(vdev, "Virtqueue size exceeded");
1517         goto done;
1518     }
1519 
1520     if (!virtqueue_get_head(vq, vq->last_avail_idx++, &head)) {
1521         goto done;
1522     }
1523 
1524     if (virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX)) {
1525         vring_set_avail_event(vq, vq->last_avail_idx);
1526     }
1527 
1528     i = head;
1529 
1530     caches = vring_get_region_caches(vq);
1531     if (!caches) {
1532         virtio_error(vdev, "Region caches not initialized");
1533         goto done;
1534     }
1535 
1536     if (caches->desc.len < max * sizeof(VRingDesc)) {
1537         virtio_error(vdev, "Cannot map descriptor ring");
1538         goto done;
1539     }
1540 
1541     desc_cache = &caches->desc;
1542     vring_split_desc_read(vdev, &desc, desc_cache, i);
1543     if (desc.flags & VRING_DESC_F_INDIRECT) {
1544         if (!desc.len || (desc.len % sizeof(VRingDesc))) {
1545             virtio_error(vdev, "Invalid size for indirect buffer table");
1546             goto done;
1547         }
1548 
1549         /* loop over the indirect descriptor table */
1550         len = address_space_cache_init(&indirect_desc_cache, vdev->dma_as,
1551                                        desc.addr, desc.len, false);
1552         desc_cache = &indirect_desc_cache;
1553         if (len < desc.len) {
1554             virtio_error(vdev, "Cannot map indirect buffer");
1555             goto done;
1556         }
1557 
1558         max = desc.len / sizeof(VRingDesc);
1559         i = 0;
1560         vring_split_desc_read(vdev, &desc, desc_cache, i);
1561     }
1562 
1563     /* Collect all the descriptors */
1564     do {
1565         bool map_ok;
1566 
1567         if (desc.flags & VRING_DESC_F_WRITE) {
1568             map_ok = virtqueue_map_desc(vdev, &in_num, addr + out_num,
1569                                         iov + out_num,
1570                                         VIRTQUEUE_MAX_SIZE - out_num, true,
1571                                         desc.addr, desc.len);
1572         } else {
1573             if (in_num) {
1574                 virtio_error(vdev, "Incorrect order for descriptors");
1575                 goto err_undo_map;
1576             }
1577             map_ok = virtqueue_map_desc(vdev, &out_num, addr, iov,
1578                                         VIRTQUEUE_MAX_SIZE, false,
1579                                         desc.addr, desc.len);
1580         }
1581         if (!map_ok) {
1582             goto err_undo_map;
1583         }
1584 
1585         /* If we've got too many, that implies a descriptor loop. */
1586         if (++elem_entries > max) {
1587             virtio_error(vdev, "Looped descriptor");
1588             goto err_undo_map;
1589         }
1590 
1591         rc = virtqueue_split_read_next_desc(vdev, &desc, desc_cache, max, &i);
1592     } while (rc == VIRTQUEUE_READ_DESC_MORE);
1593 
1594     if (rc == VIRTQUEUE_READ_DESC_ERROR) {
1595         goto err_undo_map;
1596     }
1597 
1598     /* Now copy what we have collected and mapped */
1599     elem = virtqueue_alloc_element(sz, out_num, in_num);
1600     elem->index = head;
1601     elem->ndescs = 1;
1602     for (i = 0; i < out_num; i++) {
1603         elem->out_addr[i] = addr[i];
1604         elem->out_sg[i] = iov[i];
1605     }
1606     for (i = 0; i < in_num; i++) {
1607         elem->in_addr[i] = addr[out_num + i];
1608         elem->in_sg[i] = iov[out_num + i];
1609     }
1610 
1611     vq->inuse++;
1612 
1613     trace_virtqueue_pop(vq, elem, elem->in_num, elem->out_num);
1614 done:
1615     address_space_cache_destroy(&indirect_desc_cache);
1616 
1617     return elem;
1618 
1619 err_undo_map:
1620     virtqueue_undo_map_desc(out_num, in_num, iov);
1621     goto done;
1622 }
1623 
1624 static void *virtqueue_packed_pop(VirtQueue *vq, size_t sz)
1625 {
1626     unsigned int i, max;
1627     VRingMemoryRegionCaches *caches;
1628     MemoryRegionCache indirect_desc_cache = MEMORY_REGION_CACHE_INVALID;
1629     MemoryRegionCache *desc_cache;
1630     int64_t len;
1631     VirtIODevice *vdev = vq->vdev;
1632     VirtQueueElement *elem = NULL;
1633     unsigned out_num, in_num, elem_entries;
1634     hwaddr addr[VIRTQUEUE_MAX_SIZE];
1635     struct iovec iov[VIRTQUEUE_MAX_SIZE];
1636     VRingPackedDesc desc;
1637     uint16_t id;
1638     int rc;
1639 
1640     RCU_READ_LOCK_GUARD();
1641     if (virtio_queue_packed_empty_rcu(vq)) {
1642         goto done;
1643     }
1644 
1645     /* When we start there are none of either input nor output. */
1646     out_num = in_num = elem_entries = 0;
1647 
1648     max = vq->vring.num;
1649 
1650     if (vq->inuse >= vq->vring.num) {
1651         virtio_error(vdev, "Virtqueue size exceeded");
1652         goto done;
1653     }
1654 
1655     i = vq->last_avail_idx;
1656 
1657     caches = vring_get_region_caches(vq);
1658     if (!caches) {
1659         virtio_error(vdev, "Region caches not initialized");
1660         goto done;
1661     }
1662 
1663     if (caches->desc.len < max * sizeof(VRingDesc)) {
1664         virtio_error(vdev, "Cannot map descriptor ring");
1665         goto done;
1666     }
1667 
1668     desc_cache = &caches->desc;
1669     vring_packed_desc_read(vdev, &desc, desc_cache, i, true);
1670     id = desc.id;
1671     if (desc.flags & VRING_DESC_F_INDIRECT) {
1672         if (desc.len % sizeof(VRingPackedDesc)) {
1673             virtio_error(vdev, "Invalid size for indirect buffer table");
1674             goto done;
1675         }
1676 
1677         /* loop over the indirect descriptor table */
1678         len = address_space_cache_init(&indirect_desc_cache, vdev->dma_as,
1679                                        desc.addr, desc.len, false);
1680         desc_cache = &indirect_desc_cache;
1681         if (len < desc.len) {
1682             virtio_error(vdev, "Cannot map indirect buffer");
1683             goto done;
1684         }
1685 
1686         max = desc.len / sizeof(VRingPackedDesc);
1687         i = 0;
1688         vring_packed_desc_read(vdev, &desc, desc_cache, i, false);
1689     }
1690 
1691     /* Collect all the descriptors */
1692     do {
1693         bool map_ok;
1694 
1695         if (desc.flags & VRING_DESC_F_WRITE) {
1696             map_ok = virtqueue_map_desc(vdev, &in_num, addr + out_num,
1697                                         iov + out_num,
1698                                         VIRTQUEUE_MAX_SIZE - out_num, true,
1699                                         desc.addr, desc.len);
1700         } else {
1701             if (in_num) {
1702                 virtio_error(vdev, "Incorrect order for descriptors");
1703                 goto err_undo_map;
1704             }
1705             map_ok = virtqueue_map_desc(vdev, &out_num, addr, iov,
1706                                         VIRTQUEUE_MAX_SIZE, false,
1707                                         desc.addr, desc.len);
1708         }
1709         if (!map_ok) {
1710             goto err_undo_map;
1711         }
1712 
1713         /* If we've got too many, that implies a descriptor loop. */
1714         if (++elem_entries > max) {
1715             virtio_error(vdev, "Looped descriptor");
1716             goto err_undo_map;
1717         }
1718 
1719         rc = virtqueue_packed_read_next_desc(vq, &desc, desc_cache, max, &i,
1720                                              desc_cache ==
1721                                              &indirect_desc_cache);
1722     } while (rc == VIRTQUEUE_READ_DESC_MORE);
1723 
1724     /* Now copy what we have collected and mapped */
1725     elem = virtqueue_alloc_element(sz, out_num, in_num);
1726     for (i = 0; i < out_num; i++) {
1727         elem->out_addr[i] = addr[i];
1728         elem->out_sg[i] = iov[i];
1729     }
1730     for (i = 0; i < in_num; i++) {
1731         elem->in_addr[i] = addr[out_num + i];
1732         elem->in_sg[i] = iov[out_num + i];
1733     }
1734 
1735     elem->index = id;
1736     elem->ndescs = (desc_cache == &indirect_desc_cache) ? 1 : elem_entries;
1737     vq->last_avail_idx += elem->ndescs;
1738     vq->inuse += elem->ndescs;
1739 
1740     if (vq->last_avail_idx >= vq->vring.num) {
1741         vq->last_avail_idx -= vq->vring.num;
1742         vq->last_avail_wrap_counter ^= 1;
1743     }
1744 
1745     vq->shadow_avail_idx = vq->last_avail_idx;
1746     vq->shadow_avail_wrap_counter = vq->last_avail_wrap_counter;
1747 
1748     trace_virtqueue_pop(vq, elem, elem->in_num, elem->out_num);
1749 done:
1750     address_space_cache_destroy(&indirect_desc_cache);
1751 
1752     return elem;
1753 
1754 err_undo_map:
1755     virtqueue_undo_map_desc(out_num, in_num, iov);
1756     goto done;
1757 }
1758 
1759 void *virtqueue_pop(VirtQueue *vq, size_t sz)
1760 {
1761     if (virtio_device_disabled(vq->vdev)) {
1762         return NULL;
1763     }
1764 
1765     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) {
1766         return virtqueue_packed_pop(vq, sz);
1767     } else {
1768         return virtqueue_split_pop(vq, sz);
1769     }
1770 }
1771 
1772 static unsigned int virtqueue_packed_drop_all(VirtQueue *vq)
1773 {
1774     VRingMemoryRegionCaches *caches;
1775     MemoryRegionCache *desc_cache;
1776     unsigned int dropped = 0;
1777     VirtQueueElement elem = {};
1778     VirtIODevice *vdev = vq->vdev;
1779     VRingPackedDesc desc;
1780 
1781     RCU_READ_LOCK_GUARD();
1782 
1783     caches = vring_get_region_caches(vq);
1784     if (!caches) {
1785         return 0;
1786     }
1787 
1788     desc_cache = &caches->desc;
1789 
1790     virtio_queue_set_notification(vq, 0);
1791 
1792     while (vq->inuse < vq->vring.num) {
1793         unsigned int idx = vq->last_avail_idx;
1794         /*
1795          * works similar to virtqueue_pop but does not map buffers
1796          * and does not allocate any memory.
1797          */
1798         vring_packed_desc_read(vdev, &desc, desc_cache,
1799                                vq->last_avail_idx , true);
1800         if (!is_desc_avail(desc.flags, vq->last_avail_wrap_counter)) {
1801             break;
1802         }
1803         elem.index = desc.id;
1804         elem.ndescs = 1;
1805         while (virtqueue_packed_read_next_desc(vq, &desc, desc_cache,
1806                                                vq->vring.num, &idx, false)) {
1807             ++elem.ndescs;
1808         }
1809         /*
1810          * immediately push the element, nothing to unmap
1811          * as both in_num and out_num are set to 0.
1812          */
1813         virtqueue_push(vq, &elem, 0);
1814         dropped++;
1815         vq->last_avail_idx += elem.ndescs;
1816         if (vq->last_avail_idx >= vq->vring.num) {
1817             vq->last_avail_idx -= vq->vring.num;
1818             vq->last_avail_wrap_counter ^= 1;
1819         }
1820     }
1821 
1822     return dropped;
1823 }
1824 
1825 static unsigned int virtqueue_split_drop_all(VirtQueue *vq)
1826 {
1827     unsigned int dropped = 0;
1828     VirtQueueElement elem = {};
1829     VirtIODevice *vdev = vq->vdev;
1830     bool fEventIdx = virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX);
1831 
1832     while (!virtio_queue_empty(vq) && vq->inuse < vq->vring.num) {
1833         /* works similar to virtqueue_pop but does not map buffers
1834         * and does not allocate any memory */
1835         smp_rmb();
1836         if (!virtqueue_get_head(vq, vq->last_avail_idx, &elem.index)) {
1837             break;
1838         }
1839         vq->inuse++;
1840         vq->last_avail_idx++;
1841         if (fEventIdx) {
1842             vring_set_avail_event(vq, vq->last_avail_idx);
1843         }
1844         /* immediately push the element, nothing to unmap
1845          * as both in_num and out_num are set to 0 */
1846         virtqueue_push(vq, &elem, 0);
1847         dropped++;
1848     }
1849 
1850     return dropped;
1851 }
1852 
1853 /* virtqueue_drop_all:
1854  * @vq: The #VirtQueue
1855  * Drops all queued buffers and indicates them to the guest
1856  * as if they are done. Useful when buffers can not be
1857  * processed but must be returned to the guest.
1858  */
1859 unsigned int virtqueue_drop_all(VirtQueue *vq)
1860 {
1861     struct VirtIODevice *vdev = vq->vdev;
1862 
1863     if (virtio_device_disabled(vq->vdev)) {
1864         return 0;
1865     }
1866 
1867     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
1868         return virtqueue_packed_drop_all(vq);
1869     } else {
1870         return virtqueue_split_drop_all(vq);
1871     }
1872 }
1873 
1874 /* Reading and writing a structure directly to QEMUFile is *awful*, but
1875  * it is what QEMU has always done by mistake.  We can change it sooner
1876  * or later by bumping the version number of the affected vm states.
1877  * In the meanwhile, since the in-memory layout of VirtQueueElement
1878  * has changed, we need to marshal to and from the layout that was
1879  * used before the change.
1880  */
1881 typedef struct VirtQueueElementOld {
1882     unsigned int index;
1883     unsigned int out_num;
1884     unsigned int in_num;
1885     hwaddr in_addr[VIRTQUEUE_MAX_SIZE];
1886     hwaddr out_addr[VIRTQUEUE_MAX_SIZE];
1887     struct iovec in_sg[VIRTQUEUE_MAX_SIZE];
1888     struct iovec out_sg[VIRTQUEUE_MAX_SIZE];
1889 } VirtQueueElementOld;
1890 
1891 void *qemu_get_virtqueue_element(VirtIODevice *vdev, QEMUFile *f, size_t sz)
1892 {
1893     VirtQueueElement *elem;
1894     VirtQueueElementOld data;
1895     int i;
1896 
1897     qemu_get_buffer(f, (uint8_t *)&data, sizeof(VirtQueueElementOld));
1898 
1899     /* TODO: teach all callers that this can fail, and return failure instead
1900      * of asserting here.
1901      * This is just one thing (there are probably more) that must be
1902      * fixed before we can allow NDEBUG compilation.
1903      */
1904     assert(ARRAY_SIZE(data.in_addr) >= data.in_num);
1905     assert(ARRAY_SIZE(data.out_addr) >= data.out_num);
1906 
1907     elem = virtqueue_alloc_element(sz, data.out_num, data.in_num);
1908     elem->index = data.index;
1909 
1910     for (i = 0; i < elem->in_num; i++) {
1911         elem->in_addr[i] = data.in_addr[i];
1912     }
1913 
1914     for (i = 0; i < elem->out_num; i++) {
1915         elem->out_addr[i] = data.out_addr[i];
1916     }
1917 
1918     for (i = 0; i < elem->in_num; i++) {
1919         /* Base is overwritten by virtqueue_map.  */
1920         elem->in_sg[i].iov_base = 0;
1921         elem->in_sg[i].iov_len = data.in_sg[i].iov_len;
1922     }
1923 
1924     for (i = 0; i < elem->out_num; i++) {
1925         /* Base is overwritten by virtqueue_map.  */
1926         elem->out_sg[i].iov_base = 0;
1927         elem->out_sg[i].iov_len = data.out_sg[i].iov_len;
1928     }
1929 
1930     if (virtio_host_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
1931         qemu_get_be32s(f, &elem->ndescs);
1932     }
1933 
1934     virtqueue_map(vdev, elem);
1935     return elem;
1936 }
1937 
1938 void qemu_put_virtqueue_element(VirtIODevice *vdev, QEMUFile *f,
1939                                 VirtQueueElement *elem)
1940 {
1941     VirtQueueElementOld data;
1942     int i;
1943 
1944     memset(&data, 0, sizeof(data));
1945     data.index = elem->index;
1946     data.in_num = elem->in_num;
1947     data.out_num = elem->out_num;
1948 
1949     for (i = 0; i < elem->in_num; i++) {
1950         data.in_addr[i] = elem->in_addr[i];
1951     }
1952 
1953     for (i = 0; i < elem->out_num; i++) {
1954         data.out_addr[i] = elem->out_addr[i];
1955     }
1956 
1957     for (i = 0; i < elem->in_num; i++) {
1958         /* Base is overwritten by virtqueue_map when loading.  Do not
1959          * save it, as it would leak the QEMU address space layout.  */
1960         data.in_sg[i].iov_len = elem->in_sg[i].iov_len;
1961     }
1962 
1963     for (i = 0; i < elem->out_num; i++) {
1964         /* Do not save iov_base as above.  */
1965         data.out_sg[i].iov_len = elem->out_sg[i].iov_len;
1966     }
1967 
1968     if (virtio_host_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
1969         qemu_put_be32s(f, &elem->ndescs);
1970     }
1971 
1972     qemu_put_buffer(f, (uint8_t *)&data, sizeof(VirtQueueElementOld));
1973 }
1974 
1975 /* virtio device */
1976 static void virtio_notify_vector(VirtIODevice *vdev, uint16_t vector)
1977 {
1978     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
1979     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
1980 
1981     if (virtio_device_disabled(vdev)) {
1982         return;
1983     }
1984 
1985     if (k->notify) {
1986         k->notify(qbus->parent, vector);
1987     }
1988 }
1989 
1990 void virtio_update_irq(VirtIODevice *vdev)
1991 {
1992     virtio_notify_vector(vdev, VIRTIO_NO_VECTOR);
1993 }
1994 
1995 static int virtio_validate_features(VirtIODevice *vdev)
1996 {
1997     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1998 
1999     if (virtio_host_has_feature(vdev, VIRTIO_F_IOMMU_PLATFORM) &&
2000         !virtio_vdev_has_feature(vdev, VIRTIO_F_IOMMU_PLATFORM)) {
2001         return -EFAULT;
2002     }
2003 
2004     if (k->validate_features) {
2005         return k->validate_features(vdev);
2006     } else {
2007         return 0;
2008     }
2009 }
2010 
2011 int virtio_set_status(VirtIODevice *vdev, uint8_t val)
2012 {
2013     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2014     trace_virtio_set_status(vdev, val);
2015 
2016     if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
2017         if (!(vdev->status & VIRTIO_CONFIG_S_FEATURES_OK) &&
2018             val & VIRTIO_CONFIG_S_FEATURES_OK) {
2019             int ret = virtio_validate_features(vdev);
2020 
2021             if (ret) {
2022                 return ret;
2023             }
2024         }
2025     }
2026 
2027     if ((vdev->status & VIRTIO_CONFIG_S_DRIVER_OK) !=
2028         (val & VIRTIO_CONFIG_S_DRIVER_OK)) {
2029         virtio_set_started(vdev, val & VIRTIO_CONFIG_S_DRIVER_OK);
2030     }
2031 
2032     if (k->set_status) {
2033         k->set_status(vdev, val);
2034     }
2035     vdev->status = val;
2036 
2037     return 0;
2038 }
2039 
2040 static enum virtio_device_endian virtio_default_endian(void)
2041 {
2042     if (target_words_bigendian()) {
2043         return VIRTIO_DEVICE_ENDIAN_BIG;
2044     } else {
2045         return VIRTIO_DEVICE_ENDIAN_LITTLE;
2046     }
2047 }
2048 
2049 static enum virtio_device_endian virtio_current_cpu_endian(void)
2050 {
2051     if (cpu_virtio_is_big_endian(current_cpu)) {
2052         return VIRTIO_DEVICE_ENDIAN_BIG;
2053     } else {
2054         return VIRTIO_DEVICE_ENDIAN_LITTLE;
2055     }
2056 }
2057 
2058 static void __virtio_queue_reset(VirtIODevice *vdev, uint32_t i)
2059 {
2060     vdev->vq[i].vring.desc = 0;
2061     vdev->vq[i].vring.avail = 0;
2062     vdev->vq[i].vring.used = 0;
2063     vdev->vq[i].last_avail_idx = 0;
2064     vdev->vq[i].shadow_avail_idx = 0;
2065     vdev->vq[i].used_idx = 0;
2066     vdev->vq[i].last_avail_wrap_counter = true;
2067     vdev->vq[i].shadow_avail_wrap_counter = true;
2068     vdev->vq[i].used_wrap_counter = true;
2069     virtio_queue_set_vector(vdev, i, VIRTIO_NO_VECTOR);
2070     vdev->vq[i].signalled_used = 0;
2071     vdev->vq[i].signalled_used_valid = false;
2072     vdev->vq[i].notification = true;
2073     vdev->vq[i].vring.num = vdev->vq[i].vring.num_default;
2074     vdev->vq[i].inuse = 0;
2075     virtio_virtqueue_reset_region_cache(&vdev->vq[i]);
2076 }
2077 
2078 void virtio_queue_reset(VirtIODevice *vdev, uint32_t queue_index)
2079 {
2080     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2081 
2082     if (k->queue_reset) {
2083         k->queue_reset(vdev, queue_index);
2084     }
2085 
2086     __virtio_queue_reset(vdev, queue_index);
2087 }
2088 
2089 void virtio_queue_enable(VirtIODevice *vdev, uint32_t queue_index)
2090 {
2091     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2092 
2093     /*
2094      * TODO: Seabios is currently out of spec and triggering this error.
2095      * So this needs to be fixed in Seabios, then this can
2096      * be re-enabled for new machine types only, and also after
2097      * being converted to LOG_GUEST_ERROR.
2098      *
2099     if (!virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
2100         error_report("queue_enable is only suppported in devices of virtio "
2101                      "1.0 or later.");
2102     }
2103     */
2104 
2105     if (k->queue_enable) {
2106         k->queue_enable(vdev, queue_index);
2107     }
2108 }
2109 
2110 void virtio_reset(void *opaque)
2111 {
2112     VirtIODevice *vdev = opaque;
2113     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2114     int i;
2115 
2116     virtio_set_status(vdev, 0);
2117     if (current_cpu) {
2118         /* Guest initiated reset */
2119         vdev->device_endian = virtio_current_cpu_endian();
2120     } else {
2121         /* System reset */
2122         vdev->device_endian = virtio_default_endian();
2123     }
2124 
2125     if (k->reset) {
2126         k->reset(vdev);
2127     }
2128 
2129     vdev->start_on_kick = false;
2130     vdev->started = false;
2131     vdev->broken = false;
2132     vdev->guest_features = 0;
2133     vdev->queue_sel = 0;
2134     vdev->status = 0;
2135     vdev->disabled = false;
2136     qatomic_set(&vdev->isr, 0);
2137     vdev->config_vector = VIRTIO_NO_VECTOR;
2138     virtio_notify_vector(vdev, vdev->config_vector);
2139 
2140     for(i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2141         __virtio_queue_reset(vdev, i);
2142     }
2143 }
2144 
2145 void virtio_queue_set_addr(VirtIODevice *vdev, int n, hwaddr addr)
2146 {
2147     if (!vdev->vq[n].vring.num) {
2148         return;
2149     }
2150     vdev->vq[n].vring.desc = addr;
2151     virtio_queue_update_rings(vdev, n);
2152 }
2153 
2154 hwaddr virtio_queue_get_addr(VirtIODevice *vdev, int n)
2155 {
2156     return vdev->vq[n].vring.desc;
2157 }
2158 
2159 void virtio_queue_set_rings(VirtIODevice *vdev, int n, hwaddr desc,
2160                             hwaddr avail, hwaddr used)
2161 {
2162     if (!vdev->vq[n].vring.num) {
2163         return;
2164     }
2165     vdev->vq[n].vring.desc = desc;
2166     vdev->vq[n].vring.avail = avail;
2167     vdev->vq[n].vring.used = used;
2168     virtio_init_region_cache(vdev, n);
2169 }
2170 
2171 void virtio_queue_set_num(VirtIODevice *vdev, int n, int num)
2172 {
2173     /* Don't allow guest to flip queue between existent and
2174      * nonexistent states, or to set it to an invalid size.
2175      */
2176     if (!!num != !!vdev->vq[n].vring.num ||
2177         num > VIRTQUEUE_MAX_SIZE ||
2178         num < 0) {
2179         return;
2180     }
2181     vdev->vq[n].vring.num = num;
2182 }
2183 
2184 VirtQueue *virtio_vector_first_queue(VirtIODevice *vdev, uint16_t vector)
2185 {
2186     return QLIST_FIRST(&vdev->vector_queues[vector]);
2187 }
2188 
2189 VirtQueue *virtio_vector_next_queue(VirtQueue *vq)
2190 {
2191     return QLIST_NEXT(vq, node);
2192 }
2193 
2194 int virtio_queue_get_num(VirtIODevice *vdev, int n)
2195 {
2196     return vdev->vq[n].vring.num;
2197 }
2198 
2199 int virtio_queue_get_max_num(VirtIODevice *vdev, int n)
2200 {
2201     return vdev->vq[n].vring.num_default;
2202 }
2203 
2204 int virtio_get_num_queues(VirtIODevice *vdev)
2205 {
2206     int i;
2207 
2208     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2209         if (!virtio_queue_get_num(vdev, i)) {
2210             break;
2211         }
2212     }
2213 
2214     return i;
2215 }
2216 
2217 void virtio_queue_set_align(VirtIODevice *vdev, int n, int align)
2218 {
2219     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2220     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
2221 
2222     /* virtio-1 compliant devices cannot change the alignment */
2223     if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
2224         error_report("tried to modify queue alignment for virtio-1 device");
2225         return;
2226     }
2227     /* Check that the transport told us it was going to do this
2228      * (so a buggy transport will immediately assert rather than
2229      * silently failing to migrate this state)
2230      */
2231     assert(k->has_variable_vring_alignment);
2232 
2233     if (align) {
2234         vdev->vq[n].vring.align = align;
2235         virtio_queue_update_rings(vdev, n);
2236     }
2237 }
2238 
2239 static void virtio_queue_notify_vq(VirtQueue *vq)
2240 {
2241     if (vq->vring.desc && vq->handle_output) {
2242         VirtIODevice *vdev = vq->vdev;
2243 
2244         if (unlikely(vdev->broken)) {
2245             return;
2246         }
2247 
2248         trace_virtio_queue_notify(vdev, vq - vdev->vq, vq);
2249         vq->handle_output(vdev, vq);
2250 
2251         if (unlikely(vdev->start_on_kick)) {
2252             virtio_set_started(vdev, true);
2253         }
2254     }
2255 }
2256 
2257 void virtio_queue_notify(VirtIODevice *vdev, int n)
2258 {
2259     VirtQueue *vq = &vdev->vq[n];
2260 
2261     if (unlikely(!vq->vring.desc || vdev->broken)) {
2262         return;
2263     }
2264 
2265     trace_virtio_queue_notify(vdev, vq - vdev->vq, vq);
2266     if (vq->host_notifier_enabled) {
2267         event_notifier_set(&vq->host_notifier);
2268     } else if (vq->handle_output) {
2269         vq->handle_output(vdev, vq);
2270 
2271         if (unlikely(vdev->start_on_kick)) {
2272             virtio_set_started(vdev, true);
2273         }
2274     }
2275 }
2276 
2277 uint16_t virtio_queue_vector(VirtIODevice *vdev, int n)
2278 {
2279     return n < VIRTIO_QUEUE_MAX ? vdev->vq[n].vector :
2280         VIRTIO_NO_VECTOR;
2281 }
2282 
2283 void virtio_queue_set_vector(VirtIODevice *vdev, int n, uint16_t vector)
2284 {
2285     VirtQueue *vq = &vdev->vq[n];
2286 
2287     if (n < VIRTIO_QUEUE_MAX) {
2288         if (vdev->vector_queues &&
2289             vdev->vq[n].vector != VIRTIO_NO_VECTOR) {
2290             QLIST_REMOVE(vq, node);
2291         }
2292         vdev->vq[n].vector = vector;
2293         if (vdev->vector_queues &&
2294             vector != VIRTIO_NO_VECTOR) {
2295             QLIST_INSERT_HEAD(&vdev->vector_queues[vector], vq, node);
2296         }
2297     }
2298 }
2299 
2300 VirtQueue *virtio_add_queue(VirtIODevice *vdev, int queue_size,
2301                             VirtIOHandleOutput handle_output)
2302 {
2303     int i;
2304 
2305     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2306         if (vdev->vq[i].vring.num == 0)
2307             break;
2308     }
2309 
2310     if (i == VIRTIO_QUEUE_MAX || queue_size > VIRTQUEUE_MAX_SIZE)
2311         abort();
2312 
2313     vdev->vq[i].vring.num = queue_size;
2314     vdev->vq[i].vring.num_default = queue_size;
2315     vdev->vq[i].vring.align = VIRTIO_PCI_VRING_ALIGN;
2316     vdev->vq[i].handle_output = handle_output;
2317     vdev->vq[i].used_elems = g_new0(VirtQueueElement, queue_size);
2318 
2319     return &vdev->vq[i];
2320 }
2321 
2322 void virtio_delete_queue(VirtQueue *vq)
2323 {
2324     vq->vring.num = 0;
2325     vq->vring.num_default = 0;
2326     vq->handle_output = NULL;
2327     g_free(vq->used_elems);
2328     vq->used_elems = NULL;
2329     virtio_virtqueue_reset_region_cache(vq);
2330 }
2331 
2332 void virtio_del_queue(VirtIODevice *vdev, int n)
2333 {
2334     if (n < 0 || n >= VIRTIO_QUEUE_MAX) {
2335         abort();
2336     }
2337 
2338     virtio_delete_queue(&vdev->vq[n]);
2339 }
2340 
2341 static void virtio_set_isr(VirtIODevice *vdev, int value)
2342 {
2343     uint8_t old = qatomic_read(&vdev->isr);
2344 
2345     /* Do not write ISR if it does not change, so that its cacheline remains
2346      * shared in the common case where the guest does not read it.
2347      */
2348     if ((old & value) != value) {
2349         qatomic_or(&vdev->isr, value);
2350     }
2351 }
2352 
2353 /* Called within rcu_read_lock(). */
2354 static bool virtio_split_should_notify(VirtIODevice *vdev, VirtQueue *vq)
2355 {
2356     uint16_t old, new;
2357     bool v;
2358     /* We need to expose used array entries before checking used event. */
2359     smp_mb();
2360     /* Always notify when queue is empty (when feature acknowledge) */
2361     if (virtio_vdev_has_feature(vdev, VIRTIO_F_NOTIFY_ON_EMPTY) &&
2362         !vq->inuse && virtio_queue_empty(vq)) {
2363         return true;
2364     }
2365 
2366     if (!virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX)) {
2367         return !(vring_avail_flags(vq) & VRING_AVAIL_F_NO_INTERRUPT);
2368     }
2369 
2370     v = vq->signalled_used_valid;
2371     vq->signalled_used_valid = true;
2372     old = vq->signalled_used;
2373     new = vq->signalled_used = vq->used_idx;
2374     return !v || vring_need_event(vring_get_used_event(vq), new, old);
2375 }
2376 
2377 static bool vring_packed_need_event(VirtQueue *vq, bool wrap,
2378                                     uint16_t off_wrap, uint16_t new,
2379                                     uint16_t old)
2380 {
2381     int off = off_wrap & ~(1 << 15);
2382 
2383     if (wrap != off_wrap >> 15) {
2384         off -= vq->vring.num;
2385     }
2386 
2387     return vring_need_event(off, new, old);
2388 }
2389 
2390 /* Called within rcu_read_lock(). */
2391 static bool virtio_packed_should_notify(VirtIODevice *vdev, VirtQueue *vq)
2392 {
2393     VRingPackedDescEvent e;
2394     uint16_t old, new;
2395     bool v;
2396     VRingMemoryRegionCaches *caches;
2397 
2398     caches = vring_get_region_caches(vq);
2399     if (!caches) {
2400         return false;
2401     }
2402 
2403     vring_packed_event_read(vdev, &caches->avail, &e);
2404 
2405     old = vq->signalled_used;
2406     new = vq->signalled_used = vq->used_idx;
2407     v = vq->signalled_used_valid;
2408     vq->signalled_used_valid = true;
2409 
2410     if (e.flags == VRING_PACKED_EVENT_FLAG_DISABLE) {
2411         return false;
2412     } else if (e.flags == VRING_PACKED_EVENT_FLAG_ENABLE) {
2413         return true;
2414     }
2415 
2416     return !v || vring_packed_need_event(vq, vq->used_wrap_counter,
2417                                          e.off_wrap, new, old);
2418 }
2419 
2420 /* Called within rcu_read_lock().  */
2421 static bool virtio_should_notify(VirtIODevice *vdev, VirtQueue *vq)
2422 {
2423     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
2424         return virtio_packed_should_notify(vdev, vq);
2425     } else {
2426         return virtio_split_should_notify(vdev, vq);
2427     }
2428 }
2429 
2430 void virtio_notify_irqfd(VirtIODevice *vdev, VirtQueue *vq)
2431 {
2432     WITH_RCU_READ_LOCK_GUARD() {
2433         if (!virtio_should_notify(vdev, vq)) {
2434             return;
2435         }
2436     }
2437 
2438     trace_virtio_notify_irqfd(vdev, vq);
2439 
2440     /*
2441      * virtio spec 1.0 says ISR bit 0 should be ignored with MSI, but
2442      * windows drivers included in virtio-win 1.8.0 (circa 2015) are
2443      * incorrectly polling this bit during crashdump and hibernation
2444      * in MSI mode, causing a hang if this bit is never updated.
2445      * Recent releases of Windows do not really shut down, but rather
2446      * log out and hibernate to make the next startup faster.  Hence,
2447      * this manifested as a more serious hang during shutdown with
2448      *
2449      * Next driver release from 2016 fixed this problem, so working around it
2450      * is not a must, but it's easy to do so let's do it here.
2451      *
2452      * Note: it's safe to update ISR from any thread as it was switched
2453      * to an atomic operation.
2454      */
2455     virtio_set_isr(vq->vdev, 0x1);
2456     event_notifier_set(&vq->guest_notifier);
2457 }
2458 
2459 static void virtio_irq(VirtQueue *vq)
2460 {
2461     virtio_set_isr(vq->vdev, 0x1);
2462     virtio_notify_vector(vq->vdev, vq->vector);
2463 }
2464 
2465 void virtio_notify(VirtIODevice *vdev, VirtQueue *vq)
2466 {
2467     WITH_RCU_READ_LOCK_GUARD() {
2468         if (!virtio_should_notify(vdev, vq)) {
2469             return;
2470         }
2471     }
2472 
2473     trace_virtio_notify(vdev, vq);
2474     virtio_irq(vq);
2475 }
2476 
2477 void virtio_notify_config(VirtIODevice *vdev)
2478 {
2479     if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK))
2480         return;
2481 
2482     virtio_set_isr(vdev, 0x3);
2483     vdev->generation++;
2484     virtio_notify_vector(vdev, vdev->config_vector);
2485 }
2486 
2487 static bool virtio_device_endian_needed(void *opaque)
2488 {
2489     VirtIODevice *vdev = opaque;
2490 
2491     assert(vdev->device_endian != VIRTIO_DEVICE_ENDIAN_UNKNOWN);
2492     if (!virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
2493         return vdev->device_endian != virtio_default_endian();
2494     }
2495     /* Devices conforming to VIRTIO 1.0 or later are always LE. */
2496     return vdev->device_endian != VIRTIO_DEVICE_ENDIAN_LITTLE;
2497 }
2498 
2499 static bool virtio_64bit_features_needed(void *opaque)
2500 {
2501     VirtIODevice *vdev = opaque;
2502 
2503     return (vdev->host_features >> 32) != 0;
2504 }
2505 
2506 static bool virtio_virtqueue_needed(void *opaque)
2507 {
2508     VirtIODevice *vdev = opaque;
2509 
2510     return virtio_host_has_feature(vdev, VIRTIO_F_VERSION_1);
2511 }
2512 
2513 static bool virtio_packed_virtqueue_needed(void *opaque)
2514 {
2515     VirtIODevice *vdev = opaque;
2516 
2517     return virtio_host_has_feature(vdev, VIRTIO_F_RING_PACKED);
2518 }
2519 
2520 static bool virtio_ringsize_needed(void *opaque)
2521 {
2522     VirtIODevice *vdev = opaque;
2523     int i;
2524 
2525     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2526         if (vdev->vq[i].vring.num != vdev->vq[i].vring.num_default) {
2527             return true;
2528         }
2529     }
2530     return false;
2531 }
2532 
2533 static bool virtio_extra_state_needed(void *opaque)
2534 {
2535     VirtIODevice *vdev = opaque;
2536     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2537     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
2538 
2539     return k->has_extra_state &&
2540         k->has_extra_state(qbus->parent);
2541 }
2542 
2543 static bool virtio_broken_needed(void *opaque)
2544 {
2545     VirtIODevice *vdev = opaque;
2546 
2547     return vdev->broken;
2548 }
2549 
2550 static bool virtio_started_needed(void *opaque)
2551 {
2552     VirtIODevice *vdev = opaque;
2553 
2554     return vdev->started;
2555 }
2556 
2557 static bool virtio_disabled_needed(void *opaque)
2558 {
2559     VirtIODevice *vdev = opaque;
2560 
2561     return vdev->disabled;
2562 }
2563 
2564 static const VMStateDescription vmstate_virtqueue = {
2565     .name = "virtqueue_state",
2566     .version_id = 1,
2567     .minimum_version_id = 1,
2568     .fields = (VMStateField[]) {
2569         VMSTATE_UINT64(vring.avail, struct VirtQueue),
2570         VMSTATE_UINT64(vring.used, struct VirtQueue),
2571         VMSTATE_END_OF_LIST()
2572     }
2573 };
2574 
2575 static const VMStateDescription vmstate_packed_virtqueue = {
2576     .name = "packed_virtqueue_state",
2577     .version_id = 1,
2578     .minimum_version_id = 1,
2579     .fields = (VMStateField[]) {
2580         VMSTATE_UINT16(last_avail_idx, struct VirtQueue),
2581         VMSTATE_BOOL(last_avail_wrap_counter, struct VirtQueue),
2582         VMSTATE_UINT16(used_idx, struct VirtQueue),
2583         VMSTATE_BOOL(used_wrap_counter, struct VirtQueue),
2584         VMSTATE_UINT32(inuse, struct VirtQueue),
2585         VMSTATE_END_OF_LIST()
2586     }
2587 };
2588 
2589 static const VMStateDescription vmstate_virtio_virtqueues = {
2590     .name = "virtio/virtqueues",
2591     .version_id = 1,
2592     .minimum_version_id = 1,
2593     .needed = &virtio_virtqueue_needed,
2594     .fields = (VMStateField[]) {
2595         VMSTATE_STRUCT_VARRAY_POINTER_KNOWN(vq, struct VirtIODevice,
2596                       VIRTIO_QUEUE_MAX, 0, vmstate_virtqueue, VirtQueue),
2597         VMSTATE_END_OF_LIST()
2598     }
2599 };
2600 
2601 static const VMStateDescription vmstate_virtio_packed_virtqueues = {
2602     .name = "virtio/packed_virtqueues",
2603     .version_id = 1,
2604     .minimum_version_id = 1,
2605     .needed = &virtio_packed_virtqueue_needed,
2606     .fields = (VMStateField[]) {
2607         VMSTATE_STRUCT_VARRAY_POINTER_KNOWN(vq, struct VirtIODevice,
2608                       VIRTIO_QUEUE_MAX, 0, vmstate_packed_virtqueue, VirtQueue),
2609         VMSTATE_END_OF_LIST()
2610     }
2611 };
2612 
2613 static const VMStateDescription vmstate_ringsize = {
2614     .name = "ringsize_state",
2615     .version_id = 1,
2616     .minimum_version_id = 1,
2617     .fields = (VMStateField[]) {
2618         VMSTATE_UINT32(vring.num_default, struct VirtQueue),
2619         VMSTATE_END_OF_LIST()
2620     }
2621 };
2622 
2623 static const VMStateDescription vmstate_virtio_ringsize = {
2624     .name = "virtio/ringsize",
2625     .version_id = 1,
2626     .minimum_version_id = 1,
2627     .needed = &virtio_ringsize_needed,
2628     .fields = (VMStateField[]) {
2629         VMSTATE_STRUCT_VARRAY_POINTER_KNOWN(vq, struct VirtIODevice,
2630                       VIRTIO_QUEUE_MAX, 0, vmstate_ringsize, VirtQueue),
2631         VMSTATE_END_OF_LIST()
2632     }
2633 };
2634 
2635 static int get_extra_state(QEMUFile *f, void *pv, size_t size,
2636                            const VMStateField *field)
2637 {
2638     VirtIODevice *vdev = pv;
2639     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2640     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
2641 
2642     if (!k->load_extra_state) {
2643         return -1;
2644     } else {
2645         return k->load_extra_state(qbus->parent, f);
2646     }
2647 }
2648 
2649 static int put_extra_state(QEMUFile *f, void *pv, size_t size,
2650                            const VMStateField *field, JSONWriter *vmdesc)
2651 {
2652     VirtIODevice *vdev = pv;
2653     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2654     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
2655 
2656     k->save_extra_state(qbus->parent, f);
2657     return 0;
2658 }
2659 
2660 static const VMStateInfo vmstate_info_extra_state = {
2661     .name = "virtqueue_extra_state",
2662     .get = get_extra_state,
2663     .put = put_extra_state,
2664 };
2665 
2666 static const VMStateDescription vmstate_virtio_extra_state = {
2667     .name = "virtio/extra_state",
2668     .version_id = 1,
2669     .minimum_version_id = 1,
2670     .needed = &virtio_extra_state_needed,
2671     .fields = (VMStateField[]) {
2672         {
2673             .name         = "extra_state",
2674             .version_id   = 0,
2675             .field_exists = NULL,
2676             .size         = 0,
2677             .info         = &vmstate_info_extra_state,
2678             .flags        = VMS_SINGLE,
2679             .offset       = 0,
2680         },
2681         VMSTATE_END_OF_LIST()
2682     }
2683 };
2684 
2685 static const VMStateDescription vmstate_virtio_device_endian = {
2686     .name = "virtio/device_endian",
2687     .version_id = 1,
2688     .minimum_version_id = 1,
2689     .needed = &virtio_device_endian_needed,
2690     .fields = (VMStateField[]) {
2691         VMSTATE_UINT8(device_endian, VirtIODevice),
2692         VMSTATE_END_OF_LIST()
2693     }
2694 };
2695 
2696 static const VMStateDescription vmstate_virtio_64bit_features = {
2697     .name = "virtio/64bit_features",
2698     .version_id = 1,
2699     .minimum_version_id = 1,
2700     .needed = &virtio_64bit_features_needed,
2701     .fields = (VMStateField[]) {
2702         VMSTATE_UINT64(guest_features, VirtIODevice),
2703         VMSTATE_END_OF_LIST()
2704     }
2705 };
2706 
2707 static const VMStateDescription vmstate_virtio_broken = {
2708     .name = "virtio/broken",
2709     .version_id = 1,
2710     .minimum_version_id = 1,
2711     .needed = &virtio_broken_needed,
2712     .fields = (VMStateField[]) {
2713         VMSTATE_BOOL(broken, VirtIODevice),
2714         VMSTATE_END_OF_LIST()
2715     }
2716 };
2717 
2718 static const VMStateDescription vmstate_virtio_started = {
2719     .name = "virtio/started",
2720     .version_id = 1,
2721     .minimum_version_id = 1,
2722     .needed = &virtio_started_needed,
2723     .fields = (VMStateField[]) {
2724         VMSTATE_BOOL(started, VirtIODevice),
2725         VMSTATE_END_OF_LIST()
2726     }
2727 };
2728 
2729 static const VMStateDescription vmstate_virtio_disabled = {
2730     .name = "virtio/disabled",
2731     .version_id = 1,
2732     .minimum_version_id = 1,
2733     .needed = &virtio_disabled_needed,
2734     .fields = (VMStateField[]) {
2735         VMSTATE_BOOL(disabled, VirtIODevice),
2736         VMSTATE_END_OF_LIST()
2737     }
2738 };
2739 
2740 static const VMStateDescription vmstate_virtio = {
2741     .name = "virtio",
2742     .version_id = 1,
2743     .minimum_version_id = 1,
2744     .fields = (VMStateField[]) {
2745         VMSTATE_END_OF_LIST()
2746     },
2747     .subsections = (const VMStateDescription*[]) {
2748         &vmstate_virtio_device_endian,
2749         &vmstate_virtio_64bit_features,
2750         &vmstate_virtio_virtqueues,
2751         &vmstate_virtio_ringsize,
2752         &vmstate_virtio_broken,
2753         &vmstate_virtio_extra_state,
2754         &vmstate_virtio_started,
2755         &vmstate_virtio_packed_virtqueues,
2756         &vmstate_virtio_disabled,
2757         NULL
2758     }
2759 };
2760 
2761 int virtio_save(VirtIODevice *vdev, QEMUFile *f)
2762 {
2763     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2764     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
2765     VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev);
2766     uint32_t guest_features_lo = (vdev->guest_features & 0xffffffff);
2767     int i;
2768 
2769     if (k->save_config) {
2770         k->save_config(qbus->parent, f);
2771     }
2772 
2773     qemu_put_8s(f, &vdev->status);
2774     qemu_put_8s(f, &vdev->isr);
2775     qemu_put_be16s(f, &vdev->queue_sel);
2776     qemu_put_be32s(f, &guest_features_lo);
2777     qemu_put_be32(f, vdev->config_len);
2778     qemu_put_buffer(f, vdev->config, vdev->config_len);
2779 
2780     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2781         if (vdev->vq[i].vring.num == 0)
2782             break;
2783     }
2784 
2785     qemu_put_be32(f, i);
2786 
2787     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2788         if (vdev->vq[i].vring.num == 0)
2789             break;
2790 
2791         qemu_put_be32(f, vdev->vq[i].vring.num);
2792         if (k->has_variable_vring_alignment) {
2793             qemu_put_be32(f, vdev->vq[i].vring.align);
2794         }
2795         /*
2796          * Save desc now, the rest of the ring addresses are saved in
2797          * subsections for VIRTIO-1 devices.
2798          */
2799         qemu_put_be64(f, vdev->vq[i].vring.desc);
2800         qemu_put_be16s(f, &vdev->vq[i].last_avail_idx);
2801         if (k->save_queue) {
2802             k->save_queue(qbus->parent, i, f);
2803         }
2804     }
2805 
2806     if (vdc->save != NULL) {
2807         vdc->save(vdev, f);
2808     }
2809 
2810     if (vdc->vmsd) {
2811         int ret = vmstate_save_state(f, vdc->vmsd, vdev, NULL);
2812         if (ret) {
2813             return ret;
2814         }
2815     }
2816 
2817     /* Subsections */
2818     return vmstate_save_state(f, &vmstate_virtio, vdev, NULL);
2819 }
2820 
2821 /* A wrapper for use as a VMState .put function */
2822 static int virtio_device_put(QEMUFile *f, void *opaque, size_t size,
2823                               const VMStateField *field, JSONWriter *vmdesc)
2824 {
2825     return virtio_save(VIRTIO_DEVICE(opaque), f);
2826 }
2827 
2828 /* A wrapper for use as a VMState .get function */
2829 static int virtio_device_get(QEMUFile *f, void *opaque, size_t size,
2830                              const VMStateField *field)
2831 {
2832     VirtIODevice *vdev = VIRTIO_DEVICE(opaque);
2833     DeviceClass *dc = DEVICE_CLASS(VIRTIO_DEVICE_GET_CLASS(vdev));
2834 
2835     return virtio_load(vdev, f, dc->vmsd->version_id);
2836 }
2837 
2838 const VMStateInfo  virtio_vmstate_info = {
2839     .name = "virtio",
2840     .get = virtio_device_get,
2841     .put = virtio_device_put,
2842 };
2843 
2844 static int virtio_set_features_nocheck(VirtIODevice *vdev, uint64_t val)
2845 {
2846     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2847     bool bad = (val & ~(vdev->host_features)) != 0;
2848 
2849     val &= vdev->host_features;
2850     if (k->set_features) {
2851         k->set_features(vdev, val);
2852     }
2853     vdev->guest_features = val;
2854     return bad ? -1 : 0;
2855 }
2856 
2857 int virtio_set_features(VirtIODevice *vdev, uint64_t val)
2858 {
2859     int ret;
2860     /*
2861      * The driver must not attempt to set features after feature negotiation
2862      * has finished.
2863      */
2864     if (vdev->status & VIRTIO_CONFIG_S_FEATURES_OK) {
2865         return -EINVAL;
2866     }
2867 
2868     if (val & (1ull << VIRTIO_F_BAD_FEATURE)) {
2869         qemu_log_mask(LOG_GUEST_ERROR,
2870                       "%s: guest driver for %s has enabled UNUSED(30) feature bit!\n",
2871                       __func__, vdev->name);
2872     }
2873 
2874     ret = virtio_set_features_nocheck(vdev, val);
2875     if (virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX)) {
2876         /* VIRTIO_RING_F_EVENT_IDX changes the size of the caches.  */
2877         int i;
2878         for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2879             if (vdev->vq[i].vring.num != 0) {
2880                 virtio_init_region_cache(vdev, i);
2881             }
2882         }
2883     }
2884     if (!ret) {
2885         if (!virtio_device_started(vdev, vdev->status) &&
2886             !virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
2887             vdev->start_on_kick = true;
2888         }
2889     }
2890     return ret;
2891 }
2892 
2893 size_t virtio_get_config_size(const VirtIOConfigSizeParams *params,
2894                               uint64_t host_features)
2895 {
2896     size_t config_size = params->min_size;
2897     const VirtIOFeature *feature_sizes = params->feature_sizes;
2898     size_t i;
2899 
2900     for (i = 0; feature_sizes[i].flags != 0; i++) {
2901         if (host_features & feature_sizes[i].flags) {
2902             config_size = MAX(feature_sizes[i].end, config_size);
2903         }
2904     }
2905 
2906     assert(config_size <= params->max_size);
2907     return config_size;
2908 }
2909 
2910 int virtio_load(VirtIODevice *vdev, QEMUFile *f, int version_id)
2911 {
2912     int i, ret;
2913     int32_t config_len;
2914     uint32_t num;
2915     uint32_t features;
2916     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2917     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
2918     VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev);
2919 
2920     /*
2921      * We poison the endianness to ensure it does not get used before
2922      * subsections have been loaded.
2923      */
2924     vdev->device_endian = VIRTIO_DEVICE_ENDIAN_UNKNOWN;
2925 
2926     if (k->load_config) {
2927         ret = k->load_config(qbus->parent, f);
2928         if (ret)
2929             return ret;
2930     }
2931 
2932     qemu_get_8s(f, &vdev->status);
2933     qemu_get_8s(f, &vdev->isr);
2934     qemu_get_be16s(f, &vdev->queue_sel);
2935     if (vdev->queue_sel >= VIRTIO_QUEUE_MAX) {
2936         return -1;
2937     }
2938     qemu_get_be32s(f, &features);
2939 
2940     /*
2941      * Temporarily set guest_features low bits - needed by
2942      * virtio net load code testing for VIRTIO_NET_F_CTRL_GUEST_OFFLOADS
2943      * VIRTIO_NET_F_GUEST_ANNOUNCE and VIRTIO_NET_F_CTRL_VQ.
2944      *
2945      * Note: devices should always test host features in future - don't create
2946      * new dependencies like this.
2947      */
2948     vdev->guest_features = features;
2949 
2950     config_len = qemu_get_be32(f);
2951 
2952     /*
2953      * There are cases where the incoming config can be bigger or smaller
2954      * than what we have; so load what we have space for, and skip
2955      * any excess that's in the stream.
2956      */
2957     qemu_get_buffer(f, vdev->config, MIN(config_len, vdev->config_len));
2958 
2959     while (config_len > vdev->config_len) {
2960         qemu_get_byte(f);
2961         config_len--;
2962     }
2963 
2964     num = qemu_get_be32(f);
2965 
2966     if (num > VIRTIO_QUEUE_MAX) {
2967         error_report("Invalid number of virtqueues: 0x%x", num);
2968         return -1;
2969     }
2970 
2971     for (i = 0; i < num; i++) {
2972         vdev->vq[i].vring.num = qemu_get_be32(f);
2973         if (k->has_variable_vring_alignment) {
2974             vdev->vq[i].vring.align = qemu_get_be32(f);
2975         }
2976         vdev->vq[i].vring.desc = qemu_get_be64(f);
2977         qemu_get_be16s(f, &vdev->vq[i].last_avail_idx);
2978         vdev->vq[i].signalled_used_valid = false;
2979         vdev->vq[i].notification = true;
2980 
2981         if (!vdev->vq[i].vring.desc && vdev->vq[i].last_avail_idx) {
2982             error_report("VQ %d address 0x0 "
2983                          "inconsistent with Host index 0x%x",
2984                          i, vdev->vq[i].last_avail_idx);
2985             return -1;
2986         }
2987         if (k->load_queue) {
2988             ret = k->load_queue(qbus->parent, i, f);
2989             if (ret)
2990                 return ret;
2991         }
2992     }
2993 
2994     virtio_notify_vector(vdev, VIRTIO_NO_VECTOR);
2995 
2996     if (vdc->load != NULL) {
2997         ret = vdc->load(vdev, f, version_id);
2998         if (ret) {
2999             return ret;
3000         }
3001     }
3002 
3003     if (vdc->vmsd) {
3004         ret = vmstate_load_state(f, vdc->vmsd, vdev, version_id);
3005         if (ret) {
3006             return ret;
3007         }
3008     }
3009 
3010     /* Subsections */
3011     ret = vmstate_load_state(f, &vmstate_virtio, vdev, 1);
3012     if (ret) {
3013         return ret;
3014     }
3015 
3016     if (vdev->device_endian == VIRTIO_DEVICE_ENDIAN_UNKNOWN) {
3017         vdev->device_endian = virtio_default_endian();
3018     }
3019 
3020     if (virtio_64bit_features_needed(vdev)) {
3021         /*
3022          * Subsection load filled vdev->guest_features.  Run them
3023          * through virtio_set_features to sanity-check them against
3024          * host_features.
3025          */
3026         uint64_t features64 = vdev->guest_features;
3027         if (virtio_set_features_nocheck(vdev, features64) < 0) {
3028             error_report("Features 0x%" PRIx64 " unsupported. "
3029                          "Allowed features: 0x%" PRIx64,
3030                          features64, vdev->host_features);
3031             return -1;
3032         }
3033     } else {
3034         if (virtio_set_features_nocheck(vdev, features) < 0) {
3035             error_report("Features 0x%x unsupported. "
3036                          "Allowed features: 0x%" PRIx64,
3037                          features, vdev->host_features);
3038             return -1;
3039         }
3040     }
3041 
3042     if (!virtio_device_started(vdev, vdev->status) &&
3043         !virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
3044         vdev->start_on_kick = true;
3045     }
3046 
3047     RCU_READ_LOCK_GUARD();
3048     for (i = 0; i < num; i++) {
3049         if (vdev->vq[i].vring.desc) {
3050             uint16_t nheads;
3051 
3052             /*
3053              * VIRTIO-1 devices migrate desc, used, and avail ring addresses so
3054              * only the region cache needs to be set up.  Legacy devices need
3055              * to calculate used and avail ring addresses based on the desc
3056              * address.
3057              */
3058             if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
3059                 virtio_init_region_cache(vdev, i);
3060             } else {
3061                 virtio_queue_update_rings(vdev, i);
3062             }
3063 
3064             if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3065                 vdev->vq[i].shadow_avail_idx = vdev->vq[i].last_avail_idx;
3066                 vdev->vq[i].shadow_avail_wrap_counter =
3067                                         vdev->vq[i].last_avail_wrap_counter;
3068                 continue;
3069             }
3070 
3071             nheads = vring_avail_idx(&vdev->vq[i]) - vdev->vq[i].last_avail_idx;
3072             /* Check it isn't doing strange things with descriptor numbers. */
3073             if (nheads > vdev->vq[i].vring.num) {
3074                 virtio_error(vdev, "VQ %d size 0x%x Guest index 0x%x "
3075                              "inconsistent with Host index 0x%x: delta 0x%x",
3076                              i, vdev->vq[i].vring.num,
3077                              vring_avail_idx(&vdev->vq[i]),
3078                              vdev->vq[i].last_avail_idx, nheads);
3079                 vdev->vq[i].used_idx = 0;
3080                 vdev->vq[i].shadow_avail_idx = 0;
3081                 vdev->vq[i].inuse = 0;
3082                 continue;
3083             }
3084             vdev->vq[i].used_idx = vring_used_idx(&vdev->vq[i]);
3085             vdev->vq[i].shadow_avail_idx = vring_avail_idx(&vdev->vq[i]);
3086 
3087             /*
3088              * Some devices migrate VirtQueueElements that have been popped
3089              * from the avail ring but not yet returned to the used ring.
3090              * Since max ring size < UINT16_MAX it's safe to use modulo
3091              * UINT16_MAX + 1 subtraction.
3092              */
3093             vdev->vq[i].inuse = (uint16_t)(vdev->vq[i].last_avail_idx -
3094                                 vdev->vq[i].used_idx);
3095             if (vdev->vq[i].inuse > vdev->vq[i].vring.num) {
3096                 error_report("VQ %d size 0x%x < last_avail_idx 0x%x - "
3097                              "used_idx 0x%x",
3098                              i, vdev->vq[i].vring.num,
3099                              vdev->vq[i].last_avail_idx,
3100                              vdev->vq[i].used_idx);
3101                 return -1;
3102             }
3103         }
3104     }
3105 
3106     if (vdc->post_load) {
3107         ret = vdc->post_load(vdev);
3108         if (ret) {
3109             return ret;
3110         }
3111     }
3112 
3113     return 0;
3114 }
3115 
3116 void virtio_cleanup(VirtIODevice *vdev)
3117 {
3118     qemu_del_vm_change_state_handler(vdev->vmstate);
3119 }
3120 
3121 static void virtio_vmstate_change(void *opaque, bool running, RunState state)
3122 {
3123     VirtIODevice *vdev = opaque;
3124     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3125     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
3126     bool backend_run = running && virtio_device_started(vdev, vdev->status);
3127     vdev->vm_running = running;
3128 
3129     if (backend_run) {
3130         virtio_set_status(vdev, vdev->status);
3131     }
3132 
3133     if (k->vmstate_change) {
3134         k->vmstate_change(qbus->parent, backend_run);
3135     }
3136 
3137     if (!backend_run) {
3138         virtio_set_status(vdev, vdev->status);
3139     }
3140 }
3141 
3142 void virtio_instance_init_common(Object *proxy_obj, void *data,
3143                                  size_t vdev_size, const char *vdev_name)
3144 {
3145     DeviceState *vdev = data;
3146 
3147     object_initialize_child_with_props(proxy_obj, "virtio-backend", vdev,
3148                                        vdev_size, vdev_name, &error_abort,
3149                                        NULL);
3150     qdev_alias_all_properties(vdev, proxy_obj);
3151 }
3152 
3153 void virtio_init(VirtIODevice *vdev, uint16_t device_id, size_t config_size)
3154 {
3155     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3156     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
3157     int i;
3158     int nvectors = k->query_nvectors ? k->query_nvectors(qbus->parent) : 0;
3159 
3160     if (nvectors) {
3161         vdev->vector_queues =
3162             g_malloc0(sizeof(*vdev->vector_queues) * nvectors);
3163     }
3164 
3165     vdev->start_on_kick = false;
3166     vdev->started = false;
3167     vdev->vhost_started = false;
3168     vdev->device_id = device_id;
3169     vdev->status = 0;
3170     qatomic_set(&vdev->isr, 0);
3171     vdev->queue_sel = 0;
3172     vdev->config_vector = VIRTIO_NO_VECTOR;
3173     vdev->vq = g_new0(VirtQueue, VIRTIO_QUEUE_MAX);
3174     vdev->vm_running = runstate_is_running();
3175     vdev->broken = false;
3176     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
3177         vdev->vq[i].vector = VIRTIO_NO_VECTOR;
3178         vdev->vq[i].vdev = vdev;
3179         vdev->vq[i].queue_index = i;
3180         vdev->vq[i].host_notifier_enabled = false;
3181     }
3182 
3183     vdev->name = virtio_id_to_name(device_id);
3184     vdev->config_len = config_size;
3185     if (vdev->config_len) {
3186         vdev->config = g_malloc0(config_size);
3187     } else {
3188         vdev->config = NULL;
3189     }
3190     vdev->vmstate = qdev_add_vm_change_state_handler(DEVICE(vdev),
3191             virtio_vmstate_change, vdev);
3192     vdev->device_endian = virtio_default_endian();
3193     vdev->use_guest_notifier_mask = true;
3194 }
3195 
3196 /*
3197  * Only devices that have already been around prior to defining the virtio
3198  * standard support legacy mode; this includes devices not specified in the
3199  * standard. All newer devices conform to the virtio standard only.
3200  */
3201 bool virtio_legacy_allowed(VirtIODevice *vdev)
3202 {
3203     switch (vdev->device_id) {
3204     case VIRTIO_ID_NET:
3205     case VIRTIO_ID_BLOCK:
3206     case VIRTIO_ID_CONSOLE:
3207     case VIRTIO_ID_RNG:
3208     case VIRTIO_ID_BALLOON:
3209     case VIRTIO_ID_RPMSG:
3210     case VIRTIO_ID_SCSI:
3211     case VIRTIO_ID_9P:
3212     case VIRTIO_ID_RPROC_SERIAL:
3213     case VIRTIO_ID_CAIF:
3214         return true;
3215     default:
3216         return false;
3217     }
3218 }
3219 
3220 bool virtio_legacy_check_disabled(VirtIODevice *vdev)
3221 {
3222     return vdev->disable_legacy_check;
3223 }
3224 
3225 hwaddr virtio_queue_get_desc_addr(VirtIODevice *vdev, int n)
3226 {
3227     return vdev->vq[n].vring.desc;
3228 }
3229 
3230 bool virtio_queue_enabled_legacy(VirtIODevice *vdev, int n)
3231 {
3232     return virtio_queue_get_desc_addr(vdev, n) != 0;
3233 }
3234 
3235 bool virtio_queue_enabled(VirtIODevice *vdev, int n)
3236 {
3237     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3238     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
3239 
3240     if (k->queue_enabled) {
3241         return k->queue_enabled(qbus->parent, n);
3242     }
3243     return virtio_queue_enabled_legacy(vdev, n);
3244 }
3245 
3246 hwaddr virtio_queue_get_avail_addr(VirtIODevice *vdev, int n)
3247 {
3248     return vdev->vq[n].vring.avail;
3249 }
3250 
3251 hwaddr virtio_queue_get_used_addr(VirtIODevice *vdev, int n)
3252 {
3253     return vdev->vq[n].vring.used;
3254 }
3255 
3256 hwaddr virtio_queue_get_desc_size(VirtIODevice *vdev, int n)
3257 {
3258     return sizeof(VRingDesc) * vdev->vq[n].vring.num;
3259 }
3260 
3261 hwaddr virtio_queue_get_avail_size(VirtIODevice *vdev, int n)
3262 {
3263     int s;
3264 
3265     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3266         return sizeof(struct VRingPackedDescEvent);
3267     }
3268 
3269     s = virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0;
3270     return offsetof(VRingAvail, ring) +
3271         sizeof(uint16_t) * vdev->vq[n].vring.num + s;
3272 }
3273 
3274 hwaddr virtio_queue_get_used_size(VirtIODevice *vdev, int n)
3275 {
3276     int s;
3277 
3278     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3279         return sizeof(struct VRingPackedDescEvent);
3280     }
3281 
3282     s = virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0;
3283     return offsetof(VRingUsed, ring) +
3284         sizeof(VRingUsedElem) * vdev->vq[n].vring.num + s;
3285 }
3286 
3287 static unsigned int virtio_queue_packed_get_last_avail_idx(VirtIODevice *vdev,
3288                                                            int n)
3289 {
3290     unsigned int avail, used;
3291 
3292     avail = vdev->vq[n].last_avail_idx;
3293     avail |= ((uint16_t)vdev->vq[n].last_avail_wrap_counter) << 15;
3294 
3295     used = vdev->vq[n].used_idx;
3296     used |= ((uint16_t)vdev->vq[n].used_wrap_counter) << 15;
3297 
3298     return avail | used << 16;
3299 }
3300 
3301 static uint16_t virtio_queue_split_get_last_avail_idx(VirtIODevice *vdev,
3302                                                       int n)
3303 {
3304     return vdev->vq[n].last_avail_idx;
3305 }
3306 
3307 unsigned int virtio_queue_get_last_avail_idx(VirtIODevice *vdev, int n)
3308 {
3309     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3310         return virtio_queue_packed_get_last_avail_idx(vdev, n);
3311     } else {
3312         return virtio_queue_split_get_last_avail_idx(vdev, n);
3313     }
3314 }
3315 
3316 static void virtio_queue_packed_set_last_avail_idx(VirtIODevice *vdev,
3317                                                    int n, unsigned int idx)
3318 {
3319     struct VirtQueue *vq = &vdev->vq[n];
3320 
3321     vq->last_avail_idx = vq->shadow_avail_idx = idx & 0x7fff;
3322     vq->last_avail_wrap_counter =
3323         vq->shadow_avail_wrap_counter = !!(idx & 0x8000);
3324     idx >>= 16;
3325     vq->used_idx = idx & 0x7ffff;
3326     vq->used_wrap_counter = !!(idx & 0x8000);
3327 }
3328 
3329 static void virtio_queue_split_set_last_avail_idx(VirtIODevice *vdev,
3330                                                   int n, unsigned int idx)
3331 {
3332         vdev->vq[n].last_avail_idx = idx;
3333         vdev->vq[n].shadow_avail_idx = idx;
3334 }
3335 
3336 void virtio_queue_set_last_avail_idx(VirtIODevice *vdev, int n,
3337                                      unsigned int idx)
3338 {
3339     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3340         virtio_queue_packed_set_last_avail_idx(vdev, n, idx);
3341     } else {
3342         virtio_queue_split_set_last_avail_idx(vdev, n, idx);
3343     }
3344 }
3345 
3346 static void virtio_queue_packed_restore_last_avail_idx(VirtIODevice *vdev,
3347                                                        int n)
3348 {
3349     /* We don't have a reference like avail idx in shared memory */
3350     return;
3351 }
3352 
3353 static void virtio_queue_split_restore_last_avail_idx(VirtIODevice *vdev,
3354                                                       int n)
3355 {
3356     RCU_READ_LOCK_GUARD();
3357     if (vdev->vq[n].vring.desc) {
3358         vdev->vq[n].last_avail_idx = vring_used_idx(&vdev->vq[n]);
3359         vdev->vq[n].shadow_avail_idx = vdev->vq[n].last_avail_idx;
3360     }
3361 }
3362 
3363 void virtio_queue_restore_last_avail_idx(VirtIODevice *vdev, int n)
3364 {
3365     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3366         virtio_queue_packed_restore_last_avail_idx(vdev, n);
3367     } else {
3368         virtio_queue_split_restore_last_avail_idx(vdev, n);
3369     }
3370 }
3371 
3372 static void virtio_queue_packed_update_used_idx(VirtIODevice *vdev, int n)
3373 {
3374     /* used idx was updated through set_last_avail_idx() */
3375     return;
3376 }
3377 
3378 static void virtio_split_packed_update_used_idx(VirtIODevice *vdev, int n)
3379 {
3380     RCU_READ_LOCK_GUARD();
3381     if (vdev->vq[n].vring.desc) {
3382         vdev->vq[n].used_idx = vring_used_idx(&vdev->vq[n]);
3383     }
3384 }
3385 
3386 void virtio_queue_update_used_idx(VirtIODevice *vdev, int n)
3387 {
3388     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3389         return virtio_queue_packed_update_used_idx(vdev, n);
3390     } else {
3391         return virtio_split_packed_update_used_idx(vdev, n);
3392     }
3393 }
3394 
3395 void virtio_queue_invalidate_signalled_used(VirtIODevice *vdev, int n)
3396 {
3397     vdev->vq[n].signalled_used_valid = false;
3398 }
3399 
3400 VirtQueue *virtio_get_queue(VirtIODevice *vdev, int n)
3401 {
3402     return vdev->vq + n;
3403 }
3404 
3405 uint16_t virtio_get_queue_index(VirtQueue *vq)
3406 {
3407     return vq->queue_index;
3408 }
3409 
3410 static void virtio_queue_guest_notifier_read(EventNotifier *n)
3411 {
3412     VirtQueue *vq = container_of(n, VirtQueue, guest_notifier);
3413     if (event_notifier_test_and_clear(n)) {
3414         virtio_irq(vq);
3415     }
3416 }
3417 static void virtio_config_guest_notifier_read(EventNotifier *n)
3418 {
3419     VirtIODevice *vdev = container_of(n, VirtIODevice, config_notifier);
3420 
3421     if (event_notifier_test_and_clear(n)) {
3422         virtio_notify_config(vdev);
3423     }
3424 }
3425 void virtio_queue_set_guest_notifier_fd_handler(VirtQueue *vq, bool assign,
3426                                                 bool with_irqfd)
3427 {
3428     if (assign && !with_irqfd) {
3429         event_notifier_set_handler(&vq->guest_notifier,
3430                                    virtio_queue_guest_notifier_read);
3431     } else {
3432         event_notifier_set_handler(&vq->guest_notifier, NULL);
3433     }
3434     if (!assign) {
3435         /* Test and clear notifier before closing it,
3436          * in case poll callback didn't have time to run. */
3437         virtio_queue_guest_notifier_read(&vq->guest_notifier);
3438     }
3439 }
3440 
3441 void virtio_config_set_guest_notifier_fd_handler(VirtIODevice *vdev,
3442                                                  bool assign, bool with_irqfd)
3443 {
3444     EventNotifier *n;
3445     n = &vdev->config_notifier;
3446     if (assign && !with_irqfd) {
3447         event_notifier_set_handler(n, virtio_config_guest_notifier_read);
3448     } else {
3449         event_notifier_set_handler(n, NULL);
3450     }
3451     if (!assign) {
3452         /* Test and clear notifier before closing it,*/
3453         /* in case poll callback didn't have time to run. */
3454         virtio_config_guest_notifier_read(n);
3455     }
3456 }
3457 
3458 EventNotifier *virtio_queue_get_guest_notifier(VirtQueue *vq)
3459 {
3460     return &vq->guest_notifier;
3461 }
3462 
3463 static void virtio_queue_host_notifier_aio_poll_begin(EventNotifier *n)
3464 {
3465     VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
3466 
3467     virtio_queue_set_notification(vq, 0);
3468 }
3469 
3470 static bool virtio_queue_host_notifier_aio_poll(void *opaque)
3471 {
3472     EventNotifier *n = opaque;
3473     VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
3474 
3475     return vq->vring.desc && !virtio_queue_empty(vq);
3476 }
3477 
3478 static void virtio_queue_host_notifier_aio_poll_ready(EventNotifier *n)
3479 {
3480     VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
3481 
3482     virtio_queue_notify_vq(vq);
3483 }
3484 
3485 static void virtio_queue_host_notifier_aio_poll_end(EventNotifier *n)
3486 {
3487     VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
3488 
3489     /* Caller polls once more after this to catch requests that race with us */
3490     virtio_queue_set_notification(vq, 1);
3491 }
3492 
3493 void virtio_queue_aio_attach_host_notifier(VirtQueue *vq, AioContext *ctx)
3494 {
3495     aio_set_event_notifier(ctx, &vq->host_notifier, true,
3496                            virtio_queue_host_notifier_read,
3497                            virtio_queue_host_notifier_aio_poll,
3498                            virtio_queue_host_notifier_aio_poll_ready);
3499     aio_set_event_notifier_poll(ctx, &vq->host_notifier,
3500                                 virtio_queue_host_notifier_aio_poll_begin,
3501                                 virtio_queue_host_notifier_aio_poll_end);
3502 }
3503 
3504 /*
3505  * Same as virtio_queue_aio_attach_host_notifier() but without polling. Use
3506  * this for rx virtqueues and similar cases where the virtqueue handler
3507  * function does not pop all elements. When the virtqueue is left non-empty
3508  * polling consumes CPU cycles and should not be used.
3509  */
3510 void virtio_queue_aio_attach_host_notifier_no_poll(VirtQueue *vq, AioContext *ctx)
3511 {
3512     aio_set_event_notifier(ctx, &vq->host_notifier, true,
3513                            virtio_queue_host_notifier_read,
3514                            NULL, NULL);
3515 }
3516 
3517 void virtio_queue_aio_detach_host_notifier(VirtQueue *vq, AioContext *ctx)
3518 {
3519     aio_set_event_notifier(ctx, &vq->host_notifier, true, NULL, NULL, NULL);
3520     /* Test and clear notifier before after disabling event,
3521      * in case poll callback didn't have time to run. */
3522     virtio_queue_host_notifier_read(&vq->host_notifier);
3523 }
3524 
3525 void virtio_queue_host_notifier_read(EventNotifier *n)
3526 {
3527     VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
3528     if (event_notifier_test_and_clear(n)) {
3529         virtio_queue_notify_vq(vq);
3530     }
3531 }
3532 
3533 EventNotifier *virtio_queue_get_host_notifier(VirtQueue *vq)
3534 {
3535     return &vq->host_notifier;
3536 }
3537 
3538 EventNotifier *virtio_config_get_guest_notifier(VirtIODevice *vdev)
3539 {
3540     return &vdev->config_notifier;
3541 }
3542 
3543 void virtio_queue_set_host_notifier_enabled(VirtQueue *vq, bool enabled)
3544 {
3545     vq->host_notifier_enabled = enabled;
3546 }
3547 
3548 int virtio_queue_set_host_notifier_mr(VirtIODevice *vdev, int n,
3549                                       MemoryRegion *mr, bool assign)
3550 {
3551     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3552     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
3553 
3554     if (k->set_host_notifier_mr) {
3555         return k->set_host_notifier_mr(qbus->parent, n, mr, assign);
3556     }
3557 
3558     return -1;
3559 }
3560 
3561 void virtio_device_set_child_bus_name(VirtIODevice *vdev, char *bus_name)
3562 {
3563     g_free(vdev->bus_name);
3564     vdev->bus_name = g_strdup(bus_name);
3565 }
3566 
3567 void G_GNUC_PRINTF(2, 3) virtio_error(VirtIODevice *vdev, const char *fmt, ...)
3568 {
3569     va_list ap;
3570 
3571     va_start(ap, fmt);
3572     error_vreport(fmt, ap);
3573     va_end(ap);
3574 
3575     if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
3576         vdev->status = vdev->status | VIRTIO_CONFIG_S_NEEDS_RESET;
3577         virtio_notify_config(vdev);
3578     }
3579 
3580     vdev->broken = true;
3581 }
3582 
3583 static void virtio_memory_listener_commit(MemoryListener *listener)
3584 {
3585     VirtIODevice *vdev = container_of(listener, VirtIODevice, listener);
3586     int i;
3587 
3588     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
3589         if (vdev->vq[i].vring.num == 0) {
3590             break;
3591         }
3592         virtio_init_region_cache(vdev, i);
3593     }
3594 }
3595 
3596 static void virtio_device_realize(DeviceState *dev, Error **errp)
3597 {
3598     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
3599     VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(dev);
3600     Error *err = NULL;
3601 
3602     /* Devices should either use vmsd or the load/save methods */
3603     assert(!vdc->vmsd || !vdc->load);
3604 
3605     if (vdc->realize != NULL) {
3606         vdc->realize(dev, &err);
3607         if (err != NULL) {
3608             error_propagate(errp, err);
3609             return;
3610         }
3611     }
3612 
3613     virtio_bus_device_plugged(vdev, &err);
3614     if (err != NULL) {
3615         error_propagate(errp, err);
3616         vdc->unrealize(dev);
3617         return;
3618     }
3619 
3620     vdev->listener.commit = virtio_memory_listener_commit;
3621     vdev->listener.name = "virtio";
3622     memory_listener_register(&vdev->listener, vdev->dma_as);
3623     QTAILQ_INSERT_TAIL(&virtio_list, vdev, next);
3624 }
3625 
3626 static void virtio_device_unrealize(DeviceState *dev)
3627 {
3628     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
3629     VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(dev);
3630 
3631     memory_listener_unregister(&vdev->listener);
3632     virtio_bus_device_unplugged(vdev);
3633 
3634     if (vdc->unrealize != NULL) {
3635         vdc->unrealize(dev);
3636     }
3637 
3638     QTAILQ_REMOVE(&virtio_list, vdev, next);
3639     g_free(vdev->bus_name);
3640     vdev->bus_name = NULL;
3641 }
3642 
3643 static void virtio_device_free_virtqueues(VirtIODevice *vdev)
3644 {
3645     int i;
3646     if (!vdev->vq) {
3647         return;
3648     }
3649 
3650     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
3651         if (vdev->vq[i].vring.num == 0) {
3652             break;
3653         }
3654         virtio_virtqueue_reset_region_cache(&vdev->vq[i]);
3655     }
3656     g_free(vdev->vq);
3657 }
3658 
3659 static void virtio_device_instance_finalize(Object *obj)
3660 {
3661     VirtIODevice *vdev = VIRTIO_DEVICE(obj);
3662 
3663     virtio_device_free_virtqueues(vdev);
3664 
3665     g_free(vdev->config);
3666     g_free(vdev->vector_queues);
3667 }
3668 
3669 static Property virtio_properties[] = {
3670     DEFINE_VIRTIO_COMMON_FEATURES(VirtIODevice, host_features),
3671     DEFINE_PROP_BOOL("use-started", VirtIODevice, use_started, true),
3672     DEFINE_PROP_BOOL("use-disabled-flag", VirtIODevice, use_disabled_flag, true),
3673     DEFINE_PROP_BOOL("x-disable-legacy-check", VirtIODevice,
3674                      disable_legacy_check, false),
3675     DEFINE_PROP_END_OF_LIST(),
3676 };
3677 
3678 static int virtio_device_start_ioeventfd_impl(VirtIODevice *vdev)
3679 {
3680     VirtioBusState *qbus = VIRTIO_BUS(qdev_get_parent_bus(DEVICE(vdev)));
3681     int i, n, r, err;
3682 
3683     /*
3684      * Batch all the host notifiers in a single transaction to avoid
3685      * quadratic time complexity in address_space_update_ioeventfds().
3686      */
3687     memory_region_transaction_begin();
3688     for (n = 0; n < VIRTIO_QUEUE_MAX; n++) {
3689         VirtQueue *vq = &vdev->vq[n];
3690         if (!virtio_queue_get_num(vdev, n)) {
3691             continue;
3692         }
3693         r = virtio_bus_set_host_notifier(qbus, n, true);
3694         if (r < 0) {
3695             err = r;
3696             goto assign_error;
3697         }
3698         event_notifier_set_handler(&vq->host_notifier,
3699                                    virtio_queue_host_notifier_read);
3700     }
3701 
3702     for (n = 0; n < VIRTIO_QUEUE_MAX; n++) {
3703         /* Kick right away to begin processing requests already in vring */
3704         VirtQueue *vq = &vdev->vq[n];
3705         if (!vq->vring.num) {
3706             continue;
3707         }
3708         event_notifier_set(&vq->host_notifier);
3709     }
3710     memory_region_transaction_commit();
3711     return 0;
3712 
3713 assign_error:
3714     i = n; /* save n for a second iteration after transaction is committed. */
3715     while (--n >= 0) {
3716         VirtQueue *vq = &vdev->vq[n];
3717         if (!virtio_queue_get_num(vdev, n)) {
3718             continue;
3719         }
3720 
3721         event_notifier_set_handler(&vq->host_notifier, NULL);
3722         r = virtio_bus_set_host_notifier(qbus, n, false);
3723         assert(r >= 0);
3724     }
3725     /*
3726      * The transaction expects the ioeventfds to be open when it
3727      * commits. Do it now, before the cleanup loop.
3728      */
3729     memory_region_transaction_commit();
3730 
3731     while (--i >= 0) {
3732         if (!virtio_queue_get_num(vdev, i)) {
3733             continue;
3734         }
3735         virtio_bus_cleanup_host_notifier(qbus, i);
3736     }
3737     return err;
3738 }
3739 
3740 int virtio_device_start_ioeventfd(VirtIODevice *vdev)
3741 {
3742     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3743     VirtioBusState *vbus = VIRTIO_BUS(qbus);
3744 
3745     return virtio_bus_start_ioeventfd(vbus);
3746 }
3747 
3748 static void virtio_device_stop_ioeventfd_impl(VirtIODevice *vdev)
3749 {
3750     VirtioBusState *qbus = VIRTIO_BUS(qdev_get_parent_bus(DEVICE(vdev)));
3751     int n, r;
3752 
3753     /*
3754      * Batch all the host notifiers in a single transaction to avoid
3755      * quadratic time complexity in address_space_update_ioeventfds().
3756      */
3757     memory_region_transaction_begin();
3758     for (n = 0; n < VIRTIO_QUEUE_MAX; n++) {
3759         VirtQueue *vq = &vdev->vq[n];
3760 
3761         if (!virtio_queue_get_num(vdev, n)) {
3762             continue;
3763         }
3764         event_notifier_set_handler(&vq->host_notifier, NULL);
3765         r = virtio_bus_set_host_notifier(qbus, n, false);
3766         assert(r >= 0);
3767     }
3768     /*
3769      * The transaction expects the ioeventfds to be open when it
3770      * commits. Do it now, before the cleanup loop.
3771      */
3772     memory_region_transaction_commit();
3773 
3774     for (n = 0; n < VIRTIO_QUEUE_MAX; n++) {
3775         if (!virtio_queue_get_num(vdev, n)) {
3776             continue;
3777         }
3778         virtio_bus_cleanup_host_notifier(qbus, n);
3779     }
3780 }
3781 
3782 int virtio_device_grab_ioeventfd(VirtIODevice *vdev)
3783 {
3784     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3785     VirtioBusState *vbus = VIRTIO_BUS(qbus);
3786 
3787     return virtio_bus_grab_ioeventfd(vbus);
3788 }
3789 
3790 void virtio_device_release_ioeventfd(VirtIODevice *vdev)
3791 {
3792     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3793     VirtioBusState *vbus = VIRTIO_BUS(qbus);
3794 
3795     virtio_bus_release_ioeventfd(vbus);
3796 }
3797 
3798 static void virtio_device_class_init(ObjectClass *klass, void *data)
3799 {
3800     /* Set the default value here. */
3801     VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
3802     DeviceClass *dc = DEVICE_CLASS(klass);
3803 
3804     dc->realize = virtio_device_realize;
3805     dc->unrealize = virtio_device_unrealize;
3806     dc->bus_type = TYPE_VIRTIO_BUS;
3807     device_class_set_props(dc, virtio_properties);
3808     vdc->start_ioeventfd = virtio_device_start_ioeventfd_impl;
3809     vdc->stop_ioeventfd = virtio_device_stop_ioeventfd_impl;
3810 
3811     vdc->legacy_features |= VIRTIO_LEGACY_FEATURES;
3812 
3813     QTAILQ_INIT(&virtio_list);
3814 }
3815 
3816 bool virtio_device_ioeventfd_enabled(VirtIODevice *vdev)
3817 {
3818     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3819     VirtioBusState *vbus = VIRTIO_BUS(qbus);
3820 
3821     return virtio_bus_ioeventfd_enabled(vbus);
3822 }
3823 
3824 VirtQueueStatus *qmp_x_query_virtio_queue_status(const char *path,
3825                                                  uint16_t queue,
3826                                                  Error **errp)
3827 {
3828     VirtIODevice *vdev;
3829     VirtQueueStatus *status;
3830 
3831     vdev = qmp_find_virtio_device(path);
3832     if (vdev == NULL) {
3833         error_setg(errp, "Path %s is not a VirtIODevice", path);
3834         return NULL;
3835     }
3836 
3837     if (queue >= VIRTIO_QUEUE_MAX || !virtio_queue_get_num(vdev, queue)) {
3838         error_setg(errp, "Invalid virtqueue number %d", queue);
3839         return NULL;
3840     }
3841 
3842     status = g_new0(VirtQueueStatus, 1);
3843     status->name = g_strdup(vdev->name);
3844     status->queue_index = vdev->vq[queue].queue_index;
3845     status->inuse = vdev->vq[queue].inuse;
3846     status->vring_num = vdev->vq[queue].vring.num;
3847     status->vring_num_default = vdev->vq[queue].vring.num_default;
3848     status->vring_align = vdev->vq[queue].vring.align;
3849     status->vring_desc = vdev->vq[queue].vring.desc;
3850     status->vring_avail = vdev->vq[queue].vring.avail;
3851     status->vring_used = vdev->vq[queue].vring.used;
3852     status->used_idx = vdev->vq[queue].used_idx;
3853     status->signalled_used = vdev->vq[queue].signalled_used;
3854     status->signalled_used_valid = vdev->vq[queue].signalled_used_valid;
3855 
3856     if (vdev->vhost_started) {
3857         VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev);
3858         struct vhost_dev *hdev = vdc->get_vhost(vdev);
3859 
3860         /* check if vq index exists for vhost as well  */
3861         if (queue >= hdev->vq_index && queue < hdev->vq_index + hdev->nvqs) {
3862             status->has_last_avail_idx = true;
3863 
3864             int vhost_vq_index =
3865                 hdev->vhost_ops->vhost_get_vq_index(hdev, queue);
3866             struct vhost_vring_state state = {
3867                 .index = vhost_vq_index,
3868             };
3869 
3870             status->last_avail_idx =
3871                 hdev->vhost_ops->vhost_get_vring_base(hdev, &state);
3872         }
3873     } else {
3874         status->has_shadow_avail_idx = true;
3875         status->has_last_avail_idx = true;
3876         status->last_avail_idx = vdev->vq[queue].last_avail_idx;
3877         status->shadow_avail_idx = vdev->vq[queue].shadow_avail_idx;
3878     }
3879 
3880     return status;
3881 }
3882 
3883 static strList *qmp_decode_vring_desc_flags(uint16_t flags)
3884 {
3885     strList *list = NULL;
3886     strList *node;
3887     int i;
3888 
3889     struct {
3890         uint16_t flag;
3891         const char *value;
3892     } map[] = {
3893         { VRING_DESC_F_NEXT, "next" },
3894         { VRING_DESC_F_WRITE, "write" },
3895         { VRING_DESC_F_INDIRECT, "indirect" },
3896         { 1 << VRING_PACKED_DESC_F_AVAIL, "avail" },
3897         { 1 << VRING_PACKED_DESC_F_USED, "used" },
3898         { 0, "" }
3899     };
3900 
3901     for (i = 0; map[i].flag; i++) {
3902         if ((map[i].flag & flags) == 0) {
3903             continue;
3904         }
3905         node = g_malloc0(sizeof(strList));
3906         node->value = g_strdup(map[i].value);
3907         node->next = list;
3908         list = node;
3909     }
3910 
3911     return list;
3912 }
3913 
3914 VirtioQueueElement *qmp_x_query_virtio_queue_element(const char *path,
3915                                                      uint16_t queue,
3916                                                      bool has_index,
3917                                                      uint16_t index,
3918                                                      Error **errp)
3919 {
3920     VirtIODevice *vdev;
3921     VirtQueue *vq;
3922     VirtioQueueElement *element = NULL;
3923 
3924     vdev = qmp_find_virtio_device(path);
3925     if (vdev == NULL) {
3926         error_setg(errp, "Path %s is not a VirtIO device", path);
3927         return NULL;
3928     }
3929 
3930     if (queue >= VIRTIO_QUEUE_MAX || !virtio_queue_get_num(vdev, queue)) {
3931         error_setg(errp, "Invalid virtqueue number %d", queue);
3932         return NULL;
3933     }
3934     vq = &vdev->vq[queue];
3935 
3936     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3937         error_setg(errp, "Packed ring not supported");
3938         return NULL;
3939     } else {
3940         unsigned int head, i, max;
3941         VRingMemoryRegionCaches *caches;
3942         MemoryRegionCache indirect_desc_cache = MEMORY_REGION_CACHE_INVALID;
3943         MemoryRegionCache *desc_cache;
3944         VRingDesc desc;
3945         VirtioRingDescList *list = NULL;
3946         VirtioRingDescList *node;
3947         int rc; int ndescs;
3948 
3949         RCU_READ_LOCK_GUARD();
3950 
3951         max = vq->vring.num;
3952 
3953         if (!has_index) {
3954             head = vring_avail_ring(vq, vq->last_avail_idx % vq->vring.num);
3955         } else {
3956             head = vring_avail_ring(vq, index % vq->vring.num);
3957         }
3958         i = head;
3959 
3960         caches = vring_get_region_caches(vq);
3961         if (!caches) {
3962             error_setg(errp, "Region caches not initialized");
3963             return NULL;
3964         }
3965         if (caches->desc.len < max * sizeof(VRingDesc)) {
3966             error_setg(errp, "Cannot map descriptor ring");
3967             return NULL;
3968         }
3969 
3970         desc_cache = &caches->desc;
3971         vring_split_desc_read(vdev, &desc, desc_cache, i);
3972         if (desc.flags & VRING_DESC_F_INDIRECT) {
3973             int64_t len;
3974             len = address_space_cache_init(&indirect_desc_cache, vdev->dma_as,
3975                                            desc.addr, desc.len, false);
3976             desc_cache = &indirect_desc_cache;
3977             if (len < desc.len) {
3978                 error_setg(errp, "Cannot map indirect buffer");
3979                 goto done;
3980             }
3981 
3982             max = desc.len / sizeof(VRingDesc);
3983             i = 0;
3984             vring_split_desc_read(vdev, &desc, desc_cache, i);
3985         }
3986 
3987         element = g_new0(VirtioQueueElement, 1);
3988         element->avail = g_new0(VirtioRingAvail, 1);
3989         element->used = g_new0(VirtioRingUsed, 1);
3990         element->name = g_strdup(vdev->name);
3991         element->index = head;
3992         element->avail->flags = vring_avail_flags(vq);
3993         element->avail->idx = vring_avail_idx(vq);
3994         element->avail->ring = head;
3995         element->used->flags = vring_used_flags(vq);
3996         element->used->idx = vring_used_idx(vq);
3997         ndescs = 0;
3998 
3999         do {
4000             /* A buggy driver may produce an infinite loop */
4001             if (ndescs >= max) {
4002                 break;
4003             }
4004             node = g_new0(VirtioRingDescList, 1);
4005             node->value = g_new0(VirtioRingDesc, 1);
4006             node->value->addr = desc.addr;
4007             node->value->len = desc.len;
4008             node->value->flags = qmp_decode_vring_desc_flags(desc.flags);
4009             node->next = list;
4010             list = node;
4011 
4012             ndescs++;
4013             rc = virtqueue_split_read_next_desc(vdev, &desc, desc_cache,
4014                                                 max, &i);
4015         } while (rc == VIRTQUEUE_READ_DESC_MORE);
4016         element->descs = list;
4017 done:
4018         address_space_cache_destroy(&indirect_desc_cache);
4019     }
4020 
4021     return element;
4022 }
4023 
4024 static const TypeInfo virtio_device_info = {
4025     .name = TYPE_VIRTIO_DEVICE,
4026     .parent = TYPE_DEVICE,
4027     .instance_size = sizeof(VirtIODevice),
4028     .class_init = virtio_device_class_init,
4029     .instance_finalize = virtio_device_instance_finalize,
4030     .abstract = true,
4031     .class_size = sizeof(VirtioDeviceClass),
4032 };
4033 
4034 static void virtio_register_types(void)
4035 {
4036     type_register_static(&virtio_device_info);
4037 }
4038 
4039 type_init(virtio_register_types)
4040