xref: /openbmc/qemu/hw/virtio/virtio.c (revision 93977402)
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 "qemu-common.h"
17 #include "cpu.h"
18 #include "trace.h"
19 #include "exec/address-spaces.h"
20 #include "qemu/error-report.h"
21 #include "hw/virtio/virtio.h"
22 #include "qemu/atomic.h"
23 #include "hw/virtio/virtio-bus.h"
24 #include "migration/migration.h"
25 #include "hw/virtio/virtio-access.h"
26 
27 /*
28  * The alignment to use between consumer and producer parts of vring.
29  * x86 pagesize again. This is the default, used by transports like PCI
30  * which don't provide a means for the guest to tell the host the alignment.
31  */
32 #define VIRTIO_PCI_VRING_ALIGN         4096
33 
34 typedef struct VRingDesc
35 {
36     uint64_t addr;
37     uint32_t len;
38     uint16_t flags;
39     uint16_t next;
40 } VRingDesc;
41 
42 typedef struct VRingAvail
43 {
44     uint16_t flags;
45     uint16_t idx;
46     uint16_t ring[0];
47 } VRingAvail;
48 
49 typedef struct VRingUsedElem
50 {
51     uint32_t id;
52     uint32_t len;
53 } VRingUsedElem;
54 
55 typedef struct VRingUsed
56 {
57     uint16_t flags;
58     uint16_t idx;
59     VRingUsedElem ring[0];
60 } VRingUsed;
61 
62 typedef struct VRing
63 {
64     unsigned int num;
65     unsigned int num_default;
66     unsigned int align;
67     hwaddr desc;
68     hwaddr avail;
69     hwaddr used;
70 } VRing;
71 
72 struct VirtQueue
73 {
74     VRing vring;
75 
76     /* Next head to pop */
77     uint16_t last_avail_idx;
78 
79     /* Last avail_idx read from VQ. */
80     uint16_t shadow_avail_idx;
81 
82     uint16_t used_idx;
83 
84     /* Last used index value we have signalled on */
85     uint16_t signalled_used;
86 
87     /* Last used index value we have signalled on */
88     bool signalled_used_valid;
89 
90     /* Notification enabled? */
91     bool notification;
92 
93     uint16_t queue_index;
94 
95     int inuse;
96 
97     uint16_t vector;
98     VirtIOHandleOutput handle_output;
99     VirtIOHandleOutput handle_aio_output;
100     bool use_aio;
101     VirtIODevice *vdev;
102     EventNotifier guest_notifier;
103     EventNotifier host_notifier;
104     QLIST_ENTRY(VirtQueue) node;
105 };
106 
107 /* virt queue functions */
108 void virtio_queue_update_rings(VirtIODevice *vdev, int n)
109 {
110     VRing *vring = &vdev->vq[n].vring;
111 
112     if (!vring->desc) {
113         /* not yet setup -> nothing to do */
114         return;
115     }
116     vring->avail = vring->desc + vring->num * sizeof(VRingDesc);
117     vring->used = vring_align(vring->avail +
118                               offsetof(VRingAvail, ring[vring->num]),
119                               vring->align);
120 }
121 
122 static void vring_desc_read(VirtIODevice *vdev, VRingDesc *desc,
123                             hwaddr desc_pa, int i)
124 {
125     address_space_read(&address_space_memory, desc_pa + i * sizeof(VRingDesc),
126                        MEMTXATTRS_UNSPECIFIED, (void *)desc, sizeof(VRingDesc));
127     virtio_tswap64s(vdev, &desc->addr);
128     virtio_tswap32s(vdev, &desc->len);
129     virtio_tswap16s(vdev, &desc->flags);
130     virtio_tswap16s(vdev, &desc->next);
131 }
132 
133 static inline uint16_t vring_avail_flags(VirtQueue *vq)
134 {
135     hwaddr pa;
136     pa = vq->vring.avail + offsetof(VRingAvail, flags);
137     return virtio_lduw_phys(vq->vdev, pa);
138 }
139 
140 static inline uint16_t vring_avail_idx(VirtQueue *vq)
141 {
142     hwaddr pa;
143     pa = vq->vring.avail + offsetof(VRingAvail, idx);
144     vq->shadow_avail_idx = virtio_lduw_phys(vq->vdev, pa);
145     return vq->shadow_avail_idx;
146 }
147 
148 static inline uint16_t vring_avail_ring(VirtQueue *vq, int i)
149 {
150     hwaddr pa;
151     pa = vq->vring.avail + offsetof(VRingAvail, ring[i]);
152     return virtio_lduw_phys(vq->vdev, pa);
153 }
154 
155 static inline uint16_t vring_get_used_event(VirtQueue *vq)
156 {
157     return vring_avail_ring(vq, vq->vring.num);
158 }
159 
160 static inline void vring_used_write(VirtQueue *vq, VRingUsedElem *uelem,
161                                     int i)
162 {
163     hwaddr pa;
164     virtio_tswap32s(vq->vdev, &uelem->id);
165     virtio_tswap32s(vq->vdev, &uelem->len);
166     pa = vq->vring.used + offsetof(VRingUsed, ring[i]);
167     address_space_write(&address_space_memory, pa, MEMTXATTRS_UNSPECIFIED,
168                        (void *)uelem, sizeof(VRingUsedElem));
169 }
170 
171 static uint16_t vring_used_idx(VirtQueue *vq)
172 {
173     hwaddr pa;
174     pa = vq->vring.used + offsetof(VRingUsed, idx);
175     return virtio_lduw_phys(vq->vdev, pa);
176 }
177 
178 static inline void vring_used_idx_set(VirtQueue *vq, uint16_t val)
179 {
180     hwaddr pa;
181     pa = vq->vring.used + offsetof(VRingUsed, idx);
182     virtio_stw_phys(vq->vdev, pa, val);
183     vq->used_idx = val;
184 }
185 
186 static inline void vring_used_flags_set_bit(VirtQueue *vq, int mask)
187 {
188     VirtIODevice *vdev = vq->vdev;
189     hwaddr pa;
190     pa = vq->vring.used + offsetof(VRingUsed, flags);
191     virtio_stw_phys(vdev, pa, virtio_lduw_phys(vdev, pa) | mask);
192 }
193 
194 static inline void vring_used_flags_unset_bit(VirtQueue *vq, int mask)
195 {
196     VirtIODevice *vdev = vq->vdev;
197     hwaddr pa;
198     pa = vq->vring.used + offsetof(VRingUsed, flags);
199     virtio_stw_phys(vdev, pa, virtio_lduw_phys(vdev, pa) & ~mask);
200 }
201 
202 static inline void vring_set_avail_event(VirtQueue *vq, uint16_t val)
203 {
204     hwaddr pa;
205     if (!vq->notification) {
206         return;
207     }
208     pa = vq->vring.used + offsetof(VRingUsed, ring[vq->vring.num]);
209     virtio_stw_phys(vq->vdev, pa, val);
210 }
211 
212 void virtio_queue_set_notification(VirtQueue *vq, int enable)
213 {
214     vq->notification = enable;
215     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_RING_F_EVENT_IDX)) {
216         vring_set_avail_event(vq, vring_avail_idx(vq));
217     } else if (enable) {
218         vring_used_flags_unset_bit(vq, VRING_USED_F_NO_NOTIFY);
219     } else {
220         vring_used_flags_set_bit(vq, VRING_USED_F_NO_NOTIFY);
221     }
222     if (enable) {
223         /* Expose avail event/used flags before caller checks the avail idx. */
224         smp_mb();
225     }
226 }
227 
228 int virtio_queue_ready(VirtQueue *vq)
229 {
230     return vq->vring.avail != 0;
231 }
232 
233 /* Fetch avail_idx from VQ memory only when we really need to know if
234  * guest has added some buffers. */
235 int virtio_queue_empty(VirtQueue *vq)
236 {
237     if (vq->shadow_avail_idx != vq->last_avail_idx) {
238         return 0;
239     }
240 
241     return vring_avail_idx(vq) == vq->last_avail_idx;
242 }
243 
244 static void virtqueue_unmap_sg(VirtQueue *vq, const VirtQueueElement *elem,
245                                unsigned int len)
246 {
247     unsigned int offset;
248     int i;
249 
250     offset = 0;
251     for (i = 0; i < elem->in_num; i++) {
252         size_t size = MIN(len - offset, elem->in_sg[i].iov_len);
253 
254         cpu_physical_memory_unmap(elem->in_sg[i].iov_base,
255                                   elem->in_sg[i].iov_len,
256                                   1, size);
257 
258         offset += size;
259     }
260 
261     for (i = 0; i < elem->out_num; i++)
262         cpu_physical_memory_unmap(elem->out_sg[i].iov_base,
263                                   elem->out_sg[i].iov_len,
264                                   0, elem->out_sg[i].iov_len);
265 }
266 
267 /* virtqueue_detach_element:
268  * @vq: The #VirtQueue
269  * @elem: The #VirtQueueElement
270  * @len: number of bytes written
271  *
272  * Detach the element from the virtqueue.  This function is suitable for device
273  * reset or other situations where a #VirtQueueElement is simply freed and will
274  * not be pushed or discarded.
275  */
276 void virtqueue_detach_element(VirtQueue *vq, const VirtQueueElement *elem,
277                               unsigned int len)
278 {
279     vq->inuse--;
280     virtqueue_unmap_sg(vq, elem, len);
281 }
282 
283 /* virtqueue_discard:
284  * @vq: The #VirtQueue
285  * @elem: The #VirtQueueElement
286  * @len: number of bytes written
287  *
288  * Pretend the most recent element wasn't popped from the virtqueue.  The next
289  * call to virtqueue_pop() will refetch the element.
290  */
291 void virtqueue_discard(VirtQueue *vq, const VirtQueueElement *elem,
292                        unsigned int len)
293 {
294     vq->last_avail_idx--;
295     virtqueue_detach_element(vq, elem, len);
296 }
297 
298 /* virtqueue_rewind:
299  * @vq: The #VirtQueue
300  * @num: Number of elements to push back
301  *
302  * Pretend that elements weren't popped from the virtqueue.  The next
303  * virtqueue_pop() will refetch the oldest element.
304  *
305  * Use virtqueue_discard() instead if you have a VirtQueueElement.
306  *
307  * Returns: true on success, false if @num is greater than the number of in use
308  * elements.
309  */
310 bool virtqueue_rewind(VirtQueue *vq, unsigned int num)
311 {
312     if (num > vq->inuse) {
313         return false;
314     }
315     vq->last_avail_idx -= num;
316     vq->inuse -= num;
317     return true;
318 }
319 
320 void virtqueue_fill(VirtQueue *vq, const VirtQueueElement *elem,
321                     unsigned int len, unsigned int idx)
322 {
323     VRingUsedElem uelem;
324 
325     trace_virtqueue_fill(vq, elem, len, idx);
326 
327     virtqueue_unmap_sg(vq, elem, len);
328 
329     if (unlikely(vq->vdev->broken)) {
330         return;
331     }
332 
333     idx = (idx + vq->used_idx) % vq->vring.num;
334 
335     uelem.id = elem->index;
336     uelem.len = len;
337     vring_used_write(vq, &uelem, idx);
338 }
339 
340 void virtqueue_flush(VirtQueue *vq, unsigned int count)
341 {
342     uint16_t old, new;
343 
344     if (unlikely(vq->vdev->broken)) {
345         vq->inuse -= count;
346         return;
347     }
348 
349     /* Make sure buffer is written before we update index. */
350     smp_wmb();
351     trace_virtqueue_flush(vq, count);
352     old = vq->used_idx;
353     new = old + count;
354     vring_used_idx_set(vq, new);
355     vq->inuse -= count;
356     if (unlikely((int16_t)(new - vq->signalled_used) < (uint16_t)(new - old)))
357         vq->signalled_used_valid = false;
358 }
359 
360 void virtqueue_push(VirtQueue *vq, const VirtQueueElement *elem,
361                     unsigned int len)
362 {
363     virtqueue_fill(vq, elem, len, 0);
364     virtqueue_flush(vq, 1);
365 }
366 
367 static int virtqueue_num_heads(VirtQueue *vq, unsigned int idx)
368 {
369     uint16_t num_heads = vring_avail_idx(vq) - idx;
370 
371     /* Check it isn't doing very strange things with descriptor numbers. */
372     if (num_heads > vq->vring.num) {
373         virtio_error(vq->vdev, "Guest moved used index from %u to %u",
374                      idx, vq->shadow_avail_idx);
375         return -EINVAL;
376     }
377     /* On success, callers read a descriptor at vq->last_avail_idx.
378      * Make sure descriptor read does not bypass avail index read. */
379     if (num_heads) {
380         smp_rmb();
381     }
382 
383     return num_heads;
384 }
385 
386 static bool virtqueue_get_head(VirtQueue *vq, unsigned int idx,
387                                unsigned int *head)
388 {
389     /* Grab the next descriptor number they're advertising, and increment
390      * the index we've seen. */
391     *head = vring_avail_ring(vq, idx % vq->vring.num);
392 
393     /* If their number is silly, that's a fatal mistake. */
394     if (*head >= vq->vring.num) {
395         virtio_error(vq->vdev, "Guest says index %u is available", *head);
396         return false;
397     }
398 
399     return true;
400 }
401 
402 enum {
403     VIRTQUEUE_READ_DESC_ERROR = -1,
404     VIRTQUEUE_READ_DESC_DONE = 0,   /* end of chain */
405     VIRTQUEUE_READ_DESC_MORE = 1,   /* more buffers in chain */
406 };
407 
408 static int virtqueue_read_next_desc(VirtIODevice *vdev, VRingDesc *desc,
409                                     hwaddr desc_pa, unsigned int max,
410                                     unsigned int *next)
411 {
412     /* If this descriptor says it doesn't chain, we're done. */
413     if (!(desc->flags & VRING_DESC_F_NEXT)) {
414         return VIRTQUEUE_READ_DESC_DONE;
415     }
416 
417     /* Check they're not leading us off end of descriptors. */
418     *next = desc->next;
419     /* Make sure compiler knows to grab that: we don't want it changing! */
420     smp_wmb();
421 
422     if (*next >= max) {
423         virtio_error(vdev, "Desc next is %u", *next);
424         return VIRTQUEUE_READ_DESC_ERROR;
425     }
426 
427     vring_desc_read(vdev, desc, desc_pa, *next);
428     return VIRTQUEUE_READ_DESC_MORE;
429 }
430 
431 void virtqueue_get_avail_bytes(VirtQueue *vq, unsigned int *in_bytes,
432                                unsigned int *out_bytes,
433                                unsigned max_in_bytes, unsigned max_out_bytes)
434 {
435     unsigned int idx;
436     unsigned int total_bufs, in_total, out_total;
437     int rc;
438 
439     idx = vq->last_avail_idx;
440 
441     total_bufs = in_total = out_total = 0;
442     while ((rc = virtqueue_num_heads(vq, idx)) > 0) {
443         VirtIODevice *vdev = vq->vdev;
444         unsigned int max, num_bufs, indirect = 0;
445         VRingDesc desc;
446         hwaddr desc_pa;
447         unsigned int i;
448 
449         max = vq->vring.num;
450         num_bufs = total_bufs;
451 
452         if (!virtqueue_get_head(vq, idx++, &i)) {
453             goto err;
454         }
455 
456         desc_pa = vq->vring.desc;
457         vring_desc_read(vdev, &desc, desc_pa, i);
458 
459         if (desc.flags & VRING_DESC_F_INDIRECT) {
460             if (desc.len % sizeof(VRingDesc)) {
461                 virtio_error(vdev, "Invalid size for indirect buffer table");
462                 goto err;
463             }
464 
465             /* If we've got too many, that implies a descriptor loop. */
466             if (num_bufs >= max) {
467                 virtio_error(vdev, "Looped descriptor");
468                 goto err;
469             }
470 
471             /* loop over the indirect descriptor table */
472             indirect = 1;
473             max = desc.len / sizeof(VRingDesc);
474             desc_pa = desc.addr;
475             num_bufs = i = 0;
476             vring_desc_read(vdev, &desc, desc_pa, i);
477         }
478 
479         do {
480             /* If we've got too many, that implies a descriptor loop. */
481             if (++num_bufs > max) {
482                 virtio_error(vdev, "Looped descriptor");
483                 goto err;
484             }
485 
486             if (desc.flags & VRING_DESC_F_WRITE) {
487                 in_total += desc.len;
488             } else {
489                 out_total += desc.len;
490             }
491             if (in_total >= max_in_bytes && out_total >= max_out_bytes) {
492                 goto done;
493             }
494 
495             rc = virtqueue_read_next_desc(vdev, &desc, desc_pa, max, &i);
496         } while (rc == VIRTQUEUE_READ_DESC_MORE);
497 
498         if (rc == VIRTQUEUE_READ_DESC_ERROR) {
499             goto err;
500         }
501 
502         if (!indirect)
503             total_bufs = num_bufs;
504         else
505             total_bufs++;
506     }
507 
508     if (rc < 0) {
509         goto err;
510     }
511 
512 done:
513     if (in_bytes) {
514         *in_bytes = in_total;
515     }
516     if (out_bytes) {
517         *out_bytes = out_total;
518     }
519     return;
520 
521 err:
522     in_total = out_total = 0;
523     goto done;
524 }
525 
526 int virtqueue_avail_bytes(VirtQueue *vq, unsigned int in_bytes,
527                           unsigned int out_bytes)
528 {
529     unsigned int in_total, out_total;
530 
531     virtqueue_get_avail_bytes(vq, &in_total, &out_total, in_bytes, out_bytes);
532     return in_bytes <= in_total && out_bytes <= out_total;
533 }
534 
535 static bool virtqueue_map_desc(VirtIODevice *vdev, unsigned int *p_num_sg,
536                                hwaddr *addr, struct iovec *iov,
537                                unsigned int max_num_sg, bool is_write,
538                                hwaddr pa, size_t sz)
539 {
540     bool ok = false;
541     unsigned num_sg = *p_num_sg;
542     assert(num_sg <= max_num_sg);
543 
544     if (!sz) {
545         virtio_error(vdev, "virtio: zero sized buffers are not allowed");
546         goto out;
547     }
548 
549     while (sz) {
550         hwaddr len = sz;
551 
552         if (num_sg == max_num_sg) {
553             virtio_error(vdev, "virtio: too many write descriptors in "
554                                "indirect table");
555             goto out;
556         }
557 
558         iov[num_sg].iov_base = cpu_physical_memory_map(pa, &len, is_write);
559         if (!iov[num_sg].iov_base) {
560             virtio_error(vdev, "virtio: bogus descriptor or out of resources");
561             goto out;
562         }
563 
564         iov[num_sg].iov_len = len;
565         addr[num_sg] = pa;
566 
567         sz -= len;
568         pa += len;
569         num_sg++;
570     }
571     ok = true;
572 
573 out:
574     *p_num_sg = num_sg;
575     return ok;
576 }
577 
578 /* Only used by error code paths before we have a VirtQueueElement (therefore
579  * virtqueue_unmap_sg() can't be used).  Assumes buffers weren't written to
580  * yet.
581  */
582 static void virtqueue_undo_map_desc(unsigned int out_num, unsigned int in_num,
583                                     struct iovec *iov)
584 {
585     unsigned int i;
586 
587     for (i = 0; i < out_num + in_num; i++) {
588         int is_write = i >= out_num;
589 
590         cpu_physical_memory_unmap(iov->iov_base, iov->iov_len, is_write, 0);
591         iov++;
592     }
593 }
594 
595 static void virtqueue_map_iovec(struct iovec *sg, hwaddr *addr,
596                                 unsigned int *num_sg, unsigned int max_size,
597                                 int is_write)
598 {
599     unsigned int i;
600     hwaddr len;
601 
602     /* Note: this function MUST validate input, some callers
603      * are passing in num_sg values received over the network.
604      */
605     /* TODO: teach all callers that this can fail, and return failure instead
606      * of asserting here.
607      * When we do, we might be able to re-enable NDEBUG below.
608      */
609 #ifdef NDEBUG
610 #error building with NDEBUG is not supported
611 #endif
612     assert(*num_sg <= max_size);
613 
614     for (i = 0; i < *num_sg; i++) {
615         len = sg[i].iov_len;
616         sg[i].iov_base = cpu_physical_memory_map(addr[i], &len, is_write);
617         if (!sg[i].iov_base) {
618             error_report("virtio: error trying to map MMIO memory");
619             exit(1);
620         }
621         if (len != sg[i].iov_len) {
622             error_report("virtio: unexpected memory split");
623             exit(1);
624         }
625     }
626 }
627 
628 void virtqueue_map(VirtQueueElement *elem)
629 {
630     virtqueue_map_iovec(elem->in_sg, elem->in_addr, &elem->in_num,
631                         VIRTQUEUE_MAX_SIZE, 1);
632     virtqueue_map_iovec(elem->out_sg, elem->out_addr, &elem->out_num,
633                         VIRTQUEUE_MAX_SIZE, 0);
634 }
635 
636 void *virtqueue_alloc_element(size_t sz, unsigned out_num, unsigned in_num)
637 {
638     VirtQueueElement *elem;
639     size_t in_addr_ofs = QEMU_ALIGN_UP(sz, __alignof__(elem->in_addr[0]));
640     size_t out_addr_ofs = in_addr_ofs + in_num * sizeof(elem->in_addr[0]);
641     size_t out_addr_end = out_addr_ofs + out_num * sizeof(elem->out_addr[0]);
642     size_t in_sg_ofs = QEMU_ALIGN_UP(out_addr_end, __alignof__(elem->in_sg[0]));
643     size_t out_sg_ofs = in_sg_ofs + in_num * sizeof(elem->in_sg[0]);
644     size_t out_sg_end = out_sg_ofs + out_num * sizeof(elem->out_sg[0]);
645 
646     assert(sz >= sizeof(VirtQueueElement));
647     elem = g_malloc(out_sg_end);
648     elem->out_num = out_num;
649     elem->in_num = in_num;
650     elem->in_addr = (void *)elem + in_addr_ofs;
651     elem->out_addr = (void *)elem + out_addr_ofs;
652     elem->in_sg = (void *)elem + in_sg_ofs;
653     elem->out_sg = (void *)elem + out_sg_ofs;
654     return elem;
655 }
656 
657 void *virtqueue_pop(VirtQueue *vq, size_t sz)
658 {
659     unsigned int i, head, max;
660     hwaddr desc_pa = vq->vring.desc;
661     VirtIODevice *vdev = vq->vdev;
662     VirtQueueElement *elem;
663     unsigned out_num, in_num;
664     hwaddr addr[VIRTQUEUE_MAX_SIZE];
665     struct iovec iov[VIRTQUEUE_MAX_SIZE];
666     VRingDesc desc;
667     int rc;
668 
669     if (unlikely(vdev->broken)) {
670         return NULL;
671     }
672     if (virtio_queue_empty(vq)) {
673         return NULL;
674     }
675     /* Needed after virtio_queue_empty(), see comment in
676      * virtqueue_num_heads(). */
677     smp_rmb();
678 
679     /* When we start there are none of either input nor output. */
680     out_num = in_num = 0;
681 
682     max = vq->vring.num;
683 
684     if (vq->inuse >= vq->vring.num) {
685         virtio_error(vdev, "Virtqueue size exceeded");
686         return NULL;
687     }
688 
689     if (!virtqueue_get_head(vq, vq->last_avail_idx++, &head)) {
690         return NULL;
691     }
692 
693     if (virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX)) {
694         vring_set_avail_event(vq, vq->last_avail_idx);
695     }
696 
697     i = head;
698     vring_desc_read(vdev, &desc, desc_pa, i);
699     if (desc.flags & VRING_DESC_F_INDIRECT) {
700         if (desc.len % sizeof(VRingDesc)) {
701             virtio_error(vdev, "Invalid size for indirect buffer table");
702             return NULL;
703         }
704 
705         /* loop over the indirect descriptor table */
706         max = desc.len / sizeof(VRingDesc);
707         desc_pa = desc.addr;
708         i = 0;
709         vring_desc_read(vdev, &desc, desc_pa, i);
710     }
711 
712     /* Collect all the descriptors */
713     do {
714         bool map_ok;
715 
716         if (desc.flags & VRING_DESC_F_WRITE) {
717             map_ok = virtqueue_map_desc(vdev, &in_num, addr + out_num,
718                                         iov + out_num,
719                                         VIRTQUEUE_MAX_SIZE - out_num, true,
720                                         desc.addr, desc.len);
721         } else {
722             if (in_num) {
723                 virtio_error(vdev, "Incorrect order for descriptors");
724                 goto err_undo_map;
725             }
726             map_ok = virtqueue_map_desc(vdev, &out_num, addr, iov,
727                                         VIRTQUEUE_MAX_SIZE, false,
728                                         desc.addr, desc.len);
729         }
730         if (!map_ok) {
731             goto err_undo_map;
732         }
733 
734         /* If we've got too many, that implies a descriptor loop. */
735         if ((in_num + out_num) > max) {
736             virtio_error(vdev, "Looped descriptor");
737             goto err_undo_map;
738         }
739 
740         rc = virtqueue_read_next_desc(vdev, &desc, desc_pa, max, &i);
741     } while (rc == VIRTQUEUE_READ_DESC_MORE);
742 
743     if (rc == VIRTQUEUE_READ_DESC_ERROR) {
744         goto err_undo_map;
745     }
746 
747     /* Now copy what we have collected and mapped */
748     elem = virtqueue_alloc_element(sz, out_num, in_num);
749     elem->index = head;
750     for (i = 0; i < out_num; i++) {
751         elem->out_addr[i] = addr[i];
752         elem->out_sg[i] = iov[i];
753     }
754     for (i = 0; i < in_num; i++) {
755         elem->in_addr[i] = addr[out_num + i];
756         elem->in_sg[i] = iov[out_num + i];
757     }
758 
759     vq->inuse++;
760 
761     trace_virtqueue_pop(vq, elem, elem->in_num, elem->out_num);
762     return elem;
763 
764 err_undo_map:
765     virtqueue_undo_map_desc(out_num, in_num, iov);
766     return NULL;
767 }
768 
769 /* Reading and writing a structure directly to QEMUFile is *awful*, but
770  * it is what QEMU has always done by mistake.  We can change it sooner
771  * or later by bumping the version number of the affected vm states.
772  * In the meanwhile, since the in-memory layout of VirtQueueElement
773  * has changed, we need to marshal to and from the layout that was
774  * used before the change.
775  */
776 typedef struct VirtQueueElementOld {
777     unsigned int index;
778     unsigned int out_num;
779     unsigned int in_num;
780     hwaddr in_addr[VIRTQUEUE_MAX_SIZE];
781     hwaddr out_addr[VIRTQUEUE_MAX_SIZE];
782     struct iovec in_sg[VIRTQUEUE_MAX_SIZE];
783     struct iovec out_sg[VIRTQUEUE_MAX_SIZE];
784 } VirtQueueElementOld;
785 
786 void *qemu_get_virtqueue_element(QEMUFile *f, size_t sz)
787 {
788     VirtQueueElement *elem;
789     VirtQueueElementOld data;
790     int i;
791 
792     qemu_get_buffer(f, (uint8_t *)&data, sizeof(VirtQueueElementOld));
793 
794     elem = virtqueue_alloc_element(sz, data.out_num, data.in_num);
795     elem->index = data.index;
796 
797     for (i = 0; i < elem->in_num; i++) {
798         elem->in_addr[i] = data.in_addr[i];
799     }
800 
801     for (i = 0; i < elem->out_num; i++) {
802         elem->out_addr[i] = data.out_addr[i];
803     }
804 
805     for (i = 0; i < elem->in_num; i++) {
806         /* Base is overwritten by virtqueue_map.  */
807         elem->in_sg[i].iov_base = 0;
808         elem->in_sg[i].iov_len = data.in_sg[i].iov_len;
809     }
810 
811     for (i = 0; i < elem->out_num; i++) {
812         /* Base is overwritten by virtqueue_map.  */
813         elem->out_sg[i].iov_base = 0;
814         elem->out_sg[i].iov_len = data.out_sg[i].iov_len;
815     }
816 
817     virtqueue_map(elem);
818     return elem;
819 }
820 
821 void qemu_put_virtqueue_element(QEMUFile *f, VirtQueueElement *elem)
822 {
823     VirtQueueElementOld data;
824     int i;
825 
826     memset(&data, 0, sizeof(data));
827     data.index = elem->index;
828     data.in_num = elem->in_num;
829     data.out_num = elem->out_num;
830 
831     for (i = 0; i < elem->in_num; i++) {
832         data.in_addr[i] = elem->in_addr[i];
833     }
834 
835     for (i = 0; i < elem->out_num; i++) {
836         data.out_addr[i] = elem->out_addr[i];
837     }
838 
839     for (i = 0; i < elem->in_num; i++) {
840         /* Base is overwritten by virtqueue_map when loading.  Do not
841          * save it, as it would leak the QEMU address space layout.  */
842         data.in_sg[i].iov_len = elem->in_sg[i].iov_len;
843     }
844 
845     for (i = 0; i < elem->out_num; i++) {
846         /* Do not save iov_base as above.  */
847         data.out_sg[i].iov_len = elem->out_sg[i].iov_len;
848     }
849     qemu_put_buffer(f, (uint8_t *)&data, sizeof(VirtQueueElementOld));
850 }
851 
852 /* virtio device */
853 static void virtio_notify_vector(VirtIODevice *vdev, uint16_t vector)
854 {
855     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
856     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
857 
858     if (unlikely(vdev->broken)) {
859         return;
860     }
861 
862     if (k->notify) {
863         k->notify(qbus->parent, vector);
864     }
865 }
866 
867 void virtio_update_irq(VirtIODevice *vdev)
868 {
869     virtio_notify_vector(vdev, VIRTIO_NO_VECTOR);
870 }
871 
872 static int virtio_validate_features(VirtIODevice *vdev)
873 {
874     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
875 
876     if (k->validate_features) {
877         return k->validate_features(vdev);
878     } else {
879         return 0;
880     }
881 }
882 
883 int virtio_set_status(VirtIODevice *vdev, uint8_t val)
884 {
885     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
886     trace_virtio_set_status(vdev, val);
887 
888     if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
889         if (!(vdev->status & VIRTIO_CONFIG_S_FEATURES_OK) &&
890             val & VIRTIO_CONFIG_S_FEATURES_OK) {
891             int ret = virtio_validate_features(vdev);
892 
893             if (ret) {
894                 return ret;
895             }
896         }
897     }
898     if (k->set_status) {
899         k->set_status(vdev, val);
900     }
901     vdev->status = val;
902     return 0;
903 }
904 
905 bool target_words_bigendian(void);
906 static enum virtio_device_endian virtio_default_endian(void)
907 {
908     if (target_words_bigendian()) {
909         return VIRTIO_DEVICE_ENDIAN_BIG;
910     } else {
911         return VIRTIO_DEVICE_ENDIAN_LITTLE;
912     }
913 }
914 
915 static enum virtio_device_endian virtio_current_cpu_endian(void)
916 {
917     CPUClass *cc = CPU_GET_CLASS(current_cpu);
918 
919     if (cc->virtio_is_big_endian(current_cpu)) {
920         return VIRTIO_DEVICE_ENDIAN_BIG;
921     } else {
922         return VIRTIO_DEVICE_ENDIAN_LITTLE;
923     }
924 }
925 
926 void virtio_reset(void *opaque)
927 {
928     VirtIODevice *vdev = opaque;
929     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
930     int i;
931 
932     virtio_set_status(vdev, 0);
933     if (current_cpu) {
934         /* Guest initiated reset */
935         vdev->device_endian = virtio_current_cpu_endian();
936     } else {
937         /* System reset */
938         vdev->device_endian = virtio_default_endian();
939     }
940 
941     if (k->reset) {
942         k->reset(vdev);
943     }
944 
945     vdev->broken = false;
946     vdev->guest_features = 0;
947     vdev->queue_sel = 0;
948     vdev->status = 0;
949     vdev->isr = 0;
950     vdev->config_vector = VIRTIO_NO_VECTOR;
951     virtio_notify_vector(vdev, vdev->config_vector);
952 
953     for(i = 0; i < VIRTIO_QUEUE_MAX; i++) {
954         vdev->vq[i].vring.desc = 0;
955         vdev->vq[i].vring.avail = 0;
956         vdev->vq[i].vring.used = 0;
957         vdev->vq[i].last_avail_idx = 0;
958         vdev->vq[i].shadow_avail_idx = 0;
959         vdev->vq[i].used_idx = 0;
960         virtio_queue_set_vector(vdev, i, VIRTIO_NO_VECTOR);
961         vdev->vq[i].signalled_used = 0;
962         vdev->vq[i].signalled_used_valid = false;
963         vdev->vq[i].notification = true;
964         vdev->vq[i].vring.num = vdev->vq[i].vring.num_default;
965         vdev->vq[i].inuse = 0;
966     }
967 }
968 
969 uint32_t virtio_config_readb(VirtIODevice *vdev, uint32_t addr)
970 {
971     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
972     uint8_t val;
973 
974     if (addr + sizeof(val) > vdev->config_len) {
975         return (uint32_t)-1;
976     }
977 
978     k->get_config(vdev, vdev->config);
979 
980     val = ldub_p(vdev->config + addr);
981     return val;
982 }
983 
984 uint32_t virtio_config_readw(VirtIODevice *vdev, uint32_t addr)
985 {
986     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
987     uint16_t val;
988 
989     if (addr + sizeof(val) > vdev->config_len) {
990         return (uint32_t)-1;
991     }
992 
993     k->get_config(vdev, vdev->config);
994 
995     val = lduw_p(vdev->config + addr);
996     return val;
997 }
998 
999 uint32_t virtio_config_readl(VirtIODevice *vdev, uint32_t addr)
1000 {
1001     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1002     uint32_t val;
1003 
1004     if (addr + sizeof(val) > vdev->config_len) {
1005         return (uint32_t)-1;
1006     }
1007 
1008     k->get_config(vdev, vdev->config);
1009 
1010     val = ldl_p(vdev->config + addr);
1011     return val;
1012 }
1013 
1014 void virtio_config_writeb(VirtIODevice *vdev, uint32_t addr, uint32_t data)
1015 {
1016     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1017     uint8_t val = data;
1018 
1019     if (addr + sizeof(val) > vdev->config_len) {
1020         return;
1021     }
1022 
1023     stb_p(vdev->config + addr, val);
1024 
1025     if (k->set_config) {
1026         k->set_config(vdev, vdev->config);
1027     }
1028 }
1029 
1030 void virtio_config_writew(VirtIODevice *vdev, uint32_t addr, uint32_t data)
1031 {
1032     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1033     uint16_t val = data;
1034 
1035     if (addr + sizeof(val) > vdev->config_len) {
1036         return;
1037     }
1038 
1039     stw_p(vdev->config + addr, val);
1040 
1041     if (k->set_config) {
1042         k->set_config(vdev, vdev->config);
1043     }
1044 }
1045 
1046 void virtio_config_writel(VirtIODevice *vdev, uint32_t addr, uint32_t data)
1047 {
1048     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1049     uint32_t val = data;
1050 
1051     if (addr + sizeof(val) > vdev->config_len) {
1052         return;
1053     }
1054 
1055     stl_p(vdev->config + addr, val);
1056 
1057     if (k->set_config) {
1058         k->set_config(vdev, vdev->config);
1059     }
1060 }
1061 
1062 uint32_t virtio_config_modern_readb(VirtIODevice *vdev, uint32_t addr)
1063 {
1064     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1065     uint8_t val;
1066 
1067     if (addr + sizeof(val) > vdev->config_len) {
1068         return (uint32_t)-1;
1069     }
1070 
1071     k->get_config(vdev, vdev->config);
1072 
1073     val = ldub_p(vdev->config + addr);
1074     return val;
1075 }
1076 
1077 uint32_t virtio_config_modern_readw(VirtIODevice *vdev, uint32_t addr)
1078 {
1079     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1080     uint16_t val;
1081 
1082     if (addr + sizeof(val) > vdev->config_len) {
1083         return (uint32_t)-1;
1084     }
1085 
1086     k->get_config(vdev, vdev->config);
1087 
1088     val = lduw_le_p(vdev->config + addr);
1089     return val;
1090 }
1091 
1092 uint32_t virtio_config_modern_readl(VirtIODevice *vdev, uint32_t addr)
1093 {
1094     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1095     uint32_t val;
1096 
1097     if (addr + sizeof(val) > vdev->config_len) {
1098         return (uint32_t)-1;
1099     }
1100 
1101     k->get_config(vdev, vdev->config);
1102 
1103     val = ldl_le_p(vdev->config + addr);
1104     return val;
1105 }
1106 
1107 void virtio_config_modern_writeb(VirtIODevice *vdev,
1108                                  uint32_t addr, uint32_t data)
1109 {
1110     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1111     uint8_t val = data;
1112 
1113     if (addr + sizeof(val) > vdev->config_len) {
1114         return;
1115     }
1116 
1117     stb_p(vdev->config + addr, val);
1118 
1119     if (k->set_config) {
1120         k->set_config(vdev, vdev->config);
1121     }
1122 }
1123 
1124 void virtio_config_modern_writew(VirtIODevice *vdev,
1125                                  uint32_t addr, uint32_t data)
1126 {
1127     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1128     uint16_t val = data;
1129 
1130     if (addr + sizeof(val) > vdev->config_len) {
1131         return;
1132     }
1133 
1134     stw_le_p(vdev->config + addr, val);
1135 
1136     if (k->set_config) {
1137         k->set_config(vdev, vdev->config);
1138     }
1139 }
1140 
1141 void virtio_config_modern_writel(VirtIODevice *vdev,
1142                                  uint32_t addr, uint32_t data)
1143 {
1144     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1145     uint32_t val = data;
1146 
1147     if (addr + sizeof(val) > vdev->config_len) {
1148         return;
1149     }
1150 
1151     stl_le_p(vdev->config + addr, val);
1152 
1153     if (k->set_config) {
1154         k->set_config(vdev, vdev->config);
1155     }
1156 }
1157 
1158 void virtio_queue_set_addr(VirtIODevice *vdev, int n, hwaddr addr)
1159 {
1160     vdev->vq[n].vring.desc = addr;
1161     virtio_queue_update_rings(vdev, n);
1162 }
1163 
1164 hwaddr virtio_queue_get_addr(VirtIODevice *vdev, int n)
1165 {
1166     return vdev->vq[n].vring.desc;
1167 }
1168 
1169 void virtio_queue_set_rings(VirtIODevice *vdev, int n, hwaddr desc,
1170                             hwaddr avail, hwaddr used)
1171 {
1172     vdev->vq[n].vring.desc = desc;
1173     vdev->vq[n].vring.avail = avail;
1174     vdev->vq[n].vring.used = used;
1175 }
1176 
1177 void virtio_queue_set_num(VirtIODevice *vdev, int n, int num)
1178 {
1179     /* Don't allow guest to flip queue between existent and
1180      * nonexistent states, or to set it to an invalid size.
1181      */
1182     if (!!num != !!vdev->vq[n].vring.num ||
1183         num > VIRTQUEUE_MAX_SIZE ||
1184         num < 0) {
1185         return;
1186     }
1187     vdev->vq[n].vring.num = num;
1188 }
1189 
1190 VirtQueue *virtio_vector_first_queue(VirtIODevice *vdev, uint16_t vector)
1191 {
1192     return QLIST_FIRST(&vdev->vector_queues[vector]);
1193 }
1194 
1195 VirtQueue *virtio_vector_next_queue(VirtQueue *vq)
1196 {
1197     return QLIST_NEXT(vq, node);
1198 }
1199 
1200 int virtio_queue_get_num(VirtIODevice *vdev, int n)
1201 {
1202     return vdev->vq[n].vring.num;
1203 }
1204 
1205 int virtio_get_num_queues(VirtIODevice *vdev)
1206 {
1207     int i;
1208 
1209     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
1210         if (!virtio_queue_get_num(vdev, i)) {
1211             break;
1212         }
1213     }
1214 
1215     return i;
1216 }
1217 
1218 void virtio_queue_set_align(VirtIODevice *vdev, int n, int align)
1219 {
1220     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
1221     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
1222 
1223     /* virtio-1 compliant devices cannot change the alignment */
1224     if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
1225         error_report("tried to modify queue alignment for virtio-1 device");
1226         return;
1227     }
1228     /* Check that the transport told us it was going to do this
1229      * (so a buggy transport will immediately assert rather than
1230      * silently failing to migrate this state)
1231      */
1232     assert(k->has_variable_vring_alignment);
1233 
1234     vdev->vq[n].vring.align = align;
1235     virtio_queue_update_rings(vdev, n);
1236 }
1237 
1238 static void virtio_queue_notify_aio_vq(VirtQueue *vq)
1239 {
1240     if (vq->vring.desc && vq->handle_aio_output) {
1241         VirtIODevice *vdev = vq->vdev;
1242 
1243         trace_virtio_queue_notify(vdev, vq - vdev->vq, vq);
1244         vq->handle_aio_output(vdev, vq);
1245     }
1246 }
1247 
1248 static void virtio_queue_notify_vq(VirtQueue *vq)
1249 {
1250     if (vq->vring.desc && vq->handle_output) {
1251         VirtIODevice *vdev = vq->vdev;
1252 
1253         if (unlikely(vdev->broken)) {
1254             return;
1255         }
1256 
1257         trace_virtio_queue_notify(vdev, vq - vdev->vq, vq);
1258         vq->handle_output(vdev, vq);
1259     }
1260 }
1261 
1262 void virtio_queue_notify(VirtIODevice *vdev, int n)
1263 {
1264     virtio_queue_notify_vq(&vdev->vq[n]);
1265 }
1266 
1267 uint16_t virtio_queue_vector(VirtIODevice *vdev, int n)
1268 {
1269     return n < VIRTIO_QUEUE_MAX ? vdev->vq[n].vector :
1270         VIRTIO_NO_VECTOR;
1271 }
1272 
1273 void virtio_queue_set_vector(VirtIODevice *vdev, int n, uint16_t vector)
1274 {
1275     VirtQueue *vq = &vdev->vq[n];
1276 
1277     if (n < VIRTIO_QUEUE_MAX) {
1278         if (vdev->vector_queues &&
1279             vdev->vq[n].vector != VIRTIO_NO_VECTOR) {
1280             QLIST_REMOVE(vq, node);
1281         }
1282         vdev->vq[n].vector = vector;
1283         if (vdev->vector_queues &&
1284             vector != VIRTIO_NO_VECTOR) {
1285             QLIST_INSERT_HEAD(&vdev->vector_queues[vector], vq, node);
1286         }
1287     }
1288 }
1289 
1290 static VirtQueue *virtio_add_queue_internal(VirtIODevice *vdev, int queue_size,
1291                                             VirtIOHandleOutput handle_output,
1292                                             bool use_aio)
1293 {
1294     int i;
1295 
1296     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
1297         if (vdev->vq[i].vring.num == 0)
1298             break;
1299     }
1300 
1301     if (i == VIRTIO_QUEUE_MAX || queue_size > VIRTQUEUE_MAX_SIZE)
1302         abort();
1303 
1304     vdev->vq[i].vring.num = queue_size;
1305     vdev->vq[i].vring.num_default = queue_size;
1306     vdev->vq[i].vring.align = VIRTIO_PCI_VRING_ALIGN;
1307     vdev->vq[i].handle_output = handle_output;
1308     vdev->vq[i].handle_aio_output = NULL;
1309     vdev->vq[i].use_aio = use_aio;
1310 
1311     return &vdev->vq[i];
1312 }
1313 
1314 /* Add a virt queue and mark AIO.
1315  * An AIO queue will use the AioContext based event interface instead of the
1316  * default IOHandler and EventNotifier interface.
1317  */
1318 VirtQueue *virtio_add_queue_aio(VirtIODevice *vdev, int queue_size,
1319                                 VirtIOHandleOutput handle_output)
1320 {
1321     return virtio_add_queue_internal(vdev, queue_size, handle_output, true);
1322 }
1323 
1324 /* Add a normal virt queue (on the contrary to the AIO version above. */
1325 VirtQueue *virtio_add_queue(VirtIODevice *vdev, int queue_size,
1326                             VirtIOHandleOutput handle_output)
1327 {
1328     return virtio_add_queue_internal(vdev, queue_size, handle_output, false);
1329 }
1330 
1331 void virtio_del_queue(VirtIODevice *vdev, int n)
1332 {
1333     if (n < 0 || n >= VIRTIO_QUEUE_MAX) {
1334         abort();
1335     }
1336 
1337     vdev->vq[n].vring.num = 0;
1338     vdev->vq[n].vring.num_default = 0;
1339 }
1340 
1341 void virtio_irq(VirtQueue *vq)
1342 {
1343     trace_virtio_irq(vq);
1344     vq->vdev->isr |= 0x01;
1345     virtio_notify_vector(vq->vdev, vq->vector);
1346 }
1347 
1348 bool virtio_should_notify(VirtIODevice *vdev, VirtQueue *vq)
1349 {
1350     uint16_t old, new;
1351     bool v;
1352     /* We need to expose used array entries before checking used event. */
1353     smp_mb();
1354     /* Always notify when queue is empty (when feature acknowledge) */
1355     if (virtio_vdev_has_feature(vdev, VIRTIO_F_NOTIFY_ON_EMPTY) &&
1356         !vq->inuse && virtio_queue_empty(vq)) {
1357         return true;
1358     }
1359 
1360     if (!virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX)) {
1361         return !(vring_avail_flags(vq) & VRING_AVAIL_F_NO_INTERRUPT);
1362     }
1363 
1364     v = vq->signalled_used_valid;
1365     vq->signalled_used_valid = true;
1366     old = vq->signalled_used;
1367     new = vq->signalled_used = vq->used_idx;
1368     return !v || vring_need_event(vring_get_used_event(vq), new, old);
1369 }
1370 
1371 void virtio_notify(VirtIODevice *vdev, VirtQueue *vq)
1372 {
1373     if (!virtio_should_notify(vdev, vq)) {
1374         return;
1375     }
1376 
1377     trace_virtio_notify(vdev, vq);
1378     vdev->isr |= 0x01;
1379     virtio_notify_vector(vdev, vq->vector);
1380 }
1381 
1382 void virtio_notify_config(VirtIODevice *vdev)
1383 {
1384     if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK))
1385         return;
1386 
1387     vdev->isr |= 0x03;
1388     vdev->generation++;
1389     virtio_notify_vector(vdev, vdev->config_vector);
1390 }
1391 
1392 static bool virtio_device_endian_needed(void *opaque)
1393 {
1394     VirtIODevice *vdev = opaque;
1395 
1396     assert(vdev->device_endian != VIRTIO_DEVICE_ENDIAN_UNKNOWN);
1397     if (!virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
1398         return vdev->device_endian != virtio_default_endian();
1399     }
1400     /* Devices conforming to VIRTIO 1.0 or later are always LE. */
1401     return vdev->device_endian != VIRTIO_DEVICE_ENDIAN_LITTLE;
1402 }
1403 
1404 static bool virtio_64bit_features_needed(void *opaque)
1405 {
1406     VirtIODevice *vdev = opaque;
1407 
1408     return (vdev->host_features >> 32) != 0;
1409 }
1410 
1411 static bool virtio_virtqueue_needed(void *opaque)
1412 {
1413     VirtIODevice *vdev = opaque;
1414 
1415     return virtio_host_has_feature(vdev, VIRTIO_F_VERSION_1);
1416 }
1417 
1418 static bool virtio_ringsize_needed(void *opaque)
1419 {
1420     VirtIODevice *vdev = opaque;
1421     int i;
1422 
1423     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
1424         if (vdev->vq[i].vring.num != vdev->vq[i].vring.num_default) {
1425             return true;
1426         }
1427     }
1428     return false;
1429 }
1430 
1431 static bool virtio_extra_state_needed(void *opaque)
1432 {
1433     VirtIODevice *vdev = opaque;
1434     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
1435     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
1436 
1437     return k->has_extra_state &&
1438         k->has_extra_state(qbus->parent);
1439 }
1440 
1441 static bool virtio_broken_needed(void *opaque)
1442 {
1443     VirtIODevice *vdev = opaque;
1444 
1445     return vdev->broken;
1446 }
1447 
1448 static const VMStateDescription vmstate_virtqueue = {
1449     .name = "virtqueue_state",
1450     .version_id = 1,
1451     .minimum_version_id = 1,
1452     .fields = (VMStateField[]) {
1453         VMSTATE_UINT64(vring.avail, struct VirtQueue),
1454         VMSTATE_UINT64(vring.used, struct VirtQueue),
1455         VMSTATE_END_OF_LIST()
1456     }
1457 };
1458 
1459 static const VMStateDescription vmstate_virtio_virtqueues = {
1460     .name = "virtio/virtqueues",
1461     .version_id = 1,
1462     .minimum_version_id = 1,
1463     .needed = &virtio_virtqueue_needed,
1464     .fields = (VMStateField[]) {
1465         VMSTATE_STRUCT_VARRAY_POINTER_KNOWN(vq, struct VirtIODevice,
1466                       VIRTIO_QUEUE_MAX, 0, vmstate_virtqueue, VirtQueue),
1467         VMSTATE_END_OF_LIST()
1468     }
1469 };
1470 
1471 static const VMStateDescription vmstate_ringsize = {
1472     .name = "ringsize_state",
1473     .version_id = 1,
1474     .minimum_version_id = 1,
1475     .fields = (VMStateField[]) {
1476         VMSTATE_UINT32(vring.num_default, struct VirtQueue),
1477         VMSTATE_END_OF_LIST()
1478     }
1479 };
1480 
1481 static const VMStateDescription vmstate_virtio_ringsize = {
1482     .name = "virtio/ringsize",
1483     .version_id = 1,
1484     .minimum_version_id = 1,
1485     .needed = &virtio_ringsize_needed,
1486     .fields = (VMStateField[]) {
1487         VMSTATE_STRUCT_VARRAY_POINTER_KNOWN(vq, struct VirtIODevice,
1488                       VIRTIO_QUEUE_MAX, 0, vmstate_ringsize, VirtQueue),
1489         VMSTATE_END_OF_LIST()
1490     }
1491 };
1492 
1493 static int get_extra_state(QEMUFile *f, void *pv, size_t size)
1494 {
1495     VirtIODevice *vdev = pv;
1496     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
1497     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
1498 
1499     if (!k->load_extra_state) {
1500         return -1;
1501     } else {
1502         return k->load_extra_state(qbus->parent, f);
1503     }
1504 }
1505 
1506 static void put_extra_state(QEMUFile *f, void *pv, size_t size)
1507 {
1508     VirtIODevice *vdev = pv;
1509     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
1510     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
1511 
1512     k->save_extra_state(qbus->parent, f);
1513 }
1514 
1515 static const VMStateInfo vmstate_info_extra_state = {
1516     .name = "virtqueue_extra_state",
1517     .get = get_extra_state,
1518     .put = put_extra_state,
1519 };
1520 
1521 static const VMStateDescription vmstate_virtio_extra_state = {
1522     .name = "virtio/extra_state",
1523     .version_id = 1,
1524     .minimum_version_id = 1,
1525     .needed = &virtio_extra_state_needed,
1526     .fields = (VMStateField[]) {
1527         {
1528             .name         = "extra_state",
1529             .version_id   = 0,
1530             .field_exists = NULL,
1531             .size         = 0,
1532             .info         = &vmstate_info_extra_state,
1533             .flags        = VMS_SINGLE,
1534             .offset       = 0,
1535         },
1536         VMSTATE_END_OF_LIST()
1537     }
1538 };
1539 
1540 static const VMStateDescription vmstate_virtio_device_endian = {
1541     .name = "virtio/device_endian",
1542     .version_id = 1,
1543     .minimum_version_id = 1,
1544     .needed = &virtio_device_endian_needed,
1545     .fields = (VMStateField[]) {
1546         VMSTATE_UINT8(device_endian, VirtIODevice),
1547         VMSTATE_END_OF_LIST()
1548     }
1549 };
1550 
1551 static const VMStateDescription vmstate_virtio_64bit_features = {
1552     .name = "virtio/64bit_features",
1553     .version_id = 1,
1554     .minimum_version_id = 1,
1555     .needed = &virtio_64bit_features_needed,
1556     .fields = (VMStateField[]) {
1557         VMSTATE_UINT64(guest_features, VirtIODevice),
1558         VMSTATE_END_OF_LIST()
1559     }
1560 };
1561 
1562 static const VMStateDescription vmstate_virtio_broken = {
1563     .name = "virtio/broken",
1564     .version_id = 1,
1565     .minimum_version_id = 1,
1566     .needed = &virtio_broken_needed,
1567     .fields = (VMStateField[]) {
1568         VMSTATE_BOOL(broken, VirtIODevice),
1569         VMSTATE_END_OF_LIST()
1570     }
1571 };
1572 
1573 static const VMStateDescription vmstate_virtio = {
1574     .name = "virtio",
1575     .version_id = 1,
1576     .minimum_version_id = 1,
1577     .minimum_version_id_old = 1,
1578     .fields = (VMStateField[]) {
1579         VMSTATE_END_OF_LIST()
1580     },
1581     .subsections = (const VMStateDescription*[]) {
1582         &vmstate_virtio_device_endian,
1583         &vmstate_virtio_64bit_features,
1584         &vmstate_virtio_virtqueues,
1585         &vmstate_virtio_ringsize,
1586         &vmstate_virtio_broken,
1587         &vmstate_virtio_extra_state,
1588         NULL
1589     }
1590 };
1591 
1592 void virtio_save(VirtIODevice *vdev, QEMUFile *f)
1593 {
1594     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
1595     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
1596     VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev);
1597     uint32_t guest_features_lo = (vdev->guest_features & 0xffffffff);
1598     int i;
1599 
1600     if (k->save_config) {
1601         k->save_config(qbus->parent, f);
1602     }
1603 
1604     qemu_put_8s(f, &vdev->status);
1605     qemu_put_8s(f, &vdev->isr);
1606     qemu_put_be16s(f, &vdev->queue_sel);
1607     qemu_put_be32s(f, &guest_features_lo);
1608     qemu_put_be32(f, vdev->config_len);
1609     qemu_put_buffer(f, vdev->config, vdev->config_len);
1610 
1611     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
1612         if (vdev->vq[i].vring.num == 0)
1613             break;
1614     }
1615 
1616     qemu_put_be32(f, i);
1617 
1618     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
1619         if (vdev->vq[i].vring.num == 0)
1620             break;
1621 
1622         qemu_put_be32(f, vdev->vq[i].vring.num);
1623         if (k->has_variable_vring_alignment) {
1624             qemu_put_be32(f, vdev->vq[i].vring.align);
1625         }
1626         /* XXX virtio-1 devices */
1627         qemu_put_be64(f, vdev->vq[i].vring.desc);
1628         qemu_put_be16s(f, &vdev->vq[i].last_avail_idx);
1629         if (k->save_queue) {
1630             k->save_queue(qbus->parent, i, f);
1631         }
1632     }
1633 
1634     if (vdc->save != NULL) {
1635         vdc->save(vdev, f);
1636     }
1637 
1638     /* Subsections */
1639     vmstate_save_state(f, &vmstate_virtio, vdev, NULL);
1640 }
1641 
1642 /* A wrapper for use as a VMState .put function */
1643 static void virtio_device_put(QEMUFile *f, void *opaque, size_t size)
1644 {
1645     virtio_save(VIRTIO_DEVICE(opaque), f);
1646 }
1647 
1648 /* A wrapper for use as a VMState .get function */
1649 static int virtio_device_get(QEMUFile *f, void *opaque, size_t size)
1650 {
1651     VirtIODevice *vdev = VIRTIO_DEVICE(opaque);
1652     DeviceClass *dc = DEVICE_CLASS(VIRTIO_DEVICE_GET_CLASS(vdev));
1653 
1654     return virtio_load(vdev, f, dc->vmsd->version_id);
1655 }
1656 
1657 const VMStateInfo  virtio_vmstate_info = {
1658     .name = "virtio",
1659     .get = virtio_device_get,
1660     .put = virtio_device_put,
1661 };
1662 
1663 static int virtio_set_features_nocheck(VirtIODevice *vdev, uint64_t val)
1664 {
1665     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1666     bool bad = (val & ~(vdev->host_features)) != 0;
1667 
1668     val &= vdev->host_features;
1669     if (k->set_features) {
1670         k->set_features(vdev, val);
1671     }
1672     vdev->guest_features = val;
1673     return bad ? -1 : 0;
1674 }
1675 
1676 int virtio_set_features(VirtIODevice *vdev, uint64_t val)
1677 {
1678    /*
1679      * The driver must not attempt to set features after feature negotiation
1680      * has finished.
1681      */
1682     if (vdev->status & VIRTIO_CONFIG_S_FEATURES_OK) {
1683         return -EINVAL;
1684     }
1685     return virtio_set_features_nocheck(vdev, val);
1686 }
1687 
1688 int virtio_load(VirtIODevice *vdev, QEMUFile *f, int version_id)
1689 {
1690     int i, ret;
1691     int32_t config_len;
1692     uint32_t num;
1693     uint32_t features;
1694     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
1695     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
1696     VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev);
1697 
1698     /*
1699      * We poison the endianness to ensure it does not get used before
1700      * subsections have been loaded.
1701      */
1702     vdev->device_endian = VIRTIO_DEVICE_ENDIAN_UNKNOWN;
1703 
1704     if (k->load_config) {
1705         ret = k->load_config(qbus->parent, f);
1706         if (ret)
1707             return ret;
1708     }
1709 
1710     qemu_get_8s(f, &vdev->status);
1711     qemu_get_8s(f, &vdev->isr);
1712     qemu_get_be16s(f, &vdev->queue_sel);
1713     if (vdev->queue_sel >= VIRTIO_QUEUE_MAX) {
1714         return -1;
1715     }
1716     qemu_get_be32s(f, &features);
1717 
1718     /*
1719      * Temporarily set guest_features low bits - needed by
1720      * virtio net load code testing for VIRTIO_NET_F_CTRL_GUEST_OFFLOADS
1721      * VIRTIO_NET_F_GUEST_ANNOUNCE and VIRTIO_NET_F_CTRL_VQ.
1722      *
1723      * Note: devices should always test host features in future - don't create
1724      * new dependencies like this.
1725      */
1726     vdev->guest_features = features;
1727 
1728     config_len = qemu_get_be32(f);
1729 
1730     /*
1731      * There are cases where the incoming config can be bigger or smaller
1732      * than what we have; so load what we have space for, and skip
1733      * any excess that's in the stream.
1734      */
1735     qemu_get_buffer(f, vdev->config, MIN(config_len, vdev->config_len));
1736 
1737     while (config_len > vdev->config_len) {
1738         qemu_get_byte(f);
1739         config_len--;
1740     }
1741 
1742     num = qemu_get_be32(f);
1743 
1744     if (num > VIRTIO_QUEUE_MAX) {
1745         error_report("Invalid number of virtqueues: 0x%x", num);
1746         return -1;
1747     }
1748 
1749     for (i = 0; i < num; i++) {
1750         vdev->vq[i].vring.num = qemu_get_be32(f);
1751         if (k->has_variable_vring_alignment) {
1752             vdev->vq[i].vring.align = qemu_get_be32(f);
1753         }
1754         vdev->vq[i].vring.desc = qemu_get_be64(f);
1755         qemu_get_be16s(f, &vdev->vq[i].last_avail_idx);
1756         vdev->vq[i].signalled_used_valid = false;
1757         vdev->vq[i].notification = true;
1758 
1759         if (vdev->vq[i].vring.desc) {
1760             /* XXX virtio-1 devices */
1761             virtio_queue_update_rings(vdev, i);
1762         } else if (vdev->vq[i].last_avail_idx) {
1763             error_report("VQ %d address 0x0 "
1764                          "inconsistent with Host index 0x%x",
1765                          i, vdev->vq[i].last_avail_idx);
1766                 return -1;
1767         }
1768         if (k->load_queue) {
1769             ret = k->load_queue(qbus->parent, i, f);
1770             if (ret)
1771                 return ret;
1772         }
1773     }
1774 
1775     virtio_notify_vector(vdev, VIRTIO_NO_VECTOR);
1776 
1777     if (vdc->load != NULL) {
1778         ret = vdc->load(vdev, f, version_id);
1779         if (ret) {
1780             return ret;
1781         }
1782     }
1783 
1784     /* Subsections */
1785     ret = vmstate_load_state(f, &vmstate_virtio, vdev, 1);
1786     if (ret) {
1787         return ret;
1788     }
1789 
1790     if (vdev->device_endian == VIRTIO_DEVICE_ENDIAN_UNKNOWN) {
1791         vdev->device_endian = virtio_default_endian();
1792     }
1793 
1794     if (virtio_64bit_features_needed(vdev)) {
1795         /*
1796          * Subsection load filled vdev->guest_features.  Run them
1797          * through virtio_set_features to sanity-check them against
1798          * host_features.
1799          */
1800         uint64_t features64 = vdev->guest_features;
1801         if (virtio_set_features_nocheck(vdev, features64) < 0) {
1802             error_report("Features 0x%" PRIx64 " unsupported. "
1803                          "Allowed features: 0x%" PRIx64,
1804                          features64, vdev->host_features);
1805             return -1;
1806         }
1807     } else {
1808         if (virtio_set_features_nocheck(vdev, features) < 0) {
1809             error_report("Features 0x%x unsupported. "
1810                          "Allowed features: 0x%" PRIx64,
1811                          features, vdev->host_features);
1812             return -1;
1813         }
1814     }
1815 
1816     for (i = 0; i < num; i++) {
1817         if (vdev->vq[i].vring.desc) {
1818             uint16_t nheads;
1819             nheads = vring_avail_idx(&vdev->vq[i]) - vdev->vq[i].last_avail_idx;
1820             /* Check it isn't doing strange things with descriptor numbers. */
1821             if (nheads > vdev->vq[i].vring.num) {
1822                 error_report("VQ %d size 0x%x Guest index 0x%x "
1823                              "inconsistent with Host index 0x%x: delta 0x%x",
1824                              i, vdev->vq[i].vring.num,
1825                              vring_avail_idx(&vdev->vq[i]),
1826                              vdev->vq[i].last_avail_idx, nheads);
1827                 return -1;
1828             }
1829             vdev->vq[i].used_idx = vring_used_idx(&vdev->vq[i]);
1830             vdev->vq[i].shadow_avail_idx = vring_avail_idx(&vdev->vq[i]);
1831 
1832             /*
1833              * Some devices migrate VirtQueueElements that have been popped
1834              * from the avail ring but not yet returned to the used ring.
1835              */
1836             vdev->vq[i].inuse = vdev->vq[i].last_avail_idx -
1837                                 vdev->vq[i].used_idx;
1838             if (vdev->vq[i].inuse > vdev->vq[i].vring.num) {
1839                 error_report("VQ %d size 0x%x < last_avail_idx 0x%x - "
1840                              "used_idx 0x%x",
1841                              i, vdev->vq[i].vring.num,
1842                              vdev->vq[i].last_avail_idx,
1843                              vdev->vq[i].used_idx);
1844                 return -1;
1845             }
1846         }
1847     }
1848 
1849     return 0;
1850 }
1851 
1852 void virtio_cleanup(VirtIODevice *vdev)
1853 {
1854     qemu_del_vm_change_state_handler(vdev->vmstate);
1855     g_free(vdev->config);
1856     g_free(vdev->vq);
1857     g_free(vdev->vector_queues);
1858 }
1859 
1860 static void virtio_vmstate_change(void *opaque, int running, RunState state)
1861 {
1862     VirtIODevice *vdev = opaque;
1863     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
1864     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
1865     bool backend_run = running && (vdev->status & VIRTIO_CONFIG_S_DRIVER_OK);
1866     vdev->vm_running = running;
1867 
1868     if (backend_run) {
1869         virtio_set_status(vdev, vdev->status);
1870     }
1871 
1872     if (k->vmstate_change) {
1873         k->vmstate_change(qbus->parent, backend_run);
1874     }
1875 
1876     if (!backend_run) {
1877         virtio_set_status(vdev, vdev->status);
1878     }
1879 }
1880 
1881 void virtio_instance_init_common(Object *proxy_obj, void *data,
1882                                  size_t vdev_size, const char *vdev_name)
1883 {
1884     DeviceState *vdev = data;
1885 
1886     object_initialize(vdev, vdev_size, vdev_name);
1887     object_property_add_child(proxy_obj, "virtio-backend", OBJECT(vdev), NULL);
1888     object_unref(OBJECT(vdev));
1889     qdev_alias_all_properties(vdev, proxy_obj);
1890 }
1891 
1892 void virtio_init(VirtIODevice *vdev, const char *name,
1893                  uint16_t device_id, size_t config_size)
1894 {
1895     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
1896     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
1897     int i;
1898     int nvectors = k->query_nvectors ? k->query_nvectors(qbus->parent) : 0;
1899 
1900     if (nvectors) {
1901         vdev->vector_queues =
1902             g_malloc0(sizeof(*vdev->vector_queues) * nvectors);
1903     }
1904 
1905     vdev->device_id = device_id;
1906     vdev->status = 0;
1907     vdev->isr = 0;
1908     vdev->queue_sel = 0;
1909     vdev->config_vector = VIRTIO_NO_VECTOR;
1910     vdev->vq = g_malloc0(sizeof(VirtQueue) * VIRTIO_QUEUE_MAX);
1911     vdev->vm_running = runstate_is_running();
1912     vdev->broken = false;
1913     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
1914         vdev->vq[i].vector = VIRTIO_NO_VECTOR;
1915         vdev->vq[i].vdev = vdev;
1916         vdev->vq[i].queue_index = i;
1917     }
1918 
1919     vdev->name = name;
1920     vdev->config_len = config_size;
1921     if (vdev->config_len) {
1922         vdev->config = g_malloc0(config_size);
1923     } else {
1924         vdev->config = NULL;
1925     }
1926     vdev->vmstate = qemu_add_vm_change_state_handler(virtio_vmstate_change,
1927                                                      vdev);
1928     vdev->device_endian = virtio_default_endian();
1929     vdev->use_guest_notifier_mask = true;
1930 }
1931 
1932 hwaddr virtio_queue_get_desc_addr(VirtIODevice *vdev, int n)
1933 {
1934     return vdev->vq[n].vring.desc;
1935 }
1936 
1937 hwaddr virtio_queue_get_avail_addr(VirtIODevice *vdev, int n)
1938 {
1939     return vdev->vq[n].vring.avail;
1940 }
1941 
1942 hwaddr virtio_queue_get_used_addr(VirtIODevice *vdev, int n)
1943 {
1944     return vdev->vq[n].vring.used;
1945 }
1946 
1947 hwaddr virtio_queue_get_ring_addr(VirtIODevice *vdev, int n)
1948 {
1949     return vdev->vq[n].vring.desc;
1950 }
1951 
1952 hwaddr virtio_queue_get_desc_size(VirtIODevice *vdev, int n)
1953 {
1954     return sizeof(VRingDesc) * vdev->vq[n].vring.num;
1955 }
1956 
1957 hwaddr virtio_queue_get_avail_size(VirtIODevice *vdev, int n)
1958 {
1959     return offsetof(VRingAvail, ring) +
1960         sizeof(uint16_t) * vdev->vq[n].vring.num;
1961 }
1962 
1963 hwaddr virtio_queue_get_used_size(VirtIODevice *vdev, int n)
1964 {
1965     return offsetof(VRingUsed, ring) +
1966         sizeof(VRingUsedElem) * vdev->vq[n].vring.num;
1967 }
1968 
1969 hwaddr virtio_queue_get_ring_size(VirtIODevice *vdev, int n)
1970 {
1971     return vdev->vq[n].vring.used - vdev->vq[n].vring.desc +
1972 	    virtio_queue_get_used_size(vdev, n);
1973 }
1974 
1975 uint16_t virtio_queue_get_last_avail_idx(VirtIODevice *vdev, int n)
1976 {
1977     return vdev->vq[n].last_avail_idx;
1978 }
1979 
1980 void virtio_queue_set_last_avail_idx(VirtIODevice *vdev, int n, uint16_t idx)
1981 {
1982     vdev->vq[n].last_avail_idx = idx;
1983     vdev->vq[n].shadow_avail_idx = idx;
1984 }
1985 
1986 void virtio_queue_invalidate_signalled_used(VirtIODevice *vdev, int n)
1987 {
1988     vdev->vq[n].signalled_used_valid = false;
1989 }
1990 
1991 VirtQueue *virtio_get_queue(VirtIODevice *vdev, int n)
1992 {
1993     return vdev->vq + n;
1994 }
1995 
1996 uint16_t virtio_get_queue_index(VirtQueue *vq)
1997 {
1998     return vq->queue_index;
1999 }
2000 
2001 static void virtio_queue_guest_notifier_read(EventNotifier *n)
2002 {
2003     VirtQueue *vq = container_of(n, VirtQueue, guest_notifier);
2004     if (event_notifier_test_and_clear(n)) {
2005         virtio_irq(vq);
2006     }
2007 }
2008 
2009 void virtio_queue_set_guest_notifier_fd_handler(VirtQueue *vq, bool assign,
2010                                                 bool with_irqfd)
2011 {
2012     if (assign && !with_irqfd) {
2013         event_notifier_set_handler(&vq->guest_notifier, false,
2014                                    virtio_queue_guest_notifier_read);
2015     } else {
2016         event_notifier_set_handler(&vq->guest_notifier, false, NULL);
2017     }
2018     if (!assign) {
2019         /* Test and clear notifier before closing it,
2020          * in case poll callback didn't have time to run. */
2021         virtio_queue_guest_notifier_read(&vq->guest_notifier);
2022     }
2023 }
2024 
2025 EventNotifier *virtio_queue_get_guest_notifier(VirtQueue *vq)
2026 {
2027     return &vq->guest_notifier;
2028 }
2029 
2030 static void virtio_queue_host_notifier_aio_read(EventNotifier *n)
2031 {
2032     VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
2033     if (event_notifier_test_and_clear(n)) {
2034         virtio_queue_notify_aio_vq(vq);
2035     }
2036 }
2037 
2038 void virtio_queue_aio_set_host_notifier_handler(VirtQueue *vq, AioContext *ctx,
2039                                                 VirtIOHandleOutput handle_output)
2040 {
2041     if (handle_output) {
2042         vq->handle_aio_output = handle_output;
2043         aio_set_event_notifier(ctx, &vq->host_notifier, true,
2044                                virtio_queue_host_notifier_aio_read);
2045     } else {
2046         aio_set_event_notifier(ctx, &vq->host_notifier, true, NULL);
2047         /* Test and clear notifier before after disabling event,
2048          * in case poll callback didn't have time to run. */
2049         virtio_queue_host_notifier_aio_read(&vq->host_notifier);
2050         vq->handle_aio_output = NULL;
2051     }
2052 }
2053 
2054 static void virtio_queue_host_notifier_read(EventNotifier *n)
2055 {
2056     VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
2057     if (event_notifier_test_and_clear(n)) {
2058         virtio_queue_notify_vq(vq);
2059     }
2060 }
2061 
2062 void virtio_queue_set_host_notifier_fd_handler(VirtQueue *vq, bool assign,
2063                                                bool set_handler)
2064 {
2065     AioContext *ctx = qemu_get_aio_context();
2066     if (assign && set_handler) {
2067         if (vq->use_aio) {
2068             aio_set_event_notifier(ctx, &vq->host_notifier, true,
2069                                    virtio_queue_host_notifier_read);
2070         } else {
2071             event_notifier_set_handler(&vq->host_notifier, true,
2072                                        virtio_queue_host_notifier_read);
2073         }
2074     } else {
2075         if (vq->use_aio) {
2076             aio_set_event_notifier(ctx, &vq->host_notifier, true, NULL);
2077         } else {
2078             event_notifier_set_handler(&vq->host_notifier, true, NULL);
2079         }
2080     }
2081     if (!assign) {
2082         /* Test and clear notifier before after disabling event,
2083          * in case poll callback didn't have time to run. */
2084         virtio_queue_host_notifier_read(&vq->host_notifier);
2085     }
2086 }
2087 
2088 EventNotifier *virtio_queue_get_host_notifier(VirtQueue *vq)
2089 {
2090     return &vq->host_notifier;
2091 }
2092 
2093 void virtio_device_set_child_bus_name(VirtIODevice *vdev, char *bus_name)
2094 {
2095     g_free(vdev->bus_name);
2096     vdev->bus_name = g_strdup(bus_name);
2097 }
2098 
2099 void GCC_FMT_ATTR(2, 3) virtio_error(VirtIODevice *vdev, const char *fmt, ...)
2100 {
2101     va_list ap;
2102 
2103     va_start(ap, fmt);
2104     error_vreport(fmt, ap);
2105     va_end(ap);
2106 
2107     vdev->broken = true;
2108 
2109     if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
2110         virtio_set_status(vdev, vdev->status | VIRTIO_CONFIG_S_NEEDS_RESET);
2111         virtio_notify_config(vdev);
2112     }
2113 }
2114 
2115 static void virtio_device_realize(DeviceState *dev, Error **errp)
2116 {
2117     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
2118     VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(dev);
2119     Error *err = NULL;
2120 
2121     if (vdc->realize != NULL) {
2122         vdc->realize(dev, &err);
2123         if (err != NULL) {
2124             error_propagate(errp, err);
2125             return;
2126         }
2127     }
2128 
2129     virtio_bus_device_plugged(vdev, &err);
2130     if (err != NULL) {
2131         error_propagate(errp, err);
2132         return;
2133     }
2134 }
2135 
2136 static void virtio_device_unrealize(DeviceState *dev, Error **errp)
2137 {
2138     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
2139     VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(dev);
2140     Error *err = NULL;
2141 
2142     virtio_bus_device_unplugged(vdev);
2143 
2144     if (vdc->unrealize != NULL) {
2145         vdc->unrealize(dev, &err);
2146         if (err != NULL) {
2147             error_propagate(errp, err);
2148             return;
2149         }
2150     }
2151 
2152     g_free(vdev->bus_name);
2153     vdev->bus_name = NULL;
2154 }
2155 
2156 static Property virtio_properties[] = {
2157     DEFINE_VIRTIO_COMMON_FEATURES(VirtIODevice, host_features),
2158     DEFINE_PROP_END_OF_LIST(),
2159 };
2160 
2161 static void virtio_device_class_init(ObjectClass *klass, void *data)
2162 {
2163     /* Set the default value here. */
2164     DeviceClass *dc = DEVICE_CLASS(klass);
2165 
2166     dc->realize = virtio_device_realize;
2167     dc->unrealize = virtio_device_unrealize;
2168     dc->bus_type = TYPE_VIRTIO_BUS;
2169     dc->props = virtio_properties;
2170 }
2171 
2172 static const TypeInfo virtio_device_info = {
2173     .name = TYPE_VIRTIO_DEVICE,
2174     .parent = TYPE_DEVICE,
2175     .instance_size = sizeof(VirtIODevice),
2176     .class_init = virtio_device_class_init,
2177     .abstract = true,
2178     .class_size = sizeof(VirtioDeviceClass),
2179 };
2180 
2181 static void virtio_register_types(void)
2182 {
2183     type_register_static(&virtio_device_info);
2184 }
2185 
2186 type_init(virtio_register_types)
2187