xref: /openbmc/qemu/hw/virtio/virtio.c (revision fe44dc91)
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 #include "sysemu/dma.h"
27 
28 /*
29  * The alignment to use between consumer and producer parts of vring.
30  * x86 pagesize again. This is the default, used by transports like PCI
31  * which don't provide a means for the guest to tell the host the alignment.
32  */
33 #define VIRTIO_PCI_VRING_ALIGN         4096
34 
35 typedef struct VRingDesc
36 {
37     uint64_t addr;
38     uint32_t len;
39     uint16_t flags;
40     uint16_t next;
41 } VRingDesc;
42 
43 typedef struct VRingAvail
44 {
45     uint16_t flags;
46     uint16_t idx;
47     uint16_t ring[0];
48 } VRingAvail;
49 
50 typedef struct VRingUsedElem
51 {
52     uint32_t id;
53     uint32_t len;
54 } VRingUsedElem;
55 
56 typedef struct VRingUsed
57 {
58     uint16_t flags;
59     uint16_t idx;
60     VRingUsedElem ring[0];
61 } VRingUsed;
62 
63 typedef struct VRing
64 {
65     unsigned int num;
66     unsigned int num_default;
67     unsigned int align;
68     hwaddr desc;
69     hwaddr avail;
70     hwaddr used;
71 } VRing;
72 
73 struct VirtQueue
74 {
75     VRing vring;
76 
77     /* Next head to pop */
78     uint16_t last_avail_idx;
79 
80     /* Last avail_idx read from VQ. */
81     uint16_t shadow_avail_idx;
82 
83     uint16_t used_idx;
84 
85     /* Last used index value we have signalled on */
86     uint16_t signalled_used;
87 
88     /* Last used index value we have signalled on */
89     bool signalled_used_valid;
90 
91     /* Notification enabled? */
92     bool notification;
93 
94     uint16_t queue_index;
95 
96     unsigned int inuse;
97 
98     uint16_t vector;
99     VirtIOHandleOutput handle_output;
100     VirtIOHandleOutput handle_aio_output;
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(vdev->dma_as, 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(vq->vdev->dma_as, 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     AddressSpace *dma_as = vq->vdev->dma_as;
248     unsigned int offset;
249     int i;
250 
251     offset = 0;
252     for (i = 0; i < elem->in_num; i++) {
253         size_t size = MIN(len - offset, elem->in_sg[i].iov_len);
254 
255         dma_memory_unmap(dma_as, elem->in_sg[i].iov_base,
256                          elem->in_sg[i].iov_len,
257                          DMA_DIRECTION_FROM_DEVICE, size);
258 
259         offset += size;
260     }
261 
262     for (i = 0; i < elem->out_num; i++)
263         dma_memory_unmap(dma_as, elem->out_sg[i].iov_base,
264                          elem->out_sg[i].iov_len,
265                          DMA_DIRECTION_TO_DEVICE,
266                          elem->out_sg[i].iov_len);
267 }
268 
269 /* virtqueue_detach_element:
270  * @vq: The #VirtQueue
271  * @elem: The #VirtQueueElement
272  * @len: number of bytes written
273  *
274  * Detach the element from the virtqueue.  This function is suitable for device
275  * reset or other situations where a #VirtQueueElement is simply freed and will
276  * not be pushed or discarded.
277  */
278 void virtqueue_detach_element(VirtQueue *vq, const VirtQueueElement *elem,
279                               unsigned int len)
280 {
281     vq->inuse--;
282     virtqueue_unmap_sg(vq, elem, len);
283 }
284 
285 /* virtqueue_unpop:
286  * @vq: The #VirtQueue
287  * @elem: The #VirtQueueElement
288  * @len: number of bytes written
289  *
290  * Pretend the most recent element wasn't popped from the virtqueue.  The next
291  * call to virtqueue_pop() will refetch the element.
292  */
293 void virtqueue_unpop(VirtQueue *vq, const VirtQueueElement *elem,
294                      unsigned int len)
295 {
296     vq->last_avail_idx--;
297     virtqueue_detach_element(vq, elem, len);
298 }
299 
300 /* virtqueue_rewind:
301  * @vq: The #VirtQueue
302  * @num: Number of elements to push back
303  *
304  * Pretend that elements weren't popped from the virtqueue.  The next
305  * virtqueue_pop() will refetch the oldest element.
306  *
307  * Use virtqueue_unpop() instead if you have a VirtQueueElement.
308  *
309  * Returns: true on success, false if @num is greater than the number of in use
310  * elements.
311  */
312 bool virtqueue_rewind(VirtQueue *vq, unsigned int num)
313 {
314     if (num > vq->inuse) {
315         return false;
316     }
317     vq->last_avail_idx -= num;
318     vq->inuse -= num;
319     return true;
320 }
321 
322 void virtqueue_fill(VirtQueue *vq, const VirtQueueElement *elem,
323                     unsigned int len, unsigned int idx)
324 {
325     VRingUsedElem uelem;
326 
327     trace_virtqueue_fill(vq, elem, len, idx);
328 
329     virtqueue_unmap_sg(vq, elem, len);
330 
331     if (unlikely(vq->vdev->broken)) {
332         return;
333     }
334 
335     idx = (idx + vq->used_idx) % vq->vring.num;
336 
337     uelem.id = elem->index;
338     uelem.len = len;
339     vring_used_write(vq, &uelem, idx);
340 }
341 
342 void virtqueue_flush(VirtQueue *vq, unsigned int count)
343 {
344     uint16_t old, new;
345 
346     if (unlikely(vq->vdev->broken)) {
347         vq->inuse -= count;
348         return;
349     }
350 
351     /* Make sure buffer is written before we update index. */
352     smp_wmb();
353     trace_virtqueue_flush(vq, count);
354     old = vq->used_idx;
355     new = old + count;
356     vring_used_idx_set(vq, new);
357     vq->inuse -= count;
358     if (unlikely((int16_t)(new - vq->signalled_used) < (uint16_t)(new - old)))
359         vq->signalled_used_valid = false;
360 }
361 
362 void virtqueue_push(VirtQueue *vq, const VirtQueueElement *elem,
363                     unsigned int len)
364 {
365     virtqueue_fill(vq, elem, len, 0);
366     virtqueue_flush(vq, 1);
367 }
368 
369 static int virtqueue_num_heads(VirtQueue *vq, unsigned int idx)
370 {
371     uint16_t num_heads = vring_avail_idx(vq) - idx;
372 
373     /* Check it isn't doing very strange things with descriptor numbers. */
374     if (num_heads > vq->vring.num) {
375         virtio_error(vq->vdev, "Guest moved used index from %u to %u",
376                      idx, vq->shadow_avail_idx);
377         return -EINVAL;
378     }
379     /* On success, callers read a descriptor at vq->last_avail_idx.
380      * Make sure descriptor read does not bypass avail index read. */
381     if (num_heads) {
382         smp_rmb();
383     }
384 
385     return num_heads;
386 }
387 
388 static bool virtqueue_get_head(VirtQueue *vq, unsigned int idx,
389                                unsigned int *head)
390 {
391     /* Grab the next descriptor number they're advertising, and increment
392      * the index we've seen. */
393     *head = vring_avail_ring(vq, idx % vq->vring.num);
394 
395     /* If their number is silly, that's a fatal mistake. */
396     if (*head >= vq->vring.num) {
397         virtio_error(vq->vdev, "Guest says index %u is available", *head);
398         return false;
399     }
400 
401     return true;
402 }
403 
404 enum {
405     VIRTQUEUE_READ_DESC_ERROR = -1,
406     VIRTQUEUE_READ_DESC_DONE = 0,   /* end of chain */
407     VIRTQUEUE_READ_DESC_MORE = 1,   /* more buffers in chain */
408 };
409 
410 static int virtqueue_read_next_desc(VirtIODevice *vdev, VRingDesc *desc,
411                                     hwaddr desc_pa, unsigned int max,
412                                     unsigned int *next)
413 {
414     /* If this descriptor says it doesn't chain, we're done. */
415     if (!(desc->flags & VRING_DESC_F_NEXT)) {
416         return VIRTQUEUE_READ_DESC_DONE;
417     }
418 
419     /* Check they're not leading us off end of descriptors. */
420     *next = desc->next;
421     /* Make sure compiler knows to grab that: we don't want it changing! */
422     smp_wmb();
423 
424     if (*next >= max) {
425         virtio_error(vdev, "Desc next is %u", *next);
426         return VIRTQUEUE_READ_DESC_ERROR;
427     }
428 
429     vring_desc_read(vdev, desc, desc_pa, *next);
430     return VIRTQUEUE_READ_DESC_MORE;
431 }
432 
433 void virtqueue_get_avail_bytes(VirtQueue *vq, unsigned int *in_bytes,
434                                unsigned int *out_bytes,
435                                unsigned max_in_bytes, unsigned max_out_bytes)
436 {
437     unsigned int idx;
438     unsigned int total_bufs, in_total, out_total;
439     int rc;
440 
441     idx = vq->last_avail_idx;
442 
443     total_bufs = in_total = out_total = 0;
444     while ((rc = virtqueue_num_heads(vq, idx)) > 0) {
445         VirtIODevice *vdev = vq->vdev;
446         unsigned int max, num_bufs, indirect = 0;
447         VRingDesc desc;
448         hwaddr desc_pa;
449         unsigned int i;
450 
451         max = vq->vring.num;
452         num_bufs = total_bufs;
453 
454         if (!virtqueue_get_head(vq, idx++, &i)) {
455             goto err;
456         }
457 
458         desc_pa = vq->vring.desc;
459         vring_desc_read(vdev, &desc, desc_pa, i);
460 
461         if (desc.flags & VRING_DESC_F_INDIRECT) {
462             if (desc.len % sizeof(VRingDesc)) {
463                 virtio_error(vdev, "Invalid size for indirect buffer table");
464                 goto err;
465             }
466 
467             /* If we've got too many, that implies a descriptor loop. */
468             if (num_bufs >= max) {
469                 virtio_error(vdev, "Looped descriptor");
470                 goto err;
471             }
472 
473             /* loop over the indirect descriptor table */
474             indirect = 1;
475             max = desc.len / sizeof(VRingDesc);
476             desc_pa = desc.addr;
477             num_bufs = i = 0;
478             vring_desc_read(vdev, &desc, desc_pa, i);
479         }
480 
481         do {
482             /* If we've got too many, that implies a descriptor loop. */
483             if (++num_bufs > max) {
484                 virtio_error(vdev, "Looped descriptor");
485                 goto err;
486             }
487 
488             if (desc.flags & VRING_DESC_F_WRITE) {
489                 in_total += desc.len;
490             } else {
491                 out_total += desc.len;
492             }
493             if (in_total >= max_in_bytes && out_total >= max_out_bytes) {
494                 goto done;
495             }
496 
497             rc = virtqueue_read_next_desc(vdev, &desc, desc_pa, max, &i);
498         } while (rc == VIRTQUEUE_READ_DESC_MORE);
499 
500         if (rc == VIRTQUEUE_READ_DESC_ERROR) {
501             goto err;
502         }
503 
504         if (!indirect)
505             total_bufs = num_bufs;
506         else
507             total_bufs++;
508     }
509 
510     if (rc < 0) {
511         goto err;
512     }
513 
514 done:
515     if (in_bytes) {
516         *in_bytes = in_total;
517     }
518     if (out_bytes) {
519         *out_bytes = out_total;
520     }
521     return;
522 
523 err:
524     in_total = out_total = 0;
525     goto done;
526 }
527 
528 int virtqueue_avail_bytes(VirtQueue *vq, unsigned int in_bytes,
529                           unsigned int out_bytes)
530 {
531     unsigned int in_total, out_total;
532 
533     virtqueue_get_avail_bytes(vq, &in_total, &out_total, in_bytes, out_bytes);
534     return in_bytes <= in_total && out_bytes <= out_total;
535 }
536 
537 static bool virtqueue_map_desc(VirtIODevice *vdev, unsigned int *p_num_sg,
538                                hwaddr *addr, struct iovec *iov,
539                                unsigned int max_num_sg, bool is_write,
540                                hwaddr pa, size_t sz)
541 {
542     bool ok = false;
543     unsigned num_sg = *p_num_sg;
544     assert(num_sg <= max_num_sg);
545 
546     if (!sz) {
547         virtio_error(vdev, "virtio: zero sized buffers are not allowed");
548         goto out;
549     }
550 
551     while (sz) {
552         hwaddr len = sz;
553 
554         if (num_sg == max_num_sg) {
555             virtio_error(vdev, "virtio: too many write descriptors in "
556                                "indirect table");
557             goto out;
558         }
559 
560         iov[num_sg].iov_base = dma_memory_map(vdev->dma_as, pa, &len,
561                                               is_write ?
562                                               DMA_DIRECTION_FROM_DEVICE :
563                                               DMA_DIRECTION_TO_DEVICE);
564         if (!iov[num_sg].iov_base) {
565             virtio_error(vdev, "virtio: bogus descriptor or out of resources");
566             goto out;
567         }
568 
569         iov[num_sg].iov_len = len;
570         addr[num_sg] = pa;
571 
572         sz -= len;
573         pa += len;
574         num_sg++;
575     }
576     ok = true;
577 
578 out:
579     *p_num_sg = num_sg;
580     return ok;
581 }
582 
583 /* Only used by error code paths before we have a VirtQueueElement (therefore
584  * virtqueue_unmap_sg() can't be used).  Assumes buffers weren't written to
585  * yet.
586  */
587 static void virtqueue_undo_map_desc(unsigned int out_num, unsigned int in_num,
588                                     struct iovec *iov)
589 {
590     unsigned int i;
591 
592     for (i = 0; i < out_num + in_num; i++) {
593         int is_write = i >= out_num;
594 
595         cpu_physical_memory_unmap(iov->iov_base, iov->iov_len, is_write, 0);
596         iov++;
597     }
598 }
599 
600 static void virtqueue_map_iovec(VirtIODevice *vdev, struct iovec *sg,
601                                 hwaddr *addr, unsigned int *num_sg,
602                                 int is_write)
603 {
604     unsigned int i;
605     hwaddr len;
606 
607     for (i = 0; i < *num_sg; i++) {
608         len = sg[i].iov_len;
609         sg[i].iov_base = dma_memory_map(vdev->dma_as,
610                                         addr[i], &len, is_write ?
611                                         DMA_DIRECTION_FROM_DEVICE :
612                                         DMA_DIRECTION_TO_DEVICE);
613         if (!sg[i].iov_base) {
614             error_report("virtio: error trying to map MMIO memory");
615             exit(1);
616         }
617         if (len != sg[i].iov_len) {
618             error_report("virtio: unexpected memory split");
619             exit(1);
620         }
621     }
622 }
623 
624 void virtqueue_map(VirtIODevice *vdev, VirtQueueElement *elem)
625 {
626     virtqueue_map_iovec(vdev, elem->in_sg, elem->in_addr, &elem->in_num, 1);
627     virtqueue_map_iovec(vdev, elem->out_sg, elem->out_addr, &elem->out_num, 0);
628 }
629 
630 static void *virtqueue_alloc_element(size_t sz, unsigned out_num, unsigned in_num)
631 {
632     VirtQueueElement *elem;
633     size_t in_addr_ofs = QEMU_ALIGN_UP(sz, __alignof__(elem->in_addr[0]));
634     size_t out_addr_ofs = in_addr_ofs + in_num * sizeof(elem->in_addr[0]);
635     size_t out_addr_end = out_addr_ofs + out_num * sizeof(elem->out_addr[0]);
636     size_t in_sg_ofs = QEMU_ALIGN_UP(out_addr_end, __alignof__(elem->in_sg[0]));
637     size_t out_sg_ofs = in_sg_ofs + in_num * sizeof(elem->in_sg[0]);
638     size_t out_sg_end = out_sg_ofs + out_num * sizeof(elem->out_sg[0]);
639 
640     assert(sz >= sizeof(VirtQueueElement));
641     elem = g_malloc(out_sg_end);
642     elem->out_num = out_num;
643     elem->in_num = in_num;
644     elem->in_addr = (void *)elem + in_addr_ofs;
645     elem->out_addr = (void *)elem + out_addr_ofs;
646     elem->in_sg = (void *)elem + in_sg_ofs;
647     elem->out_sg = (void *)elem + out_sg_ofs;
648     return elem;
649 }
650 
651 void *virtqueue_pop(VirtQueue *vq, size_t sz)
652 {
653     unsigned int i, head, max;
654     hwaddr desc_pa = vq->vring.desc;
655     VirtIODevice *vdev = vq->vdev;
656     VirtQueueElement *elem;
657     unsigned out_num, in_num;
658     hwaddr addr[VIRTQUEUE_MAX_SIZE];
659     struct iovec iov[VIRTQUEUE_MAX_SIZE];
660     VRingDesc desc;
661     int rc;
662 
663     if (unlikely(vdev->broken)) {
664         return NULL;
665     }
666     if (virtio_queue_empty(vq)) {
667         return NULL;
668     }
669     /* Needed after virtio_queue_empty(), see comment in
670      * virtqueue_num_heads(). */
671     smp_rmb();
672 
673     /* When we start there are none of either input nor output. */
674     out_num = in_num = 0;
675 
676     max = vq->vring.num;
677 
678     if (vq->inuse >= vq->vring.num) {
679         virtio_error(vdev, "Virtqueue size exceeded");
680         return NULL;
681     }
682 
683     if (!virtqueue_get_head(vq, vq->last_avail_idx++, &head)) {
684         return NULL;
685     }
686 
687     if (virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX)) {
688         vring_set_avail_event(vq, vq->last_avail_idx);
689     }
690 
691     i = head;
692     vring_desc_read(vdev, &desc, desc_pa, i);
693     if (desc.flags & VRING_DESC_F_INDIRECT) {
694         if (desc.len % sizeof(VRingDesc)) {
695             virtio_error(vdev, "Invalid size for indirect buffer table");
696             return NULL;
697         }
698 
699         /* loop over the indirect descriptor table */
700         max = desc.len / sizeof(VRingDesc);
701         desc_pa = desc.addr;
702         i = 0;
703         vring_desc_read(vdev, &desc, desc_pa, i);
704     }
705 
706     /* Collect all the descriptors */
707     do {
708         bool map_ok;
709 
710         if (desc.flags & VRING_DESC_F_WRITE) {
711             map_ok = virtqueue_map_desc(vdev, &in_num, addr + out_num,
712                                         iov + out_num,
713                                         VIRTQUEUE_MAX_SIZE - out_num, true,
714                                         desc.addr, desc.len);
715         } else {
716             if (in_num) {
717                 virtio_error(vdev, "Incorrect order for descriptors");
718                 goto err_undo_map;
719             }
720             map_ok = virtqueue_map_desc(vdev, &out_num, addr, iov,
721                                         VIRTQUEUE_MAX_SIZE, false,
722                                         desc.addr, desc.len);
723         }
724         if (!map_ok) {
725             goto err_undo_map;
726         }
727 
728         /* If we've got too many, that implies a descriptor loop. */
729         if ((in_num + out_num) > max) {
730             virtio_error(vdev, "Looped descriptor");
731             goto err_undo_map;
732         }
733 
734         rc = virtqueue_read_next_desc(vdev, &desc, desc_pa, max, &i);
735     } while (rc == VIRTQUEUE_READ_DESC_MORE);
736 
737     if (rc == VIRTQUEUE_READ_DESC_ERROR) {
738         goto err_undo_map;
739     }
740 
741     /* Now copy what we have collected and mapped */
742     elem = virtqueue_alloc_element(sz, out_num, in_num);
743     elem->index = head;
744     for (i = 0; i < out_num; i++) {
745         elem->out_addr[i] = addr[i];
746         elem->out_sg[i] = iov[i];
747     }
748     for (i = 0; i < in_num; i++) {
749         elem->in_addr[i] = addr[out_num + i];
750         elem->in_sg[i] = iov[out_num + i];
751     }
752 
753     vq->inuse++;
754 
755     trace_virtqueue_pop(vq, elem, elem->in_num, elem->out_num);
756     return elem;
757 
758 err_undo_map:
759     virtqueue_undo_map_desc(out_num, in_num, iov);
760     return NULL;
761 }
762 
763 /* virtqueue_drop_all:
764  * @vq: The #VirtQueue
765  * Drops all queued buffers and indicates them to the guest
766  * as if they are done. Useful when buffers can not be
767  * processed but must be returned to the guest.
768  */
769 unsigned int virtqueue_drop_all(VirtQueue *vq)
770 {
771     unsigned int dropped = 0;
772     VirtQueueElement elem = {};
773     VirtIODevice *vdev = vq->vdev;
774     bool fEventIdx = virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX);
775 
776     if (unlikely(vdev->broken)) {
777         return 0;
778     }
779 
780     while (!virtio_queue_empty(vq) && vq->inuse < vq->vring.num) {
781         /* works similar to virtqueue_pop but does not map buffers
782         * and does not allocate any memory */
783         smp_rmb();
784         if (!virtqueue_get_head(vq, vq->last_avail_idx, &elem.index)) {
785             break;
786         }
787         vq->inuse++;
788         vq->last_avail_idx++;
789         if (fEventIdx) {
790             vring_set_avail_event(vq, vq->last_avail_idx);
791         }
792         /* immediately push the element, nothing to unmap
793          * as both in_num and out_num are set to 0 */
794         virtqueue_push(vq, &elem, 0);
795         dropped++;
796     }
797 
798     return dropped;
799 }
800 
801 /* Reading and writing a structure directly to QEMUFile is *awful*, but
802  * it is what QEMU has always done by mistake.  We can change it sooner
803  * or later by bumping the version number of the affected vm states.
804  * In the meanwhile, since the in-memory layout of VirtQueueElement
805  * has changed, we need to marshal to and from the layout that was
806  * used before the change.
807  */
808 typedef struct VirtQueueElementOld {
809     unsigned int index;
810     unsigned int out_num;
811     unsigned int in_num;
812     hwaddr in_addr[VIRTQUEUE_MAX_SIZE];
813     hwaddr out_addr[VIRTQUEUE_MAX_SIZE];
814     struct iovec in_sg[VIRTQUEUE_MAX_SIZE];
815     struct iovec out_sg[VIRTQUEUE_MAX_SIZE];
816 } VirtQueueElementOld;
817 
818 void *qemu_get_virtqueue_element(VirtIODevice *vdev, QEMUFile *f, size_t sz)
819 {
820     VirtQueueElement *elem;
821     VirtQueueElementOld data;
822     int i;
823 
824     qemu_get_buffer(f, (uint8_t *)&data, sizeof(VirtQueueElementOld));
825 
826     /* TODO: teach all callers that this can fail, and return failure instead
827      * of asserting here.
828      * When we do, we might be able to re-enable NDEBUG below.
829      */
830 #ifdef NDEBUG
831 #error building with NDEBUG is not supported
832 #endif
833     assert(ARRAY_SIZE(data.in_addr) >= data.in_num);
834     assert(ARRAY_SIZE(data.out_addr) >= data.out_num);
835 
836     elem = virtqueue_alloc_element(sz, data.out_num, data.in_num);
837     elem->index = data.index;
838 
839     for (i = 0; i < elem->in_num; i++) {
840         elem->in_addr[i] = data.in_addr[i];
841     }
842 
843     for (i = 0; i < elem->out_num; i++) {
844         elem->out_addr[i] = data.out_addr[i];
845     }
846 
847     for (i = 0; i < elem->in_num; i++) {
848         /* Base is overwritten by virtqueue_map.  */
849         elem->in_sg[i].iov_base = 0;
850         elem->in_sg[i].iov_len = data.in_sg[i].iov_len;
851     }
852 
853     for (i = 0; i < elem->out_num; i++) {
854         /* Base is overwritten by virtqueue_map.  */
855         elem->out_sg[i].iov_base = 0;
856         elem->out_sg[i].iov_len = data.out_sg[i].iov_len;
857     }
858 
859     virtqueue_map(vdev, elem);
860     return elem;
861 }
862 
863 void qemu_put_virtqueue_element(QEMUFile *f, VirtQueueElement *elem)
864 {
865     VirtQueueElementOld data;
866     int i;
867 
868     memset(&data, 0, sizeof(data));
869     data.index = elem->index;
870     data.in_num = elem->in_num;
871     data.out_num = elem->out_num;
872 
873     for (i = 0; i < elem->in_num; i++) {
874         data.in_addr[i] = elem->in_addr[i];
875     }
876 
877     for (i = 0; i < elem->out_num; i++) {
878         data.out_addr[i] = elem->out_addr[i];
879     }
880 
881     for (i = 0; i < elem->in_num; i++) {
882         /* Base is overwritten by virtqueue_map when loading.  Do not
883          * save it, as it would leak the QEMU address space layout.  */
884         data.in_sg[i].iov_len = elem->in_sg[i].iov_len;
885     }
886 
887     for (i = 0; i < elem->out_num; i++) {
888         /* Do not save iov_base as above.  */
889         data.out_sg[i].iov_len = elem->out_sg[i].iov_len;
890     }
891     qemu_put_buffer(f, (uint8_t *)&data, sizeof(VirtQueueElementOld));
892 }
893 
894 /* virtio device */
895 static void virtio_notify_vector(VirtIODevice *vdev, uint16_t vector)
896 {
897     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
898     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
899 
900     if (unlikely(vdev->broken)) {
901         return;
902     }
903 
904     if (k->notify) {
905         k->notify(qbus->parent, vector);
906     }
907 }
908 
909 void virtio_update_irq(VirtIODevice *vdev)
910 {
911     virtio_notify_vector(vdev, VIRTIO_NO_VECTOR);
912 }
913 
914 static int virtio_validate_features(VirtIODevice *vdev)
915 {
916     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
917 
918     if (virtio_host_has_feature(vdev, VIRTIO_F_IOMMU_PLATFORM) &&
919         !virtio_vdev_has_feature(vdev, VIRTIO_F_IOMMU_PLATFORM)) {
920         return -EFAULT;
921     }
922 
923     if (k->validate_features) {
924         return k->validate_features(vdev);
925     } else {
926         return 0;
927     }
928 }
929 
930 int virtio_set_status(VirtIODevice *vdev, uint8_t val)
931 {
932     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
933     trace_virtio_set_status(vdev, val);
934 
935     if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
936         if (!(vdev->status & VIRTIO_CONFIG_S_FEATURES_OK) &&
937             val & VIRTIO_CONFIG_S_FEATURES_OK) {
938             int ret = virtio_validate_features(vdev);
939 
940             if (ret) {
941                 return ret;
942             }
943         }
944     }
945     if (k->set_status) {
946         k->set_status(vdev, val);
947     }
948     vdev->status = val;
949     return 0;
950 }
951 
952 bool target_words_bigendian(void);
953 static enum virtio_device_endian virtio_default_endian(void)
954 {
955     if (target_words_bigendian()) {
956         return VIRTIO_DEVICE_ENDIAN_BIG;
957     } else {
958         return VIRTIO_DEVICE_ENDIAN_LITTLE;
959     }
960 }
961 
962 static enum virtio_device_endian virtio_current_cpu_endian(void)
963 {
964     CPUClass *cc = CPU_GET_CLASS(current_cpu);
965 
966     if (cc->virtio_is_big_endian(current_cpu)) {
967         return VIRTIO_DEVICE_ENDIAN_BIG;
968     } else {
969         return VIRTIO_DEVICE_ENDIAN_LITTLE;
970     }
971 }
972 
973 void virtio_reset(void *opaque)
974 {
975     VirtIODevice *vdev = opaque;
976     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
977     int i;
978 
979     virtio_set_status(vdev, 0);
980     if (current_cpu) {
981         /* Guest initiated reset */
982         vdev->device_endian = virtio_current_cpu_endian();
983     } else {
984         /* System reset */
985         vdev->device_endian = virtio_default_endian();
986     }
987 
988     if (k->reset) {
989         k->reset(vdev);
990     }
991 
992     vdev->broken = false;
993     vdev->guest_features = 0;
994     vdev->queue_sel = 0;
995     vdev->status = 0;
996     atomic_set(&vdev->isr, 0);
997     vdev->config_vector = VIRTIO_NO_VECTOR;
998     virtio_notify_vector(vdev, vdev->config_vector);
999 
1000     for(i = 0; i < VIRTIO_QUEUE_MAX; i++) {
1001         vdev->vq[i].vring.desc = 0;
1002         vdev->vq[i].vring.avail = 0;
1003         vdev->vq[i].vring.used = 0;
1004         vdev->vq[i].last_avail_idx = 0;
1005         vdev->vq[i].shadow_avail_idx = 0;
1006         vdev->vq[i].used_idx = 0;
1007         virtio_queue_set_vector(vdev, i, VIRTIO_NO_VECTOR);
1008         vdev->vq[i].signalled_used = 0;
1009         vdev->vq[i].signalled_used_valid = false;
1010         vdev->vq[i].notification = true;
1011         vdev->vq[i].vring.num = vdev->vq[i].vring.num_default;
1012         vdev->vq[i].inuse = 0;
1013     }
1014 }
1015 
1016 uint32_t virtio_config_readb(VirtIODevice *vdev, uint32_t addr)
1017 {
1018     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1019     uint8_t val;
1020 
1021     if (addr + sizeof(val) > vdev->config_len) {
1022         return (uint32_t)-1;
1023     }
1024 
1025     k->get_config(vdev, vdev->config);
1026 
1027     val = ldub_p(vdev->config + addr);
1028     return val;
1029 }
1030 
1031 uint32_t virtio_config_readw(VirtIODevice *vdev, uint32_t addr)
1032 {
1033     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1034     uint16_t val;
1035 
1036     if (addr + sizeof(val) > vdev->config_len) {
1037         return (uint32_t)-1;
1038     }
1039 
1040     k->get_config(vdev, vdev->config);
1041 
1042     val = lduw_p(vdev->config + addr);
1043     return val;
1044 }
1045 
1046 uint32_t virtio_config_readl(VirtIODevice *vdev, uint32_t addr)
1047 {
1048     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1049     uint32_t val;
1050 
1051     if (addr + sizeof(val) > vdev->config_len) {
1052         return (uint32_t)-1;
1053     }
1054 
1055     k->get_config(vdev, vdev->config);
1056 
1057     val = ldl_p(vdev->config + addr);
1058     return val;
1059 }
1060 
1061 void virtio_config_writeb(VirtIODevice *vdev, uint32_t addr, uint32_t data)
1062 {
1063     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1064     uint8_t val = data;
1065 
1066     if (addr + sizeof(val) > vdev->config_len) {
1067         return;
1068     }
1069 
1070     stb_p(vdev->config + addr, val);
1071 
1072     if (k->set_config) {
1073         k->set_config(vdev, vdev->config);
1074     }
1075 }
1076 
1077 void virtio_config_writew(VirtIODevice *vdev, uint32_t addr, uint32_t data)
1078 {
1079     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1080     uint16_t val = data;
1081 
1082     if (addr + sizeof(val) > vdev->config_len) {
1083         return;
1084     }
1085 
1086     stw_p(vdev->config + addr, val);
1087 
1088     if (k->set_config) {
1089         k->set_config(vdev, vdev->config);
1090     }
1091 }
1092 
1093 void virtio_config_writel(VirtIODevice *vdev, uint32_t addr, uint32_t data)
1094 {
1095     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1096     uint32_t val = data;
1097 
1098     if (addr + sizeof(val) > vdev->config_len) {
1099         return;
1100     }
1101 
1102     stl_p(vdev->config + addr, val);
1103 
1104     if (k->set_config) {
1105         k->set_config(vdev, vdev->config);
1106     }
1107 }
1108 
1109 uint32_t virtio_config_modern_readb(VirtIODevice *vdev, uint32_t addr)
1110 {
1111     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1112     uint8_t val;
1113 
1114     if (addr + sizeof(val) > vdev->config_len) {
1115         return (uint32_t)-1;
1116     }
1117 
1118     k->get_config(vdev, vdev->config);
1119 
1120     val = ldub_p(vdev->config + addr);
1121     return val;
1122 }
1123 
1124 uint32_t virtio_config_modern_readw(VirtIODevice *vdev, uint32_t addr)
1125 {
1126     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1127     uint16_t val;
1128 
1129     if (addr + sizeof(val) > vdev->config_len) {
1130         return (uint32_t)-1;
1131     }
1132 
1133     k->get_config(vdev, vdev->config);
1134 
1135     val = lduw_le_p(vdev->config + addr);
1136     return val;
1137 }
1138 
1139 uint32_t virtio_config_modern_readl(VirtIODevice *vdev, uint32_t addr)
1140 {
1141     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1142     uint32_t val;
1143 
1144     if (addr + sizeof(val) > vdev->config_len) {
1145         return (uint32_t)-1;
1146     }
1147 
1148     k->get_config(vdev, vdev->config);
1149 
1150     val = ldl_le_p(vdev->config + addr);
1151     return val;
1152 }
1153 
1154 void virtio_config_modern_writeb(VirtIODevice *vdev,
1155                                  uint32_t addr, uint32_t data)
1156 {
1157     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1158     uint8_t val = data;
1159 
1160     if (addr + sizeof(val) > vdev->config_len) {
1161         return;
1162     }
1163 
1164     stb_p(vdev->config + addr, val);
1165 
1166     if (k->set_config) {
1167         k->set_config(vdev, vdev->config);
1168     }
1169 }
1170 
1171 void virtio_config_modern_writew(VirtIODevice *vdev,
1172                                  uint32_t addr, uint32_t data)
1173 {
1174     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1175     uint16_t val = data;
1176 
1177     if (addr + sizeof(val) > vdev->config_len) {
1178         return;
1179     }
1180 
1181     stw_le_p(vdev->config + addr, val);
1182 
1183     if (k->set_config) {
1184         k->set_config(vdev, vdev->config);
1185     }
1186 }
1187 
1188 void virtio_config_modern_writel(VirtIODevice *vdev,
1189                                  uint32_t addr, uint32_t data)
1190 {
1191     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1192     uint32_t val = data;
1193 
1194     if (addr + sizeof(val) > vdev->config_len) {
1195         return;
1196     }
1197 
1198     stl_le_p(vdev->config + addr, val);
1199 
1200     if (k->set_config) {
1201         k->set_config(vdev, vdev->config);
1202     }
1203 }
1204 
1205 void virtio_queue_set_addr(VirtIODevice *vdev, int n, hwaddr addr)
1206 {
1207     vdev->vq[n].vring.desc = addr;
1208     virtio_queue_update_rings(vdev, n);
1209 }
1210 
1211 hwaddr virtio_queue_get_addr(VirtIODevice *vdev, int n)
1212 {
1213     return vdev->vq[n].vring.desc;
1214 }
1215 
1216 void virtio_queue_set_rings(VirtIODevice *vdev, int n, hwaddr desc,
1217                             hwaddr avail, hwaddr used)
1218 {
1219     vdev->vq[n].vring.desc = desc;
1220     vdev->vq[n].vring.avail = avail;
1221     vdev->vq[n].vring.used = used;
1222 }
1223 
1224 void virtio_queue_set_num(VirtIODevice *vdev, int n, int num)
1225 {
1226     /* Don't allow guest to flip queue between existent and
1227      * nonexistent states, or to set it to an invalid size.
1228      */
1229     if (!!num != !!vdev->vq[n].vring.num ||
1230         num > VIRTQUEUE_MAX_SIZE ||
1231         num < 0) {
1232         return;
1233     }
1234     vdev->vq[n].vring.num = num;
1235 }
1236 
1237 VirtQueue *virtio_vector_first_queue(VirtIODevice *vdev, uint16_t vector)
1238 {
1239     return QLIST_FIRST(&vdev->vector_queues[vector]);
1240 }
1241 
1242 VirtQueue *virtio_vector_next_queue(VirtQueue *vq)
1243 {
1244     return QLIST_NEXT(vq, node);
1245 }
1246 
1247 int virtio_queue_get_num(VirtIODevice *vdev, int n)
1248 {
1249     return vdev->vq[n].vring.num;
1250 }
1251 
1252 int virtio_queue_get_max_num(VirtIODevice *vdev, int n)
1253 {
1254     return vdev->vq[n].vring.num_default;
1255 }
1256 
1257 int virtio_get_num_queues(VirtIODevice *vdev)
1258 {
1259     int i;
1260 
1261     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
1262         if (!virtio_queue_get_num(vdev, i)) {
1263             break;
1264         }
1265     }
1266 
1267     return i;
1268 }
1269 
1270 void virtio_queue_set_align(VirtIODevice *vdev, int n, int align)
1271 {
1272     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
1273     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
1274 
1275     /* virtio-1 compliant devices cannot change the alignment */
1276     if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
1277         error_report("tried to modify queue alignment for virtio-1 device");
1278         return;
1279     }
1280     /* Check that the transport told us it was going to do this
1281      * (so a buggy transport will immediately assert rather than
1282      * silently failing to migrate this state)
1283      */
1284     assert(k->has_variable_vring_alignment);
1285 
1286     vdev->vq[n].vring.align = align;
1287     virtio_queue_update_rings(vdev, n);
1288 }
1289 
1290 static void virtio_queue_notify_aio_vq(VirtQueue *vq)
1291 {
1292     if (vq->vring.desc && vq->handle_aio_output) {
1293         VirtIODevice *vdev = vq->vdev;
1294 
1295         trace_virtio_queue_notify(vdev, vq - vdev->vq, vq);
1296         vq->handle_aio_output(vdev, vq);
1297     }
1298 }
1299 
1300 static void virtio_queue_notify_vq(VirtQueue *vq)
1301 {
1302     if (vq->vring.desc && vq->handle_output) {
1303         VirtIODevice *vdev = vq->vdev;
1304 
1305         if (unlikely(vdev->broken)) {
1306             return;
1307         }
1308 
1309         trace_virtio_queue_notify(vdev, vq - vdev->vq, vq);
1310         vq->handle_output(vdev, vq);
1311     }
1312 }
1313 
1314 void virtio_queue_notify(VirtIODevice *vdev, int n)
1315 {
1316     virtio_queue_notify_vq(&vdev->vq[n]);
1317 }
1318 
1319 uint16_t virtio_queue_vector(VirtIODevice *vdev, int n)
1320 {
1321     return n < VIRTIO_QUEUE_MAX ? vdev->vq[n].vector :
1322         VIRTIO_NO_VECTOR;
1323 }
1324 
1325 void virtio_queue_set_vector(VirtIODevice *vdev, int n, uint16_t vector)
1326 {
1327     VirtQueue *vq = &vdev->vq[n];
1328 
1329     if (n < VIRTIO_QUEUE_MAX) {
1330         if (vdev->vector_queues &&
1331             vdev->vq[n].vector != VIRTIO_NO_VECTOR) {
1332             QLIST_REMOVE(vq, node);
1333         }
1334         vdev->vq[n].vector = vector;
1335         if (vdev->vector_queues &&
1336             vector != VIRTIO_NO_VECTOR) {
1337             QLIST_INSERT_HEAD(&vdev->vector_queues[vector], vq, node);
1338         }
1339     }
1340 }
1341 
1342 VirtQueue *virtio_add_queue(VirtIODevice *vdev, int queue_size,
1343                             VirtIOHandleOutput handle_output)
1344 {
1345     int i;
1346 
1347     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
1348         if (vdev->vq[i].vring.num == 0)
1349             break;
1350     }
1351 
1352     if (i == VIRTIO_QUEUE_MAX || queue_size > VIRTQUEUE_MAX_SIZE)
1353         abort();
1354 
1355     vdev->vq[i].vring.num = queue_size;
1356     vdev->vq[i].vring.num_default = queue_size;
1357     vdev->vq[i].vring.align = VIRTIO_PCI_VRING_ALIGN;
1358     vdev->vq[i].handle_output = handle_output;
1359     vdev->vq[i].handle_aio_output = NULL;
1360 
1361     return &vdev->vq[i];
1362 }
1363 
1364 void virtio_del_queue(VirtIODevice *vdev, int n)
1365 {
1366     if (n < 0 || n >= VIRTIO_QUEUE_MAX) {
1367         abort();
1368     }
1369 
1370     vdev->vq[n].vring.num = 0;
1371     vdev->vq[n].vring.num_default = 0;
1372 }
1373 
1374 static void virtio_set_isr(VirtIODevice *vdev, int value)
1375 {
1376     uint8_t old = atomic_read(&vdev->isr);
1377 
1378     /* Do not write ISR if it does not change, so that its cacheline remains
1379      * shared in the common case where the guest does not read it.
1380      */
1381     if ((old & value) != value) {
1382         atomic_or(&vdev->isr, value);
1383     }
1384 }
1385 
1386 bool virtio_should_notify(VirtIODevice *vdev, VirtQueue *vq)
1387 {
1388     uint16_t old, new;
1389     bool v;
1390     /* We need to expose used array entries before checking used event. */
1391     smp_mb();
1392     /* Always notify when queue is empty (when feature acknowledge) */
1393     if (virtio_vdev_has_feature(vdev, VIRTIO_F_NOTIFY_ON_EMPTY) &&
1394         !vq->inuse && virtio_queue_empty(vq)) {
1395         return true;
1396     }
1397 
1398     if (!virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX)) {
1399         return !(vring_avail_flags(vq) & VRING_AVAIL_F_NO_INTERRUPT);
1400     }
1401 
1402     v = vq->signalled_used_valid;
1403     vq->signalled_used_valid = true;
1404     old = vq->signalled_used;
1405     new = vq->signalled_used = vq->used_idx;
1406     return !v || vring_need_event(vring_get_used_event(vq), new, old);
1407 }
1408 
1409 void virtio_notify_irqfd(VirtIODevice *vdev, VirtQueue *vq)
1410 {
1411     if (!virtio_should_notify(vdev, vq)) {
1412         return;
1413     }
1414 
1415     trace_virtio_notify_irqfd(vdev, vq);
1416 
1417     /*
1418      * virtio spec 1.0 says ISR bit 0 should be ignored with MSI, but
1419      * windows drivers included in virtio-win 1.8.0 (circa 2015) are
1420      * incorrectly polling this bit during crashdump and hibernation
1421      * in MSI mode, causing a hang if this bit is never updated.
1422      * Recent releases of Windows do not really shut down, but rather
1423      * log out and hibernate to make the next startup faster.  Hence,
1424      * this manifested as a more serious hang during shutdown with
1425      *
1426      * Next driver release from 2016 fixed this problem, so working around it
1427      * is not a must, but it's easy to do so let's do it here.
1428      *
1429      * Note: it's safe to update ISR from any thread as it was switched
1430      * to an atomic operation.
1431      */
1432     virtio_set_isr(vq->vdev, 0x1);
1433     event_notifier_set(&vq->guest_notifier);
1434 }
1435 
1436 void virtio_notify(VirtIODevice *vdev, VirtQueue *vq)
1437 {
1438     if (!virtio_should_notify(vdev, vq)) {
1439         return;
1440     }
1441 
1442     trace_virtio_notify(vdev, vq);
1443     virtio_set_isr(vq->vdev, 0x1);
1444     virtio_notify_vector(vdev, vq->vector);
1445 }
1446 
1447 void virtio_notify_config(VirtIODevice *vdev)
1448 {
1449     if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK))
1450         return;
1451 
1452     virtio_set_isr(vdev, 0x3);
1453     vdev->generation++;
1454     virtio_notify_vector(vdev, vdev->config_vector);
1455 }
1456 
1457 static bool virtio_device_endian_needed(void *opaque)
1458 {
1459     VirtIODevice *vdev = opaque;
1460 
1461     assert(vdev->device_endian != VIRTIO_DEVICE_ENDIAN_UNKNOWN);
1462     if (!virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
1463         return vdev->device_endian != virtio_default_endian();
1464     }
1465     /* Devices conforming to VIRTIO 1.0 or later are always LE. */
1466     return vdev->device_endian != VIRTIO_DEVICE_ENDIAN_LITTLE;
1467 }
1468 
1469 static bool virtio_64bit_features_needed(void *opaque)
1470 {
1471     VirtIODevice *vdev = opaque;
1472 
1473     return (vdev->host_features >> 32) != 0;
1474 }
1475 
1476 static bool virtio_virtqueue_needed(void *opaque)
1477 {
1478     VirtIODevice *vdev = opaque;
1479 
1480     return virtio_host_has_feature(vdev, VIRTIO_F_VERSION_1);
1481 }
1482 
1483 static bool virtio_ringsize_needed(void *opaque)
1484 {
1485     VirtIODevice *vdev = opaque;
1486     int i;
1487 
1488     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
1489         if (vdev->vq[i].vring.num != vdev->vq[i].vring.num_default) {
1490             return true;
1491         }
1492     }
1493     return false;
1494 }
1495 
1496 static bool virtio_extra_state_needed(void *opaque)
1497 {
1498     VirtIODevice *vdev = opaque;
1499     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
1500     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
1501 
1502     return k->has_extra_state &&
1503         k->has_extra_state(qbus->parent);
1504 }
1505 
1506 static bool virtio_broken_needed(void *opaque)
1507 {
1508     VirtIODevice *vdev = opaque;
1509 
1510     return vdev->broken;
1511 }
1512 
1513 static const VMStateDescription vmstate_virtqueue = {
1514     .name = "virtqueue_state",
1515     .version_id = 1,
1516     .minimum_version_id = 1,
1517     .fields = (VMStateField[]) {
1518         VMSTATE_UINT64(vring.avail, struct VirtQueue),
1519         VMSTATE_UINT64(vring.used, struct VirtQueue),
1520         VMSTATE_END_OF_LIST()
1521     }
1522 };
1523 
1524 static const VMStateDescription vmstate_virtio_virtqueues = {
1525     .name = "virtio/virtqueues",
1526     .version_id = 1,
1527     .minimum_version_id = 1,
1528     .needed = &virtio_virtqueue_needed,
1529     .fields = (VMStateField[]) {
1530         VMSTATE_STRUCT_VARRAY_POINTER_KNOWN(vq, struct VirtIODevice,
1531                       VIRTIO_QUEUE_MAX, 0, vmstate_virtqueue, VirtQueue),
1532         VMSTATE_END_OF_LIST()
1533     }
1534 };
1535 
1536 static const VMStateDescription vmstate_ringsize = {
1537     .name = "ringsize_state",
1538     .version_id = 1,
1539     .minimum_version_id = 1,
1540     .fields = (VMStateField[]) {
1541         VMSTATE_UINT32(vring.num_default, struct VirtQueue),
1542         VMSTATE_END_OF_LIST()
1543     }
1544 };
1545 
1546 static const VMStateDescription vmstate_virtio_ringsize = {
1547     .name = "virtio/ringsize",
1548     .version_id = 1,
1549     .minimum_version_id = 1,
1550     .needed = &virtio_ringsize_needed,
1551     .fields = (VMStateField[]) {
1552         VMSTATE_STRUCT_VARRAY_POINTER_KNOWN(vq, struct VirtIODevice,
1553                       VIRTIO_QUEUE_MAX, 0, vmstate_ringsize, VirtQueue),
1554         VMSTATE_END_OF_LIST()
1555     }
1556 };
1557 
1558 static int get_extra_state(QEMUFile *f, void *pv, size_t size,
1559                            VMStateField *field)
1560 {
1561     VirtIODevice *vdev = pv;
1562     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
1563     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
1564 
1565     if (!k->load_extra_state) {
1566         return -1;
1567     } else {
1568         return k->load_extra_state(qbus->parent, f);
1569     }
1570 }
1571 
1572 static int put_extra_state(QEMUFile *f, void *pv, size_t size,
1573                            VMStateField *field, QJSON *vmdesc)
1574 {
1575     VirtIODevice *vdev = pv;
1576     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
1577     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
1578 
1579     k->save_extra_state(qbus->parent, f);
1580     return 0;
1581 }
1582 
1583 static const VMStateInfo vmstate_info_extra_state = {
1584     .name = "virtqueue_extra_state",
1585     .get = get_extra_state,
1586     .put = put_extra_state,
1587 };
1588 
1589 static const VMStateDescription vmstate_virtio_extra_state = {
1590     .name = "virtio/extra_state",
1591     .version_id = 1,
1592     .minimum_version_id = 1,
1593     .needed = &virtio_extra_state_needed,
1594     .fields = (VMStateField[]) {
1595         {
1596             .name         = "extra_state",
1597             .version_id   = 0,
1598             .field_exists = NULL,
1599             .size         = 0,
1600             .info         = &vmstate_info_extra_state,
1601             .flags        = VMS_SINGLE,
1602             .offset       = 0,
1603         },
1604         VMSTATE_END_OF_LIST()
1605     }
1606 };
1607 
1608 static const VMStateDescription vmstate_virtio_device_endian = {
1609     .name = "virtio/device_endian",
1610     .version_id = 1,
1611     .minimum_version_id = 1,
1612     .needed = &virtio_device_endian_needed,
1613     .fields = (VMStateField[]) {
1614         VMSTATE_UINT8(device_endian, VirtIODevice),
1615         VMSTATE_END_OF_LIST()
1616     }
1617 };
1618 
1619 static const VMStateDescription vmstate_virtio_64bit_features = {
1620     .name = "virtio/64bit_features",
1621     .version_id = 1,
1622     .minimum_version_id = 1,
1623     .needed = &virtio_64bit_features_needed,
1624     .fields = (VMStateField[]) {
1625         VMSTATE_UINT64(guest_features, VirtIODevice),
1626         VMSTATE_END_OF_LIST()
1627     }
1628 };
1629 
1630 static const VMStateDescription vmstate_virtio_broken = {
1631     .name = "virtio/broken",
1632     .version_id = 1,
1633     .minimum_version_id = 1,
1634     .needed = &virtio_broken_needed,
1635     .fields = (VMStateField[]) {
1636         VMSTATE_BOOL(broken, VirtIODevice),
1637         VMSTATE_END_OF_LIST()
1638     }
1639 };
1640 
1641 static const VMStateDescription vmstate_virtio = {
1642     .name = "virtio",
1643     .version_id = 1,
1644     .minimum_version_id = 1,
1645     .minimum_version_id_old = 1,
1646     .fields = (VMStateField[]) {
1647         VMSTATE_END_OF_LIST()
1648     },
1649     .subsections = (const VMStateDescription*[]) {
1650         &vmstate_virtio_device_endian,
1651         &vmstate_virtio_64bit_features,
1652         &vmstate_virtio_virtqueues,
1653         &vmstate_virtio_ringsize,
1654         &vmstate_virtio_broken,
1655         &vmstate_virtio_extra_state,
1656         NULL
1657     }
1658 };
1659 
1660 void virtio_save(VirtIODevice *vdev, QEMUFile *f)
1661 {
1662     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
1663     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
1664     VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev);
1665     uint32_t guest_features_lo = (vdev->guest_features & 0xffffffff);
1666     int i;
1667 
1668     if (k->save_config) {
1669         k->save_config(qbus->parent, f);
1670     }
1671 
1672     qemu_put_8s(f, &vdev->status);
1673     qemu_put_8s(f, &vdev->isr);
1674     qemu_put_be16s(f, &vdev->queue_sel);
1675     qemu_put_be32s(f, &guest_features_lo);
1676     qemu_put_be32(f, vdev->config_len);
1677     qemu_put_buffer(f, vdev->config, vdev->config_len);
1678 
1679     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
1680         if (vdev->vq[i].vring.num == 0)
1681             break;
1682     }
1683 
1684     qemu_put_be32(f, i);
1685 
1686     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
1687         if (vdev->vq[i].vring.num == 0)
1688             break;
1689 
1690         qemu_put_be32(f, vdev->vq[i].vring.num);
1691         if (k->has_variable_vring_alignment) {
1692             qemu_put_be32(f, vdev->vq[i].vring.align);
1693         }
1694         /* XXX virtio-1 devices */
1695         qemu_put_be64(f, vdev->vq[i].vring.desc);
1696         qemu_put_be16s(f, &vdev->vq[i].last_avail_idx);
1697         if (k->save_queue) {
1698             k->save_queue(qbus->parent, i, f);
1699         }
1700     }
1701 
1702     if (vdc->save != NULL) {
1703         vdc->save(vdev, f);
1704     }
1705 
1706     if (vdc->vmsd) {
1707         vmstate_save_state(f, vdc->vmsd, vdev, NULL);
1708     }
1709 
1710     /* Subsections */
1711     vmstate_save_state(f, &vmstate_virtio, vdev, NULL);
1712 }
1713 
1714 /* A wrapper for use as a VMState .put function */
1715 static int virtio_device_put(QEMUFile *f, void *opaque, size_t size,
1716                               VMStateField *field, QJSON *vmdesc)
1717 {
1718     virtio_save(VIRTIO_DEVICE(opaque), f);
1719 
1720     return 0;
1721 }
1722 
1723 /* A wrapper for use as a VMState .get function */
1724 static int virtio_device_get(QEMUFile *f, void *opaque, size_t size,
1725                              VMStateField *field)
1726 {
1727     VirtIODevice *vdev = VIRTIO_DEVICE(opaque);
1728     DeviceClass *dc = DEVICE_CLASS(VIRTIO_DEVICE_GET_CLASS(vdev));
1729 
1730     return virtio_load(vdev, f, dc->vmsd->version_id);
1731 }
1732 
1733 const VMStateInfo  virtio_vmstate_info = {
1734     .name = "virtio",
1735     .get = virtio_device_get,
1736     .put = virtio_device_put,
1737 };
1738 
1739 static int virtio_set_features_nocheck(VirtIODevice *vdev, uint64_t val)
1740 {
1741     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1742     bool bad = (val & ~(vdev->host_features)) != 0;
1743 
1744     val &= vdev->host_features;
1745     if (k->set_features) {
1746         k->set_features(vdev, val);
1747     }
1748     vdev->guest_features = val;
1749     return bad ? -1 : 0;
1750 }
1751 
1752 int virtio_set_features(VirtIODevice *vdev, uint64_t val)
1753 {
1754    /*
1755      * The driver must not attempt to set features after feature negotiation
1756      * has finished.
1757      */
1758     if (vdev->status & VIRTIO_CONFIG_S_FEATURES_OK) {
1759         return -EINVAL;
1760     }
1761     return virtio_set_features_nocheck(vdev, val);
1762 }
1763 
1764 int virtio_load(VirtIODevice *vdev, QEMUFile *f, int version_id)
1765 {
1766     int i, ret;
1767     int32_t config_len;
1768     uint32_t num;
1769     uint32_t features;
1770     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
1771     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
1772     VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev);
1773 
1774     /*
1775      * We poison the endianness to ensure it does not get used before
1776      * subsections have been loaded.
1777      */
1778     vdev->device_endian = VIRTIO_DEVICE_ENDIAN_UNKNOWN;
1779 
1780     if (k->load_config) {
1781         ret = k->load_config(qbus->parent, f);
1782         if (ret)
1783             return ret;
1784     }
1785 
1786     qemu_get_8s(f, &vdev->status);
1787     qemu_get_8s(f, &vdev->isr);
1788     qemu_get_be16s(f, &vdev->queue_sel);
1789     if (vdev->queue_sel >= VIRTIO_QUEUE_MAX) {
1790         return -1;
1791     }
1792     qemu_get_be32s(f, &features);
1793 
1794     /*
1795      * Temporarily set guest_features low bits - needed by
1796      * virtio net load code testing for VIRTIO_NET_F_CTRL_GUEST_OFFLOADS
1797      * VIRTIO_NET_F_GUEST_ANNOUNCE and VIRTIO_NET_F_CTRL_VQ.
1798      *
1799      * Note: devices should always test host features in future - don't create
1800      * new dependencies like this.
1801      */
1802     vdev->guest_features = features;
1803 
1804     config_len = qemu_get_be32(f);
1805 
1806     /*
1807      * There are cases where the incoming config can be bigger or smaller
1808      * than what we have; so load what we have space for, and skip
1809      * any excess that's in the stream.
1810      */
1811     qemu_get_buffer(f, vdev->config, MIN(config_len, vdev->config_len));
1812 
1813     while (config_len > vdev->config_len) {
1814         qemu_get_byte(f);
1815         config_len--;
1816     }
1817 
1818     num = qemu_get_be32(f);
1819 
1820     if (num > VIRTIO_QUEUE_MAX) {
1821         error_report("Invalid number of virtqueues: 0x%x", num);
1822         return -1;
1823     }
1824 
1825     for (i = 0; i < num; i++) {
1826         vdev->vq[i].vring.num = qemu_get_be32(f);
1827         if (k->has_variable_vring_alignment) {
1828             vdev->vq[i].vring.align = qemu_get_be32(f);
1829         }
1830         vdev->vq[i].vring.desc = qemu_get_be64(f);
1831         qemu_get_be16s(f, &vdev->vq[i].last_avail_idx);
1832         vdev->vq[i].signalled_used_valid = false;
1833         vdev->vq[i].notification = true;
1834 
1835         if (vdev->vq[i].vring.desc) {
1836             /* XXX virtio-1 devices */
1837             virtio_queue_update_rings(vdev, i);
1838         } else if (vdev->vq[i].last_avail_idx) {
1839             error_report("VQ %d address 0x0 "
1840                          "inconsistent with Host index 0x%x",
1841                          i, vdev->vq[i].last_avail_idx);
1842                 return -1;
1843         }
1844         if (k->load_queue) {
1845             ret = k->load_queue(qbus->parent, i, f);
1846             if (ret)
1847                 return ret;
1848         }
1849     }
1850 
1851     virtio_notify_vector(vdev, VIRTIO_NO_VECTOR);
1852 
1853     if (vdc->load != NULL) {
1854         ret = vdc->load(vdev, f, version_id);
1855         if (ret) {
1856             return ret;
1857         }
1858     }
1859 
1860     if (vdc->vmsd) {
1861         ret = vmstate_load_state(f, vdc->vmsd, vdev, version_id);
1862         if (ret) {
1863             return ret;
1864         }
1865     }
1866 
1867     /* Subsections */
1868     ret = vmstate_load_state(f, &vmstate_virtio, vdev, 1);
1869     if (ret) {
1870         return ret;
1871     }
1872 
1873     if (vdev->device_endian == VIRTIO_DEVICE_ENDIAN_UNKNOWN) {
1874         vdev->device_endian = virtio_default_endian();
1875     }
1876 
1877     if (virtio_64bit_features_needed(vdev)) {
1878         /*
1879          * Subsection load filled vdev->guest_features.  Run them
1880          * through virtio_set_features to sanity-check them against
1881          * host_features.
1882          */
1883         uint64_t features64 = vdev->guest_features;
1884         if (virtio_set_features_nocheck(vdev, features64) < 0) {
1885             error_report("Features 0x%" PRIx64 " unsupported. "
1886                          "Allowed features: 0x%" PRIx64,
1887                          features64, vdev->host_features);
1888             return -1;
1889         }
1890     } else {
1891         if (virtio_set_features_nocheck(vdev, features) < 0) {
1892             error_report("Features 0x%x unsupported. "
1893                          "Allowed features: 0x%" PRIx64,
1894                          features, vdev->host_features);
1895             return -1;
1896         }
1897     }
1898 
1899     for (i = 0; i < num; i++) {
1900         if (vdev->vq[i].vring.desc) {
1901             uint16_t nheads;
1902             nheads = vring_avail_idx(&vdev->vq[i]) - vdev->vq[i].last_avail_idx;
1903             /* Check it isn't doing strange things with descriptor numbers. */
1904             if (nheads > vdev->vq[i].vring.num) {
1905                 error_report("VQ %d size 0x%x Guest index 0x%x "
1906                              "inconsistent with Host index 0x%x: delta 0x%x",
1907                              i, vdev->vq[i].vring.num,
1908                              vring_avail_idx(&vdev->vq[i]),
1909                              vdev->vq[i].last_avail_idx, nheads);
1910                 return -1;
1911             }
1912             vdev->vq[i].used_idx = vring_used_idx(&vdev->vq[i]);
1913             vdev->vq[i].shadow_avail_idx = vring_avail_idx(&vdev->vq[i]);
1914 
1915             /*
1916              * Some devices migrate VirtQueueElements that have been popped
1917              * from the avail ring but not yet returned to the used ring.
1918              * Since max ring size < UINT16_MAX it's safe to use modulo
1919              * UINT16_MAX + 1 subtraction.
1920              */
1921             vdev->vq[i].inuse = (uint16_t)(vdev->vq[i].last_avail_idx -
1922                                 vdev->vq[i].used_idx);
1923             if (vdev->vq[i].inuse > vdev->vq[i].vring.num) {
1924                 error_report("VQ %d size 0x%x < last_avail_idx 0x%x - "
1925                              "used_idx 0x%x",
1926                              i, vdev->vq[i].vring.num,
1927                              vdev->vq[i].last_avail_idx,
1928                              vdev->vq[i].used_idx);
1929                 return -1;
1930             }
1931         }
1932     }
1933 
1934     return 0;
1935 }
1936 
1937 void virtio_cleanup(VirtIODevice *vdev)
1938 {
1939     qemu_del_vm_change_state_handler(vdev->vmstate);
1940     g_free(vdev->config);
1941     g_free(vdev->vq);
1942     g_free(vdev->vector_queues);
1943 }
1944 
1945 static void virtio_vmstate_change(void *opaque, int running, RunState state)
1946 {
1947     VirtIODevice *vdev = opaque;
1948     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
1949     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
1950     bool backend_run = running && (vdev->status & VIRTIO_CONFIG_S_DRIVER_OK);
1951     vdev->vm_running = running;
1952 
1953     if (backend_run) {
1954         virtio_set_status(vdev, vdev->status);
1955     }
1956 
1957     if (k->vmstate_change) {
1958         k->vmstate_change(qbus->parent, backend_run);
1959     }
1960 
1961     if (!backend_run) {
1962         virtio_set_status(vdev, vdev->status);
1963     }
1964 }
1965 
1966 void virtio_instance_init_common(Object *proxy_obj, void *data,
1967                                  size_t vdev_size, const char *vdev_name)
1968 {
1969     DeviceState *vdev = data;
1970 
1971     object_initialize(vdev, vdev_size, vdev_name);
1972     object_property_add_child(proxy_obj, "virtio-backend", OBJECT(vdev), NULL);
1973     object_unref(OBJECT(vdev));
1974     qdev_alias_all_properties(vdev, proxy_obj);
1975 }
1976 
1977 void virtio_init(VirtIODevice *vdev, const char *name,
1978                  uint16_t device_id, size_t config_size)
1979 {
1980     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
1981     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
1982     int i;
1983     int nvectors = k->query_nvectors ? k->query_nvectors(qbus->parent) : 0;
1984 
1985     if (nvectors) {
1986         vdev->vector_queues =
1987             g_malloc0(sizeof(*vdev->vector_queues) * nvectors);
1988     }
1989 
1990     vdev->device_id = device_id;
1991     vdev->status = 0;
1992     atomic_set(&vdev->isr, 0);
1993     vdev->queue_sel = 0;
1994     vdev->config_vector = VIRTIO_NO_VECTOR;
1995     vdev->vq = g_malloc0(sizeof(VirtQueue) * VIRTIO_QUEUE_MAX);
1996     vdev->vm_running = runstate_is_running();
1997     vdev->broken = false;
1998     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
1999         vdev->vq[i].vector = VIRTIO_NO_VECTOR;
2000         vdev->vq[i].vdev = vdev;
2001         vdev->vq[i].queue_index = i;
2002     }
2003 
2004     vdev->name = name;
2005     vdev->config_len = config_size;
2006     if (vdev->config_len) {
2007         vdev->config = g_malloc0(config_size);
2008     } else {
2009         vdev->config = NULL;
2010     }
2011     vdev->vmstate = qemu_add_vm_change_state_handler(virtio_vmstate_change,
2012                                                      vdev);
2013     vdev->device_endian = virtio_default_endian();
2014     vdev->use_guest_notifier_mask = true;
2015 }
2016 
2017 hwaddr virtio_queue_get_desc_addr(VirtIODevice *vdev, int n)
2018 {
2019     return vdev->vq[n].vring.desc;
2020 }
2021 
2022 hwaddr virtio_queue_get_avail_addr(VirtIODevice *vdev, int n)
2023 {
2024     return vdev->vq[n].vring.avail;
2025 }
2026 
2027 hwaddr virtio_queue_get_used_addr(VirtIODevice *vdev, int n)
2028 {
2029     return vdev->vq[n].vring.used;
2030 }
2031 
2032 hwaddr virtio_queue_get_desc_size(VirtIODevice *vdev, int n)
2033 {
2034     return sizeof(VRingDesc) * vdev->vq[n].vring.num;
2035 }
2036 
2037 hwaddr virtio_queue_get_avail_size(VirtIODevice *vdev, int n)
2038 {
2039     return offsetof(VRingAvail, ring) +
2040         sizeof(uint16_t) * vdev->vq[n].vring.num;
2041 }
2042 
2043 hwaddr virtio_queue_get_used_size(VirtIODevice *vdev, int n)
2044 {
2045     return offsetof(VRingUsed, ring) +
2046         sizeof(VRingUsedElem) * vdev->vq[n].vring.num;
2047 }
2048 
2049 uint16_t virtio_queue_get_last_avail_idx(VirtIODevice *vdev, int n)
2050 {
2051     return vdev->vq[n].last_avail_idx;
2052 }
2053 
2054 void virtio_queue_set_last_avail_idx(VirtIODevice *vdev, int n, uint16_t idx)
2055 {
2056     vdev->vq[n].last_avail_idx = idx;
2057     vdev->vq[n].shadow_avail_idx = idx;
2058 }
2059 
2060 void virtio_queue_update_used_idx(VirtIODevice *vdev, int n)
2061 {
2062     vdev->vq[n].used_idx = vring_used_idx(&vdev->vq[n]);
2063 }
2064 
2065 void virtio_queue_invalidate_signalled_used(VirtIODevice *vdev, int n)
2066 {
2067     vdev->vq[n].signalled_used_valid = false;
2068 }
2069 
2070 VirtQueue *virtio_get_queue(VirtIODevice *vdev, int n)
2071 {
2072     return vdev->vq + n;
2073 }
2074 
2075 uint16_t virtio_get_queue_index(VirtQueue *vq)
2076 {
2077     return vq->queue_index;
2078 }
2079 
2080 static void virtio_queue_guest_notifier_read(EventNotifier *n)
2081 {
2082     VirtQueue *vq = container_of(n, VirtQueue, guest_notifier);
2083     if (event_notifier_test_and_clear(n)) {
2084         virtio_notify_vector(vq->vdev, vq->vector);
2085     }
2086 }
2087 
2088 void virtio_queue_set_guest_notifier_fd_handler(VirtQueue *vq, bool assign,
2089                                                 bool with_irqfd)
2090 {
2091     if (assign && !with_irqfd) {
2092         event_notifier_set_handler(&vq->guest_notifier,
2093                                    virtio_queue_guest_notifier_read);
2094     } else {
2095         event_notifier_set_handler(&vq->guest_notifier, NULL);
2096     }
2097     if (!assign) {
2098         /* Test and clear notifier before closing it,
2099          * in case poll callback didn't have time to run. */
2100         virtio_queue_guest_notifier_read(&vq->guest_notifier);
2101     }
2102 }
2103 
2104 EventNotifier *virtio_queue_get_guest_notifier(VirtQueue *vq)
2105 {
2106     return &vq->guest_notifier;
2107 }
2108 
2109 static void virtio_queue_host_notifier_aio_read(EventNotifier *n)
2110 {
2111     VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
2112     if (event_notifier_test_and_clear(n)) {
2113         virtio_queue_notify_aio_vq(vq);
2114     }
2115 }
2116 
2117 static void virtio_queue_host_notifier_aio_poll_begin(EventNotifier *n)
2118 {
2119     VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
2120 
2121     virtio_queue_set_notification(vq, 0);
2122 }
2123 
2124 static bool virtio_queue_host_notifier_aio_poll(void *opaque)
2125 {
2126     EventNotifier *n = opaque;
2127     VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
2128 
2129     if (virtio_queue_empty(vq)) {
2130         return false;
2131     }
2132 
2133     virtio_queue_notify_aio_vq(vq);
2134 
2135     /* In case the handler function re-enabled notifications */
2136     virtio_queue_set_notification(vq, 0);
2137     return true;
2138 }
2139 
2140 static void virtio_queue_host_notifier_aio_poll_end(EventNotifier *n)
2141 {
2142     VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
2143 
2144     /* Caller polls once more after this to catch requests that race with us */
2145     virtio_queue_set_notification(vq, 1);
2146 }
2147 
2148 void virtio_queue_aio_set_host_notifier_handler(VirtQueue *vq, AioContext *ctx,
2149                                                 VirtIOHandleOutput handle_output)
2150 {
2151     if (handle_output) {
2152         vq->handle_aio_output = handle_output;
2153         aio_set_event_notifier(ctx, &vq->host_notifier, true,
2154                                virtio_queue_host_notifier_aio_read,
2155                                virtio_queue_host_notifier_aio_poll);
2156         aio_set_event_notifier_poll(ctx, &vq->host_notifier,
2157                                     virtio_queue_host_notifier_aio_poll_begin,
2158                                     virtio_queue_host_notifier_aio_poll_end);
2159     } else {
2160         aio_set_event_notifier(ctx, &vq->host_notifier, true, NULL, NULL);
2161         /* Test and clear notifier before after disabling event,
2162          * in case poll callback didn't have time to run. */
2163         virtio_queue_host_notifier_aio_read(&vq->host_notifier);
2164         vq->handle_aio_output = NULL;
2165     }
2166 }
2167 
2168 void virtio_queue_host_notifier_read(EventNotifier *n)
2169 {
2170     VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
2171     if (event_notifier_test_and_clear(n)) {
2172         virtio_queue_notify_vq(vq);
2173     }
2174 }
2175 
2176 EventNotifier *virtio_queue_get_host_notifier(VirtQueue *vq)
2177 {
2178     return &vq->host_notifier;
2179 }
2180 
2181 void virtio_device_set_child_bus_name(VirtIODevice *vdev, char *bus_name)
2182 {
2183     g_free(vdev->bus_name);
2184     vdev->bus_name = g_strdup(bus_name);
2185 }
2186 
2187 void GCC_FMT_ATTR(2, 3) virtio_error(VirtIODevice *vdev, const char *fmt, ...)
2188 {
2189     va_list ap;
2190 
2191     va_start(ap, fmt);
2192     error_vreport(fmt, ap);
2193     va_end(ap);
2194 
2195     vdev->broken = true;
2196 
2197     if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
2198         virtio_set_status(vdev, vdev->status | VIRTIO_CONFIG_S_NEEDS_RESET);
2199         virtio_notify_config(vdev);
2200     }
2201 }
2202 
2203 static void virtio_device_realize(DeviceState *dev, Error **errp)
2204 {
2205     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
2206     VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(dev);
2207     Error *err = NULL;
2208 
2209     /* Devices should either use vmsd or the load/save methods */
2210     assert(!vdc->vmsd || !vdc->load);
2211 
2212     if (vdc->realize != NULL) {
2213         vdc->realize(dev, &err);
2214         if (err != NULL) {
2215             error_propagate(errp, err);
2216             return;
2217         }
2218     }
2219 
2220     virtio_bus_device_plugged(vdev, &err);
2221     if (err != NULL) {
2222         error_propagate(errp, err);
2223         return;
2224     }
2225 }
2226 
2227 static void virtio_device_unrealize(DeviceState *dev, Error **errp)
2228 {
2229     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
2230     VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(dev);
2231     Error *err = NULL;
2232 
2233     virtio_bus_device_unplugged(vdev);
2234 
2235     if (vdc->unrealize != NULL) {
2236         vdc->unrealize(dev, &err);
2237         if (err != NULL) {
2238             error_propagate(errp, err);
2239             return;
2240         }
2241     }
2242 
2243     g_free(vdev->bus_name);
2244     vdev->bus_name = NULL;
2245 }
2246 
2247 static Property virtio_properties[] = {
2248     DEFINE_VIRTIO_COMMON_FEATURES(VirtIODevice, host_features),
2249     DEFINE_PROP_END_OF_LIST(),
2250 };
2251 
2252 static int virtio_device_start_ioeventfd_impl(VirtIODevice *vdev)
2253 {
2254     VirtioBusState *qbus = VIRTIO_BUS(qdev_get_parent_bus(DEVICE(vdev)));
2255     int n, r, err;
2256 
2257     for (n = 0; n < VIRTIO_QUEUE_MAX; n++) {
2258         VirtQueue *vq = &vdev->vq[n];
2259         if (!virtio_queue_get_num(vdev, n)) {
2260             continue;
2261         }
2262         r = virtio_bus_set_host_notifier(qbus, n, true);
2263         if (r < 0) {
2264             err = r;
2265             goto assign_error;
2266         }
2267         event_notifier_set_handler(&vq->host_notifier,
2268                                    virtio_queue_host_notifier_read);
2269     }
2270 
2271     for (n = 0; n < VIRTIO_QUEUE_MAX; n++) {
2272         /* Kick right away to begin processing requests already in vring */
2273         VirtQueue *vq = &vdev->vq[n];
2274         if (!vq->vring.num) {
2275             continue;
2276         }
2277         event_notifier_set(&vq->host_notifier);
2278     }
2279     return 0;
2280 
2281 assign_error:
2282     while (--n >= 0) {
2283         VirtQueue *vq = &vdev->vq[n];
2284         if (!virtio_queue_get_num(vdev, n)) {
2285             continue;
2286         }
2287 
2288         event_notifier_set_handler(&vq->host_notifier, NULL);
2289         r = virtio_bus_set_host_notifier(qbus, n, false);
2290         assert(r >= 0);
2291     }
2292     return err;
2293 }
2294 
2295 int virtio_device_start_ioeventfd(VirtIODevice *vdev)
2296 {
2297     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2298     VirtioBusState *vbus = VIRTIO_BUS(qbus);
2299 
2300     return virtio_bus_start_ioeventfd(vbus);
2301 }
2302 
2303 static void virtio_device_stop_ioeventfd_impl(VirtIODevice *vdev)
2304 {
2305     VirtioBusState *qbus = VIRTIO_BUS(qdev_get_parent_bus(DEVICE(vdev)));
2306     int n, r;
2307 
2308     for (n = 0; n < VIRTIO_QUEUE_MAX; n++) {
2309         VirtQueue *vq = &vdev->vq[n];
2310 
2311         if (!virtio_queue_get_num(vdev, n)) {
2312             continue;
2313         }
2314         event_notifier_set_handler(&vq->host_notifier, NULL);
2315         r = virtio_bus_set_host_notifier(qbus, n, false);
2316         assert(r >= 0);
2317     }
2318 }
2319 
2320 void virtio_device_stop_ioeventfd(VirtIODevice *vdev)
2321 {
2322     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2323     VirtioBusState *vbus = VIRTIO_BUS(qbus);
2324 
2325     virtio_bus_stop_ioeventfd(vbus);
2326 }
2327 
2328 int virtio_device_grab_ioeventfd(VirtIODevice *vdev)
2329 {
2330     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2331     VirtioBusState *vbus = VIRTIO_BUS(qbus);
2332 
2333     return virtio_bus_grab_ioeventfd(vbus);
2334 }
2335 
2336 void virtio_device_release_ioeventfd(VirtIODevice *vdev)
2337 {
2338     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2339     VirtioBusState *vbus = VIRTIO_BUS(qbus);
2340 
2341     virtio_bus_release_ioeventfd(vbus);
2342 }
2343 
2344 static void virtio_device_class_init(ObjectClass *klass, void *data)
2345 {
2346     /* Set the default value here. */
2347     VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
2348     DeviceClass *dc = DEVICE_CLASS(klass);
2349 
2350     dc->realize = virtio_device_realize;
2351     dc->unrealize = virtio_device_unrealize;
2352     dc->bus_type = TYPE_VIRTIO_BUS;
2353     dc->props = virtio_properties;
2354     vdc->start_ioeventfd = virtio_device_start_ioeventfd_impl;
2355     vdc->stop_ioeventfd = virtio_device_stop_ioeventfd_impl;
2356 
2357     vdc->legacy_features |= VIRTIO_LEGACY_FEATURES;
2358 }
2359 
2360 bool virtio_device_ioeventfd_enabled(VirtIODevice *vdev)
2361 {
2362     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2363     VirtioBusState *vbus = VIRTIO_BUS(qbus);
2364 
2365     return virtio_bus_ioeventfd_enabled(vbus);
2366 }
2367 
2368 static const TypeInfo virtio_device_info = {
2369     .name = TYPE_VIRTIO_DEVICE,
2370     .parent = TYPE_DEVICE,
2371     .instance_size = sizeof(VirtIODevice),
2372     .class_init = virtio_device_class_init,
2373     .abstract = true,
2374     .class_size = sizeof(VirtioDeviceClass),
2375 };
2376 
2377 static void virtio_register_types(void)
2378 {
2379     type_register_static(&virtio_device_info);
2380 }
2381 
2382 type_init(virtio_register_types)
2383