xref: /openbmc/qemu/hw/virtio/virtio.c (revision d73abd6d)
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 <inttypes.h>
15 
16 #include "trace.h"
17 #include "exec/address-spaces.h"
18 #include "qemu/error-report.h"
19 #include "hw/virtio/virtio.h"
20 #include "qemu/atomic.h"
21 #include "hw/virtio/virtio-bus.h"
22 #include "migration/migration.h"
23 #include "hw/virtio/virtio-access.h"
24 
25 /*
26  * The alignment to use between consumer and producer parts of vring.
27  * x86 pagesize again. This is the default, used by transports like PCI
28  * which don't provide a means for the guest to tell the host the alignment.
29  */
30 #define VIRTIO_PCI_VRING_ALIGN         4096
31 
32 typedef struct VRingDesc
33 {
34     uint64_t addr;
35     uint32_t len;
36     uint16_t flags;
37     uint16_t next;
38 } VRingDesc;
39 
40 typedef struct VRingAvail
41 {
42     uint16_t flags;
43     uint16_t idx;
44     uint16_t ring[0];
45 } VRingAvail;
46 
47 typedef struct VRingUsedElem
48 {
49     uint32_t id;
50     uint32_t len;
51 } VRingUsedElem;
52 
53 typedef struct VRingUsed
54 {
55     uint16_t flags;
56     uint16_t idx;
57     VRingUsedElem ring[0];
58 } VRingUsed;
59 
60 typedef struct VRing
61 {
62     unsigned int num;
63     unsigned int num_default;
64     unsigned int align;
65     hwaddr desc;
66     hwaddr avail;
67     hwaddr used;
68 } VRing;
69 
70 struct VirtQueue
71 {
72     VRing vring;
73     uint16_t last_avail_idx;
74     /* Last used index value we have signalled on */
75     uint16_t signalled_used;
76 
77     /* Last used index value we have signalled on */
78     bool signalled_used_valid;
79 
80     /* Notification enabled? */
81     bool notification;
82 
83     uint16_t queue_index;
84 
85     int inuse;
86 
87     uint16_t vector;
88     void (*handle_output)(VirtIODevice *vdev, VirtQueue *vq);
89     VirtIODevice *vdev;
90     EventNotifier guest_notifier;
91     EventNotifier host_notifier;
92     QLIST_ENTRY(VirtQueue) node;
93 };
94 
95 /* virt queue functions */
96 void virtio_queue_update_rings(VirtIODevice *vdev, int n)
97 {
98     VRing *vring = &vdev->vq[n].vring;
99 
100     if (!vring->desc) {
101         /* not yet setup -> nothing to do */
102         return;
103     }
104     vring->avail = vring->desc + vring->num * sizeof(VRingDesc);
105     vring->used = vring_align(vring->avail +
106                               offsetof(VRingAvail, ring[vring->num]),
107                               vring->align);
108 }
109 
110 static inline uint64_t vring_desc_addr(VirtIODevice *vdev, hwaddr desc_pa,
111                                        int i)
112 {
113     hwaddr pa;
114     pa = desc_pa + sizeof(VRingDesc) * i + offsetof(VRingDesc, addr);
115     return virtio_ldq_phys(vdev, pa);
116 }
117 
118 static inline uint32_t vring_desc_len(VirtIODevice *vdev, hwaddr desc_pa, int i)
119 {
120     hwaddr pa;
121     pa = desc_pa + sizeof(VRingDesc) * i + offsetof(VRingDesc, len);
122     return virtio_ldl_phys(vdev, pa);
123 }
124 
125 static inline uint16_t vring_desc_flags(VirtIODevice *vdev, hwaddr desc_pa,
126                                         int i)
127 {
128     hwaddr pa;
129     pa = desc_pa + sizeof(VRingDesc) * i + offsetof(VRingDesc, flags);
130     return virtio_lduw_phys(vdev, pa);
131 }
132 
133 static inline uint16_t vring_desc_next(VirtIODevice *vdev, hwaddr desc_pa,
134                                        int i)
135 {
136     hwaddr pa;
137     pa = desc_pa + sizeof(VRingDesc) * i + offsetof(VRingDesc, next);
138     return virtio_lduw_phys(vdev, pa);
139 }
140 
141 static inline uint16_t vring_avail_flags(VirtQueue *vq)
142 {
143     hwaddr pa;
144     pa = vq->vring.avail + offsetof(VRingAvail, flags);
145     return virtio_lduw_phys(vq->vdev, pa);
146 }
147 
148 static inline uint16_t vring_avail_idx(VirtQueue *vq)
149 {
150     hwaddr pa;
151     pa = vq->vring.avail + offsetof(VRingAvail, idx);
152     return virtio_lduw_phys(vq->vdev, pa);
153 }
154 
155 static inline uint16_t vring_avail_ring(VirtQueue *vq, int i)
156 {
157     hwaddr pa;
158     pa = vq->vring.avail + offsetof(VRingAvail, ring[i]);
159     return virtio_lduw_phys(vq->vdev, pa);
160 }
161 
162 static inline uint16_t vring_get_used_event(VirtQueue *vq)
163 {
164     return vring_avail_ring(vq, vq->vring.num);
165 }
166 
167 static inline void vring_used_ring_id(VirtQueue *vq, int i, uint32_t val)
168 {
169     hwaddr pa;
170     pa = vq->vring.used + offsetof(VRingUsed, ring[i].id);
171     virtio_stl_phys(vq->vdev, pa, val);
172 }
173 
174 static inline void vring_used_ring_len(VirtQueue *vq, int i, uint32_t val)
175 {
176     hwaddr pa;
177     pa = vq->vring.used + offsetof(VRingUsed, ring[i].len);
178     virtio_stl_phys(vq->vdev, pa, val);
179 }
180 
181 static uint16_t vring_used_idx(VirtQueue *vq)
182 {
183     hwaddr pa;
184     pa = vq->vring.used + offsetof(VRingUsed, idx);
185     return virtio_lduw_phys(vq->vdev, pa);
186 }
187 
188 static inline void vring_used_idx_set(VirtQueue *vq, uint16_t val)
189 {
190     hwaddr pa;
191     pa = vq->vring.used + offsetof(VRingUsed, idx);
192     virtio_stw_phys(vq->vdev, pa, val);
193 }
194 
195 static inline void vring_used_flags_set_bit(VirtQueue *vq, int mask)
196 {
197     VirtIODevice *vdev = vq->vdev;
198     hwaddr pa;
199     pa = vq->vring.used + offsetof(VRingUsed, flags);
200     virtio_stw_phys(vdev, pa, virtio_lduw_phys(vdev, pa) | mask);
201 }
202 
203 static inline void vring_used_flags_unset_bit(VirtQueue *vq, int mask)
204 {
205     VirtIODevice *vdev = vq->vdev;
206     hwaddr pa;
207     pa = vq->vring.used + offsetof(VRingUsed, flags);
208     virtio_stw_phys(vdev, pa, virtio_lduw_phys(vdev, pa) & ~mask);
209 }
210 
211 static inline void vring_set_avail_event(VirtQueue *vq, uint16_t val)
212 {
213     hwaddr pa;
214     if (!vq->notification) {
215         return;
216     }
217     pa = vq->vring.used + offsetof(VRingUsed, ring[vq->vring.num]);
218     virtio_stw_phys(vq->vdev, pa, val);
219 }
220 
221 void virtio_queue_set_notification(VirtQueue *vq, int enable)
222 {
223     vq->notification = enable;
224     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_RING_F_EVENT_IDX)) {
225         vring_set_avail_event(vq, vring_avail_idx(vq));
226     } else if (enable) {
227         vring_used_flags_unset_bit(vq, VRING_USED_F_NO_NOTIFY);
228     } else {
229         vring_used_flags_set_bit(vq, VRING_USED_F_NO_NOTIFY);
230     }
231     if (enable) {
232         /* Expose avail event/used flags before caller checks the avail idx. */
233         smp_mb();
234     }
235 }
236 
237 int virtio_queue_ready(VirtQueue *vq)
238 {
239     return vq->vring.avail != 0;
240 }
241 
242 int virtio_queue_empty(VirtQueue *vq)
243 {
244     return vring_avail_idx(vq) == vq->last_avail_idx;
245 }
246 
247 static void virtqueue_unmap_sg(VirtQueue *vq, const VirtQueueElement *elem,
248                                unsigned int len)
249 {
250     unsigned int offset;
251     int i;
252 
253     offset = 0;
254     for (i = 0; i < elem->in_num; i++) {
255         size_t size = MIN(len - offset, elem->in_sg[i].iov_len);
256 
257         cpu_physical_memory_unmap(elem->in_sg[i].iov_base,
258                                   elem->in_sg[i].iov_len,
259                                   1, size);
260 
261         offset += size;
262     }
263 
264     for (i = 0; i < elem->out_num; i++)
265         cpu_physical_memory_unmap(elem->out_sg[i].iov_base,
266                                   elem->out_sg[i].iov_len,
267                                   0, elem->out_sg[i].iov_len);
268 }
269 
270 void virtqueue_discard(VirtQueue *vq, const VirtQueueElement *elem,
271                        unsigned int len)
272 {
273     vq->last_avail_idx--;
274     virtqueue_unmap_sg(vq, elem, len);
275 }
276 
277 void virtqueue_fill(VirtQueue *vq, const VirtQueueElement *elem,
278                     unsigned int len, unsigned int idx)
279 {
280     trace_virtqueue_fill(vq, elem, len, idx);
281 
282     virtqueue_unmap_sg(vq, elem, len);
283 
284     idx = (idx + vring_used_idx(vq)) % vq->vring.num;
285 
286     /* Get a pointer to the next entry in the used ring. */
287     vring_used_ring_id(vq, idx, elem->index);
288     vring_used_ring_len(vq, idx, len);
289 }
290 
291 void virtqueue_flush(VirtQueue *vq, unsigned int count)
292 {
293     uint16_t old, new;
294     /* Make sure buffer is written before we update index. */
295     smp_wmb();
296     trace_virtqueue_flush(vq, count);
297     old = vring_used_idx(vq);
298     new = old + count;
299     vring_used_idx_set(vq, new);
300     vq->inuse -= count;
301     if (unlikely((int16_t)(new - vq->signalled_used) < (uint16_t)(new - old)))
302         vq->signalled_used_valid = false;
303 }
304 
305 void virtqueue_push(VirtQueue *vq, const VirtQueueElement *elem,
306                     unsigned int len)
307 {
308     virtqueue_fill(vq, elem, len, 0);
309     virtqueue_flush(vq, 1);
310 }
311 
312 static int virtqueue_num_heads(VirtQueue *vq, unsigned int idx)
313 {
314     uint16_t num_heads = vring_avail_idx(vq) - idx;
315 
316     /* Check it isn't doing very strange things with descriptor numbers. */
317     if (num_heads > vq->vring.num) {
318         error_report("Guest moved used index from %u to %u",
319                      idx, vring_avail_idx(vq));
320         exit(1);
321     }
322     /* On success, callers read a descriptor at vq->last_avail_idx.
323      * Make sure descriptor read does not bypass avail index read. */
324     if (num_heads) {
325         smp_rmb();
326     }
327 
328     return num_heads;
329 }
330 
331 static unsigned int virtqueue_get_head(VirtQueue *vq, unsigned int idx)
332 {
333     unsigned int head;
334 
335     /* Grab the next descriptor number they're advertising, and increment
336      * the index we've seen. */
337     head = vring_avail_ring(vq, idx % vq->vring.num);
338 
339     /* If their number is silly, that's a fatal mistake. */
340     if (head >= vq->vring.num) {
341         error_report("Guest says index %u is available", head);
342         exit(1);
343     }
344 
345     return head;
346 }
347 
348 static unsigned virtqueue_next_desc(VirtIODevice *vdev, hwaddr desc_pa,
349                                     unsigned int i, unsigned int max)
350 {
351     unsigned int next;
352 
353     /* If this descriptor says it doesn't chain, we're done. */
354     if (!(vring_desc_flags(vdev, desc_pa, i) & VRING_DESC_F_NEXT)) {
355         return max;
356     }
357 
358     /* Check they're not leading us off end of descriptors. */
359     next = vring_desc_next(vdev, desc_pa, i);
360     /* Make sure compiler knows to grab that: we don't want it changing! */
361     smp_wmb();
362 
363     if (next >= max) {
364         error_report("Desc next is %u", next);
365         exit(1);
366     }
367 
368     return next;
369 }
370 
371 void virtqueue_get_avail_bytes(VirtQueue *vq, unsigned int *in_bytes,
372                                unsigned int *out_bytes,
373                                unsigned max_in_bytes, unsigned max_out_bytes)
374 {
375     unsigned int idx;
376     unsigned int total_bufs, in_total, out_total;
377 
378     idx = vq->last_avail_idx;
379 
380     total_bufs = in_total = out_total = 0;
381     while (virtqueue_num_heads(vq, idx)) {
382         VirtIODevice *vdev = vq->vdev;
383         unsigned int max, num_bufs, indirect = 0;
384         hwaddr desc_pa;
385         int i;
386 
387         max = vq->vring.num;
388         num_bufs = total_bufs;
389         i = virtqueue_get_head(vq, idx++);
390         desc_pa = vq->vring.desc;
391 
392         if (vring_desc_flags(vdev, desc_pa, i) & VRING_DESC_F_INDIRECT) {
393             if (vring_desc_len(vdev, desc_pa, i) % sizeof(VRingDesc)) {
394                 error_report("Invalid size for indirect buffer table");
395                 exit(1);
396             }
397 
398             /* If we've got too many, that implies a descriptor loop. */
399             if (num_bufs >= max) {
400                 error_report("Looped descriptor");
401                 exit(1);
402             }
403 
404             /* loop over the indirect descriptor table */
405             indirect = 1;
406             max = vring_desc_len(vdev, desc_pa, i) / sizeof(VRingDesc);
407             desc_pa = vring_desc_addr(vdev, desc_pa, i);
408             num_bufs = i = 0;
409         }
410 
411         do {
412             /* If we've got too many, that implies a descriptor loop. */
413             if (++num_bufs > max) {
414                 error_report("Looped descriptor");
415                 exit(1);
416             }
417 
418             if (vring_desc_flags(vdev, desc_pa, i) & VRING_DESC_F_WRITE) {
419                 in_total += vring_desc_len(vdev, desc_pa, i);
420             } else {
421                 out_total += vring_desc_len(vdev, desc_pa, i);
422             }
423             if (in_total >= max_in_bytes && out_total >= max_out_bytes) {
424                 goto done;
425             }
426         } while ((i = virtqueue_next_desc(vdev, desc_pa, i, max)) != max);
427 
428         if (!indirect)
429             total_bufs = num_bufs;
430         else
431             total_bufs++;
432     }
433 done:
434     if (in_bytes) {
435         *in_bytes = in_total;
436     }
437     if (out_bytes) {
438         *out_bytes = out_total;
439     }
440 }
441 
442 int virtqueue_avail_bytes(VirtQueue *vq, unsigned int in_bytes,
443                           unsigned int out_bytes)
444 {
445     unsigned int in_total, out_total;
446 
447     virtqueue_get_avail_bytes(vq, &in_total, &out_total, in_bytes, out_bytes);
448     return in_bytes <= in_total && out_bytes <= out_total;
449 }
450 
451 static void virtqueue_map_iovec(struct iovec *sg, hwaddr *addr,
452                                 unsigned int *num_sg, unsigned int max_size,
453                                 int is_write)
454 {
455     unsigned int i;
456     hwaddr len;
457 
458     /* Note: this function MUST validate input, some callers
459      * are passing in num_sg values received over the network.
460      */
461     /* TODO: teach all callers that this can fail, and return failure instead
462      * of asserting here.
463      * When we do, we might be able to re-enable NDEBUG below.
464      */
465 #ifdef NDEBUG
466 #error building with NDEBUG is not supported
467 #endif
468     assert(*num_sg <= max_size);
469 
470     for (i = 0; i < *num_sg; i++) {
471         len = sg[i].iov_len;
472         sg[i].iov_base = cpu_physical_memory_map(addr[i], &len, is_write);
473         if (!sg[i].iov_base) {
474             error_report("virtio: error trying to map MMIO memory");
475             exit(1);
476         }
477         if (len == sg[i].iov_len) {
478             continue;
479         }
480         if (*num_sg >= max_size) {
481             error_report("virtio: memory split makes iovec too large");
482             exit(1);
483         }
484         memmove(sg + i + 1, sg + i, sizeof(*sg) * (*num_sg - i));
485         memmove(addr + i + 1, addr + i, sizeof(*addr) * (*num_sg - i));
486         assert(len < sg[i + 1].iov_len);
487         sg[i].iov_len = len;
488         addr[i + 1] += len;
489         sg[i + 1].iov_len -= len;
490         ++*num_sg;
491     }
492 }
493 
494 void virtqueue_map(VirtQueueElement *elem)
495 {
496     virtqueue_map_iovec(elem->in_sg, elem->in_addr, &elem->in_num,
497                         MIN(ARRAY_SIZE(elem->in_sg), ARRAY_SIZE(elem->in_addr)),
498                         1);
499     virtqueue_map_iovec(elem->out_sg, elem->out_addr, &elem->out_num,
500                         MIN(ARRAY_SIZE(elem->out_sg), ARRAY_SIZE(elem->out_addr)),
501                         0);
502 }
503 
504 int virtqueue_pop(VirtQueue *vq, VirtQueueElement *elem)
505 {
506     unsigned int i, head, max;
507     hwaddr desc_pa = vq->vring.desc;
508     VirtIODevice *vdev = vq->vdev;
509 
510     if (!virtqueue_num_heads(vq, vq->last_avail_idx))
511         return 0;
512 
513     /* When we start there are none of either input nor output. */
514     elem->out_num = elem->in_num = 0;
515 
516     max = vq->vring.num;
517 
518     i = head = virtqueue_get_head(vq, vq->last_avail_idx++);
519     if (virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX)) {
520         vring_set_avail_event(vq, vq->last_avail_idx);
521     }
522 
523     if (vring_desc_flags(vdev, desc_pa, i) & VRING_DESC_F_INDIRECT) {
524         if (vring_desc_len(vdev, desc_pa, i) % sizeof(VRingDesc)) {
525             error_report("Invalid size for indirect buffer table");
526             exit(1);
527         }
528 
529         /* loop over the indirect descriptor table */
530         max = vring_desc_len(vdev, desc_pa, i) / sizeof(VRingDesc);
531         desc_pa = vring_desc_addr(vdev, desc_pa, i);
532         i = 0;
533     }
534 
535     /* Collect all the descriptors */
536     do {
537         struct iovec *sg;
538 
539         if (vring_desc_flags(vdev, desc_pa, i) & VRING_DESC_F_WRITE) {
540             if (elem->in_num >= ARRAY_SIZE(elem->in_sg)) {
541                 error_report("Too many write descriptors in indirect table");
542                 exit(1);
543             }
544             elem->in_addr[elem->in_num] = vring_desc_addr(vdev, desc_pa, i);
545             sg = &elem->in_sg[elem->in_num++];
546         } else {
547             if (elem->out_num >= ARRAY_SIZE(elem->out_sg)) {
548                 error_report("Too many read descriptors in indirect table");
549                 exit(1);
550             }
551             elem->out_addr[elem->out_num] = vring_desc_addr(vdev, desc_pa, i);
552             sg = &elem->out_sg[elem->out_num++];
553         }
554 
555         sg->iov_len = vring_desc_len(vdev, desc_pa, i);
556 
557         /* If we've got too many, that implies a descriptor loop. */
558         if ((elem->in_num + elem->out_num) > max) {
559             error_report("Looped descriptor");
560             exit(1);
561         }
562     } while ((i = virtqueue_next_desc(vdev, desc_pa, i, max)) != max);
563 
564     /* Now map what we have collected */
565     virtqueue_map(elem);
566 
567     elem->index = head;
568 
569     vq->inuse++;
570 
571     trace_virtqueue_pop(vq, elem, elem->in_num, elem->out_num);
572     return elem->in_num + elem->out_num;
573 }
574 
575 /* virtio device */
576 static void virtio_notify_vector(VirtIODevice *vdev, uint16_t vector)
577 {
578     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
579     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
580 
581     if (k->notify) {
582         k->notify(qbus->parent, vector);
583     }
584 }
585 
586 void virtio_update_irq(VirtIODevice *vdev)
587 {
588     virtio_notify_vector(vdev, VIRTIO_NO_VECTOR);
589 }
590 
591 static int virtio_validate_features(VirtIODevice *vdev)
592 {
593     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
594 
595     if (k->validate_features) {
596         return k->validate_features(vdev);
597     } else {
598         return 0;
599     }
600 }
601 
602 int virtio_set_status(VirtIODevice *vdev, uint8_t val)
603 {
604     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
605     trace_virtio_set_status(vdev, val);
606 
607     if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
608         if (!(vdev->status & VIRTIO_CONFIG_S_FEATURES_OK) &&
609             val & VIRTIO_CONFIG_S_FEATURES_OK) {
610             int ret = virtio_validate_features(vdev);
611 
612             if (ret) {
613                 return ret;
614             }
615         }
616     }
617     if (k->set_status) {
618         k->set_status(vdev, val);
619     }
620     vdev->status = val;
621     return 0;
622 }
623 
624 bool target_words_bigendian(void);
625 static enum virtio_device_endian virtio_default_endian(void)
626 {
627     if (target_words_bigendian()) {
628         return VIRTIO_DEVICE_ENDIAN_BIG;
629     } else {
630         return VIRTIO_DEVICE_ENDIAN_LITTLE;
631     }
632 }
633 
634 static enum virtio_device_endian virtio_current_cpu_endian(void)
635 {
636     CPUClass *cc = CPU_GET_CLASS(current_cpu);
637 
638     if (cc->virtio_is_big_endian(current_cpu)) {
639         return VIRTIO_DEVICE_ENDIAN_BIG;
640     } else {
641         return VIRTIO_DEVICE_ENDIAN_LITTLE;
642     }
643 }
644 
645 void virtio_reset(void *opaque)
646 {
647     VirtIODevice *vdev = opaque;
648     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
649     int i;
650 
651     virtio_set_status(vdev, 0);
652     if (current_cpu) {
653         /* Guest initiated reset */
654         vdev->device_endian = virtio_current_cpu_endian();
655     } else {
656         /* System reset */
657         vdev->device_endian = virtio_default_endian();
658     }
659 
660     if (k->reset) {
661         k->reset(vdev);
662     }
663 
664     vdev->guest_features = 0;
665     vdev->queue_sel = 0;
666     vdev->status = 0;
667     vdev->isr = 0;
668     vdev->config_vector = VIRTIO_NO_VECTOR;
669     virtio_notify_vector(vdev, vdev->config_vector);
670 
671     for(i = 0; i < VIRTIO_QUEUE_MAX; i++) {
672         vdev->vq[i].vring.desc = 0;
673         vdev->vq[i].vring.avail = 0;
674         vdev->vq[i].vring.used = 0;
675         vdev->vq[i].last_avail_idx = 0;
676         virtio_queue_set_vector(vdev, i, VIRTIO_NO_VECTOR);
677         vdev->vq[i].signalled_used = 0;
678         vdev->vq[i].signalled_used_valid = false;
679         vdev->vq[i].notification = true;
680         vdev->vq[i].vring.num = vdev->vq[i].vring.num_default;
681     }
682 }
683 
684 uint32_t virtio_config_readb(VirtIODevice *vdev, uint32_t addr)
685 {
686     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
687     uint8_t val;
688 
689     if (addr + sizeof(val) > vdev->config_len) {
690         return (uint32_t)-1;
691     }
692 
693     k->get_config(vdev, vdev->config);
694 
695     val = ldub_p(vdev->config + addr);
696     return val;
697 }
698 
699 uint32_t virtio_config_readw(VirtIODevice *vdev, uint32_t addr)
700 {
701     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
702     uint16_t val;
703 
704     if (addr + sizeof(val) > vdev->config_len) {
705         return (uint32_t)-1;
706     }
707 
708     k->get_config(vdev, vdev->config);
709 
710     val = lduw_p(vdev->config + addr);
711     return val;
712 }
713 
714 uint32_t virtio_config_readl(VirtIODevice *vdev, uint32_t addr)
715 {
716     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
717     uint32_t val;
718 
719     if (addr + sizeof(val) > vdev->config_len) {
720         return (uint32_t)-1;
721     }
722 
723     k->get_config(vdev, vdev->config);
724 
725     val = ldl_p(vdev->config + addr);
726     return val;
727 }
728 
729 void virtio_config_writeb(VirtIODevice *vdev, uint32_t addr, uint32_t data)
730 {
731     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
732     uint8_t val = data;
733 
734     if (addr + sizeof(val) > vdev->config_len) {
735         return;
736     }
737 
738     stb_p(vdev->config + addr, val);
739 
740     if (k->set_config) {
741         k->set_config(vdev, vdev->config);
742     }
743 }
744 
745 void virtio_config_writew(VirtIODevice *vdev, uint32_t addr, uint32_t data)
746 {
747     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
748     uint16_t val = data;
749 
750     if (addr + sizeof(val) > vdev->config_len) {
751         return;
752     }
753 
754     stw_p(vdev->config + addr, val);
755 
756     if (k->set_config) {
757         k->set_config(vdev, vdev->config);
758     }
759 }
760 
761 void virtio_config_writel(VirtIODevice *vdev, uint32_t addr, uint32_t data)
762 {
763     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
764     uint32_t val = data;
765 
766     if (addr + sizeof(val) > vdev->config_len) {
767         return;
768     }
769 
770     stl_p(vdev->config + addr, val);
771 
772     if (k->set_config) {
773         k->set_config(vdev, vdev->config);
774     }
775 }
776 
777 uint32_t virtio_config_modern_readb(VirtIODevice *vdev, uint32_t addr)
778 {
779     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
780     uint8_t val;
781 
782     if (addr + sizeof(val) > vdev->config_len) {
783         return (uint32_t)-1;
784     }
785 
786     k->get_config(vdev, vdev->config);
787 
788     val = ldub_p(vdev->config + addr);
789     return val;
790 }
791 
792 uint32_t virtio_config_modern_readw(VirtIODevice *vdev, uint32_t addr)
793 {
794     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
795     uint16_t val;
796 
797     if (addr + sizeof(val) > vdev->config_len) {
798         return (uint32_t)-1;
799     }
800 
801     k->get_config(vdev, vdev->config);
802 
803     val = lduw_le_p(vdev->config + addr);
804     return val;
805 }
806 
807 uint32_t virtio_config_modern_readl(VirtIODevice *vdev, uint32_t addr)
808 {
809     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
810     uint32_t val;
811 
812     if (addr + sizeof(val) > vdev->config_len) {
813         return (uint32_t)-1;
814     }
815 
816     k->get_config(vdev, vdev->config);
817 
818     val = ldl_le_p(vdev->config + addr);
819     return val;
820 }
821 
822 void virtio_config_modern_writeb(VirtIODevice *vdev,
823                                  uint32_t addr, uint32_t data)
824 {
825     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
826     uint8_t val = data;
827 
828     if (addr + sizeof(val) > vdev->config_len) {
829         return;
830     }
831 
832     stb_p(vdev->config + addr, val);
833 
834     if (k->set_config) {
835         k->set_config(vdev, vdev->config);
836     }
837 }
838 
839 void virtio_config_modern_writew(VirtIODevice *vdev,
840                                  uint32_t addr, uint32_t data)
841 {
842     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
843     uint16_t val = data;
844 
845     if (addr + sizeof(val) > vdev->config_len) {
846         return;
847     }
848 
849     stw_le_p(vdev->config + addr, val);
850 
851     if (k->set_config) {
852         k->set_config(vdev, vdev->config);
853     }
854 }
855 
856 void virtio_config_modern_writel(VirtIODevice *vdev,
857                                  uint32_t addr, uint32_t data)
858 {
859     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
860     uint32_t val = data;
861 
862     if (addr + sizeof(val) > vdev->config_len) {
863         return;
864     }
865 
866     stl_le_p(vdev->config + addr, val);
867 
868     if (k->set_config) {
869         k->set_config(vdev, vdev->config);
870     }
871 }
872 
873 void virtio_queue_set_addr(VirtIODevice *vdev, int n, hwaddr addr)
874 {
875     vdev->vq[n].vring.desc = addr;
876     virtio_queue_update_rings(vdev, n);
877 }
878 
879 hwaddr virtio_queue_get_addr(VirtIODevice *vdev, int n)
880 {
881     return vdev->vq[n].vring.desc;
882 }
883 
884 void virtio_queue_set_rings(VirtIODevice *vdev, int n, hwaddr desc,
885                             hwaddr avail, hwaddr used)
886 {
887     vdev->vq[n].vring.desc = desc;
888     vdev->vq[n].vring.avail = avail;
889     vdev->vq[n].vring.used = used;
890 }
891 
892 void virtio_queue_set_num(VirtIODevice *vdev, int n, int num)
893 {
894     /* Don't allow guest to flip queue between existent and
895      * nonexistent states, or to set it to an invalid size.
896      */
897     if (!!num != !!vdev->vq[n].vring.num ||
898         num > VIRTQUEUE_MAX_SIZE ||
899         num < 0) {
900         return;
901     }
902     vdev->vq[n].vring.num = num;
903 }
904 
905 VirtQueue *virtio_vector_first_queue(VirtIODevice *vdev, uint16_t vector)
906 {
907     return QLIST_FIRST(&vdev->vector_queues[vector]);
908 }
909 
910 VirtQueue *virtio_vector_next_queue(VirtQueue *vq)
911 {
912     return QLIST_NEXT(vq, node);
913 }
914 
915 int virtio_queue_get_num(VirtIODevice *vdev, int n)
916 {
917     return vdev->vq[n].vring.num;
918 }
919 
920 int virtio_get_num_queues(VirtIODevice *vdev)
921 {
922     int i;
923 
924     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
925         if (!virtio_queue_get_num(vdev, i)) {
926             break;
927         }
928     }
929 
930     return i;
931 }
932 
933 int virtio_queue_get_id(VirtQueue *vq)
934 {
935     VirtIODevice *vdev = vq->vdev;
936     assert(vq >= &vdev->vq[0] && vq < &vdev->vq[VIRTIO_QUEUE_MAX]);
937     return vq - &vdev->vq[0];
938 }
939 
940 void virtio_queue_set_align(VirtIODevice *vdev, int n, int align)
941 {
942     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
943     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
944 
945     /* virtio-1 compliant devices cannot change the alignment */
946     if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
947         error_report("tried to modify queue alignment for virtio-1 device");
948         return;
949     }
950     /* Check that the transport told us it was going to do this
951      * (so a buggy transport will immediately assert rather than
952      * silently failing to migrate this state)
953      */
954     assert(k->has_variable_vring_alignment);
955 
956     vdev->vq[n].vring.align = align;
957     virtio_queue_update_rings(vdev, n);
958 }
959 
960 void virtio_queue_notify_vq(VirtQueue *vq)
961 {
962     if (vq->vring.desc && vq->handle_output) {
963         VirtIODevice *vdev = vq->vdev;
964 
965         trace_virtio_queue_notify(vdev, vq - vdev->vq, vq);
966         vq->handle_output(vdev, vq);
967     }
968 }
969 
970 void virtio_queue_notify(VirtIODevice *vdev, int n)
971 {
972     virtio_queue_notify_vq(&vdev->vq[n]);
973 }
974 
975 uint16_t virtio_queue_vector(VirtIODevice *vdev, int n)
976 {
977     return n < VIRTIO_QUEUE_MAX ? vdev->vq[n].vector :
978         VIRTIO_NO_VECTOR;
979 }
980 
981 void virtio_queue_set_vector(VirtIODevice *vdev, int n, uint16_t vector)
982 {
983     VirtQueue *vq = &vdev->vq[n];
984 
985     if (n < VIRTIO_QUEUE_MAX) {
986         if (vdev->vector_queues &&
987             vdev->vq[n].vector != VIRTIO_NO_VECTOR) {
988             QLIST_REMOVE(vq, node);
989         }
990         vdev->vq[n].vector = vector;
991         if (vdev->vector_queues &&
992             vector != VIRTIO_NO_VECTOR) {
993             QLIST_INSERT_HEAD(&vdev->vector_queues[vector], vq, node);
994         }
995     }
996 }
997 
998 VirtQueue *virtio_add_queue(VirtIODevice *vdev, int queue_size,
999                             void (*handle_output)(VirtIODevice *, VirtQueue *))
1000 {
1001     int i;
1002 
1003     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
1004         if (vdev->vq[i].vring.num == 0)
1005             break;
1006     }
1007 
1008     if (i == VIRTIO_QUEUE_MAX || queue_size > VIRTQUEUE_MAX_SIZE)
1009         abort();
1010 
1011     vdev->vq[i].vring.num = queue_size;
1012     vdev->vq[i].vring.num_default = queue_size;
1013     vdev->vq[i].vring.align = VIRTIO_PCI_VRING_ALIGN;
1014     vdev->vq[i].handle_output = handle_output;
1015 
1016     return &vdev->vq[i];
1017 }
1018 
1019 void virtio_del_queue(VirtIODevice *vdev, int n)
1020 {
1021     if (n < 0 || n >= VIRTIO_QUEUE_MAX) {
1022         abort();
1023     }
1024 
1025     vdev->vq[n].vring.num = 0;
1026     vdev->vq[n].vring.num_default = 0;
1027 }
1028 
1029 void virtio_irq(VirtQueue *vq)
1030 {
1031     trace_virtio_irq(vq);
1032     vq->vdev->isr |= 0x01;
1033     virtio_notify_vector(vq->vdev, vq->vector);
1034 }
1035 
1036 static bool vring_notify(VirtIODevice *vdev, VirtQueue *vq)
1037 {
1038     uint16_t old, new;
1039     bool v;
1040     /* We need to expose used array entries before checking used event. */
1041     smp_mb();
1042     /* Always notify when queue is empty (when feature acknowledge) */
1043     if (virtio_vdev_has_feature(vdev, VIRTIO_F_NOTIFY_ON_EMPTY) &&
1044         !vq->inuse && vring_avail_idx(vq) == vq->last_avail_idx) {
1045         return true;
1046     }
1047 
1048     if (!virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX)) {
1049         return !(vring_avail_flags(vq) & VRING_AVAIL_F_NO_INTERRUPT);
1050     }
1051 
1052     v = vq->signalled_used_valid;
1053     vq->signalled_used_valid = true;
1054     old = vq->signalled_used;
1055     new = vq->signalled_used = vring_used_idx(vq);
1056     return !v || vring_need_event(vring_get_used_event(vq), new, old);
1057 }
1058 
1059 void virtio_notify(VirtIODevice *vdev, VirtQueue *vq)
1060 {
1061     if (!vring_notify(vdev, vq)) {
1062         return;
1063     }
1064 
1065     trace_virtio_notify(vdev, vq);
1066     vdev->isr |= 0x01;
1067     virtio_notify_vector(vdev, vq->vector);
1068 }
1069 
1070 void virtio_notify_config(VirtIODevice *vdev)
1071 {
1072     if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK))
1073         return;
1074 
1075     vdev->isr |= 0x03;
1076     vdev->generation++;
1077     virtio_notify_vector(vdev, vdev->config_vector);
1078 }
1079 
1080 static bool virtio_device_endian_needed(void *opaque)
1081 {
1082     VirtIODevice *vdev = opaque;
1083 
1084     assert(vdev->device_endian != VIRTIO_DEVICE_ENDIAN_UNKNOWN);
1085     if (!virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
1086         return vdev->device_endian != virtio_default_endian();
1087     }
1088     /* Devices conforming to VIRTIO 1.0 or later are always LE. */
1089     return vdev->device_endian != VIRTIO_DEVICE_ENDIAN_LITTLE;
1090 }
1091 
1092 static bool virtio_64bit_features_needed(void *opaque)
1093 {
1094     VirtIODevice *vdev = opaque;
1095 
1096     return (vdev->host_features >> 32) != 0;
1097 }
1098 
1099 static bool virtio_virtqueue_needed(void *opaque)
1100 {
1101     VirtIODevice *vdev = opaque;
1102 
1103     return virtio_host_has_feature(vdev, VIRTIO_F_VERSION_1);
1104 }
1105 
1106 static bool virtio_ringsize_needed(void *opaque)
1107 {
1108     VirtIODevice *vdev = opaque;
1109     int i;
1110 
1111     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
1112         if (vdev->vq[i].vring.num != vdev->vq[i].vring.num_default) {
1113             return true;
1114         }
1115     }
1116     return false;
1117 }
1118 
1119 static void put_virtqueue_state(QEMUFile *f, void *pv, size_t size)
1120 {
1121     VirtIODevice *vdev = pv;
1122     int i;
1123 
1124     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
1125         qemu_put_be64(f, vdev->vq[i].vring.avail);
1126         qemu_put_be64(f, vdev->vq[i].vring.used);
1127     }
1128 }
1129 
1130 static int get_virtqueue_state(QEMUFile *f, void *pv, size_t size)
1131 {
1132     VirtIODevice *vdev = pv;
1133     int i;
1134 
1135     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
1136         vdev->vq[i].vring.avail = qemu_get_be64(f);
1137         vdev->vq[i].vring.used = qemu_get_be64(f);
1138     }
1139     return 0;
1140 }
1141 
1142 static VMStateInfo vmstate_info_virtqueue = {
1143     .name = "virtqueue_state",
1144     .get = get_virtqueue_state,
1145     .put = put_virtqueue_state,
1146 };
1147 
1148 static const VMStateDescription vmstate_virtio_virtqueues = {
1149     .name = "virtio/virtqueues",
1150     .version_id = 1,
1151     .minimum_version_id = 1,
1152     .needed = &virtio_virtqueue_needed,
1153     .fields = (VMStateField[]) {
1154         {
1155             .name         = "virtqueues",
1156             .version_id   = 0,
1157             .field_exists = NULL,
1158             .size         = 0,
1159             .info         = &vmstate_info_virtqueue,
1160             .flags        = VMS_SINGLE,
1161             .offset       = 0,
1162         },
1163         VMSTATE_END_OF_LIST()
1164     }
1165 };
1166 
1167 static void put_ringsize_state(QEMUFile *f, void *pv, size_t size)
1168 {
1169     VirtIODevice *vdev = pv;
1170     int i;
1171 
1172     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
1173         qemu_put_be32(f, vdev->vq[i].vring.num_default);
1174     }
1175 }
1176 
1177 static int get_ringsize_state(QEMUFile *f, void *pv, size_t size)
1178 {
1179     VirtIODevice *vdev = pv;
1180     int i;
1181 
1182     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
1183         vdev->vq[i].vring.num_default = qemu_get_be32(f);
1184     }
1185     return 0;
1186 }
1187 
1188 static VMStateInfo vmstate_info_ringsize = {
1189     .name = "ringsize_state",
1190     .get = get_ringsize_state,
1191     .put = put_ringsize_state,
1192 };
1193 
1194 static const VMStateDescription vmstate_virtio_ringsize = {
1195     .name = "virtio/ringsize",
1196     .version_id = 1,
1197     .minimum_version_id = 1,
1198     .needed = &virtio_ringsize_needed,
1199     .fields = (VMStateField[]) {
1200         {
1201             .name         = "ringsize",
1202             .version_id   = 0,
1203             .field_exists = NULL,
1204             .size         = 0,
1205             .info         = &vmstate_info_ringsize,
1206             .flags        = VMS_SINGLE,
1207             .offset       = 0,
1208         },
1209         VMSTATE_END_OF_LIST()
1210     }
1211 };
1212 
1213 static const VMStateDescription vmstate_virtio_device_endian = {
1214     .name = "virtio/device_endian",
1215     .version_id = 1,
1216     .minimum_version_id = 1,
1217     .needed = &virtio_device_endian_needed,
1218     .fields = (VMStateField[]) {
1219         VMSTATE_UINT8(device_endian, VirtIODevice),
1220         VMSTATE_END_OF_LIST()
1221     }
1222 };
1223 
1224 static const VMStateDescription vmstate_virtio_64bit_features = {
1225     .name = "virtio/64bit_features",
1226     .version_id = 1,
1227     .minimum_version_id = 1,
1228     .needed = &virtio_64bit_features_needed,
1229     .fields = (VMStateField[]) {
1230         VMSTATE_UINT64(guest_features, VirtIODevice),
1231         VMSTATE_END_OF_LIST()
1232     }
1233 };
1234 
1235 static const VMStateDescription vmstate_virtio = {
1236     .name = "virtio",
1237     .version_id = 1,
1238     .minimum_version_id = 1,
1239     .minimum_version_id_old = 1,
1240     .fields = (VMStateField[]) {
1241         VMSTATE_END_OF_LIST()
1242     },
1243     .subsections = (const VMStateDescription*[]) {
1244         &vmstate_virtio_device_endian,
1245         &vmstate_virtio_64bit_features,
1246         &vmstate_virtio_virtqueues,
1247         &vmstate_virtio_ringsize,
1248         NULL
1249     }
1250 };
1251 
1252 void virtio_save(VirtIODevice *vdev, QEMUFile *f)
1253 {
1254     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
1255     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
1256     VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev);
1257     uint32_t guest_features_lo = (vdev->guest_features & 0xffffffff);
1258     int i;
1259 
1260     if (k->save_config) {
1261         k->save_config(qbus->parent, f);
1262     }
1263 
1264     qemu_put_8s(f, &vdev->status);
1265     qemu_put_8s(f, &vdev->isr);
1266     qemu_put_be16s(f, &vdev->queue_sel);
1267     qemu_put_be32s(f, &guest_features_lo);
1268     qemu_put_be32(f, vdev->config_len);
1269     qemu_put_buffer(f, vdev->config, vdev->config_len);
1270 
1271     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
1272         if (vdev->vq[i].vring.num == 0)
1273             break;
1274     }
1275 
1276     qemu_put_be32(f, i);
1277 
1278     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
1279         if (vdev->vq[i].vring.num == 0)
1280             break;
1281 
1282         qemu_put_be32(f, vdev->vq[i].vring.num);
1283         if (k->has_variable_vring_alignment) {
1284             qemu_put_be32(f, vdev->vq[i].vring.align);
1285         }
1286         /* XXX virtio-1 devices */
1287         qemu_put_be64(f, vdev->vq[i].vring.desc);
1288         qemu_put_be16s(f, &vdev->vq[i].last_avail_idx);
1289         if (k->save_queue) {
1290             k->save_queue(qbus->parent, i, f);
1291         }
1292     }
1293 
1294     if (vdc->save != NULL) {
1295         vdc->save(vdev, f);
1296     }
1297 
1298     /* Subsections */
1299     vmstate_save_state(f, &vmstate_virtio, vdev, NULL);
1300 }
1301 
1302 static int virtio_set_features_nocheck(VirtIODevice *vdev, uint64_t val)
1303 {
1304     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1305     bool bad = (val & ~(vdev->host_features)) != 0;
1306 
1307     val &= vdev->host_features;
1308     if (k->set_features) {
1309         k->set_features(vdev, val);
1310     }
1311     vdev->guest_features = val;
1312     return bad ? -1 : 0;
1313 }
1314 
1315 int virtio_set_features(VirtIODevice *vdev, uint64_t val)
1316 {
1317    /*
1318      * The driver must not attempt to set features after feature negotiation
1319      * has finished.
1320      */
1321     if (vdev->status & VIRTIO_CONFIG_S_FEATURES_OK) {
1322         return -EINVAL;
1323     }
1324     return virtio_set_features_nocheck(vdev, val);
1325 }
1326 
1327 int virtio_load(VirtIODevice *vdev, QEMUFile *f, int version_id)
1328 {
1329     int i, ret;
1330     int32_t config_len;
1331     uint32_t num;
1332     uint32_t features;
1333     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
1334     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
1335     VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev);
1336 
1337     /*
1338      * We poison the endianness to ensure it does not get used before
1339      * subsections have been loaded.
1340      */
1341     vdev->device_endian = VIRTIO_DEVICE_ENDIAN_UNKNOWN;
1342 
1343     if (k->load_config) {
1344         ret = k->load_config(qbus->parent, f);
1345         if (ret)
1346             return ret;
1347     }
1348 
1349     qemu_get_8s(f, &vdev->status);
1350     qemu_get_8s(f, &vdev->isr);
1351     qemu_get_be16s(f, &vdev->queue_sel);
1352     if (vdev->queue_sel >= VIRTIO_QUEUE_MAX) {
1353         return -1;
1354     }
1355     qemu_get_be32s(f, &features);
1356 
1357     config_len = qemu_get_be32(f);
1358 
1359     /*
1360      * There are cases where the incoming config can be bigger or smaller
1361      * than what we have; so load what we have space for, and skip
1362      * any excess that's in the stream.
1363      */
1364     qemu_get_buffer(f, vdev->config, MIN(config_len, vdev->config_len));
1365 
1366     while (config_len > vdev->config_len) {
1367         qemu_get_byte(f);
1368         config_len--;
1369     }
1370 
1371     num = qemu_get_be32(f);
1372 
1373     if (num > VIRTIO_QUEUE_MAX) {
1374         error_report("Invalid number of PCI queues: 0x%x", num);
1375         return -1;
1376     }
1377 
1378     for (i = 0; i < num; i++) {
1379         vdev->vq[i].vring.num = qemu_get_be32(f);
1380         if (k->has_variable_vring_alignment) {
1381             vdev->vq[i].vring.align = qemu_get_be32(f);
1382         }
1383         vdev->vq[i].vring.desc = qemu_get_be64(f);
1384         qemu_get_be16s(f, &vdev->vq[i].last_avail_idx);
1385         vdev->vq[i].signalled_used_valid = false;
1386         vdev->vq[i].notification = true;
1387 
1388         if (vdev->vq[i].vring.desc) {
1389             /* XXX virtio-1 devices */
1390             virtio_queue_update_rings(vdev, i);
1391         } else if (vdev->vq[i].last_avail_idx) {
1392             error_report("VQ %d address 0x0 "
1393                          "inconsistent with Host index 0x%x",
1394                          i, vdev->vq[i].last_avail_idx);
1395                 return -1;
1396 	}
1397         if (k->load_queue) {
1398             ret = k->load_queue(qbus->parent, i, f);
1399             if (ret)
1400                 return ret;
1401         }
1402     }
1403 
1404     virtio_notify_vector(vdev, VIRTIO_NO_VECTOR);
1405 
1406     if (vdc->load != NULL) {
1407         ret = vdc->load(vdev, f, version_id);
1408         if (ret) {
1409             return ret;
1410         }
1411     }
1412 
1413     /* Subsections */
1414     ret = vmstate_load_state(f, &vmstate_virtio, vdev, 1);
1415     if (ret) {
1416         return ret;
1417     }
1418 
1419     if (vdev->device_endian == VIRTIO_DEVICE_ENDIAN_UNKNOWN) {
1420         vdev->device_endian = virtio_default_endian();
1421     }
1422 
1423     if (virtio_64bit_features_needed(vdev)) {
1424         /*
1425          * Subsection load filled vdev->guest_features.  Run them
1426          * through virtio_set_features to sanity-check them against
1427          * host_features.
1428          */
1429         uint64_t features64 = vdev->guest_features;
1430         if (virtio_set_features_nocheck(vdev, features64) < 0) {
1431             error_report("Features 0x%" PRIx64 " unsupported. "
1432                          "Allowed features: 0x%" PRIx64,
1433                          features64, vdev->host_features);
1434             return -1;
1435         }
1436     } else {
1437         if (virtio_set_features_nocheck(vdev, features) < 0) {
1438             error_report("Features 0x%x unsupported. "
1439                          "Allowed features: 0x%" PRIx64,
1440                          features, vdev->host_features);
1441             return -1;
1442         }
1443     }
1444 
1445     for (i = 0; i < num; i++) {
1446         if (vdev->vq[i].vring.desc) {
1447             uint16_t nheads;
1448             nheads = vring_avail_idx(&vdev->vq[i]) - vdev->vq[i].last_avail_idx;
1449             /* Check it isn't doing strange things with descriptor numbers. */
1450             if (nheads > vdev->vq[i].vring.num) {
1451                 error_report("VQ %d size 0x%x Guest index 0x%x "
1452                              "inconsistent with Host index 0x%x: delta 0x%x",
1453                              i, vdev->vq[i].vring.num,
1454                              vring_avail_idx(&vdev->vq[i]),
1455                              vdev->vq[i].last_avail_idx, nheads);
1456                 return -1;
1457             }
1458         }
1459     }
1460 
1461     return 0;
1462 }
1463 
1464 void virtio_cleanup(VirtIODevice *vdev)
1465 {
1466     qemu_del_vm_change_state_handler(vdev->vmstate);
1467     g_free(vdev->config);
1468     g_free(vdev->vq);
1469     g_free(vdev->vector_queues);
1470 }
1471 
1472 static void virtio_vmstate_change(void *opaque, int running, RunState state)
1473 {
1474     VirtIODevice *vdev = opaque;
1475     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
1476     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
1477     bool backend_run = running && (vdev->status & VIRTIO_CONFIG_S_DRIVER_OK);
1478     vdev->vm_running = running;
1479 
1480     if (backend_run) {
1481         virtio_set_status(vdev, vdev->status);
1482     }
1483 
1484     if (k->vmstate_change) {
1485         k->vmstate_change(qbus->parent, backend_run);
1486     }
1487 
1488     if (!backend_run) {
1489         virtio_set_status(vdev, vdev->status);
1490     }
1491 }
1492 
1493 void virtio_instance_init_common(Object *proxy_obj, void *data,
1494                                  size_t vdev_size, const char *vdev_name)
1495 {
1496     DeviceState *vdev = data;
1497 
1498     object_initialize(vdev, vdev_size, vdev_name);
1499     object_property_add_child(proxy_obj, "virtio-backend", OBJECT(vdev), NULL);
1500     object_unref(OBJECT(vdev));
1501     qdev_alias_all_properties(vdev, proxy_obj);
1502 }
1503 
1504 void virtio_init(VirtIODevice *vdev, const char *name,
1505                  uint16_t device_id, size_t config_size)
1506 {
1507     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
1508     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
1509     int i;
1510     int nvectors = k->query_nvectors ? k->query_nvectors(qbus->parent) : 0;
1511 
1512     if (nvectors) {
1513         vdev->vector_queues =
1514             g_malloc0(sizeof(*vdev->vector_queues) * nvectors);
1515     }
1516 
1517     vdev->device_id = device_id;
1518     vdev->status = 0;
1519     vdev->isr = 0;
1520     vdev->queue_sel = 0;
1521     vdev->config_vector = VIRTIO_NO_VECTOR;
1522     vdev->vq = g_malloc0(sizeof(VirtQueue) * VIRTIO_QUEUE_MAX);
1523     vdev->vm_running = runstate_is_running();
1524     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
1525         vdev->vq[i].vector = VIRTIO_NO_VECTOR;
1526         vdev->vq[i].vdev = vdev;
1527         vdev->vq[i].queue_index = i;
1528     }
1529 
1530     vdev->name = name;
1531     vdev->config_len = config_size;
1532     if (vdev->config_len) {
1533         vdev->config = g_malloc0(config_size);
1534     } else {
1535         vdev->config = NULL;
1536     }
1537     vdev->vmstate = qemu_add_vm_change_state_handler(virtio_vmstate_change,
1538                                                      vdev);
1539     vdev->device_endian = virtio_default_endian();
1540 }
1541 
1542 hwaddr virtio_queue_get_desc_addr(VirtIODevice *vdev, int n)
1543 {
1544     return vdev->vq[n].vring.desc;
1545 }
1546 
1547 hwaddr virtio_queue_get_avail_addr(VirtIODevice *vdev, int n)
1548 {
1549     return vdev->vq[n].vring.avail;
1550 }
1551 
1552 hwaddr virtio_queue_get_used_addr(VirtIODevice *vdev, int n)
1553 {
1554     return vdev->vq[n].vring.used;
1555 }
1556 
1557 hwaddr virtio_queue_get_ring_addr(VirtIODevice *vdev, int n)
1558 {
1559     return vdev->vq[n].vring.desc;
1560 }
1561 
1562 hwaddr virtio_queue_get_desc_size(VirtIODevice *vdev, int n)
1563 {
1564     return sizeof(VRingDesc) * vdev->vq[n].vring.num;
1565 }
1566 
1567 hwaddr virtio_queue_get_avail_size(VirtIODevice *vdev, int n)
1568 {
1569     return offsetof(VRingAvail, ring) +
1570         sizeof(uint16_t) * vdev->vq[n].vring.num;
1571 }
1572 
1573 hwaddr virtio_queue_get_used_size(VirtIODevice *vdev, int n)
1574 {
1575     return offsetof(VRingUsed, ring) +
1576         sizeof(VRingUsedElem) * vdev->vq[n].vring.num;
1577 }
1578 
1579 hwaddr virtio_queue_get_ring_size(VirtIODevice *vdev, int n)
1580 {
1581     return vdev->vq[n].vring.used - vdev->vq[n].vring.desc +
1582 	    virtio_queue_get_used_size(vdev, n);
1583 }
1584 
1585 uint16_t virtio_queue_get_last_avail_idx(VirtIODevice *vdev, int n)
1586 {
1587     return vdev->vq[n].last_avail_idx;
1588 }
1589 
1590 void virtio_queue_set_last_avail_idx(VirtIODevice *vdev, int n, uint16_t idx)
1591 {
1592     vdev->vq[n].last_avail_idx = idx;
1593 }
1594 
1595 void virtio_queue_invalidate_signalled_used(VirtIODevice *vdev, int n)
1596 {
1597     vdev->vq[n].signalled_used_valid = false;
1598 }
1599 
1600 VirtQueue *virtio_get_queue(VirtIODevice *vdev, int n)
1601 {
1602     return vdev->vq + n;
1603 }
1604 
1605 uint16_t virtio_get_queue_index(VirtQueue *vq)
1606 {
1607     return vq->queue_index;
1608 }
1609 
1610 static void virtio_queue_guest_notifier_read(EventNotifier *n)
1611 {
1612     VirtQueue *vq = container_of(n, VirtQueue, guest_notifier);
1613     if (event_notifier_test_and_clear(n)) {
1614         virtio_irq(vq);
1615     }
1616 }
1617 
1618 void virtio_queue_set_guest_notifier_fd_handler(VirtQueue *vq, bool assign,
1619                                                 bool with_irqfd)
1620 {
1621     if (assign && !with_irqfd) {
1622         event_notifier_set_handler(&vq->guest_notifier,
1623                                    virtio_queue_guest_notifier_read);
1624     } else {
1625         event_notifier_set_handler(&vq->guest_notifier, NULL);
1626     }
1627     if (!assign) {
1628         /* Test and clear notifier before closing it,
1629          * in case poll callback didn't have time to run. */
1630         virtio_queue_guest_notifier_read(&vq->guest_notifier);
1631     }
1632 }
1633 
1634 EventNotifier *virtio_queue_get_guest_notifier(VirtQueue *vq)
1635 {
1636     return &vq->guest_notifier;
1637 }
1638 
1639 static void virtio_queue_host_notifier_read(EventNotifier *n)
1640 {
1641     VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
1642     if (event_notifier_test_and_clear(n)) {
1643         virtio_queue_notify_vq(vq);
1644     }
1645 }
1646 
1647 void virtio_queue_set_host_notifier_fd_handler(VirtQueue *vq, bool assign,
1648                                                bool set_handler)
1649 {
1650     if (assign && set_handler) {
1651         event_notifier_set_handler(&vq->host_notifier,
1652                                    virtio_queue_host_notifier_read);
1653     } else {
1654         event_notifier_set_handler(&vq->host_notifier, NULL);
1655     }
1656     if (!assign) {
1657         /* Test and clear notifier before after disabling event,
1658          * in case poll callback didn't have time to run. */
1659         virtio_queue_host_notifier_read(&vq->host_notifier);
1660     }
1661 }
1662 
1663 EventNotifier *virtio_queue_get_host_notifier(VirtQueue *vq)
1664 {
1665     return &vq->host_notifier;
1666 }
1667 
1668 void virtio_device_set_child_bus_name(VirtIODevice *vdev, char *bus_name)
1669 {
1670     g_free(vdev->bus_name);
1671     vdev->bus_name = g_strdup(bus_name);
1672 }
1673 
1674 static void virtio_device_realize(DeviceState *dev, Error **errp)
1675 {
1676     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
1677     VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(dev);
1678     Error *err = NULL;
1679 
1680     if (vdc->realize != NULL) {
1681         vdc->realize(dev, &err);
1682         if (err != NULL) {
1683             error_propagate(errp, err);
1684             return;
1685         }
1686     }
1687 
1688     virtio_bus_device_plugged(vdev, &err);
1689     if (err != NULL) {
1690         error_propagate(errp, err);
1691         return;
1692     }
1693 }
1694 
1695 static void virtio_device_unrealize(DeviceState *dev, Error **errp)
1696 {
1697     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
1698     VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(dev);
1699     Error *err = NULL;
1700 
1701     virtio_bus_device_unplugged(vdev);
1702 
1703     if (vdc->unrealize != NULL) {
1704         vdc->unrealize(dev, &err);
1705         if (err != NULL) {
1706             error_propagate(errp, err);
1707             return;
1708         }
1709     }
1710 
1711     g_free(vdev->bus_name);
1712     vdev->bus_name = NULL;
1713 }
1714 
1715 static Property virtio_properties[] = {
1716     DEFINE_VIRTIO_COMMON_FEATURES(VirtIODevice, host_features),
1717     DEFINE_PROP_END_OF_LIST(),
1718 };
1719 
1720 static void virtio_device_class_init(ObjectClass *klass, void *data)
1721 {
1722     /* Set the default value here. */
1723     DeviceClass *dc = DEVICE_CLASS(klass);
1724 
1725     dc->realize = virtio_device_realize;
1726     dc->unrealize = virtio_device_unrealize;
1727     dc->bus_type = TYPE_VIRTIO_BUS;
1728     dc->props = virtio_properties;
1729 }
1730 
1731 static const TypeInfo virtio_device_info = {
1732     .name = TYPE_VIRTIO_DEVICE,
1733     .parent = TYPE_DEVICE,
1734     .instance_size = sizeof(VirtIODevice),
1735     .class_init = virtio_device_class_init,
1736     .abstract = true,
1737     .class_size = sizeof(VirtioDeviceClass),
1738 };
1739 
1740 static void virtio_register_types(void)
1741 {
1742     type_register_static(&virtio_device_info);
1743 }
1744 
1745 type_init(virtio_register_types)
1746