virtio: check VirtQueue Vring object is set
[qemu.git] / hw / virtio / virtio.c
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 "hw/virtio/virtio-access.h"
25 #include "sysemu/dma.h"
26
27 /*
28 * The alignment to use between consumer and producer parts of vring.
29 * x86 pagesize again. This is the default, used by transports like PCI
30 * which don't provide a means for the guest to tell the host the alignment.
31 */
32 #define VIRTIO_PCI_VRING_ALIGN 4096
33
34 typedef struct VRingDesc
35 {
36 uint64_t addr;
37 uint32_t len;
38 uint16_t flags;
39 uint16_t next;
40 } VRingDesc;
41
42 typedef struct VRingAvail
43 {
44 uint16_t flags;
45 uint16_t idx;
46 uint16_t ring[0];
47 } VRingAvail;
48
49 typedef struct VRingUsedElem
50 {
51 uint32_t id;
52 uint32_t len;
53 } VRingUsedElem;
54
55 typedef struct VRingUsed
56 {
57 uint16_t flags;
58 uint16_t idx;
59 VRingUsedElem ring[0];
60 } VRingUsed;
61
62 typedef struct VRingMemoryRegionCaches {
63 struct rcu_head rcu;
64 MemoryRegionCache desc;
65 MemoryRegionCache avail;
66 MemoryRegionCache used;
67 } VRingMemoryRegionCaches;
68
69 typedef struct VRing
70 {
71 unsigned int num;
72 unsigned int num_default;
73 unsigned int align;
74 hwaddr desc;
75 hwaddr avail;
76 hwaddr used;
77 VRingMemoryRegionCaches *caches;
78 } VRing;
79
80 struct VirtQueue
81 {
82 VRing vring;
83
84 /* Next head to pop */
85 uint16_t last_avail_idx;
86
87 /* Last avail_idx read from VQ. */
88 uint16_t shadow_avail_idx;
89
90 uint16_t used_idx;
91
92 /* Last used index value we have signalled on */
93 uint16_t signalled_used;
94
95 /* Last used index value we have signalled on */
96 bool signalled_used_valid;
97
98 /* Notification enabled? */
99 bool notification;
100
101 uint16_t queue_index;
102
103 unsigned int inuse;
104
105 uint16_t vector;
106 VirtIOHandleOutput handle_output;
107 VirtIOHandleAIOOutput handle_aio_output;
108 VirtIODevice *vdev;
109 EventNotifier guest_notifier;
110 EventNotifier host_notifier;
111 QLIST_ENTRY(VirtQueue) node;
112 };
113
114 static void virtio_free_region_cache(VRingMemoryRegionCaches *caches)
115 {
116 if (!caches) {
117 return;
118 }
119
120 address_space_cache_destroy(&caches->desc);
121 address_space_cache_destroy(&caches->avail);
122 address_space_cache_destroy(&caches->used);
123 g_free(caches);
124 }
125
126 static void virtio_init_region_cache(VirtIODevice *vdev, int n)
127 {
128 VirtQueue *vq = &vdev->vq[n];
129 VRingMemoryRegionCaches *old = vq->vring.caches;
130 VRingMemoryRegionCaches *new;
131 hwaddr addr, size;
132 int event_size;
133 int64_t len;
134
135 event_size = virtio_vdev_has_feature(vq->vdev, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0;
136
137 addr = vq->vring.desc;
138 if (!addr) {
139 return;
140 }
141 new = g_new0(VRingMemoryRegionCaches, 1);
142 size = virtio_queue_get_desc_size(vdev, n);
143 len = address_space_cache_init(&new->desc, vdev->dma_as,
144 addr, size, false);
145 if (len < size) {
146 virtio_error(vdev, "Cannot map desc");
147 goto err_desc;
148 }
149
150 size = virtio_queue_get_used_size(vdev, n) + event_size;
151 len = address_space_cache_init(&new->used, vdev->dma_as,
152 vq->vring.used, size, true);
153 if (len < size) {
154 virtio_error(vdev, "Cannot map used");
155 goto err_used;
156 }
157
158 size = virtio_queue_get_avail_size(vdev, n) + event_size;
159 len = address_space_cache_init(&new->avail, vdev->dma_as,
160 vq->vring.avail, size, false);
161 if (len < size) {
162 virtio_error(vdev, "Cannot map avail");
163 goto err_avail;
164 }
165
166 atomic_rcu_set(&vq->vring.caches, new);
167 if (old) {
168 call_rcu(old, virtio_free_region_cache, rcu);
169 }
170 return;
171
172 err_avail:
173 address_space_cache_destroy(&new->used);
174 err_used:
175 address_space_cache_destroy(&new->desc);
176 err_desc:
177 g_free(new);
178 }
179
180 /* virt queue functions */
181 void virtio_queue_update_rings(VirtIODevice *vdev, int n)
182 {
183 VRing *vring = &vdev->vq[n].vring;
184
185 if (!vring->num || !vring->desc || !vring->align) {
186 /* not yet setup -> nothing to do */
187 return;
188 }
189 vring->avail = vring->desc + vring->num * sizeof(VRingDesc);
190 vring->used = vring_align(vring->avail +
191 offsetof(VRingAvail, ring[vring->num]),
192 vring->align);
193 virtio_init_region_cache(vdev, n);
194 }
195
196 /* Called within rcu_read_lock(). */
197 static void vring_desc_read(VirtIODevice *vdev, VRingDesc *desc,
198 MemoryRegionCache *cache, int i)
199 {
200 address_space_read_cached(cache, i * sizeof(VRingDesc),
201 desc, sizeof(VRingDesc));
202 virtio_tswap64s(vdev, &desc->addr);
203 virtio_tswap32s(vdev, &desc->len);
204 virtio_tswap16s(vdev, &desc->flags);
205 virtio_tswap16s(vdev, &desc->next);
206 }
207
208 static VRingMemoryRegionCaches *vring_get_region_caches(struct VirtQueue *vq)
209 {
210 VRingMemoryRegionCaches *caches = atomic_rcu_read(&vq->vring.caches);
211 assert(caches != NULL);
212 return caches;
213 }
214 /* Called within rcu_read_lock(). */
215 static inline uint16_t vring_avail_flags(VirtQueue *vq)
216 {
217 VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
218 hwaddr pa = offsetof(VRingAvail, flags);
219 return virtio_lduw_phys_cached(vq->vdev, &caches->avail, pa);
220 }
221
222 /* Called within rcu_read_lock(). */
223 static inline uint16_t vring_avail_idx(VirtQueue *vq)
224 {
225 VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
226 hwaddr pa = offsetof(VRingAvail, idx);
227 vq->shadow_avail_idx = virtio_lduw_phys_cached(vq->vdev, &caches->avail, pa);
228 return vq->shadow_avail_idx;
229 }
230
231 /* Called within rcu_read_lock(). */
232 static inline uint16_t vring_avail_ring(VirtQueue *vq, int i)
233 {
234 VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
235 hwaddr pa = offsetof(VRingAvail, ring[i]);
236 return virtio_lduw_phys_cached(vq->vdev, &caches->avail, pa);
237 }
238
239 /* Called within rcu_read_lock(). */
240 static inline uint16_t vring_get_used_event(VirtQueue *vq)
241 {
242 return vring_avail_ring(vq, vq->vring.num);
243 }
244
245 /* Called within rcu_read_lock(). */
246 static inline void vring_used_write(VirtQueue *vq, VRingUsedElem *uelem,
247 int i)
248 {
249 VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
250 hwaddr pa = offsetof(VRingUsed, ring[i]);
251 virtio_tswap32s(vq->vdev, &uelem->id);
252 virtio_tswap32s(vq->vdev, &uelem->len);
253 address_space_write_cached(&caches->used, pa, uelem, sizeof(VRingUsedElem));
254 address_space_cache_invalidate(&caches->used, pa, sizeof(VRingUsedElem));
255 }
256
257 /* Called within rcu_read_lock(). */
258 static uint16_t vring_used_idx(VirtQueue *vq)
259 {
260 VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
261 hwaddr pa = offsetof(VRingUsed, idx);
262 return virtio_lduw_phys_cached(vq->vdev, &caches->used, pa);
263 }
264
265 /* Called within rcu_read_lock(). */
266 static inline void vring_used_idx_set(VirtQueue *vq, uint16_t val)
267 {
268 VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
269 hwaddr pa = offsetof(VRingUsed, idx);
270 virtio_stw_phys_cached(vq->vdev, &caches->used, pa, val);
271 address_space_cache_invalidate(&caches->used, pa, sizeof(val));
272 vq->used_idx = val;
273 }
274
275 /* Called within rcu_read_lock(). */
276 static inline void vring_used_flags_set_bit(VirtQueue *vq, int mask)
277 {
278 VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
279 VirtIODevice *vdev = vq->vdev;
280 hwaddr pa = offsetof(VRingUsed, flags);
281 uint16_t flags = virtio_lduw_phys_cached(vq->vdev, &caches->used, pa);
282
283 virtio_stw_phys_cached(vdev, &caches->used, pa, flags | mask);
284 address_space_cache_invalidate(&caches->used, pa, sizeof(flags));
285 }
286
287 /* Called within rcu_read_lock(). */
288 static inline void vring_used_flags_unset_bit(VirtQueue *vq, int mask)
289 {
290 VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
291 VirtIODevice *vdev = vq->vdev;
292 hwaddr pa = offsetof(VRingUsed, flags);
293 uint16_t flags = virtio_lduw_phys_cached(vq->vdev, &caches->used, pa);
294
295 virtio_stw_phys_cached(vdev, &caches->used, pa, flags & ~mask);
296 address_space_cache_invalidate(&caches->used, pa, sizeof(flags));
297 }
298
299 /* Called within rcu_read_lock(). */
300 static inline void vring_set_avail_event(VirtQueue *vq, uint16_t val)
301 {
302 VRingMemoryRegionCaches *caches;
303 hwaddr pa;
304 if (!vq->notification) {
305 return;
306 }
307
308 caches = vring_get_region_caches(vq);
309 pa = offsetof(VRingUsed, ring[vq->vring.num]);
310 virtio_stw_phys_cached(vq->vdev, &caches->used, pa, val);
311 address_space_cache_invalidate(&caches->used, pa, sizeof(val));
312 }
313
314 void virtio_queue_set_notification(VirtQueue *vq, int enable)
315 {
316 vq->notification = enable;
317
318 if (!vq->vring.desc) {
319 return;
320 }
321
322 rcu_read_lock();
323 if (virtio_vdev_has_feature(vq->vdev, VIRTIO_RING_F_EVENT_IDX)) {
324 vring_set_avail_event(vq, vring_avail_idx(vq));
325 } else if (enable) {
326 vring_used_flags_unset_bit(vq, VRING_USED_F_NO_NOTIFY);
327 } else {
328 vring_used_flags_set_bit(vq, VRING_USED_F_NO_NOTIFY);
329 }
330 if (enable) {
331 /* Expose avail event/used flags before caller checks the avail idx. */
332 smp_mb();
333 }
334 rcu_read_unlock();
335 }
336
337 int virtio_queue_ready(VirtQueue *vq)
338 {
339 return vq->vring.avail != 0;
340 }
341
342 /* Fetch avail_idx from VQ memory only when we really need to know if
343 * guest has added some buffers.
344 * Called within rcu_read_lock(). */
345 static int virtio_queue_empty_rcu(VirtQueue *vq)
346 {
347 if (unlikely(!vq->vring.avail)) {
348 return 1;
349 }
350
351 if (vq->shadow_avail_idx != vq->last_avail_idx) {
352 return 0;
353 }
354
355 return vring_avail_idx(vq) == vq->last_avail_idx;
356 }
357
358 int virtio_queue_empty(VirtQueue *vq)
359 {
360 bool empty;
361
362 if (unlikely(!vq->vring.avail)) {
363 return 1;
364 }
365
366 if (vq->shadow_avail_idx != vq->last_avail_idx) {
367 return 0;
368 }
369
370 rcu_read_lock();
371 empty = vring_avail_idx(vq) == vq->last_avail_idx;
372 rcu_read_unlock();
373 return empty;
374 }
375
376 static void virtqueue_unmap_sg(VirtQueue *vq, const VirtQueueElement *elem,
377 unsigned int len)
378 {
379 AddressSpace *dma_as = vq->vdev->dma_as;
380 unsigned int offset;
381 int i;
382
383 offset = 0;
384 for (i = 0; i < elem->in_num; i++) {
385 size_t size = MIN(len - offset, elem->in_sg[i].iov_len);
386
387 dma_memory_unmap(dma_as, elem->in_sg[i].iov_base,
388 elem->in_sg[i].iov_len,
389 DMA_DIRECTION_FROM_DEVICE, size);
390
391 offset += size;
392 }
393
394 for (i = 0; i < elem->out_num; i++)
395 dma_memory_unmap(dma_as, elem->out_sg[i].iov_base,
396 elem->out_sg[i].iov_len,
397 DMA_DIRECTION_TO_DEVICE,
398 elem->out_sg[i].iov_len);
399 }
400
401 /* virtqueue_detach_element:
402 * @vq: The #VirtQueue
403 * @elem: The #VirtQueueElement
404 * @len: number of bytes written
405 *
406 * Detach the element from the virtqueue. This function is suitable for device
407 * reset or other situations where a #VirtQueueElement is simply freed and will
408 * not be pushed or discarded.
409 */
410 void virtqueue_detach_element(VirtQueue *vq, const VirtQueueElement *elem,
411 unsigned int len)
412 {
413 vq->inuse--;
414 virtqueue_unmap_sg(vq, elem, len);
415 }
416
417 /* virtqueue_unpop:
418 * @vq: The #VirtQueue
419 * @elem: The #VirtQueueElement
420 * @len: number of bytes written
421 *
422 * Pretend the most recent element wasn't popped from the virtqueue. The next
423 * call to virtqueue_pop() will refetch the element.
424 */
425 void virtqueue_unpop(VirtQueue *vq, const VirtQueueElement *elem,
426 unsigned int len)
427 {
428 vq->last_avail_idx--;
429 virtqueue_detach_element(vq, elem, len);
430 }
431
432 /* virtqueue_rewind:
433 * @vq: The #VirtQueue
434 * @num: Number of elements to push back
435 *
436 * Pretend that elements weren't popped from the virtqueue. The next
437 * virtqueue_pop() will refetch the oldest element.
438 *
439 * Use virtqueue_unpop() instead if you have a VirtQueueElement.
440 *
441 * Returns: true on success, false if @num is greater than the number of in use
442 * elements.
443 */
444 bool virtqueue_rewind(VirtQueue *vq, unsigned int num)
445 {
446 if (num > vq->inuse) {
447 return false;
448 }
449 vq->last_avail_idx -= num;
450 vq->inuse -= num;
451 return true;
452 }
453
454 /* Called within rcu_read_lock(). */
455 void virtqueue_fill(VirtQueue *vq, const VirtQueueElement *elem,
456 unsigned int len, unsigned int idx)
457 {
458 VRingUsedElem uelem;
459
460 trace_virtqueue_fill(vq, elem, len, idx);
461
462 virtqueue_unmap_sg(vq, elem, len);
463
464 if (unlikely(vq->vdev->broken)) {
465 return;
466 }
467
468 if (unlikely(!vq->vring.used)) {
469 return;
470 }
471
472 idx = (idx + vq->used_idx) % vq->vring.num;
473
474 uelem.id = elem->index;
475 uelem.len = len;
476 vring_used_write(vq, &uelem, idx);
477 }
478
479 /* Called within rcu_read_lock(). */
480 void virtqueue_flush(VirtQueue *vq, unsigned int count)
481 {
482 uint16_t old, new;
483
484 if (unlikely(vq->vdev->broken)) {
485 vq->inuse -= count;
486 return;
487 }
488
489 if (unlikely(!vq->vring.used)) {
490 return;
491 }
492
493 /* Make sure buffer is written before we update index. */
494 smp_wmb();
495 trace_virtqueue_flush(vq, count);
496 old = vq->used_idx;
497 new = old + count;
498 vring_used_idx_set(vq, new);
499 vq->inuse -= count;
500 if (unlikely((int16_t)(new - vq->signalled_used) < (uint16_t)(new - old)))
501 vq->signalled_used_valid = false;
502 }
503
504 void virtqueue_push(VirtQueue *vq, const VirtQueueElement *elem,
505 unsigned int len)
506 {
507 rcu_read_lock();
508 virtqueue_fill(vq, elem, len, 0);
509 virtqueue_flush(vq, 1);
510 rcu_read_unlock();
511 }
512
513 /* Called within rcu_read_lock(). */
514 static int virtqueue_num_heads(VirtQueue *vq, unsigned int idx)
515 {
516 uint16_t num_heads = vring_avail_idx(vq) - idx;
517
518 /* Check it isn't doing very strange things with descriptor numbers. */
519 if (num_heads > vq->vring.num) {
520 virtio_error(vq->vdev, "Guest moved used index from %u to %u",
521 idx, vq->shadow_avail_idx);
522 return -EINVAL;
523 }
524 /* On success, callers read a descriptor at vq->last_avail_idx.
525 * Make sure descriptor read does not bypass avail index read. */
526 if (num_heads) {
527 smp_rmb();
528 }
529
530 return num_heads;
531 }
532
533 /* Called within rcu_read_lock(). */
534 static bool virtqueue_get_head(VirtQueue *vq, unsigned int idx,
535 unsigned int *head)
536 {
537 /* Grab the next descriptor number they're advertising, and increment
538 * the index we've seen. */
539 *head = vring_avail_ring(vq, idx % vq->vring.num);
540
541 /* If their number is silly, that's a fatal mistake. */
542 if (*head >= vq->vring.num) {
543 virtio_error(vq->vdev, "Guest says index %u is available", *head);
544 return false;
545 }
546
547 return true;
548 }
549
550 enum {
551 VIRTQUEUE_READ_DESC_ERROR = -1,
552 VIRTQUEUE_READ_DESC_DONE = 0, /* end of chain */
553 VIRTQUEUE_READ_DESC_MORE = 1, /* more buffers in chain */
554 };
555
556 static int virtqueue_read_next_desc(VirtIODevice *vdev, VRingDesc *desc,
557 MemoryRegionCache *desc_cache, unsigned int max,
558 unsigned int *next)
559 {
560 /* If this descriptor says it doesn't chain, we're done. */
561 if (!(desc->flags & VRING_DESC_F_NEXT)) {
562 return VIRTQUEUE_READ_DESC_DONE;
563 }
564
565 /* Check they're not leading us off end of descriptors. */
566 *next = desc->next;
567 /* Make sure compiler knows to grab that: we don't want it changing! */
568 smp_wmb();
569
570 if (*next >= max) {
571 virtio_error(vdev, "Desc next is %u", *next);
572 return VIRTQUEUE_READ_DESC_ERROR;
573 }
574
575 vring_desc_read(vdev, desc, desc_cache, *next);
576 return VIRTQUEUE_READ_DESC_MORE;
577 }
578
579 void virtqueue_get_avail_bytes(VirtQueue *vq, unsigned int *in_bytes,
580 unsigned int *out_bytes,
581 unsigned max_in_bytes, unsigned max_out_bytes)
582 {
583 VirtIODevice *vdev = vq->vdev;
584 unsigned int max, idx;
585 unsigned int total_bufs, in_total, out_total;
586 VRingMemoryRegionCaches *caches;
587 MemoryRegionCache indirect_desc_cache = MEMORY_REGION_CACHE_INVALID;
588 int64_t len = 0;
589 int rc;
590
591 if (unlikely(!vq->vring.desc)) {
592 if (in_bytes) {
593 *in_bytes = 0;
594 }
595 if (out_bytes) {
596 *out_bytes = 0;
597 }
598 return;
599 }
600
601 rcu_read_lock();
602 idx = vq->last_avail_idx;
603 total_bufs = in_total = out_total = 0;
604
605 max = vq->vring.num;
606 caches = vring_get_region_caches(vq);
607 if (caches->desc.len < max * sizeof(VRingDesc)) {
608 virtio_error(vdev, "Cannot map descriptor ring");
609 goto err;
610 }
611
612 while ((rc = virtqueue_num_heads(vq, idx)) > 0) {
613 MemoryRegionCache *desc_cache = &caches->desc;
614 unsigned int num_bufs;
615 VRingDesc desc;
616 unsigned int i;
617
618 num_bufs = total_bufs;
619
620 if (!virtqueue_get_head(vq, idx++, &i)) {
621 goto err;
622 }
623
624 vring_desc_read(vdev, &desc, desc_cache, i);
625
626 if (desc.flags & VRING_DESC_F_INDIRECT) {
627 if (desc.len % sizeof(VRingDesc)) {
628 virtio_error(vdev, "Invalid size for indirect buffer table");
629 goto err;
630 }
631
632 /* If we've got too many, that implies a descriptor loop. */
633 if (num_bufs >= max) {
634 virtio_error(vdev, "Looped descriptor");
635 goto err;
636 }
637
638 /* loop over the indirect descriptor table */
639 len = address_space_cache_init(&indirect_desc_cache,
640 vdev->dma_as,
641 desc.addr, desc.len, false);
642 desc_cache = &indirect_desc_cache;
643 if (len < desc.len) {
644 virtio_error(vdev, "Cannot map indirect buffer");
645 goto err;
646 }
647
648 max = desc.len / sizeof(VRingDesc);
649 num_bufs = i = 0;
650 vring_desc_read(vdev, &desc, desc_cache, i);
651 }
652
653 do {
654 /* If we've got too many, that implies a descriptor loop. */
655 if (++num_bufs > max) {
656 virtio_error(vdev, "Looped descriptor");
657 goto err;
658 }
659
660 if (desc.flags & VRING_DESC_F_WRITE) {
661 in_total += desc.len;
662 } else {
663 out_total += desc.len;
664 }
665 if (in_total >= max_in_bytes && out_total >= max_out_bytes) {
666 goto done;
667 }
668
669 rc = virtqueue_read_next_desc(vdev, &desc, desc_cache, max, &i);
670 } while (rc == VIRTQUEUE_READ_DESC_MORE);
671
672 if (rc == VIRTQUEUE_READ_DESC_ERROR) {
673 goto err;
674 }
675
676 if (desc_cache == &indirect_desc_cache) {
677 address_space_cache_destroy(&indirect_desc_cache);
678 total_bufs++;
679 } else {
680 total_bufs = num_bufs;
681 }
682 }
683
684 if (rc < 0) {
685 goto err;
686 }
687
688 done:
689 address_space_cache_destroy(&indirect_desc_cache);
690 if (in_bytes) {
691 *in_bytes = in_total;
692 }
693 if (out_bytes) {
694 *out_bytes = out_total;
695 }
696 rcu_read_unlock();
697 return;
698
699 err:
700 in_total = out_total = 0;
701 goto done;
702 }
703
704 int virtqueue_avail_bytes(VirtQueue *vq, unsigned int in_bytes,
705 unsigned int out_bytes)
706 {
707 unsigned int in_total, out_total;
708
709 virtqueue_get_avail_bytes(vq, &in_total, &out_total, in_bytes, out_bytes);
710 return in_bytes <= in_total && out_bytes <= out_total;
711 }
712
713 static bool virtqueue_map_desc(VirtIODevice *vdev, unsigned int *p_num_sg,
714 hwaddr *addr, struct iovec *iov,
715 unsigned int max_num_sg, bool is_write,
716 hwaddr pa, size_t sz)
717 {
718 bool ok = false;
719 unsigned num_sg = *p_num_sg;
720 assert(num_sg <= max_num_sg);
721
722 if (!sz) {
723 virtio_error(vdev, "virtio: zero sized buffers are not allowed");
724 goto out;
725 }
726
727 while (sz) {
728 hwaddr len = sz;
729
730 if (num_sg == max_num_sg) {
731 virtio_error(vdev, "virtio: too many write descriptors in "
732 "indirect table");
733 goto out;
734 }
735
736 iov[num_sg].iov_base = dma_memory_map(vdev->dma_as, pa, &len,
737 is_write ?
738 DMA_DIRECTION_FROM_DEVICE :
739 DMA_DIRECTION_TO_DEVICE);
740 if (!iov[num_sg].iov_base) {
741 virtio_error(vdev, "virtio: bogus descriptor or out of resources");
742 goto out;
743 }
744
745 iov[num_sg].iov_len = len;
746 addr[num_sg] = pa;
747
748 sz -= len;
749 pa += len;
750 num_sg++;
751 }
752 ok = true;
753
754 out:
755 *p_num_sg = num_sg;
756 return ok;
757 }
758
759 /* Only used by error code paths before we have a VirtQueueElement (therefore
760 * virtqueue_unmap_sg() can't be used). Assumes buffers weren't written to
761 * yet.
762 */
763 static void virtqueue_undo_map_desc(unsigned int out_num, unsigned int in_num,
764 struct iovec *iov)
765 {
766 unsigned int i;
767
768 for (i = 0; i < out_num + in_num; i++) {
769 int is_write = i >= out_num;
770
771 cpu_physical_memory_unmap(iov->iov_base, iov->iov_len, is_write, 0);
772 iov++;
773 }
774 }
775
776 static void virtqueue_map_iovec(VirtIODevice *vdev, struct iovec *sg,
777 hwaddr *addr, unsigned int *num_sg,
778 int is_write)
779 {
780 unsigned int i;
781 hwaddr len;
782
783 for (i = 0; i < *num_sg; i++) {
784 len = sg[i].iov_len;
785 sg[i].iov_base = dma_memory_map(vdev->dma_as,
786 addr[i], &len, is_write ?
787 DMA_DIRECTION_FROM_DEVICE :
788 DMA_DIRECTION_TO_DEVICE);
789 if (!sg[i].iov_base) {
790 error_report("virtio: error trying to map MMIO memory");
791 exit(1);
792 }
793 if (len != sg[i].iov_len) {
794 error_report("virtio: unexpected memory split");
795 exit(1);
796 }
797 }
798 }
799
800 void virtqueue_map(VirtIODevice *vdev, VirtQueueElement *elem)
801 {
802 virtqueue_map_iovec(vdev, elem->in_sg, elem->in_addr, &elem->in_num, 1);
803 virtqueue_map_iovec(vdev, elem->out_sg, elem->out_addr, &elem->out_num, 0);
804 }
805
806 static void *virtqueue_alloc_element(size_t sz, unsigned out_num, unsigned in_num)
807 {
808 VirtQueueElement *elem;
809 size_t in_addr_ofs = QEMU_ALIGN_UP(sz, __alignof__(elem->in_addr[0]));
810 size_t out_addr_ofs = in_addr_ofs + in_num * sizeof(elem->in_addr[0]);
811 size_t out_addr_end = out_addr_ofs + out_num * sizeof(elem->out_addr[0]);
812 size_t in_sg_ofs = QEMU_ALIGN_UP(out_addr_end, __alignof__(elem->in_sg[0]));
813 size_t out_sg_ofs = in_sg_ofs + in_num * sizeof(elem->in_sg[0]);
814 size_t out_sg_end = out_sg_ofs + out_num * sizeof(elem->out_sg[0]);
815
816 assert(sz >= sizeof(VirtQueueElement));
817 elem = g_malloc(out_sg_end);
818 trace_virtqueue_alloc_element(elem, sz, in_num, out_num);
819 elem->out_num = out_num;
820 elem->in_num = in_num;
821 elem->in_addr = (void *)elem + in_addr_ofs;
822 elem->out_addr = (void *)elem + out_addr_ofs;
823 elem->in_sg = (void *)elem + in_sg_ofs;
824 elem->out_sg = (void *)elem + out_sg_ofs;
825 return elem;
826 }
827
828 void *virtqueue_pop(VirtQueue *vq, size_t sz)
829 {
830 unsigned int i, head, max;
831 VRingMemoryRegionCaches *caches;
832 MemoryRegionCache indirect_desc_cache = MEMORY_REGION_CACHE_INVALID;
833 MemoryRegionCache *desc_cache;
834 int64_t len;
835 VirtIODevice *vdev = vq->vdev;
836 VirtQueueElement *elem = NULL;
837 unsigned out_num, in_num, elem_entries;
838 hwaddr addr[VIRTQUEUE_MAX_SIZE];
839 struct iovec iov[VIRTQUEUE_MAX_SIZE];
840 VRingDesc desc;
841 int rc;
842
843 if (unlikely(vdev->broken)) {
844 return NULL;
845 }
846 rcu_read_lock();
847 if (virtio_queue_empty_rcu(vq)) {
848 goto done;
849 }
850 /* Needed after virtio_queue_empty(), see comment in
851 * virtqueue_num_heads(). */
852 smp_rmb();
853
854 /* When we start there are none of either input nor output. */
855 out_num = in_num = elem_entries = 0;
856
857 max = vq->vring.num;
858
859 if (vq->inuse >= vq->vring.num) {
860 virtio_error(vdev, "Virtqueue size exceeded");
861 goto done;
862 }
863
864 if (!virtqueue_get_head(vq, vq->last_avail_idx++, &head)) {
865 goto done;
866 }
867
868 if (virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX)) {
869 vring_set_avail_event(vq, vq->last_avail_idx);
870 }
871
872 i = head;
873
874 caches = vring_get_region_caches(vq);
875 if (caches->desc.len < max * sizeof(VRingDesc)) {
876 virtio_error(vdev, "Cannot map descriptor ring");
877 goto done;
878 }
879
880 desc_cache = &caches->desc;
881 vring_desc_read(vdev, &desc, desc_cache, i);
882 if (desc.flags & VRING_DESC_F_INDIRECT) {
883 if (desc.len % sizeof(VRingDesc)) {
884 virtio_error(vdev, "Invalid size for indirect buffer table");
885 goto done;
886 }
887
888 /* loop over the indirect descriptor table */
889 len = address_space_cache_init(&indirect_desc_cache, vdev->dma_as,
890 desc.addr, desc.len, false);
891 desc_cache = &indirect_desc_cache;
892 if (len < desc.len) {
893 virtio_error(vdev, "Cannot map indirect buffer");
894 goto done;
895 }
896
897 max = desc.len / sizeof(VRingDesc);
898 i = 0;
899 vring_desc_read(vdev, &desc, desc_cache, i);
900 }
901
902 /* Collect all the descriptors */
903 do {
904 bool map_ok;
905
906 if (desc.flags & VRING_DESC_F_WRITE) {
907 map_ok = virtqueue_map_desc(vdev, &in_num, addr + out_num,
908 iov + out_num,
909 VIRTQUEUE_MAX_SIZE - out_num, true,
910 desc.addr, desc.len);
911 } else {
912 if (in_num) {
913 virtio_error(vdev, "Incorrect order for descriptors");
914 goto err_undo_map;
915 }
916 map_ok = virtqueue_map_desc(vdev, &out_num, addr, iov,
917 VIRTQUEUE_MAX_SIZE, false,
918 desc.addr, desc.len);
919 }
920 if (!map_ok) {
921 goto err_undo_map;
922 }
923
924 /* If we've got too many, that implies a descriptor loop. */
925 if (++elem_entries > max) {
926 virtio_error(vdev, "Looped descriptor");
927 goto err_undo_map;
928 }
929
930 rc = virtqueue_read_next_desc(vdev, &desc, desc_cache, max, &i);
931 } while (rc == VIRTQUEUE_READ_DESC_MORE);
932
933 if (rc == VIRTQUEUE_READ_DESC_ERROR) {
934 goto err_undo_map;
935 }
936
937 /* Now copy what we have collected and mapped */
938 elem = virtqueue_alloc_element(sz, out_num, in_num);
939 elem->index = head;
940 for (i = 0; i < out_num; i++) {
941 elem->out_addr[i] = addr[i];
942 elem->out_sg[i] = iov[i];
943 }
944 for (i = 0; i < in_num; i++) {
945 elem->in_addr[i] = addr[out_num + i];
946 elem->in_sg[i] = iov[out_num + i];
947 }
948
949 vq->inuse++;
950
951 trace_virtqueue_pop(vq, elem, elem->in_num, elem->out_num);
952 done:
953 address_space_cache_destroy(&indirect_desc_cache);
954 rcu_read_unlock();
955
956 return elem;
957
958 err_undo_map:
959 virtqueue_undo_map_desc(out_num, in_num, iov);
960 goto done;
961 }
962
963 /* virtqueue_drop_all:
964 * @vq: The #VirtQueue
965 * Drops all queued buffers and indicates them to the guest
966 * as if they are done. Useful when buffers can not be
967 * processed but must be returned to the guest.
968 */
969 unsigned int virtqueue_drop_all(VirtQueue *vq)
970 {
971 unsigned int dropped = 0;
972 VirtQueueElement elem = {};
973 VirtIODevice *vdev = vq->vdev;
974 bool fEventIdx = virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX);
975
976 if (unlikely(vdev->broken)) {
977 return 0;
978 }
979
980 while (!virtio_queue_empty(vq) && vq->inuse < vq->vring.num) {
981 /* works similar to virtqueue_pop but does not map buffers
982 * and does not allocate any memory */
983 smp_rmb();
984 if (!virtqueue_get_head(vq, vq->last_avail_idx, &elem.index)) {
985 break;
986 }
987 vq->inuse++;
988 vq->last_avail_idx++;
989 if (fEventIdx) {
990 vring_set_avail_event(vq, vq->last_avail_idx);
991 }
992 /* immediately push the element, nothing to unmap
993 * as both in_num and out_num are set to 0 */
994 virtqueue_push(vq, &elem, 0);
995 dropped++;
996 }
997
998 return dropped;
999 }
1000
1001 /* Reading and writing a structure directly to QEMUFile is *awful*, but
1002 * it is what QEMU has always done by mistake. We can change it sooner
1003 * or later by bumping the version number of the affected vm states.
1004 * In the meanwhile, since the in-memory layout of VirtQueueElement
1005 * has changed, we need to marshal to and from the layout that was
1006 * used before the change.
1007 */
1008 typedef struct VirtQueueElementOld {
1009 unsigned int index;
1010 unsigned int out_num;
1011 unsigned int in_num;
1012 hwaddr in_addr[VIRTQUEUE_MAX_SIZE];
1013 hwaddr out_addr[VIRTQUEUE_MAX_SIZE];
1014 struct iovec in_sg[VIRTQUEUE_MAX_SIZE];
1015 struct iovec out_sg[VIRTQUEUE_MAX_SIZE];
1016 } VirtQueueElementOld;
1017
1018 void *qemu_get_virtqueue_element(VirtIODevice *vdev, QEMUFile *f, size_t sz)
1019 {
1020 VirtQueueElement *elem;
1021 VirtQueueElementOld data;
1022 int i;
1023
1024 qemu_get_buffer(f, (uint8_t *)&data, sizeof(VirtQueueElementOld));
1025
1026 /* TODO: teach all callers that this can fail, and return failure instead
1027 * of asserting here.
1028 * This is just one thing (there are probably more) that must be
1029 * fixed before we can allow NDEBUG compilation.
1030 */
1031 assert(ARRAY_SIZE(data.in_addr) >= data.in_num);
1032 assert(ARRAY_SIZE(data.out_addr) >= data.out_num);
1033
1034 elem = virtqueue_alloc_element(sz, data.out_num, data.in_num);
1035 elem->index = data.index;
1036
1037 for (i = 0; i < elem->in_num; i++) {
1038 elem->in_addr[i] = data.in_addr[i];
1039 }
1040
1041 for (i = 0; i < elem->out_num; i++) {
1042 elem->out_addr[i] = data.out_addr[i];
1043 }
1044
1045 for (i = 0; i < elem->in_num; i++) {
1046 /* Base is overwritten by virtqueue_map. */
1047 elem->in_sg[i].iov_base = 0;
1048 elem->in_sg[i].iov_len = data.in_sg[i].iov_len;
1049 }
1050
1051 for (i = 0; i < elem->out_num; i++) {
1052 /* Base is overwritten by virtqueue_map. */
1053 elem->out_sg[i].iov_base = 0;
1054 elem->out_sg[i].iov_len = data.out_sg[i].iov_len;
1055 }
1056
1057 virtqueue_map(vdev, elem);
1058 return elem;
1059 }
1060
1061 void qemu_put_virtqueue_element(QEMUFile *f, VirtQueueElement *elem)
1062 {
1063 VirtQueueElementOld data;
1064 int i;
1065
1066 memset(&data, 0, sizeof(data));
1067 data.index = elem->index;
1068 data.in_num = elem->in_num;
1069 data.out_num = elem->out_num;
1070
1071 for (i = 0; i < elem->in_num; i++) {
1072 data.in_addr[i] = elem->in_addr[i];
1073 }
1074
1075 for (i = 0; i < elem->out_num; i++) {
1076 data.out_addr[i] = elem->out_addr[i];
1077 }
1078
1079 for (i = 0; i < elem->in_num; i++) {
1080 /* Base is overwritten by virtqueue_map when loading. Do not
1081 * save it, as it would leak the QEMU address space layout. */
1082 data.in_sg[i].iov_len = elem->in_sg[i].iov_len;
1083 }
1084
1085 for (i = 0; i < elem->out_num; i++) {
1086 /* Do not save iov_base as above. */
1087 data.out_sg[i].iov_len = elem->out_sg[i].iov_len;
1088 }
1089 qemu_put_buffer(f, (uint8_t *)&data, sizeof(VirtQueueElementOld));
1090 }
1091
1092 /* virtio device */
1093 static void virtio_notify_vector(VirtIODevice *vdev, uint16_t vector)
1094 {
1095 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
1096 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
1097
1098 if (unlikely(vdev->broken)) {
1099 return;
1100 }
1101
1102 if (k->notify) {
1103 k->notify(qbus->parent, vector);
1104 }
1105 }
1106
1107 void virtio_update_irq(VirtIODevice *vdev)
1108 {
1109 virtio_notify_vector(vdev, VIRTIO_NO_VECTOR);
1110 }
1111
1112 static int virtio_validate_features(VirtIODevice *vdev)
1113 {
1114 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1115
1116 if (virtio_host_has_feature(vdev, VIRTIO_F_IOMMU_PLATFORM) &&
1117 !virtio_vdev_has_feature(vdev, VIRTIO_F_IOMMU_PLATFORM)) {
1118 return -EFAULT;
1119 }
1120
1121 if (k->validate_features) {
1122 return k->validate_features(vdev);
1123 } else {
1124 return 0;
1125 }
1126 }
1127
1128 int virtio_set_status(VirtIODevice *vdev, uint8_t val)
1129 {
1130 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1131 trace_virtio_set_status(vdev, val);
1132
1133 if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
1134 if (!(vdev->status & VIRTIO_CONFIG_S_FEATURES_OK) &&
1135 val & VIRTIO_CONFIG_S_FEATURES_OK) {
1136 int ret = virtio_validate_features(vdev);
1137
1138 if (ret) {
1139 return ret;
1140 }
1141 }
1142 }
1143 if (k->set_status) {
1144 k->set_status(vdev, val);
1145 }
1146 vdev->status = val;
1147 return 0;
1148 }
1149
1150 bool target_words_bigendian(void);
1151 static enum virtio_device_endian virtio_default_endian(void)
1152 {
1153 if (target_words_bigendian()) {
1154 return VIRTIO_DEVICE_ENDIAN_BIG;
1155 } else {
1156 return VIRTIO_DEVICE_ENDIAN_LITTLE;
1157 }
1158 }
1159
1160 static enum virtio_device_endian virtio_current_cpu_endian(void)
1161 {
1162 CPUClass *cc = CPU_GET_CLASS(current_cpu);
1163
1164 if (cc->virtio_is_big_endian(current_cpu)) {
1165 return VIRTIO_DEVICE_ENDIAN_BIG;
1166 } else {
1167 return VIRTIO_DEVICE_ENDIAN_LITTLE;
1168 }
1169 }
1170
1171 static void virtio_virtqueue_reset_region_cache(struct VirtQueue *vq)
1172 {
1173 VRingMemoryRegionCaches *caches;
1174
1175 caches = atomic_read(&vq->vring.caches);
1176 atomic_rcu_set(&vq->vring.caches, NULL);
1177 if (caches) {
1178 call_rcu(caches, virtio_free_region_cache, rcu);
1179 }
1180 }
1181
1182 void virtio_reset(void *opaque)
1183 {
1184 VirtIODevice *vdev = opaque;
1185 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1186 int i;
1187
1188 virtio_set_status(vdev, 0);
1189 if (current_cpu) {
1190 /* Guest initiated reset */
1191 vdev->device_endian = virtio_current_cpu_endian();
1192 } else {
1193 /* System reset */
1194 vdev->device_endian = virtio_default_endian();
1195 }
1196
1197 if (k->reset) {
1198 k->reset(vdev);
1199 }
1200
1201 vdev->broken = false;
1202 vdev->guest_features = 0;
1203 vdev->queue_sel = 0;
1204 vdev->status = 0;
1205 atomic_set(&vdev->isr, 0);
1206 vdev->config_vector = VIRTIO_NO_VECTOR;
1207 virtio_notify_vector(vdev, vdev->config_vector);
1208
1209 for(i = 0; i < VIRTIO_QUEUE_MAX; i++) {
1210 vdev->vq[i].vring.desc = 0;
1211 vdev->vq[i].vring.avail = 0;
1212 vdev->vq[i].vring.used = 0;
1213 vdev->vq[i].last_avail_idx = 0;
1214 vdev->vq[i].shadow_avail_idx = 0;
1215 vdev->vq[i].used_idx = 0;
1216 virtio_queue_set_vector(vdev, i, VIRTIO_NO_VECTOR);
1217 vdev->vq[i].signalled_used = 0;
1218 vdev->vq[i].signalled_used_valid = false;
1219 vdev->vq[i].notification = true;
1220 vdev->vq[i].vring.num = vdev->vq[i].vring.num_default;
1221 vdev->vq[i].inuse = 0;
1222 virtio_virtqueue_reset_region_cache(&vdev->vq[i]);
1223 }
1224 }
1225
1226 uint32_t virtio_config_readb(VirtIODevice *vdev, uint32_t addr)
1227 {
1228 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1229 uint8_t val;
1230
1231 if (addr + sizeof(val) > vdev->config_len) {
1232 return (uint32_t)-1;
1233 }
1234
1235 k->get_config(vdev, vdev->config);
1236
1237 val = ldub_p(vdev->config + addr);
1238 return val;
1239 }
1240
1241 uint32_t virtio_config_readw(VirtIODevice *vdev, uint32_t addr)
1242 {
1243 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1244 uint16_t val;
1245
1246 if (addr + sizeof(val) > vdev->config_len) {
1247 return (uint32_t)-1;
1248 }
1249
1250 k->get_config(vdev, vdev->config);
1251
1252 val = lduw_p(vdev->config + addr);
1253 return val;
1254 }
1255
1256 uint32_t virtio_config_readl(VirtIODevice *vdev, uint32_t addr)
1257 {
1258 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1259 uint32_t val;
1260
1261 if (addr + sizeof(val) > vdev->config_len) {
1262 return (uint32_t)-1;
1263 }
1264
1265 k->get_config(vdev, vdev->config);
1266
1267 val = ldl_p(vdev->config + addr);
1268 return val;
1269 }
1270
1271 void virtio_config_writeb(VirtIODevice *vdev, uint32_t addr, uint32_t data)
1272 {
1273 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1274 uint8_t val = data;
1275
1276 if (addr + sizeof(val) > vdev->config_len) {
1277 return;
1278 }
1279
1280 stb_p(vdev->config + addr, val);
1281
1282 if (k->set_config) {
1283 k->set_config(vdev, vdev->config);
1284 }
1285 }
1286
1287 void virtio_config_writew(VirtIODevice *vdev, uint32_t addr, uint32_t data)
1288 {
1289 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1290 uint16_t val = data;
1291
1292 if (addr + sizeof(val) > vdev->config_len) {
1293 return;
1294 }
1295
1296 stw_p(vdev->config + addr, val);
1297
1298 if (k->set_config) {
1299 k->set_config(vdev, vdev->config);
1300 }
1301 }
1302
1303 void virtio_config_writel(VirtIODevice *vdev, uint32_t addr, uint32_t data)
1304 {
1305 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1306 uint32_t val = data;
1307
1308 if (addr + sizeof(val) > vdev->config_len) {
1309 return;
1310 }
1311
1312 stl_p(vdev->config + addr, val);
1313
1314 if (k->set_config) {
1315 k->set_config(vdev, vdev->config);
1316 }
1317 }
1318
1319 uint32_t virtio_config_modern_readb(VirtIODevice *vdev, uint32_t addr)
1320 {
1321 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1322 uint8_t val;
1323
1324 if (addr + sizeof(val) > vdev->config_len) {
1325 return (uint32_t)-1;
1326 }
1327
1328 k->get_config(vdev, vdev->config);
1329
1330 val = ldub_p(vdev->config + addr);
1331 return val;
1332 }
1333
1334 uint32_t virtio_config_modern_readw(VirtIODevice *vdev, uint32_t addr)
1335 {
1336 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1337 uint16_t val;
1338
1339 if (addr + sizeof(val) > vdev->config_len) {
1340 return (uint32_t)-1;
1341 }
1342
1343 k->get_config(vdev, vdev->config);
1344
1345 val = lduw_le_p(vdev->config + addr);
1346 return val;
1347 }
1348
1349 uint32_t virtio_config_modern_readl(VirtIODevice *vdev, uint32_t addr)
1350 {
1351 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1352 uint32_t val;
1353
1354 if (addr + sizeof(val) > vdev->config_len) {
1355 return (uint32_t)-1;
1356 }
1357
1358 k->get_config(vdev, vdev->config);
1359
1360 val = ldl_le_p(vdev->config + addr);
1361 return val;
1362 }
1363
1364 void virtio_config_modern_writeb(VirtIODevice *vdev,
1365 uint32_t addr, uint32_t data)
1366 {
1367 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1368 uint8_t val = data;
1369
1370 if (addr + sizeof(val) > vdev->config_len) {
1371 return;
1372 }
1373
1374 stb_p(vdev->config + addr, val);
1375
1376 if (k->set_config) {
1377 k->set_config(vdev, vdev->config);
1378 }
1379 }
1380
1381 void virtio_config_modern_writew(VirtIODevice *vdev,
1382 uint32_t addr, uint32_t data)
1383 {
1384 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1385 uint16_t val = data;
1386
1387 if (addr + sizeof(val) > vdev->config_len) {
1388 return;
1389 }
1390
1391 stw_le_p(vdev->config + addr, val);
1392
1393 if (k->set_config) {
1394 k->set_config(vdev, vdev->config);
1395 }
1396 }
1397
1398 void virtio_config_modern_writel(VirtIODevice *vdev,
1399 uint32_t addr, uint32_t data)
1400 {
1401 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1402 uint32_t val = data;
1403
1404 if (addr + sizeof(val) > vdev->config_len) {
1405 return;
1406 }
1407
1408 stl_le_p(vdev->config + addr, val);
1409
1410 if (k->set_config) {
1411 k->set_config(vdev, vdev->config);
1412 }
1413 }
1414
1415 void virtio_queue_set_addr(VirtIODevice *vdev, int n, hwaddr addr)
1416 {
1417 if (!vdev->vq[n].vring.num) {
1418 return;
1419 }
1420 vdev->vq[n].vring.desc = addr;
1421 virtio_queue_update_rings(vdev, n);
1422 }
1423
1424 hwaddr virtio_queue_get_addr(VirtIODevice *vdev, int n)
1425 {
1426 return vdev->vq[n].vring.desc;
1427 }
1428
1429 void virtio_queue_set_rings(VirtIODevice *vdev, int n, hwaddr desc,
1430 hwaddr avail, hwaddr used)
1431 {
1432 if (!vdev->vq[n].vring.num) {
1433 return;
1434 }
1435 vdev->vq[n].vring.desc = desc;
1436 vdev->vq[n].vring.avail = avail;
1437 vdev->vq[n].vring.used = used;
1438 virtio_init_region_cache(vdev, n);
1439 }
1440
1441 void virtio_queue_set_num(VirtIODevice *vdev, int n, int num)
1442 {
1443 /* Don't allow guest to flip queue between existent and
1444 * nonexistent states, or to set it to an invalid size.
1445 */
1446 if (!!num != !!vdev->vq[n].vring.num ||
1447 num > VIRTQUEUE_MAX_SIZE ||
1448 num < 0) {
1449 return;
1450 }
1451 vdev->vq[n].vring.num = num;
1452 }
1453
1454 VirtQueue *virtio_vector_first_queue(VirtIODevice *vdev, uint16_t vector)
1455 {
1456 return QLIST_FIRST(&vdev->vector_queues[vector]);
1457 }
1458
1459 VirtQueue *virtio_vector_next_queue(VirtQueue *vq)
1460 {
1461 return QLIST_NEXT(vq, node);
1462 }
1463
1464 int virtio_queue_get_num(VirtIODevice *vdev, int n)
1465 {
1466 return vdev->vq[n].vring.num;
1467 }
1468
1469 int virtio_queue_get_max_num(VirtIODevice *vdev, int n)
1470 {
1471 return vdev->vq[n].vring.num_default;
1472 }
1473
1474 int virtio_get_num_queues(VirtIODevice *vdev)
1475 {
1476 int i;
1477
1478 for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
1479 if (!virtio_queue_get_num(vdev, i)) {
1480 break;
1481 }
1482 }
1483
1484 return i;
1485 }
1486
1487 void virtio_queue_set_align(VirtIODevice *vdev, int n, int align)
1488 {
1489 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
1490 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
1491
1492 /* virtio-1 compliant devices cannot change the alignment */
1493 if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
1494 error_report("tried to modify queue alignment for virtio-1 device");
1495 return;
1496 }
1497 /* Check that the transport told us it was going to do this
1498 * (so a buggy transport will immediately assert rather than
1499 * silently failing to migrate this state)
1500 */
1501 assert(k->has_variable_vring_alignment);
1502
1503 if (align) {
1504 vdev->vq[n].vring.align = align;
1505 virtio_queue_update_rings(vdev, n);
1506 }
1507 }
1508
1509 static bool virtio_queue_notify_aio_vq(VirtQueue *vq)
1510 {
1511 if (vq->vring.desc && vq->handle_aio_output) {
1512 VirtIODevice *vdev = vq->vdev;
1513
1514 trace_virtio_queue_notify(vdev, vq - vdev->vq, vq);
1515 return vq->handle_aio_output(vdev, vq);
1516 }
1517
1518 return false;
1519 }
1520
1521 static void virtio_queue_notify_vq(VirtQueue *vq)
1522 {
1523 if (vq->vring.desc && vq->handle_output) {
1524 VirtIODevice *vdev = vq->vdev;
1525
1526 if (unlikely(vdev->broken)) {
1527 return;
1528 }
1529
1530 trace_virtio_queue_notify(vdev, vq - vdev->vq, vq);
1531 vq->handle_output(vdev, vq);
1532 }
1533 }
1534
1535 void virtio_queue_notify(VirtIODevice *vdev, int n)
1536 {
1537 VirtQueue *vq = &vdev->vq[n];
1538
1539 if (unlikely(!vq->vring.desc || vdev->broken)) {
1540 return;
1541 }
1542
1543 trace_virtio_queue_notify(vdev, vq - vdev->vq, vq);
1544 if (vq->handle_aio_output) {
1545 event_notifier_set(&vq->host_notifier);
1546 } else if (vq->handle_output) {
1547 vq->handle_output(vdev, vq);
1548 }
1549 }
1550
1551 uint16_t virtio_queue_vector(VirtIODevice *vdev, int n)
1552 {
1553 return n < VIRTIO_QUEUE_MAX ? vdev->vq[n].vector :
1554 VIRTIO_NO_VECTOR;
1555 }
1556
1557 void virtio_queue_set_vector(VirtIODevice *vdev, int n, uint16_t vector)
1558 {
1559 VirtQueue *vq = &vdev->vq[n];
1560
1561 if (n < VIRTIO_QUEUE_MAX) {
1562 if (vdev->vector_queues &&
1563 vdev->vq[n].vector != VIRTIO_NO_VECTOR) {
1564 QLIST_REMOVE(vq, node);
1565 }
1566 vdev->vq[n].vector = vector;
1567 if (vdev->vector_queues &&
1568 vector != VIRTIO_NO_VECTOR) {
1569 QLIST_INSERT_HEAD(&vdev->vector_queues[vector], vq, node);
1570 }
1571 }
1572 }
1573
1574 VirtQueue *virtio_add_queue(VirtIODevice *vdev, int queue_size,
1575 VirtIOHandleOutput handle_output)
1576 {
1577 int i;
1578
1579 for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
1580 if (vdev->vq[i].vring.num == 0)
1581 break;
1582 }
1583
1584 if (i == VIRTIO_QUEUE_MAX || queue_size > VIRTQUEUE_MAX_SIZE)
1585 abort();
1586
1587 vdev->vq[i].vring.num = queue_size;
1588 vdev->vq[i].vring.num_default = queue_size;
1589 vdev->vq[i].vring.align = VIRTIO_PCI_VRING_ALIGN;
1590 vdev->vq[i].handle_output = handle_output;
1591 vdev->vq[i].handle_aio_output = NULL;
1592
1593 return &vdev->vq[i];
1594 }
1595
1596 void virtio_del_queue(VirtIODevice *vdev, int n)
1597 {
1598 if (n < 0 || n >= VIRTIO_QUEUE_MAX) {
1599 abort();
1600 }
1601
1602 vdev->vq[n].vring.num = 0;
1603 vdev->vq[n].vring.num_default = 0;
1604 }
1605
1606 static void virtio_set_isr(VirtIODevice *vdev, int value)
1607 {
1608 uint8_t old = atomic_read(&vdev->isr);
1609
1610 /* Do not write ISR if it does not change, so that its cacheline remains
1611 * shared in the common case where the guest does not read it.
1612 */
1613 if ((old & value) != value) {
1614 atomic_or(&vdev->isr, value);
1615 }
1616 }
1617
1618 /* Called within rcu_read_lock(). */
1619 static bool virtio_should_notify(VirtIODevice *vdev, VirtQueue *vq)
1620 {
1621 uint16_t old, new;
1622 bool v;
1623 /* We need to expose used array entries before checking used event. */
1624 smp_mb();
1625 /* Always notify when queue is empty (when feature acknowledge) */
1626 if (virtio_vdev_has_feature(vdev, VIRTIO_F_NOTIFY_ON_EMPTY) &&
1627 !vq->inuse && virtio_queue_empty(vq)) {
1628 return true;
1629 }
1630
1631 if (!virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX)) {
1632 return !(vring_avail_flags(vq) & VRING_AVAIL_F_NO_INTERRUPT);
1633 }
1634
1635 v = vq->signalled_used_valid;
1636 vq->signalled_used_valid = true;
1637 old = vq->signalled_used;
1638 new = vq->signalled_used = vq->used_idx;
1639 return !v || vring_need_event(vring_get_used_event(vq), new, old);
1640 }
1641
1642 void virtio_notify_irqfd(VirtIODevice *vdev, VirtQueue *vq)
1643 {
1644 bool should_notify;
1645 rcu_read_lock();
1646 should_notify = virtio_should_notify(vdev, vq);
1647 rcu_read_unlock();
1648
1649 if (!should_notify) {
1650 return;
1651 }
1652
1653 trace_virtio_notify_irqfd(vdev, vq);
1654
1655 /*
1656 * virtio spec 1.0 says ISR bit 0 should be ignored with MSI, but
1657 * windows drivers included in virtio-win 1.8.0 (circa 2015) are
1658 * incorrectly polling this bit during crashdump and hibernation
1659 * in MSI mode, causing a hang if this bit is never updated.
1660 * Recent releases of Windows do not really shut down, but rather
1661 * log out and hibernate to make the next startup faster. Hence,
1662 * this manifested as a more serious hang during shutdown with
1663 *
1664 * Next driver release from 2016 fixed this problem, so working around it
1665 * is not a must, but it's easy to do so let's do it here.
1666 *
1667 * Note: it's safe to update ISR from any thread as it was switched
1668 * to an atomic operation.
1669 */
1670 virtio_set_isr(vq->vdev, 0x1);
1671 event_notifier_set(&vq->guest_notifier);
1672 }
1673
1674 static void virtio_irq(VirtQueue *vq)
1675 {
1676 virtio_set_isr(vq->vdev, 0x1);
1677 virtio_notify_vector(vq->vdev, vq->vector);
1678 }
1679
1680 void virtio_notify(VirtIODevice *vdev, VirtQueue *vq)
1681 {
1682 bool should_notify;
1683 rcu_read_lock();
1684 should_notify = virtio_should_notify(vdev, vq);
1685 rcu_read_unlock();
1686
1687 if (!should_notify) {
1688 return;
1689 }
1690
1691 trace_virtio_notify(vdev, vq);
1692 virtio_irq(vq);
1693 }
1694
1695 void virtio_notify_config(VirtIODevice *vdev)
1696 {
1697 if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK))
1698 return;
1699
1700 virtio_set_isr(vdev, 0x3);
1701 vdev->generation++;
1702 virtio_notify_vector(vdev, vdev->config_vector);
1703 }
1704
1705 static bool virtio_device_endian_needed(void *opaque)
1706 {
1707 VirtIODevice *vdev = opaque;
1708
1709 assert(vdev->device_endian != VIRTIO_DEVICE_ENDIAN_UNKNOWN);
1710 if (!virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
1711 return vdev->device_endian != virtio_default_endian();
1712 }
1713 /* Devices conforming to VIRTIO 1.0 or later are always LE. */
1714 return vdev->device_endian != VIRTIO_DEVICE_ENDIAN_LITTLE;
1715 }
1716
1717 static bool virtio_64bit_features_needed(void *opaque)
1718 {
1719 VirtIODevice *vdev = opaque;
1720
1721 return (vdev->host_features >> 32) != 0;
1722 }
1723
1724 static bool virtio_virtqueue_needed(void *opaque)
1725 {
1726 VirtIODevice *vdev = opaque;
1727
1728 return virtio_host_has_feature(vdev, VIRTIO_F_VERSION_1);
1729 }
1730
1731 static bool virtio_ringsize_needed(void *opaque)
1732 {
1733 VirtIODevice *vdev = opaque;
1734 int i;
1735
1736 for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
1737 if (vdev->vq[i].vring.num != vdev->vq[i].vring.num_default) {
1738 return true;
1739 }
1740 }
1741 return false;
1742 }
1743
1744 static bool virtio_extra_state_needed(void *opaque)
1745 {
1746 VirtIODevice *vdev = opaque;
1747 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
1748 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
1749
1750 return k->has_extra_state &&
1751 k->has_extra_state(qbus->parent);
1752 }
1753
1754 static bool virtio_broken_needed(void *opaque)
1755 {
1756 VirtIODevice *vdev = opaque;
1757
1758 return vdev->broken;
1759 }
1760
1761 static const VMStateDescription vmstate_virtqueue = {
1762 .name = "virtqueue_state",
1763 .version_id = 1,
1764 .minimum_version_id = 1,
1765 .fields = (VMStateField[]) {
1766 VMSTATE_UINT64(vring.avail, struct VirtQueue),
1767 VMSTATE_UINT64(vring.used, struct VirtQueue),
1768 VMSTATE_END_OF_LIST()
1769 }
1770 };
1771
1772 static const VMStateDescription vmstate_virtio_virtqueues = {
1773 .name = "virtio/virtqueues",
1774 .version_id = 1,
1775 .minimum_version_id = 1,
1776 .needed = &virtio_virtqueue_needed,
1777 .fields = (VMStateField[]) {
1778 VMSTATE_STRUCT_VARRAY_POINTER_KNOWN(vq, struct VirtIODevice,
1779 VIRTIO_QUEUE_MAX, 0, vmstate_virtqueue, VirtQueue),
1780 VMSTATE_END_OF_LIST()
1781 }
1782 };
1783
1784 static const VMStateDescription vmstate_ringsize = {
1785 .name = "ringsize_state",
1786 .version_id = 1,
1787 .minimum_version_id = 1,
1788 .fields = (VMStateField[]) {
1789 VMSTATE_UINT32(vring.num_default, struct VirtQueue),
1790 VMSTATE_END_OF_LIST()
1791 }
1792 };
1793
1794 static const VMStateDescription vmstate_virtio_ringsize = {
1795 .name = "virtio/ringsize",
1796 .version_id = 1,
1797 .minimum_version_id = 1,
1798 .needed = &virtio_ringsize_needed,
1799 .fields = (VMStateField[]) {
1800 VMSTATE_STRUCT_VARRAY_POINTER_KNOWN(vq, struct VirtIODevice,
1801 VIRTIO_QUEUE_MAX, 0, vmstate_ringsize, VirtQueue),
1802 VMSTATE_END_OF_LIST()
1803 }
1804 };
1805
1806 static int get_extra_state(QEMUFile *f, void *pv, size_t size,
1807 VMStateField *field)
1808 {
1809 VirtIODevice *vdev = pv;
1810 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
1811 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
1812
1813 if (!k->load_extra_state) {
1814 return -1;
1815 } else {
1816 return k->load_extra_state(qbus->parent, f);
1817 }
1818 }
1819
1820 static int put_extra_state(QEMUFile *f, void *pv, size_t size,
1821 VMStateField *field, QJSON *vmdesc)
1822 {
1823 VirtIODevice *vdev = pv;
1824 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
1825 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
1826
1827 k->save_extra_state(qbus->parent, f);
1828 return 0;
1829 }
1830
1831 static const VMStateInfo vmstate_info_extra_state = {
1832 .name = "virtqueue_extra_state",
1833 .get = get_extra_state,
1834 .put = put_extra_state,
1835 };
1836
1837 static const VMStateDescription vmstate_virtio_extra_state = {
1838 .name = "virtio/extra_state",
1839 .version_id = 1,
1840 .minimum_version_id = 1,
1841 .needed = &virtio_extra_state_needed,
1842 .fields = (VMStateField[]) {
1843 {
1844 .name = "extra_state",
1845 .version_id = 0,
1846 .field_exists = NULL,
1847 .size = 0,
1848 .info = &vmstate_info_extra_state,
1849 .flags = VMS_SINGLE,
1850 .offset = 0,
1851 },
1852 VMSTATE_END_OF_LIST()
1853 }
1854 };
1855
1856 static const VMStateDescription vmstate_virtio_device_endian = {
1857 .name = "virtio/device_endian",
1858 .version_id = 1,
1859 .minimum_version_id = 1,
1860 .needed = &virtio_device_endian_needed,
1861 .fields = (VMStateField[]) {
1862 VMSTATE_UINT8(device_endian, VirtIODevice),
1863 VMSTATE_END_OF_LIST()
1864 }
1865 };
1866
1867 static const VMStateDescription vmstate_virtio_64bit_features = {
1868 .name = "virtio/64bit_features",
1869 .version_id = 1,
1870 .minimum_version_id = 1,
1871 .needed = &virtio_64bit_features_needed,
1872 .fields = (VMStateField[]) {
1873 VMSTATE_UINT64(guest_features, VirtIODevice),
1874 VMSTATE_END_OF_LIST()
1875 }
1876 };
1877
1878 static const VMStateDescription vmstate_virtio_broken = {
1879 .name = "virtio/broken",
1880 .version_id = 1,
1881 .minimum_version_id = 1,
1882 .needed = &virtio_broken_needed,
1883 .fields = (VMStateField[]) {
1884 VMSTATE_BOOL(broken, VirtIODevice),
1885 VMSTATE_END_OF_LIST()
1886 }
1887 };
1888
1889 static const VMStateDescription vmstate_virtio = {
1890 .name = "virtio",
1891 .version_id = 1,
1892 .minimum_version_id = 1,
1893 .minimum_version_id_old = 1,
1894 .fields = (VMStateField[]) {
1895 VMSTATE_END_OF_LIST()
1896 },
1897 .subsections = (const VMStateDescription*[]) {
1898 &vmstate_virtio_device_endian,
1899 &vmstate_virtio_64bit_features,
1900 &vmstate_virtio_virtqueues,
1901 &vmstate_virtio_ringsize,
1902 &vmstate_virtio_broken,
1903 &vmstate_virtio_extra_state,
1904 NULL
1905 }
1906 };
1907
1908 int virtio_save(VirtIODevice *vdev, QEMUFile *f)
1909 {
1910 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
1911 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
1912 VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev);
1913 uint32_t guest_features_lo = (vdev->guest_features & 0xffffffff);
1914 int i;
1915
1916 if (k->save_config) {
1917 k->save_config(qbus->parent, f);
1918 }
1919
1920 qemu_put_8s(f, &vdev->status);
1921 qemu_put_8s(f, &vdev->isr);
1922 qemu_put_be16s(f, &vdev->queue_sel);
1923 qemu_put_be32s(f, &guest_features_lo);
1924 qemu_put_be32(f, vdev->config_len);
1925 qemu_put_buffer(f, vdev->config, vdev->config_len);
1926
1927 for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
1928 if (vdev->vq[i].vring.num == 0)
1929 break;
1930 }
1931
1932 qemu_put_be32(f, i);
1933
1934 for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
1935 if (vdev->vq[i].vring.num == 0)
1936 break;
1937
1938 qemu_put_be32(f, vdev->vq[i].vring.num);
1939 if (k->has_variable_vring_alignment) {
1940 qemu_put_be32(f, vdev->vq[i].vring.align);
1941 }
1942 /*
1943 * Save desc now, the rest of the ring addresses are saved in
1944 * subsections for VIRTIO-1 devices.
1945 */
1946 qemu_put_be64(f, vdev->vq[i].vring.desc);
1947 qemu_put_be16s(f, &vdev->vq[i].last_avail_idx);
1948 if (k->save_queue) {
1949 k->save_queue(qbus->parent, i, f);
1950 }
1951 }
1952
1953 if (vdc->save != NULL) {
1954 vdc->save(vdev, f);
1955 }
1956
1957 if (vdc->vmsd) {
1958 int ret = vmstate_save_state(f, vdc->vmsd, vdev, NULL);
1959 if (ret) {
1960 return ret;
1961 }
1962 }
1963
1964 /* Subsections */
1965 return vmstate_save_state(f, &vmstate_virtio, vdev, NULL);
1966 }
1967
1968 /* A wrapper for use as a VMState .put function */
1969 static int virtio_device_put(QEMUFile *f, void *opaque, size_t size,
1970 VMStateField *field, QJSON *vmdesc)
1971 {
1972 return virtio_save(VIRTIO_DEVICE(opaque), f);
1973 }
1974
1975 /* A wrapper for use as a VMState .get function */
1976 static int virtio_device_get(QEMUFile *f, void *opaque, size_t size,
1977 VMStateField *field)
1978 {
1979 VirtIODevice *vdev = VIRTIO_DEVICE(opaque);
1980 DeviceClass *dc = DEVICE_CLASS(VIRTIO_DEVICE_GET_CLASS(vdev));
1981
1982 return virtio_load(vdev, f, dc->vmsd->version_id);
1983 }
1984
1985 const VMStateInfo virtio_vmstate_info = {
1986 .name = "virtio",
1987 .get = virtio_device_get,
1988 .put = virtio_device_put,
1989 };
1990
1991 static int virtio_set_features_nocheck(VirtIODevice *vdev, uint64_t val)
1992 {
1993 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1994 bool bad = (val & ~(vdev->host_features)) != 0;
1995
1996 val &= vdev->host_features;
1997 if (k->set_features) {
1998 k->set_features(vdev, val);
1999 }
2000 vdev->guest_features = val;
2001 return bad ? -1 : 0;
2002 }
2003
2004 int virtio_set_features(VirtIODevice *vdev, uint64_t val)
2005 {
2006 /*
2007 * The driver must not attempt to set features after feature negotiation
2008 * has finished.
2009 */
2010 if (vdev->status & VIRTIO_CONFIG_S_FEATURES_OK) {
2011 return -EINVAL;
2012 }
2013 return virtio_set_features_nocheck(vdev, val);
2014 }
2015
2016 int virtio_load(VirtIODevice *vdev, QEMUFile *f, int version_id)
2017 {
2018 int i, ret;
2019 int32_t config_len;
2020 uint32_t num;
2021 uint32_t features;
2022 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2023 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
2024 VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev);
2025
2026 /*
2027 * We poison the endianness to ensure it does not get used before
2028 * subsections have been loaded.
2029 */
2030 vdev->device_endian = VIRTIO_DEVICE_ENDIAN_UNKNOWN;
2031
2032 if (k->load_config) {
2033 ret = k->load_config(qbus->parent, f);
2034 if (ret)
2035 return ret;
2036 }
2037
2038 qemu_get_8s(f, &vdev->status);
2039 qemu_get_8s(f, &vdev->isr);
2040 qemu_get_be16s(f, &vdev->queue_sel);
2041 if (vdev->queue_sel >= VIRTIO_QUEUE_MAX) {
2042 return -1;
2043 }
2044 qemu_get_be32s(f, &features);
2045
2046 /*
2047 * Temporarily set guest_features low bits - needed by
2048 * virtio net load code testing for VIRTIO_NET_F_CTRL_GUEST_OFFLOADS
2049 * VIRTIO_NET_F_GUEST_ANNOUNCE and VIRTIO_NET_F_CTRL_VQ.
2050 *
2051 * Note: devices should always test host features in future - don't create
2052 * new dependencies like this.
2053 */
2054 vdev->guest_features = features;
2055
2056 config_len = qemu_get_be32(f);
2057
2058 /*
2059 * There are cases where the incoming config can be bigger or smaller
2060 * than what we have; so load what we have space for, and skip
2061 * any excess that's in the stream.
2062 */
2063 qemu_get_buffer(f, vdev->config, MIN(config_len, vdev->config_len));
2064
2065 while (config_len > vdev->config_len) {
2066 qemu_get_byte(f);
2067 config_len--;
2068 }
2069
2070 num = qemu_get_be32(f);
2071
2072 if (num > VIRTIO_QUEUE_MAX) {
2073 error_report("Invalid number of virtqueues: 0x%x", num);
2074 return -1;
2075 }
2076
2077 for (i = 0; i < num; i++) {
2078 vdev->vq[i].vring.num = qemu_get_be32(f);
2079 if (k->has_variable_vring_alignment) {
2080 vdev->vq[i].vring.align = qemu_get_be32(f);
2081 }
2082 vdev->vq[i].vring.desc = qemu_get_be64(f);
2083 qemu_get_be16s(f, &vdev->vq[i].last_avail_idx);
2084 vdev->vq[i].signalled_used_valid = false;
2085 vdev->vq[i].notification = true;
2086
2087 if (!vdev->vq[i].vring.desc && vdev->vq[i].last_avail_idx) {
2088 error_report("VQ %d address 0x0 "
2089 "inconsistent with Host index 0x%x",
2090 i, vdev->vq[i].last_avail_idx);
2091 return -1;
2092 }
2093 if (k->load_queue) {
2094 ret = k->load_queue(qbus->parent, i, f);
2095 if (ret)
2096 return ret;
2097 }
2098 }
2099
2100 virtio_notify_vector(vdev, VIRTIO_NO_VECTOR);
2101
2102 if (vdc->load != NULL) {
2103 ret = vdc->load(vdev, f, version_id);
2104 if (ret) {
2105 return ret;
2106 }
2107 }
2108
2109 if (vdc->vmsd) {
2110 ret = vmstate_load_state(f, vdc->vmsd, vdev, version_id);
2111 if (ret) {
2112 return ret;
2113 }
2114 }
2115
2116 /* Subsections */
2117 ret = vmstate_load_state(f, &vmstate_virtio, vdev, 1);
2118 if (ret) {
2119 return ret;
2120 }
2121
2122 if (vdev->device_endian == VIRTIO_DEVICE_ENDIAN_UNKNOWN) {
2123 vdev->device_endian = virtio_default_endian();
2124 }
2125
2126 if (virtio_64bit_features_needed(vdev)) {
2127 /*
2128 * Subsection load filled vdev->guest_features. Run them
2129 * through virtio_set_features to sanity-check them against
2130 * host_features.
2131 */
2132 uint64_t features64 = vdev->guest_features;
2133 if (virtio_set_features_nocheck(vdev, features64) < 0) {
2134 error_report("Features 0x%" PRIx64 " unsupported. "
2135 "Allowed features: 0x%" PRIx64,
2136 features64, vdev->host_features);
2137 return -1;
2138 }
2139 } else {
2140 if (virtio_set_features_nocheck(vdev, features) < 0) {
2141 error_report("Features 0x%x unsupported. "
2142 "Allowed features: 0x%" PRIx64,
2143 features, vdev->host_features);
2144 return -1;
2145 }
2146 }
2147
2148 rcu_read_lock();
2149 for (i = 0; i < num; i++) {
2150 if (vdev->vq[i].vring.desc) {
2151 uint16_t nheads;
2152
2153 /*
2154 * VIRTIO-1 devices migrate desc, used, and avail ring addresses so
2155 * only the region cache needs to be set up. Legacy devices need
2156 * to calculate used and avail ring addresses based on the desc
2157 * address.
2158 */
2159 if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
2160 virtio_init_region_cache(vdev, i);
2161 } else {
2162 virtio_queue_update_rings(vdev, i);
2163 }
2164
2165 nheads = vring_avail_idx(&vdev->vq[i]) - vdev->vq[i].last_avail_idx;
2166 /* Check it isn't doing strange things with descriptor numbers. */
2167 if (nheads > vdev->vq[i].vring.num) {
2168 error_report("VQ %d size 0x%x Guest index 0x%x "
2169 "inconsistent with Host index 0x%x: delta 0x%x",
2170 i, vdev->vq[i].vring.num,
2171 vring_avail_idx(&vdev->vq[i]),
2172 vdev->vq[i].last_avail_idx, nheads);
2173 return -1;
2174 }
2175 vdev->vq[i].used_idx = vring_used_idx(&vdev->vq[i]);
2176 vdev->vq[i].shadow_avail_idx = vring_avail_idx(&vdev->vq[i]);
2177
2178 /*
2179 * Some devices migrate VirtQueueElements that have been popped
2180 * from the avail ring but not yet returned to the used ring.
2181 * Since max ring size < UINT16_MAX it's safe to use modulo
2182 * UINT16_MAX + 1 subtraction.
2183 */
2184 vdev->vq[i].inuse = (uint16_t)(vdev->vq[i].last_avail_idx -
2185 vdev->vq[i].used_idx);
2186 if (vdev->vq[i].inuse > vdev->vq[i].vring.num) {
2187 error_report("VQ %d size 0x%x < last_avail_idx 0x%x - "
2188 "used_idx 0x%x",
2189 i, vdev->vq[i].vring.num,
2190 vdev->vq[i].last_avail_idx,
2191 vdev->vq[i].used_idx);
2192 return -1;
2193 }
2194 }
2195 }
2196 rcu_read_unlock();
2197
2198 return 0;
2199 }
2200
2201 void virtio_cleanup(VirtIODevice *vdev)
2202 {
2203 qemu_del_vm_change_state_handler(vdev->vmstate);
2204 }
2205
2206 static void virtio_vmstate_change(void *opaque, int running, RunState state)
2207 {
2208 VirtIODevice *vdev = opaque;
2209 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2210 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
2211 bool backend_run = running && (vdev->status & VIRTIO_CONFIG_S_DRIVER_OK);
2212 vdev->vm_running = running;
2213
2214 if (backend_run) {
2215 virtio_set_status(vdev, vdev->status);
2216 }
2217
2218 if (k->vmstate_change) {
2219 k->vmstate_change(qbus->parent, backend_run);
2220 }
2221
2222 if (!backend_run) {
2223 virtio_set_status(vdev, vdev->status);
2224 }
2225 }
2226
2227 void virtio_instance_init_common(Object *proxy_obj, void *data,
2228 size_t vdev_size, const char *vdev_name)
2229 {
2230 DeviceState *vdev = data;
2231
2232 object_initialize(vdev, vdev_size, vdev_name);
2233 object_property_add_child(proxy_obj, "virtio-backend", OBJECT(vdev), NULL);
2234 object_unref(OBJECT(vdev));
2235 qdev_alias_all_properties(vdev, proxy_obj);
2236 }
2237
2238 void virtio_init(VirtIODevice *vdev, const char *name,
2239 uint16_t device_id, size_t config_size)
2240 {
2241 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2242 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
2243 int i;
2244 int nvectors = k->query_nvectors ? k->query_nvectors(qbus->parent) : 0;
2245
2246 if (nvectors) {
2247 vdev->vector_queues =
2248 g_malloc0(sizeof(*vdev->vector_queues) * nvectors);
2249 }
2250
2251 vdev->device_id = device_id;
2252 vdev->status = 0;
2253 atomic_set(&vdev->isr, 0);
2254 vdev->queue_sel = 0;
2255 vdev->config_vector = VIRTIO_NO_VECTOR;
2256 vdev->vq = g_malloc0(sizeof(VirtQueue) * VIRTIO_QUEUE_MAX);
2257 vdev->vm_running = runstate_is_running();
2258 vdev->broken = false;
2259 for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2260 vdev->vq[i].vector = VIRTIO_NO_VECTOR;
2261 vdev->vq[i].vdev = vdev;
2262 vdev->vq[i].queue_index = i;
2263 }
2264
2265 vdev->name = name;
2266 vdev->config_len = config_size;
2267 if (vdev->config_len) {
2268 vdev->config = g_malloc0(config_size);
2269 } else {
2270 vdev->config = NULL;
2271 }
2272 vdev->vmstate = qemu_add_vm_change_state_handler(virtio_vmstate_change,
2273 vdev);
2274 vdev->device_endian = virtio_default_endian();
2275 vdev->use_guest_notifier_mask = true;
2276 }
2277
2278 hwaddr virtio_queue_get_desc_addr(VirtIODevice *vdev, int n)
2279 {
2280 return vdev->vq[n].vring.desc;
2281 }
2282
2283 hwaddr virtio_queue_get_avail_addr(VirtIODevice *vdev, int n)
2284 {
2285 return vdev->vq[n].vring.avail;
2286 }
2287
2288 hwaddr virtio_queue_get_used_addr(VirtIODevice *vdev, int n)
2289 {
2290 return vdev->vq[n].vring.used;
2291 }
2292
2293 hwaddr virtio_queue_get_desc_size(VirtIODevice *vdev, int n)
2294 {
2295 return sizeof(VRingDesc) * vdev->vq[n].vring.num;
2296 }
2297
2298 hwaddr virtio_queue_get_avail_size(VirtIODevice *vdev, int n)
2299 {
2300 return offsetof(VRingAvail, ring) +
2301 sizeof(uint16_t) * vdev->vq[n].vring.num;
2302 }
2303
2304 hwaddr virtio_queue_get_used_size(VirtIODevice *vdev, int n)
2305 {
2306 return offsetof(VRingUsed, ring) +
2307 sizeof(VRingUsedElem) * vdev->vq[n].vring.num;
2308 }
2309
2310 uint16_t virtio_queue_get_last_avail_idx(VirtIODevice *vdev, int n)
2311 {
2312 return vdev->vq[n].last_avail_idx;
2313 }
2314
2315 void virtio_queue_set_last_avail_idx(VirtIODevice *vdev, int n, uint16_t idx)
2316 {
2317 vdev->vq[n].last_avail_idx = idx;
2318 vdev->vq[n].shadow_avail_idx = idx;
2319 }
2320
2321 void virtio_queue_restore_last_avail_idx(VirtIODevice *vdev, int n)
2322 {
2323 rcu_read_lock();
2324 if (vdev->vq[n].vring.desc) {
2325 vdev->vq[n].last_avail_idx = vring_used_idx(&vdev->vq[n]);
2326 vdev->vq[n].shadow_avail_idx = vdev->vq[n].last_avail_idx;
2327 }
2328 rcu_read_unlock();
2329 }
2330
2331 void virtio_queue_update_used_idx(VirtIODevice *vdev, int n)
2332 {
2333 rcu_read_lock();
2334 if (vdev->vq[n].vring.desc) {
2335 vdev->vq[n].used_idx = vring_used_idx(&vdev->vq[n]);
2336 }
2337 rcu_read_unlock();
2338 }
2339
2340 void virtio_queue_invalidate_signalled_used(VirtIODevice *vdev, int n)
2341 {
2342 vdev->vq[n].signalled_used_valid = false;
2343 }
2344
2345 VirtQueue *virtio_get_queue(VirtIODevice *vdev, int n)
2346 {
2347 return vdev->vq + n;
2348 }
2349
2350 uint16_t virtio_get_queue_index(VirtQueue *vq)
2351 {
2352 return vq->queue_index;
2353 }
2354
2355 static void virtio_queue_guest_notifier_read(EventNotifier *n)
2356 {
2357 VirtQueue *vq = container_of(n, VirtQueue, guest_notifier);
2358 if (event_notifier_test_and_clear(n)) {
2359 virtio_irq(vq);
2360 }
2361 }
2362
2363 void virtio_queue_set_guest_notifier_fd_handler(VirtQueue *vq, bool assign,
2364 bool with_irqfd)
2365 {
2366 if (assign && !with_irqfd) {
2367 event_notifier_set_handler(&vq->guest_notifier,
2368 virtio_queue_guest_notifier_read);
2369 } else {
2370 event_notifier_set_handler(&vq->guest_notifier, NULL);
2371 }
2372 if (!assign) {
2373 /* Test and clear notifier before closing it,
2374 * in case poll callback didn't have time to run. */
2375 virtio_queue_guest_notifier_read(&vq->guest_notifier);
2376 }
2377 }
2378
2379 EventNotifier *virtio_queue_get_guest_notifier(VirtQueue *vq)
2380 {
2381 return &vq->guest_notifier;
2382 }
2383
2384 static void virtio_queue_host_notifier_aio_read(EventNotifier *n)
2385 {
2386 VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
2387 if (event_notifier_test_and_clear(n)) {
2388 virtio_queue_notify_aio_vq(vq);
2389 }
2390 }
2391
2392 static void virtio_queue_host_notifier_aio_poll_begin(EventNotifier *n)
2393 {
2394 VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
2395
2396 virtio_queue_set_notification(vq, 0);
2397 }
2398
2399 static bool virtio_queue_host_notifier_aio_poll(void *opaque)
2400 {
2401 EventNotifier *n = opaque;
2402 VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
2403 bool progress;
2404
2405 if (!vq->vring.desc || virtio_queue_empty(vq)) {
2406 return false;
2407 }
2408
2409 progress = virtio_queue_notify_aio_vq(vq);
2410
2411 /* In case the handler function re-enabled notifications */
2412 virtio_queue_set_notification(vq, 0);
2413 return progress;
2414 }
2415
2416 static void virtio_queue_host_notifier_aio_poll_end(EventNotifier *n)
2417 {
2418 VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
2419
2420 /* Caller polls once more after this to catch requests that race with us */
2421 virtio_queue_set_notification(vq, 1);
2422 }
2423
2424 void virtio_queue_aio_set_host_notifier_handler(VirtQueue *vq, AioContext *ctx,
2425 VirtIOHandleAIOOutput handle_output)
2426 {
2427 if (handle_output) {
2428 vq->handle_aio_output = handle_output;
2429 aio_set_event_notifier(ctx, &vq->host_notifier, true,
2430 virtio_queue_host_notifier_aio_read,
2431 virtio_queue_host_notifier_aio_poll);
2432 aio_set_event_notifier_poll(ctx, &vq->host_notifier,
2433 virtio_queue_host_notifier_aio_poll_begin,
2434 virtio_queue_host_notifier_aio_poll_end);
2435 } else {
2436 aio_set_event_notifier(ctx, &vq->host_notifier, true, NULL, NULL);
2437 /* Test and clear notifier before after disabling event,
2438 * in case poll callback didn't have time to run. */
2439 virtio_queue_host_notifier_aio_read(&vq->host_notifier);
2440 vq->handle_aio_output = NULL;
2441 }
2442 }
2443
2444 void virtio_queue_host_notifier_read(EventNotifier *n)
2445 {
2446 VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
2447 if (event_notifier_test_and_clear(n)) {
2448 virtio_queue_notify_vq(vq);
2449 }
2450 }
2451
2452 EventNotifier *virtio_queue_get_host_notifier(VirtQueue *vq)
2453 {
2454 return &vq->host_notifier;
2455 }
2456
2457 void virtio_device_set_child_bus_name(VirtIODevice *vdev, char *bus_name)
2458 {
2459 g_free(vdev->bus_name);
2460 vdev->bus_name = g_strdup(bus_name);
2461 }
2462
2463 void GCC_FMT_ATTR(2, 3) virtio_error(VirtIODevice *vdev, const char *fmt, ...)
2464 {
2465 va_list ap;
2466
2467 va_start(ap, fmt);
2468 error_vreport(fmt, ap);
2469 va_end(ap);
2470
2471 if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
2472 virtio_set_status(vdev, vdev->status | VIRTIO_CONFIG_S_NEEDS_RESET);
2473 virtio_notify_config(vdev);
2474 }
2475
2476 vdev->broken = true;
2477 }
2478
2479 static void virtio_memory_listener_commit(MemoryListener *listener)
2480 {
2481 VirtIODevice *vdev = container_of(listener, VirtIODevice, listener);
2482 int i;
2483
2484 for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2485 if (vdev->vq[i].vring.num == 0) {
2486 break;
2487 }
2488 virtio_init_region_cache(vdev, i);
2489 }
2490 }
2491
2492 static void virtio_device_realize(DeviceState *dev, Error **errp)
2493 {
2494 VirtIODevice *vdev = VIRTIO_DEVICE(dev);
2495 VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(dev);
2496 Error *err = NULL;
2497
2498 /* Devices should either use vmsd or the load/save methods */
2499 assert(!vdc->vmsd || !vdc->load);
2500
2501 if (vdc->realize != NULL) {
2502 vdc->realize(dev, &err);
2503 if (err != NULL) {
2504 error_propagate(errp, err);
2505 return;
2506 }
2507 }
2508
2509 virtio_bus_device_plugged(vdev, &err);
2510 if (err != NULL) {
2511 error_propagate(errp, err);
2512 vdc->unrealize(dev, NULL);
2513 return;
2514 }
2515
2516 vdev->listener.commit = virtio_memory_listener_commit;
2517 memory_listener_register(&vdev->listener, vdev->dma_as);
2518 }
2519
2520 static void virtio_device_unrealize(DeviceState *dev, Error **errp)
2521 {
2522 VirtIODevice *vdev = VIRTIO_DEVICE(dev);
2523 VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(dev);
2524 Error *err = NULL;
2525
2526 virtio_bus_device_unplugged(vdev);
2527
2528 if (vdc->unrealize != NULL) {
2529 vdc->unrealize(dev, &err);
2530 if (err != NULL) {
2531 error_propagate(errp, err);
2532 return;
2533 }
2534 }
2535
2536 g_free(vdev->bus_name);
2537 vdev->bus_name = NULL;
2538 }
2539
2540 static void virtio_device_free_virtqueues(VirtIODevice *vdev)
2541 {
2542 int i;
2543 if (!vdev->vq) {
2544 return;
2545 }
2546
2547 for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2548 if (vdev->vq[i].vring.num == 0) {
2549 break;
2550 }
2551 virtio_virtqueue_reset_region_cache(&vdev->vq[i]);
2552 }
2553 g_free(vdev->vq);
2554 }
2555
2556 static void virtio_device_instance_finalize(Object *obj)
2557 {
2558 VirtIODevice *vdev = VIRTIO_DEVICE(obj);
2559
2560 memory_listener_unregister(&vdev->listener);
2561 virtio_device_free_virtqueues(vdev);
2562
2563 g_free(vdev->config);
2564 g_free(vdev->vector_queues);
2565 }
2566
2567 static Property virtio_properties[] = {
2568 DEFINE_VIRTIO_COMMON_FEATURES(VirtIODevice, host_features),
2569 DEFINE_PROP_END_OF_LIST(),
2570 };
2571
2572 static int virtio_device_start_ioeventfd_impl(VirtIODevice *vdev)
2573 {
2574 VirtioBusState *qbus = VIRTIO_BUS(qdev_get_parent_bus(DEVICE(vdev)));
2575 int n, r, err;
2576
2577 for (n = 0; n < VIRTIO_QUEUE_MAX; n++) {
2578 VirtQueue *vq = &vdev->vq[n];
2579 if (!virtio_queue_get_num(vdev, n)) {
2580 continue;
2581 }
2582 r = virtio_bus_set_host_notifier(qbus, n, true);
2583 if (r < 0) {
2584 err = r;
2585 goto assign_error;
2586 }
2587 event_notifier_set_handler(&vq->host_notifier,
2588 virtio_queue_host_notifier_read);
2589 }
2590
2591 for (n = 0; n < VIRTIO_QUEUE_MAX; n++) {
2592 /* Kick right away to begin processing requests already in vring */
2593 VirtQueue *vq = &vdev->vq[n];
2594 if (!vq->vring.num) {
2595 continue;
2596 }
2597 event_notifier_set(&vq->host_notifier);
2598 }
2599 return 0;
2600
2601 assign_error:
2602 while (--n >= 0) {
2603 VirtQueue *vq = &vdev->vq[n];
2604 if (!virtio_queue_get_num(vdev, n)) {
2605 continue;
2606 }
2607
2608 event_notifier_set_handler(&vq->host_notifier, NULL);
2609 r = virtio_bus_set_host_notifier(qbus, n, false);
2610 assert(r >= 0);
2611 }
2612 return err;
2613 }
2614
2615 int virtio_device_start_ioeventfd(VirtIODevice *vdev)
2616 {
2617 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2618 VirtioBusState *vbus = VIRTIO_BUS(qbus);
2619
2620 return virtio_bus_start_ioeventfd(vbus);
2621 }
2622
2623 static void virtio_device_stop_ioeventfd_impl(VirtIODevice *vdev)
2624 {
2625 VirtioBusState *qbus = VIRTIO_BUS(qdev_get_parent_bus(DEVICE(vdev)));
2626 int n, r;
2627
2628 for (n = 0; n < VIRTIO_QUEUE_MAX; n++) {
2629 VirtQueue *vq = &vdev->vq[n];
2630
2631 if (!virtio_queue_get_num(vdev, n)) {
2632 continue;
2633 }
2634 event_notifier_set_handler(&vq->host_notifier, NULL);
2635 r = virtio_bus_set_host_notifier(qbus, n, false);
2636 assert(r >= 0);
2637 }
2638 }
2639
2640 void virtio_device_stop_ioeventfd(VirtIODevice *vdev)
2641 {
2642 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2643 VirtioBusState *vbus = VIRTIO_BUS(qbus);
2644
2645 virtio_bus_stop_ioeventfd(vbus);
2646 }
2647
2648 int virtio_device_grab_ioeventfd(VirtIODevice *vdev)
2649 {
2650 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2651 VirtioBusState *vbus = VIRTIO_BUS(qbus);
2652
2653 return virtio_bus_grab_ioeventfd(vbus);
2654 }
2655
2656 void virtio_device_release_ioeventfd(VirtIODevice *vdev)
2657 {
2658 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2659 VirtioBusState *vbus = VIRTIO_BUS(qbus);
2660
2661 virtio_bus_release_ioeventfd(vbus);
2662 }
2663
2664 static void virtio_device_class_init(ObjectClass *klass, void *data)
2665 {
2666 /* Set the default value here. */
2667 VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
2668 DeviceClass *dc = DEVICE_CLASS(klass);
2669
2670 dc->realize = virtio_device_realize;
2671 dc->unrealize = virtio_device_unrealize;
2672 dc->bus_type = TYPE_VIRTIO_BUS;
2673 dc->props = virtio_properties;
2674 vdc->start_ioeventfd = virtio_device_start_ioeventfd_impl;
2675 vdc->stop_ioeventfd = virtio_device_stop_ioeventfd_impl;
2676
2677 vdc->legacy_features |= VIRTIO_LEGACY_FEATURES;
2678 }
2679
2680 bool virtio_device_ioeventfd_enabled(VirtIODevice *vdev)
2681 {
2682 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2683 VirtioBusState *vbus = VIRTIO_BUS(qbus);
2684
2685 return virtio_bus_ioeventfd_enabled(vbus);
2686 }
2687
2688 static const TypeInfo virtio_device_info = {
2689 .name = TYPE_VIRTIO_DEVICE,
2690 .parent = TYPE_DEVICE,
2691 .instance_size = sizeof(VirtIODevice),
2692 .class_init = virtio_device_class_init,
2693 .instance_finalize = virtio_device_instance_finalize,
2694 .abstract = true,
2695 .class_size = sizeof(VirtioDeviceClass),
2696 };
2697
2698 static void virtio_register_types(void)
2699 {
2700 type_register_static(&virtio_device_info);
2701 }
2702
2703 type_init(virtio_register_types)