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