memory: remove assertion on memory_region_destroy
[qemu/ar7.git] / hw / virtio / virtio.c
blob013979a6b82fd0501f274bcc9fd6be86a5b96b05
1 /*
2 * Virtio Support
4 * Copyright IBM, Corp. 2007
6 * Authors:
7 * Anthony Liguori <aliguori@us.ibm.com>
9 * This work is licensed under the terms of the GNU GPL, version 2. See
10 * the COPYING file in the top-level directory.
14 #include <inttypes.h>
16 #include "trace.h"
17 #include "exec/address-spaces.h"
18 #include "qemu/error-report.h"
19 #include "hw/virtio/virtio.h"
20 #include "qemu/atomic.h"
21 #include "hw/virtio/virtio-bus.h"
22 #include "migration/migration.h"
23 #include "hw/virtio/virtio-access.h"
26 * The alignment to use between consumer and producer parts of vring.
27 * x86 pagesize again. This is the default, used by transports like PCI
28 * which don't provide a means for the guest to tell the host the alignment.
30 #define VIRTIO_PCI_VRING_ALIGN 4096
32 typedef struct VRingDesc
34 uint64_t addr;
35 uint32_t len;
36 uint16_t flags;
37 uint16_t next;
38 } VRingDesc;
40 typedef struct VRingAvail
42 uint16_t flags;
43 uint16_t idx;
44 uint16_t ring[0];
45 } VRingAvail;
47 typedef struct VRingUsedElem
49 uint32_t id;
50 uint32_t len;
51 } VRingUsedElem;
53 typedef struct VRingUsed
55 uint16_t flags;
56 uint16_t idx;
57 VRingUsedElem ring[0];
58 } VRingUsed;
60 typedef struct VRing
62 unsigned int num;
63 unsigned int align;
64 hwaddr desc;
65 hwaddr avail;
66 hwaddr used;
67 } VRing;
69 struct VirtQueue
71 VRing vring;
72 hwaddr pa;
73 uint16_t last_avail_idx;
74 /* Last used index value we have signalled on */
75 uint16_t signalled_used;
77 /* Last used index value we have signalled on */
78 bool signalled_used_valid;
80 /* Notification enabled? */
81 bool notification;
83 uint16_t queue_index;
85 int inuse;
87 uint16_t vector;
88 void (*handle_output)(VirtIODevice *vdev, VirtQueue *vq);
89 VirtIODevice *vdev;
90 EventNotifier guest_notifier;
91 EventNotifier host_notifier;
94 /* virt queue functions */
95 static void virtqueue_init(VirtQueue *vq)
97 hwaddr pa = vq->pa;
99 vq->vring.desc = pa;
100 vq->vring.avail = pa + vq->vring.num * sizeof(VRingDesc);
101 vq->vring.used = vring_align(vq->vring.avail +
102 offsetof(VRingAvail, ring[vq->vring.num]),
103 vq->vring.align);
106 static inline uint64_t vring_desc_addr(VirtIODevice *vdev, hwaddr desc_pa,
107 int i)
109 hwaddr pa;
110 pa = desc_pa + sizeof(VRingDesc) * i + offsetof(VRingDesc, addr);
111 return virtio_ldq_phys(vdev, pa);
114 static inline uint32_t vring_desc_len(VirtIODevice *vdev, hwaddr desc_pa, int i)
116 hwaddr pa;
117 pa = desc_pa + sizeof(VRingDesc) * i + offsetof(VRingDesc, len);
118 return virtio_ldl_phys(vdev, pa);
121 static inline uint16_t vring_desc_flags(VirtIODevice *vdev, hwaddr desc_pa,
122 int i)
124 hwaddr pa;
125 pa = desc_pa + sizeof(VRingDesc) * i + offsetof(VRingDesc, flags);
126 return virtio_lduw_phys(vdev, pa);
129 static inline uint16_t vring_desc_next(VirtIODevice *vdev, hwaddr desc_pa,
130 int i)
132 hwaddr pa;
133 pa = desc_pa + sizeof(VRingDesc) * i + offsetof(VRingDesc, next);
134 return virtio_lduw_phys(vdev, pa);
137 static inline uint16_t vring_avail_flags(VirtQueue *vq)
139 hwaddr pa;
140 pa = vq->vring.avail + offsetof(VRingAvail, flags);
141 return virtio_lduw_phys(vq->vdev, pa);
144 static inline uint16_t vring_avail_idx(VirtQueue *vq)
146 hwaddr pa;
147 pa = vq->vring.avail + offsetof(VRingAvail, idx);
148 return virtio_lduw_phys(vq->vdev, pa);
151 static inline uint16_t vring_avail_ring(VirtQueue *vq, int i)
153 hwaddr pa;
154 pa = vq->vring.avail + offsetof(VRingAvail, ring[i]);
155 return virtio_lduw_phys(vq->vdev, pa);
158 static inline uint16_t vring_used_event(VirtQueue *vq)
160 return vring_avail_ring(vq, vq->vring.num);
163 static inline void vring_used_ring_id(VirtQueue *vq, int i, uint32_t val)
165 hwaddr pa;
166 pa = vq->vring.used + offsetof(VRingUsed, ring[i].id);
167 virtio_stl_phys(vq->vdev, pa, val);
170 static inline void vring_used_ring_len(VirtQueue *vq, int i, uint32_t val)
172 hwaddr pa;
173 pa = vq->vring.used + offsetof(VRingUsed, ring[i].len);
174 virtio_stl_phys(vq->vdev, pa, val);
177 static uint16_t vring_used_idx(VirtQueue *vq)
179 hwaddr pa;
180 pa = vq->vring.used + offsetof(VRingUsed, idx);
181 return virtio_lduw_phys(vq->vdev, pa);
184 static inline void vring_used_idx_set(VirtQueue *vq, uint16_t val)
186 hwaddr pa;
187 pa = vq->vring.used + offsetof(VRingUsed, idx);
188 virtio_stw_phys(vq->vdev, pa, val);
191 static inline void vring_used_flags_set_bit(VirtQueue *vq, int mask)
193 VirtIODevice *vdev = vq->vdev;
194 hwaddr pa;
195 pa = vq->vring.used + offsetof(VRingUsed, flags);
196 virtio_stw_phys(vdev, pa, virtio_lduw_phys(vdev, pa) | mask);
199 static inline void vring_used_flags_unset_bit(VirtQueue *vq, int mask)
201 VirtIODevice *vdev = vq->vdev;
202 hwaddr pa;
203 pa = vq->vring.used + offsetof(VRingUsed, flags);
204 virtio_stw_phys(vdev, pa, virtio_lduw_phys(vdev, pa) & ~mask);
207 static inline void vring_avail_event(VirtQueue *vq, uint16_t val)
209 hwaddr pa;
210 if (!vq->notification) {
211 return;
213 pa = vq->vring.used + offsetof(VRingUsed, ring[vq->vring.num]);
214 virtio_stw_phys(vq->vdev, pa, val);
217 void virtio_queue_set_notification(VirtQueue *vq, int enable)
219 vq->notification = enable;
220 if (vq->vdev->guest_features & (1 << VIRTIO_RING_F_EVENT_IDX)) {
221 vring_avail_event(vq, vring_avail_idx(vq));
222 } else if (enable) {
223 vring_used_flags_unset_bit(vq, VRING_USED_F_NO_NOTIFY);
224 } else {
225 vring_used_flags_set_bit(vq, VRING_USED_F_NO_NOTIFY);
227 if (enable) {
228 /* Expose avail event/used flags before caller checks the avail idx. */
229 smp_mb();
233 int virtio_queue_ready(VirtQueue *vq)
235 return vq->vring.avail != 0;
238 int virtio_queue_empty(VirtQueue *vq)
240 return vring_avail_idx(vq) == vq->last_avail_idx;
243 void virtqueue_fill(VirtQueue *vq, const VirtQueueElement *elem,
244 unsigned int len, unsigned int idx)
246 unsigned int offset;
247 int i;
249 trace_virtqueue_fill(vq, elem, len, idx);
251 offset = 0;
252 for (i = 0; i < elem->in_num; i++) {
253 size_t size = MIN(len - offset, elem->in_sg[i].iov_len);
255 cpu_physical_memory_unmap(elem->in_sg[i].iov_base,
256 elem->in_sg[i].iov_len,
257 1, size);
259 offset += size;
262 for (i = 0; i < elem->out_num; i++)
263 cpu_physical_memory_unmap(elem->out_sg[i].iov_base,
264 elem->out_sg[i].iov_len,
265 0, elem->out_sg[i].iov_len);
267 idx = (idx + vring_used_idx(vq)) % vq->vring.num;
269 /* Get a pointer to the next entry in the used ring. */
270 vring_used_ring_id(vq, idx, elem->index);
271 vring_used_ring_len(vq, idx, len);
274 void virtqueue_flush(VirtQueue *vq, unsigned int count)
276 uint16_t old, new;
277 /* Make sure buffer is written before we update index. */
278 smp_wmb();
279 trace_virtqueue_flush(vq, count);
280 old = vring_used_idx(vq);
281 new = old + count;
282 vring_used_idx_set(vq, new);
283 vq->inuse -= count;
284 if (unlikely((int16_t)(new - vq->signalled_used) < (uint16_t)(new - old)))
285 vq->signalled_used_valid = false;
288 void virtqueue_push(VirtQueue *vq, const VirtQueueElement *elem,
289 unsigned int len)
291 virtqueue_fill(vq, elem, len, 0);
292 virtqueue_flush(vq, 1);
295 static int virtqueue_num_heads(VirtQueue *vq, unsigned int idx)
297 uint16_t num_heads = vring_avail_idx(vq) - idx;
299 /* Check it isn't doing very strange things with descriptor numbers. */
300 if (num_heads > vq->vring.num) {
301 error_report("Guest moved used index from %u to %u",
302 idx, vring_avail_idx(vq));
303 exit(1);
305 /* On success, callers read a descriptor at vq->last_avail_idx.
306 * Make sure descriptor read does not bypass avail index read. */
307 if (num_heads) {
308 smp_rmb();
311 return num_heads;
314 static unsigned int virtqueue_get_head(VirtQueue *vq, unsigned int idx)
316 unsigned int head;
318 /* Grab the next descriptor number they're advertising, and increment
319 * the index we've seen. */
320 head = vring_avail_ring(vq, idx % vq->vring.num);
322 /* If their number is silly, that's a fatal mistake. */
323 if (head >= vq->vring.num) {
324 error_report("Guest says index %u is available", head);
325 exit(1);
328 return head;
331 static unsigned virtqueue_next_desc(VirtIODevice *vdev, hwaddr desc_pa,
332 unsigned int i, unsigned int max)
334 unsigned int next;
336 /* If this descriptor says it doesn't chain, we're done. */
337 if (!(vring_desc_flags(vdev, desc_pa, i) & VRING_DESC_F_NEXT)) {
338 return max;
341 /* Check they're not leading us off end of descriptors. */
342 next = vring_desc_next(vdev, desc_pa, i);
343 /* Make sure compiler knows to grab that: we don't want it changing! */
344 smp_wmb();
346 if (next >= max) {
347 error_report("Desc next is %u", next);
348 exit(1);
351 return next;
354 void virtqueue_get_avail_bytes(VirtQueue *vq, unsigned int *in_bytes,
355 unsigned int *out_bytes,
356 unsigned max_in_bytes, unsigned max_out_bytes)
358 unsigned int idx;
359 unsigned int total_bufs, in_total, out_total;
361 idx = vq->last_avail_idx;
363 total_bufs = in_total = out_total = 0;
364 while (virtqueue_num_heads(vq, idx)) {
365 VirtIODevice *vdev = vq->vdev;
366 unsigned int max, num_bufs, indirect = 0;
367 hwaddr desc_pa;
368 int i;
370 max = vq->vring.num;
371 num_bufs = total_bufs;
372 i = virtqueue_get_head(vq, idx++);
373 desc_pa = vq->vring.desc;
375 if (vring_desc_flags(vdev, desc_pa, i) & VRING_DESC_F_INDIRECT) {
376 if (vring_desc_len(vdev, desc_pa, i) % sizeof(VRingDesc)) {
377 error_report("Invalid size for indirect buffer table");
378 exit(1);
381 /* If we've got too many, that implies a descriptor loop. */
382 if (num_bufs >= max) {
383 error_report("Looped descriptor");
384 exit(1);
387 /* loop over the indirect descriptor table */
388 indirect = 1;
389 max = vring_desc_len(vdev, desc_pa, i) / sizeof(VRingDesc);
390 desc_pa = vring_desc_addr(vdev, desc_pa, i);
391 num_bufs = i = 0;
394 do {
395 /* If we've got too many, that implies a descriptor loop. */
396 if (++num_bufs > max) {
397 error_report("Looped descriptor");
398 exit(1);
401 if (vring_desc_flags(vdev, desc_pa, i) & VRING_DESC_F_WRITE) {
402 in_total += vring_desc_len(vdev, desc_pa, i);
403 } else {
404 out_total += vring_desc_len(vdev, desc_pa, i);
406 if (in_total >= max_in_bytes && out_total >= max_out_bytes) {
407 goto done;
409 } while ((i = virtqueue_next_desc(vdev, desc_pa, i, max)) != max);
411 if (!indirect)
412 total_bufs = num_bufs;
413 else
414 total_bufs++;
416 done:
417 if (in_bytes) {
418 *in_bytes = in_total;
420 if (out_bytes) {
421 *out_bytes = out_total;
425 int virtqueue_avail_bytes(VirtQueue *vq, unsigned int in_bytes,
426 unsigned int out_bytes)
428 unsigned int in_total, out_total;
430 virtqueue_get_avail_bytes(vq, &in_total, &out_total, in_bytes, out_bytes);
431 return in_bytes <= in_total && out_bytes <= out_total;
434 void virtqueue_map_sg(struct iovec *sg, hwaddr *addr,
435 size_t num_sg, int is_write)
437 unsigned int i;
438 hwaddr len;
440 if (num_sg > VIRTQUEUE_MAX_SIZE) {
441 error_report("virtio: map attempt out of bounds: %zd > %d",
442 num_sg, VIRTQUEUE_MAX_SIZE);
443 exit(1);
446 for (i = 0; i < num_sg; i++) {
447 len = sg[i].iov_len;
448 sg[i].iov_base = cpu_physical_memory_map(addr[i], &len, is_write);
449 if (sg[i].iov_base == NULL || len != sg[i].iov_len) {
450 error_report("virtio: error trying to map MMIO memory");
451 exit(1);
456 int virtqueue_pop(VirtQueue *vq, VirtQueueElement *elem)
458 unsigned int i, head, max;
459 hwaddr desc_pa = vq->vring.desc;
460 VirtIODevice *vdev = vq->vdev;
462 if (!virtqueue_num_heads(vq, vq->last_avail_idx))
463 return 0;
465 /* When we start there are none of either input nor output. */
466 elem->out_num = elem->in_num = 0;
468 max = vq->vring.num;
470 i = head = virtqueue_get_head(vq, vq->last_avail_idx++);
471 if (vdev->guest_features & (1 << VIRTIO_RING_F_EVENT_IDX)) {
472 vring_avail_event(vq, vq->last_avail_idx);
475 if (vring_desc_flags(vdev, desc_pa, i) & VRING_DESC_F_INDIRECT) {
476 if (vring_desc_len(vdev, desc_pa, i) % sizeof(VRingDesc)) {
477 error_report("Invalid size for indirect buffer table");
478 exit(1);
481 /* loop over the indirect descriptor table */
482 max = vring_desc_len(vdev, desc_pa, i) / sizeof(VRingDesc);
483 desc_pa = vring_desc_addr(vdev, desc_pa, i);
484 i = 0;
487 /* Collect all the descriptors */
488 do {
489 struct iovec *sg;
491 if (vring_desc_flags(vdev, desc_pa, i) & VRING_DESC_F_WRITE) {
492 if (elem->in_num >= ARRAY_SIZE(elem->in_sg)) {
493 error_report("Too many write descriptors in indirect table");
494 exit(1);
496 elem->in_addr[elem->in_num] = vring_desc_addr(vdev, desc_pa, i);
497 sg = &elem->in_sg[elem->in_num++];
498 } else {
499 if (elem->out_num >= ARRAY_SIZE(elem->out_sg)) {
500 error_report("Too many read descriptors in indirect table");
501 exit(1);
503 elem->out_addr[elem->out_num] = vring_desc_addr(vdev, desc_pa, i);
504 sg = &elem->out_sg[elem->out_num++];
507 sg->iov_len = vring_desc_len(vdev, desc_pa, i);
509 /* If we've got too many, that implies a descriptor loop. */
510 if ((elem->in_num + elem->out_num) > max) {
511 error_report("Looped descriptor");
512 exit(1);
514 } while ((i = virtqueue_next_desc(vdev, desc_pa, i, max)) != max);
516 /* Now map what we have collected */
517 virtqueue_map_sg(elem->in_sg, elem->in_addr, elem->in_num, 1);
518 virtqueue_map_sg(elem->out_sg, elem->out_addr, elem->out_num, 0);
520 elem->index = head;
522 vq->inuse++;
524 trace_virtqueue_pop(vq, elem, elem->in_num, elem->out_num);
525 return elem->in_num + elem->out_num;
528 /* virtio device */
529 static void virtio_notify_vector(VirtIODevice *vdev, uint16_t vector)
531 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
532 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
534 if (k->notify) {
535 k->notify(qbus->parent, vector);
539 void virtio_update_irq(VirtIODevice *vdev)
541 virtio_notify_vector(vdev, VIRTIO_NO_VECTOR);
544 void virtio_set_status(VirtIODevice *vdev, uint8_t val)
546 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
547 trace_virtio_set_status(vdev, val);
549 if (k->set_status) {
550 k->set_status(vdev, val);
552 vdev->status = val;
555 bool target_words_bigendian(void);
556 static enum virtio_device_endian virtio_default_endian(void)
558 if (target_words_bigendian()) {
559 return VIRTIO_DEVICE_ENDIAN_BIG;
560 } else {
561 return VIRTIO_DEVICE_ENDIAN_LITTLE;
565 static enum virtio_device_endian virtio_current_cpu_endian(void)
567 CPUClass *cc = CPU_GET_CLASS(current_cpu);
569 if (cc->virtio_is_big_endian(current_cpu)) {
570 return VIRTIO_DEVICE_ENDIAN_BIG;
571 } else {
572 return VIRTIO_DEVICE_ENDIAN_LITTLE;
576 void virtio_reset(void *opaque)
578 VirtIODevice *vdev = opaque;
579 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
580 int i;
582 virtio_set_status(vdev, 0);
583 if (current_cpu) {
584 /* Guest initiated reset */
585 vdev->device_endian = virtio_current_cpu_endian();
586 } else {
587 /* System reset */
588 vdev->device_endian = virtio_default_endian();
591 if (k->reset) {
592 k->reset(vdev);
595 vdev->guest_features = 0;
596 vdev->queue_sel = 0;
597 vdev->status = 0;
598 vdev->isr = 0;
599 vdev->config_vector = VIRTIO_NO_VECTOR;
600 virtio_notify_vector(vdev, vdev->config_vector);
602 for(i = 0; i < VIRTIO_PCI_QUEUE_MAX; i++) {
603 vdev->vq[i].vring.desc = 0;
604 vdev->vq[i].vring.avail = 0;
605 vdev->vq[i].vring.used = 0;
606 vdev->vq[i].last_avail_idx = 0;
607 vdev->vq[i].pa = 0;
608 vdev->vq[i].vector = VIRTIO_NO_VECTOR;
609 vdev->vq[i].signalled_used = 0;
610 vdev->vq[i].signalled_used_valid = false;
611 vdev->vq[i].notification = true;
615 uint32_t virtio_config_readb(VirtIODevice *vdev, uint32_t addr)
617 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
618 uint8_t val;
620 if (addr + sizeof(val) > vdev->config_len) {
621 return (uint32_t)-1;
624 k->get_config(vdev, vdev->config);
626 val = ldub_p(vdev->config + addr);
627 return val;
630 uint32_t virtio_config_readw(VirtIODevice *vdev, uint32_t addr)
632 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
633 uint16_t val;
635 if (addr + sizeof(val) > vdev->config_len) {
636 return (uint32_t)-1;
639 k->get_config(vdev, vdev->config);
641 val = lduw_p(vdev->config + addr);
642 return val;
645 uint32_t virtio_config_readl(VirtIODevice *vdev, uint32_t addr)
647 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
648 uint32_t val;
650 if (addr + sizeof(val) > vdev->config_len) {
651 return (uint32_t)-1;
654 k->get_config(vdev, vdev->config);
656 val = ldl_p(vdev->config + addr);
657 return val;
660 void virtio_config_writeb(VirtIODevice *vdev, uint32_t addr, uint32_t data)
662 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
663 uint8_t val = data;
665 if (addr + sizeof(val) > vdev->config_len) {
666 return;
669 stb_p(vdev->config + addr, val);
671 if (k->set_config) {
672 k->set_config(vdev, vdev->config);
676 void virtio_config_writew(VirtIODevice *vdev, uint32_t addr, uint32_t data)
678 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
679 uint16_t val = data;
681 if (addr + sizeof(val) > vdev->config_len) {
682 return;
685 stw_p(vdev->config + addr, val);
687 if (k->set_config) {
688 k->set_config(vdev, vdev->config);
692 void virtio_config_writel(VirtIODevice *vdev, uint32_t addr, uint32_t data)
694 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
695 uint32_t val = data;
697 if (addr + sizeof(val) > vdev->config_len) {
698 return;
701 stl_p(vdev->config + addr, val);
703 if (k->set_config) {
704 k->set_config(vdev, vdev->config);
708 void virtio_queue_set_addr(VirtIODevice *vdev, int n, hwaddr addr)
710 vdev->vq[n].pa = addr;
711 virtqueue_init(&vdev->vq[n]);
714 hwaddr virtio_queue_get_addr(VirtIODevice *vdev, int n)
716 return vdev->vq[n].pa;
719 void virtio_queue_set_num(VirtIODevice *vdev, int n, int num)
721 /* Don't allow guest to flip queue between existent and
722 * nonexistent states, or to set it to an invalid size.
724 if (!!num != !!vdev->vq[n].vring.num ||
725 num > VIRTQUEUE_MAX_SIZE ||
726 num < 0) {
727 return;
729 vdev->vq[n].vring.num = num;
730 virtqueue_init(&vdev->vq[n]);
733 int virtio_queue_get_num(VirtIODevice *vdev, int n)
735 return vdev->vq[n].vring.num;
738 int virtio_queue_get_id(VirtQueue *vq)
740 VirtIODevice *vdev = vq->vdev;
741 assert(vq >= &vdev->vq[0] && vq < &vdev->vq[VIRTIO_PCI_QUEUE_MAX]);
742 return vq - &vdev->vq[0];
745 void virtio_queue_set_align(VirtIODevice *vdev, int n, int align)
747 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
748 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
750 /* Check that the transport told us it was going to do this
751 * (so a buggy transport will immediately assert rather than
752 * silently failing to migrate this state)
754 assert(k->has_variable_vring_alignment);
756 vdev->vq[n].vring.align = align;
757 virtqueue_init(&vdev->vq[n]);
760 void virtio_queue_notify_vq(VirtQueue *vq)
762 if (vq->vring.desc) {
763 VirtIODevice *vdev = vq->vdev;
764 trace_virtio_queue_notify(vdev, vq - vdev->vq, vq);
765 vq->handle_output(vdev, vq);
769 void virtio_queue_notify(VirtIODevice *vdev, int n)
771 virtio_queue_notify_vq(&vdev->vq[n]);
774 uint16_t virtio_queue_vector(VirtIODevice *vdev, int n)
776 return n < VIRTIO_PCI_QUEUE_MAX ? vdev->vq[n].vector :
777 VIRTIO_NO_VECTOR;
780 void virtio_queue_set_vector(VirtIODevice *vdev, int n, uint16_t vector)
782 if (n < VIRTIO_PCI_QUEUE_MAX)
783 vdev->vq[n].vector = vector;
786 VirtQueue *virtio_add_queue(VirtIODevice *vdev, int queue_size,
787 void (*handle_output)(VirtIODevice *, VirtQueue *))
789 int i;
791 for (i = 0; i < VIRTIO_PCI_QUEUE_MAX; i++) {
792 if (vdev->vq[i].vring.num == 0)
793 break;
796 if (i == VIRTIO_PCI_QUEUE_MAX || queue_size > VIRTQUEUE_MAX_SIZE)
797 abort();
799 vdev->vq[i].vring.num = queue_size;
800 vdev->vq[i].vring.align = VIRTIO_PCI_VRING_ALIGN;
801 vdev->vq[i].handle_output = handle_output;
803 return &vdev->vq[i];
806 void virtio_del_queue(VirtIODevice *vdev, int n)
808 if (n < 0 || n >= VIRTIO_PCI_QUEUE_MAX) {
809 abort();
812 vdev->vq[n].vring.num = 0;
815 void virtio_irq(VirtQueue *vq)
817 trace_virtio_irq(vq);
818 vq->vdev->isr |= 0x01;
819 virtio_notify_vector(vq->vdev, vq->vector);
822 /* Assuming a given event_idx value from the other size, if
823 * we have just incremented index from old to new_idx,
824 * should we trigger an event? */
825 static inline int vring_need_event(uint16_t event, uint16_t new, uint16_t old)
827 /* Note: Xen has similar logic for notification hold-off
828 * in include/xen/interface/io/ring.h with req_event and req_prod
829 * corresponding to event_idx + 1 and new respectively.
830 * Note also that req_event and req_prod in Xen start at 1,
831 * event indexes in virtio start at 0. */
832 return (uint16_t)(new - event - 1) < (uint16_t)(new - old);
835 static bool vring_notify(VirtIODevice *vdev, VirtQueue *vq)
837 uint16_t old, new;
838 bool v;
839 /* We need to expose used array entries before checking used event. */
840 smp_mb();
841 /* Always notify when queue is empty (when feature acknowledge) */
842 if (((vdev->guest_features & (1 << VIRTIO_F_NOTIFY_ON_EMPTY)) &&
843 !vq->inuse && vring_avail_idx(vq) == vq->last_avail_idx)) {
844 return true;
847 if (!(vdev->guest_features & (1 << VIRTIO_RING_F_EVENT_IDX))) {
848 return !(vring_avail_flags(vq) & VRING_AVAIL_F_NO_INTERRUPT);
851 v = vq->signalled_used_valid;
852 vq->signalled_used_valid = true;
853 old = vq->signalled_used;
854 new = vq->signalled_used = vring_used_idx(vq);
855 return !v || vring_need_event(vring_used_event(vq), new, old);
858 void virtio_notify(VirtIODevice *vdev, VirtQueue *vq)
860 if (!vring_notify(vdev, vq)) {
861 return;
864 trace_virtio_notify(vdev, vq);
865 vdev->isr |= 0x01;
866 virtio_notify_vector(vdev, vq->vector);
869 void virtio_notify_config(VirtIODevice *vdev)
871 if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK))
872 return;
874 vdev->isr |= 0x03;
875 virtio_notify_vector(vdev, vdev->config_vector);
878 static bool virtio_device_endian_needed(void *opaque)
880 VirtIODevice *vdev = opaque;
882 assert(vdev->device_endian != VIRTIO_DEVICE_ENDIAN_UNKNOWN);
883 return vdev->device_endian != virtio_default_endian();
886 static const VMStateDescription vmstate_virtio_device_endian = {
887 .name = "virtio/device_endian",
888 .version_id = 1,
889 .minimum_version_id = 1,
890 .fields = (VMStateField[]) {
891 VMSTATE_UINT8(device_endian, VirtIODevice),
892 VMSTATE_END_OF_LIST()
896 static const VMStateDescription vmstate_virtio = {
897 .name = "virtio",
898 .version_id = 1,
899 .minimum_version_id = 1,
900 .minimum_version_id_old = 1,
901 .fields = (VMStateField[]) {
902 VMSTATE_END_OF_LIST()
904 .subsections = (VMStateSubsection[]) {
906 .vmsd = &vmstate_virtio_device_endian,
907 .needed = &virtio_device_endian_needed
909 { 0 }
913 void virtio_save(VirtIODevice *vdev, QEMUFile *f)
915 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
916 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
917 VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev);
918 int i;
920 if (k->save_config) {
921 k->save_config(qbus->parent, f);
924 qemu_put_8s(f, &vdev->status);
925 qemu_put_8s(f, &vdev->isr);
926 qemu_put_be16s(f, &vdev->queue_sel);
927 qemu_put_be32s(f, &vdev->guest_features);
928 qemu_put_be32(f, vdev->config_len);
929 qemu_put_buffer(f, vdev->config, vdev->config_len);
931 for (i = 0; i < VIRTIO_PCI_QUEUE_MAX; i++) {
932 if (vdev->vq[i].vring.num == 0)
933 break;
936 qemu_put_be32(f, i);
938 for (i = 0; i < VIRTIO_PCI_QUEUE_MAX; i++) {
939 if (vdev->vq[i].vring.num == 0)
940 break;
942 qemu_put_be32(f, vdev->vq[i].vring.num);
943 if (k->has_variable_vring_alignment) {
944 qemu_put_be32(f, vdev->vq[i].vring.align);
946 qemu_put_be64(f, vdev->vq[i].pa);
947 qemu_put_be16s(f, &vdev->vq[i].last_avail_idx);
948 if (k->save_queue) {
949 k->save_queue(qbus->parent, i, f);
953 if (vdc->save != NULL) {
954 vdc->save(vdev, f);
957 /* Subsections */
958 vmstate_save_state(f, &vmstate_virtio, vdev);
961 int virtio_set_features(VirtIODevice *vdev, uint32_t val)
963 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
964 VirtioBusClass *vbusk = VIRTIO_BUS_GET_CLASS(qbus);
965 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
966 uint32_t supported_features = vbusk->get_features(qbus->parent);
967 bool bad = (val & ~supported_features) != 0;
969 val &= supported_features;
970 if (k->set_features) {
971 k->set_features(vdev, val);
973 vdev->guest_features = val;
974 return bad ? -1 : 0;
977 int virtio_load(VirtIODevice *vdev, QEMUFile *f, int version_id)
979 int i, ret;
980 int32_t config_len;
981 uint32_t num;
982 uint32_t features;
983 uint32_t supported_features;
984 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
985 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
986 VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev);
989 * We poison the endianness to ensure it does not get used before
990 * subsections have been loaded.
992 vdev->device_endian = VIRTIO_DEVICE_ENDIAN_UNKNOWN;
994 if (k->load_config) {
995 ret = k->load_config(qbus->parent, f);
996 if (ret)
997 return ret;
1000 qemu_get_8s(f, &vdev->status);
1001 qemu_get_8s(f, &vdev->isr);
1002 qemu_get_be16s(f, &vdev->queue_sel);
1003 if (vdev->queue_sel >= VIRTIO_PCI_QUEUE_MAX) {
1004 return -1;
1006 qemu_get_be32s(f, &features);
1008 if (virtio_set_features(vdev, features) < 0) {
1009 supported_features = k->get_features(qbus->parent);
1010 error_report("Features 0x%x unsupported. Allowed features: 0x%x",
1011 features, supported_features);
1012 return -1;
1014 config_len = qemu_get_be32(f);
1017 * There are cases where the incoming config can be bigger or smaller
1018 * than what we have; so load what we have space for, and skip
1019 * any excess that's in the stream.
1021 qemu_get_buffer(f, vdev->config, MIN(config_len, vdev->config_len));
1023 while (config_len > vdev->config_len) {
1024 qemu_get_byte(f);
1025 config_len--;
1028 num = qemu_get_be32(f);
1030 if (num > VIRTIO_PCI_QUEUE_MAX) {
1031 error_report("Invalid number of PCI queues: 0x%x", num);
1032 return -1;
1035 for (i = 0; i < num; i++) {
1036 vdev->vq[i].vring.num = qemu_get_be32(f);
1037 if (k->has_variable_vring_alignment) {
1038 vdev->vq[i].vring.align = qemu_get_be32(f);
1040 vdev->vq[i].pa = qemu_get_be64(f);
1041 qemu_get_be16s(f, &vdev->vq[i].last_avail_idx);
1042 vdev->vq[i].signalled_used_valid = false;
1043 vdev->vq[i].notification = true;
1045 if (vdev->vq[i].pa) {
1046 virtqueue_init(&vdev->vq[i]);
1047 } else if (vdev->vq[i].last_avail_idx) {
1048 error_report("VQ %d address 0x0 "
1049 "inconsistent with Host index 0x%x",
1050 i, vdev->vq[i].last_avail_idx);
1051 return -1;
1053 if (k->load_queue) {
1054 ret = k->load_queue(qbus->parent, i, f);
1055 if (ret)
1056 return ret;
1060 virtio_notify_vector(vdev, VIRTIO_NO_VECTOR);
1062 if (vdc->load != NULL) {
1063 ret = vdc->load(vdev, f, version_id);
1064 if (ret) {
1065 return ret;
1069 /* Subsections */
1070 ret = vmstate_load_state(f, &vmstate_virtio, vdev, 1);
1071 if (ret) {
1072 return ret;
1075 if (vdev->device_endian == VIRTIO_DEVICE_ENDIAN_UNKNOWN) {
1076 vdev->device_endian = virtio_default_endian();
1079 for (i = 0; i < num; i++) {
1080 if (vdev->vq[i].pa) {
1081 uint16_t nheads;
1082 nheads = vring_avail_idx(&vdev->vq[i]) - vdev->vq[i].last_avail_idx;
1083 /* Check it isn't doing strange things with descriptor numbers. */
1084 if (nheads > vdev->vq[i].vring.num) {
1085 error_report("VQ %d size 0x%x Guest index 0x%x "
1086 "inconsistent with Host index 0x%x: delta 0x%x",
1087 i, vdev->vq[i].vring.num,
1088 vring_avail_idx(&vdev->vq[i]),
1089 vdev->vq[i].last_avail_idx, nheads);
1090 return -1;
1095 return 0;
1098 void virtio_cleanup(VirtIODevice *vdev)
1100 qemu_del_vm_change_state_handler(vdev->vmstate);
1101 g_free(vdev->config);
1102 g_free(vdev->vq);
1105 static void virtio_vmstate_change(void *opaque, int running, RunState state)
1107 VirtIODevice *vdev = opaque;
1108 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
1109 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
1110 bool backend_run = running && (vdev->status & VIRTIO_CONFIG_S_DRIVER_OK);
1111 vdev->vm_running = running;
1113 if (backend_run) {
1114 virtio_set_status(vdev, vdev->status);
1117 if (k->vmstate_change) {
1118 k->vmstate_change(qbus->parent, backend_run);
1121 if (!backend_run) {
1122 virtio_set_status(vdev, vdev->status);
1126 void virtio_instance_init_common(Object *proxy_obj, void *data,
1127 size_t vdev_size, const char *vdev_name)
1129 DeviceState *vdev = data;
1131 object_initialize(vdev, vdev_size, vdev_name);
1132 object_property_add_child(proxy_obj, "virtio-backend", OBJECT(vdev), NULL);
1133 object_unref(OBJECT(vdev));
1134 qdev_alias_all_properties(vdev, proxy_obj);
1137 void virtio_init(VirtIODevice *vdev, const char *name,
1138 uint16_t device_id, size_t config_size)
1140 int i;
1141 vdev->device_id = device_id;
1142 vdev->status = 0;
1143 vdev->isr = 0;
1144 vdev->queue_sel = 0;
1145 vdev->config_vector = VIRTIO_NO_VECTOR;
1146 vdev->vq = g_malloc0(sizeof(VirtQueue) * VIRTIO_PCI_QUEUE_MAX);
1147 vdev->vm_running = runstate_is_running();
1148 for (i = 0; i < VIRTIO_PCI_QUEUE_MAX; i++) {
1149 vdev->vq[i].vector = VIRTIO_NO_VECTOR;
1150 vdev->vq[i].vdev = vdev;
1151 vdev->vq[i].queue_index = i;
1154 vdev->name = name;
1155 vdev->config_len = config_size;
1156 if (vdev->config_len) {
1157 vdev->config = g_malloc0(config_size);
1158 } else {
1159 vdev->config = NULL;
1161 vdev->vmstate = qemu_add_vm_change_state_handler(virtio_vmstate_change,
1162 vdev);
1163 vdev->device_endian = virtio_default_endian();
1166 hwaddr virtio_queue_get_desc_addr(VirtIODevice *vdev, int n)
1168 return vdev->vq[n].vring.desc;
1171 hwaddr virtio_queue_get_avail_addr(VirtIODevice *vdev, int n)
1173 return vdev->vq[n].vring.avail;
1176 hwaddr virtio_queue_get_used_addr(VirtIODevice *vdev, int n)
1178 return vdev->vq[n].vring.used;
1181 hwaddr virtio_queue_get_ring_addr(VirtIODevice *vdev, int n)
1183 return vdev->vq[n].vring.desc;
1186 hwaddr virtio_queue_get_desc_size(VirtIODevice *vdev, int n)
1188 return sizeof(VRingDesc) * vdev->vq[n].vring.num;
1191 hwaddr virtio_queue_get_avail_size(VirtIODevice *vdev, int n)
1193 return offsetof(VRingAvail, ring) +
1194 sizeof(uint64_t) * vdev->vq[n].vring.num;
1197 hwaddr virtio_queue_get_used_size(VirtIODevice *vdev, int n)
1199 return offsetof(VRingUsed, ring) +
1200 sizeof(VRingUsedElem) * vdev->vq[n].vring.num;
1203 hwaddr virtio_queue_get_ring_size(VirtIODevice *vdev, int n)
1205 return vdev->vq[n].vring.used - vdev->vq[n].vring.desc +
1206 virtio_queue_get_used_size(vdev, n);
1209 uint16_t virtio_queue_get_last_avail_idx(VirtIODevice *vdev, int n)
1211 return vdev->vq[n].last_avail_idx;
1214 void virtio_queue_set_last_avail_idx(VirtIODevice *vdev, int n, uint16_t idx)
1216 vdev->vq[n].last_avail_idx = idx;
1219 void virtio_queue_invalidate_signalled_used(VirtIODevice *vdev, int n)
1221 vdev->vq[n].signalled_used_valid = false;
1224 VirtQueue *virtio_get_queue(VirtIODevice *vdev, int n)
1226 return vdev->vq + n;
1229 uint16_t virtio_get_queue_index(VirtQueue *vq)
1231 return vq->queue_index;
1234 static void virtio_queue_guest_notifier_read(EventNotifier *n)
1236 VirtQueue *vq = container_of(n, VirtQueue, guest_notifier);
1237 if (event_notifier_test_and_clear(n)) {
1238 virtio_irq(vq);
1242 void virtio_queue_set_guest_notifier_fd_handler(VirtQueue *vq, bool assign,
1243 bool with_irqfd)
1245 if (assign && !with_irqfd) {
1246 event_notifier_set_handler(&vq->guest_notifier,
1247 virtio_queue_guest_notifier_read);
1248 } else {
1249 event_notifier_set_handler(&vq->guest_notifier, NULL);
1251 if (!assign) {
1252 /* Test and clear notifier before closing it,
1253 * in case poll callback didn't have time to run. */
1254 virtio_queue_guest_notifier_read(&vq->guest_notifier);
1258 EventNotifier *virtio_queue_get_guest_notifier(VirtQueue *vq)
1260 return &vq->guest_notifier;
1263 static void virtio_queue_host_notifier_read(EventNotifier *n)
1265 VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
1266 if (event_notifier_test_and_clear(n)) {
1267 virtio_queue_notify_vq(vq);
1271 void virtio_queue_set_host_notifier_fd_handler(VirtQueue *vq, bool assign,
1272 bool set_handler)
1274 if (assign && set_handler) {
1275 event_notifier_set_handler(&vq->host_notifier,
1276 virtio_queue_host_notifier_read);
1277 } else {
1278 event_notifier_set_handler(&vq->host_notifier, NULL);
1280 if (!assign) {
1281 /* Test and clear notifier before after disabling event,
1282 * in case poll callback didn't have time to run. */
1283 virtio_queue_host_notifier_read(&vq->host_notifier);
1287 EventNotifier *virtio_queue_get_host_notifier(VirtQueue *vq)
1289 return &vq->host_notifier;
1292 void virtio_device_set_child_bus_name(VirtIODevice *vdev, char *bus_name)
1294 g_free(vdev->bus_name);
1295 vdev->bus_name = g_strdup(bus_name);
1298 static void virtio_device_realize(DeviceState *dev, Error **errp)
1300 VirtIODevice *vdev = VIRTIO_DEVICE(dev);
1301 VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(dev);
1302 Error *err = NULL;
1304 if (vdc->realize != NULL) {
1305 vdc->realize(dev, &err);
1306 if (err != NULL) {
1307 error_propagate(errp, err);
1308 return;
1311 virtio_bus_device_plugged(vdev);
1314 static void virtio_device_unrealize(DeviceState *dev, Error **errp)
1316 VirtIODevice *vdev = VIRTIO_DEVICE(dev);
1317 VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(dev);
1318 Error *err = NULL;
1320 virtio_bus_device_unplugged(vdev);
1322 if (vdc->unrealize != NULL) {
1323 vdc->unrealize(dev, &err);
1324 if (err != NULL) {
1325 error_propagate(errp, err);
1326 return;
1330 g_free(vdev->bus_name);
1331 vdev->bus_name = NULL;
1334 static void virtio_device_class_init(ObjectClass *klass, void *data)
1336 /* Set the default value here. */
1337 DeviceClass *dc = DEVICE_CLASS(klass);
1339 dc->realize = virtio_device_realize;
1340 dc->unrealize = virtio_device_unrealize;
1341 dc->bus_type = TYPE_VIRTIO_BUS;
1344 static const TypeInfo virtio_device_info = {
1345 .name = TYPE_VIRTIO_DEVICE,
1346 .parent = TYPE_DEVICE,
1347 .instance_size = sizeof(VirtIODevice),
1348 .class_init = virtio_device_class_init,
1349 .abstract = true,
1350 .class_size = sizeof(VirtioDeviceClass),
1353 static void virtio_register_types(void)
1355 type_register_static(&virtio_device_info);
1358 type_init(virtio_register_types)