virtio: introduce device specific migration calls
[qemu.git] / hw / virtio / virtio.c
blob7f9ac5e0b9eaa77a7310c5b0d63206acfccb8676
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"
24 * The alignment to use between consumer and producer parts of vring.
25 * x86 pagesize again. This is the default, used by transports like PCI
26 * which don't provide a means for the guest to tell the host the alignment.
28 #define VIRTIO_PCI_VRING_ALIGN 4096
30 typedef struct VRingDesc
32 uint64_t addr;
33 uint32_t len;
34 uint16_t flags;
35 uint16_t next;
36 } VRingDesc;
38 typedef struct VRingAvail
40 uint16_t flags;
41 uint16_t idx;
42 uint16_t ring[0];
43 } VRingAvail;
45 typedef struct VRingUsedElem
47 uint32_t id;
48 uint32_t len;
49 } VRingUsedElem;
51 typedef struct VRingUsed
53 uint16_t flags;
54 uint16_t idx;
55 VRingUsedElem ring[0];
56 } VRingUsed;
58 typedef struct VRing
60 unsigned int num;
61 unsigned int align;
62 hwaddr desc;
63 hwaddr avail;
64 hwaddr used;
65 } VRing;
67 struct VirtQueue
69 VRing vring;
70 hwaddr pa;
71 uint16_t last_avail_idx;
72 /* Last used index value we have signalled on */
73 uint16_t signalled_used;
75 /* Last used index value we have signalled on */
76 bool signalled_used_valid;
78 /* Notification enabled? */
79 bool notification;
81 uint16_t queue_index;
83 int inuse;
85 uint16_t vector;
86 void (*handle_output)(VirtIODevice *vdev, VirtQueue *vq);
87 VirtIODevice *vdev;
88 EventNotifier guest_notifier;
89 EventNotifier host_notifier;
92 /* virt queue functions */
93 static void virtqueue_init(VirtQueue *vq)
95 hwaddr pa = vq->pa;
97 vq->vring.desc = pa;
98 vq->vring.avail = pa + vq->vring.num * sizeof(VRingDesc);
99 vq->vring.used = vring_align(vq->vring.avail +
100 offsetof(VRingAvail, ring[vq->vring.num]),
101 vq->vring.align);
104 static inline uint64_t vring_desc_addr(hwaddr desc_pa, int i)
106 hwaddr pa;
107 pa = desc_pa + sizeof(VRingDesc) * i + offsetof(VRingDesc, addr);
108 return ldq_phys(&address_space_memory, pa);
111 static inline uint32_t vring_desc_len(hwaddr desc_pa, int i)
113 hwaddr pa;
114 pa = desc_pa + sizeof(VRingDesc) * i + offsetof(VRingDesc, len);
115 return ldl_phys(&address_space_memory, pa);
118 static inline uint16_t vring_desc_flags(hwaddr desc_pa, int i)
120 hwaddr pa;
121 pa = desc_pa + sizeof(VRingDesc) * i + offsetof(VRingDesc, flags);
122 return lduw_phys(&address_space_memory, pa);
125 static inline uint16_t vring_desc_next(hwaddr desc_pa, int i)
127 hwaddr pa;
128 pa = desc_pa + sizeof(VRingDesc) * i + offsetof(VRingDesc, next);
129 return lduw_phys(&address_space_memory, pa);
132 static inline uint16_t vring_avail_flags(VirtQueue *vq)
134 hwaddr pa;
135 pa = vq->vring.avail + offsetof(VRingAvail, flags);
136 return lduw_phys(&address_space_memory, pa);
139 static inline uint16_t vring_avail_idx(VirtQueue *vq)
141 hwaddr pa;
142 pa = vq->vring.avail + offsetof(VRingAvail, idx);
143 return lduw_phys(&address_space_memory, pa);
146 static inline uint16_t vring_avail_ring(VirtQueue *vq, int i)
148 hwaddr pa;
149 pa = vq->vring.avail + offsetof(VRingAvail, ring[i]);
150 return lduw_phys(&address_space_memory, pa);
153 static inline uint16_t vring_used_event(VirtQueue *vq)
155 return vring_avail_ring(vq, vq->vring.num);
158 static inline void vring_used_ring_id(VirtQueue *vq, int i, uint32_t val)
160 hwaddr pa;
161 pa = vq->vring.used + offsetof(VRingUsed, ring[i].id);
162 stl_phys(&address_space_memory, pa, val);
165 static inline void vring_used_ring_len(VirtQueue *vq, int i, uint32_t val)
167 hwaddr pa;
168 pa = vq->vring.used + offsetof(VRingUsed, ring[i].len);
169 stl_phys(&address_space_memory, pa, val);
172 static uint16_t vring_used_idx(VirtQueue *vq)
174 hwaddr pa;
175 pa = vq->vring.used + offsetof(VRingUsed, idx);
176 return lduw_phys(&address_space_memory, pa);
179 static inline void vring_used_idx_set(VirtQueue *vq, uint16_t val)
181 hwaddr pa;
182 pa = vq->vring.used + offsetof(VRingUsed, idx);
183 stw_phys(&address_space_memory, pa, val);
186 static inline void vring_used_flags_set_bit(VirtQueue *vq, int mask)
188 hwaddr pa;
189 pa = vq->vring.used + offsetof(VRingUsed, flags);
190 stw_phys(&address_space_memory,
191 pa, lduw_phys(&address_space_memory, pa) | mask);
194 static inline void vring_used_flags_unset_bit(VirtQueue *vq, int mask)
196 hwaddr pa;
197 pa = vq->vring.used + offsetof(VRingUsed, flags);
198 stw_phys(&address_space_memory,
199 pa, lduw_phys(&address_space_memory, pa) & ~mask);
202 static inline void vring_avail_event(VirtQueue *vq, uint16_t val)
204 hwaddr pa;
205 if (!vq->notification) {
206 return;
208 pa = vq->vring.used + offsetof(VRingUsed, ring[vq->vring.num]);
209 stw_phys(&address_space_memory, pa, val);
212 void virtio_queue_set_notification(VirtQueue *vq, int enable)
214 vq->notification = enable;
215 if (vq->vdev->guest_features & (1 << VIRTIO_RING_F_EVENT_IDX)) {
216 vring_avail_event(vq, vring_avail_idx(vq));
217 } else if (enable) {
218 vring_used_flags_unset_bit(vq, VRING_USED_F_NO_NOTIFY);
219 } else {
220 vring_used_flags_set_bit(vq, VRING_USED_F_NO_NOTIFY);
222 if (enable) {
223 /* Expose avail event/used flags before caller checks the avail idx. */
224 smp_mb();
228 int virtio_queue_ready(VirtQueue *vq)
230 return vq->vring.avail != 0;
233 int virtio_queue_empty(VirtQueue *vq)
235 return vring_avail_idx(vq) == vq->last_avail_idx;
238 void virtqueue_fill(VirtQueue *vq, const VirtQueueElement *elem,
239 unsigned int len, unsigned int idx)
241 unsigned int offset;
242 int i;
244 trace_virtqueue_fill(vq, elem, len, idx);
246 offset = 0;
247 for (i = 0; i < elem->in_num; i++) {
248 size_t size = MIN(len - offset, elem->in_sg[i].iov_len);
250 cpu_physical_memory_unmap(elem->in_sg[i].iov_base,
251 elem->in_sg[i].iov_len,
252 1, size);
254 offset += size;
257 for (i = 0; i < elem->out_num; i++)
258 cpu_physical_memory_unmap(elem->out_sg[i].iov_base,
259 elem->out_sg[i].iov_len,
260 0, elem->out_sg[i].iov_len);
262 idx = (idx + vring_used_idx(vq)) % vq->vring.num;
264 /* Get a pointer to the next entry in the used ring. */
265 vring_used_ring_id(vq, idx, elem->index);
266 vring_used_ring_len(vq, idx, len);
269 void virtqueue_flush(VirtQueue *vq, unsigned int count)
271 uint16_t old, new;
272 /* Make sure buffer is written before we update index. */
273 smp_wmb();
274 trace_virtqueue_flush(vq, count);
275 old = vring_used_idx(vq);
276 new = old + count;
277 vring_used_idx_set(vq, new);
278 vq->inuse -= count;
279 if (unlikely((int16_t)(new - vq->signalled_used) < (uint16_t)(new - old)))
280 vq->signalled_used_valid = false;
283 void virtqueue_push(VirtQueue *vq, const VirtQueueElement *elem,
284 unsigned int len)
286 virtqueue_fill(vq, elem, len, 0);
287 virtqueue_flush(vq, 1);
290 static int virtqueue_num_heads(VirtQueue *vq, unsigned int idx)
292 uint16_t num_heads = vring_avail_idx(vq) - idx;
294 /* Check it isn't doing very strange things with descriptor numbers. */
295 if (num_heads > vq->vring.num) {
296 error_report("Guest moved used index from %u to %u",
297 idx, vring_avail_idx(vq));
298 exit(1);
300 /* On success, callers read a descriptor at vq->last_avail_idx.
301 * Make sure descriptor read does not bypass avail index read. */
302 if (num_heads) {
303 smp_rmb();
306 return num_heads;
309 static unsigned int virtqueue_get_head(VirtQueue *vq, unsigned int idx)
311 unsigned int head;
313 /* Grab the next descriptor number they're advertising, and increment
314 * the index we've seen. */
315 head = vring_avail_ring(vq, idx % vq->vring.num);
317 /* If their number is silly, that's a fatal mistake. */
318 if (head >= vq->vring.num) {
319 error_report("Guest says index %u is available", head);
320 exit(1);
323 return head;
326 static unsigned virtqueue_next_desc(hwaddr desc_pa,
327 unsigned int i, unsigned int max)
329 unsigned int next;
331 /* If this descriptor says it doesn't chain, we're done. */
332 if (!(vring_desc_flags(desc_pa, i) & VRING_DESC_F_NEXT))
333 return max;
335 /* Check they're not leading us off end of descriptors. */
336 next = vring_desc_next(desc_pa, i);
337 /* Make sure compiler knows to grab that: we don't want it changing! */
338 smp_wmb();
340 if (next >= max) {
341 error_report("Desc next is %u", next);
342 exit(1);
345 return next;
348 void virtqueue_get_avail_bytes(VirtQueue *vq, unsigned int *in_bytes,
349 unsigned int *out_bytes,
350 unsigned max_in_bytes, unsigned max_out_bytes)
352 unsigned int idx;
353 unsigned int total_bufs, in_total, out_total;
355 idx = vq->last_avail_idx;
357 total_bufs = in_total = out_total = 0;
358 while (virtqueue_num_heads(vq, idx)) {
359 unsigned int max, num_bufs, indirect = 0;
360 hwaddr desc_pa;
361 int i;
363 max = vq->vring.num;
364 num_bufs = total_bufs;
365 i = virtqueue_get_head(vq, idx++);
366 desc_pa = vq->vring.desc;
368 if (vring_desc_flags(desc_pa, i) & VRING_DESC_F_INDIRECT) {
369 if (vring_desc_len(desc_pa, i) % sizeof(VRingDesc)) {
370 error_report("Invalid size for indirect buffer table");
371 exit(1);
374 /* If we've got too many, that implies a descriptor loop. */
375 if (num_bufs >= max) {
376 error_report("Looped descriptor");
377 exit(1);
380 /* loop over the indirect descriptor table */
381 indirect = 1;
382 max = vring_desc_len(desc_pa, i) / sizeof(VRingDesc);
383 desc_pa = vring_desc_addr(desc_pa, i);
384 num_bufs = i = 0;
387 do {
388 /* If we've got too many, that implies a descriptor loop. */
389 if (++num_bufs > max) {
390 error_report("Looped descriptor");
391 exit(1);
394 if (vring_desc_flags(desc_pa, i) & VRING_DESC_F_WRITE) {
395 in_total += vring_desc_len(desc_pa, i);
396 } else {
397 out_total += vring_desc_len(desc_pa, i);
399 if (in_total >= max_in_bytes && out_total >= max_out_bytes) {
400 goto done;
402 } while ((i = virtqueue_next_desc(desc_pa, i, max)) != max);
404 if (!indirect)
405 total_bufs = num_bufs;
406 else
407 total_bufs++;
409 done:
410 if (in_bytes) {
411 *in_bytes = in_total;
413 if (out_bytes) {
414 *out_bytes = out_total;
418 int virtqueue_avail_bytes(VirtQueue *vq, unsigned int in_bytes,
419 unsigned int out_bytes)
421 unsigned int in_total, out_total;
423 virtqueue_get_avail_bytes(vq, &in_total, &out_total, in_bytes, out_bytes);
424 return in_bytes <= in_total && out_bytes <= out_total;
427 void virtqueue_map_sg(struct iovec *sg, hwaddr *addr,
428 size_t num_sg, int is_write)
430 unsigned int i;
431 hwaddr len;
433 if (num_sg > VIRTQUEUE_MAX_SIZE) {
434 error_report("virtio: map attempt out of bounds: %zd > %d",
435 num_sg, VIRTQUEUE_MAX_SIZE);
436 exit(1);
439 for (i = 0; i < num_sg; i++) {
440 len = sg[i].iov_len;
441 sg[i].iov_base = cpu_physical_memory_map(addr[i], &len, is_write);
442 if (sg[i].iov_base == NULL || len != sg[i].iov_len) {
443 error_report("virtio: error trying to map MMIO memory");
444 exit(1);
449 int virtqueue_pop(VirtQueue *vq, VirtQueueElement *elem)
451 unsigned int i, head, max;
452 hwaddr desc_pa = vq->vring.desc;
454 if (!virtqueue_num_heads(vq, vq->last_avail_idx))
455 return 0;
457 /* When we start there are none of either input nor output. */
458 elem->out_num = elem->in_num = 0;
460 max = vq->vring.num;
462 i = head = virtqueue_get_head(vq, vq->last_avail_idx++);
463 if (vq->vdev->guest_features & (1 << VIRTIO_RING_F_EVENT_IDX)) {
464 vring_avail_event(vq, vring_avail_idx(vq));
467 if (vring_desc_flags(desc_pa, i) & VRING_DESC_F_INDIRECT) {
468 if (vring_desc_len(desc_pa, i) % sizeof(VRingDesc)) {
469 error_report("Invalid size for indirect buffer table");
470 exit(1);
473 /* loop over the indirect descriptor table */
474 max = vring_desc_len(desc_pa, i) / sizeof(VRingDesc);
475 desc_pa = vring_desc_addr(desc_pa, i);
476 i = 0;
479 /* Collect all the descriptors */
480 do {
481 struct iovec *sg;
483 if (vring_desc_flags(desc_pa, i) & VRING_DESC_F_WRITE) {
484 if (elem->in_num >= ARRAY_SIZE(elem->in_sg)) {
485 error_report("Too many write descriptors in indirect table");
486 exit(1);
488 elem->in_addr[elem->in_num] = vring_desc_addr(desc_pa, i);
489 sg = &elem->in_sg[elem->in_num++];
490 } else {
491 if (elem->out_num >= ARRAY_SIZE(elem->out_sg)) {
492 error_report("Too many read descriptors in indirect table");
493 exit(1);
495 elem->out_addr[elem->out_num] = vring_desc_addr(desc_pa, i);
496 sg = &elem->out_sg[elem->out_num++];
499 sg->iov_len = vring_desc_len(desc_pa, i);
501 /* If we've got too many, that implies a descriptor loop. */
502 if ((elem->in_num + elem->out_num) > max) {
503 error_report("Looped descriptor");
504 exit(1);
506 } while ((i = virtqueue_next_desc(desc_pa, i, max)) != max);
508 /* Now map what we have collected */
509 virtqueue_map_sg(elem->in_sg, elem->in_addr, elem->in_num, 1);
510 virtqueue_map_sg(elem->out_sg, elem->out_addr, elem->out_num, 0);
512 elem->index = head;
514 vq->inuse++;
516 trace_virtqueue_pop(vq, elem, elem->in_num, elem->out_num);
517 return elem->in_num + elem->out_num;
520 /* virtio device */
521 static void virtio_notify_vector(VirtIODevice *vdev, uint16_t vector)
523 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
524 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
526 if (k->notify) {
527 k->notify(qbus->parent, vector);
531 void virtio_update_irq(VirtIODevice *vdev)
533 virtio_notify_vector(vdev, VIRTIO_NO_VECTOR);
536 void virtio_set_status(VirtIODevice *vdev, uint8_t val)
538 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
539 trace_virtio_set_status(vdev, val);
541 if (k->set_status) {
542 k->set_status(vdev, val);
544 vdev->status = val;
547 void virtio_reset(void *opaque)
549 VirtIODevice *vdev = opaque;
550 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
551 int i;
553 virtio_set_status(vdev, 0);
555 if (k->reset) {
556 k->reset(vdev);
559 vdev->guest_features = 0;
560 vdev->queue_sel = 0;
561 vdev->status = 0;
562 vdev->isr = 0;
563 vdev->config_vector = VIRTIO_NO_VECTOR;
564 virtio_notify_vector(vdev, vdev->config_vector);
566 for(i = 0; i < VIRTIO_PCI_QUEUE_MAX; i++) {
567 vdev->vq[i].vring.desc = 0;
568 vdev->vq[i].vring.avail = 0;
569 vdev->vq[i].vring.used = 0;
570 vdev->vq[i].last_avail_idx = 0;
571 vdev->vq[i].pa = 0;
572 vdev->vq[i].vector = VIRTIO_NO_VECTOR;
573 vdev->vq[i].signalled_used = 0;
574 vdev->vq[i].signalled_used_valid = false;
575 vdev->vq[i].notification = true;
579 uint32_t virtio_config_readb(VirtIODevice *vdev, uint32_t addr)
581 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
582 uint8_t val;
584 if (addr + sizeof(val) > vdev->config_len) {
585 return (uint32_t)-1;
588 k->get_config(vdev, vdev->config);
590 val = ldub_p(vdev->config + addr);
591 return val;
594 uint32_t virtio_config_readw(VirtIODevice *vdev, uint32_t addr)
596 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
597 uint16_t val;
599 if (addr + sizeof(val) > vdev->config_len) {
600 return (uint32_t)-1;
603 k->get_config(vdev, vdev->config);
605 val = lduw_p(vdev->config + addr);
606 return val;
609 uint32_t virtio_config_readl(VirtIODevice *vdev, uint32_t addr)
611 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
612 uint32_t val;
614 if (addr + sizeof(val) > vdev->config_len) {
615 return (uint32_t)-1;
618 k->get_config(vdev, vdev->config);
620 val = ldl_p(vdev->config + addr);
621 return val;
624 void virtio_config_writeb(VirtIODevice *vdev, uint32_t addr, uint32_t data)
626 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
627 uint8_t val = data;
629 if (addr + sizeof(val) > vdev->config_len) {
630 return;
633 stb_p(vdev->config + addr, val);
635 if (k->set_config) {
636 k->set_config(vdev, vdev->config);
640 void virtio_config_writew(VirtIODevice *vdev, uint32_t addr, uint32_t data)
642 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
643 uint16_t val = data;
645 if (addr + sizeof(val) > vdev->config_len) {
646 return;
649 stw_p(vdev->config + addr, val);
651 if (k->set_config) {
652 k->set_config(vdev, vdev->config);
656 void virtio_config_writel(VirtIODevice *vdev, uint32_t addr, uint32_t data)
658 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
659 uint32_t val = data;
661 if (addr + sizeof(val) > vdev->config_len) {
662 return;
665 stl_p(vdev->config + addr, val);
667 if (k->set_config) {
668 k->set_config(vdev, vdev->config);
672 void virtio_queue_set_addr(VirtIODevice *vdev, int n, hwaddr addr)
674 vdev->vq[n].pa = addr;
675 virtqueue_init(&vdev->vq[n]);
678 hwaddr virtio_queue_get_addr(VirtIODevice *vdev, int n)
680 return vdev->vq[n].pa;
683 void virtio_queue_set_num(VirtIODevice *vdev, int n, int num)
685 /* Don't allow guest to flip queue between existent and
686 * nonexistent states, or to set it to an invalid size.
688 if (!!num != !!vdev->vq[n].vring.num ||
689 num > VIRTQUEUE_MAX_SIZE ||
690 num < 0) {
691 return;
693 vdev->vq[n].vring.num = num;
694 virtqueue_init(&vdev->vq[n]);
697 int virtio_queue_get_num(VirtIODevice *vdev, int n)
699 return vdev->vq[n].vring.num;
702 int virtio_queue_get_id(VirtQueue *vq)
704 VirtIODevice *vdev = vq->vdev;
705 assert(vq >= &vdev->vq[0] && vq < &vdev->vq[VIRTIO_PCI_QUEUE_MAX]);
706 return vq - &vdev->vq[0];
709 void virtio_queue_set_align(VirtIODevice *vdev, int n, int align)
711 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
712 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
714 /* Check that the transport told us it was going to do this
715 * (so a buggy transport will immediately assert rather than
716 * silently failing to migrate this state)
718 assert(k->has_variable_vring_alignment);
720 vdev->vq[n].vring.align = align;
721 virtqueue_init(&vdev->vq[n]);
724 void virtio_queue_notify_vq(VirtQueue *vq)
726 if (vq->vring.desc) {
727 VirtIODevice *vdev = vq->vdev;
728 trace_virtio_queue_notify(vdev, vq - vdev->vq, vq);
729 vq->handle_output(vdev, vq);
733 void virtio_queue_notify(VirtIODevice *vdev, int n)
735 virtio_queue_notify_vq(&vdev->vq[n]);
738 uint16_t virtio_queue_vector(VirtIODevice *vdev, int n)
740 return n < VIRTIO_PCI_QUEUE_MAX ? vdev->vq[n].vector :
741 VIRTIO_NO_VECTOR;
744 void virtio_queue_set_vector(VirtIODevice *vdev, int n, uint16_t vector)
746 if (n < VIRTIO_PCI_QUEUE_MAX)
747 vdev->vq[n].vector = vector;
750 VirtQueue *virtio_add_queue(VirtIODevice *vdev, int queue_size,
751 void (*handle_output)(VirtIODevice *, VirtQueue *))
753 int i;
755 for (i = 0; i < VIRTIO_PCI_QUEUE_MAX; i++) {
756 if (vdev->vq[i].vring.num == 0)
757 break;
760 if (i == VIRTIO_PCI_QUEUE_MAX || queue_size > VIRTQUEUE_MAX_SIZE)
761 abort();
763 vdev->vq[i].vring.num = queue_size;
764 vdev->vq[i].vring.align = VIRTIO_PCI_VRING_ALIGN;
765 vdev->vq[i].handle_output = handle_output;
767 return &vdev->vq[i];
770 void virtio_del_queue(VirtIODevice *vdev, int n)
772 if (n < 0 || n >= VIRTIO_PCI_QUEUE_MAX) {
773 abort();
776 vdev->vq[n].vring.num = 0;
779 void virtio_irq(VirtQueue *vq)
781 trace_virtio_irq(vq);
782 vq->vdev->isr |= 0x01;
783 virtio_notify_vector(vq->vdev, vq->vector);
786 /* Assuming a given event_idx value from the other size, if
787 * we have just incremented index from old to new_idx,
788 * should we trigger an event? */
789 static inline int vring_need_event(uint16_t event, uint16_t new, uint16_t old)
791 /* Note: Xen has similar logic for notification hold-off
792 * in include/xen/interface/io/ring.h with req_event and req_prod
793 * corresponding to event_idx + 1 and new respectively.
794 * Note also that req_event and req_prod in Xen start at 1,
795 * event indexes in virtio start at 0. */
796 return (uint16_t)(new - event - 1) < (uint16_t)(new - old);
799 static bool vring_notify(VirtIODevice *vdev, VirtQueue *vq)
801 uint16_t old, new;
802 bool v;
803 /* We need to expose used array entries before checking used event. */
804 smp_mb();
805 /* Always notify when queue is empty (when feature acknowledge) */
806 if (((vdev->guest_features & (1 << VIRTIO_F_NOTIFY_ON_EMPTY)) &&
807 !vq->inuse && vring_avail_idx(vq) == vq->last_avail_idx)) {
808 return true;
811 if (!(vdev->guest_features & (1 << VIRTIO_RING_F_EVENT_IDX))) {
812 return !(vring_avail_flags(vq) & VRING_AVAIL_F_NO_INTERRUPT);
815 v = vq->signalled_used_valid;
816 vq->signalled_used_valid = true;
817 old = vq->signalled_used;
818 new = vq->signalled_used = vring_used_idx(vq);
819 return !v || vring_need_event(vring_used_event(vq), new, old);
822 void virtio_notify(VirtIODevice *vdev, VirtQueue *vq)
824 if (!vring_notify(vdev, vq)) {
825 return;
828 trace_virtio_notify(vdev, vq);
829 vdev->isr |= 0x01;
830 virtio_notify_vector(vdev, vq->vector);
833 void virtio_notify_config(VirtIODevice *vdev)
835 if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK))
836 return;
838 vdev->isr |= 0x03;
839 virtio_notify_vector(vdev, vdev->config_vector);
842 void virtio_save(VirtIODevice *vdev, QEMUFile *f)
844 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
845 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
846 VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev);
847 int i;
849 if (k->save_config) {
850 k->save_config(qbus->parent, f);
853 qemu_put_8s(f, &vdev->status);
854 qemu_put_8s(f, &vdev->isr);
855 qemu_put_be16s(f, &vdev->queue_sel);
856 qemu_put_be32s(f, &vdev->guest_features);
857 qemu_put_be32(f, vdev->config_len);
858 qemu_put_buffer(f, vdev->config, vdev->config_len);
860 for (i = 0; i < VIRTIO_PCI_QUEUE_MAX; i++) {
861 if (vdev->vq[i].vring.num == 0)
862 break;
865 qemu_put_be32(f, i);
867 for (i = 0; i < VIRTIO_PCI_QUEUE_MAX; i++) {
868 if (vdev->vq[i].vring.num == 0)
869 break;
871 qemu_put_be32(f, vdev->vq[i].vring.num);
872 if (k->has_variable_vring_alignment) {
873 qemu_put_be32(f, vdev->vq[i].vring.align);
875 qemu_put_be64(f, vdev->vq[i].pa);
876 qemu_put_be16s(f, &vdev->vq[i].last_avail_idx);
877 if (k->save_queue) {
878 k->save_queue(qbus->parent, i, f);
882 if (vdc->save != NULL) {
883 vdc->save(vdev, f);
887 int virtio_set_features(VirtIODevice *vdev, uint32_t val)
889 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
890 VirtioBusClass *vbusk = VIRTIO_BUS_GET_CLASS(qbus);
891 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
892 uint32_t supported_features = vbusk->get_features(qbus->parent);
893 bool bad = (val & ~supported_features) != 0;
895 val &= supported_features;
896 if (k->set_features) {
897 k->set_features(vdev, val);
899 vdev->guest_features = val;
900 return bad ? -1 : 0;
903 int virtio_load(VirtIODevice *vdev, QEMUFile *f, int version_id)
905 int i, ret;
906 int32_t config_len;
907 uint32_t num;
908 uint32_t features;
909 uint32_t supported_features;
910 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
911 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
912 VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev);
914 if (k->load_config) {
915 ret = k->load_config(qbus->parent, f);
916 if (ret)
917 return ret;
920 qemu_get_8s(f, &vdev->status);
921 qemu_get_8s(f, &vdev->isr);
922 qemu_get_be16s(f, &vdev->queue_sel);
923 if (vdev->queue_sel >= VIRTIO_PCI_QUEUE_MAX) {
924 return -1;
926 qemu_get_be32s(f, &features);
928 if (virtio_set_features(vdev, features) < 0) {
929 supported_features = k->get_features(qbus->parent);
930 error_report("Features 0x%x unsupported. Allowed features: 0x%x",
931 features, supported_features);
932 return -1;
934 config_len = qemu_get_be32(f);
937 * There are cases where the incoming config can be bigger or smaller
938 * than what we have; so load what we have space for, and skip
939 * any excess that's in the stream.
941 qemu_get_buffer(f, vdev->config, MIN(config_len, vdev->config_len));
943 while (config_len > vdev->config_len) {
944 qemu_get_byte(f);
945 config_len--;
948 num = qemu_get_be32(f);
950 if (num > VIRTIO_PCI_QUEUE_MAX) {
951 error_report("Invalid number of PCI queues: 0x%x", num);
952 return -1;
955 for (i = 0; i < num; i++) {
956 vdev->vq[i].vring.num = qemu_get_be32(f);
957 if (k->has_variable_vring_alignment) {
958 vdev->vq[i].vring.align = qemu_get_be32(f);
960 vdev->vq[i].pa = qemu_get_be64(f);
961 qemu_get_be16s(f, &vdev->vq[i].last_avail_idx);
962 vdev->vq[i].signalled_used_valid = false;
963 vdev->vq[i].notification = true;
965 if (vdev->vq[i].pa) {
966 uint16_t nheads;
967 virtqueue_init(&vdev->vq[i]);
968 nheads = vring_avail_idx(&vdev->vq[i]) - vdev->vq[i].last_avail_idx;
969 /* Check it isn't doing very strange things with descriptor numbers. */
970 if (nheads > vdev->vq[i].vring.num) {
971 error_report("VQ %d size 0x%x Guest index 0x%x "
972 "inconsistent with Host index 0x%x: delta 0x%x",
973 i, vdev->vq[i].vring.num,
974 vring_avail_idx(&vdev->vq[i]),
975 vdev->vq[i].last_avail_idx, nheads);
976 return -1;
978 } else if (vdev->vq[i].last_avail_idx) {
979 error_report("VQ %d address 0x0 "
980 "inconsistent with Host index 0x%x",
981 i, vdev->vq[i].last_avail_idx);
982 return -1;
984 if (k->load_queue) {
985 ret = k->load_queue(qbus->parent, i, f);
986 if (ret)
987 return ret;
991 virtio_notify_vector(vdev, VIRTIO_NO_VECTOR);
993 if (vdc->load != NULL) {
994 return vdc->load(vdev, f, version_id);
997 return 0;
1000 void virtio_cleanup(VirtIODevice *vdev)
1002 qemu_del_vm_change_state_handler(vdev->vmstate);
1003 g_free(vdev->config);
1004 g_free(vdev->vq);
1007 static void virtio_vmstate_change(void *opaque, int running, RunState state)
1009 VirtIODevice *vdev = opaque;
1010 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
1011 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
1012 bool backend_run = running && (vdev->status & VIRTIO_CONFIG_S_DRIVER_OK);
1013 vdev->vm_running = running;
1015 if (backend_run) {
1016 virtio_set_status(vdev, vdev->status);
1019 if (k->vmstate_change) {
1020 k->vmstate_change(qbus->parent, backend_run);
1023 if (!backend_run) {
1024 virtio_set_status(vdev, vdev->status);
1028 void virtio_init(VirtIODevice *vdev, const char *name,
1029 uint16_t device_id, size_t config_size)
1031 int i;
1032 vdev->device_id = device_id;
1033 vdev->status = 0;
1034 vdev->isr = 0;
1035 vdev->queue_sel = 0;
1036 vdev->config_vector = VIRTIO_NO_VECTOR;
1037 vdev->vq = g_malloc0(sizeof(VirtQueue) * VIRTIO_PCI_QUEUE_MAX);
1038 vdev->vm_running = runstate_is_running();
1039 for (i = 0; i < VIRTIO_PCI_QUEUE_MAX; i++) {
1040 vdev->vq[i].vector = VIRTIO_NO_VECTOR;
1041 vdev->vq[i].vdev = vdev;
1042 vdev->vq[i].queue_index = i;
1045 vdev->name = name;
1046 vdev->config_len = config_size;
1047 if (vdev->config_len) {
1048 vdev->config = g_malloc0(config_size);
1049 } else {
1050 vdev->config = NULL;
1052 vdev->vmstate = qemu_add_vm_change_state_handler(virtio_vmstate_change,
1053 vdev);
1056 hwaddr virtio_queue_get_desc_addr(VirtIODevice *vdev, int n)
1058 return vdev->vq[n].vring.desc;
1061 hwaddr virtio_queue_get_avail_addr(VirtIODevice *vdev, int n)
1063 return vdev->vq[n].vring.avail;
1066 hwaddr virtio_queue_get_used_addr(VirtIODevice *vdev, int n)
1068 return vdev->vq[n].vring.used;
1071 hwaddr virtio_queue_get_ring_addr(VirtIODevice *vdev, int n)
1073 return vdev->vq[n].vring.desc;
1076 hwaddr virtio_queue_get_desc_size(VirtIODevice *vdev, int n)
1078 return sizeof(VRingDesc) * vdev->vq[n].vring.num;
1081 hwaddr virtio_queue_get_avail_size(VirtIODevice *vdev, int n)
1083 return offsetof(VRingAvail, ring) +
1084 sizeof(uint64_t) * vdev->vq[n].vring.num;
1087 hwaddr virtio_queue_get_used_size(VirtIODevice *vdev, int n)
1089 return offsetof(VRingUsed, ring) +
1090 sizeof(VRingUsedElem) * vdev->vq[n].vring.num;
1093 hwaddr virtio_queue_get_ring_size(VirtIODevice *vdev, int n)
1095 return vdev->vq[n].vring.used - vdev->vq[n].vring.desc +
1096 virtio_queue_get_used_size(vdev, n);
1099 uint16_t virtio_queue_get_last_avail_idx(VirtIODevice *vdev, int n)
1101 return vdev->vq[n].last_avail_idx;
1104 void virtio_queue_set_last_avail_idx(VirtIODevice *vdev, int n, uint16_t idx)
1106 vdev->vq[n].last_avail_idx = idx;
1109 void virtio_queue_invalidate_signalled_used(VirtIODevice *vdev, int n)
1111 vdev->vq[n].signalled_used_valid = false;
1114 VirtQueue *virtio_get_queue(VirtIODevice *vdev, int n)
1116 return vdev->vq + n;
1119 uint16_t virtio_get_queue_index(VirtQueue *vq)
1121 return vq->queue_index;
1124 static void virtio_queue_guest_notifier_read(EventNotifier *n)
1126 VirtQueue *vq = container_of(n, VirtQueue, guest_notifier);
1127 if (event_notifier_test_and_clear(n)) {
1128 virtio_irq(vq);
1132 void virtio_queue_set_guest_notifier_fd_handler(VirtQueue *vq, bool assign,
1133 bool with_irqfd)
1135 if (assign && !with_irqfd) {
1136 event_notifier_set_handler(&vq->guest_notifier,
1137 virtio_queue_guest_notifier_read);
1138 } else {
1139 event_notifier_set_handler(&vq->guest_notifier, NULL);
1141 if (!assign) {
1142 /* Test and clear notifier before closing it,
1143 * in case poll callback didn't have time to run. */
1144 virtio_queue_guest_notifier_read(&vq->guest_notifier);
1148 EventNotifier *virtio_queue_get_guest_notifier(VirtQueue *vq)
1150 return &vq->guest_notifier;
1153 static void virtio_queue_host_notifier_read(EventNotifier *n)
1155 VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
1156 if (event_notifier_test_and_clear(n)) {
1157 virtio_queue_notify_vq(vq);
1161 void virtio_queue_set_host_notifier_fd_handler(VirtQueue *vq, bool assign,
1162 bool set_handler)
1164 if (assign && set_handler) {
1165 event_notifier_set_handler(&vq->host_notifier,
1166 virtio_queue_host_notifier_read);
1167 } else {
1168 event_notifier_set_handler(&vq->host_notifier, NULL);
1170 if (!assign) {
1171 /* Test and clear notifier before after disabling event,
1172 * in case poll callback didn't have time to run. */
1173 virtio_queue_host_notifier_read(&vq->host_notifier);
1177 EventNotifier *virtio_queue_get_host_notifier(VirtQueue *vq)
1179 return &vq->host_notifier;
1182 void virtio_device_set_child_bus_name(VirtIODevice *vdev, char *bus_name)
1184 g_free(vdev->bus_name);
1185 vdev->bus_name = g_strdup(bus_name);
1188 static void virtio_device_realize(DeviceState *dev, Error **errp)
1190 VirtIODevice *vdev = VIRTIO_DEVICE(dev);
1191 VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(dev);
1192 Error *err = NULL;
1194 if (vdc->realize != NULL) {
1195 vdc->realize(dev, &err);
1196 if (err != NULL) {
1197 error_propagate(errp, err);
1198 return;
1201 virtio_bus_device_plugged(vdev);
1204 static void virtio_device_unrealize(DeviceState *dev, Error **errp)
1206 VirtIODevice *vdev = VIRTIO_DEVICE(dev);
1207 VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(dev);
1208 Error *err = NULL;
1210 virtio_bus_device_unplugged(vdev);
1212 if (vdc->unrealize != NULL) {
1213 vdc->unrealize(dev, &err);
1214 if (err != NULL) {
1215 error_propagate(errp, err);
1216 return;
1220 g_free(vdev->bus_name);
1221 vdev->bus_name = NULL;
1224 static void virtio_device_class_init(ObjectClass *klass, void *data)
1226 /* Set the default value here. */
1227 DeviceClass *dc = DEVICE_CLASS(klass);
1229 dc->realize = virtio_device_realize;
1230 dc->unrealize = virtio_device_unrealize;
1231 dc->bus_type = TYPE_VIRTIO_BUS;
1234 static const TypeInfo virtio_device_info = {
1235 .name = TYPE_VIRTIO_DEVICE,
1236 .parent = TYPE_DEVICE,
1237 .instance_size = sizeof(VirtIODevice),
1238 .class_init = virtio_device_class_init,
1239 .abstract = true,
1240 .class_size = sizeof(VirtioDeviceClass),
1243 static void virtio_register_types(void)
1245 type_register_static(&virtio_device_info);
1248 type_init(virtio_register_types)