4 * Copyright IBM, Corp. 2007
9 * This work is licensed under the terms of the GNU GPL, version 2. See
10 * the COPYING file in the top-level directory.
19 //#define VIRTIO_ZERO_COPY
21 /* from Linux's linux/virtio_pci.h */
23 /* A 32-bit r/o bitmask of the features supported by the host */
24 #define VIRTIO_PCI_HOST_FEATURES 0
26 /* A 32-bit r/w bitmask of features activated by the guest */
27 #define VIRTIO_PCI_GUEST_FEATURES 4
29 /* A 32-bit r/w PFN for the currently selected queue */
30 #define VIRTIO_PCI_QUEUE_PFN 8
32 /* A 16-bit r/o queue size for the currently selected queue */
33 #define VIRTIO_PCI_QUEUE_NUM 12
35 /* A 16-bit r/w queue selector */
36 #define VIRTIO_PCI_QUEUE_SEL 14
38 /* A 16-bit r/w queue notifier */
39 #define VIRTIO_PCI_QUEUE_NOTIFY 16
41 /* An 8-bit device status register. */
42 #define VIRTIO_PCI_STATUS 18
44 /* An 8-bit r/o interrupt status register. Reading the value will return the
45 * current contents of the ISR and will also clear it. This is effectively
46 * a read-and-acknowledge. */
47 #define VIRTIO_PCI_ISR 19
49 #define VIRTIO_PCI_CONFIG 20
51 /* Virtio ABI version, if we increment this, we break the guest driver. */
52 #define VIRTIO_PCI_ABI_VERSION 0
54 /* How many bits to shift physical queue address written to QUEUE_PFN.
55 * 12 is historical, and due to x86 page size. */
56 #define VIRTIO_PCI_QUEUE_ADDR_SHIFT 12
58 /* The alignment to use between consumer and producer parts of vring.
59 * x86 pagesize again. */
60 #define VIRTIO_PCI_VRING_ALIGN 4096
62 /* QEMU doesn't strictly need write barriers since everything runs in
63 * lock-step. We'll leave the calls to wmb() in though to make it obvious for
64 * KVM or if kqemu gets SMP support.
66 #define wmb() do { } while (0)
68 typedef struct VRingDesc
76 typedef struct VRingAvail
83 typedef struct VRingUsedElem
89 typedef struct VRingUsed
93 VRingUsedElem ring[0];
99 target_phys_addr_t desc;
100 target_phys_addr_t avail;
101 target_phys_addr_t used;
108 uint16_t last_avail_idx;
110 void (*handle_output)(VirtIODevice *vdev, VirtQueue *vq);
113 #define VIRTIO_PCI_QUEUE_MAX 16
115 /* virt queue functions */
116 #ifdef VIRTIO_ZERO_COPY
117 static void *virtio_map_gpa(target_phys_addr_t addr, size_t size)
120 target_phys_addr_t addr1;
122 off = cpu_get_physical_page_desc(addr);
123 if ((off & ~TARGET_PAGE_MASK) != IO_MEM_RAM) {
124 fprintf(stderr, "virtio DMA to IO ram\n");
128 off = (off & TARGET_PAGE_MASK) | (addr & ~TARGET_PAGE_MASK);
130 for (addr1 = addr + TARGET_PAGE_SIZE;
131 addr1 < TARGET_PAGE_ALIGN(addr + size);
132 addr1 += TARGET_PAGE_SIZE) {
135 off1 = cpu_get_physical_page_desc(addr1);
136 if ((off1 & ~TARGET_PAGE_MASK) != IO_MEM_RAM) {
137 fprintf(stderr, "virtio DMA to IO ram\n");
141 off1 = (off1 & TARGET_PAGE_MASK) | (addr1 & ~TARGET_PAGE_MASK);
143 if (off1 != (off + (addr1 - addr))) {
144 fprintf(stderr, "discontigous virtio memory\n");
149 return phys_ram_base + off;
153 static void virtqueue_init(VirtQueue *vq, target_phys_addr_t pa)
156 vq->vring.avail = pa + vq->vring.num * sizeof(VRingDesc);
157 vq->vring.used = vring_align(vq->vring.avail +
158 offsetof(VRingAvail, ring[vq->vring.num]),
159 VIRTIO_PCI_VRING_ALIGN);
162 static inline uint64_t vring_desc_addr(VirtQueue *vq, int i)
164 target_phys_addr_t pa;
165 pa = vq->vring.desc + sizeof(VRingDesc) * i + offsetof(VRingDesc, addr);
169 static inline uint32_t vring_desc_len(VirtQueue *vq, int i)
171 target_phys_addr_t pa;
172 pa = vq->vring.desc + sizeof(VRingDesc) * i + offsetof(VRingDesc, len);
176 static inline uint16_t vring_desc_flags(VirtQueue *vq, int i)
178 target_phys_addr_t pa;
179 pa = vq->vring.desc + sizeof(VRingDesc) * i + offsetof(VRingDesc, flags);
180 return lduw_phys(pa);
183 static inline uint16_t vring_desc_next(VirtQueue *vq, int i)
185 target_phys_addr_t pa;
186 pa = vq->vring.desc + sizeof(VRingDesc) * i + offsetof(VRingDesc, next);
187 return lduw_phys(pa);
190 static inline uint16_t vring_avail_flags(VirtQueue *vq)
192 target_phys_addr_t pa;
193 pa = vq->vring.avail + offsetof(VRingAvail, flags);
194 return lduw_phys(pa);
197 static inline uint16_t vring_avail_idx(VirtQueue *vq)
199 target_phys_addr_t pa;
200 pa = vq->vring.avail + offsetof(VRingAvail, idx);
201 return lduw_phys(pa);
204 static inline uint16_t vring_avail_ring(VirtQueue *vq, int i)
206 target_phys_addr_t pa;
207 pa = vq->vring.avail + offsetof(VRingAvail, ring[i]);
208 return lduw_phys(pa);
211 static inline void vring_used_ring_id(VirtQueue *vq, int i, uint32_t val)
213 target_phys_addr_t pa;
214 pa = vq->vring.used + offsetof(VRingUsed, ring[i].id);
218 static inline void vring_used_ring_len(VirtQueue *vq, int i, uint32_t val)
220 target_phys_addr_t pa;
221 pa = vq->vring.used + offsetof(VRingUsed, ring[i].len);
225 static uint16_t vring_used_idx(VirtQueue *vq)
227 target_phys_addr_t pa;
228 pa = vq->vring.used + offsetof(VRingUsed, idx);
229 return lduw_phys(pa);
232 static inline void vring_used_idx_increment(VirtQueue *vq, uint16_t val)
234 target_phys_addr_t pa;
235 pa = vq->vring.used + offsetof(VRingUsed, idx);
236 stw_phys(pa, vring_used_idx(vq) + val);
239 static inline void vring_used_flags_set_bit(VirtQueue *vq, int mask)
241 target_phys_addr_t pa;
242 pa = vq->vring.used + offsetof(VRingUsed, flags);
243 stw_phys(pa, lduw_phys(pa) | mask);
246 static inline void vring_used_flags_unset_bit(VirtQueue *vq, int mask)
248 target_phys_addr_t pa;
249 pa = vq->vring.used + offsetof(VRingUsed, flags);
250 stw_phys(pa, lduw_phys(pa) & ~mask);
253 void virtio_queue_set_notification(VirtQueue *vq, int enable)
256 vring_used_flags_unset_bit(vq, VRING_USED_F_NO_NOTIFY);
258 vring_used_flags_set_bit(vq, VRING_USED_F_NO_NOTIFY);
261 int virtio_queue_ready(VirtQueue *vq)
263 return vq->vring.avail != 0;
266 int virtio_queue_empty(VirtQueue *vq)
268 return vring_avail_idx(vq) == vq->last_avail_idx;
271 void virtqueue_fill(VirtQueue *vq, const VirtQueueElement *elem,
272 unsigned int len, unsigned int idx)
277 #ifndef VIRTIO_ZERO_COPY
278 for (i = 0; i < elem->out_num; i++)
279 qemu_free(elem->out_sg[i].iov_base);
283 for (i = 0; i < elem->in_num; i++) {
284 size_t size = MIN(len - offset, elem->in_sg[i].iov_len);
286 #ifdef VIRTIO_ZERO_COPY
288 ram_addr_t addr = (uint8_t *)elem->in_sg[i].iov_base - phys_ram_base;
291 for (off = 0; off < size; off += TARGET_PAGE_SIZE)
292 cpu_physical_memory_set_dirty(addr + off);
296 cpu_physical_memory_write(elem->in_addr[i],
297 elem->in_sg[i].iov_base,
300 qemu_free(elem->in_sg[i].iov_base);
306 idx = (idx + vring_used_idx(vq)) % vq->vring.num;
308 /* Get a pointer to the next entry in the used ring. */
309 vring_used_ring_id(vq, idx, elem->index);
310 vring_used_ring_len(vq, idx, len);
313 void virtqueue_flush(VirtQueue *vq, unsigned int count)
315 /* Make sure buffer is written before we update index. */
317 vring_used_idx_increment(vq, count);
321 void virtqueue_push(VirtQueue *vq, const VirtQueueElement *elem,
324 virtqueue_fill(vq, elem, len, 0);
325 virtqueue_flush(vq, 1);
328 static int virtqueue_num_heads(VirtQueue *vq, unsigned int idx)
330 uint16_t num_heads = vring_avail_idx(vq) - idx;
332 /* Check it isn't doing very strange things with descriptor numbers. */
333 if (num_heads > vq->vring.num) {
334 fprintf(stderr, "Guest moved used index from %u to %u",
335 idx, vring_avail_idx(vq));
342 static unsigned int virtqueue_get_head(VirtQueue *vq, unsigned int idx)
346 /* Grab the next descriptor number they're advertising, and increment
347 * the index we've seen. */
348 head = vring_avail_ring(vq, idx % vq->vring.num);
350 /* If their number is silly, that's a fatal mistake. */
351 if (head >= vq->vring.num) {
352 fprintf(stderr, "Guest says index %u is available", head);
359 static unsigned virtqueue_next_desc(VirtQueue *vq, unsigned int i)
363 /* If this descriptor says it doesn't chain, we're done. */
364 if (!(vring_desc_flags(vq, i) & VRING_DESC_F_NEXT))
365 return vq->vring.num;
367 /* Check they're not leading us off end of descriptors. */
368 next = vring_desc_next(vq, i);
369 /* Make sure compiler knows to grab that: we don't want it changing! */
372 if (next >= vq->vring.num) {
373 fprintf(stderr, "Desc next is %u", next);
380 int virtqueue_avail_bytes(VirtQueue *vq, int in_bytes, int out_bytes)
383 int num_bufs, in_total, out_total;
385 idx = vq->last_avail_idx;
387 num_bufs = in_total = out_total = 0;
388 while (virtqueue_num_heads(vq, idx)) {
391 i = virtqueue_get_head(vq, idx++);
393 /* If we've got too many, that implies a descriptor loop. */
394 if (++num_bufs > vq->vring.num) {
395 fprintf(stderr, "Looped descriptor");
399 if (vring_desc_flags(vq, i) & VRING_DESC_F_WRITE) {
401 (in_total += vring_desc_len(vq, i)) >= in_bytes)
405 (out_total += vring_desc_len(vq, i)) >= out_bytes)
408 } while ((i = virtqueue_next_desc(vq, i)) != vq->vring.num);
414 int virtqueue_pop(VirtQueue *vq, VirtQueueElement *elem)
416 unsigned int i, head;
418 if (!virtqueue_num_heads(vq, vq->last_avail_idx))
421 /* When we start there are none of either input nor output. */
422 elem->out_num = elem->in_num = 0;
424 i = head = virtqueue_get_head(vq, vq->last_avail_idx++);
428 if (vring_desc_flags(vq, i) & VRING_DESC_F_WRITE) {
429 elem->in_addr[elem->in_num] = vring_desc_addr(vq, i);
430 sg = &elem->in_sg[elem->in_num++];
432 sg = &elem->out_sg[elem->out_num++];
434 /* Grab the first descriptor, and check it's OK. */
435 sg->iov_len = vring_desc_len(vq, i);
437 #ifdef VIRTIO_ZERO_COPY
438 sg->iov_base = virtio_map_gpa(vring_desc_addr(vq, i), sg->iov_len);
440 /* cap individual scatter element size to prevent unbounded allocations
441 of memory from the guest. Practically speaking, no virtio driver
442 will ever pass more than a page in each element. We set the cap to
443 be 2MB in case for some reason a large page makes it way into the
444 sg list. When we implement a zero copy API, this limitation will
446 if (sg->iov_len > (2 << 20))
447 sg->iov_len = 2 << 20;
449 sg->iov_base = qemu_malloc(sg->iov_len);
451 !(vring_desc_flags(vq, i) & VRING_DESC_F_WRITE)) {
452 cpu_physical_memory_read(vring_desc_addr(vq, i),
457 if (sg->iov_base == NULL) {
458 fprintf(stderr, "Invalid mapping\n");
462 /* If we've got too many, that implies a descriptor loop. */
463 if ((elem->in_num + elem->out_num) > vq->vring.num) {
464 fprintf(stderr, "Looped descriptor");
467 } while ((i = virtqueue_next_desc(vq, i)) != vq->vring.num);
473 return elem->in_num + elem->out_num;
478 static VirtIODevice *to_virtio_device(PCIDevice *pci_dev)
480 return (VirtIODevice *)pci_dev;
483 static void virtio_update_irq(VirtIODevice *vdev)
485 qemu_set_irq(vdev->pci_dev.irq[0], vdev->isr & 1);
488 static void virtio_reset(void *opaque)
490 VirtIODevice *vdev = opaque;
500 virtio_update_irq(vdev);
502 for(i = 0; i < VIRTIO_PCI_QUEUE_MAX; i++) {
503 vdev->vq[i].vring.desc = 0;
504 vdev->vq[i].vring.avail = 0;
505 vdev->vq[i].vring.used = 0;
506 vdev->vq[i].last_avail_idx = 0;
511 static void virtio_ioport_write(void *opaque, uint32_t addr, uint32_t val)
513 VirtIODevice *vdev = to_virtio_device(opaque);
519 case VIRTIO_PCI_GUEST_FEATURES:
520 if (vdev->set_features)
521 vdev->set_features(vdev, val);
522 vdev->features = val;
524 case VIRTIO_PCI_QUEUE_PFN:
525 pa = (ram_addr_t)val << VIRTIO_PCI_QUEUE_ADDR_SHIFT;
526 vdev->vq[vdev->queue_sel].pfn = val;
530 virtqueue_init(&vdev->vq[vdev->queue_sel], pa);
533 case VIRTIO_PCI_QUEUE_SEL:
534 if (val < VIRTIO_PCI_QUEUE_MAX)
535 vdev->queue_sel = val;
537 case VIRTIO_PCI_QUEUE_NOTIFY:
538 if (val < VIRTIO_PCI_QUEUE_MAX && vdev->vq[val].vring.desc)
539 vdev->vq[val].handle_output(vdev, &vdev->vq[val]);
541 case VIRTIO_PCI_STATUS:
542 vdev->status = val & 0xFF;
543 if (vdev->status == 0)
549 static uint32_t virtio_ioport_read(void *opaque, uint32_t addr)
551 VirtIODevice *vdev = to_virtio_device(opaque);
552 uint32_t ret = 0xFFFFFFFF;
557 case VIRTIO_PCI_HOST_FEATURES:
558 ret = vdev->get_features(vdev);
559 ret |= (1 << VIRTIO_F_NOTIFY_ON_EMPTY);
561 case VIRTIO_PCI_GUEST_FEATURES:
562 ret = vdev->features;
564 case VIRTIO_PCI_QUEUE_PFN:
565 ret = vdev->vq[vdev->queue_sel].pfn;
567 case VIRTIO_PCI_QUEUE_NUM:
568 ret = vdev->vq[vdev->queue_sel].vring.num;
570 case VIRTIO_PCI_QUEUE_SEL:
571 ret = vdev->queue_sel;
573 case VIRTIO_PCI_STATUS:
577 /* reading from the ISR also clears it. */
580 virtio_update_irq(vdev);
589 static uint32_t virtio_config_readb(void *opaque, uint32_t addr)
591 VirtIODevice *vdev = opaque;
594 vdev->get_config(vdev, vdev->config);
596 addr -= vdev->addr + VIRTIO_PCI_CONFIG;
597 if (addr > (vdev->config_len - sizeof(val)))
600 memcpy(&val, vdev->config + addr, sizeof(val));
604 static uint32_t virtio_config_readw(void *opaque, uint32_t addr)
606 VirtIODevice *vdev = opaque;
609 vdev->get_config(vdev, vdev->config);
611 addr -= vdev->addr + VIRTIO_PCI_CONFIG;
612 if (addr > (vdev->config_len - sizeof(val)))
615 memcpy(&val, vdev->config + addr, sizeof(val));
619 static uint32_t virtio_config_readl(void *opaque, uint32_t addr)
621 VirtIODevice *vdev = opaque;
624 vdev->get_config(vdev, vdev->config);
626 addr -= vdev->addr + VIRTIO_PCI_CONFIG;
627 if (addr > (vdev->config_len - sizeof(val)))
630 memcpy(&val, vdev->config + addr, sizeof(val));
634 static void virtio_config_writeb(void *opaque, uint32_t addr, uint32_t data)
636 VirtIODevice *vdev = opaque;
639 addr -= vdev->addr + VIRTIO_PCI_CONFIG;
640 if (addr > (vdev->config_len - sizeof(val)))
643 memcpy(vdev->config + addr, &val, sizeof(val));
645 if (vdev->set_config)
646 vdev->set_config(vdev, vdev->config);
649 static void virtio_config_writew(void *opaque, uint32_t addr, uint32_t data)
651 VirtIODevice *vdev = opaque;
654 addr -= vdev->addr + VIRTIO_PCI_CONFIG;
655 if (addr > (vdev->config_len - sizeof(val)))
658 memcpy(vdev->config + addr, &val, sizeof(val));
660 if (vdev->set_config)
661 vdev->set_config(vdev, vdev->config);
664 static void virtio_config_writel(void *opaque, uint32_t addr, uint32_t data)
666 VirtIODevice *vdev = opaque;
669 addr -= vdev->addr + VIRTIO_PCI_CONFIG;
670 if (addr > (vdev->config_len - sizeof(val)))
673 memcpy(vdev->config + addr, &val, sizeof(val));
675 if (vdev->set_config)
676 vdev->set_config(vdev, vdev->config);
679 static void virtio_map(PCIDevice *pci_dev, int region_num,
680 uint32_t addr, uint32_t size, int type)
682 VirtIODevice *vdev = to_virtio_device(pci_dev);
686 for (i = 0; i < 3; i++) {
687 register_ioport_write(addr, 20, 1 << i, virtio_ioport_write, vdev);
688 register_ioport_read(addr, 20, 1 << i, virtio_ioport_read, vdev);
691 if (vdev->config_len) {
692 register_ioport_write(addr + 20, vdev->config_len, 1,
693 virtio_config_writeb, vdev);
694 register_ioport_write(addr + 20, vdev->config_len, 2,
695 virtio_config_writew, vdev);
696 register_ioport_write(addr + 20, vdev->config_len, 4,
697 virtio_config_writel, vdev);
698 register_ioport_read(addr + 20, vdev->config_len, 1,
699 virtio_config_readb, vdev);
700 register_ioport_read(addr + 20, vdev->config_len, 2,
701 virtio_config_readw, vdev);
702 register_ioport_read(addr + 20, vdev->config_len, 4,
703 virtio_config_readl, vdev);
705 vdev->get_config(vdev, vdev->config);
709 VirtQueue *virtio_add_queue(VirtIODevice *vdev, int queue_size,
710 void (*handle_output)(VirtIODevice *, VirtQueue *))
714 for (i = 0; i < VIRTIO_PCI_QUEUE_MAX; i++) {
715 if (vdev->vq[i].vring.num == 0)
719 if (i == VIRTIO_PCI_QUEUE_MAX || queue_size > VIRTQUEUE_MAX_SIZE)
722 vdev->vq[i].vring.num = queue_size;
723 vdev->vq[i].handle_output = handle_output;
728 void virtio_notify(VirtIODevice *vdev, VirtQueue *vq)
730 /* Always notify when queue is empty */
731 if ((vq->inuse || vring_avail_idx(vq) != vq->last_avail_idx) &&
732 (vring_avail_flags(vq) & VRING_AVAIL_F_NO_INTERRUPT))
736 virtio_update_irq(vdev);
739 void virtio_notify_config(VirtIODevice *vdev)
741 if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK))
745 virtio_update_irq(vdev);
748 void virtio_save(VirtIODevice *vdev, QEMUFile *f)
752 pci_device_save(&vdev->pci_dev, f);
754 qemu_put_be32s(f, &vdev->addr);
755 qemu_put_8s(f, &vdev->status);
756 qemu_put_8s(f, &vdev->isr);
757 qemu_put_be16s(f, &vdev->queue_sel);
758 qemu_put_be32s(f, &vdev->features);
759 qemu_put_be32(f, vdev->config_len);
760 qemu_put_buffer(f, vdev->config, vdev->config_len);
762 for (i = 0; i < VIRTIO_PCI_QUEUE_MAX; i++) {
763 if (vdev->vq[i].vring.num == 0)
769 for (i = 0; i < VIRTIO_PCI_QUEUE_MAX; i++) {
770 if (vdev->vq[i].vring.num == 0)
773 qemu_put_be32(f, vdev->vq[i].vring.num);
774 qemu_put_be32s(f, &vdev->vq[i].pfn);
775 qemu_put_be16s(f, &vdev->vq[i].last_avail_idx);
779 void virtio_load(VirtIODevice *vdev, QEMUFile *f)
783 pci_device_load(&vdev->pci_dev, f);
785 qemu_get_be32s(f, &vdev->addr);
786 qemu_get_8s(f, &vdev->status);
787 qemu_get_8s(f, &vdev->isr);
788 qemu_get_be16s(f, &vdev->queue_sel);
789 qemu_get_be32s(f, &vdev->features);
790 vdev->config_len = qemu_get_be32(f);
791 qemu_get_buffer(f, vdev->config, vdev->config_len);
793 num = qemu_get_be32(f);
795 for (i = 0; i < num; i++) {
796 vdev->vq[i].vring.num = qemu_get_be32(f);
797 qemu_get_be32s(f, &vdev->vq[i].pfn);
798 qemu_get_be16s(f, &vdev->vq[i].last_avail_idx);
800 if (vdev->vq[i].pfn) {
801 target_phys_addr_t pa;
803 pa = (ram_addr_t)vdev->vq[i].pfn << VIRTIO_PCI_QUEUE_ADDR_SHIFT;
804 virtqueue_init(&vdev->vq[i], pa);
808 virtio_update_irq(vdev);
811 VirtIODevice *virtio_init_pci(PCIBus *bus, const char *name,
812 uint16_t vendor, uint16_t device,
813 uint16_t subvendor, uint16_t subdevice,
814 uint16_t class_code, uint8_t pif,
815 size_t config_size, size_t struct_size)
822 pci_dev = pci_register_device(bus, name, struct_size,
827 vdev = to_virtio_device(pci_dev);
832 vdev->vq = qemu_mallocz(sizeof(VirtQueue) * VIRTIO_PCI_QUEUE_MAX);
834 config = pci_dev->config;
835 pci_config_set_vendor_id(config, vendor);
836 pci_config_set_device_id(config, device);
838 config[0x08] = VIRTIO_PCI_ABI_VERSION;
841 pci_config_set_class(config, class_code);
844 config[0x2c] = subvendor & 0xFF;
845 config[0x2d] = (subvendor >> 8) & 0xFF;
846 config[0x2e] = subdevice & 0xFF;
847 config[0x2f] = (subdevice >> 8) & 0xFF;
852 vdev->config_len = config_size;
853 if (vdev->config_len)
854 vdev->config = qemu_mallocz(config_size);
858 size = 20 + config_size;
860 size = 1 << qemu_fls(size);
862 pci_register_io_region(pci_dev, 0, size, PCI_ADDRESS_SPACE_IO,
864 qemu_register_reset(virtio_reset, vdev);