2 * generic functions used by VFIO devices
4 * Copyright Red Hat, Inc. 2012
7 * Alex Williamson <alex.williamson@redhat.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.
12 * Based on qemu-kvm device-assignment:
13 * Adapted for KVM by Qumranet.
14 * Copyright (c) 2007, Neocleus, Alex Novik (alex@neocleus.com)
15 * Copyright (c) 2007, Neocleus, Guy Zana (guy@neocleus.com)
16 * Copyright (C) 2008, Qumranet, Amit Shah (amit.shah@qumranet.com)
17 * Copyright (C) 2008, Red Hat, Amit Shah (amit.shah@redhat.com)
18 * Copyright (C) 2008, IBM, Muli Ben-Yehuda (muli@il.ibm.com)
21 #include <sys/ioctl.h>
23 #include <linux/vfio.h>
25 #include "hw/vfio/vfio-common.h"
26 #include "hw/vfio/vfio.h"
27 #include "exec/address-spaces.h"
28 #include "exec/memory.h"
30 #include "qemu/error-report.h"
31 #include "sysemu/kvm.h"
34 struct vfio_group_head vfio_group_list
=
35 QLIST_HEAD_INITIALIZER(vfio_group_list
);
36 struct vfio_as_head vfio_address_spaces
=
37 QLIST_HEAD_INITIALIZER(vfio_address_spaces
);
41 * We have a single VFIO pseudo device per KVM VM. Once created it lives
42 * for the life of the VM. Closing the file descriptor only drops our
43 * reference to it and the device's reference to kvm. Therefore once
44 * initialized, this file descriptor is only released on QEMU exit and
45 * we'll re-use it should another vfio device be attached before then.
47 static int vfio_kvm_device_fd
= -1;
51 * Common VFIO interrupt disable
53 void vfio_disable_irqindex(VFIODevice
*vbasedev
, int index
)
55 struct vfio_irq_set irq_set
= {
56 .argsz
= sizeof(irq_set
),
57 .flags
= VFIO_IRQ_SET_DATA_NONE
| VFIO_IRQ_SET_ACTION_TRIGGER
,
63 ioctl(vbasedev
->fd
, VFIO_DEVICE_SET_IRQS
, &irq_set
);
66 void vfio_unmask_single_irqindex(VFIODevice
*vbasedev
, int index
)
68 struct vfio_irq_set irq_set
= {
69 .argsz
= sizeof(irq_set
),
70 .flags
= VFIO_IRQ_SET_DATA_NONE
| VFIO_IRQ_SET_ACTION_UNMASK
,
76 ioctl(vbasedev
->fd
, VFIO_DEVICE_SET_IRQS
, &irq_set
);
79 void vfio_mask_single_irqindex(VFIODevice
*vbasedev
, int index
)
81 struct vfio_irq_set irq_set
= {
82 .argsz
= sizeof(irq_set
),
83 .flags
= VFIO_IRQ_SET_DATA_NONE
| VFIO_IRQ_SET_ACTION_MASK
,
89 ioctl(vbasedev
->fd
, VFIO_DEVICE_SET_IRQS
, &irq_set
);
93 * IO Port/MMIO - Beware of the endians, VFIO is always little endian
95 void vfio_region_write(void *opaque
, hwaddr addr
,
96 uint64_t data
, unsigned size
)
98 VFIORegion
*region
= opaque
;
99 VFIODevice
*vbasedev
= region
->vbasedev
;
112 buf
.word
= cpu_to_le16(data
);
115 buf
.dword
= cpu_to_le32(data
);
118 hw_error("vfio: unsupported write size, %d bytes", size
);
122 if (pwrite(vbasedev
->fd
, &buf
, size
, region
->fd_offset
+ addr
) != size
) {
123 error_report("%s(%s:region%d+0x%"HWADDR_PRIx
", 0x%"PRIx64
125 __func__
, vbasedev
->name
, region
->nr
,
129 trace_vfio_region_write(vbasedev
->name
, region
->nr
, addr
, data
, size
);
132 * A read or write to a BAR always signals an INTx EOI. This will
133 * do nothing if not pending (including not in INTx mode). We assume
134 * that a BAR access is in response to an interrupt and that BAR
135 * accesses will service the interrupt. Unfortunately, we don't know
136 * which access will service the interrupt, so we're potentially
137 * getting quite a few host interrupts per guest interrupt.
139 vbasedev
->ops
->vfio_eoi(vbasedev
);
142 uint64_t vfio_region_read(void *opaque
,
143 hwaddr addr
, unsigned size
)
145 VFIORegion
*region
= opaque
;
146 VFIODevice
*vbasedev
= region
->vbasedev
;
155 if (pread(vbasedev
->fd
, &buf
, size
, region
->fd_offset
+ addr
) != size
) {
156 error_report("%s(%s:region%d+0x%"HWADDR_PRIx
", %d) failed: %m",
157 __func__
, vbasedev
->name
, region
->nr
,
166 data
= le16_to_cpu(buf
.word
);
169 data
= le32_to_cpu(buf
.dword
);
172 hw_error("vfio: unsupported read size, %d bytes", size
);
176 trace_vfio_region_read(vbasedev
->name
, region
->nr
, addr
, size
, data
);
178 /* Same as write above */
179 vbasedev
->ops
->vfio_eoi(vbasedev
);
184 const MemoryRegionOps vfio_region_ops
= {
185 .read
= vfio_region_read
,
186 .write
= vfio_region_write
,
187 .endianness
= DEVICE_LITTLE_ENDIAN
,
191 * DMA - Mapping and unmapping for the "type1" IOMMU interface used on x86
193 static int vfio_dma_unmap(VFIOContainer
*container
,
194 hwaddr iova
, ram_addr_t size
)
196 struct vfio_iommu_type1_dma_unmap unmap
= {
197 .argsz
= sizeof(unmap
),
203 if (ioctl(container
->fd
, VFIO_IOMMU_UNMAP_DMA
, &unmap
)) {
204 error_report("VFIO_UNMAP_DMA: %d", -errno
);
211 static int vfio_dma_map(VFIOContainer
*container
, hwaddr iova
,
212 ram_addr_t size
, void *vaddr
, bool readonly
)
214 struct vfio_iommu_type1_dma_map map
= {
215 .argsz
= sizeof(map
),
216 .flags
= VFIO_DMA_MAP_FLAG_READ
,
217 .vaddr
= (__u64
)(uintptr_t)vaddr
,
223 map
.flags
|= VFIO_DMA_MAP_FLAG_WRITE
;
227 * Try the mapping, if it fails with EBUSY, unmap the region and try
228 * again. This shouldn't be necessary, but we sometimes see it in
229 * the the VGA ROM space.
231 if (ioctl(container
->fd
, VFIO_IOMMU_MAP_DMA
, &map
) == 0 ||
232 (errno
== EBUSY
&& vfio_dma_unmap(container
, iova
, size
) == 0 &&
233 ioctl(container
->fd
, VFIO_IOMMU_MAP_DMA
, &map
) == 0)) {
237 error_report("VFIO_MAP_DMA: %d", -errno
);
241 static bool vfio_listener_skipped_section(MemoryRegionSection
*section
)
243 return (!memory_region_is_ram(section
->mr
) &&
244 !memory_region_is_iommu(section
->mr
)) ||
246 * Sizing an enabled 64-bit BAR can cause spurious mappings to
247 * addresses in the upper part of the 64-bit address space. These
248 * are never accessed by the CPU and beyond the address width of
249 * some IOMMU hardware. TODO: VFIO should tell us the IOMMU width.
251 section
->offset_within_address_space
& (1ULL << 63);
254 static void vfio_iommu_map_notify(Notifier
*n
, void *data
)
256 VFIOGuestIOMMU
*giommu
= container_of(n
, VFIOGuestIOMMU
, n
);
257 VFIOContainer
*container
= giommu
->container
;
258 IOMMUTLBEntry
*iotlb
= data
;
261 hwaddr len
= iotlb
->addr_mask
+ 1;
265 trace_vfio_iommu_map_notify(iotlb
->iova
,
266 iotlb
->iova
+ iotlb
->addr_mask
);
269 * The IOMMU TLB entry we have just covers translation through
270 * this IOMMU to its immediate target. We need to translate
271 * it the rest of the way through to memory.
273 mr
= address_space_translate(&address_space_memory
,
274 iotlb
->translated_addr
,
275 &xlat
, &len
, iotlb
->perm
& IOMMU_WO
);
276 if (!memory_region_is_ram(mr
)) {
277 error_report("iommu map to non memory area %"HWADDR_PRIx
"",
282 * Translation truncates length to the IOMMU page size,
283 * check that it did not truncate too much.
285 if (len
& iotlb
->addr_mask
) {
286 error_report("iommu has granularity incompatible with target AS");
290 if ((iotlb
->perm
& IOMMU_RW
) != IOMMU_NONE
) {
291 vaddr
= memory_region_get_ram_ptr(mr
) + xlat
;
292 ret
= vfio_dma_map(container
, iotlb
->iova
,
293 iotlb
->addr_mask
+ 1, vaddr
,
294 !(iotlb
->perm
& IOMMU_WO
) || mr
->readonly
);
296 error_report("vfio_dma_map(%p, 0x%"HWADDR_PRIx
", "
297 "0x%"HWADDR_PRIx
", %p) = %d (%m)",
298 container
, iotlb
->iova
,
299 iotlb
->addr_mask
+ 1, vaddr
, ret
);
302 ret
= vfio_dma_unmap(container
, iotlb
->iova
, iotlb
->addr_mask
+ 1);
304 error_report("vfio_dma_unmap(%p, 0x%"HWADDR_PRIx
", "
305 "0x%"HWADDR_PRIx
") = %d (%m)",
306 container
, iotlb
->iova
,
307 iotlb
->addr_mask
+ 1, ret
);
312 static void vfio_listener_region_add(MemoryListener
*listener
,
313 MemoryRegionSection
*section
)
315 VFIOContainer
*container
= container_of(listener
, VFIOContainer
,
316 iommu_data
.type1
.listener
);
322 if (vfio_listener_skipped_section(section
)) {
323 trace_vfio_listener_region_add_skip(
324 section
->offset_within_address_space
,
325 section
->offset_within_address_space
+
326 int128_get64(int128_sub(section
->size
, int128_one())));
330 if (unlikely((section
->offset_within_address_space
& ~TARGET_PAGE_MASK
) !=
331 (section
->offset_within_region
& ~TARGET_PAGE_MASK
))) {
332 error_report("%s received unaligned region", __func__
);
336 iova
= TARGET_PAGE_ALIGN(section
->offset_within_address_space
);
337 llend
= int128_make64(section
->offset_within_address_space
);
338 llend
= int128_add(llend
, section
->size
);
339 llend
= int128_and(llend
, int128_exts64(TARGET_PAGE_MASK
));
341 if (int128_ge(int128_make64(iova
), llend
)) {
345 memory_region_ref(section
->mr
);
347 if (memory_region_is_iommu(section
->mr
)) {
348 VFIOGuestIOMMU
*giommu
;
350 trace_vfio_listener_region_add_iommu(iova
,
351 int128_get64(int128_sub(llend
, int128_one())));
353 * FIXME: We should do some checking to see if the
354 * capabilities of the host VFIO IOMMU are adequate to model
357 * FIXME: For VFIO iommu types which have KVM acceleration to
358 * avoid bouncing all map/unmaps through qemu this way, this
359 * would be the right place to wire that up (tell the KVM
360 * device emulation the VFIO iommu handles to use).
363 * This assumes that the guest IOMMU is empty of
364 * mappings at this point.
366 * One way of doing this is:
367 * 1. Avoid sharing IOMMUs between emulated devices or different
369 * 2. Implement VFIO_IOMMU_ENABLE in the host kernel to fail if
370 * there are some mappings in IOMMU.
372 * VFIO on SPAPR does that. Other IOMMU models may do that different,
373 * they must make sure there are no existing mappings or
374 * loop through existing mappings to map them into VFIO.
376 giommu
= g_malloc0(sizeof(*giommu
));
377 giommu
->iommu
= section
->mr
;
378 giommu
->container
= container
;
379 giommu
->n
.notify
= vfio_iommu_map_notify
;
380 QLIST_INSERT_HEAD(&container
->giommu_list
, giommu
, giommu_next
);
381 memory_region_register_iommu_notifier(giommu
->iommu
, &giommu
->n
);
386 /* Here we assume that memory_region_is_ram(section->mr)==true */
388 end
= int128_get64(llend
);
389 vaddr
= memory_region_get_ram_ptr(section
->mr
) +
390 section
->offset_within_region
+
391 (iova
- section
->offset_within_address_space
);
393 trace_vfio_listener_region_add_ram(iova
, end
- 1, vaddr
);
395 ret
= vfio_dma_map(container
, iova
, end
- iova
, vaddr
, section
->readonly
);
397 error_report("vfio_dma_map(%p, 0x%"HWADDR_PRIx
", "
398 "0x%"HWADDR_PRIx
", %p) = %d (%m)",
399 container
, iova
, end
- iova
, vaddr
, ret
);
402 * On the initfn path, store the first error in the container so we
403 * can gracefully fail. Runtime, there's not much we can do other
404 * than throw a hardware error.
406 if (!container
->iommu_data
.type1
.initialized
) {
407 if (!container
->iommu_data
.type1
.error
) {
408 container
->iommu_data
.type1
.error
= ret
;
411 hw_error("vfio: DMA mapping failed, unable to continue");
416 static void vfio_listener_region_del(MemoryListener
*listener
,
417 MemoryRegionSection
*section
)
419 VFIOContainer
*container
= container_of(listener
, VFIOContainer
,
420 iommu_data
.type1
.listener
);
424 if (vfio_listener_skipped_section(section
)) {
425 trace_vfio_listener_region_del_skip(
426 section
->offset_within_address_space
,
427 section
->offset_within_address_space
+
428 int128_get64(int128_sub(section
->size
, int128_one())));
432 if (unlikely((section
->offset_within_address_space
& ~TARGET_PAGE_MASK
) !=
433 (section
->offset_within_region
& ~TARGET_PAGE_MASK
))) {
434 error_report("%s received unaligned region", __func__
);
438 if (memory_region_is_iommu(section
->mr
)) {
439 VFIOGuestIOMMU
*giommu
;
441 QLIST_FOREACH(giommu
, &container
->giommu_list
, giommu_next
) {
442 if (giommu
->iommu
== section
->mr
) {
443 memory_region_unregister_iommu_notifier(&giommu
->n
);
444 QLIST_REMOVE(giommu
, giommu_next
);
451 * FIXME: We assume the one big unmap below is adequate to
452 * remove any individual page mappings in the IOMMU which
453 * might have been copied into VFIO. This works for a page table
454 * based IOMMU where a big unmap flattens a large range of IO-PTEs.
455 * That may not be true for all IOMMU types.
459 iova
= TARGET_PAGE_ALIGN(section
->offset_within_address_space
);
460 end
= (section
->offset_within_address_space
+ int128_get64(section
->size
)) &
467 trace_vfio_listener_region_del(iova
, end
- 1);
469 ret
= vfio_dma_unmap(container
, iova
, end
- iova
);
470 memory_region_unref(section
->mr
);
472 error_report("vfio_dma_unmap(%p, 0x%"HWADDR_PRIx
", "
473 "0x%"HWADDR_PRIx
") = %d (%m)",
474 container
, iova
, end
- iova
, ret
);
478 static const MemoryListener vfio_memory_listener
= {
479 .region_add
= vfio_listener_region_add
,
480 .region_del
= vfio_listener_region_del
,
483 static void vfio_listener_release(VFIOContainer
*container
)
485 memory_listener_unregister(&container
->iommu_data
.type1
.listener
);
488 int vfio_mmap_region(Object
*obj
, VFIORegion
*region
,
489 MemoryRegion
*mem
, MemoryRegion
*submem
,
490 void **map
, size_t size
, off_t offset
,
494 VFIODevice
*vbasedev
= region
->vbasedev
;
496 if (vbasedev
->allow_mmap
&& size
&& region
->flags
&
497 VFIO_REGION_INFO_FLAG_MMAP
) {
500 if (region
->flags
& VFIO_REGION_INFO_FLAG_READ
) {
504 if (region
->flags
& VFIO_REGION_INFO_FLAG_WRITE
) {
508 *map
= mmap(NULL
, size
, prot
, MAP_SHARED
,
510 region
->fd_offset
+ offset
);
511 if (*map
== MAP_FAILED
) {
517 memory_region_init_ram_ptr(submem
, obj
, name
, size
, *map
);
518 memory_region_set_skip_dump(submem
);
521 /* Create a zero sized sub-region to make cleanup easy. */
522 memory_region_init(submem
, obj
, name
, 0);
525 memory_region_add_subregion(mem
, offset
, submem
);
530 void vfio_reset_handler(void *opaque
)
533 VFIODevice
*vbasedev
;
535 QLIST_FOREACH(group
, &vfio_group_list
, next
) {
536 QLIST_FOREACH(vbasedev
, &group
->device_list
, next
) {
537 vbasedev
->ops
->vfio_compute_needs_reset(vbasedev
);
541 QLIST_FOREACH(group
, &vfio_group_list
, next
) {
542 QLIST_FOREACH(vbasedev
, &group
->device_list
, next
) {
543 if (vbasedev
->needs_reset
) {
544 vbasedev
->ops
->vfio_hot_reset_multi(vbasedev
);
550 static void vfio_kvm_device_add_group(VFIOGroup
*group
)
553 struct kvm_device_attr attr
= {
554 .group
= KVM_DEV_VFIO_GROUP
,
555 .attr
= KVM_DEV_VFIO_GROUP_ADD
,
556 .addr
= (uint64_t)(unsigned long)&group
->fd
,
559 if (!kvm_enabled()) {
563 if (vfio_kvm_device_fd
< 0) {
564 struct kvm_create_device cd
= {
565 .type
= KVM_DEV_TYPE_VFIO
,
568 if (kvm_vm_ioctl(kvm_state
, KVM_CREATE_DEVICE
, &cd
)) {
569 error_report("Failed to create KVM VFIO device: %m");
573 vfio_kvm_device_fd
= cd
.fd
;
576 if (ioctl(vfio_kvm_device_fd
, KVM_SET_DEVICE_ATTR
, &attr
)) {
577 error_report("Failed to add group %d to KVM VFIO device: %m",
583 static void vfio_kvm_device_del_group(VFIOGroup
*group
)
586 struct kvm_device_attr attr
= {
587 .group
= KVM_DEV_VFIO_GROUP
,
588 .attr
= KVM_DEV_VFIO_GROUP_DEL
,
589 .addr
= (uint64_t)(unsigned long)&group
->fd
,
592 if (vfio_kvm_device_fd
< 0) {
596 if (ioctl(vfio_kvm_device_fd
, KVM_SET_DEVICE_ATTR
, &attr
)) {
597 error_report("Failed to remove group %d from KVM VFIO device: %m",
603 static VFIOAddressSpace
*vfio_get_address_space(AddressSpace
*as
)
605 VFIOAddressSpace
*space
;
607 QLIST_FOREACH(space
, &vfio_address_spaces
, list
) {
608 if (space
->as
== as
) {
613 /* No suitable VFIOAddressSpace, create a new one */
614 space
= g_malloc0(sizeof(*space
));
616 QLIST_INIT(&space
->containers
);
618 QLIST_INSERT_HEAD(&vfio_address_spaces
, space
, list
);
623 static void vfio_put_address_space(VFIOAddressSpace
*space
)
625 if (QLIST_EMPTY(&space
->containers
)) {
626 QLIST_REMOVE(space
, list
);
631 static int vfio_connect_container(VFIOGroup
*group
, AddressSpace
*as
)
633 VFIOContainer
*container
;
635 VFIOAddressSpace
*space
;
637 space
= vfio_get_address_space(as
);
639 QLIST_FOREACH(container
, &space
->containers
, next
) {
640 if (!ioctl(group
->fd
, VFIO_GROUP_SET_CONTAINER
, &container
->fd
)) {
641 group
->container
= container
;
642 QLIST_INSERT_HEAD(&container
->group_list
, group
, container_next
);
647 fd
= qemu_open("/dev/vfio/vfio", O_RDWR
);
649 error_report("vfio: failed to open /dev/vfio/vfio: %m");
654 ret
= ioctl(fd
, VFIO_GET_API_VERSION
);
655 if (ret
!= VFIO_API_VERSION
) {
656 error_report("vfio: supported vfio version: %d, "
657 "reported version: %d", VFIO_API_VERSION
, ret
);
662 container
= g_malloc0(sizeof(*container
));
663 container
->space
= space
;
665 if (ioctl(fd
, VFIO_CHECK_EXTENSION
, VFIO_TYPE1_IOMMU
) ||
666 ioctl(fd
, VFIO_CHECK_EXTENSION
, VFIO_TYPE1v2_IOMMU
)) {
667 bool v2
= !!ioctl(fd
, VFIO_CHECK_EXTENSION
, VFIO_TYPE1v2_IOMMU
);
669 ret
= ioctl(group
->fd
, VFIO_GROUP_SET_CONTAINER
, &fd
);
671 error_report("vfio: failed to set group container: %m");
673 goto free_container_exit
;
676 ret
= ioctl(fd
, VFIO_SET_IOMMU
,
677 v2
? VFIO_TYPE1v2_IOMMU
: VFIO_TYPE1_IOMMU
);
679 error_report("vfio: failed to set iommu for container: %m");
681 goto free_container_exit
;
684 container
->iommu_data
.type1
.listener
= vfio_memory_listener
;
685 container
->iommu_data
.release
= vfio_listener_release
;
687 memory_listener_register(&container
->iommu_data
.type1
.listener
,
688 container
->space
->as
);
690 if (container
->iommu_data
.type1
.error
) {
691 ret
= container
->iommu_data
.type1
.error
;
692 error_report("vfio: memory listener initialization failed for container");
693 goto listener_release_exit
;
696 container
->iommu_data
.type1
.initialized
= true;
698 } else if (ioctl(fd
, VFIO_CHECK_EXTENSION
, VFIO_SPAPR_TCE_IOMMU
)) {
699 ret
= ioctl(group
->fd
, VFIO_GROUP_SET_CONTAINER
, &fd
);
701 error_report("vfio: failed to set group container: %m");
703 goto free_container_exit
;
705 ret
= ioctl(fd
, VFIO_SET_IOMMU
, VFIO_SPAPR_TCE_IOMMU
);
707 error_report("vfio: failed to set iommu for container: %m");
709 goto free_container_exit
;
713 * The host kernel code implementing VFIO_IOMMU_DISABLE is called
714 * when container fd is closed so we do not call it explicitly
717 ret
= ioctl(fd
, VFIO_IOMMU_ENABLE
);
719 error_report("vfio: failed to enable container: %m");
721 goto free_container_exit
;
724 container
->iommu_data
.type1
.listener
= vfio_memory_listener
;
725 container
->iommu_data
.release
= vfio_listener_release
;
727 memory_listener_register(&container
->iommu_data
.type1
.listener
,
728 container
->space
->as
);
731 error_report("vfio: No available IOMMU models");
733 goto free_container_exit
;
736 QLIST_INIT(&container
->group_list
);
737 QLIST_INSERT_HEAD(&space
->containers
, container
, next
);
739 group
->container
= container
;
740 QLIST_INSERT_HEAD(&container
->group_list
, group
, container_next
);
743 listener_release_exit
:
744 vfio_listener_release(container
);
753 vfio_put_address_space(space
);
758 static void vfio_disconnect_container(VFIOGroup
*group
)
760 VFIOContainer
*container
= group
->container
;
762 if (ioctl(group
->fd
, VFIO_GROUP_UNSET_CONTAINER
, &container
->fd
)) {
763 error_report("vfio: error disconnecting group %d from container",
767 QLIST_REMOVE(group
, container_next
);
768 group
->container
= NULL
;
770 if (QLIST_EMPTY(&container
->group_list
)) {
771 VFIOAddressSpace
*space
= container
->space
;
773 if (container
->iommu_data
.release
) {
774 container
->iommu_data
.release(container
);
776 QLIST_REMOVE(container
, next
);
777 trace_vfio_disconnect_container(container
->fd
);
778 close(container
->fd
);
781 vfio_put_address_space(space
);
785 VFIOGroup
*vfio_get_group(int groupid
, AddressSpace
*as
)
789 struct vfio_group_status status
= { .argsz
= sizeof(status
) };
791 QLIST_FOREACH(group
, &vfio_group_list
, next
) {
792 if (group
->groupid
== groupid
) {
793 /* Found it. Now is it already in the right context? */
794 if (group
->container
->space
->as
== as
) {
797 error_report("vfio: group %d used in multiple address spaces",
804 group
= g_malloc0(sizeof(*group
));
806 snprintf(path
, sizeof(path
), "/dev/vfio/%d", groupid
);
807 group
->fd
= qemu_open(path
, O_RDWR
);
809 error_report("vfio: error opening %s: %m", path
);
810 goto free_group_exit
;
813 if (ioctl(group
->fd
, VFIO_GROUP_GET_STATUS
, &status
)) {
814 error_report("vfio: error getting group status: %m");
818 if (!(status
.flags
& VFIO_GROUP_FLAGS_VIABLE
)) {
819 error_report("vfio: error, group %d is not viable, please ensure "
820 "all devices within the iommu_group are bound to their "
821 "vfio bus driver.", groupid
);
825 group
->groupid
= groupid
;
826 QLIST_INIT(&group
->device_list
);
828 if (vfio_connect_container(group
, as
)) {
829 error_report("vfio: failed to setup container for group %d", groupid
);
833 if (QLIST_EMPTY(&vfio_group_list
)) {
834 qemu_register_reset(vfio_reset_handler
, NULL
);
837 QLIST_INSERT_HEAD(&vfio_group_list
, group
, next
);
839 vfio_kvm_device_add_group(group
);
852 void vfio_put_group(VFIOGroup
*group
)
854 if (!group
|| !QLIST_EMPTY(&group
->device_list
)) {
858 vfio_kvm_device_del_group(group
);
859 vfio_disconnect_container(group
);
860 QLIST_REMOVE(group
, next
);
861 trace_vfio_put_group(group
->fd
);
865 if (QLIST_EMPTY(&vfio_group_list
)) {
866 qemu_unregister_reset(vfio_reset_handler
, NULL
);
870 int vfio_get_device(VFIOGroup
*group
, const char *name
,
871 VFIODevice
*vbasedev
)
873 struct vfio_device_info dev_info
= { .argsz
= sizeof(dev_info
) };
876 fd
= ioctl(group
->fd
, VFIO_GROUP_GET_DEVICE_FD
, name
);
878 error_report("vfio: error getting device %s from group %d: %m",
879 name
, group
->groupid
);
880 error_printf("Verify all devices in group %d are bound to vfio-<bus> "
881 "or pci-stub and not already in use\n", group
->groupid
);
885 ret
= ioctl(fd
, VFIO_DEVICE_GET_INFO
, &dev_info
);
887 error_report("vfio: error getting device info: %m");
893 vbasedev
->group
= group
;
894 QLIST_INSERT_HEAD(&group
->device_list
, vbasedev
, next
);
896 vbasedev
->num_irqs
= dev_info
.num_irqs
;
897 vbasedev
->num_regions
= dev_info
.num_regions
;
898 vbasedev
->flags
= dev_info
.flags
;
900 trace_vfio_get_device(name
, dev_info
.flags
, dev_info
.num_regions
,
903 vbasedev
->reset_works
= !!(dev_info
.flags
& VFIO_DEVICE_FLAGS_RESET
);
907 void vfio_put_base_device(VFIODevice
*vbasedev
)
909 if (!vbasedev
->group
) {
912 QLIST_REMOVE(vbasedev
, next
);
913 vbasedev
->group
= NULL
;
914 trace_vfio_put_base_device(vbasedev
->fd
);
918 static int vfio_container_do_ioctl(AddressSpace
*as
, int32_t groupid
,
919 int req
, void *param
)
922 VFIOContainer
*container
;
925 group
= vfio_get_group(groupid
, as
);
927 error_report("vfio: group %d not registered", groupid
);
931 container
= group
->container
;
932 if (group
->container
) {
933 ret
= ioctl(container
->fd
, req
, param
);
935 error_report("vfio: failed to ioctl %d to container: ret=%d, %s",
936 _IOC_NR(req
) - VFIO_BASE
, ret
, strerror(errno
));
940 vfio_put_group(group
);
945 int vfio_container_ioctl(AddressSpace
*as
, int32_t groupid
,
946 int req
, void *param
)
948 /* We allow only certain ioctls to the container */
950 case VFIO_CHECK_EXTENSION
:
951 case VFIO_IOMMU_SPAPR_TCE_GET_INFO
:
955 /* Return an error on unknown requests */
956 error_report("vfio: unsupported ioctl %X", req
);
960 return vfio_container_do_ioctl(as
, groupid
, req
, param
);