4 * Copyright IBM, Corp. 2008
8 * Anthony Liguori <aliguori@us.ibm.com>
9 * Glauber Costa <gcosta@redhat.com>
11 * This work is licensed under the terms of the GNU GPL, version 2 or later.
12 * See the COPYING file in the top-level directory.
16 #include "qemu/osdep.h"
17 #include <sys/ioctl.h>
19 #include <linux/kvm.h>
21 #include "qemu/atomic.h"
22 #include "qemu/option.h"
23 #include "qemu/config-file.h"
24 #include "qemu/error-report.h"
25 #include "qapi/error.h"
26 #include "hw/pci/msi.h"
27 #include "hw/pci/msix.h"
28 #include "hw/s390x/adapter.h"
29 #include "exec/gdbstub.h"
30 #include "sysemu/kvm_int.h"
31 #include "sysemu/runstate.h"
32 #include "sysemu/cpus.h"
33 #include "sysemu/sysemu.h"
34 #include "qemu/bswap.h"
35 #include "exec/memory.h"
36 #include "exec/ram_addr.h"
37 #include "exec/address-spaces.h"
38 #include "qemu/event_notifier.h"
39 #include "qemu/main-loop.h"
42 #include "sysemu/sev.h"
43 #include "sysemu/balloon.h"
45 #include "hw/boards.h"
47 /* This check must be after config-host.h is included */
49 #include <sys/eventfd.h>
52 /* KVM uses PAGE_SIZE in its definition of KVM_COALESCED_MMIO_MAX. We
53 * need to use the real host PAGE_SIZE, as that's what KVM will use.
55 #define PAGE_SIZE getpagesize()
60 #define DPRINTF(fmt, ...) \
61 do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
63 #define DPRINTF(fmt, ...) \
67 #define KVM_MSI_HASHTAB_SIZE 256
69 struct KVMParkedVcpu
{
70 unsigned long vcpu_id
;
72 QLIST_ENTRY(KVMParkedVcpu
) node
;
77 AccelState parent_obj
;
84 struct kvm_coalesced_mmio_ring
*coalesced_mmio_ring
;
85 bool coalesced_flush_in_progress
;
87 int robust_singlestep
;
89 #ifdef KVM_CAP_SET_GUEST_DEBUG
90 QTAILQ_HEAD(, kvm_sw_breakpoint
) kvm_sw_breakpoints
;
92 int max_nested_state_len
;
96 bool manual_dirty_log_protect
;
97 /* The man page (and posix) say ioctl numbers are signed int, but
98 * they're not. Linux, glibc and *BSD all treat ioctl numbers as
99 * unsigned, and treating them as signed here can break things */
100 unsigned irq_set_ioctl
;
101 unsigned int sigmask_len
;
103 #ifdef KVM_CAP_IRQ_ROUTING
104 struct kvm_irq_routing
*irq_routes
;
105 int nr_allocated_irq_routes
;
106 unsigned long *used_gsi_bitmap
;
107 unsigned int gsi_count
;
108 QTAILQ_HEAD(, KVMMSIRoute
) msi_hashtab
[KVM_MSI_HASHTAB_SIZE
];
110 KVMMemoryListener memory_listener
;
111 QLIST_HEAD(, KVMParkedVcpu
) kvm_parked_vcpus
;
113 /* memory encryption */
114 void *memcrypt_handle
;
115 int (*memcrypt_encrypt_data
)(void *handle
, uint8_t *ptr
, uint64_t len
);
117 /* For "info mtree -f" to tell if an MR is registered in KVM */
120 KVMMemoryListener
*ml
;
126 bool kvm_kernel_irqchip
;
127 bool kvm_split_irqchip
;
128 bool kvm_async_interrupts_allowed
;
129 bool kvm_halt_in_kernel_allowed
;
130 bool kvm_eventfds_allowed
;
131 bool kvm_irqfds_allowed
;
132 bool kvm_resamplefds_allowed
;
133 bool kvm_msi_via_irqfd_allowed
;
134 bool kvm_gsi_routing_allowed
;
135 bool kvm_gsi_direct_mapping
;
137 bool kvm_readonly_mem_allowed
;
138 bool kvm_vm_attributes_allowed
;
139 bool kvm_direct_msi_allowed
;
140 bool kvm_ioeventfd_any_length_allowed
;
141 bool kvm_msi_use_devid
;
142 static bool kvm_immediate_exit
;
143 static hwaddr kvm_max_slot_size
= ~0;
145 static const KVMCapabilityInfo kvm_required_capabilites
[] = {
146 KVM_CAP_INFO(USER_MEMORY
),
147 KVM_CAP_INFO(DESTROY_MEMORY_REGION_WORKS
),
148 KVM_CAP_INFO(JOIN_MEMORY_REGIONS_WORKS
),
152 #define kvm_slots_lock(kml) qemu_mutex_lock(&(kml)->slots_lock)
153 #define kvm_slots_unlock(kml) qemu_mutex_unlock(&(kml)->slots_lock)
155 int kvm_get_max_memslots(void)
157 KVMState
*s
= KVM_STATE(current_machine
->accelerator
);
162 bool kvm_memcrypt_enabled(void)
164 if (kvm_state
&& kvm_state
->memcrypt_handle
) {
171 int kvm_memcrypt_encrypt_data(uint8_t *ptr
, uint64_t len
)
173 if (kvm_state
->memcrypt_handle
&&
174 kvm_state
->memcrypt_encrypt_data
) {
175 return kvm_state
->memcrypt_encrypt_data(kvm_state
->memcrypt_handle
,
182 /* Called with KVMMemoryListener.slots_lock held */
183 static KVMSlot
*kvm_get_free_slot(KVMMemoryListener
*kml
)
185 KVMState
*s
= kvm_state
;
188 for (i
= 0; i
< s
->nr_slots
; i
++) {
189 if (kml
->slots
[i
].memory_size
== 0) {
190 return &kml
->slots
[i
];
197 bool kvm_has_free_slot(MachineState
*ms
)
199 KVMState
*s
= KVM_STATE(ms
->accelerator
);
201 KVMMemoryListener
*kml
= &s
->memory_listener
;
204 result
= !!kvm_get_free_slot(kml
);
205 kvm_slots_unlock(kml
);
210 /* Called with KVMMemoryListener.slots_lock held */
211 static KVMSlot
*kvm_alloc_slot(KVMMemoryListener
*kml
)
213 KVMSlot
*slot
= kvm_get_free_slot(kml
);
219 fprintf(stderr
, "%s: no free slot available\n", __func__
);
223 static KVMSlot
*kvm_lookup_matching_slot(KVMMemoryListener
*kml
,
227 KVMState
*s
= kvm_state
;
230 for (i
= 0; i
< s
->nr_slots
; i
++) {
231 KVMSlot
*mem
= &kml
->slots
[i
];
233 if (start_addr
== mem
->start_addr
&& size
== mem
->memory_size
) {
242 * Calculate and align the start address and the size of the section.
243 * Return the size. If the size is 0, the aligned section is empty.
245 static hwaddr
kvm_align_section(MemoryRegionSection
*section
,
248 hwaddr size
= int128_get64(section
->size
);
249 hwaddr delta
, aligned
;
251 /* kvm works in page size chunks, but the function may be called
252 with sub-page size and unaligned start address. Pad the start
253 address to next and truncate size to previous page boundary. */
254 aligned
= ROUND_UP(section
->offset_within_address_space
,
255 qemu_real_host_page_size
);
256 delta
= aligned
- section
->offset_within_address_space
;
262 return (size
- delta
) & qemu_real_host_page_mask
;
265 int kvm_physical_memory_addr_from_host(KVMState
*s
, void *ram
,
268 KVMMemoryListener
*kml
= &s
->memory_listener
;
272 for (i
= 0; i
< s
->nr_slots
; i
++) {
273 KVMSlot
*mem
= &kml
->slots
[i
];
275 if (ram
>= mem
->ram
&& ram
< mem
->ram
+ mem
->memory_size
) {
276 *phys_addr
= mem
->start_addr
+ (ram
- mem
->ram
);
281 kvm_slots_unlock(kml
);
286 static int kvm_set_user_memory_region(KVMMemoryListener
*kml
, KVMSlot
*slot
, bool new)
288 KVMState
*s
= kvm_state
;
289 struct kvm_userspace_memory_region mem
;
292 mem
.slot
= slot
->slot
| (kml
->as_id
<< 16);
293 mem
.guest_phys_addr
= slot
->start_addr
;
294 mem
.userspace_addr
= (unsigned long)slot
->ram
;
295 mem
.flags
= slot
->flags
;
297 if (slot
->memory_size
&& !new && (mem
.flags
^ slot
->old_flags
) & KVM_MEM_READONLY
) {
298 /* Set the slot size to 0 before setting the slot to the desired
299 * value. This is needed based on KVM commit 75d61fbc. */
301 kvm_vm_ioctl(s
, KVM_SET_USER_MEMORY_REGION
, &mem
);
303 mem
.memory_size
= slot
->memory_size
;
304 ret
= kvm_vm_ioctl(s
, KVM_SET_USER_MEMORY_REGION
, &mem
);
305 slot
->old_flags
= mem
.flags
;
306 trace_kvm_set_user_memory(mem
.slot
, mem
.flags
, mem
.guest_phys_addr
,
307 mem
.memory_size
, mem
.userspace_addr
, ret
);
311 int kvm_destroy_vcpu(CPUState
*cpu
)
313 KVMState
*s
= kvm_state
;
315 struct KVMParkedVcpu
*vcpu
= NULL
;
318 DPRINTF("kvm_destroy_vcpu\n");
320 ret
= kvm_arch_destroy_vcpu(cpu
);
325 mmap_size
= kvm_ioctl(s
, KVM_GET_VCPU_MMAP_SIZE
, 0);
328 DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n");
332 ret
= munmap(cpu
->kvm_run
, mmap_size
);
337 vcpu
= g_malloc0(sizeof(*vcpu
));
338 vcpu
->vcpu_id
= kvm_arch_vcpu_id(cpu
);
339 vcpu
->kvm_fd
= cpu
->kvm_fd
;
340 QLIST_INSERT_HEAD(&kvm_state
->kvm_parked_vcpus
, vcpu
, node
);
345 static int kvm_get_vcpu(KVMState
*s
, unsigned long vcpu_id
)
347 struct KVMParkedVcpu
*cpu
;
349 QLIST_FOREACH(cpu
, &s
->kvm_parked_vcpus
, node
) {
350 if (cpu
->vcpu_id
== vcpu_id
) {
353 QLIST_REMOVE(cpu
, node
);
354 kvm_fd
= cpu
->kvm_fd
;
360 return kvm_vm_ioctl(s
, KVM_CREATE_VCPU
, (void *)vcpu_id
);
363 int kvm_init_vcpu(CPUState
*cpu
)
365 KVMState
*s
= kvm_state
;
369 DPRINTF("kvm_init_vcpu\n");
371 ret
= kvm_get_vcpu(s
, kvm_arch_vcpu_id(cpu
));
373 DPRINTF("kvm_create_vcpu failed\n");
379 cpu
->vcpu_dirty
= true;
381 mmap_size
= kvm_ioctl(s
, KVM_GET_VCPU_MMAP_SIZE
, 0);
384 DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n");
388 cpu
->kvm_run
= mmap(NULL
, mmap_size
, PROT_READ
| PROT_WRITE
, MAP_SHARED
,
390 if (cpu
->kvm_run
== MAP_FAILED
) {
392 DPRINTF("mmap'ing vcpu state failed\n");
396 if (s
->coalesced_mmio
&& !s
->coalesced_mmio_ring
) {
397 s
->coalesced_mmio_ring
=
398 (void *)cpu
->kvm_run
+ s
->coalesced_mmio
* PAGE_SIZE
;
401 ret
= kvm_arch_init_vcpu(cpu
);
407 * dirty pages logging control
410 static int kvm_mem_flags(MemoryRegion
*mr
)
412 bool readonly
= mr
->readonly
|| memory_region_is_romd(mr
);
415 if (memory_region_get_dirty_log_mask(mr
) != 0) {
416 flags
|= KVM_MEM_LOG_DIRTY_PAGES
;
418 if (readonly
&& kvm_readonly_mem_allowed
) {
419 flags
|= KVM_MEM_READONLY
;
424 /* Called with KVMMemoryListener.slots_lock held */
425 static int kvm_slot_update_flags(KVMMemoryListener
*kml
, KVMSlot
*mem
,
428 mem
->flags
= kvm_mem_flags(mr
);
430 /* If nothing changed effectively, no need to issue ioctl */
431 if (mem
->flags
== mem
->old_flags
) {
435 return kvm_set_user_memory_region(kml
, mem
, false);
438 static int kvm_section_update_flags(KVMMemoryListener
*kml
,
439 MemoryRegionSection
*section
)
441 hwaddr start_addr
, size
, slot_size
;
445 size
= kvm_align_section(section
, &start_addr
);
452 while (size
&& !ret
) {
453 slot_size
= MIN(kvm_max_slot_size
, size
);
454 mem
= kvm_lookup_matching_slot(kml
, start_addr
, slot_size
);
456 /* We don't have a slot if we want to trap every access. */
460 ret
= kvm_slot_update_flags(kml
, mem
, section
->mr
);
461 start_addr
+= slot_size
;
466 kvm_slots_unlock(kml
);
470 static void kvm_log_start(MemoryListener
*listener
,
471 MemoryRegionSection
*section
,
474 KVMMemoryListener
*kml
= container_of(listener
, KVMMemoryListener
, listener
);
481 r
= kvm_section_update_flags(kml
, section
);
487 static void kvm_log_stop(MemoryListener
*listener
,
488 MemoryRegionSection
*section
,
491 KVMMemoryListener
*kml
= container_of(listener
, KVMMemoryListener
, listener
);
498 r
= kvm_section_update_flags(kml
, section
);
504 /* get kvm's dirty pages bitmap and update qemu's */
505 static int kvm_get_dirty_pages_log_range(MemoryRegionSection
*section
,
506 unsigned long *bitmap
)
508 ram_addr_t start
= section
->offset_within_region
+
509 memory_region_get_ram_addr(section
->mr
);
510 ram_addr_t pages
= int128_get64(section
->size
) / getpagesize();
512 cpu_physical_memory_set_dirty_lebitmap(bitmap
, start
, pages
);
516 #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1))
519 * kvm_physical_sync_dirty_bitmap - Sync dirty bitmap from kernel space
521 * This function will first try to fetch dirty bitmap from the kernel,
522 * and then updates qemu's dirty bitmap.
524 * NOTE: caller must be with kml->slots_lock held.
526 * @kml: the KVM memory listener object
527 * @section: the memory section to sync the dirty bitmap with
529 static int kvm_physical_sync_dirty_bitmap(KVMMemoryListener
*kml
,
530 MemoryRegionSection
*section
)
532 KVMState
*s
= kvm_state
;
533 struct kvm_dirty_log d
= {};
535 hwaddr start_addr
, size
;
536 hwaddr slot_size
, slot_offset
= 0;
539 size
= kvm_align_section(section
, &start_addr
);
541 MemoryRegionSection subsection
= *section
;
543 slot_size
= MIN(kvm_max_slot_size
, size
);
544 mem
= kvm_lookup_matching_slot(kml
, start_addr
, slot_size
);
546 /* We don't have a slot if we want to trap every access. */
550 /* XXX bad kernel interface alert
551 * For dirty bitmap, kernel allocates array of size aligned to
552 * bits-per-long. But for case when the kernel is 64bits and
553 * the userspace is 32bits, userspace can't align to the same
554 * bits-per-long, since sizeof(long) is different between kernel
555 * and user space. This way, userspace will provide buffer which
556 * may be 4 bytes less than the kernel will use, resulting in
557 * userspace memory corruption (which is not detectable by valgrind
558 * too, in most cases).
559 * So for now, let's align to 64 instead of HOST_LONG_BITS here, in
560 * a hope that sizeof(long) won't become >8 any time soon.
562 if (!mem
->dirty_bmap
) {
563 hwaddr bitmap_size
= ALIGN(((mem
->memory_size
) >> TARGET_PAGE_BITS
),
564 /*HOST_LONG_BITS*/ 64) / 8;
565 /* Allocate on the first log_sync, once and for all */
566 mem
->dirty_bmap
= g_malloc0(bitmap_size
);
569 d
.dirty_bitmap
= mem
->dirty_bmap
;
570 d
.slot
= mem
->slot
| (kml
->as_id
<< 16);
571 if (kvm_vm_ioctl(s
, KVM_GET_DIRTY_LOG
, &d
) == -1) {
572 DPRINTF("ioctl failed %d\n", errno
);
577 subsection
.offset_within_region
+= slot_offset
;
578 subsection
.size
= int128_make64(slot_size
);
579 kvm_get_dirty_pages_log_range(&subsection
, d
.dirty_bitmap
);
581 slot_offset
+= slot_size
;
582 start_addr
+= slot_size
;
589 /* Alignment requirement for KVM_CLEAR_DIRTY_LOG - 64 pages */
590 #define KVM_CLEAR_LOG_SHIFT 6
591 #define KVM_CLEAR_LOG_ALIGN (qemu_real_host_page_size << KVM_CLEAR_LOG_SHIFT)
592 #define KVM_CLEAR_LOG_MASK (-KVM_CLEAR_LOG_ALIGN)
594 static int kvm_log_clear_one_slot(KVMSlot
*mem
, int as_id
, uint64_t start
,
597 KVMState
*s
= kvm_state
;
598 uint64_t end
, bmap_start
, start_delta
, bmap_npages
;
599 struct kvm_clear_dirty_log d
;
600 unsigned long *bmap_clear
= NULL
, psize
= qemu_real_host_page_size
;
604 * We need to extend either the start or the size or both to
605 * satisfy the KVM interface requirement. Firstly, do the start
606 * page alignment on 64 host pages
608 bmap_start
= start
& KVM_CLEAR_LOG_MASK
;
609 start_delta
= start
- bmap_start
;
613 * The kernel interface has restriction on the size too, that either:
615 * (1) the size is 64 host pages aligned (just like the start), or
616 * (2) the size fills up until the end of the KVM memslot.
618 bmap_npages
= DIV_ROUND_UP(size
+ start_delta
, KVM_CLEAR_LOG_ALIGN
)
619 << KVM_CLEAR_LOG_SHIFT
;
620 end
= mem
->memory_size
/ psize
;
621 if (bmap_npages
> end
- bmap_start
) {
622 bmap_npages
= end
- bmap_start
;
624 start_delta
/= psize
;
627 * Prepare the bitmap to clear dirty bits. Here we must guarantee
628 * that we won't clear any unknown dirty bits otherwise we might
629 * accidentally clear some set bits which are not yet synced from
630 * the kernel into QEMU's bitmap, then we'll lose track of the
631 * guest modifications upon those pages (which can directly lead
632 * to guest data loss or panic after migration).
634 * Layout of the KVMSlot.dirty_bmap:
636 * |<-------- bmap_npages -----------..>|
639 * |----------------|-------------|------------------|------------|
642 * start bmap_start (start) end
643 * of memslot of memslot
645 * [1] bmap_npages can be aligned to either 64 pages or the end of slot
648 assert(bmap_start
% BITS_PER_LONG
== 0);
649 /* We should never do log_clear before log_sync */
650 assert(mem
->dirty_bmap
);
652 /* Slow path - we need to manipulate a temp bitmap */
653 bmap_clear
= bitmap_new(bmap_npages
);
654 bitmap_copy_with_src_offset(bmap_clear
, mem
->dirty_bmap
,
655 bmap_start
, start_delta
+ size
/ psize
);
657 * We need to fill the holes at start because that was not
658 * specified by the caller and we extended the bitmap only for
661 bitmap_clear(bmap_clear
, 0, start_delta
);
662 d
.dirty_bitmap
= bmap_clear
;
664 /* Fast path - start address aligns well with BITS_PER_LONG */
665 d
.dirty_bitmap
= mem
->dirty_bmap
+ BIT_WORD(bmap_start
);
668 d
.first_page
= bmap_start
;
669 /* It should never overflow. If it happens, say something */
670 assert(bmap_npages
<= UINT32_MAX
);
671 d
.num_pages
= bmap_npages
;
672 d
.slot
= mem
->slot
| (as_id
<< 16);
674 if (kvm_vm_ioctl(s
, KVM_CLEAR_DIRTY_LOG
, &d
) == -1) {
676 error_report("%s: KVM_CLEAR_DIRTY_LOG failed, slot=%d, "
677 "start=0x%"PRIx64
", size=0x%"PRIx32
", errno=%d",
678 __func__
, d
.slot
, (uint64_t)d
.first_page
,
679 (uint32_t)d
.num_pages
, ret
);
682 trace_kvm_clear_dirty_log(d
.slot
, d
.first_page
, d
.num_pages
);
686 * After we have updated the remote dirty bitmap, we update the
687 * cached bitmap as well for the memslot, then if another user
688 * clears the same region we know we shouldn't clear it again on
689 * the remote otherwise it's data loss as well.
691 bitmap_clear(mem
->dirty_bmap
, bmap_start
+ start_delta
,
693 /* This handles the NULL case well */
700 * kvm_physical_log_clear - Clear the kernel's dirty bitmap for range
702 * NOTE: this will be a no-op if we haven't enabled manual dirty log
703 * protection in the host kernel because in that case this operation
704 * will be done within log_sync().
706 * @kml: the kvm memory listener
707 * @section: the memory range to clear dirty bitmap
709 static int kvm_physical_log_clear(KVMMemoryListener
*kml
,
710 MemoryRegionSection
*section
)
712 KVMState
*s
= kvm_state
;
713 uint64_t start
, size
, offset
, count
;
717 if (!s
->manual_dirty_log_protect
) {
718 /* No need to do explicit clear */
722 start
= section
->offset_within_address_space
;
723 size
= int128_get64(section
->size
);
726 /* Nothing more we can do... */
732 for (i
= 0; i
< s
->nr_slots
; i
++) {
733 mem
= &kml
->slots
[i
];
734 /* Discard slots that are empty or do not overlap the section */
735 if (!mem
->memory_size
||
736 mem
->start_addr
> start
+ size
- 1 ||
737 start
> mem
->start_addr
+ mem
->memory_size
- 1) {
741 if (start
>= mem
->start_addr
) {
742 /* The slot starts before section or is aligned to it. */
743 offset
= start
- mem
->start_addr
;
744 count
= MIN(mem
->memory_size
- offset
, size
);
746 /* The slot starts after section. */
748 count
= MIN(mem
->memory_size
, size
- (mem
->start_addr
- start
));
750 ret
= kvm_log_clear_one_slot(mem
, kml
->as_id
, offset
, count
);
756 kvm_slots_unlock(kml
);
761 static void kvm_coalesce_mmio_region(MemoryListener
*listener
,
762 MemoryRegionSection
*secion
,
763 hwaddr start
, hwaddr size
)
765 KVMState
*s
= kvm_state
;
767 if (s
->coalesced_mmio
) {
768 struct kvm_coalesced_mmio_zone zone
;
774 (void)kvm_vm_ioctl(s
, KVM_REGISTER_COALESCED_MMIO
, &zone
);
778 static void kvm_uncoalesce_mmio_region(MemoryListener
*listener
,
779 MemoryRegionSection
*secion
,
780 hwaddr start
, hwaddr size
)
782 KVMState
*s
= kvm_state
;
784 if (s
->coalesced_mmio
) {
785 struct kvm_coalesced_mmio_zone zone
;
791 (void)kvm_vm_ioctl(s
, KVM_UNREGISTER_COALESCED_MMIO
, &zone
);
795 static void kvm_coalesce_pio_add(MemoryListener
*listener
,
796 MemoryRegionSection
*section
,
797 hwaddr start
, hwaddr size
)
799 KVMState
*s
= kvm_state
;
801 if (s
->coalesced_pio
) {
802 struct kvm_coalesced_mmio_zone zone
;
808 (void)kvm_vm_ioctl(s
, KVM_REGISTER_COALESCED_MMIO
, &zone
);
812 static void kvm_coalesce_pio_del(MemoryListener
*listener
,
813 MemoryRegionSection
*section
,
814 hwaddr start
, hwaddr size
)
816 KVMState
*s
= kvm_state
;
818 if (s
->coalesced_pio
) {
819 struct kvm_coalesced_mmio_zone zone
;
825 (void)kvm_vm_ioctl(s
, KVM_UNREGISTER_COALESCED_MMIO
, &zone
);
829 static MemoryListener kvm_coalesced_pio_listener
= {
830 .coalesced_io_add
= kvm_coalesce_pio_add
,
831 .coalesced_io_del
= kvm_coalesce_pio_del
,
834 int kvm_check_extension(KVMState
*s
, unsigned int extension
)
838 ret
= kvm_ioctl(s
, KVM_CHECK_EXTENSION
, extension
);
846 int kvm_vm_check_extension(KVMState
*s
, unsigned int extension
)
850 ret
= kvm_vm_ioctl(s
, KVM_CHECK_EXTENSION
, extension
);
852 /* VM wide version not implemented, use global one instead */
853 ret
= kvm_check_extension(s
, extension
);
859 static uint32_t adjust_ioeventfd_endianness(uint32_t val
, uint32_t size
)
861 #if defined(HOST_WORDS_BIGENDIAN) != defined(TARGET_WORDS_BIGENDIAN)
862 /* The kernel expects ioeventfd values in HOST_WORDS_BIGENDIAN
863 * endianness, but the memory core hands them in target endianness.
864 * For example, PPC is always treated as big-endian even if running
865 * on KVM and on PPC64LE. Correct here.
879 static int kvm_set_ioeventfd_mmio(int fd
, hwaddr addr
, uint32_t val
,
880 bool assign
, uint32_t size
, bool datamatch
)
883 struct kvm_ioeventfd iofd
= {
884 .datamatch
= datamatch
? adjust_ioeventfd_endianness(val
, size
) : 0,
891 trace_kvm_set_ioeventfd_mmio(fd
, (uint64_t)addr
, val
, assign
, size
,
893 if (!kvm_enabled()) {
898 iofd
.flags
|= KVM_IOEVENTFD_FLAG_DATAMATCH
;
901 iofd
.flags
|= KVM_IOEVENTFD_FLAG_DEASSIGN
;
904 ret
= kvm_vm_ioctl(kvm_state
, KVM_IOEVENTFD
, &iofd
);
913 static int kvm_set_ioeventfd_pio(int fd
, uint16_t addr
, uint16_t val
,
914 bool assign
, uint32_t size
, bool datamatch
)
916 struct kvm_ioeventfd kick
= {
917 .datamatch
= datamatch
? adjust_ioeventfd_endianness(val
, size
) : 0,
919 .flags
= KVM_IOEVENTFD_FLAG_PIO
,
924 trace_kvm_set_ioeventfd_pio(fd
, addr
, val
, assign
, size
, datamatch
);
925 if (!kvm_enabled()) {
929 kick
.flags
|= KVM_IOEVENTFD_FLAG_DATAMATCH
;
932 kick
.flags
|= KVM_IOEVENTFD_FLAG_DEASSIGN
;
934 r
= kvm_vm_ioctl(kvm_state
, KVM_IOEVENTFD
, &kick
);
942 static int kvm_check_many_ioeventfds(void)
944 /* Userspace can use ioeventfd for io notification. This requires a host
945 * that supports eventfd(2) and an I/O thread; since eventfd does not
946 * support SIGIO it cannot interrupt the vcpu.
948 * Older kernels have a 6 device limit on the KVM io bus. Find out so we
949 * can avoid creating too many ioeventfds.
951 #if defined(CONFIG_EVENTFD)
954 for (i
= 0; i
< ARRAY_SIZE(ioeventfds
); i
++) {
955 ioeventfds
[i
] = eventfd(0, EFD_CLOEXEC
);
956 if (ioeventfds
[i
] < 0) {
959 ret
= kvm_set_ioeventfd_pio(ioeventfds
[i
], 0, i
, true, 2, true);
961 close(ioeventfds
[i
]);
966 /* Decide whether many devices are supported or not */
967 ret
= i
== ARRAY_SIZE(ioeventfds
);
970 kvm_set_ioeventfd_pio(ioeventfds
[i
], 0, i
, false, 2, true);
971 close(ioeventfds
[i
]);
979 static const KVMCapabilityInfo
*
980 kvm_check_extension_list(KVMState
*s
, const KVMCapabilityInfo
*list
)
983 if (!kvm_check_extension(s
, list
->value
)) {
991 void kvm_set_max_memslot_size(hwaddr max_slot_size
)
994 ROUND_UP(max_slot_size
, qemu_real_host_page_size
) == max_slot_size
996 kvm_max_slot_size
= max_slot_size
;
999 static void kvm_set_phys_mem(KVMMemoryListener
*kml
,
1000 MemoryRegionSection
*section
, bool add
)
1004 MemoryRegion
*mr
= section
->mr
;
1005 bool writeable
= !mr
->readonly
&& !mr
->rom_device
;
1006 hwaddr start_addr
, size
, slot_size
;
1009 if (!memory_region_is_ram(mr
)) {
1010 if (writeable
|| !kvm_readonly_mem_allowed
) {
1012 } else if (!mr
->romd_mode
) {
1013 /* If the memory device is not in romd_mode, then we actually want
1014 * to remove the kvm memory slot so all accesses will trap. */
1019 size
= kvm_align_section(section
, &start_addr
);
1024 /* use aligned delta to align the ram address */
1025 ram
= memory_region_get_ram_ptr(mr
) + section
->offset_within_region
+
1026 (start_addr
- section
->offset_within_address_space
);
1028 kvm_slots_lock(kml
);
1032 slot_size
= MIN(kvm_max_slot_size
, size
);
1033 mem
= kvm_lookup_matching_slot(kml
, start_addr
, slot_size
);
1037 if (mem
->flags
& KVM_MEM_LOG_DIRTY_PAGES
) {
1038 kvm_physical_sync_dirty_bitmap(kml
, section
);
1041 /* unregister the slot */
1042 g_free(mem
->dirty_bmap
);
1043 mem
->dirty_bmap
= NULL
;
1044 mem
->memory_size
= 0;
1046 err
= kvm_set_user_memory_region(kml
, mem
, false);
1048 fprintf(stderr
, "%s: error unregistering slot: %s\n",
1049 __func__
, strerror(-err
));
1052 start_addr
+= slot_size
;
1058 /* register the new slot */
1060 slot_size
= MIN(kvm_max_slot_size
, size
);
1061 mem
= kvm_alloc_slot(kml
);
1062 mem
->memory_size
= slot_size
;
1063 mem
->start_addr
= start_addr
;
1065 mem
->flags
= kvm_mem_flags(mr
);
1067 err
= kvm_set_user_memory_region(kml
, mem
, true);
1069 fprintf(stderr
, "%s: error registering slot: %s\n", __func__
,
1073 start_addr
+= slot_size
;
1079 kvm_slots_unlock(kml
);
1082 static void kvm_region_add(MemoryListener
*listener
,
1083 MemoryRegionSection
*section
)
1085 KVMMemoryListener
*kml
= container_of(listener
, KVMMemoryListener
, listener
);
1087 memory_region_ref(section
->mr
);
1088 kvm_set_phys_mem(kml
, section
, true);
1091 static void kvm_region_del(MemoryListener
*listener
,
1092 MemoryRegionSection
*section
)
1094 KVMMemoryListener
*kml
= container_of(listener
, KVMMemoryListener
, listener
);
1096 kvm_set_phys_mem(kml
, section
, false);
1097 memory_region_unref(section
->mr
);
1100 static void kvm_log_sync(MemoryListener
*listener
,
1101 MemoryRegionSection
*section
)
1103 KVMMemoryListener
*kml
= container_of(listener
, KVMMemoryListener
, listener
);
1106 kvm_slots_lock(kml
);
1107 r
= kvm_physical_sync_dirty_bitmap(kml
, section
);
1108 kvm_slots_unlock(kml
);
1114 static void kvm_log_clear(MemoryListener
*listener
,
1115 MemoryRegionSection
*section
)
1117 KVMMemoryListener
*kml
= container_of(listener
, KVMMemoryListener
, listener
);
1120 r
= kvm_physical_log_clear(kml
, section
);
1122 error_report_once("%s: kvm log clear failed: mr=%s "
1123 "offset=%"HWADDR_PRIx
" size=%"PRIx64
, __func__
,
1124 section
->mr
->name
, section
->offset_within_region
,
1125 int128_get64(section
->size
));
1130 static void kvm_mem_ioeventfd_add(MemoryListener
*listener
,
1131 MemoryRegionSection
*section
,
1132 bool match_data
, uint64_t data
,
1135 int fd
= event_notifier_get_fd(e
);
1138 r
= kvm_set_ioeventfd_mmio(fd
, section
->offset_within_address_space
,
1139 data
, true, int128_get64(section
->size
),
1142 fprintf(stderr
, "%s: error adding ioeventfd: %s (%d)\n",
1143 __func__
, strerror(-r
), -r
);
1148 static void kvm_mem_ioeventfd_del(MemoryListener
*listener
,
1149 MemoryRegionSection
*section
,
1150 bool match_data
, uint64_t data
,
1153 int fd
= event_notifier_get_fd(e
);
1156 r
= kvm_set_ioeventfd_mmio(fd
, section
->offset_within_address_space
,
1157 data
, false, int128_get64(section
->size
),
1160 fprintf(stderr
, "%s: error deleting ioeventfd: %s (%d)\n",
1161 __func__
, strerror(-r
), -r
);
1166 static void kvm_io_ioeventfd_add(MemoryListener
*listener
,
1167 MemoryRegionSection
*section
,
1168 bool match_data
, uint64_t data
,
1171 int fd
= event_notifier_get_fd(e
);
1174 r
= kvm_set_ioeventfd_pio(fd
, section
->offset_within_address_space
,
1175 data
, true, int128_get64(section
->size
),
1178 fprintf(stderr
, "%s: error adding ioeventfd: %s (%d)\n",
1179 __func__
, strerror(-r
), -r
);
1184 static void kvm_io_ioeventfd_del(MemoryListener
*listener
,
1185 MemoryRegionSection
*section
,
1186 bool match_data
, uint64_t data
,
1190 int fd
= event_notifier_get_fd(e
);
1193 r
= kvm_set_ioeventfd_pio(fd
, section
->offset_within_address_space
,
1194 data
, false, int128_get64(section
->size
),
1197 fprintf(stderr
, "%s: error deleting ioeventfd: %s (%d)\n",
1198 __func__
, strerror(-r
), -r
);
1203 void kvm_memory_listener_register(KVMState
*s
, KVMMemoryListener
*kml
,
1204 AddressSpace
*as
, int as_id
)
1208 qemu_mutex_init(&kml
->slots_lock
);
1209 kml
->slots
= g_malloc0(s
->nr_slots
* sizeof(KVMSlot
));
1212 for (i
= 0; i
< s
->nr_slots
; i
++) {
1213 kml
->slots
[i
].slot
= i
;
1216 kml
->listener
.region_add
= kvm_region_add
;
1217 kml
->listener
.region_del
= kvm_region_del
;
1218 kml
->listener
.log_start
= kvm_log_start
;
1219 kml
->listener
.log_stop
= kvm_log_stop
;
1220 kml
->listener
.log_sync
= kvm_log_sync
;
1221 kml
->listener
.log_clear
= kvm_log_clear
;
1222 kml
->listener
.priority
= 10;
1224 memory_listener_register(&kml
->listener
, as
);
1226 for (i
= 0; i
< s
->nr_as
; ++i
) {
1235 static MemoryListener kvm_io_listener
= {
1236 .eventfd_add
= kvm_io_ioeventfd_add
,
1237 .eventfd_del
= kvm_io_ioeventfd_del
,
1241 int kvm_set_irq(KVMState
*s
, int irq
, int level
)
1243 struct kvm_irq_level event
;
1246 assert(kvm_async_interrupts_enabled());
1248 event
.level
= level
;
1250 ret
= kvm_vm_ioctl(s
, s
->irq_set_ioctl
, &event
);
1252 perror("kvm_set_irq");
1256 return (s
->irq_set_ioctl
== KVM_IRQ_LINE
) ? 1 : event
.status
;
1259 #ifdef KVM_CAP_IRQ_ROUTING
1260 typedef struct KVMMSIRoute
{
1261 struct kvm_irq_routing_entry kroute
;
1262 QTAILQ_ENTRY(KVMMSIRoute
) entry
;
1265 static void set_gsi(KVMState
*s
, unsigned int gsi
)
1267 set_bit(gsi
, s
->used_gsi_bitmap
);
1270 static void clear_gsi(KVMState
*s
, unsigned int gsi
)
1272 clear_bit(gsi
, s
->used_gsi_bitmap
);
1275 void kvm_init_irq_routing(KVMState
*s
)
1279 gsi_count
= kvm_check_extension(s
, KVM_CAP_IRQ_ROUTING
) - 1;
1280 if (gsi_count
> 0) {
1281 /* Round up so we can search ints using ffs */
1282 s
->used_gsi_bitmap
= bitmap_new(gsi_count
);
1283 s
->gsi_count
= gsi_count
;
1286 s
->irq_routes
= g_malloc0(sizeof(*s
->irq_routes
));
1287 s
->nr_allocated_irq_routes
= 0;
1289 if (!kvm_direct_msi_allowed
) {
1290 for (i
= 0; i
< KVM_MSI_HASHTAB_SIZE
; i
++) {
1291 QTAILQ_INIT(&s
->msi_hashtab
[i
]);
1295 kvm_arch_init_irq_routing(s
);
1298 void kvm_irqchip_commit_routes(KVMState
*s
)
1302 if (kvm_gsi_direct_mapping()) {
1306 if (!kvm_gsi_routing_enabled()) {
1310 s
->irq_routes
->flags
= 0;
1311 trace_kvm_irqchip_commit_routes();
1312 ret
= kvm_vm_ioctl(s
, KVM_SET_GSI_ROUTING
, s
->irq_routes
);
1316 static void kvm_add_routing_entry(KVMState
*s
,
1317 struct kvm_irq_routing_entry
*entry
)
1319 struct kvm_irq_routing_entry
*new;
1322 if (s
->irq_routes
->nr
== s
->nr_allocated_irq_routes
) {
1323 n
= s
->nr_allocated_irq_routes
* 2;
1327 size
= sizeof(struct kvm_irq_routing
);
1328 size
+= n
* sizeof(*new);
1329 s
->irq_routes
= g_realloc(s
->irq_routes
, size
);
1330 s
->nr_allocated_irq_routes
= n
;
1332 n
= s
->irq_routes
->nr
++;
1333 new = &s
->irq_routes
->entries
[n
];
1337 set_gsi(s
, entry
->gsi
);
1340 static int kvm_update_routing_entry(KVMState
*s
,
1341 struct kvm_irq_routing_entry
*new_entry
)
1343 struct kvm_irq_routing_entry
*entry
;
1346 for (n
= 0; n
< s
->irq_routes
->nr
; n
++) {
1347 entry
= &s
->irq_routes
->entries
[n
];
1348 if (entry
->gsi
!= new_entry
->gsi
) {
1352 if(!memcmp(entry
, new_entry
, sizeof *entry
)) {
1356 *entry
= *new_entry
;
1364 void kvm_irqchip_add_irq_route(KVMState
*s
, int irq
, int irqchip
, int pin
)
1366 struct kvm_irq_routing_entry e
= {};
1368 assert(pin
< s
->gsi_count
);
1371 e
.type
= KVM_IRQ_ROUTING_IRQCHIP
;
1373 e
.u
.irqchip
.irqchip
= irqchip
;
1374 e
.u
.irqchip
.pin
= pin
;
1375 kvm_add_routing_entry(s
, &e
);
1378 void kvm_irqchip_release_virq(KVMState
*s
, int virq
)
1380 struct kvm_irq_routing_entry
*e
;
1383 if (kvm_gsi_direct_mapping()) {
1387 for (i
= 0; i
< s
->irq_routes
->nr
; i
++) {
1388 e
= &s
->irq_routes
->entries
[i
];
1389 if (e
->gsi
== virq
) {
1390 s
->irq_routes
->nr
--;
1391 *e
= s
->irq_routes
->entries
[s
->irq_routes
->nr
];
1395 kvm_arch_release_virq_post(virq
);
1396 trace_kvm_irqchip_release_virq(virq
);
1399 static unsigned int kvm_hash_msi(uint32_t data
)
1401 /* This is optimized for IA32 MSI layout. However, no other arch shall
1402 * repeat the mistake of not providing a direct MSI injection API. */
1406 static void kvm_flush_dynamic_msi_routes(KVMState
*s
)
1408 KVMMSIRoute
*route
, *next
;
1411 for (hash
= 0; hash
< KVM_MSI_HASHTAB_SIZE
; hash
++) {
1412 QTAILQ_FOREACH_SAFE(route
, &s
->msi_hashtab
[hash
], entry
, next
) {
1413 kvm_irqchip_release_virq(s
, route
->kroute
.gsi
);
1414 QTAILQ_REMOVE(&s
->msi_hashtab
[hash
], route
, entry
);
1420 static int kvm_irqchip_get_virq(KVMState
*s
)
1425 * PIC and IOAPIC share the first 16 GSI numbers, thus the available
1426 * GSI numbers are more than the number of IRQ route. Allocating a GSI
1427 * number can succeed even though a new route entry cannot be added.
1428 * When this happens, flush dynamic MSI entries to free IRQ route entries.
1430 if (!kvm_direct_msi_allowed
&& s
->irq_routes
->nr
== s
->gsi_count
) {
1431 kvm_flush_dynamic_msi_routes(s
);
1434 /* Return the lowest unused GSI in the bitmap */
1435 next_virq
= find_first_zero_bit(s
->used_gsi_bitmap
, s
->gsi_count
);
1436 if (next_virq
>= s
->gsi_count
) {
1443 static KVMMSIRoute
*kvm_lookup_msi_route(KVMState
*s
, MSIMessage msg
)
1445 unsigned int hash
= kvm_hash_msi(msg
.data
);
1448 QTAILQ_FOREACH(route
, &s
->msi_hashtab
[hash
], entry
) {
1449 if (route
->kroute
.u
.msi
.address_lo
== (uint32_t)msg
.address
&&
1450 route
->kroute
.u
.msi
.address_hi
== (msg
.address
>> 32) &&
1451 route
->kroute
.u
.msi
.data
== le32_to_cpu(msg
.data
)) {
1458 int kvm_irqchip_send_msi(KVMState
*s
, MSIMessage msg
)
1463 if (kvm_direct_msi_allowed
) {
1464 msi
.address_lo
= (uint32_t)msg
.address
;
1465 msi
.address_hi
= msg
.address
>> 32;
1466 msi
.data
= le32_to_cpu(msg
.data
);
1468 memset(msi
.pad
, 0, sizeof(msi
.pad
));
1470 return kvm_vm_ioctl(s
, KVM_SIGNAL_MSI
, &msi
);
1473 route
= kvm_lookup_msi_route(s
, msg
);
1477 virq
= kvm_irqchip_get_virq(s
);
1482 route
= g_malloc0(sizeof(KVMMSIRoute
));
1483 route
->kroute
.gsi
= virq
;
1484 route
->kroute
.type
= KVM_IRQ_ROUTING_MSI
;
1485 route
->kroute
.flags
= 0;
1486 route
->kroute
.u
.msi
.address_lo
= (uint32_t)msg
.address
;
1487 route
->kroute
.u
.msi
.address_hi
= msg
.address
>> 32;
1488 route
->kroute
.u
.msi
.data
= le32_to_cpu(msg
.data
);
1490 kvm_add_routing_entry(s
, &route
->kroute
);
1491 kvm_irqchip_commit_routes(s
);
1493 QTAILQ_INSERT_TAIL(&s
->msi_hashtab
[kvm_hash_msi(msg
.data
)], route
,
1497 assert(route
->kroute
.type
== KVM_IRQ_ROUTING_MSI
);
1499 return kvm_set_irq(s
, route
->kroute
.gsi
, 1);
1502 int kvm_irqchip_add_msi_route(KVMState
*s
, int vector
, PCIDevice
*dev
)
1504 struct kvm_irq_routing_entry kroute
= {};
1506 MSIMessage msg
= {0, 0};
1508 if (pci_available
&& dev
) {
1509 msg
= pci_get_msi_message(dev
, vector
);
1512 if (kvm_gsi_direct_mapping()) {
1513 return kvm_arch_msi_data_to_gsi(msg
.data
);
1516 if (!kvm_gsi_routing_enabled()) {
1520 virq
= kvm_irqchip_get_virq(s
);
1526 kroute
.type
= KVM_IRQ_ROUTING_MSI
;
1528 kroute
.u
.msi
.address_lo
= (uint32_t)msg
.address
;
1529 kroute
.u
.msi
.address_hi
= msg
.address
>> 32;
1530 kroute
.u
.msi
.data
= le32_to_cpu(msg
.data
);
1531 if (pci_available
&& kvm_msi_devid_required()) {
1532 kroute
.flags
= KVM_MSI_VALID_DEVID
;
1533 kroute
.u
.msi
.devid
= pci_requester_id(dev
);
1535 if (kvm_arch_fixup_msi_route(&kroute
, msg
.address
, msg
.data
, dev
)) {
1536 kvm_irqchip_release_virq(s
, virq
);
1540 trace_kvm_irqchip_add_msi_route(dev
? dev
->name
: (char *)"N/A",
1543 kvm_add_routing_entry(s
, &kroute
);
1544 kvm_arch_add_msi_route_post(&kroute
, vector
, dev
);
1545 kvm_irqchip_commit_routes(s
);
1550 int kvm_irqchip_update_msi_route(KVMState
*s
, int virq
, MSIMessage msg
,
1553 struct kvm_irq_routing_entry kroute
= {};
1555 if (kvm_gsi_direct_mapping()) {
1559 if (!kvm_irqchip_in_kernel()) {
1564 kroute
.type
= KVM_IRQ_ROUTING_MSI
;
1566 kroute
.u
.msi
.address_lo
= (uint32_t)msg
.address
;
1567 kroute
.u
.msi
.address_hi
= msg
.address
>> 32;
1568 kroute
.u
.msi
.data
= le32_to_cpu(msg
.data
);
1569 if (pci_available
&& kvm_msi_devid_required()) {
1570 kroute
.flags
= KVM_MSI_VALID_DEVID
;
1571 kroute
.u
.msi
.devid
= pci_requester_id(dev
);
1573 if (kvm_arch_fixup_msi_route(&kroute
, msg
.address
, msg
.data
, dev
)) {
1577 trace_kvm_irqchip_update_msi_route(virq
);
1579 return kvm_update_routing_entry(s
, &kroute
);
1582 static int kvm_irqchip_assign_irqfd(KVMState
*s
, int fd
, int rfd
, int virq
,
1585 struct kvm_irqfd irqfd
= {
1588 .flags
= assign
? 0 : KVM_IRQFD_FLAG_DEASSIGN
,
1592 irqfd
.flags
|= KVM_IRQFD_FLAG_RESAMPLE
;
1593 irqfd
.resamplefd
= rfd
;
1596 if (!kvm_irqfds_enabled()) {
1600 return kvm_vm_ioctl(s
, KVM_IRQFD
, &irqfd
);
1603 int kvm_irqchip_add_adapter_route(KVMState
*s
, AdapterInfo
*adapter
)
1605 struct kvm_irq_routing_entry kroute
= {};
1608 if (!kvm_gsi_routing_enabled()) {
1612 virq
= kvm_irqchip_get_virq(s
);
1618 kroute
.type
= KVM_IRQ_ROUTING_S390_ADAPTER
;
1620 kroute
.u
.adapter
.summary_addr
= adapter
->summary_addr
;
1621 kroute
.u
.adapter
.ind_addr
= adapter
->ind_addr
;
1622 kroute
.u
.adapter
.summary_offset
= adapter
->summary_offset
;
1623 kroute
.u
.adapter
.ind_offset
= adapter
->ind_offset
;
1624 kroute
.u
.adapter
.adapter_id
= adapter
->adapter_id
;
1626 kvm_add_routing_entry(s
, &kroute
);
1631 int kvm_irqchip_add_hv_sint_route(KVMState
*s
, uint32_t vcpu
, uint32_t sint
)
1633 struct kvm_irq_routing_entry kroute
= {};
1636 if (!kvm_gsi_routing_enabled()) {
1639 if (!kvm_check_extension(s
, KVM_CAP_HYPERV_SYNIC
)) {
1642 virq
= kvm_irqchip_get_virq(s
);
1648 kroute
.type
= KVM_IRQ_ROUTING_HV_SINT
;
1650 kroute
.u
.hv_sint
.vcpu
= vcpu
;
1651 kroute
.u
.hv_sint
.sint
= sint
;
1653 kvm_add_routing_entry(s
, &kroute
);
1654 kvm_irqchip_commit_routes(s
);
1659 #else /* !KVM_CAP_IRQ_ROUTING */
1661 void kvm_init_irq_routing(KVMState
*s
)
1665 void kvm_irqchip_release_virq(KVMState
*s
, int virq
)
1669 int kvm_irqchip_send_msi(KVMState
*s
, MSIMessage msg
)
1674 int kvm_irqchip_add_msi_route(KVMState
*s
, int vector
, PCIDevice
*dev
)
1679 int kvm_irqchip_add_adapter_route(KVMState
*s
, AdapterInfo
*adapter
)
1684 int kvm_irqchip_add_hv_sint_route(KVMState
*s
, uint32_t vcpu
, uint32_t sint
)
1689 static int kvm_irqchip_assign_irqfd(KVMState
*s
, int fd
, int virq
, bool assign
)
1694 int kvm_irqchip_update_msi_route(KVMState
*s
, int virq
, MSIMessage msg
)
1698 #endif /* !KVM_CAP_IRQ_ROUTING */
1700 int kvm_irqchip_add_irqfd_notifier_gsi(KVMState
*s
, EventNotifier
*n
,
1701 EventNotifier
*rn
, int virq
)
1703 return kvm_irqchip_assign_irqfd(s
, event_notifier_get_fd(n
),
1704 rn
? event_notifier_get_fd(rn
) : -1, virq
, true);
1707 int kvm_irqchip_remove_irqfd_notifier_gsi(KVMState
*s
, EventNotifier
*n
,
1710 return kvm_irqchip_assign_irqfd(s
, event_notifier_get_fd(n
), -1, virq
,
1714 int kvm_irqchip_add_irqfd_notifier(KVMState
*s
, EventNotifier
*n
,
1715 EventNotifier
*rn
, qemu_irq irq
)
1718 gboolean found
= g_hash_table_lookup_extended(s
->gsimap
, irq
, &key
, &gsi
);
1723 return kvm_irqchip_add_irqfd_notifier_gsi(s
, n
, rn
, GPOINTER_TO_INT(gsi
));
1726 int kvm_irqchip_remove_irqfd_notifier(KVMState
*s
, EventNotifier
*n
,
1730 gboolean found
= g_hash_table_lookup_extended(s
->gsimap
, irq
, &key
, &gsi
);
1735 return kvm_irqchip_remove_irqfd_notifier_gsi(s
, n
, GPOINTER_TO_INT(gsi
));
1738 void kvm_irqchip_set_qemuirq_gsi(KVMState
*s
, qemu_irq irq
, int gsi
)
1740 g_hash_table_insert(s
->gsimap
, irq
, GINT_TO_POINTER(gsi
));
1743 static void kvm_irqchip_create(MachineState
*machine
, KVMState
*s
)
1747 if (kvm_check_extension(s
, KVM_CAP_IRQCHIP
)) {
1749 } else if (kvm_check_extension(s
, KVM_CAP_S390_IRQCHIP
)) {
1750 ret
= kvm_vm_enable_cap(s
, KVM_CAP_S390_IRQCHIP
, 0);
1752 fprintf(stderr
, "Enable kernel irqchip failed: %s\n", strerror(-ret
));
1759 /* First probe and see if there's a arch-specific hook to create the
1760 * in-kernel irqchip for us */
1761 ret
= kvm_arch_irqchip_create(machine
, s
);
1763 if (machine_kernel_irqchip_split(machine
)) {
1764 perror("Split IRQ chip mode not supported.");
1767 ret
= kvm_vm_ioctl(s
, KVM_CREATE_IRQCHIP
);
1771 fprintf(stderr
, "Create kernel irqchip failed: %s\n", strerror(-ret
));
1775 kvm_kernel_irqchip
= true;
1776 /* If we have an in-kernel IRQ chip then we must have asynchronous
1777 * interrupt delivery (though the reverse is not necessarily true)
1779 kvm_async_interrupts_allowed
= true;
1780 kvm_halt_in_kernel_allowed
= true;
1782 kvm_init_irq_routing(s
);
1784 s
->gsimap
= g_hash_table_new(g_direct_hash
, g_direct_equal
);
1787 /* Find number of supported CPUs using the recommended
1788 * procedure from the kernel API documentation to cope with
1789 * older kernels that may be missing capabilities.
1791 static int kvm_recommended_vcpus(KVMState
*s
)
1793 int ret
= kvm_vm_check_extension(s
, KVM_CAP_NR_VCPUS
);
1794 return (ret
) ? ret
: 4;
1797 static int kvm_max_vcpus(KVMState
*s
)
1799 int ret
= kvm_check_extension(s
, KVM_CAP_MAX_VCPUS
);
1800 return (ret
) ? ret
: kvm_recommended_vcpus(s
);
1803 static int kvm_max_vcpu_id(KVMState
*s
)
1805 int ret
= kvm_check_extension(s
, KVM_CAP_MAX_VCPU_ID
);
1806 return (ret
) ? ret
: kvm_max_vcpus(s
);
1809 bool kvm_vcpu_id_is_valid(int vcpu_id
)
1811 KVMState
*s
= KVM_STATE(current_machine
->accelerator
);
1812 return vcpu_id
>= 0 && vcpu_id
< kvm_max_vcpu_id(s
);
1815 static int kvm_init(MachineState
*ms
)
1817 MachineClass
*mc
= MACHINE_GET_CLASS(ms
);
1818 static const char upgrade_note
[] =
1819 "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n"
1820 "(see http://sourceforge.net/projects/kvm).\n";
1825 { "SMP", ms
->smp
.cpus
},
1826 { "hotpluggable", ms
->smp
.max_cpus
},
1829 int soft_vcpus_limit
, hard_vcpus_limit
;
1831 const KVMCapabilityInfo
*missing_cap
;
1834 const char *kvm_type
;
1836 s
= KVM_STATE(ms
->accelerator
);
1839 * On systems where the kernel can support different base page
1840 * sizes, host page size may be different from TARGET_PAGE_SIZE,
1841 * even with KVM. TARGET_PAGE_SIZE is assumed to be the minimum
1842 * page size for the system though.
1844 assert(TARGET_PAGE_SIZE
<= getpagesize());
1848 #ifdef KVM_CAP_SET_GUEST_DEBUG
1849 QTAILQ_INIT(&s
->kvm_sw_breakpoints
);
1851 QLIST_INIT(&s
->kvm_parked_vcpus
);
1853 s
->fd
= qemu_open("/dev/kvm", O_RDWR
);
1855 fprintf(stderr
, "Could not access KVM kernel module: %m\n");
1860 ret
= kvm_ioctl(s
, KVM_GET_API_VERSION
, 0);
1861 if (ret
< KVM_API_VERSION
) {
1865 fprintf(stderr
, "kvm version too old\n");
1869 if (ret
> KVM_API_VERSION
) {
1871 fprintf(stderr
, "kvm version not supported\n");
1875 kvm_immediate_exit
= kvm_check_extension(s
, KVM_CAP_IMMEDIATE_EXIT
);
1876 s
->nr_slots
= kvm_check_extension(s
, KVM_CAP_NR_MEMSLOTS
);
1878 /* If unspecified, use the default value */
1883 s
->nr_as
= kvm_check_extension(s
, KVM_CAP_MULTI_ADDRESS_SPACE
);
1884 if (s
->nr_as
<= 1) {
1887 s
->as
= g_new0(struct KVMAs
, s
->nr_as
);
1889 kvm_type
= qemu_opt_get(qemu_get_machine_opts(), "kvm-type");
1891 type
= mc
->kvm_type(ms
, kvm_type
);
1892 } else if (kvm_type
) {
1894 fprintf(stderr
, "Invalid argument kvm-type=%s\n", kvm_type
);
1899 ret
= kvm_ioctl(s
, KVM_CREATE_VM
, type
);
1900 } while (ret
== -EINTR
);
1903 fprintf(stderr
, "ioctl(KVM_CREATE_VM) failed: %d %s\n", -ret
,
1907 if (ret
== -EINVAL
) {
1909 "Host kernel setup problem detected. Please verify:\n");
1910 fprintf(stderr
, "- for kernels supporting the switch_amode or"
1911 " user_mode parameters, whether\n");
1913 " user space is running in primary address space\n");
1915 "- for kernels supporting the vm.allocate_pgste sysctl, "
1916 "whether it is enabled\n");
1924 /* check the vcpu limits */
1925 soft_vcpus_limit
= kvm_recommended_vcpus(s
);
1926 hard_vcpus_limit
= kvm_max_vcpus(s
);
1929 if (nc
->num
> soft_vcpus_limit
) {
1930 warn_report("Number of %s cpus requested (%d) exceeds "
1931 "the recommended cpus supported by KVM (%d)",
1932 nc
->name
, nc
->num
, soft_vcpus_limit
);
1934 if (nc
->num
> hard_vcpus_limit
) {
1935 fprintf(stderr
, "Number of %s cpus requested (%d) exceeds "
1936 "the maximum cpus supported by KVM (%d)\n",
1937 nc
->name
, nc
->num
, hard_vcpus_limit
);
1944 missing_cap
= kvm_check_extension_list(s
, kvm_required_capabilites
);
1947 kvm_check_extension_list(s
, kvm_arch_required_capabilities
);
1951 fprintf(stderr
, "kvm does not support %s\n%s",
1952 missing_cap
->name
, upgrade_note
);
1956 s
->coalesced_mmio
= kvm_check_extension(s
, KVM_CAP_COALESCED_MMIO
);
1957 s
->coalesced_pio
= s
->coalesced_mmio
&&
1958 kvm_check_extension(s
, KVM_CAP_COALESCED_PIO
);
1960 s
->manual_dirty_log_protect
=
1961 kvm_check_extension(s
, KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2
);
1962 if (s
->manual_dirty_log_protect
) {
1963 ret
= kvm_vm_enable_cap(s
, KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2
, 0, 1);
1965 warn_report("Trying to enable KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2 "
1966 "but failed. Falling back to the legacy mode. ");
1967 s
->manual_dirty_log_protect
= false;
1971 #ifdef KVM_CAP_VCPU_EVENTS
1972 s
->vcpu_events
= kvm_check_extension(s
, KVM_CAP_VCPU_EVENTS
);
1975 s
->robust_singlestep
=
1976 kvm_check_extension(s
, KVM_CAP_X86_ROBUST_SINGLESTEP
);
1978 #ifdef KVM_CAP_DEBUGREGS
1979 s
->debugregs
= kvm_check_extension(s
, KVM_CAP_DEBUGREGS
);
1982 s
->max_nested_state_len
= kvm_check_extension(s
, KVM_CAP_NESTED_STATE
);
1984 #ifdef KVM_CAP_IRQ_ROUTING
1985 kvm_direct_msi_allowed
= (kvm_check_extension(s
, KVM_CAP_SIGNAL_MSI
) > 0);
1988 s
->intx_set_mask
= kvm_check_extension(s
, KVM_CAP_PCI_2_3
);
1990 s
->irq_set_ioctl
= KVM_IRQ_LINE
;
1991 if (kvm_check_extension(s
, KVM_CAP_IRQ_INJECT_STATUS
)) {
1992 s
->irq_set_ioctl
= KVM_IRQ_LINE_STATUS
;
1995 kvm_readonly_mem_allowed
=
1996 (kvm_check_extension(s
, KVM_CAP_READONLY_MEM
) > 0);
1998 kvm_eventfds_allowed
=
1999 (kvm_check_extension(s
, KVM_CAP_IOEVENTFD
) > 0);
2001 kvm_irqfds_allowed
=
2002 (kvm_check_extension(s
, KVM_CAP_IRQFD
) > 0);
2004 kvm_resamplefds_allowed
=
2005 (kvm_check_extension(s
, KVM_CAP_IRQFD_RESAMPLE
) > 0);
2007 kvm_vm_attributes_allowed
=
2008 (kvm_check_extension(s
, KVM_CAP_VM_ATTRIBUTES
) > 0);
2010 kvm_ioeventfd_any_length_allowed
=
2011 (kvm_check_extension(s
, KVM_CAP_IOEVENTFD_ANY_LENGTH
) > 0);
2016 * if memory encryption object is specified then initialize the memory
2017 * encryption context.
2019 if (ms
->memory_encryption
) {
2020 kvm_state
->memcrypt_handle
= sev_guest_init(ms
->memory_encryption
);
2021 if (!kvm_state
->memcrypt_handle
) {
2026 kvm_state
->memcrypt_encrypt_data
= sev_encrypt_data
;
2029 ret
= kvm_arch_init(ms
, s
);
2034 if (machine_kernel_irqchip_allowed(ms
)) {
2035 kvm_irqchip_create(ms
, s
);
2038 if (kvm_eventfds_allowed
) {
2039 s
->memory_listener
.listener
.eventfd_add
= kvm_mem_ioeventfd_add
;
2040 s
->memory_listener
.listener
.eventfd_del
= kvm_mem_ioeventfd_del
;
2042 s
->memory_listener
.listener
.coalesced_io_add
= kvm_coalesce_mmio_region
;
2043 s
->memory_listener
.listener
.coalesced_io_del
= kvm_uncoalesce_mmio_region
;
2045 kvm_memory_listener_register(s
, &s
->memory_listener
,
2046 &address_space_memory
, 0);
2047 memory_listener_register(&kvm_io_listener
,
2049 memory_listener_register(&kvm_coalesced_pio_listener
,
2052 s
->many_ioeventfds
= kvm_check_many_ioeventfds();
2054 s
->sync_mmu
= !!kvm_vm_check_extension(kvm_state
, KVM_CAP_SYNC_MMU
);
2056 qemu_balloon_inhibit(true);
2069 g_free(s
->memory_listener
.slots
);
2074 void kvm_set_sigmask_len(KVMState
*s
, unsigned int sigmask_len
)
2076 s
->sigmask_len
= sigmask_len
;
2079 static void kvm_handle_io(uint16_t port
, MemTxAttrs attrs
, void *data
, int direction
,
2080 int size
, uint32_t count
)
2083 uint8_t *ptr
= data
;
2085 for (i
= 0; i
< count
; i
++) {
2086 address_space_rw(&address_space_io
, port
, attrs
,
2088 direction
== KVM_EXIT_IO_OUT
);
2093 static int kvm_handle_internal_error(CPUState
*cpu
, struct kvm_run
*run
)
2095 fprintf(stderr
, "KVM internal error. Suberror: %d\n",
2096 run
->internal
.suberror
);
2098 if (kvm_check_extension(kvm_state
, KVM_CAP_INTERNAL_ERROR_DATA
)) {
2101 for (i
= 0; i
< run
->internal
.ndata
; ++i
) {
2102 fprintf(stderr
, "extra data[%d]: %"PRIx64
"\n",
2103 i
, (uint64_t)run
->internal
.data
[i
]);
2106 if (run
->internal
.suberror
== KVM_INTERNAL_ERROR_EMULATION
) {
2107 fprintf(stderr
, "emulation failure\n");
2108 if (!kvm_arch_stop_on_emulation_error(cpu
)) {
2109 cpu_dump_state(cpu
, stderr
, CPU_DUMP_CODE
);
2110 return EXCP_INTERRUPT
;
2113 /* FIXME: Should trigger a qmp message to let management know
2114 * something went wrong.
2119 void kvm_flush_coalesced_mmio_buffer(void)
2121 KVMState
*s
= kvm_state
;
2123 if (s
->coalesced_flush_in_progress
) {
2127 s
->coalesced_flush_in_progress
= true;
2129 if (s
->coalesced_mmio_ring
) {
2130 struct kvm_coalesced_mmio_ring
*ring
= s
->coalesced_mmio_ring
;
2131 while (ring
->first
!= ring
->last
) {
2132 struct kvm_coalesced_mmio
*ent
;
2134 ent
= &ring
->coalesced_mmio
[ring
->first
];
2136 if (ent
->pio
== 1) {
2137 address_space_rw(&address_space_io
, ent
->phys_addr
,
2138 MEMTXATTRS_UNSPECIFIED
, ent
->data
,
2141 cpu_physical_memory_write(ent
->phys_addr
, ent
->data
, ent
->len
);
2144 ring
->first
= (ring
->first
+ 1) % KVM_COALESCED_MMIO_MAX
;
2148 s
->coalesced_flush_in_progress
= false;
2151 static void do_kvm_cpu_synchronize_state(CPUState
*cpu
, run_on_cpu_data arg
)
2153 if (!cpu
->vcpu_dirty
) {
2154 kvm_arch_get_registers(cpu
);
2155 cpu
->vcpu_dirty
= true;
2159 void kvm_cpu_synchronize_state(CPUState
*cpu
)
2161 if (!cpu
->vcpu_dirty
) {
2162 run_on_cpu(cpu
, do_kvm_cpu_synchronize_state
, RUN_ON_CPU_NULL
);
2166 static void do_kvm_cpu_synchronize_post_reset(CPUState
*cpu
, run_on_cpu_data arg
)
2168 kvm_arch_put_registers(cpu
, KVM_PUT_RESET_STATE
);
2169 cpu
->vcpu_dirty
= false;
2172 void kvm_cpu_synchronize_post_reset(CPUState
*cpu
)
2174 run_on_cpu(cpu
, do_kvm_cpu_synchronize_post_reset
, RUN_ON_CPU_NULL
);
2177 static void do_kvm_cpu_synchronize_post_init(CPUState
*cpu
, run_on_cpu_data arg
)
2179 kvm_arch_put_registers(cpu
, KVM_PUT_FULL_STATE
);
2180 cpu
->vcpu_dirty
= false;
2183 void kvm_cpu_synchronize_post_init(CPUState
*cpu
)
2185 run_on_cpu(cpu
, do_kvm_cpu_synchronize_post_init
, RUN_ON_CPU_NULL
);
2188 static void do_kvm_cpu_synchronize_pre_loadvm(CPUState
*cpu
, run_on_cpu_data arg
)
2190 cpu
->vcpu_dirty
= true;
2193 void kvm_cpu_synchronize_pre_loadvm(CPUState
*cpu
)
2195 run_on_cpu(cpu
, do_kvm_cpu_synchronize_pre_loadvm
, RUN_ON_CPU_NULL
);
2198 #ifdef KVM_HAVE_MCE_INJECTION
2199 static __thread
void *pending_sigbus_addr
;
2200 static __thread
int pending_sigbus_code
;
2201 static __thread
bool have_sigbus_pending
;
2204 static void kvm_cpu_kick(CPUState
*cpu
)
2206 atomic_set(&cpu
->kvm_run
->immediate_exit
, 1);
2209 static void kvm_cpu_kick_self(void)
2211 if (kvm_immediate_exit
) {
2212 kvm_cpu_kick(current_cpu
);
2214 qemu_cpu_kick_self();
2218 static void kvm_eat_signals(CPUState
*cpu
)
2220 struct timespec ts
= { 0, 0 };
2226 if (kvm_immediate_exit
) {
2227 atomic_set(&cpu
->kvm_run
->immediate_exit
, 0);
2228 /* Write kvm_run->immediate_exit before the cpu->exit_request
2229 * write in kvm_cpu_exec.
2235 sigemptyset(&waitset
);
2236 sigaddset(&waitset
, SIG_IPI
);
2239 r
= sigtimedwait(&waitset
, &siginfo
, &ts
);
2240 if (r
== -1 && !(errno
== EAGAIN
|| errno
== EINTR
)) {
2241 perror("sigtimedwait");
2245 r
= sigpending(&chkset
);
2247 perror("sigpending");
2250 } while (sigismember(&chkset
, SIG_IPI
));
2253 int kvm_cpu_exec(CPUState
*cpu
)
2255 struct kvm_run
*run
= cpu
->kvm_run
;
2258 DPRINTF("kvm_cpu_exec()\n");
2260 if (kvm_arch_process_async_events(cpu
)) {
2261 atomic_set(&cpu
->exit_request
, 0);
2265 qemu_mutex_unlock_iothread();
2266 cpu_exec_start(cpu
);
2271 if (cpu
->vcpu_dirty
) {
2272 kvm_arch_put_registers(cpu
, KVM_PUT_RUNTIME_STATE
);
2273 cpu
->vcpu_dirty
= false;
2276 kvm_arch_pre_run(cpu
, run
);
2277 if (atomic_read(&cpu
->exit_request
)) {
2278 DPRINTF("interrupt exit requested\n");
2280 * KVM requires us to reenter the kernel after IO exits to complete
2281 * instruction emulation. This self-signal will ensure that we
2284 kvm_cpu_kick_self();
2287 /* Read cpu->exit_request before KVM_RUN reads run->immediate_exit.
2288 * Matching barrier in kvm_eat_signals.
2292 run_ret
= kvm_vcpu_ioctl(cpu
, KVM_RUN
, 0);
2294 attrs
= kvm_arch_post_run(cpu
, run
);
2296 #ifdef KVM_HAVE_MCE_INJECTION
2297 if (unlikely(have_sigbus_pending
)) {
2298 qemu_mutex_lock_iothread();
2299 kvm_arch_on_sigbus_vcpu(cpu
, pending_sigbus_code
,
2300 pending_sigbus_addr
);
2301 have_sigbus_pending
= false;
2302 qemu_mutex_unlock_iothread();
2307 if (run_ret
== -EINTR
|| run_ret
== -EAGAIN
) {
2308 DPRINTF("io window exit\n");
2309 kvm_eat_signals(cpu
);
2310 ret
= EXCP_INTERRUPT
;
2313 fprintf(stderr
, "error: kvm run failed %s\n",
2314 strerror(-run_ret
));
2316 if (run_ret
== -EBUSY
) {
2318 "This is probably because your SMT is enabled.\n"
2319 "VCPU can only run on primary threads with all "
2320 "secondary threads offline.\n");
2327 trace_kvm_run_exit(cpu
->cpu_index
, run
->exit_reason
);
2328 switch (run
->exit_reason
) {
2330 DPRINTF("handle_io\n");
2331 /* Called outside BQL */
2332 kvm_handle_io(run
->io
.port
, attrs
,
2333 (uint8_t *)run
+ run
->io
.data_offset
,
2340 DPRINTF("handle_mmio\n");
2341 /* Called outside BQL */
2342 address_space_rw(&address_space_memory
,
2343 run
->mmio
.phys_addr
, attrs
,
2346 run
->mmio
.is_write
);
2349 case KVM_EXIT_IRQ_WINDOW_OPEN
:
2350 DPRINTF("irq_window_open\n");
2351 ret
= EXCP_INTERRUPT
;
2353 case KVM_EXIT_SHUTDOWN
:
2354 DPRINTF("shutdown\n");
2355 qemu_system_reset_request(SHUTDOWN_CAUSE_GUEST_RESET
);
2356 ret
= EXCP_INTERRUPT
;
2358 case KVM_EXIT_UNKNOWN
:
2359 fprintf(stderr
, "KVM: unknown exit, hardware reason %" PRIx64
"\n",
2360 (uint64_t)run
->hw
.hardware_exit_reason
);
2363 case KVM_EXIT_INTERNAL_ERROR
:
2364 ret
= kvm_handle_internal_error(cpu
, run
);
2366 case KVM_EXIT_SYSTEM_EVENT
:
2367 switch (run
->system_event
.type
) {
2368 case KVM_SYSTEM_EVENT_SHUTDOWN
:
2369 qemu_system_shutdown_request(SHUTDOWN_CAUSE_GUEST_SHUTDOWN
);
2370 ret
= EXCP_INTERRUPT
;
2372 case KVM_SYSTEM_EVENT_RESET
:
2373 qemu_system_reset_request(SHUTDOWN_CAUSE_GUEST_RESET
);
2374 ret
= EXCP_INTERRUPT
;
2376 case KVM_SYSTEM_EVENT_CRASH
:
2377 kvm_cpu_synchronize_state(cpu
);
2378 qemu_mutex_lock_iothread();
2379 qemu_system_guest_panicked(cpu_get_crash_info(cpu
));
2380 qemu_mutex_unlock_iothread();
2384 DPRINTF("kvm_arch_handle_exit\n");
2385 ret
= kvm_arch_handle_exit(cpu
, run
);
2390 DPRINTF("kvm_arch_handle_exit\n");
2391 ret
= kvm_arch_handle_exit(cpu
, run
);
2397 qemu_mutex_lock_iothread();
2400 cpu_dump_state(cpu
, stderr
, CPU_DUMP_CODE
);
2401 vm_stop(RUN_STATE_INTERNAL_ERROR
);
2404 atomic_set(&cpu
->exit_request
, 0);
2408 int kvm_ioctl(KVMState
*s
, int type
, ...)
2415 arg
= va_arg(ap
, void *);
2418 trace_kvm_ioctl(type
, arg
);
2419 ret
= ioctl(s
->fd
, type
, arg
);
2426 int kvm_vm_ioctl(KVMState
*s
, int type
, ...)
2433 arg
= va_arg(ap
, void *);
2436 trace_kvm_vm_ioctl(type
, arg
);
2437 ret
= ioctl(s
->vmfd
, type
, arg
);
2444 int kvm_vcpu_ioctl(CPUState
*cpu
, int type
, ...)
2451 arg
= va_arg(ap
, void *);
2454 trace_kvm_vcpu_ioctl(cpu
->cpu_index
, type
, arg
);
2455 ret
= ioctl(cpu
->kvm_fd
, type
, arg
);
2462 int kvm_device_ioctl(int fd
, int type
, ...)
2469 arg
= va_arg(ap
, void *);
2472 trace_kvm_device_ioctl(fd
, type
, arg
);
2473 ret
= ioctl(fd
, type
, arg
);
2480 int kvm_vm_check_attr(KVMState
*s
, uint32_t group
, uint64_t attr
)
2483 struct kvm_device_attr attribute
= {
2488 if (!kvm_vm_attributes_allowed
) {
2492 ret
= kvm_vm_ioctl(s
, KVM_HAS_DEVICE_ATTR
, &attribute
);
2493 /* kvm returns 0 on success for HAS_DEVICE_ATTR */
2497 int kvm_device_check_attr(int dev_fd
, uint32_t group
, uint64_t attr
)
2499 struct kvm_device_attr attribute
= {
2505 return kvm_device_ioctl(dev_fd
, KVM_HAS_DEVICE_ATTR
, &attribute
) ? 0 : 1;
2508 int kvm_device_access(int fd
, int group
, uint64_t attr
,
2509 void *val
, bool write
, Error
**errp
)
2511 struct kvm_device_attr kvmattr
;
2515 kvmattr
.group
= group
;
2516 kvmattr
.attr
= attr
;
2517 kvmattr
.addr
= (uintptr_t)val
;
2519 err
= kvm_device_ioctl(fd
,
2520 write
? KVM_SET_DEVICE_ATTR
: KVM_GET_DEVICE_ATTR
,
2523 error_setg_errno(errp
, -err
,
2524 "KVM_%s_DEVICE_ATTR failed: Group %d "
2525 "attr 0x%016" PRIx64
,
2526 write
? "SET" : "GET", group
, attr
);
2531 bool kvm_has_sync_mmu(void)
2533 return kvm_state
->sync_mmu
;
2536 int kvm_has_vcpu_events(void)
2538 return kvm_state
->vcpu_events
;
2541 int kvm_has_robust_singlestep(void)
2543 return kvm_state
->robust_singlestep
;
2546 int kvm_has_debugregs(void)
2548 return kvm_state
->debugregs
;
2551 int kvm_max_nested_state_length(void)
2553 return kvm_state
->max_nested_state_len
;
2556 int kvm_has_many_ioeventfds(void)
2558 if (!kvm_enabled()) {
2561 return kvm_state
->many_ioeventfds
;
2564 int kvm_has_gsi_routing(void)
2566 #ifdef KVM_CAP_IRQ_ROUTING
2567 return kvm_check_extension(kvm_state
, KVM_CAP_IRQ_ROUTING
);
2573 int kvm_has_intx_set_mask(void)
2575 return kvm_state
->intx_set_mask
;
2578 bool kvm_arm_supports_user_irq(void)
2580 return kvm_check_extension(kvm_state
, KVM_CAP_ARM_USER_IRQ
);
2583 #ifdef KVM_CAP_SET_GUEST_DEBUG
2584 struct kvm_sw_breakpoint
*kvm_find_sw_breakpoint(CPUState
*cpu
,
2587 struct kvm_sw_breakpoint
*bp
;
2589 QTAILQ_FOREACH(bp
, &cpu
->kvm_state
->kvm_sw_breakpoints
, entry
) {
2597 int kvm_sw_breakpoints_active(CPUState
*cpu
)
2599 return !QTAILQ_EMPTY(&cpu
->kvm_state
->kvm_sw_breakpoints
);
2602 struct kvm_set_guest_debug_data
{
2603 struct kvm_guest_debug dbg
;
2607 static void kvm_invoke_set_guest_debug(CPUState
*cpu
, run_on_cpu_data data
)
2609 struct kvm_set_guest_debug_data
*dbg_data
=
2610 (struct kvm_set_guest_debug_data
*) data
.host_ptr
;
2612 dbg_data
->err
= kvm_vcpu_ioctl(cpu
, KVM_SET_GUEST_DEBUG
,
2616 int kvm_update_guest_debug(CPUState
*cpu
, unsigned long reinject_trap
)
2618 struct kvm_set_guest_debug_data data
;
2620 data
.dbg
.control
= reinject_trap
;
2622 if (cpu
->singlestep_enabled
) {
2623 data
.dbg
.control
|= KVM_GUESTDBG_ENABLE
| KVM_GUESTDBG_SINGLESTEP
;
2625 kvm_arch_update_guest_debug(cpu
, &data
.dbg
);
2627 run_on_cpu(cpu
, kvm_invoke_set_guest_debug
,
2628 RUN_ON_CPU_HOST_PTR(&data
));
2632 int kvm_insert_breakpoint(CPUState
*cpu
, target_ulong addr
,
2633 target_ulong len
, int type
)
2635 struct kvm_sw_breakpoint
*bp
;
2638 if (type
== GDB_BREAKPOINT_SW
) {
2639 bp
= kvm_find_sw_breakpoint(cpu
, addr
);
2645 bp
= g_malloc(sizeof(struct kvm_sw_breakpoint
));
2648 err
= kvm_arch_insert_sw_breakpoint(cpu
, bp
);
2654 QTAILQ_INSERT_HEAD(&cpu
->kvm_state
->kvm_sw_breakpoints
, bp
, entry
);
2656 err
= kvm_arch_insert_hw_breakpoint(addr
, len
, type
);
2663 err
= kvm_update_guest_debug(cpu
, 0);
2671 int kvm_remove_breakpoint(CPUState
*cpu
, target_ulong addr
,
2672 target_ulong len
, int type
)
2674 struct kvm_sw_breakpoint
*bp
;
2677 if (type
== GDB_BREAKPOINT_SW
) {
2678 bp
= kvm_find_sw_breakpoint(cpu
, addr
);
2683 if (bp
->use_count
> 1) {
2688 err
= kvm_arch_remove_sw_breakpoint(cpu
, bp
);
2693 QTAILQ_REMOVE(&cpu
->kvm_state
->kvm_sw_breakpoints
, bp
, entry
);
2696 err
= kvm_arch_remove_hw_breakpoint(addr
, len
, type
);
2703 err
= kvm_update_guest_debug(cpu
, 0);
2711 void kvm_remove_all_breakpoints(CPUState
*cpu
)
2713 struct kvm_sw_breakpoint
*bp
, *next
;
2714 KVMState
*s
= cpu
->kvm_state
;
2717 QTAILQ_FOREACH_SAFE(bp
, &s
->kvm_sw_breakpoints
, entry
, next
) {
2718 if (kvm_arch_remove_sw_breakpoint(cpu
, bp
) != 0) {
2719 /* Try harder to find a CPU that currently sees the breakpoint. */
2720 CPU_FOREACH(tmpcpu
) {
2721 if (kvm_arch_remove_sw_breakpoint(tmpcpu
, bp
) == 0) {
2726 QTAILQ_REMOVE(&s
->kvm_sw_breakpoints
, bp
, entry
);
2729 kvm_arch_remove_all_hw_breakpoints();
2732 kvm_update_guest_debug(cpu
, 0);
2736 #else /* !KVM_CAP_SET_GUEST_DEBUG */
2738 int kvm_update_guest_debug(CPUState
*cpu
, unsigned long reinject_trap
)
2743 int kvm_insert_breakpoint(CPUState
*cpu
, target_ulong addr
,
2744 target_ulong len
, int type
)
2749 int kvm_remove_breakpoint(CPUState
*cpu
, target_ulong addr
,
2750 target_ulong len
, int type
)
2755 void kvm_remove_all_breakpoints(CPUState
*cpu
)
2758 #endif /* !KVM_CAP_SET_GUEST_DEBUG */
2760 static int kvm_set_signal_mask(CPUState
*cpu
, const sigset_t
*sigset
)
2762 KVMState
*s
= kvm_state
;
2763 struct kvm_signal_mask
*sigmask
;
2766 sigmask
= g_malloc(sizeof(*sigmask
) + sizeof(*sigset
));
2768 sigmask
->len
= s
->sigmask_len
;
2769 memcpy(sigmask
->sigset
, sigset
, sizeof(*sigset
));
2770 r
= kvm_vcpu_ioctl(cpu
, KVM_SET_SIGNAL_MASK
, sigmask
);
2776 static void kvm_ipi_signal(int sig
)
2779 assert(kvm_immediate_exit
);
2780 kvm_cpu_kick(current_cpu
);
2784 void kvm_init_cpu_signals(CPUState
*cpu
)
2788 struct sigaction sigact
;
2790 memset(&sigact
, 0, sizeof(sigact
));
2791 sigact
.sa_handler
= kvm_ipi_signal
;
2792 sigaction(SIG_IPI
, &sigact
, NULL
);
2794 pthread_sigmask(SIG_BLOCK
, NULL
, &set
);
2795 #if defined KVM_HAVE_MCE_INJECTION
2796 sigdelset(&set
, SIGBUS
);
2797 pthread_sigmask(SIG_SETMASK
, &set
, NULL
);
2799 sigdelset(&set
, SIG_IPI
);
2800 if (kvm_immediate_exit
) {
2801 r
= pthread_sigmask(SIG_SETMASK
, &set
, NULL
);
2803 r
= kvm_set_signal_mask(cpu
, &set
);
2806 fprintf(stderr
, "kvm_set_signal_mask: %s\n", strerror(-r
));
2811 /* Called asynchronously in VCPU thread. */
2812 int kvm_on_sigbus_vcpu(CPUState
*cpu
, int code
, void *addr
)
2814 #ifdef KVM_HAVE_MCE_INJECTION
2815 if (have_sigbus_pending
) {
2818 have_sigbus_pending
= true;
2819 pending_sigbus_addr
= addr
;
2820 pending_sigbus_code
= code
;
2821 atomic_set(&cpu
->exit_request
, 1);
2828 /* Called synchronously (via signalfd) in main thread. */
2829 int kvm_on_sigbus(int code
, void *addr
)
2831 #ifdef KVM_HAVE_MCE_INJECTION
2832 /* Action required MCE kills the process if SIGBUS is blocked. Because
2833 * that's what happens in the I/O thread, where we handle MCE via signalfd,
2834 * we can only get action optional here.
2836 assert(code
!= BUS_MCEERR_AR
);
2837 kvm_arch_on_sigbus_vcpu(first_cpu
, code
, addr
);
2844 int kvm_create_device(KVMState
*s
, uint64_t type
, bool test
)
2847 struct kvm_create_device create_dev
;
2849 create_dev
.type
= type
;
2851 create_dev
.flags
= test
? KVM_CREATE_DEVICE_TEST
: 0;
2853 if (!kvm_check_extension(s
, KVM_CAP_DEVICE_CTRL
)) {
2857 ret
= kvm_vm_ioctl(s
, KVM_CREATE_DEVICE
, &create_dev
);
2862 return test
? 0 : create_dev
.fd
;
2865 bool kvm_device_supported(int vmfd
, uint64_t type
)
2867 struct kvm_create_device create_dev
= {
2870 .flags
= KVM_CREATE_DEVICE_TEST
,
2873 if (ioctl(vmfd
, KVM_CHECK_EXTENSION
, KVM_CAP_DEVICE_CTRL
) <= 0) {
2877 return (ioctl(vmfd
, KVM_CREATE_DEVICE
, &create_dev
) >= 0);
2880 int kvm_set_one_reg(CPUState
*cs
, uint64_t id
, void *source
)
2882 struct kvm_one_reg reg
;
2886 reg
.addr
= (uintptr_t) source
;
2887 r
= kvm_vcpu_ioctl(cs
, KVM_SET_ONE_REG
, ®
);
2889 trace_kvm_failed_reg_set(id
, strerror(-r
));
2894 int kvm_get_one_reg(CPUState
*cs
, uint64_t id
, void *target
)
2896 struct kvm_one_reg reg
;
2900 reg
.addr
= (uintptr_t) target
;
2901 r
= kvm_vcpu_ioctl(cs
, KVM_GET_ONE_REG
, ®
);
2903 trace_kvm_failed_reg_get(id
, strerror(-r
));
2908 static bool kvm_accel_has_memory(MachineState
*ms
, AddressSpace
*as
,
2909 hwaddr start_addr
, hwaddr size
)
2911 KVMState
*kvm
= KVM_STATE(ms
->accelerator
);
2914 for (i
= 0; i
< kvm
->nr_as
; ++i
) {
2915 if (kvm
->as
[i
].as
== as
&& kvm
->as
[i
].ml
) {
2916 size
= MIN(kvm_max_slot_size
, size
);
2917 return NULL
!= kvm_lookup_matching_slot(kvm
->as
[i
].ml
,
2925 static void kvm_accel_class_init(ObjectClass
*oc
, void *data
)
2927 AccelClass
*ac
= ACCEL_CLASS(oc
);
2929 ac
->init_machine
= kvm_init
;
2930 ac
->has_memory
= kvm_accel_has_memory
;
2931 ac
->allowed
= &kvm_allowed
;
2934 static const TypeInfo kvm_accel_type
= {
2935 .name
= TYPE_KVM_ACCEL
,
2936 .parent
= TYPE_ACCEL
,
2937 .class_init
= kvm_accel_class_init
,
2938 .instance_size
= sizeof(KVMState
),
2941 static void kvm_type_init(void)
2943 type_register_static(&kvm_accel_type
);
2946 type_init(kvm_type_init
);