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 <sys/types.h>
17 #include <sys/ioctl.h>
21 #include <linux/kvm.h>
23 #include "qemu-common.h"
24 #include "qemu-barrier.h"
31 /* This check must be after config-host.h is included */
33 #include <sys/eventfd.h>
36 /* KVM uses PAGE_SIZE in it's definition of COALESCED_MMIO_MAX */
37 #define PAGE_SIZE TARGET_PAGE_SIZE
42 #define DPRINTF(fmt, ...) \
43 do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
45 #define DPRINTF(fmt, ...) \
49 typedef struct KVMSlot
51 target_phys_addr_t start_addr
;
52 ram_addr_t memory_size
;
53 ram_addr_t phys_offset
;
58 typedef struct kvm_dirty_log KVMDirtyLog
;
66 struct kvm_coalesced_mmio_ring
*coalesced_mmio_ring
;
67 int broken_set_mem_region
;
70 int robust_singlestep
;
72 #ifdef KVM_CAP_SET_GUEST_DEBUG
73 struct kvm_sw_breakpoint_head kvm_sw_breakpoints
;
75 int irqchip_in_kernel
;
83 static const KVMCapabilityInfo kvm_required_capabilites
[] = {
84 KVM_CAP_INFO(USER_MEMORY
),
85 KVM_CAP_INFO(DESTROY_MEMORY_REGION_WORKS
),
89 static KVMSlot
*kvm_alloc_slot(KVMState
*s
)
93 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
94 if (s
->slots
[i
].memory_size
== 0) {
99 fprintf(stderr
, "%s: no free slot available\n", __func__
);
103 static KVMSlot
*kvm_lookup_matching_slot(KVMState
*s
,
104 target_phys_addr_t start_addr
,
105 target_phys_addr_t end_addr
)
109 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
110 KVMSlot
*mem
= &s
->slots
[i
];
112 if (start_addr
== mem
->start_addr
&&
113 end_addr
== mem
->start_addr
+ mem
->memory_size
) {
122 * Find overlapping slot with lowest start address
124 static KVMSlot
*kvm_lookup_overlapping_slot(KVMState
*s
,
125 target_phys_addr_t start_addr
,
126 target_phys_addr_t end_addr
)
128 KVMSlot
*found
= NULL
;
131 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
132 KVMSlot
*mem
= &s
->slots
[i
];
134 if (mem
->memory_size
== 0 ||
135 (found
&& found
->start_addr
< mem
->start_addr
)) {
139 if (end_addr
> mem
->start_addr
&&
140 start_addr
< mem
->start_addr
+ mem
->memory_size
) {
148 int kvm_physical_memory_addr_from_ram(KVMState
*s
, ram_addr_t ram_addr
,
149 target_phys_addr_t
*phys_addr
)
153 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
154 KVMSlot
*mem
= &s
->slots
[i
];
156 if (ram_addr
>= mem
->phys_offset
&&
157 ram_addr
< mem
->phys_offset
+ mem
->memory_size
) {
158 *phys_addr
= mem
->start_addr
+ (ram_addr
- mem
->phys_offset
);
166 static int kvm_set_user_memory_region(KVMState
*s
, KVMSlot
*slot
)
168 struct kvm_userspace_memory_region mem
;
170 mem
.slot
= slot
->slot
;
171 mem
.guest_phys_addr
= slot
->start_addr
;
172 mem
.memory_size
= slot
->memory_size
;
173 mem
.userspace_addr
= (unsigned long)qemu_safe_ram_ptr(slot
->phys_offset
);
174 mem
.flags
= slot
->flags
;
175 if (s
->migration_log
) {
176 mem
.flags
|= KVM_MEM_LOG_DIRTY_PAGES
;
178 return kvm_vm_ioctl(s
, KVM_SET_USER_MEMORY_REGION
, &mem
);
181 static void kvm_reset_vcpu(void *opaque
)
183 CPUState
*env
= opaque
;
185 kvm_arch_reset_vcpu(env
);
188 int kvm_irqchip_in_kernel(void)
190 return kvm_state
->irqchip_in_kernel
;
193 int kvm_pit_in_kernel(void)
195 return kvm_state
->pit_in_kernel
;
198 int kvm_init_vcpu(CPUState
*env
)
200 KVMState
*s
= kvm_state
;
204 DPRINTF("kvm_init_vcpu\n");
206 ret
= kvm_vm_ioctl(s
, KVM_CREATE_VCPU
, env
->cpu_index
);
208 DPRINTF("kvm_create_vcpu failed\n");
214 env
->kvm_vcpu_dirty
= 1;
216 mmap_size
= kvm_ioctl(s
, KVM_GET_VCPU_MMAP_SIZE
, 0);
219 DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n");
223 env
->kvm_run
= mmap(NULL
, mmap_size
, PROT_READ
| PROT_WRITE
, MAP_SHARED
,
225 if (env
->kvm_run
== MAP_FAILED
) {
227 DPRINTF("mmap'ing vcpu state failed\n");
231 if (s
->coalesced_mmio
&& !s
->coalesced_mmio_ring
) {
232 s
->coalesced_mmio_ring
=
233 (void *)env
->kvm_run
+ s
->coalesced_mmio
* PAGE_SIZE
;
236 ret
= kvm_arch_init_vcpu(env
);
238 qemu_register_reset(kvm_reset_vcpu
, env
);
239 kvm_arch_reset_vcpu(env
);
246 * dirty pages logging control
248 static int kvm_dirty_pages_log_change(target_phys_addr_t phys_addr
,
249 ram_addr_t size
, int flags
, int mask
)
251 KVMState
*s
= kvm_state
;
252 KVMSlot
*mem
= kvm_lookup_matching_slot(s
, phys_addr
, phys_addr
+ size
);
256 fprintf(stderr
, "BUG: %s: invalid parameters " TARGET_FMT_plx
"-"
257 TARGET_FMT_plx
"\n", __func__
, phys_addr
,
258 (target_phys_addr_t
)(phys_addr
+ size
- 1));
262 old_flags
= mem
->flags
;
264 flags
= (mem
->flags
& ~mask
) | flags
;
267 /* If nothing changed effectively, no need to issue ioctl */
268 if (s
->migration_log
) {
269 flags
|= KVM_MEM_LOG_DIRTY_PAGES
;
271 if (flags
== old_flags
) {
275 return kvm_set_user_memory_region(s
, mem
);
278 static int kvm_log_start(CPUPhysMemoryClient
*client
,
279 target_phys_addr_t phys_addr
, ram_addr_t size
)
281 return kvm_dirty_pages_log_change(phys_addr
, size
, KVM_MEM_LOG_DIRTY_PAGES
,
282 KVM_MEM_LOG_DIRTY_PAGES
);
285 static int kvm_log_stop(CPUPhysMemoryClient
*client
,
286 target_phys_addr_t phys_addr
, ram_addr_t size
)
288 return kvm_dirty_pages_log_change(phys_addr
, size
, 0,
289 KVM_MEM_LOG_DIRTY_PAGES
);
292 static int kvm_set_migration_log(int enable
)
294 KVMState
*s
= kvm_state
;
298 s
->migration_log
= enable
;
300 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
303 if (!mem
->memory_size
) {
306 if (!!(mem
->flags
& KVM_MEM_LOG_DIRTY_PAGES
) == enable
) {
309 err
= kvm_set_user_memory_region(s
, mem
);
317 /* get kvm's dirty pages bitmap and update qemu's */
318 static int kvm_get_dirty_pages_log_range(unsigned long start_addr
,
319 unsigned long *bitmap
,
320 unsigned long offset
,
321 unsigned long mem_size
)
324 unsigned long page_number
, addr
, addr1
, c
;
326 unsigned int len
= ((mem_size
/ TARGET_PAGE_SIZE
) + HOST_LONG_BITS
- 1) /
330 * bitmap-traveling is faster than memory-traveling (for addr...)
331 * especially when most of the memory is not dirty.
333 for (i
= 0; i
< len
; i
++) {
334 if (bitmap
[i
] != 0) {
335 c
= leul_to_cpu(bitmap
[i
]);
339 page_number
= i
* HOST_LONG_BITS
+ j
;
340 addr1
= page_number
* TARGET_PAGE_SIZE
;
341 addr
= offset
+ addr1
;
342 ram_addr
= cpu_get_physical_page_desc(addr
);
343 cpu_physical_memory_set_dirty(ram_addr
);
350 #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1))
353 * kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space
354 * This function updates qemu's dirty bitmap using cpu_physical_memory_set_dirty().
355 * This means all bits are set to dirty.
357 * @start_add: start of logged region.
358 * @end_addr: end of logged region.
360 static int kvm_physical_sync_dirty_bitmap(target_phys_addr_t start_addr
,
361 target_phys_addr_t end_addr
)
363 KVMState
*s
= kvm_state
;
364 unsigned long size
, allocated_size
= 0;
369 d
.dirty_bitmap
= NULL
;
370 while (start_addr
< end_addr
) {
371 mem
= kvm_lookup_overlapping_slot(s
, start_addr
, end_addr
);
376 size
= ALIGN(((mem
->memory_size
) >> TARGET_PAGE_BITS
), HOST_LONG_BITS
) / 8;
377 if (!d
.dirty_bitmap
) {
378 d
.dirty_bitmap
= qemu_malloc(size
);
379 } else if (size
> allocated_size
) {
380 d
.dirty_bitmap
= qemu_realloc(d
.dirty_bitmap
, size
);
382 allocated_size
= size
;
383 memset(d
.dirty_bitmap
, 0, allocated_size
);
387 if (kvm_vm_ioctl(s
, KVM_GET_DIRTY_LOG
, &d
) == -1) {
388 DPRINTF("ioctl failed %d\n", errno
);
393 kvm_get_dirty_pages_log_range(mem
->start_addr
, d
.dirty_bitmap
,
394 mem
->start_addr
, mem
->memory_size
);
395 start_addr
= mem
->start_addr
+ mem
->memory_size
;
397 qemu_free(d
.dirty_bitmap
);
402 int kvm_coalesce_mmio_region(target_phys_addr_t start
, ram_addr_t size
)
405 KVMState
*s
= kvm_state
;
407 if (s
->coalesced_mmio
) {
408 struct kvm_coalesced_mmio_zone zone
;
413 ret
= kvm_vm_ioctl(s
, KVM_REGISTER_COALESCED_MMIO
, &zone
);
419 int kvm_uncoalesce_mmio_region(target_phys_addr_t start
, ram_addr_t size
)
422 KVMState
*s
= kvm_state
;
424 if (s
->coalesced_mmio
) {
425 struct kvm_coalesced_mmio_zone zone
;
430 ret
= kvm_vm_ioctl(s
, KVM_UNREGISTER_COALESCED_MMIO
, &zone
);
436 int kvm_check_extension(KVMState
*s
, unsigned int extension
)
440 ret
= kvm_ioctl(s
, KVM_CHECK_EXTENSION
, extension
);
448 static int kvm_check_many_ioeventfds(void)
450 /* Userspace can use ioeventfd for io notification. This requires a host
451 * that supports eventfd(2) and an I/O thread; since eventfd does not
452 * support SIGIO it cannot interrupt the vcpu.
454 * Older kernels have a 6 device limit on the KVM io bus. Find out so we
455 * can avoid creating too many ioeventfds.
457 #if defined(CONFIG_EVENTFD) && defined(CONFIG_IOTHREAD)
460 for (i
= 0; i
< ARRAY_SIZE(ioeventfds
); i
++) {
461 ioeventfds
[i
] = eventfd(0, EFD_CLOEXEC
);
462 if (ioeventfds
[i
] < 0) {
465 ret
= kvm_set_ioeventfd_pio_word(ioeventfds
[i
], 0, i
, true);
467 close(ioeventfds
[i
]);
472 /* Decide whether many devices are supported or not */
473 ret
= i
== ARRAY_SIZE(ioeventfds
);
476 kvm_set_ioeventfd_pio_word(ioeventfds
[i
], 0, i
, false);
477 close(ioeventfds
[i
]);
485 static const KVMCapabilityInfo
*
486 kvm_check_extension_list(KVMState
*s
, const KVMCapabilityInfo
*list
)
489 if (!kvm_check_extension(s
, list
->value
)) {
497 static void kvm_set_phys_mem(target_phys_addr_t start_addr
, ram_addr_t size
,
498 ram_addr_t phys_offset
)
500 KVMState
*s
= kvm_state
;
501 ram_addr_t flags
= phys_offset
& ~TARGET_PAGE_MASK
;
505 /* kvm works in page size chunks, but the function may be called
506 with sub-page size and unaligned start address. */
507 size
= TARGET_PAGE_ALIGN(size
);
508 start_addr
= TARGET_PAGE_ALIGN(start_addr
);
510 /* KVM does not support read-only slots */
511 phys_offset
&= ~IO_MEM_ROM
;
514 mem
= kvm_lookup_overlapping_slot(s
, start_addr
, start_addr
+ size
);
519 if (flags
< IO_MEM_UNASSIGNED
&& start_addr
>= mem
->start_addr
&&
520 (start_addr
+ size
<= mem
->start_addr
+ mem
->memory_size
) &&
521 (phys_offset
- start_addr
== mem
->phys_offset
- mem
->start_addr
)) {
522 /* The new slot fits into the existing one and comes with
523 * identical parameters - nothing to be done. */
529 /* unregister the overlapping slot */
530 mem
->memory_size
= 0;
531 err
= kvm_set_user_memory_region(s
, mem
);
533 fprintf(stderr
, "%s: error unregistering overlapping slot: %s\n",
534 __func__
, strerror(-err
));
538 /* Workaround for older KVM versions: we can't join slots, even not by
539 * unregistering the previous ones and then registering the larger
540 * slot. We have to maintain the existing fragmentation. Sigh.
542 * This workaround assumes that the new slot starts at the same
543 * address as the first existing one. If not or if some overlapping
544 * slot comes around later, we will fail (not seen in practice so far)
545 * - and actually require a recent KVM version. */
546 if (s
->broken_set_mem_region
&&
547 old
.start_addr
== start_addr
&& old
.memory_size
< size
&&
548 flags
< IO_MEM_UNASSIGNED
) {
549 mem
= kvm_alloc_slot(s
);
550 mem
->memory_size
= old
.memory_size
;
551 mem
->start_addr
= old
.start_addr
;
552 mem
->phys_offset
= old
.phys_offset
;
555 err
= kvm_set_user_memory_region(s
, mem
);
557 fprintf(stderr
, "%s: error updating slot: %s\n", __func__
,
562 start_addr
+= old
.memory_size
;
563 phys_offset
+= old
.memory_size
;
564 size
-= old
.memory_size
;
568 /* register prefix slot */
569 if (old
.start_addr
< start_addr
) {
570 mem
= kvm_alloc_slot(s
);
571 mem
->memory_size
= start_addr
- old
.start_addr
;
572 mem
->start_addr
= old
.start_addr
;
573 mem
->phys_offset
= old
.phys_offset
;
576 err
= kvm_set_user_memory_region(s
, mem
);
578 fprintf(stderr
, "%s: error registering prefix slot: %s\n",
579 __func__
, strerror(-err
));
584 /* register suffix slot */
585 if (old
.start_addr
+ old
.memory_size
> start_addr
+ size
) {
586 ram_addr_t size_delta
;
588 mem
= kvm_alloc_slot(s
);
589 mem
->start_addr
= start_addr
+ size
;
590 size_delta
= mem
->start_addr
- old
.start_addr
;
591 mem
->memory_size
= old
.memory_size
- size_delta
;
592 mem
->phys_offset
= old
.phys_offset
+ size_delta
;
595 err
= kvm_set_user_memory_region(s
, mem
);
597 fprintf(stderr
, "%s: error registering suffix slot: %s\n",
598 __func__
, strerror(-err
));
604 /* in case the KVM bug workaround already "consumed" the new slot */
608 /* KVM does not need to know about this memory */
609 if (flags
>= IO_MEM_UNASSIGNED
) {
612 mem
= kvm_alloc_slot(s
);
613 mem
->memory_size
= size
;
614 mem
->start_addr
= start_addr
;
615 mem
->phys_offset
= phys_offset
;
618 err
= kvm_set_user_memory_region(s
, mem
);
620 fprintf(stderr
, "%s: error registering slot: %s\n", __func__
,
626 static void kvm_client_set_memory(struct CPUPhysMemoryClient
*client
,
627 target_phys_addr_t start_addr
,
628 ram_addr_t size
, ram_addr_t phys_offset
)
630 kvm_set_phys_mem(start_addr
, size
, phys_offset
);
633 static int kvm_client_sync_dirty_bitmap(struct CPUPhysMemoryClient
*client
,
634 target_phys_addr_t start_addr
,
635 target_phys_addr_t end_addr
)
637 return kvm_physical_sync_dirty_bitmap(start_addr
, end_addr
);
640 static int kvm_client_migration_log(struct CPUPhysMemoryClient
*client
,
643 return kvm_set_migration_log(enable
);
646 static CPUPhysMemoryClient kvm_cpu_phys_memory_client
= {
647 .set_memory
= kvm_client_set_memory
,
648 .sync_dirty_bitmap
= kvm_client_sync_dirty_bitmap
,
649 .migration_log
= kvm_client_migration_log
,
650 .log_start
= kvm_log_start
,
651 .log_stop
= kvm_log_stop
,
656 static const char upgrade_note
[] =
657 "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n"
658 "(see http://sourceforge.net/projects/kvm).\n";
660 const KVMCapabilityInfo
*missing_cap
;
664 s
= qemu_mallocz(sizeof(KVMState
));
666 #ifdef KVM_CAP_SET_GUEST_DEBUG
667 QTAILQ_INIT(&s
->kvm_sw_breakpoints
);
669 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
670 s
->slots
[i
].slot
= i
;
673 s
->fd
= qemu_open("/dev/kvm", O_RDWR
);
675 fprintf(stderr
, "Could not access KVM kernel module: %m\n");
680 ret
= kvm_ioctl(s
, KVM_GET_API_VERSION
, 0);
681 if (ret
< KVM_API_VERSION
) {
685 fprintf(stderr
, "kvm version too old\n");
689 if (ret
> KVM_API_VERSION
) {
691 fprintf(stderr
, "kvm version not supported\n");
695 s
->vmfd
= kvm_ioctl(s
, KVM_CREATE_VM
, 0);
698 fprintf(stderr
, "Please add the 'switch_amode' kernel parameter to "
699 "your host kernel command line\n");
704 missing_cap
= kvm_check_extension_list(s
, kvm_required_capabilites
);
707 kvm_check_extension_list(s
, kvm_arch_required_capabilities
);
711 fprintf(stderr
, "kvm does not support %s\n%s",
712 missing_cap
->name
, upgrade_note
);
716 s
->coalesced_mmio
= kvm_check_extension(s
, KVM_CAP_COALESCED_MMIO
);
718 s
->broken_set_mem_region
= 1;
719 #ifdef KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
720 ret
= kvm_check_extension(s
, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
);
722 s
->broken_set_mem_region
= 0;
727 #ifdef KVM_CAP_VCPU_EVENTS
728 s
->vcpu_events
= kvm_check_extension(s
, KVM_CAP_VCPU_EVENTS
);
731 s
->robust_singlestep
= 0;
732 #ifdef KVM_CAP_X86_ROBUST_SINGLESTEP
733 s
->robust_singlestep
=
734 kvm_check_extension(s
, KVM_CAP_X86_ROBUST_SINGLESTEP
);
738 #ifdef KVM_CAP_DEBUGREGS
739 s
->debugregs
= kvm_check_extension(s
, KVM_CAP_DEBUGREGS
);
744 s
->xsave
= kvm_check_extension(s
, KVM_CAP_XSAVE
);
749 s
->xcrs
= kvm_check_extension(s
, KVM_CAP_XCRS
);
752 ret
= kvm_arch_init(s
);
758 cpu_register_phys_memory_client(&kvm_cpu_phys_memory_client
);
760 s
->many_ioeventfds
= kvm_check_many_ioeventfds();
778 static void kvm_handle_io(uint16_t port
, void *data
, int direction
, int size
,
784 for (i
= 0; i
< count
; i
++) {
785 if (direction
== KVM_EXIT_IO_IN
) {
788 stb_p(ptr
, cpu_inb(port
));
791 stw_p(ptr
, cpu_inw(port
));
794 stl_p(ptr
, cpu_inl(port
));
800 cpu_outb(port
, ldub_p(ptr
));
803 cpu_outw(port
, lduw_p(ptr
));
806 cpu_outl(port
, ldl_p(ptr
));
815 #ifdef KVM_CAP_INTERNAL_ERROR_DATA
816 static int kvm_handle_internal_error(CPUState
*env
, struct kvm_run
*run
)
818 fprintf(stderr
, "KVM internal error.");
819 if (kvm_check_extension(kvm_state
, KVM_CAP_INTERNAL_ERROR_DATA
)) {
822 fprintf(stderr
, " Suberror: %d\n", run
->internal
.suberror
);
823 for (i
= 0; i
< run
->internal
.ndata
; ++i
) {
824 fprintf(stderr
, "extra data[%d]: %"PRIx64
"\n",
825 i
, (uint64_t)run
->internal
.data
[i
]);
828 fprintf(stderr
, "\n");
830 if (run
->internal
.suberror
== KVM_INTERNAL_ERROR_EMULATION
) {
831 fprintf(stderr
, "emulation failure\n");
832 if (!kvm_arch_stop_on_emulation_error(env
)) {
833 cpu_dump_state(env
, stderr
, fprintf
, CPU_DUMP_CODE
);
834 return EXCP_INTERRUPT
;
837 /* FIXME: Should trigger a qmp message to let management know
838 * something went wrong.
844 void kvm_flush_coalesced_mmio_buffer(void)
846 KVMState
*s
= kvm_state
;
847 if (s
->coalesced_mmio_ring
) {
848 struct kvm_coalesced_mmio_ring
*ring
= s
->coalesced_mmio_ring
;
849 while (ring
->first
!= ring
->last
) {
850 struct kvm_coalesced_mmio
*ent
;
852 ent
= &ring
->coalesced_mmio
[ring
->first
];
854 cpu_physical_memory_write(ent
->phys_addr
, ent
->data
, ent
->len
);
856 ring
->first
= (ring
->first
+ 1) % KVM_COALESCED_MMIO_MAX
;
861 static void do_kvm_cpu_synchronize_state(void *_env
)
863 CPUState
*env
= _env
;
865 if (!env
->kvm_vcpu_dirty
) {
866 kvm_arch_get_registers(env
);
867 env
->kvm_vcpu_dirty
= 1;
871 void kvm_cpu_synchronize_state(CPUState
*env
)
873 if (!env
->kvm_vcpu_dirty
) {
874 run_on_cpu(env
, do_kvm_cpu_synchronize_state
, env
);
878 void kvm_cpu_synchronize_post_reset(CPUState
*env
)
880 kvm_arch_put_registers(env
, KVM_PUT_RESET_STATE
);
881 env
->kvm_vcpu_dirty
= 0;
884 void kvm_cpu_synchronize_post_init(CPUState
*env
)
886 kvm_arch_put_registers(env
, KVM_PUT_FULL_STATE
);
887 env
->kvm_vcpu_dirty
= 0;
890 int kvm_cpu_exec(CPUState
*env
)
892 struct kvm_run
*run
= env
->kvm_run
;
895 DPRINTF("kvm_cpu_exec()\n");
897 if (kvm_arch_process_async_events(env
)) {
898 env
->exit_request
= 0;
902 cpu_single_env
= env
;
905 if (env
->kvm_vcpu_dirty
) {
906 kvm_arch_put_registers(env
, KVM_PUT_RUNTIME_STATE
);
907 env
->kvm_vcpu_dirty
= 0;
910 kvm_arch_pre_run(env
, run
);
911 if (env
->exit_request
) {
912 DPRINTF("interrupt exit requested\n");
914 * KVM requires us to reenter the kernel after IO exits to complete
915 * instruction emulation. This self-signal will ensure that we
918 qemu_cpu_kick_self();
920 cpu_single_env
= NULL
;
921 qemu_mutex_unlock_iothread();
923 run_ret
= kvm_vcpu_ioctl(env
, KVM_RUN
, 0);
925 qemu_mutex_lock_iothread();
926 cpu_single_env
= env
;
927 kvm_arch_post_run(env
, run
);
929 kvm_flush_coalesced_mmio_buffer();
932 if (run_ret
== -EINTR
|| run_ret
== -EAGAIN
) {
933 DPRINTF("io window exit\n");
934 ret
= EXCP_INTERRUPT
;
937 DPRINTF("kvm run failed %s\n", strerror(-run_ret
));
941 switch (run
->exit_reason
) {
943 DPRINTF("handle_io\n");
944 kvm_handle_io(run
->io
.port
,
945 (uint8_t *)run
+ run
->io
.data_offset
,
952 DPRINTF("handle_mmio\n");
953 cpu_physical_memory_rw(run
->mmio
.phys_addr
,
959 case KVM_EXIT_IRQ_WINDOW_OPEN
:
960 DPRINTF("irq_window_open\n");
961 ret
= EXCP_INTERRUPT
;
963 case KVM_EXIT_SHUTDOWN
:
964 DPRINTF("shutdown\n");
965 qemu_system_reset_request();
966 ret
= EXCP_INTERRUPT
;
968 case KVM_EXIT_UNKNOWN
:
969 fprintf(stderr
, "KVM: unknown exit, hardware reason %" PRIx64
"\n",
970 (uint64_t)run
->hw
.hardware_exit_reason
);
973 #ifdef KVM_CAP_INTERNAL_ERROR_DATA
974 case KVM_EXIT_INTERNAL_ERROR
:
975 ret
= kvm_handle_internal_error(env
, run
);
978 #ifdef KVM_CAP_SET_GUEST_DEBUG
980 DPRINTF("kvm_exit_debug\n");
981 if (kvm_arch_debug(&run
->debug
.arch
)) {
985 /* re-enter, this exception was guest-internal */
988 #endif /* KVM_CAP_SET_GUEST_DEBUG */
990 DPRINTF("kvm_arch_handle_exit\n");
991 ret
= kvm_arch_handle_exit(env
, run
);
993 ret
= EXCP_INTERRUPT
;
994 } else if (ret
> 0) {
1002 cpu_dump_state(env
, stderr
, fprintf
, CPU_DUMP_CODE
);
1003 vm_stop(VMSTOP_PANIC
);
1006 env
->exit_request
= 0;
1007 cpu_single_env
= NULL
;
1011 int kvm_ioctl(KVMState
*s
, int type
, ...)
1018 arg
= va_arg(ap
, void *);
1021 ret
= ioctl(s
->fd
, type
, arg
);
1028 int kvm_vm_ioctl(KVMState
*s
, int type
, ...)
1035 arg
= va_arg(ap
, void *);
1038 ret
= ioctl(s
->vmfd
, type
, arg
);
1045 int kvm_vcpu_ioctl(CPUState
*env
, int type
, ...)
1052 arg
= va_arg(ap
, void *);
1055 ret
= ioctl(env
->kvm_fd
, type
, arg
);
1062 int kvm_has_sync_mmu(void)
1064 return kvm_check_extension(kvm_state
, KVM_CAP_SYNC_MMU
);
1067 int kvm_has_vcpu_events(void)
1069 return kvm_state
->vcpu_events
;
1072 int kvm_has_robust_singlestep(void)
1074 return kvm_state
->robust_singlestep
;
1077 int kvm_has_debugregs(void)
1079 return kvm_state
->debugregs
;
1082 int kvm_has_xsave(void)
1084 return kvm_state
->xsave
;
1087 int kvm_has_xcrs(void)
1089 return kvm_state
->xcrs
;
1092 int kvm_has_many_ioeventfds(void)
1094 if (!kvm_enabled()) {
1097 return kvm_state
->many_ioeventfds
;
1100 void kvm_setup_guest_memory(void *start
, size_t size
)
1102 if (!kvm_has_sync_mmu()) {
1103 int ret
= qemu_madvise(start
, size
, QEMU_MADV_DONTFORK
);
1106 perror("qemu_madvise");
1108 "Need MADV_DONTFORK in absence of synchronous KVM MMU\n");
1114 #ifdef KVM_CAP_SET_GUEST_DEBUG
1115 struct kvm_sw_breakpoint
*kvm_find_sw_breakpoint(CPUState
*env
,
1118 struct kvm_sw_breakpoint
*bp
;
1120 QTAILQ_FOREACH(bp
, &env
->kvm_state
->kvm_sw_breakpoints
, entry
) {
1128 int kvm_sw_breakpoints_active(CPUState
*env
)
1130 return !QTAILQ_EMPTY(&env
->kvm_state
->kvm_sw_breakpoints
);
1133 struct kvm_set_guest_debug_data
{
1134 struct kvm_guest_debug dbg
;
1139 static void kvm_invoke_set_guest_debug(void *data
)
1141 struct kvm_set_guest_debug_data
*dbg_data
= data
;
1142 CPUState
*env
= dbg_data
->env
;
1144 dbg_data
->err
= kvm_vcpu_ioctl(env
, KVM_SET_GUEST_DEBUG
, &dbg_data
->dbg
);
1147 int kvm_update_guest_debug(CPUState
*env
, unsigned long reinject_trap
)
1149 struct kvm_set_guest_debug_data data
;
1151 data
.dbg
.control
= reinject_trap
;
1153 if (env
->singlestep_enabled
) {
1154 data
.dbg
.control
|= KVM_GUESTDBG_ENABLE
| KVM_GUESTDBG_SINGLESTEP
;
1156 kvm_arch_update_guest_debug(env
, &data
.dbg
);
1159 run_on_cpu(env
, kvm_invoke_set_guest_debug
, &data
);
1163 int kvm_insert_breakpoint(CPUState
*current_env
, target_ulong addr
,
1164 target_ulong len
, int type
)
1166 struct kvm_sw_breakpoint
*bp
;
1170 if (type
== GDB_BREAKPOINT_SW
) {
1171 bp
= kvm_find_sw_breakpoint(current_env
, addr
);
1177 bp
= qemu_malloc(sizeof(struct kvm_sw_breakpoint
));
1184 err
= kvm_arch_insert_sw_breakpoint(current_env
, bp
);
1190 QTAILQ_INSERT_HEAD(¤t_env
->kvm_state
->kvm_sw_breakpoints
,
1193 err
= kvm_arch_insert_hw_breakpoint(addr
, len
, type
);
1199 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
1200 err
= kvm_update_guest_debug(env
, 0);
1208 int kvm_remove_breakpoint(CPUState
*current_env
, target_ulong addr
,
1209 target_ulong len
, int type
)
1211 struct kvm_sw_breakpoint
*bp
;
1215 if (type
== GDB_BREAKPOINT_SW
) {
1216 bp
= kvm_find_sw_breakpoint(current_env
, addr
);
1221 if (bp
->use_count
> 1) {
1226 err
= kvm_arch_remove_sw_breakpoint(current_env
, bp
);
1231 QTAILQ_REMOVE(¤t_env
->kvm_state
->kvm_sw_breakpoints
, bp
, entry
);
1234 err
= kvm_arch_remove_hw_breakpoint(addr
, len
, type
);
1240 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
1241 err
= kvm_update_guest_debug(env
, 0);
1249 void kvm_remove_all_breakpoints(CPUState
*current_env
)
1251 struct kvm_sw_breakpoint
*bp
, *next
;
1252 KVMState
*s
= current_env
->kvm_state
;
1255 QTAILQ_FOREACH_SAFE(bp
, &s
->kvm_sw_breakpoints
, entry
, next
) {
1256 if (kvm_arch_remove_sw_breakpoint(current_env
, bp
) != 0) {
1257 /* Try harder to find a CPU that currently sees the breakpoint. */
1258 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
1259 if (kvm_arch_remove_sw_breakpoint(env
, bp
) == 0) {
1265 kvm_arch_remove_all_hw_breakpoints();
1267 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
1268 kvm_update_guest_debug(env
, 0);
1272 #else /* !KVM_CAP_SET_GUEST_DEBUG */
1274 int kvm_update_guest_debug(CPUState
*env
, unsigned long reinject_trap
)
1279 int kvm_insert_breakpoint(CPUState
*current_env
, target_ulong addr
,
1280 target_ulong len
, int type
)
1285 int kvm_remove_breakpoint(CPUState
*current_env
, target_ulong addr
,
1286 target_ulong len
, int type
)
1291 void kvm_remove_all_breakpoints(CPUState
*current_env
)
1294 #endif /* !KVM_CAP_SET_GUEST_DEBUG */
1296 int kvm_set_signal_mask(CPUState
*env
, const sigset_t
*sigset
)
1298 struct kvm_signal_mask
*sigmask
;
1302 return kvm_vcpu_ioctl(env
, KVM_SET_SIGNAL_MASK
, NULL
);
1305 sigmask
= qemu_malloc(sizeof(*sigmask
) + sizeof(*sigset
));
1308 memcpy(sigmask
->sigset
, sigset
, sizeof(*sigset
));
1309 r
= kvm_vcpu_ioctl(env
, KVM_SET_SIGNAL_MASK
, sigmask
);
1315 int kvm_set_ioeventfd_mmio_long(int fd
, uint32_t addr
, uint32_t val
, bool assign
)
1317 #ifdef KVM_IOEVENTFD
1319 struct kvm_ioeventfd iofd
;
1321 iofd
.datamatch
= val
;
1324 iofd
.flags
= KVM_IOEVENTFD_FLAG_DATAMATCH
;
1327 if (!kvm_enabled()) {
1332 iofd
.flags
|= KVM_IOEVENTFD_FLAG_DEASSIGN
;
1335 ret
= kvm_vm_ioctl(kvm_state
, KVM_IOEVENTFD
, &iofd
);
1347 int kvm_set_ioeventfd_pio_word(int fd
, uint16_t addr
, uint16_t val
, bool assign
)
1349 #ifdef KVM_IOEVENTFD
1350 struct kvm_ioeventfd kick
= {
1354 .flags
= KVM_IOEVENTFD_FLAG_DATAMATCH
| KVM_IOEVENTFD_FLAG_PIO
,
1358 if (!kvm_enabled()) {
1362 kick
.flags
|= KVM_IOEVENTFD_FLAG_DEASSIGN
;
1364 r
= kvm_vm_ioctl(kvm_state
, KVM_IOEVENTFD
, &kick
);
1374 int kvm_on_sigbus_vcpu(CPUState
*env
, int code
, void *addr
)
1376 return kvm_arch_on_sigbus_vcpu(env
, code
, addr
);
1379 int kvm_on_sigbus(int code
, void *addr
)
1381 return kvm_arch_on_sigbus(code
, addr
);