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 /* KVM uses PAGE_SIZE in it's definition of COALESCED_MMIO_MAX */
32 #define PAGE_SIZE TARGET_PAGE_SIZE
37 #define DPRINTF(fmt, ...) \
38 do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
40 #define DPRINTF(fmt, ...) \
44 typedef struct KVMSlot
46 target_phys_addr_t start_addr
;
47 ram_addr_t memory_size
;
48 ram_addr_t phys_offset
;
53 typedef struct kvm_dirty_log KVMDirtyLog
;
61 #ifdef KVM_CAP_COALESCED_MMIO
62 struct kvm_coalesced_mmio_ring
*coalesced_mmio_ring
;
64 int broken_set_mem_region
;
67 int robust_singlestep
;
69 #ifdef KVM_CAP_SET_GUEST_DEBUG
70 struct kvm_sw_breakpoint_head kvm_sw_breakpoints
;
72 int irqchip_in_kernel
;
77 static KVMState
*kvm_state
;
79 static KVMSlot
*kvm_alloc_slot(KVMState
*s
)
83 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
84 /* KVM private memory slots */
87 if (s
->slots
[i
].memory_size
== 0)
91 fprintf(stderr
, "%s: no free slot available\n", __func__
);
95 static KVMSlot
*kvm_lookup_matching_slot(KVMState
*s
,
96 target_phys_addr_t start_addr
,
97 target_phys_addr_t end_addr
)
101 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
102 KVMSlot
*mem
= &s
->slots
[i
];
104 if (start_addr
== mem
->start_addr
&&
105 end_addr
== mem
->start_addr
+ mem
->memory_size
) {
114 * Find overlapping slot with lowest start address
116 static KVMSlot
*kvm_lookup_overlapping_slot(KVMState
*s
,
117 target_phys_addr_t start_addr
,
118 target_phys_addr_t end_addr
)
120 KVMSlot
*found
= NULL
;
123 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
124 KVMSlot
*mem
= &s
->slots
[i
];
126 if (mem
->memory_size
== 0 ||
127 (found
&& found
->start_addr
< mem
->start_addr
)) {
131 if (end_addr
> mem
->start_addr
&&
132 start_addr
< mem
->start_addr
+ mem
->memory_size
) {
140 static int kvm_set_user_memory_region(KVMState
*s
, KVMSlot
*slot
)
142 struct kvm_userspace_memory_region mem
;
144 mem
.slot
= slot
->slot
;
145 mem
.guest_phys_addr
= slot
->start_addr
;
146 mem
.memory_size
= slot
->memory_size
;
147 mem
.userspace_addr
= (unsigned long)qemu_get_ram_ptr(slot
->phys_offset
);
148 mem
.flags
= slot
->flags
;
149 if (s
->migration_log
) {
150 mem
.flags
|= KVM_MEM_LOG_DIRTY_PAGES
;
152 return kvm_vm_ioctl(s
, KVM_SET_USER_MEMORY_REGION
, &mem
);
155 static void kvm_reset_vcpu(void *opaque
)
157 CPUState
*env
= opaque
;
159 kvm_arch_reset_vcpu(env
);
162 int kvm_irqchip_in_kernel(void)
164 return kvm_state
->irqchip_in_kernel
;
167 int kvm_pit_in_kernel(void)
169 return kvm_state
->pit_in_kernel
;
173 int kvm_init_vcpu(CPUState
*env
)
175 KVMState
*s
= kvm_state
;
179 DPRINTF("kvm_init_vcpu\n");
181 ret
= kvm_vm_ioctl(s
, KVM_CREATE_VCPU
, env
->cpu_index
);
183 DPRINTF("kvm_create_vcpu failed\n");
190 mmap_size
= kvm_ioctl(s
, KVM_GET_VCPU_MMAP_SIZE
, 0);
192 DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n");
196 env
->kvm_run
= mmap(NULL
, mmap_size
, PROT_READ
| PROT_WRITE
, MAP_SHARED
,
198 if (env
->kvm_run
== MAP_FAILED
) {
200 DPRINTF("mmap'ing vcpu state failed\n");
204 #ifdef KVM_CAP_COALESCED_MMIO
205 if (s
->coalesced_mmio
&& !s
->coalesced_mmio_ring
)
206 s
->coalesced_mmio_ring
= (void *) env
->kvm_run
+
207 s
->coalesced_mmio
* PAGE_SIZE
;
210 ret
= kvm_arch_init_vcpu(env
);
212 qemu_register_reset(kvm_reset_vcpu
, env
);
213 kvm_arch_reset_vcpu(env
);
220 * dirty pages logging control
222 static int kvm_dirty_pages_log_change(target_phys_addr_t phys_addr
,
223 ram_addr_t size
, int flags
, int mask
)
225 KVMState
*s
= kvm_state
;
226 KVMSlot
*mem
= kvm_lookup_matching_slot(s
, phys_addr
, phys_addr
+ size
);
230 fprintf(stderr
, "BUG: %s: invalid parameters " TARGET_FMT_plx
"-"
231 TARGET_FMT_plx
"\n", __func__
, phys_addr
,
232 (target_phys_addr_t
)(phys_addr
+ size
- 1));
236 old_flags
= mem
->flags
;
238 flags
= (mem
->flags
& ~mask
) | flags
;
241 /* If nothing changed effectively, no need to issue ioctl */
242 if (s
->migration_log
) {
243 flags
|= KVM_MEM_LOG_DIRTY_PAGES
;
245 if (flags
== old_flags
) {
249 return kvm_set_user_memory_region(s
, mem
);
252 int kvm_log_start(target_phys_addr_t phys_addr
, ram_addr_t size
)
254 return kvm_dirty_pages_log_change(phys_addr
, size
,
255 KVM_MEM_LOG_DIRTY_PAGES
,
256 KVM_MEM_LOG_DIRTY_PAGES
);
259 int kvm_log_stop(target_phys_addr_t phys_addr
, ram_addr_t size
)
261 return kvm_dirty_pages_log_change(phys_addr
, size
,
263 KVM_MEM_LOG_DIRTY_PAGES
);
266 static int kvm_set_migration_log(int enable
)
268 KVMState
*s
= kvm_state
;
272 s
->migration_log
= enable
;
274 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
277 if (!mem
->memory_size
) {
280 if (!!(mem
->flags
& KVM_MEM_LOG_DIRTY_PAGES
) == enable
) {
283 err
= kvm_set_user_memory_region(s
, mem
);
291 /* get kvm's dirty pages bitmap and update qemu's */
292 static int kvm_get_dirty_pages_log_range(unsigned long start_addr
,
293 unsigned long *bitmap
,
294 unsigned long offset
,
295 unsigned long mem_size
)
298 unsigned long page_number
, addr
, addr1
, c
;
300 unsigned int len
= ((mem_size
/ TARGET_PAGE_SIZE
) + HOST_LONG_BITS
- 1) /
304 * bitmap-traveling is faster than memory-traveling (for addr...)
305 * especially when most of the memory is not dirty.
307 for (i
= 0; i
< len
; i
++) {
308 if (bitmap
[i
] != 0) {
309 c
= leul_to_cpu(bitmap
[i
]);
313 page_number
= i
* HOST_LONG_BITS
+ j
;
314 addr1
= page_number
* TARGET_PAGE_SIZE
;
315 addr
= offset
+ addr1
;
316 ram_addr
= cpu_get_physical_page_desc(addr
);
317 cpu_physical_memory_set_dirty(ram_addr
);
324 #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1))
327 * kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space
328 * This function updates qemu's dirty bitmap using cpu_physical_memory_set_dirty().
329 * This means all bits are set to dirty.
331 * @start_add: start of logged region.
332 * @end_addr: end of logged region.
334 static int kvm_physical_sync_dirty_bitmap(target_phys_addr_t start_addr
,
335 target_phys_addr_t end_addr
)
337 KVMState
*s
= kvm_state
;
338 unsigned long size
, allocated_size
= 0;
343 d
.dirty_bitmap
= NULL
;
344 while (start_addr
< end_addr
) {
345 mem
= kvm_lookup_overlapping_slot(s
, start_addr
, end_addr
);
350 size
= ALIGN(((mem
->memory_size
) >> TARGET_PAGE_BITS
), HOST_LONG_BITS
) / 8;
351 if (!d
.dirty_bitmap
) {
352 d
.dirty_bitmap
= qemu_malloc(size
);
353 } else if (size
> allocated_size
) {
354 d
.dirty_bitmap
= qemu_realloc(d
.dirty_bitmap
, size
);
356 allocated_size
= size
;
357 memset(d
.dirty_bitmap
, 0, allocated_size
);
361 if (kvm_vm_ioctl(s
, KVM_GET_DIRTY_LOG
, &d
) == -1) {
362 DPRINTF("ioctl failed %d\n", errno
);
367 kvm_get_dirty_pages_log_range(mem
->start_addr
, d
.dirty_bitmap
,
368 mem
->start_addr
, mem
->memory_size
);
369 start_addr
= mem
->start_addr
+ mem
->memory_size
;
371 qemu_free(d
.dirty_bitmap
);
376 int kvm_coalesce_mmio_region(target_phys_addr_t start
, ram_addr_t size
)
379 #ifdef KVM_CAP_COALESCED_MMIO
380 KVMState
*s
= kvm_state
;
382 if (s
->coalesced_mmio
) {
383 struct kvm_coalesced_mmio_zone zone
;
388 ret
= kvm_vm_ioctl(s
, KVM_REGISTER_COALESCED_MMIO
, &zone
);
395 int kvm_uncoalesce_mmio_region(target_phys_addr_t start
, ram_addr_t size
)
398 #ifdef KVM_CAP_COALESCED_MMIO
399 KVMState
*s
= kvm_state
;
401 if (s
->coalesced_mmio
) {
402 struct kvm_coalesced_mmio_zone zone
;
407 ret
= kvm_vm_ioctl(s
, KVM_UNREGISTER_COALESCED_MMIO
, &zone
);
414 int kvm_check_extension(KVMState
*s
, unsigned int extension
)
418 ret
= kvm_ioctl(s
, KVM_CHECK_EXTENSION
, extension
);
426 static void kvm_set_phys_mem(target_phys_addr_t start_addr
,
428 ram_addr_t phys_offset
)
430 KVMState
*s
= kvm_state
;
431 ram_addr_t flags
= phys_offset
& ~TARGET_PAGE_MASK
;
435 if (start_addr
& ~TARGET_PAGE_MASK
) {
436 if (flags
>= IO_MEM_UNASSIGNED
) {
437 if (!kvm_lookup_overlapping_slot(s
, start_addr
,
438 start_addr
+ size
)) {
441 fprintf(stderr
, "Unaligned split of a KVM memory slot\n");
443 fprintf(stderr
, "Only page-aligned memory slots supported\n");
448 /* KVM does not support read-only slots */
449 phys_offset
&= ~IO_MEM_ROM
;
452 mem
= kvm_lookup_overlapping_slot(s
, start_addr
, start_addr
+ size
);
457 if (flags
< IO_MEM_UNASSIGNED
&& start_addr
>= mem
->start_addr
&&
458 (start_addr
+ size
<= mem
->start_addr
+ mem
->memory_size
) &&
459 (phys_offset
- start_addr
== mem
->phys_offset
- mem
->start_addr
)) {
460 /* The new slot fits into the existing one and comes with
461 * identical parameters - nothing to be done. */
467 /* unregister the overlapping slot */
468 mem
->memory_size
= 0;
469 err
= kvm_set_user_memory_region(s
, mem
);
471 fprintf(stderr
, "%s: error unregistering overlapping slot: %s\n",
472 __func__
, strerror(-err
));
476 /* Workaround for older KVM versions: we can't join slots, even not by
477 * unregistering the previous ones and then registering the larger
478 * slot. We have to maintain the existing fragmentation. Sigh.
480 * This workaround assumes that the new slot starts at the same
481 * address as the first existing one. If not or if some overlapping
482 * slot comes around later, we will fail (not seen in practice so far)
483 * - and actually require a recent KVM version. */
484 if (s
->broken_set_mem_region
&&
485 old
.start_addr
== start_addr
&& old
.memory_size
< size
&&
486 flags
< IO_MEM_UNASSIGNED
) {
487 mem
= kvm_alloc_slot(s
);
488 mem
->memory_size
= old
.memory_size
;
489 mem
->start_addr
= old
.start_addr
;
490 mem
->phys_offset
= old
.phys_offset
;
493 err
= kvm_set_user_memory_region(s
, mem
);
495 fprintf(stderr
, "%s: error updating slot: %s\n", __func__
,
500 start_addr
+= old
.memory_size
;
501 phys_offset
+= old
.memory_size
;
502 size
-= old
.memory_size
;
506 /* register prefix slot */
507 if (old
.start_addr
< start_addr
) {
508 mem
= kvm_alloc_slot(s
);
509 mem
->memory_size
= start_addr
- old
.start_addr
;
510 mem
->start_addr
= old
.start_addr
;
511 mem
->phys_offset
= old
.phys_offset
;
514 err
= kvm_set_user_memory_region(s
, mem
);
516 fprintf(stderr
, "%s: error registering prefix slot: %s\n",
517 __func__
, strerror(-err
));
522 /* register suffix slot */
523 if (old
.start_addr
+ old
.memory_size
> start_addr
+ size
) {
524 ram_addr_t size_delta
;
526 mem
= kvm_alloc_slot(s
);
527 mem
->start_addr
= start_addr
+ size
;
528 size_delta
= mem
->start_addr
- old
.start_addr
;
529 mem
->memory_size
= old
.memory_size
- size_delta
;
530 mem
->phys_offset
= old
.phys_offset
+ size_delta
;
533 err
= kvm_set_user_memory_region(s
, mem
);
535 fprintf(stderr
, "%s: error registering suffix slot: %s\n",
536 __func__
, strerror(-err
));
542 /* in case the KVM bug workaround already "consumed" the new slot */
546 /* KVM does not need to know about this memory */
547 if (flags
>= IO_MEM_UNASSIGNED
)
550 mem
= kvm_alloc_slot(s
);
551 mem
->memory_size
= size
;
552 mem
->start_addr
= start_addr
;
553 mem
->phys_offset
= phys_offset
;
556 err
= kvm_set_user_memory_region(s
, mem
);
558 fprintf(stderr
, "%s: error registering slot: %s\n", __func__
,
564 static void kvm_client_set_memory(struct CPUPhysMemoryClient
*client
,
565 target_phys_addr_t start_addr
,
567 ram_addr_t phys_offset
)
569 kvm_set_phys_mem(start_addr
, size
, phys_offset
);
572 static int kvm_client_sync_dirty_bitmap(struct CPUPhysMemoryClient
*client
,
573 target_phys_addr_t start_addr
,
574 target_phys_addr_t end_addr
)
576 return kvm_physical_sync_dirty_bitmap(start_addr
, end_addr
);
579 static int kvm_client_migration_log(struct CPUPhysMemoryClient
*client
,
582 return kvm_set_migration_log(enable
);
585 static CPUPhysMemoryClient kvm_cpu_phys_memory_client
= {
586 .set_memory
= kvm_client_set_memory
,
587 .sync_dirty_bitmap
= kvm_client_sync_dirty_bitmap
,
588 .migration_log
= kvm_client_migration_log
,
591 int kvm_init(int smp_cpus
)
593 static const char upgrade_note
[] =
594 "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n"
595 "(see http://sourceforge.net/projects/kvm).\n";
600 s
= qemu_mallocz(sizeof(KVMState
));
602 #ifdef KVM_CAP_SET_GUEST_DEBUG
603 QTAILQ_INIT(&s
->kvm_sw_breakpoints
);
605 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++)
606 s
->slots
[i
].slot
= i
;
609 s
->fd
= qemu_open("/dev/kvm", O_RDWR
);
611 fprintf(stderr
, "Could not access KVM kernel module: %m\n");
616 ret
= kvm_ioctl(s
, KVM_GET_API_VERSION
, 0);
617 if (ret
< KVM_API_VERSION
) {
620 fprintf(stderr
, "kvm version too old\n");
624 if (ret
> KVM_API_VERSION
) {
626 fprintf(stderr
, "kvm version not supported\n");
630 s
->vmfd
= kvm_ioctl(s
, KVM_CREATE_VM
, 0);
633 fprintf(stderr
, "Please add the 'switch_amode' kernel parameter to "
634 "your host kernel command line\n");
639 /* initially, KVM allocated its own memory and we had to jump through
640 * hooks to make phys_ram_base point to this. Modern versions of KVM
641 * just use a user allocated buffer so we can use regular pages
642 * unmodified. Make sure we have a sufficiently modern version of KVM.
644 if (!kvm_check_extension(s
, KVM_CAP_USER_MEMORY
)) {
646 fprintf(stderr
, "kvm does not support KVM_CAP_USER_MEMORY\n%s",
651 /* There was a nasty bug in < kvm-80 that prevents memory slots from being
652 * destroyed properly. Since we rely on this capability, refuse to work
653 * with any kernel without this capability. */
654 if (!kvm_check_extension(s
, KVM_CAP_DESTROY_MEMORY_REGION_WORKS
)) {
658 "KVM kernel module broken (DESTROY_MEMORY_REGION).\n%s",
663 s
->coalesced_mmio
= 0;
664 #ifdef KVM_CAP_COALESCED_MMIO
665 s
->coalesced_mmio
= kvm_check_extension(s
, KVM_CAP_COALESCED_MMIO
);
666 s
->coalesced_mmio_ring
= NULL
;
669 s
->broken_set_mem_region
= 1;
670 #ifdef KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
671 ret
= kvm_ioctl(s
, KVM_CHECK_EXTENSION
, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
);
673 s
->broken_set_mem_region
= 0;
678 #ifdef KVM_CAP_VCPU_EVENTS
679 s
->vcpu_events
= kvm_check_extension(s
, KVM_CAP_VCPU_EVENTS
);
682 s
->robust_singlestep
= 0;
683 #ifdef KVM_CAP_X86_ROBUST_SINGLESTEP
684 s
->robust_singlestep
=
685 kvm_check_extension(s
, KVM_CAP_X86_ROBUST_SINGLESTEP
);
689 #ifdef KVM_CAP_DEBUGREGS
690 s
->debugregs
= kvm_check_extension(s
, KVM_CAP_DEBUGREGS
);
695 s
->xsave
= kvm_check_extension(s
, KVM_CAP_XSAVE
);
700 s
->xcrs
= kvm_check_extension(s
, KVM_CAP_XCRS
);
703 ret
= kvm_arch_init(s
, smp_cpus
);
708 cpu_register_phys_memory_client(&kvm_cpu_phys_memory_client
);
724 static int kvm_handle_io(uint16_t port
, void *data
, int direction
, int size
,
730 for (i
= 0; i
< count
; i
++) {
731 if (direction
== KVM_EXIT_IO_IN
) {
734 stb_p(ptr
, cpu_inb(port
));
737 stw_p(ptr
, cpu_inw(port
));
740 stl_p(ptr
, cpu_inl(port
));
746 cpu_outb(port
, ldub_p(ptr
));
749 cpu_outw(port
, lduw_p(ptr
));
752 cpu_outl(port
, ldl_p(ptr
));
763 #ifdef KVM_CAP_INTERNAL_ERROR_DATA
764 static void kvm_handle_internal_error(CPUState
*env
, struct kvm_run
*run
)
767 if (kvm_check_extension(kvm_state
, KVM_CAP_INTERNAL_ERROR_DATA
)) {
770 fprintf(stderr
, "KVM internal error. Suberror: %d\n",
771 run
->internal
.suberror
);
773 for (i
= 0; i
< run
->internal
.ndata
; ++i
) {
774 fprintf(stderr
, "extra data[%d]: %"PRIx64
"\n",
775 i
, (uint64_t)run
->internal
.data
[i
]);
778 cpu_dump_state(env
, stderr
, fprintf
, 0);
779 if (run
->internal
.suberror
== KVM_INTERNAL_ERROR_EMULATION
) {
780 fprintf(stderr
, "emulation failure\n");
781 if (!kvm_arch_stop_on_emulation_error(env
))
784 /* FIXME: Should trigger a qmp message to let management know
785 * something went wrong.
791 void kvm_flush_coalesced_mmio_buffer(void)
793 #ifdef KVM_CAP_COALESCED_MMIO
794 KVMState
*s
= kvm_state
;
795 if (s
->coalesced_mmio_ring
) {
796 struct kvm_coalesced_mmio_ring
*ring
= s
->coalesced_mmio_ring
;
797 while (ring
->first
!= ring
->last
) {
798 struct kvm_coalesced_mmio
*ent
;
800 ent
= &ring
->coalesced_mmio
[ring
->first
];
802 cpu_physical_memory_write(ent
->phys_addr
, ent
->data
, ent
->len
);
804 ring
->first
= (ring
->first
+ 1) % KVM_COALESCED_MMIO_MAX
;
810 static void do_kvm_cpu_synchronize_state(void *_env
)
812 CPUState
*env
= _env
;
814 if (!env
->kvm_vcpu_dirty
) {
815 kvm_arch_get_registers(env
);
816 env
->kvm_vcpu_dirty
= 1;
820 void kvm_cpu_synchronize_state(CPUState
*env
)
822 if (!env
->kvm_vcpu_dirty
)
823 run_on_cpu(env
, do_kvm_cpu_synchronize_state
, env
);
826 void kvm_cpu_synchronize_post_reset(CPUState
*env
)
828 kvm_arch_put_registers(env
, KVM_PUT_RESET_STATE
);
829 env
->kvm_vcpu_dirty
= 0;
832 void kvm_cpu_synchronize_post_init(CPUState
*env
)
834 kvm_arch_put_registers(env
, KVM_PUT_FULL_STATE
);
835 env
->kvm_vcpu_dirty
= 0;
838 int kvm_cpu_exec(CPUState
*env
)
840 struct kvm_run
*run
= env
->kvm_run
;
843 DPRINTF("kvm_cpu_exec()\n");
846 #ifndef CONFIG_IOTHREAD
847 if (env
->exit_request
) {
848 DPRINTF("interrupt exit requested\n");
854 if (kvm_arch_process_irqchip_events(env
)) {
859 if (env
->kvm_vcpu_dirty
) {
860 kvm_arch_put_registers(env
, KVM_PUT_RUNTIME_STATE
);
861 env
->kvm_vcpu_dirty
= 0;
864 kvm_arch_pre_run(env
, run
);
865 cpu_single_env
= NULL
;
866 qemu_mutex_unlock_iothread();
867 ret
= kvm_vcpu_ioctl(env
, KVM_RUN
, 0);
868 qemu_mutex_lock_iothread();
869 cpu_single_env
= env
;
870 kvm_arch_post_run(env
, run
);
872 if (ret
== -EINTR
|| ret
== -EAGAIN
) {
874 DPRINTF("io window exit\n");
880 DPRINTF("kvm run failed %s\n", strerror(-ret
));
884 kvm_flush_coalesced_mmio_buffer();
886 ret
= 0; /* exit loop */
887 switch (run
->exit_reason
) {
889 DPRINTF("handle_io\n");
890 ret
= kvm_handle_io(run
->io
.port
,
891 (uint8_t *)run
+ run
->io
.data_offset
,
897 DPRINTF("handle_mmio\n");
898 cpu_physical_memory_rw(run
->mmio
.phys_addr
,
904 case KVM_EXIT_IRQ_WINDOW_OPEN
:
905 DPRINTF("irq_window_open\n");
907 case KVM_EXIT_SHUTDOWN
:
908 DPRINTF("shutdown\n");
909 qemu_system_reset_request();
912 case KVM_EXIT_UNKNOWN
:
913 DPRINTF("kvm_exit_unknown\n");
915 case KVM_EXIT_FAIL_ENTRY
:
916 DPRINTF("kvm_exit_fail_entry\n");
918 case KVM_EXIT_EXCEPTION
:
919 DPRINTF("kvm_exit_exception\n");
921 #ifdef KVM_CAP_INTERNAL_ERROR_DATA
922 case KVM_EXIT_INTERNAL_ERROR
:
923 kvm_handle_internal_error(env
, run
);
927 DPRINTF("kvm_exit_debug\n");
928 #ifdef KVM_CAP_SET_GUEST_DEBUG
929 if (kvm_arch_debug(&run
->debug
.arch
)) {
930 env
->exception_index
= EXCP_DEBUG
;
933 /* re-enter, this exception was guest-internal */
935 #endif /* KVM_CAP_SET_GUEST_DEBUG */
938 DPRINTF("kvm_arch_handle_exit\n");
939 ret
= kvm_arch_handle_exit(env
, run
);
944 if (env
->exit_request
) {
945 env
->exit_request
= 0;
946 env
->exception_index
= EXCP_INTERRUPT
;
952 int kvm_ioctl(KVMState
*s
, int type
, ...)
959 arg
= va_arg(ap
, void *);
962 ret
= ioctl(s
->fd
, type
, arg
);
969 int kvm_vm_ioctl(KVMState
*s
, int type
, ...)
976 arg
= va_arg(ap
, void *);
979 ret
= ioctl(s
->vmfd
, type
, arg
);
986 int kvm_vcpu_ioctl(CPUState
*env
, int type
, ...)
993 arg
= va_arg(ap
, void *);
996 ret
= ioctl(env
->kvm_fd
, type
, arg
);
1003 int kvm_has_sync_mmu(void)
1005 #ifdef KVM_CAP_SYNC_MMU
1006 KVMState
*s
= kvm_state
;
1008 return kvm_check_extension(s
, KVM_CAP_SYNC_MMU
);
1014 int kvm_has_vcpu_events(void)
1016 return kvm_state
->vcpu_events
;
1019 int kvm_has_robust_singlestep(void)
1021 return kvm_state
->robust_singlestep
;
1024 int kvm_has_debugregs(void)
1026 return kvm_state
->debugregs
;
1029 int kvm_has_xsave(void)
1031 return kvm_state
->xsave
;
1034 int kvm_has_xcrs(void)
1036 return kvm_state
->xcrs
;
1039 void kvm_setup_guest_memory(void *start
, size_t size
)
1041 if (!kvm_has_sync_mmu()) {
1042 #ifdef MADV_DONTFORK
1043 int ret
= madvise(start
, size
, MADV_DONTFORK
);
1051 "Need MADV_DONTFORK in absence of synchronous KVM MMU\n");
1057 #ifdef KVM_CAP_SET_GUEST_DEBUG
1058 struct kvm_sw_breakpoint
*kvm_find_sw_breakpoint(CPUState
*env
,
1061 struct kvm_sw_breakpoint
*bp
;
1063 QTAILQ_FOREACH(bp
, &env
->kvm_state
->kvm_sw_breakpoints
, entry
) {
1070 int kvm_sw_breakpoints_active(CPUState
*env
)
1072 return !QTAILQ_EMPTY(&env
->kvm_state
->kvm_sw_breakpoints
);
1075 struct kvm_set_guest_debug_data
{
1076 struct kvm_guest_debug dbg
;
1081 static void kvm_invoke_set_guest_debug(void *data
)
1083 struct kvm_set_guest_debug_data
*dbg_data
= data
;
1084 CPUState
*env
= dbg_data
->env
;
1086 dbg_data
->err
= kvm_vcpu_ioctl(env
, KVM_SET_GUEST_DEBUG
, &dbg_data
->dbg
);
1089 int kvm_update_guest_debug(CPUState
*env
, unsigned long reinject_trap
)
1091 struct kvm_set_guest_debug_data data
;
1093 data
.dbg
.control
= reinject_trap
;
1095 if (env
->singlestep_enabled
) {
1096 data
.dbg
.control
|= KVM_GUESTDBG_ENABLE
| KVM_GUESTDBG_SINGLESTEP
;
1098 kvm_arch_update_guest_debug(env
, &data
.dbg
);
1101 run_on_cpu(env
, kvm_invoke_set_guest_debug
, &data
);
1105 int kvm_insert_breakpoint(CPUState
*current_env
, target_ulong addr
,
1106 target_ulong len
, int type
)
1108 struct kvm_sw_breakpoint
*bp
;
1112 if (type
== GDB_BREAKPOINT_SW
) {
1113 bp
= kvm_find_sw_breakpoint(current_env
, addr
);
1119 bp
= qemu_malloc(sizeof(struct kvm_sw_breakpoint
));
1125 err
= kvm_arch_insert_sw_breakpoint(current_env
, bp
);
1131 QTAILQ_INSERT_HEAD(¤t_env
->kvm_state
->kvm_sw_breakpoints
,
1134 err
= kvm_arch_insert_hw_breakpoint(addr
, len
, type
);
1139 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
1140 err
= kvm_update_guest_debug(env
, 0);
1147 int kvm_remove_breakpoint(CPUState
*current_env
, target_ulong addr
,
1148 target_ulong len
, int type
)
1150 struct kvm_sw_breakpoint
*bp
;
1154 if (type
== GDB_BREAKPOINT_SW
) {
1155 bp
= kvm_find_sw_breakpoint(current_env
, addr
);
1159 if (bp
->use_count
> 1) {
1164 err
= kvm_arch_remove_sw_breakpoint(current_env
, bp
);
1168 QTAILQ_REMOVE(¤t_env
->kvm_state
->kvm_sw_breakpoints
, bp
, entry
);
1171 err
= kvm_arch_remove_hw_breakpoint(addr
, len
, type
);
1176 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
1177 err
= kvm_update_guest_debug(env
, 0);
1184 void kvm_remove_all_breakpoints(CPUState
*current_env
)
1186 struct kvm_sw_breakpoint
*bp
, *next
;
1187 KVMState
*s
= current_env
->kvm_state
;
1190 QTAILQ_FOREACH_SAFE(bp
, &s
->kvm_sw_breakpoints
, entry
, next
) {
1191 if (kvm_arch_remove_sw_breakpoint(current_env
, bp
) != 0) {
1192 /* Try harder to find a CPU that currently sees the breakpoint. */
1193 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
1194 if (kvm_arch_remove_sw_breakpoint(env
, bp
) == 0)
1199 kvm_arch_remove_all_hw_breakpoints();
1201 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
)
1202 kvm_update_guest_debug(env
, 0);
1205 #else /* !KVM_CAP_SET_GUEST_DEBUG */
1207 int kvm_update_guest_debug(CPUState
*env
, unsigned long reinject_trap
)
1212 int kvm_insert_breakpoint(CPUState
*current_env
, target_ulong addr
,
1213 target_ulong len
, int type
)
1218 int kvm_remove_breakpoint(CPUState
*current_env
, target_ulong addr
,
1219 target_ulong len
, int type
)
1224 void kvm_remove_all_breakpoints(CPUState
*current_env
)
1227 #endif /* !KVM_CAP_SET_GUEST_DEBUG */
1229 int kvm_set_signal_mask(CPUState
*env
, const sigset_t
*sigset
)
1231 struct kvm_signal_mask
*sigmask
;
1235 return kvm_vcpu_ioctl(env
, KVM_SET_SIGNAL_MASK
, NULL
);
1237 sigmask
= qemu_malloc(sizeof(*sigmask
) + sizeof(*sigset
));
1240 memcpy(sigmask
->sigset
, sigset
, sizeof(*sigset
));
1241 r
= kvm_vcpu_ioctl(env
, KVM_SET_SIGNAL_MASK
, sigmask
);
1247 int kvm_set_ioeventfd_pio_word(int fd
, uint16_t addr
, uint16_t val
, bool assign
)
1249 #ifdef KVM_IOEVENTFD
1250 struct kvm_ioeventfd kick
= {
1254 .flags
= KVM_IOEVENTFD_FLAG_DATAMATCH
| KVM_IOEVENTFD_FLAG_PIO
,
1261 kick
.flags
|= KVM_IOEVENTFD_FLAG_DEASSIGN
;
1262 r
= kvm_vm_ioctl(kvm_state
, KVM_IOEVENTFD
, &kick
);