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
;
81 static KVMState
*kvm_state
;
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 /* KVM private memory slots */
95 if (i
>= 8 && i
< 12) {
98 if (s
->slots
[i
].memory_size
== 0) {
103 fprintf(stderr
, "%s: no free slot available\n", __func__
);
107 static KVMSlot
*kvm_lookup_matching_slot(KVMState
*s
,
108 target_phys_addr_t start_addr
,
109 target_phys_addr_t end_addr
)
113 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
114 KVMSlot
*mem
= &s
->slots
[i
];
116 if (start_addr
== mem
->start_addr
&&
117 end_addr
== mem
->start_addr
+ mem
->memory_size
) {
126 * Find overlapping slot with lowest start address
128 static KVMSlot
*kvm_lookup_overlapping_slot(KVMState
*s
,
129 target_phys_addr_t start_addr
,
130 target_phys_addr_t end_addr
)
132 KVMSlot
*found
= NULL
;
135 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
136 KVMSlot
*mem
= &s
->slots
[i
];
138 if (mem
->memory_size
== 0 ||
139 (found
&& found
->start_addr
< mem
->start_addr
)) {
143 if (end_addr
> mem
->start_addr
&&
144 start_addr
< mem
->start_addr
+ mem
->memory_size
) {
152 int kvm_physical_memory_addr_from_ram(KVMState
*s
, ram_addr_t ram_addr
,
153 target_phys_addr_t
*phys_addr
)
157 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
158 KVMSlot
*mem
= &s
->slots
[i
];
160 if (ram_addr
>= mem
->phys_offset
&&
161 ram_addr
< mem
->phys_offset
+ mem
->memory_size
) {
162 *phys_addr
= mem
->start_addr
+ (ram_addr
- mem
->phys_offset
);
170 static int kvm_set_user_memory_region(KVMState
*s
, KVMSlot
*slot
)
172 struct kvm_userspace_memory_region mem
;
174 mem
.slot
= slot
->slot
;
175 mem
.guest_phys_addr
= slot
->start_addr
;
176 mem
.memory_size
= slot
->memory_size
;
177 mem
.userspace_addr
= (unsigned long)qemu_safe_ram_ptr(slot
->phys_offset
);
178 mem
.flags
= slot
->flags
;
179 if (s
->migration_log
) {
180 mem
.flags
|= KVM_MEM_LOG_DIRTY_PAGES
;
182 return kvm_vm_ioctl(s
, KVM_SET_USER_MEMORY_REGION
, &mem
);
185 static void kvm_reset_vcpu(void *opaque
)
187 CPUState
*env
= opaque
;
189 kvm_arch_reset_vcpu(env
);
192 int kvm_irqchip_in_kernel(void)
194 return kvm_state
->irqchip_in_kernel
;
197 int kvm_pit_in_kernel(void)
199 return kvm_state
->pit_in_kernel
;
203 int kvm_init_vcpu(CPUState
*env
)
205 KVMState
*s
= kvm_state
;
209 DPRINTF("kvm_init_vcpu\n");
211 ret
= kvm_vm_ioctl(s
, KVM_CREATE_VCPU
, env
->cpu_index
);
213 DPRINTF("kvm_create_vcpu failed\n");
220 mmap_size
= kvm_ioctl(s
, KVM_GET_VCPU_MMAP_SIZE
, 0);
223 DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n");
227 env
->kvm_run
= mmap(NULL
, mmap_size
, PROT_READ
| PROT_WRITE
, MAP_SHARED
,
229 if (env
->kvm_run
== MAP_FAILED
) {
231 DPRINTF("mmap'ing vcpu state failed\n");
235 if (s
->coalesced_mmio
&& !s
->coalesced_mmio_ring
) {
236 s
->coalesced_mmio_ring
=
237 (void *)env
->kvm_run
+ s
->coalesced_mmio
* PAGE_SIZE
;
240 ret
= kvm_arch_init_vcpu(env
);
242 qemu_register_reset(kvm_reset_vcpu
, env
);
243 kvm_arch_reset_vcpu(env
);
250 * dirty pages logging control
252 static int kvm_dirty_pages_log_change(target_phys_addr_t phys_addr
,
253 ram_addr_t size
, int flags
, int mask
)
255 KVMState
*s
= kvm_state
;
256 KVMSlot
*mem
= kvm_lookup_matching_slot(s
, phys_addr
, phys_addr
+ size
);
260 fprintf(stderr
, "BUG: %s: invalid parameters " TARGET_FMT_plx
"-"
261 TARGET_FMT_plx
"\n", __func__
, phys_addr
,
262 (target_phys_addr_t
)(phys_addr
+ size
- 1));
266 old_flags
= mem
->flags
;
268 flags
= (mem
->flags
& ~mask
) | flags
;
271 /* If nothing changed effectively, no need to issue ioctl */
272 if (s
->migration_log
) {
273 flags
|= KVM_MEM_LOG_DIRTY_PAGES
;
275 if (flags
== old_flags
) {
279 return kvm_set_user_memory_region(s
, mem
);
282 int kvm_log_start(target_phys_addr_t phys_addr
, ram_addr_t size
)
284 return kvm_dirty_pages_log_change(phys_addr
, size
, KVM_MEM_LOG_DIRTY_PAGES
,
285 KVM_MEM_LOG_DIRTY_PAGES
);
288 int kvm_log_stop(target_phys_addr_t phys_addr
, ram_addr_t size
)
290 return kvm_dirty_pages_log_change(phys_addr
, size
, 0,
291 KVM_MEM_LOG_DIRTY_PAGES
);
294 static int kvm_set_migration_log(int enable
)
296 KVMState
*s
= kvm_state
;
300 s
->migration_log
= enable
;
302 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
305 if (!mem
->memory_size
) {
308 if (!!(mem
->flags
& KVM_MEM_LOG_DIRTY_PAGES
) == enable
) {
311 err
= kvm_set_user_memory_region(s
, mem
);
319 /* get kvm's dirty pages bitmap and update qemu's */
320 static int kvm_get_dirty_pages_log_range(unsigned long start_addr
,
321 unsigned long *bitmap
,
322 unsigned long offset
,
323 unsigned long mem_size
)
326 unsigned long page_number
, addr
, addr1
, c
;
328 unsigned int len
= ((mem_size
/ TARGET_PAGE_SIZE
) + HOST_LONG_BITS
- 1) /
332 * bitmap-traveling is faster than memory-traveling (for addr...)
333 * especially when most of the memory is not dirty.
335 for (i
= 0; i
< len
; i
++) {
336 if (bitmap
[i
] != 0) {
337 c
= leul_to_cpu(bitmap
[i
]);
341 page_number
= i
* HOST_LONG_BITS
+ j
;
342 addr1
= page_number
* TARGET_PAGE_SIZE
;
343 addr
= offset
+ addr1
;
344 ram_addr
= cpu_get_physical_page_desc(addr
);
345 cpu_physical_memory_set_dirty(ram_addr
);
352 #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1))
355 * kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space
356 * This function updates qemu's dirty bitmap using cpu_physical_memory_set_dirty().
357 * This means all bits are set to dirty.
359 * @start_add: start of logged region.
360 * @end_addr: end of logged region.
362 static int kvm_physical_sync_dirty_bitmap(target_phys_addr_t start_addr
,
363 target_phys_addr_t end_addr
)
365 KVMState
*s
= kvm_state
;
366 unsigned long size
, allocated_size
= 0;
371 d
.dirty_bitmap
= NULL
;
372 while (start_addr
< end_addr
) {
373 mem
= kvm_lookup_overlapping_slot(s
, start_addr
, end_addr
);
378 size
= ALIGN(((mem
->memory_size
) >> TARGET_PAGE_BITS
), HOST_LONG_BITS
) / 8;
379 if (!d
.dirty_bitmap
) {
380 d
.dirty_bitmap
= qemu_malloc(size
);
381 } else if (size
> allocated_size
) {
382 d
.dirty_bitmap
= qemu_realloc(d
.dirty_bitmap
, size
);
384 allocated_size
= size
;
385 memset(d
.dirty_bitmap
, 0, allocated_size
);
389 if (kvm_vm_ioctl(s
, KVM_GET_DIRTY_LOG
, &d
) == -1) {
390 DPRINTF("ioctl failed %d\n", errno
);
395 kvm_get_dirty_pages_log_range(mem
->start_addr
, d
.dirty_bitmap
,
396 mem
->start_addr
, mem
->memory_size
);
397 start_addr
= mem
->start_addr
+ mem
->memory_size
;
399 qemu_free(d
.dirty_bitmap
);
404 int kvm_coalesce_mmio_region(target_phys_addr_t start
, ram_addr_t size
)
407 KVMState
*s
= kvm_state
;
409 if (s
->coalesced_mmio
) {
410 struct kvm_coalesced_mmio_zone zone
;
415 ret
= kvm_vm_ioctl(s
, KVM_REGISTER_COALESCED_MMIO
, &zone
);
421 int kvm_uncoalesce_mmio_region(target_phys_addr_t start
, ram_addr_t size
)
424 KVMState
*s
= kvm_state
;
426 if (s
->coalesced_mmio
) {
427 struct kvm_coalesced_mmio_zone zone
;
432 ret
= kvm_vm_ioctl(s
, KVM_UNREGISTER_COALESCED_MMIO
, &zone
);
438 int kvm_check_extension(KVMState
*s
, unsigned int extension
)
442 ret
= kvm_ioctl(s
, KVM_CHECK_EXTENSION
, extension
);
450 static int kvm_check_many_ioeventfds(void)
452 /* Userspace can use ioeventfd for io notification. This requires a host
453 * that supports eventfd(2) and an I/O thread; since eventfd does not
454 * support SIGIO it cannot interrupt the vcpu.
456 * Older kernels have a 6 device limit on the KVM io bus. Find out so we
457 * can avoid creating too many ioeventfds.
459 #if defined(CONFIG_EVENTFD) && defined(CONFIG_IOTHREAD)
462 for (i
= 0; i
< ARRAY_SIZE(ioeventfds
); i
++) {
463 ioeventfds
[i
] = eventfd(0, EFD_CLOEXEC
);
464 if (ioeventfds
[i
] < 0) {
467 ret
= kvm_set_ioeventfd_pio_word(ioeventfds
[i
], 0, i
, true);
469 close(ioeventfds
[i
]);
474 /* Decide whether many devices are supported or not */
475 ret
= i
== ARRAY_SIZE(ioeventfds
);
478 kvm_set_ioeventfd_pio_word(ioeventfds
[i
], 0, i
, false);
479 close(ioeventfds
[i
]);
487 static const KVMCapabilityInfo
*
488 kvm_check_extension_list(KVMState
*s
, const KVMCapabilityInfo
*list
)
491 if (!kvm_check_extension(s
, list
->value
)) {
499 static void kvm_set_phys_mem(target_phys_addr_t start_addr
, ram_addr_t size
,
500 ram_addr_t phys_offset
)
502 KVMState
*s
= kvm_state
;
503 ram_addr_t flags
= phys_offset
& ~TARGET_PAGE_MASK
;
507 /* kvm works in page size chunks, but the function may be called
508 with sub-page size and unaligned start address. */
509 size
= TARGET_PAGE_ALIGN(size
);
510 start_addr
= TARGET_PAGE_ALIGN(start_addr
);
512 /* KVM does not support read-only slots */
513 phys_offset
&= ~IO_MEM_ROM
;
516 mem
= kvm_lookup_overlapping_slot(s
, start_addr
, start_addr
+ size
);
521 if (flags
< IO_MEM_UNASSIGNED
&& start_addr
>= mem
->start_addr
&&
522 (start_addr
+ size
<= mem
->start_addr
+ mem
->memory_size
) &&
523 (phys_offset
- start_addr
== mem
->phys_offset
- mem
->start_addr
)) {
524 /* The new slot fits into the existing one and comes with
525 * identical parameters - nothing to be done. */
531 /* unregister the overlapping slot */
532 mem
->memory_size
= 0;
533 err
= kvm_set_user_memory_region(s
, mem
);
535 fprintf(stderr
, "%s: error unregistering overlapping slot: %s\n",
536 __func__
, strerror(-err
));
540 /* Workaround for older KVM versions: we can't join slots, even not by
541 * unregistering the previous ones and then registering the larger
542 * slot. We have to maintain the existing fragmentation. Sigh.
544 * This workaround assumes that the new slot starts at the same
545 * address as the first existing one. If not or if some overlapping
546 * slot comes around later, we will fail (not seen in practice so far)
547 * - and actually require a recent KVM version. */
548 if (s
->broken_set_mem_region
&&
549 old
.start_addr
== start_addr
&& old
.memory_size
< size
&&
550 flags
< IO_MEM_UNASSIGNED
) {
551 mem
= kvm_alloc_slot(s
);
552 mem
->memory_size
= old
.memory_size
;
553 mem
->start_addr
= old
.start_addr
;
554 mem
->phys_offset
= old
.phys_offset
;
557 err
= kvm_set_user_memory_region(s
, mem
);
559 fprintf(stderr
, "%s: error updating slot: %s\n", __func__
,
564 start_addr
+= old
.memory_size
;
565 phys_offset
+= old
.memory_size
;
566 size
-= old
.memory_size
;
570 /* register prefix slot */
571 if (old
.start_addr
< start_addr
) {
572 mem
= kvm_alloc_slot(s
);
573 mem
->memory_size
= start_addr
- old
.start_addr
;
574 mem
->start_addr
= old
.start_addr
;
575 mem
->phys_offset
= old
.phys_offset
;
578 err
= kvm_set_user_memory_region(s
, mem
);
580 fprintf(stderr
, "%s: error registering prefix slot: %s\n",
581 __func__
, strerror(-err
));
586 /* register suffix slot */
587 if (old
.start_addr
+ old
.memory_size
> start_addr
+ size
) {
588 ram_addr_t size_delta
;
590 mem
= kvm_alloc_slot(s
);
591 mem
->start_addr
= start_addr
+ size
;
592 size_delta
= mem
->start_addr
- old
.start_addr
;
593 mem
->memory_size
= old
.memory_size
- size_delta
;
594 mem
->phys_offset
= old
.phys_offset
+ size_delta
;
597 err
= kvm_set_user_memory_region(s
, mem
);
599 fprintf(stderr
, "%s: error registering suffix slot: %s\n",
600 __func__
, strerror(-err
));
606 /* in case the KVM bug workaround already "consumed" the new slot */
610 /* KVM does not need to know about this memory */
611 if (flags
>= IO_MEM_UNASSIGNED
) {
614 mem
= kvm_alloc_slot(s
);
615 mem
->memory_size
= size
;
616 mem
->start_addr
= start_addr
;
617 mem
->phys_offset
= phys_offset
;
620 err
= kvm_set_user_memory_region(s
, mem
);
622 fprintf(stderr
, "%s: error registering slot: %s\n", __func__
,
628 static void kvm_client_set_memory(struct CPUPhysMemoryClient
*client
,
629 target_phys_addr_t start_addr
,
630 ram_addr_t size
, ram_addr_t phys_offset
)
632 kvm_set_phys_mem(start_addr
, size
, phys_offset
);
635 static int kvm_client_sync_dirty_bitmap(struct CPUPhysMemoryClient
*client
,
636 target_phys_addr_t start_addr
,
637 target_phys_addr_t end_addr
)
639 return kvm_physical_sync_dirty_bitmap(start_addr
, end_addr
);
642 static int kvm_client_migration_log(struct CPUPhysMemoryClient
*client
,
645 return kvm_set_migration_log(enable
);
648 static CPUPhysMemoryClient kvm_cpu_phys_memory_client
= {
649 .set_memory
= kvm_client_set_memory
,
650 .sync_dirty_bitmap
= kvm_client_sync_dirty_bitmap
,
651 .migration_log
= kvm_client_migration_log
,
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 int 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
));
817 #ifdef KVM_CAP_INTERNAL_ERROR_DATA
818 static int kvm_handle_internal_error(CPUState
*env
, struct kvm_run
*run
)
820 fprintf(stderr
, "KVM internal error.");
821 if (kvm_check_extension(kvm_state
, KVM_CAP_INTERNAL_ERROR_DATA
)) {
824 fprintf(stderr
, " Suberror: %d\n", run
->internal
.suberror
);
825 for (i
= 0; i
< run
->internal
.ndata
; ++i
) {
826 fprintf(stderr
, "extra data[%d]: %"PRIx64
"\n",
827 i
, (uint64_t)run
->internal
.data
[i
]);
830 fprintf(stderr
, "\n");
832 if (run
->internal
.suberror
== KVM_INTERNAL_ERROR_EMULATION
) {
833 fprintf(stderr
, "emulation failure\n");
834 if (!kvm_arch_stop_on_emulation_error(env
)) {
835 cpu_dump_state(env
, stderr
, fprintf
, CPU_DUMP_CODE
);
839 /* FIXME: Should trigger a qmp message to let management know
840 * something went wrong.
846 void kvm_flush_coalesced_mmio_buffer(void)
848 KVMState
*s
= kvm_state
;
849 if (s
->coalesced_mmio_ring
) {
850 struct kvm_coalesced_mmio_ring
*ring
= s
->coalesced_mmio_ring
;
851 while (ring
->first
!= ring
->last
) {
852 struct kvm_coalesced_mmio
*ent
;
854 ent
= &ring
->coalesced_mmio
[ring
->first
];
856 cpu_physical_memory_write(ent
->phys_addr
, ent
->data
, ent
->len
);
858 ring
->first
= (ring
->first
+ 1) % KVM_COALESCED_MMIO_MAX
;
863 static void do_kvm_cpu_synchronize_state(void *_env
)
865 CPUState
*env
= _env
;
867 if (!env
->kvm_vcpu_dirty
) {
868 kvm_arch_get_registers(env
);
869 env
->kvm_vcpu_dirty
= 1;
873 void kvm_cpu_synchronize_state(CPUState
*env
)
875 if (!env
->kvm_vcpu_dirty
) {
876 run_on_cpu(env
, do_kvm_cpu_synchronize_state
, env
);
880 void kvm_cpu_synchronize_post_reset(CPUState
*env
)
882 kvm_arch_put_registers(env
, KVM_PUT_RESET_STATE
);
883 env
->kvm_vcpu_dirty
= 0;
886 void kvm_cpu_synchronize_post_init(CPUState
*env
)
888 kvm_arch_put_registers(env
, KVM_PUT_FULL_STATE
);
889 env
->kvm_vcpu_dirty
= 0;
892 int kvm_cpu_exec(CPUState
*env
)
894 struct kvm_run
*run
= env
->kvm_run
;
897 DPRINTF("kvm_cpu_exec()\n");
900 #ifndef CONFIG_IOTHREAD
901 if (env
->exit_request
) {
902 DPRINTF("interrupt exit requested\n");
908 if (kvm_arch_process_irqchip_events(env
)) {
913 if (env
->kvm_vcpu_dirty
) {
914 kvm_arch_put_registers(env
, KVM_PUT_RUNTIME_STATE
);
915 env
->kvm_vcpu_dirty
= 0;
918 kvm_arch_pre_run(env
, run
);
919 cpu_single_env
= NULL
;
920 qemu_mutex_unlock_iothread();
921 ret
= kvm_vcpu_ioctl(env
, KVM_RUN
, 0);
922 qemu_mutex_lock_iothread();
923 cpu_single_env
= env
;
924 kvm_arch_post_run(env
, run
);
926 kvm_flush_coalesced_mmio_buffer();
928 if (ret
== -EINTR
|| ret
== -EAGAIN
) {
930 DPRINTF("io window exit\n");
936 DPRINTF("kvm run failed %s\n", strerror(-ret
));
940 ret
= 0; /* exit loop */
941 switch (run
->exit_reason
) {
943 DPRINTF("handle_io\n");
944 ret
= kvm_handle_io(run
->io
.port
,
945 (uint8_t *)run
+ run
->io
.data_offset
,
951 DPRINTF("handle_mmio\n");
952 cpu_physical_memory_rw(run
->mmio
.phys_addr
,
958 case KVM_EXIT_IRQ_WINDOW_OPEN
:
959 DPRINTF("irq_window_open\n");
961 case KVM_EXIT_SHUTDOWN
:
962 DPRINTF("shutdown\n");
963 qemu_system_reset_request();
966 case KVM_EXIT_UNKNOWN
:
967 fprintf(stderr
, "KVM: unknown exit, hardware reason %" PRIx64
"\n",
968 (uint64_t)run
->hw
.hardware_exit_reason
);
971 #ifdef KVM_CAP_INTERNAL_ERROR_DATA
972 case KVM_EXIT_INTERNAL_ERROR
:
973 ret
= kvm_handle_internal_error(env
, run
);
977 DPRINTF("kvm_exit_debug\n");
978 #ifdef KVM_CAP_SET_GUEST_DEBUG
979 if (kvm_arch_debug(&run
->debug
.arch
)) {
980 env
->exception_index
= EXCP_DEBUG
;
983 /* re-enter, this exception was guest-internal */
985 #endif /* KVM_CAP_SET_GUEST_DEBUG */
988 DPRINTF("kvm_arch_handle_exit\n");
989 ret
= kvm_arch_handle_exit(env
, run
);
995 cpu_dump_state(env
, stderr
, fprintf
, CPU_DUMP_CODE
);
997 env
->exit_request
= 1;
999 if (env
->exit_request
) {
1000 env
->exit_request
= 0;
1001 env
->exception_index
= EXCP_INTERRUPT
;
1007 int kvm_ioctl(KVMState
*s
, int type
, ...)
1014 arg
= va_arg(ap
, void *);
1017 ret
= ioctl(s
->fd
, type
, arg
);
1024 int kvm_vm_ioctl(KVMState
*s
, int type
, ...)
1031 arg
= va_arg(ap
, void *);
1034 ret
= ioctl(s
->vmfd
, type
, arg
);
1041 int kvm_vcpu_ioctl(CPUState
*env
, int type
, ...)
1048 arg
= va_arg(ap
, void *);
1051 ret
= ioctl(env
->kvm_fd
, type
, arg
);
1058 int kvm_has_sync_mmu(void)
1060 return kvm_check_extension(kvm_state
, KVM_CAP_SYNC_MMU
);
1063 int kvm_has_vcpu_events(void)
1065 return kvm_state
->vcpu_events
;
1068 int kvm_has_robust_singlestep(void)
1070 return kvm_state
->robust_singlestep
;
1073 int kvm_has_debugregs(void)
1075 return kvm_state
->debugregs
;
1078 int kvm_has_xsave(void)
1080 return kvm_state
->xsave
;
1083 int kvm_has_xcrs(void)
1085 return kvm_state
->xcrs
;
1088 int kvm_has_many_ioeventfds(void)
1090 if (!kvm_enabled()) {
1093 return kvm_state
->many_ioeventfds
;
1096 void kvm_setup_guest_memory(void *start
, size_t size
)
1098 if (!kvm_has_sync_mmu()) {
1099 int ret
= qemu_madvise(start
, size
, QEMU_MADV_DONTFORK
);
1102 perror("qemu_madvise");
1104 "Need MADV_DONTFORK in absence of synchronous KVM MMU\n");
1110 #ifdef KVM_CAP_SET_GUEST_DEBUG
1111 struct kvm_sw_breakpoint
*kvm_find_sw_breakpoint(CPUState
*env
,
1114 struct kvm_sw_breakpoint
*bp
;
1116 QTAILQ_FOREACH(bp
, &env
->kvm_state
->kvm_sw_breakpoints
, entry
) {
1124 int kvm_sw_breakpoints_active(CPUState
*env
)
1126 return !QTAILQ_EMPTY(&env
->kvm_state
->kvm_sw_breakpoints
);
1129 struct kvm_set_guest_debug_data
{
1130 struct kvm_guest_debug dbg
;
1135 static void kvm_invoke_set_guest_debug(void *data
)
1137 struct kvm_set_guest_debug_data
*dbg_data
= data
;
1138 CPUState
*env
= dbg_data
->env
;
1140 dbg_data
->err
= kvm_vcpu_ioctl(env
, KVM_SET_GUEST_DEBUG
, &dbg_data
->dbg
);
1143 int kvm_update_guest_debug(CPUState
*env
, unsigned long reinject_trap
)
1145 struct kvm_set_guest_debug_data data
;
1147 data
.dbg
.control
= reinject_trap
;
1149 if (env
->singlestep_enabled
) {
1150 data
.dbg
.control
|= KVM_GUESTDBG_ENABLE
| KVM_GUESTDBG_SINGLESTEP
;
1152 kvm_arch_update_guest_debug(env
, &data
.dbg
);
1155 run_on_cpu(env
, kvm_invoke_set_guest_debug
, &data
);
1159 int kvm_insert_breakpoint(CPUState
*current_env
, target_ulong addr
,
1160 target_ulong len
, int type
)
1162 struct kvm_sw_breakpoint
*bp
;
1166 if (type
== GDB_BREAKPOINT_SW
) {
1167 bp
= kvm_find_sw_breakpoint(current_env
, addr
);
1173 bp
= qemu_malloc(sizeof(struct kvm_sw_breakpoint
));
1180 err
= kvm_arch_insert_sw_breakpoint(current_env
, bp
);
1186 QTAILQ_INSERT_HEAD(¤t_env
->kvm_state
->kvm_sw_breakpoints
,
1189 err
= kvm_arch_insert_hw_breakpoint(addr
, len
, type
);
1195 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
1196 err
= kvm_update_guest_debug(env
, 0);
1204 int kvm_remove_breakpoint(CPUState
*current_env
, target_ulong addr
,
1205 target_ulong len
, int type
)
1207 struct kvm_sw_breakpoint
*bp
;
1211 if (type
== GDB_BREAKPOINT_SW
) {
1212 bp
= kvm_find_sw_breakpoint(current_env
, addr
);
1217 if (bp
->use_count
> 1) {
1222 err
= kvm_arch_remove_sw_breakpoint(current_env
, bp
);
1227 QTAILQ_REMOVE(¤t_env
->kvm_state
->kvm_sw_breakpoints
, bp
, entry
);
1230 err
= kvm_arch_remove_hw_breakpoint(addr
, len
, type
);
1236 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
1237 err
= kvm_update_guest_debug(env
, 0);
1245 void kvm_remove_all_breakpoints(CPUState
*current_env
)
1247 struct kvm_sw_breakpoint
*bp
, *next
;
1248 KVMState
*s
= current_env
->kvm_state
;
1251 QTAILQ_FOREACH_SAFE(bp
, &s
->kvm_sw_breakpoints
, entry
, next
) {
1252 if (kvm_arch_remove_sw_breakpoint(current_env
, bp
) != 0) {
1253 /* Try harder to find a CPU that currently sees the breakpoint. */
1254 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
1255 if (kvm_arch_remove_sw_breakpoint(env
, bp
) == 0) {
1261 kvm_arch_remove_all_hw_breakpoints();
1263 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
1264 kvm_update_guest_debug(env
, 0);
1268 #else /* !KVM_CAP_SET_GUEST_DEBUG */
1270 int kvm_update_guest_debug(CPUState
*env
, unsigned long reinject_trap
)
1275 int kvm_insert_breakpoint(CPUState
*current_env
, target_ulong addr
,
1276 target_ulong len
, int type
)
1281 int kvm_remove_breakpoint(CPUState
*current_env
, target_ulong addr
,
1282 target_ulong len
, int type
)
1287 void kvm_remove_all_breakpoints(CPUState
*current_env
)
1290 #endif /* !KVM_CAP_SET_GUEST_DEBUG */
1292 int kvm_set_signal_mask(CPUState
*env
, const sigset_t
*sigset
)
1294 struct kvm_signal_mask
*sigmask
;
1298 return kvm_vcpu_ioctl(env
, KVM_SET_SIGNAL_MASK
, NULL
);
1301 sigmask
= qemu_malloc(sizeof(*sigmask
) + sizeof(*sigset
));
1304 memcpy(sigmask
->sigset
, sigset
, sizeof(*sigset
));
1305 r
= kvm_vcpu_ioctl(env
, KVM_SET_SIGNAL_MASK
, sigmask
);
1311 int kvm_set_ioeventfd_mmio_long(int fd
, uint32_t addr
, uint32_t val
, bool assign
)
1313 #ifdef KVM_IOEVENTFD
1315 struct kvm_ioeventfd iofd
;
1317 iofd
.datamatch
= val
;
1320 iofd
.flags
= KVM_IOEVENTFD_FLAG_DATAMATCH
;
1323 if (!kvm_enabled()) {
1328 iofd
.flags
|= KVM_IOEVENTFD_FLAG_DEASSIGN
;
1331 ret
= kvm_vm_ioctl(kvm_state
, KVM_IOEVENTFD
, &iofd
);
1343 int kvm_set_ioeventfd_pio_word(int fd
, uint16_t addr
, uint16_t val
, bool assign
)
1345 #ifdef KVM_IOEVENTFD
1346 struct kvm_ioeventfd kick
= {
1350 .flags
= KVM_IOEVENTFD_FLAG_DATAMATCH
| KVM_IOEVENTFD_FLAG_PIO
,
1354 if (!kvm_enabled()) {
1358 kick
.flags
|= KVM_IOEVENTFD_FLAG_DEASSIGN
;
1360 r
= kvm_vm_ioctl(kvm_state
, KVM_IOEVENTFD
, &kick
);