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 #ifdef KVM_CAP_COALESCED_MMIO
67 struct kvm_coalesced_mmio_ring
*coalesced_mmio_ring
;
69 int broken_set_mem_region
;
72 int robust_singlestep
;
74 #ifdef KVM_CAP_SET_GUEST_DEBUG
75 struct kvm_sw_breakpoint_head kvm_sw_breakpoints
;
77 int irqchip_in_kernel
;
83 static KVMState
*kvm_state
;
85 static KVMSlot
*kvm_alloc_slot(KVMState
*s
)
89 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
90 /* KVM private memory slots */
91 if (i
>= 8 && i
< 12) {
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
;
199 int kvm_init_vcpu(CPUState
*env
)
201 KVMState
*s
= kvm_state
;
205 DPRINTF("kvm_init_vcpu\n");
207 ret
= kvm_vm_ioctl(s
, KVM_CREATE_VCPU
, env
->cpu_index
);
209 DPRINTF("kvm_create_vcpu failed\n");
216 mmap_size
= kvm_ioctl(s
, KVM_GET_VCPU_MMAP_SIZE
, 0);
218 DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n");
222 env
->kvm_run
= mmap(NULL
, mmap_size
, PROT_READ
| PROT_WRITE
, MAP_SHARED
,
224 if (env
->kvm_run
== MAP_FAILED
) {
226 DPRINTF("mmap'ing vcpu state failed\n");
230 #ifdef KVM_CAP_COALESCED_MMIO
231 if (s
->coalesced_mmio
&& !s
->coalesced_mmio_ring
) {
232 s
->coalesced_mmio_ring
=
233 (void *)env
->kvm_run
+ s
->coalesced_mmio
* PAGE_SIZE
;
237 ret
= kvm_arch_init_vcpu(env
);
239 qemu_register_reset(kvm_reset_vcpu
, env
);
240 kvm_arch_reset_vcpu(env
);
247 * dirty pages logging control
249 static int kvm_dirty_pages_log_change(target_phys_addr_t phys_addr
,
250 ram_addr_t size
, int flags
, int mask
)
252 KVMState
*s
= kvm_state
;
253 KVMSlot
*mem
= kvm_lookup_matching_slot(s
, phys_addr
, phys_addr
+ size
);
257 fprintf(stderr
, "BUG: %s: invalid parameters " TARGET_FMT_plx
"-"
258 TARGET_FMT_plx
"\n", __func__
, phys_addr
,
259 (target_phys_addr_t
)(phys_addr
+ size
- 1));
263 old_flags
= mem
->flags
;
265 flags
= (mem
->flags
& ~mask
) | flags
;
268 /* If nothing changed effectively, no need to issue ioctl */
269 if (s
->migration_log
) {
270 flags
|= KVM_MEM_LOG_DIRTY_PAGES
;
272 if (flags
== old_flags
) {
276 return kvm_set_user_memory_region(s
, mem
);
279 int kvm_log_start(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 int kvm_log_stop(target_phys_addr_t phys_addr
, ram_addr_t size
)
287 return kvm_dirty_pages_log_change(phys_addr
, size
, 0,
288 KVM_MEM_LOG_DIRTY_PAGES
);
291 static int kvm_set_migration_log(int enable
)
293 KVMState
*s
= kvm_state
;
297 s
->migration_log
= enable
;
299 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
302 if (!mem
->memory_size
) {
305 if (!!(mem
->flags
& KVM_MEM_LOG_DIRTY_PAGES
) == enable
) {
308 err
= kvm_set_user_memory_region(s
, mem
);
316 /* get kvm's dirty pages bitmap and update qemu's */
317 static int kvm_get_dirty_pages_log_range(unsigned long start_addr
,
318 unsigned long *bitmap
,
319 unsigned long offset
,
320 unsigned long mem_size
)
323 unsigned long page_number
, addr
, addr1
, c
;
325 unsigned int len
= ((mem_size
/ TARGET_PAGE_SIZE
) + HOST_LONG_BITS
- 1) /
329 * bitmap-traveling is faster than memory-traveling (for addr...)
330 * especially when most of the memory is not dirty.
332 for (i
= 0; i
< len
; i
++) {
333 if (bitmap
[i
] != 0) {
334 c
= leul_to_cpu(bitmap
[i
]);
338 page_number
= i
* HOST_LONG_BITS
+ j
;
339 addr1
= page_number
* TARGET_PAGE_SIZE
;
340 addr
= offset
+ addr1
;
341 ram_addr
= cpu_get_physical_page_desc(addr
);
342 cpu_physical_memory_set_dirty(ram_addr
);
349 #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1))
352 * kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space
353 * This function updates qemu's dirty bitmap using cpu_physical_memory_set_dirty().
354 * This means all bits are set to dirty.
356 * @start_add: start of logged region.
357 * @end_addr: end of logged region.
359 static int kvm_physical_sync_dirty_bitmap(target_phys_addr_t start_addr
,
360 target_phys_addr_t end_addr
)
362 KVMState
*s
= kvm_state
;
363 unsigned long size
, allocated_size
= 0;
368 d
.dirty_bitmap
= NULL
;
369 while (start_addr
< end_addr
) {
370 mem
= kvm_lookup_overlapping_slot(s
, start_addr
, end_addr
);
375 size
= ALIGN(((mem
->memory_size
) >> TARGET_PAGE_BITS
), HOST_LONG_BITS
) / 8;
376 if (!d
.dirty_bitmap
) {
377 d
.dirty_bitmap
= qemu_malloc(size
);
378 } else if (size
> allocated_size
) {
379 d
.dirty_bitmap
= qemu_realloc(d
.dirty_bitmap
, size
);
381 allocated_size
= size
;
382 memset(d
.dirty_bitmap
, 0, allocated_size
);
386 if (kvm_vm_ioctl(s
, KVM_GET_DIRTY_LOG
, &d
) == -1) {
387 DPRINTF("ioctl failed %d\n", errno
);
392 kvm_get_dirty_pages_log_range(mem
->start_addr
, d
.dirty_bitmap
,
393 mem
->start_addr
, mem
->memory_size
);
394 start_addr
= mem
->start_addr
+ mem
->memory_size
;
396 qemu_free(d
.dirty_bitmap
);
401 int kvm_coalesce_mmio_region(target_phys_addr_t start
, ram_addr_t size
)
404 #ifdef KVM_CAP_COALESCED_MMIO
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
);
420 int kvm_uncoalesce_mmio_region(target_phys_addr_t start
, ram_addr_t size
)
423 #ifdef KVM_CAP_COALESCED_MMIO
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
);
439 int kvm_check_extension(KVMState
*s
, unsigned int extension
)
443 ret
= kvm_ioctl(s
, KVM_CHECK_EXTENSION
, extension
);
451 static int kvm_check_many_ioeventfds(void)
453 /* Older kernels have a 6 device limit on the KVM io bus. Find out so we
454 * can avoid creating too many ioeventfds.
456 #ifdef CONFIG_EVENTFD
459 for (i
= 0; i
< ARRAY_SIZE(ioeventfds
); i
++) {
460 ioeventfds
[i
] = eventfd(0, EFD_CLOEXEC
);
461 if (ioeventfds
[i
] < 0) {
464 ret
= kvm_set_ioeventfd_pio_word(ioeventfds
[i
], 0, i
, true);
466 close(ioeventfds
[i
]);
471 /* Decide whether many devices are supported or not */
472 ret
= i
== ARRAY_SIZE(ioeventfds
);
475 kvm_set_ioeventfd_pio_word(ioeventfds
[i
], 0, i
, false);
476 close(ioeventfds
[i
]);
484 static void kvm_set_phys_mem(target_phys_addr_t start_addr
, ram_addr_t size
,
485 ram_addr_t phys_offset
)
487 KVMState
*s
= kvm_state
;
488 ram_addr_t flags
= phys_offset
& ~TARGET_PAGE_MASK
;
492 /* kvm works in page size chunks, but the function may be called
493 with sub-page size and unaligned start address. */
494 size
= TARGET_PAGE_ALIGN(size
);
495 start_addr
= TARGET_PAGE_ALIGN(start_addr
);
497 /* KVM does not support read-only slots */
498 phys_offset
&= ~IO_MEM_ROM
;
501 mem
= kvm_lookup_overlapping_slot(s
, start_addr
, start_addr
+ size
);
506 if (flags
< IO_MEM_UNASSIGNED
&& start_addr
>= mem
->start_addr
&&
507 (start_addr
+ size
<= mem
->start_addr
+ mem
->memory_size
) &&
508 (phys_offset
- start_addr
== mem
->phys_offset
- mem
->start_addr
)) {
509 /* The new slot fits into the existing one and comes with
510 * identical parameters - nothing to be done. */
516 /* unregister the overlapping slot */
517 mem
->memory_size
= 0;
518 err
= kvm_set_user_memory_region(s
, mem
);
520 fprintf(stderr
, "%s: error unregistering overlapping slot: %s\n",
521 __func__
, strerror(-err
));
525 /* Workaround for older KVM versions: we can't join slots, even not by
526 * unregistering the previous ones and then registering the larger
527 * slot. We have to maintain the existing fragmentation. Sigh.
529 * This workaround assumes that the new slot starts at the same
530 * address as the first existing one. If not or if some overlapping
531 * slot comes around later, we will fail (not seen in practice so far)
532 * - and actually require a recent KVM version. */
533 if (s
->broken_set_mem_region
&&
534 old
.start_addr
== start_addr
&& old
.memory_size
< size
&&
535 flags
< IO_MEM_UNASSIGNED
) {
536 mem
= kvm_alloc_slot(s
);
537 mem
->memory_size
= old
.memory_size
;
538 mem
->start_addr
= old
.start_addr
;
539 mem
->phys_offset
= old
.phys_offset
;
542 err
= kvm_set_user_memory_region(s
, mem
);
544 fprintf(stderr
, "%s: error updating slot: %s\n", __func__
,
549 start_addr
+= old
.memory_size
;
550 phys_offset
+= old
.memory_size
;
551 size
-= old
.memory_size
;
555 /* register prefix slot */
556 if (old
.start_addr
< start_addr
) {
557 mem
= kvm_alloc_slot(s
);
558 mem
->memory_size
= start_addr
- old
.start_addr
;
559 mem
->start_addr
= old
.start_addr
;
560 mem
->phys_offset
= old
.phys_offset
;
563 err
= kvm_set_user_memory_region(s
, mem
);
565 fprintf(stderr
, "%s: error registering prefix slot: %s\n",
566 __func__
, strerror(-err
));
571 /* register suffix slot */
572 if (old
.start_addr
+ old
.memory_size
> start_addr
+ size
) {
573 ram_addr_t size_delta
;
575 mem
= kvm_alloc_slot(s
);
576 mem
->start_addr
= start_addr
+ size
;
577 size_delta
= mem
->start_addr
- old
.start_addr
;
578 mem
->memory_size
= old
.memory_size
- size_delta
;
579 mem
->phys_offset
= old
.phys_offset
+ size_delta
;
582 err
= kvm_set_user_memory_region(s
, mem
);
584 fprintf(stderr
, "%s: error registering suffix slot: %s\n",
585 __func__
, strerror(-err
));
591 /* in case the KVM bug workaround already "consumed" the new slot */
595 /* KVM does not need to know about this memory */
596 if (flags
>= IO_MEM_UNASSIGNED
) {
599 mem
= kvm_alloc_slot(s
);
600 mem
->memory_size
= size
;
601 mem
->start_addr
= start_addr
;
602 mem
->phys_offset
= phys_offset
;
605 err
= kvm_set_user_memory_region(s
, mem
);
607 fprintf(stderr
, "%s: error registering slot: %s\n", __func__
,
613 static void kvm_client_set_memory(struct CPUPhysMemoryClient
*client
,
614 target_phys_addr_t start_addr
,
615 ram_addr_t size
, ram_addr_t phys_offset
)
617 kvm_set_phys_mem(start_addr
, size
, phys_offset
);
620 static int kvm_client_sync_dirty_bitmap(struct CPUPhysMemoryClient
*client
,
621 target_phys_addr_t start_addr
,
622 target_phys_addr_t end_addr
)
624 return kvm_physical_sync_dirty_bitmap(start_addr
, end_addr
);
627 static int kvm_client_migration_log(struct CPUPhysMemoryClient
*client
,
630 return kvm_set_migration_log(enable
);
633 static CPUPhysMemoryClient kvm_cpu_phys_memory_client
= {
634 .set_memory
= kvm_client_set_memory
,
635 .sync_dirty_bitmap
= kvm_client_sync_dirty_bitmap
,
636 .migration_log
= kvm_client_migration_log
,
639 int kvm_init(int smp_cpus
)
641 static const char upgrade_note
[] =
642 "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n"
643 "(see http://sourceforge.net/projects/kvm).\n";
648 s
= qemu_mallocz(sizeof(KVMState
));
650 #ifdef KVM_CAP_SET_GUEST_DEBUG
651 QTAILQ_INIT(&s
->kvm_sw_breakpoints
);
653 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
654 s
->slots
[i
].slot
= i
;
657 s
->fd
= qemu_open("/dev/kvm", O_RDWR
);
659 fprintf(stderr
, "Could not access KVM kernel module: %m\n");
664 ret
= kvm_ioctl(s
, KVM_GET_API_VERSION
, 0);
665 if (ret
< KVM_API_VERSION
) {
669 fprintf(stderr
, "kvm version too old\n");
673 if (ret
> KVM_API_VERSION
) {
675 fprintf(stderr
, "kvm version not supported\n");
679 s
->vmfd
= kvm_ioctl(s
, KVM_CREATE_VM
, 0);
682 fprintf(stderr
, "Please add the 'switch_amode' kernel parameter to "
683 "your host kernel command line\n");
688 /* initially, KVM allocated its own memory and we had to jump through
689 * hooks to make phys_ram_base point to this. Modern versions of KVM
690 * just use a user allocated buffer so we can use regular pages
691 * unmodified. Make sure we have a sufficiently modern version of KVM.
693 if (!kvm_check_extension(s
, KVM_CAP_USER_MEMORY
)) {
695 fprintf(stderr
, "kvm does not support KVM_CAP_USER_MEMORY\n%s",
700 /* There was a nasty bug in < kvm-80 that prevents memory slots from being
701 * destroyed properly. Since we rely on this capability, refuse to work
702 * with any kernel without this capability. */
703 if (!kvm_check_extension(s
, KVM_CAP_DESTROY_MEMORY_REGION_WORKS
)) {
707 "KVM kernel module broken (DESTROY_MEMORY_REGION).\n%s",
712 s
->coalesced_mmio
= 0;
713 #ifdef KVM_CAP_COALESCED_MMIO
714 s
->coalesced_mmio
= kvm_check_extension(s
, KVM_CAP_COALESCED_MMIO
);
715 s
->coalesced_mmio_ring
= NULL
;
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
, smp_cpus
);
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
, 0);
839 /* FIXME: Should trigger a qmp message to let management know
840 * something went wrong.
846 void kvm_flush_coalesced_mmio_buffer(void)
848 #ifdef KVM_CAP_COALESCED_MMIO
849 KVMState
*s
= kvm_state
;
850 if (s
->coalesced_mmio_ring
) {
851 struct kvm_coalesced_mmio_ring
*ring
= s
->coalesced_mmio_ring
;
852 while (ring
->first
!= ring
->last
) {
853 struct kvm_coalesced_mmio
*ent
;
855 ent
= &ring
->coalesced_mmio
[ring
->first
];
857 cpu_physical_memory_write(ent
->phys_addr
, ent
->data
, ent
->len
);
859 ring
->first
= (ring
->first
+ 1) % KVM_COALESCED_MMIO_MAX
;
865 static void do_kvm_cpu_synchronize_state(void *_env
)
867 CPUState
*env
= _env
;
869 if (!env
->kvm_vcpu_dirty
) {
870 kvm_arch_get_registers(env
);
871 env
->kvm_vcpu_dirty
= 1;
875 void kvm_cpu_synchronize_state(CPUState
*env
)
877 if (!env
->kvm_vcpu_dirty
) {
878 run_on_cpu(env
, do_kvm_cpu_synchronize_state
, env
);
882 void kvm_cpu_synchronize_post_reset(CPUState
*env
)
884 kvm_arch_put_registers(env
, KVM_PUT_RESET_STATE
);
885 env
->kvm_vcpu_dirty
= 0;
888 void kvm_cpu_synchronize_post_init(CPUState
*env
)
890 kvm_arch_put_registers(env
, KVM_PUT_FULL_STATE
);
891 env
->kvm_vcpu_dirty
= 0;
894 int kvm_cpu_exec(CPUState
*env
)
896 struct kvm_run
*run
= env
->kvm_run
;
899 DPRINTF("kvm_cpu_exec()\n");
902 #ifndef CONFIG_IOTHREAD
903 if (env
->exit_request
) {
904 DPRINTF("interrupt exit requested\n");
910 if (kvm_arch_process_irqchip_events(env
)) {
915 if (env
->kvm_vcpu_dirty
) {
916 kvm_arch_put_registers(env
, KVM_PUT_RUNTIME_STATE
);
917 env
->kvm_vcpu_dirty
= 0;
920 kvm_arch_pre_run(env
, run
);
921 cpu_single_env
= NULL
;
922 qemu_mutex_unlock_iothread();
923 ret
= kvm_vcpu_ioctl(env
, KVM_RUN
, 0);
924 qemu_mutex_lock_iothread();
925 cpu_single_env
= env
;
926 kvm_arch_post_run(env
, run
);
928 if (ret
== -EINTR
|| ret
== -EAGAIN
) {
930 DPRINTF("io window exit\n");
936 DPRINTF("kvm run failed %s\n", strerror(-ret
));
940 kvm_flush_coalesced_mmio_buffer();
942 ret
= 0; /* exit loop */
943 switch (run
->exit_reason
) {
945 DPRINTF("handle_io\n");
946 ret
= kvm_handle_io(run
->io
.port
,
947 (uint8_t *)run
+ run
->io
.data_offset
,
953 DPRINTF("handle_mmio\n");
954 cpu_physical_memory_rw(run
->mmio
.phys_addr
,
960 case KVM_EXIT_IRQ_WINDOW_OPEN
:
961 DPRINTF("irq_window_open\n");
963 case KVM_EXIT_SHUTDOWN
:
964 DPRINTF("shutdown\n");
965 qemu_system_reset_request();
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
);
979 DPRINTF("kvm_exit_debug\n");
980 #ifdef KVM_CAP_SET_GUEST_DEBUG
981 if (kvm_arch_debug(&run
->debug
.arch
)) {
982 env
->exception_index
= EXCP_DEBUG
;
985 /* re-enter, this exception was guest-internal */
987 #endif /* KVM_CAP_SET_GUEST_DEBUG */
990 DPRINTF("kvm_arch_handle_exit\n");
991 ret
= kvm_arch_handle_exit(env
, run
);
997 cpu_dump_state(env
, stderr
, fprintf
, 0);
999 env
->exit_request
= 1;
1001 if (env
->exit_request
) {
1002 env
->exit_request
= 0;
1003 env
->exception_index
= EXCP_INTERRUPT
;
1009 int kvm_ioctl(KVMState
*s
, int type
, ...)
1016 arg
= va_arg(ap
, void *);
1019 ret
= ioctl(s
->fd
, type
, arg
);
1026 int kvm_vm_ioctl(KVMState
*s
, int type
, ...)
1033 arg
= va_arg(ap
, void *);
1036 ret
= ioctl(s
->vmfd
, type
, arg
);
1043 int kvm_vcpu_ioctl(CPUState
*env
, int type
, ...)
1050 arg
= va_arg(ap
, void *);
1053 ret
= ioctl(env
->kvm_fd
, type
, arg
);
1060 int kvm_has_sync_mmu(void)
1062 #ifdef KVM_CAP_SYNC_MMU
1063 KVMState
*s
= kvm_state
;
1065 return kvm_check_extension(s
, KVM_CAP_SYNC_MMU
);
1071 int kvm_has_vcpu_events(void)
1073 return kvm_state
->vcpu_events
;
1076 int kvm_has_robust_singlestep(void)
1078 return kvm_state
->robust_singlestep
;
1081 int kvm_has_debugregs(void)
1083 return kvm_state
->debugregs
;
1086 int kvm_has_xsave(void)
1088 return kvm_state
->xsave
;
1091 int kvm_has_xcrs(void)
1093 return kvm_state
->xcrs
;
1096 int kvm_has_many_ioeventfds(void)
1098 if (!kvm_enabled()) {
1101 return kvm_state
->many_ioeventfds
;
1104 void kvm_setup_guest_memory(void *start
, size_t size
)
1106 if (!kvm_has_sync_mmu()) {
1107 int ret
= qemu_madvise(start
, size
, QEMU_MADV_DONTFORK
);
1110 perror("qemu_madvise");
1112 "Need MADV_DONTFORK in absence of synchronous KVM MMU\n");
1118 #ifdef KVM_CAP_SET_GUEST_DEBUG
1119 struct kvm_sw_breakpoint
*kvm_find_sw_breakpoint(CPUState
*env
,
1122 struct kvm_sw_breakpoint
*bp
;
1124 QTAILQ_FOREACH(bp
, &env
->kvm_state
->kvm_sw_breakpoints
, entry
) {
1132 int kvm_sw_breakpoints_active(CPUState
*env
)
1134 return !QTAILQ_EMPTY(&env
->kvm_state
->kvm_sw_breakpoints
);
1137 struct kvm_set_guest_debug_data
{
1138 struct kvm_guest_debug dbg
;
1143 static void kvm_invoke_set_guest_debug(void *data
)
1145 struct kvm_set_guest_debug_data
*dbg_data
= data
;
1146 CPUState
*env
= dbg_data
->env
;
1148 dbg_data
->err
= kvm_vcpu_ioctl(env
, KVM_SET_GUEST_DEBUG
, &dbg_data
->dbg
);
1151 int kvm_update_guest_debug(CPUState
*env
, unsigned long reinject_trap
)
1153 struct kvm_set_guest_debug_data data
;
1155 data
.dbg
.control
= reinject_trap
;
1157 if (env
->singlestep_enabled
) {
1158 data
.dbg
.control
|= KVM_GUESTDBG_ENABLE
| KVM_GUESTDBG_SINGLESTEP
;
1160 kvm_arch_update_guest_debug(env
, &data
.dbg
);
1163 run_on_cpu(env
, kvm_invoke_set_guest_debug
, &data
);
1167 int kvm_insert_breakpoint(CPUState
*current_env
, target_ulong addr
,
1168 target_ulong len
, int type
)
1170 struct kvm_sw_breakpoint
*bp
;
1174 if (type
== GDB_BREAKPOINT_SW
) {
1175 bp
= kvm_find_sw_breakpoint(current_env
, addr
);
1181 bp
= qemu_malloc(sizeof(struct kvm_sw_breakpoint
));
1188 err
= kvm_arch_insert_sw_breakpoint(current_env
, bp
);
1194 QTAILQ_INSERT_HEAD(¤t_env
->kvm_state
->kvm_sw_breakpoints
,
1197 err
= kvm_arch_insert_hw_breakpoint(addr
, len
, type
);
1203 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
1204 err
= kvm_update_guest_debug(env
, 0);
1212 int kvm_remove_breakpoint(CPUState
*current_env
, target_ulong addr
,
1213 target_ulong len
, int type
)
1215 struct kvm_sw_breakpoint
*bp
;
1219 if (type
== GDB_BREAKPOINT_SW
) {
1220 bp
= kvm_find_sw_breakpoint(current_env
, addr
);
1225 if (bp
->use_count
> 1) {
1230 err
= kvm_arch_remove_sw_breakpoint(current_env
, bp
);
1235 QTAILQ_REMOVE(¤t_env
->kvm_state
->kvm_sw_breakpoints
, bp
, entry
);
1238 err
= kvm_arch_remove_hw_breakpoint(addr
, len
, type
);
1244 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
1245 err
= kvm_update_guest_debug(env
, 0);
1253 void kvm_remove_all_breakpoints(CPUState
*current_env
)
1255 struct kvm_sw_breakpoint
*bp
, *next
;
1256 KVMState
*s
= current_env
->kvm_state
;
1259 QTAILQ_FOREACH_SAFE(bp
, &s
->kvm_sw_breakpoints
, entry
, next
) {
1260 if (kvm_arch_remove_sw_breakpoint(current_env
, bp
) != 0) {
1261 /* Try harder to find a CPU that currently sees the breakpoint. */
1262 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
1263 if (kvm_arch_remove_sw_breakpoint(env
, bp
) == 0) {
1269 kvm_arch_remove_all_hw_breakpoints();
1271 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
1272 kvm_update_guest_debug(env
, 0);
1276 #else /* !KVM_CAP_SET_GUEST_DEBUG */
1278 int kvm_update_guest_debug(CPUState
*env
, unsigned long reinject_trap
)
1283 int kvm_insert_breakpoint(CPUState
*current_env
, target_ulong addr
,
1284 target_ulong len
, int type
)
1289 int kvm_remove_breakpoint(CPUState
*current_env
, target_ulong addr
,
1290 target_ulong len
, int type
)
1295 void kvm_remove_all_breakpoints(CPUState
*current_env
)
1298 #endif /* !KVM_CAP_SET_GUEST_DEBUG */
1300 int kvm_set_signal_mask(CPUState
*env
, const sigset_t
*sigset
)
1302 struct kvm_signal_mask
*sigmask
;
1306 return kvm_vcpu_ioctl(env
, KVM_SET_SIGNAL_MASK
, NULL
);
1309 sigmask
= qemu_malloc(sizeof(*sigmask
) + sizeof(*sigset
));
1312 memcpy(sigmask
->sigset
, sigset
, sizeof(*sigset
));
1313 r
= kvm_vcpu_ioctl(env
, KVM_SET_SIGNAL_MASK
, sigmask
);
1319 int kvm_set_ioeventfd_mmio_long(int fd
, uint32_t addr
, uint32_t val
, bool assign
)
1321 #ifdef KVM_IOEVENTFD
1323 struct kvm_ioeventfd iofd
;
1325 iofd
.datamatch
= val
;
1328 iofd
.flags
= KVM_IOEVENTFD_FLAG_DATAMATCH
;
1331 if (!kvm_enabled()) {
1336 iofd
.flags
|= KVM_IOEVENTFD_FLAG_DEASSIGN
;
1339 ret
= kvm_vm_ioctl(kvm_state
, KVM_IOEVENTFD
, &iofd
);
1351 int kvm_set_ioeventfd_pio_word(int fd
, uint16_t addr
, uint16_t val
, bool assign
)
1353 #ifdef KVM_IOEVENTFD
1354 struct kvm_ioeventfd kick
= {
1358 .flags
= KVM_IOEVENTFD_FLAG_DATAMATCH
| KVM_IOEVENTFD_FLAG_PIO
,
1362 if (!kvm_enabled()) {
1366 kick
.flags
|= KVM_IOEVENTFD_FLAG_DEASSIGN
;
1368 r
= kvm_vm_ioctl(kvm_state
, KVM_IOEVENTFD
, &kick
);