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
38 #define DPRINTF(fmt, ...) \
39 do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
41 #define DPRINTF(fmt, ...) \
45 typedef struct KVMSlot
47 target_phys_addr_t start_addr
;
48 ram_addr_t memory_size
;
49 ram_addr_t phys_offset
;
54 typedef struct kvm_dirty_log KVMDirtyLog
;
62 #ifdef KVM_CAP_COALESCED_MMIO
63 struct kvm_coalesced_mmio_ring
*coalesced_mmio_ring
;
65 int broken_set_mem_region
;
68 int robust_singlestep
;
70 #ifdef KVM_CAP_SET_GUEST_DEBUG
71 struct kvm_sw_breakpoint_head kvm_sw_breakpoints
;
73 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
);
163 int kvm_irqchip_in_kernel(void)
165 return kvm_state
->irqchip_in_kernel
;
168 int kvm_pit_in_kernel(void)
170 return kvm_state
->pit_in_kernel
;
175 int kvm_init_vcpu(CPUState
*env
)
177 KVMState
*s
= kvm_state
;
181 DPRINTF("kvm_init_vcpu\n");
183 ret
= kvm_vm_ioctl(s
, KVM_CREATE_VCPU
, env
->cpu_index
);
185 DPRINTF("kvm_create_vcpu failed\n");
192 mmap_size
= kvm_ioctl(s
, KVM_GET_VCPU_MMAP_SIZE
, 0);
194 DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n");
198 env
->kvm_run
= mmap(NULL
, mmap_size
, PROT_READ
| PROT_WRITE
, MAP_SHARED
,
200 if (env
->kvm_run
== MAP_FAILED
) {
202 DPRINTF("mmap'ing vcpu state failed\n");
206 #ifdef KVM_CAP_COALESCED_MMIO
207 if (s
->coalesced_mmio
&& !s
->coalesced_mmio_ring
)
208 s
->coalesced_mmio_ring
= (void *) env
->kvm_run
+
209 s
->coalesced_mmio
* PAGE_SIZE
;
212 ret
= kvm_arch_init_vcpu(env
);
214 qemu_register_reset(kvm_reset_vcpu
, env
);
215 kvm_arch_reset_vcpu(env
);
222 * dirty pages logging control
224 static int kvm_dirty_pages_log_change(target_phys_addr_t phys_addr
,
225 ram_addr_t size
, int flags
, int mask
)
227 KVMState
*s
= kvm_state
;
228 KVMSlot
*mem
= kvm_lookup_matching_slot(s
, phys_addr
, phys_addr
+ size
);
232 fprintf(stderr
, "BUG: %s: invalid parameters " TARGET_FMT_plx
"-"
233 TARGET_FMT_plx
"\n", __func__
, phys_addr
,
234 (target_phys_addr_t
)(phys_addr
+ size
- 1));
238 old_flags
= mem
->flags
;
240 flags
= (mem
->flags
& ~mask
) | flags
;
243 /* If nothing changed effectively, no need to issue ioctl */
244 if (s
->migration_log
) {
245 flags
|= KVM_MEM_LOG_DIRTY_PAGES
;
247 if (flags
== old_flags
) {
251 return kvm_set_user_memory_region(s
, mem
);
254 int kvm_log_start(target_phys_addr_t phys_addr
, ram_addr_t size
)
256 return kvm_dirty_pages_log_change(phys_addr
, size
,
257 KVM_MEM_LOG_DIRTY_PAGES
,
258 KVM_MEM_LOG_DIRTY_PAGES
);
261 int kvm_log_stop(target_phys_addr_t phys_addr
, ram_addr_t size
)
263 return kvm_dirty_pages_log_change(phys_addr
, size
,
265 KVM_MEM_LOG_DIRTY_PAGES
);
268 static int kvm_set_migration_log(int enable
)
270 KVMState
*s
= kvm_state
;
274 s
->migration_log
= enable
;
276 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
279 if (!!(mem
->flags
& KVM_MEM_LOG_DIRTY_PAGES
) == enable
) {
282 err
= kvm_set_user_memory_region(s
, mem
);
290 /* get kvm's dirty pages bitmap and update qemu's */
291 static int kvm_get_dirty_pages_log_range(unsigned long start_addr
,
292 unsigned long *bitmap
,
293 unsigned long offset
,
294 unsigned long mem_size
)
297 unsigned long page_number
, addr
, addr1
, c
;
299 unsigned int len
= ((mem_size
/ TARGET_PAGE_SIZE
) + HOST_LONG_BITS
- 1) /
303 * bitmap-traveling is faster than memory-traveling (for addr...)
304 * especially when most of the memory is not dirty.
306 for (i
= 0; i
< len
; i
++) {
307 if (bitmap
[i
] != 0) {
308 c
= leul_to_cpu(bitmap
[i
]);
312 page_number
= i
* HOST_LONG_BITS
+ j
;
313 addr1
= page_number
* TARGET_PAGE_SIZE
;
314 addr
= offset
+ addr1
;
315 ram_addr
= cpu_get_physical_page_desc(addr
);
316 cpu_physical_memory_set_dirty(ram_addr
);
323 #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1))
326 * kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space
327 * This function updates qemu's dirty bitmap using cpu_physical_memory_set_dirty().
328 * This means all bits are set to dirty.
330 * @start_add: start of logged region.
331 * @end_addr: end of logged region.
333 static int kvm_physical_sync_dirty_bitmap(target_phys_addr_t start_addr
,
334 target_phys_addr_t end_addr
)
336 KVMState
*s
= kvm_state
;
337 unsigned long size
, allocated_size
= 0;
342 d
.dirty_bitmap
= NULL
;
343 while (start_addr
< end_addr
) {
344 mem
= kvm_lookup_overlapping_slot(s
, start_addr
, end_addr
);
349 size
= ALIGN(((mem
->memory_size
) >> TARGET_PAGE_BITS
), HOST_LONG_BITS
) / 8;
350 if (!d
.dirty_bitmap
) {
351 d
.dirty_bitmap
= qemu_malloc(size
);
352 } else if (size
> allocated_size
) {
353 d
.dirty_bitmap
= qemu_realloc(d
.dirty_bitmap
, size
);
355 allocated_size
= size
;
356 memset(d
.dirty_bitmap
, 0, allocated_size
);
360 if (kvm_vm_ioctl(s
, KVM_GET_DIRTY_LOG
, &d
) == -1) {
361 DPRINTF("ioctl failed %d\n", errno
);
366 kvm_get_dirty_pages_log_range(mem
->start_addr
, d
.dirty_bitmap
,
367 mem
->start_addr
, mem
->memory_size
);
368 start_addr
= mem
->start_addr
+ mem
->memory_size
;
370 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
);
427 static void kvm_set_phys_mem(target_phys_addr_t start_addr
,
429 ram_addr_t phys_offset
)
431 KVMState
*s
= kvm_state
;
432 ram_addr_t flags
= phys_offset
& ~TARGET_PAGE_MASK
;
436 if (start_addr
& ~TARGET_PAGE_MASK
) {
437 if (flags
>= IO_MEM_UNASSIGNED
) {
438 if (!kvm_lookup_overlapping_slot(s
, start_addr
,
439 start_addr
+ size
)) {
442 fprintf(stderr
, "Unaligned split of a KVM memory slot\n");
444 fprintf(stderr
, "Only page-aligned memory slots supported\n");
449 /* KVM does not support read-only slots */
450 phys_offset
&= ~IO_MEM_ROM
;
453 mem
= kvm_lookup_overlapping_slot(s
, start_addr
, start_addr
+ size
);
458 if (flags
< IO_MEM_UNASSIGNED
&& start_addr
>= mem
->start_addr
&&
459 (start_addr
+ size
<= mem
->start_addr
+ mem
->memory_size
) &&
460 (phys_offset
- start_addr
== mem
->phys_offset
- mem
->start_addr
)) {
461 /* The new slot fits into the existing one and comes with
462 * identical parameters - nothing to be done. */
468 /* unregister the overlapping slot */
469 mem
->memory_size
= 0;
470 err
= kvm_set_user_memory_region(s
, mem
);
472 fprintf(stderr
, "%s: error unregistering overlapping slot: %s\n",
473 __func__
, strerror(-err
));
477 /* Workaround for older KVM versions: we can't join slots, even not by
478 * unregistering the previous ones and then registering the larger
479 * slot. We have to maintain the existing fragmentation. Sigh.
481 * This workaround assumes that the new slot starts at the same
482 * address as the first existing one. If not or if some overlapping
483 * slot comes around later, we will fail (not seen in practice so far)
484 * - and actually require a recent KVM version. */
485 if (s
->broken_set_mem_region
&&
486 old
.start_addr
== start_addr
&& old
.memory_size
< size
&&
487 flags
< IO_MEM_UNASSIGNED
) {
488 mem
= kvm_alloc_slot(s
);
489 mem
->memory_size
= old
.memory_size
;
490 mem
->start_addr
= old
.start_addr
;
491 mem
->phys_offset
= old
.phys_offset
;
494 err
= kvm_set_user_memory_region(s
, mem
);
496 fprintf(stderr
, "%s: error updating slot: %s\n", __func__
,
501 start_addr
+= old
.memory_size
;
502 phys_offset
+= old
.memory_size
;
503 size
-= old
.memory_size
;
507 /* register prefix slot */
508 if (old
.start_addr
< start_addr
) {
509 mem
= kvm_alloc_slot(s
);
510 mem
->memory_size
= start_addr
- old
.start_addr
;
511 mem
->start_addr
= old
.start_addr
;
512 mem
->phys_offset
= old
.phys_offset
;
515 err
= kvm_set_user_memory_region(s
, mem
);
517 fprintf(stderr
, "%s: error registering prefix slot: %s\n",
518 __func__
, strerror(-err
));
523 /* register suffix slot */
524 if (old
.start_addr
+ old
.memory_size
> start_addr
+ size
) {
525 ram_addr_t size_delta
;
527 mem
= kvm_alloc_slot(s
);
528 mem
->start_addr
= start_addr
+ size
;
529 size_delta
= mem
->start_addr
- old
.start_addr
;
530 mem
->memory_size
= old
.memory_size
- size_delta
;
531 mem
->phys_offset
= old
.phys_offset
+ size_delta
;
534 err
= kvm_set_user_memory_region(s
, mem
);
536 fprintf(stderr
, "%s: error registering suffix slot: %s\n",
537 __func__
, strerror(-err
));
543 /* in case the KVM bug workaround already "consumed" the new slot */
547 /* KVM does not need to know about this memory */
548 if (flags
>= IO_MEM_UNASSIGNED
)
551 mem
= kvm_alloc_slot(s
);
552 mem
->memory_size
= size
;
553 mem
->start_addr
= start_addr
;
554 mem
->phys_offset
= phys_offset
;
557 err
= kvm_set_user_memory_region(s
, mem
);
559 fprintf(stderr
, "%s: error registering slot: %s\n", __func__
,
567 static void kvm_client_set_memory(struct CPUPhysMemoryClient
*client
,
568 target_phys_addr_t start_addr
,
570 ram_addr_t phys_offset
)
572 kvm_set_phys_mem(start_addr
, size
, phys_offset
);
575 static int kvm_client_sync_dirty_bitmap(struct CPUPhysMemoryClient
*client
,
576 target_phys_addr_t start_addr
,
577 target_phys_addr_t end_addr
)
579 return kvm_physical_sync_dirty_bitmap(start_addr
, end_addr
);
582 static int kvm_client_migration_log(struct CPUPhysMemoryClient
*client
,
585 return kvm_set_migration_log(enable
);
588 static CPUPhysMemoryClient kvm_cpu_phys_memory_client
= {
589 .set_memory
= kvm_client_set_memory
,
590 .sync_dirty_bitmap
= kvm_client_sync_dirty_bitmap
,
591 .migration_log
= kvm_client_migration_log
,
595 void kvm_cpu_register_phys_memory_client(void)
597 cpu_register_phys_memory_client(&kvm_cpu_phys_memory_client
);
602 int kvm_init(int smp_cpus
)
604 static const char upgrade_note
[] =
605 "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n"
606 "(see http://sourceforge.net/projects/kvm).\n";
612 fprintf(stderr
, "No SMP KVM support, use '-smp 1'\n");
616 s
= qemu_mallocz(sizeof(KVMState
));
618 #ifdef KVM_CAP_SET_GUEST_DEBUG
619 QTAILQ_INIT(&s
->kvm_sw_breakpoints
);
621 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++)
622 s
->slots
[i
].slot
= i
;
625 s
->fd
= qemu_open("/dev/kvm", O_RDWR
);
627 fprintf(stderr
, "Could not access KVM kernel module: %m\n");
632 ret
= kvm_ioctl(s
, KVM_GET_API_VERSION
, 0);
633 if (ret
< KVM_API_VERSION
) {
636 fprintf(stderr
, "kvm version too old\n");
640 if (ret
> KVM_API_VERSION
) {
642 fprintf(stderr
, "kvm version not supported\n");
646 s
->vmfd
= kvm_ioctl(s
, KVM_CREATE_VM
, 0);
649 fprintf(stderr
, "Please add the 'switch_amode' kernel parameter to "
650 "your host kernel command line\n");
655 /* initially, KVM allocated its own memory and we had to jump through
656 * hooks to make phys_ram_base point to this. Modern versions of KVM
657 * just use a user allocated buffer so we can use regular pages
658 * unmodified. Make sure we have a sufficiently modern version of KVM.
660 if (!kvm_check_extension(s
, KVM_CAP_USER_MEMORY
)) {
662 fprintf(stderr
, "kvm does not support KVM_CAP_USER_MEMORY\n%s",
667 /* There was a nasty bug in < kvm-80 that prevents memory slots from being
668 * destroyed properly. Since we rely on this capability, refuse to work
669 * with any kernel without this capability. */
670 if (!kvm_check_extension(s
, KVM_CAP_DESTROY_MEMORY_REGION_WORKS
)) {
674 "KVM kernel module broken (DESTROY_MEMORY_REGION).\n%s",
679 s
->coalesced_mmio
= 0;
680 #ifdef KVM_CAP_COALESCED_MMIO
681 s
->coalesced_mmio
= kvm_check_extension(s
, KVM_CAP_COALESCED_MMIO
);
682 s
->coalesced_mmio_ring
= NULL
;
685 s
->broken_set_mem_region
= 1;
686 #ifdef KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
687 ret
= kvm_ioctl(s
, KVM_CHECK_EXTENSION
, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
);
689 s
->broken_set_mem_region
= 0;
694 #ifdef KVM_CAP_VCPU_EVENTS
695 s
->vcpu_events
= kvm_check_extension(s
, KVM_CAP_VCPU_EVENTS
);
698 s
->robust_singlestep
= 0;
699 #ifdef KVM_CAP_X86_ROBUST_SINGLESTEP
700 s
->robust_singlestep
=
701 kvm_check_extension(s
, KVM_CAP_X86_ROBUST_SINGLESTEP
);
705 #ifdef KVM_CAP_DEBUGREGS
706 s
->debugregs
= kvm_check_extension(s
, KVM_CAP_DEBUGREGS
);
709 ret
= kvm_arch_init(s
, smp_cpus
);
714 cpu_register_phys_memory_client(&kvm_cpu_phys_memory_client
);
731 static int kvm_handle_io(uint16_t port
, void *data
, int direction
, int size
,
737 for (i
= 0; i
< count
; i
++) {
738 if (direction
== KVM_EXIT_IO_IN
) {
741 stb_p(ptr
, cpu_inb(port
));
744 stw_p(ptr
, cpu_inw(port
));
747 stl_p(ptr
, cpu_inl(port
));
753 cpu_outb(port
, ldub_p(ptr
));
756 cpu_outw(port
, lduw_p(ptr
));
759 cpu_outl(port
, ldl_p(ptr
));
770 #ifdef KVM_CAP_INTERNAL_ERROR_DATA
771 static void kvm_handle_internal_error(CPUState
*env
, struct kvm_run
*run
)
774 if (kvm_check_extension(kvm_state
, KVM_CAP_INTERNAL_ERROR_DATA
)) {
777 fprintf(stderr
, "KVM internal error. Suberror: %d\n",
778 run
->internal
.suberror
);
780 for (i
= 0; i
< run
->internal
.ndata
; ++i
) {
781 fprintf(stderr
, "extra data[%d]: %"PRIx64
"\n",
782 i
, (uint64_t)run
->internal
.data
[i
]);
785 cpu_dump_state(env
, stderr
, fprintf
, 0);
786 if (run
->internal
.suberror
== KVM_INTERNAL_ERROR_EMULATION
) {
787 fprintf(stderr
, "emulation failure\n");
789 /* FIXME: Should trigger a qmp message to let management know
790 * something went wrong.
796 void kvm_flush_coalesced_mmio_buffer(void)
798 #ifdef KVM_CAP_COALESCED_MMIO
799 KVMState
*s
= kvm_state
;
800 if (s
->coalesced_mmio_ring
) {
801 struct kvm_coalesced_mmio_ring
*ring
= s
->coalesced_mmio_ring
;
802 while (ring
->first
!= ring
->last
) {
803 struct kvm_coalesced_mmio
*ent
;
805 ent
= &ring
->coalesced_mmio
[ring
->first
];
807 cpu_physical_memory_write(ent
->phys_addr
, ent
->data
, ent
->len
);
809 ring
->first
= (ring
->first
+ 1) % KVM_COALESCED_MMIO_MAX
;
817 void kvm_cpu_synchronize_state(CPUState
*env
)
819 if (!env
->kvm_vcpu_dirty
) {
820 kvm_arch_get_registers(env
);
821 env
->kvm_vcpu_dirty
= 1;
825 void kvm_cpu_synchronize_post_reset(CPUState
*env
)
827 kvm_arch_put_registers(env
, KVM_PUT_RESET_STATE
);
828 env
->kvm_vcpu_dirty
= 0;
831 void kvm_cpu_synchronize_post_init(CPUState
*env
)
833 kvm_arch_put_registers(env
, KVM_PUT_FULL_STATE
);
834 env
->kvm_vcpu_dirty
= 0;
837 int kvm_cpu_exec(CPUState
*env
)
839 struct kvm_run
*run
= env
->kvm_run
;
842 DPRINTF("kvm_cpu_exec()\n");
845 #ifndef CONFIG_IOTHREAD
846 if (env
->exit_request
) {
847 DPRINTF("interrupt exit requested\n");
853 if (env
->kvm_vcpu_dirty
) {
854 kvm_arch_put_registers(env
, KVM_PUT_RUNTIME_STATE
);
855 env
->kvm_vcpu_dirty
= 0;
858 kvm_arch_pre_run(env
, run
);
859 qemu_mutex_unlock_iothread();
860 ret
= kvm_vcpu_ioctl(env
, KVM_RUN
, 0);
861 qemu_mutex_lock_iothread();
862 kvm_arch_post_run(env
, run
);
864 if (ret
== -EINTR
|| ret
== -EAGAIN
) {
866 DPRINTF("io window exit\n");
872 DPRINTF("kvm run failed %s\n", strerror(-ret
));
876 kvm_flush_coalesced_mmio_buffer();
878 ret
= 0; /* exit loop */
879 switch (run
->exit_reason
) {
881 DPRINTF("handle_io\n");
882 ret
= kvm_handle_io(run
->io
.port
,
883 (uint8_t *)run
+ run
->io
.data_offset
,
889 DPRINTF("handle_mmio\n");
890 cpu_physical_memory_rw(run
->mmio
.phys_addr
,
896 case KVM_EXIT_IRQ_WINDOW_OPEN
:
897 DPRINTF("irq_window_open\n");
899 case KVM_EXIT_SHUTDOWN
:
900 DPRINTF("shutdown\n");
901 qemu_system_reset_request();
904 case KVM_EXIT_UNKNOWN
:
905 DPRINTF("kvm_exit_unknown\n");
907 case KVM_EXIT_FAIL_ENTRY
:
908 DPRINTF("kvm_exit_fail_entry\n");
910 case KVM_EXIT_EXCEPTION
:
911 DPRINTF("kvm_exit_exception\n");
913 #ifdef KVM_CAP_INTERNAL_ERROR_DATA
914 case KVM_EXIT_INTERNAL_ERROR
:
915 kvm_handle_internal_error(env
, run
);
919 DPRINTF("kvm_exit_debug\n");
920 #ifdef KVM_CAP_SET_GUEST_DEBUG
921 if (kvm_arch_debug(&run
->debug
.arch
)) {
922 gdb_set_stop_cpu(env
);
924 env
->exception_index
= EXCP_DEBUG
;
927 /* re-enter, this exception was guest-internal */
929 #endif /* KVM_CAP_SET_GUEST_DEBUG */
932 DPRINTF("kvm_arch_handle_exit\n");
933 ret
= kvm_arch_handle_exit(env
, run
);
938 if (env
->exit_request
) {
939 env
->exit_request
= 0;
940 env
->exception_index
= EXCP_INTERRUPT
;
947 int kvm_ioctl(KVMState
*s
, int type
, ...)
954 arg
= va_arg(ap
, void *);
957 ret
= ioctl(s
->fd
, type
, arg
);
964 int kvm_vm_ioctl(KVMState
*s
, int type
, ...)
971 arg
= va_arg(ap
, void *);
974 ret
= ioctl(s
->vmfd
, type
, arg
);
981 int kvm_vcpu_ioctl(CPUState
*env
, int type
, ...)
988 arg
= va_arg(ap
, void *);
991 ret
= ioctl(env
->kvm_fd
, type
, arg
);
998 int kvm_has_sync_mmu(void)
1000 #ifdef KVM_CAP_SYNC_MMU
1001 KVMState
*s
= kvm_state
;
1003 return kvm_check_extension(s
, KVM_CAP_SYNC_MMU
);
1009 int kvm_has_vcpu_events(void)
1011 return kvm_state
->vcpu_events
;
1014 int kvm_has_robust_singlestep(void)
1016 return kvm_state
->robust_singlestep
;
1019 int kvm_has_debugregs(void)
1021 return kvm_state
->debugregs
;
1024 void kvm_setup_guest_memory(void *start
, size_t size
)
1026 if (!kvm_has_sync_mmu()) {
1027 #ifdef MADV_DONTFORK
1028 int ret
= madvise(start
, size
, MADV_DONTFORK
);
1036 "Need MADV_DONTFORK in absence of synchronous KVM MMU\n");
1042 #ifdef KVM_CAP_SET_GUEST_DEBUG
1045 static void on_vcpu(CPUState
*env
, void (*func
)(void *data
), void *data
)
1047 #ifdef CONFIG_IOTHREAD
1048 if (env
!= cpu_single_env
) {
1054 #else /* !KVM_UPSTREAM */
1055 static void on_vcpu(CPUState
*env
, void (*func
)(void *data
), void *data
);
1056 #endif /* !KVM_UPSTREAM */
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 on_vcpu(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
);
1271 #if defined(KVM_IRQFD)
1272 int kvm_set_irqfd(int gsi
, int fd
, bool assigned
)
1274 struct kvm_irqfd irqfd
= {
1277 .flags
= assigned
? 0 : KVM_IRQFD_FLAG_DEASSIGN
,
1280 if (!kvm_enabled() || !kvm_irqchip_in_kernel())
1283 r
= kvm_vm_ioctl(kvm_state
, KVM_IRQFD
, &irqfd
);
1291 #include "qemu-kvm.c"