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 #ifdef OBSOLETE_KVM_IMPL
51 typedef struct KVMSlot
53 target_phys_addr_t start_addr
;
54 ram_addr_t memory_size
;
55 ram_addr_t phys_offset
;
60 typedef struct kvm_dirty_log KVMDirtyLog
;
68 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
;
86 static const KVMCapabilityInfo kvm_required_capabilites
[] = {
87 KVM_CAP_INFO(USER_MEMORY
),
88 KVM_CAP_INFO(DESTROY_MEMORY_REGION_WORKS
),
94 static KVMSlot
*kvm_alloc_slot(KVMState
*s
)
98 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
99 /* KVM private memory slots */
100 if (i
>= 8 && i
< 12) {
103 if (s
->slots
[i
].memory_size
== 0) {
108 fprintf(stderr
, "%s: no free slot available\n", __func__
);
112 static KVMSlot
*kvm_lookup_matching_slot(KVMState
*s
,
113 target_phys_addr_t start_addr
,
114 target_phys_addr_t end_addr
)
118 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
119 KVMSlot
*mem
= &s
->slots
[i
];
121 if (start_addr
== mem
->start_addr
&&
122 end_addr
== mem
->start_addr
+ mem
->memory_size
) {
131 * Find overlapping slot with lowest start address
133 static KVMSlot
*kvm_lookup_overlapping_slot(KVMState
*s
,
134 target_phys_addr_t start_addr
,
135 target_phys_addr_t end_addr
)
137 KVMSlot
*found
= NULL
;
140 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
141 KVMSlot
*mem
= &s
->slots
[i
];
143 if (mem
->memory_size
== 0 ||
144 (found
&& found
->start_addr
< mem
->start_addr
)) {
148 if (end_addr
> mem
->start_addr
&&
149 start_addr
< mem
->start_addr
+ mem
->memory_size
) {
157 int kvm_physical_memory_addr_from_ram(KVMState
*s
, ram_addr_t ram_addr
,
158 target_phys_addr_t
*phys_addr
)
162 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
163 KVMSlot
*mem
= &s
->slots
[i
];
165 if (ram_addr
>= mem
->phys_offset
&&
166 ram_addr
< mem
->phys_offset
+ mem
->memory_size
) {
167 *phys_addr
= mem
->start_addr
+ (ram_addr
- mem
->phys_offset
);
175 static int kvm_set_user_memory_region(KVMState
*s
, KVMSlot
*slot
)
177 struct kvm_userspace_memory_region mem
;
179 mem
.slot
= slot
->slot
;
180 mem
.guest_phys_addr
= slot
->start_addr
;
181 mem
.memory_size
= slot
->memory_size
;
182 mem
.userspace_addr
= (unsigned long)qemu_safe_ram_ptr(slot
->phys_offset
);
183 mem
.flags
= slot
->flags
;
184 if (s
->migration_log
) {
185 mem
.flags
|= KVM_MEM_LOG_DIRTY_PAGES
;
187 return kvm_vm_ioctl(s
, KVM_SET_USER_MEMORY_REGION
, &mem
);
190 #ifdef OBSOLETE_KVM_IMPL
191 static void kvm_reset_vcpu(void *opaque
)
193 CPUState
*env
= opaque
;
195 kvm_arch_reset_vcpu(env
);
199 int kvm_irqchip_in_kernel(void)
201 return kvm_state
->irqchip_in_kernel
;
204 int kvm_pit_in_kernel(void)
206 return kvm_state
->pit_in_kernel
;
210 #ifdef OBSOLETE_KVM_IMPL
211 int kvm_init_vcpu(CPUState
*env
)
213 KVMState
*s
= kvm_state
;
217 DPRINTF("kvm_init_vcpu\n");
219 ret
= kvm_vm_ioctl(s
, KVM_CREATE_VCPU
, env
->cpu_index
);
221 DPRINTF("kvm_create_vcpu failed\n");
228 mmap_size
= kvm_ioctl(s
, KVM_GET_VCPU_MMAP_SIZE
, 0);
231 DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n");
235 env
->kvm_run
= mmap(NULL
, mmap_size
, PROT_READ
| PROT_WRITE
, MAP_SHARED
,
237 if (env
->kvm_run
== MAP_FAILED
) {
239 DPRINTF("mmap'ing vcpu state failed\n");
243 if (s
->coalesced_mmio
&& !s
->coalesced_mmio_ring
) {
244 s
->coalesced_mmio_ring
=
245 (void *)env
->kvm_run
+ s
->coalesced_mmio
* PAGE_SIZE
;
248 ret
= kvm_arch_init_vcpu(env
);
250 qemu_register_reset(kvm_reset_vcpu
, env
);
251 kvm_arch_reset_vcpu(env
);
259 * dirty pages logging control
261 static int kvm_dirty_pages_log_change(target_phys_addr_t phys_addr
,
262 ram_addr_t size
, int flags
, int mask
)
264 KVMState
*s
= kvm_state
;
265 KVMSlot
*mem
= kvm_lookup_matching_slot(s
, phys_addr
, phys_addr
+ size
);
269 fprintf(stderr
, "BUG: %s: invalid parameters " TARGET_FMT_plx
"-"
270 TARGET_FMT_plx
"\n", __func__
, phys_addr
,
271 (target_phys_addr_t
)(phys_addr
+ size
- 1));
275 old_flags
= mem
->flags
;
277 flags
= (mem
->flags
& ~mask
) | flags
;
280 /* If nothing changed effectively, no need to issue ioctl */
281 if (s
->migration_log
) {
282 flags
|= KVM_MEM_LOG_DIRTY_PAGES
;
284 if (flags
== old_flags
) {
288 return kvm_set_user_memory_region(s
, mem
);
291 int kvm_log_start(target_phys_addr_t phys_addr
, ram_addr_t size
)
293 return kvm_dirty_pages_log_change(phys_addr
, size
, KVM_MEM_LOG_DIRTY_PAGES
,
294 KVM_MEM_LOG_DIRTY_PAGES
);
297 int kvm_log_stop(target_phys_addr_t phys_addr
, ram_addr_t size
)
299 return kvm_dirty_pages_log_change(phys_addr
, size
, 0,
300 KVM_MEM_LOG_DIRTY_PAGES
);
303 static int kvm_set_migration_log(int enable
)
305 KVMState
*s
= kvm_state
;
309 s
->migration_log
= enable
;
311 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
314 if (!mem
->memory_size
) {
317 if (!!(mem
->flags
& KVM_MEM_LOG_DIRTY_PAGES
) == enable
) {
320 err
= kvm_set_user_memory_region(s
, mem
);
328 /* get kvm's dirty pages bitmap and update qemu's */
329 static int kvm_get_dirty_pages_log_range(unsigned long start_addr
,
330 unsigned long *bitmap
,
331 unsigned long offset
,
332 unsigned long mem_size
)
335 unsigned long page_number
, addr
, addr1
, c
;
337 unsigned int len
= ((mem_size
/ TARGET_PAGE_SIZE
) + HOST_LONG_BITS
- 1) /
341 * bitmap-traveling is faster than memory-traveling (for addr...)
342 * especially when most of the memory is not dirty.
344 for (i
= 0; i
< len
; i
++) {
345 if (bitmap
[i
] != 0) {
346 c
= leul_to_cpu(bitmap
[i
]);
350 page_number
= i
* HOST_LONG_BITS
+ j
;
351 addr1
= page_number
* TARGET_PAGE_SIZE
;
352 addr
= offset
+ addr1
;
353 ram_addr
= cpu_get_physical_page_desc(addr
);
354 cpu_physical_memory_set_dirty(ram_addr
);
361 #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1))
364 * kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space
365 * This function updates qemu's dirty bitmap using cpu_physical_memory_set_dirty().
366 * This means all bits are set to dirty.
368 * @start_add: start of logged region.
369 * @end_addr: end of logged region.
371 static int kvm_physical_sync_dirty_bitmap(target_phys_addr_t start_addr
,
372 target_phys_addr_t end_addr
)
374 KVMState
*s
= kvm_state
;
375 unsigned long size
, allocated_size
= 0;
380 d
.dirty_bitmap
= NULL
;
381 while (start_addr
< end_addr
) {
382 mem
= kvm_lookup_overlapping_slot(s
, start_addr
, end_addr
);
387 size
= ALIGN(((mem
->memory_size
) >> TARGET_PAGE_BITS
), HOST_LONG_BITS
) / 8;
388 if (!d
.dirty_bitmap
) {
389 d
.dirty_bitmap
= qemu_malloc(size
);
390 } else if (size
> allocated_size
) {
391 d
.dirty_bitmap
= qemu_realloc(d
.dirty_bitmap
, size
);
393 allocated_size
= size
;
394 memset(d
.dirty_bitmap
, 0, allocated_size
);
398 if (kvm_vm_ioctl(s
, KVM_GET_DIRTY_LOG
, &d
) == -1) {
399 DPRINTF("ioctl failed %d\n", errno
);
404 kvm_get_dirty_pages_log_range(mem
->start_addr
, d
.dirty_bitmap
,
405 mem
->start_addr
, mem
->memory_size
);
406 start_addr
= mem
->start_addr
+ mem
->memory_size
;
408 qemu_free(d
.dirty_bitmap
);
413 int kvm_coalesce_mmio_region(target_phys_addr_t start
, ram_addr_t size
)
416 KVMState
*s
= kvm_state
;
418 if (s
->coalesced_mmio
) {
419 struct kvm_coalesced_mmio_zone zone
;
424 ret
= kvm_vm_ioctl(s
, KVM_REGISTER_COALESCED_MMIO
, &zone
);
430 int kvm_uncoalesce_mmio_region(target_phys_addr_t start
, ram_addr_t size
)
433 KVMState
*s
= kvm_state
;
435 if (s
->coalesced_mmio
) {
436 struct kvm_coalesced_mmio_zone zone
;
441 ret
= kvm_vm_ioctl(s
, KVM_UNREGISTER_COALESCED_MMIO
, &zone
);
447 int kvm_check_extension(KVMState
*s
, unsigned int extension
)
451 ret
= kvm_ioctl(s
, KVM_CHECK_EXTENSION
, extension
);
459 static int kvm_check_many_ioeventfds(void)
461 /* Userspace can use ioeventfd for io notification. This requires a host
462 * that supports eventfd(2) and an I/O thread; since eventfd does not
463 * support SIGIO it cannot interrupt the vcpu.
465 * Older kernels have a 6 device limit on the KVM io bus. Find out so we
466 * can avoid creating too many ioeventfds.
468 #if defined(CONFIG_EVENTFD) /* && defined(CONFIG_IOTHREAD) */
471 for (i
= 0; i
< ARRAY_SIZE(ioeventfds
); i
++) {
472 ioeventfds
[i
] = eventfd(0, EFD_CLOEXEC
);
473 if (ioeventfds
[i
] < 0) {
476 ret
= kvm_set_ioeventfd_pio_word(ioeventfds
[i
], 0, i
, true);
478 close(ioeventfds
[i
]);
483 /* Decide whether many devices are supported or not */
484 ret
= i
== ARRAY_SIZE(ioeventfds
);
487 kvm_set_ioeventfd_pio_word(ioeventfds
[i
], 0, i
, false);
488 close(ioeventfds
[i
]);
496 #ifdef OBSOLETE_KVM_IMPL
497 static const KVMCapabilityInfo
*
498 kvm_check_extension_list(KVMState
*s
, const KVMCapabilityInfo
*list
)
501 if (!kvm_check_extension(s
, list
->value
)) {
510 static void kvm_set_phys_mem(target_phys_addr_t start_addr
, ram_addr_t size
,
511 ram_addr_t phys_offset
)
513 KVMState
*s
= kvm_state
;
514 ram_addr_t flags
= phys_offset
& ~TARGET_PAGE_MASK
;
518 /* kvm works in page size chunks, but the function may be called
519 with sub-page size and unaligned start address. */
520 size
= TARGET_PAGE_ALIGN(size
);
521 start_addr
= TARGET_PAGE_ALIGN(start_addr
);
523 /* KVM does not support read-only slots */
524 phys_offset
&= ~IO_MEM_ROM
;
527 mem
= kvm_lookup_overlapping_slot(s
, start_addr
, start_addr
+ size
);
532 if (flags
< IO_MEM_UNASSIGNED
&& start_addr
>= mem
->start_addr
&&
533 (start_addr
+ size
<= mem
->start_addr
+ mem
->memory_size
) &&
534 (phys_offset
- start_addr
== mem
->phys_offset
- mem
->start_addr
)) {
535 /* The new slot fits into the existing one and comes with
536 * identical parameters - nothing to be done. */
542 /* unregister the overlapping slot */
543 mem
->memory_size
= 0;
544 err
= kvm_set_user_memory_region(s
, mem
);
546 fprintf(stderr
, "%s: error unregistering overlapping slot: %s\n",
547 __func__
, strerror(-err
));
551 /* Workaround for older KVM versions: we can't join slots, even not by
552 * unregistering the previous ones and then registering the larger
553 * slot. We have to maintain the existing fragmentation. Sigh.
555 * This workaround assumes that the new slot starts at the same
556 * address as the first existing one. If not or if some overlapping
557 * slot comes around later, we will fail (not seen in practice so far)
558 * - and actually require a recent KVM version. */
559 if (s
->broken_set_mem_region
&&
560 old
.start_addr
== start_addr
&& old
.memory_size
< size
&&
561 flags
< IO_MEM_UNASSIGNED
) {
562 mem
= kvm_alloc_slot(s
);
563 mem
->memory_size
= old
.memory_size
;
564 mem
->start_addr
= old
.start_addr
;
565 mem
->phys_offset
= old
.phys_offset
;
568 err
= kvm_set_user_memory_region(s
, mem
);
570 fprintf(stderr
, "%s: error updating slot: %s\n", __func__
,
575 start_addr
+= old
.memory_size
;
576 phys_offset
+= old
.memory_size
;
577 size
-= old
.memory_size
;
581 /* register prefix slot */
582 if (old
.start_addr
< start_addr
) {
583 mem
= kvm_alloc_slot(s
);
584 mem
->memory_size
= start_addr
- old
.start_addr
;
585 mem
->start_addr
= old
.start_addr
;
586 mem
->phys_offset
= old
.phys_offset
;
589 err
= kvm_set_user_memory_region(s
, mem
);
591 fprintf(stderr
, "%s: error registering prefix slot: %s\n",
592 __func__
, strerror(-err
));
597 /* register suffix slot */
598 if (old
.start_addr
+ old
.memory_size
> start_addr
+ size
) {
599 ram_addr_t size_delta
;
601 mem
= kvm_alloc_slot(s
);
602 mem
->start_addr
= start_addr
+ size
;
603 size_delta
= mem
->start_addr
- old
.start_addr
;
604 mem
->memory_size
= old
.memory_size
- size_delta
;
605 mem
->phys_offset
= old
.phys_offset
+ size_delta
;
608 err
= kvm_set_user_memory_region(s
, mem
);
610 fprintf(stderr
, "%s: error registering suffix slot: %s\n",
611 __func__
, strerror(-err
));
617 /* in case the KVM bug workaround already "consumed" the new slot */
621 /* KVM does not need to know about this memory */
622 if (flags
>= IO_MEM_UNASSIGNED
) {
625 mem
= kvm_alloc_slot(s
);
626 mem
->memory_size
= size
;
627 mem
->start_addr
= start_addr
;
628 mem
->phys_offset
= phys_offset
;
631 err
= kvm_set_user_memory_region(s
, mem
);
633 fprintf(stderr
, "%s: error registering slot: %s\n", __func__
,
639 static void kvm_client_set_memory(struct CPUPhysMemoryClient
*client
,
640 target_phys_addr_t start_addr
,
641 ram_addr_t size
, ram_addr_t phys_offset
)
643 kvm_set_phys_mem(start_addr
, size
, phys_offset
);
646 static int kvm_client_sync_dirty_bitmap(struct CPUPhysMemoryClient
*client
,
647 target_phys_addr_t start_addr
,
648 target_phys_addr_t end_addr
)
650 return kvm_physical_sync_dirty_bitmap(start_addr
, end_addr
);
653 static int kvm_client_migration_log(struct CPUPhysMemoryClient
*client
,
656 return kvm_set_migration_log(enable
);
659 static CPUPhysMemoryClient kvm_cpu_phys_memory_client
= {
660 .set_memory
= kvm_client_set_memory
,
661 .sync_dirty_bitmap
= kvm_client_sync_dirty_bitmap
,
662 .migration_log
= kvm_client_migration_log
,
665 void kvm_cpu_register_phys_memory_client(void)
667 cpu_register_phys_memory_client(&kvm_cpu_phys_memory_client
);
670 #ifdef OBSOLETE_KVM_IMPL
674 static const char upgrade_note
[] =
675 "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n"
676 "(see http://sourceforge.net/projects/kvm).\n";
678 const KVMCapabilityInfo
*missing_cap
;
682 s
= qemu_mallocz(sizeof(KVMState
));
684 #ifdef KVM_CAP_SET_GUEST_DEBUG
685 QTAILQ_INIT(&s
->kvm_sw_breakpoints
);
687 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
688 s
->slots
[i
].slot
= i
;
691 s
->fd
= qemu_open("/dev/kvm", O_RDWR
);
693 fprintf(stderr
, "Could not access KVM kernel module: %m\n");
698 ret
= kvm_ioctl(s
, KVM_GET_API_VERSION
, 0);
699 if (ret
< KVM_API_VERSION
) {
703 fprintf(stderr
, "kvm version too old\n");
707 if (ret
> KVM_API_VERSION
) {
709 fprintf(stderr
, "kvm version not supported\n");
713 s
->vmfd
= kvm_ioctl(s
, KVM_CREATE_VM
, 0);
716 fprintf(stderr
, "Please add the 'switch_amode' kernel parameter to "
717 "your host kernel command line\n");
722 missing_cap
= kvm_check_extension_list(s
, kvm_required_capabilites
);
725 kvm_check_extension_list(s
, kvm_arch_required_capabilities
);
729 fprintf(stderr
, "kvm does not support %s\n%s",
730 missing_cap
->name
, upgrade_note
);
734 s
->coalesced_mmio
= kvm_check_extension(s
, KVM_CAP_COALESCED_MMIO
);
736 s
->broken_set_mem_region
= 1;
737 #ifdef KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
738 ret
= kvm_check_extension(s
, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
);
740 s
->broken_set_mem_region
= 0;
745 #ifdef KVM_CAP_VCPU_EVENTS
746 s
->vcpu_events
= kvm_check_extension(s
, KVM_CAP_VCPU_EVENTS
);
749 s
->robust_singlestep
= 0;
750 #ifdef KVM_CAP_X86_ROBUST_SINGLESTEP
751 s
->robust_singlestep
=
752 kvm_check_extension(s
, KVM_CAP_X86_ROBUST_SINGLESTEP
);
756 #ifdef KVM_CAP_DEBUGREGS
757 s
->debugregs
= kvm_check_extension(s
, KVM_CAP_DEBUGREGS
);
762 s
->xsave
= kvm_check_extension(s
, KVM_CAP_XSAVE
);
767 s
->xcrs
= kvm_check_extension(s
, KVM_CAP_XCRS
);
770 ret
= kvm_arch_init(s
);
776 cpu_register_phys_memory_client(&kvm_cpu_phys_memory_client
);
778 s
->many_ioeventfds
= kvm_check_many_ioeventfds();
797 static int kvm_handle_io(uint16_t port
, void *data
, int direction
, int size
,
803 for (i
= 0; i
< count
; i
++) {
804 if (direction
== KVM_EXIT_IO_IN
) {
807 stb_p(ptr
, cpu_inb(port
));
810 stw_p(ptr
, cpu_inw(port
));
813 stl_p(ptr
, cpu_inl(port
));
819 cpu_outb(port
, ldub_p(ptr
));
822 cpu_outw(port
, lduw_p(ptr
));
825 cpu_outl(port
, ldl_p(ptr
));
836 #ifdef KVM_CAP_INTERNAL_ERROR_DATA
837 static int kvm_handle_internal_error(CPUState
*env
, struct kvm_run
*run
)
839 fprintf(stderr
, "KVM internal error.");
840 if (kvm_check_extension(kvm_state
, KVM_CAP_INTERNAL_ERROR_DATA
)) {
843 fprintf(stderr
, " Suberror: %d\n", run
->internal
.suberror
);
844 for (i
= 0; i
< run
->internal
.ndata
; ++i
) {
845 fprintf(stderr
, "extra data[%d]: %"PRIx64
"\n",
846 i
, (uint64_t)run
->internal
.data
[i
]);
849 fprintf(stderr
, "\n");
851 if (run
->internal
.suberror
== KVM_INTERNAL_ERROR_EMULATION
) {
852 fprintf(stderr
, "emulation failure\n");
853 if (!kvm_arch_stop_on_emulation_error(env
)) {
854 cpu_dump_state(env
, stderr
, fprintf
, CPU_DUMP_CODE
);
858 /* FIXME: Should trigger a qmp message to let management know
859 * something went wrong.
865 void kvm_flush_coalesced_mmio_buffer(void)
867 KVMState
*s
= kvm_state
;
868 if (s
->coalesced_mmio_ring
) {
869 struct kvm_coalesced_mmio_ring
*ring
= s
->coalesced_mmio_ring
;
870 while (ring
->first
!= ring
->last
) {
871 struct kvm_coalesced_mmio
*ent
;
873 ent
= &ring
->coalesced_mmio
[ring
->first
];
875 cpu_physical_memory_write(ent
->phys_addr
, ent
->data
, ent
->len
);
877 ring
->first
= (ring
->first
+ 1) % KVM_COALESCED_MMIO_MAX
;
882 #ifdef OBSOLETE_KVM_IMPL
884 static void do_kvm_cpu_synchronize_state(void *_env
)
886 CPUState
*env
= _env
;
888 if (!env
->kvm_vcpu_dirty
) {
889 kvm_arch_get_registers(env
);
890 env
->kvm_vcpu_dirty
= 1;
894 void kvm_cpu_synchronize_state(CPUState
*env
)
896 if (!env
->kvm_vcpu_dirty
) {
897 run_on_cpu(env
, do_kvm_cpu_synchronize_state
, env
);
901 void kvm_cpu_synchronize_post_reset(CPUState
*env
)
903 kvm_arch_put_registers(env
, KVM_PUT_RESET_STATE
);
904 env
->kvm_vcpu_dirty
= 0;
907 void kvm_cpu_synchronize_post_init(CPUState
*env
)
909 kvm_arch_put_registers(env
, KVM_PUT_FULL_STATE
);
910 env
->kvm_vcpu_dirty
= 0;
913 int kvm_cpu_exec(CPUState
*env
)
915 struct kvm_run
*run
= env
->kvm_run
;
918 DPRINTF("kvm_cpu_exec()\n");
921 #ifndef CONFIG_IOTHREAD
922 if (env
->exit_request
) {
923 DPRINTF("interrupt exit requested\n");
929 if (kvm_arch_process_irqchip_events(env
)) {
934 if (env
->kvm_vcpu_dirty
) {
935 kvm_arch_put_registers(env
, KVM_PUT_RUNTIME_STATE
);
936 env
->kvm_vcpu_dirty
= 0;
939 kvm_arch_pre_run(env
, run
);
940 cpu_single_env
= NULL
;
941 qemu_mutex_unlock_iothread();
942 ret
= kvm_vcpu_ioctl(env
, KVM_RUN
, 0);
943 qemu_mutex_lock_iothread();
944 cpu_single_env
= env
;
945 kvm_arch_post_run(env
, run
);
947 kvm_flush_coalesced_mmio_buffer();
949 if (ret
== -EINTR
|| ret
== -EAGAIN
) {
951 DPRINTF("io window exit\n");
957 DPRINTF("kvm run failed %s\n", strerror(-ret
));
961 ret
= 0; /* exit loop */
962 switch (run
->exit_reason
) {
964 DPRINTF("handle_io\n");
965 ret
= kvm_handle_io(run
->io
.port
,
966 (uint8_t *)run
+ run
->io
.data_offset
,
972 DPRINTF("handle_mmio\n");
973 cpu_physical_memory_rw(run
->mmio
.phys_addr
,
979 case KVM_EXIT_IRQ_WINDOW_OPEN
:
980 DPRINTF("irq_window_open\n");
982 case KVM_EXIT_SHUTDOWN
:
983 DPRINTF("shutdown\n");
984 qemu_system_reset_request();
987 case KVM_EXIT_UNKNOWN
:
988 fprintf(stderr
, "KVM: unknown exit, hardware reason %" PRIx64
"\n",
989 (uint64_t)run
->hw
.hardware_exit_reason
);
992 #ifdef KVM_CAP_INTERNAL_ERROR_DATA
993 case KVM_EXIT_INTERNAL_ERROR
:
994 ret
= kvm_handle_internal_error(env
, run
);
998 DPRINTF("kvm_exit_debug\n");
999 #ifdef KVM_CAP_SET_GUEST_DEBUG
1000 if (kvm_arch_debug(&run
->debug
.arch
)) {
1001 env
->exception_index
= EXCP_DEBUG
;
1004 /* re-enter, this exception was guest-internal */
1006 #endif /* KVM_CAP_SET_GUEST_DEBUG */
1009 DPRINTF("kvm_arch_handle_exit\n");
1010 ret
= kvm_arch_handle_exit(env
, run
);
1016 cpu_dump_state(env
, stderr
, fprintf
, CPU_DUMP_CODE
);
1018 env
->exit_request
= 1;
1020 if (env
->exit_request
) {
1021 env
->exit_request
= 0;
1022 env
->exception_index
= EXCP_INTERRUPT
;
1029 int kvm_ioctl(KVMState
*s
, int type
, ...)
1036 arg
= va_arg(ap
, void *);
1039 ret
= ioctl(s
->fd
, type
, arg
);
1046 int kvm_vm_ioctl(KVMState
*s
, int type
, ...)
1053 arg
= va_arg(ap
, void *);
1056 ret
= ioctl(s
->vmfd
, type
, arg
);
1063 int kvm_vcpu_ioctl(CPUState
*env
, int type
, ...)
1070 arg
= va_arg(ap
, void *);
1073 ret
= ioctl(env
->kvm_fd
, type
, arg
);
1080 int kvm_has_sync_mmu(void)
1082 return kvm_check_extension(kvm_state
, KVM_CAP_SYNC_MMU
);
1085 int kvm_has_vcpu_events(void)
1087 return kvm_state
->vcpu_events
;
1090 int kvm_has_robust_singlestep(void)
1092 return kvm_state
->robust_singlestep
;
1095 int kvm_has_debugregs(void)
1097 return kvm_state
->debugregs
;
1100 int kvm_has_xsave(void)
1102 return kvm_state
->xsave
;
1105 int kvm_has_xcrs(void)
1107 return kvm_state
->xcrs
;
1110 int kvm_has_many_ioeventfds(void)
1112 if (!kvm_enabled()) {
1115 return kvm_state
->many_ioeventfds
;
1118 void kvm_setup_guest_memory(void *start
, size_t size
)
1120 if (!kvm_has_sync_mmu()) {
1121 int ret
= qemu_madvise(start
, size
, QEMU_MADV_DONTFORK
);
1124 perror("qemu_madvise");
1126 "Need MADV_DONTFORK in absence of synchronous KVM MMU\n");
1132 #ifdef KVM_CAP_SET_GUEST_DEBUG
1133 #ifndef OBSOLETE_KVM_IMPL
1134 #define run_on_cpu on_vcpu
1135 #endif /* !OBSOLETE_KVM_IMPL */
1137 struct kvm_sw_breakpoint
*kvm_find_sw_breakpoint(CPUState
*env
,
1140 struct kvm_sw_breakpoint
*bp
;
1142 QTAILQ_FOREACH(bp
, &env
->kvm_state
->kvm_sw_breakpoints
, entry
) {
1150 int kvm_sw_breakpoints_active(CPUState
*env
)
1152 return !QTAILQ_EMPTY(&env
->kvm_state
->kvm_sw_breakpoints
);
1155 struct kvm_set_guest_debug_data
{
1156 struct kvm_guest_debug dbg
;
1161 static void kvm_invoke_set_guest_debug(void *data
)
1163 struct kvm_set_guest_debug_data
*dbg_data
= data
;
1164 CPUState
*env
= dbg_data
->env
;
1166 dbg_data
->err
= kvm_vcpu_ioctl(env
, KVM_SET_GUEST_DEBUG
, &dbg_data
->dbg
);
1169 int kvm_update_guest_debug(CPUState
*env
, unsigned long reinject_trap
)
1171 struct kvm_set_guest_debug_data data
;
1173 data
.dbg
.control
= reinject_trap
;
1175 if (env
->singlestep_enabled
) {
1176 data
.dbg
.control
|= KVM_GUESTDBG_ENABLE
| KVM_GUESTDBG_SINGLESTEP
;
1178 kvm_arch_update_guest_debug(env
, &data
.dbg
);
1181 run_on_cpu(env
, kvm_invoke_set_guest_debug
, &data
);
1185 int kvm_insert_breakpoint(CPUState
*current_env
, target_ulong addr
,
1186 target_ulong len
, int type
)
1188 struct kvm_sw_breakpoint
*bp
;
1192 if (type
== GDB_BREAKPOINT_SW
) {
1193 bp
= kvm_find_sw_breakpoint(current_env
, addr
);
1199 bp
= qemu_malloc(sizeof(struct kvm_sw_breakpoint
));
1206 err
= kvm_arch_insert_sw_breakpoint(current_env
, bp
);
1212 QTAILQ_INSERT_HEAD(¤t_env
->kvm_state
->kvm_sw_breakpoints
,
1215 err
= kvm_arch_insert_hw_breakpoint(addr
, len
, type
);
1221 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
1222 err
= kvm_update_guest_debug(env
, 0);
1230 int kvm_remove_breakpoint(CPUState
*current_env
, target_ulong addr
,
1231 target_ulong len
, int type
)
1233 struct kvm_sw_breakpoint
*bp
;
1237 if (type
== GDB_BREAKPOINT_SW
) {
1238 bp
= kvm_find_sw_breakpoint(current_env
, addr
);
1243 if (bp
->use_count
> 1) {
1248 err
= kvm_arch_remove_sw_breakpoint(current_env
, bp
);
1253 QTAILQ_REMOVE(¤t_env
->kvm_state
->kvm_sw_breakpoints
, bp
, entry
);
1256 err
= kvm_arch_remove_hw_breakpoint(addr
, len
, type
);
1262 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
1263 err
= kvm_update_guest_debug(env
, 0);
1271 void kvm_remove_all_breakpoints(CPUState
*current_env
)
1273 struct kvm_sw_breakpoint
*bp
, *next
;
1274 KVMState
*s
= current_env
->kvm_state
;
1277 QTAILQ_FOREACH_SAFE(bp
, &s
->kvm_sw_breakpoints
, entry
, next
) {
1278 if (kvm_arch_remove_sw_breakpoint(current_env
, bp
) != 0) {
1279 /* Try harder to find a CPU that currently sees the breakpoint. */
1280 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
1281 if (kvm_arch_remove_sw_breakpoint(env
, bp
) == 0) {
1287 kvm_arch_remove_all_hw_breakpoints();
1289 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
1290 kvm_update_guest_debug(env
, 0);
1294 #else /* !KVM_CAP_SET_GUEST_DEBUG */
1296 int kvm_update_guest_debug(CPUState
*env
, unsigned long reinject_trap
)
1301 int kvm_insert_breakpoint(CPUState
*current_env
, target_ulong addr
,
1302 target_ulong len
, int type
)
1307 int kvm_remove_breakpoint(CPUState
*current_env
, target_ulong addr
,
1308 target_ulong len
, int type
)
1313 void kvm_remove_all_breakpoints(CPUState
*current_env
)
1316 #endif /* !KVM_CAP_SET_GUEST_DEBUG */
1318 int kvm_set_signal_mask(CPUState
*env
, const sigset_t
*sigset
)
1320 struct kvm_signal_mask
*sigmask
;
1324 return kvm_vcpu_ioctl(env
, KVM_SET_SIGNAL_MASK
, NULL
);
1327 sigmask
= qemu_malloc(sizeof(*sigmask
) + sizeof(*sigset
));
1330 memcpy(sigmask
->sigset
, sigset
, sizeof(*sigset
));
1331 r
= kvm_vcpu_ioctl(env
, KVM_SET_SIGNAL_MASK
, sigmask
);
1337 int kvm_set_ioeventfd_mmio_long(int fd
, uint32_t addr
, uint32_t val
, bool assign
)
1339 #ifdef KVM_IOEVENTFD
1341 struct kvm_ioeventfd iofd
;
1343 iofd
.datamatch
= val
;
1346 iofd
.flags
= KVM_IOEVENTFD_FLAG_DATAMATCH
;
1349 if (!kvm_enabled()) {
1354 iofd
.flags
|= KVM_IOEVENTFD_FLAG_DEASSIGN
;
1357 ret
= kvm_vm_ioctl(kvm_state
, KVM_IOEVENTFD
, &iofd
);
1369 int kvm_set_ioeventfd_pio_word(int fd
, uint16_t addr
, uint16_t val
, bool assign
)
1371 #ifdef KVM_IOEVENTFD
1372 struct kvm_ioeventfd kick
= {
1376 .flags
= KVM_IOEVENTFD_FLAG_DATAMATCH
| KVM_IOEVENTFD_FLAG_PIO
,
1380 if (!kvm_enabled()) {
1384 kick
.flags
|= KVM_IOEVENTFD_FLAG_DEASSIGN
;
1386 r
= kvm_vm_ioctl(kvm_state
, KVM_IOEVENTFD
, &kick
);
1396 #if defined(KVM_IRQFD)
1397 int kvm_set_irqfd(int gsi
, int fd
, bool assigned
)
1399 struct kvm_irqfd irqfd
= {
1402 .flags
= assigned
? 0 : KVM_IRQFD_FLAG_DEASSIGN
,
1405 if (!kvm_enabled() || !kvm_irqchip_in_kernel())
1408 r
= kvm_vm_ioctl(kvm_state
, KVM_IRQFD
, &irqfd
);
1415 int kvm_on_sigbus_vcpu(CPUState
*env
, int code
, void *addr
)
1417 return kvm_arch_on_sigbus_vcpu(env
, code
, addr
);
1420 int kvm_on_sigbus(int code
, void *addr
)
1422 return kvm_arch_on_sigbus(code
, addr
);
1426 #include "qemu-kvm.c"