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
;
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 if (s
->slots
[i
].memory_size
== 0) {
104 fprintf(stderr
, "%s: no free slot available\n", __func__
);
108 static KVMSlot
*kvm_lookup_matching_slot(KVMState
*s
,
109 target_phys_addr_t start_addr
,
110 target_phys_addr_t end_addr
)
114 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
115 KVMSlot
*mem
= &s
->slots
[i
];
117 if (start_addr
== mem
->start_addr
&&
118 end_addr
== mem
->start_addr
+ mem
->memory_size
) {
127 * Find overlapping slot with lowest start address
129 static KVMSlot
*kvm_lookup_overlapping_slot(KVMState
*s
,
130 target_phys_addr_t start_addr
,
131 target_phys_addr_t end_addr
)
133 KVMSlot
*found
= NULL
;
136 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
137 KVMSlot
*mem
= &s
->slots
[i
];
139 if (mem
->memory_size
== 0 ||
140 (found
&& found
->start_addr
< mem
->start_addr
)) {
144 if (end_addr
> mem
->start_addr
&&
145 start_addr
< mem
->start_addr
+ mem
->memory_size
) {
153 int kvm_physical_memory_addr_from_ram(KVMState
*s
, ram_addr_t ram_addr
,
154 target_phys_addr_t
*phys_addr
)
158 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
159 KVMSlot
*mem
= &s
->slots
[i
];
161 if (ram_addr
>= mem
->phys_offset
&&
162 ram_addr
< mem
->phys_offset
+ mem
->memory_size
) {
163 *phys_addr
= mem
->start_addr
+ (ram_addr
- mem
->phys_offset
);
171 static int kvm_set_user_memory_region(KVMState
*s
, KVMSlot
*slot
)
173 struct kvm_userspace_memory_region mem
;
175 mem
.slot
= slot
->slot
;
176 mem
.guest_phys_addr
= slot
->start_addr
;
177 mem
.memory_size
= slot
->memory_size
;
178 mem
.userspace_addr
= (unsigned long)qemu_safe_ram_ptr(slot
->phys_offset
);
179 mem
.flags
= slot
->flags
;
180 if (s
->migration_log
) {
181 mem
.flags
|= KVM_MEM_LOG_DIRTY_PAGES
;
183 return kvm_vm_ioctl(s
, KVM_SET_USER_MEMORY_REGION
, &mem
);
186 #ifdef OBSOLETE_KVM_IMPL
187 static void kvm_reset_vcpu(void *opaque
)
189 CPUState
*env
= opaque
;
191 kvm_arch_reset_vcpu(env
);
195 int kvm_irqchip_in_kernel(void)
197 return kvm_state
->irqchip_in_kernel
;
200 int kvm_pit_in_kernel(void)
202 return kvm_state
->pit_in_kernel
;
205 #ifdef OBSOLETE_KVM_IMPL
206 int kvm_init_vcpu(CPUState
*env
)
208 KVMState
*s
= kvm_state
;
212 DPRINTF("kvm_init_vcpu\n");
214 ret
= kvm_vm_ioctl(s
, KVM_CREATE_VCPU
, env
->cpu_index
);
216 DPRINTF("kvm_create_vcpu failed\n");
223 mmap_size
= kvm_ioctl(s
, KVM_GET_VCPU_MMAP_SIZE
, 0);
226 DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n");
230 env
->kvm_run
= mmap(NULL
, mmap_size
, PROT_READ
| PROT_WRITE
, MAP_SHARED
,
232 if (env
->kvm_run
== MAP_FAILED
) {
234 DPRINTF("mmap'ing vcpu state failed\n");
238 if (s
->coalesced_mmio
&& !s
->coalesced_mmio_ring
) {
239 s
->coalesced_mmio_ring
=
240 (void *)env
->kvm_run
+ s
->coalesced_mmio
* PAGE_SIZE
;
243 ret
= kvm_arch_init_vcpu(env
);
245 qemu_register_reset(kvm_reset_vcpu
, env
);
246 kvm_arch_reset_vcpu(env
);
254 * dirty pages logging control
256 static int kvm_dirty_pages_log_change(target_phys_addr_t phys_addr
,
257 ram_addr_t size
, int flags
, int mask
)
259 KVMState
*s
= kvm_state
;
260 KVMSlot
*mem
= kvm_lookup_matching_slot(s
, phys_addr
, phys_addr
+ size
);
264 fprintf(stderr
, "BUG: %s: invalid parameters " TARGET_FMT_plx
"-"
265 TARGET_FMT_plx
"\n", __func__
, phys_addr
,
266 (target_phys_addr_t
)(phys_addr
+ size
- 1));
270 old_flags
= mem
->flags
;
272 flags
= (mem
->flags
& ~mask
) | flags
;
275 /* If nothing changed effectively, no need to issue ioctl */
276 if (s
->migration_log
) {
277 flags
|= KVM_MEM_LOG_DIRTY_PAGES
;
279 if (flags
== old_flags
) {
283 return kvm_set_user_memory_region(s
, mem
);
286 static int kvm_log_start(CPUPhysMemoryClient
*client
,
287 target_phys_addr_t phys_addr
, ram_addr_t size
)
289 return kvm_dirty_pages_log_change(phys_addr
, size
, KVM_MEM_LOG_DIRTY_PAGES
,
290 KVM_MEM_LOG_DIRTY_PAGES
);
293 static int kvm_log_stop(CPUPhysMemoryClient
*client
,
294 target_phys_addr_t phys_addr
, ram_addr_t size
)
296 return kvm_dirty_pages_log_change(phys_addr
, size
, 0,
297 KVM_MEM_LOG_DIRTY_PAGES
);
300 static int kvm_set_migration_log(int enable
)
302 KVMState
*s
= kvm_state
;
306 s
->migration_log
= enable
;
308 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
311 if (!mem
->memory_size
) {
314 if (!!(mem
->flags
& KVM_MEM_LOG_DIRTY_PAGES
) == enable
) {
317 err
= kvm_set_user_memory_region(s
, mem
);
325 /* get kvm's dirty pages bitmap and update qemu's */
326 static int kvm_get_dirty_pages_log_range(unsigned long start_addr
,
327 unsigned long *bitmap
,
328 unsigned long offset
,
329 unsigned long mem_size
)
332 unsigned long page_number
, addr
, addr1
, c
;
334 unsigned int len
= ((mem_size
/ TARGET_PAGE_SIZE
) + HOST_LONG_BITS
- 1) /
338 * bitmap-traveling is faster than memory-traveling (for addr...)
339 * especially when most of the memory is not dirty.
341 for (i
= 0; i
< len
; i
++) {
342 if (bitmap
[i
] != 0) {
343 c
= leul_to_cpu(bitmap
[i
]);
347 page_number
= i
* HOST_LONG_BITS
+ j
;
348 addr1
= page_number
* TARGET_PAGE_SIZE
;
349 addr
= offset
+ addr1
;
350 ram_addr
= cpu_get_physical_page_desc(addr
);
351 cpu_physical_memory_set_dirty(ram_addr
);
358 #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1))
361 * kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space
362 * This function updates qemu's dirty bitmap using cpu_physical_memory_set_dirty().
363 * This means all bits are set to dirty.
365 * @start_add: start of logged region.
366 * @end_addr: end of logged region.
368 static int kvm_physical_sync_dirty_bitmap(target_phys_addr_t start_addr
,
369 target_phys_addr_t end_addr
)
371 KVMState
*s
= kvm_state
;
372 unsigned long size
, allocated_size
= 0;
377 d
.dirty_bitmap
= NULL
;
378 while (start_addr
< end_addr
) {
379 mem
= kvm_lookup_overlapping_slot(s
, start_addr
, end_addr
);
384 size
= ALIGN(((mem
->memory_size
) >> TARGET_PAGE_BITS
), HOST_LONG_BITS
) / 8;
385 if (!d
.dirty_bitmap
) {
386 d
.dirty_bitmap
= qemu_malloc(size
);
387 } else if (size
> allocated_size
) {
388 d
.dirty_bitmap
= qemu_realloc(d
.dirty_bitmap
, size
);
390 allocated_size
= size
;
391 memset(d
.dirty_bitmap
, 0, allocated_size
);
395 if (kvm_vm_ioctl(s
, KVM_GET_DIRTY_LOG
, &d
) == -1) {
396 DPRINTF("ioctl failed %d\n", errno
);
401 kvm_get_dirty_pages_log_range(mem
->start_addr
, d
.dirty_bitmap
,
402 mem
->start_addr
, mem
->memory_size
);
403 start_addr
= mem
->start_addr
+ mem
->memory_size
;
405 qemu_free(d
.dirty_bitmap
);
410 int kvm_coalesce_mmio_region(target_phys_addr_t start
, ram_addr_t size
)
413 KVMState
*s
= kvm_state
;
415 if (s
->coalesced_mmio
) {
416 struct kvm_coalesced_mmio_zone zone
;
421 ret
= kvm_vm_ioctl(s
, KVM_REGISTER_COALESCED_MMIO
, &zone
);
427 int kvm_uncoalesce_mmio_region(target_phys_addr_t start
, ram_addr_t size
)
430 KVMState
*s
= kvm_state
;
432 if (s
->coalesced_mmio
) {
433 struct kvm_coalesced_mmio_zone zone
;
438 ret
= kvm_vm_ioctl(s
, KVM_UNREGISTER_COALESCED_MMIO
, &zone
);
444 int kvm_check_extension(KVMState
*s
, unsigned int extension
)
448 ret
= kvm_ioctl(s
, KVM_CHECK_EXTENSION
, extension
);
456 static int kvm_check_many_ioeventfds(void)
458 /* Userspace can use ioeventfd for io notification. This requires a host
459 * that supports eventfd(2) and an I/O thread; since eventfd does not
460 * support SIGIO it cannot interrupt the vcpu.
462 * Older kernels have a 6 device limit on the KVM io bus. Find out so we
463 * can avoid creating too many ioeventfds.
465 #if defined(CONFIG_EVENTFD) /* && defined(CONFIG_IOTHREAD) */
468 for (i
= 0; i
< ARRAY_SIZE(ioeventfds
); i
++) {
469 ioeventfds
[i
] = eventfd(0, EFD_CLOEXEC
);
470 if (ioeventfds
[i
] < 0) {
473 ret
= kvm_set_ioeventfd_pio_word(ioeventfds
[i
], 0, i
, true);
475 close(ioeventfds
[i
]);
480 /* Decide whether many devices are supported or not */
481 ret
= i
== ARRAY_SIZE(ioeventfds
);
484 kvm_set_ioeventfd_pio_word(ioeventfds
[i
], 0, i
, false);
485 close(ioeventfds
[i
]);
493 #ifdef OBSOLETE_KVM_IMPL
494 static const KVMCapabilityInfo
*
495 kvm_check_extension_list(KVMState
*s
, const KVMCapabilityInfo
*list
)
498 if (!kvm_check_extension(s
, list
->value
)) {
507 static void kvm_set_phys_mem(target_phys_addr_t start_addr
, ram_addr_t size
,
508 ram_addr_t phys_offset
)
510 KVMState
*s
= kvm_state
;
511 ram_addr_t flags
= phys_offset
& ~TARGET_PAGE_MASK
;
515 /* kvm works in page size chunks, but the function may be called
516 with sub-page size and unaligned start address. */
517 size
= TARGET_PAGE_ALIGN(size
);
518 start_addr
= TARGET_PAGE_ALIGN(start_addr
);
520 /* KVM does not support read-only slots */
521 phys_offset
&= ~IO_MEM_ROM
;
524 mem
= kvm_lookup_overlapping_slot(s
, start_addr
, start_addr
+ size
);
529 if (flags
< IO_MEM_UNASSIGNED
&& start_addr
>= mem
->start_addr
&&
530 (start_addr
+ size
<= mem
->start_addr
+ mem
->memory_size
) &&
531 (phys_offset
- start_addr
== mem
->phys_offset
- mem
->start_addr
)) {
532 /* The new slot fits into the existing one and comes with
533 * identical parameters - nothing to be done. */
539 /* unregister the overlapping slot */
540 mem
->memory_size
= 0;
541 err
= kvm_set_user_memory_region(s
, mem
);
543 fprintf(stderr
, "%s: error unregistering overlapping slot: %s\n",
544 __func__
, strerror(-err
));
548 /* Workaround for older KVM versions: we can't join slots, even not by
549 * unregistering the previous ones and then registering the larger
550 * slot. We have to maintain the existing fragmentation. Sigh.
552 * This workaround assumes that the new slot starts at the same
553 * address as the first existing one. If not or if some overlapping
554 * slot comes around later, we will fail (not seen in practice so far)
555 * - and actually require a recent KVM version. */
556 if (s
->broken_set_mem_region
&&
557 old
.start_addr
== start_addr
&& old
.memory_size
< size
&&
558 flags
< IO_MEM_UNASSIGNED
) {
559 mem
= kvm_alloc_slot(s
);
560 mem
->memory_size
= old
.memory_size
;
561 mem
->start_addr
= old
.start_addr
;
562 mem
->phys_offset
= old
.phys_offset
;
565 err
= kvm_set_user_memory_region(s
, mem
);
567 fprintf(stderr
, "%s: error updating slot: %s\n", __func__
,
572 start_addr
+= old
.memory_size
;
573 phys_offset
+= old
.memory_size
;
574 size
-= old
.memory_size
;
578 /* register prefix slot */
579 if (old
.start_addr
< start_addr
) {
580 mem
= kvm_alloc_slot(s
);
581 mem
->memory_size
= start_addr
- old
.start_addr
;
582 mem
->start_addr
= old
.start_addr
;
583 mem
->phys_offset
= old
.phys_offset
;
586 err
= kvm_set_user_memory_region(s
, mem
);
588 fprintf(stderr
, "%s: error registering prefix slot: %s\n",
589 __func__
, strerror(-err
));
594 /* register suffix slot */
595 if (old
.start_addr
+ old
.memory_size
> start_addr
+ size
) {
596 ram_addr_t size_delta
;
598 mem
= kvm_alloc_slot(s
);
599 mem
->start_addr
= start_addr
+ size
;
600 size_delta
= mem
->start_addr
- old
.start_addr
;
601 mem
->memory_size
= old
.memory_size
- size_delta
;
602 mem
->phys_offset
= old
.phys_offset
+ size_delta
;
605 err
= kvm_set_user_memory_region(s
, mem
);
607 fprintf(stderr
, "%s: error registering suffix slot: %s\n",
608 __func__
, strerror(-err
));
614 /* in case the KVM bug workaround already "consumed" the new slot */
618 /* KVM does not need to know about this memory */
619 if (flags
>= IO_MEM_UNASSIGNED
) {
622 mem
= kvm_alloc_slot(s
);
623 mem
->memory_size
= size
;
624 mem
->start_addr
= start_addr
;
625 mem
->phys_offset
= phys_offset
;
628 err
= kvm_set_user_memory_region(s
, mem
);
630 fprintf(stderr
, "%s: error registering slot: %s\n", __func__
,
636 static void kvm_client_set_memory(struct CPUPhysMemoryClient
*client
,
637 target_phys_addr_t start_addr
,
638 ram_addr_t size
, ram_addr_t phys_offset
)
640 kvm_set_phys_mem(start_addr
, size
, phys_offset
);
643 static int kvm_client_sync_dirty_bitmap(struct CPUPhysMemoryClient
*client
,
644 target_phys_addr_t start_addr
,
645 target_phys_addr_t end_addr
)
647 return kvm_physical_sync_dirty_bitmap(start_addr
, end_addr
);
650 static int kvm_client_migration_log(struct CPUPhysMemoryClient
*client
,
653 return kvm_set_migration_log(enable
);
656 static CPUPhysMemoryClient kvm_cpu_phys_memory_client
= {
657 .set_memory
= kvm_client_set_memory
,
658 .sync_dirty_bitmap
= kvm_client_sync_dirty_bitmap
,
659 .migration_log
= kvm_client_migration_log
,
660 .log_start
= kvm_log_start
,
661 .log_stop
= kvm_log_stop
,
664 void kvm_cpu_register_phys_memory_client(void)
666 cpu_register_phys_memory_client(&kvm_cpu_phys_memory_client
);
669 #ifdef OBSOLETE_KVM_IMPL
673 static const char upgrade_note
[] =
674 "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n"
675 "(see http://sourceforge.net/projects/kvm).\n";
677 const KVMCapabilityInfo
*missing_cap
;
681 s
= qemu_mallocz(sizeof(KVMState
));
683 #ifdef KVM_CAP_SET_GUEST_DEBUG
684 QTAILQ_INIT(&s
->kvm_sw_breakpoints
);
686 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
687 s
->slots
[i
].slot
= i
;
690 s
->fd
= qemu_open("/dev/kvm", O_RDWR
);
692 fprintf(stderr
, "Could not access KVM kernel module: %m\n");
697 ret
= kvm_ioctl(s
, KVM_GET_API_VERSION
, 0);
698 if (ret
< KVM_API_VERSION
) {
702 fprintf(stderr
, "kvm version too old\n");
706 if (ret
> KVM_API_VERSION
) {
708 fprintf(stderr
, "kvm version not supported\n");
712 s
->vmfd
= kvm_ioctl(s
, KVM_CREATE_VM
, 0);
715 fprintf(stderr
, "Please add the 'switch_amode' kernel parameter to "
716 "your host kernel command line\n");
721 missing_cap
= kvm_check_extension_list(s
, kvm_required_capabilites
);
724 kvm_check_extension_list(s
, kvm_arch_required_capabilities
);
728 fprintf(stderr
, "kvm does not support %s\n%s",
729 missing_cap
->name
, upgrade_note
);
733 s
->coalesced_mmio
= kvm_check_extension(s
, KVM_CAP_COALESCED_MMIO
);
735 s
->broken_set_mem_region
= 1;
736 #ifdef KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
737 ret
= kvm_check_extension(s
, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
);
739 s
->broken_set_mem_region
= 0;
744 #ifdef KVM_CAP_VCPU_EVENTS
745 s
->vcpu_events
= kvm_check_extension(s
, KVM_CAP_VCPU_EVENTS
);
748 s
->robust_singlestep
= 0;
749 #ifdef KVM_CAP_X86_ROBUST_SINGLESTEP
750 s
->robust_singlestep
=
751 kvm_check_extension(s
, KVM_CAP_X86_ROBUST_SINGLESTEP
);
755 #ifdef KVM_CAP_DEBUGREGS
756 s
->debugregs
= kvm_check_extension(s
, KVM_CAP_DEBUGREGS
);
761 s
->xsave
= kvm_check_extension(s
, KVM_CAP_XSAVE
);
766 s
->xcrs
= kvm_check_extension(s
, KVM_CAP_XCRS
);
769 ret
= kvm_arch_init(s
);
775 cpu_register_phys_memory_client(&kvm_cpu_phys_memory_client
);
777 s
->many_ioeventfds
= kvm_check_many_ioeventfds();
796 static void kvm_handle_io(uint16_t port
, void *data
, int direction
, int size
,
802 for (i
= 0; i
< count
; i
++) {
803 if (direction
== KVM_EXIT_IO_IN
) {
806 stb_p(ptr
, cpu_inb(port
));
809 stw_p(ptr
, cpu_inw(port
));
812 stl_p(ptr
, cpu_inl(port
));
818 cpu_outb(port
, ldub_p(ptr
));
821 cpu_outw(port
, lduw_p(ptr
));
824 cpu_outl(port
, ldl_p(ptr
));
833 #ifdef KVM_CAP_INTERNAL_ERROR_DATA
834 static int kvm_handle_internal_error(CPUState
*env
, struct kvm_run
*run
)
836 fprintf(stderr
, "KVM internal error.");
837 if (kvm_check_extension(kvm_state
, KVM_CAP_INTERNAL_ERROR_DATA
)) {
840 fprintf(stderr
, " Suberror: %d\n", run
->internal
.suberror
);
841 for (i
= 0; i
< run
->internal
.ndata
; ++i
) {
842 fprintf(stderr
, "extra data[%d]: %"PRIx64
"\n",
843 i
, (uint64_t)run
->internal
.data
[i
]);
846 fprintf(stderr
, "\n");
848 if (run
->internal
.suberror
== KVM_INTERNAL_ERROR_EMULATION
) {
849 fprintf(stderr
, "emulation failure\n");
850 if (!kvm_arch_stop_on_emulation_error(env
)) {
851 cpu_dump_state(env
, stderr
, fprintf
, CPU_DUMP_CODE
);
855 /* FIXME: Should trigger a qmp message to let management know
856 * something went wrong.
862 void kvm_flush_coalesced_mmio_buffer(void)
864 KVMState
*s
= kvm_state
;
865 if (s
->coalesced_mmio_ring
) {
866 struct kvm_coalesced_mmio_ring
*ring
= s
->coalesced_mmio_ring
;
867 while (ring
->first
!= ring
->last
) {
868 struct kvm_coalesced_mmio
*ent
;
870 ent
= &ring
->coalesced_mmio
[ring
->first
];
872 cpu_physical_memory_write(ent
->phys_addr
, ent
->data
, ent
->len
);
874 ring
->first
= (ring
->first
+ 1) % KVM_COALESCED_MMIO_MAX
;
879 #ifdef OBSOLETE_KVM_IMPL
881 static void do_kvm_cpu_synchronize_state(void *_env
)
883 CPUState
*env
= _env
;
885 if (!env
->kvm_vcpu_dirty
) {
886 kvm_arch_get_registers(env
);
887 env
->kvm_vcpu_dirty
= 1;
891 void kvm_cpu_synchronize_state(CPUState
*env
)
893 if (!env
->kvm_vcpu_dirty
) {
894 run_on_cpu(env
, do_kvm_cpu_synchronize_state
, env
);
898 void kvm_cpu_synchronize_post_reset(CPUState
*env
)
900 kvm_arch_put_registers(env
, KVM_PUT_RESET_STATE
);
901 env
->kvm_vcpu_dirty
= 0;
904 void kvm_cpu_synchronize_post_init(CPUState
*env
)
906 kvm_arch_put_registers(env
, KVM_PUT_FULL_STATE
);
907 env
->kvm_vcpu_dirty
= 0;
910 int kvm_cpu_exec(CPUState
*env
)
912 struct kvm_run
*run
= env
->kvm_run
;
915 DPRINTF("kvm_cpu_exec()\n");
917 if (kvm_arch_process_irqchip_events(env
)) {
918 env
->exit_request
= 0;
922 cpu_single_env
= env
;
925 if (env
->kvm_vcpu_dirty
) {
926 kvm_arch_put_registers(env
, KVM_PUT_RUNTIME_STATE
);
927 env
->kvm_vcpu_dirty
= 0;
930 kvm_arch_pre_run(env
, run
);
931 if (env
->exit_request
) {
932 DPRINTF("interrupt exit requested\n");
934 * KVM requires us to reenter the kernel after IO exits to complete
935 * instruction emulation. This self-signal will ensure that we
938 qemu_cpu_kick_self();
940 cpu_single_env
= NULL
;
941 qemu_mutex_unlock_iothread();
943 ret
= kvm_vcpu_ioctl(env
, KVM_RUN
, 0);
945 qemu_mutex_lock_iothread();
946 cpu_single_env
= env
;
947 kvm_arch_post_run(env
, run
);
949 kvm_flush_coalesced_mmio_buffer();
951 if (ret
== -EINTR
|| ret
== -EAGAIN
) {
952 DPRINTF("io window exit\n");
958 DPRINTF("kvm run failed %s\n", strerror(-ret
));
962 ret
= 0; /* exit loop */
963 switch (run
->exit_reason
) {
965 DPRINTF("handle_io\n");
966 kvm_handle_io(run
->io
.port
,
967 (uint8_t *)run
+ run
->io
.data_offset
,
974 DPRINTF("handle_mmio\n");
975 cpu_physical_memory_rw(run
->mmio
.phys_addr
,
981 case KVM_EXIT_IRQ_WINDOW_OPEN
:
982 DPRINTF("irq_window_open\n");
984 case KVM_EXIT_SHUTDOWN
:
985 DPRINTF("shutdown\n");
986 qemu_system_reset_request();
988 case KVM_EXIT_UNKNOWN
:
989 fprintf(stderr
, "KVM: unknown exit, hardware reason %" PRIx64
"\n",
990 (uint64_t)run
->hw
.hardware_exit_reason
);
993 #ifdef KVM_CAP_INTERNAL_ERROR_DATA
994 case KVM_EXIT_INTERNAL_ERROR
:
995 ret
= kvm_handle_internal_error(env
, run
);
999 DPRINTF("kvm_exit_debug\n");
1000 #ifdef KVM_CAP_SET_GUEST_DEBUG
1001 if (kvm_arch_debug(&run
->debug
.arch
)) {
1005 /* re-enter, this exception was guest-internal */
1007 #endif /* KVM_CAP_SET_GUEST_DEBUG */
1010 DPRINTF("kvm_arch_handle_exit\n");
1011 ret
= kvm_arch_handle_exit(env
, run
);
1017 cpu_dump_state(env
, stderr
, fprintf
, CPU_DUMP_CODE
);
1018 vm_stop(VMSTOP_PANIC
);
1020 ret
= EXCP_INTERRUPT
;
1023 env
->exit_request
= 0;
1024 cpu_single_env
= NULL
;
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"