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
37 #define DPRINTF(fmt, ...) \
38 do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
40 #define DPRINTF(fmt, ...) \
44 #ifdef OBSOLETE_KVM_IMPL
46 typedef struct KVMSlot
48 target_phys_addr_t start_addr
;
49 ram_addr_t memory_size
;
50 ram_addr_t phys_offset
;
55 typedef struct kvm_dirty_log KVMDirtyLog
;
63 #ifdef KVM_CAP_COALESCED_MMIO
64 struct kvm_coalesced_mmio_ring
*coalesced_mmio_ring
;
66 int broken_set_mem_region
;
69 int robust_singlestep
;
71 #ifdef KVM_CAP_SET_GUEST_DEBUG
72 struct kvm_sw_breakpoint_head kvm_sw_breakpoints
;
74 int irqchip_in_kernel
;
79 static KVMState
*kvm_state
;
83 static KVMSlot
*kvm_alloc_slot(KVMState
*s
)
87 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
88 /* KVM private memory slots */
91 if (s
->slots
[i
].memory_size
== 0)
95 fprintf(stderr
, "%s: no free slot available\n", __func__
);
99 static KVMSlot
*kvm_lookup_matching_slot(KVMState
*s
,
100 target_phys_addr_t start_addr
,
101 target_phys_addr_t end_addr
)
105 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
106 KVMSlot
*mem
= &s
->slots
[i
];
108 if (start_addr
== mem
->start_addr
&&
109 end_addr
== mem
->start_addr
+ mem
->memory_size
) {
118 * Find overlapping slot with lowest start address
120 static KVMSlot
*kvm_lookup_overlapping_slot(KVMState
*s
,
121 target_phys_addr_t start_addr
,
122 target_phys_addr_t end_addr
)
124 KVMSlot
*found
= NULL
;
127 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
128 KVMSlot
*mem
= &s
->slots
[i
];
130 if (mem
->memory_size
== 0 ||
131 (found
&& found
->start_addr
< mem
->start_addr
)) {
135 if (end_addr
> mem
->start_addr
&&
136 start_addr
< mem
->start_addr
+ mem
->memory_size
) {
144 static int kvm_set_user_memory_region(KVMState
*s
, KVMSlot
*slot
)
146 struct kvm_userspace_memory_region mem
;
148 mem
.slot
= slot
->slot
;
149 mem
.guest_phys_addr
= slot
->start_addr
;
150 mem
.memory_size
= slot
->memory_size
;
151 mem
.userspace_addr
= (unsigned long)qemu_get_ram_ptr(slot
->phys_offset
);
152 mem
.flags
= slot
->flags
;
153 if (s
->migration_log
) {
154 mem
.flags
|= KVM_MEM_LOG_DIRTY_PAGES
;
156 return kvm_vm_ioctl(s
, KVM_SET_USER_MEMORY_REGION
, &mem
);
159 #ifdef OBSOLETE_KVM_IMPL
160 static void kvm_reset_vcpu(void *opaque
)
162 CPUState
*env
= opaque
;
164 kvm_arch_reset_vcpu(env
);
168 int kvm_irqchip_in_kernel(void)
170 return kvm_state
->irqchip_in_kernel
;
173 int kvm_pit_in_kernel(void)
175 return kvm_state
->pit_in_kernel
;
179 #ifdef OBSOLETE_KVM_IMPL
180 int kvm_init_vcpu(CPUState
*env
)
182 KVMState
*s
= kvm_state
;
186 DPRINTF("kvm_init_vcpu\n");
188 ret
= kvm_vm_ioctl(s
, KVM_CREATE_VCPU
, env
->cpu_index
);
190 DPRINTF("kvm_create_vcpu failed\n");
197 mmap_size
= kvm_ioctl(s
, KVM_GET_VCPU_MMAP_SIZE
, 0);
199 DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n");
203 env
->kvm_run
= mmap(NULL
, mmap_size
, PROT_READ
| PROT_WRITE
, MAP_SHARED
,
205 if (env
->kvm_run
== MAP_FAILED
) {
207 DPRINTF("mmap'ing vcpu state failed\n");
211 #ifdef KVM_CAP_COALESCED_MMIO
212 if (s
->coalesced_mmio
&& !s
->coalesced_mmio_ring
)
213 s
->coalesced_mmio_ring
= (void *) env
->kvm_run
+
214 s
->coalesced_mmio
* PAGE_SIZE
;
217 ret
= kvm_arch_init_vcpu(env
);
219 qemu_register_reset(kvm_reset_vcpu
, env
);
220 kvm_arch_reset_vcpu(env
);
228 * dirty pages logging control
230 static int kvm_dirty_pages_log_change(target_phys_addr_t phys_addr
,
231 ram_addr_t size
, int flags
, int mask
)
233 KVMState
*s
= kvm_state
;
234 KVMSlot
*mem
= kvm_lookup_matching_slot(s
, phys_addr
, phys_addr
+ size
);
238 fprintf(stderr
, "BUG: %s: invalid parameters " TARGET_FMT_plx
"-"
239 TARGET_FMT_plx
"\n", __func__
, phys_addr
,
240 (target_phys_addr_t
)(phys_addr
+ size
- 1));
244 old_flags
= mem
->flags
;
246 flags
= (mem
->flags
& ~mask
) | flags
;
249 /* If nothing changed effectively, no need to issue ioctl */
250 if (s
->migration_log
) {
251 flags
|= KVM_MEM_LOG_DIRTY_PAGES
;
253 if (flags
== old_flags
) {
257 return kvm_set_user_memory_region(s
, mem
);
260 int kvm_log_start(target_phys_addr_t phys_addr
, ram_addr_t size
)
262 return kvm_dirty_pages_log_change(phys_addr
, size
,
263 KVM_MEM_LOG_DIRTY_PAGES
,
264 KVM_MEM_LOG_DIRTY_PAGES
);
267 int kvm_log_stop(target_phys_addr_t phys_addr
, ram_addr_t size
)
269 return kvm_dirty_pages_log_change(phys_addr
, size
,
271 KVM_MEM_LOG_DIRTY_PAGES
);
274 static int kvm_set_migration_log(int enable
)
276 KVMState
*s
= kvm_state
;
280 s
->migration_log
= enable
;
282 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
285 if (!mem
->memory_size
) {
288 if (!!(mem
->flags
& KVM_MEM_LOG_DIRTY_PAGES
) == enable
) {
291 err
= kvm_set_user_memory_region(s
, mem
);
299 /* get kvm's dirty pages bitmap and update qemu's */
300 static int kvm_get_dirty_pages_log_range(unsigned long start_addr
,
301 unsigned long *bitmap
,
302 unsigned long offset
,
303 unsigned long mem_size
)
306 unsigned long page_number
, addr
, addr1
, c
;
308 unsigned int len
= ((mem_size
/ TARGET_PAGE_SIZE
) + HOST_LONG_BITS
- 1) /
312 * bitmap-traveling is faster than memory-traveling (for addr...)
313 * especially when most of the memory is not dirty.
315 for (i
= 0; i
< len
; i
++) {
316 if (bitmap
[i
] != 0) {
317 c
= leul_to_cpu(bitmap
[i
]);
321 page_number
= i
* HOST_LONG_BITS
+ j
;
322 addr1
= page_number
* TARGET_PAGE_SIZE
;
323 addr
= offset
+ addr1
;
324 ram_addr
= cpu_get_physical_page_desc(addr
);
325 cpu_physical_memory_set_dirty(ram_addr
);
332 #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1))
335 * kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space
336 * This function updates qemu's dirty bitmap using cpu_physical_memory_set_dirty().
337 * This means all bits are set to dirty.
339 * @start_add: start of logged region.
340 * @end_addr: end of logged region.
342 static int kvm_physical_sync_dirty_bitmap(target_phys_addr_t start_addr
,
343 target_phys_addr_t end_addr
)
345 KVMState
*s
= kvm_state
;
346 unsigned long size
, allocated_size
= 0;
351 d
.dirty_bitmap
= NULL
;
352 while (start_addr
< end_addr
) {
353 mem
= kvm_lookup_overlapping_slot(s
, start_addr
, end_addr
);
358 size
= ALIGN(((mem
->memory_size
) >> TARGET_PAGE_BITS
), HOST_LONG_BITS
) / 8;
359 if (!d
.dirty_bitmap
) {
360 d
.dirty_bitmap
= qemu_malloc(size
);
361 } else if (size
> allocated_size
) {
362 d
.dirty_bitmap
= qemu_realloc(d
.dirty_bitmap
, size
);
364 allocated_size
= size
;
365 memset(d
.dirty_bitmap
, 0, allocated_size
);
369 if (kvm_vm_ioctl(s
, KVM_GET_DIRTY_LOG
, &d
) == -1) {
370 DPRINTF("ioctl failed %d\n", errno
);
375 kvm_get_dirty_pages_log_range(mem
->start_addr
, d
.dirty_bitmap
,
376 mem
->start_addr
, mem
->memory_size
);
377 start_addr
= mem
->start_addr
+ mem
->memory_size
;
379 qemu_free(d
.dirty_bitmap
);
384 int kvm_coalesce_mmio_region(target_phys_addr_t start
, ram_addr_t size
)
387 #ifdef KVM_CAP_COALESCED_MMIO
388 KVMState
*s
= kvm_state
;
390 if (s
->coalesced_mmio
) {
391 struct kvm_coalesced_mmio_zone zone
;
396 ret
= kvm_vm_ioctl(s
, KVM_REGISTER_COALESCED_MMIO
, &zone
);
403 int kvm_uncoalesce_mmio_region(target_phys_addr_t start
, ram_addr_t size
)
406 #ifdef KVM_CAP_COALESCED_MMIO
407 KVMState
*s
= kvm_state
;
409 if (s
->coalesced_mmio
) {
410 struct kvm_coalesced_mmio_zone zone
;
415 ret
= kvm_vm_ioctl(s
, KVM_UNREGISTER_COALESCED_MMIO
, &zone
);
422 int kvm_check_extension(KVMState
*s
, unsigned int extension
)
426 ret
= kvm_ioctl(s
, KVM_CHECK_EXTENSION
, extension
);
434 static void kvm_set_phys_mem(target_phys_addr_t start_addr
,
436 ram_addr_t phys_offset
)
438 KVMState
*s
= kvm_state
;
439 ram_addr_t flags
= phys_offset
& ~TARGET_PAGE_MASK
;
443 /* kvm works in page size chunks, but the function may be called
444 with sub-page size and unaligned start address. */
445 size
= TARGET_PAGE_ALIGN(size
);
446 start_addr
= TARGET_PAGE_ALIGN(start_addr
);
448 /* KVM does not support read-only slots */
449 phys_offset
&= ~IO_MEM_ROM
;
452 mem
= kvm_lookup_overlapping_slot(s
, start_addr
, start_addr
+ size
);
457 if (flags
< IO_MEM_UNASSIGNED
&& start_addr
>= mem
->start_addr
&&
458 (start_addr
+ size
<= mem
->start_addr
+ mem
->memory_size
) &&
459 (phys_offset
- start_addr
== mem
->phys_offset
- mem
->start_addr
)) {
460 /* The new slot fits into the existing one and comes with
461 * identical parameters - nothing to be done. */
467 /* unregister the overlapping slot */
468 mem
->memory_size
= 0;
469 err
= kvm_set_user_memory_region(s
, mem
);
471 fprintf(stderr
, "%s: error unregistering overlapping slot: %s\n",
472 __func__
, strerror(-err
));
476 /* Workaround for older KVM versions: we can't join slots, even not by
477 * unregistering the previous ones and then registering the larger
478 * slot. We have to maintain the existing fragmentation. Sigh.
480 * This workaround assumes that the new slot starts at the same
481 * address as the first existing one. If not or if some overlapping
482 * slot comes around later, we will fail (not seen in practice so far)
483 * - and actually require a recent KVM version. */
484 if (s
->broken_set_mem_region
&&
485 old
.start_addr
== start_addr
&& old
.memory_size
< size
&&
486 flags
< IO_MEM_UNASSIGNED
) {
487 mem
= kvm_alloc_slot(s
);
488 mem
->memory_size
= old
.memory_size
;
489 mem
->start_addr
= old
.start_addr
;
490 mem
->phys_offset
= old
.phys_offset
;
493 err
= kvm_set_user_memory_region(s
, mem
);
495 fprintf(stderr
, "%s: error updating slot: %s\n", __func__
,
500 start_addr
+= old
.memory_size
;
501 phys_offset
+= old
.memory_size
;
502 size
-= old
.memory_size
;
506 /* register prefix slot */
507 if (old
.start_addr
< start_addr
) {
508 mem
= kvm_alloc_slot(s
);
509 mem
->memory_size
= start_addr
- old
.start_addr
;
510 mem
->start_addr
= old
.start_addr
;
511 mem
->phys_offset
= old
.phys_offset
;
514 err
= kvm_set_user_memory_region(s
, mem
);
516 fprintf(stderr
, "%s: error registering prefix slot: %s\n",
517 __func__
, strerror(-err
));
522 /* register suffix slot */
523 if (old
.start_addr
+ old
.memory_size
> start_addr
+ size
) {
524 ram_addr_t size_delta
;
526 mem
= kvm_alloc_slot(s
);
527 mem
->start_addr
= start_addr
+ size
;
528 size_delta
= mem
->start_addr
- old
.start_addr
;
529 mem
->memory_size
= old
.memory_size
- size_delta
;
530 mem
->phys_offset
= old
.phys_offset
+ size_delta
;
533 err
= kvm_set_user_memory_region(s
, mem
);
535 fprintf(stderr
, "%s: error registering suffix slot: %s\n",
536 __func__
, strerror(-err
));
542 /* in case the KVM bug workaround already "consumed" the new slot */
546 /* KVM does not need to know about this memory */
547 if (flags
>= IO_MEM_UNASSIGNED
)
550 mem
= kvm_alloc_slot(s
);
551 mem
->memory_size
= size
;
552 mem
->start_addr
= start_addr
;
553 mem
->phys_offset
= phys_offset
;
556 err
= kvm_set_user_memory_region(s
, mem
);
558 fprintf(stderr
, "%s: error registering slot: %s\n", __func__
,
564 static void kvm_client_set_memory(struct CPUPhysMemoryClient
*client
,
565 target_phys_addr_t start_addr
,
567 ram_addr_t phys_offset
)
569 kvm_set_phys_mem(start_addr
, size
, phys_offset
);
572 static int kvm_client_sync_dirty_bitmap(struct CPUPhysMemoryClient
*client
,
573 target_phys_addr_t start_addr
,
574 target_phys_addr_t end_addr
)
576 return kvm_physical_sync_dirty_bitmap(start_addr
, end_addr
);
579 static int kvm_client_migration_log(struct CPUPhysMemoryClient
*client
,
582 return kvm_set_migration_log(enable
);
585 static CPUPhysMemoryClient kvm_cpu_phys_memory_client
= {
586 .set_memory
= kvm_client_set_memory
,
587 .sync_dirty_bitmap
= kvm_client_sync_dirty_bitmap
,
588 .migration_log
= kvm_client_migration_log
,
592 void kvm_cpu_register_phys_memory_client(void)
594 cpu_register_phys_memory_client(&kvm_cpu_phys_memory_client
);
597 #ifdef OBSOLETE_KVM_IMPL
599 int kvm_init(int smp_cpus
)
601 static const char upgrade_note
[] =
602 "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n"
603 "(see http://sourceforge.net/projects/kvm).\n";
608 s
= qemu_mallocz(sizeof(KVMState
));
610 #ifdef KVM_CAP_SET_GUEST_DEBUG
611 QTAILQ_INIT(&s
->kvm_sw_breakpoints
);
613 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++)
614 s
->slots
[i
].slot
= i
;
617 s
->fd
= qemu_open("/dev/kvm", O_RDWR
);
619 fprintf(stderr
, "Could not access KVM kernel module: %m\n");
624 ret
= kvm_ioctl(s
, KVM_GET_API_VERSION
, 0);
625 if (ret
< KVM_API_VERSION
) {
628 fprintf(stderr
, "kvm version too old\n");
632 if (ret
> KVM_API_VERSION
) {
634 fprintf(stderr
, "kvm version not supported\n");
638 s
->vmfd
= kvm_ioctl(s
, KVM_CREATE_VM
, 0);
641 fprintf(stderr
, "Please add the 'switch_amode' kernel parameter to "
642 "your host kernel command line\n");
647 /* initially, KVM allocated its own memory and we had to jump through
648 * hooks to make phys_ram_base point to this. Modern versions of KVM
649 * just use a user allocated buffer so we can use regular pages
650 * unmodified. Make sure we have a sufficiently modern version of KVM.
652 if (!kvm_check_extension(s
, KVM_CAP_USER_MEMORY
)) {
654 fprintf(stderr
, "kvm does not support KVM_CAP_USER_MEMORY\n%s",
659 /* There was a nasty bug in < kvm-80 that prevents memory slots from being
660 * destroyed properly. Since we rely on this capability, refuse to work
661 * with any kernel without this capability. */
662 if (!kvm_check_extension(s
, KVM_CAP_DESTROY_MEMORY_REGION_WORKS
)) {
666 "KVM kernel module broken (DESTROY_MEMORY_REGION).\n%s",
671 s
->coalesced_mmio
= 0;
672 #ifdef KVM_CAP_COALESCED_MMIO
673 s
->coalesced_mmio
= kvm_check_extension(s
, KVM_CAP_COALESCED_MMIO
);
674 s
->coalesced_mmio_ring
= NULL
;
677 s
->broken_set_mem_region
= 1;
678 #ifdef KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
679 ret
= kvm_ioctl(s
, KVM_CHECK_EXTENSION
, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
);
681 s
->broken_set_mem_region
= 0;
686 #ifdef KVM_CAP_VCPU_EVENTS
687 s
->vcpu_events
= kvm_check_extension(s
, KVM_CAP_VCPU_EVENTS
);
690 s
->robust_singlestep
= 0;
691 #ifdef KVM_CAP_X86_ROBUST_SINGLESTEP
692 s
->robust_singlestep
=
693 kvm_check_extension(s
, KVM_CAP_X86_ROBUST_SINGLESTEP
);
697 #ifdef KVM_CAP_DEBUGREGS
698 s
->debugregs
= kvm_check_extension(s
, KVM_CAP_DEBUGREGS
);
703 s
->xsave
= kvm_check_extension(s
, KVM_CAP_XSAVE
);
708 s
->xcrs
= kvm_check_extension(s
, KVM_CAP_XCRS
);
711 ret
= kvm_arch_init(s
, smp_cpus
);
716 cpu_register_phys_memory_client(&kvm_cpu_phys_memory_client
);
733 static int kvm_handle_io(uint16_t port
, void *data
, int direction
, int size
,
739 for (i
= 0; i
< count
; i
++) {
740 if (direction
== KVM_EXIT_IO_IN
) {
743 stb_p(ptr
, cpu_inb(port
));
746 stw_p(ptr
, cpu_inw(port
));
749 stl_p(ptr
, cpu_inl(port
));
755 cpu_outb(port
, ldub_p(ptr
));
758 cpu_outw(port
, lduw_p(ptr
));
761 cpu_outl(port
, ldl_p(ptr
));
772 #ifdef KVM_CAP_INTERNAL_ERROR_DATA
773 static void kvm_handle_internal_error(CPUState
*env
, struct kvm_run
*run
)
776 if (kvm_check_extension(kvm_state
, KVM_CAP_INTERNAL_ERROR_DATA
)) {
779 fprintf(stderr
, "KVM internal error. Suberror: %d\n",
780 run
->internal
.suberror
);
782 for (i
= 0; i
< run
->internal
.ndata
; ++i
) {
783 fprintf(stderr
, "extra data[%d]: %"PRIx64
"\n",
784 i
, (uint64_t)run
->internal
.data
[i
]);
787 cpu_dump_state(env
, stderr
, fprintf
, 0);
788 if (run
->internal
.suberror
== KVM_INTERNAL_ERROR_EMULATION
) {
789 fprintf(stderr
, "emulation failure\n");
790 if (!kvm_arch_stop_on_emulation_error(env
))
793 /* FIXME: Should trigger a qmp message to let management know
794 * something went wrong.
800 void kvm_flush_coalesced_mmio_buffer(void)
802 #ifdef KVM_CAP_COALESCED_MMIO
803 KVMState
*s
= kvm_state
;
804 if (s
->coalesced_mmio_ring
) {
805 struct kvm_coalesced_mmio_ring
*ring
= s
->coalesced_mmio_ring
;
806 while (ring
->first
!= ring
->last
) {
807 struct kvm_coalesced_mmio
*ent
;
809 ent
= &ring
->coalesced_mmio
[ring
->first
];
811 cpu_physical_memory_write(ent
->phys_addr
, ent
->data
, ent
->len
);
813 ring
->first
= (ring
->first
+ 1) % KVM_COALESCED_MMIO_MAX
;
819 #ifdef OBSOLETE_KVM_IMPL
821 static void do_kvm_cpu_synchronize_state(void *_env
)
823 CPUState
*env
= _env
;
825 if (!env
->kvm_vcpu_dirty
) {
826 kvm_arch_get_registers(env
);
827 env
->kvm_vcpu_dirty
= 1;
831 void kvm_cpu_synchronize_state(CPUState
*env
)
833 if (!env
->kvm_vcpu_dirty
)
834 run_on_cpu(env
, do_kvm_cpu_synchronize_state
, env
);
837 void kvm_cpu_synchronize_post_reset(CPUState
*env
)
839 kvm_arch_put_registers(env
, KVM_PUT_RESET_STATE
);
840 env
->kvm_vcpu_dirty
= 0;
843 void kvm_cpu_synchronize_post_init(CPUState
*env
)
845 kvm_arch_put_registers(env
, KVM_PUT_FULL_STATE
);
846 env
->kvm_vcpu_dirty
= 0;
849 int kvm_cpu_exec(CPUState
*env
)
851 struct kvm_run
*run
= env
->kvm_run
;
854 DPRINTF("kvm_cpu_exec()\n");
857 #ifndef CONFIG_IOTHREAD
858 if (env
->exit_request
) {
859 DPRINTF("interrupt exit requested\n");
865 if (kvm_arch_process_irqchip_events(env
)) {
870 if (env
->kvm_vcpu_dirty
) {
871 kvm_arch_put_registers(env
, KVM_PUT_RUNTIME_STATE
);
872 env
->kvm_vcpu_dirty
= 0;
875 kvm_arch_pre_run(env
, run
);
876 cpu_single_env
= NULL
;
877 qemu_mutex_unlock_iothread();
878 ret
= kvm_vcpu_ioctl(env
, KVM_RUN
, 0);
879 qemu_mutex_lock_iothread();
880 cpu_single_env
= env
;
881 kvm_arch_post_run(env
, run
);
883 if (ret
== -EINTR
|| ret
== -EAGAIN
) {
885 DPRINTF("io window exit\n");
891 DPRINTF("kvm run failed %s\n", strerror(-ret
));
895 kvm_flush_coalesced_mmio_buffer();
897 ret
= 0; /* exit loop */
898 switch (run
->exit_reason
) {
900 DPRINTF("handle_io\n");
901 ret
= kvm_handle_io(run
->io
.port
,
902 (uint8_t *)run
+ run
->io
.data_offset
,
908 DPRINTF("handle_mmio\n");
909 cpu_physical_memory_rw(run
->mmio
.phys_addr
,
915 case KVM_EXIT_IRQ_WINDOW_OPEN
:
916 DPRINTF("irq_window_open\n");
918 case KVM_EXIT_SHUTDOWN
:
919 DPRINTF("shutdown\n");
920 qemu_system_reset_request();
923 case KVM_EXIT_UNKNOWN
:
924 DPRINTF("kvm_exit_unknown\n");
926 case KVM_EXIT_FAIL_ENTRY
:
927 DPRINTF("kvm_exit_fail_entry\n");
929 case KVM_EXIT_EXCEPTION
:
930 DPRINTF("kvm_exit_exception\n");
932 #ifdef KVM_CAP_INTERNAL_ERROR_DATA
933 case KVM_EXIT_INTERNAL_ERROR
:
934 kvm_handle_internal_error(env
, run
);
938 DPRINTF("kvm_exit_debug\n");
939 #ifdef KVM_CAP_SET_GUEST_DEBUG
940 if (kvm_arch_debug(&run
->debug
.arch
)) {
941 env
->exception_index
= EXCP_DEBUG
;
944 /* re-enter, this exception was guest-internal */
946 #endif /* KVM_CAP_SET_GUEST_DEBUG */
949 DPRINTF("kvm_arch_handle_exit\n");
950 ret
= kvm_arch_handle_exit(env
, run
);
955 if (env
->exit_request
) {
956 env
->exit_request
= 0;
957 env
->exception_index
= EXCP_INTERRUPT
;
964 int kvm_ioctl(KVMState
*s
, int type
, ...)
971 arg
= va_arg(ap
, void *);
974 ret
= ioctl(s
->fd
, type
, arg
);
981 int kvm_vm_ioctl(KVMState
*s
, int type
, ...)
988 arg
= va_arg(ap
, void *);
991 ret
= ioctl(s
->vmfd
, type
, arg
);
998 int kvm_vcpu_ioctl(CPUState
*env
, int type
, ...)
1005 arg
= va_arg(ap
, void *);
1008 ret
= ioctl(env
->kvm_fd
, type
, arg
);
1015 int kvm_has_sync_mmu(void)
1017 #ifdef KVM_CAP_SYNC_MMU
1018 KVMState
*s
= kvm_state
;
1020 return kvm_check_extension(s
, KVM_CAP_SYNC_MMU
);
1026 int kvm_has_vcpu_events(void)
1028 return kvm_state
->vcpu_events
;
1031 int kvm_has_robust_singlestep(void)
1033 return kvm_state
->robust_singlestep
;
1036 int kvm_has_debugregs(void)
1038 return kvm_state
->debugregs
;
1041 #ifdef OBSOLETE_KVM_IMPL
1042 int kvm_has_xsave(void)
1044 return kvm_state
->xsave
;
1047 int kvm_has_xcrs(void)
1049 return kvm_state
->xcrs
;
1053 void kvm_setup_guest_memory(void *start
, size_t size
)
1055 if (!kvm_has_sync_mmu()) {
1056 #ifdef MADV_DONTFORK
1057 int ret
= madvise(start
, size
, MADV_DONTFORK
);
1065 "Need MADV_DONTFORK in absence of synchronous KVM MMU\n");
1071 #ifdef KVM_CAP_SET_GUEST_DEBUG
1072 #ifndef OBSOLETE_KVM_IMPL
1073 #define run_on_cpu on_vcpu
1074 static void on_vcpu(CPUState
*env
, void (*func
)(void *data
), void *data
);
1075 #endif /* !OBSOLETE_KVM_IMPL */
1077 struct kvm_sw_breakpoint
*kvm_find_sw_breakpoint(CPUState
*env
,
1080 struct kvm_sw_breakpoint
*bp
;
1082 QTAILQ_FOREACH(bp
, &env
->kvm_state
->kvm_sw_breakpoints
, entry
) {
1089 int kvm_sw_breakpoints_active(CPUState
*env
)
1091 return !QTAILQ_EMPTY(&env
->kvm_state
->kvm_sw_breakpoints
);
1094 struct kvm_set_guest_debug_data
{
1095 struct kvm_guest_debug dbg
;
1100 static void kvm_invoke_set_guest_debug(void *data
)
1102 struct kvm_set_guest_debug_data
*dbg_data
= data
;
1103 CPUState
*env
= dbg_data
->env
;
1105 dbg_data
->err
= kvm_vcpu_ioctl(env
, KVM_SET_GUEST_DEBUG
, &dbg_data
->dbg
);
1108 int kvm_update_guest_debug(CPUState
*env
, unsigned long reinject_trap
)
1110 struct kvm_set_guest_debug_data data
;
1112 data
.dbg
.control
= reinject_trap
;
1114 if (env
->singlestep_enabled
) {
1115 data
.dbg
.control
|= KVM_GUESTDBG_ENABLE
| KVM_GUESTDBG_SINGLESTEP
;
1117 kvm_arch_update_guest_debug(env
, &data
.dbg
);
1120 run_on_cpu(env
, kvm_invoke_set_guest_debug
, &data
);
1124 int kvm_insert_breakpoint(CPUState
*current_env
, target_ulong addr
,
1125 target_ulong len
, int type
)
1127 struct kvm_sw_breakpoint
*bp
;
1131 if (type
== GDB_BREAKPOINT_SW
) {
1132 bp
= kvm_find_sw_breakpoint(current_env
, addr
);
1138 bp
= qemu_malloc(sizeof(struct kvm_sw_breakpoint
));
1144 err
= kvm_arch_insert_sw_breakpoint(current_env
, bp
);
1150 QTAILQ_INSERT_HEAD(¤t_env
->kvm_state
->kvm_sw_breakpoints
,
1153 err
= kvm_arch_insert_hw_breakpoint(addr
, len
, type
);
1158 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
1159 err
= kvm_update_guest_debug(env
, 0);
1166 int kvm_remove_breakpoint(CPUState
*current_env
, target_ulong addr
,
1167 target_ulong len
, int type
)
1169 struct kvm_sw_breakpoint
*bp
;
1173 if (type
== GDB_BREAKPOINT_SW
) {
1174 bp
= kvm_find_sw_breakpoint(current_env
, addr
);
1178 if (bp
->use_count
> 1) {
1183 err
= kvm_arch_remove_sw_breakpoint(current_env
, bp
);
1187 QTAILQ_REMOVE(¤t_env
->kvm_state
->kvm_sw_breakpoints
, bp
, entry
);
1190 err
= kvm_arch_remove_hw_breakpoint(addr
, len
, type
);
1195 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
1196 err
= kvm_update_guest_debug(env
, 0);
1203 void kvm_remove_all_breakpoints(CPUState
*current_env
)
1205 struct kvm_sw_breakpoint
*bp
, *next
;
1206 KVMState
*s
= current_env
->kvm_state
;
1209 QTAILQ_FOREACH_SAFE(bp
, &s
->kvm_sw_breakpoints
, entry
, next
) {
1210 if (kvm_arch_remove_sw_breakpoint(current_env
, bp
) != 0) {
1211 /* Try harder to find a CPU that currently sees the breakpoint. */
1212 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
1213 if (kvm_arch_remove_sw_breakpoint(env
, bp
) == 0)
1218 kvm_arch_remove_all_hw_breakpoints();
1220 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
)
1221 kvm_update_guest_debug(env
, 0);
1224 #else /* !KVM_CAP_SET_GUEST_DEBUG */
1226 int kvm_update_guest_debug(CPUState
*env
, unsigned long reinject_trap
)
1231 int kvm_insert_breakpoint(CPUState
*current_env
, target_ulong addr
,
1232 target_ulong len
, int type
)
1237 int kvm_remove_breakpoint(CPUState
*current_env
, target_ulong addr
,
1238 target_ulong len
, int type
)
1243 void kvm_remove_all_breakpoints(CPUState
*current_env
)
1246 #endif /* !KVM_CAP_SET_GUEST_DEBUG */
1248 int kvm_set_signal_mask(CPUState
*env
, const sigset_t
*sigset
)
1250 struct kvm_signal_mask
*sigmask
;
1254 return kvm_vcpu_ioctl(env
, KVM_SET_SIGNAL_MASK
, NULL
);
1256 sigmask
= qemu_malloc(sizeof(*sigmask
) + sizeof(*sigset
));
1259 memcpy(sigmask
->sigset
, sigset
, sizeof(*sigset
));
1260 r
= kvm_vcpu_ioctl(env
, KVM_SET_SIGNAL_MASK
, sigmask
);
1266 int kvm_set_ioeventfd_mmio_long(int fd
, uint32_t addr
, uint32_t val
, bool assign
)
1268 #ifdef KVM_IOEVENTFD
1270 struct kvm_ioeventfd iofd
;
1272 iofd
.datamatch
= val
;
1275 iofd
.flags
= KVM_IOEVENTFD_FLAG_DATAMATCH
;
1278 if (!kvm_enabled()) {
1283 iofd
.flags
|= KVM_IOEVENTFD_FLAG_DEASSIGN
;
1286 ret
= kvm_vm_ioctl(kvm_state
, KVM_IOEVENTFD
, &iofd
);
1298 int kvm_set_ioeventfd_pio_word(int fd
, uint16_t addr
, uint16_t val
, bool assign
)
1300 #ifdef KVM_IOEVENTFD
1301 struct kvm_ioeventfd kick
= {
1305 .flags
= KVM_IOEVENTFD_FLAG_DATAMATCH
| KVM_IOEVENTFD_FLAG_PIO
,
1312 kick
.flags
|= KVM_IOEVENTFD_FLAG_DEASSIGN
;
1313 r
= kvm_vm_ioctl(kvm_state
, KVM_IOEVENTFD
, &kick
);
1322 #if defined(KVM_IRQFD)
1323 int kvm_set_irqfd(int gsi
, int fd
, bool assigned
)
1325 struct kvm_irqfd irqfd
= {
1328 .flags
= assigned
? 0 : KVM_IRQFD_FLAG_DEASSIGN
,
1331 if (!kvm_enabled() || !kvm_irqchip_in_kernel())
1334 r
= kvm_vm_ioctl(kvm_state
, KVM_IRQFD
, &irqfd
);
1342 #include "qemu-kvm.c"