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"
30 /* KVM uses PAGE_SIZE in it's definition of COALESCED_MMIO_MAX */
31 #define PAGE_SIZE TARGET_PAGE_SIZE
36 #define DPRINTF(fmt, ...) \
37 do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
39 #define DPRINTF(fmt, ...) \
43 typedef struct KVMSlot
45 target_phys_addr_t start_addr
;
46 ram_addr_t memory_size
;
47 ram_addr_t phys_offset
;
52 typedef struct kvm_dirty_log KVMDirtyLog
;
60 #ifdef KVM_CAP_COALESCED_MMIO
61 struct kvm_coalesced_mmio_ring
*coalesced_mmio_ring
;
63 int broken_set_mem_region
;
66 int robust_singlestep
;
68 #ifdef KVM_CAP_SET_GUEST_DEBUG
69 struct kvm_sw_breakpoint_head kvm_sw_breakpoints
;
71 int irqchip_in_kernel
;
75 static KVMState
*kvm_state
;
77 static KVMSlot
*kvm_alloc_slot(KVMState
*s
)
81 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
82 /* KVM private memory slots */
85 if (s
->slots
[i
].memory_size
== 0)
89 fprintf(stderr
, "%s: no free slot available\n", __func__
);
93 static KVMSlot
*kvm_lookup_matching_slot(KVMState
*s
,
94 target_phys_addr_t start_addr
,
95 target_phys_addr_t end_addr
)
99 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
100 KVMSlot
*mem
= &s
->slots
[i
];
102 if (start_addr
== mem
->start_addr
&&
103 end_addr
== mem
->start_addr
+ mem
->memory_size
) {
112 * Find overlapping slot with lowest start address
114 static KVMSlot
*kvm_lookup_overlapping_slot(KVMState
*s
,
115 target_phys_addr_t start_addr
,
116 target_phys_addr_t end_addr
)
118 KVMSlot
*found
= NULL
;
121 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
122 KVMSlot
*mem
= &s
->slots
[i
];
124 if (mem
->memory_size
== 0 ||
125 (found
&& found
->start_addr
< mem
->start_addr
)) {
129 if (end_addr
> mem
->start_addr
&&
130 start_addr
< mem
->start_addr
+ mem
->memory_size
) {
138 static int kvm_set_user_memory_region(KVMState
*s
, KVMSlot
*slot
)
140 struct kvm_userspace_memory_region mem
;
142 mem
.slot
= slot
->slot
;
143 mem
.guest_phys_addr
= slot
->start_addr
;
144 mem
.memory_size
= slot
->memory_size
;
145 mem
.userspace_addr
= (unsigned long)qemu_get_ram_ptr(slot
->phys_offset
);
146 mem
.flags
= slot
->flags
;
147 if (s
->migration_log
) {
148 mem
.flags
|= KVM_MEM_LOG_DIRTY_PAGES
;
150 return kvm_vm_ioctl(s
, KVM_SET_USER_MEMORY_REGION
, &mem
);
153 static void kvm_reset_vcpu(void *opaque
)
155 CPUState
*env
= opaque
;
157 kvm_arch_reset_vcpu(env
);
160 int kvm_irqchip_in_kernel(void)
162 return kvm_state
->irqchip_in_kernel
;
165 int kvm_pit_in_kernel(void)
167 return kvm_state
->pit_in_kernel
;
171 int kvm_init_vcpu(CPUState
*env
)
173 KVMState
*s
= kvm_state
;
177 DPRINTF("kvm_init_vcpu\n");
179 ret
= kvm_vm_ioctl(s
, KVM_CREATE_VCPU
, env
->cpu_index
);
181 DPRINTF("kvm_create_vcpu failed\n");
188 mmap_size
= kvm_ioctl(s
, KVM_GET_VCPU_MMAP_SIZE
, 0);
190 DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n");
194 env
->kvm_run
= mmap(NULL
, mmap_size
, PROT_READ
| PROT_WRITE
, MAP_SHARED
,
196 if (env
->kvm_run
== MAP_FAILED
) {
198 DPRINTF("mmap'ing vcpu state failed\n");
202 #ifdef KVM_CAP_COALESCED_MMIO
203 if (s
->coalesced_mmio
&& !s
->coalesced_mmio_ring
)
204 s
->coalesced_mmio_ring
= (void *) env
->kvm_run
+
205 s
->coalesced_mmio
* PAGE_SIZE
;
208 ret
= kvm_arch_init_vcpu(env
);
210 qemu_register_reset(kvm_reset_vcpu
, env
);
211 kvm_arch_reset_vcpu(env
);
218 * dirty pages logging control
220 static int kvm_dirty_pages_log_change(target_phys_addr_t phys_addr
,
221 ram_addr_t size
, int flags
, int mask
)
223 KVMState
*s
= kvm_state
;
224 KVMSlot
*mem
= kvm_lookup_matching_slot(s
, phys_addr
, phys_addr
+ size
);
228 fprintf(stderr
, "BUG: %s: invalid parameters " TARGET_FMT_plx
"-"
229 TARGET_FMT_plx
"\n", __func__
, phys_addr
,
230 (target_phys_addr_t
)(phys_addr
+ size
- 1));
234 old_flags
= mem
->flags
;
236 flags
= (mem
->flags
& ~mask
) | flags
;
239 /* If nothing changed effectively, no need to issue ioctl */
240 if (s
->migration_log
) {
241 flags
|= KVM_MEM_LOG_DIRTY_PAGES
;
243 if (flags
== old_flags
) {
247 return kvm_set_user_memory_region(s
, mem
);
250 int kvm_log_start(target_phys_addr_t phys_addr
, ram_addr_t size
)
252 return kvm_dirty_pages_log_change(phys_addr
, size
,
253 KVM_MEM_LOG_DIRTY_PAGES
,
254 KVM_MEM_LOG_DIRTY_PAGES
);
257 int kvm_log_stop(target_phys_addr_t phys_addr
, ram_addr_t size
)
259 return kvm_dirty_pages_log_change(phys_addr
, size
,
261 KVM_MEM_LOG_DIRTY_PAGES
);
264 static int kvm_set_migration_log(int enable
)
266 KVMState
*s
= kvm_state
;
270 s
->migration_log
= enable
;
272 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
275 if (!!(mem
->flags
& KVM_MEM_LOG_DIRTY_PAGES
) == enable
) {
278 err
= kvm_set_user_memory_region(s
, mem
);
286 static int test_le_bit(unsigned long nr
, unsigned char *addr
)
288 return (addr
[nr
>> 3] >> (nr
& 7)) & 1;
292 * kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space
293 * This function updates qemu's dirty bitmap using cpu_physical_memory_set_dirty().
294 * This means all bits are set to dirty.
296 * @start_add: start of logged region.
297 * @end_addr: end of logged region.
299 static int kvm_physical_sync_dirty_bitmap(target_phys_addr_t start_addr
,
300 target_phys_addr_t end_addr
)
302 KVMState
*s
= kvm_state
;
303 unsigned long size
, allocated_size
= 0;
304 target_phys_addr_t phys_addr
;
310 d
.dirty_bitmap
= NULL
;
311 while (start_addr
< end_addr
) {
312 mem
= kvm_lookup_overlapping_slot(s
, start_addr
, end_addr
);
317 size
= ((mem
->memory_size
>> TARGET_PAGE_BITS
) + 7) / 8;
318 if (!d
.dirty_bitmap
) {
319 d
.dirty_bitmap
= qemu_malloc(size
);
320 } else if (size
> allocated_size
) {
321 d
.dirty_bitmap
= qemu_realloc(d
.dirty_bitmap
, size
);
323 allocated_size
= size
;
324 memset(d
.dirty_bitmap
, 0, allocated_size
);
328 if (kvm_vm_ioctl(s
, KVM_GET_DIRTY_LOG
, &d
) == -1) {
329 DPRINTF("ioctl failed %d\n", errno
);
334 for (phys_addr
= mem
->start_addr
, addr
= mem
->phys_offset
;
335 phys_addr
< mem
->start_addr
+ mem
->memory_size
;
336 phys_addr
+= TARGET_PAGE_SIZE
, addr
+= TARGET_PAGE_SIZE
) {
337 unsigned char *bitmap
= (unsigned char *)d
.dirty_bitmap
;
338 unsigned nr
= (phys_addr
- mem
->start_addr
) >> TARGET_PAGE_BITS
;
340 if (test_le_bit(nr
, bitmap
)) {
341 cpu_physical_memory_set_dirty(addr
);
344 start_addr
= phys_addr
;
346 qemu_free(d
.dirty_bitmap
);
351 int kvm_coalesce_mmio_region(target_phys_addr_t start
, ram_addr_t size
)
354 #ifdef KVM_CAP_COALESCED_MMIO
355 KVMState
*s
= kvm_state
;
357 if (s
->coalesced_mmio
) {
358 struct kvm_coalesced_mmio_zone zone
;
363 ret
= kvm_vm_ioctl(s
, KVM_REGISTER_COALESCED_MMIO
, &zone
);
370 int kvm_uncoalesce_mmio_region(target_phys_addr_t start
, ram_addr_t size
)
373 #ifdef KVM_CAP_COALESCED_MMIO
374 KVMState
*s
= kvm_state
;
376 if (s
->coalesced_mmio
) {
377 struct kvm_coalesced_mmio_zone zone
;
382 ret
= kvm_vm_ioctl(s
, KVM_UNREGISTER_COALESCED_MMIO
, &zone
);
389 int kvm_check_extension(KVMState
*s
, unsigned int extension
)
393 ret
= kvm_ioctl(s
, KVM_CHECK_EXTENSION
, extension
);
401 static void kvm_set_phys_mem(target_phys_addr_t start_addr
,
403 ram_addr_t phys_offset
)
405 KVMState
*s
= kvm_state
;
406 ram_addr_t flags
= phys_offset
& ~TARGET_PAGE_MASK
;
410 if (start_addr
& ~TARGET_PAGE_MASK
) {
411 if (flags
>= IO_MEM_UNASSIGNED
) {
412 if (!kvm_lookup_overlapping_slot(s
, start_addr
,
413 start_addr
+ size
)) {
416 fprintf(stderr
, "Unaligned split of a KVM memory slot\n");
418 fprintf(stderr
, "Only page-aligned memory slots supported\n");
423 /* KVM does not support read-only slots */
424 phys_offset
&= ~IO_MEM_ROM
;
427 mem
= kvm_lookup_overlapping_slot(s
, start_addr
, start_addr
+ size
);
432 if (flags
< IO_MEM_UNASSIGNED
&& start_addr
>= mem
->start_addr
&&
433 (start_addr
+ size
<= mem
->start_addr
+ mem
->memory_size
) &&
434 (phys_offset
- start_addr
== mem
->phys_offset
- mem
->start_addr
)) {
435 /* The new slot fits into the existing one and comes with
436 * identical parameters - nothing to be done. */
442 /* unregister the overlapping slot */
443 mem
->memory_size
= 0;
444 err
= kvm_set_user_memory_region(s
, mem
);
446 fprintf(stderr
, "%s: error unregistering overlapping slot: %s\n",
447 __func__
, strerror(-err
));
451 /* Workaround for older KVM versions: we can't join slots, even not by
452 * unregistering the previous ones and then registering the larger
453 * slot. We have to maintain the existing fragmentation. Sigh.
455 * This workaround assumes that the new slot starts at the same
456 * address as the first existing one. If not or if some overlapping
457 * slot comes around later, we will fail (not seen in practice so far)
458 * - and actually require a recent KVM version. */
459 if (s
->broken_set_mem_region
&&
460 old
.start_addr
== start_addr
&& old
.memory_size
< size
&&
461 flags
< IO_MEM_UNASSIGNED
) {
462 mem
= kvm_alloc_slot(s
);
463 mem
->memory_size
= old
.memory_size
;
464 mem
->start_addr
= old
.start_addr
;
465 mem
->phys_offset
= old
.phys_offset
;
468 err
= kvm_set_user_memory_region(s
, mem
);
470 fprintf(stderr
, "%s: error updating slot: %s\n", __func__
,
475 start_addr
+= old
.memory_size
;
476 phys_offset
+= old
.memory_size
;
477 size
-= old
.memory_size
;
481 /* register prefix slot */
482 if (old
.start_addr
< start_addr
) {
483 mem
= kvm_alloc_slot(s
);
484 mem
->memory_size
= start_addr
- old
.start_addr
;
485 mem
->start_addr
= old
.start_addr
;
486 mem
->phys_offset
= old
.phys_offset
;
489 err
= kvm_set_user_memory_region(s
, mem
);
491 fprintf(stderr
, "%s: error registering prefix slot: %s\n",
492 __func__
, strerror(-err
));
497 /* register suffix slot */
498 if (old
.start_addr
+ old
.memory_size
> start_addr
+ size
) {
499 ram_addr_t size_delta
;
501 mem
= kvm_alloc_slot(s
);
502 mem
->start_addr
= start_addr
+ size
;
503 size_delta
= mem
->start_addr
- old
.start_addr
;
504 mem
->memory_size
= old
.memory_size
- size_delta
;
505 mem
->phys_offset
= old
.phys_offset
+ size_delta
;
508 err
= kvm_set_user_memory_region(s
, mem
);
510 fprintf(stderr
, "%s: error registering suffix slot: %s\n",
511 __func__
, strerror(-err
));
517 /* in case the KVM bug workaround already "consumed" the new slot */
521 /* KVM does not need to know about this memory */
522 if (flags
>= IO_MEM_UNASSIGNED
)
525 mem
= kvm_alloc_slot(s
);
526 mem
->memory_size
= size
;
527 mem
->start_addr
= start_addr
;
528 mem
->phys_offset
= phys_offset
;
531 err
= kvm_set_user_memory_region(s
, mem
);
533 fprintf(stderr
, "%s: error registering slot: %s\n", __func__
,
539 static void kvm_client_set_memory(struct CPUPhysMemoryClient
*client
,
540 target_phys_addr_t start_addr
,
542 ram_addr_t phys_offset
)
544 kvm_set_phys_mem(start_addr
, size
, phys_offset
);
547 static int kvm_client_sync_dirty_bitmap(struct CPUPhysMemoryClient
*client
,
548 target_phys_addr_t start_addr
,
549 target_phys_addr_t end_addr
)
551 return kvm_physical_sync_dirty_bitmap(start_addr
, end_addr
);
554 static int kvm_client_migration_log(struct CPUPhysMemoryClient
*client
,
557 return kvm_set_migration_log(enable
);
560 static CPUPhysMemoryClient kvm_cpu_phys_memory_client
= {
561 .set_memory
= kvm_client_set_memory
,
562 .sync_dirty_bitmap
= kvm_client_sync_dirty_bitmap
,
563 .migration_log
= kvm_client_migration_log
,
566 int kvm_init(int smp_cpus
)
568 static const char upgrade_note
[] =
569 "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n"
570 "(see http://sourceforge.net/projects/kvm).\n";
576 fprintf(stderr
, "No SMP KVM support, use '-smp 1'\n");
580 s
= qemu_mallocz(sizeof(KVMState
));
582 #ifdef KVM_CAP_SET_GUEST_DEBUG
583 QTAILQ_INIT(&s
->kvm_sw_breakpoints
);
585 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++)
586 s
->slots
[i
].slot
= i
;
589 s
->fd
= qemu_open("/dev/kvm", O_RDWR
);
591 fprintf(stderr
, "Could not access KVM kernel module: %m\n");
596 ret
= kvm_ioctl(s
, KVM_GET_API_VERSION
, 0);
597 if (ret
< KVM_API_VERSION
) {
600 fprintf(stderr
, "kvm version too old\n");
604 if (ret
> KVM_API_VERSION
) {
606 fprintf(stderr
, "kvm version not supported\n");
610 s
->vmfd
= kvm_ioctl(s
, KVM_CREATE_VM
, 0);
613 fprintf(stderr
, "Please add the 'switch_amode' kernel parameter to "
614 "your host kernel command line\n");
619 /* initially, KVM allocated its own memory and we had to jump through
620 * hooks to make phys_ram_base point to this. Modern versions of KVM
621 * just use a user allocated buffer so we can use regular pages
622 * unmodified. Make sure we have a sufficiently modern version of KVM.
624 if (!kvm_check_extension(s
, KVM_CAP_USER_MEMORY
)) {
626 fprintf(stderr
, "kvm does not support KVM_CAP_USER_MEMORY\n%s",
631 /* There was a nasty bug in < kvm-80 that prevents memory slots from being
632 * destroyed properly. Since we rely on this capability, refuse to work
633 * with any kernel without this capability. */
634 if (!kvm_check_extension(s
, KVM_CAP_DESTROY_MEMORY_REGION_WORKS
)) {
638 "KVM kernel module broken (DESTROY_MEMORY_REGION).\n%s",
643 s
->coalesced_mmio
= 0;
644 #ifdef KVM_CAP_COALESCED_MMIO
645 s
->coalesced_mmio
= kvm_check_extension(s
, KVM_CAP_COALESCED_MMIO
);
646 s
->coalesced_mmio_ring
= NULL
;
649 s
->broken_set_mem_region
= 1;
650 #ifdef KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
651 ret
= kvm_ioctl(s
, KVM_CHECK_EXTENSION
, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
);
653 s
->broken_set_mem_region
= 0;
658 #ifdef KVM_CAP_VCPU_EVENTS
659 s
->vcpu_events
= kvm_check_extension(s
, KVM_CAP_VCPU_EVENTS
);
662 s
->robust_singlestep
= 0;
663 #ifdef KVM_CAP_X86_ROBUST_SINGLESTEP
664 s
->robust_singlestep
=
665 kvm_check_extension(s
, KVM_CAP_X86_ROBUST_SINGLESTEP
);
669 #ifdef KVM_CAP_DEBUGREGS
670 s
->debugregs
= kvm_check_extension(s
, KVM_CAP_DEBUGREGS
);
673 ret
= kvm_arch_init(s
, smp_cpus
);
678 cpu_register_phys_memory_client(&kvm_cpu_phys_memory_client
);
694 static int kvm_handle_io(uint16_t port
, void *data
, int direction
, int size
,
700 for (i
= 0; i
< count
; i
++) {
701 if (direction
== KVM_EXIT_IO_IN
) {
704 stb_p(ptr
, cpu_inb(port
));
707 stw_p(ptr
, cpu_inw(port
));
710 stl_p(ptr
, cpu_inl(port
));
716 cpu_outb(port
, ldub_p(ptr
));
719 cpu_outw(port
, lduw_p(ptr
));
722 cpu_outl(port
, ldl_p(ptr
));
733 void kvm_flush_coalesced_mmio_buffer(void)
735 #ifdef KVM_CAP_COALESCED_MMIO
736 KVMState
*s
= kvm_state
;
737 if (s
->coalesced_mmio_ring
) {
738 struct kvm_coalesced_mmio_ring
*ring
= s
->coalesced_mmio_ring
;
739 while (ring
->first
!= ring
->last
) {
740 struct kvm_coalesced_mmio
*ent
;
742 ent
= &ring
->coalesced_mmio
[ring
->first
];
744 cpu_physical_memory_write(ent
->phys_addr
, ent
->data
, ent
->len
);
746 ring
->first
= (ring
->first
+ 1) % KVM_COALESCED_MMIO_MAX
;
752 void kvm_cpu_synchronize_state(CPUState
*env
)
754 if (!env
->kvm_vcpu_dirty
) {
755 kvm_arch_get_registers(env
);
756 env
->kvm_vcpu_dirty
= 1;
760 void kvm_cpu_synchronize_post_reset(CPUState
*env
)
762 kvm_arch_put_registers(env
, KVM_PUT_RESET_STATE
);
763 env
->kvm_vcpu_dirty
= 0;
766 void kvm_cpu_synchronize_post_init(CPUState
*env
)
768 kvm_arch_put_registers(env
, KVM_PUT_FULL_STATE
);
769 env
->kvm_vcpu_dirty
= 0;
772 int kvm_cpu_exec(CPUState
*env
)
774 struct kvm_run
*run
= env
->kvm_run
;
777 DPRINTF("kvm_cpu_exec()\n");
780 #ifndef CONFIG_IOTHREAD
781 if (env
->exit_request
) {
782 DPRINTF("interrupt exit requested\n");
788 if (env
->kvm_vcpu_dirty
) {
789 kvm_arch_put_registers(env
, KVM_PUT_RUNTIME_STATE
);
790 env
->kvm_vcpu_dirty
= 0;
793 kvm_arch_pre_run(env
, run
);
794 qemu_mutex_unlock_iothread();
795 ret
= kvm_vcpu_ioctl(env
, KVM_RUN
, 0);
796 qemu_mutex_lock_iothread();
797 kvm_arch_post_run(env
, run
);
799 if (ret
== -EINTR
|| ret
== -EAGAIN
) {
801 DPRINTF("io window exit\n");
807 DPRINTF("kvm run failed %s\n", strerror(-ret
));
811 kvm_flush_coalesced_mmio_buffer();
813 ret
= 0; /* exit loop */
814 switch (run
->exit_reason
) {
816 DPRINTF("handle_io\n");
817 ret
= kvm_handle_io(run
->io
.port
,
818 (uint8_t *)run
+ run
->io
.data_offset
,
824 DPRINTF("handle_mmio\n");
825 cpu_physical_memory_rw(run
->mmio
.phys_addr
,
831 case KVM_EXIT_IRQ_WINDOW_OPEN
:
832 DPRINTF("irq_window_open\n");
834 case KVM_EXIT_SHUTDOWN
:
835 DPRINTF("shutdown\n");
836 qemu_system_reset_request();
839 case KVM_EXIT_UNKNOWN
:
840 DPRINTF("kvm_exit_unknown\n");
842 case KVM_EXIT_FAIL_ENTRY
:
843 DPRINTF("kvm_exit_fail_entry\n");
845 case KVM_EXIT_EXCEPTION
:
846 DPRINTF("kvm_exit_exception\n");
849 DPRINTF("kvm_exit_debug\n");
850 #ifdef KVM_CAP_SET_GUEST_DEBUG
851 if (kvm_arch_debug(&run
->debug
.arch
)) {
852 gdb_set_stop_cpu(env
);
854 env
->exception_index
= EXCP_DEBUG
;
857 /* re-enter, this exception was guest-internal */
859 #endif /* KVM_CAP_SET_GUEST_DEBUG */
862 DPRINTF("kvm_arch_handle_exit\n");
863 ret
= kvm_arch_handle_exit(env
, run
);
868 if (env
->exit_request
) {
869 env
->exit_request
= 0;
870 env
->exception_index
= EXCP_INTERRUPT
;
876 int kvm_ioctl(KVMState
*s
, int type
, ...)
883 arg
= va_arg(ap
, void *);
886 ret
= ioctl(s
->fd
, type
, arg
);
893 int kvm_vm_ioctl(KVMState
*s
, int type
, ...)
900 arg
= va_arg(ap
, void *);
903 ret
= ioctl(s
->vmfd
, type
, arg
);
910 int kvm_vcpu_ioctl(CPUState
*env
, int type
, ...)
917 arg
= va_arg(ap
, void *);
920 ret
= ioctl(env
->kvm_fd
, type
, arg
);
927 int kvm_has_sync_mmu(void)
929 #ifdef KVM_CAP_SYNC_MMU
930 KVMState
*s
= kvm_state
;
932 return kvm_check_extension(s
, KVM_CAP_SYNC_MMU
);
938 int kvm_has_vcpu_events(void)
940 return kvm_state
->vcpu_events
;
943 int kvm_has_robust_singlestep(void)
945 return kvm_state
->robust_singlestep
;
948 int kvm_has_debugregs(void)
950 return kvm_state
->debugregs
;
953 void kvm_setup_guest_memory(void *start
, size_t size
)
955 if (!kvm_has_sync_mmu()) {
957 int ret
= madvise(start
, size
, MADV_DONTFORK
);
965 "Need MADV_DONTFORK in absence of synchronous KVM MMU\n");
971 #ifdef KVM_CAP_SET_GUEST_DEBUG
972 static void on_vcpu(CPUState
*env
, void (*func
)(void *data
), void *data
)
974 #ifdef CONFIG_IOTHREAD
975 if (env
!= cpu_single_env
) {
982 struct kvm_sw_breakpoint
*kvm_find_sw_breakpoint(CPUState
*env
,
985 struct kvm_sw_breakpoint
*bp
;
987 QTAILQ_FOREACH(bp
, &env
->kvm_state
->kvm_sw_breakpoints
, entry
) {
994 int kvm_sw_breakpoints_active(CPUState
*env
)
996 return !QTAILQ_EMPTY(&env
->kvm_state
->kvm_sw_breakpoints
);
999 struct kvm_set_guest_debug_data
{
1000 struct kvm_guest_debug dbg
;
1005 static void kvm_invoke_set_guest_debug(void *data
)
1007 struct kvm_set_guest_debug_data
*dbg_data
= data
;
1008 CPUState
*env
= dbg_data
->env
;
1010 dbg_data
->err
= kvm_vcpu_ioctl(env
, KVM_SET_GUEST_DEBUG
, &dbg_data
->dbg
);
1013 int kvm_update_guest_debug(CPUState
*env
, unsigned long reinject_trap
)
1015 struct kvm_set_guest_debug_data data
;
1017 data
.dbg
.control
= reinject_trap
;
1019 if (env
->singlestep_enabled
) {
1020 data
.dbg
.control
|= KVM_GUESTDBG_ENABLE
| KVM_GUESTDBG_SINGLESTEP
;
1022 kvm_arch_update_guest_debug(env
, &data
.dbg
);
1025 on_vcpu(env
, kvm_invoke_set_guest_debug
, &data
);
1029 int kvm_insert_breakpoint(CPUState
*current_env
, target_ulong addr
,
1030 target_ulong len
, int type
)
1032 struct kvm_sw_breakpoint
*bp
;
1036 if (type
== GDB_BREAKPOINT_SW
) {
1037 bp
= kvm_find_sw_breakpoint(current_env
, addr
);
1043 bp
= qemu_malloc(sizeof(struct kvm_sw_breakpoint
));
1049 err
= kvm_arch_insert_sw_breakpoint(current_env
, bp
);
1055 QTAILQ_INSERT_HEAD(¤t_env
->kvm_state
->kvm_sw_breakpoints
,
1058 err
= kvm_arch_insert_hw_breakpoint(addr
, len
, type
);
1063 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
1064 err
= kvm_update_guest_debug(env
, 0);
1071 int kvm_remove_breakpoint(CPUState
*current_env
, target_ulong addr
,
1072 target_ulong len
, int type
)
1074 struct kvm_sw_breakpoint
*bp
;
1078 if (type
== GDB_BREAKPOINT_SW
) {
1079 bp
= kvm_find_sw_breakpoint(current_env
, addr
);
1083 if (bp
->use_count
> 1) {
1088 err
= kvm_arch_remove_sw_breakpoint(current_env
, bp
);
1092 QTAILQ_REMOVE(¤t_env
->kvm_state
->kvm_sw_breakpoints
, bp
, entry
);
1095 err
= kvm_arch_remove_hw_breakpoint(addr
, len
, type
);
1100 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
1101 err
= kvm_update_guest_debug(env
, 0);
1108 void kvm_remove_all_breakpoints(CPUState
*current_env
)
1110 struct kvm_sw_breakpoint
*bp
, *next
;
1111 KVMState
*s
= current_env
->kvm_state
;
1114 QTAILQ_FOREACH_SAFE(bp
, &s
->kvm_sw_breakpoints
, entry
, next
) {
1115 if (kvm_arch_remove_sw_breakpoint(current_env
, bp
) != 0) {
1116 /* Try harder to find a CPU that currently sees the breakpoint. */
1117 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
1118 if (kvm_arch_remove_sw_breakpoint(env
, bp
) == 0)
1123 kvm_arch_remove_all_hw_breakpoints();
1125 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
)
1126 kvm_update_guest_debug(env
, 0);
1129 #else /* !KVM_CAP_SET_GUEST_DEBUG */
1131 int kvm_update_guest_debug(CPUState
*env
, unsigned long reinject_trap
)
1136 int kvm_insert_breakpoint(CPUState
*current_env
, target_ulong addr
,
1137 target_ulong len
, int type
)
1142 int kvm_remove_breakpoint(CPUState
*current_env
, target_ulong addr
,
1143 target_ulong len
, int type
)
1148 void kvm_remove_all_breakpoints(CPUState
*current_env
)
1151 #endif /* !KVM_CAP_SET_GUEST_DEBUG */
1153 int kvm_set_signal_mask(CPUState
*env
, const sigset_t
*sigset
)
1155 struct kvm_signal_mask
*sigmask
;
1159 return kvm_vcpu_ioctl(env
, KVM_SET_SIGNAL_MASK
, NULL
);
1161 sigmask
= qemu_malloc(sizeof(*sigmask
) + sizeof(*sigset
));
1164 memcpy(sigmask
->sigset
, sigset
, sizeof(*sigset
));
1165 r
= kvm_vcpu_ioctl(env
, KVM_SET_SIGNAL_MASK
, sigmask
);
1171 int kvm_set_ioeventfd_pio_word(int fd
, uint16_t addr
, uint16_t val
, bool assign
)
1173 #ifdef KVM_IOEVENTFD
1174 struct kvm_ioeventfd kick
= {
1178 .flags
= KVM_IOEVENTFD_FLAG_DATAMATCH
| KVM_IOEVENTFD_FLAG_PIO
,
1185 kick
.flags
|= KVM_IOEVENTFD_FLAG_DEASSIGN
;
1186 r
= kvm_vm_ioctl(kvm_state
, KVM_IOEVENTFD
, &kick
);