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 typedef struct KVMSlot
46 target_phys_addr_t start_addr
;
47 ram_addr_t memory_size
;
48 ram_addr_t phys_offset
;
53 typedef struct kvm_dirty_log KVMDirtyLog
;
61 #ifdef KVM_CAP_COALESCED_MMIO
62 struct kvm_coalesced_mmio_ring
*coalesced_mmio_ring
;
64 int broken_set_mem_region
;
67 int robust_singlestep
;
69 #ifdef KVM_CAP_SET_GUEST_DEBUG
70 struct kvm_sw_breakpoint_head kvm_sw_breakpoints
;
72 int irqchip_in_kernel
;
76 static KVMState
*kvm_state
;
78 static KVMSlot
*kvm_alloc_slot(KVMState
*s
)
82 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
83 /* KVM private memory slots */
86 if (s
->slots
[i
].memory_size
== 0)
90 fprintf(stderr
, "%s: no free slot available\n", __func__
);
94 static KVMSlot
*kvm_lookup_matching_slot(KVMState
*s
,
95 target_phys_addr_t start_addr
,
96 target_phys_addr_t end_addr
)
100 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
101 KVMSlot
*mem
= &s
->slots
[i
];
103 if (start_addr
== mem
->start_addr
&&
104 end_addr
== mem
->start_addr
+ mem
->memory_size
) {
113 * Find overlapping slot with lowest start address
115 static KVMSlot
*kvm_lookup_overlapping_slot(KVMState
*s
,
116 target_phys_addr_t start_addr
,
117 target_phys_addr_t end_addr
)
119 KVMSlot
*found
= NULL
;
122 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
123 KVMSlot
*mem
= &s
->slots
[i
];
125 if (mem
->memory_size
== 0 ||
126 (found
&& found
->start_addr
< mem
->start_addr
)) {
130 if (end_addr
> mem
->start_addr
&&
131 start_addr
< mem
->start_addr
+ mem
->memory_size
) {
139 static int kvm_set_user_memory_region(KVMState
*s
, KVMSlot
*slot
)
141 struct kvm_userspace_memory_region mem
;
143 mem
.slot
= slot
->slot
;
144 mem
.guest_phys_addr
= slot
->start_addr
;
145 mem
.memory_size
= slot
->memory_size
;
146 mem
.userspace_addr
= (unsigned long)qemu_get_ram_ptr(slot
->phys_offset
);
147 mem
.flags
= slot
->flags
;
148 if (s
->migration_log
) {
149 mem
.flags
|= KVM_MEM_LOG_DIRTY_PAGES
;
151 return kvm_vm_ioctl(s
, KVM_SET_USER_MEMORY_REGION
, &mem
);
154 static void kvm_reset_vcpu(void *opaque
)
156 CPUState
*env
= opaque
;
158 kvm_arch_reset_vcpu(env
);
161 int kvm_irqchip_in_kernel(void)
163 return kvm_state
->irqchip_in_kernel
;
166 int kvm_pit_in_kernel(void)
168 return kvm_state
->pit_in_kernel
;
172 int kvm_init_vcpu(CPUState
*env
)
174 KVMState
*s
= kvm_state
;
178 DPRINTF("kvm_init_vcpu\n");
180 ret
= kvm_vm_ioctl(s
, KVM_CREATE_VCPU
, env
->cpu_index
);
182 DPRINTF("kvm_create_vcpu failed\n");
189 mmap_size
= kvm_ioctl(s
, KVM_GET_VCPU_MMAP_SIZE
, 0);
191 DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n");
195 env
->kvm_run
= mmap(NULL
, mmap_size
, PROT_READ
| PROT_WRITE
, MAP_SHARED
,
197 if (env
->kvm_run
== MAP_FAILED
) {
199 DPRINTF("mmap'ing vcpu state failed\n");
203 #ifdef KVM_CAP_COALESCED_MMIO
204 if (s
->coalesced_mmio
&& !s
->coalesced_mmio_ring
)
205 s
->coalesced_mmio_ring
= (void *) env
->kvm_run
+
206 s
->coalesced_mmio
* PAGE_SIZE
;
209 ret
= kvm_arch_init_vcpu(env
);
211 qemu_register_reset(kvm_reset_vcpu
, env
);
212 kvm_arch_reset_vcpu(env
);
219 * dirty pages logging control
221 static int kvm_dirty_pages_log_change(target_phys_addr_t phys_addr
,
222 ram_addr_t size
, int flags
, int mask
)
224 KVMState
*s
= kvm_state
;
225 KVMSlot
*mem
= kvm_lookup_matching_slot(s
, phys_addr
, phys_addr
+ size
);
229 fprintf(stderr
, "BUG: %s: invalid parameters " TARGET_FMT_plx
"-"
230 TARGET_FMT_plx
"\n", __func__
, phys_addr
,
231 (target_phys_addr_t
)(phys_addr
+ size
- 1));
235 old_flags
= mem
->flags
;
237 flags
= (mem
->flags
& ~mask
) | flags
;
240 /* If nothing changed effectively, no need to issue ioctl */
241 if (s
->migration_log
) {
242 flags
|= KVM_MEM_LOG_DIRTY_PAGES
;
244 if (flags
== old_flags
) {
248 return kvm_set_user_memory_region(s
, mem
);
251 int kvm_log_start(target_phys_addr_t phys_addr
, ram_addr_t size
)
253 return kvm_dirty_pages_log_change(phys_addr
, size
,
254 KVM_MEM_LOG_DIRTY_PAGES
,
255 KVM_MEM_LOG_DIRTY_PAGES
);
258 int kvm_log_stop(target_phys_addr_t phys_addr
, ram_addr_t size
)
260 return kvm_dirty_pages_log_change(phys_addr
, size
,
262 KVM_MEM_LOG_DIRTY_PAGES
);
265 static int kvm_set_migration_log(int enable
)
267 KVMState
*s
= kvm_state
;
271 s
->migration_log
= enable
;
273 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
276 if (!!(mem
->flags
& KVM_MEM_LOG_DIRTY_PAGES
) == enable
) {
279 err
= kvm_set_user_memory_region(s
, mem
);
287 /* get kvm's dirty pages bitmap and update qemu's */
288 static int kvm_get_dirty_pages_log_range(unsigned long start_addr
,
289 unsigned long *bitmap
,
290 unsigned long offset
,
291 unsigned long mem_size
)
294 unsigned long page_number
, addr
, addr1
, c
;
296 unsigned int len
= ((mem_size
/ TARGET_PAGE_SIZE
) + HOST_LONG_BITS
- 1) /
300 * bitmap-traveling is faster than memory-traveling (for addr...)
301 * especially when most of the memory is not dirty.
303 for (i
= 0; i
< len
; i
++) {
304 if (bitmap
[i
] != 0) {
305 c
= leul_to_cpu(bitmap
[i
]);
309 page_number
= i
* HOST_LONG_BITS
+ j
;
310 addr1
= page_number
* TARGET_PAGE_SIZE
;
311 addr
= offset
+ addr1
;
312 ram_addr
= cpu_get_physical_page_desc(addr
);
313 cpu_physical_memory_set_dirty(ram_addr
);
320 #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1))
323 * kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space
324 * This function updates qemu's dirty bitmap using cpu_physical_memory_set_dirty().
325 * This means all bits are set to dirty.
327 * @start_add: start of logged region.
328 * @end_addr: end of logged region.
330 static int kvm_physical_sync_dirty_bitmap(target_phys_addr_t start_addr
,
331 target_phys_addr_t end_addr
)
333 KVMState
*s
= kvm_state
;
334 unsigned long size
, allocated_size
= 0;
339 d
.dirty_bitmap
= NULL
;
340 while (start_addr
< end_addr
) {
341 mem
= kvm_lookup_overlapping_slot(s
, start_addr
, end_addr
);
346 size
= ALIGN(((mem
->memory_size
) >> TARGET_PAGE_BITS
), HOST_LONG_BITS
) / 8;
347 if (!d
.dirty_bitmap
) {
348 d
.dirty_bitmap
= qemu_malloc(size
);
349 } else if (size
> allocated_size
) {
350 d
.dirty_bitmap
= qemu_realloc(d
.dirty_bitmap
, size
);
352 allocated_size
= size
;
353 memset(d
.dirty_bitmap
, 0, allocated_size
);
357 if (kvm_vm_ioctl(s
, KVM_GET_DIRTY_LOG
, &d
) == -1) {
358 DPRINTF("ioctl failed %d\n", errno
);
363 kvm_get_dirty_pages_log_range(mem
->start_addr
, d
.dirty_bitmap
,
364 mem
->start_addr
, mem
->memory_size
);
365 start_addr
= mem
->start_addr
+ mem
->memory_size
;
367 qemu_free(d
.dirty_bitmap
);
372 int kvm_coalesce_mmio_region(target_phys_addr_t start
, ram_addr_t size
)
375 #ifdef KVM_CAP_COALESCED_MMIO
376 KVMState
*s
= kvm_state
;
378 if (s
->coalesced_mmio
) {
379 struct kvm_coalesced_mmio_zone zone
;
384 ret
= kvm_vm_ioctl(s
, KVM_REGISTER_COALESCED_MMIO
, &zone
);
391 int kvm_uncoalesce_mmio_region(target_phys_addr_t start
, ram_addr_t size
)
394 #ifdef KVM_CAP_COALESCED_MMIO
395 KVMState
*s
= kvm_state
;
397 if (s
->coalesced_mmio
) {
398 struct kvm_coalesced_mmio_zone zone
;
403 ret
= kvm_vm_ioctl(s
, KVM_UNREGISTER_COALESCED_MMIO
, &zone
);
410 int kvm_check_extension(KVMState
*s
, unsigned int extension
)
414 ret
= kvm_ioctl(s
, KVM_CHECK_EXTENSION
, extension
);
422 static void kvm_set_phys_mem(target_phys_addr_t start_addr
,
424 ram_addr_t phys_offset
)
426 KVMState
*s
= kvm_state
;
427 ram_addr_t flags
= phys_offset
& ~TARGET_PAGE_MASK
;
431 if (start_addr
& ~TARGET_PAGE_MASK
) {
432 if (flags
>= IO_MEM_UNASSIGNED
) {
433 if (!kvm_lookup_overlapping_slot(s
, start_addr
,
434 start_addr
+ size
)) {
437 fprintf(stderr
, "Unaligned split of a KVM memory slot\n");
439 fprintf(stderr
, "Only page-aligned memory slots supported\n");
444 /* KVM does not support read-only slots */
445 phys_offset
&= ~IO_MEM_ROM
;
448 mem
= kvm_lookup_overlapping_slot(s
, start_addr
, start_addr
+ size
);
453 if (flags
< IO_MEM_UNASSIGNED
&& start_addr
>= mem
->start_addr
&&
454 (start_addr
+ size
<= mem
->start_addr
+ mem
->memory_size
) &&
455 (phys_offset
- start_addr
== mem
->phys_offset
- mem
->start_addr
)) {
456 /* The new slot fits into the existing one and comes with
457 * identical parameters - nothing to be done. */
463 /* unregister the overlapping slot */
464 mem
->memory_size
= 0;
465 err
= kvm_set_user_memory_region(s
, mem
);
467 fprintf(stderr
, "%s: error unregistering overlapping slot: %s\n",
468 __func__
, strerror(-err
));
472 /* Workaround for older KVM versions: we can't join slots, even not by
473 * unregistering the previous ones and then registering the larger
474 * slot. We have to maintain the existing fragmentation. Sigh.
476 * This workaround assumes that the new slot starts at the same
477 * address as the first existing one. If not or if some overlapping
478 * slot comes around later, we will fail (not seen in practice so far)
479 * - and actually require a recent KVM version. */
480 if (s
->broken_set_mem_region
&&
481 old
.start_addr
== start_addr
&& old
.memory_size
< size
&&
482 flags
< IO_MEM_UNASSIGNED
) {
483 mem
= kvm_alloc_slot(s
);
484 mem
->memory_size
= old
.memory_size
;
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 updating slot: %s\n", __func__
,
496 start_addr
+= old
.memory_size
;
497 phys_offset
+= old
.memory_size
;
498 size
-= old
.memory_size
;
502 /* register prefix slot */
503 if (old
.start_addr
< start_addr
) {
504 mem
= kvm_alloc_slot(s
);
505 mem
->memory_size
= start_addr
- old
.start_addr
;
506 mem
->start_addr
= old
.start_addr
;
507 mem
->phys_offset
= old
.phys_offset
;
510 err
= kvm_set_user_memory_region(s
, mem
);
512 fprintf(stderr
, "%s: error registering prefix slot: %s\n",
513 __func__
, strerror(-err
));
518 /* register suffix slot */
519 if (old
.start_addr
+ old
.memory_size
> start_addr
+ size
) {
520 ram_addr_t size_delta
;
522 mem
= kvm_alloc_slot(s
);
523 mem
->start_addr
= start_addr
+ size
;
524 size_delta
= mem
->start_addr
- old
.start_addr
;
525 mem
->memory_size
= old
.memory_size
- size_delta
;
526 mem
->phys_offset
= old
.phys_offset
+ size_delta
;
529 err
= kvm_set_user_memory_region(s
, mem
);
531 fprintf(stderr
, "%s: error registering suffix slot: %s\n",
532 __func__
, strerror(-err
));
538 /* in case the KVM bug workaround already "consumed" the new slot */
542 /* KVM does not need to know about this memory */
543 if (flags
>= IO_MEM_UNASSIGNED
)
546 mem
= kvm_alloc_slot(s
);
547 mem
->memory_size
= size
;
548 mem
->start_addr
= start_addr
;
549 mem
->phys_offset
= phys_offset
;
552 err
= kvm_set_user_memory_region(s
, mem
);
554 fprintf(stderr
, "%s: error registering slot: %s\n", __func__
,
560 static void kvm_client_set_memory(struct CPUPhysMemoryClient
*client
,
561 target_phys_addr_t start_addr
,
563 ram_addr_t phys_offset
)
565 kvm_set_phys_mem(start_addr
, size
, phys_offset
);
568 static int kvm_client_sync_dirty_bitmap(struct CPUPhysMemoryClient
*client
,
569 target_phys_addr_t start_addr
,
570 target_phys_addr_t end_addr
)
572 return kvm_physical_sync_dirty_bitmap(start_addr
, end_addr
);
575 static int kvm_client_migration_log(struct CPUPhysMemoryClient
*client
,
578 return kvm_set_migration_log(enable
);
581 static CPUPhysMemoryClient kvm_cpu_phys_memory_client
= {
582 .set_memory
= kvm_client_set_memory
,
583 .sync_dirty_bitmap
= kvm_client_sync_dirty_bitmap
,
584 .migration_log
= kvm_client_migration_log
,
587 int kvm_init(int smp_cpus
)
589 static const char upgrade_note
[] =
590 "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n"
591 "(see http://sourceforge.net/projects/kvm).\n";
597 fprintf(stderr
, "No SMP KVM support, use '-smp 1'\n");
601 s
= qemu_mallocz(sizeof(KVMState
));
603 #ifdef KVM_CAP_SET_GUEST_DEBUG
604 QTAILQ_INIT(&s
->kvm_sw_breakpoints
);
606 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++)
607 s
->slots
[i
].slot
= i
;
610 s
->fd
= qemu_open("/dev/kvm", O_RDWR
);
612 fprintf(stderr
, "Could not access KVM kernel module: %m\n");
617 ret
= kvm_ioctl(s
, KVM_GET_API_VERSION
, 0);
618 if (ret
< KVM_API_VERSION
) {
621 fprintf(stderr
, "kvm version too old\n");
625 if (ret
> KVM_API_VERSION
) {
627 fprintf(stderr
, "kvm version not supported\n");
631 s
->vmfd
= kvm_ioctl(s
, KVM_CREATE_VM
, 0);
634 fprintf(stderr
, "Please add the 'switch_amode' kernel parameter to "
635 "your host kernel command line\n");
640 /* initially, KVM allocated its own memory and we had to jump through
641 * hooks to make phys_ram_base point to this. Modern versions of KVM
642 * just use a user allocated buffer so we can use regular pages
643 * unmodified. Make sure we have a sufficiently modern version of KVM.
645 if (!kvm_check_extension(s
, KVM_CAP_USER_MEMORY
)) {
647 fprintf(stderr
, "kvm does not support KVM_CAP_USER_MEMORY\n%s",
652 /* There was a nasty bug in < kvm-80 that prevents memory slots from being
653 * destroyed properly. Since we rely on this capability, refuse to work
654 * with any kernel without this capability. */
655 if (!kvm_check_extension(s
, KVM_CAP_DESTROY_MEMORY_REGION_WORKS
)) {
659 "KVM kernel module broken (DESTROY_MEMORY_REGION).\n%s",
664 s
->coalesced_mmio
= 0;
665 #ifdef KVM_CAP_COALESCED_MMIO
666 s
->coalesced_mmio
= kvm_check_extension(s
, KVM_CAP_COALESCED_MMIO
);
667 s
->coalesced_mmio_ring
= NULL
;
670 s
->broken_set_mem_region
= 1;
671 #ifdef KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
672 ret
= kvm_ioctl(s
, KVM_CHECK_EXTENSION
, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
);
674 s
->broken_set_mem_region
= 0;
679 #ifdef KVM_CAP_VCPU_EVENTS
680 s
->vcpu_events
= kvm_check_extension(s
, KVM_CAP_VCPU_EVENTS
);
683 s
->robust_singlestep
= 0;
684 #ifdef KVM_CAP_X86_ROBUST_SINGLESTEP
685 s
->robust_singlestep
=
686 kvm_check_extension(s
, KVM_CAP_X86_ROBUST_SINGLESTEP
);
690 #ifdef KVM_CAP_DEBUGREGS
691 s
->debugregs
= kvm_check_extension(s
, KVM_CAP_DEBUGREGS
);
694 ret
= kvm_arch_init(s
, smp_cpus
);
699 cpu_register_phys_memory_client(&kvm_cpu_phys_memory_client
);
715 static int kvm_handle_io(uint16_t port
, void *data
, int direction
, int size
,
721 for (i
= 0; i
< count
; i
++) {
722 if (direction
== KVM_EXIT_IO_IN
) {
725 stb_p(ptr
, cpu_inb(port
));
728 stw_p(ptr
, cpu_inw(port
));
731 stl_p(ptr
, cpu_inl(port
));
737 cpu_outb(port
, ldub_p(ptr
));
740 cpu_outw(port
, lduw_p(ptr
));
743 cpu_outl(port
, ldl_p(ptr
));
754 #ifdef KVM_CAP_INTERNAL_ERROR_DATA
755 static void kvm_handle_internal_error(CPUState
*env
, struct kvm_run
*run
)
758 if (kvm_check_extension(kvm_state
, KVM_CAP_INTERNAL_ERROR_DATA
)) {
761 fprintf(stderr
, "KVM internal error. Suberror: %d\n",
762 run
->internal
.suberror
);
764 for (i
= 0; i
< run
->internal
.ndata
; ++i
) {
765 fprintf(stderr
, "extra data[%d]: %"PRIx64
"\n",
766 i
, (uint64_t)run
->internal
.data
[i
]);
769 cpu_dump_state(env
, stderr
, fprintf
, 0);
770 if (run
->internal
.suberror
== KVM_INTERNAL_ERROR_EMULATION
) {
771 fprintf(stderr
, "emulation failure\n");
773 /* FIXME: Should trigger a qmp message to let management know
774 * something went wrong.
780 void kvm_flush_coalesced_mmio_buffer(void)
782 #ifdef KVM_CAP_COALESCED_MMIO
783 KVMState
*s
= kvm_state
;
784 if (s
->coalesced_mmio_ring
) {
785 struct kvm_coalesced_mmio_ring
*ring
= s
->coalesced_mmio_ring
;
786 while (ring
->first
!= ring
->last
) {
787 struct kvm_coalesced_mmio
*ent
;
789 ent
= &ring
->coalesced_mmio
[ring
->first
];
791 cpu_physical_memory_write(ent
->phys_addr
, ent
->data
, ent
->len
);
793 ring
->first
= (ring
->first
+ 1) % KVM_COALESCED_MMIO_MAX
;
799 void kvm_cpu_synchronize_state(CPUState
*env
)
801 if (!env
->kvm_vcpu_dirty
) {
802 kvm_arch_get_registers(env
);
803 env
->kvm_vcpu_dirty
= 1;
807 void kvm_cpu_synchronize_post_reset(CPUState
*env
)
809 kvm_arch_put_registers(env
, KVM_PUT_RESET_STATE
);
810 env
->kvm_vcpu_dirty
= 0;
813 void kvm_cpu_synchronize_post_init(CPUState
*env
)
815 kvm_arch_put_registers(env
, KVM_PUT_FULL_STATE
);
816 env
->kvm_vcpu_dirty
= 0;
819 int kvm_cpu_exec(CPUState
*env
)
821 struct kvm_run
*run
= env
->kvm_run
;
824 DPRINTF("kvm_cpu_exec()\n");
827 #ifndef CONFIG_IOTHREAD
828 if (env
->exit_request
) {
829 DPRINTF("interrupt exit requested\n");
835 if (env
->kvm_vcpu_dirty
) {
836 kvm_arch_put_registers(env
, KVM_PUT_RUNTIME_STATE
);
837 env
->kvm_vcpu_dirty
= 0;
840 kvm_arch_pre_run(env
, run
);
841 qemu_mutex_unlock_iothread();
842 ret
= kvm_vcpu_ioctl(env
, KVM_RUN
, 0);
843 qemu_mutex_lock_iothread();
844 kvm_arch_post_run(env
, run
);
846 if (ret
== -EINTR
|| ret
== -EAGAIN
) {
848 DPRINTF("io window exit\n");
854 DPRINTF("kvm run failed %s\n", strerror(-ret
));
858 kvm_flush_coalesced_mmio_buffer();
860 ret
= 0; /* exit loop */
861 switch (run
->exit_reason
) {
863 DPRINTF("handle_io\n");
864 ret
= kvm_handle_io(run
->io
.port
,
865 (uint8_t *)run
+ run
->io
.data_offset
,
871 DPRINTF("handle_mmio\n");
872 cpu_physical_memory_rw(run
->mmio
.phys_addr
,
878 case KVM_EXIT_IRQ_WINDOW_OPEN
:
879 DPRINTF("irq_window_open\n");
881 case KVM_EXIT_SHUTDOWN
:
882 DPRINTF("shutdown\n");
883 qemu_system_reset_request();
886 case KVM_EXIT_UNKNOWN
:
887 DPRINTF("kvm_exit_unknown\n");
889 case KVM_EXIT_FAIL_ENTRY
:
890 DPRINTF("kvm_exit_fail_entry\n");
892 case KVM_EXIT_EXCEPTION
:
893 DPRINTF("kvm_exit_exception\n");
895 #ifdef KVM_CAP_INTERNAL_ERROR_DATA
896 case KVM_EXIT_INTERNAL_ERROR
:
897 kvm_handle_internal_error(env
, run
);
901 DPRINTF("kvm_exit_debug\n");
902 #ifdef KVM_CAP_SET_GUEST_DEBUG
903 if (kvm_arch_debug(&run
->debug
.arch
)) {
904 gdb_set_stop_cpu(env
);
906 env
->exception_index
= EXCP_DEBUG
;
909 /* re-enter, this exception was guest-internal */
911 #endif /* KVM_CAP_SET_GUEST_DEBUG */
914 DPRINTF("kvm_arch_handle_exit\n");
915 ret
= kvm_arch_handle_exit(env
, run
);
920 if (env
->exit_request
) {
921 env
->exit_request
= 0;
922 env
->exception_index
= EXCP_INTERRUPT
;
928 int kvm_ioctl(KVMState
*s
, int type
, ...)
935 arg
= va_arg(ap
, void *);
938 ret
= ioctl(s
->fd
, type
, arg
);
945 int kvm_vm_ioctl(KVMState
*s
, int type
, ...)
952 arg
= va_arg(ap
, void *);
955 ret
= ioctl(s
->vmfd
, type
, arg
);
962 int kvm_vcpu_ioctl(CPUState
*env
, int type
, ...)
969 arg
= va_arg(ap
, void *);
972 ret
= ioctl(env
->kvm_fd
, type
, arg
);
979 int kvm_has_sync_mmu(void)
981 #ifdef KVM_CAP_SYNC_MMU
982 KVMState
*s
= kvm_state
;
984 return kvm_check_extension(s
, KVM_CAP_SYNC_MMU
);
990 int kvm_has_vcpu_events(void)
992 return kvm_state
->vcpu_events
;
995 int kvm_has_robust_singlestep(void)
997 return kvm_state
->robust_singlestep
;
1000 int kvm_has_debugregs(void)
1002 return kvm_state
->debugregs
;
1005 void kvm_setup_guest_memory(void *start
, size_t size
)
1007 if (!kvm_has_sync_mmu()) {
1008 #ifdef MADV_DONTFORK
1009 int ret
= madvise(start
, size
, MADV_DONTFORK
);
1017 "Need MADV_DONTFORK in absence of synchronous KVM MMU\n");
1023 #ifdef KVM_CAP_SET_GUEST_DEBUG
1024 static void on_vcpu(CPUState
*env
, void (*func
)(void *data
), void *data
)
1026 #ifdef CONFIG_IOTHREAD
1027 if (env
!= cpu_single_env
) {
1034 struct kvm_sw_breakpoint
*kvm_find_sw_breakpoint(CPUState
*env
,
1037 struct kvm_sw_breakpoint
*bp
;
1039 QTAILQ_FOREACH(bp
, &env
->kvm_state
->kvm_sw_breakpoints
, entry
) {
1046 int kvm_sw_breakpoints_active(CPUState
*env
)
1048 return !QTAILQ_EMPTY(&env
->kvm_state
->kvm_sw_breakpoints
);
1051 struct kvm_set_guest_debug_data
{
1052 struct kvm_guest_debug dbg
;
1057 static void kvm_invoke_set_guest_debug(void *data
)
1059 struct kvm_set_guest_debug_data
*dbg_data
= data
;
1060 CPUState
*env
= dbg_data
->env
;
1062 dbg_data
->err
= kvm_vcpu_ioctl(env
, KVM_SET_GUEST_DEBUG
, &dbg_data
->dbg
);
1065 int kvm_update_guest_debug(CPUState
*env
, unsigned long reinject_trap
)
1067 struct kvm_set_guest_debug_data data
;
1069 data
.dbg
.control
= reinject_trap
;
1071 if (env
->singlestep_enabled
) {
1072 data
.dbg
.control
|= KVM_GUESTDBG_ENABLE
| KVM_GUESTDBG_SINGLESTEP
;
1074 kvm_arch_update_guest_debug(env
, &data
.dbg
);
1077 on_vcpu(env
, kvm_invoke_set_guest_debug
, &data
);
1081 int kvm_insert_breakpoint(CPUState
*current_env
, target_ulong addr
,
1082 target_ulong len
, int type
)
1084 struct kvm_sw_breakpoint
*bp
;
1088 if (type
== GDB_BREAKPOINT_SW
) {
1089 bp
= kvm_find_sw_breakpoint(current_env
, addr
);
1095 bp
= qemu_malloc(sizeof(struct kvm_sw_breakpoint
));
1101 err
= kvm_arch_insert_sw_breakpoint(current_env
, bp
);
1107 QTAILQ_INSERT_HEAD(¤t_env
->kvm_state
->kvm_sw_breakpoints
,
1110 err
= kvm_arch_insert_hw_breakpoint(addr
, len
, type
);
1115 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
1116 err
= kvm_update_guest_debug(env
, 0);
1123 int kvm_remove_breakpoint(CPUState
*current_env
, target_ulong addr
,
1124 target_ulong len
, int type
)
1126 struct kvm_sw_breakpoint
*bp
;
1130 if (type
== GDB_BREAKPOINT_SW
) {
1131 bp
= kvm_find_sw_breakpoint(current_env
, addr
);
1135 if (bp
->use_count
> 1) {
1140 err
= kvm_arch_remove_sw_breakpoint(current_env
, bp
);
1144 QTAILQ_REMOVE(¤t_env
->kvm_state
->kvm_sw_breakpoints
, bp
, entry
);
1147 err
= kvm_arch_remove_hw_breakpoint(addr
, len
, type
);
1152 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
1153 err
= kvm_update_guest_debug(env
, 0);
1160 void kvm_remove_all_breakpoints(CPUState
*current_env
)
1162 struct kvm_sw_breakpoint
*bp
, *next
;
1163 KVMState
*s
= current_env
->kvm_state
;
1166 QTAILQ_FOREACH_SAFE(bp
, &s
->kvm_sw_breakpoints
, entry
, next
) {
1167 if (kvm_arch_remove_sw_breakpoint(current_env
, bp
) != 0) {
1168 /* Try harder to find a CPU that currently sees the breakpoint. */
1169 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
1170 if (kvm_arch_remove_sw_breakpoint(env
, bp
) == 0)
1175 kvm_arch_remove_all_hw_breakpoints();
1177 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
)
1178 kvm_update_guest_debug(env
, 0);
1181 #else /* !KVM_CAP_SET_GUEST_DEBUG */
1183 int kvm_update_guest_debug(CPUState
*env
, unsigned long reinject_trap
)
1188 int kvm_insert_breakpoint(CPUState
*current_env
, target_ulong addr
,
1189 target_ulong len
, int type
)
1194 int kvm_remove_breakpoint(CPUState
*current_env
, target_ulong addr
,
1195 target_ulong len
, int type
)
1200 void kvm_remove_all_breakpoints(CPUState
*current_env
)
1203 #endif /* !KVM_CAP_SET_GUEST_DEBUG */
1205 int kvm_set_signal_mask(CPUState
*env
, const sigset_t
*sigset
)
1207 struct kvm_signal_mask
*sigmask
;
1211 return kvm_vcpu_ioctl(env
, KVM_SET_SIGNAL_MASK
, NULL
);
1213 sigmask
= qemu_malloc(sizeof(*sigmask
) + sizeof(*sigset
));
1216 memcpy(sigmask
->sigset
, sigset
, sizeof(*sigset
));
1217 r
= kvm_vcpu_ioctl(env
, KVM_SET_SIGNAL_MASK
, sigmask
);
1223 int kvm_set_ioeventfd_pio_word(int fd
, uint16_t addr
, uint16_t val
, bool assign
)
1225 #ifdef KVM_IOEVENTFD
1226 struct kvm_ioeventfd kick
= {
1230 .flags
= KVM_IOEVENTFD_FLAG_DATAMATCH
| KVM_IOEVENTFD_FLAG_PIO
,
1237 kick
.flags
|= KVM_IOEVENTFD_FLAG_DEASSIGN
;
1238 r
= kvm_vm_ioctl(kvm_state
, KVM_IOEVENTFD
, &kick
);