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"
28 /* KVM uses PAGE_SIZE in it's definition of COALESCED_MMIO_MAX */
29 #define PAGE_SIZE TARGET_PAGE_SIZE
34 #define dprintf(fmt, ...) \
35 do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
37 #define dprintf(fmt, ...) \
41 typedef struct KVMSlot
43 target_phys_addr_t start_addr
;
44 ram_addr_t memory_size
;
45 ram_addr_t phys_offset
;
50 typedef struct kvm_dirty_log KVMDirtyLog
;
60 #ifdef KVM_CAP_SET_GUEST_DEBUG
61 struct kvm_sw_breakpoint_head kvm_sw_breakpoints
;
65 static KVMState
*kvm_state
;
67 static KVMSlot
*kvm_alloc_slot(KVMState
*s
)
71 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
72 /* KVM private memory slots */
75 if (s
->slots
[i
].memory_size
== 0)
82 static KVMSlot
*kvm_lookup_slot(KVMState
*s
, target_phys_addr_t start_addr
)
86 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
87 KVMSlot
*mem
= &s
->slots
[i
];
89 if (start_addr
>= mem
->start_addr
&&
90 start_addr
< (mem
->start_addr
+ mem
->memory_size
))
97 static int kvm_set_user_memory_region(KVMState
*s
, KVMSlot
*slot
)
99 struct kvm_userspace_memory_region mem
;
101 mem
.slot
= slot
->slot
;
102 mem
.guest_phys_addr
= slot
->start_addr
;
103 mem
.memory_size
= slot
->memory_size
;
104 mem
.userspace_addr
= (unsigned long)phys_ram_base
+ slot
->phys_offset
;
105 mem
.flags
= slot
->flags
;
107 return kvm_vm_ioctl(s
, KVM_SET_USER_MEMORY_REGION
, &mem
);
111 int kvm_init_vcpu(CPUState
*env
)
113 KVMState
*s
= kvm_state
;
117 dprintf("kvm_init_vcpu\n");
119 ret
= kvm_vm_ioctl(s
, KVM_CREATE_VCPU
, env
->cpu_index
);
121 dprintf("kvm_create_vcpu failed\n");
128 mmap_size
= kvm_ioctl(s
, KVM_GET_VCPU_MMAP_SIZE
, 0);
130 dprintf("KVM_GET_VCPU_MMAP_SIZE failed\n");
134 env
->kvm_run
= mmap(NULL
, mmap_size
, PROT_READ
| PROT_WRITE
, MAP_SHARED
,
136 if (env
->kvm_run
== MAP_FAILED
) {
138 dprintf("mmap'ing vcpu state failed\n");
142 ret
= kvm_arch_init_vcpu(env
);
148 int kvm_sync_vcpus(void)
152 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
155 ret
= kvm_arch_put_registers(env
);
164 * dirty pages logging control
166 static int kvm_dirty_pages_log_change(target_phys_addr_t phys_addr
, target_phys_addr_t end_addr
,
170 KVMState
*s
= kvm_state
;
171 KVMSlot
*mem
= kvm_lookup_slot(s
, phys_addr
);
173 dprintf("invalid parameters %llx-%llx\n", phys_addr
, end_addr
);
177 flags
= (mem
->flags
& ~mask
) | flags
;
178 /* Nothing changed, no need to issue ioctl */
179 if (flags
== mem
->flags
)
184 return kvm_set_user_memory_region(s
, mem
);
187 int kvm_log_start(target_phys_addr_t phys_addr
, target_phys_addr_t end_addr
)
189 return kvm_dirty_pages_log_change(phys_addr
, end_addr
,
190 KVM_MEM_LOG_DIRTY_PAGES
,
191 KVM_MEM_LOG_DIRTY_PAGES
);
194 int kvm_log_stop(target_phys_addr_t phys_addr
, target_phys_addr_t end_addr
)
196 return kvm_dirty_pages_log_change(phys_addr
, end_addr
,
198 KVM_MEM_LOG_DIRTY_PAGES
);
202 * kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space
203 * This function updates qemu's dirty bitmap using cpu_physical_memory_set_dirty().
204 * This means all bits are set to dirty.
206 * @start_add: start of logged region. This is what we use to search the memslot
207 * @end_addr: end of logged region.
209 void kvm_physical_sync_dirty_bitmap(target_phys_addr_t start_addr
, target_phys_addr_t end_addr
)
211 KVMState
*s
= kvm_state
;
213 KVMSlot
*mem
= kvm_lookup_slot(s
, start_addr
);
214 unsigned long alloc_size
;
216 target_phys_addr_t phys_addr
= start_addr
;
218 dprintf("sync addr: %llx into %lx\n", start_addr
, mem
->phys_offset
);
220 fprintf(stderr
, "BUG: %s: invalid parameters\n", __func__
);
224 alloc_size
= mem
->memory_size
>> TARGET_PAGE_BITS
/ sizeof(d
.dirty_bitmap
);
225 d
.dirty_bitmap
= qemu_mallocz(alloc_size
);
228 dprintf("slot %d, phys_addr %llx, uaddr: %llx\n",
229 d
.slot
, mem
->start_addr
, mem
->phys_offset
);
231 if (kvm_vm_ioctl(s
, KVM_GET_DIRTY_LOG
, &d
) == -1) {
232 dprintf("ioctl failed %d\n", errno
);
236 phys_addr
= start_addr
;
237 for (addr
= mem
->phys_offset
; phys_addr
< end_addr
; phys_addr
+= TARGET_PAGE_SIZE
, addr
+= TARGET_PAGE_SIZE
) {
238 unsigned long *bitmap
= (unsigned long *)d
.dirty_bitmap
;
239 unsigned nr
= (phys_addr
- start_addr
) >> TARGET_PAGE_BITS
;
240 unsigned word
= nr
/ (sizeof(*bitmap
) * 8);
241 unsigned bit
= nr
% (sizeof(*bitmap
) * 8);
242 if ((bitmap
[word
] >> bit
) & 1)
243 cpu_physical_memory_set_dirty(addr
);
246 qemu_free(d
.dirty_bitmap
);
249 int kvm_coalesce_mmio_region(target_phys_addr_t start
, ram_addr_t size
)
252 #ifdef KVM_CAP_COALESCED_MMIO
253 KVMState
*s
= kvm_state
;
255 if (s
->coalesced_mmio
) {
256 struct kvm_coalesced_mmio_zone zone
;
261 ret
= kvm_vm_ioctl(s
, KVM_REGISTER_COALESCED_MMIO
, &zone
);
268 int kvm_uncoalesce_mmio_region(target_phys_addr_t start
, ram_addr_t size
)
271 #ifdef KVM_CAP_COALESCED_MMIO
272 KVMState
*s
= kvm_state
;
274 if (s
->coalesced_mmio
) {
275 struct kvm_coalesced_mmio_zone zone
;
280 ret
= kvm_vm_ioctl(s
, KVM_UNREGISTER_COALESCED_MMIO
, &zone
);
287 int kvm_init(int smp_cpus
)
296 s
= qemu_mallocz(sizeof(KVMState
));
298 #ifdef KVM_CAP_SET_GUEST_DEBUG
299 TAILQ_INIT(&s
->kvm_sw_breakpoints
);
301 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++)
302 s
->slots
[i
].slot
= i
;
305 s
->fd
= open("/dev/kvm", O_RDWR
);
307 fprintf(stderr
, "Could not access KVM kernel module: %m\n");
312 ret
= kvm_ioctl(s
, KVM_GET_API_VERSION
, 0);
313 if (ret
< KVM_API_VERSION
) {
316 fprintf(stderr
, "kvm version too old\n");
320 if (ret
> KVM_API_VERSION
) {
322 fprintf(stderr
, "kvm version not supported\n");
326 s
->vmfd
= kvm_ioctl(s
, KVM_CREATE_VM
, 0);
330 /* initially, KVM allocated its own memory and we had to jump through
331 * hooks to make phys_ram_base point to this. Modern versions of KVM
332 * just use a user allocated buffer so we can use phys_ram_base
333 * unmodified. Make sure we have a sufficiently modern version of KVM.
335 ret
= kvm_ioctl(s
, KVM_CHECK_EXTENSION
, KVM_CAP_USER_MEMORY
);
339 fprintf(stderr
, "kvm does not support KVM_CAP_USER_MEMORY\n");
343 /* There was a nasty bug in < kvm-80 that prevents memory slots from being
344 * destroyed properly. Since we rely on this capability, refuse to work
345 * with any kernel without this capability. */
346 ret
= kvm_ioctl(s
, KVM_CHECK_EXTENSION
,
347 KVM_CAP_DESTROY_MEMORY_REGION_WORKS
);
353 "KVM kernel module broken (DESTROY_MEMORY_REGION)\n"
354 "Please upgrade to at least kvm-81.\n");
358 s
->coalesced_mmio
= 0;
359 #ifdef KVM_CAP_COALESCED_MMIO
360 ret
= kvm_ioctl(s
, KVM_CHECK_EXTENSION
, KVM_CAP_COALESCED_MMIO
);
362 s
->coalesced_mmio
= ret
;
365 ret
= kvm_arch_init(s
, smp_cpus
);
385 static int kvm_handle_io(CPUState
*env
, uint16_t port
, void *data
,
386 int direction
, int size
, uint32_t count
)
391 for (i
= 0; i
< count
; i
++) {
392 if (direction
== KVM_EXIT_IO_IN
) {
395 stb_p(ptr
, cpu_inb(env
, port
));
398 stw_p(ptr
, cpu_inw(env
, port
));
401 stl_p(ptr
, cpu_inl(env
, port
));
407 cpu_outb(env
, port
, ldub_p(ptr
));
410 cpu_outw(env
, port
, lduw_p(ptr
));
413 cpu_outl(env
, port
, ldl_p(ptr
));
424 static void kvm_run_coalesced_mmio(CPUState
*env
, struct kvm_run
*run
)
426 #ifdef KVM_CAP_COALESCED_MMIO
427 KVMState
*s
= kvm_state
;
428 if (s
->coalesced_mmio
) {
429 struct kvm_coalesced_mmio_ring
*ring
;
431 ring
= (void *)run
+ (s
->coalesced_mmio
* TARGET_PAGE_SIZE
);
432 while (ring
->first
!= ring
->last
) {
433 struct kvm_coalesced_mmio
*ent
;
435 ent
= &ring
->coalesced_mmio
[ring
->first
];
437 cpu_physical_memory_write(ent
->phys_addr
, ent
->data
, ent
->len
);
438 /* FIXME smp_wmb() */
439 ring
->first
= (ring
->first
+ 1) % KVM_COALESCED_MMIO_MAX
;
445 int kvm_cpu_exec(CPUState
*env
)
447 struct kvm_run
*run
= env
->kvm_run
;
450 dprintf("kvm_cpu_exec()\n");
453 kvm_arch_pre_run(env
, run
);
455 if (env
->exit_request
) {
456 dprintf("interrupt exit requested\n");
461 ret
= kvm_vcpu_ioctl(env
, KVM_RUN
, 0);
462 kvm_arch_post_run(env
, run
);
464 if (ret
== -EINTR
|| ret
== -EAGAIN
) {
465 dprintf("io window exit\n");
471 dprintf("kvm run failed %s\n", strerror(-ret
));
475 kvm_run_coalesced_mmio(env
, run
);
477 ret
= 0; /* exit loop */
478 switch (run
->exit_reason
) {
480 dprintf("handle_io\n");
481 ret
= kvm_handle_io(env
, run
->io
.port
,
482 (uint8_t *)run
+ run
->io
.data_offset
,
488 dprintf("handle_mmio\n");
489 cpu_physical_memory_rw(run
->mmio
.phys_addr
,
495 case KVM_EXIT_IRQ_WINDOW_OPEN
:
496 dprintf("irq_window_open\n");
498 case KVM_EXIT_SHUTDOWN
:
499 dprintf("shutdown\n");
500 qemu_system_reset_request();
503 case KVM_EXIT_UNKNOWN
:
504 dprintf("kvm_exit_unknown\n");
506 case KVM_EXIT_FAIL_ENTRY
:
507 dprintf("kvm_exit_fail_entry\n");
509 case KVM_EXIT_EXCEPTION
:
510 dprintf("kvm_exit_exception\n");
513 dprintf("kvm_exit_debug\n");
514 #ifdef KVM_CAP_SET_GUEST_DEBUG
515 if (kvm_arch_debug(&run
->debug
.arch
)) {
516 gdb_set_stop_cpu(env
);
518 env
->exception_index
= EXCP_DEBUG
;
521 /* re-enter, this exception was guest-internal */
523 #endif /* KVM_CAP_SET_GUEST_DEBUG */
526 dprintf("kvm_arch_handle_exit\n");
527 ret
= kvm_arch_handle_exit(env
, run
);
532 if (env
->exit_request
) {
533 env
->exit_request
= 0;
534 env
->exception_index
= EXCP_INTERRUPT
;
540 void kvm_set_phys_mem(target_phys_addr_t start_addr
,
542 ram_addr_t phys_offset
)
544 KVMState
*s
= kvm_state
;
545 ram_addr_t flags
= phys_offset
& ~TARGET_PAGE_MASK
;
548 /* KVM does not support read-only slots */
549 phys_offset
&= ~IO_MEM_ROM
;
551 mem
= kvm_lookup_slot(s
, start_addr
);
553 if ((flags
== IO_MEM_UNASSIGNED
) || (flags
>= TLB_MMIO
)) {
554 mem
->memory_size
= 0;
555 mem
->start_addr
= start_addr
;
556 mem
->phys_offset
= 0;
559 kvm_set_user_memory_region(s
, mem
);
560 } else if (start_addr
>= mem
->start_addr
&&
561 (start_addr
+ size
) <= (mem
->start_addr
+
564 target_phys_addr_t mem_start
;
565 ram_addr_t mem_size
, mem_offset
;
568 if ((phys_offset
- (start_addr
- mem
->start_addr
)) ==
572 /* unregister whole slot */
573 memcpy(&slot
, mem
, sizeof(slot
));
574 mem
->memory_size
= 0;
575 kvm_set_user_memory_region(s
, mem
);
577 /* register prefix slot */
578 mem_start
= slot
.start_addr
;
579 mem_size
= start_addr
- slot
.start_addr
;
580 mem_offset
= slot
.phys_offset
;
582 kvm_set_phys_mem(mem_start
, mem_size
, mem_offset
);
584 /* register new slot */
585 kvm_set_phys_mem(start_addr
, size
, phys_offset
);
587 /* register suffix slot */
588 mem_start
= start_addr
+ size
;
589 mem_offset
+= mem_size
+ size
;
590 mem_size
= slot
.memory_size
- mem_size
- size
;
592 kvm_set_phys_mem(mem_start
, mem_size
, mem_offset
);
596 printf("Registering overlapping slot\n");
600 /* KVM does not need to know about this memory */
601 if (flags
>= IO_MEM_UNASSIGNED
)
604 mem
= kvm_alloc_slot(s
);
605 mem
->memory_size
= size
;
606 mem
->start_addr
= start_addr
;
607 mem
->phys_offset
= phys_offset
;
610 kvm_set_user_memory_region(s
, mem
);
611 /* FIXME deal with errors */
614 int kvm_ioctl(KVMState
*s
, int type
, ...)
621 arg
= va_arg(ap
, void *);
624 ret
= ioctl(s
->fd
, type
, arg
);
631 int kvm_vm_ioctl(KVMState
*s
, int type
, ...)
638 arg
= va_arg(ap
, void *);
641 ret
= ioctl(s
->vmfd
, type
, arg
);
648 int kvm_vcpu_ioctl(CPUState
*env
, int type
, ...)
655 arg
= va_arg(ap
, void *);
658 ret
= ioctl(env
->kvm_fd
, type
, arg
);
665 int kvm_has_sync_mmu(void)
667 #ifdef KVM_CAP_SYNC_MMU
668 KVMState
*s
= kvm_state
;
670 if (kvm_ioctl(s
, KVM_CHECK_EXTENSION
, KVM_CAP_SYNC_MMU
) > 0)
677 #ifdef KVM_CAP_SET_GUEST_DEBUG
678 struct kvm_sw_breakpoint
*kvm_find_sw_breakpoint(CPUState
*env
,
681 struct kvm_sw_breakpoint
*bp
;
683 TAILQ_FOREACH(bp
, &env
->kvm_state
->kvm_sw_breakpoints
, entry
) {
690 int kvm_sw_breakpoints_active(CPUState
*env
)
692 return !TAILQ_EMPTY(&env
->kvm_state
->kvm_sw_breakpoints
);
695 int kvm_update_guest_debug(CPUState
*env
, unsigned long reinject_trap
)
697 struct kvm_guest_debug dbg
;
700 if (env
->singlestep_enabled
)
701 dbg
.control
= KVM_GUESTDBG_ENABLE
| KVM_GUESTDBG_SINGLESTEP
;
703 kvm_arch_update_guest_debug(env
, &dbg
);
704 dbg
.control
|= reinject_trap
;
706 return kvm_vcpu_ioctl(env
, KVM_SET_GUEST_DEBUG
, &dbg
);
709 int kvm_insert_breakpoint(CPUState
*current_env
, target_ulong addr
,
710 target_ulong len
, int type
)
712 struct kvm_sw_breakpoint
*bp
;
716 if (type
== GDB_BREAKPOINT_SW
) {
717 bp
= kvm_find_sw_breakpoint(current_env
, addr
);
723 bp
= qemu_malloc(sizeof(struct kvm_sw_breakpoint
));
729 err
= kvm_arch_insert_sw_breakpoint(current_env
, bp
);
735 TAILQ_INSERT_HEAD(¤t_env
->kvm_state
->kvm_sw_breakpoints
,
738 err
= kvm_arch_insert_hw_breakpoint(addr
, len
, type
);
743 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
744 err
= kvm_update_guest_debug(env
, 0);
751 int kvm_remove_breakpoint(CPUState
*current_env
, target_ulong addr
,
752 target_ulong len
, int type
)
754 struct kvm_sw_breakpoint
*bp
;
758 if (type
== GDB_BREAKPOINT_SW
) {
759 bp
= kvm_find_sw_breakpoint(current_env
, addr
);
763 if (bp
->use_count
> 1) {
768 err
= kvm_arch_remove_sw_breakpoint(current_env
, bp
);
772 TAILQ_REMOVE(¤t_env
->kvm_state
->kvm_sw_breakpoints
, bp
, entry
);
775 err
= kvm_arch_remove_hw_breakpoint(addr
, len
, type
);
780 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
781 err
= kvm_update_guest_debug(env
, 0);
788 void kvm_remove_all_breakpoints(CPUState
*current_env
)
790 struct kvm_sw_breakpoint
*bp
, *next
;
791 KVMState
*s
= current_env
->kvm_state
;
794 TAILQ_FOREACH_SAFE(bp
, &s
->kvm_sw_breakpoints
, entry
, next
) {
795 if (kvm_arch_remove_sw_breakpoint(current_env
, bp
) != 0) {
796 /* Try harder to find a CPU that currently sees the breakpoint. */
797 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
798 if (kvm_arch_remove_sw_breakpoint(env
, bp
) == 0)
803 kvm_arch_remove_all_hw_breakpoints();
805 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
)
806 kvm_update_guest_debug(env
, 0);
809 #else /* !KVM_CAP_SET_GUEST_DEBUG */
811 int kvm_update_guest_debug(CPUState
*env
, unsigned long reinject_trap
)
816 int kvm_insert_breakpoint(CPUState
*current_env
, target_ulong addr
,
817 target_ulong len
, int type
)
822 int kvm_remove_breakpoint(CPUState
*current_env
, target_ulong addr
,
823 target_ulong len
, int type
)
828 void kvm_remove_all_breakpoints(CPUState
*current_env
)
831 #endif /* !KVM_CAP_SET_GUEST_DEBUG */