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"
32 /* This check must be after config-host.h is included */
34 #include <sys/eventfd.h>
37 /* KVM uses PAGE_SIZE in it's definition of COALESCED_MMIO_MAX */
38 #define PAGE_SIZE TARGET_PAGE_SIZE
43 #define DPRINTF(fmt, ...) \
44 do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
46 #define DPRINTF(fmt, ...) \
50 typedef struct KVMSlot
52 target_phys_addr_t start_addr
;
53 ram_addr_t memory_size
;
59 typedef struct kvm_dirty_log KVMDirtyLog
;
67 struct kvm_coalesced_mmio_ring
*coalesced_mmio_ring
;
68 bool coalesced_flush_in_progress
;
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
;
83 int irqchip_inject_ioctl
;
84 #ifdef KVM_CAP_IRQ_ROUTING
85 struct kvm_irq_routing
*irq_routes
;
86 int nr_allocated_irq_routes
;
88 void *used_gsi_bitmap
;
94 static const KVMCapabilityInfo kvm_required_capabilites
[] = {
95 KVM_CAP_INFO(USER_MEMORY
),
96 KVM_CAP_INFO(DESTROY_MEMORY_REGION_WORKS
),
100 static KVMSlot
*kvm_alloc_slot(KVMState
*s
)
104 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
105 if (s
->slots
[i
].memory_size
== 0) {
110 fprintf(stderr
, "%s: no free slot available\n", __func__
);
114 static KVMSlot
*kvm_lookup_matching_slot(KVMState
*s
,
115 target_phys_addr_t start_addr
,
116 target_phys_addr_t end_addr
)
120 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
121 KVMSlot
*mem
= &s
->slots
[i
];
123 if (start_addr
== mem
->start_addr
&&
124 end_addr
== mem
->start_addr
+ mem
->memory_size
) {
133 * Find overlapping slot with lowest start address
135 static KVMSlot
*kvm_lookup_overlapping_slot(KVMState
*s
,
136 target_phys_addr_t start_addr
,
137 target_phys_addr_t end_addr
)
139 KVMSlot
*found
= NULL
;
142 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
143 KVMSlot
*mem
= &s
->slots
[i
];
145 if (mem
->memory_size
== 0 ||
146 (found
&& found
->start_addr
< mem
->start_addr
)) {
150 if (end_addr
> mem
->start_addr
&&
151 start_addr
< mem
->start_addr
+ mem
->memory_size
) {
159 int kvm_physical_memory_addr_from_host(KVMState
*s
, void *ram
,
160 target_phys_addr_t
*phys_addr
)
164 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
165 KVMSlot
*mem
= &s
->slots
[i
];
167 if (ram
>= mem
->ram
&& ram
< mem
->ram
+ mem
->memory_size
) {
168 *phys_addr
= mem
->start_addr
+ (ram
- mem
->ram
);
176 static int kvm_set_user_memory_region(KVMState
*s
, KVMSlot
*slot
)
178 struct kvm_userspace_memory_region mem
;
180 mem
.slot
= slot
->slot
;
181 mem
.guest_phys_addr
= slot
->start_addr
;
182 mem
.memory_size
= slot
->memory_size
;
183 mem
.userspace_addr
= (unsigned long)slot
->ram
;
184 mem
.flags
= slot
->flags
;
185 if (s
->migration_log
) {
186 mem
.flags
|= KVM_MEM_LOG_DIRTY_PAGES
;
188 return kvm_vm_ioctl(s
, KVM_SET_USER_MEMORY_REGION
, &mem
);
191 static void kvm_reset_vcpu(void *opaque
)
193 CPUState
*env
= opaque
;
195 kvm_arch_reset_vcpu(env
);
198 int kvm_irqchip_in_kernel(void)
200 return kvm_state
->irqchip_in_kernel
;
203 int kvm_pit_in_kernel(void)
205 return kvm_state
->pit_in_kernel
;
208 int kvm_init_vcpu(CPUState
*env
)
210 KVMState
*s
= kvm_state
;
214 DPRINTF("kvm_init_vcpu\n");
216 ret
= kvm_vm_ioctl(s
, KVM_CREATE_VCPU
, env
->cpu_index
);
218 DPRINTF("kvm_create_vcpu failed\n");
224 env
->kvm_vcpu_dirty
= 1;
226 mmap_size
= kvm_ioctl(s
, KVM_GET_VCPU_MMAP_SIZE
, 0);
229 DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n");
233 env
->kvm_run
= mmap(NULL
, mmap_size
, PROT_READ
| PROT_WRITE
, MAP_SHARED
,
235 if (env
->kvm_run
== MAP_FAILED
) {
237 DPRINTF("mmap'ing vcpu state failed\n");
241 if (s
->coalesced_mmio
&& !s
->coalesced_mmio_ring
) {
242 s
->coalesced_mmio_ring
=
243 (void *)env
->kvm_run
+ s
->coalesced_mmio
* PAGE_SIZE
;
246 ret
= kvm_arch_init_vcpu(env
);
248 qemu_register_reset(kvm_reset_vcpu
, env
);
249 kvm_arch_reset_vcpu(env
);
256 * dirty pages logging control
259 static int kvm_mem_flags(KVMState
*s
, bool log_dirty
)
261 return log_dirty
? KVM_MEM_LOG_DIRTY_PAGES
: 0;
264 static int kvm_slot_dirty_pages_log_change(KVMSlot
*mem
, bool log_dirty
)
266 KVMState
*s
= kvm_state
;
267 int flags
, mask
= KVM_MEM_LOG_DIRTY_PAGES
;
270 old_flags
= mem
->flags
;
272 flags
= (mem
->flags
& ~mask
) | kvm_mem_flags(s
, log_dirty
);
275 /* If nothing changed effectively, no need to issue ioctl */
276 if (s
->migration_log
) {
277 flags
|= KVM_MEM_LOG_DIRTY_PAGES
;
280 if (flags
== old_flags
) {
284 return kvm_set_user_memory_region(s
, mem
);
287 static int kvm_dirty_pages_log_change(target_phys_addr_t phys_addr
,
288 ram_addr_t size
, bool log_dirty
)
290 KVMState
*s
= kvm_state
;
291 KVMSlot
*mem
= kvm_lookup_matching_slot(s
, phys_addr
, phys_addr
+ size
);
294 fprintf(stderr
, "BUG: %s: invalid parameters " TARGET_FMT_plx
"-"
295 TARGET_FMT_plx
"\n", __func__
, phys_addr
,
296 (target_phys_addr_t
)(phys_addr
+ size
- 1));
299 return kvm_slot_dirty_pages_log_change(mem
, log_dirty
);
302 static void kvm_log_start(MemoryListener
*listener
,
303 MemoryRegionSection
*section
)
307 r
= kvm_dirty_pages_log_change(section
->offset_within_address_space
,
308 section
->size
, true);
314 static void kvm_log_stop(MemoryListener
*listener
,
315 MemoryRegionSection
*section
)
319 r
= kvm_dirty_pages_log_change(section
->offset_within_address_space
,
320 section
->size
, false);
326 static int kvm_set_migration_log(int enable
)
328 KVMState
*s
= kvm_state
;
332 s
->migration_log
= enable
;
334 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
337 if (!mem
->memory_size
) {
340 if (!!(mem
->flags
& KVM_MEM_LOG_DIRTY_PAGES
) == enable
) {
343 err
= kvm_set_user_memory_region(s
, mem
);
351 /* get kvm's dirty pages bitmap and update qemu's */
352 static int kvm_get_dirty_pages_log_range(MemoryRegionSection
*section
,
353 unsigned long *bitmap
)
356 unsigned long page_number
, addr
, addr1
, c
;
357 unsigned int len
= ((section
->size
/ TARGET_PAGE_SIZE
) + HOST_LONG_BITS
- 1) / HOST_LONG_BITS
;
360 * bitmap-traveling is faster than memory-traveling (for addr...)
361 * especially when most of the memory is not dirty.
363 for (i
= 0; i
< len
; i
++) {
364 if (bitmap
[i
] != 0) {
365 c
= leul_to_cpu(bitmap
[i
]);
369 page_number
= i
* HOST_LONG_BITS
+ j
;
370 addr1
= page_number
* TARGET_PAGE_SIZE
;
371 addr
= section
->offset_within_region
+ addr1
;
372 memory_region_set_dirty(section
->mr
, addr
);
379 #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1))
382 * kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space
383 * This function updates qemu's dirty bitmap using cpu_physical_memory_set_dirty().
384 * This means all bits are set to dirty.
386 * @start_add: start of logged region.
387 * @end_addr: end of logged region.
389 static int kvm_physical_sync_dirty_bitmap(MemoryRegionSection
*section
)
391 KVMState
*s
= kvm_state
;
392 unsigned long size
, allocated_size
= 0;
396 target_phys_addr_t start_addr
= section
->offset_within_address_space
;
397 target_phys_addr_t end_addr
= start_addr
+ section
->size
;
399 d
.dirty_bitmap
= NULL
;
400 while (start_addr
< end_addr
) {
401 mem
= kvm_lookup_overlapping_slot(s
, start_addr
, end_addr
);
406 /* XXX bad kernel interface alert
407 * For dirty bitmap, kernel allocates array of size aligned to
408 * bits-per-long. But for case when the kernel is 64bits and
409 * the userspace is 32bits, userspace can't align to the same
410 * bits-per-long, since sizeof(long) is different between kernel
411 * and user space. This way, userspace will provide buffer which
412 * may be 4 bytes less than the kernel will use, resulting in
413 * userspace memory corruption (which is not detectable by valgrind
414 * too, in most cases).
415 * So for now, let's align to 64 instead of HOST_LONG_BITS here, in
416 * a hope that sizeof(long) wont become >8 any time soon.
418 size
= ALIGN(((mem
->memory_size
) >> TARGET_PAGE_BITS
),
419 /*HOST_LONG_BITS*/ 64) / 8;
420 if (!d
.dirty_bitmap
) {
421 d
.dirty_bitmap
= g_malloc(size
);
422 } else if (size
> allocated_size
) {
423 d
.dirty_bitmap
= g_realloc(d
.dirty_bitmap
, size
);
425 allocated_size
= size
;
426 memset(d
.dirty_bitmap
, 0, allocated_size
);
430 if (kvm_vm_ioctl(s
, KVM_GET_DIRTY_LOG
, &d
) == -1) {
431 DPRINTF("ioctl failed %d\n", errno
);
436 kvm_get_dirty_pages_log_range(section
, d
.dirty_bitmap
);
437 start_addr
= mem
->start_addr
+ mem
->memory_size
;
439 g_free(d
.dirty_bitmap
);
444 int kvm_coalesce_mmio_region(target_phys_addr_t start
, ram_addr_t size
)
447 KVMState
*s
= kvm_state
;
449 if (s
->coalesced_mmio
) {
450 struct kvm_coalesced_mmio_zone zone
;
455 ret
= kvm_vm_ioctl(s
, KVM_REGISTER_COALESCED_MMIO
, &zone
);
461 int kvm_uncoalesce_mmio_region(target_phys_addr_t start
, ram_addr_t size
)
464 KVMState
*s
= kvm_state
;
466 if (s
->coalesced_mmio
) {
467 struct kvm_coalesced_mmio_zone zone
;
472 ret
= kvm_vm_ioctl(s
, KVM_UNREGISTER_COALESCED_MMIO
, &zone
);
478 int kvm_check_extension(KVMState
*s
, unsigned int extension
)
482 ret
= kvm_ioctl(s
, KVM_CHECK_EXTENSION
, extension
);
490 static int kvm_check_many_ioeventfds(void)
492 /* Userspace can use ioeventfd for io notification. This requires a host
493 * that supports eventfd(2) and an I/O thread; since eventfd does not
494 * support SIGIO it cannot interrupt the vcpu.
496 * Older kernels have a 6 device limit on the KVM io bus. Find out so we
497 * can avoid creating too many ioeventfds.
499 #if defined(CONFIG_EVENTFD)
502 for (i
= 0; i
< ARRAY_SIZE(ioeventfds
); i
++) {
503 ioeventfds
[i
] = eventfd(0, EFD_CLOEXEC
);
504 if (ioeventfds
[i
] < 0) {
507 ret
= kvm_set_ioeventfd_pio_word(ioeventfds
[i
], 0, i
, true);
509 close(ioeventfds
[i
]);
514 /* Decide whether many devices are supported or not */
515 ret
= i
== ARRAY_SIZE(ioeventfds
);
518 kvm_set_ioeventfd_pio_word(ioeventfds
[i
], 0, i
, false);
519 close(ioeventfds
[i
]);
527 static const KVMCapabilityInfo
*
528 kvm_check_extension_list(KVMState
*s
, const KVMCapabilityInfo
*list
)
531 if (!kvm_check_extension(s
, list
->value
)) {
539 static void kvm_set_phys_mem(MemoryRegionSection
*section
, bool add
)
541 KVMState
*s
= kvm_state
;
544 MemoryRegion
*mr
= section
->mr
;
545 bool log_dirty
= memory_region_is_logging(mr
);
546 target_phys_addr_t start_addr
= section
->offset_within_address_space
;
547 ram_addr_t size
= section
->size
;
550 /* kvm works in page size chunks, but the function may be called
551 with sub-page size and unaligned start address. */
552 size
= TARGET_PAGE_ALIGN(size
);
553 start_addr
= TARGET_PAGE_ALIGN(start_addr
);
555 if (!memory_region_is_ram(mr
)) {
559 ram
= memory_region_get_ram_ptr(mr
) + section
->offset_within_region
;
562 mem
= kvm_lookup_overlapping_slot(s
, start_addr
, start_addr
+ size
);
567 if (add
&& start_addr
>= mem
->start_addr
&&
568 (start_addr
+ size
<= mem
->start_addr
+ mem
->memory_size
) &&
569 (ram
- start_addr
== mem
->ram
- mem
->start_addr
)) {
570 /* The new slot fits into the existing one and comes with
571 * identical parameters - update flags and done. */
572 kvm_slot_dirty_pages_log_change(mem
, log_dirty
);
578 /* unregister the overlapping slot */
579 mem
->memory_size
= 0;
580 err
= kvm_set_user_memory_region(s
, mem
);
582 fprintf(stderr
, "%s: error unregistering overlapping slot: %s\n",
583 __func__
, strerror(-err
));
587 /* Workaround for older KVM versions: we can't join slots, even not by
588 * unregistering the previous ones and then registering the larger
589 * slot. We have to maintain the existing fragmentation. Sigh.
591 * This workaround assumes that the new slot starts at the same
592 * address as the first existing one. If not or if some overlapping
593 * slot comes around later, we will fail (not seen in practice so far)
594 * - and actually require a recent KVM version. */
595 if (s
->broken_set_mem_region
&&
596 old
.start_addr
== start_addr
&& old
.memory_size
< size
&& add
) {
597 mem
= kvm_alloc_slot(s
);
598 mem
->memory_size
= old
.memory_size
;
599 mem
->start_addr
= old
.start_addr
;
601 mem
->flags
= kvm_mem_flags(s
, log_dirty
);
603 err
= kvm_set_user_memory_region(s
, mem
);
605 fprintf(stderr
, "%s: error updating slot: %s\n", __func__
,
610 start_addr
+= old
.memory_size
;
611 ram
+= old
.memory_size
;
612 size
-= old
.memory_size
;
616 /* register prefix slot */
617 if (old
.start_addr
< start_addr
) {
618 mem
= kvm_alloc_slot(s
);
619 mem
->memory_size
= start_addr
- old
.start_addr
;
620 mem
->start_addr
= old
.start_addr
;
622 mem
->flags
= kvm_mem_flags(s
, log_dirty
);
624 err
= kvm_set_user_memory_region(s
, mem
);
626 fprintf(stderr
, "%s: error registering prefix slot: %s\n",
627 __func__
, strerror(-err
));
629 fprintf(stderr
, "%s: This is probably because your kernel's " \
630 "PAGE_SIZE is too big. Please try to use 4k " \
631 "PAGE_SIZE!\n", __func__
);
637 /* register suffix slot */
638 if (old
.start_addr
+ old
.memory_size
> start_addr
+ size
) {
639 ram_addr_t size_delta
;
641 mem
= kvm_alloc_slot(s
);
642 mem
->start_addr
= start_addr
+ size
;
643 size_delta
= mem
->start_addr
- old
.start_addr
;
644 mem
->memory_size
= old
.memory_size
- size_delta
;
645 mem
->ram
= old
.ram
+ size_delta
;
646 mem
->flags
= kvm_mem_flags(s
, log_dirty
);
648 err
= kvm_set_user_memory_region(s
, mem
);
650 fprintf(stderr
, "%s: error registering suffix slot: %s\n",
651 __func__
, strerror(-err
));
657 /* in case the KVM bug workaround already "consumed" the new slot */
664 mem
= kvm_alloc_slot(s
);
665 mem
->memory_size
= size
;
666 mem
->start_addr
= start_addr
;
668 mem
->flags
= kvm_mem_flags(s
, log_dirty
);
670 err
= kvm_set_user_memory_region(s
, mem
);
672 fprintf(stderr
, "%s: error registering slot: %s\n", __func__
,
678 static void kvm_region_add(MemoryListener
*listener
,
679 MemoryRegionSection
*section
)
681 kvm_set_phys_mem(section
, true);
684 static void kvm_region_del(MemoryListener
*listener
,
685 MemoryRegionSection
*section
)
687 kvm_set_phys_mem(section
, false);
690 static void kvm_log_sync(MemoryListener
*listener
,
691 MemoryRegionSection
*section
)
695 r
= kvm_physical_sync_dirty_bitmap(section
);
701 static void kvm_log_global_start(struct MemoryListener
*listener
)
705 r
= kvm_set_migration_log(1);
709 static void kvm_log_global_stop(struct MemoryListener
*listener
)
713 r
= kvm_set_migration_log(0);
717 static MemoryListener kvm_memory_listener
= {
718 .region_add
= kvm_region_add
,
719 .region_del
= kvm_region_del
,
720 .log_start
= kvm_log_start
,
721 .log_stop
= kvm_log_stop
,
722 .log_sync
= kvm_log_sync
,
723 .log_global_start
= kvm_log_global_start
,
724 .log_global_stop
= kvm_log_global_stop
,
727 static void kvm_handle_interrupt(CPUState
*env
, int mask
)
729 env
->interrupt_request
|= mask
;
731 if (!qemu_cpu_is_self(env
)) {
738 static const char upgrade_note
[] =
739 "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n"
740 "(see http://sourceforge.net/projects/kvm).\n";
742 const KVMCapabilityInfo
*missing_cap
;
746 s
= g_malloc0(sizeof(KVMState
));
748 #ifdef KVM_CAP_SET_GUEST_DEBUG
749 QTAILQ_INIT(&s
->kvm_sw_breakpoints
);
751 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
752 s
->slots
[i
].slot
= i
;
755 s
->fd
= qemu_open("/dev/kvm", O_RDWR
);
757 fprintf(stderr
, "Could not access KVM kernel module: %m\n");
762 ret
= kvm_ioctl(s
, KVM_GET_API_VERSION
, 0);
763 if (ret
< KVM_API_VERSION
) {
767 fprintf(stderr
, "kvm version too old\n");
771 if (ret
> KVM_API_VERSION
) {
773 fprintf(stderr
, "kvm version not supported\n");
777 s
->vmfd
= kvm_ioctl(s
, KVM_CREATE_VM
, 0);
780 fprintf(stderr
, "Please add the 'switch_amode' kernel parameter to "
781 "your host kernel command line\n");
787 missing_cap
= kvm_check_extension_list(s
, kvm_required_capabilites
);
790 kvm_check_extension_list(s
, kvm_arch_required_capabilities
);
794 fprintf(stderr
, "kvm does not support %s\n%s",
795 missing_cap
->name
, upgrade_note
);
799 s
->coalesced_mmio
= kvm_check_extension(s
, KVM_CAP_COALESCED_MMIO
);
801 s
->broken_set_mem_region
= 1;
802 ret
= kvm_check_extension(s
, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
);
804 s
->broken_set_mem_region
= 0;
807 #ifdef KVM_CAP_VCPU_EVENTS
808 s
->vcpu_events
= kvm_check_extension(s
, KVM_CAP_VCPU_EVENTS
);
811 s
->robust_singlestep
=
812 kvm_check_extension(s
, KVM_CAP_X86_ROBUST_SINGLESTEP
);
814 #ifdef KVM_CAP_DEBUGREGS
815 s
->debugregs
= kvm_check_extension(s
, KVM_CAP_DEBUGREGS
);
819 s
->xsave
= kvm_check_extension(s
, KVM_CAP_XSAVE
);
823 s
->xcrs
= kvm_check_extension(s
, KVM_CAP_XCRS
);
827 #ifdef KVM_CAP_PIT_STATE2
828 s
->pit_state2
= kvm_check_extension(s
, KVM_CAP_PIT_STATE2
);
831 s
->pit_in_kernel
= kvm_pit
;
833 ret
= kvm_arch_init(s
);
839 memory_listener_register(&kvm_memory_listener
);
841 s
->many_ioeventfds
= kvm_check_many_ioeventfds();
843 ret
= kvm_create_irqchip(s
);
848 cpu_interrupt_handler
= kvm_handle_interrupt
;
866 static void kvm_handle_io(uint16_t port
, void *data
, int direction
, int size
,
872 for (i
= 0; i
< count
; i
++) {
873 if (direction
== KVM_EXIT_IO_IN
) {
876 stb_p(ptr
, cpu_inb(port
));
879 stw_p(ptr
, cpu_inw(port
));
882 stl_p(ptr
, cpu_inl(port
));
888 cpu_outb(port
, ldub_p(ptr
));
891 cpu_outw(port
, lduw_p(ptr
));
894 cpu_outl(port
, ldl_p(ptr
));
903 static int kvm_handle_internal_error(CPUState
*env
, struct kvm_run
*run
)
905 fprintf(stderr
, "KVM internal error.");
906 if (kvm_check_extension(kvm_state
, KVM_CAP_INTERNAL_ERROR_DATA
)) {
909 fprintf(stderr
, " Suberror: %d\n", run
->internal
.suberror
);
910 for (i
= 0; i
< run
->internal
.ndata
; ++i
) {
911 fprintf(stderr
, "extra data[%d]: %"PRIx64
"\n",
912 i
, (uint64_t)run
->internal
.data
[i
]);
915 fprintf(stderr
, "\n");
917 if (run
->internal
.suberror
== KVM_INTERNAL_ERROR_EMULATION
) {
918 fprintf(stderr
, "emulation failure\n");
919 if (!kvm_arch_stop_on_emulation_error(env
)) {
920 cpu_dump_state(env
, stderr
, fprintf
, CPU_DUMP_CODE
);
921 return EXCP_INTERRUPT
;
924 /* FIXME: Should trigger a qmp message to let management know
925 * something went wrong.
930 void kvm_flush_coalesced_mmio_buffer(void)
932 KVMState
*s
= kvm_state
;
934 if (s
->coalesced_flush_in_progress
) {
938 s
->coalesced_flush_in_progress
= true;
940 if (s
->coalesced_mmio_ring
) {
941 struct kvm_coalesced_mmio_ring
*ring
= s
->coalesced_mmio_ring
;
942 while (ring
->first
!= ring
->last
) {
943 struct kvm_coalesced_mmio
*ent
;
945 ent
= &ring
->coalesced_mmio
[ring
->first
];
947 cpu_physical_memory_write(ent
->phys_addr
, ent
->data
, ent
->len
);
949 ring
->first
= (ring
->first
+ 1) % KVM_COALESCED_MMIO_MAX
;
953 s
->coalesced_flush_in_progress
= false;
956 static void do_kvm_cpu_synchronize_state(void *_env
)
958 CPUState
*env
= _env
;
960 if (!env
->kvm_vcpu_dirty
) {
961 kvm_arch_get_registers(env
);
962 env
->kvm_vcpu_dirty
= 1;
966 void kvm_cpu_synchronize_state(CPUState
*env
)
968 if (!env
->kvm_vcpu_dirty
) {
969 run_on_cpu(env
, do_kvm_cpu_synchronize_state
, env
);
973 void kvm_cpu_synchronize_post_reset(CPUState
*env
)
975 kvm_arch_put_registers(env
, KVM_PUT_RESET_STATE
);
976 env
->kvm_vcpu_dirty
= 0;
979 void kvm_cpu_synchronize_post_init(CPUState
*env
)
981 kvm_arch_put_registers(env
, KVM_PUT_FULL_STATE
);
982 env
->kvm_vcpu_dirty
= 0;
985 int kvm_cpu_exec(CPUState
*env
)
987 struct kvm_run
*run
= env
->kvm_run
;
990 DPRINTF("kvm_cpu_exec()\n");
992 if (kvm_arch_process_async_events(env
)) {
993 env
->exit_request
= 0;
997 cpu_single_env
= env
;
1000 if (env
->kvm_vcpu_dirty
) {
1001 kvm_arch_put_registers(env
, KVM_PUT_RUNTIME_STATE
);
1002 env
->kvm_vcpu_dirty
= 0;
1005 kvm_arch_pre_run(env
, run
);
1006 if (env
->exit_request
) {
1007 DPRINTF("interrupt exit requested\n");
1009 * KVM requires us to reenter the kernel after IO exits to complete
1010 * instruction emulation. This self-signal will ensure that we
1013 qemu_cpu_kick_self();
1015 cpu_single_env
= NULL
;
1016 qemu_mutex_unlock_iothread();
1018 run_ret
= kvm_vcpu_ioctl(env
, KVM_RUN
, 0);
1020 qemu_mutex_lock_iothread();
1021 cpu_single_env
= env
;
1022 kvm_arch_post_run(env
, run
);
1024 kvm_flush_coalesced_mmio_buffer();
1027 if (run_ret
== -EINTR
|| run_ret
== -EAGAIN
) {
1028 DPRINTF("io window exit\n");
1029 ret
= EXCP_INTERRUPT
;
1032 fprintf(stderr
, "error: kvm run failed %s\n",
1033 strerror(-run_ret
));
1037 switch (run
->exit_reason
) {
1039 DPRINTF("handle_io\n");
1040 kvm_handle_io(run
->io
.port
,
1041 (uint8_t *)run
+ run
->io
.data_offset
,
1048 DPRINTF("handle_mmio\n");
1049 cpu_physical_memory_rw(run
->mmio
.phys_addr
,
1052 run
->mmio
.is_write
);
1055 case KVM_EXIT_IRQ_WINDOW_OPEN
:
1056 DPRINTF("irq_window_open\n");
1057 ret
= EXCP_INTERRUPT
;
1059 case KVM_EXIT_SHUTDOWN
:
1060 DPRINTF("shutdown\n");
1061 qemu_system_reset_request();
1062 ret
= EXCP_INTERRUPT
;
1064 case KVM_EXIT_UNKNOWN
:
1065 fprintf(stderr
, "KVM: unknown exit, hardware reason %" PRIx64
"\n",
1066 (uint64_t)run
->hw
.hardware_exit_reason
);
1069 case KVM_EXIT_INTERNAL_ERROR
:
1070 ret
= kvm_handle_internal_error(env
, run
);
1073 DPRINTF("kvm_arch_handle_exit\n");
1074 ret
= kvm_arch_handle_exit(env
, run
);
1080 cpu_dump_state(env
, stderr
, fprintf
, CPU_DUMP_CODE
);
1081 vm_stop(RUN_STATE_INTERNAL_ERROR
);
1084 env
->exit_request
= 0;
1085 cpu_single_env
= NULL
;
1089 int kvm_ioctl(KVMState
*s
, int type
, ...)
1096 arg
= va_arg(ap
, void *);
1099 ret
= ioctl(s
->fd
, type
, arg
);
1106 int kvm_vm_ioctl(KVMState
*s
, int type
, ...)
1113 arg
= va_arg(ap
, void *);
1116 ret
= ioctl(s
->vmfd
, type
, arg
);
1123 int kvm_vcpu_ioctl(CPUState
*env
, int type
, ...)
1130 arg
= va_arg(ap
, void *);
1133 ret
= ioctl(env
->kvm_fd
, type
, arg
);
1140 int kvm_has_sync_mmu(void)
1142 return kvm_check_extension(kvm_state
, KVM_CAP_SYNC_MMU
);
1145 int kvm_has_vcpu_events(void)
1147 return kvm_state
->vcpu_events
;
1150 int kvm_has_robust_singlestep(void)
1152 return kvm_state
->robust_singlestep
;
1155 int kvm_has_debugregs(void)
1157 return kvm_state
->debugregs
;
1160 int kvm_has_xsave(void)
1162 return kvm_state
->xsave
;
1165 int kvm_has_xcrs(void)
1167 return kvm_state
->xcrs
;
1170 int kvm_has_pit_state2(void)
1172 return kvm_state
->pit_state2
;
1175 int kvm_has_many_ioeventfds(void)
1177 if (!kvm_enabled()) {
1180 return kvm_state
->many_ioeventfds
;
1183 int kvm_allows_irq0_override(void)
1185 return !kvm_enabled() || !kvm_irqchip_in_kernel() || kvm_has_gsi_routing();
1188 void kvm_setup_guest_memory(void *start
, size_t size
)
1190 if (!kvm_has_sync_mmu()) {
1191 int ret
= qemu_madvise(start
, size
, QEMU_MADV_DONTFORK
);
1194 perror("qemu_madvise");
1196 "Need MADV_DONTFORK in absence of synchronous KVM MMU\n");
1202 #ifdef KVM_CAP_SET_GUEST_DEBUG
1203 struct kvm_sw_breakpoint
*kvm_find_sw_breakpoint(CPUState
*env
,
1206 struct kvm_sw_breakpoint
*bp
;
1208 QTAILQ_FOREACH(bp
, &env
->kvm_state
->kvm_sw_breakpoints
, entry
) {
1216 int kvm_sw_breakpoints_active(CPUState
*env
)
1218 return !QTAILQ_EMPTY(&env
->kvm_state
->kvm_sw_breakpoints
);
1221 struct kvm_set_guest_debug_data
{
1222 struct kvm_guest_debug dbg
;
1227 static void kvm_invoke_set_guest_debug(void *data
)
1229 struct kvm_set_guest_debug_data
*dbg_data
= data
;
1230 CPUState
*env
= dbg_data
->env
;
1232 dbg_data
->err
= kvm_vcpu_ioctl(env
, KVM_SET_GUEST_DEBUG
, &dbg_data
->dbg
);
1235 int kvm_update_guest_debug(CPUState
*env
, unsigned long reinject_trap
)
1237 struct kvm_set_guest_debug_data data
;
1239 data
.dbg
.control
= reinject_trap
;
1241 if (env
->singlestep_enabled
) {
1242 data
.dbg
.control
|= KVM_GUESTDBG_ENABLE
| KVM_GUESTDBG_SINGLESTEP
;
1244 kvm_arch_update_guest_debug(env
, &data
.dbg
);
1247 run_on_cpu(env
, kvm_invoke_set_guest_debug
, &data
);
1251 int kvm_insert_breakpoint(CPUState
*current_env
, target_ulong addr
,
1252 target_ulong len
, int type
)
1254 struct kvm_sw_breakpoint
*bp
;
1258 if (type
== GDB_BREAKPOINT_SW
) {
1259 bp
= kvm_find_sw_breakpoint(current_env
, addr
);
1265 bp
= g_malloc(sizeof(struct kvm_sw_breakpoint
));
1272 err
= kvm_arch_insert_sw_breakpoint(current_env
, bp
);
1278 QTAILQ_INSERT_HEAD(¤t_env
->kvm_state
->kvm_sw_breakpoints
,
1281 err
= kvm_arch_insert_hw_breakpoint(addr
, len
, type
);
1287 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
1288 err
= kvm_update_guest_debug(env
, 0);
1296 int kvm_remove_breakpoint(CPUState
*current_env
, target_ulong addr
,
1297 target_ulong len
, int type
)
1299 struct kvm_sw_breakpoint
*bp
;
1303 if (type
== GDB_BREAKPOINT_SW
) {
1304 bp
= kvm_find_sw_breakpoint(current_env
, addr
);
1309 if (bp
->use_count
> 1) {
1314 err
= kvm_arch_remove_sw_breakpoint(current_env
, bp
);
1319 QTAILQ_REMOVE(¤t_env
->kvm_state
->kvm_sw_breakpoints
, bp
, entry
);
1322 err
= kvm_arch_remove_hw_breakpoint(addr
, len
, type
);
1328 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
1329 err
= kvm_update_guest_debug(env
, 0);
1337 void kvm_remove_all_breakpoints(CPUState
*current_env
)
1339 struct kvm_sw_breakpoint
*bp
, *next
;
1340 KVMState
*s
= current_env
->kvm_state
;
1343 QTAILQ_FOREACH_SAFE(bp
, &s
->kvm_sw_breakpoints
, entry
, next
) {
1344 if (kvm_arch_remove_sw_breakpoint(current_env
, bp
) != 0) {
1345 /* Try harder to find a CPU that currently sees the breakpoint. */
1346 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
1347 if (kvm_arch_remove_sw_breakpoint(env
, bp
) == 0) {
1353 kvm_arch_remove_all_hw_breakpoints();
1355 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
1356 kvm_update_guest_debug(env
, 0);
1360 #else /* !KVM_CAP_SET_GUEST_DEBUG */
1362 int kvm_update_guest_debug(CPUState
*env
, unsigned long reinject_trap
)
1367 int kvm_insert_breakpoint(CPUState
*current_env
, target_ulong addr
,
1368 target_ulong len
, int type
)
1373 int kvm_remove_breakpoint(CPUState
*current_env
, target_ulong addr
,
1374 target_ulong len
, int type
)
1379 void kvm_remove_all_breakpoints(CPUState
*current_env
)
1382 #endif /* !KVM_CAP_SET_GUEST_DEBUG */
1384 int kvm_set_signal_mask(CPUState
*env
, const sigset_t
*sigset
)
1386 struct kvm_signal_mask
*sigmask
;
1390 return kvm_vcpu_ioctl(env
, KVM_SET_SIGNAL_MASK
, NULL
);
1393 sigmask
= g_malloc(sizeof(*sigmask
) + sizeof(*sigset
));
1396 memcpy(sigmask
->sigset
, sigset
, sizeof(*sigset
));
1397 r
= kvm_vcpu_ioctl(env
, KVM_SET_SIGNAL_MASK
, sigmask
);
1403 int kvm_set_ioeventfd_mmio_long(int fd
, uint32_t addr
, uint32_t val
, bool assign
)
1406 struct kvm_ioeventfd iofd
;
1408 iofd
.datamatch
= val
;
1411 iofd
.flags
= KVM_IOEVENTFD_FLAG_DATAMATCH
;
1414 if (!kvm_enabled()) {
1419 iofd
.flags
|= KVM_IOEVENTFD_FLAG_DEASSIGN
;
1422 ret
= kvm_vm_ioctl(kvm_state
, KVM_IOEVENTFD
, &iofd
);
1431 int kvm_set_ioeventfd_pio_word(int fd
, uint16_t addr
, uint16_t val
, bool assign
)
1433 struct kvm_ioeventfd kick
= {
1437 .flags
= KVM_IOEVENTFD_FLAG_DATAMATCH
| KVM_IOEVENTFD_FLAG_PIO
,
1441 if (!kvm_enabled()) {
1445 kick
.flags
|= KVM_IOEVENTFD_FLAG_DEASSIGN
;
1447 r
= kvm_vm_ioctl(kvm_state
, KVM_IOEVENTFD
, &kick
);
1454 int kvm_set_irqfd(int gsi
, int fd
, bool assigned
)
1456 struct kvm_irqfd irqfd
= {
1459 .flags
= assigned
? 0 : KVM_IRQFD_FLAG_DEASSIGN
,
1462 if (!kvm_enabled() || !kvm_irqchip_in_kernel())
1465 r
= kvm_vm_ioctl(kvm_state
, KVM_IRQFD
, &irqfd
);
1471 int kvm_on_sigbus_vcpu(CPUState
*env
, int code
, void *addr
)
1473 return kvm_arch_on_sigbus_vcpu(env
, code
, addr
);
1476 int kvm_on_sigbus(int code
, void *addr
)
1478 return kvm_arch_on_sigbus(code
, addr
);
1482 #include "qemu-kvm.c"