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 /* This check must be after config-host.h is included */
33 #include <sys/eventfd.h>
36 /* KVM uses PAGE_SIZE in it's definition of COALESCED_MMIO_MAX */
37 #define PAGE_SIZE TARGET_PAGE_SIZE
42 #define DPRINTF(fmt, ...) \
43 do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
45 #define DPRINTF(fmt, ...) \
49 typedef struct KVMSlot
51 target_phys_addr_t start_addr
;
52 ram_addr_t memory_size
;
53 ram_addr_t phys_offset
;
58 typedef struct kvm_dirty_log KVMDirtyLog
;
66 struct kvm_coalesced_mmio_ring
*coalesced_mmio_ring
;
67 int broken_set_mem_region
;
70 int robust_singlestep
;
72 #ifdef KVM_CAP_SET_GUEST_DEBUG
73 struct kvm_sw_breakpoint_head kvm_sw_breakpoints
;
75 int irqchip_in_kernel
;
81 int irqchip_inject_ioctl
;
82 #ifdef KVM_CAP_IRQ_ROUTING
83 struct kvm_irq_routing
*irq_routes
;
84 int nr_allocated_irq_routes
;
86 void *used_gsi_bitmap
;
92 static const KVMCapabilityInfo kvm_required_capabilites
[] = {
93 KVM_CAP_INFO(USER_MEMORY
),
94 KVM_CAP_INFO(DESTROY_MEMORY_REGION_WORKS
),
98 static KVMSlot
*kvm_alloc_slot(KVMState
*s
)
102 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
103 if (s
->slots
[i
].memory_size
== 0) {
108 fprintf(stderr
, "%s: no free slot available\n", __func__
);
112 static KVMSlot
*kvm_lookup_matching_slot(KVMState
*s
,
113 target_phys_addr_t start_addr
,
114 target_phys_addr_t end_addr
)
118 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
119 KVMSlot
*mem
= &s
->slots
[i
];
121 if (start_addr
== mem
->start_addr
&&
122 end_addr
== mem
->start_addr
+ mem
->memory_size
) {
131 * Find overlapping slot with lowest start address
133 static KVMSlot
*kvm_lookup_overlapping_slot(KVMState
*s
,
134 target_phys_addr_t start_addr
,
135 target_phys_addr_t end_addr
)
137 KVMSlot
*found
= NULL
;
140 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
141 KVMSlot
*mem
= &s
->slots
[i
];
143 if (mem
->memory_size
== 0 ||
144 (found
&& found
->start_addr
< mem
->start_addr
)) {
148 if (end_addr
> mem
->start_addr
&&
149 start_addr
< mem
->start_addr
+ mem
->memory_size
) {
157 int kvm_physical_memory_addr_from_ram(KVMState
*s
, ram_addr_t ram_addr
,
158 target_phys_addr_t
*phys_addr
)
162 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
163 KVMSlot
*mem
= &s
->slots
[i
];
165 if (ram_addr
>= mem
->phys_offset
&&
166 ram_addr
< mem
->phys_offset
+ mem
->memory_size
) {
167 *phys_addr
= mem
->start_addr
+ (ram_addr
- mem
->phys_offset
);
175 static int kvm_set_user_memory_region(KVMState
*s
, KVMSlot
*slot
)
177 struct kvm_userspace_memory_region mem
;
179 mem
.slot
= slot
->slot
;
180 mem
.guest_phys_addr
= slot
->start_addr
;
181 mem
.memory_size
= slot
->memory_size
;
182 mem
.userspace_addr
= (unsigned long)qemu_safe_ram_ptr(slot
->phys_offset
);
183 mem
.flags
= slot
->flags
;
184 if (s
->migration_log
) {
185 mem
.flags
|= KVM_MEM_LOG_DIRTY_PAGES
;
187 return kvm_vm_ioctl(s
, KVM_SET_USER_MEMORY_REGION
, &mem
);
190 static void kvm_reset_vcpu(void *opaque
)
192 CPUState
*env
= opaque
;
194 kvm_arch_reset_vcpu(env
);
197 int kvm_irqchip_in_kernel(void)
199 return kvm_state
->irqchip_in_kernel
;
202 int kvm_pit_in_kernel(void)
204 return kvm_state
->pit_in_kernel
;
207 int kvm_init_vcpu(CPUState
*env
)
209 KVMState
*s
= kvm_state
;
213 DPRINTF("kvm_init_vcpu\n");
215 ret
= kvm_vm_ioctl(s
, KVM_CREATE_VCPU
, env
->cpu_index
);
217 DPRINTF("kvm_create_vcpu failed\n");
223 env
->kvm_vcpu_dirty
= 1;
225 mmap_size
= kvm_ioctl(s
, KVM_GET_VCPU_MMAP_SIZE
, 0);
228 DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n");
232 env
->kvm_run
= mmap(NULL
, mmap_size
, PROT_READ
| PROT_WRITE
, MAP_SHARED
,
234 if (env
->kvm_run
== MAP_FAILED
) {
236 DPRINTF("mmap'ing vcpu state failed\n");
240 if (s
->coalesced_mmio
&& !s
->coalesced_mmio_ring
) {
241 s
->coalesced_mmio_ring
=
242 (void *)env
->kvm_run
+ s
->coalesced_mmio
* PAGE_SIZE
;
245 ret
= kvm_arch_init_vcpu(env
);
247 qemu_register_reset(kvm_reset_vcpu
, env
);
248 kvm_arch_reset_vcpu(env
);
255 * dirty pages logging control
258 static int kvm_mem_flags(KVMState
*s
, bool log_dirty
)
260 return log_dirty
? KVM_MEM_LOG_DIRTY_PAGES
: 0;
263 static int kvm_slot_dirty_pages_log_change(KVMSlot
*mem
, bool log_dirty
)
265 KVMState
*s
= kvm_state
;
266 int flags
, mask
= KVM_MEM_LOG_DIRTY_PAGES
;
269 old_flags
= mem
->flags
;
271 flags
= (mem
->flags
& ~mask
) | kvm_mem_flags(s
, log_dirty
);
274 /* If nothing changed effectively, no need to issue ioctl */
275 if (s
->migration_log
) {
276 flags
|= KVM_MEM_LOG_DIRTY_PAGES
;
279 if (flags
== old_flags
) {
283 return kvm_set_user_memory_region(s
, mem
);
286 static int kvm_dirty_pages_log_change(target_phys_addr_t phys_addr
,
287 ram_addr_t size
, bool log_dirty
)
289 KVMState
*s
= kvm_state
;
290 KVMSlot
*mem
= kvm_lookup_matching_slot(s
, phys_addr
, phys_addr
+ size
);
293 fprintf(stderr
, "BUG: %s: invalid parameters " TARGET_FMT_plx
"-"
294 TARGET_FMT_plx
"\n", __func__
, phys_addr
,
295 (target_phys_addr_t
)(phys_addr
+ size
- 1));
298 return kvm_slot_dirty_pages_log_change(mem
, log_dirty
);
301 static int kvm_log_start(CPUPhysMemoryClient
*client
,
302 target_phys_addr_t phys_addr
, ram_addr_t size
)
304 return kvm_dirty_pages_log_change(phys_addr
, size
, true);
307 static int kvm_log_stop(CPUPhysMemoryClient
*client
,
308 target_phys_addr_t phys_addr
, ram_addr_t size
)
310 return kvm_dirty_pages_log_change(phys_addr
, size
, false);
313 static int kvm_set_migration_log(int enable
)
315 KVMState
*s
= kvm_state
;
319 s
->migration_log
= enable
;
321 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
324 if (!mem
->memory_size
) {
327 if (!!(mem
->flags
& KVM_MEM_LOG_DIRTY_PAGES
) == enable
) {
330 err
= kvm_set_user_memory_region(s
, mem
);
338 /* get kvm's dirty pages bitmap and update qemu's */
339 static int kvm_get_dirty_pages_log_range(unsigned long start_addr
,
340 unsigned long *bitmap
,
341 unsigned long offset
,
342 unsigned long mem_size
)
345 unsigned long page_number
, addr
, addr1
, c
;
347 unsigned int len
= ((mem_size
/ TARGET_PAGE_SIZE
) + HOST_LONG_BITS
- 1) /
351 * bitmap-traveling is faster than memory-traveling (for addr...)
352 * especially when most of the memory is not dirty.
354 for (i
= 0; i
< len
; i
++) {
355 if (bitmap
[i
] != 0) {
356 c
= leul_to_cpu(bitmap
[i
]);
360 page_number
= i
* HOST_LONG_BITS
+ j
;
361 addr1
= page_number
* TARGET_PAGE_SIZE
;
362 addr
= offset
+ addr1
;
363 ram_addr
= cpu_get_physical_page_desc(addr
);
364 cpu_physical_memory_set_dirty(ram_addr
);
371 #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1))
374 * kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space
375 * This function updates qemu's dirty bitmap using cpu_physical_memory_set_dirty().
376 * This means all bits are set to dirty.
378 * @start_add: start of logged region.
379 * @end_addr: end of logged region.
381 static int kvm_physical_sync_dirty_bitmap(target_phys_addr_t start_addr
,
382 target_phys_addr_t end_addr
)
384 KVMState
*s
= kvm_state
;
385 unsigned long size
, allocated_size
= 0;
390 d
.dirty_bitmap
= NULL
;
391 while (start_addr
< end_addr
) {
392 mem
= kvm_lookup_overlapping_slot(s
, start_addr
, end_addr
);
397 /* XXX bad kernel interface alert
398 * For dirty bitmap, kernel allocates array of size aligned to
399 * bits-per-long. But for case when the kernel is 64bits and
400 * the userspace is 32bits, userspace can't align to the same
401 * bits-per-long, since sizeof(long) is different between kernel
402 * and user space. This way, userspace will provide buffer which
403 * may be 4 bytes less than the kernel will use, resulting in
404 * userspace memory corruption (which is not detectable by valgrind
405 * too, in most cases).
406 * So for now, let's align to 64 instead of HOST_LONG_BITS here, in
407 * a hope that sizeof(long) wont become >8 any time soon.
409 size
= ALIGN(((mem
->memory_size
) >> TARGET_PAGE_BITS
),
410 /*HOST_LONG_BITS*/ 64) / 8;
411 if (!d
.dirty_bitmap
) {
412 d
.dirty_bitmap
= qemu_malloc(size
);
413 } else if (size
> allocated_size
) {
414 d
.dirty_bitmap
= qemu_realloc(d
.dirty_bitmap
, size
);
416 allocated_size
= size
;
417 memset(d
.dirty_bitmap
, 0, allocated_size
);
421 if (kvm_vm_ioctl(s
, KVM_GET_DIRTY_LOG
, &d
) == -1) {
422 DPRINTF("ioctl failed %d\n", errno
);
427 kvm_get_dirty_pages_log_range(mem
->start_addr
, d
.dirty_bitmap
,
428 mem
->start_addr
, mem
->memory_size
);
429 start_addr
= mem
->start_addr
+ mem
->memory_size
;
431 qemu_free(d
.dirty_bitmap
);
436 int kvm_coalesce_mmio_region(target_phys_addr_t start
, ram_addr_t size
)
439 KVMState
*s
= kvm_state
;
441 if (s
->coalesced_mmio
) {
442 struct kvm_coalesced_mmio_zone zone
;
447 ret
= kvm_vm_ioctl(s
, KVM_REGISTER_COALESCED_MMIO
, &zone
);
453 int kvm_uncoalesce_mmio_region(target_phys_addr_t start
, ram_addr_t size
)
456 KVMState
*s
= kvm_state
;
458 if (s
->coalesced_mmio
) {
459 struct kvm_coalesced_mmio_zone zone
;
464 ret
= kvm_vm_ioctl(s
, KVM_UNREGISTER_COALESCED_MMIO
, &zone
);
470 int kvm_check_extension(KVMState
*s
, unsigned int extension
)
474 ret
= kvm_ioctl(s
, KVM_CHECK_EXTENSION
, extension
);
482 static int kvm_check_many_ioeventfds(void)
484 /* Userspace can use ioeventfd for io notification. This requires a host
485 * that supports eventfd(2) and an I/O thread; since eventfd does not
486 * support SIGIO it cannot interrupt the vcpu.
488 * Older kernels have a 6 device limit on the KVM io bus. Find out so we
489 * can avoid creating too many ioeventfds.
491 #if defined(CONFIG_EVENTFD) && defined(CONFIG_IOTHREAD)
494 for (i
= 0; i
< ARRAY_SIZE(ioeventfds
); i
++) {
495 ioeventfds
[i
] = eventfd(0, EFD_CLOEXEC
);
496 if (ioeventfds
[i
] < 0) {
499 ret
= kvm_set_ioeventfd_pio_word(ioeventfds
[i
], 0, i
, true);
501 close(ioeventfds
[i
]);
506 /* Decide whether many devices are supported or not */
507 ret
= i
== ARRAY_SIZE(ioeventfds
);
510 kvm_set_ioeventfd_pio_word(ioeventfds
[i
], 0, i
, false);
511 close(ioeventfds
[i
]);
519 static const KVMCapabilityInfo
*
520 kvm_check_extension_list(KVMState
*s
, const KVMCapabilityInfo
*list
)
523 if (!kvm_check_extension(s
, list
->value
)) {
531 static void kvm_set_phys_mem(target_phys_addr_t start_addr
, ram_addr_t size
,
532 ram_addr_t phys_offset
, bool log_dirty
)
534 KVMState
*s
= kvm_state
;
535 ram_addr_t flags
= phys_offset
& ~TARGET_PAGE_MASK
;
539 /* kvm works in page size chunks, but the function may be called
540 with sub-page size and unaligned start address. */
541 size
= TARGET_PAGE_ALIGN(size
);
542 start_addr
= TARGET_PAGE_ALIGN(start_addr
);
544 /* KVM does not support read-only slots */
545 phys_offset
&= ~IO_MEM_ROM
;
548 mem
= kvm_lookup_overlapping_slot(s
, start_addr
, start_addr
+ size
);
553 if (flags
< IO_MEM_UNASSIGNED
&& start_addr
>= mem
->start_addr
&&
554 (start_addr
+ size
<= mem
->start_addr
+ mem
->memory_size
) &&
555 (phys_offset
- start_addr
== mem
->phys_offset
- mem
->start_addr
)) {
556 /* The new slot fits into the existing one and comes with
557 * identical parameters - update flags and done. */
558 kvm_slot_dirty_pages_log_change(mem
, log_dirty
);
564 /* unregister the overlapping slot */
565 mem
->memory_size
= 0;
566 err
= kvm_set_user_memory_region(s
, mem
);
568 fprintf(stderr
, "%s: error unregistering overlapping slot: %s\n",
569 __func__
, strerror(-err
));
573 /* Workaround for older KVM versions: we can't join slots, even not by
574 * unregistering the previous ones and then registering the larger
575 * slot. We have to maintain the existing fragmentation. Sigh.
577 * This workaround assumes that the new slot starts at the same
578 * address as the first existing one. If not or if some overlapping
579 * slot comes around later, we will fail (not seen in practice so far)
580 * - and actually require a recent KVM version. */
581 if (s
->broken_set_mem_region
&&
582 old
.start_addr
== start_addr
&& old
.memory_size
< size
&&
583 flags
< IO_MEM_UNASSIGNED
) {
584 mem
= kvm_alloc_slot(s
);
585 mem
->memory_size
= old
.memory_size
;
586 mem
->start_addr
= old
.start_addr
;
587 mem
->phys_offset
= old
.phys_offset
;
588 mem
->flags
= kvm_mem_flags(s
, log_dirty
);
590 err
= kvm_set_user_memory_region(s
, mem
);
592 fprintf(stderr
, "%s: error updating slot: %s\n", __func__
,
597 start_addr
+= old
.memory_size
;
598 phys_offset
+= old
.memory_size
;
599 size
-= old
.memory_size
;
603 /* register prefix slot */
604 if (old
.start_addr
< start_addr
) {
605 mem
= kvm_alloc_slot(s
);
606 mem
->memory_size
= start_addr
- old
.start_addr
;
607 mem
->start_addr
= old
.start_addr
;
608 mem
->phys_offset
= old
.phys_offset
;
609 mem
->flags
= kvm_mem_flags(s
, log_dirty
);
611 err
= kvm_set_user_memory_region(s
, mem
);
613 fprintf(stderr
, "%s: error registering prefix slot: %s\n",
614 __func__
, strerror(-err
));
616 fprintf(stderr
, "%s: This is probably because your kernel's " \
617 "PAGE_SIZE is too big. Please try to use 4k " \
618 "PAGE_SIZE!\n", __func__
);
624 /* register suffix slot */
625 if (old
.start_addr
+ old
.memory_size
> start_addr
+ size
) {
626 ram_addr_t size_delta
;
628 mem
= kvm_alloc_slot(s
);
629 mem
->start_addr
= start_addr
+ size
;
630 size_delta
= mem
->start_addr
- old
.start_addr
;
631 mem
->memory_size
= old
.memory_size
- size_delta
;
632 mem
->phys_offset
= old
.phys_offset
+ size_delta
;
633 mem
->flags
= kvm_mem_flags(s
, log_dirty
);
635 err
= kvm_set_user_memory_region(s
, mem
);
637 fprintf(stderr
, "%s: error registering suffix slot: %s\n",
638 __func__
, strerror(-err
));
644 /* in case the KVM bug workaround already "consumed" the new slot */
648 /* KVM does not need to know about this memory */
649 if (flags
>= IO_MEM_UNASSIGNED
) {
652 mem
= kvm_alloc_slot(s
);
653 mem
->memory_size
= size
;
654 mem
->start_addr
= start_addr
;
655 mem
->phys_offset
= phys_offset
;
656 mem
->flags
= kvm_mem_flags(s
, log_dirty
);
658 err
= kvm_set_user_memory_region(s
, mem
);
660 fprintf(stderr
, "%s: error registering slot: %s\n", __func__
,
666 static void kvm_client_set_memory(struct CPUPhysMemoryClient
*client
,
667 target_phys_addr_t start_addr
,
668 ram_addr_t size
, ram_addr_t phys_offset
,
671 kvm_set_phys_mem(start_addr
, size
, phys_offset
, log_dirty
);
674 static int kvm_client_sync_dirty_bitmap(struct CPUPhysMemoryClient
*client
,
675 target_phys_addr_t start_addr
,
676 target_phys_addr_t end_addr
)
678 return kvm_physical_sync_dirty_bitmap(start_addr
, end_addr
);
681 static int kvm_client_migration_log(struct CPUPhysMemoryClient
*client
,
684 return kvm_set_migration_log(enable
);
687 static CPUPhysMemoryClient kvm_cpu_phys_memory_client
= {
688 .set_memory
= kvm_client_set_memory
,
689 .sync_dirty_bitmap
= kvm_client_sync_dirty_bitmap
,
690 .migration_log
= kvm_client_migration_log
,
691 .log_start
= kvm_log_start
,
692 .log_stop
= kvm_log_stop
,
695 static void kvm_handle_interrupt(CPUState
*env
, int mask
)
697 env
->interrupt_request
|= mask
;
699 if (!qemu_cpu_is_self(env
)) {
706 static const char upgrade_note
[] =
707 "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n"
708 "(see http://sourceforge.net/projects/kvm).\n";
710 const KVMCapabilityInfo
*missing_cap
;
714 s
= qemu_mallocz(sizeof(KVMState
));
716 #ifdef KVM_CAP_SET_GUEST_DEBUG
717 QTAILQ_INIT(&s
->kvm_sw_breakpoints
);
719 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
720 s
->slots
[i
].slot
= i
;
723 s
->fd
= qemu_open("/dev/kvm", O_RDWR
);
725 fprintf(stderr
, "Could not access KVM kernel module: %m\n");
730 ret
= kvm_ioctl(s
, KVM_GET_API_VERSION
, 0);
731 if (ret
< KVM_API_VERSION
) {
735 fprintf(stderr
, "kvm version too old\n");
739 if (ret
> KVM_API_VERSION
) {
741 fprintf(stderr
, "kvm version not supported\n");
745 s
->vmfd
= kvm_ioctl(s
, KVM_CREATE_VM
, 0);
748 fprintf(stderr
, "Please add the 'switch_amode' kernel parameter to "
749 "your host kernel command line\n");
754 missing_cap
= kvm_check_extension_list(s
, kvm_required_capabilites
);
757 kvm_check_extension_list(s
, kvm_arch_required_capabilities
);
761 fprintf(stderr
, "kvm does not support %s\n%s",
762 missing_cap
->name
, upgrade_note
);
766 s
->coalesced_mmio
= kvm_check_extension(s
, KVM_CAP_COALESCED_MMIO
);
768 s
->broken_set_mem_region
= 1;
769 ret
= kvm_check_extension(s
, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
);
771 s
->broken_set_mem_region
= 0;
774 #ifdef KVM_CAP_VCPU_EVENTS
775 s
->vcpu_events
= kvm_check_extension(s
, KVM_CAP_VCPU_EVENTS
);
778 s
->robust_singlestep
=
779 kvm_check_extension(s
, KVM_CAP_X86_ROBUST_SINGLESTEP
);
781 #ifdef KVM_CAP_DEBUGREGS
782 s
->debugregs
= kvm_check_extension(s
, KVM_CAP_DEBUGREGS
);
786 s
->xsave
= kvm_check_extension(s
, KVM_CAP_XSAVE
);
790 s
->xcrs
= kvm_check_extension(s
, KVM_CAP_XCRS
);
794 #ifdef KVM_CAP_PIT_STATE2
795 s
->pit_state2
= kvm_check_extension(s
, KVM_CAP_PIT_STATE2
);
798 s
->pit_in_kernel
= kvm_pit
;
800 ret
= kvm_arch_init(s
);
806 cpu_register_phys_memory_client(&kvm_cpu_phys_memory_client
);
808 s
->many_ioeventfds
= kvm_check_many_ioeventfds();
810 ret
= kvm_create_irqchip(s
);
815 cpu_interrupt_handler
= kvm_handle_interrupt
;
833 static void kvm_handle_io(uint16_t port
, void *data
, int direction
, int size
,
839 for (i
= 0; i
< count
; i
++) {
840 if (direction
== KVM_EXIT_IO_IN
) {
843 stb_p(ptr
, cpu_inb(port
));
846 stw_p(ptr
, cpu_inw(port
));
849 stl_p(ptr
, cpu_inl(port
));
855 cpu_outb(port
, ldub_p(ptr
));
858 cpu_outw(port
, lduw_p(ptr
));
861 cpu_outl(port
, ldl_p(ptr
));
870 static int kvm_handle_internal_error(CPUState
*env
, struct kvm_run
*run
)
872 fprintf(stderr
, "KVM internal error.");
873 if (kvm_check_extension(kvm_state
, KVM_CAP_INTERNAL_ERROR_DATA
)) {
876 fprintf(stderr
, " Suberror: %d\n", run
->internal
.suberror
);
877 for (i
= 0; i
< run
->internal
.ndata
; ++i
) {
878 fprintf(stderr
, "extra data[%d]: %"PRIx64
"\n",
879 i
, (uint64_t)run
->internal
.data
[i
]);
882 fprintf(stderr
, "\n");
884 if (run
->internal
.suberror
== KVM_INTERNAL_ERROR_EMULATION
) {
885 fprintf(stderr
, "emulation failure\n");
886 if (!kvm_arch_stop_on_emulation_error(env
)) {
887 cpu_dump_state(env
, stderr
, fprintf
, CPU_DUMP_CODE
);
888 return EXCP_INTERRUPT
;
891 /* FIXME: Should trigger a qmp message to let management know
892 * something went wrong.
897 void kvm_flush_coalesced_mmio_buffer(void)
899 KVMState
*s
= kvm_state
;
900 if (s
->coalesced_mmio_ring
) {
901 struct kvm_coalesced_mmio_ring
*ring
= s
->coalesced_mmio_ring
;
902 while (ring
->first
!= ring
->last
) {
903 struct kvm_coalesced_mmio
*ent
;
905 ent
= &ring
->coalesced_mmio
[ring
->first
];
907 cpu_physical_memory_write(ent
->phys_addr
, ent
->data
, ent
->len
);
909 ring
->first
= (ring
->first
+ 1) % KVM_COALESCED_MMIO_MAX
;
914 static void do_kvm_cpu_synchronize_state(void *_env
)
916 CPUState
*env
= _env
;
918 if (!env
->kvm_vcpu_dirty
) {
919 kvm_arch_get_registers(env
);
920 env
->kvm_vcpu_dirty
= 1;
924 void kvm_cpu_synchronize_state(CPUState
*env
)
926 if (!env
->kvm_vcpu_dirty
) {
927 run_on_cpu(env
, do_kvm_cpu_synchronize_state
, env
);
931 void kvm_cpu_synchronize_post_reset(CPUState
*env
)
933 kvm_arch_put_registers(env
, KVM_PUT_RESET_STATE
);
934 env
->kvm_vcpu_dirty
= 0;
937 void kvm_cpu_synchronize_post_init(CPUState
*env
)
939 kvm_arch_put_registers(env
, KVM_PUT_FULL_STATE
);
940 env
->kvm_vcpu_dirty
= 0;
943 int kvm_cpu_exec(CPUState
*env
)
945 struct kvm_run
*run
= env
->kvm_run
;
948 DPRINTF("kvm_cpu_exec()\n");
950 if (kvm_arch_process_async_events(env
)) {
951 env
->exit_request
= 0;
955 cpu_single_env
= env
;
958 if (env
->kvm_vcpu_dirty
) {
959 kvm_arch_put_registers(env
, KVM_PUT_RUNTIME_STATE
);
960 env
->kvm_vcpu_dirty
= 0;
963 kvm_arch_pre_run(env
, run
);
964 if (env
->exit_request
) {
965 DPRINTF("interrupt exit requested\n");
967 * KVM requires us to reenter the kernel after IO exits to complete
968 * instruction emulation. This self-signal will ensure that we
971 qemu_cpu_kick_self();
973 cpu_single_env
= NULL
;
974 qemu_mutex_unlock_iothread();
976 run_ret
= kvm_vcpu_ioctl(env
, KVM_RUN
, 0);
978 qemu_mutex_lock_iothread();
979 cpu_single_env
= env
;
980 kvm_arch_post_run(env
, run
);
982 kvm_flush_coalesced_mmio_buffer();
985 if (run_ret
== -EINTR
|| run_ret
== -EAGAIN
) {
986 DPRINTF("io window exit\n");
987 ret
= EXCP_INTERRUPT
;
990 DPRINTF("kvm run failed %s\n", strerror(-run_ret
));
994 switch (run
->exit_reason
) {
996 DPRINTF("handle_io\n");
997 kvm_handle_io(run
->io
.port
,
998 (uint8_t *)run
+ run
->io
.data_offset
,
1005 DPRINTF("handle_mmio\n");
1006 cpu_physical_memory_rw(run
->mmio
.phys_addr
,
1009 run
->mmio
.is_write
);
1012 case KVM_EXIT_IRQ_WINDOW_OPEN
:
1013 DPRINTF("irq_window_open\n");
1014 ret
= EXCP_INTERRUPT
;
1016 case KVM_EXIT_SHUTDOWN
:
1017 DPRINTF("shutdown\n");
1018 qemu_system_reset_request();
1019 ret
= EXCP_INTERRUPT
;
1021 case KVM_EXIT_UNKNOWN
:
1022 fprintf(stderr
, "KVM: unknown exit, hardware reason %" PRIx64
"\n",
1023 (uint64_t)run
->hw
.hardware_exit_reason
);
1026 case KVM_EXIT_INTERNAL_ERROR
:
1027 ret
= kvm_handle_internal_error(env
, run
);
1030 DPRINTF("kvm_arch_handle_exit\n");
1031 ret
= kvm_arch_handle_exit(env
, run
);
1037 cpu_dump_state(env
, stderr
, fprintf
, CPU_DUMP_CODE
);
1038 vm_stop(VMSTOP_PANIC
);
1041 env
->exit_request
= 0;
1042 cpu_single_env
= NULL
;
1046 int kvm_ioctl(KVMState
*s
, int type
, ...)
1053 arg
= va_arg(ap
, void *);
1056 ret
= ioctl(s
->fd
, type
, arg
);
1063 int kvm_vm_ioctl(KVMState
*s
, int type
, ...)
1070 arg
= va_arg(ap
, void *);
1073 ret
= ioctl(s
->vmfd
, type
, arg
);
1080 int kvm_vcpu_ioctl(CPUState
*env
, int type
, ...)
1087 arg
= va_arg(ap
, void *);
1090 ret
= ioctl(env
->kvm_fd
, type
, arg
);
1097 int kvm_has_sync_mmu(void)
1099 return kvm_check_extension(kvm_state
, KVM_CAP_SYNC_MMU
);
1102 int kvm_has_vcpu_events(void)
1104 return kvm_state
->vcpu_events
;
1107 int kvm_has_robust_singlestep(void)
1109 return kvm_state
->robust_singlestep
;
1112 int kvm_has_debugregs(void)
1114 return kvm_state
->debugregs
;
1117 int kvm_has_xsave(void)
1119 return kvm_state
->xsave
;
1122 int kvm_has_xcrs(void)
1124 return kvm_state
->xcrs
;
1127 int kvm_has_pit_state2(void)
1129 return kvm_state
->pit_state2
;
1132 int kvm_has_many_ioeventfds(void)
1134 if (!kvm_enabled()) {
1137 return kvm_state
->many_ioeventfds
;
1140 int kvm_allows_irq0_override(void)
1142 return !kvm_enabled() || !kvm_irqchip_in_kernel() || kvm_has_gsi_routing();
1145 void kvm_setup_guest_memory(void *start
, size_t size
)
1147 if (!kvm_has_sync_mmu()) {
1148 int ret
= qemu_madvise(start
, size
, QEMU_MADV_DONTFORK
);
1151 perror("qemu_madvise");
1153 "Need MADV_DONTFORK in absence of synchronous KVM MMU\n");
1159 #ifdef KVM_CAP_SET_GUEST_DEBUG
1160 struct kvm_sw_breakpoint
*kvm_find_sw_breakpoint(CPUState
*env
,
1163 struct kvm_sw_breakpoint
*bp
;
1165 QTAILQ_FOREACH(bp
, &env
->kvm_state
->kvm_sw_breakpoints
, entry
) {
1173 int kvm_sw_breakpoints_active(CPUState
*env
)
1175 return !QTAILQ_EMPTY(&env
->kvm_state
->kvm_sw_breakpoints
);
1178 struct kvm_set_guest_debug_data
{
1179 struct kvm_guest_debug dbg
;
1184 static void kvm_invoke_set_guest_debug(void *data
)
1186 struct kvm_set_guest_debug_data
*dbg_data
= data
;
1187 CPUState
*env
= dbg_data
->env
;
1189 dbg_data
->err
= kvm_vcpu_ioctl(env
, KVM_SET_GUEST_DEBUG
, &dbg_data
->dbg
);
1192 int kvm_update_guest_debug(CPUState
*env
, unsigned long reinject_trap
)
1194 struct kvm_set_guest_debug_data data
;
1196 data
.dbg
.control
= reinject_trap
;
1198 if (env
->singlestep_enabled
) {
1199 data
.dbg
.control
|= KVM_GUESTDBG_ENABLE
| KVM_GUESTDBG_SINGLESTEP
;
1201 kvm_arch_update_guest_debug(env
, &data
.dbg
);
1204 run_on_cpu(env
, kvm_invoke_set_guest_debug
, &data
);
1208 int kvm_insert_breakpoint(CPUState
*current_env
, target_ulong addr
,
1209 target_ulong len
, int type
)
1211 struct kvm_sw_breakpoint
*bp
;
1215 if (type
== GDB_BREAKPOINT_SW
) {
1216 bp
= kvm_find_sw_breakpoint(current_env
, addr
);
1222 bp
= qemu_malloc(sizeof(struct kvm_sw_breakpoint
));
1229 err
= kvm_arch_insert_sw_breakpoint(current_env
, bp
);
1235 QTAILQ_INSERT_HEAD(¤t_env
->kvm_state
->kvm_sw_breakpoints
,
1238 err
= kvm_arch_insert_hw_breakpoint(addr
, len
, type
);
1244 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
1245 err
= kvm_update_guest_debug(env
, 0);
1253 int kvm_remove_breakpoint(CPUState
*current_env
, target_ulong addr
,
1254 target_ulong len
, int type
)
1256 struct kvm_sw_breakpoint
*bp
;
1260 if (type
== GDB_BREAKPOINT_SW
) {
1261 bp
= kvm_find_sw_breakpoint(current_env
, addr
);
1266 if (bp
->use_count
> 1) {
1271 err
= kvm_arch_remove_sw_breakpoint(current_env
, bp
);
1276 QTAILQ_REMOVE(¤t_env
->kvm_state
->kvm_sw_breakpoints
, bp
, entry
);
1279 err
= kvm_arch_remove_hw_breakpoint(addr
, len
, type
);
1285 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
1286 err
= kvm_update_guest_debug(env
, 0);
1294 void kvm_remove_all_breakpoints(CPUState
*current_env
)
1296 struct kvm_sw_breakpoint
*bp
, *next
;
1297 KVMState
*s
= current_env
->kvm_state
;
1300 QTAILQ_FOREACH_SAFE(bp
, &s
->kvm_sw_breakpoints
, entry
, next
) {
1301 if (kvm_arch_remove_sw_breakpoint(current_env
, bp
) != 0) {
1302 /* Try harder to find a CPU that currently sees the breakpoint. */
1303 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
1304 if (kvm_arch_remove_sw_breakpoint(env
, bp
) == 0) {
1310 kvm_arch_remove_all_hw_breakpoints();
1312 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
1313 kvm_update_guest_debug(env
, 0);
1317 #else /* !KVM_CAP_SET_GUEST_DEBUG */
1319 int kvm_update_guest_debug(CPUState
*env
, unsigned long reinject_trap
)
1324 int kvm_insert_breakpoint(CPUState
*current_env
, target_ulong addr
,
1325 target_ulong len
, int type
)
1330 int kvm_remove_breakpoint(CPUState
*current_env
, target_ulong addr
,
1331 target_ulong len
, int type
)
1336 void kvm_remove_all_breakpoints(CPUState
*current_env
)
1339 #endif /* !KVM_CAP_SET_GUEST_DEBUG */
1341 int kvm_set_signal_mask(CPUState
*env
, const sigset_t
*sigset
)
1343 struct kvm_signal_mask
*sigmask
;
1347 return kvm_vcpu_ioctl(env
, KVM_SET_SIGNAL_MASK
, NULL
);
1350 sigmask
= qemu_malloc(sizeof(*sigmask
) + sizeof(*sigset
));
1353 memcpy(sigmask
->sigset
, sigset
, sizeof(*sigset
));
1354 r
= kvm_vcpu_ioctl(env
, KVM_SET_SIGNAL_MASK
, sigmask
);
1360 int kvm_set_ioeventfd_mmio_long(int fd
, uint32_t addr
, uint32_t val
, bool assign
)
1362 #ifdef KVM_IOEVENTFD
1364 struct kvm_ioeventfd iofd
;
1366 iofd
.datamatch
= val
;
1369 iofd
.flags
= KVM_IOEVENTFD_FLAG_DATAMATCH
;
1372 if (!kvm_enabled()) {
1377 iofd
.flags
|= KVM_IOEVENTFD_FLAG_DEASSIGN
;
1380 ret
= kvm_vm_ioctl(kvm_state
, KVM_IOEVENTFD
, &iofd
);
1392 int kvm_set_ioeventfd_pio_word(int fd
, uint16_t addr
, uint16_t val
, bool assign
)
1394 #ifdef KVM_IOEVENTFD
1395 struct kvm_ioeventfd kick
= {
1399 .flags
= KVM_IOEVENTFD_FLAG_DATAMATCH
| KVM_IOEVENTFD_FLAG_PIO
,
1403 if (!kvm_enabled()) {
1407 kick
.flags
|= KVM_IOEVENTFD_FLAG_DEASSIGN
;
1409 r
= kvm_vm_ioctl(kvm_state
, KVM_IOEVENTFD
, &kick
);
1419 int kvm_set_irqfd(int gsi
, int fd
, bool assigned
)
1421 struct kvm_irqfd irqfd
= {
1424 .flags
= assigned
? 0 : KVM_IRQFD_FLAG_DEASSIGN
,
1427 if (!kvm_enabled() || !kvm_irqchip_in_kernel())
1430 r
= kvm_vm_ioctl(kvm_state
, KVM_IRQFD
, &irqfd
);
1436 int kvm_on_sigbus_vcpu(CPUState
*env
, int code
, void *addr
)
1438 return kvm_arch_on_sigbus_vcpu(env
, code
, addr
);
1441 int kvm_on_sigbus(int code
, void *addr
)
1443 return kvm_arch_on_sigbus(code
, addr
);
1447 #include "qemu-kvm.c"