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"
25 #include "qemu-option.h"
26 #include "qemu-config.h"
34 #include "exec-memory.h"
35 #include "event_notifier.h"
37 /* This check must be after config-host.h is included */
39 #include <sys/eventfd.h>
42 #ifdef CONFIG_VALGRIND_H
43 #include <valgrind/memcheck.h>
46 /* KVM uses PAGE_SIZE in its definition of COALESCED_MMIO_MAX */
47 #define PAGE_SIZE TARGET_PAGE_SIZE
52 #define DPRINTF(fmt, ...) \
53 do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
55 #define DPRINTF(fmt, ...) \
59 #define KVM_MSI_HASHTAB_SIZE 256
61 typedef struct KVMSlot
64 ram_addr_t memory_size
;
70 typedef struct kvm_dirty_log KVMDirtyLog
;
78 struct kvm_coalesced_mmio_ring
*coalesced_mmio_ring
;
79 bool coalesced_flush_in_progress
;
80 int broken_set_mem_region
;
83 int robust_singlestep
;
85 #ifdef KVM_CAP_SET_GUEST_DEBUG
86 struct kvm_sw_breakpoint_head kvm_sw_breakpoints
;
92 /* The man page (and posix) say ioctl numbers are signed int, but
93 * they're not. Linux, glibc and *BSD all treat ioctl numbers as
94 * unsigned, and treating them as signed here can break things */
95 unsigned irq_set_ioctl
;
96 #ifdef KVM_CAP_IRQ_ROUTING
97 struct kvm_irq_routing
*irq_routes
;
98 int nr_allocated_irq_routes
;
99 uint32_t *used_gsi_bitmap
;
100 unsigned int gsi_count
;
101 QTAILQ_HEAD(msi_hashtab
, KVMMSIRoute
) msi_hashtab
[KVM_MSI_HASHTAB_SIZE
];
107 bool kvm_kernel_irqchip
;
108 bool kvm_async_interrupts_allowed
;
109 bool kvm_irqfds_allowed
;
110 bool kvm_msi_via_irqfd_allowed
;
111 bool kvm_gsi_routing_allowed
;
113 static const KVMCapabilityInfo kvm_required_capabilites
[] = {
114 KVM_CAP_INFO(USER_MEMORY
),
115 KVM_CAP_INFO(DESTROY_MEMORY_REGION_WORKS
),
119 static KVMSlot
*kvm_alloc_slot(KVMState
*s
)
123 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
124 if (s
->slots
[i
].memory_size
== 0) {
129 fprintf(stderr
, "%s: no free slot available\n", __func__
);
133 static KVMSlot
*kvm_lookup_matching_slot(KVMState
*s
,
139 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
140 KVMSlot
*mem
= &s
->slots
[i
];
142 if (start_addr
== mem
->start_addr
&&
143 end_addr
== mem
->start_addr
+ mem
->memory_size
) {
152 * Find overlapping slot with lowest start address
154 static KVMSlot
*kvm_lookup_overlapping_slot(KVMState
*s
,
158 KVMSlot
*found
= NULL
;
161 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
162 KVMSlot
*mem
= &s
->slots
[i
];
164 if (mem
->memory_size
== 0 ||
165 (found
&& found
->start_addr
< mem
->start_addr
)) {
169 if (end_addr
> mem
->start_addr
&&
170 start_addr
< mem
->start_addr
+ mem
->memory_size
) {
178 int kvm_physical_memory_addr_from_host(KVMState
*s
, void *ram
,
183 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
184 KVMSlot
*mem
= &s
->slots
[i
];
186 if (ram
>= mem
->ram
&& ram
< mem
->ram
+ mem
->memory_size
) {
187 *phys_addr
= mem
->start_addr
+ (ram
- mem
->ram
);
195 static int kvm_set_user_memory_region(KVMState
*s
, KVMSlot
*slot
)
197 struct kvm_userspace_memory_region mem
;
199 mem
.slot
= slot
->slot
;
200 mem
.guest_phys_addr
= slot
->start_addr
;
201 mem
.memory_size
= slot
->memory_size
;
202 mem
.userspace_addr
= (unsigned long)slot
->ram
;
203 mem
.flags
= slot
->flags
;
204 if (s
->migration_log
) {
205 mem
.flags
|= KVM_MEM_LOG_DIRTY_PAGES
;
207 return kvm_vm_ioctl(s
, KVM_SET_USER_MEMORY_REGION
, &mem
);
210 static void kvm_reset_vcpu(void *opaque
)
212 CPUState
*cpu
= opaque
;
214 kvm_arch_reset_vcpu(cpu
);
217 int kvm_init_vcpu(CPUArchState
*env
)
219 CPUState
*cpu
= ENV_GET_CPU(env
);
220 KVMState
*s
= kvm_state
;
224 DPRINTF("kvm_init_vcpu\n");
226 ret
= kvm_vm_ioctl(s
, KVM_CREATE_VCPU
, env
->cpu_index
);
228 DPRINTF("kvm_create_vcpu failed\n");
234 cpu
->kvm_vcpu_dirty
= true;
236 mmap_size
= kvm_ioctl(s
, KVM_GET_VCPU_MMAP_SIZE
, 0);
239 DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n");
243 cpu
->kvm_run
= mmap(NULL
, mmap_size
, PROT_READ
| PROT_WRITE
, MAP_SHARED
,
245 if (cpu
->kvm_run
== MAP_FAILED
) {
247 DPRINTF("mmap'ing vcpu state failed\n");
251 if (s
->coalesced_mmio
&& !s
->coalesced_mmio_ring
) {
252 s
->coalesced_mmio_ring
=
253 (void *)cpu
->kvm_run
+ s
->coalesced_mmio
* PAGE_SIZE
;
256 ret
= kvm_arch_init_vcpu(cpu
);
258 qemu_register_reset(kvm_reset_vcpu
, cpu
);
259 kvm_arch_reset_vcpu(cpu
);
266 * dirty pages logging control
269 static int kvm_mem_flags(KVMState
*s
, bool log_dirty
)
271 return log_dirty
? KVM_MEM_LOG_DIRTY_PAGES
: 0;
274 static int kvm_slot_dirty_pages_log_change(KVMSlot
*mem
, bool log_dirty
)
276 KVMState
*s
= kvm_state
;
277 int flags
, mask
= KVM_MEM_LOG_DIRTY_PAGES
;
280 old_flags
= mem
->flags
;
282 flags
= (mem
->flags
& ~mask
) | kvm_mem_flags(s
, log_dirty
);
285 /* If nothing changed effectively, no need to issue ioctl */
286 if (s
->migration_log
) {
287 flags
|= KVM_MEM_LOG_DIRTY_PAGES
;
290 if (flags
== old_flags
) {
294 return kvm_set_user_memory_region(s
, mem
);
297 static int kvm_dirty_pages_log_change(hwaddr phys_addr
,
298 ram_addr_t size
, bool log_dirty
)
300 KVMState
*s
= kvm_state
;
301 KVMSlot
*mem
= kvm_lookup_matching_slot(s
, phys_addr
, phys_addr
+ size
);
304 fprintf(stderr
, "BUG: %s: invalid parameters " TARGET_FMT_plx
"-"
305 TARGET_FMT_plx
"\n", __func__
, phys_addr
,
306 (hwaddr
)(phys_addr
+ size
- 1));
309 return kvm_slot_dirty_pages_log_change(mem
, log_dirty
);
312 static void kvm_log_start(MemoryListener
*listener
,
313 MemoryRegionSection
*section
)
317 r
= kvm_dirty_pages_log_change(section
->offset_within_address_space
,
318 section
->size
, true);
324 static void kvm_log_stop(MemoryListener
*listener
,
325 MemoryRegionSection
*section
)
329 r
= kvm_dirty_pages_log_change(section
->offset_within_address_space
,
330 section
->size
, false);
336 static int kvm_set_migration_log(int enable
)
338 KVMState
*s
= kvm_state
;
342 s
->migration_log
= enable
;
344 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
347 if (!mem
->memory_size
) {
350 if (!!(mem
->flags
& KVM_MEM_LOG_DIRTY_PAGES
) == enable
) {
353 err
= kvm_set_user_memory_region(s
, mem
);
361 /* get kvm's dirty pages bitmap and update qemu's */
362 static int kvm_get_dirty_pages_log_range(MemoryRegionSection
*section
,
363 unsigned long *bitmap
)
366 unsigned long page_number
, c
;
368 unsigned int len
= ((section
->size
/ getpagesize()) + HOST_LONG_BITS
- 1) / HOST_LONG_BITS
;
369 unsigned long hpratio
= getpagesize() / TARGET_PAGE_SIZE
;
372 * bitmap-traveling is faster than memory-traveling (for addr...)
373 * especially when most of the memory is not dirty.
375 for (i
= 0; i
< len
; i
++) {
376 if (bitmap
[i
] != 0) {
377 c
= leul_to_cpu(bitmap
[i
]);
381 page_number
= (i
* HOST_LONG_BITS
+ j
) * hpratio
;
382 addr1
= page_number
* TARGET_PAGE_SIZE
;
383 addr
= section
->offset_within_region
+ addr1
;
384 memory_region_set_dirty(section
->mr
, addr
,
385 TARGET_PAGE_SIZE
* hpratio
);
392 #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1))
395 * kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space
396 * This function updates qemu's dirty bitmap using
397 * memory_region_set_dirty(). This means all bits are set
400 * @start_add: start of logged region.
401 * @end_addr: end of logged region.
403 static int kvm_physical_sync_dirty_bitmap(MemoryRegionSection
*section
)
405 KVMState
*s
= kvm_state
;
406 unsigned long size
, allocated_size
= 0;
410 hwaddr start_addr
= section
->offset_within_address_space
;
411 hwaddr end_addr
= start_addr
+ section
->size
;
413 d
.dirty_bitmap
= NULL
;
414 while (start_addr
< end_addr
) {
415 mem
= kvm_lookup_overlapping_slot(s
, start_addr
, end_addr
);
420 /* XXX bad kernel interface alert
421 * For dirty bitmap, kernel allocates array of size aligned to
422 * bits-per-long. But for case when the kernel is 64bits and
423 * the userspace is 32bits, userspace can't align to the same
424 * bits-per-long, since sizeof(long) is different between kernel
425 * and user space. This way, userspace will provide buffer which
426 * may be 4 bytes less than the kernel will use, resulting in
427 * userspace memory corruption (which is not detectable by valgrind
428 * too, in most cases).
429 * So for now, let's align to 64 instead of HOST_LONG_BITS here, in
430 * a hope that sizeof(long) wont become >8 any time soon.
432 size
= ALIGN(((mem
->memory_size
) >> TARGET_PAGE_BITS
),
433 /*HOST_LONG_BITS*/ 64) / 8;
434 if (!d
.dirty_bitmap
) {
435 d
.dirty_bitmap
= g_malloc(size
);
436 } else if (size
> allocated_size
) {
437 d
.dirty_bitmap
= g_realloc(d
.dirty_bitmap
, size
);
439 allocated_size
= size
;
440 memset(d
.dirty_bitmap
, 0, allocated_size
);
444 if (kvm_vm_ioctl(s
, KVM_GET_DIRTY_LOG
, &d
) == -1) {
445 DPRINTF("ioctl failed %d\n", errno
);
450 kvm_get_dirty_pages_log_range(section
, d
.dirty_bitmap
);
451 start_addr
= mem
->start_addr
+ mem
->memory_size
;
453 g_free(d
.dirty_bitmap
);
458 static void kvm_coalesce_mmio_region(MemoryListener
*listener
,
459 MemoryRegionSection
*secion
,
460 hwaddr start
, hwaddr size
)
462 KVMState
*s
= kvm_state
;
464 if (s
->coalesced_mmio
) {
465 struct kvm_coalesced_mmio_zone zone
;
471 (void)kvm_vm_ioctl(s
, KVM_REGISTER_COALESCED_MMIO
, &zone
);
475 static void kvm_uncoalesce_mmio_region(MemoryListener
*listener
,
476 MemoryRegionSection
*secion
,
477 hwaddr start
, hwaddr size
)
479 KVMState
*s
= kvm_state
;
481 if (s
->coalesced_mmio
) {
482 struct kvm_coalesced_mmio_zone zone
;
488 (void)kvm_vm_ioctl(s
, KVM_UNREGISTER_COALESCED_MMIO
, &zone
);
492 int kvm_check_extension(KVMState
*s
, unsigned int extension
)
496 ret
= kvm_ioctl(s
, KVM_CHECK_EXTENSION
, extension
);
504 static int kvm_check_many_ioeventfds(void)
506 /* Userspace can use ioeventfd for io notification. This requires a host
507 * that supports eventfd(2) and an I/O thread; since eventfd does not
508 * support SIGIO it cannot interrupt the vcpu.
510 * Older kernels have a 6 device limit on the KVM io bus. Find out so we
511 * can avoid creating too many ioeventfds.
513 #if defined(CONFIG_EVENTFD)
516 for (i
= 0; i
< ARRAY_SIZE(ioeventfds
); i
++) {
517 ioeventfds
[i
] = eventfd(0, EFD_CLOEXEC
);
518 if (ioeventfds
[i
] < 0) {
521 ret
= kvm_set_ioeventfd_pio_word(ioeventfds
[i
], 0, i
, true);
523 close(ioeventfds
[i
]);
528 /* Decide whether many devices are supported or not */
529 ret
= i
== ARRAY_SIZE(ioeventfds
);
532 kvm_set_ioeventfd_pio_word(ioeventfds
[i
], 0, i
, false);
533 close(ioeventfds
[i
]);
541 static const KVMCapabilityInfo
*
542 kvm_check_extension_list(KVMState
*s
, const KVMCapabilityInfo
*list
)
545 if (!kvm_check_extension(s
, list
->value
)) {
553 static void kvm_set_phys_mem(MemoryRegionSection
*section
, bool add
)
555 KVMState
*s
= kvm_state
;
558 MemoryRegion
*mr
= section
->mr
;
559 bool log_dirty
= memory_region_is_logging(mr
);
560 hwaddr start_addr
= section
->offset_within_address_space
;
561 ram_addr_t size
= section
->size
;
565 /* kvm works in page size chunks, but the function may be called
566 with sub-page size and unaligned start address. */
567 delta
= TARGET_PAGE_ALIGN(size
) - size
;
573 size
&= TARGET_PAGE_MASK
;
574 if (!size
|| (start_addr
& ~TARGET_PAGE_MASK
)) {
578 if (!memory_region_is_ram(mr
)) {
582 ram
= memory_region_get_ram_ptr(mr
) + section
->offset_within_region
+ delta
;
585 mem
= kvm_lookup_overlapping_slot(s
, start_addr
, start_addr
+ size
);
590 if (add
&& start_addr
>= mem
->start_addr
&&
591 (start_addr
+ size
<= mem
->start_addr
+ mem
->memory_size
) &&
592 (ram
- start_addr
== mem
->ram
- mem
->start_addr
)) {
593 /* The new slot fits into the existing one and comes with
594 * identical parameters - update flags and done. */
595 kvm_slot_dirty_pages_log_change(mem
, log_dirty
);
601 if (mem
->flags
& KVM_MEM_LOG_DIRTY_PAGES
) {
602 kvm_physical_sync_dirty_bitmap(section
);
605 /* unregister the overlapping slot */
606 mem
->memory_size
= 0;
607 err
= kvm_set_user_memory_region(s
, mem
);
609 fprintf(stderr
, "%s: error unregistering overlapping slot: %s\n",
610 __func__
, strerror(-err
));
614 /* Workaround for older KVM versions: we can't join slots, even not by
615 * unregistering the previous ones and then registering the larger
616 * slot. We have to maintain the existing fragmentation. Sigh.
618 * This workaround assumes that the new slot starts at the same
619 * address as the first existing one. If not or if some overlapping
620 * slot comes around later, we will fail (not seen in practice so far)
621 * - and actually require a recent KVM version. */
622 if (s
->broken_set_mem_region
&&
623 old
.start_addr
== start_addr
&& old
.memory_size
< size
&& add
) {
624 mem
= kvm_alloc_slot(s
);
625 mem
->memory_size
= old
.memory_size
;
626 mem
->start_addr
= old
.start_addr
;
628 mem
->flags
= kvm_mem_flags(s
, log_dirty
);
630 err
= kvm_set_user_memory_region(s
, mem
);
632 fprintf(stderr
, "%s: error updating slot: %s\n", __func__
,
637 start_addr
+= old
.memory_size
;
638 ram
+= old
.memory_size
;
639 size
-= old
.memory_size
;
643 /* register prefix slot */
644 if (old
.start_addr
< start_addr
) {
645 mem
= kvm_alloc_slot(s
);
646 mem
->memory_size
= start_addr
- old
.start_addr
;
647 mem
->start_addr
= old
.start_addr
;
649 mem
->flags
= kvm_mem_flags(s
, log_dirty
);
651 err
= kvm_set_user_memory_region(s
, mem
);
653 fprintf(stderr
, "%s: error registering prefix slot: %s\n",
654 __func__
, strerror(-err
));
656 fprintf(stderr
, "%s: This is probably because your kernel's " \
657 "PAGE_SIZE is too big. Please try to use 4k " \
658 "PAGE_SIZE!\n", __func__
);
664 /* register suffix slot */
665 if (old
.start_addr
+ old
.memory_size
> start_addr
+ size
) {
666 ram_addr_t size_delta
;
668 mem
= kvm_alloc_slot(s
);
669 mem
->start_addr
= start_addr
+ size
;
670 size_delta
= mem
->start_addr
- old
.start_addr
;
671 mem
->memory_size
= old
.memory_size
- size_delta
;
672 mem
->ram
= old
.ram
+ size_delta
;
673 mem
->flags
= kvm_mem_flags(s
, log_dirty
);
675 err
= kvm_set_user_memory_region(s
, mem
);
677 fprintf(stderr
, "%s: error registering suffix slot: %s\n",
678 __func__
, strerror(-err
));
684 /* in case the KVM bug workaround already "consumed" the new slot */
691 mem
= kvm_alloc_slot(s
);
692 mem
->memory_size
= size
;
693 mem
->start_addr
= start_addr
;
695 mem
->flags
= kvm_mem_flags(s
, log_dirty
);
697 err
= kvm_set_user_memory_region(s
, mem
);
699 fprintf(stderr
, "%s: error registering slot: %s\n", __func__
,
705 static void kvm_region_add(MemoryListener
*listener
,
706 MemoryRegionSection
*section
)
708 kvm_set_phys_mem(section
, true);
711 static void kvm_region_del(MemoryListener
*listener
,
712 MemoryRegionSection
*section
)
714 kvm_set_phys_mem(section
, false);
717 static void kvm_log_sync(MemoryListener
*listener
,
718 MemoryRegionSection
*section
)
722 r
= kvm_physical_sync_dirty_bitmap(section
);
728 static void kvm_log_global_start(struct MemoryListener
*listener
)
732 r
= kvm_set_migration_log(1);
736 static void kvm_log_global_stop(struct MemoryListener
*listener
)
740 r
= kvm_set_migration_log(0);
744 static void kvm_mem_ioeventfd_add(MemoryListener
*listener
,
745 MemoryRegionSection
*section
,
746 bool match_data
, uint64_t data
,
749 int fd
= event_notifier_get_fd(e
);
752 assert(match_data
&& section
->size
<= 8);
754 r
= kvm_set_ioeventfd_mmio(fd
, section
->offset_within_address_space
,
755 data
, true, section
->size
);
761 static void kvm_mem_ioeventfd_del(MemoryListener
*listener
,
762 MemoryRegionSection
*section
,
763 bool match_data
, uint64_t data
,
766 int fd
= event_notifier_get_fd(e
);
769 r
= kvm_set_ioeventfd_mmio(fd
, section
->offset_within_address_space
,
770 data
, false, section
->size
);
776 static void kvm_io_ioeventfd_add(MemoryListener
*listener
,
777 MemoryRegionSection
*section
,
778 bool match_data
, uint64_t data
,
781 int fd
= event_notifier_get_fd(e
);
784 assert(match_data
&& section
->size
== 2);
786 r
= kvm_set_ioeventfd_pio_word(fd
, section
->offset_within_address_space
,
793 static void kvm_io_ioeventfd_del(MemoryListener
*listener
,
794 MemoryRegionSection
*section
,
795 bool match_data
, uint64_t data
,
799 int fd
= event_notifier_get_fd(e
);
802 r
= kvm_set_ioeventfd_pio_word(fd
, section
->offset_within_address_space
,
809 static MemoryListener kvm_memory_listener
= {
810 .region_add
= kvm_region_add
,
811 .region_del
= kvm_region_del
,
812 .log_start
= kvm_log_start
,
813 .log_stop
= kvm_log_stop
,
814 .log_sync
= kvm_log_sync
,
815 .log_global_start
= kvm_log_global_start
,
816 .log_global_stop
= kvm_log_global_stop
,
817 .eventfd_add
= kvm_mem_ioeventfd_add
,
818 .eventfd_del
= kvm_mem_ioeventfd_del
,
819 .coalesced_mmio_add
= kvm_coalesce_mmio_region
,
820 .coalesced_mmio_del
= kvm_uncoalesce_mmio_region
,
824 static MemoryListener kvm_io_listener
= {
825 .eventfd_add
= kvm_io_ioeventfd_add
,
826 .eventfd_del
= kvm_io_ioeventfd_del
,
830 static void kvm_handle_interrupt(CPUArchState
*env
, int mask
)
832 CPUState
*cpu
= ENV_GET_CPU(env
);
834 env
->interrupt_request
|= mask
;
836 if (!qemu_cpu_is_self(cpu
)) {
841 int kvm_set_irq(KVMState
*s
, int irq
, int level
)
843 struct kvm_irq_level event
;
846 assert(kvm_async_interrupts_enabled());
850 ret
= kvm_vm_ioctl(s
, s
->irq_set_ioctl
, &event
);
852 perror("kvm_set_irq");
856 return (s
->irq_set_ioctl
== KVM_IRQ_LINE
) ? 1 : event
.status
;
859 #ifdef KVM_CAP_IRQ_ROUTING
860 typedef struct KVMMSIRoute
{
861 struct kvm_irq_routing_entry kroute
;
862 QTAILQ_ENTRY(KVMMSIRoute
) entry
;
865 static void set_gsi(KVMState
*s
, unsigned int gsi
)
867 s
->used_gsi_bitmap
[gsi
/ 32] |= 1U << (gsi
% 32);
870 static void clear_gsi(KVMState
*s
, unsigned int gsi
)
872 s
->used_gsi_bitmap
[gsi
/ 32] &= ~(1U << (gsi
% 32));
875 static void kvm_init_irq_routing(KVMState
*s
)
879 gsi_count
= kvm_check_extension(s
, KVM_CAP_IRQ_ROUTING
);
881 unsigned int gsi_bits
, i
;
883 /* Round up so we can search ints using ffs */
884 gsi_bits
= ALIGN(gsi_count
, 32);
885 s
->used_gsi_bitmap
= g_malloc0(gsi_bits
/ 8);
886 s
->gsi_count
= gsi_count
;
888 /* Mark any over-allocated bits as already in use */
889 for (i
= gsi_count
; i
< gsi_bits
; i
++) {
894 s
->irq_routes
= g_malloc0(sizeof(*s
->irq_routes
));
895 s
->nr_allocated_irq_routes
= 0;
897 if (!s
->direct_msi
) {
898 for (i
= 0; i
< KVM_MSI_HASHTAB_SIZE
; i
++) {
899 QTAILQ_INIT(&s
->msi_hashtab
[i
]);
903 kvm_arch_init_irq_routing(s
);
906 static void kvm_irqchip_commit_routes(KVMState
*s
)
910 s
->irq_routes
->flags
= 0;
911 ret
= kvm_vm_ioctl(s
, KVM_SET_GSI_ROUTING
, s
->irq_routes
);
915 static void kvm_add_routing_entry(KVMState
*s
,
916 struct kvm_irq_routing_entry
*entry
)
918 struct kvm_irq_routing_entry
*new;
921 if (s
->irq_routes
->nr
== s
->nr_allocated_irq_routes
) {
922 n
= s
->nr_allocated_irq_routes
* 2;
926 size
= sizeof(struct kvm_irq_routing
);
927 size
+= n
* sizeof(*new);
928 s
->irq_routes
= g_realloc(s
->irq_routes
, size
);
929 s
->nr_allocated_irq_routes
= n
;
931 n
= s
->irq_routes
->nr
++;
932 new = &s
->irq_routes
->entries
[n
];
933 memset(new, 0, sizeof(*new));
934 new->gsi
= entry
->gsi
;
935 new->type
= entry
->type
;
936 new->flags
= entry
->flags
;
939 set_gsi(s
, entry
->gsi
);
941 kvm_irqchip_commit_routes(s
);
944 static int kvm_update_routing_entry(KVMState
*s
,
945 struct kvm_irq_routing_entry
*new_entry
)
947 struct kvm_irq_routing_entry
*entry
;
950 for (n
= 0; n
< s
->irq_routes
->nr
; n
++) {
951 entry
= &s
->irq_routes
->entries
[n
];
952 if (entry
->gsi
!= new_entry
->gsi
) {
956 entry
->type
= new_entry
->type
;
957 entry
->flags
= new_entry
->flags
;
958 entry
->u
= new_entry
->u
;
960 kvm_irqchip_commit_routes(s
);
968 void kvm_irqchip_add_irq_route(KVMState
*s
, int irq
, int irqchip
, int pin
)
970 struct kvm_irq_routing_entry e
;
972 assert(pin
< s
->gsi_count
);
975 e
.type
= KVM_IRQ_ROUTING_IRQCHIP
;
977 e
.u
.irqchip
.irqchip
= irqchip
;
978 e
.u
.irqchip
.pin
= pin
;
979 kvm_add_routing_entry(s
, &e
);
982 void kvm_irqchip_release_virq(KVMState
*s
, int virq
)
984 struct kvm_irq_routing_entry
*e
;
987 for (i
= 0; i
< s
->irq_routes
->nr
; i
++) {
988 e
= &s
->irq_routes
->entries
[i
];
989 if (e
->gsi
== virq
) {
991 *e
= s
->irq_routes
->entries
[s
->irq_routes
->nr
];
996 kvm_irqchip_commit_routes(s
);
999 static unsigned int kvm_hash_msi(uint32_t data
)
1001 /* This is optimized for IA32 MSI layout. However, no other arch shall
1002 * repeat the mistake of not providing a direct MSI injection API. */
1006 static void kvm_flush_dynamic_msi_routes(KVMState
*s
)
1008 KVMMSIRoute
*route
, *next
;
1011 for (hash
= 0; hash
< KVM_MSI_HASHTAB_SIZE
; hash
++) {
1012 QTAILQ_FOREACH_SAFE(route
, &s
->msi_hashtab
[hash
], entry
, next
) {
1013 kvm_irqchip_release_virq(s
, route
->kroute
.gsi
);
1014 QTAILQ_REMOVE(&s
->msi_hashtab
[hash
], route
, entry
);
1020 static int kvm_irqchip_get_virq(KVMState
*s
)
1022 uint32_t *word
= s
->used_gsi_bitmap
;
1023 int max_words
= ALIGN(s
->gsi_count
, 32) / 32;
1028 /* Return the lowest unused GSI in the bitmap */
1029 for (i
= 0; i
< max_words
; i
++) {
1030 bit
= ffs(~word
[i
]);
1035 return bit
- 1 + i
* 32;
1037 if (!s
->direct_msi
&& retry
) {
1039 kvm_flush_dynamic_msi_routes(s
);
1046 static KVMMSIRoute
*kvm_lookup_msi_route(KVMState
*s
, MSIMessage msg
)
1048 unsigned int hash
= kvm_hash_msi(msg
.data
);
1051 QTAILQ_FOREACH(route
, &s
->msi_hashtab
[hash
], entry
) {
1052 if (route
->kroute
.u
.msi
.address_lo
== (uint32_t)msg
.address
&&
1053 route
->kroute
.u
.msi
.address_hi
== (msg
.address
>> 32) &&
1054 route
->kroute
.u
.msi
.data
== msg
.data
) {
1061 int kvm_irqchip_send_msi(KVMState
*s
, MSIMessage msg
)
1066 if (s
->direct_msi
) {
1067 msi
.address_lo
= (uint32_t)msg
.address
;
1068 msi
.address_hi
= msg
.address
>> 32;
1069 msi
.data
= msg
.data
;
1071 memset(msi
.pad
, 0, sizeof(msi
.pad
));
1073 return kvm_vm_ioctl(s
, KVM_SIGNAL_MSI
, &msi
);
1076 route
= kvm_lookup_msi_route(s
, msg
);
1080 virq
= kvm_irqchip_get_virq(s
);
1085 route
= g_malloc(sizeof(KVMMSIRoute
));
1086 route
->kroute
.gsi
= virq
;
1087 route
->kroute
.type
= KVM_IRQ_ROUTING_MSI
;
1088 route
->kroute
.flags
= 0;
1089 route
->kroute
.u
.msi
.address_lo
= (uint32_t)msg
.address
;
1090 route
->kroute
.u
.msi
.address_hi
= msg
.address
>> 32;
1091 route
->kroute
.u
.msi
.data
= msg
.data
;
1093 kvm_add_routing_entry(s
, &route
->kroute
);
1095 QTAILQ_INSERT_TAIL(&s
->msi_hashtab
[kvm_hash_msi(msg
.data
)], route
,
1099 assert(route
->kroute
.type
== KVM_IRQ_ROUTING_MSI
);
1101 return kvm_set_irq(s
, route
->kroute
.gsi
, 1);
1104 int kvm_irqchip_add_msi_route(KVMState
*s
, MSIMessage msg
)
1106 struct kvm_irq_routing_entry kroute
;
1109 if (!kvm_gsi_routing_enabled()) {
1113 virq
= kvm_irqchip_get_virq(s
);
1119 kroute
.type
= KVM_IRQ_ROUTING_MSI
;
1121 kroute
.u
.msi
.address_lo
= (uint32_t)msg
.address
;
1122 kroute
.u
.msi
.address_hi
= msg
.address
>> 32;
1123 kroute
.u
.msi
.data
= msg
.data
;
1125 kvm_add_routing_entry(s
, &kroute
);
1130 int kvm_irqchip_update_msi_route(KVMState
*s
, int virq
, MSIMessage msg
)
1132 struct kvm_irq_routing_entry kroute
;
1134 if (!kvm_irqchip_in_kernel()) {
1139 kroute
.type
= KVM_IRQ_ROUTING_MSI
;
1141 kroute
.u
.msi
.address_lo
= (uint32_t)msg
.address
;
1142 kroute
.u
.msi
.address_hi
= msg
.address
>> 32;
1143 kroute
.u
.msi
.data
= msg
.data
;
1145 return kvm_update_routing_entry(s
, &kroute
);
1148 static int kvm_irqchip_assign_irqfd(KVMState
*s
, int fd
, int virq
, bool assign
)
1150 struct kvm_irqfd irqfd
= {
1153 .flags
= assign
? 0 : KVM_IRQFD_FLAG_DEASSIGN
,
1156 if (!kvm_irqfds_enabled()) {
1160 return kvm_vm_ioctl(s
, KVM_IRQFD
, &irqfd
);
1163 #else /* !KVM_CAP_IRQ_ROUTING */
1165 static void kvm_init_irq_routing(KVMState
*s
)
1169 void kvm_irqchip_release_virq(KVMState
*s
, int virq
)
1173 int kvm_irqchip_send_msi(KVMState
*s
, MSIMessage msg
)
1178 int kvm_irqchip_add_msi_route(KVMState
*s
, MSIMessage msg
)
1183 static int kvm_irqchip_assign_irqfd(KVMState
*s
, int fd
, int virq
, bool assign
)
1187 #endif /* !KVM_CAP_IRQ_ROUTING */
1189 int kvm_irqchip_add_irqfd_notifier(KVMState
*s
, EventNotifier
*n
, int virq
)
1191 return kvm_irqchip_assign_irqfd(s
, event_notifier_get_fd(n
), virq
, true);
1194 int kvm_irqchip_remove_irqfd_notifier(KVMState
*s
, EventNotifier
*n
, int virq
)
1196 return kvm_irqchip_assign_irqfd(s
, event_notifier_get_fd(n
), virq
, false);
1199 static int kvm_irqchip_create(KVMState
*s
)
1201 QemuOptsList
*list
= qemu_find_opts("machine");
1204 if (QTAILQ_EMPTY(&list
->head
) ||
1205 !qemu_opt_get_bool(QTAILQ_FIRST(&list
->head
),
1206 "kernel_irqchip", true) ||
1207 !kvm_check_extension(s
, KVM_CAP_IRQCHIP
)) {
1211 ret
= kvm_vm_ioctl(s
, KVM_CREATE_IRQCHIP
);
1213 fprintf(stderr
, "Create kernel irqchip failed\n");
1217 kvm_kernel_irqchip
= true;
1218 /* If we have an in-kernel IRQ chip then we must have asynchronous
1219 * interrupt delivery (though the reverse is not necessarily true)
1221 kvm_async_interrupts_allowed
= true;
1223 kvm_init_irq_routing(s
);
1228 static int kvm_max_vcpus(KVMState
*s
)
1232 /* Find number of supported CPUs using the recommended
1233 * procedure from the kernel API documentation to cope with
1234 * older kernels that may be missing capabilities.
1236 ret
= kvm_check_extension(s
, KVM_CAP_MAX_VCPUS
);
1240 ret
= kvm_check_extension(s
, KVM_CAP_NR_VCPUS
);
1250 static const char upgrade_note
[] =
1251 "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n"
1252 "(see http://sourceforge.net/projects/kvm).\n";
1254 const KVMCapabilityInfo
*missing_cap
;
1259 s
= g_malloc0(sizeof(KVMState
));
1262 * On systems where the kernel can support different base page
1263 * sizes, host page size may be different from TARGET_PAGE_SIZE,
1264 * even with KVM. TARGET_PAGE_SIZE is assumed to be the minimum
1265 * page size for the system though.
1267 assert(TARGET_PAGE_SIZE
<= getpagesize());
1269 #ifdef KVM_CAP_SET_GUEST_DEBUG
1270 QTAILQ_INIT(&s
->kvm_sw_breakpoints
);
1272 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
1273 s
->slots
[i
].slot
= i
;
1276 s
->fd
= qemu_open("/dev/kvm", O_RDWR
);
1278 fprintf(stderr
, "Could not access KVM kernel module: %m\n");
1283 ret
= kvm_ioctl(s
, KVM_GET_API_VERSION
, 0);
1284 if (ret
< KVM_API_VERSION
) {
1288 fprintf(stderr
, "kvm version too old\n");
1292 if (ret
> KVM_API_VERSION
) {
1294 fprintf(stderr
, "kvm version not supported\n");
1298 max_vcpus
= kvm_max_vcpus(s
);
1299 if (smp_cpus
> max_vcpus
) {
1301 fprintf(stderr
, "Number of SMP cpus requested (%d) exceeds max cpus "
1302 "supported by KVM (%d)\n", smp_cpus
, max_vcpus
);
1306 s
->vmfd
= kvm_ioctl(s
, KVM_CREATE_VM
, 0);
1309 fprintf(stderr
, "Please add the 'switch_amode' kernel parameter to "
1310 "your host kernel command line\n");
1316 missing_cap
= kvm_check_extension_list(s
, kvm_required_capabilites
);
1319 kvm_check_extension_list(s
, kvm_arch_required_capabilities
);
1323 fprintf(stderr
, "kvm does not support %s\n%s",
1324 missing_cap
->name
, upgrade_note
);
1328 s
->coalesced_mmio
= kvm_check_extension(s
, KVM_CAP_COALESCED_MMIO
);
1330 s
->broken_set_mem_region
= 1;
1331 ret
= kvm_check_extension(s
, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
);
1333 s
->broken_set_mem_region
= 0;
1336 #ifdef KVM_CAP_VCPU_EVENTS
1337 s
->vcpu_events
= kvm_check_extension(s
, KVM_CAP_VCPU_EVENTS
);
1340 s
->robust_singlestep
=
1341 kvm_check_extension(s
, KVM_CAP_X86_ROBUST_SINGLESTEP
);
1343 #ifdef KVM_CAP_DEBUGREGS
1344 s
->debugregs
= kvm_check_extension(s
, KVM_CAP_DEBUGREGS
);
1347 #ifdef KVM_CAP_XSAVE
1348 s
->xsave
= kvm_check_extension(s
, KVM_CAP_XSAVE
);
1352 s
->xcrs
= kvm_check_extension(s
, KVM_CAP_XCRS
);
1355 #ifdef KVM_CAP_PIT_STATE2
1356 s
->pit_state2
= kvm_check_extension(s
, KVM_CAP_PIT_STATE2
);
1359 #ifdef KVM_CAP_IRQ_ROUTING
1360 s
->direct_msi
= (kvm_check_extension(s
, KVM_CAP_SIGNAL_MSI
) > 0);
1363 s
->intx_set_mask
= kvm_check_extension(s
, KVM_CAP_PCI_2_3
);
1365 s
->irq_set_ioctl
= KVM_IRQ_LINE
;
1366 if (kvm_check_extension(s
, KVM_CAP_IRQ_INJECT_STATUS
)) {
1367 s
->irq_set_ioctl
= KVM_IRQ_LINE_STATUS
;
1370 ret
= kvm_arch_init(s
);
1375 ret
= kvm_irqchip_create(s
);
1381 memory_listener_register(&kvm_memory_listener
, &address_space_memory
);
1382 memory_listener_register(&kvm_io_listener
, &address_space_io
);
1384 s
->many_ioeventfds
= kvm_check_many_ioeventfds();
1386 cpu_interrupt_handler
= kvm_handle_interrupt
;
1402 static void kvm_handle_io(uint16_t port
, void *data
, int direction
, int size
,
1406 uint8_t *ptr
= data
;
1408 for (i
= 0; i
< count
; i
++) {
1409 if (direction
== KVM_EXIT_IO_IN
) {
1412 stb_p(ptr
, cpu_inb(port
));
1415 stw_p(ptr
, cpu_inw(port
));
1418 stl_p(ptr
, cpu_inl(port
));
1424 cpu_outb(port
, ldub_p(ptr
));
1427 cpu_outw(port
, lduw_p(ptr
));
1430 cpu_outl(port
, ldl_p(ptr
));
1439 static int kvm_handle_internal_error(CPUArchState
*env
, struct kvm_run
*run
)
1441 CPUState
*cpu
= ENV_GET_CPU(env
);
1443 fprintf(stderr
, "KVM internal error.");
1444 if (kvm_check_extension(kvm_state
, KVM_CAP_INTERNAL_ERROR_DATA
)) {
1447 fprintf(stderr
, " Suberror: %d\n", run
->internal
.suberror
);
1448 for (i
= 0; i
< run
->internal
.ndata
; ++i
) {
1449 fprintf(stderr
, "extra data[%d]: %"PRIx64
"\n",
1450 i
, (uint64_t)run
->internal
.data
[i
]);
1453 fprintf(stderr
, "\n");
1455 if (run
->internal
.suberror
== KVM_INTERNAL_ERROR_EMULATION
) {
1456 fprintf(stderr
, "emulation failure\n");
1457 if (!kvm_arch_stop_on_emulation_error(cpu
)) {
1458 cpu_dump_state(env
, stderr
, fprintf
, CPU_DUMP_CODE
);
1459 return EXCP_INTERRUPT
;
1462 /* FIXME: Should trigger a qmp message to let management know
1463 * something went wrong.
1468 void kvm_flush_coalesced_mmio_buffer(void)
1470 KVMState
*s
= kvm_state
;
1472 if (s
->coalesced_flush_in_progress
) {
1476 s
->coalesced_flush_in_progress
= true;
1478 if (s
->coalesced_mmio_ring
) {
1479 struct kvm_coalesced_mmio_ring
*ring
= s
->coalesced_mmio_ring
;
1480 while (ring
->first
!= ring
->last
) {
1481 struct kvm_coalesced_mmio
*ent
;
1483 ent
= &ring
->coalesced_mmio
[ring
->first
];
1485 cpu_physical_memory_write(ent
->phys_addr
, ent
->data
, ent
->len
);
1487 ring
->first
= (ring
->first
+ 1) % KVM_COALESCED_MMIO_MAX
;
1491 s
->coalesced_flush_in_progress
= false;
1494 static void do_kvm_cpu_synchronize_state(void *arg
)
1496 CPUState
*cpu
= arg
;
1498 if (!cpu
->kvm_vcpu_dirty
) {
1499 kvm_arch_get_registers(cpu
);
1500 cpu
->kvm_vcpu_dirty
= true;
1504 void kvm_cpu_synchronize_state(CPUArchState
*env
)
1506 CPUState
*cpu
= ENV_GET_CPU(env
);
1508 if (!cpu
->kvm_vcpu_dirty
) {
1509 run_on_cpu(cpu
, do_kvm_cpu_synchronize_state
, cpu
);
1513 void kvm_cpu_synchronize_post_reset(CPUArchState
*env
)
1515 CPUState
*cpu
= ENV_GET_CPU(env
);
1517 kvm_arch_put_registers(cpu
, KVM_PUT_RESET_STATE
);
1518 cpu
->kvm_vcpu_dirty
= false;
1521 void kvm_cpu_synchronize_post_init(CPUArchState
*env
)
1523 CPUState
*cpu
= ENV_GET_CPU(env
);
1525 kvm_arch_put_registers(cpu
, KVM_PUT_FULL_STATE
);
1526 cpu
->kvm_vcpu_dirty
= false;
1529 int kvm_cpu_exec(CPUArchState
*env
)
1531 CPUState
*cpu
= ENV_GET_CPU(env
);
1532 struct kvm_run
*run
= cpu
->kvm_run
;
1535 DPRINTF("kvm_cpu_exec()\n");
1537 if (kvm_arch_process_async_events(cpu
)) {
1538 env
->exit_request
= 0;
1543 if (cpu
->kvm_vcpu_dirty
) {
1544 kvm_arch_put_registers(cpu
, KVM_PUT_RUNTIME_STATE
);
1545 cpu
->kvm_vcpu_dirty
= false;
1548 kvm_arch_pre_run(cpu
, run
);
1549 if (env
->exit_request
) {
1550 DPRINTF("interrupt exit requested\n");
1552 * KVM requires us to reenter the kernel after IO exits to complete
1553 * instruction emulation. This self-signal will ensure that we
1556 qemu_cpu_kick_self();
1558 qemu_mutex_unlock_iothread();
1560 run_ret
= kvm_vcpu_ioctl(cpu
, KVM_RUN
, 0);
1562 qemu_mutex_lock_iothread();
1563 kvm_arch_post_run(cpu
, run
);
1566 if (run_ret
== -EINTR
|| run_ret
== -EAGAIN
) {
1567 DPRINTF("io window exit\n");
1568 ret
= EXCP_INTERRUPT
;
1571 fprintf(stderr
, "error: kvm run failed %s\n",
1572 strerror(-run_ret
));
1576 switch (run
->exit_reason
) {
1578 DPRINTF("handle_io\n");
1579 kvm_handle_io(run
->io
.port
,
1580 (uint8_t *)run
+ run
->io
.data_offset
,
1587 DPRINTF("handle_mmio\n");
1588 cpu_physical_memory_rw(run
->mmio
.phys_addr
,
1591 run
->mmio
.is_write
);
1594 case KVM_EXIT_IRQ_WINDOW_OPEN
:
1595 DPRINTF("irq_window_open\n");
1596 ret
= EXCP_INTERRUPT
;
1598 case KVM_EXIT_SHUTDOWN
:
1599 DPRINTF("shutdown\n");
1600 qemu_system_reset_request();
1601 ret
= EXCP_INTERRUPT
;
1603 case KVM_EXIT_UNKNOWN
:
1604 fprintf(stderr
, "KVM: unknown exit, hardware reason %" PRIx64
"\n",
1605 (uint64_t)run
->hw
.hardware_exit_reason
);
1608 case KVM_EXIT_INTERNAL_ERROR
:
1609 ret
= kvm_handle_internal_error(env
, run
);
1612 DPRINTF("kvm_arch_handle_exit\n");
1613 ret
= kvm_arch_handle_exit(cpu
, run
);
1619 cpu_dump_state(env
, stderr
, fprintf
, CPU_DUMP_CODE
);
1620 vm_stop(RUN_STATE_INTERNAL_ERROR
);
1623 env
->exit_request
= 0;
1627 int kvm_ioctl(KVMState
*s
, int type
, ...)
1634 arg
= va_arg(ap
, void *);
1637 ret
= ioctl(s
->fd
, type
, arg
);
1644 int kvm_vm_ioctl(KVMState
*s
, int type
, ...)
1651 arg
= va_arg(ap
, void *);
1654 ret
= ioctl(s
->vmfd
, type
, arg
);
1661 int kvm_vcpu_ioctl(CPUState
*cpu
, int type
, ...)
1668 arg
= va_arg(ap
, void *);
1671 ret
= ioctl(cpu
->kvm_fd
, type
, arg
);
1678 int kvm_has_sync_mmu(void)
1680 return kvm_check_extension(kvm_state
, KVM_CAP_SYNC_MMU
);
1683 int kvm_has_vcpu_events(void)
1685 return kvm_state
->vcpu_events
;
1688 int kvm_has_robust_singlestep(void)
1690 return kvm_state
->robust_singlestep
;
1693 int kvm_has_debugregs(void)
1695 return kvm_state
->debugregs
;
1698 int kvm_has_xsave(void)
1700 return kvm_state
->xsave
;
1703 int kvm_has_xcrs(void)
1705 return kvm_state
->xcrs
;
1708 int kvm_has_pit_state2(void)
1710 return kvm_state
->pit_state2
;
1713 int kvm_has_many_ioeventfds(void)
1715 if (!kvm_enabled()) {
1718 return kvm_state
->many_ioeventfds
;
1721 int kvm_has_gsi_routing(void)
1723 #ifdef KVM_CAP_IRQ_ROUTING
1724 return kvm_check_extension(kvm_state
, KVM_CAP_IRQ_ROUTING
);
1730 int kvm_has_intx_set_mask(void)
1732 return kvm_state
->intx_set_mask
;
1735 void *kvm_vmalloc(ram_addr_t size
)
1740 mem
= kvm_arch_vmalloc(size
);
1745 return qemu_vmalloc(size
);
1748 void kvm_setup_guest_memory(void *start
, size_t size
)
1750 #ifdef CONFIG_VALGRIND_H
1751 VALGRIND_MAKE_MEM_DEFINED(start
, size
);
1753 if (!kvm_has_sync_mmu()) {
1754 int ret
= qemu_madvise(start
, size
, QEMU_MADV_DONTFORK
);
1757 perror("qemu_madvise");
1759 "Need MADV_DONTFORK in absence of synchronous KVM MMU\n");
1765 #ifdef KVM_CAP_SET_GUEST_DEBUG
1766 struct kvm_sw_breakpoint
*kvm_find_sw_breakpoint(CPUState
*cpu
,
1769 struct kvm_sw_breakpoint
*bp
;
1771 QTAILQ_FOREACH(bp
, &cpu
->kvm_state
->kvm_sw_breakpoints
, entry
) {
1779 int kvm_sw_breakpoints_active(CPUState
*cpu
)
1781 return !QTAILQ_EMPTY(&cpu
->kvm_state
->kvm_sw_breakpoints
);
1784 struct kvm_set_guest_debug_data
{
1785 struct kvm_guest_debug dbg
;
1790 static void kvm_invoke_set_guest_debug(void *data
)
1792 struct kvm_set_guest_debug_data
*dbg_data
= data
;
1794 dbg_data
->err
= kvm_vcpu_ioctl(dbg_data
->cpu
, KVM_SET_GUEST_DEBUG
,
1798 int kvm_update_guest_debug(CPUArchState
*env
, unsigned long reinject_trap
)
1800 CPUState
*cpu
= ENV_GET_CPU(env
);
1801 struct kvm_set_guest_debug_data data
;
1803 data
.dbg
.control
= reinject_trap
;
1805 if (env
->singlestep_enabled
) {
1806 data
.dbg
.control
|= KVM_GUESTDBG_ENABLE
| KVM_GUESTDBG_SINGLESTEP
;
1808 kvm_arch_update_guest_debug(cpu
, &data
.dbg
);
1811 run_on_cpu(cpu
, kvm_invoke_set_guest_debug
, &data
);
1815 int kvm_insert_breakpoint(CPUArchState
*current_env
, target_ulong addr
,
1816 target_ulong len
, int type
)
1818 CPUState
*current_cpu
= ENV_GET_CPU(current_env
);
1819 struct kvm_sw_breakpoint
*bp
;
1823 if (type
== GDB_BREAKPOINT_SW
) {
1824 bp
= kvm_find_sw_breakpoint(current_cpu
, addr
);
1830 bp
= g_malloc(sizeof(struct kvm_sw_breakpoint
));
1837 err
= kvm_arch_insert_sw_breakpoint(current_cpu
, bp
);
1843 QTAILQ_INSERT_HEAD(¤t_cpu
->kvm_state
->kvm_sw_breakpoints
,
1846 err
= kvm_arch_insert_hw_breakpoint(addr
, len
, type
);
1852 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
1853 err
= kvm_update_guest_debug(env
, 0);
1861 int kvm_remove_breakpoint(CPUArchState
*current_env
, target_ulong addr
,
1862 target_ulong len
, int type
)
1864 CPUState
*current_cpu
= ENV_GET_CPU(current_env
);
1865 struct kvm_sw_breakpoint
*bp
;
1869 if (type
== GDB_BREAKPOINT_SW
) {
1870 bp
= kvm_find_sw_breakpoint(current_cpu
, addr
);
1875 if (bp
->use_count
> 1) {
1880 err
= kvm_arch_remove_sw_breakpoint(current_cpu
, bp
);
1885 QTAILQ_REMOVE(¤t_cpu
->kvm_state
->kvm_sw_breakpoints
, bp
, entry
);
1888 err
= kvm_arch_remove_hw_breakpoint(addr
, len
, type
);
1894 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
1895 err
= kvm_update_guest_debug(env
, 0);
1903 void kvm_remove_all_breakpoints(CPUArchState
*current_env
)
1905 CPUState
*current_cpu
= ENV_GET_CPU(current_env
);
1906 struct kvm_sw_breakpoint
*bp
, *next
;
1907 KVMState
*s
= current_cpu
->kvm_state
;
1911 QTAILQ_FOREACH_SAFE(bp
, &s
->kvm_sw_breakpoints
, entry
, next
) {
1912 if (kvm_arch_remove_sw_breakpoint(current_cpu
, bp
) != 0) {
1913 /* Try harder to find a CPU that currently sees the breakpoint. */
1914 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
1915 cpu
= ENV_GET_CPU(env
);
1916 if (kvm_arch_remove_sw_breakpoint(cpu
, bp
) == 0) {
1921 QTAILQ_REMOVE(&s
->kvm_sw_breakpoints
, bp
, entry
);
1924 kvm_arch_remove_all_hw_breakpoints();
1926 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
1927 kvm_update_guest_debug(env
, 0);
1931 #else /* !KVM_CAP_SET_GUEST_DEBUG */
1933 int kvm_update_guest_debug(CPUArchState
*env
, unsigned long reinject_trap
)
1938 int kvm_insert_breakpoint(CPUArchState
*current_env
, target_ulong addr
,
1939 target_ulong len
, int type
)
1944 int kvm_remove_breakpoint(CPUArchState
*current_env
, target_ulong addr
,
1945 target_ulong len
, int type
)
1950 void kvm_remove_all_breakpoints(CPUArchState
*current_env
)
1953 #endif /* !KVM_CAP_SET_GUEST_DEBUG */
1955 int kvm_set_signal_mask(CPUArchState
*env
, const sigset_t
*sigset
)
1957 CPUState
*cpu
= ENV_GET_CPU(env
);
1958 struct kvm_signal_mask
*sigmask
;
1962 return kvm_vcpu_ioctl(cpu
, KVM_SET_SIGNAL_MASK
, NULL
);
1965 sigmask
= g_malloc(sizeof(*sigmask
) + sizeof(*sigset
));
1968 memcpy(sigmask
->sigset
, sigset
, sizeof(*sigset
));
1969 r
= kvm_vcpu_ioctl(cpu
, KVM_SET_SIGNAL_MASK
, sigmask
);
1975 int kvm_set_ioeventfd_mmio(int fd
, uint32_t addr
, uint32_t val
, bool assign
,
1979 struct kvm_ioeventfd iofd
;
1981 iofd
.datamatch
= val
;
1984 iofd
.flags
= KVM_IOEVENTFD_FLAG_DATAMATCH
;
1987 if (!kvm_enabled()) {
1992 iofd
.flags
|= KVM_IOEVENTFD_FLAG_DEASSIGN
;
1995 ret
= kvm_vm_ioctl(kvm_state
, KVM_IOEVENTFD
, &iofd
);
2004 int kvm_set_ioeventfd_pio_word(int fd
, uint16_t addr
, uint16_t val
, bool assign
)
2006 struct kvm_ioeventfd kick
= {
2010 .flags
= KVM_IOEVENTFD_FLAG_DATAMATCH
| KVM_IOEVENTFD_FLAG_PIO
,
2014 if (!kvm_enabled()) {
2018 kick
.flags
|= KVM_IOEVENTFD_FLAG_DEASSIGN
;
2020 r
= kvm_vm_ioctl(kvm_state
, KVM_IOEVENTFD
, &kick
);
2027 int kvm_on_sigbus_vcpu(CPUArchState
*env
, int code
, void *addr
)
2029 CPUState
*cpu
= ENV_GET_CPU(env
);
2030 return kvm_arch_on_sigbus_vcpu(cpu
, code
, addr
);
2033 int kvm_on_sigbus(int code
, void *addr
)
2035 return kvm_arch_on_sigbus(code
, addr
);