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
63 target_phys_addr_t start_addr
;
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
;
91 /* The man page (and posix) say ioctl numbers are signed int, but
92 * they're not. Linux, glibc and *BSD all treat ioctl numbers as
93 * unsigned, and treating them as signed here can break things */
94 unsigned irqchip_inject_ioctl
;
95 #ifdef KVM_CAP_IRQ_ROUTING
96 struct kvm_irq_routing
*irq_routes
;
97 int nr_allocated_irq_routes
;
98 uint32_t *used_gsi_bitmap
;
99 unsigned int gsi_count
;
100 QTAILQ_HEAD(msi_hashtab
, KVMMSIRoute
) msi_hashtab
[KVM_MSI_HASHTAB_SIZE
];
106 bool kvm_kernel_irqchip
;
107 bool kvm_async_interrupts_allowed
;
108 bool kvm_irqfds_allowed
;
109 bool kvm_msi_via_irqfd_allowed
;
110 bool kvm_gsi_routing_allowed
;
112 static const KVMCapabilityInfo kvm_required_capabilites
[] = {
113 KVM_CAP_INFO(USER_MEMORY
),
114 KVM_CAP_INFO(DESTROY_MEMORY_REGION_WORKS
),
118 static KVMSlot
*kvm_alloc_slot(KVMState
*s
)
122 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
123 if (s
->slots
[i
].memory_size
== 0) {
128 fprintf(stderr
, "%s: no free slot available\n", __func__
);
132 static KVMSlot
*kvm_lookup_matching_slot(KVMState
*s
,
133 target_phys_addr_t start_addr
,
134 target_phys_addr_t end_addr
)
138 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
139 KVMSlot
*mem
= &s
->slots
[i
];
141 if (start_addr
== mem
->start_addr
&&
142 end_addr
== mem
->start_addr
+ mem
->memory_size
) {
151 * Find overlapping slot with lowest start address
153 static KVMSlot
*kvm_lookup_overlapping_slot(KVMState
*s
,
154 target_phys_addr_t start_addr
,
155 target_phys_addr_t end_addr
)
157 KVMSlot
*found
= NULL
;
160 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
161 KVMSlot
*mem
= &s
->slots
[i
];
163 if (mem
->memory_size
== 0 ||
164 (found
&& found
->start_addr
< mem
->start_addr
)) {
168 if (end_addr
> mem
->start_addr
&&
169 start_addr
< mem
->start_addr
+ mem
->memory_size
) {
177 int kvm_physical_memory_addr_from_host(KVMState
*s
, void *ram
,
178 target_phys_addr_t
*phys_addr
)
182 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
183 KVMSlot
*mem
= &s
->slots
[i
];
185 if (ram
>= mem
->ram
&& ram
< mem
->ram
+ mem
->memory_size
) {
186 *phys_addr
= mem
->start_addr
+ (ram
- mem
->ram
);
194 static int kvm_set_user_memory_region(KVMState
*s
, KVMSlot
*slot
)
196 struct kvm_userspace_memory_region mem
;
198 mem
.slot
= slot
->slot
;
199 mem
.guest_phys_addr
= slot
->start_addr
;
200 mem
.memory_size
= slot
->memory_size
;
201 mem
.userspace_addr
= (unsigned long)slot
->ram
;
202 mem
.flags
= slot
->flags
;
203 if (s
->migration_log
) {
204 mem
.flags
|= KVM_MEM_LOG_DIRTY_PAGES
;
206 return kvm_vm_ioctl(s
, KVM_SET_USER_MEMORY_REGION
, &mem
);
209 static void kvm_reset_vcpu(void *opaque
)
211 CPUArchState
*env
= opaque
;
213 kvm_arch_reset_vcpu(env
);
216 int kvm_init_vcpu(CPUArchState
*env
)
218 KVMState
*s
= kvm_state
;
222 DPRINTF("kvm_init_vcpu\n");
224 ret
= kvm_vm_ioctl(s
, KVM_CREATE_VCPU
, env
->cpu_index
);
226 DPRINTF("kvm_create_vcpu failed\n");
232 env
->kvm_vcpu_dirty
= 1;
234 mmap_size
= kvm_ioctl(s
, KVM_GET_VCPU_MMAP_SIZE
, 0);
237 DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n");
241 env
->kvm_run
= mmap(NULL
, mmap_size
, PROT_READ
| PROT_WRITE
, MAP_SHARED
,
243 if (env
->kvm_run
== MAP_FAILED
) {
245 DPRINTF("mmap'ing vcpu state failed\n");
249 if (s
->coalesced_mmio
&& !s
->coalesced_mmio_ring
) {
250 s
->coalesced_mmio_ring
=
251 (void *)env
->kvm_run
+ s
->coalesced_mmio
* PAGE_SIZE
;
254 ret
= kvm_arch_init_vcpu(env
);
256 qemu_register_reset(kvm_reset_vcpu
, env
);
257 kvm_arch_reset_vcpu(env
);
264 * dirty pages logging control
267 static int kvm_mem_flags(KVMState
*s
, bool log_dirty
)
269 return log_dirty
? KVM_MEM_LOG_DIRTY_PAGES
: 0;
272 static int kvm_slot_dirty_pages_log_change(KVMSlot
*mem
, bool log_dirty
)
274 KVMState
*s
= kvm_state
;
275 int flags
, mask
= KVM_MEM_LOG_DIRTY_PAGES
;
278 old_flags
= mem
->flags
;
280 flags
= (mem
->flags
& ~mask
) | kvm_mem_flags(s
, log_dirty
);
283 /* If nothing changed effectively, no need to issue ioctl */
284 if (s
->migration_log
) {
285 flags
|= KVM_MEM_LOG_DIRTY_PAGES
;
288 if (flags
== old_flags
) {
292 return kvm_set_user_memory_region(s
, mem
);
295 static int kvm_dirty_pages_log_change(target_phys_addr_t phys_addr
,
296 ram_addr_t size
, bool log_dirty
)
298 KVMState
*s
= kvm_state
;
299 KVMSlot
*mem
= kvm_lookup_matching_slot(s
, phys_addr
, phys_addr
+ size
);
302 fprintf(stderr
, "BUG: %s: invalid parameters " TARGET_FMT_plx
"-"
303 TARGET_FMT_plx
"\n", __func__
, phys_addr
,
304 (target_phys_addr_t
)(phys_addr
+ size
- 1));
307 return kvm_slot_dirty_pages_log_change(mem
, log_dirty
);
310 static void kvm_log_start(MemoryListener
*listener
,
311 MemoryRegionSection
*section
)
315 r
= kvm_dirty_pages_log_change(section
->offset_within_address_space
,
316 section
->size
, true);
322 static void kvm_log_stop(MemoryListener
*listener
,
323 MemoryRegionSection
*section
)
327 r
= kvm_dirty_pages_log_change(section
->offset_within_address_space
,
328 section
->size
, false);
334 static int kvm_set_migration_log(int enable
)
336 KVMState
*s
= kvm_state
;
340 s
->migration_log
= enable
;
342 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
345 if (!mem
->memory_size
) {
348 if (!!(mem
->flags
& KVM_MEM_LOG_DIRTY_PAGES
) == enable
) {
351 err
= kvm_set_user_memory_region(s
, mem
);
359 /* get kvm's dirty pages bitmap and update qemu's */
360 static int kvm_get_dirty_pages_log_range(MemoryRegionSection
*section
,
361 unsigned long *bitmap
)
364 unsigned long page_number
, c
;
365 target_phys_addr_t addr
, addr1
;
366 unsigned int len
= ((section
->size
/ TARGET_PAGE_SIZE
) + HOST_LONG_BITS
- 1) / HOST_LONG_BITS
;
367 unsigned long hpratio
= getpagesize() / TARGET_PAGE_SIZE
;
370 * bitmap-traveling is faster than memory-traveling (for addr...)
371 * especially when most of the memory is not dirty.
373 for (i
= 0; i
< len
; i
++) {
374 if (bitmap
[i
] != 0) {
375 c
= leul_to_cpu(bitmap
[i
]);
379 page_number
= (i
* HOST_LONG_BITS
+ j
) * hpratio
;
380 addr1
= page_number
* TARGET_PAGE_SIZE
;
381 addr
= section
->offset_within_region
+ addr1
;
382 memory_region_set_dirty(section
->mr
, addr
,
383 TARGET_PAGE_SIZE
* hpratio
);
390 #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1))
393 * kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space
394 * This function updates qemu's dirty bitmap using
395 * memory_region_set_dirty(). This means all bits are set
398 * @start_add: start of logged region.
399 * @end_addr: end of logged region.
401 static int kvm_physical_sync_dirty_bitmap(MemoryRegionSection
*section
)
403 KVMState
*s
= kvm_state
;
404 unsigned long size
, allocated_size
= 0;
408 target_phys_addr_t start_addr
= section
->offset_within_address_space
;
409 target_phys_addr_t end_addr
= start_addr
+ section
->size
;
411 d
.dirty_bitmap
= NULL
;
412 while (start_addr
< end_addr
) {
413 mem
= kvm_lookup_overlapping_slot(s
, start_addr
, end_addr
);
418 /* XXX bad kernel interface alert
419 * For dirty bitmap, kernel allocates array of size aligned to
420 * bits-per-long. But for case when the kernel is 64bits and
421 * the userspace is 32bits, userspace can't align to the same
422 * bits-per-long, since sizeof(long) is different between kernel
423 * and user space. This way, userspace will provide buffer which
424 * may be 4 bytes less than the kernel will use, resulting in
425 * userspace memory corruption (which is not detectable by valgrind
426 * too, in most cases).
427 * So for now, let's align to 64 instead of HOST_LONG_BITS here, in
428 * a hope that sizeof(long) wont become >8 any time soon.
430 size
= ALIGN(((mem
->memory_size
) >> TARGET_PAGE_BITS
),
431 /*HOST_LONG_BITS*/ 64) / 8;
432 if (!d
.dirty_bitmap
) {
433 d
.dirty_bitmap
= g_malloc(size
);
434 } else if (size
> allocated_size
) {
435 d
.dirty_bitmap
= g_realloc(d
.dirty_bitmap
, size
);
437 allocated_size
= size
;
438 memset(d
.dirty_bitmap
, 0, allocated_size
);
442 if (kvm_vm_ioctl(s
, KVM_GET_DIRTY_LOG
, &d
) == -1) {
443 DPRINTF("ioctl failed %d\n", errno
);
448 kvm_get_dirty_pages_log_range(section
, d
.dirty_bitmap
);
449 start_addr
= mem
->start_addr
+ mem
->memory_size
;
451 g_free(d
.dirty_bitmap
);
456 int kvm_coalesce_mmio_region(target_phys_addr_t start
, ram_addr_t size
)
459 KVMState
*s
= kvm_state
;
461 if (s
->coalesced_mmio
) {
462 struct kvm_coalesced_mmio_zone zone
;
468 ret
= kvm_vm_ioctl(s
, KVM_REGISTER_COALESCED_MMIO
, &zone
);
474 int kvm_uncoalesce_mmio_region(target_phys_addr_t start
, ram_addr_t size
)
477 KVMState
*s
= kvm_state
;
479 if (s
->coalesced_mmio
) {
480 struct kvm_coalesced_mmio_zone zone
;
486 ret
= 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 target_phys_addr_t 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_begin(MemoryListener
*listener
)
709 static void kvm_commit(MemoryListener
*listener
)
713 static void kvm_region_add(MemoryListener
*listener
,
714 MemoryRegionSection
*section
)
716 kvm_set_phys_mem(section
, true);
719 static void kvm_region_del(MemoryListener
*listener
,
720 MemoryRegionSection
*section
)
722 kvm_set_phys_mem(section
, false);
725 static void kvm_region_nop(MemoryListener
*listener
,
726 MemoryRegionSection
*section
)
730 static void kvm_log_sync(MemoryListener
*listener
,
731 MemoryRegionSection
*section
)
735 r
= kvm_physical_sync_dirty_bitmap(section
);
741 static void kvm_log_global_start(struct MemoryListener
*listener
)
745 r
= kvm_set_migration_log(1);
749 static void kvm_log_global_stop(struct MemoryListener
*listener
)
753 r
= kvm_set_migration_log(0);
757 static void kvm_mem_ioeventfd_add(MemoryRegionSection
*section
,
758 bool match_data
, uint64_t data
, int fd
)
762 assert(match_data
&& section
->size
<= 8);
764 r
= kvm_set_ioeventfd_mmio(fd
, section
->offset_within_address_space
,
765 data
, true, section
->size
);
771 static void kvm_mem_ioeventfd_del(MemoryRegionSection
*section
,
772 bool match_data
, uint64_t data
, int fd
)
776 r
= kvm_set_ioeventfd_mmio(fd
, section
->offset_within_address_space
,
777 data
, false, section
->size
);
783 static void kvm_io_ioeventfd_add(MemoryRegionSection
*section
,
784 bool match_data
, uint64_t data
, int fd
)
788 assert(match_data
&& section
->size
== 2);
790 r
= kvm_set_ioeventfd_pio_word(fd
, section
->offset_within_address_space
,
797 static void kvm_io_ioeventfd_del(MemoryRegionSection
*section
,
798 bool match_data
, uint64_t data
, int fd
)
803 r
= kvm_set_ioeventfd_pio_word(fd
, section
->offset_within_address_space
,
810 static void kvm_eventfd_add(MemoryListener
*listener
,
811 MemoryRegionSection
*section
,
812 bool match_data
, uint64_t data
,
815 if (section
->address_space
== get_system_memory()) {
816 kvm_mem_ioeventfd_add(section
, match_data
, data
,
817 event_notifier_get_fd(e
));
819 kvm_io_ioeventfd_add(section
, match_data
, data
,
820 event_notifier_get_fd(e
));
824 static void kvm_eventfd_del(MemoryListener
*listener
,
825 MemoryRegionSection
*section
,
826 bool match_data
, uint64_t data
,
829 if (section
->address_space
== get_system_memory()) {
830 kvm_mem_ioeventfd_del(section
, match_data
, data
,
831 event_notifier_get_fd(e
));
833 kvm_io_ioeventfd_del(section
, match_data
, data
,
834 event_notifier_get_fd(e
));
838 static MemoryListener kvm_memory_listener
= {
840 .commit
= kvm_commit
,
841 .region_add
= kvm_region_add
,
842 .region_del
= kvm_region_del
,
843 .region_nop
= kvm_region_nop
,
844 .log_start
= kvm_log_start
,
845 .log_stop
= kvm_log_stop
,
846 .log_sync
= kvm_log_sync
,
847 .log_global_start
= kvm_log_global_start
,
848 .log_global_stop
= kvm_log_global_stop
,
849 .eventfd_add
= kvm_eventfd_add
,
850 .eventfd_del
= kvm_eventfd_del
,
854 static void kvm_handle_interrupt(CPUArchState
*env
, int mask
)
856 env
->interrupt_request
|= mask
;
858 if (!qemu_cpu_is_self(env
)) {
863 int kvm_set_irq(KVMState
*s
, int irq
, int level
)
865 struct kvm_irq_level event
;
868 assert(kvm_async_interrupts_enabled());
872 ret
= kvm_vm_ioctl(s
, s
->irqchip_inject_ioctl
, &event
);
874 perror("kvm_set_irq");
878 return (s
->irqchip_inject_ioctl
== KVM_IRQ_LINE
) ? 1 : event
.status
;
881 #ifdef KVM_CAP_IRQ_ROUTING
882 typedef struct KVMMSIRoute
{
883 struct kvm_irq_routing_entry kroute
;
884 QTAILQ_ENTRY(KVMMSIRoute
) entry
;
887 static void set_gsi(KVMState
*s
, unsigned int gsi
)
889 s
->used_gsi_bitmap
[gsi
/ 32] |= 1U << (gsi
% 32);
892 static void clear_gsi(KVMState
*s
, unsigned int gsi
)
894 s
->used_gsi_bitmap
[gsi
/ 32] &= ~(1U << (gsi
% 32));
897 static void kvm_init_irq_routing(KVMState
*s
)
901 gsi_count
= kvm_check_extension(s
, KVM_CAP_IRQ_ROUTING
);
903 unsigned int gsi_bits
, i
;
905 /* Round up so we can search ints using ffs */
906 gsi_bits
= ALIGN(gsi_count
, 32);
907 s
->used_gsi_bitmap
= g_malloc0(gsi_bits
/ 8);
908 s
->gsi_count
= gsi_count
;
910 /* Mark any over-allocated bits as already in use */
911 for (i
= gsi_count
; i
< gsi_bits
; i
++) {
916 s
->irq_routes
= g_malloc0(sizeof(*s
->irq_routes
));
917 s
->nr_allocated_irq_routes
= 0;
919 if (!s
->direct_msi
) {
920 for (i
= 0; i
< KVM_MSI_HASHTAB_SIZE
; i
++) {
921 QTAILQ_INIT(&s
->msi_hashtab
[i
]);
925 kvm_arch_init_irq_routing(s
);
928 static void kvm_irqchip_commit_routes(KVMState
*s
)
932 s
->irq_routes
->flags
= 0;
933 ret
= kvm_vm_ioctl(s
, KVM_SET_GSI_ROUTING
, s
->irq_routes
);
937 static void kvm_add_routing_entry(KVMState
*s
,
938 struct kvm_irq_routing_entry
*entry
)
940 struct kvm_irq_routing_entry
*new;
943 if (s
->irq_routes
->nr
== s
->nr_allocated_irq_routes
) {
944 n
= s
->nr_allocated_irq_routes
* 2;
948 size
= sizeof(struct kvm_irq_routing
);
949 size
+= n
* sizeof(*new);
950 s
->irq_routes
= g_realloc(s
->irq_routes
, size
);
951 s
->nr_allocated_irq_routes
= n
;
953 n
= s
->irq_routes
->nr
++;
954 new = &s
->irq_routes
->entries
[n
];
955 memset(new, 0, sizeof(*new));
956 new->gsi
= entry
->gsi
;
957 new->type
= entry
->type
;
958 new->flags
= entry
->flags
;
961 set_gsi(s
, entry
->gsi
);
963 kvm_irqchip_commit_routes(s
);
966 void kvm_irqchip_add_irq_route(KVMState
*s
, int irq
, int irqchip
, int pin
)
968 struct kvm_irq_routing_entry e
;
970 assert(pin
< s
->gsi_count
);
973 e
.type
= KVM_IRQ_ROUTING_IRQCHIP
;
975 e
.u
.irqchip
.irqchip
= irqchip
;
976 e
.u
.irqchip
.pin
= pin
;
977 kvm_add_routing_entry(s
, &e
);
980 void kvm_irqchip_release_virq(KVMState
*s
, int virq
)
982 struct kvm_irq_routing_entry
*e
;
985 for (i
= 0; i
< s
->irq_routes
->nr
; i
++) {
986 e
= &s
->irq_routes
->entries
[i
];
987 if (e
->gsi
== virq
) {
989 *e
= s
->irq_routes
->entries
[s
->irq_routes
->nr
];
994 kvm_irqchip_commit_routes(s
);
997 static unsigned int kvm_hash_msi(uint32_t data
)
999 /* This is optimized for IA32 MSI layout. However, no other arch shall
1000 * repeat the mistake of not providing a direct MSI injection API. */
1004 static void kvm_flush_dynamic_msi_routes(KVMState
*s
)
1006 KVMMSIRoute
*route
, *next
;
1009 for (hash
= 0; hash
< KVM_MSI_HASHTAB_SIZE
; hash
++) {
1010 QTAILQ_FOREACH_SAFE(route
, &s
->msi_hashtab
[hash
], entry
, next
) {
1011 kvm_irqchip_release_virq(s
, route
->kroute
.gsi
);
1012 QTAILQ_REMOVE(&s
->msi_hashtab
[hash
], route
, entry
);
1018 static int kvm_irqchip_get_virq(KVMState
*s
)
1020 uint32_t *word
= s
->used_gsi_bitmap
;
1021 int max_words
= ALIGN(s
->gsi_count
, 32) / 32;
1026 /* Return the lowest unused GSI in the bitmap */
1027 for (i
= 0; i
< max_words
; i
++) {
1028 bit
= ffs(~word
[i
]);
1033 return bit
- 1 + i
* 32;
1035 if (!s
->direct_msi
&& retry
) {
1037 kvm_flush_dynamic_msi_routes(s
);
1044 static KVMMSIRoute
*kvm_lookup_msi_route(KVMState
*s
, MSIMessage msg
)
1046 unsigned int hash
= kvm_hash_msi(msg
.data
);
1049 QTAILQ_FOREACH(route
, &s
->msi_hashtab
[hash
], entry
) {
1050 if (route
->kroute
.u
.msi
.address_lo
== (uint32_t)msg
.address
&&
1051 route
->kroute
.u
.msi
.address_hi
== (msg
.address
>> 32) &&
1052 route
->kroute
.u
.msi
.data
== msg
.data
) {
1059 int kvm_irqchip_send_msi(KVMState
*s
, MSIMessage msg
)
1064 if (s
->direct_msi
) {
1065 msi
.address_lo
= (uint32_t)msg
.address
;
1066 msi
.address_hi
= msg
.address
>> 32;
1067 msi
.data
= msg
.data
;
1069 memset(msi
.pad
, 0, sizeof(msi
.pad
));
1071 return kvm_vm_ioctl(s
, KVM_SIGNAL_MSI
, &msi
);
1074 route
= kvm_lookup_msi_route(s
, msg
);
1078 virq
= kvm_irqchip_get_virq(s
);
1083 route
= g_malloc(sizeof(KVMMSIRoute
));
1084 route
->kroute
.gsi
= virq
;
1085 route
->kroute
.type
= KVM_IRQ_ROUTING_MSI
;
1086 route
->kroute
.flags
= 0;
1087 route
->kroute
.u
.msi
.address_lo
= (uint32_t)msg
.address
;
1088 route
->kroute
.u
.msi
.address_hi
= msg
.address
>> 32;
1089 route
->kroute
.u
.msi
.data
= msg
.data
;
1091 kvm_add_routing_entry(s
, &route
->kroute
);
1093 QTAILQ_INSERT_TAIL(&s
->msi_hashtab
[kvm_hash_msi(msg
.data
)], route
,
1097 assert(route
->kroute
.type
== KVM_IRQ_ROUTING_MSI
);
1099 return kvm_set_irq(s
, route
->kroute
.gsi
, 1);
1102 int kvm_irqchip_add_msi_route(KVMState
*s
, MSIMessage msg
)
1104 struct kvm_irq_routing_entry kroute
;
1107 if (!kvm_gsi_routing_enabled()) {
1111 virq
= kvm_irqchip_get_virq(s
);
1117 kroute
.type
= KVM_IRQ_ROUTING_MSI
;
1119 kroute
.u
.msi
.address_lo
= (uint32_t)msg
.address
;
1120 kroute
.u
.msi
.address_hi
= msg
.address
>> 32;
1121 kroute
.u
.msi
.data
= msg
.data
;
1123 kvm_add_routing_entry(s
, &kroute
);
1128 static int kvm_irqchip_assign_irqfd(KVMState
*s
, int fd
, int virq
, bool assign
)
1130 struct kvm_irqfd irqfd
= {
1133 .flags
= assign
? 0 : KVM_IRQFD_FLAG_DEASSIGN
,
1136 if (!kvm_irqfds_enabled()) {
1140 return kvm_vm_ioctl(s
, KVM_IRQFD
, &irqfd
);
1143 #else /* !KVM_CAP_IRQ_ROUTING */
1145 static void kvm_init_irq_routing(KVMState
*s
)
1149 void kvm_irqchip_release_virq(KVMState
*s
, int virq
)
1153 int kvm_irqchip_send_msi(KVMState
*s
, MSIMessage msg
)
1158 int kvm_irqchip_add_msi_route(KVMState
*s
, MSIMessage msg
)
1163 static int kvm_irqchip_assign_irqfd(KVMState
*s
, int fd
, int virq
, bool assign
)
1167 #endif /* !KVM_CAP_IRQ_ROUTING */
1169 int kvm_irqchip_add_irqfd(KVMState
*s
, int fd
, int virq
)
1171 return kvm_irqchip_assign_irqfd(s
, fd
, virq
, true);
1174 int kvm_irqchip_add_irq_notifier(KVMState
*s
, EventNotifier
*n
, int virq
)
1176 return kvm_irqchip_add_irqfd(s
, event_notifier_get_fd(n
), virq
);
1179 int kvm_irqchip_remove_irqfd(KVMState
*s
, int fd
, int virq
)
1181 return kvm_irqchip_assign_irqfd(s
, fd
, virq
, false);
1184 int kvm_irqchip_remove_irq_notifier(KVMState
*s
, EventNotifier
*n
, int virq
)
1186 return kvm_irqchip_remove_irqfd(s
, event_notifier_get_fd(n
), virq
);
1189 static int kvm_irqchip_create(KVMState
*s
)
1191 QemuOptsList
*list
= qemu_find_opts("machine");
1194 if (QTAILQ_EMPTY(&list
->head
) ||
1195 !qemu_opt_get_bool(QTAILQ_FIRST(&list
->head
),
1196 "kernel_irqchip", true) ||
1197 !kvm_check_extension(s
, KVM_CAP_IRQCHIP
)) {
1201 ret
= kvm_vm_ioctl(s
, KVM_CREATE_IRQCHIP
);
1203 fprintf(stderr
, "Create kernel irqchip failed\n");
1207 s
->irqchip_inject_ioctl
= KVM_IRQ_LINE
;
1208 if (kvm_check_extension(s
, KVM_CAP_IRQ_INJECT_STATUS
)) {
1209 s
->irqchip_inject_ioctl
= KVM_IRQ_LINE_STATUS
;
1211 kvm_kernel_irqchip
= true;
1212 /* If we have an in-kernel IRQ chip then we must have asynchronous
1213 * interrupt delivery (though the reverse is not necessarily true)
1215 kvm_async_interrupts_allowed
= true;
1217 kvm_init_irq_routing(s
);
1222 static int kvm_max_vcpus(KVMState
*s
)
1226 /* Find number of supported CPUs using the recommended
1227 * procedure from the kernel API documentation to cope with
1228 * older kernels that may be missing capabilities.
1230 ret
= kvm_check_extension(s
, KVM_CAP_MAX_VCPUS
);
1234 ret
= kvm_check_extension(s
, KVM_CAP_NR_VCPUS
);
1244 static const char upgrade_note
[] =
1245 "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n"
1246 "(see http://sourceforge.net/projects/kvm).\n";
1248 const KVMCapabilityInfo
*missing_cap
;
1253 s
= g_malloc0(sizeof(KVMState
));
1256 * On systems where the kernel can support different base page
1257 * sizes, host page size may be different from TARGET_PAGE_SIZE,
1258 * even with KVM. TARGET_PAGE_SIZE is assumed to be the minimum
1259 * page size for the system though.
1261 assert(TARGET_PAGE_SIZE
<= getpagesize());
1263 #ifdef KVM_CAP_SET_GUEST_DEBUG
1264 QTAILQ_INIT(&s
->kvm_sw_breakpoints
);
1266 for (i
= 0; i
< ARRAY_SIZE(s
->slots
); i
++) {
1267 s
->slots
[i
].slot
= i
;
1270 s
->fd
= qemu_open("/dev/kvm", O_RDWR
);
1272 fprintf(stderr
, "Could not access KVM kernel module: %m\n");
1277 ret
= kvm_ioctl(s
, KVM_GET_API_VERSION
, 0);
1278 if (ret
< KVM_API_VERSION
) {
1282 fprintf(stderr
, "kvm version too old\n");
1286 if (ret
> KVM_API_VERSION
) {
1288 fprintf(stderr
, "kvm version not supported\n");
1292 max_vcpus
= kvm_max_vcpus(s
);
1293 if (smp_cpus
> max_vcpus
) {
1295 fprintf(stderr
, "Number of SMP cpus requested (%d) exceeds max cpus "
1296 "supported by KVM (%d)\n", smp_cpus
, max_vcpus
);
1300 s
->vmfd
= kvm_ioctl(s
, KVM_CREATE_VM
, 0);
1303 fprintf(stderr
, "Please add the 'switch_amode' kernel parameter to "
1304 "your host kernel command line\n");
1310 missing_cap
= kvm_check_extension_list(s
, kvm_required_capabilites
);
1313 kvm_check_extension_list(s
, kvm_arch_required_capabilities
);
1317 fprintf(stderr
, "kvm does not support %s\n%s",
1318 missing_cap
->name
, upgrade_note
);
1322 s
->coalesced_mmio
= kvm_check_extension(s
, KVM_CAP_COALESCED_MMIO
);
1324 s
->broken_set_mem_region
= 1;
1325 ret
= kvm_check_extension(s
, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
);
1327 s
->broken_set_mem_region
= 0;
1330 #ifdef KVM_CAP_VCPU_EVENTS
1331 s
->vcpu_events
= kvm_check_extension(s
, KVM_CAP_VCPU_EVENTS
);
1334 s
->robust_singlestep
=
1335 kvm_check_extension(s
, KVM_CAP_X86_ROBUST_SINGLESTEP
);
1337 #ifdef KVM_CAP_DEBUGREGS
1338 s
->debugregs
= kvm_check_extension(s
, KVM_CAP_DEBUGREGS
);
1341 #ifdef KVM_CAP_XSAVE
1342 s
->xsave
= kvm_check_extension(s
, KVM_CAP_XSAVE
);
1346 s
->xcrs
= kvm_check_extension(s
, KVM_CAP_XCRS
);
1349 #ifdef KVM_CAP_PIT_STATE2
1350 s
->pit_state2
= kvm_check_extension(s
, KVM_CAP_PIT_STATE2
);
1353 #ifdef KVM_CAP_IRQ_ROUTING
1354 s
->direct_msi
= (kvm_check_extension(s
, KVM_CAP_SIGNAL_MSI
) > 0);
1357 ret
= kvm_arch_init(s
);
1362 ret
= kvm_irqchip_create(s
);
1368 memory_listener_register(&kvm_memory_listener
, NULL
);
1370 s
->many_ioeventfds
= kvm_check_many_ioeventfds();
1372 cpu_interrupt_handler
= kvm_handle_interrupt
;
1388 static void kvm_handle_io(uint16_t port
, void *data
, int direction
, int size
,
1392 uint8_t *ptr
= data
;
1394 for (i
= 0; i
< count
; i
++) {
1395 if (direction
== KVM_EXIT_IO_IN
) {
1398 stb_p(ptr
, cpu_inb(port
));
1401 stw_p(ptr
, cpu_inw(port
));
1404 stl_p(ptr
, cpu_inl(port
));
1410 cpu_outb(port
, ldub_p(ptr
));
1413 cpu_outw(port
, lduw_p(ptr
));
1416 cpu_outl(port
, ldl_p(ptr
));
1425 static int kvm_handle_internal_error(CPUArchState
*env
, struct kvm_run
*run
)
1427 fprintf(stderr
, "KVM internal error.");
1428 if (kvm_check_extension(kvm_state
, KVM_CAP_INTERNAL_ERROR_DATA
)) {
1431 fprintf(stderr
, " Suberror: %d\n", run
->internal
.suberror
);
1432 for (i
= 0; i
< run
->internal
.ndata
; ++i
) {
1433 fprintf(stderr
, "extra data[%d]: %"PRIx64
"\n",
1434 i
, (uint64_t)run
->internal
.data
[i
]);
1437 fprintf(stderr
, "\n");
1439 if (run
->internal
.suberror
== KVM_INTERNAL_ERROR_EMULATION
) {
1440 fprintf(stderr
, "emulation failure\n");
1441 if (!kvm_arch_stop_on_emulation_error(env
)) {
1442 cpu_dump_state(env
, stderr
, fprintf
, CPU_DUMP_CODE
);
1443 return EXCP_INTERRUPT
;
1446 /* FIXME: Should trigger a qmp message to let management know
1447 * something went wrong.
1452 void kvm_flush_coalesced_mmio_buffer(void)
1454 KVMState
*s
= kvm_state
;
1456 if (s
->coalesced_flush_in_progress
) {
1460 s
->coalesced_flush_in_progress
= true;
1462 if (s
->coalesced_mmio_ring
) {
1463 struct kvm_coalesced_mmio_ring
*ring
= s
->coalesced_mmio_ring
;
1464 while (ring
->first
!= ring
->last
) {
1465 struct kvm_coalesced_mmio
*ent
;
1467 ent
= &ring
->coalesced_mmio
[ring
->first
];
1469 cpu_physical_memory_write(ent
->phys_addr
, ent
->data
, ent
->len
);
1471 ring
->first
= (ring
->first
+ 1) % KVM_COALESCED_MMIO_MAX
;
1475 s
->coalesced_flush_in_progress
= false;
1478 static void do_kvm_cpu_synchronize_state(void *_env
)
1480 CPUArchState
*env
= _env
;
1482 if (!env
->kvm_vcpu_dirty
) {
1483 kvm_arch_get_registers(env
);
1484 env
->kvm_vcpu_dirty
= 1;
1488 void kvm_cpu_synchronize_state(CPUArchState
*env
)
1490 if (!env
->kvm_vcpu_dirty
) {
1491 run_on_cpu(env
, do_kvm_cpu_synchronize_state
, env
);
1495 void kvm_cpu_synchronize_post_reset(CPUArchState
*env
)
1497 kvm_arch_put_registers(env
, KVM_PUT_RESET_STATE
);
1498 env
->kvm_vcpu_dirty
= 0;
1501 void kvm_cpu_synchronize_post_init(CPUArchState
*env
)
1503 kvm_arch_put_registers(env
, KVM_PUT_FULL_STATE
);
1504 env
->kvm_vcpu_dirty
= 0;
1507 int kvm_cpu_exec(CPUArchState
*env
)
1509 struct kvm_run
*run
= env
->kvm_run
;
1512 DPRINTF("kvm_cpu_exec()\n");
1514 if (kvm_arch_process_async_events(env
)) {
1515 env
->exit_request
= 0;
1520 if (env
->kvm_vcpu_dirty
) {
1521 kvm_arch_put_registers(env
, KVM_PUT_RUNTIME_STATE
);
1522 env
->kvm_vcpu_dirty
= 0;
1525 kvm_arch_pre_run(env
, run
);
1526 if (env
->exit_request
) {
1527 DPRINTF("interrupt exit requested\n");
1529 * KVM requires us to reenter the kernel after IO exits to complete
1530 * instruction emulation. This self-signal will ensure that we
1533 qemu_cpu_kick_self();
1535 qemu_mutex_unlock_iothread();
1537 run_ret
= kvm_vcpu_ioctl(env
, KVM_RUN
, 0);
1539 qemu_mutex_lock_iothread();
1540 kvm_arch_post_run(env
, run
);
1542 kvm_flush_coalesced_mmio_buffer();
1545 if (run_ret
== -EINTR
|| run_ret
== -EAGAIN
) {
1546 DPRINTF("io window exit\n");
1547 ret
= EXCP_INTERRUPT
;
1550 fprintf(stderr
, "error: kvm run failed %s\n",
1551 strerror(-run_ret
));
1555 switch (run
->exit_reason
) {
1557 DPRINTF("handle_io\n");
1558 kvm_handle_io(run
->io
.port
,
1559 (uint8_t *)run
+ run
->io
.data_offset
,
1566 DPRINTF("handle_mmio\n");
1567 cpu_physical_memory_rw(run
->mmio
.phys_addr
,
1570 run
->mmio
.is_write
);
1573 case KVM_EXIT_IRQ_WINDOW_OPEN
:
1574 DPRINTF("irq_window_open\n");
1575 ret
= EXCP_INTERRUPT
;
1577 case KVM_EXIT_SHUTDOWN
:
1578 DPRINTF("shutdown\n");
1579 qemu_system_reset_request();
1580 ret
= EXCP_INTERRUPT
;
1582 case KVM_EXIT_UNKNOWN
:
1583 fprintf(stderr
, "KVM: unknown exit, hardware reason %" PRIx64
"\n",
1584 (uint64_t)run
->hw
.hardware_exit_reason
);
1587 case KVM_EXIT_INTERNAL_ERROR
:
1588 ret
= kvm_handle_internal_error(env
, run
);
1591 DPRINTF("kvm_arch_handle_exit\n");
1592 ret
= kvm_arch_handle_exit(env
, run
);
1598 cpu_dump_state(env
, stderr
, fprintf
, CPU_DUMP_CODE
);
1599 vm_stop(RUN_STATE_INTERNAL_ERROR
);
1602 env
->exit_request
= 0;
1606 int kvm_ioctl(KVMState
*s
, int type
, ...)
1613 arg
= va_arg(ap
, void *);
1616 ret
= ioctl(s
->fd
, type
, arg
);
1623 int kvm_vm_ioctl(KVMState
*s
, int type
, ...)
1630 arg
= va_arg(ap
, void *);
1633 ret
= ioctl(s
->vmfd
, type
, arg
);
1640 int kvm_vcpu_ioctl(CPUArchState
*env
, int type
, ...)
1647 arg
= va_arg(ap
, void *);
1650 ret
= ioctl(env
->kvm_fd
, type
, arg
);
1657 int kvm_has_sync_mmu(void)
1659 return kvm_check_extension(kvm_state
, KVM_CAP_SYNC_MMU
);
1662 int kvm_has_vcpu_events(void)
1664 return kvm_state
->vcpu_events
;
1667 int kvm_has_robust_singlestep(void)
1669 return kvm_state
->robust_singlestep
;
1672 int kvm_has_debugregs(void)
1674 return kvm_state
->debugregs
;
1677 int kvm_has_xsave(void)
1679 return kvm_state
->xsave
;
1682 int kvm_has_xcrs(void)
1684 return kvm_state
->xcrs
;
1687 int kvm_has_pit_state2(void)
1689 return kvm_state
->pit_state2
;
1692 int kvm_has_many_ioeventfds(void)
1694 if (!kvm_enabled()) {
1697 return kvm_state
->many_ioeventfds
;
1700 int kvm_has_gsi_routing(void)
1702 #ifdef KVM_CAP_IRQ_ROUTING
1703 return kvm_check_extension(kvm_state
, KVM_CAP_IRQ_ROUTING
);
1709 void *kvm_vmalloc(ram_addr_t size
)
1714 mem
= kvm_arch_vmalloc(size
);
1719 return qemu_vmalloc(size
);
1722 void kvm_setup_guest_memory(void *start
, size_t size
)
1724 #ifdef CONFIG_VALGRIND_H
1725 VALGRIND_MAKE_MEM_DEFINED(start
, size
);
1727 if (!kvm_has_sync_mmu()) {
1728 int ret
= qemu_madvise(start
, size
, QEMU_MADV_DONTFORK
);
1731 perror("qemu_madvise");
1733 "Need MADV_DONTFORK in absence of synchronous KVM MMU\n");
1739 #ifdef KVM_CAP_SET_GUEST_DEBUG
1740 struct kvm_sw_breakpoint
*kvm_find_sw_breakpoint(CPUArchState
*env
,
1743 struct kvm_sw_breakpoint
*bp
;
1745 QTAILQ_FOREACH(bp
, &env
->kvm_state
->kvm_sw_breakpoints
, entry
) {
1753 int kvm_sw_breakpoints_active(CPUArchState
*env
)
1755 return !QTAILQ_EMPTY(&env
->kvm_state
->kvm_sw_breakpoints
);
1758 struct kvm_set_guest_debug_data
{
1759 struct kvm_guest_debug dbg
;
1764 static void kvm_invoke_set_guest_debug(void *data
)
1766 struct kvm_set_guest_debug_data
*dbg_data
= data
;
1767 CPUArchState
*env
= dbg_data
->env
;
1769 dbg_data
->err
= kvm_vcpu_ioctl(env
, KVM_SET_GUEST_DEBUG
, &dbg_data
->dbg
);
1772 int kvm_update_guest_debug(CPUArchState
*env
, unsigned long reinject_trap
)
1774 struct kvm_set_guest_debug_data data
;
1776 data
.dbg
.control
= reinject_trap
;
1778 if (env
->singlestep_enabled
) {
1779 data
.dbg
.control
|= KVM_GUESTDBG_ENABLE
| KVM_GUESTDBG_SINGLESTEP
;
1781 kvm_arch_update_guest_debug(env
, &data
.dbg
);
1784 run_on_cpu(env
, kvm_invoke_set_guest_debug
, &data
);
1788 int kvm_insert_breakpoint(CPUArchState
*current_env
, target_ulong addr
,
1789 target_ulong len
, int type
)
1791 struct kvm_sw_breakpoint
*bp
;
1795 if (type
== GDB_BREAKPOINT_SW
) {
1796 bp
= kvm_find_sw_breakpoint(current_env
, addr
);
1802 bp
= g_malloc(sizeof(struct kvm_sw_breakpoint
));
1809 err
= kvm_arch_insert_sw_breakpoint(current_env
, bp
);
1815 QTAILQ_INSERT_HEAD(¤t_env
->kvm_state
->kvm_sw_breakpoints
,
1818 err
= kvm_arch_insert_hw_breakpoint(addr
, len
, type
);
1824 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
1825 err
= kvm_update_guest_debug(env
, 0);
1833 int kvm_remove_breakpoint(CPUArchState
*current_env
, target_ulong addr
,
1834 target_ulong len
, int type
)
1836 struct kvm_sw_breakpoint
*bp
;
1840 if (type
== GDB_BREAKPOINT_SW
) {
1841 bp
= kvm_find_sw_breakpoint(current_env
, addr
);
1846 if (bp
->use_count
> 1) {
1851 err
= kvm_arch_remove_sw_breakpoint(current_env
, bp
);
1856 QTAILQ_REMOVE(¤t_env
->kvm_state
->kvm_sw_breakpoints
, bp
, entry
);
1859 err
= kvm_arch_remove_hw_breakpoint(addr
, len
, type
);
1865 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
1866 err
= kvm_update_guest_debug(env
, 0);
1874 void kvm_remove_all_breakpoints(CPUArchState
*current_env
)
1876 struct kvm_sw_breakpoint
*bp
, *next
;
1877 KVMState
*s
= current_env
->kvm_state
;
1880 QTAILQ_FOREACH_SAFE(bp
, &s
->kvm_sw_breakpoints
, entry
, next
) {
1881 if (kvm_arch_remove_sw_breakpoint(current_env
, bp
) != 0) {
1882 /* Try harder to find a CPU that currently sees the breakpoint. */
1883 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
1884 if (kvm_arch_remove_sw_breakpoint(env
, bp
) == 0) {
1890 kvm_arch_remove_all_hw_breakpoints();
1892 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
1893 kvm_update_guest_debug(env
, 0);
1897 #else /* !KVM_CAP_SET_GUEST_DEBUG */
1899 int kvm_update_guest_debug(CPUArchState
*env
, unsigned long reinject_trap
)
1904 int kvm_insert_breakpoint(CPUArchState
*current_env
, target_ulong addr
,
1905 target_ulong len
, int type
)
1910 int kvm_remove_breakpoint(CPUArchState
*current_env
, target_ulong addr
,
1911 target_ulong len
, int type
)
1916 void kvm_remove_all_breakpoints(CPUArchState
*current_env
)
1919 #endif /* !KVM_CAP_SET_GUEST_DEBUG */
1921 int kvm_set_signal_mask(CPUArchState
*env
, const sigset_t
*sigset
)
1923 struct kvm_signal_mask
*sigmask
;
1927 return kvm_vcpu_ioctl(env
, KVM_SET_SIGNAL_MASK
, NULL
);
1930 sigmask
= g_malloc(sizeof(*sigmask
) + sizeof(*sigset
));
1933 memcpy(sigmask
->sigset
, sigset
, sizeof(*sigset
));
1934 r
= kvm_vcpu_ioctl(env
, KVM_SET_SIGNAL_MASK
, sigmask
);
1940 int kvm_set_ioeventfd_mmio(int fd
, uint32_t addr
, uint32_t val
, bool assign
,
1944 struct kvm_ioeventfd iofd
;
1946 iofd
.datamatch
= val
;
1949 iofd
.flags
= KVM_IOEVENTFD_FLAG_DATAMATCH
;
1952 if (!kvm_enabled()) {
1957 iofd
.flags
|= KVM_IOEVENTFD_FLAG_DEASSIGN
;
1960 ret
= kvm_vm_ioctl(kvm_state
, KVM_IOEVENTFD
, &iofd
);
1969 int kvm_set_ioeventfd_pio_word(int fd
, uint16_t addr
, uint16_t val
, bool assign
)
1971 struct kvm_ioeventfd kick
= {
1975 .flags
= KVM_IOEVENTFD_FLAG_DATAMATCH
| KVM_IOEVENTFD_FLAG_PIO
,
1979 if (!kvm_enabled()) {
1983 kick
.flags
|= KVM_IOEVENTFD_FLAG_DEASSIGN
;
1985 r
= kvm_vm_ioctl(kvm_state
, KVM_IOEVENTFD
, &kick
);
1992 int kvm_on_sigbus_vcpu(CPUArchState
*env
, int code
, void *addr
)
1994 return kvm_arch_on_sigbus_vcpu(env
, code
, addr
);
1997 int kvm_on_sigbus(int code
, void *addr
)
1999 return kvm_arch_on_sigbus(code
, addr
);