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 "qemu/osdep.h"
17 #include <sys/ioctl.h>
19 #include <linux/kvm.h>
21 #include "qemu/atomic.h"
22 #include "qemu/option.h"
23 #include "qemu/config-file.h"
24 #include "qemu/error-report.h"
25 #include "qapi/error.h"
27 #include "hw/pci/msi.h"
28 #include "hw/pci/msix.h"
29 #include "hw/s390x/adapter.h"
30 #include "exec/gdbstub.h"
31 #include "sysemu/kvm_int.h"
32 #include "sysemu/cpus.h"
33 #include "qemu/bswap.h"
34 #include "exec/memory.h"
35 #include "exec/ram_addr.h"
36 #include "exec/address-spaces.h"
37 #include "qemu/event_notifier.h"
40 #include "sysemu/sev.h"
41 #include "sysemu/balloon.h"
43 #include "hw/boards.h"
45 /* This check must be after config-host.h is included */
47 #include <sys/eventfd.h>
50 /* KVM uses PAGE_SIZE in its definition of KVM_COALESCED_MMIO_MAX. We
51 * need to use the real host PAGE_SIZE, as that's what KVM will use.
53 #define PAGE_SIZE getpagesize()
58 #define DPRINTF(fmt, ...) \
59 do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
61 #define DPRINTF(fmt, ...) \
65 #define KVM_MSI_HASHTAB_SIZE 256
67 struct KVMParkedVcpu
{
68 unsigned long vcpu_id
;
70 QLIST_ENTRY(KVMParkedVcpu
) node
;
75 AccelState parent_obj
;
82 struct kvm_coalesced_mmio_ring
*coalesced_mmio_ring
;
83 bool coalesced_flush_in_progress
;
85 int robust_singlestep
;
87 #ifdef KVM_CAP_SET_GUEST_DEBUG
88 QTAILQ_HEAD(, kvm_sw_breakpoint
) kvm_sw_breakpoints
;
90 int max_nested_state_len
;
94 bool manual_dirty_log_protect
;
95 /* The man page (and posix) say ioctl numbers are signed int, but
96 * they're not. Linux, glibc and *BSD all treat ioctl numbers as
97 * unsigned, and treating them as signed here can break things */
98 unsigned irq_set_ioctl
;
99 unsigned int sigmask_len
;
101 #ifdef KVM_CAP_IRQ_ROUTING
102 struct kvm_irq_routing
*irq_routes
;
103 int nr_allocated_irq_routes
;
104 unsigned long *used_gsi_bitmap
;
105 unsigned int gsi_count
;
106 QTAILQ_HEAD(, KVMMSIRoute
) msi_hashtab
[KVM_MSI_HASHTAB_SIZE
];
108 KVMMemoryListener memory_listener
;
109 QLIST_HEAD(, KVMParkedVcpu
) kvm_parked_vcpus
;
111 /* memory encryption */
112 void *memcrypt_handle
;
113 int (*memcrypt_encrypt_data
)(void *handle
, uint8_t *ptr
, uint64_t len
);
115 /* For "info mtree -f" to tell if an MR is registered in KVM */
118 KVMMemoryListener
*ml
;
124 bool kvm_kernel_irqchip
;
125 bool kvm_split_irqchip
;
126 bool kvm_async_interrupts_allowed
;
127 bool kvm_halt_in_kernel_allowed
;
128 bool kvm_eventfds_allowed
;
129 bool kvm_irqfds_allowed
;
130 bool kvm_resamplefds_allowed
;
131 bool kvm_msi_via_irqfd_allowed
;
132 bool kvm_gsi_routing_allowed
;
133 bool kvm_gsi_direct_mapping
;
135 bool kvm_readonly_mem_allowed
;
136 bool kvm_vm_attributes_allowed
;
137 bool kvm_direct_msi_allowed
;
138 bool kvm_ioeventfd_any_length_allowed
;
139 bool kvm_msi_use_devid
;
140 static bool kvm_immediate_exit
;
142 static const KVMCapabilityInfo kvm_required_capabilites
[] = {
143 KVM_CAP_INFO(USER_MEMORY
),
144 KVM_CAP_INFO(DESTROY_MEMORY_REGION_WORKS
),
145 KVM_CAP_INFO(JOIN_MEMORY_REGIONS_WORKS
),
149 #define kvm_slots_lock(kml) qemu_mutex_lock(&(kml)->slots_lock)
150 #define kvm_slots_unlock(kml) qemu_mutex_unlock(&(kml)->slots_lock)
152 int kvm_get_max_memslots(void)
154 KVMState
*s
= KVM_STATE(current_machine
->accelerator
);
159 bool kvm_memcrypt_enabled(void)
161 if (kvm_state
&& kvm_state
->memcrypt_handle
) {
168 int kvm_memcrypt_encrypt_data(uint8_t *ptr
, uint64_t len
)
170 if (kvm_state
->memcrypt_handle
&&
171 kvm_state
->memcrypt_encrypt_data
) {
172 return kvm_state
->memcrypt_encrypt_data(kvm_state
->memcrypt_handle
,
179 /* Called with KVMMemoryListener.slots_lock held */
180 static KVMSlot
*kvm_get_free_slot(KVMMemoryListener
*kml
)
182 KVMState
*s
= kvm_state
;
185 for (i
= 0; i
< s
->nr_slots
; i
++) {
186 if (kml
->slots
[i
].memory_size
== 0) {
187 return &kml
->slots
[i
];
194 bool kvm_has_free_slot(MachineState
*ms
)
196 KVMState
*s
= KVM_STATE(ms
->accelerator
);
198 KVMMemoryListener
*kml
= &s
->memory_listener
;
201 result
= !!kvm_get_free_slot(kml
);
202 kvm_slots_unlock(kml
);
207 /* Called with KVMMemoryListener.slots_lock held */
208 static KVMSlot
*kvm_alloc_slot(KVMMemoryListener
*kml
)
210 KVMSlot
*slot
= kvm_get_free_slot(kml
);
216 fprintf(stderr
, "%s: no free slot available\n", __func__
);
220 static KVMSlot
*kvm_lookup_matching_slot(KVMMemoryListener
*kml
,
224 KVMState
*s
= kvm_state
;
227 for (i
= 0; i
< s
->nr_slots
; i
++) {
228 KVMSlot
*mem
= &kml
->slots
[i
];
230 if (start_addr
== mem
->start_addr
&& size
== mem
->memory_size
) {
239 * Calculate and align the start address and the size of the section.
240 * Return the size. If the size is 0, the aligned section is empty.
242 static hwaddr
kvm_align_section(MemoryRegionSection
*section
,
245 hwaddr size
= int128_get64(section
->size
);
246 hwaddr delta
, aligned
;
248 /* kvm works in page size chunks, but the function may be called
249 with sub-page size and unaligned start address. Pad the start
250 address to next and truncate size to previous page boundary. */
251 aligned
= ROUND_UP(section
->offset_within_address_space
,
252 qemu_real_host_page_size
);
253 delta
= aligned
- section
->offset_within_address_space
;
259 return (size
- delta
) & qemu_real_host_page_mask
;
262 int kvm_physical_memory_addr_from_host(KVMState
*s
, void *ram
,
265 KVMMemoryListener
*kml
= &s
->memory_listener
;
269 for (i
= 0; i
< s
->nr_slots
; i
++) {
270 KVMSlot
*mem
= &kml
->slots
[i
];
272 if (ram
>= mem
->ram
&& ram
< mem
->ram
+ mem
->memory_size
) {
273 *phys_addr
= mem
->start_addr
+ (ram
- mem
->ram
);
278 kvm_slots_unlock(kml
);
283 static int kvm_set_user_memory_region(KVMMemoryListener
*kml
, KVMSlot
*slot
, bool new)
285 KVMState
*s
= kvm_state
;
286 struct kvm_userspace_memory_region mem
;
289 mem
.slot
= slot
->slot
| (kml
->as_id
<< 16);
290 mem
.guest_phys_addr
= slot
->start_addr
;
291 mem
.userspace_addr
= (unsigned long)slot
->ram
;
292 mem
.flags
= slot
->flags
;
294 if (slot
->memory_size
&& !new && (mem
.flags
^ slot
->old_flags
) & KVM_MEM_READONLY
) {
295 /* Set the slot size to 0 before setting the slot to the desired
296 * value. This is needed based on KVM commit 75d61fbc. */
298 kvm_vm_ioctl(s
, KVM_SET_USER_MEMORY_REGION
, &mem
);
300 mem
.memory_size
= slot
->memory_size
;
301 ret
= kvm_vm_ioctl(s
, KVM_SET_USER_MEMORY_REGION
, &mem
);
302 slot
->old_flags
= mem
.flags
;
303 trace_kvm_set_user_memory(mem
.slot
, mem
.flags
, mem
.guest_phys_addr
,
304 mem
.memory_size
, mem
.userspace_addr
, ret
);
308 int kvm_destroy_vcpu(CPUState
*cpu
)
310 KVMState
*s
= kvm_state
;
312 struct KVMParkedVcpu
*vcpu
= NULL
;
315 DPRINTF("kvm_destroy_vcpu\n");
317 ret
= kvm_arch_destroy_vcpu(cpu
);
322 mmap_size
= kvm_ioctl(s
, KVM_GET_VCPU_MMAP_SIZE
, 0);
325 DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n");
329 ret
= munmap(cpu
->kvm_run
, mmap_size
);
334 vcpu
= g_malloc0(sizeof(*vcpu
));
335 vcpu
->vcpu_id
= kvm_arch_vcpu_id(cpu
);
336 vcpu
->kvm_fd
= cpu
->kvm_fd
;
337 QLIST_INSERT_HEAD(&kvm_state
->kvm_parked_vcpus
, vcpu
, node
);
342 static int kvm_get_vcpu(KVMState
*s
, unsigned long vcpu_id
)
344 struct KVMParkedVcpu
*cpu
;
346 QLIST_FOREACH(cpu
, &s
->kvm_parked_vcpus
, node
) {
347 if (cpu
->vcpu_id
== vcpu_id
) {
350 QLIST_REMOVE(cpu
, node
);
351 kvm_fd
= cpu
->kvm_fd
;
357 return kvm_vm_ioctl(s
, KVM_CREATE_VCPU
, (void *)vcpu_id
);
360 int kvm_init_vcpu(CPUState
*cpu
)
362 KVMState
*s
= kvm_state
;
366 DPRINTF("kvm_init_vcpu\n");
368 ret
= kvm_get_vcpu(s
, kvm_arch_vcpu_id(cpu
));
370 DPRINTF("kvm_create_vcpu failed\n");
376 cpu
->vcpu_dirty
= true;
378 mmap_size
= kvm_ioctl(s
, KVM_GET_VCPU_MMAP_SIZE
, 0);
381 DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n");
385 cpu
->kvm_run
= mmap(NULL
, mmap_size
, PROT_READ
| PROT_WRITE
, MAP_SHARED
,
387 if (cpu
->kvm_run
== MAP_FAILED
) {
389 DPRINTF("mmap'ing vcpu state failed\n");
393 if (s
->coalesced_mmio
&& !s
->coalesced_mmio_ring
) {
394 s
->coalesced_mmio_ring
=
395 (void *)cpu
->kvm_run
+ s
->coalesced_mmio
* PAGE_SIZE
;
398 ret
= kvm_arch_init_vcpu(cpu
);
404 * dirty pages logging control
407 static int kvm_mem_flags(MemoryRegion
*mr
)
409 bool readonly
= mr
->readonly
|| memory_region_is_romd(mr
);
412 if (memory_region_get_dirty_log_mask(mr
) != 0) {
413 flags
|= KVM_MEM_LOG_DIRTY_PAGES
;
415 if (readonly
&& kvm_readonly_mem_allowed
) {
416 flags
|= KVM_MEM_READONLY
;
421 /* Called with KVMMemoryListener.slots_lock held */
422 static int kvm_slot_update_flags(KVMMemoryListener
*kml
, KVMSlot
*mem
,
425 mem
->flags
= kvm_mem_flags(mr
);
427 /* If nothing changed effectively, no need to issue ioctl */
428 if (mem
->flags
== mem
->old_flags
) {
432 return kvm_set_user_memory_region(kml
, mem
, false);
435 static int kvm_section_update_flags(KVMMemoryListener
*kml
,
436 MemoryRegionSection
*section
)
438 hwaddr start_addr
, size
;
442 size
= kvm_align_section(section
, &start_addr
);
449 mem
= kvm_lookup_matching_slot(kml
, start_addr
, size
);
451 /* We don't have a slot if we want to trap every access. */
455 ret
= kvm_slot_update_flags(kml
, mem
, section
->mr
);
458 kvm_slots_unlock(kml
);
462 static void kvm_log_start(MemoryListener
*listener
,
463 MemoryRegionSection
*section
,
466 KVMMemoryListener
*kml
= container_of(listener
, KVMMemoryListener
, listener
);
473 r
= kvm_section_update_flags(kml
, section
);
479 static void kvm_log_stop(MemoryListener
*listener
,
480 MemoryRegionSection
*section
,
483 KVMMemoryListener
*kml
= container_of(listener
, KVMMemoryListener
, listener
);
490 r
= kvm_section_update_flags(kml
, section
);
496 /* get kvm's dirty pages bitmap and update qemu's */
497 static int kvm_get_dirty_pages_log_range(MemoryRegionSection
*section
,
498 unsigned long *bitmap
)
500 ram_addr_t start
= section
->offset_within_region
+
501 memory_region_get_ram_addr(section
->mr
);
502 ram_addr_t pages
= int128_get64(section
->size
) / getpagesize();
504 cpu_physical_memory_set_dirty_lebitmap(bitmap
, start
, pages
);
508 #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1))
511 * kvm_physical_sync_dirty_bitmap - Sync dirty bitmap from kernel space
513 * This function will first try to fetch dirty bitmap from the kernel,
514 * and then updates qemu's dirty bitmap.
516 * NOTE: caller must be with kml->slots_lock held.
518 * @kml: the KVM memory listener object
519 * @section: the memory section to sync the dirty bitmap with
521 static int kvm_physical_sync_dirty_bitmap(KVMMemoryListener
*kml
,
522 MemoryRegionSection
*section
)
524 KVMState
*s
= kvm_state
;
525 struct kvm_dirty_log d
= {};
527 hwaddr start_addr
, size
;
530 size
= kvm_align_section(section
, &start_addr
);
532 mem
= kvm_lookup_matching_slot(kml
, start_addr
, size
);
534 /* We don't have a slot if we want to trap every access. */
538 /* XXX bad kernel interface alert
539 * For dirty bitmap, kernel allocates array of size aligned to
540 * bits-per-long. But for case when the kernel is 64bits and
541 * the userspace is 32bits, userspace can't align to the same
542 * bits-per-long, since sizeof(long) is different between kernel
543 * and user space. This way, userspace will provide buffer which
544 * may be 4 bytes less than the kernel will use, resulting in
545 * userspace memory corruption (which is not detectable by valgrind
546 * too, in most cases).
547 * So for now, let's align to 64 instead of HOST_LONG_BITS here, in
548 * a hope that sizeof(long) won't become >8 any time soon.
550 size
= ALIGN(((mem
->memory_size
) >> TARGET_PAGE_BITS
),
551 /*HOST_LONG_BITS*/ 64) / 8;
552 if (!mem
->dirty_bmap
) {
553 /* Allocate on the first log_sync, once and for all */
554 mem
->dirty_bmap
= g_malloc0(size
);
557 d
.dirty_bitmap
= mem
->dirty_bmap
;
558 d
.slot
= mem
->slot
| (kml
->as_id
<< 16);
559 if (kvm_vm_ioctl(s
, KVM_GET_DIRTY_LOG
, &d
) == -1) {
560 DPRINTF("ioctl failed %d\n", errno
);
565 kvm_get_dirty_pages_log_range(section
, d
.dirty_bitmap
);
571 /* Alignment requirement for KVM_CLEAR_DIRTY_LOG - 64 pages */
572 #define KVM_CLEAR_LOG_SHIFT 6
573 #define KVM_CLEAR_LOG_ALIGN (qemu_real_host_page_size << KVM_CLEAR_LOG_SHIFT)
574 #define KVM_CLEAR_LOG_MASK (-KVM_CLEAR_LOG_ALIGN)
577 * kvm_physical_log_clear - Clear the kernel's dirty bitmap for range
579 * NOTE: this will be a no-op if we haven't enabled manual dirty log
580 * protection in the host kernel because in that case this operation
581 * will be done within log_sync().
583 * @kml: the kvm memory listener
584 * @section: the memory range to clear dirty bitmap
586 static int kvm_physical_log_clear(KVMMemoryListener
*kml
,
587 MemoryRegionSection
*section
)
589 KVMState
*s
= kvm_state
;
590 struct kvm_clear_dirty_log d
;
591 uint64_t start
, end
, bmap_start
, start_delta
, bmap_npages
, size
;
592 unsigned long *bmap_clear
= NULL
, psize
= qemu_real_host_page_size
;
596 if (!s
->manual_dirty_log_protect
) {
597 /* No need to do explicit clear */
601 start
= section
->offset_within_address_space
;
602 size
= int128_get64(section
->size
);
605 /* Nothing more we can do... */
611 /* Find any possible slot that covers the section */
612 for (i
= 0; i
< s
->nr_slots
; i
++) {
613 mem
= &kml
->slots
[i
];
614 if (mem
->start_addr
<= start
&&
615 start
+ size
<= mem
->start_addr
+ mem
->memory_size
) {
621 * We should always find one memslot until this point, otherwise
622 * there could be something wrong from the upper layer
624 assert(mem
&& i
!= s
->nr_slots
);
627 * We need to extend either the start or the size or both to
628 * satisfy the KVM interface requirement. Firstly, do the start
629 * page alignment on 64 host pages
631 bmap_start
= (start
- mem
->start_addr
) & KVM_CLEAR_LOG_MASK
;
632 start_delta
= start
- mem
->start_addr
- bmap_start
;
636 * The kernel interface has restriction on the size too, that either:
638 * (1) the size is 64 host pages aligned (just like the start), or
639 * (2) the size fills up until the end of the KVM memslot.
641 bmap_npages
= DIV_ROUND_UP(size
+ start_delta
, KVM_CLEAR_LOG_ALIGN
)
642 << KVM_CLEAR_LOG_SHIFT
;
643 end
= mem
->memory_size
/ psize
;
644 if (bmap_npages
> end
- bmap_start
) {
645 bmap_npages
= end
- bmap_start
;
647 start_delta
/= psize
;
650 * Prepare the bitmap to clear dirty bits. Here we must guarantee
651 * that we won't clear any unknown dirty bits otherwise we might
652 * accidentally clear some set bits which are not yet synced from
653 * the kernel into QEMU's bitmap, then we'll lose track of the
654 * guest modifications upon those pages (which can directly lead
655 * to guest data loss or panic after migration).
657 * Layout of the KVMSlot.dirty_bmap:
659 * |<-------- bmap_npages -----------..>|
662 * |----------------|-------------|------------------|------------|
665 * start bmap_start (start) end
666 * of memslot of memslot
668 * [1] bmap_npages can be aligned to either 64 pages or the end of slot
671 assert(bmap_start
% BITS_PER_LONG
== 0);
672 /* We should never do log_clear before log_sync */
673 assert(mem
->dirty_bmap
);
675 /* Slow path - we need to manipulate a temp bitmap */
676 bmap_clear
= bitmap_new(bmap_npages
);
677 bitmap_copy_with_src_offset(bmap_clear
, mem
->dirty_bmap
,
678 bmap_start
, start_delta
+ size
/ psize
);
680 * We need to fill the holes at start because that was not
681 * specified by the caller and we extended the bitmap only for
684 bitmap_clear(bmap_clear
, 0, start_delta
);
685 d
.dirty_bitmap
= bmap_clear
;
687 /* Fast path - start address aligns well with BITS_PER_LONG */
688 d
.dirty_bitmap
= mem
->dirty_bmap
+ BIT_WORD(bmap_start
);
691 d
.first_page
= bmap_start
;
692 /* It should never overflow. If it happens, say something */
693 assert(bmap_npages
<= UINT32_MAX
);
694 d
.num_pages
= bmap_npages
;
695 d
.slot
= mem
->slot
| (kml
->as_id
<< 16);
697 if (kvm_vm_ioctl(s
, KVM_CLEAR_DIRTY_LOG
, &d
) == -1) {
699 error_report("%s: KVM_CLEAR_DIRTY_LOG failed, slot=%d, "
700 "start=0x%"PRIx64
", size=0x%"PRIx32
", errno=%d",
701 __func__
, d
.slot
, (uint64_t)d
.first_page
,
702 (uint32_t)d
.num_pages
, ret
);
705 trace_kvm_clear_dirty_log(d
.slot
, d
.first_page
, d
.num_pages
);
709 * After we have updated the remote dirty bitmap, we update the
710 * cached bitmap as well for the memslot, then if another user
711 * clears the same region we know we shouldn't clear it again on
712 * the remote otherwise it's data loss as well.
714 bitmap_clear(mem
->dirty_bmap
, bmap_start
+ start_delta
,
716 /* This handles the NULL case well */
719 kvm_slots_unlock(kml
);
724 static void kvm_coalesce_mmio_region(MemoryListener
*listener
,
725 MemoryRegionSection
*secion
,
726 hwaddr start
, hwaddr size
)
728 KVMState
*s
= kvm_state
;
730 if (s
->coalesced_mmio
) {
731 struct kvm_coalesced_mmio_zone zone
;
737 (void)kvm_vm_ioctl(s
, KVM_REGISTER_COALESCED_MMIO
, &zone
);
741 static void kvm_uncoalesce_mmio_region(MemoryListener
*listener
,
742 MemoryRegionSection
*secion
,
743 hwaddr start
, hwaddr size
)
745 KVMState
*s
= kvm_state
;
747 if (s
->coalesced_mmio
) {
748 struct kvm_coalesced_mmio_zone zone
;
754 (void)kvm_vm_ioctl(s
, KVM_UNREGISTER_COALESCED_MMIO
, &zone
);
758 static void kvm_coalesce_pio_add(MemoryListener
*listener
,
759 MemoryRegionSection
*section
,
760 hwaddr start
, hwaddr size
)
762 KVMState
*s
= kvm_state
;
764 if (s
->coalesced_pio
) {
765 struct kvm_coalesced_mmio_zone zone
;
771 (void)kvm_vm_ioctl(s
, KVM_REGISTER_COALESCED_MMIO
, &zone
);
775 static void kvm_coalesce_pio_del(MemoryListener
*listener
,
776 MemoryRegionSection
*section
,
777 hwaddr start
, hwaddr size
)
779 KVMState
*s
= kvm_state
;
781 if (s
->coalesced_pio
) {
782 struct kvm_coalesced_mmio_zone zone
;
788 (void)kvm_vm_ioctl(s
, KVM_UNREGISTER_COALESCED_MMIO
, &zone
);
792 static MemoryListener kvm_coalesced_pio_listener
= {
793 .coalesced_io_add
= kvm_coalesce_pio_add
,
794 .coalesced_io_del
= kvm_coalesce_pio_del
,
797 int kvm_check_extension(KVMState
*s
, unsigned int extension
)
801 ret
= kvm_ioctl(s
, KVM_CHECK_EXTENSION
, extension
);
809 int kvm_vm_check_extension(KVMState
*s
, unsigned int extension
)
813 ret
= kvm_vm_ioctl(s
, KVM_CHECK_EXTENSION
, extension
);
815 /* VM wide version not implemented, use global one instead */
816 ret
= kvm_check_extension(s
, extension
);
822 static uint32_t adjust_ioeventfd_endianness(uint32_t val
, uint32_t size
)
824 #if defined(HOST_WORDS_BIGENDIAN) != defined(TARGET_WORDS_BIGENDIAN)
825 /* The kernel expects ioeventfd values in HOST_WORDS_BIGENDIAN
826 * endianness, but the memory core hands them in target endianness.
827 * For example, PPC is always treated as big-endian even if running
828 * on KVM and on PPC64LE. Correct here.
842 static int kvm_set_ioeventfd_mmio(int fd
, hwaddr addr
, uint32_t val
,
843 bool assign
, uint32_t size
, bool datamatch
)
846 struct kvm_ioeventfd iofd
= {
847 .datamatch
= datamatch
? adjust_ioeventfd_endianness(val
, size
) : 0,
854 trace_kvm_set_ioeventfd_mmio(fd
, (uint64_t)addr
, val
, assign
, size
,
856 if (!kvm_enabled()) {
861 iofd
.flags
|= KVM_IOEVENTFD_FLAG_DATAMATCH
;
864 iofd
.flags
|= KVM_IOEVENTFD_FLAG_DEASSIGN
;
867 ret
= kvm_vm_ioctl(kvm_state
, KVM_IOEVENTFD
, &iofd
);
876 static int kvm_set_ioeventfd_pio(int fd
, uint16_t addr
, uint16_t val
,
877 bool assign
, uint32_t size
, bool datamatch
)
879 struct kvm_ioeventfd kick
= {
880 .datamatch
= datamatch
? adjust_ioeventfd_endianness(val
, size
) : 0,
882 .flags
= KVM_IOEVENTFD_FLAG_PIO
,
887 trace_kvm_set_ioeventfd_pio(fd
, addr
, val
, assign
, size
, datamatch
);
888 if (!kvm_enabled()) {
892 kick
.flags
|= KVM_IOEVENTFD_FLAG_DATAMATCH
;
895 kick
.flags
|= KVM_IOEVENTFD_FLAG_DEASSIGN
;
897 r
= kvm_vm_ioctl(kvm_state
, KVM_IOEVENTFD
, &kick
);
905 static int kvm_check_many_ioeventfds(void)
907 /* Userspace can use ioeventfd for io notification. This requires a host
908 * that supports eventfd(2) and an I/O thread; since eventfd does not
909 * support SIGIO it cannot interrupt the vcpu.
911 * Older kernels have a 6 device limit on the KVM io bus. Find out so we
912 * can avoid creating too many ioeventfds.
914 #if defined(CONFIG_EVENTFD)
917 for (i
= 0; i
< ARRAY_SIZE(ioeventfds
); i
++) {
918 ioeventfds
[i
] = eventfd(0, EFD_CLOEXEC
);
919 if (ioeventfds
[i
] < 0) {
922 ret
= kvm_set_ioeventfd_pio(ioeventfds
[i
], 0, i
, true, 2, true);
924 close(ioeventfds
[i
]);
929 /* Decide whether many devices are supported or not */
930 ret
= i
== ARRAY_SIZE(ioeventfds
);
933 kvm_set_ioeventfd_pio(ioeventfds
[i
], 0, i
, false, 2, true);
934 close(ioeventfds
[i
]);
942 static const KVMCapabilityInfo
*
943 kvm_check_extension_list(KVMState
*s
, const KVMCapabilityInfo
*list
)
946 if (!kvm_check_extension(s
, list
->value
)) {
954 static void kvm_set_phys_mem(KVMMemoryListener
*kml
,
955 MemoryRegionSection
*section
, bool add
)
959 MemoryRegion
*mr
= section
->mr
;
960 bool writeable
= !mr
->readonly
&& !mr
->rom_device
;
961 hwaddr start_addr
, size
;
964 if (!memory_region_is_ram(mr
)) {
965 if (writeable
|| !kvm_readonly_mem_allowed
) {
967 } else if (!mr
->romd_mode
) {
968 /* If the memory device is not in romd_mode, then we actually want
969 * to remove the kvm memory slot so all accesses will trap. */
974 size
= kvm_align_section(section
, &start_addr
);
979 /* use aligned delta to align the ram address */
980 ram
= memory_region_get_ram_ptr(mr
) + section
->offset_within_region
+
981 (start_addr
- section
->offset_within_address_space
);
986 mem
= kvm_lookup_matching_slot(kml
, start_addr
, size
);
990 if (mem
->flags
& KVM_MEM_LOG_DIRTY_PAGES
) {
991 kvm_physical_sync_dirty_bitmap(kml
, section
);
994 /* unregister the slot */
995 g_free(mem
->dirty_bmap
);
996 mem
->dirty_bmap
= NULL
;
997 mem
->memory_size
= 0;
999 err
= kvm_set_user_memory_region(kml
, mem
, false);
1001 fprintf(stderr
, "%s: error unregistering slot: %s\n",
1002 __func__
, strerror(-err
));
1008 /* register the new slot */
1009 mem
= kvm_alloc_slot(kml
);
1010 mem
->memory_size
= size
;
1011 mem
->start_addr
= start_addr
;
1013 mem
->flags
= kvm_mem_flags(mr
);
1015 err
= kvm_set_user_memory_region(kml
, mem
, true);
1017 fprintf(stderr
, "%s: error registering slot: %s\n", __func__
,
1023 kvm_slots_unlock(kml
);
1026 static void kvm_region_add(MemoryListener
*listener
,
1027 MemoryRegionSection
*section
)
1029 KVMMemoryListener
*kml
= container_of(listener
, KVMMemoryListener
, listener
);
1031 memory_region_ref(section
->mr
);
1032 kvm_set_phys_mem(kml
, section
, true);
1035 static void kvm_region_del(MemoryListener
*listener
,
1036 MemoryRegionSection
*section
)
1038 KVMMemoryListener
*kml
= container_of(listener
, KVMMemoryListener
, listener
);
1040 kvm_set_phys_mem(kml
, section
, false);
1041 memory_region_unref(section
->mr
);
1044 static void kvm_log_sync(MemoryListener
*listener
,
1045 MemoryRegionSection
*section
)
1047 KVMMemoryListener
*kml
= container_of(listener
, KVMMemoryListener
, listener
);
1050 kvm_slots_lock(kml
);
1051 r
= kvm_physical_sync_dirty_bitmap(kml
, section
);
1052 kvm_slots_unlock(kml
);
1058 static void kvm_log_clear(MemoryListener
*listener
,
1059 MemoryRegionSection
*section
)
1061 KVMMemoryListener
*kml
= container_of(listener
, KVMMemoryListener
, listener
);
1064 r
= kvm_physical_log_clear(kml
, section
);
1066 error_report_once("%s: kvm log clear failed: mr=%s "
1067 "offset=%"HWADDR_PRIx
" size=%"PRIx64
, __func__
,
1068 section
->mr
->name
, section
->offset_within_region
,
1069 int128_get64(section
->size
));
1074 static void kvm_mem_ioeventfd_add(MemoryListener
*listener
,
1075 MemoryRegionSection
*section
,
1076 bool match_data
, uint64_t data
,
1079 int fd
= event_notifier_get_fd(e
);
1082 r
= kvm_set_ioeventfd_mmio(fd
, section
->offset_within_address_space
,
1083 data
, true, int128_get64(section
->size
),
1086 fprintf(stderr
, "%s: error adding ioeventfd: %s (%d)\n",
1087 __func__
, strerror(-r
), -r
);
1092 static void kvm_mem_ioeventfd_del(MemoryListener
*listener
,
1093 MemoryRegionSection
*section
,
1094 bool match_data
, uint64_t data
,
1097 int fd
= event_notifier_get_fd(e
);
1100 r
= kvm_set_ioeventfd_mmio(fd
, section
->offset_within_address_space
,
1101 data
, false, int128_get64(section
->size
),
1104 fprintf(stderr
, "%s: error deleting ioeventfd: %s (%d)\n",
1105 __func__
, strerror(-r
), -r
);
1110 static void kvm_io_ioeventfd_add(MemoryListener
*listener
,
1111 MemoryRegionSection
*section
,
1112 bool match_data
, uint64_t data
,
1115 int fd
= event_notifier_get_fd(e
);
1118 r
= kvm_set_ioeventfd_pio(fd
, section
->offset_within_address_space
,
1119 data
, true, int128_get64(section
->size
),
1122 fprintf(stderr
, "%s: error adding ioeventfd: %s (%d)\n",
1123 __func__
, strerror(-r
), -r
);
1128 static void kvm_io_ioeventfd_del(MemoryListener
*listener
,
1129 MemoryRegionSection
*section
,
1130 bool match_data
, uint64_t data
,
1134 int fd
= event_notifier_get_fd(e
);
1137 r
= kvm_set_ioeventfd_pio(fd
, section
->offset_within_address_space
,
1138 data
, false, int128_get64(section
->size
),
1141 fprintf(stderr
, "%s: error deleting ioeventfd: %s (%d)\n",
1142 __func__
, strerror(-r
), -r
);
1147 void kvm_memory_listener_register(KVMState
*s
, KVMMemoryListener
*kml
,
1148 AddressSpace
*as
, int as_id
)
1152 qemu_mutex_init(&kml
->slots_lock
);
1153 kml
->slots
= g_malloc0(s
->nr_slots
* sizeof(KVMSlot
));
1156 for (i
= 0; i
< s
->nr_slots
; i
++) {
1157 kml
->slots
[i
].slot
= i
;
1160 kml
->listener
.region_add
= kvm_region_add
;
1161 kml
->listener
.region_del
= kvm_region_del
;
1162 kml
->listener
.log_start
= kvm_log_start
;
1163 kml
->listener
.log_stop
= kvm_log_stop
;
1164 kml
->listener
.log_sync
= kvm_log_sync
;
1165 kml
->listener
.log_clear
= kvm_log_clear
;
1166 kml
->listener
.priority
= 10;
1168 memory_listener_register(&kml
->listener
, as
);
1170 for (i
= 0; i
< s
->nr_as
; ++i
) {
1179 static MemoryListener kvm_io_listener
= {
1180 .eventfd_add
= kvm_io_ioeventfd_add
,
1181 .eventfd_del
= kvm_io_ioeventfd_del
,
1185 int kvm_set_irq(KVMState
*s
, int irq
, int level
)
1187 struct kvm_irq_level event
;
1190 assert(kvm_async_interrupts_enabled());
1192 event
.level
= level
;
1194 ret
= kvm_vm_ioctl(s
, s
->irq_set_ioctl
, &event
);
1196 perror("kvm_set_irq");
1200 return (s
->irq_set_ioctl
== KVM_IRQ_LINE
) ? 1 : event
.status
;
1203 #ifdef KVM_CAP_IRQ_ROUTING
1204 typedef struct KVMMSIRoute
{
1205 struct kvm_irq_routing_entry kroute
;
1206 QTAILQ_ENTRY(KVMMSIRoute
) entry
;
1209 static void set_gsi(KVMState
*s
, unsigned int gsi
)
1211 set_bit(gsi
, s
->used_gsi_bitmap
);
1214 static void clear_gsi(KVMState
*s
, unsigned int gsi
)
1216 clear_bit(gsi
, s
->used_gsi_bitmap
);
1219 void kvm_init_irq_routing(KVMState
*s
)
1223 gsi_count
= kvm_check_extension(s
, KVM_CAP_IRQ_ROUTING
) - 1;
1224 if (gsi_count
> 0) {
1225 /* Round up so we can search ints using ffs */
1226 s
->used_gsi_bitmap
= bitmap_new(gsi_count
);
1227 s
->gsi_count
= gsi_count
;
1230 s
->irq_routes
= g_malloc0(sizeof(*s
->irq_routes
));
1231 s
->nr_allocated_irq_routes
= 0;
1233 if (!kvm_direct_msi_allowed
) {
1234 for (i
= 0; i
< KVM_MSI_HASHTAB_SIZE
; i
++) {
1235 QTAILQ_INIT(&s
->msi_hashtab
[i
]);
1239 kvm_arch_init_irq_routing(s
);
1242 void kvm_irqchip_commit_routes(KVMState
*s
)
1246 if (kvm_gsi_direct_mapping()) {
1250 if (!kvm_gsi_routing_enabled()) {
1254 s
->irq_routes
->flags
= 0;
1255 trace_kvm_irqchip_commit_routes();
1256 ret
= kvm_vm_ioctl(s
, KVM_SET_GSI_ROUTING
, s
->irq_routes
);
1260 static void kvm_add_routing_entry(KVMState
*s
,
1261 struct kvm_irq_routing_entry
*entry
)
1263 struct kvm_irq_routing_entry
*new;
1266 if (s
->irq_routes
->nr
== s
->nr_allocated_irq_routes
) {
1267 n
= s
->nr_allocated_irq_routes
* 2;
1271 size
= sizeof(struct kvm_irq_routing
);
1272 size
+= n
* sizeof(*new);
1273 s
->irq_routes
= g_realloc(s
->irq_routes
, size
);
1274 s
->nr_allocated_irq_routes
= n
;
1276 n
= s
->irq_routes
->nr
++;
1277 new = &s
->irq_routes
->entries
[n
];
1281 set_gsi(s
, entry
->gsi
);
1284 static int kvm_update_routing_entry(KVMState
*s
,
1285 struct kvm_irq_routing_entry
*new_entry
)
1287 struct kvm_irq_routing_entry
*entry
;
1290 for (n
= 0; n
< s
->irq_routes
->nr
; n
++) {
1291 entry
= &s
->irq_routes
->entries
[n
];
1292 if (entry
->gsi
!= new_entry
->gsi
) {
1296 if(!memcmp(entry
, new_entry
, sizeof *entry
)) {
1300 *entry
= *new_entry
;
1308 void kvm_irqchip_add_irq_route(KVMState
*s
, int irq
, int irqchip
, int pin
)
1310 struct kvm_irq_routing_entry e
= {};
1312 assert(pin
< s
->gsi_count
);
1315 e
.type
= KVM_IRQ_ROUTING_IRQCHIP
;
1317 e
.u
.irqchip
.irqchip
= irqchip
;
1318 e
.u
.irqchip
.pin
= pin
;
1319 kvm_add_routing_entry(s
, &e
);
1322 void kvm_irqchip_release_virq(KVMState
*s
, int virq
)
1324 struct kvm_irq_routing_entry
*e
;
1327 if (kvm_gsi_direct_mapping()) {
1331 for (i
= 0; i
< s
->irq_routes
->nr
; i
++) {
1332 e
= &s
->irq_routes
->entries
[i
];
1333 if (e
->gsi
== virq
) {
1334 s
->irq_routes
->nr
--;
1335 *e
= s
->irq_routes
->entries
[s
->irq_routes
->nr
];
1339 kvm_arch_release_virq_post(virq
);
1340 trace_kvm_irqchip_release_virq(virq
);
1343 static unsigned int kvm_hash_msi(uint32_t data
)
1345 /* This is optimized for IA32 MSI layout. However, no other arch shall
1346 * repeat the mistake of not providing a direct MSI injection API. */
1350 static void kvm_flush_dynamic_msi_routes(KVMState
*s
)
1352 KVMMSIRoute
*route
, *next
;
1355 for (hash
= 0; hash
< KVM_MSI_HASHTAB_SIZE
; hash
++) {
1356 QTAILQ_FOREACH_SAFE(route
, &s
->msi_hashtab
[hash
], entry
, next
) {
1357 kvm_irqchip_release_virq(s
, route
->kroute
.gsi
);
1358 QTAILQ_REMOVE(&s
->msi_hashtab
[hash
], route
, entry
);
1364 static int kvm_irqchip_get_virq(KVMState
*s
)
1369 * PIC and IOAPIC share the first 16 GSI numbers, thus the available
1370 * GSI numbers are more than the number of IRQ route. Allocating a GSI
1371 * number can succeed even though a new route entry cannot be added.
1372 * When this happens, flush dynamic MSI entries to free IRQ route entries.
1374 if (!kvm_direct_msi_allowed
&& s
->irq_routes
->nr
== s
->gsi_count
) {
1375 kvm_flush_dynamic_msi_routes(s
);
1378 /* Return the lowest unused GSI in the bitmap */
1379 next_virq
= find_first_zero_bit(s
->used_gsi_bitmap
, s
->gsi_count
);
1380 if (next_virq
>= s
->gsi_count
) {
1387 static KVMMSIRoute
*kvm_lookup_msi_route(KVMState
*s
, MSIMessage msg
)
1389 unsigned int hash
= kvm_hash_msi(msg
.data
);
1392 QTAILQ_FOREACH(route
, &s
->msi_hashtab
[hash
], entry
) {
1393 if (route
->kroute
.u
.msi
.address_lo
== (uint32_t)msg
.address
&&
1394 route
->kroute
.u
.msi
.address_hi
== (msg
.address
>> 32) &&
1395 route
->kroute
.u
.msi
.data
== le32_to_cpu(msg
.data
)) {
1402 int kvm_irqchip_send_msi(KVMState
*s
, MSIMessage msg
)
1407 if (kvm_direct_msi_allowed
) {
1408 msi
.address_lo
= (uint32_t)msg
.address
;
1409 msi
.address_hi
= msg
.address
>> 32;
1410 msi
.data
= le32_to_cpu(msg
.data
);
1412 memset(msi
.pad
, 0, sizeof(msi
.pad
));
1414 return kvm_vm_ioctl(s
, KVM_SIGNAL_MSI
, &msi
);
1417 route
= kvm_lookup_msi_route(s
, msg
);
1421 virq
= kvm_irqchip_get_virq(s
);
1426 route
= g_malloc0(sizeof(KVMMSIRoute
));
1427 route
->kroute
.gsi
= virq
;
1428 route
->kroute
.type
= KVM_IRQ_ROUTING_MSI
;
1429 route
->kroute
.flags
= 0;
1430 route
->kroute
.u
.msi
.address_lo
= (uint32_t)msg
.address
;
1431 route
->kroute
.u
.msi
.address_hi
= msg
.address
>> 32;
1432 route
->kroute
.u
.msi
.data
= le32_to_cpu(msg
.data
);
1434 kvm_add_routing_entry(s
, &route
->kroute
);
1435 kvm_irqchip_commit_routes(s
);
1437 QTAILQ_INSERT_TAIL(&s
->msi_hashtab
[kvm_hash_msi(msg
.data
)], route
,
1441 assert(route
->kroute
.type
== KVM_IRQ_ROUTING_MSI
);
1443 return kvm_set_irq(s
, route
->kroute
.gsi
, 1);
1446 int kvm_irqchip_add_msi_route(KVMState
*s
, int vector
, PCIDevice
*dev
)
1448 struct kvm_irq_routing_entry kroute
= {};
1450 MSIMessage msg
= {0, 0};
1452 if (pci_available
&& dev
) {
1453 msg
= pci_get_msi_message(dev
, vector
);
1456 if (kvm_gsi_direct_mapping()) {
1457 return kvm_arch_msi_data_to_gsi(msg
.data
);
1460 if (!kvm_gsi_routing_enabled()) {
1464 virq
= kvm_irqchip_get_virq(s
);
1470 kroute
.type
= KVM_IRQ_ROUTING_MSI
;
1472 kroute
.u
.msi
.address_lo
= (uint32_t)msg
.address
;
1473 kroute
.u
.msi
.address_hi
= msg
.address
>> 32;
1474 kroute
.u
.msi
.data
= le32_to_cpu(msg
.data
);
1475 if (pci_available
&& kvm_msi_devid_required()) {
1476 kroute
.flags
= KVM_MSI_VALID_DEVID
;
1477 kroute
.u
.msi
.devid
= pci_requester_id(dev
);
1479 if (kvm_arch_fixup_msi_route(&kroute
, msg
.address
, msg
.data
, dev
)) {
1480 kvm_irqchip_release_virq(s
, virq
);
1484 trace_kvm_irqchip_add_msi_route(dev
? dev
->name
: (char *)"N/A",
1487 kvm_add_routing_entry(s
, &kroute
);
1488 kvm_arch_add_msi_route_post(&kroute
, vector
, dev
);
1489 kvm_irqchip_commit_routes(s
);
1494 int kvm_irqchip_update_msi_route(KVMState
*s
, int virq
, MSIMessage msg
,
1497 struct kvm_irq_routing_entry kroute
= {};
1499 if (kvm_gsi_direct_mapping()) {
1503 if (!kvm_irqchip_in_kernel()) {
1508 kroute
.type
= KVM_IRQ_ROUTING_MSI
;
1510 kroute
.u
.msi
.address_lo
= (uint32_t)msg
.address
;
1511 kroute
.u
.msi
.address_hi
= msg
.address
>> 32;
1512 kroute
.u
.msi
.data
= le32_to_cpu(msg
.data
);
1513 if (pci_available
&& kvm_msi_devid_required()) {
1514 kroute
.flags
= KVM_MSI_VALID_DEVID
;
1515 kroute
.u
.msi
.devid
= pci_requester_id(dev
);
1517 if (kvm_arch_fixup_msi_route(&kroute
, msg
.address
, msg
.data
, dev
)) {
1521 trace_kvm_irqchip_update_msi_route(virq
);
1523 return kvm_update_routing_entry(s
, &kroute
);
1526 static int kvm_irqchip_assign_irqfd(KVMState
*s
, int fd
, int rfd
, int virq
,
1529 struct kvm_irqfd irqfd
= {
1532 .flags
= assign
? 0 : KVM_IRQFD_FLAG_DEASSIGN
,
1536 irqfd
.flags
|= KVM_IRQFD_FLAG_RESAMPLE
;
1537 irqfd
.resamplefd
= rfd
;
1540 if (!kvm_irqfds_enabled()) {
1544 return kvm_vm_ioctl(s
, KVM_IRQFD
, &irqfd
);
1547 int kvm_irqchip_add_adapter_route(KVMState
*s
, AdapterInfo
*adapter
)
1549 struct kvm_irq_routing_entry kroute
= {};
1552 if (!kvm_gsi_routing_enabled()) {
1556 virq
= kvm_irqchip_get_virq(s
);
1562 kroute
.type
= KVM_IRQ_ROUTING_S390_ADAPTER
;
1564 kroute
.u
.adapter
.summary_addr
= adapter
->summary_addr
;
1565 kroute
.u
.adapter
.ind_addr
= adapter
->ind_addr
;
1566 kroute
.u
.adapter
.summary_offset
= adapter
->summary_offset
;
1567 kroute
.u
.adapter
.ind_offset
= adapter
->ind_offset
;
1568 kroute
.u
.adapter
.adapter_id
= adapter
->adapter_id
;
1570 kvm_add_routing_entry(s
, &kroute
);
1575 int kvm_irqchip_add_hv_sint_route(KVMState
*s
, uint32_t vcpu
, uint32_t sint
)
1577 struct kvm_irq_routing_entry kroute
= {};
1580 if (!kvm_gsi_routing_enabled()) {
1583 if (!kvm_check_extension(s
, KVM_CAP_HYPERV_SYNIC
)) {
1586 virq
= kvm_irqchip_get_virq(s
);
1592 kroute
.type
= KVM_IRQ_ROUTING_HV_SINT
;
1594 kroute
.u
.hv_sint
.vcpu
= vcpu
;
1595 kroute
.u
.hv_sint
.sint
= sint
;
1597 kvm_add_routing_entry(s
, &kroute
);
1598 kvm_irqchip_commit_routes(s
);
1603 #else /* !KVM_CAP_IRQ_ROUTING */
1605 void kvm_init_irq_routing(KVMState
*s
)
1609 void kvm_irqchip_release_virq(KVMState
*s
, int virq
)
1613 int kvm_irqchip_send_msi(KVMState
*s
, MSIMessage msg
)
1618 int kvm_irqchip_add_msi_route(KVMState
*s
, int vector
, PCIDevice
*dev
)
1623 int kvm_irqchip_add_adapter_route(KVMState
*s
, AdapterInfo
*adapter
)
1628 int kvm_irqchip_add_hv_sint_route(KVMState
*s
, uint32_t vcpu
, uint32_t sint
)
1633 static int kvm_irqchip_assign_irqfd(KVMState
*s
, int fd
, int virq
, bool assign
)
1638 int kvm_irqchip_update_msi_route(KVMState
*s
, int virq
, MSIMessage msg
)
1642 #endif /* !KVM_CAP_IRQ_ROUTING */
1644 int kvm_irqchip_add_irqfd_notifier_gsi(KVMState
*s
, EventNotifier
*n
,
1645 EventNotifier
*rn
, int virq
)
1647 return kvm_irqchip_assign_irqfd(s
, event_notifier_get_fd(n
),
1648 rn
? event_notifier_get_fd(rn
) : -1, virq
, true);
1651 int kvm_irqchip_remove_irqfd_notifier_gsi(KVMState
*s
, EventNotifier
*n
,
1654 return kvm_irqchip_assign_irqfd(s
, event_notifier_get_fd(n
), -1, virq
,
1658 int kvm_irqchip_add_irqfd_notifier(KVMState
*s
, EventNotifier
*n
,
1659 EventNotifier
*rn
, qemu_irq irq
)
1662 gboolean found
= g_hash_table_lookup_extended(s
->gsimap
, irq
, &key
, &gsi
);
1667 return kvm_irqchip_add_irqfd_notifier_gsi(s
, n
, rn
, GPOINTER_TO_INT(gsi
));
1670 int kvm_irqchip_remove_irqfd_notifier(KVMState
*s
, EventNotifier
*n
,
1674 gboolean found
= g_hash_table_lookup_extended(s
->gsimap
, irq
, &key
, &gsi
);
1679 return kvm_irqchip_remove_irqfd_notifier_gsi(s
, n
, GPOINTER_TO_INT(gsi
));
1682 void kvm_irqchip_set_qemuirq_gsi(KVMState
*s
, qemu_irq irq
, int gsi
)
1684 g_hash_table_insert(s
->gsimap
, irq
, GINT_TO_POINTER(gsi
));
1687 static void kvm_irqchip_create(MachineState
*machine
, KVMState
*s
)
1691 if (kvm_check_extension(s
, KVM_CAP_IRQCHIP
)) {
1693 } else if (kvm_check_extension(s
, KVM_CAP_S390_IRQCHIP
)) {
1694 ret
= kvm_vm_enable_cap(s
, KVM_CAP_S390_IRQCHIP
, 0);
1696 fprintf(stderr
, "Enable kernel irqchip failed: %s\n", strerror(-ret
));
1703 /* First probe and see if there's a arch-specific hook to create the
1704 * in-kernel irqchip for us */
1705 ret
= kvm_arch_irqchip_create(machine
, s
);
1707 if (machine_kernel_irqchip_split(machine
)) {
1708 perror("Split IRQ chip mode not supported.");
1711 ret
= kvm_vm_ioctl(s
, KVM_CREATE_IRQCHIP
);
1715 fprintf(stderr
, "Create kernel irqchip failed: %s\n", strerror(-ret
));
1719 kvm_kernel_irqchip
= true;
1720 /* If we have an in-kernel IRQ chip then we must have asynchronous
1721 * interrupt delivery (though the reverse is not necessarily true)
1723 kvm_async_interrupts_allowed
= true;
1724 kvm_halt_in_kernel_allowed
= true;
1726 kvm_init_irq_routing(s
);
1728 s
->gsimap
= g_hash_table_new(g_direct_hash
, g_direct_equal
);
1731 /* Find number of supported CPUs using the recommended
1732 * procedure from the kernel API documentation to cope with
1733 * older kernels that may be missing capabilities.
1735 static int kvm_recommended_vcpus(KVMState
*s
)
1737 int ret
= kvm_vm_check_extension(s
, KVM_CAP_NR_VCPUS
);
1738 return (ret
) ? ret
: 4;
1741 static int kvm_max_vcpus(KVMState
*s
)
1743 int ret
= kvm_check_extension(s
, KVM_CAP_MAX_VCPUS
);
1744 return (ret
) ? ret
: kvm_recommended_vcpus(s
);
1747 static int kvm_max_vcpu_id(KVMState
*s
)
1749 int ret
= kvm_check_extension(s
, KVM_CAP_MAX_VCPU_ID
);
1750 return (ret
) ? ret
: kvm_max_vcpus(s
);
1753 bool kvm_vcpu_id_is_valid(int vcpu_id
)
1755 KVMState
*s
= KVM_STATE(current_machine
->accelerator
);
1756 return vcpu_id
>= 0 && vcpu_id
< kvm_max_vcpu_id(s
);
1759 static int kvm_init(MachineState
*ms
)
1761 MachineClass
*mc
= MACHINE_GET_CLASS(ms
);
1762 static const char upgrade_note
[] =
1763 "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n"
1764 "(see http://sourceforge.net/projects/kvm).\n";
1769 { "SMP", ms
->smp
.cpus
},
1770 { "hotpluggable", ms
->smp
.max_cpus
},
1773 int soft_vcpus_limit
, hard_vcpus_limit
;
1775 const KVMCapabilityInfo
*missing_cap
;
1778 const char *kvm_type
;
1780 s
= KVM_STATE(ms
->accelerator
);
1783 * On systems where the kernel can support different base page
1784 * sizes, host page size may be different from TARGET_PAGE_SIZE,
1785 * even with KVM. TARGET_PAGE_SIZE is assumed to be the minimum
1786 * page size for the system though.
1788 assert(TARGET_PAGE_SIZE
<= getpagesize());
1792 #ifdef KVM_CAP_SET_GUEST_DEBUG
1793 QTAILQ_INIT(&s
->kvm_sw_breakpoints
);
1795 QLIST_INIT(&s
->kvm_parked_vcpus
);
1797 s
->fd
= qemu_open("/dev/kvm", O_RDWR
);
1799 fprintf(stderr
, "Could not access KVM kernel module: %m\n");
1804 ret
= kvm_ioctl(s
, KVM_GET_API_VERSION
, 0);
1805 if (ret
< KVM_API_VERSION
) {
1809 fprintf(stderr
, "kvm version too old\n");
1813 if (ret
> KVM_API_VERSION
) {
1815 fprintf(stderr
, "kvm version not supported\n");
1819 kvm_immediate_exit
= kvm_check_extension(s
, KVM_CAP_IMMEDIATE_EXIT
);
1820 s
->nr_slots
= kvm_check_extension(s
, KVM_CAP_NR_MEMSLOTS
);
1822 /* If unspecified, use the default value */
1827 s
->nr_as
= kvm_check_extension(s
, KVM_CAP_MULTI_ADDRESS_SPACE
);
1828 if (s
->nr_as
<= 1) {
1831 s
->as
= g_new0(struct KVMAs
, s
->nr_as
);
1833 kvm_type
= qemu_opt_get(qemu_get_machine_opts(), "kvm-type");
1835 type
= mc
->kvm_type(ms
, kvm_type
);
1836 } else if (kvm_type
) {
1838 fprintf(stderr
, "Invalid argument kvm-type=%s\n", kvm_type
);
1843 ret
= kvm_ioctl(s
, KVM_CREATE_VM
, type
);
1844 } while (ret
== -EINTR
);
1847 fprintf(stderr
, "ioctl(KVM_CREATE_VM) failed: %d %s\n", -ret
,
1851 if (ret
== -EINVAL
) {
1853 "Host kernel setup problem detected. Please verify:\n");
1854 fprintf(stderr
, "- for kernels supporting the switch_amode or"
1855 " user_mode parameters, whether\n");
1857 " user space is running in primary address space\n");
1859 "- for kernels supporting the vm.allocate_pgste sysctl, "
1860 "whether it is enabled\n");
1868 /* check the vcpu limits */
1869 soft_vcpus_limit
= kvm_recommended_vcpus(s
);
1870 hard_vcpus_limit
= kvm_max_vcpus(s
);
1873 if (nc
->num
> soft_vcpus_limit
) {
1874 warn_report("Number of %s cpus requested (%d) exceeds "
1875 "the recommended cpus supported by KVM (%d)",
1876 nc
->name
, nc
->num
, soft_vcpus_limit
);
1878 if (nc
->num
> hard_vcpus_limit
) {
1879 fprintf(stderr
, "Number of %s cpus requested (%d) exceeds "
1880 "the maximum cpus supported by KVM (%d)\n",
1881 nc
->name
, nc
->num
, hard_vcpus_limit
);
1888 missing_cap
= kvm_check_extension_list(s
, kvm_required_capabilites
);
1891 kvm_check_extension_list(s
, kvm_arch_required_capabilities
);
1895 fprintf(stderr
, "kvm does not support %s\n%s",
1896 missing_cap
->name
, upgrade_note
);
1900 s
->coalesced_mmio
= kvm_check_extension(s
, KVM_CAP_COALESCED_MMIO
);
1901 s
->coalesced_pio
= s
->coalesced_mmio
&&
1902 kvm_check_extension(s
, KVM_CAP_COALESCED_PIO
);
1904 s
->manual_dirty_log_protect
=
1905 kvm_check_extension(s
, KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2
);
1906 if (s
->manual_dirty_log_protect
) {
1907 ret
= kvm_vm_enable_cap(s
, KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2
, 0, 1);
1909 warn_report("Trying to enable KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2 "
1910 "but failed. Falling back to the legacy mode. ");
1911 s
->manual_dirty_log_protect
= false;
1915 #ifdef KVM_CAP_VCPU_EVENTS
1916 s
->vcpu_events
= kvm_check_extension(s
, KVM_CAP_VCPU_EVENTS
);
1919 s
->robust_singlestep
=
1920 kvm_check_extension(s
, KVM_CAP_X86_ROBUST_SINGLESTEP
);
1922 #ifdef KVM_CAP_DEBUGREGS
1923 s
->debugregs
= kvm_check_extension(s
, KVM_CAP_DEBUGREGS
);
1926 s
->max_nested_state_len
= kvm_check_extension(s
, KVM_CAP_NESTED_STATE
);
1928 #ifdef KVM_CAP_IRQ_ROUTING
1929 kvm_direct_msi_allowed
= (kvm_check_extension(s
, KVM_CAP_SIGNAL_MSI
) > 0);
1932 s
->intx_set_mask
= kvm_check_extension(s
, KVM_CAP_PCI_2_3
);
1934 s
->irq_set_ioctl
= KVM_IRQ_LINE
;
1935 if (kvm_check_extension(s
, KVM_CAP_IRQ_INJECT_STATUS
)) {
1936 s
->irq_set_ioctl
= KVM_IRQ_LINE_STATUS
;
1939 kvm_readonly_mem_allowed
=
1940 (kvm_check_extension(s
, KVM_CAP_READONLY_MEM
) > 0);
1942 kvm_eventfds_allowed
=
1943 (kvm_check_extension(s
, KVM_CAP_IOEVENTFD
) > 0);
1945 kvm_irqfds_allowed
=
1946 (kvm_check_extension(s
, KVM_CAP_IRQFD
) > 0);
1948 kvm_resamplefds_allowed
=
1949 (kvm_check_extension(s
, KVM_CAP_IRQFD_RESAMPLE
) > 0);
1951 kvm_vm_attributes_allowed
=
1952 (kvm_check_extension(s
, KVM_CAP_VM_ATTRIBUTES
) > 0);
1954 kvm_ioeventfd_any_length_allowed
=
1955 (kvm_check_extension(s
, KVM_CAP_IOEVENTFD_ANY_LENGTH
) > 0);
1960 * if memory encryption object is specified then initialize the memory
1961 * encryption context.
1963 if (ms
->memory_encryption
) {
1964 kvm_state
->memcrypt_handle
= sev_guest_init(ms
->memory_encryption
);
1965 if (!kvm_state
->memcrypt_handle
) {
1970 kvm_state
->memcrypt_encrypt_data
= sev_encrypt_data
;
1973 ret
= kvm_arch_init(ms
, s
);
1978 if (machine_kernel_irqchip_allowed(ms
)) {
1979 kvm_irqchip_create(ms
, s
);
1982 if (kvm_eventfds_allowed
) {
1983 s
->memory_listener
.listener
.eventfd_add
= kvm_mem_ioeventfd_add
;
1984 s
->memory_listener
.listener
.eventfd_del
= kvm_mem_ioeventfd_del
;
1986 s
->memory_listener
.listener
.coalesced_io_add
= kvm_coalesce_mmio_region
;
1987 s
->memory_listener
.listener
.coalesced_io_del
= kvm_uncoalesce_mmio_region
;
1989 kvm_memory_listener_register(s
, &s
->memory_listener
,
1990 &address_space_memory
, 0);
1991 memory_listener_register(&kvm_io_listener
,
1993 memory_listener_register(&kvm_coalesced_pio_listener
,
1996 s
->many_ioeventfds
= kvm_check_many_ioeventfds();
1998 s
->sync_mmu
= !!kvm_vm_check_extension(kvm_state
, KVM_CAP_SYNC_MMU
);
2000 qemu_balloon_inhibit(true);
2013 g_free(s
->memory_listener
.slots
);
2018 void kvm_set_sigmask_len(KVMState
*s
, unsigned int sigmask_len
)
2020 s
->sigmask_len
= sigmask_len
;
2023 static void kvm_handle_io(uint16_t port
, MemTxAttrs attrs
, void *data
, int direction
,
2024 int size
, uint32_t count
)
2027 uint8_t *ptr
= data
;
2029 for (i
= 0; i
< count
; i
++) {
2030 address_space_rw(&address_space_io
, port
, attrs
,
2032 direction
== KVM_EXIT_IO_OUT
);
2037 static int kvm_handle_internal_error(CPUState
*cpu
, struct kvm_run
*run
)
2039 fprintf(stderr
, "KVM internal error. Suberror: %d\n",
2040 run
->internal
.suberror
);
2042 if (kvm_check_extension(kvm_state
, KVM_CAP_INTERNAL_ERROR_DATA
)) {
2045 for (i
= 0; i
< run
->internal
.ndata
; ++i
) {
2046 fprintf(stderr
, "extra data[%d]: %"PRIx64
"\n",
2047 i
, (uint64_t)run
->internal
.data
[i
]);
2050 if (run
->internal
.suberror
== KVM_INTERNAL_ERROR_EMULATION
) {
2051 fprintf(stderr
, "emulation failure\n");
2052 if (!kvm_arch_stop_on_emulation_error(cpu
)) {
2053 cpu_dump_state(cpu
, stderr
, CPU_DUMP_CODE
);
2054 return EXCP_INTERRUPT
;
2057 /* FIXME: Should trigger a qmp message to let management know
2058 * something went wrong.
2063 void kvm_flush_coalesced_mmio_buffer(void)
2065 KVMState
*s
= kvm_state
;
2067 if (s
->coalesced_flush_in_progress
) {
2071 s
->coalesced_flush_in_progress
= true;
2073 if (s
->coalesced_mmio_ring
) {
2074 struct kvm_coalesced_mmio_ring
*ring
= s
->coalesced_mmio_ring
;
2075 while (ring
->first
!= ring
->last
) {
2076 struct kvm_coalesced_mmio
*ent
;
2078 ent
= &ring
->coalesced_mmio
[ring
->first
];
2080 if (ent
->pio
== 1) {
2081 address_space_rw(&address_space_io
, ent
->phys_addr
,
2082 MEMTXATTRS_UNSPECIFIED
, ent
->data
,
2085 cpu_physical_memory_write(ent
->phys_addr
, ent
->data
, ent
->len
);
2088 ring
->first
= (ring
->first
+ 1) % KVM_COALESCED_MMIO_MAX
;
2092 s
->coalesced_flush_in_progress
= false;
2095 static void do_kvm_cpu_synchronize_state(CPUState
*cpu
, run_on_cpu_data arg
)
2097 if (!cpu
->vcpu_dirty
) {
2098 kvm_arch_get_registers(cpu
);
2099 cpu
->vcpu_dirty
= true;
2103 void kvm_cpu_synchronize_state(CPUState
*cpu
)
2105 if (!cpu
->vcpu_dirty
) {
2106 run_on_cpu(cpu
, do_kvm_cpu_synchronize_state
, RUN_ON_CPU_NULL
);
2110 static void do_kvm_cpu_synchronize_post_reset(CPUState
*cpu
, run_on_cpu_data arg
)
2112 kvm_arch_put_registers(cpu
, KVM_PUT_RESET_STATE
);
2113 cpu
->vcpu_dirty
= false;
2116 void kvm_cpu_synchronize_post_reset(CPUState
*cpu
)
2118 run_on_cpu(cpu
, do_kvm_cpu_synchronize_post_reset
, RUN_ON_CPU_NULL
);
2121 static void do_kvm_cpu_synchronize_post_init(CPUState
*cpu
, run_on_cpu_data arg
)
2123 kvm_arch_put_registers(cpu
, KVM_PUT_FULL_STATE
);
2124 cpu
->vcpu_dirty
= false;
2127 void kvm_cpu_synchronize_post_init(CPUState
*cpu
)
2129 run_on_cpu(cpu
, do_kvm_cpu_synchronize_post_init
, RUN_ON_CPU_NULL
);
2132 static void do_kvm_cpu_synchronize_pre_loadvm(CPUState
*cpu
, run_on_cpu_data arg
)
2134 cpu
->vcpu_dirty
= true;
2137 void kvm_cpu_synchronize_pre_loadvm(CPUState
*cpu
)
2139 run_on_cpu(cpu
, do_kvm_cpu_synchronize_pre_loadvm
, RUN_ON_CPU_NULL
);
2142 #ifdef KVM_HAVE_MCE_INJECTION
2143 static __thread
void *pending_sigbus_addr
;
2144 static __thread
int pending_sigbus_code
;
2145 static __thread
bool have_sigbus_pending
;
2148 static void kvm_cpu_kick(CPUState
*cpu
)
2150 atomic_set(&cpu
->kvm_run
->immediate_exit
, 1);
2153 static void kvm_cpu_kick_self(void)
2155 if (kvm_immediate_exit
) {
2156 kvm_cpu_kick(current_cpu
);
2158 qemu_cpu_kick_self();
2162 static void kvm_eat_signals(CPUState
*cpu
)
2164 struct timespec ts
= { 0, 0 };
2170 if (kvm_immediate_exit
) {
2171 atomic_set(&cpu
->kvm_run
->immediate_exit
, 0);
2172 /* Write kvm_run->immediate_exit before the cpu->exit_request
2173 * write in kvm_cpu_exec.
2179 sigemptyset(&waitset
);
2180 sigaddset(&waitset
, SIG_IPI
);
2183 r
= sigtimedwait(&waitset
, &siginfo
, &ts
);
2184 if (r
== -1 && !(errno
== EAGAIN
|| errno
== EINTR
)) {
2185 perror("sigtimedwait");
2189 r
= sigpending(&chkset
);
2191 perror("sigpending");
2194 } while (sigismember(&chkset
, SIG_IPI
));
2197 int kvm_cpu_exec(CPUState
*cpu
)
2199 struct kvm_run
*run
= cpu
->kvm_run
;
2202 DPRINTF("kvm_cpu_exec()\n");
2204 if (kvm_arch_process_async_events(cpu
)) {
2205 atomic_set(&cpu
->exit_request
, 0);
2209 qemu_mutex_unlock_iothread();
2210 cpu_exec_start(cpu
);
2215 if (cpu
->vcpu_dirty
) {
2216 kvm_arch_put_registers(cpu
, KVM_PUT_RUNTIME_STATE
);
2217 cpu
->vcpu_dirty
= false;
2220 kvm_arch_pre_run(cpu
, run
);
2221 if (atomic_read(&cpu
->exit_request
)) {
2222 DPRINTF("interrupt exit requested\n");
2224 * KVM requires us to reenter the kernel after IO exits to complete
2225 * instruction emulation. This self-signal will ensure that we
2228 kvm_cpu_kick_self();
2231 /* Read cpu->exit_request before KVM_RUN reads run->immediate_exit.
2232 * Matching barrier in kvm_eat_signals.
2236 run_ret
= kvm_vcpu_ioctl(cpu
, KVM_RUN
, 0);
2238 attrs
= kvm_arch_post_run(cpu
, run
);
2240 #ifdef KVM_HAVE_MCE_INJECTION
2241 if (unlikely(have_sigbus_pending
)) {
2242 qemu_mutex_lock_iothread();
2243 kvm_arch_on_sigbus_vcpu(cpu
, pending_sigbus_code
,
2244 pending_sigbus_addr
);
2245 have_sigbus_pending
= false;
2246 qemu_mutex_unlock_iothread();
2251 if (run_ret
== -EINTR
|| run_ret
== -EAGAIN
) {
2252 DPRINTF("io window exit\n");
2253 kvm_eat_signals(cpu
);
2254 ret
= EXCP_INTERRUPT
;
2257 fprintf(stderr
, "error: kvm run failed %s\n",
2258 strerror(-run_ret
));
2260 if (run_ret
== -EBUSY
) {
2262 "This is probably because your SMT is enabled.\n"
2263 "VCPU can only run on primary threads with all "
2264 "secondary threads offline.\n");
2271 trace_kvm_run_exit(cpu
->cpu_index
, run
->exit_reason
);
2272 switch (run
->exit_reason
) {
2274 DPRINTF("handle_io\n");
2275 /* Called outside BQL */
2276 kvm_handle_io(run
->io
.port
, attrs
,
2277 (uint8_t *)run
+ run
->io
.data_offset
,
2284 DPRINTF("handle_mmio\n");
2285 /* Called outside BQL */
2286 address_space_rw(&address_space_memory
,
2287 run
->mmio
.phys_addr
, attrs
,
2290 run
->mmio
.is_write
);
2293 case KVM_EXIT_IRQ_WINDOW_OPEN
:
2294 DPRINTF("irq_window_open\n");
2295 ret
= EXCP_INTERRUPT
;
2297 case KVM_EXIT_SHUTDOWN
:
2298 DPRINTF("shutdown\n");
2299 qemu_system_reset_request(SHUTDOWN_CAUSE_GUEST_RESET
);
2300 ret
= EXCP_INTERRUPT
;
2302 case KVM_EXIT_UNKNOWN
:
2303 fprintf(stderr
, "KVM: unknown exit, hardware reason %" PRIx64
"\n",
2304 (uint64_t)run
->hw
.hardware_exit_reason
);
2307 case KVM_EXIT_INTERNAL_ERROR
:
2308 ret
= kvm_handle_internal_error(cpu
, run
);
2310 case KVM_EXIT_SYSTEM_EVENT
:
2311 switch (run
->system_event
.type
) {
2312 case KVM_SYSTEM_EVENT_SHUTDOWN
:
2313 qemu_system_shutdown_request(SHUTDOWN_CAUSE_GUEST_SHUTDOWN
);
2314 ret
= EXCP_INTERRUPT
;
2316 case KVM_SYSTEM_EVENT_RESET
:
2317 qemu_system_reset_request(SHUTDOWN_CAUSE_GUEST_RESET
);
2318 ret
= EXCP_INTERRUPT
;
2320 case KVM_SYSTEM_EVENT_CRASH
:
2321 kvm_cpu_synchronize_state(cpu
);
2322 qemu_mutex_lock_iothread();
2323 qemu_system_guest_panicked(cpu_get_crash_info(cpu
));
2324 qemu_mutex_unlock_iothread();
2328 DPRINTF("kvm_arch_handle_exit\n");
2329 ret
= kvm_arch_handle_exit(cpu
, run
);
2334 DPRINTF("kvm_arch_handle_exit\n");
2335 ret
= kvm_arch_handle_exit(cpu
, run
);
2341 qemu_mutex_lock_iothread();
2344 cpu_dump_state(cpu
, stderr
, CPU_DUMP_CODE
);
2345 vm_stop(RUN_STATE_INTERNAL_ERROR
);
2348 atomic_set(&cpu
->exit_request
, 0);
2352 int kvm_ioctl(KVMState
*s
, int type
, ...)
2359 arg
= va_arg(ap
, void *);
2362 trace_kvm_ioctl(type
, arg
);
2363 ret
= ioctl(s
->fd
, type
, arg
);
2370 int kvm_vm_ioctl(KVMState
*s
, int type
, ...)
2377 arg
= va_arg(ap
, void *);
2380 trace_kvm_vm_ioctl(type
, arg
);
2381 ret
= ioctl(s
->vmfd
, type
, arg
);
2388 int kvm_vcpu_ioctl(CPUState
*cpu
, int type
, ...)
2395 arg
= va_arg(ap
, void *);
2398 trace_kvm_vcpu_ioctl(cpu
->cpu_index
, type
, arg
);
2399 ret
= ioctl(cpu
->kvm_fd
, type
, arg
);
2406 int kvm_device_ioctl(int fd
, int type
, ...)
2413 arg
= va_arg(ap
, void *);
2416 trace_kvm_device_ioctl(fd
, type
, arg
);
2417 ret
= ioctl(fd
, type
, arg
);
2424 int kvm_vm_check_attr(KVMState
*s
, uint32_t group
, uint64_t attr
)
2427 struct kvm_device_attr attribute
= {
2432 if (!kvm_vm_attributes_allowed
) {
2436 ret
= kvm_vm_ioctl(s
, KVM_HAS_DEVICE_ATTR
, &attribute
);
2437 /* kvm returns 0 on success for HAS_DEVICE_ATTR */
2441 int kvm_device_check_attr(int dev_fd
, uint32_t group
, uint64_t attr
)
2443 struct kvm_device_attr attribute
= {
2449 return kvm_device_ioctl(dev_fd
, KVM_HAS_DEVICE_ATTR
, &attribute
) ? 0 : 1;
2452 int kvm_device_access(int fd
, int group
, uint64_t attr
,
2453 void *val
, bool write
, Error
**errp
)
2455 struct kvm_device_attr kvmattr
;
2459 kvmattr
.group
= group
;
2460 kvmattr
.attr
= attr
;
2461 kvmattr
.addr
= (uintptr_t)val
;
2463 err
= kvm_device_ioctl(fd
,
2464 write
? KVM_SET_DEVICE_ATTR
: KVM_GET_DEVICE_ATTR
,
2467 error_setg_errno(errp
, -err
,
2468 "KVM_%s_DEVICE_ATTR failed: Group %d "
2469 "attr 0x%016" PRIx64
,
2470 write
? "SET" : "GET", group
, attr
);
2475 bool kvm_has_sync_mmu(void)
2477 return kvm_state
->sync_mmu
;
2480 int kvm_has_vcpu_events(void)
2482 return kvm_state
->vcpu_events
;
2485 int kvm_has_robust_singlestep(void)
2487 return kvm_state
->robust_singlestep
;
2490 int kvm_has_debugregs(void)
2492 return kvm_state
->debugregs
;
2495 int kvm_max_nested_state_length(void)
2497 return kvm_state
->max_nested_state_len
;
2500 int kvm_has_many_ioeventfds(void)
2502 if (!kvm_enabled()) {
2505 return kvm_state
->many_ioeventfds
;
2508 int kvm_has_gsi_routing(void)
2510 #ifdef KVM_CAP_IRQ_ROUTING
2511 return kvm_check_extension(kvm_state
, KVM_CAP_IRQ_ROUTING
);
2517 int kvm_has_intx_set_mask(void)
2519 return kvm_state
->intx_set_mask
;
2522 bool kvm_arm_supports_user_irq(void)
2524 return kvm_check_extension(kvm_state
, KVM_CAP_ARM_USER_IRQ
);
2527 #ifdef KVM_CAP_SET_GUEST_DEBUG
2528 struct kvm_sw_breakpoint
*kvm_find_sw_breakpoint(CPUState
*cpu
,
2531 struct kvm_sw_breakpoint
*bp
;
2533 QTAILQ_FOREACH(bp
, &cpu
->kvm_state
->kvm_sw_breakpoints
, entry
) {
2541 int kvm_sw_breakpoints_active(CPUState
*cpu
)
2543 return !QTAILQ_EMPTY(&cpu
->kvm_state
->kvm_sw_breakpoints
);
2546 struct kvm_set_guest_debug_data
{
2547 struct kvm_guest_debug dbg
;
2551 static void kvm_invoke_set_guest_debug(CPUState
*cpu
, run_on_cpu_data data
)
2553 struct kvm_set_guest_debug_data
*dbg_data
=
2554 (struct kvm_set_guest_debug_data
*) data
.host_ptr
;
2556 dbg_data
->err
= kvm_vcpu_ioctl(cpu
, KVM_SET_GUEST_DEBUG
,
2560 int kvm_update_guest_debug(CPUState
*cpu
, unsigned long reinject_trap
)
2562 struct kvm_set_guest_debug_data data
;
2564 data
.dbg
.control
= reinject_trap
;
2566 if (cpu
->singlestep_enabled
) {
2567 data
.dbg
.control
|= KVM_GUESTDBG_ENABLE
| KVM_GUESTDBG_SINGLESTEP
;
2569 kvm_arch_update_guest_debug(cpu
, &data
.dbg
);
2571 run_on_cpu(cpu
, kvm_invoke_set_guest_debug
,
2572 RUN_ON_CPU_HOST_PTR(&data
));
2576 int kvm_insert_breakpoint(CPUState
*cpu
, target_ulong addr
,
2577 target_ulong len
, int type
)
2579 struct kvm_sw_breakpoint
*bp
;
2582 if (type
== GDB_BREAKPOINT_SW
) {
2583 bp
= kvm_find_sw_breakpoint(cpu
, addr
);
2589 bp
= g_malloc(sizeof(struct kvm_sw_breakpoint
));
2592 err
= kvm_arch_insert_sw_breakpoint(cpu
, bp
);
2598 QTAILQ_INSERT_HEAD(&cpu
->kvm_state
->kvm_sw_breakpoints
, bp
, entry
);
2600 err
= kvm_arch_insert_hw_breakpoint(addr
, len
, type
);
2607 err
= kvm_update_guest_debug(cpu
, 0);
2615 int kvm_remove_breakpoint(CPUState
*cpu
, target_ulong addr
,
2616 target_ulong len
, int type
)
2618 struct kvm_sw_breakpoint
*bp
;
2621 if (type
== GDB_BREAKPOINT_SW
) {
2622 bp
= kvm_find_sw_breakpoint(cpu
, addr
);
2627 if (bp
->use_count
> 1) {
2632 err
= kvm_arch_remove_sw_breakpoint(cpu
, bp
);
2637 QTAILQ_REMOVE(&cpu
->kvm_state
->kvm_sw_breakpoints
, bp
, entry
);
2640 err
= kvm_arch_remove_hw_breakpoint(addr
, len
, type
);
2647 err
= kvm_update_guest_debug(cpu
, 0);
2655 void kvm_remove_all_breakpoints(CPUState
*cpu
)
2657 struct kvm_sw_breakpoint
*bp
, *next
;
2658 KVMState
*s
= cpu
->kvm_state
;
2661 QTAILQ_FOREACH_SAFE(bp
, &s
->kvm_sw_breakpoints
, entry
, next
) {
2662 if (kvm_arch_remove_sw_breakpoint(cpu
, bp
) != 0) {
2663 /* Try harder to find a CPU that currently sees the breakpoint. */
2664 CPU_FOREACH(tmpcpu
) {
2665 if (kvm_arch_remove_sw_breakpoint(tmpcpu
, bp
) == 0) {
2670 QTAILQ_REMOVE(&s
->kvm_sw_breakpoints
, bp
, entry
);
2673 kvm_arch_remove_all_hw_breakpoints();
2676 kvm_update_guest_debug(cpu
, 0);
2680 #else /* !KVM_CAP_SET_GUEST_DEBUG */
2682 int kvm_update_guest_debug(CPUState
*cpu
, unsigned long reinject_trap
)
2687 int kvm_insert_breakpoint(CPUState
*cpu
, target_ulong addr
,
2688 target_ulong len
, int type
)
2693 int kvm_remove_breakpoint(CPUState
*cpu
, target_ulong addr
,
2694 target_ulong len
, int type
)
2699 void kvm_remove_all_breakpoints(CPUState
*cpu
)
2702 #endif /* !KVM_CAP_SET_GUEST_DEBUG */
2704 static int kvm_set_signal_mask(CPUState
*cpu
, const sigset_t
*sigset
)
2706 KVMState
*s
= kvm_state
;
2707 struct kvm_signal_mask
*sigmask
;
2710 sigmask
= g_malloc(sizeof(*sigmask
) + sizeof(*sigset
));
2712 sigmask
->len
= s
->sigmask_len
;
2713 memcpy(sigmask
->sigset
, sigset
, sizeof(*sigset
));
2714 r
= kvm_vcpu_ioctl(cpu
, KVM_SET_SIGNAL_MASK
, sigmask
);
2720 static void kvm_ipi_signal(int sig
)
2723 assert(kvm_immediate_exit
);
2724 kvm_cpu_kick(current_cpu
);
2728 void kvm_init_cpu_signals(CPUState
*cpu
)
2732 struct sigaction sigact
;
2734 memset(&sigact
, 0, sizeof(sigact
));
2735 sigact
.sa_handler
= kvm_ipi_signal
;
2736 sigaction(SIG_IPI
, &sigact
, NULL
);
2738 pthread_sigmask(SIG_BLOCK
, NULL
, &set
);
2739 #if defined KVM_HAVE_MCE_INJECTION
2740 sigdelset(&set
, SIGBUS
);
2741 pthread_sigmask(SIG_SETMASK
, &set
, NULL
);
2743 sigdelset(&set
, SIG_IPI
);
2744 if (kvm_immediate_exit
) {
2745 r
= pthread_sigmask(SIG_SETMASK
, &set
, NULL
);
2747 r
= kvm_set_signal_mask(cpu
, &set
);
2750 fprintf(stderr
, "kvm_set_signal_mask: %s\n", strerror(-r
));
2755 /* Called asynchronously in VCPU thread. */
2756 int kvm_on_sigbus_vcpu(CPUState
*cpu
, int code
, void *addr
)
2758 #ifdef KVM_HAVE_MCE_INJECTION
2759 if (have_sigbus_pending
) {
2762 have_sigbus_pending
= true;
2763 pending_sigbus_addr
= addr
;
2764 pending_sigbus_code
= code
;
2765 atomic_set(&cpu
->exit_request
, 1);
2772 /* Called synchronously (via signalfd) in main thread. */
2773 int kvm_on_sigbus(int code
, void *addr
)
2775 #ifdef KVM_HAVE_MCE_INJECTION
2776 /* Action required MCE kills the process if SIGBUS is blocked. Because
2777 * that's what happens in the I/O thread, where we handle MCE via signalfd,
2778 * we can only get action optional here.
2780 assert(code
!= BUS_MCEERR_AR
);
2781 kvm_arch_on_sigbus_vcpu(first_cpu
, code
, addr
);
2788 int kvm_create_device(KVMState
*s
, uint64_t type
, bool test
)
2791 struct kvm_create_device create_dev
;
2793 create_dev
.type
= type
;
2795 create_dev
.flags
= test
? KVM_CREATE_DEVICE_TEST
: 0;
2797 if (!kvm_check_extension(s
, KVM_CAP_DEVICE_CTRL
)) {
2801 ret
= kvm_vm_ioctl(s
, KVM_CREATE_DEVICE
, &create_dev
);
2806 return test
? 0 : create_dev
.fd
;
2809 bool kvm_device_supported(int vmfd
, uint64_t type
)
2811 struct kvm_create_device create_dev
= {
2814 .flags
= KVM_CREATE_DEVICE_TEST
,
2817 if (ioctl(vmfd
, KVM_CHECK_EXTENSION
, KVM_CAP_DEVICE_CTRL
) <= 0) {
2821 return (ioctl(vmfd
, KVM_CREATE_DEVICE
, &create_dev
) >= 0);
2824 int kvm_set_one_reg(CPUState
*cs
, uint64_t id
, void *source
)
2826 struct kvm_one_reg reg
;
2830 reg
.addr
= (uintptr_t) source
;
2831 r
= kvm_vcpu_ioctl(cs
, KVM_SET_ONE_REG
, ®
);
2833 trace_kvm_failed_reg_set(id
, strerror(-r
));
2838 int kvm_get_one_reg(CPUState
*cs
, uint64_t id
, void *target
)
2840 struct kvm_one_reg reg
;
2844 reg
.addr
= (uintptr_t) target
;
2845 r
= kvm_vcpu_ioctl(cs
, KVM_GET_ONE_REG
, ®
);
2847 trace_kvm_failed_reg_get(id
, strerror(-r
));
2852 static bool kvm_accel_has_memory(MachineState
*ms
, AddressSpace
*as
,
2853 hwaddr start_addr
, hwaddr size
)
2855 KVMState
*kvm
= KVM_STATE(ms
->accelerator
);
2858 for (i
= 0; i
< kvm
->nr_as
; ++i
) {
2859 if (kvm
->as
[i
].as
== as
&& kvm
->as
[i
].ml
) {
2860 return NULL
!= kvm_lookup_matching_slot(kvm
->as
[i
].ml
,
2868 static void kvm_accel_class_init(ObjectClass
*oc
, void *data
)
2870 AccelClass
*ac
= ACCEL_CLASS(oc
);
2872 ac
->init_machine
= kvm_init
;
2873 ac
->has_memory
= kvm_accel_has_memory
;
2874 ac
->allowed
= &kvm_allowed
;
2877 static const TypeInfo kvm_accel_type
= {
2878 .name
= TYPE_KVM_ACCEL
,
2879 .parent
= TYPE_ACCEL
,
2880 .class_init
= kvm_accel_class_init
,
2881 .instance_size
= sizeof(KVMState
),
2884 static void kvm_type_init(void)
2886 type_register_static(&kvm_accel_type
);
2889 type_init(kvm_type_init
);